Université Denis Diderot 2012-2013
Mathématiques M1 Spectral Theory
Corrigé de l’examen duu 16 janvier 2013 Durée 3 heures. Documents interdits.
1. Let H be complex Hilbert space, A ∈ L(H) and P ∈ C[X].
(a) Recall the relation seen in the course which exists between Sp(P (A)) and SpA for a self-adjoint operator A. Show that the same result holds for an arbitrary bounded operator.
Answer — The relation is
Sp(P (A)) = {P (λ); λ ∈ SpA}.
It is proved by showing the inclusions Sp(P (A)) ⊂ {P (λ); λ ∈ SpA} and Sp(P(A)) ⊃ {P (λ); λ ∈ SpA}. For the first inclusion one takes any µ ∈ Sp(P (A)), decomposes P (X) − µ = a(X − λ
1) · · · (X − λ
n) and apply this identity by replacing X by A to conclude that µ = P (λ
j), for some λ
j∈ SpA. For the second inclusion we take any λ ∈ SpA and decompose the polynomial P(X) − P (λ) = (X − λ)Q(X) and use the same reasoning to conclude that P (λ) ∈ Sp(P(A)).
In both reasoning the same kinds of arguments are used, which do not rely on the fact that A is self-adjoint. Hence the result actually extend to any bounded operator.
(b) Prove that if P (A) = 0, then SpA is contained in the set of roots of P .
Answer — If P(A) = 0 we deduce from the previous question that {P (λ); λ ∈ SpA} = Sp(P(A)) = {0}. Hence any λ in SpA is contained in the set of roots of P . 2. Let H = L
2( R , λ) be the Hilbert space of complex-valued square integrable functions
with respect to the Lebesgue measure. Let f : R → C be a measurable and λ-locally square integrable function on R (i.e. R
K
|f|
2dλ < ∞ for all K ⊂ R compact). Define the unbounded operator M on H :
D(M ) = {ξ ∈ H : Z
|f ξ|
2dλ < ∞} (M ξ)(x) = f (x)ξ(x) ξ ∈ D(M), x ∈ R . (a) Show that M is densely defined.
Answer — It suffices to show that D(M ) contains a dense subset of H, for instance the space L
∞c(R, C) of essentially bounded measurable functions with compact sup- port. Indeed if ξ ∈ L
∞c( R , C ), then supp ξ ⊂ K, a compact subset of R and there exists some M > 0 s.t. |ξ(x)| ≤ M , for a.e. x ∈ R . Hence R
|f ξ|
2dλ ≤ M
2R
K
|f |
2dλ < +∞,
hence ξ ∈ D(M). Thus L
∞c(R, C) ⊂ D(M ). The fact that L
∞c(R, C) is dense in
H is a consequence of the fact that it contains the space C
c( R , C ) of continuous
functions with compact support, which is known to be dense in H. Alternatively
one may prove directly that L
∞c(R, C) is dense in H by proving that any function
ξ ∈ H is approximated by the sequence (ξ
n)
nwith values in L
∞c( R , C ) defined by
ξ
n(x) := 1
[−n,n](x) inf(|ξ(x)|, n)
ξ(x)nand by using Lebesgue’s dominated convergence
theorem.
(b) Let T be the unbounded operator on H defined by
D(T ) = D(M) (T ξ)(x) = f (x)ξ(x) ξ ∈ D(T ), x ∈ R . Show that T is densely defined and M = T
∗. Deduce that M is closed.
Answer — First D(T) = D(M ) is dense by the result of the previous question. The domain of T
∗is the subspace of ζ ∈ H such that the linear form
D(T ) −→ C ξ 7−→ R
R
ζ(x)f(x)ξ(x)dλ(x)
admits a continuous linear extension to H. This linear form is uniquely defined be- cause D(T ) is dense in H. By the Riesz–Fréchet theorem, this property is equiva- lent to the fact that [x 7−→ f(x)ζ (x)] belongs to L
2( R , C ). This latter condition is equivalent to ζ ∈ D(M ) and, if so, this linear map coincides with T . Hence D(T
∗) = D(T ) = D(M ) and T
∗= M . It follows that M is closed since it coin- cides with the adjoint operator of an operator defined on a dense domain.
