J. G.
Fallaque
ab,
M.
Ramos
c,
H. F.
Busnengo
c,
F.
Martín
abd and
C.
Díaz
*e
aDepartamento de Química, Módulo 13, Universidad Autónoma de Madrid, 28049 Madrid, Spain
bInstituto Madrileño de Estudios Avanzados en Nanociencia (IMDEA-nanociencia), Cantoblanco, 28049 Madrid, Spain
cInstituto de Física Rosario, CONICET and Universidad Nacional de Rosario, Bv. 27 de Febrero 210 bis, 2000 Rosario, Argentina
dCondensed Matter Physics Center (IFIMAC), Universidad Autónoma de Madrid, 28049 Madrid, Spain
eDepartamento de Química Física, Facultad de CC. Químicas, Universidad Complutense de Madrid, 28040 Madrid, Spain. E-mail: crdiaz08@ucm.es
First published on 10th September 2020
The dissociative adsorption of molecular oxygen on metal surfaces has long been controversial, mostly due to the spin-triplet nature of its ground state, to possible non-adiabatic effects, such as an abrupt charge transfer from the metal to the molecule, or even to the role played by the surface electronic state. Here, we have studied the dissociative adsorption of O2 on CuML/Ru(0001) at normal and off-normal incidence, from thermal to super-thermal energies, using quasi-classical dynamics, in the framework of the generalized Langevin oscillator model, and density functional theory based on a multidimensional potential energy surface. Our simulations reveal a rather intriguing behavior of dissociative adsorption probabilities, which exhibit normal energy scaling at incidence energies below the reaction barriers and total energy scaling above, irrespective of the reaction channel, either direct dissociation, trapping dissociation, or molecular adsorption. We directly compare our results with existing scanning tunneling spectroscopy and microscopy measurements. From this comparison, we infer that the observed experimental behavior at thermal energies may be due to ligand and strain effects, as already found for super-thermal incidence energies.
All these studies mentioned above share the use of the PBE functional. This functional has been also used to study the interaction of O2 with Cu surfaces. But, in this case, adiabatic dynamics studies show spontaneous dissociation,29–31 in contrast with the activated behavior found experimentally.32–34 However, as we have recently shown for O2/Cu(111),35,36 theoretical and experimental results can be reconciled if the revised Perdew–Burke–Ernzerhof (RPBE) functional37 is used in the DFT electronic calculations. The dependence of the dynamics calculations on the DFT functional used to compute the PES is not a surprise in molecule/surface systems (see for example ref. 38 and references therein), however such a notable change, from non-activated to activated, is more unusual. In our previous work, carried out at normal incidence using pure classical dynamics, we also showed that, due to the nature of the chemisorption wells, the inclusion of the surface degrees of freedom (DOFs) was essential to reproduce qualitatively King and Wells sticking experimental measurements;31 contrary to results obtained for other O2/metal systems, where the effect of the surface DOFs was found to be relatively small21 or even negligible.39
In our previous studies,35,36 we also simulated the sticking (molecular + dissociative adsorption) probabilities for one and two layers of Cu grown on Ru(0001). For these two systems, our simulations also reproduced qualitatively the experimental measurements at super-thermal energies.31 However, results at normal incidence, without any extra knowledge, were not enough to perform a meaningful comparison with tunneling spectroscopy and microscopy (STS-STM) measurements performed by Otero et al.40 These authors showed a very strong dependence of O2 reactivity as a function of the number of Cu layers, decreasing sharply with the increasing number of Cu layers, and related this behavior to the population of the surface electronic state.40,41
Here, we have performed a complete set of quasi-classical dynamics simulations based on the RPBE-PES published in ref. 36. Our simulations show a complex behavior of sticking probabilities as a function of the incidence angle, following normal energy scaling at low energies, and a negligible effect of the rotational internal molecular DOFs. A proper average of the simulated sticking probabilities at thermal energies has allowed us to perform a direct comparison with STS-STM experimental measurements.
