Surface Activity of Early Transition-Metal Oxycarbides: CO2 Adsorption Case Study

Theoretical studies and experiments have suggested that transition-metal carbides (TMCs) can be useful materials for carbon capture and storage or usage technologies from air sources. However, TMCs are known to become easily oxidized in the presence of molecular oxygen, and their properties jeopardized while being transformed into transition-metal oxycarbides (TMOCs), which can affect the TMCs’ chemical activity, for example, towards CO2. Here, by means of density functional theory (DFT) based calculations including dispersion, we address the possible effect of oxycarbide formation in the CO2 capture course. A careful analysis of different models shows that for group 4 TMCs (TM = Ti, Zr, Hf), their oxidation into TMOCs involves a negligible structural distortion of the outermost oxide surface layer, whereas severe rumplings are predicted for group 5 and 6 TMOCs (TM = V, Nb, Ta, Mo). The large surface distortion in the latter TMOCs results in a weak interaction with CO2, with adsorption energies below −0.2...


INTRODUCTION
The chemistry of CO 2 when adsorbed on a materials surface is nowadays an active field of research, 1,2 highly spurred by the urgent necessity to find candidate materials able to efficiently capture CO 2 , with the ultimate purpose of climate change mitigation. 3 Such materials must enable the implementation of CO 2 capture and storage (CSS) technologies, 4,5 and when possible, the CO 2 conversion into other greener added-value chemicals, through the so-called CO 2 capture and usage (CCU) processes. However, because of the high chemical stability of CO 2 , only a few selected privileged materials are able to strongly enough adsorb CO 2 so that its capture is efficient under standard conditions of temperature and CO 2 partial pressure. Furthermore, CCU processes normally involve molecular activation, achievable via electron transfer from the substrate, leading to a bent CO 2 molecule with elongated, weakened C-O bonds, 1 which severely restricts the number of potential materials for this purpose.
Transition metal carbides (TMCs) have been recently adverted as potential materials for CCS and CCU technologies, with promising results based on density functional theory (DFT) simulations coupled to the transition state theory (TST) framework. These allowed gaining reasonable estimates of adsorption and desorption rates, and by so, of the practical CO 2 partial pressure and temperature windows. 6 Earlier DFT results pointed at CO 2 activation on molybdenum carbides on different stable surfaces, 7,8 where experimental results on different molybdenum carbide phases 8-10 revealed CO 2 dissociation at room temperature, together to a catalytic CO 2 hydrogenation conversion at elevated temperatures. This indirectly already evidenced the formation of surface activated CO 2 moieties. The prognosticated CO 2 adsorption and activation on TMCs most stable (001) surfaces 6 has been recently conceptproven by X-ray photoemission spectroscopies (XPS) based on the outcome of DFT simulated surface science techniques. 11 Note in passing by that two-dimensional early transition metal carbides, known as MXenes, have been likewise recently forecasted as appealing materials for CCS and CCU, 12 and the CO 2 selective capture in N 2 /CO 2 mixtures recently experimentally confirmed. 13 Despite the promising CCS/CCU features of most stable (001) surfaces of rocksalt crystal structure TiC, ZrC, HfC (group 4), NbC, TaC (group 5), and δ-MoC (group 6) TMCs, 6,11 one has to keep in mind that these materials are longstanding known to be easily oxidized, forming surface oxide structures known as oxycarbides. Thus, oxygen is regarded as a poison on TMCs, 14-16 e.g. oxycarbide is known to undermine the Mo 2 C catalytic performance for the water gas shift (WGS) reaction, where O moieties are mostly created from the H 2 O decomposition or hydroxyl recombination. In the course of the reverse WGS, either a direct hydrogenation of CO 2 , or a rapid hydrogenation of as-created O adatoms may prevent such oxycarbide formation, as observed on α-Mo 2 C. 8 However, for the aforementioned series of TMCs, the interaction with O 2 often leads to dissociation with the potential formation of oxycarbides being greatly energetically and kinetically enhanced. [17][18][19] This is clearly exemplified in the WGS on TiC (001), where O adatoms are found to lead to a thermodynamic sink. 20 The DFT calculated energy barriers, E b , for O 2 dissociation go from very low on group 4 TMCs, ranging from 0.16 to 0.19 eV, to moderate on group 5 TMCs -0.47-0.82 eV-, being δ-MoC the most resilient, with an energy barrier of 1.15 eV. 17 23 Therefore, it is reasonable to argue that such atomic exchange, with concomitant formation of an oxycarbide surface, is likely to occur under standard catalytic working conditions, unless is avoided, e.g. hydrogenating the asgenerated the surface O. Aside, these previous studies support that surface generated O adatoms will eventually replace C in the carbide crystallographic network, rather than staying as an oxygen overlayer.
