Abstract
It is well known that there are two approaches applicable in constructing frames starting from one fixed frame. One is based on \(l^{2}\)-operator portraits by which, using a suitable bounded linear operator on \(l^{2}\), one can construct an arbitrary frame from one fixed frame. The other is based on perturbation that allows suitable perturbing a frame leaving a frame. The study of Hilbert–Schmidt frames (HS-frames) has interested some mathematicians in recent years. This paper addresses \(l^{2}\)-operator portraits and perturbations in the setting of HS-frames. We prove that the portrait of a HS-frame under a bounded invertible operator on \(l^{2}\) is still a HS-frame; present a sufficient condition on bounded operators on \(l^{2}\) which transform an \(l^{2}\)-decomposable HS-frame into another HS-frame (HS-Riesz basis, HS-frame sequence and HS-Riesz sequence); and prove that suitable perturbing a HS-frame sequence (HS-Riesz sequence) leaves a HS-frame sequence (HS-Riesz sequence). Finally, using these results we recover some conclusions on frames.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The concept of frame was proposed by Duffin and Schaeffer [18] in 1952 to study nonharmonic Fourier series. It was reintroduced by Daubechies, Grossmann and Meyer [15] in 1986. Since then, the frame theory has developed greatly and attracted more attention and research. We refer to [12, 18, 23, 39] for basic results on frames. Recall that a countable sequence \(f=\{f_{j}\}_{j\in J}\) in a separable Hilbert space \({\mathcal {H}}\) is called a frame for \({\mathcal {H}}\) if there exist constants \(0<A\le B<\infty \) such that
It is called a Bessel sequence with bound B if the right-hand side inequality of (1.1) holds. A frame for \({\mathcal {H}}\) is called a Riesz basis for \({\mathcal {H}}\) if it ceases to a frame when removing an arbitrary element. A sequence \(f=\{f_{j}\}_{j\in J}\) in \({\mathcal {H}}\) is called a frame sequence (Riesz sequence) if it is a frame (Riesz basis) for its closed linear span. For a Bessel sequence \(f=\{f_{j}\}_{j\in J}\) in \({\mathcal {H}}\), we always denote by \(T_{f}\), \(T_{f}^{*}\) and \(S_{f}\) its synthesis operator \(T_{f}:l^{2}\rightarrow {\mathcal {H}}\), analysis operator \(T_{f}^{*}:{\mathcal {H}}\rightarrow l^{2} \) and frame operator \(S_{f}: {\mathcal {H}}\rightarrow {\mathcal {H}}\), respectively, i.e.,
In last decades, various generalizations of frame have been put forward for special purposes. The concept of Hilbert–Schmidt frame (HS-frame for simplicity) was introduced by Sadeghi and Arefijamaal in [34] which is more general than that of all existing frames such as g-frame, frame of subspaces, pseudo-frame, fusion frame, outer frame, bounded quasi-projector and a class of time-frequency localization operators (see [3, 5, 7, 8, 17, 20, 29, 35, 36]).
Given two Hilbert spaces \({\mathcal {H}}_{1}\) and \({\mathcal {H}}_{2}\), we denote by \({\mathcal {B}}({\mathcal {H}}_{1},\,{\mathcal {H}}_{2})\) the set of all bounded linear operators from \({\mathcal {H}}_{1}\) to \({\mathcal {H}}_{2}\), and write \({\mathcal {B}}({\mathcal {H}}_{1})={\mathcal {B}}({\mathcal {H}}_{1},\,{\mathcal {H}}_{1})\). For a sequence \(\{\Theta _{j}\}_{j\in J}\) in \({\mathcal {B}}({\mathcal {H}}_{1},\,{\mathcal {H}}_{2})\), \(\Theta :=\sum \limits _{j\in J}\Theta _{j}\) means that, for each \(h\in {\mathcal {H}}_{1}\), \(\Theta h=\sum \limits _{j\in J}\Theta _{j}h\) with unconditional convergence. Throughout this paper, \({\mathcal {H}}\) and \({\mathcal {K}}\) are separable Hilbert spaces, J and K are countable index sets. Given \(U\in {\mathcal {B}}({\mathcal {K}})\) and an orthonormal basis \(\{e_{k}\}_{k\in K}\) for \({\mathcal {K}}\), define
It is well known that, \(\Vert U\Vert _{2}\) is independent of the choice of \(\{e_{k}\}_{k\in K}\). We call the operator U a Hilbert–Schmidt operator if \(\Vert U\Vert _{2}<\infty \). Let \({\mathcal {C}}_{2}({\mathcal {K}})\) (\({\mathcal {C}}_{2}={\mathcal {C}}_{2}({\mathcal {K}})\) for simplicity) denote the Hilbert space consisting of all Hilbert–Schmidt operators on \({\mathcal {K}}\) endowed with the inner product
Define \(\oplus {\mathcal {C}}_{2}\) as the Hilbert space:
with the inner product given by \(\langle {\mathcal {A}},{\mathcal {B}}\rangle =\sum \limits _{j\in J}\langle {\mathcal {A}}_{j},{\mathcal {B}}_{j}\rangle _{\tau }\) for \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\) and \({\mathcal {B}}=\{{\mathcal {B}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\). Write
Definition 1.1
( [16, 32, 34]) A sequence \(\{{\mathcal {G}}_{j}\}_{j\in J}\) in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) is said to be a Hilbert–Schmidt frame or simply a HS-frame for \({\mathcal {H}}\) (a HS-frame sequence in \({\mathcal {H}}\)) with respect to \({\mathcal {K}}\), if there exist constants \(0<A\le B<\infty \) such that
where A and B are called HS-frame bounds. It is called a Parseval HS-frame if \(A=B=1\) in (1.2). It is called a HS-Bessel sequence with bound B if the right-hand side inequality of (1.2) holds. It is called HS-complete if \(\{h\in {\mathcal {H}}: {\mathcal {G}}_{j}h=0{\hbox { for }} j\in J\}=\{0\}\). It is called a HS-Riesz sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if there exist constants \(0<A\le B<\infty \) such that
where A and B are called HS-Riesz bounds. It is called a HS-Riesz basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if it is a HS-Riesz sequence and HS-complete. It is called a HS-orthonormal system in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if \({\mathcal {G}}_{j}{\mathcal {G}}_{k}^{*}=\delta _{j,\,k}I\) for \(j,k\in J\). And it is called a HS-orthonormal basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if it is a HS-orthonormal system and HS-complete. Herein and after, we use I to denote the identity operator regardless of its underlying space.
Given a HS-Bessel sequence \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), we always denote by \(T_{{\mathcal {G}}}\), \(T_{{\mathcal {G}}}^{*}\) and \(S_{{\mathcal {G}}}\) its synthesis operator \(T_{{\mathcal {G}}}:\oplus {\mathcal {C}}_{2}\rightarrow {\mathcal {H}}\), analysis operator \(T_{{\mathcal {G}}}^{*}:{\mathcal {H}}\rightarrow \oplus {\mathcal {C}}_{2}\) and frame operator \(S_{{\mathcal {G}}}: {\mathcal {H}}\rightarrow {\mathcal {H}}\), respectively, i.e.,
For more details on HS-frames, we refer the reader to [4, 16, 28, 30,31,32,33,34].