Based on our previous studies35,36 showing the key role that surface temperature plays in O2/CuML/Ru(0001), we have taken into account the surface temperature by means of the generalized Langevin Oscillator (GLO) model47,48 (see ref. 36 for further details). However, unlike those previous calculations, here we have performed quasi-classical dynamics, i.e., we have included the zero point energy (ZPE) of the molecule in the calculations. At this point, we can anticipate that the unwanted phenomenon of vibrational energy leakage associated with the quasi-classical calculations plays a minor role here, because of the relatively low ZPE of O2 in the gas phase, 95.1 meV (obtained solving the 1D Hamiltonian by expanding the wave function on the basis of B-spline functions49), and the relatively high vibrational softening (the adiabatic transfer of energy from the vibrational to the translational DOF) that makes the ZPE decrease down to 56.2 meV for a molecule close to the surface. In performing the quasi-classical simulations, to obtain meaningful probabilities for the different final reactive channels, we have run up to 20000 trajectories for each set of initial incidence conditions (Ei, θi, φi, v, and J) – see Table 1, where Ei is the total incidence energy, θi and φi are the polar and azimuthal angles, respectively (see Fig. 1), and v and J are the vibrational and rotational states, respectively. At the end of each trajectory, we analyze the atomic or molecular coordinates and velocities to assign the final channel. Thus, the dissociative adsorption channel is defined by a O–O distance r ≥ 2.4 Å (around twice the equilibrium value of O2 in the gas phase) with dr/dt > 0. If the molecule bounces (change the sign of its velocity normal to the surface) five or more times before dissociation, we assign this trajectory to the trapping dissociation channel, otherwise we assign it to the direct dissociation channel. If at the end of the integration time (1 ns) the molecule remains trapped in a molecular chemisorption well, the trajectory is assigned to the molecular adsorption channel.
Study | Rotational | Vibrational | Off-normal incidence |
---|---|---|---|
φ i (deg.) | 0 | 0 | 0 |
θ i (deg.) | 0 | 0 | 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85 |
v | 0 | 1, 2, 3, 4, 5, 6 | 0 |
J | 1, 3, 9, 31 | 0 | 0 |
E i (eV) | 0.01, 0.03, 0.06, 0.09, 0.12, 0.18, 0.21, 0.24, 0.27, 0.3, 0.4, 0.5, 0.6 | 0.0001, 0.005, 0.01, 0.03, 0.06, 0.09, 0.12, 0.18, 0.21, 0.24, 0.27, 0.3, 0.4, 0.5, 0.6 | 0.06, 0.1, 0.18, 0.3 |
Number of sets | 52 | 90 | 58 |
It is also important to note that if we compare Fig. 2(a) with Fig. 6(a) of ref. 36, where reaction probabilities were obtained from pure classical trajectory calculations, we can conclude that classical (C) and quasi-classical (QC) simulations only differ by a small shift of the QC probabilities towards lower energies, roughly 40 meV. This shift is compatible with the vibrational softening experienced by the molecule when approaching the surface. This small shift does not have any significant effect when compared with typical King and Wells results measured at super-thermal energies.31 However, as we show below, the effect of vibrational softening, even as small as the present one, cannot be ignored at thermal energies.
We have also performed calculations for vibrationally excited molecules, whose presence in the experimental molecular beams, specially at thermal energies, is expected to be small, but not at super-thermal energies. The results are shown in Fig. 3. As can be seen the O2/Cu/Ru(0001) system becomes non-activated when O2 is vibrationally excited. As shown in Fig. 3(a) the total sticking probability for O2(J = 1, v = 1) is close to 0.9 at very low incidence energies. It can also be seen that trapping dissociation decreases monotonously with the incidence energy, while direct dissociation does the opposite. These results point to a very strong vibration-translation coupling, and therefore, to a very efficient transfer of energy from the vibrational to the translational motion. A similar strong coupling has been observed previously for O2/Ag(110),22 where a vibrational efficacy (i.e. the efficiency of vibrational energy to promote reaction56) was found to be larger than one. In spite of this, the O2/Ag(110) system still remains activated for v = 1. Coming back to the O2/Cu/Ru(0001) system, one can see (Fig. 3(c)) that by further increasing the vibrational energy, the sticking probability decreases at low incidence energy, and the sticking curves exhibit a subtle non-monotonous behavior. This is the consequence of the non-monotonous behavior of direct dissociation, which governs the total dissociation at high vibrational energies (see Fig. 3(b)). Similar findings have been reported for H2/metal surface systems in ref. 57, where it was shown to be the consequence of the balance between the ability of the vibrationally excited molecule to reach the appropriate orientation for dissociation and the number of accessible reaction paths. Nevertheless, as shown in Fig. 3(c), the sticking probability varies very little in the range v = 1 − v = 6. At this point, it is worth pointing out that, although molecular beam experiments for vibrationally excited O2 molecules are not available in the literature, they could be carried out using pulsed narrow bandwidth laser Raman excitation, as suggested in ref. 58, or using induced adiabatic Raman passage (SARP), as proposed in ref. 59.
