- Email: [email protected]
Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents
G Model
ARTICLE IN PRESS
CATTOD-10226; No. of Pages 11
Catalysis Today xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Catalysis Today journal homepage: www.elsevier.com/locate/cattod
Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents Lyudmila V. Moskaleva Institute of Applied and Physical Chemistry and Center for Environmental Research and Sustainable Technology, Universität Bremen, 28359 Bremen, Germany
a r t i c l e
i n f o
Article history: Received 30 December 2015 Received in revised form 25 April 2016 Accepted 16 May 2016 Available online xxx Keywords: Density functional theory Au(321) surface Propylene epoxidation Propylene oxide Selectivity Hydroperoxyl
a b s t r a c t Propylene oxide (PO) is an important bulk chemical used for synthesis of many value-added products. Much effort has been devoted to finding a simple and “green” process for PO production. Promising results in terms of activity and selectivity have been achieved for propylene oxidation on supported gold catalysts with a mixture of O2 and H2 or with O2 and H2 O. In this work a detailed transformation network of competitive reaction pathways following the initial steps of oxidation has been studied theoretically using density functional theory (DFT). The results of calculations question some of the earlier assumptions regarding the mechanism of PO formation. Surface hydroperoxo species (OOH) formed in situ are shown to be responsible for the high selectivity of propylene epoxidation with O2 on gold-based catalysts using hydrogen or water as co-reactants. © 2016 Elsevier B.V. All rights reserved.
1. Introduction Epoxidation of propylene is crucial in the manufacture of propylene oxide (PO), a valuable raw material for producing many value-added chemicals, such as polyether polyols, propylene glycol, and others [1]. At present, two major methods are used for the commercial PO production: the chlorohydrin process and the hydroperoxide processes. These processes proceed via multiple reaction steps and utilize hazardous chlorine and costly hydroperoxides as oxidants [1]. Therefore, efforts have been made to develop a “green” process that could directly and selectively epoxidize propylene to PO. Although silver supported on alumina is well known as a uniquely effective catalyst for ethylene epoxidation, Ag catalysts are far less effective for epoxidation of propylene (and other olefins with allylic C–H bonds) because of predominant total oxidation to CO2 and H2 O [2]. The low selectivity obtained for propylene epoxidation has been attributed to a weaker binding strength of allylic (–CH3 ) hydrogen in propylene than that of the vinyl hydrogen in ethylene [3]. Two other coinage metals, Au and Cu on appropriate supports, have also been tested as catalysts for propylene epoxidation [3]. As in the case of silver, the common challenge identified for all
E-mail address: [email protected]
of them is the need to suppress undesirable reaction pathways leading either to total oxidation or to the formation of other partial oxidation products, e.g., acrolein, acetone, propanal. Hayashi et al. [4], who first reported the selective epoxidation of propylene on supported 2–5 nm Au nanoparticles, proposed to use a sacrificial reductive agent (H2 ) to better control the reactivity of active oxygen species and suppress deep oxidation. This approach showed successful results in terms of the PO selectivity (∼90%), although conversions were low (only 1–2%). Since then, several groups have been working on the optimization of propylene epoxidation over gold-based catalysts and significant progress has been achieved [5–11]. Recently, it was shown that, similarly to H2 , water admixtures can also drastically increase the selectivity to PO [9]. It is therefore reasonable to expect that in both cases a surface hydroperoxide intermediate, OOH, formed in situ could play a role as a milder oxidant than atomic oxygen and could be responsible for the selective PO formation. Although a tentative mechanism of OOH formation and oxidation has been proposed by Haruta et al. [8,12], it has not yet been confirmed by unbiased computational models based on first principles calculations. Herein, we present a comprehensive theoretical investigation of the reaction network involved in the oxidation of propylene on gold on the basis of DFT. We have chosen a stepped and kinked Au(321) surface as a convenient model to study the reactions taking place on the defect-rich surface of a nanoparticle gold catalyst.
http://dx.doi.org/10.1016/j.cattod.2016.05.034 0920-5861/© 2016 Elsevier B.V. All rights reserved.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
2
This model has been previously successfully employed in a number of theoretical studies [13–19]. Although the oxide support is an indispensable component of a catalyst, in this work we focus on the metal component, assuming that the key reaction steps that determine the selectivity take place on the gold surface. This study is the first to compare and contrast the chemisorbed O and OOH species as oxidizing agents for propylene oxidation on gold. We show why the oxidation with hydroperoxo species leads to a higher PO selectivity. An important aspect that has not received enough attention in earlier theoretical studies is the easy isomerization of the commonly considered PO precursor (a five-member ring oxametallacycle, OM2 C) to propanal or acetone, which according to our calculations, always has a lower activation barrier than that required for epoxide formation. Thus, even if the H abstraction from the CH3 group of propylene could be suppressed, that still would not solve the problem of low PO selectivity. Our study shows that, quite remarkably, the propylene oxidation with surface OOH allows for at least three low-energy reaction pathways not going through that five-member-ring OM2 C intermediate, hence, suppressing the isomerization to carbonyl compounds. At the same time, the abstraction of the allylic H from propylene by adsorbed OOH has higher activation energy than in the case of H abstraction by surface O. That explains why high selectivities to PO can be achieved if chemisorbed atomic O as oxidant is replaced by a milder OOH counterpart. Furthermore, our study clearly shows that the key reaction pathways are structure sensitive. The reaction pathways taking place at low-coordinated edge Au atoms go through somewhat different intermediates than on flat terrace sites. In particular, four-member ring oxametallacycles bound on a single kink Au atom seem to be crucial precursors to epoxide formation. The paper is structured as follows. Section 2 describes the models and computational approach. Section 3.1 discusses the reaction pathways of propylene with chemisorbed O. Section 3.2 considers the oxidation with chemisorbed hydroperoxy radical and highlights an important difference to the atomic-oxygen induced pathways. Section 4 summarizes the most important findings.
The calculations have been carried out with the VASP code [23,24] using PBE functional and projector augmented wave (PAW) method [25,26] with an energy cutoff of 415 eV. The choice of the PBE functional is justified by its very good performance in describing bulk properties of transition metals [27,28]. The k point sampling was generated by the Monkhorst-Pack procedure [29]; a 5 × 5 × 1 k point mesh was used for geometry optimizations. The structures were relaxed until the force acting on each atom was ≤ 2 × 10−2 eV Å−1 . The spin polarization has been taken into account for all reactions involving paramagnetic O2 molecule. For the remaining reactions the spin-restricted formalism was used, as justified by test studies for adsorption systems on non-magnetic metallic surfaces [30]. Transition states (TSs) of the reactions were determined by applying the dimer method [31]. To verify the nature of a TS, the minima connected through the TS were identified by following the steepest descent reaction pathway downhill from the TS to the reactant and product valleys. The adsorption energy of a stable molecule Eads (X) was calculated as follows: E ads (X) = (E X∗ − E substrate − E X )
where EX* denotes the total energy of the adsorbed X* on the surface in the optimized geometry; Esubstrate and EX are the total energies of the clean surface (or a surface with pre-adsorbed co-adsorbate in the case of co-adsorption) and of the X molecule in the gas phase, respectively. With this definition, negative values of adsorption energy indicate exothermic adsorption. For a chemisorbed O atom, the adsorption energy is calculated by taking one-half the energy of a gas-phase O2 molecule plus the energy of clean Au(321) as reference. Importantly, the relative energies of various species and transition states shown in the calculated energy profiles are computed using the clean Au(321), gas-phase propylene, gas-phase O2 , and, where relevant, gas-phase H2 as reference. For example, Erel (C3 H∗6 + OOH∗ ) = EC3 H6 ∗ +OOH∗ − EAu(321) − EC3 H6 − EO2 −
2. Models and computational details In this study we use the Au(321) surface, as in Fig. 1(a,b), as a model to represent the defect-rich surface of gold nanoparticles, with the typical structural features expected for these particles, such as small terraces of (111) type, steps and kinks. Of course, the real catalyst has a more complex structure, where the particle size, shape, and the perimeter interface of metal particles and the oxide support definitely play a role in the activity. However, we concentrate on the reaction steps that determine the selectivity and which are expected to take place on the gold surface. Hence, the choice of this relatively simple but computationally affordable model is justified. A comparison with the work of Chang et al. [20], who calculated some of the reactions discussed below using a Au38 nanocluster model, shows that we predict the same chemistry and our results agree qualitatively, only the adsorption energies are much stronger on a nanoparticle surface, which is expected because of a small particle size. The slab model has been constructed using the bulk lattice parameter of Au 4.173 Å obtained from a calculation with the PBE form of the generalized-gradient approximation (GGA) for the exchange-correlation functional [21,22]. The slab model contains 14 atomic layers corresponding to the thickness of the resulting slab of ∼7.2 Å and the vacuum spacing between periodically repeated slabs of ∼8.5 Å. We use a (2 × 1) surface unit cell containing 28 metal atoms. The uppermost 14 metal atoms were allowed to fully relax, whereas the lower 14 atoms were kept fixed at their optimized bulk positions.
(1)
1 EH . 2 2 (2)
where EC3 H6 ∗ +OOH∗ is the total energy of co-adsorbed propylene and hydroperoxyl, EAu(321) is the total energy of the clean Au(321) slab EC3 H6 , EO2 and E H2 are the total energies of gas-phase propylene, O2 and H2 , respectively, and here and further on a * symbol denotes an adsorbed species. A single exception is Fig. 9(b), where gas-phase O2 and H2 O are used as reference species. Relative energies do not include zero-point energy corrections. 3. Results and discussion 3.1. Reaction of propylene with chemisorbed O atoms The epoxidation of propylene on gold and other coinage metals is believed to proceed by analogy to the accepted mechanism of ethylene epoxidation, via an oxametallacycle intermediate formed from addition of surface O to the C=C bond of propylene [32]. A competing H abstraction from the CH3 group of propylene forming surface allyl species is thought to be leading to total combustion [3,33]. Studies of propylene oxidation on oxygen covered Au(111) and Au(100) single-crystal surfaces in ultrahigh vacuum reported either CO2 and H2 O [34] or acrolein [35] as main products but no propylene oxide was detected. The differences in observed main products in Refs. [34] and [35] were attributed to the different methods used for O deposition. Earlier theoretical studies of propylene oxidation by surface O on gold [36,37] focused on the primary reaction steps leading either to metallacycles or allyl but refrained
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
3
Fig. 2. Reaction network of propylene partial oxidation with adsorbed O as oxidizing agent. The starting point is indicated with a blue frame. Reaction steps addressed in earlier studies on extended gold surfaces are marked with green arrows. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Fig. 1. (a) Side and (b) top view of the Au(321) surface. The (2 × 1) surface unit cell is indicated. (c) and (d): Most favorable adsorption sites of propylene. (e)–(h): Most favorable adsorption sites of oxygen.
