Adsorption studies on porous materials require coordinated steps such as structure preparation, force‑field selection, simulation execution, validation, and downstream analysis, making full end‑to‑end automation challenging. This work evaluates RaspAssistant, a hierarchical multi‑agent framework, for carrying out an entire adsorption‑simulation workflow from force‑field preparation through validation to separation analysis.
Across four adsorption and separation benchmarks, twelve multi‑agent workflow reproductions were performed. Scores based on workflow correctness, simulation validity, and analysis consistency ranged from 8.50 to 10.00.
A separate benchmark of eight force‑field extraction cases yielded a mean weighted score of 0.967, reflecting both parameter agreement and readiness for RASPA.
In open‑ended studies, the system first diagnosed and refined an acetylene force field; then, after validating a $CO_2$ force field against experimental data for MFI, it screened 197 pure‑silica zeolites, all of which passed automated input checks.
These results demonstrate that, given a task specification and necessary inputs, RaspAssistant can autonomously coordinate, validate, adapt, and document extended adsorption‑simulation workflows while exposing model sensitivities and methodological limits for expert review.
Review