A pyridinic Fe-N4 macrocycle models the active sites in Fe/N-doped carbon electrocatalysts

Iron- and nitrogen-doped carbon (Fe-N-C) materials are leading candidates to replace platinum catalysts for the oxygen reduction reaction (ORR) in fuel cells; however, their active site structures remain poorly understood. A leading postulate is that the iron-containing active sites exist primarily in a pyridinic Fe-N4 ligation environment, yet, molecular model catalysts generally feature pyrrolic coordination. Herein, we report a molecular pyridinic hexaazacyclophane macrocycle, (phen2N2)Fe, and compare its spectroscopic, electrochemical, and catalytic properties for ORR to a typical Fe-N-C material and prototypical pyrrolic iron macrocycles. N 1s XPS and XAS signatures for (phen2N2)Fe are remarkably similar to those of Fe-N-C. Electrochemical studies reveal that (phen2N2)Fe has a relatively high Fe(III/II) potential with a correlated ORR onset potential within 150 mV of Fe-N-C. Unlike the pyrrolic macrocycles, (phen2N2)Fe displays excellent selectivity for four-electron ORR, comparable to Fe-N-C materials. The aggregate spectroscopic and electrochemical data demonstrate that (phen2N2)Fe is a more effective model of Fe-N-C active sites relative to the pyrrolic iron macrocycles, thereby establishing a new molecular platform that can aid understanding of this important class of catalytic materials.

T he four-electron, four-proton reduction of molecular oxygen to water is the efficiency-limiting half reaction in lowtemperature fuel cells 1 . Regardless of the fuel source used, the current density output is primarily limited by the slow electron transfer kinetics of the oxygen reduction reaction (ORR) taking place at the cathode 2 . The prototypical material for catalyzing this reaction in commercial fuel cells is platinum metal (Pt) supported on carbon. However, the high cost and scarcity of Pt impedes the large-scale deployment of fuel cell devices and motivates the development of Earth-abundant electrocatalysts for oxygen reduction. These catalysts must operate at low overpotentials [3][4][5] and with high selectivity for the four-electron reduction of oxygen to water instead of the two-electron reduction process to generate hydrogen peroxide [6][7][8][9] . Since early reports of oxygen reduction catalyzed by macrocyclic first-row transition metal complexes 10,11 , there has been a global effort to develop selective and efficient ORR catalysts featuring base metal active sites 12 .
Pyrolyzed iron-and nitrogen-doped (Fe-N-C) materials are leading Earth-abundant alternatives to Pt-based ORR electrocatalysts 13 , however marked increases in catalyst performance are needed to make these materials technologically viable. Recent years have witnessed significant improvements in their performance 14 , however systematic development of these materials is hampered by limited molecular-level understanding or control of the iron active sites 15 . Fe-N-C materials are typically prepared by the high-temperature pyrolysis of finely dispersed iron salts 16 , porphyrins 17 , or phthalocyanines 18 along with a metal-organic framework (MOF) 19 or carbon-based support 20 . The uncontrolled nature of pyrolysis leads to a wide diversity of iron environments as well as extended solid iron phases in the resulting Fe-N-C materials 21,22 . This poor control, combined with the wide variability in preparative procedures, has led to longstanding uncertainty about the local structure of the iron active sites responsible for ORR, thereby impeding systematic enhancements to catalytic performance 23 .
Numerous recent studies have provided significant insight into possible Fe-N-C active site structures. While metallic iron active sites have been postulated 24,25 , mononuclear Fe-N 4 active sites are more commonly invoked to explain correlations between X-ray absorption spectroscopy (XAS) or 57 Fe Mössbauer spectroscopy with ORR activity 16 . Despite the growing consensus that Fe-N 4 sites are essential for ORR, the ligation environment of these iron sites remains uncertain 26,27 . Indeed, even though iron porphyrin and phthalocyanine complexes can be used as precursors to Fe-N-C materials, there is evidence that the core pyrrolic ligation environment of these precursors changes substantially upon pyrolysis [28][29][30][31] . In particular, the first shell Fe-N bond lengths in Fe-N-C materials have been reported to be shorter than the macrocyclic iron complex precursor used in the material synthesis 17 . Furthermore, X-ray photoelectron spectroscopy (XPS) results have pointed to the presence of metal-coordinated pyridinic nitrogen moieties as opposed to metal-bound pyrrolic nitrogens in Fe-N-C materials 32 . 57 Fe Mössbauer spectra of many iron porphyrin 33 and phthalocyanine 34 complexes differ dramatically from the main 57 Fe Mössbauer doublets assigned to the putative Fe-N 4 active sites in Fe-N-C materials [35][36][37] . In addition, atomic-resolution electron microscopy data indicate the presence of mono-dispersed iron atoms bound within the plane of graphitic carbon [38][39][40][41][42] suggesting that the ligating groups are six-membered heterocycles rather than the five-membered rings found in pyrrolic macrocycles. Based on these spectroscopic and imaging results, there is a growing body of evidence that the Fe-N 4 sites in Fe-N-C materials are ligated by pyridinic moieties fused within graphitic sheets (Fig. 1a).
Despite substantial evidence for pyridinic Fe-N 4 sites in Fe-N-C materials, nearly all molecular Fe-based ORR catalysts that are used to model the iron-containing active sites of Fe-N-C are macrocyclic complexes that feature pyrrolic coordination environments (Fig. 1b). Notwithstanding evident differences between the spectroscopic features of Fe-N-C and pyrrolic-type macrocycles, such macrocyclic complexes are routinely used as models during the analysis of X-ray absorption data to build computational fits to the experimental Fe-N-C data in order to construct toy models of the apparent Fe-N 4 sites. Such an approach could be strengthened by the creation of a pyridinic Fe-N 4 model architecture that is better able to model the average Fe-N 4 sites in Fe-N-C relative to pyrrolic iron complexes. Owing to the inherent multiscale complexity of Fe-N-C materials, no molecular complex will ever be able to perfectly reproduce their spectroscopic signatures or catalytic function. Pyrrolic model compounds have offered valuable spectroscopic and mechanistic insights and such molecular-level studies would benefit from new model complexes that feature the pyridinic ligation motifs that are now thought to exist in Fe-N-C materials.
Herein, we report the synthesis and characterization of this pyridinic Fe-N 4 macrocyclic fragment (Fig. 1c) and compare its 57 Fe Mössbauer, XPS, and XAS features and ORR performance to those of prototypical Fe-N-C, iron octaethylporphyrin, (OEP)Fe, and iron phthalocyanine, (Pc)Fe, catalysts. Through these studies, we demonstrate that the iron coordination environment in [(phen 2 N 2 )Fe] 2 O displays greater spectroscopic similarity to Fe-N-C materials than [(OEP)Fe] 2 O and [(Pc)Fe] 2 O. Furthermore, we find that (phen 2 N 2 )FeCl displays superior catalytic activity and selectivity relative to (OEP)FeCl and (Pc)FeCl, approaching the performance metrics of Fe-N-C materials. These studies establish (phen 2 N 2 )Fe complexes as better structural and functional mimics of Fe-N 4 active sites in Fe-N-C materials than legacy pyrrolic macrocycles.

