---
title: Woven Frames in Hilbert Spaces
url: https://www.emergentmind.com/topics/woven-frames
type: topic
---

# Woven Frames in Hilbert Spaces

Woven frames are families of frames indexed over a common set whose coordinatewise interlacings remain frames with uniform bounds. In the standard two-frame formulation, frames \(\Phi=\{\varphi_i\}_{i\in I}\) and \(\Psi=\{\psi_i\}_{i\in I}\) for a Hilbert space \(H\) are woven if there exist constants \(0<A\le B\) such that for every subset \(\sigma\subset I\), the mixed family \(\{\varphi_i\}_{i\in \sigma}\cup \{\psi_i\}_{i\in \sigma^c}\) is a frame for \(H\) with the same bounds \(A,B\). More generally, for finitely many frames \(F_j=\{f_{ij}\}_{i\in I}\), \(j\in[m]\), one requires that every partition \(\{\sigma_j\}_{j\in[m]}\) of \(I\) produce a mixed frame \(\bigcup_{j\in[m]}\{f_{ij}\}_{i\in \sigma_j}\) with universal bounds. The subject was introduced in “Weaving Frames” and has since developed into a substantial branch of frame theory with operator-theoretic, geometric, perturbative, time-frequency, fusion, and \(K\)-relative variants [1503.03947][1808.05095].

## 1. Foundational definition and basic equivalences

A countable family \(\{f_i\}_{i\in I}\subset H\) is a frame if there exist constants \(0<A\le B<\infty\) such that
\[
A\|f\|^2 \le \sum_{i\in I} |\langle f,f_i\rangle|^2 \le B\|f\|^2,\qquad \forall f\in H.
\]
For finitely many frames \(F_j=\{f_{ij}\}_{i\in I}\), \(j\in[m]\), the woven condition requires universal constants \(C,D>0\) such that for every partition \(\{\sigma_j\}_{j\in[m]}\) of \(I\),
\[
C\|f\|^2 \le \sum_{j\in[m]}\sum_{i\in \sigma_j} |\langle f,f_{ij}\rangle|^2 \le D\|f\|^2,\qquad \forall f\in H.
\]
A mixed family associated with a fixed partition is a weaving frame, and if only the uniform upper inequality holds, the family is Bessel woven [1808.05095].

For pairs, the definition specializes to
\[
A\|x\|^2 \le \sum_{i\in \sigma} |\langle x,f_i\rangle|^2 + \sum_{i\in \sigma^c} |\langle x,g_i\rangle|^2 \le B\|x\|^2,\qquad \forall x\in H,
\]
for every \(\sigma\subset I\). The universal character of the constants is the substantive requirement: woven-ness is stronger than the assertion that each individual interlacing is merely a frame [1503.03947][2101.07351].

An apparently weaker notion, weakly woven, asks only that every weaving be a frame, with bounds allowed to depend on the partition. For two frames this distinction collapses: weakly woven if and only if woven. This equivalence is one of the foundational structural facts of the theory and is repeatedly used in later perturbative and geometric arguments [1503.03947][2101.07351].

In finite dimensions, woven-ness reduces to spanning. If \(\mathbb H^M\) is finite-dimensional and \(\{f_i\}_{i=1}^N,\{g_i\}_{i=1}^N\) are frames, then they are woven if and only if every mixed family \(\{f_i\}_{i\in \sigma}\cup\{g_i\}_{i\in \sigma^c}\) spans \(\mathbb H^M\). In particular, if both frames are full spark and \(N\ge 2M-1\), then they are necessarily woven [1503.03947]. This finite-dimensional criterion should not be transferred naively to infinite dimensions: order and labeling matter, and even two orthonormal bases need not be woven if their labels are mismatched [1705.04166].

