The Characterisation of Hydrogen on Nickel and Cobalt Catalysts

We have investigated a series of supported and unsupported nickel and cobalt catalysts, principally using neutron vibrational spectroscopy (inelastic neutron scattering, INS). For an alumina supported Ni catalyst we are able to detect hydrogen on the metal for the first time, all previous work has used Raney Ni. For an unsupported Ni foam catalyst, which has similar behaviour to Raney Ni but with a much lower density, the spectra show that there are approximately equal numbers of (100) and (111) sites, in contrast to Raney Ni that shows largely (111) sites. The observation of hydrogen on cobalt catalysts proved to be extremely challenging. In order to generate a cobalt metal surface, reduction in hydrogen at 250–300 °C is required. Lower temperatures result in a largely hydroxylated surface. The spectra show that on Raney Co (and probably also on a Co foam catalyst), hydrogen occupies a threefold hollow site, similar to that found on Co(101¯0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10\bar{1}0$$\end{document}). The reduced surface is highly reactive: transfers between cells in a high quality glovebox were sufficient to re-hydroxylate the surface.


Introduction
The Fischer-Tropsch (FT) process is a well-established chemical reaction that utilises synthesis gas (CO and H 2 ), derived from natural gas, coal or biomass, to produce a wide-range of valuable hydrocarbon products [1,2]. Currently, worldwide fuel and chemical production is based predominantly on crude oil, although this is not sustainable. Fischer-Tropsch synthesis (FTS) could be a viable alternative; however, the FT process's economic viability is strongly linked to the oil price [3]. The development of the FT process both industrially and academically, will often follow the trail of the cost and availability of crude oil as a result. Currently there is a resurgence of interest in the FT process for synthetic fuel production, because of the decreasing lifetime of crude oil, as well as a combination of other environmental and political factors [1,3,4].
The criteria required for an effective Fischer-Tropsch catalyst is to promote both a high capacity of hydrogen adsorption as well as efficiently dissociate adsorbed CO. Several metals are effective FT catalysts such as ruthenium, nickel, iron and cobalt [1,4,5]. Of these, ruthenium is in limited supply and thus expensive and nickel catalysts result in large quantities of undesirable methane and can also produce highly toxic nickel carbonyl compounds [1,4]. (However, nickel is a very efficient hydrogenation catalyst and is widely used industrially for this purpose [6][7][8]). Consequently, only iron and cobalt based catalysts are currently used industrially as FT catalysts [1,2]. Iron is abundant and is therefore the cheapest of the active FT catalysts [3]. Iron based catalysts have a lower hydrogenation activity in comparison to cobalt and therefore, have a high selectivity towards the production of olefins and oxygenate products with promoters commonly added to further manipulate the product slate [1,2]. The current demand for an alternative route to produce lower olefins (C 2 -C 4 ), high value chemical feedstock, free from the use of crude 1 3 oil, has led to increased interest in the Fischer-Tropsch to olefins (FTO) process, in which iron-based catalysts are ideally suited [9][10][11][12][13]. However, iron-based catalyst also catalyse the water gas shift reaction and exhibit a much more complex structure under FT conditions resulting in uncertainty over the proposed active phase and operation of the catalysts [1][2][3][4].
On the other hand, cobalt catalysts exhibit a lower water-gas shift activity that makes them particularly advantageous for the conversion of natural gas derived syngas which possesses a high H 2 :CO ratio [1,2,4]. As cobalt is a relatively expensive metal, to maximise its use typically the cobalt metal (10-30 wt%) is dispersed on a high surface area stable oxidic support such as alumina, silica or titania. A second metal (0.05-1 wt%), often a noble metal (e.g. Pt, Ru, Re), is commonly added as a promoter to enhance the reduction process and to keep the Co surface 'clean' [1,2,14]. Cobalt catalysts provide selectivity towards high molecular weight paraffinic products and are generally used as low temperature FT (LTFT) catalysts. High temperature FT (HTFT) results in more methane production [1][2][3][4], as the purpose of the high temperature is to force the formation of a lighter product slate. There has been a shift away from FT processes that can produce petrol (HTFT) towards wax production, which is then hydrocracked to produce diesel [15,16].
Cobalt based catalysts exhibit a range of desirable properties which have made them an attractive class of catalysts for FT synthesis including a high hydrogenation activity and that they deactivate less rapidly than iron-based catalysts [1,2,17,18]. Cobalt catalysts were utilised in the first commercial FT plants in [1938][1939][1940][1941][1942][1943][1944][1945] in Germany and Japan and today are currently used by Sasol/QP in the Oryx GTL plant in Qatar, Shell in the SMDS plant in Malaysia as well as the Pearl plant in Qatar [1,3]. Academic and industrial research over the last 20 years has increased the knowledge base of cobalt Fischer-Tropsch synthesis (FTS) substantially by understanding the effect of crystallite size on activity, selectivity and deactivation mechanisms [19][20][21]. However, despite the process being applied industrially for close to a century, the underlying mechanisms which govern the FT reaction remain under debate.
While the interaction of carbon monoxide with cobalt has been extensively studied (e.g. [22][23][24][25]), the state of hydrogen on the surface of cobalt has been much less studied. In particular, studies with vibrational spectroscopy are very rare. This is probably because of the usual reasons: adsorbed hydrogen generally gives weak bands and often they occur in the low energy region of the spectrum that is masked by the intense bands of the support; hydrogen on platinum exemplifies this issue ( [26] and references therein). Hydrogen on single crystal Co(1010 ) under ultra high vacuum (uhv) has been studied by high resolution electron energy loss spectroscopy (HREELS) [27]. Inelastic neutron scattering (INS) spectroscopy is a form of vibrational spectroscopy that is particularly sensitive to vibrations involving hydrogen motion [25] and has been extensively used to study hydrogen on, and in, catalysts [26,[28][29][30], however, to date, it has been little used for cobalt catalysts [31,32].
At the outset of this work, our aim was to use INS spectroscopy to characterise the adsorption sites of hydrogen on a range of unsupported and supported nickel and cobalt catalysts. As we will show, this proved to be much more difficult than we expected. In the bulk of the paper, we describe the trials and tribulations encountered and we conclude with some thoughts as to why these occurred and how future work may overcome them.

