Spectral Analysis of Linear Operators (Continued)
SMA 5878 Functional Analysis II
Alexandre Nolasco de Carvalho
Departamento de MatematicaInstituto de Ciencias Matematicas and de Computacao
Universidade de Sao Paulo
March 18, 2019
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Note that, for ξ0 ∈ ρ(A), if |ξ − ξ0| < ‖(ξ0 − A)−1‖−1L(X ) we have
that ξ ∈ ρ(A) and
(ξ − A)−1 =
∞∑
n=0
(ξ0 − ξ)n(ξ0 − A)−n−1. (1)
On the other hand, if |ξ − ξ0| > ‖(ξ0 − A)‖L(X ) we have thatξ ∈ ρ(A) and
(ξ − A)−1 = −
∞∑
n=0
(ξ0 − ξ)−n−1(ξ0 − A)n (2)
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Hence, the radius of convergence of the Taylor’s series in (1) is thereciprocal of the spectral radius of the operator (ξ0 − A)−1 whilethe radius of convergence of the Laurent’s series in (2) is thespectral radius of (ξ0 − A). Thus, in the circles{λ ∈ C : |λ− ξ0| = (rσ(ξ0 − A)−1)−1} and{λ ∈ C : |λ− ξ0| = rσ((ξ0 − A))} there are points of σ(A).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
O primeiro teorema da aplicacao espectral
Next we show a version of the Spectral Mapping Theorem forpolynomials.
Seja p(λ) = anλn + an−1λ
n−1 + · · ·+ a1λ+ a0, ai ∈ C, 0 ≤ i ≤ n.
If A ∈ L(X ), we define
p(A) = anAn + an−1A
n−1 + · · ·+ a1A+ a0I
and, if B ⊂ C, we define p(B) := {p(b) : b ∈ B}.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
TheoremIf A ∈ L(X ) and p : C → C is a polynomial, then
i) σp(p(A)) = p(σp(A)),
ii) σr (p(A)) = p(σr (A))\p(σp(A)),
iii) σc(p(A)) = p(σc(A))\(p(σp(A)) ∪ p(σr (A))) and
iv) σ(p(A)) = p(σ(A)).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof: Let p(λ) = anλn + an−1λ
n−1 + · · · + a1λ+ a0, ai ∈ C,0 ≤ i ≤ n be a polinomial.
It is easy to se that for all scalar α ∈ C, and T ∈ L(X ),σ(αT ) = {αλ ∈ C : λ ∈ σ(T )} =: ασ(T ).
Hence, without loss of generality, we may assume that an = (−1)n.
If β1, · · · , βn are the roots of the polynomial q(λ) = µ− p(λ), then
µ− p(A) = q(A) = (β1 − A) · · · (βn − A). (3)
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
The proof of iv) follows immediately from i), ii) and iii).
i) If µ− p(A) is not injective, it follows from (3) that there is i0with 1 ≤ i0 ≤ n such that (βi0 − A) is not injective.
Reciprocally, if for some i0 with 1 ≤ i0 ≤ n, (βi0 − A) is notinjective, it follows from (3) that µ− p(A) is not injective. Thisshows that σp(p(A)) = p(σp(A)).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
ii) If µ ∈ σr (p(A)), µ− p(A) is injective, from (3), (βi − A) isinjective for all 1 ≤ i ≤ n.
Besides that, R(µ− p(A)) is not dense and consequently, for some1 ≤ i0 ≤ n we must have that R(βi0 − A) is not dense. It followsthat βi0 ∈ σr (A) and p(βi0) = µ.
This shows that σr (p(A)) ⊂ p(σr (A))\p(σp(A)).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
On the other hand, if µ ∈ p(σr (A))\p(σp(A)), it follows that(βi−A) is injective for all 1≤ i≤n (since σp(p(A)) = p(σp(A))and, for some 1 ≤ i0 ≤ n, R(βi0 − A) is not dense.