## 2. Operator-theoretic framework

A major development in the theory is the extension of the standard operator machinery of frame theory to woven families. For a fixed partition \(\{\sigma_j\}_{j\in[m]}\), the weaving analysis, synthesis, and frame operators are
\[
U_{\sigma_j}(f)=\{\langle f,f_{ij}\rangle\}_{i\in \sigma_j},\qquad
T_{\sigma_j}\{c_{ij}\}_{i\in \sigma_j}=\sum_{i\in \sigma_j} c_{ij}f_{ij},
\]
\[
S_{\sigma_j}f=T_{\sigma_j}U_{\sigma_j}f=\sum_{i\in \sigma_j}\langle f,f_{ij}\rangle f_{ij}.
\]
For the full family \(F=\{f_{ij}\}_{i\in I,\;j\in[m]}\), one uses the coefficient Hilbert space \(\bigoplus_{j\in[m]}\ell^2(I)\) and defines
\[
U_F(f)=\{\langle f,f_{ij}\rangle\}_{i\in I,\;j\in[m]},\qquad
T_F\{c_{ij}\}=\sum_{j\in[m]}\sum_{i\in I} c_{ij}f_{ij},
\]
\[
S_Ff=T_FU_Ff=\sum_{j\in[m]}\sum_{i\in I}\langle f,f_{ij}\rangle f_{ij}.
\]
The Gram operator is understood in the induced form \(G_F=U_FT_F\) [1808.05095].

This framework yields direct analogues of classical frame characterizations. A family of Bessel sequences \(\{f_{ij}\}_{i,j}\) is a woven frame with universal bounds \(C,D\) if and only if its woven frame operator satisfies
\[
CI_H \le S_F \le DI_H.
\]
Equivalently,
\[
C\|f\|^2 \le \langle S_Ff,f\rangle = \sum_{j\in[m]}\sum_{i\in I} |\langle f,f_{ij}\rangle|^2 \le D\|f\|^2.
\]
Thus positivity and bounded invertibility of \(S_F\) characterize woven frames exactly as in ordinary frame theory [1808.05095].

The same operator language gives reconstruction and normalization. When \(F\) is woven,
\[
f= \sum_{j\in[m]}\sum_{i\in I} \langle f,S_F^{-1}f_{ij}\rangle f_{ij}
 = \sum_{j\in[m]}\sum_{i\in I} \langle f,f_{ij}\rangle S_F^{-1}f_{ij},
\]
and \(\{S_F^{-1}f_{ij}\}_{i,j}\) is the standard dual woven family. Moreover,
\[
\{S_F^{-1/2}f_{ij}\}_{i,j}
\]
is a tight woven frame with universal bound \(1\), so every woven frame can be Parsevalized [1808.05095].

Several operator characterizations coexist in the literature. One formulation states that woven-ness is equivalent to the existence, for every partition, of a synthesis operator \(T\) satisfying \(AI_{\mathcal H}\le TT^*\); another states that a weaving is tight exactly when its mixed frame operator is \(AI_{\mathcal H}\) [1901.09432][1809.09465]. In the pair case, surjectivity of the mixed synthesis operator is also decisive: two frames are woven if and only if the synthesis operator of each weaving has range \(\mathcal H\) [1809.09465].

Operator transport laws are equally important. If \(E:H\to H\) is bounded and invertible, then
\[
E \text{ invertible } \iff \{Ef_{ij}\}_{i,j}\text{ is a woven frame},
\qquad
S_{EF}=ES_FE^*,
\]
with transformed bounds \(C\|E^{-1}\|^{-2}\) and \(D\|E\|^2\). In particular, unitary operators preserve the woven bounds [1808.05095]. At the same time, applying an invertible operator to a single frame need not produce a woven pair with the original frame; this distinction is explicitly illustrated by counterexamples [1809.09465].

## 3. Geometry, Riesz bases, and duality

The theory becomes especially rigid for Riesz bases. If two Riesz bases are woven as frames, then every weaving is in fact a Riesz basis. Consequently, a Riesz basis cannot be woven with a redundant frame: if \(\Phi\) is a Riesz basis and \(\Psi\) is a woven partner, then \(\Psi\) must also be a Riesz basis [1503.03947]. This is one of the central qualitative differences between woven Riesz bases and general woven frames.

