Selective electrochemical reduction of nitric oxide to hydroxylamine by atomically dispersed iron catalyst

Electrocatalytic conversion of nitrogen oxides to value-added chemicals is a promising strategy for mitigating the human-caused unbalance of the global nitrogen-cycle, but controlling product selectivity remains a great challenge. Here we show iron–nitrogen-doped carbon as an efficient and durable electrocatalyst for selective nitric oxide reduction into hydroxylamine. Using in operando spectroscopic techniques, the catalytic site is identified as isolated ferrous moieties, at which the rate for hydroxylamine production increases in a super-Nernstian way upon pH decrease. Computational multiscale modelling attributes the origin of unconventional pH dependence to the redox active (non-innocent) property of NO. This makes the rate-limiting NO adsorbate state more sensitive to surface charge which varies with the pH-dependent overpotential. Guided by these fundamental insights, we achieve a Faradaic efficiency of 71% and an unprecedented production rate of 215 μmol cm−2 h−1 at a short-circuit mode in a flow-type fuel cell without significant catalytic deactivation over 50 h operation.


Supplementary Note 3. Rotating ring disk electrode (RRDE) collection efficiency estimated
with an assumption of selective NO-to-NH2OH conversion on FeNC-dry-0.5.
The RRDE collection efficiency (N) was estimated from the voltammetric responses of Pt ring electrode during NORR on the FeNC-dry-0.5 disk electrode at a 1,600 rpm rotation speed in a NO-saturated 0.1 M HClO4 electrolyte (Fig. 1d), a method suggested by Koper group. 2 The disk electrode was polarised either at 0.05 VRHE or open circuit potential (OCP), a condition that FeNC-dry-0.5 catalyses NORR at the 2 nd reduction region (Fig. 1a) or NORR does not occur, respectively. When NO was reduced on the disk at 0.05 VRHE (idisk = ca. 0.57 mA), decrease in Pt ring current was observed (Δiring = ca. 0.08 mA). Note that NORR follows one-electron pathway on Pt electrode at a potential above 0.25 VRHE, producing N2O as a final product. Because the current on the Pt ring electrode is proportional to the NO concentration at the ring electrode, which is the product of N and NO concentration at the disk electrode, the decrease in the ring current can be expressed as following equation (Eq. 2).
Δiring = iring (@ Edisk = OCP) − iring (@ Edisk = 0.05 VRHE) (Eq. 2) where CNO@OCP and CNO@0.05VRHE are the NO concentration at the disk electrode polarised at OCP and 0.05 VRHE, respectively. Therefore, 'CNO@OCP − CNO@0.05VRHE' is the NO concentration consumed at the disk electrode during the NORR at 0.05 VRHE, which is proportional to the current on the FeNC-dry-0.5 disk electrode (Eq. 3). Assuming that NORR on FeNC-dry-0.5 selectively produces NH2OH at 0.05 VRHE (i.e., n = 3), the N was estimated to ca. 0.42, which is in great agreement with the calibration value ( Supplementary Fig. 10b). Therefore, this RRDE study confirms that FeNC-dry-0.5 mainly catalyses NO-to-NH2OH conversion at the 2 nd reduction region. This study further indicates that N2O and NH3 are not main NORR products on FeNC-dry-0.5 since these assumptions lead to unrealistically low and high N values of 0.14 and 0.71, respectively.

Density functional theory (DFT) calculations
DFT calculations of reaction energetics were carried out with a periodic plane-wave implementation and ultra-soft pseudo-potentials using QUANTUM ESPRESSO version 6.1 interfaced with the atomistic simulation environment (ASE). 5, 6 We applied ultra-soft pseudopotentials and the revised Perdew-Burke-Ernzerhof functional (RPBE). 7 Spin-polarised calculations were performed for which we determined optimally converged plane-wave and density cutoffs of 700 and 7,000 eV, respectively, and used a Fermi-level smearing width of 0.1 eV. Magnetic momenta were automatically optimised by QUANTUM ESPRESSO to minimise the total energy and are shown in Supplementary Table 2.
First, the 2D unit cell area of the Fe-N4/graphene system was optimised by the Broyden-Fletcher-Goldfarb-Shanno (BFGS) line search algorithm. Subsequently, adsorption energies were calculated. For this, we used a 20 Å separation of the surface slabs, and 3 × 3 × 1 Monkhorst-Pack k-point grids. 8 The self-consistent continuum solvation (SCCS) implicit solvation model as implemented in the Environ QUANTUM ESPRESSO module was used to model the presence of implicit water. The 'fitg03' (in a.u.: ρmin = 0.0001, ρmax = 0.005) solvation parameter set is the default set that has been optimised for neutral molecule solvation energies. 9 The bulk dielectric permittivity was set to εb = 6ε0 (vacuum permittivity ε0) that has been rationalised in previous studies from the highly constrained water that has been observed at various metal surfaces. [10][11][12][13][14][15][16] Cavitation and repulsive energy terms are included by introducing an energy term proportional to the cavity surface area as described in the previous literature, 9 and we here apply the parameter (α + γ) = 11.5 dyn cm −1 from the fitg03 parameter set.
Dispersive solute-solvent interactions are ignored since they depend on the cavity volume which is an ill-defined property in surface slab calculations.