Results and discussion
Synthesis of Fe-N-C and (phen 2 N 2 )Fe complexes. A prototypical Fe-N-C material was synthesized by a combination of literature methods 19,48 . In brief, Fe-N-C was prepared by pyrolysis of a mixture of iron(II) acetate, 1,10-phenanthroline, and a zinc imidazolate MOF (ZIF-8) under a reducing atmosphere (5% H 2 in Ar). After pyrolysis, the sample was washed with 0.1 M H 2 SO 4 to remove trace Fe(0). The 57 Fe Mössbauer spectrum of this material ( Supplementary Fig. 1 The bis-phenanthroline macrocycle, (phen 2 N 2 )H 2 49 , was prepared by exposure of 2,9-dichlorophenanthroline ( Supplementary  Fig. 2) to anhydrous ammonia in a Parr reactor that was held at 280°C for 24 h and then ramped to 300°C for 12 h (Fig. 2).
Metalation with iron was achieved by treating (phen 2 N 2 )H 2 with anhydrous FeCl 3 in DMF or HMPA in the presence of excess tributylamine under an N 2 atmosphere. To remove unreacted FeCl 3 and other reaction byproducts, the crude reaction mixture was washed with copious amounts of diethyl ether and dichloromethane. Due to the extremely low solubility of the (phen 2 N 2 )FeCl complex, we found that residual Fe 2 Cl 6 •(DMF) 3 (~16%) remained in the sample despite exhaustive washing with anhydrous solvents 50 . HR-MS analysis of the material reveals a prominent peak at 440.1061 m/z and a smaller peak at 475.1023 m/z, in line with the expected masses of [(phen 2 N 2 )Fe] + (440.0473 m/z) and [(phen 2 N 2 )FeCl] + (475.0161 m/z), respectively. UV-Vis data also support formation of the metalated complex with a bathochromic shift of the ligand π → π* transitions and the appearance of shoulders on the main π → π* peak ( Supplementary Fig. 3). The 57 Fe Mössbauer spectrum of (phen 2 N 2 )FeCl ( Supplementary Fig. 1, middle and Supplementary Table 2) reveals a broad quadrupole doublet assigned to (phen 2 N 2 )FeCl in the S = 3/2 state 51-54 , which differs dramatically from the parameters for (OEP)FeCl and (Pc)FeCl (Supplementary Figs. 4 and 5 and Supplementary Table 2). This assignment is consistent with our calculations on (phen 2 N 2 )FeCl, which predict an intermediate-spin ground state and is consistent with a compressed N 4 -binding pocket that stabilizes the S = 3/2 spin state (see Supplementary Note 1 for calculated energies and geometries).
To better mimic the axial O-ligation present in the material, we Soxhleted the (phen 2 N 2 )FeCl sample with ethanol. This served to generate the μ-oxo [(phen 2 N 2 )Fe] 2 O via reaction with adventitious water and base. Similar conversions are precedented for sterically unprotected porphyrin complexes 55 . While this species is catalytically inactive in the acidic electrolyte ( Supplementary  Fig. 6 Table 2). The vast differences in the 57 Fe Mössbauer spectra of (phen 2 N 2 )Fe, (OEP)Fe, and (Pc)Fe upon exchange of the axial ligand suggest that 57 Fe Mössbauer spectroscopy alone may be insufficient to distinguish pyrrolic from pyridinic ligation at putative Fe-N 4 sites in Fe-N-C materials.
XPS reveals nitrogen and iron environments similar to Fe-N-C. The nitrogen environments present in Fe-N-C, [(phen 2 N 2 )Fe] 2 Fig. 2 Syntheses of (phen 2 N 2 )H 2 , (phen 2 N 2 )FeCl, and [(phen 2 N 2 )Fe] 2 O. The (phen 2 N 2 )H 2 macrocycle was synthesized by the reaction of 2,9dichloro-1,10-phenanthroline with anhydrous ammonia. The (phen 2 N 2 )FeCl model complex was obtained by treatment of (phen 2 N 2 )H 2 with FeCl 3 in the presence of N(n-Bu) 3  all samples, component decomposition of the N 1s manifold for Fe-N-C, and N 1s peaks, assignments, and parameters for all samples). For Fe-N-C, we observed a broad N 1s XPS signal similar to that reported for other Fe-N-C materials (Fig. 3a). The N peak envelope for Fe-N-C spans 396-407 eV and can be deconvoluted into four peaks, previously assigned to oxidized (404.0 eV), graphitic/pyrrolic (401.3 eV), metal-coordinated (399.9 eV), and pyridinic nitrogen (398.3 eV) 60,61 . We note that since this sample was prepared by the pyrolysis of ZIF-8, we expect the metal-coordinated N peak to result from both iron and zinc coordination. Notably, comparing N 1s XPS binding energies for Zn II porphyrins 62 (Fig. 3b) reveal a single peak centered at 399.3 eV that can be deconvoluted into two components centered at 398.3 and 399.6 eV with a ratio of peak area values of 1:2. Based on the integration ratio and the symmetry of the (phen 2 N 2 )Fe subunits in [(phen 2 N 2 )Fe] 2 O, we assign the 398.3 peak to the bridgehead nitrogens and the 399.6 peak to the coordinated pyridinic nitrogens. In comparison, the N 1s XPS spectrum (Supplementary Fig. 12, bottom) of (phen 2 N 2 )H 2 is dominated by two peaks corresponding to bridging and pyridinic nitrogen atoms at 397.9 and 399.5 eV, respectively, in a 1:2 ratio. This indicates that both bridging and pyridinic nitrogen atoms shift to higher binding energies upon metalation with iron. Similar changes in the N 1s spectrum have been reported for free-base porphyrins upon metalation 63 2 O is composed of a sharp peak at 398.4 eV with a satellite peak centered at 400.1 eV (Fig. 3c). Analogously, [(OEP)Fe] 2 O consists of a sharp peak at 397.6 eV with a broad satellite peak centered at 399.9 eV 64,65 (Fig. 3d). While the main peak in [(Pc)Fe] 2 O appears in the region assigned to N pyridinic for Fe-N-C, it is best fit with a single component, indicating that the coordinating and non-coordinating N atoms have comparable binding energies. Thus, the exclusive presence of phthalocyanine environments in Fe-N-C would be insufficient to account for the large metal-N x peak in Fe-N-C. Notably, N 1 s peaks for (phen 2 N 2 )FeCl, (Pc)FeCl, and (OEP)FeCl ( Supplementary Fig. 11) appear within 0.1-0.4 eV of the corresponding peaks for the μoxo compounds, indicating that the identity of the axial ligand does not impact the N 1s binding energy to the same extent as does changing the metal-coordinated nitrogen atoms from pyrrolic to pyridinic. Importantly, the N 1s peaks for the coordinated pyridinic nitrogens in the (phen 2 N 2 )Fe complexes match well with the peak corresponding to the metal-coordinated pyridinic nitrogen component (399.9 eV) of Fe-N-C, whereas the N 1s peaks of the coordinated pyrrolic nitrogens in (Pc)Fe and (OEP)Fe complexes appear at significantly lower binding energies. Together, the XPS data indicate that the ironcoordinated N environment in [(phen 2 N 2 )Fe] species are electronically very similar to the metal-coordinated N environments in Fe-N-C materials.
