Ab initio molecular dynamics, machine learning and complex liquids

Using liquid water as an example I will show that deep neural networks allow us to greatly accelerate ab-initio molecular dynamics simulations without loss of accuracy. The approach naturally leads to models with different levels of coarse-graining both for the electronic ground-state and the atomic structure information. The basic procedure was outlined in two recent papers [1,2].

[1] L. Zhang, J. Han, H. Wang, R. Car, W. E, Physical Review Letters 120, 14301 (2018)
[2] L. Zhang, J. Han, H. Wang, R. Car, W. E, The Journal of Chemical Physics 149, 034101 (2018)