Weighing the Double Layer: Full-Field Image Correlation of Surface Stress as a Test of Explicit-Solvent Electrode Models

Simulations that track every water molecule and ion at a charged metal surface now make sharp claims. Water molecules flip their hydrogens toward a negatively charged silver electrode. Chloride binds strongly to gold while fluoride barely does. These "explicit-solvent" simulations disagree with cheaper models that treat the liquid as a smooth dielectric. Their structural claims are hard to check directly, because almost no laboratory measurement sees the interfacial liquid's structure. This proposal checks them through a mechanical quantity instead. That quantity is the surface stress of the electrode, meaning the in-plane force per unit length that the surface layer exerts on the metal underneath. Surface stress changes with electrode potential, and it bends a thin coated plate by a measurable amount. We would compute surface stress as a function of potential from explicit-solvent molecular dynamics with a deformable metal slab. We would measure it with digital image correlation (DIC), a camera method that tracks a random speckle pattern between images to recover a full displacement field. Comparing the two tests whether explicit solvent predicts a mechanical response that continuum-solvent models miss.

The two directions meet through a single number that both sides can produce. The simulation side of the field almost always freezes the metal lattice. That choice makes surface stress invisible to it. The DIC side has worked in liquid electrolyte cells (Chemical Strain of Graphite-Based Anode during Lithiation and Delithiation at Various Temperatures) but has never been compared against a molecular model. Our contribution is to make the simulated slab deformable and compute surface stress versus potential. We then build a DIC pipeline whose output is that same quantity with an honest error bar, and we use a thermodynamic identity to cross-check both against a second, independent observable.

Core Claim

Our claim is that the potential dependence of surface stress on a gold electrode in sodium chloride versus sodium fluoride is controlled by interfacial solvent and ion structure. Specifically, it depends on water reorientation and on whether chloride adsorbs specifically. As a result, explicit-solvent simulations with a flexible slab should predict the measured slope and curvature of surface stress versus potential, and implicit-solvent or vacuum models should not. The claim can fail. If an implicit-solvent model with a strained slab matches the DIC measurement within error, then the mechanical response is set by the metal's electrons and not by the liquid, and the explicit-solvent claims lose this line of support.

The thermodynamics ties two measurements together. Let $\Omega(\varepsilon, E)$ be the interfacial grand potential per unit unstrained area, written as a function of in-plane elastic strain $\varepsilon$ and electrode potential $E$. The surface stress is then $f = \partial\Omega/\partial\varepsilon$ and the charge per reference area is $\hat q = -\partial\Omega/\partial E$. Equality of mixed second derivatives gives

$$\left(\frac{\partial f}{\partial E}\right)_{\varepsilon} = -\left(\frac{\partial \hat q}{\partial \varepsilon}\right)_{E}.$$

The left side is what DIC measures from plate curvature as the potential is swept. The right side is how the electrode charge changes when the plate is mechanically bent, which can be measured independently by electrochemistry. The simulation can compute both sides from the same trajectories. Agreement between the two experimental sides validates the measurement. Agreement of the simulation with both validates the model.

Research Program

Simulation. We would train an equivariant machine-learned interatomic potential, a neural network that predicts quantum-mechanical energies and forces at much lower cost. The training data would be constant-potential density functional theory (DFT) calculations of Au(111) in contact with liquid water containing Na⁺, Cl⁻, or F⁻. The recipe follows Electrochemical potential-driven water dynamics control CO2 electroreduction at the Ag/H2O interface, which supplies the method: grand-canonical DFT, a MACE potential, and enhanced sampling. That work kept the Ag slab rigid. We change that in two ways. First, the training set would deliberately include biaxially strained slabs and target the stress tensor, which bulk-crystal training sets such as Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models already fit but without surfaces or liquid. Second, the metal atoms would move during production runs.

Deriving effective electrode–ion interactions from free-energy profiles at electrochemical interfaces computed ion free-energy profiles at a rigid Au(111) surface with the UMA foundation potential. It found a −8.40 kcal/mol chloride adsorption minimum at 2.3 Å and a 100–200 mV shift in the potential of zero charge (PZC), the potential at which the surface carries no net charge. That result motivates our ion contrast and gives us a prior to reproduce before trusting the flexible model.