INS Spectroscopy
INS spectra were recorded with the TOSCA [44,45] and MAPS [46,47] spectrometers at the ISIS Neutron and Muon Source [48]. As described in detail elsewhere [28,46], while the two spectrometers operate over the complete "infrared spectroscopy" range of 0-4000 cm −1 , they have different operating principles. Neutrons have mass, consequently INS spectra are a function of both energy transfer (ω, cm −1 ) and momentum transfer (Q, Å −1 ). TOSCA follows a fixed trajectory through (Q,ω) space, such that there is a single Q value for each energy transfer value. This means that energy transfer values in the O-H stretch region, ~ 3000-3700 cm −1 , are associated with large momentum transfer and the spectra are dominated by multiphonon processes whose intensity varies as Q 2n , where n ≥ 2 and the informative fundamental modes are unobservable. This has the result that TOSCA spectra are optimal below 2000 cm −1 . In contrast, with MAPS the incident beam (E i ) is a single energy selected from a polychromatic pulse of neutrons. This provides a spectrum in the range 0-E i , and the resolution is a fraction of E i (~ 1.5% for MAPS) thus, to obtain good resolution across the entire spectral range, several incident energies are required. Note that improved resolution comes with reduced spectral range. MAPS can access a wide swath of (Q,ω) space in a single measurement, thus by selecting only the small Q data, the high energy fundamentals are observable. Modes at lower energy are seen, but with poorer resolution than with TOSCA.

3
The complementarity of MAPS and TOSCA is illustrated in Fig. 1. The top part shows the S(Q,ω) map of one of the samples (Co foam as received under 900 mbar hydrogen) recorded with MAPS using E i = 650 meV (5243 cm −1 ). The solid black lines are the trajectories of the forward and backscattering detectors on TOSCA (in all the spectra shown here, these are summed and the average is displayed). Figure 1a is the MAPS data shown in the top part, summed over the range 0 ≤ Q ≤ 10 Å −1 (unless otherwise stated, this range was used for all of the MAPS spectra shown). A broad O-H stretch centred at 3540 cm −1 is easily seen, Fig. 1b, however, there is little detail at lower energies. The TOSCA data, Fig. 1c, shows much more detail at lower energy but there is no sign of the O-H stretch. Thus it is the combination of the two types of spectrometer that provides the best description of the chemical system under investigation. Table 1 lists the materials studied and their source. The Raney Co was supplied by Activated Metal Catalysts as a slurry in water. This was filtered at a Buchner pump and transferred to an aluminium cell. This was all done under a blanket of flowing argon, as Raney Co is potentially pyrophoric in air. For the Ni and Co foam catalysts (prepared under license from Alantum Europe GmbH as described in [49,50]), a wet sample was transferred into a glass vessel and treated under flowing, pure dry nitrogen (80 l/h) which was constantly removed by a vacuum pump. After 10 min the vacuum was adjusted to 400 mbar. After 20 min the catalyst was heated to 150 °C. After 3 h no more water was seen by visual inspection. After the sample was cooled to room temperature, the vacuum pump was stopped. The sample was flushed for 5 min with flowing nitrogen. The sample was transferred into a glove box and sealed into an INS can and cyclically treated with hydrogen in the range ~ 10 −3 mbar to 1220 mbar for 12 h until no further hydrogen consumption/adsorption was measured (capacitive pressure transducer, Baratron). As may be seen from Table 1, as a result of the preparation procedure, all of the unsupported metal catalysts had residual Al bound in various intermetallic phases which could not be leached by alkaline treatment. The Co/C and CoCuMn/C catalysts were prepared as described previously [51,52].

