Discrete calculus

Gustavo Lau
Discrete calculus

The purpose of this presentation is to give an introduction to a

version of calculus that applies to sequences. It is known as
finite or discrete calculus.

This relatively unknown theory has, surprisingly, many

analogous results and parallel techniques to those of traditional
calculus.

This presentation is aimed at UK Year 13 students that are

taking Further Maths A-level.
Discrete calculus

The basic objects of study in traditional calculus are functions

that have real numbers as both input and output (domain and
range):
f :R→R
In discrete calculus the functions have as input the natural
numbers:
u:N→R
We traditionally call these functions sequences and we use the
notation un instead of u(n). We use n instead of x to
emphasize that n is a natural number.
Discrete calculus

Continuous function: Discrete function (sequence):

f (x) = x2 un = n2
Derivative and difference

Let’s recall the definition of the derivative:

f (x + h) − f (x)
f 0 (x) = lim
h→0 h
In the discrete case the smallest possible h is 1:

f (x + 1) − f (x)
1
This leads to the discrete version of the derivative. The
difference of a sequence un is defined as:

∆un = un+1 − un
Difference
The difference of a sequence un is:
∆un = un+1 − un
Examples:
n 0 1 2 3 4 5 Sequence type
un 7 7 7 7 7 7 Constant
∆un 0 0 0 0 0 Zero

un 0 6 12 18 24 30 Linear
∆un 6 6 6 6 6 Constant

un 0 1 4 9 16 25 Quadratic
∆un 1 3 5 7 9 Linear

un 1 2 4 8 16 32 Geometric (Exponential)
∆un 1 2 4 8 16 Geometric (Exponential)

The differences behave similarly to the derivatives.

Geometric interpretation of the difference
The difference of a sequence un is:

∆un = un+1 − un

un = n2

∆u1 = u2 − u1 = 4 − 1 = 3 is the gradient (slope) of the line that

goes through (1, u1 ) and (2, u2 ).
In general, ∆un is the gradient of the line that goes through
(n, un ) and (n + 1, un+1 ).
This is analogous to: f 0 (x) is the gradient of the tangent of
y = f (x) that goes through (x, f (x)).
Difference of n2
The difference of un = n2 is:

∆un = (n + 1)2 − n2 = 2n + 1

We would like to obtain a similar result to

f (x) = x2 =⇒ f 0 (x) = 2x

For that purpose we define the falling factorial (power):

n2 = n(n − 1)

Then:

un = n2 =⇒ ∆un = (n + 1)2 − n2
= (n + 1)n − n(n − 1)
= 2n
Differences

Worksheet 1
The difference of a sequence un is:

∆un = un+1 − un

In general, the falling factorial (power), n to the k falling,

when k > 0, is:

nk = n(n − 1) . . . (n − k + 1)
n0 = 1
1
n−k =
(n + 1)(n + 2) . . . (n + k)
Difference of a falling factorial
The falling factorial (power), n to the k falling when k > 0, is:

nk = n(n − 1) . . . (n − k + 1)

If un = nk then:

∆un = (n + 1)k − nk
   
= (n + 1)n . . . (n − k + 2) − n(n − 1) . . . (n − k + 1)
 
= n(n − 1) . . . (n − k + 2) n + 1 − (n − k + 1)
= knk−1

Therefore:
un = nk =⇒ ∆un = knk−1
This is the discrete version of:

f (x) = xk =⇒ f 0 (x) = kxk−1

Difference of a falling factorial
When the exponent is negative the definition is:
1
n−k =
(n + 1)(n + 2) . . . (n + k)

If un = n−k then:

∆un = (n + 1)−k − n−k

1 1
= −
(n + 2)(n + 3) . . . (n + k + 1) (n + 1)(n + 2) . . . (n + k)
n + 1 − (n + k + 1)
=
(n + 1)(n + 2) . . . (n + k + 1)
−k
=
(n + 1)(n + 2) . . . (n + k + 1)
= −kn−k−1

