PhononBench:A Large-Scale Phonon-Based Benchmark for Dynamical Stability in Crystal Generation

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Science & Research — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences, Physical Sciences & Chemistry · Depth: Expert, extended

Summary

PhononBench, the first large-scale benchmark for dynamical stability in AI-generated crystals, evaluates 108,843 structures from six leading generative models using the MatterSim interatomic potential, which offers DFT-level accuracy for phonon predictions. This benchmark reveals a significant limitation: the average dynamical stability rate across all generated structures is only 25.83%, with the top-performing model, MatterGen, achieving just 41.0%. In property-targeted generation, such as band-gap conditioning with MatterGen, stability remains low at 23.5% even at the optimal 0.5 eV band-gap. Space-group-controlled generation shows higher-symmetry crystals, like cubic systems, reaching up to 49.2% stability, but the overall average is 34.4%. The study also identifies 28,119 new phonon-stable crystal structures, providing a valuable resource for future materials exploration.

Key takeaway

For AI and research scientists developing crystal generative models, you must prioritize dynamical stability beyond thermodynamic criteria. Your current models likely produce many unstable structures, with average stability around 25.83%. Integrate MatterSim-v1 for efficient phonon calculations into your workflow to screen for stability. Focus on training with large, high-quality datasets and consider explicit stability constraints or post-generation validation to improve the physical viability of generated materials.

Key insights

Current AI crystal generation models largely fail to produce dynamically stable materials, averaging only 25.83% stability.

Principles

• Dynamical stability is critical for material synthesizability.
• Large, high-quality datasets improve generative model performance.
• Higher crystal symmetry correlates with increased dynamical stability.

Method

PhononBench uses MatterSim-v1 and Phonopy to perform high-throughput phonon calculations on relaxed crystal structures, assessing dynamical stability by checking for imaginary modes.

In practice

• Use MatterSim-v1 for efficient, DFT-accurate phonon calculations.
• Prioritize generative models trained on large, high-quality datasets.
• Incorporate explicit dynamical stability screening post-generation.

Topics

• Crystal Generation Models
• Dynamical Stability
• Phonon Calculations
• MatterSim
• Materials Discovery
• AI in Materials Science

Code references

• xqh19970407/PhononBench

Best for: AI Scientist, Research Scientist

Related on AIssential

• Finetuning-Free Diffusion Model with Adaptive Constraint Guidance for Inorganic Crystal Structure Generation — Adaptive constraint guidance in diffusion models generates thermodynamically stable inorganic crystal structures with user-defined properties.
• Advancing AI for materials with MatterSim: experimental synthesis, faster simulation, and multi-task models — MatterSim advancements enable faster, more accurate, and multi-property materials simulations, validated by experimental discovery.
• De novo molecular generation with optical property preconditioning at the token level — Token-conditioned autoregressive models can generate OLED molecules with targeted optical properties, but control varies by chemical motif.
• Mixing generative AI with physics to create personal items that work in the real world — PhysiOpt combines generative AI with physics simulations to create functional, real-world 3D-printable designs.
• Neural surrogates for crystal growth dynamics with variable supersaturation: explicit vs. implicit conditioning — Explicit supersaturation conditioning in CRNNs offers superior crystal growth simulation fidelity over implicit methods.
• Coupling Language Models with Physics-based Simulation for Synthesis of Inorganic Materials — Coupling LLMs with physics-based simulations offers more viable inorganic material synthesis routes than classical algorithms, leveraging LLMs' implicit priors.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.