A geometric characterization of woven Riesz bases is given in terms of separation of subspaces. For Riesz bases \(\Phi=\{\varphi_i\}\) and \(\Psi=\{\psi_i\}\), woven-ness is equivalent to the existence of a uniform positive lower bound on the distance, or equivalently angle, between
\[
\operatorname{span}\{\varphi_i\}_{i\in \sigma}
\quad\text{and}\quad
\operatorname{span}\{\psi_i\}_{i\in \sigma^c},
\]
for all \(\sigma\subset \mathbb N\) [1503.03947]. This subspace-transversality viewpoint is sharpened in later work.

A recent characterization reformulates woven Riesz bases entirely in terms of the change-of-basis operator. In finite dimensions, if \(\{v_1,\dots,v_n\}\) and \(\{w_1,\dots,w_n\}\) are bases and \(A\) is the change-of-basis matrix from \(\{v_i\}\) to \(\{w_i\}\), then the bases are woven if and only if every central submatrix \(A_J\) indexed by \(J\subset\{1,\dots,n\}\) is invertible. In the infinite-dimensional Riesz-basis setting, the analogue is that all central compressions \(A_J=E_J^*AE_J\) must be uniformly boundedly invertible. This provides a sharp operator-theoretic criterion for woven Riesz bases [2404.02229].

The same result admits a reconstruction interpretation. In finite dimensions, if \(A\in W_n\), meaning all central submatrices are invertible, then for every \(J\subset\{1,\dots,n\}\) one can reconstruct \(x\in\mathbb C^n\) uniquely from the mixed data
\[
Y^{(J)}(x)=\{(Ax)(j)\}_{j\in J}\cup \{x(j)\}_{j\in I\setminus J}.
\]
For the Fourier matrix \(F_n\), the paper shows \(F_n\in W_n\) when \(n\) is prime, so any vector in \(\mathbb C^p\) can be reconstructed from any weaving of its coordinates and Fourier coefficients [2404.02229].

Duality supplies another axis of structure. Every Riesz basis is woven with its canonical dual [1909.08835]. More generally, if a frame \(\varphi=\{\varphi_i\}\) satisfies
\[
\|I_{\mathcal H}-S_\varphi^{-1}\|^2 < \frac{A_\varphi}{B_\varphi},
\]
then \(\varphi\) has infinitely many dual frames woven with it; analogous statements hold for approximate duals under conditions involving \(T^*S_\varphi^{-1}\) [1909.08835]. A separate operator criterion shows that if \(S_\varphi^{-1}\ge I_{\mathcal H}\) and \(S_\varphi\) commutes with all partial frame operators \(S_{\varphi_J}\), then \(\varphi\) is woven with its canonical dual [1909.08835].

These positive duality results are not exhaustive. The literature presents sufficient conditions, not a complete characterization, for when a frame is woven with its canonical, alternate, or approximate dual [1909.08835].

## 4. Perturbation, stability, and construction methods

Perturbation theory is central because woven-ness is intended to model robust reconstruction under changes of sensing or representation systems. A clean operator criterion is the following: if \(F=\{f_i\}_{i\in I}\) is a frame with optimal lower frame bound \(A_F\), and \(G=\{g_i\}_{i\in I}\) is a Bessel sequence whose synthesis operator satisfies
\[
\|T_F-T_G\|<A_F^{1/2},
\]
then \((F,G)\) is a woven pair with explicit universal bounds
\[
\big(A_F^{1/2}-\|T_F-T_G\|\big)^2
\quad\text{and}\quad
B_F+B_G.
\]
This formulation allows \(G\) to be merely Bessel, and every weaving is handled uniformly because its synthesis operator is \(T_F+(T_G-T_F)P_\sigma\) [2101.07351].