Clearly, a remaining open question concerns the impact of the formation of such oxycarbides in comparison to the chemistry of the pristine carbide surfaces. Along this discussion, the oxycarbide activity is often regarded as similar to the parent higher oxide situation, e.g., the oxycarbide phase of TiC would chemically act similar to TiO 2 , 8 although locally the oxycarbide geometric (and consequently electronic) structure can grossly differ from that of the parent bulk oxides. Furthermore, often the oxycarbide modeling is considered with fully O-covered TMC surfaces (θ O = 1 monolayer) 16,24 with just rare exceptions where a fully oxidized TMC layer is duly described. 22,23 Hence, the surface chemistry of the oxycarbide phases resulting from exposing TMCs surfaces to oxygen remains essentially hitherto unknown.
In the present work, we contribute unraveling the impact of oxycarbide formation on the CCS activity of TMCs and, hence, to determine whether these materials would feature deactivation towards CO 2 in the presence of atmospheric O 2 . To this end, we assume a limiting oxycarbide formation situation where the exposed TMC (001) surface has fully -4 -exchanged its C atoms by O, i.e., featuring a transition metal oxide layer but commensurate with atomic structure of the underlying TMC. Note, however, that such models suppose a layer-by-layer oxide growth from the parent carbide. However, one has to keep present that other mechanism may occur, such as pit growth, plus different domains with different oxide stoichiometries can coexist on a formed oxycarbide, here not tackled.
The here presented simulations, carried out by means of periodic DFT calculations on appropriate oxycarbide slab models of the (001) surface of these TMCs, show that group 4 transition metal oxycarbides (TMOCs) feature no significant surface relaxation, and by that they keep the CO 2 adsorptive and activation capabilities. The group 5 and 6 TMOCs either show no rumpling or partial/full oxygen rumpling, but in any case with very weak interaction with CO 2 . Consequently, group 4 TMCs (TiC, ZrC, and HfC) are highlighted as materials for CO 2 CCS and CCU technologies even in the presence of O 2 .

COMPUTATIONAL DETAILS
The studied TMOCs have been represented by periodic slab models The employed slabs are built from TMCs (001) surface models previously optimized with the same DFT based method, 6 but replacing the first layer C atoms by O atoms. These slab models contain four atomic layers with the two outermost fully relaxed and the two bottommost fixed as in the bulk optimized material, providing an appropriate environment to the surface region. 6 The structural optimizations are carried out within the same computational framework earlier described in the literature 6 using the Vienna ab initio simulation package (VASP), 25 and using the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional, 26 as it constitutes one of the most accurate choices in the description of TMCs. 27 Whenever necessary, the effect of dispersion related interaction terms, as in the estimate of CO 2 adsorption energies, have been included following the Grimme D3 (PBE-D3) correction. 28 The effect of the atomic core electrons on the valence electron density is taken into account through the projected augmented wave (PAW) method developed by Blöchl,29 as implemented in VASP by Kresse and Joubert. 30 The valence electron density is expanded in a plane wave basis set, with a kinetic energy cutoff of 415 eV.
