Catalytic reactions for H2 production on multimetallic surfaces: a review

Herewith, an overview is provided on the recent developments in the utilization of multimetallic catalysts to produce large amounts of molecular hydrogen, especially via the steam reforming of hydrocarbons and the water–gas shift reaction. Emphasis is given on the explanation of the problems affecting the currently used catalysts and how the addition/incorporation of other metals in available or new catalysts may lead to improved catalyst activity, selectivity and stability. We compare results from selected key examples taken from the literature where multimetallic catalysts are used for the aforementioned reactions. The methanol and ammonia decompositions are also critically analyzed, with focus on Earth-abundant metal elements.


Introduction
The potential of molecular hydrogen (H 2 ) as an energy carrier is expected to play a crucial role in future energy systems. Hydrogen can be obtained from renewable and nonrenewable sources, with different colors unofficially assigned by the North American Council for Freight Efficiency (NACFE [1]) to denote hydrogen origin. Renewable (green) hydrogen can be obtained, e.g. from water splitting into hydrogen and oxygen in an electrolyzer that is fed with renewable electricity from wind or solar farms, or from the steam reforming of inedible or waste biomass. Red/yellow hydrogen is produced by the electrolysis of water with nuclear power/grid electricity and turquoise hydrogen is produced from methane pyrolysis. Nevertheless, at present, the major fractions of the World's hydrogen arise from the steam methane reforming of natural gas or oil/naphta (78%, gray hydrogen) and from coal gasification (18%, brown hydrogen), figure 1.
As can be seen in figure 1, a successful combination of steam reforming of natural gas and water gas shift reactions will lead mostly to hydrogen gas and carbon dioxide, with the former being then used as a fuel and with the latter being (a) released to the atmosphere (eventually recycled through the natural plant respiration cycle to generate biomass), (b) used in alternative chemical processes (e.g. as an oxidant weaker than O 2 or C 1 -building block to obtain high-value chemicals), (c) sequestered by suitable underground geologic formations, or (d) repurposed for CO 2 enhanced oil recovery processes [2,3]. According to the NACFE coloring code, hydrogen from gray/brown sources but with CO 2 effectively repurposed/sequestered takes the blue color. The importance of the reforming of biomass is expected to increase significantly in forested areas since it may allow for nearly carbon neutral processes, which is of utmost importance to counter the global warming.
For hydrogen to become a viable fuel option, greater quantities must be produced cleanly and efficiently, and production costs must be lowered. The implementation of the infrastructure to distribute hydrogen is greatly dependent on a steady and cost-effective hydrogen supply and several different hydrogen carriers (e.g. ammonia, methanol, toluene/methylcyclohexane pair, etc [4]) have been considered for bulk storage and transport of hydrogen. The current annual global dedicated hydrogen production is about 70 megatons but it was recently estimated that this amount will be doubled in the next 5 years [5]. To ensure these Figure 1. Schematic view of the process of steam reforming of methane (SRM) to produce H2 and CO (on the left), with the latter being combined with steam to produce additional H2 through the water-gas shift reaction (WGSR, on the right). The WGSR is also important to convert CO from coal gasification (3 C + O2 + H2O → H2 + 3CO) to additional hydrogen. The conversion and selectivity are governed by the appropriateness of the catalysts used in each reaction. demands can be met, catalysts that can increase the activities of the reactions used for H 2 production must be found. Such catalysts must solve the problems found with conventional metal/metal oxide support catalysts, e.g. low conversion during startup/shutdown, high energy demands, and deactivation by sintering or by poisoning species like sulfur. For compact applications, pyrophoric problems inherent to some of the available catalysts also must be solved.
Recent strategies rely on the addition of other metals to existing or to newly designed metal-based catalysts to form metallic alloys, in which synergic interactions can solve at least some of the problems referred above that are associated to current commercial catalysts. However, this is not an easy task since the nature of the metallic elements, their exact proportion in the alloy, the nature of the support, the size and shape of the metal or metal alloy particles, among many others, are all important variables that may significantly affect the catalyst activity and selectivity.
In this review, we seek to examine recent scientific advances on the application of multimetallic catalysts in the production of H 2 through the water gas shift reaction (WGSR) and the steam reforming of hydrocarbons which are the current workhorses in industry for producing large quantities of this gas. We briefly discuss why the CO 2 dry reforming of methane (DRM) is less interesting for producing H 2 than the steam reforming of methane (SRM), despite the former consuming CO 2 , thus contributing to effective carbon capture. Finally, we review the importance of multimetallic catalysts in the reactions of the decomposition of methanol and ammonia to produce H 2 in more compact applications. Methanol and ammonia [6], which may be produced during periods of excess sustainable energy, are easily liquefied and, consequently, can be stored and handled more easily than liquid H 2 . Given the vast literature on the field, we could not address all the works published in the literature devoted entirely to each of the individual reactions considered and we refer the reader to other excellent reviews of steam reforming of hydrocarbons [5,7], WGSR [8], and methanol [9,10] or ammonia decompositions [11].

Steam reforming of hydrocarbons 2.1.1. Introduction
Steam reforming of hydrocarbons is the standard production method of hydrogen at an industrial scale [12,13]. It follows the general reaction: yielding syngas as its reaction product. The reaction always produces more H 2 than CO, with light hydrocarbons giving higher H 2 yields. For this reason, and due to lower investment costs, production of hydrogen is preferentially performed with light hydrocarbons [12,13]. The reaction is usually realized at 700 • C-925 • C, in the presence of a catalyst [13]. A posterior stage can be employed, at somewhat lower temperatures, where the WGSR equation (2), is used. The WGSR further increases H 2 yield, which is highly positive if the purpose of the process is to obtain pure H 2 , and not syngas. This reaction is exothermic, which allows it to be promoted at lower temperatures than the steam reforming reaction. The steam reforming reaction proceeds entirely through the action of the catalyst; both the organic compound and water are decomposed on its surface, followed by product formation and desorption. Overwhelmingly, methane (in the form of natural gas) is the hydrocarbon of choice for the production of hydrogen through steam reforming [14,15]. Dry reforming is a potential alternative to steam reforming, for generating syngas from hydrocarbons, which uses carbon dioxide, instead of water, as a reforming reagent. Just like steam reforming, dry reforming yields syngas: Unlike steam reforming, dry reforming typically yields more CO than H 2 . It is thus more adequate for applications where the purpose is production of synthesis gas, and not pure H 2 . DRM yields syngas in a 1:1 ratio. This ratio can be adjusted at a posterior stage, if necessary, although this incurs a cost [16]. Heavier hydrocarbons carry the problem of increased coke formation [17], which is a common occurrence with natural gas and biogas feeds [12]. Methane is thus the preferential hydrocarbon used in dry reforming processes. With two carbon sources instead of one, the tendency for carbon deposition is much higher in DRM than in steam reforming [18]. Furthermore, unlike with steam reforming, a high CO 2 /CH 4 ratio does not help in this situation. In fact, steam may be used as an additive in dry reforming, so as to prevent coking [19]. Since the dry reforming of hydrocarbons is less appropriate for H 2 production than the steam reforming, this process will not be reviewed here. Yet, recent accounts about multimetallic catalyst design for the DRM can be found in the literature [20][21][22][23].

Monometallic catalysts
Methane steam reforming is usually conducted using a nickel catalyst supported on Al 2 O 3 or SiO 2 [12,14,15,24,25]. Even though nickel is known to not be the most active element for this reaction [26], it has the advantage of being a relatively cheap transition metal (TM), giving a good compromise between stability, activity and cost [25,27]. Nickel has also been shown to be excellent at decomposing carbonated species, while being immune to oxidation [25,26]. Nickel-based catalysts are, however, greatly affected by coking, which has prompted much research on new, coking-resistant catalysts. This can be done either by doping well-known nickel catalysts or by using completely different metals. Methane steam reforming proceeds by decomposition of methane and water on the surface of the metal, with CH 4 decomposing into CH 3 * , CH 2 * , CH * , C * and H * adatom compounds, and H 2 O into H * , OH * and O * , where * denotes a free site on the catalyst surface [25,[28][29][30]. These decomposition products can then recombine to form a plethora of adsorbed compounds, such as COH * , COOH * , CH 2 O * , etc. A diagram detailing this process for steam reforming (SRM), dry reforming (DRM) and partial oxidation (POM) of methane is shown in figure 2.  (111) surface. Black, red and green curves present results for models of the Ni(111) surface when bare, with one and with two Au atoms, respectively. Adapted from [35], with the permission of AIP Publishing.
The presence of C * on the catalyst promotes the formation of coke. The most common strategy employed to prevent coking is to use high H 2 O (steam)/CH 4 ratios [12,31], but this generates increased costs in post-reaction separations. Thus, a lot of research has been devoted to finding catalysts that easily oxidize C * into CO [25]. Most approaches seek to compensate nickel's affinity for carbon species with oxidant elements [24,25,30]. Alternatively, a catalyst where C * is not prone to form, but CO is formed by dehydrogenation of COH * [29], would also be useful.

