Wavelet Coorbit Spaces

• Wavelet coorbit spaces are smoothness spaces defined via continuous wavelet transforms and affine group representations, bridging continuous analysis and discrete decompositions.
• They recast representation‐theoretic properties into Fourier-side decomposition spaces, establishing links with Besov, Triebel–Lizorkin, and anisotropic spaces.
• Their discretization yields stable Banach frames with coefficient norms equivalent to classical dyadic decompositions, ensuring robust signal analysis.

Wavelet coorbit spaces are representation-theoretic smoothness spaces defined from continuous wavelet transforms. In their standard form, one starts from a semidirect product $G=\mathbb{R}^d\rtimes H$, where $H\leq GL(\mathbb{R}^d)$ is a dilation group, and measures a function or distribution $f$ by the size of its wavelet transform $W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$ in a weighted mixed-norm space on $G$. Modern theory identifies these spaces with Fourier-side decomposition spaces determined by the dual action of $H$, so the continuous transform, its discretization, and the induced frequency covering become different realizations of the same structure (Feichtinger et al., 2016, Führ et al., 2014).

1. Historical formation and conceptual scope

The modern theory of wavelet coorbit spaces is rooted in the convergence of three lines of development: the Frazier–Jawerth dyadic coefficient characterizations of Besov and Triebel–Lizorkin spaces, abstract coorbit theory in the sense of Feichtinger and Gröchenig, and Fourier-side decomposition spaces. In the Frazier–Jawerth setting, a distribution $f$ is analyzed into coefficients indexed by dyadic cubes

$$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$

and membership in a smoothness space is equivalent to membership of the coefficient sequence in a solid BK-space. In modern language, this is already a Banach-frame characterization, and the Frazier–Jawerth atoms can be written in representation-theoretic form as

$$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$

for $\ell(Q)=2^{-\nu}$ (Feichtinger et al., 2016).

Coorbit theory abstracts this mechanism. Starting from a unitary representation $H\leq GL(\mathbb{R}^d)$0 on a Hilbert space, it replaces the specific dyadic analysis by the voice transform

$H\leq GL(\mathbb{R}^d)$1

and defines function spaces by requiring $H\leq GL(\mathbb{R}^d)$2 to belong to a prescribed solid function space on the group. In the wavelet case, $H\leq GL(\mathbb{R}^d)$3 is the continuous wavelet transform, so the older dyadic coefficient theory appears as a discrete prototype of a more general representation-theoretic construction. The resulting viewpoint is that wavelet coorbit spaces are not an isolated class of spaces; they sit at the intersection of affine harmonic analysis, Banach-frame theory, and Fourier decomposition methods (Feichtinger et al., 2016).

A central shift in this development is that the continuous transform is no longer treated as a purely analytic device for producing coefficients. It becomes the concrete realization of a group representation, and the corresponding smoothness space is read off from the geometry of the associated dilation action. This is why coorbit theory, decomposition spaces, and wavelet-frame decompositions are now routinely regarded as different formulations of one structural phenomenon (Feichtinger et al., 2016).

2. Representation-theoretic construction

For a closed matrix group $H\leq GL(\mathbb{R}^d)$4, the relevant affine group is

$H\leq GL(\mathbb{R}^d)$5

with left Haar measure

$H\leq GL(\mathbb{R}^d)$6

The quasi-regular representation is

$H\leq GL(\mathbb{R}^d)$7

and the associated generalized wavelet transform is

$H\leq GL(\mathbb{R}^d)$8

(Feichtinger et al., 2016).

Abstractly, if $H\leq GL(\mathbb{R}^d)$9 carries a square-integrable unitary representation $f$0, the voice transform is

$f$1

For a suitable control weight $f$2, one defines the test space

$f$3

and the anti-dual reservoir $f$4. If

$f$5

then for $f$6 the coorbit space attached to a solid translation-invariant BF-space $f$7 is

$f$8

A main theorem is that $f$9 is independent of the chosen $W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$0 (Feichtinger et al., 2016).

For wavelet coorbit spaces associated with general dilation groups, a basic model class is

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$1

with norm

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$2

The corresponding coorbit space is

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$3

for a suitable reservoir $W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$4, and if $W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$5, then

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$6

with equivalent norms (Feichtinger et al., 2016).

