Nathaniel
Leonard‡
a,
Wen
Ju‡
a,
Ilya
Sinev
b,
Julian
Steinberg
a,
Fang
Luo
a,
Ana Sofia
Varela
c,
Beatriz
Roldan Cuenya
*bd and
Peter
Strasser
*a
aThe Electrochemical Energy, Catalysis, and Materials Science Laboratory, Department of Chemistry, Chemical Engineering Division, Technical University Berlin, Berlin, Germany. E-mail: pstrasser@tu-berlin.de
bDepartment of Physics, Ruhr Universität Bochum, Bochum, Germany
cInstitute of Chemistry, National Autonomous University of Mexico, Mexico City, Mexico
dDepartment of Interface Science, Fritz-Haber Institute of the Max Planck Society, Berlin, Germany. E-mail: roldan@fhi-berlin.mpg.de
First published on 3rd May 2018
We report novel structure–activity relationships and explore the chemical state and structure of catalytically active sites under operando conditions during the electrochemical CO2 reduction reaction (CO2RR) catalyzed by a series of porous iron–nitrogen–carbon (FeNC) catalysts. The FeNC catalysts were synthesized from different nitrogen precursors and, as a result of this, exhibited quite distinct physical properties, such as BET surface areas and distinct chemical N-functionalities in varying ratios. The chemical diversity of the FeNC catalysts was harnessed to set up correlations between the catalytic CO2RR activity and their chemical nitrogen-functionalities, which provided a deeper understanding between catalyst chemistry and function. XPS measurements revealed a dominant role of porphyrin-like Fe–Nx motifs and pyridinic nitrogen species in catalyzing the overall reaction process. Operando EXAFS measurements revealed an unexpected change in the Fe oxidation state and associated coordination from Fe2+ to Fe1+. This redox change coincides with the onset of catalytic CH4 production around −0.9 VRHE. The ability of the solid state coordinative Fe1+–Nx moiety to form hydrocarbons from CO2 is remarkable, as it represents the solid-state analogue to molecular Fe1+ coordination compounds with the same catalytic capability under homogeneous catalytic environments. This finding highlights a conceptual bridge between heterogeneous and homogenous catalysis and contributes significantly to our fundamental understanding of the FeNC catalyst function in the CO2RR.
While the investigation of the CO2RR on solid-state single-site MNC catalysts is a new and emerging research direction, the CO2RR has been extensively studied on molecular porphyrin-containing complexes. The complexing of transition metals in nitrogen-containing frameworks for CO2RR dates back to the 70 s, when Meshitsuka et al. observed CO2RR over Co phthalocyanine.14 In 1980, Fisher and Eisenberg measured CO current efficiencies of over 50% on various Ni and Co macrocycles.15 More recently, Zhang et al. have measured CO efficiencies of 90% and partial current densities of around 9 mA cm−2 on Co phthalocyanine supported on carbon nanotubes.16 Although these macrocycles do not contain precious metals, it cannot be said that they are cheap. An important development step was the production of similar materials by the heat treatment of separate metal and nitrogen precursors.17 Initially developed for oxygen reduction, these metal–nitrogen–carbon (MNC) catalysts provide a less expensive alternative to metal macrocycles.12,17–23 The present work builds on these earth-abundant MNC catalysts by improving performance and increasing the scientific community's understanding of these active and stable catalytic materials.