from exploring the complete reaction network. Herein, we investigated the more extended reaction network, Fig. 2. Particularly important are isomerization pathways of oxametalacycles to carboxylic compounds. As demonstrated earlier in a DFT study of propylene epoxidation over Ag(111) and Ag(100) surfaces combined with Raman spectroscopy and mass spectrometry [38] and shown below, these reactions are in direct competition with the epoxide formation. The reaction scheme in Fig. 2 refers to the case of oxidation with pure O2 (which adsorbs dissociatively) or with preadsorbed atomic O in the case of model studies on single-crystal surfaces, where no H2 or H2 O are added to the reactants stream. The dissociative chemisorption of O2 , see Section 3.2, is probably the rate-limiting step, since all further steps lie energetically lower than the gas-phase propylene and O2 . In Section 3.2 we will explain why the addition of H2 or H2 O to O2 not only facilitates the O2 activation but also results in new possibilities for selective epoxide formation.
and donor orbitals of the metal [39]. The adsorption energies of propylene are at −12 kJ mol−1 on top of the 8-fold coordinated Au atom and −55 kJ mol−1 on top of the 6-fold coordinated Au atom, which is much stronger than the values reported for the adsorption on the Au(111) surface (ca. −5 kJ mol−1 ) [36,37] but much weaker than the adsorption energy on the apex site of a Au38 nanoparticle (-100 kJ mol−1 ) [20] or on top of a single Au adatom on the Au(111) surface (–75 kJ mol−1 ) [37]. Propylene and oxygen atoms, when adsorbed close to each other (but not binding to the same Au atom), co-adsorb favorably with a slight adsorption energy gain of 4–13 kJ mol−1 with respect to the individually adsorbed species. Clearly, low-coordinated atoms at step edges of a gold surface bind the reactants and intermediates stronger than the highercoordinated terrace sites, but not too strong so as to poison the catalyst. Therefore, in the following we mainly discuss the reaction pathways at step edges. Elementary steps leading to allyl and oxametallacycle intermediates have been previously studied on the flat Au(111) [36,37] and on Au(111) with vacancies and adatoms, as well as on straight monoatomic steps of the Au(211) surface [37]. Below we comment on how the current results compare to those from earlier studies.
3.1.1. C3 H6 and O adsorption In our earlier work [17] we have demonstrated that atomic oxygen on the Au(321) surface prefers adsorption sites immediately adjacent to the step edge. These sites are either of the three-fold fcc or the two-fold (“bridge”) type, Fig. 1(e–h). The adsorption energies relative to the gas-phase O2 range from −12 to +4 kJ mol−1 . Similarly, propylene binds preferentially above low-coordinated metal atoms along the step edge in a so-called “-mode” through an interaction of the alkene and * orbitals with the acceptor
3.1.2. Epoxide formation via oxametallacycle intermediates An oxametallacycle is an intermediate attached to a metal surface in a M–O–C–C–M type linkage. In the context of widely studied ethylene epoxidation on silver, two types of oxametallacycles have been considered in the literature: a five-member ring containing two metal atoms (OMME) and a four-member ring containing a single metal center (OME). Earlier DFT calculations on Ag(111) [32] and Ag(110) [40] surface models in combination with the high resolution electron energy loss spectroscopy (HREELS) [32] concluded
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 4
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 3. Energetic profiles (kJ mol−1 ) for the reaction of chemisorbed propylene and O via primary (a) and secondary (b) oxametalacycle intermediates on the Au(321) surface. A * symbol indicates adsorbed species. Only the lowest-energy pathways identified are shown. See also Figs. S1 and S2 for alternative pathways.
that OME, although slightly less thermodynamically favorable than OMME, should be a more likely precursor of ethylene oxide formation on silver. The analogous four-member oxametallacycle structure predicted by Barteau et al. [41,42] as an intermediate for styrene epoxidation on Ag(111) surface was later supported experimentally by Madix group [43,44]. To differentiate between the two types of metallacycles, in this work we refer to a metallacycle with a five-member ring as OM2 C, while the four-member ring is denoted as OM1 C. Examples of both structures are shown in Figs. 3 and 4. Previous DFT studies on the Au(111) surface [36,37] considered only the formation of OM2 C-type intermediates and their further
ring contraction to epoxide. According to our calculations using the stepped Au(321) surface, the OM1 C intermediates lie significantly higher in energy than OM2 C (by 36–70 kJ mol−1 ). Therefore, in the first reaction step the formation of a five-member ring is more favorable. However, our calculations suggest that the ring contraction of OM1 C to epoxide has a lower-lying transition state (see alternative pathways in Figs. S1, S2); therefore, OM1 C is identified in this work to be a key intermediate on the way to epoxide formation, in agreement with the conclusions reached by Barteau et al. [32,40] regarding the mechanism of ethylene epoxidation on Ag surfaces.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 4. Atomic geometries of the key minima and transition states along the lowest energy pathways of epoxide formation via primary (left) and secondary (right) oxametallacycle intermediates, corresponding to energetic profiles of Fig. 3.
5
We explored the addition of chemisorbed oxygen to either C1 or C2 atom of propylene forming a primary (OM2 C ) or a secondary (OM2 C ) oxametallacycle intermediate, respectively. The relative energies of the intermediates and transition states for the lowestenergy pathways of epoxide formation identified on Au(321) are shown in Fig. 3. Basically, the pathways passing through primary and secondary metallacycles are qualitatively very similar, with not so high energy differences, except for the initial addition step, which renders the OM2 C formation kinetically more favorable. In agreement with the earlier work of Baker et al. [37], we found lower activation energies (by 5–19 kJ mol−1 ) associated with the formation of OM2 C compared to OM2 C ; however, the formation of OM2 C is slightly more exothermic, by 3–15 kJ mol−1 . The most thermodynamically favorable OM2 C structures identified bind directly at the step edge, with the C atom attached at a kink site and with the O atom attached to another kink and an 8-fold coordinated Au between the kinks in a bridge mode. Alternatively, C atom can attach at an 8-fold coordinated Au atom and O at a 6-fold coordinated kink Au atom. In this case, the energy of the resulting OM2 C is slightly higher, see Figs. S1 and S2 . The transition states on the way to OM2 C and OM2 C (TS1) have very similar structures, Fig. 4. The surface oxygen atom interacts with one of the carbon atoms of the double bond in the plane perpendicular to the molecular plane of propylene. The C–O distance of the forming bond is 2.17 Å and the C–C distance of the double bond is increased to 1.4 Å. A lower barrier for OM2 C formation might result from an electron donating effect of a methyl group attached to the reactive C atom of the alkene. Not all OM2 C intermediates can be converted to epoxide directly. All OM2 C structures shown in Fig. 3 except one, the OM2 C along the magenta path, first need to pass through a four-member metallacycle OM1 C, where both C and O are attached to the same low-coordinated kink Au atom. The direct pathways, when existing, generally have a higher-lying transition state, see also Figs. S1, S2. The energetically lowest transition state structures for epoxide formation have been found for step-wise pathways going via OM1 C intermediates, whose structures due to a smaller ring size are geometrically closer to the final epoxide product. Remarkably, in all these structures the Au atom incorporated in the oxametallacycle is the kink atom with the lowest coordination number. This Au atom has extra free bonding capacities enabling the formation of multiple bonds with the –C–C–O– backbone of an oxametallacycle, and lowering the energy of the corresponding transition state for epoxide ring closure, TS2, which is geometrically quite similar to the OM1 C intermediate, Fig. 4. All oxametallacycles can undergo a 1,2-H shift reaction forming carboxylic compounds propanal or acetone. The geometries of transition states and products corresponding to the lowest-energy pathways according to Fig. 3 are given in Fig. 5. The transition state geometries present an early character, nearer to the reactant oxametalacycle than to the product, consistent with the strongly exothermic reaction. Notably, for all OM2 C structures the C–H activation to form carboxylic compounds has a lower activation barrier than the highest transition state on the way to becoming an epoxide, see Figs. 3, S1, and S2, providing another reason why with O2 alone as an oxidant no epoxide is actually formed on Au catalysts. Interestingly, this energetic preference is reversed for OM1 C. The formation of carboxylic compounds from OM1 C has higher activation energy than the conversion to epoxide, see the discussion in Section 3.2.2. Therefore, if OM1 C was formed directly from adsorbed propylene and oxygen, its ring closure to form epoxide would have a higher rate than the isomerisation to propanal or acetone. However, we have not found a direct pathway for OM1 C formation from adsorbed O. Even if such a path exists, it probably has higher activation energy than that for OM2 C formation because the latter lies energetically much lower than OM1 C. Hence, OM2 C
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 6
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
oxidation using the Au38 cluster model [20]. However, for such a direct addition to happen, the propylene molecule must approach the O atom from above, so as to make the pz orbital of O interact with the * of the former, see the geometrical arrangement of calculated transition states in Fig. S3. Therefore, this pathway is of the Eley-Rideal type, where propylene reacts with the adsorbed O directly from the gas-phase or from a weakly physisorbed state. We have calculated the activation barriers of 37–52 kJ mol−1 with respect to chemisorbed O and physisorbed propylene for a direct propylene formation, see Fig. 6. These values are in good agreement with the earlier calculated value of 31 kJ mol−1 using the Au38 cluster model [20]. Because an Eley-Rideal mechanism results in a very low pre-exponential factor (due to an extremely short time scale of a gas-surface collision), whereas the calculated activation barrier is significant, this route to PO formation is expected to have an insignificantly low rate in comparison with the competing Langmuir-Hinshelwood-type pathways.
Fig. 5. Atomic geometries of the transition states and products for propanal (left) and acetone(right) formation via primary (left) and secondary (right) oxametallacycle intermediates, corresponding to energetic profiles of Fig. 3.
