Velocityformer: Broken-Symmetry-Matched Equivariant Graph Transformers for Galaxy Velocity Reconstruction

Forum topic · 小凯 · 2026-05-22

Summary

Velocityformer (arXiv:2505.15983) is an equivariant graph transformer architecture introduced by Tilman Troester, David Mirkovic, and Veronika Oehl (2025) for reconstructing galaxy velocities from spectroscopic surveys, a key requirement for measuring the kinematic Sunyaev-Zeldovich (kSZ) effect. The signal-to-noise ratio of kSZ measurements scales directly with the correlation coefficient r between reconstructed and true velocities, making reconstruction accuracy critical. The central design insight is to align the model's inductive bias with the broken symmetry of the observational data. The authors show this symmetry matching consistently improves performance across all model sizes and training volumes. Velocityformer improves r by 35% over the standard linear theory baseline and outperforms machine learning baselines at every data volume, demonstrating that symmetry-aware architectures can substantially boost kSZ cosmology measurement quality.

Overview

This post introduces the paper Velocityformer: Broken-Symmetry-Matched Equivariant Graph Transformers (arXiv, published 2025-05-20) by Tilman Troester, David Mirkovic, and Veronika Oehl.

arXiv link: 2505.15983
Field: Machine Learning applied to cosmology

Abstract (translated)

Accurate 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. The authors 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 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: Precise kSZ measurements depend on accurate galaxy velocity reconstruction from spectroscopic surveys; SNR scales directly with the correlation coefficient r.
Method: Velocityformer, an equivariant graph transformer whose inductive biases are matched to the broken symmetry of the observational data.
Result: Consistent improvement across model sizes and training volumes; 35% higher r versus the standard linear theory baseline; beats ML baselines at every data volume.

Source: arXiv:2505.15983