Metal-Catalysed Reactions of Hydrocarbons / 05-Introduction to the Catalysis of Hydrocarbon Reactions

These represent the absolute minimum of information required, and in some cases it is possible to move on from them, at least in a speculative fashion. Modern experimental methods, to be mentioned briefly in Section 5.7, sometimes reveal the structures of possible or probable adsorbed intermediates, and theoretical techniques have a role as well. The concept of the most abundant surface intermediate (MASI)22 has proved useful for some systems, but caution is necessary because that species is not necessarily the vital reactive intermediate; indeed it may only be abundant and observable because it is unreactive. Such appears to be the case with ethylidyne in the hydrogenation of ethene; the much less abundant π-adsorbed form may well be the reactive state. It is also necessary to consider, in the case of supported metals, exactly where the reaction is taking place. We shall meet at least one example of a reaction occurring on the support using hydrogen atoms arriving by spillover from the metal (Chapter 10). Formulating a rate expression when this situation occurs needs great care.

Reaction schemes such as 5.G are very incomplete accounts of mechanism. A fuller statement will show symbolically the adsorbed intermediates as in 5.C and 5.D; the existence of such intermediate species may have to be inferred, because they do not necessarily lead to an observable product. Thus, in the reaction sequence 5.I, X and Y may be seen in the gas phase, depending on the relative values of the

rate constants, or Z may be the only detectable product. These possibilities occasion much discussion as to whether the straight A to Z process occurs at a different kind of site, and perhaps by a different mechanism, from that giving X and Y as well. This question arises when discussing the hydrogenation of ethyne (Chapter 9) and the hydrogenolysis of n-butane (Chapter 13). While it is perfectly possible that a reaction may proceed differently at distinct sites, and may follow different kinetics, the Principle of Economy of Hypothesis, otherwise known as Occam’s Razor, requires us first to explore the simplest option, which is the ‘rake’ mechanism shown in Scheme (5.I).

5.4. THE IDEA OF THE ACTIVE CENTRE43–48

The term ‘active centre’ (or ‘active site’, which is the same thing) was first employed by Sir Hugh Taylor in his far-sighted papers published in 1925 and 1926.49 They should be required reading for all students of heterogeneous catalysis, because their thinking has guided generation after generation of researchers in the

field, and their message is still very relevant. The central theme is that the surface of a typical metal contains atoms of various co-ordination number, and that each class of reaction will only proceed at a place named the ‘active centre’ where there is an atom, or a group of atoms, of the appropriate type. Each reaction or reaction class proceeding on a given surface will have its own specific requirement, the stringency of which may vary from one extreme, where only a small fraction of the surface is active, to the other, where all sites are suitable and the whole surface is active. In Taylor’s words,49 The amount of surface that is catalytically active depends on the reaction catalysed. The first type of reaction was originally called ‘demanding’ and the latter ‘facile’,19,50,51 but ‘facile’ has a pejorative connotation, and these terms were later replaced by the still not entirely satisfactory ‘structuresensitive’ and ‘structure-insensitive’.19,24,52 The usefulness of these names, and their possible refinement, will be considered again in a moment.

Taylor’s picture of a typical surface, containing a highly unstable arrangement of atoms having mainly low co-ordination numbers,49 is one to which we would not now subscribe, but considering the simple case of atoms on the surface of a cube helps to make the point. The corner atoms might be the locus of a structure-sensitive reaction, while those in the sides (excluding edge atoms), being the majority except for the smallest cubes, could be responsible for a structure-insensitive reaction. An active centre may therefore be regarded as having a specified number of atoms having defined co-ordination numbers, and held in an arrangement peculiarly effective for the reaction in question.53

The number of active centres is therefore generally less than the total number of surface atoms,54 the ratio of the two being termed the Taylor fraction FT (or Taylor ratio). A.A. Balandin attempted to make Taylor’s ideas more precise by proposing a Multiplet Hypothesis55,56 by which each type of reaction required a ‘multiplet’ of several atoms to be an active centre. So for example to chemisorb a benzene molecule either for its hydrogenation or in its formation by dehydrogenation, a hexagonal multiplet with one central atom as found in the fcc(111) plane would be needed. If the number of atoms composing the multiplet (or ensemble as it was called by Kobozev57,58) is the Balandin number NB, then

