Chun-Ran
Chang
ab,
Zhi-Jian
Zhao
b,
Klaus
Köhler
b,
Alexander
Genest
b,
Jun
Li
a and
Notker
Rösch
*bc
aDepartment of Chemistry and Key Laboratory of Organic Optoelectronics and Molecular Engineering of Ministry of Education, Tsinghua University, Beijing 100084, China
bDepartment Chemie and Catalysis Research Center, Technische Universität München, 85747 Garching, Germany
cInstitute of High Performance Computing, Agency for Science, Technology and Research, 1 Fusionopolis Way, #16-16 Connexis, Singapore 138632, Singapore. E-mail: roesch@mytum.de
First published on 27th September 2012
We studied computationally the leaching of palladium from the ideal surface Pd(111) and its various structural defects at different coverages of CO, using density functional calculations on slab models. Accordingly, the energy required for leaching of a single Pd atom from a bare surface is quite large, at least ∼270 kJ mol−1. In a CO atmosphere at low density, PdCO is predicted to be the leaching species; this process was calculated to require at least 225 kJ mol−1, somewhat less than the leaching of a bare metal adatom, 268 kJ mol−1. The energies required for either leaching process (at low CO density), yielding single Pd atoms or PdCO subcarbonyl, correlate in a linear fashion with the coordination number of the Pd center to be leached. At high CO coverage, leaching of Pd subcarbonyl species, Pd(CO)x (x = 2, 3), was calculated to be thermodynamically favorable in several cases, providing direct theoretical evidence for the feasibility of Pd leaching in a dense CO atmosphere. In a qualitative fashion, we also explored possible leaching mechanisms, comprising one or two steps.
Even more common are leaching processes when solid catalysts are applied in a liquid medium. Here, leaching of the active metal into the surrounding liquid reaction mixture is often assumed to be the result of solvolysis of metal–oxygen bonds by which the catalyst is attached to the support, e.g., silica or alumina. Here as well, it is of key importance to understand what is happening when metal atoms are leached from metal surfaces or metal nano-particles. Leaching does not only cause loss of catalyst activity, but results in contaminated products and represents an economic burden on any catalytic process. Interesting timely examples are related to solid catalysts, e.g., supported noble metals in liquid phase reactions for fine chemical synthesis and pharmaceutical applications,6 as well as to electrocatalysis, e.g., leaching of Pt in low temperature fuel cells.7,8
Palladium is one of the most interesting transition metals applied as catalysts in organic synthesis. Because of a unique combination of various properties relevant to catalytic cycles, the element and its compounds are the catalysts of choice for a number of rather diverse reactions, such as hydrogenation, oxidation, and C–C coupling reactions of Heck9–12 and Suzuki13,14 types. In many cases these catalysts have been applied as supported (or also unsupported) palladium nano-particles in liquid phase. In particular for the latter reaction type, i.e., carbon–carbon coupling reactions of Heck and Suzuki types, a variety of authors did not regard leaching primarily as a negative phenomenon as we have discussed exclusively up to now, but proposed Pd leaching as a prerequisite for high catalytic activity in those reactions.15–23 These studies revealed that Pd nanoparticles serve as a reservoir of soluble, truly active species in a catalytic cycle. In many cases Pd is re-deposited onto the solid surface (support or Pd nano-particles) at the end of the reaction. In contrast, several recent papers reported that Heck and in particular Suzuki coupling reactions (can) occur at the surface of Pd nanoparticles.24–29 Thus the nature of the catalytically active species in these coupling reactions is the subject of intense debate.30 While leaching of Pd into solution (and re-deposition) has definitively been proven in many experiments, mechanisms of leaching remain largely unclear in detail.
The motivation of this paper is to examine elementary steps of such leaching processes at an atomic level by quantum chemical methods. Leaching processes at solid–liquid interfaces obviously are rather complex as they involve solid surfaces, solvent, substrates and additives. Therefore, it seems reasonable to approach modeling such challenging processes by addressing a simpler model system at a solid–gaseous interface, namely catalyst deactivation due to loss of metallic species into the gas phase by chemical reactions as mentioned above.
In such model systems, carbon monoxide, CO, can serve as a suitable ligand or “transporting agent”, referring to a CVT half reaction. Carbon monoxide is commonly used as a specific probe for studying properties of catalysts. The chemisorption of carbon monoxide on transition metals has intensely been studied, both experimentally and theoretically.31–39 Accordingly CO is able to form volatile carbonyl species with transition metal atoms, e.g., M(CO)x, with M = Ni, Pd, Pt; x = 1–4. However, nickel tetracarbonyl Ni(CO)4 was the only metal carbonyl to be prepared or purified at room temperature and atmospheric pressure directly from gaseous CO and the metal, in the so-called Mond process.40 Using time-lapsed STM, Besenbacher et al. observed CO induced Ni separation from a Au–Ni surface alloy when exposed to a high pressure of CO.41 Pd subcarbonyls can also be prepared through the interaction of CO with stepped surfaces of palladium at 298 K and a CO pressure of 1.3 × 10−4 Pa by Pulsed Field Desorption Mass Spectroscopy (PFDMS),42 providing an experimental indication for Pd leaching in a CO atmosphere.
