Neural Network Potential for Adsorption Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for simulating adsorption processes, such as the grand canonical Monte Carlo (GCMC) method, face challenges in balancing calculation efficiency and accuracy, particularly when dealing with high-performance adsorbents and complex microstructures.
Innovation Solution
An information processing device is developed, equipped with a trained model that outputs physical property values when molecular information is input. This device performs simulations using a neural network potential (NNP) to define molecular and adsorbent models, allowing for calculations under arbitrary activity and temperature conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If quantum chemical calculation is used, then prediction accuracy is improved, but calculation time and cost increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training a neural network model using quantum chemical calculation data before actual simulations. This pre-computed model stores the relationship between molecular structures and adsorption energies, allowing rapid predictions during GCMC simulations without performing expensive quantum calculations in real-time, thus resolving the contradiction between accuracy and calculation time
Solution Approach 2:
The patent uses copying by creating a neural network model that replicates the behavior of quantum chemical calculations. Instead of performing actual quantum chemical computations during simulations, the system copies the essential predictive capabilities of quantum methods into a trained neural network, enabling fast and accurate adsorption energy predictions
2Productivity
If classical force field calculation is used, then calculation speed is improved, but treatment of chemical reactions and accuracy are limited
Solution Approach 1:
The patent applies parameter changes by transitioning from fixed classical force field parameters to flexible neural network parameters. The neural network model learns optimal parameters from quantum chemical data, enabling it to accurately represent both physical adsorption (van der Waals interactions) and chemical adsorption (bonding reactions), thus improving accuracy while maintaining the computational efficiency of classical methods
3Measurement precision
If advanced modeling is used for quantum chemical calculation, then prediction accuracy is improved, but device complexity and calculation cost increase
Solution Approach 1:
The patent applies taking out by extracting the complex quantum chemical calculation process from the simulation workflow and replacing it with a pre-trained neural network model. The neural network encapsulates the complex quantum mechanical relationships in a simplified computational form, maintaining prediction accuracy while dramatically reducing modeling complexity and calculation costs during actual simulations
Data Source
AI summary
An information processing device includes a memory and a processor. The memory stores information on a trained model that outputs a physical property value when information on a molecule is input. The processor defines a molecular model representing a target molecular structure and an adsorbent model representing a structure of an adsorbent, performs a simulation in which the trained model is used at least in part in a first model in which the molecular model is placed around the adsorbent model under arbitrary activity and temperature conditions, and acquires an adsorption volume and adsorption structure as a result of the simulation.