Mathematics for Data Science 1

Prof. Madhavan Mukund

Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 01
Natural Numbers and their operations

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, welcome to the 1st week of Mathematics 1 for Data Science. So, we are going to start
with some very basic things which you probably know; right from the beginning we are
going to start talking about numbers. So, in this 1st module what we are going to talk about is
natural numbers and integers.

(Refer Slide Time: 00:30)

So, as you probably remember from as young as you were in school when you first came
across numbers, we use numbers mainly for counting. So, for instance if we see 7 balls like
this and then we see 7 pencils like this, then we need to know that these are the same number
of things and for this we use this number 7. So, 7 represents what is common to these two
objects that there are 7 balls and 7 pencils. So, 7 is an abstract concept in that sense and it
refers to a quantity.

So, we all of course, know the numbers 1, 2, 3, 4 and all that. So, when we see a number of
things, we can count them. But perhaps the most important number of all which is of Indian
origin is 0. So, it is quite important to have a way to represent something when there is
nothing to count because without a 0, we cannot use our place numbering system that we use
to manipulate numbers.

So, these numbers starting with 0 are what are often called the natural numbers. Now there is
some confusion in some books and many books will actually use only 1, 2, 3, 4 to represent
the natural numbers. So, we use N to represent the set of natural numbers and in case there is
any confusion whether 0 is included in this set or not, now sometimes people will not include
0 in the set of natural numbers.

So, sometimes to emphasize that we are using 0, we will actually put the subscript 0 below
the N right. So, we will write either N or N 0, but whenever we are talking about natural
numbers, it always includes a 0. Now what can we do with natural numbers? Well we can
add them, we can subtract them, we can multiply them, we can divide them. So, these are the
normal arithmetic operations which you have studied in school.

(Refer Slide Time: 02:43)

But what is really interesting from a mathematics perspective is, when we take natural
numbers and we perform an operation on them, do we always get a natural number? So, if we
add two natural numbers, do we get a natural number? If we subtract a number from another,
we get a natural number? If we multiply them, do we get a natural number? If we divide one
by another, do we get a natural number?

So, the first operation which fails this test is subtraction because if we subtract a larger
number from a smaller number, so supposing we take 6 and subtract it from 5; then we go
below 0 right. If you have 5 things and we take away 6 things, we will be cannot take away 6
things that is what subtraction means. So, we need to expand the scope of our numbers to
allow these operations to work sensibly and this is how we get the negative numbers.
So, we had the positive numbers 0, 1, 2 the non-negative technically because 0 is neither
positive nor negative. So, we had the positive numbers 1, 2, 3, 4. We added a 0 to account for
the fact that we are counting nothing and now we add symmetrically on the other side
negative numbers -1, -2, -3. So, this is just to illustrate why we get them of course, this is
something that you should know from school.

So, this set which is the natural numbers extended with a negative numbers is what we call
the integers and we use Z to indicate the set of integers. So, we have N the set of natural
numbers which starts at 0 and goes forward 0, 1, 2, 3, 4 and we have the integers which start
at no at minus infinity and go to plus infinity. So, these are both infinite sets, but the natural
numbers have a starting point 0 and the integers extend to infinity in both directions.

So, it is very convenient mentally to think of the integers as forming this kind of a sequence
where on the left you have the very small ones and on the right you have the very long ones
and this is normally called the number line. So, as you go from left to right, the numbers are
increasing and this is how the integers are arranged.

(Refer Slide Time: 04:27)

So, we said that subtraction takes us away from natural numbers and we brought the integers.
So, now, let us look at the other two operations that we talked about multiplication and
division. So, let us start with multiplication. So, when we say 7 × 4 what we are really saying
is take 7 objects and make 4 copies of them. So, for instance on the right, we have those 7
balls that we started with and then we have made 4 copies of them. So, if we want to know
how many balls are here, then we have 7 from the first group, 7 from the second group and so
on. So, we have 4 groups of 7 and this is if we add it up going to be 28.

(Refer Slide Time: 05:02)

So, in general this is how we multiply when we take a number m and multiply it by n, what
we are doing is we are making n copies of m. So, we are taking m + m + m… n times. So, in
this sense multiplication is repeated addition.

So, we often use this time sign the × sign for multiplication, but this is often cumbersome
when we write out equations. So, sometimes we replace this time sign by a . and sometimes
we write nothing at all. So, if we just write two symbols together, we do not write this
normally for numbers because imagine that if I write 7 4 like this, then you do not know
whether it is a number 74 or its 7 × 4. So, if we have numbers, we will normally write a dot
explicitly between them like 7 × 3. But when we have a names like m or n standing for
numbers, then if we write mn; we assume that it is one number m multiplied by another
number n.
(Refer Slide Time: 05:56)

Now, we have integers, an integers have signs they are positive and negative numbers. So, we
have to remember that when we multiply numbers with signs, the resulting number also has a
sign and there is a sign rule which basically says that if we have one negative number
multiplied by one positive number, then the result is a negative number. So, let us assume
that m is a positive number so, - m is a negative number. So, say - 7 × 4 would be - 28. On
the other hand if I take (- 7) × (- 4), then the two negations will cancel, and I will get 28.

So, if you have an even number of minus signs, you get a positive number; if you have an
odd number of minus signs, you get a negative number.
(Refer Slide Time: 06:40)

Now just like we have repeated addition, we can also do repeated multiplication. So, instead
of doing m plus m, we can takes m times m and this is called m squared and the reason that it
is called m squared is visible in the picture here. So, we have now here 6 balls and 6 balls. So,
we have 6 × 6 right. So, this means that we can arrange these 6 times 6 balls in a square and
this is why we call this square.

(Refer Slide Time: 07:15)

So, this notation m 2stands to the fact that, m is multiplied by itself twice. Now if you
multiply it by self 3 times, then we get a cube. So, here for instance we have 3 balls by 3 balls
and then we have a height a stack of 3 such balls. So, we have a square of 3 by 3, 9 balls and
we have 3 stacks of these one on top of the other. So, this naturally forms a cube so, m × m ×
m is written m 3.

(Refer Slide Time: 07:37)

Now, unfortunately we live in a 3-dimensional world and we cannot imagine objects which
have more than 3 dimensions. So, our vocabulary stops with cube. So, in general if we have
m , then we write m × m × m…, k times and we just say it is m , we do not have a fancy name
k k

for it. We just say it is the kth power of m ok. So, to emphasize multiplication is repeated
addition and exponentiation as we have seen is repeated multiplication.
(Refer Slide Time: 08:07)

So, now let us come to division. So, you would have seen this familiar problem in school.
You have a certain number of objects and you want to divide them among certain number of
people. So, for example, supposing you have 20 mangoes and you want to give them to 5
friends. So, how many mangoes does each friend get?

(Refer Slide Time: 08:31)

So, here on the right we have this picture and then, what you do is well you start by
distributing one mango to each friend right. So, you take out 5 mangoes and you give them to
each of your friends. So, now, you have given away 5 mangoes and you have only 15
mangoes left so, you repeat the process. Among the 15 mangoes, you give away 5 to your
friends one each and now your 15 mangoes have become 10 and do it one more time and
your 10 mangoes have become 5, do it a third time or fourth time rather and the 5 mangoes
are now gone.

So, after 4 rounds of distributing mangoes, each time giving one mango each so, 5 mangoes
per round, you have got rid of your 20 mangoes so, 20 ÷ 5 is 4. So, here as we have
illustrated, division is actually repeated subtraction. You keep subtracting by the number you
are trying to divide and finally, if you hit 0, then you have divided it exactly.

Well, what if you had only 19 mangoes? Now you know very well that 19 mangoes cannot be
evenly divided into 4 into 5 groups. So, if you would start distributing like we had above the
first three rounds would go fine; you would come from 19 to 14 from 14 to 9 and then you
will come from 9 to 4 and now you have only 4 mangoes left and you have 5 friends so, you
cannot give 1 each.

So, we have managed to distribute 3 times and we have 4 left over. So, formally this is
written as saying that the quotient the number of times you can actually divide without
getting into a fractional part is 3 and the remainder that is after you have a little bit left over
which you cannot subtract one more time is the remainder is 4. So, for 19 ÷ 5, the quotient is
3 and the remainder is 4.

Now, very often we will need to use this remainder and there is a notation for remainder. So,
this is this notation called modulus. So, modulus is another word for remainder and it is
written as mod. So, 19 mod 5 is the same as the remainder when 19 is divided by 5. So,
instead of saying the remainder of 19 divided by 5 is 4, we will often say 19 mod 5 is 4.
(Refer Slide Time: 10:38)

So, with this notation, we can now define what is a factor. So, a factor is a number which
divides a bigger number evenly without any remainder. So, a | b, if b mod a is 0. Remember
what this mean is means is that if b is divided by a, there is no remainder and we write this
with this vertical bar |. So, on the left is the smaller number, on the right is the bigger number.
So, a divides b this is what this is supposed to say and the other way of thinking about it is
that b is some multiple of a. So, b if a | b then a.k=b ok. So, we have some multiple the some
number of times that a goes into b. So, therefore, b is a multiple of a.

(Refer Slide Time: 11:27)

So, here are some examples we have already seen that 4 |20 because 4 × 5 is 20, 7 | 63
because 7 × 9 is 63, 32 |1024 because 32 × 32 is 1024 and so on.

Now, the symbol that we use for not being a divisor is just to put a stroke across that vertical
line. So, 4 does not divide 19 because there is no way to multiply anything by 4 and get 19.
Similarly, 9 does not divide 100 evenly because we get 9 ×11 = 99 and then we go 108.

So, we say formally that a is a factor of b if a | b right. So, a | b is the same as saying that a is
a factor of b and it is easy to see that factors must come in pairs because if a | b then, a goes
into b some k times. So, k | b right so, k × a = b so, both k is a factor and a is a factor. So, for
instance, if you take a number 12 then 1 is a factor because 1 divides everything and in fact,
for every number n, 1 × n is n so, the pair for 1 is always the number itself.

Now in this case, 12 is divisible by 2 and 2 goes in 6 times. So, the pair 2, 6 form 2 factors 6
times 2 is 12, 2 × 6 is 12 and similarly 3 × 4. Now, of course, there is an important side
condition which is that sometimes the pair is the same as the number itself and this happens
when the number actually happens to be a perfect square that is, it is some number multiplied
by itself. So, for instance consider 36 so, 36 is 6 × 6. So, if you look at the factors of 36 and
group them in pairs, then we have 1 and 36, we have 2 and 18, we have 3 and 12, we have 4
and 9 and finally, we have the factor 6, but 6 is multiplied by 6. So, 6 does not produce a new
factor as its pair, it is just itself.

So, another way of thinking about it is that, if you have something which is not a square you
will have an even number of factors, you will have 2 + 2 + 2 +2. If something is a square, you
will have an odd number of factors, you will have 2 + 2 +2 and finally, when you come to the
number of which it is a square that number will come only once in the list of factors.
(Refer Slide Time: 13:45)

So, once we talk about factors, we come to a very interesting class of numbers which are the
prime numbers. So, a prime number is one which has no factors other than 1 and itself. So, 1
is a factor always and 1 × n is n. So, for we try to usually write p for a prime number. So, a
prime number has only two factors 1 and p.

Now, it is important that it must have two factors, two separate factors. So, one technically is
not a prime because it has only one factor one itself because 1×1 is 1 and so, the only factor
that 1 has is 1. So, the smallest prime actually is 2 because it has two factors 1 and itself 2
and no other factors. 3 is also a prime because it has only 2 factors 1 and 3, 2 does not go into
3 and so on.

So, we are all familiar with the smaller prime numbers. So, 2 is the first prime number, 3 is
the next prime number, then 5, then 7. Notice that, after 2 no even numbers can be primes
because they are all multiples of 2 and so, 2 divides them. Now we come to 9 and 9 is not a
prime number because it is a multiple of 3, but 11 is a prime number and so on.

So, there is actually one clever way which is call the sieve of Eratosthenes to generate prime
numbers which is whenever you discover a prime, you knock off all the numbers which are
multiples of it. So, we can do this for instance to get all the prime numbers from 1 to 100. So,
what we do is we first lay out a grid like this right, we know that 1 is not a prime so, the first
prime that we have as a candidate is 2 right. So, this is how the sieve of Eratosthenes works,
you lay out the numbers in a grid and now we can try and mark off all the prime numbers
which are up to 100.

So, we know that 1 is not a prime so, we leave 1 off the grid and we start with 2. So, 2 is our
first prime number and what the sieve of Eratosthenes says is you knock off all multiples of
2. So, you knock off all the even numbers and of course, now you can do it in one shot so,
you can knock off this whole column, this whole column so, all these numbers are not prime
ok.

So, now once you have you have a target so, we are looking only up to 100. So, up to 100 we
have knocked off all the powers of 2 or all the multiples of 2. So, now, we look at the first
number which is not been marked off and we notice that 3 is a prime because 3 is not yet
marked off. So, now, we start mark off multiples of 3, some of them are already marked off
because they are multiples of 2. So, 6 is already gone, but 9 is also gone, 12 is already gone,
but 15 is also gone and so on.

So, we can mark off all the other multiples of 2 which are not multiples of 3 and so on right.
So, we get this kind of a picture and now having done this assuming we have done it all the
way, then we will come and find that 5 is a prime right. So, this is the process by which if you
want to know count all the primes up to a certain number n, you can write out all the numbers
up to n and starting at the left you can take the first unmarked number, call it a prime and
mark all its multiples to the right as non primes and the next unmarked number will be the
next prime and so on.
(Refer Slide Time: 16:56)

Now, this is not necessarily an efficient way to do the prime numbers, but this is a good way
to generate them without missing out any. One of the important facts that we use all the time
is that every number can not only be factorized as we have seen into a number of different
pairs of factors it can actually factorize uniquely into the prime numbers that form it.

So, for instance if we look at 12, we said that 12 was 2 times 6, it was also 4 times 3, it was 1
times 12 and so on, but fundamentally it has 3 prime factors 2 2 again and 3. So, depending
on how we combine them for instance, we can get 4 × 3 or we can get 2 × 6 and so on, but 2
× 2 × 3 is the absolute unique way of writing 12 as a product of prime numbers and using our
exponentiation notation, we can condense this and put the 2 2’s together and say it is22 ×3.

Similarly, if we take a number like 126, then it is 2 × 3, 6 × 3, 18 ×7 ok. So, the prime factors
are precisely 2 3 twice and 7 and we can write this as 2× 32 × 7. So, this is very important
because we use it implicitly along a lot and we will see later how we use this.
(Refer Slide Time: 18:10)

So, this is called the prime factorization right. So, every integer can be decomposed into a
product of primes in a unique way.

(Refer Slide Time: 18:20)

So, to summarize we started with a natural numbers which we use for counting which are the
numbers 0, 1, 2, 3, 4 and so on. Then, we extended these numbers with a negative numbers
and this gave us the set of integers. So, the integers include all the natural numbers as well as
the negative numbers 0, 1, 2, 3 and so on -1, -2, -3 and so on. We saw some basic arithmetic
operations on these the usual addition, subtraction, multiplication, division and
exponentiation.

We also looked at what happens when we divide integers and we do not want to look at
fractions, then we talk about the quotient which is the integer number of times that the
dividend goes into the number and the remainder is also written as a mod b. So, using this
notation of a mod b, we can talk about divisibility which we write with a vertical bar. So, a | b
if a mod b is 0. So, the factors of a number are those numbers which divide it and a prime
number has exactly two factors 1 and itself and we can always decompose any integer
uniquely into the list of factors, prime factors that multiply out to form that number.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture - 02
Rational Number

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, now first lecture on Numbers; we looked at natural numbers and integers. So, now, let see
what happens when we try to divide. So, let us look at the rational numbers.

(Refer Slide Time: 00:23)

So, we said that we cannot represent 19 / 5 as an integer because we cannot find a number k
such that 5 × k is 19. So, as we know the way we deal with this is to represent this quantity as

4
a fraction. So, we say that 19 /5 is 3 . So, this number is an example of a rational number.
5
So, rational number what we usually called fractions in school, a rational number is

p
something that can be written as , ; where, p and q are both integers. So, as you probably
q

p
remember from school, the number on the top is called the numerator. So for, ; p is called
q
the numerator and q is called the denominator.

So, just like we had the symbols N and Z to represent the natural numbers and the integers,
we have a special symbol which is somewhat unusual which is Q. So, Q stands for the
rational numbers and again, to just say it is a special Q, we write these double lines on sides.
So, this Q with these fat boundaries denotes the rational numbers. So, one thing about the
rational numbers is that the same number can be written in many different ways. Now, this is
not true of integers. Of course, we are not talking about changing base from binary to decimal
or something.
But if you write a 7, there is only one way to write 7 fix, if you are fix the notation that you
are using for writing numbers. With rational numbers, this is not true because there are many

p p
ways of writing such that is actually a same number. So, for instance if we take the
q q

3 3 6 30
number , then we all know that is the same as and this is the same as . So, when we
5 5 10 50
take a rational number and multiply it by something the same quantity on the top and the

3 6
bottom so, , 3 × 2 and 5 × 2, we get the same number; or 3 × 10 and 5 × 10, we get the
5 10

30
same number . So, this is sometimes a nuisance, but it is also sometimes useful.
50

3 3 2 3
Now, there is no reasonable way to compare two numbers like say and or and . If we
5 4 5 4
have two fractions which have different denominators, there is no way to directly compare
them. So, the only way to compare them is to somehow convert them into equivalent
fractions such that they have the same denominator. So, the usual way is just to find a number
such that both the denominators multiply into that number rather factors of that number.
Now, you can find the smallest such number which is called the least common multiple; but
you can find any number of this form.

(Refer Slide Time: 03:05)

 3 3
So, for instance, if you want to add and , now you cannot do that directly; but you know
5 4
that 20 is a number which divides which is a multiple of both 5 and 4. So, you can represent

3 12 3
as equivalently as ; you can represent equivalent. So, this is equivalent and this is
5 20 4
equivalent. So, you have converted these numbers into a different fraction of the same
number; but this new representation has the same denominator.

And now once, the two denominator that the same, you can add the numerators and you can

27
get (12 + 15)/20 is . So, this kind of manipulation requires the denominators to be the
20
same and therefore, it is actually extremely useful that we can write the same rational number
in many different ways. The same is to we want to compare two numbers.

(Refer Slide Time: 03:54)

3 3
If we want to check whether is bigger or smaller than , there is no way to do it directly.
5 4
What we have to do is again take the denominators and make them the same and then, say

12 15
that is less than because you are dividing something 20 parts and you are taking 12 of
20 20
them that is less than taking 15. Now, as I said there is no reason why this must be the
smallest one. So, for instance you could take a bigger number like 100, right. So, 5 goes into
100 and 5 goes 4 also goes into 100.
 3 60 3 75
So, we could also say that is the same as and, is the same as and therefore,
5 100 4 100

60 75 3 3
since 60 is less than 75; is less than and therefore, is less than . So, it is not
100 100 5 4
really important that the denominator is the smallest common multiple of the two
denominators; but it must be some common multiple so that you can bring it all to a common
number that you can then compare.

(Refer Slide Time: 04:49)

So, we saw that representation is not unique for rational numbers. So, how do we find
actually the best way to represent a rational number? So, normally if you are not using it for
some arithmetic operation of some comparison, we would prefer to have it in a reduced form.
So, the reduced form of a rational number is one, where there are no common factors

p
between the top and the bottom. So, is of the form, where we cannot find any factor f such
q
that f | p and f | q.

18 3
So, for instance, if we take , then its reduced form will be . Notice that 3 is of the form 3
60 10
× 1 and 10 is of the form 5 × 2 × 1. So, therefore, there is no common factor between the top
and the bottom and therefore, this is in reduced form.
(Refer Slide Time: 05:42)

So, this is called the greatest common divisor problem. So, we want to find the largest
number which divides both the top and the bottom; both the numerator and the denominator;
divide them both by this and then come to something in the reduced form. So, in this case,
what we are saying is that the gcd of 18 and 60 is actually 6 and we can do this using our
prime factorization that we talked about before.

So, if we look at prime factorization for 18, then 18 is 2 × 3 × 3 right; its 2 × 3 is 6 and 6 × 3
is 18 and the prime factorization of 60 is 2 × 2 ×3 × 5; its 4 × 3, 12 and 12 × 5. So, now, you
can look at what are common. So, we have one 2 here and one 2 here. So, we can say that this
is part of the same factor, we have one 3 here and another 3 there. The second 2 is not present
in the first term.

So, we have a 2 and 3 and 18 which are factors. We have a 2 and 3 in 60 which are factors
and this gives us the fact that 6 is a common factor. There is no bigger common factor
because we want to assemble a bigger common factor, we have to pull out one more prime
from each side; but there is no prime left which is present on both sides. 3 is there in 18; 2
and 5 are there on 60, but we do not have a matching one of the other side right.
(Refer Slide Time: 06:59)

So, this way, the common prime factors are one 2 and one 3 and so, 2 × 3 equal to 6 is the
gcd. Now, this is not the best way to find the gcd, there are more efficient ways to find the
gcd. But this intuitively tells us what the gcd is. You take the prime factorization of both the
numbers and you collect together all the primes that occur in both the numbers, the same
number of times.

(Refer Slide Time: 07:23)

So, here is another interesting property about rational numbers. Now, for each integer, we
know intuitively that there is something which is the next integer and the previous integer. If
I tell you 22 and ask you what is the next integer? Then, you will know it is 23. What is the
previous one? It will be 21. So, for every integer m, the next one is m + 1 and the previous
one is m - 1 and it does not matter, if this is positive or negative. So, for instance if I am at
17, then the next integer is 18, the previous one is 16; right. If I am at -1, then the next integer
is 0 and the previous integer is -2. So, I can always take the integer that I am at, add 1 and get
the next integer, subtract 1 you will get the previous integer.

(Refer Slide Time: 08:10)

So, the property of this next and previous is that there is nothing in between right. So, there is
no integer between m and m +1, there is no integer between m and m -1. So, that is what next
means, it is not some bigger integer or some smaller integer. It is the immediate neighbor in
the integer of the in this number line. Now, what about rationals? Is it possible to talk about
the next and the previous rational number? Now, it turns out that this is not possible for a
very simple reason.

So, between any two rationals, we can always find another one because we can always take
the average of 2 numbers. So, remember that if you take the average of any 2 numbers, then it
must be between those 2 numbers right because it is the sum of the numbers divided by 2. So,
the average cannot be smaller than both or cannot bigger than both. So, if the 2 numbers are
not the same, then it must lie strictly between them. If the numbers are the same, then the
average is the same.
So, if somebody has 37 marks and 37 marks, then their average marks is 37. But if they have
37 marks and 52 marks, even without calculating the average, you know that their average is

m
bigger than 37, but smaller than 52; right. So, in the same way, if I give you 2 fractions
n

p m p
and and I tell you that is smaller than . Remember that in order to do this, we would
q n q
have to normally get the denominators to be the same and so on.

m p m
But supposing I know that is smaller than . So, I know that say is here and I know that
n q n

p m p p
say is here and supposing you claim that and are adjacent, that is is the next rational
q n q q

m
after . Well, I will say no; let me take these 2 numbers and find its average right. So, this
n

a
average now is also a rational number because you can also represent it as right. If you just
b

m p
workout this plus divided by 2, you can simplify this whole expression and you will get
n q

a
a new number which is also of the form .
b

So, this is also a rational number and this rational number as we argued must be between the
2 numbers and therefore, between any 2 rational numbers by just taking the average of the
mean of the 2 numbers, I can find another one.
(Refer Slide Time: 10:17)

So, in other words, the rational numbers are dense right. So, dense in the usual sense, so
dense just means that they are closely packed together. So, basically you cannot find any gaps
in the rational numbers because any between any 2 rational numbers, you will find another
rational number and this is not true of the integers because we saw that in the number line,
there is a gap between m and m + 1, there is no integer there right. So, we say that the rational
numbers are dense and conversely, we say that the integers and the natural numbers are
discrete. So, a discrete set has this kind of next property and a dense set has no next property
between any 2 numbers, will find another number right.

(Refer Slide Time: 10:56)

To summarize, we use this funny symbol Q to denote the rational numbers and a rational
number is just the ratio. So, that is where it comes from actually; so, ratio. So, rational
number comes from the word ratio and so, it is a ratio of 2 integers p divided by q. Now,
there is no unique representation of a rational number because we can multiply both the
numerator and the denominator by the same quantity and get a new rational number which is
exactly the same in terms of the quantity that it represents.

And we use this fact for things like arithmetic and comparisons, but if we really want to talk
about rational numbers in a canonical way, in a unique way; then, we get this reduced form,
where we cancel out the common factors using prime factorization. So, that we get a number
whose gcd of the numerator and the denominator is 1.

And finally, we saw that we cannot talk about the next or the previous rational number
because between any 2 rational numbers, there is another rational number. In particular, if
you take the average of the 2 numbers, you will find a number that is in between. So, unlike
the integers and the natural numbers which are discrete for which next and previous makes
sense; for the rational numbers, there is no such quantity.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture - 03
Real and Complex Number

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, we started with the natural numbers and the integers and then, we moved on to the

p
rational numbers which are defined as ; where p and q are both integers.
q

(Refer Slide Time: 00:23)

So, we decided that the rational numbers are dense right and that means that on this number
line between any two rationals, you can find a rational. So, if I want to now talk about this
number line, then I know that if I take any two positions, then I will find a rational between
them and I will find a rational between them and so on. So, it makes sense to ask this
question which is that if I take any two points and the rational between them any two points,
then is this entire number line composed only of rational numbers. Of course, some of those
rational numbers are integers.

7
So, an integer is a rational number because I can write 7; for instance, as right. So, this is of
1

p
the form . So, any rational number which in reduced form as denominator 1 is an integer;
q
so, an integer is a special case of a rational number. So, do all the rational numbers fill up this
number line? That is the question.
(Refer Slide Time: 01:12)

So, it turns out this is not the case. So, remember that a square of a number is the number
multiplied by itself. So, if I take a number m and multiply it by itself, I get m 2 which is m ×
m and if I take this operation and turn it around, then the square root of a number is that
number r such that r × r is equal to m right. So, I want to find out which number, I have to
square in order to get m and that is called the square root.

So, if we take the so called perfect squares, like 1, 4, 9, 16, 25 and so on their square roots are
integers. So, 12 is 1. So, the √1 is 1; 22is 4, so the √4 is 2; 52is 25, so √25 is 5; 16 2is 256, so √
256 is 16 and so on. So, some integers are clearly squares of other integers and so, you can
get the square root and find an integer. Now, what happens if something is not a square right?
So, supposing I take a number which is not a square like 10 and I take its square root, I know
that the square root is not an integer, its somewhere between 3 and 4 because 32is 9 and 42is
16. Question is, is it a rational number or not?
(Refer Slide Time: 02:36)

So, what happens to the square roots of integers that are not perfect squares? So, the smallest
such number which is not a perfect square because 1 remember is a perfect square, 1 × 1 is 1.
The smallest such number that is not a perfect square is actually 2 and it is one of the very old

p
results that the √2 cannot be written as . This was certainly known to the ancient Greeks, in
q
fact, to Pythagoras and one way to do this is to see that you can actually draw a line of; so,
this is not an unreal number in that sense right.

So, you can actually draw a line of this length because if you take a square, whose sides are 1
right. So, this is 1, then if you remember your Pythagoras theorem; then, the hypotenuse of
this triangle is going to be √ 12 +12 ,technically which is √2. So, I can actually physically draw
a line whose length is √2. So, this is a very real quantity.

On the other hand, for reasons that we will not described here, but there will be a separate

p
lecture explaining this for if you are interested. √2 cannot be written as a rational number .
q
So, here is a number which is a very measurable quantity, I can actually draw this quantity as
a length. At the same time, it does not fit into this number line of rational numbers which
seems to cover all the rational numbers, all the numbers because they are dense.
(Refer Slide Time: 03:56)

So, √2, since it is not a rational number right and yet it exists is called an irrational number
and these numbers which constitute all the rational numbers and the real irrational numbers
together are called the real numbers. So, the real numbers are denoted by this double line R.
So, we had N for the natural numbers, Z for the integers, Q for the rational numbers and now,
we have the real numbers R.

So, the real numbers extend the rational numbers by these so called irrational numbers which

p
are very much on the number line, but which cannot be written on the form . Now, it is not
q
difficult to argue that like the rationals, the real numbers are dense for the very same reason.
Because if you have two real numbers r and r ' such that r is smaller than r prime, then you
can just take their average r + r ' divided by 2. This must be a number πwhich is bigger than r
and it is smaller than r ' and therefore, it must lie between them. So, between any 2 real
numbers, you will find another real number. So, the real numbers are also dense.
(Refer Slide Time: 05:00)

So, there are some irrational numbers which we use a lot in mathematics and which you have
probably come across; one of them is this famous number π which comes when we are
talking about circles. Because it is the ratio of the circumference to the diameter and this is an
invariant. π is always; the circumference divided by diameter for any circle is π ok.

p
So, π is an irrational number. We cannot write it in the form and it has this. If you write it
q
in this decimal form, it has this infinite decimal expansion. Another number which is very
popular as an irrational number is this number e which is used for natural logarithms. So, it is
2.7182818 and so on right. So, there are a lot of irrational numbers. So, √2 as we have seen
as an irrational number. It will turn out that square root of anything, √3 is also an irrational
number, √6 is also an irrational number.

Anything which is not a perfect square, its square root is actually an irrational number. But
many of these numbers are not very useful to us, but π and e are certainly very useful
irrational numbers. So, now, we have seen that we can find more numbers on the line than
just the rationals and these are the real numbers. So, do we stop here? Well, let us look at the
square root operation which we use in order to claim that there are irrational numbers. So,
what happens if we now take the square root of a negative number like -1?

So, remember that we had a sign rule for multiplication. The sign rule for multiplication said
that if I multiply any two numbers, then if the two signs are the same that is their two
negative signs or two positive signs, I will get a positive sign in the answer. Only if the two
signs are different, if I have one minus sign and one plus sign, will I get a negative answer.
So, if I want to multiply two numbers and get a -1, one of them must be negative and one
must not be negative. But by definition, a square root is a number which is multiplied by
itself, the same number has to be multiplied by itself. So, it will have the same sign.

So, any square root which multiplies by itself must give me a positive number. So, if I take a
negative number, there is no way to find a square root for it. So, if we want to find square
roots for negative numbers, we have to create yet another class of numbers called complex
numbers. So, complex numbers extend the real numbers, just like real numbers extend the
rational numbers and rational numbers extend the integers and so on. But the good news for
you is that we do not have to look at complex numbers for this course.

(Refer Slide Time: 07:35)

So, to summarize, a real numbers extend the rational numbers by adding the so called

p
irrational numbers which cannot be represented of the form and a typical example of an
q
irrational number is the square root of an integer that is not a perfect square. So, √2 for
example is not a rational number and this is also of the case was √3, √5, √6 and so on. So,
except for the perfect squares, none of the square roots are actually rational numbers. Now,
just like we said that the rational numbers are dense because the average of any two rational
numbers is a rational number. Similarly, the real numbers are dense because the average of
any two real numbers is a real number.

So, we have a progression in terms of numbers. So, every natural number that we started with
is also an integer because the integers extend the natural numbers with negative quantities.
Now, every integer is also a rational number because we can think of every integer as a ratio

p
; where, the denominator is 1. And finally, every rational number is a real number because
q
we said that the real numbers include all the rational numbers plus all the irrational numbers.
And finally, we said that there are even things beyond rational numbers like complex
numbers, but we will not discuss them.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture - 04
Set Theory

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, we have seen numbers; we have seen natural numbers, we have seen integers, rationals,
reals and we have loosely talked of them as sets of numbers. So, let us try to understand little
more clearly what we mean by a set.

(Refer Slide Time: 00:26)

So, at its basic level a set is a collection of items. So, for instance, we could have a set called
the days of the week which has 7 members; Sunday, Monday, Tuesday, Wednesday,
Thursday, Friday and Saturday or we could take a number like 24 and list out the factors of
24 and call this a set. So, we have 1, 2, 3, 4, 6, 8, 12 and 24.

So, if you count, there are 8 factors that 24 has or we could take all the prime numbers up to a
certain limit. Supposing, we want to know the prime numbers below 15, then we know that
we do not have 1; but 2, 3, 5, 7 are the single digit prime numbers and then, 7, 11 and 13.

So, 2, 3, 5, 7, 11, 13 are all the primes below 15. So, this is how we talk about sets informally
as just collections of items. Of course, as we have seen sets can be infinite and in particular,
the infinite sets that we deal with very commonly are those which consists of the different
types of numbers.

Remember that N this funny N stands for the natural numbers that is 0, 1, 2, 3, 4. Z stands
for the integers. So, that is the natural numbers along with the negative integers like -1, -2, -3
and so on. Q is a peculiar symbol for the rational numbers, these are the fractions those

p
numbers which we can write as ; where, p and q are both integers.
q

And finally, R is a set of real numbers. So, the real numbers includes all the rationals all the
fractions, but also numbers that cannot be represented as fractions, such as the square root of
2 and other irrational numbers like π and e.

So, in all these things that we have seen above, it looks like there is some kind of condition
which requires a set to have some uniformity; either a set consists of numbers or a set
consists of days of the week or something like that. But actually mathematically there is no
constraint on this. A set can have any kind of members, even a mixed membership; there is
no uniformity of type.

So, for instance, we could enumerate the set of objects that appear in a painting. Now, here is
a particularly famous painting, where it is not so easy to enumerate the objects because its
drawn in a very abstract way. This is a painting called Three Musicians by Pablo Picasso. But
we could see roughly that there are three people and that there are some musical instruments
and so on and if you look very carefully, you will even find a dog.

So, notice that there is no commonality. There are people, there are musical instruments,
there are chairs, there are tables, there are animals and so on. So, a set in particular can have
any kind of members, it does not matter if they are mixed in type.
(Refer Slide Time: 03:00)

So, one of the important differences between say set and a sequence or a list is that the order
in which we identify a set does not matter. So, normally when we talk of numbers, we tend to
list them in a particular way; but as the set it does not matter. So, for instance, if you take the
set of cricketers; Kohli, Dhoni and Pujara. If you reorder this set as Pujara, Kohli and Dhoni,
it is the same set right.

So, the sequence in which you list the members of a set does not matter and for that matter, if
you happened to accidentally write the same member twice, it does not change the set. So, in
this particular set if we add Kohli a second time, as the set it does not matter. Though of
course, if you are a cricket fan maybe you would like Kohli to bat twice for you.

So, when we look at a set, we might ask a basic question as to how many members it has. So,
the cardinality of a set is the number of items in the set and if it is a finite set, we can just
count the items. So, for instance, if you look at the factors that we listed of 24, then we can
count them and say that this has cardinality 8.

Sometimes, it may not be obvious that a set is finite. You might remember from geometry
that a regular polygon is one, where the all the sides are equal and all the angles are equal. So,
the smallest regular polygon is an equilateral triangle in which we have 3 sides all equal and
3 internal angles of 60 degrees each. Then, we move to four sides we get a square, then we
get regular pentagons, hexagons, heptagons, octagons and so on.
So, for any number of sides, you can draw a regular polygon with that many sides with equal
angles on the inside. So, there is no limit. The set of regular polygons is infinite. But if we
move to three dimensions, the corresponding notion to a regular polygon is what is called a
platonic solid. In a platonic solid, first of all you have surfaces or sides each side is a regular
polygon and all these regular polygons meet at the same angle in three dimensions.

Now, it turns out that though you might imagine that there are infinitely many regular
polygons in two dimensions, there are only 5 platonic solids in three dimensions. So, this is
an example of a set which turns out to be finite, even though there is no reason for it to be
finite. So, these 5 platonic solids are the tetrahedron which has triangles.

The cube which we have which has squares and then, we have an octahedron which has 8
sides which are triangles. Then, we have a dodecahedron with 12 sides and an icosahedron
with 20 sides and there are no other regular solids, surprisingly it turns out.

Now, cardinality is quite easy to determine for a finite set, but what about for an infinite set?
Remember that, we said that we wanted to go from integers to rational numbers because we
want to talk about what happens when we divide 2 integers and the answer is not an integer
and it is clear to us from our discussion that integers were discrete, we can talk about a next
number and a previous number. So, there are gaps in the integers and rational were dense;
between any 2 rational numbers, there is another rational number.

So, intuitively, it seems like we are adding things to the integers to get rational numbers. But
can we make it formal in terms of cardinality? Are there more rational numbers than there are
integers? And what happens, when we go from Q to R when we go from rational numbers to
real numbers? So, remember that the real numbers, we had introduced because they were
numbers such as the √2 which could not be represented as a fraction.

So, clearly there are some rational numbers which are real and some real numbers which are
not rational and therefore, we have a bigger set; but again, R is really bigger than Q . So, this
is a separate discussion, there will be a small separate lecture about this. But there is a way to
measure cardinality of infinite sets, but it is not as straight forward as it is for finite set as you
would imagine.
(Refer Slide Time: 06:58)

So, how do we describe a set? Well, we have already seen that for a finite set, we can just list
out the members of the set explicitly. So, we can write out 3 numbers; Kohli, Dhoni, Pujara
or 8 members the factors of 24. So, the normal notation for a list of items which form a set is
to use these curly braces and to separate the items by commas.

Now, in many books and even in our lectures we will see notation like 0, 1, 2 …indicating
that there is an infinite set of elements to be added which follows some kind of a pattern. So,
this looks a way of listing out an infinite set, but you must understand that this is only an
informal notation, this is not a formal notion.

So, you cannot write … and claim that you are listing out a infinite set. So, in fact, you need
some other way of doing it and we will come to that as we go along in this lecture.

Now, it said seems reasonable that if a set is a collection of items, then we can collect
anything and make it a set. It turns out that this is not quite true and this is particularly, a
problem when we move to infinite sets. So, we have seen some infinite sets of numbers like
naturals and reals and so on; but in general, if you take an infinite collection of objects, it may
or may not form a set. In particular, Bertrand Russell showed that there is a problem, if we
collect all the sets together and call it a set.

So, if we have a set of all sets, then we have a problem and this is something which is called
Russell’s Paradox which we will discuss in the separate lecture, but you must be careful to
note that though the notion of a set is intuitive and it seems natural that any collection of
objects is a set, we have to actually be a little careful in mathematics, if we are using sets in
order to define what is a set and what is not a set.

But given that whatever we will see in our course, we will be fairly straight forward. So,
whenever we see a collection of numbers or a collection of objects of mathematical
description, we can safely assume that they are sets.

So, again some terminology. So, we have talked of different things items in a set, members of
a set and so on. So, the most formal notation for the members of a set is an element. So, a set
consists of elements and we write this membership of an element in a set using this ∈ notion.
So, we have this ∈ notation which stands for element of. So, when we write x ∈X, we mean
that small x is a member of the set capital X.

So, example 0 is a member of the natural numbers right. So, 0 ∈ Nis what we use. So, we can
see for instance that 5 is an integer, but √2 as we claimed is not a rational number. So, an
element of symbol with the line across it, means not an element of. So, 5 is an element of
integer set and √2 is not a member of the set of rationals.

(Refer Slide Time: 10:02)

So, moving on from elements, we can compare sets by asking whether one set is included in
another set and this is called a subset. So, X ⊆ Y, if every element of X is also an element of
Y and this is written using this subset notation ⊆. So, you have this familiar notation X ⊆Y.
(Refer Slide Time: 10:27)

So, for example, if we take just 2 out of the 3 players were listed before saying Kohli and
Pujara; then, this set forms the subset of our original set Kohli, Dhoni and Pujara. Similarly,
if we take all the natural numbers and collect only the prime numbers. So, remember that the
prime number is a number whose only factors are 1 and the number itself. So, it has exactly 2
factors; 1 and p and then, p is a prime number.

So, since some many numbers are not prime, primes is a subset of natural numbers. Since, the
integers extend the natural numbers with the negative numbers, we can say that the natural
numbers are included in the integers. So, N ⊆ Z. Similarly, we extended Z to Q. So, the set
of integers is a subset of the rationals and the set of rationals is a subset of reals.

So, if you wanted to draw it, we could draw it in this particular way. So, we can raw a large
circle representing the reals, a small circle inside right in the center representing the natural
numbers and if one circle is included in another circle, it means that this circle is a subset of
the circle outside it. So, here you can see that the natural numbers are a subset of the integers
and then, from the integers, we can say that there are subset to the rationals and the rationals
are a subset of the real numbers.

So, this kind of a diagram, where we represent a set by a boundary. So, this is a very abstract
diagram. We are not in this case for example, listing out the elements of the set we are just
indicating the extent of the set saying that the set are extends beyond Q and everything that is
in Q is sitting inside R.
So, these are what are called Venn diagrams. So, a Venn diagram is a very useful way to
picturize a set and relationships between sets; is one set a subset of another, is one set not a
subset of another and so on.

(Refer Slide Time: 12:20)

So, we often use Venn diagrams pictorially in order to represent sets. So, notice that every set
is a subset of itself because remember the definition of a subset set that X ⊆Y, if every
member of X is also a member of Y. So, since every element of X is also an element of X,
trivially as a extreme case of this definition, every set is a subset of itself.

So, this in fact, gives us an important notion which looks obvious; but it is not so obvious,
when are two sets equal. So, two sets are equal if and only if, they are actually the same set of
elements. So, one way to check that two sets are equal is to check that everything in the first
set belongs to the second set. So, X ⊆Y and everything in the second set belongs to the first
set. So, Y ⊆X.

So, often this happens when we have two different ways of looking at the same set of objects.
We have two different descriptions of the same set of objects and we want to check whether
they are equal or not. Then, using the first description, we argue that everything which
satisfies the first description also satisfies the second description and vice versa.

So, though this looks fairly obvious for finite sets, when it comes to infinite sets we have
sometimes have to argue in an indirect way. So, this although it is an obvious statement is
very important that X = Y provided X ⊆Y and Y ⊆X. So, sometimes we want to distinguish
between the case, when X is really a proper subset of Y; that means, it does not include all of
Y and that it is possibly equal to Y.

So, the subset equal to notation that we have right allows both. When we write X ⊆X, what
we are saying is that it is a subset, but it is actually equal. So, we are allowing both cases. So,
if you want to talk about proper subsets, sometimes we use a different notation.

So, we might either drop the equal to sign , just write the subset sign ⊂or we might explicitly
like we said not element of right. So, we are saying that this is not equal to. So, we are
dropping the equal to from below the subset.

Now, this is a bit dangerous. Second symbol this not equal to this is always correct. This is
sometimes used both ways. So, you have to be bit careful when we look at books when you
see the single subset without the equal to whether they mean subset and equal to or proper
subset.

(Refer Slide Time: 14:45)

So, we know for instance that the natural numbers is a proper subset of the integers because
the negative numbers are not there. Similarly, the integers are clearly a proper subset of the
rationals and because the irrational numbers are not rational, the rational are a proper subset
of the real numbers.
So, in most interesting cases, we will be looking at proper subsets. Sometimes, we will
emphasize it by adding this cross against the equal to and sometimes, we will not and very
often from context we will know whether we are talking about proper subsets or we are
talking about subset which allow the full set.

(Refer Slide Time: 15:16)

Now, there is a very important set just like the 0 is very important in numbers, there is a very
important set which is important set theory. It is the equivalent of 0. It is the set which has no
elements. So, the set which has no elements is called the empty set and is written ∅. It is
basically you can think of it as a 0 with a line across it. So, this Greek letter phi, symbolizes
the empty set; so, it has no elements.
(Refer Slide Time: 15:44)

Now, what may not be very obvious is that this empty set is actually a subset of any set.
Remember that we said that X ⊆ Y, if every element of X is also every is also an element of
Y. Now, you might argue that an empty set has no elements. So, why is this true? Well, when
we say for every and there is nothing in the set, then for every something is true right.

So, if I say that all birds with 3 legs have pink beaks, then this is actually true because we can
imagine that there are no birds with three legs and therefore, every bird which actually has 3
legs will have a pink beak. But since, there are no birds with 3 legs this is actually true.

So, these kinds of vacuous statements as they are called will hold for sentences which use the
word all where the set is empty. So, in particular, every element of the empty set because
there are none. So, every element that could be in the empty set is also an any set X that you
build. So, this empty set is a subset of every possible set. Now, though we have talked about
elements and sets.

So, they are two different categories of objects. So, we have numbers and the numbers belong
to a set of the type N or Q or R or Z; a set can clearly contain other sets. So, there is no
restriction saying that the members of a set or the elements of a set must be some kind of
discrete and indivisible objects.

So, one of the important sets of sets that we would like to look at is what is called the
Powerset. So, we talked a subset. So, supposing we want to enumerate all the subsets. So,
here is a two element set a comma b. So, what are all the subsets? Well, we already saw that
the empty set is always a subset. So, that is one subset.

The set itself for any X, X ⊆ X. So, X equal to {a, b}. So, we have these two subsets which
come just from the fact that empty is the subset of every set and the set itself is a subset. And
then, we have two proper subsets either we can include the a and exclude the b or include the
b and exclude the a. So, there are four subsets of X and if we group together these four
subsets into a larger set, then we get the Powerset.

Now, notice that this itself is the set right. So, we do not write. So, this is different from this.
The first is a set consisting of one element, namely the set consisting of the empty set. The
lower thing is the empty set alone which is the set with no elements. So, if we put a brace
around the empty set symbol, then we create a set with one element.

So, for instance, if you ask what is the power set of the empty set right. So, we know that the
empty set has a power set which contains the empty set. So, we have at least one empty set as
one element of the power set and there is nothing else right.

So, the full set is also the empty set, but if you duplicate an element, it is a same thing. So, in
fact, the power set of the empty set is a set consisting of just one element, namely the empty
set itself. So, just remember this, that the empty set on its own denotes a set with no elements,
but an empty set with the brace around it is not the same thing. It is a set consisting of one
element, namely the empty set.
(Refer Slide Time: 19:03)

So, we saw above that if we have two elements, then the power set had four elements. So, in
fact, one can generalize this and say that if we have n elements, then we would have 2n
subsets. So, for instance, if we had a, b, c right, then we would have 1 subset which is empty.
We would have 3 subsets which are one element each and then, we would have 3 more
subsets which are 2 elements each a, b a, c and b, c and finally, we would have the set itself
right.

So, these are the only subsets. If you add these up, this is 8 which is 23. You can check that if
you do it for a, b, c, d; then, you would have 24 , 16. So, why is it that a set with n element
should have 2nsubsets, no more no less?
(Refer Slide Time: 19:55)

So, here is one argument. Supposing we have n elements in the set. So, let us just call these
without describing what they are specifically as x 1, x2 up to x n. So, we have n distinct elements
x 1to x n. Remember these must be different because you cannot duplicate elements in the set.
So, now, we want to construct a subset.

So, how do you construct a subset? Well for each element xi , we have to either include the set
include xi in the subset or exclude xi subset. So, we have to make a choice for each xi right.

So, overall, we have to make n choices right. For each xi we have to decide whether to include
it or exclude it from the subset. So, we have two different choices for each elements. So, we
have two ways to decide whether to do something with x 1, keep it or leave it; x2 keep it or
leave it. So, then we have two times two choices for x 1and x2 together; two times two choices
for x 1, x2 , x 3 together.

So, in general, if we have n such choices where each choice involves two options, then we
have 2 into 2 into 2, n times 2nchoices. So, each of these choices gives us different subset. So,
whenever we make a different choice, we will either leave out i from the set or put an xi . So, it
will differ from the choice, where we do the other thing. So, each choice generates a separate
subset. So, there are exactly 2n subsets.
Here is another way of looking at subsets and getting to the same result. So, we can actually
think of subsets in terms of binary numbers. So, let us again think of our n element set x 1to x n
right. So, now, supposing we look at n digit binary number. So, digit actually is comes from
decimal. So, we say bit for binary digit. So, n bit binary number. So, remember in a binary
number system, we have 0’s and 1’s and the place values represent powers of two.

So, we have the unit digits is units as usual. The next digit 2 to the power 0 is a is number of
twos, number of fours, number of eights. So, it is like the decimal thing is in base 10. This is
in base 2. So, now, if we look at n bit binary numbers, then for instance, if we look at 3 bit
binary numbers, then we have 8 of them.

We can start with 0 0 0, then 0 0 1, 0 1 0 and so on up to 1 1 1 and again, the reason that there
are 2 to the n, n bit numbers is because for each bit we can choose to put 0 or 1. So, we have
two choices for the first bit, two choice for second bit and so on.

So, it is not surprising that an n bit binary number can represent 2 to the n different values
from 0 to 2 to the n minus 1, if we think of them as numbers. Now, we are interested in n bit
binary numbers as representing subsets. So, what we will look at is the ith bit and say that the
ith bit represents the choice that we made.

If we chose to keep xi in our subset, we will call it 0. If we chose to we will call it 1 for
example. And if we choose to omit xi from our set, we will call it 0. So, 0 represents the
choice, where we leave out xi ; 1 represent the choice, where we keep xi

So, supposing we have this four elements set a, b, c, d; then, if we look at the binary sequence
or the bit sequence 0 1 0 1, the first 0 corresponds to a, so it says leave out a. The second 0
corresponds to c, so it says leave out c and for b and d we have put a 1. So, it says keep b and
keep b. So, it says leave out a, keep b, leave out c, keep d. So, this 0 1 0 1 as a binary
sequence corresponds to the set b comma d.

What does 0 0 0, the all 0 sequence say? The all 0 sequence says every xi in the set is omitted
from the subset. So, this is precisely the subset which is the empty set because it has no
elements and what about the all 1 sequence? Well, the all 1 sequence says every xi that we
have is included in the final subset. So, this is the set itself. So, remember that these are the
two extreme subsets; the empty set and the set itself and all the other ones come in between.
So, from this, we can see that every n bit number represents one sequence of choices. So, this
gives us 2nchoices because there are precisely2n , n bit numbers. So, hopefully with this, you
are now clear about the fact that any finite set with n elements has exactly 2n subsets.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Lecture - 05
Construction of Subsets and set operations

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

Now, let us talk about subsets in the infinite context. So, how do we talk about subsets of the
numbers in a precise way? So, this is something called set comprehension. So, this is just
some jargon. So, a set comprehension is just a term used for this which we have sometimes
seen and which we will now review. So, if we want to talk about the set of even integers, the
set of even integers are those integers which when divided by 2 have a remainder 0. So,
remember that the remainder is called mod. So, x mod 2 is the remainder when divided by 2.

So, if x mod 2 is 0 it means that when we divide x by 2 there is no remainder. So, any such x
is an even number. So, this notation that we have written is actually the set comprehension
notation. So, let us try and separate out the different parts and understand what is going on.

So, when we use set comprehension first of all we can only do set comprehension when we
have a starting set. So, we have to begin with a set and construct a subset of that set. So, the
first thing says that we want to take all x in Z. So, this here says that we are looking at
elements from an existing set in this case this set is a set of integers. Then it says I want to
take all elements and apply some condition to decide whether to keep that number or not. So,
that is the second part of the right hand side.

So, we have the first part which tells us which set we are looking at the second part which
tells us what condition we want. So, we are really saying x in Z such that x mod 2 is 0 and
finally, with this bar and this left hand side we are saying collect together all the x which
satisfy this. So, this overall this notation says collect all the x for which x is in Z such that x
mod 2 is 0 or in other words x is even. So, this is set comprehension notation and this is
formally how you define a subset of an infinite set. Remember that we cannot list out the
elements in an infinite set.

Now we assume that we already have a set like Z or N or Q or R for which we know what
elements are. So, we do not have to describe how to pick out element we know what those
elements are. What we are now giving is a description of how to choose elements which
satisfy a given property. So, let us look at some more examples. So, for instance let us look at
perfect squares.

So, remember that we said an integer is a perfect square if its square root is also an integer.
So, for instance 25 is a perfect square because the square root is 5, but 26 is not a perfect
square because there is no integer which multiplied by itself is 26. So, here is a set
comprehension notation of the perfect square.
So, first of all remember square number has to be positive. We already discussed that
negative numbers cannot be squares because when we multiply 2 numbers by them to the
same numbered by itself the, 2 numbers will have the same sign. So, either it will be minus
into minus is plus or it will be plus into plus is plus because the multiplication rule says that if
the 2 numbers you are multiplying have the same sign the outcome is always positive. So,
first of all we can only have positive numbers. So, instead of looking at integers, it suffices to
look at the natural numbers.

So, we say for all m which are natural numbers such that the square root of m is also a natural
number. So, this is that the square root of m also belongs to the set N collect all such m right.
So, we are collecting all the m. So, this will give us if we write it out explicitly 1 will fall into
this set, the next number that will fall into the set is 4, then 9 and then 16 and then 25 and so
on right. So, the notation in blue is a succinct way of writing this informal infinite list which
starts with 1 and goes on. So, we are pulling out the numbers from N one by one; checking if
they are perfect squares and if so we are enumerating them.

(Refer Slide Time: 04:07)

We also talked about rationals in reduced form. We said that there are many different ways of
writing the same rational number because if we multiply the numerator and the denominator
by the same quantity, the number we are representing does not change. And we use this fact
in order to make denominators same when we did comparisons or arithmetic like addition and
subtraction. So, what are the actual rationals in reduced form. So, this is a subset of the
 3 6
rationals. For example, is in reduced form, is not in reduced form because I can; cancel
5 10

3
the 2 and get .
5

So, if we want numbers and rationals in reduced form first of all we pick up any 2 numbers
which are integers. Remember that a rational is actually a pair a numerator and a denominator

p p
which are integers. So, every rational looks like this right, but we do not want any such .
q q

p
We want such that they do not have any common divisors other than 1. So, recall the gcd is
q
the greatest common divisor; it is the largest number that divides both p and q and what we
want is that p and q have no numbers which can be divided into them other than 1. And if the

p
gcd of p and q is 1 then is a rational and it is in reduced form because the gcd is 1 right.
q

So, this is another example of set comprehension in order to define an interesting subset of
the rationals.

(Refer Slide Time: 05:34)

One of the things that we will often use with respect to numbers is to define intervals of
numbers between something and something else. So, for instance if you are looking at the
integers; we might want the integers from some lower limit to some upper limit. This for
example, is an expression which describes the integers between -6 and +6 right. So, it says I
want all z which belong to the set of integers such that z is above -6 greater than equal to -6
and less than or equal to 6. Now, we could split this for instance into two conditions. We
could also say z is bigger than -6 and z is smaller than 6 and so on

So, the way in which we write this condition which applies to the thing may vary and all of
them could be equivalent to each other. So, we will not be very pedantic about what syntax
we used to write there. So, for instance in the previous case here, we could have just read
written x is even instead of x mod 2 is 0 ok. So, we will not worry too much, but it is just that
we have this format where we take the underlying set, we pick out all elements, make it
satisfy condition. If it satisfy the condition, it belongs to the subset.

(Refer Slide Time: 06:44)

So, intervals are more interesting when we talk about real numbers and one of the intervals
that we really often want to talk about is the interval between 0 and 1. So, 0 to 1 is quite
interesting because we will often talk about probabilities for instance and probabilities range
between 0 and 1. So, what can we do between 0 and 1? Well first of all we can take all the
real numbers between 0 and 1 including both 0 and 1 and this is called the closed interval.

Closed interval means in this case, it includes the endpoints. So, if I draw this as a number
line for instance. So, normally I have 0 1, 2 ,-1 and so on. So, this is my number line. So, then
this closed interval says I want all the numbers from 0 to 1 including 0 and 1. So, this is my
closed interval right. So, what we write is take all r in the set of reals such that 0≤r≤
1.

(Refer Slide Time: 07:46)

So, r must be between 0 and 1 it could be 0 and it could be 1. If we want to exclude the
endpoints, then we get what is called an open interval and the way we draw an open interval;
if we want to draw it in a pictorial way is to emphasize that the endpoints are missing by
drawing a circle there.

So, we draw a circle to indicate that those are not included. So, if we so I have to fill in the
circle corresponding to the endpoints that endpoint is included in our interval. If we do not
fill it in it is not included, but formally it is just a set defined using set comprehension and
whether it is open or closed depends on whether the inequality has an equal to or not whether
it is strictly less than or it is less than equal to whether it is strictly greater than or greater than
equal to.
(Refer Slide Time: 08:29)

Now, there is nothing to stop us from including one endpoint and not including the other. So,
we had an closed interval which had both endpoints, we had an open interval which had both
endpoints missing. And we could say for instance that an interval is left open. So, it is all
numbers between 0 and 1; it does not allow us to use 0, but 1 is included. So, in notation we
will use this. So, the notice that we use this round bracket for open and we use the square
bracket for closed. So, here obviously we will use a round bracket for the open end and a
square bracket for the closed end. So, the left is open. So, we call this a left open interval.

So, left open interval has all numbers which are strictly bigger than 0, but less than equal to 1.
So, correspondingly you could have a right open interval. And what would this be? This
would be all the r such that r belongs to a set of reals. Now, 0 ≤ r we are allowed to include 0,
but we should not include 1 right. So, this is the right open interval. So, this will be an
important part of many discussions. So, you should be aware of these intervals as
representing sets of points in particular a subset of the reals which can be defined using set
comprehension.
(Refer Slide Time: 09:41)

So, finally, let us look at some simple operations on sets which we are all familiar with. So,
the first one is union. So, the union of two sets just combines them into a single set. So,
suppose we have a, b, c as one set and we combine it with c, d, e then we get a single set. And
notice that we have some elements which may appear in both sets and they appear only once
in the final set because remember that a set has no duplicates right. So, in the union if we take
sets which have some common elements across the two sets, they get represented exactly
once in the final set.

So, therefore, the cardinality of the union will in general be less than the cardinality of the
two sets put together. So, here we have two-three element sets, we take the union we get a
five element set not a six element set because there are some elements which are common
and the symbol for union is this ∪ right. So, X ∪ Y and if we go back to our Venn diagram;
so, remember that we used when diagrams in order to informally look at sets and we talked
about subsets. So, here we have a Venn diagram which represents the left hand side set is X,
the right hand set is Y and the picture suggests that X is not a subset of Y and Y is not a
subset of X, but there may be some overlap. So, this is the general case right.

Generally speaking if I give you two sets, there will be some elements which belong only to
X some elements should belong only to Y and some which belong to both. So, this kind of a
picture with two overlapping circles or ellipses is a particularly general picture of two sets
represented as Venn diagrams. Even though we are not specifying what the elements are this
is a picture. So, here for instance if we wanted to write out these elements in this particular
set if you wanted to write we have a here, b here, c here, d here and e here.

So, what this means is that if we look at the circles a, b, c belongs to the left circle c, d, e
belongs to the right circle, but we put c in the portion which is covered by both circles to
indicate that it is in the common portion. So, this grey shaded area in this particular case
represents the union of two sets.

(Refer Slide Time: 11:43)

So, the corresponding thing which takes up only the elements which occur in both sets as you
know is called intersection. So, intersection is written with the upside down version of the
union sign right. So, X intersection Y is written like X∩Y . So, here for instance we look at
elements which are on both sides. So, we have a, b, c, d intersection a, d, e, f. So, a is
common to both, b is not there on the right hand side, c is not there on the right hand side, d
is common to both and if you go to the right hand side e is not there on the left hand side f is
not there. So, only a and d are surviving intersection.

So, again if we draw this out as a Venn diagram on the right, the shaded portion which is the
area which is overlapped by both the circles is the intersection. So, in this particular case we
would write a here because it is in both b here. Notice the order is not important and in an
Venn diagram if we actually put the elements the position is not important. So, I can put them
anywhere and then I put e here and f there for instance. So, this is a pictorial representation of
the two sets on the left. The shaded area corresponds to the intersection and the non-shaded
portions are those which are in one set, but not in the other.

(Refer Slide Time: 12:53)

Another operation on sets is called set difference. So, in set difference we take two sets and
we want to know what is there in the first set that is not there in the second set. So, for
instance we want to know which are the real numbers which are not rational right. So, then
we would write in this notation which are the real numbers which are not rational right or
which are the rational numbers which are not integers. So, this is a common thing that we
might want to do.

So, we write either this direct subtraction which is the normal minus sign or we write this
back slash kind of notation \ to indicate the set difference. So, it is all elements in the first set
which are not in the second set. So, here for instance if you look at the first set a is there, but
a is also there in the second set. So, a is not counted, b is there, but b is not there in the
second set. So, b is in the set difference c is there c is not there in the second set. So, c is in
the set difference, but d for instance appears here. So, d is not counted.

So, here we have that the first set minus the second set has b and c because those are the 2
elements in the first set which are not in the second set. Now, this is like subtraction not
symmetric in the sense that you know that 3 - 5 is not the same as 5 - 3 unlike 3 + 5 right. So,
3 + 5 is the same as 5 + 3, but 3 - 5 is not the same as 5 - 3. So, if I take union for instance,
then Y ∪X = X ∪Y right and Y ∩X = X ∩Y because this it does not matter which side you
take from.

Because finally, you are going to look at all elements which I has a common to both side or
included in both sides. Now here if I take the reverse if I take a, d, e, f right and I subtract out
the elements from a, b, c, d; then I would see that again a would disappear. So, the same
elements disappeared because the common part is the same. So, a would disappear and d
would disappear because these are the parts which are on both sides, but what survives now is
e, f right.

So, when I do it in the other way around, I get the elements on the right hand side which are
not on the left hand side. So, in the set difference the order of the sets in the expression
matters. X \ Y is not the same as Y \ X just like in subtraction and here we have a picture
right. So, this shows us this picture. It says that you take everything in X and you remove
everything that all includes. So, in particular you remove all these elements which are in the
intersection and that gives us X \ Y.

(Refer Slide Time: 15:19)

And finally, we often talk about the complement. We say those numbers that are not prime.
So, those numbers that are not prime in particular are called composite numbers. So,
composite number is defined to be a number which has factors other than 1 and same. So, any
number which is not prime has more than 2 factors. So, such a number is called a composite
number. So, clearly a number is either a prime or it is not a prime.
So, either it is prime or it is a composite. So, the composite numbers are disjoint from the
primes and they are all the numbers that are not prime. So, this is what we mean by
complement. Complement means the opposite side it means everything else, but complement
is not very straightforward in set theory because complement with respect to what.

So, if I say numbers that are not prime, but I do not tell you in what set I am talking about this
thing. If I look at complement in for example, in the reals; it will include all numbers like π
and e and √ 2and so on and that is not what you mean right. When I say the complement of
the primes; you are not thinking of rational numbers, irrational numbers and so on. You are
thinking of integers or in particular you are talking about natural numbers which are not
primes right. So, we would always want to define what is called a universe ok.

So, we need a universe with respect to which we are going to complement. So, if we say that
the complement of prime numbers in the universe of natural numbers, then we get the
composite numbers. So, when we say primes for instance we see this Venn diagram on the
right, we see primes as a subset of the natural numbers. So, then the grey shaded area is all
the composite numbers right. But if this was not this, but R then we would have various thing
we would have √ 2, e and so on sitting here which is not what we intend.

So, whenever you use the word complement, you must make sure that you have specified
complement with respect to what. What is the overall set with respect to which you are
negating the set that you have and that is very important.

(Refer Slide Time: 17:18)

So, let us wrap up this lecture. So, we are all familiar with sets as an informal term which we
have come across from school level and a set is a standard way to represent a collection of
mathematical objects. So, it is very important to be familiar with the terminology of sets
element of subset of and so on and also the notation the curly brace listing out the elements
set comprehension and so on. So, sets may be finite or infinite. An infinite sets are actually
quite tricky and interesting and most of the interesting sets that we are going to look at will be
infinite because very often we will be thinking of sets in terms of numbers, but we will also
be thinking in terms of finite things.

For instance we talked about we could talk about for instance a time table then we might
want to know the set of stations at which the train stops or we might want to look at a
shopping list and we might want to look at the set of items that the store has in its inventory.
So, sets are a very useful way to talk about collections of objects infinite collections are
important because numbers are infinite, but other finite collections are also important from a
computational and data science point of view.

So, we saw that we have some useful notation like set comprehension which allows us to
define subsets of infinite sets and we have these standard operations on sets like union,
intersection, set difference and complement which allow us to take sets and combine them in
many different ways. So, it is important that you get used to all these notions as I said
because these notions are used implicitly throughout mathematics and these are not difficult
notions is just a question of understanding the notation and understanding exactly what
happens when you apply each of these operations.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute
Lecture- 5A
Sets: Examples
So, we have seen some definitions of Sets and some operations on them. So, let us look at more
examples to get familiar with the notation and the terminology of sets.

(Refer Slide Time: 00:25)

So, remember that a set is a collection of items and when we write out a set, if it is a finite set,
then we can just enumerate the items in the set by writing them within curly braces. On the other
hand, if we have an infinite set, we really can not write out all the elements even though
informally, we put dot, dot, dot to indicate a sequence, if that sequence is not very regular. For
example, supposing it is a set of prime numbers, which does not have a clear pattern, then it is
not very easy to represent it explicitly like this.

So, we saw that there will be another notation called set comprehension that we will come to.
But, before that let us talk about the two basic relationships between sets and membership of a
set. So, membership is denoted by this element of relation. So, small x typically denotes a
member or an element of a set, and capital X usually denotes a set itself. So, when we write x
belongs to X like this, what we mean is the element x belongs to X.

So, for example, the number 5 belongs to set of integers, and √2 does not belong to the set of
rationals for instance. Subset on the other hand, says that one set is included in another set, so
everything that belongs to X belongs to Y. So, for instance, all the prime numbers are natural
numbers, so the primes are a subset of the naturals. Every natural number is an integer, so the
natural numbers are a subset of the integers. Similarly, the integers are a subset of the rationals
and the rationals are subset of the reals.

And we draw this using these Venn diagrams where we draw these ovals or circles or boxes
representing the extent of a set, it is a picture of a set. And then depending on whether a box
intersects another box or it sits inside a box, it indicates whether the first set is a subset of the
other one or they overlap and so on. So, in this particular diagram which also has colors, we have
indicated the subset relationship between the different types of numbers that we have studied, the
naturals, the integers, the rationals, and the reals.

And finally, one very useful thing to know about sets is the power set. So, when we take a set, we
can enumerate all its subsets. So, remember that we have just defined a subset. And in particular,
we have this special subset called the empty set, which is a subset of every set. The empty set has
no elements in it, but we needed it for technical reasons, and it is a subset of every set. And in
addition, if you have 2 elements set {a,b}, then the subsets could be the individual elements, the
set containing a and the set containing b or the entire set itself.

So once again, just like the empty set is a subset of everything the set itself is also a subset of
itself. And we argued that for a finite set with n elements, we will always have 2n subsets. So
here for instance we have 2 elements, so we have 22 = 4 subsets. So, this is just a review of what
we have already seen.

(Refer Slide Time: 03:06)

Now, let us look at this set comprehension notation, which is what we said we would use when
we have to describe infinite sets which cannot be written down explicitly. So, this was a typical
example. So, supposing we want to write down the set of all the squares of the even integers. So,
the even integers are -2, +2, 0 as an even, -4, +4, and so on. But if we square them, then we know
that (− 2)2 is the same as 22 is 4.

So, the set on the right which is written in this informal dot, dot, dot notation has 02 , 22 , 42 ,
62 and so on. So, how would we write this out? Well, this is that notation on the left, which says
that we take every x which belongs to the integers, check whether it is even, whether (x mod 2 =
0), and then square it. So, let us just break this up into parts so that we remember exactly what is
happening.

So first, in the set comprehension notation, we have a generator. A generator says that we are
taking elements from an existing set, so we can only build new sets from old sets. So, we already
have a set of integers, and we are going to try out every integer in the set, so, that is what x
element of Z says, is try every x ∈ Z, so Z generates this set. Now, all the x's that come out are
not interesting to us. So, we want to filter out those that are useful, that satisfy a given property.

In this case, the property that we are looking for is that the number is even. So, we want those x
which come out of Z through the generator, such that they satisfy the property that x when
divided by 2 has remainder 0, which is the property that x is even. And finally, with these x, we
do not want to keep them as they are, we want to transform them. So, on the left-hand side of this
vertical bar, this is the left-hand side are the actual elements of the set.

The elements of the set are generated right, then filtered through some conditions, which rule out
the ones we do not want, and when the ones we keep, we can transform them. In this case, we
want the squares, we do not want the even numbers, we want their squares. So, if you look on the
right, this is what happened.

So, when we started the generating process, we had all the integers, then we filtered out, and we
got only the even ones, and now we transform them. So, for each even number, we produced its
square. And now in this process, you will notice that (− 2)2 = 4 and 22 = 4 also. So, some
elements will disappear because we do not keep duplicates.
(Refer Slide Time: 05:35)
So finally, when we go through this, we end up with this sequence 4, 0, 4. And then in this, we
will throw away all the elements on the left, and we get the number sequence on the top. So, this
is how set comprehension works.

So, we can write filters in many different ways as long as it is unambiguous, we will not be very
particular about the language we use so long as there is no question about what we mean. So, for
instance, we looked at this example, we have rational numbers, but some rational numbers are
4 2
not in reduced form. For instance, if I write 10
, then I should actually think of this as 5
, because
2
it is 5
⨯ 22 = 10
4
.

2 4
So, I have actually multiplied both the numerator and the denominator by 2, to go from 5
to 10
,
but it is the same rational number. So, we want the numerator and the denominator to not have
any common divisors, which is the same as saying that their greatest common divisor is 1, that is
nothing other than 1 divides both the top and the bottom of the fraction. So, if we take all the
p
rational numbers, so we generate all the possible rational numbers q
, which belong to the set of

rationals.

Then, we filter out those which have no common divisor between the numerator and the
denominator and we keep only those, we do not transform it in any way, we just keep it here. So,
here the transformation is just to keep it as it is, this is sometimes called the identity
transformation. The identity just takes an input and produces the output the same as the input.

So, this gives the set of rationals in reduced form. So, here we have used a function, GCD. Even
though we have not formally defined it here, we assume that people understand what GCD
means. So, this is what we mean by saying that we can write the filter in any reasonable way, as
long as people understand what it means.

Another example, we looked at are intervals. So, here we want the real numbers, which start
from -1 including -1, and go up to but not including 2. So, in this case, we will use less than and
less than equal to, so we will take all the reals. So, we take every possible real number, but we
are not interested in all the reals, so we check whether it is greater than or equal to -1, so it
includes -1 and everything above it. So, it cuts off everything which is strictly smaller than -1.

But, we also do not want it to cross 2, so we stop below 2, so it should be greater than equal to -1
or and less than 2 and if so, again we keep it without any transformation. And this notation on
the top, the square bracket, and round bracket are indications of whether the endpoint is included
or not. So, -1 endpoint is included, +2 endpoint is not included.

(Refer Slide Time: 08:14)

So, let us see why we would actually want set comprehension notation. So, let us extend our first
example of squares of the even numbers to cubes. So, cube is just a number multiplied by itself 3
times. So, square is x⨯x, a cube is x⨯x⨯x, 3 times. So, if we want the cubes of the first 5 natural
numbers, we can write it out explicitly like this, we can take this generator and generate the first
5 natural numbers as 0, 1, 2, 3, 4. Remember that, in our terminology natural numbers start with
0, even though in some books, you will find that natural numbers start with 1, we always assume
natural numbers start with 0.

So, the first 5 natural numbers are 0, 1, 2, 3, 4. So, this is our generator, take every n in this and
transform it to n3 without doing any further filtering. We are not asking for the first 5 odd
numbers or the first 5 numbers which have some other property, we just take, taking the first 5
numbers. Now, imagine that we change this question to the first 500 natural numbers, then
though we can write it out explicitly, it is rather tedious.

So, we have to replace the small list of 5 numbers by a long list of 500 numbers. And remember,
we are not really allowed to write dot, dot, dot if we are being mathematically precise. So, we
actually have to physically write out these 500 numbers. Now, this is not terribly convenient. On
the other hand, we can define the first 500 numbers quite easily using set comprehension.
So, we can say, give me all the natural numbers, that is the generator, but restrict the natural
number to be less than 500. So, remember that the first 500 natural numbers are going to be 0 up
to 499. So now, this says that this set X is actually this long set here which we have written
explicitly. So, we have replaced that very long and tedious expression by a much more compact
expression, which captures exactly the same set. So now, we can have a much more readable
version of these cubes of the first 500 natural numbers.

As an intermediate set, we generate the set X, set X = {n | n ∈ N, n < 500 }. And then we take
this as the generator and we say, okay, take every n which belongs to this X. So now, we know
that x is restricted to 0 to 499. And then, take the cubes of these numbers, so we get n cubed in
this range. So, this is one other use of set comprehension, which is to make our definitions more
readable and understandable and less tedious to write.
(Refer Slide Time: 10:37)
So, let us look at one more round of examples. So, we saw this before, we talked about perfect
squares. So, we said that some integers are squares of other integers and some integers are not
squares. In particular, those which are not squares, their square roots are actually irrational. We
proved for instance, in our supplementary lecture, that the √2 is irrational. So, perfect square is
an integer such that its square root is also an integer. So, this is what this says, give me all the
integers, which satisfy the condition that their square root is also an integer.

So, the square root of small z also belongs to a set of integers, give me all set Z and call it a
perfect square. Now, notice that the square must be positive, we have already discussed this
because you multiply 2 negative numbers, you get a positive number, you multiply 2 positive
numbers, you again get a positive number. So, in fact, a perfect square must always be
non-negative, it could be 0.

So, we could as well assume that the target set is generated by the set of natural numbers. And
that, we are only interested in the positive square root, so remember that, 4 has 2 square roots,
the number 4 is either (-2)⨯(-2), or 2⨯2, but it is sufficient to know that one of its square roots is
an integer because the other one will just be the same with a minus sign.

So, we can as well define the same set of perfect squares in terms of the natural numbers, we
generate all the natural numbers whose square roots are also a natural numbers. Now, we can
turn this around and replace the filter by a condition. So, we know that every natural number
when it is squared will give us a natural number. So, all the perfect squares will be generated in
that form, take a natural number, square it. So, instead of looking for those numbers whose
square root is a natural number, we can just take every natural number and square it.

So, we just generate all the natural numbers and without filtering them, we just take the output
square. So, this also gives us 0, 12 , 22 , 32 and so on. So, these are all different ways of writing
the same thing. In one case, we replace the generating set from integers to natural numbers
because of the property of perfect squares. In another case, we transformed the filter into a
transformation. So, instead of putting a condition on the numbers that we are generating, we took
all the numbers and then squared them to get the actual perfect squares.

Now we could extend the notion of perfect squares to other sets of numbers. For instance,
rationals can also admit a definition of a perfect square, so a rational will be a perfect square if it
is a square of another rational. In particular, a rational could be an integer, but we will now
9
integers can also be above and below the line, so we could have 16
, for instance as a rational
32 3
number, which is 42
, so, 4
⨯ 34 = 16
9
.

So, we might want to say that this is a perfect square in the world of rationals. And not

everything is a perfect square because since √2 cannot be represented a rational, it is easy to

 p 2
check that half cannot be represented with a form ( q ) . So, not every rational in this sense is a

perfect square, some are, some are not. So, we can again change the definition above and replace
Z and N by Q and get a reasonable definition of perfect squares in a different domain of
numbers.

So, we can say give me all the rationals q such that √q is also a rational. Or using the second
form, we can say take all the rationals and square them. So, take every q, which is a rational q
and give me q 2 . So, this says that depending on how you choose the generator, you might
generate the same set, or you might generate a different set. So, it is important to specify all the
parts of a set comprehension correctly, so that there is no ambiguity and so that you get the set
that you mean to get.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute
Lecture- 5B
Examples of Set Operations and Counting Problems
(Refer Slide Time: 00:15)
So, the other operations that we saw on sets are union, intersection, and complement, which we
represented using Venn diagrams as shown here. So, the union takes two sets and combines them
and removes the duplicates. So, the overlapping part between the two diagrams represents the
common element. So, in this case, we would have this common element c over here, and then we
have had a and b over here, and we would have d and e over here because d and e belongs only
to Y, a, b belongs only to X.

Conversely, we can take only those things which are common to the two and in this case, we
have a and d over here, and then we know that b and c are only on the left and e and f are only on
the right. So, the intersection tells us the elements which are common to the two sets. Set
difference tells us what is on the left but not on the right.

And finally, the complement can be taken if we have an overall universe that is a full set to talk
about. And with respect to that set, we can ask which elements are not in the set that we are
looking at. So for instance, if we are looking at the natural numbers as a whole, the primes are a
subset of the natural numbers, the complement of the primes are all those natural numbers that
are not primes.

Now, remember that the complement matters, because if we take the complement of the primes,
for example, with respect to the real numbers, we will get all sorts of other numbers which are
not even integers. So, whenever we define the complement, we need to define the universe that
we are talking about.

(Refer Slide Time: 01:37)

So, this leads us to a class of problems that you might come across, which can be solved nicely
using these Venn diagrams. So, these Venn diagrams are not just pretty pictures, they are actually
useful ways to reason about these problems. So, here is a typical problem that you could come
across. So, you have a class in which 30 students have taken physics, and 25 students have taken
biology, but 10 have actually taken both physics and biology, but there are also 5 who have taken
neither of these two subjects.

So, these are the facts that are given to you. There are 30 students taking physics, 25 taken
biology, 10 take both, 5 take neither, the question is how many students are there in the class. So,
using Venn diagram notation, you can represent the fact that there are two sets of students, those
who take physics and those who take biology by representing them by two sets, say P and Q.
And we know that some take both, so, there is an intersection so these two sets overlap.

Now, from the data that we are given, we know that the overlap has 10 students, so we can write
a number 10 in the intersection to indicate that there are 10 students who take physics and take
biology. Now, we know that 30 students took physics overall and we have already accounted for
10 of them because they have all taken both physics and biology. So, there are 20 students who
have taken physics, but have not taken biology.
So, this in our set notation is the set difference, it is the difference between P and B, how many
elements are in P which are not in B, how many students have taken physics who have not taken
biology. And we have a symmetric thing on the right hand side. So, we know that there are 10
students who have taken both but 25 students take biology. So, there must be 15 students who are
in B∖P, these are students who took biology and did not take physics.

So, in this way, we can populate the three regions of the Venn diagram with numbers indicating
how many students are in each of these regions at 10 in the intersection, 20 on the left hand side,
15 on the right hand side. But, this is not the entire class because with respect to the entire class
we have to take the number who are in the complement, those who have taken neither physics
nor biology, and these are 5 students who are outside P ∪ B.

Now, technically one should draw outside this the complement to indicate the entire class but just
for convenience, I have not done that, but this entire complement outside this contains 5
elements. So, totally from this, we can see that there are 4 regions of interest. We have the P∖B
region physics but not biology, we have the B∖ P region, biology but not physics, we have the P
∩ B region taking both, and we have the complement, taking neither, and these are all disjoint
from each other.

So, now if we add up the students across these, we get the exact number of students. And in this
case it is 5 + 20 + 10 + 15 = 50. So, there are actually 55 students taking physics and biology
together, but the total class strength is only 50. And actually only 45 students are taking these
subjects because 5 are not taken either of them.

(Refer Slide Time: 04:43)

So, here is a variation where the data for the problem is given in a different way. So now, you are
told the class strength 55, you are told that 32 students took physics and of them 11 took physics
and biology and you are also told that 7 took neither. So, the question is how many took biology
but not physics. So again, we draw a Venn diagram and from the previous question, we know
that we can put 11 in the intersection, because that is the number who took both.

And since there are 32 who took physics, we can subtract out these 11 and say that P∖B is 21 and
in the complement, we have 7. So, the question now is how many are in B∖P, which I have
marked by x, but now we know the total. So, we know that the four numbers together, add up to
the total which is 55. So, 7 + 21 + 11+ x = 55. So, if we solve for x, we get that x = 16. So, we
can deduce that 16 students have taken biology but not physics in this situation.

So here is yet another version of this. So, we have 60 students in the class. So again, we know
the total number of students in the class, we are told that 35 students took biology, 35 students
took physics, and 30 took biology, and 10 took neither. So now, we are trying to calculate the
intersection, how many people took both subjects. So again, let us use this notation which we
introduced when we first introduced sets.

So, this perpendicular bar on the side of a set indicates the size of the set. So, this is the
cardinality of a set, cardinality is the number of elements, so the cardinality of Y is denoted by
putting Y inside these bars. So, what we are told is that the set P has cardinality 35. That is a set
of students who have taken physics overall, including those who have taken both, set B has 30
and 35 plus 30, there are 65 students who have taken in the union, of these I mean, have taken
these together.

But we also know that there are 60 students in the class of whom 10 have taken neither. So, the
actual union has only 50 elements. So, there are totally 65 people who are taking either physics
or biology or both, but this total number actually spans only 50 students, so some of them must
be taking both and are being counted twice. So, this must be the difference of the two.

So, 15 of these people must be counted twice, otherwise we would not have this mismatch. So, if
we draw the diagram for this, this is how it comes out. We have 15, that we calculated for the
intersection by taking the total number, realizing the 10 have taken neither, and then computing
the difference between the number who should have taken both the subjects from those who are
actually registered for either one or both of the subjects. So, these are three different examples
using Venn diagrams to indicate how you can solve these kinds of counting problems.
(Refer Slide Time: 07:37)

So, to summarize, we use set notation because it is a very useful and precise way to talk about
collections of objects. And if we use it nicely, it is also a concise way sometimes instead of
writing out a long sequence of values, we can actually describe it using a condition. So, this is
typically where we use set comprehension.

So, remember that set comprehension has three parts, some of which may not be used. So, you
always have a generator, a basic set from which you are creating new sets, you may have a filter
which takes out some elements from the generated set and throws them away and keeps only
those that satisfy the condition.

And finally, you may have a transformation which takes these filtered elements and does
something to make them into the elements that you want, for example, the squares of the even
numbers. And then we also saw that Venn diagrams are not just simple doodles that you draw to
indicate sets, Venn diagrams can actually be very useful for calculating properties about sets,
especially numerical problems about sets. So, it is important to be able to draw the proper Venn
diagram to indicate which groups of sets overlap, how they overlap, and which parts are empty,
and so on.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 06
Relations

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:14)

So, we have seen Sets, now let us move on to Relations.

(Refer Slide Time: 00:17)

As we saw a set is a collection of items and we can construct new sets from old sets. So, we
can take unions combine two sets into one. We can take intersections, take the common
elements. We can take the difference that is take the elements of X which are not in Y and if
we define the universe with respect to which we are working, we can define the complement
those elements that are not in X.

Now, in general we are interested in carving out subsets of a set and so, we use the set
comprehension notation. So, what this does is it takes a base set and takes elements of that
set, then it applies some condition those elements we are interested in and then it collects
them all together. So, we can take all the integers which are divisible by 2 or not divisible by
2 in this case, so we get the odd one; so, those where the remainder is 1.

Or we can take all fractions in which the numerator and the denominator have no common
divisor or we can take for instance the real numbers which lie in an interval with [3,17).
(Refer Slide Time: 01:17)

So, now, we will see a new way to combine sets to form new sets and this is called the
Cartesian product. And, in the Cartesian product basically what we do is we take two sets and
we take one element from each and form a pair. So, A × B as it is called is the set of all pairs
which we write with this normal bracket notation (a,b) such that the first element a comes
from the big the set A and the second element comes from the set B.

So, for instance, if A is the set {0, 1} and B is a set {2, 3} then all possible pairs we can form
in the Cartesian product a 0 combined with 2. So, (0, 2), (0 ,3) and then 1 combined with 2, (1
, 2) and (1,3). So, we have four possible pairs.

Now, in sets we said that the order of the element is not important, but of course, when we
are doing this kind of a pairing, then we know that the left set comes from the left part of the
product and the right element comes from the right part of the product. So, for example, (0 ,1)
is not equal to (1 , 0). So, here we have to respect the order when we talk about a pair.

Now, if we have sets of numbers right, then we normally visualize the product as a space
which we draw familiarly as a graph. So, for instance if we take N × N then we draw N × N
as this grid, where on the x-axis you have one copy of N , on the y-axis you have another
copy of N . And, for example, if you want to look at the pair (2 , 3), then such that the x-
coordinate is 2 and the y-coordinate is 3 and you get this point and similarly, if you look at
the point (5 ,6); you get this point right.
So, you can take the first coordinate plot it on the x-axis; take the second coordinate plot it on
the y-axis and where those two points meet in the grid is the point that we are interested in.
So, this is one way of visualizing a binary relation on numbers.

(Refer Slide Time: 03:00)

And, we can do the same thing if you are using say the reals, in which case the grid points
that we are going to plot will have real coordinates and not just natural number coordinates.

(Refer Slide Time: 03:10)

So, now we have this Cartesian product which consists of all possible pairs of the two sets
and as we did with set comprehension we might want to pick out some of these sets some of
these pairs and this is what we call a relation.

So, we combine this Cartesian product operation with set comprehension. So, for instance,
we can take all pairs of numbers which are natural numbers (m ,n), but we want to insist that
the second number is 1 plus the first number.

(Refer Slide Time: 03:40)

So, we get for instance (0, 1) because the second number one is 0 + 1; (2, 3) because 3 is 2 +
1, (17, 18) and so on. And, if we plot these points alone on the right then we get these so, we
get a subset of the overall points and this these points satisfied this set comprehension
condition.
(Refer Slide Time: 03:59)

Another example would be pairs again of natural numbers (d,n), where d is a factor of n.
Remember, d is a factor of n means that if I divide n by d, I get remainder 0. So, for instance
2 is a factor of 82, 14 is a factor of 56. So, these will be points in our relation. So, this is what
is called a binary relation. So, formally it is a subset of the product. So, we take the Cartesian
product all possible pairs and then we apply some kind of a condition which filters out the
pairs of interest to us and it gives us therefore, a subset of pairs and this is what we call a
relation.

Now, to denote the pairs that belonged to the relation either we can give the name of the
relation as a set and say that (a,b) ∈R or sometimes to say that a is related to b, we use R as a
kind of operator. We say a is related by R to b and so, we write a R b. So, these are two
notations which you might see in different books and they mean exactly the same thing.
(Refer Slide Time: 04:57)

So, let us let us look at some other examples of relations outside the numbers. So, supposing
you have a school in which there are some teachers and some courses to be taught. So, T is
the set of teachers; C is the set of courses that are being offered in this term, then you need to
describe which teachers are teaching which courses. So, we would have an allocation relation
A which is a subset of all possible pairs T × C.

So, every teacher and principle could be teaching every course, but of course, this is not
normally the case. We do not have all teachers teaching all courses, we have some teachers
teaching some courses. So, we would specifically say take every pair of possible teacher
course pairs, then we take out those were precisely the teacher T is actually teaching the
course C and we collect those together to form this allocation relation.

So, here is a different graphical way of describing a relation not in terms of the grid and the
graph that we have learned when we do graphs in school. So, this is also called a graph, but
this is a graph in which we have some nodes representing the elements on the set. So, on the
left hand side we have five teachers, on the right hand side we have four courses and the
arrows from the left hand side to the right hand side connect the pairs which are in the
relation. So, we see that Kumar teaches maths; Deb teaches history and so on.

So, this is a useful way of visualizing relations on finite sets and we will see this often as we
go along. Another example of a similar type of a relation is that between a parent and a child
specifically let us look at mothers and children. So, if we have a set of people in a country,
then we can take the set of all pairs of people and then isolate from that pairs in which the
first element of the pair is the mother of the second element. So, we want (m,c) which
belongs to P × P such that m is the mother of c.

(Refer Slide Time: 06:49)

So, let us go back to numbers. So, supposing we want to plot all points which are in R × R
which are at a distance 5 from (0, 0) which is normally called the origin. So, one thing you
need to know for this we probably you should have learned this at some point is that if I take
a point (a ,b) and calculate its difference from (0 , 0). So, this is calculated using the
Pythagoras theorem and it comes out to be √ a2 +b2.

So, in other words, the relation we are looking for in this case is all (a,b) whose distance from
(0 , 0) is 5. So, all (a, b) and R cross R such that √ a2 +b2is equal to 5. So, here are some of
the points (0,5) for instance you can see (0, 5) is there because the sum is 0 plus 25 and the
square root of that is 5. (3 ,4) is there because 3 squared is 9, 4 squared is 16, 9 plus 16 is 25,
square root to 25 is again 5.

So, interestingly these points if we plot every such point in R × R which satisfies this actually
defines a circle of radius 5 with center at (0 , 0). So, relations can define interesting geometric
shapes and very often we do deal with geometric shapes in this relational form because it is
easier to manipulate than looking at pictures. Now, depending on how we are going to view a
relation, we can look at it in different ways.
So, remember that we looked at rationals in reduced form. So, we said that a rational in
reduced form has p / q such that p and q are integers and the gcd is 1 right; that means, that
they do not have a greatest common divisor other than 1. But, we can also think of this as a
relation on integers itself. We want all pairs of integers. So, every rational is really a pair of
integers, the numerator and the denominator and we want every pair of integers where the
gcd is 1, that is, there is no common divisor.

(Refer Slide Time: 08:51)

So, we do not have to restrict our self to binary relations. The Cartesian product notation
extends to multiple sets. Let us look at three sets for instance. Remember, Pythagoras
theorem which says that the square on the hypotenuse is the sum of the squares on the
opposite sides. So, what values of a, b and c could be the sides of a right triangle are
determined by Pythagoras’s theorem.

So, we would say that a, b and c is a valid triple in the Pythagoras sense if (a, b, c) belongs to
N × N × N . So, here we now have three copies of N and a, b, c must all be nonzero. They
must all be positive length we do not want to have triangles in which one line one side is
collapsed to a point and we want the constraint that a2 +b2 =c2 .
(Refer Slide Time: 09:41)

Here is another example. Suppose, we look at squares on the plane squares with real corners
right. So, a corner is a point (x , y) which is in R × R . So, we define the x coordinate and the
y coordinate that defines the corner of a square and we want four corners which together form
a square if we connect them by lines. So, for instance, if you look on the right the four blue
dots correspond to a square which is cornered at (0, 0); (0, 2); (2, 0) and (2, 2).

The red square is also red points also define a square because this is a rotated square, but then
if you rotate it vertically; you will turn out that this diamond is actually a square. So, there are
many such four sets of points which form the corners of squares and we might be interested
in all such four sets of points. So, now, we have a relation which involves four sets of points,
but each point itself is a pair of real numbers; it is an x and a y.
(Refer Slide Time: 10:35)

So, square if we think of it as a relation is actually a relation on R 2 that is the first corner
times R 2 the second corner times R 2 the third corner and the fourth corner R 2 again. So, this is
actually either a relation on eight copies of R or if you want to group it four copies of pairs of
R.

So, this just says that we can take relations on arbitrary an arbitrary number of copies of a set
and we get larger and larger from pairs, we move to triples we move to quadruples and in
general if we have n copies we call this an n tuples.

(Refer Slide Time: 11:11)

So, there are some special binary relations which pop up all over the place. So, it is useful to
know their names. The first one is called the identity relation and as you would expect, the
identity relation maps every element to itself. So, if I take A × A, so, first of all the identity
relation is defined on two copies of the same set because identity means equality. So, I take A
× A. So, this has all kinds of pairs (a , b), where both a and b belong to A and now, I want the
condition that a = b.

So, in other words, I want things of the form (a , a). So, if I plot this for instance on the
natural numbers and N × N , then I get (0, 0); (1, 1) and so on, and these are the points which
are drawn on the right in this grid.

(Refer Slide Time: 12:05)

Now, point of notation we sometimes it is tedious to write this notation as it is says us (a,b)
time in n. So, we do not want to you know have to write out this long thing. So, sometimes
we simplify this by saying I want all pairs such that a comma a belongs to A × A. So, what
we are really saying is that the second day and the first day must be the same. So, we are
collapsing the equality and this. Now, this is not technically correct, but this is often used in
order to simplify the notation.

And, sometimes we might drop the product altogether. We might just say we want all pairs (a
,a)where a comes from the set A. So, in other words we are pulling out one copy of the
element from the set and then we are constructing a pair by taking two copies of it. So, all of
these are equivalent ways of writing this although only the first one technically follows the
notation that we are using to introduce relations.

Now, there are some properties that relations may have. The first one is called reflexivity. So,
reflexivity refers to the fact that an element is related to itself. So, a reflexive relation is one
in which for every element a; (a,a) belongs to R. So, in other words based on what we just
wrote above, it means that the identity relation is included in R. So, it does not mean that is
the only thing. The identity relation has only the reflexive elements. A relation that is
reflexive will have the identity pairs and it will have other pairs, but it must have all the
identity pairs to be called reflexive.

A symmetric relation for instance is one where if (a, b) is there, then (b, a) must be there. So,
for instance looking at reflexive relations, one example is the division relation. So, if we
provided we make sure that the numbers are not 0, then we know it is reflexive because every
number divides itself. So, if we take the reflect division relation as the relation that we
introduced in the first part of this lecture that would be reflexive because a divides a for every
a which is not 0.

Similarly, symmetric relations if we look at pairs where the greatest common divisor is 1, in
other words they have no common divisors. This is what happens for example, in reduced
fractions, then it does not matter whether we write it as (a , b) or (b, a). So, if (a, b) has
greatest common divisor 1, so does (b , a). So, (a, b) and (b, a) must both either be there in
the relation or neither will be there.

Similarly, if we look at this which is asking about the absolute value so, it is saying give me
all numbers a and b such that a - b is either 2 or -2. So, the absolute value takes the difference
and removes the negative sign. Now, we see that for instance if (5, 7) is there, then (7, 5)
must be there because they both have the same difference depending on how we write it.
Normally, in subtraction we have a sign difference, but because we are taking the absolute
value there is no difference actually between these two. So, this absolute value relation also if
we fix is a symmetric relation.
(Refer Slide Time: 14:51)

A third property that relations may have and which are useful is called transitivity. So,
transitivity says that if we have two pairs which are related such that they share an elements.
So, a is related to b and b is related to c, then a must be related to c. So, again our divisibility
is a relation. So, supposing we say that 2 | 6 and we say that 6 | 36, then from this we can
conclude that 2 | 36 as well, right.

Similarly, if we take less than if we say that 3 < 10 and 10 < 28, then we know from this that
3 < 28. So, this is transitivity.

(Refer Slide Time: 15:34)

So, if we want to draw it pictorially if we have three elements a, b and c and this arrow
remember we had this graph notation which says a is related to b and b is related to a, then
this dashed line represents the requirement for transitivity a must be related to c.

(Refer Slide Time: 15:49)

Now, we saw symmetry. So, symmetry says that if (a, b) is in R, then (b, a) must also be in R.
Anti-symmetry says something different it says if (a, b) is in R, then (b, a) should not be in R.
So, less than for example, which was transitive above is also anti-symmetric. If you take
strictly less than, if a is strictly less than b; then it cannot be that be strictly less than a. So,
this is an anti symmetric relation, but anti symmetry does not require that one of the two must
be there. It only says that if one pair is there the opposite pair should not be there ok.

Similarly, if we look at our mother and children example; obviously, if p is the mother of c
then c cannot be the mother of p ok. Now, there may be p and c such that neither p is the
mother of c nor is c is the mother of p. So, that is allowed. We do not insist that every pair (p,
c) must be related one way or another, but if it is related one way it should not be related the
other way is what anti-symmetry says.
(Refer Slide Time: 16:46)

So, if we combine some of these conditions, we get an interesting class relations called
equivalence relations. So, equivalence relation is something that is reflexive, symmetric and
transitive. So, as an example supposing we connect together all numbers which have the same
remainder modulo 5. So, for instance 7 has a remainder 2 with respect to 5 and so does 22.
So, 7 and 5 would be related in this way if we define the relationship as having the same
remainder modulo 5.

Now, notice that if two numbers have the same remainder modulo 5; that means, that going
from one number to the other you are going in multiples of 5. So, for instance 22 -7 is 15
right. So, this is this modulo arithmetic. So, if you add the number that you are dividing by,
then you get the same remainder and so, in set notation we can say that the integers modulo 5
are all pairs a, b such that b - a mod 5 is 0. In other words, we are not asking what is the
actual remainder of b and a, we are just saying that b and a are separated by a multiple of 5
therefore, they must have the same remainder modulo 5.

Now, this divides the integers into five groups if I based on the remainder. So, there are the
group of numbers which are divisible by 5, they have remainder 0. Those like 6, 11 and all
which have remainder 1; 7, 12 and all which one remainder 2 and so on. So, we have five
possible remainders 0, 1, 2, 3, 4 and therefore, this divides the set of integers into five disjoint
classes.
As an example of modulo arithmetic that we are all familiar with, consider what happens
when we look at a normal clock. Now, a normal clock measures time from 0 to 12 and then
cycles around again. So, though there are 24 hours in a day, the clock is actually partitioning
these 24 into two sets where we have 0 and 12 as same, 1 and 13 as same and so on right. So,
2 am and 2 pm, there is no distinction on the clock.

So, the clock is actually showing us this equivalence class of hours regarding am and pm as
being equal and we have to know from context whether the clock is showing am or pm. So,
the main thing to note about an equivalence relation is that it partitions a set. It partitions a set
into disjoint groups, all of the elements within a group are equivalent and all of the elements
outside across groups are not equivalent to each other.

So, the groups of equivalent elements that we formed through an equivalence relation are
called equivalence classes. So, this might look a little abstract now, but equivalence classes
really represent a kind of equality and sometimes we are happy to work with this equality in
terms of equivalence relations rather than actual equality and it has very much the same
properties as equality does.

(Refer Slide Time: 19:28)

So, to summarize as we have seen a Cartesian product can generate n-tuples of elements from
n sets. So, if we have X 1, X 2 , up to X n , n sets these can be different or the same, then we can
take one element from each set and form an n-tuple x 1, x2 , up to x n. And, when we now pick
out some particular subset of these n-tuples, we get a relation. So, for instance, if we take
pairs from N × R and we want the second element of the pair the real number to be the
square root of the first element, then we get N × R such that r = √m.

So, here on the right we have seen we show one picture of this. So, there are some elements
like (2, √ 2 ), (4 , 2) , (7, √ 7 ) and so on. Now, just notice that in this picture the y-axis is
elongated compared to the x-axis. So, this is not in some sense to scale in both dimensions
because the square root function behaves like this.

(Refer Slide Time: 20:27)

So, we have seen that there are some properties that we would like to record of binary
relations – reflexivity, symmetry, transitivity and sometimes anti-symmetry. And, using
reflexivity, symmetry and transitivity together we get what is called an equivalence relation,
an equivalence relations partition sets into equivalence classes which behave like equality.

Thank you.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 07
Functions

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:14)

(Refer Slide Time: 00:17)

So, closely related to relations are functions. So, what is the function? A function is a rule
that tells us how to convert an input into an output. So, for instance suppose we want a
function that given an x returns as x2, then this is one way to write the rule. We write this
symbol which says x maps to x2; given an x it is transformed to x2, but more conventionally
we also give a name to the function. So, in this case we can call it square(x).

So, square(x) takes a parameter x as input and it produces as output; some value which
transforms this parameter, in this case x2.

(Refer Slide Time: 00:56)

So, we can plot x versus x2 by putting all the points where the second coordinate is the
function value of the first coordinate. So, if we look at x2 for instance, it forms this up you
know inverted parabola shape which you should be familiar with. And notice that because for
instance 22 is the same as (-2)2, there is a symmetry about the y axis.

So, for instance 22 is the same as (-2)2, and 32 would be the same as (-3)2 and so on.

(Refer Slide Time: 01:31)

So, when we define a function, we have to be careful about specifying what set we take the
input from and what sets the output produces. So, the input set is called the domain. So, for
instance the domain of square as we have defined it above is a set of reals, so we can take the
square of any real number.

Now the output when we apply square, we know that it is going to be a real number; so the
codomain as it is called is the output set of possible values is called the codomain, in this case
is the reals. But of course, we know that when we square a number; even if the input is
negative, the output is going to be positive. So, even though the codomain is a set of all reals,
we cannot get all reals as output of the square function. So, there is a separate name for that
called the range.

So, the range of a function is a subset of the codomain; the range tells us what values the
function can actually take. So, in this case the range of the square function is the non-negative
reals. So, this is all real numbers greater than equal to 0 which is sometimes written like this
and if you want to explicitly write it out; it is the set of all r in the set of reals such that r >= 0.

So, in order to specify a function abstractly and describe its domain and codomain, we
usually write that f which is the name that we give to an arbitrary function is a function from
X the domain to Y the codomain. So, this notation f : X to Y tells us without telling us what
the function is actually doing; it tells us on what sets it operates, what is the input set and
what is the output set.

(Refer Slide Time: 03:08)

So, the close connection between functions and relations is that we can associate with every
function f a relation R f ; and R f is merely all the pairs of inputs and outputs that the function
allows. So, for example, with our square functions sq we have R sq as all pairs (x,y), such that
y is equal to x2.

So, this is actually sometimes simplified by saying y is equal to x2. So, we do not write out
f(x) and then say f(x) is y; we just directly say y is equal to x2 to denote that the output is the
square of the input. So, this is an implicit notation, where we are implicitly naming the output
for each x as y. So, notice that if we talk about it as a relation; remember that a relation is a
subset of the Cartesian product of two sets. So, in this case, the Cartesian product is formed
by the domain of the function and the range of the function, and then the relation is a subset
of the domain X the range.
So, what are some properties of this relation? Well, first of all when we define the domain of
a function, we really mean that the function is defined at every possible value in that domain.
So, for every x and domain of the function f, there must be a valid value f(x); so there must
be a y such that (x, y) belongs to the relation R f . The other property is that this is a rule for
producing an output from an input; so there can be no confusion about what the output is.

So, for each x that we feed in as a domain value to the function, there must be as exactly one
output value f(x) that we get out. So, there is only one y in the codomain, such that (x,y)
belongs to R f . And in fact, we saw in the lecture on relations that, we would draw relations
by plotting the points which form part of the relation. So, technically when we are drawing a
graph of a function as we have done here for this parabola, we are actually drawing all the
points which satisfy the relation R f .

So, plotting a graph is the same for functions and relations; because implicitly we are plotting
the relation that corresponds to a given function.

(Refer Slide Time: 05:23)

So, let us look at some other functions that we will encounter as we go along. So, if we have
a function of the form something x + something. So, mx + c, then this defines a line. So, then
the like we see a line 3.5x + 5.7. And what we will see as we go along in this course is that,
the quantity which multiplies x is called the slope and it determines the angle at which the
line goes; and the other quantity which is without x determines the intercept.
So, notice that if you set x = 0, then the first term goes to 0; this gets cancelled out, if x is 0.
So, the answer will be 5.7. So, when x is 0, you get 5.7. So, what the second term tells us is
where this line crosses the y axis.

(Refer Slide Time: 06:09)

So, if we change these two values, we get different lines. So, for instance if we change the
intercept and keep the slope the same; then we get a line which has the same slope it is
parallel, it is at the same angle. But now the intercept is -1.2; so it crosses the y axis lower, so
the whole line is shifted to the right.

(Refer Slide Time: 06:28)

On the other hand if we keep the intercept the same; but we change the slope, we get a
different slanted line. So, here we have reduced the slope from 3.5 to 2; so it is a shallower
line and the green line passes through exactly the same point 5.7 as the previous one, but it
has a shallower slope.

(Refer Slide Time: 06:49)

And we can change both and in fact, we can put a negative slope; so if you have a negative
slope, it comes down rather than going up, so we have this line coming here. And notice that
it crosses at 2.5, so that is the intercept. So, by changing the values of the slope and the
intercept, we get many different lines and many different functions.
(Refer Slide Time: 07:07)

And for all of these functions that we have defined the domain is the set of reals, the
codomain is the set of reals; but also because we can intuitively see that the line goes from

way down -∞ to way up +∞ whether it is going up or down, it can take all values in the real.
So, not only is the codomain equal to R, it is also the range.

(Refer Slide Time: 07:28)

So, here is another function x maps to √x. The first question is, is this a function? So,
remember that for a function, we need it to be defined on every input value and we also
needed to have a unique output. So, remember that when we square a negative number, we
get the same as when we square the positive version; so 52 and (-5)2 are both 25.

So, technically if we take √25, we cannot determine whether we are talking about +5 or -5.

So, when we write √x as a function, our convention is that we are taking the positive square
root. So, the function on the right plots the positive square root; if we were to take the
negative square root, then it would be a symmetric curve going below. And now if we take
both these together, then this is not a function; because if we take any x value, we have two
possible outputs for this which is not allowed. So, we are taking by convention the positive
square root.

(Refer Slide Time: 08:23)

Now what is the domain of this function? Well it depends on what we allow the codomain to
be. We have seen that negative numbers cannot have real square roots; no real number can
multiply itself to produce a negative number, because of the law of signs for multiplication.
So, if we insist that the output should be a real number, then the domain of this function, the
function can only be defined when the input is not negative. So, we have this set which we
defined before; the set of reals bigger than or equal to 0.

On the other hand, if we move to the set of complex numbers which we said we are not going

to describe in detail; the set of complex numbers includes √-1 and implicitly through that the
square root of all negative numbers. So, once we allow complex numbers as the output of our
function, then we can define square root on all the real numbers.

So, the notion of domain and range is kind of flexible depending on how we are going to use
the function. So, we have to be very clear when we are using a function what context we are
using it in.

(Refer Slide Time: 09:24)

Now we saw when we looked at relations that there are some properties of relations which
are interesting like reflexivity, symmetry and so on. Similarly there are properties of
functions which are interesting; the first interesting property of function is whether it is one to
one, whether it is injective.

What this means is; if I give you different inputs, does the function always produce different

outputs? If x 1≠ x2 , is it guaranteed that f( x¿¿1) ¿≠ f( x¿¿2) ¿? So, if we look at the linear

function that we saw before the line, then we can see that it is injective; because if we change
x, we move along the line to a new point. So, no two x points, point to the same y point; so
therefore, this is an injective function.

If on the other hand, we take a parabola as function which of the other form something
squared, so 7 x2 for instance. Then we already saw that f(a) is the same as f (-a), so there will
be two points; the plus version and the minus version, both of which has the same output. So,
it is not the case that distinct outputs produce distinct, inputs produce distinct outputs; so the
square function is not injective.

(Refer Slide Time: 10:32)

On the other side we talked about the distinction between the codomain and the range; we
said that the codomain is the set of values into which the function produces answers, but the
range is the actual set of values of the functions can take.

So, the question is, whether or not all values in the codomain are actually touched by the
function and this is called surjectivity or onto. So, the range of a surjective function is in fact
equal to the codomain, which says that for every y which is in the possible codomain of f;
there is actually an x in the domain of f, such that f (x) = y.

Now, once again if we take a line, then this is surjective; because if I pick any point y, I can
find a point x, I can solve for x for example, which gives me that y. On the other hand if I
take a parabola, in this case we have shifted the parabola up, so it is 5 x2 +3 . Then we can see
that, first of all a parabola with no shifted, if I did not have this +3 term; then we know that it
can only take positive values, because x2 will always be a non-negative number.

Now if I further add +3, it can only take values 3 and above; so this definitely is not
surjective, the domain codomain is a set of all reals, but the actual range is only if the reals
which are bigger than or equal to 3. Similarly if I take this 7 √ ❑function, then we know that
even if we take the codomain to be R; so we only take square roots of positive numbers. We
know that we will never get a negative answer, because by convention we have taken positive
square roots.

So, this is again not a surjective function. So, these are two important properties of functions,
are they injective is it one is to one; if I give you different inputs, do I get different outputs
and is it surjective, is it onto, does every possible output have a corresponding input that
maps to it.

(Refer Slide Time: 12:25)

So, if you combine these two, you get something called a bijective function. So, a bijective
function is something with where there is a one to one correspondence between the domain
and the codomain.

So, every x in the domain maps to a distinct y in the codomain and every y in the codomain
has a unique x that maps to it. So, from the statement it looks clear that this corresponds to
injectivity and surjectivity. So, actually this is the theorem that a function is bijectuve if and
only if it is both injective and surjective.

Now this may look obvious, but actually only one direction is obvious, from the definition,
we can see that if a function is bijective; it must be injective, because it says every x maps to
a distinct y, so no two x will map to the same y.

It also says it is surjective, because it says every y in the codomain has a unique pre image.
So, the fact that a bijection implies injectivity and surjectivity is part of the definition; the
other way requires a small argument. So, supposing a function is injective and surjective, we
have to show that it is bijective. So, for this, we have to guarantee first that every x maps to
unique y; but this is guaranteed because the function is injective, injectivity says if I have two

inputs x 1and x2 which are not the same, f ( x1 ¿ ≠ f ( x2 ¿ . So, this is fine.

What about surjectivity? So, surjectivity says that everything in the output comes from some
input not necessarily unique; but if two things map to the same output right, if two things map
to the same output, if I have a y such that I have x 1and x2 mapping to the same y. So, if it has
even, if a surjective function if the output has two pre images; then these two pre images do
not satisfy injectivity. So, if I combine surjectivity in the presence of injectivity, I know that
the pre image is unique; and therefore these two conditions guarantee that I have a bijection.

(Refer Slide Time: 14:15)

So, an important use of bijection is to count the items in a set. So, remember we said that the
cardinality of a set is the number of items and if you have a finite set, we can count them.
Now supposing somebody gives you two large sacks filled with marbles or balls and ask you
to check whether the two sacks have the same number of balls each. So, think of these sacks
as sets and these balls are a large number of elements.

Now, you could of course, count the marbles in each sack, but this is a bit tedious; because
we know that as we are keeping track of these small objects, we often lose count or miss
count or add one or plus one. So, at the end, we have to be doubly sure that we have counted
correctly, so we will count it a number of times. So, counting the marbles in each sack and
then checking if the two counts are equal is a tedious process and it is error prone, if we do it
manually.

Now, here is a manual process which is less error prone. Supposing we put our hand into
each sack and pull out a marble from each sack and put it away somewhere; then we put our
hands again in and take out one marble each again and put it away somewhere. So, with each
move, we are taking out one marble from each sack. So, what can we say; well if the two
marbles sacks get empty together, then we pulled out one from each. So, we have actually
established that there is a one to one correspondence between the marbles in the first sack and
the marble in the second sack.

If on the other hand when we find one sack is empty and the other sack is not empty; this
means that up to this point, we pulled out an equal number of marbles from both sacks and
now one sack has extra marble, so they were not equal. So, in this way establishing a
bijection is equivalent to saying that two sets have the same cardinality. So, for finite sets this
is a convenience; but for infinite sets this is the only way in order to establish that the
cardinality is the same.

So, for instance supposing we want to know whether the number of lines that we can draw is
the same as the number of points on this plane R X R . So, R X R is a set of all points that
you can draw on this plane and the number of lines we can draw is a number of such straight
lines that we can draw; are these the same? Now it may not seem obvious how to argue this
one way or another; but remember that we said that every line can be represented by a
function of the form mx + c. And we also said that if you change m, you get a new line and if
you change c, you get a new line. So, m and c together uniquely define a line.

So, since m and c together uniquely define line; every pair (m, c) defines a line and every line
defines a pair (m, c), so there is a one to one bijection between the lines and the pairs of
points on this plane. So, actually the number of lines is the same as R X R. So, think about it,
because this may not be obvious at first sight; but by establishing a bijection in this way, we
can say that the number of lines that we can draw on a plane are equal to the number of points
on a plane.
(Refer Slide Time: 17:09)

Now, suppose we extend this argument; if we take any two points right, if we take two points
say x 1and x2 , we can draw a unique line passing through these points. So, this is a well known
fact from geometry.

(Refer Slide Time: 17:23)

So, we know that the number of lines has the same cardinality as R X R that is what we
claimed in the previous argument. Now we say that every pair of points defines a line. So,
can we say that every pair of points therefore, has the same cardinality? So, remember this is
a pair of points.
So, we have one point here and one point here. So, do we say that every pair of points has the
same cardinality as the set of all points? So, it is R 2X R 2 the same as R XR, is this an
argument for that? So, important thing is to ensure that we have a bijection; the problem is
that this is not a bijection, because along any line we have many points, right.

So, if I take these two points, indeed it forms a unique line; but I get the same line if I take
these two points for instance. So, it is not the case that every pair of points that I pick
generates a different line. So, unless I can show you that pairs of points, different pairs of
points generate different lines; I do not get a one is to one correspondence between pairs of
points and lines, and therefore this bijection breaks down.

(Refer Slide Time: 18:24)

So, whenever we are trying to use a bijection to describe some kind of a correspondence and
count points especially in an infinite set, count elements of an infinite set, compare infinite
sets against each other; you must make sure that the function you are defining is really a
bijection.
(Refer Slide Time: 18:39)

So, to summarize a function gives us a rule to map inputs to outputs. And with each function
we have to specify three sets; we have to specify the domain, so the function must be defined
on every set in the element of the domain set, the codomain what are the output elements
supposed to look like and the range which was actually the output assumed by the function
once we applied.

So, not all elements in a codomain may actually be attainable by the function; the range is
those elements which you can reach through the function. With each function we can
associate a binary relation consisting of all pairs (x , y), such that y = f (x). Then we saw
some interesting properties that we would like to prove for functions in order to make use of
them; one is injectivity that is every pair of distinct inputs produces distinct outputs, so this is
one is to one. And surjectivity which says actually that the codomain and the range match;
everything that I could possibly generate, can in fact be generated by applying the function.

Then we saw that a bijection combines these two. So, a bijection gives us something which is
an injection and a surjection; something that is one to one and onto. And once we have a
bijection between two sets, we can actually argue that the two sets have the same cardinality
and this is often the only way to prove that two infinite sets have the same cardinality.

Thank you.
 Mathematics for Data Science 1.
Professor Madhavan Mukund.
Department of Computer Science
Mathematical Institute, Chennai.
Lecture-7A.
Relations: Examples.
So, earlier we defined relations as subsets of elements of a Cartesian product which have special
properties. So, let us take a look at relations again and understand why we are so interested in
relations.

(Refer Slide Time: 00:25)

So, remember that a Cartesian product takes all pairs of elements from a collection of sets. In
particular, if you say A cross B, you are taking 2 sets A and B, and you are taking every pair of
elements of the form small a small b such that the first small a comes from capital A and small b
comes from capital B. The order is important, the first element in the pair comes from the first
set, the second comes from the second set. So concretely, let us look at these 2 sets.

So, suppose A = {1, 4, 7}, so it has 3 elements, and B = {1, 16, 49}. So, if you now look at A ×
B, it looks at every pair. So, if you can take this one, and combine it with 1, 16 and 49 to get
this. Then you can take this 4 and combine it again with 1, 16 and 49 to get these for 3 pairs, and
finally you take 7 and then you combine it again with 1, 16 and 49 to get these pairs.

So, it is easy to see that if you have m elements in the first and n elements in the second, every
one of those m elements is paired with every one of the n elements, so you get m × n pairs. Now,
the first thing to remember is that the Cartesian product is ordered. So, there is a first and there is
a second. So, if you reverse this and say B × A, you do not get the same set of pairs, every pair is
reverse. So, (16,1) replaces (1,16), (49,1) replaces (1,49). So, this is the first thing to remember
about Cartesian products.

The other thing to remember is that there is no relation, there is no constraint on what you can
take the Cartesian product of. You can very easily take the Cartesian product of a set with itself.
So, the set to itself is not just pairs of identical elements, but also pairs of non identical elements.
So, if you take B × B, you get Of course, (1,1), (16,16), (49,49). But you also get the dissimilar
pairs like (1,16), (16,49), (49,16), and so on.

So, this is an example with 2 sets, but there is nothing to restrict us to 2 sets. So, in general, a
Cartesian product can take a large number of sets and gives us tuples. So, for instance, if we take
3 sets, we get these triples, each element has 3, each element in the Cartesian product has 3
elements in order.

So, here for instance, if I do A × B × A, I take every element in A, combine it with every element
in B and then with A again. So, I have 1 from A, 1 from B and 1 from A. Then I have 1 from A,
1 from B and 4 from A, the second copy of A and the first copy of A are different.
So, I have (1,1,1), (1,1,4), (1,1,7), then I move to the second element of B, I have (1,16,1),
(1,16,7). Now, ultimately the Cartesian product is a set, so it does not matter in what order I write
these triples. But to order to write them down systematically, it is convenient to write them down
in this particular way, where we go through each set one by one, otherwise, we may miss out on
something. So, the reason we need Cartesian products is because they are the building blocks of
relations.

So finally, what we want is not all these pairs or triples, but some of them which are of interest to
us. So, for example, from the first Cartesian product A × B, we may be interested in the pairs
where each element from A is paired with a corresponding position B. So, the first element in A
is paired only with the first position in B, second with the second and so on. So, we might want
to say that we want S, a set which is a subset of A × B, which from those 9 different pairs picks
out only 3 of them of interest, (1,1), (4,16) and (7,49).

Now, if as in this case, there is some way of describing this, which is more abstract, you can also
use a set comprehension. So, we can talk in terms of positions or observe that in this particular
case, the second element is always a square of the first element. So, we could also write this as
the set of pairs (a,b), where (a,b) comes from A × B, so we are generating every possible pair in
the Cartesian product.

But then we are filtering, remember that we had these filters, so we are filtering it so that we only
retain those pairs for which the second component B is the square of the first component. So, this
is how relations are defined. They are typically defined as subsets of the Cartesian product. And
we can either write out the subset explicitly or try to express it implicitly using the set
comprehension notation.

(Refer Slide Time: 4:57)

So, we saw some examples. So, let us look at these examples again more carefully, some
examples from numbers. So, divisibility is an important relation when we are talking about
natural numbers or integers. So, divisibility talks about pairs of natural numbers, such that the
first one divides the second one. So, we want (d,n) such that d divides n, remember this notation,
this perpendicular bar for numbers denotes, this is not the same as the one that we use in set
comprehension.

So, here it is an operation, arithmetic operation which says d divides n, so if I divide n by d, there
is no remainder, it is a 0, d perfectly divides n. So, this would have this divisibility relation
would have pairs like (7,63) because 7 × 9 = 63, or (17,85), because 17 × 5 = 85, and so on. So,
we have a large number of pairs of divisors and numbers which the divisors divide equally,
evenly. So, this we can write in our set comprehension notation because this is an infinite set, so
we have no other way of listing everything.

(Refer Slide Time: 06:09)

So, we take all pairs N × N, (d,n), such that d | n. So, this is our filter. So, we want to generate
everything of this form, but filter out under the condition that d must be a divisor of n and keep
all such pairs. And this we can call d, the divisibility relation.
(Refer Slide Time: 6:2 1)

Now, this is the relation on pairs of natural numbers, so we only get positive divisors. If we
extend it to integers, then we will get even negative divisors. We know that (-7) × (-9) is also 63,
because the 2 negative signs will cancel out. So, if you extend the generating set from N to Z,
from the natural numbers to the integers, then we get a larger set of divisor pairs. So, we get
minus and plus elements for the same pairs that we had in the original relation.

Here is another example. Let us look at what we call prime powers. So, a prime power is
something that is a prime multiplied by itself for a certain number of times. So, for instance, we
can say that 55 = 3125. So, 52 = 25, 54 rather, 52 , 53 = 125, and 54 = 625. So, 625 is a prime
power, similarly 343 = 74 , so it is a prime power and so on. Why is (3,1) in this relation because
anything to the power 0 is 1 by definition. So, 30 = 1, in fact, anything to the power, so any
number to the power 0 is 1. This is by definition. So, for every number comma 1 will be a prime
power.
(Refer Slide Time: 7:41)

So, if you want to define prime powers, it is useful to first define primes. So, one way we can
define primes is to say, give me a natural number, such as the factors of the natural number
consists of exactly 2 elements, 1 and the number itself. And because in sets, we do not
distinguish duplicates, in this definition, if I just say factors(p) = {1, p}, it includes a case where
p is 1, because factors(1) = {1,1}, which is just 1. But I do not want to count 1 as a prime
number. So, we also specify that P is not 1. So, this is the set of primes.

And now, we can say the set of prime powers is the set of all pairs in P × N, where P is defined
above, P × N, such that n is the power of p. So, n = pm for some m, which is a natural number,
which could be 0. That is why we get (3,1). So, this is an example that we also talked about. It is
saying that when you are writing the set comprehension, you can write these kinds of statements.

So, you do not have to be very precise about what you are writing mathematically in terms of
notation, as long as the understanding is clear, there is no ambiguity about what you mean. So,
you can write words like for some, you can also write it in a mathematical notation using
symbols for there exists and for all and so on, but it is not necessary. As long as you are precise,
you can use set comprehension notation in a flexible way.
(Refer Slide Time: 9:00)
So, these are relations in a formal sense. But why are we so interested in relations especially in
the context of computing and data. So, let us look at relations which go beyond numbers. So,
here is an example. Supposing we are talking about an airline, which serves a set of cities and we
are interested in the routes that this airline serves. So, let us C be the set of cities where the
airline operates. So clearly, the airline operates between some pairs of cities, but not all of them.

So, some of these cities are connected by direct flights and for other situations, you have to take
a hopping flight which goes from city A to city B and then from city B to C. So, let us look at
that subset D of direct flights between cities in C. So, this is an example of a relation. Not every
pair of cities is connected by a direct flight. So, if you take all possible pairs of cities, some of
them are connected by direct flights, and some are not. So, this way, information about an
airline's route is really a relation in the sense that we mean.

Now, we have defined certain properties of relations, we said that the relation is reflexive. Now,
this is useful to ask this question because we are talking about a relation between a set and itself.
So, we can ask whether every element in the set is related to itself or is not related to itself. So,
reflexive means that always we have (a,a) in D, for all, for every a. And irreflexive means,
exactly the opposite of this is never in D and for all A.
So, the question is, in terms of direct flights, is this going to be a reflexive relation and
irreflexive relation or neither. Well, it is easy to see that this should not be reflexive. Because we
do not expect an airline to actually operate a flight which takes off from an airport and then lands
immediately in the airport. And in fact, we would precisely like it to be irreflexive, that is, this
should never happens.

So, this should not be reflexive because we do not want every airport to serve itself and we want
it to be irreflexive because we want no airport to serve itself. So, this is an example of an
irreflexive relation. Now, is it a symmetric relation? So, symmetric relation says that whenever I
have a pair of cities in the relation, then I will also have the reverse pair in the relation. So, if I
can fly from one city to another directly, then I can also fly back.

So, concretely for instance, if I take any 2 cities a direct supposing there is a direct route from
Bangalore to Delhi, then is there always a direct flight back from Delhi to Bangalore. Now, if
you think about airlines, this is usually the case. But actually, if you look at domestic flights in
particular, this is typically true only for the bigger cities, it will certainly be true for all the metro
cities and the largest state capitals and so on. But if you look at smaller cities, this is not
necessarily in the case.

For instance, it is quite common for airlines to serve 3 cities in a triangular route. So, you might
have a flight that takes you from Chennai to Madurai, but if you want to come back from
Madurai to Chennai, you cannot fly back directly, but you may have to fly to Salem and then
come. So, between these 3 cities you can get from one to another, either directly or indirectly
depending in which direction you are going. So, this relation is going to be irreflexive but not
necessarily symmetric, it depends on the context.

(Refer Slide Time: 12:29)

Now, one thing you can do is to extend this to a table. So, here is a useful table that we might
want to keep, which might be used to derive other things such as how long it takes to fly or how
expensive a ticket is like to be. So, here we are just recording a fact which is what is the flying
distance between a pair of cities. So, this table says that if the source is Bangalore and the
destination is Chennai, it is 290 kilometers, whereas if the source is Chennai and the destination
is Delhi, it is 1752 kilometers.

So, for every direct flight which our airline operates, you can record this distance and put it in a
table. So, what is important to recognize and this is why relations are so useful in computing and
data is a table is just a relation. So, every column represents a potential set of values. Here, the
first column represents a possible city, so it is taken from the set C, the second column is also
taken from the set C, the third column is a natural number.

If you take pairs of cities which are the same, you could put 0, so it could be from Delhi to Delhi
it is 0. So, in general, you have all possible pairs of cities and all possible numbers, but only
some of them are interesting. Namely, when I have 2 cities which are actually connected by a
flight and the distance the number is actually the real distance. So, it is a relation on C×C×N.

As we said, some relations are useless so we would not record them even though we know them.
We know that for every city, the flying distance from the city to itself is 0, so there is no reason
to record it in the table. The other thing is that unlike our direct flight’s relation, this is actually a
symmetric relation. So, first of all, we will only keep direct flights because we do not want
indirect flights. But distances are definitely symmetric.

So, it doesn't really matter whether there is a direct flight from Chennai to Delhi and back or
whether there is a direct flight from Chennai to Madurai and not back. It is enough to record the
distance from Chennai to Delhi and Chennai to Madurai once each. I do not have to keep the
distance from Delhi to Chennai separately as you can see above, in this example, Chennai to
Delhi and Delhi to Chennai are both exactly the same distance 1752 because that is how
distances work, distances are symmetric.

So, if we have symmetric entries, in a practical sense, when we represent a relation as a table, we
can save on space by not recording the symmetric entries and making a note separately that this
relation is symmetric. So, that is why it is important to know the property of the relation. It is not
just an abstract question, is this reflexive, is this irreflexive, it is actually a practical
consideration, a symmetric relation can be represented by only half the entries in the relation, the
other half followed by symmetry.
(Refer Slide Time: 15:12)

So, let us go further with this. So, another place where we often encounter tables are, for
instance, when looking at data about people. Let us look at students. So, typically a college
would record or a school would record information about students in this form. So, they would
assign a roll number, then they would record maybe the name, the date of birth, and there would
typically be other personal information like maybe their home address, phone number, and so on.

So here, what is important is that some columns are not natural in the sense. So, we know that
everybody has a name and they are born on a particular date, but this roll number is actually
assigned to them by the school or college. And this is something which is designed to be unique,
so no 2 students get the same roll number. So, this kind of column is called a key. And this is
because we want to identify, define each student directly and individually without getting
confused about which student we are talking about.

And unfortunately, the other columns are not keys, 2 students could have the same name. And it
is even possible for 2 students to have the same name and the same date of birth. So, we cannot
rely on the fact that the other columns will uniquely distinguish. So now, if we have a unique roll
number for every student, then each row is identified by the roll number. So, we can actually
think about the row as being something where if I give you the roll number, you can tell me
which row it is and give me the other values in that thing.

So, this is more like a function. A function says given an input give me a unique output. So,
given a roll number, tell me all the values associated with the roll number, the name, the date of
birth, and so on. So, this kind of a stored table is also called sometimes a set of key value pairs,
given the key there is a unique value. I can change the value for a given key by updating it. But if
I add a new entry, I have to add a new key so there is no confusion.

(Refer Slide Time: 17:03)

So, usually a school or college will maintain more than one table of this kind. For instance, there
might be a separate table, where we maintain the marks of the student or the grades of a student
in the courses that they do. And here for conciseness, we might keep only the roll numbers and
the subject names and not the names of the students. So, for instance, in the second table, we
have the roll number, subject and the grade. Here is a typical requirement when we have to
generate a report card.

The grade card has, the grade table has the roll number and the subject and the grade but it does
not tell us who the student is. And that is, for example, it may be difficult for an outsider who
except for the student themselves to know whose roll number belongs to whom, because nobody
would recognize these strange character sequences. So, we want a table that looks like this which
has the roll number and extra column with the name which is not there in the grade table which
is taken from the first table and then we want the subject and the grade.

