Explaining the structure sensitivity of Pt and Rh for aqueous-phase hydrogenation of phenol

Upgrading bio-oils derived from waste biomass is a route to sustainably produce fuels and chemicals and reduce our dependency on petrochemicals.^1–3 Currently, the aqueous-phase hydrogenation step in biomass conversion is too capital- and energy-intensive to compete economically with fossil fuels.^4,5 Many studies have explored electrocatalytic hydrogenation (ECH) and thermocatalytic hydrogenation (TCH) to understand the reaction mechanism and improve the efficiency of bio-oil conversion.^6–9 The most active ECH and TCH catalysts are platinum group metals,^6,10–16 typically dispersed as nanoparticles onto supports to increase fractional exposure and improve catalyst utilization. However, the surfaces of these nanoparticles have multiple exposed facets, not all of which are active for hydrogenation.^10,11 Thus, decreasing the particle size to increase the catalyst surface area may actually decrease the catalyst performance and utilization because a lower fraction of active sites may be exposed at small particle sizes. A better understanding of why certain catalyst sites are active for hydrogenation is required to effectively upgrade the many compounds present in the bio-oil. We hypothesize that the adsorption energies of the bio-oil compound and hydrogen on the active site largely affect the catalytic activity, as expected from the Sabatier principle;¹⁷ however, it may not be the sole predictor (i.e., intrinsic differences in kinetics between surfaces may also affect the rate). Ideally, the aqueous-phase hydrogenation activity of a given molecule on a catalyst surface could be qualitatively estimated based on simple adsorption energy calculations,^18,19 which would accelerate the computational screening of active catalysts for aqueous-phase reactions on metal surfaces. Thus, it is important to know if adsorption energies alone are sufficiently accurate as descriptors to predict trends in hydrogenation kinetics.

In this work, we use computational and experimental methods to identify the active sites of Pt and Rh for ECH and TCH of phenol, a model bio-oil compound, and explain why these sites are active. We study phenol because it is a representative of phenolic compounds within many bio-oil mixtures,^3,20,21 and Pt^10–12 and Rh^10,13–15 are among the most well-studied catalysts for ECH and TCH of phenol. Phenol hydrogenation is structure-sensitive on Pt, and the Pt(111)-like sites [i.e., (111) terraces] have been shown to be the active facet for ECH and TCH;^10,11 however, the reason why only this Pt facet is active for phenol ECH and TCH is unclear. In addition, it is unclear if phenol ECH and TCH are also structure-sensitive on Rh or what the active site is for Rh. One hypothesis is that certain facets are active due to weaker adsorption of phenol, which prevents phenol from poisoning the catalyst surface and blocking hydrogen adsorption under typical operating conditions.^10–15 

Herein, we show that ECH and TCH of phenol are structure-sensitive on Rh, similar to what has been shown for Pt. We test the hypothesis that certain Pt or Rh facets are active due to weaker phenol adsorption. Computational and experimental data for phenol ECH and TCH on Pt and Rh corroborate our hypothesis that the adsorption energy of phenol influences the reaction rate on the active site, but other factors, namely, the activation energies, also contribute to the activity. We emphasize that although the adsorption energy of the reacting compound influences the reaction rate, it is not the sole descriptor of a catalyst’s activity.

The electrocatalytic and thermocatalytic hydrogenation of phenol on both Pt/C and Rh/C have been reported to follow a Langmuir–Hinshelwood (LH) mechanism. This LH mechanism involves the competitive adsorption of phenol and hydrogen followed by sequential surface reactions of adsorbed hydrogen (H*) and the adsorbed organic to form cyclohexanone and cyclohexanol.^10,15 The major difference between ECH and TCH of phenol is the source of adsorbed hydrogen. For ECH, hydrogen equivalents are produced on the catalyst surface by reducing protons from the aqueous solution using an applied cathodic potential. For TCH, the H* is derived from dissociative adsorption of H₂. ECH and TCH of phenol have similar apparent activation energies and product distributions at applied potentials where the ECH rate is comparable to that of TCH at 1 bar H₂, implying the same rate-determining step (RDS).^13–15 The surface reaction between adsorbed phenol and H* has been proposed as the RDS for ECH and TCH on Pt/C, indicating that the turnover frequency (TOF) is controlled by the rate constant (k) and equilibrium coverages of phenol (θ_P) and hydrogen (θ_H), according to the LH model described by the following equation:

Because the surface reaction is rate-determining, θ_P and θ_H are assumed to be quasi-equilibrated and controlled by the aqueous equilibrium adsorption constants on the active site (K_P and K_H, where P = phenol and H = hydrogen). The TOF can be written in terms of the bulk concentration of phenol divided by the standard concentration of 1M (C_P) and the aqueous hydrogen concentration (C_H) in the following equation:

The C_H is equivalent to [P_H2/(1 bar)]^1/2 for TCH, where P_H2 is the pressure of hydrogen in equilibrium with the solution. For ECH, C_H is the concentration of H⁺ divided by 1M.¹⁰ For TCH, K_H is the equilibrium adsorption constant of H* from ½ H₂. For ECH, K_H is also the equilibrium adsorption constant of H*, but from H⁺ and e⁻, thus, K_H increases with more negative applied potential due to the increased thermodynamic driving force to form H*. The potential dependence of phenol ECH TOFs on Pt/C and Rh/C is qualitatively explained by Eq. (2) by considering this change in K_H.¹⁰ Briefly, the TOF increases with more negative applied potential at low hydrogen coverages and decreases with more negative applied potential at high hydrogen coverages due to competitive adsorption with phenol.¹⁰ 