A concrete specialization is the one-dimensional affine group

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$7

with left Haar measure

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$8

The wavelet representation is

$W_\psi f(x,h)=\langle f,\pi(x,h)\psi\rangle$9

and

$G$0

Using the affine weights

$G$1

one obtains

$G$2

and within this weighted-$G$3 framework the paper identifies

$G$4

with equivalent norms (Zimmermann, 2024).

In the more general integrably admissible setting, one often replaces the abstract anti-dual reservoir by the Fourier-adapted distribution space

$G$5

and defines

$G$6

This formulation is tailored to reducible as well as irreducible quasi-regular representations and restores analyzing-vector independence by restricting the analyzer class to $G$7 (Führ et al., 2019).

3. Dilation groups, admissibility, and analyzing wavelets

The defining geometric datum of a wavelet coorbit space is the dual action of the dilation group. For $G$8, it is

$G$9

or equivalently $H$0, depending on convention. The quasi-regular representation is irreducible and square-integrable exactly when there exists $H$1 such that the orbit

$H$2

is open and conull, and the stabilizer

$H$3

is compact. Such groups are called admissible dilation groups (Feichtinger et al., 2016).

This orbit criterion is the basic classification principle for generalized wavelet transforms. In dimension $H$4, standard examples are the diagonal group

$H$5

with dual orbit $H$6, the similitude group

$H$7

with orbit $H$8, and the shearlet-type groups

$H$9

whose dual orbit is

$f$0

(Feichtinger et al., 2016).

Analyzing wavelets are selected either abstractly, through the classes $f$1 and $f$2, or by explicit Fourier-side conditions. A basic sufficient criterion is

$f$3

so every compactly frequency-supported smooth wavelet inside the open orbit is a “better vector” and therefore yields Banach frames and atomic decompositions (Feichtinger et al., 2016).

For compactly supported wavelets, the decisive concept is vanishing moments on the blind spot $f$4. A function $f$5 has vanishing moments on $f$6 of order $f$7 if $f$8 and

$f$9

The orbit-adapted envelope

$$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$0

measures simultaneously the distance to the blind spot and decay at infinity. Integrals built from

$$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$1

encode the interaction of orbit geometry and dilation, and lead to explicit criteria ensuring that sufficiently many vanishing moments plus enough smoothness imply $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$2 (Führ, 2013).

A later simplification replaces direct estimates on $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$3 by comparison inequalities for the pullback

$$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$4

If one can control $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$5, $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$6, $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$7, and $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$8 by powers of $$Q_{\nu,k}=\left\{x\in\mathbb{R}^d:2^{-\nu}k_i\le x_i<2^{-\nu}(k_i+1),\ \forall i=1,\dots,d\right\},$$9, then temperate embeddedness and strong temperate embeddedness follow automatically, yielding explicit atom criteria for broad classes of groups, in particular all irreducibly admissible abelian dilation groups and generalized shearlet dilation groups (Führ et al., 2014).

In dimension $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$0, the orbit condition is especially effective: every admissible dilation group has a temperately embedded dual orbit. As a consequence, the general vanishing-moment criteria apply uniformly to all two-dimensional admissible dilation groups, including shearlet groups (Führ, 2012).

4. Fourier-side structure and decomposition spaces

The decisive bridge from wavelet coorbits to Fourier analysis is the frequency-side expression for the wavelet transform. For the quasi-regular representation,

$$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$1

Hence, if $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$2 is supported in a bounded set $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$3, then $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$4 only sees frequencies in

$$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$5

This is the structural reason why wavelet coorbit spaces admit a Fourier-side description by coverings induced from the dual action (Feichtinger et al., 2016).

A decomposition space is built from an admissible covering $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$6 of an open frequency region, a BAPU $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$7, an exponent $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$8, and a weighted sequence space $$\psi_Q(x)=2^{-\nu d/2}\psi_\nu(x-x_Q)=[\pi(x_Q,2^{-\nu})\psi](x)$$9. In the Fourier-side form used for wavelet coorbits,

$\ell(Q)=2^{-\nu}$0

and

$\ell(Q)=2^{-\nu}$1

(Feichtinger et al., 2016).

