Uncovering the reaction mechanism behind CoO as active phase for CO2 hydrogenation

Transforming carbon dioxide into valuable chemicals and fuels, is a promising tool for environmental and industrial purposes. Here, we present catalysts comprising of cobalt (oxide) nanoparticles stabilized on various support oxides for hydrocarbon production from carbon dioxide. We demonstrate that the activity and selectivity can be tuned by selection of the support oxide and cobalt oxidation state. Modulated excitation (ME) diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) reveals that cobalt oxide catalysts follows the hydrogen-assisted pathway, whereas metallic cobalt catalysts mainly follows the direct dissociation pathway. Contrary to the commonly considered metallic active phase of cobalt-based catalysts, cobalt oxide on titania support is the most active catalyst in this study and produces 11% C2+ hydrocarbons. The C2+ selectivity increases to 39% (yielding 104 mmol h−1 gcat−1 C2+ hydrocarbons) upon co-feeding CO and CO2 at a ratio of 1:2 at 250 °C and 20 bar, thus outperforming the majority of typical cobalt-based catalysts.


Supplementary Methods 1.Catalyst preparation
The BET surface areas and pore volumes of the support materials were determined from N2 adsorptiondesorption isotherms measured at -196°C using a Micromeritics TriStar II PLUS instrument. Prior to the physisorption measurements, the samples were dried at 300°C overnight in N2 flow. The resulting physical properties of the support materials and the exact amounts of chemicals used during the incipient wetness impregnation (IWI) preparation procedure can be found in Supplementary Table 1. All support oxides were purchased from companies (Supplementary Table 1), apart from CeO2, which was synthesized in-house by homogeneous deposition precipitation (HDP) using urea as a precipitation agent. In 1.7 L of deionized water, 50 g cerium (III) nitrate hexahydrate (99.99% trace metals basis, Sigma-Aldrich) and 27 g of urea were dissolved. The mixture was added to a double-walled glass vessel and heated to 90°C for ~20 h while stirring at 600 rpm. The sample was washed three times with 500 mL deionized water, dried at 60°C, and then at 120°C. Calcination was performed at 500°C for 2 h (5°C min -1 ramp).

Operando Raman micro-spectroscopy
The operando Raman micro-spectroscopy experiments were carried out with a Horiba Xplora Raman microscope equipped with a 532 nm laser (output of 0.53 MW), a 1200 nm grating and a 50x objective. The range was set to 100-2800 cm -1 , the acquisition time was 10 s and 5 accumulations were performed. The catalyst sample was placed in a THMS600 cell from Linkam Scientific Instruments. The Linkam cell was connected to a Bronkhorst mass flow controller (MFC) for gas input. The output of the cell was connected to an OmniStar mass spectrometer from Pfeiffer Vacuum to analyze the gaseous products. In a typical experiment about 10-15 mg of catalyst sample was placed in the Linkam cell. The temperature was ramped to 250°C (suffix: -ox) or 450°C (suffix: -red) with 10°C min -1 under a gas mixture of Ar:H2 40:20 mL min -1 and held there for 1 h. Then, the cell was either held at or cooled down to 250°C and the gas flow was switched to Ar:H2:CO2 in a ratio of 40:10:2.5 mL min -1 . The temperature was held on this for one hour. During this hour, the gaseous products were continuously analyzed by the mass spectrometer. Raman spectra were collected before and during the reaction.

H2-temperature programmed reduction (TPR)
H2-TPR measurements were performed using a Micromeritics AutoChem II 2920. Samples were placed on quartz wool into a U-tube quartz reactor. The gas mixture consisted of 5% H2 in Ar with a total gas flow of 40 ml min -1 . H2-TPR was carried out by heating with 5°C min -1 up to 1000°C for all cobalt-based catalysts and held for 30 min at this temperature. A constant initial sample weight of 0.05 g was used and H2 consumption was continuously monitored by a thermal conductivity detector. To assess reducibility, a centroid was calculated for each data set. This was done by calculating the integral (MATLAB) of the H2-TPR data for each sample and then taking ½ of that.