Although metal leaching in a CO atmosphere was scarcely reported, CO adsorption-induced structural changes can be found in many experimental reports. By studying Pd deposition onto a model alumina surface in a CO atmosphere, Freund et al. found that Pd carbonyls formed may substantially change the properties of deposited metal particles, indicative of a strong interaction between Pd and CO.43 Using electronic structure calculations and Monte-Carlo simulations, Shan et al. revealed that PdAu alloy surfaces can change from Au-rich to Pd-rich due to strong interactions between Pd and CO.44 A recent experimental study revealed also leaching of Pd atoms from Pd–Au particles.45 Similarly, CO induced changes of the surface composition were reported on CuPt alloys.46 CO driven morphological changes also take place on gold surfaces. Hrbek and co-workers observed that low-temperature adsorption of CO on gold surfaces results in a transformation of hexagonal islands into round ones and the formation of nano-sized gold particles.47 Chemisorption of CO molecules was found to induce significant structural changes in small gold clusters, from two- to three-dimensions.48
Inspired by the findings just described and the resulting discussions, we explored in the present computational work the possibility of Pd leaching from solid Pd surfaces by means of CO chemisorption using density functional theory (DFT). The results from these simple models shall form the basis for a better understanding of more complex systems (heterogeneous liquid-phase catalysis) and for estimating the as yet unclear influence of entropic effects. The model surfaces employed in this study on palladium consist of the commonly used (111) surface, defects on (111), surfaces with kink and step structures, exposing Pd atoms to be leached with different coordination numbers (Fig. 1). We systematically studied Pd leaching from bare Pd models as well as models with low or high CO coverage. Based on our computational results, we demonstrate that Pd leaching is facile from low-coordinated sites at high CO coverage, in many cases accompanied by a release of energy. We will also discuss possible elementary steps of Pd leaching mechanisms on models of high CO coverage.
![]() | ||
Fig. 1 Sketches of the periodic slab models. (a) Pd1 on Pd(111), CN = 3; (b) Pd3 line defect on Pd(111), CN = 4; (c) Pd3 triangle on Pd(111), CN = 5; (d) kink at Pd(865), CN = 6; (e) step at Pd(221), CN = 7. Atoms colored in dark blue represent additional atoms on Pd(111), kink atoms of Pd(865), and step atoms of Pd(221). For Pd(865) and Pd(221), the upper layer of atoms are in light blue, terrace atoms are in gray. |
We applied periodic models of surface slabs with four atomic layers, with a vacuum region of at least 1 nm between the slabs. Atoms in the “bottom” two layers of the slabs were kept fixed at the theoretical bulk-terminated geometry (Pd–Pd = 280 pm). The adsorbates were bound at the “upper” side of the slab models. The two “top” layers of metal atoms were allowed to relax during geometry optimizations, together with the adsorbate(s), until the force on each atom was less than 2 × 10−4 eV pm−1. To take charge transfer between the adsorbate and the substrate into account, the IDIPOL tag was set to 3. When carrying out a normal mode analysis, we tightened the SCF energy convergence from (standard) 10−4 eV to 10−6 eV.
For some of the crucial Pd leaching steps, we calculated the changes in the Gibbs free energy under standard conditions.56 We checked the accuracy of the standard approach for vibration frequencies, as implemented in the program VASP, for the low-energy modes of the adsorbates, i.e., for frustrated translational (FT) and rotational (FR) modes, by evaluating frequencies from fits to point-wise generated energies along the modes. Only for the mode of lowest energy (e.g. 41 cm−1 for the T + B adsorption complex; see below), we determined a notable uncertainty in the frequency. However, the resulting uncertainty in the entropy term of the free energy was less than 2 kJ mol−1 at 298 K, an acceptable error margin for the problem under study. Thus we treated low-frequency modes in a standard fashion.
First we examined models with low CO coverage (1/12 ML), i.e., one CO adsorbed on the models just introduced. For each model, we explored all types of CO adsorption sites with regard to the Pd atom to be leached; we chose the most stable adsorption complex for further inspection. The adsorption energy Ead of an adsorbate was determined as Ead = Eads/sub − (Eads + Esub); when convenient, we also shall discuss desorption energies Edes = −Ead. Here Eads/sub is the total energy of the slab model, covered with the adsorbate in the optimized geometry; Eads is the total energy of the adsorbate in the gas phase (ground state); Esub is the total energy of the clean substrate. With these definitions, negative values of Ead imply a release of energy or a favorable process.
