Molecule Simulator

RealQM, hard partition: every electron (or opposite-spin pair, 2hom) owns a domain; domains meet at free walls placed by density continuity. Heavy atoms are pseudo-atoms: inner shells frozen into a core of charge Zx with hard radius rc fitted to the sum of the ionisation energies of the electrons carried. Solver molecule.js (WebGPU), box 12 a0 (LiH 20), h = 0.1.
Judged by geometry. Bond lengths and shapes are the result; binding (atomisation) energies are shown as a diagnostic. Where an atom carries many valence electrons the binding carries a large offset set by the composition and nearly independent of geometry (CH4: 3.2 Eh, constant to 0.03 over R 2.05–2.45), which cancels at fixed composition and in processes that keep every atom's bonding.
Article: RealQM for Molecules: Validation (PDF) · requires WebGPU (Chrome/Edge 113+, Safari 17+) · a run takes 2–15 min

Best cases — run and check

Each row: the recorded molecule and reference energies, the binding they give, experiment. Run recomputes one piece live; the row then shows the live energy, its difference from the record, and the binding from live values where available. The page under the table is the live solver (its 3D boundary button shows the free domain walls).
Live run: idle

Homonuclear X2 — axisymmetric solver

Exact reduction for linear molecules (2D z,ρ grid), run from the command line: python3 tools/axisym_linear.py <preset> R=…. Only a σ pair is representable. R-scan 0.8–1.3 Rexp; lengths a0, energies Eh. Runs and densities: linear molecule viewer.

Linear molecules — all-2hom chains, axisymmetric solver

One 2hom σ pair per bond, a fixed plane at each heavy nucleus, the same pseudo-atoms as above (no refit per molecule). python3 tools/axisym_linear.py chain spec=H-C-N bonds=2.0135,2.1792. Re from a parabola through the 0.9/1.0/1.1 × Rexp points. LiH and BeH2 differ from the 3D runs above (BeH2 0.284 vs 0.248): quote only after a mesh ladder. NCCN and HC4H were too jagged at h 0.1 (plane offsets) to read.

Energy offset and the range

Bond lengths are in range: every X–H bond comes out a few per cent short (CH4 −4%, residual forces on H inward everywhere). Where a heavy atom carries its whole valence the binding is two to three times too large (CH4 1.40 vs 0.67, NH3 0.95 vs 0.47), by an amount attached to each heavy atom (ethane twice methane), so it cancels in processes that keep each atom's bonding. Shapes and free angles are not in range: CH4 has no preference for the tetrahedron over the square, and the H2O angle closes or opens depending on whether the lone pairs are frozen. Where lone pairs press on a bond from both sides the geometry changes too: O2 over-binds twofold with a short bond and F2 has no minimum. Correction 2026-11-14: above five domains molecule.js used a total-Poisson path whose electron repulsion is 2.5–4 Eh low in CH4; the earlier “3.2 Eh offset, bond +10%, tetrahedron preferred” came from it. Every live run here now uses the per-electron path, and the quoted energies were checked by an independent recomputation from the converged density (article, Solvers section).
Other pages: free molecule builder (proteins, dynamics) · earlier mol_fast simulator (archived) · gallery