In this paper, we define the g-Riesz-dual of a given special operator-valued sequence with respect to g-orthonormal bases for a separable Hilbert space. We demonstrate that the g-R-dual keeps some synchronous frame properties with the operator-valued sequence given. We also display some Schauder basis-like properties of the g-R-dual in the light of the properties of the given sequence. In particular, the g-R-dual can be characterized by the use of another sequence, related to the given sequence. Finally, a special sequence is constructed to build the relationship between an operator-valued sequence and a g-Riesz sequence.
Hinweise
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
1 Introduction
Duality principles in Gabor theory play a fundamental role in analyzing the Gabor system. In [1], the authors described the concept of the Riesz-dual of a vector-valued sequence and illustrated the common frame properties for the give sequence and its R-dual. The conditions under which a Riesz sequence can be a R-dual of a given frame are investigated in [2]. In this paper, we are interested in the duality principles for g-frames. In [3], the g-R-dual was first defined, and some frame properties of g-R-dual were exhibited by the properties of the given operator-valued sequence. In this paper, our definition of g-R-dual in Sect. 2 is much weaker, and we characterize the g-R-dual with the analysis operator. The properties of the g-completeness, g-w-linearly independent, g-minimality of the g-R-dual is accounted in Sect. 3. In Sect. 4, we construct a sequence with a g-Riesz sequence and a given operator-valued sequence to consider the g-R-dual in a different way.
Throughout this paper, we use \(\mathbb{N}\) to denote the set of all natural numbers, and assume that \(\{H_{i}\}_{i\in\mathbb{N}}\) is a sequence of closed subspaces of a separable Hilbert space K, H is a separable Hilbert space. Denote by \(\{A_{i}\}_{i\in\mathbb{N}}\), or for short \(\{A_{i}\}\), a sequence of operators with \(A_{i}\in B(H,H_{i})\) for any \(i\in\mathbb{N}\). Suppose that \(B(H,H_{i})\) denotes the collection of all the bounded linear operators from H into \(H_{i}\), \(i\in\mathbb{N}\). Denote by \(\bigoplus_{i\in \mathbb{N}}{H_{i}}\) the orthogonal direct sum Hilbert space of \(\{ H_{i}\}_{i\in\mathbb{N}}\), \(\{g_{i}\}:=\{g_{i}\}_{i\in\mathbb{N}}\) for any \(\{g_{i}\}_{i\in\mathbb{N}}\in\bigoplus_{i\in\mathbb {N}}{H_{i}}\).
Anzeige
In [10], Sun raised the concept of a g-frame as follows. Let \(A_{i}\in B(H,H_{i})\), \(i\in\mathbb{N}\). If there exist two constants \(a, b\) such that
$$ a \Vert f \Vert ^{2}\leq\sum_{i\in\mathbb {N}} \Vert A_{i}f \Vert ^{2}\leq b \Vert f \Vert ^{2},\quad \forall f\in H, $$
we call \(\{A_{i}\}\) a g-frame for H. We call \(\{A_{i}\}\) a tight g-frame for H if \(a=b\). Specially, if \(a=b=1\), \(\{A_{i}\} \) is called a Parseval g-frame for H. If the inequalities above hold only for \(f\in \overline{\operatorname{span}} \{A^{*}_{i}H_{i}\} _{i\in\mathbb{N}}\), we call \(\{A_{i}\}\) a g-frame sequence for H. If only the right-hand inequality above holds, then we say that \(\{ A_{i}\}\) is a g-Bessel sequence for H. If \(\overline{\operatorname{span}} \{A^{*}_{i}H_{i}\}_{i\in\mathbb{N}}=H\), we say that \(\{A_{i}\}\) is g-complete in H. If \(\{A_{i}\}\) is g-complete and such that
we call \(\{A_{i}\}\) a g-Riesz basis for H. If the g-completeness is not satisfied, it is called a g-Riesz sequence for H. As we know, if \(\{A_{i}\}\) is a g-frame for H, we define \(S_{A}f=\sum_{i\in\mathbb{N}}A^{*}_{i}A_{i}f\) for any \(f\in H\), then \(S_{A}\) is a well-defined, bounded, positive, invertible operator by [10]. We call \(S_{A}\) a frame operator of \(\{A_{i}\}\). Another basic fact is that \(\{\widetilde{A}_{i}\}_{i\in\mathbb{N}}=\{ A_{i}S_{A}^{-1}\}_{i\in\mathbb{N}}\) is a g-frame for H, we call it a canonical dual g-frame of \(\{A_{i}\}\). Extensively, by [8], if a g-frame \(\{B_{i}\}\) for H such that \(f=\sum_{i\in\mathbb {N}}B^{*}_{i}A_{i}f\) for every \(f\in H\), we say that it is a dual g-frame of \(\{A_{i}\}\). Recently, g-frames in Hilbert spaces have been studied intensively; for more details see [4‐10] and the references therein.
In the following we introduce some definitions and lemmas connected with the g-bases in Hilbert space which will be needed in the paper.
\(\{A_{i}\}\) is a g-orthonormal sequence for H, i.e., \(\langle A^{*}_{i}g_{i},A^{*}_{j}g_{j}\rangle=\delta _{ij}\langle g_{i},g_{j}\rangle\) for any \(i, j\in\mathbb{N}\), any \(g_{i}\in H_{i}, g_{j}\in H_{j}\).
(2)
\(\{A_{i}\}\) is g-complete in H.
We call \(\{A_{i}\}\) a g-orthonormal basis for H. Obviously, (2) is equivalent to that \(\{A_{i}\}\) is a Parseval g-frame for H by [5, Corollary 4.4], when (1) holds. Specially, if \(\{A_{i}\} \) only satisfies \(A_{i}A^{*}_{j}=0\) for any \(i, j\in\mathbb{N}\), \(i\neq j\), \(\{A_{i}\}\) is called a g-orthogonal sequence for H.
Anzeige
The g-orthonormal basis is a special case that itself is g-biorthonormal. The following result shows that for the g-Riesz basis there also exists a g-biorthonormal sequence.
LetHbe a separable Hilbert space, \(\{H_{i}\}_{i\in\mathbb{N}}\)be a sequence of separable Hilbert spaces. Then there exists a sequence\(\{ \varGamma_{i}\}\), which is a g-orthonormal basis forHif and only if\(\operatorname{dim}H=\sum_{i\in\mathbb{N}}\operatorname{dim}H_{i}\).
The concept of g-bases in Hilbert space is a generalization of the Schauder basis. Let \(\{A_{i}\}\). If for any \(f\in H\), there is a unique sequence \(\{g_{i}\}_{i\in\mathbb{N}}\) with \(g_{i}\in H_{i}\) for any \(i\in\mathbb{N}\) such that \(f=\sum_{i\in\mathbb{N}}A_{i}^{*}g_{i}\), we call \(\{A_{i}\}\) a g-basis for H. If \(\{A_{i}\}\) is a g-basis for \(\overline{\operatorname{span}} \{A_{i}^{*}H_{i}\}_{i\in\mathbb {N}}\), \(\{A_{i}\}\) is called a g-basic sequence for H. Moreover, If \(\sum_{i\in\mathbb{N}}A^{*}_{i}g_{i}=0\) for \(\{g_{i}\} \in\bigoplus_{i\in\mathbb{N}}{H_{i}}\), then \(g_{i}=0\), we call \(\{A_{i}\}\)g-w-linearly independent. If \(A^{*}_{j}g_{j}\notin \overline{\operatorname{span}}_{i\neq j} \{A^{*}_{i}g_{i}\}_{i\in \mathbb{N}}\) for any \(\{g_{i}\}\in\bigoplus_{i\in\mathbb {N}}{H_{i}}\) such that \(g_{i}\in H_{i}\), \(g_{i}\neq0\), any \(i\in \mathbb{N}\), we call \(\{A_{i}\}\)g-minimal. For more details as regards g-bases see [4].