Experimental
Pt promoted Co/SiO 2 and Co/Al 2 O 3 Fischer-Tropsch catalysts were prepared via the wet impregnation of silica and γ-alumina with solutions of Co(NO 3 ) 2 ·6H 2 O and [Pt(NH 3 ) 4 ](NO 3 ) 2 . The total pore volume of the oxide support and maximum solubility of cobalt nitrate in water limit the Co loading that can be achieved in a single   (25 mL) and stirred on a rotary evaporator at 60 °C for 10 min at atmospheric pressure. The silica support (Davisil Grade 62, pore size 150 Å, pore volume 1.15 cm 3 /g, 20.0276 g) was then added to the Co solution. A further 15 mL of water was required to form a slurry, which was stirred on the rotary evaporator for a further 10 min. The rotary evaporator pressure was then reduced to 280 mbar for 30 min. The temperature of the water bath was increased to 75 °C for 1.5 h, then to 85 °C for 1 h and finally the pressure was reduced to 50 mbar, in 10 mbar increments, for 3 h. A dark pink powder (32.4097 g) was obtained and calcined at 250 °C (ramp rate 1 °C/min) in a vertical calcination tube for 6 h under a continuous air flow (1000 mL/min) to yield Co 3 O 4 -SiO 2 (22.7431 g). For the second impregnation, cobalt nitrate (11.4126 g) and tetraammineplatinum(II) nitrate (1.7 mL of the same stock solution) were dissolved in water (35 mL) and stirred on a rotary evaporator at 60 °C for 10 min at atmospheric pressure. Calcined material from the first impregnation (20.0531 g) was added to the Co solution and the mixture was stirred for a further 10 min. The rotary evaporator pressure was then reduced to 280 mbar for 30 min. The temperature of the water bath was increased to 75 °C for 1.5 h, then to 85 °C for 1 h and finally the pressure was reduced to 50 mbar, in 10 mbar increments, for 3 h. The material was calcined in a vertical tube furnace for 6 h under an air flow (1000 mL/min) to yield Co 3 O 4 -SiO 2 (22.9085 g). The material was characterised by BET (surface area: 205 m 2 /g, pore size: 134 Å), XRD (Panalytical X'Pert Pro X-ray diffractometer, Co Kα1,2 radiation from 10° to 90° (2θ), scan rate 1.34°/min, at ambient temperature) (Co 3 O 4 , 15.5(1) nm) and H 2 chemisorption (Micromeritics ASAP 2020) (43 m 2 /g).
For the Co/Pt on γ-alumina catalyst, cobalt (II) nitrate hexahydrate (15.8921 g) and tetraammineplatinum(II) nitrate (1 mL of a 31 mM stock solution) were dissolved in water (25 mL) and stirred on a rotary evaporator at 60 °C for 10 min at atmospheric pressure. The γ-alumina support (Puralox (Sasol), pore size 127 Å, pore volume 0.5272 cm 3 /g, 20.0056 g) was then added to the Co solution and stirred on the rotary evaporator for a further 10 min. The rotary evaporator pressure was then reduced to 280 mbar for 30 min. The temperature of the water bath was increased to 75 °C for 1.5 h, then to 85 °C for 1 h and finally the pressure was reduced to 50 mbar, in 10 mbar increments, for 3 h. A dark pink powder (30.8973 g) was obtained and calcined at 250 °C (ramp rate 1 °C/min) in a vertical calcination tube for 6 h under a continuous air flow (1000 mL/min) to yield Co 3 O 4 -Al 2 O 3 (23.2340 g). For the second impregnation, cobalt nitrate (11.4121 g) and tetraammineplatinum(II) nitrate (1.7 mL of the same stock solution) were dissolved in water (35 mL) and stirred on a rotary evaporator at 60 °C for 10 min at atmospheric pressure. Calcined material from the first impregnation (20.0132 g) was added to the Co solution and the mixture was stirred for a further 10 min. The rotary evaporator pressure was then reduced to 280 mbar for 30 min. The temperature of the water bath was increased to 75 °C for 1.5 h, then to 85 °C for 1 h and finally the pressure was reduced to 50 mbar, in 10 mbar increments, for 3 h. The material was calcined in a vertical tube furnace for 6 h under an air flow (1000 mL/min) to yield Co 3 O 4 -Al 2 O 3 (22.5938 g). The material was characterised by BET (surface area: 95 m 2 /g, pore size: 103 Å), XRD (Co 3 O 4 , 13.7(1) nm) and H 2 chemisorption (56 m 2 /g).
Infrared spectra of the Co 3 O 4 -SiO 2 and Co 3 O 4 -Al 2 O 3 materials were recorded using a Thermo Nicolet Avatar spectrometer (4 cm −1 resolution, 64 scans). The samples (ca 150 mg) were reduced in a custom-built DRIFTS chamber under flowing hydrogen (20 mL/min) at 425 °C (ramp rate of 20 °C/min) for 3 h. Spectra were measured after cooling to room temperature under a hydrogen atmosphere. The sample spectrum was ratioed against a background obtained of dried KBr packed into the DRIFTS chamber.
The catalysts were subjected to a variety of treatments as listed in Table 2. All of these were carried out using a rig [53] designed to handle the large quantities (30-50 g for metal-only catalysts, 5-10 g for supported catalysts) required for INS measurements [28,29,46]. Most of the INS spectra were recorded from samples in aluminium cells that are of the same design as the Inconel cells described previously [53], so allow hydrogen to be flowed through them. They are indium-wire sealed and have maximum operating conditions of 1 barg at 120 °C. The Inconel cells have the advantage that they are rated to 500 °C at 20 barg [53], so allow reduction to be carried out at higher temperatures. In some cases, spectra were recorded with the sample in Inconel cell. The advantage of the Al cells is that they have a much lower background than stainless steel or Inconel cells [53]. After activation, transfers between cells were carried out in an argon-filled glove box (MBraun UniLab MB-20-G, [

