Frame operator and dual frames

[21/100%]
Show
LaTeX

Suppose that {xn}\left\{ x_{n}\right\} is a Bessel sequence that is a basis for a Hilbert space HH. Let {yn}\left\{ y_{n}\right\} be the biorthogonal sequence, and let BB be a Bessel bound.

?
(a)

Show that

∀x∈H,1B∥x∥2≤∑n∣⟨x,yn⟩∣2 \forall x \in H, \quad \frac{1}{B}\left\| x\right\| ^{2} \leq \sum _{n}\left| \left\langle x, y_{n} \right\rangle \right|^{2}

We say that {yn}\left\{ y_{n}\right\} has a lower frame bound of B−1B^{-1}; compare Definition 8.2. Note that {yn}\left\{ y_{n}\right\} need not be a Bessel sequence; see Exercise 7.1(h).

(b)

Show that for all N∈NN \in \mathbb {N} and c1,…,cN∈Fc_{1}, \ldots , c_{N} \in \mathbf{F} we have

1B∑n=1N∣cn∣2≤∥∑n=1Ncnyn∥2 \frac{1}{B} \sum _{n = 1}^{N}\left|c_{n}\right|^{2} \leq \left\| \sum _{n = 1}^{N} c_{n} y_{n}\right\| ^{2}

Let {xn}\left\{ x_{n}\right\} be an AA-tight frame for a Hilbert space HH. Show that the frame operator for {xn}\left\{ x_{n}\right\} (see Definition 7.3) is S=AIS = A I, and use this to show that x=A−1∑⟨x,xn⟩xnx = A^{-1} \sum \left\langle x, x_{n} \right\rangle x_{n} for x∈Hx \in H.

?
Problem 8.11 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 8 "Frames in Hilbert Spaces"

Let v1,…,vnv_{1}, \ldots , v_{n} be vectors in Fd\mathbf{F}^{d}. Give direct proofs of the following statements.

?
(a)

{v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} is a Bessel sequence in Fd\mathbf{F}^{d}, its synthesis operator corresponds to multiplication by the matrix RR that has v1,…,vnv_{1}, \ldots , v_{n} as columns, and its analysis operator is C=R∗C = R^{*}, the Hermitian of the matrix RR.

(b)

The following are equivalent: (i) {v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} spans Fd\mathbf{F}^{d}, (ii) S=C∗CS = C^{*} C is positive definite, (iii) {v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} is a frame for Fd\mathbf{F}^{d}. Further, in case these hold, the optimal frame bounds for {v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} are λ1,λd\lambda_{1}, \lambda_{d}, where λ1\lambda_{1} is the smallest eigenvalue of SS and λd\lambda_{d} is the largest eigenvalue.

(c)

The following are equivalent: (i) {v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} is linearly independent, (ii) G=CC∗G = C C^{*} is positive definite, (iii) {v1,…,vn}\left\{ v_{1}, \ldots , v_{n}\right\} is a Riesz sequence in Fd\mathbf{F}^{d}.

Given a sequence {xn}\left\{ x_{n}\right\} in a Hilbert space HH, show that {xn}\left\{ x_{n}\right\} is a frame with frame bounds A,BA, B if and only if Sx=∑⟨x,xn⟩xnS x = \sum \left\langle x, x_{n} \right\rangle x_{n} is a well-defined positive linear mapping of HH into HH that satisfies AI≤S≤BIA I \leq S \leq B I.

?

This exercise will give an alternative proof that the frame operator SS is a topological isomorphism.

?
(a)

Show that if U,V∈B(H)U, V \in \mathcal{B}(H) are positive and U≤VU \leq V then ∥U∥≤∥V∥\left\| U\right\| \leq \left\| V\right\|.

