Homework 5

Functional Analysis Fifth Homework
Tianyu Tao
April 14, 2015
1. Algebra of n n complex matrices:
Proof. (a). Let mT (x) be the minimal polynomial of T , we have by the Cayley-Hamilton
theorem that mT (T ) = 0 and it is the polynomial with the least degree having this
property, thus the dimension of A is equal to the degree of mT (call it n): the matrices
I, T, T 2 , . . . , T n1 is a basis for this algebra by the minmality of mT .
(b). Let , first, clearly is determined by its values on the generators of A,
namely I and A, necessarily (I) = 1, so we only need to determine what c := (T ) is.
Well, we know mT (T ) = 0 the zero matrix, so 0 = (mT (T )) = mT ((T )) = mT (c),
since the roots of the minimal polynomial are precisely the eigenvalues of T , let the
eigenvalues of T be 1 , . . . , l , then there should be l functionals i with i (T ) = i .
2. Local algebras:
Proof. (a). Suppose the Jordan Canonical form of T is
i 1
B1
..
.
.
i . .
, with Bi =
.
.
.. 1
..
i
Br
where we assumed there are r eigenvalues. For i = 1, 2, . . . , r, define Ai to be the
algebra generated by the ith Jordan block Bi and identity matrix (whose size and
position matches that of Bi ). Then A is the direct sum of the Ai s since T is the sum of
its Jordan blocks. Each Ai is maximal, since there is only one multiplicative functionals
on Ai , namely i (Bi ) = i by the discussion in problem 1.
(b). A matrix T is diagonalizable if and only if all of its Jordan blocks are trivial, i.e.
the blocks are of size 1 1. If this happens, the multiplicative functionals i defined
above is the trivial one, since in that case the Ai is simply the algebra generated by
1 1 identity matrix, so isomorphic to C.
On the other hand if all the Ai are isomorphic to C, then Ai does not have any nontrivial maximal ideals, then the corresponding Jordan block are all 1dimensional,
hence T becomes diagonalizable.
1
Homework 4
Functional Analysis Spring 2015
Tianyu Tao
3. Generic properties of matrices

Proof. Let us do this one by one:
a. By the second problem this question is just asking whether a generic marix is
diagonalizable. This is true, let O = {T Mn (C), T is diagonalizable}, we need
to show that O is open and dense in Mn (C). Since Mn (C) is a finite dimensional
Banach space, we may choose an appropriate norm on it to induce the topology,
in this case we use the max norm: kT k = max{|Tij |} if T = (Tij ), note however,
when talk about Banach algebras, the max norm is not sub-multiplicative, so in
some case we need to use equivalence of norms to convert result about max norm
into results about the usual operator norm.
Suppose T O, so T is diagonalizable, we want to show for some > 0, all
matrices S with kS T k < will be diagonalizable. We shall use the formula
det(A + B) = det(A) + O()
so that we have the det( S) = det( T + T S), write T S = B with
kBk < 1, then we see
det( S) = det( T ) + O(),
so that the characteristic polynomial of S is a small perturbation of T , by Rouches
theorem, a simple zero becomes a simple zero under a small perturbation.
Now to show O is dense, we pick any T O, consider the ball B (T ) = {S :
kS T k < } with > 0 given, we need to show B (T ) contains a diagonalizable
matrix. We may assume T is already in its canonical form, then it is easy to find
S B (T ) which is diagonalizable: just replace the Jordan block
i 1
.
i . .
..
i 1
1
.
i 2 . .
by
..
.
1
1
i
i l
where i are chosen small so that the sum of i is less than , the latter block has
a characteristic polynomial with distinct root, as desired.
b. This is true as well, let
V = {(T, S) : the only invariant subspace shared by T, S are trivial},
we need to show V is open and dense in Mn (C) Mn (C).
In part (a) we showed the set of diagonalizable matrix D Mn (C) is open and
dense, so DD is open and dense in Mn (C)Mn (C), so if we can show V (DD)
is open and dense in D D we will be done. In fact, we can even replace D by D0 ,
which is to be defined as the set of matrices having distinct eigenvalues in Mn (C)
2
April 14, 2015
Homework 4
Tianyu Tao
by a similar argument above, I will just use D to mean the collection of all such
matrices.
First, to show V (D D) is open in D D, assume T and S are diagonalizable
matrix which does not share a common non-trivial invariant subspace. From linear
algebra, we know the restriction of diagonalizable matrix on invariant subspaces
are still diagonalizable, so in particular the invariant subspace for a diagonalzable
matrix with distinct eigenvalue are just the eigenspaces and direct sums of them.
So if T and S does not share a non-trivial invariant subspace, it is equivalent to
say that all the eigenvectors and the direct sums of i with 1 < i < n eigenvectors
are all distinct, under a small perturbation of T and S, this property should be
preserved, I should fill in some details...
(although in the real n = 3 the case is pretty clear: fix two set of two distinct
vectors in R3 , they span two planes which only intersect in one straight line, and
if we perturb the two set of basis the planes they span is only perturbed a little,
so remain only intersect only in straight line...)
Similar idea is applied to deduce that V (D D) is also dense in D D: if
B (T ) B (S) is some open set in D D with (S, T ) D D, if T and S already
share no non-trivial invariant subspace, then by above we find other pairs which
does not share invariant subspace; if T and S share a common invariant subspace,
then for some i with 1 < i < n, the directed sum of some i eigenvectors of T and
S are the same, again, a small perturbation would make the direct sum become
unequal, thus finding some (T, S) V inside this open set in D D...
c. This is true, by part (b), the set of pairs (T, S) Mn (C) Mn (C) which share
no non-trivial invariant subspace are open and dense, by Burnsides theorem on
matrix algebra, which states the only irreducible algebra (subalgebra which has
no non-trivial invariant subspace) in Mn (C) is Mn (C) itself, all such pairs (T, S)
together with I must generate the whole algebra Mn (C).
4. Compactness of multiplication operator:

