Huu Chuong
Nguyën‡
a,
Felipe Andrés
Garcés-Pineda‡
a,
Mabel
de Fez-Febré
ab,
José Ramón
Galán-Mascarós
*ac and
Núria
López
*a
aInstitute of Chemical Research of Catalonia (ICIQ), The Barcelona Institute of Science and Technology, Av. Països Catalans 16, Tarragona 43007, Spain. E-mail: nlopez@iciq.es; jrgalan@iciq.es
bDepartament de Quimica Fisica i Inorganica, Universitat Rovira i Virgili, Marcel.lí Domingo s/n, Tarragona, E-43007, Spain
cICREA, Passeig Lluis Companys, 23, Barcelona 08010, Spain
First published on 30th January 2020
The oxygen evolution reaction (OER) is the major bottleneck to develop viable and cost-effective water electrolysis, a key process in the production of renewable fuels. Hematite, all iron α-Fe2O3, would be an ideal OER catalyst in alkaline media due to its abundance and easy processing. Despite its promising theoretical potential, it has demonstrated very poor OER activity under multiple experimental conditions, significantly worse than that of Co or Ni-based oxides. In the search for improving hematite performance, we have analysed the effect of doping with redox vs. non-redox active species (Ni or Zn). Our results indicate that Zn doping clearly outperforms Ni, commonly accepted as a preferred dopant. Zn-doped hematite exhibits catalytic performances close to the state-of-the-art for alkaline water splitting: reaching 10 mA cm−2 at just 350 mV overpotential (η) at pH 13, thus twenty times that of hematite. Such a catalytic enhancement can be traced back to a dramatic change in the reaction pathway. Incorporation of Ni, as previously suggested, decreases the energetic barrier for the OER on the available centres. In contrast, Zn facilitates the appearance of a dominant and faster alternative via a two-site reaction, where the four electron oxidation reaction starts on Fe, but is completed on Zn after thermodynamically favoured proton coupled electron transfer between adjacent metal centres. This unique behaviour is prompted by the non-redox character of Zn centres, which maintain the same charge during OER. Our results open an alternative role for dopants on oxide surfaces and provide a powerful approach for catalytic optimisation of oxides, including but not limited to highly preferred all-iron oxides.
Earth abundant transition metal oxides are excellent OER catalysts under alkaline conditions.3–5 Among them, the most efficient are typically based on Ni or Co,6 with other metals incorporated at low concentrations, or as dopants7 to enhance performance. Although the role of iron is supposed to be fundamental for many of these OER catalysts, surprisingly, all-iron oxides have exhibited, until now, significantly poor performance, appearing as the worst OER catalysts in the series.8 This is the case for α-Fe2O3 (hematite, or most commonly known as “rust”), a low cost, highly stable, and non-toxic oxide.9–11 Extensive studies have been done on tuning hematite's electronic structure to improve its catalytic activity on water splitting. Among them, doping with metal12–17 and non-metal18–20 impurities has been specially addressed. In general, the incorporation of transition metals has been regarded as an effective approach to enhance the catalytic activity of α-Fe2O3.21–24 Unfortunately, results on hematite have been modest, when compared with the state-of-the-art. The best hematite-based OER electrocatalyst was obtained in 98%-hematite nano-sphericons doped with atomically dispersed Ce and Ni/NiO nanoparticles.25 Significantly, the introduction of close-shell (non-redox) active dopants has been overlooked in all these previous electrocatalytic studies.
