Machine Learning for Multi-Scale Molecular Simulation and Design

Explore advanced machine learning techniques for multi-scale molecular simulation and design in this comprehensive talk. Delve into coarse-grained modeling methods that extend time and length scales in molecular simulations. Learn about a novel multi-scale graph neural network approach for simulating coarse-grained molecular dynamics with large time steps, demonstrating effectiveness in complex systems like single-chain coarse-grained polymers and multi-component Li-ion polymer electrolytes. Discover how this method achieves structural and dynamical property recovery at significantly higher speeds compared to classical force fields. Investigate an innovative approach to designing Metal-organic frameworks (MOFs) for carbon capture applications, combining diffusion-based generative modeling with coarse-grained representations of building blocks. Gain insights into motivations behind accelerating molecular dynamics, past efforts in the field, and the challenges of extending simulation scales. Explore topics such as graph clustering coarse-graining, long time-integration steps, and simulation screening of large chemical spaces. Conclude with a discussion on spatio-temporal coarse-graining tradeoffs and participate in a Q&A session to deepen your understanding of these cutting-edge techniques in molecular simulation and design.

- Intro
- Motivation
- Past efforts in accelerating MD
- Extending MD simulation scales
- Graph clustering coarse-graining
- Long time-integration step
- Simulation screening a large chemical space
- Experiments
- Spatio-temporal coarse graining tradeoffs
- Q+A