Results
In the following sections we describe the results obtained for each of the catalysts in turn. As we were unable to source a pure cobalt Raney Co catalyst, we used a Raney Co (81) Ni(16)Al(3) material. As a check on our methodology, the spectrum of hydrogen on nickel analogues of some of the cobalt catalysts used here was also measured. All previous work involving INS studies of hydrogen on nickel have utilised Raney Ni [40][41][42][43], thus the use of supported catalysts is novel and industrially relevant. The studies of the Ni catalysts will be presented first and then the cobalt systems. Figure 2 shows the spectra of the supported nickel catalysts. Figure 2a-c shows the Ni/Al 2 O 3 catalyst under hydrogen, after evacuation and the difference spectrum. This shows a broad, weak feature at 940 cm −1 that is assigned to hydrogen on Ni(111) facets [40][41][42][43].

Ni/Al 2 O 3 and Ni/C
In contrast, the only feature for the Ni/C catalyst (Fig. 2d) is a weak mode at 650 cm −1 . (The strong features below ~ 400 cm −1 are due to the steel can). The spectrum is unusual in that carbon supports typically give welldefined spectra (e.g. [55][56][57]). This is because the carbon is not present as a standard support material, rather it is the residue from the precursor to controlled decomposition to generate finely divided metal on an unconventional carbon "support" so as to provide good metal dispersion [52]. The difference spectrum shows only a very broad band centred at ~ 600 cm −1 that does not match that expected for hydrogen on nickel. This may be the bending mode of hydroxyls on the Ni, but without MAPS data, this assignment is very tentative.

Ni Foam
The foam catalysts (Ni and Co) are an evolution of the Raney process. In particular, due to their high porosity and low filling weight, substantially less metal mass is required as compared to other activated metal catalyst solutions for fixed-bed applications. The Evonik Metalyst MC™ 9 foam catalysts offer high activity and selectivity, especially for mass transfer limited industrial fixed-bed hydrogenation processes [49,50]. Figure 3 shows the spectra of the Ni foam sample. In the top of the figure, the features below ~ 400 cm −1 are due to the steel can. From the MAPS spectrum in the O-H stretch region (Fig. 3a), it is apparent that residual hydroxyls are present, although the amount is small and that ~ 50% are lost on evacuation at 200 °C (Fig. 3c). The spectra clearly show the presence of adsorbed hydrogen (broad band from 400 to 1200 cm −1 ). This is more clearly seen in the difference spectrum, (Fig. 3d). Room temperature evacuation removes the physisorbed H 2 (Fig. 3f). The overnight evacuation at 200 °C removes the adsorbed hydrogen (Fig. 3e). There are hints of peaks at 900 and 1030 cm −1 , which by comparison to Raney Ni [40][41][42][43], see Fig. 4, are assigned to hydrogen on (111) facets. However, there is also a long tail to low energy, whereas on Raney Ni there were only weak peaks.
These are assigned to hydrogen on "non-(111)" facets. The greater intensity of these shows that there is a much larger degree of surface heterogeneity in this sample as compared to Raney Ni. A crude estimation, based on the relative area of hydrogen on the (111) sites to the total area, would indicate that there are approximately equal numbers of (111) and non-(111) sites in the foam catalyst, whereas for Raney Ni the ratio is at least 5:1 (111) to non-(111). We note that there is 10% Al present in the Ni foam catalyst (Table 1), however, this is similar to that found for Raney Ni [58], so probably does not account for the different surface structures of the foam and Raney Ni catalysts. HREELS investigations of the high coverage phases of hydrogen on Ni(110) [35] show modes at 637, 871 and 1065 cm −1 . On Ni(100) [36,37] the perpendicular mode (symmetric stretch) is at 645 cm −1 and is essentially dispersionless (i.e. there is no change in transition energy as a function of wavevector), whereas the parallel mode (asymmetric stretch) is strongly dispersed, being at 710 cm −1 at the surface Brillouin zone Γ-point and at 850 cm −1 at the X-point. As INS spectroscopy observes all wavevectors [28], this will result in the mode being spread across this range. As most of the additional intensity (as compared to Raney Ni) occurs in the range 500-850 cm −1 , this suggests that the surface of the foam catalyst has a much larger number of (100) facets than Raney Ni, which is largely composed of (111) facets.

