You are on page 1of 7

Math 206, Spring 2016

Assignment 13 Solutions

Due: April 22, 2016

Part A.
(1) Consider the subspace spanned by {1, x, x2 , x3 , x4 } within C[1, 1]. Use the Gram-Schmidt process to
create an orthonormal basis for this subspace. (Of course, you should use the standard inner product
on C[1, 1].)
Solution. Let v0 = 1, v1 = x, v2 = x2 , v3 = x3 , v4 = x4 . Well produce an orthonormal collection
{u0 , u1 , u2 , u3 , u4 } from this using Gram-Schmidt. The only way in which our computations will diverge
from the Gram-Schmidt algorithm in Rn is that we will replace the dot product with the standard inner
product on C[1, 1].
Computing u0 is the easiest. We will use the fact that kv0 k =
u0 =

qR

1
1

1 dx =

2. We then have

1
1
1
v0 = 1 = .
kv0
2
2

Now for u1 , we start by computing


1

Z


1
1
x dx = x 0 = x.
2
2
1
R1
To produce u1 we need to normalize this result; observe that kxk2 = 1 x x dx = 32 , and so

1
3
u1 = p
x = x.
2
2/3
v1 hv1 , u0 iu0 = x

To compute u2 we first consider


v2 hv2 , u0 iu0 hv2 , u1 iu1 = x2



Z 1
1
3
1
3
x2 dx
x2 x dx x
2
2
2
2
1
1

Z

2 1
0
3 2
1
= x2 .
3
2

=x

Now we need to normalize the result, which requires us to compute kx2 31 k =


Hence were left with


45
1
2
u2 =
x
.
3
8

qR
1

(x2 13 )2 dx =
1

8
45 .

Now we compute u3 . Well need to know the following inner products:




Z 1
1
3
hv3 , u0 i =
x
dx = 0
2
1
!

Z 1
3
6
3
x dx =
hv3 , u1 i =
x
5
2
1

!
Z 1
45
1
3
2

x
dx = 0.
hv3 , u2 i =
x
3
8
1

http://palmer.wellesley.edu/~aschultz/w16/math206

Page 1 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Hence we produce u3 by normalizing the vector v3

6
5 u1

= x3

Due: April 22, 2016



6 3
x
5
2

= x3 53 x:



3
175
3

=
x

x
.
2
5
8
x3 53 x dx

x3 35 x

u3 = qR
1
1

We are finally ready to produce u4 . Again, well need to compute some inner products:
Z

hv4 , u0 i =

x4

hv4 , u1 i =
1

x4

hv4 , u2 i =
1

hv4 , u3 i =

x4


1
2

dx =
5
2
!
3
x dx = 0
2


!
45
1
4 2
2

x
dx = .
3
8
7 5

!

175
3

dx = 0.
x3 x
5
8

Now we need to normalize

v4 = v4



4 2
1 6
1
2
u0 u2 = x4
x2
.
5
5 7
3
7 5

This vector has


kv4 k2 =

128
11025

We get
r
u4 =

11025
128




1 6
1
4
2
x
x
.
5 7
3

(This problem required approximately 13% of my soul to complete.)

(2) Prove that the only eigenvalues of an orthogonal matrix are 1 and 1.
Solution. Suppose that A is orthogonal and is an eigenvalue; let x be an associated nonzero
eigenvector. This means that Ax = x. Since A is orthogonal we know that
kxk = kAxk = kxk = ||kxk,
and since x 6= 0 we know that kxk =
6 0. Hence in the above equation we can cancel the factors of kxk
on both sides to deduce that || = 1. This gives the desired result.

(3) Complete problems 16 and 17 from section 5.4.
http://palmer.wellesley.edu/~aschultz/w16/math206

Page 2 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Due: April 22, 2016

Solution. First we solve problem 16. Suppose that A Rrc . Then we have
rank(A) = dim(im(A))

(theorem earlier in semester)

= r dim(im(A) )

(complementarity of dimensions of orthogonal complements)

= r dim(ker(A ))

(our given fact)

= dim(im(A ))

(rank nullity)

= rank(A ).

For problem 17, it is true that rank(A) = rank(AT A). To see this, recall that we already know that
ker(A) = ker(AT A) from class, and so dim(ker(A)) = dim(ker(AT A)). By rank nullity we have that
dim(ker(A)) = m rank(A), and likewise we have dim(ker(AT A)) = m rank(AT A). Setting these two
expressions equal to each other gives the desired result.

(4) Complete problem 19 from section 5.4.
Solution. Since A is invertible, the unique least square solution to Ax = b is
x = AT A

1

AT b.