(c) Compute M
∗.
Answer — By changing f 7−→ f , we change T to M . Hence by applying the previous results, we deduce that M
∗= T .
(d) Show that sp(M ) = EssIm(f ) where
EssIm(f ) = {λ ∈ C : ∀ > 0 λ(f
−1(B (λ, ))) > 0}.
Answer — We first show that EssIm(f ) ⊂ Sp(M ). Let λ ∈ EssIm(f ). Then for any n ∈ N
∗, λ[f
−1(B(λ, 1/n))] > 0. Let A
nbe a compact subset of f
−1(B(λ, 1/n)) such that λ(A
n) > 0. Set ξ
n:= 1
An/ p
λ(A
n), then R
|ξ
n|
2dλ = λ(A
n)
−1R
An
dλ = 1.
Moreover, ∀x ∈ A
n, |f(x) − λ| <
1n, so f |
Anis bounded by |λ| + 1/n. Hence T ξ
n, defined by (T ξ
n)(x) = f(x)ξ
n(x), is in L
2(R, C), so ξ
n∈ D(T ). Now
k(T − λ)ξ
nk
2H= Z
R
|f (x) − λ|
2|ξ
n(x)|
2dλ ≤ 1 n
2Z
An
|ξ
n(x)|
2dλ = 1 n
2.
Since kξ
nk
2H= 1, this shows that T − λ does not have a bounded inverse. Hence λ ∈ Sp(T ).
Let us now show that ( C \ EssIm(f)) ⊂ ( C \ Sp(T )), i.e. that, for any λ which does not belong to EssIm(f), T − λ is invertible. Indeed if λ 6∈ EssIm(f ), then ∃ε > 0 such that λ[f
−1(B(λ, ε))] = 0, i.e. λ({x; |f (x) − λ| < ε}) = 0. Hence, for a.e. x ∈ R ,
1
|f(x)−λ|
≤
1ε. This proves that the operator ξ 7−→
f−λξis bounded, but this operator is the inverse of T − λ. Hence λ 6∈ Sp(T ).
(e) Show that if f is continuous then sp(M ) = f ( R ).
Answer — Because of the previous question, it suffices to show that, if f is conti- nuous, EssIm(f )) = f (R). On the one hand, if λ ∈ EssIm(f )), then, ∀ε > 0, f
−1(B(λ, ε)) has a non vanishing measure, thus is non empty, i.e. f ( R ) ∩ B(λ, ε) 6= ∅.
Hence λ ∈ f ( R ) (note that, here, we do not need f to be continuous). On the other
hand, if λ ∈ f (R), then ∀ε > 0, B (λ, ε) ∩ f (R) 6= ∅, i.e. f
−1(B(λ, ε)) 6= ∅. Since f is
continuous f
−1(B (λ, ε)) is also open, hence it has a non vanishing measure.
(f) Let A ⊂ R be a Borel subset such that λ(A) < ∞, ν be the finite measure on R defined by ν(B) = λ(A ∩ B) for all B Borel subset of R . Observe that the integral with respect to ν is
Z
gdν = Z
A
gdλ.
Let f
∗(ν) the measure image of ν by f i.e., f
∗(ν)(B) = ν (f
−1(B)) for all Borel subset B ⊂ C . Let ξ = 1
A∈ H. Suppose that f is bounded and continuous. Show that M is bounded and the spectral measure µ
ξof M associated to ξ is µ
ξ= f
∗(ν).
Answer — The fact that M is bounded follows from the fact that f is bounded, i.e.
there exists M > 0 such that |f (x)| ≤ M , ∀x ∈ R . The spectral measure µ
ξis defined by : ∀ϕ ∈ C( R , C ) (or for any Borelian function)
Z
ϕdµ
ξ= hξ, ϕ(M )ξi = Z
R
ξ(x)ϕ(f (x))ξ(x)dλ = Z
A
ϕ ◦ f dλ = Z
ϕ ◦ f dν.
