2026 · arXiv · Preprint

C2NP: A Benchmark for Learning Scale-Dependent Geometric Invariances in 3D Materials Generation

Can Polat , Erchin Serpedin , Mustafa Kurban , Hasan Kurban

Research highlights

C2NP: testing the transition from crystals to nanoparticles

Measure whether generative models preserve structure and symmetry across changes in physical size.

Why we investigated it

Bulk unit cells and finite nanoparticles differ through surfaces and truncation, so bulk accuracy does not establish scale generalization.

What the work contributes

C2NP tests both radius-conditioned nanoparticle generation and recovery of bulk lattice parameters and space groups.

What the evidence shows

Across more than 170,000 configurations, evaluated models can have low losses yet fail joint structure and symmetry tests under shift.

Build on this work

Use size-based splits and both task directions to diagnose geometric failures in materials generators.

  • Crystal generation
  • Nanoparticles
  • Scale generalization
  • Symmetry recovery
Abstract
Generative models for materials have achieved strong performance on periodic bulk crystals, yet their ability to generalize across scale transitions to finite nanostructures remains largely untested. We introduce Crystal-to-Nanoparticle (C2NP), a systematic benchmark for evaluating generative models when moving between infinite crystalline unit cells and finite nanoparticles, where surface effects and size-dependent distortions dominate. C2NP defines two complementary tasks: (i) generating nanoparticles of specified radii from periodic unit cells, testing whether models capture surface truncation and geometric constraints; and (ii) recovering bulk lattice parameters and space-group symmetry from finite particle configurations, assessing whether models can infer underlying crystallographic order despite surface perturbations. Using diverse materials as a structurally consistent testbed, we construct over 170,000 nanoparticle configurations by carving particles from supercells derived from DFT-relaxed crystal unit cells, and introduce size-based splits that separate interpolation from extrapolation regimes. Experiments with state-of-the-art approaches, including diffusion, flow-matching, and variational models, show that even when losses are low, models often fail geometrically under distribution shift, yielding large lattice-recovery errors and near-zero joint accuracy on structure and symmetry. Overall, our results suggest that current methods rely on template memorization rather than scalable physical generalization. C2NP offers a controlled, reproducible framework for diagnosing these failures, with immediate applications to nanoparticle catalyst design, nanostructured hydrides for hydrogen storage, and materials discovery. Dataset and code are available at https://github.com/KurbanIntelligenceLab/C2NP.
Cite
@misc{polat2026c2np,
  title = {C2NP: A Benchmark for Learning Scale-Dependent Geometric Invariances in 3D Materials Generation},
  author = {Can Polat and Erchin Serpedin and Mustafa Kurban and Hasan Kurban},
  year = 2026,
  journal = {arXiv},
  doi = {10.48550/arxiv.2601.19076},
  url = {https://arxiv.org/pdf/2601.19076},
}