Fig. 6. Energetic profiles (kJ mol−1 ) for a direct reaction Eley-Rideal (ER) type between the gas-phase propylene and chemisorbed atomic O. A * symbol indicates adsorbed species.
will form in the first C3 H6 * + O* addition step and will preferentially react to form propanal or acetone but not epoxide. To sum up, the reaction between surface O and propylene through oxametallacycle intermediates to epoxide is energetically feasible with the first transition state being the highest in energy, 50–55 kJ mol−1 relative to co-adsorbed C3 H6 and O. However, in addition to the competing H abstraction from the methyl group by surface O (Section 3.1.4), the isomerization of OM2 C intermediates to carboxylic compounds is also energetically favored over the desired epoxide product, thus revealing another obstacle on the way of improving the PO selectivity. 3.1.3. Direct epoxidation of C3 H6 In addition to the widely accepted oxametallacycle route, we have also considered a direct formation of epoxide by a reaction of propylene with chemisorbed O, Figs. 6 , S3. This type of reaction has been proposed in an earlier theoretical study of propylene
3.1.4. Hydrogen abstraction by chemisorbed O and further reactions of allyl Consistent with the earlier study of Roldan et al. on the Au(111) surface [36], an abstraction of a hydrogen atom from the methyl group by adsorbed atomic O was found to have the lowest activation barrier of all primary pathways, 11–45 kJ mol−1 relative to co-adsorbed reactants, with the transition states lying in the most favorable cases below the gas-phase reactants by up to 50 kJ mol−1 , see Fig. 7. We explored different positions of co-adsorbed reactants on the surface and the corresponding reaction pathways. As seen from Fig. 7, the energy profiles and particularly the activation barrier heights vary considerably. The energies of the initial states (coadsorbed propylene and O) lie in most cases higher than those in Fig. 3. This is because in Fig. 3, only the pathways with the lowest overall energy have been shown (see Figs. S1 and S2 for other low-energy pathways), whereas Fig. 7 shows pathways with a larger variation of initial and product geometries. It can be generally observed that the pathways where oxygen in the initial state is bound in a bridge position at a step edge (Path 1 and Path 4, Fig. 8) have lower activation energies of 11–14 kJ mol−1 and are strongly exothermic by 53–69 kJ mol−1 , which can be explained by the fact that a 2-fold coordination is preferred by OH in the final state. In contrast, pathways where O is initially adsorbed at a 3-fold fcc site (Path 2, Path 3, and Path 5, Fig. 8) have higher activation energies, of 22–45 kJ mol−1 and are exothermic by only 10–24 kJ mol−1 . In the final state, OH binds at a 3-fold site (or at a pseudo 2-fold in the case of Path 3) which is not its most favorable binding mode, explaining relatively high energies of the final state and higher activation energies. Path 3 is somewhat special because propylene is interacting with two O atoms due to the translational symmetry of a unit cell, but the described trends still hold for this case as well. The transition state structures of the two groups are also somewhat different. For the latter group, both the O–H and C–H distances in the TS are close to 1.3 Å, i.e., the transferred H atom is in the middle between the C and O centers. This geometry is very similar to that reported in earlier studies on the Au(111) surface [36,37] and on Au(111) with defects and on Au(211) [36], where the Habstracting surface O was in all cases also placed at a 3-fold fcc site. In contrast, if O is bound at a 2-fold bridge site at a step edge, the TS structure is more reactant-like, with a longer O–H distance of ∼1.4–1.45 Å and a shorter C–H distance of ∼1.2 Å. This is in line with the Hummond postulate [45], since reactions of this group are strongly exothermic with low activation energy. Although allyl intermediate is often referred to as precursor to total combustion, that is an arguable point because according to Fig. 7 allyl can readily react to form allyl alcohol or acrolein − the products which are moderately strongly bound and can eas-
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
7
Fig. 7. Energetic profiles (kJ mol−1 ) for allylic H abstraction by the surface O and the subsequent reaction steps on the Au(321) surface. A * symbol indicates adsorbed species.
Fig. 8. Atomic geometries of the reactants, products and transition states along the various pathways of allylic hydrogen abstraction from propylene by surface O, corresponding to energetic profiles of Fig. 7.
ily desorb from the surface. The geometries of intermediates and transition states on the way to allyl alcohol and acrolein are shown in Figs. S4 and S5 . Allyl alcohol is known to convert to acrolein on O-covered Au(111) [35]. This explains why acrolein is the main partial oxidation product observed on gold-based catalysts [35,46,47]. More likely precursors to total combustion are oxametallacycles, which lie lower in energy than allyl and have high activation barriers for further conversion to epoxide, propanal, or acetone, whereas
dehydrogenation to even more strongly bound species, such as acetonyl (see Fig. S2) is expected to be facile. Before closing this section, it is worth commenting on the role of oxygen coverage for epoxidation selectivity. Earlier studies on ethylene epoxidation over silver-based catalysts emphasized the importance of high O coverage [48] and perhaps even a surface oxide structure [49,50] to create weakly bound electrophilic forms of oxygen responsible for epoxide formation in contrast to nucleophilic O atoms adsorbed at low coverage, which are active in C–H
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
8
Fig. 9. Energetic profiles (kJ mol−1 ) for the formation of surface OOH species through a gold-mediated reaction between (a) H2 and O2 and (b) H2 O and O2 on the Au(321) surface. A * symbol indicates adsorbed species.
activation. The chloride promoters were suggested to play a role similar to subsurface O in lowering the binding energy of oxygen atoms and making them more electrophilic [48]. In the case of gold catalysts, however, high coverage of oxygen is not achievable because of low dissociation probability and low binding energy of O2 . Also, such structures would be thermodynamically unstable on gold at experimentally relevant conditions [19]. At the same time, the nucleophilic properties of surface oxygen increase in the sequence Cu < Ag < Au [36] explaining why the C–H activation by surface O has the lowest activation energy on gold. This is in line with the observation that silver catalysts alloyed with gold selectively produce acrolein [47]. 3.2. Reaction of propylene with chemisorbed hydroperoxo species (OOH) 3.2.1. Formation of hydroperoxo species on gold Oxygen alone cannot selectively oxidize propylene to PO on gold for the reasons discussed in the preceding sections. However, if hydrogen or water is added to O2 as co-reactants, high PO selectivities can be achieved and the oxidation reaction proceeds already at low temperature (30–120 ◦ C) [4,9] although for better conversions the temperature should be increased to 160–200 ◦ C [5,6,11]. It has been suggested that hydroperoxo species OOH may be responsible for selective epoxide formation in this case [8,12]. Surface OOH can be formed according to the following reactions: H2 + ∗ → 2H∗
(3a)
H ∗ + O2 ∗ → OOH∗
(3b)
H2 ∗ + O2 ∗ → OOH ∗ + H∗
(4)
H2 O ∗ + O2 ∗ → OOH ∗ + OH∗
(5)
Fig. 9 shows the calculated energetic profiles for OOH* formation from the surface-mediated reaction between O2 and H2 or O2 and H2 O on our model Au(321) surface. Significantly, in both cases the activation energy of the rate-limiting steps is much lower than for O2 dissociation. It should be noted, however, that the actual activation energies of the rate-limiting steps are probably even somewhat
lower than the values calculated on Au(321) due to an assisting effect of the oxide support. In fact, Green at al. [51] recently have shown on the basis of DFT calculations that O2 -assisted H2 dissociation at the Au/TiO2 perimeter sites (whereby a Ti-OOH intermediate is formed) involves an activation barrier of 15 kJ mol−1 whereas on pure gold we calculate the activation energies of ca. 20 kJ mol−1 . Our value is very close to that calculated for the same reaction by Boronat et al. on a Ag13 cluster [52], and our optimized TS geometry is also similar to that of ref. [52]. For the formation of OOH* through the reaction between co-adsorbed water and O2 we calculate an activation energy of 29 kJ mol−1 , in close agreement with Chang et al. [20], who calculated using the Au38 cluster model an activation energy of 26 kJ mol−1 for this process. Because the activation energy required for OOH* formation is low (and significantly lower than that needed for O2 dissociation), the reaction conditions can be tuned in such a way that the direct O2 dissociation does not take place or is very slow, whereas the main pathway of O2 activation is via the formation and further reactions of the hydroperoxide. 3.2.2. Reaction of C3 H6 with surface OOH Although in the mechanism proposed by Haruta et al. OOH* reacts with propylene on the TiO2 support [8,12], Chang et al. [20] have recently reported DFT calculations on a model Au38 nanoparticle, which demonstrated that also gold alone can catalyze propylene oxidation. Our results presented in the following concur with the findings of Chang et al. [20] but we have identified another, even lower lying path for epoxide formation. We have calculated several possible pathways for propylene reaction with adsorbed OOH, Fig. 10. All of them were chosen to take place at the step edge to make a direct comparison to the surface O-induced pathways possible and also because we expect these steps to be among the most energetically favorable in analogy to the energetic preference of C3 H6 * + O* pathways. Path 1 (not considered previously) proceeds with analogy to the lowest OM2 C pathway for the reaction between propylene and surface O whose energy profile is shown in Fig. 2 (B) in magenta color. In the first step (TS1) the hydroperoxyl attaches to the middle C atom of propylene forming an analog of an oxametallacycle OM2 C but with an
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
9
Fig. 10. Energetic profiles (kJ mol−1 ) for the reaction of propylene and hydroperoxyl and the subsequent reaction steps on the Au(321) surface. A * symbol indicates adsorbed species.
OH group attached to the O atom of the cycle, see Fig. 11 (Path 1). An important advantage for selective PO formation is that this “hydroperoxametallacycle” HO-OM2 C cannot so easily isomerize to acetone. In the second step (TS2) the OH group splits off leaving a metallalcycle OM1 C , which in the third step converts to epoxide as discussed in Section 3.1.2. We expect that there should be a similar pathway going over OM2 C but we have not considered it here since the energy of the first and highest transition state (TS1) is expected to be higher for HO-OM2 C than for HO-OM2 C in analogy to surface O-derived pathways as summarized in Fig. 3. There are two other, more direct pathways to epoxide, denoted as Path 2 and Path 3 in Fig. 10. Both of them start with the propylene molecule adsorbed at an 8-fold coordinated Au atom, Fig. 11. Therefore, the initial state (co-adsorbed propylene and OOH) has a higher energy in Path 2 and Path 3, Fig. 10. In Path 2 the short oxametallacycle OM1 C is formed directly, with the activation barrier of 28 kJ mol−1 . The corresponding transition state, TS3, lies 44 kJ mol−1 higher than TS1. Alternatively, epoxide can be directly formed along Path 3 and TS4. Transition states TS3 and TS4 have about the same height; therefore, we cannot judge on the basis of energies alone which one would be preferred. Pathways similar to Path 2 and Path 3 have also been identified on the Au38 nanoparticle, with the activation energies of 80 and 95 kJ mol−1 for the analogs of TS3 and TS4, respectively, relative to co-adsorbed propylene and OOH [20]. These values would agree well with ours if we took the lowest co-adsorbed state for propylene and OOH (from Path 1) as reference (in ref. [20] propylene is adsorbed at a low-coordinated apex site in the initial state, similar to a kink site of Au(321)). In any case, Path 1 seems to be the lowest-energy pathway identified so far and it lies lower in energy than the respective lowest-energy path on the C3 H6 * + O* potential energy surface. Remarkably, all three identified pathways for oxidation with OOH* bypass the formation of an intermediate that can easily isomerize to acetone. The short oxametallacycle OM1 C can be converted to acetone but the activation barrier in this case is much higher than that for epoxide formation. The TS structure for Htransfer to form acetone is shown in TS6 of Fig. 12. The TS structure has an even more early character than the analogous TS from the
large-ring OM2 C , consistent with the stronger exothermicity of this transformation from OM1 C . The activation barriers for transformation to acetone are almost of the same height from OM1 C and OM2 C but because the epoxide formation is so facile from OM1 C , the conversion to acetone can no longer be competing with it at low temperature. Also the H abstraction from the methyl group of propylene by OOH* has higher activation energy than that of H abstraction by surface O. The transition state for this reaction, TS5, is calculated to lie slightly above TS1 as can be seen in Fig. 10. The activation energy for H abstraction with OOH* is as high as 54 kJ mol−1 . It should be kept in mind, however, that some of the OOH* species formed can dissociate to OH* + O* before reacting with propylene, though with a relatively high activation barrier of ∼65 kJ mol−1 calculated with our surface model as can be seen in Fig. 9. Further, two surface OH species can recombine forming H2 O and surface O, with a low activation barrier of only 10 kJ mol−1 , Fig. 9. Therefore, even in the presence of hydrogen or water it will probably not be possible to completely replace surface O species by OOH, but it is important to be able to reduce the concentration and reactivity of the former.