Multimetallic catalysts
Alloying of Ni with another metal has been considered for increasing catalyst resistance. Most nickel-based multimetallics involve alloying nickel with a noble metal, such as gold, platinum, or silver. Related TMs like copper and cobalt are also common. For the most part, the work with these multimetallics focuses on coking inhibition, while retaining or increasing activity.
Noble metals have very well-established anti-coking properties, as both monometallic [32] and multimetallic catalysts. Nearly all the work developed in this area has focused on Ni-Au catalysts, due to their high activity and strong resistance to carbon deposition. We will focus most of the rest of the section discussing Ni-Au catalysts, how research on them has evolved, and what the mechanisms underlying their properties are. As will be shown, it is not entirely clear what lies at the root of the anti-coking properties of Ni-Au catalysts.
The study of Ni-Au bimetallic catalysts for CH 4 steam reforming was kickstarted by Nørskov et al [33][34][35][36][37][38][39][40]. In a series of pioneering studies, feasibility and attractive anti-coking properties were established. In 1993 [33], they observed that Au atoms deposited on a Ni (110) surface, at room temperature (at which the two metals are nominally immiscible), tended to merge onto the surface layer of Ni atoms, forming a surface alloy. A later study showed that this surface alloy is only stable up to 850 K-900 K, the temperature after which the Au atoms start diffusing into the bulk [34]. A density functional theory (DFT) study [35] showed that CH 4 decomposition on the Ni(111) surface had its dissociation barrier increased by 16 kJ mol −1 when next to an Au atom (figure 3), which would indicate that activity is greatly diminished by the previously identified surface alloying. The authors attribute this to an energetic lowering of the local density of d-states on the Ni atoms in contact with Au atoms. Another study [36] determined that the presence of Au atoms on a Ni (111) surface also influences the probability of 'sticking' upon impact of a CH 4 molecule with the surface. Sticking probability decreases by ∼95% with an Au coverage as low as 0.2 monolayer (ML). This is in agreement with the results of a previous study [34], where decreased interaction with the surface upon increase in Au coverage was observed not only for CH 4 , but also for CO and D 2 . At Au coverages above 0.5 ML, CO and D 2 were barely adsorbed by the surface.
In a 1998 letter [37], an explicit case is made for Ni-Au industrial catalysts for steam reforming. The authors build on their previous work with a DFT study, whereby it was observed that chemisorption of carbon, C * , was much more favorable in pure Ni(111) than in Ni-Au alloys, with energy differences of up to 190 kJ mol −1 . It was thus possible to predict that a Ni-Au catalyst should be much less prone to graphite formation than a pure Ni one. Experimental testing of this hypothesis was performed with a 16.5 weight % (wt%) Ni/MgAl 2 O 4 catalyst, doped with 0.3 wt% Au. Analysis with extended x-ray absorption fine structure spectroscopy (EXAFS) showed that the Au atoms formed a surface alloy with Ni. Testing with n-butane steam reforming confirmed that, unlike pure Ni catalysts, the Ni-Au catalyst did not suffer from deactivation due to graphite formation. The same authors were granted a patent the next year [38], where a Ni-Au catalyst with 0.01/30% Au/Ni weight ratio, prepared by sequential impregnation, was proposed for the steam reforming of hydrocarbons 'without any carbon formation' . Once again, EXAFS showed that Au is deposited on the surface of Ni nanoparticles inside the catalyst.
In 2001, the same group [39] applied the sequential impregnation technique to create Ni-Au bimetallic catalysts, using SiO 2 and MgAl 2 O 4 as supports. Transmission electron microscopy (TEM) and energy dispersive spectrometry (EDS) allowed the observation of bimetallic nanoparticles in both supports. However, with SiO 2 , pure Au nanoparticles are also formed, separate from the bimetallic nanoparticles. Comparison of calculated x-ray absorption fine structure (XAFS) spectra and experimental EXAFS spectra for Ni-Au/SiO 2 samples shows that only ∼1/4 of the Au atoms participate in a surface alloy, with the rest forming separate pure Au phases. Monte Carlo simulations indicate that Au atoms accumulate on the edge and kink sites between low-index surface planes of the Ni nanoparticles. Thermogravimetric analysis (TGA) of MgAl 2 O 4 -supported catalysts used in the steam reforming of n-butane revealed that, under conditions where heavy carbon deposition occurs for the Ni catalysts, Ni-Au catalysts do not show carbon deposits.
In a highly influential theoretical study from 2002, Nørskov et al [40] used DFT and Monte Carlo simulations to show that carbon deposition and graphite formation tend to occur at step edges of Ni nanoparticles, and not at flat faces. They also determined that atoms from additives such as gold are more stable at precisely the same step sites, confirming the results from Monte Carlo simulations, from this and from their previous study [39]. The main function of these additives is thus blocking access to the highly reactive step sites, not in promoting oxidation of deposited carbon. This effect replicates that of H 2 S, which poisons Ni catalysts in a way that prevents carbon deposition, and may be used as an additive in steam reforming processes [41,42]. However, this also impacts catalyst activity, as step sites are also the most reactive towards CO and H 2 formation. A posterior study [43] also showed that Ni nanoparticle steps are much more active than terraces for CH 4 sticking and decomposition. Ni-Au catalysts would thus offer a compromise between improved resilience to carbon deposition and decreased activity for CH 4 decomposition. In the 2002 study [40], the authors reach the conclusion that an incomplete additive coverage would prevent graphite formation, while keeping the catalyst's activity. It is also observed that the Au atoms tend to be evenly spread on the surface of the Ni nanoparticles, not forming clusters. Similarly to what happened in their previous study [39], an EXAFS analysis of a Ni-Au/MgAl 2 O 4 catalyst showed that Au nanoparticles are formed apart from the Ni nanoparticles. This time, 44% of Au atoms are involved in the surface alloy with Ni. These proportions seem to be related to the relative amounts of Au and Ni used during catalyst synthesis.
In 2006, in one of the first studies on Ni-Au catalyst performed by other groups, Chin et al [44] performed an extensive characterization of nanoparticle surface in an Au-doped Ni/MgAl 2 O 4 catalyst. Through H 2 volumetric chemisorption, the authors were able to confirm the strong reducing effect of Au atoms on the surface of Ni nanoparticles. The presence of alloyed Au leads to decreasing chemisorption of H 2 by several times the amount of Au atoms, thus showing that their effect goes beyond site blockage. Temperature-programmed desorption (TPD) experiments with N 2 O confirmed this, by showing that the presence of Au on the surface prevented the decomposition of N 2 O and the formation of NiO. EXAFS tests confirmed that alloyed Au atoms have a very low coordination number, in agreement with the hypothesis that they tend to stay at the surface, away from the bulk. This work agrees with the results of previous experimental and theoretical studies performed by the Nørskov group [33,37,40].
Triantafyllopoulos and Neophytides [45] synthesized a Ni-Au/YSZ catalyst, with a Ni:Au ratio of 50:1, which was used to study CH 4 dissociative adsorption. The authors observed a strong inhibition of graphitic carbon formation on the surface of the catalyst, together with an increase in the activation energy for CH 4 dehydrogenation/CH x rehydrogenation, making CH x species more stable and easier to oxidize to CH x O. Decreased formation of graphitic carbon is thus attributed to a decrease in C * formation, non-inhibition of graphite formation or increased C * oxidation.
A study by Keane et al [46], used the reductive deposition technique to generate an Au-doped Ni/Al 2 O 3 catalyst. Energy dispersive x-ray analysis of metal nanoparticles showed that, in a catalyst with an 8.2 Ni/Au atomic ratio, surface composition was close to 1:1. This supports the observations of Nørskov et al, regarding the formation of surface alloys of the nominally immiscible nickel and gold.
Using Au-doped Ni/CeO 2 (Gd 2 O 3 ) as anode material in a solid oxide fuel cell, Niakolas et al [47] showed conclusively that the addition of Au inhibits coke formation. When performing a TGA of materials with different levels of Au, the results clearly show that the presence of Au greatly diminishes the rate of carbon deposition, and increased Au content enhances this effect (figure 4). The Au was added by the deposition-precipitation method, with the presence of Au nanoparticles with tens of nanometers in diameter being confirmed by scanning electron microscopy (SEM). The existence of metallic Au in the material was confirmed by x-ray diffraction (XRD) analysis. However, it is interesting to notice that the XRD results show that metallic Au, while certainly present before the reduction step of the synthesis procedure, seems to disappear afterwards. SEM analysis also fails to find the gold nanoparticles after the reduction step, performed with 80% H 2 /Ar at 800 • C, indicating that some major structural shift occurs in the material during the reduction step, making it unclear exactly what is behind the anti-coking properties of the catalyst.
Lazar et al [48] tested the activity of a Ni/γ-Al 2 O 3 catalyst, synthesized with the co-impregnation method, by measuring CH 4 conversion in non-repeatable single runs. The authors observed a marked effect in the catalyst's performance upon the addition of noble metals. A 1%-Au/8%-Ni catalyst led to an increased activity from 40% to 60% CH 4 conversion, at 450 • C. At higher temperatures, the improvement in performance starts to disappear, and above 600 • C it shifts, with the Au-doped catalyst performing slightly worse than the original. The authors speculate that this effect may be due to alloying of the two metals, originally existing in separate nanoparticles. This hypothesis is in agreement with previous work [39,40] suggesting Au atoms would tend to occupy the more reactive sites on the surface of Ni nanoparticles. The original catalyst was also doped with Ag, with very different results from Au doping; the Ag/Ni catalysts were almost inert, and only at 700 • C did their activity start to approach that of the other two catalysts. The authors also tested catalyst stability, by testing performance after 48 h of continuous run, at 500 • C and 700 • C. At 500 • C, Au doping strongly stabilized catalyst performance, while the original catalyst started showing signs of degradation. At 700 • C, the opposite happened, with the original catalyst having nearly identical performance, while the Au-doped one showed a slight decrease. If Au nanoparticles do exist at 500 • C, but tend to disappear at 700 • C, these results seem to indicate that coke inhibition stems from the activity of segregated nanoparticles, and not deposition on the surface of the Ni nanoparticles, with the latter actually being detrimental.
A follow-up study by the same group provided similar results [49], with some notable additional information. Exhaust gases were analyzed with a gas chromatograph, allowing them to determine the proportion of CO 2 and CO. Ni-Au catalysts strongly favor CO 2 formation, so they do not just have higher CH 4 conversion, but also higher H 2 production, through the WGSR. The authors were also able to determine, through TGA of used catalysts, that Ni-Au and Ni-Ag catalysts are not only far less susceptible to carbon deposition, but the kind of carbon that is deposited is different. Standard deactivation from carbon deposition involves the formation of graphitic crystallites, which the presence of Au and Ag hinders, with the small amount of carbon deposited being amorphous. However, these catalysts were still subject to degradation, at high temperatures, which the authors attributed to site blockage on the surface of Ni nanoparticles, induced by the formation of a surface alloy from initially segregated nanoparticles. The idea that Ni-Au alloys are only formed in solid-supported catalysts at high temperatures is reinforced by a characterization study conducted by Maniecki et al [50], where x-ray photoelectron spectroscopy (XPS) and XRD spectra indicate that Au and Ni co-exist as an alloy on the surface of the catalyst only after it is exposed to high temperatures.
In their review of 2013, Wu et al [18], taking a lead from the work of Maniecki [50] and Lazar [48,49], conceived of a mechanism for the formation of Ni-Au alloys in bimetallic catalysts prepared by the co-impregnation method, whereby effective reduction of NiO and temperature are the decisive factors determining the extent and kind of alloying that takes place, where, originally, the two metals were segregated in separated nanoparticles. This view is well represented in the scheme in figure 5. This alloying effect, induced by reduction and high temperatures, would neatly explain the 'disappearance' of gold nanoparticles in the work of Niakolas et al [47], occurring precisely after a reduction step at high temperatures. Interestingly, the suggested cut-off temperature for the formation of the alloy, 1153 K, is remarkably close to the critical point of the miscibility gap of 1089.5 K reported by Wang et al [51]. However, this temperature is also similar to those used in post impregnation calcination and reduction [39,[44][45][46]48], so studies suggesting both the surface alloy and the segregated Au nanoparticles perspective employ the kind of temperatures Wu et al suggest give rise to the alloying effect.
Focusing again on the supposed activity of step sites, Arevalo et al [29] investigated the reasons behind the strong binding of C * on Ni nanoparticle step sites, by using dispersion-corrected DFT-based first principles calculations. Comparing the interaction energy of the several decomposition fragments of CH 4 on Ni and Ru steps, the authors determined that C * + 4H * is the most favored species for Ni, but not for Ru, which rather favors CH * + 3H * . Their explanation for this is that C * at Ni steps can interact with five Ni atoms, four at the surface and one beneath the surface, while with Ru it interacts only with four surface atoms. In the case of Ni, when the subsurface atom is replaced for an Au atom, the C * interaction is greatly decreased, and CH * + 3H * is once again the most stable state (figure 6). However, it is unclear how this could explain the anti-coking effect of Au, as the authors do not present any mechanism for the formation of the 4Ni + 1Au configuration on the step sites of Ni nanoparticles. Indeed, many other authors [33,37,40,44] have shown that the Ni-Au alloy formed tends to be concentrated on the surface, with only high temperatures allowing Au diffusion into the bulk [44].
Following the Ni nanoparticle modification perspective, de Oliveira Rocha et al [52] performed an in-depth study of surface modification of a Ni/Al 2 O 3 catalyst through sequential impregnation with Au. By using diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) of adsorbed CO, the authors were able to study the changes that Au impregnation induced on the surface of Ni nanoparticles in the catalyst. Among many subtle changes, it was possible to ascertain the preferential alloying of Au atoms on Ni atoms with low coordination (i.e. steps, corners, and surface irregularities). However, there would still occur some Au deposition on Ni(111) atoms, which increases local electron density, thus promoting a preferential adsorption of CO in the linear form (with faster kinetics), instead of bridge form. Testing of the catalyst in a fixed bed reactor showed that the apparent activation energy of the steam reforming reaction was greatly affected by Au impregnation, which is attributed to a decrease in available low coordination Ni atoms. The authors thus point to two factors underlying the decreased carbon deposition in Au-doped Ni catalysts; deposition in low coordination sites hinders CH 4 decomposition, and deposition in high coordination sites promotes faster CO desorption. At low temperatures, CO formation is typically the rate determining step (rds) [26], so Au deposition has the effect of 'the steps of methane activation and carbon oxidation becoming equilibrated.' While it is entirely certain that Au atoms are sometimes deposited on top of Ni nanoparticles forming a surface alloy [39,40,44,52], and that on other occasions segregated Au nanoparticles have been detected [47][48][49], it is frequently impossible to ascertain whether both forms co-exist in the same material. The study by Molenbroek et al [39] is perhaps the only one that, by employing TEM/EDS, comes close to excluding the formation of independent nanoparticles. Still, it seems to us that the mechanisms and conditions behind the formation of segregated Au and Ni nanoparticles, a surface alloy, or both, has yet to be unveiled. Furthermore, the way segregated Au nanoparticles are able to condition the formation of graphitic carbon remains a mystery. It is possible that the studies where formation of segregated nanoparticles was detected failed to detect the additional formation of Ni nanoparticles with a Ni-Au surface alloy, which would function in the way Nørskov et al have discovered. Without further studies on this issue, we believe this should be the default hypothesis.
Xu et al [53] developed a computational approach to screen the potential of a large number of TM alloys for the reaction of SRM by separately performing microkinetic modeling of all the elementary reactions involved, employing energies of reaction intermediates and transition states from scaling relations using the binding energies of carbon and oxygen as descriptors. After having established a volcano-shaped relationship for the reaction as a function of the two descriptors above, they were able to predict the rates under operando conditions of binary alloys based on the A 3 B or AB (A and B are TMs) stoichiometries, with stability close to the stabilities of the bulk pure A and B metals and stability against oxidation. They found that A 3 B-like alloys derived from the cheap Ni, Fe and Co elements were very active and stable. Importantly, Ni 3 Fe and Co 3 Ni alloys displayed activity either with AA or AB terminations. Encouragingly, NiFe and NiCo alloys were experimentally found to display interesting activities for the reaction. However, it seems that the exact composition of the catalyst has a decisive role, and alloying, by itself, is not a solution. In fact, the effect of Fe on NiFe catalysts has been found to have both positive and negative influence in the catalytic activity in the reforming of hydrocarbons [54]. On the one hand, the activation and dissociation of C-H and C-C bonds occur mostly on surface Ni atoms. Thus, the alloying with Fe will lead to a decrease in the number of the most active sites. On the other hand, Fe is oxidized more easily than Ni and the iron oxide formation was associated to suppressing of carbon deposition and assistance in reactions between carbonaceous with oxide species. Therefore, while computational studies are certainly relevant for fast screening of materials, further experimental optimization of catalysts under operando conditions is needed.

