Frames and operator representations of frames Ole Christensen Joint - - PowerPoint PPT Presentation

frames and operator representations of frames
SMART_READER_LITE
LIVE PREVIEW

Frames and operator representations of frames Ole Christensen Joint - - PowerPoint PPT Presentation

Frames and operator representations of frames Ole Christensen Joint work with Marzieh Hasannasab HATA DTU DTU Compute, Technical University of Denmark HATA: Harmonic Analysis - Theory and Applications https://hata.compute.dtu.dk/ Ole


slide-1
SLIDE 1

Frames and operator representations of frames

Ole Christensen Joint work with Marzieh Hasannasab

HATA – DTU DTU Compute, Technical University of Denmark HATA: Harmonic Analysis - Theory and Applications https://hata.compute.dtu.dk/ Ole Christensen Jakob Lemvig Mads Sielemann Jakobsen Marzieh Hasannasab Kamilla Haahr Nielsen Ehsan Rashidi Jordy van Velthoven

August 17, 2017

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 1 / 27

slide-2
SLIDE 2

Frames and overview of the talk

  • If a sequence {fk}∞

k=1 in a Hilbert spaces H is a frame, there exists

another frame {gk}∞

k=1 such that

f =

  • k=1

f, gkfk, ∀f ∈ H. Similar to the decomposition in terms of an orthonormal basis, but MUCH MORE flexible.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 2 / 27

slide-3
SLIDE 3

Frames and overview of the talk

  • If a sequence {fk}∞

k=1 in a Hilbert spaces H is a frame, there exists

another frame {gk}∞

k=1 such that

f =

  • k=1

f, gkfk, ∀f ∈ H. Similar to the decomposition in terms of an orthonormal basis, but MUCH MORE flexible.

  • We will consider representations of frames on the form

{fk}∞

k=1 = {Tnf1}∞ n=0 = {f1, Tf1, T2f1, · · · },

where T : H → H is a linear operator, possibly bounded.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 2 / 27

slide-4
SLIDE 4

Frames and overview of the talk

  • If a sequence {fk}∞

k=1 in a Hilbert spaces H is a frame, there exists

another frame {gk}∞

k=1 such that

f =

  • k=1

f, gkfk, ∀f ∈ H. Similar to the decomposition in terms of an orthonormal basis, but MUCH MORE flexible.

  • We will consider representations of frames on the form

{fk}∞

k=1 = {Tnf1}∞ n=0 = {f1, Tf1, T2f1, · · · },

where T : H → H is a linear operator, possibly bounded. Main conclusion: Frame theory is operator theory, with several interesting and challenging open problems!

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 2 / 27

slide-5
SLIDE 5

Bessel sequences

Definition A sequence {fk}∞

k=1 in H is called a Bessel sequence if there exists

a constant B > 0 such that

  • k=1

|f, fk|2 ≤ B ||f||2, ∀f ∈ H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 3 / 27

slide-6
SLIDE 6

Bessel sequences

Definition A sequence {fk}∞

k=1 in H is called a Bessel sequence if there exists

a constant B > 0 such that

  • k=1

|f, fk|2 ≤ B ||f||2, ∀f ∈ H. Theorem Let {fk}∞

k=1 be a sequence in H, and B > 0 be given. Then {fk}∞ k=1

is a Bessel sequence with Bessel bound B if and only if T : {ck}∞

k=1 → ∞

  • k=1

ckfk defines a bounded operator from ℓ2(N) into H and ||T|| ≤ √ B.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 3 / 27

slide-7
SLIDE 7

Bessel sequences

Pre-frame operator or synthesis operator associated to a Bessel sequence: T : ℓ2(N) → H, T{ck}∞

k=1 = ∞

  • k=1

ckfk The adjoint operator - the analysis operator: T∗ : H → ℓ2(N), T∗f = {f, fk}∞

k=1.

