Concluding remarks: Reaction mechanisms in catalysis: perspectives and prospects

We consider the current status of our understanding of reaction mechanisms in catalysis in the light of the papers presented in this Discussion. We identify some of the challenges in both theoretical and experimental studies, which we illustrate by considering three key reactions.


Introduction
This Faraday Discussion has clearly demonstrated both the importance of developing an understanding of the mechanisms of catalytic reaction and the rapid growth in our ability to develop detailed mechanistic models. The Discussion has also highlighted the challenges in the eld. This article cannot attempt to present a comprehensive survey, but will focus on four themes which have been particularly prominent in the Discussion. The rst concerns the role of theory and modelling; the second, the relationship between structure and mechanism; thirdly we consider developments in in situ/operando techniques; and nally, we focus on three key reactions: methanol to gasoline (MTG); ammonia synthesis; and CO 2 hydrogenation. We conclude by considering the future prospects for the eld. Both these studies reveal the power of DFT techniques in identifying key intermediates and in producing mechanistic energy proles. However, the rapid growth in the utilisation of DFT raises a number of issues that we will now discuss in more detail.

Accuracy of DFT methods
As argued above, DFT has unquestionably a major role to play in computational catalysis, but it is nevertheless important to note that it is an approximate method and that there are inaccuracies in energies and more especially activation energies calculated using standard functionals. The paper of Yang and Yang in this Discussion (DOI: 10.1039/C9FD00122K) makes an important contribution by allowing us to quantify the impact of these uncertainties on proposed mechanisms; and in many applications the inaccuracies may be unimportant as only qualitative information and trends are needed. But if more quantitative information is required, the limitations of standard DFT approaches may become problematic. One approach is to use more sophisticated functionals, especially hybrid functionals which mix in a proportion of exact exchange into the exchange correlation functional, thereby providing improved accuracy at the cost of greater computational expense. An alternative approach is to use quantum mechanical/ molecular mechanical (QM/MM) methods, which are discussed in greater detail below.

QM/MM methods
The application of atomistic modelling techniques in the study of heterogeneous catalysis has been dominated by the use of periodic methods; and in DFT studies Fig. 1 The QM/MM methodology in applications to solid state and surface modelling. The inner region (red) is described by a QM technique, while the outer region (blue) is modelled using an MM description. An appropriate description of the interface region (green) between the QM and MM regions is important. In modelling ionic crystals, an outer layer of point charges (grey) is added to ensure that the electrostatic potential in the QM region is accurately described. the use of periodic, plane wave methods has been very extensive, with the use of the VASP 1 code being widespread. QM/MM methods adopt an alternative approach, as illustrated in Fig. 1. The concept is simple. The region of interest, which in the case of a catalytic process will be the active site, reacting molecule and surrounding atoms, are described quantum mechanically (QM) and embedded in a more approximate interatomic potential or molecular mechanical (MM) description, which provided the chemistry is localised within the QM regions will be adequate. The method has some distinct advantages over periodic approaches: rst, it avoids the articial interactions between periodically repeated sorbed molecules; secondly, there is a signicant technical advantage in calculations of e.g. ionisation potentials and proton affinities, in that the vacuum level is clearly dened, which is less apparent with periodic methods; but in the present context, perhaps the most signicant advantage is that it is possible to use very high level quantum mechanical methods in the QM region, which would be impractical with periodic techniques. An excellent example is the landmark publication of Piccini et al., 2 in which by the use of QM/MM techniques with a high level (MP2) QM core, achieved chemical accuracy in modelling the reaction of methanol with alkenes in zeolites. Such accuracy is needed to calculate accurate reaction rates, although for many applications identifying trends and qualitative aspects, lower accuracy may, as noted, be quite acceptable.
The use of QM/MM methods in catalytic science will continue to grow, while periodic methods will, of course, continue to make wide ranging and important contributions.

Microkinetic modelling
This increasingly used technique explores the mechanistic consequences of calculated reaction energy proles by linking the calculated activation energies to a Monte-Carlo routine. It allows the relative contributions of different mechanistic pathways to be explored as a function of temperature and pressure and for patterns of reactant and product coverages on reacting surfaces to be predicted.  Fig. 2, as well as the relevant adsorption and diffusion processes. These were then input into the kinetic Monte-Carlo (KMC) algorithm available in the Zacros code. 5 The analysis shows that under reaction conditions, the surface is dominated by adsorbed CO and H. Moreover, the work gives valuable mechanistic insights. It shows that the most common events are H 2 O and OH decomposition which are reversible; it nds that OH radicals react with CO to form a carboxyl species, which can subsequently decompose to generate CO 2 ; it also shows that O formed by the dissociation of OH can react directly with CO to form CO 2 . It appears therefore the WGS reaction on Pd involves both direct oxidation and carboxyl pathways. Many other examples will be found in the recent literature illustrating the power of this approach.

