Influence of a Cu-Zirconia Interface Structure on CO2 Adsorption and Activation

We have screened different Cu-ZrO2 interface structures and analysed the influence of the interface structure on CO2 binding strength using density functional theory calculations. Our results demonstrate that a Cu nanorod favours one position on both tetragonal and monoclinic ZrO2 surfaces, where the bottom Cu atoms are placed close the lattice oxygens. CO2 prefers a bent bidentate configuration at the interface and the molecule is clearly activated being negatively charged. Altogether, our results highlight that CO2 adsorption and activation depend sensitively on the chemical composition and atomic structure of the interface used in the calculations.


I. INTRODUCTION
2][3] Due to the highly oxidized state, thermodynamic stability, and unreactive nature of CO 2 , economical, active, and selective catalysts are mandatory and the chemical conversion and the economical utilization of CO 2 is a notable scientific and technical challenge. 1merous experimental and computational studies have shown that CO 2 reduction takes place at a metal-oxide interface, [4][5][6][7][8][9][10] which is also an active domain for many other industrially important catalytic reactions 11 such as the water-gas-shift reaction 12,13 and CO oxidation 14,15 just to mention but a few.These reactions have been reported to take place over a variety of metal-oxide interfaces with diverse chemical nature and composition e.g., Au-TiO 2 , [15][16][17] Cu-ZnO, 4,9 Rh-ZrO 2 , 13,18 FeO-Pt, 19 Pd-Co 3 O 4 , 20,21 Pt-SiO 2 22,23 , and others 24 .
As the experimental characterization of interface structures at the atomic level is demanding, density functional theory (DFT) modelling is extensively used to obtain microscopic information about chemical and structural properties of interfaces 4,13,15,17,25,26 and to establish structure-performance relationships 25 .While the catalyst models used in DFT calculations must be firmly based on real catalytic systems, simplifications are mandatory to reduce the computational burden.The key feature of a catalyst to be captured by the model is an active site.Typically, the employed models vary depending on the chemical composition and nature of the active site.For metal-only active sites, periodic surface slab models are commonly used, [27][28][29][30] whereas active sites consisting of a metal-oxide interface are often represented by oxide-supported metal clusters 8,[31][32][33][34] or infinitely long nanorods 13,15,25,26 .
Among the possible chemical transformations, CO 2 conversion to methanol is particularly interesting due to the potential of methanol as a future energy carrier. 2,359][40] Recent studies indicate that a reaction mechanism and selectivity are determined by adsorption energies of key reaction intermediates. 5,8,26,31Two central reaction steps, namely H 2 dissociation 41,42 and CO 2 activation, are strongly associated to the Cu-ZrO 2 interface. 5,38,43The reaction is inferred to be structure sensitive on Cu and the synergy between Cu and support oxides is responsible for enhanced reactivity 4,5,[44][45][46][47][48] .In calculations, however, adsorption characteristics may sensitively depend on a constructed catalyst model, and small differences between relatively similar active sites may introduce large variations in adsorption energies as shown, e.g., for CO adsorption zirconia supported metal clusters. 33Computationally CO 2 conversion to methanol has been recently studied employing a Cu 38 cluster model supported on a m-ZrO 2 (111) surface 8 and a Cu nanorod model, which is composed of three layers of stacked Cu(100) facets on t-ZrO 2 (101) 26 .A similar nanorod model was also used for a Au-MgO interface to address a water-gas-shift reaction. 12In another computational study, the catalytic properties of various metal-MgO interfaces were considered using rods that also consist of (100) terminated slices but cut in a different orientation to better match the symmetry of MgO(100). 25CO 2 reduction on SiO 2 and TiO 2 supported Pt was, in turn, investigated using a Pt 25 cluster, also composed of stacked (100) facets and exposing (111) microfacets towards the interface. 31A differently shaped Rh nanorod, terminated by (111) facets on each side was employed for a water-gas-shift reaction on ZrO 2 13 and a similar model was used for Au-TiO 2 to address CO oxidation 15 as well as low temperature H 2 oxidation 49 .As one would expect, the symmetry and periodicity of the underlying support oxide naturally influences the interface and must therefore be considered.For highly symmetric oxide surfaces, such as MgO(100), the orientation of a deposited nanorod has a vanishingly small impact on the interface.For less symmetric oxides, the position and orientation of the nanorod can substantially change the interface structure.For example, a monoclinic ZrO 2 (111) surface displays a less symmetric crystal structure than a tetragonal ZrO 2 (101) surface.One more feature that may affect interface reactivity is strain effects 25 , which together with defects in catalytic metal particles have been suggested to significantly impact on catalytic efficiency. 25,50,51This is hardly surprising as it is well established from the numerous studies that straining of metal surfaces changes their reactivity observed as shifts in d-band centers. 30,52Artificial strain effects may emerge in the construction of atomic models for metal -oxide interfaces, because the lattice mismatch between metal and oxide will introduce strain along the nanorod as it has to meet the periodicity set for the support surface.
In this study, we focus on CO 2 adsorption and activation on various Cu-ZrO 2 interface structures to shed light on the influence of the interface properties on CO 2 binding.
Specifically, we considered two Cu nanorod models with different geometries as well as both tetragonal (t-ZrO 2 (101)) and monoclinic (m-ZrO 2 (111)) zirconia surfaces.A comprehensive screening of nanorod positions was performed on the oxide and CO 2 characteristics were analysed for differently strained interfaces.The adsorption process was broken down to distinct electronic contributions, and they were used to attempt to establish general trends between interface structure and its ability to adsorb and activate CO 2 .