Frennet’s free potential site58–60 is a closely related concept to that of the active centre, as is Campbell’s true ensemble requirement.61

It should now be clear that the attribute of ‘sensitivity’ belongs to a catalytic system comprising reactants plus catalyst: it cannot be safely assumed that because a reaction appears structure-insensitive on one type of catalyst it will necessarily be so on another. Thus for example we might expect alkene hydrogenation to be ‘insensitive’ on nickel, but ‘sensitive’ on copper, where only a few surface atoms are able to assist. It should therefore be possible to classify metals accordingly to

their structure sensitivity for a given reaction, but there has been little progress in this direction to date.

We have noted elsewhere a strong desire on the part of scientists to categorise an observation by placing it in one of only two boxes, feeling that thereby they have advanced their understanding. The affixing of the labels ‘structure-sensitive’ or ‘structure-insensitive’ to a system is in fact an oversimplification, and conflicts with Taylor’s idea49 of an infinite variability of sensitivity with reaction type. We may conceive the typical surface as having atoms belonging to one of a number of sub-sets containing atoms of specified co-ordination number, or of a defined mix of different co-ordination numbers (e.g. one corner and one edge). We might therefore more usefully speak of a degree of structure sensitivity, depending on the number of the sub-sets that qualify to contribute active centres. This concept has been given quantitative expressions in the context of particular size variation by David Avnir24,62,63 using fractal analysis. A reaction system is assigned a reaction dimension DR defined by the equation

where (r /t −1) is the turnover frequency and R the mean particle radius. This equation affords convincingly linear plots for a number of reactions, and the values of DR are often nearly integral: they can be interpreted in terms of the involvement of a specific class of surface atom (corner, edge, plane etc.) according to the manner of their expected occurrence as a function of size, as expounded by van Hardeveld and Hartog.64–66 Fractional numbers denote the involvement of more complex atomic groupings. While the validity of this approach has been seriously questioned,24 its value lies in attaching shades of grey to a concept that is otherwise seen only in black and white.

Assessment of sensitivity and, for a ‘sensitive’ reaction, identification of the active centre and estimation of their number, have been prime objectives of much research, but definitive answers are unfortunately elusive. One contributing factor to the difficulties encountered is the likely mobility of surface atoms, especially those on small particles and at high temperatures, where the surface may be in a semi-molten state (Section 2.5.3). Mobility may occur to a lesser extent even on single crystal surfaces.67 Somorjai has advanced the idea that active centres do not pre-exist, but are formed by adaptation of the surface to the reaction being catalysed, suitable sites being created by reconstruction. This process undoubtedly occurs on single crystals, and there is evidence that strongly adsorbing molecules such as alkynes can withdraw atoms from their normal lattice positions (see Chapter 9). Structure sensitivity was first recognised by a decrease in activity for hydrogenolysis of neopentane relative to its isomerisation after heat treatment of a Pt/C (Spheron 6) catalyst, although little change in particular size was seen.50,51

It was thought that active centres were removed by annealing, i.e. by smoothing an initially rough surface, although other explanations are possible.

For single crystal surfaces, a reaction is deemed ‘insensitive’ if its rate is about the same on all low Miller index planes, but since these differ from small metal particles in not having atoms of very low co-ordination number, the term face sensitivity should be used in this case. Two further approaches to the general problem have been tried: (1) systematic variation of particle in supported metal catalysts, and (2) alteration of the composition of the surface of bimetallic catalysts, either supported or unsupported (Section 5.7). These lead respectively to particle size sensitivity and ensemble size sensitivity, but the three types are not necessarily exactly the same.

