Quantitative spectral theory of dilation-adapted time-frequency localization operators: plunge-region profiles, corrected density laws, and generator decay (Daubechies-Paul, affine setting)

Daubechies and Paul (Inverse Problems, 1988) constructed time-frequency localization operators adapted to dilations instead of translations and modulations: using affine coherent states whose Fourier transforms are supported on a half-line, localization sets in the hyperbolic upper half-plane, and a commuting second-order differential operator, they obtained band-pass filters with explicit Laguerre-type eigenfunctions and closed-form eigenvalues. They explicitly left unresolved three coupled limitations: the generators decay only polynomially in time (about $|x|^{-2}$ for the base choice), causing large edge effects and a wide plunge region; the semiclassical/Nyquist counting heuristic fails to match the actual spectral counts; and the dependence of the plunge width and localization uniformity on the two-parameter generator family $(\beta,\gamma)$ and on the admissible set geometry was not analyzed. Later literature has resolved adjacent problems (an affine Nyquist rate for discrete coherent-state systems via von Neumann dimension; a Landau-type eigenvalue count for mixed-state affine localization operators; a characterization of when such counting laws hold by unimodularity and Folner sequences; Faber-Krahn-type optimality and optimal norm bounds for wavelet localization operators), and a sharp plunge-profile program exists only for the Weyl-Heisenberg case. The surviving open core is a quantitative spectral theory for the pure-state dilation-adapted localization operators themselves: eigenvalue-distribution and plunge-width asymptotics with their dependence on the generator family and set geometry, a corrected density law with quantified error terms, and the construction (or exclusion) of generators with faster temporal decay that keep the spectral problem exactly solvable.

Background

Concentrating a signal simultaneously in time and frequency is a classical problem of harmonic analysis. For the time-band limiting operator $P_T\mathcal{F}^{-1}P_\Omega\mathcal{F}P_T$ (project onto a time interval, then onto a frequency band, then back), Slepian, Pollak and Landau showed that the eigenvalues exhibit a sharp phase transition: roughly $|T||\Omega|/2\pi$ eigenvalues are close to $1$, followed by a narrow transition band — the plunge region — after which the eigenvalues fall rapidly to $0$; the number of essentially concentrated orthogonal modes equals the phase-space volume, the prediction of the Nyquist density, and Landau and Widom later gave sharp two-term asymptotics for the plunge.

Daubechies' geometric phase-space approach recasts such operators using coherent states — families of phase-space-labeled copies of a single template function obtained by a group action. For the Weyl-Heisenberg group (time-frequency shifts) one defines, for a window $g$ and a region $\Omega$ in the time-frequency plane, the localization operator $A_\Omega^g f=\int_\Omega V_g f(z)\,\pi(z)g\,dz$, where $V_g f$ is the short-time Fourier transform of $f$ and $\pi(z)$ the time-frequency shift by $z$; for the Gaussian window and rotationally symmetric $\Omega$ these operators commute with a harmonic-oscillator-type differential operator, which makes their spectra explicit (Hermite eigenfunctions, incomplete-gamma eigenvalues), and Heil, Ramanathan and Topiwala, and De Mari, Feichtinger and Nowak established uniform eigenvalue estimates in the general Weyl-Heisenberg case.

Daubechies and Paul carried out the analogue for the affine group $\mathbb{R}\rtimes\mathbb{R}^+$, whose elements $(b,a)$ act on signals by translation and dilation $T_b D_a$ — the group underlying the wavelet transform. For an admissible wavelet (a function whose dilated translates generate the wavelet coefficients $\langle\psi_{b,a},f\rangle$ with a resolution of the identity), restricting the wavelet coefficients to a set $S$ in the upper half-plane $\{(b,a):a>0\}$ yields a dilation-adapted localization operator; because dilations act on frequencies as scalings, these operators are band-pass, cutting out both very low and very high frequencies. The upper half-plane carries hyperbolic geometry (the geometry of constant negative curvature preserved by translations and dilations), and the paper's localization sets are hyperbolic discs $S_C$ and their two-parameter relatives. The generators chosen there have Fourier transforms supported on one half-line and proportional to $k e^{-k}$, so that in the time domain the generator decays only like $|x|^{-2}$ (a Laplace-transform pair). The key structural fact is that the resulting localization operators commute with a second-order differential operator $T=-k\,\mathrm{d}^2/\mathrm{d}k^2-\mathrm{d}/\mathrm{d}k+k+1/k$ acting on the Fourier transform, a hyperbolic analogue of the harmonic oscillator related to the Maass (hyperbolic Landau-level) Hamiltonian on the half-plane; its square-integrable eigenfunctions are Laguerre functions (orthogonal polynomial eigenfunctions on the half-line), giving fully explicit eigenvalues and multiplicities, with geometric-type eigenvalue decay governed by the ratio $(C-1)/(C+1)$ of the hyperbolic disc. Counting eigenvalues above a threshold then grows linearly in the size parameter $C$ of the set, but with a constant that does not match the naive Nyquist/semiclassical count based on the phase-space area $|S_C|$; the authors attribute the mismatch to the non-uniform localization of the generators and to edge effects, i.e. to the wide plunge region.

