Classical Force Field Parameter Design Using Deep Learning Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing force fields face challenges in balancing computational cost, learning cost, and versatility for simulating chemical reactions and adsorption, with first-principles calculations being too costly and classical force fields requiring high learning costs and low versatility.
Innovation Solution
A design device and method that utilizes a deep learning force field to infer physical property values and determine parameters for a classical force field, leveraging trained neural networks to simulate first-principles calculations efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If first-principles calculation is used for behavior analysis of chemical reactions or adsorption, then high accuracy and versatility are achieved, but computational cost becomes prohibitively high with a practical limit of about 100 atoms
Solution Approach 1:
A deep learning force field is introduced as an intermediary between first-principles calculation and classical force field. The deep learning model learns from first-principles calculation data and provides accurate potential energy surface predictions with computational cost comparable to classical force fields, thus mediating between accuracy and computational efficiency
Solution Approach 2:
The deep learning force field creates a copy of the first-principles calculation results by training on its data. Instead of performing expensive first-principles calculations repeatedly, the system uses the trained deep learning model to copy and reproduce first-principles accuracy at much lower computational cost
2Use of energy by moving object
If classical force field is used for simulation, then computational cost is very low enabling analysis of macromolecules with about one million atoms, but learning cost is high and versatility is low due to requiring many first-principles calculations to determine parameters
Solution Approach 1:
First-principles calculations are performed in advance to generate training data for the deep learning force field. This preliminary action creates a reusable model that captures the accuracy of first-principles calculations without requiring repeated expensive calculations, thereby improving versatility while maintaining low computational cost
Solution Approach 2:
The system changes from using fixed classical force field parameters determined by expensive fitting procedures to using adaptive deep learning parameters that are automatically learned from first-principles data. This parameter transformation enables the model to achieve first-principles accuracy with computational efficiency
3Measurement precision
If deep learning force field is trained to reproduce first-principles calculation results, then high accuracy and versatility are achieved, but computational cost or memory utilization does not reach that of the classical force field
Solution Approach 1:
The system uses a computationally inexpensive deep learning force field that can be trained once on first-principles data and then reused many times. The initial training cost is amortized over numerous simulations, making the effective computational cost per simulation comparable to or better than classical force fields while achieving superior accuracy
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A design device for designing a classical force field includes an inference unit configured to infer, using a deep learning force field that has learned results of a first-principles calculation of a selected chemical structure, physical property values of the chemical structure necessary for creating the classical force field; and a determination unit configured to determine parameters of a function system of the classical force field, using the inferred physical property values of the chemical structure.