Catalysis Carbon Dissolution and Segregation in Platinum

Recent experimental studies evidenced C dissolution in Pt nanoparticles after CH 4 decomposition, and its low temperature seggregation to form surface graphene, highlighting a graphene growth from below, with indications of an easier C transfer in between surface and subsurface regions at Pt grain boundaries, alhtough the ultimate atomistic mechanism remained unclear. A plausible explanation is provided here by exploring and comparing C incorporation on Ni, Pd, and Pt (111) surfaces by density functional (DF) calculations on slabs models at a low coverage regime, evaluating energetic stability and subsurface sinking kinetic feasiblity. Four DF functionals have been used, avoiding possible biased results. All show that C atoms occupy octahedral subsurface sites ( oss ) in Ni(111), with high sinking energy barriers of 80-90 kJ mol -1 , whereas both oss and tetrahedral subsurface sites ( tss ) can be occupied on Pd(111), with low sinking energy barriers of 20-50 kJ mol -1 . The oss sites are strongly disfavoured on Pt(111), whereas tss sites are found to be isoenergetic to surface sites, with low subsurface sinking energy barriers of 27-41 kJ mol -1 . Calculations on Pt 79 and Pt 140 nanoparticle models reveal how tss sites are more stabilized at low-coordinated sites, where subsurface sinking energy barriers drop to values of ~17 kJ mol -1 . Results explain experimentally observed C dissolution and segregation in Pt systems, more favoured at grain boundarie, as well at the growth from below and formation of double layers. Present results as well open a gate for profiting of small quantities of C placed at the subsurface region to tune the surface catalytic activity of Pt nanoparticle based catalysts.


Introduction
The interplay of Carbon and Platinum is paramount in many scientific and technological fields. For instance, Pt nanoparticles synthesized over high surface area, microporous carbon based materials are nowadays widespread used as oxygen reduction reaction (ORR) catalysts in fuel cell technologies. 1 As far as carbon-based nanotechnologies are concerned, Pt single crystal grains have been recently proposed as a template for the controlled graphene tessellation on their surface. 2 However, in heterogeneous catalysis, Carbon is also longstanding regarded as a catalyst poison. Coke, e.g. graphitic carbon, generated in the course of a surface catalysed reaction involving organic moieties, is deposited over the metal catalyst surface, blocking the active sites and so severely decimating its performance. 3 Catalyst deactivation is therefore an undesired yet regretfully common phenomenon, with a high, negative impact in chemical industry. Because of this with many research endeavours are addressed at its mitigation, even to the point of prevention, including the set up of poison 4 and sinter resistant 5 catalysts, as well as catalyst regeneration procedures. 3 However, recent studies have shown that low contents of some catalyst surface poisons are actually beneficial, allowing for the selective isolation of intermediates and final products with a given desirable stereochemistry. 6,7 This poison-to-promoter role change seems to be case for C 1 atomic species on Pd catalysts employed for hydrocarbon dehydrogenation processes, 8 where the other group X transition metals Ni and Pt are employed as catalysts as well. [9][10][11] Indeed, a higher performance has been found to be connected to the existence of subsurface C species at hydrocarbon reaction working conditions, with theoretical simulations backing up the experimental determination of carbon residues, which, in the form of either substitutional or interstitial impurities, affect the catalyst selectivity.
