Correlated interfacial water transport and proton conductivity in perfluorosulfonic acid membranes

Significance

Fuel-cell proton-exchange membranes (PEMs) are central to hydrogen fuel-cell technology. The mechanism and optimization of proton transport in these materials is intimately linked with hydration and water transport. Studies of water transport in PEMs have previously been limited to macroscopic permeability measurements. Our work connects the macroscopic proton transport in PEMs to the molecular water arrangement: Two distinct types of water, differing in their degree of molecular coordination, transit the membrane at very different rates. The more weakly coordinated water species consistently exhibits faster transit across five different PEMs and shows a direct correlation to the proton conductivity, suggesting that designing PEMs to maximize the fraction of undercoordinated water is key for maximizing overall transport.

Abstract

Water must be effectively transported and is also essential for maximizing proton conductivity within fuel-cell proton-exchange membranes (PEMs). Therefore, identifying relationships between PEM properties, water transport, and proton conductivity is essential for designing optimal PEMs. Here, we use coherent Raman spectroscopy to quantify real-time, in situ diffusivities of water subspecies, bulk-like and nonbulk-like (interfacial) water, in five different perfluorosulfonic acid (PFSA) PEMs. Although the PEMs were chemically diverse, water transport within them followed the same rule: Total water diffusivity could be represented by a linear combination of the bulk-like and interfacial water diffusivities. Moreover, the diffusivity of interfacial water was consistently larger than that of bulk-like water. These measurements of microscopic transport were combined with through-plane proton conductivity measurements to reveal the correlation between interfacial water transport and proton conductivity. Our results demonstrate the importance of maximizing the diffusivity and fractional contribution of interfacial water to maximize the proton conductivity in PFSA PEMs.

Proton-exchange membranes are the central, performance-limiting component of hydrogen fuel-cell (PEMFC) technology for numerous reasons. A long-standing challenge in large-scale PEMFC implementation concerns water management within PEMs. Water structure and transport are intimately related to PEMFC performance through regulation of water management (i.e., its spatial distribution) and by affecting proton transport within perfluorosulfonic acid (PFSA) PEMs (1–3). Rational development of high-performance PEMs has remained challenging because it requires a fundamental knowledge of the link between membrane chemistry, structure, and transport properties. The current industry standard PEM is Nafion, which consists of hydrophobic Teflon chains with ether-linked side chains that are terminated by hydrophilic sulfonic acid head groups (4). The polymer architecture of Nafion leads to hydrophilic–hydrophobic phase separation, resulting in self-assembly of ionic clusters (domains) of a few nanometers in diameter (4–8).

The understanding of Nafion morphology has largely converged to a view of a hydrophobic membrane matrix containing nanoscale, hydrophilic domains with widths up to 4 nm in diameter (9). Water is believed to reside in the hydrophilic domains and facilitate proton transport through the membrane. The nanoscale width of the domains makes Nafion an intriguing system regarding transport properties because nanoscale confinement may strongly influence the water chemistry and transport in the membrane (10–12). Indeed, vibrational spectroscopy of water––which reflects the local water chemical environment through changes in the shape of OH vibrational bands––in Nafion by different groups has shown that water hydrogen bonding (coordination) is strongly perturbed in Nafion compared with “normal” or bulk water (13–15). More recent work on water properties in nanometer-sized Nafion mimics, sodium bis(2-ethylhexyl)sulfosuccinate (aerosol OT) micelles, by Fayer and coworkers (16, 17), has established the existence of two water species with very different vibrational signatures and dynamics in these micelles: bulk water and interfacial water. The same group has proposed a core–shell model of water structure in confinement where bulk-like water exists at the core of the micelles, and interfacial (shell) water exists at the micelle interface (18). Benziger and coworkers (19, 20) have proposed a similar core–shell model for water structure in Nafion channels, and our previous results have experimentally confirmed the existence of two subspecies of water––bulk-like water (bulkW) and nonbulk-like water (nonbulkW) or interfacial water––in Nafion and similar interfacial geometries (21, 22). That said, the functional implication (if any) of these different water populations on, e.g., transport properties of Nafion is unknown. Previous work by Song et al. (23) using pulsed field gradient NMR spectroscopy with different spin labels has suggested that water near the hydrophobic portion of Nafion exhibits much faster diffusivity (more than four orders of magnitude larger) than bulkW in Nafion, but the relationship between these diffusivities and overall water transport is not clear. Moreover, a recent study using NMR, neutron scattering, and simulations showed a somewhat contradictory result: subdiffusive transport of water confined to ionic domains in various PFSA PEMs, including Nafion and Aquivion membranes (24). It has, in summary, been challenging to experimentally address the combined effect of PFSA architecture and water structure (and dynamics) on water and proton transport properties of these media.

