Controlling reaction pathways of selective C–O bond cleavage of glycerol

The selective hydrodeoxygenation (HDO) reaction is desirable to convert glycerol into various value-added products by breaking different numbers of C–O bonds while maintaining C–C bonds. Here we combine experimental and density functional theory (DFT) results to reveal that the Cu modifier can significantly reduce the oxophilicity of the molybdenum carbide (Mo2C) surface and change the product distribution. The Mo2C surface is active for breaking all C–O bonds to produce propylene. As the Cu coverage increases to 0.5 monolayer (ML), the Cu/Mo2C surface shows activity towards breaking two C–O bonds and forming ally-alcohol and propanal. As the Cu coverage further increases, the Cu/Mo2C surface cleaves one C–O bond to form acetol. DFT calculations reveal that the Mo2C surface, Cu-Mo interface, and Cu surface are distinct sites for the production of propylene, ally-alcohol, and acetol, respectively. This study explores the feasibility of tuning the glycerol HDO selectivity by modifying the surface oxophilicity.

In the Supplementary Figure 3, the sharp peak at 260 K was from the cracking pattern of glycerol. A sharp peak at 365 K was observed on the Cu-terminated surface, and a broad peak at 569 K was observed on the Mo2C surface.

Correlating Computational Predictions to TPD and HREELS Experimental Results
Gibbs free energy profiles, calculated at a temperature of 350 K and under UHV conditions (Pgas = 10 -9 atm), for the minimum energy pathways of glycerol decomposition on the three active sites investigated are provided in the Supplementary Fig. 13. For the interface sites, only the allyl alcohol formation pathway is shown to improve clarity. On the Mo sites, all O-H and C-O bond cleavages are consistently exergonic, and only the final protonation step is endergonic. This is due to strong adsorption of oxygen atoms on the Mo sites, and these oxygen atoms can only be removed at higher temperatures and by adding excess H2 as shown in the previous work. 1 Although desorption of propylene is slightly endergonic by 0.06 eV at 350 K, it becomes exergonic above 400 K which agrees well with the TPD experiments that display the largest peak for propylene desorption at 411 K. In our model, propylene is adsorbed on the sites neighboring the adsorbed O and H atoms (see Supplementary Fig. 7, structure 8) for which the adsorption energy is calculated as -1.50 eV (Supplementary Table 3). Our calculations predict that the adsorption of propylene can be further stabilized by 0.6-0.8 eV on sites further away from adsorbed O and H atoms which is consistent with a stronger adsorption energy of -2.95 eV (calculated with RPBE functional) reported for ethylene on Mo2C (0001) surface 2 . However, such empty Mo sites might not be available in the present case due to the presence of strongly adsorbed O atoms that are dissociation products from glycerol. In addition, the Cu/Mo2C/Mo(110) surfaces used in the current TPD experiments are pre-dosed with hydrogen which could further affect the availability of empty Mo sites and the stability of the reactant and product on the surface. Free energy calculations under UHV conditions (T= 350 K; PH2 = 10 -9 atm) suggest that the average adsorption free energy of H calculated with reference to the energy of H2 is -0.49 eV when 8 hydrogen atoms are adsorbed on the Mo sites of the catalyst model used here (which corresponds to a hydrogen coverage on the Mo sites of 0.5 ML). The adsorption of glycerol was further tested in the presence of 8 hydrogen atoms, and we found that the adsorption is destabilized by 0.6 eV in the presence of adsorbed H atoms on neighboring sites ( Supplementary Fig. 14). Thus, the free energy of adsorption in the presence of H atoms becomes endergonic ( 350 = 0.15 eV) compared to an exergonic adsorption of glycerol predicted in the absence of H atoms (Supplementary Fig. 13). This again agrees with the TPD experiments that display a glycerol desorption peak on the hydrogen pre-dosed surfaces at a low temperature of 280 K. However, the presence of H atoms is not expected to change the trend observed in the competing elementary reactions. At an H atom coverage lower than 1 on the Mo sites, the H atoms can easily move between different Mo sites with a small barrier of 0.2-0.3 eV, making the Mo sites available for glycerol dissociation reactions.