Water-gas shift reaction 2.2.1. Introduction
The term water gas stands for an equimolar mixture of carbon monoxide and hydrogen that is produced by passing steam over heated coke, following the general reaction: In the 19th century, the reaction described by equation (4) was employed, for example, to enrich the gas generated for residential and commercial use in heating and lighting. More recently, as pointed out above, the production of steam gas is based on the SRM: providing a high molar ratio of H 2 over CO. In both cases, the aim is to obtain hydrogen-rich streams for numerous applications, namely, as a reactant in fuel cells or in the ammonia synthesis (see below). The WGSR, equation (2), discovered by Felice Fontana in 1780 [55], is a reversible chemical reaction between carbon monoxide and steam to produce carbon dioxide and hydrogen. The WGSR has been carried out in industry in two steps using two different temperatures. This is because the reaction is exothermic, hence favoring formation of products at low temperatures, but its kinetics is very slow. Hence, a step at high temperature (high-temperature shift, HTS) is employed to reach the equilibrium faster, and then a step at a lower temperature (low-temperature shift, LTS) is used to achieve almost full conversion of CO. The HTS uses a catalyst based on iron oxide (chromium oxide has been also employed for catalyst stabilization and copper or other supported metals may be present) that works at temperatures well above 300 • C (the catalyst is inactive below this temperature). The reaction mixture from the HTS is then cooled to ∼200 • C in the LTS reactor, which employs catalysts based on copper (conventionally, Cu/ZnO/Al 2 O 3 ) which become inactive above 250 • C. In the HTS step, the content of CO in the feed stream is usually reduced to 2% and in the LTS step it is reduced to <0.5%. In fact, from the equilibrium constant as a function of the temperature [55], It is clear that the equilibrium CO content is 20 times lower at 200 • C than at 400 • C. Because the total number of moles is the same in the reactants and in the products, pressure cannot be employed to force the conversion towards the desired products but is often employed to increase the reaction rate, hence allowing the equilibrium to be attained faster. Cryogenic, membrane, separation with solvents or sorbents are common strategies used for separating CO 2 from gas mixtures. Due to its high energy efficiency and because CO 2 interacts much stronger with most adsorbents than H 2 , pressure swing adsorption is the method of choice for separating carbon dioxide from hydrogen, leading to hydrogen purities over 99.9%. While the catalysts above have been used for several decades, they suffer from several shortcomings. For example, although catalysts based on Fe/Cr oxides are stable against sulfur poisoning, that is only the case if the amounts of sulfur are small. Also, if metallic iron species are present in the Fe/Cr catalyst, methanation of carbon monoxide via: takes place, leading to hydrogen consumption. The presence of Cr(VI) in the HTS catalysts is also a problem because this ion is carcinogenic to humans. Similarly, Cu-based catalysts are sulfur-sensitive and Cu crystallites are prone to thermal sintering. Thus, strong effort is being made to find catalysts with improved characteristics for the WGSR. Apart from being active for the WGSR, the required catalyst must have thermal and air stability for long term activity, and has to be cheap. Therefore, obvious choices are catalysts mainly constituted by cheap metals or metal oxides incorporating highly dispersed/diluted (expensive) metals. Potential elements for catalytic applications are the abundant Cu, Fe, Zn, Al or Cr, which are already found in the industrial WGSR catalysts, and Co or Ni. While metallic copper is the most active pure metal catalyst [56], it might be possible to obtain improved catalysts by mixing elements (or more complex materials) that display distinct characteristics, i.e. stronger or weaker adsorptions of WGSR reactants, intermediates and products than those found for metallic copper, until reaching an optimal balance, i.e. the top of the volcano. It is generally accepted that the WGSR proceeds through one of the following reaction routes, viz. the (a) redox, (b) formate or (c) carboxyl mechanisms (figure 7). The redox pathway is characterized by the complete dehydrogenation of water on the catalyst surface, while the associative formate and carboxylic routes are named so because they lead to the formation of these intermediate species upon the reaction of carbon monoxide with surface hydroxyl species.
Importantly, the first step of the WGSR, i.e. the dissociation of water into H and OH adsorbed species (figure 7), was found to be the rate determining one on Cu and on Au surfaces or nanoparticles [58][59][60]. It was predicted from DFT calculations for the WGSR on planar and stepped copper surfaces that the associative mechanism through the carboxyl intermediate, where carboxyl dehydrogenation is assisted by co-adsorbed hydroxyl species, is energetically more favorable.

