Chemtrain: Learning Deep Potential Models via Automatic Differentiation and Statistical Physics

This article introduces chemtrain, a modular and customizable software framework for training neural network (NN) potential models, which enhance molecular dynamics (MD) simulations by accurately modeling interactions at different scales. chemtrain addresses the challenges of generating sufficient and precise training data through a combination of top-down (macroscopic experimental data) and bottom-up (microscopic simulation data) learning strategies. It supports advanced training algorithms such as Force Matching (FM), Relative Entropy Minimization (RM), and Differentiable Trajectory Reweighting (DiffTRe), leveraging the JAX library for automatic differentiation, scalability, and computational efficiency. By providing an object-oriented interface, chemtrain simplifies complex workflows, including combining pre-training methods with fine-tuning or active learning approaches. The framework’s flexibility is demonstrated through examples of parameterizing both all-atomistic models (e.g., titanium) and coarse-grained models (e.g., alanine dipeptide in water), showcasing its ability to fuse experimental and simulation data for accurate potential models.

For more details, please continue reading the full article under the following link:


In general, if you enjoy reading this kind of scientific news articles, I am always keen to connect with fellow researchers in materials science, including the possibility to discuss about any potential interest in the Materials Square cloud-based online platform ( www.matsq.com ), designed for streamlining the execution of materials and molecular modelling simulations!

The Materials Square platform in fact provides extensive web-based functionalities for computational chemistry/materials research and training/education purposes, as detailed also in the following PDF brochure: https://www.materialssquare.com/wp-content/uploads/Materials_Square_Brochure_2022-compressed_1674181754.pdf

Many thanks for your interest and consideration,

Dr. Gabriele Mogni
Technical Consultant and EU Representative of Virtual Lab Inc., the parent company of the Materials Square platform
Website: Home | Virtual Lab Inc.
Email: gabriele@simulation.re.kr

#materials #materialsscience #materialsengineering #computationalchemistry #modelling #chemistry #researchanddevelopment #research #MaterialsSquare #ComputationalChemistry #Tutorial #DFT #simulationsoftware #simulation