Predicting Ti K-edge XANES spectra across three TiO2 polymorphs¶
System: Local (ML inference) · Type: ML spectral prediction · Outcome: Success
The ask¶
(This campaign predates verbatim prompt logging — the request below is reconstructed from the campaign's recorded intent, not a direct quote.)
"Predict Ti K-edge XANES spectra for the three main TiO2 polymorphs (anatase, brookite, rutile) and benchmark them against experimental reference standards."
What happened¶
Trinity pulled all three TiO2 polymorph structures from Materials Project and compared their predicted Ti K-edge XANES spectra — via a machine-learned spectral-prediction model — against experimental reference standards, using five different similarity metrics plus an energy-shift optimization. All three predictions were correctly matched to their corresponding experimental standard. One subtlety: brookite and anatase were nearly tied on one similarity metric, and a derivative-based metric was needed as an unexpected tiebreaker to correctly tell them apart — a genuinely useful methodological finding for future spectral-matching work.
Results¶
- Anatase: best correlation 0.971 with its reference standard.
- Brookite: best correlation 0.944, correctly distinguished from anatase using a derivative-based tiebreaker metric.
- Rutile: best correlation 0.953.
- All three polymorphs correctly ranked and matched to their reference; all metrics exceeded the 0.9 similarity threshold set as the success criterion.