Double-Tier ML Interatomic Simulation for Boundary Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning interatomic potential (MLIP) methods face challenges with scalability, long-range effects, and discontinuities in potential energy surfaces, as well as difficulties in capturing complex boundary conditions, leading to high computational costs and inaccuracies in molecular dynamics simulations.
Innovation Solution
A double tier machine learning approach is employed, using a high-fidelity MLIP in a central region and a lower-fidelity MLIP in a surrounding region, with interpolation between them, to dynamically evolve atomic interactions and simulate complex boundary conditions efficiently, while maintaining accuracy and reducing computational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ab-initio calculations (DFT) are used to obtain accurate estimations of interaction energies, then measurement precision is improved, but computational cost increases significantly
Solution Approach 1:
The simulation system is segmented into multiple spatial regions with different fidelity levels. A central high-fidelity region uses ab-initio calculations for accurate interaction energy estimation, while surrounding low-fidelity regions use computationally cheaper models. This segmentation allows the system to achieve high accuracy where needed without the prohibitive computational cost of applying ab-initio methods throughout the entire system.
Solution Approach 2:
Different regions of the simulation system are assigned different levels of computational fidelity according to local requirements. The central region maintains high quality ab-initio calculations for accurate physical state determination, while peripheral regions use simplified models. This local differentiation optimizes the balance between measurement precision and computational energy consumption.
2Measurement precision
If a single high-fidelity MLIP is used throughout the entire simulation region, then measurement precision is improved, but device complexity and computational cost increase
Solution Approach 1:
The simulation domain is divided into a central high-fidelity region and surrounding low-fidelity regions. This spatial segmentation allows the system to use computationally expensive high-fidelity MLIPs only where necessary for accurate physical state determination, while using simpler, less expensive models in peripheral regions, thereby reducing overall device complexity.
Solution Approach 2:
Instead of applying high-fidelity MLIPs to the entire simulation region (excessive action), the system applies them only to the central region where high measurement precision is required (partial action). This partial application of high-fidelity methods reduces computational resource consumption while maintaining accuracy where it matters most.
3Measurement precision
If ab-initio calculations are applied to large systems, then measurement precision is improved, but productivity decreases due to high computational cost
Solution Approach 1:
By segmenting the simulation system into high-fidelity and low-fidelity regions, the method enables accurate physical state determination in the central region while maintaining simulation efficiency in peripheral regions. This allows large systems to be simulated with acceptable productivity by avoiding the prohibitive computational cost of applying ab-initio methods to all atoms throughout the entire system.
Solution Approach 2:
High measurement precision is maintained locally in the central region through ab-initio calculations, while computational efficiency is prioritized in peripheral regions through the use of simplified models. This local differentiation enables the system to achieve both accuracy and productivity by applying high computational resources only where physically necessary.
Data Source
AI summary
A machine learning simulation method of determining a physical state of interaction between atoms from one or more physical properties of the atoms is disclosed. The method including dynamically evolving a first subset of atoms via a first machine learning model within a central high-fidelity region based on the one or more physical properties of the atoms. The method further includes dynamically evolving a second subset of the atoms via a second machine learning model with a remaining low-fidelity region based on the one or more physical properties of the atoms. The method also includes dynamically evolving a third subset of atoms located between the central high-fidelity region and the remaining low-fidelity region based on an interpolation of the first and second machine learning models to determine the physical state between the atoms.