To speed-up the CO 2 adsorption calculations, an initial screening of all sites on the different surface models was carried out using a 5×5×1 Monkhorst-Pack k-points grid, 31 where most stable situations were refined reoptimized using a denser 9×9×1 grid.
Convergence criteria for electronic and structural updates were 10 -6 eV and 0.01 eV Å -1 respectively. The adsorbed CO 2 optimized geometries were characterized as minima of the potential energy surface by frequency analysis through Hessian matrix construction and -5 -diagonalization, with the elements of the adsorbate related Hessian matrix estimated from finite differences of analytical gradients with atomic displacements of 0.03 Å length. In some CO 2 physisorbed cases, one or two small spurious imaginary frequencies remain; always well below 50 cm -1 , and related to substrate non-fully frustrated translational or rotational modes.
At the strict convergence criteria used, these geometries can be safely reported as minima of the potential energy surface. 32,33 The interaction between CO 2 and the TMOC surfaces has been further studied by means of charge density difference (CDD) analysis, simulation of the associate infrared (IR) spectrum, and electron localization functions (ELF); computational details on these methods simulations can be found elsewhere. 11 Note that although the employed models describe oxycarbides with a full monolayer (ML) of oxygen coverage (θ O = 1 ML), the CO 2 coverage considered is low enough (θ CO 2 = 0.125 ML) so as to disregard lateral interactions between periodically repeated CO 2 adsorbates. We should remind that for either O or CO 2 , the coverage is defined as the number of entities with respect to the number of transition metal atoms exposed on the surface. The CO 2 adsorption energies, !"# !" ! , are estimated as where !" ! /!"#$ is the energy of the TMOC (001) surface slab with the adsorbed CO 2 , !"#$ is the energy of the optimized yet pristine TMOC (001) surface slab, and !" ! the energy of the CO 2 molecule optimized in an asymmetric unit cell of 9×10×11 Å dimensions.
Within this definition, the more negative the E ads , the stronger the adsorption is. Unless stated otherwise, all reported E ads values include the zero point energy (ZPE) correction. On the basis of the calculated harmonic vibrational frequencies, adsorption/desorption rates for CO 2 from the different surfaces have been obtained in the framework of TST as described in previous works. 6,32,33 For a fair comparison between TMC and TMOC surfaces -exhibiting different adsorption sites for CO 2 -, a value of 0.139 nm² was used for an estimate of the adsorption site area on all surfaces. In addition, a conservative sticking factor of 0.2 was assumed for the CO 2 impingement on both TMC and TMOC surfaces. 32,33

Pristine TMOCs (001) surfaces
First, we consider the energetics and geometric structure of the TMOC (001) surfaces For each TMOC (001) surface Table 1 reports the surface energy difference between the optimized structure for each three models relative to the most stable one. Note that after relaxation some surface models evolved to different structures, i.e. the half rumpled situations for group IV TMOCs relaxed to the non-rumpled model. For all the other cases, the structures remained as such yet relaxed, being indeed minima of the potential energy landscape.
Analyzing the results, for group 4 TMOCs, the oxycarbide formation does not lead to any noticeable rumpling, as the full rumpling situations are higher in energy, with the sole exception of HfC oxycarbide, for which both surface structures are close in energy. The unrumpled, half-rumpled, and fully-rumpled situations are competitive in the (group 5) VC oxycarbide, whereas the fully rumpled situation appears to be the likely model for the rest of group 5 and 6 TMOCs. Note in passing by that even though unrumpled situations were explicitly optimized for group 5 and 6 TMOCs, these could spontaneously evolve to fully rumpled in the course of CO 2 adsorption.
