Same-Run Reference Ratios for Predicting How Self-Lubricating Anti-Ice Coatings Hold Up Outdoors

Many coatings stay slippery by keeping a thin liquid layer at their surface, and ice cannot grip a liquid. A well-known example is a hydrogel that swells with ambient moisture into a film that will not freeze. Such a coating is only useful if the film keeps coming back after it is sheared off, dried out or frozen, over and over. That is a regeneration problem. Labs measure it with icing and de-icing cycles, but rigs differ in shear rate, humidity, water purity and loading. A coating that ranks first in one lab can rank fifth in another, and neither ranking need match what happens on a roof or a power line. This is the same lab-to-field gap that dogs drug development, where cell-culture results fail to predict the whole organism.

Particle physics has a standard way to handle a problem with this structure. Testing whether electrons, muons and taus feel the weak force equally is hard if you measure one decay rate at a time, because poorly known strong-force factors swamp the signal. Physicists therefore measure ratios of two decays of the same parent particle. The shared factors cancel, and only a known, calculable mass-dependent correction remains. We propose the same move for coatings. Every lab or field test would run a fixed reference coating in the same run, and each coating would be reported as the ratio of its degradation to the reference's. We claim this cancels rig-specific factors and makes lab rankings carry over to the field, but only among coatings that keep their liquid layer by the same physical mechanism. Across mechanisms, a correction derived from the physics is needed, just as the physics ratios need their mass correction.

What the evidence base contains, and what it does not

The research file we were given contained material on liquid-layer lubrication and the opening of the lepton-universality material. It contained no papers on adsorbent regeneration or on in vitro to in vivo translation. Those two directions are therefore used here as framing, without citations. Regeneration supplies the cycle-based view of the liquid layer as something that is used up and restored. Translation supplies the evaluation question of whether a cheap test predicts the expensive one. The file had no section describing prior attempts at this intersection. I found no evidence that same-run reference ratios have been used to transfer lab rankings of anti-ice coatings to the field, but I cannot rule it out from this file alone.

The lubrication literature in the file motivates the proposal and also sets its central constraint. It shows that interfacial liquid layers persist through quite different mechanisms, and those mechanisms respond differently to their surroundings. Cao et al. blend a sodium polyacrylate hydrogel into polypropylene. Ions in the hydrogel disrupt water's hydrogen bonding, so on exposure to moisture or frost the surface forms a non-freezing lubricating layer, with ice adhesion of 5.13 to 18.95 kPa, below the 20 kPa threshold at which ice falls off under gravity or wind. The authors note that the formulation has only been validated in polypropylene. A second mechanism comes from Ma et al., who keep water from freezing in an electrolyte gel down to −30 °C by adding ethylene glycol, which competes for hydrogen bonds. A third comes from the liquid-infused pipeline surfaces reviewed by Fan et al., which replenish their slippery film by drawing light fractions from the passing oil.

These three routes regenerate differently. The hydrogel needs ambient water to rebuild its layer. The glycol gel carries its antifreeze inside and loses it over time. The infused surface needs a supply from the flow. Geology shows the same divide. Tajika et al. attribute the long runout of a 2013 rock avalanche to pore water squeezed out of saturated ground, a film that collapses once the water drains away. Keresztúri et al. model films on Martian dunes that form through surface forces, which requires no squeezing and no bulk water source. A humidity or drainage difference between rigs will therefore scale the hydrogel's degradation much more than the glycol gel's. The rig factor does not multiply every coating by the same amount, and that is exactly where a naive ratio fails.

Core Claim

Write the measured degradation of coating $c$ in run $r$ as $D_{cr} = \theta_c \cdot \kappa_{m(c), r}$. Here $\theta_c$ is the coating's intrinsic durability. $\kappa_{m,r}$ is the run's nuisance factor as experienced by mechanism class $m$, meaning how harsh that run's shear, humidity and ice type were for coatings of that class.

If a reference coating $c_0$ of the same class runs alongside, the ratio $D_{cr}/D_{c_0 r} = \theta_c/\theta_{c_0}$ does not depend on the run. Across classes the ratio picks up a leftover term $\kappa_{m,r}/\kappa_{m',r}$. Our claim is that this leftover can be predicted from a small number of measured run conditions. The candidates are relative humidity during recovery, minimum temperature, and peak shear stress, chosen because they govern the regeneration routes above. The bet is that a physics-informed ratio, a same-class ratio plus a fitted correction across classes, predicts field rankings far better than absolute lab numbers do, and nearly as well as heavy statistical calibration while needing much less data.