Many aspects of MNC catalysts have already been discussed in the literature. Often this discussion has centered around the MNC catalysts for the oxygen reduction reaction (ORR), and one of the more contentious debates has concerned the identification of the active sites. Active site identification has been a particularly problematic subject in the literature due to the high porosity and diverse surface chemistry of MNC catalysts. For ORR, a number of active sites have been postulated including pyridinic nitrogen,24–33 metal centers coordinated with nitrogen atoms,13,24–27,34–36 and metallic particles protected by layers of carbon.37,38 Although much of the work has concerned ORR, we can still adopted some ideas from the literature of the possible validity of these active sites for CO2RR. With reference to the first suspect, pyridinic nitrogen, Wu et al. showed CO2RR on various nitrogen-doped carbon foams leading to an increased CO production with increasing pyridinic nitrogen content. Additionally, their DFT calculations indicated that pyridinic species could be highly active sites for CO2RR.39 Considering metal center sites, a variety of nitrogen coordinations have been postulated for ORR, from relatively simple MeN2 or MeN4 sites, to double metal centers such as Me2N5.34,35 The evidence for these sites' activities towards CO2RR is twofold. First, the work on metal macrocycles mentioned previously would indicate that these similar MeNx structures should also be active.14,15 In a recent study done by Ju and Bagger, the Metal–N4 motif was proposed as the active site for the CO2RR process.40 The second piece of evidence is the higher activity of metal-containing catalysts in comparison to metal free.12 The last active site possibility, carbon encapsulated nanoparticles, has been investigated for oxygen reduction reaction.37,38 With regard to CO2RR, the metallic content has been thought to be primarily involved in the competing H2 generation.41,42 Thus, achieving high M–Nx active site density is still challenging by simply using excessive metal precursors, since this rather enhances the formation of metallic/inorganic particles, primarily producing the unwanted H2.41
In the present work, a number of FeNC catalysts will be explored that were prepared by polymerizing aniline around a carbon support in the presence of an iron salt.12,19–22 The heat treatment and subsequent acid washing to remove the excess metal content result in an active CO2RR catalyst.12 However, here we consider FeNC catalysts made in the presence of a variety of additional nitrogen precursors, in part to create a diversity of surface chemistries, in part to control the surface active site such as M–Nx and pyridinic N. These secondary chemical N-sources make the catalysts not only catalytically more active, but they allow a variation of the physico-chemical properties of the resulting FeNC materials. Thanks to their varying surface chemistries we uncover previously unexplored CO2RR activity–property relationships and offer new insight in the identity of the active catalyst structure.18,19 Particular emphasis is given to the chemical state and the structure of the catalytic Fe center during the CO2RR catalysis. To get this information, we apply operando X-ray absorption spectroscopy and uncover an unusual, yet catalytically very selective Fe(I) oxidation state, which is probably responsible for the formation of CH4.
For the synthesis, 3 ml of aniline are added into 0.5 liter of 1 M HCl along with 5 g FeCl3 and 5 g ammonium persulfate. At this point a secondary nitrogen precursor was added if desired. For the control catalyst (CTRL) no secondary nitrogen precursor was added. The quantity of the secondary precursor is calculated to add 0.333 moles of nitrogen. Various secondary nitrogen precursors were chosen to represent common nitrogen precursors19,43–50 with varying size and nitrogen contents as summarized in Table S1 in the ESI.† This resulted in 7 g melamine (MEL) or cyanamide (CM), 10 g urea (UREA), or 23.6 g nicarbazin (NCB). After one hour of stirring, 0.4 g of pretreated carbon was added. This pretreated carbon has been ultrasonically dispersed in 50 ml of DI water. The resulting mixture was stirred for 48 hours and then dried. After drying, the mixture was ball-milled and heat treated with a ramp of 30 °C per minute to 900 °C and kept at this temperature for one hour in a nitrogen atmosphere. After heat treatment, the material was refluxed in 2 M H2SO4 overnight and rinsed to neutral pH via vacuum filtration. After this acid wash, a second identical heat treatment was performed. At least a second acid wash and third heat treatment were performed on each sample. After this third heat treatment, XRD was used to determine whether the sample had been cleaned of excess residual Fe (usually in the form of FeS). If the sample is not clean, a third acid wash and fourth heat treatments were performed (this was the case for CTRL and NCB). These materials have also been explored as oxygen reduction catalysts in a related paper.
