Laureates: Junior Prizes of the sCs faLL Meeting 2018 CHIMIA 2019, 73, No. 4 239 doi:10.2533/chimia.2019.239 Chimia 73 (2019) 239–244 © Swiss Chemical Society What Can We Learn from First Principles Multi-Scale Models in Catalysis? The Role of the Ni/Al2O3 Interface in Water-Gas Shift and Dry Reforming as a Case Study Lucas Foppa§*a, Kim Larmierb, and Aleix Comas-Vivesac § SCS-Metrohm Award for best oral presentation in Catalysis Science & Engineering Abstract: Computational first principles models based on density functional theory (DFT) have emerged as an important tool to address reaction mechanisms and active sites in metal nanoparticle catalysis. However, the common evaluation of potential energy surfaces for selected reaction steps contrasts with the complexity of reaction networks under operating conditions, where the interplay of adsorbate populations and competing routes at reaction conditions determine the most relevant states for catalyst activity and selectivity. Here, we discuss how the use of a multi-scale first principles approach combining DFT calculations at the atomistic level with kinetic models may be used to understand reactions catalyzed by metal nanoparticles. The potential of such an approach is illustrated for the case of Al2O3-supported Ni nanoparticle catalysts in the water-gas shift and dry reforming reactions. In these systems, both Ni nanoparticle (metal) as well as metal/oxide interface sites are available and may play a role in catalysis, which depends not only on the energy for critical reaction steps, as captured by DFT, but also on the reaction temperature and adsorbate populations, as shown by microkinetic modelling and experiments. Keywords: Density functional theory · Metal/oxide interfaces · Microkinetic modeling · Supported metal nanoparticle catalysts Lucas Foppa studied engineering at odic density functional theory (DFT) methods to efficiently solve the Universidade Federal do Rio Grande the electronic structure problem in solids and surfaces boosted do Sul (UFRGS) in Brazil and at the École the use of computational methods to address these questions, with Centrale Marseille in France. After a MSc in the advantage of giving direct access to the underlying electronic chemistry at UFRGS, he joined ETH Zürich structure at the atomistic level.[2] for a PhD devoted to the first principles The first principles computational approach to investigate reac- modelling of metal nanoparticle catalysis. tion mechanisms on NP catalysis typically concentrates on the eval- He is currently a Swiss National Science uation of potential energy surfaces (PESs) for key selected elemen- Foundation post-doc fellow at the Theory tary steps of the reaction of interest (Fig. 1a.1), which are known, Department of the Fritz-Haber Institute of or more often assumed, to be rate and/or selectivity-determining. the Max Planck Society in Germany. More specifically, DFT is used to determine the energy of reac- tants, intermediates and transition states (TSs) adsorbed on (slab) 1. First Principles Modelling of Nanoparticle Catalysis surface or NP models for the catalyst surface sites. The relevance Major industrial processes in chemistry are carried out in many of mechanisms is then discussed based on the energetics of the se- instances with supported transition-metal nanoparticle (NP) cata- lected reaction steps. The most likely pathways are those involving lysts.[1] Some examples are water-gas shift (WGS) and methane not too weak nor too strong binding of key reaction intermediates reforming used for hydrogen and syngas (CO/H2) production, re- (thermodynamic factor) and/or those associated to low-energy TSs spectively, the Fischer-Tropsch synthesis of hydrocarbons and the with hence affordable energy barriers (kinetic factor). When the full Haber-Bosch process for ammonia production. Elucidating the catalytic cycle can be evaluated by DFT, the energy span model, detailed reaction mechanisms and associated surface active sites well established for homogeneous catalysis, is also often applied.