The frame operator: S : H → H, Sf = TT∗f =

  • k=1

f, fkfk. The series defining S converges unconditionally for all f ∈ H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 4 / 27

slide-8
SLIDE 8

Frames

Definition: A sequence {fk}∞

k=1 in H is a frame if there exist constants

A, B > 0 such that A ||f||2 ≤

  • k=1

|f, fk|2 ≤ B ||f||2, ∀f ∈ H. A and B are called frame bounds. Note:

  • Any orthonormal basis is a frame;
  • Example of a frame which is not a basis:

{e1, e1, e2, e3, . . . }, where {ek}∞

k=1 is an ONB.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 5 / 27

slide-9
SLIDE 9

The frame decomposition

If {fk}∞

k=1 is a frame, the frame operator

S : H → H, Sf =

  • f, fkfk

is well-defined, bounded, invertible, and self-adjoint.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 6 / 27

slide-10
SLIDE 10

The frame decomposition

If {fk}∞

k=1 is a frame, the frame operator

S : H → H, Sf =

  • f, fkfk

is well-defined, bounded, invertible, and self-adjoint. Theorem - the frame decomposition Let {fk}∞

k=1 be a frame with frame

  • perator S. Then

f =

  • k=1

f, S−1fkfk, ∀f ∈ H. It might be difficult to compute S−1 !

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 6 / 27

slide-11
SLIDE 11

The frame decomposition

If {fk}∞

k=1 is a frame, the frame operator

S : H → H, Sf =

  • f, fkfk

is well-defined, bounded, invertible, and self-adjoint. Theorem - the frame decomposition Let {fk}∞

k=1 be a frame with frame

  • perator S. Then

f =

  • k=1

f, S−1fkfk, ∀f ∈ H. It might be difficult to compute S−1 ! Important special case: If the frame {fk}∞

k=1 is tight, A = B, then S = A I and

f = 1 A

  • k=1

f, fkfk, ∀f ∈ H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 6 / 27

slide-12
SLIDE 12

General dual frames

A frame which is not a basis is said to be overcomplete. Theorem: Assume that {fk}∞

k=1 is an overcomplete frame. Then there exist

frames {gk}∞

k=1 = {S−1fk}∞ k=1

for which f =

  • k=1

f, gkfk =

  • k=1

f, S−1fkfk, ∀f ∈ H.

  • {gk}∞

k=1 is called a dual frame of {fk}∞ k=1.

  • The excess of a frame is the maximal number of elements that can be

removed such that the remaining set is still a frame. The excess equals dim N(T) - the dimension of the kernel of the synthesis operator.

  • When the excess is large, the set of dual frames is large.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 7 / 27

slide-13
SLIDE 13

General dual frames

Note: Let {fk}∞

k=1 be a Bessel sequence with pre-frame operator

T : H → ℓ2(N), T{ck}∞

k=1 = ∞

  • k=1

ckfk [T∗f = {f, fk}∞

k=1]

and {gk}∞

k=1 be a Bessel sequence with pre-frame operator

U : H → ℓ2(N), U{ck}∞

k=1 = ∞

  • k=1

ckgk [U∗f = {f, gk}∞

k=1]

Then {fk}∞

k=1 and {gk}∞ k=1 are dual frames if and only if

f =

  • k=1

f, gkfk, ∀f ∈ H, i.e., if and only if TU∗ = I.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 8 / 27

slide-14
SLIDE 14

Key tracks in frame theory:

  • Frames in finite-dimensional spaces;
  • Frames in general separable Hilbert spaces
  • Concrete frames in concrete Hilbert spaces:
  • Gabor frames in L2(R), L2(Rd);
  • Wavelet frames;
  • Shift-invariant systems, generalized shift-invariant (GSI) systems;
  • Shearlets, etc.
  • Frames in Banach spaces;
  • (GSI) Frames on LCA groups
  • Frames via integrable group representations, coorbit theory.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 9 / 27

slide-15
SLIDE 15

Key tracks in frame theory:

  • Frames in finite-dimensional spaces;
  • Frames in general separable Hilbert spaces
  • Concrete frames in concrete Hilbert spaces:
  • Gabor frames in L2(R), L2(Rd);
  • Wavelet frames;
  • Shift-invariant systems, generalized shift-invariant (GSI) systems;
  • Shearlets, etc.
  • Frames in Banach spaces;
  • (GSI) Frames on LCA groups
  • Frames via integrable group representations, coorbit theory.

