Catalyst Screening Using Descriptor Maps and Two-Stage Energy Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for selecting catalysts are inefficient and time-consuming, particularly when using first principles calculation based on density functional theory (DFT), and prediction methods lack accuracy, making it difficult to find catalysts with good production efficiency for target products.
Innovation Solution
A method involving the selection of catalysts using descriptors such as intermediate and transition state structures, creating maps to represent reactivity, and performing calculations with machine learning potentials to efficiently screen candidate substances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If first principles calculation based on DFT is used to calculate the energy required for synthesizing the target product, then the calculation accuracy is improved, but the calculation time increases significantly making it difficult to apply to large numbers of catalysts
Solution Approach 1:
The patent segments the catalyst evaluation process into two distinct stages: (1) a first screening stage using machine learning models that provide rapid but less accurate energy predictions, and (2) a second verification stage using DFT-based first principles calculations that provide high accuracy but are computationally expensive. This segmentation allows the system to efficiently evaluate large numbers of catalysts by applying the appropriate level of computational rigor to each candidate.
Solution Approach 2:
The patent applies preliminary action by performing machine learning-based energy predictions on all candidate catalysts before subjecting them to time-consuming DFT calculations. The machine learning model serves as a preliminary filter that identifies promising catalysts based on their structural and compositional features, thereby preparing a reduced set of candidates for the subsequent high-accuracy DFT verification stage.
2Speed
If a general regression model using substance parameters as descriptors is used to predict the energy required for synthesizing the target product, then the calculation speed is improved, but the prediction accuracy is limited and energy prediction performance is insufficient
Solution Approach 1:
The patent merges two different prediction approaches into a unified two-stage evaluation system. The first stage uses a machine learning regression model that leverages substance parameters and descriptors to provide rapid energy predictions. The second stage combines these results with DFT-based first principles calculations to verify and refine the predictions. This merging allows the system to achieve both high calculation speed and high prediction accuracy by strategically combining the strengths of both methods.
Solution Approach 2:
The patent introduces an intermediary verification mechanism where machine learning predictions serve as an intermediate step between initial catalyst screening and final high-accuracy DFT evaluation. The machine learning model acts as an intermediary that quickly narrows down the candidate pool, making the subsequent DFT calculations more efficient and effective.
3Loss of information
If transition state structures are calculated for various catalysts using DFT-based first principles calculation, then the catalytic reaction mechanism understanding is improved, but the calculation time becomes prohibitively long making it virtually impossible to calculate for thousands of catalysts
Solution Approach 1:
The patent segments the transition state analysis into a two-tier approach: (1) a preliminary screening tier using machine learning models that can rapidly estimate transition state energies based on catalyst descriptors, and (2) a detailed verification tier using DFT-based calculations that provide comprehensive mechanistic understanding but are computationally intensive. This segmentation enables the system to evaluate transition states for many more catalysts than would be possible with DFT alone.
Solution Approach 2:
The patent applies preliminary action by using machine learning models to predict transition state energies and identify promising catalysts before performing detailed DFT-based transition state calculations. This preliminary assessment based on catalyst descriptors and structural features allows the system to focus computational resources on the most promising candidates, thereby maintaining mechanistic understanding while dramatically reducing the overall calculation time required.
Data Source
AI summary
A method for selecting a catalyst includes selecting, as a descriptor, an energy of an intermediate structure or a transition state structure included in an elementary reaction of a catalytic reaction for producing a target product from raw materials, creating a map representing a relationship between the descriptor and a reactivity of the catalyst, calculating the descriptor related to the catalytic reaction using candidate substances in a state where the candidate substances are fixed, creating a first plot map using the descriptor, selecting first screened candidate substances from the candidate substances based on the first plot map, calculating the descriptor for the catalytic reaction using the first screened candidate substances in a state where a surface of the first screened candidate substances is relaxed, creating a second plot map using the calculated descriptor, and selecting second screened candidate substances based on the second plot map.