CO2-temperature programmed desorption (TPD)
CO2 Temperature Programmed Desorption (TPD) measurements were carried out on a Micrometrics ASAP2920 instrument equipped with a thermal conductivity detector. In a typical experiment, 100 mg of sample (SiO2, Al2O3, TiO2, or CeO2; see Supplementary Table 1 for details) was loaded in a quartz tube and dried in situ by ramping with 5°C min -1 to 300°C in a He flow and remained at that temperature for 60 min. Subsequently, the sample was cooled down to 60°C; at this temperature pulses of 10% CO2 in He of 25 cm 3 /min were applied. After saturation with CO2, the sample was outgassed for 30 min at 60°C to ensure removal of physisorbed CO2. Finally, the sample was heated to 700°C with a ramp of 5°C min -1 to measure CO2 desorption. To assess support basicity, the integral of the CO2-TPD data was calculated for each data set and divided by the BET surface area of the respective sample.

Density functional theory (DFT) calculations
Quantum-chemical calculations in this work were performed using a planewave density function theory (DFT) approach with the projector-augmented wave (PAW) method 1,2 in conjunction with a Perdew-Becke-Ernzerhof (PBE) exchange-correlation functional 3  The Co(111) , Co(110) and CoO(100) surfaces were modeled using a (3x3) surface, with 6, 4 and 6 metal layers, respectively. A Monck-horst-Pack mesh of k-points of (5x5x1) for Co(110) and (3x3x1) for Co(111) and CoO(100) were used 8 . A vacuum layer of 15 Å perpendicular to the surface was employed to avoid the spurious interaction of neighboring supercells. To avoid the build-up of a large dipole moment between neighboring supercells, the adsorbates were placed on both sides of the surface slabs retaining a point of inversion. All atoms were allowed to relax. Partial occupancies were determined using a first-order Methfessel-Paxton scheme with a smearing width of 0.2 eV and 0.03 eV for the Co and CoO slabs, respectively 9 . Electronic convergence was set to 10 -5 eV, and geometries were converged to 10 -4 eV using a conjugant gradient algorithm for the Co systems and a quasi-Newton algorithm for CoO system. For the gas-phase calculation of CO2, the molecule was placed in a 10x10x10 Å unit cell. Gaussian smearing with a width of 2x10 -5 eV was used for electron smearing and only the gammapoint was used to sample the Brillouin zone.
The adsorption energy, ads , is defined as follows: where slab+adsorbate represents the total energy of the optimized adsorbate on the surface, slab is the energy of the nickel slab and adsorbate is the energy of the adsorbate in the gas phase. The total density of states was calculated in the energy range of -30 eV to 15 eV over 4500 grid points. Then, the partial charge density corresponding to the energy interval of each molecular orbital was calculated. The contour plots of the electron density from these intervals were plotted on a cutting plane running parallel through CO2 adsorbed on the top of the slab.

Operando modulation excitation (ME) diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS)
A picture of the operando modulated excitation (ME) diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) setup and a schematic drawing of the cell used during the experiments can be found in Supplementary Figure 1. The cell was designed to represent a plug flow reactor and has a reduced dead volume, which allows for fast exchange of the gas atmosphere and ensures suitability for transient experiments 10 . The gas flows were controlled with mass flow controllers (MFCs), which were calibrated in advance. During the modulation experiments, the gas inlet was controlled with fast switching valves (solenoid valves) that were operated using an automated script. The experiments were carried out as graphically represented in in Supplementary Figure 2 by admitting to the catalyst sample equally long (60 s) pulses of CO2/H2 and H2 (modulation period, H2/CO2=3) at the selected temperature 250°C while recording 120 spectra (1 spectrum/s) at 80 kHz scanner velocity and at 4 cm -1 resolution. One experiment consisted of 10 modulation periods. Prior to the modulation experiments, the sample (typically 35-40 mg with a grain size of 250-63 μm) was heated to 250°C at 10°C min -1 in H2/N2 (H2/N2=1; total flow rate 40 mL min -1 ) to create the CoO oxidation state. Then, at 250°C 10 modulation periods were performed by alternating flows of CO2:H2 (H2/CO2 =3; 6 mL min -1 H2, 2 mL min -1 CO2, 12 mL min -1 N2 total flows) and H2 (6 mL min -1 H2 and 12 mL min -1 N2 total flows). During each period of 120 s, 120 spectra (1000-4000 cm -1 with 4 cm -1 spectral resolution) were recorded. The 10 modulation periods of 120 s each resulted in a 20 min experiment. After the modulation experiment, the sample was heated to 450°C at 10°C min -1 in H2/N2 (H2/N2=1; total flow rate 40 mL min -1 ) and was held there for 1 h to reduce the CoO nanoparticles to metallic Co. Then, the sample was cooled to 250°C at 10°C min -1 and the modulation experiment described above was repeated. During the experiments, the gaseous products were analyzed using an Omnistar quadrupole mass spectrometer from Pfeiffer Vacuum using the following m/z values: 2 (H2), 4 (helium), 14 The external stimulus is a sinusoidal or square-wave function with frequency ω. Once the full set of time-resolved data (time domain) is measured along a given number of modulation periods, each spectrum is averaged along the modulation periods to obtain a set of average time-resolved spectra (spectrum i is the average of spectrum i from each modulation period) 12 . Demodulation by PSD transforms the set of averaged time-resolved data into a set of phase-resolved data (phase domain). In this set of data, the active species appear with a phase delay ϕ instead of appearing with a time delay Δt. The phase delay contains additional kinetic information about the system that is often hidden in the kinetic information delivered by the time-resolved data (for example because of strong band overlap). The phase-resolved amplitude spectra in Figure 2b in the main text were obtained by taking the absolute maximum at every single wavenumber in the phase-resolved spectra. The result is that only positive signals are displayed for the sake of clarity in the discussion.