The resulting electrode was inserted into a CO2-saturated, 0.1 M KHCO3 solution. The electrochemical reduction of CO2 was carried out in a two compartment cell divided by a polymer electrolyte membrane (NR212). The electrochemical data were acquired using a SP-200 potentiostat (Biologic). Cyclic voltammetry was firstly carried out on the Fe–N–C catalysts in CO2-saturated 0.1 M KHCO3 at various scan rates (20 mV s−1, 15 mV s−1, 10 mV s−1, 5 mV s−1 and 1 mV s−1) to estimate the double layer (DL) capacitance, which is usually thought to be proportional to the electrochemical surface area (ECSA). The potential cycling was performed between −0.1 and 0.42 VRHE to avoid the faradaic process (Fig. S6†). By extracting the double layer current densities at +0.16 V vs. RHE (middle of the E scanning range), we were able to quantify the double layer capacitance of each catalyst under electrochemical conditions and found that it shows a perfect agreement with the BET surface area, indicating that the BETSA and ECSA are corresponding with each other (Fig. S7†).
For a CO2RR measurement, the potential is scanned at a rate of 5 mV s−1 from 0.05 vs. RHE to the working potential (between −0.5 V and −1.0 V) and then kept at this value for 60 minutes. The sample of the gas was analyzed after 15 and 60 minutes of constant potential with a Gas Chromatograph (Shimadzu GC 2014, Detectors: FID & TCD). Liquid products were measured via HPLC (Agilent 1200, Detector: RID) and a liquid GC (Shimadzu 2010 plus, Detector: FID). All potentials are corrected for ohmic resistances unless otherwise noted.
Nitrogen 1s XPS region scans are shown in Fig. 2a. All samples show a similar structure with two dominating peaks at 398.7 and 401.3 eV, indicating prevalence of pyridinic and pyrrolic nitrogen in the structure. A spectral valley between those peaks, where Nx–Fe species are reported,25 is shallower in the PANI-MEL sample, pointing to a higher concentration of Fe–porphyrin moieties in those samples. Indeed, a more detailed analysis of the N 1s regions, exemplified by the PANI MEL sample in Fig. 2b is summarized in Table 1. The N 1s spectra of Fe–protoporphyrin from Sigma-Aldrich is shown in Fig. S9.† Fe 2p spectra of all PANI samples show a similar structure with Fe 2p3/2 having an intense peak centered at 711 eV and a weak satellite observed around 715.5 eV (Fig. S10†). The structure detected is similar to the shape of Fe 2p3/2 previously reported for ferrous oxide (FeO).56,57 It is noteworthy that there are no hints of Fe–N moieties reported at 708 eV (Fig. S11†).58 Altogether, the analysis of the Fe 2p3/2 spectra indicates that the most iron seen by XPS is in the form of oxidic Fe(II), while Fe–Nx species, probed indirectly in N 1s spectra (Fig. 2b and S9), must be lying either in the deeper layers or in pores, not accessible to XPS at Fe 2p due to the lower value of the corresponding inelastic mean free path of photoelectrons.59
Fig. 2 (a) High resolution N 1s XPS data of PANI samples with different N precursors. (b) Example of the deconvolution of a N 1s spectrum acquired for the MEL sample. |
Sample | Pyridinic | Nx–Fe | Pyrrolic | Graphitic | N–Ox |
---|---|---|---|---|---|
NCB | 25.0 | 12.8 | 44.1 | 13.0 | 5.1 |
CTRL | 28.4 | 12.6 | 45.9 | 10.5 | 2.6 |
CM | 33.8 | 12.7 | 42.1 | 8.6 | 2.8 |
MEL | 31.8 | 17.7 | 36.9 | 10.1 | 3.5 |
UREA | 32.7 | 12.4 | 40.3 | 10.7 | 3.9 |
Fe–PP (Sigma-Aldrich) | — | 100 | — | — | — |