4. Conclusions We have investigated the complex reaction network for propylene epoxidation with surface O and OOH as oxidants, including important side reactions. We have shown that in the reaction of propylene with chemisorbed atomic oxygen not only the formation of allylic species but also isomerization reactions of metallacycles to carboxylic compounds are responsible for low epoxide selectivity. Furthermore, five-member ring oxametallacycles are energetically deeply lying intermediates on the global potential energy surface with relatively high activation barriers for conversion to products that can desorb from the surface, including epoxide and carboxylic compounds. Therefore, OM2 C oxametallacycles, which can further easily dehydrogenate and undergo further O addition, decarbonylation and decarboxylation steps, are likely intermediates on the way to total combustion. In contrast, allylic species are shown to
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 10
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 12. Atomic geometries of the key minima and transition states leading to allyl (left) and acetone (right) formation corresponding to energetic profiles of Fig. 10.
tion) but also proceeds along a reaction path of an overall lower energy with respect to the gas-phase reagents. Acknowledgements We acknowledge the financial support from the German Research Foundation (DFG) within the Project No. MO 1863/2-2 and MO 1863/4-1. The calculations were performed at the HPC Cluster HERO, located at the University of Oldenburg and funded by the DFG through its Major Research Instrumentation Program (INST 184/108-1 FUGG) and the Ministry of Science and Culture of the Lower Saxony Federal State, and in part with resources provided by the North-German Supercomputing Alliance (HLRN), Project No. hbc00018. Appendix A. Supplementary data Fig. 11. Atomic geometries of the key minima and transition states leading to epoxide formation along the various pathways of propylene reaction with hydroperoxyl, corresponding to energetic profiles of Fig. 10.
have low barriers for conversion to acrolein or allyl alcohol. The former is found as major partial oxidation product on gold catalysts in the absence of hydrogen or water. We have shown that the addition of hydrogen or water to the reacting gas mixture readily produces surface hydroperoxo species with activation energies much lower than that required for O2 dissociation. Therefore, the propylene partial oxidation reaction can proceed at milder oxidation conditions, avoiding deep oxidation. The present study clearly shows that the oxidation with hydroperoxyl is not only more selective than the reaction with surface atomic oxygen (due to the suppression of propanal and acetone formation and due to a higher activation energy of allylic hydrogen abstrac-
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.cattod.2016.05. 034. References [1] D.L. Trent, Propylene Oxide, Kirk-Othmer Encyclopedia of Chemical Technology, John Wiley & Sons Inc., 2000. [2] J.W. Medlin, J.R. Monnier, M.A. Barteau, J. Catal. 204 (2001) 71–76. [3] J. Huang, M. Haruta, Res. Chem. Intermed. 38 (2012) 1–24. [4] T. Hayashi, K. Tanaka, M. Haruta, J. Catal. 178 (1998) 566–575. [5] A.K. Sinha, S. Seelan, S. Tsubota, M. Haruta, Angew. Chem. Int. Ed. 43 (2004) 1546–1548. [6] B. Taylor, J. Lauterbach, W.N. Delgass, Appl. Catal. A 291 (2005) 188–198. [7] L. Cumaranatunge, W.N. Delgass, J. Catal. 232 (2005) 38–42. [8] J. Huang, T. Akita, J. Faye, T. Fujitani, T. Takei, M. Haruta, Angew. Chem. Int. Ed. 48 (2009) 7862–7866. [9] M. Ojeda, E. Iglesia, Chem. Commun. (2009) 352–354.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
[10] S. Lee, L.M. Molina, M.J. López, J.A. Alonso, B. Hammer, B. Lee, S. Seifert, R.E. Winans, J.W. Elam, M.J. Pellin, S. Vajda, Angew. Chem. Int. Ed. 48 (2009) 1467–1471. [11] J. Huang, T. Takei, T. Akita, H. Ohashi, M. Haruta, Appl. Catal. B 95 (2010) 430–438. [12] J.J. Bravo-Suárez, K.K. Bando, J. Lu, M. Haruta, T. Fujitani, T. Oyama, J. Phys. Chem. C 112 (2008) 1115–1123. [13] J.-H. Franke, D.S. Kosov, J. Chem. Phys. 139 (2013) 224709. [14] J.L.C. Fajín, J.R.B. Gomes, M.N.D.S. Cordeiro, Langmuir 29 (2013) 8856–8864. [15] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, J.Phys. Chem. C 112 (2008) 17291–17302. [16] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, J. Catal. 289 (2012) 11–20. [17] L.V. Moskaleva, S. Rohe, A. Wittstock, V. Zielasek, T. Kluner, K.M. Neyman, M. Baumer, Phys. Chem. Chem. Phys. 13 (2011) 4529–4539. [18] L.V. Moskaleva, V. Zielasek, T. Klüner, K.M. Neyman, M. Bäumer, Chem. Phys. Lett. 525–526 (2012) 87–91. [19] L.V. Moskaleva, T. Weiss, T. Klüner, M. Bäumer, J. Phys. Chem. C 119 (2015) 9215–9226. [20] C.-R. Chang, Y.-G. Wang, J. Li, Nano Res. 4 (2011) 131–142. [21] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865–3868. [22] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 78 (1997), 1396–1396. [23] G. Kresse, J. Furthmüller, Comput. Mater. Sci. 6 (1996) 15–50. [24] G. Kresse, J. Hafner, Phys. Rev. B 49 (1994) 14251–14269. [25] P.E. Blöchl, Phys. Rev. B 50 (1994) 17953–17979. [26] G. Kresse, D. Joubert, Phys. Rev. B 59 (1999) 1758–1775. ˜ [27] P. Janthon, S.M. Kozlov, F. Vines, J. Limtrakul, F. Illas, J. Chem. Theory Comput. 9 (2013) 1631–1640. ˜ [28] P. Janthon, S. Luo, S.M. Kozlov, F. Vines, J. Limtrakul, D.G. Truhlar, F. Illas, J. Chem. Theory Comput. 10 (2014) 3832–3839. [29] H.J. Monkhorst, J.D. Pack, Phys. Rev. B 13 (1976) 5188–5192. [30] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, F. Illas, J. Chem. Theory Comput. 8 (2012) 1737–1743.
11
[31] G. Henkelman, H. Jónsson, J. Chem. Phys. 111 (1999) 7010–7022. [32] S. Linic, M.A. Barteau, J. Am. Chem. Soc. 124 (2002) 310–317. [33] D. Torres, N. Lopez, F. Illas, R.M. Lambert, Angew. Chem. Int. Ed. 46 (2007) 2055–2058. [34] K.A. Davis, D.W. Goodman, J. Phys. Chem. B 104 (2000) 8557–8562. [35] X. Deng, B.K. Min, X. Liu, C.M. Friend, J. Phys. Chem. B 110 (2006) 15982–15987. [36] A. Roldan, D. Torres, J.M. Ricart, F. Illas, J. Mol. Catal. A: Chem. 306 (2009) 6–10. [37] T.A. Baker, B. Xu, S.C. Jensen, C.M. Friend, E. Kaxiras, Catal. Sci. Tech. 1 (2011) 1166–1174. [38] A. Pulido, P. Concepción, M. Boronat, A. Corma, J. Catal. 292 (2012) 138–147. [39] F.A. Carey, R.J. Sundberg, Advanced Organic Chemistry, 4th ed., Kluwer, New York, 2002. [40] J.W. Medlin, M.A. Barteau, J. Phys. Chem. B 105 (2001) 10054–10061. [41] M. Enever, S. Linic, K. Uffalussy, J.M. Vohs, M.A. Barteau, J. Phys. Chem. B 109 (2005) 2227–2233. [42] H. Piao, M.C.N. Enever, K. Adib, J. Hrbek, M.A. Barteau, Surf. Sci. 571 (2004) 139–145. [43] A. Klust, R.J. Madix, Surf. Sci. 600 (2006) 5025–5040. [44] L. Zhou, R.J. Madix, J. Phys. Chem. C 112 (2008) 4725–4734. [45] G.S. Hammond, J. Am. Chem. Soc. 77 (1955) 334–338. [46] N.W. Cant, W.K. Hall, J. Phys. Chem. 75 (1971) 2914–2921. [47] P.V. Geenen, H.J. Boss, G.T. Pott, J. Catal. 77 (1982) 499–510. [48] R.A. van Santen, Theoretical Heterogeneous Catalysis, World Scientific Singapore, 1991. [49] M.-L. Bocquet, A. Michaelides, D. Loffreda, P. Sautet, A. Alavi, D.A. King, J. Am. Chem. Soc. 125 (2003) 5620–5621. [50] M.O. Özbek, I. Önal, R.A. van Santen, ChemCatChem 3 (2011) 150–153. [51] I.X. Green, W. Tang, M. Neurock, J.T. Yates, Angew. Chem. Int. Ed. 50 (2011) 10186–10189. [52] M. Boronat, A. Pulido, P. Concepcion, A. Corma, Phys. Chem. Chem. Phys. 16 (2014) 26600–26612.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
ARTICLE IN PRESS
CATTOD-10226; No. of Pages 11
Catalysis Today xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Catalysis Today journal homepage: www.elsevier.com/locate/cattod
Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents Lyudmila V. Moskaleva Institute of Applied and Physical Chemistry and Center for Environmental Research and Sustainable Technology, Universität Bremen, 28359 Bremen, Germany
a r t i c l e
i n f o
Article history: Received 30 December 2015 Received in revised form 25 April 2016 Accepted 16 May 2016 Available online xxx Keywords: Density functional theory Au(321) surface Propylene epoxidation Propylene oxide Selectivity Hydroperoxyl
a b s t r a c t Propylene oxide (PO) is an important bulk chemical used for synthesis of many value-added products. Much effort has been devoted to finding a simple and “green” process for PO production. Promising results in terms of activity and selectivity have been achieved for propylene oxidation on supported gold catalysts with a mixture of O2 and H2 or with O2 and H2 O. In this work a detailed transformation network of competitive reaction pathways following the initial steps of oxidation has been studied theoretically using density functional theory (DFT). The results of calculations question some of the earlier assumptions regarding the mechanism of PO formation. Surface hydroperoxo species (OOH) formed in situ are shown to be responsible for the high selectivity of propylene epoxidation with O2 on gold-based catalysts using hydrogen or water as co-reactants. © 2016 Elsevier B.V. All rights reserved.
