Saved in:
Bibliographic Details
Main Authors: Hyun, Lee, Yoon, Sohee, Park, Jinwoo, Chae, Sue In, Park, Seongeon, Ahn, Jooyeon, Jung, Yebin, Chung, Youjung, Chang, Hogeun, Park, Sujin, Kang, Myeonginn, Kim, Jina, Kim, Ho-Gyeong, Jeong, Myeonghun
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.00768
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916984413749248
author Hyun, Lee
Yoon, Sohee
Park, Jinwoo
Chae, Sue In
Park, Seongeon
Ahn, Jooyeon
Jung, Yebin
Chung, Youjung
Chang, Hogeun
Park, Sujin
Kang, Myeonginn
Kim, Jina
Kim, Ho-Gyeong
Jeong, Myeonghun
author_facet Hyun, Lee
Yoon, Sohee
Park, Jinwoo
Chae, Sue In
Park, Seongeon
Ahn, Jooyeon
Jung, Yebin
Chung, Youjung
Chang, Hogeun
Park, Sujin
Kang, Myeonginn
Kim, Jina
Kim, Ho-Gyeong
Jeong, Myeonghun
contents AI-driven materials discovery that couples automated experimentation with algorithmic decision-making requires process aware recipe to property predictors that are accurate, calibrated, and physically admissible. We approach this as a reasoning problem with large reasoning models (LRMs). To instill reasoning capability into language models, we curate reasoning traces from a teacher model to train a student model. However, most training pipelines select reasoning traces using binary correctness or learned preference signals that poorly reflect physical admissibility. We introduce Physics-aware Rejection Sampling (PaRS), a training-time trace selection scheme that favors traces consistent with fundamental physics and numerically close to targets, with lightweight halting to control compute. We instantiate our framework with a large student model fine-tuned on traces synthesized by a larger teacher model, and evaluate under matched token budgets against various rejection sampling baselines. Our method improves accuracy and calibration, reduces physics-violation rates, and lowers sampling cost relative to baselines. These results indicate that modest, domain-aware constraints combined with trace-level selection provide a practical path toward reliable, efficient LRMs for process-aware property prediction and closed-loop materials design.
format Preprint
id arxiv_https___arxiv_org_abs_2509_00768
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Aligning Reasoning LLMs for Materials Discovery with Physics-aware Rejection Sampling
Hyun, Lee
Yoon, Sohee
Park, Jinwoo
Chae, Sue In
Park, Seongeon
Ahn, Jooyeon
Jung, Yebin
Chung, Youjung
Chang, Hogeun
Park, Sujin
Kang, Myeonginn
Kim, Jina
Kim, Ho-Gyeong
Jeong, Myeonghun
Artificial Intelligence
Materials Science
Computation and Language
AI-driven materials discovery that couples automated experimentation with algorithmic decision-making requires process aware recipe to property predictors that are accurate, calibrated, and physically admissible. We approach this as a reasoning problem with large reasoning models (LRMs). To instill reasoning capability into language models, we curate reasoning traces from a teacher model to train a student model. However, most training pipelines select reasoning traces using binary correctness or learned preference signals that poorly reflect physical admissibility. We introduce Physics-aware Rejection Sampling (PaRS), a training-time trace selection scheme that favors traces consistent with fundamental physics and numerically close to targets, with lightweight halting to control compute. We instantiate our framework with a large student model fine-tuned on traces synthesized by a larger teacher model, and evaluate under matched token budgets against various rejection sampling baselines. Our method improves accuracy and calibration, reduces physics-violation rates, and lowers sampling cost relative to baselines. These results indicate that modest, domain-aware constraints combined with trace-level selection provide a practical path toward reliable, efficient LRMs for process-aware property prediction and closed-loop materials design.
title Aligning Reasoning LLMs for Materials Discovery with Physics-aware Rejection Sampling
topic Artificial Intelligence
Materials Science
Computation and Language
url https://arxiv.org/abs/2509.00768