The degree of rumpling can be quantified by measuring the O and TM heights of the TMOCs models compared to the parent TMC (001) surfaces. 6,34 The corresponding displacements, Δh, obtained from PBE geometry optimizations, are displayed in Table 2

CO 2 adsorption
-7 -Following a previous procedure, 6 the CO 2 molecule has been adsorbed on the TMOCs slab models, by placing the molecule parallel to the surface, at a height of 1.6 to 2 Å, over different high-symmetry surface positions, contemplating top, bridge, and hollow positions, with two different CO 2 orientations along the surface planes on each site (see Figure 2). The geometry optimizations lead to two distinct situations depending on the TM group. On group 5 and 6 TMOCs, the geometry optimization of the adsorbed CO 2 on the fully rumpled models (NbC, TaC, and δ-MoC oxycarbides), and on the competitive unrumpled, half-rumpled and fully-rumpled models of the VC oxycarbide lead to physisorption situations with E ads values smaller than -0.05 eV (PBE), which are slightly larger (below -0.27 eV) when including dispersions (PBE-D3); a summary of results is presented in Table 3. In these cases, the CO 2 molecule remains essentially straight and parallel to the surface, see Figure S3 in SI and structural data in Table 3, in line with the results for other surfaces where CO 2 interaction is weak. Hence, the above-commented rumpling, which brings the structure closer to the parental transition metal oxides, makes the positively charged metal centers less accessible to stabilize the negatively charged O atoms of an hypothetical CO 2 δentity, which seems to be key point in the predicted CO 2 physisorption.
However, the situation is markedly different for group 4 TMOCs, where a few competitive chemisorption situations are found. Figure 3 displays the most stable adsorption situation on TiOC, while the corresponding situations for HfOC and ZrOC are reported in the Figures S1 and S2 of the Supporting Information (SI), respectively. The molecular structure of the chemisorbed CO 2 on these TMOCs surfaces closely resembles those obtained on nonoxidized pristine TMC (001) surfaces. 6 However, here most cases involve a four-fold atom coordination, or indentation. In the case of TiOC, the most stable coordinations correspond to MOMO bidentate and tridentate sites, as shown in Figure 3. The most stable configuration for CO 2 on ZrOC displays the same MOMO tridentate as on TiOC but also another distinct tridentate MMMO situation (see Figure S1 in SI) is found. Notice also that CO 2 adsorbed on The above interpretation is further supported by the Bader charge analysis, which reveals large TMOC→CO 2 charge transfers in the 1.0-1.9 e range, as gathered in Table 3.
Note also that charge density difference (CDD) and electron localization function (ELF) profiles are similar to those corresponding to activated CO 2 on TMC (001) surfaces, 11 as shown in Figure 3. The present CDD pictures are fully consistent with surface metal and O centers favorably interacting with the Lewis acidic and basic carbon and oxygen atoms of CO 2 . 35 Furthermore, charge depletion at the TMOC surface and charge accumulation regions on the CO 2 molecule support the charge transfer mechanism above discussed. Finally, the ELF plots highlight an important electron pair density between the oxycarbide surface O atom directly involved in the CO 2 adsorption and the CO 2 carbon atom, evidencing a newly formed covalent bond, and therefore suggesting the emergence of a carbonate (CO 3 !-) type of entity.
Such carbonate-type of interaction is classically observed for CO 2 when chemisorbed on highly ionic rocksalt alkaline earth oxides, such as MgO or CaO, where previous studies showed that the CO 2 adsorbs in a monodentate and tridentate fashion as the here exposed TopO and MMO sites on TiOC. 36,37 This goes along with the observed interaction of CO 2 on the rocksalt type of titanium dioxide, as TiO is known to be more ionic than titanium dioxide TiO 2 , 38 and hence, behaving quite as an alkaline earth oxide such as MgO or CaO.

