Surface Energy Manuscript

This repository contains various scripts/notebooks we used to create the results in our paper.

Table of Contents

• Prerequisites
• Usage
  • Cleavage energy dataset
  • Train a CGCNN model

Prerequisites

• Crystal Graph Convolutional Neural Networks (CGCNN)

• Additional packages required for cgcnn environmnet:

• PyTorch
• scikit-learn
• pymatgen
• AdamW

Usage

Cleavage energy dataset

Located in the cleavage_energy_dataset folder.

We have included a pickel file that contain our cleavage energy data, along with a Jupyter notebook (read_data.ipynb).

Train a CGCNN model

Located in the train_CGCNN_model folder.

We have included the cgcnn we used, and random_assignment method. random_assignment_method.ipynb notebook splits the data randomly into 8:2 training: test set, and uses CGCNN to train a model. You need to clone the CGCNN repository and install all the prerequisite packages in order to run these notebooks

We have included the optimized paramters we used for CGCNN: "atom_fea_len": 43,
"batch_size": 87,
"step": "0.1",
"epochs": 218,
"h_fea_len": 114,
"log_learning_rate": -6.465085550816676,
"n_conv": 8,
"n_h": 3,
"max_num_nbr": 12,
"optimizer": "AdamW"