Mapping free energy regimes in electrocatalytic reductions to screen transition metal-based catalysts† †Electronic supplementary information (ESI) available: Supplemental thermochemical calculations, electrochemical data and simulations and Cartesian coordinates of DFT-optimized intermediates. See DOI: 10.1039/c9sc01766f

The free energy landscape of catalytic intermediates in the two-electron reduction of CO2 and proton donors is mapped with density functional theory to screen catalyst candidates from a library of transition metals and ligands.


Introduction
The multi-electron reduction of abundant feedstocks such as protons and CO 2 to fuels and value-added products 1-3 requires catalysts that can rapidly mediate these reductions with high selectivity and energy efficiency. Several earth-abundant heterogeneous electrocatalysts such as metallic copper can catalyze CO 2 reduction, but oen suffer from poor selectivity and energy efficiency. [4][5][6][7] Furthermore, the challenge of knowing the exact chemical nature of the catalytic active site on metal surfaces makes the rational optimization of such catalysts difficult. On the other hand, with transition metal complexes, one can in principle tune the relative energies of the catalytic intermediates by simply changing the metal and ligand combination, choosing from a vast library to favor one stoichiometric product over another. While heterogeneous catalysts may continue to be more practical, molecular complexes serve as excellent model systems to advance single-site catalyst design.
The multi-electron reduction of CO 2 to methanol or C 2 products is the ultimate goal, but currently no strategies exist to precisely design catalysts that can achieve these transformations rapidly with optimal selectivity. Therefore, as a rst step towards this process, we choose the two-electron reduction of protons to H 2 and of CO 2 to HCO 2 À and CO (eqn (1)-(3)) as the target reactions for catalyst design, given the availability of mechanistic information for these individual reactions. Furthermore, these transformations involve common intermediates, which provide useful handles to reduce the complexity of the design process. Interestingly, there are many more known single-site molecular electrocatalysts for H 2 and CO production [8][9][10][11] than for HCO 2 À production, [12][13][14][15][16][17] and even the few examples of the latter involve precious metals and require high overpotentials.
HA + CO 2 + 2e À / HCO 2 À + A À (1) 2HA + 2e À / H 2 + 2A À HA + 2CO 2 + 2e À / CO + HCO 3 À + A À In order to optimize catalytic activity, our two main strategies involve (a) establishing the relative free energies of the stoichiometric products for a given proton donor and (b) minimizing the free energy corrugation of the catalytic intermediates relative to zero driving force as illustrated in Scheme 1. The latter strategy is central to achieving the desired catalytic activity as intermediates that are high energy or deep trap states in the catalytic pathway will deactivate the catalysts. While the relative rates of different processes will be ultimately governed by the relative heights of transition states, the rst screen of catalysts described in this work aims to understand how free energy corrugations of key catalytic intermediates are determined by the transition metal and the ligand environment.
A signicant recent development in transition metalmediated catalysis has been the application of density functional theory-based computations to model the reactivity of transition metal complexes. [18][19][20] DFT-based methods have proven to predict with reasonable accuracy redox potentials, pK a 's and ligand-exchange equilibria of transition metal complexes in aprotic polar solvents that are well described by simple polarizable continuum models. 18,[21][22][23][24] These methods have been employed to model post-facto the catalytic landscape in a single family of catalysts with relatively small perturbations to the ligand eld. [25][26][27] While the resulting changes in reactivity are extremely useful for the system under study, there is little insight offered into the reactivity of a larger set of complexes. Accelerating the screening of metal and ligand choices towards predicting the optimal catalyst candidate with such methods requires the identication of suitable thermodynamic and kinetic descriptors.
