Reaction Path Search Using Trained Models

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current methods for reaction path search, such as the ADDF and AFIR methods, face high calculation costs and limitations in molecule size and number due to the need for extensive first-principles calculations, making them inefficient for classical force fields and prone to search failures.

Innovation Solution

An information processing device utilizing trained models to search for reaction paths by outputting physical quantities like energy and Hessian matrices, reducing the need for frequent high-cost calculations through Hessian updates and artificial force-induced reactions, allowing for faster and more accurate searches.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If first-principles calculation is used to calculate energy, force, or Hessian while moving atomic nuclei, then calculation accuracy is improved, but calculation cost increases enormously

Engineering Contradiction:

Improvecalculation accuracyVSAvoidcalculation cost

Core Design Contradiction:

Measurement precisionVSUse of energy by stationary object

Solution Approach 1:

The patent segments the calculation process by using trained models for routine energy and force calculations, reserving first-principles calculations only for verification and critical steps, thus dividing the computational workload between fast approximations and accurate but expensive methods

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates trained models that copy the behavior of first-principles calculations, allowing the system to use these trained copies for most calculations while maintaining the option to verify with actual first-principles calculations when needed

Inventive Principle:

Principle #26Copying

2Extent of automation

If ADDF method is used to search for reaction paths, then automated search capability is improved, but calculation cost increases due to anharmonic downward distortion calculation

Engineering Contradiction:

Improveautomated search capabilityVSAvoidcalculation cost

Core Design Contradiction:

Extent of automationVSUse of energy by stationary object

Solution Approach 1:

The patent uses trained models that copy the computational results of expensive anharmonic calculations, enabling automated reaction path search without repeatedly performing the costly anharmonic downward distortion calculations

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent performs preliminary training of models using a limited set of first-principles calculations before the actual reaction path search, so that during the automated search phase, the pre-trained models can provide fast predictions without requiring additional expensive calculations

Inventive Principle:

Principle #10Preliminary action

3Adaptability or versatility

If AFIR method is used for reaction path search, then search coverage is improved, but difficulty in parameter selection and accuracy verification increases

Engineering Contradiction:

Improvesearch coverageVSAvoidparameter selection complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements feedback mechanisms where the trained model predictions are continuously verified against first-principles calculations at key points, allowing the system to self-correct and maintain accuracy while exploring diverse reaction paths

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system uses the trained models to automatically guide the reaction path search process, reducing the need for manual parameter tuning and expert intervention while maintaining comprehensive search coverage

Inventive Principle:

Principle #25Self-service

4Productivity

If high-speed methods are used for reaction path search, then productivity is improved, but search failure risk increases

Engineering Contradiction:

Improvesearch speedVSAvoidsearch success rate

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent prepares trained models in advance that are robust to various molecular configurations, cushioning against potential search failures by having pre-learned knowledge ready to handle diverse reaction scenarios

Inventive Principle:

Principle #11Beforehand cushioning (Prior cushioning)

Solution Approach 2:

The system continuously monitors search progress and compares model predictions with first-principles calculations, providing feedback that allows real-time correction of potentially failed search trajectories while maintaining high overall productivity

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS20250021705A1Information processing device, information processing method, and non-transitory computer readable medium

Publication Date: 2025.01.16 PREFERRED NETWORKS INC

AI summary

An information processing device includes one or more memories; and one or more processors. The one or more processors are configured to search for a reaction path by using one or more trained models that, when receiving an input of a three-dimensional arrangement of two or more atoms forming a molecule, output a physical quantity regarding the molecule.