[3] in NP catalysis, at the molecular level, is therefore crucial in order The energy span, which is generally the difference between the to rationalize macroscopic observations and to enable the rational highest and lowest energy points of the PES, integrates both the design of catalysts with tunable activity and selectivity towards thermodynamic and kinetic criteria (Fig. 1a.2) and could, in prin- the desired products. In addition to experimental approaches to ciple, be used as an indicator of the turnover frequency (TOF) and investigate reaction mechanisms and active sites on NP catalysts, thus of the relevance associated to a given reaction route: the lower the recent development of high performance computing and peri- the reaction span is, the higher the TOF should be. *Correspondence: Dr. L. Foppaa E-mail: foppa@inorg.chem.ethz.ch a Department of Chemistry and Applied Biosciences, ETH Zurich, Vladimir-Prelog-Weg 1–5, CH-8093 Zurich b IFP Energies Nouvelles, Rond-Point de l’échangeur de Solaize, BP3, 69360 Solaize, France c Current address: Department of Chemistry, Universitat Autònoma de Barcelona, 08193 Cerdanyola del Vallès, Catalonia, Spain 240 CHIMIA 2019, 73, No. 4 Laureates: Junior Prizes of the sCs faLL Meeting 2018 Nevertheless, it is not straightforward to know how selected energy of surface species obtained from DFT combined with sta- parts of the PES or even how the PES of the full catalytic cycle tistical mechanics for the estimation of Gibbs free-energies (via relates to the overall activity or selectivity of the catalyst, since the partition functions) is used to compute the reaction rates of all actual rates of elementary steps during the reaction depend not the elementary steps of the reaction network at reaction condi- only on the energetics of the process, but also on the concentration tions (temperature and pressure). The gas-phase concentrations of reaction intermediates, which in the case of surface reactions and coverages of adsorbed species are then explicitly determined corresponds to adsorbates populations or coverages (Fig. 1b.1). by solving a system of differential equations describing the evolu- As a consequence, reaction intermediates and TSs whose energy tion of reaction rates in a model reactor, whose boundary condi- actually affect the overall (observed) reaction rate and selectivity tions are the feed compositions (Fig. 2).[5] The interplay of all depend on the relative population of surface species and on the pre- the elementary steps included in the microkinetic model with the ferred reaction routes, which may be hard to infer based solely on reaction conditions determines the composition of adsorbed spe- examining PESs since they are sensitive to reaction conditions, for cies and major reaction pathways, which in turn affect the catalyst instance temperature, pressure and chemical potential of the gas- activity (conversion) and selectivity. phase. In particular, surface sites may be poisoned or inhibited by Contrary to the analysis of the PESs for selected reaction steps, the strong adsorption of reaction intermediates during the reaction, no assumption needs to be made regarding the intermediate(s) suppressing the rate for important reaction steps. Furthermore, or TS(s) that control activity and selectivity when a microkinetic even if PESs are drawn as a linear succession of elementary steps model is available. Instead, sensitivity analyses[6] can be per- occurring in a determined sequence, in most cases