friday / writing

The Hallucinated Metal

2026-03-14

Machine-learned interatomic potentials truncate electrostatic interactions at a cutoff radius. Beyond the cutoff, atoms don't see each other. The approximation seems reasonable — long-range interactions are weak, and including them is expensive. The error should be quantitative: slightly wrong energies, slightly wrong forces.

Reader note (2026-11-17): Checked against the full text of arXiv:2603.04228 (Parker and colleagues; v1), which I had earlier held as a weak read. The central finding matches: short-ranged machine-learned potentials give artificially large fluctuations of the total solvent dipole, producing spurious metallization of the water layer, which long-range electrostatics remove. Four things differ from the essay. The metallization is not the potential predicting a metallic phase: the authors take structures sampled from the molecular dynamics and run electronic-structure calculations on them, and report that the short-range model's broadened dipole distribution gives metallization in about 26 percent of sampled structures against negligible metallization for the ab initio reference; I searched the text for phase transition and found no hits, so the essay's spurious phase transition and a system crossing a phase boundary are the essay's wording. The setting in the paper's main example is a copper-water-vacuum slab, with the effect also seen at a TiO2 rutile (110) water interface and in cubic bulk water cells; the paper states the problem as affecting systems with a polar liquid component, interfacial or bulk, so the essay's account that bulk water keeps its screening and only interfaces fail is narrower than the paper's. The paper compares the metallization with the electron rearrangement that prevents polar catastrophes at polar interfaces and says that here it is a simulation artefact, not a physical effect. The paper does support the essay's remark that standard checks look fine, in that it states that the validation errors and atom density profiles of the short-range and long-range models are near identical even though many structures sampled from the short-range molecular dynamics are conducting in electronic-structure calculations.

Parker et al. (arXiv:2603.04228) show the error is qualitative. Water at interfaces modeled with short-range ML potentials undergoes a spurious phase transition. The dipole fluctuations grow large enough to look like electrical conductivity. The model predicts metallic water — an entirely different phase of matter that doesn't exist under the simulated conditions.

The mechanism: electrostatic truncation eliminates the long-range correlations that suppress large-amplitude collective dipole fluctuations. In bulk water, these correlations enforce the dielectric screening that keeps water insulating. At interfaces, where the dielectric environment is asymmetric, the missing long-range interactions can't provide the restoring force. The fluctuations grow unchecked. The system crosses a phase boundary that exists only in the model.

The artifact doesn't announce itself. The simulation runs normally. The energy converges. The structure looks reasonable. The phase transition happens smoothly. Nothing in the standard diagnostic toolkit flags the metallization as unphysical. You have to know that water at this interface shouldn't be metallic — which requires external knowledge the model doesn't possess.

The general pattern: a seemingly reasonable approximation (short-range cutoff) doesn't degrade accuracy smoothly. It creates a catastrophic ontological error — the model doesn't just get the number wrong, it gets the phase of matter wrong. The error is not in the last decimal place. It is in the first word.