Eksamen Solutions TMA4230 2005-05-31 S

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

TMA4230 Functional Analysis, 31 May 2005

Suggested solution
Text in small type, like this, is not part of the solution, but rather comments on the solution.

Problem 1
a) The closed graph theorem states that, if X and Y are Banach spaces and T : X Y is a linear operator
with closed graph in X Y , then T is bounded.
An operator with a closed graph is itself called closed, so an even briefer statement of the theorem is possible
and acceptable. Kreyszig considers the case where T is defined only on a closed subspace of X . This is only
apparently more general, since a closed subspace of a Banach space is itself a Banach space. But of course,
this formulation is also acceptable. It is necessary for both X and Y to be Banach spaces, as counterexamples
can be constructed if either of them is not complete.

b) First, ker P is a closed subspace: This is true of the kernel of every bounded linear operator, whether
or not they are projections. From P 2 = P we get P P x = P x, so that P y = y if y = P x im P . It follows
that im P = ker(I P ), which is a closed subspace. If y im P ker P then y = P y = 0, so y = 0. Thus
im P ker P = {0}. And we can always write x = P x + (I P )x with P x im P and (I P )x ker P , so
that X = im P + ker(I P ).
c) By assumption, every x X is uniquely written x = y + z with y Y and z Z . If P is a projection
with image Y and kernel Z , then we are forced to have P y = y and P z = 0, so P x = P (y + z) = y.
Moreover, this is well defined P because of the uniqueness of the decomposition of x.
To prove the linearity of P is straightforward, and Ill skip it here.

To prove that P is bounded, we use the closed graph theorem. If x X we write as before x = y + z
with y Y , z Z , and note that P x = y. Thus we get a pair (x, y) = (y + z, y) in the graph of P , and
we note that every member of the graph of P can be written this way. Put differently, a pair (x, y)
belongs to the graph of P if and only if y Y and x y Z .
Let now (x n , y n ) belong to the graph of P for each n, and assume x n x and y n y in the norm
topology, i.e., kx n xk 0 and ky n yk 0.
Now, since Y is closed, y Y is an immediate consequence. And since x n y n xy and x n y n Z ,
and moreover Z is closed, we get x y Z . Thus the limit (x, y) belongs to the graph of P . We have
proved that P has a closed graph, and therefore the boundedness of P .

Problem 2

a) If A is an algebra with unit, x A, and p is a polynomial, then p(x) = p (x) .

b) We apply the identity of the previous question. Since now p(x) = 0, we get p(x) = {0}, so that

p (x) = {0}. That is, (x) is contained in the zero set of p, which is a finite set.

Special case p(t ) = t 4 1: The zero set of p is {1, +1, i , +i }.

soln.tex,v 1.2

TMA4230 Functional Analysis, 20050531

Suggested solution

Problem 3
a) Any bounded self-adjoint operator T on a Hilbert space H can be written
Z
T = d E
R

where (E )R is the spectral family of T .


The spectral family satisfies these properties: Each E is a self-adjoint projection, E E if < ,
E = 0 for < m and E = I for M where m and M are the smallest and largest elements of (T ),
and E is the strong operator limit of E as + .
The integral is the limit in the norm topology of the RiemannStieltjes sums
n
X
k=1

k (E k E k1 )

where 0 < 1 < < n , 0 < m, n M , k1 k k for k = 1, . . . , n, and the limit is taken as the
norm maxk |k k1 | goes to zero.
Finally,
Z

f (T ) =

f () d E

for any continuous function f on (T ).


The latter definition requires values of f outside (T ), for example in any gaps in the spectrum. This is remedied by expanding the definition of f in a suitable way, for example by making it affine in each such gap. The
exact manner of extension is unimportant, in the limit.

b) It is known that E is constant in each subinterval of the complement of the spectrum. There are
n + 1 such intervals, yielding n + 1 different values of E . By the continuity from the right of the
spectral map, the value of each E k is the value of E in the interval to the right of k . Thus

< 1 ,

0
E =

k < k+1 ,

E
k

k = 1, . . . , n 1,

n .

If we write P 1 = E 1 and P k = (E k E k1 ) for k = 2, . . . , n the integral becomes


Z
n
X
T = d E =
k P k .
R

k=1

The following question was not supposed to be here. It had been replaced by a different question, but then
an early version was delivered to the exam office by mistake.
The students were told around 11:30, and told that they could skip this question and still get a full score.

c) The definition of U t becomes


Ut = e

i tT

e i t d E.

The adjoint becomes


U t

=e

i t T

e i t d E .

We compute the product by multiplying the integrands pointwise: e i t e i t = 1 and integrating:


R
R
R
U t U t = U tU t = R d E = I . Similarly U s U t = R e i s e i t d E = R e i (s+t ) d E = U s+t . Finally, to compute the derivative consider
Z i s
U s U t
e ei t
=
d E
st
st
R
and let s t . Then the integrand converges to the derivative i . Moreover the convergence is uniform, so we can take the limit inside the integral and get dU t /d t = i T U t . (The factor i was missing
from the problem statement.)
2

soln.tex,v 1.2

You might also like