From that it follows that µ− p(A) is injective but R(µ− p(A)) isnot dense and µ ∈ σr (p(A)).
This shows that p(σr (A))\p(σp(A)) ⊂ σr (p(A)) and the proof ofii) is complete.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
iii) If µ ∈ σc(p(A)) then µ /∈ σp(p(A)) = p(σp(A)) and it followsthat (βi − A) is injective, 1 ≤ i ≤ n.
Also,µ /∈σr (p(A))=p(σr (A))\p(σp(A))andR(βi−A)=X , 1≤ i≤n.
Since µ /∈ρ(A) it follows that R(βi0 − A)(X , for some 1 ≤ i0 ≤ n.
If µ ∈ p(σc(A))\p(σp(A)) ∪ p(σr (A)) then µ = p(βi ) withβi /∈ σp(A) ∪ σr (A), 1 ≤ i ≤ n.
Hence (βi − A) is one-to-one and has dense image, 1 ≤ i ≤ n.
Since βi0 ∈ σc(A), for some 1 ≤ i0 ≤ n, we must have thatR(βi0 − A) ( X and (µ− p(A)) is one-to-one, has dense imageand R(µ− p(A)) ( X . This proves that µ ∈ σc(p(A)).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Example
Let X = ℓ2(C), T : ℓ2(C) 7→ ℓ2(C) the linear operator defined by
T ({x1, x2, x3, · · · }) = {x1, 0, x2, x3, . . .} and p(λ) = λ2 − λ. It iseasy to see that 0 ∈ σr (T ) and therefore p(0) = 0 ∈ p(σr (T )). Onthe other hand, we see that p(0) = 0 /∈ σr (p(T )), since
p(T )({x1, x2, x3, · · · }) = ({0, 0,−x2, x2 − x3, x3 − x4, · · · })
and p(T ) is not injective operator.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Next we give an alternative proof of
rσ(A) = limn→∞
‖An‖1/nL(X ) = inf
n≥1‖An‖
1/nL(X )
using the previous theorem.
In fact, σ(An) = {zn : z ∈ σ(A)} and rσ(A)n = rσ(A
n) ≤ ‖An‖L(X )
and rσ(A) ≤ ‖An‖1/nL(X ). Hence,
rσ(A) = lim supn→∞
‖An‖1/nL(X ) ≤ inf
n∈N‖An‖
1/nL(X ) ≤ lim inf
n→∞‖An‖
1/nL(X ).
So, the limit exists and rσ(A) = infn∈N ‖An‖1/nL(X ).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
DefinitionLet X be a Banach space over C and A ∈ L(X ). We say that A is
nilpotent if there is a n0 ∈ N such that An0 = 0 and that A is
quasi-nilpotent if ‖An‖1n
L(X )
n→∞−→ 0.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Example
Let T :ℓ1(C)→ℓ1(C) defined by
T (x1, x2, x3, · · · )=(0, x1,x2
2,x3
3,x4
4, · · · ).
It is easy to see that ‖T n‖L(ℓ1) ≤1n! and therefore T is
quasinilpotent and σ(T ) = {0}.
ExerciseLet X be a Banach space over C, A ∈ L(X ) a given nilpotent
operator and 0 6= λ ∈ C, compute (λ− A)−1.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Dual operators
Next we recall the definition of dual operator. Let X and Y beBanach spaces over the field K (R or C) with duals X ∗ and Y ∗.
If Z is either X or Y , z∗ ∈ Z ∗ we denote its values at a vectorz ∈ Z by 〈z , z∗〉Z ,Z∗ .
The map Z ×Z ∗ ∋ (z , z∗) 7→ 〈z , z∗〉Z ,Z∗ ∈ K is called duality map.
For simplicity of notation, if no confusion may arise, we will write〈z , z∗〉 instead of 〈z , z∗〉Z ,Z∗ .