geometrical and electronic structure, may favour the incorporation of other moieties, as found to be the case for H in Pd nanoparticles, 19 eventually allowing, by indirect control of the amount of subsurface H, for critically speeding up or slowing down the on-going surface hydrogenation reaction catalysed on Pd nanoparticles. 20 However, despite being of the same group X, subsurface C species in Pt have been historically disregarded, despite of the experimental evidence of their existence. C impurities are long known to segregate from bulk during Pt single crystal preparation. 21 This has been recently profited to synthesize surface graphene from C dissolved in Pt nanocrystallites, involving methane (CH 4 ) chemical vapour deposition, its surface full dehydrogenation and C atomic dissolution at high temperatures, followed by a low temperature C segregation. 2,22,23 These experiments provided compelling evidence of C dissolution and a graphene synthesis mechanism from below, providing a response to a longstanding argument about graphene formation from surface aggregation or subsurface segregation. 22 Aside, C absorption/segregation was found to be more facile at Pt grain boundaries. Therefore, tuning the working conditions allowing for C concentration control as well as of the balance between deposition, and absorption and desorption rates permit to tailor the formation of a continuous graphene layer, few-layer graphene, up to graphite nanocrystallites. This seems to be a key also for the appearance of islands of C impurities beneath the as-formed graphene sheet, misaligned with respect to the graphene layer, known also to form from below. These resemble the so-known double layer model proposed for graphene flakes on Pt(111) surface, also summoned long ago to explain the separation of the carbon layer, now known as graphene, from it. 24,25 The double layer model has been backed up by theoretical simulations based density functional (DF) theory. 26 However, the crucial question of how could C atoms slide into the graphene/Pt interface remained open. Experimentally, C solubility in Pt at 1000 ºC is 1.14 at. %, 27 similar to that of Ni (1.26 at. %), 28 and actually both molten Pd and Pt feature a similar solubility for C. 27 So, the growth from below appears as an appealing simple explanation for it. However, this possibility has never been validated from the theoretical point of view. Indeed octahedral subsurface (oss) site occupation has been found to be endothermic by ~60 kJ mol -1 , according to DF calculations on Pt(111) surface slab model. 26 In addition the lower Pt ductility compared to Ni and Pd, as a result from a stronger Pt metal bonding, implies a cost for structural relaxation nine times larger than Pd or Ni, and so this factor plays against C incorporation, despite the higher melting Pt is appealing for graphene synthesis control. 2 All in all, C dissolution in Pt remains an open question where the diverse puzzle pieces do not fit together. 29 To provide an unbiased answer to this and to fill the necessary lack of mechanistic detail on C dissolution and segregation processes on Pt systems, we carried out a full state-of-the-art DF of them on Ni, Pd, and Pt surfaces, thus evaluating the stability of all subsurface and surface species, as well as the subsurface sinking energy barriers. Through this systematic study a complete picture of the energetics and kinetics of subsurface C moieties formation is gained. The obtained results at different levels of theory show that while C atoms prefer to occupy oss sites in Ni, they occupy both oss and tetrahedral subsurface sites (tss) on Pd, and, more importantly, they also can occupy tss sites in Pt. For Pt, tss stability is further strengthened at undercoordinated sites, as shown on Pt 79 and Pt 140 nanoparticles, accompanied by a subsurface sinking energy barrier reduction. Present results strong support the existence of subsurface C in Pt systems and explain to the above-commented graphene segregation on Pt surfaces, the double layer model, and the preferential absorption on Pt grain boundaries. Control of subsurface C can be envisaged as a way to tune the surface Pt catalytic activity, specially at low coordinated sites, thus contributing into the rational design of tailored Pt-based heterogeneous catalysts.

Computational details
The DF calculations have been carried out using the VASP code exploding periodic boundary conditions. 30 Geometries and total energies reported in the following were optimized using four different exchange-correlation (xc) functionals, including  35 was not used since provides essentially same results as VWN for transition metals. 36,37 The underlying reason to explore many xc functionals is to better assure an unbiased outcome, since e.g. VWN and CA are known to overestimate interatomic interactions, whereas RPBE for instance is known to underestimate them. 36 Valence electrons density was expanded in a plane wave basis set with a 415 eV cutoff for the kinetic energy and the projector augmented wave method was used to describe the interactions between core and valence electrons. 38 Spin polarized calculations were carried out for magnetic Ni, and non-spin polarized for nonmagnetic Pd and Pt metals. A 54 metal atoms (3×3) supercell slab was used to model the (111) surface of the three metals. The slab models contained six atomic layers with nine atoms per layer. The three bottom layers of the slab were kept fixed at the optimized yet bulkterminated geometry, while the other three upper layers were allowed to further relax during geometry optimization, together with the adsorbed/absorbed carbon atom. The reciprocal space was sampled with 6×6×1 Γ-centered k-point grid and calculations were performed using a Gaussian smearing of 0.2 eV energy width to speed up convergence, yet final energies where extrapolated to 0 K (no smearing). In addition, Pt 79 and Pt 140 nanoparticle models were also used allowing a comparison among extended regular surfaces and low coordinated sites at Pt nanoparticle grain boundaries. For a depiction of Pt 79 and Pt 140 cuboctahedral nanoparticles we refer to the literature. 26,39 Briefly, Pt nanoparticles have been calculated by placing them in a cubic box imposing a minimum distance of 10 Å in between translationally repeated nanoparticles and considering the Γ point only. The same procedure was used to compute the energy of the isolated C atom, hence an asymmetric box of 9×10×11 Å has been used to assure the correct orbital occupancy featuring a triplet state. In all cases, geometry optimizations were performed until all forces acting on relaxed atoms became less than 0.03 eV Å -1 . Several positions are possible for the adsorption or absorption of C atoms on the fcc (111) metal surfaces. Herein we sampled three adsorption and three absorption sites, shown in Fig. 1: Top site and fcc and hcp three-fold hollow sites for the adsorption, and oss and two types of tss for subsurface C. Note that oss is located below an fcc surface site, whereas tss is located below a surface hcp hollow site. The tss' site is located just below a surface top metal atom. Given the employed supercell, calculations imply a C (sub)surface coverage of 1 / 9 monolayer (ML). The activation energy barriers associated to a surface-tosubsurface diffusion were determined by scanning the potential energy profile along the line that connects the local minima of the surface and subsurface impurity atoms. Thus estimations of Transition States (TSs) were searched in a pointwise fashion along the path connecting adsorption and absorption configurations, where height of the carbon atom, defined with respect to the most distant frozen metal layer, was fixed, whereas all other degrees of freedom were allowed to fully relax. Subsequently, each of these approximate TSs was refined allowing C displacement along surface/facet direction by a quasi-Newton method, until forces acting on atoms were below 0.03 eV Å -1 . The resulting TS structures have been characterized by vibrational frequency analysis, certifying their saddle point nature.