Since A has orthonormal columns in this particular problem, we know that AT A = I, and hence





1
1 0 0
1
1
x = AT b =
=
.
0 1 0
1
1

Part B.
(1) Prove that the set


cos(t) cos(2t) cos(3t)


, , ,

sin(t) sin(2t) sin(3t)


, , ,

is an orthonormal collection in C[, ] (using the inner product we gave in class). For a given M N
and f C[, ], find the coefficients a0 , a1 , , aM and b1 , , bM so that
f = a0 +

M
X

an cos(nt) +

n=1

M
X

bn sin(nt)

n=1

is as close to f as possible. (I.e., if f is the function one gets from any other choice of coefficients in
such a linear combination, then kf f k < kf f k.)
http://palmer.wellesley.edu/~aschultz/w16/math206

Page 3 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Solution. To show that the given set is orthonormal, we need


identities hold:


Z 
1
1

dt = 1
2
2



Z 
cos(nt)
cos(nt)

dt = 1
for all



Z 
sin(nt)
sin(nt)

dt = 1
for all



Z 
1
cos(nt)

dt = 0
for all



Z 
1
sin(mt)

dt = 0
for all



Z 
cos(nt)
sin(mt)

dt = 0
for all



Z 
cos(mt)
cos(nt)

dt = 0
for all





Z
sin(mt)
sin(nt)

dt = 0
for all

Due: April 22, 2016


to show that the following integral

nN
nN
nN
mN
n, m N
n, m N
n, m N.

The all have mostly the same flavor, so well just show how one derives one of these relations. Let
n, m N be given, and consider
Z
cos(nt) sin(mt) dt.

Observe that if we can show this quantity is 0, then the sixth equation above holds.
First suppose that n = m. In this case we can use the substitution u = sin(nt); then we have
du = n cos(nt) dt, and the bounding values t = and t = both corresponds to u = 0 (since sine of
any integer multiple of is zero). Hence we can rewrite the integral as
Z 0
1
du = 0.
u=0 n
Now suppose that n 6= m. Well
use integration
by parts (the familiar identity from integral calculus
R
R
that says informally that u dv = uv v du) to compute this integral. Well choose u = sin(mt)
and dv = cos(nt) dt. We then have du = m cos(mt) dt and v = n1 sin(nt). Hence integration by parts
gives


Z
Z
Z
sin(mt) sin(nt)
m
m
cos(nt) sin(mt) dt =

cos(mt) sin(nt) dt =
cos(mt) sin(nt) dt.
n
n

Well use integration by parts again, this time with u = cos(mt) and dv = sin(nt) dt. This gives
du = m sin(mt) dt and v = n1 cos(nt). Integration by parts gives
!


Z
Z
m
m
cos(mt) cos(nt)
m

cos(mt) sin(nt) dt =

cos(nt) sin(mt) dt
n
n
n
n

 Z

m
m
=

cos(nt) sin(mt) dt
n
n
2 Z
m
m
cos(nt) sin(mt) dt.
= 2
n n
http://palmer.wellesley.edu/~aschultz/w16/math206

Page 4 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Due: April 22, 2016

Combined with our previous equation, we then have


Z
Z
m2
cos(nt) sin(mt) dt = 2
cos(nt) sin(mt) dt.
n

Since n 6= m by assumption, this equality can only hold if the given integral is zero.
Now we tackle the second part of the problem. Let

 
 

1
cos(t) cos(2t)
cos(M t)
sin(t) sin(2t)
sin(M t)
, , ,
, , ,
V = span

2
The Pythagorean theorem tells us that amongst all vectors in V , the one closest to f is projV f . Hence
if we can find coefficients a0 , a1 , , aM and b1 , , bM so that
M
X

projV (f ) = a0 +

an cos(nt) +

n=1

M
X

bn sin(nt),

n=1

then this vector will satisfy the desired property for f from the problem.
To find the relevant coefficients, note that our spanning set for V is orthonormal, and hence an
orthonormal basis for V . Hence we have




M 
M 
X
1
cos(nt) cos(nt) X
sin(nt) sin(nt)
1
+

+
f,
f,
projV (f ) = f,

2
2 n=1
n=1
Z



M Z
M Z
X
1
cos(nt)
cos(nt) X
sin(nt)
sin(nt)
1

=
dt +
f (t)
dt
+
f (t)
dt
f (t)