Given a matrix \(M=(m_{i,\,j})_{i,\,j\in \mathbb {N}}\) and two sequences \(\{f_{i}\}_{i=1}^{\infty }\), \(\{g_{i}\}_{i=1}^{\infty }\) in a Hilbert space \({\mathcal {H}}\), \((g_{1},g_{2},\cdots )=(f_{1},f_{2},\cdots )M\) means that, for each \(j\in \mathbb {N}\), \(\sum \limits _{i=1}^{\infty }m_{i,\,j}f_{i}\) is convergent and \(g_{j}=\sum \limits _{i=1}^{\infty }m_{i,\,j}f_{i}\). It is well known that permutating a frame for \({\mathcal {H}}\) leaves a frame for \({\mathcal {H}}\). And by a standard argument, given a frame for \({\mathcal {H}}\), its image under an arbitrary bounded linear bijection on \({\mathcal {H}}\) is still a frame for \({\mathcal {H}}\). But the latter excludes permutation. For example, if \(\{e_{1}, e_{2},\cdots \}\) is an orthonormal basis for \({\mathcal {H}}\), then
and its permutation
are both Parseval frames for \({\mathcal {H}}\). However, there exists no \(T\in {\mathcal {B}}({\mathcal {H}})\) such that \(g_{i}=Tf_{i}\) for all \(i\in \mathbb {N}\). Indeed,
for an arbitrary \(T\in {\mathcal {B}}({\mathcal {H}})\). Fortunately,
where \(M= \begin{pmatrix} M_{1} &{} 0 \\ 0 &{} I \end{pmatrix}\), \(M_{1}= \begin{pmatrix} 0 &{} c \\ 1 &{} 0 \end{pmatrix}\), c is an arbitrary constant. Obviously, M defines a bounded linear operator on \(l^{2}\), and M is invertible if and only if \(c\ne 0\). More general, \(l^{2}\)-operator images of a frame contains its all permutated versions. Aldroubi in [2] studied \(l^{2}\)-operator portraits of frames. The author characterized all bounded linear operators on \(l^{2}\) that transform frames into other frames for \({\mathcal {H}}\). Specifically, given a frame \(\{f_{i}\}_{i=1}^{\infty }\) for \({\mathcal {H}}\), the author characterized all bounded linear operators \({\widetilde{M}}\) on \(l^{2}\) such that \(\{g_{i}\}_{i=1}^{\infty }\) defined below is a frame for \({\mathcal {H}}\):
This method was also applied to g-frame setting in Guo [22]. Observe that \({\widetilde{M}}\) in (1.6) is linear. We can think that \(\{g_{i}\}_{i=1}^{\infty }\) is a linear perturbation of \(\{f_{i}\}_{i=1}^{\infty }\). On the other hand, nonlinear perturbation (usually so-called perturbation) of frames was extensively studied ([6, 9,10,11, 13, 14, 19, 24, 25]). It can be considered as an extension of the classical Paley–Wiener Theorem in [39] which reads: Let \(\{x_{i}\}_{i=1}^{\infty }\) be a basis for a Banach space X, then an arbitrary sequence \(\{y_{i}\}_{i=1}^{\infty }\) in X satisfying the following condition is a basis for X: there exists a constant \(0\le \lambda <1\) such that
Recall from [6, 9, 11, 14] that the Paley–Wiener type theorem on frames says that, given a frame (frame sequence) \(\{f_{i}\}_{i=1}^{\infty }\) in a Hilbert space \({\mathcal {H}}\), a sequence \(\{g_{i}\}_{i=1}^{\infty }\) in \({\mathcal {H}}\) is a frame (frame sequence) in \({\mathcal {H}}\) if there exist constants \(\lambda _{1},\,\lambda _{2},\,\mu \ge 0\) such that
where \(0\le \lambda _{2}<1\), and \(\lambda _{1}\), \(\mu \) satisfy some technical conditions. It was generalized to g-frame, K-g-frame and HS-frame settings ( [1, 21, 27, 32, 36, 38]). By [6, Theorem 2], if \(\{f_{i}\}_{i=1}^{\infty }\) is a frame with lower bound A and \(\max (\lambda _{1}+\frac{\mu }{\sqrt{A}},\,\lambda _{2})<1\) in (1.7), then \(\{g_{i}\}_{i=1}^{\infty }\) is a frame.
Suppose \(\{f_{i}\}_{i=1}^{\infty }\) and \(\{g_{i}\}_{i=1}^{\infty }\) are as in (1.4) and (1.5). Then \(\left\| \sum \limits _{i=1}^{\infty }c_{i}(f_{i}-g_{i})\right\| =1\) if \(c_{2}=1\) and \(c_{i}=0\) for \(i\ne 2\). Thus (1.7) cannot holds for arbitrary positive constants \(\lambda _{1}\), \(\lambda _{2}\) and \(\mu \) satisfying \(\max (\lambda _{1}+\mu ,\,\lambda _{2})<1\). This suggests to some extent that the perturbation technique cannot cover \(l^{2}\)-operator portrait method.
Motivated by the above works, this paper addresses \(l^{2}-\)operator portraits and perturbations under the setting of HS-frame sequences. It is well known that an orthonormal basis for a Hilbert space may represent its all elements. However, Example 2.1 below shows that a HS-orthonormal basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) need not represent all elements in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\), not to mention a HS-frame. So, to study frame properties of \(l^{2}\)-operator portraits of HS-frames (HS-frame sequences), we naturally restrict us to \(l^{2}\)-decomposable HS-operator sequences.
Definition 1.2
Given a HS-orthonormal basis \({\mathcal {E}}=\{{\mathcal {E}}_{j}\}_{j\in J}\) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), a sequence \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) is said to be \(l^{2}(J)\)-decomposable with respect to \({\mathcal {E}}\) is there exists \(U=(u_{k,\,j})_{k,\,j\in J}\in {\mathcal {B}}(l^{2}(J))\) such that
Remark 1.1
By Lemmas 2.2 and 2.4 below, \({\mathcal {G}}\) is well defined and is a HS-Bessel sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
From [11, 14, 37], the “gap between two subspaces” is important in studying perturbation of frame sequences. It is also useful in this paper. Let V, W be two closed subspaces of \({\mathcal {H}}\). When \(V\ne \{0\}\), the gap from V to W is defined by
Define \(\delta (V,\,W)=0\) when \(V=\{0\}\). From [26] we know that \(\delta (V,\,W)=\delta (W^{\perp },V^{\perp })\), \(0\le \delta (V,\,W)\le 1\), and that \(\delta (V,\,W)\) need not equal to \(\delta (W,\,V)\). Recall from [16, Lemma 2.5] and [32, Theorem 4.3] that a HS-Bessel sequence is a HS-frame sequence if and only if its synthesis operator has closed range, and that suitably perturbing a HS-frame \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) leaves a HS-frame \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\), where \(\mathrm{range}(T_{{\mathcal {F}}})\subset \mathrm{range}(T_{{\mathcal {G}}})\) automatically holds. But we need not have \(\mathrm{range}(T_{{\mathcal {F}}})\subset \mathrm{range}(T_{{\mathcal {G}}})\) if \({\mathcal {G}}\) is a HS-frame sequence but not a HS-frame. This suggests us to study stable perturbation of HS-frame sequences with the help of gap between subspaces associated with \(T_{{\mathcal {F}}}\) and \(T_{{\mathcal {G}}}\).
The rest of this paper is organized as follows. In Sect. 2, we prove that an arbitrary bounded invertible operator on \(l^{2}(J)\) transforms a HS-frame into another HS-frame (see Theorem 2.1); and present a sufficient condition on bounded operators on \(l^{2}(J)\) which transform an \(l^{2}(J)\)-decomposable HS-frame into another HS-frame (HS-Riesz basis, HS-frame sequence and HS-Riesz sequence) (see Theorem 2.2). In Sect. 3, we prove that suitably perturbing a HS-frame sequence (HS-Riesz sequence) leaves a HS-frame sequence (HS-Riesz sequence). In Sect. 4, we give a remark that shows our results can recover some known conclusions on frames.