2 Duality for g-frame
Before giving the definition of g-R-dual, we introduce a lemma which is related to the g-Bessel sequence.
Lemma 2.1
The sequence\(\{A_{i}\}\)is a g-Bessel sequence forHif and only if\(\sum_{i\in\mathbb{N}}A^{*}_{i}g_{i}\)is convergent for any\(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}}{H_{i}}\), and is also equivalent to that\(\sum_{i\in\mathbb {N}} \Vert A_{i}f \Vert ^{2}<\infty\)for every\(f\in H\).
Proof
Suppose \(\sum_{i\in\mathbb{N}}A^{*}_{i}g_{i}\) is convergent for any \(\{g_{i}\}\in\bigoplus_{i\in\mathbb {N}}{H_{i}}\). For any \(n\in\mathbb{N}\), \(\{g_{i}\}\in \bigoplus_{i\in\mathbb{N}}{H_{i}}\), we define \(T_{n}: \bigoplus_{i\in\mathbb{N}}{H_{i}}\rightarrow H, T_{n}\{g_{i}\}=\sum_{i=1}^{n}A^{*}_{i}g_{i}\). Thus \(T_{n}\) is bounded evidently. Since \(\{T_{n}\}_{n\in\mathbb{N}}\) converges to T in the strong operator topology as \(n\rightarrow \infty\), where \(T\{g_{i}\}=\sum_{i\in\mathbb {N}}A^{*}_{i}g_{i}\) for every \(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}}{H_{i}}\). Then T is bounded by the uniform boundedness principle in Banach space. The rest follows directly. □
For a g-Bessel sequence \(\{A_{i}\}\), we can define the analysis operator as \(\theta_{A}: H\rightarrow \bigoplus_{i\in\mathbb{N}}{H_{i}}, \theta_{A}f=\{A_{i}f\}_{i\in\mathbb{N}}\text{ for any }f\in H\), which is well defined and bounded obviously by Lemma 2.1.
Definition 2.2
Let \(\{\varLambda_{i}\}\), \(\{\varGamma_{i}\}\) be two g-orthonormal bases for H. Suppose a sequence \(\{A_{i}\}\) such that \(\sum_{i\in \mathbb{N}} \Vert A_{i}\varLambda^{*}_{j}g_{j} \Vert ^{2}<\infty\) for any \(j\in \mathbb{N}\), any \(g_{j}\in H_{j}\). We define
We call \(\{{\mathcal{A}}_{i}\}\) a g-R-dual sequence of \(\{A_{i}\}\).
Remark 2.3
By [4, Theorem 4.4], for any \(j\in\mathbb{N}\), \({\mathcal {A}}_{j}\) is well defined if and only if \(\{A_{i}\varLambda^{*}_{j}g_{j}\} _{i\in\mathbb{N}}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\) for any \(g_{j}\in H_{j}\), i.e., \(\{A_{i}Q_{j}f\}_{i\in\mathbb{N}}\in \bigoplus_{i\in\mathbb{N}}{H_{i}}\) for any \(f\in H\), i.e., \(\{ A_{i}\}\) is a g-Bessel sequence for \(\operatorname{ran}Q_{j}\) by Lemma 2.1, where \(Q_{j}\) is the orthogonal projection from H onto \(\overline{\operatorname{ran}} \varLambda_{j}^{*}\). Obviously, \(\{A_{i}\}\) may not be a g-Bessel sequence for H. The condition of our definition is weaker than that in [3, Definition 1.13]. Thus Definition 2.2 is equivalent to \({\mathcal{A}}_{j}=\sum_{i\in\mathbb {N}}\varLambda_{j}A_{i}^{*}\varGamma_{i}\) for any \(j\in\mathbb{N}\). By Definition 1.1, we get \(\varGamma_{k}{\mathcal {A}}^{*}_{j}=A_{k}\varLambda^{*}_{j}\) for every \(i, k\in\mathbb{N}\).
The following exhibits that the sequence \(\{A_{i}\}\) satisfying Definition 2.2 shares the common properties with its g-R-dual \(\{{\mathcal{A}}_{i}\}\). Similar results are referred to in [3, Theorem 2.2].
Theorem 2.4
Let\(\{A_{i}\}\)satisfy Definition 2.2, \(\{{\mathcal{A}}_{i}\}\)be its g-R-dual defined in Definition 2.2. Then\(\{A_{i}\}\)is a g-Bessel sequence forHif and only if\(\{{\mathcal{A}}_{i}\}\)is a g-Bessel sequence forH. Moreover, they have the same upper bound.
Proof
For every \(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}}{H_{i}}\), let \(f=\sum_{i\in\mathbb{N}}\varLambda^{*}_{i}g_{i}\), \(h=\sum_{i\in\mathbb {N}}\varGamma^{*}_{i}g_{i}\). Suppose \(\{A_{i}\}\) is a g-Bessel sequence for H and has an upper bound b. Since \(\theta_{\varLambda}, \theta_{\varGamma}: H\rightarrow \bigoplus_{i\in\mathbb{N}}{H_{i}}\) are unitary,
By Lemma 2.1, \(\{{\mathcal{A}}_{i}\}\) is a g-Bessel sequence for H and has an upper bound b. The converse is similar. □
When \(\{A_{i}\}\) is a g-Bessel sequence, there exists a unitary equivalence between \(\{\varLambda_{i}S_{A}^{\frac{1}{2}}\}\) and the R-dual \(\{{\mathcal{A}}_{i}\}\).
Theorem 2.5
Let\(\{A_{i}\}\)be a g-Bessel sequence forH, \(\{{\mathcal{A}}_{i}\}\)be its g-R-dual defined in Definition 2.2. Then
there exists an isometric operatorTfrom\(\overline{\operatorname{ran}} S_{A}^{\frac{1}{2}}\theta_{\varLambda}^{*}\)onto\(\overline{\operatorname{ran}} \theta_{\mathcal{A}}^{*}\)such that\({\mathcal {A}}_{i}T=\varLambda_{i}S_{A}^{\frac{1}{2}}\)for any\(i\in \mathbb{N}\).
Proof
(1) Since \(\{A_{i}\}\) is a g-Bessel sequence for H, so is \(\{{\mathcal {A}}_{i}\}\) by Theorem 2.4. Then, for any \(i, j\in\mathbb {N}\), any \(g_{i}\in H_{i}, g_{j}\in H_{j}\), we have
(3) Define \(T^{*}:\operatorname{ran}\theta_{\mathcal{A}}^{*}\rightarrow \operatorname{ran}S_{A}^{\frac{1}{2}}\), \(T^{*}(\sum_{i\in\mathbb {N}}{\mathcal{A}}^{*}_{i}g_{i})=\sum_{i\in\mathbb{N}}S_{A}^{\frac {1}{2}}\varLambda_{i}^{*}g_{i}\) for any \(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\). It is easy to verify \(T^{*}\) is well defined by (2). We can extend T to an isometric operator from \(\overline{\operatorname {ran}} S_{A}^{\frac{1}{2}}\theta_{\varLambda}^{*}\) onto \(\overline{\operatorname {ran}} \theta_{\mathcal{A}}^{*}\). We still denote the operator as T for convenience. □
In the following results we show the properties of g-R-dual in the case that \(\{A_{i}\}\) is a g-frame sequence by the corresponding analysis operators. The results are similar to the conclusions in [3, Corollary 2.6].
