The Role of Surface-Bound Dihydropyridine Analogues in Pyridine-Catalyzed CO2 Reduction over Semiconductor Photoelectrodes

We propose a general reaction mechanism for the pyridine (Py)-catalyzed reduction of CO2 over GaP(111), CdTe(111), and CuInS2(112) photoelectrode surfaces. This mechanism proceeds via formation of a surface-bound dihydropyridine (DHP) analogue, which is a newly postulated intermediate in the Py-catalyzed mechanism. Using density functional theory, we calculate the standard reduction potential related to the formation of the DHP analogue, which demonstrates that it is thermodynamically feasible to form this intermediate on all three investigated electrode surfaces under photoelectrochemical conditions. Hydride transfer barriers from the intermediate to CO2 demonstrate that the surface-bound DHP analogue is as effective at reducing CO2 to HCOO– as the DHP(aq) molecule in solution. This intermediate is predicted to be both stable and active on many varying electrodes, therefore pointing to a mechanism that can be generalized across a variety of semiconductor surfaces, and explains the observed electrode dependence of the photocatalysis. Design principles that emerge are also outlined.


1) DFT Computational Details
All computations were performed using the NWCHEM 6.6 simulation package 1 with the hybrid B3LYP 2,3 exchange-correlation (XC) functional and Grimme's D2 4 semi-empirical dispersion correction. Continuum solvation was treated with the "Solvation Model based on solute electron Density" (SMD), 5 where the default NWCHEM parameters were employed for all atomic radii as well as for the dielectric constant of water (ε = 78.4). Inner core electrons/nuclei and outer-core/valence electrons of Ga, Cd, Te, Cu, and In were respectively represented by the following effective core potentials (ECPs) and double zeta (DZ) basis sets: Stuttgart (28, MWB) + DZ, 6 Stuttgart (28, MWB) + DZ, 7 Stuttgart (46, MWB) + DZ, 6 Stuttgart (10, MWB) + DZ, 8,9 and Stuttgart + DZ (46,MWB). 6 The numbers in parentheses are the number of core electrons replaced by the ECP. MWB is the designation in the Stuttgart database indicating that the ECP is derived through fitting to multi-electron, quasi-relativistic data sets.
Geometries were optimized with a quasi-Newton-Raphson algorithm (as available in NWCHEM 6.6) with a force convergence criterion of 0.00045 Ha bohr -1 . Minimum energy structures were verified with frequency analyses to ensure that all imaginary modes were eliminated. Transition state (TS) structures were also verified with frequency analyses, in which all reported TS structures have one imaginary frequency, and where optimization along both directions of the imaginary mode leads to minima corresponding to super-molecule structures characteristic of the reactant and product states (i.e., CO 2 and HCOOweakly adsorbed above the 2-PyH -* and Py * surface models). Note that these super-molecule reactant states vary from the reference state used to determine the effective free energy barriers reported in the main text (i.e., there the effective reference state consists of CO 2(aq) in solution at infinite separation from the surface). Numerical Hessians were employed to calculate frequencies, which were calculated with finite displacements (± 0.01 Å) of all atoms except the fixed pseudo-hydrogen saturators (see section 3 below).

2) Calculation of Standard Reduction Potentials (SRPs)
Standard reduction potentials were calculated from reaction free energies employing a previously established methodology. 12,13 The SRP is calculated with the equation: where ‫ܧ‬ is the SRP, n is the number of electrons transferred in the reduction, m is the number of protons involved in the reaction, F is the Faraday constant, T is the temperature (T = 298 K at standard state), R is the gas constant, and pH = 5.2. ‫ܩ∆‬ is the reaction free energy, which is calculated with the expression: where ‫ܩ‬ ௗ௨ௗ and ‫ܩ‬ ௧௧ are the free energies of the reduced product and the reactant (including implicit solvation, as well as all translational, rotational, and vibrational contributions), respectively, and ‫ܩ‬ ି is the free energy of an electron in solution, as determined empirically from the standard hydrogen electrode (SHE = -4.281 V). 14 ‫ܩ‬ ுା is the free energy of a proton in solution, determined empirically as -11.72 eV. 15 All reduction potentials are reported relative to the standard calomel electrode (0 V-SHE = -0.244 V-SCE). Statistical mechanical expressions for the ideal gas, harmonic oscillator, and rigid rotor were employed to derive translational, vibrational, and rotational thermochemical properties. Translational and rotational contributions were considered to be zero for species adsorbed on cluster model surfaces with no translational or rotational degrees of freedom. The appropriateness of the employed implicit solvation model for calculating SRP values of the species considered in this study was demonstrated previously by Keith and Carter. 12,16 The computed SRP for the reduction of PyH + (aq) to PyH• (aq) using this approach is corroborated in the works of multiple separate research groups employing both theory [17][18][19] and experiment, 20 demonstrating the feasibility of this method.
Finally, Keith and Carter 21 further demonstrated that SRP values for the Py-derived species considered in this study calculated with B3LYP/aug-cc-pVDZ and with high level (U)CCSD(T)-F12/aug-cc-pVTZ-F12 approached varied by less than 0.1 V. Thus, the B3LYP functional is an appropriate choice for the calculation of adsorption energies and SRP values presented in this manuscript.