Computational work has arguably more successfully targeted the transport mechanisms and relation between water and proton transport in PFSA PEMs (reviewed in refs. 25 and 26). An important quality of these theoretical studies is the derived mechanistic insight into the proton transport mechanism for differing hydration and polymer structure. In particular, reactive multistate empirical valence band (MS-EVB) simulations have shown that the two primary mechanisms for proton transport, the Grotthuss proton-hopping mechanism (or structural diffusion) and vehicular diffusion are anticorrelated to some degree in the confined PSFA geometry compared to bulk water where the two mechanisms appear to be additive (27, 28). The anticorrelation makes it more difficult to distinguish between the contributions of the Grotthuss and vehicular components. Further simulations by Voth and coworkers (29) that included the complete Grotthuss physics using reactive dynamics with MS-EVB went on to show that the sulfonic acid-to-sulfonic acid hydration shell proton-transfer mechanism is dominant in the PFSA geometry. The authors showed that excess protons at hydration levels of 5, 9, and 12 were almost always (95% of the time) within two water solvation shells of at least one acidic head group, which themselves have overlapping solvation shells that facilitate Grotthuss-type proton transport in PFSA nanoconfined water pools. Increasing the hydration level changes the size and connectivity of the hydration layers of the sulfonic acids; however, increased hydration does not lead to a significant increase of the hydrated protons in the center (or core) of a water pool beyond two hydration shells from sulfonic acid groups. This suggests that sulfonic-acid-assisted passing of protons is the dominant mechanism for proton transport, even at higher hydration levels, in contrast to previous work using an empirical, nonreactive force field that showed that high-speed proton transport in PFSAs was reminiscent of transport in bulk water (30). In that study, the authors found that high-speed proton transport occurred predominantly in the core of PFSA PEM channels—presumably much farther away from sulfonic acid groups––at high hydration levels because strong proton interaction with sulfonic acid groups could slow transport (30). Interestingly, the width and geometry of the water layer drastically changed the proton solvation and overall water structure as well (31). An overview of the literature shows that while numerous simulations have largely clarified how protons transport in nanoconfined PFSA materials, the relation between water––or water subspecies––transport and proton transport has not experimentally been demonstrated.

Vibrational spectroscopy is an effective method to identify water subspecies in situ; however, measuring molecular transport properties in real time in a microscopic volume is experimentally demanding. Here, we present real-time, in situ, label-free measurement of water and water subspecies diffusion in fully hydrated (saturating λ-condition) PEMs using broadband coherent anti-Stokes Raman scattering (BCARS) spectroscopy. We study five different PEMs, differing in chemical composition and processing and show that the apparent diffusion coefficient (ADC) for nonbulkW (interfacial water) is always larger than that of bulkW. By combining our vibrational spectroscopy with macroscale ac impedance measurements of through-plane conductivity under the same conditions, we found that the diffusivity of nonbulkW was directly correlated with proton conductivity in PFSA PEMs, independent of membrane chemistry.

Results

H₂O–D₂O Exchange Experiments with PFSA Membranes.

We measured in situ, time-dependent water transport in Nafion and other PFSA membranes using a microfluidic Y valve to initiate H₂O–D₂O exchange within an immobilized PFSA membrane in our BCARS microscope, as in our previous work (21). BCARS is a nonlinear analog of Raman spectroscopy that provides a quantitative Raman-like (RL) vibrational spectrum (700–4,000 cm⁻¹), after appropriate processing (SI Appendix, Experimental Methods), with millisecond time resolution from a femtoliter volume in the sample. Fig. 1 (Top Left) shows a schematic of the measurement wherein a PFSA membrane, in this case, dispersion cast Nafion 212 (N212, equivalent weight = 1,100 g mol⁻¹), was initially saturated by D₂O, representing a D₂O-equilibrated state. After switching the Y valve to initiate exchange, BCARS spectra from the membrane were acquired at 100-ms intervals. Switching the microfluidic channel solution to H₂O took ∼5 s in our gravity-driven, laminar flow conditions (SI Appendix, Fig. S1), after which H₂O diffused into the membrane and D₂O in membrane diffused out (Movie S1).