Before proceeding, we introduce some notations and notions for later use. For a HS-frame \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), another HS-frame \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is called an alternate dual HS-frame of \({\mathcal {G}}\) if
In particular, \(\{{\mathcal {G}}_{j}S_{{\mathcal {G}}}^{-1}\}_{j\in J}\) is a dual of \({\mathcal {G}}\) by [34, Remark 3.8], which is called the canonical dual HS-frame of \({\mathcal {G}}\) in [4]. If \({\mathcal {G}}\) is a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), by a standard argument, \(\overline{\mathrm{span}}\{({\mathcal {G}}_{j}S_{{\mathcal {G}}}^{-1})^{*}{\mathcal {A}}_{j}:{\mathcal {A}}_{j}\in {\mathcal {C}}_{2}\}_{j\in J}=\overline{\mathrm{span}}\{{\mathcal {G}}_{j}^{*}{\mathcal {A}}_{j}:{\mathcal {A}}_{j}\in {\mathcal {C}}_{2}\}_{j\in J}\). Thus \(\{{\mathcal {G}}_{j}S_{{\mathcal {G}}}^{-1}\}_{j\in J}\) is also a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with HS-frame bounds \(B^{-1}\) and \(A^{-1}\). In this paper, \(P_{M}\) always denotes the orthogonal projection from an appropriate space to the closed subspace M. For a bounded linear operator U, we denote by \(\ker (U)\), \(\mathrm{{range}}(U)\) and \(U^{*}\) its kernel, range and adjoint operator, respectively, and by \(U^{\dagger }\) its pseudo-inverse if U is of closed range in addition. For a matrix U, we denote by \({\overline{U}}\) and \(U^{*}\) its conjugate and conjugate transpose, respectively.
2 \(l^{2}(J)\)-operator Portraits
This section focuses on \(l^{2}(J)\)-operator portraits of HS-frames (HS-frame sequences). We prove that “HS-frame sequence” property is preserved under bounded invertible operators on \(l^{2}(J)\), and present a sufficient condition on bounded operators on \(l^{2}(J)\) which transform an \(l^{2}(J)\)-decomposable HS-frame into another HS-frame (HS-Riesz basis, HS-frame sequence and HS-Riesz sequence). For this purpose, we need to give some lemmas. Before this, we first given an example that a HS-operator does not represented by a HS-orthonormal basis. This explains why we introduce Definition 1.2.
Example 2.1
Let \({\mathcal {H}}\) be a Hilbert space with \(\{e_{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) being an orthonormal basis for \({\mathcal {H}}\), and \({\mathcal {K}}=l^{2}(\mathbb {N})\) with \(\{w_{k}\}_{k\in \mathbb {N}}\) be the standard orthonormal basis that is, for every k, \(w_{k}\) is 1 in the \(k-th\) position and 0 in other positions. Define \(\{\xi _{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) by
where \(U=\left( \begin{array}{cc} U_{1} &{} 0 \\ 0 &{} I \end{array} \right) \), \(U_{1}=\left( \begin{array}{cc} \frac{1}{\sqrt{2}} &{} \frac{1}{\sqrt{2}} \\ \\ -\frac{1}{\sqrt{2}} &{} \frac{1}{\sqrt{2}} \end{array} \right) \). Then \(\{\xi _{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) is an orthonormal basis for \({\mathcal {C}}_{2}\) by [30, Theorem2.1], and define the operator sequence \(\{{\mathcal {G}}_{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) by
Then \(\{{\mathcal {G}}_{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) is a Parseval HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\). By [31, Theorem 2.1], there exists a Hilbert space \(\widetilde{{\mathcal {H}}}={\mathcal {H}}\bigoplus {\mathcal {M}}\supset {\mathcal {H}}\) and a HS-orthonormal basis \(\{\widetilde{{\mathcal {G}}}_{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\) for \(\widetilde{{\mathcal {H}}}\) with respect to \({\mathcal {K}}\) such that
Define \({\widetilde{\Lambda }}\in {\mathcal {B}}(\widetilde{{\mathcal {H}}},\,{\mathcal {C}}_{2})\) by
where \({\widetilde{\xi }}\) is defined by
Then we claim that there exists no \(c=\{c_{(i,\,j)}\}_{(i,\,j)\in \mathbb {N}^{2}}\in l^{2}\) such that
We prove it by contradiction. Suppose (2.6) holds for some \(c\in l^{2}\). Observe that
by (2.1)–(2.3). It follows that
However, using (2.4) and (2.5) we have
It contradicts (2.7).
The following lemma reveals the connection between the pseudo-inverse of the product of two matrices and their pseudo-inverses.
Lemma 2.1
Let U be an \(l\times n\) matrix and V be an \(n\times m\) matrix. Suppose \(\mathrm{{range}}(V)\subset \mathrm{{range}}(U^{*})\). Then \((UV)^{\dagger }x=V^{\dagger }U^{\dagger }x\) for \(x\in \mathrm{{range}}(UV)\). Furthermore, \(\Vert (UV)^{\dagger }\Vert \le \Vert U^{\dagger }\Vert \Vert V^{\dagger }\Vert \).
Proof
Arbitrarily fix \(x\in \mathrm{{range}}(UV)\). Then
by the definition of pseudo-inverse. This implies that
due to \(\ker (V)\subset \ker (UV)\). Since \(\mathrm{{range}}(V)\subset \mathrm{{range}}(U^{*})=\ker (U)^{\perp }\), we have
Observe that \(\mathrm{range}(U^{\dagger })=\ker (U)^{\perp }\) and \(U^{\dagger }U=P_{\mathrm{range}(U^{\dagger })}\) by [12, Lemma 2.5.2]. It follows that \(U^{\dagger }U=P_{\ker (U)^{\perp }}\). Similarly, \(V^{\dagger }V=P_{\ker (V)^{\perp }}\). This implies that
Next we prove \(\Vert (UV)^{\dagger }\Vert \le \Vert U^{\dagger }\Vert \Vert V^{\dagger }\Vert \). Arbitrarily fix \(y=y_{1}+y_{2}\in \mathbb {C}^{l}\) with \(y_{1}\in \mathrm{{range}}(UV)\) and \(y_{2}\in \mathrm{{range}}(UV)^{\perp }\). Then \((UV)^{\dagger }y=(UV)^{\dagger }y_{1}\) by the definition of pseudo-inverse. Replacing x in (2.11) by \(y_{1}\) leads to
This leads to \(\Vert (UV)^{\dagger }\Vert \le \Vert U^{\dagger }\Vert \Vert V^{\dagger }\Vert \) by the arbitrariness of y. \(\square \)
The following lemma shows that \(\sum \limits _{j\in J}c_{j}{\mathcal {G}}_{j}\) is well defined and belongs to \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) whenever \(c\in l^{2}(J)\) and \(\{{\mathcal {G}}_{j}\}_{j\in J}\) is a HS-Bessel sequence.
Lemma 2.2
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-Bessel sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bound B. Then, for all \(c=\{c_{j}\}_{j\in J}\in l^{2}(J)\), \(\Theta =\sum \limits _{j\in J}c_{j}{\mathcal {G}}_{j}\) is well defined, and \(\Theta \in {\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\).
Proof
Arbitrarily fix \(h\in {\mathcal {H}}\). Then
for \(c\in l_{0}(J)\), where B is the HS-Bessel bound of \({\mathcal {G}}\). This leads to the lemma due to \(l_{0}(J)\) being dense in \(l^{2}(J)\). \(\square \)
The following lemma shows that a bounded linear operator on \(l^{2}(J)\) induces a bounded linear operator on \(\oplus {\mathcal {C}}_{2}\).