several interme- formed to identify surface species whose stability affect the over- diates are part of multiple competing routes within very complex all reaction rate or selectivity. The degree of rate control (DRC) reaction networks (Fig. 1b.2) whose interplay is difficult to assess analysis proposed by Campbell, for instance, identifies rate-deter- from reaction energy diagrams. This may be especially the case mining steps (RDS) based on the evaluation of the degree of rate when multiple reaction sites are available (e.g. NP terraces, defects control quantity (XRC,i)[6b] for each elementary step of the reaction as well as support and NP/support interface sites) and reaction in- network. XRC,i is the relative increase in the overall reaction rate termediates may migrate from one type of site to another. that is obtained if the reaction rate constant for the elementary The use of kinetic models derived from DFT calculations (also step i is increased infinitesimally, while keeping the reaction rate known as microkinetic models) applied to NP catalysis is an al- constants and the thermodynamic equilibrium constants for all the ternative to tackle the complexity of reaction networks under re- other elementary steps fixed. A positive (resp. negative) XRC,i value action conditions, since the contribution of adsorbate population indicates that increasing the rate of the step i results in global reac- is introduced into the calculated rate in addition to the energetic tion rate enhancement (resp. suppression). The elementary step i term provided by DFT and all reactions in the network are consid- is said to be rate-determining (RDS) if XRC,i > 0 or inhibitive if XRC,i ered simultaneously.[4] Within microkinetic models, the electronic < 0, while zero DRC values indicate non-RDSs, i.e. steps whose Fig. 1. (a) Common approaches to address reaction mechanisms and active sites on NP catalysts using DFT calculations: (a.1) evaluation of PESs for selected reaction steps, focusing on the adsorption (Gibbs) free-energy (Gads) of intermediates (adsorbed species denoted by A*) and activation free-energies (Gact) to achieve TSs and (a.2) Evaluation of free-energy span (Gspan) of full catalytic cycles. (b) Limitations of PES analysis and span model: (b.1) the rate of surface reactions depends on adsorbate population (coverage of the reacting species) and (b.2) reaction networks typically display interconnected pathways, with common intermediates. In the figure, G corresponds to Gibbs free-energy, r.c. (reaction coordinate) to an arbitrary reaction coordinate, ∆Sact and ∆Hact to activation entropy and enthalpy, respectively, R to the universal constant of the gases and T to the temperature. Laureates: Junior Prizes of the sCs faLL Meeting 2018 CHIMIA 2019, 73, No. 4 241 Fig. 2. Microkinetic models as a part of a multi-scale approach to link the atomistic (first principles calculations) with the macroscopic (experiments) scales. In the figure, ∆Sact and ∆Hact correspond to activation entropy and enthalpy for a given reaction elementary step, respectively, T to the tem- perature, P to the pressure, feed to the initial composition of reactants, ri to the rate of the elementary surface reaction i, Ai* to the coverage of ad- sorbates taking part in reaction i and R to the universal constant of the gases. infinitesimal rate increase do not impact the global reaction rate. 2. Ni/Al2O3 Interfaces in Water-Gas Shift and Dry Similarly, a degree of thermodynamic rate control XTRC,i can be Reforming of Methane evaluated to identify reaction intermediates whose stability affect The combination of DFT calculations with microkinetic the overall rate.