PXRD-Bench: How Close Are We to Determining Crystal Structures from Powder Diffraction?
To fill this gap, we introduce PXRD-Bench, a unified benchmark that extends evaluation to unknown per-element atom counts and recovery of both structures in two-phase mixtures.
ProofPaper ↗
Key points
- Determining crystal structures from powder X-ray diffraction (PXRD) would enable routine characterization of newly synthesized materials, including those absent from reference databases.
- To provide a strong reference baseline, we also introduce Bragglite, a PXRD-to-structure model developed through a systematic study of PXRD modeling choices.
- Further analysis yields several key findings: some models fail to benefit from experimental PXRD, while generalization to unseen structural prototypes and recovery from two-phase mixtures remain challenging.
- Together, our benchmark, systematic studies, and Bragglite clarify the capabilities and limitations of current PXRD-to-structure models.
Sources (1)
- [1]PXRD-Bench: How Close Are We to Determining Crystal Structures from Powder Diffraction?arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 09:34 AM
To fill this gap, we introduce PXRD-Bench, a unified benchmark that extends evaluation to unknown per-element atom counts and recovery of both structures in two-phase mixtures.
Determining crystal structures from powder X-ray diffraction (PXRD) would enable routine characterization of newly synthesized materials, including those absent from reference databases.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026Redesigning source control at Anthropic
- Oct 12, 2026Claude Managed Agents Q&A: Evals, CI/CD and Multi-Tenancy (Part 4 of 4) — Anthropic
- Oct 12, 2026Scheduled Agents, Dreaming and Q&A: Claude Managed Agents (Part 3 of 4) — Anthropic
- Oct 12, 2026Inside the AI Economy: What Stripe's Data Reveals — Nilofer Rajpurkar, Stripe
- Oct 12, 2026X-Ragent: Tool-Augmented Reinforcement Learning for Evidence-Grounded Chest X-Ray Reasoning
- Oct 10, 2026Satya Nadella says we should assume all AI models are ‘compromised’