Research Group HATA DTU (Harmonic Analysis - Theory and Applications , Technical University of Denmark), https://hata.compute.dtu.dk/

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 9 / 27

slide-16
SLIDE 16

Key tracks in frame theory:

  • Frames in finite-dimensional spaces;
  • Frames in general separable Hilbert spaces
  • Concrete frames in concrete Hilbert spaces:
  • Gabor frames in L2(R), L2(Rd);
  • Wavelet frames;
  • Shift-invariant systems, generalized shift-invariant (GSI) systems;
  • Shearlets, etc.
  • Frames in Banach spaces;
  • (GSI) Frames on LCA groups
  • Frames via integrable group representations, coorbit theory.

Research Group HATA DTU (Harmonic Analysis - Theory and Applications , Technical University of Denmark), https://hata.compute.dtu.dk/ An Introduction to frames and Riesz bases, 2.edition, Birkh¨ auser 2016

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 9 / 27

slide-17
SLIDE 17

Towards concrete frames - operators on L2(R)

Translation by a ∈ R: Ta : L2(R) → L2(R), (Taf)(x) = f(x − a). Modulation by b ∈ R : Eb : L2(R) → L2(R), (Ebf)(x) = e2πibxf(x). All these operators are unitary on L2(R).

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 10 / 27

slide-18
SLIDE 18

Towards concrete frames - operators on L2(R)

Translation by a ∈ R: Ta : L2(R) → L2(R), (Taf)(x) = f(x − a). Modulation by b ∈ R : Eb : L2(R) → L2(R), (Ebf)(x) = e2πibxf(x). All these operators are unitary on L2(R). Gabor systems in L2(R): have the form {e2πimbxg(x − na)}m,n∈Z for some g ∈ L2(R), a, b > 0. Short notation: {EmbTnag}m,n∈Z = {e2πimbxg(x − na)}m,n∈Z

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 10 / 27

slide-19
SLIDE 19

Gabor systems in L2(R)

It is known how to construct frames and dual pairs of frames with the Gabor structure {EmbTnag}m,n∈Z = {e2πimbxg(x − na)}m,n∈Z

  • Typical choices of g: B-splines or the Gaussian.
  • {EmbTnag}m,n∈Z can only be a frame if ab ≤ 1.
  • If {EmbTnag}m,n∈Z is a frame, then it is a basis if and only if ab = 1.
  • Gabor frames {EmbTnag}m,n∈Z are always linearly independent, and they

have infinite excess if ab < 1.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 11 / 27

slide-20
SLIDE 20

Dynamical Sampling

Introduced in papers by Aldroubi, Davis & Krishtal, and Aldroubi, Cabrelli, Molter & Tang. Further developed in papers by Aceska, Aldroubi, Cabrelli, C ¸ akmak, Kim, Molter, Paternostro, Petrosyan, Philipp. Let H denote a Hilbert space, and A a class of operators T : H → H. For T ∈ A and ϕ ∈ H, consider the iterated system {Tnϕ}∞

n=0 = {ϕ, Tϕ, T2ϕ · · · }.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 12 / 27

slide-21
SLIDE 21

Dynamical Sampling

Introduced in papers by Aldroubi, Davis & Krishtal, and Aldroubi, Cabrelli, Molter & Tang. Further developed in papers by Aceska, Aldroubi, Cabrelli, C ¸ akmak, Kim, Molter, Paternostro, Petrosyan, Philipp. Let H denote a Hilbert space, and A a class of operators T : H → H. For T ∈ A and ϕ ∈ H, consider the iterated system {Tnϕ}∞

n=0 = {ϕ, Tϕ, T2ϕ · · · }.

