Reinterpreting Simulation Results for New Materials Applications with a Large Language Model
Year: 2026
Type: preprint
Abstract: In materials discovery, computational results generated for one application may retain value for deciding what to investigate next. Here we use a large language model to reinterpret archived adsorption calculations in the context of framework chemistry and pore structure across three metal–organic framework case studies. Xenon affinity directly guided Xe/Kr selection, while dilute-limit methane affinity transferred more effectively than finite-pressure uptake. We then asked whether archived results could suggest new applications without a predefined target. CO2 results prompted four separation directions, CO2/N2, CO2/CH4, C2H2/CO2, and C2H6/CO2, none of which was recovered when the same materials lacked the archived CO2 information. Nine of 15 CO2-informed material–application pairs ranked in the top quarter of corresponding classical screening distributions. Archived simulations can therefore guide what to investigate next, from selecting materials for a known target to opening a new application.
Source: ChemRxiv
Author: Jihan Kim
Institution: Korea Advanced Institute of Science and Technology
Language: English
Subfield: Inorganic Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
Awards: G1364446810, G1611679478, G1730441869
OpenAlex ID: W7214668940