Chapter 3.13

Unbounded Operators

[25/0%]
Show
LaTeX
Problem 3.13.1

The associative law (T1T2)T3=T1(T2T3)\left(T_{1} T_{2}\right) T_{3} = T_{1}\left(T_{2} T_{3}\right) has been used freely throughout this chapter. Prove it. Prove also that T1⊂T2T_{1} \subset T_{2} implies ST1⊂ST2S T_{1} \subset S T_{2} and T1S⊂T2ST_{1} S \subset T_{2} S.

?
Problem 3.13.2

Let TT be a densely defined operator in HH. Prove that TT has a closed extension if and only if D(T∗)\mathscr {D}\left(T^{*}\right) is dense in HH. In that case, prove that T∗∗T^{* *} is an extension of TT.

?
Problem 3.13.3

By Theorem 13.8,D(T∗)={0}13.8, \mathscr {D}\left(T^{*}\right) = \left\{ 0\right\} for a densely defined operator TT in HH if and only if G(T)\mathscr {G}(T) is dense in H×HH \times H. Show that this can actually happen.

Suggestion: Let {en:n=1,2,3,…}\left\{ e_{n}: n = 1,2,3, \ldots \right\} be an orthonormal basis of HH; let {xn}\left\{ x_{n}\right\} be a dense subset of HH; define Ten=xnT e_{n} = x_{n}; and extend TT linearly to D(T)\mathscr {D}(T), the set of all finite linear combinations of the basis vectors ene_{n}. Show that the graph of this TT is dense in H×HH \times H.

?
Problem 3.13.4

Suppose TT is a densely defined, closed operator in HH, and T∗T⊂TT∗T^{*} T \subset T T^{*}. Does it follow that TT is normal?

?
Problem 3.13.5

Suppose TT is a densely defined operator in HH, and (Tx,x)=0(T x, x) = 0 for every x∈D(T)x \in \mathscr {D}(T). Does it follow that Tx=0T x = 0 for every x∈D(T)x \in \mathscr {D}(T)?

?
Problem 3.13.6

If TT is an operator in HH, define

N(T)={x∈D(T):Tx=0}. \mathcal{N}(T) = \left\{ x \in \mathscr {D}(T): T x = 0\right\} .

If D(T)\mathscr {D}(T) is dense, prove that

N(T∗)=R(T)⊥∩D(T∗). \mathcal{N}\left(T^{*}\right) = \mathscr {R}(T)^{\perp } \cap \mathscr {D}\left(T^{*}\right).

If TT is also closed, prove that

N(T)=R(T∗)⊥∩D(T) \mathscr {N}(T) = \mathscr {R}\left(T^{*}\right)^{\perp } \cap \mathscr {D}(T)

This generalizes Theorem 12.10.

?
Problem 3.13.7

Consider the following three boundary value problems. The differential equation is

f′′−f=g f^{\prime \prime }-f = g

where g∈L2([0,1])g \in L^{2}([0,1]) is given. The choices of boundary conditions are

(i) f(0)=f(1)=0f(0) = f(1) = 0.

(ii) f′(0)=f′(1)=0f^{\prime }(0) = f^{\prime }(1) = 0.

(iii) f(0)=f(1)f(0) = f(1) and f′(0)=f′(1)f^{\prime }(0) = f^{\prime }(1).

Show that each of these problems has a unique solution ff such that f′f^{\prime } is absolutely continuous and f′′∈L2([0,1])f^{\prime \prime } \in L^{2}([0,1]). Hint: Combine Example 13.4 with Theorem 13.13.

Do this also by solving the problems explicitly.

?
Problem 3.13.8
?
(a)

Prove the self-adjointness of the operator TT in L2(R)L^{2}(R), defined by Tf=if′T f = i f^{\prime }, with D(T)\mathscr {D}(T) consisting of all absolutely continuous f∈L2f \in L^{2} such that f′∈L2f^{\prime } \in L^{2}.

Hint: You may need to know that f(t)→0f(t) \rightarrow 0 as t→±∞t \rightarrow \pm \infty for every f∈D(T)f \in \mathscr {D}(T). Prove this. Or prove more, namely, that every f∈D(T)f \in \mathscr {D}(T) is the Fourier transform of an L1L^{1}-function.