As discussed previously, the discrepancy between surface and bulk iron contents (detected by ICP and XPS as shown in Fig. 1) indicates the presence of iron species not seen by surface sensitive techniques, e.g. covered by a carbon layer or isolated in pores of the support. To investigate the nature of those species, X-ray absorption spectroscopy (XAFS) measurements were carried out ex situ. X-ray absorption near edge structure (XANES) spectra of selected samples (Fig. 3a) indicate similar chemical state and coordination of iron. The spectra show a pre-edge feature at ca. 7114 eV, corresponding to a 1s → 3d electronic transition typical for Fe3+ in an octahedral local environment. An intense feature above the absorption edge, between 7126 and 7159 V (so-called white line) has however no similarities with the most common iron oxides (see Fig. S12† for comparison), but is well in line with the results published for similar materials.60,61 The EXAFS spectra plotted on Fig. 3b (Fe K-edge k2-weighted EXAFS data of Fe–PANI samples in k-space and analysis as exemplified by CM sample are shown in Fig. S13 and 14†), despite looking somewhat alike, have distinct differences in both, peak positions and intensities. The first peak, originating from a light backscatterer, is observed at 1.46 Å (uncorrected for a phase shift) in the PANI-MEL sample and shifts towards shorter distances in PANI(CTRL) and PANI-CM, 1.42 and 1.39 Å (uncorrected) correspondingly. At the same time, the peak intensity is similar in PANI-MEL and PANI-CM samples, while the PANI sample shows a considerably smaller peak. The second backscattering feature between ca. 2.0 and 3.0 Å (uncorrected) is worth special attention. Its location is somehow similar to Fe–Fe backscattering in both common iron oxides with bcc structure (Fig. S12b†), although neither matches exactly in peak position. Zitolo et al. assigned a similar structure to crystalline Fe2N, formed during pyrolysis in NH3.60 In our case, formation of nitrides was not observed by any other method. To obtain further details on the local Fe environment, the EXAFS spectra were fitted using an Fe–porphyrin structure62 as model and the results are summarized in Table 2. It is seen that the first coordination shell around Fe can be well described by the FeN4 moiety. Slight deviations from 4-fold coordination are explained by the interference with iron oxide species which were detected by our surface sensitive XPS method and should be present in the as prepared ex situ measured samples. The second next neighbor peak observed at ca. 2.4–2.5 Å (uncorrected) is indeed well described by carbon from a porphyrin structure with the real bond distance close to the reference of 3.0 Å. The corresponding coordination number however is significantly lower than 8 in crystalline porphyrin, indicating a highly disordered structure. The latter is also supported by the substantially higher Debye–Waller factors obtained for the PANI samples as compared to the reference iron protoporphyrin sample.
Fig. 3 (a) Fe K-edge XANES and (b) EXAFS spectra of selected Fe–PANI samples, dotted lines in (b) show fitted models. |
CNFe–N | r Fe–N, Å | σ 2Fe–N 10−3, Å2 | CNFe–C | r Fe–C, Å | σ 2Fe–C 10−3, Å2 |
---|---|---|---|---|---|
CM | |||||
3.8(3) | 1.98(1) | 7.5(6) | 1.3(2) | 3.0(1) | 10.3(9) |
CTRL | |||||
3.5(3) | 1.99(1) | 8.5(7) | 1.3(2) | 2.8(1) | 9.2(8) |
MEL | |||||
3.8(2) | 2.01(1) | 7.3(6) | 0.8(1) | 2.7(2) | 9.8(9) |
Fe–PP (Sigma-Aldrich) | |||||
4.0 | 1.93(1) | 4.7(6) | 8.0 | 3.2(1) | 4.6(9) |
Now that the catalysts have been described structurally and chemically, this information can be used to better understand the catalysts electrochemical performance. Fig. 4 shows the CO production rate (a) and CO faradaic efficiency (b) of the various catalysts. From the CO production rate a kinetic region can be identified by the strong increase in performance with decreasing potential between −0.45 and −0.6 V vs. RHE. At lower potentials this kinetic region gives way to a plateau with maximum production rates of over 5 mA cm−2 for the melamine and cyanamide catalysts. The lack of potential dependence indicated by this plateau suggests that the rate limiting step has shifted to some non-electrochemical process. The faradaic efficiency towards CO production, shown in Fig. 4b, shows peak faradaic efficiencies occurring around −0.6 V vs. RHE with the top performing catalysts being melamine with 85% maximum CO efficiency.