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Let A : D(A) ⊂ X → Y be a densely defined operator.
The dual operator A∗ : D(A∗) ⊂ Y ∗ → X ∗ of A is the linearoperator defined by: D(A∗) is the set of all y∗ ∈ Y ∗ such thaty∗ ◦ A : D(A) ⊂ X → K is bounded; that is, the set of all y∗ ∈ Y ∗
for which there exists w∗ ∈ X ∗ satisfying
〈Ax , y∗〉 =(y∗◦A)(x) = w∗(x) = 〈x ,w∗〉, ∀ x ∈ D(A). (4)
If y∗ ∈ D(A∗) we define A∗y∗ := w∗ where w∗ is the uniqueelement of X ∗ satisfying (4).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
ExerciseIf X is a Banach space and A : D(A) ⊂ X → Y is a densely
defined linear operator, show that A∗ : D(A∗) ⊂ Y ∗ → X ∗ is a
closed linear operator.
We start with some basic results about dual operators.
LemmaLet X and Y be Banach spaces over K and A ∈ L(X ,Y ); then,A∗ ∈ L(Y ∗,X ∗) and ‖A‖L(X ,Y ) = ‖A∗‖L(Y ∗,X∗).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof: For all y∗ ∈ Y ∗, y∗◦A is a bounded linear functional andtherefore it determines a unique element x∗ ∈ X ∗ for which〈x , x∗〉 = 〈Ax , y∗〉, for all x ∈ X . It follows that D(A∗) = Y ∗.
Also,
‖A∗‖L(Y ∗,X∗) = sup‖y∗‖Y ∗≤1
‖A∗y∗‖X∗ = sup‖y∗‖Y ∗≤1
sup‖x‖X≤1
|〈x ,A∗y∗〉|
= sup‖x‖X≤1
sup‖y∗‖X∗≤1
|〈Ax , y∗〉| = sup‖x‖X≤1
‖Ax‖Y
= ‖A‖L(X ,Y ).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
LemmaLet X be a reflexive Banach space K. If A : D(A) ⊂ X → X is
closed and densely defined then D(A∗) is dense in X ∗.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof: If D(A∗) is not dense in X ∗ there exists x0 ∈ X such thatx0 6= 0 and 〈x0, x
∗〉 = 0 for all x∗ ∈ D(A∗).
Since the graph of A is closed and does not contains (0, x0), fromHahn-Banach Theorem, there are x∗1 and x∗2 in X ∗ such that〈x , x∗1 〉 − 〈Ax , x∗2 〉 = 0 for all x ∈ D(A) and 〈0, x∗1 〉 − 〈x0, x
∗2 〉 6= 0.
It follows that x∗2 6= 0, 〈x0, x∗2 〉 6= 0, x∗2 ∈ D(A∗) and A∗x∗2 = x∗1 .
This implies that 〈x0, x∗2 〉 = 0 which is a contradiction. Thus
D(A∗) is dense in X ∗.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
ExerciseExhibit an example of a densely defined operator,
A : D(A) ⊂ X → X such that D(A∗) ( X.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
ExerciseThe annihilator of a subset M ⊂ X is the set
M⊥ = {x∗ ∈ X ∗ : 〈x , x∗〉 = 0,∀x ∈ M} and the annihilator of
M∗ ⊂ X ∗ is the set (M∗)⊥ = {x ∈ X : 〈x , x∗〉 = 0,∀x∗ ∈ M∗}.We know that if M ⊂ X is a vector space then (M⊥)⊥ = M.
