Reference

Model limits

This document describes what the PROTONYX engine does not model. Stating the limits is what makes the rest credible. The engine returns confident numbers within a validated envelope and is deliberately silent when your design sits outside it, letting the results stand and the warnings speak.

The validated envelope

The engine is validated near 80 °C, well hydrated, at moderate pressure. Confidence decays outside this band — not because the maths is wrong, but because the physical assumptions it rests on no longer hold.

  • Temperature: Below 10 °C and above 100 °C, hydration assumptions become uncertain. Above 100 °C with liquid feed, the tool flags that pressurisation is needed but does not model it.
  • Membrane hydration: Mist feed is modelled through a fixed 0.85 hydration factor relative to liquid. The tool does not resolve the kinetics of water loss through the membrane or the transient behaviour of a drying cell.
  • Current density: Above 2 A/cm², thermal limits become significant and are not modelled; results run optimistic. At or within 20% of the limiting current density, the concentration term rises sharply and the model depends on a literature-typical figure unless you supplied a measured one.
  • Pressure range: Thermodynamic Nernst shifts are captured. Above ~10 bar absolute, or with a differential pressure above ~10 bar, the tool flags that bubble behaviour, mass transport, crossover and membrane stress under pressure are not modelled.
  • Where the physics is simplified

    Membrane conductivity uses a generic temperature correction

    Enter conductivity on the conventional PFSA reporting basis — 30 °C, 100 % RH, approximately 0.1 S/cm for Nafion 117. The engine corrects it to operating temperature with the Springer et al. (1991) Arrhenius term, σ(T) = σ₃₀ · exp[1268 · (1/303.15 − 1/T)]: a factor of exactly 1 at 30 °C and about 1.46 at 60 °C.

    Two consequences. The correction is generic, not fitted per membrane, so a membrane whose conductivity rises differently with temperature will be off by that difference. And a value already measured at operating temperature would be corrected twice, overstating conductivity and understating ohmic loss; the tool warns when a high conductivity is paired with a high operating temperature, which is the signature of that mistake.

    Activation overpotential uses Tafel, floored at j₀

    Overpotential is the Tafel form: η = (b / 2.303) · log₁₀(j / j₀), where b is the Tafel slope. Below the exchange current density j₀, the equation yields η < 0 — activation is negative, which is physically allowed but makes the model singular at zero current. The engine floors the current density at j₀, so below that point η = 0 and there is no activation loss.

    This produces a kink at j₀ visible on a log plot. It is not the Butler–Volmer asinh form, which would smooth the curve; we use Tafel because the reference physics (`docs/engine.pdf`) uses it, and any deviation breaks parity with the golden-vector suite. Real cells do exhibit a smooth low-current regime, so if you operate just above j₀ the model may understate activation. This is small in practice — j₀ is typically 1 µA/cm² for the anode and 1 mA/cm² for the cathode — but it is worth knowing.

    Geometry is active area only

    The engine takes a single length `a` (cell side, mm) and assumes a square active area. In-plane collector resistance, aspect ratio, feed-point distance, and land:channel geometry are not modelled. The in-plane resistance is lumped into the contact-resistance input instead, so if you have measured it or estimated it from your design, that is where it goes.

    Mass transport is literature-typical

    The limiting current density is modelled using the standard diffusion equation. Its value is a user input, `jLim`, and it is literature-typical — representative of liquid-fed PEM at 1–3 bar and moderate flow. A real limiting current depends on diffusion-layer thickness, porosity, tortuosity and reactant concentration, none of which are engine inputs.

    At or above jLim the concentration overpotential is clamped so the reported voltage is a floor, not a prediction. If you are operating within 20% of jLim and have not measured your own limiting current, the result depends on a number nobody fitted to your stack. The tool warns you when you are in that band.

    What is not modelled at all

    Gas crossover and hydrogen safety

    Faradaic efficiency is assumed to be unity. Hydrogen crossover through the membrane rises with differential pressure and temperature, reducing actual efficiency and creating a safety margin that the tool does not compute. Pressurised stacks in particular have a turndown limit below which crossover becomes significant; the engine makes no statement about it.

    This is the most consequential omission on the page, so it is worth being concrete about the size of it. Crossover is close to a fixed flux set by pressure, so as a *fraction* of production it grows without bound as current falls. A reference model of a 50 µm membrane at 30 bar puts hydrogen in the oxygen stream at roughly 3% at 0.5 A/cm², 6% at 0.25 A/cm² and 15% at 0.1 A/cm² — the lower flammability limit being 4%. Faradaic efficiency over the same range falls from 98% to 92%.

