Neural network force field computational training routines for molecular dynamics computer simulations

Patent metadata

Field	Value
Publication number	US-12094580-B2
Application number	US-202117484448-A
Country	US
Kind code	B2
Filing date	Sep 24, 2021
Priority date	Sep 24, 2021
Publication date	Sep 17, 2024
Grant date	Sep 17, 2024

Abstract

A computational method for training a neural network force field (NNFF) configured to simulate molecular and/or atomic motion within a material system. The method includes the step of receiving molecular structure data of a molecule in the material system. The method also includes optimizing a geometry of the molecule using the molecular structure data and a density functional theory (DFT) simulation to obtain DFT optimized geometry data. The method further includes optimizing the geometry of the molecule using the molecular structure data and a classical force field (FF) simulation to obtain FF optimized geometry data. The method also includes outputting NNFF training data comprised of the DFT optimized geometry data and the FF optimized geometry data. The NNFF training data is configured to train an NNFF for simulating molecular and/or atomic molecular and/or atomic motion within the material system.

First claim

What is claimed is:

1. A computational process for training a neural network force field (NNFF) configured to simulate molecular and/or atomic motion within a material system, the process comprising:
receiving molecular structure data of a molecule in the material system;
optimizing a geometry of the molecule using the molecular structure data and a density functional theory (DFT) simulation to obtain DFT optimized geometry data;
optimizing the geometry of the molecule using the molecular structure data and a classical force field (FF) simulation to obtain FF optimized geometry data; and
outputting NNFF training data comprised of the DFT optimized geometry data and the FF optimized geometry data, the NNFF training data is configured to train the NNFF for simulating molecular and/or atomic motion within the material system; and
training the NNFF with the NNFF training data to simulate molecular and/or atomic motion within the material system while reducing the training cost of molecular dynamics data.

2. The computational process of claim 1 , wherein the FF optimized geometry data includes one or more normal mode displacements of the molecule.

3. The computational process of claim 1 , wherein the classical FF simulation includes a harmonic analysis.

4. The computational process of claim 1 , wherein the classical FF simulation is selected from the group consisting of: a universal force field (UFF), a classical force field, a reactive force field (ReaxFF), SchNet, fast learning of atomistic rare events (FLARE), and a graph neural network force field (GNNFF).

5. The computational process of claim 1 , wherein the DFT optimized geometry data includes an equilibrium structure of the molecule.

6. The computational process of claim 1 , wherein the NNFF training data includes one or more trajectories for the molecule.

7. The computational process of claim 1 , wherein the material system is a portion of one of the following systems: a fuel cell, a water desalination system, a catalysis system, a coating system, and a battery system.

8. A computational process for training a neural network force field (NNFF) configured to simulate molecular and/or atomic motion within a material system, the process comprising:
receiving molecular structure data of a molecule in the material system, the molecular structure data includes an irrational structure of the molecule;
solvating the irrational structure of the molecule using a classical FF simulation or an ab initio molecular dynamics (MD) simulation to obtain solvation trajectory data of the molecule;
outputting NNFF training data comprised of the solvation trajectory data, the NNFF training data is configured to train the NNFF for simulating molecular and/or atomic motion within the material system; and
training the NNFF with the NNFF training data to simulate molecular and/or atomic motion within the material system to minimize the irrational structure predicted by MD simulations.

9. The computational process of claim 8 , further comprising isolating the irrational structure of the molecule before the solvating step.

10. The computational process of claim 8 , further comprising minimizing the solvation trajectory data to obtain minimized, solvation trajectory data, and the NNFF training data includes the minimized, solvation trajectory data.

11. The computational process of claim 8 , further comprising obtaining one or more forces and/or energies of the molecule using a density functional theory (DFT) simulation and the solvation trajectory data.

12. The computational process of claim 8 , wherein the classical FF simulation is selected from the group consisting of: a universal force field (UFF), a reactive force field (ReaxFF), SchNet, fast learning of atomistic rare events (FLARE), and a graph neural network force field (GNNFF).

13. The computational process of claim 8 , wherein the irrational structure of the molecule is a predicted high energy irrational structure of the molecule.

14. The computational process of claim 8 , wherein the material system is a portion of one of the following systems: a fuel cell, a water desalination system, a catalysis system, a coating system, and a battery system.

15. A computational process for training a neural network force field (NNFF) configured to simulate molecular and/or atomic motion within a material system, the process comprising:
receiving molecular structure data of a molecule in the material system, the molecular structure data is predicted from a density functional theory (DFT) simulation, and the molecular structure data includes one or more tracked bond lengths, bond angles, and/or chemical structures;
evaluating the molecular structure data to obtain geometry error data of the one or more tracked bonds, angles, and/or structures;
outputting NNFF training data comprised of the geometry error data, the NNFF training data is configured to train the NNFF for simulating molecular and/or atomic motion within the material system; and
training the NNFF with the NNFF training data to simulate molecular and/or atomic motion within the material system to promote accurate structure predictions without additional molecular dynamics simulation datapoints.

16. The computational process of claim 15 , wherein the evaluating step includes evaluating the molecular structure data using the NNFF to obtain a DFT energy of the material system, the NNFF training data comprises the DFT energy.

17. The computational process of claim 15 , wherein the evaluating step includes minimizing the geometry error data using the NNFF to obtain minimized geometry error data, the NNFF training data comprises the minimized geometry error data.

18. The computational process of claim 15 , wherein the molecular structure data includes a weight for each of the one or more tracked bond lengths, bond angles, and/or chemical structures.

19. The computational process of claim 15 , wherein the one or more tracked bonds, angles, and/or structures includes at least one bond length and at least one bond angle.

20. The computational process of claim 15 , wherein the material system is a portion of one of the following systems: a fuel cell, a water desalination system, a catalysis system, a coating system, and a battery system.

Assignees

• Bosch Gmbh Robert

Inventors

• Ganeshan Karthik
• Gadelrab Karim
• Kornbluth Mordechai
• Mailoa Jonathan

Classifications

• G16C60/00 Primary

  Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation · CPC title

• G06N3/08

  Learning methods · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G16C10/00

  Computational theoretical chemistry, i.e. ICT specially adapted for theoretical aspects of quantum chemistry, molecular mechanics, molecular dynamics or the like · CPC title