Proton Exchange Membrane Electrolysis Overview

• Proton exchange membrane electrolysis is an electrochemical water-splitting process that uses a solid polymer electrolyte to separate hydrogen and oxygen efficiently.
• The technology leverages high current densities, compact stack designs, and rapid dynamic response under variable renewable power while facing challenges in catalyst stability and membrane durability.
• Integration of advanced materials, modeling techniques, and degradation-aware operations is essential for scalable, resource-efficient deployment amid iridium supply constraints.

Proton exchange membrane electrolysis (PEMEL), also termed proton exchange membrane water electrolysis (PEMWE), is an electrochemical water-splitting technology in which a solid polymer electrolyte conducts protons from anode to cathode while separating the product gases. In PEMEL, water is oxidized at the anode to oxygen, protons, and electrons; the protons cross the membrane and the electrons traverse the external circuit to the cathode, where hydrogen is evolved. The technology is distinguished by high current density, compact stacks, and rapid dynamic response under variable renewable power, but the same acidic polymer-electrolyte architecture that enables those advantages also imposes severe constraints on catalyst stability, membrane durability, porous transport design, and long-term materials availability (Wortmann et al., 3 Sep 2025, Chen et al., 2024).

1. Electrochemical architecture and operating regime

The elementary reaction set in PEMEL is conventionally written as

$$\mathrm{H_2O \rightarrow \tfrac{1}{2} O_2 + 2H^+ + 2e^-},$$

$\mathrm{2H^+ + 2e^- \rightarrow H_2},$

$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$

The membrane is the central electrolyte layer between anode and cathode. It must transport protons, block electrons, prevent $\mathrm{H_2}$ and $\mathrm{O_2}$ crossover, and operate under hot, wet, acidic, oxidizing conditions over long lifetimes (Tran et al., 26 Jan 2026).

Relative to alkaline electrolysis, PEMEL is repeatedly characterized as offering higher power density or high current density, fast dynamic response, compact system design, and, in some formulations, high-pressure operation and high-purity hydrogen. Those attributes explain its importance for coupling electrolysis to intermittent wind and solar, for flexible operation under fluctuating electricity supply, and for operation over a wide current-density range (Wortmann et al., 3 Sep 2025, Chen et al., 18 May 2026, Schofield et al., 2024).

At the cell and MEA level, the architecture is not limited to membrane and catalyst layers. On the anode side, the porous transport layer (PTL) must simultaneously conduct electrons, transport liquid water to reaction sites, remove evolved oxygen, and mechanically support the membrane electrode assembly. In consequence, PEMEL performance is governed by coupled multiphase, multicomponent transport and interfacial phenomena across the membrane electrode assembly, rather than by catalyst kinetics alone (Goswami et al., 6 Jan 2026).

2. Functional materials and their constraints

The acidic environment sharply narrows the admissible material set. For the oxygen evolution reaction in strongly acidic conditions, iridium oxide remains the preferred anode catalyst because it combines catalytic performance with stability under highly oxidizing potentials, and the analyzed literature states that no viable substitute currently exists. The same literature emphasizes the classic activity–stability problem in acidic OER catalysis: materials such as $\mathrm{Ru/RuO_2}$ can be very active, but stability is insufficient, whereas iridium oxides are the practical benchmark because they resist dissolution far better (Wortmann et al., 3 Sep 2025).

The membrane is equally central and is treated not as a passive separator but as a multi-objective materials problem involving proton conductivity, water uptake, gas crossover resistance, mechanical integrity, thermal stability, and electronic insulation. The widely used benchmark is Nafion, but the membrane-design literature also identifies Nafion’s “forever-chemistry” character, high cost, and the practical undesirability of requiring high water uptake and $\mathrm{RH} \simeq 100\%$ for peak conductivity (Tran et al., 26 Jan 2026).

