Event Generation with Parallel Langevin Sampling and Learned Stein Diagnostics

2026-06-16 · Source: stat.ML updates on arXiv.org · Field: Science & Research — Physical Sciences & Chemistry, Mathematics & Computational Sciences, Artificial Intelligence & Machine Learning · Depth: Expert, long

Summary

A novel event generation method addresses the computational bottleneck in precision collider phenomenology, particularly for high-multiplicity final states at facilities like the high-luminosity Large Hadron Collider. This approach utilizes many parallel underdamped Langevin chains, extracting a single terminal state from each to produce unweighted events while circumventing within-chain autocorrelation. A learned Stein discrepancy (LSD) serves as a data-driven convergence diagnostic, estimating the relaxation time. Applied to tree-level u\bar{u}\to Z+ng event generation, the method demonstrates that relaxation needs only a modest number of exact-target Langevin steps, with growth mildly increasing with multiplicity. Furthermore, neural-network surrogate initialization significantly reduces the required exact matrix-element and gradient evaluations. The technique was validated using an Ornstein–Uhlenbeck process in d=4 and d=16 dimensions, showing LSD accurately tracks analytic convergence and the importance of a boundary function for sharp phase-space cuts.

Key takeaway

For Research Scientists and Machine Learning Engineers facing computational bottlenecks in high-multiplicity event generation, you should consider adopting parallel Underdamped Langevin Algorithm (ULA) sampling. This method, combined with learned Stein discrepancy diagnostics, offers a path to efficiently generate unweighted events without the limitations of rejection sampling. Integrating neural network surrogates for initial chain equilibration can further reduce expensive matrix-element and gradient evaluations, significantly accelerating your simulation workflows.

Key insights

Parallel Langevin sampling with learned Stein diagnostics efficiently generates unweighted collider events, overcoming rejection sampling limitations.

Principles

• Parallel MCMC avoids within-chain autocorrelation.
• Learned Stein discrepancy quantifies MCMC convergence.
• Surrogates accelerate equilibration for expensive targets.

Method

Evolve parallel Underdamped Langevin Algorithm (ULA) chains with Metropolis-Hastings corrections and momentum updates. Use a learned Stein discrepancy to determine relaxation time and neural network surrogates for initial acceleration.

In practice

• Implement ULA for unweighted event generation.
• Use LSD to monitor MCMC chain convergence.
• Pre-train surrogates to reduce initial gradient evaluations.

Topics

• Event Generation
• Langevin Sampling
• Stein Discrepancy
• Monte Carlo Methods
• Collider Phenomenology
• Neural Network Surrogates

Code references

• rbvh/ula_stein_evt_gen

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Stein’s method, learning and inference -or- how to really monitor convergence and thin chains — Stein's method, leveraging scores and operators, offers a robust approach for convergence monitoring and advanced probabilistic inference.
• More Than Image Generators: A Science of Problem-Solving using Probability | Diffusion Models — Diffusion models combine deep learning with Langevin sampling to generate images by iteratively following noisy gradients of an image's probability distribution.
• Dimension-Uniform Discretization Analysis of Preconditioned Annealed Langevin Dynamics for Multimodal Gaussian Mixtures — Dimension-uniform control in annealed Langevin dynamics requires specific discretization and preconditioning for stability.
• Automating the Expert Eye: A System-Agnostic Deep Learning Framework for Rare Event Discovery in Imbalanced Force Spectroscopy — Deep learning automates rare event discovery in SMFS, overcoming extreme data imbalance with high accuracy and interpretability.
• Theoretical guidelines for annealed Langevin dynamics in compositional simulation-based inference — Theoretical guidelines for annealed Langevin dynamics hyperparameters ensure sampling accuracy in compositional SBI, favoring Linhart's method.
• LESS Is More: Mutual-Stability Sampling for Diffusion Language Models — Adaptive sampling based on token stability significantly enhances diffusion LLM inference efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.