Modeling plasma-induced surface charge effects on CO2 activation by single atom catalysts supported on reducible and irreducible metal oxides

The accumulation of negative surface charge on catalytic surfaces in the presence of low-temperature plasma (LTP) could influence catalytic performance. However, it is difficult to disentangle the role of surface charging and other LTP catalytic effects in experiment. Herein, we use density functional theory (DFT) modeling to understand the effect of plasma-induced surface charging on CO2 activation by atomically dispersed single atom (SA) catalysts on both reducible and irreducible metal oxide supports. We model CO2 adsorption strength and CO2 dissociation barriers for Co1, Ni1, Cu1, Rh1, Pd1, and Ag1 SAs on both reducible and irreducible supports, namely, CeO2(100), TiO2(101), and γ-Al2O3(110), to elucidate trends. We find that accumulated surface charge on the SA increases the CO2 adsorption strength and decreases the CO2 dissociation barrier for all studied SA/support combinations. For both charged and uncharged (neutral) systems, SAs on the reducible CeO2(100) support generally adsorb CO2 more weakly compared to when on irreducible supports like γ-Al2O3(110). SAs on γ-Al2O3(110) typically have larger barriers for CO2 dissociation for both charged and uncharged systems compared to TiO2(101) and CeO2(100). The magnitude of surface charging effects on CO2 binding energies and dissociation barriers depends sensitively on both the SA and the support. In some cases, the CO2 activation trends qualitatively change between neutral and charged systems for a fixed SA across different supports. This DFT modeling study demonstrates that surface charging should be considered in strong electric fields because it can have a large effect on molecule adsorption and bond-breaking on catalytic surfaces.


Introduction
Low-temperature plasma (LTP) enhanced catalysis is a growing field of research that has reported a variety of synergistic effects for increased activity or selectivity compared to thermal catalysis, including operating beyond the thermochemical equilibrium limit [1][2][3]. LTP in direct contact with a catalyst or generated upstream from the catalyst promotes conversion of challenging reactions such as N 2 to ammonia [4] and CO 2 to fuels at lower thermal temperatures [5]. The optimal catalyst under LTP conditions is often different than the optimal catalyst for thermocatalytic conditions, opening new materials spaces for exploration [4]. However, mechanistic knowledge of LTP-enhanced catalysis at the atomic level is lacking in many cases. Greater understanding of the LTP effect on the catalyst structure and properties as well as the role of LTP-generated species in the reaction mechanism is needed to guide catalyst design.
Atomistic modeling using quantum mechanics can give insight into LTP-enhanced catalytic phenomena [6][7][8]. The plasma-solid interface involves many phenomena that may affect catalytic performance (e.g. radical species, ions, excited vibrational species, electric field) that atomistic modeling can help understand. Because of the complexity of treating all possible phenomena [1], however, atomistic modeling studies thus far have focused on probing one LTP-based phenomenon at a time. Modeling each LTP/catalyst interaction in isolation helps to assess their relative impact and deconvolute the LTP effects on catalysis. For example, the role of plasmainduced vibrational excitations of N 2 to enhance ammonia synthesis has been studied by density functional theory (DFT) and microkinetic modeling [4]. Vibrational excitations were suggested to greatly enhance N 2 dissociation rates and NH 3 production [4], yet the important role of radical N and H species cannot be excluded [4,7,9,10]. Vibrationally excited states of H 2 and CH 4 were explicitly modeled using molecular dynamics [11], finding that the presence of non-equilibrium vibrational states had a greater impact on catalytic activity for terrace surfaces compared to stepped surfaces. Modeling predicted that radical impingement of plasma-generated atomic N and O species onto Pt films gave increased production of NO compared to the conventional Langmuir-Hinshelwood reaction of adsorbed molecular * N 2 and * O 2 [12]. The role of surface charging has also been examined by atomistic modeling [13,14], but is less explored compared to the effect of vibrational excitations and radical formation.
