Fast, accurate, and tunable: Advancing battery materials innovation with Schrödinger’s Machine Learning Force Fields

The problem:

Developing next-generation energy storage solutions requires a deep understanding of complex, multiscale phenomena—from ion transport in electrolytes to the reactive formation of the solid-electrolyte interphase (SEI). Historically, researchers have been forced to choose between two extremes: the high accuracy but prohibitive computational cost of Density Functional Theory (DFT), or the speed of classical force fields that often lack the “physics” necessary to capture reactive events or complex chemistries. This “simulation gap” delays time-to-market and limits the ability to explore a vast chemical space.

The solution:

This webinar introduces Schrödinger’s state-of-the-art machine learning force field (MLFF) framework, featuring the MPNICE (Message Passing Network with Iterative Charge Equilibration) and QRNN (Charge Recursive Neural Network) architectures. By combining the accuracy of physics-based modeling with the transformative speed of machine learning, Schrödinger provides a “best-of-both-worlds” solution that eliminates traditional trade-offs. We will present live demos showcasing applications of MLFFs for accurate modeling of complex systems including liquid and solid-state electrolytes.

Key highlights:

• Rapid Efficiency: Utilize GPU-accelerated engines like Desmond to accelerate the MD simulations, enabling accurate modeling of complex systems like electrolyte formulations and cathode coatings
• Near-DFT Accuracy at Scale: Achieve quantum-level precision for energy and force predictions while simulating large systems at timescales previously reserved for classical MD
• Unrivaled Tunability: Unlike “black-box” models, Schrödinger’s MLFFs are highly customizable, allowing researchers to incorporate explicit electrostatics and iterative charge equilibration to model ionic liquids and battery interfaces with high fidelity
• Seamless Usability: Integrated within the intuitive Schrödinger Materials Science platform, these tools allow users to deploy advanced digital workflows without machine learning expertise