The free energy of the transition metal hydride intermediate relative to reactants, products and other intermediates, an important thermodynamic quantity that determines the driving force towards hydride transfer to CO 2 and protons to produce HCO 2 À and H 2 respectively, has been well studied as an important descriptor of catalytic activity. [28][29][30][31] For instance, we previously showed, using experimentally calibrated DFT, how specic ligand environments change the free energy of CO 2 insertion into a ruthenium hydride complex. 24 The hydride complex we were studying despite being an active transfer hydrogenation catalyst for ketones 32 was not found to be a catalyst for CO 2 reduction due to product inhibition. In another case, a cyclopentadienyl Ru complex 33  , under electrocatalytic conditions, CO is the major product. 38 Therefore, the intermediates leading to CO production thermodynamically and kinetically compete with metal hydride chemistry, motivating the need for computational catalyst design approaches to go beyond hydricity as a single descriptor.
In this work, we explore a diverse library of transition metals and ligands and show the broader utilization of DFT in the catalyst design process. We compute the standard state free energies of the metal hydride and metal-bound CO intermediates relative to the stoichiometric reactants and products for metal complexes with diverse ligand environments around three earth-abundant d 6 transition metal ions (Mn(I), Fe(II) and Co(III)), to screen for optimal metal-ligand combinations. This coarse level of screening based on just two standard state thermodynamic descriptors, using the BP86 density functional, 33,39 weeds out bad candidates effectively and identies two bipyridines (bpy) and a pyridine (py) ligand around Fe(II) as promising candidates among the 36 complexes studied. Fortuitously the two redox-active bipyridyl units are predicted to provide optimal reduction potentials for accumulating the two electrons needed to form the hydride. Because the simple [Fe II (bpy) 2 (py)(CH 3 CN)] 2+ complex would not be experimentally Scheme 1 Catalyst design strategies. stable due to ligand scrambling at ambient temperatures, a pentadentate iron complex, [Fe(bpy2PYMe)S] 2+ (bpy2PYMe ¼ 1-(2-pyridyl)-1-bis-(6-2,2 0 -bipyridyl)ethane, S ¼ CH 3 CN), reported previously by Long et al. 40 was synthesized. [Fe(bpy2-PYMe)S] 2+ was indeed found to electrocatalytically reduce protons as well as CO 2 at modest overpotentials with no catalyst trapping by CO. While the calculated reduction potentials and the free energies of the metal hydride and carbonyl intermediates are validated by the experimental results, the binding energy of CO 2 to the singly reduced Fe complex was found to be energetically favorable in contrast to the predictions, resulting in the uphill formation of CO instead of the thermodynamically favored HCO 2 À . This work, therefore, provides an efficient rst step of catalyst design, by computationally screening against bad candidates via the mapping of free energies of key intermediates in the catalytic pathway. More work is needed to further narrow the candidate space through better modelling of transition metal-CO 2 interactions as a function of the redox state of the transition metal complex. To the best of our knowledge, this is the rst reported application of DFT for the a priori design of molecular electrocatalysts.

Thermodynamics of the stoichiometric reactions
For the two-electron reduction of CO 2 to HCO 2 À and of protons to H 2 in the standard states of reactants and products, the relative driving force in acetonitrile as a function of the pK a of the stoichiometric proton donor, HA, is shown in eqn (4). They are based on the thermochemical analyses shown in Table S1. † DG CO 2 R À DG Given the difference in proton stoichiometry between eqn (1) and (2), plotting DG CO2R À DG HER versus the pK a of the proton donor HA in acetonitrile yields a straight line that crosses zero at a pK a of ca. 24 (Fig. S1 †). With weaker acids in acetonitrile (pK a > 24) there will therefore be a thermodynamic bias for CO 2 reduction to HCO 2 À versus proton reduction. 