We define the calculated adsorption/absorption energies as; where !/!"#$% is the total energy of the metal model with the C atom either adsorbed or absorbed, ! is the energy of an isolated carbon atom, and !"#$% is the energy of the optimized clean substrate: Ni, Pd, and Pt (111) surface slab models, or Pt 79 or Pt 140 nanoparticles. With this definition, stable adsorption/absorption corresponds to positive !"#/!"# values.

C adsorption/absorption on/in metal surfaces
Let us first start with the energetic stability of C atoms on Ni, Pd, and Pt (111) surfaces, as well as in their first subsurface layer. Adsorption and absorption energies are listed in Table 1, and plotted in Fig. 2. From the obtained results some general features are clearly evident. First, both PW91 and PBE xc functionals provide essentially same results, with discrepancies of, at most, 6 kJ mol -1 , thus well within standard DF accuracy of ~10 kJ mol -1 , and in accordance, for instance, the very similar performance of PBE and PW91 in transition metal bulk properties. 36 Along this line, VWN results always depict stronger adsorption/absorption situations, whereas the opposite applies for RPBE, again in line with previous results on transition metal bulks, 36 yet one clearly notices that, aside from the bonding strengthening/weakening, relative positions of ad/absorption sites are consistently predicted, and so, any of the tested xc functionals is depicting the very same chemical situation. Moreover, in all cases, surface adsorption top site is markedly less stable than hcp or fcc hollows -by more than 210 kJ mol -1 in the best case-and so has not been further considered as a possible site for adsorbed C atoms. Last but not least, notice that adsorption and absorption energies feature values above 500 kJ mol -1 whatever the metal or the employed xc functional, thus showing a highly strong interaction, in agreement with previous DF studies. 26,40,41 Focusing the attention, adsorption energies for C on Ni(111) surface are of the same order of magnitude than lowcoverage reported values reported in the literature ranging 640-700 kJ mol -1 , as obtained at PW91 or PBE levels, 40 , at any of the DF levels employed in the present work. However, the largest change occurs subsurface. Both tss and oss sites are more stable than the respective surface hcp and fcc sites. Indeed, even tss' is competitive to surface sites, being considered isoenergetic -differences in interaction strength below 6 kJ mol -1 -and actually in fully agreement with previous calculations at RPBE level by Nykänen et al. 45 Indeed, oss sites are much more stable than corresponding surface fcc sites, by 60-76 kJ mol -1 , depending on the xc functional, and so, more stabilized than on Ni. This fact aligns with an experimentally observed Pd-C interphase, 46  , value reproduced here with any of the explored xc functionals. 26 The change on the adsorption site preference is alleged to be due to a subsurface electron density steric repulsion. Aside, following the same Ni→Pd trend, E ads values for hcp and fcc sites are 10-30 kJ mol -1 larger than for Pd, and so, adsorption strength along the group nicely correlates with d-band centre and/or upper d-band edge electronic descriptors. 48 In addition, adsorption strengths are perfectly in line with those obtained by Michaelides and Hu using the PW91 functional. 49 The other remarkable point is that oss is sensibly destabilized with respect fcc surface site, as clearly observed in Fig. 2. The destabilization is large, by 50-60 kJ mol -1 , and in perfect agreement with previous estimates at RPBE level. 26 Clearly, the longer Pt-Pt distance is not translated in a better accommodation of C in the larger oss sites, and actually is rather detrimental. Indeed, an optimum oss environment for C seems to be achieved using Pd, but on Pt this slightly larger bond length appears to be excessive, and the interaction gets weakened. Quite surprisingly, C at subsurface tss site in Pt(111) appears to be strengthened. Here, at PW91, PBE, and RPBE xc levels, tss is slightly more stable than the above surface hcp site, by 3-6 kJ mol -1 , although less stable by 3 kJ mol -1 at VWN level. Indeed, it appears that space is playing a key role in subsurface stabilization, which accords with VWN results, where Pt-Pt distances are sensibly shorter, and so tss is less stabilized. The most surprising result here is that, for the first time, tss subsurface site in Pt(111) is found to be at least competitive to any of the surface fcc or hcp sites for C atoms. However, so far only thermodynamic aspects have been  Please do not adjust margins Please do not adjust margins contemplated. In the next section we inspect whether kinetic aspects would inhibit or favour such a subsurface occupancy.