Monometallic catalysts
The role of the different components of the industrial WGSR catalyst is not clear but the availability of metallic Cu species seems to be important, both in the LTS and in the HTS catalysts. For example, Yahiro et al [61] used the conventional impregnation method to prepare several catalysts based on Cu-ZnO supported on oxides with high surface area, namely, Al 2 O 3 , MgO, SiO 2 -Al 2 O 3 , SiO 2 -MgO, β-zeolite and CeO 2 , and found that the reducibility of CuO is an important factor controlling the activity of the catalyst with the best activities found when JRC-ALO-8 alumina was employed as the support, followed by JRC-ALO-5 alumina and CeO 2 . Interestingly, the addition of CuO to the HTS Fe 3 O 4 /Cr 2 O 3 catalyst promotes the reaction but with contradictory views. Rhodes and Hutchings [62] proposed that metallic Cu is not present and that the catalytic improvements are caused by Cu 2+ species acting in solid solution, which modify the electronic properties of the Fe 3 O 4 /Cr 2 O 3 catalyst and lead to a significant decrease in the activation energy from 118 to 75-80 kJ mol −1 . However, Puig-Molina et al [63] performed in-situ XAFS studies of a Cu-promoted Cr-stabilized iron oxide catalyst at 380 • C and elevated pressures and concluded that metallic Cu forms at 250 • C, promoting the WGSR activity that is due to the magnetite surface incorporating Cr 3+ species. The formation of metallic Cu species at temperatures above 250 • C was confirmed by Estrella et al [64] from in-situ time-resolved XRD, XAFS, and atomic pair distribution function analysis of the behavior of CuFe 2 O 4 and Cu/Fe 2 O 3 catalysts under WGSR conditions. More recently, Zhang et al [65] carried out a comprehensive experimental and computational study with the aim of identifying the WGSR active sites on copper-based catalysts. These authors developed an experimental strategy to synthesize uniform cubes, octahedra and rhombic dodecahedra Cu nanocrystals, with well-defined {100}, {111}, and {110} surfaces and size distributions of 1000 ± 150, 1056 ± 207, and 595 ± 113 nm, respectively, from the corresponding Cu 2 O Figure 8. Arrhenius plots of the WGSR rates for 2.7% Pt/Al2O3, 5% Pt/CeO2, 2% Pt/TiO2, and 4% Pt/Mo2C catalysts. Adapted with permission from [67]. Copyright (2011) American Chemical Society. nanocrystals. They then found the cube/octahedron to be the most/least active which was attributed to a higher {100}/{111} facets ratio in the former than in the latter. Moreover, the experimental findings suggest that all the elementary surface reactions occur at the Cu-Cu suboxide interface of the Cu(100) surface. Separate DFT calculations employing Cu(100) and Cu (111) surfaces, bare or with Cu 2 O islands to model the Cu/Cu oxide interface, confirmed that the reactions were in general more facile on the Cu(100) model with Cu 2 O islands, with reactants, intermediates and products of the WGSR adsorbing on the metallic atoms in the vicinity of Cu 2 O [65]. The oxygens of the oxide assume an important role in interacting with and then adsorbing the hydrogen atom originating from the water dissociation, the rds on metal surfaces [58][59][60]. In fact, when the reaction takes place at the Cu/Cu oxide interface, the energy barrier is 0.4 eV, significantly smaller than the value (1.2 eV) calculated on Cu(100) and on several other surfaces [66].
Schweitzer et al [67] compared the role of different supports, namely, non-reducible Al 2 O 3 , reducible CeO 2 and TiO 2 , and Mo 2 C, in the catalysis of the WGSR by platinum. Notably, as can be seen in figure 8, the Pt/Al 2 O 3 catalyst is clearly less active than those using cerium or titanium oxides supports, and the three oxides are less interesting than the molybdenum carbide support.
The increased activity of platinum catalysts when supported on reducible oxides than on non-reducible ones was observed in other studies. For example, Phatak et al [68] compared the WGSR on Pt catalysts supported on ceria and alumina, at p = 1 atm, T = 180 • C-345 • C, and found that the turnover frequency (TOF) was 30 times higher on ceria-supported Pt than on the corresponding alumina-supported catalysts. Results in the same direction were obtained by Panagiotopoulou and Kondarides [69] upon comparison of several reducible, namely, TiO 2 , CeO 2 , La 2 O 3 , and YSZ, and non-reducible, namely, Al 2 O 3 , MgO, and SiO 2 , metal oxides. In particular, the latter authors found that the TOF of CO conversion did not depend on metal loading, dispersion or crystallite size, but depended strongly on the nature of the metal oxide carrier, with the catalytic activity being 1-2 orders of magnitude higher when the active metal (e.g. Pt or Ru) is supported on reducible oxides [69]. In brief, the easier extraction of oxygen from reducible oxides increases the activity of the reaction since this facilitates the oxidation of CO molecules adsorbed on the supported metal. The support is then oxidized by oxygen originating from water. Therefore, in such situations, not only the supported metal but also the partially reduced metal from the oxide actively participated in the reaction, thus resembling a multimetallic catalyst as those reviewed below.