The literature is full of attempts to deduce the nature of an active centre by systematically changing the particle size in a supported metal catalyst.44,68–70 The argument runs as follows. Alteration of the mean particle size, for example, by changing the metal content, the method of preparation, or the thermal pre-treatment, will alter the numbers of atoms having a particular co-ordination number, or the numbers of atoms forming specific groupings, in a way that can be predicted by the use of models such as those advanced by van Hardeveld and Hartog.64 The so-called B5 site, comprising five atoms in one of two configurations at a step, has claimed particular attention.29,65 Oles Poltorak70 also contributed to early work in this field, coining the term mitohedrical region to describe the size range where proportions of edge and corner atoms change quickly. This has also been termed the mesoscopic region, as being intermediate between the microscopic and the macroscopic. Correlation with the corresponding change in specific rate may lead to some idea as to the type of active centre responsible for the reaction. It is undoubtedly true that specific rates of many catalysed reactions do change in a regular manner with mean particle size, even if this is only expressed as an H/Ms ratio; the specific rate may either increase or decrease or pass through a maximum (hardly ever a minimum) with increasing size (see Figure 5.10).

Figure 5.10. Possible forms of dependence of TOF or areal rate on particle size: 1, no dependence; 2, negative/antipathetic dependence; no dependence; 3, positive/sympathetic dependence; 4, TOF passes through maximum.

There are however a number of difficulties and dangers in this procedure.19,71

(1) As noted in Sections 2.4 and 5.2.3, the particle size distribution may be binodal, a fact only revealed by TEM; in this case, mean size as determined by selective chemisorption has no meaning, and the observed activity will be the sum of those given by two quite different kinds of particle. (2) Additionally, the H/Ms ratio is a function of size for very small particles;72 hydrogen chemisorption is therefore not a reliable method for size estimation. (3) Most importantly, in a supported metal catalyst there will inevitably be a size distribution, even if it is mononodal, and the population of surface atoms of a selected co-ordination number will vary greatly across the distribution. As noted earlier (Section 2.41), only the mean surface co-ordination number changes smoothly with size, and use of numbers derived from regular model particles having a complete outer layer can be very misleading. (4) Methods of preparation designed to effect size changes may bring about other differences, notably to the concentration of residual impurities such as chloride ion. (5) Many other properties alter with size besides surface structure as discussed earlier (Section 2.5). (6) Finally, as we have seen, surface mobility may vitiate entirely the possibility of a strictly geometric factor in catalysis. Thus while we may speak of a particle-size (or dispersion) sensitivity, it would be unwise to attribute this to a specific structural property. This strongly negative note is regretted, but it is essential: the concept of sensitivity will be re-visited in the next Section with a somewhat more positive result.

A further cautionary word must be added. Conclusions concerning the particle size sensitivity of a reaction are often based on a quite inadequate amount of experimentation, usually just the rate measured under a single set of conditions. There is therefore no knowing how dependent the conclusion is on the choice of those conditions, and whether the use of other reactant pressures or temperatures would have led to other conclusions. This frequent economy of effort is hard to understand, as the necessary physical characterisation is much more expensive and time-consuming, and is quite inexcusable now that micro-processor-controlled equipment is so freely available. A salutary example is provided by benzene hydrogenation over Ni/SiO2, where with increasing particle size the rate at room temperature increases, but above 453 K it decreases.58 There are other indications of a change in rate-controlling step between these temperatures (see Chapter 10).

Accepting the existence of ‘active sites’ implies also that there are also ‘inactive centres’ or perhaps ‘overactive centres’ that are quickly inactivated by the destructive chemisorption that occurs so easily with hydrocarbons. The accidental or deliberate removal of such sites by autogenic toxins (i.e. ‘carbonaceous deposits’, ethylidyne etc.), or by other poisonous species such as sulfur compounds, allows the remaining ‘active centres’ to exhibit reactions of a structure-insensitive type that could not take place while the overactive centres were in existence. This situation may arise not only through surface heterogeneity, but also on a plane surface by operation of the Principle of Maximum Occupancy (Section 4.2) by which the preferred first reaction is that which utilises the largest size of active

centre by forming the greatest possible number of bonds to the surface. There remain in between these species, which are usually dehydrogenated forms of the reactant hydrocarbon (Section 4.6) but may be other toxins, small sites on which other reactions can proceed; so for example ethene hydrogenation is thought to proceed typically on a relatively few sites not occupied by ‘spectator species’ such as ethylidyne (Chapter 7). Active centres isolated by strongly-held species or other site-blocking species may also catalyse reactions that are impossible on clean surfaces; thus ruthenium catalysts, which are normally very active for hydrogenolysis of alkanes, can when suitably poisoned give very high selectivities for skeletal isomeration (Chapter 14).