And here, we see why it is important to have keys because we have this name Payal Ghosh,
which is ambiguous, there are 2 Payal Ghosh’es. And in fact, they have 2 different entries in this
table because they have 2 different roll numbers. So, the Payal Ghosh who got an A in
mathematics is not the same as the Payal Ghosh who got a B in physics. So, this is an operation
which combines these 2 tables. And remember that a table is a relation.

(Refer Slide Time: 18:23)

So, this operation, which combines 2 tables is also an operation which combines 2 relations, and
it is an important operation in computing and in data science called a Join. So formally, a Join
takes tuples from 2 relations and combines them on common values. So here, for instance, you
take any arbitrary roll number, name and date of birth from students, you take any arbitrary roll
numbers subject and grade from grades, but you want that the roll number in the roll number of
the 2 sides belongs the same.

So, the r comes from students and the r′ comes from grades and you want r = r′. And if this is the
case, then you put out a new tuple, which combines the n from the left hand side throws away the
date of birth, we are not interested in preserving the date of birth, keeps the n and keeps the
subject and the grade s and g and of course keeps the roll number which is the same on both
sides.

So, this will ensure that we do not get rows merged, where they correspond to 2 different
students. So, the marks for Abhay, or the grade for Abhay will not be merged with the name and
date of birth for Jeremy Pinto, because they have 2 different roll numbers. So, this is called the
Join and this is a very important operation on relations, and therefore on tables. And this is
something that we use implicitly all the time.
(Refer Slide Time: 19:39)

So, to summarize, a relation describes special tuples in a Cartesian product. And what is really
important for us from a computing and data science point of view is that we work with tables all
the time and tables are really relations. So, that is why relations play such a central role in many
of the things that we are going to look at. So, it is important to get the terminology of relations
right.

And when we combine information on tables, these are actually operations on relations such as
the Join operation that we described, this is only one kind of Join we may have different types of
operations, which we will see in other courses later on. But please, keep in mind that tables are
relations. Thank you.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Department of Computer Science
Mathematical Institute, Chennai
Lecture-1.7B
Function: Examples
So, let us take a closer look at functions now.

(Refer Slide Time: 0:19)

So, remember that a function is a rule that map's inputs to outputs. So for instance, if we are
looking at numbers, a function could take an input x and map it to x2 , which can also write
given a name saying g( x) is equal to x2 , which says g is the name of a function, which when
it takes an input of the form x produces an output of the form x2 .

And with such functions, we have a notion of a domain that is what are the inputs that are
allowed, the set from which we take inputs. Codomain, what is the set to which the outputs
belong and range which is the actual outputs that this input set generates for this given rule.
So, for instance, we have for this function this relation associated with it, all pairs x comma y
such that x and y are reals. So, the domain and the codomain are both reals, the rule is y
equals x2 , so that is the filter that we put, we only want such pairs.

And if we plot all the points which belong to the relation, we get this graph on the right. And
this actually tells us that the range of the function even though the codomain is all reals, the
range of the function actually keeps this function above 0, so we only get non negative reals
as outputs. Now, we are not restricted to looking at functions on numbers, we can also look at
functions on other sets.
So, for instance, if we look at the set of all people in the universe, in the world, in the
country, in any range of geographical regions, we can look for the function mother which
says, given a person, this will map the person uniquely to the mother of that person. So, this
is a function because every person has 1 mother. So, in this lecture, and in general, when we
are talking about functions in this course, we will look more at functions on numbers. So, let
us look at these a little more closely. What are the questions that we really want to ask about
functions on numbers?

(Refer Slide Time: 2:09)

So, one of the basic questions is, what are the ranges of the values that we can get. So, in
other words, we have a core domain. But what is the range of values that we can actually
achieve through the function. So as we saw, this square function, f ( x)=x2 is always positive,
so we always get something between 0 and −∞, there is no upper bound, but we never get
something which is negative.
(Refer Slide Time: 2:35)

On the other hand, if we take a cubic function of this form f ( x)=x3 −3 x 2 +5, then when x
becomes very small, the x3 becomes very small because the cube of a negative number is a
negative number. So, cube have a large negative number, I mean magnitude, the

(−1000)∗(−1000)∗(−1000)=−10 . So, as we go into negative, large negative values, we

−9

can at large negative outputs, same for large positive values. So, this has a range from minus
−∞ to +∞.

(Refer Slide Time: 3:05)

And then there are some functions like the trigonometric function sin x, which oscillate
between an upper bound and lower bound. So, if you take sin x, usually it is between +1 and -
1. If we take 5sin x, then it will be between−5 and +5. So, this has a bounded range. Even
though we consider all possible inputs, we never go outside this range from −5 and +5.

(Refer Slide Time: 3:28)

Now, within the range of values that it can take, we are often interested in specific points, in
particular, where the value are a minimum and where they are maximum. So, for instance,
this function that we have seen before f ( x)=x2, it is clear from the graph on the right that at
0 the output is 0 and at all other points is bigger than 0, so it attains its minimum value at 0.
And because it keeps growing indefinitely in both sides, there is no maximum value.

(Refer Slide Time: 3:57)

Now, the cubic function we said grows arbitrarily small as we go to the negative inputs and
arbitrary large. So, there is actually no maximum and minimum, but it has an interesting
behavior in between because it zigzags it goes up, comes down and goes up again. So, there
is something called a local maximum and a local minimum. So, at x=0, it turns around, so it
achieves a maximum value and starts falling briefly and then at x=2 it turns around again.
So, it achieves a local minimum and goes up again. So, we are interested in finding out where
these local maxima and minima are for various reasons.

(Refer Slide Time: 4:32)

And similarly, if we look at something like sin x, then it has, of course, local minima and
maxima, −¿5 is a local minimum and +¿5 is a local maximum, it is also a global minimum
and maximum because these are the maximum and minimum values that the function can
ever attain. And now, these values are actually attained infinitely often periodically as we go
from left to right.
(Refer Slide Time: 4:53)

Another thing which we are interested in about functions is how fast they grow. Thus one
function grow faster than another. So, if you look at our 2 functions, f ( x)=x2, and

f ( x)=x −3 x +5, and we look at their 2 graphs, then it is very clear that the red line,
3 2

although initially on the right, it is below the green line, it overtakes it, and after that, it is
never going to be below the green line. So, in this way, the cubic function grows faster than
the square function.

Now, why is this interesting? Well, we often see this informally stated in various contexts.
So, let us look at a context which is relevant for you. So, let G( y) be the number of data
science students graduating in a year y. So, as the year increases, so we go from 2020 to
2021, and so on, the value G takes a certain number and hopefully because courses are
growing, this number is increasing.

At the same time, there are jobs being created in data science. So, let J ( y) be the number of
new data science jobs in a year . Now, ideally, you would like that these 2 are comparable,
that the jobs are growing because the number graduates is growing and vice versa. If the
number of jobs increases more than the number of graduates then there is a demand for
graduates and of course, more graduates will opt to study data science. So, you would expect
a demand for this kind of course.

Of course, the unfortunate case might happen the other way around, if suddenly there is a
slump in demand, then people who graduate with a degree in data science will not be
employable and then there will be a reverse trend. So, these are some of the reasons why
when we look at data, we are interested in comparing the growth rate of functions and we will
look at this in the context of the functions that we study mathematically.

(Refer Slide Time: 6:33)

So, to summarize, we will typically study functions over numbers. And we are looking at
many properties of these functions which are interesting to us, for instance, the range of
outputs, where these functions attain local minima and local maxima and what are their
relative growth rates and many other things which we will come across as we go along.
Thank you.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 08
Prime Numbers

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, when we looked at the natural numbers, we talked about divisibility and we talked about
the prime numbers. So, we know that the prime numbers start with 2, 3, 5, 7 and so on. So,
how many prime numbers are there?

(Refer Slide Time: 00:25)

So, remember that a prime number is something that has only two factors 1 and itself. Now, it
must have exactly two factors. So, 1 is not a prime. So, the first few prime numbers are 2, 3,
then 4 is not a prime – because 4 is divisible by 2, then 5, again 6 is not a prime and so on.
So, the question is, is this set of numbers these prime numbers is it a finite set or are there
infinitely many prime numbers?

Now, if there is a finite set of prime numbers, there will be a largest prime number. So, the
same question can be asked by asking is there a largest prime? So, if it is a finite set, in that
finite set, there will be a largest one. And if there is a largest one, then below that largest one
there are only finitely many numbers, so there can only be finitely many primes. So, asking
whether the set of primes is finite is the same as asking whether there is a largest prime.

So, what we are going to see is a version of a proof that goes back to Euclid from about 300
BCE, which shows that there cannot be a largest prime. And as we argued if there is no
largest prime, then it must be that the set of primes is actually an infinite set.
(Refer Slide Time: 01:35)

So, to go ahead with this we need a basic fact about divisibility. So, this says that if a number
divides a+b and it also divides a, then it must divide b. So, let us look at an example. So,
supposing you say that 7 divides 21, and I write 21 was 14+7 then 7 also divides 14, and
therefore, it also divides 7. Similarly, if I say 6 divides 36+24 which is 60; then since 6
divides 36, it must also divide 24 right. So, this is not very difficult to prove. So, let us prove
it just to get a feel of how such proofs go.

(Refer Slide Time: 02:17)

So, since n divides the sum a+b, a+b can be written as a multiple of n. So, let us call it u
times n. Similarly, since we have assumed that n divides a, a can also be written as a multiple
of n; let us call it v X n. So, what we are told is that any a + b is u X n for some u, a itself is v
X n for some v. And the question is b also some multiple of n does n divide b?

Well, because of what we have just discussed a + b can be written as v n + b, because a is v n,

and the sum v n + b which is the same as a + b is in fact u n. So, now, we can do some simple
rearrangement. So, we can take u n = v n + b, and just take the v n to the other side and we
get u n - v n = b and so b is (u – v) times n. So, this simply proves to us that if a number
divides a sum and it divides one part of that sum, it also must divide the other part of the sum.
And we will use this in order to show Euclid’s result.

(Refer Slide Time: 03:21)

So, what Euclid said is that suppose the list of primes is finite. So, if it is finite, then we can
list them out and it is a finite set, so it is some p 1to pk . We do not have to be in any particular
order. we can assume that p 1 is the smallest one; it is 2, p2is 3 and so on. But it does not
really matter as long as this exhaustively completes all the primes.

Now, we construct a new number which is the product of all these primes, you multiply all
these primes by themselves to each other and then we add 1 right. So, n is p1X p2X ...X p k +1.
So, now, the question is what is the status of n? So, since we have assumed that the list of
primes is finite, n must be a composite number, because this is not one of the primes that we
had before right, it is bigger than all of them because it is the product of all of them plus 1.
Now, since it is a composite number it must have a factor other than 1 and itself. And because
we have listed out all the primes one of the primes among them must be a factor. So, let us
assume that p j is a factor. So, p j divides n right. So, there is one in this p1to p k , there is a p j
which divides n. But on the other hand, let us look at this part right. The first part the first
part is the product of all the prime So, p j appears in that product.

So, if it is one of the factors of the product, it must divide the product right. So, p j divides n,
and p j also divides one part of the sum. So, remember what we said that if some number n
divides a + b and if some number n divides a also, then n must divide b. So, in this case a+b
is the product of the primes plus 1, and a itself is a product of the primes and we have argued
that there is one prime p j which divides both of these. So, therefore, by that divisibility result
that we showed in the previous slide p j must divide 1. But of course, we know that p j is a
number bigger than 1, it cannot divide 1. And so we have a contradiction right.

(Refer Slide Time: 05:21)

So, what is the contradiction? Well we assume that n was a new number was a composite
number because we have exhausted all the primes, but in fact, it cannot be composite because
then we cannot find a proper divisor for it among the primes. Therefore, n itself must be a
prime. And notice by construction n is actually bigger than all these. So, it also shows that
there is no largest prime, because for any set of primes we can always construct a larger
prime. So, this is essentially what Euclid did.
(Refer Slide Time: 05:47)

So, we know more about prime numbers. So, prime numbers are very mysterious because
their distribution is kind of unclear, but they also have important properties as we will see.
So, prime numbers have been extensively studied in mathematics in an area called number
theory. So, one of the things that is studied about prime numbers is how they are distributed.
So, as we go a larger and larger in the set of natural numbers, how frequently do we find
primes?

So, π(x) is supposed to denote the number of primes that is smaller than any given number x.
So, for instance, π (4) would be 2, because 2 and 3 are the only 2 primes below 4; π (10)
would include 2, 3, 5 and 7. So, π (10) would be 4 and so on.

Now, as you go larger and larger, the gaps between the primes become larger. And in fact,
you can prove amazing things like the prime number theorem which says that π(x) is
approximately x / log x for large values of x. Now, it does not matter if you do not understand
what this means, but it is important to understand that this is a very significant type of
argument that you can give about the distribution of a set of numbers which is quite in a way
randomly distributed.

Now, in terms of modern applications of primes, it might seem that primes are very strange
things, and we would only need to study them a number theory. In fact, the famous
mathematician G. H. Hardy once said that he was very proud of the fact that he did number
theory and nothing that he studied had any application. Well, it is not quite true because
primes as we will see are actually quite useful.

So, one of the questions that you might want to ask is given a number check whether it is a
prime. Now, of course, there is a brute force way of doing it which is to try and enumerate all
the factors by looking at all the numbers below n and dividing n by them, but that is not
considered to be an efficient way to do it. And in fact, this was proved by three Indian
computer scientists from IIT Kanpur, Manindra Agrawal, Neeraj Kayal, and Nitin Saxena in
2002, and it is one of the breakthrough results in theoretical computer science in the history
of the subject.

So, checking whether a number is prime can be done efficiently. But what about the other
question, if I know a number is not a prime, can I factorize it? So, we know number is not a
prime, but how do I find two non-prime, two non-trivial factors that is not 1 or itself. Now, it
turns out that there is no efficient way to do this. So, this is quite paradoxical. We can check
whether a number is prime or not, but if it is not a prime we can factorize it fast. And this in
fact is the reason why we are so concerned about prime numbers, because we would like to
find numbers which are not prime, but which are actually products of large primes. So, their
factors are only large prime numbers, and this is a very important in cryptography.

And cryptography in this sense is something which affects not just you know military secrets,
but it affects us in day-to-day life because whenever we do electronic commerce our
transactions are protected by cryptography to prevent unauthorized transactions from being
executed on our behalf or to prevent them from being tampered with they are all encrypted.
And a lot of this encryption is based on the existence of large prime numbers, and the fact
that factorizing the product of two large primes is difficult. So, prime numbers though they
are very exotic in number theory are actually a very, very important part of our day-to-day
life.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 09
Why is a number irrational?

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

When we looked at the different types of numbers, we started with the natural numbers, move

p
to the integers, then to the rationals which are expressed as . And then we argued that the
q
rationals do not exhaust all the numbers that we need; and in particular, we claim that the √ ❑
cannot be expressed as a rational numbers, so it is what is called an irrational number. So, let
us try and ask why √ ❑is an irrational number.

(Refer Slide Time: 00:37)

So, the discovery of irrational numbers actually is attributed to the ancient Greeks; and in
particular, it comes from Pythagoras. So, remember that in Pythagoras’s theorem which you
must have studied in school. If you have a right angled triangle, then the square on the
hypotenuse that is the square on the long diagonal side – this one, has an area which is the
sum of the squares on the other side. So, in other words, if you have a right angled triangle
and you measure the three sides, you get a2+ b2=c 2. So, from this, knowing a and b, you can
compute c.

So, in particular, if you draw a square which has one and one as its two sides, then this must
be the √ ❑which is the √ ❑. So, you can actually physically draw if you assume that you can
measure out a unit length using some kind of a measure, then by drawing a square, you can
actually construct a length √ ❑.
(Refer Slide Time: 01:35)

So, for Pythagoras it was very important to understand how to describe the √ ❑as a rational
number, and he and his followers and many times many years trying to prove that in fact it
could be expressed as a rational number. Much after Pythagoras, about 50-60 years after
Pythagoras, one of his followers Hippasus is claimed to have proved that √ ❑is irrational this
was around 500 BCE.

Now, the followers of Pythagoras had a very mystical idea about numbers, and they felt that
numbers could solve everything. And in particular they were very keen that rational rational
numbers should form the basis of all of what we could call it modern day time science and
philosophy. So, the followers of Pythagoras were really shocked by this discovery of
Hippasus, they found it to be a, I mean they could not argue with it; at the same time they felt
that this discovery could not be revealed to the public because they felt it was very
dangerous. So, in fact, it is said that they allegedly drowned him in the sea to prevent this
from being made public. So, the √ ❑being irrational has a rather colorful history. And let us
see now how Hippasus proved that this was actually the case.
(Refer Slide Time: 02:41)

So, let us assume as in many of our arguments. Let us assume that √ ❑was rational. So, if it is
rational, then we know that it can be written as a ratio or fraction of two integers p and q; and
in particular we can assume that it is in reduced form. So, p and q have no common divisor,

p
their gcd is 1. So, if we take √ ❑is equal to , and we square both sides, then √ ❑times √ ❑is 2
q

p p p2 p
2
on the left hand side, and times is 2 . So, we get 2 is equal to 2 . So, we can cross
q q q q
multiply as usual, take the q 2from the denominator on the right hand side to the left hand side
numerator, and we get 2q 2 is equal to p2.

So, what is p2? p2 is p X p. And if it is of the form 2 times something, then it is an even
number, because an even number is something which has 2 as a factor. So, p2has 2 as a
factor. So, p2is an even number. Now, it is a basic fact about natural numbers that if you
multiply two odd numbers, you get an odd number; and if you multiply two even numbers,
you get an even number. So, if p2is even, and p2 is p X p, then both p and p – the two copies
must both be even; so p must be an even number in other words.

So, if p is an even number, then we can write p as 2 times something because p is even p
must be of the form two times something say 2a right. So, from this initial assumption, we
have concluded that the numerator of this fraction which represents √ ❑is actually an even
number of the form 2a.
So, now, let us substitute in this equation for p2right. So, p2is (2a)2 is 4 times a 2. So, now 4a2
is equal to 2q 2. So, now, we can cancel right. So, we can take this 2, and this 2, and cancel it.

(Refer Slide Time: 04:39)

So, we have in other words that q2is 2a 2. And if q2is 2a 2, then by the same argument as
before q2is also even, and so q must be even. And therefore, q can be written as the form of 2
times some other number b. So, we have that p is of the form 2 times a and q was of the form
2 times b. But what this means is that the gcd of p and q must be at least 2, because both of
them are even numbers. So, they are both multiples of 2. So, we claimed initially that the gcd
of p and q is 1. We said that they were actually both in reduced form. So, there was no
common factor other than 1. And now we have shown that if we assume that we in fact
generate 2 as a common factor. So, this cannot be the case. So, the only contradiction that we

p
can resolve with this is by assuming that could not have been there. So, therefore, √ ❑
q

p
cannot be represented by any reduced fraction .
q
(Refer Slide Time: 05:32)

So, this argument of Hippasus is a common way of arguing things in mathematics right. To
show that some fact capital P holds you first assume that not P holds, it is negation holds. So,
we wanted to show that there is no way that √ ❑cannot be expressed as rational. So, we said
let us assume the negation. Let us assume that √ ❑can in fact be express as a rational, and
then you take that assumption and derive a contradiction. And since you cannot accept a
contradiction, your assumption must be wrong and therefore, what you tried to prove
originally was correct.

So, in fact, it is not just √ ❑that is irrational, √ ❑is also irrational. Now, 4 is a perfect square.
So, we know that √ ❑ is 2. What about√ ❑; that is also irrational. So, among the integers
among the natural numbers we have the perfect squares 1, 4, 9, 16, 25 and so on which
consists of 12, 22, 32, 4 2, 52and so on. So, a perfect square is one whose square root is also a
natural number.

Now, it turns out that anything which is not a perfect square has an irrational square root, and
the proof is not exactly the same because we have used a property of 2, and evenness in this
proof, but with a very similar argument you can show this is the case. So, therefore, there are
a lot of irrational numbers that you can generate just by taking square roots of non-perfect
squares.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute

Week - 01
Lecture – 10
Set versus Collections

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, we have looked at sets, and we said that a set loosely speaking is a collection of items.
And then we made some remarks in that lecture that not everything can be thought of with a
set. So, let us ask whether every collection is in fact a set, and if not, why not?

(Refer Slide Time: 00:29)

So, as we said a set is a collection of items. And when set theory was investigated formally
starting from the late 1800s, the idea was to make set theory a foundation of mathematics. So,
let us try to briefly understand what that means. So, we wanted to the mathematicians of the
time wanted to start off with very basic things and build up all of mathematics from that, and
they felt that set theory was a good place to start.

So, some of the mathematicians who are involved in this was Georg Cantor and Richard
Dedekind from the 1870. So, this is a mistake, this is not the 1970s of course, but the 1870s.
(Refer Slide Time: 01:05)

So, one is aspect of this foundational nature of set theory is for insists how do you generate
numbers if you have only sets. So, one of the things that you need if you start with set theory
is the empty set. So, you have it for free. So, what they said is that 0 can be thought of as the
empty set.

So, we are going to use sets to represent numbers, and we are going to use the empty set to
stand for 0. So, what is 1? Well, 1 is a set that consists of 0 and the set containing 0; in other
words it is a set containing the empty set, and the set containing empty set. So, remember that
the set containing empty set this is not the same as this right. The empty set has no elements;
the set containing the empty set has one element.
(Refer Slide Time: 01:52)

Similarly, 2 would be the set which contains 1 in the representation above, and the set
containing 1. So, it is a bit tedious to write out. So, I have not expanded it. But you just take
the expression for one in terms of the empty set replace it twice, and you get the number 2.
And in this way for any number j plus 1, you can get it from the number j by taking the
representation of j adding the set containing the representative j putting it into a new set.

So, these are the natural numbers as expressed using sets starting from the empty set. And
then you can actually define set theoretic ways of combining these two define, the addition of
two numbers and this format to get a new number which is the sum and the product and so
on. So, this is what it means to use something like set theory as a foundation of mathematics.
(Refer Slide Time: 02:37)

So, basically set theory assumes that you have the empty set, and then you have basic set
building operations. For instance, you can take the union of sets, you can take the intersection
of sets, you can take the Cartesian product which we saw when we were looking at relations.
And you can of course to set comprehension which is that you can take some elements from a
set which satisfy a condition and build a subset.

So, now into this picture came Bertrand Russell and he asked whether this would make sense
or not. So, here we come back to our fundamental question is every collection a set? In
particular he asked can there be a set of all sets? So, remember that sets are objects just like
anything else. So, we can collect them together. So, is this collection of all sets in fact a set?

Well, supposing it is a set, then we can do the following. We can apply set comprehension
right, and we can pick out some sets from this collection of all sets. So, we will call capital S,
the subset of all sets that do not contain themselves. So, this is a subset of this hypothetical
set of all sets. So, this capital S is a set because we have applied set comprehension to the set
of all sets. So, we have the set of all sets. And among all sets we have pulled out those sets
which do not contain themselves. So, this is the condition we have applied, and this is
allowed by set comprehension.

Now, the question is does the set that we have constructed belong to itself, does S belong to
S? Well, if it does belong to itself, then it does not satisfy its own definition because elements
of S should not contain themselves. So, S cannot belong to itself, because if it did it would
contradict to way we have pulled out S from the set of all sets. But if it does not belong to
itself, then that is also a contradiction, because then S does not belong to S and by the
condition that we have applied to pull out sets S must be included in that condition.

So, either way we have a paradox; we have a contradiction. So, S can neither belong to itself
nor can it not belong to itself. And this is called Russell’s Paradox. He was the first person
who published this and made it publicly known, but this was also independently discovered
by another well known set theorist of the time called Ernst Zermelo. So, what this really tells
us? If you remember our argument is that we made some assumption, and then from that
assumption we realized that we have a contradiction or an observed situation.

So, something must be wrong in one of our assumptions. And here it turns out that the
assumption that goes wrong in all these is the assumption that there is a set of all sets. If we
did not have a set of all sets, we could not have done the set comprehension, and therefore,
we would not have reached this observed conclusion.

(Refer Slide Time: 05:23)

So, what Russell’s Paradox really tells us is that, not every collection can be called as set in
particular the set of all sets does not exist. So, he went through an exercise of trying to
formulate a different version of set theory which he called type theory and so on, but in
modern mathematics typically if you are not sure that what you are dealing with is a set then
it is safer to just called such a collection a class. So, a class is just a collection of objects
which does not have any of the implied properties that you expect from the sets.
So, this paradox as we said came in the context of set theory being used as a foundation of
mathematics. And, this seem to casts doubts on whether it could be used at all. So, it had a
major impact on this whole mathematical exercise of deriving mathematics from logical
foundations which went on into the 20th century which we will not be able to discuss here
unfortunately, but it is a fascinating subject in its own right.

For us what we have to be clear about is that whenever we use sets we must make sure that
we always start with sets that we have and build new sets from existing sets. So, we can
assume that the numbers are sets. So, we have the set of natural numbers, the set of integers,
the set of rationals, the set of reals and so on. And, whenever we construct a new set we just
have to verify that the set that we started with to construct the new set was already a set.

So, we take a Cartesian product or a union or set comprehension, we always start with old
sets and make new sets. So, those old sets must be well-defined. So, in other words, we
should not manufacture sets out of thin air such as the set of all sets.
 Mathematics for Data Science 1
Prof. Madhavan Mukund
Department of computer science
Chennai Mathematical Institute

Week - 01
Lecture – 11
Degrees of infinity

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, when we looked at the sets of numbers, we said that we have various kinds of infinite sets
– the natural numbers, integers, reals, the rationals, some of them are discrete, some of them
are dense. And the question that we asked was whether they all have the same size, or there
are more of one than another?

(Refer Slide Time: 00:29)

So, the question that we want to ask is, are there degrees of infinity? So, we know that for a
set the cardinality denotes a number of elements, and if it is a finite set we just have to count
these elements. So, for a finite set, there is no problem about cardinality which is the count
the number of elements and we are done.

We get a natural number which is the cardinality of the set. Now, the question is what do we
do for infinite sets. So, let us look at the natural numbers for instance. So, in which we move
from the natural number to the integers, we added negative numbers. So, clearly we have
added an infinite set of numbers we roughly doubled the set. So, is the set of natural numbers
are same as the integer number in size or not?

Similarly, when we move from the integers to the rationals, we move from a discrete set
where we had a next and previous element to a dense set where between any two element
there is an another element. So, this suggests that there should be more rational than reals
rationals than integers, but is that true or not?
And finally, when we move from rationals to real numbers we added a whole bunch of

p
irrational numbers which cannot be expressed in the form . So, clearly the real numbers
q
have a large number of new things which are not in the rationals. So, again is the set of reals
larger than the set of rationals or not? So, this study of the cardinality of infinite sets was
actually undertaken by Georg Cantor in the 1870s. And as we have seen when we studied
functions the correct way to compare the cardinality of infinite sets is to use a bijection.

So, what is the bijection? The bijection is one-to-one and an onto function. In other words, it
allows us to map one set to another set in such a way that two elements are always map to
two different elements and everything on the other side is map 2 from something here that is
the onto part. So, it is one-to-one no onto elements map to the same one, and it is onto no
element on the right hand side is missed out.

So, intuitively what this allows us do through this function this bijection is to pair up the
elements from the one side with the elements from the another side. So, I take an element on
the left hand side, through the bijection I pair up it with an element of right hand side. And
because it is one-to-one and onto, this pair this paring actually exhaustibly covers all the
elements in both sides or nothing is left out.

So, we have paired up everything and therefore, the two sides have the same cardinality. So,
this is the technique that we will investigate in order to resolve these questions about the
cardinalities of the infinite sets of numbers that we have discussed above.
(Refer Slide Time: 02:55)

So, our starting point is the set of natural number, because this is the first infinite set that we
have to begin with. When we start counting we realized that there is no largest number
because we can always add 1. And so if we take all the finite numbers that we can used to
count, we get an infinite set called the natural numbers.

Now, supposing we find a bijection between the set of natural numbers and some other set X,
does not matter what this set is, but supposing there is a bijection. We can pair of the natural
numbers with the elements of X. This means that we can actually effectively enumerate the
elements of X, we can take the number paired with 0, f (0) and call that the beginning of X,
then f (1) is an X element, f (2) and so on.

And because we are doing this kind of enumerating X, we can count X in a way via f and so
we call any such set countable. So, countable set is one which can be bijectively paired up
with the set of natural numbers. So, when we are looking at other sets, we will first check
whether they are countable or if not we have to argue that they cannot be counted.
(Refer Slide Time: 04:02)

So, let us begin with set of integers and show that it is countable. So, why should be, why
should it not be countable, or why should it be a surprise if it is countable? Well, because Z
extends N with negative integers right. So, for every, if you do not count 0 in the calculation,
for every positive natural number there is a corresponding negative integer in Z.

So, Z is referring twice as big as N ; for +1 you have -1; +2 you have -2 and so on right. So, it
seems contradictory that you can double the set, and still have the set of the same size that
you started with. So, the question now is for Z to be countable, can we set up a bijection
between the natural numbers and Z?

So, let us look at Z as we do on the number line. So, it starts from some -∞ and then it

comes to -4, -3, -2, -1, 0, 1, 2, 3, 4 and continues. So, we start our enumeration at 0.

So, we enumerate 0, the 0 of Z as the 0th element, then we map 1 to +1, map 2 to -1.
What do we do next? Well, we map 3 to +2, and 4 to -2.

So, we keep zigzagging to the right hand to the left, we count Z by starting with the center
moving right one, moving left one, moving right one, moving left one. So, in this way we
could now enumerate the number +3 as 5, -3 as 6, +4 as 7, -4 as 8. So, in this way we can
actually enumerate Z effectively. So, f (0) is 0 as we saw. If i is odd for example, 1 then f (i)

i+1
is .
2
So, f(1) for instance is (1 + 1)/ 2 = 1; f(3) = (3 + 1) / 2 = 2 and so on. So, if f is odd, I have

i+1 i
. And if it is even like 2, then I take - . So, I take -2/2 which is -1. If it is 4, I get -4/2
2 2
which is -2 right. So, we have actually given an effective way of assigning a position in some
sense or count to every number in Z, and this shows that the set of integers is actually
countable.

(Refer Slide Time: 06:17)

(Refer Slide Time: 06:20)

Now, what about the rationals? One reason why we might suspect that the rationals are not
countable is because the rationals we saw a dense between any two rational numbers there is
an another rational number.

Whether the integers and the rational numbers are discrete, you can always find a next
number; and in the case of integers you can always find a previous number. For natural
numbers 0 has no previous number, every other number has a previous and a next. So, given
that rationals are dense and the integers are discrete, the question is are there more rationals
than there are integers?

Now, there is an obvious bijection between pairs of integers and rationals because that is
what a rational is, rational is a pair of integers p upon q. So, I can take a pair (p , q) and Z

p
cross Z and directly connect it in an bijective way to the fraction . So, every pair gives a
q
unique rational number, every rational number gives me a unique pair.

There is no surprise here, there are no we are not talking about reduce forms of for example,

1 2 3
we have different numbers like , then we have , and , these are all different rational
10 20 30
numbers they may represent the same value, but they represent different pairs. So, this is a
clear bijection between Z cross Z and Q. So, Z cross Z has the same size of Q.

So, if we are looking at the cardinality of Q, we can also look at the cardinality of Z cross Z.
Because if we can measure the cardinality the size of Z cross Z, then through this bijection, Q
must have the same size, there is no need to separately measure the size of Q.

So, instead of Z cross Z just to make the picture easier to see, we will actually do N cross N ,
and then I will show you how to extend it to Z cross Z. So, here is a picture of N cross N . So,
remember that we think of N cross N in a two-dimensional grid and at each point (i, j) I have
a dot representing the pair (i, j). So, for instance this pair, this pair is (5 , 4), because it comes
from the 5 and the 4 over here right. So, every dot in this pair in this grid is a pair in N cross
N.
(Refer Slide Time: 08:22)

Now, I am going to enumerate this in a particular way. So, here is a one enumeration. So, you
start with the 0th element as the element at the bottom left corner what is normally called the
origin. Then you enumerate the first diagonal right, so you go from here and then you go
right and then you go up. So, you enumerate in this way then you continue.

(Refer Slide Time: 08:47)

So, you started from here went up, then you up there, and come back down again right. So,
you can slice this thing like this right. So, you can slice this gird like this, and enumerate it
diagonal by diagonal.
(Refer Slide Time: 08:58)

So, this gives us an effective enumeration of N cross N .

(Refer Slide Time: 09:04)

But we can also enumerate in different ways. For instance, we can enumerate in these larger
and larger squares. So, we can start here, then finish this, then do this, then do this, then do
this and so on right. So, long if we do not miss out any point in the grid we are done.
(Refer Slide Time: 09:21)

So, this shows us that N cross N is something that we can enumerate. Now, how would we
do it for Z cross Z? Well, it is very simple. If I had Z cross Z, I would also have points on
this side, and I would also have points below right. So, I would have points to the left and
below 0 because I would had negative numbers.

So, now, if I wanted to enumerate Z X Z, I would start here, then I would do this, and I
would complete this diamond, then I would go here, and then go here, and then complete this
diamond and so on right. So, instead of doing just the diagonal, I would extend the diagonal
around to form a diamond, and in this way I would start from the center and spiral out so that
I enumerate all the numbers in Z X Z. So, N X N can be enumerated as we saw, and this can
be easily extended to Z X Z.
(Refer Slide Time: 10:10)

So, therefore, the set of rational numbers though it is dense and then it looks superficially to
be much larger than the set of integers, actually both the integers and the rational numbers
have the same number of elements which is quite surprising, but it is true.

(Refer Slide Time: 10:26)

So, for all the infinite sets we have seen are countable right. Of course, the natural

numbers are countable by definition, and then we saw integer are also countable,

and the rational are also countable. So, what about the real numbers? So, how did

we get to the real numbers? We took the rationals and then we added all these
irrational numbers like √2, π, e and so on. So, Cantor showed that R actually is not

countable. So, let us see how this proof works.

So, actually he did not, he did have a separate proof that R is not countable, but later on he
made another proof which is easier to present which starts with the different set. So, instead
of looking at R, we will look at something which looks quite different. We will look at
infinite sequences over 0, 1. So, an infinite sequence of a 0, 1 is just something like you just
keep writing down 0 or 1 infinitely many times without stopping right.

So, what Cantor argued is that this set is not something that you can count. So, supposing you
can enumerate the infinite sequences over 0, 1, then on the right to see some enumeration; we
are not looking at a particular enumeration in some particular order. We are just saying is
there any enumeration at all, so that I can write down the 0-th sequence. So, this is the 0th
sequences, this is f (0) in some sense, this is f (1), this f (2) and so on.

So, I have just written f (0) as s0, and f (1) as s1, and so on. And each sequence has positions
which I have written b for bits because these are binary digits 0 or 1. So, each sequence has
an infinite sequence of bits which characterize what it is, and no 2 rows are the same they are
all different infinite sequences of 0s and 1s right. So, hypothetically this table is an
enumeration of such sequences. So, if this is a enumeration of all such sequences, can be
derive a contradiction? So, this is how Cantor derived a contradiction.

(Refer Slide Time: 12:26)

So, he said let us take each row and reverse the bit. And which bit to be reversed? Well, if we
are in the ith row, then we reverse the ith bit. So, in the first row which is s 0, we reverse b 0 ,
in the second row. So, if you want go back, so this was 0, so we are here at 0 0 1 0. So, after
flipping, it becomes 1 1 0 1 right. So, what we are doing is in 0th row, we are flipping b0; in
row s1 we are flipping b1; in s2 we are flipping b2, and so on.

(Refer Slide Time: 13:03)

So, now this gives us a new sequence which we can read off diagonally right. The sequence
consists of the red numbers which we have got by flipping the number at the i-th position in
the i-th sequence. What can be say about this sequence? Well, first of all it is an infinite 0, 1
sequence.
(Refer Slide Time: 13:20)

But this infinite 0, 1 sequence cannot be any of the rows in my table, because by construction
if it is a row in my table it must be s j for some j, but at position j, s j has been flipped. So, this
cannot be s j because if I had s j already in my table if the sequence is already in my table, the
new sequence has the j-th bit flipped. So, diagonal sequence differs from each si at bi, and
therefore, this new sequence that I have constructed cannot be part of the enumeration.

Now, it is important that we are shown this regardless of what the enumeration looks like, we
have not made any assumption about the order in which we are enumerating. We have said
no matter what sequence you have in mind in terms of enumeration, you would have to be
able to write down the sequences one after the other table in a sequence of rows.

However you write it down, I will be able to construct this new diagonal sequence by taking
the i-th bit in the i-th row and flipping it. So, however you enumerate it, I get a new sequence
which is not part of your enumeration. Therefore, there is no possible way of enumerating 0,
1 sequences.

So, as we said this is not the question we asked, the question we asked is are the real numbers
enumerable, are real numbers countable? And what we have actually argued is that 0,
1 sequences, infinite 0, 1 sequences are not countable. So, from here how do we get to the
real number?
(Refer Slide Time: 14:38)

Well, it is one way to do this is to just think of these 0, 1 sequences as actually decimal
fraction. Now, we know that we can write things like 10.3 and 6.28 and so on. So, now, we
just restrict our self to writing in decimal fractions of the form 0 point something where
everything on the right hand side of the decimal point is either a 0 or 1.

So, here is an example right. So, this is an example of a 0, 1 sequence represented as a

decimal fraction. So, since each sequence is different, each such decimal fraction represents a
different number.

(Refer Slide Time: 15:20)

And these are all real numbers between 0 and 1, because they all have an integer part which
is 0, and then we have something which is of course, we could have exactly 0 if we have all
0s ok. So, we definitely do not have, all we do not cannot get to 1, but we can think of these
as numbers between 0 and 1.

So, each such sequence represents a different point in the interval 0 to 1. So, this is an
injective function right. So, this is an injection that is a one-to-one function from infinite
sequences 0, 1 to the interval (0, 1). Now, the interval (0, 1) is a very small fraction of the
reals.

So, what this argument tells us is that in fact even this very small fraction of the reals is not
countable because the set of underlying 0, 1 sequences not countable. So, if this even this
small fraction of the reals is cannot be enumerated, then R itself cannot be countable right.
So, this is an indirect argument saying that not saying thatR itself is not countable directly,
but saying that there is a small part of R, which is not countable. And since R is much more
than that, if the small part cannot be counted we have no hope of counting the whole thing.

(Refer Slide Time: 16:26)

So, to summarize any set that has a bijection from N is what we call a countable set. And we
showed that the set of integers in the set of rationals are countable by describing a strategy to
enumerate the sets. Now, this argument is due to Cantor which builds this diagonal sequence
called diagonalization and has been used in many other proofs involving infinity after that.
So, the proof of diagonalization by Cantor shows that the set of real numbers is not countable.
So, notice that the set of real numbers is not countable and the set of rationals is countable.
What it does to the rationals to create the real numbers? We added the irrational numbers. So,
actually the set of irrational numbers that we have added to the rationals must be itself
uncountable, because we cannot take two countable sets and add them up and get an
uncountable set. So, in other words, there are vastly more irrational numbers than there are
rational numbers that is what it tells us.

Now, one question that we could ask is, is there anything in between? So, these are sets that
we have been using intuitively. So, we have counted them. But can we construct something
for instance which is not countable, but which is smaller than the reals right? So, is there such
an infinite set?

Now, it turns out that this is a very non-trivial question. This question was actually posed
when Cantor came up with this proof in the late 1800s, and it remained a very central opened
question it was called the continuum hypothesis.

So, if you look at cardinal numbers in the finite sense, we have 1, 2, 3, 4, 5. So, we have a
kind of small jumps between them, but we have a continuous sequence of numbers. Now, we
seem to have this big jump between the real number the integers of the natural number and
the real numbers, is there something in between or is it so, is there a continuum of these
infinite numbers or these big jumps?

And this continuum hypothesis was a very important open question in set theory. And in the
1960s Paul Cohen actually showed that this is a question which cannot be proved or
disproved. So, this is what is called independent. So, this is a fact which is independent of set
theory using the axiom of the set theory, no way that you can either prove or disprove it.

So, both the fact that there is a such a set, and there is not such a set are consistent. So, these
infinite sets lead to a lot of interesting questions, some of them are quite mindboggling, and
they are quite counterintuitive. But if you are interested in these things, it is well-worth
looking into them.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 12
Rectangular Coordinate system

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, hello students, today we are going to see some elements of coordinate geometry. Now, let
us try to identify these elements as axes, points and lines. We have already seen in basic
geometry what are points, lines and planes. So, we will further study this and we will study
some algebraic properties using coordinate geometry of these particular geometric objects.

(Refer Slide Time: 00:40)

So, in that context first we need to revise our Rectangular Coordinate System; why is
rectangular coordinate system important and how we can study. Given a point on a plane;
given a point on a plane you want to describe how this plane be how this point behaves or
what is the location of this point. Now, if I want to consider this point and I want to describe
the position of the point as of now I cannot say anything more than, this point is slightly
towards the right top of the plane.

Now, if I introduce a horizontal line over here, then I can say the point is in the upper half of
the plane. This gives a slightly better visibility to the point or slightly better description of a
point. Now, if I consider a real number system associated with this line then I can say the
point lies in 0 to 5, if I plot two perpendicular lines between 0 to 5 then I will get this point.

This is much better. Now, these perpendicular lines can also be replaced with one
perpendicular line which is this which has a real number system associated with it. Now,
when a real number system is associated with this point, then what you can actually see is if I
can consider this, this particular structure or this particular square which is enclosed within 5
on the vertical line and 5 on the horizontal line; I am giving a much better description of a
point.

Then I can enhance this further by putting up the grid lines. These grid lines now typically in
this case locate the exact location of the point. So, what is the exact location of the point over
here? If you look at this exact location of the point is on the horizontal line if you travel 3
units in one direction, horizontal direction and 4 units in the vertical direction then you will
reach this point.

So, I can also name this point as in the horizontal direction I have to travel 3 units and in the
vertical direction I have to travel 4 units. So, I can name this point as 3 comma 4 that will be
a precise description of this point. So, in turn what we have seen just now is a reference
system through which we are able to specify the location of a point in a specific manner. Let
us analyze this reference system that we have introduced.

Now, in horizontal direction I have to travel 3 units and in vertical direction I have to travel 4
units; that means, I am actually specifying the coordinates in X direction and coordinates in
vertical direction. So, in particular these horizontal directions and vertical directions are
called X axis and Y axis respectively.

So, if you look at this horizontal direction, you can see the vertical line cuts the horizontal
line into two parts; positive part of X axis and negative part of X axis. Similarly, the vertical
line is cut by the horizontal line into two parts. On the upper side we have a positive part of Y
axis and on the lower side we have a negative part of Y axis.

So, this is a typical structure which is called coordinate plane ok. Now, let us come to the
nomenclature of this particular coordinate plane. As I mentioned if I am travelling 3 units in
horizontal direction; I will call that as X coordinate and if I am travelling 4 units in vertical
direction, I will call that as Y coordinate. Hence, the name coordinates.

These two lines X axis and Y axis meet each other at a 90 degrees angle; that means, both the
lines are perpendicular to each other. Therefore, the name rectangular; recta means right in
Latin so, rectangular means 90 degrees coordinate system; that means, a rectangular
coordinate system. So, let us revise what we have studied just now in words.
The horizontal line is called X axis, it allows you to move from left to right. The vertical line
is called Y axis which allows the movement up and down, then there comes a point of
intersection of these two axes which is called origin. The point of intersection of these two
axes is called origin and if you look at the coordinates of these, then any point on this
particular plane can be denoted by a ordered pair (x, y).

You can see one blue point is also popping up now. Now, how to describe a point using a
coordinate plane? So, for example, given a point (3, 4) how will I locate this point? So, if you
look at this (3, 4), we have already seen how to locate it. We have travelled 3 units in
horizontal direction and 4 units in vertical direction therefore, (3, 4).

Now, suppose you are given another point which is (-5, 2), then this x coordinate
corresponding x coordinate is negative; that means, I have to go to the left of the vertical line.
That means, I have to travel here a 5 units distance which is - 5 and on the positive side of Y
axis I have to travel that is up upper up upper half divided by X axis I have to travel 2 units
which will give me the point (-5 ,2).

So, this is how we can uniquely describe points using coordinate plane. Now, when I was
when we were studying these two points (3, 4) and (-5, 2), you can easily see with respect to
this coordinate axes you can have 4 parts of the coordinate plane.

(Refer Slide Time: 07:24)

Let us study those parts in detail in the next slide. So, next slide is this coordinate plane.
Now, I have identified 4 points in all 4 parts of the coordinate plane. So, you can see the first
point P which lies in the positive side of X axis and positive side of Y axis has positive x and
y coordinates which is given by quadrant I. So, any point in this plane, in this particular
quarter will have positive X and positive Y axis.

Now, in general as a mathematical psychology we move in a anti-clockwise direction. So,

now, I can move in a anti-clockwise direction to the next one fourth part, next quarter of the
coordinate plane. And, see that my X axis has negative values and my Y axis has positive
values. All points which have this form of values are called points on the second quadrant or
the quadrant the one the quarter of this particular coordinate plane is called quadrant II.

Next we come in a anti-clockwise direction to the third side that is this. So, if you look at the
point R which is lies in this particular quadrant is (-3, -4); that means, the x value is negative
and the y value is negative. Therefore, (-3,-4) is a point which lies in quadrant III.

Remember it is easy to remember this that quadrant I and quadrant II, quadrant III that is odd
quadrants have same parity of x and y coordinates. And, quadrant II and quadrant IV have
opposite parity of x and y coordinates. So, let us go to quadrant IV, you can see a point S lies
in quadrant IV which has coordinates 4 and -3. Now, this 4 and -3 which denotes x
coordinate is positive and y coordinate is negative such a classification comes in quadrant IV.

So, this is how a coordinate plane is come split into four quadrants. Now, a question may
arise in your mind; suppose I have this point which is (5, 0). Now, in which quadrant this
point lie? The answer is this point does not lie in any of the quadrants. This point lies on the
X axis. Similar question can be asked for a point (0, 5). The point does not lie on any
quadrant, but lies on the Y axis.

So, based on this particular understanding, a coordinate plane its subdivided into first is four
quadrants, two are axes. Let us try to see what are the typical features of the quadrants and
these axes. Quadrant I, you will have x and y coordinates which are positive. Quadrant II, you
will have x coordinate which is negative y coordinate which is positive. Quadrant III, you
will have both negative values. Remember odd quadrants will have same parity that is
quadrant I is positive, quadrant III is negative.
Now, quadrant IV will have positive and negative, x coordinate which is positive, y
coordinate which is negative. Then comes the split into axes. So, on the X axis you will have
points which can either take positive values or negative values for x coordinates and 0 for y
coordinate. On the Y axis you will have points which can take positive and negative values
for y coordinates, but 0 for x coordinate. Now, there remains only one point which is the
point of intersection which is identified as origin ok.

So, this completes our understanding of the coordinate system. Why quadrants, quadrant
system is helpful? Sometimes you have been given several points to plot. Now, those points
if you look at them closely, you need not have to divide the system in a equally distance
manner, like this manner. You may have many points in quadrant I, in that case you can scale
this, you can bring this to the bottom right corner, bottom left corner and just focus on
quadrant I.

So, if you have a good understanding of quadrants, you may be able to graph the functions
better, graph the points better; that is why the coordinate system is important. This ends our
discussion on coordinate system.
 Mathematics for Data Science-1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 13
Distance formula

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:14)

So, after coordinate system, let us try to identify one classical problem.
(Refer Slide Time: 00:20)

That is if I have a point and somebody ask me a point is located here; let us say point is (3,4).
And somebody ask me what is the distance of this point from the origin? So, in this particular
slide our goal is to find the distance of a point P which is (3,4) from the origin.

That essentially reduces to finding the length of this line segment which is joining points O
and P. So, is there any classical tool that is of my help? Suppose, now if this point is either
lying on X axis or Y axis, let us say if this point is say (3,0) ok. If this is the point that is of
interest to me; do I know how to find the distance of this point? The answer is yes I know, I
just need to calculate the units that are in horizontal direction.

Suppose the point is on Y axis, then do I know how to calculate the distance of this particular
point from Y axis? The answer is again yes I know, I just need to calculate the number of
units that I need to travel to reach this point. So, if the point lies on X axis and Y axis, I know
how to calculate the distance of a point. Now, if the point is lying anywhere in the coordinate
plane, how to find a distance is a question.

For that, let us try to understand the situation, that if I know if somehow I can understand this
with respect to this coordinate axis. This particular position with respect to these coordinate
axis then I will be able to give the answer to find the distance between the two points. So, let
us try to do one thing that is let us try to get the image of this point (3, 4) on X axis.
So, how will I get the image of this point (3,4) onto the X axis? The easiest way is you drop a
perpendicular on X axis, that intersects the X axis at point (3,0). Once this is done then you
can actually drop a perpendicular and see that it forms a right angled triangle with X axis in
place and a vertical line in place; you have a right angled triangle. Do you know any theorem
in our conventional geometry that relates this particular structure?

You know Pythagoras Theorem or Pythagorean Theorem, that relates this particular structure.
In a right angled triangle the hypotenuse length of the hypotenuse is given by square root of
its adjacent sides; square root of squares of the lengths of the adjacent sides. So, we will try to
use this for finding the distance of a point from the origin. So, by the Pythagorean Theorem, I
know OP2is actually equal to OQ2 +QP 2.

Now, the exercise that we did orally just before starting this problem will help us to
understand what is OQ2. So, what is OQ? OQ is a part of X axis, OQ is a line segment which
is a part of X axis. What is the length of OQ? We have already discussed that, that length is 3
units. Similarly, if you look at QP; what is QP? QP is parallel to Y axis. So, it is as good as
projection of Y axis projection onto Y axis.

So, what is the length of this particular line segment which is QP? That is 4 units; so I know
the length of OQ and I know the length of QP. Therefore, by Pythagorean Theorem, I know
the length of OP. So, what will be the length of OP? It will be √ ❑. So, 32is 9, 4 2is 16
therefore, this will give me 25; 16+9 and positive square root of it will give me number 5.

Now, has it anything special to do with point (3,4) or can I generalize this? The answer is yes,
it has nothing special to do with point (3,4). I could have started with point P which is (x , y)
and then projected this onto X axis or I figured out the image onto X axis which will be (x ,
0). And therefore, the length of OQ will be x and the length of QP will be from 0 to y units;
that means, y units.

So, length of QP will be y units and therefore, the formula OP = √ ❑ would have been
possible. So, let us try to take this particular example and try to generalize this problem to
finding the distance between any two points.
(Refer Slide Time: 06:02)

So, distance between any two points. So, again the setup is pretty common. Our goal is to
find the distance between any two points P ( x ¿ ¿1 , y1 )¿ and R ( x ¿ ¿2 , y 2) ¿. How will you
find the distance between any two points? Let us see the points on the graph, then the things
will be more specific. My ( x ¿ ¿1 , y1 )¿ is (5, 6) and ( x ¿ ¿2 , y 2) ¿ is (-1 , 2).

Now, if I look at these two points, I want to find the distance between these two points. So,
once easy way to find a distance between these two points is to construct a right angle
triangle. But, now because the point is not located on X axis, this (-1 , 2) is not located on
any of the axis; I cannot say drop a perpendicular to X axis.

So, the actual way that I should do here is I will drop a perpendicular to X axis which will
intersect at (5 , 0). And, then to this line I will drop a perpendicular from the point R minus
(1,2) and which will intersect this, this particular line which will be the perpendicular to X
axis, at where the y coordinate will be 2 and x coordinate will be 5. So, this point will be (5,2)
and then I will get a right angled triangle.

By skipping these steps, we can straight away say that you construct a right angled triangle
with a right angle at point Q which is ( x ¿ ¿1 , y 2)¿. Just relate this( x ¿ ¿1 , y 2)¿, if you use this
terminology is the point (5,2). So, you can draw a right angle triangle using (5,2) ok.

So, this way we need not have to specify steps that you have to draw two perpendiculars and
all; because the point may as well lie in the third quadrant. And, in that case dropping
perpendicular to X axis may not help, you have to extend the perpendicular beyond X axis.
So, it is always better to consider this kind of structure, that is construct a right angled
triangle with right angle at point Q which is ( x ¿ ¿1 , y 2)¿ .

Then it does not matter where the point actually lies. Now, once the right angle triangle is in
place, the same theory that we used Pythagorean Theorem will come into play. And, by
Pythagorean Theorem if I want to find the length of PR, I know PR 2=QR2 + PQ 2. Can I
calculate the length of QR and PQ? The answer is yes I can calculate, because, the line
segment QR is actually parallel to X axis and the line segment PQ is parallel to Y axis.

Therefore, this is as good as computing the length on X axis and this is as good as computing
the length on Y axis; hence what we will get is. So, how to compute the length? It is basically
the change in x coordinates. So, how far the x coordinates have changed? So, while
computing the length parallel to X axis always remember you should go from left to right,
that is when you are subtracting you should take the highest value first that is 5 – (-1).

So, the length of this will be 6 units and while subtracting or while finding the length in a
vertical direction go from bottom to up. That means, you subtract the value that is highest in
Y direction to the value that is lowest in Y direction. So, here 6 - 2 will give me 4 and in the
X direction 5 – (-1) will give me 6 units. So, this is how we will calculate the length of these
two line segments.

And therefore, I can easily find the length of PR; while calculating the length because we are
in this particular case, we are considering squares. It does not matter whether you consider x 1
first or x2 first because, anyway we are squaring even if you get the negative value, you will
be squaring it.

So, in particular in this case where the coordinates are ( x ¿ ¿1 , y1 )¿ and ( x ¿ ¿2 , y 2) ¿, I will

take ( x ¿ ¿2−x 1)2 ¿; does not matter which one is bigger. And ( y ¿ ¿ 2− y1 )2 ¿ , does not matter
which one is bigger.

And I will take a positive square root of it. Therefore, my length PR for this particular
example will be 6 2+ 42; 62 =¿ 36, 42=¿ 16 together they will give 52 is 2 √ ❑. So now, we
have established a general formula which is called distance formula for finding the distance
between any two points on a coordinate plane.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 14
Section formula

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

Now, let us take up the next concept. Now, we have handled two points; now let us take three
points.

(Refer Slide Time: 00:22)

And, let us say those three points lie on a line and given that the point P cuts the line segment
AB in the ratio m : n. Our goal is to know the coordinates of point P; this will give us the
Section Formula. So, this is the graphical representation of the points. So, there are two, there
is a line segment AB and point P cuts this line segment in the ratio m : n.
How will you find the coordinates of point P? This is the question; let us bring in our
coordinate system. So, let the coordinates of A and B are ¿, y 1 ¿ and ¿, y 2 ¿, the coordinates of
A are ¿, y 1 ¿ coordinates of B are¿, y 2 ¿,. I do not know what P is, let us assume it has some
coordinates which are x and y ok. So, let us bring in them in the coordinate system which is
this.

(Refer Slide Time: 01:29)

Let us try to understand this particular coordinate system by putting up some triangles
around. So, what I have done is I have actually constructed two triangles using the same logic
that I used in the distance formula. If the coordinates of point P are (x,y), then I will construct
a right angled triangle in this direction; where the x coordinate will be x and y coordinate will
be the coordinate of y coordinate of A that is y 1.

Similarly, I will do the same thing with respect to point B; I will drop a perpendicular which
will meet at this particular point. So, basically I will drop a perpendicular which will meet the
X axis and again I will draw a perpendicular here. But, let us for sake of simplicity we have
constructed a right angle triangle, where the y coordinate of this point will be y and the x
coordinate of this point will be the x coordinate of point B which will be x2 , ( x2 , y ).
With this understanding we can proceed further and see that the triangles, these two triangles
are similar to each other. How? First of all let us see this line is parallel to X axis and this line
is parallel to X axis as well. Therefore, these two are parallel lines and this is a transversal
that is passing through these two parallel lines. Therefore, these two angles the angle A and
angle P will be same or equal.

Next these two are right angles, then we know the sum of the angles in a triangle is 180
degrees, therefore this angle, angle B must be equal to angle P. Therefore, triangle AQP must
be similar to triangle PRB by angle test that essentially means I have their sides in some ratio,
correct. So, for simplicity I have plotted these points with some coordinate references.

So, this is A is (2,2), B is (8,8); then whatever I mentioned the coordinates of Q are (x,2) and

AP
coordinates of R are (8,y). So, now these two things will be in some ratio that is , these
PB

AQ QP
are the hypotenuse of these two right angle triangles is equal to and this thing is right
PR RB

AP AQ PQ
or you can see is equal to which is equal to .
PB PR BR

AP
Now, I already know have a ratio m : n. So, their ratio is m by n that is already known to
PB
us, that is given to me. Now, can I calculate the length of AQ and PR? The answer is yes,
because AQ is parallel to x axis. It is just subtracting the highest x coordinate from the low.

So, it will be x -2 in the figure and in our theory it is x - x 1. Similarly, you can look at PR; it
will be 8 - x or in our theory it will be x2 - x. For y axis or the lines that are parallel to y axis
PQ and BR you can see you will go to the highest value that is y - 2 or y - y 1and the other one
BR will have y 2− y right.

m x−x1 y− y1
So, together I will have a representation of this form: = = . Now, take one
n x 2−x y2 − y

m
equality at a time; that means, is equal to let us say these the consider these x coordinates.
n
So, we will just cross multiply them, rearrange them you will get what x is equal to.
 m
In a similar manner just take is equal to this, these y coordinates ratio and then cross
n
multiply and rearrange them. You will get the following values which are given by

m x2 + n x1 m y2 +n y 1
x= . And, similarly y= . This gives me the section formula, when a point
m+ n m+n
divides the line in the ratio m : n.

Another interesting question is suppose I know the coordinates of x and y; can I find in what
ratio the line divides? Obviously, yes because you know the coordinates of the line, you just
need to use this formula for finding the ratio ok; that will be more clear when you solve more
problems ok.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 15
Area of triangle

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:15)

After section formula let us try to understand the three points when they are not on one line
that is when they form a triangle. So, you have been given three points, and you know they
are not collinear points and therefore, they will form a triangle. And the question can be how
to find the area of triangle using the coordinate system.

So, let us try to see that using the coordinate system. So, there is some triangle ABC and I
want to find the area of triangle ABC. Let the coordinates of that triangle be ( x1 , y 1 ),( x2 , y2 )
and ( x3 , y 3 ). Once I have these coordinates, I can plot it here. You can see on the right there
is an image of a triangle.

Now, how to find the area of this triangle? Now, whatever I discussed so far everything
actually relied on dropping a perpendicular to X - axis and finding the area of the geometric
object that is formed. In earlier cases, it was just a triangle. Now, if we follow that theory
then you can easily see that I need to do something like dropping a perpendicular to x axis.
So, I have dropped perpendiculars to X - axis.

(Refer Slide Time: 01:35)

Now, I have generated some figures. What are the figures that I have generated? In particular,
I have generated 3 trapeziums, trapezium ADFC that is the biggest one encompassing
everything. Then, you can look at trapezium ADEB, then you can look at the trapezium
BEFC.

Now, my triangle is trapped in between these trapeziums. So, let us try to make our
understanding crystal clear. If I want to find the area of triangle ABC, then I need to first
consider the biggest possible quadrilateral or trapezium that is ADFC and eliminate the areas
of two smaller trapeziums that is ADEB and BEFC. And whatever I am left with is the area
of triangle ABC.

Now, do I know how to find the area of trapezium? Yes, I know. The formula is half times
sum of parallel sides into the height of the trapezium. So, we need to quantify how will these
quantities be calculated? Let us consider trapezium ADFC, if I consider a trapezium the
ADFC then what are the parallel sides of this trapezium? Side AD and side FC.

So, I will take average of these two parallel sides that is half of AD plus FC. Then, what is a
height? Height should have a perpendicular distance, so that is X - axis. So, I know the
distance will be DF.

So, let us take the general coordinate system rather than using this coordinate system. What
are the coordinates of A and D? So, A has coordinates ( x1 , y 1 )and after dropping a
perpendicular on X – axis the y coordinate will vanish and therefore, the coordinate of D will
be ( x1 ,0). So, what will be the length of AD? It will be purely in terms of y that is y 1 .

Similar, thing is applicable for CF. So, it will be nothing but y 3; so, area of ADFC,

1 1
A rea ( ADFC ) = ( AD + FC ) × DF= ( y1 + y 3 ) × DF .
2 2

Now, what is the length of line segment DF or FD? Highest minus the lowest. So, in this case
our F is (8 ,0) or ( x3 , 0 )and the point D is ( x1 ,0). So, it is ( x3 −x 1 ).

1 1
A rea ( ADFC ) = ( AD + FC ) × DF= ( y1 + y 3 ) ×(x 3− x1 ) .
2 2

In a similar manner, I can actually see a smaller trapezium that is ADEB, smaller
quadrilateral that is ADEB and the height of that quadrilateral will be the length of ED which
is 2 in this case or x2 −x1in the coordinate system. So, this is what our understanding of the
length is. In a similar manner, the sum of lengths of parallel sides is y 1 + y 2.

1 1
A rea ( AD EB ) = ( AD + EB ) × D E= ( y 1 + y2 ) ×( x2 −x1 ) .
2 2

In a similar manner you can compute BEFC.

 1 1
A rea ( BEFC ) = ( BE +CF ) × E F= ( y2 + y3 ) ×( x3 −x2 ) .
2 2

Now, using this you can compute the area of the triangle which can be easily seen to be in
this form. So, I have just taken this example and computed these values. So, the values are
effectively in this particular case the length of AD was 2.

The length of CF was 3 so

1 1 1
Area ( abc )= ( 2+3 ) × 4− ( 2+1 ) ×2− ( 1+3 ) × 2=10−3−4=3 square units .
2 2 2

Now, if you look at this particular thing and rewrite this expression you will get a very nice
expression. You can juggle with this expression and try to simplify it by taking a cross
products and you will come up with the expression of this form.

1
A ( ∆ ABC )= ∨x1 ( y2 − y3 ) + x2 ( y3 − y 1 ) + x 3 ( y1 − y 2 )∨¿
2

The absolute sign is just to ensure that the area value should not be negative, but the
calculation still remains same. And you need to consider one caution here that all the vertices
of a triangle in an anticlockwise direction then only this formula is valid. So, I have
considered area of a triangle.

Now what we have seen so far is given two points how to find the distance between two
points, given three points if they are collinear, we have found the section formula that can
help us to find their ratios or the coordinates of the middle point. Now, if the points are non-
collinear, we have seen how to compute the area of the triangle using the coordinate system.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 16
Slope of a Line

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:14)

So, after looking at area of the triangle using coordinates, let us now focus our attention to
again a two-point system and one-dimensional objects that is a line.
(Refer Slide Time: 00:24)

We have already seen in our basic classes that two points uniquely determine line. Now, if I
want to characterize a line, and if I give you two points, I should be able to find a line passing
through these two points. How is the geometric object algebraically related to the coordinate
geometry? That is what we want to explore now, to explore that I need a concept of a slope of
a line.

So, what essentially is the slope of a line? In a vague manner, what we understand by slope of
a line? If you look at this coordinate plane which is displayed here. If I am moving some
units in x directions; the question can be asked with respect to this change in x direction what
is the corresponding change in y direction.

So, if I want to answer that question then I need to consider a ratio of change in y direction to
change in x direction; some people call it as rise by run ratio, run is in the horizontal
direction, rise is in the vertical direction. So, you can consider slope of a line as a rise by run
ratio. So, let us try to make this work concept clearer by showing some examples.

So, now, here is a line with two points given onto it. Again, our standard conventional
method we will construct a right-angled triangle using these two points. Now, the question
that I posed is what is a rise by a run can be answered over here. For example, you look at
this right-angled triangle, what is happening? This is the movement of a line in moving from
one point to another point in y direction, this vertical length is the direction, is the movement
of a line in moving from one point to other point in y direction and this horizontal line is a
movement in x direction while moving from point A to B on a line.

So, essentially what I need to capture is the change in y direction that is from point ( 4 ,−2 ) to
point ( 4 , 4 ), that is −6 and moving in x direction from (−2 ,−2) to ( 4 ,−2 ) that means, −6
here also. So, the slope of a line can be equal to 1. This we can make it more precise by
giving some formal definitions.

So, if I want to find the slope of a line given the coordinate plane, I can always identify these
two points as ( x1 , y 1 ) and ( x2 , y 2 ). I will construct a right-angled triangle which intersects the
point at( x 2 , y 1 ). And once I constructed, as I mentioned you know what is the change in x
direction and what is the corresponding change in y direction, therefore, you can actually
compute the ratio of this. But while computing the ratio, you can also think remember some
concept from trigonometry.

For example, when I constructed this right-angle triangle there is some angle formed over
here this arc denotes that angle. Let us call that angle as theta. Now, what I am saying is
change in y upon change in x, but can you relate some quantity related to this trigonometric
ratio that is tan of θ, right. So, what I can say is my m or the slope of a line is MB by AM

y1 − y 2
which is and which is also equal totan θ.
x1 −x2

So, I have defined one thing that is m which is the ratio of these two, but which in turn turned
out to be equal totan θ. So, if it is tan θ, see here it does not matter whether I take
y 1− y2 ∨ y2 − y 1whatever I am doing I should do synonymously. For example, if I have taken
y 2− y1then I should take x2 −x1or if I have taken y 1− y2 then I should take x1 −x2.

So, it does not matter which order you are swapping because finally you are taking the ratio
so whatever you are doing you do it asynchronously, so that there will not be any confusion.
So,m=tan θ. Now, I have introduced two terminologies here m and θ. So, let us define them
properly. This m is called slope of a line, which is the topic of this discussion. And then this θ
is called inclination of a line with respect to positive X - axis measured in an anti-clockwise
direction.
Now, somebody may say I have drawn this angle over here, but if you look at this particular
line, this line is parallel to X - axis. And this line is intersecting X - axis here, that means
even if I consider this angle, this angle also will be θfrom the basics of geometry, correct.

So, now the question can be asked how far the θ can go? So, to answer that question let us try
to see if I am considering a θ then θ can be equal to 0, θ equal to 90 degrees tan is not
defined. As you can see tan of 90 is not defined, but it can go up to 180 degrees. So, the
variation of θ allowed is 0 to 180 degrees.

(Refer Slide Time: 07:42)

So, now let us have a look at the salient features of the slope of a line. In particular, let us see
if the line is parallel to X - axis the angle of inclination is 0 degrees; therefore, the slope of a
line should be 0. Now, if the angle is 90 degrees; that means, 90 degrees with respect to X -
axis; that means, eventually I am on Y - axis or in fact, I am on Y - axis in such case tan 90 is
undefined, right. Therefore, slope is undefined.

As you can see if I have an angle which is 90 degrees that is Y - axis; that means, x is equal
to constant is the equation of the line. And you cannot have any movement in y direction or
you can have infinite movement in y direction without any change in x direction. That itself
creates a problem therefore, the slope is undefined for theta is equal to 90 degrees or the
inclination is equal to 90 degrees.
So, with respect to inclination there is another definition of slope. If theta is the inclination of
a line l then tan theta is called slope or gradient of the line. This is the second definition of
our slope of a line which matches exactly with the original definition, but there will be some
glitch, there may be some confusion, ambiguity.

So, let us try to resolve that ambiguity because this theta is the angle made with respect to
positive X - axis. And theta not equal to 90 degrees I can define m=tan θ. That is perfectly
fine and it is well-defined over there whenever it is not equal to 90 degrees. What is the
ambiguity? The ambiguity can be shown in the figure. For example, now what is θover here?
θ over here is actually this particular angle.

Now, if I you look at this particular angle which is θ you can see that this is an obtuse angle.
Now, how to evaluate a tan of this angle? We already know some methods, but will that
contradict with our definition of slope. That is the question. So, if I use the rise by run
formula or the change in y to up on change in x formula, how will I figure out the slope? So,
the answer is I will simply drop a perpendicular or I will construct a right-angle triangle with
right angle at point M which is (−4 ,−2).

In that case, I will be interested in this angle that is angle at A in our older definition or this
angle is essentially equal to 180−θ. So, let us go further. This angle is equal to this angle.
What is the measurement of this angle? It is180−θ. That means, if I want to find a slope

δy
according to our definition that is or change in y by change in x, then I need to consider
δx
the angle of this particular structure that is tan(180−θ). So, m=tan(180−θ).

Now, what is tan(180−θ)? If you use simple trigonometric formula you will get tan(180−θ)
is nothing, but −tan θ. But what is −tan θ? You can easily see what is −tan θwhich will be

y1 − y 2
. So, in short, our formula for slope is consistent no matter which definition we use,
x1 −x2
therefore a slope of a line is uniquely determined given a line.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 17
Parallel and perpendicular lines

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:15)

Now the question can be asked that if a line is given to me, I can uniquely determine the
slope, but if a slope is given to me can I uniquely determine a line? That is the next question
that I will put up. In any sense the question asks can there be many lines with same slope?
The answer can be seen in this GIF image.

If you look at this image closely what we have done is? We have fixed one line and we know
how to compute the slope of this line we have a it will be minus 1 based on the coordinates.
Now, the blue line that is revolving around is actually having the same inclination as the
orange line.

Now, the orange line and blue line have the same inclination; that means, tan of those
inclinations will be same, will match and hence there can be infinitely many parallel lines
which have a same slope. So, the answer to this question, can slope of a line uniquely
determine a line? The answer is no, you cannot uniquely determine a line given the slope of a
line or the inclination of the line.

Now, why do we study the concept of slope or whatever we studied how it is helpful? The
helpfulness of this concept is just what we discussed in this graphical image, what we are
seeing is if the inclinations are same the line better be parallel. So, for parallel lines I can use
this concept and derive a condition of slope. Similarly, I can do by rotating them by 90
degrees; that means, I we can consider the perpendicular lines and I can consider general two
lines intersecting each other and see what condition I can derive based on the slope.

So, I want to explore the usefulness of slope. So, to explore this I will first figure out the
condition for parallel lines and I will figure out the condition for perpendicular lines, in due
course we will find the relation between slopes of two lines and their intersection and their
angles of intersection. This is what we will do in next few minutes.
(Refer Slide Time: 03:15)

So, let us go to the next characterization of parallel lines via slope. Now as you can see in this
image there are two parallel lines, they have same inclination, but they are not unique that is
what we figured out. So, if I play this video you can see again, this is similar to what we have
seen in the last video.

So, I have something which is moving around and there can be infinitely many lines, what
remains constant is the inclination, the inclination is same if I have parallel lines. So, let us
try to see whether we you can derive something. So, let to put it in a proper context.

Let orange line be l1and the blue line be l2 be two non-vertical lines. Why non-vertical lines?
Vertical lines have angle of 90 degrees for which the concept of slope is undefined,
inclination 90 degrees for which the concept of slope is undefined. So, what I need is non-
vertical lines. So, considered two non-vertical lines with slopes m 1∧m 2given the slopes their
inclinations αand βrespectively.

Now, if you have been given that l1is parallel to l2then α =β, inclinations are same that is
what we have seen in the figure and that is what we discussed in the last slide also. So, if
α =β then naturally tan α =tan β , once tan α =tan β ; what is tan α ? It is the slope of line11
that is m 1 and tan βis the slope of line l2which is m 2. Therefore, clearly the slopes are equal,
m 1=m2.
The converse that is assumed that, if the slopes are equal thentan α =tan βby a definition.
Now, tan α =tan βdoes that imply α is equal to β? In our case because we are restricting the
inclinations to vary from 0 to 180 degrees the value of tan is uniquely determined. And
therefore, because α ∧βlie in 0 to 180 degrees α =β which resolves the problem; that means,
their inclinations are same. That means the two lines are parallel. So, l1 is parallel to l2.

So, what is a characterization of parallel lines? That means, if I want to say two non-vertical
lines l1 and l2 are parallel then it suffices to check whether their slopes are equal or not. If
they are parallel then the slopes better be equal and if the slopes are equal then we have
parallel lines. Now similar characterization we are searching for in perpendicular lines. So, let
us go and try to figure out this characterization for perpendicular lines.

(Refer Slide Time: 06:42)

Let us try to visualize, what are the perpendicular lines? So, here are two perpendicular lines
one l1and l2let us take the orange line as l1 and blue line as l2. So, l1 will have slope m 1, l2 will
have slope m 2 angle of inclination of l1 is α then inclination of β, if it is perpendicular to line
l1is 90+ α which is β . And, then you may play with the tangent function of it and you can get
something which is very interesting.

So, let us try to figure out what is that interesting thing that we are getting. So, to put it
formally let l1 and l2 be two non-vertical lines because I cannot work with vertical lines θ
equal to 90 degrees, the concept of slope is not defined which slopes m 1, m 2 inclinations α
and β respectively, no problem in this. If l1 is perpendicular to l2 as is the case in this a figure
I have β is equal to90+ α.

So, if I want to figure out the relation between the slopes of 1 1 and l2 then it is a good idea to
take tangent of β. So, let us take that. So, tan β=tan(90+α ) , but tan(90+ α ) if you use that

−1
simple formula that is available to you is −cot α which also can be written as .
tan α

But what is tan α? tan β is the slope of a line l2 which is m 2 and tan αis the slope of a line l1

−1
which is m 1. So, what we have just now derived is m 2= or m 1 m2 =−1. That means, if you
m1
take two slopes if you take slopes of two lines take a product of them and if you get the
quantity to be equal to −1; that means, you have got a perpendicular line.

But right now, we have not proved that result, what we have proved just now is if l1 is
perpendicular to l2 then the product of the slopes better be -1. Now I want to prove if the
product of the slopes is -1 then the lines are perpendicular, how will I go about this? Exactly
the way we went for parallel lines.

−1
So, m 1 m2 is−1 then I; obviously, tan α tan β=−1; that means, tan βwill be equal to or
tan α
tan α =−cot βbut what is −cot β ? tan(90+ β )or either it will be this way or it will be the
other way so, tan(90−β). So, −cot β is either tan(90+ β ) or tan(90−β), in any case the
difference between α and β is 90 degrees.

Therefore, l1 is perpendicular to l2. Hence, we have proved a characterization that if two non-
vertical lines are perpendicular to each other, the product of their slopes is equal to -1 which
can be written in this form. Two non-vertical lines l1 and l2 are perpendicular if and only if
m 1 m2 =−1 or you can verbally write product of their slopes is equal to - 1.

So, this is the characterization of the perpendicular lines via slope. So, what we have seen so
far is the characterization of parallel lines by slope and characterization of perpendicular lines
via slope, what if they are not parallel or perpendicular and they intersect just like that? If
they are not parallel then they better intersect each other.
(Refer Slide Time: 11:37)

So, in general if I want to have an intersection of two lines and I know the slopes of those two
lines. Can I talk about the angle of intersection of these two lines? The answer is yes. So, here
is the relation of angles between the two lines and their slopes. So, what I want to say if once
I show the figure it will be clear.

As of now let us understand I have two non-vertical lines with slopes m 1 and m 2, inclinations
α 1and α 2 respectively. And, l1 and l2 intersect each other, they are not parallel so they will
intersect somehow and they are not perpendicular also. So, they intersect in angles ϕ and θ
are the adjacent angles that are formed by l1 and l2, if they intersect in a perpendicular manner
the adjacent angles will be 90 degrees each. So, that is not an interesting case because we
have resolved that case.

So, now, if they intersect at any angle then this figure will look like this; let us first
understand this figure. So, there are two lines l1 and l2. So, l1 has angle of inclinationα 1, l2 has
inclination α 2 these two lines intersect over here near y coordinate ϕand they have two
angles; one is θ, another one is ϕ.

So, these two angles are adjacent angles. What can you say about the angle θ that is formed?
As you can see the angle α 2 is obtuse and α 1 is slight acute. So, the angle θ is actually α 2
minus α 1 provided α 1 and α 2are not equal to 90 degrees. Why? Because I cannot consider
vertical lines as simple as that. So, the angle is 90 not equal to 90 degrees, θ=α 2−α 1.
So, if I want to talk in terms of slopes of these lines, I better take tangent function and apply it
to the angle θ. So, let me do it. So tan θ=tan (α 2−α 1 ). Take a standard trigonometric formula

tan α 2 −tan α 1
of tan(α 2 −α 1), you will get . But what is tan α 2? tan α 2is nothing but the slope
1+ tan α 1 tan α 2
of line l2 which is m 2 and tan α 1 it is slope of line l1 which is m 1.

m2 −m 1
Therefore, the answer to this is . So, I know what is tan θ, now you can look at the
1+m 1 m2
angle ϕ which is 180−θ. So, I can similarly derive a relationship for tan ϕwhich is
tan(180−θ), we have already seen, this is −tan θ .So, that m 2−m1 will be swapped to m 1−m2
denominator remains the same, the condition m 1 m2 ≠−1remains the same because they
should not be perpendicular.

In this case we have figured out what is the relation of tan of that angle with respect to the
slopes of the lines. So, this finishes our discussion on two lines. Now another interesting
question that comes is, what if the three points are collinear, then how will the slopes be
interpreted? Imagine three points are collinear then what happens is their slopes must be
equal because they are all lying on the same line right and there is one common point.

So, if A, B, C are collinear slope of AB is equal to slope of BC and therefore, all of them
must be collinear. So, if there is any common point in between those three points the slopes
are equal, the points are collinear, that is called the relation of collinearity using slopes.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 18
Representation of a Line-1

(Refer Slide Time: 00:06)

(Refer Slide Time: 00:14)

So, what we have seen so far is, what is a relation of the slope with respect to line and we
have exploited certain how we can use the slope to determine whether the lines are parallel
perpendicular. And, if I know the slope of the line then how will I find a slope of two non -
vertical lines, then how will I find the relation between angles and other properties.

(Refer Slide Time: 00:41)

Now, we will come to the Representation of a Line, as we have already seen slope cannot
represent a line uniquely. So, what is it that, that is required for representing a line uniquely?
So, this raises two questions, how to represent a line uniquely? And the second question is,
given any point of how will you decide whether that point lies on the line or not?

So, in order to answer these two questions, let us take the first question first and rephrase it.
So, if I want to represent a line uniquely, then what I need to figure out is, I need to figure out
a condition or a definite expression which will describe the line in terms of its coordinate
plane. So, for a given line l I should be able to find a definite condition or expression which
describes the line in terms of coordinate plane. That is in terms of the coordinates or to be
more precise what should be the condition on the coordinates in order to describe the line l.

If I can understand what is this condition then the second question is automatically answered
because if the coordinates of P are given to me and they satisfy the condition or expression
for the line l then they must lie on the line l otherwise they do not lie on the line l, then it is
just a simple job of checking whether that condition is satisfied or not. So, with this in mind
we will try to answer the first question that is how to represent a line uniquely?
Now, what kind of lines we have seen so far? We have seen lines which are similar to X -
axis, lines which are similar to Y - axis; those are typically horizontal and vertical lines.

(Refer Slide Time: 02:37)

Let us first understand, what is a horizontal line. So, a line is said to be a horizontal line if it
is of this form, now this line can be infinitely many. So, you can have infinitely many
horizontal lines as can be seen from the video. Now, how to represent this line uniquely is my
question. So, let us say I need to find this line or the condition for this line, how can I find the
condition for this line?

So, let us first define this line as a horizontal line and let us say horizontal line is a horizontal
line if and only if it is parallel to X - axis, this is our definition of a horizontal line. Now, if I
want to specify this line uniquely what do I need to know? I need to know the distance of this
line or the location of this line from X - axis, that is I need to know the y coordinate of this
line you can see here. So, I want to locate this line or the value that it takes on Y - axis if I
want to specify this line.

Let us say this value is given to be a then I know it is a horizontal line. So, all points will lie
at a same distance from X - axis therefore, all points will satisfy the condition y=a. You take
any point on this line it will satisfy the condition y=a.

So, in case of horizontal lines what I have done is I have identified the condition that is y=a.
So, what will be the condition on points? The points will be of the form ( x ,a), x can be any
value, but the y coordinate of that point will be fixed that is a. In a similar manner we can
consider vertical lines.

(Refer Slide Time: 05:02)

So, what is a vertical line? You can see this in this image and in this video, we can see all
these kind lines are called vertical lines. So, how will I identify these vertical lines? First, I
will define the vertical lines a line is a vertical line if and only if it is parallel to Y - axis.
Now, to specify the location what do I need? To locate a line, I need to know the distance of
this line from Y - axis ok. So, that essentially means what value it takes on x coordinate or X
- axis.

So, how will you identify this? You just need to identify the one point in this particular line
let us say this is the point and I need to see what is the distance of this point from X - axis, if
that is b, then all points of the form (b , y) will be lying on this line; all points of the form
(b , y) will be lying on this line. And therefore, the equation of the line the expression for the
line will have a form x=b.

I mentioned the all points will be of the form (b , y). So, if I get two points where the y
coordinate where the x coordinate is fixed and I know it is a line then I know it is parallel to
Y - axis or it is a vertical line right. In a similar manner the other one is parallel to X - axis
and it will be a horizontal line. Let us make it more crystal clear by solving one example.
(Refer Slide Time: 06:49)

So, here is an example where a question is given to you want to find the equation or
expression for the lines parallel to the axis and passing through point (5,7). Now, the lines
are passing through point (5,7) and it is also given that they are parallel to axes. So, a line
which is parallel to X - axis is known as horizontal line, a line which is parallel to Y - axis is
known as vertical line. So, essentially this question asks you to find one horizontal line and
one vertical line.

So, let us go to the coordinate plane, this is the coordinate plane let us locate the point (5,7),
it will be somewhere here. Now let us first focus on identifying the horizontal line. What is a
horizontal line? A line which is parallel to X - axis is a horizontal line. So, a line which is
parallel to X - axis, then what do I need to know? Its distance from X - axis, the distance is 7
according to this particular expression because (5,7) is a point on that line.

So, the distance is 7 so, the line must appear somewhere here, now further the next question
is I want to find a vertical line that passes through point (5,7). So, now I need to know the
distance of a line from X - axis. So, I will locate point 5 over here and all points on the line
on that particular line will be of the form (5, y). So, (5,7) will also fall on that line. So, this
is the line; so, this is how we will find the lines.

Now what are the typical equations of the line? So, the horizontal line will be y=7 and the
vertical line will be x=5. This is how we will study horizontal and vertical line. So, what is a
vertical line? Vertical line has inclination as at 90 degrees, and therefore, the slope of this line
is not defined remember this in mind.

Another point which is horizontal line it never intersects actually X - axis, but the inclination
of this line with respect to X - axis is 0 degrees therefore, it will have a slope 0. So, we have
eliminated the cases where the slope does not exist or slope is 0, now we need to identify
similar kind of expressions for lines which are not vertical. So, let us go further and identify
such expressions.

(Refer Slide Time: 09:48)

So, here as we already know that slope cannot uniquely determine line, then the question is
which slope if I give you some more information can you determine the line? So, here in this
case what we are identifying is we are giving a point and giving a slope and then we are
asking a question can we solve this problem or can we find a unique expression for a given
line? So, the question is for a non-vertical line l vertical line we do not have to consider
because the slope is not defined. So, for a non-vertical line l with slope m and a fixed-point P

( x0 , y0 ) on the line can we find the equation or the algebraic representation of a geometric
object that is line is the question.

So, here what are the things that we know? We know slope and we know a point on the line.
So, in order to answer this question, we know that two points uniquely determine a line. So,
let us take another point Q ( x , y ) . I do not know the coordinates of these points, but I assume
that this point lies on line l. Now, I know from the definition of slope that I have defined
change in y by change in x the slope of a line is given by. So, what are the two points now? Q
and P.

y− y 0
So, change in y will be y− y o and change in x will be x−x0. So, I know m= , this is
x−x 0
what I know from my definition. It has nothing to do with tan θ even if you have it you can
find out what is tan θ, but since nothing is known in specific we cannot find the tan θ, but
tan θis anyway given to you in terms of slope.

y− y 0
So, now I have m= . So, how will I find the condition on x and y? Just cross multiply
x−x 0
this x−x0 , you will get an expression which is y− y 0 =m(x−x0 ). This condition uniquely
identifies my line, there cannot be any other line satisfying this condition.

So, therefore, any point that lies on this particular line that is P ( x , y) that lies on this
particular line, it must satisfy the condition that is given here. This form of expression is
called point slope form. So, this is a point slope form of equation which essentially says that
give me one point and slope of the line I will give you the equation of a line.

The beauty is the geometric object now can be represented in terms of the equation, initially
when we started, we tried to represent a point which is a geometric object in terms of
coordinate plane and the coordinates of the point. Now we are giving infinite set of points
having certain condition that is a geometric object of line how you can represent it
algebraically using the equation of a line. So, this is point slope form. Let us try to see how
we can use the point slope form in our problem solving.
(Refer Slide Time: 13:40)

So, now, I have been asked to find the equation of a line which passes through point (5, 6)
and has slope of −2. Here the interesting thing is slope is negative. So, let us identify the
point (5, 6) on the coordinate plane and now I want to identify the line that passes through
this with slope of -2. So, now I have a formula for point slope form, I can use that formula
and I can straight away derive it for let us try Q ( x , y) is an arbitrary point.

So, I need two points to identify a line. So, Q ( x , y ) with the arbitrary point on this line then

y−6
using point slope formula we simply substitute −2= , slightly rearrange the terms; what
x−5
you will get is y−6=2(5−x). If you simplify this you will get the expression y=16−2 x.

Now, let us try to see, if I want to know this value of x what point what value of y will satisfy
this equation. Let us put x is equal to 3 here if I put x is equal to 3 here then I get y is equal to
10 after simplifying this I will get y=10.So, that means, the point (3, 10) should lie on this
particular line. So, let us see that (3, 10) is here and now you know from basic geometry that
two points uniquely identify a line. So, you can just draw a line using your ruler passing
through these two points this is the line that we are expecting.

So, the question did not ask you to draw a graph, but drawing graph always verifies whether
you have found a correct answer or not. So, it is better to cross check using graphs. So, the
answer to the question is the equation of the line passing through point (5, 6) and slope -2 is
y=16−2 x . Now, suppose somebody decides not to give me slope and somebody says that
now you have been given only two points; can you find the equation of line?

The answer is; obviously, yes because given two points I can always determine the slope
right for example, in our earlier case when we defined slope I need to figure out what is
change in y and what is change in x using these two points and that will give me slope to be
equal to -2. And therefore, I can always use this formulation to find the equation of the line,
but you can use this knowledge and derive another form that is equation of a line two - point
form.

(Refer Slide Time: 17:00)

So, given two points the question is can you determine the line uniquely which should be
possible and through our basic knowledge of geometry we already know that two points
uniquely determine a line, now we will see that in our coordinate geometry. So, the
assumption is let the line l pass through points P and Q with coordinates ( x1 , y 1 ) and ( x2 , y 2 ).

To start with this, I will take another point R which is arbitrary point because I want to find
the condition in terms of coordinates. So, whenever I want to find the equation of line I will
start with an arbitrary point. So, R ( x , y ) is an arbitrary point on the line l. Now, look at these
three points P, Q, and R they all lie on one line therefore, the points P, Q, and R are collinear
yes; so, points P, Q, and R are collinear.
Therefore, suppose I consider only these two points P and R, using these two points P and R,
I can easily figure out the slope of a line. If I consider points P and Q, I also know the slope
of a line; now because these points are collinear what can you say about slope of line PR and
slope of line PQ, both must be same or equal? So, slope of PR is equal to slope of PQ because
they are collinear.

So, if this is the case, then what is slope of PR? You can easily figure out P is this ( x1 , y 1 )and
R is ( x , y). So, the slope of PR first you consider change in y, y− y 1 upon change in x that is
x−x1that is slope of PR. What is slope of PQ? PQ is( x1 , y 1 ) and ( x2 , y 2 ). So, y change in y is
y 2− y1 and change in x is x2 −x1.

(Refer Slide Time: 19:21)

y− y 1 y2 − y 1
Therefore, I will get the equation of this form = .
x−x1 x2 −x1

So, again if you look at it closely this particular thing is nothing, but the slope of a line m and
we are doing things which are very similar to slope point form. But instead of counting it
explicitly we are counting it as a ratio and then you rearrange the term and you will get this
expression because you just take this denominator on the other side and you will get this
expression.

Now, this line is again uniquely characterized and therefore, any point that lies on this line
must satisfy this condition. So, if your point is R lies on this line then it must satisfy this
condition and this form is called two - point form. So, remember these are the formulas that
we are deriving; first was slope line formula, second is two - point form.

(Refer Slide Time: 20:38)

Let us understand this formula better by solving some examples. So, let us take one example
where I want to find the equation of a line that is passing through two points
( 5 ,10 )∧(−4 ,−2). Let us identify these two points on a coordinate plane ( 5 ,10 ) ,(−4 ,−2). I
want to find the equation of this line.

So, I will use another point Q which is an arbitrary point and it has a coordinate( x , y), I will
use the two - point form. So, using two- point form what should I get? So, I am taking, this

−2−10
point P. So, ( y−10 )= ( x−5). So, always remember this order does not matter I can
−4−5
always start with this as well. So, change in y is 10−(−2) and change in x is 5−(−4) in both

12 4
cases my answer to this particular fraction will be which is .
9 3

So, it does not matter whether you take this as ( x1 , y 1 ) or you take this as ( x1 , y 1 ), you will
always get the same answer. So, if you simplify this you will get the expression of a line

4
because as I mentioned the slope was . So, you just simplify this you will get the expression
3
of a line3 y=4 x+10. So, this will be the line that is passing through these two points.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 19
Representation of a Line-2

(Refer Slide Time: 00:05)

(Refer Slide Time: 00:14)

Now let us go ahead and try to figure out some spatial variations where the calculations
become extremely easy for. These two forms are primary two-point form and slope-point
form. So, when you consider slope point form you can also consider a special case that is
slope-intercept form. So, this is the methodology that we will use for considering slope-
intercept form, before that let me define what is an intercept.

So, let l be the line with slope m that cuts Y - axis at point c. Then this c is called y intercept
of the line l. So, what is the meaning that it cuts Y - axis at c? The y coordinate of that point
is c and the x coordinate is 0; that means, any point that it cuts through Y - axis of line l will
be of the form (0 , c) and that ( 0, c ) will lie on line l.

Now we have our slope point form instead of having any point ( x , y) you have a specific
point which is (0 , c). So, I apply the slope point form or point slope form in this expression.
What you will get instead of y− y 0 you have y−cwhich is equal to m, m is the slope of the
line m times x−x0. What is x 0? Zero.

So, so we will get y−c=mx and therefore, I will get a form y=mx +c, this is a standard form
that we generally deal with when we are dealing with straight lines. So, you have got a slope-
intercept form which is of the form y=mx +c.

The interesting fact is the calculations are very simple whenever you are given the slope-
intercept form. For example, now if you know the y intercept is at c and the slope is m you do
not have to do any calculations, but straight away write this expression that is y is equal to
take the slope m, take the intercept c; y=mx +c will be your answer.

Therefore, the calculations simplify significantly when you are considering a slope-intercept
form. If the intercept is not available then you may have to go to that point slope form and
figure out what it is. Now there can be if the line cuts Y - axis the line can as well-cut X -
axis. So, there can be another variation of this formulation that is if a line l with slope m cuts
X - axis at point d. Then d will be called as x intercept of the line l.

If d is called as x intercept of the line l then how will this point lie on the line l or what are the
coordinates of the line that intersects X - axis and line l? So, what is the point of intersection?
That will be (d , 0) and this (d , 0) lies on line l. So, I will again use the point slope form of the
line.
So, if I want to use point slope form y−0=m¿) will be the answer. So, that will be the form
y=mx−md. So, let us try to use this and solve some problems for finding the equations of
the line using slope-intercept form.

(Refer Slide Time: 04:22)

1
So, typically some example like this. So, I want to find the equation of a line with slope is
2

−3
and y intercept is . Remember here things are very easy because you just need to know
2

1 −3
mx +c. So, what is m? m is and c is . So, upfront I can tell you orally this, the equation
2 2

1 3
of the line will be y= x− . Let us verify the result using the graphics and all other things.
2 2

−3
So, here is the y intercept of this particular line. So, here the y intercept is at point . Now
2

1 3
slope is half correct. So, the equation of line you can easily see is y= x− . So, let us try to
2 2

1 3
figure out what is the x intercept of this line. So, y= x− . So, the x intercept of this line is
2 2
3. So, the question could have been asked that find the equation of a line with slope half and
x intercept equal to 3 that also can be a question and the answer will be same.
So, let us see what is the next question that is find the equation of a line with slope half, but x
intercept is 4 it is not 3. So, it is definitely not a same line because x intercept is 4, but the
slope is half. So, can you relate it to some of the concepts? The slope is half; that means, the
slopes are equal, we have seen that if the slopes are equal then lines must be parallel to each
other.

So, therefore, I can easily see that the line must be parallel to this line with some different
intercept which is at 4 for this the intercept is 3 so, intercept is 4. So, what can be the y
intercept can also be an interesting question. We will answer it later. Right now, let us see
how we can answer the question that is asked here. Find the equation of line with slope half
and x intercept 4.

So, according to our formulation y=mx−md. So, where d is the intercept that is 4 so and this

1
is half. So, y= x−4 is the equation of this line.
2

(Refer Slide Time: 07:00)

You can simplify this which will give you 2 y− x+4=0. So, this will be the expression for
the line. This is the slope-intercept form of the line, now we can go to two-point form that is
suppose I have been given x intercept and y intercept how will I identify the line.
(Refer Slide Time: 07:33)

So, let us now go to the form of intercept that is intercept form, how to find equation of line
when you have been given two intercepts x and y. So, let us formulate the hypothesis,
suppose a line makes x intercept at a, y intercept at b, then naturally the coordinates of these
two points are ( a, 0 )∧(0 ,b). So, we will use two- point form to derive the equation of line.

So, I will take this point as the first point therefore, the y coordinate is 0. So,

b−0
( y−0 )= ( x−a).
0−a

y −x
Now, if you divide this expression throughout by b then you will get = +1. Because this
b a
has a minus sign shift it to the left hand side and you will get this expression which is

y x , now you see how beautiful is this expression; x intercept is a so, below x you put a
+ =1
b a
y intercept is b. So, below y you put b.

Therefore, there is nothing to memorize, it is just a simple trick that

x y
+ =1 that is how you will get the intercept form. So, it is very easy
x−intercept y−intercept
to solve the problems if you remember this trick.
Now, let us take one example where we need to find this. So, find the equation of line having
x intercept at -3 and y intercept at 3. So, you do not have to do any complicated calculations,

x y
you can simply say + =1, multiply throughout by 3 you will get the expression y=x+3
−3 3
.

So, let us verify whether this satisfies because it is always better to verify using graph. So, x
intercept is -3 y intercept is 3, the line that passes through these two points is y=x+3. This is
what the intercept form is, it is very simple and you can practice more and more problems.

That is all for today.

 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02
Tutorial 01
(Refer Slide Time 00:19)

(())(0:18) In this tutorial we are going to look at the problems which are related to contents of
week 2, that is to do with straight lines and all these topics here.

(Refer Slide Time 00:34)

So, we will start with our first question. The data provided here is, there is a company which is
selling mobile phones and it all begins in March 2019. In March 2019, the selling price was
8000 and it was sold at 8000 rupees, mobile A was sold at 8000 rupees from March until June.
After that, due to increasing demand, the company decided to increase the price by 250 each
month, so they are selling better.

So, they have decided to increase their price by 250 rupees every month. This went on until a
new mobile B was launched at a lesser price, competition at a lesser price was launched in
January. So, because of this the selling price of A dropped at a rate of 500 per month, from
January till March 2020, so 2 months it had decreased. We are expected to demonstrate a clear
graph of this. For that let us look at this graph.

(Refer Slide Time 01:58)

What we need to realize about situations like this is, the 𝑥 and 𝑦 axis do not necessarily
represent the same units. So, we have along the 𝑥 axis 1 unit is 1 month, however along the 𝑦
axis 1 unit is, let us take about 250 rupees. So, 1 month and 250 rupees are not the same thing,
so please remember this in situations like this. Now, because we are beginning from March let
us take the starting month to be March, then this is April, May, and so on.

So, our entire problem deals with this 1 year span from March 2019 to March 2020, so this will
be along our 𝑥 axis. And now along the 𝑦 axis, if we took each unit to be 250, then this is 250
and this is 500 and so on, the 8000 will be beyond our screen. So, to better represent our
situation, we are going to introduce a zigzag here to indicate that a lot of values have been
compressed into this little space. So, we are going to start from 8000 and this is going to be
8250, 8500, so on. And now we begin to mark out the points that we have, we know that in
month of March the price was 8000, so this is the point for the month of March.

And then in April, May, and June the price stayed constant so it is been like this. And this
portion can be represented using a horizontal line and this line is y is equal to 8000. Beyond
that, the price had been increasing by 250 every month so in July we will be here, August here,
September here, this will be October, this will be November, this is December, and this is
January.

So, this segment can be indicated by this line, in order to find out the equation of this line we
use the 2 point form, so we first write 2 points on this line segment. You could choose this one
which is August, and for that let us number our months now, so March will be 0, April is 1,
May is 2, June is 3, this is 4 and this is 5. So, our price point here it is (5, 8500).

I am ready to take another month, so let us take October, this is the seventh month from March
2019, so this point becomes (7, 9000). Using these 2 points, we can find the equation of the
𝑦−𝑦1 𝑦 −𝑦
line by employing the 2 point form of the line equation, = 𝑥2 −𝑥1 , where 𝑥1 , 𝑦1 and
𝑥−𝑥1 2 1

𝑦−8500 9000−8500
𝑥2 , 𝑦2 are two points on the line segment. So, here we can see it as = . So, this
𝑥−5 7−5
500
would be equal to = 250.
2

So that implies 𝑦 − 8500 = 250(𝑥 − 5), which finally gives us the line equation to be 𝑦 =
250𝑥 − 1250 + 8500 plus this line is y = 250𝑥 + 7250. Moving on, the next 2 months, the
price dropped by 500 each month. So, here we are at 9750, then for February we should be at
9250, so this will be our point for February and then the next month again 500 drop we will
reach here, which is 8750. And this line segment also corresponds to a straight line, which also
we can find using the 2 point form.

So, this point here is, let us number the months completely, this is 8, this is 9, this is 10, this is
11, this is 12. So, this point here, which is January is the tenth month, and the 𝑦 axis gives us
9750 whereas this point here, this is the twelvth month, and it corresponds to 8750. And again,
we would like to know the line equation for this and we use the 2 point form again.

𝑦−8750 9750−8750 1000

So, this is = that gives us = −500. Plus we have 𝑦 − 8750 = −500𝑥 +
𝑥−12 10−12 −2

6000. That gives us 𝑦 = −500𝑥 + 14750, so this is our new length. And this is a clear graph
of the situation and the given question.

(Refer Slide Time 09:18)

For the part B of this question, it is asked, what is the price of mobile A in December. So, this
is the December month, which would be this point here which has a price of 9500, so this is
our answer for B. And then in C it has asked, calculate the slope of mobile A’s price from
January to March 2020, so we want the slope of this segment here and this slope we had already
9750−8750 1000
calculated, it was 𝑚 = = = −500.
10−12 −2

And because of the negative slope you can see that it is a decreasing function, which is what is
happening, the price had fallen at 500 per month. Lastly, we have being asked, what is the price
of mobile A in March 2020, so this is March 2020, this is a point and we have already found
the price which is 8750 that is our part.
 Mathematics for Data Science 1
Indian institute of Technology, Madras
Week 02
Tutorial 02
(Refer Slide Time 00:16)
In the second question, the reserved triangular field ABC, whose coordinates are given. And if
watering costs rupees 10 per unit square, so they are giving the cost of watering the field, and
it is so and so amount per unit square that is area, how much would you have to pay for the
whole field? So, we would like to find out the area of the field. And then if the fencing wire
around the field costs rupees 5 per unit, how much should he have to pay for 3 rounds of fencing
around this field that is find the perimeter, so find the area and perimeter of this particular field.
(Refer Slide Time 01:08)

So, if we consider this to be our origin, the triangle is made up of these points, (1, 1), this will
be (9, 1) and this is (1, 7). So, this is our triangle, this is A, this is B, and this is C and you can
see that AC is completely vertical, its 𝑥 coordinate remains the same, it is 1, and AB is
completely horizontal, its 𝑦 coordinate remains the same, which is 1.

So, this is a right angled triangle. Now, we could use the area of triangle formula, the area will
|𝑥1 (𝑦2 −𝑦3 ) + 𝑥2 (𝑦3 −𝑦1 )+ 𝑥3 (𝑦1 −𝑦2 )|
be , which in this case is going to be, you can take any of these
2

points to be 𝑥1 𝑦1 and the others to, the next one to be 𝑥2 𝑦2 and 𝑥1 to be 𝑥3 𝑦3 . The, how you
choose 𝑥1 𝑦1 𝑥2 𝑦2 and 𝑥3 𝑦3 does not matter, the order is what is important. Applying this
formula, we get our 𝑥1 , 𝑦2 , 𝑦3 is 1.

|1(1−1) + 9(1−7)+ 1(7−1)| |−48| 48

So, = = , and that is 24 square unit. However, the same problem
2 2 2

could be approached in a slightly different way which is, if I observe that this is a right angle
triangle, I could just do half into base into height. And here the base would be the length AB,
that is half into AB for which the height would be the length AC.

And now since AB is horizontal, you can directly take the length to be 𝑥2 − 𝑥3 which is the
1
difference in the 𝑥 coordinates, so × 8 × 6 we have 24 square unit. So, this worked out
2

because our triangle is a right-angled triangle. So now the cost of watering is supposed to be
the area into cost of watering per square unit which is 10 rupees, so total cost of watering is
equal to 24 into 10 that is rupees 240.
(Refer Slide Time 05:18)

For the second part of the question, we require the perimeter of this triangle because fencing is
done along the perimeter, and they have to do 3 rounds of fencing at the rate of rupees 5 per
unit. So, we first find the perimeter.

(Refer Slide Time 05:39)

Perimeter would simply be AB plus BC plus CA, which is AB is clear it is 8 units, CA is also
clear which is 6 units, BC needs to be figured out and BC we find out using the Euclidean
distance, that is the √(𝑥2 − 𝑥1 )2 + (𝑦2 − 𝑦1 )2 that is the square of the difference in 𝑥
coordinates plus the square of the difference in 𝑦 coordinates, the whole under root.
So, this gives us √(−8)2 + 62 = √64 + 36 = √100 = 10. So, we have 10. So, this quantity
is 10 and thus our perimeter is also 24 units and we need wiring for fencing around 3 rounds.
So, we will require 24 into 3 is equal to 72 unit of wire and then each unit has been fixed a
price of 5 rupees. So, we have 72 into 5 is rupees 360 is the cost of fencing.
 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02
Tutorial 03
(Refer Slide Time 00:17)

In the third question, the two friends positioned at these two locations and both of them go to
a position P. The speeds are given, and the time of their meeting is given, then what should be
this position P given that 1unit distance is equal to 1 kilometre. So first let us look at their
positions.
(Refer Slide Time 00:55)

So, this point is the origin, and now among the 2 friends, 1 Abdul is at (-2,2), so this is -2 here,
and this is 2 here. So, Abdul is here, A (-2,2). And we have the other one Ram at (4,10), which
is this is 4 on the x axis and this is 10 on the y axis, so Ram is here (4,10). It says they are
moving towards each other, so this is a path they take, where Abdul is moving this way and
Ram is moving this way.

And what we know about their movement is, Abdul is moving at 60 kmph and Ram is moving
at 90 kmph, so Ram is faster and they are meeting in 4 minutes. If 1 unit is a kilometre, we
have 60×4/60 because it is 4 minutes and the units are in hours kilometre per hour, so we do
4/60 is equal to 4km.
So, Abdul is moving 4 km, whereas Ram is moving 90×4/60, which is 6 km, so they meet
somewhere in this region and we would like to know that point. And that point we can achieve
through the section formula; we do not actually need to find the distances. And for applying
the section formula, what we need to know is the ratio of how this point cuts the line segment
AR. And that ratio we can use it in this way.

So, we know that this length is supposed to be 4 km and this length is supposed to be 6 km
which means the ratio is 4:6 that is 2:3. So, we now apply the section formula, which is
(𝑚 𝑥 + 𝑚 𝑥 )/(𝑚 +𝑚 ). This will be the x coordinate of that point and
(𝑚 𝑦 + 𝑚 𝑦 )/(𝑚 + 𝑚 ) will be the y coordinate of that point. So, let us call this point B,
so this is the formula for B, so we get the point B is applying 𝑚 is, this is the ratio 𝑚 : 𝑚
and this is (𝑥 , 𝑦 ) and this is (𝑥 , 𝑦 )

So, we have, 𝑚 𝑥 would be 2×4 + 𝑚 𝑥 would be 3×(-2) the whole by 𝑚 +𝑚 is 5 and (𝑚

𝑦 would be 2×10 + 𝑚 𝑦 would be 3×2)/5 again. So that gives us 8-6 = 2, 2/5 is 0.4, and
2×10 = 20, 3×2 = 6 or 26 /5 = 5.2. So, B is (0.4,5.2). We can check with our intuition, this point
that we marked out actually has an x coordinate between 0 and 1 and a y coordinate between 5
and 6. So the point we are looking for is (0.4,5.2).
 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02 - Tutorial 04
(Refer Slide Time 00:16)

Now, fourth question, there is a line which is represented by 7y = 56 - 8x.

(Refer Slide Time 00:27)

Let us first draw this line, so this is our origin and our line equation is 7y = 56 - 8x. In order to
draw this line, in order to find out the curve, we need two points, two points are enough. And
the easiest way to find out these two points is to work with the intercepts, that is when this line
cuts the X-axis and when it cuts the Y-axis. So when, it is cutting the X-axis, y will be 0, so we
just take the Y-coordinate to be 0, and we write 0 = 56 - 8x and to denote that this is the
intercept, I am going to call it 𝑥 and that gives us 8𝑥 = 56 and that gives us 𝑥 = 7. So, the
x- intercept is 7 which is here. So, (7,0) is one point. And now, for the other point, we take x
to be 0 and thus we can say 7y is equal to 56. Again, for the intercept, I am going to use 𝑦 ,
56 - 0, therefore 𝑦 , , the y intercept is 8. So, this point here, which is (0,8), this is our y-
intercept.

(Refer Slide Time 02:09)

So, this is a straight line, we have been given 7 y = 56 - 8x. It passes through (7,0) and (0,8).
Now, for a mirror image, what happens is, and here we are treating the Y-axis as the mirror, so
Y-axis is the mirror, you are at the same distance from your mirror as your reflection. So, your
reflection will be at the exact distance from the mirror on the opposite side as you, so for
example, if we take our (0.7,0) on the other side, which is this point that is (-7,0), that would
be the reflection of (7,0) with respect to the Y-axis as the mirror. However, (0,8), since it is
already on the Y-axis, its reflection is going to coincide with itself, so this is the other point of
the reflection.

And thus, the mirror image for this line is going to be this other line which passes through these
two points, (-7,0) and (0,8). For finding the equation of this line, we can use the two point form.
And when we apply the values, we get (y – 0) / (x +7) = (8 – 0) / (0 + 7), which gives us 7y =
8 x + 56. So, the mirror image line if you have to write it in the same form as the other one, 7y
= 56 + 8 x.

(Refer Slide Time 04:40)

Now, in the next part of the question, they are asking if A is the set of all elements inside the
area enclosed by these two lines and the X-axis. So, we are looking at this triangle, and in this
triangle, we have being asked what is the set of Y coordinates of the points in set A.

(Refer Slide Time 05:03)

So, all possible Y coordinates in this set. So, every point within this triangle and on the triangle
itself count, and as you can clearly see the least Y coordinate here is 0, and the maximum Y
coordinate here is 8. So, the set of Y coordinates is going to be the closed interval [0, 8], because
we are considering the triangle also to be part of this set, not just the points inside the triangle
interior to the triangle, we are considering the triangle also to be part of the set. So, this is the
answer for part A. And for part B we have what is the set of X coordinates of the points in set
A, and again, we look for the least and the maximum here, the least is -7 and the maximum is
7. And every value in between is there so this would be again the closed interval [-7 ,7].
 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02 - Tutorial 05

(Refer Slide Time 00:15)

Now 5th problem, Mary has subscribed to a cell phone plan with 400 free minutes, a 50 rupee
monthly fee and 20 paisa for every additional minute over 400. And the question is, what is
her bill amount if she uses 700 minutes?

(Refer Slide Time 00:42)

So let us put down our variables here. So there is 400 free minutes and there is a 50 rupee
charge per month and we have 20 paisa that is 0.2 rupees per minute over 400 minutes. Now,
our independent variable is the number of minutes, the bill is dependent on the number of
minutes, so our x variable is number of minutes and the y variable is bill amount. And what
we know is for every month the bill amount will always have a 50 rupee charge, and on top
of that you are being charged 0.2 for every minute over 400, which means if x is the total
number of minutes, then (x – 400) (0.2) will be the charge for the additional minutes.

This is the fixed charge whereas this is the additional minutes charge, so we get a linear
equation which is y = 50 + x/5(because 0.2 is 1/5) - 80 which is then (x/5) - 30 . If we
simplify it further, we get 5y = x - 150 => x - 5y - 150 = 0. This is the equation that relates
our bill amount to the number of minutes. So, Mary is using 700 minutes per month and we
need the bill amount for that. So, if we substitute x = 700, we get, 700 - 5y - 150 = 0, this
gives us 5y = 550 which implies y = Rs. 110. This is the bill amount for Mary.
 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02 - Tutorial 06

(Refer Slide Time 00:17)

For our 6th problem we have these 4 points given to us. Let us first plot them out on a graph.

(Refer Slide Time 00:32)

And this will be 0, we have a (-4, 4), so this is -1, this is -2, this is -3, this is -4, this is 1, 2, 3
and 4, so this point here is our K (-4, 4). And then we have (6.5, 6.5), this is 1, 2, 3, 4, 5, 6, 7,
this here is 6.5 and 5, 6 and 7, this here is 6.5, here we are with L (6.5, 6.5), then we have a
(2, -2), -1, this is -2. So this point here is our M (2, -2). And lastly, we have (-5, -5), -4 and
-5, this is our point.
(Refer Slide Time 01:48)

Now we are told that R is the point of intersection of KM and LN, and it is known to cut the
line segment KM in this ratio, 4 is to 2 ratio, so let us identify R.

(Refer Slide Time 02:04)

So from our diagram, it appears to be the origin. Lets verify this, so we need this to be in the
ratio of 4:2. So when we use the section formula, which is the coordinates of a point cutting a
line segment in a ratio, m1 : m2 would be this, (m1x2 + m2x1) /(m1 + m2). And then we have
(m1y2 + m2y1) /(m1 + m2). So in this context, R is going to be ( (4(2) + 2 (-4))/ 6 , (4(-2) +
2(4))/6 ). And these 2 cancel out because it is 8 - 8, these two also cancel because -8 +8. So it
is true, R the point is the origin. Moving on then, we have two other points, P and Q given to
be (4, 0), (0, -7), so these are on the axis. So this point here is P(4, 0), and this is -6, this is -7
so this point here would become Q, which is (0, -7).

(Refer Slide Time 03:58)

Lets look at the options, RP and RQ are parallel, this is one option, lets verify. Now clearly,
this is 90°, PQ, PR is perpendicular to RQ and not parallel. So this is definitely wrong and
this is definitely right. Is there adequate information for finding the relation between RP and
RQ? Yes, we have just found the relation, so there has been adequate information.

Now let us look at ∠ LRP +∠ PRM. So we are interested in this angle plus ∠ PRM. So this
sum is the total ∠ LRM, so we need to know what is the angle between LR and RM. So let us
look at the slope of LR. So this slope if I call it m1, this is equal to (6.5 - 0) /(6.5 - 0), which is
1, which is basically tan 45°, so this angle here it is 45°.

And now let us look at this angle here, which is PRM. Then, if we look at the slope here,
which is m2 that is (-2 - 0)/(2 - 0), which is -1, which is equal to tan -45°, therefore this angle
here is -45° because we are going clockwise from the horizontal. So in sum, we know that
∠ LRM is 45°+ 45°, leading us to see that this is 90°.

So this is true, which means the following statement is false, so this is false. And here we
have again adequate information for finding the relation between LRP and PRM. Clearly, we
have four options correct, so none of the above is not correct.
 Mathematics for Data Science 1
Indian Institute of Technology, Madras
Week 02 - Tutorial 07
(Refer Slide Time 00:15)

In the 7th question we have two points, one is the origin O, and some other point P (7, 3). And
this line segment OP is being rotated. So first, lets mark out these points.
(Refer Slide Time 00:32)

So first lets mark out these points, we have this is the origin and this is 1, 2, 3, 4, 5, 6, 7, this
is 1, 2, 3. So our point P is here, this is P (7, 3) and this is the origin of course O. And we
have this line segment OP given to us. Now, OP is being rotated by 360° about the x axis, lets
see what that means. So every point on OP is going around the x axis in a circle, that is what
rotation is, rotation is a combined circular motion of many particles, here we have this point
let us take P, P goes around the x axis reaching this bottom point, then it circles back to itself.
So you would see the circle if you looked at it from the right. From the screen’s perspective,
this is what it will look like. And this is the case with every point on this line.

Suppose I took this point, this is just oppositely going to go till here in this circle and return
back. So, every point is doing this circle, which means on this side we actually have the
mirror image of OP with respect to the x axis which looks something like this. So, we have
these circles being formed due to the rotation and as you can see, the final shape it appears to
be a cone that is what has happened. Take a line segment and you rotate it about some central
axis, you obtain a cone about that central axis.

(Refer Slide Time 02:53)

And they have given us the volume of a cone, volume of a cone is (4/3)πR2h, where R is the
radius of the base circle and h is the height of the cone.
(Refer Slide Time 03:16)

So in the cone we have obtained the radius, base radius is this quantity, which is R = 3
because it is a y coordinate of the point P that is a distance of point P from the x axis. And
likewise, if you observe the height of this cone, that quantity h = 7 because the x coordinate
of point. In this way, we can obtain the volume of our cone using the formula that is given V=
(4/3)πR2h. We are going to approximate pi to be 3.14 or 22 by 7. So, this is roughly equal to
(4/3)(22/7)(9)(7) so 7 and 7 cancels of, 3 and 9 gives us 3. So we get 264 cubic units, so this
is our volume of the cone.

(Refer Slide Time 04:54)

Now in the second part of this question, it is being said that the rotation is done around the y
axis instead of the x axis, so what will this look like?
(Refer Slide Time 05:08)

So, this OP is going to have a mirror image about the y axis which is going to look like this.
So that means, our point P is going around in a circle to reach this opposite point here and it
is coming back to itself. So this would become the base circle for our new cone which is
obtained by rotation about the y axis. Now as you can see, this value is already the height so
height is 3 now, whereas the radius is basically 7. So, our height is 3 and radius is 7 so these
values have changed. So, if we call this quantity, this volume to be V2, V2 is (4/3)πR2h, which
will be roughly equal to (4/3)(22/7)(7)(7)(3). So, 3 and 7 cancel off here and we get this is
equal to 616 cubic units. So, this is the volume if OP is rotated about the y axis 360°, so that
cone’s volume is 616 cubic units.
(Refer Slide Time 06:58)

This problem gets progressively more complex, we are now adding the new point (14, 6)
which is along the line segment OP, it is on the extension of OP, and then PQ is rotated about
x axis by 360°

(Refer Slide Time 07:25)

So, lets see what is happening here, so our point (14, 6) Q, (14, 6) is here, which gives us PQ
as this line segment. And now, they are saying that PQ is being rotated about the x axis which
will result in the mirror image in this way. This is -1, -2, -3, -4, -5, and -6. So, we are here
now, I think we can call this point T and this point is S. So, we have ST in this way again, so
what we see here the rotated geometry. So, for reference we are going to take one more point
here, which kind of moves around. So, what we are seeing here this is what is called the
frustum of a cone. This is a cut-off portion of a larger cone which would be QO rotated about
the x axis.

(Refer Slide Time 08:56)

So, thus as you can see, the volume we require is the frustum of the cone which is this region
and this volume is the result of subtracting this volume from the total cone. So, we already
know this volume OP as that cone’s volume to be 264 cubic units. So, the blue shaded region
that would be the volume of the large cone that is of OQ rotating about x axis and that we can
calculate as V3 = (4/3)πR2h, where this is approximately equal to (4/3)(22/7)(6)(6)(14). So, 3
one’s 3 two’s, 7 one’s 7 two’s, so we have 2,112 cubic units. So the volume we require is
going to be 2,112 – 264 = 1,848 cubic units.
 Mathematics for Data Science 1
Week 02
Tutorial 01

(Refer Slide Time 00:14)

The 8th problem is pretty interesting. So, we have Sania who hears a sound in a night, and
she comes out to her balcony, which is at a height of 80 feet from the ground.

(Refer Slide Time 00:29)

So, let this be our tower, which has a height 80 feet. So, if we take this point to be origin (0,
0), Sania is here, which would be (0, 80). And she uses a torch light which makes angles θ
and α with the ground, so the rays from the torch light make angles between these two.

So, this angle here, this is θ and this angle here it is α . And the two thieves, their heights are
given and they are standing at these distances from the buildings. So, thief T1 is somewhere
here and T2 is here, what is given to us is this distance is 37.5. So, T1 is (37.5, 0), and this
distance is 50 feet, so this is 37.5 feet, this is 50 feet. So, T2 will be the point (50, 0). And we
are also given to understand that T1 is standing at a certain height, T2 is standing at a certain
height, which are roughly the same; one is 5 feet, the other is 5.3 feet.

In our diagram, we have drawn the rays of light as though they are passing away from the 2
thieves, however that we need to find out. So, if tan θ is 2, and tan α is 16/9, can Sania see
any of the thieves ? so it is given to us that tan θ = 2 whereas, tan α = 16/9.

So, that means we can find the slope of this line, which is the lowest ray of the torch, and this
line which is the farthest ray from the torch, and these slopes would be m1 =-2 and the minus
is because the standard angle here, which is angle from the posture x axis is actually 180 o - θ.
As you can see, it is clearly a line with the negative slope.

Likewise, this also is 180o - α, thus this slope m2 = -16/9. In our diagram, we have drawn it as
though the 2 thieves are safe. But this is only a rough schematic diagram, we did not draw θ
and α accurately. What we need to do now is to check if the line from Sania to the head of
thief 1 or the line from Sanya to the foot of thief 2. If these 2 lines have slopes between m1
and m2, then the 2 thieves are likely to be seen. So, we need to calculate these slopes, let us
call the head of thief 1 as H1 and that point will be (37.5, 5.3). So, slope of SH1 = (80-5.3)/
(0-37.5), which is -(74.7 / 37.5), which is roughly -1.992.

And slope of ST2, which is to the foot of thief 2 is (80 - 0) / (0 - 15), which is equal to -1.6.
So, I want to call this m3 and this is m4. And here m2 is roughly equal to -1.78. So, clearly m3
is greater than m1 and lesser than m2, but m4 is greater than m2 and also greater than m1 which
means m4 that is the foot of thief 2 is not visible to Sania, the actual light cone looks
something like this. So, thus we can say the head of thief 1 is visible whereas thief 2 is not
visible.

(Refer Slide Time 7:19)

And this light cone that we have drawn earlier it is wrong.

(Refer Slide Time 7:28)

Now, she moves her torch so that she can see the ground from a distance of 48 feet. Can she
see thieves or not?

(Refer Slide Time: 07:32)

That would mean she is able to see from some point here to some point beyond. From the
diagram, it is pretty clear that thief 2 is going to be visible, we do not know if thief 1 will be
visible though we have to check for thief 1's head. So, this point that we are talking about,
which is let us call it P gives us a slope with S as SP, the slope is equal to (80 - 0) / (0 - 48)
because point P is basically (48, 0).
So, that gives us - (80 / 48) which is divisible by 16, both of them are divisible by 16. This
would be 5 and this would be 3, so this is roughly -1.67. That would give us m3 is let us call
this now m5. m3 is lesser than m5. And that means the head of thief 1 is not visible now, but
thief 2 is visible.
 Mathematics for Data Science 1
Week-02 Tutorial-09
(Refer Slide Time: 0:12)

For our 9 th problem, we have 2 colleagues Ramesh and Suresh, and their office starts at 9:30
AM, Suresh starts at 8:50, Ramesh starts at 9, and they both go at equal speed. At 9:20 they
decide to increase their speeds in order to reach their office on time, which is at 9:30 and this
increase in speed was 30 kilometer per hour each, and they manage to reach the office on
time. So, the timer begins at 8:50 AM, which means our origin is corresponding to 8:50 AM.

And since we know that Suresh started at 8:50 path A must belong to Suresh and Ramesh
started a little late, so this here should be 9 AM. So, B, the path B corresponds to Ramesh's
journey, which gives us option A is correct. Of course, this is wrong because both paths do
not belong to Suresh. This is also wrong because path A does not belong to Ramesh, both
paths do not belong to Ramesh and Ramesh has a path Suresh has a path so all of these
options are wrong, only option A is right.
(Refer Slide Time: 1:54)

Now, in the second part, we are being asked the final position t, d, where t must be in minutes
and d must be in kilometers. So, what, so this is not actually the position, is a coordinate in
this particular graph regarding the final position of Ramesh and Suresh respectively.

(Refer Slide Time: 2:24)

So, we know that, Suresh started at 8:50 and he traveled till 9:30. That means, Suresh
traveled for 40 minutes, whereas, Ramesh started at 9 AM and reached office at 9:30 AM. So,
Ramesh started for 30 minutes.
(Refer Slide Time: 2:50)

However, both of them reached at the same time, which means, this point and this point in the
graph, both of them have the same x coordinate. Now, that clearly rules out this, this, this and
this, because none of these have the same x coordinate.

(Refer Slide Time: 3:18)

In terms of the number of kilometers traveled, Suresh goes at 60 kmph for an hour till 9:20.

So, in an hour he must have covered 30 kilometers and then for 10 minutes, he goes at a

speed of an additional 30 kmph, so, 90 kilometer per hour for 10 minutes. So, 10 minutes is

an hour, gives us 15 km. So, overall Suresh covered 45 km, so this point it must be 40,

45.
Whereas, Ramesh also covered the same 15 km in those 10 minutes but in the initial time of
established it is only 20 minutes, he did not cover 30, he instead covered 20 minutes is of an

hour 60 gives us 20 km. So, Ramesh covered 20 km + 15 km s giving us 35 kilometer

overall. So, this point here it is 40, 35.

(Refer Slide Time: 4:48)

So, our correct option is this one.

 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras – Chennai

Lecture-20
General Equation of Line

So, far in our journey we have studied how to represent on line which is a geometric object in
algebraic manner using various forms of equations. This is a time to recollect; what are the forms
of equations that we have studied and understand some common properties commonalities in that
equation of line and give a general equation of line which will be helpful for further analysis. So,
let us see what are the different forms of line; equations of line that we have studied.
(Refer Slide Time: 00:49)

So, in particular we had two forms one is two-point form another when its slope point form. So,
first I will list the slope point form, a specialized version of this is slope intercept form where
instead of a point you have been given x intercept or a y-intercept. Then we have also studied
two-point form given two points how to uniquely determine a line and a specialized version of
that is nothing but intercept form.

So we can quickly review these forms like slope point form we have a point ( x0 , y0 ) which is
given to us and a slope m that is given to us. So, we come up with an equation when we give the
algebraic representation of this line with slope y with slope m and point ( x0 , y0 ) , we will come up

with a representation as ( y− y 0 ) =m(x−x0 ). When you come to slope intercept form suppose the
x intercept is given to me if I have been given an x intercept then the y coordinate of that point
will be 0.

So let us say x intercept is d, in that case my equation from slope point form as slope intercept
form is a specialized version of slope in point form. My equation will become ( y−0 )=m( x−d)
if the intercept is at d. So, y=m( x−d), in a similar manner so the y intercept is given to me and
that intercept is at c then my y 0 will be replaced by c and x 0 will be replaced by 0 therefore I will
come up with an equation y=m x+c that is what is listed here, given a y-intercept and given an x
intercept the equation has a form y=m( x−d).

Let us come to two point form we have also seen during the course that this two point form is
closely related to slope point form. We also know that given any two points on a line we can
determine the slope of a line so in this particular expression m will be replaced by the ratio of the
change in y upon change in x. Therefore the two point form will be just replica of this instead of
m you will have the difference between y-axis difference between the coordinates of y-axis and
difference between the coordinates of x-axis that will be given in this form.
y 2− y1
( y− y1 )= ( x−x1 )
x2−x 1

Now remember here the points given are not ( x0 , y0 ) so the points given are ( x1 , y 1 ) and ( x2 , y 2 )
therefore x 0 is replaced by x 1 and y 0 is replaced by y 1 and this thing is nothing but a replacement
of m that is how these two forms are also closely related. In an intercept form you will get two
intercepts x intercept let us say x intercept is a and y intercept is b then how will these forms
change?

If x intercept is a that means I have a point (a ,0) so my ( x1 , y 1 ) will be nothing but x1 =a , y 1=0

b −b
and ( y2 − y 1) will be b, x2 −x1 will be minus a, so is equal to x, and if you simplify that
−a a
 x y
you will come up with a very simple expression of the form + =1. So, here there is a no-
a b
brainer nothing to remember below x you write x-intercept below y you write y-intercept and
equate it with 1.

Now if you look at all these forms there is one common feature, let us take the slope point form
given a point ( x0 , y0 ) this x 0∧ y 0is fixed. The slope of a line is fixed. So, now what we are
identifying is we are identifying in a condition in the form of (x, y) what these coordinates
should satisfy. So, the variables are x and y.

If you look at all these forms the same feature is visible, the variables are x and y and I have an
expression of the form some constant times y some constant times x and added with another
constant. Let us take this feature for example y− y 0 =m ( x−x 0 )now I want to differentiate
between variables and constant. So, I can simply write this as y−mx= y0 −m x0. y 0 −m x 0 will be
the constant associated with this particular equation and y and one variable y is associated with
real coefficient 1 and variable x is associated with real coefficient -m.

So in particular I can have a general form of the equation and similar story is true for all this. For

1
example, if you come here, with variable x, is a real coefficient that is associated, with variable
a

1
y, that is a real coefficient that is associated and the constant is c. So, I can discuss same things
b
about all these features but one thing is common that I can have a general form of equation
which will be of the form Ax+ By +C=0.

Now let us identify this particular general form with our various expressions like slope point
form, the way I discussed the slope point form we already know. In this case we have assumed
that b is equal to 1 but I can as well multiply by a constant term throughout the equation and we
will have the same equation. So, assuming this holds true let us discuss about this particular
expression. So, in this case you can easily see if I relate this equation with this equation that is
you rewrite this as y−mx= y0 −m x0.
In that case you can have this expression which will give the value of m when you compare with

−A
respect to this expression as and value of y 0 −m x 0, now remember this is a constant term
B

−C
because all these are constants. So, y 0 −m x 0= . If you are able to understand this then you
B
can easily understand the slope intercept form. Because in the slope-intercept form, y=mx +cyou
have y-intercept which is c therefore your y 0 will be replaced by c and x 0 will be replaced by 0

−A
so if you look at this expression m will still remain , when I am identifying this equation m
B

−A −A
will still remain by minus
B
, y 0 is identified with C−
B ( )
x 0is 0 so this becomes irrelevant

−C
so y 0 is c so c= . In a similar manner you can do for x-intercept and you will get these
B
expressions.