Multimetallic catalysts based on copper
As stated in the previous sub-section, subtle changes in the nature of the different species in the catalyst may lead to important improvements in the catalysis of the WGSR. Therefore, significant efforts were made to analyze the influence of a second metal in the catalysis of this reaction.
WGSR catalysts based on monometallic (Ni and Cu) and bimetallic (Ni-Cu, 1:1 alloy confirmed by x-ray absorption spectroscopy, XAS) particles anchored on SiO 2 were synthesized via an in-situ self-assembly core-shell precursor route by Ang et al [70]. The CO conversion measured at 400 • C for catalysts with a total of 10% metal loading, prepared with a small amount of oleic acid (OA) to aid metal dispersion, was 97% in the case of the bimetallic 5Ni5Cu/SiO 2 catalyst, 92% in the case of 10Ni/SiO 2 and 84% in the case of 10Cu/SiO 2 , but at lower temperatures, e.g. 250 • C, only monometallic copper catalysts could convert CO. Lower conversions were obtained for catalysts prepared without the addition of the dispersant, suggesting that smaller particle sizes enhance WGSR activity. Moreover, as can be seen in figure 9 for the bimetallic 5Ni5Cu/SiO 2 catalyst, while the CO conversion of the sample prepared with OA remained stable for a runtime of 75 h, the CO conversion is halved in the case of the catalyst prepared without the dispersant, which suggests less stable metal-support interactions with subsequent metal sintering in the latter.
Catalysts prepared with/without the dispersant displayed H 2 yields of 50%/45% but the monometallic nickel catalysts were not selective and led to formation of small amounts of methane. The observations were then rationalized by Ang et al upon analysis of in-situ DRIFTS results collected at different temperatures and CO-temperature-programmed-reduction-mass spectroscopy data. In the catalysts containing Cu, CO adsorption occurs at Cu sites which are known to be less active than Ni sites for CO cleavage, hence avoiding the formation of carbon precursors that are eventually hydrogenated to methane. Also, Ni segregation is suppressed in the bimetallic catalysts, otherwise, methane would form as in the 10Ni/SiO 2 . The formation of smaller metal clusters when OA is employed in the catalyst preparation enhances metal-support interactions, hence increasing the catalyst stability. Also, XAS results suggest that NiO species occurring at the bimetallic particle-support interface remain unreduced even at 450 • C, conferring increased stability because of the formation of stronger particle-support interactions. Concomitantly, the CO adsorption is stronger than in the case of large particles and the CO surface species become closer to hydroxyl groups on the SiO 2 support, clearly suggesting a synergic behavior in the multimetallic catalyst.
Qualitatively similar results were found by Saw et al [71] for Ni/Cu catalysts supported on ceria, upon variation of the molar ratio of copper and nickel from 0 to 1 with 10 wt% loadings of their metal nitrates; the catalysts obtained were denoted 10Ni, 9Ni1Cu, 5Ni5Cu, 1Ni9Cu and 10Cu. The 5Ni5Cu/CeO 2 catalyst was found to exhibit the highest reaction rate with the least methane formation, which was attributed to the formation of a Ni-Cu alloy phase and the stabilization of the CO adsorbed species. Kinetic studies revealed that one-site carboxyl mechanism could be the main reaction pathway with formate acting as spectator. However, other possibilities, e.g. combination of both carboxyl/redox and formate/redox routes could not be ruled out.
Arbeláez et al [72] investigated the WGSR over structured Cu, Ni, and bimetallic Cu-Ni supported on activated carbon catalysts. Two bimetallic catalysts with fixed Cu:Ni molar ratios of 2:1 and 1:2 were prepared. Again, the results pointed out to the formation of a Cu/Ni alloy with Cu having the important role of mitigating the methanation activity of Ni by controlling the CO adsorption behavior and dissociation ability as suggested by Saw et al [71], consequently favoring the shift process. Interestingly, CO conversion was found to be the highest in the case of the Cu-rich bimetallic catalyst with activity comparable to those demonstrated by catalysts based on noble metals deposited on ceria. The higher activity of Cu-rich bimetallic catalysts towards the WGSR was also reported by Jha et al [73] who studied mesoporous NiCu/CeO 2 catalysts with weight percentage ratios of Ni/Cu from (1:1) to (1:4). Importantly, while metallic phases of Cu and Ni were observed in the XRD patterns of Cu/CeO 2 and Ni/CeO 2 , respectively, after being tested in the WGSR at temperatures up to 550 • C, in the case of the reduced NiCu(1:4)/CeO 2 catalyst only the metallic phase of copper was found, which was interpreted as a result of Cu segregation to the surface layer, a point of view that is supported by the results of Lin et al [74].
More recently, Pastor-Pérez et al [75] revealed that bimetallic NiCu formulations are not always advantageous for the WGSR. In fact, for catalysts prepared on a CeO 2 /Al 2 O 3 support, the monometallic Cu catalyst was more active in both model and post-reforming WGSR mixtures at low-medium temperatures than the bimetallic CuNi systems. Only for medium-high temperatures does the bimetallic CuNi(1:1) outperform the stability levels reached with the monometallic formulation, becoming an interesting choice even when start-up/shutdowns operations are considered during the catalytic experiments. These observations agree with those of Ang et al [70] presented above but using a SiO 2 substrate. Computational studies were also performed to obtain a detailed structural and energetical understanding of the WGSR. Liu et al [76] performed DFT calculations, with the Perdew-Burke-Ernzerhof (PBE) functional and Gaussian-type orbitals, for the WGSR occurring on TM 1 @Cu 12 particles, with TM = Co, Rh, Ir, Ni, Pd, Pt, Cu, Ag, and Au, i.e. with the TM atom at the center of the particle, surrounded by 12 copper atoms. They found that the reactions proceed through the carboxyl mechanism, while the redox and formate routes arise as minor channels in the WGSR on the Cu-based particles, a finding that agrees with calculated data on extended pure-Cu surfaces [59]. On the TM 1 @Cu 12 particles, the formation of carboxyl species (COOH) from co-adsorbed CO and OH was found to be the rds when TM = Ag, while the dissociation of the carboxyl species into co-adsorbed CO 2 and H species is the rds for TM = Co, Ni, Rh, Pd, Ir, and Au, and the formation of co-adsorbed CO 2 and H 2 from CO 2 and H species is the rds for TM = Cu and Pt. These authors combined the energetic span model of Kozuch and Shaik [77] with the calculated free energy profiles to find the turnover-determining transition state (TDTS) and the turnover-determining intermediate (TDI) which govern TOF and selectivity. Interestingly, upon the consideration of the energetic span model, the TDTS and the TDI are the same for all systems, respectively, the transition state corresponding to the formation of carboxyl species (COOH) from co-adsorbed CO and OH and the state for co-adsorbed CO, H and OH species. The highest TOF was calculated for the Co 1 @Cu 12 cluster, with the group 9 elements, Co, Rh, and Ir, having TOFs that are higher than those for TM from groups 10 or 11, hence, more promising catalyst candidates for the WGSR.
Luo et al [78] investigated the WGSR on PdCu/CeO 2 also by means of DFT calculations, employing the PBE approach with D3 dispersion corrections and plane-wave basis sets. The calculations considered an unsupported PdCu 3 (111) slab and a PdCu particle, with eight Pd and 12 Cu atoms per unit cell, that was deposited on a slab of CeO 2 . In the case of the unsupported PdCu 3 (111) slab, water was found to adsorb very weakly (E ads = −0.05 eV, the negative value meaning favorable adsorption) while the barrier for its dissociation was quite large (E act = 1.42 eV), which suggests that water dissociation is difficult, as confirmed by microkinetic simulations to be the rds. The water dissociation barrier on the PdCu/CeO 2 interface is significantly reduced to E act = 0.24 eV, suggesting a more active catalyst for dissociating water, but where the dissociation of the hydroxyl species is slow. The calculations also show that CO adsorption is stronger on Pd than on Cu, which suggests that Pd plays a key role in the WGSR by capturing CO molecules from the gas-phase.
Saqlain et al [79] studied the WGSR on metallic Cu(100) and bimetallic Cu-Au(100) surfaces with a DFT approach (Perdew-Wang 91 (PW91) functional) and plane-wave basis sets. The results of their calculations predicted that the bimetallic surface is superior to the Cu(100) surface for the catalysis of the WGSR, since the decomposition of water, which is difficult on the metallic surface, becomes energetically more favorable on the bimetallic surface. The charge transfer from the exposed Cu layer to the underneath Au layer was suggested to be the main reason for the enhanced activity, with CO oxidation following the redox path on both the surfaces. On the bimetallic surface, the rds was linked to the O-H disproportionation reaction leading to O and H 2 O.
The computational studies above clearly suggest that active WGSR catalysts must be able to easily dissociate the water molecule. Therefore, several researchers focused their attention on the reaction of water dissociation on targeted catalyst models instead of calculating the full WGSR mechanism. Ghosh et al [80] used the PBE functional and plane-wave basis sets to study the water adsorption and dissociation on Cu/Ni bimetallic surfaces with varying concentration of the two metals. Overall, their calculations show that the presence of Ni in Cu(111) surfaces decreases the activation energy barriers for water dissociation, with values decreasing from 1.14 eV on Cu(111) to 0.61 eV in the case of a Ni monolayer on top of Cu (111), but the adsorption energies are almost constant with values between −0.12 and −0.18 eV. In absolute value, the activation energies are larger than the adsorption energies, suggesting that water molecules will desorb intact without dissociating upon temperature increase.
Bimetallic catalyst models based on Cu (110) and Cu(111) surfaces, with one outermost Cu atom replaced by Ni, Rh, Ir or Ag, were considered in another DFT study by Fajín et al [81] performed with the PW91 functional and plane-waves. The selection of the dopant elements was based on previous studies by the same authors [82] that suggested surfaces of Ni, Rh, Ir and Ag to be quite interesting for dissociating a water molecule. Upon comparison of the adsorption energies and of the activation energy barriers on the bimetallic surfaces with those calculated on parent metallic surfaces, these authors found a cooperative effect (i.e. more pronounced effect in the bimetallic surfaces than in the corresponding parent pure metallic surfaces) that stabilizes the reactants and products of the water dissociation reaction and leads to a decrease of the energy barriers associated to the cleavage of the O-H bond in the water molecule. Importantly, the absolute values of the adsorption energies for the water molecule become larger than the activation energy barriers on the bimetallic surfaces, suggesting that the water molecule will effectively dissociate into co-adsorbed H and OH species, with Rh@Cu(110) and Ir@Cu (110) being very interesting cases with adsorption energies/activation energy barriers of −0.51/0.31 eV and −0.55/0.21 eV, respectively. Fajín et al [83] extended their computational studies to trimetallic catalyst models and found that the replacement of two outermost atoms by one Al and one Zn atom in extended copper surfaces dramatically improves the activity of the copper surface towards the dissociation of the water molecule. Among the 25 trimetallic catalyst models considered in their work [83], the calculations predict a facile dissociation of the water molecule onto the (AlZn)@Cu(111) surface, which is aligned with the experimental results by Boumaza et al [84] where the potential of a Cu 0.5 Zn 0.5 Al 2 O 4 spinel oxide catalyst for the WGSR was demonstrated. The high activity of the ternary Cu/Zn/Al catalyst suggests that the active phase of the Cu/ZnO/Al 2 O 3 commercial catalyst might embody localized patches of a trimetallic alloy containing these metal atoms [85]. A spinel phase but with Fe instead of Al (Cu 0.15 ZnFe 2 ) was also found to be quite active for the WGSR [86]. In fact, the trimetallic catalyst was found to be more active than bimetallic Cu-Zn and Cu-Fe oxide catalysts which was suggested to be a consequence of the lower temperature required for reducing CuO in the spinel catalyst than in the bimetallic oxides [86], which is well aligned with the suggestions by Yahiro et al [61] discussed above. However, in the works by Khan et al [87] and Natesakhawat et al [88], where several different Fe 2.73 M 0.27 O 4 (M = Al, Ce, Co, Cr, Cu, Ga, Mn, Ni and Zn)-type spinels were studied, the complexity of the process became evident, since different aspects, e.g. improved covalency of the Fe III ↔ Fe II redox couple, sintering restriction, decrease of the temperature for Fe 2 O 3 → Fe 3 O 4 reduction, or mobility of lattice oxygen, may be differently influenced by the added metal.

Multimetallic catalysts based on platinum
Platinum particles dispersed over metal oxide supports such as Al 2 O 3 , CeO 2 , ZrO 2 , CeO 2 -ZrO 2 , or TiO 2 display interesting activity for the WGSR at low temperatures [68,[89][90][91][92][93][94][95][96][97]. Therefore, it is not surprising to find studies where the effect of the addition of a second metal in the catalytic activity towards the WGSR was analyzed. For example, Aragao et al [98] used the method of controlled surface reactions (CSRs) to incorporate Fe into Pt/SiO 2 with 5 wt% Pt loading, with Pt to Fe molar ratios between 1:0.05 and 1:0.2. The deposition of Fe was suggested to occur near uncoordinated atoms at the edges and corners of the Pt nanoparticles and/or activated hydroxyl groups at the interface with the support. Notably, no Fe deposition could be detected on the bare SiO 2 support, clearly suggesting that OH groups on the bare support were not having the same characteristics as those in Pt/SiO 2 . The catalytic activities of the bimetallic FePt/SiO 2 catalysts are compared with those of the monometallic catalysts, bare and physically mixed, in table 1. The influence of the support was also analyzed upon additional comparison with carbon supported Pt/C and Pt 1 Fe 0.2 /C catalysts. As can be seen, when Fe is added to the Pt-based catalysts, the CO uptake decreases but the CO 2 production rates increase, clearly showing the synergic influence of iron in the WGSR catalysis. Also, the comparison of results for similar catalysts but with silica or carbon supports clearly suggests the participation of the hydroxyl groups on the silica surface in the catalysis of the WGSR. Aragao et al suggested that in the case of the monometallic Pt/SiO 2 and Pt/C catalysts, the reaction follows the associative mechanism, i.e. a Langmuir-Hinshelwood type process where the adsorbed CO reacts with surface OH, generated by H 2 O activation, and progressing through carbonate/carboxyl or formate intermediates. In the case of the iron containing catalysts, the Fe promoting role was explained in terms of the redox properties of Fe, where coordinatively unsaturated ferrous sites confined near the Pt-SiO 2 interface decrease the energy barriers for O 2 activation and reaction with CO adsorbed on a nearby Pt site. Nevertheless, an associative mechanism in which the coordinatively unsaturated ferrous sites are highly active for H 2 O activation and dissociation could not be disregarded.
The CSR method was also used by Sener et al [99] to prepare MoPt bimetallic catalysts on carbon and silica supports. As in the case of the FePt bimetallic catalysts, the catalytic activities of silica-supported catalysts are higher than the ones displayed by the carbon-supported samples. Interestingly, they found that the deposition of Mo onto Pt/C reaches saturation for a composition of Mo:Pt = 0.32 and, beyond this point, the additional Mo species are deposited on the carbon support. Moreover, the TOFs on the PtMo/C catalysts were found to scale linearly with the estimates of surface Mo mole fraction, which suggests that surface Pt-Mo interfaces are the active sites for the WGSR.
Rhenium was found to enhance the activity and stability of Pt-based catalysts, even at very low temperatures [100][101][102]. The experimental study of Duke et al [103] aimed to understand the origins of the activity enhancement when compared with monometallic Pt-supported catalyst, by controlling the effects of Re oxidation and bimetallic composition, and confirming the absence of cluster sintering during reaction. Bimetallic catalysts having a large fraction of surface Pt atoms were found to be more active than bimetallic catalysts having a large fraction of surface Re atoms, which was suggested to be a consequence of the lower activity of Re towards the WGSR. However, the lower activity of Re does not explain why bimetallic catalysts with large fractions of surface Pt atoms are also more active than pure Pt catalysts. Duke et al [103] confirmed that only metallic Re is present in the case of the more active catalysts with large fractions of surface Pt atoms and that when ReO x species were present the catalytic activity was reduced. From infrared absorption reflection spectroscopy experiments on pure Pt and Pt-Re surfaces, they found that the CO coverage is greater on Pt. This information was used to propose that Re species avoid catalyst poisoning by CO, as found by others [104][105][106], thus enhancing the catalytic activity. DFT calculations also demonstrate that CO binds more strongly to Pt than to Pt-Re. In particular, the study by Detwiler et al [107] confirmed previous evidence that Re is inserted within the Pt layers, leading to the formation of a Pt skin instead of the formation of a ReO x layer. The presence of subsurface Re species lowers the d-band center of the exposed Pt atoms which results in weaker adsorbate binding.