Fig. 1.

Measuring water transport in situ in PFSA membranes using real-time vibrational microscopy. (A) RL intensity of the OD-stretching vibration (green) decreases while the intensity of the OH-stretching vibration (red) increases during the H₂O–D₂O exchange. All RL spectra were normalized to the maximum value in the respective regions. (B) OD (green) and OH (red) concentrations vs. time obtained by integrating the respective spectral regions; solid black lines are fits to the Fickian diffusion model described in SI Appendix. The concentration profiles were normalized to their maximum value. (C–E) Time-dependent spectral amplitudes showing water subspecies contributions during H₂O–D₂O exchange in N212. All spectra are divided by the maximum of the OH band at t = 30 s (after exchange is complete) to show different temporal evolution.

The OH and OD concentrations were obtained from the integrated amplitudes of the OH (2,918–3,810 cm⁻¹) and OD (2,182–2,780 cm⁻¹) stretching vibrations, respectively (Fig. 1B). The time-dependent OH and OD concentrations show that the diffusion process in N212 was complete within ∼30 s of initially detecting the OH signal. Fitting the time-dependent OH and OD traces with an analytical solution to Fick’s second law (see SI Appendix, Fig. S2 and Eq. S3 for fitting details), allowed us to quantify the transport by extracting an ADC from the traces. The ADC for the integrated OH amplitude [ADC(OH) = 3.75 ± 0.10 × 10⁻¹⁰ m²/s (mean ± SD)] was statistically identical to the ADC of the integrated OD amplitude [ADC(OD) = 3.89 ± 0.20 × 10⁻¹⁰ m²/s (mean ± SD)] in N212 (SI Appendix, Fig. S3) from n = 6 independent N212 samples. These numbers are quantitatively consistent with our previous measurements using a slightly modified experimental and fitting protocol (21) and are consistent with other values reported for water diffusion in Nafion (SI Appendix, Table S1). Moreover, the reverse experiment––D₂O diffusing in an H₂O equilibrated membrane––shows statistically identical results (SI Appendix, Fig. S3), which confirms that OH and OD exchange simultaneously. Accordingly, both the OH and OD regions can, in principle, be used to analyze the diffusivity of water in Nafion; however, for simplicity, we focus on the OH region for experiments of H₂O diffusing into a D₂O-equilibrated membrane for the remainder of this work. Finally, we note that experiments from different depths (relative to the channel–membrane interface) and lateral positions within N212 showed that ADC(OH) was not position dependent (SI Appendix, Figs. S2 and S3), as expected for a nanoporous membrane that is uniform over micrometer length scales (the size of our focal volume).

Quantifying Heterogeneous Diffusivity of Water in PFSA Membranes.

We have previously shown that the overall OH-stretch intensity of water in Nafion can be decomposed into two subspecies contributions: bulkW and (nonbulkW). BulkW is represented by the RL response of MilliQ water, and the nonbulkW spectrum was deduced by analyzing many RL spectra of two different Nafion membranes at multiple hydration states (SI Appendix, Fig. S4) with a constrained classical least-squares (CCLS) global fitting algorithm (see SI Appendix, Fig. S5 for fitting details). Inspection of the nonbulkW shape (SI Appendix, Fig. S5, orange) shows that it is clearly less hydrogen-bonded compared with bulkW with the main peak being strongly blue-shifted to higher frequency.

In addition to bulkW and nonbulkW subspecies, exchange of D₂O for H₂O will necessarily result in an intermediate HOD species, with a distinct vibrational response, which also contributes to the OH-stretch intensity (32–34). Therefore, to analyze the shape of the time-dependent RL spectra, a third spectral component for HOD is required. We deduced the HOD RL spectrum using the CCLS algorithm on RL spectra from a series of D₂O/H₂O mixtures (SI Appendix, Figs. S6 and S7). Using the deduced HOD lineshape, the fitted concentration of H₂O, D₂O, and HOD from the mixtures was quantitatively consistent with those predicted by the established binomial mixing for isotopic scrambling (SI Appendix, Fig. S8) (33, 34), confirming the accuracy of the HOD lineshape. In principle, one would expect a bulk and non-bulk HOD subspecies for Nafion, and therefore, we further checked whether the binomial mixing held for H₂O, D₂O, and HOD in N212 equilibrated in different H₂O/D₂O mixtures using the deduced HOD lineshape. Also here, good agreement was observed between the decomposition concentrations, in this case using bulkW, nonbulkW, and HOD lineshapes, and those predicted by the binomial mixing rule (SI Appendix, Fig. S9). These results show that the OH-stretch vibrational band in Nafion during H₂O–D₂O exchange can be described in a consistent manner using the sum of three lineshapes: bulkW, nonbulkW, and HOD.

