ACHIEVING DYNAMIC ACCURACY IN MACHINE-LEARNED CG POTENTIALS BY MODULATING POTENTIAL ENERGY LANDSCAPE
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Molecular dynamics, Neural network potential, Graph Neural Network, Neural ODE, Coarse-graining, Dynamics
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: This paper introduces a neural network-based coarse-grained water model (a single pseudo-atom model) optimized to model the potential energy landscape with the aim of reproducing the structure and dynamics of a reference water model.
Abstract: In this paper, we introduce a coarse-grained (CG) model designed to reproduce the structure and dynamics of all-atom systems. Our approach combines a graph neural network potential and a high-frequency potential energy surface landscape function to effectively capture essential features of the fine-grained atomistic model. The Neural-Network potential accurately captures complex atomic inter- actions using learned representations and can be effectively parameterized to re- produce distribution functions from high-fidelity all-atom (AA) simulations. Nev- ertheless, such parameterization inherently smoothens out the AA Energy land- scape, resulting in the loss of information required for capturing the system dy- namics. We, therefore, provide a route to enrich the ML CG potentials for bulk systems by emulating the AA landscape within the mapped CG ensemble by aug- menting the GNN potential with a high-frequency potential term, thereby pro- viding an accurate representation of CG dynamics as well as the structure. We demonstrate the utility of our framework by reproducing the Radial Distribution Function (RDF) and the mean-square displacement (MSD) of various AA and CG systems. Notably, we apply our methodology to coarse-grain the widely used SPC/E water model, thereby providing compelling evidence of the fidelity of our model to coarse-grain complex systems, which include electrostatic and multi- body effects. Our work signifies a significant step towards more efficient and accurate simulations of complex systems using coarse-grained methodologies.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4342
Loading