CrystalJev: thinking fast and slow with atomistic foundation models for materials discovery

2026-10-04

AI Summary

This study addresses the computational inefficiency of atomistic foundation models in materials screening by proposing a decision framework inspired by dual-process "fast and slow thinking." Methodologically, the model is restructured from a slow simulator into a rapid decision-maker that predicts stability via a single forward pass. A novel threshold-energy-to-probability mapping is introduced, integrated with value-of-information theory to trigger costly relaxation calculations only when necessary, while frozen interatomic potentials and probability calibration ensure reliability. Experimental results demonstrate that this approach reduces screening costs by an order of magnitude while achieving accuracy comparable to full relaxation. In prospective testing, the prediction error for stability rates remains below 2.1%, highlighting the framework’s practical efficacy for accelerated materials discovery.

Abstract

Atomistic foundation models triage millions of hypothetical materials but are used as slow simulators, their thresholded energies taken at face value. They are better read as fast decision-makers. CrystalJev queries a frozen interatomic potential once per unrelaxed structure and answers typed questions with calibrated probabilities, finite-sample guarantees and a rule for when to think slowly. Across 65 Matbench Discovery models, a 'stable' call is a probability in disguise, explained by a model's errors and the candidate population. Once trained, one forward pass decides nearly as well as a relaxation at a thirtieth of its cost, and a value-of-information theory sends slower computation only where decisions can change. The same layer answers electronic, mechanical and molecular questions. In a registered prospective test with 700 new density-functional calculations, single-pass forecasts calibrated only on existing data over-stated the stable fraction of unseen candidates (5.8%) by at most 2.1 percentage points.

Problem

• atomistic foundation models
• materials discovery
• uncertainty quantification
• computational efficiency
• decision-making

Innovation

• Atomistic foundation models
• CrystalJev
• Probability calibration
• Value-of-information theory
• Materials discovery