A PEMEL-oriented screening specification for fluorine-free membrane discovery was reported as follows (Tran et al., 26 Jan 2026):

Property	Desired value	Nafion reference
Proton conductivity $\sigma$	$> 0.1$ S/cm @ $T=80^\circ$C, RH = 100%	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$0 S/cm
Water uptake $\mathrm{2H^+ + 2e^- \rightarrow H_2},$1	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$2 wt%	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$3 wt%
Young’s modulus $\mathrm{2H^+ + 2e^- \rightarrow H_2},$4	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$5 MPa	50–220 MPa
Glass transition $\mathrm{2H^+ + 2e^- \rightarrow H_2},$6	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$7 K	396–398 K
Decomposition $\mathrm{2H^+ + 2e^- \rightarrow H_2},$8	$\mathrm{2H^+ + 2e^- \rightarrow H_2},$9 K	553 K
$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$0	$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$1 Barrer	1.1–34.3 Barrer
$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$2	$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$3 Barrer	9.3–65.0 Barrer
Band gap $\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$4	$\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$5 eV	—

Catalyst-layer loading reduction is not treated as a free design variable. The resource-analysis literature notes that very low iridium loading can compromise catalyst-layer homogeneity, catalyst utilization, and lifetime; degradation becomes more problematic; and reducing dissolution rates is essential if lower iridium loadings are to coexist with acceptable 10- to 14-year lifetimes. A present-day state-of-the-art around 750 $\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$6 is reported, literature values for current iridium-specific power density span roughly 0.34–2.0 mg/W in 2024, and roughly $\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$7 is cited as the minimum for a 10-year lifetime (Wortmann et al., 3 Sep 2025).

PTL design introduces a second materials trade-off. In single-layer PTLs, open porous networks facilitate mass transport but incur large voltage penalties from PTL–ACL contact resistance. Bilayer architectures with dense microporous layers reduce those losses by simultaneously improving transport, contact, and structural stability, and stratified multilayer stacks with fine pores near the ACL and highly porous backing layers deliver superior performance at high current densities. In one validated comparison at $\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$8, the reported cell voltages were $\mathrm{H_2O \rightarrow H_2 + \tfrac{1}{2} O_2}.$9 V for a single-layer PTL, $\mathrm{H_2}$0 V for a bilayer, and $\mathrm{H_2}$1 V for a trilayer (Goswami et al., 6 Jan 2026).

Claims of non-noble acidic OER replacement remain exploratory. The study “Electrooxidation of a cobalt based steel in LiOH: a non-noble metal based electro-catalyst suitable for durable water-splitting in an acidic milieu” reported $\mathrm{H_2}$2 overpotential at $\mathrm{H_2}$3 at pH 1 and weight loss of $\mathrm{H_2}$4 after $\mathrm{H_2}$5, but it was a three-electrode half-cell study in sulfuric acid rather than a PEM membrane-electrode assembly demonstration (Schäfer et al., 2017).

3. Transport, interfacial losses, and degradation physics

PEMEL transport is governed by a strong coupling among membrane hydration, ionic resistance, gas evolution, and interfacial contact. The 0-D stack optimization literature writes total cell voltage as

$\mathrm{H_2}$6

with ohmic loss represented through membrane thickness and conductivity,

$\mathrm{H_2}$7

In this representation, the economics of dispatch, the thermal balance, and stack lifetime all depend on the same voltage decomposition (Schofield et al., 2024).

Operando infrared imaging has made the spatial nonuniformity of hydration explicit. In a microfluidic PEM water electrolyzer chip, higher anolyte $\mathrm{H_2}$8 concentrations increased standard deviations in current densities and produced stronger water diffusion gradients. The study resolved wet, hybrid, and dry regions within the same operating device and concluded that local membrane hydration can vary strongly depending on channel wetness, which implies localized conductivity variations and current-density heterogeneity. Because the microfluidic cell lacked a porous transport layer, the results were interpreted as direct evidence that PTLs are needed to distribute water evenly within the channels (Chevalier et al., 17 May 2025).

The PTL literature quantifies a related misconception: maximizing porosity does not necessarily minimize cell voltage. Over $\mathrm{H_2}$9 to 0.75, permeability rose strongly and mass-transport loss at $\mathrm{O_2}$0 decreased from $\mathrm{O_2}$1 V to $\mathrm{O_2}$2 V, yet the ohmic contribution increased from $\mathrm{O_2}$3 V to $\mathrm{O_2}$4 V because PTL–ACL contact resistance dominated. This establishes that PEMEL PTLs must be designed as interfacial-transport systems rather than as bulk porous media alone (Goswami et al., 6 Jan 2026).

Membrane degradation is both a performance and a safety issue. The PINN-based degradation study models membrane thinning by a first-order degradation law coupled to cell-voltage evolution and emphasizes the feedback sequence in which oxygen crosses the membrane, hydrogen peroxide forms, radicals attack Nafion, the membrane thins, and crossover increases further. Although a thinner membrane might appear to reduce ionic resistance, the model assumes conductivity degrades as

$\mathrm{O_2}$5

so the net effect is an increase in ohmic resistance and voltage over time (Polo-Molina et al., 19 Jun 2025).