The paper also introduces a two-parameter family of generators, with Fourier transforms of the form $k^{\beta}e^{-k^{\gamma}}$ (parameters $\gamma>0$, $\beta>(\gamma-1)/2$) and a corresponding commuting differential operator $T_{\beta,\gamma}$, but leaves the quantitative dependence of plunge width and localization uniformity on $(\beta,\gamma)$ unanalyzed. In what follows, a Folner sequence is a sequence of subsets of a group whose boundary becomes negligible relative to their size under group translations; a group is non-unimodular when its left and right invariant Haar measures differ, which is the case for the affine group. The affine group's non-unimodularity is now known to be the structural reason why naive Euclidean-type counting laws fail in the wavelet setting.

Problem Statement

For time-frequency localization operators adapted to dilations — the affine (wavelet) coherent-state localization operators introduced by Daubechies and Paul — establish a quantitative spectral theory, addressing three coupled facets of a single objective. The three facets are logically linked within this one theory, and facet (2)'s criterion includes its quantified leading correction and error terms. (1) Generator design: construct affine coherent-state generators whose temporal decay is faster than any polynomial decay ($\sim|x|^{-2}$ for the base choice, whose Fourier transform is proportional to $k e^{-k}$ on a half-line) while preserving the analytic tractability of the spectral problem — that is, the existence of a commuting differential operator and closed-form eigenfunctions and eigenvalues — or prove that no such generator exists within the exactly solvable framework. (2) Corrected density law: determine for which generators, localization sets, and growth regimes the number of eigenvalues above a threshold $1-\delta$ is asymptotically given by the phase-space (hyperbolic) volume of the localization set, quantify the leading correction when it is not, and thereby give a general criterion for when the semiclassical/Nyquist density prediction holds or fails for dilation-adapted localization operators, together with the error terms that the known leading-order mixed-state and discrete-density results do not provide. (3) Plunge width and parameter dependence: determine how the width of the plunge region — the range of eigenvalue indices where eigenvalues drop from near $1$ to near $0$ — and the uniformity of localization depend on the generator parameters $(\beta,\gamma)$ and on the geometry of the admissible localization sets, in particular establishing criteria that guarantee narrow plunge regions or that restore the density prediction in an appropriate limit. The common objective is the dilation-side counterpart, for the pure-state affine operators, of the sharp quantitative theory (Landau-Widom-type two-term asymptotics and sharp plunge bounds) that is available for time-band and Weyl-Heisenberg localization operators.

