Molecular Adsorption Simulation with Neural Network Potentials
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing molecular simulation methods, such as density functional theory (DFT) and Monte Carlo methods, face high computing costs and difficulties in simulating adsorption when molecular shapes change or chemical bonds break, making it challenging to screen many molecules effectively.
Innovation Solution
An information processing device employs a neural network potential (NNP) method and structure optimization techniques like BFGS to simulate the adsorption of monomolecules on a solid surface, allowing for the rapid and accurate determination of stable adsorption sites and angles, even when chemical bonds form or break, by using parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If quantum theory methods such as DFT are used for structure optimization, then accuracy of adsorption characteristics is improved, but computing cost becomes very high
Solution Approach 1:
The patent segments the simulation process into two distinct stages: (1) a pre-computation stage using quantum theory methods (DFT) to generate reference data for training a machine learning model, and (2) a production stage using the trained machine learning model for rapid prediction. This segmentation allows accurate quantum calculations to be performed only once for model training, while subsequent predictions use the computationally efficient machine learning model, thereby resolving the contradiction between accuracy and computing cost.
Solution Approach 2:
The patent performs preliminary action by pre-computing adsorption characteristics using quantum theory methods to train a machine learning model before actual molecular screening. The trained model is then reused for predicting adsorption characteristics of multiple molecules, eliminating the need to perform expensive quantum calculations repeatedly. This preliminary computation resolves the technical contradiction by paying the high computing cost once to enable fast, accurate predictions thereafter.
2Quantity of substance
If Monte Carlo method is used to consider multiple molecules adsorption, then adsorption characteristics of multiple molecules are obtained, but it is difficult to simulate adsorption where molecular shape changes or chemical bonds break
Solution Approach 1:
The patent replaces the traditional Monte Carlo method with a machine learning model-based approach. The machine learning model, trained on quantum theory calculations, can predict adsorption characteristics including scenarios where chemical bonds break and molecular shapes change, without being constrained by the fixed internal coordinates limitation of Monte Carlo methods. This substitution resolves the contradiction by maintaining multi-molecule simulation capability while gaining versatility in handling bond breaking and shape changes.
3Device complexity
If internal coordinates of molecule are fixed in Monte Carlo method, then computational simplicity is maintained, but adsorption where molecular shape changes significantly cannot be simulated
Solution Approach 1:
The patent substitutes the Monte Carlo method with a machine learning model that can handle variable molecular geometries. The machine learning model predicts adsorption characteristics for molecules with changing shapes and broken bonds without requiring fixed internal coordinates, thus maintaining computational efficiency while dramatically increasing adaptability to simulate shape changes.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables fast and accurate simulation of multiple molecules adsorbed on a solid surface, predicting adsorption energy distribution and properties like film quality and surface friction, with reduced computational burden.
Implementation Method 1
an adsorption phenomenon of molecules constituting the gas or liquid to a solid surface occurs
Implementation Method 2
chemisorption due to formation of chemical bonds such as covalent bonds
Implementation Method 3
physisorption due to van der Waals forces
Data Source
AI summary
An information processing device includes a memory and a processor. The processor is configured to: define a molecular model representing a target molecular structure and a solid surface model; acquire an adsorption site and a position and angle at which the molecular model of a monomolecule approaches the solid surface model by executing a simulation of placing the molecular model of the monomolecule on the solid surface model; execute a simulation of making a plurality of the molecular models adsorb on a plurality of the adsorption sites; and acquire a structure and adsorption energy distribution when the molecular model of multiple molecules is adsorbed on the solid surface model.