[6d] The DRC analysis is thus equivalent to the models provided a deeper understanding of the WGS (Eqn. introduction of infinitesimal perturbations to the energy of each (1)) and the dry reforming of methane (DRM, Eqn. (2)) reac- TS or intermediate of a given reaction network while monitoring tions catalyzed by Ni NPs supported on alumina. The WGS is changes on the outcome of the reaction, i.e. the conversion, as an important industrial technology for hydrogen production[7] illustrated in Fig. 3. The information obtained from a DRC analy- and the DRM allows the transformation of natural gas into syn- sis is valuable for catalyst design, since it directly shows which gas (CO/H2 mixture) in the industrially-relevant gas-to-liquid intermediate(s) and TS(s) need to be stabilized (or de-stabilized) chain for the production of synthetic fuels,[8] being an inter- for the desired effect on the overall rate and selectivity. Indeed, esting alternative to the commonly used steam reforming of the use of microkinetic models associated to DRC analysis has methane which enables the conversion of CO2 into value-added become standard in catalysts screening studies and also provided products. a broader mechanistic understanding for numerous reactions cata- lyzed by metal NPs as compared to the analysis of PESs alone.[4] CO(g) + H2O(g) = CO2(g) + H2(g) ∆rH° = –41.1kJ.mol–1 (1) CH4(g) + CO2(g) = 2 H2(g) + 2 CO(g) ∆rH° = 247.4 kJ.mol–1 (2) The two reactions are both catalyzed by Ni NPs supported on Al2O3 and present several common intermediates and elementary steps (Fig. 4a), namely the CO2-to-CO (or the reverse, for WGS) transformation.[9] However, because of very diverse thermody- namics, they are performed at significantly different temperatures, 300 and 650 °C for WGS and DRM, respectively, which could affect the preferred reaction mechanisms. In Al2O3-supported Ni NP catalysts, like in the majority of industrial NP catalysts, an oxide material is used to ensure the stability on stream, which provides additional interface sites on the metal/oxide boundary whose role in catalysis is often invoked and consequently highly debated (Fig. 4b).[10] Therefore, a detailed understanding of WGS and DRM mechanisms and especially the active sites (e.g. sites Fig. 3. DRC analysis pictured as infinitesimal perturbations to the Gibbs free-energy of each i-th TS and intermediate (int.) of the reaction net- on the bare Ni NP vs. the Ni/Al2O3 interface) is desirable, but work to evaluate the degree of rate control (XRC,i) and degree of thermo- also challenging due to the complexity of the WGS and DRM dynamic rate control (XTRC,i) quantities, respectively, associated to each reactions, displaying several elementary steps which may all take reaction step. In the figure, G corresponds to Gibbs free-energy and r.c. place on each of the available surface sites and are all potentially to an arbitrary reaction coordinate. rate-determining. 242 CHIMIA 2019, 73, No. 4 Laureates: Junior Prizes of the sCs faLL Meeting 2018 2.1 Evaluation of PES by DFT to Determine the Active Sites The interaction of several molecules involved in the WGS and DRM reaction networks with Ni vs. Ni/Al2O3 interface sites has been evaluated by DFT using Ni cluster models sup- ported on Al2O3 surfaces. Interface sites were shown to strongly adsorb CO[11] or activate H2, CH4,[12] H2O and CO2[13] molecules with lower energy barriers compared to bare Ni sites located on the metal NP surface, presumably making the overall WGS and DRM easier at the interface. In particular, CO2 direct activa- tion, forming CO and O fragments in one elementary step, was shown to be much less energy-demanding on Ni/Al2O3 inter- face sites containing Lewis acidic Al atoms of the (110) γ-Al2O3 surface in the vicinity of the NP compared to bare metal sites on the Ni55 cluster model with ca. 1 nm diameter (PESs shown in Fig. 5a).