In the particular case where ϕ = 1
B(where B is a Borelian subset of C ), Z
1
Bdµ
ξ= Z
1
B◦ f dν = Z
R
1
f−1(B)dν = ν(f
−1(B )) = f
∗(ν)(B) and we deduce that µ
ξ(B) = R
1
Bdµ
ξ= f
∗(ν )(B), for all Borelian subset B. This implies µ
ξ= f
∗(ν).
3. Let H be a complex Hilbert space and T ∈ L(H) be self-adjoint.
(a) Show that : ∀n ∈ N , ||T
2n|| = ||T ||
2n.
Answer — We first show this result for n = 1, i.e. ||T
2|| = ||T ||
2. By using the fact that T
2is self-adjoint and a result from the course, we have
||T
2|| = sup
||x||≤1|hx, T
2xi| = sup
||x||≤1|hT
∗x, T xi| = sup
||x||≤1|hT x, T xi|
= sup
||x||≤1||T x||
2=
sup
||x||≤1||T x||
2= ||T ||
2.
Similarly, by using the fact that T
2nis self-adjoint, we show that ||T
2n|| = ||T
2n−1||
2. We conclude by recursion.
(b) Show by recursion that the following property P(n) is true for any n ∈ N : P (n) : ∀k ∈ N such that 0 ≤ k ≤ 2
n, ||T
k|| = ||T ||
k. [Hint : for 2
n< k ≤ 2
n+1, consider ` := 2
n+1− k and T
`T
k.]
Answer — The property P(0) is straightforward (||T
0|| = ||T||
0et ||T || = ||T ||).
Assume that we have shown P (n) let us prove P (n + 1) : it is enough to show that
||T
k|| = ||T||
kfor 2
n< k ≤ 2
n+1. Let ` := 2
n+1− k, then 0 ≤ ` < 2
n. Hence by using the result of the previous question, we get
||T ||
2n+1= ||T
2n+1|| = ||T
`T
k|| ≤ ||T
`|| ||T
k||.
We can then use P (n) with `, i.e. ||T
`|| = ||T||
`. We hence obtain
||T ||
`+k≤ ||T ||
`||T
k||
which gives by simplification ||T ||
k≤ ||T
k||. The reverse inequality ||T ||
k≥ ||T
k|| is
actually true for any operator. We deduce ||T ||
k= ||T
k||.
Conclusion : we have shown that, for any self-adjoint operator T ∈ L(H),
∀n ∈ N, ||T
n|| = ||T ||
n.
4. Let H be a complex Hilbert space. The goal of this exercise is to show that there are no bounded self-adjoint
1operators P, Q ∈ L(H) such that [P, Q] = λ1
H, where 1
H∈ L(H) is the identity operator, λ is a complex constant
2and [P, Q] := P Q − QP .
(a) Assume that H has a finite dimension n. Show, by using a simple argument, that there are no operators P, Q ∈ L(H) such that [P, Q] = 1
H.
Answer — On the one hand the trace of [P, Q] vanishes, on the other hand the trace of 1
His n.
(b) In the following we assume that the dimension of H is infinite. Show that, for any pair of operators P, Q ∈ L(H), we have :
∀n ∈ N
∗, [P
n, Q] =
n
X
j=1
P
n−j[P, Q]P
j−1.
Answer — We write that [P
n, Q] = P
nQ − QP
n= (P
nQ − P
n−1QP ) + (P
n−1QP − P
n−2QP
2) + · · · + (P QP
n−1− QP
n)
= P
nj=1
P
n−j(P Q − QP )M
j−1.
(c) We argue by contradiction and we assume that there exists operators P, Q ∈ L(H) such that [P, Q] = 1
H. By using the result of the previous question and of the previous exercise, show that, for any n ∈ N
∗,
2||P || ||Q|| ≥ n and conclude to a contradiction.