Lemma 2.3
Suppose \(U=(u_{k,\,j})_{k,\,j\in J}\in {\mathcal {B}}(l^{2}(J))\). Define \(\Lambda :\oplus {\mathcal {C}}_{2}\rightarrow \oplus {\mathcal {C}}_{2}\) by
Then \(\Lambda \) is bounded on \(\oplus {\mathcal {C}}_{2}.\)
Proof
First we claim that \(\sum \limits _{j\in J}c_{j}{\mathcal {A}}_{j}\) is well defined if \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\) and \(c=\{c_{j}\}_{j\in J}\in l^{2}\). Indeed, for an arbitrary finite subset \({\widetilde{J}}\) of J,
This implies that \(\sum \limits _{j\in J}c_{j}{\mathcal {A}}_{j}\) converges unconditionally. Since \(U\in {\mathcal {B}}(l^{2}(J))\), we have \(U^{*}\in {\mathcal {B}}(l^{2}(J))\). It follows that each column \(\{\overline{u_{k,\,j}}\}_{j\in J}\) of \(U^{*}\) belongs to \(l^{2}\). Thus, for each \(k\in J\), \(\sum \limits _{j\in J}\overline{u_{k,\,j}}{\mathcal {A}}_{j}\) is well defined. Next we prove that \(\Lambda \) is a bounded operator on \(\oplus {\mathcal {C}}_{2}.\)
Let \(\{\xi _{m}\}_{m\in M}\) be an orthonormal basis for \({\mathcal {C}}_{2}\). Arbitrarily fix \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0} {\mathcal {C}}_{2}\), then there exists a sequence \(\{c_{m}^{(j)}\}_{j\in J,\,m\in M}\in l^{2}\) such that \({\mathcal {A}}_{j}=\sum \limits _{m\in M}c_{m}^{(j)}\xi _{m}\) for \(j\in J\), where \(\{c_{m}^{(j)}\}_{j\in J}\in l_{0}\) for each \(m\in M\). It follows that
and
Thus
Therefore, \(\Lambda \) is bounded on \(\oplus {\mathcal {C}}_{2}\) due to \(\oplus _{0}{\mathcal {C}}_{2}\) being dense in \(\oplus {\mathcal {C}}_{2}\). \(\square \)
The following lemma shows that every \(l^{2}(J)\)-decomposable sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) is a HS-Bessel sequence.
Lemma 2.4
An arbitrary \(l^{2}(J)\)-decomposable sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) is a HS-Bessel sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
Proof
Suppose \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) is an \(l^{2}(J)\)-decomposable sequence with respect to a HS-orthonormal basis \({\mathcal {E}}=\{{\mathcal {E}}_{j}\}_{j\in J}\), and
where \(U=(u_{k,\,j})_{k,\,j\in J}\in {\mathcal {B}}(l^{2}(J))\). Then
Next we prove \({\mathcal {G}}\) is a HS-Bessel sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\). Arbitrarily fix \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0} {\mathcal {C}}_{2}\). Then
by (2.12). Let \(\{\xi _{m}\}_{m\in M}\) be an orthonormal basis for \({\mathcal {C}}_{2}\). Then there exists a sequence \(\{c_{m}^{(j)}\}_{j\in J,\,m\in M}\in l^{2}\) such that \({\mathcal {A}}_{j}=\sum \limits _{m\in M}c_{m}^{(j)}\xi _{m}\) for \(j\in J\), where \(\{c_{m}^{(j)}\}_{j\in J}\in l_{0}\) for each \(m\in M\). It follows that
Observe that
by [16, Lemma 2.1] due to \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0}{\mathcal {C}}_{2}\). Collecting (2.13)-(2.15), we obtain that
This implies that \({\mathcal {G}}\) is a HS-Bessel sequence due to \(\oplus _{0}{\mathcal {C}}_{2}\) being dense in \(\oplus {\mathcal {C}}_{2}\). \(\square \)
The following lemma focuses on HS-frame property of \(l^{2}(J)\)-decomposable HS-operator sequences.
Lemma 2.5
Given a HS-orthonormal basis \({\mathcal {E}}=\{{\mathcal {E}}_{j}\}_{j\in J}\) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be an \(l^{2}(J)\)-decomposable sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) with respect to \({\mathcal {E}}\) associated with \(U=(u_{k,\,j})_{k,\,j\in J}\in {\mathcal {B}}\left( l^{2}(J)\right) \), i.e.,
Then \({\mathcal {G}}\) is a HS-frame (HS-Riesz basis, HS-frame sequence, HS-Riesz sequence) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if U is bounded and surjective (bijective, of closed range, injective and of closed range) on \(l^{2} (J)\).
Proof
By Lemma 2.4, \({\mathcal {G}}\) is a HS-Bessel sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\). By a simple computation,
where \(\Lambda \) is as in Lemma 2.3. Observe that \(\Lambda \) is bounded by Lemma 2.3, and \(T_{{\mathcal {G}}}\), \(T_{{\mathcal {E}}}\) are also bounded. It follows that
due to \(\oplus _{0} {\mathcal {C}}_{2}\) being dense in \(\oplus {\mathcal {C}}_{2}\). By [32, Proposition 2.13 and Theorem 2.15] and [16, Lemmas 2.5 and 2.6], we need to prove that \(T_{{\mathcal {G}}}\) is surjective (bijective, of closed range, injective and of closed range) when U is surjective (bijective, of closed range, injective and of closed range). Since \(T_{{\mathcal {E}}}\) is unitary, by (2.16), \(T_{{\mathcal {G}}}\) is surjective (bijective, of closed range, injective and of closed range) if and only if \(\Lambda \) is surjective (bijective, of closed range, injective and of closed range). So it is enough to prove that \(\Lambda \) is surjective (injective, of closed range) if U is surjective (injective, of closed range). Let \(\{\xi _{m}\}_{m\in M}\) be an orthonormal basis for \({\mathcal {C}}_{2}\). Then
and
Case 1. U is surjective.
Arbitrarily fix \({\mathcal {B}}=\{{\mathcal {B}}_{k}\}_{k\in J}\in \oplus {\mathcal {C}}_{2}\). Then
Define \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\) by
where
Then
by (2.19). So \({\mathcal {A}}\) is well defined, and
By (2.17), (2.18) and a simple computation, we have
with
Observing \({\overline{U}}\,{\overline{U}}^{\dagger }=I\) leads to \(\Lambda {\mathcal {A}}={\mathcal {B}}\). Thus \(\Lambda \) is surjective.
Case 2. U is injective.
Suppose \(\Lambda {\mathcal {A}}=0\) for some \({\mathcal {A}}\in \oplus {\mathcal {C}}_{2}\). Then
and
by (2.17) and (2.18). This leads to
due to \({\overline{U}}\) being injective. Equivalently, \({\mathcal {A}}=0\). Thus \(\Lambda \) is injective.
Case 3. U is of closed range.
Suppose there exists a sequence \(\{{\mathcal {A}}^{(n)}\}_{n\in \mathbb {N}}\subset \oplus {\mathcal {C}}_{2}\) with \({\mathcal {A}}^{(n)}=\{{\mathcal {A}}_{j}^{(n)}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\) for \(n\in \mathbb {N}\) such that
Then
by (2.17), (2.18) due to \(\{\xi _{m}\}_{m\in M}\) being an orthonormal basis for \({\mathcal {C}}_{2}\). Collecting (2.20) and (2.21) leads to
Since U is of closed range, so is \({\overline{U}}\). Then, for each \(m\in M\), there exists a \(\{c_{m}^{(j)}\}_{j\in J}\in \ker ({\overline{U}})^{\perp }\) such that
equivalently,
Define \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}\) by
Observe that \({\overline{U}}\,{\overline{U}}^{\dagger }=P_{\mathrm{{range}}({\overline{U}})}\). Similarly to Case 1, we have \({\mathcal {A}}\in \oplus {\mathcal {C}}_{2}\) and \(\Lambda {\mathcal {A}}={\mathcal {B}}\). Thus \(\Lambda \) has closed range. The proof is completed. \(\square \)
The following theorem shows that a bounded invertible operator on \(l^{2}(J)\) transforms a HS-frame into another HS-frame.