Plasma impinging onto a catalyst surface causes an accumulation of negative surface charge, which has implications for catalytic performance [15][16][17]. If the surface charging is sufficiently large, catalyst activity and selectivity can be modified from changing the binding energy of molecules (thus changing species coverages) and by increasing intrinsic kinetics by decreasing activation barriers of elementary reaction steps. The nature of the support and catalyst particle size can affect the importance of surface charging on heterogeneous catalysis. For example, DFT modeling was used to probe how plasma-induced surface charging changes the binding energy and activation of CO 2 on atomically dispersed metal ions (i.e. single atom (SA) catalysts) and nanoclusters on metal oxide supports. Specifically, Ti 1 , Cu 1 and Ni 1 SAs on γ-Al 2 O 3 (110) [13], as well as Cu 5 and Ni 5 clusters supported on anatase TiO 2 (101) [14], were studied. Plasma-induced surface charging dramatically increased CO 2 adsorption strengths and decreased activation barriers for * CO 2 dissociation to * CO and * O for both SAs and clusters in comparable magnitudes. It is important to know if these findings are general across many SA catalysts and supports, or if they depend sensitively on the SA/support combination.
Herein, we computationally study various SA catalyst systems and compare charged (LTP-enhanced catalysis) and uncharged (thermal catalysis) systems to elucidate trends across the periodic table with respect to plasma-induced surface charging and catalytic reduction of CO 2 . We study SA catalysts because they typically show high selectivity toward the reverse water-gas shift reaction, CO 2 + H 2 ⇌ CO + H 2 O, relative to nanoparticles [18,19]. We predict how the surface charging of irreducible and reducible supports from LTP can modulate the binding energy and activation of CO 2 and intermediates during CO 2 reduction by SA catalysts. We study six different SAs and three different support materials to span a range of electronic properties for our catalyst and broaden the applicability of our resulting predictions. The six SA catalysts selected are Co 1 , Ni 1 , Cu 1 , Rh 1 , Pd 1 , and Ag 1 because they are common catalysts and systematically vary in d-electronic configuration across the 3d and 4d transition metal series. The three supports studied, i.e. CeO 2 (100), TiO 2 (101), and γ-Al 2 O 3 (110), are selected because they have different levels of reducibility and are broadly used catalyst supports. This collection of systems allows us to build a correlation between CO 2 adsorption and dissociation for charged and uncharged catalysts. We also examine the extent of electron delocalization and charge distribution depending on the reducibility of different supports. We observe the effect of the extra electron charge on the energy of the transition states for CO 2 dissociation relative to the intermediates by comparing the dissociation activation barriers for each system. Insights from these studies could be used to guide selection of single metal atom and support type for LTP-enhanced CO 2 reduction, as well as demonstrate that surface charging should be considered in models of LTP catalysis with strong electric fields.

Methods
DFT calculations were conducted using the CP2K software [20]. An example input file is provided in the Supporting Information. Catalyst models consisted of atomically dispersed SAs (M 1 ) from Groups IX, X, and XI (i.e. Co 1 , Ni 1 , Cu 1 , Rh 1 , Pd 1 , Ag 1 ) on three different metal oxide supports that differ in reducibility, namely, CeO 2 (100), anatase TiO 2 (101), and γ-Al 2 O 3 (110) (from most to least reducible). These chosen surface facets offer stable binding locations for SA [21][22][23]. Supports are periodic in the xy-plane and were arranged as supercells approximately 10 × 10 × 15 Å. The bottom two layers of each support were fixed in the position of the bulk lattice, while the remaining layers could relax during geometry optimization. Starting lattice parameters were taken from experimental data [24][25][26] and then allowed to geometrically relax using the chosen modeling parameters. The double-zeta valence polarized (DZVP) basis sets were used for all atoms, which were optimized from molecular interactions [27], combined with a plane wave basis set with 1200 Ry cutoff. Convergence criteria of inner loop (electronic) and outer loop (ionic) iterations was set to 10 −5 au. The Perdew-Burke-Ernzerhof (PBE) functional with Hubbard + U and D3 dispersion [28] corrections was used for all calculations. For CeO 2 , an effective value of U-J = 5.0 eV for the Ce 4f electrons [29,30], and for TiO 2 an effective U-J = 2.5 eV was used for the Ti 3d electrons [31]. No U correction was used for Al 2 O 3 because this support does not have strongly correlated electrons. The Martyna-Tuckerman Poisson solver was used [32], which allowed for defining periodicity in the xy plane while avoiding a periodic z boundary. This formalism avoided the unphysical self-interaction of vertically stacked slabs and counterions, which would provide unrealistic results for charged simulations [33]. Transition states for neutral systems were found using the Climbing Image Nudged Elastic Band (NEB) method [34] with six intermediate images, and for charged systems we used single point calculations starting from the same transition state geometries found for neutral systems.
