Importance of the decoration in shaped cobalt nanoparticles in the acceptor-less secondary alcohol dehydrogenation

Kamila Kaźmierczakab, Raj Kumar Ramamoorthyc, Arthur Moissetc, Guillaume Viauc, Arnaud Violad, Marion Giraudd, Jennifer Perond, Lorette Sicardd, Jean-Yves Piquemald, Michèle Bessonb, Noémie Perret*b and Carine Michel*a
aUniv Lyon, ENS de Lyon, CNRS UMR 5182, Laboratoire de Chimie, Université Claude Bernard Lyon 1, 46 Allée d'Italie, 69342, Lyon, France. E-mail: carine.michel@ens-lyon.fr
bUniv Lyon, CNRS, UMR5256, IRCELYON, Université Claude Bernard Lyon 1, 2 avenue Albert Einstein, 69626, Villeurbanne, France. E-mail: noemie.perret@ircelyon.univ-lyon1.fr
cLPCNO CNRS, INSA, UPS UMR 5215, Université de Toulouse, 135 avenue de Rangueil, 31077 Cedex 4, Toulouse, France
dITODYS, CNRS, UMR 7086, Université de Paris, 15 rue J.-A. de Baïf, F-75013 Paris, France

Received 28th February 2020 , Accepted 14th April 2020

First published on 30th April 2020


Metal catalysts are essential in the production of fuels and chemicals. Nonetheless, tailoring the exposed active sites to achieve the maximal theoretical conversion is still a great challenge. In the case of structure-sensitive reactions, such as the attractive acceptor-less alcohol dehydrogenation, playing on the exposed metallic sites appears as an appealing strategy. Still, this approach requires advanced preparation protocols and is even more difficult to implement for supported non-noble metal catalysts which easily undergo sintering. Using the polyol method, we synthesized fourteen different cobalt catalysts, which consist of unsupported shaped nanoparticles stabilized by adsorbed carboxylate ligands. Their shape and the amount of ligands were characterized by combining TEM and TGA-N2 measurements. These catalysts were found to be active in the 2-octanol dehydrogenation conditionally upon an organic layer limited to 1 to 2 monolayers. Moreover, they were fully selective towards the desired ketone and H2. The active catalysts were stable, with no leaching or modification of the shape during the reaction. Periodic DFT computations predict a greater activity of the pristine open-type facet than of the compact one, but this is not confirmed experimentally with no clear correlation between the activity expressed in turnover number and the amount of a given type of site as quantified by TEM. Further modeling including the organic layer shows that the presence of ligands reduces the sensitivity to the metallic structure. Nonetheless, these ligands generate a catalytic pocket, similar to the one found in enzymes, that interacts with the alcohol substrate through H-bonding. This pocket is the most adapted to the alcohol dehydrogenation on the open-type facet, which is mainly exposed on rods. This detailed understanding paves the way to an improved design of bespoke unsupported catalysts considering simultaneously the structure and the nature of the ligand.


1. Introduction

For structure-sensitive reactions, the performance of metal-supported catalysts can be improved by playing on the size of the metallic nanoparticles (NPs).1 In particular, low coordinated metallic sites can have a better catalytic ability than high coordinated ones, and their relative concentration increases with the decrease of NP size. For instance, corner sites were shown to be 200 times more active than terrace sites for Au supported over hydrotalcite (HT) for alcohol dehydrogenation, but they are also 4 times less present on NPs of 2 nm size and almost 200 times less on 12 nm size NPs.2 Obtaining supported NPs that are small enough (below 2 nm) to expose a noticeable amount of low coordinated sites appears a difficult task when using conventional routes such as wet impregnation, especially when turning to abundant non-noble metals, i.e., Co, Ni and Cu.3–5 Another approach to increase the number of low coordinated metallic sites is to design NPs that are shaped to expose mainly open-type facets. For instance, Xu et al.6 demonstrated that Ag nanocubes were much more active in the oxidation of styrene than near-spherical NPs (4 times more active) and platelets (14 times more active). This effect was related to a higher activity of the Ag (100) open-type facet, predominantly present on nanocubes, in comparison with the Ag (111) close-packed surface, present on both near-spherical NPs and platelets. Besides the activity, the selectivity can also be affected by the type of exposed surface. It was reported for benzene hydrogenation over Pt NPs7 that the Pt (100) surface favors complete hydrogenation to cyclohexane, while the Pt (111) facet promotes partial hydrogenation to cyclohexene. These tailored NPs can be prepared with solvothermal or hydrothermal processes using not only noble but also non-noble metals.8–10 They were recently acknowledged as a promising new class of catalysts, yet still requiring more investigations and understanding.11–14