In contrast, non-redox doping has been studied in detail in the corresponding hematite-based photoelectrodes. α-Fe2O3 is a semiconductor with a band gap of 2.0–2.2 eV, well-aligned for overall water splitting, with a theoretical conversion efficiency of 12%. Despite its promising features, experimental studies delivered moderate results due to its poor conductivity and high electron–hole recombination rates and to the large OER overpotential needed to carry out water oxidation on a hematite surface.9–11 Doping with non-redox cations such as Zn2+, Sn4+ or Ti4+ has been used to facilitate charge transport, enhancing electrical conductivity and photoelectrochemical performance. Zn2+ doping, for instance, transforms the weakly n-type hematite into a p-type semiconductor, as required for a successful photocathode.26,27 In comparison, Zn doping should be detrimental in the case of a photoanode. Interestingly, very recent studies have reported that, although anodic photocurrent decreases upon Zn doping when compared with a more suitable n-type dopant,28 Zn doping results in a significant decrease in the OER onset overpotential, even in the dark.29 From these surprising results, Ferapontova et al. suggested that Zn doping was directly enhancing the water oxidation catalysis on hematite surfaces, but the origin of this effect was not identified.30
This is precisely the aim of the present work, where we report a systematic study on OER catalysis promoted by α-Fe2O3, as modified with redox and non-redox active aliovalent dopants. In pure catalytic studies (in the dark), our results indicate that non-redox dopants promote a superior OER activity, thanks to the appearance of an alternative and faster multi-centre OER mechanism, only accessible upon the presence of such close-shell centres on the oxide surface. These findings obtained from the combination of experimental and computational data appear to be valid for other transition metal oxides as OER catalysts, opening alternative optimisation strategies into the complex and crucial problem of water splitting catalysis.
Then, we tested the OER activity by linear sweep voltammetry (LSV) in 0.1 M KOH solution (pH 13) (see Fig. 2a and S2†). The (Fe2O3:Ni) electrode activity in basic media increases with Ni doping, starting at very low Ni concentrations (<1%), until reaching a maximum OER activity at 3% doping, obtaining a current density of 5.6 mA cm−2 at 1.57 V vs. RHE when, under the same conditions, hematite only delivers 0.5 mA cm−2. Above 3% doping is not beneficial, and performance is essentially constant up to 8% doping (Fig. 2b).
Fig. 2 (a) Ni and Zn doped hematite current density vs. overpotential. (b, left) Current density vs. metal concentration taken at 350 mV overpotential and (b, right) most stable structures from DFT (from top to bottom) 5% Ni:Fe2O3, 10% Ni:Fe2O3, 5% Zn:Fe2O3, and 10% ZnFe2O3. Notice that Ni dopants sit in lower layers when concentration is increased but Zn dopants always prefer to be on top. The same color code as in Fig. 1. |
On the other hand, the activity of Zn-doped hematite (Fe2O3:Zn) follows a different trend. Very low Zn concentration shows a negligible effect on catalytic performance. However, above 4% Zn, OER activity progressively improves reaching a current density over 9 mA cm−2 at 1.57 V vs. RHE (350 mV overpotential) at maximum doping. This is approximately twenty times higher than that of undoped hematite α-Fe2O3, and significantly higher also in comparison with the Ni-doped samples.
In order to determine if the apparently different catalytic activity could be due to physicochemical modifications of the material upon doping, and not directly related to an enhanced catalytic activity, we carried out several additional characterisation experiments. First, it is important to note that all electrochemical measurements were iR-compensated, so conductivity improvement could not explain the OER current boost. We also looked into the effect of doping on the preferential growth of active crystalline faces. Analysis of the XRD data indicates that only Ni doping induces some variations in the texture of the crystallites, whereas Zn doping has no effect (Fig. S3†). We also performed nitrogen adsorption/desorption analysis that confirms that the exposed surface area is barely affected by doping (Fig. S4†). The same conclusion was found from double layer capacitance data for all materials. Capacitance, proportional to the surface density of active sites, shows very minor changes and follows the same trend upon doping for both, Ni and Zn (Fig. S5†). Finally, although ZnO is highly soluble in alkaline media, we did not find significant Zn leaching during OER experiments with Zn:Fe2O3. Less than 0.05% of the total Zn content (Table S1†) was found in the electrolyte solution after 2 hour water electrolysis at 10 mA cm−2 in KOH (0.1 M). This small Zn leaching was also observed in the XPS analysis, where the Zn+2/Fe+3 ratio decreases slightly after electrochemical performance (Fig. S6†). This negligible leaching evidences that the incorporated Zn in the hematite structure is chemically stable. In summary, all data indicate that indirect effects, such as improved charge transport or larger surface area, cannot justify the enhanced electrocatalytic performance observed, suggesting that the origin should be associated with an improved catalytic mechanism.
