Small Angle Neutron Scattering in McStas: optimization for high throughput virtual experiments

TL;DR

This paper introduces optimized small angle neutron scattering components in McStas, enabling high-throughput virtual experiments with improved performance and the ability to generate large datasets for machine learning applications.

Contribution

Development of neutron scattering components in McStas incorporating polydispersity and orientation effects, optimized for parallel computing and high-throughput virtual experiments.

Findings

Benchmarking shows significant acceleration with GPU and multi-core implementations.

Models enable rapid generation of large virtual datasets for machine learning.

Performance improvements facilitate high-throughput neutron scattering simulations.

Abstract

In this work we present the development of small angle scattering components in McStas that describe the neutron interaction with 70 different form and structure factors. We describe the considerations taken into account for the generation of these components, such as the incorporation of polydispersity and orientational distribution effects in the Monte Carlo simulation. These models can be parallelized by means of multi-core simulations and graphical processing units (GPUs). The acceleration schemes for the aforementioned models are benchmarked, and the resulting performance is presented. This allows for the estimation of computation times in high-throughput virtual experiments. The presented work enables the generation of large datasets of virtual experiments that can be explored and used by machine learning algorithms.