Saved in:
Bibliographic Details
Main Authors: Brutti, Pierpaolo, Durastanti, Claudio, Mari, Francesco
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.20498
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915758831828992
author Brutti, Pierpaolo
Durastanti, Claudio
Mari, Francesco
author_facet Brutti, Pierpaolo
Durastanti, Claudio
Mari, Francesco
contents Diffusion models provide a principled framework for generative modeling via stochastic differential equations and time-reversed dynamics. Extending spectral diffusion approaches to spherical data, however, raises nontrivial geometric and stochastic issues that are absent in the Euclidean setting. In this work, we develop a diffusion modeling framework defined directly on finite-dimensional spherical harmonic representations of real-valued functions on the sphere. We show that the spherical discrete Fourier transform maps spatial Brownian motion to a constrained Gaussian process in the frequency domain with deterministic, generally non-isotropic covariance. This induces modified forward and reverse-time stochastic differential equations in the spectral domain. As a consequence, spatial and spectral score matching objectives are no longer equivalent, even in the band-limited setting, and the frequency-domain formulation introduces a geometry-dependent inductive bias. We derive the corresponding diffusion equations and characterize the induced noise covariance.
format Preprint
id arxiv_https___arxiv_org_abs_2601_20498
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Spectral Diffusion Models on the Sphere
Brutti, Pierpaolo
Durastanti, Claudio
Mari, Francesco
Probability
Machine Learning
60H10, 42C10, 62H11
Diffusion models provide a principled framework for generative modeling via stochastic differential equations and time-reversed dynamics. Extending spectral diffusion approaches to spherical data, however, raises nontrivial geometric and stochastic issues that are absent in the Euclidean setting. In this work, we develop a diffusion modeling framework defined directly on finite-dimensional spherical harmonic representations of real-valued functions on the sphere. We show that the spherical discrete Fourier transform maps spatial Brownian motion to a constrained Gaussian process in the frequency domain with deterministic, generally non-isotropic covariance. This induces modified forward and reverse-time stochastic differential equations in the spectral domain. As a consequence, spatial and spectral score matching objectives are no longer equivalent, even in the band-limited setting, and the frequency-domain formulation introduces a geometry-dependent inductive bias. We derive the corresponding diffusion equations and characterize the induced noise covariance.
title Spectral Diffusion Models on the Sphere
topic Probability
Machine Learning
60H10, 42C10, 62H11
url https://arxiv.org/abs/2601.20498