For addressing high coverages of CO, we selected the three models Pd1, Pd3 line, and Pd(865). We made some assumptions regarding the exact coverage and the distribution of the CO adsorbates to limit the modeling effort. At low coverage, three-fold hollow sites were reported to be most favorable for CO adsorption on Pd(111).58 With increasing coverage, a small fraction of molecules preferably adsorbs at bridge and top sites, as shown in various experimental59–61 and theoretical62 studies. Experimentally the saturation coverage of CO on Pd(111) was determined at 0.75 ML;59,63,64 the adsorption pattern, as revealed by low-energy electron diffraction, is a (2 × 2)-3 CO structure63 which, in a (2 × 2) surface unit cell, comprises one CO molecule at a top position and two molecules at hollow positions. Higher values of saturation coverage were reported on other surfaces: 0.80 ML on Pd(100), 0.94 ML on Pd(110), and 0.89 ML on Pd(210).65
Guided by these experimental findings, we constructed high coverage models of 0.75 ML (9 adsorbates per unit cell; see above) invoking two principles. (i) Priority was given to hollow sites and a small fraction of adsorbates were assigned to bridge and top sites; (ii) adsorbates should be distributed as uniformly as possible throughout the surface unit cell. For convenience, we refer to the CO molecules as CO-L or CO-E depending on whether they directly interact with a Pd atom to be leached and are located in its environment. Structures of surface models are named by the adsorption sites of CO-L. For example, a system with 3 CO-L molecules, one each at a top (T), a bridge (B), and a hollow (H) site, is labeled as T + B + H. For isomers of the same CO-L adsorption type on one surface model designators (1), (2), etc. are added. Besides the adsorption energy and desorption energy introduced above in the low coverage case, two further types of energies are used to characterize high-coverage models. The total adsorption energy, Ead(9CO) of 9 CO molecules in one unit cell, refers to the total energy release when adsorbing all these 9 CO molecules at the bare surfaces. The average CO adsorption energy is obtained as Ead(9CO)/9.
For each of the three models Pd1, Pd3 line, and Pd(865), we explored several distributions of 9 CO adsorbates, the most stable of which are shown in Fig. 2. Apparently the adsorbates are well dispersed, with CO–CO distances in a reasonable range. Most inter-adsorbate distances, as measured by the C–C distances, are larger than 300 pm; only 10–20% fall in the range 270–300 pm. The lateral distribution of CO on the Pd surface may be subject to van-der-Waals interactions, not well represented in standard GGA functionals.66–69 With a DFT-D approach we estimated the interaction solely between CO molecules at ∼8 kJ mol−1.70,71 The interaction of CH2 groups of alkanes features a dispersion contribution of ∼10 kJ mol−1 in their interaction with a Pt(111) surface; this energy value may be taken as upper bound for the dispersion interaction of CO and the metal surface.67,72
![]() | ||
Fig. 2 Selected high coverage models with 9 CO molecules per unit cell. (a) Pd1 at Pd(111); (b) Pd3 line defect on Pd(111); (c) kink defect of Pd(865). The sketches depict the most stable structures calculated. |
CO adsorption predominantly occurs at hollow sites, with the ratio of hollow:
bridge
:
top sites as 6
:
1
:
2 for the Pd1 model, 5
:
4
:
0 for the Pd3 line model, and 7
:
1
:
1 for the Pd(865) model. Average CO adsorption energies for the most stable configurations were calculated as −145 kJ mol−1, −146 kJ mol−1, and −147 kJ mol−1, respectively. Note that these values are higher (by absolute value) than the average CO adsorption energy, −141 kJ mol−1, calculated for the experimental (2 × 2)–3 CO structure. This may have been expected as the larger surface area of the defect models, compared to the ideal (111) surface, admits a reduction of the inter-adsorbate repulsion. Recently dense layers of CO on Pt(111) and Rh (111) surfaces have been studied in some detail in kinetic Monte Carlo studies.73
In the high coverage case, ∼0.75 ML, we mainly focused on leaching of Pd(CO)2. As stated above, low-coordinated defects might accommodate more CO molecules than the ideal (111) surface. Therefore, we introduced one more CO molecule per unit cell, resulting in a CO coverage of ∼0.83 ML with 3 CO molecules directly interacting with the Pd atom to be leached. For that higher CO coverage, we explored in addition leaching of Pd(CO)3.
Pdx(s) → Pdx−1(s) + Pd(g) | (1) |
In the optimized structures, the small Pd moieties adsorbed on Pd(111) (Pd1, Pd3 line, Pd3 triangle, Fig. 1a–c) uniformly exhibit shorter interatomic distances than Pd bulk.74,75 For example, the distances between the Pd adatom of the Pd1 defect and its three nearest-neighbor Pd atoms of the (111) surface are shortened by 20 pm compared to the Pd–Pd distance of 280 pm, calculated for bulk Pd. In the models with Pd3 line and Pd3 triangle defects, the nearest-neighbor distances within these adsorbate moieties are 15–20 pm shorter than the bulk reference. Such shortening of bonds in metal clusters with decreasing particle size was demonstrated in several experimental76,77 and theoretical38,78,79 studies. Metal clusters contract notably as the surface-to-volume increases; this phenomenon can be rationalized as result of surface tension or, alternatively, by invoking bond order conservation.38,79
Energetically, the calculated desorption energies Edes(Pd) from the bare Pd models, or the reaction energies of eqn (1), range from 268 kJ mol−1 to 449 kJ mol−1 (Table 1). These values indicate that Pd leaching from bare Pd models will only occur under extreme conditions. As expected, the desorption energy of a Pd atom increases with increasing CN, exhibiting a very good linear correlation (Fig. 3). Recall that the average cohesive energy per atom of Pdn clusters has been calculated to exhibit such linear dependence on the average CN.74 The lowest desorption energy of a Pd atom, 268 kJ mol−1 for the model Pd1, implies that leaching of even a fully exposed single Pd atom from a bare Pd(111) surface is unlikely under normal conditions. All other models yield even higher desorption energies, rendering Pd leaching even more difficult.