For wavelet coorbit spaces associated with an admissible dilation group $\ell(Q)=2^{-\nu}$2, Führ and Voigtlaender proved the isomorphism

$\ell(Q)=2^{-\nu}$3

where, for a well-spread family $\ell(Q)=2^{-\nu}$4 and a bounded open set $\ell(Q)=2^{-\nu}$5,

$\ell(Q)=2^{-\nu}$6

Thus the mixed-norm condition on the continuous wavelet transform is exactly equivalent to a decomposition-space norm for the frequency covering generated by the dual action (Feichtinger et al., 2016).

The same result was formulated directly as a Fourier-transform isomorphism: $\ell(Q)=2^{-\nu}$7 with $\ell(Q)=2^{-\nu}$8 obtained by transplanting

$\ell(Q)=2^{-\nu}$9

from the group to the dual orbit. This gives a purely Fourier-analytic description of wavelet coorbit spaces and explains why their basic examples are Besov-type spaces (Führ et al., 2014).

This identification has several consequences. First, it shows that the actual structure of a wavelet coorbit space is encoded by the induced frequency covering, not merely by the existence of a continuous transform. Second, it makes embeddings and equivalence problems accessible through general decomposition-space criteria. Third, it extends beyond the classical irreducible setting: for integrably admissible groups, one again obtains coorbit spaces realized as Besov-type decomposition spaces, and for one-parameter groups generated by expansive matrices these coincide precisely with the corresponding anisotropic Besov spaces (Führ et al., 2019).

5. Discretization, Banach frames, and coefficient norms

One of the central achievements of coorbit theory is the passage from continuous transforms to discrete coefficient systems. Abstractly, if $H\leq GL(\mathbb{R}^d)$00, then sufficiently dense and well-separated sample sets $H\leq GL(\mathbb{R}^d)$01 generate a Banach frame: $H\leq GL(\mathbb{R}^d)$02 with coefficients in a suitable solid BK-space $H\leq GL(\mathbb{R}^d)$03. The coefficient norm is equivalent to the coorbit norm, and this is the abstract analogue of the Frazier–Jawerth mechanism (Feichtinger et al., 2016).

In Zimmermann’s weighted-$H\leq GL(\mathbb{R}^d)$04 framework for the affine group, the discrete sets are regular affine lattices

$H\leq GL(\mathbb{R}^d)$05

with associated atoms

$H\leq GL(\mathbb{R}^d)$06

For $H\leq GL(\mathbb{R}^d)$07, the discretized weight is

$H\leq GL(\mathbb{R}^d)$08

Under sufficient vanishing moments, smoothness, and local oscillation assumptions, this yields a Banach frame for $H\leq GL(\mathbb{R}^d)$09, with coefficient norm

$H\leq GL(\mathbb{R}^d)$10

equivalent to the coorbit norm and hence to the corresponding homogeneous Besov norm (Zimmermann, 2024).

For general matrix dilation groups, the discretization theorem has the form “any sufficiently fine sampling.” Once $H\leq GL(\mathbb{R}^d)$11, there exists a neighborhood $H\leq GL(\mathbb{R}^d)$12 such that every $H\leq GL(\mathbb{R}^d)$13-dense relatively separated family $H\leq GL(\mathbb{R}^d)$14 yields

$H\leq GL(\mathbb{R}^d)$15

with unconditional convergence in $H\leq GL(\mathbb{R}^d)$16, and

$H\leq GL(\mathbb{R}^d)$17

The key point is that, once the analyzing wavelet satisfies the explicit decay, smoothness, and orbit-adapted vanishing-moment assumptions, the resulting discrete system works simultaneously for whole scales of coorbit spaces, including homogeneous Besov spaces and shearlet coorbit spaces (Führ, 2013).

This discretization theory is conceptually important because it shows that wavelet coorbit spaces are not only defined by continuous transforms. They admit stable atomic decompositions, Banach frames, and coefficient-space norms comparable to classical wavelet coefficient characterizations. In that sense, the continuous group-theoretic construction and the discrete wavelet expansion are two levels of the same theory (Feichtinger et al., 2016).

6. Examples, equivalence, and later extensions

For the affine group in one dimension, the weighted-$H\leq GL(\mathbb{R}^d)$18 theory recovers the diagonal Besov scale

$H\leq GL(\mathbb{R}^d)$19

while the mixed-norm formulation

$H\leq GL(\mathbb{R}^d)$20

is available after extending the coefficient spaces from weighted Lebesgue spaces to mixed norms (Zimmermann, 2024). In the isotropic higher-dimensional setting, the similitude group yields homogeneous Besov spaces, and in the modern decomposition-space formulation ordinary isotropic wavelets, anisotropic wavelets, and shearlets are all treated as instances of wavelet coorbit spaces attached to different dilation groups (Feichtinger et al., 2016).