Comparing Fig. 4 with the chemical characterizations, it is evident that the addition of the secondary nitrogen precursor has impacted both chemistry and catalyst performance. In order to understand this performance better, surface area effects must be understood. There is a strong correlation between BET surface area and performance as shown in Fig. 5. This relationship is not surprising considering the fundamental role of real surface area in heterogeneous catalysis. To reach a deeper understanding of catalysts behavior, the current can be normalized to the real surface area. This will allow a comparison of specific current densities with surface chemistries obtained from XPS results.
Fig. 5 Trends of CO current density on the plateau and in the kinetic region (−0.53 V vs. RHE) varying with BET specific surface area. |
For a better understanding of the intrinsic catalytic activity of these Fe–N–C catalysts for CO2-to-CO conversion, specific current densities were calculated by normalizing the current in the kinetic region by the BET surface area, subsequently correlated to the chemical make-up of the surface as measured by XPS. The data show particularly good correlations with N–Fe species (black squares) and pyridinic nitrogen species (open circles) as shown in Fig. 6a. The N–Fe correlation suggests potentially active Fe–Nx sites, and the pyridinic nitrogen trend is in accordance with its catalytic properties hypothesized by Wu et al. (both mentioned in the Introduction).12,39 Comparing the N–Fe and N-pyridinic correlations, it can be observed that some catalysts have relatively more N–Fe (MEL and CTRL) and some relatively more N-pyridinic (UREA and CM). This observation leads to the hypothesis that both constituents contribute to the catalytic activity. For this reason, the authors also include a correlation of the specific current density with the sum of N-pyridinic and N–Fe content (black triangles). The linear fit of the summed data set shows a higher R2 than either of the other fits. This fit improvement suggests that both sites are likely active.
Fig. 6 Trends of CO current densities in the kinetic region (−0.53 V vs. RHE) varying with (a) various pyridinic nitrogen and N–Fe and (b) surface Fe content. Current density is normalized to the specific surface area as calculated by the BET method. BET-normalized CO current densities in the kinetic region as a function of other functionalities are shown in Fig. S15 and 16.† Free energy diagrams from CO2 to CO over FeNx and pyridinic-N sites are shown in Fig S17,† data which are adapted from ref. 40 and 63. |
In addition to the hypothetical nitrogen active sites, literature results have also suggested that trapped iron content near the surface may be an active site for this type of catalysts.37,41,42 For comparison, Fig. 6b shows the correlation of specific activity with total iron from XPS. The iron has a poorer correlation than either of the nitrogen constituents, suggesting that this data does not support the hypothesis that encapsulated Fe is an active site for CO2RR. This uncertainty is compounded by the fact that the Fe peak is small and hard to quantify from XPS and that the samples contained not only Fe–N species, but also FeOx species at/near the surface. The alternative theory that the metallic Fe content is a H2-generation site is also hard to prove.41,42 In this case, the melamine based catalyst would be expected to show the highest H2 faradaic efficiencies (lower CO efficiencies), but this is certainly not the case. In fact, MEL has the highest CO efficiency even though it also has the highest surface metal content.
Operando XAFS data were collected to elucidate possible changes in the oxidation state and local coordination of Fe. Fe K-edge XANES spectra taken under reaction conditions in CO2 saturated KHCO3 shown in Fig. 7a display a shift in edge position at −1.1 V vs. RHE (red trace). This shift is similar to that found on related catalysts at around +0.75 V vs. RHE under acidic conditions.64 That shift was connected to 2+/3+ active site redox behavior which had important implications concerning adsorbate bond strength.64 Similarly, it is likely that the shift observed in the operando XANES data is correlated to a 1+/2+ redox transition. This redox behavior has been observed for various iron-based macrocycles at similar potentials.65–69 These changes in active site oxidation state and coordination are also supported by changes observed in EXAFS as seen in Fig. 7b. With decreasing potential, the corresponding spectrum shows a slight decrease of both backscattering features at 1.44 Å (uncorrected), assigned to N/O, and shoulder at 2.35 Å (uncorrected), previously shown to correspond to Fe–C. Thus, the Fe–N/O coordination number decreases from 4.3 to 3.9, while the Fe–C coordination number decreases from 2.6 to 1.8 (see Tables 3 and S4† for details). The changes observed can be assigned to the reduction of surface iron oxides, detected by XPS, and increasing disorder in the material under reaction conditions.