Theorem 2.6
Let\(\{A_{i}\}\)satisfy Definition 2.2, \(\{{\mathcal{A}}_{i}\}\)be its g-R-dual defined in Definition 2.2. Then\(\{A_{i}\}\)is a g-frame sequence forHif and only if\(\{{\mathcal{A}}_{i}\}\)is a g-frame sequence forHwith the same frame bounds. Specially, in this case the following are equivalent:
(1)
\(\{A_{i}\}\)is a g-frame forHwith the frame bounds\(a, b\).
(2)
\(\{{\mathcal{A}}_{i}\}\)is a g-Riesz sequence forHwith the frame bounds\(a, b\).
The case of the g-Bessel upper bound we get easily by Theorem 2.4. We now show the case of the lower bound in a similar way as the proof of Theorem 2.4.
Because \(\{A_{i}\}\), \(\{{\mathcal{A}}_{i}\}\) are g-Bessel sequences, we easily have \(\theta_{A}=\theta_{\varGamma}\theta_{\mathcal {A}}^{*}\theta_{\varLambda}\). Then \(g\in\operatorname{ker}\theta_{A}\) if and only if \(g\in\operatorname{ker}\theta_{\mathcal{A}}^{*}\theta _{\varLambda}\), i.e., \(\theta_{\varLambda}g\in\operatorname{ker}\theta _{\mathcal{A}}^{*}\). Hence, \(g\in(\operatorname{ker}\theta_{A})^{\bot}\) if and only if \(\theta_{\varLambda}g\in(\operatorname{ker}\theta_{\mathcal {A}}^{*})^{\bot}\) since \(\theta_{\varLambda}\) is unitary.
Evidently, \(\{A_{i}\}\) is a g-frame sequence for H if and only if for any \(f\in\operatorname{ran}\theta^{*}_{A}\), one has \(a \Vert f \Vert ^{2}\leq\sum_{i\in\mathbb{N}} \Vert A_{i}f \Vert ^{2}= \Vert \theta _{A}f \Vert ^{2}\leq b \Vert f \Vert ^{2}\), i.e.,
$$ a \Vert \theta_{\varLambda}f \Vert ^{2}= \bigl\Vert \theta_{\mathcal {A}}^{*}\theta_{\varLambda}f \bigr\Vert ^{2}\leq b \Vert f \Vert ^{2}=b \Vert \theta_{\varLambda}f \Vert ^{2}, $$
which is equivalent to \(\{{\mathcal{A}}_{i}\}\) is a g-frame sequence for H.
The equivalence of (1) and (2) is obvious since \((\operatorname{ker}\theta _{A})^{\bot}=\{0\}\) if and only if \((\operatorname{ker}\theta_{\mathcal {A}}^{*})^{\bot}=\{0\}\) by the proof above.
(1) ⇒ (3). Let \(\{A_{i}\}\) be a g-frame for H with the frame bounds \(a, b\). Take P as an arbitrary orthogonal projection on H. For any \(f=f_{1}+f_{2}\in H\), where \(f_{1}\in\operatorname{ran}P, f_{2}\in\operatorname{ker}P\), we have
The following result was given in [3, Theorem 4.1], we here give a simple illustration by the use of the analysis operators.
Lemma 2.7
Let\(\{A_{i}\}, \{B_{i}\}\)be two g-frames forH, \(\{{\mathcal{A}}_{i}\}\), \(\{{\mathcal{B}}_{i}\}\)be their g-R-dual sequences defined in Definition 2.2, respectively. Then\(\{A_{i}\}\)is a dual g-frame of\(\{B_{i}\}\)if and only if\(\langle{\mathcal{A}}^{*}_{i}g_{i}, {\mathcal{B}}^{*}_{j}g_{j}\rangle =\delta_{ij}\langle g_{i},g_{j}\rangle\)for any\(i, j\in\mathbb{N}\), any\(g_{i}\in H_{i}\), \(g_{j}\in H_{j}\).
Proof
By Definition 2.2, we get \(\theta_{\mathcal{A}}=\theta _{\varLambda}\theta^{*}_{A}\theta_{\varGamma}\), \(\theta_{\mathcal {B}}=\theta_{\varLambda}\theta^{*}_{B}\theta_{\varGamma}\). Then \(\theta_{\mathcal{A}}\theta_{\mathcal{B}}^{*}=\theta_{\varLambda }\theta^{*}_{A}\theta_{B}\theta_{\varLambda}^{*}\). Obviously, \(\theta^{*}_{A}\theta_{B}=I\) if and only if \(\theta _{\mathcal{A}}\theta_{\mathcal{B}}^{*}=I_{\bigoplus_{i\in \mathbb{N}}{H_{i}}}\), i.e., \(\langle{\mathcal{A}}^{*}_{i}g_{i}, {\mathcal{B}}^{*}_{j}g_{j}\rangle=\delta_{ij}\langle g_{i},g_{j}\rangle\) for any \(i, j\in\mathbb{N}\), any \(g_{i}\in H_{i}\), \(g_{j}\in H_{j}\). □
The following shows that the g-R-dual of the canonical dual g-frame is the “minimal” and has the “smallest distance” with \(\{A_{i}\}\) among the g-R-duals of all the alternate dual g-frames, which is a generalization of the result in [3, Theorem 4.5].
Theorem 2.8
Let\(\{A_{i}\}\)be a g-frame forH, \(\{\widetilde{A}_{i}\}\)be the canonical dual g-frame of\(\{A_{i}\}\), \(\{B_{i}\}\)be a dual g-frame of\(\{A_{i}\}\). \(\{{\mathcal{A}}_{i}\}\)and\(\{{\mathcal{B}}_{i}\}\)are the corresponding g-R-duals defined in Definition 2.2, respectively. Then the following are equivalent:
(1) ⇔ (2). By [3, Theorem 4.4], we obtain \({\mathcal{B}}_{i}=\widetilde{{\mathcal{A}}}_{i}+\Delta_{i}\) for any \(i\in\mathbb{N}\), where \(\{\Delta_{i}\}\) is a g-Bessel sequence for H such that \(\operatorname{ran}\theta_{\Delta}^{*}\subset(\operatorname {ran}\theta_{\mathcal{A}}^{*})^{\bot}\). Then, for every \(\{g_{i}\}\in \bigoplus_{i\in\mathbb{N}} {H_{i}}\), we get
Specially, if we take \(\{\delta_{ij}g_{i}\}_{j\in\mathbb{N}}\), then \(\Vert {\mathcal{B}}_{i}^{*}g_{i} \Vert \geq \Vert {\widetilde{{\mathcal {A}}}}_{i}^{*}g_{i} \Vert \). Hence, \(B_{i}=\widetilde{A}_{i}\) if and only if \(\Delta_{i}=0\) for any \(i\in\mathbb{N}\).