Comparison of different dilation groups has become a major topic. In dimension $H\leq GL(\mathbb{R}^d)$21, irreducibly admissible groups fall into three model types—similitude, diagonal, and shearlet $H\leq GL(\mathbb{R}^d)$22—and coorbit equivalence is governed almost completely by the open dual orbit: if the orbit has $H\leq GL(\mathbb{R}^d)$23 or $H\leq GL(\mathbb{R}^d)$24 connected components, orbit equality determines the coorbit class, whereas in the $H\leq GL(\mathbb{R}^d)$25-component case of shearlet type, coorbit equivalence forces equality of the groups themselves (Asharaf et al., 6 Aug 2025). In dimension $H\leq GL(\mathbb{R}^d)$26, all irreducibly admissible matrix groups were classified up to conjugacy and finite-index extension; every such group has compactly supported admissible vectors, and all except one exceptional case, labeled $H\leq GL(\mathbb{R}^d)$27, have compactly supported atoms $H\leq GL(\mathbb{R}^d)$28 for every polynomially bounded control weight (Currey et al., 2016).

A later coarse-geometric theory treats both irreducible and reducible quasi-regular representations through integrably admissible dilation groups. In that framework, coorbit equivalence is characterized by equality of the essential frequency support and weak equivalence of the induced coverings, and for connectivity-respecting groups it is equivalent to a quasi-isometry between the large-scale geometries of the corresponding frequency-covering models. One consequence is that, among one-parameter groups, only the essentially isotropic case is coorbit equivalent to an irreducibly admissible group such as $H\leq GL(\mathbb{R}^d)$29 (Führ et al., 2024).

This reducible perspective is complemented by a Fréchet-target extension of coorbit theory in which the usual $H\leq GL(\mathbb{R}^d)$30 test space is replaced by a Fréchet space $H\leq GL(\mathbb{R}^d)$31, notably

$H\leq GL(\mathbb{R}^d)$32

That framework admits reproducing representations whose kernels are not $H\leq GL(\mathbb{R}^d)$33-integrable, including Shannon wavelets, whose kernel belongs to $H\leq GL(\mathbb{R}^d)$34 for every $H\leq GL(\mathbb{R}^d)$35 but not to $H\leq GL(\mathbb{R}^d)$36. In this sense, classical integrable-kernel wavelet coorbit theory is a special case of a broader reproducing-kernel theory (Dahlke et al., 2014).

There are also non-group and non-Euclidean extensions. In the warped time-frequency framework, logarithmic warping

$H\leq GL(\mathbb{R}^d)$37

produces continuous wavelet transforms, and the resulting coorbit spaces coincide with the classical wavelet coorbit spaces after the reciprocal change of scale parameter (Holighaus et al., 2015). For generalized shearlet dilation groups, dilational symmetries of the entire coorbit scale are characterized by quasi-isometries of the induced frequency metric, and for these groups the compatible dilations are exactly the normalizer: $H\leq GL(\mathbb{R}^d)$38 (Führ et al., 2023). Over disconnected local fields $H\leq GL(\mathbb{R}^d)$39, the affine group

$H\leq GL(\mathbb{R}^d)$40

again supports a full wavelet coorbit theory; the homogeneous Besov spaces $H\leq GL(\mathbb{R}^d)$41 are identified as coorbits, every nonzero $H\leq GL(\mathbb{R}^d)$42 yields tight wavelet frames on suitable discrete subsets of $H\leq GL(\mathbb{R}^d)$43, and orthonormal wavelet bases generated from $H\leq GL(\mathbb{R}^d)$44 are unconditional bases for all associated coorbit spaces (Abhinav et al., 10 Aug 2025).

A persistent point of interpretation is that wavelet coorbit spaces are not exhausted by the phrase “spaces defined by a mixed $H\leq GL(\mathbb{R}^d)$45-condition on a continuous wavelet transform.” That formulation captures the norm, but not the structure. The deeper invariant is the action of the dilation group on frequency space, which determines admissibility, the orbit geometry, the induced decomposition covering, the discretization, and the comparison theory between different wavelet systems (Feichtinger et al., 2016).