In the field of agentic scientific machine learning (SciML), large language model (LLM) agents can discover surrogate models and select one using an automated scoring system, typically based on an error metric. However, a low error does not ensure that the predicted fields meet critical physics criteria for mechanics, such as boundary conditions, superposition, stiffness scaling, or causality.
To address this, we introduce Physics-Audited Agentic SciML (PA-SciML), a verification-first workflow. This workflow fixes a scoring evaluator before the search, derives reviewable machine-checkable physics requirements, checks the outputs of each trained candidate, and separately searches prescribed input ranges or measured load-history spans for high-violation cases without reference solution fields.
A surrogate is reported as verified only under the stated checks. When enabled, the workflow also incorporates advisory numerical probes before training and tests one modeling change at a time to record which isolated edits are associated with score improvements.
In the reported computational solid mechanics numerical examples, the static elasticity run selects a surrogate with lower validation error than the error-only baseline, while both selected models pass common linear-elastic checks.
In the transient elastodynamics run, an error-only baseline with similar mean error fails a stricter causality check by responding to future parts of the loading history, whereas the selected surrogate passes the stated checks.
The main distinction lies in per-candidate physics evidence on predicted fields, not a richer aggregate score.
Blogger's Review: PA-SciML significantly enhances the reliability of agentic machine learning models by integrating a physics auditing mechanism. This approach emphasizes not only the predictive accuracy of models but also their adherence to fundamental physical principles, which holds substantial practical significance and research value.