Methanol decomposition 2.3.1. Introduction
Methanol (CH 3 OH) contains 13 wt% of hydrogen [108] and is a potential hydrogen carrier (from the capture of CO 2 and from water splitting H 2 ) as comprehensively documented by Frei et al [109]. Its decomposition for hydrogen production can be represented as follows, CH 3 OH → CO + 2H 2 (8) i.e. each mole of methanol decomposes into two moles of hydrogen. This reaction is endothermic, with ∆H = +91.7 kJ mol −1 (around 1 eV) at 298 K [12]. The resulting CO may be combined with steam to produce additional H 2 via the WGSR equation (2). One of the difficulties of using methanol to produce hydrogen to fuel the engine of a vehicle are the high temperatures required, a problem which affects especially Northern countries. In fact, while the methanol synthesis requires low temperatures and high pressures, its decomposition is performed at ambient pressure and higher temperatures [110].

Monometallic catalysts
The most studied catalysts for methanol decomposition are metallic: Cu, Ni, Pt and Pd [111][112][113][114][115][116][117][118]. In particular, Pt has been the subject of much scrutiny and controversy, namely due to the debate regarding the dissociation reaction pathway ( figure 10). Using high-resolution electron loss spectroscopy and thermal desorption spectroscopy, the clean Pt (111) surface was reported to allow methanol decomposition into CO and H 2 at 140 K, while the same surface covered by oxygen instead forms water and methoxy (CH 3 O), which then require a higher temperature to decompose into the final products (CO and H 2 ) [119]. This has been confirmed by several theoretical and experimental works [111-113, 120, 121]. Using different experimental techniques (secondary ion mass spectrometry and thermal programmed desorption), it has also been shown that methanol can decompose on Pt(111) by first breaking the C-O bond, forming a methyl group [122], a similar mechanism as the one reported at high temperatures on the (1 × 1)Pt(110) surface [123]. Regardless of the mechanism leading to the complete methanol decomposition, the greatest concern lies in the fact that CO adsorbs very strongly on the Pt surface, quickly reducing its catalytic capability [124][125][126][127]. This CO poisoning can be mitigated by mixing platinum with another metal, by surface coverage, alloying, doping, etc. The quest for a suitable combination served as motivation for most of the works on methanol decomposition on Pt, for at least the last decade [128][129][130].

Multimetallic catalysts
One of the most popular bimetallic catalysts for methanol decomposition is Au/Pt. One reason for this is that, on these bimetallic surfaces, the energy (and, consequently, the temperature) required for CO desorption reduces to less than a half of that on pure Pt. In fact, on Pt (111) and Pt (100), the CO desorption temperatures are 400 K and 525 K, corresponding to desorption energies of around 1.12 and 1.41 eV, respectively [132]. If, instead, one considers surfaces with the same Miller indexes, but with approximately one half of the surface Pt atoms substituted by Au atoms, the CO desorption temperatures are considerably lower (174 K and 237 K, respectively), and so are the desorption energies (0.46 and 0.62 eV, respectively) [133]. This result, predicted by DFT calculations and confirmed by TPD experiments, implies that Au/Pt bimetallic surfaces are not as easily poisoned by CO as pure Pt surfaces. The same study also showed that the Au/Pt surface is stable in the conditions required for CO desorption. Yuan et al [117] employed DFT calculations to study methanol decomposition on the Au(111) surface, with some Au surface atoms replaced by Pt in an hexagonal fashion, and predict an alternative reaction mechanism, which ultimately may not lead to the production of CO. By comparing two general decomposition reaction paths-(a) breaking the O-H bond first vs (b) breaking it last-they found that the former contains one sizeable barrier to overcome, while the latter contains three (see figure 11 and table 2). In the same study, it was also shown that the two paths are very unlikely to cross each other in one of the intermediate reactions. The highest energy barrier for reaction path (a) is of 0.98 eV and corresponds to the O-H bond scission, while the ensuing C-H scissions require less energy. From the data in table 2, we can conclude that on PtAu(111), after CH 3 OH becomes CH 3 O, the latter should easily decompose to CH 2 O, which is then harder to break. This contrasts with the clean Pt(111) surface where, after O-H scission, the decomposition reaction easily continues to form CO, due to the extremely lower energy barriers. The values in table 2 corresponding to methanol decomposition path (b) show that the alloying effect has quite a weak impact. Here it should be noted that the calculated adsorption energies of methanol on the Pt(111) and PtAu (111) surfaces are very low, 0.33 and 0.29 eV, respectively [112]. Both these values are lower than the energy barriers of the first reaction steps shown in table 2 and, therefore, methanol is more likely to desorb from the surface and lose its first H atom through some other mechanism, such as a collision, than when adsorbed. The reactions can then proceed on the catalytic surface because the absolute binding energies of both CH 3 O and CH 2 OH on either surface are all higher than 1.50 eV, favoring dehydrogenation rather than desorption, especially through path (a). One crucial difference between Pt(111) and PtAu(111) is the barrier for CH 2 O dehydrogenation, which is much higher in the bimetallic surface. This difference, added to the moderate adsorption energy of CH 2 O on PtAu(111) (0.54 eV), suggests that CH 2 O could be directly oxidized by atmospheric oxygen into formic acid, instead of forming CO.
Pure Pt and Pt/Au clusters have also been theoretically studied in terms of methanol dehydrogenation and CO desorption [134]. Qualitatively, the results coincide with those of previous calculations and experiments [117,129,132]; it was found that, although on Pt/Au the energy of the dissociation rate-limiting step is harder to overcome, the final CO desorption is much easier on Pt/Au. On the Pt 3 cluster, the dissociation was predicted to begin with a C-H bond scission with an energy barrier of 0.56 eV, while on PtAu 2 the reaction begins by breaking the O-H bond, with an energetic cost of 0.93 eV. The latter is compensated by a CO desorption energy of 0.19 eV, much lower than the 1.31 eV calculated for Pt 3 .
The alloying effect of Pt/Au on TiO 2 for methanol decomposition was finally studied in detail, both theoretically and experimentally, by Tenney et al [135]. There is indeed very strong synergy between all the Figure 11. Visual representation of the methanol dehydrogenation process along two pathways, each beginning (left) or ending (right) with the O-H bond breaking. Color code for blue and green pathways as in figure 10. From top to bottom, hydrogen bonds in the adsorbed methanol molecule are sequentially broken until only CO is left. The numbers represent energies of adsorption, migration, or energy barriers. Gray and yellow spheres represent atoms of platinum or gold, respectively. Labels in red indicate the positions of the carbon, oxygen, or hydrogen atoms above the surface. Adapted from [117], with the permission of AIP Publishing. Table 2. Energy barriers as calculated in [117], in eV, for methanol decomposition, following two paths: O-H bond breaking (a) as the first step vs (b) as the last step.

Reaction
Pt ( It was shown that Pt/Au clusters with more than 50% Au contain only Au atoms on the surface. As previous studies had found, the first bond to be broken in the process is O-H, a scission facilitated by the presence of TiO 2 . However, if only TiO 2 was present, the resulting methoxy (CH 3 O) would likely recombine with the hydrogen and desorb without reaction. The role of Au is to promote H 2 desorption at low temperature (<300 K) to prevent the reaction from stopping at this stage. The reaction on pure Au clusters yields CH 2 O rather than CO as its main product. Nevertheless, on Pt/Au, CH 3 O connects to the surface via its O atom, which induces the migration of Pt from under Au to the surface and allows the dehydrogenation to complete. The dehydrogenation activity of this cluster is very similar to that of pure Pt clusters. The advantage of the alloy is that the CO desorption temperature decreases with increasing Au fraction (in particular, it decreases by about 25 K at an Au fraction of 25%) [136]. Therefore, a delicate balance between the amount of Pt and Au needs to be found: if there is too much Au, the dehydrogenation will not be completed, but too much Pt yields too strong a binding energy for CO, poisoning the catalyst. Let us now turn our attention to bimetallic catalysts with Pt and metals other than Au. Guo et al studied the substitutional doping of Ni on Pt [131]. This doping was found to shift the dehydrogenation path from first breaking one C-H bond to begin with an O-H bond scission. Charge transfer from Ni to Pt was found to weaken the CO poisoning. Removing CO from the PtNi cluster was predicted to require 0.85 eV, while the same operation on Pt was calculated at between 0.89 and 1.51 eV. More recently, a theoretical study on the PtRu(111) surface over boron-doped graphene found that adding Ru to Pt weakens the CO adsorption, and that the graphene sheet improves the catalytic activity by lowering the methanol dehydrogenation energy barriers [137,138]. The CO adsorption energy on PtRu(111) on boron-graphene was calculated at −1.81 eV, which is considerably lower than on Ru(0001) (−2.30 eV [139]) or PtRu(111) (−2.10 eV [140]). Also, recently, Pt/Ir was demonstrated as an alternative catalyst for methanol decomposition with reduced CO poisoning with respect to pure Pt, with two advantages: weaker CO binding energy (by 0.26 eV) and more favorable CO oxidation by OH − .
It should be noted that catalysts involving noble metals, such as Ru, Ir, Pt, or Au, may not realistically be employed at a significant scale. However, they can be regarded as benchmarks, whose performance as catalysts may be achieved or surpassed by research on mono or multimetallic catalysts. Therefore, it is not surprising to see that other researchers devoted their attention to alloying of noble metals with abundant ones as Cu, Ni or Co. Upon a combination of DFT calculations and thermochemical scaling relationships, Mehmood et al [141] estimated the thermochemistry and kinetics of the methanol decomposition on pure metallic or bimetallic clusters with four M atoms (M = Ag, Au, Co, Cu, Rh, Pd and Pt). From a volcano plot of the methanol decomposition using the adsorption energies of O and C as descriptors, they found that the Pd 1 Cu 3 and Pd 2 Co 2 subnanometer metal clusters were close to the predicted maximum of the volcano, with predicted rates higher than those predicted for Rh 4 , Pd 4 or Pt 4 clusters. Further research is certainly needed to analyze whether it is possible to develop improved catalysts based on abundant metals for the methanol decomposition reaction.