C subsurface sinking
In this section we investigate the feasibility of C subsurface migration by estimating the subsurface sinking process activation energy. To this end TSs have been located following the hcp↔tss and fcc↔oss paths and appropriately characterized. Forth and back activation energies are encompassed in Table 2, and subsurface diffusion profiles in Fig. 3. From the results it is clear that subsurface diffusion profiles for any path are generally equivalent when calculated with any of the employed functionals. Actually, TSs heights for critical steps may differ by up to ~15 kJ mol -1 , but clearly this is biased by the relative position of initial and final states, as expected from the Brønsted-Evans-Polanyi (BEP) relationships on transition metals, 50,51 and the fact that these are actually not heavily influenced by the employed level of theory. 52 All that said, it is to highlight that on Ni(111) the C atoms are thermodynamically driven to occupy oss sites, yet a high barrier of 80-90 kJ mol -1 is needed to be overcome.
The relatively high energy barriers are in line with the experimental observation of C subsurface diffusion at temperatures above 700 K. 53 42 clearly showed that such high barriers are an artefact of using an exceedingly small p(2×2) supercell. In fact, the lateral displacement of surface Ni atoms in the TS implies a strong slab tension, which raises the TS energy. Using the PW91 functional, these authors estimated an energy barrier using a p(2×2) supercell of 164 kJ mol -1 , and a value of 71 kJ mol -1 using a p(3×3) supercell, which nicely correlates with present results using the same cell dimension. The higher experimental energy barriers are probably due to the posterior bulk diffusion of the subsurface C atoms, estimated to be 156-166 kJ mol -1 at PW91 level, 42,56 which would act as a rate limiting step.  When comparing Ni to Pd, one observes in Pd that both oss and tss sites can be kinetically occupied, since subsurface diffusion energy barriers are similar in height. Indeed, the tss sites, despite being less stable than oss, display lower energy barriers compared to barriers towards oss by ~16 kJ mol -1 , in full agreement with previous PW91 estimations on VWN structures. 41 Note that, whatever the subsurface site, C incorporation implies overcoming energy barriers of 20-50 kJ mol -1 . This signifies that C penetration is easier than C adatom surface diffusion, which has an energy barrier of almost 70 kJ mol -1 as obtained at PW91 level, 13 and being also smaller than the dehydrogenation barriers in C atom formation from methane 57 or ethylene, 58 being as high as 125 and 154 kJ mol -1 as obtained at RPBE and PW91 levels, respectively. So, whenever C atoms are on the Pd(111) surface, they sink subsurface, being neither their formation nor their surface diffusion the limiting step towards the formation of a Pd-C phase. The larger malleability of Pd compared to Ni and the larger intermetallic distances are both at the origin of the increased subsurface stability and kinetic feasibility for subsurface C species. Note that this correlates with a smaller experimentally determined subsurface diffusion energy barrier of 107-155 kJ mol , thus fitting the above commented experimental values, and backing up a bulk diffusion rate limiting step, rather than the actual subsurface diffusion.