Defecta | CNa | E des b (Pd) | Sitec | E ad c (CO) | ΔEd | E des e (PdCO) | ΔGdesf (PdCO) |
---|---|---|---|---|---|---|---|
a Defects as described in Section 3 and coordination numbers CN of the Pd atom to be leached. b Reaction energy of the leaching of a Pd atom from bare Pd models, eqn (1). c Most stable site of CO adsorption and adsorption energy, eqn (2). d Overall reaction energy of subcarbonyl formation from CO in the gas phase, eqn (4). e Reaction energy of PdCO leaching from a CO covered surface, eqn (3). f Change in Gibbs free energy, eqn (3), at 298.15 K, 105 Pa. g Various defects on Pd(111): Pd1, Pd3 line defect, and Pd3 triangle, see text. h Kink defect. i Step defect. j Ideal surface. | |||||||
Pd1g | 3 | 268 | Bridge | −186 | 39 | 225 | 166 |
Pd3 lineg | 4 | 307 | Bridge | −197 | 78 | 275 | 218 |
Pd3 triangleg | 5 | 319 | Hollow | −214 | 90 | 304 | 245 |
Pd(865)h | 6 | 361 | Hollow | −194 | 132 | 326 | 266 |
Pd(221)i | 7 | 392 | Hollow | −197 | 163 | 363 | 303 |
Pd(111)j | 9 | 449 | Hollow | −197 | 220 | 417 | 356 |
![]() | ||
Fig. 3 Calculated desorption energies Edes of Pd (triangles) and PdCO (circles) on bare Pd surfaces or surfaces at low CO coverage, for various defects (Table 1), characterized by the coordination number (CN) of the Pd atom to be leached. The linear correlation coefficients R2 are 0.991 (Pd) and 0.988 (PdCO). |
Pdx(s) + CO(g) → CO(ad)/Pdx(s) | (2) |
Subsequently PdCO desorbs into the gas phase from the adsorption complex formed, entailing the energy change Edes(PdCO):
CO(ad)/Pdx(s) → Pdx−1(s) + PdCO(g) | (3) |
The specific positions for CO adsorption on each of our models are displayed in Fig. S1 of the ESI.† For most models, including the Pd3 triangle, Pd(865), Pd(221), and Pd(111), three-fold hollow sites were calculated to be preferred for CO adsorption; with bond lengths Pd–C = 205–207 pm and C–O = 119–120 pm. However, for the models Pd1 and Pd3 line, bridge sites involving low-coordinated Pd atoms are most favorable for CO adsorption, with Pd–C = 196–198 pm, i.e. about 10 pm shorter than at hollow sites.
![]() | ||
Fig. 4 CO adsorption and PdCO desorption processes at low CO coverage. Ead(CO): adsorption energy of CO; Edes(PdCO): desorption energy of PdCO; ΔE: reaction energy of Pdx(s) + CO(g) → Pdx−1(s) +PdCO(g), eqn (4). |
The adsorption energies Ead(CO) range from −185 kJ mol−1 to −215 kJ mol−1 for the most stable sites on the models studied (Table 1). It is well known that defects with low-coordinated atoms exhibit (in absolute terms) higher CO adsorption energies than regular sites of Pd(111).80 If CO adsorbs at top sites of our various models, the adsorption interactions indeed are 13–25 kJ mol−1 stronger than at top sites of Pd(111), at coverage 1/12 ML (Table S1, ESI†). However, among the most stable sites listed in Table 1, only on a Pd3 triangle CO is more strongly bound, by 17 kJ mol−1, than on Pd(111). CO adsorption energies of the most stable sites of the other models are rather close to the value calculated for Pd(111).
CO adsorbates facilitate Pd leaching compared to the results calculated for bare Pd surfaces (Section 4.1). At low CO coverage, the overall leaching process can be written as the sum of eqn (2) and (3):
Pdx(s) + CO(g) → Pdx−1(s) + PdCO(g) | (4) |
Accordingly, the reaction energy ΔE of eqn (4) can be written as ΔE = Ead(CO) + Edes(PdCO). Compared to the energy changes Edes(Pd), eqn (1), reaction energies ΔE of eqn (4) are decreased by the stabilization energy, 229 kJ mol−1 (Table 1), of a Pd atom due to ligation by a CO molecule in the gas phase:
Pd(g) + CO(g) → PdCO(g) | (5) |
The energy changes of the individual leaching steps, Edes(Pd) [eqn (1)] and Edes(PdCO) [eqn (3)], are still quite similar. CO assistance reduces the energy of Pd leaching only slightly, by 15–43 kJ mol−1 (Table 1). The largest effect, reduction by 43 kJ mol−1, was calculated for a Pd1 defect on Pd(111). The adsorbed CO weakens the Pd–Pd bonds which are broken in the leaching process, thus lowering the leaching energy. Interestingly, leaching energies of PdCO subcarbonyls also increase in a quite linear fashion with the CN of Pd to be leached (Fig. 3).
To reduce the gap between computational and actual experimental conditions, we also calculated the Gibbs free energy change ΔGdes(PdCO) of the leaching step, eqn (3), at 298.15 K and 105 Pa, applying thermodynamic corrections to Edes(PdCO). Vibrational frequencies of Pdx−1 before and after PdCO leaching turned out to be quite similar; therefore, the corresponding thermodynamic corrections are negligible, hence the Gibbs free energy of Pdx−1 remains almost the same during the reaction. Therefore, ΔGdes(PdCO) essentially equals the difference in Gibbs free energies between the PdCO moiety in the adsorption complex and the gas phase. According to our models, a free energy change of at least 166 kJ mol−1 (Pd1 model) is required for PdCO removal (Table 1), indicating that PdCO leaching is still difficult to achieve under standard conditions.