The iron environments present in Fe-N-C, [(phen 2 N 2 )Fe] 2   In the N 1s spectra, pyridinic and bridging nitrogens are depicted in blue, metal-coordinated nitrogens are depicted in magenta, graphitic/pyrrolic nitrogens are depicted as yellow, oxidized nitrogens are shown in purple and the π-π* satellites are shown in dark cyan. In the Fe 2p spectra, the 2p 3/2 Fe(II) peak is shown in blue and the 2p 3/2 Fe(III) peak is given in magenta. The Fe 2p 1/2 peak is given in aqua blue. Multiplet peaks for the 2p 3/2 and 2p 1/2 peaks are given in green and dark cyan, respectively.
samples, we also observe significant asymmetry on the main Fe 2p 3 Fig. 15), and that residual FeCl 3 salts are likely subsumed into these peaks 69 Fig. 16 and Supplementary  Table 8). Notably, the peak in Fe-N-C assigned to Fe(III) is 2.7 eV positive of the corresponding Fe(III) peak for [(phen 2 N 2 )Fe] 2 Table 7) as well as an unknown degree of heterogeneity in the local coordination environment of each iron, the XANES line shapes have been examined for a wide variety of pyrolyzed Fe-N-C catalysts, including Fe-N-C derived from iron pyrrolic macrocycles, and features of the XANES have been correlated to pyrolysis temperature and ORR activity 26,70,71 .
In particular, the intensity of the first main-edge peak, C, has been found to positively correlate with increased pyrolysis temperature as well as improved ORR activity, whereas the intensities of the main-edge shoulder, B, and the higher energy main-edge peak, D, have been shown to anticorrelate with both pyrolysis temperature and ORR activity (  (Fig. 4a, blue), which shows a higher intensity for C, and lower intensities for B and D. The intensity of pre-edge peak, A, has also been shown to positively correlate with ORR activity 70 . We find that the preedge peak line shape for [(phen 2 N 2 )Fe] 2 O matches closely with that of Fe-N-C, but that the peak intensity for both is lower than that for [(OEP)Fe] 2 2 O do so nine and eleven times, respectively. Furthermore, the similarity between Fe-N-C and [(phen 2 N 2 )Fe] 2 O is also reflected in the k 2 -weighted Fourier transform EXAFS radial distribution spectra (Fig. 4c) 55,75,76 . Importantly, these Fe-N bond lengths reflect not only the N-N separation in the ligand, but also the degree of pucker of the iron center out of the equatorial plane. Thus, this distance alone cannot distinguish pyridinic from pyrrolic ligation and, indeed, we see similar Fe-N vectors for [(phen 2 N 2 )Fe] 2 O and [(Pc)Fe] 2 O. Furthermore, for [(phen 2 N 2 )Fe] 2 O, the 1.97 Å Fe-N distance extracted from EXAFS modeling is substantially higher than the 1.86 Å distance predicted from the calculation if the iron were to reside in the N 4 plane for the phen 2 N 2 ligand ( Supplementary Fig. 20), further evincing that the iron center is puckered. This reasoning is also in line with a computational XAS study that showed an enhancement of the main-edge peak C (Fig. 4a, blue) as the displacement of iron from the N 4 plane was increased 70 . Consistent with the Fe-N-C material having a putative OH x axial ligand, the EXAFS fits return an Fe-O scattering path of 2.04 Å, slightly larger than the range of Fe-O scattering paths, 1.73-1.79 Å, extracted for the molecular μ-oxo complexes.
Taking the raw k 2 -weighted EXAFS oscillations, the FT-EXAFS spectra, and the degree of iron puckering into account, the XAS data highlight that, relative to (Pc)Fe and (OEP)Fe, (phen 2 N 2 )Fe is an improved model of the iron sites in Fe-N-C catalysts. We stress that the similarity in the k-space EXAFS data should not be taken to mean that (phen 2 N 2 )Fe has an identical structure to Fe-N-C, but that the aggregate scattering environment of the iron centers in (phen 2 N 2 )Fe is qualitatively similar to the average cumulative scattering environment of the iron centers in Fe-N-C. We acknowledge this resemblance could be coincidental and introduced by countervailing changes in axial and equatorial scattering interactions. Nonetheless, the data indicate that the iron centers in (phen 2 N 2 )Fe exist in a qualitatively similar average environment to the iron centers in Fe-N-C.
Fe(III/II) potentials are defined by ligating and proximal N atoms. All three model compounds as well as the Fe-N-C material were evaluated electrochemically as thin films supported on glassy carbon electrodes. In a typical preparation, Fe-N-C powders were dispersed in a 7:2:1 combination of CH 2 Cl 2 , ethanol, and 5 wt% Nafion solution (75 wt% ethanol and 20 wt% water), respectively. The resulting inks were dropcast onto electrodes and allowed to dry in air to generate a well-adhered catalyst film. A similar procedure was used for (phen 2 N 2 )FeCl, (OEP)FeCl, and (Pc)FeCl with inclusion of Vulcan carbon powder to enhance film conductivity. We expect that any residual FeCl 3 salts present in the (phen 2 N 2 )FeCl sample will readily dissolve into the acidic electrolyte and will not impact the electrochemical results. In line with this expectation, we find that FeCl 3 /Vulcan inks are inactive for ORR ( Supplementary Fig. 21).