Research Program

We would assemble a panel of roughly 18 coatings, six from each of the three mechanism classes. Each class gets one designated reference coating, and the polypropylene hydrogel of Cao et al. serves as the hydrogel-class reference. Every coating goes through icing and de-icing cycles on three lab rigs that differ on purpose: push-off adhesion, centrifuge adhesion, and a shear-flow rig with controlled humidity. Ice adhesion is logged after every cycle.

The primary endpoint is cycles-to-failure, the first cycle at which adhesion exceeds the 20 kPa self-shedding threshold from Cao et al. The same panel goes to at least three outdoor sites for a winter: one humid and maritime, one dry and cold, and one with freeze-thaw cycling. Each site logs adhesion and its local conditions on the same schedule. The field cycles-to-failure is the "in vivo" outcome to be predicted.

The analysis fits the multiplicative model above. It tests whether the cancellation holds within a class by checking whether ratios vary across rigs. It then fits the correction across classes as a low-dimensional function of the logged run conditions and checks it on held-out sites.

Baselines

The naive reference point is to rank coatings by their absolute lab cycles-to-failure on a single rig. This is how coatings are reported today, as in the single-number adhesion results of Cao et al. Each rig is scored on its own. It needs no extra runs, so it is the floor any method must clear.

The strongest competitor is a mixed-effects calibration model. It regresses field cycles-to-failure on absolute lab results and gives each rig and each site its own random offset. It is the natural statistician's answer and could match our approach given enough runs. To keep the comparison fair, it sees exactly the same measurements, including the reference coatings' data, which it may use as covariates. Its random-effect structure and covariates get the same tuning effort as our correction function, selected by leave-one-site-out validation.

There is no reported result found on Lacuna for transferring anti-ice coating rankings from lab to field, so no published number exists to beat. The only reported anchor is the 5.13 to 18.95 kPa adhesion range of Cao et al., which we use to sanity-check our rigs on the reference coating. Our method counts as winning if it beats the naive ranking by at least 0.2 in Spearman rank correlation with field rankings, and matches the mixed-effects model within its bootstrap interval when the number of runs is cut to a third.

Two ablations isolate what the method relies on. In the first, the reference coating is run in a separate session rather than the same run, which removes the shared nuisance factor while keeping everything else. If the ratio still works this way, cancellation is not the reason it works. In the second, a single reference is used for all classes, with no correction across classes. This tests whether the mechanism grouping actually matters.

Why not just…?

Why not just build one more realistic lab rig? No lab rig reproduces every field condition, and the lubrication literature shows that the critical variables differ by mechanism. The hydrogel depends on humidity, the glycol gel on depletion of its antifreeze, and the infused surface on its supply source. A rig tuned for one class will still mis-scale the others. Our design measures this directly: if a single rig's absolute numbers transfer as well as the ratios, the proposal is unnecessary.

Why not just fit a calibration model and skip the reference coating? This is the real threat. With many runs, random effects for each rig absorb the same nuisance factors that the ratio cancels. Our advantage should show up in the small-data regime, and in cases where the nuisance factor varies run to run within one rig, which a per-rig offset cannot capture. The comparison at reduced run counts settles this. If calibration ties us at every data size, we will report that the reference coating adds nothing beyond bookkeeping.

Why not just take ratios against one universal reference? That assumes the run factor is the same for every coating. The class-dependent regeneration routes argue against this, and the single-reference ablation tests it directly. In the physics analogy, a ratio of decays with very different kinematics still needs its mass correction. Here, skipping the correction should show up as systematic misranking across classes at the dry site.

Decisive Checks

The idea fails if same-run ratios within a class vary across rigs about as much as absolute values do, since that would mean cancellation does not occur. We predict the opposite: the coefficient of variation of ratios within a class should be below half that of absolute values.

The idea gains its strongest support from a specific crossing prediction at the dry, cold site. The hydrogel coatings should fall in field rank relative to the glycol gels by an amount the humidity-based correction predicts in advance. The uncorrected ratio should miss this shift.