Computational chemistry has reached a critical point; the combination of pre-exascale amounts of GPU compute, chemically accurate interaction potentials, and effective accelerated sampling techniques has given it the ability to provide near-exact solutions to the equations of quantum and statistical mechanics.
In this project, the goal is to push this ability to its limits and characterize the phase diagram of ice across a large range of thermodynamic conditions, including negative pressures and ultra-low temperatures. The physics of ice polymorphs constitutes a long-standing challenge due to the electronic complexity of hydrogen-bonded networks, intrinsic and pervasive proton disorder, and large nuclear quantum effects.
The project team proposes the first computational workflow which can fully accommodate all this complexity in a near-exact manner, both in terms of the description of the interatomic interactions as well as the wavefunction character of the nuclei and all entropic contributions to the free energy.
To achieve this, the research team relies on a combination of transfer-learned equivariant interaction potentials, accurate post-HF calculations, and a variety of accelerated sampling techniques. Upon completion, the results of this project can guide experimental research towards new ice polymorphs, and set a gold standard for phase stability calculations which are broadly applicable within materials science.
Ghent University, Belgium.