−C
So m as I mentioned c= and for getting d you just put x 0=d and y 0 =0, you will get this
B
expression. So, same exercise can be done for two point form and intercept form remember this
m will be replaced by a ratio of these two differences. So, m is replaced by a ratio of these two
differences there is no ( x0 , y0 ) there will be ( x1 , y 1 ) therefore you will have an expression of this
form.

−C −A
But remember this is common everywhere the slope is everywhere so essentially, we
B B

x y
have got one simple general equation. Similar things you can do for + =1 that is intercept
a b

−C −C
form and you will get a= , b= . So, what we have seen here is an exact matching one-to-
A B
one correspondence of a general equation with respect to this equation.
Now why should I consider general equation? Remember when we figured out this
representation our assumption was these are non vertical lines. For vertical lines our slope do not
exist but in this case if you; and those lines are where the slope do not exist those lines are
vertical lines. They are of the form x is equal to some constant. If you look at this equation which
is a general form of this equation you just put B to be equal to 0 you will get Ax+ c=0 that

−C
means x is equal to some constant x is equal to , you will get that is what our intercept form
A
also reveals.

So all these lines are actually vertical lines, so this general equation is capable of handling
vertical lines also, horizontal lines are anyway handled here because if you put m is equal to 0
the horizontal line is handled. While we were deriving these forms we were always assuming
non-vertical lines. So, non-vertical lines are covered as well as vertical lines are covered
therefore this equation is a general form of equation of a line.

Also, in your earlier classes you might have studied this as a polynomial in without this equal to
0 Ax + By + C is a polynomial in two variables and it is a linear polynomial in two variables.
Therefore you will hear a term called linear equation in two variables. So, in particular if this has
to represent general form of a equation of line then A and B cannot be simultaneously equal to 0.

If A and B are simultaneously equal to 0 then I am actually equating constant with a zero which
is invalid therefore the assumption will always be A and B cannot be simultaneously equal to 0.
Though individually they can be equal to 0 for example you can put A is equal to 0 then you will
get y is equal to some constant which is a line parallel to x axis. You can put B is equal to 0 then
you will get a line x is equal to constant, x is equal to constant is parallel to y axis.

So now we will bring up a definition that any equation of the form Ax + B y + C = 0 where A
and B are not equal to 0 simultaneously individually they can be 0 or they can be nonzero as well
is called general linear equation because we are handling a linear polynomial which is equated to
0 so it is an equation, general linear equation or general equation of a line. So, what we are
summarizing here is a polynomial in two variables or and general linear equation in two
variables gives you line.

So this is the identification of a geometric object called straight line with an algebraic
representation of general linear equation. So, this will give us both the strength in our analysis
because now you do not have to discuss about the line. But you can as well discuss about its
algebraic representation or you can start with an algebraic representation of a line and then
discuss about the geometric properties of the line. How let us see in the next slide.
(Refer Slide Time: 14:30)

So, here is an example, the example gives you a question that the equation of a line is
3 x−4 y+12=0. Now I do not know how this line behaves now I want to see how this linear
equation represents a line. So, when I talk about a line what is the natural question we will talk
about what are the two points that uniquely determine this line or you can ask what is the slope
of a line and give me one point on a line because we have slope-intercept form or we have two
point form any of them should be usable.

So in order to discuss about the geometric aspects we can ask a question that find the slope or x
intercept or y intercept of a line. So, how will you find this the job is pretty simple let us go back
and revisit the previous slide which will make the job very simple. Suppose I want to determine
the x-intercept and y-intercept then I have this intercept form right which says that a is the x-
intercept and b is the y-intercept.

Now you I have been given an equation in this form which is Ax+ B y+C=0 so I can image
lately consider this equation and consider the values of a and b which is −C / A and B is equal to

−C . So, let us go and do the same thing on the on the our; now our problem so we have
B
identified Ax+ B y+C=0. So, what is A, A is 3, B is minus4, C is positive 12. So, what should
be my x intercept A as you have seen in the previous slide is −C / A.

12
So what is C? which is 3 so my a is 4, and a minus sign associated with it so a=−4. In a
A

12
similar manner you can talk about y intercept which is =−3 but a minus sign because it is
−4

−C so it will be 3. So, now we can readily answer the question what is on a x-intercept and y-
B
intercept.

Now the question comes what is the slope of a line. So, for slope of a line you can use the slope
intercept form y=mx +c. So, identify this equation in the form of y=mx +c so if you look at this

3 12
equation, I should push this 4y to the right hand side that gives me y= x+ . So, my m should
4 4

3
be this is the answer. So, slope intercept form you have y is equal to 3 by 4 x plus 3 so the
4

3
slope is naturally , this easy is our calculation.
4

Now we have identified an algebraic object as a geometric object. Now let us see what we can do
further and we can actually verify this graphically you know although it may be correct it is
 3
always better to verify it graphically. So, slope is x intercepts should be - 4 and y intercept
4
should be 3 if you want to satisfy the equation of this line this should happen right.

So this is how we have drawn so the x-intercept is -4, y intercept is 3 and the line passes through
this. Now you pick for verification purposes you can pick any point on this line and you can put
the values of the coordinates into the equation of a line and verify that it will give you the value 0
that will be the identification that your answer is correct.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras – Chennai

Lecture-21
Equation of Parallel and Perpendicular Lines in General Form

(Refer Slide Time: 00:19)

Let us look at next example which is another application of a general form of equation of a line.
The example is stated in the form of a question that is if I have been given two lines
a1 x+b1 y +c1 =0, a2 x+b2 y+c2 =0, b1 , b2 ≠ 0. What does this mean? That means the lines are
non-vertical. b1 ,b2 ≠ 0 means the lines are non-vertical you can verify for yourself.

Now two such lines are parallel if a1 b2 =a2 b1and perpendicular if a1 a2 +b1 b2 =0. This is an
interesting application of general form of equation of line. And if you recollect, we have derived
some characterization of line in terms of slope. So, let us try to see this problem so let me first
identify if I want to characterize parallel and perpendicular lines what should I do?

What is a parallel line, how will I identify a parallel line when I will have their slopes to be equal
and how will I identify a perpendicular line, when the product of the slopes of the two lines is -1?
So, if you remember this then the job reduces to finding the slopes of the two lines. Can I find a
slope of these lines? Let us first consider this line a1 x+b1 y +c1 =0. You should be immediately
able to identify this with slope point form which is y=mx +c.

So if I want to adjust this equation in the form of y=mx +c then what should I do? Because b1 is
nonzero I can divide throughout by b1 and shift this coordinate of y to their right-hand side of the

−a1 c1 −a1
equation. So, I will get y= − . So what is the slope . A similar trick you can apply
b1 b1 b1

−a2
here and therefore you will get m 2= . So using slope intercept form you have got
b2

−a1 −a2
m 1= ∧m2 = .
b1 b2

Now let us recollect the famous fact because b1 ∧b2 are not equal to 0 we are not considering
vertical lines. So, two non-vertical lines are parallel if and only if their slopes are equal. So, what
you will do you will just put m 1=m2because you have been given that the lines are parallel. So if

a1 a 2
you put m 1=m2 , minus sign will cancel each other = . Multiply both sides by b1 b2, b1 , b2 are
b1 b 2
nonzero.

So multiply both sides by b1 b2, you will get a1 b2 =a2 b1. Therefore, the lines are parallel then
a1 b2 =a2 b1. In a similar manner we also know something about perpendicular lines that the
product of their slopes is -1, if the lines are perpendicular. So, just multiply m 1 ,m 2 and equated to

a1 a2
-1. Minus sign will cancel each other so you will get × =−1.
b1 b2
So take the denominator on the right hand side that is b1 b2, so a1 a2 =−b1 b2 which essentially
means a1 a2 +b1 b2 =0. Therefore, we have proved the result. So, now what we have done right
now is we have related our result about the characterization of perpendicular and parallel line via
slope to a general form of equation and this is the new condition that we are coming up with if
the lines are parallel and you have been given to two non -vertical lines and their general forms
then you just need to check that a1 b2 =a2 b1 for the lines to be parallel and a1 a2 +b1 b2 =0 for the
lines to be perpendicular. This you can consider as another characterization of parallel and
perpendicular lines using a general form of the equation of lines.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras – Chennai

Lecture-22
Equation of a Perpendicular Line Passing Through a Point

(Refer Slide Time: 00:15)

So, now you have been presented with an equation of line and a point and the question is you
find the equation of a line perpendicular to the line x−2 y +3=0 and passing through the point
(−1, 2). So, in this case let us identify the general form of the equation that is Ax+ By +C=0 and
you can easily see that A=1, B=−2 , C=3.

−A −1
Therefore, the slope of the given line the line that is given to you will be which is that
B −2

1 1
will be . So, the slope of the given line m 1= , now if at all a line is perpendicular to it then
2 2
you already know that the product of the slopes is -1. So if the product of the slopes is -1 then

−1 1
m 1 m2 =−1 that is m 2= . So m 1 is which will give me m 2=−2.
m1 2
So now the problem reduces to the slope of a given line is -2 and it passes through point (−1, 2)
and I want to find the equation of a line that is passing through point (−1, 2) and has slope -2.
So, use the slope point form y− y 0 =m ( x−x 0 ), y 0 is 2 so you can easily see y−2=−2 ( x+1 ), ¿,
so ( x +1 ) , rearrange the terms so this 2 will get cancelled constant therefore I will get the equation
of a line to be y=−2 x or in a general form you can write this as −2 x + y=0.

So, let us try to figure out whether the line which we have actually found is perpendicular or not.
So, the orange line is the line for which the equation is given x−2 y +3=0 the point (−1, 2) is
displayed in the graph and the line passing through it is also displayed and you can clearly see
the angle that is made is 90 degrees therefore the lines are perpendicular and our answer is
correct. So, our verification test has passed.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras – Chennai

Lecture-23
Distance of a point from the given line

(Refer Slide Time: 00:16)

So, we have verified the answer now what you can see here is (-1, 2) is a point which is lying on
the line which is perpendicular to the given line. An interesting question can be asked that what
is the distance of this; point from the given line. Let us try to answer that question in the next
slide.
(Refer Slide Time: 00:40)
So, the question is given any point I have another line the point is not collinear to the line then
what is the distance of that point from a line. So, let us take that as our goal. To be precise we are
interested in finding the distance of the point P which has coordinates ( x1 , y 1 ) of this point from
the line l which has equation Ax+ By +C=0 and this is a general form of equation.

Now how will we proceed? So, if I want to understand the location of P, I need to do some
analysis because Ax+ By +C=0 is a completely geometric object. So, I need to understand this
line in terms of its geometric concepts. So, what are the geometric concepts associated with this
line they are slope x intercept y intercept or points on this particular line. So, let us identify those
things first.

So if we assume that A and B both are not equal to 0 then I can rewrite this equation in the form

x y −C
of intercept form that is + =1 so in that case my a is actually that is x intercept and my
a b A

−C
small b will be which is y intercept. So, I have identified this line as I have identified the 2
B
points and these 2 points uniquely determine the line. So, I know how the line is located.
Let us try to visualize this line in terms of the graph of a function. So, as you can see I have

−C −C
mentioned that x intercept is
A
so it is mentioned as a point Q which is
A( )
, 0 , y intercept is

−C C
B ( )
which is identified here so 0 ,− , the point P is located here it may be located anywhere
A

but right now the point P is located here it has coordinates ( x1 , y 1 ). So, now I want to identify a
distance of this point from this line, the line joining the points Q and R.

So, what is the distance? It should be the shortest distance from the line, so the shortest distance
in this case if you move along this line the shortest distance in this case is a point where the point
is actually perpendicular to the line. So, what I want to say is the shortest distance is the one
which is the perpendicular distance. So, the entire question reduces to how to find this
perpendicular distance PM.

So, let us try to see what are the geometric objects associated with this. So, you can see from the
dotted lines the geometric object that I can associate with this particular distance is a triangle
PQR. Now if I want to find the distance PM, I can take help of this triangle PQR so that I will be
able to find the distance PM. So, how will I do that? First you see if I want to compute the area
of triangle PQR what do I need to know?

I need to know the base and the height and the area of a triangle is half base into height. So, half

1
base into height means ×QR × PM . I do not know what is PM. But we have already seen in
2
this course how to find area of a triangle where its coordinates are given. So, even though I do
not know what is PM I know how to compute the area of a triangle. The next question is do I
know how to compute the length QR?

Yes of course because this is x intercept this is y intercept and these are the lines which are the
distances on x and y axis and all of them form a right-angled triangle. So, by Pythagorean
theorem I will be able to find the length of QR. So, I can reformulate the question as
 A ( ∆ PQR ) . So, now I know how to compute the length PM if I know how to compute
PM =2×
QR
area of triangle and how to compute the length of line segment QR both of which I know.

So, let us go ahead and try to compute area of triangle PQR. So, here is our formula for area of
triangle PQR which has coordinates ( x1 , y 1 ) , ( x2 , y2 ) ,∧( x 3 , y 3 ). So, let us start with ( x1 , y 1 ), the

( x1 , y 1 ) is the first coordinate remember you will always take this in anti-clockwise direction. So,
I will start with this coordinate then I will go to R and then I will go to Q. So, this is ( x1 , y 1 ) this

is ( x2 , y 2 ) and this is ( x3 , y 3 ) according to the notation that is given in the formula.

So you will see x1 ( y 2 − y3 ) so x 1 is first coordinate it will remain x 1 because P has coordinate

−C −C
( x1 , y 1 ), y 2 is
B (
, y 3 is zero. So, you will get x1
B )
−0 then the next term that is x2 , x2 here is

−C
0 so this entire thing vanishes then you go to x 3 , what is x3 ? x3 is , into y 1, which is y 1 as it
A

is − ( −CB ) so ( y + CB ), this is how I got the formula.

1

So, if you look at this formula closely you can actually take C common from all within the mod

sign so you can take |CB |, denominator has terms containing B and AB. So, if you want to take
those terms out you multiply throughout by AB or you find the LCM is AB and you take AB out

1 |C|
so you will get × ×|A x1 + B y 1 +C|, remember this is the term corresponding to general
2 |B|
form of the equation.

Now we have seen how to compute area of triangle PQR. Next, we will see how to compute the
length QR. But length QR is actually very easy because I have a point Q which has only x-
coordinate and I have a point R which has only y-coordinate. So, it will be as if computing the
 2 2
C C these are the 2 sides of the triangle and this QR is the
distance of length QR is
√ 2+ 2
A B
hypotenuse of that right-angle triangle.

So, again you can simplify this to amend to this form so you can take out C common so you will
get |C|, you take A and B common you will get |AB| and then you will get √ A 2 +B2 which is
which is in the numerator and now if you look at this form PM which is the length of the line

A ( ∆ PQR )
segment PM is 2× , so just now it is just a matter of feeding the values this half will
QR
get cancelled with this 2 and area of triangle PQR is this and QR is this therefore this constants
also will vanish because they are same.

|A x 1 +B y 1 +C|
And you will get the formula to be equal to this is how you will calculate a
√ A 2 +B2
perpendicular distance of a point from a line. Now this idea can be helpful in finding one more
thing that is a distance between two parallel lines.
(Refer Slide Time: 10:22)

So, the question can be asked is I have two parallel lines what is the distance between two
parallel lines. So, let us take the set up because the lines are parallel l1 and l2 they have common
slope or the same slope, so their slope is m. Then you can use the slope point form which is
y=mx +c 1. Now I want to use the previous concept that I have introduced distance of a point
from a line. So, I will first identify this with x intercept. So, what will be the x intercept in this
case?

If you identify this line it is very easy to see go back to the general form and figure out that x-

−c1 −c1
intercept is , because B=1 here B=−m and C=−c1 , so the intercept is this . Let us take
m m
another line that is l2 it has same slope identify it with our standard form
A=−m, B=1 ,∧C=– c2. So, now given x-intercept what are the coordinates of this x-intercept

−c1 .
( , 0)
m

So now the problem reduces to finding the distance of this point from this line ok. So, by using

−c1
the distance of a point from a line formula where the point is ( , 0), you just need to substitute
m

|A x 1 +B y 1 +C|
this point ( x1 , y 1 ) into this formula for the distance of a line which is given as .
√ A 2 +B2
c1
So, my point x1 is− substituted here, y 1 is 0 substituted here you will get the formula to be
m

|C1 −C2|
equal to 2 2
so in this case √ A 2 +B2, B was 1, A is -m so it is √ 1+m 2.
√ A +B

Now you can actually identify this formula in the general equation form also. So, in the general

−C 1 −C 2
form instead of B = 1 we have slope which is equal to -A/B and c 1= and c 2= . So, this
B B
I am matching with both equations in general form these are slope point forms but now if you
match these equations with a general form you will get this description of the line where you
have Ax+ By +C 1 as one line is equal to 0 as one equation of line.
Ax+ By +C 2 =0 as equation of the second line. So, in that case this is the form and therefore now
you just substitute these values into this expression. So, this m will be replaced by - A/B so you
will get √ A 2 +B2 here and some |AB| will come out common and therefore finally that will

|C1 −C2|
cancel off with this B and you will get the expression of the form ,C 1 ∧C 2 belong to
√ A2 + B2
general form of equation.

So, this gives us a clear-cut understanding of the interconnection between the slope point form
and general form of equation and we have figured out what is a distance between two parallel
lines using distance of a point from a line formula.
(Refer Slide Time: 14:42)

So, now we will solve some examples to concretize the concepts so here are the examples in line.
So, you have been asked to find a distance of a point (3,−5) from the line 3 x−4 y−26=0. So,

|A x 1 +B y 1 +C|
in this case you just need to apply the formula, what is a formula, . So, what is A,
√ A 2 +B2
B, and C here is the key question? ( x1 , y 1 ) is known to be (3,−5). So, A is 3, B is -4, C is -26
then you just need to apply that formula which will give the denominator square root of 25 it will
give me 5 the numerator will be 3.
In a similar manner you can ask a question what is the distance between two parallel lines

|C1 −C2|
3 x−4 y+7=0, 3 x−4 y+5=0. So, what is a formula that we have derived it is it is
√ A2 + B2
very straightforward. So, what is C 1 here? C 1 the first line the constant term is 7 the second line

7−5
the constant term is 5. So, it will be modulus of 2 2 . So, you will get the answer to be 2 /
√ 3 + (−4 )
5.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras – Chennai

Lecture-24
Straight Line Fit

Welcome friends welcome back so far what we have seen is a distance of a line from a point
distance between two parallel lines. But the question now we can ask is, is that the only distance
that we can seek as a distance of a point from the line. To demonstrate this let me give you one
example where the paradigm will change as we will compare several points set of points and we
will compare the distance from those set of points to the line and the paradigm change that I want
to say is you will think differently how the distance will change from a line.

So, let us take one simple example this example is related to a small experiment that you might
have conducted in your lab.
(Refer Slide Time: 01:16)

It is a physics experiment which says V =IR that is voltage is equal to current times the
resistance. Voltage is equal to current times the resistance you all know this is a law this is the
law of physics where voltage is measured in volts current is measured in amperes and resistance
in ohms. Now the experiment that a physics teacher asked you to conduct is you have to verify
this law or using this law can you compute the resistance of a particular equipment.

So now what you will do is you will actually relate this with our equation of a straight line. So, if
I want to relate this with the equation of a straight line then what will happen you see V is
voltage so on the right hand side you can replace this voltage by y then the current that is
delivered to the circuit or the equipment you can denote it by say x and you want to determine
the resistance which is an unknown so you can put it as m.

And what is the constant? The constant is 0, so you can relate this with the equation y=mx ,
where y is the voltage, x is the current and m is the resistance and the whole purpose is to
determine this resistance over here m. So, the setup is ready the lab technician has arranged a set
up and you just have to go and perform the experiment and verify this phenomenon. So, the
catch over here is you want to determine what is a resistance.

So, the lab technician was very kind he has given you a priori information that there are only two
kinds of resistors our lab has one has a resistance of 1 ohm another one has a resistance of 2
ohms. This is the information that is given to you. Also notice the fact that this line is passing
through the origin that means (0, 0) is one point why (0, 0) should be a one point because there is
no current then there is no voltage this is our assumption.

So (0, 0) is one point and this line is passing through the origin so if I look at a mathematical
theory that I have studied so far I can safely assume if I get one reading if I get one reading from
that circuit that will help me in understanding the behavior and I can safely go and tell what is
the resistance of this particular equipment. Let us try to see how this assumption works out over
here.

Now this is the data you have conducted some experiments you have observed some data so it is
like you have passed a current of 1 ampere and you received the output of 2 volts here you can
say you have passed the current of 5 amperes and you have received output which is 4 volts and
so on and so forth. So, this is how it is working on. Now we want to identify what is the correct
line that will fit because I know from theory that this is a line passing through the origin.

So in particular if I tell you this line which is (1, 2) and (0, 0) then I will get the equation of line
using a slope point form or point - point form we also know that the intercept is (0, 0) so slope
intercept form y=mx +c, where c is 0 you can easily see the line that passes through this point is
y=2 x. But with the same register you also got these readings. So, let us see based on the lab
technicians’ knowledge if we draw two lines, they will be seen they will be visible like this,
interesting.

So, if I take only one observation and stop my experiment, I will get the line y=2 x. But if I go
for more experimentation then I am getting a line which seems to be similar to y=x. Now what
is it that is happening here, which line is a better fit. So, I need to answer this question because
this line actually passes through the point (0, 0) and y=2 x, this line is not passing through any
of the points.

So which line is better that is a natural question that comes to our mind? So, we will try to
answer this question mathematically. So, how will I answer this question mathematically? Let us
zoom in and consider our notion of a perpendicular distance. What is a perpendicular distance?
You will actually drop a perpendicular from this point to this point and you will compute the
distance of a line. Is that distance a correct distance? Geometrically it is a correct distance that is
a distance of a line.

But in this context that we are taking real-world context what is happening here is if I pass a
current of let us say 7 amperes this particular line is saying I should get a voltage of 7 volts but
actually I got a voltage of 8 volts. So, now I may not be interested if I drop a perpendicular from
this point to this point because this line is y=x it may cross this line at point 7.5. I am not
interested what is the value of y at point 7.5.

I should be interested in what is the value of y at point 7 because I have passed the current of 7
amperes not 7.5 amperes. So, the perspective of distance changes here because I want to find the
distance for this particular value of x from the line and the point how to go about then we will
not consider a perpendicular distance. This is a paradigm shift that I was talking about at the
beginning of the video.

So now I will not consider this thing but I will consider this distance that is a vertical distance the
distance that is parallel to y axis that is what I will consider. So, once I consider the distance that
is parallel to y axis, I have to consider these distances. So, again coming back to the question
which line is the best-fit line I can consider similar distances over here. And I can consider
similar distances over the blue line.

So which line is the best fit? We will try to answer this question mathematically. So,
mathematically we have seen that perpendicular distance will not fetch me any result directly.
So, I need to consider the distances that are parallel to y axis.
(Refer Slide Time: 08:57)

So, let us formalize this in a real term, this is the data that was shown in the picture. So, for 1
ampere you have got 2 volts current. For 5 ampere you got 4 volts current, for 7 you got 8, 8 you
got 9, 9 you got 8.7 and 10 you got 9. So, there is no direct relation between y and x; you cannot
figure out the y=x is visible over here but something is there which is making that line pass very
close to all these points.
This is the this is the demonstration, so y=2 x is way apart and we are assuming that the
hypothesis given by the lab technician is correct. So, I want to mathematically formulate this
problem. There are two lines y=x∧ y=2 x both pass through the origin, so current 0 voltage 0
hypothesis is correct. Now you have the set of observations xi’s and yi’s. I want to compute
which line is better.

So, let us try to see if I consider the sum of the differences, what do I mean by some of the
differences? If I consider y=x is a valid equation of line then I will consider y i−x i, that is the
distance between the line y and x because here y is equal to x if I input xi my point that I will get
is also xi because y=x i and the actual output that I have got is y i so I will consider y i−x i as one
coordinate and y i−2 xi as another difference that will be a point over here y i−2 xi it will be a
point over here.

But if I just consider the differences the problem is the differences may cancel each other some
differences may be positive some differences may be negative. so, I do not want those
differences to cancel out each other so what I will do is I will take square of them. So, in

6 6
particular we can define the sum square difference that is ∑ ( yi −xi ) 2∧∑ ( yi−2 xi ) 2 .
i=1 i =1

Now what this difference is calculating? It is calculating the difference between y i∧x i in the first
case and y i∧2 x iin the second case that is the error that we have made when we actually saw the
output on the error the equipment has made or in our recording the error which is made in
2 2
whatever way that is the error made. So, ( yi −xi ) and ( yi −2 xi ) right. Now what do you think
which one will be better the one that will be better which will have a least difference.

So, you can actually put in these values and compute these differences and square them sum over
them you will get the first difference is 5.09 and the second difference is 328.49. In this situation
what should be our conclusion? Our conclusion should be that the difference where the
difference is least that is 5.09 this must be a better line as compared to this line that essentially
reduces to a conclusion that y is equal to x is a better line as compared to y is equal to 2x which
is pretty evident intuitive from the figure as well.

So, you can see this figure you can see this chunk of points that are located around y=x and
therefore the resistance of the equipment that is given to us must be 1 ohm that should be our
conclusion. So, I want to introduce a notion of this kind to handle the real-world problems. So,
let us see what is that notion? In this case you were very lucky the lab technician has given you
the set of points or the resistance values there are only two resistance values.
(Refer Slide Time: 13:49)

But real life is not that lucky, so there they may not give you the set of values, and you want to
find out what is the best line that is passing through these set of points. In that case this notion of
a distance of a set of points from a line may help. So, what is this notion? First of all, we know
one notion is perpendicular distance but that perpendicular distance may not be of much use
when we are coming to the real-world perspective.

In that case we talk about the distances that are parallel to y axis from the distance of a points

that are parallel to y axis. So, in particular if you have been given n points { ( x i , yi )|i=1, 2 , … , n}.
You just plot this equation y=mx +c. Now remember here this equation is valid when it is not a
vertical line. If it is a vertical line this equation is not valid. And if it is a vertical line you do not
need such a complicated procedure to estimate it.
So y=mx +c is our standard equation of line which is a slope point form or slope-intercept form
to be precise and then as in the previous case we have defined the squared sum of the distance of
the set of points from the line. So, in the previous case y i−x i but in this case what should it be

n
2 2
( yi −m x i−c ) and you have to sum over all of them. So, ∑ ( yi −m x i−c ) .
i=1

So, we call this as sum squared error or some squared distance, sum squared error so the
abbreviation is SSE.
n
2
SSE=∑ ( yi −m x i −c )
i =1

Now the fact is when we are handling a general problem, we do not know what will be m and
what will be c.
(Refer Slide Time: 16:08)

So, our goal should be if I want to find the best line, I want to find the best line passing through
this point what should be my goal. So, these raises two questions if I have some square, I want to
know the value of m I want to know the value of c? So, given the set of points how to find a line
that fits the given set of points remember now I am not uniquely determining the line I am saying
but that fits the given set of points.
The line may not pass through any of the points in this particular case in other words y=mx +c
so what is the equation of the line that best fits the given set of points. This will mean I need to
find an equation of a line y=mx +c and then the question can be reframed into two questions that
is what do I mean by the value of m and c that best fits the line and then I have to define what is
the best fit according to me.

Obviously, the best fit according to me will be the sum squared error minimization. And so, if I
define SSE in this manner then I want to find the values of m and c that minimize SSE but this is
right now beyond our scope as so far, we have handled only linear terms. But if you look at these
terms, they appear to be in the form of squares of something. So, we need to divide some
strategies in order to find this minimization for m and c so with that we will see in few upcoming
videos of the course, thank you.
 Mathematics for Data Science 1
Week-03 Tutorial - Point of Intersection of two lines
(Refer Slide Time: 0:14)

Hello mathematics students. In this tutorial, we are going to learn to find the point of
intersection of two given lines. So, you have two-line equations given to you. Let us call one
𝑎𝑎1 𝑥𝑥 + 𝑏𝑏1 𝑦𝑦 + 𝑐𝑐1 = 0, let this be line l1, and line l2 is a 𝑎𝑎2 𝑥𝑥 + 𝑏𝑏2 𝑦𝑦 + 𝑐𝑐2 = 0. And we try to find
out the point at which these two lines intersect. And that would basically be the solution the
(𝑥𝑥, 𝑦𝑦)which satisfies l1 and l2 as well. It is easier to observe this process with example. So,
let us take 2 example lines and find out where they intersect.

So, for our examples, let us take l1 is 2𝑥𝑥 + 3𝑦𝑦 − 12 = 0, whereas 5𝑥𝑥 − 10𝑦𝑦 + 5 = 0. So,
when we have these 2 line equations, how do we solve for x and y. So, the best thing to do is
to eliminate one variable, either x or y and get a single equation in the other variable. So, what
I mean by that, and this could be done in 2 ways. One way is called substitution. In substitution,
in order to remove one variable, we basically express the other in terms of it.

For example, if I wanted to eliminate the 𝑦𝑦 variable, what I do is I express 𝑥𝑥 in terms of 𝑦𝑦. So,
I get all externs on 1 side, so 2𝑥𝑥 is on one side, and the other terms non 𝑥𝑥 terms on the other
12−3𝑦𝑦
side, which will give me 12 − 3𝑦𝑦. This would then indicate that 𝑥𝑥 is 2
, and then I take this

representation of 𝑥𝑥 in terms of 𝑦𝑦, and substitute it into this equation. What that gives us is,
12−3𝑦𝑦
suppose I substituted it, now I will get 5 � 2
� − 10𝑦𝑦 + 5 = 0.
 15𝑦𝑦
So we get 30 − 2
− 10𝑦𝑦 + 5 = 0. That is essentially taking the 𝑦𝑦 common I am going to get
15𝑦𝑦 𝑦𝑦
− 2
+ 35 = 0, canceling off the 35, so I get 1 here, 1 here, that would indicate 2
= 1, this

implies 𝑦𝑦 = 2. So because we eliminated the 𝑥𝑥 here, we got an equation which is entirely in 𝑦𝑦,
which lets us solve for 𝑦𝑦, and we get the value of y.

Now, to obtain 𝑥𝑥, we simply have to substitute this value of 𝑦𝑦 in this representation of 𝑥𝑥, so we
12−(6)
will 𝑥𝑥 = 2
= 3. Which means the solution for these 2 line equations is (3, 2), 𝑥𝑥 = 3 and

𝑥𝑥 = 2. And we can verify this quite immediately by substituting these values into the equations,
I will get 2(2) + 3(1) − 12 = 0. Likewise, 5(3) − 10(2) + 5 = 0. So it is fairly clear that
(3, 2) is the solution which satisfies both linear equations.

(Refer Slide Time: 4:43)

Another method of doing the same thing, which is to solve these 2 equations, we call it
elimination. And in elimination, what we do is we again, take these 2 equations, which is2𝑥𝑥 +
3𝑦𝑦 − 12 = 0, and 5𝑥𝑥 − 10𝑦𝑦 + 5 = 0. We again choose to eliminate either of these variables,
because we earlier eliminated 𝑥𝑥 and got an equation in 𝑦𝑦, now I am going to eliminate 𝑦𝑦 and
get an equation 𝑥𝑥. And for that, what we do is, we multiply this entire equation by the 𝑦𝑦
coefficient in this equation, which is minus 10.

So, I am going to multiply this whole thing with minus 10. And we multiply this entire equation
with the 𝑦𝑦 coefficient here in the other equation, that is 3. What that will give us is this would
give us minus −20𝑥𝑥 − 30𝑦𝑦 + 120 = 0. And this gives us 15𝑥𝑥 − 30𝑦𝑦 + 15 = 0. And now
what is to be observed is this is −30𝑦𝑦 and this is also −30𝑦𝑦, because here we multiply 3 with
−10 and here we multiplied −10 with 3.

And that lets us cancel these off, if I subtracted this whole equation from the previous one now.
So that will result in −30𝑦𝑦 by −30𝑦𝑦 getting canceled, and here, I will get −35𝑥𝑥 + 108 = 0.
105
And this would indicate that 𝑥𝑥 = 35
= 3. And now I can substitute 𝑥𝑥 = 3 in either of those

equations. If I substituted in the second one, I would get 5(3) − 10𝑦𝑦 + 5 = 0, this indicates
20
15 + 5 = 10𝑦𝑦, which gives us 𝑦𝑦 = 10 = 2. So, we got our value back, the point back, which

is (3, 2). This is the point of intersection of these 2 lines.

(Refer Slide Time: 7:41)

So, if we plotted these, these are our line equations, let us take the first one, I will reduce this
𝑥𝑥 𝑦𝑦
to intercept form, which will have to be to 2𝑥𝑥 + 3𝑦𝑦 = 12 is going to give us 6
+ 4 = 1. So, 𝑥𝑥

intercept is going to be 6, this and the 𝑦𝑦 intercept is going to be 4, which is this and so our line
is this is our l1. Now, if we try to plot the other equation, here, again, I will get 5𝑥𝑥 − 10𝑦𝑦 +
𝑥𝑥 𝑦𝑦
5 = 0, −1 + 1� = 1.
2

So, here we have this is the 𝑥𝑥 intercept, whereas this is the 𝑦𝑦 intercept 0.5 here. So, this is our
line equation 2. And clearly the intersection is happening here at this point, which is you can
see this is (3, 2). So, in this way, you can try to find the point of intersection of any 2 given
lines. However, you are likely to run into a bit of trouble in 2 cases, and let us see those 2 cases.

(Refer Slide Time: 9:21)

Consider these 2 line equations, 11 is still 2𝑥𝑥 + 3𝑦𝑦 − 12 = 0, whereas 5𝑥𝑥 − 7.5𝑦𝑦 + 10 = 0.
If we try to solve this using the substitution method, for example, we would get, I would, let us
say I try to eliminate the variable 𝑥𝑥 in which case I should be doing to 2𝑥𝑥 − 12 = −3𝑦𝑦, which
12−2𝑥𝑥 2𝑥𝑥
would indicate 𝑦𝑦 = 3
=4− 3
. And substituting this in l2, I will get from l2, this is from

l1.

15 2𝑥𝑥
And now in l2, if I substituted this, I would get 5𝑥𝑥 + 2
�4 − 3
� + 10 = 0. This gives us 5𝑥𝑥 +

30 − 5𝑥𝑥 + 10 = 0. And you see that 5𝑥𝑥 and −5𝑥𝑥 cancels and we come at the strange
contradiction where 40 = 0. And this is not okay right. We know that 40 ≠ 0. So, there is
some contradiction we are arriving at.

And what does this contradiction indicate? It indicates that there is no point for which these 2
lines meet. So, you cannot find a point of intersection for these 2 lines. So why is that? That is
because they are parallel. If we plotted these lines,

(Refer Slide Time: 11:51)

We know that for l1, the intercepts are 6 and 4, respectively. So this is 1, 2, 3, 4, 5, 6. So this
is our intercept for l1, 𝑥𝑥 intercept for l1 and 𝑦𝑦 intercept for l1 is 4 1 ,2, 3 and 4. For l2, we have
𝑥𝑥 𝑦𝑦
to see now for l2, we get 5𝑥𝑥 − 7.5𝑦𝑦 + 10, which indicates (−2) + (−4/3) = 1.

4
So, in 𝑥𝑥 = −2, so this would be our point and in 𝑥𝑥 = − 3is a little below −1, which is about

one third the way from −1 and −2. So, this would be it. If we plotted these lines now we see
that these are, in fact, parallel lines. They just do not meet anywhere, which is why when you
try to solve for a point of intersection, you get a contradiction. So here, we can say that there
is no solution for this system of linear equations.
(Refer Slide Time: 13:30)

Now, in the third case, let us look at a line equation which is our l2 earlier that was 5𝑥𝑥 − 10𝑦𝑦 +
5 = 0. And there is some other equation l4 let us call it, which is 25𝑥𝑥 − 50𝑦𝑦 + 25 = 0. So,
when we solve for these 2 equations, now let me try the elimination method. So, I am going to
get 2 equations, then one is 125𝑥𝑥 − 250𝑦𝑦 + 125 = 0. And here I am going to get another one,
125𝑥𝑥 − 250𝑦𝑦 + 125 = 0.

We have the same coefficient for y. So if I attempted to subtract this equation entirely, I will
get 0. So, I have this statement, which is always true. Unlike the previous case where it was
never true, 40 was never going to be equal to 0, here I get a statement, which is always true,
which is 0 = 0, independent of the coordinates of 𝑥𝑥 and 𝑦𝑦.

And this means something similar to the previous case, but not exactly the same. What is
happening here is since this is always true, it means there are infinite solutions for these 2
equations. If you observe what is actually happening is l2 and l4 are the same line, which is
why we got this entirely identical equations, both of these, let us call this equation 5 and let us
call this equation 6. And we see that equation 5 and equation 6 are the same, there is no
difference, which means our 2 original lines are coinciding.

If they are the same line, then we will get infinitely many points which satisfy both of them.
So we have infinitely many solutions for these 2 lines. So whatever 𝑥𝑥 you take, you are going
to get a solution for that 𝑥𝑥. So in the graph, this is what is going to look like.

(Refer Slide Time: 16:28)

We know the intercepts of our l2, which is −1 and 𝑦𝑦 intercept was half, so this would be our
l1, it is passing through (−1, 0), and also (0, 1/2). And as we had found earlier, it is passing
through (3, 2) as well. Now let us consider the other equation. Now let us consider the other
𝑥𝑥 𝑦𝑦
equation which is l4, and we will have 25𝑥𝑥 − 50𝑦𝑦 = −25. This gives us (−1) + (1/2) = 1.

So, again we get the same intercepts. Thus, l2 will have to coincide entirely with l1. And that
is what is happening, they are the same line. So, we get infinitely many solutions when we get
a true statement, an always true statement independent of 𝑥𝑥 and 𝑦𝑦 in case of the same line, that
is both line equations are representing the same line.
 Mathematics for Data Science 1
Week-03
Tutorial-01
(Refer Slide Time: 0:16)

Hello, mathematics students. This is a tutorial for week 3, where we will be doing more straight
line concepts problems. Primarily, this is the syllabus that has been covered here. Let us begin
with our first question.

(Refer Slide Time: 0:31)

There is a company with two kinds of equipment, A and B. And they have work lives of 3 years
and 4 years respectively. So, work life of A is, let us call it 𝑊 is 3, 𝑊 is 4 years. Further, the
values of equipment A and B decrease yearly according to these equations. These are our
equations, where 𝑣 is supposed to be the value of A and 𝑣 is supposed to be the value of B
in thousands, respectively, and x is the number of years for which that value is applicable.

So, what are the costs of the equipments? So, the cost of the equipments would be 𝑣 and 𝑣
values when x is equal to 0, that is, when you just bought it, what is the value of the equipment.
So, we just take x is equal to 0 and from this we get 0.5𝑣 - 62.5 = 0, this would give us 𝑣 is
equal to, to indicate that this is the initial time I am going to make it 𝐴 , 𝑣 so yes, this is 𝑣
and that is 62.5 /12.5, which is equal to 5.

Therefore, the cost of A, I will call it 𝐶 is rupees 5000. Now, let us work with B. Same thing
again, we take x is equal to 0. So, we have 12𝑣 - 72 = 0, this will imply 𝑣 again we are
calling 𝑣 to indicate the initial cost that would be 72 /12 which is equal to 6. So, 𝐶 , the
cost of B is rupees 6000. Going further, we are asked what are the yearly depreciations of the
two equipments.

So, yearly depreciation basically means how much value is decreasing each year. So, let us
look at that. Here, in this case, x is number of years, whereas y is the value. So, what is being
asked in a yearly depreciation is the change in y for a unit change in x, which is basically just
a slope. Because slope is changing y, ∆y by changing x. So, when ∆x is equal to 1, ∆y is equal
to the slope.

(Refer Slide Time: 3:56)

So, we can find this by just finding the slope for each of those two linear equations. And for
the slope, we convert our equations to the y = mx + c form, then the m is going to be the slope.
So, one equation is 5x + 12.5𝑣 - 62.5 = 0. This would indicate that 12.5𝑣 = -5 x + 62.5. Going
further then, it will have 𝑣 = -5x / 12.5 + 62.5 /12.5, we had already seen it to be equal to 5.

So, that is equal to - 0.4 x + 5 = 𝑣 . So here, we are, our m in the equation is basically -0.4. So,
this is the reduction in one year, -0.4 ×1000 because we are taking everything in thousands, so,
that is basically -400. So, this is the depreciation, 400 is the depreciation every year for the
company one, we can also verify this by looking at the values of 𝑣 for year one.

So, when x = 1 we have 5 +12.5𝑣 - 62.5 = 0, this gives us 𝑣 = 57.5 / 12.5 which is equal to
4.6. So, 𝑣 was originally 5, that means it was originally 5000 rupees and after 1 year it became
4.6 which is 4600 rupees. So, the difference is 400 rupees. So, that is the yearly depreciation
for the first equipment.

(Refer Slide Time: 6:48)

Now, let us look at the second equipment now second equipment the equation was 6x + 12𝑣
- 72 = 0. Again, if we put this to the y = mx+ c form, the slope intercept form we will be getting
first we have to do 12𝑣 = -6 x + 72. This indicates 𝑣 = -0.5 x + 6, thus -0.5 is the slope
here. Which means 500 rupees is the yearly depreciation.
(Refer Slide Time: 7:42)

In the last part, they said that the company will buy back the equipment after its work life. And
Vijay has a requirement of such equipment for 12 years. Which kind of equipment will cost
him lesser.

(Refer Slide Time: 7:58)

So, in the case of the first equipment, let us call it case A, and here let us have case B to
consider. And in case A the initial cost was 5000 rupees and each year there is a decrease of
400 rupees. So, in first year we lose this much, in the second year we lose another 400 rupees
and at the end of the third years, there is a loss of another 400 rupees. And we are aware that 3
years is a worklife for A, whereas for B it is 4 years. This is to say that at the end of 3 years,
the value of the machine is 3800 rupees.
So, if now, Vijay buys the equipment afresh, then and the company is buying back this 3800.
All that Vijay needs to spend now is rupees 1200 and this way he gets an additional 3 years.
So, with 5000 he got 3 years and now another 3 years this way. So, in order to get 12 years
with equipment A, the total money that Vijay will require to spend is 5000, which is the initial
first 3 years and from then on 3 years plus 3 years plus 3 years because it is totally 12 years.

So, 3 times 1200, that is rupees 8600 in case of A, whereas in B, B is more expensive. So, we
have 6000 and every year there is a loss of 500 rupees in value and this is required to be done
4 times because the work life for B is 4 times. So, we are effectively subtracting 2000 rupees
from the original value. So, we have 4000 at the end of it, which means for the first 4 years
there is an expenditure of 6000 but then, for the remaining 8 years, there has been only 2000
each.

This is so because the product's value is already 4000 rupees and in order to get a new version
of equipment B, Vijay only has to spend 2000 rupees. So, total expenditure in this case is going
to be 10,000 rupees because 6000 + 2000 + 2000. Here, we are not supposed to forget one thing
though, that is the end of, after these 3 years, 3 years pass, at the end of 12 years, he can sell it
off for 3800. So, we are supposed to further subtract 3800 here and likewise here, we can sell
it off for 4000. So, here we get rupees 4800 whereas, here we get rupees 6000. So, the
expenditure is clearly lesser for A. So, A would be the good choice for Vijay.
 Mathematics for Data Science 1
Week-03
Tutorial-02
(Refer Slide Time: 0:16)

And for our second question, there are 2 lines, and these are the equations, which represent
our lines. And a line l3 is parallel to l1 , and passes through (-5, 0). Now we can find l3 , by
using the point slope form, we already have the point, which is (-5, 0). And we can also find
the slope from l1 slope, we already have l1 . And we can write, so l1 is this,
6 x +12 y  72 = 0, which tells us that 12 y = 6 x + 72.

x 1 1
And that gives us y =  + 6 so, the slope here is  , because y = mx + c. So, slope is  .
2 2 2
y 0 1
Now if we did point slope form on this, we would get =  which indicates
x+5 2
2 y = x  5 . So therefore, x + 2 y + 5 = 0 is basically our line l3 . And now if we look further,

we have line l 4 which is passing through this point, and it is perpendicular to l 3 .

1
So, if we took this to be m1 =  = m3 because m1 and m3 are the same slope. And let us
2
consider the slope of l4 to be m4, so we can say m3  m4 = 1 , because they are
1
perpendicular, that would indicate m4 =  , which is basically 2. So we now have the
m3

slope of l4 . And it also goes through this point.

(Refer Slide Time: 2:57)

5
y+
So again, using point slope form, we have 2 = 2, that would indicate y = 2 x  5 So this
x 2.
is our l4 .

(Refer Slide Time: 3:14)

Now, the question is being asked is, what is the cardinality of A, which is a set of all points
common to at least 2 of the mentioned lines.
(Refer Slide Time: 3:28)

For that, let us try to draw our lines on the graph. 6 x +12 y  72 = 0 would give us if x = 0, it
gives us y = 6 which means some point let us call this here is (0,6), it goes through this point.

And if y = 0, you get x = 12 . So that would be some point here. So, our l1 is this line. And

now we know l3 is parallel to this line. So l3 , if we, again did the same thing of putting

y = 0, x becomes  5 , which is somewhere here.

So, as you can see, I am doing this on a rough estimate. I am not trying to be accurate, but
even a rough estimate can work out here, because you might not always find graph paper
when you require it. So often developing an intuition for the rough estimates is a good idea to
solve problems. Now, this is one point and when x = 0, y becomes -2.5, which is somewhere
like this. So we have (0, -2.5). As you can probably see from our last rough estimate itself
that these do appear to be parallel, they seem to be in the same direction.

Now, l 2 if we look into it with a similar logic, we can see that l 2 can be reduced to
y x
 = 1. So in our intercept form, we can now tell that if I made this plus, this becomes  5,
6 5
so the x intercept is -5, which is this point, again, and y intercept is 6, so that is this point. So,
l 2 , in fact, passes through these 2 points. So, this is our l 2 . So, this was l1 now, this is l3 and

this is l 2 .
Lastly, let us reduce our l4 into the intercept form, we get 4 x  2 y = 5, therefore,
x y
+ = 1. So, when we look at this then 5/4 is a quantity just a little greater than 1, so
5/ 4  5/ 2
it is probably somewhere here and 5/2 is a 2 and a half basically. So, -2.5, so this and this
plus we have something like this happening. So, overall there are four points, which are
common to any pair of these four lines.

(Refer Slide Time: 7:02)

So, our question, the cardinality of A, where A is a set of all points common to at least 2 of
the mentioned lines. So, that would be 4, there are 4 points of intersection here. Now, if R is a
relation, and it is the set of all points inside the region bounded by these 4 lines. So, here we
are, when we say relation, we are basically saying every point in the set when is taken as a
ordered pair like this (x, y), then x would be from the domain of the relationship and y would
be from the co-domain.

So, this is seen as a relation from the set of x values and to the set of y values. And now, we
are asked to find the range and domain of relation R, which is to basically find when we say
range, all the possible y values and the domain is all the possible x values.
(Refer Slide Time: 8:16)

So, here in this region that we are looking at, the possible y values would be between this
value and this value. So, all possible y values are between -2.5 and 6, whereas the possible x
values are between this point and this point, that is between -5 to some particular quantity,
which is the x coordinate of this point. And that point is the intersection of l1 and l4 . So, let

us try to solve l1 and l4 to find that point of intersection.

(Refer Slide Time: 9:03)

5
We know that this as l1 , and this is l4 and from l4 , we know that y is basically 2 x  . If
2
we substituted this into l1 we would get 6 x +122 x  5 / 2  72 = 0. This would give us
6 x + 24 x  30  72 = 0. That indicates 30 x = 102 which indicates x = 3.4. Correspondingly, y
would then be 2  3.4  2.5, because 5/2 is 2.5, which gives us 6.8 - 2.5, which is equal to 4.3.

(Refer Slide Time: 10:24)

So this point here is (3.4, 4.3) and we only require the x value. So the x values range from -5
to 3.4.

(Refer Slide Time: 10:42)

However, one important thing we need to look for here now is the region bounded by these 4
lines, but excluding the lines themselves.
(Refer Slide Time: 10:52)

Which means -2.5 and 6 themselves do not fall into our domain because we are not interested
in the points on the curve. So this point is on the curve, this point is on the curve, but it is not
inside, similarly, for each of these, because they are the border points. So, -5 is not an x value
inside the domain. Similarly, 3.4 is not a value inside the domain. So, our domain is the
5,3.4. Likewise, -2.5 is not a y value inside the range and 6 is also not a y value inside the
range, so our range is (-2.5, 6).

(Refer Slide Time: 11:50)

Lastly, there is a line l5 represented by this equation given to us find the cardinality of set B,

which has all the points common to l1 and l5 .

(Refer Slide Time: 12:04)

Let us look at l1 and l5 . l5 is given as x + 2 y = 12 . Now, if we applied our intercept form

again, we would get x / 12 + y / 6 = 1. Let us look at that x / 12 indicates x intercept of l 2 ,

y / 6 indicates y intercept of 6. So, we see that l5 is basically the same line as l1 , indeed if

you multiply this whole equation with 6, you will just get the form of l1. Therefore, l1 and l5
are the same lines.

(Refer Slide Time: 12:54)

Then, the question is asking, find the cardinality of set B, which has all the points common to
the lines l1 and l5 . There are infinite points because they are the same line. So, the
cardinality of set B is infinite.
 Mathematics for Data Science 1
Week-03
Tutorial-03
(Refer Slide Time: 0:14)

Now, third question, you have two friends Lincoln and Lila who purchase shares of two
companies. Lincoln purchases six shares of a company M and one share of company N and
overall spends 400. This can be encapsulated as if the company M's share price is Pm and for

n that is Pn we can say that 6Pm + Pn = 400. Then for Lila there is four shares of Company
M and three shares of Company N coming to 360.

So, for Lila we have 4Pm + 3Pn = 360. How much did each of them spend on n? So, we need

to know what is Pn and 3Pn , that is what we are interested in. To find the values of Pm and

Pn we will require to solve these two linear equations. However, we only required to find Pn
because the question is only pertaining to the company N's shares. So, we can work towards
eliminating the Pm variable from these two equations.

So, we can multiply this equation by 4 and this one by 6 because 4  6 = 24, 6  4 = 24 and
that way we should be able to subtract 24 Pm . So, we are going to get from the first equation

24Pm + 4Pn = 1600, whereas, from the second equation we get 24Pm +18Pn = 2160. Now, if
we subtract second equation from the first we get these two canceling off and here we get
 14Pn = 560.
And this indicates that Pn = 560 / 14 because we can cancel out the plus and the plus and that

is equal to 40. So, Pn is 40 rupees per share. And now since Lincoln has purchased only one
share, Lincoln spent only 40 rupees on company N, whereas, Lila spent three times that
which is rupees 120.
 Mathematics for Data Science 1
Week-03
Tutorial-04
(Refer Slide Time: 0:15)

For our fourth question, we want the equation of a line which is perpendicular to this line,
and is at this distance from the origin. So, from y  5x = 0, we get y = 5 x, so therefore, the

slope m1 is 5. And if our line is perpendicular to it, then our line m2 must be  1/ m1 , which
x
is equal to -1/5. So, we know that our line is some y =  + C.
5

If we kind of simplify it, we are going to get 5 y + x = C, this C is not the same thing as the
previous C, I have just used that as C because it is an arbitrary constant, which is yet to be
determined, otherwise it should have been 5C. Anyway, now we have to find the value of this
C in this equation. For that, we are going to use the next bit of information that is given to us,
which is the distance from the origin.

So, this line has this distance from the origin. So the distance from a point formula is
ax1 + by1 + c
, where (x1, y1) is the point from which we are measuring the distance for this
2 2
a +b

line. So, in our case, (x1, y1) is (0, 0) because we are doing from the origin. So in our case,

we get modulus of
 c , So, modulus of  c is just the same thing as c.
a2 + b2
And root of a 2 + b 2 , in our case comes out to be 25 +1, , that is 26 . So we have
c , this is given out to be
1
, which would imply c = 1 , and that would imply c ± 1. So,
26 26
we get two answers.

(Refer Slide Time: 2:49)

What are the two answers? One is for c being +1, we have 5 y + x = 1. And in the other case,

we get 5 y + x = 1. for the other choice. So, how does this happen, what is actually
happening here to try to plot our lines?

(Refer Slide Time: 3:08)

So, we have two lines, which we were looking at, which is 5 y + x = 1 and 5 y + x = 1 and
y x
from this we get the intercepts to be, the intercept form of this would be + = 1. And in
1/ 5 1
y x
this case, we get + = 1. So, in one case, we have a y intercept of 1/ 5. So, let us
 1/ 5  1
assume this is 1/5, then x intercept is 1 which is 5 times of that, so that so it must be
somewhere here, so this would be (1, 0) and this is (0, 1/5).

And our line is going through these two points, giving us something like this. Let us call this
1
l1 and where do we get the distance from the origin, we get it when we measure it
26
perpendicularly from the origin. Now, let us look at the other equation. So -1/5, so, this
should be exactly below this this way and this is -1, so this would be exactly opposite in this
way at the same distance.

So now we have these two points, so we can also construct this line, which goes this way.
1
And as you can see, they are both parallel and exactly opposite to that you get this
26
1
distance which is again perpendicular distance and it is also at . So, we have two lines
26
which satisfy our requirements, one is 5 y + x = 1, the other is 5 y + x = 1.
 Mathematics for Data Science 1
Week-03
Tutorial-05
(Refer Slide Time: 0:14)

In this problem, should be our fifth question. Suppose to find the area of ABC, there are three
points here. So, we need to make that triangle, our triangle would look something like this.
But to find the area, we are supposed to calculate a base and its corresponding height. So, we
are not supposed to use the formula which involves the three coordinates, instead, we will
take any of these sides to be the base. So, let me take AC to be the base. So, we need to find
the base length, which is AC, that would be by Euclidean distance formula,
√(4 − 6)2 + (2 − 7)2 .

So, this comes out to be √4 + 25. So, that gives us √29 is the base. Now the altitude, the
height from B would be something like this, let us call this point D and this is 90 degrees, so
B to D that length would be the height. So, BD is going to be the distance of the point B from
the line AC, the shortest distance of point B, the line AC. So, for this we can use the distance
formula of a point from a straight line. However, we first need to find out the equation of AC.

𝑦−7 7−2 5
For that, let us use the 2 point form because we have 2 points, we will get = = =
𝑥−6 6−4 2

2.5 . Anyway, if we cross multiply, we get 2y-14=5x-30, which gives us the equation to be
5x-2y-16=0 and the distance of (0,5) which is our B from this particular line. So, this line is
our 5x-2y-16=0 .
 |ax1 +by1 +c|
So, that distance can be calculated from the formula, which is the . So here a is our 5,
√𝑎 2 +b2

b is -2 and c is -16. So, substituting and ( x1,y1 )is our coordinates of B, this is x1 and this is
|0−10−16| |−26|
y1, so the coordinates of B. So here we get 25+4
, which then gives us , | − 26| is then
√29

26. So, this would be the height.

(Refer Slide Time: 3:48)

Combining these two quantities, we get our area as half into base into height, which will then
1 1 26
be 2 × AC × BD, which then gives us 2 × √29 × , √29 and √29 cancels off, 2 cancels with 26
√29

giving us 13. So, we get 13 square units as the area of our triangle ABC.
 Mathematics for Data Science 1
Week 03 – Tutorial 06
(Refer Slide Time: 00:14)

Sixth question. We have Junaid who is traveling on a road represented by the equation x+y-
𝑥 𝑦
10=0. So, in the graph if we plot that, we can see that x+y=10, which gives us 10
+ 10 = 1,

which means the x intercept and y intercept are both equal to 10. So, if this is 10 and this is
also 10, so this would be (10,0). Whereas this is (0,10) and the line that passes through them
is the road that Junaid is traveling on.

So, this is the line that Junaid is traveling on and he calls Ravi, asking him to meet on the
same road. But Ravi is at this point. So, that would be 5, it would be somewhere here halfway
and 1 would be somewhere here. So, this is 1 and this would become our location of Ravi,
that is (5,1) and Ravi wishes to cover the minimum distance to Junaid’s road. So, we know
that minimum distance is achieved when you go perpendicular that is normal to the other line.

So, we can see that Ravi goes along this path and intersects, that path intersects somewhere
over there, let us call this point P and he arrives at this point P in 2 minutes and we are being
asked, what is Ravi’s speed? So, we need to first find out, assuming Ravi’s original location
(5,1) is R, if we find out what RP is, then we should be able to find out the speed. So, RP is
basically the shortest distance of R from this particular line. So, we can calculate it from that
|ax1 +by1 +c|
formula, which is .
√𝑎 2 +b2
 |5+1−10|
And now, here a is 1, b is also 1, c is -10. So, we have this is equal to , because a is 1
√1+1
|−4| 4
and b is 1, the squares are also 1. So, we get = , which is then equal to 2√2 units and
√2 √2

now it is given that 1 unit is equivalent to √2 kilometres.

So, that means the distance in kilometre RP in km is equal to 2√2 × √2, that is 4 km and Ravi
2 1
has taken 2 minutes. If we write it in hours, t in hours is then 60 that is 30hours. So, the speed
4
would be distance by time, which is 1 which gives us 120 kmph, kilometre per hour. This is
30

the speed that Ravi travels at.

 Mathematics for Data Science 1
Week 03 – Tutorial 07
(Refer Slide Time: 0:14)
In our seventh question, we have this interesting thing, where there are two anthropology
students and they are calculating the relationship between the length of the femur and the
height of a female adult using fossilised bones. So, what is exactly happening here? What
is femur?

From Wikipedia, we can see that the femur is the thigh bone, which is this particular bone.
So, what is happening is, in our question, there are fossilised bones and these anthropology
students, anthropologists try to study the nature of humans and their societies as they were
evolving.

So, here we have fossilised bones and suppose we have the femur of what we know to be a
female adult, then we are estimating the height of that female adult from the length of the
femur bone, from the thigh bone. So, it is given that this relationship is linear, we have
. Both use the data given below, so this is the data that is available. We have the
femur length and the height of the adult female.

So, from this we are trying to develop this model and Chetan has found ,
whereas Raju has calculated m to be 2.1 and n to be 72. So, both of them agree on n, this
parameter is already fixed. It is the m that we are trying to see, whose m is better. So, in terms
of linear equation, m is basically the slope of the line. So, how do we do this? We want to use
the concept of Sum Squared Error, which we call SSE.

So, in both cases, we are going to look at what is being predicted in terms of height and what
is the actual data. So, let s look at case one, let us look at Chetan’s case here and Raju’s case
here. In terms of Chetan’s case, we would have the , where m is 2, so we have
∑ and we sum it over how many items 1, 2, 3, 4.

So, let us call this and i goes from 1 to 4 and in case of Raju’s measurements, this error
would be again, i goes from 1 to 4 and we have∑ . So, I think we just
need to do the calculations now. So, let us look at this here, so this is case 1. So, this is case 2,
this is case 3, this is case 4, is 38 and is 147.

So, when we put in 38 here we get 2 times 38 is 76 + 72 - 147 the whole square and then in
case 2, we have 40 and 150 as and . So, we will get 80 + 72 - 150 the whole square and
then we have, is 42 and is 155. We have (84 + 72 – 155) 2 and lastly, we have is 44
and is 160. So, we have (88 + 72 – 160)2.

So, this is the total sum squared error for Chetan. Whereas in case of Raju, we would get 2.1
times the same thing. So, 2.1 times 38 is 79.8 + 72 - 147 the whole square and in case 2 we
get 84 + 72 - 150 the whole square + in 3 we get 88.2 + 72 - 155 the whole square and lastly,
in case 4, we have 92.4 + 72 - 160 the whole square. We calculate these values then we get 12
+ 2 2 + 1 2 + 0 2.

So, this is 6. So, sum square error for Chetan is 6. Whereas in Raju’s case we would have 4.8
2
+ 6 2 + 5.2 2 + 4.4 2. Now clearly, in this error, there is a 6 2 which has to be greater than 6,
which means Raju’s error is much more than Chetan’s error. Therefore, Chetan’s line fit is
better.
 Mathematics for Data Science 1
Week 03 – Tutorial 08
(Refer Slide Time: 00:14)

In our eighth question, we want the equation of a line which is parallel to this given line. So,
let us first plot the line that is given, lets plot this line. If you look at that line, it is
, which gives us . So, the x intercept and y intercepts can be marked out as

1.6s. So, if we take this to be 1 and this to be 2, so this is - 1 roughly and this is - 2, roughly.

Yeah, this might be our intercept, which is , 0) and our y intercept, again, if we take this to

be 1 and this to be 2, 1.25 is somewhere likely here.

So, this is probably our y intercept, (0, . As you can see, we are doing a thoroughly rough

plotting, we do not always have to be very accurate with our plotting. This is only for an
indication. So, this would be our line, and now we have another line given
to us, which is .

Clearly, these two lines are parallel to each other because they have the same slope and that
slope would be, we write it as y is equal to, we will get 3x / 4, so the slope is and it is

passing through the origin, because if I put x = 0 and y = 0, the line equation is satisfied, that
is there is no constant term. So, this line is our and we are trying to find a line
that is parallel to these 2 and it is at a distance of 1 unit from the line.

Let us name these lines as well. Let us call this and this is . I am going to erase the
intercepts to make it look a little clear and now, we can find our equation and for that, we will
use the formula of separation between 2 parallel lines. So, that two parallel lines and we write
them with the same coefficients.

So, and the other one would be . This is how two

parallel lines would look like. You can reduce them to have the same coefficients for x and y,
like in this case so this is 3 and this is - 4, this is also 3 and this is also - 4. In this case, the
separation between these two parallel lines would be the modulus divided by
(√ . So, the equation we are looking for, the line we are looking for also is going to
be some 3x - 4y + c = 0.

So, its separation from our l1 is going to be applying the formula modulus of √ .

is 25. Therefore, you have modulus of , which is what is expected to be 1 unit. That gives

us modulus of c - 5 = 5.

Now the model is, indicates that there are two possible values here, one could be c - 5 if
,. Because then c - 5 would be positive, and the other would be 5 - c if c < 5. So, what
we get is two separate solutions, one is c - 5 = 5 or 5 - c = 5, in which case we get c is 10 or 0.
So, we have 2 lines, one is , the other is our , this is our line.

So now, because of this, we can say that this length between, the separation between these
two lines is now one unit and therefore, the other line, which is our 3x - 4y + 10 = 0 is going
to come on the other side of , which is going to look like this. So, this line is our and it

will have intercepts equal to, this should be 0, which is 2.5 and this would be (- ,0), this is

our other plan.

And now, we should also find out what the value of a is, because a would be the distance
between the lines and that is what they are saying, its units away from our and we
know that this is one unit and this is also one unit. So, this total length is going to be 2 units.
So, a = 2.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 4.1 A
Quadratic functions
Welcome students, today we are going to start the new topic in our syllabus that is Quadratic
functions. Before starting this new topic, let us revise what we have studied so far. We started
with some simple geometric objects like points and lines, after studying points and lines
geometrically we plotted them on coordinate plane and seen how to derive the algebraic
equation of a geometric line.

When we have seen the algebraic equation of a geometric line, we got a form of the form
, it is also known as linear function that we have seen in last few lectures. And if
you recall recollect it from the first week where you have studied functions, this is
is a linear function.

Now, we want to enhance our knowledge further and add 1 more intrication or 1 more
complexity in this particular function and that is why we are studying quadratic function.
Here we will take an approach where we will first state the algebraic form and then derive its
geometric properties as opposed to what we did in straight lines. So, let us start with
quadratic functions.

(Refer Slide Time: 01:38)

The first question is, how will I define quadratic functions? The answer to this question is
given in this slide. So, a quadratic function is described by an equation of the form
, where is a crucial condition. Why? If it simply reduces to a linear
function. Let us talk about the name quadratic function. The name quadratic function is
derived from 1 foreign language where the quadra term, actual word quadratic term means
square and quadratic means related to square.

So, a quadratic function is a function that is related to square of the variable as can be seen
from the definition, it has a term containing a . So, if then it does not have a term
containing square so it no longer remains a quadratic function and it is a linear function
which is equivalent to a straight line as a geometric object.

So, we will put a condition that that means we are studying a quadratic equation. The
next question is how to plot a graph of this function. So, this equation is actually composed of
3 terms, let us describe them 1 by 1 that is a , this term is a quadratic term.

As I mentioned earlier, when , the term survives and that term is a linear term.
And finally, if you put , only term that survives is c so that is nothing but a constant
term. So, a quadratic equation can be split into 3 parts. If is not there, then I know how to
handle this term on a coordinate plane, it just simply represents an equation of a non-vertical
line.

So, I know how to handle these terms. So, what if the a term remains that is ? We can
graph this particular function and graph of any quadratic function will be called as parabola.
Graph of any quadratic function will be known as parabola. So, what are the important
features of parabola? In order to do that we first need to plot the parabola.

So, what is the best way to do it? We have already seen to graph any function what we need
to do is, we need to take the value of , put it in the formula and evaluate it and get the
values of . So, consider all ordered pairs and plot them on the coordinate plane so that they
satisfy this function, is the best way to handle it. For example, let us take this let us take for
example, when and and , let us take that particular function that is .

In that case what I will do is, I will put , I will get back 0. So, I will plot a point ,I
will take I will get 1, I will get 1, and then then I will get 4 and if I take
again I will get 4. So, can be easily plotted by joining these points smoothly
this is the curve , this is how we plot our quadratic function.
(Refer Slide Time: 5:35)

Let us take 1 example. Let us say I want to graph a function . How will I
graph this function? In 3 steps. First, I will generate a table of ordered pairs satisfying the
given function. Second, I will plot those points on the coordinate plane. Once I plot those
points on the coordinate plane, I will connect a smooth curve joining the 2 points, this is the
recipe for drawing a function. Let us draw it here, for that I have computed some points you
can verify by yourself, if you put the value of , you will get and on
solving you will get 1.

You take the value you will get 1, for x square you will get and 1 in the constant
term. So, together they will cancel and you will get 0. Similarly, you can compute for
it is 1 and for it is 4. Now, our job is to consider a coordinate plane and plot these
points so I have plotted these points.

So, these points are plotted and now I need to draw a graph, which is connecting all these
points. Now, here you remember I have plotted these 3 points, how will I know the shape of
this graph in this zone? That is a major question that you can ask, but this parabola is
somewhat symmetric in a sense, suppose I take this point, what is the point here, the point is
(1, 4).

Now, if I consider this point which is where it takes the value 0 and consider the point 1 it
is 2 units apart. So, somewhere in this where will come, which is 2 units apart from -1, the
value of the parabola will be again 4. I will keep the cursor here, see. So, there is some kind
of symmetry underlying this particular function, we need to understand that symmetry in a
better way.

So, what essentially is happening is, if I consider this point which is the bottom of the curve,
and if I draw a straight line, which is the line , then if you look at all these points for
every point there is a similar point on at the same distance from this particular point.
This also can be called as a symmetry of a parabola. We will study this later in the next slide.
So, right now, our job is to graph a function which we have plotted and let us explore further
properties of this parabola like this symmetry, what is the meaning of the symmetry and all
those things.

(Refer Slide Time: 09:00)

So, there are a few important observations, if you consider equation of ,

these are all parabolas, I have shown you two parabolas , both
parabolas have axis of symmetry. Inevitably all parabolas will have an axis of symmetry that
is, what is axis of symmetry. Let us go to the previous slide and see.
(Refer Slide Time: 9:28)

The axis of symmetry over here, as I mentioned was . If I take this graph paper and
fold along , then the curves that we have plotted here must exactly match each other
that gives us a recipe to draw a parabola.

(Refer Slide Time: 9:50)

So, all parabolas will have axis of symmetries that is if you take a graph paper containing the
parabola, and if you fold it along the axis of symmetry, the portions of the parabola on both
sides will exactly match with each other, this is the beauty of a parabola. So, now if I know
how the parabola appears on one side, I know how the parabola appears on the other side of
the axis of symmetry. It is a pure reflection of whatever is happening on one side.
Then, the point this axis of symmetry as we have seen in the previous graph, the point at
which this axis of symmetry meets parabola, we will call that point as a vertex of the
parabola. This is again a nomenclature we will call that point as a vertex of the parabola and
the point at which , if you put , then the point at which the coordinate is taken is
called the value c or you can simply refer to the equation , put , that will be
the y intercept which will be given by .

These 3 points play a crucial role in graphing the parabola. How? Let us do it one by one, Let
us say, our quadratic function is where , you can easily figure out that the
intercept of this point by putting the value 0 in .

Now, I want to know the axis of symmetry, this plays a crucial role. So, I will derive the
expression for axis of symmetry later but right now you memorize this equation as as
this needs some algebraic skills which we do not have right now. So, I will derive it later. But
right now, you understand that .

Remember, the equation of the quadratic function is given by , and will not
play any role in this and and will play a role. So, it is is the axis of symmetry and
where the graph meets this parabola, it is called vertex. So, the coordinate of the vertex is
obviously, because the axis of symmetry has . So, the coordinate of the vertex is

Let us see how this knowledge helps us in understanding how to draw a parabola. So, there
are 3 steps in drawing the parabola, first you need to generate a table of values, but if you
generate a table of values only on one side and you do not have a table of values on vertex,
then you may not be able to draw the parabola appropriately, that is way the knowledge of
these facts is important, so let us see how to draw a parabola by example.
(Refer Slide Time: 13:07)

, I want to graph this function. So, I will reiterate on previous points. So,
what is the intercept, intercept is 9, because if you put , the y intercept is 9. Next, I
want to know the equation for axis of symmetry. So in this case, what is , is 8, is 1, so

is , which will give me axis of symmetry to be , intercept is 9, axis of symmetry is

so I can evaluate the coordinate that is ( ) will be the vertex. How this
comes, you just substitute over here in this expression, you will get the value to be equal
to .

So, now with these 3 terms, how will I draw the function? So, now I know that around vertex
I need to find the points. So, based on this, I will draw a table. So, around I have simply
taken three points, fourth point is already with me, (0, 9) is the 4th point. So, around the
around the point I have taken the values so which is the value of when you
substitute in the function , already known and , . So, I have 3 points and the point
(0, 9).

So, I will plot these points on a graph paper, take a graph paper, plot the axis of symmetry
because around this the curve should be symmetric, take these 3 points, these 3 points are
here and I know (0, 9) is another point. So, it should be somewhere here (0, 9). Now, let us
plot a graph. So, now we have plotted a graph with much ease because of the knowledge of
axis of symmetry I know where the point where the minimum has occurred or the vertex
point is that is the beauty of axis of symmetry. So, this is how you will be able to plot any
function any quadratic function given to you, this is about the graphing of a function.
(Refer Slide Time: 15:48)

Let us try to see, is this the only shape that is possible that is the upward shape. Let us try to
figure out whether is this a quadratic function first of all, , the answer is yes, ,
and . Now, in this case, let us try to figure out the 3 summaries that is what will be
the intercept for this? intercept will be 1, what will be the axis of symmetry for this
because b is 0, it does not matter what is the value of it will be 0.

So, is the axis of symmetry that is axis is the axis of symmetry for this particular
function. And the vertex is (0, 1). In this case, we are not really getting much information
because what this is saying is (0, 1) is the intercept that is (0, 1) is the coordinate, axis of
symmetry is 0 that means (0, 1) is the vertex as well, right?

But still this information will suffice because I know I have to find the points around 0. So,
let Figure out the points around 0; , 0 and 1, these are the 3 points, their y coordinates
respectively are 0, 1, 0. Now you see there is a change, earlier we were only dealing with
positive side of axis or the axis where the curve is opening up, here the curve is opening
down. For example, if I plot an axis of symmetry over here, which is axis and if I plot these
3 points, these 3 points look like this that means the curve will go downward. So, the curve is
opening down, why this has happened.

In earlier examples, if you look at it closely then this the form, general form of this
expression , in all of them was equal to 1, and in this particular expression
therefore, the curve is actually opening down instead of opening up, this point needs
to be noted.
(Refer Slide Time: 18:15)

So, let us know this point and figure out what happens when this is greater than 0 or is less
than 0. So, that leads us to the next question that is maximum and minimum values. So, the
coordinate of the vertex of a given quadratic function is minimum or maximum value attained
by the function.

Do you all agree with this, we have seen 3 to 4 graphs of the functions, first we have seen
where it goes to bottom and 0 is the minimum value, then we have seen
which again gave us 0 value, then finally we have seen Third graph that we have seen the
last graph that is , because the value of a was negative, it was going downward, and
that will give me the maximum value and all the values are below that value.

So, the coordinate of the vertex of a given quadratic function gives us the minimum or
maximum value attained by the quadratic function. In particular, given any graph given any
function where . The graph of this quadratic function if a is greater
than 0 will open upwards and will have a minimum value.

If a is less than 0, the graph will open downwards and will have the maximum value and
there will be either maximum or minimum values, not both, this is the beauty of the quadratic
function. Another thing that you can see is the range of the quadratic function, if you relate to
your weak 1 background, where you are discussing about the domain-codomain range, so the
range of this quadratic function will be.

So, let us say a is greater than 0, then it attains the minimum value then it will be the
minimum value and all of the real line that is above the minimum value. And if a is less than
0, then it will be the maximum value and an entire real line which is below that particular
thing, I can denote this using the set theoretic notation as it is a set of real numbers R
intersected with a set of all , these values such that is greater than or equal to
min when , or if it is set of real numbers intersected with such that is
less than or equal to the maximum value that has achieved.

So, let us try to visualize this. For example, if your graph looks like this. So, in
particular the range of the value, range of values is from this point to upward. So, this is the
entire real line above this value. Similarly, if , the range of the values that is taken by
this function is this, if you relate this to domain codomain terminology, what is the domain of
this quadratic function, it is an entire real name and range is restricted to some subset of real
life.

(Refer Slide Time: 21:51)

So, we will try to improve upon this concept using this example.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 4.1 B
Examples of Quadratic Functions
(Refer Slide Time: 00:16)

Let us say, this example we have been given a function and we are asked
to determine whether has minimum or maximum values, if so what is the value and you
need to state the domain and range of . Let us first attempt the second question, what is a
domain? Domain of is enter real line we do not have to worry, what is the range of ?

Let us take this function identify so . Since , the function

opens up, if the function opens up then it will have a minimum value. So, the answer to first
question is whether has minimum or maximum value, it has a minimum value, once it has a
minimum value it cannot have maximum value, if so what is the value?

You need to figure out what is the vertex of this particular parabola. So, what is the formula
for vertex of the parabola, , , so is which will give me minus 3. Sorry,

this is wrong, it should give me , , it should give me , which is written wrong here,
but the graph is correct here where we are getting is the vertex. So, if you substitute
, what do you get? and therefore, the value of this is nothing but 0.
So, this is wrong it, should be .

And obviously, the range if it has a minimum value, the range is minimum upwards, this is
the entire real line above this minimum. So, that is R intersected with such that
 . So, we have understood how to find the minimum and maximum values of a function, if
is negative you can similarly find the maximum value.

(Refer Slide Time: 02:58)

So, let us try to make this example more realistic. So, let us take 1 realistic example. Where a
tour bus in Chennai serves 500 customers per day, they charge rupees 40 per person. Now,
they want to revamp their strategies, so the owner of the bus service estimate that the
company would lose 10 passengers per day for each Rs 4 hike in the fare.

So, if they hike the fare by 4 rupees, then they will lose 10 customers per day, this is the
estimate. Now, the company wants to maximize the profit, so how much should be the fair in
order to maximize the income of the company is the question. So, let us try to answer this
question using our knowledge of quadratic equation.

So, let us say, 1 unit of hike is 4 rupees so let denote the number of Rs 4 fare hikes. So,
what will this impact? This will impact the number of passengers because we are losing 10
passengers per fare hike. So, what will be the corresponding fair price for the passenger? It
will be , 40 rupees is the fees that we are charging per person, the company charging
per person and if I hike the fare it will be four times , this will be charged per person.

Now, the number of passengers with this hike if you increase x units, that means, you will
lose 10 passengers every units increase. That means is the passengers that still
remain. So, in this case, the income of the company will be the number of passengers into the
fare, they have charged so that is . If you open this, open the bracket
and multiply them, then you will get the expression to be . This is
the income.

Now, the company wants to maximize the profit, first of all after getting this quadratic
equation, can you tell me is the maximum possible? The answer is yes, and why the answer is
yes, because it lies in the coefficients and . So, what is here? , and
. Because , a < 0 so, the curve will open downwards that means the
maximum is possible.

And what will be the maximum value attained then that is what we have to figure out. So
now, the next question is okay. So, where this maximum will be attained? The maximum is
possible, maximum will be attained on the vertex, coordinate of the vertex will give me the
maximum. So, I will simply figure out what is the coordinate of the vertex, coordinate of
the vertex is point of intersection of the axis of symmetry, what is the axis of symmetry
, what is ? 1600, is 20,000 and is 0.

So, which will give me 20, so that is what 20 is yes. And of maximize coordinate

the maximum fair that we will get is 36,000. Right now, how much we are earning, how
much the company is earning, it is 500 customers they are serving, where everybody is
paying 40 rupees so they are simply earning only 20,000 rupees that is when you do not
increase any fare , you get 20,000. So, the main question is how much the fare should
be?

Now, what we are suggesting here by solving this problem is there should be a 20 units of
hike of rupees 4 each that means, what we are suggesting is there should be 80 rupees hike in
the fare. So, the new fare for the company should be 40 plus four times that is 20. So,
and that is what is the recommended hike by the company. So, now every
person should be charged 120 rupees as opposed to 40 rupees and then the company will be
profitable and you may have to serve less customers. This is how we are using real life, we
are using quadratic equations to solve real life situations.
(Refer Slide Time: 8:37)

Now, let us go back to our linear functions, where we studied the slopes of the lines. What
was the slope of a line? Slope of a line was change in by change in . Let us see what the
concept of slope has to do with a quadratic function. Let us try to analyse that. So, my goal in
this set of slides is given a quadratic function where , how to
determine the slope of a function .

So, in order to generalize this notion of slope of a function, we will first recall what we do
know about linear function. So, if you look at a linear function which is y which is equal to
, we know that this represents the slope and can be calculated by
considering a ratio of change in upon change in .

We have spent a lot of time in understanding the slope and when I consider a linear function,
I also know that the slope remains constant okay. We also know that the slope is nothing but
of some inclination and that inclination is with positive axis. I want to relate all these
concepts and try to figure out what is the slope of this quadratic function. Let us go ahead, we
will use a similar analogy for a quadratic function and define the slope of a quadratic
function. First let us take one example to discuss this concept of slope.
(Refer Slide Time 10:25)

Let us take our standard prototype example. We are trying to answer this question, .
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 4.1 C
Slope of a Quadratic Function

(Refer Slide Time: 00:15)

So, in this situation, I want to know what is the slope of this curve and is it a constant or which
variable or what else? So, we want to answer this question. So, first, we need to plot this
function, for plotting the function, we know what is the axis of symmetry for this b is 0. So, so
𝑦 − 𝑎𝑥𝑖𝑠 is the axis of symmetry and it will be symmetric about 𝑦 − 𝑎𝑥𝑖𝑠. Minimum value
will be 0 as it can be seen.

So, I will take a symmetry about 𝑦 = 0 that is, I have taken −2, −1, 0, 1, 2, these are the points
then I have evaluated the ordered pairs, that is 4, 1, 0, 1, 4. The symmetry is clearly visible in
these. Now, what is the definition of slope? It is change in 𝑦 upon change in 𝑥.

So, if you look at the left-hand side, the first column, the change in 𝑥 is constant. It is 1 all the
time so I will use this notation, and I will go ahead and figure out what is the difference between
𝑦𝑖 values because the denominator is always 1, it suffices to take the difference between 𝑦𝑖
values.

So, the first value is −3, 1 −4 is −3, 𝑦𝑖 − 𝑦𝑖−1 . 0 − 1 = −1, 1 − 0 = 1,4 − 1 = 3, so I got the
changes in 𝑦 with respect to 1 unit change in 𝑥 so this is the slope, but where does this slope
lie or at what point is this slope? Because if it is a straight line, I know the slope is constant.
So, in order to understand this let us go to a figure and try to understand.

This is a curve, 𝑦 = 𝑥 2 . Now, when I consider these 2 points 𝑦𝑖 − 𝑦𝑖−1 , what I am actually
doing is, I am assuming a straight line connecting these 2 points and I am calculating slope for
it. So, I have assumed all these straight lines and I have calculated the slope for it. Is this a
slope for a curve? No, basically not because it is a slope for that straight line.

So, now how will I identify this slope? So, if at all I want to decide what is the slope of the line
if you look at our old definition the change in 𝑦 by change in 𝑥 also associated with 𝑡𝑎𝑛 𝜃, the
𝑡𝑎𝑛 𝜃 plays a crucial role, what is 𝜃? 𝜃 is the angle of inclination. So, if I consider any point
over here, and if I draw the inclination of, if I draw a line passing through that point and if I
measure the inclination of that point with this positive 𝑥 − 𝑎𝑥𝑖𝑠 then I will get a slope because
the definition of 𝑡𝑎𝑛 𝜃 was not dependent on the line per se, it was dependent on that line on
that particular inclination.

So, 𝑡𝑎𝑛 of that is still a slope of a line. So, let us try to use this idea and see what we can get.
So now, I have identified 1 point let us say this point is actually (1.5, 2.25) because I am
considering a curve, which is 𝑦 = 𝑥 2 . What will be the slope of a line at this point? We can ask
this question but if you look at this line, this vertical line, this vertical line and if I slide this
vertical line slightly for this point, then this is nothing but a tangent to this curve, it passes
through it only once.

Let us try to actually plot that line. Yes so, once we have plotted this line, this is the tangent to
that curve and the line is actually parallel to this line and the point is 1.5. This gives me a hint
that this is something like you have −3, the point is, you have a slope between these 2 points
as 3, you have a point which is 1.5 and if I divide this point, this particular difference by that
point I am getting 2. Then let us look at these differences, what are these differences, the
difference is −1 − (−3) = 2,1 − (−1) = 2 so, all these differences are 2.

If you look at the second differences of these points, there are 2 that means there is some
relation, 3 and 1.5, 1.5 times 2 is 3. So, I can safely assume that this point 1.5 is actually a
midpoint of 1 and 2 on the 𝑥 − 𝑎𝑥𝑖𝑠 and therefore, whatever value is given to it is actually the
value of the slope of a curve. And in particular, if I go here, for example, if I go here, and if I
talk about the point 0 and 1, then what I will get is a point 0.5, the midpoint of this. Again, I
can do a similar exercise, I can draw a line and the line again will be parallel to this line and at
a point 0.5, I will get the line with a slope 1.

In a similar manner if I go here, I will get a line with a slope −1, in a similar manner here, I
will get a line with a slope − 3 and therefore, I can safely conclude that the slope of this
particular curve is 2𝑥. How? I have computed it. So, let us now verify our hypothesis. So, let
us take a point 0, consider any 2 points about 0, let us take symmetric points because I need a
symmetry.

So, let us take the point (1, -1), what is the slope of this line? It is horizontal line, so the slope
should be 0 and that is what this slope is. So, in particular, I can verify for all points if I consider
a point, let us say a, a is used here. Let us say if I consider a point z then I will go z +u, z−u, I
will consider those 2 things and I will assume their values, draw a straight line joining them
and whatever is the value of the slope for that straight line will be the value of slope for my
point. This is a beautiful idea that can be generalized to a general quadratic curve.

(Refer Slide Time: 8:33)

So, let us answer the general question that is, I want to find a slope of a quadratic function𝑎𝑥 2 +
𝑎𝑥 2 + 𝑏𝑥 + 𝑐 where 𝑎 ≠ 0. So, we will simply take 5 set of points, standard 5 set of points,
−2, −1, 0, 1, 2, I will just substitute these values in the function. So, I will get a corresponding
values of 𝑦𝑖 ′𝑠, which are here, 4𝑎 − 2𝑏 + 𝑐 , and𝑎 + 𝑏 + 𝑐 , 𝑐, 4𝑎 − 2𝑏 + 𝑐 . I will take the first
differences of these two, those are given here and then I will take one more difference of these
two, all these differences will turn out to be 2𝑎.
Now, if I look at the points which are here, and if I consider the midpoint of this midpoint of
these 2 that is 1.5 so−2𝑎 × 1.5 + 𝑏 will give me the answer to my question that what is the
slope of that particular value, because if you look at this −3, −3 is actually 2 times 1.5. This
−1 is actually 2 times −0.5, 1 is 2 times 0.5, 3 is again 2 times 1.5 so I am essentially getting
the slope of all these values that means, my answer to the question that the slope of this curve
quadratic function is 2𝑎𝑥 + 𝑏.

Now, from the table it is very clear the 2 way comes here,𝑎𝑥 + 𝑏 I have derived it because this
is a value containing 1.5 in the middle, so this is 2 times 1.5 So, that is 𝑎𝑥 so 2𝑎𝑥 that is what
this is 2𝑎𝑥 + 𝑏. Now, we can do some interesting observations, we have already seen around
point 0 for 𝑦 = 𝑥 2 , the slope was flat it was 0. So, when will that happen? Right.

So, you can equate this 2𝑎𝑥 + 𝑏 = 0, slope 0 means the function has reached its minimum or
maximum, slope is 0. So, when will that happen? That is 𝑥 = −𝑏. This is one of the reasons
2𝑎
why 𝑥 = −𝑏 is the value of the minimum or maximum, because the slope reaches the value 0.
2𝑎
So, here what actually slope, calculates?

Slope actually calculate the rate of change with respect to 𝑥 and a rate of change of 𝑦 with
respect to change in 𝑥. So, if the rate of change is becoming 0, that means the function has
reached its minimum or maximum. So, this justifies the idea that why a quadratic function
should have a minimum or maximum value at the point𝑥 = −𝑏.
2𝑎

Still, that point is pending where we want to find why the axis of symmetry is 𝑥 = −𝑏 and we
2𝑎
will come to it later. But as you can see here, the slope of a quadratic function is significantly
different from slope of a line, slope of a line is constant, whereas the slope of a function
quadratic function 𝑓 is no longer a constant. In fact, it is variable that is 2𝑎𝑥 + 𝑏. It depends
on 𝑎and 𝑏, not on the constant 𝑐 , which is expected.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 4.4
Solution of quadratic equation using graph

In today’s video, we are going to learn what are Quadratic Equations. And once we set up the
Quadratic Equation, we are going to see, what are the solutions of the Quadratic Equations,
that are called roots of the Quadratic Equation and how to solve these Quadratic Equations,
using the technique that we have demonstrated, for quadratic functions. That is Graphing
technique. So, let us start.

(Refer Slide Time: 00:42)

So, first of all, let us understand what is a quadratic equation and how it is related to quadratic
function. So, here is a definition. If a quadratic function is set to be equal to a value, then the
result is called quadratic equation. So, let us see one example. For example, 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 =
0, is one quadratic equation, where 𝑎 ≠ 0.

In the similar manner, 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 = 5, is another quadratic equation. Obviously a should

not be equal to 0. Now, once we get the Quadratic Equation, if the coefficients, what are the
coefficients. Coefficients are like a, b and c. These are called coefficients of the Quadratic
Equation.
If the coefficients are from set of integers, which we have studied in week 1. So, if a, b, c, the
coefficients are integers, and on the righthand side, it is equated to 0. That is, you have an
equation, 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 = 0, where a is not equal to 0 and a, b, c are integers. Then the
Quadratic Equation is said to be in the standard form. So, on this slide, we have seen two
definitions. One, what is Quadratic Equation. Quadratic Equation is nothing, but a quadratic
function, where it is equated to some value.

And what is a standard form of Quadratic Equation? That is 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 = 0, where 𝑎, 𝑏, 𝑐 ∈

ℤ and a is not equal to 0. Then the Quadratic Equation is said to be in a standard form.

(Refer Slide Time: 02:37)

Now, once we have a Quadratic Equation in standard form, we can discuss about roots of the
Quadratic Equation or zeroes of the functions. And we will see how the concept of roots of
Quadratic Equation and zeroes of quadratic function are related, in this slide. So, the solutions
of the Quadratic Equation are called roots of the equation.

What do I mean by that? If 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 = 0, then what is the value of x, that gives me 0, is

called the solution to the Quadratic Equation. And also, that value of x will also be known as
root of the Quadratic Equation. So, this way we get the root of the Quadratic Equations.

So now, which way you can find the roots of the Quadratic Equations? One method, which is
very easy. If you have a quadratic function associated with this Quadratic Equation, then you
just plot the quadratic function and find its zeroes. What is a zero of a quadratic function? Zero
of a quadratic function is nothing but its x intercept. So, in particular, if you observe that, zeroes
of the functions are x intercepts of its graph and these are the solutions to the related equation,
𝑓(𝑥) = 0, at these points?

So, if you are having a quadratic function, what you need to do is just plot it and see where it
intersects x axis. If it intersects x axis, then you got the solution or the root of the Quadratic
Equation. So, let us try to see this through some examples.

(Refer Slide Time: 04:32)

So, here are some examples. So, the question is to find the roots of the following equations.
First equation is 𝑥 2 + 6 𝑥 + 8 = 0. Second one is 𝑥 2 + 2 𝑥 + 1 = 0. And third one is 𝑥 2 +
1 = 0. Now, we will take these equations one by one. So, essentially what we are proposing
is, we want to chart these equations or we want to plot these expressions on a graph paper.

So, if you recollect from our last few videos, in order to plot a quadratic function, we need to
understand the axis of symmetry of the quadratic function. So, let us take the first example,
where you have 𝑥 2 + 6𝑥 + 8. Now, I want to understand, what is the axis of symmetry of this
particular function.

Let us see. So, in this case, for our standard notation, our related quadratic function is 𝑥 2 +
6𝑥 + 8. So, 𝑎 = 1, 𝑏 = 6 𝑎𝑛𝑑 𝑐 = 8. So, y intercept is obviously 8. And axis of symmetry
𝑏 6
is 𝑥 = − 2𝑎, which obviously means it is− 2 , which is -3. So, axis of symmetry is x = -3.
So, axis of symmetry is x = -3 and a, the value of a is positive. So, what are the things that we
can conclude from our previous videos? That is, if a > 0, the curve opens up, the graph of the
function opens up. It attends the minimum.

And the axis of symmetry in this particular example is x = -3. So, the simplest thing that we
can do here is, put x is equal to minus 3 in this expression. And you will see that, the expression
will take a negative value. That means the y value taken is negative. That means if the y value
taken is negative, you can easily see, the curve opens up. That means it will intersect x axis in
two points.

Now, we want to guess those two points. Without plotting, right now based on our visual
interpretation of this curve, can we guess the two points? Okay. So, -3, the value is negative.
That means, for -3 it is negative. Then let us check it for x is equal to -2. If you substitute x is
equal to - 2, you will get 4 − 12 + 8 = 0. So, one root I have got, which is -2. If -2 is one root,
-3 is one, -3 is axis of symmetry. That means, at a distance one apart from this, there will be
another root. That means - 4 will be the second root.

Wow. So, we were able to understand, that -4 and -2 will be the roots of this equation, without
even drawing, just on the basis of what we have understood. So, what we have understood here
is, -2 and - 4 will take the value 0 and for x is equal to -3, you will get one negative value. And
based on that, you have prepared a table. And therefore, you can plot this graph easily. Right?

So, we will graph the related quadratic function, using axis of symmetry and vertex. We have
already discussed this. So now, axis of symmetry x = - 3, the roots are - 4 and - 2. And therefore,
the Quadratic Equation given here, 𝑥 2 + 6𝑥 + 8 has two real solutions, two real roots. How
will the graph look like? It is very easy. We have already imagined the graph. Yes, so this is
the graph, where - 4 is a point here and -2 is a point here. -4, - 2 are the roots. And here, it
achieves the minimum, which is - 3.

So, you can easily plot this graph. Let us go to the second equation. Now, in this second
equation, again we will consider the associated quadratic function. What is the associated
𝑏
quadratic function? 𝑥 2 + 2𝑥 + 1. What will be the axis of symmetry for this? − 2𝑎 , that will
𝑏
be - 1. Because b is 2 and a is 1. So − 2𝑎 = −1.

So, x = - 1, is the axis of symmetry for this particular quadratic function. Let us substitute the
value of x = - 1, in this quadratic function. So, you will get (−1)2, which is 1, 2 × −1, which
is – 2, + 1. So, you will get 0. Oh! so, -1 itself is a zero. Correct? But that is a point of the
vertex, where it achieves the minimum. So, there using axis of symmetry, you can conclude
that there cannot be any other point, other than - 1, where it will take the value 0. Because that’s
the point, where the vertex arises.

That means the axis of symmetry for the second equation is x is equal to -1. a is greater than 0.
So, the curve opens up. And therefore, it achieves the minimum. And therefore, the roots are
-1 and -1. What is the value at -1? It is 0. So, that is that itself is a root. And therefore, it has
only one real root, which is repeated twice. So, in particular, the graph of a function will look
like this.

Now, the next problem is very interesting.𝑥 2 + 1, where if you compare this with a standard
form of the equation, 𝑎𝑥 2 + 𝑏𝑥 + 1 , then you will get b to be equal to 0. That means this curve
or this, the graph of this function will be symmetric about x = 0, that is y axis. And since a is
greater than 0, the curve will open upwards. So, the curve is opened up.

Now, a > 0, it will achieve the minimum value. Where it will achieve the minimum value? At
the vertex. So, what is the vertex of this particular function? Because x = 0, so, where it, you
substitute x is equal to 0 here. So, that value is 1,𝑥 2 + 1 = 1. So, (0, 1), so 1 is the minimum
value of this function. Can this function be equal to 0 then? It cannot be. So, this will give us
the answer, that axis of symmetry x is 0. There are no real roots for this particular function,
because it never intersects x axis.

And the function will look like this. So, this in short summarizes, what are possible solutions
in any scenario, 𝑎𝑥 2 + 𝑏𝑥 + 𝑐 = 0 is given to you. In particular, if 𝑎𝑥 2 + 𝑏𝑥 + 𝑐, if you are
able to find the vertex and the vertex takes the negative value and a is greater than 0, the curve
opens up. So, it will have two roots which are real numbers. If the curve opens up, but the value
at the vertex is 0, then it has only one root.

And if the curve opens up and it is above the X-axis, that is it takes a positive value on the
vertex, y coordinate of the vertex is positive. Then it will never intersect x axis. In the similar
manner, you it is for you to study, that when a is less than 0, what will happen. So, I can give
you the rough interpretation. If a is less than 0 and it achieves the maximum on the vertex.

And if that maximum is positive, then it will have two real roots. If a is less than 0 and at the
vertex, the value is 0, then it will have a single real root. And if a is less than 0 and it is below
X- axis, then it will have no real roots. So, these are the scenarios, that we can cover using this,
this graphing technique.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 01

(Refer Slide Time: 00:15)

Hello mathematics students in this tutorial, we are going to look at problems related to the
topics covered in weak four. And so, these are the topics and we will begin with our first
question.

(Refer Slide Time: 00:26)

Here, we would like the minimum value of y for this particular quadratic function. And first,
let us put down the quadratic function in its standard form, the standard form would be 𝑦 =
𝑎𝑥 + 𝑏𝑥 + 𝑐. In which case, our particular equation, the one that is given here would give
us 𝑎 = 1, 𝑏 = 1 and 𝑐 = 2. We are looking at the minimum value. Now, because the
𝑥 square coefficient a is 1 that is a is greater than 0.

So, our parabola will be in this form, if a were lesser than 0, it would be inverted, it would be
a downturned parabola, but right now it is in this form, and the minimum value is going to
occur at this point, which is the vertex, which we know to be . And so, we know our vertex

for this particular equation is . And the value of y at would be the minimum.

So, I can write the minimum is equal to 𝑦 ( ), which in this case is 𝑦( ). And if I substitute

that, I would get (− ) – + 2, which is essentially - + 2, which gives us 2 - , which is

( )
equal to , which is equal to and that is essentially 1.75. So, this point, here it is now we

know it to be (-0.5, 1.75).

Now, they are asking us for the 𝑥- intercept and this is what we need to observe about the
𝑥 −intercept.

(Refer Slide Time: 2:57)

Point (-0.5, 1.75) assuming this is 1 and this is 2, this is -1 of course, so this is negative side
and this is -2. So, -0.5 is going to be somewhere here and on the Y- axis, this would be 1 and
this would be 2, 1.75 is somewhere here so our vertex point is here.
And from here, we know that this is an upward parabola, which is going to be something like
this. And that means it never touches the X- axis at all. There is no 𝑥 −intercept for this
parabola.

And lastly, it is asked to find the length of the line segment on the straight line passing through
the y- intercept of the given curve and the point (-2, 4). So, (-2, 4) is somewhere over here, and
we need to find this point here the 𝑦 −intercept. And the 𝑦 −intercept is easy to obtain, since
our curve is y = 𝑥 + 𝑥 + 2. 𝑦 − intercept is obtained when the curve cuts the Y- axis that is
when 𝑥 = 0 so 𝑦(0) = 2, therefore, our intercept is actually (0, 2).

And the point we are looking at is (-2, 4) and it is the length of this line segment that we require.
And that line segment we will get by using the Euclidean distance formula, it will be
(−2 − 0) + (4 − 2) which is essentially √4 + 4 , that is root √8, which is 2√2, units.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 02

(Refer Slide Time: 00:14)

Now, second question we are going to have, this quadratic functions curve touches the X-axis
exactly at 1 point. And for that what is the value of 𝑘 supposed to be? First observation should

be that the vertex is given to us. The vertex, which is , here 𝑎 = 1, 𝑏 = −6, and 𝑐 = 𝑘,

thus the vertex is , which is that is 3. So, this is 1, this is 2 and this is 3, our vertex is on

this particular line that is 𝑥 = 3. And we are told that it touches the X- axis, the parabola
touches the X-axis at precisely 1 point.

We also can see that a is positive, so this is an upward turn parabola, upturned parabola. And
if it touches the X-axis at exactly 1 point that is only possible when the vertex is right here on
the X-axis itself, and from here, our parabola looks something like this. That means, for this
condition to be satisfied at the vertex, 𝑦 = 0 that is 𝑦 (3) = 0. And that is equal to 9 −
18 + 𝑘 = 0. This gives us 𝑘 = 9 that is it so 𝑘 = 9. When that happens, our equation is
𝑦 = 𝑥 – 6𝑥 + 9 and it has its vertex at (3, 0).
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 03

(Refer Slide Time: 00:14)

In question 3, there is a path 𝑥 = 𝑦 − 6𝑦 + 8. So, if we observe here, we are basically

saying 𝑥 is a function of 𝑦. And that function is quadratic, we have 𝑦 − 6𝑦 + 8. So, we are
now switching the axis and so our parabola is expected to look something like this, or like this,
because this is the X-axis and this is the Y-axis. Now, we see that the coefficient of y square,
which is 𝑎 is 1, and 𝑏 is -6, the coefficient of 𝑦 and lastly, the constant term 𝑐 is 8.

Since, 𝑎 is greater than 0, we expect that this is an upturned parabola. In the case of upturn in
X, what we mean is it is towards the positive X-axis. So, our parabola is expected to be
something like this. Of course, it could be moving about, we do not know where exactly it cuts
the axis or where the point is. And for that, we will have to go further. They are saying this 2
stops on the line 𝑥 = 0 that is on Y-axis and of course, these will be this point and this point,
basically, if we looked at it in terms of our standard 𝑦 = 𝑓 (𝑥) these are what are the roots of
our equation.

And Arav's home is at the origin, Arav lives at the origin so this is where Arav is. How much
minimum distance will Arav have to cover in order to catch the train? So, the question is simple,
you have two routes for your 𝑥 = 𝑓(𝑦), and these routes will be on the Y-axis now, because
we have switched the axis and which route is closer to Arav's home that is which route is closer
to the origin. So, let us try to find out now, let us try to plot this particular graph and let us see
where the tool train stops are. From the equation, we know that the vertex will be , which
( )
is again , that is 3.

So, here we are basically saying 𝑦 = 3 is the vertex. So, this is 1, this is 2, then this is our
𝑦 = 3 and thus, the vertex will be along the line, 𝑦 = 3, the axis of symmetry is 𝑦 = 3. So,
this is our axis of symmetry, 𝑦 = 3. And for plotting the graph, we are now going to look at
various points, which will be 3 and 1 to the other side of 3, 2 and 1 to this side of 3, 4 and then
5, and then 1, this should give us a reasonable idea of what the graph looks like.

So, 𝑓 (3) at the vertex, what is the x-coordinate that would be 𝑓(3) = 9 − 18 + 8 = −1,
so 𝑥 = −1, which is going to be somewhere around here, this is our vertex. And 𝑓 (2) will be
equal to 𝑓(4) because of symmetry. So, if I just substitute 2, I will get 4 − 12 + 8 = 0, ok,
that is good so we now have roots, we know that on 2 this point, and at 4 our curve is going to
intersect the Y-axis.

So, if you want, we can further look at what is 𝑓(1) which is also equal to 𝑓(5), that is going
to give us 1 − 6 + 8 = 3, so we got to be somewhere over here, for these two points we are
going to get somewhere here and thus our quadratic parabola looks like this. And we know for
a fact that the routes are 𝑦 = 2 and 𝑦 = 4. Clearly 𝑦 = 2 is closer to the origin. So, the
minimum distance that Arav will have to cover is 2 units that is from the origin to this particular
point, and this is the distance he will have to cover.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 04

(Refer Slide Time: 00:15)

In the fourth question, there is some data of a vehicle, and a student fitted a curve for the
vehicle's speed 𝑥. So, this is our variable 𝑥 and its fuel economy mileage in kilometre per litre
as 𝑓(𝑥). So, it is a function of 𝑥 and this function is given in this way, we are going to use it
for 𝑦 which means if we reduce it to the standard form, we will get 𝑦 = 𝑓(𝑥) = 𝑥 −

+ 30. So, we have the situation where a the coefficient of 𝑥 square is , which is equal

to −0.025. And 𝑏 is the coefficient of 𝑥 which is and that is , therefore 2.2, and lastly,

𝑐 = 30.

Now, we may observe that the 𝑥 square coefficient is negative so this is a downturn parabola,
which is why they are asking what is the maximum economy. So, at the vertex, you will get
the maximum fuel economy so we need to find the vertex. And we know that the vertex is at x
.
is equal to , which in our case is then ×( .
. This is probably better than in fractions.
)

So, if we write it down in fractions, we have −𝑏 = and this will be into and is

then −40 itself, because a is . So, we have the 40 and the 40 cancelling off and minus and
minus become plus 2, and 88 will give us 44. So, we have the vertex that is we get the maximum
fuel economy at a speed of 44 kilometres per hour. And what is the maximum economy at this
×
particular speed that we can calculate from our equation directly we have 𝑓(44) = −
×
+ 30 so this is 4, 10s a 4, 11s.

This is also 4, 10s and 4, 11s and we get 96.8 − 48.4 + 30, which is then further equal to
96.8 is two times 48.4 so you will get 48.4 + 30 giving us 78.4 kilometre per litre. So, we
can say that this is our maximum fuel economy which is achieved at this particular speed.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 05
(Refer Slide Time: 0:15)

Our fifth problem looks a little complicated, but let us go one by one. And here we have the
production rate of a material which is being made in a factory depends and it two factors 𝑓1 and
𝑓2 as 𝑅 = 𝑓1 𝑓2 . And these two factors, they are the functions of the purity of the raw material.
And that variable is 𝑥, 𝑥 is the purity of the raw material. And both these functions are given
to be linear 𝑓1 (𝑥) = 𝑎𝑥 + 𝑏, 𝑓2 (𝑥) = −𝑐𝑥 + 𝑑. And it is given that 𝑎, 𝑏, 𝑐, 𝑑 are all positive.

And it is asked find the purity of material, that is the value of 𝑥 for which the production is
maximum. So, let us understand what is being done here. We have two linear functions and the
rate of production 𝑅 = 𝑓1 𝑓2 , which will then 𝑅 = (𝑎𝑥 + 𝑏)(−𝑐𝑥 + 𝑑) = −𝑎𝑐𝑥 2 + 𝑎𝑑𝑥 −
𝑏𝑐𝑥 + 𝑏𝑑 = −𝑎𝑐𝑥 2 + (𝑎𝑑 − 𝑏𝑐)𝑥 + 𝑏𝑑.

We are told that 𝑎, 𝑏, 𝑐, 𝑑 are all positive, and that indicates the coefficient of 𝑥 2 is negative
because the negative of 𝑎𝑐 and that means this is a quadratic function whose parabola is
downturned, therefore, we will be able to get a maximum value at some point and this is going
−𝑏
to be at the vertex, we know that this is going to be at the vertex. So, the vertex is at , that is
2𝑎

because here we have 𝑎, 𝑏, 𝑐, 𝑑 already.

 −(𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝑜𝑓 𝑥) (𝑎𝑑−𝑏𝑐) 𝑎𝑑−𝑏𝑐
Let us write it down more carefully, that is the 2(𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝑜𝑓 𝑥 2 ) = − = , is where
2(−𝑎𝑐) 2𝑎𝑐

we will get the vertex. And since we know that the maximum is going to occur at this particular
𝑎𝑑−𝑏𝑐
𝑥, we get the 𝑥 = .
2𝑎𝑐
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial – 06

(Refer Slide Time: 0:14)

In our sixth question, we are given this particular quadratic function 𝑓𝑓(𝑥𝑥) = −𝑥𝑥 2 + 8𝑥𝑥 + 6.
And we are told that two points P and Q, which are on this parabola such that they are two units
away from the axis of symmetry. So, let us try to find out what the axis of symmetry is for this
parabola. Our equation is 𝑓𝑓1 (𝑥𝑥) = −𝑥𝑥 2 + 8𝑥𝑥 + 6. And that would mean, in a standard form
𝑎𝑎 = −1, 𝑏𝑏 = 8, and 𝑐𝑐 = 6.

−𝑏𝑏
And that would give us the vertex is at 𝑥𝑥 = 2𝑎𝑎
, which in our case will then become 𝑎𝑎 = −1,

𝑏𝑏 = 8, so we will get 4. And the functions value at 4 is 𝑓𝑓1 (4) = −(4)2 + 8 × 4 + 6 = 22. So,
the vertex is (4, 22). Further, we are told that P and Q are two units away from the axis of
symmetry. So, the axis of symmetry is along 𝑥𝑥 = 4, which means P and Q will be at 𝑥𝑥 = 2 and
𝑥𝑥 = 6, 4 − 2 and 4 + 2.

So, these points are going to be P(2, 𝑓𝑓1 (2)) = (2, −4 + 16 + 6) = (2, 18). And the point Q
is going to be 𝑃𝑃(6, 𝑓𝑓1 (6)) and from symmetry we know that this is also going to be 18, so 𝑃𝑃(6,
18). And it is now told to us that the triangle PVQ is rotated 180 degrees about its axis of
symmetry and we are being asked the curved surface area of the resulting cone. So, let us look
at what this looks like.
(Refer Slide Time: 3:08)

So, now let us suppose that this point here, let us call this our (4, 22), in that case 18 is 4 units
below, so this will be the horizontal line passing through 18 and 2 will be here. So, (2,18) is
here and this gives us (6,18) is here. This is (2,18) and this is (6,18). And that gives us a parabola
which looks something like this, obviously a smoother curve than I have drawn, but something
like this. And the triangle we are interested in is an isosceles triangle, which looks roughly like
this.

This is the triangle that is being rotated 180 degrees about its axis of symmetry and its axis of
symmetry is 𝑥𝑥 = 4. I am erasing the parabola in order to focus on the triangle alone. If this
triangle were to be rotated, this point which is our P, this is our Q, this point P basically goes
around and reaches Q, whereas Q comes around and reaches P. And in this way, we have a
cone in our hands and we want the curved surface area and that would be this region and the
base circle is this flat surface below this is the base circle.

And we are interested in the curved region whose surface area is given to be 𝜋𝜋𝜋𝜋𝜋𝜋. So, what is
𝑟𝑟, 𝑟𝑟 is the radius of the base circle. Which is basically then this quantity, this is 𝑟𝑟, which we
can tell is 4 − 2, so it is 2. And what is 𝑙𝑙 over here, that is the slant height, which is basically
this height, that height can be obtained as the hypotenuse of this base radius and height here,
which is as we can see 4 units. So, 𝑙𝑙 = √22 + 42 = √20 = 2√5.

So, we have 𝑟𝑟 = 2 and 𝑙𝑙 = 2√5, this gives us a curved surface area is 𝜋𝜋 × 2 × 2√5 = 4√5𝜋𝜋
square units.
(Refer Slide Time: 6:46)

For the part B of our question we have another curve which is also quadratic and whose roots
are basically 4 repeated. So, 𝑓𝑓2 (𝑥𝑥) = (𝑥𝑥 − 4)(𝑥𝑥 − 4). So, x being equal to 4 makes 𝑓𝑓2 (4) =0.
So, therefore, our root is 4 and it is repeated because coming twice here. So, let us now try to
look at what they are asking. Now, let A be the set of all points inside the region bounded by
these curves, including the curves. So, we are saying the region bounded by these curves and
including the curves.

And they would like the range of y coordinates of points in it. We know already that (4, 22) is
the vertex for our previous parabola. And it also passed through (2, 18) and (6, 18). And about
this new parabola, the 𝑓𝑓2 (𝑥𝑥), we know that 4 is repeated root so there is only 1 root and
therefore, at 4, that is 22, this would be 21, this is 20, this is 19, 18 17, 16, 15, 14, 13, 12, 1, 11,
10, 9, 8, 7, 6, 5, 4, 3, 2, 1, and 0. So, this is going to be the repeated root and the vertex of our
other parabola.

So, if one parabola is like this, 𝑓𝑓1 had negative 𝑥𝑥 2 coefficient so it is a downturned parabola,
then the other parabola 𝑓𝑓2 (𝑥𝑥) = (𝑥𝑥 − 4)2 is an upturned parabola which is going to be
something like this. So, these curves are going to intersect in some way this way. And we are
interested in the range of 𝑦𝑦 − 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐. So that would be, what are all the 𝑦𝑦 −
𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 possible in this region.

So, if this is the region we are looking at, then clearly this is the upper bound of our 𝑦𝑦 −
𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 and this is a lower bound. So, 𝑦𝑦 − 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 in our region range between 0
and 22. And they said including the curve, so 0 is also included, 22 is also included, so we can
write the same thing as 𝑦𝑦 𝜖𝜖 [0, 22].
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 07

(Refer Slide Time: 0:04)

In question number 7, we have one relation given this way, y2 = 4ax. And they are asking a
very simple question, is y a function of x. So, we have y2 = 4ax. And the interesting thing
about square roots is, if I did the square root of 1, it is not just 1, it is actually ±1. So, both
(+1) = 1, which is also equal (−1) .

So in this case, we need to consider the fact that y =±√4𝑎𝑥. Which means for the same x, I
might have 2 different y's. So, put it this way, I am basically saying f(x) assuming it is a
function is equal to √4𝑎𝑥 and f(x) is also equal to the −√4𝑎𝑥. And this is not allowed, for a
single element in the domain, for a function, you should have only one image in the range.
But here we have 2 different images for the same element in the domain. Therefore, this is
not a function.
(Refer Slide Time: 1:55)

If we looked at it in terms of the plot, we have y2 = 4ax, that is what we are trying to plot.
And for x = 0, we get y = 0. So, this curve passes through the origin definitely. And for the
next x value, I am going to take a, so therefore, y2 = 4a2, which gives 𝑦 = ±2𝑎. So, if a is
positive, this is (a, 0), then 2a is going to be somewhere here like this, and -2a is going to be
somewhere here like this.

And so, we have a parabola which goes something like this. And if a were to be negative,
then this would have been (a, 0) and we would have a similar parabola in the negative
direction. Either way, it is pretty clear that for a given value, you have two corresponding y
values, for a given value of x you have two corresponding y values and that is not allowed for
a function.

Independently f(x) =√4𝑎𝑥 , which is this part of the curve, that can be treated as a function
and f(x) = −√4𝑎𝑥 , for convenience let us call this as f1 and this is f2. This is also possible to
be treated as a function independently, but their combination which gives us this relation, that
is not a function.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Week - 04
Tutorial - 08

(Refer Slide Time: 0:15)

For our eighth question we have an advertiser who is analyzing the growth of likes for their
new ad on YouTube. She analyzed that the increase in likes in a given second is equal to 4
times 𝑡𝑎𝑣 , where 𝑡𝑎𝑣 is the midpoint of the time interval, that is the average time in that time
interval. And so we are given an example to explain what this is. The increase in likes from 3
seconds to 4 seconds. So, from the time 𝑡 = 3 to the time 𝑡 = 4, there is a number of increase
in likes, which is equal to 4 × 3.5 and 3.5 is the midpoint of 3 and 4.

So, one way to write this is, let us look at time 𝑡 seconds and the time 𝑡 + 1 seconds. Then it is
given to us that the likes at time 𝑡 + 1, so, number of likes is a function of time. So, 𝑙(𝑡 + 1) −
(𝑡+𝑡+1)
𝑙(𝑡) = 4 × 𝑡𝑎𝑣 = 4 × = 4𝑡 + 2, this is the difference in the likes from time 𝑡 seconds
2

to 𝑡 + 1 seconds.

Now, it is further given to us that this particular function is a quadratic function. So, 𝑙(𝑡 + 1) =
𝑎(𝑡 + 1)2 + 𝑏(𝑡 + 1) + 𝑐 and 𝑙(𝑡) = 𝑎𝑡 2 + 𝑏𝑡 + 𝑐. Then 𝑙(𝑡 + 1) − 𝑙(𝑡) = 𝑎𝑡 2 + 2𝑎𝑡 + 𝑎 +
𝑏𝑡 + 𝑏 + 𝑐 − (𝑎𝑡 2 + 𝑏𝑡 + 𝑐) = 2𝑎𝑡 + 𝑎 + 𝑏

(Refer Slide Time: 3:08)

This quantity is basically equal to 2𝑡 + 4. So, we are saying that 2𝑎𝑡 + 𝑎 + 𝑏 = 2𝑡 + 4. Now,
what are we supposed to acknowledge here is that the term with the 𝑡 in it, that is the time
dependent term is going to be same on both sides. Whereas the term which is constant is going
to be same on both sides.

Thus, we are saying 2𝑎𝑡 = 4𝑡 and 𝑎 + 𝑏 = 2. This gives us 2 times t and t cancelled. So, we
know 𝑎 = 2 and that would imply 𝑏 = 2 − 𝑎 = 0.
(Refer Slide Time: 4:20)

And our question is asking us what is the value of 𝑏. So, we know this is equal to 0. Second
question, the second part of the question is asking what is the total number of likes at the end
of 60 seconds.

(Refer Slide Time: 4:49)

That would be impossible to calculate because we have the values of 𝑎 and 𝑏, so we know that
our 𝑙(𝑡), in this case we want 𝑙 of 60. 𝑙(𝑡) = 𝑎𝑡 2 + 0𝑡 + 𝑐 = 2𝑡 2 + 𝑐. But we do not know
what 𝑐 is. So, 𝑙(60) = 2 × 602 + 𝑐. Now, if we made further interpretations that there were 0
likes at time 𝑡 = 0. So, if 𝑙(0) = 0 then 𝑐 = 0. So, this is a particular assumption we are
making, we are assuming that the timer started when the likes were 0 and that would imply
your 𝑙(𝑡) = 2𝑡 2 .
So 𝑙(60) = 2 × 60 × 60 = 7200, that is 7200 likes at the end of 1 minute.

(Refer Slide Time: 6:18)

And lastly, for Part C, we are being asked the domain of the function is [𝑘, ∞), what is the
value of 𝑘. We know that 𝑙(𝑡) = 2𝑡 2 + 𝑐. Now only real requirement we have is that our likes
be greater than or equal to 0. So, 𝑙(𝑡) ≥ 0 → 2𝑡 2 + 𝑐 ≥ 0. Another thing we have is clearly
that 𝑙(0) = 𝑐 ≥ 0, because at 0 time, it is not like you can have negative likes. So, 𝑐 ≥ 0.

Now we know that 𝑡 2 ≥ 0 and now we also found that 𝑐 ≥ 0. So, 2𝑡 2 + 𝑐 ≥ 0, which means
any time that is 0 or greater than 0. So, we are looking at the timer being started at a particular
time and from there on, if this is 0 from there on your function is well defined and the number
of likes will be greater than or equal to 0. So, the domain will be all the time from 0 seconds to
∞.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 27
Solution of Quadratic Equation Using Factorization

(Refer Slide Time: 00:15)

So, in the last video, we have seen how to find the roots of a Quadratic Equation by
graphing the associated quadratic function.
(Refer Slide Time: 00:37)

In this video, we will see how to find the solution of a quadratic equation by a well
known method called factoring method. For that, we will define one new form of a
quadratic function that is intercept form.

What is an intercept form? If which is a quadratic function is written in this

form , where and are called binomials ok. Where this p and q are nothing but x
intercepts of the quadratic function.

So, essentially what you have done is, you have seen a graph of a function and you have
located the two intercepts x intercepts of the function. Whenever the expression is

possible in this form you are writing it. So, , and this form is called the
intercept form.

So, let us try to see one example of intercept form which is let us say you have been

given this intercept form . And you also know that these 1 and 5 are x
intercepts of the quadratic function, that means, you have been given two values. Can
you find the third value? The answer is yes.

So, at point 1, when , the value is 0; at point 5, the value is 0. Now, using the logic that I
gave you in the previous video, you can actually see that there will be some axis of
symmetry between this 1 and 5, because 1 and 5 both take value 0 right. So, the axis of
symmetry will be nothing but the distance between these two points divided by 2. So,

. I am sorry it will not be a distance, it is just the sum of these two points
divided by 2, average of these two points, that should be the correct terminology. So, it

should be average of two points. So,

Now, will be the axis of symmetry for this particular function if at all the this are
the roots of the this are the x intercepts of the function right. So, now, I have given you
is the axis of symmetry. So, just substitute the value 3 in this particular graph in
this particular equation, and you will get 3 - 1 = 2, and 3 – 5 = -2, that means,
. So, you got three values. What are those three values? (1,
0), (5, 0), (3, -12).

So, based on this information, you can easily plot the graph which will look like this. As
you can see this is the value -12 here, value -12 is here and 0, 0. So, you can easily plot
this graph. You can connect the smooth curve using this.

Now, the main question is once given this kind of expression, how to convert this
expression into a standard form? So, that is the question that I will post down. How will
you convert the intercept form into the standard form? Just by multiplying the two
binomials. So, for multiplying the two binomials, we have one rule which I will state in
the next slide.
(Refer Slide Time: 04:17)

So, how the conversion from intercept form to standard form will happen using a method
called FOIL method which is described below. So, the product of two binomials as I

already mentioned that and these are the two binomials. So, the product
of two binomials is sum of the product of first terms the outer, the inner and the last
terms. So, let me make it more precise by demonstrating it.

So, let us consider this expression which is . Now, what I will do is I will
first take the first term of this expression, and the first term of this expression, and
multiply them together that is the first term over here by the sum of the product of the
first terms I mean this term.

Then I will take the inner term ok, then I will take the outer term, sorry, then I will take
the outer term that is is the outer term here, sorry, not is not the outer term, is
the inner term. You have which is the outer term, and which is the outer term. So,
now you just multiply them together which gives me which is the outer term
product of the outer term.

Then you take the inner terms that is and . So, , this is the inner term. And
are the last terms. First term, outer term, inner term and last term that way we will
multiply these things together. That means, I will get the first term as ; second term

is ; and the third term as .

Now, if you look at, so basically the is the term which is the coefficient of ,
is the is the term which is the coefficient of , and is the term
which is the which is the constant term right.

So, now, a quick observation you can make is if you look at the product of this first term
and the last term, what will you get , so it is , the product is right.

So, the product of the coefficient of and the coefficient of constant is . In a

similar manner, if you consider the coefficients of , a d and b c are the coefficients of
x, , so if you take product of these two terms, again it will be . This is a
crucial observation which we will need while converting a standard form to intercept
form and vice versa.

Just remember this the product of the coefficient of and the coefficient of constant is
. And the product of the two terms of the coefficients of is ; both of them
are . So, this we will use to convert our expression into standard form, and convert
our expression in intercept form in various ways. So, that observation is very crucial for
us.

(Refer Slide Time: 08:03)

So, let us do take one example and see how we can apply our knowledge which we have
gained in this particular video along with the previous videos to solve this problem. So,

the question is the write a quadratic equation with roots 2, and -4 in the standard

form ok. So, let us recollect what is a standard form. Standard form is of ;
and a, b, c all are integers; and . This is the standard form; we have already seen
that ok.

So, now, if I want to write this, we will use our knowledge about intercept form, and we

can easily write this expression as because and -4 are the roots.
Yes, but this equation is not in the standard form. So, now in the previous slide, we have
seen that in order to convert this into a standard form, we will use a FOIL method. So, let

us try to use a FOIL method. So, the what is here? that is a is 1, b is , c is

1, d is 4 ok.

Now, you use FOIL method that is first terms. So, first terms is , so it will retain

, Then , so and a is 1, 4 that is the product here

this is the term which have coefficient of , and then which is the term here.
So, this is successful application of FOIL method.

Now, let us rewrite all these things that is you can sum this and write the sum that is ,

so , so right. Is this equation in the standard form?

No, because for standard form a, b, c, all must be integers. So, what I will do is, I will
multiply this equation with 3 on both sides. So, if I multiply on both sides with 3, then I

get that is the solution to this question. So, the quadratic equation in

standard form is . So, we have solved.

(Refer Slide Time: 11:09)

Let us now go further and try to see how I will convert a standard form into an intercept
form. Again we will use FOIL method, but in a reverse manner. So, I want to convert a

function quadratic function which is given to me to intercept form, that

means, I want to write . So, how will I convert this?

So, what I will do is, I will take this particular function , and apply FOIL
method to it. How to apply FOIL method to it? I will equate this to be equal to

. Then based on FOIL method, I have this expression which is sorry

. Now, remember we have done some observations that is the

product of this and this is right.

So, now, I can equate this equation with this equation. So, term containing will be

equated with term containing . So, I will get , and . Then

from this expression I can also derive an information that is that is the product of
the first and the last term and the product of the terms contained in the sum is 30. So,
; and .

Now, my job becomes crucial. My job is to guess what those two terms will be ad and bc
right. So, that their product is 30, and if you sum over them, then it must be -13. For that
I will use the prime factorization theorem that was introduced in week-1. So, if you look
at this expression 30, I will get prime factors as .

Now, I want the product to be equal to 30, and I want the sum to be equal to -13. So,
based on this, what I can derive is if at all this, this term has to be negative, I should have
some negative factors over here and both of them should be negative factors. In
particular if I combine 5 and 2, I will get 10 and 3, and 10 + 3 =13, but it is not giving
me -13.

So, I will use a trick that multiplication of two negative numbers will become a positive

number. So, it is which will give me 30; at the same time, it will be the sum
will be -13. So that means, my ad is -10; bc is -3. It does not matter, you can switch also.
You can write bc as -3, and ad as -10 also, it does not matter.

So, now I will substitute these values into this expression, essentially I will rewrite this

expression. So, I will write this expression as ok. Then what I will do
is I will look at the first two terms; first two terms, and I will take the greatest common
factor from these two terms that is . So, I will take , whatever is remaining I will

put in a bracket that is .

Here also I will do take the greatest common factor out that is -3, so . Now,

you can see these s are same. So, essentially this expression will come if I have

. Now, is this in the intercept form? No, still it is not in the intercept form.

What is the intercept form? It is . So, I will just divide everything by 5 in

this expression and take the 5 out. So, this is the intercept form.

So, using FOIL method, I have converted this expression into an intercept form. An
expression was given to me in standard form; I have converted it into intercept form. Let
us see few more examples as this concept is quite intricate. You may need some practice,
you solve as many problems as possible, but I will give you some demo cases, so that it
will be easy to distinguish for you.
(Refer Slide Time: 16:05)

So, let us take let us say you have you have been asked to solve this equation; .
Now, here you do not need, you do not really need a FOIL method. What you need is,

just rearrange and you just take out the greatest common factor which is . So,

this will give you . So, if I want to solve this, I know and are
the things. So, 0 and 8 are the roots of this given quadratic equation. Simple, this solves
our problem for such a simple case, where the constant term is absent right.

Now, let us take another example, . Now, in this case, you will use FOIL
method obviously but the essence of FOIL method reduces to that the coefficients of
are of the form , and the product of this and this is . So, the product
is 4, and is -4, this is what it reduces to ok.

So, if abcd is 4, and ad + bc is -4, is there any other way out 4 can be factorized only in
one way that is ; . And ad + bc is -4, that means, both of them should be
negative . So, ad is -2, bc is -2. Substitute it in the master equation where you

can write . So, you have substituted it in master equation.

Now, you go ahead and take out the greatest common factors out, the first expression
will have out, the second expression will have 2 out. Then again these are product of
binomials. So, it will be = 0 given in the expression. So, what is the root of this
equation? 2; 2 is the real repeated root of this equation.

Let us go ahead and solve one more example. . This is quite interesting 25,
you can see is a perfect square 5, and I want to find the root of this equation. Again I will
use FOIL method, abcd is -25, and ad + bc is 0 right. 5 is a perfect square. So, is
the factorization, but -25 is there. So, one will be, one 5 will be with a positive sign,
another 5 will be with a negative sign. So, ad can be +5, and bc can be -5 or vice versa, it
does not matter.

Substitute this substitute this knowledge into this expression. So, you take

, take out the greatest common factors that is x and 5 respectively, you
will get this kind of expression. And then you just rewrite them as product of two

numbers that is , and you have solved. Remember in all these

expressions we have written this in intercept form.

So, all these expressions are written in intercept form. Once you write an expression in
intercept form, it is very easy to find the roots of the equation, or in fact once you write
in the intercept form you have already figured out the roots of the equation.
 Mathematics for Data Science 1
Week 05 - Additional Lecture
(Refer Slide Time: 00:34)

Hello, everyone, today we will discuss a small topic related to the forms of parabola. In other
words we are going to see the relation between the standard form of a parabola and the vertex form
of the parabola, we already know the standard form of a parabola which is given by 𝑦 = 𝑎𝑥 2 +
𝑏𝑥 + 𝑐 where these 𝑎, 𝑏, 𝑐 belong to real and 𝑎 ≠ 0. From this standard form we will try to derive
the vertex form of a parabola.

Now, from this equation we know that the coordinate of the vertex of the parabola is vertex will
−𝑏 𝑏2
be 𝑥 coordinate will be and 𝑦 coordinate will be 𝑐 − 4𝑎, this we already see in the previous
2𝑎
−𝑏
lecture. Now, let us denote this coordinate of the vertex as (ℎ , 𝑘). So, our ℎ will be nothing but 2𝑎
𝑏2
and 𝑘 will be 𝑐 − 4𝑎 .We have obtained the required data, now let us start the deriving.