Theorem 2.1
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\). Suppose \(\Gamma =(\gamma _{k,\,j})_{k,\,j\in J}\) is a bounded operator on \(l^{2}(J)\). Define \({\mathcal {F}}_{j}=\sum \limits _{k\in J}\gamma _{k,\,j}{\mathcal {G}}_{k}\) for \(k\in J\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) when \(\Gamma \) is invertible on \(l^{2}(J)\).
Proof
By Lemma 2.2, \({\mathcal {F}}_{j}\) is well defined and \({\mathcal {F}}_{j}\in {\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) for \(j\in J\). Let \(\{\xi _{m}\}_{m\in M}\) be an orthonormal basis for \({\mathcal {C}}_{2}\). Then
for \(h\in {\mathcal {H}}\). Also observe that \(\sum \limits _{k\in J}\left| \langle {\mathcal {G}}_{k}h,\,\xi _{m}\rangle _{\tau }\right| ^{2}\le \sum \limits _{k\in J}\Vert {\mathcal {G}}_{k}h\Vert _{2}^{2}<\infty .\) It follows that
When \(\Gamma \) is invertible, there exist constants \(0<c_{1}\le c_{2}<\infty \) such that
This together with (2.22) leads to
Thus \({\mathcal {F}}\) is a HS-frame due to \({\mathcal {G}}\) being a HS-frame. The proof is completed. \(\square \)
Remark 2.1
In Theorem 2.1, if \({\mathcal {G}}\) is a HS-frame sequence, then replacing \({\mathcal {H}}\) by \(\overline{\mathrm{span}}\{{\mathcal {G}}_{j}^{*}{\mathcal {A}}_{j}:{\mathcal {A}}_{j}\in {\mathcal {C}}_{2}\}_{j\in J}\) leads to \({\mathcal {F}}\) being a HS-frame sequence with
The following theorem characterizes the HS-frame property of the \(l^{2}(J)\)-operator portraits of HS-frames.
Theorem 2.2
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be an \(l^{2}(J)\)-decomposable sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) with respect to a HS-orthonormal basis \({\mathcal {E}}=\{{\mathcal {E}}_{j}\}_{j\in J}\) associated with \(U=(u_{j,\,k})_{j,k\in J}\in {\mathcal {B}}(l^{2}(J))\), i.e.,
Suppose \({\mathcal {G}}\) is a HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\), and \(V=(v_{j,k})_{j,k\in J}\in {\mathcal {B}}(l^{2}(J))\). Let \({\mathcal {F}}_{j}=\sum \limits _{k\in J}v_{k,\,j}{\mathcal {G}}_{k}\) for \(j\in J\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a HS-frame (HS-Riesz basis, HS-frame sequence, HS-Riesz sequence) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) if UV is bounded and surjective (bijective, of closed range, injective and of closed range) on \(l^{2} (J)\).
Proof
By a standard argument, we have \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is an \(l^{2}(J)\)-decomposable sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\) with respect to \({\mathcal {E}}\) associated with \(UV\in {\mathcal {B}}(l^{2}(J))\). Applying Lemma 2.5 to UV gives the theorem. \(\square \)
The following is an example of Theorem 2.2.
Example 2.2
Let \({\mathcal {E}}=\{{\mathcal {E}}_{j}\}_{j\in \mathbb {N}}\) be a HS-orthonormal basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\). Define \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in \mathbb {N}}\) by
where \(U_{k}\) are \((l_{k}-l_{k-1})\times (n_{k}-n_{k-1})\) matrices such that \(rank(U_{k})=l_{k}-l_{k-1}\) for \(k\in \mathbb {N}\) (write \(l_{0}=n_{0}=0\)). Then
Obviously, \(\Vert U\Vert \le \sup \limits _{k\in \mathbb {N}}\Vert U_{k}\Vert \). Arbitrarily fix \(Y=(Y_{k})_{k\in \mathbb {N}}\in l^{2}\) with \(Y_{k}\in \mathbb {C}^{l_{k}-l_{k-1}}\). Since \(rank(U_{k})=l_{k}-l_{k-1}\), we have \(U_{k}U_{k}^{\dagger }Y_{k}=Y_{k}\) for \(k\in \mathbb {N}\). Take \(X=\left( U_{k}^{\dagger }Y_{k}\right) _{k\in \mathbb {N}}\). Then
and
Thus U is bounded and surjective on \(l^{2}(\mathbb {N})\). It follows that \({\mathcal {G}}\) is a HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) by Lemma 2.5.
Define matrix \(V=diag(V_{1}, V_{2}, \cdots )\) satisfying
where \(V_{k}\) are \((n_{k}-n_{k-1})\times (m_{k}-m_{k-1})\) matrices such that \(\mathrm{{range}}(V_{k})\subset \mathrm{{range}}(U_{k}^{*})\) for \(k\in \mathbb {N}\) with \(l_{0}=m_{0}=0\). Let \({\mathcal {F}}_{j}=\sum \limits _{k\in J}v_{j,k}{\mathcal {G}}_{k}\) for \(j\in \mathbb {N}\). Then we have the following claims.
Claim 1. UV is bounded on \(l^{2}(\mathbb {N})\), and UV is surjective (injective) if and only if each \(U_{k}V_{k}\) with \(k\in \mathbb {N}\) is surjective (injective).
Obviously, \(\Vert UV\Vert \le \left( \sup \limits _{k\in \mathbb {N}}\Vert U_{k}\Vert \right) \left( \sup \limits _{k\in \mathbb {N}}\Vert V_{k}\Vert \right) <\infty \), UV is injective if and only if each \(U_{k}V_{k}\) with \(k\in \mathbb {N}\) is injective. Suppose each \(U_{k}V_{k}\) with \(k\in \mathbb {N}\) is surjective. Arbitrarily fix \(Y=(Y_{k})_{k\in \mathbb {N}}\) in \(l^{2}\) with \(Y_{k}\in \mathbb {C}^{l_{k}-l_{k-1}}\). Since \(U_{k}V_{k}\) is surjective and \(\mathrm{{range}}(V_{k})\subset \mathrm{{range}}(U_{k}^{*})\), we have that
and
for \(k\in \mathbb {N}\) by Lemma 2.1. Take \(X=(X_{k})_{k\in \mathbb {N}}\) with \(X_{k}=(U_{k}V_{k})^{\dagger }Y_{k}\). Then
by (2.23), and
Thus UV is surjective. Obviously, UV being surjective implies each \(U_{k}V_{k}\) with \(k\in \mathbb {N}\) being surjective. Claim 1 therefore follows.
Claim 2. UV is of closed range. Suppose \(\{X^{(n)}\}_{n\in \mathbb {N}}\) is a sequence in \(l^{2}\) such that
for some \(Y\in l^{2}\), where \(X^{(n)}=\left( X_{k}^{(n)}\right) _{k\in \mathbb {N}}\), \(Y=(Y_{k})_{k\in \mathbb {N}}\) with \(X_{k}^{(n)}\in \mathbb {C}^{m_{k}-m_{k-1}},\) \(Y_{k}\in \mathbb {C}^{l_{k}-l_{k-1}}\). Then
Since each \(U_{k}V_{k}\) is a \((l_{k}-l_{k-1})\times (m_{k}-m_{k-1})\) matrix, it is of closed range. Take \(X_{k}=(U_{k}V_{k})^{\dagger }Y_{k}\) for \(k\in \mathbb {N}\). Then
Let \(X=(X_{k})_{k\in \mathbb {N}}\). Then \(X\in l^{2}\) similarly to Claim 1. Obviously, \(UVX=Y\). Thus UV is of closed range.
