Let CSP Be Your ANCHOR: Adaptive Crystal Search over Frozen Structure Priors
Abstract
De novo crystal generation (DNG) models decide where to search in composition space and how to generate structures with one set of weights. We argue that discovery is better served by separating the two. A crystal structure prediction (CSP) model is a physical prior that should be improved by likelihood training, while rewards, including novelty measured against the search's own history, should act on a search over compositions. We introduce ANCHOR, a GRPO composition policy trained with multi-objective rewards around a frozen CSP model, and continuous adaptive novelty (CAN), a graded novelty score against known structures and a growing discovery history. Using the frozen CSP model as a fixed ruler under one evaluator, we test where adaptation should act. Replacing DNG compositions with ANCHOR's policy on the same CSP backbone raises MSUN from 11.4% to 47.6% and SUN from 1.1% to 22.1% at 99.9% formula uniqueness. Fine-tuning DNG models directly on the same rewards instead moves their composition marginal without raising their on-hull fraction. We show that KL-regularized fine-tuning of a DNG model can only reweight chemistry the pretrained model already supports by a bounded factor, while unregularized DNG fine-tunes move toward known or less stable chemistry. Even a stability-only reward routed into ANCHOR's CSP backbone roughly halves SUN relative to the frozen backbone, whereas likelihood training on structures found during search can improve a CSP backbone. Under MatterGen's evaluation pipeline, ANCHOR raises state-of-the-art MSUN from 29.2% to 41.3%, transfers without retraining to two further CSP backbones, and reaches 47.1% after distillation into Crystalite-CSP. As with any model optimised against a potential, its on-hull rate depends on that potential.
Related
- DiffGA: Steering Frozen Diffusion Models for Materials Design via Evolutionary Search
- Crystal structure of URu4Si2, a new Ru-rich phase in the U–Ru–Si ternary system
- Bridging powder and multi-crystal diffraction with basis-adaptive texture tomography
- Interplay between crystal structure and magnetism in CeCrB4
- Text-guided flow matching enables sample-efficient crystal structure generation
BibTeX
@misc{hovmand2026your,
title = {{Let CSP Be Your ANCHOR: Adaptive Crystal Search over Frozen Structure Priors}},
author = {Emma Lei Hovmand and Jonas Elsborg and Melih Kandemir and Arghya Bhowmik},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.33407},
eprint = {2609.33407},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.33407},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Let CSP Be Your ANCHOR: Adaptive Crystal Search over Frozen Structure Priors
AU - Emma Lei Hovmand
AU - Jonas Elsborg
AU - Melih Kandemir
AU - Arghya Bhowmik
PY - 2026
DA - 2026/09/30
PB - arXiv
DO - 10.48550/arxiv.2609.33407
UR - https://arxiv.org/abs/2609.33407
AB - De novo crystal generation (DNG) models decide where to search in composition space and how to generate structures with one set of weights. We argue that discovery is better served by separating the two. A crystal structure prediction (CSP) model is a physical prior that should be improved by likelihood training, while rewards, including novelty measured against the search's own history, should act on a search over compositions. We introduce ANCHOR, a GRPO composition policy trained with multi-objective rewards around a frozen CSP model, and continuous adaptive novelty (CAN), a graded novelty score against known structures and a growing discovery history. Using the frozen CSP model as a fixed ruler under one evaluator, we test where adaptation should act. Replacing DNG compositions with ANCHOR's policy on the same CSP backbone raises MSUN from 11.4% to 47.6% and SUN from 1.1% to 22.1% at 99.9% formula uniqueness. Fine-tuning DNG models directly on the same rewards instead moves their composition marginal without raising their on-hull fraction. We show that KL-regularized fine-tuning of a DNG model can only reweight chemistry the pretrained model already supports by a bounded factor, while unregularized DNG fine-tunes move toward known or less stable chemistry. Even a stability-only reward routed into ANCHOR's CSP backbone roughly halves SUN relative to the frozen backbone, whereas likelihood training on structures found during search can improve a CSP backbone. Under MatterGen's evaluation pipeline, ANCHOR raises state-of-the-art MSUN from 29.2% to 41.3%, transfers without retraining to two further CSP backbones, and reaches 47.1% after distillation into Crystalite-CSP. As with any model optimised against a potential, its on-hull rate depends on that potential.
ER -