II. COMPUTATIONAL METHODS
All density functional theory (DFT) calculations were carried out using the BEEF-vdW exchange-correlation functional 53 in the projector-augmented wave (PAW) 54 formalism as implemented in the GPAW 55 package.The core electrons of all elements were described by PAW setups in the frozen-core approximation.A maximum spacing of 0.18 Å was used for the real-space grid basis, and the reciprocal space was sampled at the Γ point.Periodic boundary conditions were used in two directions.A Hubbard U correction 56 of 2.0 eV was applied to the d-orbitals of the zirconium atoms.The geometry optimizations were performed using the BFGS algorithm as implemented in the Atomic Simulation Environment (ASE). 57The computed electronic structures were analyzed by the Bader partitioning method 58 using code written by Tang et al. 59 to obtain the distribution of partial charges on individual atoms.The density of states was analyzed to locate the d-band centers for the purpose of investigating their importance to the reactivity of the Cu nanorods according to the d-band model. 60e interface models were built by placing a Cu nanorod over the most stable facets of monoclinic and tetragonal zirconia surfaces, m-ZrO 2 (111) and t-ZrO 2 (101), as supports (see Fig. 1 for an overview).We adopted two Cu nanorod models, similar to those that have been used in previous publications, 13,25,26   are (111) and (100) of which ( 111) is slightly more stable. 61

A. Model system screening
Figure 1 illustrates the catalytic model systems for both nanorod models and supports.
Screening of the rod position was carried out by displacing the nanorods in two directions on the support: along and perpendicular to the nanorod, while keeping the orientation fixed.
To scan the surface, we used a grid of 0.7 Å steps along the nanorod axis, and 1.1 Å or 1.5 Å steps perpendicular to the nanorod for the t-ZrO 2 or m-ZrO 2 , respectively.The CO 2 binding was studied by attaching the molecule to one of the bottom-row Cu atoms at the reactive interface.
The oxide support was described by a slab model, the thickness of which was set to two stoichiometric layers.This approximation is necessary to reduce the computational burden especially in the larger cells (see below).We consider the two-layer model sufficient to reveal general trends when comparing the nanorod models, and justified, since in preliminary evaluations with two to four layers, the Cu-zirconia binding energy showed only minor variation.We also determined that the d-band centers of the interfacial copper atoms varied insignificantly between different slab thicknesses.However, the CO 2 adsorption energy depends on slab thickness and Test calculations showed that a thicker slab would enhance the binding.In geometry optimizations, the bottom layer of the zirconia slab was kept frozen to its initial bulk geometry, while the top zirconia layer, the Cu nanorod, and the possible CO 2 adsorbate were allowed to relax until the maximum residual force was reduced below 0.02 eV Å−1 .