Consequences on CCS and CCU
According to the calculated CO 2 adsorption energies on the oxycarbide models described in the previous, the oxidation of TMC (001) surfaces would unchain dramatic consequences for most of the systems. While group 5 and 6 NbC, TaC, and δ-MoC will chemisorb and activate CO 2 at normal conditions of temperature and CO 2 partial pressure 6 ⎯ even VC if exposed to sufficiently high partial pressure 11 ⎯, a point experimentally observable by infrared (IR) spectroscopies with signals in the 1100-1600 cm -1 range, 11 the corresponding oxidized (VOC, NbOC, TaOC and MoOC oxycarbide) surfaces would lead to physisorbed CO 2, and, therefore, to no IR active signals. On the contrary, the formation of the TiOC, ZrOC, and HfOC oxycarbides would just imply a shift of the IR active bands as shown in Figure 4. Given the distorted nature of the CO 2 bonding in these oxycarbide surface models, the shift appears to be especially significant in the asymmetric stretching, ν as , of the C-O bonds. As an example, the TopC CO 2 adsorption IR feature on TiC (001) surface, predicted to be centered at 1514 cm -1 , which would significantly blueshift by 128 cm -1 on TopO situation to a value of 1642 cm -1 , see Figure 4. On the contrary, the symmetric stretching, ν s , would barely redshift by 18 cm -1 . In this sense vibrational spectroscopies would constitute a straightforward way of detecting and distinguishing activated CO 2 on carbide and oxycarbide surfaces, or, alternatively, to detect the presence of surface oxidation of a TMC using CO 2 as a probe molecule.
Last, the noticeable strengthening of the CO 2 adsorption on group 4 TMOC (001) surfaces has consequences on the transition temperature ranges for switching from CO 2 capture to release, which would move to values higher than for the corresponding TMC. This is shown on Figure 5 for most stable CO 2 adsorption on group 4 TMOCs, with adsorption/desorption rates estimated from transition state theory 6 and considering the ZPEcorrected PBE and PBE-D3 adsorption energies as fringe situations. Within this procedure we evaluated the turning temperatures for three CO 2 partial pressures ( !" ! ); i) the nowadays atmospheric CO 2 partial pressure of 40 Pa, 39 ii) a !" ! = 15·10 3 Pa pressure benchmark used in exhaust gases, 40 and iii) a standard pressure of 1 bar (10 5 Pa), relevant in the pure CO 2 stream generation after CCS. 41 The results reveal that, under O 2 atmospheric conditions, CO 2 would be stronger adsorbed on the TMOCs than on the TMCs and, hence, it will be more difficult to be released. Note that going from TMC to TMOC shifts, the turning ranges increase by 150-190 K, and stretch over a wider range, even more acute in the HfC based case. Nevertheless, the higher surface activity of group 4 oxycarbides, their carbide passivating nature, and their enhanced CO 2 activation capabilities would make them excellent, realistic choices for the CO 2 posterior conversion, e.g., for CO 2 hydrogenation towards methanol, although further research has to be devoted on this matter.

SUMMARY AND CONCLUSIONS
In this work TMOCs surface models have been built, analyzed, and used to study oxycarbide formation implications in the CO 2 adsorption and activation on the stable (001) surfaces of different transition metal carbides (TMCs and most stable CO 2 physisorbed situations on the TMOC surface models ( Figure S3). The Supporting Information is available free of charge on the ACS Publications website at DOI:     surface slab models. Color scheme as in Figure 1.   (001) surfaces. Peaks are marked with their vibrational frequency (in cm -1 ). Relative intensities (Rel. int.) of the asymmetric stretch, ν as , were overall low and have been slightly increased for better visibility.

XXXXXXXXXXX
-19 - Figure 5. Calculated rates of adsorption, r ads , and desorption, r des , for CO 2 on TiOC, ZrOC, and HfOC (001) surface models as a function of the temperature, and at three partial pressures of CO2: 40, 15·10 3 , and 10 5 Pa. For each adsorption situation, limiting values for strongest PBE-D3 (dotted lines) and weakest PBE chemisorption cases (solid lines) are considered. The marked points T 1 -T 6 show how adsorption/desorption turning temperature ranges have been acquired (depicted below). For convenience, similar ranges are also shown for TiC, ZrC, HfC.