Metadynamical simulations
Metadynamics is a powerful tool, which allows the investigation of activated processes which would be inaccessible to conventional molecular dynamics, enabling free energy barriers to be calculated. A review of its application to catalytic processes is available from Van Speybroeck et al. 6 A recent example is provided by the work of Nastase et al., 7 who applied metadynamical simulation techniques to the study of the reactivity of methanol in ZSM-5. Ab initio dynamics rst established that methanol clusters can deprotonate framework acid sites. Metadynamical simulations were then able to model the energetics of framework methoxylation. Methanol loading is found strongly to inuence the activation energy of this key reaction, especially in the presence of a second acid site; the second site polarises the hydrogen bonded network and can reduce the activation barrier by $40 kJ mol À1 . The commonest approach is to represent the solvent as a continuum; but with the continuing growth of computer power, explicit solvent modelling will become increasingly viable. QM/MM techniques will also have an important role to play in meeting this challenge.

Structure and mechanism
Our comments here will be brief. We wish simply to emphasise that accurate structural models are essential for the development of detailed mechanistic models, which poses a particular challenge in heterogeneous catalysis, where there are oen uncertainties in the structure and composition of surfaces or of nano-particle structures. A good illustration of the latter is provided by the work of Malta et al. 8 on acetylene hydrochlorination catalysis by a supported Au catalyst; the work showed that, in contrast to the previous widely held view that the active sites were gold nano-particles, single gold cations were present in the operating catalyst.
The complexity of oxide surface structures is illustrated by both experimental 9 and computational 10 studies of the polar surfaces of zinc oxide, which despite the simplicity of the bulk structure are highly complex with triangular vacancy clusters as illustrated in Fig. 3. The need for detailed structural models emphasises the role of advanced characterisation techniques, particularly those highlighted in the next section. An important recent development is the ability to undertake "in reactor" and spatially resolved experiments. The former is nicely illustrated by the paper in this Discussion of Nieuwelink et al. (DOI: 10.1039/D0FD00006J) which was able to probe single particles in a microreactor during multiphase hydrogenation reactions. To illustrate the latter, we refer to the recent work of Decarolis et al. 11 who performed isothermal spatially resolved measurements on Pd/Al 2 O 3 during NH 3 selective catalytic oxidation (SCO). Their approach is illustrated in Fig. 4; and the results show clearly that in situ and operando experiments that rely on single point analysis can miss vitally important information. The study nds that at the inlet of the reactor, the Pd speciation is predominantly PdN x ; that as the reaction exotherm progresses towards the outlet, the bulk interstitial N become mobile and the presence of a surface nitride structure is identied; that whilst the nitride is present there is still appreciable selectivity toward N 2 ; and as the particles become more oxidised, there is a change in selectivity toward NO x products. Finally, at 400 C, the data suggest that not only the NH 3 -SCO but also an NH 3 selective catalytic reduction like behaviour is happening on the catalyst, converting the NO produced by the oxidation to N 2 O. The need for full spatial analysis is increasingly apparent and will assume additional importance as our ability to perform this type of experiment develops.

Key systems: mechanistic challenges
The three key catalytic systems and processes discussed here will illustrate some of the progress and challenges in understanding reaction mechanisms in catalysis.

Methanol to gasoline (MTG)
This widely studied reaction catalysed by zeolite ZSM-5 is discussed in detail in the introductory article of Hutchings (DOI: 10.1039/D1FD00023C). Here we focus on one aspect: the role of carbenes and related species in the formation of the initial C-C bond, the understanding of which is possibly the greatest challenge in unravelling the complex mechanism of MTG catalysis. As argued by Hutchings, several pieces of evidence have pointed to a signicant role of carbene like species with further support being provided by recent synchrotron based IR studies of Minova et al. 12 However, calculations of Sinclair and Catlow 13 found high formation energies of 215-232 kJ mol À1 for the deprotonation of methoxy groups to form carbenes. The problem has recently been re-examined by Nastase et al. 14,15 using QM/MM techniques. The carbene structures examined are illustrated in Fig. 5. The calculations nd even higher energies than those obtained previously. They show, moreover, that both methoxy and carbene species are immobile. The results suggest that, if, as indicated by experiment, carbenes play a signicant role, the mechanism cannot be stepwise; rather a concerted process involving the simultaneous formation and reaction of carbenes may operate. Clearly there is need for more computational and experimental work if we are to understand the crucially important mechanism of the rst C-C bond formation.