By Theorem 2.2 and Claims 1 and 2, we have
(i) Suppose \(m_{k}= l_{k}\) and \(rank(U_{k}V_{k})=l_{k}-l_{k-1}\) for \(k\in \mathbb {N}\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in \mathbb {N}}\) is a HS-Riesz basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
(ii) Suppose \(rank(U_{k}V_{k})=l_{k}-l_{k-1}\) for \(k\in \mathbb {N}\), and there exists at least one \(k_{0}\in \mathbb {N}\) such that \(m_{k_{0}}-m_{k_{0}-1}>l_{k_{0}}-l_{k_{0}-1}\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in \mathbb {N}}\) is a HS-frame but not a HS-Riesz basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
(iii) Suppose there exists at least one \(k_{0}\in \mathbb {N}\) such that \(rank(U_{k_{0}}V_{k_{0}})<l_{k_{0}}-l_{k_{0}-1}\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in \mathbb {N}}\) is a HS-frame sequence but not a HS-frame for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
(iv) Suppose \(rank(U_{k}V_{k})=m_{k}-m_{k-1}\) for each \(k\in \mathbb {N}\), and there exists at least one \(k_{0}\in \mathbb {N}\) such that \(m_{k_{0}}-m_{k_{0}-1}<l_{k_{0}}-l_{k_{0}-1}\). Then \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in \mathbb {N}}\) is a HS-Riesz sequence but not a HS-Riesz basis for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\).
3 Perturbation of HS-Frame Sequences
This section focuses on stable perturbations of HS-frame sequences (HS-Riesz sequences). For this purpose, we quote following lemmas from [11, 14, 26].
Lemma 3.1
[26] Let \({\mathcal {H}}_{1}\) and \({\mathcal {H}}_{2}\) be two Hilbert spaces, and \(U\in {\mathcal {B}}({\mathcal {H}}_{1},{\mathcal {H}}_{2})\). Then
(i) \(\gamma (U)=\inf \Big \{\Vert Uh\Vert :h\in (\ker U)^{\perp },\Vert h\Vert =1 \Big \}>0\) if and only if \(\mathrm{{range}}(U)\) is closed.
(ii) If \(\mathrm{{range}}(U)\) is closed, then \(\Vert U^{\dagger }\Vert =\gamma (U)^{-1}\).
Lemma 3.2
[11, Theorem 2.2] Given Hilbert spaces \({\mathcal {H}}_{1}\) and \( {\mathcal {H}}_{2}\), and let \(T,\,U\in {\mathcal {B}}({\mathcal {H}}_{1},{\mathcal {H}}_{2})\). Suppose that \(\delta =\delta \left( \ker (T),\ker (U)\right) <1\), and there exist numbers \(0\le \lambda _{1}<1\), \(\lambda _{2}>-1\) and \(\mu \ge 0\) such that
Then
(i) \(\gamma (U)\ge \frac{(1-\lambda _{1})\gamma (T)\sqrt{1-\delta ^{2}}-\mu }{1+\lambda _{2}}\).
(ii) If \(\mathrm{{range}}(T)\) is closed and \(\lambda _{1}+\frac{\mu }{\gamma (T)\sqrt{1-\delta ^{2}}}<1\), then \(\mathrm{{range}}(U)\) is closed and
Lemma 3.3
[14, Lemma 2.1] Let \({\mathcal {H}}\) be a Hilbert space. Then \(\delta (V,W)=\Vert P_{V}P_{W^{\perp }}\Vert \) for all closed subspaces V and W of \({\mathcal {H}}\).
The following theorem gives a perturbation theorem of HS-frame sequences, where the gap between the kernel spaces of synthesis operators is involved.
Theorem 3.1
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds A and B. Suppose that \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a sequence in \({\mathcal {B}}({\mathcal {H}},{\mathcal {C}}_{2})\), and that there exist constants \(\lambda _{2}\in [0,1)\) and \(\lambda _{1}\), \(\mu \ge 0\) such that
for \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0}{\mathcal {C}}_{2}\). Then \({\mathcal {F}}\) is a HS-Bessel sequence with bound \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
If furthermore \(\delta =\delta \left( \ker (T_{{\mathcal {G}}}),\,\ker (T_{{\mathcal {F}}})\right) <1\) and \(\lambda _{1}+\frac{\mu }{\sqrt{A(1-\delta ^{2})}}<1\), then \({\mathcal {F}}\) is a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with lower bound \(\left( \frac{(1-\lambda _{1})\sqrt{A(1-\delta ^{2})}-\mu }{1+\lambda _{2}}\right) ^{2}\).
Proof
By the same procedure as in [32, Theorem 4.3], \({\mathcal {F}}\) is a HS-Bessel sequence with bound \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\). It follows that
by (3.1) due to \(\oplus _{0}{\mathcal {C}}_{2}\) being dense in \(\oplus {\mathcal {C}}_{2}\). Next we prove that \({\mathcal {F}}\) satisfies the lower bound condition if \(\delta <1\) and \(\lambda _{1}+\frac{\mu }{\sqrt{A(1-\delta ^{2})}}<1\). Since \({\mathcal {G}}\) is a HS-frame sequence, \(\mathrm{{range}}(T_{\mathcal {G}})\) is closed by [16, Lemma 2.5]. Then, by [32, Proposition 2.13], and Lemma 3.1,
and thus
Collecting (3.2), (3.4) and Lemma 3.2 leads to the fact that \(\mathrm{{range}}(T_{{\mathcal {F}}})\) is closed, and
This implies that \({\mathcal {F}}\) is a HS-frame sequence by [16, Lemma 2.5]. Observe that its optimal lower bound is \(\left\| T_{{\mathcal {F}}}^{\dagger }\right\| ^{-2}\) by [32, Proposition 2.13]. By (3.3) and (3.5),
The proof is completed. \(\square \)
The following theorem deals the perturbation of HS-Riesz sequences, it is an application of Theorem 3.1.
Theorem 3.2
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-Riesz sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds A and B. Suppose that \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a sequence in \({\mathcal {B}}({\mathcal {H}},{\mathcal {C}}_{2})\), and that there exist constants \(\lambda _{2}\in [0,\,1)\) and \(\lambda _{1}\), \(\mu \ge 0\) such that \(\lambda _{1}+\frac{\mu }{\sqrt{A}}<1\) and
for \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0}{\mathcal {C}}_{2}\). Then \({\mathcal {F}}\) is a HS-Riesz sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
Proof
Since \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) is a HS-Riesz sequence, it is a HS-frame sequence and \(\ker (T_{{\mathcal {G}}})=0\) by [16, Lemma 2.6]. This implies that
By Theorem 3.1, \({\mathcal {F}}\) is a HS-frame sequence with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\), and (3.6) holds for all \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\). By [16, Lemma 2.6], we only need to prove \(\ker (T_{{\mathcal {F}}})=0\) to finish the proof. Next we do it. For \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\),
It follows that
Thus \(\ker (T_{{\mathcal {F}}})=0\). The proof is completed. \(\square \)
Using the gap between the range spaces of synthesis operators, the following theorem gives a perturbation theorem of HS-frame sequences.