(2) ⇔ (3). By Lemma 2.7, for any \(i\in \mathbb{N}\), we obtain
3 Characterization of the Schauder basis-like properties of g-R-dual
Suppose \(\{A_{i}\}\) is a g-Bessel sequence for H, \(\{{\mathcal {A}}_{i}\}\) is its g-R-dual defined in Definition 2.2. We will characterize the Schauder basis-like properties (g-completeness, g-w-linearly independence, g-minimality) of \(\{{\mathcal{A}}_{i}\}\) in terms of \(\{A_{i}\}\).
Theorem 3.1
Let\(\{A_{i}\}\)be a g-Bessel sequence forH, \(\{{\mathcal{A}}_{i}\}\)be its g-R-dual defined in Definition 2.2. Then the following are equivalent:
(1) ⇔ (2). By Definition 2.2, \(\theta _{\mathcal{A}}^{*}=\theta_{\varGamma}^{*}\theta_{A}\theta_{\varLambda }^{*}\). For arbitrary \(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\), we have \(\{g_{i}\}\in\operatorname{ker}\theta_{\mathcal {A}}^{*}\) if and only if \(\theta_{\varLambda}^{*}\{g_{i}\}\in\operatorname {ker}\theta_{A}\). Then \(\{A_{i}\}\) is g-complete if and only if \(\operatorname{ker}\theta_{\mathcal{A}}^{*}=\{0\}\), i.e., \(\{{\mathcal {A}}_{i}\}\) is g-w-linearly independent.
(2) ⇔ (3). It is evident as \(\Vert \theta _{A}x_{n} \Vert ^{2}= \Vert \theta_{\mathcal{A}}^{*}\theta_{\varLambda}x_{n} \Vert ^{2}\). □
Now we have the next special result. By [4, Theorem 5.2], if \(\{A_{i}\}\) is a g-frame sequence for H, the existing condition of the g-biorthonormal sequence means the minimality of \(\{A_{i}\}\).
Theorem 3.2
Let\(\{A_{i}\}\)be a g-Bessel sequence forH, \(\{{\mathcal{A}}_{i}\}\)defined in Definition 2.2be its g-R-dual. If there exists a sequence\(\{\Delta_{i}\}\)which is g-biorthonormal with\(\{{\mathcal {A}}_{i}\}\)such that\(\Delta_{i}^{*}\)is injective for any\(i\in \mathbb{N}\), then
(1)
there are constants\(0< c_{i}\leq1\)for arbitrary\(i\in\mathbb{N}\)such that\(\Vert c_{i}g_{i} \Vert \leq \Vert \sum_{j\in \mathbb{N}}{\mathcal{A}}_{j}^{*}g_{j} \Vert \)for any\(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\);
(2)
there are constants\(0< a_{i}\)for arbitrary\(i\in \mathbb{N}\)such that
Take arbitrary \(h_{i}\in H_{i}\) and \(\Vert h_{i} \Vert =1\) and let \(c_{i}=\min\{ 1, \frac{1}{ \Vert \Delta_{i} \Vert }\}\) for every \(i\in\mathbb{N}\). Since \(\langle{\mathcal{A}}_{i}^{*}g_{i}, \Delta_{j}^{*}g_{j}\rangle =\delta_{ij}\langle g_{i}, g_{j}\rangle\) for any \(i, j\in\mathbb {N}\), \(g_{i}\in H_{i}\)\(g_{j}\in H_{j}\), we have
The converse is evident since \(\Vert a_{i}g_{i} \Vert ^{2}\leq \Vert \{a_{i}g_{i}\} \Vert ^{2}\). □
In the following we illustrate that the g-R-dual \(\{{\mathcal{A}}_{i}\} \) is a g-basic sequence by the properties of \(\{A_{i}\}\), which also shows the conclusion of Theorem 2.6 from another perspective. It can be realized as a kind of g-completeness of \(\{{\mathcal{A}}_{i}\}\).
Theorem 3.3
Let\(\{A_{i}\}\)be a g-frame sequence forH, \(\{{\mathcal{A}}_{i}\}\)defined in Definition 2.2be its g-R-dual. Let\(P_{n}\)be the orthogonal projection fromHonto\(N_{n}:=\overline{\operatorname{span}} \{ \varLambda^{*}_{i}H_{i}\}_{i=1}^{n}\)for any\(n\in\mathbb{N}\). Then the following are equivalent:
(1)
\(\{{\mathcal{A}_{i}}\}\)a g-basic sequence forH.
(2)
There exists a constant\(0< b<\infty\)such that\(\sum_{i\in\mathbb{N}} \Vert A_{i}P_{n}f \Vert ^{2}\leq b\sum_{i\in\mathbb{N}} \Vert A_{i}f \Vert ^{2}\)for any\(n\in\mathbb{N}\), any\(f\in H\).
(3)
There exists a constant\(0< b<\infty\)such that\(S_{AP_{n}}\leq bS_{A}\)for any\(n\in\mathbb{N}\), where\(S_{AP_{n}}\)is the frame operator of the g-Bessel sequence\(\{A_{i}P_{n}\}_{i\in \mathbb{N}}\).
Let \({\mathbb{I}}=\{j\in{\mathbb{N}}:{\mathcal{A}}_{j}^{*}=\theta _{\varGamma}^{*}\theta_{A}\varLambda^{*}_{j}\neq0\}\). Without loss of generality, we can suppose \({\mathcal{A}}_{i}\neq0\) for any \(i\in \mathbb{N}\).
(1) ⇔ (2). By [4, Theorem 3.3], \(\{{\mathcal {A}_{i}}\}\) is a g-basic sequence for H if and only if there exists a constant \(0< b<\infty\) such that, for arbitrary \(n\leq m\), any \(\{ g_{i}\}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\), one has
where \(x=\sum_{i=1}^{m}\varLambda_{i}^{*}g_{i}\). Since \(P_{n}\varLambda _{i}^{*}=0\) for every \(i\in\mathbb{N}\) such that \(n< i\leq m\), \(\sum_{i=1}^{n}\varLambda_{i}^{*}g_{i}=P_{n}x\). Similarly, we have \(\Vert \sum_{i=1}^{n} {\mathcal{A}}_{i}^{*}g_{i} \Vert ^{2}=\sum_{i\in\mathbb {N}} \Vert A_{i}P_{n}x \Vert ^{2}\).
(2) ⇔ (3). (2) is equivalent to \(\langle S_{AP_{n}}f, f\rangle=\langle\theta_{A}P_{n} f, \theta _{A}P_{n} f\rangle\leq b\langle Sf,f\rangle\) for any \(f\in H\), which is obvious.
By [4, Lemma 2.16], \(\{{\mathcal{A}}_{i}\}\) is a g-Riesz sequence for H. Then \({\mathcal{A}}_{i}\neq0\) for any \(i\in\mathbb {N}\). By Definition 2.2, we have \({\mathcal{A}}_{i}^{*}=\theta ^{*}_{\varGamma}\theta_{A}\varLambda^{*}_{i}\). Then \(\theta_{A}\varLambda ^{*}_{i}\neq0\), i.e., \(\sum_{i\in\mathbb{N}} \Vert A_{i}\varLambda _{i}^{*}g_{i} \Vert ^{2}\neq0\) for any \(i\in\mathbb{N}\), \(g_{i}\in H_{i}\). Hence,
Therefore, we only need to show the g-completeness of \(\{A_{i}\}\) in H.