The same paper derives a corollary for perturbations induced by an operator: if \(T\in B(\mathcal H)\) and
\[
\|(I_d-T)T_F\|<A_F^{1/2},
\]
then \(F\) and \(T(F)=\{Tf_i\}\) are woven [2101.07351]. It also proves that if \((F,G)\) is already woven and \(U,V\) are invertible operators sufficiently close in the sense that
\[
\|U^{-1}V-I_d\|^2 < \frac{C}{B_G}
\qquad\text{or}\qquad
\|V^{-1}U-I_d\|^2 < \frac{C}{B_F},
\]
where \(C\) is the optimal lower woven bound, then \(U(F)\) and \(V(G)\) remain woven [2101.07351].

A geometric necessary condition for a woven pair is expressed through angles between subspaces. If \((F,G)\) is woven, then
\[
\sup_{\sigma\subset I} c\big(N(T_{F,G}),R(Q_\sigma)\big)<1,
\]
where \(T_{F,G}(x\oplus y)=T_Fx+(T_G-T_F)y\) and \(Q_\sigma\) is an oblique projection. The paper explicitly emphasizes that this condition is necessary, not sufficient; it guarantees only a uniform frame-sequence type lower bound, not spanning of the whole space [2101.07351].

Earlier perturbation criteria remain influential. One sufficient condition says that if
\[
\left\|\sum_{i\in I} c_i(\varphi_i-\psi_i)\right\|
\le
\lambda_1\left\|\sum_{i\in I}c_i\varphi_i\right\|
+\lambda_2\left\|\sum_{i\in I}c_i\psi_i\right\|
+\mu\|(c_i)\|_{\ell^2},
\]
with
\[
\lambda_1\sqrt{B_\varphi}+\lambda_2\sqrt{B_\psi}+\mu<\sqrt{A_\varphi},
\]
then \(\varphi\) and \(\psi\) are woven [1909.08835]. Another operator-closeness result states that if a frame \(\{\varphi_i\}\) and a bounded operator \(T\) satisfy
\[
\|I_{\mathbb H}-T\|^2 < \frac{A}{B},
\]
then \(\{\varphi_i\}\) and \(\{T\varphi_i\}\) are woven [1503.03947].

Construction theory complements perturbation theory. One finite-dimensional method shows that if \(\{f_i\}_{i\in[m]}\) is a frame with \(f_{m+1}=0\), then \(\{f_i-f_{i+1}\}_{i\in[m]}\) is also a frame and the two are woven [1809.09465]. Another construction uses idempotent operators: if \(F\in\mathcal L(\mathcal H)\) satisfies \(F^2=F\), has closed range, and \(R(F)=R(F^*)\), then for a frame \(\{f_i\}\) of \(R(F^*)\), the family \(\{Ff_i\}\) is also a frame for \(R(F^*)\) and is woven with \(\{f_i\}\) [1809.09465]. The hypotheses are essential; the same paper provides counterexamples when \(F^2=F\) fails or \(R(F)\neq R(F^*)\) [1809.09465].

## 5. Extensions beyond ordinary frames

The woven paradigm extends systematically to operator-valued, fusion, and \(K\)-relative settings. In fusion-frame language, a weighted family of closed subspaces \(\{(W_i,\nu_i)\}_{i\in I}\) is a fusion frame if
\[
C\|f\|^2 \le \sum_{i\in I} \nu_i^2\|P_{W_i}f\|^2 \le D\|f\|^2.
\]
Two such families are woven fusion frames if every partition-induced mixture is again a fusion frame with universal bounds [1808.03765]. The main bridge theorem states that woven fusion frames are equivalent to woven ordinary frames obtained from weighted local frame sequences or weighted orthonormal bases spanning the constituent subspaces [1808.03765]. The same paper develops perturbation criteria, restriction to closed subspaces by intersection, and a woven Riesz decomposition notion for families of subspaces [1808.03765].

