Our approach outperforms traditional theories, captures non-local structure, scales beyond training & maintains thermodynamic consistency. It makes cDFT practical for electrolytes and pushes it toward realistic chemical systems—like salts & water.
#multiscalemodelling #AI #machinelearning #physics

Accurately modeling ionic fluids remains a century-old challenge due to competing long-range Coulomb forces & short-range sterics, beyond Debye-Hückel or Poisson-Boltzmann. We combine liquid state theory & a ML-based framework for simple liquids to learn free energy functionals for electrolytes.

Our findings establish "dielectrocapillarity'' -- the use of electric field gradients to control confined fluids -- as a powerful tool for controlling volumetric capacity in nanopores.

Congrats to @annatbui.bsky.social & check out her talk at APS @apsphysics.bsky.social next week!

In the application paper (arxiv.org/abs/2503.09855), we employ the theory with deep learning methods, to demonstrate, from first principles, that dielectrophoretic coupling enables tunable control over the liquid-gas phase transition, capillary condensation, and fluid uptake into porous media.

where we treat the charge density as an observable of the system, with the intrinsic Helmholtz free energy functional dependent upon both density and electrostatic potential. Expressions for the coupling between number and charge densities emerge naturally in this formalism.

macroscopic bodies, their use is questionable when relevant length scales become comparable to a system's natural correlation lengths, as commonly occurs in, e.g., biological systems, nanopores, and microfluidics. Our approach is based on the recently proposed hyperdensity functional theory,

In the theory paper (arxiv.org/abs/2503.09768), we present a first principles theory for electromechanics in fluids. Electromechanical phenomena describes the response of the number density to electric fields. While continuum theories are successful in describing electromechanics in