Application Of Supervised Learning In Condensed Matter Physics

With the continuous progress of computing science,machine learning has been widely applied in the field of condensed matter physics.Machine learning is similar to human learning.Human learning is through learning knowledge.When humans come into contact with new things,they can make judgments and solutions based on the knowledge they have learned.Machine learning is when a machine learns from an input sample to produce a "black box".When some new samples are input into the "black box",the machine can make judgments based on the learned "black box".The learning ability of the machine is also related to these training samples.The quality of the training samples determines the generalization ability of the machine to new samples.This is also similar to human learning.The quality of knowledge often determines whether people can be better at new things.In the introduction,we introduce the research background,research significance and research status of supervised learning in machine learning in condensed matter physics.In the first part of our work,we give an application of machine learning in the field of materials science:we use a fully connected neural network to predict the conversion efficiency limit of tandem solar cells,and find out through the known results of high-throughput calculations The conversion efficiency limit of the specific material,the input result of this network only depends on the band gap of the top and bottom cell materials of the tandem solar cell.We first use the combination of part of the top and bottom battery band gaps as the network training data;the conversion efficiency limit calculated based on the diode equation under the AM 1.5 standard is used as the known answer to train the neural network.Through the training of the neural network,we can successfully predict the conversion efficiency limit of more band gap combinations.This method can also be used as the back end of the high-throughput material design network,which can effectively help find more efficient and cheap materials.In the second part of our work,we introduced the application of machine learning in the phase transition of matter:we used the Monte Carlo method to generate three types of samples under different system sizes:1)percolation samples under periodic boundary conditions;2)Samples that percolate under non-periodic boundary conditions but not percolation under periodic boundary conditions;3)Samples that do not percolate under non-periodic boundary conditions.We use these samples to train the convolutional neural network,and the trained convolutional neural network can accurately identify the samples in these three situations.According to the percolation probability of different samples,the phase transition point of percolation can be determined by the prediction result of the convolutional network.At the same time,we also defined the percolation probability in different directions.The prediction of these percolation probabilities through the convolutional network can give the phase transition point of percolation in a self-consistent manner.This allows us to use machine learning to study other physical phase transition problems and it provides good ideas and directions.Finally,we give the final summary and outlook of the paper.