For \(K\)-frames, the lower bound is measured by \(\|K^*f\|^2\):
\[
A\|K^* f\|^2 \le \sum_{i\in I} |\langle f, \phi_i \rangle|^2 \le B\|f\|^2.
\]
A family of \(K\)-frames is \(K\)-woven if every partitioned mixture is again a \(K\)-frame with uniform \(K\)-frame bounds. The central operator characterization is
\[
AKK^* \le M_\sigma M_\sigma^*,
\]
where \(M_\sigma\) is the synthesis operator of the weaving determined by \(\sigma\). In the two-frame case, weakly \(K\)-woven and \(K\)-woven are equivalent, and the paper proves a Paley–Wiener type perturbation theorem in which quantitative closeness of two \(K\)-frames yields \(K\)-woven-ness [1710.09562].

The corresponding subspace-based theory is that of \(K\)-fusion frames. A weighted collection \(\{(W_i,w_i)\}_{i\in I}\) is a \(K\)-fusion frame if
\[
A\|K^*f\|^2 \le \sum_{i\in I} w_i^2\|P_{W_i}f\|^2 \le B\|f\|^2.
\]
Two such families are woven if
\[
A\|K^*f\|^2 \le \sum_{i\in \sigma} w_i^2\|P_{W_i}f\|^2 + \sum_{i\in \sigma^c} v_i^2\|P_{V_i}f\|^2 \le B\|f\|^2
\]
for every \(\sigma\subset I\). The theory includes transport under operators \(T\), equivalence with ordinary woven fusion frames on \(R(K)\) when \(R(K)\) is closed, Paley–Wiener type operator perturbation, and erasure criteria [1912.07530].

The generalized-frame, or \(g\)-frame, setting displays both continuity with and divergence from ordinary weaving. A \(g\)-frame is a sequence of operators \(\{\Lambda_m\in L(H,H_m)\}\) satisfying
\[
A\|h\|^2 \le \sum_{m\in\mathbb N}\|\Lambda_m h\|^2 \le B\|h\|^2.
\]
Two \(g\)-frames are woven when every subset \(\sigma\subseteq\mathbb N\) satisfies
\[
A\|h\|^2 \le \sum_{m\in \sigma}\|\Lambda_m h\|^2+\sum_{m\in \sigma^c}\|\Omega_m h\|^2 \le B\|h\|^2.
\]
A key result is that weaving \(g\)-frames are characterized by weaving of the induced ordinary frames \(\{\Lambda_m^*f_{n,m}\}\) and \(\{\Omega_m^*g_{n,m}\}\). Another is that a \(g\)-frame and any dual \(g\)-frame are woven, with universal bounds
\[
\min\left\{\frac1{2B_1},\frac1{2B_2}\right\}
\quad\text{and}\quad
B_1+B_2.
\]
At the same time, some rigidities of ordinary Riesz-basis weaving fail in the generalized setting: a \(g\)-frame can weave with a \(g\)-Riesz basis without itself being a \(g\)-Riesz basis, and a weaving of two \(g\)-exact frames need not be \(g\)-exact [2001.09432].

The continuous end of the theory is represented by woven continuous controlled \(K\)-g-fusion frames. These are measurable families \(\{(F_i(x),\Lambda_i(x),v_i(x))\}_{i\in[m],\,x\in X}\) over a measure space \((X,\mu)\) satisfying, for every partition \(\{\sigma_i\}\) of \(X\),
\[
A\|K^*f\|^2 \le \sum_{i=1}^m\int_{\sigma_i} v_i^2(x)\,
\big\langle \Lambda_i(x)P_{F_i(x)}Uf,\Lambda_i(x)P_{F_i(x)}Tf\big\rangle\,d\mu(x)
\le B\|f\|^2.
\]
The theory includes operator characterizations, invariance under invertible maps, restriction to subspaces, deletion principles, sufficient conditions via positivity and orthogonality, and a perturbation theorem [2111.03389].