[13a] Based on these results, both the WGS and DRM reactions could be proposed to occur preferentially via mecha- nisms operating on the metal/oxide interface. The PESs for the full catalytic cycles for the WGS and DRM reactions including the main proposed mechanisms were also evaluated on Ni and Ni/Al2O3 sites by DFT (Fig. 5b), from which the reaction free- energy span could be evaluated.[13b] Whereas the WGS reaction on Ni NP sites involves a free-energy span of 166 kJ.mol–1, the reaction on the Ni/Al2O3 site is associated with 222 kJ.mol–1. In the case of DRM, the free-energy spans correspond to 225 and 246 kJ.mol–1 for Ni NP and Ni/Al2O3 sites respectively. Therefore, based on the span model, the interface should be less active than the Ni sites. This is in contrast to the proposed Fig. 4. (a) Representation of the main WGS (in red) and DRM (in blue) role of the interface in favoring the WGS and DRM reactions reaction mechanisms with common intermediates (CO2*, CO* and O*) based on the easier activation of CO2 and H2O on the interface highlighted. (b) Schematic representation of nanoparticle (Ni) and inter- compared to Ni as shown by the DFT-evaluated PESs, so one face (Ni/Al2O3) sites present in Al2O3-supported Ni NP catalysts. Ni, Al may wonder which of these contradictory interpretations is the and O atoms are shown in blue, magenta and red, respectively, in (b). correct one. Adapted from ref. [13b]. Fig. 5. Atomistic modeling: reaction energy profiles evaluated by DFT on nanoparticle (NP, in blue) and interface (IN, in red) sites present on Al2O3- supported Ni NP models. CO2 adsorption and activation: (a.1) free energy profiles (at 300 °C) and (a.2) structure of the identified TSs. Ni, Al, O and C atoms are shown in blue, magenta, red and gray, respectively. (b) Free energy profiles evaluated at 300 and 650 °C for the full catalytic cycles of WGS and DRM, respectively, with indicated free energy spans. Adapted from refs [13b] and [13a]. Laureates: Junior Prizes of the sCs faLL Meeting 2018 CHIMIA 2019, 73, No. 4 243 2.2 From PES to Formation Rates Using Microkinetic surface reactions describing transformations on the Ni NP as well Modeling as on the Ni/Al2O3 interface. The set of surface reactions contains In order to have a better description of the catalytic process, the most favorable WGS and DRM mechanisms on Ni identified including the effect of the different temperatures of WGS and by previous DFT studies.[9,17] The same model was used for the DRM on reaction mechanisms and preference for metal vs. metal/ evaluation of both WGS and DRM reactions. Further computa- oxide interface active sites as well as the complex reaction net- tional details are available elsewhere.[13b] works, the DFT data was used to construct a mean-field microki- From the steady-state rates of each individual reaction step netic model.[13b] Periodic DFT calculations were carried out with of WGS and DRM reaction networks obtained from the micro- the VASP code,[14] using the PAW method (plane-wave basis set kinetic model (Fig. 6a), the main reaction pathway is associated with pseudopotentials) and the PW91 exchange-correlation func- with the Ni/Al2O3 interface for the WGS, whereas both Ni NP and tional.