Vacancy diffusion barrier dataset for NNK model

This data repository contains the data for training the neural network model in the following paper. Xing, B., Rupert, T. J., Pan, X., & Cao, P. (2024). Neural network kinetics for exploring diffusion multiplicity and chemical ordering in compositionally complex materials. Nature Communications, 15(1), 3879. The repository has the following directories: material_models:This directory consists 46 DUMP format files representing 46 distinct atomistic models. The compositions are listed in the text file "dirnames.txt". These compositions are used for preparing dataset for training ml models including on-lattice representation of local atomistic configuration and corresponding diffusion barrier from NEB calculation. The "data" folder contains on-lattice representations and the "labels" folder contains the diffusion barriers. data:This directory consists of 46 NPY format files, each of which corresponds to the on-lattice representations of local atomistic configurations from one composition. Each NPY file contains a four-dimensional matrix which serves as the input to ml models. The matrix dimension is 16000x9x9x9, containing 16000 on-lattice representations of distinct local atomistic configurations, each of which has the dimension of 9x9x9. The file name takes the format as *.npy, where the asterisk denotes the index of composition. The compositions are listed in the text file named as "dirnames.txt". labels:This directory consists of 46 NPY format files, each of which corresponds to the diffusion barriers from one composition. The 46 compositions are listed in the text file named as "dirnames.txt". Each NPY file contains a vector with 16,000 barriers corresponding to 16,000 on-lattice representations from the "data" folder. For example, "0.npy" in the "labels" folder consists the labels (diffusion barriers) corresponding to on-lattice representations from "0.npy" in the "data" folder. scripts:This directory consists of scripts for generating on-lattice representations of local atomistic configurations.本数据集仓库包含用于训练下述论文中神经网络模型的相关数据。 论文信息：Xing, B.、Rupert, T. J.、Pan, X. 与 Cao, P. 于2024年发表于《自然·通讯》（Nature Communications）第15卷第1期第3879页的论文《用于探索成分复杂材料中扩散多重性与化学有序性的神经网络动力学》（原英文标题：Neural network kinetics for exploring diffusion multiplicity and chemical ordering in compositionally complex materials）。 本仓库包含以下目录： material_models：该目录包含46个DUMP格式文件，分别对应46种独立的原子尺度模型。成分信息已在文本文件"dirnames.txt"中列出。上述成分被用于构建机器学习模型的训练数据集，包括局部原子构型的格上表示（on-lattice representation）与来自NEB计算（Nudged Elastic Band）的对应扩散势垒。其中，"data"文件夹存储格上表示数据，"labels"文件夹存储扩散势垒标签数据。 data：该目录包含46个NPY格式文件，每个文件对应一种成分的局部原子构型格上表示。每个NPY文件均为一个四维矩阵，作为机器学习模型的输入数据。该矩阵维度为16000×9×9×9，包含16000组不同局部原子构型的格上表示，每组格上表示的维度为9×9×9。文件名采用"*.npy"格式，其中星号代表成分的索引。成分信息已在文本文件"dirnames.txt"中列出。 labels：该目录包含46个NPY格式文件，每个文件对应一种成分的扩散势垒数据。46种成分的信息已在文本文件"dirnames.txt"中列出。每个NPY文件均为一个包含16000个势垒的向量，与"data"文件夹中16000组格上表示一一对应。例如，"labels"文件夹中的"0.npy"即为"data"文件夹中"0.npy"对应的格上表示的扩散势垒标签。 scripts：该目录包含用于生成局部原子构型格上表示的脚本文件。