Data analysis
The spectra were analyzed using LabSpec6 (Raman), MATLAB (ME DRIFTS), and Origin 9.

Kinetic parameters for Co-TiO2-ox and Co/TiO2-red
The kinetic parameters reaction order (n, m) and apparent activation energy (Ea) were determined by varying the reactant (CO2, H2) concentration and the temperature, respectively. For these experiments the high-pressure catalytic testing setup, as described in the Methods section of the main text, was used. The catalyst was either treated at 250°C in H2 for 1 h to obtain the CoO phase (Co/TiO2-ox) or at 450°C in H2 for 1 h to obtain the metallic Co phase (Co/TiO2-red). Then, at T=250°C and P=20 bar, the H2/CO2 ratio was varied at 2, 3, and 4 while the H2 concentration was kept constant and the H2/CO2 ratio was varied at 2 and 3 while the CO2 concentration was kept constant. This was repeated three times in order to obtain standard deviations. The measured intrinsic reaction rates were used to calculate the reaction order (n) in CO2 and (m) in H2 according to the rate law (Supplementary Equation (1)) 13 , where k is the rate constant, n the reaction order in CO2, and m the reaction order in H2. Besides, the temperature was varied between 200 and 280°C at P=20 bar and the measured intrinsic rates were used to calculate the apparent activation energy (Ea) according to the Arrhenius equation (Supplementary Equation (2)) 13 , where k is the rate constant, A is the pre-exponential factor, R the universal gas constant, and T the absolute temperature in Kelvin.

Thermodynamic calculations
Thermodynamic calculations were performed using the software HSC 9.6.1, in the Gem equilibrium composition module, based on Gibbs free energy minimization 14 .

Operando Raman micro-spectroscopy
Operando Raman micro-spectroscopy was used to study the oxidation state of cobalt (see Supplementary Figure 5). Co3O4 has a peak at 690 cm -1 , whereas a CoO peak is observed at 675 cm -1 , and the vibrations of metallic cobalt are not observable, as they are Raman-inactive 15 . The catalysts treated at 250°C in H2 (suffix:ox) contained primarily CoO at 250°C during the CO2 hydrogenation reaction (Supplementary Figure 6), as indicated by the peak positions around 675 cm -1 . The catalysts treated at 450°C in H2 (suffix: -red) contained metallic cobalt during CO2 hydrogenation at 250°C, as indicated by the absence of a peak 675-690 cm -1 .
For the reducible supports, TiO2 and CeO2, (partial) reduction of the support was observed in the Raman spectra. For example, CeO2 displayed a Ce 4+ -O-Ce 4+ wagging vibration at 458 cm -1 prior to the reaction. The CeO2 lattice expands slightly upon reduction, as Ce 3+ is larger than Ce 4+ , causing the wagging vibrational frequency to shift down 16 . We observed indeed a shift from 458 cm -1 in the fresh Co/CeO2 catalyst to e.g. 452 cm -1 during CO2 hydrogenation at 250°C for Co/CeO2-ox. A similar effect was observed for TiO2, as e.g. the peak at 519 cm -1 (Eg) 17 prior to reaction shifted to 513 cm -1 during CO2 hydrogenation at 250°C for Co/TiO2-ox.   Figure 5) carbon deposition was observed during the CO2 hydrogenation reaction, as indicated by the appearance of the D and G bands in the Raman spectra, typically found between 1200 and 1600 cm -118 . The formation of carbonaceous materials on the catalyst surface, could lead to catalyst deactivation 19,20 . From the mass spectra recorded during the operando Raman experiments (Supplementary Figure 7), the CO2 signal (m/z=44, orange line) indeed seemed lower for the CoO catalysts compared to the metallic Co catalysts. Carbon deposition on the CoO catalysts could thus be (one of) the reason(s) for their lower activity. Though, for all catalysts the CO2 signal decreased with temperature, indicating that more CO2 was converted at higher temperatures. Besides, methane (m/z=16, magenta line) and H2O (m/z=18, purple line) were followed during the operando Raman experiments. Their increase indicates increasing activity. The hydroxymethyl fragment (m/z=31, green line) was followed as a measure for oxygenated products. However, the hydroxymethyl fragment remained constant at 0 during all experiments and no prove was found for the formation of oxygenates. Small amounts of ethane or CO were indicated by the variation of the m/z=28 fragment (blue line in Supplementary Figure 7).

