Katie's paper on selective adsorption live on arXiv

First and foremost — well done to Katie for writing her first paper!

In this paper, we show how the hyper‑DFT approach to mean‑field theory (see our previous work) can be used for a “train once, learn many” strategy, greatly simplifying the machine‑learning component of neural cDFT. In particular, we demonstrate how the properties of a binary fluid can be investigated by learning the free‑energy functional only for a repulsive single‑component reference system.

But this paper is more than a methodological development. We find that the presence of an azeotrope in the fluid’s phase diagram influences adsorption under thermodynamic conditions far from liquid–vapour coexistence. We also show that adsorption is the thermodynamic driving force for pore selectivity, and that a completely unselective state corresponds to an extremum in the surface tension.