All the elementary reactions involved in the allyl alcohol formation pathway at the interface sites are exergonic since these reactions occur on the Mo sites. The role of the Cu atoms is primarily to block the active site for further C-O dissociation. Also, the Cu atoms are active for the isomerization of allyl alcohol to propanal as shown in Fig. 4 of the paper. The free energies of these reactions are slightly above the free energies of the propylene formation pathway, mainly due to the absence of the third Mo-O bond at the interface. Furthermore, the stability of adsorbed oxygen atoms on similar 3-fold hollow Mo sites was examined at various distances to the Cu/Mo2C interface. We found that the adsorption of oxygen atoms closer to the interface sites are 0.6-1 eV less stable than those adsorbed farther away from the interface (Supplementary Fig. 15). This suggests that it is easier to remove the adsorbed oxygens at the interface which again agrees with the experimental observations from HREELS spectra that a 0.3 ML Cu/Mo2C/Mo(110) has a weaker interaction with O than Mo2C/Mo(110) (Fig. 2 of the paper).
The free energy profile for the formation of acetol on the Cu sites is significantly above the energy profiles for propylene and allyl alcohol formation, and the energy of the most stable state in this pathway (CH2OH-CO-CH2*, 5) is only 0.66 eV lower than the initial state ( Supplementary  Fig. 13). Calculations revealed that the formation of H2O from the final state (O*+2H*) is exergonic by 0.05 eV at 350 K temperature and thus, the adsorbed oxygen can be easily removed. This is consistent with the results from HREELS spectra that atomic oxygens were not observed on Cu (Fig. 2 of the paper) and also from TPD spectra ( Supplementary Fig. 3) that H2O is easily removed from Cu surfaces. The highest energy state in the free energy profile is found to be the transition state corresponding to the first O-H dissociation process ( = 1.00 eV) which appears to be the rate-limiting process for acetol formation.
In order to further examine the removal of adsorbed oxygen atoms from the three different active sites, activation barriers were calculated for the H2O formation process, both in the presence and absence of excess H2 (see Supplementary Table 1). On the Mo sites, hydrogenation of either O* or OH* with adsorbed H* was found to be highly endothermic with barriers of about 2 eV (reactions 1 & 2). However, direct dissociation of gas phase H2 onto O* was found to be favorable with an activation barrier of only 0.38 eV. Similarly, the formation of H2O via the disproportionation reaction of two neighboring OH* was also found to possess a barrier of only 0.75 eV. These results are consistent with the earlier report by Ren et al. that the adsorbed O on the Mo sites can be removed by excess H2. 3 On the Cu/Mo2C interface sites, we examined the possibility of H2 dissociation on the Cu sites and two subsequent H atom spillovers from the Cu to the interface oxygen adsorbed on the Mo sites to form H2O. Calculations revealed that the dissociation of gas phase H2 on Cu is an exothermic process and the spillover of H atoms from Cu to interface O requires overcoming barriers of only about 1 eV, suggesting that these processes are feasible. The direct dissociation of gas phase H2 onto O* (reaction 9) and OH disproportionation (reaction 10) were also examined at the interface site for which the barriers were calculated as 0.39 and 0.61 eV, respectively. H2O formation is thus favorable both on the Mo sites and at the interface sites via OH disproportionation provided that empty Mo sites are available at neighboring sites that can promote the formation of OH groups by direct dissociation of gas phase H2 onto adsorbed O.