Suppose there exists \(f\in H\), \(f\neq0\) such that \(\langle A_{i}^{*}g_{i}, f\rangle=0\) for arbitrary \(i\in\mathbb{N}\), \(g_{i}\in H_{i}\). Obviously, there is a sequence \(\{f_{i}\}\in\bigoplus_{i\in\mathbb{N}} {H_{i}}\) such that \(f=\sum_{i\in\mathbb {N}}\varLambda_{i}^{*}f_{i}\). Assume \(k\in{\mathbb{N}}\) is the smallest positive integer such that \(f_{i}\neq0\). Then \(P_{k}f=\varLambda _{k}^{*}f_{k}\). We get
\(\{A_{i}\}\)is exact (i.e., if it ceases to be a g-frame whenever any one of its elements is removed), and the canonical dual g-frame is biorthonormal with\(\{ A_{i}\}\).
Proof
The equivalence of (1), (2), (3) can be obtained by [4, Lemma 2.16]. By [9, Corollary 2.6], we get the equivalence of (3) and (6). Since \(\{A_{i}\}\) is a g-frame, we get \(\sum_{i\in\mathbb{N}} \Vert {\mathcal{A}}_{i}x_{n} \Vert ^{2}= \Vert \theta _{A}^{*}\theta_{\varGamma}x_{n} \Vert ^{2}\). Then (5) holds if and only if \(\theta_{A}^{*}\) is injective, i.e., (3) holds.
Similarly, by Definition 2.2, we have \(\theta_{\mathcal {A}}=\theta_{\varLambda}\theta_{A}^{*}\theta_{\varGamma}\). For any \(f\in H\), we obtain \(f\in\operatorname{ker}\theta_{\mathcal{A}}\) if and only if \(\theta_{\varGamma}f\in\operatorname{ker}\theta_{A}^{*}\). Thus we get the equivalence of (3), (4) by Theorem 2.6. □
4 G-R-dual and the g-orthogonal sequence
4.1 The characterization of g-R-dual
Let \(\{\varLambda_{i}\}\) be a g-orthonormal basis for H. In this section we mainly investigate the conditions under which a g-Riesz sequence \(\{{\mathcal{A}}_{i}\}\) is the g-R-dual of a g-frame \(\{A_{i}\} \). We denote \(\{\widetilde{{\mathcal{A}}}_{i}\}\) as the canonical dual g-frame of \(\{{\mathcal{A}}_{i}\}\), which is also a g-Riesz sequence. Define \(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta_{\widetilde{\mathcal {A}}}\) for any \(i\in\mathbb{N}\). Then
Evidently, \(\{C_{i}\}\) is a g-Bessel sequence for H. Let \(M=\operatorname {ran}\theta_{\mathcal{A}}^{*}\). Thus \(\operatorname{ran}\theta _{C}^{*}\subset M\). By Lemma 1.2, we also get \({\mathcal{A}}_{j}C^{*}_{i}=\varLambda _{j}A_{i}^{*}\) for any \(i\in\mathbb{N}\).
Proposition 4.1
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal {A}}_{i}\}\)be a g-Riesz basis forM, \(\{\widetilde{{\mathcal {A}}}_{i}\}\)be the canonical dual g-frame of\(\{{\mathcal{A}}_{i}\}\)inM, whereMis a closed subspace ofH. For any sequence\(\{ A_{i}\}\), we have the following:
(1)
There exists a sequence\(\{\varGamma'_{i}\}\)such that\(A_{i}=\varGamma'_{i}\theta_{\mathcal{A}}^{*}\theta_{\varLambda}\)for any\(i\in\mathbb{N}\), i.e., \(A^{*}_{i}g_{i}=\sum_{j\in \mathbb{N}}\varLambda_{j}^{*}{\mathcal{A}}_{j}{\varGamma'}_{i}^{*}g_{i}\)for any\(g_{i}\in H_{i}\).
(2)
The sequence\(\{\varGamma'_{i}\}\)satisfying\(A_{i}=\varGamma'_{i}\theta_{\mathcal{A}}^{*}\theta_{\varLambda}\)can be written as\(\varGamma'_{i}=C_{i}+D_{i}\)for every\(i\in\mathbb{N}\), where\(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta_{\widetilde{\mathcal {A}}}\), \(D_{i}\in B(H,H_{i})\)and\(\operatorname{ran}D^{*}_{i}\subset M^{\bot}\).
(3)
If\(H=M\), the sequence\(\{\varGamma'_{i}\}\)satisfying\(A_{i}=\varGamma'_{i}\theta_{\mathcal{A}}^{*}\theta_{\varLambda }\)has the unique solution\(\varGamma'_{i}=C_{i}\)for any\(i\in\mathbb {N}\), where\(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta_{\widetilde {\mathcal{A}}}\).
Proof
(1) Since \(A_{i}^{*}g_{i}=\sum_{j\in\mathbb{N}}\varLambda _{j}^{*}\varLambda_{j}A_{i}^{*}g_{i}\) for any \(i\in\mathbb{N}\), \(g_{i}\in H_{i}\) and \({\mathcal{A}}_{j}C^{*}_{i}=\varLambda _{j}A_{i}^{*}\), we have \(A_{i}^{*}g_{i}=\sum_{j\in\mathbb{N}}\varLambda _{j}^{*}{\mathcal{A}}_{j}C^{*}_{i}g_{i}\). We take \(\varGamma'_{i}=C_{i}\).
(2) For any \(i\in\mathbb{N}\), take arbitrary operator \(D_{i}\in B(M^{\bot}, H_{i})\). Obviously, \(\operatorname{ran}D^{*}_{i}\subset M^{\bot }\) is satisfied. Let \(\varGamma'_{i}=C_{i}+D_{i}\). Since \(M=\operatorname {ran}\theta_{\mathcal{A}}^{*}\), by (1), we have
For the converse, suppose \(A_{i}=\varGamma'_{i}\theta_{\mathcal {A}}^{*}\theta_{\varLambda}\) for any \(i\in\mathbb{N}\). By (1), \(C_{i}\theta_{\mathcal{A}}^{*}\theta_{\varLambda}=A_{i}\). Let \(D_{i}=\varGamma'_{i}-C_{i}\). Hence, \(D_{i}\theta_{\mathcal {A}}^{*}\theta_{\varLambda}=0\). Since \(M=\operatorname{ran}\theta_{\mathcal{A}}^{*}\), \(M\subset\operatorname {ker}D_{i}\). Thus \(\operatorname{ran}D^{*}_{i}\subset M^{\bot}\).
(3) If \(H=M\), we have \(D_{i}=0\) for any \(i\in\mathbb{N}\) from (2). □
Proposition 4.1 did not have any assumption on \(\{A_{i}\}\) or use any relationship between \(\{A_{i}\}\) and \(\{{\mathcal{A}}_{i}\}\).
The next result exhibits that \(\{C_{i}\}\) and \(\{A_{i}\}\) have the common properties.