∴ un = n−k =⇒ ∆un = −kn−k−1

Difference of a geometric (exponential) sequence
If un is a geometric sequence we have:
un = cn =⇒ ∆un = cn+1 − cn = cn (c − 1)
In particular if c = 2:
un = 2n =⇒ ∆un = 2n
This is the discrete version of the exponential function:
f (x) = ex =⇒ f 0 (x) = ex
Another way to see that 2 is the discrete version of e is:
n   n ∞
X n X n(n − 1) . . . (n − r + 1) X nr
2n = (1 + 1)n = = =
r r! r!
r=0 r=0 r=0

This is similar to the exponential function’s MacLaurin series:

∞
X xr
ex =
r!
r=0
Properties of differences
If un and vn are sequences and c is a constant:
∆(un + vn ) = ∆un + ∆vn (Linearity 1)
∆(cun ) = c∆un (Linearity 2)
∆(un vn ) = (∆un )vn+1 + un ∆vn
= (∆un )vn + un ∆vn + ∆un ∆vn (Product rule)
 
un (∆un )vn − un ∆vn
∆ = (Quotient rule)
vn vn vn+1

These are similar to:

(f + g)0 (x) = f 0 (x) + g 0 (x) (Linearity 1)
0 0
(cf ) (x) = cf (x) (Linearity 2)
0 0 0
(f.g) (x) = f (x)g(x) + f (x)g (x) (Product rule)
 0
f f 0 (x)g(x) − f (x)g 0 (x)
(x) = (Quotient rule)
g [g(x)]2
Geometric proof of the product rule

∆vn un ∆vn

(∆un )vn+1
vn+1
un vn
vn

un ∆un
un+1

∆(un vn ) = un+1 vn+1 − un vn = (∆un )vn+1 + un ∆vn

Indefinite sum

The indefinite integral, also known as the antiderivative, is the

inverse operation to the derivative. That is, to evaluate the
integral:
Z
f (x)dx

we search for a function F such that f (x) = F 0 (x).

The discrete version is the indefinite sum, also known as the

antidifference. It is the inverse operation to the difference. To
evaluate the sum:
X
un
n

we search for a sequence U such that un = ∆Un .

Indefinite sum
Another way to describe the indefinite integral is:
Z
f 0 (x)dx = f (x) + C

and the analogous for the indefinite sum is:

X
∆un = un + C
n

Leibniz introduced the integral symbol based on the long letter

S (a letter that fell out of use in the 19th century) because he
thought of the integral as an infinite sum of infinitesimal
summands. He even used the name ’calculus summatorius’,
the name integral calculus appeared later.

It was common to use S for summation until Euler’s capital

sigma, the Greek S, became popular.
Indefinite sums

Worksheet 2
To evaluate the sum:
X
un
n

we search for a sequence U such that un = ∆Un .

Elementary differences and sums
Sequence Difference Sum
X
un ∆un un
n
c 0 cn

n2
n 1
2

nk+1
nk knk−1 , k 6= −1
k+1

cn
cn cn (c − 1)
c−1
n
X 1
n−1 −n−2 Harmonic number Hn = ,
r
r=1
x
1
Z
the discrete version of ln(x) = dr
1 r
Properties of indefinite sums
If un and vn are sequences and c is a constant:
X X X
(un + vn ) = un + vn (Linearity 1)
X X
(cun ) = c un (Linearity 2)
X X
un ∆vn = un vn − vn+1 ∆un (Sum by parts)

These are similar to:

Z Z Z
[f + g](x)dx = f (x)dx + g(x)dx (Linearity 1)
Z Z
cf (x)dx = c f (x)dx (Linearity 2)
dv du
Z Z
u dx = uv − v dx (Integration by parts)
dx dx

Note that the sum and integration by parts formulas come from
the respective product rules.
Definite sum
The fundamental theorem of calculus: If f (x) = F 0 (x) then
Z b b
f (x)dx = F (x) = F (b) − F (a)

a a

has this discrete version: If ur = ∆Ur then

b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a

This is the method of differences to compute telescoping sums:

b
X b
X
ur = ∆Ur
r=a r=a
= (Ua+1 − Ua ) + (Ua+2 − Ua+1 ) + · · · + (Ub+1 − Ub )
= Ub+1 − Ua
Fundamental theorem of sum calculus
If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a