On the Cu sites, the activation barriers calculated for H2O formation in the absence of excess H2 was found to be >1 eV (reactions 12 & 13 in Supplementary Table 1), however, these barriers are smaller than the corresponding barriers on the Mo sites. The calculated barriers for the stepwise hydrogenation of O* (1.05 eV) and OH* (1.36 eV) are similar to those reported for different terminations of Cu surfaces 4,5 . Here again, direct dissociation of excess gas phase H2 onto adsorbed O* (reaction 14) and the OH disproportionation to form H2O (reaction 15) were found to be more favorable than the stepwise hydrogenation processes. The endothermicity of H2O desorption from the three sites decreases in the order, Mo site (0.94 eV) > interface site (0.82 eV) > Cu site (0.35 eV), which is consistent with the decrease in the desorption temperature observed in the TPD spectra for H2O desorption when going from pure Mo2C to 1 ML Cu/Mo2C (Supplementary Fig.  3).
These results suggest that H2O formation from adsorbed O on the three surfaces are possible in the presence of excess H2 with activation barriers of less than 1 eV. Since the partial pressure of H2O at UHV conditions is as low as 10 -12 atm, H2O can easily desorb from these sites. However, under experimental reactor conditions, i.e., ambient pressure conditions, all the Mo sites could be occupied by oxygen atoms. In fact, constrained ab initio thermodynamic analysis carried out in the presence of an equimolar H2/H2O gas phase for our current catalyst model suggested that all the exposed Mo sites are occupied by oxygen atoms at a temperature of 500 K. For this model (1 st structure in Supplementary Fig. 16), the oxygen vacancy formation free energy was found to range from 0.4 to 1.2 eV (O* + H2 → Ovac + H2O, T = 500 K; Pgas=1 atm) when going from the Cu/Mo2C interface site to the sites away from Cu. Thus, we examined the possibility of H spillover from Cu to the interface oxygens to form H2O. The calculated barriers for the 1 st and 2 nd H spillover process were found to be similar but slightly more favorable on the fully oxygen covered model compared to those calculated for the model with lower oxygen coverage (Supplementary Table 1). The free energy profile calculated at a temperature of 500 K suggested that the removal of interface oxygen is feasible when the PH2/PH2O is slightly above 10 2 which can easily be achieved experimentally. The effective free energy barrier of 1.36 eV corresponds to an approximate oxygen removal rate of 10 -1 s -1 , indicating that the interface oxygens can be removed from an "oxycarbide" surface and the Cu/Mo2C catalyst should remain active at more practical reactor conditions.

Surface regeneration under UHV condition
To study the role of Cu modifier in catalyst regeneration, sequential TPD experiments were performed on three surfaces, Mo2C, 0.5 ML Cu/Mo2C and 1 ML Cu/Mo2C, between 200 K and 500 K. Supplementary Fig. 5(a) shows the desorption of propylene on the Mo2C surface. In the second TPD experiment, the propylene desorption peak shifts to 470 K and the intensity reduces, suggesting a lower C-O bond cleavage activity of the Mo2C surface after the first TPD experiment. In contrast, on the 1 ML Cu/Mo2C surface ( Supplementary Fig. 5(b)), the Cu site maintains activity in three sequential TPD experiments. In Supplementary Fig. 5(c), the 0.5 ML Cu/Mo2C also loses activity toward allyl alcohol formation after the first TPD experiment. In the second TPD experiment, a small peak of propanal is observed at 402 K ( Supplementary Fig. 5(d)). The TPD results under the UHV condition suggest that the Mo2C surface and Cu-Mo2C interface are less stable than the Cu surface, which is due to the strong oxophilicity of the Mo site.
However, these experiments were performed under UHV conditions, in which gas-phase H2 is not involved. In a real catalytic HDO process with a high H2 pressure, gas-phase H2 should help remove surface oxygen and maintain a higher hydrogen coverage. In a previous study, Ren et al. studied the HDO reaction of propanal on powder Mo2C catalyst using a flow reactor. 3 The catalyst quickly deactivated in the absence of H2. With co-feed H2, a steady-state propanal conversion of approximately 50% was achieved at 573 K, suggesting that oxygen was removed by H2. This was also confirmed by the production of water at this temperature. These results are also consistent with DFT results in the current study of lower activation barriers for oxygen removal in the presence of gas-phase H2 (Supplementary Table 1)."