Notes on Fourier-Bessel wavelets
Marcel Venturotti, Georgios Exarchakis
cs.LG · cs.CV · math.NA
Quick verdict
Constructs Gaussian-enveloped Fourier-Bessel wavelets from Neumann disk harmonics with sound, self-contained derivations; its uniform-frequency advantage rests on illustrative experiments only.
Summary
These notes construct a wavelet family on the plane from the Neumann eigenfunctions of the Helmholtz equation on the unit disk: Bessel functions J_m(λ_{m,k}ρ) with angular factors e^{imφ}, enveloped by a fixed-width Gaussian, where the eigenvalues (roots of J'_m, found by Muller's method) are approximately linearly spaced with spacing tending to π. The authors derive the zero-mean correction K_{0,k}=e^{-λ²σ²/2} for the m=0 order, explicit L2 normalisation constants via Weber's exponential integrals, a frequency-domain peak normalisation, and a closed-form Fourier representation using the Jacobi-Anger expansion. The only empirical evidence is illustrative: in one compared configuration the frame ripple ratio B/A is 2.2530 for the proposed bank (m=3, k=8, σ=1) versus 6.2377 for Solid Harmonics (J=3, L=5), and a parameter grid search shows similarly lower values. The experiments carry no task validation; frame proofs, the choice of σ, and downstream utility remain open. All derivations are explicit and a companion library (fbscatnet) is released, making the numerical results readily reproducible.
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 1
● low 3
4 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- The Neumann eigenvalues λ_{m,k} (positive roots of J'_m) on the unit disk are asymptotically linearly spaced, with spacing converging to π.
- Gaussian-enveloped disk harmonics with the correction K_{0,k}=e^{-λ²σ²/2} for m=0 form a zero-mean wavelet family with explicit closed-form L2 normalisation constants (Eqs. 87, 92).
- The family admits a closed-form 2D Fourier representation (Theorem 5.1) that separates into an angular factor (−i)^m e^{imφ} and a radial modified-Bessel response.
- In the tested configurations, Fourier-Bessel banks achieve flatter frequency coverage (lower ripple ratio B/A) than Solid Harmonic banks (Fig. 9).
Strengths
Self-contained, verifiable derivations
The Bessel derivative identity (Lemma 1), the zero-mean correction K_{0,k}=e^{-λ²σ²/2} (Prop. 4.1 via Weber's first exponential integral), the L2 constants (Props. 4.2-4.3), and the closed-form Fourier representation (Thm. 5.1 via Jacobi-Anger) are each derived step by step from the series definitions, and internal cross-checks hold (e.g., Eq. 98-100 recombine exactly into the bracket of Eq. 92).
Numerically trustworthy eigenvalue tables
The Table 2 Neumann eigenvalues (e.g., λ_{0,1}=3.8317, λ_{1,1}=1.8412, λ_{2,1}=3.0542) match the known roots of J'_m to four decimals, and the McMahon expansion (Eq. 72-73) is correctly used only as an initializer and spacing interpretation, with the caveat (Remark 3.3) that actual construction uses the computed roots.
Unusually honest scoping
Reproducibility by construction
Weaknesses
Empirical evidence is a single illustrative configuration
Frame properties unproven and possibly unattainable as stated
π-spacing is asymptotic only, and only radial
Thin novelty positioning and prior-art coverage
No analysis of the Gaussian width σ
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a set of theoretical notes, explicitly scoped as such (Sec. 1, 'Scope of these notes'): a self-contained derivation of a Gaussian-windowed Fourier-Bessel wavelet family built on the disk harmonics of Shaqfa et al. [9]. The claimed contributions are (1) the wavelet construction with an m=0 zero-mean correction, (2) closed-form L1/L2 and peak normalisation constants, (3) a closed-form Fourier-domain representation, and (4) preliminary numerical evidence that linear radial eigenvalue spacing yields more uniform frequency coverage than Solid Harmonics. The authors are commendably careful: they repeatedly state that the construction does not replace dyadic scaling and that the scattering-theoretic properties of dyadic wavelets (Lipschitz continuity, deformation stability) 'have not been established' for the new family.
The central mathematical premise is sound and independently checkable: the Neumann eigenvalues of the unit disk are the positive roots of J'_m, and McMahon's expansion (Eq. 72-73) makes the asymptotic π spacing immediate. I spot-checked the numbers: Table 2's λ values (3.8317, 7.0156, 10.1735, ...; 1.8412, 5.3314, ...) match the known roots of J'_m to four decimals, and the +1.35%→+0.12% (m=0) versus +11.10%→+0.33% (m=1) convergence pattern correctly reflects the −(4m²+3)/(8β) correction term.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 4 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What are Fourier-Bessel wavelets?
A family of 2D wavelets built from the eigenfunctions of the Helmholtz equation on the unit disk with Neumann boundary conditions: Bessel radial oscillations J_m(λ_{m,k}ρ) with angular factors e^{imφ}, multiplied by a Gaussian envelope. Each wavelet is indexed by angular order m and the root index k, and their Fourier transforms have a ring-like shape whose radial peak sits near the eigenvalue λ_{m,k}.
Why do the authors use non-dyadic, linearly spaced frequencies instead of standard wavelet scales?
The Neumann eigenvalues of the disk become approximately equally spaced (spacing tending to π) as the root index grows, so indexing filters by eigenvalue gives nearly uniform coverage of radial frequencies. This is offered as a complement — not a replacement — to dyadic scaling, motivated by reconstruction-style tasks where balanced frequency coverage may be preferable to the low-frequency emphasis of dyadic wavelets.
How do Fourier-Bessel wavelets compare to Solid Harmonic wavelets?
In the paper's single illustrative experiment, the Fourier-Bessel bank (m=3, k=8, σ=1) reached a frequency-coverage ripple ratio B/A of 2.2530 versus 6.2377 for a Solid Harmonic bank (J=3, L=5), with a parameter grid search showing the same ordering. This is not a frame proof or a task benchmark; both constructions remain valid and no application performance is measured.
Is the code for these wavelets available?
Yes. The paper states the construction is implemented in the Python library fbscatnet on GitHub (github.com/Smee18/FourierBesselWavelets), which is intended to reproduce the derivations and figures and to use the wavelets inside scattering networks, similarly to Kymatio.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- The provided text appears complete including references, but figures were recovered as embedded text only; the plotted coverage maps in Fig. 9a-b and 3D wavelet shapes could not be visually inspected.
- Fig. 5 and Fig. 7 axes are labelled k=0..4 while the text defines k starting at 1 — likely a display convention that could not be verified from the text.
- The frequency-coverage mask, bank sizes, and exact ratio computations in Fig. 9 are described only in captions, so the comparison's fairness was assessed from the stated configuration rather than the raw data.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.