Finally, we will mention that the reaction probabilities barely depend on the surface temperature in the 250–550 K range.
For θi higher than θtes the reaction probabilities follow normal energy scaling. This can be clearly seen in Fig. 5, where we have plotted the sticking probabilities as a function of the normal energy, Eicos2(θi). We can observe normal energy scaling up to E⊥ = 75 meV. This result has important implications when comparing with experimental results, as discussed in Section 3.3.
Fig. 5 Quasi-classical total sticking probabilities for O2(v = 0, J = 0) on Cu1ML/Ru(0001) as a function of the normal energy. |
Finally, we should point out that the results shown in this section have been obtained for molecular incidence along the azimuthal angle φi = 0° (see Fig. 1). However, we have checked that, qualitatively, similar results are obtained for an incidence along other φi angles. We have also verified that these results do not depend on the surface temperature within the range of 250–550 K.
(1) |
In ref. 40, results for O2/CunML/Ru(0001) for up to n = 4 monolayers are available. However computing sticking probabilities beyond n = 1 is not an easy task. A single monolayer of Cu grows pseudomorphically on Ru(0001), but adding a second monolayer leads to a unit cell periodicity (16 × ).64 The periodicity is lost for more than two monolayers. Therefore, building a DFT-based PES for O2/CunML/Ru(0001) with n > 1 is virtually unapproachable from the computational point of view. However, as shown in ref. 36, we can reasonably estimate the sticking probability for O2/Cu2ML/Ru(0001) by shifting the sticking probabilities obtained for O2/Cu1ML/Ru(0001) by an amount equal to the difference between the minimum reaction barrier for n = 1 and the average minimum reaction barrier for n = 2. As discussed in ref. 36, in the (16 × ) unit cell we can distinguish three zones (see Fig. 9 of ref. 36), fcc, dislocation, and hcp, characterized by three different minimum reaction barriers, 30, 56, and 57 meV, respectively. Thus, the shift δE is computed as:
(2) |
The accuracy of this estimation method relies, beyond the barrier heights, on the differences between the PESs of O2/Cu1ML/Ru(0001) and O2/Cu2ML/Ru(0001). In ref. 36, we showed that the shapes of 2D cuts corresponding to the fcc region of O2/Cu2ML/Ru(0001) were very similar to those obtained for O2/Cu1ML/Ru(0001), and the shapes of the 2D cuts corresponding to the hcp and dislocation regions of O2/Cu2ML/Ru(0001) were also qualitatively similar to those obtained for the fcc region, the barrier heights being the only appreciable difference. Therefore, once the molecules overcome the minimum reaction barriers in O2/Cu2ML/Ru(0001), they are expected to follow a similar reaction path as in O2/Cu1ML/Ru(0001). Hence, the dynamics results are expected to be similar except for a shift reflecting the different reaction barrier heights. The validity of this estimation method is supported by the good qualitative agreement between classical theoretical results and Kings and Wells experimental measurements (see Fig. 11 of ref. 36). The quasi-classical total sticking probabilities for O2/Cu2ML/Ru(0001) estimated using this procedure are shown in Fig. 6. It is worth noticing that, if the shapes of the PES for the surface terminations (fcc, hcp, and dislocation) had been very different, we would have to follow a more general (and computationally much more expensive) procedure, consisting of (i) the calculation of the PES for the three systems, (ii) the computation of the corresponding sticking probabilities, and (iii) performing a weighted average of the former probabilities.
Fig. 6 Quasi-classical total sticking probabilities of O2(v = 0, J = 0) on Cu1ML/Ru(0001) (black triangles) and Cu2ML/Ru(0001) (red circles) as a function of the normal incidence energy. |
Finally, we have applied eqn (1) to the sticking probabilities shown in Fig. 6 to obtain the thermally averaged sticking probabilities (S0), which can be directly compared with the experimental data of ref. 40. We show this comparison in Fig. 7. One can see that our theoretical simulations reproduce fairly well the variation of S0 as a function of the number of Cu monolayers. In the case of two monolayers the agreement is even quantitative, although we should keep in mind that this probability is just an estimation. It is worth noticing that experimental uncertainties are higher for the one-monolayer system. For the sake of completeness, we have also included S0 values obtained from classical sticking calculations. These probabilities also show the right trend, but are lower than the experimental values and those obtained from the quasi-classical calculations.