Proposition 4.2
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal {A}}_{i}\}\)be a g-Riesz basis forMwith the frame boundscandd, \(\{\widetilde{{\mathcal{A}}}_{i}\}\)be the canonical dual g-frame of\(\{{\mathcal{A}}_{i}\}\)inM, whereMis a closed subspace ofH. For a sequence\(\{A_{i}\}\), define\(C_{i}=A_{i}\theta_{\varLambda }^{*}\theta_{\widetilde{\mathcal{A}}}\), for any\(i\in\mathbb{N}\), we have
(1)
If\(\{A_{i}\}\)is a g-Bessel sequence forHwith the upper boundb, then\(\{C_{i}\}\)is a g-Bessel sequence forHwith the upper bound\(bc^{-1}\). Moreover, for any\(\{g_{i}\}\in\bigoplus_{i\in\mathbb {N}}{H_{i}}\), we have
$$ c \biggl\Vert \sum_{i\in\mathbb {N}}C^{*}_{i}g_{i} \biggr\Vert ^{2}\leq \biggl\Vert \sum_{i\in\mathbb {N}}A^{*}_{i}g_{i} \biggr\Vert ^{2}\leq d \biggl\Vert \sum _{i\in\mathbb {N}}C^{*}_{i}g_{i} \biggr\Vert ^{2}. $$
Specially, \(\{A_{i}\}\)is g-w-linearly independent if and only if\(\{ C_{i}\}\)is g-w-linearly independent.
(2)
If\(\{A_{i}\}\)is a g-frame forHwith the frame bounds\(a, b\), then\(\{C_{i}\}\)is a g-frame forMwith the frame bounds\(ad^{-1}, bc^{-1}\).
(3)
If\(\{A_{i}\}\)is a g-Riesz basis forHwith the frame bounds\(a, b\), then\(\{C_{i}\}\)is a g-Riesz basis forMwith the frame bounds\(ad^{-1}, bc^{-1}\).
(4)
If\(\{C_{i}\}\)is a g-Bessel sequence forHwith the upper bound\(b_{1}\), then\(\{A_{i}\}\)is a g-Bessel sequence forHwith the upper bound\(b_{1}d\).
(5)
If\(\{C_{i}\}\)is a g-frame forMwith the frame bounds\(a_{1}, b_{1}\), then\(\{A_{i}\}\)is a g-frame forHwith the frame bounds\(a_{1}c, b_{1}d\).
(6)
If\(\{C_{i}\}\)is a g-Riesz basis forMwith the frame bounds\(a_{1}, b_{1}\), then\(\{A_{i}\}\)is a g-Riesz basis forHwith the frame bounds\(a_{1}c, a_{1}d\).
Proof
(1) Since \(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta_{\widetilde{\mathcal {A}}}\) for any \(i\in\mathbb{N}\), for every \(f\in H\), we have
Moreover, because \(\theta_{C}^{*}=\theta_{\widetilde{\mathcal {A}}}^{*}\theta_{\varLambda}\theta_{A}^{*}\), for any \(\{g_{i}\}\in \bigoplus_{i\in\mathbb{N}}{H_{i}}\), we have
(3) Suppose \(\{A_{i}\}\) is a g-Riesz basis for H. Since \(\{C_{i}\}\) is a g-frame for M by (2) and is g-w-linearly independent by (1), \(\{C_{i}\}\) is a g-Riesz basis for M by [4, Lemma 2.16]. The frame bounds can be obtained by (2).
The rest is similar to the above. □
From the above, \(\{C_{i}\}\), \(\{A_{i}\}\) have the same properties, but the bounds may not be common.
Corollary 4.3
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal{A}}_{i}\}\)be a g-orthonormal basis forM, whereMis a closed subspace ofH. For a sequence\(\{A_{i}\}\), define\(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta_{\widetilde{\mathcal{A}}}\)for any\(i\in\mathbb{N}\), we have:
(1)
\(\{C_{i}\}\)is a g-Bessel sequence forHif and only if\(\{A_{i}\}\)is a g-Bessel sequence forHwith the same bound.
(2)
\(\{C_{i}\}\)is a g-frame forMif and only if\(\{ A_{i}\}\)is a g-frame forHwith the same bounds.
(3)
\(\{C_{i}\}\)is a g-Riesz basis forMif and only if\(\{A_{i}\}\)is a g-Riesz basis forHwith the same bounds.
Proof
Take \(c=d=1\) by the proof of Proposition 4.2, which can be obtained directly. □
Let \(\{{\mathcal{A}}_{i}\}\) be a g-Riesz basis for M, where M is a closed subspace of H. Let \({\mathscr{A}}_{i}={\mathcal {A}}_{i}S_{\mathcal{A}}^{-\frac{1}{2}}\) for any \(i\in\mathbb{N}\), where \(S_{\mathcal{A}}\) is the frame operator of \(\{{\mathcal{A}}_{i}\} \). Then \(\{{\mathscr{A}}_{i}\}\) is a g-orthonormal basis for M. Let \(\{\varLambda_{i}\}\) be a g-orthonormal basis for H and \(\varTheta=\theta _{\varLambda}^{*}\theta_{\mathscr{A}}\). Obviously, \(\varTheta: M\rightarrow H\) is unitary and \({\mathscr{A}}_{i}=\varLambda_{i}\varTheta\). Then we have the following result.
Proposition 4.4
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal {A}}_{i}\}\)be a g-Riesz basis forMwith the frame bounds\(c, d\), whereMis a closed subspace ofH, \(\{A_{i}\}\)be a g-frame forHwith the frame bounds\(a, b\). Define\(C_{i}=A_{i}\theta_{\varLambda }^{*}\theta_{\widetilde{\mathcal{A}}}\)for every\(i\in\mathbb{N}\). Then the following are equivalent:
By Proposition 4.2, \(\{C_{i}\}\) is a g-frame for M. Since \(\theta_{C}=\theta_{A}\theta_{\varLambda}^{*}\theta_{\widetilde {\mathcal{A}}}\) and \(\theta_{\widetilde{\mathcal{A}}}=\theta_{\varLambda }\varTheta S_{\mathcal{A}}^{-\frac{1}{2}}\), we have \(S_{C}=S_{\mathcal{A}}^{-\frac{1}{2}}\varTheta^{*}S_{A}\varTheta S_{\mathcal {A}}^{-\frac{1}{2}}\). Obviously, \(S_{C}=P\) if and only if \(S_{\mathcal{A}}=\varTheta ^{*}S_{A}\varTheta\), where P is the orthogonal projection from H onto M. □
If \(\{A_{i}\}\) is a tight g-frame for H with the bound a. Let \(\{ {\mathcal{A}}_{i}\}\) be a tight g-Riesz basis for M with frame bound a. Then \(S_{A}=aI\), \(S_{\mathcal{A}}=aP\). Thus Proposition 4.4(2) holds obviously. Then we get Corollary 4.6 directly.
Proposition 4.5
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal {A}}_{i}\}\)be a g-Riesz basis forM, whereMis a closed subspace ofH. If\(\{A_{i}\}\)is a g-frame forH, define\(C_{i}=A_{i}\theta _{\varLambda}^{*}\theta_{\widetilde{\mathcal{A}}}\)for any\(i\in\mathbb {N}\). Then the following are equivalent:
(1)
If\(\{{\mathcal{A}}_{i}\}\)is the g-R-dual sequence of\(\{A_{i}\}\)with respect to two g-orthonormal bases\(\{\varLambda_{i}\} \), \(\{\varGamma_{i}\}\).
(2)
There exists a g-orthonormal basis\(\{\varGamma_{i}\} \)forHsuch that\(A_{i}=\varGamma_{i}\theta_{\mathcal{A}}^{*}\theta _{\varLambda}\)for every\(i\in\mathbb{N}\).
(3)
There exists a g-orthonormal basis\(\{\varGamma_{i}\} \)forHsuch that\(C_{i}=\varGamma_{i}P\)for every\(i\in\mathbb{N}\), wherePis the orthogonal projection fromHontoM.