Cobalt Catalysts
In the bulk, cobalt exhibits two polymorphs [59], hexagonally close-packed (hcp) and face-centered cubic (fcc), with the hcp phase transforming to the fcc phase above 422 ± 5 °C [60]. It has been reported that for cobalt particles less than 40 nm, the predominant phase should be fcc [61]. However, the temperature of reduction, addition of promoters and the support influence the relative amounts of the phases. In particular, with alumina the fcc phase is obtained [62,63], while on silica fcc [64] or a mixture of both phases was obtained [62,63]. Given the pyrophoric nature of the reduced materials, XRD was not possible with our catalysts. However, given the similarity of the supported catalysts to the literature materials, it is probable that our alumina and silica catalysts are largely fcc cobalt. For the unsupported materials it is less clear; the spectra would suggest that they are hcp cobalt but this is unproven.

Co/SiO 2
The original plan was to prepare the sample by reducing the catalyst under flowing hydrogen at 200 °C and cool under a hydrogen flow. This would be measured on TOSCA. The sample would then be dehydrogenated and measured again  to generate a background. Unfortunately, due to a breakdown in the proton accelerator used for neutron production [48], it was not possible to measure the sample immediately after hydrogenation and it was several months before it could be measured on TOSCA. As shown in Fig. 5a and d, the spectrum is dominated by water (libration at 600 cm −1 and bend at 1660 cm −1 ) and hydroxyls (bend at 1100 cm −1 and O-H stretch at 3340 cm −1 ) [65][66][67], suggesting either a slow reaction with the support via spillover or, more likely, that the can had leaked over time. Following this, the sample was reduced and re-hydrogenated again and re-measured giving a clean result, shown in Fig. 5b and e. The sample was then dehydrogenated for the background measurement shown in Fig. 5c.
The difference spectrum of the re-hydrogenated sample, Fig. 5e, shows a very broad band centred at 500 cm −1 , with maxima at 320, 465, 675 cm −1 . From the literature on Co-H vibrations using HREELS, only the feature at 675 cm −1 is vaguely consistent [27] with the assignment to Co-H. It is possible that the modes are Co-H related or it is also possible they are due to hydroxyls on the cobalt or on the catalyst support. We will return to this point in the Discussion.
The infrared spectrum of the reduced catalyst, Fig. 5f, shows the presence of isolated (sharp band at 3740 cm −1 ) and hydrogen-bonded hydroxyls (broad band at 3365-3720 cm −1 ), both on the silica. The spectrum also illustrates an advantage of INS spectroscopy for the study of supported catalysts: the region below 2000 cm −1 is dominated by Si-O stretch and bend modes rendering observation of species on the metal very difficult. In contrast, INS spectroscopy can access the entire 0-4000 cm −1 range, albeit with inferior resolution. The complementarity of the techniques is obvious. Figure 6 shows the succession of spectra acquired for the Raney Co catalyst with 650 and 300 meV incident energy on MAPS after subtraction of the empty can. If it was apparent from the 650 meV spectrum that the treatment had not resulted in any change to the spectrum, to save time the 300 meV spectrum was not recorded.

