Introduction

As our energy consumption continues to escalate, it is increasingly important that energy conversion and chemical production technologies progress alongside. Improving the efficiency of all involved electrochemical steps and coupling processes that require the input of energy to drive pertinent reactions with green sources, such as wind and solar power, provides a gateway to a renewable future. Inspired by this outlook and motivated by the remarkable success of the Li-ion battery technology, scientists and engineers are exploring advanced electrochemical transformations, many of which involve light, earth-abundant elements, such as C, H, N, O, and S to derive chemical conversion materials with high energy density. Heavily involved is the chemistry of oxygen, because the release of stored chemical energy is performed via the oxidation of the “fuel” using molecular oxygen. For example, molecular O2 is the oxidant responsible for converting the fuel (H2) to water and generating electricity in the hydrogen fuel cell (Fig. 1b). The reverse process, thermodynamically uphill water electrolysis generates the fuel and oxidant (Fig. 1c). Similarly, methanol is used as fuel in the methanol fuel cell (Fig. 1d), and may be regenerated via reduction of CO2 (Fig. 1e). Finally, some light metals, like Zn, operate as anodes in metal-air batteries (Fig. 1f). In addition to energy conversion applications, electrochemical conversions are explored for direct synthesis of industrially relevant chemicals. For example, the electrochemical reduction of oxygen or nitrogen to hydrogen peroxide or ammonia, respectively, (Fig. 1g) represents a promising alternative for the costly anthraquinone and Haber–Bosch processes currently used on large scales.

Fig. 1: Structures of electrocatalytic nanocarbons and schemes of electrochemical devices.
figure 1

a Representative structures of the three EN types. Hydrogen fuel cell (b) and water electrolyzer (c) where EN is incorporated to accelerate the ORR, OER, and HER. A methanol fuel cell (d) and CO2 reduction cell (e) which make use of EN for the MOR, CO2RR, ORR and OER. A zinc-air battery (f) where EN functions as a cathode electrocatalyst for the ORR and OER. A nitrogen reduction or oxygen reduction cell (g) where EN is being investigated for the NRR and 2e− ORR for electrochemical production of ammonia and hydrogen peroxide, respectively.

Full size image

Electrocatalytic nanocarbon (EN), defined here as any form of chemically modified graphitic nanocarbon, has emerged as an excellent electrode material for the chemical transformations central to the applications described above. The last several years have witnessed an explosion of scientific studies reporting excellent performance of EN in accelerating the oxygen evolution reaction (OER)1, oxygen reduction reaction (ORR)4. Inspired by this, many successful studies on SACs for ORR have been reported5,6,34,35,36. For example, McCloskey used a mild reduction of graphene oxide to generate EN material with oxygen groups, that convert oxygen to hydrogen peroxide at low onset overpotential (10 mV) and high selectivities across an entire potential range (100% Faradaic efficiency, Fig. 2e)34. While epoxide functional groups were identified as the catalytically active moiety in this study (using in situ Raman spectroscopy), recent reports indicate that carbonyl groups are more catalytically active toward two-electron ORR36. First-row transition metal based SAC catalysts have also shown excellent performance in electrocatalytic H2O2 formation37,38,39. These studies point to an important role that the SAC coordination environment plays, where presence of O was found to have large effect on the product selectivity. For example, Zhang et al, synthesized a Ni-SAC with mixed N and O coordination sites based upon the molecular Jacobsen’s catalyst with NiN2O2 coordination37. Activity towards two-electron ORR was benchmarked against an analogous Ni-based SAC with only nitrogen atoms in the coordination environment (NiN4). As measured by ring current on the rotating ring disk electrode, the oxygen-containing NiN2O2 catalyst showed much improved selectivity toward two-electron ORR (96%) compared with the all-nitrogen NiN4 (62%). The likely role of oxygen atoms in the coordination sphere is to tune active site oxophilicity, such that adsorbed *OOH intermediates are preferentially protonated at the proximal O, rather than distal, preventing O–O bond cleavage. The presence of oxygen itself, as either a first or second shell ligand, can have a major effect on active site oxophilicity in SACs37,38,39.