Answer — First method (assuming that P and Q are self-adjoint) : Because of the previous question and the hypothesis [P, Q] = 1
H, we have [P
n, Q] = nP
n−1. Hence, by using the result of Exercise 1,
n||P ||
n−1= ||nP
n−1|| = ||[P
n, Q]|| ≤ 2||P
n|| ||Q|| = 2||P||
n||Q||,
which gives us, by dividing by ||P||
n−1, that n ≤ 2||P|| ||Q||. Since n can be chosen arbitrarily, this is impossible, since P and Q are bounded.
Second method (without assuming that P and Q are selfadjoint) :
nkP
n−1k = k[P, Q]k = kP
n−1P Q − QP P
n−1k ≤ kP
n−1k(kPk kQk + kQk kP k)
implies n ≤ 2kP k kQk.
1. Sorry, this hypothesis was missing : the aim of this question was to use the result of the previous exercise on self-adjoint operators. We shall see, however, that the hypothesis thatP andQbe self-adjoint is not essential.
2. The reason for this other modification of the hypothesis (we set λ= 1in the original exam) is that, ifP andQ are self-adjoint, then[P, Q]is necessarily skew adjoint, i.e. satisfies[P, Q]∗ =−[P, Q], so that condition [P, Q] =λ1H makes sense iffλis complex imaginary.
(d) Give an example of a complex Hilbert space H and two non bounded operators P and Q such that there exists a dense vector subspace V ⊂ H such that ∀ϕ ∈ V , [P, Q]ϕ = ϕ
Answer — We may choose H := L
2(R, C), D(P ) = H
1(R, C), P = i
dxd, D(Q) = {f ∈ L
2( R , C ); R
R
|xf (x)|
2dx < +∞} and Q defined by (Qf )(x) = xf (x).
5. Let H := `
2(Z, C) ' `
2(Z). We denote by (
n)
n∈Zthe canonical Hilbertian Hermitian orthogonal basis of H (i.e.
nis the sequence which vanishes for all relative integer, excepted for n, for which it takes the value 1). We note L, R ∈ L(H) the operators defined by :
L
n=
n−1et R
n=
n+1, ∀n ∈ Z and A := L + R ∈ L(H).
(a) Compute L
∗and R
∗. Deduce that A is self-adjoint.
Answer — For all x = P
n∈Z
x
nnand y = P
n∈Z
y
nn, we have Ly = P
n∈Z
y
nn−1= P
n∈Z
y
n+1nand thus :
hx, Lyi = X
n∈Z
x
ny
n+1= X
n∈Z
x
n−1y
n= hRx, yi,
since Rx = P
n∈Z
x
nn+1= P
n∈Z
x
n−1n. Hence L
∗= R. Similarly we obtain that R
∗= L. We deduce then that A is self-adjoint.
(b) We note U : `
2( Z ) −→ L
2( R / Z ) the unitary operator defined by : (U
n) (θ) = e
i2πnθ,
∀n ∈ Z (Fourier series isomorphism). For all function m ∈ L
∞( R / Z ) we note m b ∈ L L
2( R / Z )
the multiplication operator defined by :
∀f ∈ L
2(R/Z), ( mf b ) (θ) = m(θ)f (θ), p.p.
Find U LU
−1and U RU
−1and show that they coincide with multiplication operators by functions to be precised. Deduce U AU
−1.
Answer — For any function f ∈ L
2( R / Z ) the Fourier series of which is f (θ) = P
n∈Z
a
ne
i2πnθ, we have U
−1f = P
n∈Z
a
nnand thus LU
−1f = P
n∈Z
a
nn−1, which implies :
U LU
−1f (θ) = X
n∈Z
a
ne
i2π(n−1)θ= e
−i2πθX
n∈Z
a
ne
i2πnθ= e
−i2πθf(θ).
