Subdyadic time–frequency analysis: Gabor frames, modulation spaces, and Miyachi multipliers

Vicente Vergara · 2025

Read Subdyadic time–frequency analysis: Gabor frames, modulation spaces, and Miyachi multipliers by Vicente Vergara (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-17740.txt.

Introduction

Vicente Vergara builds a new time frequency framework tuned to wave like phase functions that spread out over time. Standard tools split phase space in a regular dyadic way, while his subdyadic geometry adapts the split to how dispersive waves actually behave.

On that geometry he constructs stable Gabor frames, families of wave packets with controlled overlap, and defines matching modulation spaces for measuring signals. He proves key properties of those spaces and shows that a class of multipliers stays bounded on them. The framework also recovers a classic description of where a signal is rough in space and direction.

Catalog record

CREATORVicente Vergara
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE135 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-17740

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Subdyadic time–frequency analysis: Gabor frames, modulation spaces, and Miyachi multipliers Vicente Vergara Abstract We present a time–frequency framework adapted to dispersive phase functions via a subdyadic geometry in phase space. On top of this geometry we construct stable Gabor frames with quantitative control of overlap, almost orthogonality, and off-diagonal decay. Based on these frames we introduce modulation spaces consistent with the subdyadic scale and establish window and lattice independence, identifications in the Hilbertian case, duality, and natural inclusion relations. Within this setting we develop a theory for two-sided Miyachi multipliers, relying on discrete almost diagonalization and Wiener–Jaffard type results for well-localized matrices, and obtain boundedness on weighted modulation spaces. Finally, we define a Gabor-type wavefront set adapted to the subdyadic geometry and prove its invariance and ellipticity with respect to smooth order-zero pseudodifferential operators. Taken together, these results provide a unified tool both for global microlocal analysis and for the design of stable numerical schemes in high-frequency regimes. Keywords : Subdyadic Gabor frames; dispersive modulation spaces; Miyachi multipliers; Gabor wavefront set; pseudodifferential operators; time–frequency analysis. AMS MSC 2020 : Primary 42C15, 42B25, 35S05; Secondary 42B15, 35Q41, 46F12. 1 Introduction The Gabor transform (short-time Fourier transform, STFT) plays a central role in time–frequency analysis, as it provides a representation that is simultaneously localized in space and frequency and quantitatively reflects the uncertainty principle; its formalism allows one to construct stable frames, define and characterize modulation spaces, and design robust algorithms for feature extraction, compression, and filtering in signals and solutions to PDEs [ 9 , 8 , 10 ] . Beyond its foundational role in signal processing, the STFT and Gabor frames offer an efficient bridge between the geometry of phase space and numerical discretization: the possibility of adapting the window and the lattice to the dynamics of the problem enables sparse representations and natural preconditioners, with a direct impact on global microlocal analysis and on high-frequency computational methods. In a recent development, Jaming and Rathmair have shown that Gaussian Gabor frames allow one to tackle phase retrieval problems in a stable way by means of convex optimization techniques [ 16 ] . In their work, the modulus of the STFT (the spectrogram) sampled on a uniform lattice is interpreted as a family of quadratic measurements which, after a lifting procedure, lead to a matrix completion problem solved by semidefinite programming. The underlying geometry is Euclidean and isotropic, and the Gaussian window plays a crucial role in connecting the STFT with the Bargmann transform and entire functions of several complex variables. From this perspective, our subdyadic framework provides an alternative family of time–frequency frames, adapted to dispersive phases of the form |\xi|^{\alpha} and to the region \{|\xi|^{\alpha}\geq 1\} , which is natural for designing dispersive variants of these reconstruction (or numerical preconditioning) schemes in situations where the classical Euclidean geometry does not adequately reflect the dynamics of the problem. Dispersive phenomena governed by radial phases of the form |\xi|^{\alpha} , with \alpha>0 , exhibit an intrinsic anisotropy in phase space that is not well captured by the classical dyadic paradigm. In the frequency region |\xi|\sim 2^{j} , the relevant scale for operators with essentially linear phase in \xi naturally leads to cells (\Delta x,\Delta\xi)\sim(2^{-j},2^{j}) . However, when the phase is |\xi|^{\alpha} , the interplay between curvature and oscillation requires a subdyadic resolution : the effective frequency size becomes |\xi|^{1-\alpha} and its spatial dual |\xi|^{\alpha-1} . This redistribution of time–frequency resolution has materialized in refined subdyadic tools (square functions and maximal