Rust-accelerated powder X-ray diffraction simulation for high-throughput and machine-learning-driven materials science
Journal of Applied Crystallography · 01.08.2026 · 59 (4) · 1344-1350
DOI: 10.1107/s1600576726005273 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Applied Crystallography, online erschienen am 22.07.2026. Verfasst von Miroslav Lebeda und 3 weiteren Autor:innen. Beteiligte Einrichtung: Faculty of Mechanical Engineering Czech Technical University in Prague Technická 4 16607Prague 6 Czechia. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: České Vysoké Učení Technické v Praze.
Abstract (Originalsprache)
High-throughput powder X-ray diffraction (XRD) simulations are a key prerequisite for generating large datasets used in the development of machine-learning models for XRD-based materials analysis. However, the widely used pymatgen powder XRD calculator, implemented entirely in Python, can be computationally inefficient for large-scale workloads, limiting throughput. We present XRD-Rust , a Rust-accelerated implementation of the pymatgen powder XRD calculator that maintains compatibility with the existing Python-based workflows. The method retains pymatgen for crystal structure handling, while reimplementing the computationally intensive parts of the XRD calculation in the Rust language, with optional further acceleration via single instruction, multiple data (SIMD) vectorization and multi-threaded execution across reflections using the Rayon library. Performance benchmarking on two large crystallographic datasets, the Materials Cloud Three-Dimensional Structure Database (MC3D, 33142 structures) and the Crystallography Open Database (COD, 515181 structures), demonstrates substantial speedups. For the MC3D, XRD-Rust achieves a median serial SIMD speedup of 10.8× [median absolute deviation (MAD) of 1.8×], increasing to 15.1× (MAD 3.8×) with eight-thread parallel execution, with a maximum runtime reduction from 40.5 s to 0.9 s. For the COD, the median serial SIMD acceleration reaches 10.7× (MAD 4.2×), while eight-thread parallel execution yields 19.5× (MAD 10.0×), with a maximum reduction from 1437 to 1 min. SIMD vectorization alone provides additional performance gains ranging from a few percent to several tens of percent, depending on the workload and execution mode. Overall, these benchmarks demonstrate that XRD-Rust significantly accelerates powder XRD simulations compared with the original pymatgen implementation, enabling efficient high-throughput dataset generation and improving performance in interactive diffraction analysis applications.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Faculty of Mechanical Engineering Czech Technical University in Prague Technická 4 16607Prague 6 Czechia.
- Faculty of Nuclear Sciences and Physical Engineering Czech Technical University in Prague Trojanova 339/13 12000Prague 2 Czechia.
- Institute of Physics of the Czech Academy of Sciences Na Slovance 2 18200Prague 8 Czechia.
ORCID-Profile: Miroslav Lebeda, Jan Drahokoupil, Petr Veřtát, Petr Vlčák
Förderung
- České Vysoké Učení Technické v Praze
Zitieren
APA
Lebeda, M., Drahokoupil, J., Veřtát, P., & Vlčák, P. (2026). Rust-accelerated powder X-ray diffraction simulation for high-throughput and machine-learning-driven materials science. Journal of Applied Crystallography, 59(4), 1344-1350. https://doi.org/10.1107/s1600576726005273
Vancouver
Lebeda M, Drahokoupil J, Veřtát P, Vlčák P. Rust-accelerated powder X-ray diffraction simulation for high-throughput and machine-learning-driven materials science. Journal of Applied Crystallography. 2026;59(4):1344-1350. doi:10.1107/s1600576726005273
BibTeX
@article{Lebeda2026,
author = {Lebeda, Miroslav and Drahokoupil, Jan and Veřtát, Petr and Vlčák, Petr},
title = {Rust-accelerated powder X-ray diffraction simulation for high-throughput and machine-learning-driven materials science},
journal = {Journal of Applied Crystallography},
year = {2026},
volume = {59},
number = {4},
pages = {1344--1350},
doi = {10.1107/s1600576726005273}
}
EUVIMED