Key questions:

  • Can {Tnϕ}∞

n=0 be a basis for H for some T ∈ A, ϕ ∈ H?

  • Can {Tnϕ}∞

n=0 be a frame for H for some T ∈ A, ϕ ∈ H?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 12 / 27

slide-22
SLIDE 22

Dynamical Sampling and standard operator theory

Consider a bounded operator T : H → H. Recall:

  • A vector ϕ ∈ H is cyclic if span{Tnϕ}∞

n=0 = H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 13 / 27

slide-23
SLIDE 23

Dynamical Sampling and standard operator theory

Consider a bounded operator T : H → H. Recall:

  • A vector ϕ ∈ H is cyclic if span{Tnϕ}∞

n=0 = H.

This is much weaker than the condition that {Tnϕ}∞

n=0 is a frame.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 13 / 27

slide-24
SLIDE 24

Dynamical Sampling and standard operator theory

Consider a bounded operator T : H → H. Recall:

  • A vector ϕ ∈ H is cyclic if span{Tnϕ}∞

n=0 = H.

This is much weaker than the condition that {Tnϕ}∞

n=0 is a frame.

  • A vector ϕ ∈ H is hypercyclic if {Tnϕ}∞

n=0 is dense in H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 13 / 27

slide-25
SLIDE 25

Dynamical Sampling and standard operator theory

Consider a bounded operator T : H → H. Recall:

  • A vector ϕ ∈ H is cyclic if span{Tnϕ}∞

n=0 = H.

This is much weaker than the condition that {Tnϕ}∞

n=0 is a frame.

  • A vector ϕ ∈ H is hypercyclic if {Tnϕ}∞

n=0 is dense in H.

This is too strong in the frame context - it excludes that {Tnϕ}∞

n=0 is a

Bessel sequence.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 13 / 27

slide-26
SLIDE 26

Dynamical Sampling

Recall the key questions in dynamical sampling:

  • Can {Tnϕ}∞

n=0 be a basis for H for some T ∈ A, ϕ ∈ H?

  • Can {Tnϕ}∞

n=0 be a frame for H for some T ∈ A, ϕ ∈ H?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 14 / 27

slide-27
SLIDE 27

Dynamical Sampling

Recall the key questions in dynamical sampling:

  • Can {Tnϕ}∞

n=0 be a basis for H for some T ∈ A, ϕ ∈ H?

  • Can {Tnϕ}∞

n=0 be a frame for H for some T ∈ A, ϕ ∈ H?

Dual approach by C. & Marzieh Hasannasab:

  • When does a given frame {fk}∞

k=1 has a representation

{fk}∞

k=1 = {Tnϕ}∞ n=0

for some operator T : H → H?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 14 / 27

slide-28
SLIDE 28

Dynamical Sampling

Recall the key questions in dynamical sampling:

  • Can {Tnϕ}∞

n=0 be a basis for H for some T ∈ A, ϕ ∈ H?

  • Can {Tnϕ}∞

n=0 be a frame for H for some T ∈ A, ϕ ∈ H?

Dual approach by C. & Marzieh Hasannasab:

  • When does a given frame {fk}∞

k=1 has a representation

{fk}∞

k=1 = {Tnϕ}∞ n=0

for some operator T : H → H?

  • Under what conditions is such a representation possible with a bounded
  • perator T?
  • What are the properties of such frames?
  • What are the properties of the relevant operators T?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 14 / 27

slide-29
SLIDE 29

Dynamical sampling

Plan for the rest of the talk:

  • A sample of results from the literature
  • A characterization of the frames that have a representation

{fk}∞

k=1 = {Tnϕ}∞ n=0 for some operator T : span{fk}∞ k=1 → H.

  • Characterizations of the case where T can be chosen to be bounded.
  • Properties of {fk}∞

k=1 and properties of T.

  • Open problems.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 15 / 27

slide-30
SLIDE 30

Results from the literature

  • If T is normal, then {Tnϕ}∞

n=0 is not a basis (Aldroubi, Cabrelli,

C ¸ akmak, Molter, Petrosyan).