Micro-kinetic model
A micro-kinetic model was created from the forward and backward rates of the following reaction steps (where Fe II refers to the whole catalyst): The pressure of NO gas was chosen as 0.5 bar (corresponding to the value that can be estimated from the actually measured NO concentration and Henry constant, Supplementary Note 2). Steps 1-5 are for NH2OH formation, 1-4 and 6-7 for NH3 formation, and 1 and 9-12 for N2O formation. Reaction energetics for proton-coupled electron transfer (PCET) steps were evaluated based on the computational hydrogen electrode (CHE). 23 In addition, the impact of non-vanishing surface charge density/electric field was accounted for by first converting the actual electrode potential to a surface charge density. Assuming a constant double layer capacitance Cdl (ca. 20 μF cm −2 for graphene), 24,25 the surface charge density σ is generated according to: where E is the applied electrode potential and E PZC is the potential of zero charge (PZC) which we measured here to be 0 V vs. standard hydrogen electrode (SHE) (Supplementary Fig. 35).
Then, the parabolic fits of the surface charge density dependence of Supplementary Fig. 16 were used to relate the potential to a field-stabilisation of the intermediate. This procedure was adapted from a previous publication to which the interested reader is referred to. 17 We used the CATMAP simulation package to solve the coupled mean-field rate equations in the steady-state assumption. 26 Electrochemical barriers were estimated by referencing to the reversible potential of each electrochemical reaction step. 27 The rate constant for a particular elementary reaction step i is given as: with the pre-exponential factor A, the applied voltage E, and the activation energy Ga,i. The activation energy can be expressed with reference to the reversible potential of the step as: For the remaining steps which do not involve the transfer of a proton-electron pair (NO adsorption, NO dimerisation, and the desorption reactions), a zero kinetic barrier was assumed.
For NO adsorption and dimerisation, this assumption is valid, because the thermodynamic driving force is dominating over the relevant potential range with a strongly down-hill adsorption of NO and uphill formation of Fe II -N2O2 − . We did, however, also test a 0.5 eV barrier for NO adsorption, which we saw to only reduce the current densities slightly at the highest overpotentials ( Supplementary Fig. 22). For the desorption all barriers were assumed to be zero, but a 0.5 eV barrier was tested for all 3 barriers to estimate the impact of the approximation. For NH2OH adsorption, the introduction of a desorption barrier shifts the NH2OH partial current downward and NH2OH desorption can become limiting. This, however, would be inconsistent with the experimental pH trends, which is why we conclude that the barrier must be smaller and less important. For NH3, the desorption barrier was found to have no influence. For N2O, the desorption barrier let to a decrease of all currents, however, outside the experimentally relevant potential range.
The active site density was needed to convert the turn over frequency provided by microkinetic modeling to a current density. We calculated it based on the experimental data of the Despite of our best efforts to estimate the active site density, it is extremely difficult to quantify how many active sites are actually exposed on the surface and able to participate in the catalytic reaction. Moreover, our Fe-N-C catalyst is a highly porous material, where a mass transport effect cannot be fully ignored even at the low overpotential regime. We thus found some quantitative discrepancy between theory and experiment ( Supplementary Fig. 22d). Of particular, experimental current density showed an early leveling off behavior in comparison to the theoretical data, which can be attributed to the substantial mass transport effect. However, even with such approximations, our micro-kinetic model can correctly describe that N2O is formed at lower and NH2OH/NH3 at higher overpotentials. Most importantly, it also described the experimentally observed inverse pH trends on an RHE scale by the rate-limiting steps and charge dependence, and also the key intermediates proposed from theory were successfully captured from our additional in situ IR experiments   Table 1). 29 Error bars indicate the uncertainty from the Mössbauer fitting parameters.