Cyclic voltammograms of Fe-N-C, (phen 2 N 2 )FeCl, (OEP)FeCl, and (Pc)FeCl recorded in the absence of O 2 provide insight into the redox potential of the metal center. For (OEP)FeCl (Fig. 5, orange), we observe a redox feature with E 1/2 = 0.27 V (all potentials are reported vs the reversible hydrogen electrode, RHE). We note that the Fe(III/II) redox couple of (OEP)FeCl is quite broad with a peak-to-peak separation of 315 mV. We attribute this to sluggish Fe(III/II) redox self-exchange through the film, but nonetheless, we note that the observed E 1/2 value is comparable to other carbon-adsorbed iron porphyrin compounds 17,56,77,78 . For (Pc)FeCl (Fig. 5, aqua), we observe a broad redox feature with E 1/2 = 0.61 V ( Supplementary Fig. 22), similar to reported values for carbon-supported iron phthalocyanine complexes 79 . For (phen 2 N 2 )FeCl (Fig. 5, blue), we observe a reversible redox wave at 0.59 V. Based on the known Nernstian pH dependence of Fe(III/II) couples for adsorbed porphyrin complexes 80 as well as the quasi-Nernstian behavior of the redox wave for (phen 2 N 2 )Fe ( Supplementary Fig. 23), we assign this redox process to the one-electron, one-proton reduction of Fe (III)-OH to Fe(II)-OH 2 .
For Fe-N-C (Fig. 5, black), we observe a large double-layer capacitance, consistent with the high surface area of the material, and a pronounced redox wave at 0.63 V (Fig. 5). Although in situ spectroscopic data are required for an unambiguous assignment, we tentatively attribute this redox wave to proton-coupled electron transfer (PCET) reactions of edge-localized quinone and hydroxyl moieties, rather than to an Fe(III/II) redox process. Although this assignment may seem counterintuitive, we favor it for several reasons. First, quinone and hydroxyl PCET reactions are known to occur in the~0.6 V potential region on carbon powders 81 . Second, integrating the observed 0.6 V redox wave yields a charge that is 3.4 times larger than the expected charge for a 1-electron redox process of the surface exposed iron centers in the Fe-N-C material 82 . Third, the potential of the wave is comparable to the value for (Pc)Fe and (phen 2 N 2 )Fe and this is inconsistent with the expectation that the extended π framework in Fe-N-C should raise the Fe(III/II) potential 83  . We postulate that the actual Fe(III)-OH/Fe(II)-OH 2 redox process for this Fe-N-C material is obscured by the large double-layer charging background and the likely distribution of iron redox potentials. We also note that some iron redox centers may not engage in PCET reactivity and would not be expected to display a redox wave 87 . The redox potentials extracted from the model compounds provide valuable information about the factors that may contribute to the redox potentials of the iron centers in Fe-N-C materials. Despite the fact that both (Pc)Fe and (OEP)Fe have pyrrolic ligating atoms, the redox potential for (Pc)Fe is~0.3 V positive of the redox potential of (OEP)FeCl. The principle structural differences between (Pc)Fe and (OEP)Fe are the presence of N vs C atoms in the meso positions and the presence of a more extended π system in (Pc)Fe. One study has found that iron octaethyltetraazaporphyrin possesses a similar Fe(III/II) redox potential to (Pc)Fe 88 , suggesting that the presence of electronegative meso-N atoms is the principal contributor to the dramatic positive shift observed. Notably, whereas (phen 2 N 2 )FeCl possesses two non-ligating nitrogen atoms in the local environment, (Pc)FeCl has four, yet their Fe(III/II) potentials are comparable. This highlights that the less electron-donating pyridinic coordination environment in (phen 2 N 2 )FeCl is an ideal structural motif for pushing the Fe(III/II) positive. Together, the data suggest that for the iron centers in the Fe-N-C material, the effective redox potential is due, not only to the ligating heterocycles, but also the nitrogen population near the Fe center 16,89,90 . Thus, even Fe-N 4 sites with identical primary coordination spheres may display a distribution of redox properties depending on the population and position of nonligating proximal nitrogen atoms.
The observed Fe(III/II) redox waves result from only a fraction of the catalyst loaded into the dropcast film. The integrated charge in the Fe(III/II) redox waves of (phen 2 N 2 )FeCl, (OEP) FeCl, and (Pc)FeCl catalyst film corresponds to 7%, 43%, and 2% of the total catalyst loading for each compound, respectively (Supplementary Table 10). We attribute the higher electroactive percentage in (OEP)FeCl film to the improved solubility of the porphyrin catalyst, which leads to superior catalyst dispersion within the film. Nonetheless, these PCET redox potentials provide a good indication that the iron centers in (phen 2 N 2 )Fe and (Pc)Fe are more electropositive than those in (OEP)Fe.
(Phen 2 N 2 )FeCl catalyzes ORR with similar activity to (Pc)FeCl. (Phen 2 N 2 )FeCl is a potent catalyst for the ORR. In O 2 -saturated 0.1 M HClO 4 , the reversible surface redox wave at 0.59 V (Fig. 6, black) is replaced by a large catalytic wave that displays an onset potential of 0.75 V (all onset potentials correspond to a current density of -0.1 mA cm −2 and all electrochemical performance parameters are collected in Supplementary Table 11) and reaches a current plateau at~0.5 V (Fig. 6, red). This catalytic wave spans the same potential range as the reversible redox feature in the absence of O 2 suggesting that ORR catalysis is mediated by the Fe (III/II) redox process (Fig. 6, inset). We also find that the Fe center is critical for activity. The unmetalated (phen 2 N 2 )H 2 ligand has activity comparable to the carbon background (Fig. 7, green) in acidic media with an onset potential of 0.34 V (Fig. 7, magenta) and activity slightly higher than background in alkaline media with an onset potential of 0.75 V, 140 mV negative of (phen 2 N 2 ) FeCl ( Supplementary Figs. 24 and 25). Catalytic onset for (phen 2 N 2 )FeCl (Fig. 7, blue) Fig. 26b and Supplementary Table 12). This positive shift in onset potential and the corresponding increase in TOF for (phen 2 N 2 )FeCl and (Pc)FeCl relative to (OEP)FeCl is consistent with the positive shift of the corresponding Fe(III/II) redox potentials for the former compounds. These results are in line with prior work that identified scaling relationships in ORR activity among adsorbed porphyrins and phthalocyanines, and showed that catalytic activity is positively correlated with the Fe(III/II) potential 91,92 . While a detailed mechanistic analysis of the ORR profile of these catalysts is beyond the scope of this work, the observed correlation among adsorbed Fe-N 4 compounds could be rationalized by the following simplified mechanistic sequence: In this sequence, at any given E, the rate of the reaction will be given by the product, K 1 × k 2 , of the equilibrium constant, K 1 , for the Fe(III)-OH to Fe(II)-OH 2 conversion, Eq. (1), and the rate constant, k 2 , for rate-limiting O 2 binding, Eq. (2). If k 2 is roughly insensitive to the local structure of the Fe center, then the reaction rate will be predominantly gated by the Fe(III/II) redox process and the higher redox potentials will correlate with more available Fe(II) at a given E, resulting in faster catalysis. We acknowledge that the above mechanistic picture is speculative and will require more studies to validate, but it nonetheless provides a simple framework for rationalizing the observed reactivity trend. Furthermore, our data provide the first experimental evidence that changing from pyrrolic to pyridinic coordination does not dramatically alter the scaling relationship between ORR activity and Fe(III/II) potential. This provides further support for the notion that Fe-N-C materials contain pyridinic Fe-N 4 sites with sufficiently high redox potentials to mediate ORR at relatively low overpotentials.
