Neural Network Molecular Dynamics for Chemical Reaction Acceleration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network potentials (NNPs) face challenges in dynamically simulating chemical reactions within a practical computational timeframe, making it difficult to handle processes like polymerization and decomposition using molecular dynamics (MD) simulations.
Innovation Solution
An information processing apparatus that identifies target atoms for chemical bonding, applies action and additional forces using a neural network, and executes molecular dynamics simulations to accelerate chemical reactions by applying boost potentials.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If molecular dynamics simulation is performed using neural network potentials, then calculation speed is improved, but chemical reactions cannot be observed within practical computational timeframe
Solution Approach 1:
The patent applies preliminary action by pre-identifying target atoms that are likely to undergo chemical bonding before the molecular dynamics simulation begins. This allows the simulation to focus computational resources on atoms most likely to react, thereby accelerating the observation of chemical reactions within practical computational timeframes while maintaining the speed advantages of neural network potentials
2Measurement precision
If density functional theory is used for electronic state simulation, then calculation accuracy is improved, but computational time becomes excessively long
Solution Approach 1:
The patent applies local quality by using neural network potentials that are specifically trained to provide high-accuracy predictions for local atomic environments and chemical bonding scenarios. This allows the system to maintain calculation accuracy comparable to density functional theory for relevant chemical processes while achieving the speed improvements needed for practical computational timeframes
Data Source
AI summary
An information processing apparatus according to an embodiment includes at least one memory, and at least one processor. The at least one processor is configured to: identify target atoms subject to chemical bonding among atoms; acquire information regarding a first action force acting on each of the atoms, the information being generated by inputting an atomic structure of the atoms into a neural network; acquire information regarding a first additional force to be applied to at least one of the target atoms; and execute a molecular dynamics simulation for the atoms using the information regarding the first action force, the information regarding the first additional force, and position information of the atoms.