B. Strain of the nanorod
With the present DFT model, we obtain a bulk Cu lattice constant of 3.69 Å which leads to 2.61 Å nearest-neighbor Cu-Cu distance prior to modifications.The nanorods were created by repeating periodic Cu 8 minimum nanorod units having length of one atom.Our computationally determined zirconia lattice constants can be found from Table S1.To explore possible implications of the artificial strain on the metal-oxide interface, we studied five different nanorod lengths on tetragonal ZrO 2 and three on monoclinic ZrO 2 surfaces, see Table S2 for numerical details of the cell sizes.
Because the calculations have to be periodic along the length of the rod, the Cu-Cu distance modified in that direction was always adjusted accordingly to meet the periodicity of the given surface unit cell.We define the strain as positive when the nanorod is stretched and negative when it is compressed relative to the computationally optimized bulk Cu lattice constant.The nanorods illustrated in Fig. 1 are those with the lowest strains, with t-ZrO 2 support producing a −0.72 % and m-ZrO 2 a −1.02 % strain.Overall, the strain varies from −7.3 to +8.1 % between the different surface models investigated.

C. Energy decomposition
The energy decomposition was set to characterize different contributions in the adsorption of CO 2 at the catalytic sites.First, the adsorption energy of CO 2 was computed from the total energies as where E CO 2 /Cu/ZrO 2 stands for the full system, E Cu/ZrO 2 for the bare ZrO 2 -supported Cu nanorod, and E CO 2 for the gas-phase (linear, inactivated) CO 2 .The gas-phase reference was computed in a large non-periodic cell.We will use the adsorption energy difference ∆∆E ads to compare the different interface model systems.Then, to exclude the contribution of atomic relaxations from the above CO 2 -Cu/ZrO 2 bond strength, we computed the total electronic interaction energy as Here, asterisks stand for reference configurations, where all the atomic positions were fixed to those optimized for the full system.The adsorption and electronic interaction energies in Eqs. ( 1) and ( 2) thus differ by an energy penalty required by the deformation of the catalyst and the CO 2 molecule upon adsorption.Again, for comparison, we define ∆∆E tot el to assess the difference between the models.
The electronic interaction energies between the CO 2 molecule and the catalyst components Cu and ZrO 2 were separated according to Again, the atomic coordinates of the CO 2 and the isolated Cu nanorod and ZrO 2 were fixed to those optimized for the full system.Summing Eqs. ( 3) and ( 4) together accounts for the pairwise contributions to the three-body interaction in Eq. ( 2).Finally, to account for the missing contribution to the total electronic interaction, we define an excess energy which describes the change in electronic interaction energy due to the synergy effect of the metal-oxide interface.