Potential | CNFe–N | r Fe–N/Å | σ 2Fe–N 10−3, Å2 | CNFe–C | r Fe–C/Å | σ 2Fe–C 10−3, Å2 |
---|---|---|---|---|---|---|
−0.5 VRHE | 4.3(8) | 2.00(2) | 6.8(3) | 2.6(8) | 3.0(2) | 8.5(7) |
−0.9 VRHE | 4.2(8) | 2.00(2) | 7.0(4) | 2.4(5) | 3.0(2) | 8.7(7) |
−1.1 VRHE | 3.9(9) | 2.00(3) | 7.2(4) | 1.8(4) | 3.1(2) | 9.2(8) |
The redox transition could have a significant impact on binding energies and reaction mechanisms. One of the interesting mechanistic questions concerning these catalysts is the role and prevalence of CO poisoning. It has been suspected that strong CO binding is one of the inhibitors of higher CO2RR performance as well as the cause of CH4 production. For that reason, the CO2 consumption as a function of potential was considered in the hope of seeing significant changes in catalyst behavior. This was accomplished by calculating the ratio of CH4 production to total CO2 consumed as shown in Fig. 7c. Observable in this figure is a strong increase in CH4 production between −0.9 and −1.1 V vs. RHE. The correspondence of this increase with the changes in coordination and redox behavior indicated by EXAFS suggests that the binding behavior of reactants may indeed be modified during this potential change. This observation is consistent with a proposed mechanism for photochemical methane formation on a potentially similar iron-based macrocycle catalyst.68 In the proposed mechanism the Fe 1+/2+ transition plays an integral role in the conversion of CO into CH4via a formyl intermediate.68 This theory is consistent with our observations of increased CH4 production, changed oxidation state, and decreased coordination. All together, we hypothesize that these FeNx sites would provide enough room and proper binding energy for proton binding, facilitating the intermediate CHO*, which might be the rate-limiting step for CH4 formation. Despite the lack of systematic theoretical simulations, this mechanistic change has significant implications for future catalysts development. Specifically, for actives sites that are nitrogen-coordinated iron complexes, it is possible to adjust the iron center redox potential by adjusting ligand number/strength. This adjustment of iron center redox behavior or coordination number could be used to synthesize catalysts with higher CH4 yields. Conversely, it may be possible to inhibit CO-poisoning by tuning iron complexes to have lower Fe 1+/2+ redox potentials. Unfortunately, such a study is difficult on the present set of catalysts and would require a following work that looked at CO2RR on a set of Fe-macrocycles with varying metal center electron density of states. Such a study is outside the scope of this work.
Towards our second goal of increasing understanding of MNC catalysts we have three additional conclusions. Firstly, high specific surface areas are important to catalytic activity. This suggests that reaction rates are limited by adsorption and/or kinetics (i.e. surface events). Secondly, the comparison of specific current density with XPS data on the surface indicates that both N–Fe and pyridinic species are likely actives sites. Finally, operando EXAFS results indicate a reduction in metallic content between −0.9 and −1.1 V vs. RHE which corresponds with the redox potential of Fe 1+/2+. This event coincides with an increase in CH4 production, which suggests a change in active site behavior. The authors hypothesize that this change in behavior corresponds to a change in reaction mechanism that results in the onset of CH4 production.
Footnotes |
† Electronic supplementary information (ESI) available: Additional information on the secondary nitrogen precursors used as well as addition al XPS and EXAFS data is included. See DOI: 10.1039/c8sc00491a |
‡ Nathaniel Leonard and Wen Ju contribute equally. |
This journal is © The Royal Society of Chemistry 2018 |