Solvent can influence the hydrogenation rate of organic molecules by changing their adsorption thermodynamics and hydrogenation barriers of elementary steps. Upon adsorption, organic molecules with large footprints, such as aromatics, must displace solvent molecules at the solvent/metal interface. The aqueous-phase adsorption energies of aromatic compounds are weaker than those measured in the gas phase because of the energetic penalty of solvent displacement.^18,22 For example, the adsorption free energies of phenol on Pt and Rh are over 150 kJ mol⁻¹ weaker in the aqueous phase compared to the gas phase due to the displacement of multiple water molecules upon adsorption.^18,19,23 At room temperature, the weaker adsorption energy of phenol in the aqueous phase increases reaction rates compared to the gas phase because of more balanced coverages of phenol and hydrogen. The water solvent can lower barriers through two ways: first, water solvation can reduce the energy of transition states more than reactant species, and second, water can directly participate in transition states to facilitate bond breaking and formation. In aqueous-phase hydrogenation reactions on metals, water is predicted to lower the barriers for the hydrogenation of C=C bonds by less than 10 kJ mol⁻¹ through solvation.^24–26 In contrast, water is predicted to lower the barriers of O–H bond breaking by >35 kJ mol⁻¹ through direct participation in transition states.^24–29 

The LH model [Eq. (2)] has been shown to approximate the rates of ECH and TCH of phenol on platinum group metals.^10,14,30 Equation (2) has been used to fit the TOF measured over a range of phenol concentrations for the TCH of phenol on Pt/C.¹⁰ The extracted K_P value on the active site was compared with K_P values of stepped facets and Pt(111)-like sites extracted independently from adsorption isotherms on polycrystalline Pt.^10,31 The active facet for phenol hydrogenation was attributed to Pt(111)-like sites due to similar K_P values. However, whether the activity of Pt(111) is due to fast intrinsic hydrogenation kinetics (larger k) or optimal coverages of the adsorbed reactants (higher θ_Pθ_H) is unknown. In the case of Rh, even the active site and reason for its activity are unknown, despite Rh/C being a commonly studied metal for phenol hydrogenation.^10,13–15

Here, we elucidate which Rh facet is active for phenol hydrogenation and explain the origin of the high TOF on the active sites of Pt/C and Rh/C. Using density functional theory (DFT) modeling, microkinetic simulations, and experimental kinetic measurements, we address whether the active site for phenol hydrogenation is a step or terrace facet and how intrinsic kinetics (i.e., transition state energies) and the phenol adsorption energy govern the phenol hydrogenation activity of Pt/C and Rh/C. We predict that (111) terraces of Pt/C and Rh/C are the active sites for phenol hydrogenation due to faster intrinsic kinetics and weaker adsorption of phenol compared to (221) stepped facets. We report that the TOF on Rh/C for ECH of phenol increases with particle size, similar to what has been shown previously on Pt/C, consistent with the active sites for phenol hydrogenation being the (111) terraces. These results indicate that synthesis techniques to preferentially expose (111) terraces on supported nanoparticles should be used to optimize the utilization of metals, increase bio-oil hydrogenation rates, and reduce the cost of bio-oil valorization.

The Vienna Ab initio Simulation Package (VASP) was used for all DFT calculations^32–35 along with the Atomic Simulation Environment (ASE) interface.³⁶ The Perdew–Burke–Ernzerhof functional³⁷ with the semi-empirical D3 dispersion correction³⁸ was chosen because of its reasonable accuracy and tractable computational expense.³⁹ The projector augmented wave method and a plane wave kinetic energy cutoff of 400 eV were used for all the calculations.^40,41 A first-order Methfessel–Paxton smearing scheme with a 0.2 eV smearing width was used.⁴² 

The Pt(111), Pt(221), Rh(111), and Rh(221) surfaces were modeled using 3 × 3 × 4 slabs consisting of 36 metal atoms. The bottom two layers of each surface were fixed in their bulk lattice coordinates, and the top two layers could relax during geometry optimization. Metal slabs were separated by a 20 Å vacuum. The bulk lattice parameters of Pt (3.925 Å) and Rh (3.972 Å) were determined by relaxing four-atom face centered cubic unit cells of both metals. A 16 × 16 × 16 Monkhorst–Pack k-point grid was used when determining the bulk lattice parameters.⁴³ For surface calculations, a 5 × 5 × 1 Monkhorst–Pack k-point grid was used. All DFT calculations were non-spin polarized. Spin polarization was found to have a negligible effect on the adsorption energy of phenol on Pt and Rh. Geometries were optimized until the electronic energy and ionic forces were converged to within 10⁻⁵ eV and 0.01 eV Å⁻¹. The climbing-image nudged elastic band (CI-NEB) method was used to find transition states of the hydrogenation reaction.⁴⁴ CI-NEB images were optimized until the electronic energies and ionic forces were within 10⁻⁵ eV and 0.05 eV Å⁻¹. The free energies for phenol and all reaction intermediates were evaluated in the harmonic limit at 300 K (including translation, rotational, and vibrational free energy contributions). The free energies of gas-phase species were evaluated using ideal gas thermodynamic corrections at 300 K and 1 bar.