41 For example, phenol (pK a ¼ 29.14) 42 should disfavor H 2 while acetic acid (pK a ¼ 23) 42 should favor H 2 . Under weakly acidic conditions, the two-electron, one-proton reduction of CO 2 to CO produces bicarbonate (eqn (3)). Due to the lack of experimental free energies for this reaction in acetonitrile, we estimate an upper bound based on the two-proton reduction of CO 2 to CO and water (eqn S1 †). We estimate this latter process (eqn S1 †) to be roughly uphill by ca. 4.8 kcal mol À1 relative to H 2 production under water levels of $1 mM in acetonitrile (eqn S2 †), in agreement with reported values. 43 Catalytic intermediates in the two-electron reductions The reactants, products and catalytic intermediates involved in the two-electron reduction of CO 2 and protons as expected from chemical precedent are shown in Scheme 2, mediated by a generic transition metal complex [M-S] n+ , where M is a transition metal in a six-coordinate ligand environment in which S is a labile solvent ligand (CH 3 CN in this case). The chemical potential of electrons is set to a value of À1.6 V vs. Fc + /Fc with phenol as the proton donor that makes CO 2 and HCO 2 À ergoneutral in the standard state, H 2 at a relative free energy of 6.7 kcal mol À1 (eqn (4)), and CO with an upper bound of ca. 11.5 kcal mol À1 (see the ESI †). 44 Two-electron reduction of the metal complex [M-S] n+ with the dissociation of S is followed by protonation at the metal center to form the metal hydride intermediate. CO

In silico screening of catalyst candidates
Within this class of reactivity, we rst calculate two key thermodynamic quantities, viz. the free energy of the metal hydride ðDG MÀH Þ and the metal carbonyl intermediates ðDG MÀCO Þ relative to the thermodynamically favorably product with weak proton donors, HCO 2 À . Given the same proton stoichiometry in eqn (1) and (3) independent of the choice of the proton donor, which will merely shi the relative energies of the stoichiometric products (Table S3 †). Scheme 3 lists the families of complexes we chose for a comparison of their reactivity within our design framework. They encompass d 6 earth-abundant metal ions (Mn(I), Fe(II) and Co(III)) and a diverse set of strong eld ligands including polypyridines, phosphines, amines and carbonyls to enhance metal-ligand binding.
The two thermodynamic quantities DG MÀH and DG MÀCO relative to HCO 2 À are plotted versus each other in Fig. 1 for all the complexes studied. These values are independent of the choice of the proton donor due to the equal proton stoichiometry in eqn (1) and (3). For selective and ergoneutral HCO 2 À production, the optimal value of DG MÀH is zero and DG MÀCO is greater than zero, and vice versa for CO production. DG MÀCO values greater than 10 kcal mol À1 , for example, would inhibit the formation of [M-CO] n+ at room temperature. 45 First, independent of the choice of metal, it is evident that all the data points fall roughly on a straight line for the cyclopentadienyl, pincer and bis-bipyridine frameworks. The slopes of the linear ts are all about À0.5, meaning that as the free energy of [M-CO] n+ decreases and the free energy of [M-H] (nÀ1)+ goes up. This is reasonable as an increase in electron density at the metal center is expected to increase the [M-H] (nÀ1)+ free energy but lower the [M-CO] n+ free energy due to increased back-bonding into the p* orbital of the carbonyl ligand. For the same metal, however, changing the ligands causes minor deviations from the best t line, presumably due to different extents of sigma and pi interactions in these complexes. Second, all the Co complexes are clustered in the upper-le quadrant of the plot while all the Mn complexes are clustered in the lower-right quadrant across all the families of ligands. This suggests that while CO does not thermodynamically trap the Co(III) complexes, the corresponding hydrides are all very stable relative to HCO 2 À and therefore unreactive towards CO 2 .
Conversely, in the Mn(I) systems, the hydrides lie higher than HCO 2 À in relative energy, and therefore will react with CO 2 , but the corresponding carbonyl intermediates are very stable potentially leading to catalytic trap states. It is evident that of the three metals, the complexes of Fe are the best potential catalyst candidates as their DG

MÀH and DG
MÀCO values lie closest to the origin.