Nonetheless, these solvothermal preparations usually require the presence of ligands to stabilize and direct the NP growth.15,16 If the NPs are not supported afterwards, those ligands are necessary to prevent aggregation in solution. Yet, the presence of ligands is very often found to be detrimental to the catalytic activity of metal NPs. For instance, Au25/C was shown to be less and less active in the aerobic oxidation of benzyl alcohol as the dodecanethiolate surface coverage increased.17 The catalytic activity vanished once a full surface coverage was reached. However, ligands were also found to have a valuable impact in some cases, especially on reaction selectivity,18–20 a positive influence attributed to steric,21 orientation,22 and electronic23 effects. For example, the ligand density can limit the possible orientations of poly-functional molecules when accessing the catalytic sites. This was nicely illustrated recently by Medlin and co-workers on the hydrogenation of cinnamaldehyde on Pt/Al2O3, a typical reaction where selectivity towards the unsaturated alcohol is key. When the catalyst was modified using 3-phenylpropanethiol ligands,22 the chemoselectivity of the reaction was considerably improved, with a raise of the selectivity towards unsaturated alcohol from 25% to 95%. The limited access to the surface sites and the structural proximity of the cinnamyl aldehyde to the chosen ligand (3-phenylpropanethiol) were invoked to rationalize this strong improvement. The same strategy was later applied to furfuryl alcohol hydrogenation.24 The desired hydrodeoxygenation to methylfuran is increased by adjusting the thiolate ligand chain. Using density functional theory (DFT), the thiolate ligands were found to suppress the unwanted decarbonylation route while stepped sites were still available for the sought hydrodeoxygenation. Computational investigations were also key to understand the differential impact of mono- vs. bidentate phosphines on the decarbonylation of fatty alcohols to α-olefins catalyzed by Pd. While the monodentate completely passivated the catalyst, the polydentate favored a highly selective activity. Ortuño and López25 demonstrated that the higher flexibility of the bidentate ligands was essential to create catalytically active pockets.

To prepare decorated shaped NPs, the polyol method26 (one of the solvothermal protocols) appears as an appealing procedure that is able to generate a variety of shapes and is scalable to the preparation of several grams of powder per batch at laboratory scale. We recently focused on the preparation of Co NPs using it.27,28 By varying the preparation parameters, multiple shapes were obtained, like nanorods,27,29,30 nanowires,31 sea-urchin-like particles,27 cubic particles,27 hourglass-like particles30 and platelets.29,30 Combining experimental and periodic DFT investigations, we also rationalized the role of the carboxylate ligand in the shape control.29 Changing its concentration in the growth solution tunes its chemical potential and hence the relative stability of decorated crystallographic surfaces, and thereby the shape of the Co NPs.

Acceptor-less alcohol dehydrogenation is a very appealing reaction, since it is atom efficient and it leads to high-value added products, namely a ketone and H2.32 However, it requires an elevated temperature to proceed and an anaerobic atmosphere to adopt the desired acceptor-less mechanism.33 Many different supported noble and non-noble metal catalysts were reported as active in this reaction, i.e. Pt,34 Pd,35 Ru,36,37 Re,38 Rh,39 Au,2,40,41 Ag,42–45 Cu,46–49 Ni (ref. 50 and 51) and Co.52 Regarding the unsupported metal catalysts, Re NPs were reported as active towards acceptor-less alcohol dehydrogenation.53 Also, we have recently shown that Co nanorods stabilized by carboxylate (laurate) ligands have good catalytic properties for this reaction, with a high chemoselectivity towards a broad substrate scope including linear and cyclic alcohols and good recyclability (3 runs with no decrease in activity and selectivity and no modification of the anisotropic shape).54 The potentialities of these catalysts are clearly not fully exploited yet. Playing with the shape and the type of exposed facets appears as an interesting path to follow, since two very recent studies highlighted that the alcohol dehydrogenation reaction is structure sensitive on Cu. He et al.55 showed that the Cu(211) stepped surface obtained after surface re-structuring of Cu supported catalysts exhibits a higher activity towards ethanol dehydrogenation than the close-packed Cu(111) surface. Using periodic DFT calculations, Hoyt et al.56 demonstrated that dehydrogenation reaction intermediates are more strongly bonded on the Cu (111)/(100) step edges than on the Cu (111) surface. The related activation energies are also lower on step edges, making those sites clearly more active than the close-packed surface sites.

To fully exploit the potentialities of decorated shaped NPs, improved understanding is needed to grasp the nature of the catalytic sites made up of exposed (under)-coordinated metal atoms surrounded by organic ligands. In this work, we took advantage of the flexibility of the polyol process to prepare a variety of Co NPs decorated with carboxylate ligands and test these non-supported catalysts towards 2-octanol dehydrogenation. The NP shape was intentionally tuned to expose various facets in different amounts to probe the structure sensitivity of the alcohol dehydrogenation on Co. The carboxylate chain length was also modified to investigate its impact on the catalytic activity, following the work of Medlin and co-workers on the hydrogenation of cinnamaldehyde.22 Lastly, the influence of the presence of ligands on the exposed catalytic sites on the activity was rationalized through DFT computations.