The electrocatalytic data establish that (phen 2 N 2 )FeCl and (Pc) FeCl have superior activity relative to (OEP)FeCl, whereas both remain at a deficit to Fe-N-C. While the Fe(III/II) redox potential is difficult to discern for Fe-N-C, the positive shift of the Fe(III) peak in the XPS suggests that the iron centers in Fe-N-C are even more electropositive than those in (phen 2 N 2 )FeCl and (Pc)FeCl. This change may contribute to the higher onset potentials of 0.85 and 0.90 V observed for the Fe-N-C catalyst in acidic and alkaline media, respectively 17 . We stress that although (Pc)FeCl and (phen 2 N 2 )FeCl display comparable activity for ORR, (Pc)FeCl is an inferior structural model on the basis of the spectroscopic results presented above. In aggregate, the spectroscopic as well as electrochemical results strongly suggest that (phen 2 N 2 )FeCl is the more effective model complex for mimicking iron-containing active sites in Fe-N-C relative to pyrrolic complexes such as (Pc) FeCl. Furthermore, it follows that observing high catalytic ORR performance is insufficient evidence to establish a molecular complex as an accurate and comprehensive molecular model for the active sites in Fe-N-C materials. The deficit in activity relative to Fe-N-C may also result, in part, from weak interaction between the molecular orbitals of the model and the band states of the carbon host. Notwithstanding, the fact that a discrete pyridinic molecular complex such as (phen 2 N 2 )FeCl displays an ORR onset within 150 mV of champion Fe-N-C materials suggests enormous opportunities for rational molecular design of highly active Febased ORR catalysts based on pyridinic ligation motifs that structurally model Fe-N-C.
(Phen 2 N 2 )FeCl mirrors the selectivity trends of Fe-N-C. Furthermore, (phen 2 N 2 )FeCl displays high selectivity for the fourelectron reduction of O 2 that is comparable to Fe-N-C and greater than that of (OEP)FeCl or (Pc)FeCl. Using rotating ring disk electrode (RRDE) voltammetry (Supplementary Figs. 27 and 28) we determined the percent H 2 O 2 produced and electrons transferred as a function of applied potential for all four catalysts ( Fig. 8 and Supplementary Figs. 29 and 30). For (phen 2 N 2 )FeCl, H 2 O 2 production stabilizes rapidly, reaching values of 0.1% and 2.7% in acid and alkaline media at 0.40 V, respectively. These values correspond to 3.99 and 3.95 electrons transferred in each condition, respectively. Indeed, in both acidic and alkaline media, the selectivity of (phen 2 N 2 )FeCl in the transport-limited region surpasses that of Fe-N-C at the same potential. Further, in stark contrast to (phen 2 N 2 )FeCl, both (Pc)FeCl and (OEP)FeCl display significantly lower selectivity in acid media, generating 6.6% and 12.5% H 2 O 2 , respectively, at 0.40 V. In 0.1 M NaOH, (Pc)FeCl displays similar selectivity relative to (phen 2 N 2 )FeCl; however, both are slightly more selective than (OEP)FeCl ( Supplementary  Fig. 29). We stress that the RRDE results are best interpreted in qualitative rather than quantitative terms because apparent H 2 O 2 yields are strongly dependent on catalyst loading, film morphology, ink composition, and carbon support identity [93][94][95] . Nevertheless, these results highlight the ability of a carbon-adsorbed pyridinic Fe-N 4 coordination environment to facilitate selective O 2 -reduction pathways in both acidic and alkaline media.
Whereas Fe-N-C materials are relatively stable, we observe limited stability of the (phen 2 N 2 )FeCl catalyst with activity decaying in acidic media over the course of several slow-scan cyclic voltammograms (15-20 min). Analogous deactivations are also observed for the pyrrolic complexes and are documented in the literature [96][97][98] . These observations highlight the important role of the carbon framework in increasing the relative stability of Fe-N 4 sites in Fe-N-C materials against oxidative and protolytic decomposition induced by the acidic conditions and the presence of parasitic amounts of H 2 O 2 99 . Indeed, we posit that extending the aromatic periphery around the (phen 2 N 2 )FeCl active site could enhance stability and allow for the bottom-up synthesis of robust Fe-based ORR catalysts.
The results reported herein provide a molecular perspective on the structure and oxygen reduction reactivity of Fe-N 4 active sites in Fe-N-C materials. We have synthesized a new (phen 2 N 2 )Fe fragment and demonstrated that it is a superior molecular model for the putative Fe-N 4 active sites in Fe-N-C materials compared to exhaustively studied iron porphyrin and phthalocyanine complexes. Electrochemical studies establish that ORR activity is positively correlated to the Fe(III/II) potential and that, while (OEP)Fe has too low of a potential to mediate efficient catalysis, both phthalocyanine-and pyridine-based ligation environments in (Pc)Fe and (phen 2 N 2 )Fe can generate a positive shift of the Fe (III/II) potential and a corresponding increase in ORR activity. Among the model complexes, the (phen 2 N 2 )Fe platform is unique in displaying both high activity and near 100% selectivity for four-electron oxygen reduction in acidic media, the very characteristics found in Fe-N-C materials. Since high electrochemical performance does not necessarily imply structural similarity, our spectroscopic results provide additional discrimination. In particular, the pyridinic ligation environment in (phen 2 N 2 )Fe leads to XPS and XAS signatures that are remarkably similar to those of Fe-N-C materials and notably distinct from those in either (OEP)Fe or (Pc)Fe. Taken in aggregate, the spectroscopic similarity and the electrochemical activity and selectivity trends indicate that, relative to (OEP)Fe or (Pc)Fe, (phen 2 N 2 )Fe is a significantly improved structural and functional model of Fe-N-C materials. Since the (phen 2 N 2 )Fe core can be readily derivatized 100,101 and used as the building block for functional catalytic materials, the molecular model complex developed here provides a powerful platform with which to advance the synthesis and understanding of single-site heterogeneous electrocatalysts for critical energy conversation reactions.