(b)

Let SS be the frame operator for a frame {xn}\left\{ x_{n}\right\} that has frame bounds A,BA, B. Prove the operator inequalities

0≤I−1BS≤B−ABI 0 \leq I-\frac{1}{B} S \leq \frac{B-A}{B} I

and

0≤I−2A+BS≤B−AB+AI 0 \leq I-\frac{2}{A+B} S \leq \frac{B-A}{B+A} I

and use either one of these to show that SS is a topological isomorphism.

Prove Theorem 8.18: let {xn}\left\{ x_{n}\right\} be a frame for HH with analysis operator CC.

?
(a)

If {yn}\left\{ y_{n}\right\} is a Bessel sequence in HH that is an alternative dual for {xn}\left\{ x_{n}\right\}, then {yn}\left\{ y_{n}\right\} is a frame and {xn}\left\{ x_{n}\right\} is an alternative dual frame for {yn}\left\{ y_{n}\right\}.

(b)

A Bessel sequence {yn}\left\{ y_{n}\right\} is the canonical dual frame for {xn}\left\{ x_{n}\right\} if and only if it is an alternative dual and range⁡(C)=range⁡(C~)\operatorname {range}(C) = \operatorname {range}(\widetilde{C}), where C~\widetilde{C} is the analysis operator of {yn}\left\{ y_{n}\right\}.

Complete the proof of Theorem 8.19: let MM be a closed subspace of HH, let {xn}\left\{ x_{n}\right\} be a frame for HH, and let PP be the orthogonal projection of HH onto MM.

?
(a)

{Pxn}\left\{ P x_{n}\right\} is a frame for MM with the same frame bounds as {xn}\left\{ x_{n}\right\}, and {Px~n}\left\{ P \widetilde{x}_{n}\right\} is an alternative dual frame for {Pxn}\left\{ P x_{n}\right\}.

(b)

{Px~n}\left\{ P \widetilde{x}_{n}\right\} is the canonical dual frame for {Pxn}\left\{ P x_{n}\right\} if and only if PS=SPPS = SP, where SS is the frame operator for {xn}\left\{ x_{n}\right\}.

Let {xn}\left\{ x_{n}\right\} be a frame for a Hilbert space HH. Let C,R,SC, R, S be the analysis, synthesis, and frame operators for {xn}\left\{ x_{n}\right\}, and let C~,R~\widetilde{C}, \widetilde{R} be the analysis and synthesis operators for the canonical dual frame {x~n}\left\{ \widetilde{x}_{n}\right\}.

?
(a)

Show that C~=CS−1\widetilde{C} = C S^{-1} and R~=S−1R\widetilde{R} = S^{-1} R.

(b)

Show that the orthogonal projection PP of ℓ2\ell^{2} onto range (C)(C) is Pc=CR~c=CS−1Rc={⟨∑ncnx~n,xk⟩}k∈N,c=(cn)∈ℓ2P c = C \widetilde{R} c = C S^{-1} R c = \left\{ \left\langle \sum_{n} c_{n} \widetilde{x}_{n}, x_{k} \right\rangle \right\}_{k \in \mathbb {N}}, \quad c = \left(c_{n}\right) \in \ell^{2}.

Let {xn}\left\{ x_{n}\right\} be a frame for a Hilbert space HH and let C:H→ℓ2C: H \rightarrow \ell^{2} be its analysis operator. Let {δn}\left\{ \delta_{n}\right\} be the standard basis for ℓ2\ell^{2}, and prove that the following two statements are equivalent.

?
(a)

{yn}\left\{ y_{n}\right\} is an alternative dual frame for {xn}\left\{ x_{n}\right\}.

(b)

There is a bounded left-inverse VV of CC such that yn=Vδny_{n} = V \delta_{n} for all n∈Nn \in \mathbb {N} (that is, V:ℓ2→HV: \ell^{2} \rightarrow H is bounded, VC=IV C = I, and yn=Vδny_{n} = V \delta_{n}).