According to the Catalytic Cycle (Figure 5.1), an active centre is a place where the reactants are adsorbed adjacently, ready to react. We may therefore attempt a working definition of an active centre as a location on the surface of a catalyst where reactants may be adsorbed in the right way and with the best strengths to give the desired products efficiently. It is not necessary for the reactants to adsorb directly onto the active centre; one or both may adsorb elsewhere and then diffuse to the active point; this is what happens in spillover catalysis. There are without doubt cases where two reactants do not adsorb in competition on the same type of site; each has its own requirement, and it may be more closely restricted for one reactant than for the other. The active centre will then contain two sites of different character, on which each reactant is adsorbed non-competitively. Kinetic analysis should reveal when this situation arises.

Correct estimation of the number of reacting centres is essential for the accurate measurement of turnover frequency. Although this is very hard to do in practice, a mental picture of what constitutes a reacting centre may be helpful. Imagine a reaction in a flow system proceeding under quite steady conditions, and freeze the motion of time: the number of reacting centres is the number of places at which at that instant of time the reactants have progressed so far along the reaction co-ordinate through the transition state that formation of products is inevitable. The literature records only one attempt to estimate this number by a transient response method: using14C-labelled ethene, the fraction of exposed metal atoms constituting reactive centres during its hydrogenation was for Pt/SiO2 50% and for Ir/SiO2 17%.51

5.5. THE USE OF BIMETALLIC CATALYSTS

In the search for information on the composition of active centres, and for materials of improved catalytic performance, very much use has been made of bimetallic catalysts (see Further Reading at the end of the chapter). The term is preferred to ‘alloy’ as in many cases the degree of intimacy of the components is uncertain, and in some cases interesting behaviour is found with systems exhibiting

only very limited mutual solubility: it implies however that both components are in the zero-valent state during use, although this requirement does not prevent discussion of systems where this is not fully proved, or where perhaps one component is only partially metallic. There is indeed a very narrow and somewhat arbitrary dividing line between ‘promotion’ and the effects shown in bimetallic catalysts. We should note too that although multi-element formulations often appear in the patent literature, the presence of two components poses quite sufficient difficulties for the academic scientist.

Bimetallic catalysts have had outstanding success in industrial applications, most notably in petrochemistry and in petroleum reforming, where the combinations Pt-Sn, Pt-Ir and Pt-Re have found widespread use, and have been pervasively studied in both industrial and academic laboratories. Their success cannot be assigned to a single cause, but rather to a number of favourable factors working in concert; and it is largely to this success that we owe our extensive knowledge of other bimetallic systems through a kind of scientific spillover.

Early research on bimetallic catalysts12 used mainly simple reactions such as parahydrogen conversion and formic acid decomposition, and mainly unsupported forms such as wires, films, foils and powders, since there was no certain way of making supported bimetallic catalysts and no known means of characterising their surfaces. Its motivation was to detect an electronic factor in catalysis, and favoured systems were those in which there was a continuous range of solid solutions; and therefore, it was thought (mistakenly) that there would be a monotonic change in the electron: atom ratio of all component atoms. Significant changes in activity were indeed observed and correlated with composition (see for example Section 3.4.2), but unfortunately the interpretation placed on them was not right. Revision of these views was made necessary by the findings made by electron spectroscopy (Section 2.5.4), which has complicated rather than simplified the task of understanding the results. This phase of research was followed in the period 1970–1990 by the use of supported bimetallics or more complex reactions such as those of hydrocarbons that modelled those occurring in petroleum reforming: very notable contributions were made by John Sinfelt, Wolfgang Sachtler, Vladimir Ponec, John Clarke and many others. One of Sinfelt’s major discoveries was that extended mutual solubility is not a prerequisite for a useful bimetallic system, because a surface ‘alloy’ formed for example by copper atoms on the surface of a ruthenium or osmium particle was stable and usable, and this has led to the widespread use of surface alloy formed by vapour deposition of atoms of one metal onto a single crystal surface of another (Section 1.3.1).

