0% found this document useful (0 votes)
14 views

B4 2Lecture14P

This document provides an overview of spectral theory for bounded linear operators on Banach and Hilbert spaces. It defines the spectrum and different parts of the spectrum. It also discusses properties of the spectrum for normal operators.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
14 views

B4 2Lecture14P

This document provides an overview of spectral theory for bounded linear operators on Banach and Hilbert spaces. It defines the spectrum and different parts of the spectrum. It also discusses properties of the spectrum for normal operators.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 20

B4.

2 Functional Analysis II
Lecture 14

Luc Nguyen
luc.nguyen@maths

University of Oxford

HT 2021

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 1 / 19


In the last 3 lectures

Convergence/Divergence of Fourier series

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 2 / 19


In this lecture

The ABCs of spectral theory for bounded linear operators.


Spectra of normal operators.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 3 / 19


Basic definitions
Definition
Let X be a complex Banach space and T ∈ B(X ).
The spectrum σ(T ) of T is the set of complex numbers λ such
that λI − T has no inverse in B(X ).
The resolvent set ρ(T ) of T is the complement of σ(T ) in C.
If λ ∈ ρ(T ), then Rλ (T ) = (λI − T )−1 is called the resolvent of
T at λ.
Facts from FA1:
σ(T ) is closed, non-empty and contained in {|λ| ≤ kT k}.
Gelfand’s formula: The spectral radius is

rad(σ(T )) := sup |λ| = lim kT n k1/n = inf kT n k1/n .


λ∈σ(T ) n→∞ n

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 4 / 19


Refined definitions

Definition
Point spectrum: λ ∈ σp (T ) if λI − T is not injective.
λ is called an eigenvalue of T and the non-trivial elements of
Ker(λI − T ) are called the eigenvectors of T .
Residual spectrum: λ ∈ σr (T ) if λI − T is injective and its
range Im(λI − T ) is not dense in X .
Continuous spectrum: λ ∈ σc (T ) if λI − T is injective and its
range Im(λI − T ) is a proper dense subset of X .
Approximate point spectrum: λ ∈ σap (T ) if there is a sequence
(xn ) ⊂ X such that kxn k = 1 and kTxn − λxn k → 0.
λ is called an approximate eigenvalue of T .

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 5 / 19


A schematic diagram

σc (T )

σp (T )

σr (T )

The spectrum of a bounded linear operator.

σ(T ) is split as σp (T ) ∪ σc (T ) ∪ σr (T ).
σp (T ), σc (T ), and σr (T ) are mutually disjoint.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 6 / 19


A schematic diagram

σc (T )

σp (T )

σr (T )

σap (T )

The spectrum of a bounded linear operator.

σ(T ) is split as σp (T ) ∪ σc (T ) ∪ σr (T ).
σp (T ), σc (T ), and σr (T ) are mutually disjoint.
σp (T ) is a subset of σap (T ).
We will now see that σc (T ) is also a subset of σap (T ).

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 6 / 19


σc (T ) ⊂ σap (T )
Lemma
Let T ∈ B(X ) be a bounded linear operator on a Banach space X .
Then σc (T ) ⊂ σap (T ).

Proof
Suppose by contradiction that there exists λ ∈ σc (T ) \ σap (T ).
λ ∈ σc (T ) means that λI − T is injective and its range
Y = Im(λI − T ) is dense in X . In particular, λI − T considered
as a map from X into Y is bijective and has an inverse, say
U : Y → X.
λ∈/ σap (T ) means that there is some c > 0 such that
k(λI − T )xk ≥ c for all x ∈ X , kxk = 1.
Equivalently, this means k(λI − T )xk ≥ ckxk for all x ∈ X .
Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 7 / 19
σc (T ) ⊂ σap (T )

λ ∈ σc (T ) \ σap (T ).
Y = Im(λI − T ) is dense in X and λI − T considered as a map
from X into Y has an inverse U : Y → X .
k(λI − T )xk ≥ ckxk for all x ∈ X .
This means that ky k ≥ ckUy k for all y ∈ Y , i.e. U is bounded.
Since Y is dense in X , U extends to Ū ∈ B(X ).
We are now in position to deduce a contradiction: If p ∈ X \ Y
and pn ∈ Y such that pn → p, then Upn → Ūp and so

(λI − T )Ūp = lim (λI − T )Upn = lim pn = p.


n→∞ n→∞

This shows that p belongs to Y , a contradiction.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 8 / 19


σr (T ) ⊂ σp (T 0)
Lemma
Let T ∈ B(X ) be a bounded linear operator on a Banach space X .
Then σr (T ) ⊂ σp (T 0 ) where T 0 ∈ B(X ∗ ) is the dual of T .

Proof
Suppose λ ∈ σr (T ) so that Y = Im(λI − T ) is a proper
non-dense subspace of X .
By the Hahn-Banach theorem, there is an element ` ∈ X ∗ ,
k`k∗ = 1 such that `(Ȳ ) = 0.
Now, for all x ∈ X we have

(λI − T 0 )` (x) = ` (λI − T )x = 0.


 

This means (λI − T 0 )` = 0, i.e. λ is an eigenvalue of T 0 .


Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 9 / 19
Example
Example
Let X be a complex Banach space and T ∈ B(X ). Show that if λ is
on the (topological) boundary of σ(T ), then λ ∈ σap (T ).

Suppose by contradiction that λ ∈


/ σap (T ). As seen earlier, this
means that there is some c > 0 such that

k(λI − T )xk ≥ ckxk for all x ∈ X .