Let T∈B(H)T \in \mathcal{B}(H) be a Hilbert-Schmidt operator on a Hilbert space HH (see Definition B.7). Show that if {xn}\left\{ x_{n}\right\} is a frame for HH with canonical dual frame {x~n}\left\{ \widetilde{x}_{n}\right\}, then the Hilbert-Schmidt norm of TT is given by ∥T∥HS2=∑⟨Txn,Tx~n⟩\left\| T\right\|_{\mathrm{HS}}^{2} = \sum \left\langle T x_{n}, T \widetilde{x}_{n} \right\rangle.

?

Prove Corollary 8.23.

?
Remark.
?

Corollary 8.23: If {xn}\left\{ x_{n}\right\} is a frame for a Hilbert space HH, then the following statements are equivalent. (a) {xn}\left\{ x_{n}\right\} is an exact frame. (b) {xn}\left\{ x_{n}\right\} and {x~n}\left\{ \widetilde{x}_{n}\right\} are biorthogonal. (c) ⟨xn,x~n⟩=1\left\langle x_{n}, \widetilde{x}_{n} \right\rangle = 1 for all nn.

Show that if {xn}\left\{ x_{n}\right\} is an exact frame for a Hilbert space HH then it has a unique dual frame (the canonical dual).

?
Problem 8.26 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 8 "Frames in Hilbert Spaces"

Let {xn}\left\{ x_{n}\right\} be a frame for a Hilbert space HH. Let C,RC, R be the analysis and synthesis operators for {xn}\left\{ x_{n}\right\}, and let C~,R~\widetilde{C}, \widetilde{R} be the analysis and synthesis operators for the canonical dual frame {x~n}\left\{ \widetilde{x}_{n}\right\}.

?
(a)

Show that range⁡(C)=range⁡(C~)\operatorname {range}(C) = \operatorname {range}(\widetilde{C}).

(b)

Show that the pseudoinverse R†R^{\dagger } of the synthesis operator RR is R†=C~R^{\dagger } = \widetilde{C}, so

R†x=C~x={⟨x,x~n⟩}={⟨x,S−1xn⟩},x∈H R^{\dagger } x = \widetilde{C} x = \left\{ \left\langle x, \widetilde{x}_{n} \right\rangle \right\} = \left\{ \left\langle x, S^{-1} x_{n} \right\rangle \right\} , \quad x \in H
(c)

Show that the optimal frame bounds for {xn}\left\{ x_{n}\right\} are

A=∥C~∥−2=∥S−1∥−1 and B=∥C∥2=∥S∥ A = \left\| \widetilde{C}\right\| ^{-2} = \left\| S^{-1}\right\| ^{-1} \quad \text{ and } \quad B = \left\| C\right\| ^{2} = \left\| S\right\|

Prove Theorem 8.32.

?
Remark.
?

Theorem 8.32: Given a sequence {xn}\left\{ x_{n}\right\} in a Hilbert space HH, the following statements are equivalent. (a) {xn}\left\{ x_{n}\right\} is a Riesz basis for HH. (b) The analysis operator Cx={⟨x,xn⟩}C x = \left\{ \left\langle x, x_{n} \right\rangle \right\} maps HH bijectively onto ℓ2\ell^{2}. (c) The synthesis operator Rc=∑cnxnR c = \sum c_{n} x_{n} is well defined for each c=(cn)∈ℓ2c = (c_{n}) \in \ell^{2}, and maps ℓ2\ell^{2} bijectively onto HH. (d) There exists an orthonormal basis {en}\left\{ e_{n}\right\} for HH and a bounded bijection T∈B(H)T \in \mathcal{B}(H) such that Ten=xnT e_{n} = x_{n} for every nn. (e) {xn}\left\{ x_{n}\right\} is an ω\omega-independent frame. (f) {xn}\left\{ x_{n}\right\} is an ℓ2\ell^{2}-independent frame, i.e., it is a frame and if ∑cnxn=0\sum c_{n} x_{n} = 0 for some sequence (cn)∈ℓ2(c_{n}) \in \ell^{2} then cn=0c_{n} = 0 for every nn.