Let’s solve a classic example of the method of differences using

the fundamental theorem of sum calculus:
b
X 1
r=a
r(r + 1)
!
1 1 1 1
Ur = − =⇒ ∆Ur = − − − =
r r+1 r r(r + 1)
b
b+1 !
X 1 1  1 1 1 1
∴ =−  =− − − =− +
r=a
r(r + 1) r b+1 a b+1 a
a
Definite sums

Worksheet 3
Fundamental theorem of sum calculus: If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a
Fundamental theorem of sum calculus

If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a
n
X
(2r + 1) = ?
r=0

To evaluate this using the fundamental theorem of sum

calculus, we need to find Ur such that ∆Ur = 2r + 1:

Ur = r2 =⇒ ∆Ur = (r + 1)2 − r2 = 2r + 1
n
X n+1
∴ (2r + 1) = r2  = (n + 1)2 − 02 = (n + 1)2

0
r=0
Fundamental theorem of sum calculus
If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a
b
X
cr = ?
r=a
cr cr+1 cr
Ur = =⇒ ∆Ur = − = cr
c−1 c−1 c−1
b
b+1
X cr  cb+1 − ca
∴ cr = =

c − 1 c−1

r=a a
b
X cb+1
−1
In particular, cr = (Geometric series)
c−1
r=0
Fundamental theorem of sum calculus

If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a
b  
X r
=?
r=a
k−1
 
r
Ur =
k
     
r+1 r r
∆Ur = − = (Pascal’s rule)
k k k−1
b    b+1    
X r r  b+1 a
∴ = = −
k − 1 k k k

r=a

a
Fundamental theorem of sum calculus
If ur = ∆Ur then
b
X b+1
ur = Ur  = Ub+1 − Ua

a
r=a
Xn
rk = ?
r=0
rk+1
Ur = =⇒ ∆Ur = rk
k+1
n+1
n k+1 
X r (n + 1)k+1
∴ rk = =

k + 1 k+1

r=0 0

The continuous version of this is:

Z n
nk+1
xk dx =
0 k+1
Sums of powers

We can use
n
X (n + 1)k+1
rk =
k+1
r=0

to compute sums of powers:

n n
X X (n + 1)2
r= r1 =
2
r=0 r=0
(n + 1)n
=
2
Sums of powers

n
X (n + 1)k+1
rk =
k+1
r=0

What about the sum of squares?

r2 = r(r − 1) = r2 − r
∴ r2 = r2 + r1
n n
X X (n + 1)3 (n + 1)2
r2 = (r2 + r1 ) = +
3 2
r=0 r=0
(n + 1)n(n − 1) (n + 1)n
= +
3 2
n(n + 1)(2n + 1)
=
6
Sums of powers

Worksheet 4

n
X (n + 1)k+1
rk =
k+1
r=0
n n
X X (n + 1)3 (n + 1)2
r2 = (r2 + r1 ) = +
3 2
r=0 r=0

Using this technique compute the sum of cubes:

n
X
r3 = ?
r=0
Sums of powers

n
X (n + 1)k+1
rk =
k+1
r=0

r3 = r(r − 1)(r − 2) = r3 − 3r2 + 2r

r3 = r3 + 3r2 − 2r = r3 + 3(r2 + r1 ) − 2r1
∴ r3 = r3 + 3r2 + r1
n n
X
3
X (n + 1)4 3(n + 1)3 (n + 1)2
r = (r3 + 3r2 + r1 ) = + +
4 3 2
r=0 r=0
(n + 1)n(n − 1)(n − 2) 3(n + 1)n(n − 1) (n + 1)n
= + +
4 3 2
n2 (n + 1)2
=
4
Sum by parts
The sum by parts formula for definite sums is:
b
X b
X
ur ∆vr = ub+1 vb+1 − ua va − vr+1 ∆ur
r=a r=a

For example, to evaluate this sum by parts:

n
X
2r3r
r=0

we let ur = r and ∆vr = 2 · 3r , then ∆ur = 1, vr = 3r and:

n
X n
X
r n+1 0
2r3 = (n + 1)3 −0·3 − 3r+1 · 1
r=0 r=0
n
X 3n+1 − 1
= (n + 1)3n+1 − 3 3r = (n + 1)3n+1 − 3
3−1
r=0
 1 3
= 3n+1 n− +
2 2
Sum by parts