[15] Kinetic modeling simulations were performed using the Ni/Al2O3 sites are active for the DRM. The DRC analysis shows Chemkin® software[16] and a plug-flow reactor model. Within the that the most RDSs of WGS and DRM are not the activation of mean-field approximation, the system is considered well-mixed, reactants, but the dissociation of adsorbed hydroxyl groups (de- i.e. the reaction rates are calculated from the average coverage noted OH*) and the CH* oxidation reaction with O*, respectively. of surface species considering that they occupy adjacent sites on The surface hydroxyl (OH*) and oxygen (O*) participating in the surface of the catalyst allowing their reaction. The effect of these RDSs are much more available at the Ni/Al2O3 interface due the formation of separated domains of adsorbed species, which to their favorable binding to Al2O3 Lewis acidic Al atoms on the could decrease the probability of finding adsorbates in adjacent perimeter of the Ni NP. During WGS at 300 °C, the NP is highly sites thus decreasing the reaction rate, is not modeled within this poisoned with adsorbed CO* and lacks OH* groups, so that the approximation. The kinetic model is dual-site and includes two presence of interfacial OH* boosts the overall reaction rate at the different surface sites simultaneously, corresponding to the Ni interface. Conversely, during DRM at 650 °C, the coverage of NP surface to the Ni/Al2O3 interface. It contains six adsorption adsorbed CO* on the Ni NP is rather low and O* is readily avail- elementary steps describing the adsorption and desorption of re- able within the metal, making the presence of interfacial O* not actants and product molecules on each of the sites as well as 27 significant to the overall catalytic activity. Fig. 6. (a) Kinetic modeling: net reaction rates for the main individual elementary steps of WGS and DRM reaction networks calculated with a DFT- derived dual-site (nanoparticle and interface) microkinetic model for Al2O3-supported Ni NP catalysts. Thick arrows indicate the preferred reaction channel. WGS reaction conditions: 300 °C, 1 bar, feed composition (molar fraction) 0.10 H2O, 0.10 CO, 0.80 N2. CO conversion: 10%. DRM reaction conditions: 650 °C, 1 bar, feed composition (molar fraction) 0.45 CH4, 0.45 CO2, 0.1 N2; CH4 conversion: 25%. (b) Experiments: (b.1) Measured reac- tion rates as a function of NP diameter (dp) for Al2O3-supported Ni catalysts (logarithm scale) for WGS reaction at 300 °C (CO conversions between 12 and 15%) and DRM reaction at 650 °C (CH4 conversions between 27 and 29 %). The experimental feed compositions are the same as those used in the kinetic model. (b.2) Evolution of rates and fraction of types of exposed metallic sites. Adapted from ref. [13b]. 244 CHIMIA 2019, 73, No. 4 Laureates: Junior Prizes of the sCs faLL Meeting 2018 2.3 Experimental Verification of Active Site and the European Social Fund for a Ramon y Cajal research contract Requirements (RYC-2016-19930). We also acknowledge Prof. Christophe Copéret for If the active sites for WGS and DRM are different, as proposed fruitful discussions. by the microkinetic model, the reaction rates must depend on the proportion of interface sites present in Al2O3-supported Ni NP cata- Received: January 15, 2019 lysts and thus on the NP size in a different manner for each of the reactions. This is because the evolution of the proportion of perim- [1] a) G. A. Somorjai, L. Yimin, ‘Introduction to Surface Chemistry and eter sites at the metal/support interface and surface sites with the NP Catalysis’, Wiley, 1994; b) J. P. Hindermann, G. J. Hutchings, A. diameter is different and decays as ~dp–1.9 and ~dp–0.9, for perimeter Kiennemann, Catal. Rev. 1993, 35, 1. and surface sites, respectively.[10b,18] In fact, γ-Al2O3-supported Ni [2] a) J. K. Norskov, T. Bligaard, J. Rossmeisl, C. H. Christensen, Nat. Chem. catalysts with different NP sizes behave differently in WGS and 2009, 1, 37; b) J. K. Nørskov, F. Abild-Pedersen, F. Studt, T. Bligaard, Proc. DRM reactions when the reaction rates are plotted against the aver- Nat. Acad. Sci. USA 2011, 108, 937. [3] a) L. Falivene, S. M. Kozlov, L. Cavallo, ACS Catal. 2018, 8, 5637; b) S. age diameter of the catalyst particle (Fig. 6b.1).