Bayesian Optimization for Ultra-Incompressible Crystal Structure Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current materials discovery methods, such as density functional theory (DFT), are limited by high computational costs and poor scalability, restricting the exploration of vast chemical and structural spaces, and rely on expensive ab initio calculations for obtaining equilibrium crystal structures, which hinders the application of machine learning models for predicting material properties across diverse chemical and structural spaces.
Innovation Solution
The integration of a Bayesian optimization with symmetry relaxation (BOWSR) algorithm and graph neural networks, like the MatErials Graph Network (MEGNet) formation energy model, allows for the generation of equilibrium crystal structures without relying on DFT calculations, enabling efficient screening of candidate crystal structures for exceptional properties like ultra-incompressibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If density functional theory (DFT) calculations are used to obtain equilibrium crystal structures, then accuracy of structural prediction is improved, but computational cost and time consumption increase significantly
Solution Approach 1:
The patent applies preliminary action by using Bayesian optimization to pre-relax crystal structures and predict equilibrium configurations before machine learning property predictions are performed. This preliminary structural relaxation step enables subsequent ML models to operate on accurate equilibrium structures without requiring expensive ab initio calculations at every stage, thus reducing overall computational time while maintaining prediction accuracy
Solution Approach 2:
The patent introduces an intermediary approach by employing machine learning models trained on DFT data to approximate equilibrium crystal structures and their properties. These ML models serve as intermediaries between full DFT calculations and final material screening, enabling rapid evaluation of vast chemical and structural spaces with accuracy comparable to DFT but at fraction of the computational cost
2Reliability
If ab initio calculations are performed to obtain equilibrium crystal structures, then reliability of structural data is improved, but scalability to vast chemical spaces deteriorates
Solution Approach 1:
The patent applies copying by training machine learning models on datasets generated from ab initio calculations. Once trained, these models can rapidly predict properties for new crystal structures without requiring repeated expensive calculations. This copying approach allows the system to evaluate thousands of candidate structures with reliability comparable to the training data while achieving high screening throughput across vast chemical spaces
Solution Approach 2:
The patent changes parameters by transitioning from first-principles quantum mechanical calculations to machine learning predictions for property evaluation. This parameter change in the computational methodology maintains reliability through careful model training and validation while dramatically improving productivity by enabling rapid screening of large numbers of candidate materials
3Productivity
If machine learning models are applied to predict material properties across diverse chemical spaces, then productivity of materials discovery is improved, but accuracy of predictions deteriorates without accurate equilibrium structures
Solution Approach 1:
The patent applies preliminary action by performing Bayesian optimization to relax crystal structures and predict equilibrium configurations before machine learning property predictions are performed. This preliminary structural relaxation ensures that ML models operate on accurate equilibrium structures, maintaining prediction accuracy while enabling high-throughput screening of diverse chemical spaces
Solution Approach 2:
The patent implements feedback by using machine learning models trained on accurate DFT data to guide the search for equilibrium crystal structures. The models provide feedback on predicted properties and stability, enabling iterative refinement of structural predictions while maintaining both high productivity in exploring chemical spaces and high accuracy in property predictions
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
This approach significantly improves the accuracy of machine learning-based property predictions, enabling the rapid screening of a vast solution space for materials exhibiting exceptional mechanical properties, such as ultra-incompressibility, and facilitates the synthesis of novel materials like MoWC2 and ReWB via in-situ reactive spark plasma sintering.
Implementation Method 1
a first machine learning model trained to generate, based at least on a first crystal structure, an equilibrium crystal structure corresponding the first crystal structure by at least iteratively searching a solution space including a plurality of possible variations of the first crystal structure for a variation of the first crystal structure having a minimum formation energy
Implementation Method 2
determining, based at least on the equilibrium crystal structure, one or more properties of the first crystal structure
Implementation Method 3
facilitates the synthesis of novel materials like MoWC2 and ReWB via in-situ reactive spark plasma sintering
Data Source
AI summary
A method for machine learning enabled material design may include applying a first machine learning model trained to generate an equilibrium crystal structure corresponding a crystal structure generated, for example, by performing an elemental substation. The first machine learning model may generate the equilibrium crystal structure by iteratively searching a solution space including possible variations of the crystal structure for a variation having a minimum formation energy. The searching may be constrained to variations having a same symmetry as the crystal structure. Properties of the crystal structure may be determined, for example, by applying a second machine learning to the equilibrium crystal structure. The crystal structure may be identified as a candidate for synthesis based the properties of the crystal structure, such as an above-threshold elastic modulus corresponding to an ultra-incompressibility. Various materials identified using this method and related systems and computer program products are also provided.