Since λ is on the boundary of σ(T ), there exists λn ∈ ρ(T ) such


that λn → λ. By the above, we have for large n that
1
k(λn I − T )xk ≥ ckxk for all x ∈ X .
2
2
This gives kRλn (T )k ≤ c
for large n.
Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 10 / 19
Examples
2
kRλn (T )k ≤ c
for large n.
Now
 
λI − T = (λn I − T ) + (λ − λn )I = (λn I − T ) I + (λ − λn )Rλn (T ) .
| {z }
=:Sn

By the above estimate, kSn k < 1 for large n. This implies that
I + Sn and hence λI − T are invertible, contradicting the fact
that λ ∈ σ(T ).

Remark
It can also be shown that if λ is on the boundary of σ(T ), then
Im (λI − T ) 6= X .

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 11 / 19


Hilbert settings
Let X be a (complex) Hilbert space and T ∈ B(X ).
Note: (λI − T )∗ = λ̄I − T ∗ .
T T∗

λ ∈ σp (T ) Ker(λI − T ) 6= 0 Im(λ̄I − T ∗ ) isn’t dense


(⇒ λ̄ ∈ σp (T ∗ ) ∪ σr (T ∗ ))

λ ∈ σc (T ) Ker(λI − T ) = 0, Ker(λ̄I − T ∗ ) = 0,
⇔ λ̄ ∈ σc (T ∗ ) Im(λI − T ) is dense Im(λ̄I − T ∗ ) is dense

λ ∈ σr (T ) Ker(λI − T ) = 0, Ker(λ̄I − T ∗ ) 6= 0
Im(λI − T ) isn’t dense (⇒ λ̄ ∈ σp (T ∗ ))
Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 12 / 19
σ(T ) = σap (T ) ∪ σp0 (T ∗) = σ 0(T ∗)

In particular, we have
Theorem
Let X be a complex Hilbert space and T ∈ B(X ). Then

σ(T ) = σap (T ) ∪ σp (T 0 ) = σap (T ) ∪ σp0 (T ∗ ) = σ 0 (T ∗ )

where σp0 (T ∗ ) = {λ : λ̄ ∈ σp (T ∗ )} and σ 0 (T ∗ ) = {λ : λ̄ ∈ σ(T ∗ )}.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 13 / 19


Example
Example
Let X be a (complex) Hilbert space and T ∈ B(X ). Show that if
λ ∈ σ(T ), then there exists (xn ) ⊂ X , kxn k = 1 such that
hTxn , xn i → λ. In other words,
σ(T ) ⊂ {hTx, xi : kxk = 1} ⊂ C.

If λ ∈ σap (T ), then we can select (xn ) with kxn k = 1 such that


(λI − T )xn → 0. By Cauchy-Schwarz’ inequality, this implies
λ − hTxn , xn i = hλxn − Txn , xn i → 0
and so hTxn , xn i → λ.
/ σap (T ), then λ̄ ∈ σp (T ∗ ) and so there is an x with
If λ ∈
kxk = 1 and T ∗ x = λ̄x. This implies with xn = x that
hTxn , xn i = hTx, xi = hx, T ∗ xi = hx, λ̄xi = λ.
Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 14 / 19
Normal operators

Definition
Let X be a complex Hilbert space. An operator T ∈ B(X ) is called
normal if TT ∗ = T ∗ T .

Proposition
Let X be a complex Hilbert space and T ∈ B(X ). T is normal if
and only if kTxk = kT ∗ xk for all x ∈ X .

Proof
(⇒) This is straightforward:

kTxk2 = hTx, Txi = hx, T ∗ Txi = hx, TT ∗ xi = hT ∗ x, T ∗ xi = kT ∗ xk2 .

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 15 / 19


Normal operators

Proof
(⇐) Suppose that kTxk = kT ∗ xk for all x ∈ X .
By polarisation, we have hTx, Ty i = hT ∗ x, T ∗ y i for all x, y ∈ X .
Now we reverse the argument we did above

hx, T ∗ Ty i = hTx, Ty i = hT ∗ x, T ∗ y i = hx, TT ∗ y i.

This implies that T ∗ T = TT ∗ .

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 16 / 19


Normal operators

Corollary
Let X be a complex Hilbert space. If T ∈ B(X ) is normal then
Ker T = Ker T ∗ .
Proof: This is because kTxk = kT ∗ xk.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 17 / 19


Spectra of normal operators

Corollary
Let X be a complex Hilbert space. If T ∈ B(X ) is normal then
σr (T ) = ∅ and σ(T ) = σap (T ).

Proof
It suffices to show that σr (T ) is empty.
We knew that σr (T ) ⊂ σp (T 0 ) = σp0 (T ∗ ).
Now, note that λI − T is also normal. By the previous corollary,
we have that λ̄I − T ∗ is injective if and only if λI − T is
injective, i.e. σp0 (T ∗ ) = σp (T ).
So σr (T ) ⊂ σp (T ). As these two sets are disjoint, this is
possible only if σr (T ) is empty.

Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 18 / 19


Orthogonality of eigenvectors of normal operators
Proposition
Let X be a complex Hilbert space and T ∈ B(X ) be normal. If x
and y are eigenvectors of T corresponding to different eigenvalues,
then hx, y i = 0.

Proof
Suppose Tx = λx and Ty = λ̃y . We have
λhx, y i = hλx, y i = hTx, y i = hx, T ∗ y i.
As λ̃I − T is normal, we have 0 = k(λ̃I − T )y k = k(λ̃I ¯ − T ∗ )y k
¯ .
and so T ∗ y = λ̃y
It follows that
¯ i = λ̃hx, y i.
λhx, y i = hx, T ∗ y i = hx, λ̃y
Since λ 6= λ̃, this implies hx, y i = 0.
Luc Nguyen (University of Oxford) B4.2 FA II – Lecture 14 HT 2021 19 / 19

You might also like