Fig. 7 Thermally averaged sticking probabilities of O2/CuML/Ru(0001) as a function of the number of monolayers ML. Red: circles classical results; green triangles: quasi-classical calculations; black squares: experimental data from ref. 40. |
This qualitative agreement between theory and experiment, at thermal energy, allows us to conclude that the surface state population is not the only possible explanation for the sharp decrease of the O2 sticking probabilities when moving from Cu1ML/Ru(0001) to Cu2ML/Ru(0001), as previously suggested,40,41 because our PES does not describe such a state properly due to the limited number of Ru(0001) layers included in our DFT calculations of the PES. Instead, the sharp decrease resulting from our calculations is mainly due to strain effects, as already pointed out in our previous work at super-thermal energies.36 Indeed, in Cu1ML/Ru(0001), the Cu overlayer adopts the lattice parameter of Ru(0001) (2.71 Å), and this strain, together with the modification of the electronic structure induced by the binding of the Cu atoms to the Ru atoms (ligand effect), causes a substantial reduction of the reaction barrier compared to that of Cu(111) – a pure Cu surface with the same symmetry as Cu1ML/Ru(0001) – from 97 meV to 26 meV. In Cu2ML/Ru(0001), ligand effects become negligible, and the strain is slightly relaxed, since the average lattice parameter decreases to ≈2.63 Å, while the average minimum reaction barrier increases up to almost 49 meV, leading to the steep drop of the total sticking probability. As the number of Cu layers increases, the average lattice parameter is expected to go on decreasing, and the average minimum reaction barrier to increase even more, thus causing a further decrease in the sticking probability, until the system becomes a Cu(111)-like system.
The current analysis completes our previous study carried out for super-thermal energies.36 The whole picture shows that our theoretical model, based on quasi-classical dynamics and DFT-based PESs modeled considering a low number of metal layers, is able to qualitatively describe the interaction between O2 and CunML/Ru(0001) (n = 1, 2) from thermal40 to super-thermal31 energies.
Finally we can conjecture about the origin of the remaining disagreement between theory and experiment, beyond possible experimental inaccuracies. Plausible sources of inaccuracies of our theoretical simulations could be due to the following:
• A poor description of surface phonons: to include the effect of the surface temperature, we have used the GLO model, which accounts for the molecule-surface energy exchange, thermal fluctuations and energy dissipation to the bulk, but does not account for the individual motions of the surface atoms. However, as shown for H2 dissociation on Cu(111),65 by means of ab initio molecular dynamics (AIMD), the surface atom motion may have a measurable influence on both the barrier heights and locations, which would have an influence on the sticking probabilities and, therefore, on the thermallyaveraged sticking probability. Here, we should point out that running AIMD for an O2/metal surface system is not an easy task, due to the possibility of introducing unphysical spin-flipping from triplet to singlet multiplicity states of the molecule before it reaches the reaction barriers. This phenomenon may confer extra energy to the molecule to overcome the reaction barriers and dissociate regardless of the incidence energy, which may lead the system to behave as non-activated.
• We cannot rule out possible non-adiabatic mechanisms, such as electron–hole excitation or charge transfer from the surface to the molecule. Based on previous results for O2/metal surfaces,20,28 electron–hole excitations are expected to play a minor role. Charge transfer, on the other hand, cannot be totally ruled out, although, in view of the qualitative agreement with the experimental data, it seems to play a less important role than in the case of O2/Al(111).16–19
• A third source of inaccuracy could be the semi-local exchange–correlation functional used in the calculations. Although the RPBE functional employed here yields much better results than the PBE functional, previously used to study this system, more accurate results could be obtained using the specific reaction parameter (SRP) strategy,66 which has been already tested for a number of H2/metal surface (see, for example ref. 67–69 and references therein) and CHD3/metal surface systems.70 However, to properly implement this strategy, more experimental data would be needed than currently available.
This journal is © the Owner Societies 2021 |