Problem 8.32 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 8 "Frames in Hilbert Spaces"

Let {xn}\left\{ x_{n}\right\} be a frame for a Hilbert space HH and let {yn}\left\{ y_{n}\right\} be a frame for a Hilbert space KK. Inspired by Definition 4.19, declare {xn}\left\{ x_{n}\right\} and {yn}\left\{ y_{n}\right\} to be equivalent if there exists a topological isomorphism T:H→KT: H \rightarrow K such that Txn=ynT x_{n} = y_{n} for every nn.

?
(a)

Show that {xn}\left\{ x_{n}\right\} and {yn}\left\{ y_{n}\right\} are equivalent if and only if range (CX)=\left(C_{X}\right) = range (CY)\left(C_{Y}\right), where CX,CYC_{X}, C_{Y} are the analysis operators for {xn}\left\{ x_{n}\right\} and {yn}\left\{ y_{n}\right\}.

(b)

Show that part (a) is equivalent to:

∀(cn)∈ℓ2,∑cnxn=0⟺∑cnyn=0. \forall \left(c_{n}\right) \in \ell ^{2}, \quad \sum c_{n} x_{n} = 0 \Longleftrightarrow \sum c_{n} y_{n} = 0.
(c)

Exhibit a frame {xn}\left\{ x_{n}\right\} and a permutation σ\sigma of N\mathbb {N} that fixes all but finitely many elements of N\mathbb {N} such that {xn}\left\{ x_{n}\right\} is not equivalent to {xσ(n)}\left\{ x_{\sigma (n)}\right\}. Show that this cannot happen if {xn}\left\{ x_{n}\right\} is an exact frame.

(d)

Exhibit a frame {xn}\left\{ x_{n}\right\} and a sequence of signs εn=±1\varepsilon_{n} = \pm 1 such that {xn}\left\{ x_{n}\right\} is not equivalent to {εnxn}\left\{ \varepsilon_{n} x_{n}\right\}. Show that this cannot happen if {xn}\left\{ x_{n}\right\} is an exact frame.

Remark Parts (c) and (d) illustrate some of the weaknesses of this notion of equivalence for frames. A more robust notion of frame equivalence was introduced by Balan and Landau in [BL07].

Let HH be an infinite-dimensional separable Hilbert space. Show that if MM is an infinite-dimensional closed subspace of ℓ2\ell^{2}, then there exists a frame {xn}\left\{ x_{n}\right\} for HH such that range⁡(C)=M\operatorname {range}(C) = M, where CC is the analysis operator.

?
Problem 8.41 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 8 "Frames in Hilbert Spaces"

Prove Lemma 8.47: let {xn}\left\{ x_{n}\right\} be a Bessel sequence with analysis operator CC and synthesis operator R=C∗R = C^{*}, and suppose U,VU, V are operators such that VC=CUV C = C U.

?
(a)

U(ker⁡(C))⊆ker⁡(C)U\left(\operatorname {ker}(C)\right) \subseteq \operatorname {ker}(C).

(b)

V∗(ker⁡(R))⊆ker⁡(R)V^{*}\left(\operatorname {ker}(R)\right) \subseteq \operatorname {ker}(R).

(c)

V(range⁡(C))⊆range⁡(C)V\left(\operatorname {range}(C)\right) \subseteq \operatorname {range}(C).

(d)

U∗(range⁡(R))⊆range⁡(R)U^{*}\left(\operatorname {range}(R)\right) \subseteq \operatorname {range}(R).

Suppose that g∈L2(R)g \in L^{2}(\mathbb {R}) is such that T(g)\mathcal{T}(g) is a frame sequence in L2(R)L^{2}(\mathbb {R}), and let S:V0(g)→V0(g)S: V_{0}(g) \rightarrow V_{0}(g) be its frame operator.