The critical question now addressed was whether the results could be understood simply in terms of the effect of the ‘inert’ metal on the mean surface ensemble size of the active partner, or whether there was some movement of electrons between the components, modifying the properties of the more active one. Once again, effort was concentrated on bimetallics formed from Groups 10 and

11, because the additional electron brought about a catastrophic decrease in activity, and because many of the combinations showed complete miscibility. It has been the aspiration of many workers in catalysis to find an unambiguous answer to the straightforward question: ensemble size effect or ligand effect? The answer is not however always so simple as the question, and unfortunately most scientists have failed to appreciate that the best answer may be ‘a little one and a lot of the other.’ We may anticipate the conclusion of a great body of experimental work by saying that, when a bimetallic system is formed by two metals differing greatly in catalytic activity but only slightly in electronic constitution (e.g. Ni-Cu; Pd-Ag), ensemble size is the dominant factor; but when the two metals differ significantly in electronic structure (eg. Pt-Sn; Pt-Re) then the ligand effect may also be significant.

Calculation of the number of pairs, triplets or larger ensembles of one kind of atom randomly dispersed on a plane surface containing two kinds is a simple application of binomial theorem.13 Use of the results in real systems is however predicated on a number of assumptions, and conditions that have to be met.13 These may be enumerated as follows.

The bimetallic system must be homogeneous, comprising a single phase: phase separation may occur below a critical temperature as with the NiCu/SiO2 catalysts.58 Ordered superlattices may occur at certain compositions (e.g. Pt3Cu).

There must be no short-range ordering, i.e. no preferred formation of clusters of one component in the bulk or on the surface. This condition is quite well met with Ni-Cu and perhaps other Group 10–11 bimetallics, but the greater the disparity in electronic structure the more likely it is to occur.

If segregation of one component to the surface is thought likely, and in theory it is nearly always possible, the surface composition must be measured, for example, by XPS.

The possibility of reactionor chemisorption-induced heterogeneity has to be recognised, as the component interacting most strongly with the reactants may be drawn to the surface. The surface composition has therefore to be checked after reaction, as well as before.

In the case of small particles and stepped or kinked surfaces, the component of lower surface energy should segregate preferentially to the site of the lowest co-ordination number, where it makes greatest impact on the total energy; and as its concentration is increased, it will occupy progressively edges, rough surfaces and open planes (e.g. fcc(100)), and finally closepacked planes.13 Relevant calculations have been performed for a number of systems and these have shown that the distinction between each class of site is not clear cut, but depends on the difference between the energies of energetically adjacent sites.

Finally it must be remembered that, as particle size is decreased, there is a rising minimum concentration of one component needed to cover the surface, even if its tendency to surface segregation is total. Thus for example at 50% dispersion, less than 50% of either component will not be sufficient to cover the surface fully. For this reason, if there is a broad size distribution the surface composition of all particles may not be the same.

The quantitative significance to be attached to observations of changes in rate with composition is therefore somewhat problematic, and indeed there are further points of uncertainty to be considered. Enhanced activity is sometimes seen on adding a Group 11 element to an active one, and this may well result from a reduced rate of deactivation by elimination of the larger ensembles on which toxic deposits originate. It is therefore important to record initial activities and deactivation rates if proper comparison is to be made. It is also possible that atoms of the Group 11 metal adjacent to an ensemble of active atoms can join in the reaction, i.e. hetroatomic sites may be initiated, and in the limit spillover to areas of the ‘inactive’ component may contribute to the reaction. These possibilities are however not easily confirmed, but the fall in the heat of hydrogen chemisorption on Ni-Cu/SiO2 catalysts with increasing copper content may be one indicator of it.12 Alternatively, hydrogen atoms may be forced into less energetic sites involving only nickel. Failure to recognise surface heterogeneity and its effect on the location of Group 11 atoms may have important consequences: if for example the active centre demands an active atom of low co-ordination number (CN), and if these are quickly replaced in preparations containing some Group 11 metal, then activity will quickly decrease as the active sites are annihilated, and the size of the active ensemble may be thus greatly over-estimated. The same effect arises when say a single nickel atom is replaced by a copper atom, if several other nickel atoms are thereby eliminated from participating in an active centre.73 These considerations may explain the unreasonably large values of Balandin number reported in the work of G.-A. Martin and his associates.58 This tendency for low CN sites to be first occupied by atoms that lower the particles energy has however been used constructively by Bernard Coq and his colleagues to synthesise catalysts in which such sites are absent: marked changes in reaction specificity occur in consequence.

