Spectral Theorem for compact self-adjoint operators
Philippe Jaming
Universit ´e de Bordeaux
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
Master Math ´ematiques et Applications
M1 Math ´ematiques fondamentales & M1 Analyse, EDP, probabilit ´es Lecture : Introduction to spectral analysis
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 1 / 13
Preliminary warning
This video is a complement to the lecture notes available at
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
1. Statement of the Spectral Theorem 2. An example
3. Toolbox
Preliminary warning
This video is a complement to the lecture notes available at
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
1. Statement of the Spectral Theorem 2. An example
3. Toolbox
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 2 / 13
Preliminary warning
This video is a complement to the lecture notes available at
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
1. Statement of the Spectral Theorem 2. An example
3. Toolbox
Preliminary warning
This video is a complement to the lecture notes available at
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
1. Statement of the Spectral Theorem 2. An example
3. Toolbox
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 2 / 13
Spectral Theorem for compact self-adjoint operators
Aim: give the main tools to prove the spectral theorem Theorem (Spectral Theorem)
H separable, infinite dimensional, Hilbert space. T : H → H compact, self-adjoint operator.
∃ (e k ) k∈ N orthonormal basis of H;
∃ (λ k ) k∈ N real, λ k → 0;
Tx = X
k∈ N
λ k hx , e k ie k .
Spectral Theorem for compact self-adjoint operators
Aim: give the main tools to prove the spectral theorem Theorem (Spectral Theorem)
H separable, infinite dimensional, Hilbert space. T : H → H compact, self-adjoint operator.
∃ (e k ) k∈ N orthonormal basis of H;
∃ (λ k ) k∈ N real, λ k → 0;
Tx = X
k∈ N
λ k hx , e k ie k .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 3 / 13
Spectral Theorem for compact self-adjoint operators
Aim: give the main tools to prove the spectral theorem Theorem (Spectral Theorem)
H separable, infinite dimensional, Hilbert space. T : H → H compact, self-adjoint operator.
∃ (e k ) k∈ N orthonormal basis of H;
∃ (λ k ) k∈ N real, λ k → 0;
Tx = X
k∈ N
λ k hx , e k ie k .
Spectral Theorem for compact self-adjoint operators
Aim: give the main tools to prove the spectral theorem Theorem (Spectral Theorem)
H separable, infinite dimensional, Hilbert space. T : H → H compact, self-adjoint operator.
∃ (e k ) k∈ N orthonormal basis of H;
∃ (λ k ) k∈ N real, λ k → 0;
Tx = X
k∈ N
λ k hx , e k ie k .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 3 / 13
Spectral Theorem for compact self-adjoint operators
Aim: give the main tools to prove the spectral theorem Theorem (Spectral Theorem)
H separable, infinite dimensional, Hilbert space. T : H → H compact, self-adjoint operator.
∃ (e k ) k∈ N orthonormal basis of H;
∃ (λ k ) k∈ N real, λ k → 0;
Tx = X
k∈ N
λ k hx , e k ie k .
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 4 / 13
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 4 / 13
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 4 / 13
Stability
Throughout H=Hilbert space, T : H → H bounded linear.
Observation
If T = T ∗ and E ⊂ H closed subspace invariant through T : T (E ) ⊂ E then E ⊥ is also stable through T
Proof.
x ∈ E , y ∈ E ⊥ then Tx ∈ E and
hTy , x i =
* y
|{z}
∈E
⊥, Tx
|{z}
∈E
+
= 0
that is Ty ∈ E ⊥ .
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 5 / 13
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 5 / 13
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 5 / 13
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Kernel and image
ker T is stable through T . Observation
If T = T ∗ , H = ker T ⊕ Im T , i.e. ker T = Im T ⊥ i.e. (ker T ) ⊥ = Im T . Proof.
x ∈ Im T ⊥ means ∀y ∈ H
0 = hx , Ty i = hTx , y i
⇔ Tx = 0 i.e. x ∈ ker T .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 5 / 13
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 6 / 13
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 6 / 13
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Reduction to one-to-one operators
π 0 , ⊥-proj on ker T , I − π 0 ⊥-proj on Im T then (with previous observation)
T = π 0 T π 0 + (I − π 0 )T (I − π 0 ) both are self adjoint and compact if T is.
(I − π 0 )T (I − π 0 ) is T : Im T → Im T and is one-to-one.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 6 / 13
Spectral theorem - comments
one can also interpret this as T having a matrix of the form
ker T E 1 E 2 · · ·
e 1,1 , . . . , e n
1,1 e 1,2 , . . . , e n
2,2
ker T 0 0 0 · · ·
e 1,1 .. . e n
1,1
0
λ 1 0
. ..