(b)

Fix g∈L2(R)g \in L^{2}(R). Use Theorem 13.13 to prove that the equation

f′′−f=g f^{\prime \prime }-f = g

has a unique absolutely continuous solution f∈L2f \in L^{2}, which has f′∈L2f^{\prime } \in L^{2}, f′′∈L2f^{\prime \prime } \in L^{2}, and f′f^{\prime } absolutely continuous.

Prove also, by direct calculation, that

f(x)=−12∫−∞xet−xg(t)dt−12∫x∞ex−tg(t)dt f(x) = -\frac{1}{2} \int _{-\infty }^{x} e^{t-x} g(t) d t-\frac{1}{2} \int _{x}^{\infty } e^{x-t} g(t) d t

This solution can also be found by means of Fourier transforms.

Problem 3.13.9

Let H2H^{2} be the space of all holomorphic functions f(z)=∑cnznf(z) = \sum c_{n} z^{n} in the open unit disc that satisfy

∥f∥2=∑n=0∞∣cn∣2<∞ \left\| f\right\| ^{2} = \sum _{n = 0}^{\infty }\left|c_{n}\right|^{2} < \infty

Show that H2H^{2} is a Hilbert space which is isomorphic to ℓ2\ell^{2} via the one-to-one correspondence f↔{cn}f \leftrightarrow \left\{ c_{n}\right\}.

Define V∈B(H2)V \in \mathscr {B}\left(H^{2}\right) by (Vf)(z)=zf(z)(V f)(z) = z f(z). Show that VV is the Cayley transform of the symmetric operator TT in H2H^{2}, given by

(Tf)(z)=i1+z1−zf(z) (T f)(z) = i \frac{1+z}{1-z} f(z)

Find the ranges of T+iIT+i I and of T−iIT-i I; show that one is H2H^{2} and one has codimension 1. (Compare with Example 13.21).

?
Problem 3.13.10

With H2H^{2} as in Exercise 9, define VV now by

(Vf)(z)=zf(z2) (V f)(z) = z f\left(z^{2}\right)

Show that VV is an isometry which is the Cayley transform of a closed symmetric operator TT in H2H^{2}, whose deficiency indices are 0 and ∞\infty.

?
Problem 3.13.11

Prove part (c) of Lemma 13.18.

?
Problem 3.13.12
?
(a)

In the context of Theorem 13.24, how are the operators Ψ(f+g)\Psi (f+g) and Ψ(f)+Ψ(g)\Psi (f)+\Psi (g) related?

(b)

If ff and gg are measurable and gg is bounded, prove that Ψ(g)\Psi (g) maps Df\mathscr {D}_{f} into Df\mathscr {D}_{f}.

(c)

Prove that Ψ(f)=Ψ(g)\Psi (f) = \Psi (g) if and only if f=gf = g a.e. [E], that is, if and only if

E({p:f(p)≠g(p)})=0 E(\left\{ p: f(p) \neq g(p)\right\} ) = 0
Problem 3.13.13

Is the operator CC that occurs in the proof of Theorem 13.33 normal?

?
Problem 3.13.14

Prove that every normal operator NN in HH, bounded or not, has a polar decomposition

N=UP=PU N = U P = P U

where UU is unitary, PP is self-adjoint, P≥0P \geq 0. Moreover, D(P)=D(N)\mathscr {D}(P) = \mathscr {D}(N).

?
Problem 3.13.15

Prove the following extension of Theorem 12.16: If T∈B(H)T \in \mathscr {B}(H), if MM and NN are normal operators in HH, and if TM⊂NTT M \subset N T, then also TM∗⊂N∗TT M^{*} \subset N^{*} T.

?
Problem 3.13.16

Suppose TT is a closed operator in H,D(T)=D(T∗)H, \mathscr {D}(T) = \mathscr {D}\left(T^{*}\right), and ∥Tx∥=∥T∗x∥\left\| T x\right\| = \left\| T^{*} x\right\| for every x∈D(T)x \in \mathscr {D}(T). Prove that TT is normal. Hint: Begin by proving that

(Tx,Ty)=(T∗x,T∗y)(x∈D(T),y∈D(T)) (T x, T y) = \left(T^{*} x, T^{*} y\right) \quad (x \in \mathscr {D}(T), y \in \mathscr {D}(T))
?
Problem 3.13.17

Prove that the spectrum σ(T)\sigma (T) of any operator TT in HH is a closed subset of CC. (See Definition 13.26). Hint: If ST⊂TS=IS T \subset T S = I, and S∈B(H)S \in \mathscr {B}(H), then S(I−λS)−1S(I-\lambda S)^{-1} is a bounded inverse of T−MT-M, for small ∣λ∣\left|\lambda \right|.