III. RESULTS AND DISCUSSION
A. Binding of the minimum-strain Cu nanorods on ZrO 2 supports To evaluate how the structure and position of a nanorod on a zirconia support influences the catalytic properties of the formed interface, we carefully analyzed the binding of the two nanorods by screening their positions on the zirconia.The heat-maps given in Fig. 2 show how the binding energy of a nanorod depends on its position on the surface.Specifically, we plot the relative energies of both ( 100) and ( 111) models with respect to their most stable structures on both tetragonal (a and b) and monoclinic (c and d) supports.We find that the variation of the nanorod position along its length has only a minor influence on the binding energy whereas moving the nanorod in the perpendicular direction across the zirconia surface introduces substantial energy changes.On t-ZrO 2 , the energy difference between the most stable and the least stable nanorod position is 1.4 eV for the (100) model and 2.0 eV for the (111) model.On m-ZrO 2 the corresponding value is 2.2 eV for both models.
We ascribe the large positional effect on the energy to the strong interaction between the bottom Cu atoms and the surface O anions.more stable than the (100) model.On t-ZrO 2 , the relative stability of the models differ by 0.11 eV per nanorod unit, while on m-ZrO 2 the difference as large as 0.32 eV.In the gas-phase, the (111) model is 0.33 eV/nanorod unit more stable than the (100) model.
However, as this energy difference decreases on the tetragonal zirocnia surface, it implies that the interaction with the support stabilises particularly the (100) Cu nanorod.We link the surface-specific stabilization effect to the Cu-O distances, which are shorter on tetragonal zirconia than on monoclinic zirconia, see Fig. S4 a) and S5.The relative stabilities of (100) and (111) nanorods on zirconia surfaces are 0.37 eV/nanorod unit and 0.24 eV/nanorod unit, respectively and they further demonstrate the ability of the tetragonal surface to better stabilise Cu than the monoclinic surface.Even when considering the adsorption of a single Cu atom, we see that adsorption is about 0.3 eV more exothermic on tetragonal zirconia.
These are in line with a previous experimental study 39 , which reports stronger Cu-ZrO 2 interactions on tetragonal zirconia than on monoclinic.To gain more detail understanding for the energy variation seen in the heat map plots and to link the binding energy to the microscopic structure, we plotted the energy change with respect to the most stable (111)-

B. Strain effects on nanorod binding
A less optimal oxide surface cell size in calculations can introduce artificial strain effects on the nanorod and these, in turn, can influence the computed adsorption characteristics.
Therefore, we analyzed strain effects more closely, for computational details see Table S2.
First, the position of the differently strained nanorods were screened over both zirconia surfaces.We find that the preferred nanorod positions are almost identical to those given in Figs. 1, S2, and S3 for the ideal interfaces, and thus we limit our study on the most stable nanorod positions.In particular, we aim to understand changes in nanorod binding energies due to strain and elucidate the difference between the tetragonal and monoclinic ZrO 2 surfaces with special emphasis on the tetragonal surface.
The last column in Table I shows that the total binding energy (∆E b ) is more exothermic for the (100) model and the tetragonal surface than for the (111) model and the monoclinic surface, but it follows no obvious trend.To understand the origin of this variation better, the thermodynamic cycle (see schematic Figure 3) was devised to analyse the different components.The cycle divides the Cu-ZrO 2 binding energy into four contributions.The first three steps constitute the changes in the gas-phase nanorod and the last one measures the pure electronic interaction with zirconia.
Step 1 describes the change for the optimisation of interatomic Cu-Cu distances whereas step 2 gives the energy change originating from strain effects and step 3 represents the energy change due to structure deformation arising from the interaction between the nanorod and zirconia.The remaining contribution, step 4, defines the pure electronic binding interaction between the deformed Cu nanorod and zirconia.
The energy changes accompanying the thermodynamic steps are collected in Table I.
Overall the contributions from steps 1-3 are small compared to step 4, which dominates.
The slight energy decrease seen in step 1 for all the models indicates that the Cu bulk lattice constant is not optimal for the gas-phase nanorod.In fact, the Cu-Cu distance decreases from the bulk value 2.61 Å to 2.58 Å for both nanorod models on m-ZrO2 2 , and to 2.55 (2.57) Å for the (100) model ((111) model) on t-ZrO2 2 .Other minor variations in Cu-Cu distances and energies in step 1 originate from the differences in the computational cell shapes.As expected, increasing strain leads to positive (endothermic) energy change in step 2 as mainly also does nanorod deformation in step 3.While irregularities introduced by the support make it difficult to predict clear trends, the larger strain is accompanied by stronger deformation and this is especially clear for the two most strained (100) rods.
Cu nanorod binding energy on zirconia is dominated by electronic interaction (step 4), which is more exothermic for t-ZrO 2 than for m-ZrO 2 resulting most likely from the shorter Cu-O distance as suggested also for the minimum strain interfaces.As illustrated in Fig. S6, the introduced strain correlates well with the electronic interaction energy, where the compression of the nanorod decreases the electronic interaction between Cu and zirconia, while the expansion of the nanorod enhances it.No clear correlation is seen when attempting to link binding energies to strain.Multiple factors may contribute to this, the main reason being the irregular structural deformations during structure optimization.The d-band-center analysis supports the stronger interaction between Cu and tetragonal zirconia as the shifts in the d-band center are larger for tetragonal than monoclinic surfaces.The net shift, see Table S3, is to lower energies and it is dominated by the "ligand" effect introduced by the oxide, whereas the "strain" effect, including deformation, constitutes minor contribution.

