Substitution-Based Analysis of Structural Novelty for Generative Models of Materials
2026-06-22 • Machine Learning
Machine Learning
AI summaryⓘ
The authors studied AI models that create new inorganic crystal designs to see if these models really explore new types of crystals or just recycle old known ones. They found that most AI-generated crystals are either very similar to ones the AI already learned or can be made by simple substitutions in known structures. This recycling is especially common in crystals with high symmetry, while AI does explore more new structures in less symmetrical crystals. The authors highlight that current AI models tend to stick close to familiar crystal patterns rather than discovering completely new ones.
generative AIinorganic crystalscrystal symmetryelemental substitutionstructural prototypesmetastabilitytraining datacrystal designstructural fingerprintsmemorisation
Authors
Masahiro Negishi, Aron Walsh
Abstract
There has been rapid progress in generative artificial intelligence (AI) models for inorganic crystal design, which can efficiently generate large numbers of candidate compounds after being trained on databases of known crystals. However, it remains unclear whether they genuinely expand the accessible materials search space beyond conventional strategies such as elemental substitution within known structure types. We address this question by developing a workflow to assess whether AI-generated crystals are duplicates of training structures, reproducible by elemental substitution, or unmatched by either criterion. Applying this workflow to representative generative models reveals that 81-92% of chemically valid and metastable generated crystals are either training duplicates or substitution-derived structures. This tendency is particularly strong in high-symmetry crystal systems, even though many possible structural prototypes remain unexplored. Further analysis of the underlying structural fingerprints shows that low-symmetry structures beyond duplication or substitution can be interpreted as interpolation in training-data-rich regions, while high-symmetry duplicates appear to result from memorisation in training-sparse regions. Our findings highlight a limitation in the current generation of models that exhibit a bias towards known structural prototypes in the high symmetry regions, but enable wider exploration of the low-symmetry structural space.