With spectral shapes for the primary water subspecies in the membrane, we could further determine their respective diffusivities. Time-dependent RL spectra of H₂O–D₂O exchange in N212 were decomposed using nonnegative least-squares according to $S_{\text{meas}}=C_{\text{bulkW}}\times S_{\text{bulkW}}+C_{\text{nonbulkW}}\times S_{\text{nonbulkW}}+C_{\text{HOD}}\times S_{\text{HOD}}+E$, where $S_{meas}$ is the experimental RL spectrum, $C_X$ is the fractional concentration of subspecies X, $S_X$ is the subspecies X RL spectrum, and E is the error between the experimental and calculated spectrum. The time-dependent contributions of the different components to the overall OH-region spectra from N212 during the H₂O–D₂O exchange are shown in Fig. 1 C–E. Notably, the nonbulkW (and HOD) component reached its maximum value much earlier (at t = 7.4 s) than bulkW (∼35 s), demonstrating the larger mobility of the nonbulkW water species. Fig. 2A shows the C vs. t curves for each subspecies in N212 from all RL spectra in a representative exchange experiment. The residuals from these decompositions (Fig. 2A, Insets, gray curves) are less than 5% of the total spectral amplitude, and adding a fourth component to the decomposition did not improve the fit (SI Appendix, Figs. S10 and S11). Fig. 2A confirms the observation that nonbulkW (orange) appears earlier (at t ∼ 3 s) than bulkW (dark blue at t ∼ 7 s) in N212. The concentration of HOD (light blue) at first increases with the appearance of nonbulkW and decreases to zero upon complete exchange, as expected, since it is a transient species that should disappear as the membrane completely exchanges D₂O for H₂O. Similar water transport was observed in three individual experiments within one N212 membrane and across n = 6 different N212 membrane samples.

Fig. 2.

Heterogeneous water transport in PFSA membranes. (A) Decomposed concentration profiles of bulkW (blue), nonbulkW (orange), and HOD (cyan) (see SI Appendix, section 6 for decomposition details) in N212 during H₂O–D₂O exchange; the total normalized OH signal is shown in black. Red lines show fits to 1D Fickian model for bulkW and nonbulkW. (A, Bottom) Measured RL spectra (black) and fitted spectra (red) at times indicated by gray vertical lines. RL spectra of subspecies of bulkW (blue), nonbulkW (orange), and HOD (blue), as well as residuals (gray), are shown in the spectral plots. (B) ADC of bulkW (blue) and nonbulkW (orange) computed from fitting concentration profiles for different PFSA membranes. Error bars are SDs of the ADCs from three measurements in each of n = 6 independent membrane pieces for each PFSA PEM. (C) Total water (OH) diffusivity in PFSA membranes is a linear combination of bulkW and nonbulkW ADCs. Measured values (red bars) and calculated values (gray bars) as a linear combination of ADC(bulkW) and ADC(nonbulkW) weighted by their fractional concentration in the respective membrane under H₂O-saturated conditions.

Before analyzing the concentration profiles of the subspecies with the 1D diffusion model, we performed a control experiment to verify that the isotopic scrambling reaction H₂O + D₂O ⇌ 2HOD was sufficiently fast to not influence our measurements on the millisecond timescale. We measured H₂O diffusion into a 4.2-mm-deep × 1-mm-wide D₂O-filled aperture (SI Appendix, Fig. S12), decomposed the time-dependent OH region RL spectra into HOD and bulkW, and calculated the ADC(bulkW) from the 1D Fickian model (SI Appendix, Eq. S3). Should isotopic exchange strongly influence the bulkW signal we detect, one would expect our ADC analysis considering only diffusion (as opposed to reaction–diffusion) to disagree strongly with the mutual diffusion coefficient of water. We found an ADC(bulkW) ∼2.2 ± 0.4 × 10⁻⁹ m²/s (n = 3 independent experiments) from this experiment, which is nearly identical to the accepted H₂O mutual diffusion constant (D_H2O ∼ 2.3 × 10⁻⁹ m²/s) reported in experimental and theoretical studies (35–39). This close agreement demonstrates that our diffusion-only analysis accurately captures the H₂O/D₂O transport phenomena in the exchange experiments. Thus, the 1D Fickian diffusion model was used to calculate the ADC of bulkW and nonbulkW from their respective concentration profiles.