Co-adsorbed hydrogen is typically present on the catalyst surface in acidic solutions.^45–48 The Gibbs free energies of adsorbed phenol with co-adsorbed hydrogen on the (111) terraces and (221) steps of Pt and Rh were calculated as a function of hydrogen coverage at 300 K and 1 bar H₂. The system with the lowest free energy on each surface was chosen to model the phenol hydrogenation reaction. On the (111) surfaces, hydrogen was placed in the same configurations used previously to model benzene hydrogenation with co-adsorbed hydrogen.⁴⁹ The lowest energy configuration on Pt(111) and Rh(111) consisted of phenol adsorbed in the bridge-30 configuration with 4/9 monolayer (ML) coverage of co-adsorbed hydrogen (Figs. S1 and S2). On Pt(221) and Rh(221), phenol adsorbed at the step edge with 4/9 ML of hydrogen co-adsorbed on Pt(111) and 6/9 ML of hydrogen co-adsorbed on Rh(111) (Figs. S3 and S4). The relative free energy of adsorbed phenol with co-adsorbed hydrogen as a function of hydrogen coverage is shown in Fig. S5.

Mean-field microkinetic simulations were performed using the MKMCXX software⁵⁰ to predict TOFs of phenol hydrogenation on Pt and Rh. Here, we assume that adsorption of all species requires a single site in the microkinetic model. In reality, phenol adsorbs to multiple atoms, and thus, it would be more appropriate to require phenol to adsorb on an ensemble of surface atoms. Nonetheless, as we show below, the single site microkinetic model gives predictions that are qualitatively consistent with experimental observations on Pt. Forward and reverse reaction rate constants were calculated using harmonic transition state theory and DFT-computed activation free energies. Thermochemistry corrections were included for all reaction intermediates and transition states using the harmonic approximation. A constant pre-exponential factor of 6.25 × 10¹² s⁻¹ calculated using $kBTh$ at 300 K was used. Adsorption and desorption of phenol, hydrogen, and cyclohexanone were also treated using an Arrhenius model. Adsorption and desorption were assumed to be barrierless. In the microkinetic model, we use partial pressures of phenol and hydrogen at the same ratio as what is in solution experimentally.^51,52 The concentration of phenol was 20 mM, and the concentration of hydrogen in solution (0.78 mM) was estimated using Henry’s law for hydrogen in equilibrium with 1 bar H₂. The total system pressure was 1 bar.

Although the solvent affects the hydrogenation kinetics of aromatic molecules, we found that implicit solvation using VASPsol^53,54 for phenol hydrogenation had a negligible effect on the reaction energies and transition states and, thus, was not considered further (Fig. S6). Experimental aqueous-phase adsorption free energies of phenol on Pt and Rh were used in the place of DFT-computed free energies as inputs to the microkinetic model to capture the effects of solvation, solvent displacement, and changes in entropy upon adsorption of phenol. Due to the challenges in explicitly treating solvation of reactants and intermediates along each step in the reaction mechanisms, we assume that all species have solvation energies similar to phenol. The desorption energy of cyclohexanone was chosen to maintain thermodynamic consistency in the overall reaction free energy. The reaction free energy (−44 kJ mol⁻¹) was calculated using the reaction P_(aq) + 2H_2(g) ⇌ CHO_(aq), where the aqueous-phase free energies of phenol (P) and cyclohexanone (CHO) are from gas-phase free energies calculated from DFT that were shifted for solvation using Henry’s law constants of the species.⁵⁵ The average adsorption free energy of H* along the reaction energy diagram was used for the microkinetic model. Cyclohexanone was modeled as the sole product of phenol hydrogenation because the formation of cyclohexanone dominates the product distribution under experimental conditions with low phenol conversion (<10% conversion).¹³ 

Chemicals were obtained from Sigma-Aldrich and used as received. Aqueous sodium acetate buffer solution (3M, pH = 5.2 ± 0.1) was used as the supporting electrolyte for the ECH of phenol (≥99.0%). Ethyl acetate (≥99.8%) was used as the solvent for extraction, and dimethoxybenzene (99%) was used as the internal standard for gas chromatography. N₂ (>99.99%, Cryogenic Gases) was used to remove dissolved oxygen from the reactor. H₂ (ultra-high purity grade, Cryogenic Gases) was used for TCH. All water was purified with a Milli-Q system up to a resistivity of 18.2 MΩ cm. Rh/C purchased from Sigma-Aldrich (1, 3, 5, and 10 wt. %) was used for all ECH studies at a phenol concentration of 20 mM. For the TCH studies as a function of phenol concentration, the rates at low (10 mM) and high (900 mM) phenol concentrations were low enough that product quantification was difficult with the low total amount of metal for these catalysts. Therefore, a higher Rh metal loading, 20 wt. % Rh/C from Fuel Cell Store, was used for all TCH studies.

X-Ray Diffraction (XRD) analysis was done on the 20 wt. % Rh/C using a Rigaku Miniflex x-ray diffractometer with Cu Kα radiation (λ = 1.5418 Å) and a Ni filter. Samples were scanned between the 2θ range of 10° < 2θ < 90° at a rate of 5°/min. The Scherrer equation was used to calculate the average crystallite size of the 20 wt. % Rh/C as 2.9 nm using the full width at half maximum of the Rh(111), Rh(220), and Rh(311) diffraction peaks in Fig. S7. The XRD instrument line broadening was negligible compared to the estimated particle size and was, thus, ignored for calculating the particle size. For the TCH studies using this same 20 wt. % Rh/C, the number of surface metal atoms used to normalize the TOF was estimated from the particle size obtained by XRD (see the supplementary material). The particle diameters of Rh/C used to determine the particle size dependence for ECH of phenol were estimated to be 2, 5, 7, and 10 nm for the 1, 3, 5, and 10 wt. % loading, respectively, from transmission electron microscopy (TEM). Ground catalyst samples were suspended in ethanol, and microliters of the catalyst suspension were applied on a copper grid coated with carbon and dried before a TEM measurement was performed with a JEOL JEM-2011 electron microscope operating at 120 keV accelerating voltage. For further details and sample images, see Ref. 56. The number of surface metal atoms for the 1, 3, 5, and 10 wt. % Rh/C catalysts used for the particle size studies was determined from H₂ chemisorption as described previously.^13,56 The metal dispersions, or moles of surface Rh per total moles of Rh, of the 1, 3, 5, and 10 wt. % Rh/C samples were 0.43, 0.2, 0.135, and 0.1, respectively. Details on the weight loading of Rh supported on carbon, the commercial source of the catalyst, the reaction for which each catalyst was used, average particle sizes from TEM, XRD, and H₂ chemisorption and dispersion from H₂ chemisorption and XRD are shown in Table SI.