This is also hinted by the Tafel plots (Fig. S7†). Upon Ni doping, the Tafel slope remains almost constant at 36 ± 2 mV dec−1. In contrast, upon Zn doping, the slope drops very fast, down to 26 ± 2 mV dec−1. Since the Tafel slope is directly related to the rate limiting step during the reaction, this change suggests a change in the reaction mechanism upon Zn doping.
To first explain the behaviour observed in Fig. 2b, we have computed the stability of the structure doped with Ni and Fe. Our results from DFT show that the first Ni dopants prefer to be on the top layer, making them available to participate in the catalytic process. However, as the dopant concentration increases, the next dopants prefer to be on the second layer, with a detrimental effect on OER catalysis (Table S7†). For the Zn doping, our DFR results show that Zn always prefers to stay on the top layer up to even 10% doping. These results are in accordance with the experimental measurements.
The aliovalent doped hematite was represented by a slab of the lowest energy surface (0001) and a (2 × 2) supercell containing 5% (Zn and Ni) and 10% as an approximation for the 8% Zn-doped hematite. Initially, the dopants replace Fe ions in the lattice and an extensive configurational search (1466 structures)41 to localise the lowest energy distribution was performed with PBE+U. An exact 8% computational doping would have increased the number of possible configurations to sample and requires enormous supercells; therefore the 10% was considered instead (see the ESI and Fig. S8† for more details).
The effect of the dopants in the mechanism of electrocatalytic water oxidation on hematite electrodes can be very complex.42 There are two main OER mechanisms in the literature: water nucleophilic attack (WNA) and interaction of two M–O units (I2M).43–45 However, ab initio molecular dynamics simulations have shown that the WNA is energetically more favorable on the (0001) surface of hematite.46 Rossmeisl et al.47,48 showed that alkaline and acidic models for the OER have equivalent intermediates during the reaction cycles and mostly the resting state of the surface under reaction conditions is different.49 In alkaline solutions, the WNA is a four-step process48 with the following reaction network:
OH− + * ⇌ *OH + e− | (1) |
*OH + OH− ⇌ *O + H2O(l) + e− | (2) |
*O + OH− ⇌ *OOH + e− | (3) |
*OOH + OH− ⇌ * + O2(g) + H2O(l) + e− | (4) |
For each step involving the (*OH, *O, and *OOH) intermediates we computed the corresponding free energy ΔG using the computational hydrogen electrode.49–51 The OER potential UOER = max[ΔG1, ΔG2, ΔG3, ΔG4] was determined for the pure Fe2O3 system, and Fe2O3:M (M = Ni, Zn and 5–10%). The simplest OER path would be one in which all the intermediates interact with only one type of active site (either Fe or the dopant M).
Our results in Table 1 show that the second process, i.e. the oxidation of the *OH species to *O with the release of a proton and an electron (ΔG2), is a very energy demanding step in all cases (Fig. S9†). The Ni-doped system would exhibit the best OER activity at any doping level if no other paths were possible, despite (i) appearing to be the worst for ΔG1 and ΔG4, and (ii) the Zn-doped system being the best for ΔG3. In the same line, the limited effect of Ni doping that reaches a maximum at just 3% is consistent with the DFT model, since the first Ni dopants prefer to be on the surface, but additional Ni atoms go preferably into the lower layer when the concentration increases (see Table S7†). Thus, increasing the doping level beyond a given threshold may not improve the activity.
Thus, the superior effect on catalytic performance observed for the Zn-doped system must arise from a different mechanism. One plausible possibility is the appearance of a multi-center model, where the initial reaction steps take place on the Fe sites, but with the reaction concluding on an adjacent Zn site. In this alternative model, the OER starts on an iron site for steps 1 and 2, reaching the Fe–*O state, previous to the nucleophilic attack (Fig. 3b). Then, the reaction continues on an adjacent Zn center, with steps 3 and 4, as compiled in Fig. 3a. This alternative appears to be thermodynamically even more favorable, once the crucial PCET step between both sites is allowed.