(4)
\(\{C_{i}\}\)is a Parseval g-frame forMand\(\operatorname{dim}\operatorname{ker}\theta_{C}^{*}=\operatorname{dim}M^{\bot}\).
(1) ⇒ (2) By Definition 2.2, we have \({\mathcal {A}}_{i}^{*}=\theta_{\varGamma}^{*}\theta_{A}\varLambda_{i}^{*}\) for every \(i\in\mathbb{N}\). Hence, \(A_{i}=\varGamma_{i}\theta_{\mathcal {A}}^{*}\theta_{\varLambda}\).
(2) ⇒ (1) It is obvious by Definition 2.2. The equivalence of (2) and (3) can be obtained by Proposition 4.1.
(3) ⇒ (4) For any \(\{g_{i}\}\in\bigoplus_{i\in \mathbb{N}}{H_{i}}\), we have
Obviously, \(\{g_{i}\}\in\operatorname{ker}\theta_{C}^{*}\) if and only if \(\theta_{\varGamma}^{*}\{g_{i}\}\in M^{\bot}\). Then \(\operatorname {dim}\operatorname{ker}\theta_{C}^{*}=\operatorname{dim}M^{\bot}\) as \(\theta _{\varGamma}\) is unitary. Evidently, \(\{C_{i}\}\) is a Parseval g-frame for M.
(4) ⇒ (3) Suppose \(\{C_{i}\}\) is a Parseval g-frame for M. Let \(K=M\oplus(\operatorname{ran}\theta_{C})^{\bot}\), \(T_{i}=C_{i}\oplus P_{i}Q^{\bot}\) for any \(i\in\mathbb{N}\), where \(Q, P_{i}\) are the orthogonal projection from \(\bigoplus_{i\in \mathbb{N}}{H_{i}}\) onto \(\operatorname{ran}\theta_{C}\), \(H_{i}\), respectively, for every \(i\in\mathbb{N}\). It is easy to get \(\{T_{i}\} \) is a g-orthonormal basis for K by [7, Theorem 4.1].
Since \(\operatorname{dim}\operatorname{ker}\theta_{C}^{*}=\operatorname{dim}M^{\bot }\), there exists a unitary operator \(V: M^{\bot}\rightarrow\operatorname {ker}\theta_{C}^{*}\). Let \(\varGamma_{i}=T_{i}(P\oplus V)=C_{i}\oplus P_{i}Q^{\bot}V\) for every \(i\in\mathbb{N}\). As \(P\oplus V: M\oplus M^{\bot}\rightarrow M\oplus(\operatorname{ran}\theta _{C})^{\bot}\) is unitary, where P is the orthogonal projection from H onto M, we see that \(\{\varGamma_{i}\}\) is a g-orthonormal basis for H by [6, Theorem 3.5]. Obviously, we have \(C_{i}=\varGamma_{i}P\). The equivalence of (4), (5) is direct by Proposition 4.4. □
By Proposition 4.5, we can also get the following corollary, which was showed in [3, Theorem 2.7].
Corollary 4.6
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal{A}}_{i}\}\)be a tight g-Riesz basis forMwith the frame bounda, whereMis a closed subspace ofH. If\(\{A_{i}\}\)is a tight g-frame with the frame bounda. Then there exists a g-orthonormal basis\(\{\varGamma_{i}\}\)forHsuch that\(\{{\mathcal{A}}_{i}\}\)is the g-R-dual of\(\{A_{i}\}\)with respect to two g-orthonormal bases\(\{\varLambda_{i}\}\), \(\{\varGamma_{i}\} \)if and only if\(\operatorname{dim}\operatorname{ker}\theta_{C}^{*}=\operatorname {dim}M^{\bot}\), where\(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta _{\widetilde{\mathcal{A}}}\)for any\(i\in\mathbb{N}\).
Proof
By Proposition 4.2(3), \(\{C_{i}\}\) is a Parseval g-frame for M. It is obvious by Proposition 4.5. □
Corollary 4.7
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal{A}}_{i}\}\)be a g-Riesz basis forM, \(\{\widetilde {{\mathcal{A}}}_{i}\}\)be the canonical dual g-frame of\(\{{\mathcal {A}}_{i}\}\)inM, whereMis a closed subspace ofH. If\(\{A_{i}\} \)is a g-frame forH. Define\(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta _{\widetilde{\mathcal{A}}}\)for any\(i\in\mathbb{N}\). For any\(\{ g_{i}\}\in\bigoplus_{i\in\mathbb{N}}{H_{i}}\), let\(g=\theta_{\varLambda}^{*}\{g_{i}\}\in H\), \(h=\theta_{\mathcal{A}}^{*}\{ g_{i}\}\in M\). Then there exists a g-orthonormal basis\(\{\varGamma_{i}\} \)forHsuch that\(\{{\mathcal{A}}_{i}\}\)is the g-R-dual of\(\{ A_{i}\}\)with respect to two g-orthonormal bases\(\{\varLambda_{i}\}\), \(\{ \varGamma_{i}\}\)if and only if\(\sum_{i\in\mathbb {N}} \Vert A_{i}g \Vert ^{2}= \Vert h \Vert ^{2}\)and\(\operatorname{dim}\operatorname{ker}\theta _{C}^{*}=\operatorname{dim}M^{\bot}\).
The result now follows from Proposition 4.5 directly. □
4.2 The construction of orthogonal sequence
Now we will construct a sequence \(\{\varGamma'_{i}\}\) such \(A_{i}=\sum_{j\in\mathbb{N}}\varGamma'_{i}\widetilde{{\mathcal{A}}}_{j}^{*}\varLambda _{j}\), which is characterized in Proposition 4.1.
Proposition 4.8
Let\(\{\varLambda_{i}\}\)be a g-orthonormal basis forH, \(\{{\mathcal {A}}_{i}\}\)be a g-Riesz basis forM, \(\{\widetilde{{\mathcal {A}}}_{i}\}\)be the canonical dual g-frame of\(\{{\mathcal{A}}_{i}\}\)inM, whereMis a closed subspace ofH. If\(\operatorname{dim}M^{\bot }=\sum_{i}\operatorname{dim}H_{i}=\infty\), we have:
(1)
For any sequence\(\{A_{i}\}\), there exists a g-w-linearly independent sequence\(\{\varGamma'_{i}\}\)such that\(A_{i}=\sum_{j\in\mathbb{N}}\varGamma'_{i}\widetilde{{\mathcal {A}}}_{j}^{*}\varLambda_{j}\)for every\(i\in\mathbb{N}\).
(2)
For any g-Bessel sequence\(\{A_{i}\}\), there exists a norm-bounded and g-w-linearly independent sequence\(\{\varGamma'_{i}\} \)such that\(A_{i}=\sum_{j\in\mathbb{N}}\varGamma'_{i}\widetilde {{\mathcal{A}}}_{j}^{*}\varLambda_{j}\)for every\(i\in\mathbb{N}\).
(3)
For any operator sequence\(\{A_{i}\}\), there exists a g-orthogonal sequence\(\{\varGamma'_{i}\}\)such that\(A_{i}=\sum_{j\in\mathbb{N}}\varGamma'_{i}\widetilde{{\mathcal {A}}}_{j}^{*}\varLambda_{j}\)for every\(i\in\mathbb{N}\).