A subset M∗ ⊂ X ∗ is said total if (M∗)⊥ = {0}. Show that, if
A : D(A) ⊂ X → X is a closed densely defined operator then,
D(A∗) is total.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
TheoremLet A : D(A) ⊂ X → X be a linear densely defined operator. Then
ρ(A) = ρ(A∗) and ((λ− A)−1)∗ = (λ− A∗)−1,∀λ ∈ ρ(A)
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof: From the definition of dual (λI − A)∗ = λI ∗ − A∗. If λ− A
is injective and has dense image, let us show that
(1) ((λI − A)−1)∗(λI ∗ − A∗)x∗ = x∗, ∀x∗ ∈ D(A∗) and
(2) (λI ∗ − A∗)((λI − A)−1)∗x∗ = x∗, ∀x∗ ∈ D(((λI − A)−1)∗).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof of (1): If x ∈ R(λ− A), x∗ ∈ D(A∗), then
〈x , x∗〉 = 〈(λI −A)(λI −A)−1x , x∗〉 = 〈(λI −A)−1x , (λI ∗−A∗)x∗〉.
It follows that (λI ∗ − A∗)x∗ ∈ D(((λI − A)−1)∗)(R(λI ∗ − A∗) ⊂ D(((λI − A)−1)∗)) and, from the fact thatR(λI − A) = X , we have
((λI − A)−1)∗(λI ∗ − A∗)x∗ = x∗, ∀x∗ ∈ D(A∗).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Proof of (2): If x∗ ∈ D(((λI − A)−1)∗) and x ∈ D(A), then
〈x , x∗〉 = 〈(λI−A)−1(λI−A)x , x∗〉 = 〈(λI−A)x , ((λI−A)−1)∗x∗〉.
Hence ((λI − A)−1)∗x∗ ∈ D(λI ∗ − A∗) and, from the fact thatD(A) = X , we have
(λI ∗ − A∗)((λI − A)−1)∗x∗ = x∗, ∀x∗ ∈ D(((λI − A)−1)∗).
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Now we can complete the proof of the theorem. If λ ∈ ρ(A),(λI − A)−1 is bounded and we have that ((λI − A)−1)∗ ∈ L(X ∗).
From (1) and (2) it follows that (λI ∗ − A∗)−1 = ((λI − A)−1)∗
and λ ∈ ρ(A∗).
If λ ∈ ρ(A∗), note that A∗ is closed and, consequently,(λI ∗ − A∗)−1 ∈ L(X ∗).
We already know that λI − A is densely defined.
Let us show that λI − A is injective and has dense image.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
To see that λI − A is injective note that, if x ∈ D(A) is such that(λ− A)x = 0 and x∗ ∈ D(A∗), then
0 = 〈(λI − A)x , x∗〉 = 〈x , (λI ∗ − A)∗x∗〉.
Since R(λI ∗ − A∗) = X ∗ we have that x = 0 and therefore λI − A
is injective.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
Now, to see that λI − A has dense image note that, if x∗ ∈ X ∗ issuch that 0 = 〈(λI − A)x , x∗〉 for all x ∈ D(A), then x∗ ∈ D(A∗)and 0 = 〈x , (λI − A)∗x∗〉 for all x ∈ D(A).
Since D(A) is dense in X , it follows that (λI − A)∗x∗ = 0 and,since λ ∈ ρ(A∗), it follows that x∗ = 0. With this we have provedthat R(λI − A) is dense in X .
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II
Spectral Analysis of Linear Operators (Continued)Bounded Linear OperatorsDual operators
To conclude that λ ∈ ρ(A), it remains to prove that (λI − A)−1 isbounded. If x∗ ∈ X ∗ = R(λI ∗ − A∗) ⊂ D(((λI − A)−1)∗) andx ∈ R(λI − A), from (1) and (2), we have that
|〈(λI − A)−1x , x∗〉| = |〈x , ((λI − A)−1)∗x∗〉| = |〈x , (λI ∗ − A∗)−1x∗〉|
≤ ‖(λI ∗ − A∗)−1‖ ‖x∗‖ ‖x‖
From this it follows that (λ− A)−1 is bounded, proving thatλ ∈ ρ(A) and completing the proof.
Alexandre Nolasco de Carvalho ICMC - USP SMA 5878 Functional Analysis II