Neural surrogates for crystal growth dynamics with variable supersaturation: explicit vs. implicit conditioning

2026-04-23 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Mathematics & Computational Sciences, Engineering & Applied Sciences · Depth: Expert, quick

Summary

Matteo Rigoni, Daniele Lanzoni, Francesco Montalenti, and Roberto Bergamaschini developed Convolutional Recurrent Neural Network (CRNN) surrogate models for simulating crystal growth dynamics, specifically addressing variable supersaturation. These models were trained on time-sequence datasets derived from numerical integration of Allen-Cahn dynamics, which incorporates faceting through kinetic anisotropy. Two distinct network architectures were explored: one implicitly infers supersaturation from an input mini-sequence of evolution frames to predict continuation, while the other explicitly takes the supersaturation parameter and a single initial frame to predict the entire sequence. Systematic testing revealed that explicit parameter conditioning yields superior results, accurately reproducing ground-truth profiles. The implicit mini-sequence approach achieved comparable performance only with larger training datasets. The models demonstrated strong conditioning to supersaturation, accurately reflecting its impact on growth rates and faceted morphology, and proved scalable to 256 times larger domains and over 10 times longer sequences with minimal error.

Key takeaway

For AI Scientists developing crystal growth simulations, prioritizing explicit parameter conditioning in your neural network architectures is crucial for achieving high-fidelity results. If explicit input is not feasible, be prepared to utilize significantly larger training datasets to approach comparable accuracy with implicit conditioning methods. Your models can then scale effectively to larger domains and longer sequences, enhancing simulation capabilities.

Key insights

Explicit supersaturation conditioning in CRNNs offers superior crystal growth simulation fidelity over implicit methods.

Principles

• Explicit parameter conditioning improves model accuracy.
• Larger datasets can compensate for implicit conditioning.
• CRNNs scale effectively for crystal growth simulations.

Method

Two CRNN architectures were developed: one implicitly infers supersaturation from an input mini-sequence, and another explicitly takes supersaturation as a direct input alongside an initial frame to predict crystal growth sequences.

In practice

• Use explicit parameter inputs for higher fidelity.
• Consider larger datasets for implicit conditioning.
• Apply CRNNs for scalable crystal growth modeling.

Topics

• Neural Surrogates
• Crystal Growth Dynamics
• Convolutional Recurrent Neural Networks
• Supersaturation
• Allen-Cahn Dynamics

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.
• Neural network surrogates with uncertainty quantification for inverse problems in partial differential equations — DeepGaLA uses uncertainty-aware neural network surrogates for scalable, reliable Bayesian inference in inverse problems, outperforming traditional methods in efficiency.
• Overcoming the Limits of Finite Difference Method; Physics-Informed Neural Network for Noisy High-Dimensional Heat Diffusion — PINNs overcome classical numerical method limitations in noisy, high-dimensional heat diffusion, maintaining accuracy and efficiency.
• Force-free molecular dynamics through autoregressive equivariant networks — TrajCast uses equivariant MPNNs to directly predict molecular states, enabling faster, data-efficient MD simulations with strong generalization.
• RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning — Deep chemical understanding and structured training enable compact LLMs to outperform larger models in reaction prediction.
• Preserving Plasticity in Continual Learning via Dynamical Isometry — Dynamical isometry, where layer-wise Jacobian singular values remain near one, is key to preserving plasticity in continual learning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.