[19] The obtained Kozuch, S. Shaik, Acc. Chem. Res. 2011, 44, 101. dependence of WGS and DRM reaction rates on the NP diameter [4] a) J. A. Dumesic, ‘The Microkinetics of heterogeneous catalysis’, American indicates that WGS is more efficiently promoted by sites residing Chemical Society, Washington, DC, USA, 1993; b) S. Raimondeau, D. G. in the Ni/Al2O3 perimeter whereas DRM is rather related to the total Vlachos, Chem. Eng. J. 2002, 90, 3; c) L. J. Broadbelt, R. Q. Snurr, Appl. Catal., A. 2000, 200, 23; d) P. Stoltze, Prog. Surf. Sci. 2000, 65, 65; e) D. number of exposed metal atoms, since the evolution of the reaction G. Vlachos, AlChE J. 2012, 58, 1314; f) M. Salciccioli, M. Stamatakis, S. rates with the diameter of the catalysts (~dp–0.5 and ~dp–1.4 for WGS Caratzoulas, D. G. Vlachos, Chem. Eng. Sci. 2011, 66, 4319. and DRM respectively) is similar to the evolution of the proportion [5] H. S. Fogler, ‘Elements of chemical reaction engineering’, Prentice-Hall, of perimeter sites and surface sites with the NP diameter (Fig. 6b.2). 1986. Therefore, both experiments and microkinetic modelling confirm [6] a) C. T. Campbell, ACS Catal. 2017, 7, 2770; b) C. T. Campbell, Top. Catal. 1994, 1, 353; c) C. T. Campbell, J. Catal. 2001, 204, 520; d) C. Stegelmann, the different active site requirement in these reactions. A. Andreasen, C. T. Campbell, J. Am. Chem. Soc. 2009, 131, 8077. [7] C. Ratnasamy, J. P. Wagner, Cat. Rev. 2009, 51, 325. 3. Perspectives and Conclusion [8] a) M. De Falco, G. Iaquaniello, G. Centi, ‘CO2: A Valuable Source of The multi-scale approach based in microkinetic models to un- Carbon’, Springer London ; Imprint: Springer, 2013; b) D. Pakhare, J. Spivey, Chem. Soc. Rev. 2014, 43, 7813. derstand catalytic processes catalyzed by metal NPs via the ab initio [9] L. Foppa, M.-C. Silaghi, K. Larmier, A. Comas-Vives, J. Catal. 2016, 343, route presents several approximations that may be severe in some 196. cases and which therefore should be further addressed. Firstly, since [10] a) Z.-J. Zhao, Z. Li, Y. Cui, H. Zhu, W. F. Schneider, W. N. Delgass, F. reaction rates depend exponentially on the enthalpy and entropy of Ribeiro, J. Greeley, J. Catal. 2017, 345, 157; b) M. Shekhar, J. Wang, W.-S. surface reactions (Fig. 1b.1), small energy or entropy differences Lee, W. D. Williams, S. M. Kim, E. A. Stach, J. T. Miller, W. N. Delgass, F. H. Ribeiro, J. Am. Chem. Soc. 2012, 134, 4700; c) S. Aranifard, S. C. will produce large differences in rates. In this respect, the evalua- Ammal, A. Heyden, J. Catal. 2014, 309, 314; d) S. C. Ammal, A. Heyden, tion of energy with DFT may need to be done beyond the typically J. Catal. 2013, 306, 78; e) J. H. Bitter, K. Seshan, J. A. Lercher, J. Catal. used generalized gradient approximation (GGA) for the exchange- 1997, 171, 279; f) J. H. Bitter, K. Seshan, J. A. Lercher, J. Catal. 1998, 176, correlation energy for improved accuracy (e.g. meta-GGA or hy- 93; g) Z. Liu, D. C. Grinter, P. G. Lustemberg, T.-D. Nguyen-Phan, Y. Zhou, S. Luo, I. Waluyo, E. J. Crumlin, D. J. Stacchiola, J. Zhou, J. Carrasco, H. brid functionals).[10a,20] Furthermore, when anharmonic effects or F. Busnengo, M. V. Ganduglia-Pirovano, S. D. Senanayake, J. A. Rodriguez, degrees of freedom other than the vibrational ones (translations, Angew. Chem. 2016, 128, 7581; h) P. G. Lustemberg, P. J. Ramírez, Z. rotations) are relevant for adsorbate entropy, the commonly used Liu, R. A. Gutiérrez, D. G. Grinter, J. Carrasco, S. D. Senanayake, J. A. harmonic approximation may introduce considerable errors.[21] Rodriguez, M. V. Ganduglia-Pirovano, ACS Catal. 2016, 6, 8184. Both these options represent, however, a sensible increase in the [11] a) G. Pacchioni, N. Rösch, Surf. Sci. 1994, 306, 169; b) L. G. V. Briquet, C. R. A. Catlow, S. A. French, J. Phys. Chem. C 2010, 114, 22155. computational cost, which may be prohibitive depending on the [12] J. Li, E. Croiset, L. Ricardez-Sandoval, J. Phys. Chem. C 2013, 117, 16907. size of the model system and complexity of the reaction network. [13] a) M.