Article pubs.acs.org/IC
Understanding the Lewis Acidity of Co(II) Sites on a Silica Surface Deven P. Estes,† Amanda K. Cook, Erwin Lam, Louise Wong,‡ and Christophe Copéret* Department of Chemistry and Applied Biosciences, ETH Zürich, Vladimir-Prelog Weg 1-5, CH-8093 Zürich, Switzerland S Supporting Information *
ABSTRACT: Heterogeneous catalysts consisting of isolated transition-metal sites dispersed on the surface of metal oxide supports are commonly used in the chemical industry. Often their reactivity relies on the Lewis acidity of the active sites on the surface of the catalyst. A recent report from our group showed that silica-supported Co(II) sites, prepared via surface organometallic chemistry, are active in both alkene hydrogenation and alkane dehydrogenation, possibly linked to the Lewis acidity of the Co(II) sites. Here we use molecular probes and analogues to both qualitatively and quantitatively model the Lewis acidity of the surface sites. Some sites do not bind probe molecules like carbon monoxide, tetrahydrofuran, and olefins, while others exhibit a continuum of Lewis acidities. This is consistent with variations in the coordination environment of Co. These results suggest that only the most Lewis acidic sites are involved in dehydrogenation and hydrogenation, consistent with catalyst poisoning studies.
■
INTRODUCTION Isolated metal sites supported on metal oxide surfaces are an important class of industrial catalysts used in a variety of applications such as ethylene polymerization, alkene metathesis, and alkane dehydrogenation, to name a few.1 However, these catalysts are not well understood because of the polydispersity of the surface sites with a range of reactivities. In fact, it is often the case that heterogeneous catalysts have a broad distribution of sites, associated with a broad range of reactivity and a small fraction of active sites. In some cases, such as in the Phillips ethylene polymerization catalyst,2 this distribution of sites provides the unique properties in the thus-formed polymer. In most cases, however, it leads to low selectivity and an overall low efficiency of metal utilization. In order to improve chemical processes, an important step is a molecular understanding of surface structures and the building of robust structure−activity relationships for rational catalyst design. One possible approach to shedding some light on this problem has been to use surface organometallic chemistry (SOMC).3 SOMC, coupled with the thermolytic molecular precursor approach,4 allows the synthesis of “naked” surface sites with relatively well-defined coordination spheres as models of these industrial catalysts.5 Using this approach, we synthesized well-defined models of industrial polymerization and metathesis catalysts, with ca. 50−60% active sites.5a,6 However, these catalysts also consist of a distribution of surface sites, the reactivity of which depends on the local coordination environment thought to be caused by the amorphous character of silica.7 While this method still gives a distribution of sites, the increased number of active surface sites has allowed us to obtain a more detailed understanding of the reaction mechanism and to detect in some instances reaction intermediates. © XXXX American Chemical Society
Hock and co-workers as well as our own group recently found that Co(II) sites on silica (1) surfaces catalyze alkane dehydrogenation, alkene hydrogenation, and alkyne trimerization.5e,8 Characterization of these surface sites showed that they consisted of monomeric high-spin tetrahedral Co(II) sites, on average, where the two neutral O ligands correspond to Si−O−Si bridges. However, little to no information about the coordination environment of specific Co sites was obtained.5e We proposed that these reactions all involve C−H activation by a 1,2-addition-type mechanism (eq 1). In this mechanism, the
ability of a site to heterolytically split a C−H bond is directly related to its Lewis acidity.9 In view of the presence of different types of surface sites in these model systems, we have decided to investigate the Lewis acidic properties of these surface sites. A typical procedure for assessing the qualitative acidity of a support is to adsorb molecular probes onto the surface of the catalyst. Various probes exist for Lewis acidity such as pyridine,10 carbon monoxide (CO),11 dihydrogen (H2) at 77 K,11a water, or tetrahydrofuran (more common in homogeneous catalysis).12 This strategy qualitatively confirms the existence of both Brønsted and Lewis acidic sites on the catalyst surface. Here we apply this strategy in order to both qualitatively and quantitatively characterize the Lewis acidic surface sites in Received: February 17, 2017
A
DOI: 10.1021/acs.inorgchem.7b00443 Inorg. Chem. XXXX, XXX, XXX−XXX
Article
Inorganic Chemistry
parameters for each gas from both models are given in Table S2. The fits are shown, with Q values in units of moles of gas per moles of Co.
catalyst 1. We can qualitatively determine the type of interaction by using spectroscopic handles (such as IR spectroscopy) to understand the electronic interaction of the probe molecule with the surface sites. We use gas and liquid titration methods (such as constant-volume chemisorption) to quantitatively assess both the number and strength of Lewis acidic sites on the surface of 1. We gain further insight on the structural origin of the Lewis acid properties of the various sites by combining this with studies of ligand binding to molecular models and density functional theory (DFT) calculations. Poisoning studies ultimately show the relationship between the number of active sites in catalysis and their Lewis acidity properties.