A 125 ml jacketed glass batch reactor was used to perform TCH of phenol. 10 mg of 20 wt. % Rh/C catalyst was added to Millipore water in the cell and then sparged with a flow of N₂ for 30 min at 100 ml min⁻¹ under continuous stirring at 500 rpm to remove oxygen and disperse the nanoparticles. A flow of H₂ at 70 ml min⁻¹ into the cell for 30 min was used to activate the catalyst, saturate the solution with H₂, and bring the headspace pressure to 1 bar H₂. Before introducing phenol, the H₂ flow was reduced to 25 ml min⁻¹ and kept constant throughout the reaction to maintain the reactor pressure at 1 bar H₂ and prevent air from entering the reactor. The temperature was kept at 25 °C using a refrigerated/heated bath circulator (Fisher Scientific). To attain a desired phenol concentration in the cell, a particular concentration of phenol in water, already sparged with N₂ to remove dissolved O₂, was introduced into the cell. A syringe connected to PEEK tubing was used to avoid possible contamination from a metal needle. The pH of the solution with phenol was 5.2–5.4, which is the same as the pH of phenol in acetate buffer used for ECH. Previous studies of TCH of phenol on Pt/C show that at this pH, the phenol TCH activity is essentially independent of the presence of acetate.¹⁴ The reactor mixture was stirred using a stir bar at 500 rpm for the duration of the reaction. At this stir rate, limitations on the measured rate due to mass transfer were eliminated (Fig. S8). The reaction start time was recorded immediately after the phenol was introduced into the cell. All measurements were performed under differential conditions (<10% conversion of phenol).

A two-compartment batch electrolysis cell was used for ECH studies.⁵⁶ The cathodic and anodic compartments were separated by a Nafion 117 membrane. The carbon felt (Alfa Aesar, >99.0%, 3 × 1.5 cm² with 3.2 mm thickness) used as the working electrode was first presoaked in water and connected to a 3 mm diameter graphite rod (Sigma-Aldrich, 99.99%). For each of the 1, 3, 5, and 10 wt. % Rh/C catalysts, 10 mg catalyst was added to the acetate buffer supporting electrolyte in the cathodic compartment. The carbon felt connected to a graphite rod was inserted into the cell, and the catalyst was loaded into the felt by stirring the catalyst/acetate buffer mixture for 30 min at 500 rpm until the catalyst infiltrated the carbon felt.¹³ The effectiveness factor of 0.97 for the carbon felt (3.2 mm thick) used in this work indicates that the felt was thin enough to avoid internal mass transfer limitations.¹¹ The reference electrode was an Ag/AgCl double junction electrode, and the counter electrode was a high-surface area Pt mesh. Acetate anions are reported to have negligible adsorption on Pt and Rh at the cathodic potential used in this work [i.e., −0.1 V vs RHE (reversible hydrogen electrode)].⁵⁷ Hence, we assume that ECH is not impacted by competitive adsorption between the reactants and acetate. Before ECH, the supporting electrolyte was sparged with N₂, which was also used as a blanket to prevent O₂ from entering the cell. A current of −40 mA was applied for 30 min to the working electrode to polarize the catalyst.

To measure the ECH rate, phenol solution already sparged with N₂ was added to the cell to make a final concentration of 20 mM, and the reaction was performed at a fixed potential using a Bio-Logic VSP-300. The solution resistance was measured using impedance spectroscopy based on the real part of the impedance at a frequency of 200 kHz. During the ECH measurements, only 85% of the solution resistance was compensated automatically because a higher fraction of compensation results in instability in the applied voltage due to the limitations of the potentiostat controller.⁵⁸ The remaining 15% of the solution resistance was manually corrected following the measurement and used to calculate the final iR-compensated applied potential of −0.1 V vs RHE that is reported. The pH of acetate buffer and phenol solution in the cathodic compartment was measured before and after the reaction for each catalyst metal weight loading, and no change was observed.

The TCH and ECH reactions were monitored by removing 0.5 ml aliquots every 3 min from the batch reactor. Three sequential liquid–liquid extractions were performed to transfer phenol and its associated hydrogenation products from the aqueous phase to an organic phase. In each extraction step, the aqueous phase was mixed with 1 ml of pure ethyl acetate to transfer phenol and its associated hydrogenation products from the aqueous phase to the organic phase, and then, the two liquids were separated. Three extractions were sufficient to transfer all organics from the aqueous-phase aliquot to the organic phase because no phenol or product was detected upon additional extraction. Remaining water was removed from the organic phase with anhydrous Na₂SO₄ (Sigma-Aldrich 99%). 20 µl of pure dimethoxybenzene was added as an internal standard to 1 ml of the dried organic phase. 1 µl of the resulting solution was injected to an Agilent Varian 450 gas chromatograph equipped with a flame ionization detector. Carbon balances were greater than 90% for all reported data. Rates were calculated using the moles of cyclohexanone and cyclohexanol detected to determine the moles of phenol converted. At least four time points were used to determine the rate. TOFs were calculated based on the rate of cyclohexanol and cyclohexanone formed per surface metal atom as described in Ref. 13. Additional details on TOF calculations are given in the supplementary material.