Defect | Adsorption sitea | n b (CO) | E totad c (CO) | E (10)ad d (CO) | E des e [Pd(CO)2] | E des e [Pd(CO)3] | ΔGdesf [Pd(CO)2] | ΔGdesf [Pd(CO)3] |
---|---|---|---|---|---|---|---|---|
a CO adsorption sites on the Pd atom to be leached. T – top site; B – bridge site; H – hollow site. b Number of CO molecules per surface unit cell. c Total adsorption energy of all adsorbed CO molecules, per surface unit cell. d Adsorption energy of the tenth CO molecule. e Desorption energy of Pd(CO)x, x = 2, 3, eqn (3). f Corresponding change in the Gibbs free energy change. g Isomer of T + B (1). h Tenth CO molecule added to the model T + B (1). i Isomer of T + T + B (1); tenth CO molecule added to the model T + B (2). j Tenth CO molecule added to the model B + B model. k Tenth CO molecule added to the model T + B. l Tenth CO molecule added to the model B + H. | ||||||||
Pd1 | T + B (1) | 9 | −1304 | 48 | −33 | |||
T + T | 9 | −1249 | −22 | −97 | ||||
T + B (2)g | 9 | −1243 | −28 | −110 | ||||
T + T + B (1)h | 10 | −1327 | −23 | −14 | −22 | −91 | −102 | |
T + T + B (2)i | 10 | −1252 | −9 | −70 | −112 | −145 | −198 | |
Pd3 line | B + B | 9 | −1315 | 141 | 57 | |||
T + B | 9 | −1278 | 104 | 24 | ||||
T + B + Bj | 10 | −1357 | −42 | 109 | 90 | 28 | −2 | |
T + T + Bk | 10 | −1269 | 9 | −3 | 3 | −77 | −79 | |
Kink at Pd(865) | B + H | 9 | −1325 | 152 | 65 | |||
B + B | 9 | −1322 | 119 | 37 | ||||
T + B | 9 | −1302 | 99 | 16 | ||||
T + H | 9 | −1297 | 123 | 40 | ||||
T + B + Hl | 10 | −1367 | −42 | 91 | 101 | 9 | 9 |
For a variety of optimized high-coverage configurations of Pd1 and Pd3 (line), one recognizes that the Pd atoms to be leached are notably shifted, by 30–100 pm, from their original positions on the corresponding bare surfaces through interaction with several ligands CO-L. In fact, this phenomenon can already be observed at ∼0.75 ML coverage. For instance, one end of the Pd3 line defect bends upward, with the angle Pd–Pd–Pd = 160°. At higher coverage, e.g. for ∼0.83 ML, the adsorbate-induced deformations of the Pd defects are even more distinct (Fig. 5). Direct interactions with three CO ligands extend Pd–Pd nearest-neighbor distances of the Pd1 model from initially 260 pm to 294, 304, and 348 pm (Fig. 5a). Upon high coverage by CO, the three atoms of the Pd3 line defect moved from three-fold hollow sites to bridge sites, with the end-on Pd atom lifted from the surface. In this context, recall that in Ni–CO systems a Ni center at a kink site is detached by 80 pm when two CO molecules are attached in top positions.41 These notable structure changes strongly indicate the possibility of subsequent Pd leaching. On the other hand, there is no dramatic change in the morphology of the Pd(865) model at neither coverage studied; this result reflects the higher CN of the kink Pd atom and the relatively stable surface configuration.
![]() | ||
Fig. 5 Optimized structures of defects on a Pd surface at high CO coverage (10 molecules per unit cell). (a) Three adsorbed CO molecules on Pd1 at Pd (111). (b) Three adsorbed CO molecules on the end atom of a Pd3 line defect on Pd(111). |
For all systems studied, the calculated values of the total adsorption energies Ead(9CO) fall in a rather narrow range, from −1243 kJ mol−1 to −1367 kJ mol−1. The average adsorption energies of CO molecules are ∼60 kJ mol−1 smaller (by absolute value) than calculated for the low-coverage models. For example, in the most preferred model T + B + B of a Pd3 line defect, the average CO adsorption energy, −136 kJ mol−1, is 61 kJ mol−1 less favorable than in the corresponding low-coverage model, −197 kJ mol−1 (Table 1). This reduction can be rationalized by repulsive CO–CO interactions at reduced inter-adsorbate distances as they occur at high CO coverage. To this effect, bonding competition among Pd–CO interactions may also contribute.81
In all high-coverage models, desorption energies Edes[Pd(CO)x] of Pd(CO)x subcarbonyls (x = 2, 3) are dramatically reduced, by ∼150 kJ mol−1, with respect to the values calculated at low CO coverage (Table 2). In some cases, desorption energies of Pd(CO)x even become negative, especially for the defects Pd1 and Pd3 (line). Thus leaching of Pd(CO)x moieties becomes thermodynamically favorable. In the following, we will discuss in turn the desorption processes for the three models studied.