Surface charge interactions were implemented similar to that of Bal et al [13], where an extra electron charge was given to the slab surface and was countered by a proton in the vacuum layer for net charge neutrality. The H + counterion was given a null basis set to force the extra electron to associate with the surface rather than the H + itself. The net charge on H + was confirmed to be +1.0 e in each system by Hirshfeld charge analysis. The negative charge was allowed to distribute among the remaining atoms of the slab, and charge localization on the SA was calculated using Hirshfeld charge. The electric field strength was tuned by changing the height of the proton above the surface. For uncharged systems, the vacuum layer above the support surface was ∼25 Å (total cell height of 40 Å). For charged systems, the proton was placed as a counterion 25 Å above the surface, and the total vacuum space is 85 Å (total cell height of 100 Å). This modeling protocol results in a mean electric field strength of 1.58 V Å −1 , and an equivalent surface charge density of −0.14 C m −2 . This field strength is similar to the peak electric field observed experimentally within nanodischarges of porous catalyst materials [35,36].

Results and discussion
Although there have been no experimental studies exploring solely the effect of plasma-induced surface charging on catalyst performance, outside of the plasma catalysis field there is research into tunable surface chemistry by controlling the surface charge of the catalyst through applied voltage and electric field [37,38]. This research may have some analog to our LTP catalysis systems and may help isolate the relative effect that surface charge has outside of other LTP phenomena at plasma-solid interfaces. One experimental study showed significant increases in binding energy (up to 0.62 eV) for isopropyl alcohol on amorphous alumina/graphene/HfO 2 dielectric/p-type Si when inducing a positive 0.162 e charge per active site [39]. With negative charge instead of positive, the impact on adsorption may depend on the adsorbate characteristics and polarity but still have a strengthening effect on adsorption. There are also parallels between LTP-induced surface charging and electrocatalytic studies that model charge on surfaces. Some studies explicitly add electrons to the system similar to our own methodology while operating with canonical Kohn-Sham DFT [40,41], whereas others make use of grand canonical DFT that includes a potential in the quantum mechanical calculations, which effectively induces a surface charge [42,43]. These studies differ in the use of a solvent, either as a dielectric background or with explicit solvent molecules that interact with the electronic field. The charge in electrocatalysis is also a result of an equilibrium reached with the grand canonical potential provided, while LTP is an inherently non-equilibrium state.
To our knowledge, SA catalysis and LTP catalysis have not been studied together experimentally. This study therefore focuses primarily on the difference between systems with and without LTP-induced surface charge, rather than the activity inherent to SAs compared to clusters or larger particles. When examining the rate enhancement by plasma catalysis for CO 2 reduction, it is important to know what role surface charging plays in that enhancement. Herein, we initially discuss our models of the atomically dispersed SA catalysts on metal oxide supports. Next, we analyze the electronic structure and charge of these SA/support systems under surface charging conditions and link these electronic properties to their CO 2 binding energy and CO 2 activation capabilities. We then discuss implications of these results on the field of LTP-enhanced catalysis.
The geometries of likely locations of the SAs on the metal oxide supports are modeled using DFT. The model geometries of each M 1 /support pair are shown in figure 1, including a schematic of a charged system with the H + counterion in the vacuum above the surface ( figure 1(g)). The binding locations are chosen based on a review of stable SA locations proposed under non-plasma conditions [18,21,44,45]. We did not investigate if the plasma-induced surface charge would change the preferred SA binding location because the focus of this work is to directly compare how the charge itself can affect the characteristics of CO 2 binding energy and activation barriers, and therefore we aim to minimize the effects of confounding variables of different SA geometries. SA binding locations are held constant for each of the six different metals to isolate the effects of different geometries within each support material. We model M 1 on CeO 2 (100) in the hollow site between four surface O atoms, approximating a square planar geometry preferred by most metal atoms on ceria step sites (figures 1(a) and (d)) [21]. M 1 on TiO 2 (101) is modeled as bridged between a pair of two-coordinated O atoms, with varying levels of interaction with a three-coordinated O atom beneath depending on the metal identity (figures 1(b) and (e)). For M 1 on Al 2 O 3 (110), the M 1 replaced a surface three-coordinated Al atom (figures 1(c) and (f)), because this motif was predicted to be a preferred geometry for stable SAs in recent studies of γ-Al 2 O 3 (110) [44,45].
