top of page

Github repo for the atom-motif dual network (AMDNet)

As inspired by Pauling’s rule, we created a motif-centric learning framework by combining motif information with the atom-based graph neural networks. The atom-motif dual network (AMDNet) outperforms the state-of-the-art models in predicting the electronic structures of metal oxides. This work illustrates the route toward the fundamental design of machine learning architecture for complex materials by incorporating beyond-atom physical principles. 

The github repo link for AMDNet is here.

For the citation to the code, please refer to the following articles:


H. R. Banjade, S. Hauri, S. Zhang, F. Ricci, W. Gong, G. Hautier, S. Vucetic, Q. Yan, “Structure motif centric learning framework for inorganic crystalline systems”, Sci. Adv., 7, eabf1754 (2021)

Github repo for the contrastive learning ​framework for density functional (DFCL)


Incorporation of physical constraints into machine learning density functional design is crucial to generalize neural network (NN) based exchange correlation (XC) functionals. In the enclosed manuscript, we demonstrate that contrastive learning is a computationally efficient and flexible method to incorporate physical constraints in NN based density functional design. We propose a density functional contrastive learning framework to incorporate the uniform density scaling property of electron density for exchange energies, by pretraining on 10,000 molecules in the QM9 database and transferring to a downstream task to predict exchange energies from volumetric electron densities. The work demonstrates that contrastive learning may serve as an adaptive and effective method to incorporate physical constraints of density function theory into the ML model design process. 

The github repo link for DFCL is here.

For the citation to the code, please refer to the following articles:

W. Gong, T. Sun, H. Bai, S. Chowdhury, P. Chu, A. Aryal, J. Yu, H. Ling, J. P. Perdew, Q. Yan, "Incorporation of density scaling constraint in density functional design via contrastive representation learning", arXiv:2205.15071 (2022) 

2D materials database with symmetry-incorporated electronic structure information

We are building a 2D materials database based on ~2000 nonmetallic 2D materials extracted from the C2DB database, calculated using the HSE06 functional. What is unique about the datahbase is that all the symmetry information (irreducible representations) of electronic states have been extracted and stored in the database. This set of material information will be essential for quantum material discovery and design.

The link to the database is here

bottom of page