Proof. Suppose Tm is compact but m is not equal to 0 a.e. we put E = {m 6= 0}, by
assumption, (E) > 0. Then, consider H = L2 (E, |E ), this is a closed subspace of
X = L2 (, ), hence a Hilbert space as well.
Notice H is infinite dimensional: by assumption on the measure , since (E) > 0,
there is a measurable F1 E such that 0 < (F1 ) < (E), continue, since then
(E\F1 ) = (E) (F1 ) > 0, we extract F2 E\E1 with 0 < (F2 ) < (E\F1 ), then
we find F3 E\(F1 F2 ), and so on, then consider the functions Fn , by construction,
Fi are P
disjoint, and none of the Fi is 0 (a.e.), thus the Fn s are linearly independent:
if 0 =
an Fn (), plug in Fi will force ai = 0.
Since we assumed Tm is compact on X, its restriction on Tm |H : H X is also
compact. Because on E, m 6= 0, we have Tm is injective, also, Tm |H maps H onto H:
if h H, then put g = m1 h, clearly g H, so Tm |H(g) = mg = h. Therefore, by
3
April 14, 2015
Homework 4
Tianyu Tao
the open-mapping theorem, Tm |H have a bounded inverse on H, but then the identity
operator on H is also compact, hence the unit ball on H is compact, which shows H
is finite dimensional, which is a contradiction to our first observation, so necessarily
m = 0 a.e. for Tm to be compact.
5. Invertible elements in subalgebra and the whole algebra :
Proof. (i). Since D and are connected open set in C, by the maximal modulus
principle, supzD |f (z)| and supz |f (z)| occurs at the boundary of D and . If
f (z) A, then kf | kB = supz |f (z)| = supzD |f (z)| = supzD |f (z)| = kf kA s
shows the claim.
(ii). Consider the function f (z) = z, this is invertible in B, whose inverse is of course
g(z) = 1/z, note indeed g B, and f is certainly in A, so f A B , but
f
/ A , since the only candidate for inverse is 1/z, which is not in A: it is not a
holomorphic function on D.
6. Index of Fredholm operator :

Proof. (a) By standard ODE results, ker L is just the space for homogeneous solutions,
which has dimension n; for each g X0 , we can use the variation of parameter
formula to get a solution to the equation Lf = g for some f Xm , then it follows
L is actually surjective, so the codimension is 0, hence the index of L, which is
the difference of dim ker L and dim X0 /L(Xm ), is equal to m.
(b) To determine the range of L, we need to find all g X0 such that g = f 0 for
some f X1 . We know that any holomorphic function g on an open disk has a
primitive there, so anyway f exist locally, but this seems to depend on the domain
, so I am not sure...
Of course the kernel is the set of all functions f with f 0 = 0 and f can be extended
continuously to the boundary of , which are just constants.
April 14, 2015

Homework 5

Uploaded by

Homework 5

Uploaded by

Functional Analysis Fifth Homework

Functional Analysis Spring 2015

3. Generic properties of matrices

April 14, 2015

Functional Analysis Spring 2015

4. Compactness of multiplication operator:

April 14, 2015

Functional Analysis Spring 2015

6. Index of Fredholm operator :

April 14, 2015

You might also like