Velocityformer: An Equivariant Graph Transformer for Galaxy Velocity Reconstruction in kSZ Effect Studies

Forum topic · 小凯 · 2026-05-22

Summary

Velocityformer is an equivariant graph transformer architecture designed to reconstruct galaxy velocities from spectroscopic surveys, enabling precise measurement of the kinematic Sunyaev-Zeldovich (kSZ) effect. Because kSZ signal-to-noise ratio scales directly with the correlation coefficient r between reconstructed and true velocities, improving velocity reconstruction is critical for cosmology. The key innovation is matching the model's inductive bias to the broken symmetry of the observational data. This symmetry-matching approach delivers consistent gains across all model sizes and training volumes, improving the correlation coefficient r by 35% over the standard linear theory baseline and outperforming existing ML baselines at every data volume tested. The work demonstrates that encoding known physical symmetries into graph transformer architectures yields substantial improvements for cosmological inference tasks. The paper is authored by Tilman Troester, David Mirkovic, and Veronika Oehl, posted to arXiv on May 20, 2025.

Overview

Research area: Machine Learning for Cosmology Authors: Tilman Troester, David Mirkovic, Veronika Oehl Published: 2025-05-20 arXiv: 2505.15983

Abstract

Precise measurement of the kinematic Sunyaev-Zeldovich (kSZ) effect requires accurate reconstruction of galaxy velocities from spectroscopic surveys. The signal-to-noise ratio (SNR) of kSZ measurements scales directly with the correlation coefficient r between reconstructed and true velocities.

We introduce Velocityformer, an equivariant graph transformer architecture designed to match the specific symmetry of the observational data. Matching the model inductive bias to the data's broken symmetry consistently improves performance across all model sizes and training volumes, with Velocityformer improving r by 35% over the standard linear theory baseline and outperforming ML baselines at every data volume.

Key Points

Problem: Accurate galaxy velocity reconstruction is essential for precise kSZ effect measurements, since kSZ SNR scales linearly with the correlation coefficient r between reconstructed and true velocities.
Method: Velocityformer is an equivariant graph transformer whose architectural inductive bias is explicitly matched to the broken symmetry structure present in real observational data.
Result: The symmetry-matched architecture improves r by 35% over the standard linear theory baseline.
Robustness: Gains hold consistently across all model sizes and training-data volumes examined.
Versus ML baselines: Velocityformer outperforms competing machine learning approaches at every data volume tested.

Significance

The work demonstrates that explicitly encoding known physical symmetries into graph transformer architectures provides a meaningful inductive bias for cosmological velocity-field reconstruction, yielding gains that persist across model and data scales.