Skip navigation
Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp01hm50tv65x
Title: Machine learning for multi-scale molecular modeling: theories, algorithms, and applications
Authors: Zhang, Linfeng
Advisors: E, Weinan
Car, Roberto
Contributors: Applied and Computational Mathematics Department
Keywords: Deep Potential
Enhanced Sampling
Machine Learning
Molecular Dynamics
Multi-scale Molecular Modeling
Reinforced Dynamics
Subjects: Applied mathematics
Computational chemistry
Issue Date: 2020
Publisher: Princeton, NJ : Princeton University
Abstract: In recent years, machine learning has emerged as a promising tool for dealing with the difficulty of representing high dimensional functions. This gives us an unprecedented opportunity to revisit theoretical foundations of various scientific fields, develop new schemes, improve existing methodologies, and solve problems that were too complicated for conventional approaches to address. In this dissertation, we identify a list of such problems in the context of multi-scale molecular modeling and propose machine learning based strategies to boost simulations with {\it ab initio} accuracy to much larger scales than conventional approaches. We consider two representative challenges: 1) how to go from many-electron-ion to atomistic systems, for which the key has been a general and efficient representation of the potential energy surface generated by electronic structure models; 2) how to go from atomistic to coarse-grained systems, for which one is interested in the free energy of the coarse-grained variables as well as the associated dynamical behavior. Our strategies follow two seemingly obvious but non-trivial principles: 1) machine learning based models should respect important physical constraints like symmetry; 2) to build truly reliable models, efficient algorithms are needed to construct a minimal but truly representative training data set. We use these principles to construct the Deep Potential model for the potential energy surface, the Deep Potential Molecular dynamics (DeePMD) which is a new paradigm for performing {\it ab initio} molecular dynamics, a concurrent learning scheme (DP-GEN) for generating the data set on the fly, algorithms for constructing the Wannier centers (Deep Wanner) and for efficiently exploring the free energy landscape (Reinforced Dynamics), as well as a machine learning-based coarse grained molecular dynamics model (DeePCG), etc. Applications of these models and algorithms are presented for problems in chemistry, biology, and materials science. Finally, we present our efforts on developing related open-source software packages, which have now been widely used worldwide by experts and practitioners in the molecular simulation community.
URI: http://arks.princeton.edu/ark:/88435/dsp01hm50tv65x
Alternate format: The Mudd Manuscript Library retains one bound copy of each dissertation. Search for these copies in the library's main catalog: catalog.princeton.edu
Type of Material: Academic dissertations (Ph.D.)
Language: en
Appears in Collections:Applied and Computational Mathematics

Files in This Item:
File Description SizeFormat 
Zhang_princeton_0181D_13353.pdf17.82 MBAdobe PDFView/Download


Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.