𝑏
We have 𝑦 = 𝑎𝑥 2 + 𝑏𝑥 + 𝑐, I will take a common from these two terms, I will get 𝑎 (𝑥 2 + 𝑎 𝑥) +
𝑏2 𝑏 𝑏2 𝑏2
𝑐 , also I will add and subtract 4𝑎2 to this term, I will get 𝑎(𝑥 2 + 𝑎 𝑥 + 4𝑎2 − 4𝑎2 )+c.

𝑏
Now, also I will multiply with 2 and divide by 2 here, now I will rewrite this 𝑎(𝑥 2 + 2 × 𝑥 × 2𝑎 +
𝑏 𝑏2 𝑏 𝑏
(2𝑎)2 − 4𝑎2 )+c. So, if we observe this 𝑥 2 + 2 × 𝑥 × 2𝑎 + (2𝑎)2 , this is in the form of 𝑝2 + 2𝑝𝑞 +

𝑞 2 , we can write this as (𝑝 + 𝑞)2 .

 𝑏 𝑏2 𝑏 2
So, writing like that, we get (𝑥 + 2𝑎)2 − (2𝑎)2 + 𝑐. Now, if I multiply 𝑎 we get 𝑎[(𝑥 + 2𝑎) −

𝑏2 𝑏 2 𝑏2
]+c, a and this cancelled, finally I will obtain this will be equal to 𝑎 (𝑥 + 2𝑎) + 𝑐 − 4𝑎.
4𝑎2

𝑏2 𝑏
If we observe 𝑐 − 4𝑎 is 𝑘 here and 2𝑎 will be − ℎ, so if we substituted ℎ and 𝑘 in this equation we

get 𝑦 = 𝑎(𝑥 − ℎ)2 + 𝑘 , this is the vertex form, vertex form of the form of a parabola, where this
(ℎ , 𝑘) is the coordinate of the vertex of the given parabola, this is the standard form and from this
standard form we derived the vertex form.

(Refer Slide Time: 05:43)

So, we have got the vertex form of the parabola which will be like this 𝑎(𝑥 − ℎ)2 + 𝑘 where (ℎ , 𝑘)
is the vertex of the parabola. Now, let us see one example to understand this vertex form clearly.
So, suppose we have an equation of a parabola given like this 𝑦 = 3𝑥 2 + 6𝑥 + 9 now we try to
write in vertex form, so I will take 3 common from the first two terms, I will get 𝑥 2 + 2𝑥 + 9, so
in order to make this a perfect square I will add 1 and subtract 1.

So, 3(𝑥 2 + 2𝑥 + 1 − 1) + 9, so 3 times this can be written as 3((𝑥 + 1)2 − 1) + 9 , which gives
us (3(𝑥 − (−1))2 + 6). So, we have got the equation 𝑦 = (3(𝑥 − (−1))2 + 6). So, if we equate
it with this vertex form we get ℎ = −1 and 𝑘 = 6. So, our vertex will be at point (−1 , 6) is the
vertex of the given parabola.
So, we will just cross verify it, we know that if we have a standard form we can calculate the x
−𝑏
coordinate of the vertex, so x coordinate of this vertex will be 𝑥 = , so here 𝑏 is 6 and 𝑎 is 3, so
2𝑎
𝑏2
if I substitute that−6 𝑏𝑦 2 × 3 which I will get 𝑥 = −1 and we know the y coordinate as 𝑐 − 4𝑎

here we have c is 9 − 𝑏 is 6 so 𝑏 2 is 36 / 4𝑎 is 3 so 4 9’s 9 3’s, so I will get 6. So, my vertex point

will be at (− 1 , 6), if we solve y, solve through standard form also.

(Refer Slide Time: 09:02)

Now, let us see one more example, find the equation of a parabola such that it passes through the
origin and the vertex of the parabola is at (1 , 2). So, as we know the vertex form given by y is
equal to a times x minus h whole square plus k, here we have given that (ℎ , 𝑘) is nothing but (1 ,
2).

So, if we substitute that our equation will be simplified to 𝑎(𝑥 − 1)2 + 2, also it is given that this
equation passes through the origin that means 0 , 0 should satisfy this equation. So, if we substitute
it we get the value of a, so 0 = 𝑎(0 − 1)2 + 2, this implies 0 = 𝑎(−1)2 + 2, which implies again
𝑎 = − 2.

So, our final equation of the parabola will be equal to 𝑦 = −2(𝑥 − 1)2 + 2, if you open that 𝑦 =
−2𝑥 2 + 4𝑥 − 2 + 2, so this will be get cancelled and 4𝑥 − 2𝑥 2 , so 𝑦 = 4𝑥 − 2𝑥 2 is the equation
of the parabola that passes through the origin and the vertex of this parabola will be at (1 , 2).
Thank you.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Solution of quadratic equation using Square method
Indian Institute of Technology, Madras
Lecture 28
(Refer Slide Time: 00:14)

In this video we are going to learn one more interesting method called Solving quadratic
equations or for finding the roots of quadratic equation. The method named completing the
Square method also it has a very good connection with a very well-known or very popularly
known as Quadratic formula. So, we will explore that connection towards the end of this video.
(Refer Slide Time: 00:44)

So, let us start I will demonstrate this method through some examples. So, let us first understand
or revise what we done in the earlier stage or in the earlier video. We have used a method of
Factoring that I have called as old method. So, let us take this example If you

use the method of factoring you need to identify what is this term 24 and 1. So, so - 24
is the product of the leading coefficient and the constant term and .

So, I have this setup which is and . So, I will essentially use a prime
factorization theorem and get the prime factors of -24 so that if you rearrange the prime factors
in such a way that the sum should be equal to 10. One such rearrangement is 12 and -2. So, ad is
12 and bc is -2. So, I got this and then based on our factorization technique I will substitute this
12 and 2 for this coefficient if x and I will get this expression which is .

Then I will use the greatest common factor technique that is I will take out x common, 2
common from the last 2 terms and I will get this expression. So, finally, I got
and therefore, I will get the roots of the equation are -12 and 2. Now, somebody came up and
thought, that why should I bother what is this last term? It is just a constant right, so I will
replace this constant with something and I will work on it.
So, from that particular though comes the new method, which is the method of completing the
square. So, what that person did is just rewrite this expression into this form that is
. Now, the next question that person asked is if I look at this , do I know something that
will make this particular expression as a complete square. So, what do I mean by complete
square let us understand?

So, complete square means if I want to write then what I need to do here is to
add some number and subtract some number or to add this same number on both sides right. So,
in this case if you look at this expression that is which is . So, this a is the
number that I am looking for. Now, in this case if I consider this expression and if I want to add
a number which will typically be what that should be, Is the first question.

So, to answer that let us equate this 10 to this 2a, so therefore is the answer. So,
what will be ? will be which is 25. So, now I got a number to add and subtract
from both sides. So, what I will do is I will add the number 25 on both sides so once I add the
number 25 on both sides for this expression. I get , it turns out here in this
case that the number is 49 which is also a perfect square.

But it need not be the case all the time. So, now what I know here is this number this particular
expression is nothing but from this formula. Formula that is given here and then what is

other side is . So, I can rewrite this expression as wonderful.

So, I got something in terms of squares, now had it been only one square then in the situation
was easy I would have equated to . But there will be two situations because both the
terms on the left-hand side and on the right-hand side are squares. Now, you just write
and that will give us four cases.

But if you look at these two expressions, they will eventually reduce to the same two expressions
that is . So, it is sufficing to consider only two equations . Once I have
considered this then I know the solution right, so it is just a matter of substituting the values and
doing some little bit of algebra. So, you subtract 5 from both sides so which will give
me 2, and which will give me -12.

These are the roots of the quadratic equation, these exactly match with the roots that we have got
-12 and 2, and here 2 and -12. So, the solution set is same therefore now it is a personal choice
which method to prefer but what is a choice that is available if you have some difficulty in
factoring this. Let us say this is not 24 but some absurd number and you have some difficulty in
factoring this finding prime factorization.

What you are doing here is you are not using this particular property when you are doing this
example. When you are solving this example through this method you are not using this
particular property so you can get rid of this property and you do not have to worry about. One
note of caution is you cannot use this method when the number given here becomes negative in
this side because square root of a negative number is not defined.

So, after adding this a and the number still remains negative you cannot use this method because
according to this method there is no real solution whereas we do not know. So, this method had
some limitations but it is quite powerful in solving the problems okay. We will come to how to
overcome the limitations in a certain in the next few slides and we will see its beautiful
connection with the quadratic formula.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 29
Quadratic formula

(Refer Slide Time: 00:15)

So, let us now go when the right-hand side is not a perfect square. In this case the right-
hand side was a perfect square. So, when right hand side is not a perfect square, let us

take this expression where; . So, I have already done the first step, I

have added it. So, you can see the left-hand side is nothing, but which is equal to
32 and 32 is not the perfect square.

So, in such case what will happen? So, you will equate you will go in a similar manner

, you will take a positive square root of the left-hand side and .

can be decomposed into . So, is 4. So, it is .

So, you will get 2 two roots; are the roots of the quadratic equation and they

are irrational roots because, is an irrational number ok. It is interesting to verify this
result using a graph because, that will give us the clear cut understanding where does this

are mapped. The two green dots over here represent the location of the roots ok.
So, this is how you will solve a quadratic equation using the method of completing the
squares.

Now let us explore the relation between quadratic equations method finding the roots
using quadratic equations. Sorry; finding the roots of quadratic equations using
completing the square method and its connection with quadratic formula.

(Refer Slide Time: 02:05)

So, for this let us take a general quadratic function equated to 0 that is; .I
n is the second step is because . I can easily divide a I can easily divide by a; so
that will give us the second step.
Now, as far as we can understand what we have here is, is the constant term. So, as

per if we go by our method of completing the square, we will push this on the other

side. So, it will take a negative sign that is what you are seeing here .

And, now, was the term was the term, if it is a complete square will have

come. So, I will get a term which is our a in the earlier expression that will be . So,

will be , which is the term that I will add on both sides. So, I have added on

both sides .

Now, look at this expression carefully, what is this expression? This is and

this is some constant. So, I will use this that is; is equal to we can rearrange

this term, is the LCM. So, you just multiply by 4a over here you will get

there and divide by 4a so you will get . This is what you will get if you
rearrange these terms hm. Sorry, it is wrong.

It is it is. This is wrong. It should be and then once you take the square

root of this then you will get .

So, effectively using the same method of completing the square, the root of this equation

will be ; as easy as this. So, this method is very powerful; this is

what we have done using method of completing the squares and it gives us a general
formula which is called quadratic formula. So, this formula is known as quadratic
formula and the term over here in the square root is known as discriminant.

What is the quadratic formula? This complete expression on the right-hand side is a

quadratic formula. And the term in the square root since is called the
discriminant. Why? Because it discriminates.

Let us see, if this ; that simply means we have two real roots. If this

; that means, we have only one repeated root. And if , then we

are actually taking square root of a negative number which will go to the complex
domain. So, it has no real roots.

So, let us summarize this method or the summary of this method into a table.

(Refer Slide Time: 06:03)

So, value of the discriminant suppose and the discriminant is a
perfect square; that means, I know the square root of it then we have two real rational

roots. If , but it is not a perfect square; then I have two real irrational roots.

We have already seen in week 1 that real number line is divided into rational numbers

and irrational numbers. So, this is the splitting which will help. So, if , but

not a perfect square I will get irrational number. If is a perfect square, I will

get a rational number. If , then I will get one real rational root. And if

, I do not have any root.

Let us demonstrate it through some graphs which we have seen which we have already

seen. So, here is the example; where . These are the two roots of this
quadratic equation which are given here.

Let us say , this is the root only one root it has and it is repeated. And if

, our example was . So, in this case right it never touches the

minimum; so you will get this particular expression ok. , it never really
touches the x axis. That is the verification that we have something like this ok.

So, let us go to the next slide which is actually yeah of course, these are the conditions
that are required where a, b, c are rational numbers because, I am telling that this is this
is will be a rational root. So, if it is not rational then what you need to do is; you suppress
this, you do not need to say anything about this.

If they are rational numbers then whatever I am saying over here is true and whatever I
am saying over here is true. If they are not rational numbers still you will have two real
roots, but I cannot say whether they will be rational or irrational. And you will you may

have one real root, but it can be irrational also. For example, or and
then still you will get root as a which is an irrational number.
So, that is so in order to distinguish between rational and irrational you need a condition
that a, b, c are rational numbers. If you do not want to distinguish between rational and
irrational numbers you do not need this condition. You can have a, b, c as real numbers;
only condition that will prevail is .

(Refer Slide Time: 09:21)

So, let us go ahead and see some examples where I will use the discriminant formula or
quadratic formula to distinguish between the roots. So, the question itself says; find the
value of discriminant for each equation and then describe the number and type of roots
for the equation.

So, let us take the first example is . So, I want to evaluate .

So, it is b is -12, a is a is 9 and c is 4. So, I want to evaluate for this. In a

similar manner let us take the next equation which ; where b will be
16, a will be 2 and c will be 33.

Let us evaluate for the first equation that is; . So, 9 4s are 36,

36 4s are 144 and is also 144. So, 144 - 144, if you refer to the previous table it has
only one repeated rational root.
You go to the second example ; b is 16, a is 2 and c is 33. So, if you look at it

is 256, is 264 yes. So, I got 256 - 264; that means, I got -8. Therefore, the

, and hence it will have no real root ok. This is the summary of using the
discriminant method.

The discriminant method or the quadratic formula actually gives you a number of ways
to handle the problem. So, in short what we have seen today is, we can solve an equation
given that I know the values of a, b and c using the quadratic formula. So, let us
summarize what are all the methods that we have studied in this particular example ok.

(Refer Slide Time: 11:45)

So, the summary of concepts is; let us say I have a method which is called graphing
method. This is the method which we started with. When do we use this method? The
graphing method actually unless your solutions are integers will not give you a good
result.

But it is a best method to verify your results or verify the results that you have actually
found algebraically. If there is any calculation mistake or something it will be revealed
very easily. So, the graphing method is very helpful when you want to verify the result,
but you can also use it to find roots of the equation occasionally.
Similarly, factoring method also suffers from the disadvantage that; it may not be helpful
if the factors are not easily visible. For example, you may get the constant term to be
equal to 26.2 in the quadratic equation. The constant term is 26.2 and then, you may have
tough time in visualizing the factors.

So, in such cases factoring method need not be used, but it is very helpful if the constant
term is 0 or the factors are easy to find you can actually guess the factors if the nice
numbers like 49, 24 all these nice numbers are there then you can very well go with
factorization method.

The method of completing the squares all the time it works; it is very easy when b is
even. Otherwise, you will have problems if the right-hand side in the method of
completing the square is negative, that is you will go to complex domain and then you
may have some problems which we are not dealing with in this course. So, for us it can
be always used when b is even ok.

Now last method is quadratic formula, which is derived from which is derived from
method of completing the square. And this gives you all the time this is this will give the
answer all the time it is always helpful for. So, for our purposes when we are studying
these methods these two methods; completing the square and quadratic formula will
always give the answer irrelevant of whether the coefficients are rational numbers,
irrational numbers, or some absurd numbers ok.

Now, let us go to one more concept which is called axis of symmetry. I have not given
any derivation about axis of symmetry yet.
(Refer Slide Time: 14:33)

So, let us start with axis of symmetry. We already know while graphing the graphing the
quadratic function it is very important to know the axis of symmetry. And we have

boldly claimed at is the axis of symmetry. Now I will answer the question why

is the axis of symmetry. This is an application of method of completing the

square.

So, let us assume that I have been given a general quadratic function .
. So, I will pull out a common and therefore, my expression will become

. Now, when I was completing the square I was throwing on the right
hand side, but this time that provision is not there.

So, I will retain only thing is I will split the entire expression. So, when I split the

entire expression, I will get a as it is and this expression as it is here should come;
that means, I will add and subtract ok. And therefore, I will get this
expression. Once I get this expression, I will recognize this term this entire term so these

three terms together as ok.

Now you look at this equation correctly. For example, this is a quadratic function written
in a different form. What is this number? This is some constant ok. This is some constant
and now you look at this number this is actually deciding the symmetry around x
symmetry on x axis.

If you put as one vertical line; everything because it is symmetric about that

y axis, everything will be symmetric about . This expression defines the

symmetry of the relation or the symmetry of the function because this is nothing, but just
a constant on y axis.

Therefore, this x is equal x. So, basically you will write . So, that

vertical line is the axis of symmetry for this expression. So, the symmetry about .
Therefore, this is known as axis of symmetry that answers the quadratic equation axis of

symmetry question. Why is the axis of symmetry?

This ends our topic on quadratic functions and quadratic equations.

 Mathematics for Data Science 1
Week 05 - Tutorial 01
(Refer Slide Time: 0:14)

Hello Mathematics students, in this week’s tutorial we will look at question related to quadratic
functions. In our first question here we have two quadratic functions given to us and they intersect
each other at two points and what will be there 𝑥 coordinates. Clearly if they intersecting each
other that means the x and y will be same. So, that is mean 𝑦1 = 𝑦2 and this is what we are trying
to solve for.

So, 𝑎1 𝑥 2 + 𝑏1 𝑥 + 𝑐 should be equal to 𝑎2 𝑥 2 + 𝑏2 𝑥 + 𝑐 and the 𝑥 is supposed to be same. So,

anyway we can cancel off the c here. So this gives us (𝑎1 − 𝑎2 )𝑥 2 + (𝑏1 − 𝑏2 )𝑥 = 0. This would
imply this is us 𝑥[(𝑎1 − 𝑎2 )𝑥1 + (𝑏1 − 𝑏2 )] = 0. So, this corresponds to two different solutions.
So, if we took this part to be 0 then 𝑥 = 0 as one solution.

𝑏 −𝑏
And the next will give us 𝑥 = −(𝑎1 −𝑎2 ). So, this is a product of two terms and that product of two
1 2

terms is 0 which means either of two terms has to be 0. So, one solution is x being 0 and the other
one you get this as the solution. Now this is only a valid solution if 𝑎1 and 𝑎2 are not equal because
a denominator cannot be 0. Therefore, 𝑎1 ≠ 𝑎2 is a condition that needs to be satisfied.

𝑏 −𝑏
So, these are the two 𝑥 coordinates one is 0 and the other is 𝑎1−𝑎2 .
2 1
 Mathematics for Data Science 1
Week 05 - Tutorial 02
(Refer Slide Time: 0:14)

We are supposed to use this information this particular table to solve question 2 and 3. We will do
a question 2 now. And this table will give us the variation of approximate temperature T. So, this
is a temperature T in ℃ at a particular place with time small t, so this is the time. So, the time and
the respective temperatures are given in this table.

And Anshu fit a quadratic equation for temperature during day time as 𝑇(𝑥) = −0.4𝑥 2 + 5𝑥 +
25. Where 𝑥 is the number of hours after 8 am. So, 𝑥 begins from 0 for 𝑥 = 0. If we wrote
additionally here this is 0, this is 1, this is 2, this is 3, this is 4, 5, 6, 7, 8, 9, 10, 11 and 12. So, we
have x going from 0 to 12. If she will not, so if Anshu will not go out of her home the temperature
is greater than 40 degrees.

So, greater than 40 degrees and Anshu will not go out of the home. Then what is the minimum
time gap when she will not go out? Which means what is the time when the temperature is greater
than 40. And this is on the basis of this particular quadratic equations. So, we are essentially trying
to solve this as 𝑇(𝑥) > 40. So, that means −0.4𝑥 2 + 5𝑥 + 25 > 40.
(Refer Slide Time: 2:23)

And that would indicate that 0.4𝑥 2 − 5𝑥 + 15 < 0. So, if we took all the LHS to the RHS this is
what you will get and this is an upward facing parabola. So, the parabola will be like this and we
are looking for the portion where you have the value, the y value to less than 0. So, that would be
−𝑏±√𝑏 2 −4𝑎𝑐
happen between the roots for this we find out the roots using the formula .
2𝑎

And how did we know that this parabola is upward facing because a is greater than 0, a here is 0.4
b is -5 and c is 15. So, these roots will come out to be 5 ± √25 − 4 × 0.4 × 15 that is 16×0.4 that
is 6 ×4 that is 24. So, divided by 2𝑎 is 0.8.
(Refer Slide Time: 3:46)

So, our roots are 5 plus or minus 1 divided 0.8 which is one is 6/ 0.8 and the other is 4/ 0.8. So,
that gives us the roots as 7.5 and 5. So, these are the roots 5 and 7.5 and that means, this condition
that is the temperature being greater than 40 is satisfied between 5 hours and 7.5 hours. That would
be from here till somewhere in between here that is 15, 30. So, from 1 pm to 3:30 pm is the time
suggested by the curve fit that Anshu has drawn but clearly this is wrong because it is already 43
here, and 48, here and 46, and 43, and 40, and 40 so it is a much larger duration where the
temperature is greater than 40 degrees Celsius. So, this particular curve fit is pretty bad.
 Mathematics for Data Science 1
Week 05 - Tutorial 03

(Refer Slide Time: 0:14)

A third Question is related to the second question. So, in case you have not the second question
please go back and see it. And here we are trying to say that instead of fitting a quadratic we can
fit two linear equations l1 and l2. They have already provided us with the two equations which are
this and this. So, 𝑙1 : 𝑦 = 3𝑥 + 25 and 𝑙2 : 𝑦 = −3𝑥 + 67 and the curves are already given.

Now the question is asking us to draw rough sketch of the quadratic equation that was fit and the
vertex is also provided for us with respect to these two lines. So, I think it is useful if we can just
mark out the points here. So, (0 , 30),(7 , 46) and (12 , 32) are already given. The remaining ones
were, this is (1, 32), this is (2 , 34), this one is (3 , 36), this one is (44 , 40), this is (5 , 43), this is
(6 , 46). So, this question has a problem here this should be (7, 48) that point.

And this is (8, 46) again and this is (9 , 43) and this point is (10 , 40) this is (11 , 35) and that may
have (12 , 32). So, these are points and for us to do the rough sketch. The vertex is at 6.25 so the
vertex should somewhere here and it is at 40.625. So, this is the horizontal (4, 40), then vertex is
somewhere here. So, clearly our quadratic is below the points that we have been given. And this
being the 𝑥 2 coefficient is −0.4 which is less than 0 so it is a down turned parabola.
And let us look at the two points that we know for sure 0 and 𝑥 = 0 this parabola is going to give
us 25 the quadratic equation going to give us 25. Which is definitely below so somewhere here it
appearing to be intersecting with this line. So let us look at, we have 3𝑥 + 25 so at 𝑥 = 0 the
quadratic equation and the l1 line meet. And 𝑥 = 12, we have 0.4 into −0.4 × ( 144 ) + 60 + 25
which give us about 27.4, I think.

So, that is below this somewhere here so our quadratic is going to look like this. Something like
this so it is quite inaccurate for the given data. So, this is a very bad curve fit.
 Mathematics for Data Science 1
Week 05 - Tutorial 04

(Refer Slide Time: 0:14)

This is the pretty straight forward question, we are given a quadratic equation and we are asked to
find the roots and also calculate the sum and product of roots. So, the roots we are going to get
−𝑏±√𝑏2 −4𝑎𝑐
from the formula again which is . So, that gives us, here in this case 𝑎 is 5, 𝑏 is 8 and
2𝑎
−8±√64−4×5
𝑐 is 1. So, we have .
10

−8+√44 −8−√44 −4+√11

So, we have and . And if we simplify it taking 2 common out, you will get
10 10 5
−4−√11
because the 4 comes out of the square root and becomes 2. And .
5
(Refer Slide Time: 1:44)

−4 √11
And the sum of these roots is if you just add them up you will ge𝑡 + . So, these get canceled
5 5
−8
so you get is the sum. And in terms of product you basically doing (𝑎 + 𝑏) × (𝑎 − 𝑏) so you
5
−4 √11 2 16−11 1 −8
will get ( 5 )2 − ( ) . So, that gives us = 5. So, this is the product of the roots. and
5 25 5
1
are just sum and product of the roots respectively.
5

(Refer Slide Time: 2:43)

The question is asking us to prove that the sum and product of roots for any quadratic equation
−𝑏 𝑐
will be this and this respectively. and respectively. So, all we need to do for this is to sum
𝑎 𝑎

−𝑏 √𝑏 2 −4𝑎𝑐 −𝑏 √𝑏 2 −4𝑎𝑐
+ . This we are summing with 2𝑎 − a. So, clearly these two cancel off and you
2𝑎 2𝑎 2𝑎
−2𝑏 −𝑏
are left with , 2 and 2 cancel off and you have is sum of roots.
2𝑎 𝑎

−𝑏
And when we do the product again it is in the (𝑎 + 𝑏)(𝑎 − 𝑏) form so we will get( 2𝑎 )2 −
√𝑏2 −4𝑎𝑐 2 𝑏2 (𝑏 2 −4𝑎𝑐) 𝑏 2 −(𝑏 2 −4𝑎𝑐)
( ) which gives us 4𝑎2 − . So, that gives us , 𝑏 2 − 𝑏 2 cancels off then
2𝑎 4𝑎2 4𝑎2
𝑐
we have 4 4 going away 𝑎 𝑎𝑛𝑑 𝑎 going away so you were left with 𝑎. So, this is the product of

roots for a quadratic equation.

 Mathematics for Data Science 1
Week 05 - Tutorial 05
(Refer Slide Time: 0:14)

In this question we have 2 sets capital M and capital N which are sets of all values of small m
and small n respectively such that these two equations have always two distinct real roots each,
then find the sets M and N. Let us finish this part first. So, for a quadratic equation 𝑎𝑥 2 + 𝑏𝑥 +
𝑐 = 0 to have distinct real roots, the discriminant which is basically the value 𝑏 2 − 4𝑎𝑐 > 0.

So, for this first equation that would be 𝑚2 − 16 > 0 and simultaneously, we need for the
second equation 𝑛2 − 4 > 0. So, 𝑚2 > 16 and 𝑛2 > 4 and this would imply m is positive and
greater than 4 or m is negative and lesser than −4. And here this would imply similarly n is
positive and greater than 2 or n is negative and lesser than −2. So, these are all the possible
values for which you will have two real distinct roots for these equations.
(Refer Slide Time: 2:02)

So, your set M would be the union of two intervals, one is a (−∞, −4) ∪ (4, ∞). And set N is
similarly (−∞, −2) ∪ (2, ∞). Now the next part of the question, C is a set of integers and
values of m and n are to be chosen randomly from C, then define the set C such that both
equations have two distinct real roots each.

So, this is necessarily one single set we are taking and m and n should be chosen from that set.
So, we clearly cannot have m being −2 or 2 or even −3 or 3. The set we are looking for is
some sort of an intersection of capital M and capital N because both small m and small n should
be drawn from this. And in this case that intersection will just be the set capital M because M
is necessarily a subset of N.

However, C is also given out to be a set of integers, so it is not just the intersection of m and n,
it is the set of integers which belong to the intersection and this case that intersection is only
capital M where therefore, we have this set coming up as C.
 Mathematics for Data Science 1
Week 05 - Tutorial 06
(Refer Slide Time: 0:19)

In our sixth question, there is a sniper who shoots a bullet at some inclination from the ground
towards a bird flying in the − 𝑥 direction, some bird is flying in the − 𝑥 direction at a constant
height of 1600 feet. Because of gravity, the path of the bullet is projected as shown in this
diagram. So, this is the bullet, it is going in this particular parabolic path and this is the bird
which is going in the − 𝑥 direction at a constant height of 1600 feet.

Now, they have given the height y of the bullet at t seconds as this function, this is a quadratic
1
function 𝑦 = 𝑢𝑦 𝑡 − 2 𝑔𝑡 2 , where 𝑢𝑖 is the initial vertical speed and that is also given here, it is

equal to 400 feet per second and the value of 𝑔 is also given here, 32 𝑓𝑒𝑒𝑡/𝑠 2 . And then, the
distance travelled by the bullet in 𝑥 direction is given by 𝑥 = 𝑢𝑥 𝑡 and 𝑢𝑥 = 𝑢𝑦 = 400 feet per
second, neglecting the effect of the wind and everything, find the position of hitting?

Where will the bullet hit the bird and that would be here where y = 1600 for the bullet. So, let
1 1
us use the y equation and the y equation is 𝑢𝑦 𝑡 − 𝑔𝑡 2 . So, 𝑢𝑦 𝑡 − 𝑔𝑡 2 = 𝑦, so we know 𝑦 is
2 2
1
supposed to be 1600 and 𝑢𝑖 is 400, so we get 400𝑡 − 2 𝑔, 𝑔 𝑖𝑠 32𝑡 2 . So, 2 ones and 2 16s, now

you can cancel off 16 here with this is equal to and this becomes 100 and this becomes 25.
(Refer Slide Time: 2:34)

So, we get a quadratic equation which is 𝑡 2 − 25𝑡 + 100 = 0 and if we solve for the roots of
this equation, we get the time when y is 1600 and we will get 2 times because y is 1600 twice
on this path. So, we will get 𝑡1 and 𝑡2 , we are looking for 𝑡1 because that is where the bullet
−𝑏±√𝑏2 −4𝑎𝑐
will hit the bird. So, your two roots are using the formula , here you will get it as
2𝑎
25±√625−400
.
2

25±√225 25±15 25−15

So, that gives us which then given us . So, we have one solution, 𝑡1 = and
2 2 2
25+15
𝑡2 = . So, this is equal to 5 and this is equal to 20. Clearly, 𝑡1 =5 seconds is where our
2

bullet will hit the bird. This is 𝑡1 =5 seconds. And we already know the y coordinate of this
place so, for finding the position what is left is to find the x coordinate which we will get from
𝑥 = 𝑢𝑥 𝑡 where, 𝑢𝑥 is already given to be 400.

(Refer Slide Time: 4:23)

So, 𝑥 = 400 × 𝑡1 which is 5 and that is equal to 2000 feet.

(Refer Slide Time: 4:35)

Thus, the 𝑥 coordinate for the point of hitting is 2000 and the 𝑦 coordinate is 1600 feet. And
this is the point where it hits.
 Mathematics for Data Science 1
Week 05 - Tutorial 07
(Refer Slide Time: 0:14)

In this question there are these two curves 𝐶1 and 𝐶2 which are both quadratic curves and there
is this line l which is passing through these two intersection point. So, line l is passing through
the intersection points of these two parabolas. They are asking find 𝐶1′ and 𝐶2′ , the curves of the
functions 𝐹1′ and 𝐹2′ which are reflections of 𝐶1 and 𝐶2 respectively around l which means for
𝐶1 the reflection would be something like this, about l it would be something like this and for
𝐶2 the reflection would be something like this and these are what we are trying to find out, 𝐶1′
and 𝐶2′ .

So, this should be 𝐶2′ and this would be 𝐶1′ . For all of these, we have to first find the line l and
that we can find when we solve for the equality of these two functions. So, we are taking 𝑥 2 −
6𝑥 = −2𝑥 2 + 12𝑥. And that gives us 3𝑥 2 − 18𝑥 = 0 and that further gives us 𝑥 (3𝑥 − 18) =
0 that indicates 𝑥 = 0 or 𝑥 = 6.

So, this point has coordinate 𝑥 = 0 and this point has coordinate 𝑥 = 6. We need to find the
y coordinates for these points now. For that we substitute 𝑥 = 0 and we get in this equation or
this equation I wrote this and we get 𝑦 = 0. So, this point is essentially the origin. Whereas,
for this point we substitute 𝑥 = 6 and we get 36 − 36 which is 0. So, this point would then
be (6, 0).

So, essentially this is a horizontal line which is 𝑦 = 0, 𝑙 is 𝑦 = 0. So, now we are just looking
for reflections about the 𝑥 axis because 𝑦 = 0 as the 𝑥 axis. And that would give us directly
the negative coefficients of the same things. So, 𝐶1′ would then be 2𝑥 2 − 12𝑥 whereas, 𝐶2′
would now be −𝑥 2 + 6𝑥. Thank you.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 30
Polynomials

Let us introduce Polynomials. So, today we are going to see how the polynomials look
like, how they behave. So, let us start with polynomials. Let us go ahead and see what
expressions do we call as polynomial.

(Refer Slide Time: 00:29)

So, for that let us take the first point where we will take a Layman’s perspective and, we
will try to understand what a Layman will think of a polynomial. In order to do that let us
first take a Layman’s perspective and see what Layman will think. So, for a Layman, a
polynomial it is some kind of mathematical expression which is a sum of several
mathematical terms.

Then we asked Layman what do you mean by mathematical terms? The answer is each
term in this expression, each term in this expression, that is mathematical term in this
expression can be a number, a variable, or a product of several variables.

So, according to Layman each term this mathematical term can include a number, a
variable, or product of several variables. These are the things that are allowed. So, basically
then, let us take one example for this. And let us see if I have 3𝑥 this is a number and some
variable. So, I have a constant 3, I have some number like 𝑥 , I have some term like 𝑥 𝑦,
all this contribute to something called polynomial ok.

Now, take a more significant number that is say 𝑥 + 4𝑦 + 2𝑧 + 10; will this contribute
to be a polynomial? Yes, because it is sum of a number which is 10 here, a variable. There
are many variables 1, 2, 3, there are three variables, and product of several variables; in
particular here we have 𝑥 and here we have 𝑦 . So, this also qualifies to be a polynomial.

Then according to this, suppose what I will write is here, let us say some expression of the

form 𝑡 + 𝑡; is this expression a polynomial? Layman will say yeah, it can be a polynomial

because, 𝑡 , if you square this number you will get this, correct. So, this is one variable
this is another variable and therefore, we are actually having a polynomial.

So, then we went and asked mathematician, what is a mathematician’s perspective of a

polynomial?

(Refer Slide Time: 03:35)

So, that we will define as a definition. So, a mathematician said a polynomial is nothing,
but an algebraic expression in which only arithmetic is addition, subtraction, multiplication
and this is interesting, he mentions it as natural exponents of the variables. Natural, by
natural I mean the way we defined a set of natural numbers in our first week I mean natural
means 0, 1, 2 and so on.
So, this is my set of natural numbers which include 0, if you want to emphasize it you can
put it as 𝑁 . Otherwise, you can call this set as set of whole numbers or set of non-negative
integers. So, the definition can be twisted like this will have non-negative integer
exponents. If you do not want any ambiguity, we can say that non-negative integer
exponents of the variables.

So, then we if we go back to that earlier expression which is 𝑡 + 𝑡, the all other
expressions will qualify to be a variable, but this expression will not qualify to be a

variable, why? Because this 𝑡 by definition is not a natural number, it is a rational

number. We will come to it later, but this cannot qualify definitely so this cannot qualify
as a polynomial if we go by this definition.

We have already seen many examples. Let us re-iterate them; one example was constant
3, another was 3𝑥 , another one was 3𝑥 + 𝑦 + 4𝑧 + 10, all these are qualified to be
polynomials. But this expression does not qualify to be a polynomial. So, this we will
consider later as well and we will give our rational reason why it is not a polynomial.

So, we now we know what is polynomial. Now, it is time to see why is the name
polynomial? Why do we call this as polynomial? That is what we will see now. So, let us
go ahead and see something about the nomenclature. Why do we call them polynomials?

(Refer Slide Time: 06:17)

So, polynomials essentially is derived from two words one word is poly, second word is
nomen. This is a Greek word this word and this has roots in Latin. The word poly
essentially means many and the word nomen essentially means names or a terms. So, in
our case it turns out to be terms. So, an expression having many terms is called polynomial
ok.

Now, each term of the because it has many terms each term of the polynomial will be
called as monomial, each term of a polynomial will be called as monomial. Then, if the
polynomial has only two terms then you will call it as binomial. If the polynomial has only
three terms then you will call it as trinomial.

So far, if you can label them you can label them, but in general we will treat them as
polynomials. And remember that we will include this also; a monomial is also a
polynomial for us. We will not distinguish between monomial and polynomial. Of course,
monomial enjoy some different set of properties, but we will keep them with polynomials.

So, let us take one example. For example, a polynomial in one variable can be represented
as 𝑎 𝑥 this is the highest term, 𝑎 𝑥 , 𝑎 𝑥 , 𝑎 right. I am assuming that this 𝑎 ’s
not equal to 0. Otherwise the if they are 0, then the polynomial may extend to infinity. I
do not want that. So, I am assuming that these 𝑎 ′s are not 0.

Now, if you can rewrite this using the notation of summation in this manner and therefore,
this 𝑎 will have a specific name and it will be called as coefficient of the term. Because
this is a polynomial in one variable, 𝑥 is the variable we are interested in 𝑥 is the variable,
and this 𝑚 is the exponent of the variable.

Now, remember in order that this term to be a polynomial, this 𝑚 should always be a
natural exponent; by natural I mean the one that is non-negative integer. If it is not
nonnegative integer then I cannot classify this as a polynomial ok. Let us go to the next
step and see some examples of polynomials and try to identify whether the given
expressions are polynomials or not.
(Refer Slide Time: 09:35)

So, here is the question about identification of polynomials. Identify whether the
followings are polynomials or not. The 1st one; 𝑥 + 4𝑥 + 2, what can you say about
this? So, the first let us take term by term 𝑥 , 4𝑥 and 2. So, the all these are monomials
involved in this polynomial.

So, when I take 𝑥 it is nothing, but variable 𝑥 multiplied by 𝑥. So, it is a product of two
variables. So, this is ok. When I take 4𝑥 it is a number and a variable. So, this is also I do
not have any and finally, this is just a number. So, together and the expression given it is
sum of these that is; 𝑥 + 4𝑥 + 2, expression given it is sum of this. Therefore, this is a
valid polynomial form.

Let us go ahead ok. Again, the same expression has come 𝑥 + 𝑥 . So, now, you look at

the terms that are involved 𝑥 and 𝑥 . Now, if you look at the terms involved 𝑥 and 𝑥 ,
then this it is simply a variable raised to 1, 𝑥 , right. So, I do not have any problem, this is

a valid term because there is no issue with this; 𝑥 this term is not a valid term because, it
has rational exponent. So, this is not correct.

So, this second expression 2 is not a polynomial, why? We need to justify we need to write
a reason because, the 2nd monomial has rational exponent. This is an interesting
observation. So, this does not qualify to be, to be a polynomial. So, I can erase this. This
is not a polynomial.
Now, some people may say that what is a big deal? I can put 𝑥 , let’s say 𝑡 and you can
rewrite this expression as 𝑡 + 𝑡 , but remember when I am putting 𝑥 raised to half as t, I
am putting an explicit assumption on this 𝑥 that is; this 𝑥 should be greater than or equal
to 0. So, I cannot define this polynomial on the entire real line.

So, we will refrain from doing such assumptions and therefore, it would not be a
polynomial. Let us go to the next example, this example ok. So, let me erase the previous
ones so that I will have some space.

(Refer Slide Time: 13:15)

We will write all the terms as usual 𝑥, 𝑦, 𝑥𝑦, 𝑎𝑛𝑑 𝑥 . We will analyse the each
monomial one by one, is this valid? Let me change the colour. Is this valid? Yes, it is valid
because it is just a variable 𝑦? Yes, it is valid just a variable; product of several variables;
product exponent natural exponent of single variable? Yes, it is valid.

So, according to me this and this qualify as a polynomial. And this do not qualify as a
polynomial. So, we have identified what are the polynomials and how they look like. So,
our identification part is complete. In particular we are dealing with polynomials having
real coefficients because, all the numbers that I am giving you are real numbers.

So, just remember this fact we are only handling polynomials with real coefficients. If you
go to the further branches of mathematics you may have polynomials with simply integer
coefficients, you may have polynomials with complex coefficients, we are not dealing with
them. So, this is how we will identify whether a polynomial whether a given expression is
a polynomial or not.

(Refer Slide Time: 14:51)

Let us go ahead and try to describe what are the types of polynomials, that we can
encounter. We have already seen them; we are just enlisting them for the sake of
completeness. So, there can be polynomials in one variable which will typically look like
∑ 𝑎 𝑥 . So, that let me rewrite it.

So, that was our expression; summation over ∑ 𝑎 𝑥 . So, this particular thing falls
into that category. What is what will be here in this particular case 𝑎 = 1, 𝑎 = 1 and all
others like 𝑎 , 𝑎 and 𝑎 , all of them are 0. So, this is how we will describe the polynomial.
So, this is a polynomial in one variable.

You can encounter a polynomial in two variables. For example, we have already seen some
examples this is a polynomial in two variables and you can have similar expression, but
now, you will have 𝑎 𝑏 and 𝑎 or something of that sort, to indicate the powers of
these exponents. So, we will not indulge into a mathematical representation of this, but
you can have polynomials in two variables.

In a similar manner you can have polynomials in three variables or more than two
variables. So, here is an example of a polynomial in more than two variables. And these
are the types of polynomials that you may encounter with real coefficients. So, this
summarizes the topic of representation of polynomials.

Now, let us go ahead and see some further properties of these polynomials.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 31
Degree of Polynomials

(Refer Slide Time: 00:15)

So, in particular, if I want to tell something about a Polynomial, an important property is

a degree of the polynomial. So, what is the degree of the polynomial? For demonstration
purposes, let me take one example. Let us say my example is 3𝑥 + 4𝑥 𝑦 + 10𝑦 + 1,
this is my example.

Then, I say this is the example. So, if I want to decide the degree of the polynomial, we
have already seen each term itself is a polynomial. So, 3𝑥 is one; 4𝑥 𝑦 is one; 10𝑦 is
one and this 1 is one. So, I want to identify the degree of each term as well.

In each term, there are many variables. For example, you would look at this term, if you
look at this term then there are 2 variables. So, I want to have a complete understanding.
So, in order to define the degree of a polynomial, I will start with defining the degree of
the variable. So, the exponent on the variable in a term, the exponent on the variable in a
term is called degree of that variable in that term. So, for demonstration purposes, let us
take the expression 4𝑥 𝑦 .
So, in this particular expression or in this particular monomial, how many variables are
involved? One variable is 𝑥, second variable is 𝑦. So, what I am saying is now in this term,
the degree of 𝑥; the degree of 𝑥 let me abbreviate it as degree; degree of 𝑥 is 2 and the
degree of 𝑦, variable 𝑦 is also 2, ok. So, this is how I will describe the degree of the
variable.

(Refer Slide Time: 02:39)

Now, let us take this term as a term and say what is the degree of this term, right. So, let
me erase this particular portion which is actually blocking our view ok. So, the degree of
that term, this term, we have already seen the degree of 𝑥 is something and degree of 𝑦 is
something.

Degree of 𝑥 was 2 and degree of 𝑦 was 2, the degree of the term is the sum of the degrees
of those variables in the term. That means if I look at this expression which is 4𝑥 𝑦 , then
and I ask for the degree of this term, then it is essentially degree of 𝑥 plus degree of 𝑦 that
is 2 + 2 which is equal to 4.

So, degree of this term the second term in this expression is 4, fine. Now, we will answer
the question, what is the degree of a polynomial? So, the degree of the polynomial is the
largest degree of among these all the terms of any one of the terms with nonzero
coefficients or terms will exist only when there are nonzero coefficients.
So, let us try to see how we can solve this problem. So, we will try to list all the degrees.
So, if I take the first term that is 3𝑥 , second term is 4𝑥 𝑦 , then the next term is 10𝑦 and
the last term is 1 which is the constant ok.

So, now, we will talk in terms of degrees. So, what is the degree of this particular term? It
has only one variable 𝑥 which has which is raised to the second power. So, the degree of
this term is actually 2. What is the degree of this term? We have already seen here, the
degree of this term is 4.

What is the degree of this term? The degree of this term is again 𝑦, means 𝑦 . So, the
exponent is 1, Interesting. Now, what is the degree of this term? Now, remember this is
an expression in two variables; 𝑥 and 𝑦. So, what then, I will ask a question what is the
degree of 𝑥 and what is the degree of 𝑦?

Now, you can also see that 1𝑥 𝑦 = 1. So, degree of 𝑥 is naturally equal to 0 and degree
of y is also equal to 0, right. Therefore, I can write the degree of this expression is 0 ok.
Now, which one is the largest among these four? 0, 1, 2, 4? 4 is the largest. So, the degree
of this particular polynomial is 4. So, this degree is actually 4, then write it here 4. So, this
is a polynomial of degree 4 ok.

(Refer Slide Time: 06:31)

So, in this contest in while finding the degree of this particular polynomial, we have seen
two things. What are those two things? If the coefficient is if the variable is 𝑥, then this is
𝑥 , if it is a constant, then 𝑐 × 𝑥 . In our case because the polynomial was having 2
variables, it is 𝑐𝑥 𝑦 .

Interesting question comes when we try to see polynomials at some other things. Let us
say if I want to describe 0, when 𝑐 is nonzero, it is ok, but if 𝑐 = 0, then what? Then, you
can see 0 = 0 + 0𝑥 + 0𝑥 + 0𝑥 … , the matter is complicated further and so on right.

It will continue. So, if the point given, then this number is 0, then we will call this as 0
polynomial and we cannot define the degree of this polynomial because for a degree, we
need a nonzero coefficient, just remember this in mind.

Therefore, the degree of 0 polynomial is always undefined this is an interesting fact which
will be used when we use the division algorithm. The degree of 0 polynomial is undefined.
So, you can use it in a more interesting manner that is what I can say. So, the degree; this
is what? Degree of 0 polynomial is undefined.

(Refer Slide Time: 08:19)

So, we have understood what is the degree of the polynomial. So, in particular, based on
the degrees, now we have introduced one classification. So, based on the degrees, how the
polynomials can be classified. So, if the polynomial has degree 0, then it is constant and
this constant can never be equal to 0.
This is an important assumption. Then, if the polynomial is of degree 1, linear polynomial,
then you will have a polynomial in this form. When I write this, then I should write 𝑎 ≠
0, if I have a quadratic polynomial, if the polynomial ok.

So, here these are the polynomials in one variable, then I am considering a linear
polynomials. If I am considering a polynomial of the form 𝑥 + 𝑦, this is still a linear
polynomial; but it is a polynomial in 2 variables. So, you can also encounter such
polynomials in linear, but the crucial fact is degree is 1.

Second one is a quadratic polynomial which is of this form and here you can have
polynomial in two variables, three variables or whatever way you want. Then, you will get
a cubic polynomial which will have all terms containing degree 3, the highest term, highest
monomial will have degree 3.

So, this is these are the examples of degree 3 polynomials. Similarly, degree 4
polynomials are called quartic polynomials and they will be given in this form and
similarly, degree 5 polynomials are called quintic or quantic polynomials which will be
represented with degree 5 polynomials, right.

So, and in general, you have a general term which is called polynomial. So, to be if you
want to be specific, you can use this classification and say it is a quadratic polynomial,
then you are giving more information about it.

This is what today’s lecture meant to be. So, we have introduced the topic of polynomials.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 32
Algebra of polynomials: Addition & Subtraction

In this video, we will start with polynomials and we will try to do some Algebra with
Polynomials. Or in other words you can say we will try to understand some operations
on polynomials like Addition and Subtraction. So, let us move on.

(Refer Slide Time: 00:34)

In order to simplify our calculations, we will only focus on polynomials in one variable;
whereas all the operations that we are discussing can be done on polynomials with
multiple variables. In order to pinpoint the thing, we will recollect how polynomials in
one variable look like.

So, a polynomial of degree n in one variable can be represented in this form;

n n−1
an x +a n−1 x + .. .. +a1 x+a0 . So, you can actually correlate this with, let us say this is
a monomial with 0 degree, this is a monomial with 1 degree and so on, if you go on this
way, this is the monomial with n th degree. In order that this polynomial to qualify as
a polynomial with n th degree, we need something, we need one condition; that
condition is actually I need this to be a polynomial of the term of degree n to be non-
zero.

So, that forces me to write an ≠0 , this is a condition that require that is required for
writing a polynomial of nth degree. Remember here the argument is only one that is the
variable is only one x. So, I can also assign this as something called p ( x ) , and now
you can as well treat this p ( x ) as a function of one variable which is interesting.

So, if you assign this as a function of one variable, the next question is; how is this
function, what is the domain and co-domain and range of this function? So, the function
runs from ℝ to ℝ . So, it is a function from real line to real line; whereas the range
typically depends on function.

For example, if I take a function like let us say p1 ( x ) is one function, which is a1 x+a0
and if I take this function, then it is a linear function; we have already seen this function,
this is an equation of a real line, equation of a line. And if a 1 is not equal to 0, then this
function actually represents a real line. If a1 ≠0 , it also represents a real line, but it is
some constant.

So, it is a horizontally real line. So, now, the range of this function for a1 ≠0 is entire

real line. Whereas if you look at some other function, let us say p2 ( x ) =a 2 x 2 +a1 x+a0 .
Now, this particular function represents a parabola, which we have seen in our topic on
quadratic functions.

And you know depending on the sign of a2 , the parabola can open upward or
downward; if it opens upwards, the range is the minimum value and any point beyond
that; if it opens downwards, the range is the maximum value and any point below it. So,
depending on the choice of the function, the ranges may differ. We will deal with
polynomials as function when we will study the graphing of polynomials.

Right now, we are interested in algebraic properties of this polynomial. So, we will focus
ourselves on the algebra of the polynomials that is addition, subtraction, multiplication,
division.
(Refer Slide Time: 04:53)

So, let us move ahead and try to understand addition of polynomials. We have done this
in some sense, for example, currently the polynomial that we have written that
n n−1
an x +a n−1 x + .. . , is also addition of some kind; but it is addition of monomials. So,
let us try to see, if I have been given to two polynomials, how will I add them?

To help us in understanding and developing general theory for addition of polynomials,

we will consider these three examples. Remember, the first example is actually one
polynomial added with another monomial; second one both are two polynomials, but
there are no clashing terms, like they the exponents are different for both the
polynomials, you can check and the third one has few clashing terms.

So, we will demonstrate the addition of polynomials through these three things and we
will formalize this into a theory. So, let us start with the first expression,
p ( x ) =x 2 + 4 x+ 4 . So, we are starting with this, this particular expression. So, p ( x ) is
x 2 . So, if I am writing x 2 , then it essentially means I am multiplying this with 1, the
coefficient is 1; if I am starting with 4 x , then I do not have to do anything and this is 4.

So, in this case in our standard form for a quadratic polynomial, what is a standard form

for a quadratic polynomial? a2 x 2 +a1 x+a0 this is our standard form. So, in this

particular thing, you can identify a2 =1 ,a1 =4 and a0 =4 .

In a similar manner, I will look at this particular expression which is q ( x ) . Now, you
notice the fact that q ( x ) is just a constant polynomial, q ( x ) do not have any terms
which are related to square or related to a linear term. And, I want to add this polynomial
to a given expression.

So, how will I add? So, let us bring in the terms related to square term and related to
linear term; if I bring in those terms, the associated coefficients will be 0 right, the
associated coefficients will be 0. So, I can write this term as 0 x 2 +0 x+10 .

Now, because of this, let us write it in a generalized setting; b2 x 2 +b1 x+b0 , right. So,
now, I am trying to add these two polynomials. So, what is our recipe? We will consider
the terms with like powers that is like exponents, ok. So, let me try to add the things.

So, if I consider this, this particular expression that is given here. So, I have 1 x 2 , 1
minute. So, let me bring in my mouse pointer here. So, I have 1 x 2 + 0 x 2 . So, 1 + 0, I
will get again singleton x 2 ; then 4 + 0 that will give me 4 x , 4 +10 will give me 14. So,
essentially I can see that this expression should have a formulation which is of the form
x 2 +4 x+14 .

So, if I now try to do it in a more general settings, then how will I compare with this
general setting. Let us see it here. So, I want to add these two. So, just add. So, in a

similar manner, if I add these two; what I am getting is ( a 2 +b2 ) x 2 + ( a1 +b1 ) x+a0 +b0 ,

just to remember this format. So, what I am writing here is essentially.

Another point that to note with this example is; the first one was a polynomial of degree
2, the second expression q ( x ) was a polynomial of degree 0. Now, the resultant
expression that is p ( x ) +q ( x ) , what is the degree of this polynomial? It is a polynomial
of degree 2. So, it is the maximum of degree of the first polynomial and degree of the
second polynomial. So, we have roughly understood the settings that we need maximum
of 1 and 2, let me write the findings in a different way.
(Refer Slide Time: 10:49)

So, if I have a polynomial of degree m and let us say if I have a polynomial of degree
m and degree n, these are the two polynomials; and if they satisfy a relation that
m is less than n , then the resultant polynomial will have a degree n . If I have a
polynomial where degree m is equal to n ok, then the resultant polynomial will
still have a degree n .

And if I have a case where m is greater than n , then the resultant what will be the; so
if we switch this q ( x ) to p ( x ) and p ( x ) to q ( x ) , this case will happen and the
resultant polynomial will have degree m. So, just remember this in mind; that means
it is always maximum of m and n , if the resultant polynomial is having
polynomials underlying polynomials with different degrees. So, with this understanding,
let us attack the second problem.

So, the second problem has p ( x ) which is x which is a polynomial of degree 4. So, I
have written all other terms which were not there in the polynomial by multiplying with
0. In a similar manner, I have written the second polynomial q ( x ) which is a polynomial
of degree 3; but we want the maximum degree to survive right or is essentially in this
expression the maximum degree will survive, therefore I have this kind of expression.

So, the resultant polynomial we know from our discussion will be a polynomial of
degree 4, and therefore I need to consider all the terms that correspond to each of the
degrees. So, what is that term corresponding to degree 4? In the first expression that is
 p ( x ) is 1, the coefficient is 1 and the term corresponding to degree 4 in the second
polynomial that is q ( x ) is degree 0.

So, I will get 1 + 0, which is 1; 1 x 4 . In a similar manner you can see, for x 3 it is 1 x 3 ;
2 2
x there is no survivor both are 0, so 0 x , 4+ 0 x and 0 + 1 times 1. So, it is 1. So, the
resultant that you are interested in is x 4 +x 3 +4 x+ 1 . Again I will reiterate, this time it

will be; if you consider a generalized polynomial, it will be a 4 x 4 +a 3 x 3 +a2 x2 +a1 x+a 0 .

And in a similar manner q ( x ) will b 4 x 4 +b 3 x 3 +b 2 x 2 +b1 x+b0 . And if you sum over
them, what we have written in yellow is essentially sum of
a 4 +b 4 =1,a 3 +b 3 =1 ,a1 +b1 =4 ,a0 +b 0 =1 ; a2 and b2 will sum to 0, because it does
not have any non-zero coefficient. So, this is how we will handle the third, this is how
we have handled the second problem.

And the term containing the highest degree survive, therefore the degree of polynomial is
4, ok. So, let us go back to the third problem, let us come ahead and solve the third
problem; p ( x ) and q ( x ) , again similar setting highest degree is degree 3. So, the term
corresponding to degree 3 will survive. So, the polynomial with lower degree, I will
bring it to degree 3. So, essentially, I will multiply with coefficient 0 for a degree 3 term.

Again by same logic, I will add the two terms and therefore, I will get the corresponding
answer. So, there were cross terms, like the term corresponding to x 2 was crossing; for
example, both polynomials had terms corresponding to x 2 .

So, you can see the difference here, we are just adding 2 + 1, 1 + 2. So, all these things
are happening and together we are writing the result x 3 +3 x 2 +3 x+2 . So, from this we
can derive, if you are clear with these three examples then we can derive a general
formula; otherwise pause and look at each of the terms, you will be able to understand
the general formula in a bit better manner if you pause and review these additions.

So, let us come to the general formula, you must have paused and understood the

n
additions. So, if I have a polynomials of the form ∑ ak x k . And if you assume that
k= 0

an ≠0 , then this is a polynomial of degree n.

 m
In a similar manner you have taken a second polynomial q ( x )=∑ b j x j . Now it does
j=0

not matter whether m is greater than n or m is less than n, this particular

thing will give you the answer, ok. What is the resultant? So, if I want to add these two
polynomial functions p ( x ) and q ( x ) , then p ( x ) +q ( x ) will essentially show this kind of
representation.

So, what we are actually saying is, choose which one is the maximum m or n ok?
Whichever is the maximum? Take that maximum. So, it is maximum of m and n ;
sum will run from k is equal to 0 to n . Let us say for argument purposes this is the
highest degree that is n is the highest degree, then match the degree of the highest
degree for other coefficients, for example, for j equal to m+ 1 to n put all
b j ' s to be equal to 0.

If you do so, this is what we have done here. So, in this case the first degree was 2 and
the second degree of was 0. So, in this case we matched the degree and substituted all the
coefficients here to be equal to 0. So, you do a similar thing over here in general and then
just add the coefficients (a k + bk ) x k , and this should give you the final answer. So, this
is in fact an algorithm for adding the polynomials, so this is algorithm for adding the
polynomials.

What is the, what are the steps in the algorithm? First identify degrees of both
polynomials, choose the polynomial with highest degree that will be the degree of the
resultant polynomial. Take the polynomial of least degree that is step 2, take the
polynomial of least degree, add all the coefficients which are of the degree higher than
the polynomial and multiply them with coefficients 0.

Once you do that you are ready to do the addition, add the two using this formulation.
So, this is how you can program, you can actually program into a computer for addition
of polynomials. Now, let us try to understand this with subtraction. What is the
difference between subtraction and addition? Both are essentially same, but in
subtraction you are multiplying the second polynomial by -1.
(Refer Slide Time: 19:19)

We have already seen how subtraction will happen. So, here is a quick overview of these
examples. So, this is, these are the polynomials, same polynomials now we are
subtracting. And what we are doing by subtracting? I mean we have multiplied with -1,
just look at here all these terms.

The procedure is exactly the same, it is just that first we have to multiply by -1 and put
the polynomial appropriately. So, let us start with first example, but it will be a quick
run, because p(x) is this. So, there is no change, but I want to subtract p(x) from,
I want to subtract q ( x) from p(x) . So, this polynomial will be multiplied with -1
that is what is done here, so −q ( x) .

So, correspondingly all coefficients are negated, just look at these terms; all coefficients
are negated and therefore. Because there were no cross terms, so you will not find any
difference in the first two terms; but the significant difference is there in the third term
which is actually -6. In a similar manner, take the second question and you are
multiplying q (x) with -1.

So, −0 x 4−x 3−0 x 2−0 x−1 , right. Again because there were no cross terms, there
were no additions. So, this also will have a minimal effect where the second term, these
terms will be with the negative sign, right. So, this -1 was there. So, this will have a
negative sign.
So, there is a minimal impact. The third example that we have taken will face a major
impact; for example, p(x)−q(x ) . Now, you take this −q ( x) here, if you take
3
that −q (x) here, then p( x)−q(x ) ; the first term will be as it is x , because it
is coming from this point here there was a clash of x 2 . So, 2 in when we added, it was
2
2 + 1. So, everything became 3; here it is (2−1) x . In a similar manner (1−2) x
and then the final term was -2.

3 2
So, essentially you got the expression in this form, which is x + x −x −2 ; but the key
principles remain the same, except for multiplying with -1. Multiplying with -1, because
that was a polynomial of degree 0 will not change the degree of the polynomial. So, once
it is not changing the degree of the polynomial, all the rules which were possible for
addition remain intact.

For example, you have to choose the degree which is maximum of the polynomial; there
is no change in the degree except for the multiplication of a minus sign. So, that
multiplication of minus sign is absorbed here. So, now, in the new rule it will be
p(x)−q(x ) will be k is equal to 0 to maximum of m and n there is a
remained intact; and earlier when we were adding it was a( k )+ b(k ), now it is
a(k )−b(k ), ok.

So, I hope you have understood addition and subtraction of the polynomials, both are
essentially same and the resultant what we are getting is again a polynomial. In the next
video, we will take a closer look at multiplication of polynomials.

Thank you.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture - 33
Algebra of polynomials: Multiplication

(Refer Slide Time: 00:14)

In this video, we will learn how to multiply two polynomials. Let us start with basics of
multiplication of polynomials. We already know how to multiply two binomials. For
example, if you have been given two binomials of the form 𝑎𝑥 + 𝑏 + 𝑐𝑥 + 𝑑, then you
know how to multiply these two binomials that is we will use the foil method. However,
in this context, we want to generalize the settings for multiplication of polynomials of
arbitrary degree.

So, let us see, let us start with some simple monomials with through examples. So, here is
a polynomial given to you 𝑝(𝑥) = 𝑥 + 𝑥 + 1 and 𝑞 𝑥 = 2𝑥 . The question is do I know
how to multiply these two polynomials? Remember this one is called monomial, it has
only one term. So, a standard rule of multiplication will mean we have seen this in our
quadratic functions that I will consider the product in this manner.
Once I consider the product in this manner, what we will do is we will try to multiply each
term of this 2𝑥 with each term of this polynomial. So, there are three terms. And for each
term this 2𝑥 will be multiplied. So, if I do that the law of exponents will apply.

For example, 𝑥 × 𝑥 will mean 𝑥 . So, once I apply we apply the law of exponents and
add the exponents, obviously, 2 was a constant coefficient of 𝑥 which will be multiplied
throughout the expression. And therefore, the resultant is this which we can simplify as
2 𝑥 + 2 𝑥 + 2 𝑥 . This is how we will multiply a monomial.

Now, as you can see the this polynomial has three terms 1, 2 and 3. So, it is not a binomial;
it is a trinomial. So, my foil method will not work here. So, foil method will work only for
these kind of expressions which are binomials. So, let us go ahead and try to consider a
similar expression that is a quadratic expression and another binomial, and try to see how
can I extend the basis of foil method right.

So, here is a binomial 2 𝑥 + 1. And here is a general polynomial quadratic polynomial

which is 𝑥 + 𝑥 + 1 same. Now, what will you do? So, naturally you will consider
𝑝 (𝑥) × 𝑞 (𝑥) which will be written in this form. Now, if I want to extend the basis
whatever I did for monomial, that means, I need to convert this into two monomials.

So, what are those two monomials? One monomial is 2𝑥; another monomial is 1. So, if I
treat them separately that is if I write them in this manner, let me erase this, that is I have
written them in this manner.

Then what can I do about it, that means, now this turned out to be a same expression instead
of 𝑥 , here it is 𝑥 that is all is the difference right. So, whatever I did here, I can do it here.
And the last term is actually multiplied with 1 which it suppressed because multiplication
with 1 will not change anything. So, I do not have to worry about the last term.
(Refer Slide Time: 04:05)

Now, I will multiply this 2𝑥 with all the terms in for of 𝑝(𝑥) = 𝑥 + 𝑥 + 1 which is
similar to this particular thing. So, I will get 2 𝑥 + 2𝑥 + 2𝑥 + 𝑥 + 𝑥 + 1. Now,
the job is very simple.

You can treat this as one polynomial, and this one as a second polynomial, and then we
have to add. How we add polynomials? We will add polynomials by matching the
exponents, matching the exponents of 𝑥. So, if I want to add these two polynomials, what
will I do, I will simply match the exponents and I will add them which is given here.

(Refer Slide Time: 05:02)

So, in this case 2 𝑥 , there is no competing term for 𝑥 . So, it remains 2; 𝑥 comes here
and here, therefore, I added the two which gives me 2+1, in a similar manner the terms
containing x are these two. So, I have added these two, so 2+ 1 𝑥 + 1 which is similar to
what we have seen in the last video of addition of polynomials. And therefore, we get the
answer to be equal to 2𝑥 + 3𝑥 + 3𝑥 + 1.

So, effectively what we have done is we know how to multiply the terms term by term.
And finally, if at all I want to seek an extension of a foil method, it will be a term by term
multiplication of polynomials, that means, you take the polynomial of least degree and
multiply it with the polynomial of highest degree term by term, add those term match the
powers and then write your answer. So, this is one prototype that we can follow for finding
multiplication of polynomials or result of the multiplication of polynomials.

Now, the next question is can I generalize this method or can I answer it programmatically,
that means, can I give a simple formula for what the coefficient of one part 𝑥 will be?
For example, in this case can I give a general formula what will be the coefficient of 3𝑥
provided I know polynomials 𝑝(𝑥) and 𝑞(𝑥). So, to answer that, let us go ahead and try to
find a general formulation of this form of this formula.

(Refer Slide Time: 07:03)

Let us go ahead. And if you are asked given one quadratic polynomial and one linear
polynomial, you are asked to compute 𝑝(𝑥) × 𝑞(𝑥), how will you go about this? This is
what our task is now. So naturally I will write 𝑝(𝑥) × 𝑞(𝑥), and then I will convert each
of them into monomials that is one monomial will be 𝑏 𝑥, and second monomial will be
𝑏 .

In this case, what will happen is we will simply multiply them as a separate term by term
multiplication. So, in earlier case our 𝑏 was 1 when we studied one example. But here we
are considering a general expression, and none of the expressions are 0 that is what we are
assuming none of the coefficients at 𝑎 , 𝑎 , 𝑎 , 𝑏 and 𝑏 none of them are 0.

For example, if you consider 𝑏 = 0, then this term itself will vanish the second term itself
will vanish; you will not have the second term. So, we are assuming that all terms remain
in the loop ok. So, now it simple, the job is multiplying these two polynomials, and you
will get some answers that is ok, but now our main worry is to find a pattern in these
answers ok.

So, now, when I multiplied this, if you look at this particular expression that is
(𝑎 𝑏 𝑥 +𝑎 𝑏 𝑥 + 𝑎 𝑏 𝑥 ) + (𝑎 𝑏 𝑥 + 𝑎 𝑏 𝑥 + 𝑎 𝑏 ) . Here you take a pause
and examine the terms. For example, this term contains the coefficient of 𝑥 , this is 2 +
1.

So, 𝑥 . So, in that case, what is happening here is if you look at the suffixes of the
coefficients this is a2, this is 𝑏 , so together they will sum to 3. In a similar manner, you
look at this term which contains 𝑥 . And you look at the suffixes of the coefficients that is
𝑎 𝑏 , together they will sum to the exponent that is a 1+1=2. So, this should be a
coefficient of 𝑥 .

Then if this logic is correct, what should be the coefficient of a constant? The coefficient
of the constant that is 𝑥 . So, the coefficient of the constant must be 𝑎 𝑏 . In a similar
manner you can ask the question what is a coefficient of 𝑥? If you asked that question, you
will naturally get the answer you collect all the in all the coefficients such that their suffixes
will sum to 1 that is 𝑎 𝑏 + 𝑏 𝑎 . So, is there anything called 𝑏 𝑎 ? Yes, it is here.

So, this what we have actually done is we have figured out a pattern; that means, if I want
to find the coefficient of 𝑥 , then better the sum should be some 𝑎 𝑏 , so that they both
will sum, they both will sum to it is not equal to the this is I am saying 𝑥 raise to coefficient
of 𝑥 will be equal to of the will be of the form 𝑎 +𝑏 . So, with this understanding, let
us go further and try to rewrite this sum ok.

So, once I have rewritten this sum, my analogy is further amplified. For example, if you
look at the coefficient of 𝑥 , yes, it was it is 𝑎 𝑏 and 𝑎 𝑏 which is the coefficient of 𝑥 ,
so that also means this means if I can sum over this 𝑗 from 0 to what point to a point where
I want the sum the exponent is raised to k, then I will get all possible combinations where
sum is actually 𝑘.

In a similar manner, you can pause this video and verify whether you are getting the same
expression for 𝑥 and all others right. So, with this understanding, I am ready to generalize
this demonstration or this theory for a polynomial of an arbitrary order.

Let us consider polynomials of degree n and m, and try to find the general answer for them,
and that answer will be in this form. So, if you are given a polynomial of degree 𝑛, 𝑝(𝑥),
and if you are given another polynomial of degree 𝑚, 𝑞(𝑥), let us say 𝑚 ≠ 𝑛.

Even if 𝑚 = 𝑛 it does not matter, but for our purposes let us take 𝑚 ≠ 𝑛, then what will
be the coefficient of each of the 𝑥 ’s? The coefficient is actually given here, ∑ 𝑎𝑏
this is what we have figured out in this expression is the coefficient of 𝑥 .

Then the question is how far the degree will go? The degree will go till 𝑚 + 𝑛 𝑚 ≠ 𝑛; if
𝑚 = 𝑛 then the degree will go to 2 𝑛 that is ok. So, 𝑘 = 0 to 𝑚 + 𝑛, and each of the
coefficient of 𝑥 will be ∑ 𝑎𝑏 . Now, let us demonstrate this idea with one example.
Let us go ahead and see one example of this idea.
(Refer Slide Time: 13:36)

So, now, you have been given two polynomials two quadratic polynomials and you are
asked to compute the multiplication of these two polynomials. One way is very simple you
will go with term by term multiplication, and it simply means you have to multiply the
terms of second polynomial with the first polynomial in a term by term fashion, or you can
actually use the formula that I have given you in the previous slide. So, you can pause this
video, and try to compute by yourself or you can go along with me.

So, let us recall that formula again that is 𝑝(𝑥) is equal to sum a, so my polynomial is a
polynomial of degree 𝑛, and 𝑞 (𝑥) is a polynomial of degree 𝑚. In this case, in this
particular example, the polynomial the first polynomial is of degree 2 as well as the second
polynomial is of degree 2.

So, in order to find the product of these two polynomials, what do we need to find is we
simply need to find the coefficients of 𝑥 . So, let us first identify what are 𝑎 ’s and what
are 𝑏 ’s, 𝑗 is a dummy index. So, it does not matter.

So, let us first identify what are 𝑎 ’s and 𝑏 ’s. So, 𝑎 as you can see is 1, 𝑏 is 1, a 1 is 1
again, 𝑏 is 2, correct, this is correct, and then 𝑎 and 𝑏 both are 1. So, I have enlisted all
the coefficients of this particular expression, 𝑝 (𝑥) and expressions 𝑝 (𝑥) and 𝑞 (𝑥). Now,
we need to use this formula, then this formula which gives me the sum. So, let us use this
formula and figure out.
Remember, all the coefficients that are not listed here. For example, what will be 𝑎 , if at
all, I will write 𝑎 , what will be 𝑎 in this expression? It will be 0. What will be 𝑎 in this
expression? It will be 0. So, all the coefficients that are not listed here are 0s. Keep this in
mind and try to answer the question.

So, now, computation of coefficient; it is very easy. So, let us start with 0th degree term
that is constant term. So, here 𝑘 = 0. So, the summation will actually go from 𝑗 = 0 to 0,
that means, it will have only one term which is 𝑎 𝑏 .

What is 𝑎 𝑏 ? Look here 1 into 1, so it will give you 1 ok. Let us go for a degree 1 term.
So, 𝑗 is equal to 0 to 1, 𝑗 is equal to 0 to 1, so it will have, 𝑎 𝑏 + 𝑎 𝑏 these two terms are
there. So, let us compute them through this table 𝑎 is 1, 𝑏 is 1, so this will retain 1. 𝑎 is
1; 𝑏 is 2, so it will give you 2. So, together it is 1+ 2 = 3.

Let us go for a second order term that is the monomial with degree 2. So, in this case, 𝑗
will run from 0 to 2. So, I will have 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , this is
correct. Just go ahead and compute these terms, 𝑎 is 1, 𝑏 is 1, so you will get 1, 𝑎 is 1,
𝑏 is 2, so you will get 2. And 𝑎 𝑏 that is 𝑎 is 1, 𝑏 is 1, so you will get another 1. So,
you will get the sum to be 4.

Let us go for a third term 𝑥 term, and just simply substitute this. So, we need to find all
possible combinations. So, if it is a degree 3 term and we start with 𝑎 , it will be 𝑎 𝑏 ,
𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , these are the terms. And then you simply compute them.

Remember here now we came up with 𝑏 . What is 𝑏 ? 𝑏 is not listed here, that means, b
𝑏 must be 0. In a similar manner here 𝑎 must be 0 correct. So, these 2 terms are chopped
off right away they are 0. So, let us focus on the other 2 terms the first term you can easily
verify because 𝑏 is 1, and 𝑎 is 1. And 𝑎 𝑏 , 𝑏 is 2, 𝑎 is 1, so it will be 2. So, 1+ 2=3;
this is correct.

Now, the final term is a degree 4 term, correct. If you do a term wise multiplication, what
you will come up with is because the degree 4 will be contributed by the highest order
terms.
So, you will simply multiply 𝑥 × 𝑥 , and you will get only 1 term. But in this formulation
what we are doing here is we are taking all possible terms of degree 4. So, even though
they are 0, we will first list them, and we will put them as 0s.

So, now, when we consider degree 4 term, I will get 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 , 𝑎 𝑏 and

𝑎 𝑏 . So, all these terms are here. And most of the terms will obviously, be 0 only 1 term
is a contributor.

For example, 𝑎 𝑏 is 0, 𝑎 𝑏 will be 0, 𝑎 𝑏 is 0, 𝑎 𝑏 is 0. Why? Because 𝑏 , 𝑏 , 𝑎 , 𝑎

all are 0 only term that will contribute is 𝑎 𝑏 which will be 1×1, so 1. So, this gives us a
clear cut answer, and this is a systematic way to multiply two polynomials.

Therefore, the resultant polynomial 𝑝(𝑥) × 𝑞(𝑥) simply write the terms from this table, so
this is a coefficient of 𝑥 is 1, so the constant term 1 is here coefficient of 𝑥 is 3, so 3𝑥 is
here. So, in a similar manner 𝑥 coefficient of 𝑥 is 4. So, you will get 4𝑥 here ok; so 3𝑥
correct.

So, this is also done. And then 𝑥 has only 1 term as 1, so 𝑥 . Therefore, you got the
resultant polynomial to be equal to this. Now, remember one side note the multiplication
of two polynomials will always fetch you a polynomial again ok. Next operation is division
which we will see in the next video, but the division of two polynomials will not always
lead to a polynomial. We will see that in the next video.

Bye for now.

Thank you.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 34
Algebra of polynomials: Division

(Refer Slide Time: 00:15)

In this video, let us have look at Division of polynomials. What is a division of

polynomial? We have already familiar with division of polynomials, but we have not
done in a rigorous manner and we do not know all possible cases that can occur while
considering division of polynomials, that is why it is important to look at division of
polynomials.

We know some cases like for example, if I have been given a polynomial say
2
a2 x + a1 x +a 0 and if I am told that if this polynomial is divided by a constant say

2
a2 x a 1 x a 0
c . Then, I know what is the resultant polynomial. It will be + + that
c c c
will be the polynomial. So, this case, we are already familiar with.

Now, let us go to one more level of extension. Suppose, this polynomial is divided by a
monomial; that means, we are considering a division of a polynomial by a monomial.
Monomial means, the polynomial that contains only one term, only one variable term.

2
3 x +4 x +3
So, in that case, let us take this example. So, , ok.
x

So, notice few factors here. In this case, when I am considering a division of two
polynomials, the numerator and the denominator; the denominator should always have a
degree smaller than the degree of the numerator. If it is not the case, let us say the
numerator has degree m and the denominator has degree n , then what I am saying
is the degree of the numerator m should always be greater than or equal to n . If it
is not the case, then the division is not possible ok.

For example, let us consider one case, where I am considering a constant polynomial let
us say 4 and I am dividing it by some polynomial which is 2 x +1 . Here, I cannot
divide this; I cannot divide by this polynomial because there is no corresponding x
term. Here it is x0 .

So, I cannot divide this polynomial because the degree of the polynomial plays a crucial
role. So, in this case, the division is not possible, I have to keep this function as it is. Let
us keep this point in our mind and consider division of polynomials. So, now, I am
dividing a polynomial with a monomial, how will you handle this?

(Refer Slide Time: 03:09)

So, a monomial is simply x here. So, what I will do is I will split this with this
addition sign, I will split each of them in separate terms. So, I will consider the term

3 x2
that will give me a term, when I consider this term it will give me a term 3x .
x

4x
When I consider , I will get a term 4.
x

3
Now, as I mentioned earlier when I consider the term , the degree of 3 is a constant
x
because 3 is a constant polynomial the degree is 0. So, I cannot divide this polynomial.

3
So, this will automatically influence this decision that it will remain as it is, that is .
x
So, these are some key things while dividing polynomial by a monomial.

Now, the key idea is I want to divide a polynomial with another polynomial. Let me
erase this first. So, now, I want to divide a polynomial with another polynomial. So, how
will I go about this?

That is I want to find something of this sort. Let us address this question in a video ok.
So, apparently, I do not have any practical way to divide this right now; but from
whatever theory I learnt about quadratic functions, can I derive something? That is what
the question is. So, we will try to figure out some more methods in this video.

(Refer Slide Time: 04:51)

So, let us continue and take a question, this is the numerator that is given to me divided
by another polynomial which is x+ 1 and I want to figure out what this will be equal
to? Let me take this polynomial over here and try to figure out what this polynomial will
be equal to. So, now, I have 3 x2 + 4 x+ 1 and it is divided by x+ 1 . Now, if I want
to divide the numerator by the denominator, what I should see is ok, the denominator has
the highest degree which is x . The numerator has highest degree which is x 2 and
now, how will I be able to get rid of the denominator for some at least for some terms?

So, in that quest, what I will see is I will simply take the first term over here and the first

2
3x
term over here and I will see like monomial, I will see what is . This I can do very
x
easily because both are monomials. So, x vanishes with this square and I will I am
left with 3 x . So, next thing that I will do is I will consider 3 x( x+1) . So, this
actually gives me the answer 3 x2 +3 x . Now, I will try to figure out this term in the
expression that is given in the numerator.

So, if I want to figure out the expression that is given in the numerator, I can easily split
this 4 x as 3 x+ x . If I can do so, that means, I can take this term and based on this

3 x 2 +3 x+ x +1
logic, I can actually write this as .
x +1

Now, I can intelligently split the term over here and I can divide this and I can take this
as a separate term and divide this. So, now, you can readily see the answer will be here,
3 x( x+1) that will get cancel off with x+ 1 and over here, it will be 1. So, the
answer is 3 x+1 . Therefore, such a division is possible, ok.

Let us verify whether the answer is 3 x+1 ; yes. So, I have demonstrated you how to
divide a polynomial using simple method by the method of factorization that we have
already used. Now, this is because x+ 1 was the factor of 3 x2 + 4 x+ 1 .

What if x+ 1 is not a factor of 3 x2 + 4 x+1 , what would have happened? Let us use
this example to understand our findings. So, let me take a eraser and let me write that this
instead of 1, let me put ok, let all other things remain constant, what makes it a factor;
that x+ 1 .
So, I will simply what I will simply do is I will simply change the term to 4 ok. Now,
x+ 1 is no longer a factor, still I will continue with the same method, I will take this

3x 2
3x by x. So, I can consider this x +3 x , only difference is this will be x+ 4 .
x
2
In that case, what happens is 3 x +3 x . So, that will give me 3x into let me

1( x +1+3)
rewrite this as . So, that again gives me an edge that is this is nothing but
x +1

1(x +1)+3
3 x+ getting cancelled.
x+ 1

3
So, this will remain as . So, this is how even if it is not a factor, I can divide the
x +1
polynomial. Now, as we have started by giving some for addition, multiplication,
subtraction, we have given some algorithms. So, now, we need to identify such algorithm
for division of polynomials. To do that, let us first solve this complicated problem in this
simple manner and try to derive an algorithm and try to derive an algorithm for by
solving this problem.

(Refer Slide Time: 10:15)

So, the problem is, I want to divide the terms divide a polynomial
p(x)=x 4 +2 x 2 +3 x+ 2 by q ( x)=x 2 + x+ 1 . So, the p(x) is a polynomial of
degree 4; q (x) is a polynomial of degree 2. So, how will I go about this? So, again, I
 4
will apply a strain strategy that is I will start by writing x plus remember here, it
directly goes to 2 x2 . So, but I want the term containing x3 also to be present
3 2 2
0 x +2 x + 3 x +2 and this term is divided by x + x +1 .

So, remember our first step in the last example was first you take the first term over here
and take the first term over here. Then, take a consider a division of these monomials.
2 2 2
This will give me x . So, what I will do now is, I will consider x ( x + x +1) . This
4 3 2
will give me actually x +x + x .

Now, as per our earlier strategy while solving this problem, we have adopted a strategy
that I will add these terms over here. So, if I add these terms over here, then I will
subtract appropriate terms over here. So, in this case, x 4 is already there, x3 was
3
not there and here, x is there.

So, I need to subtract that x3 from this expression and then, I need to subtract from
2
2x , I will split this into two. So, let us rewrite this expression, that is the numerator
of this expression, x 4 is already there. In order to cancel the denominator, I need x
3

over here.

3 2 3 3
So, I need to add x +x ok; but this x was not present here, it was 0x . So,
naturally the next step will be to eliminate x 3 from here. So, that it will retain a legacy
3 2 2
of this term. So, if I have eliminated x , then it is 2x of which 1x , I have
taken out, so this will be another x 2+3 x +2 as it is.

So, let me write that term as it is 3 x+2 and now, if I divide this term by x 2+ x +1 ,
then what I will get here is take these first three terms and keep the remaining term as it
is ok. So, if I do that, then what will happen is this term x 2 will come out as common
plus now, what happens?

This term vanishes, this term vanishes, this term because x 2 , I can take out common;
2
from these three terms, I can take out x common that is what I have written and it
cancels with the denominator. So, whatever is remaining are the remaining term that is
3 2
−x + x +3 x+ 2 and this thing is divided by x 2+ x +1 .
Now, is our division over? No, because the numerator over here has a higher degree than
the denominator. Therefore, our division is not over. So, again, I will follow a similar
step, I will simply change the color so that I will have a better view ok. So, let us change
the color and have a better view of this.

So, let me write it here from this; from this step, I can go here and say ok. So, this is in
fact, equal to x2 plus now you look at this term −x 3 and x 2 . So, you divide

−x 3
2
which will give you −x . So, in this case, you will multiply
x
−x ( x 2+ x +1) .

2 3 2
So, if you multiply −x ( x + x +1) , what you will get over here is −x −x −x , this
is what you will get. So, you write this term as it is, that is −x 3−x 2−x .

3
Now, from this term, you adjust the terms. So, −x is already there, so I do not have
2 2
to compensate for this term. But there is a plus x , there is a plus x and here
2 2
there is a −x . So, that will give me plus 2x because I am compensating for this
extra −x 2 added in this term, then there is a −x and over here it is plus 3 x .

So, I have to add one x for this −x . So, that will give me plus 4 x plus and there
2
is no competition for a constant term upon x + x +1 . Now, you can take out x
common and this will cancel off, this term will cancel off with this term by taking x
2
common. So, it is x −x is in common plus what you are left with here is
2 2
2 x +4 x+2 upon x + x +1 ok. Again, you will apply a similar procedure that is

2 x2
you will actually divide . So, you will get 2. So, essentially what you; so, when
x2

2
2x
you do that, when you divide 2
, you will get 2.
x

2
So, when you will multiply this number by 2, let me write it here that is 2(x + x+1)
2
ok. So, in this case, what you will get is (2 x +2 x+ 2) of which 2 x 2 is already
2 2
there. So, I will continue over here itself x −x+ 2 x is already there, 2 x 2 . So, let it
be 2 x 2 +2 x , over here there is plus 4 x . So, I can split 2x over here plus 2 x
plus 2. So, that will again come plus 2 as it is here. So, what is remaining now is 2 x
upon x 2+ x +1 .

So, now if you look at this term, what you will get is you can take out 2 common and this
will cancel off with this denominator and therefore, the final expression, I am running
short of space. So, let me erase some terms over here. Let me erase some terms over here
so that I will get some space.

(Refer Slide Time: 19:43)

2x
So, you can rewrite this as to be equal to x 2−x+ 2+ 2
. This will be the final
x + x+1
answer to this division ok. So, this is how we can actually do a division of two
polynomials ok. Let us remove this and see whether to verify whether we have got the
final answer to be correct or not. So, I have removed it, you must have noted the answer.
(Refer Slide Time: 20:21)

2 2x 2
And the final answer that we have got here is x −x+ 2+ ;q( x )=x + x+1 . Yes,
q (x)
so I have got the correct answer. So, here while doing this, we have derived one
algorithm which we will emphasize in the next slide.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 35
Division Algorithm

(Refer Slide Time: 00:14)

So, let us go to the next slide and emphasize the algorithm that we have derived just now.
In order to understand the algorithm, you need some terminology. For example, this
p(x) is called the divident; the q ( x) is called the divisor. The term that you get
over here, here is called the term that you get over here is called the quotient. And the
2x that you have got is called the remainder.

Remember you will declare something as a reminder only when the degree of the
denominator is higher than the degree of the numerator, this is the strategy that we will
follow.

So, now, you are very clear about the terminology, the numerator is the divident, the
denominator is the divisor, the term the polynomial term that you get after dividing is
called the quotient, and the rational and the remainder is something that where the degree
of the numerator is smaller than the degree of the denominator.
This is also called a rational function. If you look at polynomial as a function, then
division of two polynomials is a rational function, only condition that we are enforcing is
q ( x) cannot be equal to 0, this is the condition is which is always in place. Let me
eliminate this and let us go and study the algorithm.

(Refer Slide Time: 01:56)

So, for division of polynomials, we will use the following division algorithm which we
have derived just now, where in the first step what we will do is we will arrange the
terms in the descending order of the degree, and add the missing exponent with 0 as a
coefficient.

Then after adding the missing 0, 0 as a coefficient after adding the missing exponents,
next what we will do is, we will take the first leading terms or the leading monomials,
and we will divide the dividents monomial, the leading monomial of the divident and the
leading monomial of the divisor together. And we will get some number which is which
we will call as quotient, temporary quotient and that quotient we will multiply with our
divident.

Once we multiply with our divident, what we will actually do is we will subtract that
from the original expression for the polynomial that is our numerator. Whatever is
remaining, we will treat that as the next divident. Once we treat that divident, then we
will check if the degree of that new divident is higher than the degree of the denominator
or divisor. If yes, then we will continue with the procedure; if no, we will terminate the
procedure; this is how we will give the division algorithm.

Let us understand this division algorithm by using one example. So, here is an example.
This is the numerator 2 x 3 +3 x 2+1 divided by 2 x +1 , and I want to find the answer
to this question. Let us figure out how to find the answer.

So, in the earlier quest, what I did is I have used the standard numerator denominator.
Now, there is a popular method for division of the polynomial which is called long
division, which works in a similar manner and the same division algorithm works, but
you will have a better handle over the terms.

So, in this long division, what you will do is you will put a parenthesis over here, and
you will put 2 x +1 outside the parenthesis, and you will put this term that is
2 x 3 +3 x 2 . Now, remember the first step plus 0 x+1 ok. So, this is how we will
write. Now, according to our standard terminology, what we will do is we will take the
leading terms 2x and 2 x 3. So, somewhere in the rough you do that. What is
3 2.
2x divided by 2 x ? This will give you x .

2
So, you write x over here, multiply x 2 with 2 x +1 . Once you multiply x
2

with 2 x +1 , write that term over here, 2 x 3 + x 2 . Now, according to our algorithm
divide the first term of the dividend by the first term of the divisor and get the monomial
that monomial is x 2 over here. Next step, multiply the monomial with the divisor and
subtract the result from the divident. So, this is the result from the divident, result from
multiplication, and you are subtracting it from the divident.

So, this will cancel off. So, this will give me 0 and 3 x2 −x2 will give me
2
2 x +0 x +1 . This is the result ok. So, now, this result, I will check whether the degree
of this result this polynomial that I have obtained is greater or smaller than this ok, that is
what we will do. Check if the resultant polynomial has a degree less than the divisor that
is not true.

So, we will go to step 2. What is the step 2? Which is this, divide the first term, first term

2 x2
of this divident with this that is you will divide . So, what you will get here is
2x
 x . So, you will simply add x over here. And then you will multiply that x with
2 x +1 . Once you do that, you will get 2 x 2 + x . So, you write here 2 x 2 + x .

Then what is the next step? You subtract it from the result, so minus, minus 2 x2
vanishes, this gives me −x+ 1 , ok. So, −x+ 1 , again I will go to the same step
because this degree is same, it is not less than the degree of the denominator.

x 1
So, I will again follow the same procedure; which will give me . So,
2x 2

1 1
naturally I will add over here. And once I add over here, when I multiply
2 2

1 1 1
with 2 x +1 , what I will get here is x+ . So, I will write that x+ .
2 2 2

But remember over here the thing was −x . So, I should what I should have done is I

−x −1
should have multiplied -1 to the x that means, . So, the answer is , and
2 2

−1 −1
you will multiply over here, so −x this will not be plus this will be , so
2 2

1 1
−x− which will be given a negative sign. So, this will be x+ . So, I will get
2 2

3 3
the answer to be equal to ok. So, the answer is .
2 2

So, what is what will be the resultant answer? This should not be plus, 1 minute, let me
make it very clear. This cannot be plus; this should be minus, because I have to multiply

−1 3 2 1
with . And here it is the remainder is . So, what I got here is x + x−
2 2 2

3
and the as a quotient, and the remainder is .
2
(Refer Slide Time: 09:23)

3
1 2
So, let me rewrite it again that is I got x 2+ x− + , this is what I got. Let me
2 2 x+ 1

3
1 2
verify this result. And we have demonstrated the algorithm, yes, x 2+ x− + ,
2 2 x+ 1
this is how we will consider division of polynomials in general.
 Mathematics for Data Science 1
Week 06 - Tutorial 01
(Refer Slide Time: 00:16)

Hello, mathematics students. In this week's tutorials, we will look at some questions based on
polynomials and the algebra of polynomials. In this question, we have two quadratic equations,
which are 𝑝(𝑥) and 𝑔(𝑥), presumably equal to 0, and they have the roots, −1 + 1, − 5 + 5
respectively. Then the degree of the polynomial 𝑝(𝑥) × 𝑔(𝑥) is three, it is not because you have
two quadratic equations, and you are multiplying them.

So, the 𝑥 2 terms will have to necessarily multiply, so (𝑎1 𝑥 2 + 𝑏1 𝑥 + 𝑐1 ) × (𝑎2 𝑥 2 + 𝑏2 𝑥 + 𝑐2 ),

when you multiply these, this term, and this term will have to be multiplied and you are going to
get (𝑎1 𝑎2 𝑥 4 ), so the degree has to be 4, which is this. So, B is correct. And then we have the sum,
is equal to, so we need to find the respective quadratic equations now for this, so this would be
(𝑥 + 1) × (𝑥 − 1), the other would be (𝑥 + 5) × (𝑥 − 6).

So, this gives us this is, 𝑥 2 − 1. And this is essentially 𝑥 2 − 𝑥 − 30. So, when we add these two,
we get 𝑝(𝑥) + 𝑔(𝑥) = 2𝑥 2 − 𝑥 − 31. So, C is correct, and that would imply D is wrong. And now
we are looking at the difference 𝑝(𝑥) − 𝑔(𝑥) and that would give us 𝑥 2 − 1 − (𝑥 2 − 𝑥 − 30) =x
square minus 1 minus of x square minus x minus 30, which is 𝑥 2 − 1 − 𝑥 2 + 𝑥 + 30. So, 𝑥 2 − 𝑥 2
cancel off and you have 𝑥 + 29. So, E is wrong and F would be correct.
 Mathematics for Data Science 1
Week 06 – Tutorial 02
(Refer Slide Time: 00:16)

In question number two, there is this polynomial, 3𝑥 4 − 8𝑥 3 + 16𝑥 2 − 10 and is divided by

another polynomial 𝑥 2 − 𝑝, then the remainder comes out to be − 8 𝑥 − 𝑐. They are saying find
the value of 𝑝 and 𝑐. So, let us do the division, then, we have 3𝑥 4 − 8𝑥 3 + 16𝑥 2 − 10 and here
we have 𝑥 2 − 𝑝.

So, this gives us 3𝑥 2 to start with and so this will be 3𝑥 4 − 3𝑝𝑥 2 , so we should write it there,
−3𝑝𝑥 2 . And this goes off, and we get −8𝑥 3 + this becomes +. So, 16 + 3𝑝𝑥 2 − 10. So, now we
have −8𝑥 coming up here, which gives us −8𝑥 3 + 8𝑥, so +8𝑝𝑥 and this goes off again.
So, we have 16+3𝑝𝑥 2 − 8𝑝𝑥 − 10 . So, we again multiply by 16 plus 3p here, and that gives us
16+3𝑝𝑥 2 . And there is no 𝑥 term, we get minus 16𝑝 − 3𝑝2 , then this of course cancelled again.
So, we are left with −8𝑝𝑥 − 10 + 16𝑝 − 3𝑝2 , because this is being subtracted.

So, they are saying this remainder is− 8 𝑥 − 𝑐. And that is equal to −8𝑝𝑥 − 10 + 16𝑝 − 3𝑝2 . So,
the 𝑥 terms have to be the same here, which gives 𝑝 = 1. And then c would be the negative of
−10 + 16𝑝 + 3𝑝2 , which is equal to the negative of −10 + 16 + 3. So, that is the negative of 9,
and so we get − 9.

And that will indicate that none of the options are correct. So, this probably was supposed to be
− 9. We observed that option A and option C are in fact the same thing. So, one of this was
probably supposed to be − 9. Anyway, so our answer is that 𝑝 = 1, and 𝑐 = 9.
 Mathematics for Data Science 1
Week 06 - Tutorial 03
(Refer Slide Time: 00:16)

Now, we have this problem, which of the following polynomials should be added to the polynomial
𝑝 (𝑥) to make it divisible by 𝑥 + 9. So, we need to recognize that it is not necessary that there is
only one polynomial that you add, because since it is only divisibility, we can add a number of
polynomials to 𝑝(𝑥) and make it divisible by 𝑥 + 9. So, we have to check for each of these cases.

So let us see, or what we can additionally do is, we can look at the remainder that we get by
dividing 𝑝(𝑥) with this and then see what to do with that remainder. So, if we did the division,
now, we have 2𝑥 3 + 23𝑥 2 + 40𝑥 and we are dividing it with 𝑥 + 9. So, start with 2𝑥 2 , so we get
2𝑥 3 + 18𝑥 2 . So, this cancels off, this gives us 5𝑥 2 + 40𝑥.

So, we do +5𝑥 additionally, then we get 5𝑥 2 + 45𝑥, so negative and negative so we are left with
−5𝑥 and then that gives us a −5 additionally here, so we have −5𝑥 − 45, therefore these two go
off and we are left with 45 as our remainder. So, 𝑝(𝑥) is essentially (𝑥 + 9) into the quotient +45.
So, if we subtracted 45 from 𝑝( 𝑥), we will get divisibility by (𝑥 + 9).

So, B is necessarily correct. Let us look at what happens if we added A, if we added A, 𝑝(𝑥) +
2𝑥 2 + 9𝑥 is some multiple of some product of (𝑥 + 9), and some quadratic plus 2𝑥 2 + 9𝑥 + 45.
So, unless 2𝑥 2 + 9𝑥 + 45 is divisible by (𝑥 + 9), 𝑝(𝑥) would not be divisible by (𝑥 + 9).

So, what we should really be checking is 2𝑥 2 + 9𝑥 + 45. Is it divisible by (𝑥 + 9)? And the direct
way to check it is to substitute 𝑥 = − 9, so you will get 2 × 81 + 9 × − 9 + 45 = 162 − 81 +
45, which is greater than 0, it is not equal to 0. So, no, A does not give us divisibility by (𝑥 + 9).

What happens if we added 5𝑥, we get 5𝑥 + 45. So, we have this 45 remainder, so we are getting
5𝑥 + 45, which is equal to 5(𝑥 + 9), which is directly divisible by (𝑥 + 9). So, this is correct too,
c is also correct. And what happens if we added 𝑥 2 − 126 , then we would get 𝑥 2 − 126+45 as
the additional part upside from (𝑥 + 9) into that quadratic, so this is equal to 𝑥 2 − 81, which is
equal to (𝑥 + 9 )(𝑥 − 9). So, (𝑥 + 9 ) is dividing this particular polynomial. So, we can add 𝑥 2 −
26𝑥 − 126 also, and get divisibility by (𝑥 + 9 ).
 Mathematics for Data Science 1
Week 06 - Tutorial 04

(Refer Slide Time: 0:15)

In this question we have 3 polynomials, 𝑃(𝑥), 𝑄(𝑥) and 𝑅(𝑥) and their degrees are given to be
2, 3 and 4 respectively. Which are the most suitable, although not necessarily exact
representation of ℎ(𝑥) where ℎ(𝑥) is a polynomial in 𝑥 and it is given
𝑃(𝑥)×𝑄(𝑥)−𝑄(𝑥)×𝑅(𝑥)+𝑅(𝑥)×𝑃(𝑥)
as . So, what we need to do here is to identify the degree of the
𝑃(𝑥)+𝑃(𝑥)𝑄(𝑥)

numerator and the denominator.

Numerator degree 𝑃(𝑥) × 𝑄(𝑥) will give 2 + 3 = 5 that would be the degree of 𝑃(𝑥) × 𝑄(𝑥),
the degrees will add up and when we look at −𝑄(𝑥) × 𝑅(𝑥), then again the degrees will add
up which will give us 3 + 4 = 7, so this is from −𝑄(𝑥) × 𝑅(𝑥) and then 𝑅(𝑥) × 𝑃(𝑥) gives
2 + 4 = 6. This is 𝑅(𝑥) × 𝑃(𝑥) degree.

And in the denominator 𝑃(𝑥) anyway has degree of 2 and 𝑃(𝑥) × 𝑄(𝑥) we have seen has
degree of 5. So, since we are adding all these polynomials together, the degree of the entire
numerator is the maximum which is 7. So, we have 7 as a degree of the numerator and 5 as the
degree of the denominator. Since it is a division, the powers will have to subtract, so degree of
ℎ(𝑥) = 7 − 5 = 2. So, ℎ(𝑥) is a quadratic and that would indicate B and C are possibly the
curves because these look like quadratic curves. A and D are definitely straight lines.
 Mathematics for Data Science 1
Week 06 - Tutorial 05
(Refer Slide Time: 0:14)

There are 6 flat, 6 of them, thick iron sheets each of length, breath and thickness 𝑥 + 4, 𝑥 +
𝑥 2𝑥+6 𝑥+4
3 and 𝑥 respectively and they are melted to make solid boxes of dimensions 2, , . How
3 5

many solid boxes can be made this way? So, basically the volume will have to be equal. So,
first we find the volume of our 6 sheets put together that would be 6 [(𝑥 + 4) × (𝑥 +
3 ) × 𝑥] and this would be equal to the volume of the solid boxes.

𝑥 2𝑥+6 𝑥+4
So, let us say there are 𝑛 solid boxes and then the volume of each is 2, and . So, now
3 5

this 𝑥 and this 𝑥 cancels and this 𝑥 + 4 and this numerator here cancels and 2𝑥 + 6 is (𝑥 +
𝑛×2
3) × 2 so, this is one time and this is 2 times. So, what we get is 6 = 2×3×5. So, 2 and 2 also

cancels. This implies 𝑛 = 6 × 3 × 5 and that is 90. So, you get 90 boxes overall.
 Mathematics for Data Science 1
Week 06 - Tutorial 06
(Refer Slide Time: 0:14)

In this question, let 𝑥 be the number of years since the year 2000, so 𝑥 = 0 denotes the year
2000. And the total amount generated in lakhs by selling a product is given by 𝑇(𝑥). So, this
is a polynomial which has the variable as a number of years since 2000, and the different cost
of the particular years are given here. So, purchase cost is this polynomial, transportation cost
is this polynomial, miscellaneous cost is this polynomial.

So, we now have to find out the profit for that year. So, that would just be 𝑇(𝑥) minus all these
cost. So, it is 5𝑥 4 + 3𝑥 3 + 𝑥 2 + 𝑥 − (𝑥 4 + 𝑥 3 + 𝑥 2 ) − (𝑥 3 + 𝑥 2 + 𝑥) − (0.5𝑥 2 + 0.5𝑥). So,
this would be the total profit and for that we now have to look at the each x power term.

So, 𝑥 4 , there are 2 terms, 5𝑥 4 and −𝑥 4 . So, we get 4𝑥 4 and 𝑥 3 there are 3 terms, 3𝑥 3 , and −𝑥 3
and −𝑥 3 here. So, we get 𝑥 3 and 𝑥 2 x square terms there are 4, there is this 𝑥 2 and then there
is this −𝑥 2 and another −𝑥 2 and minus −0.5𝑥 2 .

So, that will give us minus 1.5 x square because this and this cancels off and then we get
−1.5𝑥 2 . And lastly the 𝑥 term there is 𝑥 and − 𝑥 which cancels off and −0.5 𝑥. So, −0.5 𝑥.
So, this would be the total profit for that year.
 Mathematics for Data Science 1
Week 06 - Tutorial 07
(Refer Slide Time: 0:14)

In this question, we have a company which is producing a product A through 3 processes and
the cost of production are given as 𝑀1 (𝑥), 𝑀2 (𝑥) and 𝑀3 (𝑥). These are the 3 cost of
production. They have given us polynomials of 𝑥. So, what is 𝑥? x is the cost of raw material
per kilo. And now they are also giving us the waste management cost as 𝑊1 (𝑥), 𝑊2 (𝑥) and
𝑊3 (𝑥). What will be the effective manufacturing cost?

So, effective manufacturing cost simply has to be the sum of these. So, 𝐸1 (𝑥) = 𝑀1 (𝑥) +
𝑊1 (𝑥), so that is going to be so, 𝑀1 is here, 𝑊1 is here. So, 𝑀1 would be 100𝑥 3 and there is
no 𝑥 3 term in 𝑊1 , so we first write down 100𝑥 3 , then there is an 𝑥 2 term here, 20𝑥 2 , there is
also an 𝑥 2 term here, 0.01. So, their sum will give us 20.01𝑥 2 and then there is no 𝑥 term in
𝑀1 . There is an 𝑥 term here, so you get − 0.008 𝑥. So, this is also done. And then lastly we
have the constant term which is+ 10. So, this is the effective manufacturing cost for process 1.

Likewise, process 2 would be 𝑀2 (𝑥) + 𝑊2 (𝑥). So, here this is 𝑀2 (𝑥), this starts with an 𝑥 4
and 𝑊2 also has an 𝑥 4 term. So, we have 20.01𝑥 4 plus there is no 𝑥 3 term in 𝑀2 , there is an
𝑥 3 term here, so there is no, this is not plus, we write 0.001𝑥 3 , then the 𝑥 2 term there is a 10𝑥 2
here and a 0.01𝑥 2 here. So, + 10.001𝑥 2 and lastly there is a constant term and no constant term
over there. So, this is −20. So, this is the 𝐸2 .

And then 𝐸3 is going to be 𝑀3 (𝑥) + 𝑊3 (𝑥) and in 𝑀3 there is just 2 terms, the 𝑥 3 terms and a
constant. 𝑊3 , there is only one term which is the 𝑥 2 term. So, we just write all of them together,
𝑥 3 + 0.01𝑥 2 + 20 . So, this is the effective manufacturing cost for the third process and the
three of them are given here.

(Refer Slide Time: 3:35)

Now, what is the ratio of effective manufacturing cost of first and third processes when the cost
of raw material per kg is rupees 1? So, basically we are looking for the ratio 𝐸1 (1): 𝐸3 (1) which
is then we just substitute 1 in the E1 term, so we get [100 + 20.01 − 0.008 + 10]: [1 + 0.01 +
20] is to, so we get 130.002 is to 21.01. So, we have to put this down as a number 130.002
divided by 21.01 is roughly 6.18762.

Then we have the third question, third part of this question which says, which asks which of
the processes M1, M2 and M3 should the company chose when the cost of raw material per kg
is 10? So, the company should chose the cheapest process. So, we have to find out 𝐸1 (10),
𝐸2 (10) and 𝐸3 (10). And then if we looked at that 𝐸1 (10) =100𝑥 3 is 1 lakh plus 20.01𝑥 2 is
2001 − 0.08 + 10. This is then 102010.92.

Moving on then E2(10) is 200100. So, 𝑥 4 = 04 so, this is what we get and this is 200100, so
𝑥 4 = 104 , so this is what we get and this is 200100. We see that the remaining smaller terms,
𝑥 3 , 𝑥 2 and constant term have small coefficients as well, 0.001 and 10.001. So, they will not
really impact the value very much. So, we know that this is already larger than 𝐸1 (10), so we
do not consider it.

Let us look at 𝐸3 (10) which is then 1000 + 1 + 20, so this is just 1021 rupees. So, 𝐸3 (10) is
the least which is why the company should chose the third process as their process when x is
equal to 10 rupees per kilo.
 Mathematics for Data Science 1
Week 06 - Tutorial 08
(Refer Slide Time: 0:14)

Our last question we are looking at the best fit for some data. So, this is the fit we have obtained a
fifth degree polynomial for this data, these 4 points and they are asking what is the value of c, c is
the constant term here. What is the value of c if this curve has to be the best fit using sum squared
error? So, let us assume this curve is 𝑓(𝑥) = 2𝑥 5 − 4𝑥 4 − 3𝑥 + 𝑐. So, we are going to have to
also put up the 𝑓(𝑥) value, so 𝑓(0) is then c because everything else is power of 𝑥, so 𝑓(0)=c, and
then we have to look at 𝑓(1) which is 2 − 4 − 3 + 𝑐 that is equal to 𝑐 − 5.

So, here this is 𝑐 − 5 and then 𝑓(2) is 2 × 32 − 4 × 16 − 6 + 𝑐, now 2 × 32 is 64, 4 × 16 is

64, so these two cancel off, so you get 6 𝑐 − 6. And lastly, we have 𝑓(3) which is 2 × 243 −
4 × 81 − 3 × 3 + 𝑐 so that gives us 𝑐 + 153, so here it will be 𝑐 + 153. So, for finding SSE we
are going to have to do 𝑓 (𝑥 − 𝑦) or (𝑦 − 𝑓(𝑥))2.

So, (𝑦𝑖 − 𝑓(𝑥𝑖 ))2 and we are going to sum it from 𝑖 = 1 to 4 and that gives us (𝑐 − 0)2 +
(𝑐 − 1)2 + (𝑐 − 1)2 + (𝑐 + 1)2 + (𝑐 + 2)2 So, this is the sum squared error.

(Refer Slide Time: 3:08)

And we get 𝑐 2 + 𝑐 2 + 1 − 2𝑐 + 𝑐 2 + 1 + 2𝑐 + 𝑐 2 + 4 + 4𝑐, so this − 2 𝑐 and this + 2 𝑐 cancels
off and we arrive at 4𝑐 2 + 4𝑐 + 5, this is our sum squared error it is a quadratic in c and for
minimum and this is also an upward facing quadratic because the coefficient of 𝑐 2 > 0, so it will
be a parabola like this and the minimum occurs at this point which is the vertex of the parabola
−𝑏 −1 −1
and that we know is 2𝑎 , here − 𝑏 = −4 and 𝑎 = 4 so 2𝑎 = 8 so you get , so for 𝑐 = , we get
2 2

the minimum sum squared error. Thank you.

 Mathematics for Data Science 1

Prof. Neelesh S Upadhye

Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 36
Graphs of Polynomials: Identification and Characterization

Hello friends, in this video, we will take up our next mission of about understanding the
polynomials. This mission is given a graph of a function, whether we can identify the
given function is a polynomial or not. If you have been given a polynomial equation,
how will you put it on a graph paper?

So, the mission is twofold. First, If you have been given a graph of a function, you will
identify whether this function is a polynomial or not. If yes, the, we will answer the
second question that is can I derive the algebraic equation of this polynomial? The
second part of the mission is we want to identify how the graph looks like if I have the
equation of the polynomial. So, let us begin our journey about understanding the Graphs
of the Polynomials.

(Refer Slide Time: 01:17)

So, first of all let us recollect from our earlier experience that is linear functions and
quadratic functions. If I am as linear functions and quadratic functions themselves are
graphs of the functions. So, when I am plotting these two functions or when I am putting
them on the graph paper, what happens? There you will never feel any abrupt jerk while
drawing these functions. If you are trying to draw, you, for example, if you are trying to
draw a line, then what you will do is you will simply draw a line, and then on graph
paper. And there would not be any jerk for drawing the line.

In a similar manner if you are asked to plot a quadratic curve, you will find a axis of
symmetry, and around the axis of symmetry you will do something like this, this is let us
say this is the graph right, that means, the curve that you are trying to draw is has always
been smooth. So, that one feature we can record in our mind. And say that the, if I have
been given a polynomial function, the polynomial function must be smooth that means, I
should be able to join the points effortlessly without having any jerk.

If there is any corner or edge in the graph then it better not be a polynomial function.
Another thing is you can draw these graphs without lifting your pen; you can draw these
graphs without lifting your pen that means, these graphs always are continuous.

(Refer Slide Time: 03:04)

So, let us try to list these properties; first if you have polynomial of second degree or
higher even a linear degree, the graphs do not have sharp corners that is the graphs are
always smooth curves, this is a first feature that we will notice if I have been presented
with graph of a function.

Second thing is polynomial functions always display graphs that have no breaks that is
what I meant; the graph that I am drawing is always going to be continuous curve. Or
you can say in better words, it is curves with no breaks are called continuous, and
therefore the function itself will be continuous.

So, let us identify through two graphs. Let us have this graph. Now, is this a polynomial
function? Does it satisfy the first criteria? That is it should be a smooth curve, is it a
smooth curve? Yes, it is a smooth curve. But if you look at this point, then it had some
sharp corner over here, this corner is very sharp. And therefore, I cannot qualify this as
polynomial function; this is not a polynomial function.

Let us have a look at the next graph which is this. Now, here I can use my free hand skill
to draw a curve and I can actually find out how I can draw better curve. For example, if I
start drawing this curve, then I can easily pass through this. So, you will all the
transitions are very smooth, because the transition is very smooth I can easily identify
this to be a polynomial function.

(Refer Slide Time: 05:06)

Therefore, this qualifies to be a polynomial function, whereas this does not qualified to
be a polynomial function. Let us take a quick look at some other graphs and see whether
they will qualify as polynomial function or not.

(Refer Slide Time: 05:22)

So, let us go ahead. And pose a question, which of the graphs given below represent
polynomial functions? So, one by one I will unfold the graph, and we will argue for
whether they are polynomial functions or not. This is the first graph. As I mentioned
earlier, this also qualifies to be a polynomial function.

For example, if I want to draw a curve across this, I can easily draw a curve without
lifting my pen and therefore, it qualifies to be continuous, and it has no breaks in
between. So, therefore, it is continuous and it does not have any sharp edges, and the
graph seems to be free hand. Therefore, it qualifies to be a polynomial function. So, my
answer to this question is yes, this is a polynomial function ok.

Let us go ahead with the next graph ok. Now, what about this graph? Of course, we have
argued for say similar graph in the earlier page that this graph does not seem to be a very
neat graph, and it has a corner over here. This is the corner point of the graph. And
therefore, this disqualifies to be a polynomial function; this is not a polynomial function.
Again let me reiterate this was a valid polynomial function.
Next graph, let us look, let us try to see the next graph. This graph more or less seems to
be a graph of a line, because it is a graph of a line. You can see this is also smooth. The
transition is very smooth. So, again this will qualify as a polynomial function at least as
far as the graph is visible. This is a graph of a line and it is a linear polynomial. So, it
qualifies to be a polynomial function.

Let us go to the next graph ok. So, this graph is actually smooth. I can draw a curve over
here, but at this point let me erase this graph; actually, you do it here. At this point, at
this juncture, there is some problem. What is the problem? Over here, if I am drawing a
curve over here, then I have to lift my pen come to a point 0 and then start drawing it.

So, this defeats the criteria that the graph should be continuous. Though the curves are
very smooth, but at this point, this in this juncture, the problem is you cannot have a
drawing without lifting your pen. Therefore, this will disqualify to be a polynomial
function. This is not a polynomial function.

So, we have identified what is a polynomial function and what is not a polynomial
function. Generally, whenever you have several ups and downs in the functions, we will
estimate them or we will guess them to be a polynomial function if there is no corner as
given in this graph, second graph to be precise.

(Refer Slide Time: 09:15)

And if there is no there should not be any corner of this kind and there should not be any
discontinuity of this kind, ok. Then we can easily safely say that the given function is a
polynomial function ok.

And if you are looking at the ups and downs, these up and down of a function, those are
the typical features of polynomial functions. With this knowledge, we are ready to
handle polynomial functions because now, if you have been given a function on a graph
paper you can identify whether the given function is a polynomial function or not. The
next important criteria for about polynomial functions like quadratic equations is our
identity our ability to identify zeros; zeros of the function.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 37
Zeroes of Polynomial Functions

(Refer Slide Time: 00:15)

So, let us focus on Zeros of Polynomial Functions. So, for clarity, let us recall what is zero
of a polynomial function. If 𝑓 is a polynomial function, then the values of 𝑥 for which
𝑓(𝑥) = 0 is called zero of f. A value 𝑥 of for which 𝑓(𝑥) = 0 is called zero of f.

Now, when we studied quadratic functions, we had several methods of identifying the
zeros of the quadratic functions. For example, we actually tried to graph the quadratic
function because we knew some techniques, we actually plotted set of ordered pairs on a
graph paper and join the curve smoothly, then we identified it is crucial to identify axis of
symmetry and around axis of symmetry you can plot and wherever it intersects x axis, we
will call that as a zero of a function. This is how we identified quadratic zeros of quadratic
functions.

Another way that we used which will be helpful here is; factoring the quadratic function
into factors given a quadratic function identify the factors and write the polynomial into
intercept form. If you are able to do that, then you have again identified zeros of the
polynomial because when you said that quadratic function to be equal to zero and if it is
in a factored form, all the coefficients corresponding to that factor will be all the numbers
corresponding to that factor will be zeros of the polynomial function.

So, now, we will focus on the factoring component of polynomial functions. So, if the
equation of the polynomial function can be factored, then we can set that each factor to be
equal to 0 and solve for zeros; this is an important step. But it as we have seen in quadratic
functions, this is not always possible.

In such case, if you put some random values, if you throw in some random values in the
function and you get something like 𝑥 = 𝑎, you will get the value to be 0 that is also
helpful.

Then, you can guess that is a factor and you can use the previous video to divide the
polynomial by (𝑥 − 𝑎) which will give you the remainder term and that remainder, you
can actually figure out whether you can; you can consider factoring for that remainder or
not all these things are possible or the other factor it is not remainder sorry it is the other
factor. So, these are some possible ways.

Up to quadratic equations, we had some easy ways out easy way out like given the equation
of a quadratic function, we can use this method to find x intercept because x for x
intercepts, we get zeros that is what I explained earlier also. So, you can find x intercepts
and you will easily get this. You can use the similar technique of finding x intercepts for a
general polynomial function also, but it is very difficult to plot a polynomial function ok.

Given a graph of a polynomial function and you have identified based on our previous
criteria, you have identified that this is a polynomial function, you can guess what are the
zeros of the polynomial function that way this statement helps. But, if you go for higher
order polynomials that is general polynomials, this can become messy, it can be really
challenging.

Quadratic equations can be easily solved using quadratic formula we have a solution for
quadratic equations. But, the cubic and four-degree polynomials have some formulae
which you may study in your tutorials, but they are not easy enough to remember. And,
for higher degree polynomials, you do not have any idea of how to approach finding zeros
of the polynomial functions, you have to go by trial and error method and whatever
knowledge you have about square, quadratic, linear and cubic polynomials.
(Refer Slide Time: 05:00)

So, let us summarize what we have; what we have discussed just now. If I want to identify
zeros of polynomial functions, the factoring technique is a crucial technique. So, what you
can do is you can look at the polynomial and if you look at the polynomial, there is one
easy way out that if you can identify the greatest common factor that is the greatest
monomial that can be taken out common you can use that technique.

Once, if there is no such technique, if once that is available, the polynomial is more or less
manageable, then you can use the technique of factor by grouping. So, you can create
groups in that and see whether a anything is coming out common that is another technique.
Another thing is you can instead of handling groups, you can decide to handle three terms
at a time so, that is a trinomial factoring. This will be helpful when you have very high
degree polynomial. So, these are the common methods for factoring the polynomials.

Once you can factor the polynomials each of them can be equated to 0 by writing a
polynomial in a factored form. And then finally, if you are not very sure, then you can use
some graphical tools which are available these days on computer or on the net one such
tool is; Desmos which we are using in our presentations.

So, you can use those tools to determine the intercepts. In these tools basically, you will
give of equation of a function and it will be graphed they will give the they will project
the graph of a function right. So, this is our zeros of the polynomials and factoring play a
crucial role.
(Refer Slide Time: 06:53)

To understand this, let us see how to find x-intercept of a polynomial function by factoring.
So, what we have discussed just now is we have set the equation that is 𝑓(𝑥) = 0 in order
to facilitate factoring 𝑓(𝑥) = 0, then if the polynomial is given in factor form; factored
form then equate each of them to be equal to 0 which we have seen for quadratic case also.

If it is not given in factor factored form, first in that you will look for is you take out some
common monomial that is available in all the terms if that is that is there and you have
taken out or if that is not there still you can go to the second step that is whatever at the
rest of the terms you can factor them into factorable binomials or trinomials, you look for
try to look for combinations which we have done successfully for quadratic equations well
doing the factoring. So, you can do a similar thing over here.

And then finally, set each factor equal to zero that will give you the x-intercept. This is
the; this is the strategy that we will follow for finding x-intercept of polynomial function
by the method of factoring.
(Refer Slide Time: 08:21)

So, let us look at this example where we will follow the steps of the algorithm. So, the a
question says; find x-intercepts of a function 𝑥 6 − 8𝑥 4 + 16𝑥 2 . So, as per our algorithm
or as per the steps given in the previous slide, I will set 𝑓(𝑥) = 0 that is; 𝑥 6 − 8𝑥 4 +
16𝑥 2 = 0.

Now, you look at greatest common factor, a monomial that is common in all these terms
that is 𝑥 2 . So, what I will do is I will separate out this 𝑥 2 , I have taken out this 𝑥 2 and now,
you look at the other factor that is 𝑥 4 − 8𝑥 2 + 16.

Now, this factor can be related to our quadratic equation of the form 𝑡 2 − 8𝑡 + 16. Can I
factor this quadratic equation because there is no term corresponding to 𝑥1 and there is no
there are no odd terms essentially. So, I can use this and I can leverage the skill of quadratic
equations to solve this equation and from quadratic equation point of view, I know this is
(𝑡 − 4)2 = 0. So, instead of t here, it is 𝑥 2 . So, that will give me 𝑥 2 × (𝑥 2 − 4)2 = 0 .

Now everything is looks in the form of 𝑥 2 . So, what are the values of x? What are the
feasible values of x? Those will be the x-intercepts. So, you can put 𝑥 2 is so, this will give
me 𝑥 2 = 0 or 𝑥 2 − 4 = 0. So, 𝑥 2 − 4 can further be factored into (𝑥 − 2)(𝑥 + 2) = 0.
And with this understanding, I can write 𝑥 = 0, 2, −2 are the intercepts of f x-intercepts
of 𝑓ok.
(Refer Slide Time: 11:09)

Now, as per the last step in the algorithm, you want to verify this result. How will you
verify this result? Using the technology; so using Desmos, I have drawn this graph and
you can verify that 𝑥 = −2 which is here, 𝑥 = 0 which is here and 𝑥 = 2 which is here
are all x-intercepts of a polynomial function given by these 𝑓(𝑥) ok. So, this is how we
will identify x-intercepts.

Let us understand this strategy by looking at one more example.

(Refer Slide Time: 11:49)

So, now here, we have been asked to find x-intercept of a polynomial function which is a
cubic polynomial function 𝑥 3 − 4𝑥 2 − 3𝑥 + 12 fine. So, as per our set up, this first step
is set 𝑓(𝑥) = 0. So, you have set 𝑓(𝑥) = 0 that essentially gives me 𝑥 3 − 4𝑥 2 − 3𝑥 +
12 = 0.

Then, the second a step if you have any common monomial, there is no common monomial
because the last term is a constant term so, you cannot figure out a common monomial.
Then is there any pattern? Can you look at two-two terms each binomials or trinomials
because there are four terms, it is better to look at binomial terms.

So, if you look at the first two terms, you can see that you can throw out 𝑥 2 as a common
thing, if you throw out 𝑥 2 as a common thing, then you will be stayed with (𝑥 − 4) as a
term as a one factor. And if you look at these two terms, then again if you take out 3
common - 3 common, then you will get (𝑥 − 4). So, using the technique of binomial,
binomials in this case, I am able to see this kind of factoring possible.

Good, that essentially means I can rewrite this expression as (𝑥 2 − 3)(𝑥 − 4) = 0. Then,
I want to solve this (𝑥 2 − 3) that is all is remaining which is a quadratic equation. So,
you can easily solve using quadratic formula or a factoring, but here in this case, I know
the factors so, that will be (𝑥 − √3) and (𝑥 + √3) = 0.

(Refer Slide Time: 14:14)

And therefore; therefore, the solution of this quadratic equation is well known that is 𝑥 =
4, +√3, −√3 are the x-intercepts of the function. The final step is I want to verify using
some technology or a graphing tool. This is the graph of a function.

So, in this case, you can easily verify there are three roots: first root this one which is a
occurs, it occurs at , −√3, this one this is , √3, this one is 4. So, these are the four these are
the three roots of a cubic polynomial. Roots or x-intercepts or a zeros of a cubic polynomial
ok.

(Refer Slide Time: 15:05)

So, let us go ahead and see; let me remove this blocks. Another example where I want, I
am interested in finding x-intercepts as well as y-intercepts of a polynomial function which
is given in a factored form.

So, the polynomial is given in factored form. So, visually you will be able to guess the
roots. So, as a standard set up, we will set 𝑔(𝑥) = 0. Once you said 𝑔(𝑥) = 0, it is very
clear that 𝑥 = 1 and 𝑥 = −3 are the x-intercepts of 𝑓.

What about y-intercept? What is the y-intercept at all? So, y-intercept is where x is given
to be 0. So, simply substitute 𝑥 = 0 in the expression of 𝑔(𝑥), you will get 𝑔(0), which
is 0 × (−1); the whole square that is 1+0+3 that is 3 so, 1 × 3 is 3 so, your 𝑔(0) = 3.
So, this is how you will figure out x-intercepts and y-intercepts of the function and this is
the graph of that function. Using technology, I have verified.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 38
Graphs of Polynomials: Multiplicities

(Refer Slide Time: 00:14)

So, now, so far, we have mastered two skills; given a graph of a polynomial function, I
know whether the given function is I can identify whether the given function is a
polynomial function or not, ok. The second thing that we have seen is from algebraic
expression of polynomial function whether it is in factored form or non-factored form, I
have some set of rules or algorithm which will help me to identify, the roots of the
polynomial or the zeros of the polynomial.

So, with this knowledge, can I explicitly write a polynomial function, or do I need to know
something more about it? That is what the question that is troubling us. For example, the
knowledge about x-intercepts in this case, and the knowledge of y-intercept, is this helping
us to understand how the polynomial will look like?

For example, how will I decide the polynomial is going down from here, polynomial is
going up from here, and it will stay going up forever, or when will this kind of shape come,
the curve when will the rise and fall will happen, I do not know anything about this right
now. What I know is simply the function should be smooth, yes, this is a smooth function.
But, from this graph can I write this equation? Seems to be difficult right now, but we will
get the handle over it in due course of time.

(Refer Slide Time: 02:11)

So, let us now look at the x-intercepts or identifying the x-intercepts using graph. So, you
have been given a polynomial function. You have used a technology to identify the graph,
but still you are not convinced.

And, you want to try it by your hand. So, how will you do it? That is the question. So, if I
want to find the x-intercept, the given polynomial is a cubic polynomial, fine. So, this
polynomial is not given in a factored form; I cannot find greatest common factor that is
not possible.

Then, if I want to find something which is like binomial thing that is (𝑥 + 4), but the rest
the other term is (𝑥 − 6). So, I am not actually getting these two tricks done. So, there is
no way in which, I can factor this polynomial. So, one crude way, if you do not know how
to go about, is to plot the pair of values as we have done in quadratic case.

So, simply find out what are the function values at some points. So, these are some standard
points, I have drawn them symmetrically 0, 1, 2, -1, and - 2. When I considered these two
points, because the function is very nice, I accidentally came across two zeroes that are -
2 and 1, good.
So, -2 and 1 are the x-intercepts of f which is clear from the table. Now, can I use this
knowledge to find the third zero? The answer is yes. And, we know the long division. So,
what you do is you consider (𝑥 + 2) as one factor and (𝑥 − 1) as another factor. You
multiply (𝑥 + 2)(𝑥 − 1) and treat that as a divisor, and take 𝑓(𝑥) as a dividend, and do
the long division.

If you do the long division; you will get, you may pause the video and try for yourself;
otherwise you will get the third 0 to be equal to 𝑥 = − 3; that is 𝑥 + 3 is another factor.
And, this is a cubic polynomial, so it cannot have any other factor, can I have at most 3
roots. So, you got this (𝑥 + 3), (𝑥 + 2), and (𝑥 − 1) as the factored form of this equation.
Then, it is easy to plot the equation along with this table of values.

So, what you will do is, you will simply put up the points, you will simply put up the point.
So, over here I know something and I know I have figured out the third root to be equal to
𝑥 = − 3; therefore, I can put that point as well.

So, this is 𝑥 = − 3 and, how to draw a line passing through; so, the next step is joining the
line. So, you can draw join a smooth line passing through these points, then at this point it
will turn up; at this point it will turn up, but to connect to this point, it has to go down, and
then I do not have any idea. So, right now I can draw only up to this, right.

We will analyze further and see the cubic polynomial cannot turn more than two times, so
that we will that we will come later. But, right now I can draw it this way. So, let us see
what our graphical tool gives us. Yes, where the previous image and this image are slightly
perturb but they are exactly matching. So, this is how the behaviour of the function will
be.

So, now we have slightly better edge over drawing the graph of a function, when we have
been given a equation of this form, but still I do not know why this is not turning up or
why this is not coming down, I need to understand these things in a better way by using
some analytical tools. For a moment, we got the correct graph.
(Refer Slide Time: 06:52)

So, let us move ahead and try to see the behaviour of the graphs around the intercepts. For
that, it is important to know the multiplicities of factor. So, in particular, graphs behave
differently for at various x-intercepts.

We can go back to the previous case, where everything is of linear order that is 𝑥 + 3,
𝑥 + 2 and 𝑥 + 1, the graph was behaving like this. If you go further back; why would this
graph behave like this? Over here, when 𝑥 = −3 is a factor, the graph the graph was
actually like a straight line and over here it turned around.

(Refer Slide Time: 07:33)

So, why should it turn around the factor? So, for example, 𝑥 + 3 is one factor that is -3,
and here 𝑥 − 1 is one factor. But, for this factor it turned around; and for this factor it
crossed, it cross the x-axis. So, why is this happening? So, I need to have a deep
understanding of this. For that, we will discuss the next that is what we will discuss in the
next slide. So, is it related to the function being appearing the factor appearing multiple
times? That is what we will try to see.

So, as mentioned as shown in the earlier slide, that the graph can cross over the horizontal
axis or it may bounce off; that means, it will touch and go up that is tangential to that axis.
So, why is this happening at x-intercept? That, so in that case let for that making the
understanding clear we will write a polynomial in a factored form which is
(𝑥 − 1)2 (𝑥 + 2)3 (𝑥 + 4) , right. And, let us draw that polynomial using technology or
graphing tool, ok.