We have computed the possibility to transfer the Fe–*O active species into Zn–*O to link the two paths in a bifunctional mechanism.51 The reaction can proceed through an O–Fe–O–Zn–OH substructure, where the H from the hydroxyl is transferred to the Fe–O moiety via effective proton-coupled electron transfer (PCET) between both metal centres.52 The barrier for such a process is only 0.33 eV. As suggested by the remaining electron density difference on the oxygen neighbours in the inset of Fig. 3a, the proton moves in the solution, assisted by a water molecule53 while the electron moves through the surface. Thus, the PCET step effectively moves the reaction site from Fe to Zn, regenerating the Fe–OH into its resting state, and generating a Zn–*O site, that offers a more thermodynamically favored pathway for the two remaining steps of the OER. In contrast, after the second oxidation towards the OER, an intermediate PCET process does not generate an analogous Ni–*O site. The most favoured pathway ends up in a non-oxyradical Ni = O configuration. This is demonstrated in Fig. 4 that shows how the residual charge resides either on the oxygen neighbour of Zn (a), or on the Ni center (b), leaving low radical character for the terminal oxygen in the latter case (see Fig. S10 and S11†). The results point out that the non-redox nature of the divalent dopant plays a major role in improving the electrocatalytic performance. The reason is that close-shell Zn2+ cannot accommodate/release any further electrons. Thus, the oxo M–*O structure has a more robust radical nature thus making the lifetime of the radical species longer and more likely to have a water nucleophilic attack. This aspect was overlooked in previous proposals of multiple-site mechanisms for the OER, where the dopant was improving the activity of the original atom.22,54,55 It is also worthy to mention that our DFT calculations confirm that Zn dopants preferentially sit on the surface even at high concentrations (see Fig. S8†), also supporting the different Ni vs. Zn doping effects. This agrees with the XRD analyses that showed how Ni-doping affects the crystal structure (Fig. S3†).
We correlated the XPS data for the O 1s signal in Fig. 4c with the electrocatalytic performance in Fig. 2. For the Ni-doped system, the current density increases when the area of the O 1s signal increases. The total area of the O 1s XPS signal can be directly related to oxygen coverage, as a good indicator of surface area. This observation concludes that a larger surface area enhances the electrocatalytic performance. However, the trend is the opposite for the Zn-doped system. The increased activity correlates with a decrease in the O 1s signal. A simple model for PCET using 2D random walk indicates that the probability for PCET decreases with the surface area, for a given number of exposed Zn (see the ESI† in Section 2D; random walk for PCET for the mathematical development and other details).
Fig. 5 Volcano plot for the overpotential as a function of the energy difference ΔG*O − ΔG*OH computed without ZPE or entropic correction. The gray dotted line corresponds to the top volcano for redox doped hematite in agreement with the previous literature.24 The color code is the same as that in Fig. 1. The rectangle zones correspond to experimental measurements for Zn:Fe2O3 in green and Ni:Fe2O3 in orange. Underlined names correspond to the site where the OER occurs. A circle corresponds to a Fe site and a square to a dopant site. Both sites underlined or a triangle means PCET. The inset shows the linear scaling between the binding energies of *OOH as a function of *OH. The highest dotted line corresponds to the redox doped hematite and presents an offset of +3.2 eV. The solid line corresponds to an offset of +2.8 eV for the non-redox dopants. The lower dashed line corresponds to the ideal catalyst. All lines have a slope of 1. |
