Daniel Hedman^1,2*, Tom Rothe^3*, Gustav Johansson^2*, Fredrik Sandin^4*, J. Andreas Larsson^2*, Yoshiyuki Miyamoto^1*

1 Research Center for Computational Design of Advanced Functional Materials, National Institute of Advanced Industrial Science and Technology (AIST), Central 2, 1-1-1 Umezono, Tsukuba, Ibaraki, 305-8568, Japan

2 Applied Physics, Division of Materials Science, Department of Engineering Sciences and Mathematics, Luleå University of Technology, SE-971 87 Luleå, Sweden

3 Institute of Physics, Faculty of Natural Sciences, Chemnitz University of Technology, 09126 Chemnitz, Germany

4 Machine Learning, Embedded Intelligent Systems Lab, Department of Computer Science, Electrical and Space Engineering, Luleå University of Technology, SE-971 87 Luleå, Sweden

* Corresponding authors emails: daniel.hedman@ltu.se, tom.rothe@s2015.tu-chemnitz.de, gustav.johansson@ltu.se, fredrik.sandin@ltu.se, andreas.1.larsson@ltu.se, yoshi-miyamoto@aist.go.jp

DOI10.24435/materialscloud:6h-yj [version v1]

Publication date: Nov 11, 2020

How to cite this record

Daniel Hedman, Tom Rothe, Gustav Johansson, Fredrik Sandin, J. Andreas Larsson, Yoshiyuki Miyamoto, CA-9, a dataset of carbon allotropes for training and testing of neural network potentials, Materials Cloud Archive 2020.144 (2020), doi: 10.24435/materialscloud:6h-yj.

Description

The use of machine learning to accelerate computer simulations is on the rise. In atomistic simulations, the use of machine learning interatomic potentials (ML-IAPs) can significantly reduce computational costs while maintaining accuracy close to that of ab initio methods. To achieve this, ML-IAPs are trained on large datasets of images, meaning atomistic configurations labeled with data from ab initio calculations. Focusing on carbon, we have created a dataset, CA-9, consisting of 48000 images labeled with energies, forces and stress tensors obtained via ab initio molecular dynamics (AIMD). We use deep learning to train state-of-the-art neural network potentials (NNPs), a form of ML-IAP, on the CA-9 dataset and investigate how training and validation data can affect the performance of the NNPs. Our results show that image generation with AIMD causes a high degree of similarity between the generated images, which has a detrimental effect on the NNPs. However, by carefully choosing which images from the dataset are included in the training and validation data, this effect can be mitigated. We end by benchmarking our trained NNPs in real-world applications and show we can reproduce results from ab initio calculations with an accuracy higher than previously published ML- or classic IAPs.

Materials Cloud sections using this data

Files

File name	Size	Description
Readme.txt MD5md5:ef25015cf52d749a74d34aa4580c61be	2.4 KiB	Readme file
scripts.zip MD5md5:47504db1933a414ff4277da31c05e29e	3.4 KiB	Python scripts used to read data from VASP and train neural network potentials
datasets.zip MD5md5:a3739f0fd8d107195afb9d975a804f65	453.4 MiB	Datasets for training and testing of neural network potentials
NNPs.zip MD5md5:efb81a67ccd0171c9009e32c86986238	13.1 MiB	The best trained neural network potentials for each dataset

License

Files and data are licensed under the terms of the following license: Creative Commons Attribution 4.0 International.
Metadata, except for email addresses, are licensed under the Creative Commons Attribution Share-Alike 4.0 International license.

External references

Keywords

CA-9 Dataset Machine learning Interatomic potential Carbon Neural network potential