Hence U LU
−1coincides with the multiplication the multiplication operator by the function θ 7−→ e
−i2πθ. Similarly U RU
−1coincides with the multiplication operator by the function θ 7−→ e
i2πθ. We deduce that U AU
−1= U LU
−1+ U RU
−1coincides with the multiplication operator by the function θ 7−→ e
i2πθ+ e
−i2πθ= 2 cos(2πθ), i.e. U AU
−1= 2 cos(2πθ). \
(c) Let ψ ∈ H be different of 0 and g := U ψ. We note F
ψ:= {P (A)ψ| P ∈ C [X]} =
Vect
C{A
nψ| n ∈ N }.
Show that the U F
ψ, the image of F
ψby U , is equal to : P
+g := {f g| f ∈ P
+}, where P
+is the subspace of L
2( R / Z ) of polynomials in e
i2πθand e
−i2πθwhich are even function of θ.
Answer — A consequence of the previous question is that, ∀n ∈ N , U (A
nψ) = U A
nU
−1U ψ = (U AU
−1)
ng = (2 cos(2πθ))
ng,
hence, more generally, for any polynomial P ∈ C [X], U(P (A)ψ) = P (2 cos(2πθ))g.
We deduce that U F
ψcoincides with {P (2 cos(2πθ))g| P ∈ C [X]}. Hence it suffices to remark
3that the set Q := {P (2 cos(2πθ))| P ∈ C [X]} coincides with P
+Lastly the closure of P
+dans L
2( R / Z ) is the set of even functions of the variable θ.
(d) Let ψ and g be as in the preceding question. We define h ∈ L
2( R / Z ) by : h(θ) = 2i sin(2πθ)g(−θ).
Show that, ∀f
1, f
2∈ P
+, hf
1g, f
2hi
L2= 0.
Answer — For any functions f
1, f
2∈ P
+, we have : hf
1g, f
2hi
L2=
Z
R/Z
f
1(θ)g(θ)f
2(θ)2i sin(2πθ)g(−θ)dθ,
which vanishes since the functions θ 7−→ f
1(θ)f
2(θ) and θ 7−→ g(θ)g(−θ) are even and the function θ 7−→ 2i sin(2πθ) is odd, so that hf
1g, f
2hi
L2is equal to the value of the integral of an odd function.
(e) Deduce from the preceding questions that A does not admit a cyclic vector.
Answer — For any ψ ∈ H \ {0}, let g := U ψ and χ := U
−1h ∈ H, where h is defined in terms of g as in the previous question. Then, because of Parseval’s inequality, for any polynomials P
1, P
2∈ C [X],
hP
1(A)ψ, P
2(A)χi = hU P
1(A)ψ, U P
2(A)χi
L2= hf
1g, f
2hi
L2,
where f
1, f
2∈ P
+are such that f b
1= U P
1(A)U
−1and f b
2= U P
2(A)U
−1(recall that we denote by f b the multiplication operator by the function f). Hence, by the result of the previous question, hP
1(A)ψ, P
2(A)χi = 0. This implies that F
ψ⊥ F
χ. Since F
χ6= {0}, we deduce that F
ψis not dense in H, i.e. that ψ is not cyclic for A.
3. Let us show it by recursion on N : the set PN :=n P
−N≤n≤Nanei2πnθ|an∈C, a−n=an
o
is equal to QN :={P(2 cos(2πθ))|degP ≤N}. ForN = 0, the result is straightforward. Assume thatPN−1=QN−1 let us show thatPN=QN. We use the identity :
(ei2πθ+e−i2πθ)N =
N
X
j=0
N!
(N−j)!j!ei2π(2j−N)θ=ei2πN θ+e−i2πN θ+RN(θ), (1) whereRN(θ) =PN−1
j=1 N!
(N−j)!j!ei2π(2j−N)θ. It is clear thatRN∈ PN−1=QN−1(note in particular thatRN(θ) = RN(−θ)). We deduceei2πN θ+e−i2πN θ∈ QN and(ei2πθ+e−i2πθ)N ∈ PN. Any element ofPN readsa(ei2πN θ+ e−i2πN θ) +α(θ), whereα∈ PN−1, and is thus contained in QN. Similarly any element ofQN readsb(ei2πθ+ e−i2πθ)N+β(θ), whereβ∈ QN−1, and is thus also contained inPN.