The experimental TCH rate data were fitted by minimizing the sum of the squared errors between the experimental TOF at different phenol concentrations and the TOF predicted from the reaction model in Eq. (2), in the same way as described in Ref. 10. To minimize the difference, the values of k, K_P, and K_HC_H were optimized, and these are the values reported. For simplicity in fitting the kinetic model, the values of k, K_P, and K_HC_H were assumed to be constant with phenol concentration (i.e., Langmuir adsorption). We have previously shown that equilibrium adsorption isotherms on Rh were best fit assuming an adsorbate interaction term of zero (i.e., Langmuir model),¹⁹ supporting our assumption of Langmuir adsorption when fitting the kinetic data here.

By fitting phenol hydrogenation rate data to an LH model, K_P of the active site can be extracted and compared with known K_P values on different facets to determine the identity of the active site. This approach has been used on Pt/C to determine that Pt(111) is the active site for phenol hydrogenation,¹⁰ but similar analysis has not yet been performed on Rh. Here, we measure phenol hydrogenation rates on Rh/C under the same TCH conditions as performed for Pt/C (i.e., 1 bar H₂)¹⁰ and extract K_P on Rh/C by fitting kinetic data to Eq. (2). We compare the extracted K_P on the active site with K_P values previously extracted from an aqueous-phase phenol adsorption isotherm on a Rh wire at the same temperature and pH as our kinetic measurements here.¹⁹ 

The TOF in Fig. 1 shows a phenol concentration dependence typical of a competitive LH mechanism, where at low concentrations of phenol, the reaction order in phenol is positive due to the phenol coverage being low, and at high concentrations of phenol, the reaction order is negative due to the phenol coverage being too high and poisoning the metal surface. By minimizing the error between the experimental TOF values and the predicted TOF values from Eq. (2), we determine k, K_P, and K_HC_H.

We compare the phenol adsorption equilibrium constant, K_P, on the active site from TCH rate measurements to the K_P values extracted independently from the adsorption isotherm fitting to understand the active site on Pt and Rh (Table I). K_P on the active site of Pt/C was measured previously to be 33, which corresponds to a Gibbs free energy of adsorption (∆G_P) of −8.7 kJ mol^–1, by fitting kinetic data using Eq. (2).¹⁰ K_P values of 38 and 120 000 have been measured previously for two distinct adsorption sites on polycrystalline Pt by fitting to a phenol adsorption isotherm constructed by measuring the fraction of hydrogen underpotential deposition (H_UPD) sites inhibited by phenol adsorption as a function of phenol concentration.³¹ Phenol adsorption sites with the lower K_P values adsorb phenol more weakly, and sites with the higher K_P values adsorb phenol more strongly. We refer to the adsorption equilibrium constants of the weak and strong adsorption sites as K_P,weak and K_P,strong, respectively. We refer to the corresponding Gibbs free energies as ∆G_P,weak and ∆G_P,strong. The weaker phenol adsorption site of Pt is attributed to the (111) terraces, and the stronger phenol adsorption site is attributed to (110)-like, (100)-like, and step sites based on the measured cyclic voltammograms (CVs).^10,31 As discussed in Ref. 10, the similarity between the K_P value on the active site of Pt/C from kinetic measurements (K_P = 33) and K_P,weak from the adsorption isotherm fitting (K_P,weak = 38) implies that the active facet for phenol hydrogenation on Pt/C is a Pt(111)-like terrace, which adsorbs phenol more weakly than stepped facets.

Similar to Pt, polycrystalline Rh has two distinct adsorption sites with different equilibrium constants for phenol—the values of K_P,weak and K_P,strong on the Rh wire were measured previously from adsorption isotherm fitting to be 17 and 41000, respectively (Table I).¹⁹ The comparable K_P value on the active site of Rh from Fig. 1 (K_P = 38) and K_P,weak from adsorption isotherm measurements (K_P,weak = 17) suggests that the facet that adsorbs phenol more weakly is active for phenol TCH on Rh/C. The three orders of magnitude difference between K_P,strong and K_P of the active site of Rh indicates that this stronger adsorption site does not contribute significantly to the catalytic turnover. Unlike on Pt, where the H_UPD peaks of (111) terraces and step sites are distinguishable, the CVs on Rh do not indicate whether (111) terraces or steps are weak or strong adsorption sites. Therefore, we cannot determine from CVs alone whether the active site of Rh is the (111) terrace or if stepped facets [e.g., Rh(221) steps] are responsible for TCH activity. In Sec. III C, we describe our first principles calculations and microkinetic modeling to determine the active Rh facet.

It is unclear whether Pt(111) is intrinsically more active than the stepped facets (i.e., larger k) or whether the activity of Pt(111) is from differences in adsorption energies between the weak and strong adsorption sites. To clarify the origin of the activity of the Pt(111) facet, we study the phenol hydrogenation reaction on the Pt(111) terrace and Pt(221) step at 0 V vs RHE (i.e., 1 bar H₂ at 300 K) using DFT modeling and mean-field microkinetic simulations. This is identical to the methodology for calculating the TCH rate at 1 bar H₂ from a computational perspective. Our calculations of phenol hydrogenation on Pt(100) in the gas phase show that Pt(100) is over four orders of magnitude less active than Pt(111) and, thus, is not considered further (Fig. S9). As discussed earlier, the surface reaction for TCH and ECH is hypothesized to be the same,^13–15 so we assume ECH and TCH share the same active site.