Supplementary
Activation energies (Ea) were calculated via Arrhenius plots from the CO2 conversion 21 as measured with mass spectrometry at different temperatures during the operando Raman experiments (Supplementary Figure 8).
For the CoO catalysts, the activation energy decreased with increasing support reducibility in the order of Co/SiO2>Co/Al2O3>Co/TiO2>Co/CeO2. For the metallic Co catalysts, the activation energies were typically lower compared to their CoO counterpart, except from Co/CeO2, where Co/CeO2-red had a higher activation energy than Co/CeO2-ox.  Figures 5 and 7).
The first peak is typically assigned to the reduction of Co3O4 to CoO (@250-290°C), while the second peak is ascribed to the reduction of CoO to metallic Co (@300-420°C) 22 . For Co/Al2O3, two other peaks were clearly visible at 673 and 958°C, which have been ascribed to cobalt aluminates species. Similarly, for Co/SiO2, an additional peak was observer around at 635°C, which could indicate the reduction of cobalt silicates 22,23 .
Supplementary Figure 9. H2-TPR profiles of the cobalt-based catalyst under study recorded with a thermal conductivity detector (TCD).

CO2-temperature programmed desorption
The CO2-TPD profiles for the SiO2, Al2O3, TiO2, and CeO2 support materials are displayed in Supplementary  Figure 10. Different types of basic sites were identified: weak (<150°C), medium (150-350°C), and strong (>350°C) 24 . The TPD peaks generally shift to higher temperature and increase in intensity when the basic sites become stronger and increase in quantity 25 . Both the non-reducible supports SiO2 and Al2O3 were dominated by strong basic sites 26 , with their main desorption peaks at 430°C and 393°C, respectively. The reducible supports displayed different basic sites. For TiO2, two peaks were observed: one at 199°C, which was assigned to bidentate carbonate decomposition from medium strength basic sites, and a smaller peak at 390°C, which was attributed to monodentate carbonate decomposition from strong basic sites. For CeO2, we observed a peak at 135°C, ascribed to bidentate carbonate decomposition from weak-medium strength basic sites, and another peak at 397°C, interpreted as monodentate carbonate decomposition from strong basic sites 24 . The total amount of desorbed CO2 was estimated from the integral of the CO2-TPD peak area. The CO2 surface density quantification was used as a measure for the basicity and increases from SiO2<Al2O3<CeO2<TiO2 (Supplementary Table 4), which is conform with earlier reports in the literature 27,28 . Though, some studies report that CeO2 was more basic compared to TiO2 29,30 , it has also been reported that the basicity decreased with increasing calcination temperature of the metal oxide 31 and is consequently a more complex cohesion than just the chemical element.
Supplementary Figure 10. CO2-TPD profiles of the support materials used in this study recorded with a thermal conductivity detector (TCD).

Density functional theory calculations
To gain a theoretical understanding of the differences in CO2 adsorption on CoO versus metallic Co, we performed density functional theory (DFT) calculations on the face-centered cubic (FCC) CoO and metallic Co surface facets. We performed geometry optimizations of Co(110), Co(111), and CoO(100), the most active facets, both with and without adsorbed CO2. The obtained structures are visualized in Supplementary Figure  11. The geometry optimization of CO2 on the Co(111) surface resulted in a stable structure with a positive adsorption energy of 35.4 kJ/mol, indicating an endothermic process and it is likely that CO2 activation does not take place on this facet 32 33 . Moreover, the support material, which we excluded from our calculations, plays a significant role in the resulting adsorption energy 34 .