The electrochemical CO2 reduction reaction (CO2RR) presents a convenient opportunity to help decelerate rising atmospheric greenhouse gas levels. Coupling this process with renewable energy sources presents a sustainable method of recycling atmospheric CO2 for production of valuable industrial chemicals and fuels. Most commonly, CO2RR proceeds through a 2e− process to afford CO, an important feedstock for further conversion into organic fuels and chemicals. Heteroatom do** of graphitic carbon induces significant charge or spin redistribution and significant electronic structure modification, making them an impressive platform to improve CO2RR selectivity towards CO. For example, graphene foams with N-defect sites were found to perform CO2RR electrocatalysis with maximum Faradaic efficiency for CO (FECO) of 85% at −0.58 V vs. RHE. X-ray photoelectron spectroscopy (XPS) and electrochemical data revealed that samples with the highest pyridinic-N content displayed the highest FECO values and at lowest overpotentials. Computational evidence agreed with these results, revealing the active sites to be electronegative, pyridinic-N sites40. Similarly, N and F dual-doped holey graphene achieved an FECO of 90% at −0.6 V vs. RHE in CO2-saturated 0.1 M KHCO3 (Fig. 2f).73. As such, it is important to utilize multiple spectroscopies, each providing complementary information about the metal centers and the associated coordination sphere. It is also essential that in situ electrochemical studies are reported, since these studies provide important mechanistic insights. When used in situ, changes to EXAFS and XANES spectra can be representative of structural distortions and changes in active site electronic structure, consequences of active site, and substrate interaction/hybridization. Similarly, spectroscopies like FTIR and Raman are sensitive to the vibrational modes of a sample and can thus monitor interactions between the substrate, active site, and support. Coupling together a full suite of ex situ and in situ measurements can result in confident assignment of active site geometric structure along with the observance of some intermediate states.

Insights from molecular models

In parallel with research on catalytically active EN materials, molecularly precise models have been extensively studied as homogeneous electrocatalysts for OER/ORR/CO2RR/NRR. Several advantages make them ideal systems for mechanistic studies: well-defined structures that are characterized with high fidelity, structural tunability that enables optimization of catalytic activity through rational design of active species, and accessibility of techniques to study catalytic mechanisms. Insights gained from those studies are envisioned to guide design of future catalytic materials. One of the most important insights from these studies is that catalytic motifs are never single atoms. For example, multielectron chemistry, such as OER, is challenging for a single metal catalytic site to accommodate. While late precious transition metals, such as Ru, can support such large oxidation state changes (from 2+ to 5+)74, earth-abundant, first row transition metals, such as Co, usually undergo only one or two-electron oxidation. Thus, OER often involves bimetallic sites, such as the Co-cubane model presented in (Fig. 4a)75. Extensive spectroscopic and electrochemical studies of Co-cubane showed that two Co active sites undergo a sequence of PCET steps to form intermediate structures 2–5 (Fig. 4a), followed by the rate-determining O–O bond formation step from intermediate 5. Such bimetallic catalysis is believed to occur in many OER catalysts, such as enzymatic Mn-based clusters76 and artificial transition metal oxide catalysts77, confirming the importance of cooperative effects between multimetallic sites in electrocatalysis.

Fig. 4: Catalytic mechanisms deduced from molecular models.
figure 4

a Co-cubane OER catalysts and its catalytic mechanism. Catalyst undergoes four-electron oxidation shared between two Co-centers yielding an oxyl intermediate 5 which then forms O–O bond75. b NRR asymmetric mechanism considered as a possible pathway for nitrogenases78. c ORR mechanism accepted for metal-free GQDs showing carbanion (3) intermediate which is crucial for O2 activation. However, if pH is lowered, protonation becomes fast yielding intermediate 5 and shutting off the catalysis79. d CO2RR mechanism showing branching between CO and formate-forming cycles from intermediate 3. Transition metal used in complex determines the path catalysis undertakes85.