For the Pd1 model, the desorption energies of Pd(CO)x vary from 48 kJ mol−1 to −112 kJ mol−1, the latter value indicating an exothermic process. Therefore, the Pd1 system with its low coordination number (CN = 3) is the most favorable defect for Pd leaching (Table 2). We checked two aspects that may affect the desorption energy of Pd(CO)x fragments, especially for models with a CO coverage of 0.75 ML: (i) the adsorption modes of CO-L molecules, and (ii) the distribution of the CO-E molecules. The isomeric systems T + B (1) and T + B (2) have the same CO-L adsorption modes, but differ in their distributions of CO-E molecules on the Pd(111) terrace. In system T + B (2), Edes[Pd(CO)2] is 76 kJ mol−1 lower than that for T + B (1), but overall the system T + B (2) is 61 kJ mol−1 less stable than T + B (1); see the values of Ead(9CO) (Table 2). On the other hand, the configurations T + T and T + B (2) have the same distribution of CO-E molecules, but differ in the adsorption modes of the CO-L molecules; still, their total adsorption energies and the resulting Pd(CO)2 leaching energies are quite similar. Therefore, the value of Ead(9CO) seems to depend on the adsorption modes of CO-E molecules, resulting in different leaching energies of Pd(CO)x.
To further identify aspects in the distribution of CO-L or CO-E that benefit Pd leaching at high CO coverages, we studied two extreme cases. Assuming only two CO-L molecules on the Pd atom to be leached, but without any CO-E on the (111) terrace, we calculated the desorption energies of Pd(CO)2 moieties as high as 183 kJ mol−1 (T + B) and 133 kJ mol−1 (T + T). Therefore, two directly interacting molecules CO-L do not suffice for leaching Pd(CO)2. Hence CO-E molecules indeed play an important role in the overall process. Desorption of Pd(CO)2 becomes easier at even higher coverage, 0.83 ML, with 10 CO molecules per 4 × 3 surface unit cell. The systems T + T + B (1) and T + T + B (2) were modeled by adding a third CO molecule in the structures T + B (1) and T + B (2), respectively, on top of the Pd atom to be leached. The adsorption energies for this tenth CO adsorbate were calculated to be quite low, −23 kJ mol−1 and −9 kJ mol−1 (Table 2), but the resulting values of Edes[Pd(CO)2] were further reduced, from 48 kJ mol−1 to −14 kJ mol−1 and from −22 kJ mol−1 to −70 kJ mol−1 (Table 2). On these very densely covered surfaces, Pd–Pd bonds are weakened further as a result of new Pd–CO interactions; in addition, also CO–CO repulsion between CO-E and CO-L molecules will increase the driving force for PdCOx leaching.
For the models with Pd3 line defects, we calculated desorption energies Edes[Pd(CO)x] (x = 2, 3) ranging from 141 kJ mol−1 to −3 kJ mol−1, in general higher than for the Pd1 models. The adsorption energy of the tenth CO on T + T + B is 9 kJ mol−1, indicating a slightly endothermic process. This low adsorption energy for the tenth CO can be rationalized by a distortion of the surface structure around the Pd center to be leached, after adding the tenth CO (Fig. 5b).
With the kink model Pd(865), we calculated leaching energies for Pd(CO)x (x = 2, 3) ranging from 91 kJ mol−1 to 152 kJ mol−1 (Table 2). These values are somewhat higher than those determined for the Pd1 and Pd3 line models, consistent with the trend for leaching of PdCO at low CO coverage. Although the values of Edes[Pd(CO)x] remain positive on Pd(865), these processes are by more than 200 kJ mol−1 less endothermic than leaching of PdCO at low coverage (Tables 1 and 2).
Similar DFT model calculations have been carried out41 on CO-induced Ni leaching from Ni(754) kink surfaces at a CO coverage of 0.57 ML. [Note that the (754) surface of an fcc metal has the same kind of kink defect as the (865) surface, but the terraces are wider in the latter case.] Using data from that work41 one estimates a reaction energy of ∼60 kJ mol−1 for the process Ni(kink)(CO)2(ad) + CO(g) → Ni(CO)3(g). This value compares very well with the reaction energy, 59 kJ mol−1, of the analogous process on Pd, Pd(kink)(CO)2(ad) + CO(g) → Pd(CO)3(g) (Table 2). The numerical similarity of these results suggests similar propensities for the leaching of Ni(CO)3 and Pd(CO)3. At first glance, the surface coverages of the two processes appear very different, 0.57 ML in the Ni–CO system vs. 0.75 ML in the Pd–CO system. However, the surface density of CO adsorbates in the Ni–CO system is only ∼5% smaller if one accounts for the calculated metal–metal distances: 248 pm for Ni, 280 pm for Pd. Just as on Pd(111), a higher coverage on Ni(111) facilitates leaching, e.g., as found experimentally for a CO coverage of 0.625 ML on Ni82,83 (equivalent to ∼0.78 ML on Pd, i.e., ∼5% lower than the higher coverage of 0.83 ML used in some of the present models). These results, together with the extreme conditions for the experimental leaching of Pd subcarbonyls,40,42 imply that CO-induced leaching is easier from Ni than from Pd surfaces.
