Compiling Chemical Knowledge into Executable Descriptors for Materials Prediction
Abstract
Materials prediction depends critically on how scientific knowledge is represented, yet many governing considerations exist only as natural-language heuristics that conventional learners cannot use. We introduce CRISP, a large language model-assisted framework that treats representation construction as a rule-space exploration and compilation problem: it repeatedly samples target-relevant chemical rules without access to structures, labels or data splits, consolidates related concepts, and compiles each into an executable scalar descriptor supplied to a conventional learner. For positive-unlabeled inorganic-crystal synthesizability, CRISP outperformed expert-curated and generic structural representations under a shared learner and surpassed purpose-built synthesizability models, with its advantage most pronounced under structural-size and chemical-family shifts. Infrequently generated rules contributed complementary predictive information, showing that generation frequency does not determine utility. The same workflow yielded competitive representations for formation energy and ionic conductivity while revealing task-dependent limits for shear modulus, establishing a dataset-blind, auditable route from broad chemical knowledge to transferable computational representations.
Related
- From Autonomous Knowledge Generation to Generalizable Property Prediction: A Precision-First Pipeline for Electro-Optic Materials Discovery
- Autonomous Analysis of Thermoelectric Materials Big Data by Multi-Agentic AI and Construction of a Knowledge Base and Specialized AI
- Agentic schema-guided extraction of materials process knowledge from scientific literature
- Knowledge-Guided Construction of Machine Learning Potentials for Reactive Behavior in Energetic Materials
- Executable verification through formalized expert reasoning in astronomical spectroscopy
BibTeX
@misc{choi2026compiling,
title = {{Compiling Chemical Knowledge into Executable Descriptors for Materials Prediction}},
author = {Jaehwan Choi and Kunik Jang and Seongmin Kim and Shuan Chen and Kyungju Nam and Seung Hyo Noh and Donghwi Kim and Yousung Jung},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.27587},
eprint = {2608.27587},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.27587},
}
RIS
TY - UNPB
TI - Compiling Chemical Knowledge into Executable Descriptors for Materials Prediction
AU - Jaehwan Choi
AU - Kunik Jang
AU - Seongmin Kim
AU - Shuan Chen
AU - Kyungju Nam
AU - Seung Hyo Noh
AU - Donghwi Kim
AU - Yousung Jung
PY - 2026
DA - 2026/08/27
PB - arXiv
DO - 10.48550/arxiv.2608.27587
UR - https://arxiv.org/abs/2608.27587
AB - Materials prediction depends critically on how scientific knowledge is represented, yet many governing considerations exist only as natural-language heuristics that conventional learners cannot use. We introduce CRISP, a large language model-assisted framework that treats representation construction as a rule-space exploration and compilation problem: it repeatedly samples target-relevant chemical rules without access to structures, labels or data splits, consolidates related concepts, and compiles each into an executable scalar descriptor supplied to a conventional learner. For positive-unlabeled inorganic-crystal synthesizability, CRISP outperformed expert-curated and generic structural representations under a shared learner and surpassed purpose-built synthesizability models, with its advantage most pronounced under structural-size and chemical-family shifts. Infrequently generated rules contributed complementary predictive information, showing that generation frequency does not determine utility. The same workflow yielded competitive representations for formation energy and ionic conductivity while revealing task-dependent limits for shear modulus, establishing a dataset-blind, auditable route from broad chemical knowledge to transferable computational representations.
ER -