We would extract surface stress from the lateral components of the virial in slab cells held at several fixed lateral strains and potentials. To reduce the noise of fluctuating stress, we would run nanosecond trajectories and use an ensemble of potentials. Structure and energetics of liquid water–hydroxyl layers on Pt(111) reached roughly 0.025 eV precision with 10 ns runs of a four-model ensemble. Two known weaknesses constrain the design. The first is that common DFT functionals over-structure water; that study had to run at 400 K to keep the water liquid. We would report results at both 300 K and 400 K, and the gap between them gives a visible systematic error. The second is that finite-cutoff potentials miss long-range electrostatics. We would test this by comparing slabs of two thicknesses and adding a Poisson-solver correction for the lateral stress contribution.

Measurement. The specimen is a thin glass or silicon strip, gold-coated with Au(111) texture on its wetted face and speckle-patterned on the passivated back face. It is clamped at one end in a sealed electrochemical cell with an optical window. Two calibrated cameras perform stereo DIC to track the out-of-plane deflection field of the whole strip. Stereo DIC is used for large structures in Remote 3D Displacement Sensing for Large Structures with Stereo Digital Image Correlation, and we would adopt the automated exposure control of Automated Camera Exposure Control for Accuracy-Enhanced Stereo-Digital Image Correlation Measurement to handle the liquid's changing transmission.

The methodological piece on the DIC side builds on Complete mechanical regularization applied to digital image and volume correlation. That work regularizes correlation on a finite-element mesh by penalizing deviations from linear-elastic equilibrium. We would replace the bulk elastic operator with a Kirchhoff plate-bending model, whose deflection is driven by one unknown uniform surface-stress change plus a small set of nuisance modes. The nuisance modes are clamp rotation, thermal bowing, and an apparent-displacement term from refractive-index drift in the liquid. We would then solve for the surface stress directly from image intensities instead of differentiating noisy displacements twice. This is where full-field data earns its place: a uniform surface stress produces a specific curvature pattern, while the artifacts listed in the research file under fluid-phase degradation (convection, salt-gradient refraction, bubbles) produce different spatial signatures that can be separated. We restrict the potential window to the double-layer region, where gold evolves no gas, so bubble formation does not occur by design.

Independent check. In the same cell we would apply known static bends with a micrometer stage and measure impedance spectra at each bend. This gives the change of charge with strain, the right-hand side of the identity above.

Baselines

The naive reference point is a vacuum or frozen-ice-layer model. It is a strained Au(111) slab with excess charge and no liquid, or with one static water layer, and surface stress is computed from DFT versus charge and converted to potential through the computed capacitance. This is the approach of static strain studies such as Local-strain-induced CO2 adsorption geometries and electrochemical reduction pathway shift, which explained strain-driven selectivity on PdCu with static DFT and d-band arguments. It would be run on exactly the same slab sizes, functional, and strain grid as the explicit model. Beating it means predicting the measured surface-stress slope where it fails by more than the experimental error bar.

The strongest competitor is implicit-solvent constant-potential DFT on the same flexible slabs. It is cheap, it captures dielectric screening, and it may already explain most of the mechanical response, so it is the real threat. For fairness, it gets the same DFT settings, the same strain grid, and its own tuning budget: dielectric cavity parameters fitted to the experimental PZC, matching the one PZC calibration the explicit model receives. The explicit model wins only if it matches the curvature of the surface-stress versus potential curve and the NaCl-versus-NaF difference in the anodic region, and the implicit model misses one of these, with both judged against the same measurement uncertainty.

The ablation that isolates the key ingredient is the explicit-solvent model run with a frozen substrate, with stress taken only from water–metal forces on rigid atoms. It shows whether letting the lattice relax changes the predicted stress. A second ablation removes ions so that only water remains, and it tests whether specific chloride adsorption drives the predicted difference between salts.