References

1. I. Daubechies and T. Paul, 'Time-frequency localisation operators-a geometric phase space approach: II. The use of dilations', Inverse Problems 4 (1988) 661-680, DOI 10.1088/0266-5611/4/3/009, https://doi.org/10.1088/0266-5611/4/3/009
2. I. Daubechies, 'Time-frequency localization operators: a geometric phase space approach', IEEE Transactions on Information Theory 34 (1988) 605-612, DOI 10.1109/18.9761, https://doi.org/10.1109/18.9761 (companion Weyl-Heisenberg treatment)
3. D. Slepian and H. O. Pollak, 'Prolate Spheroidal Wave Functions, Fourier Analysis and Uncertainty - I', Bell System Technical Journal 40 (1961) 43-63, DOI 10.1002/j.1538-7305.1961.tb03976.x, https://doi.org/10.1002/j.1538-7305.1961.tb03976.x
4. H. J. Landau and H. O. Pollak, 'Prolate Spheroidal Wave Functions, Fourier Analysis and Uncertainty-III: The Dimension of the Space of Essentially Time- and Band-Limited Signals', Bell System Technical Journal 41 (1962) 1295-1336, DOI 10.1002/j.1538-7305.1962.tb03279.x, https://doi.org/10.1002/j.1538-7305.1962.tb03279.x
5. H. J. Landau and H. Widom, 'Eigenvalue distribution of time and frequency limiting', Journal of Mathematical Analysis and Applications 77 (1980) 469-481, DOI 10.1016/0022-247X(80)90241-3, https://doi.org/10.1016/0022-247X(80)90241-3
6. C. Heil, J. Ramanathan and P. Topiwala, 'Phase space localization operators: asymptotics of singular values', Proceedings of the IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis (1994) 194-196, DOI 10.1109/tfsa.1994.467260, https://doi.org/10.1109/tfsa.1994.467260
7. F. De Mari, H. G. Feichtinger and K. Nowak, 'Uniform Eigenvalue Estimates for Time-Frequency Localization Operators', Journal of the London Mathematical Society 65 (2002) 720-732, DOI 10.1112/s0024610702003101, https://doi.org/10.1112/s0024610702003101
8. L. D. Abreu and M. Dörfler, 'An inverse problem for localization operators', Inverse Problems 28 (2012) 115001, DOI 10.1088/0266-5611/28/11/115001, https://doi.org/10.1088/0266-5611/28/11/115001
9. S. Ghobber, 'Boundedness and compactness of localization operators on the affine group', Journal of Pseudo-Differential Operators and Applications 10 (2019) 515-533, DOI 10.1007/s11868-019-00304-3, https://doi.org/10.1007/s11868-019-00304-3
10. E. Berge, S. M. Berge, F. Luef and E. Skrettingland, 'Affine quantum harmonic analysis', Journal of Functional Analysis 282 (2022) 109327, DOI 10.1016/j.jfa.2021.109327, https://doi.org/10.1016/j.jfa.2021.109327 (arXiv:2102.13006, https://arxiv.org/abs/2102.13006)
11. L. D. Abreu and M. Speckbacher, 'Affine density, von Neumann dimension and a problem of Perelomov', Advances in Mathematics 407 (2022) 108564, DOI 10.1016/j.aim.2022.108564, https://doi.org/10.1016/j.aim.2022.108564 (arXiv:2012.11314, https://arxiv.org/abs/2012.11314)
12. S. Halvdansson, 'Quantum harmonic analysis on locally compact groups', Journal of Functional Analysis 285 (2023) 110096, DOI 10.1016/j.jfa.2023.110096, https://doi.org/10.1016/j.jfa.2023.110096 (arXiv:2210.08314, https://arxiv.org/abs/2210.08314)
13. J. P. G. Ramos and P. Tilli, 'A Faber-Krahn inequality for wavelet transforms', Bulletin of the London Mathematical Society 55 (2023) 2018-2034, DOI 10.1112/blms.12833, https://doi.org/10.1112/blms.12833 (arXiv:2205.07998, https://arxiv.org/abs/2205.07998)
14. F. Nicola and P. Tilli, 'The norm of time-frequency and wavelet localization operators', Transactions of the American Mathematical Society 376 (2023) 7353-7375, DOI 10.1090/tran/8990, https://doi.org/10.1090/tran/8990 (arXiv:2207.08624, https://arxiv.org/abs/2207.08624)
15. L. D. Abreu and M. Speckbacher, 'Donoho-Logan large sieve principles for the wavelet transform', Applied and Computational Harmonic Analysis 74 (2025) 101709, DOI 10.1016/j.acha.2024.101709, https://doi.org/10.1016/j.acha.2024.101709 (arXiv:2210.13056, https://arxiv.org/abs/2210.13056)
16. A. Kulikov, 'Exponential lower bound for the eigenvalues of the time-frequency localization operator before the plunge region', Applied and Computational Harmonic Analysis 71 (2024) 101639, DOI 10.1016/j.acha.2024.101639, https://doi.org/10.1016/j.acha.2024.101639 (arXiv:2306.12430, https://arxiv.org/abs/2306.12430)
17. F. Riccardi, 'An optimal estimate for the norm of wavelet localization operators', arXiv:2502.10228 (2025), https://arxiv.org/abs/2502.10228
18. S. Halvdansson, 'Empirical plunge profiles of time-frequency localization operators', Applied and Computational Harmonic Analysis 81 (2026) 101825, DOI 10.1016/j.acha.2025.101825, https://doi.org/10.1016/j.acha.2025.101825 (arXiv:2502.11805, https://arxiv.org/abs/2502.11805)
19. F. Schroth, 'Eigenvalue accumulation for operator convolutions on locally compact groups', arXiv:2603.08141 (2026), https://arxiv.org/abs/2603.08141