1. Introduction Epoxidation of propylene is crucial in the manufacture of propylene oxide (PO), a valuable raw material for producing many value-added chemicals, such as polyether polyols, propylene glycol, and others [1]. At present, two major methods are used for the commercial PO production: the chlorohydrin process and the hydroperoxide processes. These processes proceed via multiple reaction steps and utilize hazardous chlorine and costly hydroperoxides as oxidants [1]. Therefore, efforts have been made to develop a “green” process that could directly and selectively epoxidize propylene to PO. Although silver supported on alumina is well known as a uniquely effective catalyst for ethylene epoxidation, Ag catalysts are far less effective for epoxidation of propylene (and other olefins with allylic C–H bonds) because of predominant total oxidation to CO2 and H2 O [2]. The low selectivity obtained for propylene epoxidation has been attributed to a weaker binding strength of allylic (–CH3 ) hydrogen in propylene than that of the vinyl hydrogen in ethylene [3]. Two other coinage metals, Au and Cu on appropriate supports, have also been tested as catalysts for propylene epoxidation [3]. As in the case of silver, the common challenge identified for all
E-mail address: [email protected]
of them is the need to suppress undesirable reaction pathways leading either to total oxidation or to the formation of other partial oxidation products, e.g., acrolein, acetone, propanal. Hayashi et al. [4], who first reported the selective epoxidation of propylene on supported 2–5 nm Au nanoparticles, proposed to use a sacrificial reductive agent (H2 ) to better control the reactivity of active oxygen species and suppress deep oxidation. This approach showed successful results in terms of the PO selectivity (∼90%), although conversions were low (only 1–2%). Since then, several groups have been working on the optimization of propylene epoxidation over gold-based catalysts and significant progress has been achieved [5–11]. Recently, it was shown that, similarly to H2 , water admixtures can also drastically increase the selectivity to PO [9]. It is therefore reasonable to expect that in both cases a surface hydroperoxide intermediate, OOH, formed in situ could play a role as a milder oxidant than atomic oxygen and could be responsible for the selective PO formation. Although a tentative mechanism of OOH formation and oxidation has been proposed by Haruta et al. [8,12], it has not yet been confirmed by unbiased computational models based on first principles calculations. Herein, we present a comprehensive theoretical investigation of the reaction network involved in the oxidation of propylene on gold on the basis of DFT. We have chosen a stepped and kinked Au(321) surface as a convenient model to study the reactions taking place on the defect-rich surface of a nanoparticle gold catalyst.
http://dx.doi.org/10.1016/j.cattod.2016.05.034 0920-5861/© 2016 Elsevier B.V. All rights reserved.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
2
This model has been previously successfully employed in a number of theoretical studies [13–19]. Although the oxide support is an indispensable component of a catalyst, in this work we focus on the metal component, assuming that the key reaction steps that determine the selectivity take place on the gold surface. This study is the first to compare and contrast the chemisorbed O and OOH species as oxidizing agents for propylene oxidation on gold. We show why the oxidation with hydroperoxo species leads to a higher PO selectivity. An important aspect that has not received enough attention in earlier theoretical studies is the easy isomerization of the commonly considered PO precursor (a five-member ring oxametallacycle, OM2 C) to propanal or acetone, which according to our calculations, always has a lower activation barrier than that required for epoxide formation. Thus, even if the H abstraction from the CH3 group of propylene could be suppressed, that still would not solve the problem of low PO selectivity. Our study shows that, quite remarkably, the propylene oxidation with surface OOH allows for at least three low-energy reaction pathways not going through that five-member-ring OM2 C intermediate, hence, suppressing the isomerization to carbonyl compounds. At the same time, the abstraction of the allylic H from propylene by adsorbed OOH has higher activation energy than in the case of H abstraction by surface O. That explains why high selectivities to PO can be achieved if chemisorbed atomic O as oxidant is replaced by a milder OOH counterpart. Furthermore, our study clearly shows that the key reaction pathways are structure sensitive. The reaction pathways taking place at low-coordinated edge Au atoms go through somewhat different intermediates than on flat terrace sites. In particular, four-member ring oxametallacycles bound on a single kink Au atom seem to be crucial precursors to epoxide formation. The paper is structured as follows. Section 2 describes the models and computational approach. Section 3.1 discusses the reaction pathways of propylene with chemisorbed O. Section 3.2 considers the oxidation with chemisorbed hydroperoxy radical and highlights an important difference to the atomic-oxygen induced pathways. Section 4 summarizes the most important findings.
The calculations have been carried out with the VASP code [23,24] using PBE functional and projector augmented wave (PAW) method [25,26] with an energy cutoff of 415 eV. The choice of the PBE functional is justified by its very good performance in describing bulk properties of transition metals [27,28]. The k point sampling was generated by the Monkhorst-Pack procedure [29]; a 5 × 5 × 1 k point mesh was used for geometry optimizations. The structures were relaxed until the force acting on each atom was ≤ 2 × 10−2 eV Å−1 . The spin polarization has been taken into account for all reactions involving paramagnetic O2 molecule. For the remaining reactions the spin-restricted formalism was used, as justified by test studies for adsorption systems on non-magnetic metallic surfaces [30]. Transition states (TSs) of the reactions were determined by applying the dimer method [31]. To verify the nature of a TS, the minima connected through the TS were identified by following the steepest descent reaction pathway downhill from the TS to the reactant and product valleys. The adsorption energy of a stable molecule Eads (X) was calculated as follows: E ads (X) = (E X∗ − E substrate − E X )
where EX* denotes the total energy of the adsorbed X* on the surface in the optimized geometry; Esubstrate and EX are the total energies of the clean surface (or a surface with pre-adsorbed co-adsorbate in the case of co-adsorption) and of the X molecule in the gas phase, respectively. With this definition, negative values of adsorption energy indicate exothermic adsorption. For a chemisorbed O atom, the adsorption energy is calculated by taking one-half the energy of a gas-phase O2 molecule plus the energy of clean Au(321) as reference. Importantly, the relative energies of various species and transition states shown in the calculated energy profiles are computed using the clean Au(321), gas-phase propylene, gas-phase O2 , and, where relevant, gas-phase H2 as reference. For example, Erel (C3 H∗6 + OOH∗ ) = EC3 H6 ∗ +OOH∗ − EAu(321) − EC3 H6 − EO2 −
2. Models and computational details In this study we use the Au(321) surface, as in Fig. 1(a,b), as a model to represent the defect-rich surface of gold nanoparticles, with the typical structural features expected for these particles, such as small terraces of (111) type, steps and kinks. Of course, the real catalyst has a more complex structure, where the particle size, shape, and the perimeter interface of metal particles and the oxide support definitely play a role in the activity. However, we concentrate on the reaction steps that determine the selectivity and which are expected to take place on the gold surface. Hence, the choice of this relatively simple but computationally affordable model is justified. A comparison with the work of Chang et al. [20], who calculated some of the reactions discussed below using a Au38 nanocluster model, shows that we predict the same chemistry and our results agree qualitatively, only the adsorption energies are much stronger on a nanoparticle surface, which is expected because of a small particle size. The slab model has been constructed using the bulk lattice parameter of Au 4.173 Å obtained from a calculation with the PBE form of the generalized-gradient approximation (GGA) for the exchange-correlation functional [21,22]. The slab model contains 14 atomic layers corresponding to the thickness of the resulting slab of ∼7.2 Å and the vacuum spacing between periodically repeated slabs of ∼8.5 Å. We use a (2 × 1) surface unit cell containing 28 metal atoms. The uppermost 14 metal atoms were allowed to fully relax, whereas the lower 14 atoms were kept fixed at their optimized bulk positions.
(1)
1 EH . 2 2 (2)
where EC3 H6 ∗ +OOH∗ is the total energy of co-adsorbed propylene and hydroperoxyl, EAu(321) is the total energy of the clean Au(321) slab EC3 H6 , EO2 and E H2 are the total energies of gas-phase propylene, O2 and H2 , respectively, and here and further on a * symbol denotes an adsorbed species. A single exception is Fig. 9(b), where gas-phase O2 and H2 O are used as reference species. Relative energies do not include zero-point energy corrections. 3. Results and discussion 3.1. Reaction of propylene with chemisorbed O atoms The epoxidation of propylene on gold and other coinage metals is believed to proceed by analogy to the accepted mechanism of ethylene epoxidation, via an oxametallacycle intermediate formed from addition of surface O to the C=C bond of propylene [32]. A competing H abstraction from the CH3 group of propylene forming surface allyl species is thought to be leading to total combustion [3,33]. Studies of propylene oxidation on oxygen covered Au(111) and Au(100) single-crystal surfaces in ultrahigh vacuum reported either CO2 and H2 O [34] or acrolein [35] as main products but no propylene oxide was detected. The differences in observed main products in Refs. [34] and [35] were attributed to the different methods used for O deposition. Earlier theoretical studies of propylene oxidation by surface O on gold [36,37] focused on the primary reaction steps leading either to metallacycles or allyl but refrained
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
3
Fig. 2. Reaction network of propylene partial oxidation with adsorbed O as oxidizing agent. The starting point is indicated with a blue frame. Reaction steps addressed in earlier studies on extended gold surfaces are marked with green arrows. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Fig. 1. (a) Side and (b) top view of the Au(321) surface. The (2 × 1) surface unit cell is indicated. (c) and (d): Most favorable adsorption sites of propylene. (e)–(h): Most favorable adsorption sites of oxygen.