?
(a)

Show that S(Tkf)=Tk(Sf)S\left(T_{k} f\right) = T_{k}(S f) for each f∈V0(g)f \in V_{0}(g). Use this to show that the canonical dual frame in V0(g)V_{0}(g) is T(h)\mathcal{T}(h) where h=S−1gh = S^{-1} g.

(b)

Show that

f∈V0(g)⟺∃m∈L2(T) such that f^=mg^ a.e.  f \in V_{0}(g) \Longleftrightarrow \exists m \in L^{2}(\mathbf{T}) \text{ such that } \widehat{f} = m \widehat{g} \text{ a.e. }
Problem 11.3 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 11 "Gabor Bases and Frames"

Let G(g,a,b)\mathcal{G}(g, a, b) be a Gabor frame for L2(R)L^{2}(\mathbb {R}).

?
(a)

Show that the frame operator SS commutes with MbnM_{b n} and TakT_{a k} for all k,n∈Zk, n \in \mathbb {Z}, and use this to show that S−1S^{-1} also commutes with MbnM_{b n} and TakT_{a k}.

(b)

Show that the canonical dual frame of G(g,a,b)\mathcal{G}(g, a, b) is the Gabor frame G(g~,a,b)\mathcal{G}(\widetilde{g}, a, b) where g~=S−1g\widetilde{g} = S^{-1} g.

(c)

Suppose that G(g,a,b)\mathcal{G}(g, a, b) is a frame for L2(R)L^{2}(\mathbb {R}). Show that G(g,a,b)\mathcal{G}(g, a, b) is a Riesz basis if and only if ⟨g,g~⟩=1\left\langle g, \widetilde{g} \right\rangle = 1.

(d)

Show that the canonical Parseval frame of a lattice Gabor frame is another lattice Gabor frame. Specifically, if G(g,a,b)\mathcal{G}(g, a, b) is a frame for L2(R)L^{2}(\mathbb {R}) and we set g♯=S−1/2gg^{\sharp } = S^{-1 / 2} g, where SS is the frame operator, then G(g♯,a,b)\mathcal{G}\left(g^{\sharp }, a, b\right) is a Parseval frame for L2(R)L^{2}(\mathbb {R}).

Problem 11.6 from Heil C. "A Basis Theory Primer" [Hei11]More problems in Chapter 11 "Gabor Bases and Frames"

Assume that the hypotheses of part (a) of Theorem 11.4 are satisfied, i.e., 0<ab≤1,supp⁡(g)⊆[0,b−1]0 < a b \leq 1, \operatorname {supp}\left(g\right) \subseteq \left[0, b^{-1}\right], and equation (11.3) holds. Prove the following statements about the frame G(g,a,b)\mathcal{G}(g, a, b).

?
(a)

The frame operator is pointwise multiplication by b−1G0b^{-1} G_{0}, i.e., Sf=b−1G0fS f = b^{-1} G_{0} f for f∈L2(R)f \in L^{2}(\mathbb {R}).

(b)

The canonical dual frame is G(g~,a,b)\mathcal{G}(\widetilde{g}, a, b) where g~=bg/G0\widetilde{g} = b g / G_{0}.

(c)

If ab=1a b = 1 then G(g,a,b)\mathcal{G}(g, a, b) is a Riesz basis for L2(R)L^{2}(\mathbb {R}).

(d)

If 0<ab<10 < a b < 1 then G(g,a,b)\mathcal{G}(g, a, b) is a redundant frame for L2(R)L^{2}(\mathbb {R}).

Show that if W(ψ,a,b)\mathcal{W}(\psi , a, b) is a wavelet frame for L2(R)L^{2}(\mathbb {R}), then its frame operator SS commutes with DanD_{a^{n}} for n∈Zn \in \mathbb {Z}. What does this imply about the structure of the canonical dual frame?

?

References

[Hei11] Heil C. A Basis Theory Primer. Birkhäuser, 2011.