  • If T is unitary, then {Tnϕ}∞

n=0 is not a frame (Aldroubi, Petrosyan).

  • If T is compact, then {Tnϕ}∞

n=0 is not a frame (C., Hasannasab, Rashidi).

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 16 / 27

slide-31
SLIDE 31

A positive result from the literature

Example (Aldroubi, Cabrelli, Molter, Tang) Consider an operator T of the form T = ∞

k=1 λkPk, where Pk, k ∈ N, are rank 1 orthogonal projections

such that PjPk = 0, j = k, ∞

k=1 Pk = I, and |λk| < 1 for all k ∈ N.

  • There exists an ONB {ek}∞

k=1 such that

Tf =

  • k=1

λkf, ekek, f ∈ H.

  • Assume that {λk}∞

k=1 satisfies the Carleson condition, i.e.,

inf

k

  • j=k

|λj − λk| |1 − λjλk| > 0.

  • Letting ϕ := ∞

k=1

  • 1 − |λk|2ek, the family {Tnϕ}∞

n=0 is a frame for H.

  • Concrete case: λk = 1 − α−k for some α > 1.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 17 / 27

slide-32
SLIDE 32

The key question, formulated in ℓ2(N)

Motivated by all the talks on Hilbert/Hankel/Helson/Toeplitz matrices: Key question: Identify a class of infinite non-diagonalizable matrices A = (am,n)m,n≥1 and some vector ϕ ∈ ℓ2(N) such that the collection of vectors {Anϕ}∞

n=0 = {ϕ, Aϕ, A2ϕ, · · · }

form a frame for ℓ2(N).

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 18 / 27

slide-33
SLIDE 33

Existence of the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Proposition (Hasannasab & C., 2016): Consider a frame {fk}∞

k=1 for an

infinite-dimensional Hilbert space H. Then the following are equivalent:

  • There exists a linear operator T : span{fk}∞

k=1 → H such that

{fk}∞

k=1 = {Tnf1}∞ n=0,

  • The sequence {fk}∞

k=1 is linearly independent.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 19 / 27

slide-34
SLIDE 34

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

  • If {fk}∞

k=1 is a Bessel sequence, the synthesis operator

U : ℓ2(N) → H, U{ck}∞

k=1 := ∞

  • k=1

ckfk is well-defined and bounded.

  • The kernel of U is

N(U) =

  • {ck}∞

k=1 ∈ ℓ2(N)

  • k=1

ckfk = 0

  • .
  • Consider the right-shift operator T on ℓ2(N), defined by

T (c1, c2, · · · ) = (0, c1, c2, · · · ).

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 20 / 27

slide-35
SLIDE 35

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Theorem (Hasannasab & C., 2017): Consider a frame {fk}∞

k=1. Then the

following are equivalent: (i) The frame has a representation {fk}∞

k=1 = {Tnf1}∞ n=0 for some bounded

  • perator T : H → H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 21 / 27

slide-36
SLIDE 36

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Theorem (Hasannasab & C., 2017): Consider a frame {fk}∞

k=1. Then the

following are equivalent: (i) The frame has a representation {fk}∞

k=1 = {Tnf1}∞ n=0 for some bounded

  • perator T : H → H.

(ii) {fk}∞

k=1 is linearly independent and the kernel N(U) of the synthesis

  • perator U is invariant under the right-shift operator T .

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 21 / 27

slide-37
SLIDE 37

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Theorem (Hasannasab & C., 2017): Consider a frame {fk}∞

k=1. Then the

following are equivalent: (i) The frame has a representation {fk}∞

k=1 = {Tnf1}∞ n=0 for some bounded

  • perator T : H → H.

(ii) {fk}∞

k=1 is linearly independent and the kernel N(U) of the synthesis

  • perator U is invariant under the right-shift operator T .