We predict the binding energy of CO 2 on each SA/support to determine how the adsorption strength depends on plasmainduced surface charging. All geometries of adsorbates on the SA/support systems have been uploaded to the Novel Materials Discovery database, and the energies are listed in table S1 of the supporting information. CO 2 adsorption is stronger on charged systems (figure 2(a)) compared to uncharged, neutral, systems ( figure 2(b)). Stronger CO 2 adsorption can allow for longer retention times on the surface, and strongly chemisorbed CO 2 often exhibits a bent configuration that is more active for reduction reactions [46,47]. The geometries of bound CO 2 tend to adopt this bent structure when chemisorbed onto the single metal atom, with the O-C-O angle deformation ranging from 23 • to 41 • . Notably, this bent structure is not observed on charged or neutral Ag 1 /CeO 2 or Ag 1 /Al 2 O 3 (bond angle deformation between 5 • and 10 • ), which would indicate a weaker adsorption and less activated CO 2 molecule. Charged and neutral geometries differ in their average bond distances, with CO 2 being bound more closely to the single metal atom when charged (average C-M 1 bond distance is 2.25 Å for neutral vs. 2.09 Å for charged).
A binding energy trend emerges for the different M 1 species that is shared between neutral and charged systems. The binding energy is strongest for SAs in Group IX (Co 1 and Rh 1 ), then decreases going across the periodic table, that is, Co 1 > Ni 1 > Cu 1 , and Rh 1 > Pd 1 > Ag 1 . This binding energy trend is present for both 3d and 4d transition metals. Our predictions follow observed trends seen for CO 2 adsorption on metal surfaces, with noble metals Cu and Ag generally binding CO 2 more weakly compared to less noble metals like Rh [48,49]. This binding energy trend is present for all three supports studied. It is important to note the work by Bal et al that modeled SA Ni 1 and Cu 1 on γ-Al 2 O 3 (110) with and without LTP-induced surface charging [13]. Their findings agree well with our own trends, demonstrating a two or three-fold increase in binding energy of CO 2 when charge is added (mean electric field strength of 1.58 V Å −1 , surface charge density of −0.06 C m −2 ), as well as CO 2 adsorbing more strongly to Ni 1 than Cu 1 in both charged and neutral cases. The value of our predicted CO 2 adsorption is greater than that reported previously for Cu 1 /γ-Al 2 O 3 (110) [13], but the SA placement on γ-Al 2 O 3 (110) in each study differs and our surface charge density is more negative.
The CO 2 binding energy trend across the three supports shows comparable binding energies between M 1 /TiO 2 and M 1 /Al 2 O 3 , whereas M 1 /CeO 2 systems adsorb CO 2 more weakly. This trend can be rationalized by the geometry of the SA coordination environments on each support. The SA on CeO 2 (100) is more fully coordinated, surrounded by four lattice oxygens. In contrast, the SA on TiO 2 (101) and Al 2 O 3 (110) are only coordinated with three lattice oxygens and can more readily chemisorb CO 2 with a stronger bond. The reducibility of CeO 2 could also be playing a role in the SA coordination, causing the negatively charged SA to be bound more strongly to the reducible CeO 2 support and less capable of chemisorbing CO 2 .