Full size image

In addition to cooperativity of multiple metal centers, the best catalyst performance requires synergism between ligands. First and second coordination spheres play important roles in catalytic cycles such as providing pathways for protons, electrons, substrates, and products or adjusting local pH to the required value for particular catalytic steps. A beautiful example illustrating ligand participation can be found in enzymatic NRR involving Mo-nitrogenase. While the exact catalytic pathway is still not known, a plausible mechanism that is consistent with experimental evidence78 and computational studies53 is presented in Fig. 4b. The catalyst consists of a large FeMo cofactor, whose role is to facilitate accumulation of electrons and protons needed for reduction, provide a binding site for dinitrogen and reduce it to two molecules of ammonia. Interestingly, the enzymatic reaction involves eight PCET steps (only six are necessary for NRR), where the additional two PCET processes are needed for the release of H2 and H2S needed to generate coordination vacancies at Fe-centers needed for N2 binding and activation (Steps 1→3, Fig. 4b, the structures show the replacement of one of the sulfur belt atoms with N2)78. Subsequent PCET processes generate NH3 molecules, whose release is coupled with the return of the sulfur belt atom to the Fe-centers (Steps 4→1). This dynamic Fe–S metal–ligand equilibrium represent an important mechanism of providing highly reactive vacancies for substrate binding.

One feature in common to all aqueous electrocatalysis is the PCET mechanism. Transferring protons simultaneously with electrons avoids charge buildup, and therefore circumvents formation of high energy intermediates. A simple inspection of the four mechanisms presented in Fig. 4 illustrates the importance of PCET by showing that most electrons are transferred in concert with protons. However, in some cases PCET can prevent desired reactivity. Li and coworkers79 studied well-defined N-containing graphene quantum dots as models for ORR catalysis by N-doped nanocarbon (Fig. 4c). The authors observed that ORR catalysis occurs only at pH > 11, in agreement with many reports of efficient ORR by N-doped MGEN in basic media3. Based on the Pourbaix diagrams, the authors identify anionic intermediate 2 as the key intermediate activating molecular oxygen towards reduction to hydroperoxide 3. Reactive sites are identified as carbon atoms adjacent to nitrogen atoms, which is in a good agreement with ORR studies on analogous metal-free models80,81,82,83. Subsequent computational investigation of the catalytic cycle84 identified an unusual ether-like structure 4 as an intermediate, that ensures the four-electron pathway. The lack of catalytic activity at pH < 11 was assigned to fast protonation following two-electron one-proton PCET to form neutral intermediate 5, which is not sufficiently reactive to activate O2.

Product selectivity is one of the most challenging issues in electrocatalysis, particularly in the case of CO2RR where many products can be formed, such as CO, H2, formate, oxalate, methanol, etc. Selectivity toward one of the products is dictated by the energies of relevant metal-substrate adducts. For example, a recent study involving Co and Fe complexes 1 and 2 (Fig. 4d) showed how strength of C–O bonds in M–COOH intermediates control the product branching pathways85. Specifically, the authors found that Co-complex 1 reduces CO2 to CO, while Fe-complex 2 selectively forms HCOO−. Their mechanistic studies reveal that both complexes, upon two-electron reduction, react with CO2 through the carbon atom to form intermediate 3 (Fig. 4d)85. However, electron-rich Co(II) destabilizes the C-O bond through backbonding, leading to its breakage and formation of CO and H2O. Since Fe-complex 2 is a poor π-donor, intermediate 4 isomerizes into the O-bonded derivative, which leads to formation of the formate ion85. Product selectivity is also an important issue in ORR, where four-electron (H2O) and two-electron (H2O2) products are observed. For example, ORR catalyzed by Fe-porphyrins proceeds through a peroxo intermediate, the key structure for H2O/H2O2 product selectivity. If the distal oxygen is protonated a second time, the O–O bond will be cleaved and consequently four-electron reduction to water occurs. If, instead, the proximal oxygen is protonated, the O–O bond will be conserved, leading to predominantly H2O2 formation86,87. While Fe-based macrocycles could give both two-electron and four-electron reduced products, mononuclear Co-based macrocyclic catalysts usually selectively produce H2O2. A recent study by the Stahl group has shown that protonation of proximal oxygen is responsible for H2O2 formation, similar to Fe-based catalysts88.