from exploring the complete reaction network. Herein, we investigated the more extended reaction network, Fig. 2. Particularly important are isomerization pathways of oxametalacycles to carboxylic compounds. As demonstrated earlier in a DFT study of propylene epoxidation over Ag(111) and Ag(100) surfaces combined with Raman spectroscopy and mass spectrometry [38] and shown below, these reactions are in direct competition with the epoxide formation. The reaction scheme in Fig. 2 refers to the case of oxidation with pure O2 (which adsorbs dissociatively) or with preadsorbed atomic O in the case of model studies on single-crystal surfaces, where no H2 or H2 O are added to the reactants stream. The dissociative chemisorption of O2 , see Section 3.2, is probably the rate-limiting step, since all further steps lie energetically lower than the gas-phase propylene and O2 . In Section 3.2 we will explain why the addition of H2 or H2 O to O2 not only facilitates the O2 activation but also results in new possibilities for selective epoxide formation.
and donor orbitals of the metal [39]. The adsorption energies of propylene are at −12 kJ mol−1 on top of the 8-fold coordinated Au atom and −55 kJ mol−1 on top of the 6-fold coordinated Au atom, which is much stronger than the values reported for the adsorption on the Au(111) surface (ca. −5 kJ mol−1 ) [36,37] but much weaker than the adsorption energy on the apex site of a Au38 nanoparticle (-100 kJ mol−1 ) [20] or on top of a single Au adatom on the Au(111) surface (–75 kJ mol−1 ) [37]. Propylene and oxygen atoms, when adsorbed close to each other (but not binding to the same Au atom), co-adsorb favorably with a slight adsorption energy gain of 4–13 kJ mol−1 with respect to the individually adsorbed species. Clearly, low-coordinated atoms at step edges of a gold surface bind the reactants and intermediates stronger than the highercoordinated terrace sites, but not too strong so as to poison the catalyst. Therefore, in the following we mainly discuss the reaction pathways at step edges. Elementary steps leading to allyl and oxametallacycle intermediates have been previously studied on the flat Au(111) [36,37] and on Au(111) with vacancies and adatoms, as well as on straight monoatomic steps of the Au(211) surface [37]. Below we comment on how the current results compare to those from earlier studies.
3.1.1. C3 H6 and O adsorption In our earlier work [17] we have demonstrated that atomic oxygen on the Au(321) surface prefers adsorption sites immediately adjacent to the step edge. These sites are either of the three-fold fcc or the two-fold (“bridge”) type, Fig. 1(e–h). The adsorption energies relative to the gas-phase O2 range from −12 to +4 kJ mol−1 . Similarly, propylene binds preferentially above low-coordinated metal atoms along the step edge in a so-called “-mode” through an interaction of the alkene and * orbitals with the acceptor
3.1.2. Epoxide formation via oxametallacycle intermediates An oxametallacycle is an intermediate attached to a metal surface in a M–O–C–C–M type linkage. In the context of widely studied ethylene epoxidation on silver, two types of oxametallacycles have been considered in the literature: a five-member ring containing two metal atoms (OMME) and a four-member ring containing a single metal center (OME). Earlier DFT calculations on Ag(111) [32] and Ag(110) [40] surface models in combination with the high resolution electron energy loss spectroscopy (HREELS) [32] concluded
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 4
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 3. Energetic profiles (kJ mol−1 ) for the reaction of chemisorbed propylene and O via primary (a) and secondary (b) oxametalacycle intermediates on the Au(321) surface. A * symbol indicates adsorbed species. Only the lowest-energy pathways identified are shown. See also Figs. S1 and S2 for alternative pathways.
that OME, although slightly less thermodynamically favorable than OMME, should be a more likely precursor of ethylene oxide formation on silver. The analogous four-member oxametallacycle structure predicted by Barteau et al. [41,42] as an intermediate for styrene epoxidation on Ag(111) surface was later supported experimentally by Madix group [43,44]. To differentiate between the two types of metallacycles, in this work we refer to a metallacycle with a five-member ring as OM2 C, while the four-member ring is denoted as OM1 C. Examples of both structures are shown in Figs. 3 and 4. Previous DFT studies on the Au(111) surface [36,37] considered only the formation of OM2 C-type intermediates and their further
ring contraction to epoxide. According to our calculations using the stepped Au(321) surface, the OM1 C intermediates lie significantly higher in energy than OM2 C (by 36–70 kJ mol−1 ). Therefore, in the first reaction step the formation of a five-member ring is more favorable. However, our calculations suggest that the ring contraction of OM1 C to epoxide has a lower-lying transition state (see alternative pathways in Figs. S1, S2); therefore, OM1 C is identified in this work to be a key intermediate on the way to epoxide formation, in agreement with the conclusions reached by Barteau et al. [32,40] regarding the mechanism of ethylene epoxidation on Ag surfaces.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 4. Atomic geometries of the key minima and transition states along the lowest energy pathways of epoxide formation via primary (left) and secondary (right) oxametallacycle intermediates, corresponding to energetic profiles of Fig. 3.
5
We explored the addition of chemisorbed oxygen to either C1 or C2 atom of propylene forming a primary (OM2 C ) or a secondary (OM2 C ) oxametallacycle intermediate, respectively. The relative energies of the intermediates and transition states for the lowestenergy pathways of epoxide formation identified on Au(321) are shown in Fig. 3. Basically, the pathways passing through primary and secondary metallacycles are qualitatively very similar, with not so high energy differences, except for the initial addition step, which renders the OM2 C formation kinetically more favorable. In agreement with the earlier work of Baker et al. [37], we found lower activation energies (by 5–19 kJ mol−1 ) associated with the formation of OM2 C compared to OM2 C ; however, the formation of OM2 C is slightly more exothermic, by 3–15 kJ mol−1 . The most thermodynamically favorable OM2 C structures identified bind directly at the step edge, with the C atom attached at a kink site and with the O atom attached to another kink and an 8-fold coordinated Au between the kinks in a bridge mode. Alternatively, C atom can attach at an 8-fold coordinated Au atom and O at a 6-fold coordinated kink Au atom. In this case, the energy of the resulting OM2 C is slightly higher, see Figs. S1 and S2 . The transition states on the way to OM2 C and OM2 C (TS1) have very similar structures, Fig. 4. The surface oxygen atom interacts with one of the carbon atoms of the double bond in the plane perpendicular to the molecular plane of propylene. The C–O distance of the forming bond is 2.17 Å and the C–C distance of the double bond is increased to 1.4 Å. A lower barrier for OM2 C formation might result from an electron donating effect of a methyl group attached to the reactive C atom of the alkene. Not all OM2 C intermediates can be converted to epoxide directly. All OM2 C structures shown in Fig. 3 except one, the OM2 C along the magenta path, first need to pass through a four-member metallacycle OM1 C, where both C and O are attached to the same low-coordinated kink Au atom. The direct pathways, when existing, generally have a higher-lying transition state, see also Figs. S1, S2. The energetically lowest transition state structures for epoxide formation have been found for step-wise pathways going via OM1 C intermediates, whose structures due to a smaller ring size are geometrically closer to the final epoxide product. Remarkably, in all these structures the Au atom incorporated in the oxametallacycle is the kink atom with the lowest coordination number. This Au atom has extra free bonding capacities enabling the formation of multiple bonds with the –C–C–O– backbone of an oxametallacycle, and lowering the energy of the corresponding transition state for epoxide ring closure, TS2, which is geometrically quite similar to the OM1 C intermediate, Fig. 4. All oxametallacycles can undergo a 1,2-H shift reaction forming carboxylic compounds propanal or acetone. The geometries of transition states and products corresponding to the lowest-energy pathways according to Fig. 3 are given in Fig. 5. The transition state geometries present an early character, nearer to the reactant oxametalacycle than to the product, consistent with the strongly exothermic reaction. Notably, for all OM2 C structures the C–H activation to form carboxylic compounds has a lower activation barrier than the highest transition state on the way to becoming an epoxide, see Figs. 3, S1, and S2, providing another reason why with O2 alone as an oxidant no epoxide is actually formed on Au catalysts. Interestingly, this energetic preference is reversed for OM1 C. The formation of carboxylic compounds from OM1 C has higher activation energy than the conversion to epoxide, see the discussion in Section 3.2.2. Therefore, if OM1 C was formed directly from adsorbed propylene and oxygen, its ring closure to form epoxide would have a higher rate than the isomerisation to propanal or acetone. However, we have not found a direct pathway for OM1 C formation from adsorbed O. Even if such a path exists, it probably has higher activation energy than that for OM2 C formation because the latter lies energetically much lower than OM1 C. Hence, OM2 C
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 6
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
oxidation using the Au38 cluster model [20]. However, for such a direct addition to happen, the propylene molecule must approach the O atom from above, so as to make the pz orbital of O interact with the * of the former, see the geometrical arrangement of calculated transition states in Fig. S3. Therefore, this pathway is of the Eley-Rideal type, where propylene reacts with the adsorbed O directly from the gas-phase or from a weakly physisorbed state. We have calculated the activation barriers of 37–52 kJ mol−1 with respect to chemisorbed O and physisorbed propylene for a direct propylene formation, see Fig. 6. These values are in good agreement with the earlier calculated value of 31 kJ mol−1 using the Au38 cluster model [20]. Because an Eley-Rideal mechanism results in a very low pre-exponential factor (due to an extremely short time scale of a gas-surface collision), whereas the calculated activation barrier is significant, this route to PO formation is expected to have an insignificantly low rate in comparison with the competing Langmuir-Hinshelwood-type pathways.
Fig. 5. Atomic geometries of the transition states and products for propanal (left) and acetone(right) formation via primary (left) and secondary (right) oxametallacycle intermediates, corresponding to energetic profiles of Fig. 3.