Worksheet 5
The sum by parts formula for definite sums is:
b
X b
X
ur ∆vr = ub+1 vb+1 − ua va − vr+1 ∆ur
r=a r=a

Exercise: evaluate this sum by parts:

n
X
r2r
r=0
Sum by parts
The sum by parts formula for definite sums is:
b
X b
X
ur ∆vr = ub+1 vb+1 − ua va − vr+1 ∆ur
r=a r=a

To evaluate this sum by parts:

n
X
r2r
r=0

we let ur = r and ∆vr = 2r , then ∆ur = 1, vr = 2r and:

n
X n
X
r n+1 0
r2 = (n + 1)2 −0·2 − 2r+1 · 1
r=0 r=0
n
X
= (n + 1)2n+1 − 2 2r
r=0
n+1 n+1
= (n + 1)2 − 2(2 − 1)
n+1
= (n − 1)2 +2
Linear differential and difference equations
In this context the analogous to the derivative is the shift
operator: Eun = un+1 . The method to solve linear equations is
extremely similar:
Differential equation Difference equation
y 00 − 5y 0 + 6y = 0 un+2 − 5un+1 + 6un = 0
Assume y = emx Assume un = mn and m 6= 0
then y 0 = memx then un+1 = mn+1
and y 00 = m2 emx and un+2 = mn+2
∴ m2 emx − 5memx + 6emx = 0 ∴ mn+2 − 5mn+1 + 6mn = 0
∴ the auxiliary equation is ∴ the auxiliary equation is
m2 − 5m + 6 = 0 m2 − 5m + 6 = 0
(m − 2)(m − 3) = 0 (m − 2)(m − 3) = 0
∴ the general solution is ∴ the general solution is
y = Ae2x + Be3x un = A · 2n + B · 3n

where A and B are arbitrary constants.

Linear differential and difference equations with
constant coefficients

Vector space

y = A e2x + B e3x un = A 2n + B 3n v = A i + B j

Vector spaces are the subject of linear algebra.

Linear differential and difference equations

In both cases the general solution depends on the roots of the

auxiliary equation:

AE’s Differential Difference

roots equation equation
Assume y = emx Assume un = mn and m 6= 0
Distinct
(R or C)
α 6= β y = Aeαx + Beβx un = Aαn + Bβ n
Equal
α y = (Ax + B)eαx un = (An + B)αn
Complex
p ± iq y = epx (A cos qx + B sin qx)
ρe±iθ un = ρn (A cos nθ + B sin nθ)
Linear differential and difference equations

If the equation is not homogeneous:

general solution = complimentary function + particular integral.

The boundary conditions for difference equations are usually

some initial values.

In Further Maths you study the case of order 2. This can be

generalized to higher orders.

Difference equations are also called recurrence relations. A

sequence satisfying a linear recurrence with constant
coefficients is called a constant-recursive sequence.
Linear difference equations
Example: Find a formula for the periodic sequence:
11, 1, 3, 1, 11, 1, 3, 1, . . .

un+4 = un with u0 = 11, u1 = 1, u2 = 3, u3 = 1

Assuming un = mn and m 6= 0, we get:

mn+4 = mn
m4 = 1 (Auxiliary equation)
∴ m1 = 1, m2 = −1, m3 = i, m4 = −i (Roots of unity)
n n n n
∴ un = A · 1 + B(−1) + Ci + D(−i) (General solution)
A + B + C + D = 11 (u0 = 11)
A − B + Ci − Di = 1 (u1 = 1)
A+B−C −D =3 (u2 = 3)
A − B − Ci + Di = 1 (u3 = 1)
n n n
∴ un = 4 + 3(−1) + 2i + 2(−i)
Linear difference equations

Assume un = mn and m 6= 0. If the auxiliary equation has

distinct roots α and β then the general solution is:

un = Aαn + Bβ n

Exercise: Using this technique and assuming that the first term
is u0 , find a formula for the periodic sequence:

20, 10, 20, 10, 20, 10, . . .