Ammonia decomposition 2.4.1. Introduction
Around 150 million tons of ammonia (NH 3 ) are produced every year around the world [11]. This compound has long been known as an excellent choice for hydrogen production, storage and transport. Indeed, liquid ammonia contains 17.8 wt% hydrogen, amounting to 50% more hydrogen per volume than liquid hydrogen itself [142]. The fact that a pressure of around 10 atm is enough to liquefy ammonia at 298 K makes its liquid-phase storage relatively cheap, decreases the size and weight of containers, and allows for a simple way to store hydrogen at a higher volumetric density than liquid H 2 .
The global reaction of ammonia (NH 3 ) decomposition to form hydrogen (H 2 ), is endothermic, and the corresponding enthalpy of reaction at 298 K is 91.86 kJ mol −1 , or nearly 1 eV. High temperatures and low pressures facilitate this process, and in practice this reaction is normally performed at around 850 • C [12]. Alternatives include burning a fraction of the resulting hydrogen in order to help maintain the high temperature. The trade-off is the reduction of the reaction yield, as some of the hydrogen is replaced by the product(s) of combustion and water vapor, even if the combustion is done in a carbon-free atmosphere to prevent the formation of side products such as carbon monoxide or dioxide [12].

Monometallic catalysts
In reality, ammonia decomposition reaction occurs in steps, where NH 3 is successively dehydrogenated, forming N and H, which recombine as N 2 and H 2 , respectively [143], NH * + * → N * + H * Figure 12. Representation of the general mechanism of ammonia adsorption, decomposition, and product desorption on a catalyst surface. The blue and white spheres represent nitrogen and hydrogen atoms, respectively, while the remaining spheres are surface atoms. Adapted with permission from [144]. Copyright (2019) American Chemical Society.
These reactions represent, respectively, the adsorption of ammonia, the first, second and third dehydrogenations of adsorbed ammonia, and the recombination and desorption of two nitrogen or hydrogen adatoms (as shown in figure 12). A suitable catalyst must therefore be such that the binding of the nitrogen atom of NH x is strong enough for the dehydrogenation to occur, but weak enough to allow the resulting nitrogen adatoms to recombine as N 2 and desorb from the surface. This trade-off is crucial in finding the ideal catalyst for ammonia decomposition.
In this review, we focus on metallic catalysts (which, in fact, are currently the default choice) but, in the case of ammonia decomposition for hydrogen production, other choices, such as nitrides and carbides of Co, Cr, Fe, Mn, Ni, Ti and V, or alkali metal amides are available. The former have been found to show catalytic activity similar to that of metals [145,146], while the latter seem promising to promote ammonia decomposition at temperatures lower than 450 • C, contrasting with the temperatures above 600 • C required to achieve 100% conversion efficiency on metal catalysts. For further details on ammonia decomposition on non-metallic catalysts, please refer to the very recent reviews of Lamb et al [11] and Mukherjee et al [147].
Ammonia decomposition using metallic catalysts has been investigated for so long that the activity trend for different metals has been known since 1823, before the word 'catalysis' was applied to these processes [148,149]. Metallic catalysts for ammonia decomposition are so well established that some works do not even mention other types of catalysts. According to Häussinger et al [12], the typical catalysts are nickel-or iron-based. However, other metals present higher catalytic activity. The list of studied monometallic catalysts increased over time and, in 2004, Ganley et al experimentally probed the activity of 13 metals supported over alumina (Al 2 O 3 ) at T = 580 • C [150], measured as TOFs, and with results summarized in table 3. In all theoretical and experimental works, the consensus is that ruthenium yields the highest catalytic activity among all monometallic catalysts, as shown in figure 13. Other studies give slightly different orders, depending on the support, but Ru retains its merit [151,152]. Nevertheless, research did not end here, as it was found that the activity of Ru at lower temperatures decreases considerably. For instance, at 450 • C, the TOF of Ru supported on carbon nanotubes becomes less than 50% [153].
While in ammonia synthesis via the Haber-Bosch process, the rds is known to be the dissociation of the N 2 triple bond, the limiting step in ammonia decomposition depends on the metallic catalyst. On precious metals, such as Ru, Rh, Ir or Pd, the rds is the breaking of N-H bonds, while for nonprecious metals, like Ni, Co or Fe, the limiting step is N 2 desorption, one of the last steps of the process [155]. This is the reason behind the famous 'volcano plot' obtained in [154], which correlates the catalytic activity of several metals with the energy barrier for N-H bond scission. This plot shows that such a correlation exists separately for precious and nonprecious metals: the TOF increases with the N-H bond scission activation energy on the latter and decreases on the former [147,151,154]. The adsorption energy of nitrogen on each metal correlates as a volcano with both the TOF for ammonia synthesis and ammonia decomposition (with shifted maxima), and shows that Ru is the best catalyst for both processes among pure metals [151]. The volcano shape implies that there is an ideal value for nitrogen binding energy which leads to the highest ammonia synthesis activity. With this in mind, in 2005, Boisen et al [152] predicted that a combination of two metals  (one where the N adsorption energy is too high and another where it is too low), such as CoMo, should display an activity close to the optimal one (see figure 13). Their experiments confirmed that, at low NH 3 concentrations, CoMo displays a better catalytic activity for ammonia synthesis than Co or Mo individually, and better than Ru. This bimetallic mixture, supported on MCM-41 (Mobil composition of matter No. 41 type silica), was later also used for ammonia decomposition and, in this case, an ideal Co/Mo ratio of 7/3 was found [156]. This process of combining two elements hoping to obtain properties intermediate between the original ones was called 'periodic table interpolation' and suggested to be a general phenomenon in the field of catalysis, in reactions where the catalytic activity does not change monotonically with the interaction energy, such as the case of volcano-like relationships. In addition, the interpolation will only produce the desired effect if the resulting surface contains both chemical elements. If, as usually happens, one of the elements becomes too dominant on the surface, either due to the intrinsic properties of the mixture or due to interacting too strongly with the adsorbate, the interpolated properties might not be observed [157]. This ground-breaking work motivated the prediction of the ammonia decomposition activity of bimetallic catalysts, based on periodic table interpolation.

Multimetallic catalysts
Encouraged by the success of Boisen et al in producing a bimetallic catalyst for ammonia decomposition with better activity than Ru via periodic table interpolation of the volcano plot, scientists began studying other combinations of metals using both computational and experimental techniques. In 2010, Hansgen et al studied theoretically and experimentally several configurations of Ti, V, Cr, Mn, Fe, Co, and Ni [143]. The catalytic activity for ammonia decomposition was compared on surfaces containing Table 4. Calculated nitrogen binding energies of some TMs on platinum [143,158]. The second and third columns show the values for a single layer of the metal (M) on the subsurface and on the surface of platinum (Pt), respectively, in kcal/mol. The last column shows values for monometallic systems.