To better explain the reported dependence of CO 2 adsorption strength on the accumulated charge in figure 2, we conducted partial density of states (PDOS) and Hirshfeld charge analysis for each catalyst system. The PDOS analysis shows the energy and filling of each orbital with and without the extra charge, giving insight into why the binding energy differs so strongly based on surface charge. The Hirshfeld charge analysis assigns the calculated electron density to each atom center based on its free atom density at the corresponding distance from the atomic nucleus [50]. This analysis estimates the excess charge associated with each atom compared to its neutral state, which elucidates how much extra charge is taken on by the SA when the system is given an extra electron. Here we have performed a Hirshfeld charge analysis on each SA system with CO 2 adsorbed. The Hirshfeld charge on the SA catalyst depends on both the electronic structure of the SA-CO 2 complex and the support material, shown in figure 3(a) (neutral) and figure 3(b) (charged). All values for Hirshfeld charge are listed in the SI in table S1. As expected, the charge on the SA becomes more negative when an excess electron is added, shown by the difference in Hirshfeld charge with and without the excess electron ( figure 3(c)). For neutral systems, each SA carries a negative charge naturally, ranging from −0.4 to −1.0 e. With an extra electron, each adopts a further negative charge, ranging from −0.6 to −1.3 e in total. The Cu 1 and Ag 1 take on the least amount of charge, averaging −0.19 and −0.18 e respectively, compared to the neutral state. This aligns with the noble nature of bulk Cu and Ag due to their fully occupied d-valence shells. The other SAs from Group IX and X take on more negative charge, averaging close to −0.3 e in net difference. There is not a robust qualitative trend based on the support type, with some SAs having little dependence on support (e.g. Ni 1 ) whereas others vary more strongly (e.g. Pd 1 and Ag 1 ). The quantitative values of these predictions depend on the SA binding location and support structure, but the qualitative aspects of this analysis should be quite general.
To better understand the effect of the extra charge on CO 2 binding, we examined the PDOS of the carbon s and p orbitals in * CO 2 in figure 3(d), comparing with and without plasma-induced surface charging. Energies are centered with the Fermi level set to zero. Generally, the charged systems (solid lines) all experience a downshift in energy from the neutral systems (dotted lines), caused by the excess charge resulting in a higher degree of filling for higher energy orbitals. This charging has the effect of lowering the relative energy of the bonding states between the adsorbed * CO 2 and the single metal atom and should result in stronger adsorption of CO 2 to the SA. To support the qualitative description of the shift in orbital energy due to excess charge, we include arrows showing the change in average energy level across the depicted range from −3 eV to 5 eV. Average values were calculated by estimating the area under each curve with equal width rectangles (0.05 eV wide), and weighting the energy by the peak height, shown in equation (1) below: where E i is the energy at interval i and h i is the height at that energy. All arrows show the downshift in energy mentioned previously. For M 1 /CeO 2 , the magnitude of the shift is generally smaller than for M 1 /TiO 2 and M 1 /Al 2 O 3 . The largest shifts occur for Rh 1 /TiO 2 (2.24 eV to 0.52 eV) and Ag 1 /TiO 2 (3.30 eV to 0.74 eV). Figure S1 demonstrates that this change in average energy level due to charging (figure 3(d)) does not correlate directly with the amount of excess charge taken up by the SA from the Hirschfeld analysis ( figure 3(c)), as Rh 1 /TiO 2 shows a large degree of excess charge (-0.53 eV) while Ag 1 /TiO 2 has very little (-0.10 eV).
We investigate the effect of surface charging on the CO 2 activation barrier to dissociate to * CO and * O. The data in  figure 4 shows the CO 2 dissociation activation barriers for each SA/support for both neutral and charged systems. All values for dissociation barriers calculated by NEB are given in table S1. Initial state geometries for NEB calculations are the same as the CO 2 adsorption geometries, and final state geometries were calculated with one O atom dissociating onto a nearby lattice metal atom, and CO remaining bound to the single metal atom. Notably, the charged systems ( figure 4(b)) all have lower CO 2 dissociation barriers than their uncharged counterparts ( figure 4(a)). The difference in dissociation barriers is shown directly in figure 4(c).
In each case, the extra electron decreases the activation barrier for CO 2 dissociation-this lowering of the barrier arises because the surface charging stabilizes the partial C-O bonds present in the transition states relative to the reactant state. Further, the CO 2 dissociation barriers follow a familiar trend from left to right between Group IX, X, and XI, with the Group IX metals having the lowest barriers (average of 0.54 eV for charged Rh 1 , and 0.60 eV for charged Co 1 ) and the noble metals of Group XI having the highest barriers (average of 0.85 eV for charged Ag 1 , and 0.93 eV for charged Cu 1 ) (e.g. average 49% increase from neutral Co 1 to Cu 1 ). This trend is present in both neutral and charged systems, although it is slightly less pronounced in the charged systems (e.g. average 49% increase from neutral Co 1 to Cu 1 , average 38% increase from charged Co 1 to Cu 1 ). The trend with metal identity has an inverse correlation with the CO 2 binding energy and the Hirshfeld charge, which both followed a downward trend in absolute value from Co 1 > Ni 1 > Cu 1 , and Rh 1 > Pd 1 > Ag 1 . This observation supports the hypothesis that a more intense localized charge on the SA would cause a stronger binding energy and lower dissociation barrier. SA catalysts supported on Al 2 O 3 tend to have the highest CO 2 dissociation barriers, whereas CeO 2 and TiO 2 perform similarly to each other and swap between having the lowest or second lowest barrier depending on the metal. The added surface charge does have  a larger effect on the barriers of Group XI metals Cu and Ag, seen in the difference in barriers for the neutral and charged systems ( figure 4(c)).