One of the key parameters that determines catalytic performance is the overpotential. Intuition tells us that catalyst reduction potential is directly related to its reactivity: if a more negative potential is needed for catalyst reduction, the reduced form will be more reactive. Indeed, this scaling relationship has been observed in many systems, including hydride-based catalysts for CO2RR and HER (Fig. 5a), where standard reduction potentials associated with hydride generation scale linearly with their hydricities89,90,91. Importantly, this scaling relationship dictates the minimum overpotential that a catalyst can achieve: if the line passes through the point with zero overpotential (as is the case for HER catalysis by metal-based hydrides, Fig. 5a), chemical tuning of the structure will enable discovery of an ideal catalyst. If, on the other hand, the line does not cross the point of zero overpotential (as is the case for formate formation, Fig. 5a), even the best catalyst will exhibit an overpotential. That minimum overpotential is 1.1 V for metal-based hydrides and 1.5 V for metal-free hydrides (Fig. 5a). Discovering ways of breaking such scaling relationships is a crucial aspect of catalysis research and often requires partitioning the catalyst into separate chemical units that can be tuned independently. For example, the overall hydride transfer needed for formate formation can be achieved using a catalyst composed of separate electron and proton sources. Since reduction potentials and pKa values exhibit different scaling relationships (Fig. 5b)92, an ideal catalyst with zero overpotential can be more readily discovered by independently tuning the co-catalytic units.

Fig. 5: Thermodynamic properties of molecular catalysts.
figure 5

a The scaling relationships between the thermodynamic hydricity (ΔG) and the first reduction potential E1 of the metal-free (orange) and metal-based hydrides (blue). Adapted and reprinted with permission from ref. 89. b Two-electron redox potential (blue triangles) and pKa (red squares) contributions to ΔG (black circles) of metal hydrides plotted by their average redox potentials. Adapted and reprinted with permission from ref. 92. c Pourbaix diagram for Rh molecular catalysts (red triangles) and Rh attached to glassy carbon electrode (GCC-Rh, blue circles). The black solid line denotes the thermodynamic potential of HER. The orange region denotes the potential-pH region in which the model is catalytically active, while the blue region shows conditions where GCC-Rh catalyzes HER97. d Proposed interfacial free energy diagrams for GCC-Rh electrodes. The gray area represents the filled band states of the electrode, while beige denotes the unfilled band states. The dotted horizontal black line indicates the EF of the electrode. The electrical double layer, EDL, is labeled by a vertical dotted black line. The redox potential of the molecule (E1/2(RhIII/I)) and the potential for formation of a Rh−H species at the GCC site (E(Rh + H+/Rh−H)) are illustrated with dotted gray and blue lines, respectively. The electrostatic potential is shown with the dotted red line, and the potential of zero free charge (EPZFC), is indicated with a dotted gray line. Adapted and reprinted with permission from ref. 97, https://pubs.acs.org/doi/full/10.1021/jacs.9b04981. All future permission should be directed to the ACS.

Full size image

While the fields of homogeneous and heterogeneous electrocatalysis have been advancing on parallel tracks, there is still a large knowledge gap between them—it is not known what fundamentals govern the heterogenous inner-sphere electron transfer (ISET) nor how mechanisms of electrocatalysis change once molecular catalysts are placed in strong electronic communication with conductive electrodes. Recent studies indicate that interesting changes in faradaic efficiencies and product selectivity can take place. For example, immobilization of Co-phthalocyanine onto a conductive CNT support yielded reduction of CO2 to methanol in comparison to two electron reduction to CO for the solution-phase molecule9. The authors attribute this change in reactivity to restricted diffusion of the adsorbed catalyst: unlike the homogeneous model, the immobilized catalyst cannot diffuse away from the electrode surface, which enables multielectron transfers needed for formation of methanol9. In addition to immobilization effects, adsorption of the catalyst to the carbon support alters its electronic properties, as was observed in Fe-based ORR catalysts93. Specifically, it was shown that the incorporation of molecular Fe-phthalocyanine catalysts onto graphitic supports using pyrolytic methods improves ORR performance due to the electron-withdrawing effect of the carbon pi-system93. There has been an avenue of studies exploring larger conjugated structures, graphene nanoribbons (GNR), with Re catalysts grafted onto edges. It has been shown that this extended conjugation lowers the overpotential for CO2RR to CO compared to smaller models94. Surprisingly, the metal experienced no oxidation state change in the process and the overall catalytic reaction was described, as a transfer of electron density from a two-electron reduced ligand to CO295. Similar phenomena was observed when Re catalysts were embedded on the edges of glassy carbon electrodes96.