2
2 n=1

n=1






Z
Z
Z
M
M
X
X
1
1
1
f (t) dt 1 +
f (t) cos(nt) dt cos(nt) +
f (t) sin(nt) dt sin(nt).
=
2


n=1
n=1
So we should choose
Z
1
a0 =
f (t) dt
2

an =

f (t) cos(nt) dt

bn =

f (t) sin(nt) dt


(2) Suppose that h, i is an inner product on Rn . Prove that there exists a symmetric matrix A so that
hx, yi = xT Ay. Prove that the eigenvalues of A are all positive. Give an example of a symmetric matrix
A so that the function (x, y) = xT Ay is not an inner product.
Solution. Define A to be the matrix whose i, jth entry is hei , ej i. Since the inner product is symmetric,
so too is the matrix A: its i, jth entry is hei , ej i, which is the same as hej , ei i.
Assuming for the time being that we know hx, yi = xT Ay, why must A have positive eigenvalues?
If 0 were an eigenvalue and x a corresponding nonzero eigenvector, then we would know that
hx, xi = xT Ax = xT (x) = (xT x) = (x x) = kxk2
(where the dot product and magnitude at the end of our series of equations are the usual ones in
Rn ). But we know that kxk2 > 0 since the dot product is an inner product, and hence we conclude that
hx, xi = kxk2 < 0. This would contradict one of the axioms of being an inner product.
http://palmer.wellesley.edu/~aschultz/w16/math206

Page 5 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Due: April 22, 2016

Now well prove that hx, yi = xT Ay for the matrix A above. To prove this, write
x = x1 e1 + + xn en
y = y1 e1 + + yn en .
We then have
hx, yi = hx1 e1 + + xn en , y1 e1 + + yn en i
= (hx1 e1 , y1 e1 + + yn en i) + ( + hxn en , y1 e1 + + yn en i)
= (hx1 e1 , y1 e1 i + + hx1 e1 , yn en i) + + (hxn en , y1 e1 i + + hxn en , yn en i)
= (x1 y1 he1 , e1 i + + x1 yn he1 , en i) + (xn y1 hen , e1 i + + xn yn hen , en i)
=

n X
n
X

xi yj hei , ej i.

j=1 i=1

On the other hand, our matrix A Rnn has its i, jth entry as hei , ej i, so we get

he1 , e1 i he1 , en i
y1


..
..
..
xT Ay = x1 xn
.
.
.

x1

xn

hen , e1 i hen , en i
yn

yn hen , e1 i
y1 he1 , e1 i


..
..

+ +

.
.
yn hen , en i
y1 hen , e1 i

= (x1 y1 he1 , e1 i + + xn y1 hen , e1 i) + + (x1 yn he1 , en i + + xn yn hen , en i)


=

n X
n
X

xi yj hei , ej i.

i=1 j=1

Our two calculations agree.

(3) Suppose that A Rcr and that x Rc and y Rr . Prove that y Ax = AT y x.


Solution. We have seen earlier in the semester that z w = zT w, at least assuming one is willing to
identify the 1 1 matrix on the right with an element in R. Observe also that since the matrix on the
right is 1 1, it is equal to its own transpose. Combining these two facts we have
T
y Ax = yT Ax = yT Ax = xT AT (yT )T = xT AT y = x AT y.

(4) In class we proved that for any A Rrc we have im(A) = ker(AT ). Prove for any matrix A we have
ker(A) = im(AT ).
Solution. Let A be a given matrix. Since the identity im(M ) = ker(M T ) holds for any matrix M ,
it also holds for M = AT . In this case it gives
im(AT ) = ker((AT )T ) = ker(A).
If we take the orthogonal complement of both sides and use the fact that (V ) = V for any vector
space V , then we recover the desired equality.


http://palmer.wellesley.edu/~aschultz/w16/math206

Page 6 of 7

Math 206, Spring 2016

Assignment 13 Solutions

Due: April 22, 2016

(5) Complete problem 37 in section 5.4.


Solution. [Note: in my solution I assume that log means natural log, since thats what most
mathematicians mean. It looks like the book intended log to mean the logarithm with base 10. If one
makes the appropriate substitutions to the calculations below for log base 10, you recover the answer
from the book for part (a). The difference gets washed away when we write d = kat .]
For part (a), we have data points {(35, log(35)), (46, log(46)), (59, log(77)), (69, log(133))}, so we want
to solve the system

1 35
log(35)


1 46 c0
log(46)

=
1 59 c1
log(77) .
1 69
log(133)
From class we know that if A is invertible, the unique least squares solution for Ac = b is given by
c = (AT A)1 AT b.
Plugging this expression in for our A and b we have


3.193
c
.
0.013
So our fitting function is log(d) = 2.107 + 0.039t.
For part (b), if we exponentiate our fit equation for log(d) we get
d = e2.107 e0.039t 8.2231.04t .
For part (c), our answer to part (b) implies that we should expect
d 8.22351.0488 259.
Why the discrepancy between predicted screen numbers (259) and actual screen numbers (79)? Presumably the people who designed displays for airplanes realized at some point that pilots cant read a
zillion screens effectively, and so they likely condensed information into a smaller number of screens. Its
also possible that the advent of computerization meant that some data could be handled by a computer
and not seen by the pilots.


http://palmer.wellesley.edu/~aschultz/w16/math206

Page 7 of 7

You might also like