So, now some crucial things, let us identify the factors first; 𝑥 − 1 that is 𝑥 = 1, this is
the factor that we are talking about. Then, 𝑥 = − 2, this is the factor that we are talking
about and 𝑥 + 4 this is the factor that we are talking about −4.

Now, at these points, what is happening, what exactly is happening at these points? So,
when I consider the factor (𝑥 − 1)2 , because it is quadratic and if I recollect the graph of
a quadratic function, it behaved some it is not to the scale, but it behaved something like
this, right. It will never cross x-axis.

So, a similar feature is visible over here, when I consider the graph of this function. So, if
I consider a graph of this function, because 𝑥 − 1 is coming twice; it is square (𝑥 − 1)2
is there; so, what I am getting is the behaviour is of the quadratic nature, ok.
(Refer Slide Time: 10:39)

Now, let us look at that (𝑥 + 2)3 . What is a graph of (𝑥)3 ? A graph of (𝑥)3 is somewhat
like this; it crosses x-axis y is equal to x cube, it crosses x-axis. So, now that behaviour is
evident when I consider that instead of x, I consider (𝑥 + 2)3 that behaviour is evident
over here. It actually cuts and crosses x-axis.

And, if you look at the third factor that is 𝑥 + 4 which is 𝑥 = − 4, it is behaving like a
straight line that is also. So, what is happening here? I have two things; one and this one.
So, in these both cases we have odd degree polynomials and, the odd degree polynomials
as we know actually cross x-axis. And, in this case I have an even degree polynomial
which is actually bouncing off the x-axis, this is a typical feature.

So, if I have even degree what we are saying is, if the polynomial is a, the factor is of even
degree, then it will bounce off; that means, it will not cross x-axis but, if the polynomial
as odd degree, then it will actually cross x-axis. So, these are the two typical features that
we will employ while for plotting the functions which are of polynomial nature once we
understand the factors.
(Refer Slide Time: 12:25)

So, in the next slide, I have given a general description of these factors, you can go through
these slides later, but it is essentially the same that I have said just now, ok.

(Refer Slide Time: 12:38)

So, now for identifying zeros and their multiplicity. What do I mean by multiplicity? How
often that factor is appearing. In the previous case the factor 1 was appearing twice, factor
minus 2 was appearing thrice, and other factor was appearing only ones. So, if I want to
identify the zeros and their multiplicities, I should look at the shapes of the curves. For
example, if you look at the first graph, here the degree of the polynomial 𝑛 = 1, here
𝑛 = 2, 𝑦 = 𝑥 2 this is, and here 𝑛 = 3.

As mentioned earlier, it is more; it is more evident now, that when I have odd degrees, the
curve actually passes through x-axis, when I have even degrees we can draw 𝑦 = 𝑥 4 , but
this will be slightly broad and it will cut the x-axis, it will be somewhat like this. Let us
not get into that. But, for odd degrees you will get something of this form, or even degrees
you will get the bouncing off pattern and for odd degrees you will get a pattern which is
actually crossing, one minute.

Let me reiterate this that; this is very as this is very important. If you have an odd degree
polynomial, then you are almost sure you are sure to cross x-axis. If you have even degree
polynomial, you will never cross x-axis at that point, you will simply bounce off from x-
axis, ok. So, that gives us some more clarity. So, if the zeros of the polynomial or the factor
has even multiplicities, the graph will touch or is tangent to x-axis or zeros with odd
multiplicities, the graphs cross or intersect the x-axis.

Now, if you look at the even powers which are 4, 6 and 8, how will you guess, what is the
strength of the power? So, in that case the graph will still touch x-axis it will bounce off,
but which with each increasing even power it will appear to be flatter and flatter while
approaching the zero and leaving from the zero. For example, the base will broaden; in
this case, it will be something like this if it is 𝑥 4 ; 𝑥 6 further flattening.

If in a similar manner when you have odd powers like 5, 7 and then the graph will appear
to be more flat over here, and while leaving also it will leave slowly and then it will decay
at very fast rate. So, this is the typical feature from the bulge at these intervals you can
actually guess the multiplicity of a polynomial. That is the importance of this slide.

So, now we have added one more weapon in our arsenal that is we will identify the
multiplicities of the zeros. First we identify zeros. So, at the step zero is we identified given
a function where whether it is a polynomial function or not. Then, we identified the x-
intercept of the function that is zeros of the functions or roots of the functions.

After identifying roots of the functions, roots how many times repeated that is what, we
have identified here in this by using the graphical tools. This is quite powerful. And, you
can use it more often to understand the polynomial function. When you will actually solve
some problems on identifying the polynomial functions, you will get a better hold of it.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 39
Graphs of Polynomials: Behavior at X-intercepts

(Refer Slide Time: 00:14)

So, let us go ahead and look at the Graphical Behavior of Polynomials at X-Intercepts. So,
in particular if the given polynomial has a factor of the form (𝑥 − 𝑎)𝑚 , this 𝑚 is called the
multiplicity of the polynomial. And you will say 𝑥 = 𝑎, is a zero of a polynomial 𝑓 with
multiplicity 𝑚. This is to fix the terminology.

Now, the graph of a polynomial function will touch, but not cross x axis at zeros with even
multiplicities and the graph will cross x axis at zeros with odd multiplicity. We have
iterated it enough number of times.

Also, one important thing is the degree of the polynomial cannot exceed the sum of the
multiplicities, or the sum of the multiplicities is always less than or equal to the degree of
the polynomial function, this is quite common sense right. If it exceeds, if it actually
exceeds the polynomial degree then it is a polynomial of higher degree.

So, then now why it is not equal to, can be one question. The sum of multiplicities you
will say will always be equal to the degree of the polynomial function. For that we need to
understand that all the roots all the zeros of the polynomials or the only way we are
identifying zeros of the polynomials is by identifying the x intercepts.

So, all x intercepts are real roots of the polynomials, but as in the quadratic case we have
seen that some of the polynomials, some of the quadratic equations do not have real roots.
In such cases the x intercepts are not visible.

I will demonstrate it further through some examples.

(Refer Slide Time: 02:35)

So, let us try to see. So, given a polynomial of degree n, that is a graph of a polynomial of
a degree n. We want to identify zeros and their multiplicities; this is our goal. So, you are
you have been told that this polynomial is of degree n and this is the graph of the
polynomial, how, what will you do about it?

So, in that you will look at all the coordinates, where the graph touches x-axis, you take
them. If the graph touches x-axis and bounces off the x-axis then, it is a zero with even
multiplicity. If the graph actually crosses x axis, it is a zero with odd multiplicity. And
finally, when you will conclude you have to take care that because the polynomial is of
degree n, the sum of the multiplicities should never exceed the actual degree of the
polynomial.
Another thing if the graph crosses x axis and appears almost linear at the intercept, then it
is a single order; that means, it appeared only once; it is a linear function. And finally, that
is what I explained the sum of the multiplicities is no greater than this fine.

(Refer Slide Time: 04:10)

So, let us go ahead and see some examples and let us see whether we can apply these
principles in action. So, use the graph of a function of degree 6. So, it is a degree 6
polynomial, which is given to you and this is the graph, wonderful.

So, now you can easily see −2, 0 and 2. So, x intercepts are −2, 0 and 2 fine. Then if you,
let us start from left; so, at x is equal to minus 2, how is the behavior? It is more like a
straight line, it is more like a straight line. So, at minus 2, I feel the behavior is linear or it
is a onetime event.

At 0, what is happening is; it is having this S shape, somewhat twisted S shape. And that
is indicative of odd degree that is indicative of odd degree and degree can be 3 or 5 or 7 or
9 I do not know right, but I have been given that the polynomial is of order 6. So, at most
it can have a degree 5 right, the multiplicity 5.

Now, look at this particular junction, it actually bounces off the x axis. So, this is a typical
trait of even degree polynomial. So, what can be the degree? It can be 2 or 4 right; so it
can be 2 or 4 ok. Now, we need to collate this information. So, x intercept 𝑥 = −2 is linear,
there is no doubt about it. 𝑥 = 0, you have odd degree 3 or 5 is the degree and 𝑥 = 2 it is
even degree.

Now, together sum of the degree should be equal to 6, of which this 1 is fixed. So, now, I
can assign 3 or 5. If I assign 5 then 1 + 5 = 6 and if 1 + 5 = 6 that essentially means;
there is no root of the form 𝑥 = 2 that is not possible. So, I have to assign 3 over here.
Once I assign 3 over here, then I do not have any other choice, but to choose 2 over here.

So, therefore, I have identified the multiplicities of the factors like 𝑥 = −2 will have a
multiplicity of 1, 𝑥 = 2 has multiplicity of 2 and 𝑥 = 0 has multiplicity of 3. So, this is
how we will identify the factors, and identify the zeros of the functions, and identify their
multiplicities. This is much better for drawing the graph of a function.

Still, I have not answered a question that why this should go to infinity and why this also
should go to infinity; those things are not clear, but we will come to them later. Right now,
we have much better understanding about zeros of the polynomial functions and their
multiplicities and how we can use them to understand the function.

Now, another interesting thing that you can ask yourself is ok, I have seen this function
and I know the multiplicities of zeros and everything. Can I use this knowledge to actually
tell what is the equation of the function or not. We can try our hand on it, but we may not
be able to because 𝑥 = 0 is with multiplicities 3.

(Refer Slide Time: 08:39)

That gives me 𝑥 3 . 𝑥 = 2 with multiplicity 2 that gives me (𝑥 − 2) the whole square and
𝑥 = −2 with multiplicity 1 which will give me (𝑥 + 2) ok.

So, now I can say that this is the polynomial function that is graphed here, but how will I
verify? So, for that I need something which is non-zero. So, you can choose some point
and check whether this is there, but there is a catch over here.

It need not match the values that are given here. So, what we will do is though the factors
are correct the polynomial is of degree 6 here the degree is 6, we will put some unknown
a over here and we will determine this a by putting the actual values.

We will come to it later when we have better understanding about the behavior of this
kind, but right now you keep this in this point in mind; that we do not know this a. So, we
cannot actually give the exact equation of the polynomial with reference to these points.

Though we know the form of the polynomial, but we do not have accuracy up to the exact
matching on the coordinate plane with numbers. Just remember this point and this juncture
and let us go ahead and do some other problems.

(Refer Slide Time: 10:16)

So, in the next problem is of similar nature. I have a polynomial of degree 4 and I have
been projected with a graph of this polynomial function.
Now, this is an interesting example because, you have a polynomial of degree 4. In earlier
case we had a polynomial of degree 6 and all 6 roots were actually visible. In this case I
have a polynomial of degree 4, but there is only one root that is visible or one 0 that is
visible.

So, what is that number? It is 2 ok. And based on our understanding of the algorithms what
we know is the graph actually bounces off; the graph actually bounces off the x axis which
is 2. So, this is a typical trait of an even degree polynomial, or even degree even
multiplicity. So, in this case I can say it can have a multiplicity of 2 or 4. I cannot exceed
beyond 4 because the given polynomial has degree 4 only ok.

So, now because the given polynomial has degree 4, it is safe to assume that this
polynomial is of degree 4 right. But if this polynomial is of degree 4 you can see there is
a perturbation of the shape over here. It is not the shape of a polynomial of degree 4, for
example; 𝑦=𝑥 4 will not be in this form. So, I can rule out the degree 4 constraint.

Therefore, I do not have any other choice, but to say that the polynomial is of degree 2,
the multiplicity of this particular factor 0 is 2; that means, I have a factor of the form
(𝑥 − 2)2 the whole square that is all I can say in this case. Yeah, 𝑥 = 2, is of even degree
2 or 4 and hence the function based on the reasoning that I have given it must have a factor
of (𝑥 − 2)4.

Let us understand this graphically as well. This is a function, the blue line over here, is
actually (𝑥 − 2)2 . As you can see it passes very closely to through the function and the
other line is (𝑥 − 2)4 , other graph the green line is (𝑥 − 2)4 .

Now, if you look at graph of (𝑥 − 2)4 closely there is no possibility of changing the shape.
You can scale with that unknown a, but you cannot change the shape of the function. So,
this graph is actually ruled out. So, graphically we have understood why we are ruling out
and this graph is not ruled out. So, this is somewhat familiar. So, it will have some factor
of this form.

Now, one exercise for you is this graph actually though it is a polynomial of degree 4, the
way we have constructed is we have multiplied (𝑥 2 + 1) with (𝑥 − 2)2 .
And if you look at this particular factor x square plus 1 because you, now what you can do
is you can actually consider this (𝑥 − 2)2 as one factor. And you can see whether this 𝑥 2
you will get the same graph by multiplying this. The beauty of this example is that this
(𝑥 2 + 1) has no real roots.

Therefore, the degree though the degree of the polynomial is 4, I was not able to find the
two missing roots; those are not in the real domain. They are in the complex domain. So,
those things are there that is why the question that the sum of the multiplicities will always
be less than or equal to the degree of the polynomial.

So, let us say there are three factors 𝑚1 , 𝑚2 , 𝑚3 are the multiplicities. Then 𝑚1 + 𝑚2 +
𝑚3 ≤ 𝑛; that answers this question. So, in this case I was actually able to find for these
real roots only 2. So, 2 is less than or equal to 4. So, that is why the; this is the example,
why I cannot say that the sum of the multiplicities is always equal to the degree of the
polynomial.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 46
Graphs of Polynomials: End behavior

(Refer Slide Time: 00:15)

So, now we have understood how multiplicities affect the polynomial and how we are able
to find the multiplicities of the polynomial functions with some factors, correct? Still we
do not have an answer to a question that what why what is deciding this behavior that this
function will go to infinity, this function will go up as usual, how this behavior is decided,
we do not have any answer for that.

Let us try to understand that through end through what is called end behavior of the
polynomials. So, the next slide is actually the end behavior of the polynomials ok. So, let
us go to the next slide, it is end behavior of the polynomials.
(Refer Slide Time: 01:09)

So, what is an end behavior of the polynomial? In order to understand end behavior, let us
define an end behavior properly based on our understanding of quadratic equations. So,
when we studied quadratic functions, we looked at the term of the form 𝑎2 𝑥 2 + 𝑎1 𝑥 + 𝑎0
right and then, we talked about 𝑎2 𝑥 2 whether 𝑎2 > 0 or 𝑎2 < 0, then we decided the
behavior of the function.

If 𝑎2 > 0, then we said yes, if 𝑎2 > 0, the function will take its minimum and therefore, it
will go from both sides to infinity. If 𝑎2 < 0, then the function will take its maximum and
from both sides, it will go down and it will be unbounded. Now, from the graphs that you
have seen in the earlier lectures as well as in this lecture, it is very clear; it is very clear
that these functions, polynomial functions are either increasing or decreasing based on the
way they wish right.

So, for example, it can be like this also. So, or it can go like this also or it can be a straight
line as well, if it is linear or it can move like this. All these are polynomial functions. So,
now, we want to have a better understanding. So, what is the behavior of a function after
it has passed through all the roots is the question right, that is the term that was troubling
us a lot.

So, for quadratic equations, we have decided that it is basically based on a 2 because
quadratic equation the highest degree is 2. So, a 2 it is. So, now, in a similar manner, if I
want to consider a polynomial function 𝑓 (𝑥)=𝑎𝑛 𝑥 𝑛 + 𝑎𝑛−1 𝑥 𝑛−1 + ⋯ + 𝑎0 . Then, the
behavior should be decided by this term.

Why should I make this claim? Because if you look at 𝑥 𝑛 , what we are looking for is as
the value of 𝑥 increases or as the value of 𝑥 decreases. Now, it is not in that zone, where it
is passing through many roots. So, it has passed through all its possible roots and now,
after that how the function will behave? There is no determining factor right.

So, in such case, the only determining factor is the term 𝑎𝑛 𝑥 𝑛 ; why? Because for large
values of 𝑥 this term 𝑥 𝑛 will dominate all other terms corresponding to 𝑥; 𝑥 𝑛 raised to n
will dominate 𝑥 𝑛−1 , and so on that is when 𝑥 is becoming large. When 𝑥 is becoming
small that is 𝑥 is tending to −∞, the term 𝑥 will be the small 𝑥 𝑛 will be the smallest
possible term or if we that n is of even degree, still it will be the largest possible term.

In any case, the behavior of 𝑎𝑛 𝑥 𝑛 will play a dominant role in identifying the behavior
beyond roots of the polynomials or beyond zeros of the polynomials. This behavior we
will call as end behavior of a function and for polynomial functions, it is determined by
the leading terms that is 𝑎𝑛 𝑥 𝑛 .

(Refer Slide Time: 05:17)

If this 𝑎𝑛 > 0, and 𝑥 𝑛 that 𝑛 is a even power exponent is even, then as 𝑥 increases or
decreases, it is very similar to quadratic. As 𝑥 increases or decreases, 𝑓(𝑥) will always go
to infinity. If 𝑎𝑛 < 0 𝑛 is an even exponent, then whether 𝑥 increases or decreases, 𝑓 (𝑥)
will go to −∞. It will go on decreasing. Good. Then, what if 𝑎𝑛 𝑥 𝑛 that 𝑛 is the exponent
which is of odd power or exponent is odd. What happens?

If 𝑎𝑛 > 0, then as the function increases, 𝑓(𝑥) also increases. If 𝑎𝑛 > 0 and it is of odd
power as 𝑥 increases, 𝑓(𝑥) also increases; as 𝑥 decreases, 𝑓(𝑥) also decreases and both
are going to infinity; one is going to ∞, another one is going to −∞. They both are
unbounded. Similar thing can be argued for 𝑎𝑛 < 0. So, in order to improve our
understanding, I have tabulated this zone; 1 minute, let me remove this part ok.

(Refer Slide Time: 07:07)

So, this is the better understanding. So, now, you look at the leading term 𝑎𝑛 𝑥 𝑛 So, this is
referring to 𝑛, 𝑛 is of even degree, 𝑛 is of odd degree. So, if 𝑛 is of even degree and 𝑎𝑛 >
0, x tending to ∞, 𝑥 becoming larger and larger, 𝑓(𝑥) will become ∞; 𝑓(𝑥) will also
increase. 𝑎𝑛 > 0, x tending to −∞; that means, 𝑥 is becoming smaller and smaller and
smaller; but because the polynomial is of even degree, it will again go to ∞.

In a similar manner, if 𝑎𝑛 > 0 and the polynomial the leading exponent is of odd degree,
then as x tends to ∞, 𝑓(𝑥) tends to ∞. You can imagine a function of the form 𝑥 3 .
Similarly, if 𝑎𝑛 > 0, x tends to −∞, 𝑓(𝑥) will also go to −∞because 𝑓(𝑥) will also keep
decreasing.

Remember polynomials of odd degree crossover x axis, if you link that point to this, then
naturally it is very easy to and visualize the behavior of the polynomials. I will demonstrate
these two graphs again, once again to reiterate the point. If 𝑎𝑛 < 0, now 𝑎𝑛 < 0; that
means, 𝑥 becoming larger, 𝑓(𝑥) the term, the leading term of 𝑓(𝑥) will be negative more
and more negative.

So, 𝑓(𝑥) will tend to −∞; but if 𝑥 is becoming smaller and smaller, the exponent is of
even degree, still 𝑓(𝑥) will again go to −∞ because 𝑎𝑛 < 0. Come back to odd degree,
here the exact replica of what we have done for odd degree when 𝑎𝑛 > 0 will happen.

So, in this case when 𝑎𝑛 > 0, 𝑥 tending to ∞, we will make bring this 𝑓(𝑥) to go to ∞;
but in this case, it will bring it to −∞ and similar case is true for the other part that is x
tending to −∞, 𝑓(𝑥) will tend to ∞. Let us visualize it through graphs.

Let us take this first block even degree 𝑎𝑛 > 0. Imagine a function of the form 𝑥 2 or 𝑥 4
as x tends to ∞; both of them are going up. Just remember this figure that will clear this
understanding.

Let us go to odd degree with 𝑎𝑛 > 0, as 𝑥 tends to ∞, here this is going up. This is going
down right. Just imagine a figure of 𝑥 3 for the convenience. When 𝑎𝑛 > 0, just imagine a
figure of −𝑥 2 or −𝑥 4 , both of them should naturally go down. That is what is written here
as well. In a similar manner, just consider −𝑥 3 .

So, whatever was going down, will go up and whatever was going up, will go down that
is what I meant when I said this. So, now we have much better hold over end behavior of
polynomials. Now, you can look at the graph of a polynomial function and you by looking
at the end behavior, you can say whether the polynomial, the leading term of the
polynomial is of odd degree or even degree.

That is one more understanding, one more level of understanding that we have achieved
through understanding this end behavior. But that is not over. We further need better
understanding of the functions.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 42
Graphs of Polynomials: Graphing and Polynomial creation

(Refer Slide Time: 00:14)

So, now I want to use all this knowledge to plot a Polynomial Function or graph a
Polynomial Function.

(Refer Slide Time: 00:19)

So, let us reiterate what are the things that we have seen. For graphing the polynomial
function, one way is to find the tabular form and try to graph it as in a crude manner. More
knowledgeable way is, you follow this algorithm that is find x intercept, y intercept if
possible because it may happen that they do not have any real roots and you may not be
able to get x-intercept, all the x-intercepts right.

Then for graphing it is helpful to check the symmetry that is; if 𝑓(𝑥) and 𝑓(−𝑥) are same
if it is an even degree polynomial; that means, you have symmetry about y axis. If it is an
odd function you can check whether they are symmetric about origin that is 𝑓(−𝑥) =
−𝑓(𝑥). Typical case is the first symmetry is 𝑦 = 𝑥 2 , it is an even degree polynomial and
it is symmetric. So, once you have drawn here for − 𝑥 you have to just keep the mirror
image.

That is how it helps in graphing. In a similar manner a 𝑦 = 𝑥 3 is a odd degree polynomial

and 𝑓(−𝑥) = −𝑓(𝑥). Therefore, whatever you got about origin if you reflect about origin
then you will be able to retain the same shape; you do not have to compute explicitly. This
is the way this checking of symmetry helps.

Next identify the zeros; x intercepts we have already identified. So, you have identified
the zeros. Then you identify their multiplicities. If you identify the multiplicities of the
polynomials you know the behavior of the polynomials at x intercept. You just recollect;
multiplicity the sum of the multiplicities of all zeros cannot exceed the degree of the
polynomial that you have to keep in mind. After identifying the multiplicity you know the
behavior at the zeros of the polynomial function.

Now, you want to know the behavior beyond zeros of the polynomial function that is; the
end behavior. So, end behavior you can use the leading term and you can identify the
behavior. Remember the table that we have shown for identifying the end behavior.

And finally, you use the end behavior the behavior at intercepts to sketch the graph.
Turning points - the number of turning points can be identified we may not be able to
locate exactly where the turning point is. For that, you need the tools of calculus to identify
the exact location of a turning point.

And when you identify those when you roughly estimate the turning points; kindly ensure
that the number of turning points do not exceed one less than the degree of the polynomial.
So, if the degree of the polynomial is 𝑛, the number of turning point should not exceed n
− 1 ok. And finally, you can use technology to sketch the graph. So, use graphing tools
like Desmos or some other tools for graphing the function ok.

(Refer Slide Time: 04:07)

So, let us see this in action. So, here is an example I want to sketch a graph of this
polynomial function; −(𝑥 + 2)2 (𝑥 − 5). Obviously, I have figured out oh it is a
−(𝑥 + 2)2 . So, the first thing that we; so, I want to graph this function. So, the first thing
that I want to find is the x intercept, because it is given in factored form it is no brainer;
𝑥 = − 2 which is this point 𝑥 = − 2 and then (𝑥 − 5). So, 𝑥 = 5 which is this point.

These two are the zeros of the polynomial functions; 𝑥 = − 2 has multiplicity 2 and it is
an even degree polynomial. So, over here the behavior of function I am trying to sketch it
will come from here, it will go from here. So, I know the behavior of the function is of this
form, it will just pass through the axis.

And 𝑥 = −5 I do not know the exact values, but roughly it will be more of the linear form
and it will pass through the point − 5, then it will come down over here and then it will
pass through this point. So, up to this I am ok. Now, you look at this polynomial if you
look at this polynomial, then the polynomial will be a cubic polynomial; it will have a
negative term.
So, essentially 𝑎3 < 0 ok. So, the end behavior of this polynomial because 𝑎3 < 0 as 𝑥 →
∞ this function will tend to infinity. Yes, and as 𝑥 → −∞ the function will naturally go up
like this.

So, this is the vague understanding of the behavior. If I want to get more precise on what
values this is roughly the shape of the function. If I want to get more precise on what values
the function takes, I can consider the y intercept as well that is I will put 𝑦; 𝑥 = 0. So, it
will be 22 4 22 4 yes, and into − 5 that will give me − 20 + 20. So, this intercept that I
have drawn is wrong. It should be somewhere here + 20. So, let me erase this and redraw
the function again.

Let us take the eraser. So, it may not go this high as well. So, over here the font behavior
of the function ok; so, let me again go back to the marker. And the function may cut here
itself pass through this point and join this point. Yes, so, this bulge will not be there
because this function is linear over here. It may be of this form. So, let me again erase this
part yeah.

So, let us see. So, we have identified the end behavior ok, final check that number of
turning points. The function is cubic, so it can have at most two turning points, there are
only two turning points: one is here, one is here fine. So, let us see whether whatever we
have said is correct or not. So, let me hide this first.

So, x intercept is -2 and 5, no problem. 𝑥 = − 2 has multiplicity 2. So, the quadratic

behavior should be plotted there. Yes, 𝑥 = 5 has multiplicity 1. So, a linear behavior is
plotted here assume this is a line. So, linear behavior is a plot must be plotted here. And,
then 𝑓(𝑦) intercept 𝑓(0) is 20 which we corrected we were not correct in the initial stages
and the leading term is −𝑥 3 .

So, therefore, the odd degree polynomial with negative leading coefficient has the
following end behavior; as 𝑥 → ∞ 𝑓(𝑥) → −∞, this is the behavior that we have plotted;
𝑥 → ±∞ 𝑓(𝑥) → ∞, this is also correct. And 𝑓 can have at most 3 − 1 = 2 turning points,
this is the behavior right.

So, now, I was roughly ok in drawing the graph of a function. This is because I do not
exactly know the behavior of the turning points. So, I will be roughly ok in drawing the
graph of a function, but not exactly. If you want to be more precise you can actually
tabulate the values around some critical points and then you can figure out. This is when
the formula is given to you.

Now, the question can be asked that what if the formula is not given to you, but you have
been given only a function. And from the graph you need to identify the polynomial.

(Refer Slide Time: 10:17)

In such cases one theorem which will help you a lot, I will not use this theorem in a rigorous
manner. But it will help you a lot, is intermediate value theorem because we are dealing
with continuous functions. This intermediate value theorem is valid for all continuous
functions.

What this theorem says is, a polynomial function is a continuous function. So, let 𝑓 be a
polynomial function, then the intermediate value theorem states that; if
𝑓(𝑎) 𝑎𝑛𝑑 𝑓(𝑏) have opposite signs; that means. So, let us say 𝑓(𝑎) > 0, and 𝑓(𝑏) < 0
and 𝑎 > 𝑏, then there exist at least one c between a and b such that 𝑓(𝑐) = 0; that is
essentially the meaning.

For example, I have this coordinate plane my value of 𝑓(𝑎) is here, and 𝑓(𝑏) is here, and
the function that is given to me is a continuous function right. So, finally, it has to pass
through the x axis to reach the value here right. So, in such cases we will say that; this is
the 0 of the polynomial that is what we are calling as c, 𝑓(𝑐).
So, you using this you when you are actually having trouble in finding the zeros of the
function, you can actually evaluate two values any two values of opposite signs. And if
you evaluate any two values of opposite signs, then you know that there is some root some
0 in between that will improve that you will gain a confidence by doing these things.

So, this is an important theorem in mathematics, intermediate value theorem. You can use
this to find the roots of the polynomial when you are having difficulty in identifying the
roots of the polynomial. So, you simply put 𝑓(𝑎) and 𝑓(𝑏) if they have opposite sign, then
there is at least one root in between; that is the meaning. You can use this theorem to your
advantage.

(Refer Slide Time: 12:37)

So, using this theorem we can actually derive a formula for polynomial function. You use
this theorem to identify the zeros; rest of the methodology is similar. So, how to derive a
formula for polynomial functions? So, given a graph of a polynomial, how to find with in
coordinate axis? You have all the numbers attached to it, then the question can be asked
as to how to find the polynomial function the algebraic expression of a polynomial
function?

So, in that case our modus operandi is similar to what we have done. Find the x-intercepts
from the graph. Find the factors of the polynomial, this we already know. Understand the
behavior of x intercepts around x intercepts to get more understanding of the x intercepts
that is zeros of the polynomial about their multiplicities.
So, you will find multiplicity of each factor. Once you have gained understanding identify
the end behavior that also you have to do. Next, after doing that you find the least degree
polynomial containing these factors. What are the factors? Those are x intercepts that you
have figured out. You have also seen the end behavior, so the least degree polynomial
which will give you that particular function behavior.

Once the least degree polynomial is figured out you use any point on the graph that is why
the coordinate axis is important, the numbers are important. You use any point on the
graph, in particular y intercept is the easiest and in that case you can determine the stretch
factor.

The stretch factor over here is the unknown a that I have told you while figuring out the
factors in one of the examples. So, that is the stretch factor. It will be more clear when we
will solve the examples ok. So, this is our recipe for attacking the problem of deriving the
formula given a graph.

(Refer Slide Time: 14:58)

So, let us try to apply this recipe to one example. So, write the formula of a polynomial
given in the graph, the graph is here ok. So, I will go around and try to find the x intercepts
of this graph. So, one x intercept is here which I think is 𝑥 = 1 and other x intercept is -2.
So, − 2 and 1 are the x intercepts; y intercept over here is−2, 0− 2 is the y intercept. So,
we have identified x and y intercepts.
The graph actually seem to have two turning points. So, the least degree if it has two
turning points the least degree polynomial will be because 𝑛 − 1 = 2. So, the least degree
polynomial should be cube degree 3 polynomial right ok. And since it is crossing over this
end from end behavior also you will have some understanding that it is yeah, it should be
an odd degree polynomial. So, therefore, the polynomial may be of degree 3, correct.

It should be an odd degree polynomial; it has only two turning points. So, the least degree
of the polynomial is 3. Now what you will do next? Next I want to identify the
multiplicities, that is 𝑥 = 1 it the function more or less seems to be linear and at 𝑥 = − 2,
the function more or less seems to be quadratic.

So, it is very easy in this case because, 𝑥 = − 2 is a even degree behavior, x is equal to
because it is bouncing off. So, it is a even degree behavior and 𝑥 = 1 is linear behavior
and the polynomial is of degree 3 or more, but odd degree. So, the first instance is you
guess the function to be of the form (𝑥 + 2)2 (𝑥 − 1). So, now, I have not yet used the
information that the intercept the y intercept is happening at − 2, correct.

So, that information I have to use now because that is the function value that I have. These
are the based-on factors we are basically equating to 0 right. So, the a may be missed out.
So, where the non-zero value comes you should be able to figure out. You can you are free
to choose any value, but for me it is better to choose y intercept.

So, y intercept is − 2, we have already seen that, but if you put what is y intercept? It is
𝑓 (0). So, if you put this in the function form the value of 0 in the function form over here
you will get actually this is to be equal to −4𝑎. So, if you are getting this to be equal to
1
−4𝑎, then −4𝑎 must be equal to − 2; that means, 𝑎 = 2 great.

1 1
So, if 𝑎 = you substitute this value into the function. So, 𝑓(𝑥) = (𝑥 + 2)2 (𝑥 − 1),
2 2

fantastic. So, you have you got an algebraic expression. Now, to match this algebraic
expression, you use the technology that is graphing tool to plot the function and you can
verify the result for yourself that yes, this is the function that we have actually plotted ok.

So, this is the complete understanding of two-step mission that is; given an algebraic
expression how to graph the polynomial function. Given the graph of a polynomial
function, how to write an algebraic expression of a polynomial function. This ends our
topic on polynomial functions.

Thank you.
 Mathematics for Data Science 1
Week 07 - Tutorial 01
(Refer Slide Time: 0:14)

Hello mathematics students. In this tutorial we are going to look at questions based on graphs of
polynomials. So, in this question there is this polynomial 𝑝(𝑥) whose graph is given here and we
are supposed to comment on the following statements, the number of turning points, so that is easy
so there is a turn here, 1, 2, 3, 4, 5, 6 and 7. So, there are 7 turning points. And then we are asked
the number of roots, so roots would be where the polynomial touches or cuts the 𝑥 axis so that is
1, 2, 3 and 4 and 5, so there are 5 distinct roots.

Now, what is the minimum possible degree? Minimum possible degree of this polynomial based
on the number of roots. So, the minimum possible degree would be the same as the number of
roots so if there are 𝑛 roots to a polynomial then it should have a degree of at least 𝑛, so 5 is the
minimum possible degree of this polynomial based on the roots. But now they are asking what is
the minimum possible degree based on the turning points.

So, here we see this thing a straight line has no turning points, a quadratic equation has 1 turning
point and a cubic would have 2 turning points at most likewise a quartic that is a fourth degree
polynomial would have 3 turning points at most. So, if you have 𝑛 turning points, then the
minimum possible degree of the polynomial would be 𝑛 + 1. So, here that is 8.
Now, what would be the minimum degree of the polynomial given all the information we have?
Then you know that it has to be at least 8 the five which is on the basis of the roots is a lesser
number than it and we know already that it has to be at least 8, so the minimum degree of the
polynomial should be the greater of these two which is 8 because 6 and 7 and 5 are not allowed on
the basis of turning points.

And then we are being asked what is the end behavior and the coefficient of the highest degree
term. So, the end behavior shows that the polynomial is coming from ∞ and going to ∞ which
means the degree of the polynomial is definitely even. So, we can say that it is an even degree
polynomial.

So, as you can see we have just drawn these basic raw curves for the linear and quadratic and cubic
and quartic polynomial. So, linear which is an odd degree polynomial it comes from -∞ and it
goes to +∞ whereas quadratic a parabola here it is coming from -∞ and it is going to -∞ .

So, when it is even degree you see that the ends of the curves are in the same directions. Similarly,
for quartic here this is coming from ∞ and going to ∞, whereas for a cubic this is coming from -∞
and going to +∞. So, here this is coming from ∞ and going to infinity.

(Refer Slide Time: 4:17)

Therefore, this is a even degree polynomial and the coefficient of the highest degree term. So, the
coefficient of the highest degree term determines whether the behavior of the polynomial as x
increases whether it is going to +∞ or -∞ , if the coefficient of the highest degree term is positive,
it goes to +∞. So, if this is going to +∞ so this has to be positive coefficient for the highest degree
term.
 Mathematics for Data Science 1
Week 07 Tutorial 02

(Refer Slide Time: 00:15)

Now second question there is newly laid road which follows the path of this polynomial about
some coordinate system, from 𝑥 = −5 to 𝑥 = 20. And railway track is laid along the 𝑥 axis. So
how many level crossings are there? So what we are interested in is; how many times does the 𝑥
axis cut this polynomial? And for that we have to find the roots of this polynomial because roots
give when the polynomial is touching or cutting the 𝑥 axis.

Now this is of quartic forth degree polynomial multiplied with the quadratic polynomial, so the
degree is 6, so at best we could have 6 roots but let us find out what these roots are. The easy way
−𝑏±√𝑏 2 −4𝑎𝑐
to start is to first find the roots of the quadratic, so that would be using . We get
2𝑎
−15±√225−200 −15±√25
, which is that is essentially 5 or 10.
2 2

10 20
So you get or so 5 or 10 those are the two roots and they are both within the given range.
2 2

Anyway now we look at the other part, the quartic part. So here we have 𝑥 4 − 5𝑥 3 + 6𝑥 2 + 4𝑥 −
8. In this situations what is typically suggested is that we do a little bit of trial an error, we try out
with the basic small integers and we see if we can find any roots at all.
(Refer Slide Time: 02:26)
So let us start with 𝑝(0), 𝑝(0) is −8 which is clearly ≠ 0, so 0 is not a root then we have 𝑝 (1)
which is 1 − 5 + 6 + 4 − 8 = − 2 which is again ≠0, so not a root. Then we try 𝑝(−1) and we
get 1 + 5 + 6 − 4 − 8, so this is equal to 12 − 12 = 0. So yes 𝑝(−1) gives you 0 which means
we have another root that is - 1.

So let us note down our roots that we have found here, roots we have found so far are 5, 10 and -
1. Now going back to our trial and error let us try 𝑝 (2) and 𝑝 (2) gives us 16 − 8 × 5 = 40 +
6 × 4 = 24 + 8 − 8. So we get 16 + 24 = 40, 40 − 40 = 0, so this is 0. So we have another
root that we have found. So we now have two roots for our quartic and those two roots give us
another quadratic which is (𝑥 + 1)(𝑥 − 2) that is 𝑥 2 − 𝑥 − 2 .
So if we divide our quartic with quadratic we will get the other quadratic within it. So here we
have 𝑥 4 − 5𝑥 3 + 6𝑥 2 + 4𝑥 − 8 and we divide it with 𝑥 2 − 𝑥 − 2 so here go 𝑥 2 so 𝑥 4 − 𝑥 3 +
(𝑚 − 2)𝑥 2 -,+ and + cancel this of you get −4𝑥 3 + 8𝑥 2 + 4𝑥 − 8.

And then we do −4𝑥 times this, −4𝑥 3 + 4𝑥 2 + 8𝑥, so + - and - cancel this and here we have
4𝑥 2 − 4𝑥 − 8. And that is just 4 times this, so + 4. So our quartic, so is basically x(𝑥 2 − 𝑥 −
2)(𝑥 2 − 4𝑥 + 4) and this gives the quartic and additionally we have to also multiply for our p of
𝑥 we have to multiply the other quadratic which is 𝑥 2 − 15𝑥 + 50, this one.

So this is p of 𝑥 totally, an if we further separate it out into all its roots we get this one as we know
is 𝑥 + 1 into 𝑥 - 2 and this is if you notice 𝑥 - 2 to the whole square, so (𝑥 − 2)(𝑥 − 2) and then
here this we have found the roots already which is 𝑥 − 5 into what was the other root; the other
root was 10, 𝑥 − 10.

So these are our roots and the coefficient of 𝑥 power 6 will be positive clearly. So therefore this
is an even degree polynomial and thus if we have to sketch the graph it look something like this,
it comes from infinity and what is the least lowest root here, the lowest root is - 1. So at - 1 if we
draw this as the 𝑥 axis at - 1 you have one root it crosses the 𝑥 axis and then it goes around and
it comes to 2.

But 2 is a triple root, so what happens with a root if it is a single root it crosses the 𝑥 axis but if it
is a double root it will just touch the 𝑥 axis and come back but since it is a triple root it actually
crosses the 𝑥 axis. So here we do have a crossing and then afterwards at 5 and 10 we will have,
so this will be for two this will be for - 1, this will be for 5 and this will be for 10. This is just a
rough plotting of the graph.

The question was how many times does it intersect the 𝑥 axis, so we have to draw this basic
sketch and we find that the intersection are 4. If 𝑥 - 2 was not a triple root, if it were a double root
or a quadruple root like if it is there are 2 times or 4 times then the graph would be very different.
It would be 𝑥 - 1 would still be the same but at 2 you would not actually see a intersection, you
will just see a touching. It would not be a cut.

So therefore we have to check how many times the √2 occurs. Since it is an odd number of times
we can say it is actually cutting 𝑥 axis and that gives us a number of level crossings is 4. And
how many turning points are there? Now we can look at our graph and quickly tell; 1, 2 and 3; 3
turning points.
 Mathematics for Data Science 1
Week 07 - Tutorial 03

In this question Saraswati bought an 8 gram gold chain for rupees 40000, so we can presume that
1 gram is 5000 rupees on first November. And after 10 months that is August 2021, she sold the
chain and bought a new 10 gram gold chain by paying an additional 10000 rupees. Suppose, the
rate of the gold per gram is denoted by 𝐺(𝑡) and it is a function of time 𝐺(𝑡) is given to be this
cubic polynomial here and we are taking 𝑡 is to be 0 at the time when Saraswati bought her first
gold chain. So, t is a number of months since her buying her first gold chain.

Now, what is the when 𝐺(𝑡) is a polynomial of the rate for both used and new good. So, all gold
has the same rate as what we are considering and what is the rate of gold per gram when she sold
her first chain. So, after 10 months at 𝑡 = 10 is what we are really looking for. So, that means we
are looking for 𝐺(10) and that gives us 0.07 × 1000 − 1.4 × 100 + 7 × 10 + 5 and this is 70 −
140 + 70 + 5. So, that is actually 5. So, the rate is back to 5000 per gram. So, it is again rupees
5000 per gram.

Now, if she had sold the first gold chain after 6 months how much extra would she have paid for
buying the 10 grams gold chain? So, after 6 months we have to find the price, the rate, so that
would be 𝐺(6) and that is 0.07 × 63 is 216 − 1.4 × 36 + 7 × 6 + 5. And then we get this is
15.12 − 1.4 × 36 = 50.4 + 42 + 5 = 42 + 5 = 47 + 15.12 = 62.12 that is 62.12 − 50.4 =
11.72 that would give us then the rate is 11720 rupee per gram. And Saraswati is selling 8 grams
at this price that would mean she basically has to pay for the additional 2 gram and that would be
2 × 11720 which is equal to rupees 23440, this is how much she pays extra for her 10 gram gold
chain.
 Mathematics for Data Science 1
Week 07- Tutorial 04

(Refer Slide Time: 0:14)

A skydiver jumps out of a plane travelling at 3000 meter above sea level. And when she was about
500 meter above the sea level, she opens a parachute. And then she dives into the sea and reaches
30-meter-deep into the sea. Then she swims and reaches a sea coast from there she takes a
helicopter and reaches her home as shown in the figure.

So this figure is between the height and time there is no 𝑥 coordinate in this. This is only about the
y coordinate taking sea level to be 0 and the time. So initially she is way above sea level, so this
point is going to be our 3000 because as where she is jumping from and then she is dropping quite
quickly and then she slowly dropping after she opens a parachute. And she reaches under the sea
so here it is negative till she goes to the point where it is - 30 meter below sea level.

And from there she swims out and she takes a helicopter and she goes. So we are supposed to see
which of these option are correct and the range of the curve so formed is - 30 to 3000 which is true
- 30 to 3000 is her total 𝑦 coordinate range. And the domain of the curve will be the time taken for
the entire journey that is true so your curve starts here when she jumps till the point she reaches
home.

So this is correct this is also correct. Number of turning points are 5 so I see only 1, and 2, 3 turning
points. So this is not maybe this is a turning point I am not able to say because it appears to go a
little like this and then bend down. So probably this is a turning point. But either way there are not
5 they are less than 5 so this is wrong. And then the degree of the polynomial formed by the curve
will be at least 6.

Now we have to look at the roots here. So let us take this root this is a single root it just cuts the
𝑥axis like this, whereas this is a more, if the 𝑥 axis like this, it is kind of touching it this way and
that only happens if your root is a triple root at least. So it cannot happen for single root and it does
not happen for any even powered root because for root which occurs even number of times you
would not cut the 𝑥 axis.

So this has to be at least a triple root. So this is a single root the first one is the single root this one
we are assuming it is a triple root because they are asking for minimum degree at least so we are
looking for what it what the number of roots is in the minimum. And here there is a root which
occurs an even number of times because it is touching the 𝑥 axis and turning around it is not
actually crossing the 𝑥 axis.

So this is at least a double root so we have one +3 +2. So we have at least 6 roots therefore the
degree also has to be at least 6. So this is also correct.
 Mathematics for Data Science 1
Week 07- Tutorial 05

(Refer Slide Time: 0:14)

In this question, we are looking at an Electrocardiogram which is often called ECG or EKG. This
is a recording of electrical changes that occur in your heart during a cardiac cycle. So here we have
some ECG shown to us as a polynomial. And they are asking identify the number of turning points.
Ok? So that will be 1, this is 2, this is 3, 4, 5, 6, and here this is not a turning point, it is flattening
out like this and rising. Therefore it is not a turning point. We already have 1,2,3,4,5,6 so this is
7,8,9. That means we have 9 turning points. And then they are asking for the minimum sum of
multiplicities for that we look at the roots and so this one is directly cutting through this root is
directly cutting through the axis.

So the multiplicity of this is 1 and here this it is touching and coming back so it has to have an
even multiplicity. So the minimum is 2 and then here again this and this are both 1 each +1, this
1 also should be 1. And here we see this flat lined situation which occurs when you have an odd
multiplicity but not 1. So the minimum there would be 3 and then this is a 1 this is a 1 and this also
has to be 1.

So plus 1 + 1 + 1 which gives us all put together 3, 4, 5, 6, 9, 10, 11, 12, 12. So the minimum
sum of multiplicities is 12. No electric activity that is flat lined surface ECG usually indicates the
death of a person. This is as shown in the figure after 𝑈1 so after 𝑈1 we presume that it is a, it is
basically along the 𝑥 axis what polynomial will it be called for the domain after 𝑈1 . Clearly a 0
polynomial. It is simply 𝑦 is equal to a constant. So it has no degree and it is a 0 polynomial.
 Mathematics for Data Science 1
Week 07- Tutorial 06

(Refer Slide Time: 0:14)

So here we have a company’s profit varies according to the months. So they are going to have a
profit versus time along in the profit in thousands for year 2018 is represented by this polynomial
𝑝(𝑥) is equal to this quartic polynomial fourth power polynomial where 𝑥 represents the month
number starting from January as 𝑥 = 1. So January is 𝑥=1. The company declares the month as a
golden month if the profit is ≥150 thousand.
So golden month is when 𝑝(𝑥) ≥ 150. Find out how many months the company enjoyed the
golden month in the year 2018. So how many times does this happen? Alright, and there is a hint
also given to us where this cortex is apparently equal to. So they have basically given us the roots
of the polynomial and a does not matter, 𝑎 > 0.

And so we can also tell what a is 𝑎 has to be minus 0.211 because that is the coefficient of 𝑥 4 here
and minus a will be the coefficient of 𝑥 4 and the RHS therefore 𝑎 has to be equal to 0.211, −𝑎 =
− 0.211. Therefore, 𝑎 = 0.211. Anyway so now given that we already have the roots, the roots
are essentially 1.7, 4.117, 8.776, and 11.189. So these are the roots and the coefficient of the
highest power of 𝑥 is negative.

It also even so our curve is going to be something like this. Where the 𝑥 axis cutting it here and
that would mean this point is 1.7, this is 4.117, this is 8.776, and this is 11.189. So all of this is
given to us but what we are supposed to find is to related to 𝑝(𝑥) ≥ 150. So given that 𝑝(𝑥) is all
this stuff plus 5 and here we have the same terms of 𝑥 − 145, we can see that this particular
polynomial given to us is simply 𝑝(𝑥) − 150.

So whenever this polynomial that we have drawn here is greater than it. We have a golden month
so that would be month 2, 3, and even 4, then 5, 6, 7, and 8 do not come in and here we would
have month 9, and 10, and also 11. So as you can see this, this, this 3 and again 3 here. So it is
overall 6 months during which we can also tell which ones is Feb, March, and April, and here this
is September, October, and November.

So 6 months are the golden months in that year for this company.
 Mathematics for Data Science -1
Week 07 - Tutorial 07

(Refer Slide Time: 0:14)

In this question, we are given a polynomial 𝑝(𝑥) which is a product of a quadratic with a
monomial and another monomial. And the quadratic has some variable 𝑘 in it, capital K is the
set of values of this small 𝑘, choose a correct option if 𝑝(𝑥) always has 4 real roots but they
need not be distinct and we already know that 5 and 3 are roots because of these two
monomials. So, what is remaining is that our quadratic equation also should have roots.

And for that the discriminant which is 𝑘 2 − 16 should be ≥0. That would indicate 𝑘 2 ≥ 16,
thus 𝑘, the magnitude of 𝑘 ≥ 4. If 𝑘 ≥ 4 you get a repeated root you get the same root twice,
so what corresponds which option corresponds to this is a because you go from −∞ to −4 and
then 4 to +∞ and their union and 4 and − 4 are with closed intervals therefore, they are
included.
 Mathematics for Data Science -1
Weel 07-Tutorial 08

(Refer Slide Time: 0:14)

Question 8 is very closely related to question 7, we again have the same, quadratic into
monomial into monomial is the same polynomial and again the same set is given to us. Now
we have to see the correct option for 𝑝(𝑥) to have four distinct real roots, that means a root
should not be equal to each other and that is a catch.

So, we have already seen that 𝑘 2 − 16 the discriminant being equal to 0 will give us equal
roots. So, this case is not done, this time the discriminant has to be greater than 0, so that would
indicate |𝑘| > 4, so you will have (−∞, −4)⋃(4, ∞) for the quadratic condition. The other
condition here is that the roots for the quadratic should not be equal to 5 or 3.
(Refer Slide Time: 1:24)

−𝑘±√𝑘 2 −16
So, these routes which are this should not be equal to 5 or 3.
2

(Refer Slide Time: 1:46)

−𝑘±√𝑘 2 −16
So, for finding that condition let us start with = 5 let us start with this and you get
2

−𝑘 ± √𝑘 2 − 16 = 10 and that would mean 10 + 𝑘 = ±√𝑘 2 − 161.

Now if you square this we do not need to worry about the plus or minus, so let us square it and
we will reach 100 + 𝑘 2 − 16, 𝑘 2 and 𝑘 2 goes away. So we get 20𝑘 = − 116 and that would
imply 𝑘 = − 5.8. So when 𝑘 = − 5.8 the root of the quadratic part will be equal to 5.
(Refer Slide Time: 2:52)

So the root of this part will be equal to 5 and that is not allowed, so we should somehow
eliminate 5.8 from this set, − 5.8 from this set.

(Refer Slide Time: 3:08)

−𝑘±√𝑘 2 −16
And further let us check for three case where ≠ 3 so we first check when is it equal
2

to 3 and that gives us −𝑘 ± √𝑘 2 − 16 = 6 that gives us ±√𝑘 2 − 16 = 6 + 𝑘 and that further

gives us 36 + 𝑘 2 + 12𝑘 = 𝑘 2 − 16. So 𝑘 2 and 𝑘 2 canceled off and that gives us 12𝑘 = −52
−52 −13
this implies 𝑘 = which is essentially for 3 and 4 13, so .
12 3

(Refer Slide Time: 4:09)

 −13
So, 𝑘 should not be also be equal to , so which of these options does that is here we see
3
−13
option c is goes from (−∞, −5.8)⋃(−5.8, ) and keeping it open interval we are basically
3
−13
exploding −5.8 and similarly the open interval on the side on in this and this is essentially
3
−13
excluding and lastly we are doing the union with 4, ∞. So, this is correct, we are excluding
3
−13
all values from −4 and 4 and also excluding −5.8 and also excluding .
3
 Mathematics for Data Science -1
Week 07-Tutorial 09

(Refer Slide Time: 0:14)

For our last question, there is a company and they are making a particular product and the
demand of the particular product is us 𝑑(𝑥) , the production of the same product is 𝑝(𝑥) for 12
months, where 𝑥 is the number of months after January and for January we are taking 𝑥 is equal
to 0. And then they have given us 𝑑(𝑥) − 𝑝(𝑥), as a polynomial and this is essentially a
quadratic multiplied by a monomial by another monomial and another monomial.

So, we have a fifth degree polynomial here 𝑑(𝑥) − 𝑝(𝑥), then find out which months should
company reduce production after March. So, reduced production would mean 𝑝(𝑥) is greater,
that means 𝑑(𝑥) − 𝑝(𝑥) < 0 and we are interested in those situations where this curve is less
than 0 and so we just try to graph this curve and 𝑥 2 + 1 has no real roots.

So, the only roots here are 2 and 5.8 and 11.6, so our curve is going to look something like and
since 𝑎 is positive, then since the coefficient of the highest power is positive we will have this
how polynomial go to ∞ as 𝑥 goes to ∞ and then we have a situation like this because this odd
polynomial it goes to -∞ here and there are only three roots, there is 2 and 5.8 and 11.6.

So, we are looking for when is it negative and that we have to do only in the 0 to 12 range
because we are only looking for months of 1 year, actually even 12 is not correct because we
are starting from 0 we are only going till 11. So, if the root is 11.6 then 11 is somewhere here
and this is a 5.8, 6 is somewhere here and so these are the months where you get negative. Now
six is not June it is actually July because Jan is taken to be 0.

So, we have July and then August is 7, then September would be 8 and October is 9, November
is 10 and December is 11 and all these months you have the function being, the polynomial
being lesser than 0. So, these are the months where they should reduce production and they are
mentioning after March, if we looked at it before March, then you would have also have to
consider January and 0 and 1 which is January and February but since they are specifically
saying after March this is the set of months where the company should reduce production,
thank you.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 8.1
One-to-One Function: Definition & Tests

(Refer Slide Time: 00:14)

Hello students. Today, we are going to start a new unit called exponential functions and
logarithmic functions. In that our goal today is to understand exponential functions. For
studying exponential functions, a concept of a One-to-One Function is very much relevant.

So, we will first study that concept. Then, we will go to a definition of exponential function
and properties of exponential functions as we studied for polynomial and straight line case.
Then, we will define one interesting function which is called the natural exponential
function, and I will justify why it is called natural and what are the properties that it shares
with the mathematics world.
(Refer Slide Time: 01:11)

So, let us start and I think it is time to start and let us start. So, first let us go to a concept
of one-to-one function. In order to define the concept of one-to-one function, let us quickly
recall, what is a concept of a function? So, whenever I talk about function, I will talk about
𝑦𝑦 = 𝑓𝑓(𝑥𝑥). This 𝑓𝑓 when I say I need some domain let us say A and I need some co-domain
B.

In general function can be defined on any two sets, but for us let us consider to simplify
our understanding. So, that we will understand in terms of coordinate plane A and B to be
subsets of real line, ok. So, now my domain according to this definition is A, and co-
domain is B. So, now, what is the function? Function is a relation between one set to the
other set. So, it is a mapping that assigns values from one set to the values of other set.
(Refer Slide Time: 02:16)

Let us describe this function as something like this, ok. So, let us see this is the set A which
we will call as domain, and this is the; this is the set B that we will call as co-domain. This
is A this is B. What does the function do? If we feed some value of 𝑥𝑥 from this set, it will
process the value and spit out or give us 𝑓𝑓(𝑥𝑥).

It is like a popcorn machine; you are feeding in the corn and getting out the popcorn. So,
this is how the function works. Now, let us see, what are the cases that can happen?
Suppose, for same value we are getting different outputs that, is the case, when you
consider one 𝑥𝑥 more than one 𝑓𝑓(𝑥𝑥), ok.
(Refer Slide Time: 03:10)

So, now, is this a function is a question mark. We will soon answer the question. Then, it
can so happen that you have fed two different values and you are getting the same output,
ok. Is this a function? We will answer the question. And, for every single output that you
produce, there is a unique output that is produced by 𝑓𝑓, that is a one 𝑓𝑓(𝑥𝑥) only.

So, let us analyze these three again is this a function this. So, let us analyze all these three
things together. So, let us start with the first case that is, one 𝑥𝑥 more than one 𝑓𝑓(𝑥𝑥). The,
what do I mean by one 𝑥𝑥 more than one 𝑓𝑓(𝑥𝑥). Suppose, I have put 𝑥𝑥1 . If I put 𝑥𝑥1 as a value
if ones gave me 𝑓𝑓(𝑥𝑥1 ), other time it gave me 𝑓𝑓(𝑥𝑥1 ) which are not equal.

Is this a function? Is this a well-behaved function? It is not. So, I will say no, no this is not
a function. Therefore, the first thing the first case I will say is not a function. Because, we
are dealing with coordinate plane, let us understand this function with the coordinate plane.
(Refer Slide Time: 04:37)

So, what happens here is suppose, I have been I am taking the value one value 𝑥𝑥1 on
𝑥𝑥 −axis. Let us say this is 𝑥𝑥 −axis, let us say this is 𝑦𝑦 −axis. And now, I am taking one
value of 𝑥𝑥1 . So, one I once I fed in I got something which is 𝑓𝑓(𝑥𝑥1 ) here. And, other time I
fed in I got something which is 𝑓𝑓(𝑥𝑥1 ) here, ok. Now, what is this? This is actually while
if at all some curve I have to draw, it can be like this.

And, what happens here is for same value of 𝑥𝑥 you are getting two different values. Then,
your function is not well behaved, because I do not know which value will come when I
fed in 𝑥𝑥1 . So, in that case how will you identify whether something is a function or not?
Sometimes, if you have a graph of a function the way it is given; it is very easy to see;
what is not a function.

For example, if you take a line which is 𝑥𝑥 = constant. If you take your line which is 𝑥𝑥 =
constant and if I draw that line vertically. For example, let us say here this is the line 𝑥𝑥 =
constant; 𝑥𝑥 = 𝑥𝑥1 in fact I have drawn. So, if I draw this line vertically, then I can see that
there are two points at which, this line intersects the graph of a function.

When such a thing happens, we say a vertical line test that is, this is the vertical line right
it is a parallel to 𝑦𝑦 −axis. Therefore, this line passes through two points; that means, there
is something fishy about the function and we will use this as a vertical line test. And say
that, because vertical line test fails, this particular function is not a function sorry, vertical
line test fails and hence, this cannot be a function.
Can you imagine such a function where can you imagine such a relation where this is not
a function. For example, ok. Let us not imagine right now. Let us come to the next case
and generalize it to the other set up. So, now let us see here. So, first thing is not a function
that is very clear. Now, let us take the second case which is more than one 𝑥𝑥 and one 𝑓𝑓(𝑥𝑥),
ok.

(Refer Slide Time: 07:41)

So, let us understand it on a paper. So, they are more than one 𝑥𝑥, 𝑥𝑥1 and 𝑥𝑥2 , both are not
equal, but somehow, they give the output which is equal, ok. In such case, what happens?
So, as usual general this is domain set A, this is a co-domain B; 𝑓𝑓 is processing unit and I
have processed I have given I have fed in two different values of exercise, but the output
produced is same.
(Refer Slide Time: 08:27)

So, is this a function or not? The answer is this is a function. And, in this case let us see,
how it happens? So, let us have some understanding about 𝑥𝑥1 and 𝑥𝑥2 . So, naturally 𝑥𝑥1 ≠
𝑥𝑥2 and I got 𝑓𝑓(𝑥𝑥1 ) which is this and I got 𝑓𝑓(𝑥𝑥2 )which is this, ok. Both are same.

And, then how will the function look like? I can join a curve like this, where the function
actually passes through these two points and I have a curve like this. So, if this is the curve
then these two points are same. And, do you call this as a function? Yes, we call this as a
function. Based on our understanding of the function lecture we call this as a function. In
this case, something interesting has happened. Let me analyze it in a more thorough
manner.

For example, when I considered the first case let me go to the first case. I have drawn a
vertical line and I said that, because of this vertical line I can say this is not a function; I
can say this is not a function. Now, the similar graph has appeared over here, but now if I
draw a horizontal line I have a horizontal line, which passes through two points and I am
saying it is a function, correct?

So, if I rotate this graph by say 90 degrees and flip it over then, what I am getting is a
graph similar to this function. So, this actually helps me in understanding that, if I want to
write 𝑦𝑦 as a function of 𝑥𝑥, I am able to write it. But, if I want to write 𝑥𝑥 as a function of 𝑦𝑦
that is simply just flip this by rotate this by 90 degrees and flip the 𝑦𝑦 axis. That will give
you the exact understanding of the picture. And, from this to this I cannot go, ok. And
therefore, the horizontal line if I draw it passes through more than one points.

So, what will happen is; suppose, if I take; if I take 𝑥𝑥1 here, if I take a point in this domain
𝑦𝑦1 . And, if I try to setup another processing unit let us say 𝑔𝑔, and if I feed that 𝑦𝑦1 into g,
what I get if I use this 𝑓𝑓, I will not get something which is similar that is I will get
something called 𝑔𝑔(𝑦𝑦1 ) once. And, if I feed in again 𝑦𝑦1 , I will get something else as 𝑔𝑔(𝑦𝑦1 ).
That is what is happening.

For example, here if you locate this point, this point it is say 𝑦𝑦. Then, once you feed 𝑦𝑦 into
𝑔𝑔 then, 𝑔𝑔(𝑦𝑦) you will get as 𝑥𝑥1 and if you feed it other time 𝑔𝑔(𝑦𝑦), you will get as 𝑥𝑥2 .
Something interesting happens. What I am trying to do is, I am trying to reverse the
function and I an easily say that this function is not reversible.

If this is a interesting point which will help us in gaining more understanding of

exponential functions. So, if this function is not reversible and a horizontal line test
actually detects whether a function is reversible or not; so, this horizontal line test fails;
horizontal line test actually fails in this case.

So, what happens here? Here, when you applied vertical line test that is case 1, it was not
at all a function. Here it is a function, but our conclusion is it is not reversible. Let us look
at this 3rd case, where only one 𝑥𝑥 is there and one 𝑓𝑓(𝑥𝑥) is there. So, here is 𝑥𝑥, 𝑓𝑓(𝑥𝑥).

(Refer Slide Time: 13:42)

So, if I substitute 𝑥𝑥1 and 𝑥𝑥2 as two different values, then I will get through this processing
unit, I will get 𝑓𝑓(𝑥𝑥1 )and I will get 𝑓𝑓(𝑥𝑥2 ) and both of them will not be equal to each other
for 𝑥𝑥1 ≠ 𝑥𝑥2 . That is the only way it can happen right; one 𝑥𝑥, one 𝑥𝑥 to one 𝑓𝑓(𝑥𝑥). So, if I
take different 𝑓𝑓(𝑥𝑥) I will get different values, ok.

(Refer Slide Time: 14:12)

So, what will be a typical behavior of such functions? Let us try to figure out. Let us say
this is the; this is one function. So, is this function one-to-one? The answer is yes, because
if I take 𝑥𝑥1 𝑥𝑥2 here this is 𝑓𝑓(𝑥𝑥1 ), this is 𝑓𝑓(𝑥𝑥2 ). So, for every one, if 𝑥𝑥1 ≠ 𝑥𝑥2 , 𝑓𝑓(𝑥𝑥1 ) is not
going to be equal to 𝑓𝑓(𝑥𝑥2 ). Such function is called one-to-one function. And is it a
function? It is a properly defined function, yes.

So, now, I can summarize using this as, this is a function. In earlier case, we actually
characterized whether it is reversible or not. What can you say about this new function that
you have defined? This function; so, now because for 𝑥𝑥1 ≠ 𝑥𝑥2 , 𝑓𝑓(𝑥𝑥1 ) ≠ 𝑓𝑓(𝑥𝑥2 ). So, if I
start this as 𝑦𝑦1 and if I start with this as 𝑦𝑦2 , I can easily retrace back 𝑥𝑥1 𝑥𝑥2 that is 𝑔𝑔 there
will be some function 𝑔𝑔 such that, 𝑥𝑥1 = 𝑔𝑔(𝑦𝑦1 ) and 𝑥𝑥2 = 𝑔𝑔(𝑦𝑦2 ).

So, this function in some sense is actually reversible. We will come to this point in more
detail in the next section. Right now, remember that the function that is one-to-one is
reversible and it is reversible. We need to be more precise and I will give you a word for
reversible function in the later lectures. But right now, it is an important observation that
there are three cases. These three cases actually deal with a function.
The first one is not a function; first one is not a function, why? Because we had dealing
with the coordinate plane. The vertical line test fails when then you pass a vertical line that
vertical line will pass through two points or more than two points, then it is not a function.

So, then we said the second case. The second case was more than one 𝑥𝑥 and one 𝑓𝑓(𝑥𝑥). In
this case, what happened is we were able to find a function properly, but we were not able
to revert the function. So, we said the function is not reversible. And, on what basis we
have said? We have said on the basis of horizontal test, ok. And, we have related this with
our first case that is, if the function is not reversible then the other part that is g is not also
is not a function as well.

Third case where we have one 𝑥𝑥 and one 𝑓𝑓(𝑥𝑥) we showed that it is a function and it is
reversible also. This brings us to a question that, one-to-one functions how easy are they
to identify; how easy are they to identify? So, let us properly define a one-to-one function.
One in the domain, one in the co-domain and there is a clear cut association that is given
by two.

(Refer Slide Time: 18:23)

So, our definition; give a function 𝑓𝑓 is said to be one-to-one if, for every 𝑥𝑥1 ≠ 𝑥𝑥2 which
belong to the domain of a function, 𝑓𝑓(𝑥𝑥1 ) ≠ 𝑓𝑓(𝑥𝑥2 ), ok. The other interpretation is, if
𝑓𝑓(𝑥𝑥1 ) = 𝑓𝑓(𝑥𝑥2 ), this should imply 𝑥𝑥1 = 𝑥𝑥2 this is the other interpretation of definition, but
we will use this as a definition. But, sometimes it may be difficult to prove this thing in
that case, you can prove the one written in the orange.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 8.2
One-to-one Function: Examples & Theorems

(Refer Slide Time: 00:14)

So, let discuss some examples of a functions that are one to one and not one to one. So,
for this let us first take 𝑓(𝑥) = |𝑥|; is this function one to one?
(Refer Slide Time: 00:20)

Try let us try to answer this question. So, let me write this function properly. So, if 𝑓(𝑥) =
𝑥 for 𝑥 ≥ 0 and −𝑥 for 𝑥 < 0. So, it is actually a straight line on a passing through the
origin like this 40 at a 45 degree angle and the −𝑥 is this line ok. So, it is a V shape 90
degrees V; so, is this function one to one? First of all let us let us not take the argument,
first of all is this a function (𝑥) = |𝑥| ?

Pass a vertical line, take a vertical line and pass it through this; is there at if there is any
point where two points more than one points pass through this function pass through that
line then it is not a function. So, vertical line test is successful therefore, it is a function.
Vertical line test says that it is a function succeeds and we know it is a function ok. Now,
the question is the function one to one? Right. So, you pass a horizontal line. So, let me
pass one horizontal line somewhere, let us take this horizontal line. Now, is the function
one to one?

For 𝑥1 ≠ 𝑥2 , I got the same 𝑓(𝑥) ok. So, how will I prove it is not one to one? Let us take
a value which is say 2 and −2; these are the two values, 𝑓(2) = 2 = 𝑓(−2). Therefore,
this is not going to be a one to one function. So, it is not one to one function. So, our
conclusion is it is not one to one function. Then do we know functions that are one to one?
(Refer Slide Time: 03:05)

So, since we have taken 𝑓(𝑥) = |𝑥|, let us take a function 𝑓(𝑥) = 𝑥; is this function one
to one? It is a straight line passing through the origin, is this function one to one?

Let us take horizontal, first let us check whether this is a function, take a horizontal line,
pass it through this pass it horizontally, the line parallel to 𝑥 −axis. So, just drag 𝑥 −axis
up and down; do you see any point touching more than one point? No. So, it is a valid
function; then, sorry yeah you have to pass the vertical line first ok.

Start with 𝑓(𝑥) = 𝑥, take a vertical line which is 𝑦 −axis, slide it to the left, slide it to the
right. Do you see any where it has more than one points? No. So, it is a valid function.
Then take a horizontal line, pass it from the top to bottom; see if you are getting any two
points together for on that line; no. Therefore, this function is actually one to one because
𝑥1 ≠ 𝑥2 implies 𝑓(𝑥1 ) ≠ 𝑓(𝑥2 ) which is more or less expected right.

Because 𝑓(𝑥) = 𝑥 therefore, 𝑥1 ≠ 𝑥2 will give 𝑓(𝑥1 ) ≠ 𝑓(𝑥2 ). So, what about it is an
exercise then what about if you take a cubic functions? So, cubic function will pass like
this sorry, it is not a correct diagram of a cubic function. So, cubic function let us change
the color as well, cubic function will have something like this, symmetry will be retained
and then this will go down.

So, if this function, now check whether this function is one to one or not. Again the
exercise is very similar, pass a let the 𝑥 − axis go up and down, see if you are finding any
two points together. So, let us say this function is 𝑓(𝑥) = 𝑥 3 and now you can easily make
out that for 𝑥1 ≠ 𝑥2 , 𝑓(𝑥1 ) ≠ 𝑓(𝑥2 ). So, again through horizontal line test, I have detected
that the function is one to one.

(Refer Slide Time: 05:59)

So, let us write this particular test as a theorem. If any horizontal line intersects the graph
of a function in at most one point, then the function is one to one ok. So, then what we will
show here, if you want the proof of this what we will show here is if the function is not
one to one then it will intersect some horizontal line will intersect the graph of a function
in more than one point ok. So, that is very easy to prove. So, I will prove it graphically.
(Refer Slide Time: 06:41)

So, if the function is not one to one, let us say this is 𝑥 −axis, this is 𝑦 −axis. If the function
is not one to one, I can take this point and call this as 𝑥1 and I can take this point as call
this as 𝑥2 . This is how I can make function not one to one and then pass a curve passing
through these two points and pass the horizontal line over here which we have done several
times now by now.

And therefore, 𝑓(𝑥1 ) and 𝑓(𝑥2 ) are same, they both are same. Therefore, the function is
not one to one, that essentially proves the point that if a horizontal line intersects the graph
of a function in at most one point then 𝑓 is one to one good.
(Refer Slide Time: 07:37)

So, we are good to go now. Next thing that we will come is can we identify the class of
functions that are one to one? So, what class of functions can you immediately think are
one to one? For example, we have also seen some functions like if 𝑥1 ≤ 𝑥2 then 𝑓(𝑥1 ) ≤
𝑓(𝑥2 ) or let us not put this strict equality; let us put this way strictly increasing.

So, what does what do I mean by ok; let us can we question is can we identify the class of
functions that are not one to one? So, I can; obviously, think of function of this form 𝑥1 <
𝑥2 , 𝑓(𝑥1 ) < 𝑓(𝑥2 ). Let me plot it and then the my imagination will work fine. So, this
function is something like if 𝑥1 is to the left of 𝑥2 then 𝑓(𝑥1 ) should always be less to the
left of 𝑓(𝑥2 )

Or, if you are plotting it on the 𝑦 −axis then 𝑓(𝑥1 ) below 𝑓(𝑥2 ), this is the intuition and
you can draw line joining these two points. Let it go ahead and this is true for every 𝑥1 , 𝑥2
belonging to A this is true; then I am done.

But, this function have a name that is they are called increasing functions ok. In a similar
manner, if I multiply this function with minus sign. Then I will get a function which is
decreasing function and that can be written as 𝑥1 < 𝑥2 employs 𝑓(𝑥1 ) > 𝑓(𝑥2 ) and this is
called decreasing function.

Now, you look at any increasing function and apply your horizontal line test. What is the
horizontal line test? Just now we have seen that if you take the horizontal line, roll it across
the axis across 𝑦 − axis and there should not be more than one point intersecting that line
at any given point in time ok. So, this increasing function and decreasing function will
satisfy this phenomena.

(Refer Slide Time: 10:47)

And therefore, we can easily write this as through horizontal line test that, if 𝑓 is an
increasing function or a decreasing function then 𝑓 is one to one.

(Refer Slide Time: 10:59)

Let us see one decreasing function as well. What happens when the function is decreasing?
As I go from left to right there is a 𝑥1 is here, 𝑥2 is here. As I go from left to right, I get 𝑥1
here and now according to the condition 𝑓(𝑥1 ) > 𝑓(𝑥2 ). So, it will be somewhere here
and I can have a curve passing through this point in this manner ok.

This is true for every 𝑥1 and 𝑥2 belonging to the domain. And therefore, using our line test,
horizontal line test we can easily see that the function whether it is increasing or
decreasing, they are one to one. This gives us a big class of functions that are one to one.
So, one to one functions are not rare to find, one to one functions are abundant in nature
and they are reversible as well.

With this insight, we will go to our next topic which is exponential functions.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 8.3
Exponential Functions: Definitions

(Refer Slide Time: 00:14)

Welcome back. Next topic is Exponential Function. So, in this topic, what we will see is
first we will identify with some known terminology that is exponent. We have already
know seen this exponent. Where we allowed integer powers and then while defining the
exponents, we allowed rationals also. So, when we defined exponents, they were of the
form 𝑎𝑟 and we always assume 𝑎 > 0 and 𝑟 ∈ ℚ.

Now, I want to define an exponential function. So, as the name suggest, exponential has
to do something with the exponent. So, what we are doing here when I am considering a
function of this form, I am raising something, some number to the power of 𝑎 where a will
be popularly called as base and 𝑟 is the exponent. So, this is base and this is exponent.

Now, if I want to define exponential functions on real line, then it is mandatory for me to
define this 𝑎𝑟 for 𝑟 ∈ ℝ\ℚ. This is real line minus set of rational numbers that is I am
talking about set of irrational numbers. So, I do not know as of now what is a definition of
exponent form of set of irrational numbers ok.
The next question that we have seen is why is 𝑎 > 0 that for which you know the answer.
Let us say 𝑎 = −1 now𝑎1/2 = (−1)1/2 = 𝑖 which belongs to complex set of complex
numbers, but I do not want to deal with complex number so, I am avoiding a to be greater
than 0. In general, you can define 𝑎 to be a negative number and then deal with complex
numbers. We do not want to indulge into that conversation. So, I do not; I do not want to
define 𝑎 < 0.

(Refer Slide Time: 03:12)

So, 𝑎 < 0 is eliminated now the question is 𝑎𝑟 and r belongs to irrational where 𝑟 ∈ ℝ\ℚ
what will happen in this case? Or how will I define rational number?
(Refer Slide Time: 03:24)

To be precise, let us ask a question that is can I define 2√2 or 5𝜋 ok. So, in this case, there
is no direct way to answer this question, but I will definitely have a strategy which is a
calculus-based strategy to answer this question.

(Refer Slide Time: 03:52)

Let us consider this 𝜋 and the value the numerical approximation of 𝜋 is actually we all
know 𝜋 is an irrational number and 3.14592635… and this thing is non repeating it will
continue till infinity right. So, now, what I need to understand is from what I know, can I
define the number 5𝜋 ? So, anyway I cannot define it accurately right now.
So, based on my understanding, I am asking you a question that is 53 ? Right if so, then
next question is 53.1 defined? So, what I am doing here is this if yes, can I define 53.14 ?
Now, you remember all these a approximations are actually a rational approximations 3 is
a rational number, 3.1 is 31 by 10 which is again a rational number, 3.14 is 314 by 100
which is again a rational number and I can go on like this there is 3.141 and so on.

So, if I continue this way, I will reach somewhere; I will reach somewhere and that
somewhere I will call as 5𝜋 . So, in principle, I can actually define 𝑎 raised to irrational
number. This you will study when you will study a topic of sequences which is outside the
scope of this syllabus. So, we will assume that you have to trust me on this that 5𝜋 is well
defined.

In a similar manner, you can do an exercise for 2√2 . So, √2 = 1.41 … and something. So,
again you will go with 21 is defined 21.4 is defined, 21.41 is defined and so on and you will
reach somewhere that is 2√2.

So, this way we are very clear that 𝑎 𝑥 is defined for 𝑥 ∈ ℝ. This sets up the platform for
defining an exponential function, this is very important a raised to 𝑥 is well defined for
𝑥 ∈ 𝑟.

This answer is given by convergence of sequences which is outside the scope of the
syllabus, but we know that it exist for sure. So, I am guaranteeing the existence of 5𝜋 ;
existence of 5𝜋 is assured. In case you are interested, you can take a basic course in analysis
or; in analysis or calculus where you will study these things ok.
(Refer Slide Time: 07:27)

So, now let us go to let us recall all of these laws you already know simple laws of
exponents. Earlier, we have we knew the laws of exponents for only rational numbers.
Now, we are talking about the real numbers. So, 𝑠, 𝑡 ∈ ℝ, 𝑎, 𝑏 > 0, 𝑠 and 𝑡 will play a
role of exponents, 𝑎 and 𝑏 will play a role of bases ok. So, then it is very easy to prove you
might have proved. 𝑎 𝑠 𝑎𝑡 = 𝑎 𝑠+𝑡 .

Remember here, product here is becoming addition here these are crucial points (𝑎 𝑠 )𝑡 =
𝑎 𝑠𝑡 . So, 𝑎 raised to operation is becoming a product here. (𝑎𝑏) 𝑠 = 𝑎 𝑠 𝑏 𝑠 and then;
1 1 𝑠
obviously, you need to know that 1𝑠 = 1 for every 𝑠 ∈ ℝ, 𝑎−𝑠 = 𝑎𝑠 = (𝑎) , 𝑎0 = 1.

Remember where your 𝑎 > 0 because 00 is undefined ok.

(Refer Slide Time: 09:18)

So, with this understanding, we have revised laws of exponents which will which we will
use the left and right. So, you better remember all these laws and therefore, we are ready
to set a framework of exponential function. So, here is our definition. An exponential
function in the standard form is given by 𝑓(𝑥) = 𝑎 𝑥 , where 𝑎 > 0, 𝑎 ≠ 1. These are new
condition that we have introduced.

We have seen why 𝑎 > 0, but here they are saying 𝑎 ≠ 1. So, this needs further analysis,
we will analyze it in due course. So, right now, if you look at the values of 𝑎, 𝑎 > 0; that
means, all these values are allowed and 𝑎 > 1; that means, all these values are also
allowed. Bearing the values 0 and 1 right.

So, the first from the definition, the first observation that you can figure out is because you
have bared the value 0 and 1, the function 𝑓(𝑥) = 𝑎 𝑥 will have a domain which is entire
real line. For every , 𝑥 ∈ ℝ, we should be able to compute 𝑎 𝑥 ok.

Then, let us analyze this is then observation: why 𝑎 ≠ 1? Let us put 𝑎 = 1. So, 𝑓(𝑥) = 1𝑥 ,
but from the laws of exponent what you know? 1𝑠 = 1.

Therefore, 1𝑥 = 1 in fact; it is nothing, but a constant function. I am not interested in

handling a constant function right which nothing, but a horizontal line 𝑦 = 1 is the graph
of a function; I am not interested in this. So, let us not call this as exponential function that
is what we are saying in the definition.
So, hence forth, we will never talk about 𝑎 = 1, 𝑎 = 0 or 𝑎 < 0. So, if you have a real
line, you will have an expression of this form where you are talking about this interval,
open interval and this interval, which is an infinite interval. So, you have two
characterizations which is 0 < 𝑎 < 1 , 𝑎 > 1 these are the two characterizations that you
got over this thing.

(Refer Slide Time: 12:18)

Now it will be interesting to use some graphical tool and see what are some functions of
this kind look like. So, here is an exercise that I will give you. Use some graphing tool like
Desmos and plot these functions together. For example, you plot the functions given in 1
using Desmos we just put 𝑓(𝑥) is equal to this, 𝑓(𝑥) is equal to this and this and plot all
these three graphs together without any understanding about the behavior of the function
you plot all three of them together.

Then, use the 2nd graph and put all these three things together. Identify the properties of
the graph that is through which points they pass through is there any difference in the
graphs of 1 and 2. So, identify all these properties like we did in polynomials and after
doing that again return back to this video and we will see some of the functions that are
given here by a graph and we will analyze those functions. So, right now you pause the
video, come back in the next video.
 Mathematics for Data Science 1
Prof. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras

Lecture – 8.4
Exponential Functions: Graphing

(Refer Slide Time: 00:15)

Welcome back. So, I hope you must have done your exercises and you must have
developed some understanding about the exponential functions. Let us try to collect
recollect that understanding through 2 examples given here.
(Refer Slide Time: 00:31)

So, let us first take 1 a which is 𝑓𝑓(𝑥𝑥) = 2𝑥𝑥 . If you have used DESMOS, you must have
got the figure of the function. But prior to receiving the figure of the function, let us see
what should be the domain of a function.

We have already discussed in greater detail that the domain of a function can be a ℝ, entire
real line. Now, if you look at this function which is 2𝑥𝑥 , this 2 > 1 and the 2𝑥𝑥 > 20 which
is equal to 1, 2𝑥𝑥 > 20 whenever x is positive correct.

Now, because 𝑥𝑥 > 0, then 2𝑥𝑥 > 20 ok. So, if 𝑥𝑥 < 0, what will happen? 2𝑥𝑥 , when 𝑥𝑥 < 0
will always be less than 1. This is also possible. But when this 2 raised to; can this 2𝑥𝑥
become negative? No. So, it is always greater than 0.

So, if you have this understanding, then you can easily write the function has a range which
is (0, ∞). So, there is a split from when you consider a point 1, there is something
happening at point (0, 1) right. What is (0, 1)? (0, 1) actually is an 𝑦𝑦 −intercept ok,
something is happening at (0, 1) because I have put 0 here for then it is I am getting 1.

So, (0, 1) is also 𝑦𝑦 −intercept and there is something happening which is going below 0.
Is going below 1, your graph is going below 1 and therefore, this particular thing is going
down, but it never goes below 0. This is an interesting fact because if you consider 2𝑥𝑥 , it
never goes below 0.
It cannot go to a negative number. Therefore, will it touch the 𝑋𝑋 − axis? It will not touch
𝑋𝑋 − axis. In fact, 𝑥𝑥 − intercept is nil ok, but it is approaching 0. So, the something that is
approaching 0, so 𝑥𝑥 −intercept is actually it will never touch it; but it will actually go along
that line. So, this 𝑦𝑦 = 0, it will touch at infinity ok. So, such a thing, we call as horizontal
asymptote ok.

So, such a thing you call as horizontal asymptote. So, with this understanding, these are
the things that I can make out directly without looking at the graph. So, let us now look at
the graph ok, before going to that, let us see what happens to the end behavior. End
behavior of a function as 𝑥𝑥 → ∞.

So, as 2𝑥𝑥 , you consider a function 2𝑥𝑥 as 𝑥𝑥 increases, this also increases. In fact it increases
at a rapid rate than 𝑥𝑥. So, this also should tend to infinity and as 𝑥𝑥 → −∞, we have already
figured out 𝑦𝑦 = 0 is the horizontal asymptote. So, 2𝑥𝑥 will actually go to 0 ok.

(Refer Slide Time: 05:29)

Then, the question that we used to quantify while considering the function, what are the
roots of this function. So, do they have any roots? In fact, using graphical method, it is
very clear that it never touches 0. So, there are no roots and the functions increase and
decrease.

So, the domains of increase and decrease like polynomials, we studied domains of increase
and decrease; but here, I think my claim is no need to identify the domains of increase and
decrease. Why? Because you look at a function 2𝑥𝑥 , let us take 𝑥𝑥1 ≠ 𝑥𝑥2 or 𝑥𝑥1 < 𝑥𝑥2 , without
loss of generality, we can take this. Then, what can you say about 2𝑥𝑥1 and 2𝑥𝑥2 ?

See 𝑥𝑥1 < 𝑥𝑥2 , so naturally if it is raised to the power 2; 2𝑥𝑥1 and 2𝑥𝑥2 , this relation should
hold. So, what I am saying is the function is actually an increasing function and increasing
functions are 1 to 1. Therefore, I do not have any doubt that the increase and decrease, it
is only increasing; throughout the real line, the function is only increasing.

(Refer Slide Time: 07:19)

So, let us look at the graph of a function 𝑓𝑓(𝑥𝑥) = 2𝑥𝑥 . Let us identify the points. So, here
you can identify a point right. So, this point we have seen as 𝑦𝑦 − intercept and that point
was (0, 1) right. Then, the one in this case, let us look at this point which is 1 and where
will it go? It will actually tell you 2.

So, the point is (1, 2), the second point ok. So, these 2 points are very special points, they
tell you something. So, in particular, had it not been 2𝑥𝑥 , but 𝑎𝑎 𝑥𝑥 , then that point would have
been (1, 𝑎𝑎) and if you mimic this graph over here 𝑦𝑦 𝑥𝑥 is over here ok, this is a point 1, this
is the point 0 and this is the point which is 𝑎𝑎.

So, that says 𝑎𝑎 > 1; this relation is there, is greater than 0 yeah and therefore, the graph
was a point which lies here, which is here right. As 𝑥𝑥 → ∞, this graph actually goes to
infinity; as 𝑥𝑥 → −∞, this graph goes to 0. These two points is these two point and this is
an increasing function.
As you come from left to right, it increases. So, this is an increasing function, 𝑦𝑦 = 0 is the
horizontal asymptote, that is very clear ok. The range of a function is 0 to infinity, that is
also very clear. The domain of a function is entire real line, ℝ.

So, we have got all the details necessary for finding this. Now, what it so special about 2𝑥𝑥 ,
if I replace this 2 with 3, still I will have 𝑦𝑦 −intercept to be 0, 1 because 30 is also 1 and I
will again have domain of 𝑓𝑓 to be equal to ℝ; range of 𝑓𝑓 to be equal to 0 infinity; no 𝑥𝑥 −
intercept; 𝑦𝑦 = 0 will be horizontal asymptote; 𝑥𝑥 → ∞, 𝑎𝑎 𝑥𝑥 → ∞, 𝑥𝑥 → −∞, 𝑎𝑎 𝑥𝑥 → 0. There
are no roots. The function is only increasing.

(Refer Slide Time: 10:19)

And therefore, I will state this as a fact that every 𝑓𝑓(𝑥𝑥) = 𝑎𝑎 𝑥𝑥 , for 𝑎𝑎 > 1 will have same
properties as 2𝑥𝑥 . So, I do not there is no need to draw different different values. The
behavior is same only the values will change.

For example, in this case, where you have seen the graph of this (1, 2) is a point; (1, 2) is
a point, suppose I consider 3𝑥𝑥 , (1, 3) will be the point. So, only the values are changing;
but the shape, the behavior, everything else that is listed here remains the same. Therefore,
you do not have to draw a graph every time, only thing is you need to evaluate the values
in general.
(Refer Slide Time: 11:15)

So, what is the graph of 𝑓𝑓(𝑥𝑥) = 𝑎𝑎 𝑥𝑥 in general? It is this way for 𝑎𝑎 > 1. So, remember that
line that we have drawn which is that the line for 𝑎𝑎, where we have eliminated these 2
points such as 0, this is 1, we have identified what is the case for 𝑎𝑎 > 1. You have also
identified the case, where 0 < 𝑎𝑎 < 1. So, let us go back and see what happens when 0 <
𝑎𝑎 < 1. So, if 𝑎𝑎 lies here how is the behavior? So, you have already analyzed.

(Refer Slide Time: 11:59)

 1 𝑥𝑥
And let us take this function as 𝑔𝑔(𝑥𝑥) and take it to be 𝑔𝑔(𝑥𝑥) = 𝑎𝑎 𝑥𝑥 and this is �5� . Now,

you do not really have to draw this graph, what you can do is ok. So, 𝑔𝑔(𝑥𝑥) = 5−𝑥𝑥 . So,
here 𝑥𝑥 is replaced by −𝑥𝑥. So, what will be the change in the behavior?

So, when 𝑥𝑥 is replaced by −𝑥𝑥, you know its reflection across 𝑌𝑌 − axis, you have solved
many examples in the assignments. This 𝑌𝑌 − axis, this is 𝑥𝑥; then when I put it as −𝑥𝑥, it
will be simply reflected along 𝑌𝑌 − axis.

(Refer Slide Time: 12:53)

So, if you look at this graph and try to draw a graph of this function, then it should be
something like coming from here going here, it should be something like this, it should
actually look like a reflection along 𝑌𝑌 − axis. So, let us try to show it as reflection ok. This
will actually go very close, but never touch.

So, let me erase this ok. So, this is how it will look like. So, without actually thinking about
1 𝑥𝑥
anything else, you can simply draw a graph of �5� ; but still let us try to do it in regular

set up.
(Refer Slide Time: 13:43)

So, what will be the domain of this function? The domain of this function is very clear
because we have used it several times, the domain of this function will be real line. Range,
nothing changes; (0, ∞) because it is a reflection across 𝑌𝑌 − axis. So, let us look at this
function.

So, the domain will be ℝ; range will be (0, ∞). What will be the 𝑦𝑦 − intercept? Because it
is a reflection, so 𝑦𝑦 − intercept would not change, so it will be 0, 1 only. 𝑥𝑥 − intercept will
be nil, there would not be any 𝑥𝑥 − intercept.

(Refer Slide Time: 14:25)

And therefore, no roots and what about the end behavior? End behavior is like 𝑥𝑥 → ∞,
𝑥𝑥 → −∞. So, when 𝑥𝑥 → ∞, the end behavior will be because it is a reflection you see.

So, when 𝑥𝑥 → ∞ there, it was going to ∞. So, and 𝑥𝑥 → −∞, function 5𝑥𝑥 would have
1 𝑥𝑥
behaved, it will go to 0. So, that reflection will make this 𝑎𝑎 𝑥𝑥 or �5� whatever is the
1 𝑥𝑥
function �5� , let me do it properly.

1 𝑥𝑥 1 𝑥𝑥
So, this will make �5� to go to 0 and this function �5� will go to infinity ok. Good. Then,

because it is a reflection, the increasing thing will become decreasing. So, there is no
intelligence here. So, this will be in fact a decreasing function wonderful. So, we have
analyzed everything without taking much efforts. This is the beauty of once you
understand the functions on graphical plane.

(Refer Slide Time: 16:03)

1 𝑥𝑥
So, here is the graph of a function which is given to us �5� , you also might have plotted

and naturally, the we will analyze whether it coincides with our thing. So, this is a point
1
(0, 1), now it is 5. So, your point will be somewhere here, sorry this is 5. So, the point 1 is
1
here and this point is 5.
 1
So, �1, 5�, this is done. Then, as 𝑥𝑥 → ∞ , this function goes to 0. As 𝑥𝑥 → −∞ that is this

way, this function actually goes to ∞ and this function is decreasing. From left to right if
you come, you are actually coming down. So, it is a decreasing function. So, this
completely gives us an understanding of what the graph of a function will look like.

(Refer Slide Time: 17:17)

Also, the same fact is true that every 𝑓𝑓(𝑥𝑥) = 𝑎𝑎 𝑥𝑥 , where 0 < 𝑎𝑎 < 1 has same properties as
1 𝑥𝑥
�5� . Therefore, it is a representative class. So, you do not have to worry about the because

it is a representative class, you have to worry about all other functions. All other functions
will have a similar behavior.
(Refer Slide Time: 17:47)

So, we have done a lot, let us summarize these things in a neat table which is this. So, this
is the summary of the table. So, if I have been given a function 𝑓𝑓(𝑥𝑥) = 𝑎𝑎 𝑥𝑥 , then to be more
precise, let me draw a line here. This is a line; it does not look like a line, but assume that
this is a line.

This is the point 1, then I am talking about 0 < 𝑎𝑎 < 1 that this zone. In this zone, the
domain of a function is ℝ; range of a function is (0, ∞) . There are no 𝑥𝑥 − intercepts, no;
𝑦𝑦 −intercept is 0, 1. Horizontal asymptote 𝑦𝑦 = 0 is there. The function is decreasing. The
end behavior as 𝑥𝑥 → ∞ , 𝑓𝑓(𝑥𝑥) → 0; as 𝑥𝑥 → −∞, 𝑓𝑓(𝑥𝑥) → ∞ correct.

Then, you look at the function which is 𝑎𝑎 > 1, domain is real line, range is (0, ∞), nil; (0,
1), 𝑦𝑦 − intercept is (0, 1). Horizontal asymptote is 𝑦𝑦 = 0. The only distinguishing feature
is the function is increasing here and a function is decreasing here and because it is
increasing and decreasing, the end behavior changes that is because it is decreasing, it will
decrease to 0 because it is bounded below by 0 and because this is increasing, it will
increase to infinity, but here it will go to 0 ok.
(Refer Slide Time: 19:21)

Then finally, you see the prototypes, just look at the graphs of these two functions ok. This
ends our topic on exponential functions. Now, we will introduce something which is called
natural exponential function in the next video.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 47
Natural Exponential Function

(Refer Slide Time: 0:14)

Hello friends, in this video we are going to talk about yet another important function among all
exponential functions is natural exponential function. So, basically the theory of natural
exponential function is derived from calculus. So, in order to understand how relation natural
functional, natural exponential function arises, we need to study the theory of limits. In
particular, this natural exponential function is dependent on something raised to the exponent
that something is in irrational number that is called 𝑒.

And I will make sure by the end of this video you will understand why this number 𝑒 is very
important. In particular, when we talk about number 𝑒 or ratio or a limit of some quantity is
important which is shown here. So, from the theory of limits it is known that whenever you are
1
talking about (1 + 𝑛)𝑛 this particular limit it actually converges to e.

So, now unless you understand the concepts of limits, you may not be able to have complete
understanding of this concept, but still I will give you some intuition behind this number e. So,
though, as I mentioned earlier, the existence of 𝑒 is actually studied in the field of calculus. For
that you may have to do the course which is maths 2, Math for Data Science 2 and you have to
agree with me on certain facts without knowing them or you have to trust me that 𝑒 is an
irrational number and 𝑒 is approximately equal to 2.71828 and so on. It is an irrational number.
So, it will go to never ending decimal representation, it will continue on the right.

(Refer Slide Time: 2:26)

So, these are the facts about e. Now the question that we asked at the beginning of the video is
why is ‘e’ so important? So, to answer that question, let us first look at the behaviour of this
particular number as a limit. So, when I say that n goes without bounds, the number the this
1
particular function 𝑓(𝑛) = (1 + 𝑛)𝑛 converges to e. What do I mean? Let me put it in a proper

formal way.

So, I have generated a table over here for our convenience and let us understand. So, when I
1
substitute this n, the value of 𝑛 = 1, this number (1 + 𝑛)𝑛 is simply 2. When I substitute 𝑛 =

10, the number becomes 2.5937. So, does that mean this function will go without bounds? The
answer is no that is why we get the convergence. And such type of questions are studied in
Calculus.

So, when you substitute the further values of n that is, n=100 you have substituted, you got
2.7048. When you substitute 𝑛 = 1000, you will get 2.7169. Now, you can see that you
approaching closer to the ideal value of e. And when you put n =10000, you get 7181 still
because we are writing up to 4 decimal places, we are not really very close to it, but we will
be, we are very close to it, but we are not at that point.

But when I put n is equal to 1 lakh, then I get a value of 𝑒 which is 2.7182 and that is actually
exact representation of this number 𝑒 up to 4 decimal places, correct up to 4 decimal places.
Now, if you go on further and put higher and higher values 𝑒 like you can put it to be a 1
million, 10 lakhs and then you will see you will get further improvement, but because we are
focusing only on 4 digits after decimal, we this requirement is enough for us, this much
calculation is enough for us. Another thing, another aspect in which this 𝑒 becomes very crucial
in accounts this interested calculations.

(Refer Slide Time: 4:57)

You must have heard the term by now of continuous compounding and continuous
compounding actually means the taking ratios with respect to 𝑒 or taking exponents with
respect to e. So, let me demonstrate to you in this manner. Let us say you have invested rupee
1 in a bank and bank is offering 1 % interest rate and you have invested it for 1 year. So, in that
case what is the answer? 1 plus 1 upon 100 raised to 1, this is the answer, 1.01 1 % you will
get if you have invested rupee 1, you will get 1 paisa of interest.

1
So, now if you go on like this you will have something like (1 + 100)1 raised to so whatever

number of years you have invested in raised to n. Now, when you actually look at the procedure
of the bank, banks do not give you the interest which are given annually, but they credit the
interest quarterly. So, in that case what you need to understand is the interest rate is actually
given in a quarter, so it is computed on quarter and whatever interest you have accumulated,
that interest will be taken into account for the next quarter.

So in that case, basically what bank is doing is bank is taking this interest rates which is 0.01
and it is actually dividing it into 4 quarters, 4 parts because they are giving you a quarterly
interest and then you are actually getting this multiplied in this fashion. So here, if you look at
the interest rate, instead of 1, I have 0.01 as the number and for 0.01 I got this number. So, if
the bank decides, so I will revise the bank decides that I am revising, bank is revising the
interest rate every month, then what will happen?

0.01 12
Then the same logic for single year, for single year remember (1 + ) . So, if the bank starts
12

revising the interest rate infinitely often, then we are actually talking about something like
0.01 𝑛
(1 + ) and this 𝑛 → ∞. In this case, according to our judgement, according to this, this
𝑛

number was 1 here, now it is 0.01 and this number converge to e. So, based on this
understanding, if you apply the same logic and try to calculate this thing, then it will converge
to 𝑒 raised to 0.1.

This is an interesting revelation. That means, if you invest rupee 1, you just take that rupee 1 ∗
𝑒 0.01 that will be the interest accumulated along with the original capital in are bank if the bank
follows continuous compounding. This is how whenever you study finance, you calculate the
interest rate. So, this is for the period of 1 year. Now if you add the period in terms of time,
then it will be 𝑒 0.01𝑡 . So, this is how 𝑒 becomes important. Let us now replace this 1 % by a
generic number which is 𝑥.
𝑥
So, what I am talking about now is (1 + )𝑛 and now from the discussion that we have done
𝑛
𝑥
this will converge to 𝑒 and when I add the time that is it is more than 1 year, then I have
something like 𝑛𝑡 and that is where I will get 𝑥𝑡 . So, these are simple understanding why the
number 𝑒 is very important. 𝑒 typically comes when you are considering a continuous
compounding.

I hope I have made the relevance of the number e, irrational number 𝑒 very clear and it is an
irrational number and its exact value is given by this particular expression. It is not exact but it
is approximate which is suitable for our purposes.
(Refer Slide Time: 9:53)

Now let us go further and understand what is a function that we have defined here and it is why
it is called natural exponential function? So, this number 𝑒 as I mentioned now naturally comes
when you are considering continuous compounding. It also comes very naturally in the field of
Differential Equations which is also relaying on our calculus. So, this number 𝑒 has a special
name when you consider differential equations as a area which is called Euler’s Number.

So, you can Google and you can search the meaning of Euler’s number and why it is relevant.
So, that is how this 𝑒 is called a natural exponential. So, now let us formally define the function
that we have just now seen which is 𝑒 raised to 𝑥𝑡 as a natural exponential function.

(Refer Slide Time: 10:50)

So, a natural exponential function is defined as 𝑓(𝑥) = 𝑒 𝑥 . Then, you may ask a question, what
are some interesting properties of this natural exponential function? Now, the properties will
be very similar to the exponential function that we have studied, but it is special in some sense.
We will see its specialness in a when we will study some special properties of this natural
exponential function 𝑒 𝑥 . Let us list all the properties.

Domain of f, domain of this function will be set of real numbers and range of the function is
positive real line that is 0 to ∞. As you have seen e, the value of 𝑒 is 2.7182 so 𝑒 is natural
greater than 1.

(Refer Slide Time: 11:51)

So, if you recollect whatever we studied for exponential functions, you will get the graph of
exponential function in this manner where there are two typical points 0,1 is one typical point
that it passes through and it will always pass through 1,e. As x tends to infinity, the function
goes without bounds, as 𝑥 → −∞, the function asymptotically goes to the x axis so y = 0 is the
horizontal asymptote for this function, we have already seen that. For general exponential
function same properties hold true.

Now what makes 𝑒 special? And what is something special that is not true with general
exponential function. So, in this case, if you look at the point (1,𝑒) and if you draw a tangent
to a line tangent to the curve, that is a line passing through this particular point, the slope of
this line will be e, that is very special. So, 𝑒 is the slope of the line that is tangent to the curve
y is equal to 𝑒 raised to x at (1,e). So, that is one thing.
Then if you look at the area that is covered under this curve from −∞ to 1, that area is actually
e, the irrational number e. This you will learn when you will study calculus in maths 2. So, that
is very important.

(Refer Slide Time: 13:38)

And the third thing that is very important which will not happen in general with other
1
exponential function is if you draw a curve f(x) is e, if you draw a function f( x) = 𝑥. So, you

may be familiar with that function, it will be something like this and something like this. And
in this particular case, if you look at the area under the curve in the in the range 1 to e, this
particular area, this area is a unit area for f( x) =x and remember this 𝑒 is an irrational number
so still it will be a unit area.

Why it is so? this is a matter of calculus to explore, but these are the things, these are some of
the things that makes the function f( x) =𝑒 𝑥 special function. Let us now understand this
function better by considering an example which will deal with our real life problems.
(Refer Slide Time: 14:43)

So, here is an example which says that let R be the percentage of people who respond to affiliate
links under YouTube descriptions and the purchase and they purchase the product in t minutes
and that particular purchasing thing is a function of time so it is given as the 𝑅𝑡 = 50 −
100𝑒 −0.2𝑡 . So, let me give you a brief understanding of the problem.

So, now when you watch some video on YouTube if you, the speaker in the YouTube says that
there are some affiliate links below in the description. Now, if you click on that link and go to
the affiliate site, then what you will do is, either you will purchase or you will not purchase. If
you will purchase, the speaker or the channel owner will get some amount of commission.

Now, here the person who is actually giving the affiliate links is interested in finding the
number of people who are responding in t minutes. So, he has devised a function which is
available in YouTube statistics so based on the data available, he has derived a function, we
are taking the function as it is. So, that function is 𝑅𝑡 = 50 − 100𝑒 −0.2𝑡 .

Now, he is interested in answering these questions. What percentage of people responding after
10 minutes? So, how many percentage of people responded after 10 minutes? Then, based on
this function, what is the highest percentage expected? And the third question is how long
before 𝑅𝑡 > 30 %? The response rate being 30 % is also a good enough rate.

(Refer Slide Time: 16:54)

So, because you are just putting some affiliate links. So, let us try to understand what
percentage or people will respond after 10 minutes? That means, I want to essentially evaluate
the function as 𝑅10 . So, if I substitute this, it will be 50 − 100𝑒 −0.2×10. that is simply if you
rewrite this as 50 -100 times, this is 2/ 10 which can be simplified to 𝑒 −2 . And then you can
actually calculate the function 𝑒 to by value of 𝑒 −2 and you can put the value, that value is
36.46. So, this you can do it using calculator.

Now, let us look at the second question. What is the highest percentage expected? Now you
have to think about this function which is 50 − 100𝑒 −0.2𝑡 . Now, you look at the function which
is 𝑒 −0.2𝑡 or 𝑒 𝑥 .
(Refer Slide Time: 18:19)

You already know the graph of the function which is 𝑒 raised to x, 𝑓(𝑥) = 𝑒 𝑥 . Now, how will
the function look like when you are talking about 𝑓(𝑥) = 𝑒 −𝑥 has a graph of this form, roughly
this form. Now, when you are talking about − of x, when you are talking about − of x, you are
actually talking a reflection of this graph along this so that will give you some graph of this
kind, it will never cross x axis but it will go this way.

So, now you have a understanding of how the graph of 𝑒 −𝑡 will look like. But here are some
scaling versions, scaled versions like 100, this is 100 and this is 50. So, now this graph is
actually multiplied with − 100, this graph is actually multiplied with −100, but multiplying
with − 100 will again, what it will? It will actually keep the graph in a similar manner but it
will actually because it is multiplied with − 100, it will shift in some sense like this.
(Refer Slide Time: 19:33)

One minute let me chose and erase it. So, it will shift like, it will flip here and it will shift like
this. And then, now when you are adding 50 to it, this graph will actually go up by 50 units.
So, this way the changes will happen to the graph and finally graph will look something like
this. You can actually check for yourself. So, basically first multiplying this − sign will have
an effect of reflecting the graph along y axis, then multiplied with − 100 will reflect the graph
along x axis and then adding 50 will shift the graph by 50 unit.

So, you have a fairly good understanding of the graph. Now, you just apply your knowledge
that what is the highest percentage expected? So, in this case, if you understand this, the
horizontal asymptote over here is actually shifted to 50 units because you are transferring to 50
and the graph actually let me clear up the image.
(Refer Slide Time: 20:49)

You can use your graphing tool also and verify that this is the graph. So the graph will look
somewhat like this and it will asymptotically reach to 50 units. So, the highest percentage that
is expected will be 50 %. It will not exceed 50 % based on the graphical analysis. Let us analyse
this graph, instead of graphically analysing, let us look at this function 𝑒 raised to − 0.2t, this
function in itself will never exceed 1 and as x tends to infinity, this function will actually tend
to 0.

That means, whether I am multiplying by 100 or I am multiplying by 10000, as 𝑡 → ∞, this

function has to go to 0. So, this entire thing will go to 0 and therefore, 50 is the maximum that
I can achieve. Therefore, my question b is answered as 50 %. Now let us look at this particular
thing, how long it takes before 𝑅𝑡 exceeds 30 %?
(Refer Slide Time: 22:06)

So, in this case you just look at the graph of the function, so as I have already described this
will, this is how the graph of the function will look like. Now, there are two ways in which you
can solve. Let us try to see whether we can go ahead formally and solve this. So, essentially I
have been given that how long till 𝑅𝑡 >30 %? That means 𝑅𝑡 = 30, find are value of t such that
𝑅𝑡 = 30?

So, 30 = 50 − 100𝑒 −0.2𝑡 . So, 30 − 50 will give me something like − 20 = − 100𝑒 −0.2𝑡 . Now,
this is, these − signs will cancel themselves off so, this is this and then you simply rewrite this
20 1
expression is nothing but 5 = 𝑒 −0.2𝑡 . Now, I have to stop here because right now I do not
100
1
have any ways to see what t will be when 5 = 𝑒 −0.2𝑡 . No analytical way is possible.

Then, what I will do is, so analytically I am stopping here and if I somehow I am able to figure
out how to find t is equal to something, then I can answer this question. But let us now try to
compute this graphically.
(Refer Slide Time: 23:47)

So, in this case,𝑅𝑡 = 30 is this point. So, now if you go along this line and then you map this
onto x axis, remember x axis is nothing but the value of t. So, in this case, now you look at the
mesh, this roughly turns out to be 8 that means, t will be approximately equal to 8 minutes. So,
this is how we can without even solving the expression for this, graphically solve the expression
for exponential functions. So, this is one live demo of that. That is all for now. So, we will meet
in the next video where we will actually try to understand how to solve this particular problem.
Thank you.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 48
Composite Functions
(Refer Slide Time: 0:16)

Hello students, today we are going to learn the concept of composite functions, what do you mean
by a composite function? So, let me motivate this with an example. For example, it is known that
you are a very good bargainer and your friend wants to buy a computer. So, your friend takes you
to a computer store, so this is a computer store, in this computer store there are two offers available.
So, something is on sale, all items are on sale and there are two offers available, one offer is you
will get 85% of the price, whatever you buy you will get the product at 85% of the price. And the
other offer is you will get flat 3000 off on the MRP, the maximum retail price you will get 3000
off. So, these are the two offers that are available.

Obviously because you are a good bargainer, you bargain with a salesperson and you strike a deal
that is the computer that you want to buy will be given to you at 85% of the price and of the
amount, once the 85% of the price is decided further 3000 will be given you as a discount. So,
there is a discount of rupees 3000 as well as you are getting 85% of the price.

Now, this kind of thing when we write mathematically can be considered as composite functions,
you are in fact using these kind of tricks in a day-to-day life. So, let us see what happens when we
put this mathematically and how composite functions arise. So, let us say the first draft that is 85%
of the price. So, can I represent this as a function?

(Refer Slide Time: 2:20)

So, for cleanliness let us write let 𝑥denote the item price. So, let 𝑥denote the item price which is
the MRP, you can write maximum retail price and on that you are getting 15% discount that is
85% of the price you are getting. So, I can write this particular offer as f(x) which is nothing but
0.85 times x.
Now, the other offer that is on in this particular computer store is this. So, I can write this as g of
𝑥to be equal to if 𝑥is the MRP I will subtract 3000 rupees from 𝑥, so these are the two offers that
are available. Now, what we did is we want best of both the offers. Now, when a store is offering
these two offers it is safe to assume that you may not have any item that is less than 3000 rupees,
you may not have any item on sale which is less than 3000 rupees, so your 𝑥will always be greater
than 3000.

Another thing that you can assume that because the store is offering you this kind of thing, that is
85% of the price -3000, the store has already taken care of that they do not have to pay back any
money, that means after giving the 85% of the price, the price should be greater than 3000, so all
these conditions are assumed implicitly, which we will deal with them in later when we will
formulate a problem.

(Refer Slide Time: 4:23)

So, now the offer that you got if I want to write this offer mathematically I can write this as some
function h(𝑥) which is equal to it is 85% of the price -3000. So, now when we are dealing with
functions in mathematics it is good to see if I have some correspondence of the function h with
these functions f and g, this is the question that we are trying to answer when we are studying
composite functions.

So, let us first see what is being done over here, that is if I use this f then it is 0.85 x, so if I want
to do something like this then I can write this as h(x) is equal to f(x)−3000 is that a safe assumption
to do? Yes, of course because f (𝑥) is 0.85 𝑥 so what I am essentially doing is, I am for this
particular term I am substituting f (x), so it is a perfectly valid guess, fine.

Now, if you treat this f, if you treat this f (𝑥) as one argument like 𝑥then what you are actually
doing, you are actually saying it is 𝑥minus 3000 that means instead of this 𝑥 had it been f (𝑥) you
would have written f (𝑥) −3000. So, I will use that knowledge and I will try to do, I will try to
rewrite this as, this is g times f(x). Is this acceptable? Let us redo the math.

For example, what is g times f(x)? So, if you look at g of, f of, g(x), so whatever is 𝑥 you will write
that 𝑥− 3000 or whatever, let me put it this way if g had some box inside it then I will write that
box -3000. So, in particular, in that box right now f(x) is written, so I will substitute it as f (𝑥)
minus 3000, done?
And what is f(x)? Now, f(x) as you know is nothing but 0.85× 𝑥 . Therefore, I can rewrite this
function as g(f(x)). In mathematics you will rewrite this as g(f( x)), so my h(x) can also be written
in terms of g and f in this fashion. So, this is the motivation for composition of two functions. So,
in particular what we have seen is a practical example, we motivated it through a practical example
of a computer store which is offering two kinds of sales, one is 85% of the price, another one is
flat 3000 off on the MRP. So, after doing this you can easily guess that how will, how will I
evaluate this function, how will I evaluate this function, that is what we have to see.

(Refer Slide Time: 8:07)

So, in particular let us say your 𝑥in this particular function is say you can take it to be 14000 let us
say, 14000 is your 𝑥and you are asked to calculate g o f(x). So, how will you calculate? It is very
simple, you will first insert g(f(x)). So, what is f(x)? f of 𝑥 is nothing but point okay, let us follow
the same notion the way we followed, so in particular in this case this is what will happen, this is
going to be equal to 𝑓(𝑥) − 3000.

What is 𝑓(𝑥)? f(𝑥) is going to be 0.85 × (𝑥 − 3000), so I will substitute the value 14000 over
here which will give me, so since my 𝑥 is 14000 I will plug this value in, so I am calculating g of
f of 14000. What will be g of f of 14000? Again you have to do a similar calculation which will
give me 0.85 multiplied with 14000 -3000 so this I think comes out to be 11900 just check if I am
calculating it correctly -3000 which will give me 8900, 3000, 900 as it is 11 -3, 8, yes, so the final
answer is 8900, this is what, this is actually, what I have just now shown is evaluation of a
composite function which is g o f, what is, which is actually h, there is nothing special in this, it is
just a nomenclature that we are using.

But this kind of composition helps you in understanding lot of things. So, let me formally define
what is the composition of a function and how we are going to handle them mathematically.
Because composition of a function as you must have seen is again a function. So, natural questions
about domain, range will arise and we will try to answer them as and when they come.

(Refer Slide Time: 11:01)

So, let me formally define the composition of functions. What is the composition of function? So,
in particular we can write as the composition of functions f and g composition of the functions,
there are two, at least two functions you need, functions f and g or we can write the composition
of the function f with g that is also a valid terminology is denoted by, I have already defined this
notation f o g and is defined by f o g, this is one function of x, so you can write this as f(g(x)).
(Refer Slide Time: 12:34)

So, naturally the next question is what should be the domain of this function, so that we will answer
as the domain of the composite function f o g, let me write it here, f o g is actually the set of, is the
set of all 𝑥such that the two conditions we require and they are pretty evident, as we go further we
will realize how these two conditions are evident.

So, the first condition is 𝑥 is in the domain of g and second condition is it will be about 𝑥so if g
𝑥is something that you are figuring out. Now, that g(x) should be in the domain of f, g(x) is in the
domain of f. So, now why these two conditions are required that is what we need to figure out. For
that you need to focus on this particular component f (g(x)). Let us use this particular component
and try to answer the question.
(Refer Slide Time: 14:21)

So, I have, if when I talk about f o g(x), what I am talking about is f(g(x)). Now, let us look at the
first condition. If I want something to be in the domain of f o g that means it should be well defined,
so when I input the value it should give me the output, if there is some ambiguity then it is not a
properly defined function. So, let us say why this condition 𝑥 is in the domain of g.

What if 𝑥 is not in the domain of g? g of 𝑥 is not defined, because g is defined only over domain
of g, so g of 𝑥is not defined and therefore you need this condition that 𝑥should be in the domain
of g. Now, when I am using this composite function, I am applying f to the value that is obtained
by applying g, so it is g of 𝑥that is playing the part.

So, now if this g of 𝑥 that is the value of 𝑥which is in the domain of g if that particular value g of
𝑥is not in the domain of f then again this f(g(x)) is not defined. Therefore, I need g of 𝑥also to be
in the domain of f. So, in particular you can visualize it this way. So, if I have 𝑥then there is a map
which maps everything that map is g and that maps it to a value called g of x.

Now, this g(x) should be in the domain of f because I will take this value to a function which is
f(g(x)). So, this is another value and what is the application? f is the application, we are applying
the function f to the value g(x), if this 𝑔( 𝑥) is not, does not belong to domain of f then our function
is not defined.
So, you can actually remember this diagram by using this particular, this belongs to, what it
belongs to? It belongs to domain of g, this particular thing actually belongs to domain of f, this is
my abbreviation for domain and this is nothing but the range of f, so this will be in the range of f
but it can be smaller than the range of f because g (𝑥) may not cover the entire domain of f.

So, it can be smaller but this will belong to range of f or if you want to visualize it in a better
manner there is something which is box, you feed an input to this box x, g is this box and it will
throw out g(x), so when you feed x, this will spit out g(x). Now, for g(𝑥) to be fed into another let
us say this is a trapezium and that trapezium is a function f, this is a machine, function machine
that is g(x).

If this g(𝑥) is not in the domain of f, then this machine is unable to produce the output, so I need
this g(𝑥) to be in the domain of f in order to get the output which is f(g(x)). So, this is how you
can always remember how to compute the composition of two functions and what are the necessary
steps that are required to compute this composition.
 Mathematical for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 49
Composition Functions: Examples
(Refer Slide Time: 00:15)

So, we have understood the theory, roughly the theory behind the function, composition,
composite functions or composition of two functions. So, it's time to get some practice.

(Refer Slide Time: 00:29)

So, let me start with an example. And in that example, let us take you have been given two
functions f(x)=3x-4, and g(x), which is equal to let us say 𝑥 2 these are the two functions that
are given, then you are asked to find two things one is g o f(x), and the other one is obviously
f o g(x), how to find this? Let us start, let us start with a solution.

So, what can be the solution let us take this function. So, let me write it properly, it is g o f(x).
So, as per our theory, we have to write this as g(f(x)). So, g(f(x)), you can treat this as, what is
f(x) now? 𝑓(𝑥) = 3𝑥 − 4, and g(x) is x square. So, naturally g(f(x)), so, you go to this function,
you treat this g as g. So, let me write it here, you treat this g as a g of a box, and g(x) is nothing
but this box squared. So, in particular, if I want to write something about this function, this box
right now has an argument which is f(x).

So, I will simply write this as f(x) squared, that is all. Now, the entire process is simplified. So,
now, you do not have to worry about what g is, now it simply f(x)2 what is the f(x) fit that when
you in and you will get (3𝑥 − 4) 2. Another way to handle this is you can simply write g o f(x)
as g(f(x)) fit in the value of f(x) that is 𝑔(3𝑥 − 4) and what is 𝑔(3𝑥 − 4) as per our question,
it is 𝑥 2 . So, 3𝑔(3𝑥 − 4) will be (3𝑥 − 4) 2.So, anyway whichever way is convenient to you,
you proceed and you will get this answer correct.

So, what I have done here is I have replaced f(x) in this particular case, I have replaced 𝑓(𝑥) =
3𝑥 − 4 in this particular case I have written f(x) and replaced what is g(x). So, both ways you
can go now.

(Refer Slide Time: 04:01)

Let us go to the second problem that is a f o g(x) and f o g(x) is again can be written as f of g(x).
Clear, there is no question, then there are two ways let us go it the first way, what is f(g(x))?
So, what is f(x) here? 𝑓(𝑥) = 3𝑥 − 4 here. So, I will write this as to be equal to 3𝑔(𝑥) − 4.

So again, let me be very clear about this there should not be any confusion in this. So, what is
𝑓(∆)? ∆ is an argument. So, this ∆triangle will be 3∆ − 4. So, now this triangle is replaced
with g(x), that is all. Therefore, your answer is 3x-3g(x)- 4. But what is g(x)? Again, go back
to the question g(x) is 𝑥 2 So, substituted here that means it will be 3𝑥 2 − 4 and this is the final
answer for you in terms of f o g(x). So, we are seen how to write the compositions in both ways
g o f and f o g.

(Refer Slide Time: 05:26)

So, here is a quick exercise for you pause the video, do the exercise and get back the get the
answer. So, 𝑓(𝑥) = 𝑥 + 1 and 𝑔(𝑥)=𝑥 2 − 1. Then simply find g o f(x) and f o g(x). This is an
exercise you stop and get the answer. It will be a good practice to revise the concepts.
 Mathematics for Data Science 1
Professor Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture 50
Composite Functions: Domain
(Refer Slide Time: 00:15)

Let us now go further and talk about how to determine the domain of composite functions. So,
this will be any important question that is determination of domain of a composite function.

(Refer Slide Time: 00:33)

Determination of the domain let us say for domain for composite function, how will you
determine this? So, I have let us say f o g(x) = f(g(x)), we are talking about all functions that
are real value. So, in order to determine the domain there must be some rules that you should
follow I will list the rules and that essentially says the following rules, the following rules must
be followed and therefore, the following values must be excluded from input values of x.

So, this is again in concordance with what we have seen earlier that if you remember we have
seen some conditions right, where 𝑥 should be in the domain of g and g(x) should be in the
domain of f. So, again what we are discussing now is in concordance with that, but here we
were seeing what are the possible values.

Now, what we are seeing is what are the possible exclusions, that means, what value should be
excluded from the input values. So, there are basically two rules the first rule which
corresponds to the first rule of this that 𝑥 should be in the domain of g that means, 𝑥 if 𝑥 is not
in the domain of g then I cannot include it then 𝑥 cannot be in the domain of the function f o g.

So, I am talking about 𝑓 𝑜 𝑔, when you talk about g o f, you will talk about the 𝑥belonging to
domain of f implies. So, 𝑥 does not belong to the domain of f implies 𝑥 does not belong to
domain of g o f. So, just remember the function the order in which they are taken it matters and
in the similar manner, when I talked about g(𝑥) belonging to the domain of f. So, the set of all
x's such that g 𝑥does not belong to the domain of f.

So, this is the set that you need to be careful about this set must not be included in domain of
our function f o g that is a composite function otherwise, we will have some ambiguity. So, in
order to eliminate the ambiguity, we need to follow these two rules strictly very strictly. So, let
me demonstrate how these rules can fail and then we will I will demonstrate it through an
example and let me take that example as that is write it here.

(Refer Slide Time: 04:38)

2
So example, so I have been given a function 𝑓(𝑥) = 𝑥−1 and another function that is given to
3
me is g(x)= 𝑥 and you want to find f o g(x) and you also need to find a domain of this function

f o g. What is the domain? Domain if you recollect from your week 1 it is nothing but the set
of allowed values for which the function is well defined whatever input values you are fitting
into the function, this function should be well defined this is the domain This is the notion of
domain.
So, let us first see what is f o g(x)? And let us see if it gives you some hints about what can
happen, correct? So, what is f o g(x)? Simply apply our definition it is f(g(x)), fine no confusion
2 2
in this, then again you use that 𝑓(⊡) = ⊡−1. So, that gives me 𝑔(𝑥)−1.

3 3
Now, what is g(x), it is 𝑥. So, substitute what is g(x)? So, it will be 𝑥−1, simplify this assume
2𝑥
𝑥is not equal to 0 and simplify this you will get 3−𝑥. So, this is my f o g(x). Now, the question

the second question that is asked is, so I have given answer what is a f o g(x).

(Refer Slide Time: 06:52)

2𝑥
So, my 𝑓𝑜𝑔(𝑥) = 3−𝑥. Now, if you look at this function, if you look at this function, you can

simply see that at 𝑥=3 this function is not defined, because the denominator is becoming 0. So,
6 / 0 is undefined. So, this function is not defined at 𝑥=3. So, the domain of this function must
exclude 3 that is very well known.

But let us now see because of composition if I am eliminating any points, so here you look at
this function which is f(x). And you look at this function which is g(𝑥) and I am calculating f o
g(x). So, if 𝑥 does not belong to domain of g, then that function that particular value of 𝑥 should
not belong to domain of f o g that is the first rule that we have to implement.
(Refer Slide Time: 08:06)
So, rule 1, what is the rule 1? If 𝑥 does not belong to the domain of g that must imply 𝑥 does
not belong to the domain of f o g. So, what is that point? Let us look at what is g(x)? 𝑔(𝑥) =
3
. So, this function is well defined only when 𝑥 ≠ 0. So, 𝑥 ≠ 0 not equal to 0. So, 𝑥 = 0 cannot
𝑥

belong to domain of g. So, 𝑥 = 0 do not belong to domain of g. So, naturally I will enforce
that 𝑥equal to 0 should not belong to domain of f o g.

So, now, you may come up with some argument that when you look at this function, when you
look at this function, if I substitute 𝑥 = 0 if I substitute 𝑥=0, I am getting 0/3. Then this function
is well defined because the answer is 0. That is what your argument will be. But no, why? I
will tell you because when we when, when we were while we were coming to this particular
form, what we were doing actually is we were multiplying a numerator and denominator by 𝑥
or we are taking assuming 𝑥not equal to 0.

We are taking this 𝑥 on the numerator on the numerator side and multiplying by 𝑥and that is
where we have reached this point. If we had not assumed 𝑥 ≠ 0, then we would not have
reached this point. Therefore, 𝑥 ≠ 0 is a valid condition still even when you cannot see
anything visible over here, because I am composing the 2 functions where 𝑥 ≠ 0 is outside the
domain.

(Refer Slide Time: 10:28)

So, let us come to the next rule 2 that rule 2 was if g(x) does not belong to the domain of f then
I am having a problem. So, that rule we have figured out like 𝑥 says that g(x) does not belong
to the domain of f must be excluded. So, let us look at our function f what is our function f it is
2 upon 𝑥minus 1 in this case 𝑥=1 I have a function where the denominator is 0.

2
So, 𝑥 is equal. So, let me write for the sake of completeness 𝑓(𝑥) = 𝑥−1 this is well defined

when 𝑥 ≠ 1. So, so this also this point 𝑥 ≠ 1 should also be eliminated from the domain of f o
g. So, what should be the domain of a f o g? All other points the function f and g are well
defined. So, domain of f o g must be set of all x’s belonging to real line such that 𝑥 ≠ 0 and
𝑥 ≠ 3, this comma means and or if you want me to be precise, I will write.
(Refer Slide Time: 12:02)

Another quick exercise that you can do in order to verify whether you have understood the
concept of composition of function and the domain is you have been given 2 functions 𝑓(𝑥) =
1 1
and 𝑔(𝑥) = 𝑥 and you are asked to find f o g(x) and the domain of f o g. So, you can
𝑥+1

quickly solve this problem and check whether you have understood what we are supposed to
understand. Thank you.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture No. 51
Inverse Functions

(Refer Slide Time: 0:14)

Hello Students, in the last video we have stumbled upon one concept, where we could not proceed.
Then we came to…Let us go to the last videos last slide. So, here if you look at this particular
concept we actually stopped while computing. And why we stop while computing is
because we did not have enough information on, how to write t is equal to something given this
equation.

So when, what we did we found escape out by plotting the lines across 𝑥 and y a𝑥is, horizontal
and vertical lines and figured out that the answer is 8 minutes. And that is how we concluded this
is 8 minutes. Now when we start such a thing analytically that is 𝑅𝑡 is given to be 30. What is the
value of t? We want to answer such questions then we need to look at the function R and we need
to understand whether this function is reversible or not.

Which is the case, in this case, in this particular function because we were able to map it uniquely.
So, what are the important trades of this function 𝑅𝑡 ? 𝑅𝑡 was a one to one function and it was
increasing function. Hence, it was one to one. Therefore, we were able to find a reversal of the
value 30 to the value of t which is 8.

(Refer Slide Time: 1:58)

So, in order to find such reversible functions, we need to understand the theory which we will
discuss now is the theory of inverse functions. So, when I talk about inverse functions, I am talking
about functions from domain which is real line co domain which is real line. So, a function is
define from real line to real line, then the immediate question that comes to our mind, are all
functions reversible? And the immediate answer is a very well-known function that we have seen
is, 𝑓(𝑥) = 𝑥 2 .
Now this function is not reversible because it feels to pass the horizontal line test, if you remember.
So, 𝑦 = 𝑥 2 , if I try to plot, it will be something like this. Very close to something like this. And
when I pass a horizontal line through this it passes through 2 points. And let these points be 2 and
−2 . And that essentially means this, when I feed in the value 2, it will give you four. And when I
feed in the value - 2, it will give you the answer to be equal to 4.

Now if this function is reversible, when I feed the value to 4 it can spit out the two values 2 and
−2. So, it is not uniquely spitting out the value. Therefore, this function is termed as not reversible
function. Such functions we cannot study the inverse properties or the properties of inverse
functions. However, if you restrict the domain of this function instead of real line to only positive
half of the real line, then you will get one to one correspondence between the values on 𝑥 a𝑥is and
y a𝑥is and then you can talk about inverse of these functions, when it is defined from 0 to ∞.

Now let us look, then the question is, this function is not reversible then which functions are
reversible? That is a question that we can ask now, in order to answer this question, we need to
study some class of functions. So, in last few videos we have already seen that one to one functions
are nice functions. Any function that is either increasing or decreasing is one to one and therefore
we can look at one to one functions for the class of reversible functions. So, here is our answer
that we will start looking at the class of one to one functions.
(Refer Slide Time: 4:51)

Let us look at a simple function a linear function g(𝑥) is equal to 4𝑥. Is this function reversible or
not? So in order to answer this question, let us look at 𝑔(𝑥) = 4𝑥. So, you can put 𝑦 = 4𝑥. If you
look at 𝑦 = 4𝑥 from our basic understanding of linear equations or rather than linear equations an
equation of a straight line. This is a straight line passing through origin having slope 4. So, if I
𝑦
want to find a point 𝑥 on this a𝑥is then I will simply transform this as = 𝑥 and this transformation
4
𝑥
is unique. Therefore, I can write some function let us say r(𝑥) as 4. And this function will actually

be giving be the inverse of this.