The case of Pt(111) seems just the contrary as to Ni(111). Because of the above commented instability of C at oss sites, the energy barriers towards it are rather large, in the 95-105 kJ mol -1 range, whereas the surface emerging ones are sensibly small, in between 35-55 kJ mol -1 . So, occupancy of oss sites is hindered both because of thermodynamic and kinetic reasons. Surprisingly, a completely different situation is found for the 6 | Catal. Sci. Technol., 2016, 00, 1-9 This journal is © The Royal Society of Chemistry 20xx Please do not adjust margins Please do not adjust margins tss sites. Thermodynamically, they are of similar energy to hcp surface sites, and the subsurface energy barriers -and so the surface emerging ones-are relatively small as in the Pd case, here of 25-45 kJ mol -1 height. This implies a possible occupancy of tss sites at room temperature, and undoubtedly at methane or ethylene dehydrogenation working conditions, 2,22,23 where rate limiting steps have been found to be of 120 kJ mol -1 for methane, as obtained from RPBE, 57,61 and 143 kJ mol -1 from ethylene as obtained from PW91 DF calculations. 58 As happened with Pd, the surface diffusion among fcc and hcp sites was previously estimated on Pt(111) to be of 70-90 kJ mol -1 at RPBE level, and so, diffusion would be no limitation in C subsurface occupancy of tss sites. Thus, so far, we have shown how C atoms can entrench on Pt(111) surfaces occupying subsurface tss sites even at low temperatures. Apparently, the hcp↔tss route, displaying low energy barriers, is the gate both for the C dissolution and segregation processes, switched on/off by the C chemical potential at the surface and subsurface regions. Moreover, the stability of C atoms in tetrahedral interstitial sites inside Pt seems to play a major role in C incorporation on Pt based composites. The above commented experimental similarity between Pd and Pt concerning C affinity, 27 and the carbon dissolution and segregation phenomena, including the double layer model, is so now understood. 2,[21][22][23][24][25] At this point, one may wonder, given the above stated similarity in between Pd and Pt, whether subsurface C could be further stabilized at Pt low-coordinated sites, as found for Pd. In the latter, subsurface diffusion has been found to be essentially barrierless at edges and corners of Pd nanoparticles boundaries, 41 driven by a further accommodation in such sites.
This links to a significant low energy cost relaxation to accommodate subsurface C on these sites, and so supporting previous experimental statements suggesting existence of carbonaceous deposits at edges between (111) facets of Pd nanoparticles. 62 Such a large reduction in subsurface sinking energy barriers has also been found on Pd(211) steps according to the RPBE simulations by Nykänen et al. 45 Furthermore, the existence of such subsurface C atoms in Pd nanoparticles has been found to promote H incorporation both thermodynamically and kinetically. 19,20,63 Indeed, Babar and coworkers suggested that some sites of Pt nanoclusters may become more active once C is adsorbed, and eventually stronger bind H. 64 The question mark here is whether such subsurface C atoms at lowcoordinated sites in Pt nanoparticles would be further stabilized, further backing up their usage to tune, for instance, on-going surface catalysed (de)hydrogenation or oxygen reduction reactions. This is further addressed in the following section.

Subsurface C at low-coordinated sites
In order to ascertain whether C atoms are further stabilized at low coordinated sites of Pt nanoparticles, we employ two nanoparticles: Pt 79 and Pt 140 , which are within the regime of properties scalable to bulk. 65 For details of such cuboctahedral nanoparticles we refer to previous literature. 26,57 Since all the tested xc functionals provide very similar results, we restricted these more time consuming calculations to the use of the PBE functional. Furthermore, from Fig. 3, it is clear that tss sites are actually favoured towards C occupation, and so, only hcp and tss sites have been sampled on Pt 79 and Pt 140 models, as shown in Fig. 4. that such enhancement can have different origins; on one side the Pt 79 nanoparticle is at the fringe limit of the scalable regime, and so, such adsorption energy values could be simply related to this particular size & shape. Moreover, the nm size implies a quantum confinement, which would enhance the very chemical activity, and even more, site coordination is known to play a key role in modulating the adsorption energy. 66 All that considered, C adsorption seems to be slightly improved at corners and edges of the Pt 79 nanoparticle. However, this does not necessarily apply to tss sites. Here, C at tss just below corner site of Pt 79 is slightly destabilized by 5 kJ mol -1 compared to tss in Pt(111), whereas C at tss below the edge site is stabilized 12 kJ mol -1 . Note that in such subsurface sites the lower coordination and material flexibility goes for a larger C stabilization, although the reduced Pt interatomic distances imply smaller tss cavity, and so has an opposite effect. In any of the explored sites the subsurface diffusion implies a barrier of 30-40 kJ mol -1 , and so, very similar to the value of 31 kJ mol -1 obtained on the Pt(111) slab model, which seems to indicate that both factors are compensated. The situation