Eur. Phys. J. Spec. Top.
https://doi.org/10.1140/epjs/s11734-025-02015-x

Regular Article

Reconstructing sparticle masses at the LHC using generative machine learning

¹ Kavli IPMU (WPI), UTIAS, The University of Tokyo, 277-8583, Kashiwa, Chiba, Japan
² Department of Physics, Indian Institute of Technology Patna, 801106, Patna, Bihar, India

^a This email address is being protected from spambots. You need JavaScript enabled to view it.

Received: 4 July 2025
Accepted: 5 October 2025
Published online: 17 October 2025

Abstract

We explore a generative-model framework to infer the masses of heavy particles from detector-level data over a broad parameter space. Our model combines a transformer-based detector encoder and a diffusion neural network. We first apply our model to a new physics scenario involving the pair production of wino-like chargino–neutralino, $pp\to {\tilde{\chi }}_1^{\pm }{\tilde{\chi }}_2^0$, in the $1\ell +2\gamma +jets$ channel at the high-luminosity LHC (HL-LHC). We find that our framework can achieve mass reconstruction efficiency of $\gtrsim 70\%$ for the lightest neutralino ${\tilde{\chi }}_1^0$ and $\gtrsim 40\%$ for the second-lightest neutralino ${\tilde{\chi }}_2^0$, for a mass tolerance of $\mathrm{\Delta }m=30$GeV, across the entire parameter space accessible at the HL-LHC. We further extend our analysis to a different scenario with $pp\to {\tilde{\chi }}_1^{\pm }{\tilde{\chi }}_1^{\mp }+{\tilde{\chi }}_1^{\pm }{\tilde{\chi }}_2^0$ pair production at the HL-LHC in the 4ℓ+Ɇ_T channel, and for a fixed value of $m_{{\tilde{\chi }}_2^0}$, we obtain reconstruction efficiencies $\gtrsim 80\%$ over a wide range of $m_{{\tilde{\chi }}_1^0}$ for $\mathrm{\Delta }m=30$GeV.