On the Cobalt Carbide Formation in a Co/TiO2 Fischer–Tropsch Synthesis Catalyst as Studied by High-Pressure, Long-Term Operando X-ray Absorption and Diffraction

Operando X-ray absorption spectroscopy (XAS) and X-ray diffraction (XRD) were performed on a Co/TiO2 Fischer–Tropsch synthesis (FTS) catalyst at 16 bar for (at least) 48 h time-on-stream in both a synchrotron facility and a laboratory-based X-ray diffractometer. Cobalt carbide formation was observed earlier during FTS with operando XAS than with XRD. This apparent discrepancy is due to the higher sensitivity of XAS to a short-range order. Interestingly, in both cases, the product formation does not noticeably change when cobalt carbide formation is detected. This suggests that cobalt carbide formation is not a major deactivation mechanism, as is often suggested for FTS. Moreover, no cobalt oxide formation was detected by XAS or XRD. In other words, one of the classical proposals invoked to explain Co/TiO2 catalyst deactivation could not be supported by our operando X-ray characterization data obtained at close to industrially relevant reaction conditions. Furthermore, a bimodal cobalt particle distribution was observed by high-angle annular dark-field scanning transmission electron microscopy and energy-dispersive X-ray analysis, while product formation remained relatively stable. The bimodal distribution is most probably due to the mobility and migration of the cobalt nanoparticles during FTS conditions.


S1.2 X-ray Diffraction
shows the X-ray diffraction (XRD) pattern of the calcined (fresh, yellow line) and reduced (grey line) 10 wt.% Co/TiO 2 catalyst diluted with diamond powder (microcrystalline powder, ̴ 1 µm, Sigma Aldrich) in a ratio of 1:3 by mass. Phases of bulk compounds are represented by the stick plots below the diffractograms. The metallic cobalt is divided in two phases, Co-fcc (dark green and *) and Co-hcp (light green and o), and for clarity those sticks are shown with an offset. The diamond peaks are indicated with the ◊. From this pattern it is clear that the strongest metallic cobalt peaks overlap with the diamond peaks. Figure S3 shows the X-ray diffraction (XRD) pattern of the calcined (fresh, yellow line) and reduced (grey line) undiluted 10 wt.% Co/TiO 2 catalyst. Here the metallic cobalt peaks are visible for the reduced catalyst.    Table S1 Co/Ti ratio of fresh catalyst determined from EDX in the STEM for several regions. The average ratio was determined to be 0.10 with a standard deviation of 0.025.

S1.4 Reduction: XRD and XAS
Reduction of the ̴ 10 wt.% Co/TiO 2 diluted with diamond dust is shown in Figure S4 and S5 for XRD and XAS, respectively. In both cases, the temperature was increased to 400 °C in pure H 2 (10 °C/min) at atmospheric pressure and was held at temperature for 2 h. The reduction follows two steps. In the first reduction step Co 3 O 4 species are reduced to CoO. The second step follows after Co 3 O 4 is depleted and represents the reduction of CoO to metallic Co.
In the inset of Figure S4, the intensity sticks of different cobalt phases are indicated to show the two-step pathway. The peaks representing Co 3 O 4 (yellow sticks) disappear and the CoO peaks are coming up (red sticks). The latter disappear as well as the metallic cobalt peaks arise (dark and light green sticks). From analysis by Rietveld refinement in the Brucker TOPAS software, the averaged particle size was calculated at 15-20 nm. This same pathway from Co 3 O 4 to CoO and metallic Co is clearly visible in Figure S5 for the XAS experiment. Due to unforeseen circumstances at the beamline, the reduction had to be performed twice (partially), prior the Fischer-Tropsch synthesis. Figure S5a shows the linear combination fitting done in the Athena software, making use of the fresh catalyst (representing the Co 3 O 4 phase), a CoO reference and a metallic cobalt foil reference, where panel Figure S5b shows the XANES spectra during reduction. Figure S6 shows both reductions. It can be seen that the catalyst only re-oxidized to CoO after exposure to air and was easily reduced again. Figure S4 In situ X-ray diffraction patterns of the reduction (at 400 °C for 2 h with 3 mL/min pure H 2 ) of the ̴ 10 wt.% Co/TiO 2 diluted 1:3 with diamond powder by mass. The two-step pathway from Co 3 O 4 to CoO to metallic Co is clearly visible. The inset S 8 Figure S5 a) The two step pathway from Co 3 O 4 to CoO to metallic Co is clearly visible, when a linear combination fitting is done using the spectrum of the fresh catalyst (as Co 3 O 4 ), a CoO reference and a metallic Co foil reference (Co-fcc). Due to unforeseen circumstances the catalyst was exposed to air and a second reduction was done prior FTS, indicated by the dashed line. b) In situ X-ray absorption spectra of the first reduction (at 400 °C with 5 mL·min -1 pure H 2 ) of the ̴ 10 wt.% Co/TiO 2 catalyst diluted 1:3 with diamond powder by mass. Figure S6 a) Shows the first reduction and b) the second, after unforeseen circumstances prior the Fischer-Tropsch synthesis. As can be seen, the catalyst only re-oxidized to CoO and was easily reduced again.

