First-hand information for everyone
A new neural classical density functional theory (cDFT) framework can compute CO₂ adsorption curves for all pressures from a single calculation, eliminating the need for expensive grand‑canonical simulations. The method uses machine‑learned interatomic potentials (MLIPs) to generate accurate ab initio energy surfaces, which the neural network then feeds into a density‑functional model.
The authors applied the approach to more than 5,000 experimentally known MOFs. Four distinct isotherm shapes emerged, linked to pore size and binding site arrangement. These shapes, not just binding strength measured by the Henry constant, determine which MOFs are best for vacuum or temperature swing capture at low pressure versus pressure swing at higher pressure.
The chief advantage is speed: ordinary grand‑canonical simulations coupled with high‑level quantum calculations are prohibitively costly, so screening is usually limited to low‑pressure uptake. The neural cDFT bypasses that bottleneck, providing full loading curves efficiently.
While the approach shows great promise for high‑throughput screening, it relies on trained MLIPs, and its accuracy depends on the quality of those potentials.