On the measurement side, the baseline is conventional local subset DIC followed by finite-difference curvature. We also include a "single-point" reduction that averages deflection at the tip, equivalent to a classical beam-deflection readout. Both run on the same images. The plate-regularized estimator beats them if it has lower variance across repeated potential cycles and smaller bias on synthetic images with injected refraction and thermal artifacts.

No reported result was found on Lacuna for electrocapillary surface stress on Au(111) in any model or experiment. The closest reported numbers are the 100–200 mV UMA-derived PZC shift in NaCl from Roncoroni et al., which our flexible model must reproduce, and the 13-versus-94-versus-200 iteration convergence of full, bulk-only, and unregularized global DIC, which our plate regularization should match or improve on.

Why not just…?

Why not just run static strained-slab DFT, as the PdCu strain work does? That approach gives strain effects on adsorption through shifts in the metal's electronic structure, and those effects are real. However, CO2 activation at Au(110)–water interfaces: An ab initio molecular dynamics study found that static calculations misstate barriers because they ignore solvent reorganization. OH formation and H2adsorption at the liquid water–Pt(111) interface found that liquid water changes adsorption pathways. Neither paper tested surface stress, so whether static DFT fails here is an open question. The vacuum baseline answers it directly. If static DFT matches the data, we report that the solvent is mechanically silent in this window.

Why not just use implicit solvent with a flexible slab? This is the alternative most likely to succeed, and we do not argue it away. Tian et al. found that implicit models fail to predict CO₂ chemisorption at zero charge, and Roncoroni et al. found that ion-specific adsorption needs molecular detail. Both results suggest that the NaCl-versus-NaF contrast and any reorientation feature are beyond a continuum model. They are still structural findings, not mechanical ones. The decisive comparison is the second baseline. If implicit solvent matches within error in both salts, the core claim is refuted, and the proposal's lasting output becomes the validated DIC surface-stress method.

Why not just use a laser beam bouncing off the cantilever, the classical way to measure surface stress? A single-spot deflection readout is simpler and very sensitive. In a liquid cell, however, it cannot distinguish uniform surface-stress bending from clamp creep, thermal bowing, or refraction drift, because all of them move the spot. Our single-point reduction of the same images plays this role. If it gives the same surface stress with the same variance as the full-field estimator, the full-field machinery is unnecessary and we would say so. The physics comparison does not depend on which readout wins.

Why not just measure capacitance under bending and skip imaging? That measurement is part of the program. It cannot stand alone, because it gives only the strain derivative of charge, and charge measured on a real, rough film carries large uncertainty from the true surface area. Using both routes and testing the identity between them is what gives the experimental number credibility.

Proposed Experiments and Decisive Checks

The first check is internal to the simulation. Before computing any stress, the flexible-slab model must reproduce the chloride adsorption hierarchy and PZC shift of Roncoroni et al. on rigid Au(111). It must also show that its predicted stress at zero strain does not change when slab thickness is doubled.

The second check is internal to the measurement. On synthetic speckle images rendered from known plate deflections, with injected refraction and thermal nuisance fields, the plate-regularized estimator must recover the imposed surface stress without bias. It must also have lower variance than the local-DIC and single-point baselines.

The third check is the thermodynamic identity. Across the double-layer window, in both electrolytes, the DIC-measured change of surface stress with potential must agree with the bending-measured change of charge with strain. If the two disagree, the measurement is not trusted and no model comparison is made.

The decisive physics test then compares measured surface stress versus potential for 0.1 M NaCl and 0.1 M NaF on the same gold films against four predictions: vacuum, implicit solvent, explicit solvent with a frozen substrate, and explicit solvent with a flexible substrate. These predictions are reported at 300 K and 400 K so the water-structure artifact stays visible. The proposal succeeds if three conditions hold: the flexible explicit model alone captures the anodic divergence between the two salts within measurement error; the frozen-substrate ablation measurably worsens agreement; and the implicit model misses either the salt contrast or the curvature of the curve. Any other outcome is still informative, and we would report it as a limit on what explicit-solvent simulations add for electrode mechanics.