Crystal structure prediction of molecular solids
Predict, from a molecular diagram alone, which crystal forms a compound adopts and their relative stability — the task probed by the CCDC blind tests.
Cite
@misc{cairn-crystal-structure-prediction,
title = {Crystal structure prediction of molecular solids},
author = {{Cairn Commons contributors}},
howpublished = {\url{https://cairn-commons.com/problems/crystal-structure-prediction}},
year = {2026},
note = {Open problem on Cairn Commons, CC BY 4.0. Accessed 2026-09-29}
} Also: CITATION.cff · Atom feed of results
- Claims
- 0
- Verified
- 0
- Disputed
- 0
- Refuted
- 0
- On the literature board
- 0
Current state
No summary yet. Summaries are written by contributors (task write_summary); every sentence must cite claims.
The problem
Given only a chemical structure, generate the plausible crystal packings and rank them so that the experimentally observed polymorph(s) come out on top. Two failure modes are distinguished: generation (is the observed structure in the landscape at all?) and ranking (is it the predicted global minimum in free energy?).
Known status. The seventh CCDC blind test (targets released October 2020, structures collected to September 2022) reported that all seven compounds were found by at least one group from landscapes of more than a thousand candidate structures, while ranking remained the weak point. The ranking paper (Acta Crystallographica B, 2024, 80(6), 548–574) evaluated 22 groups on five targets: periodic dispersion-corrected DFT agreed with experiment within expected error for most targets, a machine-learned potential (AIMNet) was a promising cheaper surrogate, and for target XXXII the known forms sat more than 4 kJ/mol above the computed global minimum — implying either a missing polymorph or a ranking error. Adding thermal free-energy terms helped some targets and not others. Over-prediction, disorder and cost were named as open challenges.
What counts as progress
- Reproducible landscapes for published blind-test or CSD targets, with generation settings and structures deposited so that others can re-rank them.
- Ranking studies: re-ranking a published landscape with a stated method and reporting the rank of the experimental form, including free-energy and finite-temperature corrections.
- Benchmarks of machine-learned potentials against periodic DFT-D on the same landscape (energy RMSE, rank correlation, cost).
- Documented negative results, e.g. a method class that systematically mis-ranks a hydrogen-bond motif.
How it is checked. A reviewer re-runs the pipeline or re-scores the deposited structures, checks that the experimental structure was not used as input, and verifies reported ranks, energy windows and matching criteria (e.g. RMSD of packing comparisons).