Proof
(1) Since \(\operatorname{dim}M^{\bot}=\sum_{i\in\mathbb{N}}\operatorname {dim}H_{i}\), there exists a g-orthonormal basis \(\{E_{i}\}\) for \(M^{\bot}\) by [5, Theorem 3.1] with \(E_{i}\in B(M^{\bot}, H_{i})\) for any \(i\in\mathbb{N}\). Let \(W_{i}=\overline{\operatorname {ran}} {E_{i}^{*}}\) for any \(i\in\mathbb{N}\). Then \(M^{\bot}= \bigoplus_{i\in\mathbb{N}}W_{i}\) and \(E_{i}: W_{i}\rightarrow H_{i}\) is unitary. Let \(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta _{\widetilde{\mathcal{A}}}\) for any \(i\in\mathbb{N}\). Then \({\mathcal {A}}_{i}E_{j}^{*}=0\) and \(C_{i}E_{j}^{*}=\sum_{k\in\mathbb {N}}A_{i}\varLambda_{k}^{*}\widetilde{{\mathcal{A}}}_{k}E_{j}^{*}=0\).
Since there exists an invertible operator \(D_{i}: W_{i}\rightarrow H_{i}\) for any \(i\in\mathbb{N}\), we see that \(D_{i}E_{i}^{*}+C_{i}E_{i}^{*}=D_{i}E_{i}^{*}\in B(H,H_{i})\) is invertible. Let \(\varGamma'_{i}=D_{i}+C_{i}\in B(H, H_{i})\). Obviously, \(\varGamma'_{i}\neq0\).
For any \(\{g_{i}\}\in\bigoplus_{i\in\mathbb{N}}H_{i}\), if \(\sum_{i\in\mathbb{N}}{\varGamma}_{i}^{'*}g_{i}=0\), then, for any \(j\in\mathbb{N}\), we have
(2) By the proof of (1), we can choose \(D_{i}\) such that \(\Vert D_{i} \Vert =1\) (if not, we choose \(D'_{i}= \frac{D_{i}}{ \Vert D_{i} \Vert }\)) for any \(i\in\mathbb{N}\). By Proposition 4.2, \(\{C_{i}\}\) is a g-Bessel sequence for M. Suppose the upper bound of \(\{C_{i}\}\) is b. Then \(\Vert C_{i} \Vert \leq b\). Hence, for every \(i\in\mathbb{N}\), \(g_{i}\in H_{i}\), we have
(3) By Proposition 4.1, the sequence \(\{\varGamma'_{i}\}\) such that \(A_{i}=\sum_{j\in\mathbb{N}}\varGamma'_{i}\widetilde {{\mathcal{A}}}_{j}^{*}\varLambda_{j}=\varGamma'_{i}\theta_{\widetilde {{\mathcal{A}}}}^{*}\theta_{\varLambda}\) can be written as \(\varGamma '_{i}=C_{i}+D_{i}\), where \(C_{i}=A_{i}\theta_{\varLambda}^{*}\theta _{\widetilde{\mathcal{A}}}\), \(\overline{\operatorname{ran}} D^{*}_{i}\subset M^{\bot}\) for any \(i\in\mathbb{N}\). For every \(i, j\in\mathbb{N}, i\neq j\), \(g_{i}\in H_{i}\), \(g_{j}\in H_{j}\), we have
We will use the following inductive procedure to construct \(\{D_{i}\}\) such that \(\overline{\operatorname{ran}} D^{*}_{i}\subset M^{\bot}\) and \(D_{j}D^{*}_{i}=-C_{j}C^{*}_{i}\) for every \(i,j\in\mathbb{N}\), \(i\neq j\). Let \(T_{ij}=-C_{i}C^{*}_{j}\in B(H_{j},H_{i})\). Then \(T_{ij}^{*}=T_{ji}\). Let \(I_{i}\) be the identity on \(H_{i}\).
(1) Let \(D_{1}^{*}=E_{1}^{*}\).
(2) Let \(D_{2}^{*}=E^{*}_{1}X_{1,2}^{*}+E_{2}^{*}\), where \(X_{1,2}^{*}=T_{12}\).
Obviously, \(D_{1}D_{2}^{*}=E_{1}E^{*}_{1}X_{1,2}^{*}+E_{1}E_{2}^{*}=T_{12}\). Then \(\varGamma'_{1}{\varGamma'_{2}}^{*}=0\).
3) For any \(k\in\mathbb{N}\), assuming that we have gotten operators \(D_{1}, D_{2}, \ldots, D_{k}\) in terms of \(X_{i,k}\in B(H_{i},H_{k})\) (\(i=1,\ldots, k-1\)) such that \(D_{k}^{*}=\sum_{i=1}^{k-1}E^{*}_{i}X_{i,k}^{*}+E_{k}^{*}\). Then, for \(k+1\), we define \(D_{k+1}\) by \(D_{k+1}^{*}=\sum_{i=1}^{k}E^{*}_{i}X_{i,k+1}^{*}+E_{k+1}^{*}\), where operators \(X_{i,k+1}\ (i=1,2,\ldots,k)\) are given by the following equation:
Obviously, we can obtain \(X_{i,k+1}\in B(H_{i},H_{k+1})\) (\(i=1,\ldots, k\)). Thus we have constructed the sequence \(\{D_{i}\}\) and obtained \(\{ \varGamma'_{i}\}\) by \(\varGamma'_{i}=C_{i}+D_{i}\) for any \(i\in\mathbb {N}\). Then \(\{\varGamma'_{i}\}\) such that \(\varGamma'_{i}{\varGamma '}_{j}^{*}=0\) for every \(i,j\in\mathbb{N}\) with \(i\neq j\).
Lastly, we show the sequence \(\{\varGamma_{i}'\}\) satisfies the desired condition: \(A_{i}=\sum_{j\in\mathbb{N}}\varGamma_{i}'\mathcal {A}_{j}^{*}\varLambda_{j}\) for all \(i\in\mathbb{N}\).
Since \((\operatorname{ker}D_{i})^{\bot}=\overline{\operatorname{ran}} D^{*}_{i}\subset M^{\bot}\) and \(\overline{\operatorname{ran}} \widetilde{{\mathcal {A}}}_{j}^{*}\subset M\) for any \(i, j\in\mathbb{N}\), we have
Hence, \(D_{i}\widetilde{{\mathcal{A}}}_{j}^{*}=0\) for any \(i, j\in \mathbb{N}\). On the other hand, since \(C_{i}=A_{i}\theta_{\varLambda }^{*}\theta_{\widetilde{\mathcal{A}}}\) for any \(i\in\mathbb{J}\), we get \({\mathcal{A}}_{j}C^{*}_{i}=\varLambda_{j}A_{i}^{*}\). By \(A_{i}^{*}g_{i}=\sum_{j\in\mathbb{N}}\varLambda_{j}^{*}\varLambda _{j}A_{i}^{*}g_{i}\) for any \(g_{i}\in H_{i}\), any \(i\in\mathbb{N}\), we have \(A_{i}^{*}g_{i}=\sum_{j\in\mathbb{N}}\varLambda _{j}^{*}{\mathcal{A}}_{j}C^{*}_{i}g_{i}\). So \(\sum_{j\in \mathbb{N}}C_{i}\widetilde{{\mathcal{A}}}_{j}^{*}\varLambda_{j}=A_{i}\) for any \(i\in\mathbb{N}\). Then
The authors would like to express their gratitude to the reviewers for their helpful comments and suggestions.
Availability of data and materials
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.