Methods
General synthetic considerations. Where indicated, synthetic procedures were carried out under an inert atmosphere in a nitrogen-filled glovebox. All other synthetic transformations were carried out on the benchtop without effort to exclude dioxygen or moisture. Anhydrous 1,10-phenanthroline (99%), 1,3-dibromopropane (98%), tert-butanol (99%), phosphorus pentachloride (98%), and phthalonitrile (98%) were obtained from Alfa Aesar and used as received. Potassium tert-butoxide (97%) and phosphoryl chloride (99%) were obtained from TCI and Sigma Aldrich, respectively and used as received. Iron(III) chloride (98%) was purchased from Strem Chemicals and stored in a nitrogen-filled glovebox. Compressed anhydrous ammonia (99.995%) was purchased from Airgas. N,N-dimethylformamide was deoxygenated and dried using a Glass Contour System (SG Water USA, Nashua, NH) and stored in the glovebox over 4 Å molecular sieves. D 2 O, CDCl 3 and d-TFA were purchased from Sigma-Aldrich or Cambridge Isotope Laboratories and were used without further purification. 2, 3, 7, 8, 12, 13, 17, 18-Octaethyl-21H, 23H-porphine iron(III) chloride, (OEP)FeCl, was purchased from Frontier Scientific and used without further purification. Iron(II) phthalocyanine for use as a model complex in XAS data analysis was purchased from Alfa Aesar and sublimed at 600°C under reduced pressure in a single-zone tube furnace for 3 h. Routine NMR spectra were recorded on Varian Mercury 300, Bruker Avance III 400, and Varian Inova 500 spectrometers. Chemical shifts for 1 H and 13 C{ 1 H} spectra are reported in ppm downfield of TMS, with spectra referenced using the chemical shifts of the solvent residuals. Spectra recorded in d-TFA were referenced to a benzene solvent residual by including a small capillary of C 6 D 6 in the sample tube. UV-Visible spectra were measured in a 1-cm quartz cell on a Varian Cary 50 spectrophotometer. MALDI-TOF mass spectra were obtained using Bruker Autoflex LRF and Bruker Omniflex instruments operating in reflectron mode, and the peaks reported are the mass of the most intense peak in the isotope envelope. MALDI-TOF samples consisted of analyte mixed with 1,4-bis (5-phenyl-2-oxazolyl)benzene (POPOP) as a desorption matrix. Elemental analyses and inductively coupled plasma mass spectra (ICP-MS) were performed by Robertson Microlit Laboratories (Ledgewood, NJ).
Synthesis of Fe-N-C. Fe-N-C was synthesized by a hybrid method adapted from previously published procedures 19,48 . 0.019 g (0.11 mmol), Fe(OAc) 2 , 0.206 g (1.14 mmol), 1,10-phenanthroline, and 0.804 g (3.53 mmol) ZIF-8 MOF were mixed together in an agate mortar for 15 min. To remove excess water and oxygen, the prepyrolysis mixture was placed under vacuum for 30 min at 90°C and subsequently held under vacuum for a further 12 h at room temperature. 0.856 g of the precatalyst powder was weighed into an alumina boat which was positioned at the midpoint of an alumina tube centered in a single-zone tube furnace. The alumina tube was sealed with steel endcaps, evacuated briefly, and then continuously purged with argon. The furnace was heated to 1000°C at a ramp rate of 10°C min −1 and maintained at 1000°C for 60 min. The furnace was then allowed to return to room temperature. Subsequently, the alumina tube was evacuated briefly and then continuously purged with 5% hydrogen in argon. The furnace was ramped to 800°C at a rate of 10°C min −1 and maintained at 800°C for 30 min. The sample was then allowed to return to room temperature. The resulting crude Fe-N-C powder was then sonicated in 30 mL of 0.1 M H 2 SO 4 for 60 min. The black suspension was filtered and washed with water until the eluant pH was neutral. The sample was subsequently washed with acetone to remove excess water and the product was dried overnight at 60°C to yield 219.2 mg of Fe-N-C as a black powder. ICP-MS: 1.07% Fe. 57  Synthesis of 6,7-dihydro-5H-1,4-diazepino[1,2,3,4-Imn] [1,10]phenanthroline-4,8-diium bromide (2). Compound 2 was synthesized via alkylation of 1,10phenanthroline using an excess of 1,3-dibromopropane in refluxing toluene. 10.787 g (59.86 mmol) 1,10-phenanthroline was suspended in 105 mL toluene. The mixture was stirred and heated to 70°C and, subsequently, 28 mL (55.47 g, 275 mmol, 4.59 equiv.) 1,3-dibromopropane was added. The temperature was increased to 120°C and the mixture was refluxed for 4 h. After allowing the reaction vessel to cool, the mixture was filtered and washed with hexanes to yield 18.688 g (82%) of compound 2 as a yellow powder. The observed NMR peaks in D 2 O matched the literature 102 .
Synthesis of 6,7-dihydro-5H-1,4-diazepino[1,2,3,4-Imn] [1,10]phenanthroline-3,9-dione (3). Compound 3 was prepared using the modified workup described in the literature 103 . 5.061 g (13.24 mmol) of compound 2 was suspended in 85 mL tert-butanol and sonicated for 15 min. The mixture was then stirred while 6.026 g (53.7 mmol, 4.06 equiv.) KO t Bu was added in small portions over the course of 15 min. The reaction mixture was then heated to 40°C and stirred for 12 h. The solvent was removed by rotary evaporation and the crude material was suspended in 100 mL of water. The product was extracted with 100 mL of chloroform and 5% methanol by volume was added to the chloroform solution. The mixture was then filtered through a plug of silica on a glass frit and the silica was washed with chloroform until the eluant was colorless. The combined washings were concentrated to dryness by rotary evaporation to yield 1746 g (52%) of compound 3 as a brown solid. The observed NMR peaks in CDCl 3 matched those reported in the literature 102 .
Synthesis of 2,9-dichloro-1,10-phenanthroline (4). Under inert atmosphere, a mixture of 2.29 g (9.06 mmol) 3 and 3.77 g (18.1 mmol, 2.00 equiv.) PCl 5 was suspended in 35 mL POCl 3 . The mixture was then refluxed under inert atmosphere for 14 h. After cooling to room temperature, the solution was poured onto 600 mL of ice and the resulting mixture was neutralized with aqueous ammonium hydroxide (30%) to a pH > 8. After neutralization, the resulting precipitate was collected by filtration. The solid was washed with water and dried overnight at elevated temperature to yield 1.99 g (89%) of compound 4 as a light brown powder. The observed NMR peaks in CDCl 3 matched those given in the literature 102 .