𝑥 𝑦
So, let us take this function, if this function ℎ(𝑥) = 4. So, I do not need to write r(𝑥)= 4. ℎ(𝑥) =
𝑥
. Now if I start with this function and I want to get value of 𝑥, what should I do? I will write, so I
4
𝑥
will write 𝑦 = 4 and in that case I will get 4𝑦 = 𝑥. And therefore, I will get another function which

is say 4𝑥 = 𝑠(𝑥). So, essentially what we have seen is this g(𝑥) and h(𝑥) have something in
common. So, let us recollect the notion of composition of two functions, and try to answer this
question.

For e𝑥ample, if I consider the function goh(𝑥). Now this function is again a function and it will
simply operate like 𝑔 of h(𝑥). And once you start with 𝑔 of h(𝑥), what you will do is, you will treat
h(𝑥) as an argument of 𝑔 and put the values of h(𝑥) inside. So, let us try to understand this, so it is
like g(h(𝑥)) is actually, what is g(𝑥)? 4 of 𝑥. So, it will be 4 × h(𝑥). Now what is ℎ(𝑥)? ℎ(𝑥) is
𝑥
nothing but So, 4 times 𝑥 by 4 which will give me 𝑥. So, what
4
this function is, this function actually gives me identity function. 𝑔𝑜ℎ(𝑥) = 𝑥 and a similar manner

(Refer Slide Time: 8:08)

I can start thinking about hog(𝑥). now in this case, if you recollect the notion of composition of
𝑔(𝑥)
functions studied in week one, h( g(𝑥)). So, if h of g(𝑥) I will simply see what is h(𝑥). and
4

therefore what is g(𝑥)? It is 4𝑥 therefore I will get 4𝑥/ 4 which is actually equal to
𝑥. And therefore this is also equal to identity function of 𝑥. So, now to summarize what I got is
G o h(𝑥)= 𝐼(𝑥) = ℎ𝑜𝑔(𝑥). Now this becomes our definition of inverse function. And let us define
it formally as the definition of inverse function.
(Refer Slide Time: 9:16)

So, here is a definition of inverse function. The inverse function inverse of a function f, we denote
it by .𝑓 −1 is actually a function this is our notation 𝑓 −1 is actually a function such that.𝑓 −1 𝑓(𝑥) or
I can rewrite this as 𝑓 −1 𝑓(𝑥) = 𝑥. Now here is a typical thing that comes for all 𝑥 belonging to
domain of 𝑓 which is equal to range of 𝑓 −1 . And 𝑓(𝑓 −1 (𝑥)) or you can write this as 𝑓(𝑓 −1 (𝑥))
being equal to 𝑓(𝑓 −1 (𝑥)) = 𝑥 for all 𝑥 belonging to domain of 𝑓 −1 and range of 𝑓 −1

So, right now when I did this particular calculation I have assume that everything goes from real
line to real line there was no such event. Because this function is define from real line to real line.
And this function is also define from real line to real line. So, there was no consideration for
domain and ranges. But sometimes it may so happen that your original function maybe define, let
us say 𝑓 is define from ℝ to[0,∞). If such a definition is there, then you need to worry about the
domain of a function and the range of a function. Because here the domain of 𝑓 is ℝ and range of
𝑓 is 0 to ∞.

So, if I talk about 𝑓 −1 of this, then naturally I cannot go over entire real line. I have to go over 0
∞ and then I have to come to ℝ. So, this is how it will be define and therefore, the domain of 𝑓
will become the range of 𝑓 will become the domain of 𝑓 and the domain of 𝑓 will become the
range of 𝑓 −1 . This is the typical factor that you need to always remember. Now let us go ahead
and improve our understanding about one to one functions.
(Refer Slide Time: 12:00)

So, if the given function is one to one function then 𝑓 −1 always e𝑥ist for f. Now the notion may
1
confuse you. So, let me give you one precise warning that the notion 𝑓 −1does not mean 𝑓. This is
1
very important. Because you may quite often confuse 𝑓 −1 with . So, whenever we want to
𝑓

discuss in this course or in Mathematics, whenever we talk about 𝑓 −1 (𝑥) it is simply means it an
inverse function.

1
So, this is an inverse function and whenever you want to talk about the 𝑓. Then you should talk
1
about 𝑓(𝑥) − 𝑓 −1 . So, this 𝑓(𝑥) − 𝑓 −1 = 𝑥 and this 𝑓 −1 is actually has a meaning 𝑓 −1 (𝑥) with

this you always remember. Now if 𝑓 is one to one function 𝑓 −1 always e𝑥ist for f. This you have
to trust me. I cannot prove it right now with the current tools, so 𝑓 −1 always e𝑥ists.
(Refer Slide Time: 13:11)

3
Let us take one e𝑥ample g(𝑥) is equal to 𝑥 cube and 𝑔−1 of 𝑥 is √𝑥. This you can write as 𝑥 raise
to 1 by 3 as well. So that, this is simple to verify. So, now you want to verify that the given functions
are actually inverses of each other. So, in this case let us first identify the domains, it is a real line
to and range is real line. So, naturally for inverse also it’s real line to real line. Now question about
it. So, let us talk about 𝑔−1 𝑔(𝑥).

Now if you recollect the definition of inverse function then naturally the inverse function is a
function such that all this combinations, all this combinations should produce 𝑥 𝑓 −1 (𝑓(𝑥)) or
𝑓(𝑓 −1 (𝑥)) f. So, let us talk about 𝑔−1 𝑔(𝑥)). So, let us keep 𝑔−1 intact and put what is 𝑔(𝑥) which
is 𝑥3. Now this you substitute the function 𝑔−1 of 𝑥 as 𝑥1/3 then this becomes (𝑥3)1/3. Then
multiplication of indices 𝑎𝑚𝑛 applicable, so it will 𝑥. So, one way it is true.

Now the second way also you have to check. So, what you will do now, is you just write 𝑔 within
the bo𝑥 you write 𝑔−1 (𝑥) here, 𝑥 raise to 1 by 3. And then simply put the function 𝑔 so 𝑥 raise to
1 by 3 the whole thing raise to 3 which again a raise to m n. So, this will also give you 𝑥 domain
and ranges we have already seen. So, whatever the conditions of that therefore 𝑔 and 𝑔−1 are
inverses of each other. So, 𝑔−1 is inverse of 𝑔 .
(Refer Slide Time: 15:24)

Let us take this e𝑥ample where we are suppose to verify, whether 𝑓 and 𝑔 are inverses of each
other. So, let us try to verify, you can check the domain and co ranges of these functions. I will
simply start with 𝑓(𝑔(𝑥)). So, if I start with 𝑓 (𝑔(𝑥)) as per our notion what we will do? We will
simply keep this 𝑔 𝑥 in place wherever 𝑓 has an argument 𝑥. So, we will take this and we will put
𝑔 𝑥 wherever 𝑥 is written there.
3𝑥+5
𝑔(𝑥)−5 −5 3𝑥+5−5(1−2𝑥) 13𝑥
So, let us do that e𝑥ercise that is, 2𝑔(𝑥)+3 = 1−2𝑥
3𝑥+5 = 2(3𝑥+5)+3(1−2𝑥) = = 𝑥. this is 𝑓 o g(𝑥).
2× +3 13
1−2𝑥

Now what is g(𝑥)? g(𝑥) is this so let us go ahead and substitute those values over, those functions
3𝑥+5
−5
in place of 𝑔(𝑥). So, it is 1−2𝑥
3𝑥+5 . So, now it is a matter of your Algebra just simplify this. So,
2× +3
1−2𝑥

denominator both have 1 − 2𝑥 in common, so multiply the numerator by 1 − 2𝑥 and denominator

3𝑥+5−5(1−2𝑥)
by 1−2𝑥. So, that we will get rid of this. So, it will be2(3𝑥+5)+3(1−2𝑥).

So, what is a question, we have to verify that 𝑓 is the inverse of 𝑔 So, essentially I want to come
up with a number with a function which is 𝑥 𝑓 o g(𝑥)= 𝑥, this is my end goal just remember this.
Now you can simply (multi) simplify this 3𝑥 + 5 - 5 will get rid of this + 5. Let me change the
color over here. So, this will get this one will get rid of this then this is - 2𝑥 - 10𝑥 and - 10𝑥 will
become + 10𝑥. Because of this - sign and then 3. So, I will simply get here 13𝑥.
Now you look at the denominator which is 2 into 3𝑥 that is, 6𝑥 then look at the corresponding term
here - 6𝑥. So, this 𝑥, terms corresponding to 𝑥 will vanish and 3 and 2 × 5 = 10. So, I will give
get the denominator to be equal to 13 and that will give me 𝑥 as my answer. So, 𝑓 o g(𝑥) is verified.
Does this complete, will this complete our verification of whether 𝑓 is the inverse of g? No, because
I want to check whether 𝑔 is also the inverse of f. Then only the verification will be complete.

(Refer Slide Time: 19:03)

So, let us go ahead and do that, that is we will consider 𝑔 𝑓 of 𝑥 and that should give me 𝑥 that is
my end goal. So, now you look at what is a function 𝑔 and put 𝑓 𝑥 as it is everywhere. So, it is 3
times 𝑓 𝑥 + 5 1 - 2 times 𝑓 𝑥. What is the ne𝑥t step take the functional form of 𝑓 𝑥 and substitute
it in the e𝑥pression.

So, 3 into 𝑥 - 5 upon 2𝑥 + 3 + 5 upon 1 - 2 times 𝑥 - 5 upon 2𝑥 + 3 and then again the same logic
applies multiply both sides by the 2𝑥 + 3 and then you will get 3 times 𝑥 - 5 + 5 times 2𝑥 + 3 to
be, upon 2 1 is there. So, 2𝑥 + 3 as it is - 2 into 𝑥 - 5. Let us look at the simplified form let me
change the color. So, 3𝑥 + 10𝑥 that will give me 13𝑥 here 3 into 5 - 15 + 5 into 3 + 15. So, this is
taken care of vanished upon again the same logic applies 2𝑥 - 2𝑥 will vanish 2 into 5 will give me
10 and this 3 will give me 13.
Therefore, I got this domain and ranges you have already verified for yourself and therefore
process is now complete because 𝑓 of g(𝑥) is 𝑥 𝑔 o 𝑓 of 𝑥 is again 𝑥. So, we can verify that 𝑓 is
inverse of 𝑔 as stated. So, this completes our discussion on inverse functions.

(Refer Slide Time: 21:40)

Now it is important to understand graphically what the inverse function is or how the graph of 𝑓
and 𝑓 −1 is changes. So, we already have a wage understanding of the graph of 𝑓 and 𝑓 −1 . Now let
us look at it formally, so if I know something about 𝑓 or the graph of f, then given a value of a. I
am able to calculate 𝑓 of a and a 𝑓 of a is the payer which we call as graph of 𝑥, graph of f. You
look at 𝑓 −1 , what happens when you talk about 𝑓 −1 , here 𝑓 of a is actually on y a𝑥is and a is on 𝑥
a𝑥is.

So, when you look at the inverse function the values on y a𝑥is actually get convert into values of
𝑥 a𝑥is. And the values on 𝑥 a𝑥is will actually get converted into values on y a𝑥is. So, this is the
mapping that we have given. So, if you start with 𝑓 of a which is y then you will talk about
𝑓 −1 (𝑦)of y and when you talk about 𝑓 −1 (𝑦) you will actually get it to be equal to 𝑎. Because 𝑦 =
𝑓(𝑎). and this is how the entire circle is complete. So, in particular if 𝑎 and 𝑓(𝑎) is on the graph
of 𝑓 then 𝑓(a), 𝑎 is the graph of 𝑓 −1 . That is obvious.
(Refer Slide Time: 23:17)

So, let us look at, let us imagine this is the graph. This is the graph of a straight line a, fa. So, you
plot a line 𝑦 = 𝑥 here. This is a line in 𝑦 = 𝑥 and this is a point which is on the graph of 𝑓 (𝑥). So,
now you are saying that where will this point be, when I talk about 𝑓 inverse. So, then we are also
answering this question that wherever 𝑎 was there it will be f(a) now. And wherever 𝑓(𝑎) was
there now there will be 𝑎.

So, in this case you just take this distance and you plot it, you just take the distance on the y a𝑥is
and choose the distance on 𝑥 a𝑥is here and take the distance on 𝑥 a𝑥is for this point and put that
distance over here. That means it will be somewhere here. And therefore, the point will be
somewhere here and this point is actually 𝑓(𝑎). So, what we are actually doing when we are
plotting the graph is actually we are reflecting our original function, in the original function is
somewhat like this. Let us say, so it is somewhat like this.

Then what we are doing is we are actually reflecting it along y a𝑥is and it will be very similar
function. Which will look like this. So, this is how the graph of inverse function will look like it is
actually a reflection along y is equal to 𝑥 or reflection along a function y 𝑓(𝑥)=𝑥.
(Refer Slide Time: 25:13)

So, in particular the graph of 𝑓 and 𝑓 −1 are symmetric across the line y= 𝑥. This is what you have
to remember all the time. If you want, you can prove the theorem but there is nothing it just a
graphical prove that, if I want to compute this particular point and if I know that the inverse of this
function e𝑥ist, then you just take length on y and plot it across 𝑥 and length of, length in 𝑥 direction
plot it across y direction. That is what I did a this is actually the prove of the theorem that the graph
of 𝑓 and 𝑓 −1 are symmetric on y =𝑥 line. That completes our topic on inverse function. In the ne𝑥t
video we will deal with the inverse functions in a more restricted manner that is, inverse of this
e𝑥ponential functions.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture No. 52
Logarithmic Functions

(Refer Slide Time: 0:14)

So, in this video we are going to look at the inverse of exponential function. In the last video we
have seen the inverse of a general function and we have concluded that if the function is one-to-
one, then the finding the inverse of a function is very easy. So, let us focus on inverse of
exponential function in this video and see its properties graph or how it is graphed and a various
other properties about domain and range of these inverse functions for exponential functions.

So, let us recall our notion of exponential function, we started with a function which is a function
will be called as exponential function if it is written in the form 𝑓(𝑥) = 𝑎 𝑥 where there were some
conditions on a, for example, a should be greater than 0 and a cannot be equal to 1, a greater than
0 is a typical condition which we need because otherwise we have to deal with complex random,
complex variables which is out of scope of this course.

So, we are putting a to b greater than 0 and a≠1 is the condition because if you put a=1, then
𝑓(𝑥) = 1𝑥 which is 1 for all of them, so it is not an interesting function to study. So, whenever
these conditions are enforced we know that our exponential function 𝑓(𝑥) = 𝑎 𝑥 is one-to-one and
because every one-to-one function has the inverse this function also has the inverse, there is
nothing special about it. And that inverse we will define as logarithmic function. So, naturally
since we are talking about exponential function with base a so we will talk about logarithmic
function with base a.

(Refer Slide Time: 2:16)

So, here is a definition of a logarithmic function. The definition says that the logarithmic function
to the base a in the standard form is given by y=log 𝑎 𝑥 . So, remember this function is represented
by log to the base a and 𝑥 is the argument of the function, so this is the definition of a function or
this is replacing f, 𝑓 −1 (𝑥)and then 𝑥 is the argument and we are plotting it along y axis and is
defined to be the inverse of the function 𝑓(𝑥) = 𝑎 𝑥 .

So, 𝑓 −1 (𝑥)is actually log 𝑎 𝑥 , is this simple. So, now we need to understand what will be the
domain and codomain or range of this function that is an important thing that we need to
understand. So, in order to that let us try to devise some rule so that we will have a track of what
is exactly happening when we are talking about logarithmic function and how it is related to
exponential function.
(Refer Slide Time: 3:34)

So, there is a one to one correspondence between logarithmic function and an exponential function
which is expressed by this relation y=log 𝑎 𝑥 if and only if 𝑥= 𝑎 𝑦 or for more precision you can
write this as log 𝑎 𝑥 =y then you can actually virtually assume this 7 rule that is you start from the
base, go to the right hand side and come back that means what we are saying is you start with a,
go to the right hand side, that right hand side is raise to the power and that should give you 𝑥, that
is what this rule is.

So, this is simple technique to remember known as 7 rule. So, you can use this 7 rule to memorize
the one-to-one correspondence between log and the exponential function. You can easily see that
by definition if I write 𝑥= 𝑎 𝑦 , then I want to know the value of y, I should be able to get it by
taking the log of this function 𝑥.

So, this is the mathematical definition of our logarithmic function. To make this mathematical
definition precise we need to understand some prototypes that is whether this function we have
defined it to be the inverse of f but whether this function is actually the inverse of f or not that is
what we need to figure out.

So, as stated earlier we can actually check these two rules 𝑓(𝑓 −1 (𝑥)) = 𝑥 and 𝑓 −1 (𝑓(𝑥)). So,
what is𝑓(𝑓 −1 (𝑥))? As I mentioned earlier 𝑓 −1 (𝑥) is nothing but log 𝑎 𝑥 and f is 𝑎 𝑥 so you just
 −1 (𝑥)
substitute 𝑎 𝑓 . What is that? 𝑎log𝑎 𝑥 . Now, what this should be? You use this one to one
correspondence from here to here and here to here and you will get this to be equal to 𝑥.

In a similar manner you can apply it to f of 𝑥 and𝑓 −1 (𝑥) f inverse, so 𝑓 −1 (𝑓(𝑥)) is log 𝑎 𝑓(𝑥) but
what is 𝑓(𝑥)? It is 𝑎 𝑥 and therefore log 𝑎 𝑎 𝑥 = 𝑥.

(Refer Slide Time: 6:27)

Now, in order to understand this completely I need to understand the domain of log function and
range of log function and the range of log, range of exponential function and the domain of
exponential function. So, let us understand this particular thing. We have already seen what is the
domain of 𝑎 𝑥 , so we already know domain of 𝑎 𝑥 because 𝑥 can be entire real line and then it maps
this domain onto the range of 𝑎 𝑥 that range cannot take negative values, this is what we have seen
when we studied.

So, it was 0 to ∞, so this should be clear before going to the range of log function. So, if at all the
logarithmic function is to be defined, this if you recollect this should become domain of log to the
base a and this should become the range of log to the base a, so the this is the crux of the definition
of inverse. So, when this is satisfied you are done.

So, essentially your log function will be defined from 0 to ∞ to real line. That means in the domain
it cannot have negative values, it cannot have 0 as well and in the range it will have the entire real
line that is what is written here in this case that is domain of log to the base a is actually range of
𝑎 𝑥 which is 0, ∞ and domain of 𝑎 𝑥 is actually the range of log to the base a which is real line, the
entire real line.

These are the two important points which will help you in understanding the domains of the
functions which are derived from these functions that is logarithmic functions or exponential
functions. So, these, all these things you should always remember the valid ranges and domains of
the function. So, this completes our verification that logarithm function the way we have defined
is actually an inverse of exponential function. Once the verification is complete let us dwell more
and find the domain of the derived functions, derived, by derived functions means composition of
basic logarithmic function.

(Refer Slide Time: 9:04)

For example, let us take an example of f 𝑥 which is log to the base 4 of 1 -𝑥. Now, log to the base
4 is actually a function which has a domain. What is the domain of this function? The domain of
this function is actually 0 to ∞. Now, that means the argument that is supplied to this function log
to the base 4 cannot be 0, or it cannot be a negative value. So, based on this understanding from
the definition of our log function you can look at this function which is f of 𝑥 and look at the
argument of the function 1 -𝑥.

According to this definition 1 -𝑥 must be strictly greater than 0. This will happen if and only if my
1>𝑥, 1>𝑥 and because 1 -𝑥 needs to be greater than 0 can 𝑥 be less than 0, if you look at 𝑥 to be
less than 0, 1 -𝑥 will actually be greater than 0. So, the only condition that we require over here is
my function should be defined that is domain of this function f should be equal to, it cannot include
1, 1 to, it is not 1 to ∞, this is how we commit mistakes.

So, domain of f is 𝑥 should always be less than 1 that means the domain of this function should be
here -∞ to 1 and it cannot go beyond 1 this is what our understanding is about this function. Now,
let us go and enhance our understanding in finding the domain of a function which is slightly more
complicated than this function.

(Refer Slide Time: 11:26)

So, our question is to find the domain of this function g. In order to find the domain of this function
g, let us first understand what is the domain of the function log to the base 3. Now, this function is
defined when the argument given is between 0 to ∞. So, now I want the argument of this function
which is this gx to be between 0 to ∞. So, what I should do is I want this 1+𝑥 upon 1 -𝑥 trapped
between 0 to ∞ that means it should be greater than 0. Now, when this can happen?

So, naturally let us split the real line into some parts 𝑥 is≠1 is already given to you, so 𝑥 cannot
take the value 1, this is a point 0, this is a point 1, let, for safety let us put the point -1 as well here.
And now 𝑥 cannot be equal to 1, so this point is actually deleted, so this point cannot be there.
Then, 1 -𝑥 should, if 1 -𝑥>0 that means my 𝑥<1 the function is defined.
So, I have this in the similar manner -∞ to 1 but let us not go for -∞ because there is in the numerator
there is 1+𝑥, so this 1+𝑥, it can become, it can take a negative value when 𝑥<-1 and if 𝑥<-1 this 1
-𝑥 will become positive. So, I have to rule out that part as well. So, this -1 to 1 is rule, -∞ to -1 is
ruled out, -1 will give me the value 0 so -1 is also ruled out and therefore I am only left with the
interval of this form which is -1 to 1.

So, based on the arguments and based on this domain I know that the domain of this function is
valid only between -1 to 1. Now, you may say why not 0? 0 will not cause any problem because if
you look at the function, if you substitute 𝑥=0 you will get log to the base 3 of 1 which is a positive
number and therefore it is well defined. So, the domain of this function is nothing but -1 to 1, this
is how we need before trying to solve any problem related to logarithms we need to first verify
whether it is, what the problem that we are willing to solve is defined in a proper domain or not.

Most of the times when you try to formulate a problem the problem may not be defined in a proper
domain and then solving that problem is a meaningless exercise. So, just to ensure that always
your problem is defined in a valid domain. So, this ends the verification of this. Now, let us take
one more example which will actually help you in understanding the reversibility of log and
exponential function.
(Refer Slide Time: 15:06)

So, here is an example where we are actually demonstrating the reversibility of a log function or
the inverse of a log function. So, y=log to the base 3 of 𝑥. We assume that everything is well
defined and this 𝑥 belongs to 0 to ∞. In that case this y will belong to the real line and if I want to
write 3𝑦 then I will write 𝑦 as 33log3 𝑥 .

By definition, by definition this function is the inverse of the log function. Therefore, you will get
this to be equal to 𝑥 and therefore your 3𝑦 =𝑥. Now, how this helps in your calculations? Suppose,
you know some number 1.32 = 𝑚 and you want to identify this m. Then you can actually take the
log of this function, log of this function which is the inverse of this and which will be equal to log
to the base 1.3 of m and if you equate these two what you get here is 2 being equal to log to the
base 1.3 of m.

Why is it so? Because 1.3 square we have taken the log so this is like 𝑎 𝑥 and you are simplifying
it. So, 𝑎 𝑥 , 𝑎 log𝑎 𝑥 is actually 𝑥. So, you will get the number 2 naturally. So, this is how the log
thing helps.

(Refer Slide Time: 17:15)

And here what the, the fact that we have used is 𝑎𝑢 = 𝑎𝑣 𝑓𝑜𝑟 𝑎 > 0 and 𝑎 ≠ 1 implies 𝑢 = 𝑣. If
you use this fact and you are asked to find the log to the base 3 of 1 by 9, then you can easily find.
Let us see how. So, you start with log to the base 3 of 1 by 9. Now, you look at this 9 and 3. If you
1
look at 3 square that will give you 9 isn't it and that also implies3−2 will give me 9. So, I will
1
simply use the fact that log 3 3−2 = 9.

So, but this is an inverse function, this is like 3 raise this particular thing is like 3𝑥 , log 3 3𝑥 is again
going to be 𝑥, so you will get -2 to be the answer, there this is how you can solve some problems
very easily when you can identify the base is actually multiple of this particular argument. So, this
is the use of log we will deal with it in more detail when we will solve the problems on logarithms.
Now, for a moment we have identified what is the inverse function of our exponential function, it
is logarithmic function to the same base as exponential function.
(Refer Slide Time: 19:23)

Let us try to look at the graph of the inverse function that is graph of f 𝑥 equal to log to the base a
of 𝑥. How will it look like? If you remember the graphs of exponential functions, the graphs of
exponential functions were having two discriminations, like if you take a the line from 0 to ∞, then
there was some split at 1 and from 0 to 1 when there is, the value of a lies in 0 to 1, the graph was
different and from this side onwards that is a>1 the graph was different.

So, let us first imagine those graphs and let us recollect from the previous video what was the
interpretation of the graph of the inverse function. If you recollect from the previous video, the
graph of the inverse function is nothing but the reflection of the original function f along the line
y=𝑥 or the mirror image of the function.
(Refer Slide Time: 20:25)
So, let us look at the exponential function first when 0<a<1 and a>1. So, this is the graph when
0<a<1. Now, I have made it big enough so that you can understand better and the blue line is the
line y=𝑥. Now, if I want to translate the mirror image of this function how will I translate?

Let us take one point so let us take a point 0, 1 over here, the translation of that point will be 1, 0
over here and then take this point over here, I should not draw any point here because it may
confuse you so the translation of that point in this zone is a point over here and a point over here,
and similarly you go on translating and connect the two lines.

For example, here if I go on translating this point then the translation will actually go to some place
over here and if you take one more point over here then the translation will actually go to the other
quadrant which is 2 units below this and over here. So, the graph of this function will actually look
something like this, it will pass through the same nodal point and it will pass through this and then
on y axis it will be very flat, very close to the y axis and so on, so this is how the graph of the
function will look like because it is a mirror image.

So, this is how it will look like, it is not an asymptote but because the graph paper is over I am not
able to draw. In a similar manner this is the case when 0<a<1. So, I have drawn the graph in the
next sheet which is a green line you can see this green line actually matches with this green line, I
have slightly shifted the graph paper in order to have a better visibility.
Now, you can actually see this is the original function, this is the new inverse function and this is
the line y=𝑥, so you can see the correspondence of the inverse function with respect to the original
function, all this is possible because our function is one-to-one.

(Refer Slide Time: 23:07)

Now, if you look at, again look at the graph of a function where a>1 then this is the graph of a
function here there are no overlaps, so it is relatively easy to draw the graph. For example, I can
choose this point over here if I go one unit from here I should get something like this here so it is
a reflection along 𝑥 axis, so it will be relatively easy to draw the graph here, this point reflected
here that point will be reflected here and then I can draw that, I can join the curve like this and it
will be exact mirror image of the original function and it will be going close to this particular
function.

(Refer Slide Time: 24:03)

So, roughly this will be the graph, I have drawn the full proof graph on the next graph paper which
is here. So, now you can easily visualize the graphs of both the functions, let us zoom out and see
all of them together all 4 graphs together. So, these are all 4 graphs handled together, so my graph
actually looks like this graph for both the cases, so this is how it is easy to draw the graphs of
inverse functions once we know the graph of the original function.
In the next, this is, that is all for this video. In the next video what we will see is we will try to use
our knowledge of logarithmic functions and try to see how the formulation of a mathematical
problems becomes easy when we consider logarithmic functions, even though there is a limitation
that logarithmic function is defined only from 0 to ∞ not on the real line. Thank you.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Management Studies
Indian Institute of Technology, Madras
Lecture No. 53
Logarithmic Functions: Graphs

(Refer Slide Time: 00:14)

Hello friends, in the last video, we have seen how to graph a function 𝑓(𝑥) = log 𝑎 𝑥. In
particular, we have seen two kinds of graphs or two kinds of divisions, when our a is between
0 and 1 the graph has one form and when a was > 1 the graph has the other form.

(Refer Slide Time: 0:44)

In particular, what we have seen is if the graph of exponential function for a < 1 there is the
graph of a raise to 𝑥 for a < 1 is given by a red line, then you can actually reflect this graph
along the blue line which is 𝑦 =𝑥 and get the corresponding graph forlog10 𝑥 of 𝑥 when 0 < a
< 1.

In a similar manner, when a is > 1 the matters are, matter is very easy. And you see there is
because there is no intersection, you can simply reflect the red line along the blue line to get
the green line and the final graphs will look like this. So, this is the this is the case where my a
is > 1 and this is the case where 0 < a is < 1. In particular, we have already seen some something
similar in the, when we studied exponential functions. In particular, we will try to list all the
properties of this graph of exponential function.

(Refer Slide Time: 01:52)

So, let us start our journey of listing the properties of graph of logarithmic function. So, first
thing is the domain of the function we as it is an inverse function of exponential function is
(0, ∞) and the range of a function is real line, when you studied exponential function, the
intercept was (0,1) here the 𝑥 intercept is (1, 0) and there was no 𝑥 intercept here, there is no 𝑦
intercept, because simply because it is reflection along 𝑦 =𝑥 line.

Then you had (vertic) in when you studied exponential function, you had a horizontal
asymptote that is 𝑥 axis was your asymptote in this case, you will have a vertical asymptote
and that is 𝑥 =0 is the line it is easily visible over here. For example, if you look at this green
line, the vertical asymptote is towards the positive side of ∞ that is positive side of 𝑦 axis. And
if you look at this particular picture, it is towards negative side of 𝑦 axis.

So, these are the typical features that you can you will understand when you look at the graph
of a logarithmic function, then naturally this is the inverse function of a one to one function.
So, it is one to one and it passes through two points. If you recollect, the exponential function
was passing through 0.01 and 1 a. So, naturally this function will pass through points 1 0 and
a 1 all the time. So, these are the two static points whenever you consider a graph of a
logarithmic function.

As it is visible from the graph for 0 < a < 1 this green curve is actually a decreasing function.
And for a > 1, this green curve is actually an increasing function. So, that those properties
naturally boiled down to for 0 < a < 1 the function is decreasing and for a > 1 the function is
increasing.

(Refer Slide Time: 4:07)

So, these are important properties of graph of logarithmic function. So, while drawing the
graphs of logarithmic function in a standard form, you should always remember whether you
are satisfying these properties or not, that is a cross check whether your answer is correct or
not.
(Refer Slide Time: 04:29)

So, let us enhance our knowledge by taking an example of drawing a graphs drawing a graph
is not correct, or drawing graphs of the functions that drawing the graphs of the
functions,𝑓(𝑥) = − log 4 (𝑥 + 1) and 𝑔(𝑥) =log 1 (−𝑥 + 1).
4

Now, you remember the domains the domains of the function. So, here if I want to draw a
graph, let us take the function 𝑓(𝑥) here If I want to take a graph of 𝑓(𝑥), so, first let us
understand how the graph of log 4 𝑥will look like in order to understand this let us go to the
properties is a the base > 1 or < 1 this is the first concern, so my base is > 1.

So, the function should be increasing okay naturally the function is one to one and my curve
should pass through (1,0) and (4,1) correct and I should have vertical asymptote at 𝑥 =0
naturally the function is actually increasing that means, I will come from down to up and
therefore, and obviously 𝑥 intercept is (1,0) domain of 𝑥 is 0 to ∞ range of f is ℝ.
(Refer Slide Time: 06:02)

So, keeping all these things in mind, let me draw a quick sketch of this graph log 4 (𝑥). So, let
me change the colour and it will be like this this is an asymptote, then it will pass through this
point and you know what this point is and then it will be like this, this is the basic understanding
it should pass through point (1,0) this is the point (1, 0) and (a, 1).

So, 4, 1 should be the point that it should pass through. So, naturally let us see this 1, 4 on 𝑥
axis and this is 1 on 𝑦 axis. So, as it passes through (4,1). So, this is these are 4 units this is 1
unit. So, this is 1 unit and that is 4 units on 𝑥 axis this is 1 unit. So, this is how the graph will
look like, but you have not been given a graph of log 4 𝑥.

So, what is happening it is going to 𝑥+1. So, now, log 4 (𝑥 + 1), how will it look like? That is
the next question. So, what we are doing is we are shifting it on 𝑥 axis. So, whatever value 𝑥
was taking now, 𝑥+1 is taking that is a shift along this direction and shift along this direction
of 1 unit. That means, that simply means, again you can easily quickly draw the graph, rough
sketch of the graph whatever was happening for this particular thing, everything will shift and
let me write use a green colour.

Now, instead of having 𝑦 = 𝑥, instead of having this, this particular 𝑦 axis as an intercept, I
will have a new horizontal vertical asymptote which will be at 1 unit apart and my curve will
pass through this this will be my new asymptote and my curve will pass through this and it will
behave like this simple. So, instead of 1, 0, everything is translated by 1 unit. So, I will have 0,
0 the values on 𝑦 axis will not change values on 𝑥 axis will change. So, everything is translated
by your 1 unit. So, I will have a point 0, 0 and this point where it intersects it will be 3,1. So,
now this is my new graph of locked to the base 4 of 𝑥+1.

Now, the twist that is added is the-sign. So, that means it this is the graph of 𝑦 =log 4 (𝑥 + 1).
Now, if I add a-sign to this, the 𝑦 will become the -y. So, now reflection along 𝑦 axis so, the
well what I have d1 is reflection along 𝑦 axis. What I actually meant was reflection along 𝑥
axis. So, the graph of this function should reflect along the 𝑥 axis when I substitute 𝑦 = − 𝑦.

(Refer Slide Time: 9:55)

Therefore, the graph will simply twist itself along the file along the line 𝑦 along 𝑥 axis and
therefore, the upside will go down and the downside will go up and therefore, the function will
look like this, that is all. So, as you can easily see, the function will look like this the point 0,
0 will remain intact and then point 3,1. So, 3 on 𝑥 will become (3,-1), rest of the things will
remain intact. And naturally this is how the graph will look like that is all.
(Refer Slide Time: 10:45)

So, let us go to the next problem that is 𝑔(𝑥) = log 1 (−𝑥 + 1) so, now we will first look at,
4

can I draw a graph of log 1 (𝑥). The answer is yes I can draw and it will be simply this kind of
4

1
graph. So, let me draw it a quick snapshot of that graph log 4 (𝑥) by quick snapshot will give

me something like this. So, this is the point and this will increase keep on increasing and it
1
should pass through 1, 0 and a, 1. So, that point will be here somewhere. So, 1, 0 and a 1 so, ,
4
1
1 and 1, 0 correct. So, somewhere here it will be 4 and 1 so this is a point 1 here, fine.

Now, the twist is this particular function is twisted with log 1 (−𝑥). So, wherever 𝑥 is there, you
4

are replacing the values with-𝑥 that simply changes the paradigm that is, you are taking a
reflection along the 𝑦 axis. So, now essentially when I substitute−𝑥, this will be an asymptote
this will still remain an asymptote and the graph will switch like this it is an exact mirror image
of this along 𝑦 axis.

1 −1
So, naturally the point 0, 1 will become 0-1, 0-1 and that , 11 will become ( , 1). So, that
4 4

also will be there. So, this is how the graph will look like now, you are adding a twist to the
problem by adding+1. So, what will this thing this operation do. So, earlier 𝑦 = log 1 (−𝑥). was
4

there. Now, adding+1 will simply shift these values along 𝑦 axis in upward direction 1 level
up.
(Refer Slide Time: 13:10)

So, naturally the graph will look something like this. So, the point 1,0 will now be shifted to
1 1
1,1 that is this point and 4,1 will be shifted to 4,2 that is this point. So, these are the two points.

So, we are able to map those two points and therefore, the verification of graph is complete and
this is the current graph.

(Refer Slide Time: 13:52)

So, now in the next thing that we want to introduce to you is like logarithmic function to the
general base a we have some special logarithmic function that is called natural logarithmic
function which is which involves the notion of Euler’s constant that is e and this is very special
as the natural exponential function is special.
So, this function is defined in a separate way as the natural logarithmic function and it is defined
as log 𝑒 𝑥, where the base is e we have already seen the importance of e in past few videos. But,
to add to the speciality, we have some special notation for this log 𝑒 𝑥 it is always denoted by
ln 𝑥, where l stands for logarithm and n stands for natural base.

(Refer Slide Time: 14:50)

And whenever we write this as ln of 𝑥 that simply means, I am talking about the natural
logarithm of 𝑥 that is log 𝑒 𝑥. So, hence forth whenever we discuss about natural logarithmic
function, we have to use this notation ln of 𝑥 it is quite standard simple verification, you can
actually check ln of e raise to 𝑥 is 𝑥 for 𝑥 belonging to ℝ which is the domain of 𝑒 𝑥 and 𝑒 ln 𝑥
is 𝑥 for 𝑥 belonging to positive real line, which is the domain of ln x.
(Refer Slide Time: 15:39)

In a similar manner, there is some other log there are some other logarithms, what is called this
called natural log something else is called common logarithm, which has to do with our decimal
representation and common logarithm is actually denoted as log without any base. So, that
means, it is log10 𝑥 = log 𝑥.

So, in general, the common in common terminology, you may consider when there is no
mention of a log, you may consider this log is to the base 10 and if something like ln 𝑥 is
written, that is log 𝑒 𝑥. So, this is what you these are the important things you need to remember.
In olden days when we used to use calculators, there were two separate keys associated with
this 1 key was ln and another key was log.

So, in these keys, they were commonly referring to log10 𝑥if I am talking about log and log to
the natural base if I am talking about ln, so just remember these are two commonly used
logarithms, which we will use quite often in our daily practice. Calculators distinguish them
with ln and log and we prefer that you also distinguish them with ln and log. This finishes the
topic of natural logarithm, natural logarithm nothing is special. It is just a way of taking the
logs with e as a base.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Management Studies
Indian Institute of Technology, Madras
Lecture No. 54
Solving Exponential Equations

(Refer Slide Time: 00:18)

So, let us try to go back and solve our exponential equations. So, this is where the logarithms
will come handy. In particular, we have seen one property of logarithm, for example, if I am
talking about 𝑎 𝑥 ., and if I talk about log 𝑎 𝑎 𝑥 . I get 𝑥, this property we have to emphasise and
track everything in terms of this property and solve the exponential equations while solving the
using logarithms.

So, let us try to get hands on these exponential equations using the logarithm. So, first you have
to solve this equation naturally we have to solve for 𝑥. So, 2 raise, the equation is 2𝑥+1 = 64..
Now, as you know earlier that this is not more of logarithm, but more of inspection. So, 64
seems to be a nice number. So, you can write 64 as 16×4 is 64 but 16 is nothing but 24 × 22 .
So, what you got here is 26 . So, my 64 can be written as 26 .

Now, I have been asked to solve for this equation that is 2𝑥+1 = 64. So, essentially 2𝑥+1 = 26
if and only if, so you can take this 1 2 out 2𝑥 = 25 , hit the function with a logarithm and use
the property. So, use the property log 𝑎 𝑥 = 𝑥 hit the function with logarithm and you know
logarithm is one to one function.
So, nothing changes what should be my a, it should be 2, then I naturally I will get
log 2 2𝑥 = log 2 25 . Using this property, you can actually write this as 𝑥 =5 there is my answer,
I have solved this expression. So, my answer to this question is 𝑥 =5.

(Refer Slide Time: 03:08)

Let us look at the second example. This is also an exercise in computation where I will try to
match these 2 but it becomes slightly complicated because as you can see, it involves the term
containing 𝑥 2 . So, again the question is to solve for 𝑥. So, let us solve for 𝑥. So, let me first
2)
simplify the right-hand side, 𝑒 (𝑥 = (𝑒 𝑥 )2 using law of indices will become 2𝑥 and this 1 by
𝑒 3 will become-3.

Again, you use the natural log take ln on both sides. If you take ln on both sides using the same
property that I mentioned earlier, I will get 𝑥 2 = 2𝑥 − 3. That simply means, I have a quadratic
equation which says x^2+2x-3 =0. Use our knowledge of quadratic functions or quadratic
equations. And in this case, the quadratic equation I think can be solved with by a very easy
solution that is 𝑥+3x+3×𝑥-1.

This is simply by factoring you will just look at the lectures on quadratic functions and see how
to solve the equation by factoring. Actually, I will split this× 𝑥 2 + 3𝑥 − 𝑥 − 3 that will give
me-1 as common and the first factor is 𝑥+3. So, this is how I will solve done. So, now, what I
have as an answer is 𝑥 = −3 𝑎𝑛𝑑 𝑥 = 1.

Now, remember, we are solving some equations, where the domains and co domains may not
be defined properly, then we may land up in infeasible solution. So, always check whether
these two functions will work here or not, let us substitute this 𝑥 = −3 over here and you can
verify that it perfectly works and 𝑥 =1 also perfectly works. Therefore, these both are actually
the solutions to the problem.

(Refer Slide Time: 05:50)

Let us complicate the matters a bit more by putting in this kind of equation. So, let me write
this equation and now, obviously, the question is to solve for 𝑥 9𝑥−2 × 3𝑥+1 − 27 = 0. So,
here if you look at the equations closely before actually handling them, you can see that there
is some common feature between this exponent and this exponent. What is that I have
something like 32 = 9𝑥 , correct and 3(𝑥+1) , I can take the+1 out, so that these 2 will become
6. So, 6 times 6 × 3𝑥 .

Now, this thing you can rewrite as3𝑥 2 . the whole square using this trick, it is very easy now
to see that this particular thing is 3𝑥 2 − 6 × 3𝑥 − 27 = 0. Now, there are 2 ways this
equation is actually very similar to a quadratic equation or you can reword it as it is a quadratic
inform equation. Now, using this equation, you can actually solve for 3𝑥 not for 𝑥.

So, let us go ahead and put 3𝑥 . as t and is 𝑡 2 − 6𝑡 − 27 = 0 again resort to a method which is
like factoring. So, 27 can be factored×9 × 3, 9 3s are 27 with the-sign, so, it will be 𝑡 2 − 9𝑡 +
3𝑡 − 27 which is going to be equal to 0 this will give me this is if and only if t-9×t+3 equal to
0 correct. Now, the question comes t+9 =0. So, I have solved it for t what is t? t is 3𝑥 . So, I
have to resubstitute that and if I substitute that, then I get 3𝑥−9 × 3𝑥+3 = 0..

Now, you have to be a little bit careful. So, this simply means using the factor logic 3𝑥 =
9 𝑜𝑟 − 3.. Now, this should give you an alarm in your head that this particular thing is not
possible for any real 𝑥. So, this option is infeasible. So, I cannot solve for this, what about this?
Do I know something about this? Again, simple thing is you hit with a log you will get the
answer or in this case, it is more obvious that 𝑥 =2 should be the answer.

So, now once you got this you substitute this 𝑥 =2 in the original equation and verify that it
satisfies the equation that is that will be a good cross check. For example, 92 is 81, 32 is 33 =
27. This is 27, 27×2 is 54, 54+27 which should give you 81? Yes. So, it is a verification and
we have solved this problem successfully, but remember here the occurrence of infeasible
solution.

(Refer Slide Time: 10:17)

Let us take it a step further and see whether I can do something about an equation of this form,
which is 5𝑥−2 , 3𝑥+2 unlike previous problems, the bases are not the same. So, what should we
do about it is the question. So, in this case also we can actually rely on logarithm and we can
blindly hit with a logarithm, but in this case, let us hit both sides of the equation with natural
logarithm or you can take common logarithm it does not matter 5𝑥−2 =ln 33𝑥+2 .

Now, if you have noticed that when you hit with the log of the same base the other thing gets
vanished, but if you hit with a log of some other base that number will remain unperturbed, but
this 𝑥-2 and 3x+2 will come in front that is 𝑥-2 ln 5 =3x+2 ln 3. Now, this ln of 5 and ln of 3
are nothing but merely some numbers which are getting multiplied with 𝑥 and 2. So, they are
just constants.

So, now my equation is actually a linear equation let us try to simplify this. So, here there are
there is 3x here there is 𝑥 so, we what we will do is we will take both parties on one side that
is the parties corresponding to 𝑥 on one side and parties corresponding to constant on one side.
So,-2 times ln of 5 will remain as it is+2 when it comes here becomes-2 and ln of 3. So,+ln of
3 which will be equal to 3x was already there multiplied with ln of 3 and from here comes 𝑥
which is-x times ln of 5.

So, now what I got here is 3x-x times ln of 5, I want 𝑥 to come out common So, I can process
it further which will give me 𝑥 as a common factor 3 times ln of 3-ln of 5 and here it is-2 times
now, ln of 5+ln of 3 you will learn ahead it will be ln of 15. And if you look at this particular
expression this equation will not, this expression in the square bracket will not be equal to 0.
So, I can take this expression here another thing that you may notice is 3 here, can be raised to
the power of 3 over there and you can further simplify.

15
So, you will get in particular 𝑥 = −2 ln 3 ln 3 = ln 27 − ln 5 which can further be simplified to,
1
if you want you will have more precise structure later 152 which will be 225, 2 and because of
1
ln
this-sign it will be 225
27 .
ln
5

So, this is again a number and this resolves the problem for 𝑥. Whether these numbers are
feasible? Yes, they are very much feasible and therefore there is no visibility violation over
here. This we will learn in a bit later how to do all these calculations, but as of now, this is the
answer.
(Refer Slide Time: 15:12)

Now, let us look at this particular example, where you cannot solve on your own, but graphical
solution may yield a better answer. So, for that let us look at 𝑥+𝑒 𝑥 = 2. Now, if I try to hit this
with a log, but before hitting this with log because I am encountering a + sign, let me put rewrite
this equation as e raised to 𝑥 equal to 2-x.

Now, hit this expression with a log, so you will get 𝑥 =ln of 2-x and you are struck you cannot
go anywhere. Now, if I want to solve this equation again the only thing that right now we are
aware of is I will hit it with exponent and I will again get back the same equation. So, it is of
no use. So, let us focus on this equation and let us let us try to graph this equation. So, in
particular I have a function which is 𝑙𝑛 2 − 𝑥 − 𝑥 = 0. Now, let us try to use some graphical
tool like Desmos to graph these equations.
(Refer Slide Time: 16:50)

So, let us go ahead and use Desmos for graphing the equation and naturally the choice of log
is only up to B. So, I will use ln or I will use whatever so, ln of 2-x you can see the graph-x.
Now, the point where it cuts 0 is the solution to the problem and you can actually check using
Desmos that the point is actually 0.443.

So, let us go back and draw and tell everybody that 𝑥 =0.443 is the approximate solution to this
problem, but in such problem this is the best that you can do. So, today we have seen how to
solve exponential problems using graph or using algebraic methods. Thank you.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture No. 55
Logarithmic Functions: Properties - 1

(Refer Slide Time: 00:14)

So, in this video we are going to look at further properties of logarithmic functions. So, when we
look at the logarithmic functions in general we have a standard set of conditions that are imposed
that is 𝑎𝜖(0,1) and a > 0. And we already know few properties of logarithmic function namely
logarithmic function is actually an inverse of exponential function which is conveyed through this
property that is if you take 𝑎log𝑎 𝑥 , then you will get x back and if you use 𝑎 𝑥 and take log 𝑎log𝑎 𝑥
you will get x back.

So, essentially our logarithmic function is an inverse of exponential function. Another thing that
you need to recall based on the graphical representation of logarithms is log to the base a of 1 is 0,
that means we have already located that point 1, 0 is on the graph of the logarithmic function
independent of whether a < 1 or a is bigger than 1. Another thing that you need to remember or
recollect is log to the base a of a is equal to 1. What does this mean?

The point a1 is on the graph, which you have also seen, so when you are talking about logarithmic
function, these two points are on the graph, when you are talking about exponential function
because it is a reflection along y is equal to x axis 0, 1 and 1a are the points that you will look for
exponential functions. So, this we have done enough. So, these are the basic logarithmic properties
that we are already aware of.

(Refer Slide Time: 02:17)

Now, let us go further and explore something which is called laws of logarithm. Prior to that now
because we have mentioned log to the base a of 𝑎 𝑥 is x, you can also ask a question that what if I
𝜋
have been given a function like this, which is 3log3 2 , so based on this you can use this particular
formulation which is where a is 3 log to the base 3 of pi by 2 naturally this should be equal to pi
by 2. So, the sometimes some complicated numbers may simplify in this way, it is simple
demonstration of use of these properties.

Another thing that you can also see is say let us say 4 raised to log to the base 4 of 1, now you
already know log to the base 4 of 1 is 0 based on this property and therefore it is 4 raised to 0
which will naturally give you 1, so all these simple simple tricks you should solve, you should
solve more and more problems and gain more confidence in while using the logarithms, because
while solving the problems on logarithms and exponential functions applying a log function or
applying an exponential function will play a crucial role while solving the problems.

So, let us focus on some simple laws of logarithm and in fact when I, why these are called laws?
Because these are the principles for which the logarithms were invented by Napier. So, let us see
what are the laws of logarithms.
(Refer Slide Time: 04:07)

So, in order to define laws of logarithms first we need to restrict to the valid zone, so we will
restrict to the value zone in such a way that my a is between 0 and 1 open interval or my a > 1,
then because the logarithmic function is always defined on the positive side that means the
argument that is applied to logarithmic function is positive, so my M and N are actually the
arguments for the logarithmic function, so they are always positive.

And here is one more thing that is some are real number r is given to you. If all these conditions
are satisfied then there are 3 laws, we will see each of them and we will try to prove each of them.
So, first law is, logarithm of a product of two positive numbers is sum of the logarithms, verbally
you can state this as logarithm to the base a of product of two numbers positive numbers is nothing
but the sum of the logarithms of the individual numbers.

In a similar manner if you go ahead and do some simplifications you can also come up with a
second law that is, logarithm of a quotient of two positive numbers is nothing but difference
between the logarithms of those two numbers. In a similar manner this is not a new law but we
will state it for the sake of clarity that logarithm of reciprocal of a number is nothing but negative
of a logarithm of the original number.

And the fourth one which uses this number r, which we have defined here and remember this is
any real number, then the logarithm of M raised to r or any number raised to r any positive number
is to r is nothing but r times logarithm of that number. So, when many astronomist where doing
some calculations and they wanted to do the product of the two distances which are very high in
the power of 1032 or something like that. And in that case the multiplication of two numbers
becomes a tedious task, so in order to handle these tasks they have actually invented these
logarithms.

So, if you search on the Google why the logarithm were invented you will come to know about
many references from astronomy where they are successfully using logarithms. And remember
they were doing this in around eighteenth century, so there was an absence of computational
power. So, these laws were helpful and therefore they are governed as laws of logarithms. Now,
let us try to prove each of them, one by one, let us take the first law which is logarithm of the
product of two numbers positive numbers is equal to logarithm of is equal to some of the logarithm
of these two numbers.

(Refer Slide Time: 07:32)

So, in order to prove this let us put A = log 𝑎 𝑀 and B is equal to log to the base B=log 𝑎 𝑁. And
now what you do is you actually consider you actually consider A+B, so my A+B is nothing but
log to the base a of Mlog 𝑎 𝑀+log 𝑎 𝑁, so what I will do now is I will simply go back and consider
some properties of logarithms that I have already considered.

So, I will consider these kind of properties and let us see how this property can help me in proving
this particular identity. So, I have used that particular property and let us say I have raised this as
𝑎 𝐴+𝐵 . In this case, what I am using? I am using actually an exponential function is inverse of
logarithmic function, so left hand side I have raised to the power a, so naturally right hand side
will also be raised to the power a, so I will get this+log 𝑎 𝑁, no confusion here.

Now, you can actually see this particular thing when I am looking at this particular thing what I
am getting over here is a raised to 𝑎log𝑎 𝑁 . So, now what is this actually, if you look at our
definition of a raised to log to the base a of M, you will get this to be equal to M and you will get
this a raised to 𝑎 log𝑎 𝑁 = 𝑁, so you got MN.

So, now what you got here is MN and 𝑎 𝐴+𝐵 is MN, so what I have got here is 𝑎 𝐴+𝐵 = 𝑀𝑁. Now,
if you look at what you want to prove, if you want to prove the left hand side is log to the base a
of MN, so how will I get that? Again use the similar property which is given here and by using
this property take logarithm on both sides. So, you take log to the base a of a raised to A+B which
is equal to log to the base a of MN.

Now, my claim is we have proved this result, how? Because log 𝑎 𝑎𝐵 is nothing but A+B and I am
saying A+B=log 𝑎 𝑀𝑁, now what is a? Just substitute what we have put A as, log to the base a of
M and B as log 𝑎 𝑁. Therefore, I can rewrite this as log to the base a of M+log to the base a of N
is equal to log to the base a of MN. Clear.

So, my first result is proved first law is proof, now you can easily guess what modification do I
need to make for proving the second law. So, if you look at the second law that is log to the base
a of M upon N is log to the base a of M-log to the base a of N. So, if you want to prove this what
modifications you need to do? You simply use this A and B, there is no change in this only thing
that you will have is A-B over here.

If you have A-B over here all the things correspondingly all the things will change and instead of
𝑀
MN what you will get is 𝑁 , rest of the things are just same you can practice it as an exercise and

therefore you will get a similar result which says log 𝑎 𝑀log to the base a of a raise to M a raised
𝑀
to A-B is equal to log 𝑎 𝑁 . And you are done, then you will again apply the you again use the

inverse function property of logarithms and cancel with this off and you will get the second result.

(Refer Slide Time: 13:29)

So, I have not doing it properly, but you can easily derive it from it. So, let us write the second
result, this is the first result, first law of logarithm, second law of logarithm actually follows by
𝑀
replacing+sign with a-sign. So, log 𝑎 𝑀 − log 𝑎 𝑀 = log 𝑎 𝑁 .

Now, you look at the third result which actually talks about log to the base a of 1 upon N. now, in
this case you simply apply the second rule and you simply apply the second law logarithm that we
have just proved and where the M is equal to 1. So, naturally I will get log to the base a of 1-log
to the base a of N. Now, what is log to the base a of 1? We have already seen that 1, 0 is on the
graph of a log, so based on this particular property we can easily conclude that this particular thing
is going to be 0, so my final answer should be-log to the base a of N, that is all.

So, on let us look at what is the fourth result, log 𝑎 𝑀𝑟 = 𝑟 log 𝑎 𝑀. How will you prove this? Again
you will apply the same modus operandi, that is you will first isolate some term and then you look
at the term.
(Refer Slide Time: 15:33)

So, let us first look at this r belonging to set of natural numbers. If r belong to set of natural
numbers, how will you proceed? The answer is very easy, what is set of natural numbers? In our
case in our course we have defined set of natural numbers to be equal to 0, 1, 2 and so on. So, in
this case our if r belongs to this particular set then you can easily see that I can write this log of M
raised to r as log of log to the base a of M into M into M this is done r times, this is done r times
and then I can apply the law of or the multiplication rule of the logarithm which is this.

And I can simply get this as log 𝑎 𝑀𝑟 = log 𝑎 (𝑀 … . . 𝑀)[𝑟 𝑡𝑖𝑚𝑒𝑠] = 𝑟 log 𝑎 𝑀. And therefore you
will get the answer to be equal to this is equal to log to the base a of M r times. Now, if r is set of
rational numbers, this is a situation becomes tricky still can be managed, but this will not prove,
we will assume for the sake of convenience and naturally the proof or the answer lies when you
study calculus when r belongs to set of real numbers, you can actually construct a sequence of
rationals which will converge to real number.

So, what I have done here is I have partially proved this is partially proved the law of logarithm
when r belongs to set of natural numbers. When you study the math 2, or math for data science 2,
in that you will come to know how to handle these particular objects. But for us what is important
is if you give me log 𝑎 𝑀𝜋 = 𝜋 log 𝑎 𝑀.

This is this particular property of log will be very handy while solving the problems, you imagine
an irrational number which was in the index of some positive number is taken into the
multiplication with respect to log, so this simplifies the calculations significantly and as I already
mentioned when we discuss exponential function how the number e has arrived natural exponential
function how the number e has arised, the same logic applies when we will prove the result for a
set of real numbers. So, let us not get into those details right now, but for our purposes this law is
very crucial and we will use this law left and right.
 Mathematics for Data Science
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture No. 56
Logarithmic Functions: Applications

(Refer Slide Time: 00:15)

So, let us now go ahead and use these laws of logarithms and try to see some simple problems,
how the problems can be simplified using logs or how the problems can be made complicated
using logs.

(Refer Slide Time: 0:31)

So, let us see the first problem, it looks an ugly sum or ugly product and we have taken log and
how the process is simplified when you use the properties of logarithm that you have studied just
now or the laws of logarithm that you have studied just now. So, let us go ahead and do that.

So, this particular thing let me write this to be equal to, so first you identify or isolate the terms
that you identify can be separated. So, first term that I can separate out this is x cube, the second
term that I can separate out is this square root and the third term is numerator. So, essentially this
particular term if you look at can be split into 3 components, so I want end result that I want to get
after simplification should have 3 components.

𝑀
So, first I will apply the quotient rule that is log 𝑎 𝑁 , so in this case this fetches me
1
log 𝑎 𝑥 3 . (𝑥 2 + 1)2 − log 𝑎 (𝑥 + 3)4 . So, here I have not used any other rule so this is simply the
quotient rule that I have used that is the second rule that we have derived.

Let us go ahead, and see what we can do with the first term that is this term. So, now you can see
is simply see these terms can be characterize into two terms this is the first term and this is the
1
second term. So, I can write this particular thing aslog 𝑎 𝑥 3 + log 𝑎 (𝑥 2 + 1)2 . So, to identify the
raise to half is inside the log I have put it this way then minus you look at this term again.
Now in this term something is raise to the power 4. Do I have any rule for indices of the law,
indices within the argument of logarithm? Yes, we have just now proved it for set of natural
numbers. So, you can use this that rule and say that this is equal to −4log 𝑎 (𝑥 + 3).

(Refer Slide Time: 3:20)

Let us go ahead and do a similar thing for the other two terms that are listed here then we will get
the final answer that is 3log 𝑎 (𝑥 + 3). 3 times log to the base a of x plus half times here I am using
it for rational numbers which I have not proved 0.5log 𝑎 (𝑥 2 + 1) − 4log 𝑎 (𝑥 + 3). So, this is how
I can simplify. Now in general, when you study logarithms people generally get confused between
the product becoming the sum. So, here we have some terms like this. Now these terms cannot be
handled with log.
(Refer Slide Time: 04:16)

So, right now let me give you a note of caution or a warning so to speak that is let me write this as
warning. Generally though it is obvious but while doing the calculation people used this rule;
log 𝑎 𝑀 + 𝑁 ≠ log 𝑎 𝑀 + log 𝑎 𝑁. So, this is what people use and this try to solve the problem so
that they think they will simplify but remember this is not equal to that, why? Because we have
just now proved that this is nothing but log 𝑎 𝑀𝑁. So, these two things are different.

In a similar manner you can have a quotient rule that is log 𝑎 𝑀 − 𝑁 ≠ log 𝑎 𝑀 − log 𝑎 𝑁. Because
𝑀
this is actually equal to log 𝑎 𝑁 . So, just remember this warning because generally in the when you

are in the fighting spirit you are trying to solve the problem you tend to make these mistakes and
which will ruin your entire answer. So, this is with extra star marked I am emphasizing that these
two are not equal.

Now let us try to see how we can simplify, sorry we have here seen how we can simplify our life
using logarithms where everything is now almost linear terms except for this quadratic term. Now
the next question that can be asked is can you combine using logs? The answer is yes, so if you do
not want to see such a big expression or you want to have a nice compact expression, the question
is can you combine? The answer is yes, and now let us handle the terms one by one and merge the
terms.
So, first term let us take these first two terms, what are these two terms? One is 2 log 𝑎 𝑥. So, you
have already seen log 𝑎 𝑥 𝑟 = 𝑟 log 𝑎 𝑥. Apply that in reverse so you will get this particular term as
log 𝑎 𝑥 2 , do not stop there you just apply the product rule now. 2log 𝑎 𝑥 + log 𝑎 9can actually be
merged as we can use this rule and say this is equal to 9𝑥 2 .

Now let us look at the next term which is this, next two terms in fact and there is negative sign so
𝑥 2 +1
naturally a quotient rule will come and you will have something like log 𝑎 . Can you combine
5
9𝑥 2 (𝑥 2 +1)
these two? Again apply the product rule and you will get this to be equal to log 𝑎 . So, this
5

is how you can simplify your life while studying logarithms by giving a combined expression.

So, you can use this to simplify your life give a long expression with positive signs when it will
help when you have lot of large numbers to be calculated. You can also combine the logarithmic
terms and combine the expression in a compact form when this will help when you have lot of
small-small terms that are unnecessarily occupying the space. So, when extremely large terms the
simplification will help when extremely small terms will come the matter will be simplified by
combining the terms. So, these are two simple avenues where you can actually do something.

(Refer Slide Time: 08:41)

So, I have already warned you about this log 𝑎 𝑀 + 𝑁, this is nothing you cannot get anything out
of this, this is nothing. In a similar manner log 𝑎 𝑀 + 𝑁 you cannot use reverse exponential or any
other form to get something out of this while solving the problems of logarithms. So, just whenever
these such terms come be aware and do not apply the things blindly.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology Madras
Lecture No. 57
Logarithm Function: Properties -2

(Refer Slide Time: 00:14)

Now, we already know some facts, but it is better to revise them when we are actually handling
the problems. So, here is one such theorem that says that for a between 0 and 1, open interval 0
and 1 and for a > 1 and M and N > 0. So, these are essentially taking care of the conditions that we
are in the valid domain of logarithm.

Then, if M = N we know that log 𝑎 𝑀 = log 𝑎 𝑁. How this is derived? Because if you recollect we
have already proved that our function logarithmic function is one to one function that means for
every element in the domain there exists a single element in the co-domain or the range.

So, based on that you can actually see that if M = N log 𝑎 𝑀 = log 𝑎 𝑁. Then, another thing that we
want to see is two important values of logarithms. That is log, whenever we are talking about log
to the base a, this a is typically limited to or restricted to when we are solving any engineering
problem or any any stuff that involves some kind of engineering discipline, we generally restrict
to two numbers that are e that is Euler's constant, which we have mentioned and 10.
So, when whenever you are talking about these two numbers when you are talking about log 𝑒 𝑀
we will give a notation ln and when we are talking about log to the base 10, we will give a notation
in log without any base. Why we are doing this? These two have special names also natural
logarithm and common logarithm. Now, this natural logarithm is as I already mentioned, it comes
naturally in the theory of calculus and therefore it has some special value and if you are using a
scientific calculator, then you will identify this natural logarithm as ln.

So, they are there are specific provisions given in scientific calculator because it is a natural
logarithm and then if you look at the scientific calculator you can open on your computer as well,
you will have some symbol of this kind, log this is log to the base 10 which represents our decimal
system. Because in decimal system, everything is given in powers of 10. So, therefore this is called
a common logarithm because we commonly used our decimal system and this is called a natural
logarithm.

So, these are the two important logarithms that we can study in the entire life. Now, the problem
that comes is why only these two why not others. So, here is my bold claim that only these two
logarithms will suffice for studying logarithms and why?

(Refer Slide Time: 03:53)

The answer to why is given by this particular theorem that is change of base rule. What let us,
understand, what does this rule says, this is also called another law of logarithms. So, if you are
handling x that is > 0 and if you are a is between 0 and 1 or if you are a is > 1. Remember this is a
base of a logarithm. So, I am talking about change of base. So, I will talk about another base which
is b, which is between 0 and 1 and b is > 1 this is called, this will be called as old base and this
will be called as new base.

Now, if I want all the calculations to be done with respect to new base. If I want all the calculations
to be done with respect to new base, then this theorem gives the answer, how should I go about
this. So, the answer is for any 𝑥 > 0 you consider log to the base a of x then what do you do, choice
the new base that you want simply write the argument of the function over here in the numerator
and the base in the denominator and write logs with appropriate base that you prefer.

So, in particular if you give me if this theorem is valid, we will prove this theorem is valid if this
theorem is valid, then what we are actually talking about is you give me logarithm to any base, I
will convert that logarithm into logarithm to the base 10 or logarithm to the base e and I will
compute accordingly, that is the beauty of this theorem and therefore we have two important
logarithms. So, we will stick to only those two important logarithms one is to the base e which is
called natural logarithm. Another one is to the base 10, which is called common logarithm.

So, therefore, with this assumption, even our scientific calculators are designed to have only two
keys that are ln and log they do not generally talk about log to the any base. If you have advanced
scientific calculator, then it may talk about log to any base, but these two logarithms will suffice
in general. Because what you are actually talking about is simple change in the basis is just a
multiplication by a constant that log that constant will be given by this particular number, 1 upon
this particular number. So, let us go and start proving this result.
(Refer Slide Time: 06:51)

So, for proving it is very easy if you understand the basic logic in all these proofs for proving laws
of all laws of logarithm, what we are using is a key factor that is exponential function is inverse of
logarithmic function that is all. So, in particular if I want to prove this, let us say M = log 𝑎 𝑥 and
N =log 𝑏 𝑥, R = log 𝑎 𝑏.

So, what I am doing is I am mapping all these terms in terms of M, N so this particular term is N
and this particular term is R. Now, because it is this, you look at the other description of logarithmic
function. So, when you are talking about logarithmic function you are basically asking a question,
if I have been given a base a, to what power I should raise this base a so that I will get x this is the
question that we ask and answer to that question is given by M. So, essentially this means if I apply
an exponential function, which is a raise to M, 𝑎log𝑎 𝑥 , then I should get back a raise to M = x this
is what I should get back.

And in a similar manner I will use this and I will I will use it for the second term and I will get
𝑏 𝑁 = x and I will get 𝑏 𝑅 = a. So, now what is happening is, if your number is a can be written in
the form of 𝑏 𝑅 , if you are number a can be written in the form of 𝑏 𝑅 , then you can as well put this
particular thing into this expression, that means I can the write a raise to M as 𝑏 𝑅𝑀 = x that is
justified.

Now, what you have actually done you actually started with what. Let us, simplify this and this is
actually equal to 𝑏 𝑅𝑀 = x. Now, based on our understanding you simply hit this function with
log 𝑏 𝑏 𝑅𝑀 = log 𝑏 𝑥, no confusion in this but this is actually inverse of this log function and is the
exponential unction.

So, I will get back RM= log 𝑏 𝑥. Let us, substitute what is R and M and then we will resolve. So,
this will happen if and only if, what is R? R is log 𝑏 𝑎 and this is log 𝑎 𝑀 is log to the base a of x
solve. Let me erase it and rewrite it again. Which = log 𝑏 𝑥. So, I am I have justified what I stated.

So, when I want to switch from log to the base a or log to the base b, it is simply a multiplication
by a constant that is log 𝑏 𝑎and therefore, our result is actually proved and what is what is our result
that my log 𝑎 𝑥 is actually log 𝑏 𝑥which goes in the numerator and log 𝑏 𝑎, which is a proportionality
constant comes in the denominator, that is all. So, now you forget about all other bases and you
simply try working with natural logarithm or common logarithm that is the key idea that we will
follow. So, let us use this fact and try to prove certain things like this.

(Refer Slide Time: 11:58)

For example. Now, if you have been asked some question like, what is log 5 89? So, you do not
have to go into much detail that what is log to the base 5? You can simply use natural logarithm
ln 89
and use the change of base formula and get the answer to be use your calculator it has a natural
ln 5

key, which is ln, compute it you will get the answer to be equal to 2.78.

Somebody gives you any absurd number some irrational number log √2 √5. Still you do not have
ln √5
to worry just apply ln √5, ln √2 change of base formula you will get ln that is this thing is going
√2
to the numerator argument is going to the numerator base is going to the denominator and you
know ln is nothing but natural base log with natural base. So, this will become 2.32 that is all. So
again, you can use scientific calculator and get the answer. So, this is how they were calculation
simplifies no matter what base is given to you, you can easily solve all the problems.

𝑥 ln 𝑥
Now, sometimes people confuse with this kind of identity ln 𝑎 they simply write it as ln 𝑎, which

is not true. So, just this is just a warning that I want to give in particular. So, this identity is not at
all true. And therefore, you have to be careful while solving the problems. See in the verge of
solving the problems you made tend to do these kind of mistakes, which will ruin your entire
answer. So, in the next video, we will come up with better versions of graphing the logarithmic
function to any base using natural logs, or some other variations. Right now we will stop here.
 Mathematics for Data Science 1
Professor. Neelesh S Upadhye
Department of Mathematics
Indian Institute of Technology, Madras
Lecture No. 58
Logarithmic Equations

(Refer Slide Time: 0:14)

Hello students, in this video what we are going to do is, we are trying to look at the logarithmic
function and how to solve equations using logarithmic functions. So, this is the goal of this
particular lecture, so let us first get a simple πcture, we have already seen how to plot a logarithmic
function. Suppose now you have been asked to plot a graph of a function which is log 2 𝑥.

If you have a graphing calculator, sometimes the old version of graphing calculators do not allow
log 2 𝑥 to be taken and in that case what you have is either log10 𝑥or log 𝑒 𝑥which we have denoted
by ln 𝑥, this is not e, ln 𝑥. So, only these two things are available to you and you can plot these
two things, then can you plot the function log to the base 2 of 𝑥? This is what the first exercise that
we will do when we try to solve the problems, so f(𝑥) is log to the base 2 of 𝑥.

Now using the change of base formula which we derived in the last class, you can easily convert
this function × a function with log to the base 10 or log to the base e. For convenience I will choose
ln 𝑥
log to the base e. So in this case, I can simply convert using my change of base formula, this is ln 2

2. That simply means, what I am doing is, I actually need to plot ln 𝑥 and scale it appropriately
 1 1
with a constant which is ln 2. We can easily evaluate the value of ln 2 using any calculator and this

is just multiplied with ln 𝑥.

So, the graph will more or less have similar features of ln 𝑥, only thing is it is scaled appropriately.
You can try your hand in plotting this graph. This is a clear cut demonstration of the usability of
the change of base formula. So, now you need not have to bother about, to what base the function
is given. You can simply convert the function × log to the base e or log to the base 10.

This is I have roughly, I have already told this is called common logarithm and this is called natural
logarithm. That is why the name ln and once again I reiterate ln means this is log 𝑒 𝑥. I will differ
from this notation and I will use this notation for convenience and it is a standard convention in
mathematics to write this as ln. So, this is a simplest application of graphing any log to the any
base of a function.

(Refer Slide Time: 3:29)

Let us now come to somewhat tricky application of this, that you have to prove that log to the base
2 of π + 1 by log to the base 6 of π is > 2. How will you prove this? Now the key tool while solving
all the logarithmic equations or inequalities is exponentiation. So, we will use that tool over here.
So, let us start and try to solve this problem. First of all, if you notice in this particular problem,
one base is 2, another base is 6, I do not want this. I want everything to the same base.
So, I will use the same principle that I have used here, that is change of base formula. So, let us try
1 1
to take the left hand side, LHS, which is nothing but + log 𝜋, apply the change of base
log2 𝜋 6

log 𝜋
formula. So, log 2 𝜋 can naturally become log𝑒 2 log to the natural base e of π upon log to the natural
𝑒

ln 2
base e of 2, which essentially gets converted × ln 𝜋. This is again simple application of change of

base formula.

ln 6
In a similar manner, I can write this as ln 𝜋l, to be very precise what I did is, I substituted log 2 𝜋 =
ln 𝜋
and because this thing was in the denominator because the original fraction was 1 over this, so
ln 2

this is 1 over that and therefore, it will give rise to this particular number. Fine. So, this is done,
you do not have to worry about this.

(Refer Slide Time: 5:43)

So, now you can see something amazing has happened. The denominator now is actually ln 𝜋,
denominator is same. So, I can rewrite this expression as, let us rightly write this expression as
ln 2+ln 6
. Wonderful. Now do I know something about the laws of logarithm, this is 𝑙𝑜𝑔 𝑎 + 𝑙𝑜𝑔 𝑏,
ln 𝜋
ln 12
we have already solved this. This is 𝑙𝑜𝑔 𝑎𝑏, so I know this is ln(2 × 6) = .
ln 𝜋
 ln 12
Now this is about LHS. So, LHS actually simplify it to . Now the question is whether this thing
ln 𝜋

is > 2, I do not know still. Let us say, assume this thing holds true, then how will you proceed? So,
let us try to do it in this fashion.

(Refer Slide Time: 6:55)

I will again consider this particular inequality and try to prove it, try to see whether this is true or
ln 12
not, l ln 𝜋 > 2 I do not know whether this is true or not. But I know ln 𝜋 is a non-zero number. So,

I can simply take ln 𝜋, is ln π positive or negative? It is positive, so I can simply multiply

throughout by ln π, so ln 12 > 2 × ln π.

Now I have one law again to my aid that is multiplication rule, so log of 𝑥 𝑎 is a×log of 𝑥, that rule
I will use and this in fact will become=ln π2. So, now I am checking ln 12 is > ln π2d or not. This
is the question that we are asking again. So, all these are questions, we have not yet proved
anything. Now both side logs are to the same base, so I can exponentiate this, I can exponentiate
this with e, Euler’s number, Euler’s number.

So, if I exponentiate this, then because exponentiation, the operation of exponentiation is nothing
but applying exponential function to a particular argument that is monotone, it is monotonically
2
increasing. So, I will have inequalities intact, therefore 𝑒 ln 12 ≥ 𝑒 (ln 𝜋) .

So, now you can simply go ahead and try to solve this particular problem. What is 𝑒 ln 12 this
particular thing, 𝑒 ln 12 ? 𝑒 log𝑒 12 , so I already have proved that 𝑎 log𝑎 𝑥 is nothing but 𝑥, this we
already know using our inverse function definition. So, I will apply this definition over here and
therefore, my 𝑒 ln 12 will simply convert to 12 which is >𝜋 2 π or not? Is it true or not, we have to
check.

So, now you look at the way we started, we started with some complicated inequality which is
given here, the inequality, if you include this, some complicated term on the right hand side some
number was there and now we made it tailored to our understanding which is, whether 12 is >=π2
or not? Now how to prove this? Very easy, what is π basically? It is 3.141 something, something.
This number is strictly > 3.15, strictly smaller than 3.15.

So, the π2, the π is smaller than this, therefore, π2 will also be smaller than 3.152. And 3.152 will
not exceed 10, you can check for yourself, it will not exceed 10. Therefore, this inequality which
is under question is certainly true because this π2 will always be less than 10, which is less than 12
and therefore, naturally this inequality holds true. So, I do not need to put a question mark over
here, I do not need to put a question mark over here, and all these inequalities are true.

And therefore, we have proved that this particular inequality in particular is true. This is the way
we will solve a question which is using logarithms. Now you can simply identify what technique
we have used, we have used 3 laws of logarithm; first law of logarithm that we used is change of
base, second law is the multiplication rule, next the multiplication rule repeated again, here
multiplication rule repeated again and then the third law, it is not a law, it is actually the definition
of inverse function that we have used. Using these 3 together we were able to find our answer.
(Refer Slide Time: 12:01)

Let us try to solve some slightly complicated problem like this. Let us say you are asked to find,
solve log to the base 0.5 of 𝑥 = the base 0.5 of 4. So, what do I mean by solve, this is actually solve
for 𝑥. So, you are interested in finding the feasible values of 𝑥 which will satisfy this equation.
Now at the beginning you may be worried about this 0.5 in the denominator because if you
remember for 0 less than a, if a is the base, a less than 1, the behavior of log function was somewhat
different.

Do I really need to worry about it, is the first question? Before worrying about anything, let us try
to simplify this expression, so what is this equation saying, let us write this 2×log to the base 0.5
of 𝑥 = log to the base 0.5 of 4. Now first thing, if you look at the left hand side, is there any rule
that I can apply? Yes, I can apply that power law, power rule. So, I can simply convert this ×
log 0.5 𝑥 2 and then let it be as it is, so it is log 0.5 4.

Now if you look at this particular expression and if you look at the denominator which is 0.5, here
also 0.5, not denominator, base. So, if you look at this base, you can easily say that the base is
common. Therefore, the exponentiation trick which we have floated in the last class that is 𝑎log𝑎 𝑥
is 𝑥, so this trick will work. And therefore, I will exponentiate this with respect to 0.5. So, I will
2
write 0.5log0.5 𝑥 = 0.5log0.5 4 . What does this mean?

(Refer Slide Time: 14:31)

This simply means I will get here 𝑥2 = 4, 𝑥2 = 4 and now because 𝑥 = 4, based on my knowledge
about the quadratic function I know 𝑥2 - 4 = 0 has two roots, that is 𝑥 = + or - 2, these are the two
roots that are available. Now that means I have two solutions to this particular problem 𝑥 = - 2 and
𝑥 = + 2.

The next question that you should ask is are these both solutions feasible when I substitute them ×
this expression? So, what is log to the base 0.5 of - 2? When you look at the logarithmic function,
it is defined only on the positive real line, it is not defined on negative real line. So, log to the base
0.5 of - 2 is indeterminate, you cannot determine the value, the function is not defined, it is outside
the domain. So, this value you can easily chuck off.
And therefore, your solution, solution to your problem is 𝑥 = 2. This is the solution for the
logarithmic equation which we are finding, so solve for 𝑥, the answer will be 𝑥 = 2. You can simply
substitute it here and check, you put it to 22 is 4 and log to the base 0.5 of 4 = the base 0.5 of 4.
What if you put - 2? This is not valid. Correct. So, this way you have to verify once you get the
final answer.

(Refer Slide Time: 16:22)

Let us handle somewhat more difficult problem which is again going in a similar line but
exponentiation will again help you, but it will reveal some important traits over here. So, let us
look at this particular problem where the LHS is log to the base 8 of 𝑥 + 1 and log to the base 8 of
𝑥 - 1. Let us try to simplify, let us start with LHS. We want to solve for 𝑥, so solving, taking LHS
will not help. So, let us take the entire, entire thing that is log to the base 8 of 𝑥 + 1 + log to the
base 8 of 𝑥 - 1 = 1.

It is important to write the equation as it is, because if you write the equation as it is, you will
understand the nitigrities of this equation. So, it is good practice to write once, repeat once
whatever is written there, therefore I am writing this. This, do I know any rule, any law of logarithm
which will help me to simplify this? Yes, I know multiplication rule, log of m + log of n = log of
mn. So, I will apply that rule and I will get log to the base 8 of 𝑥 + 1 × 𝑥 - 1 = 1.

Now what can I do? What is the way to simplify? Now I want to get rid of factor of law, so how
will I get rid of this log factor? I will exponentiate. So, what I will do is, I will to the, what is the
base 8, so 8log8 of this particular factor, I can rewrite this as 𝑥2 - 1, which is easy for me to do
and then this = 81. Remember if you do not write this step, you may miss out on this, you may
write this to be=1. So, just write all these things.

Then 8 log to the base 8 will vanish and therefore you will get 𝑥 2 − 1 = 8. Do I know something
like this? We can, I know that 𝑥 2 − 9 = 0. That means 𝑥 = + or - 3 are the possible solutions. Now
is any, so in the earlier case when 𝑥 = + or - 3 were the solutions is there any, so for example here
in this case - 2 was eliminated. So, here also you need to do a similar check, if you put 𝑥 = + 3, if
you put 𝑥 = + 3, then what will happen?

(Refer Slide Time: 19:25)

This 3 + 1, 4 and - 3 - 1, sorry + 3 - 1 which is 2 and therefore, this is 4, and therefore it is a valid
expression. So, it is in the domain. If you put 𝑥 = - 3 and - 3 + is - 2. This is not a log, what happens
here - 3 - is - 4, which is also not a log for putting × log. So, - 3 cannot be in the domain.

So, only thing that is possible is 𝑥 = 3 should be there in the domain, again when you have solved
the equation, it is better to verify, so you write here log 8 3 + 1 which is 4 and 3 - 1 which is 2 = 1.
This is valid, so you have good answer. So, 𝑥 = 3 is the answer to this question. Let us move ahead
and solve one question which has different bases.
(Refer Slide Time: 20:32)

So, here is a question which is log to the base 3 of 𝑥 + log to the base 4 of 𝑥 = 4. Now what can
you do? Because you cannot use your trick of exponentiation, because this bases are different. So,
what should you do? The first thing is to make the bases equal, how will you make it? One formula
is change of base formula, so you apply change of base formula, so as I mentioned it is better to
write the expression once more, log to the base 3 of 𝑥 + log to the base 4 of 𝑥 = 4.

ln 𝑥
If I want to apply change of base formula, I will simply use this as log to the natural base, so ln 3+
ln 𝑥 1 1
= 4. So, ln 𝑥 is taken in common and therefore I have ln 3+ ln 4 = 4. Now you can notice that
ln 4

this particular within the brackets is a number which is non-zero.

(Refer Slide Time: 22:13)

Therefore, I can actually take this number on the other side. So, let me simplify this, ln 𝑥 = 4 times,
1 1
when I take this on the other side it will be a reciprocal of that, so ln 3+ ln 4, this is what it will be.

So, can I simplify this? Yes, I can simplify this further, which will give me 4 times, so ln 3 + ln 4
will be in denominator and in the numerator it will give me ln 3×ln 4. Fine.

Then I need to do something which is let us say, I will do it in this fashion, ln 3 × ln 4 cannot be
combined × anything, they will remain as it is. But what can be done is, this is ln 3 × ln 4 upon ln
12, 4 × 3, I have used the multiplication rule. So, let this, all these things remain as it is and because
ln 4
it is ln, I will exponentiate and I will get 𝑥 = 𝑒 4 ×ln 3 ×ln 12.

If you want you can merge these 4 with one of these ln’s and write, with one of these ln’s and write
this as ln 34 or ln 44 , whatever you are comfortable with, but I will leave this as it is. So, this is
the solution and it is a perfectly valid solution. This is how you will solve a problem which has
different bases. So, this is a perfectly valid solution. Let us go ahead and try to solve one more
example, which is, which looks somewhat ambiguous.
(Refer Slide Time: 24:41)

In a sense you have been given ln 𝑥2 = (ln 𝑥) 2. How will you solve this problem? That is a first
question. So, we need to resolve to some methodology, let us look at this particular problem and
try to simplify the things over here. So, here you typically, because this isd, you typically need
some knowledge about quadratic functions in order to solve this problem. Let us try to understand
this.

So, let me write ln 𝑥2 = (ln 𝑥) 2, in such a problem can you figure, the first question is, are the bases
common? Yes, the bases are common, so there is no problem with this. Now what is happening is,
here the term is ln 𝑥, here the term is ln 𝑥2. Can I get the terms which both are in the form of ln 𝑥,
then I can do something with it. So, I will simply ask that question and what comes to my aid is
the multiplication rule or the power law, ln ak is k×ln a. So, 2×ln 𝑥 = (ln 𝑥) 2.

Now comes the real power or the real strength. What is happening? It is 2×ln 𝑥, now the argument
both are same, so you can treat this as composition of functions. So, what is happening is, if you
write ln 𝑥 = t, you are actually talking about 2t = t. So, what I am doing is, I am putting ln 𝑥 = t,
this I am defining. Fine. So, this particular thing is fine.
(Refer Slide Time: 26:55)

So, what I will do here, very easy then, it is 2t -, basically 2t - t2 = 0, take out one t common, so t
× 2 - t = 0 and therefore, either t will be=0 or t will be=2. These are the two possible solutions. So,
now, but what is t? According to our substitution it is ln 𝑥, so that means I am saying ln 𝑥 = 0,
when is ln 𝑥 0? And ln 𝑥 = 2, so when is ln 𝑥=2? So, these are the two questions that we have
asked. So, let us say or.

And then in both cases you have a natural log, is not it? So, I can exponentiate this particular
function, so 𝑒 𝑥 = 𝑒 0 . You already know logarithmic function has a point where it passes through
0 and 𝑒 0 is, 𝑒 ln 𝑥 will be 𝑥 itself, 𝑒 0 will be 1, in this case exponentiating will give me 𝑒 ln 𝑥 = 𝑒 2
by default I wrote, so I will write it as 𝑒 2 . So, that will give me 𝑥 = 𝑒 2 or 𝑥 = 1. This is what the
answer is.
(Refer Slide Time: 28:50)

So, in this case 𝑥 = 1 is one answer or 𝑥 = e2 is another answer. It is good to go to the original
problem and check whether these conditions are satisfied in the original problem or not. So, first
let us check 𝑥 = 1, so if 𝑥 = 1, ln 1 = ln 12. What is ln 1? You know the answer log of 1 is 0 and ln,
so 02 is 0. And then if you put 𝑥 = e2, then ln 𝑒 2 , that is 4 and ln 𝑥 that is ln e2 is 2 and 22 is also 4.

So, we have verified that both these are valid answers. So, what I have said just now is ln 12 is 1,
so it is 0. 1 is 1, so ln 𝑥2 is 0. Similarly, ln 1 is 0, so 02 is 0. So, the first 𝑥 = 1 is the solution. In the
similar manner, you substitute 𝑥 = e2 and you can plot.

Now it is good think if you can use a calculator, graphing calculator like Desmos and plot these
two curves and see how, where they are intersecting. We have actually given the points of
intersection of these two curves. These are the points of intersection of these two curves. You can
verify for yourselves and I will see you in the next class. Thank you.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Indian Institute of Technology, Madras
Lecture 59
Introduction to Graphs
(Refer Slide Time: 0:09)

So, for the next few weeks we will be looking at graphs, so this will be the concluding
section of this maths course, so for the next three weeks we will looking at graph.

(Refer Slide Time: 0:24)

So, we saw graphs in the first week when we were talking about relations, we said that we
can take a relation, so what is the relation? We take two set they could be the same set, we
take the Cartesian product all pairs. So, we take A×B and then we take some subset of that
Cartesian product, so some pairs, out of the total set of pairs and we say that these pairs are
related.

And then we said that we could visualize this, so for example, supposing our set A is a set
of teachers, so let us call it T and the set C is set of courses that are being offered in the
current semester, then we could have a relation which captures, which teachers are teaching
which course.

So, we have T× C as a set of all possible pairs where the first element is the teacher and
the second element is a course and then this relation A, which is a kind of course allocation
relation describes how teachers are assigned to courses in the current semester. So, what
we said was we could draw this relation as a pictorial form, so we could create these nodes
or dots, representing each element in our set.

So since, there are two different types of sets, the set of features instead of courses we write
them in two columns like this, so we have 5 teachers and we have 4 courses and whenever
a teacher is teaching a course, we connect that teacher to the corresponding course node
through an arrow, so in this case, since there are 5 teachers and 4 courses, clearly there
must be at least one course which has been taught by 2 different teachers in this case you
can see that maths is being taught by Sheila and Kumar.

So, what we are going to do in the next three weeks is to look at this picture that we have
drawn earlier just as visualization of a relation, we are going to look at these pictures, these
pictures are called graphs and we are going to formally analyze what we can do with these
graphs.
(Refer Slide Time: 2:19)

So, to begin with a graph consists of a set of nodes or vertices and edges between them. So
typically, a graph therefore, has two components a set of vertices and a set of edges, so
vertices is the plural of vertex. So, we have one vertex, many vertices, so we use
interchangeably, either the node or vertex, as a name of for these elements and then what
edges do is that they connect these vertices. So, notice in this graph that we had drawn
before we had earlier two set, we had the set of teachers and we had the set of courses, and
we were taking a relation which is a subset of T×C.

But once we put it into the graph we lose or we do not really care about the distinction
between T and C, T and C together form the set V of vertices of the corresponding graph,
so now there are nine vertices, there is no real separation between the five that came from
the teacher set and the four that came from the course set and then the edges were those
that the original relations represented namely, those teachers which are teaching the
courses.

But in general, E is just a binary relation on the vertices so, it connect some vertices to
some other vertices. So, this graph, for example, has a direction, right a teacher teaches a
course, we do not have a corresponding edge from a course back to a teacher. So, if v, v′
is an edge it does not necessarily mean that v′, v is an edge in this particular relation it does
not even make sense for v′v to be an edge but there could be other relations where it does
make sense as we will see.

So, this kind of a graph is called a directed graph, so we have an order, we have a starting
vertex for each edge and we have an ending vertex for each edge and you are suppose to
go from the start to the end, you cannot go backwards, so think of it like a one way road.
So, there is a one-way road from the start vertex to the end vertex, so every edge is of this
form, right, it is a pair, so there is a start and there is end, so this is how you should think
of an edge. So pictorially we just draw it as a line with an arrow but mathematically, it is
an element of E × E so it is a pair, so the teacher course graph is directed.

(Refer Slide Time: 4:40)

On the other hand, supposing we are just looking at a bunch of people and we are trying to
capture, which of them are friends of each other. So, this now becomes an undirected
graphs so presumably if Sheila is a friend of Badri, then Badri is also a friend of Sheila,
because friendship is not a one directional thing, you cannot be my friend if I am not your
friend.

So, in this case, we do not have an arrow, we just have pairs, which represent in some sense
a symmetric relation, if you remember we talked about symmetric relations. If a,b is in the
relation, then b,a is also in the relation. So, this is what we have here, we have a symmetric
graph, which says that if v v′is an edge then v′v is also an edge.

So, here we have seen that there are two types of graphs that we can have, both of them are
defined in terms of vertices and edges. In the first case, if the edge relation is not symmetric
then we specifically record the direction and we call it a directed graph. If the edge relation
is symmetric and this happens very often, so it is useful to think of this as a separate case,
of course, we could always represent this by having edges in both directions, nothing to
stop us from creating an edge saying that Sheila is a friend of Badri and Badri is a friend
of Sheila and having an edge going from Sheila to Badri and one going back, but is much
more convenient to just draw a single edge with no arrows indicating that this is symmetric.
So, since this is an important special case, this is usually treated separately in graphs and it
is called an undirected graph.

(Refer Slide Time: 5:58)

So, what can we do with this graph, other than just visualizing the pictorial relationship
between the people? Supposing, we have a situation where Priya needs some help and it
turns out that actually Radhika is in a position to provide this help, but as you can see from
the graph, there is no direct connection between Priya and Radhika, so Priya and Radhika
are not friends, so Priya may not even be aware of the fact that Radhika can be a source of
help.
So, what do we do in real life? In real life when we have a problem we reach out to our
friends and we say I have a problem do you know somebody who could help me. So, in
this case Priya could reach out to her friends who are in this particular graph Aziz and
Feroze and then one of them, presumably can reach out to their friends or both of them and
so on and eventually somebody will hit upon Radhika. So, one possible scenario is that
Priya told Aziz about her problem, Aziz told Badri about this problem that Priya has and
Badri says ‘Oh, I know that Radhika can solve this problem, so let me put Priya in touch
with Radhika’.

So, what we have constructed through the friend relation is a path. It is a sequence,
connecting Priya to Radhika even though there is no direct relationship between Priya and
Radhika. On the other hand, if Priya had asked Feroze, then Feroze might have propagated
this question to Kumar and independently Kumar is also a friend of Radhika, so Kumar
would have found out the same thing and told Feroze for why does not Priya contact
Radhika. So, this is something that you can do once you have the graphical representation
of the relationship, you can look for these long-distance connections, which are called
paths.

So formally in a graph, a path is a sequence of vertices, you have a starting vertex, say v1
and an ending vertex say vk and what you want to do is go from v1 to vk by following a
sequence of connected edges. So, v1, v2 should be an edge. So, in this case v2 will be a
friend of v1, but then v2, v3 is also an edge, so v3 is a friend of v2 and so on. So, you have
k minus 1 edge connecting these k vertices, so you can go from 1 to k following these edges
and this is called a path.

So, one thing is that there is no description in the previous definition as to whether vi in
the path can be the same as a later vj in the path. In other words, you go to someplace in
the graph and then you come back to that place and then proceed. Now, of course, you can
imagine that this is never necessary.

So here is an example. I want to connect Kumar to Sheila but instead of going directly as I
would here, right, so this would be a direct connection, so let me draw that in a different
color maybe so instead of drawing a direct connection from Kumar to Sheila, I actually
took this roundabout route of going around and going. So technically, in graph theory, this
is not a path.

So, path should not repeat a vertex, if you have a sequence which starts at the vertex and
ends with another vertex and possibly repeats vertices along the way, the graph theoretic
for that is a walk, so walk is a more general type of path, in a path we usually assume that
there are no repeated vertices and if there are no vertices and there are only some n vertices
in the graph, every graph has a finite number of vertices and usually we call it n as the
number, then clearly a path cannot have more than n vertices because if I have n plus 1
vertices in my path some vertices shall repeat.

So, this means also in terms of edges the longest accurate path that is path in the strict sense
can have at most n minus 1 edges. So, we have at most n minus 1 edges in a path, where
the size of v is n.

(Refer Slide Time: 9:44)

So, that was an example that we did in the friend graph, but of course, you can also do
paths in directed graphs. So, let us not look at that previous directed graph which is a bit
boring because there is no way to go from a teacher to a course and then go anywhere, you
just get stuck. So, a more common, directed graph is something that represent say
transportation network.
If you have ever looked at an airline map or railway map, you might find graphical
representations of the routes that the airline or the railway services. So, for instance, here
this might represent an abstract picture of 10 cities which are served by some airline. So,
if we assume that this is super impose roughly on a map of India then v0 somewhere in the
north, so let us assume maybe that v0 represents Delhi and v9 the tenth vertex in this graph
maybe represent some city in the south, let us say it is Madurai.

Now, there are arrows in this graph, indicating that not all flights go between both cities in
both directions, typically between large cities like say between Delhi and Mumbai or
between Bangalore and Chennai if you have a flight going in one direction by an airline
you also have a flight in the reverse direction, so you can assume that if you can go in one
direction, you can also go in another direction. But very often in smaller sectors airlines
might operate these kinds of triangular routes, right.

So there is a route, which serves three cities, but it does not go back and forth between each
pair of cities, it starts in one city goes to the second one, goes to the third one and returns
to the first. So, for instance, in this case if you want to go from v3 to v5 you have to sit
through a halt in v6.

So, now given this graph, the same question that we asked before about how Priya would
discover that Radhika can help her. In this case we might ask a more direct and natural
question saying that I am in Madurai in v9 and is it possible on this airline to travel from
v9 to Delhi which is v0?
(Refer Slide Time: 11:43)

So, I need to find a path and of course, if you look at this picture here is a possible path, I
can go from v9 to v8 and then v5 and then to v7 and then back to v4 and then up to v0. So,
notice that some of these edges we drew with two arrows like this and this. So, this indicates
that it is a directed graph, but we explicitly have an edge in both directions. We have an
edge from v4 to v0 and we have an edge v0 to v4 where as these where there is only one
arrow, one arrowhead is a unidirectional edge, that is I can go from v3 to v6 but I cannot go
back from v6 to v3 in this graph.

Now, this is not the only way to go, obviously, so we could for instance, instead of doing
this, we could, for instance, at the v5 we could have followed this path up to v5, up to there
we have no choice because from v9 we can only go to v8 and from v8 we can only go back
to v9 which is not very useful or we can go on to v5 but at v5 there is an option to go to v3,
right and v3 is connected both ways to v4, so I can come back to v4 so this is another way
to reach v4.

So, there are two ways to get to the v4 either via v3 or via v7 both of them have roughly the
same number of, same number of cities in between so there is not much advantage, but
there are multiple paths and we can discover them.
(Refer Slide Time: 13:02)

So, in graph theory what we say is that a vertex is reachable from another vertex, if we can
find a path, so we say that v is reachable from u, if there is a path from u to v. So, some of
the questions that we might be interested in a graph, of course, the first question is, is a
vertex v reachable from u?

This is the kind of question that we asked just now about Madurai and Delhi or about Priya
and Radhika, so if Radhika can help, is there a way that Priya can find out about it through
her friends networks, so this is a reachability question for a specific pair of vertices.

Now, given that this is possible, you might still want to find out the best possible way to
do it in terms of say the shortest number of flights. So, when you log into something like
MakeMyTrip or any of these travel websites, it will offer you a number of flights direct
flight or one hop flight, a two-hop flight and you might prefer to go by a flight which has
fewer stops, so that you do not have to waste your time waiting while the plane is on the
ground.

So, we might ask for the shortest path, now shortest path for us right now is just in terms
of number of edges or number of intermediate vertices, but later on we will see that we
could also associate some kind of distance or time with each leg and then we could ask for
the shortest path not in terms of the number of hops but in terms of some quantity that we
are measuring as we travel, say the time or the distance.

Now, we could also ask a more general question which is that if I started u were all can I
reach? So, in particular, if I know this, if I know where all I can reach, then I can answer
the first question because if v is one of those vertices that I can reach, then v is reachable
from u, so this is a more general questions then asking whether a specific vertex is
reachable is asking where all can I go from a given starting vertex.

And then we can ask more general questions about the graph as a whole, so is the graph
connected, can I go from everywhere to everywhere? Now, in an undirected graph this
basically means that if I start at one vertex, I can reach every other vertex, in a directed
graph is a little more complicated, I may be able to go from one vertex to every other vertex
but I may not be able to come back. So, I need to go from every vertex to every other
vertex.

(Refer Slide Time: 15:15)

So, let at look at this graph for instance. Supposing, we take this v4 to v3 flight and make
it to one directional flight. Now, is it still possible to go? In the earlier graph actually you
can check that you can go from everywhere to everywhere, because the crucial edges so
we have this section where you can go from everywhere to everywhere, you have this
section which you can go from everywhere to everywhere, you have this section where you
can go from everywhere to everywhere and all these sections are connected by these bi
directional things, so from each of these components you can go to every other component
via v4 and then within that component, you can go around in a circle of three and reach
any place you want.

So, this original graph is definitely connected, now what happens if you break this bi
directional connectivity to v4 by saying by saying you can only go from v4 to v3? You
cannot go back. Now, earlier we could take this as an escape route from v3 to go from this
triangle to any other triangle but now we can still go down from v5 to v7 and then proceed.

So, even if we make v4 to v3 a one directional edge, there is no problem, this graph is
connected in the sense that from every city we can reach every other city. However, if we
take another edge from v4, say v4 to v0 and make that a single direction then we have an
issue, because go we can go from everywhere here, we can go up through that edge, if we
are at the top, we cannot come back down, right there is no way to leave that v0, v1, v2
triangle, because there are no edges leaving them. So, now the graph has become not
connected in one specific way which is the vertices v0, v1, v2 cannot reach the other vertices.

Everything else can reach, so notice that is a symmetric, the other vertices can reach. So,
we are not solving these questions, we are just posing these as typical questions that you
might want to ask once you have a graph presented to you as a representation or some
information so that information, we started with the motivation that it came from a relation,
but it could also be some natural information like this like friendships, or it could be like
airline routes.
(Refer Slide Time: 17:19)

So, to summarize a graph represents a relationship between entities, so in the graph these
entities are represented as nodes or vertices, and the relationship that we are trying to
capture is represented by edges between these and these edges might be directed or
undirected. So as a directed graph, we saw one example, which was the airline route.
Another example involving people can be family relationships. For example, if you have a
group of people and you want to connect people who are parent child. So, if A is a parent
of B, you want to say that A is related to B.

Now, clearly this is asymmetric if A is parent of B, then there is no way that B can be a
parent of A. So, this could be a graph that you might have seen pictorially in the form of
family trees, so people sometimes represent relationships within the family, in terms of a
graph where they draw edges between parents and children and then they have a way of
connecting people who are married together and all that, but a family tree is a kind of a
graph and that would be typically directed because parents’ child is asymmetric relation.

On the other hand, as we saw if you have a friend relation it becomes an undirected graph.
So, we have these two fundamental types of graphs and the problem that we have looked
at right now which is of interest is within the graph to identify paths and through paths, talk
about reachability and connectedness. So, we will explore these problems more
systematically in the lectures to come.
 Mathematics for Data Science
Professor Madhavan Mukund
Indian Institute of Technology, Madras
Lecture 60
Some General Graph Problems
(Refer Slide Time: 0:16)

So, in our first lecture we introduced the concept of a graph, so we said that a graph consists
of a set of vertices and a set of edges, so the edges are just pairs of vertices, so edge relation
is a subset of v×v. So, for example, we had this directed graph on the right which represents
airline roots.

And then we said that a path in a graph is a sequence of edges leading from one vertex to
another vertex without any vertex being repeated in between. So, here we see a path from
v9 to v0, so each edge must be an extension of the previous edge, so we go from v9 to v8,
then we go from v8 to v5 and so on, right.
(Refer Slide Time: 0:55)

And then we talked about reachability saying that we might want to ask whether a vertex
u, starting from a vertex u we can reach a vertex v by finding a path. So, at this point the
only problem that we have really looked at in graphs is reachability and this is really one
problem which we will spend some time on but before we get into more details about how
to calculate reachability in a graph. I would like to show you that graphs have much more
interesting problems than just reachability associated to them. So, having a graph
representation of a problem allows you to deal with very many different scenarios.
(Refer Slide Time: 1:35)

So, let us start with a problem which does not appear to be connected to graphs at all, that
is how to color a map. So, typically when you see a political map of the world or of a
country like India you will see that each political unit has a different color, but not all have
the same have different colors for instance you might find a color that is repeated like you
can see in this map for instance that some colors like light blue and green are repeated in
different places.

So, the rule is that normally when you color some state in the map or a country in a map it
must have a different color from all the countries or states which share a border, so there
is no confusion at the border because one side is colored one way and the other side is
colored the other way.

So, this is the rule for map coloring, so one question that you might ask and at the moment
it seems like an ideal mathematical question is how many colors do we need? Now, clearly
if I have a certain number of states, I can use a different color for every state, so I have an
upper bound, right, if I have say 27 states then I need 27 colors, if I have 27 colors each of
them will get a different color there is no question of two states sharing a boundary having
the same color because no two states are the same color.
But maybe I do not need 27 colors, can I do better than that? So, here is where a graph
comes in. So, how do we create a graph to represent this problem and what is the problem
that we are trying to solve on the graph. So, to create a graph what we do is create these
vertices and what are our vertices in this case? The vertices are the states, okay or if it is a
map of the world or the countries.

Now, what is the edge relation? The edge relation is going to represent when two states
share a border, when they are neighbors and must be colored differently. So, we connect
all states which share a border and then we get these black lines connecting these black
dots. So, for every state there is a black dot and every state is connected to its neighboring
state black dots, this is our underlying graph.

Now, our goal is to associate a color with every state on the map which is the same as
associating a color with every black dot in this graph. So, we start maybe by assigning a
color in this case red to Uttar Pradesh, now the rule for map coloring tells us that if Uttar
Pradesh is red then all the neighboring states must be a different color other than red. So,
anything which is connected to Uttar Pradesh in the graph, any node that is connected to
Uttar Pradesh must have a different color.

So, we start with this color for Uttar Pradesh and then we can start coloring its neighbors,
so for instance we might choose a different color green for Uttarakhand and we might
choose blue in this case for Haryana. So, proceeding in this way we go to the neighbors of
these and we use a different color but notice now that once we go from Haryana to
Rajasthan, we can reuse the color green because green has been used for Uttarakhand and
Uttarakhand and Haryana and Rajasthan do not share a border, so there is no confusion, so
we could reuse a color if it is not being used for any of the neighboring states.

So, we proceed in this way so we could again now reuse red for Punjab because Punjab is
not connected to Uttar Pradesh, remember Uttar Pradesh was originally red, so we could
not take any neighbor of Uttar Pradesh but since Uttar Pradesh was not connected to Punjab
we do not have to worry, we can reuse it but finally now when we come to Himachal, we
have a problem because Himachal is now surrounded by three neighbors which have
already been assigned three different colors, Punjab is red, Haryana is blue and Uttarakhand
is green, so we have to choose a fourth color, say yellow for this.

So, we keep proceeding in this way and we can now whenever we need a new color, we
use it, whenever we can reuse a color, we reuse it. So, we can come up with a more say
you know less expensive coloring in terms of number of colors for this, by coloring these
nodes in this way.

(Refer Slide Time: 5:51)

Now, notice that we do not really need that map anymore once we have constructed this
graph which describes the connectivity of the states in terms of which ones share a border
we could always as well start coloring the graph using this rule, that if I color a node with
one color, I cannot color any of the neighboring nodes, any edges connected to it must lead
to different colors.

Now, here you can see one advantage which is when we are working with the physical map
we have to stare at the borders and it depends on how well the map has been drawn to be
able to distinguish because sometimes we have borders which meet at a corner and
technically across a corner this coloring rule does not apply.

For example, if four states which happens actually in the United States, if four states meet
like this. Then I can actually use the same color, I can use say red, red, blue, blue this is
legal as far as map coloring goes. So, if two states or two countries touch only at one point
then they are not considered to be sharing a body.

So, this might depend on how the border is drawn maybe that is the picture that you see
but actually if you blow it up it actually looks like this, and then I have a problem, because
now I cannot use blue, blue because there is a segment of border which is common to these
two states.

But once I have transferred this information to the graph then I do not have this confusion
anymore and in fact I can take this graph and I can even distort it, right. So, this is the
original graph somewhat faithfully representing the geometry of the underlying country.

(Refer Slide Time: 7:16)

And now I can take some of these nodes which are bunched up and move them far apart so
that I can draw my coloring better. So, this is one advantage of moving to the graph which
is that the graph abstractly represents the relevant information, so we can now work with
whatever format of that information is convenient in the graph without worrying about the
original format in which the information came.

So, here in this case the geography or the geometry of the actual state boundaries is no
longer important, we just need the connectivity saying which states are neighbors of which
state.
(Refer Slide Time: 7:50)

So, abstractly we have transformed our map coloring problem to what is called a graph
coloring problem, so in a graph coloring problem we have a graph which consists of some
vertices and edges as before and separately we have a set of colors and we want to do a
coloring, so what is the coloring?

A coloring is a function which assigns to every vertex a color from the set C and the rule
is that if I have a pair of vertices connected by an edge their end points should have different
colors. So, u, v is an edge, the color of u should be different from the color of v, this is
what graph coloring demands.

And the question that we were asking is given a set a particular graph, if I do not fix the set
of colors in advance, like we were doing there when we did our example, we started with
one color red and then we were forced to we chose another color green and then we were
forced to we choose another color blue, then when we were forced to we chose a fourth
color yellow and so on.

So, if I add colors only as I need them how many colors will I need, right, what is the
minimum number of colors I need for this specific graph? So, it turns out that this problem
has actually been well studied for these graphs which come from maps. So, there is a very
well known theorem which is very hard to prove called the Four Color Theorem which
says that for graphs which are derived in the way we showed from geographical maps, four
colors suffice. So, technically these are what are called planar graphs. So, planar graph is
something where if I draw the edges, right, they will not cross, so this is a planar graph.

If I draw this edge for instance, then these two edges are crossing but this is not necessarily
a problem because I can actually take this edge is crossing, right and I can actually draw it
around. So, this is still a planar graph but now if I try to connect for instance some a third
thing which is outside here and I try to connect it across these then I will have a problem,
right. So, some graphs cannot be drawn on a sheet of paper without edges crossing and
these are called non-planar graphs.

Now, it turns out when you have a map laid out and you start connecting them obviously
the map cannot, you cannot have share a border with something which is far away, so
therefore, a map will always be a planar graph, right. So, not all graphs are planar, so the
question that graph coloring asks is the general case, right.

Now, from our perspective the question is why do we care, I mean map coloring itself
seems to be maybe a particularly specialized application where we want to look at this
coloring problem and translate it to graphs but why should we care in general, how many
colors we need for a graph? So, it turns out that graph coloring actually is very useful in a
number of different cases.
(Refer Slide Time: 10:39)

So, one typical case is in classroom scheduling, so supposing we are running a school and
we need to determine how many classrooms we need in order to run our classes. Now, this
depends on the time table, right. So, we have some time tables and we have some courses
and let us assume that this is a graph, not a chart representing, so this is the time of the day,
right. So, this might be like 9 o'clock, 10 o'clock, 11 o'clock, 12 o'clock, 1 pm, 2 pm, 3 pm.
So, across the day we have these different lecture slots and we have different courses which
occupy different slots in our time table.

Now, clearly if maths is running from 9 to 12 and English starts at 11, then English must
be in a different classroom from maths. Similarly, if history is running from 1 to 3 and
science already started at 12 and goes on till 2, history and science cannot be in the same
classroom. So, if we have overlapping slots then the corresponding classes need different
classrooms.

So, the question is what is the minimum number of classrooms I need in order to run all
these classes without having any scheduling conflicts? Now, like before we could assign a
separate classroom, we could have a separate English classroom, we can have a separate
math classroom, we can have a separate history classroom, a separate science classroom
and the problem is solved but we want to optimize, we do not need to want to necessarily
allocate a separate class, for every classroom for every course.
(Refer Slide Time: 12:01)

So, as before now we can draw a graph, so in this graph we have nodes which are our
courses and now the edge relation represents overlaps. So, an overlap says that these two
courses share a time slot and therefore, they cannot be both scheduled in the same
classroom. So, for us now colors are classrooms, so here is a situation that we have four
different colors assigned to these four different nodes saying that we are going to have
every class running in a different classroom.

But now we observe that maths and history do not overlap. So, since maths and history do
not overlap, using graph coloring we can see that the same color can be assigned to both
the maths node and the history node. So, in this way many scheduling problems can be
actually converted to graph coloring problems.
(Refer Slide Time: 12:50)

Now, here is another problem. Supposing a hotel wants to install security cameras, so they
want to put off cameras in the corridors of a hotel and as you know in many hotels corridors
are very neatly aligned, they are all straight lines but there might be a maze of corridors
which meets at different corners. So, let us assume that if you put a security camera at a
particular corner, it can monitor every corridor that meets at that intersection.

So, now the question is what is the minimum number of cameras that you need to monitor
all the corridors on the floor? So, once again we can go to graph theory, so we can represent
the floor plan of the hotel for that floor as a graph, so here the points of interest are these
intersections because clearly it is to our advantage to put a camera at an intersection
because if I put it at an intersection that camera can monitor multiple corridors and I want
to cover all the corridors.

So, I put vertices and intersections and my edges now connect these intersections, there are
segments of the corridor which run from one intersection to another intersection. And now
my question is one of called, something called a vertex cover which is that I want to choose
a subset of these intersections, such that if I put a camera at this subset then every corridor
segment is covered. So, in graph theory this is called a vertex cover question, so vertex
cover basically says that I want to choose a subset of vertices such that if I choose that
subset then those vertices cover all the edges in my graph.
So, let us look at this graph on the right, so it has 6 vertices named v0 to v5, so maybe this
represents the corridors in our hotel, so maybe I choose to put a camera at v2. So, if I choose
to put a camera at v2, then this covers these 4 corridor segments but it does not cover the
segment from v0 to v1, so I have to put one more camera, I can choose to put it at v0 or at
v1, so let me say I put it at v1 and now I have a vertex cover.

So, my vertex cover is v1 and v2, if I choose v1 and v2 as my locations for my cameras it
covers all the corridors, so this is a problem in graph theory which you can solve
independent of the source. So, this is one motivation but you can come up with other
situations where the solution that you require is a vertex cover.

A similar situation could be for instance if you want to locate ambulances at intersections,
so that they can reach all localities fast, so if you want to cover all localities with
ambulances, where should you place the ambulances in your city map so that every locality
is served within a reasonable amount of time.

(Refer Slide Time: 15:33)

Here is yet another problem. So, supposing there is a school, a famous school of dance and
they are going to put up a show which consists of a number of group dances, so in a group
dance obviously there are a number of dancers who participate but the school as a whole
has over a period of time rehearsed many such dances and some dancers take part in more
than one dance. So, if I look at all the dances that the school could possibly put up there
are overlaps between the dances in terms of which dancers are required.

So, now the problem is to organize a cultural program and in this cultural program because
of costume changes and other constraints we would like each dancer to participate in at
most one dance, so we do not want a dancer to take part in a dance and then to go back
have to change and come back and take part in another dance with a different costume. So,
given that we have some information about the dances and which dancers are required for
each dance, can we come up with a large set of dancer which we can put in this cultural
program, so that no dancer has to dance twice during that program.

So, in this particular case the graph will consist of vertices which represent the dances and
now an edge will represent an overlap in terms of the dance group between two dances, so
if two dances share a common dancer, then we cannot put both dances in the program
because one dancer would have to dance in both dances and that is not allowed by the rules
that we have just stated. So, this is our graph now and now we want to find what is called
an independent set, so we want to find a set of vertices such that there is no edge between
any two vertices in that set.

So, we want to pick a set of dances so that no two dances in the set that we have chosen
for the program share a dancer and therefore, require a dancer to dance twice. So, here is
an example with 8 nodes, so now supposing we pick v2 to be in our independent set, right,
then because I cannot use anything which is connected to it, it means that v6 cannot be part
of my independent set, I can no longer use dance in v6, I cannot use the dance in v3, I cannot
use the dance in v1. So, what can I do? Maybe I can choose v5, now v5 already rules out v6
and v1 which I have already gone but it also rules out v8 but I can now do v7 which has no
further constraints and finally I can do v4.

So, in this particular scenario I could choose actually four of these vertices such that there
is no edge between any of them, so this is what is called an independent set, right. So, the
independent set here, one independent set is v2, v5, v7, v4 of course, there is a symmetric
independent set I can treat the black edges also, black vertices also as an independent set
and so on.
(Refer Slide Time: 18:31)

So, final example of this kinds of problems that we can do with graphs, let us look at a
problem which is called matching. So, supposing we are assigning class projects and the
teacher allows the project to be done by either one person, student individually or by two
people but there is a constraint that if two people participate in a project then they must be
friends because if they do not get along with each other then the project would not get done.

So, let us assume that like we had before we have a graph which describes friendships,
what we want to do is given this graph of friends, we know who's friends with whom, we
want to find a good allocation of groups, right, we want to find pairs but these must be
pairs. So, if I have three people who are all friends of each other a, b and c are all friends
of each other, if I make a, b a pair then b cannot be a partner of c, c has to find a different
partner, c cannot partner with a, c cannot partner with b, right. So, this is what is called a
matching.

So, a matching is a subset of edges which is mutually disjoint, that is if I pick one edge and
I pick another edge they do not share any vertex, right. So, here for instance if I pick this
edge, right then I cannot pick the edges here, here or here because all of them either touch
v0 or they touch v2, so I can touch, I can pick any of these three, okay. So, this is the problem
that we have and this is called a matching.
So, for instance you might ask for what is called a maximal matching? So, maximal
matching like we started this one for instance, right. So, at this point when I have done,
when I have chosen this one, right I rule out some vertices but there are some edges but
there are still some edges which are permitted, so I can pick one of them and create one
more pair. So, for instance I might pick this one but now this rules out this vertex, this edge
and this rules out this edge also because now those edges have common vertices v3 and v4.

So, now at this point all the edges have been ruled out or included and I cannot proceed, so
this is what I call a maximal matching, right. A maximal matching is one which cannot be
extended by adding any more pair without violating some condition.

(Refer Slide Time: 20:48)

So, here for instance I what I drew as a maximal matching but here is another maximal
matching, right. If I take v0, v1 then this has knocked off these edges, right and now if I
take v2, v4 it has knocked off these edges because v4 is connected to both of them. So, I
cannot add any more edges and I am stuck making only two pairs among these six students.
Now, ideally if there were 6 students, I should hopefully be able to make three pairs and
that is what we call a perfect matching.

So, perfect matching is one which is a matching but also connects every vertex in the graph,
so every vertex is part of some pair. So, here is a perfect matching on this graph there are
six edges, I mean six vertices and there are three edges and these three edges are mutually
disjoint. So, this is the third type of, fourth type of problem that we can see.

(Refer Slide Time: 21:35)

So, what we want to emphasize is that graphs are not just about connectivity and
reachability. So, reachability and connectivity are actually very important problems in
graphs but that is not the beginning and end of graphs, there are very many interesting
problems that you can frame once you put your problem in a graph theoretic sense.

So, we saw graph coloring which we saw an example with scheduling, we saw vertex cover
where we saw an example with allocation of say security cameras, we saw this independent
set problem which in our case was about having a maximum number of dances where only
one dancer can only dance in one dance during the program and then we saw this matching
problem of allocating groups within a class.

So, although we will not necessarily look at all these problems in detail in this course it is
important to understand why there is so much emphasis on graphs and procedures and
algorithms involving graphs because the underlying representation of a graph actually is a
very rich representation and by solving problems in this abstract world of graphs you can
actually solve a number of concrete problems in one shot without having to solve them
individually.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Chennai Mathematical Institute
Lecture: 61
Representation of graphs
(Refer Slide Time: 00:14)

So, we have been talking about graphs and problems on graphs, we started with reachability.
And then we talked about very complicated problems like graph colouring, and vertex cover,
and so on. So, now let us get back to reachability, and connectivity and ask a more fundamental
question, which is how do we actually work with these graphs in a mathematical setting. So,
remember that a graph consists of 2 sets, or either a set and a relation, a set of vertices, and an
edge relation. And let us focus on reachability back for now. So, a path in a graph is a sequence
of connected edges. And we say that V is reachable from u, there is a path from u to v.
So, as humans, if we see a graph like this, then what we will do is take this picture and stare at
it, and extract this graph in some sense visually. So, we will take a graph like this. And so, you
can maybe by trial and error, but by exploration, we will do this. So, we can see that there is a
path from V9 to V0.

The problem is that we want to operate on this mathematically, we do not want to have this
picture because there is no way to formally represent this picture and how you operate on this
picture, in terms of a procedure that we can execute. So, how do we represent this picture in a
way that we can compute reachability, for example.

(Refer Slide Time: 01:30)

So, first, we need to represent this graph, in a way which is more amenable to computation than
a picture like the one on the right. So, one way is to use what is called an Adjacency matrix.
So, let us assume that the set of vertices consists of n vertices. So, this is the normal convention
and graph theory that we always use small n to represent the number of vertices, and small m
very often is used to represent the number of edges.

So, in particular, if we say that the vertices are n in number, then we will usually just for
simplicity, number them, and call them 0 to 𝑛 − 1, you can also call them 1 to n. But in
computing, it is very common to start numbering at 0, So, we will call it 0 to n minus 1. And
of course, when we actually have real vertices, like, names of cities, like in this case, Delhi and
Madurai and all that, then we will actually use some kind of a table to map the actual vertex
names to the set.
So, for instance, we might have a table which says a Delhi is V 0 and Madurai is V 9 and so
on. So, Delhi is 0, Madurai is 9, and So, on. So, in this representation, where vertices are now
numbers between 0 and 𝑛 − 1, and edge is a pair of numbers. So, an edge is a pair i,j, where i
and j lie in the range 0 to n minus 1. And we usually assume that we do not have edges like
this. We do not have the so-called self-loop, So, we do not have an edge from i to i.

So, we usually assume that 𝑖 ≠ 𝑗. So, when we have an edge i,j, they both are in the range 0
to n minus 1, because that is our set of vertices, but i is ≠j. So, given this, we now have what is
called an adjacency matrix. An adjacency matrix, we have rows and columns numbered from
0 to n minus 1. And then you put an entry in this matrix, 1, if there is an edge from i to j,
otherwise, it is 0. So, the default is 0, and you put an 1 wherever there is a matrix.

(Refer Slide Time: 03:31)

So, if we look at this graph on the right. Here is the corresponding adjacency matrix. So, first
of all, because we have 10 vertices numbered 0 to 9, we have 10 rows and 10 columns, and the
headers in red and the column headers and the row names on the left are in red, indicating that
this is row I in column j. So, now we take an edge, say, for example, we take an edge from 8
to 5.

So, 8 to 5 says that the row 8, and the row, column 5 must be a 1. So, if I look at row 8, and I
look at column 5, then I get a 1 at this position. So, that is how I fill entries in this matrix. So,
you can check that this matrix represents all the edges in the graph. So, if you take any edge in
this graph.

So, if you go here, for instance, and you look at this graph, and you say, 0 is connected to 1
and 4, will indeed 0 is connected to 1 and to 4. And because the edge, this bi-directional edge
between 0 and 4 is actually 2 edges as we said, So, this is not an undirected graph. It is a
directed graph. So, in this directed graph, we are just using a shortcut to represent it 2 edges by
putting an arrow at both ends, but actually, there is a separate edge from 4 to 0.

So, there is an edge from 0 to 4, and there is an edge from 4 to 0, whereas, there is an edge
from 0 to 1, but there is no corresponding edge here from 1 to 0 because there is no backward
edge on that group. So, this is the adjacency matrix for this directed graph. Now, we can take
the same airline route matrix.
(Refer Slide Time: 05:07)

And supposing we assume that all the routes are actually bi-directional. That is, whenever the
airline flies from one city to another, it also flies back. So, then we do not need to record these
edges anymore. And then it is better, as we said, to take such a graph where the edges are all
symmetric, and explicitly call it an undirected graph, rather than recording arrows in both
directions.

So, we will draw it in this fashion where we just draw an edge as a line connecting these 2
vertices with no arrows. And now if we look at this graph, every time there is an edge from i
to j, there must necessarily be an edge from j to i because it is asymmetric edge relation. And
if you look at the matrix, then if you go across down this main diagonal, and then if you look
at any entry like this entry here, then if we look at the symmetric entry on the other side must
be the same.

If there is an edge from 6 to 3, there must be an edge from 3 to 6. Similarly, there is an edge
from 2 to 1, there must be an edge from 1 to 2. So, this is now our thing, this is an edge from 0
to 4, there must be an edge 4 to 0, there must be an edge 0 to 4. So, an adjacency matrix is very
simple. So, we just create a row and a column for every node in your graph. And then at the
intersection of the corresponding row and a column, if there is no edge, you put a 0, if there is
an edge, you put a 1, that is all there is to it.
(Refer Slide Time: 06:23)

So, now we want to compute with this matrix. So, the whole purpose of doing this is now that
this matrix on the right, for instance, suppose to represent the same picture as this undirected
graph, So, we have thrown away the picture. And now we are looking at only the matrix. So,
in this, if we look at the undirected graph, for instance, then if I want to know the neighbours
of I, that is, which are the vertices which is I is connected by an edge, then I go to the row, for
example, I am looking for neighbours of 6, I go to the row, where 6 is the starting point.

And then I find that there is a 1 entry in column 3 and column 5. So, this says that the neighbours
of 6 are 3 and 5, which if you go up is indeed the case, the neighbours of 6 are 3 and 5. So,
without looking at the picture, I can just, in some sense, mechanically analyze this table, or this
matrix, and get the same information that I would get by looking at the picture. And this is
what we need because there is no way that we can actually design a computational procedure,
which looks at the picture and then makes decisions based on the picture.

So, if you have a directed graph is slightly more complicated. Because in a directed graph,
remember that we have outgoing edges and incoming edges. So, the notion of a neighbour is
slightly more complex. So, we have rows which represent outgoing edges. So, if I take the row
for 6, So, the row for 6 has an entry pointing to 5.
(Refer Slide Time: 07:58)

So, now we are looking at this picture, So, we have an entry point into 5 because there is an
edge from 6 to 5, but the edge from 3 is coming in. It is not an outgoing edge.

(Refer Slide Time: 08:08)

So, if we look at this graph, for the directed case, then we have to look at the column for 6, we
have to see where which things end in 6, So, the column for 6 has an entry in row 3. So, there
is an edge from 3 to 6. So, the rows represent outgoing edges. And the columns represent
incoming edges. Now, in an undirected graph, these are symmetric, if I have an outgoing edge
to j, then i must have an incoming edge from j. So, both of these will have the same information.

So, actually, I can also, find the neighbours of a vertex in an undirected graph by looking at the
column o, it is conventional look at the row I. But the column has the same information. But
in a directed graph, this is different. The outgoing edges are in the row, the incoming edges are
on the column. So, the number of neighbours has a name and graph there, it is called a degree.
So, the degree of a vertex is the number of edges, which are incident on that vertex that is
number of edges, which start from that vertex in an undirected sense.
So, here, for instance, we saw that the degree of 6 is 2, because if I look at the row for 6, or if
I look at the column for 6. So, if I look at the column for 6, also, I find that there are 2 incoming
edges. So, to speak from 3 to 5, 3, and 5, and there are 2 outgoing edges from 6 to 3 and 5, this
undirected. So, there is a uniform notion of degree, whether you are counting edges is coming
in or going out, it does not matter. So, the degree of a vertex is the number of edges, and notice
that each edge must go to a different vertex.

So, if you count the vertex you are starting at there are n minus 1 other vertices. So, the degree
could be 0, in which case this vertex is not connected to anybody. I am a friendless soul, for
example. Or I am in a city which is not served by this airline. So, I do not have any edges. So,
I could have degree 0, or I could at most a degree n minus 1, everybody in the class is my
friend.

So, I am 1 person, and all the n minus 1 other people are my friend. So, I have degree n minus
1, or I have a direct flight to every other city on the network. So, this is the case. So, the degree
is between 0 and, so, 0 less than equal to degree is less than equal to 𝑛 − 1. So, this is something
that you should remember.

(Refer Slide Time: 10:13)

Now, if I have a directed graph, this notion of degree now gets split because I have incoming
edges and outgoing edges. So, typically we will talk about the in-degree and out-degree. So,
the degree of 6 in the undirected setting was 2, because there were 2 edges, but actually, 1 was
pointing out to 5, and 1 was pointing in from 3. So, we can say that the end degree is 1 and the
out-degree is 1.
(Refer Slide Time: 10:33)
So, our goal, as we said, was to compute. So, we want to do something with this graph. So, in
this particular case, how can I use this Adjacency matrix to check whether Delhi which is the
vertex 0 is reachable from Madurai, which is a vertex 9. So, we do the natural thing, which is
we start at 9, and then we see where all we can go.

So, we first mark that from 9, we can go to 9. So, by marking what I mean is, I will now take
the row entry, and I will change the colour from red to blue. So, 9 is now reachable. So, now I
can look at the neighbours of 9, in this particular case, there is only 1 neighbour 8. And if 9 is
reachable, and I can fly from 9 to 8, then 8 is also reachable. So, I would mark 8 as reachable.

Now, what do I do, I have to start from 8 and see where all I can go. So, I Just have to
systematically repeat this procedure. So, I have to systematically mark all the neighbours of
marked vertices. So, 8 was marked. So, now 8 has 3 neighbours 5, 7, and 9. Now, notice that I
do not have to refer to the picture, I am only referring to a table, I just have to look at the row
for 8. And if I look at the row for 8, I know what the neighbours are at 8.

So, I do not have to go back to the picture. And imagine anything, this is just a mechanical
analysis of this table. So, I look at this, and this tells me that I must now mark 5, I must mark
7. And I must mark 9, but 9 is already marked, so, it would not have any effect. So, the next
step is to mark these new guys as also being visited or reachable.

(Refer Slide Time: 12:14)

So, I mark 5, 7 and 9, as also reachable. So, now I have from 9, I can reach 5, 7, and 8, other
than 9 itself. So, now I have not been very careful to keep track of it. But I know that I have
explored the neighbours of 8. But I have now discovered 2 new neighbours, which I could
reach 5 and 7. So, I pick 1 of them say pick 5. So, I look at the neighbours of 5. So, 5 has his
neighbours 3, 6, 7 and 8. So, I have to know Mark 3, 6, 7 and 8 for which 7 and 8 are already
known. So, I would mark 3 and 6.

(Refer Slide Time: 12:47)

Now, once again, I have now as unexplored things I had marked 7, but I have not looked at the
numbers of 7, I have marked now 3 and not looked at it and. So, 5, 8, and 9 have been explored
that is I marked them, and then I also mark their neighbours. So, let me again go to the smallest
unmark things with 3.

(Refer Slide Time: 13:11)

So, I look at 3 and I look at its neighbours. So, the neighbours of 3 are 4, 5, and 6. So, this
means I was marked 4, 5, and 6, of which 5 and 6 were already marked before. So, nothing
happens, but 4 gets marked. And now the smallest unexplored vertex is 4. So, I look at the
outgoing neighbours of 4, and I get 0, 3, and 7, and therefore I must Mark 0, 3, and 7. And, So,
once I have marked 0, 3, and 7, I can stop because this was my target, my target was to find
out whether 0 is reachable from 9.

