Data for "Hybrid physics-constrained learning of density functionals to bridge particle and continuum scales"

Molecular dynamics reference data accompanying the article “Hybrid physics-constrained learning of density functionals to bridge particle and continuum scales” (E. Monti, P. Yatsyshin, K. Gkagkas, A. B. Duncan, Communications Materials, 2026). All simulations are of the Lennard-Jones truncated and shifted fluid (r_c = 2.5σ) in the NVT ensemble with LAMMPS (22 Jul 2025 - Update 3); all quantities are in reduced Lennard-Jones units. planar/ — wall adsorption in a 60×20×20 box. binned_density/<wall>/<N>/<T>/: time-averaged wall-normal density profiles \rho(x) with standard errors (CSV), for the LJ 9–3 training wall Vext1 (ε_w = 1.2, σ_w = 1.2; N = 8000), the LJ 9–3 validation wall Vext2 (ε_w = 1, σ_w = 2; N = 4000–8000), the Morse wall morse and the repulsive harmonic wall with uniform body force harmonic (N = 8000), at T = 0.55–1.00. The five Vext1profiles at T = 0.55, 0.65, 0.75, 0.85, 0.95 are the training set of the paper; all others are validation data (Figs. 2, 8; Supplementary Fig. S3). lammps_trajectories/<wall>/<N>/<T>/: production-run LAMMPS dump files and random seeds from which the profiles were binned (N = 8000 only). drop/ — sessile nanodroplets on the two LJ 9–3 walls, in a 120×120×60 box. binned_density/<wall>_<N>/: time-averaged density fields ρ(r, z) binned in cylindrical coordinates (CSV), for Vext1_N48k and Vext2_N48k (T = 0.55–0.90) and Vext2_N24k (T = 0.55–0.85), with the spherical-cap fits and their uncertainties (Figs. 4–6; Supplementary Fig. S9). lammps_trajectories/<wall>_<N>/<T>/: production-run dump files, restarts, seeds, logs and the binning and surface-extraction outputs used to produce the density fields and contact angles. the simulation parameters are given in Methods and Supplementary Sections S3 and S5 of the article. The code to train the functional, solve the Euler–Lagrange equations, compute coexistence and surface tensions, together with the parameters of the trained functional and of the sixteen ablation models, is in the GitHub repository https://github.com/classical-DFT-lab/code-cdft-manuscript-2026.