appears to change when moving to a larger nanoparticle, and so, a more representative model for Pt nanoparticle grain boundary regions. Pt 140 is well within the size region of properties scalable to bulk, and here quantum size effects are attenuated compared to Pt 79 . This explains how adsorption on C atom on hcp sites of the (111) facet implies a bonding strength smaller than those sampled cases on Pt 79 . Compare for instance how E ads values in Fig. 4  , thus going for a subsurface C presence at defects of Pt nanoparticle systems and Pt steps even at low temperatures, and a much easier transfer for C atoms from and towards the subsurface region. This result is in perfect agreement with the observed preference of C to absorb on Pt at grain boundaries, as well as the growth from them. 22 In that sense, a similar mechanism for H ad/absorption modulation via subsurface C, as observed on Pd nanoparticle systems, seems plausible, and offers a simple yet intuitive explanation to the experimental observations of Babar and coworkers. 64 The effect of subsurface C in Pt may have also as well implications on oxygen reduction reactions; to name few, defective Pt(111) facets have been theoretically proposed as a way of improving Pt catalysts for the ORR, 66 and, inevitably, C impurities, when present, would reside subsurface of such defective higher active sites, modifying the surface on-going reaction. Moreover, oxygen subsurface impurities are known to be generated in Pt based catalysts during ORR, 67 including Pt@Pd and Pt@Ni systems, 68,69 and argued to play a determinant role in a successful operation by destabilizing surface hydroxyls due to electrostatic repulsion, even though subsurface O formation is a highly endothermic process, only achieved with a high surface chemical potential. 70 Clearly subsurface C, also negatively charged, could be easier formed with forecasted similar effect on surface processes.

Conclusions
Carbon is a usual support for Pt nanoparticle catalysts, yet carbon deposits formed in Pt surface poison the nanoparticle performance. Subsurface C present in other group X metals Ni and Pd are known to be a critical factor in graphene segregation processes, but also, in low contents, improving the surface catalytic activity either being more active than surface C moieties, by desirably tuning the surface catalytic activity, or by enhancing the metal incorporation of other reacting species, such as found for H on Pd nanoparticles. Recently. strong experimental evidence of C dissolution in Pt systems has been found, 2 confirming a graphene formation by C segregation from below, although the detailed atomistic mechanism remained unveiled.
Here all the experimental observations are explained via a thorough density functional study. We fully explored atomic C incorporation on Ni, Pd, and Pt (111) surface systems on proper supercell slab models, allowing for determining the relative stability of C atoms at surface and subsurface sites at a low coverage regime, while estimated subsurface sinking energy barriers permit one determining the kinetic role on the process and the working conditions at which such processes are feasible.
The results have been obtained using four different exchange-correlation functionals to avoid possible bias on the results obtained from a particular choice, yet all the employed methodologies deliver the same picture: C atoms are thermodynamically driven to occupy octahedral subsurface sites (oss) in Ni(111), despite of featuring high energy barriers of 80-90 kJ mol -1 , and so, these would only be occupied at high temperature working conditions, as experimentally observed. Both oss and tetrahedral subsurface sites (tss) would be occupied on Pd(111) but at lower temperatures, as energy barriers for subsurface occupancy are in the 20-50 kJ mol -1 range. Last but not least, according to present calculations, oss sites are hindered on Pt(111), being 50-60 kJ mol -1 less stable than immediately superior fcc surface sites, whereas tss sites are found to be essentially isoenergetic to immediately superior surface hcp sites. Thus, tss subsurface sites are probably sampled on Pt systems, given that subsurface sinking energy barriers are low, of 27-41 kJ mol -1 height, explaining the similar C dissolution capability of Pt compared to Pd, and the low temperature segregation to form surface graphene.
The analysis of structural effects shows how the size of subsurface space sites is related to the interatomic metal distances and the relaxation upon subsurface occupancy. This is at the origin of the stabilization in tss and oss sites. Last but not least, such tss occupancy has been further found on Pt 79 and Pt 140 nanoparticles, representative of larger Pt nanoparticle systems. Here the smaller room due to nanoparticle contraction is counteracted by an enhanced chemical activity. Compared to Pt(111) extended surfaces or terraces, C at tss sites near low-coordinated sites of the nm sized Pt 140 particle are slightly more stabilized but, more importantly, subsurface energy sinking drops to values of ~17 kJ mol -1 , implying a low-temperature occupancy of tss sites by C, in a similar fashion to what was observed on similar metal Pd, and explaining the observation of a stronger absorption and segregation of C atoms at Pt grain boundary regions. Present results reveal the possible modulating role of C