Material molecule composition generation method based on countercurrent

Abstract

The invention discloses a material molecule composition generation method based on countercurrent. According to the method, the countercurrent model is introduced into the field of material molecules, and the purpose of learning bidirectional mapping between material molecule components and potential variables is achieved, so that the purpose of controlling the characteristics of the generated material molecule components by regulating and controlling potential representation of the material molecules is achieved. In order to generate a chemically effective material molecule composition, an original material molecule is first expressed as a unique corresponding matrix, then the corresponding molecule expression matrix is converted into a potential expression through a plurality of reversible linear transformations, and the potential expression can be close to a known priori distribution Pz-N(0, I); and during generation, only one potential representation needs to be randomly sampled from the Pz, and then the corresponding material molecule composition can be generated through inverse mapping. The model Matlflow is experimented on a universal material molecule data set oqmd, in one-time 340,000 times of random sampling generation, the model still achieves the unique rate of 80.1% and the novelty rate of 100%, and compared with the unique rate of 68.09% of the most advanced material molecule composition generation method at present, the result is greatly improved.

Description

technical field [0001] The invention belongs to the field of material molecular composition exploration, and especially focuses on exploring novel material molecular composition. Background technique [0002] Inorganic materials are indispensable in our lives. According to the statistics of related papers on materials science, in the past century, about 2×10 5 A variety of inorganic materials are used in all aspects of life, ranging from energy storage, life production, and medical care to various fields [1]. It can be seen that the exploration of new materials is an area that needs to be focused on in contemporary scientific exploration. [0003] However, given the breadth of elements that make up inorganic chemical molecules in the periodic table, the size of the entire material space is virtually infinite. For such a large material space, traditional search methods mainly rely on chemical intuition and empirical rules obtained from prior knowledge to design reasonable m...

AI Technical Summary

Problems solved by technology

However, the deep learning model used still has the problem that the model is difficult to train, the generated material molecules are unstable, or the generated material molecules are invalid and unreasonable.

Images