Further findings from our computational models at high CO coverage are worth noting. (i) In most of the models studied, leaching of Pd(CO)x tends to be more favorable if CO adsorbs on top of the Pd atom to be leached. In such adsorption complexes leaching does not require breaking additional Pd–CO bonds which will be necessary for CO adsorption at bridge or hollow adsorption sites. However, adsorption on top occurs less frequently due to the higher adsorption energy at bridge or hollow sites;58 recall the ratio 2:
1 of hollow and top sites occupied at 0.75 ML in the structure (2 × 2)–3 CO.63 (ii) Comparing the systems at 0.75 ML and 0.83 ML, after adding the tenth CO molecules, we calculated a notably lower leaching energy for Pd(CO)2, by 32–107 kJ mol−1. Apparently, the newly formed Pd–CO interactions further weaken the Pd–Pd interactions. (iii) Leaching of Pd(CO)2 and Pd(CO)3 at 0.83 ML is expected to be competitive as judged by the similar calculated reaction energies. (iv) Compared to the average adsorption energy of CO (∼120 kJ mol−1), the relatively small value of less than 42 kJ mol−1 (in absolute terms), calculated for the adsorption of the tenth CO, i.e., the third ligand at the Pd atom to be leached, suggests that the Pd atom to be leached is almost saturated.
Experimentally,84 0.75 ML CO covered Pd(111) surfaces can be produced at 300 K and 105 Pa. To simulate practical conditions, we addressed the thermodynamics of the leaching step. As for the models of low CO coverage, we evaluated thermodynamic corrections only for the desorption of Pd(CO)x (x = 2, 3). In both cases, free energy corrections at 298.15 K, 105 Pa, are about −85 kJ mol−1, mainly due to entropy contributions (Table 2). Accordingly, on a Pd1 model, all reaction free energies are negative, suggesting that leaching of Pd(CO)x is thermodynamically favorable at high CO coverage. On the model of the Pd3 line defect, Pd(CO)x can be leached only at ∼0.83 ML. On the Pd(865) kink model, all reaction free energies remain positive, demonstrating that leaching of Pd(CO)x is getting more difficult as the CN increases. In summary, these computational results indicate that Pd leaching via subcarbonyl formation is thermodynamically feasible under experimental conditions when a Pd surface with suitable defects is exposed to a dense CO atmosphere.
We discuss here mainly the results for the system T + B (1); the changes in the other three cases are quite similar. Before Pd(CO)2 completely desorbs into the gas phase, the Pd atom to be leached, together with two CO-L ligands, at first shifts from its original hollow site to a bridge site, continuing to a top site. Afterwards it desorbs into the gas phase, suggesting that the three Pd–Pd bonds are broken one by one. The C–Pd–C angle in Pd(CO)2 gradually opens from 97° to 160° as the subcarbonyl moiety departs from the surface. When the CO-E molecules re-arrange on the Pd(111) surface, one CO shifts from a top to a bridge site, to reduce CO–CO repulsion after desorption of Pd(CO)2.
According to the energy profiles of the various Pd1 leaching processes (Fig. 6), the desorption barrier of Pd(CO)2 can be estimated as ∼85 kJ mol−1 for T + B (1), ∼35 kJ mol−1 for T + T, ∼40 kJ mol−1 for T + T + B (1), and ∼35 kJ mol−1 for T + T + B (2). These relatively low barriers corroborate that leaching of Pd(CO)2 in a dense CO atmosphere can take place under experimental conditions. The higher barrier of the system T + B (1) is consistent with the finding that an additional Pd–C bond (of the bridging CO) has to be cleaved. All barriers are passed when the Pd center is 100–150 pm above its original position. At a height of 300 pm, Pd(CO)2 can be considered to be completely desorbed into the gas phase as judged from the subsequent flat energy profiles.
![]() | ||
Fig. 6 Total energy change ΔE as Pd(CO)2 leaches into the gas phase from Pd1 on Pd(111) at high CO coverage, as a function of the change Δh in the height of the leaching Pd center above its original location. Black curves denote systems with 9 CO molecules per surface unit cell: T + B (1) (filled circles), T + T (empty circles); red curves denote systems with 10 CO molecules per surface unit cell: T + T + B (1) (filled triangles), T + T + B (2) (empty triangles). To keep the four systems comparable, the energy of a CO molecule in the gas phase is added to the results of the models T + B (1) and T + T with 9 CO molecules. For the designations of the adsorption complexes, see text and Table 2. |
![]() | ||
Fig. 7 Pathways for the diffusion of a Pd atom (or a PdCO moiety) approximated key positions along the routes A → B → C. (a) Pd(1) of a Pd3 line defect on Pd(111) moving away from the defect. (b) Pd(1) from a kink site of Pd(865) moving to the terrace. |
![]() | ||
Fig. 8 Change of the total energy as a Pd atom (or a PdCO moiety) moves on an otherwise bare surface along the diffusion pathways sketched in Fig. 7 as a function of the Pd(1)–Pd(2) distance. (a) Pd3 line defect on Pd(111). (b) Kink defect of Pd(865). Red curve (circles): diffusion of a Pd atom on a bare surface; black curve (squares): diffusion of a PdCO moiety from the starting structure with one CO at the bridge Pd(1)–Pd(2); blue curve (triangles): diffusion of PdCO from a starting structure with one CO on top of Pd(1). |
In a model of the Pd3 line defect on an otherwise bare Pd(111) surface, Pd(1) was assumed to diffuse to a hollow site at location C (Fig. 7a) via locations A and B, passing a barrier of ∼55 kJ mol−1. Subsequent diffusion steps of Pd(1) on Pd(111) are much easier, with barriers calculated as low as 12 kJ mol−1 between two hollow sites. In the low coverage case, adsorbed CO will affect the diffusion barrier. With CO adsorbed on top of Pd(1), the initial barrier of Pd(1)CO diffusion was estimated as ∼45 kJ mol−1, i.e., ∼10 kJ mol−1 lower than on a surface free of CO adsorbates (Fig. 8a). This reduced barrier reflects the weaker Pd–Pd bonds due to the Pd–CO interaction. In contrast, for CO at a bridge site between Pd(1) and Pd(2), we estimated the diffusion barrier of Pd(1)CO ∼30 kJ mol−1 higher than for Pd diffusion on a bare Pd(111) surface, due to the fact that a second Pd–C bond has to be cleaved.