Dynamic operation adds another degradation pathway through the anode catalyst layer. A temporal multiscale model for fluctuating operation hypothesizes a coupling among oxygen evolution, catalyst dissolution, and hydrogen permeation from cathode to anode. In that framework, permeating hydrogen can chemically reduce oxidized iridium surface species in the ACL; the reduced state is then more susceptible to dissolution; and the resulting loss of electrochemical surface area depends on profile shape as well as average load. The same study uses this fast–slow separation to explain why long-horizon durability under periodic forcing can be simulated with much lower computational cost (Dominguez et al., 2024).

4. Modeling, diagnostics, and control frameworks

PEMEL modeling has increasingly converged on hybrid frameworks that combine electrochemical structure with statistical or machine-learning components. A general formulation was proposed as the “Ladder of Knowledge-integrated Machine Learning,” with three levels: Level 1 – Interpolation, Level 2 – Extrapolation, and Level 3 – Representation. In the PEMWE case studies, activation losses were treated as a degradation indicator, with the governing relation

$\mathrm{O_2}$6

and knowledge was injected through signal decomposition, physics-informed loss terms, or unit-constrained symbolic discovery (Chen et al., 2024).

At the mechanistic end of the spectrum, the membrane-thinning PINN was presented as the first application of Physics-Informed Neural Networks to membrane degradation in PEM electrolyzers. The framework coupled two ordinary differential equations, one for membrane thinning and one for voltage evolution, and outperformed a conventional ANN on synthetic long-horizon extrapolation. Reported testing RMSE values were $\mathrm{O_2}$7 V for voltage and $\mathrm{O_2}$8 cm for membrane thickness, compared with $\mathrm{O_2}$9 V and $\mathrm{Ru/RuO_2}$0 cm for the ANN baseline (Polo-Molina et al., 19 Jun 2025).

At the monitoring end, transformer-based virtual electrochemical characterization was proposed for uninterrupted state-of-health assessment. The method uses an encoder-decoder transformer with patch-based time-series tokenization to reconstruct polarization curves from routine operational data sampled at $\mathrm{Ru/RuO_2}$1 Hz. Across four longitudinal runs lasting up to 478 hours, the model reported about a 10× reduction in mean squared error relative to a vanilla transformer for polarization-curve reconstruction, but the study remained a proof of concept: models were trained independently for each run, results were reported on a validation partition rather than a separate held-out test set, and cross-stack generalization was not established (Chen et al., 18 May 2026).

Control research reflects the same emphasis on dynamics. In a hybrid hydrogen electrolyzer–supercapacitor system, PEMEL was assigned the middle-frequency portion of transient power, between the high-frequency supercapacitor branch and the low-frequency alkaline branch. The PEMEL branch used a dynamic integral droop law,

$\mathrm{Ru/RuO_2}$2

while the grid-side inverter imposed an inertia-emulation response. In hardware-in-the-loop tests, PEMEL power moved from about $\mathrm{Ru/RuO_2}$3 to about $\mathrm{Ru/RuO_2}$4 with settling times of $\mathrm{Ru/RuO_2}$5 and $\mathrm{Ru/RuO_2}$6 in step-up and step-down disturbances, respectively, illustrating how PEMEL’s dynamic response is now being treated as a system service rather than only as a process load (Lin et al., 3 Jan 2026).

5. Scale-up, degradation-aware operation, and iridium-limited deployment

The economics of PEMEL depend strongly on whether degradation is embedded directly in design and dispatch. A dynamic optimization study for a grid-connected PEM electrolyzer supplying $\mathrm{Ru/RuO_2}$7 found that including usage-based degradation raised the levelized cost of hydrogen from $\mathrm{Ru/RuO_2}$8, increased total CAPEX from $\mathrm{Ru/RuO_2}$9 million, reduced utilization from $\mathrm{RH} \simeq 100\%$0 to $\mathrm{RH} \simeq 100\%$1, and shortened the replacement interval to $\mathrm{RH} \simeq 100\%$2 years. Under a 2030 mid-case, the same framework projected $\mathrm{RH} \simeq 100\%$3, $\mathrm{RH} \simeq 100\%$4 thousand cells, storage of $\mathrm{RH} \simeq 100\%$5 days, and a replacement interval of $\mathrm{RH} \simeq 100\%$6 years (Schofield et al., 2024).

