Characterizing LLM Code Generation for Computational Simulations
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 576-585
DOI: 10.1609/aaaiss.v10i1.43458 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Shital Adhikari und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Computational simulations constitute a fundamental component of engineering practice and have widespread industrial applications. However, a key bottleneck lies in the transition from theoretical formulations to implementing computational simulations, which remains a time-consuming process for researchers, demanding extensive domain knowledge about numerical methods, computational physics as well as computer programming. Large language models (LLMs), with their approximate world knowledge and proven code-generation ability, present a great opportunity to automate computational simulation design. Although current LLMs are capable of generating general-purpose code, their ability to generate reliable computational simulation code is yet untested. In this work, we study the performance of current state-of-the-art LLMs in generating simulation source code for PDE-governed systems under a Generator-Corrector agent-based framework in which the LLM is guided by residual feedback.
Zitieren
APA
Adhikari, S., Kapoor, T., & Muralidhar, N. (2026). Characterizing LLM Code Generation for Computational Simulations. Proceedings of the AAAI Symposium Series, 10(1), 576-585. https://doi.org/10.1609/aaaiss.v10i1.43458
Vancouver
Adhikari S, Kapoor T, Muralidhar N. Characterizing LLM Code Generation for Computational Simulations. Proceedings of the AAAI Symposium Series. 2026;10(1):576-585. doi:10.1609/aaaiss.v10i1.43458
BibTeX
@article{Adhikari2026,
author = {Adhikari, Shital and Kapoor, Taniya and Muralidhar, Nikhil},
title = {Characterizing LLM Code Generation for Computational Simulations},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {576--585},
doi = {10.1609/aaaiss.v10i1.43458}
}
EUVIMED