The slopes of the ts in all three plots are roughly the same but the y-intercepts vary. Notice that changing the overall charge only moves you along the best t line. In the cyclopentadienyl and pincer complexes ( Fig. 1A and B), for example, the y-intercept is negative (ca. À5 kcal mol À1 ) suggesting that [M-CO] n+ will be dominant irrespective of the other ligands. The bis-bipyridine family of complexes, on the other hand, have a positive y-intercept of ca. 5 kcal mol À1 (Fig. 1C), resulting in a ca. 10 kcal mol À1 spread, which is greater than the ca. 5 kcal mol À1 error associated with such calculations. The bisbipyridine iron complex with L ¼ pyridine (DG MÀH ¼ À1:0, DG MÀCO ¼ 4:0) is therefore an interesting candidate to test the predictions for ergoneutral metal hydride formation and the lack of catalyst trapping by CO, within an error of AE5 kcal mol À1 in the calculations (vide infra).
While the exact origin of the different intercept in the case of the bis-bipyridine family compared to the cyclopentadienyl and pincer families is unclear, we hypothesize that having a high degree of pyridyl ligation could play a role, presumably from a good balance between the s-donating and p-accepting ability of pyridyl ligands. The former is responsible for conferring nucleophilicity on the metal center leading to a higher DG MÀH , and the latter for destabilizing [M-CO] n+ through competition for p-back donation from the metal center.  To further probe this hypothesis, we modeled the effect of increasing the p-accepting ability of the pyridyl ring on DG MÀH and DG MÀCO for the [Fe(bpy) 2 (py)S] 2+ system (Fig. 2). Firstly, conjugating the pyridine with one of the bipyridine ligands to make a terpyridine (tpy) ligand reduces the pyridine to Fe-CO dihedral angle (C-N-Fe-C) from 30 to zero, and aligns the p* orbital of the pyridine with the Fe-CO axis, which destabilizes the carbonyl ligand through competition for back-bonding. Consistent with this prediction, DG MÀCO for the [Fe(bpy)(tpy)] 2+ system increases by ca. 3 kcal mol À1 with almost no effect on DG MÀH . Secondly, substitution at the para-position of the pyridine ring with a nitro group has a similar effect on DG MÀH and DG MÀCO . Consistent with the known trapping of the [Ru(bpy)(tpy)] 2+ system by CO, 38 the Ru analogs have much lower values of DG MÀCO owing to the greater back-bonding ability of Ru compared to Fe. While the p-accepting nature of the ligand trans to CO in a metal complex is regularly invoked as a determinant for thermodynamic stability (the trans inuence), our results highlight a substantial cis inuence on the energy of the carbonyl (but not of the hydride), similar to effects we have seen previously. 24

Electrocatalytic activity of [Fe(bpy2PYMe)(CH 3 CN)] 2+
In order to experimentally test these predictions, we require a pentadentate ligand to prevent ligand scrambling. Long et al. recently reported the synthesis and characterization of the pentadentate ligand 'bpy2PYMe' (Scheme 4). 40 This ligand closely matches the desired ligand environment in Fig. 1C for iron (L ¼ pyridine), and therefore we synthesized and studied the electrocatalytic activity of the complex [Fe(bpy2PYMe)(CH 3 CN)] 2+ .
The calculated free energies of the catalytic intermediates for the different two-electron reduction processes (eqn (1)-(3)) mediated by the complex [Fe(bpy2PYMe)(CH 3 CN)] 2+ are shown in Fig. 3, with phenol as the stoichiometric proton donor. In addition to possessing low corrugations with respect to DG MÀH and DG MÀCO , viz. 1.7 kcal mol À1 and 7.4 kcal mol À1 respectively, the most endergonic on-path intermediates for this system are within 5 kcal mol À1 . This complex is therefore likely to avoid high thermodynamic barriers and operate close to the standard potential for the reduction reactions. We note that in Fig. 3, the metal formate intermediate is predicted to be a product inhibitor. Previous work suggested that formate can be labilized with the modication of the solvent composition to provide H-bond stabilization. 12,24 The cyclic voltammogram of [Fe(bpy2PYMe)(CH 3 CN)] 2+ in acetonitrile is shown in Fig. 4. The complex undergoes two oneelectron reductions at À1.55 V and À1.66 V vs. Cp 2 Fe +/0 (Cp 2 Fe ¼ ferrocene). The experimental redox potentials match up perfectly with those reported by Long and coworkers. 40 2+ complex is in negligible concentration, which is in agreement with our calculations (Fig. 3).