?
Problem 3.13.18

Put ϕ(t)=exp⁡(−t2)\phi (t) = \exp \left(-t^{2}\right). Define S∈B(L2)S \in \mathscr {B}\left(L^{2}\right), where L2=L2(R)L^{2} = L^{2}(R), by

(Sf)(t)=ϕ(t)f(t−1)(f∈L2) (S f)(t) = \phi (t) f(t-1) \quad \left(f \in L^{2}\right)

so that (S2f)(t)=ϕ(t)ϕ(t−1)f(t−2)\left(S^{2} f\right)(t) = \phi (t) \phi (t-1) f(t-2), etc. (Note that SS is presented in its polar decomposition S=PUS = P U).

Find S∗S^{*}. Compute that

∥Sn∥=exp⁡{−(n−1)n(n+1)12}(n=1,2,3,…) \left\| S^{n}\right\| = \exp \left\{ -\frac{(n-1) n(n+1)}{12}\right\} \quad (n = 1,2,3, \ldots )

Conclude that SS is one-to-one, that R(S)\mathscr {R}(S) is dense in L2L^{2}, and that σ(S)={0}\sigma (S) = \left\{ 0\right\}. Define TT, with domain D(T)=R(S)\mathscr {D}(T) = \mathscr {R}(S), by

TSf=f(f∈L2) T S f = f \quad \left(f \in L^{2}\right)

Prove that σ(T)\sigma (T) is empty.

?
Problem 3.13.19

Let T1,T2,T3T_{1}, T_{2}, T_{3} be as in Example 13.4, put

D(T4)={f∈D(T1):f(0)=0} \mathscr {D}\left(T_{4}\right) = \left\{ f \in \mathscr {D}\left(T_{1}\right): f(0) = 0\right\}

and define T4f=if′T_{4} f = i f^{\prime } for all f∈D(T4)f \in \mathscr {D}\left(T_{4}\right).

Prove the following assertions:

?
(a)

Every λ∈C\lambda \in \mathbb {C} is in the point spectrum of T1T_{1}.

(b)

σ(T2)\sigma \left(T_{2}\right) consists of the numbers 2πn2 \pi n, where nn runs through the integers; each of these is in the point spectrum of T2T_{2}.

(c)

R(T3−λI)\mathscr {R}\left(T_{3}-\lambda I\right) has codimension 1 for every λ∈C\lambda \in \mathscr {C}. Hence σ(T3)=C\sigma \left(T_{3}\right) = \mathscr {C}. The point spectrum of T3T_{3} is empty.

(d)

σ(T4)\sigma \left(T_{4}\right) is empty.

Hint: Study the differential equation if′−λf=gi f^{\prime }-\lambda f = g. This illustrates how sensitive the spectrum of a differential operator is to its domain (in this case, to the boundary conditions that are imposed).

Problem 3.13.20

Show that every nonempty closed subset of CC is the spectrum of some normal operator in HH (if dim⁡H=∞\operatorname {dim} H = \infty).

?
Problem 3.13.21

Define Q(t)∈B(L2)Q(t) \in \mathscr {B}\left(L^{2}\right), where L2=L2(R)L^{2} = L^{2}(R), by

(Q(t)f)(s)=f(s+t) (Q(t) f)(s) = f(s+t)
?
(a)

Prove that each Q(t)Q(t) is unitary.

(b)

Prove that {Q(t)}\left\{ Q(t)\right\} satisfies the conditions stated in Definition 13.34.

(c)

If AA is the infinitesimal generator of {Q(t)}\left\{ Q(t)\right\}, prove that f∈D(A)f \in \mathscr {D}(A) if and only if ∫∣yf^(y)∣2dy<∞\int \left|y \hat{f}(y)\right|^{2} d y < \infty (where f^\hat{f} is the Fourier transform of ff) and that Af=f′A f = f^{\prime } for all f∈D(A)f \in \mathscr {D}(A).