operators) that have proved effective in weighted harmonic analysis [ 1 ] and that are naturally connected to the theory of Miyachi multipliers [ 20 , 21 ] and to the classical body of Littlewood–Paley inequalities and extrapolation [ 25 ] . More precisely, let \alpha,\beta\in\mathbb{R} . We consider multipliers m:\mathbb{R}^{d}\setminus\{0\}\to\mathbb{C} supported in \{\xi\in\mathbb{R}^{d}:|\xi|^{\alpha}\geq 1\}. We say that m satisfies the Miyachi condition if |D^{\gamma}m(\xi)|\;\lesssim\;|\xi|^{-\beta+|\gamma|(\alpha-1)},\qquad\text{for all }\xi\text{ with }|\xi|^{\alpha}\geq 1, (1) for all multi-indices \gamma\in\mathbb{N}^{d} with |\gamma|\leq N_{d} , where N_{d}=\Big\lfloor\frac{d}{2}\Big\rfloor+1. The implicit constant may depend on d,\alpha,\beta and N_{d} , but not on \xi . A more flexible version, suitable for the analysis of Beltrán and Bennett [ 1 ] , consists in requiring only a local quadratic control of the derivatives. Let \mathcal{B} be a family of Euclidean balls in \mathbb{R}^{d} . For a ball B\subset\mathbb{R}^{d} we write r(B) for its radius and \mathrm{dist}(B,0) for the distance from the center of B to the origin. Given a multiplier m supported in \{|\xi|^{\alpha}\geq 1\} , we say that it satisfies a Hörmander–Miyachi type condition if \sup_{B}\;\mathrm{dist}(B,0)^{\,\beta+(1-\alpha)|\gamma|}\left(\frac{1}{|B|}\int_{B}|D^{\gamma}m(\xi)|^{2}\,d\xi\right)^{1/2}<\infty, (2) for all multi-indices \gamma with |\gamma|\leq N_{d} , where the supremum is taken over all \alpha -subdyadic balls B , that is, \mathrm{dist}(B,0)^{\alpha}\geq 1\qquad\text{and}\qquad r(B)\sim\mathrm{dist}(B,0)^{1-\alpha}, where r(B) denotes the radius of B . These balls provide a natural covering of the region \{|\xi|^{\alpha}\geq 1\} by blocks whose radial scale is precisely |\xi|^{1-\alpha} . The condition ( 2 ) is strictly weaker than the pointwise condition ( 1 ) above, but it still captures the same oscillatory and growth regime of the symbols. This is due to the fact that it measures the regularity of m in an averaged sense over each ball, and the weight \mathrm{dist}(B,0)^{\,\beta+(1-\alpha)|\gamma|} compensates the expected growth/decay of D^{\gamma}m . In analogy with the g -functions developed by Stein [ 24 ] , the authors of [ 1 ] define subdyadic variants of the g -functions, constructed using convolution with a function \varphi_{t} . The corresponding subdyadic square function is defined by g_{\alpha,\beta}(f)(x)=\left(\int_{t^{\alpha}\leq 1}\int_{|x-y|\leq t^{1-\alpha}}|f\ast\varphi_{t}(y)|^{2}\;t^{-(1-\alpha)d-2\beta}\,dy\,\frac{dt}{t}\right)^{1/2}. For \alpha=0 and \beta=0 (with a suitable choice of \varphi ) one recovers a variant of the classical Littlewood–Paley g -function. The parameter \beta introduces a radial weight in the t variable. By analogy with the classical g^{*}_{\lambda} function [ 24 ] , one introduces g^{*}_{\alpha,\beta,\lambda}(f)(x)=\left(\int_{t^{\alpha}\leq 1}\int_{\mathbb{R}^{d}}|f\ast\varphi_{t}(y)|^{2}\left(1+\frac{|x-y|}{t^{1-\alpha}}\right)^{-d\lambda}\;t^{-(1-\alpha)d-2\beta}\,dy\,\frac{dt}{t}\right)^{1/2}, with \lambda>0 . By construction, g^{*}_{\alpha,\beta,\lambda} is a pointwise majorant of g_{\alpha,\beta} , and for \lambda sufficiently large it enjoys better analytic properties (for instance, in weighted estimates). Two-sided Miyachi condition Let m:\mathbb{R}^{d}\setminus\{0\}\to\mathbb{C} . We say that m satisfies the two-sided Miyachi condition if there exist \sigma>d/2 and \beta\in\mathbb{R} such that, for every multi-index |\gamma|\leq\lfloor\sigma\rfloor , |D^{\gamma}m(\xi)|\lesssim\begin{cases}|\xi|^{-\beta+(\alpha-1)|\gamma|},&|\xi|\geq 1,\\[3.0pt] |\xi|^{-\beta_{0}+(\alpha_{0}-1)|\gamma|},&0<|\xi|<1,\end{cases} (3) with parameters (\alpha,\beta) at infinity and (\alpha_{0},\beta_{0}) near the origin (see [ 20 , 21 ] ). The corresponding symbol class naturally includes the model symbols m_{\alpha,\beta}(\xi)=|\xi|^{-\beta}e^{\,i|\xi|^{\alpha}}, which were studied in depth by Hirschman [ 12 ] , Wainger [ 27 ] , Fefferman [ 7 ] , Fefferman and Stein [ 6 ] , Miyachi [ 21 ] , among others. The classical results of Miyachi ensure that, under this condition, one obtains optimal L^{p}\to L^{q} bounds for the continuous linear operator T_{m}:=\mathcal{F}^{-1}[m\mathcal{F}(\cdot)] , with relations between (p,q,\alpha,\beta) that can be expressed in terms of Sobolev-type conditions. As a consequence of the main results obtained by Beltrán and Bennett [ 1 ] on the subdyadic functions g_{\alpha,\beta,\lambda},g^{*}_{\alpha,\beta,\lambda} , they prove the following result. Theorem 1.1 (Corollary 11, [ 1 ] ) . If m:\mathbb{R}^{d}\setminus\{0\}\to\mathbb{C} satisfies ( 3 ) for every \gamma\in\mathbb{N}^{d} with |\gamma|\leq N_{d} , then \int_{\mathbb{R}^{d}}|T_{m}f|^{2}w\lesssim\int_{\mathbb{R}^{d}}|f|^{2}M^{2}\mathcal{M}_{\alpha,\beta}M^{4}w, (4) where \mathcal{M}_{\alpha,\beta}w(x):=\sup_{(y,r)\in\Lambda_{\alpha}(x)}\frac{1}{|B(y,r)|^{1-2\beta/d}}\int_{B(y,r)}w, and \Lambda_{\alpha}(x)=\{(y,r)\in\mathbb{R}^{d}\times(0,\infty):|x-y|\leq r^{\,1-\alpha}\}. Here, M denotes the classical Hardy–Littlewood maximal operator and M^{k} stands for the k -fold composition of M with itself. This work proposes a

Read the complete text