On both Pt(111) and Pt(221), the first hydrogenation step was modeled to occur on the ortho carbon followed by the hydrogenation of the meta carbon to form 1,3-cyclohexadienol, in accord with a prior study of phenol on Pt.²⁷ We predict that the third and fourth hydrogenation steps occur at the para carbon and meta carbon, respectively, to form 1-cyclohexenol. We model the tautomerization of 1-cyclohexenol to form cyclohexanone; however, tautomerization can occur at an earlier point along the reaction pathway. DFT-predicted geometries of this hydrogenation mechanism on Pt(111) are shown in Fig. 2(a), and the geometries on Pt(221) are shown in Fig. S10. The corresponding energetics of the surface reactions relative to aqueous-phase phenol and gas-phase hydrogen, with a surface hydrogen coverage of 4/9 ML, are given in Fig. 2(b). Because of the difficulties in accurately predicting aqueous-phase organic adsorption energies computationally, we instead use the experimental values for phenol adsorption from Table I as a calibration [i.e., to determine the relative energy of (ii)]. The dotted lines refer to the assumption that the modeled Pt site is “weak” adsorption (∆G_P,weak), and the solid lines refer to the assumption that the modeled Pt site is “strong” adsorption (∆G_P,strong). Although, as discussed above, we know that Pt(111) corresponds to the “weak” adsorption (∆G_P,weak), and thus, the dotted black line is the “correct” pathway for Pt(111), and the solid blue line in Fig. 2(b) is the “correct” pathway for Pt(221); we show both the cases here to understand the effect of this initial adsorption step on the overall rate. All subsequent energies [i.e., (iii)–(xvi)] are plotted based on the calculations relative to the adsorbed phenol, using the assumption that any solvation effects on phenol adsorption are the same as those for the adsorbed intermediates and transition states. The final energy of (xvii) is from the thermodynamic value of the difference in energy of phenol and cyclohexanone in the aqueous phase to maintain thermodynamic consistency for all plotted pathways. The surface-adsorbed H* consumed was replenished after each hydrogenation step to maintain the initial hydrogen coverage over the course of the reaction.

Hydrogen co-adsorption has been found to reduce hydrogenation barriers for various organic compounds on platinum group metals,^49,61,62 but has not yet been studied for phenol hydrogenation on Pt. On average, we find that co-adsorption of hydrogen decreases the activation energies of hydrogenation and tautomerization on Pt(111) by 20 kJ mol⁻¹ (Fig. S9). However, we find that the Pt(111) terrace is more active than the Pt step regardless of the presence or absence of co-adsorbed hydrogen (Fig. S9). Although co-adsorbed hydrogen does not change the activity ordering of Pt terraces and steps, this may not always be the case for other catalysts (Fig. S9 shows that hydrogen co-adsorption may change the predicted active site for Rh).

When looking at the reaction energy diagram for phenol hydrogenation on Pt(111) and Pt(221) [Fig. 2(b)], there are differences beyond the phenol adsorption energy that contribute to the kinetics. Even if the phenol adsorption energy were the same [i.e., comparing the solid blue and black lines in Fig. 2(b) to one another and the dotted lines to each other], the transition state barriers are often higher on Pt(221) than on Pt(111). These correspond to intrinsic kinetic differences on the two surfaces, rather than solely coverage effects.

To predict whether Pt(221) or Pt(111) is more active, we construct a mean-field microkinetic model to predict TOFs for phenol hydrogenation on Pt(111) and Pt(221) based on the reaction energetics in Fig. 2(b). The energies used for each adsorption and elementary reaction step in the microkinetic model of each facet are given in the supplementary material (Table SIII). The calculated TOFs between 290 and 310 K from the microkinetic simulations of phenol hydrogenation on Pt(111) and Pt(221) in Fig. 3 show that Pt(111) is the active site. The reaction on both the facets is modeled using both the weak and strong phenol adsorption free energies to clarify how differences in phenol adsorption on the two facets affect phenol hydrogenation kinetics. Pt(111) is more active than Pt(221) regardless of whether the reaction is modeled with ΔG_P,weak or ΔG_P,strong. Considering that ΔG_P,weak is attributed to Pt(111)-like terraces and ΔG_P,strong is attributed to stepped facets,³¹ our computational results predict that Pt(111) is over thirteen orders of magnitude more active than Pt(221) at 300 K and 0 V vs RHE. Even when the reaction is modeled with ΔG_P,weak on both the facets, Pt(111) is still over thirteen orders of magnitude more active than Pt(221), highlighting that the activity differences between Pt(111) terraces and steps are largely due to intrinsic kinetic differences and not only due to differences in phenol adsorption on Pt(111) and step sites. Our predicted TOFs based on the microkinetic modeling support previous experimental results that Pt(111) is the active facet for phenol hydrogenation.¹⁰ The joint factors of adsorption energy and intrinsic kinetic activity contributing to the activity of a catalyst site help explain why Pd/C is less active than Pt/C for phenol hydrogenation despite their similar adsorption energies.¹⁰ 

The predicted apparent activation energies of the reactions are extracted from the slopes of the Arrhenius plots in Fig. 3. The predicted activation energy of 102 kJ mol⁻¹ on Pt(111) (ΔG_P,weak = −9 kJ mol⁻¹) and 168 kJ mol⁻¹ on Pt(221) (ΔG_P,strong = −29 kJ mol⁻¹) are much higher than the experimental apparent activation energy of 29 kJ mol⁻¹ on Pt/C at 300 K.¹⁵ Besides the commonly attributed error in predicted energetics due to using an approximate exchange-correlation functional, the high computed apparent activation energy relative to the experimental value may arise from the exclusion of explicit solvation effects in our atomistic model.