S1.5 Catalytic Testing
Figure S7 Catalytic testing done by Shell Global Solution B.V., Amsterdam. a) shows the selectivity towards C 1 (methane) and C 5+ during a 60+ h catalytic run at 220 °C, 13 bar and a H 2 /CO ratio of 1. b) shows the numbers for different selected hydrocarbon regions within the run. c) Represents the n-paraffin/ 1-olefin ratio for the hydrocarbons C 5-10 .
S 10 S2 Extra Information on the Undiluted XRD Experiment The selectivity towards those different products showing a preference towards methane. c) Comparison of the production of olefins and paraffins. Note that C 5+ products were not quantified in this experiment

S 11
S3. Extra information on Operando X-ray Absorption Spectroscopy

S3.1 Experimental procedure for first operando XAS experiment
The length of the catalyst bed was approximately 5 mm. The capillary was loaded in to a custom built operando XAS cell. 1 A graphite ferrule was used to hold a thermocouple in place in the gas stream, fed into the quartz capillary, pressed against the quartz wool ensuring correct temperature measurement and catalyst bed stability. The flow system was leaked checked with helium at 15 bar prior to the start of the experiment. The cobalt in the catalyst was reduced at ambient pressure at 400°C under a flow of 50% H 2 , balance helium, at a flow rate of 3 ml·min -1 with a ramp rate of 10 °C·min -1 . The cell was then cooled to 220°C and pressurized to 14 bar in a 1:1 H 2 :CO at a total flow rate of 3 ml·min -1 . A mass spectrometer (Hiden Analytical QGA) was used to determine and quantify reaction products on-line in the gas stream during the XAS experiments ( Figure S8). was not stable anymore and started to slowly decrease, probably due to wax formation in the gas lines.
S 12 S 13 S 14 Figure S11 shows the Co K-edge XANES of for Co 3 O 4 , CoO, Co-hcp, Co-fcc and Co 2 C references, as well as k and R space, used in this manuscript.  S 16 Table S2: EXAFS refinement of Co foil EXAFS in Figure S13 with analogous color-coding:

S3.3.2 k 2 space during FTS and modelling
k space during the 47 h FTS experiment. Figure 14a shows everything during the experiment, while b represents the modelling (described below) of the start and the end of FTS, as well as after decarburization.