(iii) For some dual frame {gk}∞

k=1 (and hence all),

fj+1 =

  • k=1

fj, gkfk+1, ∀j ∈ N.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 21 / 27

slide-38
SLIDE 38

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Theorem (Hasannasab & C., 2017): Consider a frame {fk}∞

k=1. Then the

following are equivalent: (i) The frame has a representation {fk}∞

k=1 = {Tnf1}∞ n=0 for some bounded

  • perator T : H → H.

(ii) {fk}∞

k=1 is linearly independent and the kernel N(U) of the synthesis

  • perator U is invariant under the right-shift operator T .

(iii) For some dual frame {gk}∞

k=1 (and hence all),

fj+1 =

  • k=1

fj, gkfk+1, ∀j ∈ N. In the affirmative case, letting {gk}∞

k=1 denote an arbitrary dual frame of

{fk}∞

k=1, the operator T has the form

Tf =

  • k=1

f, gkfk+1, ∀f ∈ H.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 21 / 27

slide-39
SLIDE 39

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Corollary: Any Riesz basis {fk}∞

k=1 has the form {fk}∞ k=1 = {Tnf1}∞ n=0 for

some bounded operator T : H → H. Surprisingly, the availability of such a representation fails for frames with finite excess: Proposition: Assume that {fk}∞

k=1 is a frame and {fk}∞ k=1 = {Tnf1}∞ n=0 for a

linear operator T. If {fk}∞

k=1 has finite and strictly positive excess, then T is

unbounded.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 22 / 27

slide-40
SLIDE 40

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Note: The properties of a frame with a representation {fk}∞

k=1 = {Tnf1}∞ n=0

(linear independence and infinite excess) match precisely the properties of Gabor frames {EmbTnag}m,n∈Z for ab < 1!

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 23 / 27

slide-41
SLIDE 41

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Note: The properties of a frame with a representation {fk}∞

k=1 = {Tnf1}∞ n=0

(linear independence and infinite excess) match precisely the properties of Gabor frames {EmbTnag}m,n∈Z for ab < 1! Can a Gabor frame {EmbTnag}m,n∈Z with ab < 1 be represented on this form with a bounded operator T?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 23 / 27

slide-42
SLIDE 42

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Note: The properties of a frame with a representation {fk}∞

k=1 = {Tnf1}∞ n=0

(linear independence and infinite excess) match precisely the properties of Gabor frames {EmbTnag}m,n∈Z for ab < 1! Can a Gabor frame {EmbTnag}m,n∈Z with ab < 1 be represented on this form with a bounded operator T? No!

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 23 / 27

slide-43
SLIDE 43

Boundedness of T in the representation {fk}∞

k=1 = {Tnf1}∞ n=0

Note: The properties of a frame with a representation {fk}∞

k=1 = {Tnf1}∞ n=0

(linear independence and infinite excess) match precisely the properties of Gabor frames {EmbTnag}m,n∈Z for ab < 1! Can a Gabor frame {EmbTnag}m,n∈Z with ab < 1 be represented on this form with a bounded operator T? No! Question: Can a Gabor frame {EmbTnag}m,n∈Z with ab < 1 be represented on the {EmbTnag}m,n∈Z = {anTng}∞

n=0

for some scalars an > 0 and a bounded operator T : L2(R) → L2(R)?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 23 / 27

slide-44
SLIDE 44

A comment on the indexing

Let us index the frames by Z instead of N0 :

  • Frames with a representation {fk}∞

k=1 = {Tkf0}k∈Z have a similar

characterization as for the index set N0.

  • The conditions for boundedness of T are similar.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 24 / 27

slide-45
SLIDE 45

A comment on the indexing

Let us index the frames by Z instead of N0 :

  • Frames with a representation {fk}∞

k=1 = {Tkf0}k∈Z have a similar

characterization as for the index set N0.