The Surendranath group has explored molecular catalysts grafted to edge-planes of carbon electrodes and discovered significant changes in PCET behavior97,98. For example, Rh-based HER catalysts grafted to carbon electrodes, exhibit catalytic behaviors over a full pH = 0–14 region, in stark contrast to the homogeneous catalyst, which operates only in the acidic region (Fig. 5c). These results were rationalized as follows: the molecular model undergoes PCET and efficient catalysis in the acidic region, while catalysis is prevented in the basic region by electron transfer processes that are not coupled with proton transfer. The grafted catalyst, on the other hand, is in strong electronic communication with the conductive electrode, which prevents undesired “non-coupled” electron transfer processes from taking place. Instead, PCET proceeds in the entire pH region, controlled thermodynamically by the point of zero free charge of the electrode (Fig. 5d). Similarly, it has recently been shown that strong coupling between glassy carbon electrodes can be achieved through basal plane modification with nitrogen-doped GNRs. When the GNRs are determined to be in strong communication with the electrode surface, every ET across the entire pH region (0–14) was coupled to PT99. These results point to an important difference between homogeneous and heterogeneous catalysts: while isolated electron transfer events can take place in homogeneous catalysts, they are always coupled by a compensating transfer of charged groups, e.g., protons, in the case of heterogeneous catalysis.

Insights from computational studies

Computational density functional theory (DFT) studies of electrocatalytic processes were pioneered by Norskov and co-workers in their studies of ORR catalysis by transition metal and oxide surfaces100, and have been applied to EN, like those shown in Fig. 6a. In their approach, an overall n-electron reduction or oxidation reaction is divided into n proton-coupled one-electron transfer processes, each resulting in formation/breakage of key reaction intermediates at the electrode surface (such as *OH, *O, and *OOH in OER/ORR, Fig. 6b). Instead of calculating Gibb’s free energies of protons and electrons, the method relates these to the calculated free energy of molecular hydrogen using the electrochemical potential of the standard hydrogen electrode (SHE, H+ + e− \(\rightleftarrows\) 1/2H2). Reaction overpotential is calculated by adding free energies for individual steps, leading to a simple computational method that enables screening large numbers of model catalysts. Even though this purely thermodynamic model neglects kinetic barriers associated with, e.g., proton and electron transfer, the method has been shown to provide predictive insights into a range of electrocatalytic reactions, including ORR, OER, CO2RR, and NRR100,101,102,103,104.

Fig. 6: Volcano relationships from EN computational studies.
figure 6

a EN cells which have been investigated in computational studies. b Thermodynamic plot which shows the free energy of each elemental step for ideal (blue) and non-ideal (red and gray) catalysts for the ORR at O and 1.23 V applied potential. c, d ORR volcano plots for single-SACs and bimetallic-SACs, showing the theoretical onset potentials (\({U}_{{\rm{RHE}}}^{{\rm{onset}}}\)) versus OH adsorption free energies (\({\triangle G}_{{\ast }_{{\rm{OH}}}}\)). Adapted and reprinted with permission from refs. 108,109. e ORR volcano plot for MGEN constructed from plotting the theoretical exchange current density (\({j}_{0}^{{\rm{theory}}}\)) versus free energies of OOH adsorption (\({\triangle G}_{{\ast }_{{\rm{OOH}}}}\)). Adapted and reprinted with permission from ref. 110, https://pubs.acs.org/doi/10.1021/ja500432h, all future permissions should be directed to the ACS. f OER volcano plot for single-metallic SACs, showing the theoretical onset potentials (\({U}_{{\rm{RHE}}}^{{\rm{onset}}}\)) versus OH adsorption free energy (\({\triangle G}_{{\ast }_{{\rm{OH}}}}\)). Adapted and reprinted with permission from ref. 108. g NRR volcano plot for C3N4 SACs constructed from plotting theoretical limiting potentials versus free energy of N adsorption (\({\triangle G}_{{\ast }_{{\rm{N}}}}\)). Adapted and reprinted with permission from ref. 111. h Selectivity plot for the NRR vs. HER for SACs constructed by plotting free energy of N adsorption (\({\triangle G}_{{\ast }_{{\rm{N}}}}\)) versus free energy of H adsorption (\({\triangle G}_{{\ast }_{{\rm{H}}}}\)). Adapted and reprinted with permission from ref. 112.