Upon the addition of 0.3 M phenol (pK a ¼ 29.14 in acetonitrile 42 ) to the CV solution containing 2 mM [Fe(bpy2PYMe)(CH 3 CN)] 2+ , there is a slight enhancement in the current at the second reduction potential suggesting that H 2 evolution is taking place through the formation of a metal hydride and subsequent protonation. Saturating this solution with CO 2 (ca. 0.28 M) 47 results in a higher (ca. 2.5-fold) current enhancement near the second reduction potential (ca. À1.66 V vs. Cp 2 Fe +/0 ), indicating that CO 2 is a substrate for an electrocatalytic process mediated by [Fe(bpy2PYMe)(CH 3 CN)] 2+ . Assuming a 2 mM concentration of phenoxide and a 300 mM concentration of phenol 48 we calculate that this electrocatalytic process is occurring at an overpotential of ca. 200 mV, from eqn (4) with pK a ¼ 29.14 for phenol. Scan rate dependent ratios of the current in the presence of CO 2 to the peak current in its absence (eqn (5), Fig. 4A (inset) and S2 †) yield a pseudo-rst order rate constant of ca. 2.4 s À1 for the two-electron electrocatalytic process involving the reduction of CO 2 . Unexpectedly, there is also a positive shi of the onset potential of the rst peak in the CV with no corresponding current increase at this potential. Upon the addition of CO 2 in the absence of phenol (Fig. S3 †) a similar positive shi in the onset potential is observed in the CV. Both these results suggest that there is  energetically favorable CO 2 binding accompanying the rst electrochemical reduction of [Fe(bpy2PYMe)(CH 3 CN)] 2+ , giving rise to the positive shi in the onset in the presence of CO 2 . On this basis, we assign the oxidative feature around ca. À1.35 V vs. Cp 2 Fe +/0 to the [Fe(bpy2PYMe)(CO 2 )] 2+/+ couple. Fig. S4 † shows that the experimental data can be approximately simulated by CO 2 binding and a small amount of reductive deoxygenation at extremely negative potentials, presumably due to carbonate formation. 33 This unexpected result of favorable CO 2 binding to the singly reduced intermediate [Fe(bpy2PYMe)] + , contrary to the DFT-based predictions in Fig. 3, is discussed below.
The effect of proton strength on the electrocatalytic behavior of [Fe(bpy2PYMe)(CH 3 CN)] 2+ was explored with acetic acid (pK a ¼ 23.51 in acetonitrile) 42 as the proton donor instead of phenol. In the cyclic voltammograms (Fig. 4B), the electrocatalytic currents near the second reduction potential of [Fe(bpy2PYMe)(CH 3 CN)] 2+ are over three times greater than with phenol. However, the overpotential, as compared to the phenol case, is about 200 mV higher, as E HER ¼ ÀDG HER =2F ¼ À1:4 V vs: Cp 2 Fe þ=0 in the standard state (Table S1 †). With 0.5 M acetic acid, a pseudo-rst order rate constant greater than 150 s À1 (the actual rate is limited by the cell resistance) was estimated from scan-rate dependent CVs (Fig. S2 †). Upon the addition of CO 2 (ca. 0.28 M) there is a greater current enhancement and positive shi of the onset potential (Fig. S5 †), qualitatively similar to what was observed with phenol. Bulk electrolysis of a 2 mM solution of [Fe(bpy2PYMe)(CH 3 CN)] 2+ was rst performed with 0.3 M phenol and 0.1 M KPF 6 as the supporting electrolyte in CO 2saturated CD 3 CN (3 mL solution) at a potential of À1.65 V vs. Cp 2 Fe +/0 near the onset of the second reduction wave (Fig. 4A,   Fig. 3 DFT-calculated energy level diagram (standard state) for CO 2 reduction to HCO 2 À (in black) and CO (in grey), and proton reduction to H 2 (in grey), mediated by [Fe(bpy2PYMe)(CH 3 CN)] 2+ , with phenol as the proton donor.  see the Experimental details). Decomposition of the Fe species in solution preceded any appreciable turnover, and therefore no products were detected. The electrolysis experiment was then repeated with 0.3 M acetic acid as the proton donor instead of phenol. A 1 mL aliquot of the headspace was injected into a gas chromatograph to detect and quantify gaseous products (Fig. S6 †). Based on the calibration of the GC peak areas with 1% gas standards, the amounts of gaseous products were estimated to be 8.1 Â 10 À6 moles of CO and 2.6 Â 10 À5 moles of H 2 . They correspond to 9% and 30% of the total charge passed in the electrolysis experiment respectively for two-electron stoichiometry, suggesting that other reduced products are present.