■
Q total =
Q 1K1P 1 + K1P
+
Q 2K 2P 1 + K 2P
Q total = Q maxKP RT / E0
(2) (3)
THF Titration by NMR. A known amount of 1 (∼30 mg) was weighed into a J. Young NMR tube under Ar along with a known amount of ferrocene as the internal standard (∼20 mg), and C6D6 was added by syringe (0.5 mL). Small amounts of a stock solution of 0.0205 M THF in C6D6 were added to the slurry and allowed to equilibrate for 5−10 min, and then a quantitative (60 s recycle delay) 1 H NMR spectrum was taken to quantify the contents of the solution. The results are plotted in Figure 1.
EXPERIMENTAL SECTION
All experiments were performed under an inert atmosphere using glovebox (Ar), Schlenk (Ar), or high-vacuum (10−5 mbar) techniques. Pentane was dried by passage through two columns of activated alumina, degassed prior to use, and tested by adding drops of a sodium/benzophenone ketyl solution in tetrahydrofuran (THF). THF was distilled from purple sodium/benzophenone under Ar and stored over activated molecular sieves. C6D6 was vacuum distilled from purple sodium/benzophenone. 1H NMR spectra were obtained at 300 MHz. Transmission IR spectra were recorded on a Bruker Alpha FT-IR spectrometer. This was done using self-supporting silica disks and high-vacuum techniques as described elsewhere.3a The molecular complex Co2(OSi(OtBu)3)4 (2) was synthesized by reacting Co(HMDS)2(THF) with 2 equiv of HOSi(OtBu)3. Subsequent grafting on highly dehydroxylated silica (Aerosil 200, dehydroxylated at 700 °C), followed by thermal treatment at 500 °C under high vacuum, provide silica (1), as previously reported.5e Preparation of Co2(OSi(OtBu)3)4(THF) (3). 2 (200 mg) was first dissolved in minimal pentane (∼1 mL). Then, the addition of 2 drops of neat THF resulted in a color change from green to blue-purple and the almost immediate formation of crystals. These crystals were redissolved by the addition of a bit more pentane, and then the mixture was allowed to stand at −40 °C for 1 week, after which blue plates had formed in the solution. These crystals were used for X-ray crystallography, the results of which are shown in Table S1. We were not able to characterize this molecule by NMR because it was not possible to isolate this molecule in solution. Redissolving the blue crystals for NMR analysis yielded a solution with properties similar to those of the THF free molecule, making NMR characterization impractical. Dissolution of the base-free dimer in any amount of THF (including neat THF) results in a mixture of species, which exchange with one another rapidly on the NMR time scale. Elem anal. Calcd: C, 50.22; H, 9.40. Found: C, 49.86; H, 9.02. IR (KBr pellet, cm−1): 2973, 2926, 2899, 2864 (CH stretching), 1472, 1386, 1360 (CH bending) 1245, 1192, 1059 (s), 1015, 986, 950, 915, 827, 694, 488. UV−vis (diffuse reflectance; λmax, nm): 463, 513, 580, 680 (Figure S24). Gas Adsorption. Chemisorption experiments were performed using a Bel-Max instrument from BelJAPAN. Gases were passed through purification columns to remove oxygen and water prior to use. Before the measurement, the samples (approximately 0.1 g) were transferred from the glovebox to the Bel-Max using an airtight cell. Both adsorption and desorption curves were obtained in order to check for reversibility. The results of these adsorption experiments were compared to those of bare silica in order to account for physisorption to the silica surface. The curves obtained were fit to a variety of binding isotherms and found to fit best to either a double Langmuir model (eq 2) or a Freundlich isotherm (eq 3) (vide infra). The parameter E0 from eq 3 represents the difference in adsorption enthalpy between the most strongly and most weakly binding sites and can be extracted from the exponent of the fitting. However, in the Freundlich model, the parameters Qmax and K are fit as one component, and thus it is not possible to extract them individually by fitting alone. However, if we assume that Qmax in eq 3 is equal to the sum of Q1 and Q2 from eq 2, then we can extract the equilibrium constant for the fit from the Freundlich isotherm. The fitted
Figure 1. (Top) Chemisorption binding curves for THF (red, THF on 1; black, THF on SiO2; blue, background-corrected THF adsorption on 1). (bottom) CO (red) and ethylene (blue) binding to 1, with CO and ethylene fit to a double Langmuir isotherm model. Titration of Active Sites in Ethylene Hydrogenation. A known amount of catalyst 1 was suspended in pentane, and an appropriate amount of a 4-methylpyridine solution (0.062 M in pentane) was added in order to achieve the appropriate number of equivalents of poison. This material was stirred for 10 min, and then pentane was distilled away under high vacuum. This material was placed in a flow reactor (less than 50 mg per run), and the activity of the catalyst was measured at 200 °C, with the H2 flow varying between 3.4 × 10−6 and 1.9 × 10−5 mol of H2 s−1 and a constant ethylene flow of 1.4 × 10−6 mol of ethylene s−1. The results are plotted in Figure S17. DFT Calculations. All calculations were performed with Gaussian09(d1).13 Geometry optimizations were done using the LANL2DZ basis set and the corresponding effective core potential for Co and the 6-31g(d)14 basis set for the remaining atoms in combination with the UB3LYP15 functional. Co was considered in its S = 3/2 spin state without superexchange coupling between the Co dimers. This was confirmed by SQUID measurements on 2 (Figure S23) and B
DOI: 10.1021/acs.inorgchem.7b00443 Inorg. Chem. XXXX, XXX, XXX−XXX
Article
Inorganic Chemistry the relative energies of lower spin states (Table S3). The nature of the geometry found was confirmed by analyzing the vibrational frequencies at the same level of theory. Single-point calculations were done by using the LANL2DZ16 basis set and the corresponding effective core potential for Co and the 6-311+g(d,p)17 basis set for the remaining atoms in combination with the UB3LYP-3D functional.