Theorem 3.3
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds A and B, \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) be a sequence in \({\mathcal {B}}({\mathcal {H}},\,{\mathcal {C}}_{2})\). Write \(G=\overline{\mathrm{span}}\{{\mathcal {G}}_{j}^{*}{\mathcal {A}}_{j}:{\mathcal {A}}_{j}\in {\mathcal {C}}_{2}\}_{j\in J}\), \(F=\overline{\mathrm{span}}\{{\mathcal {F}}_{j}^{*}{\mathcal {A}}_{j}:{\mathcal {A}}_{j}\in {\mathcal {C}}_{2}\}_{j\in J}\). Suppose that \(\delta (F,\,G)<1\) and there exist constants \(\lambda _{2}\in [0,\,1)\) and \(\lambda _{1},\,\mu \ge 0\) such that
and
for \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0}{\mathcal {C}}_{2}\). Then \({\mathcal {F}}\) is a HS-frame sequence in \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
Proof
By the proof of Theorem 3.1, \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a HS-Bessel sequence with bound \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\), and (3.7) holds for all \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus {\mathcal {C}}_{2}\). Since \({\mathcal {F}}\) is a HS-Bessel sequence, we can define a bounded operator \(U:{\mathcal {H}}\rightarrow {\mathcal {H}}\) by
We first to show that the operator U is invertible. Since \({\mathcal {G}}\) is a HS-frame sequence, its canonical dual \(\{{\mathcal {G}}_{j}S_{{\mathcal {G}}}^{-1}\}_{j\in J}\) is also a HS-frame sequence with bounds \(B^{-1}\) and \(A^{-1}\). Arbitrarily fix \(h=h_{1}+h_{2}\) with \(h_{1}\in G\), \(h_{2}\in G^{\perp }\). Then
Also \(\Vert P_{F}P_{G^{\perp }}\Vert =\delta (F,\,G)\) by Lemma 3.3. It follows that
for \(h\in {\mathcal {H}}\). Then, by [6, Lemma 1], U is invertible due to \(0\le \left( \lambda _{1}+\frac{\mu }{\sqrt{A}}\right) ^{2}+\delta (F,\,G)^{2}<1\) and \(0\le \lambda _{2}<1\). Next we prove that \({\mathcal {F}}\) has a positive lower HS-frame bound to finish the proof. Arbitrarily fix \({\widetilde{h}}\in F\). Suppose \(U^{-1}{\widetilde{h}}=f_{1}+f_{2}\) with \(f_{1}\in G\) and \(f_{2}\in G^{\perp }\). Then, by the definition of U, we have
This implies that \(U^{-1}{\widetilde{h}}=f_{1}\in G\), \({\widetilde{h}}\in F\). Thus
Now we estimate \(\Vert U^{-1}{\widetilde{h}}\Vert \) \(({\widetilde{h}}\in F)\). By (3.8), we have that
and thus
for \({\widetilde{g}}\in G\). It follows that
Substituting \({\widetilde{g}}\) for \(U^{-1}{\widetilde{h}}\) in (3.10) gives
Collecting (3.9) and (3.11) leads to
equivalently,
The proof is completed. \(\square \)
Observe that \(\delta \left( F,\,G\right) =0\) if \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) is a HS-frame in Theorem 3.3. As an immediate consequence, we have the following corollary which appeared in [32, Theorem 4.3 and Corollary 4.5].
Corollary 3.1
Let \({\mathcal {G}}=\{{\mathcal {G}}_{j}\}_{j\in J}\) be a HS-frame (HS-Riesz basis) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds A and B. Suppose that \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) is a sequence in \({\mathcal {B}}({\mathcal {H}},{\mathcal {C}}_{2})\), and that there exists constants \(\lambda _{2}\in [0,1)\) and \(\lambda _{1}\), \(\mu \ge 0\) such that \(\lambda _{1}+\frac{\mu }{\sqrt{A}}<1\) and
for \({\mathcal {A}}=\{{\mathcal {A}}_{j}\}_{j\in J}\in \oplus _{0}{\mathcal {C}}_{2}\). Then \({\mathcal {F}}\) is a HS-frame (HS-Riesz basis) for \({\mathcal {H}}\) with respect to \({\mathcal {K}}\) with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
4 A Remark
In this section, we show that, using Theorems 2.1, 3.1-3.3, we can recover some results on frames.
Given a sequence \(f=\{f_{j}\}_{j\in J}\) in \({\mathcal {H}}\), we associate it with a HS-sequence \({\mathcal {F}}=\{{\mathcal {F}}_{j}\}_{j\in J}\) in \({\mathcal {B}}({\mathcal {H}},\,{\mathbb {C}})\) by
Then \({\mathcal {F}}_{j}^{*}c_{j}=c_{j}f_{j}\) for \(j\in J\) and \(c_{j}\in \mathbb {C}\). It follows that f and \({\mathcal {F}}\) has similar frame properties. Specifically, f is a Bessel sequence (frame, frame sequence, Riesz basis, Riesz sequence) for \({\mathcal {H}}\) if and only if \({\mathcal {F}}\) is a HS-Bessel sequence (HS-frame, HS-frame sequence, HS-Riesz basis, HS-Riesz sequence) for \({\mathcal {H}}\) with respect to \(\mathbb {C}\). In this case f and \({\mathcal {F}}\) have the same synthesis operator, i.e., \(T_{f}=T_{{\mathcal {F}}}\), and thus \(\ker (T_{f})=\ker (T_{{\mathcal {F}}})\), \(\mathrm{{range}}(T_{f})=\mathrm{{range}}(T_{{\mathcal {F}}})\).
Let \(\Gamma =(\gamma _{k,\,j})_{k,\,j\in J}\) define an operator in \({\mathcal {B}}(l^{2}(J))\). Applying (2.22) in Theorem 2.1, we have
It leads to the following corollary which appeared in [2, Theorem 4].
Corollary 4.1
Let \(g=\{g_{j}\}_{j\in J}\) be a frame for \({\mathcal {H}}\), and \(\Gamma =(\gamma _{k,\,j})_{k,\,j\in J}\) be a bounded linear operator on \(l^{2}(J)\). Define \(f_{j}=\sum \limits _{k\in J}\gamma _{k,\,j}\,g_{k}\) for \(j\in J\). Then \(f=\{f_{j}\}_{j\in J}\) is a frame for \({\mathcal {H}}\) if and only if there exists a constant \(\alpha \) such that
Applying Theorems 3.1–3.3, we have the following Corollaries 4.2–4.4 that appeared in [11, Theorem 3.2] and [14, Theorems 3.1 and 3.2].
Corollary 4.2
Let \(g=\{g_{j}\}_{j\in J}\) be a frame sequence in \({\mathcal {H}}\) with bounds A and B. Let \(f=\{f_{j}\}_{j\in J}\) be a sequence in \({\mathcal {H}}\) and suppose that there exist numbers \(\lambda _{2}\in [0,1)\) and \(\lambda _{1}\), \(\mu \ge 0\) such that
for \(c=\{c_{j}\}_{j\in J}\in l_{0}(J)\). Then f is a Bessel sequence with bound \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
If furthermore \(\delta =\delta \left( \ker (T_{g}),\,\ker (T_{f})\right) <1\) and \(\lambda _{1}+\frac{\mu }{\sqrt{A(1-\delta ^{2})}}<1\), then f is a frame sequence in \({\mathcal {H}}\) with lower bound \(\left( \frac{(1-\lambda _{1})\sqrt{A(1-\delta ^{2})}-\mu }{1+\lambda _{2}}\right) ^{2}\).