In the 1 H NMR spectrum of the electrolyte solution, the peak corresponding to HCO 2 À at ca. 8.2 ppm 49 was not observed.
Decomposition of the Fe species in solution prevented comprehensive analyses of the liquid phase products. With acetic acid as the proton donor, H 2 and HCO 2 À are predicted to be roughly ergoneutral in the standard state (Fig. S1 †). However, under the electrolysis conditions, there is a ca. 200 mV extra driving force towards H 2 from the higher proton activity due to the absence of a stoichiometric conjugate base, as well as from homoconjugation effects. 42,48 These effects along with the second reduction potential of [Fe(bpy2PYMe)(CH 3 CN)] 2+ , À1.66 V vs. Cp 2 Fe +/0 (and the electrolysis potential), being ca. 400 mV more negative than the potential for H 2 production under the electrolysis conditions (vide supra), together explain the observed H 2 in the electrolysis. We hypothesize that the lack of the thermodynamically favored product, HCO 2 À , under the electrolysis conditions, is due to the rate of CO 2 insertion into the metal hydride being signicantly slower than the rate of protonation of the metal hydride. CO 2 insertion into hydrides is typically a slow process (ca. 1.82 Â 10 À2 M À1 s À1 for [Ru(tpy)(bpy)H] + at 298 K in acetonitrile 50 ). The incorporation of proton-directing groups into the ligand backbone to accelerate the rate of CO 2 insertion into the metal hydride, as we have explored in previous work, 24 would favor the HCO 2 À pathway.
The production of CO as a minor product indicates direct binding of CO 2 during the electrochemical reduction of [Fe(bpy2PYMe)(CH 3 CN)] 2+ . The direct CO 2 binding pathway, evidenced by the positive shi of the rst reduction wave in the CV in Fig. S3, † competes with hydride formation (Scheme 2). However, DFT calculations at the current level of theory suggest that the binding of CO 2 aer the rst reduction is uphill by ca. 7 kcal mol À1 (Fig. 3). The same calculation using the hybrid functional B3LYP instead of BP86 suggests that the CO 2 binding step aer the rst reduction is downhill by about 4 kcal mol À1 . It is, therefore, possible that while BP86 captures redox potentials and the free energy of key intermediates such as the metal hydrides and metal carbonyls with reasonable accuracy, it underestimates the free energy of CO 2 binding to the reduced Fe-complexes. Therefore, a catalyst search algorithm that employs the free energies of the metal hydride and carbonyl intermediates alone (Fig. 1) will effectively weed out bad candidates based on these two descriptors. However, within this narrowed space of candidates, favorable CO 2 binding to the metal center could lead to CO production, as evidenced in the case of [Fe(bpy2PYMe)(CH 3 CN)] 2+ . This motivates further renement and experimental validation of DFT methods to adequately model the thermodynamics of CO 2 binding to reduced metal complexes, in order to map the free energies of the whole range of intermediates in a catalytic pathway.