C. CO 2 adsorption and activation
In order to estimate the significance of an interface site to CO 2 adsorption and activation, we conducted a thorough screening of the available sites for the structures explored in section III A. We find that the CO 2 adsorbs preferably in a bent configuration for the minimumstrain interface structures as shown in Figs. 4 and 5. Our results also demonstrate that adsorption energies depend sensitively on the nanorod model and the interface structure.
Figure 4 displays the minimum-strain Cu-t-ZrO 2 interface for which the most exothermic CO 2 adsorption energies were computed to be −1.01 and −0.52 eV for the (100) and (111) models, respectively.The adsorption energies for other interfacial sites along these interfaces can be found from Table S4.A previous DFT study 26 reports as large as a −1.78 eV adsorption energy for the CO 2 with the (100) model.We associate the large difference with two factors, firstly a +0.41 eV gas-phase correction was applied to CO 2 , 26 , and secondly, to the fact that the employed interface model has substantial, 5.08 %, strain, which strongly impacts CO 2 adsorption as discussed below.where the CO 2 is at the interface but binds only to the Cu atoms.
The most stable CO 2 adsorption geometries are structurally similar for both zirconia surfaces and nanorod models.Interaction between CO 2 and the interface introduces structural deformations to the Cu nanorods.While in the case of the t-ZrO 2 surface, the distortion of Cu edge is minor, on m-ZrO 2 CO 2 clearly pulls out one Cu atom from the both nanorods, see Fig. 5.
Table II summarizes the energy contributions defined in Eqs. ( 1)-( 5) and displays the origin of the variation of CO 2 adsorption energies from one minimum-strain interface to the other.We first focus on electronic interaction energies, which exclude all the structural deformations.For example, the adsorption energy difference (∆∆E ads ) between the (100) and (111) models is 0.49 eV for the tetragonal surface, whereas the pure electronic, aka binding, interaction energy difference, ∆∆E tot el , shows a larger, 0.75 eV, value.Interestingly, for the monoclinic surface ∆∆E ads is larger than ∆∆E tot el (0.9 vs 0.5 eV).The opposite behaviour is ascribed to nonidentical atomic relaxations for different CO 2 -interface systems.In general, we attribute the observed adsorption energy differences to different electronic interaction energies, which are clearly more exothermic for the (100) model on both tetragonal and monoclinic zirconia.The interaction energies from the CO 2 -Cu and CO 2 -ZrO 2 subsystems, which measure the binding between CO 2 and metal and CO 2 and oxide, do not indicate that one nanorod model or zirconia crystal structure is favoured over the other.The substantial synergy between the Cu and zirconia is demonstrated by the exothermic values of ∆E exc for all the studied system.
Next, a CO 2 adsorption site at the (111)-ZrO 2 interface is further explored by moving the nanorod across the oxide surfaces.Fig. 6 displays the heat maps summarizing the energy variation with respect to the most stable adsorption structure at the interface.The plots show, that for many nanorod positions, CO 2 does not either adsorb at all or adsorption is energetically very unfavourable.In order to interpret the heat map information, we analyzed the calculated CO 2 adsorption structures.For a favourable CO 2 adsorption, it seems to be crucial to have lattice Zr cations sufficiently close to the Cu edge to ensure that the oxygens of the CO 2 molecule can interact with them.The nanorod positions, where surface anions are closer to the Cu edge than the Zr cations lead to unfavourable CO 2 adsorption due to repulsive interactions between the oxygen atoms and anions.The central role of the zirconia support is further highlighted by the fact that, without the support, CO 2 only physisorbs maintaining a linear structure on a Cu(111) surface as well as on our (111) and (100) nanorods.This agrees well with the previous DFT results, which also demonstrate  CO 2 physisorption on Cu(111) and Cu(533) surfaces. 