MIDTERM EXAM 1

MATH 312, SECTION 001

Problem 1
[25 points] The LU decomposition of a matrix A is given as follows:
   
1 0 0 1 4 3
L=  −2 1 0 and U = 0 2 6 .
−3 1 1 0 0 3
Part a. [3 points] What is A?
Answer. A is just the matrix product of L and U, so
  
1 0 0 1 4 3
A = −2
 1 0 0 2 6
−3 1 1 0 0 3
 
1 4 3
= −2
 −6 0 .
−3 −10 0
Part b. [3 points] Write down the sequence of row operations which takes A to U when
performing Gaussian elimination.

Answer. Since A = LU, we must have L−1 A = U so the matrix which accomplishes our
row operations is L−1 . One could compute this inverse via Gauss-Jordan elimination and
so forth, but there is an easier way: we can just look at what L does to U and reverse
that. Looking at the entries below the diagonal in L, we see L performs the following
three operations:
R20 = R2 − 2R1 , R30 = R3 − 3R1 and R300 = R30 + R20 .
So, the sequence of row operations which takes us from A to U will be:
R20 = R2 + 2R1 , R30 = R3 + 3R1 and R300 = R30 − R20 .
That’s it, we’re done!
h i
1
Part c. [10 points] Describe how you would solve Ax = 2 as two triangular systems1.
3
Can we solve these two triangular systems in any order, or must one of them be solved
before the other? Explain your answer.
h i h i
1 1
Answer. Well, Ax = 2 means LUx = 2 . So if we set Ux = y, then we are solving
3 3
the two triangular systems h i
1
Ly = 2 and Ux = y.
3
Of course, we must solve the L-system first to get y, otherwise there will be no right hand
side to solve the U system!
1You don’t actually have to solve anything: just explain how you’d set up the two triangular systems
1
2 MATH 312, SECTION 001

Part d. [5 points] Compute the inverse of U using Gauss-Jordan elimination.

We have
1 4 3 | 1 0 0
 

[U | Id] = 0 2 6 | 0 1 0 .
0 0 3 | 0 0 1
Perform the following operations: R10 = R1 − 2R2 and then R20 = R2 − 2R3 to get
1 0 −9 | 1 −2 0
 
0 2 0 | 0 1 −2 .
0 0 3 | 0 0 1
Finally, R10 = R1 + 3R3 gives us
1 0 0 | 1 −2 3
 
0 2 0 | 0 1 −2 ,
0 0 3 | 0 0 1
and all that remains to do is scale R2 by 1/2 and R3 by 1/3. Now U−1 is the right hand
side of the following.
1 0 0 | 1 −2 3
 
0 1 0 | 0 1/2 −1 .
0 0 1 | 0 0 1/3
Part e. [4 points] Compute A−1 , or explain why A is not invertible.

Answer. Note that L is always invertible, being a product of elementary matrices; and
from Part d we know that U is invertible as well. Since A = LU, not only must A
be invertible, but we must also have A−1 = U−1 L−1 . We already have U−1 from Part
d, so it remains to compute L−1 . We multiply the matrices corresponding to the row
operations in Part b:
   
1 0 0 1 0 0 1 0 0
L−1 = 0 1 0 0 1 0 2 1 0
0 −1 1 3 0 1 0 0 1
 
1 0 0
= 2 1 0 .
1 −1 1
So, we now have
    
1 −2 3 1 0 0 0 −5 −3
A−1 = U−1 L−1 = 0 1/2 −1 2 1 0 =  0 −3/2 −1 .
0 0 1/3 1 −1 1 1/3 −1/3 1/3

Problem 2
[40 Points] The matrix B equals MR, where
   
2 2 0 1 0 3 2
M= 0  2 2 and R = 0 1 1 0 .
2 0 2 0 0 0 0
You may also use the fact that
 
1/4 −1/4 1/4
M−1 =  1/4 1/4 −1/4
−1/4 1/4 1/4
 MIDTERM EXAM 1 3

Part a. [5 points] Find a basis for the row space C(BT ).

T
 The
 pivoted
  rows of R provide a basis for the row space of B, so a basis for C(B ) is
1 0
0 and 1 .
3 1
2 0

Part b. [10 points] Find a basis for the column space C(B).
h i h i
1 0
Answer. A basis for the column space of R is given by the pivot columns 0 and 1 ;
0 0
the matrix M takes this to a basis for the column space of B. But the multiplication of
M with hthese
i vectors
h i only extracts the corresponding columns from M, so one basis for
2 2
C(B) is 0 and 2 .
2 0

Part c. [10 points] Find a basis for the null space N(B).