We found that ADC(bulkW) = 2.5 ± 0.3 × 10⁻¹⁰ m²/s and ADC(nonbulkW) = 10.1 ± 0.9 × 10⁻¹⁰ m²/s (mean ± SD), respectively, from n = 6 independent N212 samples. To determine how membrane properties such as equivalent weight (EW) or polymer architecture influence water transport, H₂O–D₂O exchange experiments were conducted in various other PFSA-based membranes: Nafion 117 (N117), Nafion 1035 (N1035), Aquivion E87 (E87), and Aquivion E98 (E98) membranes. N117 is formed from the chemically identical polymer (EW = 1,100 g mol⁻¹) as N212 but is processed using melt extrusion. The extrusion processing results in different physical properties such as membrane thickness, ionic channel size, water–membrane interaction, and proton conductivity as reported previously (21, 40, 41). N1035 (EW = 1,000 g mol⁻¹), also extrusion processed, contains more SO₃H-terminated head groups per gram of polymer compared with N117. Aquivion E87 (EW = 870 g mol⁻¹) and E98 (EW = 980 g mol⁻¹) are also extrusion processed but have a different polymer architecture from Nafion-based membranes as their side chains are shorter.

The RL spectra from H₂O–D₂O exchange experiments on the fully hydrated PFSAs were decomposed similarly as those from N212 with the same spectral shapes for bulkW, nonbulkW, and HOD. This was justified given the good agreement between the static H₂O hydrated spectra of each membrane and the spectral fits (SI Appendix, Fig. S13). The extracted ADC(bulkW) and ADC(nonbulkW) of N117 were 2.8 ± 0.6 × 10⁻¹⁰ m²/s and 10.4 ± 0.3 × 10⁻¹⁰ m²/s (n = 6 independent N117 samples), respectively––statistically identical to those measured in N212 (see SI Appendix, Fig. S14A for statistics). The subspecies diffusivities measured in E87 and E98 (SI Appendix, Fig. S14B) were also equivalent: ADC(bulkW) = 3.9 ± 0.4 × 10⁻¹⁰ m²/s and ADC(nonbulkW) = 6.6 ± 1.3 × 10⁻¹⁰ m²/s (n = 6 independent E87 and E98 samples). N1035 exhibited the fastest diffusivity among all of the PFSA membranes, and the ADC(bulkW) and ADC(nonbulk) of it were 5.8 ± 0.7 × 10⁻¹⁰ m²/s and 14.5 ± 2.0 × 10⁻¹⁰ m²/s (n = 7 independent experiments), respectively (Fig. 2B).

The overall water diffusivity––ADC(OH)––in N1035 was (statistically) significantly faster by 1.8-fold compared with N212, E87, and E98 (Fig. 2C, red bars and SI Appendix, Fig. S3) and by 1.3-fold compared with N117. We previously proposed that the sum of the ADCs for bulkW and nonbulkW weighted by their respective fractional concentration in the membrane (when fully hydrated with H₂O) could reproduce the measured ADC(OH) (21). From the fractional concentration of each subspecies at long times after exchange (effectively an equilibrium H₂O state), we indeed found that a weighted sum of the ADC(bulkW) and ADC(nonbulkW), $\text{ADC}\left(\text{OH}\right)\text{'}=C_{\text{bulkW}}\times \text{ADC}\left(\text{bulkW}\right)+C_{\text{nonbulkW}}\times \text{ADC}\left(\text{nonbulk}\right)$, where $C_X$ was the fractional concentration of species X at equilibrium in H₂O, almost perfectly reproduced the total measured ADC(OH) in all five PFSA samples (Fig. 2C, gray bars).

NonbulkW Transport and Membrane Proton Conduction Are Correlated.

