Solid Electrolyte Conductivity Prediction Using MTP and MD
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting lithium ion conductivity of solid electrolytes, such as ab initio molecular dynamics (AIMD) and molecular dynamics simulations, face challenges like significant differences between predicted and experimental values, limited accuracy at room temperature, and high computational costs, as well as the requirement for specific structural and compositional correspondence.
Innovation Solution
A method involving machine learning to simulate crystal structures, create training sets using AIMD, calculate Moment Tensor Potential (MTP) specific to the simulated structures, and predict lithium ion conductivity through molecular dynamics simulations, allowing for rapid and precise calculations, particularly for argyrodite-type crystal structures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ab initio molecular dynamics (AIMD) method using density functional theory (DFT) is used to predict lithium ion conductivity, then calculation can be performed without requiring specific potential data, but the predicted values show significant difference from experimental values and computational cost is high
Solution Approach 1:
The patent introduces machine learning models as an intermediary between DFT calculations and molecular dynamics simulations. The ML models are trained on DFT data to learn the relationship between crystal structures and lithium ion conductivity, then use this learned knowledge to make accurate predictions without requiring direct DFT calculations for each new material, thus improving both accuracy and efficiency
Solution Approach 2:
The patent performs preliminary DFT calculations and machine learning training on a dataset of known solid electrolytes before making predictions on new materials. This preliminary action creates a trained ML model that can quickly predict conductivity for new crystal structures without requiring expensive DFT calculations for each case
2Adaptability or versatility
If ab initio molecular dynamics (AIMD) method using density functional theory (DFT) is used to predict lithium ion conductivity, then calculation can be performed without requiring specific potential data, but computational cost is high
Solution Approach 1:
The patent introduces machine learning models as an intermediary between DFT calculations and molecular dynamics simulations. The ML models are trained on DFT data to learn the relationship between crystal structures and lithium ion conductivity, then use this learned knowledge to make accurate predictions without requiring direct DFT calculations for each new material, thus improving both accuracy and efficiency
Solution Approach 2:
The patent creates a computational model that copies the essential physics and chemistry of DFT calculations through machine learning. Instead of performing expensive quantum mechanical calculations for each new material, the ML model replicates the DFT results based on patterns learned from training data, dramatically reducing computational cost
3Use of energy by moving object
If molecular dynamics simulations are used to calculate lithium ion conductivity, then computational cost is reduced and room temperature prediction is possible, but calculation cannot be made unless there is a potential corresponding to the structure and composition
Solution Approach 1:
The patent introduces machine learning models as an intermediary that generates appropriate potential functions for molecular dynamics simulations based on crystal structure information. The ML models learn the relationship between crystal structures and suitable potential parameters from training data, automatically providing the necessary potentials for new materials without requiring manual parameterization
Solution Approach 2:
The patent enables the molecular dynamics simulation system to self-generate the necessary potential functions by using machine learning models that automatically determine appropriate potentials based on input crystal structure information. This eliminates the need for external manual potential selection or fitting for each new material system
4Measurement precision
If research on synthesis of solid electrolytes with various crystal structures and compositions is conducted to develop high conductivity materials, then material performance is improved, but time and effort required is excessive
Solution Approach 1:
The patent creates a computational model that copies the complex relationship between crystal structure, composition, and lithium ion conductivity through machine learning. Once trained on experimental and theoretical data, the model can quickly predict conductivity for new material compositions and structures, replacing time-consuming trial-and-error synthesis and characterization cycles
Solution Approach 2:
The patent performs preliminary screening and prediction using machine learning models before actual material synthesis. Researchers can use the trained model to identify promising crystal structures and compositions that are likely to exhibit high conductivity, thereby prioritizing which materials to synthesize and characterize experimentally, significantly reducing overall research time
Data Source
AI summary
Disclosed is a method of rapidly and precisely predicting lithium ion conductivity of a solid electrolyte. The method may include simulating a crystal structure of the solid electrolyte and creating training sets based on the crystal structure for machine learning; calculating a potential specific to the simulated crystal structure by machine learning using the training sets; and predicting the lithium ion conductivity of the solid electrolyte from the potential using molecular dynamics simulations.