Recognition Hexatic Phase In Complex Plasmas Using Convolutional Neural Networks

Hexatic phase is a special phase with properties between solid and liquid in two-dimensional systems,which has important implications for studying phase transition processes and properties.However,the hexatic phase has not been observed in the complex two-dimensional plasma phase transition process.This work aims to discover the hexatic phase in the two-dimensional complex plasma phase transition process using molecular dynamics simulation and convolutional neural network models for identification.The temperature range was distinguished through clustering analysis and gyration tensor parameter statistics.Finally,the hexatic phase was successfully discovered.First,molecular dynamics simulation was used to obtain images of particle positions at different temperatures,which were preprocessed with Gaussian filtering,random cropping,scaling,rotation,and normalization,and used to train a binary classification model for solid and liquid phases.The Res Net18 convolutional neural network model was trained,which can completely distinguish between the solid and liquid phases.Secondly,the trained binary classification model was applied to the heating process of two-dimensional complex plasma,and 512-dimensional semantic features were extracted.Clustering analysis methods such as k-means,birch,Mini-Batch K-Means,and Gaussian clustering were used to perform multiple clustering on the heating images.The silhouette coefficient and CH value were considered to evaluate the clustering model,and k-means clustering was chosen to obtain the approximate temperature range of the hexatic phase.Then,the statistical of gyration tensor parameter and the statistical of position tensor parameter were calculated to further distinguish the solid,liquid,and hexatic phases within the temperature range obtained from clustering.Finally,molecular dynamics was used to collect images of the hexatic phase in the temperature range,which were used to train a three-classification model for solid,liquid,and hexatic phases.Similarly,the Res Net18 convolutional neural network model was trained,which can effectively identify the hexatic phase.The three-classification model was used for experimental data classification,and the hexatic phase was successfully discovered in the experimental data.The approximate classification results were obtained by calculating statistical of gyration tensor parameter,which were consistent with the model identification results.This study successfully identified the hexatic phase through the combination of molecular dynamics simulation and deep learning algorithms,providing new ideas and methods for phase transition research and having certain significance for understanding the phase transition process in two-dimensional complex plasma systems.