Answer. If we label the variables of R4 by w, x, y and z then we see in R that the

columns corresponding to y and z are free whereas those corresponding to w and x are
not. Expressing w and x in terms of the free variables (using the first two – pivoted –
rows of R) gives
w = −3y − 2z and x = −y.
 −3   −2 
−1
Therefore, a basis for N(B) is given by 1 and 00 .
0 1

Part d. [10 points] Find a basis for the left null space N(BT ). Hint: you might need
M−1 for this part.
A basis for N(BT ) is given by extracting those rows of M−1 which correspond to the
zero rows of R.
 Since
 only the third row of R is zero, our basis is given by the third row
−1/4
of M−1 , i.e., 1/4 .
1/4

Part e. [5 points] State the fundamental theorem of linear algebra, and show that B
satisfies it.
The FTLA states the following: let A be any m × n matrix whose rank (or dim C(A))
equals r. Then, dim N(A) = n − r, dim C(AT ) = r and dim N(AT ) = m − r. In our case,
B is a 3 × 4 matrix with rank 2 (from Part b), and the corresponding dimensions are
• dim N(B) = 4 − 2 = 2 from Part c,
• dim C(BT ) = 2 from Part a, and
• dim N(BT ) = 3 − 2 = 1 from Part d.
Thus, B satisfies the FTLA.

Problem 3
[14 Points] Let B = MR be the matrix from Problem 2.
h 2 i
Part a. [7 points] Find all solutions to Bx = v when v = −2 0 . Hint: you don’t have

to compute the RREF of B | v from hscratch: you |

   −1

can just use R M v .
1 −1 1
ih 2 i h 0 i
−1
Answer. The vector M v equals 1/4 1 1 −1 −2 0 or −1 1 . The augmented RREF
−1 1 1
[B | v] is therefore [R | M−1 v], or
1 0 3 2 | 0
 
0 1 1 0 | 1  .
0 0 0 0 | −1
4 MATH 312, SECTION 001

It is clear from the last row that this system has no solutions: no linear combination of
zeros can produce that minus one.
h i
2
Part b. [7 points] Find all solutions to Bx = v when v = 0 . Hint: see the hint given
2
for Part a. h i
1
Answer. This time the vector M−1 v equals 0 , which is much more promising. The
0
augmented RREF [B | v] is [R | M−1 v], or
1 0 3 2 | 1
 
0 1 1 0 | 0 .
0 0 0 0 | 0
Labeling the variables as w, x, y and z as in Part b of Problem 2, we have to satisfy the
following two equations which express the pivot variables in terms of the free ones:
w = 1 − 3y − 2z and x = −y.
So, the general solution is given by all choices of y and z in the following sum:
     
1 −3 −2
0 −1 0
  +   y +   z.
0  1  0
0 0 1
As you might expect, the last two terms are exactly the null space of B!

Problem 4
[21 points, 3 points each] In each of the following cases, clearly mark the statement as
true or false. Please also explain your answers in order to receive credit for this problem!

a. If a 3 × 4 matrix has a RREF with only three pivots, then its rows are linearly
dependent.
False. There are three rows with three pivots, so they can’t be dependent.

b. If a 3 × 4 matrix has a RREF with three pivots, then its columns must span R3 .
True. The pivot columns must be exactly the standard basis vectors for R3 .

c. If a matrix A satisfies Ax = 0 for some x 6= 0 then A cannot be invertible.

True. If A were invertible, then A−1 would send 0 to x 6= 0.

d. The set A consisting of the X axis, the Y axis, the line y = x and the line y = −x
forms a subspace of R2 .
False. This set contains (1, 0) and (0, 2) but not the sum (1, 2).

e. If a vector k lies in the null space of AT and if Ax = b then A(x + k) also equals b.
False. This would be true if k was in the null space of A, not AT . If A is not a square
matrix, then we may not even be able to add x and k because the dimensions won’t
match up.

f. The product of three invertible 3 × 3 matrices is always invertible.

True. If M = ABC and all the matrices on the right side are invertible, then M−1 =
C−1 B−1 A−1 .
 MIDTERM EXAM 1 5

g. If E is the elementary matrix which adds 3 times Row 1 to Row 2, then E2 adds 9
times Row 1 to Row 2.
False. The matrix E2 just performs this operation twice, with the end result of adding
6 (not 9) times Row 1 to Row 2.