Raney Co: 1st Sample
It is apparent from Fig. 6 that only the 250 °C reduction made any significant difference to the catalyst. Drying it at 100 °C, irrespective of whether helium or hydrogen was used, resulted in a hydroxylated surface. Figure 7 shows this in more detail. In the O-H stretch region, there is a sharp peak at 3630 cm −1 and a broad one at ~ 3400 cm −1 ; in the bending region there is an intense, broad peak at 850 cm −1 and a weaker peak at 1700 cm −1 . The peaks are all assigned as surface hydroxyl modes: O-H stretch, Co-O-H bend and its first overtone respectively. The only alternative assignment is to adsorbed water. The 3630 cm −1 mode is too high to be water and agrees with that found by HREELS for hydroxyls on Co 3 O 4 (111) [68]. As seen in Fig. 6, the 3400 cm −1 peak is close to that found for ice, its larger transition energy would require it to be more weakly hydrogen bonded than water. This assignment would also require that that at least some of the 850 cm −1 band is due to the librational modes of water, which are more usually found to peak at 600-700 cm −1 [69], and crucially, that the 1700 cm −1 was the H-O-H bending mode. 1700 cm −1 would be unusually high for the water bending mode, which is usually found at 1600-1650 cm −1 [69]. Examination of the momentum transfer, (Q, Å −1 ) enables this possibility to be eliminated [70]. Figure 8a shows the Q-dependence for the 800 cm −1 mode, 8b that of the 1700 cm −1 feature and 8c that of the bending mode of water. It is apparent that the 1700 cm −1 mode peaks at a larger value of Q (Q max ) than either of the other two modes, thus it cannot be the water bending mode.
Further, its Q max = of 11.8 Å −1 is ~ √2 larger than Q max = 8.9 of the 800 cm −1 mode, consistent with it being After reduction in H 2 at 250 °C, the spectrum has dramatically changed, Fig. 7b. The amount of hydroxyls has been reduced by ~ 60%, although there still appear to be two populations present. The O-H stretch for one of these matches that of ice, however, the lower energy modes are not consistent with this. In the low energy region three weak bands are seen at 910, 1140 and 1690 cm −1 . There must be a Co-O-H bending mode present, it could be one of the three seen here, or it could be hidden by the tail of the peak at 200 cm −1 . Figure 9 shows the spectra for the Co/Al 2 O 3 sample. The infrared spectrum after reduction, Fig. 9a, shows three peaks in the O-H stretch region at 3728, 3668 and 3587 cm −1 . Again, the region below 1200 cm −1 is inaccessible because of adsorption by the support. The INS spectra Fig. 9b, c, d closely resembles that seen for activated η-Al 2 O 3 [71]: O-H stretch at 3720 cm −1 and Al-O-H bend at 890 cm −1 . Hence the additional peak at 420 cm −1 , together with the tail at 3550 cm −1 , must be assigned to hydroxyls on the cobalt nanoparticles. The interaction with H 2 results in new bands at 640, 870, 1610 and 3290 cm −1 . Subtraction of the spectrum of the reduced catalyst from that of the H 2 dosed sample, gives the spectrum shown in Fig. 9d. Comparison with that of ice, Fig. 9e, shows that this is water. That the subtraction works so well in the O-H stretch region, shows that the interaction is with the hydroxyls on the cobalt, as evidenced by the disappearance of the band at 420 cm −1 .