The overall water transport is determined by physical properties of a membrane: porosity, tortuosity, and size of ionic domains, which in turn are collectively determined by the chemical structure of the PFSA polymer and the membrane processing method (21, 30, 42, 43). Changing EW and polymer side-chain architecture of PEMs resulted in different porosities, different degrees of water uptake for each of the membranes (SI Appendix, Table S2), and varying fractional amounts of bulkW and nonbulkW (SI Appendix, Fig. S15). Excluding N212, which was the only membrane processed via drop casting, the total ADC(OH) of the extruded membranes (E87, E98, N1035, and N117) was linearly related to the ADC(nonbulkW), despite the different polymer chemistries and fractional contributions of bulkW and nonbulkW in the different membranes (Fig. 3A). This result demonstrates that the overall water transport can be tuned by changing the diffusivity of nonbulkW; the same is not true for bulkW (SI Appendix, Fig. S16A).

Fig. 3.

Linear correlations between ADC(OH) and (A) ADC(nonbulkW), (B) through-plane proton conductivity; (C) linear correlation between ADC(nonbulkW) and through-plane proton conductivity. Black, extruded Aquivion membranes with short side chains; red, extruded Nafion-based membranes with long side chains.

To determine how water transport is related to the functional property of PFSAs: proton conductivity, we measured the through-plane conductivity of the membranes in the fully hydrated state (analogous to our water transport measurements) using AC impedance spectroscopy with a custom-built cell (see SI Appendix, Figs. S17–S20 and Table S3 for experimental details, raw impedance data, and model fits). Fig. 3B shows the through-plane proton conductivity versus the total ADC(OH) for each PEM and demonstrates that these variables are clearly correlated––again surprisingly independent of the differences in membrane chemistry. Faster water diffusivity facilitates increased proton conductivity, and, because of the linear correlation between ADC(OH) and ADC(nonbulkW), the diffusivity of nonbulkW appears to be fundamentally linked to through-plane proton conductivity in PFSAs.

Discussion

Time-lapsed vibrational spectra during H₂O–D₂O exchange were used to quantify the water and water subspecies diffusivity by analysis of the change of the OH-stretching vibration. The overall water diffusivity in PFSA membranes was found to be ∼10-fold slower than water transport in bulkW, consistent with previous measurements of water diffusivity in PFSA membranes determined by other methods (SI Appendix, Table S1) (19, 44–47). The 10-fold discrepancy between the overall water diffusivity in membranes versus water diffusivity in water is likely to occur because of the physical barriers in the material: the tortuosity (or connectivity) and constrictions of ionic domains (48).

Looking closely at fractional concentrations of water subspecies during the H₂O–D₂O exchange experiments (Fig. 2A and SI Appendix, Fig. S13 E–H), one sees that the [bulkW]/[nonbulkW] ratio was substantially reduced compared with the infinite-time (equilibrium) situation for all PFSA membranes measured. Such a long-lived nonequilibrium (5–7 s in Fig. 2A) suggests a physical separation of nonbulkW and bulkW. If the two types of water coexisted within the same ionic domain, one would expect exchange dynamics between these populations after more than 100–1,000 ps (18). This would lead to instantaneous equilibration on our (millisecond) measurement timescale and effectively indistinguishable ADC for both bulkW and nonbulkW, which is not observed experimentally (Fig. 2B). Further support for physical separation of bulkW and nonbulkW is provided by the accurate calculation of the ADC(OH) by ADC(OH)′––a weighted linear combination of ADC(nonbulkW) and ADC(bulkW) by the amount of each subspecies in each respective membrane, which indicates that the subspecies transit the membrane as essentially independent entities.

The finding that nonbulkW exhibits faster transport than bulkW in all investigated PFSAs (Fig. 2B) is qualitatively similar to that seen previously (even in Nafion membranes), showing that “slippery” water exhibits faster diffusivity than bulkW (49, 50). Closer inspection of the nonbulkW spectral shape shows that, indeed, it exhibits slippery features. The main peak is strongly blue-shifted and narrower relative to that of bulkW (Fig. 2A, blue and orange spectra), demonstrating it is clearly less strongly hydrogen bonded (and less coupled) to other water molecules compared with bulkW (51, 52). Together with the previous argument that nonbulkW is physically separate from bulkW, we propose that nonbulkW is the dominant species in smaller ionic channels, where the ability for water molecules to hydrogen bond to other water molecules is reduced, while bulkW is dominant in larger channels or clusters (Fig. 4). This assignment is consistent with coarse-grained, Monte Carlo dissipative particle dynamics simulations (53) that quantitatively reproduced the measured water diffusion coefficients and work by Feng et al. (31) showing diffusion rates strongly depend on confinement geometry.