28,37 2 adsorption on the bare zirconia surface leads to carbonate (CO 3 ) formation with the lattice oxygen.We find this process exothermic by −0.57eV on t-ZrO 2 and by −0.59 eV onm-ZrO 2 .In the case of a (111) model, the carbonate formation is thermodynamically slightly more favourable than CO 2 adsorption at the interface, which makes these two simultaneous reactions competing.On the other hand, CO 2 adsorption at the interface is clearly preferred to carbonate formation for the (100) model.Previously computed 62 carbonate formation energy on the same monoclinic zirconia surface is about −1.13 eV, which is substantially more exothermic than the value reported here.We ascribe this energy difference to the different exchange and correlation functional and to the thicker zirconia slab, not computationally feasible for the present screening study.Note that under the reaction conditions, zirconia is partially covered by OH groups, which are known to react with CO 2 to create an extremely stable formate species.6,13,62,63 We close the discussion on CO 2 adsorption and activation by considering that at the strained metal-oxide interfaces.Similar to the minimum-strained nanorods, CO 2 is acti- vated via electron transfer and the molecule adopts a bent adsorption configuration.The bidentate binding is preferred, while monodentate geometries also appear at slightly higher energies.CO 2 adsorption energies and the electronic interaction energy components for the strained interfaces are presented in Table II.While the data does not allow to make comprehensive conclusions for all the interfaces, we can say that for the (111) model the straining and compressing of the nanorod leads to more exothermic electronic interaction energies.
However, one has to be careful with the nanorods exposed to larger (± > 7%) strain, as they experienced significant structural deformations during the atomic structure optimization, which reduced the atomic coordination number of some Cu atoms, typically those interacting with the C atom leading to enhanced CO 2 adsorption.Therefore, these systems are largely omitted from the detailed discussion.In general, adsorption and total electronic interaction energies do not correlate.This means that adsorption energy difference and total electronic interaction energy difference, ∆∆E ads and ∆∆E tot el , differ for a considered system pair.Again, this can be attributed to diverse deformations of the nanorod and the molecule giving system-specific positive deformation energies.In addition, we see for all the systems strong synergy between metal and oxide, which is reflected by exothermic values of ∆E exc and highlights the unique nature of the interface.Table II also shows that interfaces built from the (100) nanorod give more exothermic CO 2 adsorption energies than its (111) counterpart.
In order to understand the variation of CO 2 adsorption energy from one system to the other, we performed d-band center analysis separately to each steps in the thermodynamic cycle.The formation of a metal-oxide interface has been considered as a two-step process 25 including the "strain" and "ligand" contributions similar to bimetallic systems. 64The "strain effects include both changes in Cu-Cu distances along the nanorod as well as structural deformation of the nanorod due to interaction with zirconia whereas the ligand effect describes the electronic interactions between Cu and zirconia.The first three steps in the thermodynamic cycle contribute to the shift of the d-band center due to strain effects ∆ strain d while the shift of the d-band center for step 4 (∆ ligand d ) measures the change resulting from the ligand effect.The overall impact of the support to the Cu nanorod is a clear shift of a dband center to lower energies for all the models, see Table S3 explicit numerical values.The ligand effects clearly dominate and the negative value of ∆ ligand d for all the interface models highlight that binding interaction between the Cu and zirconia shifts the d-band center to lower energies.The contribution of the strain effects to the shift of the d-band center is negligible being positive for some interfaces and negative for some others.Table S3 shows that CO 2 adsorption energies can not be rationalized with the shift of the d-band center as no correlation can be established between the adsorption energy and the total shift in the d-band center.We believe that the complex metal-oxide interaction effects at the interface together with structural deformations, especially in the case of the less stable (100) rod, make the d-band model insufficient to explain CO 2 adsorption energies.
Altogether, our DFT results highlight that CO 2 adsorption and activation depends sensitively on the atomic structure and composition making only few site geometries favourable for CO 2 .Moreover, care must be taken when building computational interface models as artificial strain enhances CO 2 adsorption and similar effects might be present for reaction intermediates as well, let alone that the possible strain effects may affect the activation barriers for elementary steps taking place at the interface.Structure sensitivity of CO 2 adsorption suggests that not all the Cu-ZrO 2 interfaces at real-world catalytic systems are active towards CO 2 chemistry.This is because supported nanoparticles present various interface sites with different composition and atomic structure and their direct structural optimization is infeasible.

