Matter structure description method applicable to machine learning potential energy surface construction

Abstract

The invention belongs to the technical field of computational chemistry and physics, and particularly discloses a matter structure description method applicable to machine learning potential energy surface construction. A series of characteristic functions can be constructed by the aid of input atomic coordinates and can be used as input information, and global potential energy surface data of matter systems can be trained by the aid of machine learning processes so as to obtain machine learning potential energy surfaces. Inter-atomic bond lengths and bond angles are used as basic variables ofthe characteristic functions, and power functions, truncation functions, spherical harmonic functions and triangle functions are combined with one another to construct atomic ambient environments with structure information of bond making, coordination and the like; the global potential energy surface data of the matter systems come from quantum-mechanical computation and contain large quantitiesof energy, force and stress information of different matter structures. The matter structure description method has the advantages that the characteristic functions have characteristics of coordinaterotation invariance, atomic exchange invariance, first-order and second-order derivative continuity and the like, and accordingly the matter structure description method is applicable to complicated multi-element matter systems; high-dimensional machine learning potential energy surfaces obtained by means of training on the basis of the characteristic functions can be used for material structure search, reaction mechanism prediction research and the like.

Description

technical field [0001] The invention belongs to the technical field of computational chemistry and physics, and specifically relates to a material structure description method suitable for machine learning high-precision potential energy surface construction, which is used to numerically distinguish the surrounding environment of atoms in a material system. Background technique [0002] Material structure prediction and chemical reaction path search are the core tasks of contemporary physical and chemical computational simulation research, and play an important role in understanding and predicting the thermodynamic and kinetic properties of materials. Due to the complexity of the material system, currently feasible computational simulation schemes generally rely on electronic structure calculations based on quantum mechanics to establish the relationship between the geometric structure and energy of the system, that is, the potential energy surface. However, the potential fu...

AI Technical Summary

Problems solved by technology

This method overcomes the shortcomings of insufficient accuracy of the previous empirical potential function and the low efficiency of the potential energy surface of quantum mechanics. It has strong portability and versatility, and can achieve good prediction ability for different material systems.

Images