A very simple periodic sequence
Find a formula for the periodic sequence:
20, 10, 20, 10, 20, 10, . . .
un+2 = un with u0 = 20, u1 = 10

Assuming un = mn and n 6= 0, we get mn+2 = mn and:

m2 = 1 (Auxiliary equation)
∴ α = 1, β = −1 (Roots of unity)
n n n
∴ un = A · 1 + B(−1) = A + B(−1) (General solution)
A + B = 20 (u0 = 20)
A − B = 10 (u1 = 10)
∴ A = 15, B = 5
∴ un = 15 + 5(−1)n

In general, if the period is n we would have the n-th roots of 1.

Linear differential and difference equations with
constant coefficients

Worksheet 6
Assume un = mn and m 6= 0. If the auxiliary equation has
distinct roots α and β then the general solution is:

un = Aαn + Bβ n .

If the roots are the same, α, then the general solution is:

un = (An + B)αn .
Fibonacci sequence
un+2 = un+1 + un
u0 = 0, u1 = 1
Assuming un = mn and n 6= 0, we get:
mn+2 = mn+1 + mn
m2 = m + 1 (Auxiliary equation)
√ √
1+ 5 1− 5
∴α= ,β = (Golden ratio and its conjugate)
2 √ 2 √ 
1+ 5 n 1− 5 n
  
∴ un = A +B (General solution)
2 2
A+B =0 (u0 = 0)
√ √
1+ 5 1− 5 2
A +B = 1, A − B = √ (u1 = 1)
2 2 5
√ √
1 1+ 5 n 1 1− 5 n
   
∴ un = √ − √
5 2 5 2
This closed-form expression is known as Binet’s formula.
Is it all a coincidence?
A more general definition of a difference of a function f is:

∆h f (x) = f (x + h) − f (x)

where h is a positive number.

The difference that we have studied is the particular case when
h = 1:
∆un = ∆u(n) = ∆1 u(n)
Using ∆h we can rewrite the definition of the derivative like this:

∆h f (x)
f 0 (x) = lim
h→0 h
This equality explains why the properties of differences and
derivatives are so similar. There is an analogous relationship
between sums and integrals, just check the definition of
Riemann integrals.
Applications and discrete mathematics

Numerical calculus: discretize differential equations to solve

them numerically. For example, Euler’s method consists of
approximating a function by a sequence defined by a
recurrence relation.

Computer science: recurrence relations are used to analyze

recursive algorithms.

Many branches of mathematics have discrete and continuous

versions: probability distributions, Fourier transforms and
dynamical systems among many others.

The appearance of digital, as opposed to analog, computers in

the 20th century increased the importance of discrete
mathematics.
Summary
Continuous calculus Discrete calculus
Functions Sequences

Derivative Difference
xk nk

Integral Sum

Fundamental theorem Method of differences

of calculus

Differential equations Difference equations

(Recurrence relations)
emx mn

Laplace transform Generating function

References

I George Boole. Calculus of Finite Differences. 1860. The

first textbook on the subject. On page 164 it describes the
discrete version of the integrating factor method to solve
first order differential equations.
I Murray R. Spiegel. Calculus of Finite Differences and
Difference Equations. Schaum’s outlines, 1971. It has a
detailed description of general discrete calculus, that is in
terms of h.
I Ronald L. Graham, Donald E. Knuth, and Oren Patashnik.
Concrete Mathematics: A Foundation for Computer
Science. 1994. In the title, by Concrete they mean a blend
of CONtinuous and disCRETE!
I Art of Problem Solving, Why Discrete Math Is Important.
References
I A. I. Markushevich. Recursion sequences. Mir, 1975. Part
of the great Soviet collection Little Mathematics Library.
Available at urss.ru.
I Robert Ghrist, University of Pennsylvania. Discrete
Calculus videos: Sequences I, Sequences II, Differences
and Discrete Calculus.
I Richard Feynman, Feynman lectures. California Institute of
Technology. Newtons Laws of Dynamics. It includes nice
examples of discretization of differential equations.
I Wikipedia: Finite difference, Indefinite sum,
Recurrence relation, Linear difference equation,
Constant-recursive sequence.
I You can find this presentation at gustavolau.com.