Although the microkinetic modeling simulations reveal that the higher activity of Pt(111) compared to stepped facets is due to Pt(111) having both faster intrinsic kinetics and weaker phenol adsorption, there are some discrepancies between our predicted RDS and that in our LH model [Eq. (2)]. Microkinetic simulations predict that the fourth hydrogenation step on Pt(111) has the highest degree of rate control (Fig. S11).⁶³ This prediction supports that a surface hydrogenation step, not adsorption or desorption, is the RDS for phenol hydrogenation, but is different than our assumption in Eqs. (1) and (2) that the first hydrogenation step is the RDS. Explicit treatment of solvent effects along the reaction coordinate would lead to a more accurate prediction of the RDS. Recent experimental and computational work has suggested that the second hydrogen addition to phenol on Pt is the RDS at temperatures of 433–473 K, because H/D exchange of phenol was found to be fast compared to the hydrogenation rate, implying a quasi-equilibrated first hydrogenation step.⁶⁴ Because of the uncertainty in which hydrogenation step [1–4 in Fig. 2(a)] is the RDS, a more general interpretation of Eq. (2) is to describe a surface reaction between hydrogen and either phenol or a partially hydrogenated phenol intermediate, as postulated in Ref. 10.

Comparing K_P from our experimental TCH kinetic measurements with the K_P,weak and K_P,strong values extracted from adsorption isotherms indicates that the active facet for phenol hydrogenation on Rh is the facet that adsorbs phenol more weakly (Table I). However, experimental CV adsorption isotherms do not reveal whether K_P,weak (and, hence, the active facet) may be attributed to the Rh(111) terraces or stepped facets.¹⁹ To identify the active facet of Rh and explain the origin of the activity, we computationally model the hydrogenation of phenol to cyclohexanone on Rh(111) and Rh(221) with co-adsorbed hydrogen using the same mechanism considered for Pt. Similar to the observation on Pt, hydrogen co-adsorption reduces activation energies on Rh(111) by 8 kJ mol⁻¹ on average (Fig. S9). The geometries on Rh(111) are shown in Fig. 4(a), and the geometries on Rh(221) are shown in Fig. S12. On Rh, it has been proposed that phenol may be hydrogenated through a mechanism that involves the conversion of phenol (C₆H₅OH) to phenoxy (C₆H₅O) followed by sequential hydrogenation to form cyclohexanone (Scheme S1).⁶⁵ We model this mechanism (Fig. S13) and predict it to be slow compared to the mechanism in Fig. 4(a), which involves phenol hydrogenation to 1-cyclohexenol followed by tautomerization to form cyclohexanone. We use the same methodology as described for Fig. 2(b) to plot the reaction energy diagram on Rh in Fig. 4(b), again using the experimentally measured adsorption energies of phenol. Unlike for phenol, here we do not know already whether Rh(111) or Rh(221) corresponds to “weak” or “strong” adsorption. Like on Pt, the data in Fig. 4(b) show that there are differences in the reaction energy profile between Rh(111) and Rh(221) apart from the phenol adsorption energy, despite the same mechanism being investigated on both the surfaces.

We construct a mean-field microkinetic model to predict TOFs for phenol hydrogenation on Rh(111) and Rh(221) based on the energetics in Fig. 4(b) and predict that Rh(111) is the active site for phenol hydrogenation. The energies used for each adsorption and elementary reaction step are given in the supplementary material (Table SIV). By analyzing the predicted TOFs as a theoretical Arrhenius plot (Fig. 5), we show that Rh(111) is more active than Rh(221) for phenol hydrogenation at temperatures between 290 and 310 K. The Rh(111) is more active than Rh(221) regardless of whether the reaction on Rh(111) is modeled using ∆G_P,weak or ∆G_P,strong. Our experimental data (Table I) indicate that the active site for phenol hydrogenation on Rh is the site that adsorbs phenol more weakly. When modeling phenol ECH using the weak phenol adsorption energy on both the facets, we predict that Rh(111) is over six orders of magnitude more active than Rh(221). These results demonstrate that Rh(111) must be the weak adsorption site of Rh due to the relative inactivity of Rh(221). Although the microkinetic model predicts Rh(111) with ∆G_P,strong to have the highest activity due to the more balanced coverages of hydrogen and phenol, the stronger adsorption site being the active facet is inconsistent with experimental measurements. Modeling the effect of solvation on the adsorption energy of hydrogen would improve the predicted activity ordering between ∆G_P,weak and ∆G_P,strong on Rh(111).

Based on Fig. 5, the apparent activation energy for ECH of phenol on Rh(111) (ΔG_P,weak = −7 kJ mol⁻¹) is 142 kJ mol⁻¹ and for Rh(221) (ΔG_P,strong = −26 kJ mol⁻¹) is 181 kJ mol⁻¹. Similar to Pt, the apparent activation energies from the DFT-based mean-field microkinetic models are significantly larger than the experimentally measured value of 23 kJ mol⁻¹ in the absence of mass transport limitations on Rh/C.¹⁵ Similar to Pt, the fourth hydrogenation step is predicted to have the highest degree of rate control on Rh(111) and Rh(221) (Fig. S11).