S3.3.3 Modeling of first, last and decarburized sample
Prior to the XAFS data analysis the initial treatment of the experimentally obtained raw data was conducted using the Athena software following the standard procedure. Spectra were energy calibrated (using the Co foil that was simultaneously scanned during the experiment), merged, and normalized.  S 18 intensity of the last scan sample can indicate towards an incomplete formation of perfect Co 2 C phase or the add mixture of secondary metallic Co phase.
Following the standard data treatment procedures in Athena interface the energy dependent absorption spectra were converted to k space and later the k 3 weighted χ(k) spectra were Fourier transformed to R space (χ(R)). EXAFS modelling was done over the k range 3.0 -11.7 Å -1 and a R range 1.  From the structural point of view, it is difficult to discriminate between the Co-hcp and Co-fcc phases by EXAFS modelling as both structures have similar Co -Co atomic arrangements in the 1 st and 2 nd coordination shells ( Figure S12 and Table S3). The 3 rd coordination shell of the Co-hcp phase has lesser intensity compared to Co-fcc phase because of lesser C.N and in the S 19 4 th co-ordination shell of Co-fcc has higher intensity because of multiple scattering paths. 3,4 When using cif files to generate scattering paths, the cif file for Co-hcp was used instead of Co-fcc as the linear combination fitting results showed that the majority (~ 80%) of Co-hcp phase present in the First scan and Decarburized catalysts. Attempts to model the data using Co-fcc structure files were also done, but the overall R-factor was worse for each model. This suggests, but is not conclusively proven, that the Co phase present under FTS conditions and after decarburization is Co-hcp instead of Co-fcc. Depending upon the quantity of the mixed phases the coordination number of the above mentioned scattering paths varies with reaction conditions. This has been quantified here in details. The values of the EXAFS best fitted parameters for the First scan, Decarburized and Last Scan samples are summarized in Tables S4-6, respectively.
Modelling of the EXAFS data was an iterative process. Given the strong scattering path at 2.5 Å in the FTS 1 hr and decarburized scans the two spectra were modelled with only scattering S 20 paths from Co-hcp. Likewise, the similarity of the FTS 47 h scan to that of bulk Co 2 C initially resulted in a model containing scattering paths from Co 2 C. These models were able to account for some features in the EXAFS but not all of them. The inclusion of a second phase (either Co 2 C or Co-hcp) was required to obtain the best fit models as observed by the improvement of the quality of the data fitting or minimizing the mismatch between the experimental data and theoretically generated spectra as well as an improvement in the error of the models.
To model the mixed phase EXAFS simultaneous fitting, with shared parameters, was performed. The FT-EXAFS spectra and their imaginary parts fitted with the corresponding theoretically generated model are depicted in the Figure S15b. In addition to this, a rough estimation about the phase fraction of the metallic Co and Co 2 C was obtained from the data fitting by multiplying the bulk coordination of each scattering paths to a phase factor x for Co 2 C and (1-x) for metallic Co. The phase fraction between the Co 2 C and metallic Co phases were obtained near to 60:40 ratios.
The coordination number of the Co -C scattering path for the FTS 1 h and Decarburized scans is less than 1 while it is near to 2 for the FTS 47h scan, indicating a lesser Co carbide fraction during the beginning of FTS and after decarburization compared to 47 h of FTS. The value of Co -C coordination number and atom to atom distances match to the previously reported results for freshly carburized catalysts. 5 The value of the Co -Co coordination number in the 1 st Co-CO scattering path is close to the bulk coordination and much higher for the FTS 1 h and decarburized scans than the FTS 47 h scan indicating a dominant metallic Co present in these samples. The scattering paths at longer R distances have reduced coordination numbers in comparison to the bulk materials possibly as a result of particle size effect of the nanoparticles.

S 21
As a conclusion, the work shows that Co 2 C scattering paths can be detected in the first hour of FTS and that the Co nanoparticles are partially converted to cobalt carbide during the 47 h FTS experiment. Decarburization at 300°C converts the majority of the carbide phase back to metallic Co, though there is some residual carbide present.

S3.4 Extra information on the whole range linear combination fitting
LSLC is performed over a range of -20 to +50 eV around the absorption edge on the normalized spectra. The reference spectra where weighted between 0 and 1 but not forced to sum to 1.
No noise was added and combinatorics was set to use all the selected references (Co-fcc, Cohcp and Co 2 C in case of FTS conditions and Co 3 O 4 , CoO, Co-fcc and Co-hcp in case of reduction conditions).

S3.5 Cauchy wavelet transform analysis
Continuous Cauchy wavelet transform (CCWT) [7] was used to visualize the changes in k-and R-space during the in-situ XAS FTS experiment. The CCWT was performed using a k-range of k = 0-11.7 Å and a R-range of R = 0.-6 for all EXAFS spectra. S 32

S5.1 Pressure and Temperature profile
After 48 h of FTS, the H 2 /CO mixture was replaced by pure H 2 in order to decarburize the catalyst (still at 16 bar and 220 °C). After several hours the temperature was increased by 25 °C every hour until 300 °C and until a fully decarburized catalyst was obtained. Then, pure CO at 220 °C was flowed in order to carburize again until the beam time was over. The temperature and pressure profile of the whole experiment, including decarburization and re-carburization is shown in Figure S22.