IV. CONCLUSIONS
We have investigated the properties of a Cu-ZrO 2 interface and its ability to adsorb and activate CO 2 using density functional theory calculations.Specifically, two Cu nanorod models were explored on m-ZrO 2 (111), and t-ZrO 2 (101) surfaces.We observed that the (111) nanorod model is always more stable than the (100) one regardless of whether it is supported by zirconia or not.Tetragonal ZrO 2 stabilises both nanorod models more than the monoclinic ZrO 2 , which is likely due to the more exposed oxygen anions of the t-ZrO 2 (101) surface.Our calculations demonstrate that the stability of the nanorod depends sensitively on its local chemical environment on ZrO 2 and results from the fact that Cu atoms avoid the interaction with surface cations and prefer to minimize a Cu-anion nearest neighbour distance.Compression along the nanorod enhances binding to the zirconia while tension of the nanorod weakens the interaction with the studied oxide surfaces.
Our results demonstrate that the employed Cu-ZrO 2 interface model significantly impacts the adsorption characteristics of the CO 2 molecule.In general, the interfaces built using the (100) nanorod adsorb CO 2 more strongly compared to the interfaces created with the (111) nanorods.The activation of CO 2 is seen as increased Bader charge and a bent adsorption configuration.Applying strain to the nanorod enhances electronic interaction with the CO 2 , which is not always reflected in more exothermic adsorption energies due to structural deformation effects.The excess interaction energy originating from the synergy between the metal and the support is strongly exothermic for the all studied systems, highlighting the importance of the meal-oxide interface.Furthermore, depending on how the nanorod is positioned on the ZrO 2 , CO 2 might not adsorb at the interface at all.Overall, not only the chemical composition but also the diverse structural features of an interface can impact the adsorption characteristics of reacting molecules and consequently the computed activity and selectivity profiles.Therefore, when building computational models for catalytic reactions taking place at the metal-oxide interface, care should be taken in constructing interfaces and identifying interfacial active sites.TABLE S3.The shifts of the d-band centers computed for interface Cu atoms at the differently strained nanorod for each step in the thermodynamic cycle.For each case, the shift ∆ Sx d is calculated with respect to the gas-phase Cu nanorod at the initial state of the the given step in the cycle, Sx refers to to the particular step.The sum of ∆ Sx d for steps from 1 to 3 corresponds to the shift of the d-band center introduced by the strain effects while the ∆ S4 d term stands for the shift of the d-band center due to the ligand effect introduced by the support.The positive (negative) values refer to energy change to higher (lower energies).All the energy changes are given in eV   download file view on ChemRxiv Gell_supplementary_material.pdf (47.22 MiB)