Co/C
The Co/C and CoCuMn/C (see §2.2.5) catalyst materials are characterized by high metal loading with the metal nanoparticles finely dispersed in a non-graphitic carbon matrix. This is readily apparent from the TEM and the particle size distribution, Fig. 10, that shows the carbon to be highly disordered and most of the particles below 28 nm. Cobalt particles larger than 98 nm are not considered to be significant as there is less than 3% of them. The slightly higher ratios of particles in the intervals 140-150 nm and 42-56 diameter particles are most likely caused by clustering/agglomeration effects; TEM results do not indicate differing types of Co particles. The larger particles and their corresponding small surface area will not contribute significantly to the spectra. Figure 11a-c show the spectra of the Co/C sample. It can be seen that all three spectra are very similar apart from the intense peak at 46 cm −1 . As with the Ni/C catalyst (see §2.1.1) and consistent with the TEM, what is striking is that the spectra do not resemble that of a typical carbon support [55][56][57] or that of graphite, Fig. 11d [72]. The only evidence for the carbon is a very weak C-H stretch mode at 2950 cm −1 . This is characteristic of sp 3 C-H rather than the sp 2 C-H expected for hydrogen-terminated graphenelike sheets (> 3000 cm −1 ) [73]. The "as received" sample also shows a weak peak at 3490 cm −1 assigned to residual hydroxyls. The low intensity shows that there are very few of these and they are largely removed on evacuation at 200 °C. The difference spectra, Fig. 11e-g, show the effects of evacuation more clearly. Subtracting either of the evacuated spectra gives very similar results: there is an intense peak at 46 cm −1 , a very broad (> 200 cm −1 ) feature maximising at ~ 200 cm −1 and a weak, sharp peak at 116 cm −1 . The last of these is almost exactly at the transition energy of the of the J = 0 → 1 rotational transition of free H 2 and is the same energy as seen for H 2 physisorbed on graphite [74]. Accordingly it is assigned to H 2 on the support and the broad feature maximising at 200 cm −1 is characteristic of H 2 recoil [75]. The intense band at 46 cm −1 is interesting. Two possibilities are: (i) it is the Co-O-H out-of-plane bending mode, (ii) it is H 2 on Co. The first of these appears unlikely as the O-H stretch is very weak and this mode has not been seen on any of the other cobalt catalysts. In particular, the Co/SiO 2 and Co/Al 2 O 3 spectra suggest that the Co-O-H bending mode occurs around 400 cm −1 and there is no evidence for this.   [76] and Co exchanged AlPO with J = 0 → 1 rotational transition at 58.4 cm −1 [77]) are known that give energies as low as this but not for metals. For the J = 0 → 1 rotational transition to fall to 46 cm −1 from 118 cm −1 would require the H-H bondlength to increase to ~ 1.2 Å from 0.746 Å. Such bondlengths are known for Kubas-type dihydrogen complexes [78]. This implies a strong interaction between the Co nanocluster and H 2 that does not dissociate it. Figure 12 shows the spectra of the CoCuMn/C catalyst. As with the Ni/C and Co/C catalysts there is no indication of the carbon support, in either the C-H deformation or stretch regions. A weak O-H stretch mode is observed at 3590 cm −1 and an intense, broad feature centred at 800 cm −1 with its first overtone at 1650 cm −1 . The difference spectrum highlights this more clearly. Figure 13 shows the difference spectra for the Co foam sample. From Fig. 13a it is apparent that the adsorbed species is disordered water (librational modes at ~ 800 cm −1 , water scissors at 1650 cm −1 , O-H stretch at 3530 cm −1 ), the corresponding TOSCA spectra, Fig. 13b, show the librational modes and also the  translational modes (peak at 49 cm −1 ). The sharp peak at 49 cm −1 is at a similar position to that seen for the Co/C sample, Fig. 11e, f, g, however, both the shape and position are different and, crucially, there are no additional features in Fig. 11f, whereas there are for the Co foam sample. Hence we assign the two features to different species. This again emphasises the reactivity of an activated cobalt surface.

Co Foam
Re-hydrogenation of the sample after evacuation at 200 °C generates the difference spectrum shown in Fig. 13c. This shows an extremely weak, broad band that stretches from 500 to 1600 cm −1 , the weakness is undoubtedly a reflection of the low surface area of the material. However, there are no hydroxyls generated by the addition of hydrogen and the most reasonable assignment is to surface hydrogen. The considerable width of the band is probably a reflection of what is likely to be a highly disordered surface, as was found for the Ni foam sample (see Sect. 4.1.2).

Raney Co: 2nd Sample
Our experience with the Raney Co sample was that a high temperature reduction was required in order to remove the oxide. Accordingly, a second sample was dried and then reduced at 300 °C. This sample was then measured in the Inconel cell and subsequently pressurized to 1 bar H 2 . Figure 14 shows the results. The difference spectrum for the 650 meV data shows a broad band centred at ~ 880 cm −1 , that is more clearly seen in the 300 meV data. Crucially, the 650 meV data shows nothing in the O-H stretch region, so the feature cannot be hydroxyls. Comparison with the transition energies found for the high coverage phase of hydrogen on Co 1010 [27] shows that the envelope encompasses the modes and that there are sub-maxima at, or close to, the Co-H modes found for Co 1010 .