When the LSR are converted into the volcano plots (main Fig. 5), the overpotential (y-axis) is written in terms of the energy difference of *O and *OH intermediates (x-axis). Similar approaches combining experiments and theory have been reported for other systems in ref. 57–60. The redox dopants follow the dotted volcano, similar to that of Carter24 with Ni being the closest to the top. However, the non-redox volcano, shown by the continuous line, presents a higher top and thus a lower overpotential. When the *O–*OH energies of the combined Zn–Fe sites are considered (triangles in Fig. 5) the points are close to the top of the volcano with a value of 350 mV. Remarkably, the observed experimental values lie in the same range as the DFT prediction. The estimated binding energies also agree with the computed ones (Fig. S22†). Finally, as described before, the different doping-dependence observed, with a maximum for Ni (3%) and increasing from Zn, due to the ability of Zn to be accumulated on the surface while Ni subsurface positions are preferred at high loadings. Also Ferapontova et al. found better enhancement in photocatalysis at higher Zn dopant contents, showing that the onset of the photocatalytic phenomena might be linked to the active catalytic role in the OER.30
The combination of experimental and theoretical investigations on the electrocatalytic performance, active surface area and morphology indicates that the improved OER catalytic properties of Zn-doped hematite should be linked to an alternative mechanistic pathway, including a thermodynamically favoured PCET process. The direct participation of the non-redox Zn centres as active sites during the O–O bond formation allows a higher top volcano to be reached for oxides that could not be achieved without it. This mechanism involving two sites is not suitable for redox-active dopants, like Ni, since the PCET does not activate a more favorable pathway, as Zn does. Until now, high-valent metal-oxo centres were exclusively regarded as the key to the high OER catalytic activity of metal oxides.66 Thus, this novel strategy opens interesting possibilities and could inspire new dopant strategies for enhancing the activity of other transition metal oxides for OER electrocatalysis, since Zn-doping has not been carefully studied in common binary/ternary metal oxides for OER electrocatalysis.67,68 Very recently, the positive effect of Zn-doping on OER activity in alkaline media has been independently reported for the case of NiFeOxHy.69 Of course, optimum doping levels and mechanistic considerations will depend on the structure and electronics of other host matrices. Such studies are in progress.
Fe2O3 and doped Fe2O3 were obtained using the combustion method.31 A starting aqueous solution of 10 g of Fe(NO3)3·9H2O and 0.94 g of glycine in 150 mL of deionised water was prepared. This was followed by the addition of the dopants NiCl2·6H2O and ZnCl2 changing the doping concentration to 1.5%, 3%, 5%, 6.5% and 8%. After mixing the precursors with glycine, the solution was homogenised with a magnetic stirrer until total dissolution. Afterwards, the solution was heated up to 200 °C until the solvent was totally evaporated and glycine combusted. This flamy combustion process was accompanied by vigorous emission of gases (CO2, N2 and water vapor). The resulting porous dark solid was recovered and mechanically milled in an agate ball mill for 15 minutes at 25 Hz.
A 877 Titrino Plus pH-probe (Metrohm) was used to measure the experimental pH for each measurement. The pH value was used to calculate the thermodynamic water oxidation potential by employing the Nernst equation:
(5) |
The overpotential (η) was calculated by subtracting the thermodynamic water oxidation potential from the applied experimental potential (Eapp).
(6) |
All potentials reported in this manuscript were converted to the NHE reference scale using E(NHE) = E(Hg/HgO) + 0.140 V. Unless otherwise stated, the solution electrolyte used for all electrochemical tests was 0.1 M KOH (pH 13). Before any electrochemical measurement, a current interrupt (0.5 mA) was applied to each electrode set-up, with a frequency of 0.2 s ten times to measure the Ohmic drop. An iR compensation of 80 ± 10 Ω was found to be the average for all electrodes, independent of the doping level. Before linear sweep voltammetry (LSV) experiments, a break-in protocol was also applied: N2 was bubbled through the electrolyte for 15–30 min to remove O2, and then inert gas was supplied above the electrolyte. The potential was cycled between 1.3 V vs. RHE and 1.7 V RHE at 75 mV s−1 until successive measurements were stable and reproducible. To collect LSV data, O2 was bubbled for 30 min (until OCP stabilisation). Then, LSV curves were measured at 1 mV s−1 starting from OCP to 1.7 V vs. RHE and back to 1.3 V vs. RHE. The LSV measurements were repeated until successive cycles showed good reproducibility (typically 2–3 cycles).
Footnotes |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c9sc05669f |
‡ These authors contributed equally. |
This journal is © The Royal Society of Chemistry 2020 |