Synthesis of (phen 2 N 2 )H 2 . Preparation of (phen 2 N 2 )H 2 has been previously described in the literature 45,49,104 , however an alternate method is presented here. 0.547 g (2.19 mmol) 2,9-dichlorophenanthroline was added to a cylindrical glass insert and enclosed at room temperature in a 100 mL Parr bomb equipped with a rupture disk. The bomb was purged with nitrogen for 10 min and was then pressurized with anhydrous ammonia at 114 psi. The bomb was then immersed in a bed of aluminum beads and heated to 280-300°C with a heating mantle for 24 h. Following additional heating for 12 h at 300°C, the bomb was allowed to return to room temperature. The residual ammonia was vented and the crude product was dissolved in a mixture of trifluoroacetic acid, acetic acid and methanol and stirred for 30 min. The mixture was neutralized and basified with 4 M NaOH and was stirred for an additional 12 h. The precipitate was filtered and sequentially washed with ethanol and chloroform until the washings were colorless. The isolated material was then dried at 60°C for 12 h. The filtrate was concentrated in vacuo and heated to 300°C under inert atmosphere for a period of 24 h in a 500 mL round bottom flask affixed to a reflux condenser. After cooling to room temperature, the solid derived from the filtrate was treated with an identical dissolution, precipitation, and washing sequence as before to yield a second crop. In total, 0.280 g (0.506 mmol, 66% combined yield) of 3bH,10bH-  (20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) mg for molecular samples and 120 mg for Fe-N-C) were prepared by mixing each sample powder with Paratone-N oil. With the exception of the (phen 2 N 2 )FeCl sample, each sample was prepared outside the glovebox. The (phen 2 N 2 )FeCl sample was prepared similarly except the sample preparation process was carried out inside a nitrogen-filled glovebox and the sample was frozen with liquid nitrogen before handling outside the glovebox. All isomer shifts are reported relative to α-Fe metal at 298 K. All data were processed, fitted, and analyzed using an in-house software package for IGOR Pro 6 (Wavemetrics, Lake Oswego, OR).
X-ray photoelectron spectroscopy. XPS experiments were performed on Ther-moFisher Aluminum K Alpha+ (ESCA) or ThermoFisher Nexsa X-ray spectrometer systems with monochromatic aluminum K α X-ray sources (1486.68 eV). Samples for analysis were prepared by distributing a small amount of Au powder (for referencing) on conductive carbon tape and spreading the analyte on top. The analyte was mixed with the Au powder to allow for simultaneous detection of analyte and the Au powder. All experiments were run with the flood gun on to prevent sample charging. All data were collected using a 400 μm, 72 W focused X-ray beam at a base pressure of 2 × 10 −7 millibar or lower. Survey scans were collected at a pass energy of 200 eV and step size of 1 eV. High-resolution scans were collected with a pass energy of 50 eV and a step size of 0.1 eV. All data were analyzed with the Thermo Avantage software package (v5.987). The Au 4f 7/2 peak arising from the Au powder was assigned an energy of 84.0 eV and used as an internal binding energy reference for all spectra. In all cases the recorded spectral data were not modified with a smoothing algorithm. High-resolution spectra were fit by application of a Shirley-type (Smart) background and Gaussian/Lorentzian line-shapes of 30% Gaussian shape. The Simplex fitting algorithm was used in all cases. XPS ratios were taken from survey spectra. We expect these values to be reliable because survey spectra were collected with a high pass energy (200 eV), relatively high step size (1 eV) and because the spectra display a high signal to noise ratio. The [(phen 2 N 2 )Fe] 2 O sample shows a substantial amount of residual chloride relative to iron. This is attributed to variation caused by the insoluble nature of (phen 2 N 2 )Fe complexes which allows small amounts of chloride salts to be carried through the metalation and Soxhlet procedures. A possible assignment in this case is a small amount of triethylammonium chloride, [Et 3 NH]Cl. For the O 1s XPS, the carbon tape gives rise to a set of background oxygen peaks including aromatic C = O (~531 eV, green), aliphatic C-O (~532 eV, blue) and aromatic C-O (~533 eV, yellow) 108 that dominates the O 1s manifold. This background results from gaps in the microcrystalline molecular film that exposes the background carbon support.
X-ray absorption spectroscopy. XAS experiments were performed at the 10-BM beamline at the Advanced Photon Source (APS) at Argonne National Laboratory. All measurements were performed at the Fe K edge (7.112 keV) in transmission mode in fast scan from 250 eV below the edge to 550 eV above the edge. Samples were pressed into a stainless-steel sample holder. For catalysts containing 1 wt% Fe, 12 mg was loaded. For catalysts containing 8-10 wt% Fe,~1 mg diluted in boron nitride to reach a total of~10 mg was loaded. The sample holder was placed in a leak-tight sample cell with gas flow capabilities and purged with He at room temperature. The cell was sealed and transferred to the beamline to be scanned.
The data were interpreted using WinXAS 3.1 software to find the coordination number (CN) and bond distance (R) using standard procedures. The phase and amplitude functions for Fe-N and Fe-O were extracted from theoretical Feff6 calculations. Theoretical phase and amplitude files were created for the Fe-N (CN = 1, R = 1.94 Å) and Fe-O (CN = 1, R = 1.99 Å) scattering pairs obtained from known reference compounds. Least squared fits of the first shell of r-space and isolated q-space were performed on the k 2 -weighted Fourier transform data over the range 2.71-10 Å −1 in each spectrum to fit the magnitude and imaginary components. The fits were calibrated using the Fe foil (CN = 8, R = 2.54 Å). During the fit of the Fe foil, the CN was fixed to 8 and an amplitude reduction factor (S o 2 ) of 0.64 was obtained. The S o 2 was fixed as 0.64 to fit all other EXAFS data. WinXAS software provides error calculations for each parameter, generated from a correlation matrix during the least squared fit. These values are on the order of +/−10% for the CN, +/−5% for R, and +/−0.0001 for the Debye-Waller factor (s 2 ). The s 2 stands for the mean square variation in the bond distances.
Computational details. All DFT calculations were performed using the ab-initio software package Q-Chem 109 , the meta-hybrid functional TPSSh 110 and the basis set 6-31+G* [111][112][113] . To model solvation, the implicit solvation model IEF-PCM with a dielectric constant of 78.4 was used 114 . All molecular images were generated using the software package VESTA 115 . The geometries of (phen 2 N 2 )Fe(III)Cl and [(phen 2 N 2 )Fe(III)] + were optimized with spin states S = 1/2, 3/2, and 5/2. The most stable spin state was found to be S = 3/2 in both cases.