Fig. 4.

Model for ionic domains in PFSAs showing how nonbulkW and bulkW can contribute to overall water transport and still exhibit distinct diffusivities. (A) Smaller ionic water channels contain only nonbulkW (orange) and larger ionic water clusters and channels are dominated by bulkW (blue). (B, Bottom) Schematic showing the structure and binding of water in large versus small ionic water domains. BulkW (blue) is the primary species in larger domains while smaller domains contain mainly nonbulkW (orange). BulkW maintains interactions characteristic of bulkW while nonbulkW forms a less-coordinated water network confined by the polymer (gray and black) in the smaller domains.

The observed faster ADC of nonbulkW could arise from two possible mechanisms: (i) the network of channels containing nonbulkW is always less tortuous than the bulkW, i.e., in all five PEMs we tested or (ii) nonbulkW, being confined in small ionic channels, diffuses by a different transport mechanism, as has been suggested for water in highly confined geometries. Water transport in confined geometries is proposed to occur via water wires (or clusters) that exhibit lower barriers to hydrogen-bond breaking and water rotation compared with bulkW due to water molecules in the wire (cluster) having fewer hydrogen bonds on average (54). The (subnanometer) diameter of narrow ionic channels in PFSAs meets the criterion to confine water on the same scale, so such a transport mechanism is certainly plausible.

The model in Fig. 4, with narrow channels containing the faster-diffusing nonbulkW, combined with the data in Fig. 3 showing correlated nonbulkW transport and proton conductivity, has potential implications for the proton transport mechanism in PFSA materials. Specifically, our results and model are consistent with the mechanism for proton transport in PFSAs proposed by Savage et al. (29): Narrow channels will facilitate passing hydrated protons from head group to head group via confined nonbulkW with high efficiency. Interestingly, the larger diffusion coefficient of water in the narrow channels could further contribute to the proton mobility by increasing the vehicular proton diffusion since vehicular proton diffusion is purportedly correlated with vehicular water diffusion. As the size of the water-containing structures shrinks in PFSA PEMs, these structures will contain more (faster-diffusing) nonbulkW, which should also enhance vehicular proton transport.

Independent of the exact mechanism responsible for accelerated transport of water and protons in confined channels, experimental evidence has demonstrated that transport of both species increases in confinement (55, 56). A thorough study where one can explicitly vary head group density [similar to that simulated by Jang et al. (57)] and channel width independently is necessary to disambiguate these effects for PFSA transport.

Conclusion

Water transport in PFSA PEMs is heterogeneous, with chemically unique water subspecies exhibiting surprisingly distinct diffusivities. By measuring transport in real time and spectrally distinguishing the water subspecies, we observed that the weaker-coordinated, interfacial water subspecies exhibited significantly larger mobility than the bulkW water in all PFSA membranes investigated, suggesting that weakly hydrogen-bonded water always diffuses faster than bulkW water in PFSA materials. The overall water diffusivity in PFSA membranes can be quantitatively described by a linear combination of the bulk-like and nonbulk-like diffusivities weighted by the corresponding fractional amount of each species. Finally, we show that the functional property of PFSA membranes––proton conductivity––is directly linked with nonbulk-like, interfacial water transport. Our work highlights the connection of the nanoscale channel architecture (channel connectivity/tortuosity) and water confinement on the transport properties of PFSA PEMs. Therefore, we believe that future PFSA designs that maximize the amount of nonbulk-like (by, e.g., producing a larger fraction of smaller ionic channels at the expense of larger domains) and maximize its diffusivity (by increasing porosity via membrane chemistry and production) will yield the highest proton conductivity PEMs.

Methods

Nafion 117 (183 μm thick, 360 g m⁻²; DuPont), Nafion 212 (50.8 μm thick, 100 g m⁻²; DuPont), Nafion 1035 (89 μm thick, 175 g m²; Chemours), Aquivion E87 – 05S (50 μm thick, 1.93 g cm³; Solvay), and Aquivion E98 – 05 (50 μm thick, 1.93 g cm³; Solvay) were used in this work. Further experimental and analytical methods for CARS spectroscopy and microscopy are described in SI Appendix, Experimental Methods.