The same study makes explicit that degradation-aware dispatch shifts the optimum toward larger stacks, lower average current densities, and less hydrogen storage. Its empirical degradation law is current-density dependent,

$\mathrm{RH} \simeq 100\%$7

so high-current operation may be economically attractive in simple arbitrage models yet destructive once lifetime consumption is internalized (Schofield et al., 2024).

At larger scale, however, PEMEL becomes a materials-availability problem as much as a techno-economic one. The iridium-demand analysis writes annual primary iridium demand as

$\mathrm{RH} \simeq 100\%$8

thereby including both new installations and recursive end-of-life replacements. With global iridium production fixed at about $\mathrm{RH} \simeq 100\%$9, the study found that meeting net-zero targets would require both significant improvements in catalyst efficiency and access to roughly $\sigma$0 of global iridium production annually. Under conservative assumptions, a $\sigma$1 PEMEL market share is infeasible; shortages could arise as early as 2030; and long-term needs beyond 2040 are significantly underestimated if replacement cycles are ignored (Wortmann et al., 3 Sep 2025).

These results alter the interpretation of PEMEL scale-up. Under Business-as-Usual deployment, PEMEL may be manageable if catalyst thrift improves sufficiently. Under the IEA Net Zero Emissions pathway, in which PEMEL captures $\sigma$2 of the electrolyzer market and reaches 1468 GW by 2050, feasibility narrows to a small set of favorable conditions: rapid reduction of iridium-specific power density, recycling near $\sigma$3, and a relatively favorable share of world iridium supply. This suggests that PEMEL competitiveness is conditional rather than absolute and that alkaline or AEM electrolysis may need to absorb applications where PEMEL’s fast dynamic response is less essential (Wortmann et al., 3 Sep 2025).

6. Emerging directions, controversies, and open problems

One major research direction is fluorine-free membrane discovery. An AI-based design strategy combined virtual forward synthesis with Gaussian Process Regression models for proton conductivity, water uptake, Young’s modulus, glass transition temperature, thermal decomposition temperature, $\sigma$4 permeability, $\sigma$5 permeability, and band gap. The workflow screened nearly 66 million synthesizable polymers, imposed chemistry constraints such as No halogen species, No amide $\sigma$6, and Must contain sulfonate $\sigma$7, and reported 1,738 promising PEM candidates while rediscovering known membranes such as Nafion, Pemion, and sPPO (Tran et al., 26 Jan 2026).

A second direction is catalyst-layer re-architecture. The proton-electron coupled catalyst concept was demonstrated in PEMFCs and electrochemical hydrogen pumps rather than PEMEL, but it showed that an ionomer-free catalyst layer can maintain proton conduction while strongly reducing local gas-transport resistance. A plausible implication is that PEMEL cathodes, and eventually PEMEL anodes with different materials, could benefit from catalyst/support structures that integrate protonic and electronic pathways rather than relying on bulk ionomer-filled pore space. The same paper also makes clear that the demonstrated materials—PtNi on sulfonated CNTs—are not directly transferable to the PEMEL anode, where carbon corrosion and acid-stable OER catalysis remain decisive constraints (Zhang et al., 23 May 2026).

Several persistent misconceptions are therefore not supported by the current literature. Lower iridium loading alone is not automatically beneficial, because utilization, homogeneity, dissolution, and lifetime may all deteriorate (Wortmann et al., 3 Sep 2025). Higher PTL porosity alone is not automatically beneficial, because interfacial contact resistance can dominate mass-transport gains (Goswami et al., 6 Jan 2026). Apparent success of a non-noble acidic OER electrode in sulfuric acid is not equivalent to a PEM membrane-electrode assembly demonstration (Schäfer et al., 2017). And ionomer-free catalyst-layer architectures are not yet validated for PEMEL anodes, even when they are compelling in adjacent proton-exchange devices (Zhang et al., 23 May 2026).

Across these strands, PEMEL emerges as a technology defined by coupled constraints rather than by a single bottleneck. High current density, compactness, and dynamic response explain its strategic importance, but scalable deployment depends simultaneously on membrane chemistry, PTL architecture, catalyst-layer utilization, degradation-aware operation, recursive replacement accounting, and access to one of the scarcest industrial metals. The contemporary research trajectory therefore points less toward a single breakthrough component than toward coordinated advances in materials efficiency, interfacial design, closed-loop recycling, health monitoring, and resource-aware system planning (Wortmann et al., 3 Sep 2025).