S5.5 Extra information on Cluster 3
As can be seen from Figure 5, 8 and S26, cluster 3 has a lower carbide average than cluster 2. This could be explained due to the decarburization we apply later on during the experiment. were fitted with the reference spectra of Co-hcp, Co-fcc and Co 2 C and the linear combination result is shown in e).
With TOS the amount of Co 2 C phase is increasing and metallic Co decreasing. The opposite holds for the decarburization.

S 36
Note that the following data is normalized and mean centered and therefore does not show an edge jump.
Cluster 3 consists of 6 spectra, three of them (24-26 h TOS) were recorded between cluster 2 or 4 and the other three between cluster 6 and 7 (see Figure S27a). The time periods of 24-26 h and 59-61 h are clearly visible in Figure S27b (red rectangles) due to the lack of spectral features. The less detectable features can be seen in Figure S27c/d (black and yellow line, respectively) compared to the surrounding clusters. This can be well explained by the fact that the features of clusters 2 and 4 and clusters 6 and 7 are almost perfectly anticorrelated. A mix of those 2 phases will flatten all features and this shows as a transition phase between a pair of clusters. It means the phase transition between these 2 clusters is much slower than our time resolution. It seems that spectra 24-26 are not exactly/purely a transition region, however, as they more closely resemble cluster 4 than 2. This is probably due to the sample realignment that was conducted in this time window (the hutch was entered to replace the CO detector necessitating a halt to collecting the XAS spectra). The signal was temporarily lost but recovered within those 3 hours (see Figure S28). The phase transition between hours 59-61 on the other hand is real and is made more visible in the score plots in Figure S28. The top graph shows the 3D score plot and the bottom the 2D with only PC1 and PC2. The numbers correspond to the hours TOS during the experiment and the colours correspond to the clusters assigned. The arrows indicate the time series in which spectra were recorded and were added to guide the eye. Cluster 3 (numbers 24-26 and 59-61) is manually "separated" by the striped lines nicely visualizing what happened. One can clearly see that the spectra recorded between 59-61h are a transition between the clusters 6 and 7 (green and dark blue, respectively) and that spectra 24-26h actually connect clusters 2 and 4 (orange and dark yellow, respectively) but have been clustered together with the spectra connecting clusters 6 and 7 because their S 37 spectral features (or better, the absence of them) caused by the event that took place as described above, made them similar to the transition phase between clusters 6 and 7. This is an example for the challenges of long-term and high-pressure experiments and how careful data analysis can detect, isolate, and explain unforeseen events that might take place. If we had not done the clustering, we would also not have noticed this transition phase happening between 59-61h TOS. Figure S27. a) Indicates cluster 3 and the corresponding hours TOS. b) Shows the contour plot of all the spectra with TOS where the red rectangles indicate the times corresponding to cluster 3. As can be seen here, those appear to be almost featureless. This is made more visible in c) and d), where the average of the spectra 24-26h and 59-61h are plotted together with the clusters they connect.
S 38 are a transition between the clusters 6 and 7 (green and dark blue, respectively) and that spectra 24-26h should connect clusters 2 and 4 (orange and dark yellow, respectively), but due to the events that took place in this time period these spectra are almost featureless and have therefore been clustered together with the transition phase between clusters 6 and 7.

S6.1 Spent images + histograms of sizes
Cobalt species had a wide size distribution ranging from sub-nm species up to 50 nm particles. Due to the low contrast, proper particle size analysis is not trivial. The analysis is blind to sub-nm species and does not include larger agglomerates (>5nm) that were seen elsewhere in the sample. Excluding sub-nm species and large agglomerates (>5 nm), the average particle size can be roughly estimated to be around 3 nm. Figure S29. HAADF-STEM and EDX images of the spent catalyst including the Co particle size histograms taken from chemical maps of the spent catalyst after decarburization and re-carburization. Average size for the left is 2.5 nm (± 0.7 nm), for the right 3.8 nm (±1.5 nm).

S6.2 Carbon mapping of spent exp-2
Carbon has mostly uniform distribution over the titania. Some Co particles exhibit enhanced concentration of coke, but not all metal particles display such high coke accumulation, as can be seen in Figure S30. Here, two regions are shown of which region one shows a cobalt particle with a high accumulation of carbon deposit and region 2 does not.