XDFT: AI Agent Diagnoses DFT-Experiment Band-Gap Mismatches
A new AI agent called XDFT automatically identifies why density functional theory (DFT) calculations misclassify materials as metallic when experiments show them as semiconductors. The system draws hypotheses from a curated catalogue, runs first-principles tests, and updates a Bayesian posterior. On a benchmark of 124 materials, XDFT resolved 70 of 90 mismatch cases (78%), far outperforming a random baseline (19%) and a static LLM (2%).
Key facts
- Standard DFT misclassifies electronic ground states of correlated and structurally complex compounds.
- Each mismatch encodes a specific non-ideality: magnetic ordering, electron correlation, alternative polymorph, or defect.
- XDFT is a closed-loop agent that diagnoses the mismatch automatically.
- It draws candidate hypotheses from a curated catalogue and executes first-principles tests.
- It updates a global Bayesian posterior over hypothesis usefulness from each verdict.
- Verified benchmark of 124 materials.
- XDFT identifies a resolving mechanism for 70 of 90 mismatch cases (78%).
- Uniform-random baseline achieved 19%, static LLM ordering achieved 2%.
Entities
—