  • The conditions for boundedness of T are similar.
  • The standard operators in applied harmonic analysis easily leads to such

frames represented by bounded operators:

  • Translation: {Tkϕ}k∈Z = {(T1)kϕ}k∈Z;
  • Modulation: {Embϕ}m∈Z = {(Eb)mϕ}m∈Z;
  • Scaling: {Dajϕ}j∈Z = {(Da)jϕ}j∈Z;

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 24 / 27

slide-46
SLIDE 46

A comment on the indexing

Let us index the frames by Z instead of N0 :

  • Frames with a representation {fk}∞

k=1 = {Tkf0}k∈Z have a similar

characterization as for the index set N0.

  • The conditions for boundedness of T are similar.
  • The standard operators in applied harmonic analysis easily leads to such

frames represented by bounded operators:

  • Translation: {Tkϕ}k∈Z = {(T1)kϕ}k∈Z;
  • Modulation: {Embϕ}m∈Z = {(Eb)mϕ}m∈Z;
  • Scaling: {Dajϕ}j∈Z = {(Da)jϕ}j∈Z;

Question: Consider a Gabor frame {EmbTnag}m,n∈Z for L2(R) with ab < 1. Does the frame {EmbTnag}m,n∈Z have a representation {EmbTnag}m,n∈Z = {Tnϕ}∞

n=−∞

for a bounded operator T : L2(R) → L2(R) and some ϕ ∈ L2(R)?

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 24 / 27

slide-47
SLIDE 47

Frames of the form {fk}∞

k=1 = {Tnf1}∞ n=0 are special!

Theorem (C., Hasannasab, Rashidi, 2017) Assume that {Tnϕ}∞

n=0 is an

  • vercomplete frame for some ϕ ∈ H and some bounded operator T : H → H.

Then the following hold: (i) The image chain for the operator T has finite length q(T). (ii) If N ∈ N, then TNϕ ∈ span{Tnϕ}∞

n=N+1 ⇔ N ≥ q(T).

For any N ≥ q(T), let V := span{Tnϕ}∞

n=N. Then the following hold:

(iii) The space V is independent of N and has finite codimension. (iv) The sequence {Tnϕ}∞

n=N+ℓ is a frame for V for all ℓ ∈ N0.

(v) V is invariant under T, and T : V → V is surjective. (vi) If the null chain of T has finite length then T : V → V is injective; in particular this is the case if T is normal.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 25 / 27

slide-48
SLIDE 48

A way to achieve boundedness - multiple operators

Theorem (Hasannasab & C., 2016) Consider a frame {fk}∞

k=1 which is

norm-bounded below. Then there is a finite collection of vectors from {fk}∞

k=1, to be called ϕ1, . . . , ϕJ, and corresponding bounded operators

Tj : H → H, such that {fk}∞

k=1 = J

  • j=1

{Tn

j ϕj}∞ n=0.

Remark: The assumption that {fk}∞

k=1 is norm-bounded below can not be

removed.

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 26 / 27

slide-49
SLIDE 49

References

Aceska, R., and Yeon Hyang Kim, Y. H.: Scalability of frames generated by dynamical operators. arXiv preprint arXiv:1608.05622 (2016). Aldroubi, A., Cabrelli, C., Molter, U., and Tang, S.: Dynamical sampling. Appl.

  • Harm. Anal. Appl. 42 no. 3 (2017), 378-401.

Aldroubi, A., Cabrelli, C., C ¸ akmak, A. F., Molter, U., and Petrosyan, A.: Iterative actions of normal operators. J. Funct. Anal. 272 no. 3 (2017), Cabrelli, C., Molter, U, Paternostro, V., and Philipp, F.: Dynamical sampling on finite index sets. Preprint, 2017. Christensen, O., and Hasannasab, M.: Frame properties of systems arising via iterative actions of operators Preprint, 2016. Christensen, O., and Hasannasab, M.: Operator representations of frames: boundedness, duality, and stability. Int. Eq. & Op. Theory, to appear. Christensen, O., and Hasannasab, M.: An open problem concerning operator representations of frames. Preprint, 2017

(DTU ) IWOTA, Chemnitz 2017 August 17, 2017 27 / 27