    Two things follow. Our production and efficiency figures are optimistic at low current density, by a margin that widens as you turn down. And the question an operator actually asks about a pressurised stack — how far can I turn this down before the oxygen stream reaches the flammability limit — is one this tool cannot answer at all. Size your turndown against a model that computes crossover, not against this one.

    Mist feed

    Mist feed is modelled as a fixed reduction in membrane hydration (a factor of 0.85 against liquid feed), with water activity at the electrode held at 1.0 exactly as for liquid. Water-supply limits, membrane dry-out, droplet transport into the catalyst layer and the operating envelope in which a mist-fed cell stays hydrated are not modelled. The electrostatic stage acts on mist feed only through a hydration gain. Treat mist-fed results as a comparison against the same cell on liquid feed, not as an absolute prediction.

    Novel subsystems are hypotheses

    The acoustic, electrostatic and magnetic subsystems act in the model through reduced bubble coverage, improved hydration on mist feed, and — for the magnetic stage with a ferromagnetic catalyst — a direct spin-polarisation voltage reduction capped at 50 mV. Those mechanisms are physically motivated, but their coefficients have not been validated against measured cells. The improvement over the baseline cell that every result reports is a prediction to test on hardware, not an established effect, and every result that enables a subsystem says so in its assumptions.

    Bubble behaviour under pressure

    Denser gas at higher pressure means smaller bubbles and different detachment. The limiting current and concentration overpotential do not have a pressure dependence in this model, so the real net effect of pressurising can differ from the Nernst voltage shift alone — and can run in the opposite direction. The Nernst term moves the reversible floor; everything else stays fixed.

    Validation carries no fitted uncertainty

    The tool compares measured voltage against the model. The validation grading is deliberately coarse — within 20 mV is "close agreement", within 50 mV is "moderate", above 50 mV is "large" — because the model carries no fitted uncertainty. Cell-level physics is PROTONYX's own and measured, but no error model has been calibrated against a large measurement set. The residuals you see are real disagreements; they are not error bars and should not be treated as calibration windows.

    Efficiency, kWh/kg H₂, production rate and any other derived quantity are never compared against measurement. Only voltage is graded. If your measured efficiency differs from the model's, the model parameters may be wrong, your measurement may have unmeasured loss mechanisms, or both.

    Degradation is linear

    Voltage degradation is modelled as a linear fade: 5 µV/h is typical, and you can edit it. This is deliberately linear. Real degradation accelerates under high current density and high temperature, but fitting those accelerators would require data we do not have. Above 2 A/cm² or 80 °C the tool warns that the projection to end-of-life is optimistic.

    Dynamic and transient behaviour

    The engine is a steady-state model. It returns a point on the polarisation curve at your chosen current density and temperature. It makes no statement about start-up, load-following, cycling, or any time-dependent behaviour. If your application involves rapid transients or cycling, the model is not suitable.

    Spatially resolved fields

    No computational fluid dynamics, no in-plane potential distribution, no gas-phase concentration gradients. Bubble coverage — the single spatially-averaged field the model uses — is an empirical function of current density and subsystem intensity; it is not resolved.

    The provenance division

    Cell-level physics is PROTONYX's own and measured. The polarisation curve — anode activation, cathode activation, membrane ohmic resistance, contact resistance, concentration overpotential — rests on measured material properties and stack design.

    Everything below the cell boundary — stack interconnect resistance, compression work, pump and cooling power, degradation rate, and every other balance-of-plant constant — is literature-typical. These are engineering estimates and not measured PROTONYX data. Every one is user-editable. Changing a scale-up assumption re-runs the system efficiency immediately. This division is the product's credibility position: the tool makes no claim to predict system efficiency better than you can with standard engineering models, but it does resolve the cell and quantifies what the innovation bought.

    How limits are surfaced

    Every result carries two classes of constraint:

  • Warnings appear inline and unmissable at the top of the results panel. They flag when the design sits outside the validated envelope or depends on a literature-typical value in a sensitive spot. A warning does not make the result wrong, but it does let you know which assumptions matter.
  • Assumptions list every substituted default — membrane conductivity not user-supplied, limiting current density not user-measured, scale-up parameters not user-edited. They are collapsed by default but never hidden. Open the panel to see what the numbers rest on.
  • A result is never displayed without both.