Abstract:Scattering transforms are a new type of summary statistics recently developed for the study of highly non-Gaussian processes, which have been shown to be very promising for astrophysical studies. In particular, they allow one to build generative models of complex non-linear fields from a limited amount of data, and have also been used as the basis of new statistical component separation algorithms. In the context of upcoming cosmological surveys, such as LiteBIRD for the cosmic microwave background polarization or Rubin-LSST and Euclid for study of the large scale structures of the Universe, the extension of these tools to spherical data is necessary. We develop scattering transforms on the sphere and focus on the construction of maximum-entropy generative models of several astrophysical fields. We construct, from a single target field, generative models of homogeneous astrophysical and cosmological fields, whose samples are quantitatively compared to the target fields using common statistics (power spectrum, pixel probability density function and Minkowski functionals). Our sampled fields agree well with the target fields, both statistically and visually. These generative models therefore open up a wide range of new applications for future astrophysical and cosmological studies; particularly those for which very little simulated data is available. We make our code available to the community so that this work can be easily reproduced and developed further.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to develop scattering transforms applicable to spherical data and construct maximum - entropy generative models based on these transforms to simulate and generate astrophysical fields. Specifically, the research aims to: 1. **Extend scattering transforms to spherical data**: Existing scattering transforms are mainly applied to two - dimensional planar data, while cosmological analysis, especially the next - generation all - sky surveys (such as LiteBIRD, Euclid space telescope and Vera C. Rubin Observatory), needs to handle spherical data. Therefore, these tools must be extended to spherical data. 2. **Construct generative models**: Through scattering transform statistics, construct generative models under the maximum - entropy framework to generate new approximate realizations from limited data. These models can be used to simulate complex nonlinear fields, such as the large - scale structure of the universe, microwave background radiation polarization, etc. 3. **Verify the effectiveness of generative models**: Verify the statistical and visual accuracy of generative models through quantitative comparison between generated samples and target fields (using common statistics such as power spectra, pixel probability density functions and Minkowski functionals). 4. **Provide new tools for future research**: These generative models can provide new applications for future astrophysical and cosmological research, especially when the available simulation data is very limited. ### Key technical points - **Scattering Transforms**: This is a new type of summary statistic for studying highly non - Gaussian processes. It is constructed through continuous wavelet convolutions and point - by - point nonlinear operations (such as modulus operations) and can capture the characteristics of complex nonlinear fields. - **Wavelet Transform on the Sphere**: To adapt to spherical data, the research defines directional spherical convolutions and designs wave functions that satisfy good spatial locality and asymptotic non - correlation. - **Maximum Entropy Generative Models**: By minimizing the loss function, use the gradient descent method to optimize the generative model in the harmonic space so that it is as close as possible to the statistical characteristics of the target field. ### Application prospects These generative models can not only help scientists better understand and simulate complex astrophysical phenomena, but also provide data augmentation for machine - learning applications when training data is insufficient, thus promoting research progress in more fields.

Generative models of astrophysical fields with scattering transforms on the sphere

Generative models of astrophysical fields with scattering transforms on the sphere

Scattering transforms on the sphere, application to large scale structure modelling

Scattering Spectra Models for Physics

Assessing Superspheroids in Modeling the Scattering Matrices of Dust Aerosols

A point cloud approach to generative modeling for galaxy surveys at the field level

Modeling the 3-point correlation function of projected scalar fields on the sphere

Spatially Resolved Galaxy-Dust Modeling with Coupled Data-Driven Priors

Point cloud approach to generative modeling for galaxy surveys at the field level

Physically Based Real‐Time Rendering of Atmospheres using Mie Theory

Sparse Bayesian mass-mapping using trans-dimensional MCMC

Beyond mie theory

Scattering from an external field in quantum chromodynamics at high energies: from foundations to interdisciplinary connections

Generalized Lorenz-Mie theory and simulation software for structured light scattering by particles

Galactic Cosmic-ray Scattering due to Intermittent Structures

Simulation-based inference with scattering representations: scattering is all you need

Seismic scattering regimes from multiscale entropy and frequency correlations

Stochastic modelling of cosmic ray sources for diffuse high-energy gamma-rays and neutrinos

Fast emulation of anisotropies induced in the cosmic microwave background by cosmic strings

A Unified Model of Cosmic Ray Propagation and Radio Extreme Scattering Events from Intermittent Interstellar Structures