18.

06 Exam III Professor Strang May 7, 2014

Your PRINTED Name is:

Please circle your section:

Grading 1:
R01 T 10 36-144 Qiang Guang
R02 T 10 35-310 Adrian Vladu
R03 T 11 36-144 Qiang Guang 2:
R04 T 11 4-149 Goncalo Tabuada
R05 T 11 E17-136 Oren Mangoubi
R06 T 12 36-144 Benjamin Iriarte Giraldo
R07 T 12 4-149 Goncalo Tabuada 3:
R08 T 12 36-112 Adrian Vladu
R09 T 1 36-144 Jui-En (Ryan) Chang
R10 T 1 36-153 Benjamin Iriarte Giraldo 4:
R11 T 1 36-155 Tanya Khovanova
R12 T 2 36-144 Jui-En (Ryan) Chang
R13 T 2 36-155 Tanya Khovanova
R14 T 3 36-144 Xuwen Zhu
ESG T 3 Gabrielle Stoy
 1. (28 points) This question is about the differential equation
   
dy 5 2 1
= Ay = y with y(0) =
dt 8 5 0

(a) Find an eigenvalue matrix Λ and an eigenvector matrix S so that A =

SΛS −1 . Compute the matrix exponential etA by using etΛ .
(b) Find y(t) as a combination of the eigenvectors of A that has the
correct value y(0) at t = 0.
Solutions:

(a) det(A − λI) = 0 ⇔ λ2 − 10λ + 9 = 0. Eigenvalues  λ1 = 1

1
and λ2 = 9. The eigenvector associated to λ1 is v1 =
−2
 
1
and the eigenvector associated to λ2 is v2 = . The matrix
2
   
1 1 2 −1
S= and S −1 = 14 . Finally, etA = SetΛ S −1 =
−2 2 2 1
 1 t 1 9t 1 t 1 9t

2 e + 2 e − 4 e + 4 e
.
−e + e − 2 e + 12 e9t
t 9t 1 t
     
1 1 1
(b) y(0) = =a +b . This implies that a = 12 and
0 −2 2
   
1 1
b = 12 . Hence, y(t) = aeλ1 t v1 + beλ2 t v2 = 12 et + 21 e9t .
−2 2

2
2. (a) (24 points) Suppose a symmetric n by n matrix S has eigenvalues λ1 >
λ2 > . . . > λn and orthonormal eigenvectors q1 , . . . , qn .
If x = c1 q1 + c2 q2 + · · · + cn qn show that xT x = c21 + · · · + c2n and
xT Sx = λ1 c21 + · · · + λn c2n .
xT Sx
(b) What is the largest possible value of R(x) = xT x for nonzero x ?
Describe a vector x that gives this maximum value for this ratio R(x)?
Solutions:

(a) Since the eigenvectors are orthonormal, one has xT x = (c1 q1 + · · · +

cn qn )T (c1 q1 + · · · + cn qn ) = c21 q1T q1 + · · · + c2n qnT qn = c21 + · · · + c2n .
On the other hand, xT Sx = (c1 q1 + · · · + cn qn )T S(c1 q1 + · · · + cn qn ) =
(c1 q1 +· · ·+cn qn )T (λ1 c1 q1 +· · ·+λn cn qn ) = λ1 c21 q1T q1 +· · ·+λn c2n qnT qn =
λ1 c21 + · · · + λn c2n .
2 2
(b) Using (a), R(X) = λ1 cc11+···+λ n cn
+···+cn . Since λ1 > λ2 > · · · > λn , R(X)
is maximal when c2 = · · · = cn = 0 and c1 6= 0. In this case the
largest value of R(x) is λ1 and the associated vector x is any non-zero
multiple of q1 .

3
 3. (24 points)
(a) Show that the matrix S = AT A is positive semidefinite, for any matrix
A. Which test will you use and how will you show it is passed?
(b) If A is 3 by 4, show that AT A is not positive definite.
Solutions:

(a) Energy test. For every vector x we have xT Sx = xT AT Ax =

(Ax)T (Ax) = ||Ax||2 ≥ 0. Hence, S is positive semidefinite.
(b) Since A is 3×4, one has dim(C(A)) ≤ 3 and dim(N (A)) ≥ 1. Hence,
there exists a non-zero vector v such that Av = 0. As a consequence,
AT A is not positive definite.

4
 4. (24 points)
(a) Show that none of the singular values of A are larger than 3.
 
1 1 1
A =  0 1 1 .
0 0 1

(b) Why does B = AQ have the same singular values as A? (Q is an orthog-

onal matrix.)
Solutions:
 
1 1 1
(a) AT A =  1 2 2 . Hence, tr(AT A) = 6. However, AT A is posi-
1 2 3
tive semidefinite, therefore all the eigenvalues√
are nonnegative. This
implies that 0 ≤ λi ≤ 6 and hence that σi ≤ 6 ≤ 3.
(b) Since B T B = QT AT AQ, the matrixes B T B and AT A are similar.
This implies that they have the same eingenvalues
√ and therefore
that B and A have the same singular values σi = λi .

5
Scrap Paper