The use of bismuth as an inert site-blocking atom on a single crystal surface (Pt(111)) has produced interesting results.61 Increasing its surface concentration lowers the chance of dehydrogenation or decomposition of cyclic hydrocarbons as against their chance or desorbing unchanged upon heating: cycloalkanes are affected most, then aromatics, and cycloalkenes least. However the point is strongly urged that one has to distinguish between the number of atoms upon which a molecule can adsorb unchanged and the number needed to produce the state which will upon desorption give the dehydrogenated or decomposed products plus gaseous hydrogen. This latter number, the true ensemble requirement, of course

includes the number to accommodate the hydrogen atoms that are formed; it is of the order of five to ten for cyclopentene. However such numbers are only reliably derived from the effect of site-blocking upon the dehydrogenation/desorption ratio when this is very small. Campbell concludes61 with the useful caveat that

One cannot in general assess even qualitatively which reactions require a larger number of free sites simply by observing which reaction are poisoned more rapidly by the addition of a site-blocking agent such as bismuth. One must therefore seriously question even the qualitative conclusions about relative ensemble requirements that have previously been obtained in this manner.

It has often been suggested that a single atom of an active metal in a matrix of an inactive one can perform catalytic functions;74,75 the thought originated with N.I. Kobozev’s concept of the active ensemble,57 and we shall meet one or two cases when it seems very likely. Such an atom is most likely to suffer a ligand effect from its neighbours, but this does not necessarily imply an actual transfer of charge; it is quite possible that it simply experience a change in the occupancy of its electron energy levels: indeed the broadening of energy bonds as revealed by UPS58 strongly supports the idea. An analogous effect is seen with the ‘giant’ magnetic moments shown by iron and cobalt atoms when in the midst of palladium or platinum atoms.58 Charge transfer in the Pt-Re, Pt-Mo and Pt-W systems is much more likely, and a model of the Pt-Re/Al2O3 catalyst76 in which a core of rhenium atoms modifies a surface layer of platinum atoms carries a good deal of conviction. Bimetallic catalysts formed by metals of the same Group rarely produce any surprises.

The point made earlier about the need adequate experimentation before drawing conclusions is just as relevant here as when considering particle size effects. The existence of a true ligand can only be detected if there is a distinct change in the kinetic parameters, particularly the activation energy: extensive work58 on Ni/SiO2 and Ni65Cu35/SiO2 shows this is not so in this case and the nature of the active centre remains the same. We shall find other examples of such behaviour. If however the activation energy changes with composition, then the shape of the rate dependence on it will be temperature-dependent, and the shapes at the high and low temperature may support conflicting theoretical models.

It would be quite wrong to conclude this discussion on a totally negative note. The use of bimetallic catalysts has indeed drawn a very clear qualitative distinction between structure-sensitive and structure-insensitive reactions, as was shown clearly by Sinfelt’s comparison77–81 of the rate dependence of cyclohexane dehydrogenation and ethane hydrogenolysis on the copper content of Ni-Cu catalysts: the former was surprisingly constant over a wide range of concentration, while the latter fell steeply.13 This marked difference in behaviour was confirmed by the work of Clarke,82 and of many others, using various bimetallic compositions. There is thus general similarity between a reaction’s sensitivity to ensemble size and to particle size, since the availability of suitable large ensembles of active