Full size image

It is immediately obvious from the energy diagram in Fig. 6b (blue line) that the ideal catalysts exhibit the same electrochemical potential, equal to the standard reduction potential of the process of interest, for each one-electron transfer step of the overall n-electron transfer process. Any deviation from that balance, leads to an overpotential, as illustrated in Fig. 6b for ORR involving catalysts that bind *OOH too strongly (gray) or too weakly (red)101,102. Unfortunately, the energies for one-electron steps cannot be tuned individually because, from the perspective of the surface, adsorbate molecules are very similar and thus bind in similar fashions. For example, any OER/ORR catalyst that increases the energy associated with formation of *OH species will also increase the energy of formation of *OOH intermediates, resulting in scaling relationships with two important consequences. First, the overall catalytic process can be studied using a single descriptor, the Gibb’s free energy of a selected one-electron transfer step. A plot of catalytic overpotential as a function of this descriptor generates well-known volcano plots with the most active catalyst sitting atop the volcano. These binding energies can be further correlated to electronic structure descriptors, which explain bond formation between adsorbate and catalyst. A formative descriptor proposed by Norskov and Hammer is d-band theory, which describes adsorbate binding to transition metal catalysts based on position, shape, and filling of the metal d-band. Generally, metal d-bands which are high in energy, relative to Fermi level (EF), will generate catalyst-adsorbate antibonding orbitals which are also high in energy, i.e., less filled, making strong interactions and vice versa105. The d-band theory provides a framework by which we understand catalyst–adsorbate interactions through chemically intuitive electronic structure orbital descriptors. The second important consequence of Norskov’s scaling relationships is that, even for the catalyst at the top of the volcano plot, there may exist an overpotential which cannot be eliminated using simple electrocatalytic transition metal-based surfaces (0.37 V for ORR and OER)101,102. This intrinsic overpotential has prompted computational scientists to propose utilization of co-catalysts, that can selectively tune binding energies of individual intermediates106,107.

Similar computational approaches have been applied to EN. Activity of transition metals was probed in several nanocarbon metal-free, monometallic, and bimetallic environments, as shown in Fig. 6a. Substrate binding sites for metal-free systems are usually carbon atoms neighboring the heteroatom, metal sites for monometallic species, and one or two metals for bimetallic species (binding sites are marked with arrows, Fig. 6a). Similar scaling relationships were identified, leading to a single descriptor, such as the binding energy for *OH (in OER/ORR) or *N (in NRR) that predicts reactivity and produces volcano plots (Fig. 5c–g). For ORR, the most active investigated catalysts, which reside at the top of the volcano plots are Fe-pyridine-N4 (overpotential = 0.42 V, Fig. 6c)108, bimetallic CoPtN8V4 (overpotential = 0.30 V, Fig. 6d)109, and N-doped graphene (overpotential = 0.36 V, Fig. 6e)110. Additionally, the most active catalysts for the OER and NRR were identified to be Ir-pyrrole-N4 SAC (overpotential = 0.30 V, Fig. 6f)108 and Ru-g-C3N4 (overpotential = 0.478 V, Fig. 6g)111.