The diffusion pathway of the kink atom Pd(1) on a bare Pd(865) surface may be separated into two parts. First, atom Pd(1) crosses a barrier of ∼75 kJ mol−1 to reach position A (Fig. 7b), a relatively stable position. Further diffusion takes place to reach the neighboring hollow site C, after having crossed a barrier of ∼65 kJ mol−1, close to location B. At position A, Pd(1) forms a new bond with atom Pd(3), leading to the Pd3 quasi-line structure Pd(1)–Pd(3)–Pd(4) which also presents a local minimum along the energy profile (Fig. 8b). As this Pd(1)–Pd(3) bond needs to be broken during the second diffusion step, from A to C, the approximate diffusion barrier is by ∼50 kJ mol−1 higher than the diffusion barrier of the analogous step, A to C, calculated with the model of the Pd3 line defect, where a Pd atom at position A no longer is affected by strong lateral interactions. As discussed above, the fragment Pd(1)CO encounters lower barriers due to the CO ligand adsorbed on top, ∼65 kJ mol−1 for the first step and ∼40 kJ mol−1 for the second step, but Pd(1)CO diffusion becomes more difficult when its CO is initially bound in a bridging fashion between Pd(1) and Pd(2).
In summary, on models of Pd3 line defects and Pd(865), Pd diffusion occurs with barriers estimated as ∼55 kJ mol−1 and ∼75 kJ mol−1, respectively. With CO adsorbed on top of Pd(1), the diffusion encounters lower barriers, by ∼10 kJ mol−1. Inspecting the data for direct leaching (Table 2), one notes that leaching energies for Pd(CO)x in general fall in the range 90–140 kJ mol−1 for Pd3 line defects, and 90–150 kJ mol−1 on Pd(865). The corresponding leaching barriers should be higher. Comparing barriers for diffusion of Pd(CO)x to these estimated barriers of direct leaching (Table 2), one concludes that from Pd3 line defects or kink sites of Pd(865) two-step leaching is more likely than one-step leaching.
For models of high CO coverage, we calculated leaching energies of Pd(CO)x (x = 2, 3) significantly lower (by ∼150 kJ mol−1) compared to the corresponding energies from models of bare Pd surfaces or surfaces with low CO coverage. In particular for Pd1 and Pd3 line defects, reaction energies for the leaching of Pd(CO)x become negative in several cases, suggesting thermodynamically favorable processes. Two factors were suggested to contribute to the reduced Pd(CO)x leaching energy: (i) weakening of Pd–Pd bonds when Pd–CO bonds are formed; (ii) CO–CO repulsion. In addition, calculated Gibbs free energies of several processes under standard conditions are exoergic, making Pd leaching plausible in practical experiments.42
Finally, we explored mechanisms of Pd leaching from surfaces with high CO coverage. We estimated barriers of Pd(CO)2 leaching from Pd1 defects at 35–85 kJ mol−1. For the desorption of Pd(CO)x from Pd3 line defects on Pd(111) and kink defects on Pd(865) we suggested mechanisms that involve two steps: diffusion of a Pd atom from a Pd3 line defect or the kink of Pd(865) to form isolated Pd atoms on Pd(111) terraces, followed by the actual leaching of these newly formed Pd1 centers after adsorption of several CO ligands. The barrier for surface diffusion of Pd atoms (on a bare surface) was estimated as ∼55 kJ mol−1 from Pd3 line defects and as ∼75 kJ mol−1 from kinks of Pd(865). For both defects, the barriers of PdCO diffusion were estimated to be ∼10 kJ mol−1 lower. Two-step pathways seem more likely as the corresponding barriers are lower than most of the reaction energies estimated for (direct) leaching of Pd(CO)x; the latter energies can be as high as 141 kJ mol−1 for Pd3 line defects and 152 kJ mol−1 for kinks of Pd(865).
In summary, with model calculations, we provided for the first time direct theoretical insight into Pd leaching in a CO atmosphere and revealed that leaching can occur when a Pd surface with defects is exposed to high pressure of CO.
Footnote |
† Electronic supplementary information (ESI) available: Structures of the most stable CO adsorption site at low coverage and of selected high coverage models, Cartesian coordinates of optimized structures. See DOI: 10.1039/c2cy20441j |
This journal is © The Royal Society of Chemistry 2012 |