## 6. Time-frequency constructions and broader packetized generalizations

Time-frequency analysis has supplied some of the most concrete woven constructions. For single-window Gabor systems
\[
\mathcal G(\phi_j,\Gamma)=\{\pi(\gamma)\phi_j:\gamma\in\Gamma\},
\qquad
\pi(\gamma)\phi(t)=M_\omega T_x\phi(t)=e^{2\pi i\omega t}\phi(t-x),
\]
woven-ness asks that every partition of the common index set \(\Gamma\) produce a frame. This idea is extended to multi-window Gabor systems
\[
\mathcal G(\{\phi_k^j\}_{k\in I},\Gamma),\qquad j=1,\dots,M,
\]
where at each phase-space point one chooses an entire local window family rather than a single atom [1705.04166].

The decisive tool in this setting is the time-frequency localization operator
\[
H_{m,\phi}f = V_\phi^*\, m\, V_\phi f,
\qquad
V_\phi f(\gamma)=\langle f,\pi(\gamma)\phi\rangle.
\]
The paper establishes two sufficient criteria ensuring that families of variable-window localization operators satisfy
\[
A\|f\|_2^2 \le \sum_{\gamma\in\Gamma}\|H_{\eta_\gamma,\phi_{i(\gamma)}}f\|_2^2 \le B\|f\|_2^2.
\]
One is a Hilbert-space norm criterion: if \(\|\phi_1-\phi_2\|_2<\frac12\), then corresponding families of localization operators remain uniformly stable. The other is a pointwise phase-space criterion comparing \(V_{\varphi_0}\phi_i\) to the Gaussian STFT \(V_{\varphi_0}\varphi_0\) [1705.04166].

After spectral truncation of the localization operators, the retained eigenfunctions generate multi-window Gabor frames, and because the criteria hold for arbitrary assignments \(\gamma\mapsto \phi_{i(\gamma)}\), all such frames are woven. This yields explicit woven examples from Hermite functions associated with localization on elliptic domains; the same framework extends to chirped or rotated Gaussians and generalized Hermite-type eigenfunctions [1705.04166].

A newer generalization replaces vectors by subspaces and introduces information packets: families of closed subspaces \(\{W_j\}_{j\in J}\) such that every \(f\in\mathcal H\) admits an unconditionally convergent expansion
\[
f=\sum_{j\in J} f_j,\qquad f_j\in W_j.
\]
Frames and fusion frames occur as special cases, but information packets are strictly more general. Woven information packets are defined by requiring that every partition-based recombination of \(N\) such families remain an information packet [2509.07503].

This broader notion yields explicit wavelet and Gabor constructions. For wavelets, if \(\widehat\psi\) is bounded, compactly supported, and satisfies local two-sided power bounds near the origin, then for every \(N\ge 2\), the \(N\) collections
\[
\{V_{Nj}\}_{j\in\mathbb Z},\;
\{V_{Nj+1}\}_{j\in\mathbb Z},\;
\dots,\;
\{V_{Nj+N-1}\}_{j\in\mathbb Z}
\]
are woven information packets, where
\[
V_j=\overline{\operatorname{span}\{D_{a^j}T_{kb}\psi\}_{k\in\mathbb Z}}.
\]
For Gabor systems, the paper proves that for any fixed \(N\in\mathbb N\), one can find a Gabor frame split into \(N\) woven information packets, but no single Gabor frame can work for all \(N\) because the density condition forces \(abN\le 1\) [2509.07503].

These time-frequency and packetized constructions underscore a recurring theme of the field: woven-ness is easiest to verify when a highly structured geometry permits one to control arbitrary recombinations uniformly. This suggests a broader interpretation of woven frames as a theory of partition-uniform stability, with ordinary frames, Riesz bases, Gabor systems, fusion frames, \(K\)-relative systems, and information packets all appearing as instances of the same organizing principle.

Source: https://www.emergentmind.com/topics/woven-frames