To obtain a better chemical intuition for the observed trends, other descriptors have been identified based on electronic properties of the material. For example, catalytic efficiency for NRR was found to correlate with the integrated crystal orbital Hamilton population (ICOHP) descriptor to model the interaction between an adsorbate atom and a solid surface. Hamiltonian-weighting of the solid density-of-states gives partitioned band structure energies, which are separated into bonding, non-bonding, and anti-bonding states and displayed relative to EF. Calculating the ICOHP of all states below EF provides chemical information about bond strength between the adsorbate and surface. The ICOHP was shown to be well-aligned with ΔG*N and thus activity towards NRR111. For a broad range of catalysis, Zeng established a highly convenient, generalized universal descriptor (φ) for SACs which showed a linear relationship with, for example, ΔG*OH. Briefly, φ accounts for the number of d-electrons residing on the metal active site after charge redistribution from ligation and intermediate adsorption. φ includes contributions of electronegativity (E), number of nearest neighbor ligands (η) and d-electrons (θ) from the ligands, adsorbates and metal center. According to the universal descriptor, a larger φ results in larger filling of metal d-states, which depends on its position relative to EF108.

Chemical selectivity of the catalyst was also investigated, particularly in systems where competing HER is thermodynamically feasible (NRR and CO2RR). For example, the key selectivity determining step in NRR vs. HER was identified as binding of *N or *H, where surfaces with strong nitrogen binding are more selective toward NRR. Even though metal surfaces show strong preference for *H binding with few exceptions among early transition metals, SACs exhibit HER suppression because of the ensemble effect112. In effect, the ensemble effect, resulting from interactions between single metal atoms and surrounding ligands, precludes H-adsorption to preferred hollow or bridge sites and only allows for top site adsorption. However, HER is still a competing pathway for most SAC111. By plotting ΔG*N vs. ΔG*H, one can identify the most selective catalysts for NRR (Fig. 6h), which were found to be Mn-SV-N3, V-pyridine-N4, and Ti-pyridine-N4112. Improved selectivity towards NRR of SAC compared to metal catalysts is further rationalized by the difference in mechanisms of HER on those surfaces. On metal catalysts HER proceeds though easier Tafel mechanism with H–H coupling occurring between adjacent metal sites. On SAC that pathway is hindered because adjacent metal centers are far away and do not allow coupling of 2H* species. Therefore, HER follows the more difficult Heyrovsky pathway making NRR relatively competitive with HER.

Future directions

One of the principal challenges impeding development of superior EN materials is the knowledge gap, which exists between solid heterogenous electrocatalysts and their molecular counterparts. Further challenging the situation are elevated, pyrolytic temperatures, which are often employed for synthesis of graphitic carbons, in which precise control over chemical functionalization is lost making mechanistic investigations implausible. As such, EN with well-characterized chemical structures make ideal platforms for facile mechanistic investigations and should continue to be pursued. Synthetic techniques which circumvent these high temperatures and can produce desired functionalities will make valuable contributions to EN research.

Some inspiring strategies for synthesis of EN worth-exploring use chemical additives that can effectively lower the barrier of graphitic precursor decomposition, and adventitious functional groups on graphitic carbons for surface modifications. For example, when metal fluorides are added during synthesis, they release fluorine and have been shown to improve kinetics associated with carbon feedstock decomposition in the growth of graphene monoliths, revealing a path to reducing extreme temperature parameters.113 Surface oxide species, e.g., quinones, carboxylates, and alcohols, native to pre-existing graphitic carbons are common reactive functional groups used in named organic reactions. Making use of these inherent reactive sites, like the reaction between o-quinone groups and 1,2-diamines, are a simple way of covalently grafting molecular sites to conductive carbons. With well-defined catalytic motifs, fundamental studies of heterogenous ISET have been made possible114.

Further enhancement of spectroscopic and microscopic methods, both in situ and ex situ, for probing EN before, during and after catalysis will be of benefit to our mechanistic knowledge. Undoubtedly, coupling together complementary techniques, i.e., hyphenated techniques, may radically improve structural characterization of EN. For example, synchrotron X-ray scanning tunneling microscopy (SX-STM) combines together the elemental sensitivity of X-rays and sub-nanometer spatial resolution of STM115. Spectroscopic signatures belonging to adsorbed intermediates can be detected via, for example, Raman and FTIR spectroscopies, however, the observed state will be that of the kinetically slowest intermediate. The main bottleneck to improved spectroelectrochemical investigations is the limitations on temporal resolution of the chosen spectroscopy—chemical processes which are being monitored are happening at a much faster timescale than what can feasibly be sampled. If time-resolutions of optical instruments can continue to be improved, the range of catalytic intermediates we are able to observe may grow116.