Nitrogen activation
There can be no more striking example of the global importance of catalytic science and technology than ammonia synthesis, with the nutritional requirements of half the world's population being dependent on nitrogen xed by the Haber Bosch process, the energy intensive nature of which has, however, provided a stimulus for the development of catalysts that can operate under milder  The key problem in nitrogen activation is the strength of the nitrogen-nitrogen triple bondthe strongest chemical bond. Haber Bosch catalysis works by a "brute force" mechanism in which this bond must rst dissociate before the dissociated nitrogen atoms are hydrogenated. Is a more subtle and gentle mechanism possible? Work of Zeinalipour-Yazdi et al. 17 suggests that this may be the case. In computational work on the Co 3 Mo 3 N catalyst, the mechanism summarized in Fig. 6 was proposed. It involves the initial formation of surface vacancies, which can trap and activate nitrogen molecules, which are then hydrogenated without dissociation. The N-N bond is only broken when it has been degraded from a triple to a single bond. The development of this kind of associative mechanism may, we consider, be the answer to achieving nitrogen activation under relatively mild conditions.

CO 2 hydrogenation
As is well known, effective catalytic processes for carbon dioxide hydrogenation are essential for the development of sustainable chemicals and fuels. Mechanisms of hydrogenation are, however, widely debated; see, for example, the recent review of Nitopi et al. 18 The Cu/ZnO catalyst used in methanol synthesis from both CO 2 and syngas has been intensively studied, without the emergence of clear   7 Schematic depicting all of the mechanistic pathways for CO 2 hydrogenation to methanol over Cu(110) and Cu(100) surfaces investigated by Higham et al. 19 Intermediates framed with orange borders correspond to those unique to the CO hydrogenation processes subsequent to CO 2 dissociation, whilst the intermediates framed with purple borders are unique to direct CO 2 hydrogenation processes. The species and intermediates framed with black borders are common to all of the mechanistic pathways investigated. The mechanism defined by the red arrows corresponds to the least activated pathway as determined by the calculations, with the reaction proceeding through CO 2 hydrogenation to formate, formic acid, and oxy hydroxymethylene, which dissociated to yield formaldehyde, which can subsequently be hydrogenated to methanol via either methoxy or hydroxymethylene, with the former being marginally less activated overall. Reproduced from ref. 19 with permission from the Royal Society of Chemistry. consensus as to mechanistic aspects. Recently, as a precursor to a mechanistic investigation of the Cu/ZnO system, a detailed computational (DFT) study was reported of the mechanism of CO 2 hydrogenation over metallic copper surfaces. 19 The mechanistic possibilities are summarized in Fig. 7. There are two distinct routes: one via CO 2 dissociation and subsequent hydrogenation of the resulting CO; the other is non-dissociative. The two routes converge at later stages.
The rst and crucial step is CO 2 adsorption and activation. Interestingly both physisorbed and chemisorbed states are possible. The latter is metastable with a bent conguration, and with electron transfer from the metal activating the adsorbed molecule. Hydrogen dissociation is mildly exothermic with a low activation barrier. The calculations suggest that the non-dissociative route is more favourable. Reaction and activation energies for individual steps for this mechanism on the Cu(110) surface are given in Table 1. The results show that HCOO* formation is less activated than COOH* formation; and that HCOOH* formation from both HCOO* and COOH* is the least activated subsequent process. Hence OCH 2 OH* formation can be achieved via either the HCOO* or COOH* route through HCOOH* hydrogenation and subsequent dissociation to formaldehyde; the next steps are common with the CO pathway. The results also show signicant differences between the reactivity on different surfaces. Current work is extending the mechanistic analysis using models of Cu nanoclusters adsorbed on polar ZnO surfaces recently developed by Higham et al. 20 The work is assisting us to unravel the mechanistic complexities of this widely studied reaction.

Conclusions and future prospects
This Discussion volume has clearly shown that the eld is developing rapidly and that we have an increasingly powerful set of computational and experimental techniques for probing reaction mechanisms in catalysis. Computational studies will become increasingly predictive with our growing ability to model real complex systems and with improvements in the accuracy of the calculations. Experimental work will have a growing reliance on advanced characterization techniques used in operando; and experiment and theory will be used increasingly in a concerted manner. An improved ability in understanding mechanisms will be vital in optimising and designing catalytic processes.

Conflicts of interest
I have no conicts of interest to declare.