-C. Silaghi, A. Comas-Vives, C. Copéret, ACS Catal. 2016, 6, 4501; b) Another challenge is the representation of adsorbate–adsorbate L. Foppa, T. Margossian, S. M. Kim, C. Müller, C. Copéret, K. Larmier, A. interactions on highly-covered surfaces[22] and adsorbate gradi- Comas-Vives, J. Am. Chem. Soc. 2017, 139, 17128. [14] a) G. Kresse, J. Hafner, Phys. Rev. B 1994, 49, 14251; b) G. Kresse, J. ents,[23] which can be tackled with the use of models beyond the Hafner, Phys. Rev. B 1993, 47, 558; c) G. Kresse, J. Furthmüller, Comput. mean-field approximation (e.g. kinetic Monte-Carlo). In particular, Mater. Sci. 1996, 6, 15. the energetics of a process occurring at high coverages may not be [15] J. P. Perdew, J. A. Chevary, S. H. Vosko, K. A. Jackson, M. R. Pederson, D. simply extrapolated from low-coverage calculations and may need to J. Singh, C. Fiolhais, Phys. Rev. B 1992, 46, 6671. be more carefully evaluated in some cases (e.g. by using ab initio mo- [16] ‘Chemkin-Pro 15131’, Reaction Design, San Diego, 2013. [17] C. Fan, Y.-A. Zhu, M.-L. Yang, Z.-J. Sui, X.-G. Zhou, D. Chen, Ind. Eng. lecular dynamics methods).[24] Finally, the choice of model system is Chem. Res. 2015, 54, 5901. also a difficult task when numerous facets on metal NPs[25] or metal/ [18] M. Cargnello, V. V. T. Doan-Nguyen, T. R. Gordon, R. E. Diaz, E. A. Stach, oxide interfaces[26] are possible, not to mention NP size effects.[27] R. J. Gorte, P. Fornasiero, C. B. Murray, Science 2013, 341, 771. In spite of such limitations, the use of first principles multi- [19] T. Margossian, K. Larmier, S. M. Kim, F. Krumeich, A. Fedorov, P. Chen, C. R. Müller, C. Copéret, J. Am. Chem. Soc. 2017, 139, 6919. scale models including kinetic analyses, which take into account [20] S. Mallikarjun Sharada, T. Bligaard, A. C. Luntz, G.-J. Kroes, J. K. Nørskov, the reaction conditions and the complexity of intricate reaction J. Phys. Chem. C 2017, 121, 19807. networks, has become a powerful tool to distinguish mechanisms [21] a) M. Jørgensen, H. Grönbeck, J. Phys. Chem. C 2017, 121, 7199; b) G. Piccini, and active sites operating in transition metal NPs, as illustrated M. Alessio, J. Sauer, Angew. Chem. Int. Ed. 2016, 55, 5235; c) L. H. Sprowl, C. here for the case of metal/oxide interfaces in the WGS and DRM T. Campbell, L. Árnadóttir, J. Phys. Chem. C 2016, 120, 9719; d) A. Bajpai, P. Mehta, K. Frey, A. M. Lehmer, W. F. Schneider, ACS Catal. 2018, 8, 1945. reactions catalyzed by Ni NPs supported on Al2O3. Further devel- [22] a) M. Zhou, B. Liu, Ind. Eng. Chem. Res. 2017, 56, 5813; b) A. C. Lausche, opments of computational methods and realistic model systems A. J. Medford, T. S. Khan, Y. Xu, T. Bligaard, F. Abild-Pedersen, J. K. ought to lead to a more comprehensive understanding of catalytic Nørskov, F. Studt, J. Catal. 2013, 307, 275. processes catalyzed by metal NPs via first principles. [23] M. Stamatakis, D. G. Vlachos, ACS Catal. 2012, 2, 2648. [24] L. Foppa, M. Iannuzzi, C. Copéret, A. Comas-Vives, ACS Catal. 2018, 8, 6983. Acknowledgements [25] K. Reuter, C. P. Plaisance, H. Oberhofer, M. Andersen, J. Chem. Phys. 2017, ETH Zürich (research grant ETH42 14-1 and career seed grant 146, 040901. SEED-21 16-2), the Holcim Foundation, and the Swiss competence cen- [26] P. Mehta, J. Greeley, W. N. Delgass, W. F. Schneider, ACS Catal. 2017, 7, ter for energy research (SCCER heat and energy storage) are acknowl- 4707. edged for financial support. A. C.-V. acknowledges the Spanish MEC [27] L. Foppa, C. Copéret, A. Comas-Vives, J. Am. Chem. Soc. 2016, 138, 16655.