To corroborate our predictions that Pt(111) and Rh(111) are the active sites for phenol TCH and ECH, we measured the particle size dependence of the ECH TOF and Faradaic efficiency on Rh/C particles with diameters between 2 and 10 nm [Fig. 6(a)]. The TOFs for the ECH of phenol on Pt/C over a similar particle size range from Ref. 11 are reproduced in Fig. 6(a). On both Pt/C and Rh/C, larger particles are predicted to have a higher fraction of (111) sites from the cuboctahedron model,^11,66,67 while smaller particles have a higher fraction of step-like features.^68,69 These larger particles have higher TOFs and higher Faradaic efficiencies than smaller particles. At small enough particle sizes (∼2 nm), we observe practically no catalytic activity, indicating that the step sites are inactive for the electrocatalytic hydrogenation of phenol, consistent with our modeling predictions. Although the fraction of (100) terraces will also increase with particle size according to the cuboctahedron model, the (100) sites of Pt and Rh adsorb phenol stronger than (111) terraces.^19,31 Because our combined kinetic and adsorption measurements show that the active site should adsorb phenol weakly, the (100) sites are not the active sites. Additionally, we predict (100) sites to have lower gas-phase hydrogenation activity compared to (111) sites (Fig. S9). Therefore, we attribute the increase in activity and Faradaic efficiency with particle size to the increasing fraction of the active (111) terraces that dominate the catalytic turnover.

Decreasing particle size to reduce the mass of expensive metals required (i.e., Pt and Rh) is desired only down to a certain particle size, below which the decrease in fraction of active sites counteracts the benefit from a higher fractional exposure. The catalyst utilization, the phenol hydrogenation rate per gram of the catalyst, is shown in Fig. 6(b). The smallest particles of Pt (∼3 nm) have lower utilizations compared to larger particles; however, the catalyst utilization plateaus at further increasing particle sizes. We expect there to be a maximum in catalyst utilization for Pt within the 3–10 nm particle diameter range and hypothesize that increasing the particle diameter beyond 10 nm, while possibly increasing the TOF, would decrease the catalyst utilization, and, thus, be undesired for practical applications. The utilization of Rh increases monotonically in the 3–10 nm particle diameter range, which is unexpected as we would assume a plateau or a maximum to occur, similar to what is seen for Pt, if a cuboctahedron model of the particle size-facet distribution is correct.⁷⁰ This lack of a matching catalyst utilization maximum on Pt and Rh in Fig. 6(b) implies that the distribution of terrace sites with particle size may differ between Pt and Rh and that higher utilizations of Rh may be achieved on nanoparticles sizes >10 nm. However, we do hypothesize that a maximum in catalyst utilization will occur on Rh at large enough particle sizes.

Synthesizing shape-specific catalysts to preferentially expose (111) sites is a potential route to improve the TOF of the structure-sensitive phenol hydrogenation reaction and improve catalyst utilization.^71–73 The insights here for phenol hydrogenation extend to structure-sensitive hydrogenation of other oxygenated aromatics relevant to bio-oils, such as benzaldehyde. The TOF of benzaldehyde ECH on Pt/C increases with particle size,¹¹ and the (111) terraces are predicted to adsorb benzaldehyde weaker than step sites.¹⁹ Thus, the particle size dependence for benzaldehyde is explained by the increase in the fraction of (111) terraces with particle size, as is the case with phenol.

Both Pt and Rh are structure-sensitive for TCH and ECH of phenol with the active site being (111) terraces. Although (111) terraces are the weaker phenol adsorption sites, which enables more equal coverages of phenol and hydrogen reactants, those surfaces are also intrinsically more active, i.e., they have lower hydrogenation barriers. This finding shows that although the adsorption energy of phenol is important for predicting kinetics, it is not the sole predictor, potentially explaining why other metals that adsorb phenol similarly to Pt and Rh (e.g., Pd) have different kinetics. It is unknown whether weaker adsorption to the active site is a general phenomenon for the hydrogenation of oxygenated aromatics, but other catalytic reactions, such as benzaldehyde hydrogenation, also show a similar structure sensitivity. Further work is needed to identify if (111) terraces, which adsorb a wide range of organics more weakly than step sites, are also more intrinsically active for a range of bio-oil model compounds.

See the supplementary material for details on computational and experimental methods, turnover frequency normalization, mass transfer considerations, and DFT-predicted geometries and energetics for input into the microkinetic model.

This work was supported by the National Science Foundation (Grant No. 1919444). Computing resources were generously provided by the Department of Energy, National Energy Research Scientific Computing Center (Contract No. DE-AC02-05CH11231 using NERSC award No. BES-ERCAP0021436). Kinetic measurements for the particle size dependence of Rh/C were taken as part of the Chemical Transformation Initiative at Pacific Northwest National Laboratory (PNNL), conducted under the Laboratory Directed Research and Development Program at PNNL, a multiprogram national laboratory operated by Battelle for the U.S. Department of Energy. O.Y.G would like to acknowledge support by the U.S. Department of Energy (DOE), Office of SContract No. DE-AC02-05CH11231 using NERSC award No. BES-ERCAP0021436cience, O?ce of Basic Energy Sciences, Division of Chemical Sciences, Geosciences, and Biosciences (FWP 47319).

The authors declare no conflict of interest.

The data that support the findings of this study are available within this article and its supplementary material. Microkinetic modeling input scripts are made available on GitHub at https://github.com/isaiah-barth/Rh-Facet-Dependence. Raw data files and computed geometries have been uploaded to NOMAD under DOI: https://dx.doi.org/10.17172/NOMAD/2022.01.14-1.