Peak assignments
All peak assignments for the phase-resolved amplitude spectra in Figure 2 in the main text and Supplementary  Figure 12 can be found in Supplementary Mass spectrometry signals during ME DRIFTS Supplementary Figure 13. Mass spectrometry signals of CH4 (black, m/z=15) and C2+ hydrocarbons (red, m/z=29) during the ME DRIFTS experiments on (from left to right) Co/SiO2, Co/Al2O3, Co/TiO2, and Co/CeO2. The top panel displays the CoO-containing (suffix: -ox) catalysts and the bottom panel the metallic Co (suffix:red) catalysts. For Co/Al2O3-red the CO signal (gray, m/z=28) was significant and added as well.

Kinetics from phase delays
The phase-resolved amplitude spectra and corresponding phase shifts for the set of cobalt-based catalysts can be found in Supplementary Figure 14a and b, respectively. The phase shifts for selected species (adsorbed CO, (bi)carbonates, and formates) can be found in Supplementary Figures 15 and 16. The slopes of the desorption branches taken from the first 10 s after turning the CO2 gas off can be found for the same selected species in Figure  Supplementary Figure 15. Kinetic information derived from the phase shift. a Schematic representation of the H-assisted mechanism, dominant for CoO (suffix: -ox) catalysts, and the direct dissociation mechanism, dominant for metallic Co (suffix: -red) catalysts. b Phase-resolved amplitude spectra for Co/TiO2-ox and Co/TiO2red. Phase shifts of selected species from the phase-resolved amplitude spectra for c Co/TiO2-ox and d Co/TiO2red. Both samples showed carbonate, formate (*HCO2 2indicated in red; 1609-1615 cm -1 ), and formyl species. Co/TiO2-ox additionally showed *CH2 species. For Co/TiO2-red, *CO (gray; 1980 cm -1 ) displayed faster kinetics (smaller phase shift) than the carbonate, formate, and formyl species. The direct dissociation mechanism was thus faster than the H-assisted mechanism. Figure 16. Phase shifts for (from left to right) Co/SiO2, Co/Al2O3, and Co/CeO2, derived from the phase-resolved amplitude spectra. The top panel displays the CoO-containing (suffix: -ox) catalysts and the bottom panel the metallic Co (suffix: -red) catalysts. Co/SiO2-ox only showed gaseous CO, while Co/SiO2-red only displayed *CO. Co/Al2O3-ox and Co/CeO2-ox showed carbonate, formate (*HCO2 2indicated in red; 1609-1615 cm -1 ), and formyl species. The metallic Co catalysts (bottom) displayed *CO (gray; 1980-1994 cm -1 ), which had mostly faster kinetics (smaller phase shift) compared to the carbonate, formate, and formyl species. Figure 17. Slopes of the desorption branches of the time-resolved DRIFT spectra after turning off the CO2 gas for (from left to right) Co/SiO2, Co/Al2O3, and Co/CeO2. The top panel displays the CoO-containing (suffix: -ox) catalysts and the bottom panel the metallic Co (suffix: -red) catalysts. Co/SiO2-ox only showed gaseous CO, while Co/SiO2-red only displayed *CO. Co/Al2O3-ox and Co/CeO2-ox showed carbonate, formate (*HCO2 2indicated in red; 1609-1615 cm -1 ), and formyl species. The metallic Co catalysts (bottom) displayed *CO (gray; 1980-1994 cm -1 ), which had faster kinetics (steeper slopes) compared to the carbonate, formate, and formyl species. Figure 18. Determination of the slopes of the desorption profiles of selected species in the time resolved DRIFT spectra. In the row 1 and 3, the temporal dependence of the intensity of the selected species is plotted, while that of the corresponding ln(A/A0) values is shown in row 2 and 4. A is the intensity of the species and A0 is the intensity of the species at t=60 s (i.e. when the CO2 is turned off). Suffix -ox denotes the CoO catalysts and suffix -red denotes the metallic Co catalysts.