Fig. 6. Energetic profiles (kJ mol−1 ) for a direct reaction Eley-Rideal (ER) type between the gas-phase propylene and chemisorbed atomic O. A * symbol indicates adsorbed species.
will form in the first C3 H6 * + O* addition step and will preferentially react to form propanal or acetone but not epoxide. To sum up, the reaction between surface O and propylene through oxametallacycle intermediates to epoxide is energetically feasible with the first transition state being the highest in energy, 50–55 kJ mol−1 relative to co-adsorbed C3 H6 and O. However, in addition to the competing H abstraction from the methyl group by surface O (Section 3.1.4), the isomerization of OM2 C intermediates to carboxylic compounds is also energetically favored over the desired epoxide product, thus revealing another obstacle on the way of improving the PO selectivity. 3.1.3. Direct epoxidation of C3 H6 In addition to the widely accepted oxametallacycle route, we have also considered a direct formation of epoxide by a reaction of propylene with chemisorbed O, Figs. 6 , S3. This type of reaction has been proposed in an earlier theoretical study of propylene
3.1.4. Hydrogen abstraction by chemisorbed O and further reactions of allyl Consistent with the earlier study of Roldan et al. on the Au(111) surface [36], an abstraction of a hydrogen atom from the methyl group by adsorbed atomic O was found to have the lowest activation barrier of all primary pathways, 11–45 kJ mol−1 relative to co-adsorbed reactants, with the transition states lying in the most favorable cases below the gas-phase reactants by up to 50 kJ mol−1 , see Fig. 7. We explored different positions of co-adsorbed reactants on the surface and the corresponding reaction pathways. As seen from Fig. 7, the energy profiles and particularly the activation barrier heights vary considerably. The energies of the initial states (coadsorbed propylene and O) lie in most cases higher than those in Fig. 3. This is because in Fig. 3, only the pathways with the lowest overall energy have been shown (see Figs. S1 and S2 for other low-energy pathways), whereas Fig. 7 shows pathways with a larger variation of initial and product geometries. It can be generally observed that the pathways where oxygen in the initial state is bound in a bridge position at a step edge (Path 1 and Path 4, Fig. 8) have lower activation energies of 11–14 kJ mol−1 and are strongly exothermic by 53–69 kJ mol−1 , which can be explained by the fact that a 2-fold coordination is preferred by OH in the final state. In contrast, pathways where O is initially adsorbed at a 3-fold fcc site (Path 2, Path 3, and Path 5, Fig. 8) have higher activation energies, of 22–45 kJ mol−1 and are exothermic by only 10–24 kJ mol−1 . In the final state, OH binds at a 3-fold site (or at a pseudo 2-fold in the case of Path 3) which is not its most favorable binding mode, explaining relatively high energies of the final state and higher activation energies. Path 3 is somewhat special because propylene is interacting with two O atoms due to the translational symmetry of a unit cell, but the described trends still hold for this case as well. The transition state structures of the two groups are also somewhat different. For the latter group, both the O–H and C–H distances in the TS are close to 1.3 Å, i.e., the transferred H atom is in the middle between the C and O centers. This geometry is very similar to that reported in earlier studies on the Au(111) surface [36,37] and on Au(111) with defects and on Au(211) [36], where the Habstracting surface O was in all cases also placed at a 3-fold fcc site. In contrast, if O is bound at a 2-fold bridge site at a step edge, the TS structure is more reactant-like, with a longer O–H distance of ∼1.4–1.45 Å and a shorter C–H distance of ∼1.2 Å. This is in line with the Hummond postulate [45], since reactions of this group are strongly exothermic with low activation energy. Although allyl intermediate is often referred to as precursor to total combustion, that is an arguable point because according to Fig. 7 allyl can readily react to form allyl alcohol or acrolein − the products which are moderately strongly bound and can eas-
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
7
Fig. 7. Energetic profiles (kJ mol−1 ) for allylic H abstraction by the surface O and the subsequent reaction steps on the Au(321) surface. A * symbol indicates adsorbed species.
Fig. 8. Atomic geometries of the reactants, products and transition states along the various pathways of allylic hydrogen abstraction from propylene by surface O, corresponding to energetic profiles of Fig. 7.
ily desorb from the surface. The geometries of intermediates and transition states on the way to allyl alcohol and acrolein are shown in Figs. S4 and S5 . Allyl alcohol is known to convert to acrolein on O-covered Au(111) [35]. This explains why acrolein is the main partial oxidation product observed on gold-based catalysts [35,46,47]. More likely precursors to total combustion are oxametallacycles, which lie lower in energy than allyl and have high activation barriers for further conversion to epoxide, propanal, or acetone, whereas
dehydrogenation to even more strongly bound species, such as acetonyl (see Fig. S2) is expected to be facile. Before closing this section, it is worth commenting on the role of oxygen coverage for epoxidation selectivity. Earlier studies on ethylene epoxidation over silver-based catalysts emphasized the importance of high O coverage [48] and perhaps even a surface oxide structure [49,50] to create weakly bound electrophilic forms of oxygen responsible for epoxide formation in contrast to nucleophilic O atoms adsorbed at low coverage, which are active in C–H
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
8
Fig. 9. Energetic profiles (kJ mol−1 ) for the formation of surface OOH species through a gold-mediated reaction between (a) H2 and O2 and (b) H2 O and O2 on the Au(321) surface. A * symbol indicates adsorbed species.
activation. The chloride promoters were suggested to play a role similar to subsurface O in lowering the binding energy of oxygen atoms and making them more electrophilic [48]. In the case of gold catalysts, however, high coverage of oxygen is not achievable because of low dissociation probability and low binding energy of O2 . Also, such structures would be thermodynamically unstable on gold at experimentally relevant conditions [19]. At the same time, the nucleophilic properties of surface oxygen increase in the sequence Cu < Ag < Au [36] explaining why the C–H activation by surface O has the lowest activation energy on gold. This is in line with the observation that silver catalysts alloyed with gold selectively produce acrolein [47]. 3.2. Reaction of propylene with chemisorbed hydroperoxo species (OOH) 3.2.1. Formation of hydroperoxo species on gold Oxygen alone cannot selectively oxidize propylene to PO on gold for the reasons discussed in the preceding sections. However, if hydrogen or water is added to O2 as co-reactants, high PO selectivities can be achieved and the oxidation reaction proceeds already at low temperature (30–120 ◦ C) [4,9] although for better conversions the temperature should be increased to 160–200 ◦ C [5,6,11]. It has been suggested that hydroperoxo species OOH may be responsible for selective epoxide formation in this case [8,12]. Surface OOH can be formed according to the following reactions: H2 + ∗ → 2H∗
(3a)
H ∗ + O2 ∗ → OOH∗
(3b)
H2 ∗ + O2 ∗ → OOH ∗ + H∗
(4)
H2 O ∗ + O2 ∗ → OOH ∗ + OH∗
(5)
Fig. 9 shows the calculated energetic profiles for OOH* formation from the surface-mediated reaction between O2 and H2 or O2 and H2 O on our model Au(321) surface. Significantly, in both cases the activation energy of the rate-limiting steps is much lower than for O2 dissociation. It should be noted, however, that the actual activation energies of the rate-limiting steps are probably even somewhat
lower than the values calculated on Au(321) due to an assisting effect of the oxide support. In fact, Green at al. [51] recently have shown on the basis of DFT calculations that O2 -assisted H2 dissociation at the Au/TiO2 perimeter sites (whereby a Ti-OOH intermediate is formed) involves an activation barrier of 15 kJ mol−1 whereas on pure gold we calculate the activation energies of ca. 20 kJ mol−1 . Our value is very close to that calculated for the same reaction by Boronat et al. on a Ag13 cluster [52], and our optimized TS geometry is also similar to that of ref. [52]. For the formation of OOH* through the reaction between co-adsorbed water and O2 we calculate an activation energy of 29 kJ mol−1 , in close agreement with Chang et al. [20], who calculated using the Au38 cluster model an activation energy of 26 kJ mol−1 for this process. Because the activation energy required for OOH* formation is low (and significantly lower than that needed for O2 dissociation), the reaction conditions can be tuned in such a way that the direct O2 dissociation does not take place or is very slow, whereas the main pathway of O2 activation is via the formation and further reactions of the hydroperoxide. 3.2.2. Reaction of C3 H6 with surface OOH Although in the mechanism proposed by Haruta et al. OOH* reacts with propylene on the TiO2 support [8,12], Chang et al. [20] have recently reported DFT calculations on a model Au38 nanoparticle, which demonstrated that also gold alone can catalyze propylene oxidation. Our results presented in the following concur with the findings of Chang et al. [20] but we have identified another, even lower lying path for epoxide formation. We have calculated several possible pathways for propylene reaction with adsorbed OOH, Fig. 10. All of them were chosen to take place at the step edge to make a direct comparison to the surface O-induced pathways possible and also because we expect these steps to be among the most energetically favorable in analogy to the energetic preference of C3 H6 * + O* pathways. Path 1 (not considered previously) proceeds with analogy to the lowest OM2 C pathway for the reaction between propylene and surface O whose energy profile is shown in Fig. 2 (B) in magenta color. In the first step (TS1) the hydroperoxyl attaches to the middle C atom of propylene forming an analog of an oxametallacycle OM2 C but with an
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
9
Fig. 10. Energetic profiles (kJ mol−1 ) for the reaction of propylene and hydroperoxyl and the subsequent reaction steps on the Au(321) surface. A * symbol indicates adsorbed species.
OH group attached to the O atom of the cycle, see Fig. 11 (Path 1). An important advantage for selective PO formation is that this “hydroperoxametallacycle” HO-OM2 C cannot so easily isomerize to acetone. In the second step (TS2) the OH group splits off leaving a metallalcycle OM1 C , which in the third step converts to epoxide as discussed in Section 3.1.2. We expect that there should be a similar pathway going over OM2 C but we have not considered it here since the energy of the first and highest transition state (TS1) is expected to be higher for HO-OM2 C than for HO-OM2 C in analogy to surface O-derived pathways as summarized in Fig. 3. There are two other, more direct pathways to epoxide, denoted as Path 2 and Path 3 in Fig. 10. Both of them start with the propylene molecule adsorbed at an 8-fold coordinated Au atom, Fig. 11. Therefore, the initial state (co-adsorbed propylene and OOH) has a higher energy in Path 2 and Path 3, Fig. 10. In Path 2 the short oxametallacycle OM1 C is formed directly, with the activation barrier of 28 kJ mol−1 . The corresponding transition state, TS3, lies 44 kJ mol−1 higher than TS1. Alternatively, epoxide can be directly formed along Path 3 and TS4. Transition states TS3 and TS4 have about the same height; therefore, we cannot judge on the basis of energies alone which one would be preferred. Pathways similar to Path 2 and Path 3 have also been identified on the Au38 nanoparticle, with the activation energies of 80 and 95 kJ mol−1 for the analogs of TS3 and TS4, respectively, relative to co-adsorbed propylene and OOH [20]. These values would agree well with ours if we took the lowest co-adsorbed state for propylene and OOH (from Path 1) as reference (in ref. [20] propylene is adsorbed at a low-coordinated apex site in the initial state, similar to a kink site of Au(321)). In any case, Path 1 seems to be the lowest-energy pathway identified so far and it lies lower in energy than the respective lowest-energy path on the C3 H6 * + O* potential energy surface. Remarkably, all three identified pathways for oxidation with OOH* bypass the formation of an intermediate that can easily isomerize to acetone. The short oxametallacycle OM1 C can be converted to acetone but the activation barrier in this case is much higher than that for epoxide formation. The TS structure for Htransfer to form acetone is shown in TS6 of Fig. 12. The TS structure has an even more early character than the analogous TS from the
large-ring OM2 C , consistent with the stronger exothermicity of this transformation from OM1 C . The activation barriers for transformation to acetone are almost of the same height from OM1 C and OM2 C but because the epoxide formation is so facile from OM1 C , the conversion to acetone can no longer be competing with it at low temperature. Also the H abstraction from the methyl group of propylene by OOH* has higher activation energy than that of H abstraction by surface O. The transition state for this reaction, TS5, is calculated to lie slightly above TS1 as can be seen in Fig. 10. The activation energy for H abstraction with OOH* is as high as 54 kJ mol−1 . It should be kept in mind, however, that some of the OOH* species formed can dissociate to OH* + O* before reacting with propylene, though with a relatively high activation barrier of ∼65 kJ mol−1 calculated with our surface model as can be seen in Fig. 9. Further, two surface OH species can recombine forming H2 O and surface O, with a low activation barrier of only 10 kJ mol−1 , Fig. 9. Therefore, even in the presence of hydrogen or water it will probably not be possible to completely replace surface O species by OOH, but it is important to be able to reduce the concentration and reactivity of the former.