In the case of CO 2 conversion, a strong CO 2 binding energy is usually beneficial for increasing the overall reaction rate because of higher coverage of * CO 2 to react on the catalyst surface [51]. There have also been connections between CO 2 adsorption strength and CO 2 dissociation barriers through linear scaling relations [52]. The ability to link CO 2 adsorption energy to CO 2 activation and catalyst geometry and electronic structure is valuable to guide catalyst understanding and design. Here we show a correlation between adsorption energy, dissociation barrier, and excess charge on each SA/support in figure 5. The correlation indicates how the effect of charge carries across each SA/support system. Figure 5(a) shows the trend between Hirshfeld charge on the SA and CO 2 binding energy, where more negative charge accumulation correlates with stronger CO 2 adsorption. The trend across metals is most strongly linearly correlated for both the M 1 /TiO 2 and M 1 /Al 2 O 3 systems (both R 2 = 0.53). Ceria supported metals have a smaller slope and are less linearly correlated (R 2 = 0.38). The CO 2 dissociation barrier follows a linear trend with Hirshfeld charge ( figure 5(b)), where a more negative charge on the SA correlates with a lower barrier. M 1 /Al 2 O 3 carries the strongest linear correlation (R 2 = 0.73), whereas M 1 /CeO 2 has the weakest linear correlation (R 2 = 0.45). Because CO 2 binding energy and CO 2 dissociation barrier both correlate with the Hirshfeld charge, they also have a moderate linear correlation with each other, shown in figure 5(c). Systems with stronger CO 2 adsorption also have a lower CO 2 dissociation barrier, as expected from linear scaling relations. From this analysis we confirm that the excess charge strongly affects the CO 2 binding energy and dissociation barrier of supported SA catalysts, and that the amount of charge correlates qualitatively with each of those properties. The strength of the correlation depends on the support properties, with the most reducible CeO 2 having the weakest correlation between adsorption energy, dissociation energy, and charge, and the least reducible Al 2 O 3 support having the strongest correlation. However, many of these properties will depend on the SA binding location and support facet, thus more studies are needed to draw more generalizable conclusions for these correlations.

Conclusions
We performed density functional theory modeling of Co 1 , Ni 1 , Cu 1 , Rh 1 , Pd 1 , and Ag 1 SAs on both reducible and irreducible supports, i.e. CeO 2 (100), TiO 2 (101), and γ-Al 2 O 3 (110) to elucidate how plasma-induced surface charging affects CO 2 adsorption strength trends and is linked to CO 2 dissociation to CO and O. We connect these trends to the PDOS and the reducibility of the support. We find that accumulated surface charge on the SA increases the CO 2 adsorption strength and decreases the CO 2 dissociation barrier for all studied SA/support combinations, consistent with prior computational studies for other SA, nanocluster, and support combinations. We show a correlation between adsorption energy, dissociation barrier, and excess charge on each SA/support, where more negative charge accumulation correlates with stronger CO 2 binding energy and lower CO 2 dissociation barrier. The findings presented herein give a better understanding of how catalyst surface charging could enable manipulation of surface coverages and intrinsic kinetics. We find that it is important to include the effects of surface charging when designing DFT models of plasma systems, and any future multiscale model that tries to predict adsorption behavior for LTP catalysts should consider these effects. Better understanding and control of the electron density at catalyst surfaces will enable manipulation of surface chemistry for optimal rate and selectivity to desired products. We believe that a more detailed characterization of plasma-induced surface charge in situ, while challenging, would greatly benefit the future investigation of plasma catalysis.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://dx.doi.org/10. 17172/NOMAD/2022.11.15-1. Data will be available from 11 February 2023.