Corollary 4.3
Let \(g=\{g_{j}\}_{j\in J}\) be a Riesz sequence in \({\mathcal {H}}\) with bounds A and B. Let \(f=\{f_{j}\}_{j\in J}\) be a sequence in \({\mathcal {H}}\) and suppose that there exist numbers \(\lambda _{2}\in [0,1)\) and \(\lambda _{1}\), \(\mu \ge 0\) such that \(\lambda _{1}+\frac{\mu }{\sqrt{A}}<1\) and
for \(c=\{c_{j}\}_{j\in J}\in l_{0}(J)\). Then f is a Riesz sequence in \({\mathcal {H}}\) with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
Corollary 4.4
Let \(g=\{g_{j}\}_{j\in J}\) be a frame sequence in \({\mathcal {H}}\) with bounds A and B. Let \(f=\{f_{j}\}_{j\in J}\) be a sequence in \({\mathcal {H}}\). Write \(G=\overline{\mathrm{span}}\{g_{j}\}_{j\in J}\), \(F=\overline{\mathrm{span}}\{f_{j}\}_{j\in J}\). Assume that \(\delta (F,\,G)<1\) and there exist constants \(\lambda _{2}\in [0, 1)\) and \(\lambda _{1},\mu \ge 0\) such that
and
for \(c=\{c_{j}\}_{j\in J}\in l_{0}(J)\). Then f is a frame sequence in \({\mathcal {H}}\) with bounds \(\left( \frac{(1-\lambda _{1})\sqrt{A}-\mu }{1+\lambda _{2}}\right) ^{2}\) and \(\left( \frac{(1+\lambda _{1})\sqrt{B}+\mu }{1-\lambda _{2}}\right) ^{2}\).
References
Abdollahpour, M.R., Khedmati, Y.: g-duals of continuous g-frames and their perturbations. Results Math. 73(4), 15 (2018)
Aldroubi, A.: Portraits of frames. Proc. Am. Math. Soc. 123(6), 1661–1668 (1995)
Aldroubi, A., Cabrelli, C., Molter, U.: Wavelets on irregular grids with arbitrary dilation matrices and frame atoms for \(L^{2}({\mathbb{R} }^{d})\). Appl. Comput. Harmon. Anal. 17(2), 119–140 (2004)
Arefijamaal, A.A., Sadeghi, Gh.: von Neumann-Schatten dual frames and their perturbations. Results Math. 69(3–4), 431–441 (2016)
Asgari, M.S., Khosravi, A.: Frames and bases of subspaces in Hilbert spaces. J. Math. Anal. Appl. 308(2), 541–553 (2005)
Cazassa, P.G., Christensen, O.: Perturbation of operators and applications to frame theory. J. Fourier Anal. Appl. 3(5), 543–557 (1997)
Casazza, P.G., Kutyniok, G.: Frames of subspaces. Wavelets, frames and operator theory, Contemp. Math., Vol. 345, Amer. Math. Soc., Providence, RI, 87–113 (2004)
Casazza, P.G., Kutyniok, G., Li, S.: Fusion frames and distributed processing. Appl. Comput. Harmon. Anal. 25(1), 114–132 (2008)
Christensen, O.: A Paley-Wiener theorem for frames. Proc. Am. Math. Soc. 123(7), 2199–2201 (1995)
Christensen, O.: Frame perturbations. Proc. Am. Math. Soc. 123(4), 1217–1220 (1995)
Christensen, O.: Operators with closed range, pseudo-inverses, and perturbation of frames for a subspace. Canad. Math. Bull. 42(1), 37–45 (1999)
Christensen, O.: An Introduction to Frames and Riesz Bases. Second edition, Birkhäuser (2016)
Christensen, O., Hasannasab, M.: Operator representations of frames: boundedness, duality, and stability. Integral Equ. Op. Theory 88(4), 483–499 (2017)
Christensen, O., Lennard, C., Lewis, C.: Perturbation of frames for a subspace of a Hilbert space. Rocky Mountain J. Math. 30(4), 1237–1249 (2000)
Daubechies, I., Grossmann, A., Meyer, Y.: Painless nonorthogonal expansions. J. Math. Phys. 27, 1271–1283 (1986)
Dong, J., Li, Y.-Z.: Duality principles in Hilbert-Schmidt frame theory. Math. Methods Appl. Sci. 44(6), 4888–4906 (2021)
Dörfler, M., Feichtinger, H.G., Gröchenig, K.: Time-frequency partitions for the Gelfand triple \((S_{0}, L^{2}, S^{\prime }_{0})\). Math. Scand. 98(1), 81–96 (2006)
Duffin, R.J., Schaeffer, A.C.: A class of nonharmonic Fourier series. Trans. Am. Math. Soc. 72, 341–366 (1952)
Favier, S.J., Zalik, R.A.: On the stability of frames and Riesz bases. Appl. Comput. Harmon. Anal. 2(2), 160–173 (1995)
Fornasier, M.: Quasi-orthogonal decompositions of structured frames. J. Math. Anal. Appl. 289(1), 180–199 (2004)
Guo, X.: Perturbations of invertible operators and stability of g-frames in Hilbert spaces. Results Math. 64(3–4), 405–421 (2013)
Guo, X.: Operator parameterizations of g-frames. Taiwanese J. Math. 18(1), 313–328 (2014)
Heil, C.: A Basis Theory Primer. Birkhäuser/Springer, New York (2011)
Javanshiri, H., Hajiabootorabi, M., Mardanbeigi, M.R.: The effect of perturbations of frames on their alternate and approximately dual frames. Math. Methods Appl. Sci. 45(4), 2058–2071 (2022)
Jiang, Z., Yang, S.Z., Zheng, X.W.: Perturbation and construction of almost self-located robust frames with applications to erasure recovery. Math. Methods Appl. Sci. 44(8), 7333–7342 (2021)
Kato, T.: Perturbation Theory for Linear Operator. Springer-Verlag, New York (1984)
Li, D.W., Leng, J.S., Huang, T.Z., Sun, G.M.: On sum and stability of g-frames in Hilbert spaces. Linear Multilinear Algebra 66(8), 1578–1592 (2018)
Li, Y.-N., Li, Y.-Z.: Hilbert-Schmidt frames and their duals. Int. J. Wavelets Multiresolut. Inf. Process. 19(5), 2150011 (2021)
Li, S., Ogawa, H.: Pseudoframes for subspaces with applications. J. Fourier Anal. Appl. 10(4), 409–431 (2004)
Li, Y.-Z., Zhang, X.-L.: Frame properties of Hilbert-Schmidt operator sequences, Mediterr. J. Math., accepted
Li, Y.-Z., Zhang, X.-L.: Dilations of (dual) Hilbert-Schmidt frames. Ann. Funct. Anal. 13(3), 20 (2022)
Poria, A.: Approximation of the inverse frame operator and stability of Hilbert-Schmidt frames. Mediterr. J. Math. 14(4), 22 (2017)
Poria, A.: Some identities and inequalities for Hilbert-Schmidt frames. Mediterr. J. Math. 14(2), 59,14 (2017)
Sadeghi, Gh., Arefijamaal, A.A.: von Neumann-Schatten frames in separable Banach spaces. Mediterr. J. Math. 9(3), 525–535 (2012)
Sun, W.: G-frames and g-Riesz bases. J. Math. Anal. Appl. 322(1), 437–452 (2006)
Sun, W.: Stability of g-frames. J. Math. Anal. Appl. 326(2), 858–868 (2007)
Wang, Y.J., Zhu, Y.C.: G-frames and g-frame sequences in Hilbert spaces. Acta Math. Sin. 25(12), 2093–2106 (2009)
Xiang, Z.Q.: Some new results on the construction and stability of K-g-frames in Hilbert spaces. Int. J. Wavelets Multiresolut. Inf. Process. 18(5), 2050034, 19 (2020)
Young, R.M.: An Introduction to Nonharmonic Fourier Series. Academic Press, New York (1980)
Acknowledgements
The authors would like to extend their sincere gratitude to referees for their careful reading this article and valuable comments, which helped to greatly improve the readability of this article.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Rosihan M. Ali.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supported by National Natural Science Foundation of China (Grant No. 11971043)
Rights and permissions
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Zhang, XL., Li, YZ. Portraits and Perturbations of Hilbert–Schmidt Frame Sequences. Bull. Malays. Math. Sci. Soc. 45, 3197–3223 (2022). https://doi.org/10.1007/s40840-022-01375-0
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40840-022-01375-0