binding sites, one strongly and one weakly binding. The strong binding event accounted for 0.19 mol of C2H4·mol of Co−1 and 0.28 mol of C3H6·mol of Co−1 at 25 °C. The weak binding events accounted for binding of 0.23 mol of C2H4·mol of Co−1 and 0.18 mol of C3H6·mol of Co−1. All of the fitted parameters are given in Table S2. The more complex Freundlich isotherm (eq 3) also adequately modeled the data. In contrast to the double Langmuir model used above, the Freundlich isotherm assumes that the surface consists of a distribution of sites having an exponential distribution of binding energies.19 The parameter K in eq 3 represents the equilibrium binding constant for the most weakly binding sites, while the parameter E0 represents the difference in the binding energy between the most strongly and most weakly adsorbing sites.19b,20 In our case, this could be attributed to a continuous variation of the Lewis acidities of the surface sites in which a few sites are very Lewis acidic, while the bulk of the sites either are mildly acidic or have very low acidity. This could be caused by the structural inhomogeneity of the sites on the amorphous silica surface (vide infra). This model matches our recent results from periodic calculations on silica slabs that show that the reactivity of surface sites varies from very exothermic to moderately endothermic depending on the structural variations around the surface metal ion.7a It also parallels what we have observed on Ln-doped silica, where a distribution of sites was observed by luminescent t1 decay.21 The binding of olefins can occur to metals that are either Lewis acidic or strong π donors.22 In order to better characterize the nature of the interaction of the Co surface sites with ligands, we studied the adsorption of CO. We reported previously that IR spectroscopy shows that, upon exposure of the catalyst to 10 Torr of gaseous CO, a new band appears in the IR spectrum at 2184 cm−1.5e This blue-shifted band is atypical for metal complexes that bind CO, but it is quite common for metal sites on metal oxide or zeolite surfaces.11 Such a blue shift indicates that CO is interacting with very Lewis acidic or partially charged species, showing that the Co surface sites are poor π donors and most likely Lewis acidic. The fact that only one CO frequency is visible argues against the presence of two distinct sites. We monitored the equilibrium binding of CO to the surface sites similarly to what was done above for olefins. We observed once again that a simple Langmuir isotherm was not adequate to model the data. Both the two-site Langmuir model and the Freundlich isotherm are able to model the data obtained from chemisorption. Roughly the same number of sites bind CO or olefins. By the two-site model, 0.28 mol of CO·mol of Co−1 binds strongly and 0.24 mol of CO·mol of Co−1 binds weakly. The number of sites that strongly bind olefins and CO is very similar to the number of sites that bind THF in solution. Interestingly, the lower quantity of THF able to bind to surface sites is probably due to the competition between the surface binding and the interaction of THF with solvent molecules via van der Waals interactions, which would be energetically competitive with coordination to the more weakly Lewis acidic sites. These data suggest that the binding of all of these ligands occurs at the same sites by a Lewis acid/base interaction. The models used to fit the ligand binding to the surface suggest that there are a variety of sites on the surface that bind ligands with different binding energies, likely caused by local structural variations at the Co sites. Siloxide complexes, specifically those containing the −OSi(OtBu)3 ligand, often serve as models for surface species.
■
RESULTS AND DISCUSSION We first examined the binding of THF to 1. We titrated a dilute THF solution into a slurry of 1 in C6D6. By comparison with the internal standard, we were able to measure the amount of excess THF in the supernatant solution after equilibration of the mixture over the course of 5−10 min (Figure 1). The addition of less than 0.3 equiv of THF per Co surface site resulted in no free THF (