4. Conclusions We have investigated the complex reaction network for propylene epoxidation with surface O and OOH as oxidants, including important side reactions. We have shown that in the reaction of propylene with chemisorbed atomic oxygen not only the formation of allylic species but also isomerization reactions of metallacycles to carboxylic compounds are responsible for low epoxide selectivity. Furthermore, five-member ring oxametallacycles are energetically deeply lying intermediates on the global potential energy surface with relatively high activation barriers for conversion to products that can desorb from the surface, including epoxide and carboxylic compounds. Therefore, OM2 C oxametallacycles, which can further easily dehydrogenate and undergo further O addition, decarbonylation and decarboxylation steps, are likely intermediates on the way to total combustion. In contrast, allylic species are shown to
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11 10
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
Fig. 12. Atomic geometries of the key minima and transition states leading to allyl (left) and acetone (right) formation corresponding to energetic profiles of Fig. 10.
tion) but also proceeds along a reaction path of an overall lower energy with respect to the gas-phase reagents. Acknowledgements We acknowledge the financial support from the German Research Foundation (DFG) within the Project No. MO 1863/2-2 and MO 1863/4-1. The calculations were performed at the HPC Cluster HERO, located at the University of Oldenburg and funded by the DFG through its Major Research Instrumentation Program (INST 184/108-1 FUGG) and the Ministry of Science and Culture of the Lower Saxony Federal State, and in part with resources provided by the North-German Supercomputing Alliance (HLRN), Project No. hbc00018. Appendix A. Supplementary data Fig. 11. Atomic geometries of the key minima and transition states leading to epoxide formation along the various pathways of propylene reaction with hydroperoxyl, corresponding to energetic profiles of Fig. 10.
have low barriers for conversion to acrolein or allyl alcohol. The former is found as major partial oxidation product on gold catalysts in the absence of hydrogen or water. We have shown that the addition of hydrogen or water to the reacting gas mixture readily produces surface hydroperoxo species with activation energies much lower than that required for O2 dissociation. Therefore, the propylene partial oxidation reaction can proceed at milder oxidation conditions, avoiding deep oxidation. The present study clearly shows that the oxidation with hydroperoxyl is not only more selective than the reaction with surface atomic oxygen (due to the suppression of propanal and acetone formation and due to a higher activation energy of allylic hydrogen abstrac-
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.cattod.2016.05. 034. References [1] D.L. Trent, Propylene Oxide, Kirk-Othmer Encyclopedia of Chemical Technology, John Wiley & Sons Inc., 2000. [2] J.W. Medlin, J.R. Monnier, M.A. Barteau, J. Catal. 204 (2001) 71–76. [3] J. Huang, M. Haruta, Res. Chem. Intermed. 38 (2012) 1–24. [4] T. Hayashi, K. Tanaka, M. Haruta, J. Catal. 178 (1998) 566–575. [5] A.K. Sinha, S. Seelan, S. Tsubota, M. Haruta, Angew. Chem. Int. Ed. 43 (2004) 1546–1548. [6] B. Taylor, J. Lauterbach, W.N. Delgass, Appl. Catal. A 291 (2005) 188–198. [7] L. Cumaranatunge, W.N. Delgass, J. Catal. 232 (2005) 38–42. [8] J. Huang, T. Akita, J. Faye, T. Fujitani, T. Takei, M. Haruta, Angew. Chem. Int. Ed. 48 (2009) 7862–7866. [9] M. Ojeda, E. Iglesia, Chem. Commun. (2009) 352–354.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034
G Model CATTOD-10226; No. of Pages 11
ARTICLE IN PRESS L.V. Moskaleva / Catalysis Today xxx (2016) xxx–xxx
[10] S. Lee, L.M. Molina, M.J. López, J.A. Alonso, B. Hammer, B. Lee, S. Seifert, R.E. Winans, J.W. Elam, M.J. Pellin, S. Vajda, Angew. Chem. Int. Ed. 48 (2009) 1467–1471. [11] J. Huang, T. Takei, T. Akita, H. Ohashi, M. Haruta, Appl. Catal. B 95 (2010) 430–438. [12] J.J. Bravo-Suárez, K.K. Bando, J. Lu, M. Haruta, T. Fujitani, T. Oyama, J. Phys. Chem. C 112 (2008) 1115–1123. [13] J.-H. Franke, D.S. Kosov, J. Chem. Phys. 139 (2013) 224709. [14] J.L.C. Fajín, J.R.B. Gomes, M.N.D.S. Cordeiro, Langmuir 29 (2013) 8856–8864. [15] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, J.Phys. Chem. C 112 (2008) 17291–17302. [16] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, J. Catal. 289 (2012) 11–20. [17] L.V. Moskaleva, S. Rohe, A. Wittstock, V. Zielasek, T. Kluner, K.M. Neyman, M. Baumer, Phys. Chem. Chem. Phys. 13 (2011) 4529–4539. [18] L.V. Moskaleva, V. Zielasek, T. Klüner, K.M. Neyman, M. Bäumer, Chem. Phys. Lett. 525–526 (2012) 87–91. [19] L.V. Moskaleva, T. Weiss, T. Klüner, M. Bäumer, J. Phys. Chem. C 119 (2015) 9215–9226. [20] C.-R. Chang, Y.-G. Wang, J. Li, Nano Res. 4 (2011) 131–142. [21] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 77 (1996) 3865–3868. [22] J.P. Perdew, K. Burke, M. Ernzerhof, Phys. Rev. Lett. 78 (1997), 1396–1396. [23] G. Kresse, J. Furthmüller, Comput. Mater. Sci. 6 (1996) 15–50. [24] G. Kresse, J. Hafner, Phys. Rev. B 49 (1994) 14251–14269. [25] P.E. Blöchl, Phys. Rev. B 50 (1994) 17953–17979. [26] G. Kresse, D. Joubert, Phys. Rev. B 59 (1999) 1758–1775. ˜ [27] P. Janthon, S.M. Kozlov, F. Vines, J. Limtrakul, F. Illas, J. Chem. Theory Comput. 9 (2013) 1631–1640. ˜ [28] P. Janthon, S. Luo, S.M. Kozlov, F. Vines, J. Limtrakul, D.G. Truhlar, F. Illas, J. Chem. Theory Comput. 10 (2014) 3832–3839. [29] H.J. Monkhorst, J.D. Pack, Phys. Rev. B 13 (1976) 5188–5192. [30] J.L.C. Fajín, M.N.D.S. Cordeiro, J.R.B. Gomes, F. Illas, J. Chem. Theory Comput. 8 (2012) 1737–1743.
11
[31] G. Henkelman, H. Jónsson, J. Chem. Phys. 111 (1999) 7010–7022. [32] S. Linic, M.A. Barteau, J. Am. Chem. Soc. 124 (2002) 310–317. [33] D. Torres, N. Lopez, F. Illas, R.M. Lambert, Angew. Chem. Int. Ed. 46 (2007) 2055–2058. [34] K.A. Davis, D.W. Goodman, J. Phys. Chem. B 104 (2000) 8557–8562. [35] X. Deng, B.K. Min, X. Liu, C.M. Friend, J. Phys. Chem. B 110 (2006) 15982–15987. [36] A. Roldan, D. Torres, J.M. Ricart, F. Illas, J. Mol. Catal. A: Chem. 306 (2009) 6–10. [37] T.A. Baker, B. Xu, S.C. Jensen, C.M. Friend, E. Kaxiras, Catal. Sci. Tech. 1 (2011) 1166–1174. [38] A. Pulido, P. Concepción, M. Boronat, A. Corma, J. Catal. 292 (2012) 138–147. [39] F.A. Carey, R.J. Sundberg, Advanced Organic Chemistry, 4th ed., Kluwer, New York, 2002. [40] J.W. Medlin, M.A. Barteau, J. Phys. Chem. B 105 (2001) 10054–10061. [41] M. Enever, S. Linic, K. Uffalussy, J.M. Vohs, M.A. Barteau, J. Phys. Chem. B 109 (2005) 2227–2233. [42] H. Piao, M.C.N. Enever, K. Adib, J. Hrbek, M.A. Barteau, Surf. Sci. 571 (2004) 139–145. [43] A. Klust, R.J. Madix, Surf. Sci. 600 (2006) 5025–5040. [44] L. Zhou, R.J. Madix, J. Phys. Chem. C 112 (2008) 4725–4734. [45] G.S. Hammond, J. Am. Chem. Soc. 77 (1955) 334–338. [46] N.W. Cant, W.K. Hall, J. Phys. Chem. 75 (1971) 2914–2921. [47] P.V. Geenen, H.J. Boss, G.T. Pott, J. Catal. 77 (1982) 499–510. [48] R.A. van Santen, Theoretical Heterogeneous Catalysis, World Scientific Singapore, 1991. [49] M.-L. Bocquet, A. Michaelides, D. Loffreda, P. Sautet, A. Alavi, D.A. King, J. Am. Chem. Soc. 125 (2003) 5620–5621. [50] M.O. Özbek, I. Önal, R.A. van Santen, ChemCatChem 3 (2011) 150–153. [51] I.X. Green, W. Tang, M. Neurock, J.T. Yates, Angew. Chem. Int. Ed. 50 (2011) 10186–10189. [52] M. Boronat, A. Pulido, P. Concepcion, A. Corma, Phys. Chem. Chem. Phys. 16 (2014) 26600–26612.
Please cite this article in press as: L.V. Moskaleva, Theoretical mechanistic insights into propylene epoxidation on Au-based catalysts: Surface O versus OOH as oxidizing agents, Catal. Today (2016), http://dx.doi.org/10.1016/j.cattod.2016.05.034