0 λ 1
0 e 1,2
.. . e n
2,2
0
λ 2 0
. ..
0 λ 2
0 0
. .. 0
. ..
0 . ..
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 8 / 13
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 8 / 13
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 8 / 13
Eigenvalues are real
Observation
If T = T ∗ , and λ eigenvalue: ∃x ∈ H, x 6= 0 s.t. Tx = λx then λ ∈ R . Proof.
λkx k 2 = hλx , x i = hTx , x i = hx, Txi
= hx, λx i = ¯ λkxk 2 .
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 9 / 13
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 9 / 13
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 9 / 13
Eigenspaces are orthogonal
Observation
If T = T ∗ , λ 6= µ, E λ = ker(T − λI), E µ corresponding eigenspaces then E λ ⊥ E µ
Proof.
x ∈ E λ , y ∈ E µ
λhx , y i = hλx , y i = hTx , y i = hx, Ty i
= hx , µy i = ¯ µhx , y i = µhx, y i
E µ ⊂ E λ ⊥ and E µ =eigenspace for eigenvalue µ of (I − π λ )T (I − π λ ) i.e.
of T restricted to E λ ⊥
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 10 / 13
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 10 / 13
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 10 / 13
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 10 / 13
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 10 / 13
Eigenvalues exist
Lemma
If T 6= 0 is compact, self-adjoint the either kT k or −kT k is an eigenvalue
Proof: α := kT k then
α 2 = sup{kTxk 2 : kxk = 1} = sup{hTx , Tx i : kxk = 1}
= sup{ D T 2 x , x
E
: kx k = 1}.
∃x n , kx n k = 1,
T 2 x n , x n
→ α 2 .
T 2 x n − α 2 x n
2
= T 2 x n
2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
≤ kT k 4 kx n k 2 − 2α 2 < D
T 2 x n , x n E
+ α 4 kx n k 4
= 2α 4 − 2α 2 < D
T 2 x n , x n
E → 0.
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is an eigenvalue;
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 11 / 13
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is
an eigenvalue;
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is an eigenvalue;
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 11 / 13
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is
an eigenvalue;
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is an eigenvalue;
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 11 / 13
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is
an eigenvalue;
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is an eigenvalue;
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 11 / 13
Eigenvalues exist - proof continued
Proof continued.
T compact ⇒ T 2 compact, x n ∈ B,
T 2 x n − α 2 x n
2 → 0
Go to subsequence (T 2 x n
k) converges x n
kalso. x the limit, kx k = 1 and
T 2 x − α 2 x
= 0 i.e. (T 2 − α 2 )x = 0.
Rewrite this as (T + αI)(T − αI)x = 0
Case 1: (T − αI)x = 0 that is Tx = αx , α is an eigenvalue.
Case 2: y := (T − αI)x 6= 0 then (T + αI)y = 0 i.e. Ty = −αy , −α is
an eigenvalue;
Eigenspaces have finite dimension
Observation
If T is compact, each eigenspace E λ (λ 6= 0) has finite dimension.
Proof.
(e k ) k∈I O.N.B of E λ , Te k = λe k . If k 6= `, kTe k − Te ` k = |λ| √ 2 6= 0.
Can not extract convergent subsequence (if I infinite) Contradicts T compact.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 12 / 13
Eigenspaces have finite dimension
Observation
If T is compact, each eigenspace E λ (λ 6= 0) has finite dimension.
Proof.
(e k ) k∈I O.N.B of E λ , Te k = λe k . If k 6= `, kTe k − Te ` k = |λ| √ 2 6= 0.
Can not extract convergent subsequence (if I infinite)
Contradicts T compact.
Eigenspaces have finite dimension
Observation
If T is compact, each eigenspace E λ (λ 6= 0) has finite dimension.
Proof.
(e k ) k∈I O.N.B of E λ , Te k = λe k . If k 6= `, kTe k − Te ` k = |λ| √ 2 6= 0.
Can not extract convergent subsequence (if I infinite) Contradicts T compact.
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 12 / 13
Eigenspaces have finite dimension
Observation
If T is compact, each eigenspace E λ (λ 6= 0) has finite dimension.
Proof.
(e k ) k∈I O.N.B of E λ , Te k = λe k . If k 6= `, kTe k − Te ` k = |λ| √ 2 6= 0.
Can not extract convergent subsequence (if I infinite)
Contradicts T compact.
That’s all!
Thank you for your attention!
Next video: The proof.
http://www.u-bordeaux.fr/˜ pjaming/enseignement/M1.html
Philippe Jaming (Universit ´e de Bordeaux) Spectral Theorem 2 Master Math & Applications 13 / 13