Researchers at Google subsidiary DeepMind have open-sourced DM21, a neural network model for mapping electron density to chemical interaction energy, a key component of quantum mechanical simulation. DM21 is also available as an extension to the PySCF simulation framework, according to InfoQ. In a blog post, the DeepMind team shared:
“By expressing the functional as a neural network and incorporating these exact properties into the training data, we learn functionals free from important systematic errors — resulting in a better description of a broad class of chemical reactions.”
The team specifically addresses two long-standing problems with traditional functionals. That are the delocalization error and the spin symmetry breaking.
The company has used a neural network to represent the functional and tailoring its training dataset. The idea is to capture the fractional electron behavior expected for the exact functional. They have also found that they could solve the problems of delocalization and spin symmetry-breaking.
The function also showed itself to be accurate on broad, large-scale benchmarks. That is suggesting that this data-driven approach can capture aspects of the exact functional that have thus far been elusive, according to publication in DeepMind’s blog.