The solution–diffusion model was born in the 1960s, when researchers studying cellulose acetate reverse osmosis membranes had no way to see inside a polymer film at the molecular scale. Because no instrument could resolve the subnanometer structure, scientists assumed the membrane was a homogeneous, pore-free medium. The model describes transport in three steps: molecules partition from the feed into the polymer according to their solubility, diffuse independently through the matrix, and desorb into the permeate. Crucially, it assumes that free volume inside the polymer exists only as isolated, transient cavities created by thermal fluctuations of the polymer chains, never as continuous channels. This simplicity became the model’s greatest strength, allowing generations of researchers to compare membranes using just two numbers: solubility and diffusivity.

But the model’s experimental foundation, according to the new analysis, is shakier than textbooks suggest. The classic evidence came from stacked-membrane experiments, in which researchers measured water content layer by layer after permeation and found a gradient decreasing from the feed side to the permeate side. That gradient was interpreted as the concentration difference driving diffusion. The Rice-led team, drawing on their own molecular dynamics simulations and continuum poromechanical modeling, offers a different explanation: the gradient is a mechanical artifact of membrane compaction. Under hydraulic pressure, the pressure inside the membrane falls continuously from feed to permeate, so the polymer skeleton is compressed progressively harder toward the permeate side, squeezing out porosity along the way.

To settle the question experimentally, the researchers repeated stacked-membrane experiments with Nafion 117 ion-exchange membranes, the same material used in recent work by Freeman and co-workers. If the gradient were truly diffusion-driven, releasing the pressure should erase it within minutes as water redistributes. Instead, after more than 24 hours of permeation, depressurization, and a 10-minute relaxation, a pronounced porosity gradient remained. Soaking the individual layers in deionized water for three hours did not remove it either. Thickness measurements confirmed progressively greater irreversible compaction toward the permeate side, and short 15-minute permeation runs produced essentially no gradient at all. Flux–pressure cycling from 30 to 120 bar and back caused roughly a 20 percent irreversible loss in water permeance, independent evidence of permanent structural deformation. The gradient, the authors conclude, is not the driving force for transport but the structural fingerprint of pressure-driven flow through a compressible, interconnected pore network.

The structural evidence goes far beyond compaction experiments. Positron annihilation lifetime spectroscopy can measure the size and distribution of free-volume elements; small-angle neutron and X-ray scattering reveal nanoscale heterogeneity and solvent-rich domains; wide-angle X-ray scattering probes chain packing; and synchrotron-based X-ray absorption and emission spectroscopy show that the hydrogen-bond network of water confined in polyamide membranes is distorted relative to bulk water. Neutron-scattering studies have even resolved distinct populations of bound and mobile water coexisting within the active layer of reverse osmosis membranes. Molecular dynamics simulations complement these measurements by mapping the three-dimensional, solvent-accessible free volume and testing whether pathways actually connect across the membrane. Together, these techniques consistently indicate that membranes long classified as dense are riddled with connected, dynamic, subnanometer porosity.

Importantly, the authors stress that

The pervaporation case deserves particular attention because it forces a reconsideration of what happens inside a membrane when liquid and vapor coexist on either side. In the classical picture, a liquid feed dissolves into the polymer, diffuses across, and evaporates only upon reaching the downstream surface, where vacuum or sweep gas maintains near-zero partial pressure. The Perspective argues instead that spectroscopic and simulation evidence points to liquid-like and vapor-like domains coexisting within the membrane interior, meaning the liquid–vapor transition occurs inside the material rather than exclusively at its downstream boundary. This internal phase transition reframes pervaporation as a process whose selectivity is set partly by where and how that phase change unfolds within the pore network, a question the solution–diffusion framework was never designed to address.

Gas separation in glassy polymers offers a complementary line of evidence. Glassy polymers are frozen far below their relaxation timescales, so their free-volume elements are comparatively rigid and long-lived, making them ideal systems for probing whether transport proceeds through connected channels or isolated cavities. Recent molecular simulations described in the article show that permeability and selectivity in these materials emerge from pore connectivity, molecular sieving at constrictions, and the strength of gas–wall interactions, with adsorption-mediated hopping between sites replacing the notion of a homogeneous diffusion matrix. This mechanistic picture naturally explains long-standing empirical observations, such as the strong dependence of gas permeability on polymer chain rigidity and the distinctive sieving behavior of highly rigid ladder-type polymers, without requiring ad hoc corrections to diffusivity coefficients.