What I have seen is by systematically marking everything which is reachable in 1 statement, 2

steps and so, on, I have found out a way of reaching 0 from 9. So, I may or may not be able to
reach 1 and 2, I do not need to do it for this particular calculation, if I wanted to find out what
all is reachable from 9, I would continue and see if 1 and 2 get marked. But in this particular
case, my only goal was to find out whether 0 is reachable from 9, I can stop.
(Refer Slide Time: 14:08)

So, the other situation is that it is perhaps not possible to reach 0 from 9 and what will happen
there is that after I mark everything that can be marked, I will find that 0 is still not marked.
So, if at the end of this process, where I have marked all the vertices and all the neighbours of
all the vertices and I cannot mark anything more, and I find that some vertex remains unmarked,
then that vertex is not reachable from where I started.

(Refer Slide Time: 14:34)

So, abstractly, what we said is we mark the starting vertex of the source vertex is reachable,
and we systematically mark neighbours of mark vertices and we stop when the target becomes
marked. So, this is what we computed. And we did this using the matrix that was the important
thing that we did not go back to the picture and try to follow edges on the graph as a picture,
but rather we scan the rows and we did some recolouring of the row headers. And in this
process, we were able to explicitly compute whether we could reach 0 from 9 or not.

So, we had a kind of ad hoc strategy, which said that we will mark some things, and then we
will pick up the smallest thing we have not explored and all that, but we did not systematically
tell, how to do that. So, we actually need to elaborate that strategy more carefully, in order to
get an actual computational procedure out of this.

So, how do you systematically explore the mark neighbours? So, it turns out that there are 2
broad strategies for this. So, one is called breadth-first, which is what we were doing in a sense,
but not really, which is that you propagate the marks and layers. So, we look at things that are
reachable in 1 step. And then from 1 step, we look at things reachable in 2 steps, and so, on.

And the other strategy is called depth-first, which is I find one place, I can go to, there may be
3 places I can go to, but instead of going to the second place, I go to the place I could go to first
and from there, I start exploring further where I can go. So, then I keep going down that path.
And then eventually, I hit a dead-end, saying that there is no more places I can reach. And then
I go back, and I say, Okay, now let me pick up the second place I could have started with and
see where all I can go from there.

So, this is called depth-first search. So, you go as far as you can go in 1 direction, and your
backup. And then you take the second direction back up until you run out of directions. And
you keep doing this at every point. And this is called depth-first search. So, we will study
breadth-first and depth-first search in more detail as we go along.
(Refer Slide Time: 16:28)

But before we do that, let us get back to this notion of how to represent a graph. So, the strategy
that we have seen so, far is to use this adjacency matrix. So, the problem with the adjacency
matrix is that as we have seen before, if you look at the adjacency matrix, here, you will see
that there is a large number of 0s and 0s convey no information to us, it is only the once which
are useful to us. So, the number of once is relatively small compared to the number of 0s. So,
0s are non-edges, and once are edges. And we are only interested only in the edges.

So, the size of this adjacency matrix, if I have n vertices is going to be n squared, regardless of
how many edges I actually have. And now this is not a real problem in the long run, or in the
extreme case, because you could have about n squared edges. So, if you have an undirected
graph, then every pair of vertices will actually be an edge. So, you will have in terms of
undirected edges, you will have n into n minus 1 by 2 because every edge is counted twice. But
if you look at the matrix, it will actually have n into n minus 1 entry, because i j will be in the
entry, and j i will also be in their entry. So, i j and j i represent the same edge. So, the number
of actual edges is half that, but the number of entries is going to be n into 𝑛 − 1.

And of course, it was a directed graph. Also, you have n into n minus 1 should not be by 2. So,
in both cases, you could have about n squared edges, but in typically you will not have n
squared. So, typically a graph will have much fewer than n2 entries. And if you have much
fewer than n2 entries, then it is not clear that having this large matrix is the best way to represent
a graph.

So, the other option is to just directly represent the neighbours of each matrix of each vertex in
a list. So, this is what is called an adjacency list. So, in an adjacency list, what you do is you
write down for each vertex, which are the neighbours of that vertex. So, it is most sensible in
a directed graph.

So, let us look at this directed graph here. So, it says that 0 is connected to 1 and 4. So, again,
0, you put this list 1, 4, similarly, 5 is connected to 3, and 7. So, against 5, you put 3 and 7. So,
for each vertex in your graph, you just record what would have been in the adjacency matrix,
what will be in the 1s in the row for that vertex.

So, if you look at row 5 in the vertex in the adjacency matrix, for this directed graph, it will
have 2 once at 3 and 7, So, instead of writing all the 0s on the other 8 positions, we Just write
the 1 position as 3 and 7. So, this is an adjacency list. So, this is an alternative way of presenting
a graph, and we can work with this representation as well.
(Refer Slide Time: 19:06)
So, on the right-hand side, you see now, the 2 representations for that particular graph we have
been looking at the top is the adjacency matrix for that directed airline graph. And the bottom
is the adjacency list for the directed airline graph. And it is obvious from this picture, that the
adjacency list is typically much more compact in terms of the amount of space that it occupies.

But of course, you have to do different things when you compute with these 2 things. So, for
instance, if I want to check whether a vertex j is a neighbour of vertex I, is there an edge from
I to j, in the adjacency matrix, I just have to look at the cell i,j, and check whether it is 1 or not.

So, assuming that I can look up every entry in the matrix in constant time, in the same amount
of time, then checking with as an edge between i and j is taking the same amount of time for
every i and j. On the other hand, if I want to check in the adjacency lists matrix representation.
Whether, say, for example, this is an edge from 8 to 9, then I have to go to 8. And then I have
to walk down this list. In this case, I have to first check that the first entry is 5, and then I have
to look at 9 and so, on. So, I have to go through the entire list for a given vertex to determine
whether or not there is a neighbour. So, it is a little bit more expensive.

On the other hand, if I want to know all the neighbours of i, then the adjacency list directly
gives it to us. So, if there are 5 neighbours, there are 5 entries anyway, I look at 5 entries, I will
find them if there are 2 end neighbours, I will get 2 entries. On the other hand, in an adjacency
matrix, regardless of how many real neighbours there are, you have to scan the entire row,
because you do not know whether there is a 1 coming up or not.

So, you cannot stop and say okay, after 7, I do not believe there are any more neighbours. So,
you have to go. For example, if you started 8, I cannot go up to this point and say, Okay, I
found 1 neighbour, and there are no more, you have to keep going because you might find a
neighbour at last position.

So, regardless of what is the actual degree or out-degree, in this case, because it is a directed
graph, regardless of the actual degree of the vertex, you have to spend order n time in order to
collect all the neighbours of a given node, in an adjacency list, the time you take at each node
is actually proportional to the degree. Now, in practice, many graphs will have a small degree,
a given node will not be connected to very many other nodes.

So, therefore, if you have a procedure, which is proportional to degree rather than the number
of nodes, it often works much faster. That is why it is useful to have this representation. So,
there are trade-offs. So, it is not always the case that 1 is better than the other. So, typically, an
adjacency list will save space. But there are some situations where it will incur some additional
computation and vice versa. So, if you can make do with an adjacency matrix, and it is very
simple to work with it, but then you have to do the scanning of rows and columns.

On the other hand, with a decency list, some things are not so convenient. For example, imagine
how you would find out where there is an incoming edge in a directed graph, because an
incoming edge will be recorded in the list for the other one, so, you have to go there and look
at that list. So, there are other things that you have to do indirectly, with an adjacency list
representation. But it is important to recognize that there are these 2 different ways of
representing graphs and both of them are used in algorithms.

(Refer Slide Time: 22:15)

So, to summarize, what we have seen is that, we cannot just think of a graph as a picture to
describe a procedure on it, we need a representation. And we need a way of writing it down in
a way that we can manipulate mathematically. And we came up with 2 different
representations.

So, 1 is the adjacency matrix, which is a matrix of n cross n for n vertices, which says that A i
j is 1 if there is an edge from i to j, otherwise, it is 0. And the other 1 which is more compact,
in general, if we have not very many edges is what is called adjacency list, where for each
vertex, we list out the neighbours of that vertex in the list.

And with either 1 of them, we did an example using the adjacency matrix, you can do now
systematic computation. So, it is a systematic exploration of whether or not a vertex V is
reachable from a vertex U. So, we will look at more such things, in particular, we will look at
these 2 strategies which we described mentioned but did not do in detail which is breadth-first
search and depth-first search and then we will also, look at other properties of graphs that you
can compute using these 2 representations.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Chennai Mathematical Institute
Lecture: 62
Breadth-first search
(Refer Slide Time: 00:14)

So, we have been looking at this question of reachability in a graph. So, we said that to find
whether a vertex is reachable for target vertex is reachable from a source vertex, we
systematically explore the graph, we mark the source vertex, and then we go to its neighbours
mark its neighbours and keep doing this systematically until the target becomes marked.
(Refer Slide Time: 00:36)

So, what we saw last time was that, in order to do this as a procedure, we have to have a way
of representing the graph, which is not a picture. So, we came up with these two different
representations. The adjacency matrix has a row and a column for every vertex. So, remember
that we are assuming that our vertices are always numbered 0 to n minus 1 for convenience.

So, we have an entry at row i column j, it indicates whether or not there is an edge from i to j.
So, if it is a 1, it means there is an edge if there is no 1, if it is a 0, it means there is no edge.
And then we observe that this could be a wasteful representation. If there are not that many
edges, a number of entries in this matrix are 0. So, instead, we could Just record an adjacency
list for each vertex, just record the neighbours.
(Refer Slide Time: 01:23)

So, if we go back to this graph that we had just now, So, if we look at V4, for example, So, V4
has an outgoing edge to V0, and outgoing is to V3, it also has an incoming edge from V0 and
V3, and then it has an outgoing edge and incoming it from V7, So, the neighbours of 4 are 0, 3,
and 7.

(Refer Slide Time: 01:40)

So, if you look at this adjacency matrix, it says that in the neighbours of 4, if you look at the
row 4, it has once at positions 0, 3, and 7. And here, it does not have. So, there is 1 entry missing
here. So, it should be 0, 3, and 7. So, this is how we would represent the matrix as either the
graph is either an adjacency matrix or adjacency list. So, what we are going to look at now is
the second part of the story, which is having represented the graph in this way where we can
manipulate it, how do we actually systematically explore the graph.

So, we did a kind of ad hoc exploration last time using the adjacency matrix. But this time, we
want to do it systematically. And there are 2 systematic ways to do this 1 is called breadth-first
and 1 is called depth-first. So, first, we will look at breadth-first.

(Refer Slide Time: 02:33)

So, when we explore a graph, breadth-first, we do it level by level. So, we start with the vertex,
then we identify all the neighbours have this vertex, that is all the edges, which connect this
vertex to its neighbours, we look at those endpoints and say that these are now all reachable
from this vertex, then we go to those neighbours, and look at what is reachable from there. So,
these are things which are 2 levels away. So, 1 level away at the neighbours of the source
vertex, 2 levels away are the neighbours of the neighbours, and so, on.

Now, what happens, of course, is that we might end up with a situation where we start with 5,
and then we identify its neighbours as say, 3, 8, and 7. And then we go to 3, and then we
identify its neighbours as 4, 6, and 1, or not 4 and 6. And now i look at the neighbours of 6.
And it says, oh 5 is a neighbour of 6. But we started with 5. So, we need to record that a vertex
has been visited. And we need to make sure that we do not visit a vertex twice. Otherwise, we
will go round and round this kind of a triangle or a cycle a number of times.

(Refer Slide Time: 03:39)

So, we need to do this visiting and exploring. So, exploring means we go from there to 1 more
level of neighbours. So, we need to do this visiting and exploring exactly 1 per vertex. So, we
need to maintain some information. So, we need to maintain information about which vertices
have been marked as visited, that is, which note vertices have been marked as neighbours of
something we have already seen.

And in the process of exploration, these have been marked, but we still have not looked at their
neighbours. So, there is a subsequent process after visiting a vertex of exploring its neighbours.
So, whether or not such a visited vertex remains to be explored. So, we need to keep 2 things.
Has it been visited, and has it been explored? So, of course, it will be explored only after it is
visited. But if it has been visited and not explored, that means this is some pending work that
we have to do.

(Refer Slide Time: 04:26)

So, as we said, we will always assume that we have n vertices, we call them 0 to n minus 1.
So, here is an undirected graph with 9 vertices, 10 vertices, numbered 0 to 9. So, what we will
do is record the visited information as a function. If you are thinking about it in programming
terms in terms of the computational thinking course, you can think of it as a dictionary whose
keys are 0 to 9. It does not really matter, but it is actually mathematically a function for each
vertex, it tells us true the vertex has been visited or false the vertex has not been visited so, far.

So, initially, we assumed that no vertex has been visited because we have not explored the
graph at all. So, initially visited of V is false for every vertex. And now, we also separately
have to look at the vertices which have been visited that have been marked true by visited, but
which have not been explored. So, exploration means looking at the neighbours of that vertex.
So, we have not yet looked at the neighbours of that vertex. So, we have to keep a record of
this. So, this is a sequence of vertices which have been so, far visited but not yet explored.

And we will keep this in a special kind of sequence called a queue. So, the queue has exactly
the same meaning that you associate with an English queue. So, if you are standing in a line to
buy a ticket, a queue forms you join at the end of the line, the person at the front of the line
buys a ticket and moves away, the next person moves up to the front of the line. And gradually
as you move forward, when you reach the head of the queue, you get your turn. So, the same
way we will maintain these vertices in a queue. As they get visited, they will be added to the
queue, when their turn comes, they will be explored.
So, exploring a vertex technically means the following, we want to look at all the edges which
are outgoing from that vertex. So, we look at every edge i j, which is there in the graph. And if
we have not yet visited j, then we mark j as being visited, and we add j to this queue of
unexplored vertices. So, j has been marked now is visited. So, it is due to be explored when it
is turned comes. So, we put it at the end of the queue.

(Refer Slide Time: 06:26)

So, suppose, we start our BFS from a vertex j. So, what we will do is initially we will set visited
of this vertex to be true, because we start there, So, we have visited that vertex, and now we
have to explore it. So, what we do is we just put it into the queue, because our procedure is
going to be to systematically explore all the vertices which are in the queue. So, we set the
visited of j to true and we add j to this queue. And now, what we do repeatedly is we keep
removing the vertex at the head of the queue. So, this queue, as i said is a line of vertices
waiting to be explored.

So, we pick up the next 1, which is waiting, which is at the head of the queue the front of the
queue, and explore it. And exploring it, as we said, is to check whether its neighbours which
are visited, its neighbours have been visited or not. So, if a neighbour has not been visited, we
mark it as visited and put it in the queue.

So, how do we stop? Well, if there is nothing in the queue to visit anymore, to explore anymore,
this means that we have visited vertices and all the vertices we have visited, we have explored
and there is nothing left to do. So, when the queue becomes empty, this breadth-first search as
it is called BFS terminates.
(Refer Slide Time: 07:47)

So, let us try an exploration of this graph. So, let us assume that we start at this vertex 7. So,
we are going to start with 7. So, as we said, initially, we have set visited to false for all the
vertices. And initially, we have this queue. So, the queue, I am going to assume has the left
side as the front and the right side as the end. So, the head and the tail of the queue, we joined
from the current tail of the queue, and we leave from the head of the queue. So, initially, the
queue is empty, because we have not started anything yet. And initially, everything is
unmarked. So, everything has visited set to false. So, we are starting from 7.

So, the first thing we do is that we mark 7, as visited. So, we mark 7 as visited. And Just to
illustrate, we have also marked it on the graph. And now we have also put 7 in the queue,
saying that 7 needs to be explored. So, we mark 7 as visited and added to the queue. Now, the
real breadth-first search starts, which is you pick up the first element in the queue, and explore
it. So, we pick up the 7, explore it. So, what are the neighbours of 7, the neighbours of 7 are 4,
5 and 8. So, in this process, 4, 5, and 8 get marked. And now they also get added to the queue.

So, I put them in some order, I have just put them in this particular case in ascending order, it
does not matter, i could put it as 8,5,4 and 5,4,8. But is just convenient to put it in some fixed
orders, I always put them from smallest to largest. So, 4, 5, and 8 were neighbours of 7, they
had not been visited. So, now I have marked them as visited and added them to the queue. And
notice the 7 has gone from the queue because 7 has been explored. So, 7 is no longer in the
queue. Now, the first vertex in the queue to be explored is 4. So, the next step is to pick up the
first vertex in the queue and explore it. So, I explore 4.
(Refer Slide Time: 09:35)
So, if i look at 4, it has neighbours, 0 and 3, and 7. But 7 is already been visited. So, i do not
have to do anything about 7, i pick up the 2 which have not been visited, which is 0 and 3, and
set their visited status to true and i add them in the queue in some order. In this particular case,
as I said, I will put 0 before 3 Just because it is a smaller number. So, now i have finished with
4. So, 4 has left the queue. But from the previous rounds 5 and 8 are still pending.

So, 7 added 4, 5, and 8, i finished 4, 5, and 8 are still pending. So, 0 and 3 will have to wait
their turn, they will have to wait until we are finished. So, in some sense, 4, 5, and 8 were 1
level away. So, until I finish all the vertices, which are 1 level away, I am not going to explore
the vertices, which are added at level 2, namely, 0 and 3. So, what I do is I know to pick up 5
and explore it. And in the process, I look at the neighbours of 5, so, there are 3, 7, and 6. But 3
and 7 have already been visited so far, so, we do not have to look at them again. But 6 is new.

So, I marked 6, and I put it in the queue. Similarly, now i will pick up 8. And from 8 again, I
have vertices, which are 5, 7, and 9. But since 5 and 7 were already visited, what is added now
is 9 and 9 gets put into the queue. So, now I finally finished the level 1 vertices in the queue
and have come to the 0 which is added at level 2.

(Refer Slide Time: 11:04)

So, I explored the 0. So, 0 has 3 neighbours, 1, 2, and 4, but 4 is already visited earlier. In fact,
4 is where we came to 0 from in some sense. So, we now mark, 1 and 2. So, now you can see
actually that everything has been marked, and 1 and 2 get into the queue. So, at this point, in
principle, everything has been marked and you could stop.

But this is not how BFS stops, BFS stops by checking that every visited vertex has been
explored. So, what we will do now is we will go to 3 and explore it, but we find that all the
neighbours of 3 are already visited. So, exploration does not add anything to the queue, it only
removes the 3 from the queue.

Next, we explore 6 similarly, all the neighbours of 6 have already been explored. So,
exploration of 6 only remove 6 from the queue does not add anything to the queue does not
mark anything new. Similarly, we have to explore 9 again, nothing new, explore 1, again,
nothing new, explore 2, again, nothing new.

And now the queue is empty, I have run out of work to do. So, I have processed every vertex
that I visited during my breadth-first search. And as a result, I have marked all the visit vertices,
which I could visit starting from 7. In this particular case, you can see obviously, from this
graph, that everything can be reached by some parts of the other from 7. So, everything is
marked as true. So, this is breadth-first search.
(Refer Slide Time: 12:27)

So, now we know for instance, that we can reach 1 from 7 because everything was marked as
true. So, visitor, 1 is true. So, clearly, there was a way to go from 7 to 1. But this information
that we have recorded in this visited vertex does not tell us anything about how to do this. So,
if we have visited have j equal to true, after we have explored breadth-first search for I, we
know that j is reachable from i. But we do not have any information about the path. So, this is
now something that we can fix. So, how did we get to i, j, we reached j, because we explored j
from some k.

So, if we keep track of how we reached each vertex, we can work backward and extract the
path. So, visited j was set to true when we explored some vertex k. So, we can say that the
parent of j, the reason that j got added to the visited set was k. So, now if we go follow back
from j to k, through this parent link, k itself would have been added because of some other
vertex. So, we can look at parent of k.

So, maybe parent of k, some vertex L. So, we go to L, and we look at parent of L, everything
was eventually traced back to the starting vertex. So, that is what marked the first set of vertices
visited. So, in this way, we will walk backward and find the reverse path, which we can then,
of course, enumerate in the forward direction and be done with it.
(Refer Slide Time: 13:51)
So, this can be done along with breadth-first search with no extra cost, except that we record
more information. So, as before we start by setting visited to false, we are again starting with
7. So, that we Just have the same familiar computation to do. But now we have this extra
column called parent, which records the parent that is the vertex from which this merge vertex
was marked as visited. So, as before, we start by marking 7 and adding it to the queue.

And then when we process 7, we add 4, 5, and 8 as marked, but we also set the parent of 4, 5,
and 8 to be 7 to indicate that I came from here. So, let me draw an arrow like this. So, it says
that I came from here, that is what this parent is saying that I came this way. So, whichever
way you want to think of the arrow is going, it is pointing to the vertex which market. Now,
when I process the 4, similarly, it is going to mark 0, and 3, and for 0 and 3.

Now the parent becomes 4 because that is how they got mark 4 and 0 and 3 did not get marked
by 7 they got mark by 4. Same way, when I explore 5, the parent of 6 becomes 5. Same way,
when I explore 8, the parent of 9 becomes 8, because I got to 8 and 9 from 8. And finally, when
I come to 0, then the parents of 1 and 2 becomes 0 because 0 marks them. Notice that 7 does
not have a parent. And that is because we started the search at 7. So, 7 got visited, because we
initiated the breadth-first search at 7 not because of some other vertex marked it.

So, except for the Source vertex, all the other visited vertices will have a parent node marked
in there. And now we can recover this information. So, we come back to the end. And finally,
after we have explored everything we have emptied, the queue and breadth-first search is over.
Now, we can ask for instance, what is the path from 7 to 6. So, the path from 7 to 6, that
breadth-first search discovered, well, I go to 6, and 6 says the parent of 6 is 5. So, 6 says I came
from here. And 5 now says I came from 7.

So, by following these parent licks, and traceback this path from 6 to 7, and I read this path in
reverse and I get 7, 5, 6. So, here is a longer path, what is the path from 7 to 2. So, 2 says I
came from 0, 0 says I came from 4, 4 says i came from 7. So, therefore reading it backward, it
is now 7, 4, 0 and then 2. So, in this way, by keeping this parent information as we are exploring
the graph, we also record the path for every reachable vertex or a path for every reachable
vertex from the source vertex.
(Refer Slide Time: 16:45)

So, how about the distance. So, we have explained that BFS explores neighbours level by level.
So, in some sense, the level of a vertex indicates the earliest that I can get to that vertex from
the source vertex. So, all the vertices which are level 1 could be reached directly in 1 edge from
the source vertex. Anything which is at level 2 is reachable from level 1 but was not reachable
directly. So, I need two edges to reach it, and so on. So, the level gives us some notion of
distance from the source vertex. And can we do that? So, it turns out that we can modify our
breadth-first search.

So, instead of just keeping this true false information for visited, we can replace it by this level
information. So, every vertex which is reachable will have a level, which is 0 for the source
vertex, and it has 1 or more for every other vertex that is reachable. So, since the minimum
level is 0, for any reachable vertex, I can initialize the level to be minus 1.

So, any vertex whose level at the end remains minus 1 is as good as something whose visited
value was false. That is, i never reached it. Otherwise, what I do is I set the level to be 0 for the
source vertex, assuming that I am starting from i. And whenever we visit a vertex j from vertex
k, we already have a level assigned to k.

So, we assign the level of j to be the level of k plus 1. So, it should be clear that because we are
doing this level by level, the length of the shortest path, in terms of number of edges that we
have to travel is given by the level. So, if I look at level of j, I know exactly how many edges I
need to take to get from i to j, there might be longer ways, but there is no shorter way.
(Refer Slide Time: 18:35)
So, here is that breadth-first search again. So, I have now replaced this column visited by this
column level. And as we said before, we put all the levels to be initially minus 1. So, now we
start with our favourite starting point 7. So, we start by setting this level to be 0. So, instead of
setting it to be true, as visited, we set its level to be 0. Now, when we explore 7, its neighbours
get to level 1. So, 4, 5, and 8 are all at level 1. Now, when we explore the neighbour of 4, I get
to 0 and 3. So, these now have level 0032 because the level of 4 was already 1, so, it is 1 plus
1 is 2.
(Refer Slide Time: 19:17)

Now if I explore 5, though I am exploring 5, after 4, the level of 5 was the same as the level of
4. So, what I can reach from 5 is also at level 2, it is not that it is level has increased because I
am exploring it after 4. So, both 4 and 5 were added at the same time. So, they were both at
level 1 I Just happen to have happened to choose to process 4 before 5. So, 6 also becomes
level 2.

Similarly, when I look at 8, 8 was also originally a level 1 vertex, so 9 which is reachable from
it also becomes. So, at this point, if we look, we can see that the 2 vertices which have not yet
been visited, are the ones which have level minus 1. So, that is, of course going to be fixed
next. When i explore 0, so notice that explorer 0 means that 0 has level two. So, these two
become 3, because in order to reach 1 and 2, I have to first go to 0, and 0 was already 2 edges
away. So, this gives us the breadth-first search. And as before, so, this is about paths, but as
before, we could also record the, we have the parent vertices.

(Refer Slide Time: 20:23)

So, you can get the path. But we can also get the shortest distance. So, you can keep only the
level only the parent both whatever you want.

(Refer Slide Time: 20:33)

So, to summarize breadth-first search is a systematic strategy to explore a graph level by level.
So, remember that we said that broadly, the way we explore a graph is to systematically
propagate these marks from a source vertex to the neighbours, to the neighbours, and so, on.
But we need a way to do this. So, that we do not go around and around in cycles. So, we need
a way to do it. So, that we terminate sensibly.
So, breadth-first search is one such strategy. So, what we do is we record the vertices which
have been visited, and we maintain this queue of visited vertices, which are yet to be explored.
So, exploration means explore the neighbours, then we saw that we can add parent information
to recover the path. And we can maintain level information to record the distance, which is the
shortest path in terms of number of edges. And what we will see is that in general, when we
have a graph, we could record more information than just the edge.

For instance, if you had airline time graph, or a railway graph representing the route of railway
network or an airline network, then typically with each edge, you would have associated with
Some number, which we will abstractly call a cost. Now the cost could be a real cost, it could
be the price of a ticket to go from the station to that station, or from this airport at that airport.

But the cost could also be a distance, it could also be how many kilometres this route travels.
Or it could be time. For instance, if you have a train, the same distance could be covered in
different times depending on the quality of the tracks and the number of stops in between.

So, in general, to go from one point to another point, you have to traverse real-time or pay
Some real money, or spend Some distance traveling, say, supposing it is a road network, then
it will give you a measure of how many liters of petrol you would expect to consume complete
this distance.

So, in these kinds of graphs, which are called waited graphs, the shortest paths are no longer
just in terms of the number of edges, we could have a short edge, a single edge, which has a
very high cost. It could be that a direct flight costs much more than a flight which goes by an
intermediate airport because the airline is trying to encourage people to take these flights to
unpopular airports in order to fill the planes. So, it is not always the shortest number of edges
is also the shortest path. So, we will explore this separately when we look at waited graphs.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Depth first search
(Refer Slide Time: 0:14)

So, we had said that there are two systematic ways to explore a graph and we earlier looked at
breadth first search which explore a graph level by level using a queue to maintain the
unexplored vertices as we go along. The other strategy that I would use depth first search.

(Refer Slide Time: 0:31)

So, in depth first search we start from a vertex i and we pick anyone of its neighbor is not been
explored in breadth first search. We pick all its neighbors and put them into a queue, here we
pick any unexplored neighbor j and now what we do is we basically suspend the exploration of i
and start exploring j instead. So then we look at j, we look at the neighbor of j, we pick again
some neighbor of j which has not been explored.

We suspend j and go to the next vertex and we keep doing this until we run out of vertices that
we can reach down this path. And now when I reach a vertex through this process which I cannot
explore any further, I come back along the path I have taken. And find the first point where there
was another choice for an unexplored. So you back trap to the nearest suspended vertex that still
has an unexplored neighbor.

And then you explore that neighbor and so on. So here unlike in breadth first search where we
had to keep track of these vertices which have been visited level by level and then put them into
a queue and then we process this queue in this first in first out manner. So the queue, the vertices
which are added level 1 get processed before the ones at level 2 and they get process before the
ones at level 3 and so on.

In a depth first search if I have walked down some distance then I will come back to the point
where I last stopped and see there was something else I could do from there and then I will walk
back and so on. So I need not a first in first out but what is called a last in first out. So the last
vertex has suspended is the first that I restore and check again. So this is called a stack, so these
are 2 very fundamental ways of organizing sequences. A queue is a first in first out structure and
that is used in breadth first search and a stack as we will see is used in this depth first search.
(Refer Slide Time: 2:24)

So, let us try to explore this depth first search, so for this our stack will again like our queue
grow from left to right, our stack will again grow from left to right. So when I add this thing to
the stack I will put it to the right. And now unlike a queue when I remove things from stack I will
remove it from the right not from the left. Remember in a queue we had a head here and we had
tail here.

So, we put things here and we brought them out here instead here I am going to grow the stack
this way and also remove it from this end. So, that is going to be our strategies so for a change
instead of 7 let suppose we start our depth first search from vertex 4.
(Refer Slide Time: 3:04)

So, we first start as before by marking 4 as visited. And now we have to pick a neighbour of 4
and suspend 4 and start exploring that neighbour instead. So the neighbour of 4 has 0, 3 and 7 so
we can pick anyone of them so let us pick the smallest one. So I suspend 4 and I start exploring 0
instead. So now I look at the neighbours of 0 and explore them if they have not been visited.

So notice that now visited this 2 for 4 and 4 and the stack has 4 in it but now I am going to
suspend 0 and pick up one of its unexplored neighbour say one. So now I have marked 1 and 4.
And so the stack is growing from left to right remember. So 0 has come on top of 4 in the stack.
So now I have to explore 1, so explore 1 and I find it has only 1 neighbour which is unexplored
namely 2. So I suspend 1 and I start exploring 2. Now when I look at 2, 2 has no neighbuors left
to explore because it has only 2 neighbours 0 and 1 both have which I have already been visited.
(Refer Slide Time: 4:13)

So, I start moving back. So I back track from 2 to the most recently visited one which is 1 and
see whether 1 has anything more to be done. But 1 has nothing more to be done because 1 had
only 2 neighbours I have visited both of them.

(Refer Slide Time: 4:26)

So, I back track to the previous one which is 0. So I ask whether 0 has anything more to be done.
So notice that from 0 I have visited 1 and from 1 I visited 2. So though when I started with 0 I
had not seen 2 yet by that time I have come back to 0 through back tracking 2 has already been
visited. So at the current state of 0 it has no unvisited neighbours. So 4 was visited before it came
to 0.

1 was visited from 0 and 2 was indirectly visited from 1 and therefore is no longer available. So I
have to back track from 0 as well. So I back track from 0 back to 4 and now I ask whether 4 has
anything interesting to say. And 4 says yes I have an another vertex to visit which is 3. So I
explore this 3 and I suspend the 4. So now I have finish this whole section and I have come here.

(Refer Slide Time: 5:21)

So now from 3 I have two neighbours 5 and 6. And I saw a suspend 3 and maybe I explore 5.
And then from 5 I have I mean 2 unexplored neighbour 6 and 7. So maybe I suspend 5 and I
explore 6. So now when I come to this point I have 4 which triggered 3 which triggered 5 which
triggered 6. So I have 4, 3, and 5 on the stack and 6 has no new neighbours to explore because
both 3 and 5 are already visited.
So, I will back track from 6 to the previous one which is 5 and ask whether 5 has anything more
to do. Well, 5 does it has 7 and 8. So I will perhaps pick 7, so I will again suspend 5 this is the
new suspension of 5 first time I suspended it because I have explored 6 when I came back to 5
and I have suspended it again to explore 7. 7 has a neighbour 8. So I will suspend 7 and explore
8. 8 has a neighbour 9.

So, I will suspend 8 and explore 9 and at this point if you look at the visited matrix everything
this list, everything here has been marked true. But I still have this long queue a long stack of
things which I have suspended. So I have to make sure nothings is missed out. So I will now
look at 9 and 7, 9 has nothing left explore because it only had 1 neighbour 8 from which it came.

(Refer Slide Time: 6:47)

So, that is already marked. So I will go back to 8 but 8 has nothing more to do because 5 and 7
were also marked but 8 was triggered by 7. So I will go back to 7, 7 was triggered by 5, so I will
go back to 5, 5 was triggered by 3 so I will go back to 3, 3 was triggered by 4 so I have come
back to the empty stack. So I have nothing on the suspended list I have also now like before from
4 I explored 0 and 1 but indirectly through this 6 I have already explored 7.
So, 4 also has nothing to be done. So I say 4 is terminated and I quite. So this was how depth
first search work. So in a way you can imagine depth first search the way you kind of browse on
the internet. You start reading something interesting then you click on a link because you sees
interesting you follow that and before you know where you are you had started reading
something and you are somewhere far away.

So, then you have to go back follow back this links and go back to where you started and
continue. So that is more or less what depth first search is doing. So you find the first interesting
vertex and you go there. You keep distance suspension go there. Then find the first interesting
vertex go there and so on.

(Refer Slide Time: 7:43)

So, depth first search finds these long paths like we said the path it found from 4 to 7 was it said
that 4 triggered 3 in case we could have kept this parent information which we did not but for
everything which is marked as visited like in the breadth first search. You could also mark in
depth first search why it was marked visited. So you could keep this parent information. So we
said that 4 marked 3, 3 marked 5, 5 marked 6 then came back and then 5 marked 7.
So, we found this long path 4 to 3 to 5 to 7 if we are kept the parent information you would have
said parent of 7 is 5, parent of 5 is 3 and parent of 3 is 4. But this is obviously not the shortest
path. So it does not do what breadth first search does not terms of shortest paths. So it seems a bit
strange that we use this kind of indirect way of exploring when we have something which seems
to be a better one namely breadth first search.

It turns out that actually depth first search is very useful for other things. So one thing that we
can do in depth first search is keep track of how we visit these vertices. So what I can do, we will
do it formally next time but informally I can say that we keep a counter. So we say that when the
counter is 0, we entered 4. When the counter was 0 we entered 4 and from 4 we entered the
vertex 0. So at this point the counter became 1.

From 0 we entered 2, entered 1 so at this point a counter became 2 from 1 we entered this. So
this point the counter become 3. Now, we finished because the nothing to do at 2. So we finished
2 at the counter was 4, when we finished 1 the counter is 5, when we finished 0 the counter is 6.
And now we have come back to 4 so now from 4 when we explore 7 the 3 the counter is 7 and so
on. So in this way we can keep a counter incremented every time we enter a vertex for the first
time and record it against that vertex as the incoming number of that vertex.

And then every time we finish processing a vertex that is all its edges have been visited then all
its neighbours have been visited then we mark it saying that this is when we finished it. So we
started 1 vertex 1 at counter 2, we finished vertex 1 at counter 5, we started vertex 0 at counter 1
we finished vertex 0 at counter 6. So notice that the counter value tells you something. It tells
you that 1 was visited after 0 because 1 started at 2, 0 started at 1 and 0 finished after 1 because 1
finished at 5 and 0 finished at 6.
(Refer Slide Time: 10:19)

So, this DFS numbering is very useful and we can use it to find many interesting things. So there
are some things that we talk about last time as problems on graph. We talked about coloring, we
talked about matching and so on. But sometimes you also want to finds special vertices. So for
example, supposing there is a critical vertex in your graph. So imagine an electrical network.

Supposing there is one power station which is very critical that anything happen to this power
station then the entire electrical network will get divided into 2 disconnected portions. So this is
called a cut vertex or sometimes called an articulation points. So in this network for example if
this is a airline network. And we are imaging that some airport is unavailable because of bad
weather.

So, supposing there is a cyclone, so supposing vertex 4 the airport at city number 4 is knocked
out of service because of a cyclone then you can see that there is no way to get from these
vertices to these vertices. So we can still travel between 3 to 9. We can still travel between 0 1 2
but we cannot travel from 0, 1, 2 section to this. So now the graph has become disconnected. So
would we discover that 4 is such a vertex. For instance, 3 is not a such a vertex if I remove 3 the
graph still remains connected by this outer path.
(Refer Slide time: 11:37)

So, this 3 is not a, so if I remove 3 there is no (())(11:39). If I remove 5 also there is no problem,
I can still go from 4 to 6 this way and I can come to these, this way. So which are the articulation
points of cut vertices, you can discover that using DFS numbering. A related thing is this is for
vertices, related question is for edges. So are there edges which are critical for me.

(Refer Slide Time: 11:59)

If there is a root which I cannot follow will that disconnect things for me? So now you can say
that this route is okay because it is not a critical route because of if I cannot go from 4 to 3
directly I can still follow an indirect path and go from 4 to 3.
(Refer Slide Time: 12:16)

But again if I knock off this one, if I knock off the route from 4 to 0 then it turns out that 0 1 2 is
disconnect from the rest of the network so this is called a bridge. So these kinds of properties of
graphs, these cut vertices, bridges and many other interesting things can be discovered as a
byproduct of depth first search.

So, you do a depth first search then you do this DFS numbering which we will describe in a later
class. And using this DFS numbering you can actually find out interesting properties of the
graph. So that is why depth first search though it does not find shortest paths it finds out
interesting structure within a graph.
(Refer Slide Time: 12:51)

So, to summarize DFS is a different strategy from breadth first search, is another systematic
strategy to explore a graph. So what we do is we start for the vertex, suspend it go to an unvisited
neighbour, suspend it go to an unvisited neighbour and so on. And in order to keep track of these
suspended vertices and how to resume them in a systematic way we use a stack. So we use this
last in first out data structure in order to keep track of how to resume vertices when we come
back from a terminated computation.

And we saw that although not in detail just informally, we saw that if we keep track of the
sequence in which we visit vertices, when we finish, when we entered, when we finish we get
this DFS numbering and with this DFS numbering we can actually uncover some structural
properties of graphs which are quite interesting.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Applications of BFS and DFS-1
(Refer Slide Time: 0:15)

So, we have looked at breadth first search and depth first search as two ways to systematically
explore a graph.

(Refer Slide Time: 0:21)

So, what we are going to look at now is how to go beyond just reachability, so what we have
done so far is starting with a vertex how to find out what all we can reach in that graph, and we
said that BFS and DFS are two systematic strategies to do this. So when we do BFS, we do it
level by level, so we start with its vertex, we go to its nearest neighbors, then from those nearest
neighbors we go to the next set of neighbors and so on.

So in this process, one of the things we said is that BFS will discover the shortest path, the
shortest level for every reachable vertex because it is processing the graph layer by layer in some
sense, everything which is reachable in one round, in two edges and three edges and so on. And
of course the whole point of calling it breadth-first search is that we need a systematic way to do
this, so we had this queue which kept track of how to make sure that we explore all the vertices
in this level by level order.

So everything at level one is put into the queue and it will get processed before everything at
level two and this will guarantee that everything is reached in the shortest number of levels. Now
DFS was a very different strategy, it was in some sense an aggressive strategy, the moment it
moved to a neighbor, it would suspend the current vertex and then it will start exploring the
neighbor. I mean in a way it is a bit like how when we start looking up information on the
internet, right?

So, we start looking for something and then before we finish reading the article we find an
interesting link and we go to that link and we start reading that link, that has another link and we
go to that link and so on. So eventually we have to remember to come back to where we were
reading in the first place.

So DFS is like that, you start with a vertex, look at any vertex that is neighboring it which you
can explore, go down that path and only when you run out of things to see you come back to the
original vertex. So you keep this stack of suspended vertices in DFS. So the question that we are
going to address in this lecture is what more we can do then just reachability with BFS and DFS.
So, is BFS and DFS only for reachability or are there more interesting things that we can do?
(Refer Slide Time: 2:16)
So, first aspect of a graph that we will explore is that of connectivity. So we say that a graph, an
undirected graph is connected if every vertex is reachable from every other vertex. So you can on
the right two graphs, the first graph is clearly connected from any vertex of 0 1 2 3 4, you can go
to every other vertex. But if you look at the bottom you have clearly two disconnected
components.

So, you have this left hand side component and you have this right hand side component and
there is no way you can go from here to there and in fact 5 on its own is also isolated from
everything else. So that is not a component on its own, so really technically you have these
components which are, so you have this component, everything here is reachable from itself.
Here this component, everything inside this component is reachable from within that component
and finally we have this component which consists of just one vertex

So, in a disconnected graph we can identify these components. So what we want to do is put
these red border around this component and this blue border around the second component and
then find also in particular these isolated vertices which are what you might think of is trivial
components. So we said that technically there are no self-edges in a graph, we said that we do
not assume there are edges from i to i.

So, when we say a single vertex is a component we are just saying that it cannot reach anything
and so there is nothing else that it can be connected to which can come back to it. So these are
the trivial components. So our goal is to see how BFS or DFS can help us identify these
components.

(Refer Slide Time: 3:54)

So, when we are doing one of these, either BFS or DFS, what we do is, just like we kept track of
extra information like in BFS we kept track of the level number and the parent information and
even in DFS we said we could keep track of parent information, so we will have one more
component number that we keep track of. So we have a number which we are going to use to
label these components, so we are going to call it component 0, component 1, component 2 and
so on.
So, we have a component number and we attach component numbers to vertices. So initially we
want to do it for the whole graph, so we might as well start at vertex 0, we want to find out how
many components are there in this graph. So we start at vertex 0, with either BFS or DFS it does
not matter, and then this new quantity that we are going to assign to vertices which is the
component number, we initialize this quantity to 0.

So now, when we do a BFS or a DFS from vertex 0, we will reach some vertices and all of these
vertices will be connected because they are all reachable from 0 and therefore they are reachable
from each other, in fact can reach it from 0, if I can reach say 0 to i, and I can reach 0 to j, then I
can go from j to i by going back to 0 and then because these are undirected edges, you can
always go backwards.

So, if I can go from 0 to some vertex i and I can go to 0 to some vertex j, then there is a
connection from i to j also. So everything that you get in a single component, in a single scan of
BFS or DFS is going to form a connected component. So what you do is that, while you are
performing the scan you remember this component number is 0 and you just assign component
number 0 to all of these. This is an extra piece of information that you keep just like we keep
visited v, we keep component of v and we just keep assigning component of v equal to 0 for all
these vertices.

Now, if you are in the connected graph like in the first case, all the vertices would have been
covered. But if you are in a disconnected graph like in the bottom case, there are some vertices
after you have reached everything that you can reach from 0, there are some vertices which are
not yet marked as visited. So this means that they are not in component 0, they are in another
component, so you have to find that.

So, you have to pick any one of them, you have to pick any one of the vertices which are not
visited and start a new breadth-first or depth first search from there. So in particular let us pick
the smallest one, so supposing we identify the two in the second graph, in the first graph there is
nothing to do because everything is already being visited, but here we identify that vertex 2 is a
candidate to start of a new BFS or DFS because it was not visited when I started from 0.

But now this is going to be a different component. So we cannot call the component that we are
going to discover starting from 2, the same as the component we already discovered. So we have
to increment the component number, so now we are looking at vertices which will be called
component 1. So we perform this breadth-first search or depth first search, we will find that all
those 6 vertices on the right hand side are reachable from 2. And as we are done with the first
component while we are doing this we will attach the current component number which is 1 to
all of these.

So, now after two DFSs or two BFSs I have identified component 0 and component 1, but at this
point I still have an unvisited vertex, so I will repeat this. So in general there may be many
components, so each time I finish one component I will look at the remaining unvisited vertices
and start yet another DFS or BFS from safer instance the smallest numbered vertex in that set.

So, we just keep repeating this, in this case we only have to do it one more time because we have
only one such vertex, vertex 5 and because we cannot go anywhere from 5 this BFS rapidly stops
and we get a component 2. So remember each time we repeat we also increment the component
number, so we increment the component number and start a fresh scan. So this is how we can
build in a component discovery algorithm to BFS or DFS.

So, while we are doing BFS and DFS we can also discover what are the connected components
in the graph and label them. So we now know at the end of this, that for example 9 and 4 in the
bottom graph are both in component 0, so they are in the same component, whereas 9 and 10
have different component number 0 and 1, so 9 and 10 are not in the same component. So it is
important that we can at the end of this look at two vertices and decide are they connected to
each other or not without having to start another breadth-first search from those vertices and then
try to again connect them.
(Refer Slide Time: 8:27)
So, related to breadth-first search and depth first search is also the idea of a cycle. So a cycle as
you could imagine is something which is circular, so a cycle in a graph is a path which starts at
some vertex and then comes back to that same vertex. So remember that we said technically that
a path does not repeat vertices, so technically this is a walk because we start at a vertex, traverse
some edges and come back to the same vertex.

So, if you look at the bottom graph for instance, 4 then 8 then 9 and then back to 4. So this is a
cycle. Here is a more complicated cycle, 2 to 3 to 7, so I start at 2, then I go to 3 and then I come
to 7, but then I go to 10 and then I come back to 6 and then I come back to 7 and then I come
back to 2. So in this case the walk not only repeats the starting and the ending point 2 but it also
repeats 7 along the way. So this is also called a cycle.

But though you can repeat vertices, you cannot repeat edges, so you cannot claim that this is a
cycle, I went to 2 and then I came back, this is not a cycle. So a cycle cannot go back and forth
along the same edge, otherwise every edge will become a cycle and we do not really intend that.
So finally what we are typically interested in a what are called simple cycle. So simple cycle is
like this one.

So, this one is a simple cycle because it went from start to end and came back to start rather, so
the only vertex that was seen twice was a starting vertex when we closed the cycle. Whereas, this
one was not simple because we went down and then we came back up and then we visited this 7
twice, so it is actually two simple cycles which have been joined at 7. So we can do 2 3 7 and
then we can do 7 10 and 6 7.

So, if a graph does not have any cycles, then it is called acyclic. So this graph is acyclic because
there are no cycles in it. Whereas, the graph on the bottom is not acyclic, so we do not really call
it acyclic graph, we only interest, we are interested in whether it is acyclic or it is not acyclic. So
these graph has cycles and what we would like to do now is say that if a graph has cycles how do
we find these cycles. That is our goal.

(Refer Slide Time: 10:51)

So, we start by looking at what is called as tree. So tree is a minimally connected graph. So here
look at this example here, so this acyclic graph that we drew, so it is connected because we saw
that it is one connected component. It is also acyclic and in particular now if we drop any edge
like for example if we drop the 2 4 edge, then this graph will become disconnected. So if I want
to minimally connected, I need to draw at least these many edges in this case there are five
vertices, 0 to 4 and I have drawn four edges.

So, it turns out that when we explore for instance the tree using BFS, then the edges that we use
to visit new vertices, so I start at a vertex and I visit a neighbor if the neighbor is not visited, so
then I will count that edge as being visited by BFS. So the edges that BFS visits actually form a
tree. So if you look at this acyclic graph on the top, it actually visits all the edges because the
graph itself is a tree and there are not fewer edges which form a tree in that case.

Now, if you have a graph like the bottom one which has multiple components then technically
you have to start BFS each time from a new component. So it is not one tree because the tree as a
whole would connect the entire graph, but the entire graph is not connected. So each component
gets connected by a tree. So here for instance BFS does not visit this edge, so this edge is outside
the BFS tree.

Similarly, here, this edge and this edge are outside the BFS tree. Now from English since we say
that a forest is a collection of trees, in this kind of a thing also we talk about multiple trees are
forming a forest. So technically what BFS does is it discovers a collection of trees or a forest
inside the graph. So some useful facts about trees, so the first fact is that if you have n vertices
then a tree on n vertices will have n minus 1 edges.

Now, it does not specify which edges, so you can connect them in many different ways, but
whichever way you connect them you will have to use n minus 1 vertices. So just as an example
supposing we have four vertices, so one way to connect this in a tree is to connect it like this in a
single path. So this is one way to do it. And of course we can choose different paths, so we could
connect it this way or we could connect it this way and so on.

But a different way to connect it which is not a path is to connect it all to one vertex like this, so
this is also a tree. But notice that all these ways of connecting these four vertices, so that each
one is connected to the other but in a minimal way, all of them have three edges. So whenever
you have n vertices, you will have exactly n minus 1 edges and of course the tree is acyclic.
Because if you just think of it as a minimally connected graph, then if it has a cycle for instance
if I had another vertex, another edge from here to here, then I have two ways of going from all
the vertices on the cycle.

Two ways of going from 2 to 1, two ways of going from 0 to 4 and so on. So if I delete one, for
example now if I delete this vertex or this edge, then I still have a connected graph. So if I had a
cycle, then it cannot be minimal because some edge along the cycle can be removed and I will
still be able to reach everything by going around the cycle the other way. So we are now going to
prove this fact, but it is useful to know that these are all equivalent ways of thinking of a tree, so
tree is a minimally connected undirected graph, a tree is necessarily acyclic.

So, an acyclic connected undirected graph is a tree. And on the other hand if it is connected and
it has n minus 1 edges, then it is a tree and if it is a tree, it has n minus 1 edges. So just remember
all these, because these are all different ways of thinking about a tree. So now coming back to
our question, our question was how do we detect whether the graph we have is acyclic or not. So
what we saw is that we have acyclic as like the first one, then the BFS tree that we get covers all
the edges.

So, if we do not cover all the edges on the graph through our tree, if there are non-tree edges,
then those non-tree edges if we add them must form a cycle and that is exactly what we get, so
we can detect a cycle by searching for non-tree edges. So as we are going along just like we
mark vertices as visited, we can also if we want mark edges as visited or we can keep track of
them by looking back afterward we have done it to see which edges we use, say the parent thing
also gives us the visited.

So, if I say parent of i is j, that means I went from j to i to visit i, so therefore ji was a tree edge.
So whichever way we can recover the tree edges at the end of our BFS and then any non-tree
edge if it exists must form a cycle. So here we have these non-tree edges, so 6 7; 3 7; 7 10 and 8
9; which are not part of the BFS tree that we constructed. So since there is at least one such edge
there must be a cycle in particular in this case both these components have a cycle.

(Refer Slide Time: 15:53)

So, we could do this using DFS also in the same way, but we mentioned last time that DFS also
comes with a different type of strategy to keep track of it called DFS numbering. So let us build a
DFS tree for this same graph and let us describe formally how this numbering works through an
example. So we initially maintained a separate counter for numbering when we enter and exit a
vertex and we will start with 0.

So, we increment this counter every time we enter a vertex, every time we start exploring it and
every time we leave the vertex, every time we finish exploring it. So this will become clearer as
we go along, so in this process, this counter value is assigned to the vertex as an incoming
number and then when we leave the updated counter value is assigned as an outgoing number.
So we have a pre number and a post number for every vertex.

So, let us try and do a DFS for this graph here, starting at vertex 0. So we initialize our counter to
0, so we are starting, so the black indicates the vertex number and this purple number is the pre
number, a number of our DFS counter when we entered vertex 0 was 0. Now we explore for
instance 0 to 1. We have to explore the unvisited vertices which are neighboring 0, so we can do
1.

So, now we increment our counter and say and we entered vertex 1 with counter 1. So now when
we come to vertex 1, we have no further vertices from 1 which are unexplored, the only neighbor
of 1 which I can go to is 0 but 0 is already been visited, so now I am going to leave 1. But before
I leave 1, I will increment the counter, so I will leave 1, I will assign it the post number of 2, and
now in my stack I am coming back to 0. So I am back at 0.

So, 0 is not finished because 0 has another neighbor which is unexplored which is 4. So I will
increment the counter and enter vertex 4 with the pre number 3. So notice that this number is
increasing 0, then 0 plus 1 is 1, then 1 plus 1 is 2, then I went back but did I, I did not leave 0, I
am still processing 0. So I will assign a post number to 0 only when I have finished all the
neighbors of 0, in this case I am not finished, I have gone down another path.

So, 0 1 2 and now I assign the pre number 3 to this vertex. So now I am at 4, so 4 has two
unexplored neighbors other than 0, because 0 is already explored. So I go to the smaller one say
8, so again I increment my number from 3 to 4 and I enter 8. So I enter 8 with the pre number 4
and now from 8 I can go to 9, because 9 is not yet been visited. So from 8 I enter 9 with pre
number 5, so each time I am just incrementing this one counter, the pre and post is the same
counter is being incremented whether I go in or go out.

So, at 9 I get stuck, because I have only two neighbors 4 and 8, both of which have been visited.
So 9 now terminates, saying I have finished processing 9, so I increment to 6 and I get out of 9
and comeback to 8. At 8 again I have nothing more to do, so I increment to 7, get out of 8 and
come back to 4. At 4 I have done 8, but I cannot do 9, because 9 was visited through 8, so 4 is
also done, so I will assign my post number to 8 and get out of 4.
And now I come back to 0. Now at 0 I had two neighbors 1 and 4 both are done and I am now
finished with 0, so I will exit 0, so this is now my first component that I have discovered, starting
from 0 and coming back to 0 and I entered and exited each vertex by updating that counter. Now
I go to an unmarked vertex 2 and I continue the same numbering, so I enter 2 with vertex, with
pre number 10, from 2 I go to 3 with pre number 11, from 3 I go to 7 with, so I am following this
path.

So I have the smallest neighbor, then the smallest neighbor, then the smallest neighbor and then I
will get stuck. So 2 will go to 3, 3 will go to 7, 7 will go to 6, 6 will go to 10 and then at 10 I am
stuck, because I cannot go back to any vertex which is not visited, so I will exit 10 with counter
15. Then I will come back to 6, again 6 has neighbors 2 7 and 10 all of which have been seen, so
I will exit 6.

I will come back to 7, now 7 has another neighbor 11 which I have not yet explored. So when I
come back to 7, I am not done, instead I start exploring vertex 11 with counter 17. Now 11
obviously has nowhere to go after that, so I exit 11 with 18. Now I am done with 7, so 7 becomes
exited with 19, 3 is exited to 20 and finally I come back to 2 and the other vertices which are
neighboring 2 namely 7 and 6, have already been explored through 3, so there is nothing more to
be done at 2, so I exit vertex 2 with 21.

So, at this point I have visited all these vertices and I have visited all these vertices. So this
vertex 5 remains, so I have to start a third DFS as we did before. So I started with a new counter
value 22, because I finished the last one with 21 and then I immediately exit because 5 has
nothing to do. So this is, at this point we have not use in these numbers, we will see soon why we
are going to use these numbers, but at this point it is just to show that when we are doing BFS we
construct in a tree, when we are doing DFS also we construct a collection of trees is one for each
component.

And it is now useful to actually describe the order in which this tree was drawn. How did we add
edges to the tree and when did we back track up the tree? That is what we are keeping track of
for this pre and post number. So ignoring the numbering, now there are some edges which did
not come into the tree. So these are these red edges here, so we have an edge for example from 4
to 9, which did not come into the tree because our tree went 0 to 1, 0 to 4 and then 4 to 8 and
then 8 to 9.
So, since I covered 9 through 8, I never got to explore 9 from 4, so 4 to 9 is a non-tree edge.
Similarly, 2 to 7 and 2 to 6 are non-tree edges, because I did not explore them and so is 7 to 10.
Because I went from 7 to 6 to 10, so I never explored the edge 7 to 10. So these 4 edges are non-
tree edges, and each of them as you can see creates a cycle, so exactly like in the BFS, in an
undirected graph, in a DFS also all the non-tree edges create a cycle.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Chennai Mathematical Institute
Lecture No. 65
Applications of BFS and DFS-2

(Refer Slide Time: 0:4)

So, now what happens in a directed graph? So, in directed graph a cycle also is directed, that is I
must go around from a vertex through a set of neighbors and come back, but following the same
direction. So, it is like going around a circle in a one-way street, you have to follow the one-way
street, you cannot suddenly go down in one-way street the wrong way. So for instance, I can go
from 0 to 2 then I can go from 2 to 3, and then I can come back from 3 to 0. So, this is the cycle in
the directed sense because I am going forwards at every step.

On the other hand, if I try to go from 0 to 5, and then from 5 to 1, and then I try to come back from
1 to 0, this is not allowed because 1 to 0 is not in the same sense. So, if I ignore the direction there
is a cycle 0 5 1 0, but with directions there is no such cycle, so we are interested in directed cycles.
(Refer Slide Time: 1:08)

So, again we again do a DFS and do the DFS numbering, it is exactly the same, there is no
difference in DFS whether it is directed or undirected, so we follow the same protocol for these
pre and post numbers, so we start at 0 in this case. We are starting at 0 and then we are going to
systematically explore its neighbors. So, we enter 0 with pre number 0, we enter its first neighbor
with pre number 1 and now we explore 1.

So, we enter 4 with pre number 2, from 4 we have many ways to go, so the first way is 5, so we
enter 5, so we have now come down this way. So, we have gone from here and we have reached
here. So, we enter 5 with pre number 3, from 5 we can go to 7 with pre number 4. And now notice
that 7 has no outgoing vertices, edges at all. So, from 7 I cannot go anywhere because it has no
outgoing edges, so from 7 I exit with number 5. So, I have a post number 5 and I come back to 5.

Now I ask whether we can do anything from 5, well 5 had only two outgoing edges, back to 1 and
forward to 7. But 1 was already covered and 7 has just been covered, so 5 also exits with number
6. Now I come back to 4, so 4 I explored 5, 7 is no longer available because 7 is already done
through 5, but this long edge from 4 to 6 is there, so I enter 6 with pre number 7. Again from 6 the
only thing I could have done is go to 7 which I have already seen, so I exit 6 with number 8.

(Refer Slide Time: 2:36)

Now I am done with 4, so I exit 4 with number 9. Now I am also done with 1, so I exit 1 with
number 10, because 1 had only one outgoing edge to 4. So, I come back to 0, so I in some sense I
started here I went here, then in the process I explored everything on this side. So, now I go to the
right side, so I explore the other neighbor of 0 which is 2 by entering it with number 11 because
number 11 because I exited last one with number 10.

From 2 I can go to 3 with entry number 12, from 3 I cannot go anywhere else, I can only go to 6
or 0 which are both already visited, so I exit 3 with number 13. I go back and exit 2 with number
14, and I go back and I exit 0 with number 15. So, you should just do the sanity check, there are 8
vertices here, so there should be 8 into 2 is equal to 16 numbers.

Every time I enter a vertex and I exit a vertex, so if I start at 0, I must end with 15 and I indeed do
end with 15. If you do not end with 15 then there is a problem. So, if you start with n, vertex with
n graph, graph with n vertices and you do a DFS, then at the end the last exit number should be 2n
minus 1.
(Refer Slide Time: 3:46)

So, here we have tree edges which is what we have drawn when we were drawing these, so all
these edges which we followed when we did the DFS are the tree edges. So, this is now a directed
tree but otherwise is the same structure as before, so it is tree which connects all the vertices that
we visited. But the non-tree edges now come in different flavors.

So, one type of non-tree edge is one which follows the direction of the tree, so it goes from a higher
node in the tree to a lower node in the tree. So, the non-tree edge is in the same direction as the
path that it is by passing in some sense you are kind of short circuiting, you are going like a flyover
on a road, you are going over some intersections and reaching a later point. So, 0 to 3 is a forward
edge, so is 4 to 7.

So these are not part of the tree, but they traverse the tree in the same direction as the edges that
they are skipping. The converse would be a backward edge, it goes up a path in the tree. So it goes
from a lower node in the tree to a higher node but again along a path which already exists. So, I
am going from 5 back to 1, and there is a path from 1 to 4 to 5 or I am going from 3 back to 0, and
there is a path from 0 to 2 to 3.

And finally there could be edges which cut across different branches of the tree, for instance, I can
go from 6 to 7, so 6 is on this branch and 7 is on this branch, so it is not that I am going up or down
from 6 to 7, I am going across. So, these are called cross edges. So, these are the 3 types of edges
that you could have in a directed graph which are not in the DFS tree; forward edges, back edges
and cross edges.

(Refer Slide Time: 5:28)

So, now if you look at this carefully, so let us look at this forward edge 0 to 5. So 0 to 5 is a forward
edge and the other edges that it was corresponding to are 0 to 1, 1 to 4 and 4 to 5. So, this was the
new edge that I added. So, clearly adding this new edge to the existing path did not create a cycle,
because it is going in the wrong direction. So, a forward edge does not create a cycle.

(Refer Slide Time: 5:59)

On the other hand, a backward edge does create a cycle. So, if I go from 1 to 4, and then I find that
there is a backward edge from 5 to 1, then there is a cycle. Similarly, if I go from 0 to 3 via 2 and
then I find that there is a backward edge from 3 to 0, then there is a cycle. So, forward edges do
not create cycle, backward edges do create cycles and you can also check that cross edges will not
create a cycle.

(Refer Slide Time: 6:26)

So, cross edges will actually go down different, different paths and so therefore they do not create
a cycle. So, for instance, if I look at this here the cross edge from 6 to 7, so I have a 0 1 4 5 7 path,
so I have this path and I have 0 2 6 path. And now these two paths are going in opposite direction,
so no matter how I connected this way or that way, so I have a path like this and I have a path like
this, so there is no way that I can connect these two paths either left to right or right to left, form a
cycle, because the paths themselves are going in the wrong direction.

So, what we want to do now is to identify not just the non-tree edges, so in the undirected case it
was very simple, every non-tree edge indicated a cycle. Now in the directed case we are saying it
little more subtle than that, it is not enough to just to find a non-tree edge, you must find a non-
tree back edge. So, how do we know which of the edges which are not in my DFS tree are forward
edges, which are back edges and which are cross edges.
(Refer Slide Time: 7:31)

So, the problem is that of classifying these non-tree edges. And this is the first instance where we
will actually use these pre and post numbers. So, if I have a forward edge from u to v, so in this
case I have this forward edge here, from 0 to 5, then we will look at the interval, I say that I entered
5 at 3 and I left at 6. So, I have an interval 3 to 6, during this period I was processing 5, I was going
to neighbors of 5 and so on. And when I finished processing 5 I was at 6.

And look at 0, 0 has an interval which is from 0 to 15. So, I started processing 0 when the counter
was 0 and I finished it when it was 15, so what this says is that the entire processing of vertex 5
happened before I finished vertex 0. So, vertex 5 was processed as a part of vertex 0 processing,
so if now the back edge, if the edge goes from the bigger interval to the smaller interval, then it
means it is a forward edge, because it is a vertex which was processed earlier to a vertex which
was processed later, because the interval is smaller.
(Refer Slide Time: 8:42)

On the other hand, if I have a back edge, then it is exactly the reverse. So, I am going from say 3
which has an interval 12 to 13, and I am going back to 0 which has its interval 0 to 15. So since I
am going backwards, again this indicates that I did the processing of 3 while I was doing the
process of 0. So, therefore, this is an edge back from 3 to 0, 3 happened later, so now because the
edge is reversed, it is asking that the starting interval is included in the ending interval or it rather
the ending interval is included in the starting.

So the, if I am going back from v to u, from u to v, so then I want that this interval is smaller than
this interval. So, the ending interval is bigger than the starting interval.
(Refer Slide Time: 9:37)

And finally if I have a cross edge, you will see that these are actually disjoint, because they are
happening on different things, so I finished processing one path, so I came down this thing I
finished it and then I went back and started here. So, when I started on the right hand side path, I
had finished the left hand side path, so all these numbers are exhausted. So, basically the out has
happened before the in there, so the two intervals will be disjoint. So, this is how we can use this
pre and post numbers with the vertices to discover which of the non-tree edges are back edges and
therefore, we can decide whether or not are directed graph actually has a cycle.

(Refer Slide Time: 10:09)

So, just like cycles have to take directions into account, so does the notion of connectivity. So, we
said that in an undirected graph, we said a graph is connected if every vertex can be reached from
every other vertex. And now in a directed graph we have to ask whether I can go following the
directions. So, I say that a pair of vertices i and j are strongly connected if I go from i I can go to j
and then I can come back from j to i by a different path. So, in this case I say i and j are strongly
connected, if they were not strongly connected, it is possible I can go from i to j but I cannot come
back.

So for instance, if I look at, say for instance 0 and 1 in this case, I can go from 0 to 1 by following
this path, but there is no way to go from 1 to 0. Because I cannot go from 1 except 4 and I cannot,
basically the only way to come back to 0 in this graph is to come by a 3, because that is the only
incoming edge to 0 and there I cannot reach from 1 to 3 by any path. So, 0 and 1 are connected in
one direction but not connected backwards and therefore they are not strongly connected.

(Refer Slide Time: 11:25)

So, therefore, a correct notion of component that we need for a directed graph is one where not
just that every pair of vertices is connected but every pair of vertices in that component is strongly
connected. I can go from everywhere to everywhere and come back, so I can go anywhere in that
component without worrying about where I am starting. So, this is what is called an SCC or a
Strongly Connected Component.
So, you can see that for 3 vertices, like we have in this a 3 vertex strongly connected component
is just a cycle. So, basically if I have a, if I have a directed cycle, if I have something like this, then
this will be a strongly connected component. But I could have more edges, I could have something
like this and so on. It does not matter if there are more edges, but there should be a minimum
number of edges so that I can go from anywhere to anywhere in both directions.

So, in this particular graph 1 4 5 forms a cycle because I can go around this in this direction and
reach anywhere from anywhere. Similarly, 0 2 3 forms a cycle, but 7 and 6 now are stuck on their
own because if I leave 6 I cannot come back to 6 from any of these paths, if I leave 7 I cannot
come back to 7, I cannot leave 7 at all in fact because 7 has no outgoing edges. So therefore the
strongly connected components in this graph are the ones which I marked in red.

(Refer Slide Time: 12:40)

So, what we are not going to cover in this particular course but maybe at a later stage is that this
DFS numbering that we just did can also be used to compute these strongly connected components.
So, we saw that it can be used to compute back edges and detect cycles, but it can also be used to
detect strongly connected components.
(Refer Slide Time: 12:58)

So to summarize, we saw that BFS and DFS are primary strategies for reachability in a graph, but
what we have seen now is that we can do much more with BFS and DFS. So, the first thing we
can do is identify the connected components in an undirected graph. So, by doing BFS or DFS we
first uncover a tree, so we identify some edges in the graph which we process during the BFS and
DFS, this form a tree. And any edge which is outside this tree must form a cycle with the edges in
the tree, so any time there are non-tree edges in the graph after I finish my BFS or DFS, we can
generate, we can say that there is definitely a cycle in the graph.

However, for directed graph we saw that this is little bit more complicated, so we have 3 types of
non-tree edges; forward edges, back edges and cross edges. And of these only the back edges
generate cycles, so this is one instance where we used this DFS numbering this pre and post
numbering for vertices in order to detect which of these non-tree edges are back edges.

And finally, we described the notion of a strongly connected component, and although we did not
actually calculate them, we claim that the DFS numbering that we have done can also be used to
identify the strongly connected components. So, a strongly connected component, remember has
one where every pair of vertices is reachable from each other. So, we mentioned last time and we
will not elaborate on this, but DFS numbering can be used for many other things, so one thing it
can do is identify the so called critical vertices.
So, cut vertex or an articulation point is one which if I destroy it, it will disconnect the graphs. So,
if I remove it from the graph, the remaining graph falls apart, so this is critical for instance if you
are looking at a network of say a power network, if there is a power station which is relaying power
and if it goes out of action and the power network now disconnects, so two parts do not get power
from each other, then that is something that we have to be extra careful to protect.

So, these are also important things to discover in your graph. Similarly, there could be cut edges,
if I cut a connection between two nodes, then the graph falls apart, so these are called bridges. So,
articulation points and bridges can also be identified in a graph using this notion of DFS
numbering. And finally, this idea that we are looking for cycles in a graph is particularly important
in the directed sense, so there is a very important class of graphs which we will look at next week,
which is called directed acyclic graphs.

So, a directed acyclic graph as it suggests is a directed graph which has no cycles in it. Now a
directed acyclic graph is very often used to represent these kind of prerequisites or preconditions
or dependencies. So, supposing I want to describe for instance the course contents of this program
and I say that you have to do maths 1 before you do maths 2, and if I have to do stats 1 before you
do stats 2 and maybe there is no correlation between maths 1 and stats 1, so you can do them in
any order and you can postpone one to the other, but you cannot do maths 2 before maths 1.

So those are, so I have M1 and I have an edge saying M1 must be before M2 and I have one saying
that S1 must be before S2. And now maybe I have something which says that in the third semester
there is an math for ML in which I must have completed both M2 and S2. And now I ask in what
order you can do these things? So, clearly you can do math for ML only after you finish all these
4 courses, but these 4 courses you can be a little flexible, you can do S1 for instance, after M2 or
you can do M1 after S1.

So, these kind of properties about which sequences can be compatible with a set of dependencies
are used very often and this can be done by analyzing this directed acyclic graphs which we will
do later on.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Indian Institute of Technology, Madras
Lecture No. 66
Complexity of BFS and DFS

(Refer Slide Time: 00:14)

So, having seen some of the applications that we can achieve using BFS and DFS. Let us go back
a little bit and look at the connection between BFS and DFS and these representations that we
talked about Adjacency Matrices and Adjacency Lists.
(Refer Slide Time: 00:28)

So, let us just formally remember what BFS does. So BFS explores a graph level by level. So, we
maintain 2 pieces of information we maintain this flag called visited, which indicates whether a
vertex have been visited or not. And we keep this queue of unexplored vertices. So initially, we
mark all vertices as unvisited, we start from a vertex j. So, we start by setting that to be visited, set
its value of visited to true, add j to the queue, and then we repeatedly process the queue.

And by processing the queue, what we mean is we take out the first element in the queue, look at
its neighbors, and if there is any neighbor, which is unvisited, we mark it as visited, and push that
neighbor back into the queue, so it will get processed later on. And finally, when the queue gets
empty, BFS terminates.

On the other hand, with Breadth first search, we do a kind of aggressive or, impatient traversal.
So, we start with i, and we visit 1 neighbor j. And now once we visited 1 neighbor, j, instead of
going back to i and continuing with another neighbor of i we, we suspend i, and we go to neighbors
of j. The first time we find a neighbor of j again, we will suspend j, and go to a neighbor, that
neighbor k and continue.

And at this point, when we finish, and we cannot go further, we go back. And then we have
traversed back all the suspended vertices until we find 1 which is not finished, and look for another
neighbor, and so on. So here, we keep track of this visited information like in BFS, but we also
have the stack, which remembers the suspended vertices, so we can go back in the correct sequence
Last In First Out.

(Refer Slide Time: 02:05)

So, what I want to talk about is how much time this takes as a function of the size of the graph that
we are trying to explore. So typically, in a graph, there are 2 parts, there is the vertex set, and there
is the edge set, which is a subset of the pairs of vertices. So, edge relation is a subset of V cross V.
Now, the vertex set is usually denoted as having size n. So, this tells you how many nodes or
vertices there are in the graph.

Now the same set of vertices, you can draw many edges, or we can draw a few edges. So, actually,
the parameter of the number of edges is independent, in a sense, as a measure of how complicated
the graph is. So, this is usually denoted by a small m. So, m is usually the number of edges, n is
the number of vertices. So, we saw that, if you have a tree, that is a minimally connected graph,
then you will have n minus 1 edges. So, we can have interesting graphs in which the number of
edges is roughly the same as the number of vertices, you have n vertices, you have n minus 1
edges.

On the other hand, if you connect everything to everything, then for every pair of vertices, you
have an edge. And this gives us n into n minus 1 by 2, which is n choose 2 vertices. So, this is
about n square so some something like n square, so we can have either order n vertices or order n
square vertices. So therefore, the number of edges, so the number of edges forms and somewhat
independent parameter in our calculation.

So, now let us look at both BFS and DFS and see how they do this. So, the first thing is that they
visit and explore every vertex exactly once, that is the purpose of that flag visited, it makes sure
that we never go back to a vertex which is already visited and try to visit it a second time. So
therefore, we will visit and explore each vertex exactly once. So, that whole thing happens in n
times. There are n times when we visit and explore vertices. Now, what is exploring a vertex mean?

Exploring a vertex essentially means looking at all its neighbors. So, if we are looking at all the
neighbors of a vertex, we are looking at all the edges which are outgoing from that vertex. So, the
question is, how long does it take us to do this, when we are actually doing it computationally not
doing it by looking at the picture and doing it by hand. So, we said that there are 2 representations
that we have of the graph.

The first is this adjacency matrix in the adjacency matrix, the entries are 01. And the ij th entry
indicates whether there is an edge from vertex i to vertex j. So, if there is an edge, it is 1 if there is
no edge is 0. So, if we want to look at the outgoing edges of a vertex i, the only way we can do it
in an adjacency matrix is to walk down the entire row for i. So, these are also look at a i,1, i,2, i,3
and so on up to a,i,0 i,1 up to i,(𝑛 − 1) if the vertices are numbered 0 to ,(𝑛 − 1).

So, we will have to look up the entire row so, whether or not i has many neighbors, or few
neighbors are no neighbors at all. I mean, of course, in an undirected graph, it must have at least if
we had reached i during this thing, it would have an incoming thing. So, at least one neighbor, but
if it is a disconnected vertex from where we are starting, like we saw a connect a component, which
has only a single vertex, then it may have nothing at all.But we would not know that until we see
the entire row for i.

So, regardless of how many neighbors i actually has, we have to spend time proportional to n, to
discover all these neighbors. So, this means that I have overall and processing n vertices. And for
each vertex, I have to scan n entries in this matrix. So, 𝑛 × 𝑛, so I have to do something
proportional to n square, in order to do Depth first search or Breadth first search.
On the other hand, if I use an Adjacency list, then I have for each vertex an explicit list of its
neighbors. So, if it has a lot of neighbors, at least to be long, if it has very few neighbors, and this
will be short, but I will spend no more and no less time than I need to scan this list. So, if I have k
neighbors, I have to look at k entries. But I do not need to spend more, I do not need to spend n
steps looking for k entries when k is small.

The now, the problem with this is that the degree of the vertex as we call it, right degree is the
number of vertices which are number of edges, which are incident at a vertex, the degree of a
vertex varies. So, maybe for vertex 1, I had a small degree vertex 2, I had a big degree and so on.
So, if I want to count how many steps it takes across these n vertices, I have to add up the degrees,
I have to look at how much time it takes degree 1, how much time it took for degree 2, and so on.
So, how do I get a good way of estimating what this adds up to?

(Refer Slide Time: 06:48)

So, the question is, when I am processing an Adjacency list, I do work proportional to degree of
each vertex added up. When I process vertex 0, I will look at all its neighbors. So, I will spend
time proportional degree of 0 and I do vertex 1, the same thing with vertex i the same thing. So,
we are really interested in identifying what the some of the degrees in an undirected graph
represents. So, what is a degree?

A degree indicates a number, which is the number of things going out. So, if I have i, then if I have
so many things going out, then I get a contribution of 4 from i to the sum of the degrees, because
it has 4 outgoing things. But each of these things will go and terminate also in j. So, if I look at
this edge i,j, it adds 1 to the degree of i, it also adds 1 to the degree of j. So, each edge contributes
to degree of both i and j.

And every number that I get in the degree must come from some edge. So, since there are m edges,
and each edge contributes to the degree count of both the starting point and the ending point, the
sum of the degrees must actually be 2 × 𝑚. For each, each vertex each edge, 1 to m, it contributes
1 plus 1. So, it is 2 times.

(Refer Slide Time: 08:16)

So, now we can see that if you have BFS or DFS with an adjacency list, then you make n steps,
you take n steps to visit each vertex. And then across the n vertices, you take the, the sum of the
degrees or 2 m steps to explore all of them. So, this says that the time that overall that you spend
is proportional to n plus m, you have to visit all the vertices because for instance, if the entire graph
is disconnected, there are no edges at all, m is 0, but it does not mean that you will finish your DFS
or BFS in 0 time, you have to visit all the n things.

So, you have to spend n steps looking at all the vertices and across the vertices, the contribution
of that vertex to your work is the degree of that vertex. So, across the vertices is the summation of
the degrees, that is where this m term comes. Whereas as we saw, if you did this with an adjacency
matrix, it does not matter really what the degrees are, and what m is, you will end up having to
spend n steps for every vertex, because you cannot find out the neighbors of a vertex without
looking at the entire row. So, you end up taking time n square.

So, the whole distinction is between n square and n plus m. So, we said that m could be small. So,
m could be like n, like in a tree, we have only n minus 1 vertices, or m could be large, it could, in
principle be n square. If it is n square, then these 2 are roughly the same. But if it is small, then we
have a difference between something which is proportional to n, and something which is
proportional to n square.

And that is why adjacency lists can be beneficial for doing our thing. So, if m is proportional to n,
then we have a big saving by using adjacency list representations. So, though adjacency matrix is
fine in terms of understanding what is going on in terms of mechanically computing something, it
can actually cost us not just because it is a large thing, and we are keeping a lot of 0s in it. But
also, because in order to extract this interesting information about what are the neighbors, we have
no better way than to walk down the entire row.

(Refer Slide Time: 10:19)

So, here are some more interesting things that is useful to remember about degrees. So, remember
that the degree of a vertex can be obtained from an adjacency matrix by just looking at its row. So,
if you just count the number of 1s in the adjacency matrix in a row, you get the degree of the
vertex. And if it is an undirected graph, you can also get it from the columns, because it is the
number of incoming edges and the number of outgoing is the same, because it is a symmetric kind
of edge graph.

And if you have an adjacency list, then the degree is just the length of the list associated with i, it
is all the neighbors of i. Now, we already calculated that the sum of the degrees is 2 n. So, 2 times
anything must be an even number. So, the sum is an even number, because it is 2 times the number
of edges. So, the number of edges themselves will be odd, but the sum of the degrees is going to
be 2 times m, and therefore the sum of the degrees is an even number.

So now, you should remember that, if you have an odd number plus an odd number, then I will get
an even number. So, for example, if I do 3 plus 17, then I get 20. But if I do an odd number plus
an even number, then I get an odd number. So, you can think of the fact that if I am pairing up an
odd number in 2's, then I get 1 leftover element, that is why it is odd. And if I am pairing up an
even number in 2s, I get no leftover elements.

So if I take them together and pair it off in 2s, then I have 1 leftover element, that is why it is odd.
Whereas if I have 2 odd numbers, then each of them contributes 1 leftover element, I can take
those leftover elements and pair them up and I get a new pair, and so it is even, so, odd plus odd
is even or plus even is odd. So, this means that if I spot an odd degree vertex in my graph, it cannot
be the only 1 there must be another 1 right because otherwise the sum of the degrees will become
odd everything else is even plus 1 odd number will be an odd number.

So, if there is an odd number here, there must be an odd number somewhere else, so I can pair
them off. Similarly, if I find a third odd vertex there must be a fourth odd vertex. So, for every odd
degree vertex, there must be another odd degree vertex. So in other words, the number of degrees
or the odd vertex or degree vertices must themselves be even, if I have an odd number of vertices
with odd degree, then the overall sum of the degrees will be odd which is not possible.

So remember also that the degree of a vertex can be any number between 0 and n minus 1. So, 0
happens when this vertex is actually disconnected from the entire graph, n minus 1 happens when
it is connected to every 1 of the remaining n minus 1 vertices. So, it is not connected to itself
obviously, we said no self loops, but it could be connected to everything else. So, the special case
where every vertex is connected to every other vertex is what is called a Complete graph.
(Refer Slide Time: 13:15)

So, for instance, on a 3 vertices graph, a complete graph is a triangle, if a 4 vertex graph, then I
must connect everything to everything. So, I have a square and then I also have the diagonals.
Similarly, if I have a 5 vertex graph, then I will have a pentagon with all these diagonal things. So,
this is what is called a complete graph. So, in a complete graph, the degree of every vertex is n
minus 1 and I will actually have n into n minus 1 by 2 because every pair is connected. So, I have
n choose 2 edges.

(Refer Slide Time: 13:44)

Now many graphs that we encounter in practical problems like the 1s we discussed about graph
coloring or vertex cover, so on. In many reasonable situations, you can actually say that a particular
node will have no more than a certain number of neighbors for example, remember the graph
coloring problem for the security cameras? So, clearly, if I give you a particular building, then I
know that a given intersection will not have more than a certain number of corridors fixing I cannot
have a large number of corridors at some point.

So, there will be some upper bound saying that no corridor, no intersection has more than 5
corridors which meet there or if we are looking at some, some other problem, which is say for
instance, placing ambulances at an intersection, then you want to know how many roads meet at
that intersection. Or if you are looking at say, the timetabling problem you want to know how
many different courses can be scheduled in the same slot. Now, obviously, if you have a fixed
number of courses in your curriculum, not more than that many can be there.

So, very often the degree is actually independently bounded by some number, you do not have
arbitrarily large degree. So, the number the degree bounded by an external constraint. And if you
have now a constraint on the degree, it says that each degree is k, then the total sum of the degrees
can be at most k times the number of vertices k times n. And since this is twice the number of
edges, the number of edges must be 𝑘 × 𝑛/2.

So in other words, if you have a bounded degree graph, where every edge has a bounded degree,
which is independent of the size of the graph, then the total number of edges cannot be more than
linear, cannot be more than some function of n. So therefore, you should be working with
adjacency lists. And finally, if you have directed graphs, we said that it is no longer enough to talk
about degree because there are edges coming in, and there are edges going out, which are quite
independent of each other. So, we must talk about the in degree and the out degree.

So, in this case, because each edge contributes to 1 in degree and out degree, the sum of the in
degrees is the number of edges and the sum of the out degrees is also the number of edges. So,
together the in degrees plus out degrees is 2m, but they get partitioned into 2 quantities which add
up to m each.
(Refer Slide Time: 16:06)

So, what we have seen is that if we do an analysis of how the BFS and DFS work with respect to
exploring the neighbors of a vertex, if we use an adjacency matrix, it turns out that regardless of
how many neighbors a vertex has, we must scan the entire row and therefore the time taken by
BFS and DFS becomes proportional to n times n, I have to process n vertices and for each vertex,
I have to scan n elements in that row.

On the other hand, if we use an adjacency list, then the time taken to process a vertex is exactly
the number of neighbors it has. So, it is across the vertices is the sum of the degrees and this gives
us an overall timing, which is proportional to n plus m. And we also saw that there is a large
variation in m. So, m can be linear, like in a tree, or it could be quadratic, like in a complete graph.
So, it is important to be able to distinguish and use the appropriate representation in particular use
adjacency list whenever we can.

Another situation where we get small number of edges is when we have a bounded degree. So if
we, if we have some other constraints on our problem, it says that the number of edges coming out
of a given node cannot be more than a certain number independent of the total number of edges.
Then we have necessarily a graph which has only a linear number of edges. So again, an adjacency
list representation would work best.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Indian Institute of Technology, Madras
Lecture No. 67
Directed Acyclic Graphs

(Refer Slide Time: 00:14)

So, last week we looked at graphs with cycles. So, we saw that we can use our depth first search
and breadth first search to find cycles and graphs. And in particular, we looked at directed cycles.
So, we said that if in a directed graph, you have a cycle, then the cycle must follow a uniform
direction. So for example, in this graph here, we see that this is a cycle because you can go from 0
to 2, 2 to 3 and back to 0.

But although this other one here on the left looks like a cycle, it is not because if you go from 0 to
5, and then 5 to 0, then this edge from 0 to 1 is in the opposite direction, so we cannot follow that
direction. So, that is not a directed cycle. Whereas for instance, this is a direct cycle. And so this
is what we looked at last time. And what we said is that we can use DFS, Depth First Search to
find these directed cycles. Because when we do DFS, we will construct a tree to begin with.

So, we have all the edges which are passed are part of the tree which have been used during DFS.
And then among the non tree edges, we have 3 different types, so we have these forward edges,
which go forward in the tree, we have back edges, which go up the tree from a later node to an
ancestor. And then we have cross edges which go across branches. And we said that only the back
edges actually generate cycles.

And by using this DFS numbering by recording the time at which we enter each vertex to process
it, and we exit after finishing processing the vertex by looking at these DFS pre and post numbers,
we said that we could analyze the tree and look at all the non tree edges and decide which category
they belong to. So in this way, we can find out all the non tree edges. And the back edges in
particular, which are the cycle forming edges. So, now the question is, why were we so worried
about cycles in directed graphs to begin with?

(Refer Slide Time: 02:00)

So, let us look at a general problem, where we have some things to do some tasks, and there are
some dependencies between the tasks. So, as an example, suppose there is a startup, which is trying
to move into some new office space. So, there is some brand new office space, and the startup
needs to set up this office before it can move in. So, this office space is completely unfinished. It
is a new building just constructed, just the bricks are there.

And so what we need to do is a number of things, we need to lay the floor tiles, we need to plaster
and paint the walls, we also need to lay pipes, these conduits as they are called. In order to take
wires from here to there. So, there are wires of two types there are electrical wires, but also there
are networking cables for computers, there are also telephone cables and so on, and these cannot
go in the same conduit because they interfere.
So, we will have separate conduits for electrical wires, separate conduits for telecom equipment,
then of course, you have to put in the wiring. And you have to also after you finish the wiring of
the electrical things, you have to put in the fittings, you have to put in the lights, the fans, the
switches, and so on. So, now, these are all activities which need to be done, but clearly they cannot
be done in arbitrary order. So for instance, we have these constraints.

So, we need to put the conduits into the wall and the floor. So typically, these conduits run along
the wall to the plug points and to the ceiling. And then across the room, they will flow travel
underneath the floor. So, before you put the tiles and before you plaster the walls, you need to put
the conduits otherwise, obviously you have to break the tiles or break the walls, which is not a
good idea.

Then before you paint the walls, you must of course plaster through the wall. But typically you
also like to paint before you lay the tiles because you expect that the person who is laying the tiles
might mess up the walls by putting cement when they are laying the tiles. Whereas of course, tiles
are usually washable. So, if you are painting it and some paint falls on the tiles, It is not a problem.
And clearly you would like to finish the painting before you start putting in the wires into these
conduits.

The reason is that if you have the wires hanging out loose when they are painting then the paint
will go and gum up the wires and then you will have also paint going into these cracks. So,
normally you seal up these conduits with something and then you paint it and then you open it up
and put push the wires through these conduits. So wiring cabling happens after the paint and clearly
you cannot put your fittings in or you cannot put plug points and you cannot put lights and fans
unless the wires are there. So, you can finish the electrical wiring only after installing the fittings.

So, we are going to model this as a directed graph. So, the vertices are going to be the tasks that
we have to perform. So, all these tasks on the left laying the tiles plastering the walls and so on.
And an edge is going to denote a dependency. So, an edge from t to u says that t has to be finished
before you can be started. So, in our case, for instance, you have to lay the tiles before you paint
the walls.
(Refer Slide Time: 04:53)

So, if we take this particular thing and draw the graph, so first we have the vertices so we have
these vertices corresponding to the activities we have two different types of conduits, electrical
and, and Telecom. And we have tiling and plastering. We have painting. We have two kinds of
wiring, we have electrical wiring and telecom cabling. And finally we have the electrical fittings.
So, these are all the activities. These are the nodes in my graph. And now I have these constraints.

So, the first constraint says that I must lay all the conduits before I do tiling and plastering. So,
from each of the conduit nodes, vertices, I have an edge to each of the other two vertices one to
tiling and one to plastic. Then it says you must finish plastering and prep and doing the tiling before
you paint the walls. So, from both the tile and the plaster edges, we have nodes we have an edge
to painting.

Similarly, from painting, we have an edge wiring and cabling because these happen only after
painting and finally we can do the fittings only after we do the wiring. So, this is a directed graph,
which we have constructed from the task constraints given to us.
(Refer Slide Time: 05:56)

So, what is our problem now our problem is we need to complete these tasks in a way that respects
the dependency, so we must make sure that the painter comes only after the tiling and the plastering
are done. So, here is a possible way of doing it. So, we start with things which do not have any
dependencies, the conduits can be laid right at the beginning. Once they are both done, we can do
tiling and then plastering.

Then once both of these are done, we can do painting and then we can do the wiring first the
electrical wiring then the telecom cabling, and finally we can put the electrical fittings. So, this is
a sequence in which we can complete these tasks so that whenever we come up to take up a task,
all the tasks which needed to be done before that are already done. But this is not the only such
sequence of course.

For instance, we could have done the waiting in the opposite order. It does not matter whether we
do the telecom cabling (bef) conduiting before or after the electrical conduiting. Similarly, we
could take up the tiling after the plastering, because plastering and tiling do not depend on each
other. And similarly, we can do the electrical fittings even before we do the telecom cabling,
because they go through different conduits and they do not interfere with each other.

Another question we might ask so, this first question is how do we sequence these in a way that
does not violate these constraints? The second question is, what is the best way to do this?
Supposing we could do things which are independent of each other at the same time. For instance,
we could ask the person who's putting the tiles to work alongside the guy who is plastering because
we said that tiling and plastering can happen together.

Similarly, we can have the person doing the electrical conduiting, working alongside the person
who is doing the telecom conduiting. Similarly, we can have the wiring and the cabling done at
the same time. So, if we can do all this, if we can optimize this so that things which are not
dependent are done in parallel, then how soon can we finish this? How many days will it take to
complete all these tasks, following these dependencies?

(Refer Slide Time: 07:47)

So, if we look at this graph, formally, it is a directed graph. But more importantly, it is acyclic, we
do not have any cycles in this because cycles represent dependencies, if a depends on b and b
depends on a then which do you do first. So, what we are trying to do is to find a scheduled, which
enumerates these vertices in an order, such that in that sequence in the list in which we enumerate
these vertices.

If a task i must be done before a task j according to the dependencies, then it must appear before j
in the sequence. So, every time we have a dependency, an edge in the graph, that edge, the starting
point of the edge has been listed before the ending point indicating with the starting task finished
before the second task began. So, this problem formally in a directed graph is called a Topological
Sort. So, what we want to do is topologically sort this.
And the second thing is to discover how long we need to move. And then this way, I do not find
essentially the longest path. So for instance, we could say that if we start from here, then it is going
to take us four steps from starting the conduiting to finishing the cabling. But if we go along this
path, for instance, this actually says that we need to do 5 things in a sequence. We cannot do these
any faster.

Because plastering can be done only after the conduiting, painting can be done only of the
plastering, then wiring and then the fittings. So, we are trying to find the length of this longest
path. So, these are the two formal problems that we have with DAGS, topological sorting, and
longest path.

(Refer Slide Time: 09:19)

So, to summarize, directed a cyclic graphs are a natural way to represent dependencies. The
direction of the edge indicates the direction of the dependency, what must come before what. The
fact that it is acyclic follows from the fact that if you have a cycle of dependencies, if i depend on
u, and u depend on somebody else, and that person depends on me, then we all depend on each
other, so we cannot get started.

So, if I am waiting for you to finish and you are waiting for somebody else to finish and that person
is waiting for me to finish, who goes first. So, these cycles cannot be these dependencies cannot
form a cycle therefore, it must be a directed and acyclic graph. And these arise in many contexts.
So, we saw this context where we had to finish a room. It could also represent for instance the
sequence in which you take courses to complete a degree.

So, courses usually come with prerequisites. So, you cannot do Maths 2 before Maths 1, maybe
you cannot do the ML for computation or Computing for ML course, unless you have finished
Python programming and both the math courses and both the stats courses and so on. So, now, if
you have prerequisites like this, then find a sequence in which you can take the courses to complete
the degree.

Cooking is another constraint, a place with a lot of constraints, you need to first of course, make
sure you have the ingredients. So, there will be typically a list of ingredients, then there is some
processing to be done before you have to chop some things you have to make some things you
have to grind some things and so on. And then after that there is a specific sequence in which
things go into the pot. So, you put some oil and then you do something else and so on.

So, there are certain things that can be done in parallel 1 person can be chopping the vegetables
while somebody else is grinding up something, but there are some things which have to follow a
sequence. So, cooking recipes also impose a natural dependency on the tasks in order to prepare a
dish. And finally, the kind of problems that we looked at is like a typical project a construction
project or any other large project which has many phases, and these phases, some of them can be
done in parallel some have to be done in sequence.

And once we have modeled these things as DAGS, we can solve a similar problem that arises
across all these different applications by a uniform problem on DAGS, namely topological sorting
and longest paths. So, this is what we will be looking at.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Indian Institute of Technology, Madras
Lecture No. 68
Topological Sorting

(Refer Slide Time: 00:14)

So, we have motivated the use of DAGS by saying that they are useful for representing tasks and
dependencies.

(Refer Slide Time: 00:21)

And 1 of the things that we need to do in a DAGS in a directed acyclic graph is to arrange the
vertices in a list that respects the dependencies. So, we said this is a Topological sort. In a
Topological sort of the vertices, we sequentially list out the vertices in such a way that every time
there is a directed edge from i to j, i must appear before j in the sequence that we list out.

So, in terms of the applications that we are thinking of, if we think of these as tasks and
dependencies, then this type of a topological sort represents a feasible schedule. A schedule in
which no task appears before all the tasks it depends on are already completed. So, when you come
to do something, everything you need to do before that is already done.

(Refer Slide Time: 01:04)

So, the first thing to notice is that if you have cycles, then you cannot do this. If you have a graph
with directed cycles, we informally argued that if i depend on somebody else to finish, and
somebody else depends on me to finish, then we cannot do anything because we are waiting for
each other to finish. But formally, let us see what it means. So, in the topological sort, it is not just
for every edge, for every edge i comma j, it is clear that topological sort requires i to appear before
j.

But in general, if I have a path of dependencies, if i depends on k and k depends on j, then i must
come before k and k must come before j. So therefore, transitively i must come before j. So,
anytime there must there is a path from i to j it must be listed before j. So now, if I have a cycle,
in a directed graph, it means that I can go from some i to some other j. So remember, a cycle cannot
be just something which goes from i to i without going to any other vertex, we need to go at least
cross 1 edge. So, I must go from i to some other vertex j and then come back. So, there is a path
from i to j and a path back from j to i.

Now, by the previous requirement, if there is a path from i to j, then the topological sort is obliged
to put i before j. But since there is also a path from j to i, then we have to put j before i. But clearly
in a given sequence either i can come before j or j can come before i we cannot have both these
constraints satisfied in the sequence, and therefore this would be impossible. So, that is why if we
have a cycle of dependencies, there is no way to order them in a feasible sequence, such that each
task appears only after everything it depends on has happened before.

So, what we are going to do now is to show the other side of this, so what we said is if it is not a
directed acyclic graph, so it is directed, but with cycles, then topological sorting is always
impossible. On the other hand, what we need to argue is that if you give me a feasible set of
constraints, in the sense that there are no cycles, there are no cyclic dependencies, it is a DAG,
then I will always be able to complete it in some reasonable order. So, every DAG can be
topologically sorted.

(Refer Slide Time: 03:14)

So, how would we go about doing this? So, clearly I have to begin somewhere, so the first thing I
have to do is list out a task, which has no dependencies, it does not require anything else to be
done. So, there must be a vertex with no dependencies, there may be more than 1. If you look at
this graph on the right, we see that 0 has no incoming edges. So, 0 does not depend on anything. 1
has no incoming edges. So, I could start with 0 or I could start with 1. But in general, I need to
find such a vertex which has no incoming dependencies to start with.

As I complete the dependencies, a 3later vertex with depends on a few things now becomes
available, because everything that needed to be done before that is done. So, as long as we can find
vertices whose dependencies have already been listed, we can then list these. So, this is our general
strategy. So, we first start with something which has no dependencies. And every time we have a
dependency satisfied, we strike it off the list. So, then vertices, which do have dependencies,
eventually, all the dependencies have already been listed, and then I can list them.

So, in order to apply the strategy, we need to of course, guarantee that there will be a starting vertex
with no dependencies. Otherwise, there is no way we can start. And then we also have to guarantee
that eventually, every vertex which has dependencies will find all those dependencies listed out.
So, we have to make sure that every vertex, remember that we said that whenever we have a DAG,
we can do a topological sort, this is our claim.

And this strategy above says that we are going to do this by starting with a vertex with no
dependencies. So, first we have to show that such a vertex exists. And then we have to argue that
as we progress, we are definitely going to eliminate all the vertices in our into our list and finally
finished by doing all the tasks or listing them all out in this topological sort. So, we need to
somehow justify these claims in order to proceed with this strategy.
(Refer Slide Time: 05:09)

So, remember that in a directed graph, we talk about the indegree and the out degree of a vertex as
opposed to just the degree. So, in an undirected graph, the degree of a vertex refers to the number
of edges which are incident on that vertex, how many edges have that vertex as an endpoint. But
in a directed graph, these edges could either come in, or they could go out. So, the indegree is how
many edges are pointing into a vertex v.

What we are looking for is a vertex with no dependencies, that means nothing is pointing into it v
does not depend on anything, so there is no edge of the form u comma v. So, why must there be
such a vertex? So, the claim is that every time must have such a vertex within degrees 0. So, let us
suppose we have vertices within degree not 0, so pick any 1. So, we start with the vertex v, which
has indegree greater than 0. Since it has indegree greater than 0, there is at least 1 edge coming
into that vertex. So we can follow that edge backwards and go to a preceding vertex. So, we can
go back from v from a proceeding vertex.

So, let us say supposing we start here, we say that this vertex, has some incoming edge. So, I go
across this vertex, I go backwards and I come to this vertex. Now I say this vertex also has a
nonzero number of incoming things it has integrated in 0. So, I will keep doing that. So, I will start
here, then I will go back here and then maybe I will go back here. And then maybe I will go back
here. And then I stop, because I cannot go back any further.
So, in this particular case, this graph is acyclic and I have stopped at 0 by starting at 6. If I started
at 7, I could follow a different path. For example, I could go from 7, I could go to say, 4 and then
0, or from 6, I could have gone from 5 to 2 to 1 and so on. But whichever way I do it, eventually
all these paths will have to stop. And why is this the case?

(Refer Slide Time: 07:11)

Well, supposing I started some v, let me call it v1, or v0, if you want it, then I come back to another
vertex, which is v1. And then I cannot stop because v1 has indegree greater than 0, so I have to go
back to another vertex which is v2. Now, if I hit the same vertex again, then I have a cycle. So, let
me assume that v1 is different from v0 and v2 is different from v1 and v0. So I am continuously
hitting new vertices as they go along. If I do not hit a new vertex, every time I go backwards, I
have already found a cycle and I know this graph has no cycles, but on the other hand, this graph
has only n vertices.

So, if I started v0 and I do 1 step, I get back to v1, if I do 2 steps, I get to v2. So, if I do 𝑛 − 1 steps,
I have reached (𝑣𝑛 − 1). So, after I have done this, if I do it one more time, if I do an nth step
backwards, then I cannot find a new vertex anymore, which I have not seen before, because all the
n vertices in my graph have already been traversed somewhere in that path that I have seen so far.
So therefore, the new vertex must be going back somewhere here. So, it must be one of the vertices
already seen. So, there must be a cycle.
So, there is a directed cycle. And since it cannot be a directed cycle, this cannot happen. So, this
is a complicated way of proving something is called proof by contradiction. So, you say assume
that everything has an nonzero indegree, then I can find a path, which is arbitrary length, in
particular of length n, which will visit 𝑛 + 1 vertices. And since n + 1 vertices must repeat a
vertex, there must be a cycle and this cannot happen. So, this is why we will always have a starting
point, we will have always have a starting point, which is an indegree vertex with a vertex with
indegree 0.

(Refer Slide Time: 08:56)

So, this claim is now a fact. So, that was a proof that we have a vertex which is guaranteed to have
indegree 0. So, in this particular graph, as we said, we have the vertices labeled 0 and 1, which
both have no edges pointing into them. Now, what do we do? Well, we list it out, because we start
from there. And once we list it out, we kind of pretend that it is no longer a constraint because it
has been done.

So, if it was a constraint for somebody, for example, if we look at vertex 4, for instance, so vertex
4 requires vertex 0 to be completed, but if I list out 0 saying 0 is done, now, 4 has no longer any
constraints, because this constraint is gone. So, this was a claim that as we go along, the constraints
will go away until you can list out the later vertices. So, if I delete that vertex that I just found, and
all the edges from j, what happens?
So, if I delete this, and then I delete all the edges that point out of 0, then I am left with a smaller
graph in which there is 1 less vertex, and a few small, fewer edges depending on how many edges
I had connected to that original vertex. But notice that in this process, I have only removed edges
from a directed graph. And if the original graph had no cycles, this must also have no cycles,
because I have not put back any edge between 2 vertices which are not already connected. So,
what remains after this is again a DAG. So, I take a DAG, I remove any vertex from it, it remains
a DAG it may not be connected for if I have done it badly, but at least it cannot have any cycles.
So, it will be directed and it will be acyclic.

And therefore, by the same argument in the new DAG, that is, after I have believed this constraint
has been satisfied, I must again have some vertex with indegree 0. So, at every stage, I have a
DAG, whenever I have a DAG, by that earlier argument, there must be a vertex with indegree 0.
So, at every stage, there are at least 1 vertex which I can remove. And I keep doing this and after
n minus after n stages, I must have removed all the vertices. So, this is how the procedure works.

So, we repeat this process until all the vertices are listed. And we are guaranteed that at every
stage, start with a DAG, remove a vertex of degree 0, I have another DAG, therefore, I have another
vertex of indegree 0, remove that I have 1 more, and so on. So, that is why this process is
guaranteed to make progress, and is guaranteed to exhaust all the vertices in my DAG.

(Refer Slide Time: 11:24)

So, the first step to implement this as an algorithm is to compute the indegree. So, assume that we
have the graph presented to us as usual as an adjacency matrix, then we know that the incoming
edges are in the columns. So, remember that the row i has all the outgoing edges from i, and the
column i has every entry of the form j comma i, which is edge from j to i. So, if I look at the
column i it has all the entries for the incoming edges.

So, if I just walk down column i and add up the 1s, remember that this matrix has 0 1 entries, if I
just add up the 1s, I will get the indegree. So here, we have got this graph and by doing this one
scan, although we can do it pictorially in this particular case, by doing this one scan, we can count
the incoming arrows. So for instance, this has degree indegree 2, because there are 2 edges coming
in this as indegree 4, because there are four edges coming in. But this is not something we have to
do with the picture, we can actually just mechanically do it using the adjacency matrix.

So, now that we have this, now we can compute. So, we have an alternative, a second list, in some
sense a list of indegrees of every vertex, we are, we know in advance that there must be at least 1
of these vertices which has indegree 0, we do not know which 1, but we can find it by just scanning
down the list and looking for a 0. So, we go down the list and we look for a 0 and perhaps we
decide to pick so there are 2 in this case, as we know.

So, we have both what is 0 and 1. So, let us suppose we choose to do this one. So remember, our
procedure says list it out and remove it from the graph. So, we list it out here to the bottom is our
list, and we remove it from the graph. So, the edges, which are now pointing out of vertex 1 have
been removed. And in this process, the targets of those edges indegree has reduced. So, when I do
this, I had I had an edge I claim like this.

So, this indegree and this indegree now will change. So, when I remove that vertex, I must also
simultaneously update the indegrees, I do not have to scan all the indegrees. Again, I only have to
look at the row for i the vertex i just deleted as i, I only look at the row for i and every ij that I have
as an edge, I look at the degree of j and I reduce it by 1. So in this case, I had 1 to 2 and 1 to 7. So
I go to vertex 2 and reduces indegree by 1, I go to vertex 7 and reduces indegree by 1. Now again,
I have a DAG a smaller dag, again, I must have a vertex, at least 1 of indegree 0 here, I have no
choice, I have only this one. So, I remove that and list it out. And now again, these are the three
vertices which were getting edges from the vertex 0, which has this just deleted.
(Refer Slide Time: 14:10)

So, now I have to reduce they are indegrees by 1. Now I have a wide variety of vertices with
indegree 0, which I can enumerate next, because all of them depended on 1 and 2. So, I pick any
1 of them. So, for instance, I picked 3, the one in the middle, I picked 3 and I remove it, so I list it
out. And now 3 was pointing in this direction and this direction, so this 2 will have to reduce and
this 3 will have to reduce. So, I reduce them. Now again, I have 2 choices, 2 and 4. So, perhaps I
do 2 next.
(Refer Slide Time: 14:42)

So, I take 2, and now I reduce the indegree of 5 by 1. Now I have these 2 candidates to enumerate
next, so perhaps I seek vertex 5. And then I have to reduce the indegree of 6 by 1. And now maybe
I do 6 next.
(Refer Slide Time: 15:02)
And then I reduce the indegree of 7 by 1. But I still cannot enumerate 7 because it has indegree 1,
but 4 is left with indegree 0. So, I can enumerate for next, reduce the indegree of 7 to 0. And
finally, I can enumerate 7.

(Refer Slide Time: 15:22)

So, if we look at our original graph, this is what the graph looked like. So, what we have said is
that we did this first, then this, then this, then this, then this, and this, then this. And then. So, this
is the sequence in which we enumerated it, perhaps not the most obvious sequence that you would
have thought of, we might have thought of doing it top to down to 01, and then maybe 234, and
then 567. But this is a valid sequence.
(Refer Slide Time: 15:52)

So, what if we had adjacency lists instead of adjacency matrix, so we said did an adjacency matrix
representation, we can find the incoming edges by looking at the relevant column, we look at
column i, and we have done it. In adjacency list, we only have outgoing edges. If I look at the list
for i, it has only edges pointing out of i. So, how do I get the edges pointing into i. Well, you do
not do it in one shot.

So, for the column thing, you compute the indegree of i in one shot by looking at the column for
i, here you look across all the lists, so you will start with vertex 0, and look at the list of things that
0 is pointing to. So, 0 is pointing to, in this case, 2, 3, and 4. So, what you will do is you will have
separately indegree of 2, indegree of 3 and indegree of 4. So, when I see this 2, I will do a + 1 here,
when I see this 3, I will do a + 1 here, when I see this 4, I will do a + 1 here.

Now I come to 1, so 1 has outgoing 2 and 7. So, now I will do another + 1 here, and in 7, I will do
a + 1, and so on. So, you basically scan all the lists from top to bottom. And for each outgoing as
you see, you go to the corresponding target indegree and incremented by 1. So, even if you have
an adjacency list, we can do a simple scan, and update all the indegree to start with. After that it is
only a matter of checking the indegree as you are deleting them. So, it does not matter whether it
is incoming or outgoing.

So, we have the usual caveats as we had before, that is if you are doing an adjacency matrix,
everything takes order n time because you have to look at all the outgoing edges or all the things
but as we are doing something, which is the adjacency list, you can do it in time proportional to
the total number of edges.

(Refer Slide Time: 17:20)

So to summarize, we have seen from the earlier lecture, that directed acyclic graphs are a very
natural way to represent dependencies. And one of the fundamental problems in such a situation
is to find a feasible schedule, find a sequence in which you can perform the tasks or do whatever
we need to do, which does not violate any of the constraints. So, something must be listed before
something that that depends on it.
So, what we observed is that in any DAG, there has to be at least one vertex which has no
dependencies, there is something within the graph version on the directed acyclic graph
representing the dependencies has indegree 0, so we can list it because it has nothing that has to
come before it. And eliminating a vertex from a DAG gives us back another DAG smaller DAG,
possibly disconnected.

So, we could have a DAG for instance, we look like this, this is a DAG. So, it says that I must do
this, say this is my first task. And I must do this before 2 and 3. Now once I have done this, now I
have a DAG, which consists of just 2 and 3. So, we might end up with a disconnected graph, but
it is still without cycles. And therefore, by the same logic, it must have something of indegree 0
and so we can keep repeating and that is why this process works.

Now, the other thing to notice is that more than one topological sort is possible. So we saw that
when we looked at that example of how to set up the room, we said that for instance, tiling the
floor and plastering the walls can be done in either order. And in particular, what happens is that
when we end up with multiple choices for indegree 0 vertices.

So, if we look at our previous example, for instance, I could have started with 0 or with 1 we chose
to start with 1 if we started with 0 we would have got a slightly different sequence. Similarly, we
had a situation when we had 2 3 and 4 all available to us with indegree 0. So, we chose to do 2
first we could have done 3 first, we could have done 2 first we could have done 4 first.

So, whenever we have multiple degree vertices with integral 0 topological sort does not necessarily
force us to take one or the other we might choose to take the smallest one in which case we get
one particular order, but there are multiple orderings possible. So, this is a thing that we need to
remember that Topological sort produces a sequence which is compatible, but this is by no means
the only sequence there are multiple topological orderings possible.

In particular, if you have no dependencies if I have all the tasks are independent, that any ordering
is possible. So, if I have this is basically if I have n independent tasks, then I would have 𝑛!
orderings. So, the number of topological orderings can be very large so we are not really interested
in computing at this point the number of topological ordering we are also interested in finding one
of them.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Indian Statistical of Technology, Madras
Lecture 69
Longest Paths in DAGs
(Refer Slide Time: 0:09)

So, we are looking at directed a cyclic graphs as representations of sets of tasks with
dependencies. And we said that there are two natural problems on this. One, is to find a
sequence of feasible sequence in which I can do the tasks and this was topological sort.
The other one was to try and find out how many steps I need to perform these tasks in an
optimal way.

(Refer Slide Time: 0:36)

So, this is my DAG. So, if I do a topological sort and I list out the vertices such that
whenever there is an edge from i to j, i appears before j and this gives me a feasible
schedule. Now, we may be interested in finding out how fast we can do this if tasks which
have no dependencies can be done together.

So, for instance, in a topological sort, we would have to put 0 before 1 or 1 before 0 as a
sequence. But if we had no dependencies, and it is possible to do 0 and 1 parallelly, then
we could do 0 and 1 at the same time, if there are resources to do both. So, a good example
is when you are taking courses, so when you are taking courses, you do not do just one
course in a semester, you can do many courses. And as long as the number of courses which
are available for you to take are reasonable, you can do all of them, say three, or four or
five, maybe in a semester.

So, if the DAG represents prerequisites between the courses, and each course takes a
semester, so you can finish and go to the next course only in the next semester, then the
natural question to ask is how many semesters do I need to complete the remaining
requirements? So, I have a set of requirements, and they have some prerequisites between
them, how many semesters do I need from now to finish the program satisfying these
requirements. So, this is the problem that we want to solve now.

So, in this particular case, for instance, as I said, we can do 0 and 1 together. So this can be
done in the first instance, then I can do 2, 3 and 4. In these are courses, then in the first
semester, I can do 0 and 1 course, in the second semester, I can do courses 2, 3, and 4. But
now I am stuck because I cannot do 7 until I finished 6, I cannot do 6 until I finished 5, 5 I
can do because all the prerequisites are 5, namely 2 and 3 are done. So, in the third semester
I can do 5, then in the fourth semester I can do 6 and finally in the fifth semester I can do
7.

So, if this was my sequence, I mean, this was a DAG representing my prerequisites, then I
cannot do better than a sequence of 5 semesters. So this is the problem that we want to
compute.
(Refer Slide Time: 2:41)

So formally, it consists of finding the longest path in a DAG. So, what we want to do is
really compute the longest path to every vertex, and then if we have that, then the longest
path among these will be the longest path overall, if I compute the longest path from one
of the starting points, so when can I do course 6, when can I do course 3? If I know this,
then among all these, if I have the maximum, some course requires me to do it in the fifth
semester, then I know overall, I need five semesters. So, one way to solve the longest path
problem is to solve this question of computing the longest path to each vertex.

So, with the assumption that the longest path to an initial vertex, so the path is how many
semesters before in the course that example, how many semesters do I have to wait? So,
with 0 waiting, I can do anything which is indegree 0, so that is a good starting point.

So, in this case, I could do these two things right to begin with, so this is indegree 0. Now
what happens next. So, if I have indegree which is not 0, then supposing I look at this
vertex, it can happen only after 2 and 3 have happened. So, if I know that the longest path
to 2 is some k and the longest path to 3 is some l, then I must wait maximum of k and l, I
cannot finish 2 and 3 until max of k and l happen.

So, supposing this is fourth semester, fourth semester, and this is going to happen in the
fifth semester then, only in the sixth semester can I get to course number 5, so, this + 1, so,
the longest path to i is going to be 1 + the longest path to every incoming neighbor of i. So,
remember this set comprehension notation so, this is the set of all the numbers longest path
j for j, i in the edge set, all the incoming edges coming into i, I take that, take the maximum
of all of those and add 1 because I now have to go to the next semester.

(Refer Slide Time: 4:54)

So, the longest pathway to i is 1 + the maximum of all the longest paths to the incoming
neighbors. So, to compute this, I need to know the neighbors, longest paths for all the
incoming neighbors. So, if I know that, then I can take the maximum of those and add 1 to
it. But how do I know those? Well, I know those if I have already calculated those before
this, and I would have calculated those before this if I calculate them following the
topological sort.

So, if I sort out the sequences according to the dependencies, then by the time I come to i,
and I want to compute the longest path to i, all the incoming neighbors of i should already
have been listed before i. And if I am computing longest path as I listed, okay, then that
information about all these incoming neighbors will be available to me when I come to i.
So, that is the strategy that we are going to do, we are going to compute longest path in
topological order. So, we are going to compute the longest path to every vertex as we
compute topological sort, in fact.
(Refer Slide Time: 5:59)

So, more formally supposing this is a topological ordering of i, of V, right. All the
neighbors of some vertex ik, so i0 to in - 1 is some reordering of the numbers 0 to n - 1.
So, the vertices are 0 to n - 1 and I have rearranged them in some sequence, and that
sequence is i0 to in - 1. If I look at a particular entry, the kth entry in this was a k + 1 entry
ik, then all its neighbors in the graph must appear before it.

So, if I compute from left to right, then I can compute for each of these ik, the longest path
based on the values that I have already computed to the left. And since I am doing this as
I am going along, I do not have to actually enumerate all the vertices before I compute, as
I come to a vertex I already have the information to compute its longest path information,
so I can do the overlapping computation of the topological sort and the longest path, I can
do them at the same time. While I am computing topological sort, I can simultaneously
compute the longest.
(Refer Slide Time: 6:57)

So, how do we do this? As before, we compute the indegree of every vertex, right and we
also initialize this longest path, okay to be 0 for all vertices, because initially I do not know
anything. So, the blue number indicates my current knowledge about the longest path, right,
and the red number is the indegree which we are using for topological sort.

So, now we do this topological sort, as we go along and as we go along we update, so the
first vertex, remember we did last time, we will do the same order, we picked 0, I mean,
pick 1 which has indegree 0. So, if you have 1 which has indegree 0, then two things
happen. One is we are going to update the indegree of everything which is pointing out of
it. But we are also going to now update, we have now, we know definitively information
about vertex 1. So, this maximum has a kind of monotonic property.

If I have the maximum of a set, it will only get more if I go along, so if I already know that
the incoming vertex 1 has longest path 0, I cannot have of course, 0 is in this case a very
simple value because a path cannot have negative values, but I cannot, the maximum
cannot go below 0.

So, I already know that if I know that the, if I have frozen the value for the incoming vertex
1 as having longest path 0, then I already know that the value for 2 must be 1 + that and I
can keep updating this. So, I can compute the max of those things incrementally and keep
moving on.

(Refer Slide Time: 8:30)

So, what I will do is I will remove that vertex as before and now I will update as I said,
these two entries. So, both of these entries, so the 2 will become 1 for the incoming indegree
of vertex 2, the 4 will become 3 for indegree of vertex 7. But for both of them, the longest
path will now go from 0 to 1 because I know that it is at least 1.

(Refer Slide Time: 8:52)

Next, we did this one. So, when we do this one, we will now update as before the indegrees
of this, but we will also update these. Now notice that for 2, vertex 2 it already believes its
longest path is 1 and if I took the new information from 0, it is 1 + 0, which is 1, so nothing
is going to happen, so 1 is going to remain 1, but for 3 and 4, where I had previously
believed, I mean this without any justification that the longest path is 0. I am now going to
update it to 1 + 0 as 1.

(Refer Slide Time: 9:29)

So, when I remove the 0, the next step is I make all the indegree 0 because they have all
now got no vertices pointing into them, but I also update the longest path for 3 and 4. I
have also updated for 2 but no change happened because 2 already knew that it was of
longest path 1.
(Refer Slide Time: 9:49)

Next, I will pick 2 in this, next I pick 3, sorry, this is what we did last time. So, notice that
on the bottom we are keeping track of this information, right, so we are keeping track of
both the topological order, the order this, so the top row indicates the vertex number as I
enumerated, so I first enumerated vertex 1, then I enumerated vertex 0 and in the lower
row we are keeping track of the longest path, which incrementally we are getting a final
value for every vertex as we enumerate.

So, now output vertex 3 and this is going to update these values, so 3 has longest path 1, 7
believes his longest its path is 1, but now it must be at least 1 + 1, 2. So, 7 is going to
change similarly, 5 is going to now move from 0 to 1 and of course, the indegree are going
to reduce.
(Refer Slide Time: 10:35)

So, I remove the 3 and then I increment the longest path here, this should be, so this is also
notice that see, the path here was 1, so this goes directly from 0 to 2, it goes to something
which is 1 + the maximum known incoming thing. So, the maximum known incoming
thing is now 1, so it goes to 2 and this was already 1 and it goes to 2 because, so it is not
that it is incrementing, it is computing 1 + max, so this goes to 2, this also goes to 2 and
the degree has reduced.

So, we are overlapping this topological sort, with this longest path computation. So, in the
topological sort, we are decrementing the indegree each time we remove an edge, in the
max, in the longest path computation we are doing a max. So, sometimes it changes,
sometimes it does not change, when it does change, it might change by more than 1 and so
on, you have to keep track of what was the incoming max that you saw so far and add 1 to
it.
(Refer Slide Time: 11:39)

So, next we did 2, so if we do 2, then because 1 + 1 is 2, nothing is going to happen here

but the indegree of 5 will reduce, so I remove 2 and now the indegree of 5 reduces but its
longest path does not change.

(Refer Slide Time: 11:43)

Next, I do 5 and when I do 5, this now influences this, and this says this must become 3,
right, because the incoming edge to 6 has longest path 3 at the other end, 2 at the other end.
So, the incoming path to 3 must be at least 2 + 1.
(Refer Slide Time: 12:00)

So, now the degree of 6 will reduce from 1 to 0, but the length of the longest path will go
from 0 to 3 and notice as before that we are, as we are enumerating the vertices, we are
also enumerating the longest path, because now I have enumerated it, its longest path is
known. Next, we did 6, so when we remove 6, then this will go to 1, but this will go up to
4, so I remove 6.

(Refer Slide Time: 12:25)

And now 7 jumps from longest path 2 to longest path 4 because I have discovered this path
of length 3 coming up to 6, which is one of the incoming neighbors. But I cannot enumerate
7 yet, because its indegree is not yet 0.
(Refer Slide Time: 12:42)

So finally, when I output 4, at this point, the indegree of 7 becomes 0 but there is no change
to the length of the longest path, because the longest path does not come through 4 it has
already come through 6, and 7 has already discovered that. It is just that I cannot finalize
it until I actually reach the enumeration of 7.

(Refer Slide Time: 13:02)

So finally, I list out 7 and now I have the longest path with every vertex listed below.
(Refer Slide Time: 13:07)

So, if you go back to the graph, you can verify for instance, that the longest path to 7 goes
through say 1, 2, 3, so this is one longest path so 1, 2, 3, 4. So, basically the longest path is
in terms of number of edges, at how many things I have to do before I come here, how
many semesters I have to work before I do course number 7, there are multiple longest
paths in general, so in this particular case, there is only one longest path to 7, I guess you
can find but no, you can also find the longest path which goes this way. So, you can take a
path, which goes from 0 to 3 to 5 to 6 to 7, so this is also a longest path of length. So, just
the fact that you have a longest path, does not mean that is a unique one.

(Refer Slide Time: 13:51)

So, just to reiterate that directed acyclic graphs are a natural way to represent dependencies.
So, we saw before the topological sort is how to get a feasible schedule, right, how to
extract a sequence in which I can do these tasks, such that all dependencies are satisfied
before I come to a task.

But now, we also saw this problem of how to compute the shortest duration that I need if I
am allowed to do tasks in parallel. So, that is this longest path problem and we said that the
longest path can be computed in an overlapping way with the topological sort. Because
once we process of all the dependent vertices of a vertex, then we have enough information
to process that vertex itself.

So, in the same sequence that we enumerated, we can also associate with each vertex, its
longest path and incrementally build this up in parallel, so we do not have to take any extra
work, while we are doing topological sort we can also compute longest path. Now, you
might ask whether longest path makes sense in a graph with cycles because when I have
cycles, I can go round and round.

I can go to a vertex and then come back to this place and go ahead but if you remember we
made a distinction between a path and a walk. So, we said a path cannot repeat a vertex, a
walk can repeat a vertex. So, going around the cycle and then proceeding technically is a
walk not a path. So, if you use this literal interpretation of path, so we can have longest
paths in cyclic graphs also, it makes sense, what is the longest sequence of edges I can
travel before I repeat a vertex? This could be at most n - 1 but is it n - 1? So, that is a
question. So, you could ask the same question about longest paths in a graph which has
cycles.

So, in a directed a cyclic graph, we came up with a reasonably efficient algorithm which
processed each vertex only once, and then depending on how we represented it as an
adjacency list or a adjacency matrix, we had to do some scanning of that to find out the
incoming and outgoing edges. But beyond that, it is a very reasonable algorithm, it takes
time proportional to n or n + m, I mean n squared or n + m depending on how you are doing
it.
What happens in this general case, well, in this general case, actually turns out to be
surprisingly hard. So, if you have cycles in your graph, you can definitely define the notion
of a longest path by looking at only paths in which vertices do not repeat, right. So, you
know, that is going to be n - 1, but to know whether it is n - 1 or not is surprisingly hard.

So, actually there is no known efficient way to do this. Of course, you can do it by brute
force, by examining every possible path in the graph and counting its length and then taking
the maximum and essentially what we do not know right now is there any better way than
that to do it, right. So, there is a huge gap between the longest path problem and DAGs, in
the longest path problem and directed graphs which are not DAGs.

So, in directed graphs which are not DAGs, the longest path problem is very difficult
computationally compared to the relatively simple problem that we have when we have
DAGs.
 Mathematics for Data Science 1
Professor Madhavan Mukund
Indian Institute of Technology, Madras
Lecture 70
Transitive Closure
(Refer Slide Time: 0:18)

One of our original motivations for looking at graphs was to visualize relations. So, let us
go back to relations. So, supposing we have a relation R on a set S. So, relation R on a set
S remember is a subset of the Cartesian project S cross S, so S cross S is all pairs S1, S2
taken from S and some subset of these form a relation.

So, concretely, for instance, supposing we have a set of people, maybe a family, and then
we want to find out the family tree in some sense, then we might represent as a relation
when two people are related as parent and child. So, we can say that p, q belongs to R so,
R is the parent relation, whenever p is the parent of q, so, p, q belongs to the relation R, if
p is a parent of q.

Now, given this parent relation a very natural question is, what is the grandparent relation,
what is the great grandparent relation and so on. So, in general what is the ancestor relation.
So, we have is so and so is p an ancestor of q, is in the family tree is q a descendant of p?
So, how would we do this? Well, to find out whether q is a descendant of p or p is an
ancestor of q, we have to trace some sequence of relationship, we have to find a child of p
and for that child, we have to find child of that child and so on or we have to find a parent
of p.

So, in this case, we are looking for ancestor so, p to q is a parent child relationship. So, we
start with some p let us call it R0, we have to find the sequence R0, R1, Rn we have to find
the sequence of people says that R0 is a parent of R1, R1 is a parent of R2, R2 is a parent of
R3 and so on, Rn-1 is a parent of Rn, so this is an ancestor sequence. So, R0 is an ancestor
of everybody to the right in sequence and particular if R0 is P and Rn is qr, these are our
two people we are interested in finding out, then we have established that p is an ancestor
of q.

So, this is a new relation. So, this relates pairs of people who are connected by a sequence
of parent relations. So, this has a name, this is called the transitive closure. So, transitivity
says that if A is related to B and B is related to C, then A must be also related to C, this is
the definition of a transitive relation.

So, parent is not a transitive relation, so what would happen if we made parent a transitive
relation, if we forced parent to be a transitive relation, so, we said that whenever somebody
is related to somebody and somebody is related to the other person, first person is related
to third person, then that is the ancestor relation.

So, this is what happens when you close, we make the parent relation closed under
transitivity, we force transitivity onto it, we compute what is called a transitive closure,
okay, this is what we have. So, we normally denote this by R + , R+ means that we have to
apply R one or more times to go from p to q. So this is a transitive closure of R.
(Refer Slide Time: 3:32)

So, our question is, of course, how are we going to calculate this transitive closure in a
systematic way? So, recall that we can represent any such relation as a directed graph in
general. So, we represent, the vertices represent the set S, V=S and each edge represents a
pair in the relation. So, if u,V is a member of the relation R, then we put an edge from u to
V and this is the only case in which you put an edge from u to V. So, the edges are exactly
the pairs which are related by R.

(Refer Slide Time: 4:07)

So, what we have defined as R+ can be calculated in the graph as being a path, right, so we
can say in this case that V9 is related by R+ to V0, because there is a path from V9 to V0.

(Refer Slide Time: 4:25)

So, how do we find these paths? Well, essentially, we want to find all such pairs for which
V9 is related and we said that if you do not focus on a single path, we can just calculate
reachability, what all is reachable from every vertex and this we can do using breadth first
search and depth first search. So, in the R+ case, we are interested in finding the
reachability for every i in every j, right, we want to know for every i and every j, whether
it falls into R+ or not. So, we have to compute this reachability for every starting point.

So, one way to do this is to just perform this BFS, DFS starting systematically we started
the 0th vertex, you perform BFS, then you know everything which is of the form 0, i, then
you start at 1 and you perform, notice that this is a directed graph. So, you have to do it
because if there is a, if 0 can reach j, it is not obvious or is not required that j can read 0 in
a directed graph. So, you have to perform this from every vertex to find out what all pairs
fall into R +. So, this is one way to do this using what we already know.

(Refer Slide Time: 5:25)

So, here is another strategy, so the other strategy is to look at the adjacency matrix. So, in
this adjacency matrix, we put a 1 if there is an edge from i to j. But another way of thinking
of an edge is that there is a path of length 1 from i to j. A single edge is a path of length 1
and what we want is to create an similar adjacency matrix for this expanded list.

So, remember that R+ is also relation, so R+ we said is also a relation, so this also
correspond to some new graph G plus, because every relation we can draw a graph, just
add the same set of vertices in this case, but now put an edge if there is an R+ relation
value, relationship between i and j.

So, this G+ will correspond to an adjacency matrix A plus, so, that is this A+ here, we
want an matrix A+ whose ij entry is 1 , if there is a path of length 1 or more from i to j, so,
if there is an edge directly, so an ancestor, a parent is an ancestor, parents parent is an
ancestor path of length 2, parents, parents parent is an ancestor, so that is length of 3 and
so on.