3) Cluster Models
GaP (111) and CdTe(111) surfaces were represented with the cluster models derived in our previous study. 22 These models were cleaved from the periodic structures of a (2×2) reconstruction featuring one surface Ga/Cd vacancy per unit cell, as this reconstruction was predicted to be stable under photoelectrochemical conditions ( Figure S1). The resulting clusters have 24 P/Te atoms, with 12 residing in the surface layer and 12 residing in the sub-surface layer. They have 21 Ga/Cd atoms, with nine residing in the surface layer and 12 in the subsurface layer (i.e., the surface layer has one Ga/Cd vacancy per four P/Te atoms). Innocent dangling bonds at the cluster boundary were saturated using a pseudo-hydrogen capping scheme, in which each pseudo-hydrogen cap has a core charge modified to represent the atom-type it is replacing. Core charges of Z = +3/4 e, Z = +5/4 e, Z = +2/4 e, and Z = +6/4 e are used when replacing Ga, P, Cd, and Te, respectively. The core charge is determined from the number of valence electrons divided by the number of bonds in the stoichiometric bulk (i.e., for Ga there are three valence electrons divided by four bonds in the zinc blende structure). This stoichiometry yields a neutral singlet with no dangling bonds. The cluster model of the CuInS 2 (112) surface was derived from the periodic geometry of a (2×2) surface reconstruction, in which there is one Cu In anti-site defect per (2×2) surface cell. This defect was predicted to be thermodynamically stable using the same methodology that was applied to the GaP and CdTe surfaces. 23 The resulting cluster has 12 Cu atoms, four In atoms, and 16 S atoms in the surface layer, and has eight Cu atoms, eight In atoms, and 16 S atoms in the sub-surface layer (i.e., there are four Cu In anti-site defects in the equivalent (4×4) surface layer). Pseudo-hydrogen saturators were employed with core charges of Z = +1/4 e, Z = +3/4 e, and Z = +6/4 e when replacing Cu, In, and S, respectively. Pseudo-hydrogens were positioned by replacing a cation/anion in the geometry of the optimized extended surface, followed by an optimization of all pseudo-hydrogen bond lengths (where all atoms in the cluster are frozen and pseudo-hydrogen saturators are relaxed along their bond to the nearest atom in the cluster). Pseudo-hydrogens were then frozen in these positions during all subsequent geometry optimizations and vibrational frequency calculations. We note that these models of the surface assume that restructuring of the surface caused by exposure to visible light and the applied voltage is negligible. The voltages applied in the photo-electrochemical experiments are generally low-with the onset of Py-enhanced CO 2 reduction demonstrated to occur at potentials as low as -0.2 V vs. SCE. 24 We therefore do not expect that voltage-induced restructuring of the surface will play a significant role in these systems.
The cluster model approach was extensively benchmarked and validated by Keith et al.,13 where it was demonstrated that the cluster model approach yields adsorption energies that are 6 very similar to those computed with a periodic surface model. The cluster model approach was further refined, extended, and validated by Senftle et al. 22,23 These studies show that the cluster models yield adsorption energies that are typically within ~0.2 eV of the values computed with periodic surface models. These studies also show that the B3LYP XC functional yields adsorption energies consistent with the widely used PBE 25 XC functional.   (111) and (b) CuInS 2 (112) surfaces. The red (blue) isosurface indicates electron density depletion (accumulation), and the isosurface level corresponds to 0.003 ebohr -1 . Figure S4. Side view of the TS geometry of a HT from the 2-PyH -* intermediate to CO 2 over the (a) CdTe (111) and (b) CuInS 2 (112) surfaces. Pseudo-hydrogen saturators are omitted for clarity. Figure S5. (a-b) Top (left) and side (right) view of the geometry of 2-PyH -* adsorbed on the reconstructed GaP (111)  6) Reaction energy diagrams for mechanisms proceeding through 2-PyH -* or DHP * Figure S7. CO 2 reduction pathways proceeding through HT from surface-bound 2-PyH -* (blue) or DHP * (orange) intermediates on GaP(111). Species labeled with an * are adsorbed on the surface. (Inset) Side view of the TS-(DHP* + CO 2 ) transition state structure.

7)
Transition state for the reaction: H -* + CO 2 * + HCOOon GaP(111) Figure S8. Side view of the TS geometry of a HT directly from the reconstructed GaP (111) surface. Pseudo-hydrogen saturators are omitted for clarity.

8) Interaction between Py, H 2 O and β-Ga 2 O 3
In this section we consider the possible role of a native surface oxide on the photo-

9) Cartesian coordinates and total DFT energies of all reported geometries
All geometries are provided in cartesian coordinates (Å). Core charges (units of e) of pseudohydrogen saturators are indicated in brackets.