ZrO 2
structure versus the average minimum Cu-O distance between the bottom Cu atoms and topmost surface anions for different nanorod positions over both surfaces.Figs.S4 a and S5 clearly display that the (111) model is more stable than the (100) model and for both nanorod models the shorter distance corresponds to more stable structure.Furthermore,the average distance of the entire Cu nanorod (bottom) from the zirconia is 2.5 Å on t-ZrO 2 and 3.1 Å on m-ZrO 2 .The shorter average C-O distances and shorter nanorod-oxide distance can be explained with more exposed and symmetrically sitting lattice oxygens on the tetragonal surface.The optimization of bottom Cu atom positions with respect to surface oxygens is more challenging on the monoclinic surface owing to the larger asymmetry the surface and the fact that the surface anions are located deeper in the topmost surface layer.Additionally, we note that structural deformations in Cu nanorods originating from the interaction with the zirconia support show no correlation with the relative energy of the system as can be seen from FigureS4 b).

FIG. 3 .
FIG. 3. A thermodynamic cycle for the binding energy decomposition for a Cu nanorod on ZrO 2 .

FIG. 4 .
FIG. 4. Front and side views of the most stable CO 2 adsorption geometry at the minimum-strain Cu-t-ZrO 2 interfaces for the (100) model (panels a and c) and the (111) model (b and d).

FIG. 5 .
FIG. 5. Front and side views of the most stable CO 2 adsorption geometry at the minimum-strain Cu-m-ZrO 2 interfaces for the (100) model (panels a and c) and the (111) model (b and d).

Figure 5
illustrates the CO 2 adsorption on the Cu-m-ZrO 2 interface, where the computed CO 2 adsorption energies are −1.38 eV for the (100) nanorod and −0.44 eV for the (111) nanorod.For a Cu 38 -m-ZrO 2 (111) interface, CO 2 absorption energy is −1.86 eV 8 being substantially more exothermic than what we found in this work.This time, the observed adsorption energy difference is attributed on one hand to the different exchange and correlation functional, and on the other hand to the 38-atom cluster geometry.In another computational work, an adsorption energy of −0.69 eV was reported for a slightly thinner (111)-type Cu nanorod on a stepped m-ZrO 2 (212) surface, 7 CO 2 prefers a bidendate geometry, i.e., it binds to the Cu via the carbon atom and has both oxygen atoms pointing down towards the support cations with the O-C-O angle being close to 120 • .A substantial deviation from the linear gasphase structure and a slight, 0.14 Å, elongation of the C-O bonds clearly indicate CO 2 activation.This is further supported by the Bader charge analysis showing that the CO 2 molecule gains 1.2 |e| upon adsorption, which is in line with the previously reported value. 8

FIG. 6 .
FIG. 6.Heat maps for CO 2 adsorption upon varying the position of the (111) nanorod across the a) t-ZrO 2 and b) m-ZrO 2 surfaces.The nanorod lies parallel to the vertical axis and is moved along the horizontal axis.The energy difference is given with respect to the most stable adsorption structure.The vertical axis numbers represent Cu edge atoms to which CO 2 was attempted to bind.Gray color corresponds to the nanorod positions, which do not bind or activate CO 2 .

FIG. S4 .
FIG. S4.Structural analysis of the Cu (111) (blue) and (100) (orange) nanorods on m-ZrO 2 where each dot represents a structure from the heatmaps c) and d) in Fig. 2 : a) the average nearest neighbour distance for the bottom layer of each Cu nanorod to the closest O atom is shown together with the relative energy with respect to the minimum energy structure of the 111 nanorod, b) the total deformation of the Cu nanorod structure is displayed together with its energy.The deformation is the sum of the distance difference of each Cu atom in the nanorod to its initial strained bulk position.

TABLE I .
Nanorod binding energies (in eV) at each stage of the thermodynamic cycle shown in Fig.3.∆E b stands for binding energy of a nanorod to the zirconia surface and step5 in the thermodynamic cycle.To facilitate comparison, the energies are divided by the number of rod units in the corresponding cell.

TABLE II .
CO 2 adsorption energies (∆E ads ) and electronic interaction energies (∆E tot el , ∆E Cu el , ∆E Zr el , and ∆E exc ), defined in Eqs.(2)-(5) for different models and strain values.All the energy values are given in eV.

TABLE S1 .
Computationally determined lattice constants for bulk zirconia and the dimensions of the repeating units of the surface slabs

TABLE S2 .
Combinations of slab sizes (number of repeating units) and Cu nanorod lengths chosen to introduce the strains effect.The exact cell dimensions were chosen so that the ZrO 2 would not

TABLE S5 .
CO 2 adsorption energies at the interface Cu sites for differently stressed (111) nanorods