(d)

Prove that σ(A)\sigma (A) is the imaginary axis. More precisely, show that A−λIA-\lambda I is one-to-one for every λ∈C\lambda \in \mathscr {C}, that λ\lambda lies in the resolvent set of AA if and only if λ\lambda is not pure imaginary, and that the range of A−λIA-\lambda I is a proper dense subspace of L2L^{2} if λ\lambda is pure imaginary.

Hint: g∈R(A−λI)g \in \mathscr {R}(A-\lambda I) if and only if g∈L2g \in L^{2} and also g^(y)/(iy−λ)\hat{g}(y) /(i y-\lambda ) is in L2L^{2}.

Problem 3.13.22

If f∈H2f \in H^{2} (see Exercise 9) and f(z)=∑cnznf(z) = \sum c_{n} z^{n}, define

[Q(t)f](z)=∑n=0∞(n+1)−tcnzn(0≤t<∞) [Q(t) f](z) = \sum _{n = 0}^{\infty }(n+1)^{-t} c_{n} z^{n} \quad (0 \leq t < \infty )

Show that each Q(t)Q(t) is self-adjoint (and positive). Find the infinitesimal generator AA of the semigroup {Q(t)}\left\{ Q(t)\right\}. Is AA self-adjoint? Show that AA has pure point spectrum, at the points log⁡1,log⁡12,log⁡13,…\log 1, \log \frac{1}{2}, \log \frac{1}{3}, \ldots.

?
Problem 3.13.23

For f∈L2(R),x∈R,0<y<∞f \in L^{2}(R), x \in R, 0 < y < \infty, define

[Q(y)f](x)=1π∫−∞∞y(x−ξ)2+y2f(ξ)dξ [Q(y) f](x) = \frac{1}{\pi } \int _{-\infty }^{\infty } \frac{y}{(x-\xi )^{2}+y^{2}} f(\xi ) d \xi

and put Q(0)f=fQ(0) f = f. Show that {Q(y):0≤y<∞}\left\{ Q(y): 0 \leq y < \infty \right\} satisfies the conditions imposed in Definition 13.34 and that ∥Q(y)∥=1\left\| Q(y)\right\| = 1 for all yy.

Find the domain of the infinitesimal generator AA of {Q(y)}\left\{ Q(y)\right\}, and prove that

Af=−Hf′ A f = -H f^{\prime }

where HH is the Hilbert transform (Chapter 7, Exercise 24). Prove that −A-A is positive and self-adjoint.

?
Remark.
?

The integral represents a harmonic function in the upper half-plane, with boundary values ff. The semigroup property of {Q(y)}\left\{ Q(y)\right\} can be deduced from this, as well as from a look at the Fourier transforms of the functions Q(y)fQ(y) f.

Problem 3.13.24

Show that every isometric operator in HH has a closed isometric extension.

?
Problem 3.13.25

On the other hand, show that some symmetric operators in HH have no closed symmetric extension by completing the following outline.

Let {e1,e2,e3,…}\left\{ e_{1}, e_{2}, e_{3}, \ldots \right\} be an orthonormal basis of HH. Let XX be the set of all finite sums ∑αiei\sum \alpha_{i} e_{i}, subject to the condition ∑αi=0\sum \alpha_{i} = 0. Prove that XX is a dense subspace of HH. Define U∈B(H)U \in \mathscr {B}(H) by

U(∑1∞αiei)=α1e1−∑2∞αiei U\left(\sum _{1}^{\infty } \alpha _{i} e_{i}\right) = \alpha _{1} e_{1}-\sum _{2}^{\infty } \alpha _{i} e_{i}

and let VV be the restriction of UU to XX. Then VV is an isometry, with D(V)=X\mathscr {D}(V) = X, and I−VI-V is one-to-one on XX. Hence VV is the Cayley transform of a symmetric operator TT. Any closed symmetric extension of TT would correspond to a closed isometric extension V1V_{1} of VV, with I−V1I-V_{1} one-to-one. But D(V)\mathscr {D}(V) is dense in HH; hence VV has only one closed isometric extension, namely UU, and I−UI-U is not one-to-one.

?