Discussion
As stated in the Introduction, the aim was to characterise the adsorption sites of hydrogen on a range of nickel and cobalt catalysts. The work with the nickel catalysts, demonstrates that our methodology is reliable as we were able to observe adsorbed hydrogen on a supported Ni catalyst, which to our knowledge is the first time this has been achieved; all previous work has used Raney Ni [40][41][42][43]. The novel Ni foam catalyst is clearly much more heterogeneous in terms of surface sites than Raney nickel as judged by the much larger number of non-(111) sites.
The supported cobalt catalysts demonstrate the reactivity of the hydrogenated surface in that even though the transfers between sample cans was carried out in a high quality argon-filled glove box ([O 2 ], [H 2 O] < 1 ppm), the spectra consistently showed the presence of water and hydroxyls (Figs. 6,9,13).
The Co/C catalyst is intriguing because of the presence of the band at 46 cm −1 , the only credible assignment for this, is that it is H 2 coordinated to Co. While Kubas-type dihydrogen complexes have modes at this energy, the additional modes expected for complexed dihydrogen are not seen here. We did not observe it on all the catalysts. However, the mode is best seen with TOSCA (and similar instruments) as this region is routinely recorded with this type of spectrometer and TOSCA was not used for all the samples. MAPS is capable of observing transitions at low energy, but it requires the use of E i < 100 meV. With the higher energies used here (250 and 650 meV), the feature is hidden by the elastic line. It is noteworthy that the Co/C was not transferred between cans.
Our initial treatment of Raney Co was based on our past experience with Raney Ni [43], where temperatures above 100 °C are generally avoided in order to prevent sintering. From Fig. 6a-c, drying in either He or H 2 at 100 °C removes  (a-b). The vertical bars in the lower part are the transition energies at which the Co-H modes are found for the high coverage phase of hydrogen on Co 1010 at the Brillouin zone Γ-point [27] water but leaves a hydroxylated surface. Only when reduction is carried out in flowing H 2 at 250 °C or above are the hydroxyls significantly reduced, although they are not eliminated. This is in agreement with a recent operando XANES study of a Co-Ni-Re/γ-Al 2 O 3 catalyst [79] that found that reduction from Co 3 O 4 to CoO started at 260 ± 10 °C and the subsequent reduction from CoO to Co(0) occurred above 340 ± 10 °C.
The spectrum of Raney Co after high temperature reduction, shows weak features in the region where the Co-H modes are found for the high coverage phase of hydrogen on Co 1010 [27]. Based on this experience, we repeated the measurements with Raney Co but with much more stringent conditions: reduction at 300 °C. The INS spectrum was measured and the cell pressurised to 1 bar H 2 . All of the operations were carried out in an Inconel can, which is why the background in Fig. 14 is so high. The difference spectra show a broad band centred at ~ 800 cm −1 , but crucially, there is no O-H stretch present in the 650 meV data. This means that the band cannot be due to water or hydroxyls and instead is unambiguously assigned to hydrogen on the cobalt surface. The sub-maxima agree very well with the transition energies for H on Co 1010 ; the width of the feature can be ascribed to the greater heterogeneity of the Raney Co(81)Ni(16)Al(3) surface, coupled with the presence of vibrational dispersion in the modes, as is commonly found for adsorbed hydrogen on metal surfaces: e.g. H on Pt black [80], H on Ni(111) [81], H on W(100) [82]. Table 3 lists the modes observed and their assignments.
The stability of the hydroxyls on Raney Co has implications for the only previous INS study of hydrogen on Raney Co by Chojecki et al. [31]. Figure 15 shows their results. Overlaid on the lower part of the figure are the 300 meV  [31] of Raney cobalt a after drying and b after equilibration with 0.2 bar H 2 . Overlaid on the lower part are our data for Raney cobalt dried at 100 °C recorded with 300 meV incident energy. Reproduced from [31] with permission of Springer Nature spectra of our first Raney Co sample after drying at 100 °C (Fig. 7c).
In [31], the spectrum after adsorption of H 2 was assigned to a variety of adsorption sites based on calculations of small cobalt clusters (four or five Co atoms, peak positions indicated). We note that these are unlikely to be a good model of an extended metal surface. However, the broad underlying peak exactly matches what we see in the dried Raney Co and that has been unequivocally assigned to hydroxyls (Fig. 7a-c). The spectra suggest that Chojecki et al.'s activation conditions (473 K for 6 h in vacuum) resulted in a partially dehydroxylated surface that was immediately re-hydroxylated on exposure to hydrogen. The upshift of the ~ 800 cm −1 band in Fig. 15a to ~ 900 cm −1 in Fig. 15b is consistent with increased hydrogen bonding being present.

Conclusions
There are several lessons to be learned from this work. The first is that an oxidic and/or hydroxylated Co surface is very resistant to reduction. An extended period of reduction at > 250 °C is required to remove most of the hydroxyls. It is also clear that the clean surface is highly reactive; our work suggests that even very small amounts of oxygen results in hydroxylation of the surface. This means that transfers between cells, even in a good quality glove box, should be avoided.
It is striking that we were only unambiguously successful in seeing Co-H modes on the unsupported materials. On the supported catalysts we only saw hydroxyls and it is tempting to speculate that this is because of facile spillover on the supported systems that is not possible on the unsupported ones. This is clearly highly contentious because the sample handling (reduction conditions, transfers between cans) is critical and it is quite possible that we have not yet found the "sweet spot" for the supported Co catalysts.
More generally, this work shows that for catalysts that may have hydroxyls present, it is essential to use direct geometry INS spectrometers to check for the presence of O-H stretch modes. Their presence or absence allows a clear-cut distinction to be made between hydroxyls and surface hydrides. There are several examples in the literature where the analysis is predicated on the absence of O-H stretch modes in systems measured by TOSCA-like instruments. Figure 1 graphically demonstrates that this is not a valid assumption: "absence of evidence is not evidence of absence".