The practical stakes of this reframing are considerable. If transport proceeds through connected subnanometer pores, then the levers that matter most for membrane design are geometric and chemical rather than phenomenological: pore size distribution, network connectivity, tortuosity of the pathways, and the chemistry of the pore walls that governs solvent–polymer and gas–polymer interactions. This shifts the design paradigm from fitting solubility and diffusivity parameters after the fact toward deliberately engineering the architecture of the free-volume network itself. For polyamide reverse osmosis membranes, for example, it suggests that controlling the nanovoid structure formed during interfacial polymerization, and how that structure responds to sustained pressure, may matter as much as the polymer chemistry that determines water affinity.

The framework also carries implications for how membrane aging and compaction are understood. The irreversible permeance losses observed under cyclic pressurization, on the order of twenty percent in the experiments highlighted, become interpretable as progressive collapse of pore connectivity rather than as a mysterious drift in an effective diffusivity. Similarly, fouling and solvent-induced swelling can be reexamined as phenomena that alter pore geometry and wall chemistry in real time, which helps explain why membrane performance often depends strongly on operating history in ways that equilibrium sorption and diffusion coefficients cannot capture.

Methodologically, the convergence across PALS, SANS, SAXS, WAXS, synchrotron X-ray spectroscopy, and molecular dynamics is itself significant. Each technique probes a different length and timescale, from picosecond chain fluctuations to static nanoscale morphology, and their agreement on the existence of connected free-volume networks reduces the likelihood that any single measurement is an artifact. Continued progress will likely depend on coupling these experimental probes with simulations that can resolve how pore networks reconfigure under realistic operating pressures and solvent compositions, and on developing characterization methods capable of mapping connectivity, not merely cavity size, in operating membranes.

The authors are careful to note that the two conceptual frameworks need not be mutually exclusive at every scale. The solution–diffusion model remains a useful correlation tool, and continuum-level descriptions built on it have guided decades of engineering practice. The argument is instead about mechanism: the molecular picture underlying those correlations is better represented by viscous flow and hopping through connected nanoscale pathways than by diffusion through a truly continuous dense matrix. Recognizing this distinction matters because models that encode the wrong mechanism can mislead extrapolation, particularly as new membrane materials push into regimes of tighter sieving, higher pressures, and more aggressive solvents where the difference between the two pictures becomes experimentally consequential.

What emerges from the synthesis is a unifying thread across processes that have traditionally been treated separately. Reverse osmosis, organic solvent nanofiltration, pervaporation, and glassy polymer gas separation have each accumulated their own empirical models and design heuristics. Framing all of them around interconnected subnanometer pore networks provides a common physical substrate: pressure-driven viscous flow in liquid separations, internal phase coexistence in pervaporation, and sieving with adsorption-mediated transport in gas separation are all manifestations of the same underlying structural reality. Whether this perspective ultimately reshapes how next-generation membranes are designed will depend on translating the structural picture into quantitative, predictive tools, but the conceptual realignment it proposes is already substantial, touching everything from how compaction is interpreted to how the internal state of a pervaporation membrane is defined.

Subject of Research: Transport mechanisms and pore-network structure of dense polymer membranes used in water, solvent, and gas separations

Article Title: Interconnected pore networks as a unifying framework for transport in dense polymer membranes

Article References: Wang, R., Fan, H., Qian, J., Wang, L., Li, Y., & Elimelech, M. (2026). Interconnected pore networks as a unifying framework for transport in dense polymer membranes. Frontiers in Membrane Science and Technology, 5, Article 1955546. https://doi.org/10.3389/frmst.2026.1955546

Image Credits: AI Generated

DOI: 10.3389/frmst.2026.1955546

Keywords: polymer membranes, pore networks, solution-diffusion model, reverse osmosis, organic solvent nanofiltration, pervaporation, gas separation, molecular dynamics, free volume, membrane compaction, molecular sieving, desalination