So, therefore we have any path of length 1 or more between i and j that will be an edge in
the R+ relation and in the G+ graph, and we want to compute A+ from A, so that is our goal.
So, we know one way to do it, which is to go to breadth first search, and then fill in A+
from breadth first search by doing breadth first search from every vertex. But the question
is, can we do it directly using just A?
(Refer Slide Time: 7:08)

So, we have now on the left, we have A and A denotes paths of length 1. So, as the first
step let us try to compute paths of length two, right. So, paths of length 2 go from i to some
k, to some j. So, this is a path of length 2, we could also be back to itself, path of length 1
cannot go but we could have a path of length 2, which goes from i to i. So, this is the matrix
that we want to call now A2 for instance, A2 so A, which we can think of A1 if you want,
represents paths of length 1, A2 represents paths of length 2.
(Refer Slide Time: 7:55)

So, how do we fill in the entries of A2 from A? So, A2 ij is 1 if there is some k such that Aik
is 1. So, there is i to k there is a path of length 1 and k to j there is a path of length 1. So, I
look at this entry, why is there an A2 entry from 0 to 0? Well, I claim that there is a 0, 1
entry, not a 0, 1 entry sorry, there is a 0, 4 entry, so I have a path from 0 to 4, because of
length 1, and then I have a 4, 0 entry. So, if I choose k to be 4 and i to be 0 and j to be 0,
then I get A2 ij as 1 because for k =4 I have this.

(Refer Slide Time: 8:47)

So, in general, what I have to do is I have to look at, so if I do it systematically, I will start,
I want to fill in the row for 0. So, I look at all this, I look at the first entry, I can go, where
can I go via 1? So, if I say 0 to 1, then I look at the 1th at the entry, so 1 can go to 2, so
therefore I have an entry 0, 2. So 0 to 1 to 2, then 0 can go to 4, so I should have. Sorry,
yes, so 0 to 1 cannot go anywhere else, so 1 has only one outgoing edge so through 1 I
cannot go anywhere else. So, 0 can go to 4, so I now look at the outgoing edges from 4, so
0 to 4 to 0, right, so that is how I get it.

Then I have 4 goes to 3, so I have 0 to 4 to 3, so I get this entry and finally I have 4 goes
to 7 so I have 0 to 4 to 7, so I get this entry. So, in this way, I can compute all the entries
of the form A2 0,k, by finding the intermediate values by looking for something that 0
goes, 0 to k and then from k to j.

(Refer Slide Time: 10:02)

So, I can do the same thing for 1 now. So, I want to find all the entries of the form 1,j, so I
look at the, so 1 has only one outgoing thing, going to 2, and 2 has only one outgoing thing
going to 0, so the only new thing I discovered is 1 to 2 to 0, right, and that is the same.
(Refer Slide Time: 10:25)

Now, I will do the same thing for 2, so 2 has only one outgoing thing to 0. But 0 has
outgoing edges to 1 and to 4. So, I get 2 to 0 to 1 and 2 to 0 to 4. So, I get 1, so this should
not be here, this will be here, I get 2 to 1 and 2 to 4.

(Refer Slide Time: 10:47)

So, in this way, I can do this for all the entries, I can for every row, I can take the outgoing
edges 3 to k and the outgoing edges k to j, and add an entry 3 to j, right. So, by scanning
these two rows in this matrix I can compute A2 matrix. So, A2 represents all the paths of
length 2.
So, notice that the paths of length 2 do not subsume the paths of length 1, right, so for
instance, we had a path of length 1 from 0 to 1, but we have no path of length 2 from 0 to
1. So, these are paths strictly of length 2, they are not of length 0 or not of length 1, so A,
the first A has edges, the second one has paths of length 2, so I can go in length 2 from 0
to 0. But the fact that I can go from 0 to 1 in length 1 does not mean I can go from 0 to 1
length two. So that is paths of length 2.

(Refer Slide Time: 11:35)

So now, how do we go to path of length 3? Well, if I have a path of length 3, it must be of

this form, i, k1, k2 to j, right, so there must be some two things in between. So, I can split it
up whichever way I want, I can either take this point and say that I have a path of length 2,
followed by a path of length 1, or if I want, I could do it the other way, which is I could, I
could split it up here and say I have a path at this point.

Sorry, and say I have a path of length 1 followed by path of length 2, right. So basically, a
path of length 3 can be decomposed as two+ one or one+ two. And I already have explicit
matrices for 2 and 1, I know all the paths of length 1 are represented in A, I know all the
paths of length 2 are represented in A2.

So, I can say now A3, i,j is 1 if there is some k for instance, where there is a path of length
2 from A to k and there is a path of length 1 from k to j. So, now earlier, I looked at A and
within A I look for two entries. Now, I look at an entry in A2 and I look for an entry in A
and I try to match them up, I tried to find a k such that from i to k have an entry 1 in the A2
matrix, and from k to j, I have an entry in the A matrix. So this gives me A3, so you can do
the same calculation as before and you can come up with a new matrix. So I started with
A, I did one pass over A and I got A2. Now using A and A2, I can get A3.

(Refer Slide Time: 13:07)

So, now I can go from 3 to 4 in the same way, if I have a path of length 4, then it can be
decomposed as a path of length 3 followed by 1 edge. So, I can take the entries in A3 and
combine them with entries in A, so I can look for a k such that A3 i, k is 1 and A k, j is 1.
So, this now gives me A4 i, j, which will be 1 provided there is a path of length 3 from
some i to some intermediate k, followed by a path of length 1.

Now, you could also do this as 1+ 3, you can do it as 2+ 2, but let us just follow this general
rule, where we break it up into one less or + one. So, in general, if we keep going, right, so
if I want to, I already know paths of length l, and I want to extend it to l+ 1, then I will say
that Al+ 1 i, j is 1. If I already know that there is a k for which there is a path of length l
and I can extend it by 1 edge, a path of length 1 from k to j. So, I can go from i following
some l steps to k and then I can go from k to k in one step. So, then I can go therefore from
here to here in l+ 1 steps.
So, we just do the same thing again and again. The first time we are doing A combined
with A, second time we are doing A2 combined with A, in general we do Al combined with
A and each time I will go from here I get A2, from here I get A3, from here I get Al+ 1 and
so on. So, I can keep on building this matrix, which captures longer and longer paths.

(Refer Slide Time: 14:44)

So, now where do we stop? How long do we go on? Well, here we know that if there is a
path at all, then that path cannot have more than n-1 edges, because once I have traversed
n-1 edges, I have seen n-1 different vertices other than the starting point and therefore,
anything beyond that must repeat a vertex, so there must have been a shorter path.

So, therefore, if there is a path at all from i to j, it cannot have length more than 𝑛 − 1, so
I can stop with. So, once I have computed A to the 𝑛 − 1, right, I have got everything of
interest, I have got all paths of length 1, 2, 3, 4, up to n-1. And any path which is longer
than n-1 cannot be new. I mean, since it cannot contribute any new information to me about
whether or not i and j are connected by the relation or not.
(Refer Slide Time: 15:37)

So, remember that the reason we are doing this is for the transitive closure. So, we said that
i, j is in the transitive closure R+, if there is a path from i to j in the corresponding graph
for R. And we have observed many times that the length of this path is at most 𝑛 − 1. So,
therefore, we can combine all this information, right, so, we have the original A which is
the same as A1, right, path of length 1, then from that we computed as A2, then A3, A4 and
so on and up to the An-1.

So, I have this n-1 matrices, which gives me all information about paths from length 1 to
length n-1. So, what do I want to do? I want to say that there is an edge from i to j in the
R+ relation. If there is an edge somewhere in one of these, right, and I am going to write it
in this complicated way, I am going to say it is a maximum of the i, j entry in all the matrices
from k =1 to k =n-1, notice that this is strictly less than it. So, k goes from 1, 2 up to 𝑛 − 1
because it starts at 1 l, A to the l sorry, l goes from one to n-1, right.
(Refer Slide Time: 16:50)

So, this is, so, what does this mean? So, notice that each and each entry is 0 or 1. So, every
entry of these, these are all 0, 1 matrices, either there is a path of that length or there is not
a path of that length. So now, when I am taking this max, it is basically checking if all of
them are 0, that is there is no path of length 1, there is no path of length 2, there is no path
of length 3 and so on, the max is also going to be 0.

So, there will be an A+ entry which is 0, there is no path, but if there is a 1 anywhere, right,
in any one of those l positions from 1 to n-1, if any one of them is 1, then the max will
become 1, if there are many paths, there are path of length 3 and 7, it will still be 1 because
max of 1 and 1 will remain 1.

So, by taking max we are just recording is there at least 1, 1 in that sequence or not?
Sequence meaning across all these matrices in the ij th position is there at least one of these
matrices which has position value 1 at i, j. If so, the max will give me 1, if all of them are
0 max will give me 0.

So, in that sense, this A+ entry captures the fact that there exists some length path between
1 and n-1, between i and j and if it is 0, it means there is no such path, right, and we know
that if there is no path of length n-1, there cannot be a longer path because if there is a path
it must have at most length n-1, anything longer than that will be looping and will be
redundant.

So, therefore, A+ i,j is 1, if and only if there is a path from i to j and in particular, this path
must always be bounded by length n-1. So, what we have done, we can actually reformulate
it in a way that is called matrix multiplication, which we will not do right now.

But it is important to know that this, what we did is a very tedious calculation rows and
columns and all that is actually a very standard operation on matrices. So in this form, we
can write it as a sort of multiplication of matrices. It is not exactly what I have written here,
but for the purpose of this lecture, this is fine.

So, you can actually believe that this operation, the reason we are doing this with matrices
is that this operation on matrices is actually a standard mathematical operations on
matrices. So, though we have done this column and row chasing explicitly, saying we look
for a k here, we look for a k there, actually, you can actually do it directly as a matrix
operation. So therefore, it is a very standard operation is not something new.

(Refer Slide Time: 19:09)

So, to summarize, the transitive closure tells us so, this would be an R+ on top, the transitive
closure tells us whether there is a sequence of intermediate elements which connect two
elements, right. So, I have to start at p and go through multiple R edges to reach q, an
example of this was our ancestor relation. So, the ancestor from the parent, so a sequence
of parent edges generates the ancestor edge.

So clearly, since we visualize relations as graphs, this corresponds to a path, right, and in
general, these are directed edges because these relations are not assumed to be symmetric,
like the parent relation is certainly not symmetric, if A is parent of B clearly B is not a
parent of A, right. So therefore, in general, you follow a path in a directed sense, and this
is just a reachability question in graphs.

And we know that we can do this by repeatedly doing BFS and DFS from every starting
point, but what we have seen in this lecture is that alternatively, we can take the adjacency
matrix and do a form of matrix multiplication, we can do a form of matrix multiplication
to go from A to A2 to A3 and so on and stop with A to the n-1 because A to the n-1 records
paths of length n-1 which is the maximum length path, which is useful to us to find out
whether two edges, two nodes are connected and once we have got this we can take a look
for a 1 in any one of these n-1 matrices and if so, declare A+ ij to be 1.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Department of Computer Science
Indian Institute of Technology, Madras
Lecture No. 71
Matrix Multiplication
(Refer Slide Time: 00:14)

We have used matrices to represent graphs. So, we have this adjacency matrix. And when we
looked at the transitive closure problem, we suggested that it might be useful to describe the
transitive closure computation, using some operations directly on matrices. So, let us look at
matrix operations. In particular, we want to lead up to this notion of matrix multiplication.

(Refer Slide Time: 00:36)

So, a matrix in general is a 2-dimensional table of values, usually numbers, and it has a certain
number of rows, because it is 2 dimensions. So, it has rows and it has columns. So, usually, it
has some r rows and some c columns. r and c could be different numbers. And we write this as
r × c, or we call it an r×c matrix. So, r is the number of rows c is the number of columns. And
as usual, we will number the rows starting from 0 and the column starting from 0. So, the rows
are 0,1,2, up to r-1, and the columns are 0,1,2, up to c-1.

So, in the example that we have been looking at the adjacency matrix, r=c=n, which is the
number of vertices in our graph. So, we have 1 row and 1 column for every vertex in the graph.
So, we number our columns and rows is 0 to n-1. In this particular case, the entries in our
matrix are 0 and 1, but in general, they could be any numbers, even other values in general, but
we will look mainly at matrices where the entries are numbers.

So, for a completely different example, let us consider that we want to record some information
about freight traffic by rail between some cities. So, let us take say the 6 biggest cities in India,
Bangalore, Chennai, Delhi, Hyderabad, Kolkata, and Mumbai. And suppose we have
information about the volume of traffic, say in terms of 1000s of tonnes of millions of tonnes,
whatever is the appropriate metric for this. So, we have some information. So, we want to
represent this metric.

So, the first thing is that every city will correspond to an entry in the matrix in a row and the
column. So, because we are normally numbering our rows and columns from 0 to n-1, or 0 to
r-1, and 0 to c-1 in this case, again, notice that it is going to be a square matrix. So, it is going
to have 6 rows and 6 columns. Let us just number them in alphabetical order. So, 0 is Bangalore,
1 is Chennai, 2 is Delhi, 3 is Hyderabad, 4 is Kolkata and 5 is Mumbai, it does not really matter
in what order we put it.

So, this could be the matrix. So, this is not a symmetric matrix, as they say. So, there is a certain
amount of traffic say, from Bangalore to Chennai, but there is a different amount of traffic from
Chennai to Bangalore, because in general, there is no reason why the amount of freight traffic
should be the same in both directions. Notice that along the diagonal, there is no freight traffic
recorded from a city to itself, because that does not make any sense for us. So, this is another
way of using a matrix where we are representing some information about freight between pairs
of cities.
(Refer Slide Time: 03:02)

So, now, suppose we have this information, and it is counted over a certain period of time. So,
in India, you may know that the financial year starts in April. So, April 1’st is the beginning of
the financial year and march thirty first is the end. So, maybe for the first half of the financial
year, from April 1 to September thirtieth, we have a certain number of amounts of information
about this fr8 volumes, and that is the matrix on the left. So, this is the first 1, so, this is the
first half year and then we have a corresponding thing for the second half year. So, the numbers
are different.

So, for instance, here, there is only 247 whereas here in the same entry from Bangalore to
whatever was 4, you have Kolkata I think, there you have 399. So, now, a natural thing would
be to compute the freight volume for the full year given this information about the freight
volume in the first half of the year, and the fr8 volume in the second half of the year. And quite
clearly, the freight volume for any pair of cities is going to be the sum of the 2.

So, if you want to know how much traffic there has been from Delhi, which is city 2 to Calcutta,
which is city 4, you look up the entry for the first half of the year, it says okay, there were 595
million tonnes or whatever it is from Delhi to Calcutta in the first half. And then the second
half, there are 326. So, clearly put together it is 595+326, which is 921. And this is how we
would compute this information.
(Refer Slide Time: 04:22)

So, some notation. So, if we have a matrix, then remember our rows are numbered 0 to r-1 and
our columns are 0 to c-1. So, we will look at the I’th row in the J’th column, and that will give
us 1 entry in the matrix, where the row intersects the column and we will refer to that as M i
comma j. So, M i comma j is the entry in the matrix at row i and column j.

So, now here is our information about the 2 half years again, presented on the right-hand side.
So, these are the 2 matrices that were given. And we want to compute the aggregate volume
for the whole year into a new matrix, call it C. So, I have used A for the first half year, B for
the second half year, and now we want to compute a matrix C, which represents the total
volume for the whole year. And as we said before, the ij’th entry, the entry in row I column, J
of C is going to be the sum of the ij’th entries in A and B. It is quite obvious.

So, if we want to look at a pair of cities and find out the total traffic between the 2 cities, you
add up what traffic was there in the first half of the year to what traffic was there in the second
half of the year. So, when we do this, we are doing this for every element. So, we are taking
this element adding this element and getting this element, we are taking this element, this
element and getting this element. So, we are doing it in the same way for every element of the
matrix. And so, we can just concisely say that we are adding these 2 matrices. So, this is called
matrix addition.

So, when we have a matrix A and a matrix B, and they both have the same size, the same
number of rows and the same number of columns, then for every position in A there is a
corresponding position in B. If I look at row ij, in column J A, there is a row in column J and
B because they have the same size. And So, I can now add these entries and put it into a third
matrix of the same size. And I will just write C =A+B. So, C =a+B captures this fact that for
every row and every column, I am adding the entries in A and B to put them into C. So,
remember that we were not so interested in adding matrices but multiplying them. So, what
should we do for multiplying matrices?

(Refer Slide Time: 06:25)

So, can we multiply matrices the same way? So, let us take simpler numbers. So, here are 2
matrices A and B. And suppose we do the same rule, the same rule would say instead of adding
Aij, into Bij multiply Aij, by Bij. So, here, for instance, if I look at C 1, 0, it is A 1, 0, 4 ×B 1, 0,
5, 4 × 5 is 20. Similarly, if I look at, say for instance, A 3, 2, it is 5, B 3, 2 is 4. So, again, it is
20. If I look at this entry, it is 9 × 4 is 36. So, I have just multiplied entries, one at a time.

So, this could be a candidate for the operation of matrix multiplication is the obvious way in
which we can take addition and replace addition by multiplication.
(Refer Slide Time: 7:14)

Now, it turns out that this particular way of manipulating matrices, there is nothing wrong with
it, it is just that it does not turn out to be very useful in practice, addition made sense for us.
Multiplication, like this usually does not give us any useful number. So, there is a different
operation, which is given the name matrix multiplication, which is a little more complicated
than this.

So, we want to compute the multiplication A × B, where A and B are matrices. So, again, we
have to tell us a rule for computing the ij’th entry of the product. So, the product of A and B
has, again, entries i comma j. So, what is the value that goes into C ij? So, here is the
complicated rule. The complicated rule is that you look at row I. So, you look at row. So, this
is my matrix A and my matrix B. So, if I want to compute the ij’th entry in C, I look at row i
in A and I look at column j in B.

So, I look at the first entry here, what is the first entry here it is i comma 0. And look at the first
entry there. What is the first entry there it is row 0, column j. So, I multiply those 2 numbers.
Now I move to the right, and I look at the next number here. That is 1, i comma 1. And I look
at the next number in that column, that is 1 comma j. And I multiply those 2 numbers. So, I
keep walking down the row and A, A’th row at the same time I walked down the j’th column
and in B, and every point I stop and I multiply the 2 numbers that I get, I take all these numbers
and add them up. It seems really complicated, but this is how matrix multiplication works.
(Refer Slide Time: 8:56)

So, let us do an example. So, supposing I take these 2 matrices, and I want to find out, what is
the final entry going to be for C 1, 3, that is row 1, column 3 of the final matrix, what is it going
to be? Well, I have to look at row 1 in A, and I have to look at row 3 and B. And then I have
to start one by one. So, first, I multiply 4 by 8. So, that is my first component, then I have 0,
which is nice, 0 × 2 is 0, no problem, then I have 9 × 1 is 9. And then I have again, 7 × 0,
another 0, so, I get 32. And I get 9, so, 32+9, so, I get 41.

So, this is how I compute 1 entry. And this is just 1 entry. And I have to do this for every
position. So, for every position in the output, I have to scan a row of the input and a column of
the second matrix or row of the first matrix and add up those that many, I have to multiply each
individual pair and add them all up. So, I have to do a lot of work to compute 1 entry.
(Refer Slide Time: 9:55)

And if you do this, then you get this matrix, let us look at some other entries. And see, how did
we get this 46? How did we get C 1 1? Well, So, C 1, 1 means I have to look at row 1. And I
have to look at column 1. So, I take 4 × 1 is 4, then I get 0 × 0, then I get 9 × 0 is 0. And then
I get 7 × 6 is 42. So, this is how I computed it. And then I finally got 46. So, by any entry, I
can do that, so, I can take this 54 and do the same thing.

(Refer Slide Time: 10:50)

So, I want now, if I want this 1, then I want to go to the, I want to go to 2 comma 3. So, I want
to go to row 2, column 3, and now I have 8 × 6 is 48,+3 × 2 is 6,+0 × 1 is 0,+1 × 0 is 0, so, I
get 54. So, this is how I compute the product.
(Refer Slide Time: 11:05)

So, what we have described is called the matrix product. So, this thing on the right, by
computing for every. So, what we have done in particular is a square matrix. So, r is n, c is n.
So, we say C ij is A i 0, B 0 j +, A i 1 B 1 j up to A i n-1, the last column, and B n-1 j, the last row in
the column J. So, there is a simple way to write this in mathematics notation, which is that you
replace this entire summation by the summation sign. So, we say, let this second position range
from 0 to N-1. K. So, we take the summation over all k ranging from 0 to N-1 of Aik × Bkj.
So, this is just a shortcut for writing that long sum without having to specify every term and
writing dot dot dot.

So, in general, we do not need these matrices to be squared, what we want is? That this thing
is well defined, for this thing to be well defined. When I am walking down row i and A, I must
be able to take the same number of steps in column j and B. So, therefore, if I look at the length
of a row in A, if I process some number of elements in that row, then I must be able to find a
matching element for each of them to multiply and then add it. So, the length of a row in A
must be the same as the length of a column and B.

So, remember that we write these things as r by c. So, if I look at A. So, it has some r1 × c1, and
look at B, it has some r2 × c2. So, what is the length of a row in a matrix? The length of a row
is a number of columns. The number of rows is how many rows there are, but I want to know
how far can I go in a row. So, that is the length of a row in A is c1. What is the height of a
column in B? It is the number of rows, I go from the top row to the bottom row. So, that is r2.
So, what I want is that r2=c1.
So, what we are saying is that if A is m×n, then this n must be the number of rows in B and B
can have any number of columns. So, we can take 2 matrices, which have different shapes,
provided the number of columns in A=the number of rows in B. Only then I can do this the
summation correctly.

(Refer Slide Time: 13:26)

So, let us look at an example. So, what happens here is supposing now I have something which
is 3 × 4, I have 3 rows and 4 columns. And here I have 4 rows and 3 columns. So, now when I
take an entry here, and I multiply it by an entry here, I get 1 entry here. So, I get 6 × 530,+6 ×
848,+3 × 2, 6, which is 84. So, I have an entry in C for every row and column, which is there
as a row of A and a column of B.

So, finally, I end up with a matrix which has as many rows as A and as many columns as B.
So, if I start with m by n, and multiply it by an n by p matrix and up with an m by p matrix, in
our context, remember that we are dealing with adjacency matrix. So, now very limited context
in which we are using it for graphs. This does not matter because we are always going to do
square matrices, but matrix multiplication in general does not require the matrices to be square.
It only requires this correspondence between the columns of A and the rows of B.
(Refer Slide Time: 14:33)

So, now let us get back to the problem which motivated all this, the problem of transitive
closure. So, remember that transitive closure was trying to capture the fact that 2 vertices in a
graph. So, it started off with a relation, we said that we have a relation and we want to find out
whether 2 objects in our set are related by a sequence of pairs in the relation. And then we said
we will model it as a graph. And now every relational pair in our relation modelled as an edge
in the graph.

So, finding a sequence of relational pairs, and our relation is the same as finding a sequence of
edges in our graph. So, it is the same as the reachability problem. So, can we find out which all
pairs are reachable from each other? So, we started with an adjacency matrix. So, this is just
some arbitrary adjacency matrix, we do not really need to worry about what this graph
represents, because once we have the adjacency matrix, we said we can compute directly with
it.

So, the property that we know is that adjacency matrix captures paths of length 1, in other
words, direct relationships, which are already given to us and the R relation we started with it.
So, the edge relation is the relation R that we are trying to visualize. So, we know all things
which are directly related because that is given to us. Now we want to go from that to say things
which are related indirectly in 1 step.

So, we look at what we call A2, which is all pairs, which are connected by a path of length 2,
by length 2, we mean there are 2 edges, I need to traverse 2 relationships to go from i to j. And
we said that if you want to go from i to j in 2 steps, that means there must be an intermediate
vertex, and we call that k. So, there must be some k, such that I go from i to k, and then I go
from k to j. So, that was our A2.

So, now, how do we put this in the framework of matrix multiplication, that is what we are
trying to do. So, we are not doing multiplication addition over our normal numbers. We are
doing multiplication and addition, over the values true and false. So, we are doing something
which is called Boolean algebra. So, let us just write it down and understand what we are doing.
So, remember that these things indicate that there is an edge, I can write this as true or there is
no edge, this is false.

So, this is saying that there is an age from 0 to 4, there is not an edge from 1 to 5. So, the 0 is
interpreted as the nonexistence of an edge. So, if this is the answer to the question is, is there
an edge from 1 to 5? The answer is no, false there is no such edge, therefore, the answer is 0.
And is there an edge from say 3 to 4. So, the entry a 3 4 is 1, and therefore the answer is true.
So, we are working with these 2 values only. So, this is a 0 1 matrix. So, 0 represents false 1
represent true. So, this is our starting point.

And now the operations that we are interested in over Boolean values are AND and OR. So, if
I want to say OR, OR of 2 Boolean values is true provided at least 1 of them is true. So, only
if both of them are false is the answer false. So, if I now represent this OR by +, so, this is
where now this matrix multiplication idea is going to come. So, this is the algebra. So, we are
now representing the values by 0 1. So, first of all, we have removed true and false from our
Boolean algebra, from our Boolean set and we have replaced it by 1 for true and 0 for false.
(Refer Slide Time: 18:08)

Now we are taking operations and replacing them by what we think of as arithmetic operations.
So, we are taking the logical OR which says false or false is false and false or anything any
other pair gives true by saying that, if + is or and 0 is false, then 0+0 is clearly 0. 0+1 is 1 and
1+1, 1+0 is, is 1. This might even follow from the fact that there are numbers, obviously, even
as integers, 1+0 is 1 and 0+1 is 1.

What is perhaps surprising from a numeric point of view is that 1+1 is also 1. So, clearly, we
are thinking of these as integers, 1+1 would be 2, but 2 is not really a value that we are dealing
with, we are dealing with only true and false. So, what this is saying is that true or true,=true.
So, we are taking this and making it+we are taking this and making it 1 taking this and making
it 1 and saying this is 1. So, this is a kind of, strange kind of arithmetic we are doing but it is
justified because the underlying interpretation is in terms of logical values and logical
operations.

The other interesting operation for us is AND. So, AND is kind of symmetric to OR in the
sense that in AND we need both to be true. So, the only interesting case is 1×1. So, 1 × 1 is 1,
that means true and true is true. And if any of them is 0, then the answer is false. So, false in
anything is false. So, both are false, or 1 is false, it does not matter. So, now this is our setup.
So, we are working with the 0 1 matrix. And now we have these operations multiplication
which represents AND, and OR, which is represented by a +.
(Refer Slide Time: 19:34)

So, now let us look at what this says? It says for some k. So, A2 ij is 1. If for some k, I can find
Ak ik =1 and A kj =1. So, for some case, so, what are the possible values of k? k could be 0, k
could be 1. So, k has to be one of the vertices. So, k could be 0, K could be 1, k could be 2,
anything up to n-1.

So, this expression, this sentence here translates to saying either k is 0, in which case I have A1
0 is true and A0 j is true, or I have k =1 in case A i 1 is true and A1 j is true and so, on. Or
finally, the last possibility is that it is A i n-1 and A n-1 j. So, if any of these pairs A I k, k A j is
true simultaneously true, then there is an edge from i to j, If more than 1 is true, it is still true.

If I have multiple ways of going from i to j in 2 steps, it is still okay, I need at least 1. So, this
is just the expression for this left-hand side definition, written out using the logical operations
and an order and interpreting the entries in the original adjacency matrix A as true and false.
But notice that this is the expression if we write it using this algebra here, we replace every
AND by × and we replace every OR by +. So, this is an algebraic way of writing out this ANDs
and ORs.
(Refer Slide Time: 21:04)

So, once we have written this out using ANDs and ORs, it is very clear that we have written
out a matrix multiplication entry in terms of this new interpretation of ANDs and ORs. So, A2
ij is a summation of all k from 0 to n-1, A i k × A k j. This is exactly the equation we wrote for
an arbitrary matrix product, we said C IJ is a summation from K =0 to n-1 of A i k × B k j.

So, that is the reason why A square is just in matrix terms, A × A provided we think of a as
having Boolean values 0 1. And we think of OR as +, and, AND as multiplication. So, using
that interpretation, and those rules for+and ×, we get precisely A × A. Now proceed, next step
in transitive closure, we look at paths of length 3, and we said a path of length 3, for instance,
could break up as a path of length 2, followed by a path of length 1. So, I need to go from I to
k in 2 steps, and then from k to j in 1 step.

So, then, in the same logic, we can compute that this is the all over all k of this. So, it is a
summation over all k of A2 ik × A k j, because A2 captures the length 2 path, A captures the
length 1 paths. Now of course, we could also decompose this length 3 paths as a length 1 path
from I to k and a length 2 parts from k to j is the same thing, I just have to reduce that path of
length 3 to 2 things.

I already know either 2+1, or 1+2. So, it could be A2× A, it could also be A × A2, the
intermediate k could be first finding an entry in A and then finding an entry from k to j in a
square, it does not matter. So, A3 is either A2× A or a 10 A2. Notice that here, when I did A
square, I did not have an option, because both are A. So, it does not, there is only 1 way to
decompose, a 2-length path is 1+1, because that is the only information I have. But for a 3-
length path, I can do 1+2, or I can do 2+1.
Now, in general, we said if I want to, if I already knew paths of length l, then I can extend to
l+1 by again multiplying by A. So, I can take all paths, which are Connect pairs, which are
connected by a path of length l, extended by 1 more thing, and then I get all paths of length
l+1. So, each successive length path can be got by matrix multiplication. Again, we could also
invert it, we can first take 1 step and then take l steps. And of course, you could do other things
also, supposing l is 7, you could first take 3 steps, and then take 4 steps. But it is convenient to
write it in a uniform way. So, that we know how to do it systematically from 1, 2, 3, onwards.

So, this is how we get paths of different lengths by consistently multiplying 1 more time by A.
So, we take the matrix, we already have for L steps, multiply 1 more time by A, So, each time
we multiply by A, we take the matrix we have computed in the previous step and multiply 1
more time base. So, it is A × A × A × A. So, implicitly, this thing contains a lot of A × A × A
in it L ×. So, there are l × that, and then 1 more time.

So, this now gives us paths of specifically of length 1, length 2, and 3 and all that but what we
said is we wanted to know whether there was a path of any length. And then we observed that
of course, because it is a graph. If there is a path at all, there must be a part of at most length
n-1 because if it is longer than that, then there will be a loop which you can remove, because
there are only n vertices in the whole thing.

So, if I take n-1 steps, starting from a vertex, and each time I go to a new vertex, there are only
n-1 new vertices I can go to, if I take 1 more step, then the next vertex must be 1 of the n
vertices I have already seen before, and that would be useless. So, therefore, it is sufficient to
look for this matrix of length, path lengths up to N-1. So, what we want is, is there a path of
length 1 or link 2 or length 3 or this but each of them is represented by a matrix and OR is +.

And so, now we can go back to matrix addition, we want for each ij, we want to check whether
A ij is 1, or A2ij is 1, or A3ijs 1, or A 4 ij is 1 dot dot dot upto A n-1 ij is 1, we want to check if any
of them is 1. And that is the same as doing++++. So, this is why we get the transitive closure,
which we wrote as A to the power +, the transitive closure is just the sum the matrix sum of
these n-1 matrix is the A, A, which is this is really A to the power 1 implicitly.

So, A which is the 1 length path, A2 which is the 2 length path, and so, on. And, that is how we
got from transitive closure to matrix multiplication, by using the entries in A as Boolean values
and interpreting the 2 operations+and × as logical OR, and logical AND.

(Refer Slide Time: 25:49)

So, to summarize, what we have seen is that we can take matrices and operate them as a whole.
And depending on what operation we are doing, the operation is defined differently. So, matrix
addition is very simple. It is just defined element wise. And it tells us that the ij’th entry of the
final sum is the sum of the 2 ij’th entries in the initial matrices. And for this, we need that A
and B are compatible in the sense that they have the same number of rows and the same number
of columns. So, then we get C=A+B.

On the other hand, if we want to do a matrix multiplication, then we have to take for each entry
ij. In the final thing, we have to go through row i and column j of A and B, and then pairwise,
multiply all the terms and then add them up. So, that is the summation. And for this, we need
that the number of rows a number of columns in a must be =the number of rows.

So, if I walk right, some n ×, then I must be able to walk down some n ×. So, n is the number
of columns in A, it must be =the number of rows and B. So, we have this constraint that A and
B must agree on this component, the number of columns in A and the number of rows and B.
And finally, we end up with a matrix, which has as many rows as A and as many columns as
B.

(Refer Slide Time: 27:05)

And then we saw that, if we interpret the adjacency matrix as a Boolean matrix, the values 0
and 1 as being representing true and false, and we think of the operations AND and OR as
being represented by multiplication and addition, then we can think of our transitive closure
computation as initially doing a bunch of matrix products. So, we get A to the power l+1 as Al
× A.

And finally, we add up all the matrices we have computed from A1 the original 1 up to A(n-1) ,
where n is the number of vertices in our graph.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.1
Shortest Paths in Weighted Graphs

(Refer Slide Time: 0:18)

So now, let us look at a new type of graph called a weighted graph. So, remember that in a
graph like this we have seen that a systematic way to explore this graph is breadth first
search and breadth first search explores this graph level by level and therefore because it
does it level by level, it discovers the vertices reachable from the starting vertex at
successively longer distances and therefore, the BFS computes the shortest path in terms
of number of edges to every reachable vertex.

Now in practice we often assign some values to the edges, so if you look at for example, a
road map then you might see some numbers against each section of road representing the
length of the road. Similarly, if you look at say a railway map or a airline map you might
see the time it takes to do a segment of a journey or it could be the distance or it could be
even the cost you know how much does this ticket cost.

So, these numbers indicate some more abstract information about the length of an edge
than just the fact that I take one edge, so it is not enough for me to say that I took two
flights, I need to know how long these flights are. So, if I take a hopping flight in a short
distance say I go from Chennai to Bangalore to Mangalore, right it is not the same as taking
a hopping flight which takes me from Chennai to Delhi to San Francisco. So, the two are
still two length travels but one is enormously longer than the other in time.

So, this is the kind of information that we want, so we want to take this kind of a weighted
graph, so what is a weighted graph formally? A weighted graph is just a graph, so we have
a set of vertices, we have a set of edges but we have weights, so weight formula is a
function, right, a function which takes every edge that is present in the graph and assigns
it some real number.

So, notice that we are not claiming at this moment that the real number is positive or
negative, we will discuss what negative weights will mean but most of the interesting things
that we can think of the weights will be 0 or more. So, we can think of a 0 cost edge
sometimes the cost an edge which is not there has a 0 cost edge or sometimes it may be
there but typically in any reasonable scenario weights are positive but we will see a
situation where weights could actually be negative and make sense.

So, the first thing is we have to, we are going to work with these graphs and we have an
adjacency matrix way of working with a graph in general which records the presence of an
edge. So, how do we record a weighted graph, so what is our representation of weighted
graphs? So, in adjacency matrix what we can do is that whenever we normally put a 1
saying there is an edge, instead of the 1 we can put the weight, so assuming that there are
no 0 weights, then wherever there is a 0 there is no edge, wherever there is a 1 there is a
weight.

So, if you look at this graph here for instance, if I look at the edge for instance 4 to 5, then
I look at the entry 4, 5 and now I have a 50 there rather than just a 1. So, this is a very
simple way to represent weighted graphs just take the adjacency matrix and at each entry i
comma j put the weight of the edge i comma j and if there is no edge or if the weight is 0
put a 0.
(Refer Slide Time: 3:18)

So, our interest is initially to compute shortest paths in such graphs, so we have these
weighted graphs, where we have some edge weights between edges and we want to find
the shortest path, so what is the shortest path now? For us the shortest path will be the sum
of the weight, so for example, if I take this path, then the weight of that path, the length of
that path is 80 plus 70 is 150. If, I take this path for instance, the length of that path is 60.
More interestingly if I take this path the length of that part is 15 whereas the direct path
from here to here is just is 50. So, going from 4 to 5 via 6 is actually shorter than going
from 4 to 5.
(Refer Slide Time: 4:02)

So, in general this is the situation that in a weighted graph, the weighted shortest path does
not need to have the minimum number of edges in the unweighted sense. So, if I look at
the very beginning of this graph from 0 I can go from 0 to 2 in one step but then I pay a
cost of 80, whereas if I go from 0 to 1 and 1 to 2, so I take two steps then actually I get a
cost which is 10 plus 6 which is much less 16. So, the shortest path from 0 to 2 is actually
via 1, an indirect path.

(Refer Slide Time: 4:30)

So now, what are we going to try and solve with these weighted graphs? Well there are
more than one type of shortest path problem, so the first type of shortest path problem is
one where we start at a fixed vertex and we want to find out the distance to every other
vertex, so this is called the single source shortest path problem. So, we would start a fixed
vertex and find out how long it is from this vertex to every other vertex, why is this an
interesting problem?

Well, there are many applications where this is interesting for instance, suppose you are a
manufacturer, so you have a factory and you make things and now you have to take things
from your factory, your finished products and distribute them to the shops where they are
sold, the retail outlets. So, you have a single source your factory and then you have to find
the most efficient way, so the shortest path in terms of whatever you are measuring the cost
of travelling or the time it takes to travel or the distance it takes to travel, whatever is the
cost that you want to count towards the transportation cost, you would like to find the
shortest transportation cost from your factory to every one of your retail outlets. So, this is
a single source shortest path to every other vertex.

Alternatively for instance, you could be a courier company, so what happens in a courier
company is that they have these flights between cities, so all the packages which go to say
Delhi come from different destinations and they land in Delhi and they go to a centralized
clearing facility in Delhi. So, overnight you might have flights coming from Calcutta, from
Bombay, from Bangalore, from Chennai and all that all this information, all these packages
come to Delhi and now they have to deliver them out.

So, the starting point, the distribution center where all these things are initially brought in
from the airport or air cargo wherever they come, that is a single source and now they have
to now find the most efficient way to distribute it to all the destinations where these
packages have to be. So, the single source shortest path probably has a number of
applications and therefore it is an interesting version to solve.

On the other hand, sometimes we want to know something about every pair. Now, of
course, you could take the single source thing and start from every vertex and find every
other vertice distance and then you will get all pairs but generally there may be a better
way to find the distance between all pairs. So, for every i and every j we want to find out
the shortest distance from i to j. So, the single source is fixed and starting point, say fix
vertex 7, and from 7 what is the shortest distance to everything?

Now, this is for every i and j, from not only from 7 to j, I want to know from 9 to j, from
11 to j, from 7 to 11, from 9 to 11 and so on and this is the kind of thing typically that say
if you are managing a booking site and somebody says I want to go from city A to city B,
then you have to be able to provide in terms of the cost or time or some metric the cheapest
way to go from A to B.

So, somebody might say that I want to reach there as fast as possible, so what is the shortest
flight? Some people might say I want the cheapest ticket. So, based on whatever is the cost
that you are associating, then you will have to find and then you need to be able to do this
for any pair because you have customers who can be going from anywhere to anywhere.
So, this is the all pair shortest path problem, so these are the two problems that we will
initially look at in the context of weighted graphs, single source shortest paths and all pair
shortest paths.

(Refer Slide Time: 7:45)

Now at the beginning I alluded to the fact that we have this function, so we said that we
have this function which takes every edge and gives us an arbitrary number and in principle
there is nothing to prevent this number, this weight of an edge from being negative, so what
if I am thinking of this cost, what would be a reasonable scenario where I could have
negative edge weights?

(Refer Slide Time: 8:12)

So now, imagine that you are an Uber driver or an Ola driver or any one of these cab
companies, so you have a certain amount of hours when you drive your cab and at a certain
point you want to start heading home and reach home roughly when you finish your work
rather than eventually traveling across the entire city empty, so this is all, I mean many of
us in days when we have taken Ubers, I found that towards the end of the day for example
it is little difficult sometimes to get long distance things if the driver is not living in that
direction and they would say ‘No, no sorry I am going in the opposite direction because I
am heading home’.

So, here is a taxi driver trying to head home at the end of the day, so maybe he has an hour
of service left and he wants to try and optimize where he lands up at the end of this hour,
so he has a minimum amount of time to travel home. So, now he has to take a choice, so
he has to start looking for customers maybe right, so if he takes a road which has very few
customers then he will be losing money, so there will be a cost that he is paying, so that is
a positive cost.
On the other hand, if you travel on a road with many customers then you are likely to find
somebody who will hail you for a ride, you might get a call so therefore you will earn
money, so you have a negative cost, so where you are not taking customers, you are paying
for driving the car, you are paying of petrol cost and other costs and therefore, you are
losing money so that is a positive cost, where you are gaining money, where you are earning
money it is a negative cost and you want to obviously get more money, so you want to
reduce the cost you want to make the negatives more than the positive. So, you want to
find actually a route towards home which minimizes the cost, so this is one example where
negative edge weights make sense.

Now what happens to our shortest path problem in the presence of negative edge weights?
So, the problem is not so much with negative edge weights but negative cycles. So,
supposing I have somewhere in my graph, a cycle which has something like minus 3, plus
2, minus 1 and plus 2 for example, if I go around the cycle then what do I do? I add up the
weight, so I get minus 3 plus 2 is minus 1, right, minus 1 plus, so maybe I should maybe
make this also plus 1, so minus 3 plus 2 is minus 1, minus 1 minus 1 minus 2, minus 2 plus
1 is minus 1, so the total weight of the cycle is minus 1. In other words if I go around the
cycle once I reduce my cost by 1.

So, if I am supposing I am going from A to B, so there could be some arbitrary weights

here W1 and W2, so I do W1 plus W2 to go from here to there but if I want to reduce it I
go around the cycle uselessly once and I get minus 1, I go around it again and I get minus
2, I go around it again and I get minus 3, so I can go around the cycle as many times as I
want, and keep reducing my cost, I can make it as negative as I want and it does not make
sense because I mean we are actually taking longer and longer paths but because the cycle
has a negative cost we are able to do this, so this is clearly something which is, which
makes the problem undefined.
(Refer Slide Time: 11:10)

So, if we have a negative cycle we can just go round and round that cycle and then there
are no shortest paths anymore on anything which goes past that cycle because every time I
want to reduce the cost just go around cycle once. But if I do not have negative cycles then
it is fine. So, when a graph has negative cycles, shortest paths are not defined but if you
have negative edges you might have some edges which are negative but you do not have
negative cycles, then it is fine. So, if you have negative edges but no negative cycles you
can still do shortest paths but you have to then be careful that your algorithm does not
depend on the edges being positive.
(Refer Slide Time: 11:45)

To, summarize we have looked at what is called a weighted graph, so in a weighted graph
we attach to every edge a cost, a number, which we call the weight and these can be
described for us in our adjacency matrix by entering the cost or the weight of every edge
in place of a 1. So, then once we have these edge weights, then we can measure the length
of a path in terms of the weight. So, not just how many steps I take in terms of edges but
what is the total sum of the weights across these edges.

And so now I get a new notion of shortest path which is probably more natural from the
way we think about graphs representing sort of spatial things at least, so we get the sum of
all the edges that we traverse but the weights of the sums, not just the number of edges and
we saw that this now will give us something which is not necessarily the same as the
shortest path in terms of number of edges. We could have a shorter path which has a longer
higher cost as compared to a longer path.

So, we said that there are two types of shortest path problems with at least two types which
we will find interesting, one is the single source path where we start at a fixed vertex and
we want to find out where we can, how fast we can go to every other vertex, so this is for
example the delivery problem for a courier company or we have the all pairs problem which
is typically the type of problem that you need to solve if you run a travel agency, you need
to be able to tell somebody from anywhere to anywhere what is the best way to go.
And finally, we looked at this peculiar problem of negative edge weight, so we gave a
justification that there can be reasonable situations which are modeled by negative edge
weights and if we still want to be able to compute shortest paths in the presence of negative
edge weights, what we need to ensure is that there are no negative cycles because if we
have negative cycles, then the shortest path is not defined but we do not have negative
cycles even if we have negative edge weights, we can hope to find shortest paths.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.2
Single Source Shortest Paths

So, we are looking at weighted graphs, and we are looking at shortest path problems on
weighted graphs and we said that there are two types of shortest path problem that we will
focus on, the single source shortest path and the all pair shortest path. So, let us look at
how we can compute single source shortest paths in weighted graphs.

(Refer Slide Time: 0:32)

So, remember that a weighted graph is just a graph, which has in addition, this weight
function, this weight function assigns to every edge some real number. So, here on the right
we have a graph where at every edge we have a number written against it, saying, for
example that the weight of the edge from 1 to 4 for instance, is 20.
(Refer Slide Time: 0:53)

So, in the single source shortest path problem, we want to look at these edge weights, so
now we are taking paths as representing the edges underneath as representing the weights
underneath, so if I take a part, for instance, if I take this path, then the total length of this
path the 6 plus 70, it is not two the edges but the weights attached to it that I have to add
up. So, we want to find some source vertex, it could be any vertex, I start from there and I
want to find the shortest path to every other vertex in the graph.

(Refer Slide Time: 1:23)

So, in order to solve this we will first assume that edge weights are all non negative, that
is we could have 0 weights, but we certainly do not have negative weights, otherwise
algorithms that we are going to look at will not work and we will also explain why it will
not work. So, for now we are looking at graphs like the one here, where all the edge weights
are either 0 or positive, not negative.

(Refer Slide Time: 1:48)

So, let us say that our single source in this particular example is 0. So, we want to find the
shortest path from 0 to every other vertex in this graph. So, how would we do this? So, one
way to imagine this or to visualize how the algorithm works, is to imagine that these are
all some kind of pipelines carrying oil and all these nodes are actually oil depots which are
storage tanks full of oil.

So, now imagine what happens if we set fire to this thing, so we set fire here. So, what will
happen is that the fire will start traveling away from 0 along the pipeline because the
pipeline also has oil, so you burn the tank and the pipeline burns. But of course, if you have
ever burned firecrackers you know that it takes time for the fire to go along the thread, the
wick. So, it takes time for the fire to travel, but it travels in all directions at the same speed.
So, the fire, fire is traveling in this direction towards 1, it is also traveling in this direction
towards 2, but the length of the edge determines which one will burn next.
(Refer Slide Time: 2:59)

So now, if the fire is traveling at a uniform speed, then after a certain amount of time, right,
after some amount of time this one will burn. At that point, because this is 10 and that is
80 the fire would have only reached up to about here. About one eighth of the way.

(Refer Slide Time: 3:16)

And now, once the next vertex burns, it will start fires in these two directions, so now I
have a fire running in that direction, this thing is fully burned and so on. So, this is the
intuitive idea that we want to capture. So, let us see how it works.
(Refer Slide Time: 3:34)

So, we start with this thing, saying that initially when time is 0, we burn the zeroth vertex
and now the fire we said starts moving in these two directions. So, if we imagine the fire
burning then after some, let us assume that fire moves at one unit of this, one unit length
per one unit time. So, here it has to move 10 units of length, so in 10 units of time it reaches
vertex 1. Now in 10 units of time as we have indicated here, it has traveled some short
distance along the edge from 0 to 2 as well, but it is not yet anywhere near 2.

Now that this is burned, now something else is going to happen because I am going to end
up having fire going in these two directions. So, after 6 units of time from this, vertex 2 is
going to catch fire. Notice that this fire which started from 1 is still growing but it is
nowhere near 2 and meanwhile, this thing has also started traveling towards 4. So, what is
going to happen next is that this fire is going to reach this at time 10 plus 20, remember it
is assuming the fire is moving at 1 unit distance per unit time.
(Refer Slide Time: 4:48)

So, at t equal to 30 vertex 4 catches fire and meanwhile, this fire which is at 2 has started
moving there and this fire now is kind of reached a little bit less than halfway from 0 to 2
because it has (())(5:02) units so we have already spent 30 times units so far, so suppose to
be about three eighths of the way from 0 to 2.

And now what happens here is that this fire now starts going in these two directions, so
this fire is going here, this fire is going here, these 3 edges have all burned and now two
new fire start, so which one is going to reach the next one? Well, this is only 5 units of
time, so in 5 more units of time the fire from 4 will reach 6.
(Refer Slide Time: 5:30)

So, at t equal to 35 vertex 6 is going to burn. Meanwhile, this thing is started, so in 5 units
of time this fire has reached a little distance, this has made more progress, this has made
more progress and then from here, a separate fire starts going towards vertex 5 in addition
to the old fire which is coming from 4.

(Refer Slide Time: 5:52)

So, now the separate fire is going to reach in 10 units of time so at t equal to 45 our vertex
5 is going to burn at this point, we have gone down 15 units of time have passed since 4
was burned, so this fire has only moved 15 by 50 of the way, about one third of the way
there and meanwhile, these things are still going.

(Refer Slide Time: 6:10)

And then finally, what happens is that at time 86 this fire which started from here reaches
here, at this point everything has burnt, so all our vertices have burned but all the edges
have also burned it so happens, it does not have to happen always, but now we have
discovered the earliest time at which every vertex burns and now we can check that these
earliest times are actually the shortest paths.

So, we have discovered some interesting shortest paths for instance, we have discovered
this shortest path, we have also discovered this shortest path, so the shortest path to 5 is not
0, 1, 4, 5 but 0, 1, 4, 6, 5. So, this is how this algorithm works. So, let us try to understand
how we would actually do this calculation that we said by drawing these burning pipelines,
so how do you keep track of when the pipeline is going to burn next.
(Refer Slide Time: 7:00)

So, what we do is we compute at every given point when each vertex is expected to burn.
Based on the information that we have so far about which pipelines are burning and which
oil depots have burned, we compute the expected time to burn, now among all the vertices,
the one which is expected to burn next is going to burn next. So, at that point we burn it,
so we update the vertex which is burned. And once a vertex burns, it starts. Remember,
new fires along its neighbors towards, so its neighbors.

So, we have to check whether any of its neighbors will burn now faster because of this, so
this is what we saw, so we said that when 0 burned, we started a fire here, so we could
expect that this will burn at time 80. But after time 10 when this burn, a new fire starts so
we said if this burns at 10, then this is actually not going to burn 80, this is going to burn
at 16.

So, each time a vertex burns, we figure out something about the neighbors, this is going to
burn at 30, at least cannot burn beyond 30 because I already know that a fire is started at
time 10 and it is going to reach there in 20 units, similarly a fire has started here at 10 and
it is going to reach here in 6 units. So, 2 cannot burn any later than 16, maybe it will burn
even earlier but not later than 16. So, each time we burn a vertex, we update the burn time.
(Refer Slide Time: 8:19)

So, let us see how this goes. So, initially we know nothing, so initially when we start this
whole process every graph, every vertex is going to burn at some indefinite time in the
future, so as far as we are concerned it is never going to burn, at time infinity. Maybe the
fire is never going to reach there, we have no idea. Now we said, we will start at a single
source, so for the single source we know that the burn time is 0, because that is when we
start ticking, the clock starts ticking at 0 so, at time 0 we burn our source vertex which in
our case is vertex 0.

Now, this is where we start doing this, so we update the burn time of the neighbors. So,
once we have burned 0 at times 0, we have to look at its neighbors namely 2 and 1 and say,
do I know something better, what do I know about 2? Well, as far as I know 2 is never
going to burn but now I know that 2 is going to burn at least by time 80, after time 80 it is
there is no chance that 2 is not burnt.

Similarly, 1 will burn at least by time 10, because 0 is burned and the fire has started
moving in that direction. So, I update these two entries, these two entries can now be
updated with better information I can reduce it to 80 and 10.
(Refer Slide Time: 9:24)

Now, I look at all the vertices that I know about and I see which one is going to burn next.
So, I look for the smallest unburned vertex, I look for one with the smallest time to burn.
So, the smallest time to burn at this point is 10. So, I go to 10 and I say okay you are going
burn next. Now, the same thing happens when I burn 10, it starts two new fires if you
remember, so it will start a fire back of course there is no fire going back because 0 is
already burned but in the directions where there is no, where the vertices are not burnt I get
new information.

I already believe that 2 will burn at time 80 but now I can tell that it cannot stay more than
time 16. And similarly, 4 which was previously not known to burn at any time at all. Now
I know it definitely burned by 30. So, I can update this entry as neighbors of 1, this entry
and this entry to be now 16, and 30.
(Refer Slide Time: 10:19)

Now, again I look among the unburned vertices, so these are my unburned vertices, I look
among my unburnt vertices for the one which has the minimum time to burn. Now this is
the absolute time from time equal to 0. It is not 16 more units of time but at t equal to 16,
which is 6 units of time from now, because right now I am at unit, I am at time 10, 10 is
when this happens. So, in 6 more units something is going to happen and that is going to
be 2.

(Refer Slide Time: 10:48)

So, at times 16 2 burn, when 2 burn, I look at its neighbors, which are not burn, namely 3
and I say 3 which I never knew was going to burn at all, now I know it is going to burn at
86. But I get no new information on this side because 2 is not connected to anybody else,
so I have no further information yet about 5 and 6. As far as I know, 5 and 6 are never
going to burn.

Now, again I look at those vertices which are not burned and I look for the minimum. And
I find that vertex 4 is going to burn at time 30. So, I burn it, so now time is 30. So, now I
have information about 5 and 6 because once I have burned 4 I have started these two new
fires, which are going here. So, this tells me that vertex 6 is going to burn at time 30 plus
5 and vertex 5 is definitely going to burn by 30 plus 50, so I can update those to 80 and 35.
So, this is 30 plus 50 is 80 and this is 30 plus 5 is 35. So, now again I have the same thing
which are the ones who did not burn, 35, 80 and 86. So, which is the next one to burn
clearly vertex 6.

(Refer Slide Time: 11:52)

So, I burned vertex 6, now having burn vertex 6 again, I have, this should been a 6 here, I
do not know how it became 35, so this 6 now passes a new fire here. So, since this was a
35, I know that this is going to burn at 45, my previous information was 80, so I have to
remove the 80 and make it 45, with each time I improve my estimate if I can based on new
information that I get.
(Refer Slide Time: 12:19)

So, now I improve that 80 to 45 and having improved that 80 to 45. Now I burn it because
among the things which are not burned is 60, 86 and 45. So, when I burn 5, I get no new
information about 3 or anybody else, because it is not connected. So, finally, 86, 3 burns,
so this is my algorithm, this algorithm although we have described it picturesquely in terms
of burning vertices and all that, you can keep track of all this information in a matrix and
keep track of what is burned we can keep updating this time to burn and so on by looking
at the edges in my graph and finding out what are the neighbors, what is the current burning
time and what is the expected new burning time and so on.

So, it is a fairly straightforward thing to do if you have the matrix and if you have this
information about these burning times, we are not going to describe this algorithm in
precise detail now, later on we will study it in the course on algorithms. So, this algorithm
is very well known it is due to a very well known computer scientists called Edsger
Dikjstra, so this is called Dikjstra’s algorithm. This is the Dikjstra’s algorithm for the single
source shortest path.
(Refer Slide Time: 13:25)

So, why does this work, I mean why is this kind of update reasonable? So, the idea is that
every time the new shortest path that we find extends an earlier shortest path, right, so if
we look back, we go back to this. So, the shortest path to 86 is an extension of the shortest
path to 2. So, basically the shortest path to 2 is this way. So, I get the shortest pathway
here, I cannot get a shortest path to 86 which goes the other way by a longer path, so every
prefix, every beginning segment of a shortest path is also a shortest path.

It cannot be that I find the shortest path to a vertex and then to go to its neighbor I find the
shortest path with bypasses I mean find the longer path to this vertex and goes. So, every
shortest path, actually extends, an earlier shortest path. So, what we are going to show as a
kind of correctness proof is to assume that at every point when we have burned some
vertices the numbers we have associated, the burning times we have associated are the
shortest time, so by induction this is correct, because the starting point is a source vertex
which we label with 0.

So, when we label the starting point with time 0, obviously that is the shortest time at which
it burns because that is the shortest path from a vertex to itself is length 0. So now, at any
given point we have some vertices which are burned and we have some vertices which are
yet to burn and with each of them we have an estimate. So, we have some vertices in this
case s, x and y, which are burnt and now suppose our next vertex to burn with the shortest
burning time is v and it is connected to x. So, what we are assuming therefore is that the
shortest path to x now extends by x to v to shortest path to v.

So, basically the time to v is going to be the time to x plus the weight of the edge xv, I
mean that is basically what we are saying, there is no faster way to get to v than to first get
to x, which is the time it takes for x to burn plus the cost of going from x to v. So, now the
question is, is this going to be the final value for v, can we discover, because now we are
going to burn it, so this is going to get burned. So, this burn thing is going to include and
by our induction assumption once we burn a vertex, its value is correct.

So, is this value correct, that is what we want to know. So, if we choose to burn it now, are
we being hasty, are we going to discover a better path to V later on, can it happen?

(Refer Slide Time: 16:03)

So, supposing there is a better path later on so where will that better pat come? It will come
from some other vertex which is not yet burnt because among the burnt vertices we already
found the best path. So, at a later stage supposing we find a path from y to v via some w,
we burn a new vertex w and through that new burning vertex we somehow magically
discovered that there is a shorter path to v.

Now, can this happen? So, first of all, why did we burn w after v? We burned w after v
because w has a burning time which is not smaller than v, it could be exactly equal but it
is greater than or equal to it. So, the time at this point is the time is already equal to or
greater than v and then I have to come from here. So, I have to spend some time, because
there is some cost involve with that edge, it could be 0. So, I certainly I could get the same
value, maybe by some magic, the burning time of w is exactly equal to the burning time of
v, it cannot be smaller because otherwise we burned w before v. So, it is at least as big,
maybe it is not bigger but maybe equal to, but even if it is equal to, at best, then I can come
in 0 cost from w to v because edges are not negative.

So, the edge from w to v has a weight which is bigger than equal to 0. So, this is crucial,
so if we did not have this, then the decision to burn v based on its distance from x could
have been premature.

(Refer Slide Time: 17:21)

So, this tells us that we cannot use Dikjstra’s algorithm, if we have negative edge weights
because this could happen, I might discover a long path after I burnt a vertex which
becomes shorter, because there are some negative edge weights which cancel out the initial
thing, so it could be that the, this is, so w is bigger than v, but because this is negative, the
cost of coming back from w to v actually drops the cost of v below the cost that I had got
when I burnt it.
So, for this reason we have to assume, when we run Dikjstra algorithm that the edge
weights are bigger than equal to 0 otherwise this strategy for updating is not going to always
work.

(Refer Slide Time: 18:00)

So, to summarize, we have found an algorithm to compute single source shortest paths
provided the edges have non negative weights. And the way to think about this algorithm
is to use this fire analogy, so we set fire to every, we set fire to the initial vertex, think of it
as an oil tank, and then think of the edges as pipelines. So, this oil, fire that starts at source
vertex spreads at a uniform rate through these pipelines, and then we can calculate through
this pipeline uniform burning the rate at which, the time at which every vertex burns, and
when we can systematically keep track of this, you will get the shortest path to every vertex.
So, this is Dikjstra’s algorithm, which works single source shortest paths for weighted
graphs with non negative edge weights.
 Mathematics for Data Sciences 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.3
Single Source Shortest Paths with Negative weights
So, we are studying weighted graphs and we started looking at the shortest path problem, in
particular, the single source shortest path problem and we saw Dijkstra’s algorithm, and we
said that it will work if we do not have negative weights. So, let us see what happens when we
have negative weights.

(Refer Slide Time: 00:28)

So, first, let us recall Dijkstra’s algorithm and look at it a little more formally than we did last
time. So, remember that we thought of Dijkstra’s as I build them operating in this burning
pipeline story. So, we had these vertices as oil depots, and we had the edges as pipelines. And
if we set fire to a source vertex, then the fire spreads at a uniform pace along all the pipelines,
and then we try to calculate the order in which each of the vertices will catch fire and propagate
the fire further along new edges.

So, in the process of doing this as an algorithm, what we do is we keep track of which vertices
have already burned, that is the vertices for which we have already computed the shortest
distance. And we have an estimate about how long it is going to take to reach the others so that
we have an, what we call the expected burn time, which we keep updating as we go along.

So formally, we keep track of these two things, the burn status and the expected burn time. So
initially, we set the burn status to be false. So, let us call it B of j. So, for every B of i for every
vertex i, B of i is initially false. And just to keep track of an auxiliary quantity, we will let UB.
So, UB is just a set of unburned vertices here. So, UB is just stands for unburned. So UB is
those k for which Bk is false.

So initially, UB is a set of all vertices. And the way we start this algorithm is we set the burning
time of the source vertex which we are assuming to be 0 in this case. We assume that the
burning time with the source vertex is 0 and everything else has vertex burning time infinity
because we have no information. And then what we do is as long as there are unburned vertices,
in this case, everything is unburned. So of course, there are unburned vertices, you pick up one
of the unburned vertices, which has a minimum expected burn time.

So initially, there is only one that is the source. So, you pick up the source vertex, and you
update its status, you update the status of the vertex, you pick up saying it is now burned. So,
therefore, in some sense, it is expected burn time is frozen as whatever it was now, and
correspondingly, the unburned vertices will reduce because this particular vertex that you have
chosen to burn now is no longer unburned.

Now, more importantly, what you do is you look at every outgoing edge from this new vertex,
so you just burn j. So, you look at every j k edge and check if k is unburned, that is, we still
have not fixed its distance, then you update it is distance to be the minimum the distance you
already know. So, this is what you already know about k plus the new information that you get,
if you process this edge j k, which may or may not have been taken into account before
obviously, because you have not reached this day before this. So, you look at what is the time
that you burnt j and how much time it will take from j to k and this might well turn out to be
smaller, so this is Dijkstra’s algorithm.
(Refer Slide Time: 03:14)

So now, we look at this correctness proof and we argued why we need this edge weights to be
non-negative. So, we said that we are incrementally discovering shortest paths. So, every new
shortest path extends an earlier shortest path. And inductively, the burnt vertices are those for
which that distance has been computed and frozen, we are never going to update those things
again. So, at last, at some point in the algorithm, we have a big set like this, so we have this big
set of burnt vertices.

So, all the vertices in this set, so we have the starting vertex, which now I have called s just to
denote it and then we have various vertices x, y, and all that which we have learned so far. And
what Dijkstra’s algorithm now says is among the remaining, look for the one with the smallest
expected burn time, and it will turn out that that will be connected to some vertex in the set, so
it will be connected by an edge.

So, it will be the minimum of something plus the edge from x to v. And the argument was that
if we now choose to burn v and add it to the set and freeze it is time at the current time, we will
not be making a mistake and that is because if we find a new path to be later, it will come
through another w, but that w must also start from inside the burnt vertices. So, when I burnt
v, w had a higher expected burn time. So, otherwise, I would have been w.

So, when I get here, if I look at the cost of going to w plus the cost of going from w to v it
cannot be less than what I already have for v and this crucially depends on the fact that this
edge from w to v is not negative, because if it was a large negative edge, then going by a w, I
could suddenly save a lot of cost when coming to v, so that is the crucial thing. So, the argument
does not work, so we cannot freeze the cost of vertex the first time we burn it, if we are allowed
to revisit it by a negative edge later on. So, that is why Dijkstra’s algorithm requires non-
negative edge weights.

(Refer Slide Time: 05:11)

So, the difficulty is precisely this, that we stop considering updates once we burn a vertex. So,
what happens if we start allowing updates even after we have burnt a vertex? So, then the
notion of burning does not really make sense. So, this analogy that we have does not really
make sense, but it is a plausible strategy and why is it a plausible strategy? This plausible
strategy because though we are allowing negative edge weights, we do not have any negative
cycles.

So, this means that if I am going from say, the starting vertex to some vertex x, it cannot help
me to go through a loop because every loop is guaranteed to have a non-negative weight.
Because there are no negative cycles. So, if I want the shortest path from any starting vertex,
from the starting vertex to any other vertex x, I may as well assume that that path is really a
path in the sense of, we have defined that is it has no loops, it has no repeated vertices.
(Refer Slide Time: 06:15)

So, if it has no repeated vertices, then what we can consider is in terms of the length of the
path, not in terms of the weights of the path alone, but we have the shortest path, which takes
us from say 0 (())(0:27) source vertex to a vertex k. So, not only is the weight minimum, but
the reason that this is the path I chose is because there are no shorter paths which have the same
weight or less, so it need not be the shortest path overall, we saw examples where you could
have one edge, which takes me with 80. But then if I take two edges, I might go with 10 plus
6 16.

So, the shortest path might well be a roundabout path. But what we are saying is that if there
are, there is a path of length 2 which is shortest there is no path of length 1, that is what this
means. So, this is the, so l, if I have to do this, I have to take these l steps in order to get to k,
and there is no better way of getting to k to achieve this cost. So, this is not the minimum
number of edges going from 0 to k without considering weights, but it is the minimum number
of edges and the shortest weight if you consider weights.
(Refer Slide Time: 07:14)

Now here, we again go back to our old argument, which says that, okay if this is my path, then
what happens when I come to j1, could I have come to j1 any better than using the weight w1?
Well, if I could have come by a different route and come less than w1, then I could come to j1
and then continue with the same path to k, so if I could come to j1 earlier than I have now, then
I could use that path plus the path from j1 to k and get a shorter path to k.

So, if this is the shortest path to k, then this must also be the shortest path to j1, the shortest
path to j2, the shortest path to j3, and all that. And so, every prefix of this path has to itself be
a shortest path. So, this gives us a starting point to think of an algorithm to deal with negative
edge weights. So, in some sense, once we have updated this one, once we have found the
shortest path to jl minus 1 by some algorithm, then we know that the update that we get for k
is going to be a final update, there is not going to be a better one than that, because, if I keep
decreasing, I can only decrease up to the shortest path, I cannot go below the shortest path.

And when will I hit the shortest path when my nearest neighbour which feeds that shortest path
is also frozen. So, in some sense, if my neighbour’s shortest path is known, then my shortest
path will be known in the next step. So, therefore, for that neighbour, that neighbours,
neighbour should have been known in the previous step. So, we can try to see if we can fix
these shortest paths one at a time, if I can fix j1 then I can fix j2, we can fix j2, then I can fix
j3, and so on.

And once I fix jl minus 1, then I can fix k. So, what we want really is an algorithm, which tells
us that after we have done l updates, so an update for us is what we did when we burnt a vertex,
when we burnt a vertex we reset the burning time to be the minimum of what we already had
plus the new burning time we discovered through the recently discovered vertex. So, if we can
guarantee that after we have done this l times, we have made sure that there are all paths of
length l are at a minimum, there are no shorter weight paths of length l, then since there are at
most n minus 1 edges, then if we do this n minus 1 times there are no paths without repeating
vertices and no strong notion of a path there are no paths which have more than n minus 1
edges in them because once you take the nth step, you have to repeat a vertex.

So, if we have this property that we can update and make sure that after l updates, all the paths
of length l in terms of number of edges have achieved their minimum that is there are no shorter
ways to go l edges or less, you have to take maybe one more edge and take a negative edge that
is different, but you cannot get that l edges. This is then we can guarantee that we can see sort
of first find all shortest paths.

So, this is like a combination of breadth-first search which finds shortest paths by length of
path, and our weighted thing which finds it by length of weight. So, what we are saying is that
up to this path length in terms of edges, there are no shorter paths in terms of weight. And this
is the kind of property that we want.

(Refer Slide Time: 10:14)

So, this algorithm is called the Bellman Ford Algorithm. So, it is a much simpler algorithm, in
some sense, than Dijkstra’s to think of, although it requires a little bit of understanding like we
just did to see why it works. So, all we do is we like Dijkstra, this expected burn time, so we
keep track of the distance to every vertex as far as we know so far. So initially, the distance to
the source vertex, which again, we assume is 0, the source vertex is 0, and its distance is 0, and
everybody else distance is assumed to be infinity.
And now comes the update. What Bellman Ford does is just n minus 1 times it just blindly
updates everything. So, it takes every edge and it looks at the starting point at the edge and the
ending point of the edge. So, I have an edge, which goes from say, some j to some k. So, there
is currently at this iteration, there is some distance that I have associated with j and then I have
associated some distance with k.

And I have what I would get if I take this edge and append it to j. So, this is a candidate to
replace D of k. So that is what we do, we just check for every k, whether the weight that the
distance that we are currently assumed for k, is it smaller than the distance that I would get if I
take one of my neighbours and add that edge weight from that neighbour to that distance. So,
I just blindly do this n minus 1 times and the claim is that this will give you the shortest path.

(Refer Slide Time: 11:48)

So, this works for both directed and undirected graphs. So, the example we did for Dijkstra’s
algorithm was for undirected graphs, but you could as well do it for directed graphs. Because
anyway, we are following edges in one direction only when we compute the shortest thing. So,
let us look at this example, so this is a directed graph, it has some negative edge weights, like
minus 4, minus 1, and there are arrows in this, so there are some so you can go for example,
from 0 to 1, but you cannot come back from 1 to 0, and so on. So, let us see how this Bellman
Ford algorithm would work on this.

(Refer Slide Time: 12:18)

So, what we do is we keep recomputing this distance D of v. So, remember that D of v is the
best distance I know of right now for vertex v. So, in this case, there are 8 vertices from 0 to 7.
And we are going to iterate this thing after initialization, n minus 1 times. So, there are 8
vertices, so n is 8. So, we are going to run this algorithm, this iteration 7 times, and then 7
times, it should hopefully give us the shortest path from 0 to every other vertex.

So, we initialize it by setting the distance of 0 to the vertex 0 to 0, and everything else to infinity
and this is our initialization for Bellman Ford. So initially, we know nothing about how to get
to any other vertex. And now we do this update. So, now we look at every vertex or we look at
every edge is how the Bellman Ford algorithm says, we look at this edge and we say, what do
I know about the starting point plus this weight versus the ending point?

So, the update is compare D of 0 to D of 0, D of 1 to D of 0 plus the weight of 0,1. So, what
should I put here, is a question. So, should I leave it as what it is, or should I update it by some
new information that I have got about the edge coming into it. Of course, I could do it for D of
2 also, but then I know that for D of 2 and D of 6, nothing has happened because everything is
infinity, so is really D of 0, which carries some importance at this stage.
(Refer Slide Time: 13:42)

So, if we do this, we find that from 0 I can get to 1, and from 0 I can get to 7. And therefore,
the entries for 1 and 7, get updated from infinity, which is what I knew before to 0 plus 10, in
the case of 1, and 0 plus 8 in the case of 7. So, I updated to the time to reach 0 plus the weight
of the edge from 0 to that vertex and everything else is infinity because I cannot reach it from
0 at this point. But now, I have some information about 0, 1, and 7 so, in the next step, I can
look for any vertex, which is either connected to 7 or connected to 1. So, not that one, but say
this one. So, 1 is connected to 5.

(Refer Slide Time: 14:25)

So, in the next step, what happens is that because 1 is connected to 5 and I know that I can
reach 1 at time 10, then in 10 plus 2 I can reach 5 in time 12 earlier I believe it was infinity.
So, I can replace it by 12. In the same way, if I look at 6 for instance, earlier I thought it was
infinity but now I know that I can reach 7 and time 8, and 7 plus 1 is 9, 8 plus 1 is 9. So,
therefore, I can reach vertex 6 in time 9. So, these two things which are connected to the vertices
are recently burned, get updated. So, we keep doing this. So, now we have burned, 6, and 5,
not burned but we have updated 6 and 5 in addition to 0, 1, and 7. So now, we will find new
paths because 6 also has outgoing edges.

(Refer Slide Time: 15:13)

So, in particular now, what we will find is that there is a strange phenomenon that we have
discovered, which is that if I come from 6, so remember that 6 was, had been assigned 9 before,
so, right now these are the numbers that we have everything else is infinity. So, now, if I am at
9, and if I take this negative edge, then I can come from 0 to 1 at cost 5. So, instead of going
directly in cost 10, which is what I had earlier assumed, I could take this roundabout route, and
I could do 8 plus 1 minus 4, and come there and 5.

So, this is the kind of update that Dijkstra’s algorithm would not have discovered it because it
would have really frozen that thing at some point, saying that it is already given, and therefore,
I will not update. So, this 10 becomes 5, what about 2? Well, now 2, I can reach from 5. So, it
is 12 minus 2, so this becomes 10. And 5 itself, now is interesting, because earlier, I had to
come this way. And that was costing me 12. But now because I can come from 6 directly, I can
come to 6 and 9, and then 6 to 5 will give me so this has become 8.

So, in this way, we keep updating, so now after this, now that I know 2, I can even update 3
because 3 is reachable from 2. So, if I can reach 2 in time 10, I can reach 3 in time 11. But of
course, 2 itself has now got a better route. Because having come to 5 in time, 8 now.
Remember, it was 12 and then it became 8. Now I can go from 5 to 2 in times 6. So, each time
I am looking at the previous row, so the fact that 2 gets updated from 10 to 6 now does not yet
reflect in the fact that 3 should be updated, so 3 is updated from infinity to 11 because I knew
that 2 was 10 before I did.

So, all these updates are happening at this time based on the previous times information. So, I
am not doing it in sequence in that sense, so though I calculate that the distance to 2 is 6, I do
not use it to calculate the distance to 3 is 7 yet, I will do it in the next round.

(Refer Slide Time: 17:28)

So, in the next round, I discovered that 2 was frozen at 6, and therefore now D has become 7.
And finally, I have also found something that reaches vertex 4 because now I have got this path
which goes this way or there are other paths also. So, we have this path also, which goes this
way, and so on, so we have many paths which come to 4, but I only now reach that and I
calculate 14. So, I keep iterating this, and I get slightly better paths everywhere.

And finally, after I have done this 7 times, I have discovered a stable thing and you can calculate
that if you do this one more time you should get no updates. There should be no shorter paths
if there are no negative cycles. So, this is how the Bellman Ford algorithm works. It just keeps
updating every vertex every time and you do it a fixed number of times, which is the number
of vertices in your graph, minus 1. And once you have done that, you are guaranteed that all
the paths have stabilized.

(Refer Slide Time: 18:22)

So, what would happen if there was a negative cycle? Well, the path would not stabilize, there
would be a way to take a path longer than n, n minus 1 edges go beyond that and go around the
cycle and get still shorter. So, if I iterate this one more time, and the distances decrease, then I
know that there is something wrong. So, this is one way. So, either you can assume there are
negative cycles and keep running it or you can run it.

And then when you come to nth iteration, which you normally should not need, you need to
stop with n minus 1, but you can run it one extra iteration. And see if you get a decrease in the
distance. And if you get a decrease in the distance, that means there was a negative cycle. So,
check that the nth update should not reduce any D of v. If it does not reduce once, then after
that, once the D of v is stabilized, there is going to be no update because the previous update is
just going to propagate. So, once I get the column repeating, there is going to be no change
further on. So, if the column changes at the nth step, then you know that you had a negative
cycle.
(Refer Slide Time: 19:19)

So, we saw Dijkstra’s algorithm and Dijkstra’s algorithm assumed non-negative weights and
the reason that we needed that property was because of this strategy we used to freeze the
distance to burn vertices once they were burned. So, we never looked for updates to that. So,
we should not have found any new updates through negative edges. Otherwise, that strategy is
not correct.

But assuming that we do not have negative cycles, remember if we have negative cycles, the
notion of a shortest path is not defined, because you can go round and round the cycle and you
can make a shortest path as short as you want. So, the shortest path is defined only when there
are no negative cycles, even if there are negative edge weights. And in such a case, what we
just said is that the shortest path is a path, it cannot involve a loop and every prefix of that
shortest path is also a shortest path.

And you can use this in this Bellman Ford algorithm to iteratively find the longest, the shortest
paths of length 1, of length 2, length 3, and so on. And after n minus 1 iteration, you have
automatically found shortest paths to everything. So, in a way, Bellman Ford is a much simpler
algorithm to think of, it is just a blind iteration, which is n minus 1 times you keep updating it,
you do not have to keep track of anything that was burned and you do not have to keep track
of expected burn time separately and all these things. You just keep updating every time you
make an update, you look at all the neighbours and update them again.

And the property of this, this fact that the shortest paths are monotonic in the sense that every
path is an extension of a shortest path guarantees that after n minus 1 steps all these updates
will converge unless you have a negative cycle. So, you can also use this algorithm to find
negative cycles in that sense, if you go through this whole process and you do it one more time
and you find a decrease, then you have a negative cycle.
 Mathematics for Data Sciences 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.4
All-Pairs Shortest Paths
So, we are looking at shortest paths and weighted graphs.

(Refer Slide Time: 0:17)

And we said that there are two types of problems that we might look at in terms of shortest
paths, the one that we have looked at already is the one which is called the single-source
shortest path problem, which asks you to find the shortest path from this some designated
starting vertex, source vertex to every other vertex and we imagined two scenarios where this
might be useful.

So, if you have a kind of factory, which is shipping out finished products to retail outlets, or if
you have a courier company, which has got a centralized delivery facility from where they have
to ship out their things for delivery, both of these would l[i,k]e to know the shortest route from
wherever they are shipping out things, either the finished product or the courier deliveries to
all the other locations in their transportation graph.

On the other hand, another natural problem is to find the shortest distance between every pair
of vertices, and this we said would be very natural if you are running something l[i,k]e a travel
site, somebody wants to book a ticket from somewhere to somewhere else, you should have
the information about the shortest route from this starting point to the ending point, regardless
of which two points they are.
So, you cannot say I am only booking tickets from this city and not from another city. So, that
does not make any sense. So, if you have a travel booking site, it should be able to give you
optimum routes in terms of cost, or time, or distance, or whatever, from anywhere to anywhere.
So, we have seen two algorithms with a single source shortest path problem. Dijkstra’s
algorithm which works if there are no negative weights, and the Bellman-Ford algorithm,
which works even if there are negative weights. But of course, remember that negative cycles
are always forbidden because with negative cycles, the whole thing does not make sense at all,
the idea of a shortest path is meaningless.

So, how do we find all pair shortest paths? So, we can find the shortest path from a single
vertex to every other vertex. So, we can just iterate this by starting it from every vertex. So, I
started from 0, I get all paths from 0, I go to 1, and I started from 1 and I get all paths from 1.
If I do this n - 1 times, I will have the shortest path from every starting vertex to every ending
vertex. So, this is one way we could do this is just to repeatedly run Dijkstra’s or Bellman-
Ford, across all starting points. But what we are trying to see is whether there is another way
to do this, which does not involve this kind of going back and restarting the calculation. So,
that is what we will look at in this lecture.

(Refer Slide Time: 02:28)

So, we will begin with something that we saw earlier, namely, the transitive closure. So,
remember that the transitive closure, we discussed in the context of unweighted graphs and we
said the transitive closure, is the reachability calculation. So, we wanted to find out just l[i,k]e
all-pairs shortest paths, we wanted to find out all pairs reachability given i and j, can I get from
i to j in this graph? Is there a path from i to j. So, again, there, we said that we could have run
BFS or DFS from every starting vertex and got it. But instead, we ran this other algorithm,
which uses this matrix approach.

So, we started with this adjacency matrix and we said that an adjacency matrix represents
implicitly edges, which are paths of length 1. And then we wanted to use this matrix and the
operation of matrix multiplication to bootstrap this from length of 1, paths of length 1 to paths
of length 2, and so on. So, we said a path of length 2 decomposes are two paths of length 1. So,
if I want to find out the path of length 2 from i to j, I just have to check for some intermediate
k, with that I can go from i to k in one step, and then k to j and another step.

And we can generalize this and we discussed why this is matrix multiplication. We can
generalize this to l steps. If I know how to get an l steps from i to j, then I know how to get an
l + 1 steps from i to j, because I go to an intermediate k and l steps, and then from k to j in one
step. So, by decomposing l + 1 step as l steps followed by one step, I can again do a matrix
multiplication and do Al times A is equal to Al + 1.

And finally, in the transitive closure, we want to look for a path of any length. But paths
remember, are always bounded by length n - 1 because there is no need to go through a loop to
read something. So, it is sufficient to calculate path length 1 to n - 1, and then take the sum, the
sum, if you remember, was the OR, we said that this is the OR of so if there is a 1, if A[i,j] is
1 in any of these matrices, then there is a path of length 1,2,3, or up to n - 1 and therefore A
+
[i,j] the transitive closure reports one.

So, we are going to slightly reformulate this calculation of transitive closure, so we are going
to do so. This is a kind of inductive definition, so, we first find transitive closure by first finding
paths of length 1, paths of length 2, and so on and then we take the cumulative effect of all this
by doing the + at the end. Now, we are going to do another inductive thing but we are going to
use a slightly different way of calculating simpler paths and complicated paths not in terms of
path links, but which vertices we will go through.

So, we will just to distinguish it from the earlier one, let us call it B. So, earlier we had A sub
A to the K. If I said A k[i,j]. So, this was referring to paths of length k, it says that there is a
path of length k from i to j. So, here I have something else, B, k[i,j], does not refer to the length,
it refers to which vertices I am allowed to visit on the way from i to j, and not obliged to visit
all of them. But I cannot visit anything outside the set. So, if I say k as my superscript, it says,
I am only allowed to go through vertices, 0 to k - 1, I am not allowed to visit any other vertex
going from i to j.
So, if I have this constraint, can I reach i to j. So, it could be a direct thing in which I do not go
to anything, it could be 3 of these vertices, it could be 2 of these vertices, it could be any
number, but it cannot be outside the set. Now, remember, it is going to be a path. So, I never
want to visit a vertex in the set more than once. But that does not matter so much, instead of
telling me that I can get from i to j without going outside the set 0 to k - 1.

And this constraint, of course, applies only to the vertices a visit in between, so i and j need not
be between 0 and k - 1, i could be anything j could be anything. So, I am just saying that when
I leave i and before I reach j in between I should not see anything, which is outside the set 0 to
k - 1. So, in this setting, what would B of 0 mean? So, B of 0 by this definition, would be some
0 up to - 1, because k is 0. So, this is saying that I am not allowed to visit any vertex in between
because the set of vertices I visit in between can have at most number - 1, but I know that my
vertices start with 0, so there is no vertex in the set, which starts between 0 and ends before -
1.

So, therefore, B 0[i,j] says that I cannot have any intermediate vertex because no intermediate
vertex can satisfy this constraint, that its index is - 1 or smaller. So, therefore, B 0 of[i,j]
precisely captures the fact that I can go from i to j without visiting an intermediate vertex means
there is an edge from i to j. So, B 0 is just my adjacency matrix A. So, just l[i,k]e in the earlier
case, we started off with paths of length 1, and we said paths of length 1 are precisely our
adjacency matrix A. Here, we are saying paths which do not pass through any intermediate
vertices are exactly captured by an adjacency matrix A. So, the starting point of this induction
is A in both cases.

So, how would we proceed? Well, I want l[i,k]e before, I want to calculate what would be B k
+ 1. So, B k + 1 says that I am allowed to use 0 to k. So, if I am allowed to use 0 to k from i to
j, then there are two possibilities. One is that I do not need this k at all, I could already do it
without 0 to k. So, that means that Bk[i,j] is already 1, it means that without going to k, just
staying within 0 to k - 1, I have a possibility of going from i to j or I need to use k.

But if I need to use k, then remember now we use this property that we implicit, that we are
going to visit k only once there is no point in visiting k multiple times. So, if I need to use k, I
need to get from i to k. And I need to get from k to j. But if I am using up k and going from i
to k, then I cannot be using k in between, I am using k only once overall. So, going from i to
k, I do not see k, going from k to j again, I do not see k. So, I can find out whether I can go
from i to k using only the vertices 0 to k - 1.
And similarly, with that, I can go from k to j using only the vertices 0 to k - 1. And then I can
pick these two paths. So, I have a path from i to k and I have another path from k to j. And now
together, this gives me a new path from i to j, which visits k, I am forced to go to k. But in
between, I do not do anything outside 0 to k - 1 except when I hit K. So, this is my condition
now. So, I say that B k + 1[i,j] is 1 either if it is already 1 in Bk.

So, BK[i,j] is 1 meaning I do not need K, or I go to k explicitly, in which case I go from i to k

in Bk - in k - 1, using k - 1 vertices, vertices up to k - 1. And then I go from k to j using vertices
up to k - 1. So, this gives me the inductive calculation and this is a different way of calculating
the transitive closure.

(Refer Slide Time: 09:23)

So, this particular algorithm is the more standard algorithm that you see in books and this is
called Warshall’s algorithm. So, this is Warshall’s algorithm. So, we use the other algorithm,
because it is nicer in terms of describing why it is matrix multiplication. So, this requires a little
more work to put it in that framework of matrix multiplication. But essentially, it is capturing
this inductive property of finding more and more complicated paths.

So, here the complication is a number of which are the different vertices you have to see in
between rather than the total number. The total number is what the other one is talking about
the length of the path. Here, is restricting you to some subset of vertices and that subset keeps
growing. Eventually, you can use all the vertices. So, this is the thing around the formula. Now,
so B 0 of[i,j] is just A[i,j]. And B k + 1[i,j] is 1 if either Bk[i,j] is 1, so we already can get from i
to j without using the K’th vertex, or Bk [i,k] is 1 and Bk [k,j] is 1.
So, this should look very familiar because it looks very similar to the other one. This is the
interpretation of this superscript attached to the matrix is different. So, in this now, when do
we stop? Well, when the superscript becomes n, it says I am allowed to use any vertex from 0
to n - 1, it should be I am allowed to use any vertex from 0 to n - 1 that means I am allowed to
use any vertex at all. That means there is no constraint so, I have some path of the other.

So, if I do this up to Bn, so in the earlier case also added it up to A1 to A n-1. So, here if I do it
up to Bn, starting from B 0, then I am done. So, Bn is the same as A +. And now what we are
going to do is we are going to see that this shortest path algorithm that we have not the shortest
path this transitive closure algorithm that we have due to Warshall’s is very easy to extend to
the shortest path problem for all pairs.

So, remember, the transitive closure is the equivalent of all pairs reachability. Dijkstra’s
algorithm is the equivalent of BFS, except that we have weights. So, weighted reachability is
Dijkstra’s. Now, weighted all pairs reachability is all pair shortest path. So, we are going from
transitive closure, which is unweighted all pair reachability to weighted all pairs reachability.
So, with weights, we are asking what are the shortest paths.

(Refer Slide Time: 11:39)

So, we will use a similar notation. So, let us call SP for shortest path. So, shortest path with
superscript k between i and j is the length of the shortest path provided I stay within vertices 0
to k - 1 when going from i to j. So, that is what this means. So, earlier, it was saying that there
is a path. Now, I am saying among the paths that are there, what is the shortest length, I am
keeping track of the shortest length, not just the fact that there is a path but there is a path of
minimum length and keeping track of that minimum length.
So, what is the base case? Well, if there is no possibility of going through an intermediate
vertex, then we saw that B0[i,j] was just A[i,j]. In this case, I do not want the edge or not edge,
I want to know the weight of the edge. So, SP0 [i,j] is W[i,j]. Remember that W is our weight
matrix. So, W tells us for each edge, what is the value? So, W[i,j] is the weight function. So,
W[i,j] tells us how much cost is there with this edge.

Now, of course, we have to take care of the fact that there is no cost and no edge carefully. So,
what we will do is we will assume that when we have no edge the weight is somehow assigned
to a very large number in particular, we can treat it mathematically we can treat it to be infinity.
So, if there is no edge, SP 0[i,j] will report infinity, if there is an edge, it will report the weight
of the edge.

(Refer Slide Time: 13:06)

Now, the update rule is slightly different from the transitive closure rule. So, this is the slide
that you need. So, either I can go from i to j without using k. So, SP k + 1[i,j] is either SP k[i,j],
or I go using k, and then I have to check the cost. Now earlier, I just want to check whether this
was there, or that was there, I just needed to do the logical or either I can go without k or I can
go with k and I take the OR.

Here, I have to say how much does it cost me to go without k? And how much can I gain by
going with k? So, I either take the cost of going with k which is the shortest path to k from i
followed by the shortest path from k to j, or I take the shortest path without going through
vertex k at all and I will take the minimum of these two. So, this is my update rule for the
shortest path matrix.
And as usual, so this again, this same typo, but the shortest path matrix, if I now calculate up
to the nth step, then it will tell me the shortest path among all paths, the length of the shortest
path among all paths that go through any intermediate vertex from 0 to n - 1 and that covers all
possible intermediate vertices. So, this will be the overall shortest path from i to j.

(Refer Slide Time: 14:21)

So, this particular algorithm is called the Floyd-Warshall algorithm. So, the original algorithm
for transitive closure is due to Warshall and Floyd is the person who adopted this algorithm for
shortest paths, all pair shortest paths, so jointly it is called the Floyd-Warshall algorithm. So,
let us go back to this graph, which we have seen before. So, this is a graph with directions and
with negative vertices. So, I have these negative edges. So, I have some negative edges here
and there.
So, now, SP 0 is the adjacency matrix of this graph, is A[i,j] where I replace every entry the
weight. So, if I look at 0 to 10, 0 to 1, it has weighed 10. For instance, if I look at 4 to 5 for
instance, it has weight - 1, and so on. So, this is just the adjacency matrix for this graph, where
every entry[i,j] represents the weight and if there is no edge, then it is infinity. So, we have a
lot of infinity entries, we have very few entries, which are not infinity.

(Refer Slide Time: 15:20)

So, now what is SP 1, so SP 1 represents all things that you can reach going through the set 0.
But look at 0, see 0 has no incoming edges, I cannot go to 0 and then go from 0 anywhere.
Because 0 is such that you can only go out from 0, so I cannot go from i to 0 and then 0 to
anywhere. So, intermediate, 0 as an intermediate vertex is not useful in this graph, there is
nothing I can go to via 0. So, therefore, SP 1 is the same as SP 0, that is if I am allowed to go
through 0, again, nothing. If I am allowed to go through 0, again, nothing.

So, this is a certain difference between this and the transitive closure thing. So, there are paths
of length 1, and then there are paths of length 2, but the paths of length 2 cannot go through 0,
because if I go through paths of length 2 through 0, I have to enter 0, and I cannot enter 0,
because the directions are all pointing out of 0. So, in this case, the first iteration produces no
change.

(Refer Slide Time: 16:19)

Now, this was my first iteration. So, this is SP 1, which I just calculated is the same as SP 0.
Now I want to calculate SP2, SP2tells me I can go through 0 and 1. So, now I can go through
1. So, there are edges which come into 1, so where can I go by going into 1 and then out of 1
because I cannot go back to 0 anyway. So, I can go, for instance, from 6 to 5 via 1, or I can go
from say 2 to 5 via 1. So, these are two of the things that I should get new.

(Refer Slide Time: 16:51)

And indeed, when you see that, you will see that I get from, so the 3 incoming edges are 0, 6,
and 2, these are the three incoming edges to 1. So, these give me new possibilities of going
from 0 to somewhere from 2 somewhere and 6 to somewhere, but the only thing I can reach
from 1 is 5. So, the all the updates happen in column 5 because that is the target. So, from 0 to
5, I now find that there is a shortest path, which is 10 + 2, so I go 10 and then 2. Similarly, from
2 to 5, I find the shortest path, which is 1 + 2 is 3. And similarly, from 6 to 5, I find the shortest
path, which is - 4 + 2 is - 2. So, this is my first update.

(Refer Slide Time: 17:37)

Now, let us do one more update. So, I have now shortest path. So, these were the updates,
which I got when I did the first update, and now I can now I have discovered that I can reach
2, I can reach 6, and I can reach 5, no, the new thing I have discovered is I can reach 5. So,
now I obviously want all I can reach from 5. So, from 5, for example, I can go back to 2, but
from 2, I can go to 3 and so on.
(Refer Slide Time: 18:04)

So, in the next round, I get a lot of updates, so I get updates from 5 to 1 because now I can
come to 5. And then I can go to 1 using - 1. And so why is that the case because I can go around
this way. So, I can go up and go this way. And this gives me - two + 1, which is not the one I
want, I think, 5 to 1, - 2 + 1 is - 1, that is the 1.

So, in this way, I can now update everything one more time. And I can keep doing this. And if
I do this, now there are 8 vertices in this. So, if I do this up to B to the power 8, then I will have
all paths which go to 0 to 7. And you can work it out, I am not going to work it out, you will
get a matrix which gives you all the shortest paths from everywhere to everywhere. So, this is
the Floyd-Warshall algorithm.

(Refer Slide Time: 18:59)

So, to summarize, we started with Warshall’s algorithm. So, Warshall’s algorithm is an
alternative way to compute transitive closure. So, it is an iterative transitive closure algorithm.
Earlier, we did it in terms of lengths of paths and now we have done it in terms of which
intermediate vertices are I am allowed to visit while traveling from i to j. So, that is the
difference between Warshall’s algorithm and the way we had formulated in terms of paths of
length 1, 2, 3, and so on.

So, Bk[i,j] is 1, if I can only reach i to j using only vertices 0 to k - 1. So, when we adapt
Warshall’s algorithm for shortest paths, what we do is we replace that calculation of OR, I can
either reach it without using k or I can use it reach it with using k, I replace that OR by
minimum, what is the minimum I can do without using k, what is the minimum I can do with
using k and I use that to update this thing from i to j. So, this is called the Floyd-Warshall
algorithm and it should work as we saw with negative weights provided there are no negative
cycles.
 Mathematics for Data Sciences 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.5
Minimum Cost Spanning Tress
So, we have looked at shortest paths, both the single source version and the all-pairs version
with and without negative weights. And now in the context of weighted graphs, we move to a
different problem, which is the problem of computing minimum cost spanning trees.

(Refer Slide Time: 00:31)

So, to motivate this problem, let us look at a couple of examples. So, here is the first example.
So, supposing you are in a district, which has been hit by a cyclone, and many of the roads are
damaged. So, immediately after the cyclone, of course, the first priority is to restore the roads.
But you also want to restore the roads in such a way that everybody can move around as quickly
as possible.

So, you do not want to start at one end of the district and move sequentially to the other end of
the district, what you want to do is prioritize the roads to be repaired, so that everybody is
connected to everybody as fast as possible. So, which set of roads should you restore, so that
connectivity across the district is maximally restored, rather than individual parts being
connected and other parts being disconnected?

Here is another context. So, suppose you are an internet service provider. So, you provide
internet connectivity to a large number of customers in different cities, and then your customers
are demanding reliability. They are saying that in some cases, because of some damage, either
due to an accident or due to some construction or something, if a cable between two cities gets
cut, then their services cut.

So, you want to lay a parallel cable to ensure that if one cable is cut, the other cable still works.
But at the same time, you want to do this in such a way that you do not spend too much laying
parallel cables everywhere, you do not want to double up every cable in your network, you
want to double up sufficient number of cables, such that between any two locations on your
network, there is a redundant route.

So, you are not obliged to put a double cable between every pair of nodes or every pair of cities
on a network, only enough of them so that everyone is guaranteed to be connected to everyone
else even if one link fails. So, this is a related problem. So, these are both problems will feed
into this problem of finding a spanning tree.

(Refer Slide Time: 02:15)

So, a spanning tree essentially asks us how do we take a graph which is connected and retain a
minimum set of edges so that it remains connected. So, a minimum set of edges that is
connected is a tree. So, we said that a tree is a connected acyclic graph, and we will talk about
trees in more detail in this lecture. But the intuition is that if you want to connect n nodes in a
minimal way, what you end up doing is connecting them in such a way that there are no
redundant paths, there are no cycles, so this is a tree.

So, if you add an edge to a tree, you add redundancy, so you get a loop. If you remove an edge
from a tree because it is kind of minimal, if you remove an edge from a tree, the tree will fall
apart, it is no longer going to be connected that is why it is a minimal acyclic connected graph.
So, what we want in this situation both in the road situation and in the telecom situation, that
ISP situation is that we want to connect a subset of the nodes. Now, we want to say, we want
to restore a subset of the roads, or we want to double up a subset of the links such that
everything is connected to everything.

So, we want to find a subset of the edges in the original graph, which if I deal with them, either
by repairing the roads or by upgrading them to a double link, I will end up connecting
everything to everything. So, here on the right for instance is a graph and this red thing is a
spanning tree. So, it is a spanning tree. So, spanning tree is something that connects all the
vertices, so it spans the graph, it touches every vertex in the graph and it is a tree so it is a subset
of the edges it touches every vertex in the graph, it is a tree so, the red edges here form a
spanning tree.

Now, this spanning tree is not going to be unique. So, here is another spanning tree. So, this
orange where it is now also this one is also a spanning tree. So, the earlier one was one which
went this way and now we have one which goes this way. So, we have two different spanning
trees you can have multiple spanning trees. So, you could also have a spanning tree which goes
like this for it this is also spanning tree.

(Refer Slide Time: 04:22)

So, our interest is weighted graphs. So, supposing our goal is not just that we want to find a
subset of roads to fix or a subset of edges telecom links to double up, but there is a cost
associated with this. So, laying a road depending on the location and various other features,
laying road may not be the same cost all over the place. Similarly, there may be difficulty in
laying cables in some places, not in other places.
So, now if we have a difficulty or a cost or some kind of measure associated with every edge
that we want to deal with, can I find a shortest or minimum cost way of doing this? So, I will
want to find a minimum cost spanning tree? So, it is not. So, we saw that there could be many
different spanning trees. So, it is not just any old spanning tree, but a spanning tree, who if I
look at the cost of the all the edges, which I am adding to that tree, so that is the way I am going
to define.

So, remember, when we had a shortest path and a weighted graph, we added the cost of all the
edges in the path. So, here we are constructing a tree in a graph and we are going to take all the
edges that fall into the tree and say that is the total cost I am going to spend if I am going to
build this tree, if I am going to repair these roads, or if I am going to develop these cables, this
is going to be my total cost. So, I want to find the minimum spanning tree and this is called a
minimum cost spanning tree.

So, if I look at this example, for instance, so here is one spanning tree. So, this spanning tree
has cost 18+ 6 24+ 17 94+ 20 14. Now, we can easily check that this is not a minimum cost
spanning tree, in this case, is small graph, because we can construct this green tree for instance,
which has a shorter cost. So, this is 18+ 6 24, 44, this is not this is 52. So, this is actually 28+
6 44 52. But if I take out this and I put this instead, so this is also a spanning tree. So, this is a
spanning tree also for this graph and this you will check has cost 44. It is 28+ 16 is 44. So,
among all the trees that I can draw on this particular graph, it turns out that 44 is the best one.

(Refer Slide Time: 06:28)

So, in order to come up with algorithms or strategies to discover minimum costs spanning trees,
we will do some basic facts about trees and these will be useful in general. So, it is very good
to write them down once and for all so that you know them and you remember them so that
you are aware of what you are doing when you are dealing with trees. So, as far as we are
concerned, the basic definition of a tree is that it is a connected graph and it is acyclic this is
all we are told you are given n vertices, the graph on n vertices is connected, and it has no
cycle. So, we are assuming this is an undirected graph. So, it has no undirected cycles. What
can you conclude from this?

So, this is a tree, a tree is just a connected graph, which is acyclic. So, the first thing we will
conclude is that if the graph had n vertices, then the tree must have exactly n-1 edges, not more,
not less, it has exactly n-1 edges. So, here is one argument why that is the case. So, we know
that this graph is connected. So, remember that we talked about connected components. So, as
an undirected graph, this whole graph, that I am given initially, is a single component because
it is connected.

But now I also know that it is acyclic. So, it is acyclic, I claim that if I delete an edge, then it
must disconnect the graph because if it did not disconnect the graph, then if I delete an edge,
and I can still go across that edge from i to j via some other route, so I have deleted an edge ij,
in the tree that is given to me, before I deleted the edges connected after deleted if it is still
connected, it means there is still a way to go from it.

But if there is still a way from a go from i to j, does not involve this edge ij. So, if I add back
this edge, ij, then I can go from i go to j by the other path and then come back on the site. So,
there is like, but I also know that the tree is acyclic. So, therefore, it must be the case that when
I remove an edge from a tree, the tree will fall apart into two components. It cannot fall by
more than two components, because there is only one edge only connects two parts. So, the
whole thing was one component, it is like a cut one thread, and the whole thing falls apart and
two pieces.

Now I have two connected things. I cut one more edge, what will happen, one of these two will
fall into two more things. So, every time I cut an edge, I create an extra component, though I
make one component or two components, the other components are unchanged. So, every time
I delete an edge, I am going to create one more component. But how many components can I
create?

Well, I claim that at most, I can create n components because there are only n vertices. Finally,
the minimum component is disconnected vertices isolated by itself with no connections. So, I
started with one component, and I ended up with n components. And every time I did+ 1, so
how many times can I go from 1 to n to n+ 1, n-1 times, so I could only delete n-1 edges. So,
this is one argument saying that every tree on n vertices must have exactly n-1 edges.

So, this says there are no more than n-1. And you can obviously argue that if I had fewer than
n-1, then at some point, this thing would have got disconnected earlier. And that is also a
contradiction. The other flip side to this is that if I add something to this tree, then it will create
a cycle. So, in some sense, this is a minimum connected graph, it is a minimal connected graph,
that is adding more edges will only complicate the situation in terms of connectivity.

So, adding an edge is essentially symmetric to what we said before. So, we said before that if
you delete an edge, you must split the graph into two otherwise it would have been a cycle.
Now if I add an edge, I know that i and j are already connected in the tree. So, if I add an edge
by the same logic, I have created a cycle. So, therefore, whenever I add an edge to a tree, it
creates a cycle.

(Refer Slide Time: 10:13)

The third fact is that between any two points in a tree, there is only one way to go. There is
only one path between any two vertices in a tree. This is not true, in general, as we have seen
in many graphs, you can go many ways. For instance, when you are calculating shortest paths,
we found alternative paths, which got us shorter weights, and so on. But in a tree, this is not
possible in a tree, I can only go from i to j in one way it is connected, guaranteed, but it is
connected by only one way.

So, we will just look at a pictorial thing. So, supposing there are two ways to go from i to j. So,
the argument is that if there are two ways to go from i to j, then somewhere in between, there
must be something like this, a structure like this, where the two paths diverged, and the two, so
the two paths might divergent at i and j itself, it might be that I have to completely separate
paths i to j.

But whichever way if I can go to i one way and come back the other way, either on the entire
full circuit or somewhere in between, there must be the cycle where I can go around the cycle.
So, if I have multiple paths from i to j, there must be a cycle somewhere. So, these are these 3
facts about trees. So, it has exactly n-1 vertices. If I add an edge, it creates a cycle, and there is
a unique path between any two vertices.

(Refer Slide Time: 11:21)

So, to combine this, we can say that, if I give you any two of these conditions, then the graph
is a tree. So, if I tell you that the graph is connected, and acyclic, what is the definition of tree,
I showed you that it has n-1 edges, it is connected as n-1 edges, then it must be acyclic, if it is
acyclic and n-1 edges must be connected. So, if I tell you any two of these three facts, you can
conclude that the graph you are looking at is a tree. So, this is a very useful thing to remember
when you are going forward.
(Refer Slide Time: 11:49)

So, we are going to use some of these facts in order to design algorithms for this problem that
we are considering, which is to build a minimum cost spanning tree. So, remember, a minimum
cost spanning tree is a tree which touches every vertex of the given graph by taking a subset of
edges, which covers all the vertices. And among those, you want a tree in which the sum of the
edge costs that you have used to build this tree is minimum.

So, there are two strategies that one can think of to do this. And we will look at two algorithms
follow these strategies. The first strategy is to start from a single vertex or a smallest single
edge and grow a tree. So, you try to build a tree incrementally, you start, and then you keep
building a tree. So, you start at an edge make another tree, add an edge, make a bigger tree, to
add an edge and it does not make a tree you do not consider it. So, you just grow a tree, so we
will look at it is called Prim’s algorithm.

The other way is to take a disconnected thing and connect it into a tree. So, initially, you can
say that all the vertices are apart and you say, let me take a small edge and connect to things.
So, now I have got the starting point. Let me take two other edges, two other vertices, connect
them and then let me connect this to that. So, you build a tree by kind of grouping together the
components rather than growing one tree. So, this is called Kruskal’s algorithms. We will see
both of these in detail. So, you will understand the difference between these two strategies and
see how they work.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.6
Minimum Cost Spanning Tress: Prim’s Algorithm

So, we are looking at minimum cost spanning tress and weighted graphs and we said there are two
natural strategies which are in simplified by two standard algorithms. So, the first one is Prim’s
Algorithm.

(Refer Slide Time: 0:25)

So, we have a weighted graph, so a graph with a weight function which assigns a number to every
edge. And let us assume that the graph is connected because otherwise, if is not connected to begin
with, we cannot super impose a tree on it. What we want is a spanning tree, a tree which is the
subset of the edges which connects the graph and if there are not enough edges to connect the
graph to begin with, we do not even have a starting point.

So, assuming the graph is connected and it has weights. We want to find a minimum cost spanning
tree which connects all the vertices in V. So, the strategy that we are going to adopt is to
incrementally grow it. So, we start with the smallest edge and we will keep growing the tree by
adding the smallest edge that we can to the current tree while retaining a tree. So, let us look at
this example that we saw before.
So, supposing we start with the smallest edge. So, the smallest edge is the edge between 1 and 3.
Now we want to grow this tree. So, the smallest edge with which so we need to grow it meaning
we have to choose an edge which will extent this tree. So, we cannot choose for instance this edge
right now because this edge is not connected to the tree. That is the smallest edge over all but it is
not connected to the tree. I want to grow the tree. So, I can take any one of the edges coming out
of here. But not that one that one is out I cannot do that one.

So, I can take any one of the edges which is leaving the tree and extend it. So, I take the smallest
among those which is 0 to 1. And so now I have a tree which has 2 edges 1 to 3 and 0 to 1. Now I
can take any edge which is leaving this tree. So, the smallest among them is this edge with weight
18. But unfortunately when I add this edge with 18 I create a cycle. And this is not a tree. So, I
cannot do this, so I have to throw this away. And I have to go to the next one.

So, I go to the next one which is the 20. And now I have tree which connects four of the five
vertices. And now finally because I have reach 2 and now allowed to add this edge. Because now
this edge is connected to the tree that I have constructed so far. So, now add that edge and I will
get this tree which if you remember we have said last time that this tree has weight 44. And you
can check that we have actually found that one, so this is Prim’s Algorithm.

(Refer Slide Time: 2:34)

So, formally what Prim's Algorithm does is it incrementally builds an minimum cost spanning tree.
So, it keeps track over set of vertices which have been added and keeps track of set of edges which
have been added because just because of vertex has been added does not mean that all the edges
between them have been added. So, if you look at previous one for instance when we have added
3 and 0, it does not mean that the edge 0, 3 is added. So, we have to separately note which edge is
belong to the tree.

It is not just enough to say, it is not like a Dijkstra’s algorithm where we said we have burned these
vertices so we are done with them. We need to know which are the edges were used in order to
construct the tree. So, we keep track of these two things separately TV, the tree vertices. And TE,
the tree edges. So, initially everything is empty. The way we describe it we will start with the
smallest edge. So, overall we look at all the edges in the graph and we pick the smallest edge that
edge let it be called, let it be from i to j.

So, once I do that, then I am started with the minimum tree. Let a tree consisting of just one edge.
So, the TV, the vertices in the tree are i and j and the edge is this edge e which I just added which
is from i to j. And now what I will do at each step is I will take an edge which starts in T and goes
out of T. So, that is the reason we could not introduce when we had this graph already drawn the
reason we could not choose this edge is because it both end points are already in the tree that we
have constructed. So, we need an edge like this or like this which starts in the tree and end outside
the tree.

So, we choose a minimum weight edge such that u belongs to tree and the outside the other edge
end point of the edge belongs to the other side. It is not in the tree. Among all these edges which
has starting from the tree and going out you take the smallest one and add it. Once you add it you
have added a new vertex to the tree. So, TV gets expanded to add v in it and TE gets expanded to
add this edge. So, this is the algorithm for Prim.
(Refer Slide Time: 4:28)

So, again going back to that example so this is how we start. We start with both sets empty and
now we take the smallest edge overall which is this one. This is how we start the algorithm. So,
when we start the algorithm we say that 1 and 3 form a tree. So, the tree vertex at TV has 1 and 3
and the edge set consist of the edge 1, 3. Now I look at the smallest vertex which has one end point
and TV and the other end point outside and it turns out to be the edge 0, 1. So,, I add 0 to TV and
I add 0, 1 or 1, 0 because I am drawing it from 1 to 0 to this edge set.

Now I cannot do 18 because it is not inside. So, the next stage that I can do is 1, 2. Because I need
to find an edge which is inside to outside. So, this edge is not allowed because I have this condition
that you say for example 0 must be in the tree vertex set and 3 must be outside. But both 0 and 3
are inside. So, I have to drop that edge. So, I can only look at edges which leave the tree and go
out. So, among those 1, 2 is the next one. So, I get 2 now in my tree set. And my tree edge set has
1, 2.

And finally, from here I can get 8. You are use that edge with weight 8 and get 4 into my edge set
in a vertex set. And 2, 4 in the edge sets. So, now I have got all, so I did this n-3 times as the
important thing to do. So, I basically started with 1, 3 and I did 1, 2, 3 times. So, did this n-2 times.
So, I had five vertices I already started with two vertices. So, I have n-two vertices to go. Every
time I do add an edge I had one more vertex into my set.
So, after I do 10-2 times my vertex set has covered all the vertices remember to assume that the
graph is connected otherwise it is not going to do work. They not be into able to connect it. So,
this seems a little bit, I mean at one side seems reasonable another side it one might ask why is this
going to work. I mean why does this particular strategy actually give me guaranteed the shortest
smallest tree overall.

(Refer Slide Time: 6:27)

So, to do this let us prove some small graph theory fact call the minimum separator lemma. So,
what this says that supposing I take my graph. And I partition it partition it means I divide the
vertices in two sets U and W. Now I look at all the edges which have end points on opposite side
of this. So, there will be some e1 there might be some other edge e2 and so on. There might be
multiple edges which go across this partition.

So, they have one end point in U the other end point in W. So, among them let us assume that we
pick one which is smallest. So, let us just for moment assume all of them have different weight.
So, we will see what to do with different with they have equal weight later on. But let us assume
that they have smallest one supposing one of these has actually smallest. So, maybe this is the
smallest one.

Then what is lemma says is that no matter how I construct an MCST on this graph that particular
edge e3 has to be in MCST. So, the intuition is that somewhere in my graph I will be separating U
from in my tree I would be separating U from W and the best way to connect U to W is via e3. So
I must use e3 in my MCST. So, this is the claim. So, let us see why this is true.

(Refer Slide Time: 7:49)

So, as I said let us assume for now that all the edge weight in our graph are actually different from
each other. So, it just simplifies the argument a little bit. So, this is the situation we are looking at.
So, we have a tree and then this says for every partition. So, this is a universal property it says no
matter how I partition may think as U and W. This property must hold so let me just assume this
is some arbitrary partition on my graph. So, I have split my vertices into two sets which are disjoint
which together cover all set.

So, partition means exactly that partition, if I partition my toys with my assistor then I have to take
some and she has to take some and both of us get one or the other toy. And nobody leave gets left
no toy gets left out. So, partition just means that the two sets are disjoint and the union is the full
set. So, that is what partition is. So, U and W together cover all the vertices and there is no overlap.

So, now the question is that supposing I have a tree and I look at this graph this edge which I am
promised must be in the tree. It says that the smallest edge which connects a vertex in U to a vertex
in W. The smallest edge connecting this partitions must be in my tree. So, supposing it is not in
my tree. So, I am assuming the for the sake of contradiction that I have built a tree which excludes
this vertex which a lemma promises me should be in my tree.
Now other hand I have built a tree. So, this hypothetically this capital T is an MCST. It is a tree, it
is a connected graph on the underline vertices. So, there is no problem going from u to w. the small
u to small w, there must be path because it is connected. So, this path starts in the left partition.
And it ends in the right partition. So, imagine that this is river separating two sides of the city you
cannot go from this side to that side without crossing the river.

So, I am not using this edge e which I have want but I have to cross somewhere. So, there must be
some other edge. Let me call it f where this path crosses. You cannot go from the left side from
the red side to the purple side from capital U to capital W without crossing from this side to that
side via some edge. And if I assume that I have not taken the edge e which I am interested in which
is the smallest edge connecting U to W. I must have taken some other edge.

So, this is what the picture looks like. So, now we can see what happens. So, this is connected, so
now supposing I had drop this edge and instead keep this edge. Then I claim that everything that
was connected before is still connected anything that I can reach I can now go from there and
reach. So, thing that I could reach by following this U to W edge I have this long path which goes
via u, w and coming back to w prime and I can do it.

So, therefore, the connectivity does not change but I have replaced f by a smaller edge. So,
therefore if I replace f by e I have got a smaller or a cheaper spanning tree in terms of cost. So, this
contradicts the assumption that I have started out with a minimum cost spanning tree. I started of
with an MCST and I have told you get a smaller one. So, therefore, this could not being the case.

So, either T was not in MCST or T was an MCST but these assumptions that e did not belong as
false. So, therefore we have establish through this contradiction this lemma that says that if I take
any partition of vertex set and I find the smallest edge going back in fourth across those that
partition that must belong to every MCST that you build. So, if you choose that edge you are okay
if you do not choose that edge you are in trouble. So, just lets dispose of this case of distinct edge
weights before we look at Prim’s Algorithm.
(Refer Slide Time: 11:25)

So, what if two edges have the same weight it is not a big problem. Because you just need to have
a strategy to choose one or the other. So, you need to arbitrarily decide it one has smaller than the
other in order to make this thing work. Which one will go into every tree? So, if you fix a strategy,
so fix a strategy basically you assume that you have numbered the vertices in some fixed way from
0 to m-1.

And then you just decide that the ordering in such that either an edge. So, this is the numbering,
so e with number i and f with number j now. So, supposing e and f are two edges I would have
assigned each of them a number a different number between 0 and m-1. So, I look at now e comma
i and f comma j and if the weight of e is smaller than the weight of j, I declare that e is this edge is
smaller. But if they have equal weight than I will look at the index and I will say the index i is
smaller than the index j then e comma i is smaller than f comma j. So, this gives me a weigh of
kind of ordering all the equal vertices.

So, then my the lemma will say that it must include the smallest, smallest in terms of the ordering.
So, that is what we say so it is not a big deal so it can be done. So, this lemma holds in general
even if the graph has weights which repeat. So, now what is the impact of this lemma on Prim’s
Algorithm.
(Refer Slide Time: 12:36)

So, in Prime’s Algorithm, what we have at any given point is the set of vertices which are in the
tree. So if I look at the set of vertices which are in the tree. And the set of vertices which are not
in the tree. Let me call it W so this is the set-operation. So, V-TV, so I used it there also. So, this
means all the elements which I get by removing the elements of TV from V. S, W and TV together
have all the vertices and they are disjoint so it is a partition.

So, therefore, now if I look at what Prim’s Algorithm does, it says I am currently looking at a tree
which I have built and I am looking at all the edges which go out. Which connect my current tree
to vertices are not in the tree. And I pick the smallest one. But that is nothing but this minimum
separator. For this particular partition I am picking the smallest one. So, therefore, the one that I
am picking has to belong to us. So, I am not picking anything wrong.

So, every, every edge that Prim’s Algorithm picks is guaranteed to belong to every MCST by this
lemma. And since it does that and it picks exactly n-1 edges overall. All the edges are necessary
and therefore they all are part of an MCST. So, that is correctness argument. So, in fact there is a
slightly strongest statement we can make. If we look at a vertex V and we look at everything else
the whole set-V.

Then if I look at the vertices the edges coming out of V they connect the vertex V to its neighbors
and all those neighbors belong to the other partition. So, by this lemma among those edges which
are coming out of a vertex all of them are disconnect, connecting the partition containing V alone
to the remaining things. The smallest edge leaving a vertex is this minimum separator.

So minimum separator separating V alone form everything else and by this lemma that is smallest
edge must belong to every MCST. So, basically if I started a vertex and I look at all the edges
which are connected to it. And I pick the smallest one then that smallest one is guaranteed to be in
every spanning tree, every minimum spanning tree. So, actually therefore it does not really matter
that Prim’s Algorithm started with this minimum cost edge that is the bonus.

The minimum cost edge is of course going to be the minimum separator between the partition of
the two end points. So, that is fine but you can start anywhere. So, you can start with any vertex
and we know that from that vertex the smallest edge leaving it is a minimum cost separator between
it and the rest. So, I can start with the vertex TV set to be just a single vertex and no adjust. And
then I can apply Prim's Algorithm.

What Prim's Algorithm will first discover is the smallest edge which connects V to one of its
neighbors. So, that will by my first edge and so on. So, Prim's Algorithm will work from any
starting point, the first iteration will pick the minimum cost edge leaving v which by this lemma
is correct.
(Refer Slide Time: 15:53)

So, Prim’s Algorithm is a natural way to build a minimum cost spanning tree starting at any vertex
as we saw. The way we first presented it was starting with the minimum cost edge but starting at
any vertex you can build an MCST because of this minimum separator lemma. So, at each edge
what you do is you take the tree have already constructed. And we pick the minimum edge
connecting that tree to the rest.

So, we extend a tree one edge at a time. And at each point because we have going from inside to
outside the new edge is guaranteed to keep it a tree. And finally every edge that we get is
guaranteed to be required by the minimum separator lemma. So, every edge that we add was
required to be in the tree and therefore overall we have added no useless edges. So, we must have
got a minimum cost spanning tree.
 Mathematics for Data Science 1
Professor. Madhavan Mukund
Chennai Mathematical Institute
Lecture No. 12.7
Minimum Cost Spanning Tress: Kruskal’s Algorithm

We are looking at minimum cost spanning tress and we saw one strategy the Prim's strategy which
tries to incrementally grow a tree starting with one vertex or one edge until you get an overall tree
which is minimum cost. The other strategy is called Kruskal's Algorithm.

(Refer Slide Time: 0:32)

So, remember that we have working with weighted undirected graphs. So, we have a weight
function associating a number with every edge. And we assume the graph is connected. And we
want to find the minimum cost spanning tree which connects all the vertices in V. So, in Kruskal’s
Algorithm what we do is we start with all the vertices disconnected forming n components. And
then we try to merge components. We try to connect components by the smallest edge that we
have which connects to components 2.

So, let us do an example and then we will do it in more detail. So, this our familiar example so
here the first thing that we do in Kruskal's Algorithm is to sort the edges in ascending order. So,
we will sort the edges in ascending order and then we start with the smallest edge. So, the smallest
edge is in this case. So, initially we are imagining that we have this disconnected graph consisting
of these five vertices which are just siting an isolation. Then we bring this one edge in and we will
create this component.

So, now we have 4 components one component has the vertices 1 and 3. And the other three are
these isolate components. Now in Prim's Algorithm you would take this component and you would
extend the tree. In Kruskal's Algorithm you just take the smallest edge which connects two
components and makes them into a larger component. So, in this case I jump from 6 to 8. That is
the next smallest edge and it is connecting two components which are separate at the moment.

(Refer Slide Time: 1:57)

So, I am allowed to add that and I get now these two components I have three components, I have
now constructed this kind of a graph. Where I have 0 separate 1 and 3 and then 2 and 4. Now what
is the next one? The next one is this 10, so it will connect 0 to 1.
(Refer Slide Time: 2:16)

So, I am allowed to add it. So, now I have this kind of situation, so I still have overall disconnected
graph it is not a tree yet. And now I have to decide what to do. So, I first is 6, 8, and 10. So, the
next vertex in ascending order of cost is 18.

(Refer Slide Time: 2:38)

Who I discover now is at when I try to add 18 it does not connect two different components. It
connects two vertices in the same component. So, it forms a cycle and this is not good. So,
Kruskal's Algorithm says include an edge if it does not create a cycle. So, 18 gets discarded. So,
we do not take the edge 0, 3 and we proceed to the next stage. So, the next stage in ascending order
is 20.

(Refer Slide Time: 3:01)

So, we add that and we have found essentially the same spanning tree we found from Prim's
Algorithm but in a different sequence. Remember in Prim's Algorithm we first added 6 then we
added 10 then we added 20 and finally we added 8 because we could only add 8 when we reached
one end point of 8 that namely this vertex 2. Where as in Kruskal's Algorithm we added 8 upfront
and we create these disjoint components. And as we go along this components kind of merge
together and they form a tree.
(Refer Slide Time: 3:28)

So, here is the little bit more formal definition of Kruskal's Algorithm. So, we have assume that
our edges are arranged in this ascending order. So, e0 is the smallest edge this is e1, e0, e1 upto
em-1. So, I sorted in ascending order by weight. And now what you do is you keep track of like
we did not Prim's Algorithm. We keep track of the set of edges that we have added. And implicitly
the set of edges also tells us what are the components.

So, you can if you are trying to actually write this as code you will actually keep track for the
components also which is a little bit TDS when you are programming this. But mathematically we
know the edges then we can compute the components by just doing kind of the reachability on
each component. And finding out which components are. So, initially the set of edge is empty.
And now we scan all the edges from the smallest edge to the largest edge.

And if adding it creates a loop or a cycle we skip it otherwise we add it. So, here is the little bit
more complicated graph to try and see how this works. So, these are my 0 to 6 or 7 vertices and I
have some 8 edges in them. So, I sort the edges so this is my smallest edge and this is my second
smallest edge and so on and this is my third. Now I have three edges which are of equal size. So,
these 3 edges, so I fix some ordering we discuss that if you have equal weight edges you just fix
some ordering on them.

So, I have fixed some ordering which is basically based the lexicographic ordering of the end
points. So, 0, 1 comes before 4, 5 because 0 comes in 4, 4 and 4, 5 comes before 4, 6 because 5
come before 6. So, I have just chosen this ordering you can choose any ordering. So, you fix an
order in which you are going to process the vertices. Such that this an ascending order if the equal
vertices you choose some way to group them so that they are in some fixed order. So, initially,
now my edge set is empty and I pick the smallest one.

(Refer Slide Time: 5:21)

So, now I process it from smallest to largest. So, I pick the smallest one no problem it does not
create any I have got no components before this. So, I now created one component, so I can add it
and I have 5, 6 as the tree edge. Now I look at the next stage so remember this is useful that I have
already sorted it. So, I know the next stage I have to process is 1, 2. So, I look at 1, 2 again no
problem it connects two different components so I am fine. Then I look at 0, 1. 0, 1 also connects
two different components so I am fine.

So, now I have come to the 10 block. So, now I look at 4, 5. 4, 5 also connects two different things
because 5, 6 was m component and 4 will m component I am fine. Now I come to 4, 6 and 4, 6
now is actually lying within a component. So, I cannot use 4, 6, 5 skip it. So, I do not add to the
edge set. So, I leave the edge set unchanged.
(Refer Slide Time: 6:17)

Now I come to 0, 2 same problem 0, 2 is connecting two vertices which are already in the dame
component. So, I have to skip that also.

(Refer Slide Time: 6:27)

So, now after 18 I go to 20. And that is not a problem because it takes this component and this
component and connects them they are two different components. So, I add that and finally that
edge 70 which connects 2 to 3 is added. So, this is how Kruskal's Algorithm works.
(Refer Slide Time: 6:40)

So, the reason that Kruskal's Algorithm is correct is exactly the same reason that Prim's Algorithm
is correct is because of this minimum separator lemma remember what the minimum separator
lemma says if I take my graph and I split it so that there are some vertices on this side and some
on the other side. There are this partition into U and W no matter how I partition it. If none sit and
look at all the edges which cross and then among these. If I pick the smallest one then that smallest
one must belong to every MCST.

(Refer Slide Time: 7:14)

So, in our think the edges that we have found so far in Kruskal's Algorithm partition the vertices
into connected components. So, now initially each vertex is in a separate component. And when I
add an edge u, w it merges the components of u and w. So, basically the thing is that if I am keeping
track of which vertices are in the same component and I can do that incrementally. Even I add an
edge I can grow the component I can say the component containing u now contains w also. So, it
is not a problem.

So, I do this as I go along, so if I discover that the edge I want to add actually falls within the
component both end points is the same component. It will then form a cycle and then I discover it.
On the other hand if it connects two different components then we can apply this lemma to argue
that what we are doing is correct because we look at the component containing the starting point.
So, let capital U be the component containing small u. And let W be the rest.

Now W has because we are assuming that U and W are in u small u and small w are different
components, I have U which contains the small u and this W which is the rest it contains small w.
Now we are processing these edges in ascending order. So, since these are in different components
I have not yet connected these two. So, I have not found any edges yet between 0 between capital
U and capital W.

Because if I have found them I would have already connected these components. So, the reason
they are not connected is because I have not found them yet. But among the edges which I have
not looked at the edge I am looking at right now is the smallest one. Because I am doing it in
ascending order. So, this forms a partition and we have forming this we are scanning in ascending
order.

So, this edge that I am looking at must be the smallest edge connecting capital U to capital W. So,
it is satisfies this property of this minimum separator. It is a minimum separator of capital U and
capital W. So, what Kruskal's Algorithm would do? (())(9:15) pick it up an edit. And it is correct
because its separator says that any such edge which is the minimum separator between this
partition and that partition. And must be there in every MCST. So, Kruskal's Algorithm is correct
for the same reason that Prim's Algorithm is correct because of this minimum separator lemma.
(Refer Slide Time: 9:33)

So, the difference between Prim's Algorithm and Kruskal's Algorithm is basically Kruskal's
Algorithm is kind of assembles the tree bottom up. So, it takes all this just the connected things
and then it goes around putting them together. Whereas, the Prim's Algorithm starts somewhere
and it grows a tree gradually to cover the whole thing. So, they have different strategies but both
of them owe that correctness to that one lemma which says that whenever I partition the vertices
into two disjoint sets.

The smallest edge connecting these two partitions must be there in every MCST. Now if there are
repeated edge weights then we already saw in the unweighted case that there are many spanning
tress. If there are repeated the same weight repeats, then we may not get a unique spanning tree.
So, for instance supposing I take a very simple graph which is just this 3 vertex graph and I put
weight 3 here, here and here.

Then, any of these would be a spanning tree that I could take this pair of edges or I could take this
pair of edges or I could make this pair of edges. So, all of them would be minimum cost spanning
tress. And where it comes in our algorithm means when we say choose the minimum cost edge
remember we have said that we might actually have to specify it as some f comma j and so on.

So, this ordering that we choose will decide which one will get picked up. So, that is why we get
different choices. So, here you can see that this triangle on four vertices. And three vertices has
actually three spanning tress. If I do a more complicated thing for instance supposing I take a
square. Pick the four two diagonals then what are the spanning tress well the spanning tress. So,
there are some obvious tress like this one. So, this is a spanning tree. When I take 3 edges around
4 vertices or spanning tree will have 3 edges.

But these are also spanning tress. Two edges and a diagonal this z is also a spanning tree. Two
edges and diagonal connecting them. How many of them are there well there are 4 ways of going
around the outside I can have this, I can have this, I can have this. Then there are some 4 ways of
choosing the corner to include and the then the z can also be in many ways. So, it is many different
orientations. So you can see that with 6 edges I can get a anonymous number of spanning.

(Refer Slide Time: 11:48)

So, in general different choices lead to different spanning trees. And there are not unique edge
weight, then I could get a very large number of spanning. So, depending how I have chosen to
order this equal edge weights. Prim's Algorithm and Kruskal's Algorithm will pick one particular
one out of these. It will not give you all of them it will give you one of them. And if they are
disjoint in the sense that it not disjoint if they are distinct that is if all the edge weights are different
then using the minimum separator lemma you can argue that every choice on Prim’s Algorithm is
forced. Every choice in Kruskal's Algorithm is forced and they will give you exactly the same
spanning trees.

So, as long as the edge weights are disjoint it does not matter whether use Prim's or Kruskal's you
will get the same spanning tree. But if the edge weights repeat then depending on how you choose
to order the vertices. The two algorithm might produce different spanning trees with the same
minimum cost but different set of edges. So, keep that in mind.