Metal
Pt On the one hand, all of the subsurface configurations showed nitrogen binding energies below 90 kcal mol −1 , so they were discarded as catalysts for ammonia decomposition. On the other hand, a single layer of some of the metals over Pt showed very promising results. While one layer of Ti, V, Cr or Mn on Pt gives too high a binding energy (above 175 kcal mol −1 ), the surfaces of Ni-Pt-Pt, Co-Pt-Pt and Fe-Pt-Pt have binding energies remarkably close to the ideal value of 134 kcal mol −1 . These authors synthesized the Ni/Pt bimetallic system by depositing Ni at room temperature or at 600 K, which led to the Ni-Pt-Pt or to the Pt-Ni-Pt configurations, respectively. These systems were then studied using TPD and high-resolution electron energy loss spectroscopy. The Ni-Pt-Pt surface was shown computationally (and later experimentally) to be more stable than the subsurface one (e.g. Pt-Ni-Pt). Furthermore, upon exposure to oxygen, the subsurface configuration rearranged itself into the surface one, both at low pressures (around 10 −10 bar) and at atmospheric pressure [159][160][161]. The experiments also confirmed other theoretical predictions: the Ni-Pt-Pt surface was found to be very active for ammonia decomposition, as both its nitrogen desorption temperature and ammonia dehydrogenation barrier (the two rate-determining steps in ammonia decomposition) are lower than those of Ru. This surface was predicted to display activity for steady-state decomposition below 600 K.
Similarly to Ni-Pt-Pt, the N binding energies on Co-Pt-Pt and Fe-Pt-Pt surfaces are very close to the desired value of 134 kcal mol −1 . In table 4, Cu/Pt is the only bimetallic combination for which both Pt-M-Pt and M-Pt-Pt configurations represent an improvement towards reaching the desired N adsorption value of 134 kcal mol −1 , when compared to the corresponding monometallic catalyst. In contrast, for Fe, Co and Ni, only the M-Pt-Pt configuration shows an advance towards 134 kcal mol −1 , while the Pt-M-Pt configuration displays a value farther from 134 kcal mol −1 than the monometallic catalyst. For this reason, Co-Pt-Pt, Fe-Pt-Pt, and both Cu multimetallic surfaces were analyzed by the same group in another work [158]. In that study, it was confirmed that the N binding energy correlates well with the catalytic activity of bimetallic systems, and it was found that Cu bimetallic surfaces are inadequate for ammonia decomposition. In contrast, the Co/Pt and Fe/Pt were predicted to be more active catalysts for ammonia decomposition than the corresponding monometallic materials. Unlike the case of Ni-Pt-Pt, the surface configurations of Cu/Pt, Co/Pt and Fe/Pt are not always the most stable ones when compared to the subsurface structures. In fact, the Pt-Cu-Pt stacking is the most stable one regardless of the surface nitrogen coverage, while, according to the calculations, the Co-Pt-Pt and Fe-Pt-Pt configurations are only the ground state for nitrogen coverages of over 0.35 ML. Since the subsurface configurations are ineffective at decomposing ammonia, the reaction needs to occur at nitrogen coverages above this value in order to stabilize the M-Pt-Pt configuration. As stated by the authors, the experiments were carried out under ultrahigh vacuum, where the nitrogen coverage saturation is around 0.3 ML, hampering the stabilization of the desired surface configuration. There is, however, hope in the form of material edges and defects, which were not taken into account in the calculations, as these likely affect the thermodynamic and kinetic relative stability of the surface and subsurface configurations [162,163]. The energy barriers for N-H bond scissions, despite not being the rate-determining step for ammonia decomposition on nonprecious metals, also play a relevant role in the process and as a matter of fact correlate approximately linearly with the corresponding TOF. The dehydrogenation energy barriers on selected surfaces [159], calculated using the method described in [164], are shown in table 5. These barriers are significantly lower than, for instance, the ones calculated for Pd-Cu, or for Pd-doped Cu(100) [165,166], but similar to the calculations of Novell-Leruth et al for Pt, Pd and Rh [167]. If one excludes the monometallic Cu surface, all linear correlation coefficients between the values in tables 4 and 5 are above 0.95 implying that, on surfaces with higher N adsorption energies, the NH 3 dehydrogenation energy barriers are lower. This also hints that, like the N binding energies, these barriers can be used to predict the TOF of ammonia decomposition on bimetallic surfaces.
All kinds of combinations of metals under different conditions are still being tested, as ammonia decomposition on multimetallic catalysts is an extremely hot topic. In the following paragraphs, we examine some works that were published very recently.
Huang et al compared the performance of Ni, Co and Ni/Co metallic catalysts supported on Ce 0.6 Zr 0.3 Y 0.1 O 2 (CZY) [168]. It was found that the TOF is maximized at a Ni/Co mass ratio of 1/9. The resulting TOF of around 0.67 s −1 is higher than the ones measured for Ni or Co on CZY (0.52 and 0.54 s −1 , respectively), but still much lower than the TOF on some monometallic catalysts (see table 3). This higher catalytic activity was attributed to an increase in surface area.
An experimental comparison between the catalytic activity of Ni, Ru and Ni/Ru supported on Al 2 O 3 and CeO 2 showed that the initial activity of Ru/CeO 2 is the highest among all the analyzed combinations but, as happens with all studied Ru-containing catalysts, it deactivates too early due to metal sintering in the case of Ru/Al 2 O 3 or Ru volatilization on Ru/CeO 2 [169]. Additionally, CeO 2 was demonstrated to be a better support for ammonia decomposition catalysts than Al 2 O 3 . The Ni/CeO2 catalyst (with a TOF of 0.03 s −1 at 400 • C, lower than the 0.4 s −1 measured on Ni/SiO 2 at the same temperature [170]) ended up being the recommended one, due to its higher activity and catalytic stability, along with its lower cost, and because the bimetallic Ni/Ru/CeO 2 showed no improvement over Ru/CeO 2 . Around the same time, Vacharapong et al tried a similar approach: a Ni catalyst over an Al 2 O 3 support doped with Ce, including magnetic inducement (to aid in controlling the composition and uniformity of Ce) [171]. Both the doping and the magnetic field were found to increase the ammonia conversion percentage and TOF. When both changes are taken into account, the ammonia conversion percentages measured on magnetically induced Ni on Ce-doped Al 2 O 3 were almost twice as high as on Ni/Al 2 O 3 , with the largest effect coming from the doping rather than from the magnetic inducement.
The plasma-assisted ammonia decomposition catalytic activities of monometallic Fe, Co, Ni and Mo catalysts were compared with those of some of their bimetallic combinations (Fe-Co, Mo-Co, Fe-Ni and Mo-Ni), deposited on SiO 2 [172]. In terms of ammonia conversion percentage, Co proved to be the best among the monometallic catalysts, and Fe-Ni the best among the bimetallic ones, with NH 3 conversions of 48% and 60%, respectively, at 460 • C. Curiously, Fe-Ni was the only bimetallic surface which displayed a higher conversion percentage than the parent metals, at any temperature between 330 • C and 460 • C. Given its promising capabilities, Fe-Ni was then tested with different molar ratios, and the activity was found to peak at 61% on 6Fe-4Ni, showing excellent durability over a 200 h long period of continuous operation. The 6Fe-4Ni catalyst also exhibited the highest synergy with plasma: at 500 • C, over 99.9% of ammonia was decomposed, at a TOF of around 3.46 s −1 . Optical emission spectra characterization showed that plasma pre-activates NH 3 into NH 2 and NH and accelerates both the initial ammonia adsorption on metals and the final recombinative desorption of N adatoms. Consequently, the combination of plasma technology with a relatively cheap metal catalyst proved promising in the quest for better catalysts for ammonia decomposition.
One of the most recent studies involves quinary nanoparticles made up of Co, Mo, Fe, Ni and Cu [173]. Tests at 500 • C revealed that the ammonia conversion percentage and the TOF critically depend on the Co/Mo ratio, with a volcano-type relationship with the N adsorption energy. While for ratios ⩽ 1 the TOF was found lower than on Ru/TiO 2 (3.08 s −1 ), including more Co than Mo resulted in TOFs of 5.45 and 7.00 s −1 for ratios of 9/5 and 11/3, achieving conversions of 64.5% and 100%, respectively. The Co/Mo ratio in these quinary nanoparticles can therefore be adjusted to optimize their catalytic activity under different experimental conditions.

Final remarks and future directions
The review of the literature concerning the utilization of multimetallic catalysts in some important reactions for H 2 production shows that the added metals have different roles in the catalytic processes. The metal alloying alters the band structure, which necessarily changes the reactivity of the material, leading to different catalysis-relevant changes, such as adsorption mechanisms and strength, as well as alternative surface reaction paths. The supports have been found to also have important roles in the reactions: by competing with the metallic particles for adsorption of some intermediate species, the poisoning of active metal sites can be avoided; in other cases, the support induces changes in the catalytic properties of the metal particles which are beneficial.
In the case of the WGSR, an important synergic effect was found when alloying copper and nickel. While the shift activity of nickel is greater than that of copper, nickel is found to be less selective and some formation of methane occurs. The alloying of the two metals leads to high reaction rates while retaining methane formation, i.e. copper is essential to control the high activity of nickel towards the cleavage of the C-O bond of carbon monoxide, thus suppressing undesirable side effects. Results of computational studies suggest that the incorporation of very small amounts of other abundant metals, e.g. Zn or Al, may dramatically improve the activity of copper towards the WGSR. The addition of Zn shifts to lower values the reduction temperature of CuO to Cu while the main role of Al is to increase the catalyst stability. In the case of the supported catalysts, catalysts using reducible oxide supports are much more active than those employing non-reducible oxides.
The steam reforming of hydrocarbons relies essentially on nickel-based catalysts. However, these catalysts are prone to coke formation. This problem can be somewhat prevented upon the utilization of high H 2 O/CH 4 ratios but, unfortunately, this generates increased costs in post-reaction separations. Therefore, it is of utmost importance to find catalysts that can oxidize adsorbed carbon into CO. Multimetallic catalysts alloying nickel with not only noble metals, such as gold, platinum, or silver, but Earth-abundant metals such as copper and cobalt have also been suggested in the literature to inhibit coke formation while retaining or increasing activity. Apparently, the metal element added to the nickel catalyst prefers to interact with very low-coordinated Ni atoms, which are highly reactive sites that promote formation of carbon deposits. Therefore, in this specific catalytic process, the added metals do not lead to an increase of the intrinsic activity of nickel towards the steam reforming but have an important role of avoiding coke formation, improving the stability and resistance of the catalyst.
Traditionally, Pt is the most studied and promising catalyst for methanol decomposition into hydrogen and carbon monoxide. One major problem of using pure Pt is the quick CO poisoning of its surface, caused by strong CO adsorption. This motivated the still ongoing research of multimetallic alternatives, which attempt to keep the catalytic potential of monometallic surfaces for dehydrogenating methanol, while mitigating the CO saturation. Among these, Pt/Au is quite popular for its success in greatly reducing CO poisoning and in some cases providing reaction paths which avoid the production of CO.
Studies of bimetallic materials as catalysts for ammonia decomposition show that they have the potential to be more active than Ru-based materials. New combinations of metals and technologies are still being discovered, with the material design being further complicated by the unpredictability of some properties. While some of the studies reviewed here used the simple periodic table interpolation to form bimetallic compounds, this approach sometimes yields results worse than those of the parent metals. Employing first-principles calculations to predict and tune nitrogen binding energies has been shown to often lead in the desired direction, due to the established volcano-like relationship between this quantity and the TOF of catalysts. However, this area is still relatively new and further tests are still needed to confirm the reproducibility and stability of the results available so far.
Regrettably, the unifying feature of some of these reactions is the reliance on precious metals, given their resistance against oxidation and selectivity. Naturally, since these metals are rare by definition, more sustainable alternatives are required. This motivates the use of first-row TMs, such as Cu-and Ni-based bimetallic materials. Encouragingly, these non-precious TMs are also catalytically active, in special when alloyed with other Earth-abundant metals which can lead to further improvements if when wisely combined.
Despite the significant knowledge gathered from the experimental and computational studies reported in the literature, additional work is needed. Experimentally, studies are needed to fully understand the compositions of the catalysts under operando conditions, either the support termination, its interface with the metal particles and the amount of segregation at the surface of the metal particles. This information is crucial for developing structural models that can be used in parallel computational studies. In our view, the synergic combination of experimental and computational works is important to advance from trial and error synthesis to de facto tailored design of catalysts.

Data availability statement
No new data were created or analyzed in this study.