Research map / Probabilistic and scientific modeling / Molecules and proteins
Quantitative Assessment of Crystal Structure Prediction
Sergio Rincón, Gabriel González, Nicolás Andrade, Rafael Velasquez, Sebastian Ojeda, Juanita Puentes, Nicolás Aparicio, Paula Cárdenas, Pablo Arbelaez
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a quantitative evaluation framework for crystal structure prediction that addresses threshold-sensitive metrics, neglect of physical plausibility, and polymorph handling.
Problem
Existing crystal structure prediction (CSP) evaluation relies on threshold-sensitive or incomparable metrics, focuses mainly on geometric accuracy while ignoring physical plausibility, and does not account for polymorphism.
Approach
A quantitative assessment framework for CSP that replaces threshold-based, hard-to-compare metrics with more comparable ones, adds physical plausibility checks alongside geometric accuracy, and accounts for polymorphs. Details beyond the abstract are not available.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply explicit SE(3) alignment before geometric distances (ablation) | OPAL-Bench protocol · Carbon-24, MP-20, MPTS-52 | geometric distances: 25-60% lower distances | Improved |
| apply explicit SE(3) alignment (ablation) | OPAL-Bench protocol · Perov-5 | geometric distances: fourfold reduction | Improved |
| evaluate against all known polymorphs instead of one reference (ablation) | OPAL-Bench protocol · Perov-5, MP-20, MPTS-52 | geometric distances: below 10% | No clear effect |
| evaluate against all known polymorphs (ablation) | OPAL-Bench protocol · Carbon-24 | geometric distances: 30-40% lower | Improved |
| use OPAL-CSP instead of baselines | OPAL-CSP · MP-20 | Chamfer distance: 3.866 -> 3.700 (vs FlowMM) | Improved |
| use OPAL-CSP instead of baselines | OPAL-CSP · MPTS-52 | physical metrics: FE 0.150 -> 0.113 | Improved |
| use OPAL-CSP instead of CrystalFlow | OPAL-CSP · MPTS-52 | Chamfer distance: 4.275 -> 4.325 | Hurt |
| geometric performance versus physical plausibility | all CSP methods · Perov-5, MP-20, MPTS-52 | correlation of geometric and physical distances | No clear effect |
| FlowLLM geometric competitiveness | FlowLLM · MPTS-52 | formation energy: 3264 eV/atom median error | Hurt |
| use ground-truth symmetry templates instead of CSPML templates | DiffCSP++, SGEquiDiff · Perov-5, MP-20, MPTS-52 | geometric distances: MP-20 Chamfer 4.104 -> 3.949 (DiffCSP++) | Improved |
| use ground-truth templates instead of CSPML | DiffCSP++, SGEquiDiff · Carbon-24 | geometric distances | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FlowMM — FlowMM: Generating Materials with Riemannian Flow Matching (ICML 2024) · better on 4 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- All that structure matches does not glitter (NeurIPS 2025 · similarity 0.89)
- Fast Organic Crystal Structure Prediction with Unit Cell Flow Matching (NeurIPS 2026 · similarity 0.89)
- EP-Flow: Disordered Crystal Structure Prediction without Site-level Annotation (NeurIPS 2026 · similarity 0.88)
- Local-Global Associative Frames for Symmetry-Preserving Crystal Structure Modeling (NeurIPS 2025 · similarity 0.87)
- Scalable Diffusion for Materials Generation (ICLR 2024 · similarity 0.87)
- A Diffusion-Based Pre-training Framework for Crystal Property Prediction (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