Kinetic parameters for Co-TiO2-ox and Co/TiO2-red
For the best performing catalyst in our study, Co-TiO2, we additionally determined a set of kinetic parameters at P=20 bar in both the CoO and metallic Co state. The apparent activation energy (Ea) for CO2 hydrogenation was slightly lower for Co/TiO2-ox, 113 ± 3, compared to Co/TiO2-red, 122 ± 5 (Supplementary Figures 19, 20 and Table 1 in the main text). This is in line with the better performance of Co/TiO2-ox compared to Co/TiO2-red. To gain more insights, we also determined the reaction orders in CO2 and in H2 for both samples. The higher reaction order in CO2 of Co/TiO2-ox (0.38 ± 0.09) compared to Co/TiO2-red (0.15 ± 0.04) indicated that a strongly adsorbed intermediate derived from CO2 on the Co/TiO2-red surface, most likely adsorbed CO, hinders the reaction 47 . For metallic Co, a reaction order of 0.14 in CO2 has been reported previously for the CO2 hydrogenation reaction 49 . Besides, the reaction orders in H2 were almost completely opposite for the 2 samples: a positive order of 1.24 ± 0.40 for Co/TiO2-ox versus a negative order of -1.15 ± 0.07 for Co/TiO2-red. This particularly substantiates the hypothesis that the Co/TiO2-ox catalyst, following the H-assisted mechanism, benefits from a higher partial pressure in H2. On the other hand, the Co/TiO2-red catalyst, following mainly the direct dissociation mechanism, benefits from a lower partial pressure in H2, as H2 may be competing with adsorbed CO, the most important intermediate in the direct dissociation mechanism.
Supplementary Figure 19. Arrhenius plots for Co/TiO2-ox (left, pre-treated at 250°C in N2/H2=2) and Co/TiO2red (right, pre-treated at 450°C in N2/H2=2). The activities were measured between 200 and 280°C after 1 h of stabilization with at least 4 GC injection points at each temperature point.
Supplementary Figure 20. Arrhenius plots for selected products CH4, C2+, and CO for Co/TiO2-ox (top, pretreated at 250°C in N2/H2=2) and Co/TiO2-red (bottom, pre-treated at 450°C in N2/H2=2). The activities were measured between 200 and 280°C after 1 h of stabilization with at least 4 GC injection points at each temperature point. Note that CO was not produced below 240°C.

Thermodynamics
The thermodynamic feasibility of the different cobalt phases metallic Co, CoO, Co3O4, and Co2C was assessed with thermodynamic calculations. During CO2 hydrogenation, co-feeding CO2 and CO, and FTS (Supplementary Figure 21a-c), metallic Co and CoO were both thermodynamically feasible. Co3O4 and Co2C, however, were not present in significant amounts during the simulation at 250°C while varying the pressure between 1 and 20 bar. Additionally, under FTS conditions we varied the H2/CO ratio between 0 and 4 at constant temperature (250°C) and pressure (20 bar) (Supplementary Figure 21d). We found that the formation of Co2C was only thermodynamically feasible at H2/CO<1. Though, at typical cobalt-based FTS conditions (H2/CO=2) metallic Co and CoO were the dominant phases. Co3O4 was again not significantly present during the simulation.

Long-term stability testing for Co/TiO2
Long-term stability was tested for Co/TiO2-ox (Figure 5cd in the main text) and Co/TiO2-red (Supplementary Figure 22). The tests were performed at 250°C and 20 bar for 150 h in total: first for 50 h under CO/CO2 cofeeding conditions (CO2/CO=2) and then for 100 h under CO2 hydrogenation conditions (H2/CO2=3). For Co/TiO2ox (Figure 5cd in the main text), during the 50 h of co-feeding, the total carbon conversion started at ~18% and stabilized after about 10 h to ~16%, while the C2+ selectivity started at ~40% and stabilized at ~35%. For the following 100 h of CO2 conversion only, the conversion started at ~7.0% and remained ~4.5% after 100 h, while the C2+ selectivity increased from ~10% in the first few h to ~20% after 100 h, indicating that the activity loss over time was mostly related to a decrease in methane production. For Co/TiO2-red (Supplementary Figure 22), during the 50 h of co-feeding, the total carbon conversion started at ~18% and decreased to ~16% after 50 h, while the C2+ selectivity started at ~10% and increased to ~13%. For the following 100 h of CO2 conversion only, the conversion remained stable at ~5.0% during the 100 h, while the C2+ selectivity remained stable around ~5.0% during the 100 h. After 150 h time-on-stream, we verified with XRD that Co/TiO2-ox contained CoO and Co/TiO2-red contained metallic cobalt (FCC) (Supplementary Figure 23).