Conclusions
In this work we have shown how the calculation of two thermodynamic parameters, viz. the relative free energies of the metal hydride and the CO-bound intermediates by using DFT streamlines the search for an appropriate transition metal and ligand environment for catalyzing the multi-electron electrochemical reduction of CO 2 or protons at low driving forces. We applied this catalyst screening approach on a library of common ligands around three earth-abundant d 6 metal ions and tested the in silico predictions via the synthesis and electrochemical studies of an Fe-based complex, predicted to avoid trapping by CO, form a metal hydride of appropriate free energy, and possess optimal redox potentials. These predictions were validated, and we found the iron-based electrocatalyst to be active towards CO 2 reduction as well as H 2 production at the predicted potentials. The model, however, underestimated the free energy of CO 2 binding to the reduced metal complex. We highlight the need for renements to the DFT-methods to adequately capture the free energy of this latter step that leads to CO production, in order to further narrow the candidate space. This work is, to the best of our knowledge, the rst application of DFT as a screen for molecular electrocatalysts across diverse molecular environments and paves the way forward for computations to take the lead in identifying promising catalysts for electrocatalytic transformations.

Experimental details
The ligand bpy2PYMe and its Fe(II) complex were prepared as per reported protocols. 40  All electrochemical experiments were performed inside a N 2 glovebox tted with a CO 2 feedthrough using a SP-200 potentiostat from Bio-Logic Co. For the cyclic voltammetry experiments, a glassy carbon electrode (3 mm diameter) from BASi Inc. was used in combination with a Ag/AgNO 3 (10 mM)/TBAPF 6 (100 mM) reference electrode and a Pt wire counter electrode in a sealed cell. The pseudo-rst-order rate constant k cat for the catalytic waves in Fig. 4 was determined using eqn (5) 12 (n is the scan rate).
Bulk-electrolysis was performed in an 'H-cell' with two compartments, one for the working and reference electrodes and the other for the counter electrode, separated by a porous glass frit. The working electrode was a glassy carbon rod (BASi Inc.), and the reference and counter electrodes were the same as used in the cyclic voltammetry experiments. Both compartments were sealed with rubber septa and stirred vigorously during electrolysis. The entire process was carried out in a N 2 -lled glove box. Headspace product analyses were performed using a Shimadzu gas chromatograph tted with a 100 mL sample injection loop, an FID, a TCD and a CarbonPLOT column with N 2 carrier gas. Percent H 2 and CO were determined with one-point calibration with a 1% gas standard purchased from Sigma-Aldrich Co.

Computational details
All calculations based on Kohn-Sham density functional theory were performed using the Gaussian 09 51 (Rev D. 01) soware package. The BP86 (ref. 52 and 53) functional was used for all the calculations in combination with a double-zeta 6-31+G* [54][55][56] basis set on all the p-block elements, and the LANL2DZ 57 effective core potential on the transition metals. The relevant intermediates were rst optimized in the gas phase followed by a harmonic analysis on the stationary point to obtain enthalpic, entropic and zero-point energy corrections to the electronic energy in the standard state, as implemented in Gaussian 09. The solvation energy was then determined with a single point calculation with a polarizable continuum model (SMD) for acetonitrile. 58 Standard state corrections were made to the free energies to account for the change in going from 1 mol per 24.46 L (gas phase) to 1 M (solution phase). This level of theory provided excellent agreement with experimental values of the reduction potentials versus the ferrocene/ferrocenium couple in acetonitrile (eqn (6)-(8)), as well as hydricities for representative complexes in each family of complexes listed in Scheme 3. 37,59,60 Ox n+ + Cp 2 Fe # Red (nÀ1)+ + Cp 2 Fe + For all the complexes reported in this work, the lowest spin states (singlets for the even-electron intermediates and doublets for the odd-electron intermediates) were found to be the thermodynamically favorable states. This is expected for the local density functional BP86 given the choice of strong-eld ligands. 20 The B3LYP 52,61-63 functional was employed in one instance for comparison of the CO 2 binding energies to the singly reduced Fe complex. Simulations of the electrochemical data were performed with Digielch 64 using a planar semi-innite 1D diffusion model (see the ESI †).

Conflicts of interest
The authors declare no competing nancial interests.