General electrochemical methods. Sodium hydroxide (99.99%, semiconductor grade), potassium ferricyanide (ACS grade) and potassium sulfate (99.99%, semiconductor grade) were obtained from Sigma Aldrich and used as received. 5 wt% perfluorinated Nafion® resin solution was obtained from Ion Power Inc. and used as received. Aqueous electrolyte solutions were prepared with deionized water purified to a resistivity of 18.2 M Ω-cm using a Milliq water purification system (Millipore Corporation). Vulcan VXC72R carbon powders were obtained from Cabot Corporation and used as received. Glassy carbon disk electrodes were obtained from Pine Research Instrumentation, Inc. Hg/HgO and Hg/Hg 2 SO 4 reference electrodes were obtained from CH Instruments, Inc. A titanium mesh was used as the counter electrode for all electrochemical experiments. The titanium counter was constructed of titanium wire (99.9%) and titanium gauze (40 mesh) obtained from Alfa Aesar and was treated with aqua regia prior to use.
All electrochemical experiments were conducted at ambient temperature using a Biologic VSP 16-channel potentiostat and a three-electrode electrochemical cell. In all cases, the Ti counter electrode was isolated from the working compartment by a porous glass frit. Hg/HgO and Hg/Hg 2 SO 4 reference electrodes were used for all experiments conducted in alkaline and acidic aqueous electrolytes, respectively. The Hg/HgO reference was stored in 3 M NaOH solution between measurements. Electrode potentials measured in aqueous alkaline media were converted to the RHE scale using the following relationship: E(RHE) = E(Hg/HgO) + 0.094 V + 0.059*(pH) V. The Hg/Hg 2 SO 4 electrode was stored in saturated K 2 SO 4 solution between measurements. Electrode potentials measured in aqueous acidic media were converted to the RHE scale using the relationship: E(RHE) = E(Hg/Hg 2 SO 4 ) + 0.65 V + 0.059*(pH) V. Reference electrodes were periodically checked against a pristine reference to ensure against potential drift.
Routine electrochemical measurements were performed with a 5 mm diameter glassy carbon disk working electrode rotated at 2000 rpm. All scan rates were 5 mV s −1 to allow for equilibration and to approximate steady state catalytic operation. Cyclic and linear sweep voltammograms were initiated at the open circuit potential. In all cases, the potential was swept in the negative direction. Solution resistance values were <10 Ω and were ignored. All voltammograms were recorded without iR compensation. Solutions were sparged for 10 min before each experiment with either N 2 or O 2 and for another 10 min upon changing between gases. During an experiment, the headspace of the cell was continually flushed with the working gas to minimize solution contamination by atmospheric gases. All experiments were conducted using 10 μL of catalyst ink dropcast onto a polished glassy carbon RDE. Catalytic onset potentials were taken to be the potential at which the observed current density reached -0.1 mA cm −2 . Buffers used for the pH-dependence study of the (phen 2 N 2 )FeCl ink consisted of a modified Britton-Robinson buffer (50 mM each of NaHSO 4 , Na 2 HPO 4 and B(OH) 3 ) that was adjusted to the desired pH value with concentrated HClO 4 or NaOH.
Preparation of Fe-N-C ink. Fe-N-C inks for use in electrochemical studies were prepared by sequentially adding CH 2 Cl 2 , EtOH, and 5 wt% Nafion solution to 11 mg Fe-N-C catalyst in a ratio of 7:2:1 to give a final volume of 1 mL and an Fe-N-C concentration of 11 mg mL −1 . The resulting mixture was sonicated for 16 min to give the product ink. Inks were produced in 1 mL batches. Superior results in terms of carbon powder dispersion and ORR activity were obtained when CH 2 Cl 2 was added to the precursor powder first, followed by EtOH and Nafion solution. The Fe-N-C data presented in Fig. 5 and Supplementary Fig. 25 was collected at an Fe-N-C concentration of 4.7 mg mL −1 in order to limit the double-layer capacitance, which can obscure redox peaks. See Supplementary Table 10 for nominal dropcast film mass loadings.
Preparation of (phen 2 N 2 Fe)Cl, (Pc)FeCl and (OEP)FeCl Inks. Inks for use in electrochemical studies were prepared by suspending~1.1 mg mL −1 catalyst (either porphyrin-or (phen 2 N 2 )-based) in~2.5 mg mL −1 Vulcan VXC72R carbon. CH 2 Cl 2 , EtOH and 5 wt% Nafion solution are added sequentially in a ratio of 7:2:1. The resulting mixtures were sonicated for 16 min to give the product inks. The inks were produced in 1 mL batches in all cases. Note that superior results in terms of carbon powder dispersion and ORR activity were obtained when CH 2 Cl 2 was added to the solid precursor mixture first followed by EtOH and Nafion solution. See Supplementary Table 10 for nominal dropcast film mass loadings.
Rotating ring disk electrode voltammetry. RRDE voltammetry was performed with a Pine rotator using a ChangeDisk tip. RRDE experiments conducted in acidic media were performed with a platinum ring while experiments conducted in alkaline media were performed with a gold ring 116 . The Pt and Au rings were cleaned by cyclic voltammetry conducted at 50 mV s −1 over potential ranges of 1 to −0.8 V and 1.1 to −0.7 V, respectively, vs the Hg/Hg 2 SO 4 reference in 0.1 M HClO 4 . The potential was swept over the appropriate range until the cyclic voltammograms overlaid exactly.
The proportion of H 2 O 2 produced during ORR was quantified using the following relationship 95 : In this case, i ring and i disk are the currents at the ring and disk electrodes, respectively and N is the collection efficiency. The number of electrons transferred as a function of applied potential was calculated using the following relationship 117 : The collection efficiency constant, N, was calculated at the end of each group of experiments under a given set of conditions by examining the Fe(III/II) couple via chronoamperometry. The ring was polarized at 1.5 V vs RHE while the disk was held at 0.1 V vs RHE and currents were monitored for 60 seconds. The current values over the last 10 s were averaged and taken as the values for i ring and i disk . A 60 s chronoamperogram was also run with the ring as the working electrode to give the background oxidative current at 1.5 V vs RHE. Averaging the last 10 s of the background trace gave the value of i ring0 . Collection efficiency measurements were conducted using a blank glassy carbon disk. Fe 2 (SO 4 ) 3 and K 3 Fe(CN) 6 were used as the redox agents for the collection efficiency measurements in acid and base, respectively. The collection efficiency was calculated using Eq. (5) and found to be 10-11% at the Pt ring and 22-23% at the Au ring 116 : Since calculated %H 2 O 2 values are unreliable when disk currents are small (near the ORR onset region), each %H 2 O 2 trace is shown starting where dependable data are available and noise in the measurement has been minimized. Although each catalyst is distinct, this point was located between 50-100 mV past the ORR onset potential in all cases.

Data availability
Data generated or analysed during this study are included in this article and/or its supplementary information files. Raw data are available from the authors on request.