2. Materials and methods

2.1. Synthetic procedures

The list of all materials used is given in the ESI in section 1.1.
2.1.1. Preparation of cobalt precursors. Several cobalt(II) precursors were prepared with different long-chain carboxylates: heptanoate, octanoate, decanoate, laurate and palmitate, respectively denoted as H, O, D, L and P. The detailed procedures are given in the ESI. Briefly, it consists in reacting under vigorous stirring the desired carboxylic acid with a Co(II) salt in a basic (NaOH) aqueous medium. Pink precipitates are recovered, washed with distilled water, and dried in an oven. Depending on the drying procedure, phases with different hydration states were obtained for the same carboxylate compound (ESI).
2.1.2. Polyol syntheses of Co particles. Co particles with different sizes and shapes were synthesized following adapted procedures reported elsewhere.27 It consists in reacting a cobalt(II) carboxylate compound in a basic polyol(1,2-propanediol or 1,2-butanediol) at 175 °C using RuCl3 as a seeding agent. The detailed procedures are given in the ESI. The morphologies and mean dimensions of the particles can be varied depending on several reaction parameters such as the nature of the polyol, the nature of the Co(II) carboxylate, its drying procedure, the concentration of the carboxylate and the nature of the seeding agent (hydrated or anhydrous RuCl3). In this work, rods, diabolos and platelets were prepared. The following labelling scheme was chosen: type of ligand – shape – order number (when necessary), where the type of ligand indicates which carboxylate was used as a capping ligand (vide supra) and the shape corresponds to rods (R), diabolos (D) or platelets (P).

2.2. Nanoparticle characterization

2.2.1. TEM. Transmission electron microscopy (TEM) characterization was performed using a JEOL JEM-1011 instrument equipped with a LaB6 filament and operating at 100 kV. The images were collected using a 4008 × 2672 pixel CCD camera (Gatan Orius SC1000). The mean particle sizes were determined by a statistical analysis of at least 200 particles. All samples were prepared by evaporating a drop of diluted suspension in ethanol on a carbon-coated copper grid. Specific surface areas exposed by metal in the samples (SSAC) were calculated using the mean dimensions determined by TEM and assuming simple geometrical models, as explained in the ESI.
2.2.2. TGA. TGA measurements were performed under N2 to assess the mass of organic ligands of the catalysts (ΔmTGA-N2). They were performed using TGA-DSC 1 Stare System Mettler Toledo apparatus, in the temperature range 20–1000 °C at a heating rate of 10 °C min−1 and 50 mL min−1 total flow of gas.
2.2.3. XRD. X-ray diffraction analyses were carried out using a PANalytical X'pert Pro diffractometer equipped with a Co anode (λ = 1.7889 Å) and an X'celerator detector. The sizes of coherent diffraction domains were determined using MAUD software,57 which is based on the Rietveld method combined with Fourier analysis, and is well adapted for broadened diffraction peaks. Corrections for instrument broadening were made using a polycrystalline silicon standard from PANalytical.
2.2.4. N2 physisorption. The specific surface area of the samples (cobalt nanoparticles covered with ligands) was measured by N2 physisorption at 77 K using an ASAP 2020 apparatus from Micromeritics. The surface area (SSABET) was calculated using the Brunauer, Emmet and Teller (BET) equation in the relative pressure range (0.05–0.25). Before the analysis, samples were degassed under high vacuum (<150 mPa) for 12 h at 120 °C.
2.2.5. Surface coverage. For a given sample, the surface coverage with organic ligands was evaluated as the ratio between the amount of ligands (nligands, in mol) for a given mass (msample), as quantified by TGA, and the surface of exposed metallic Co (SA, in m2) for the same mass of sample, as derived from the TEM image analysis. The established accuracy of this value is ±2.5 × 10−6 mol m−2.
image file: d0cy00390e-t1.tif

image file: d0cy00390e-t2.tif

image file: d0cy00390e-t3.tif

2.3. Catalytic tests

2.3.1. Catalytic test conditions. Catalytic tests were performed to assess the activity of Co shaped NPs in the dehydrogenation of (±)-2-octanol in n-decane. They were conducted for 24 h at 145 °C using 25 mg of catalyst (1 mol%) in 45 mL of liquids, with the alcohol concentration equal to 0.95 mol L−1, in a 100 mL semi-batch glass reactor with a constant flow of inert gases (90% Ar and 10% N2, total flow 30 mL min−1) and mechanical stirring (750 rpm). The reactor was coupled with a gas chromatograph (Shimadzu GC-2010, Supelco Carboxen-1010 PLOT column, thermal program: isotherm, 50 °C, Ar carrier gas, TCD detector) to quantify the H2 production during the course of the experiment. Liquid aliquots of 0.50–0.75 mL were collected during the reaction and further analysed by gas chromatography (Shimadzu, GC-2010, column ZB-FFAP, thermal program: gradient 40 °C → 230 °C, 20 °C min−1, isothermal 230 °C, 10 min, N2 carrier gas, FID detector) to measure the concentration of alcohol and of the corresponding carbonyl product (2-octanone) in the reaction solution.
2.3.2. Analysis of the catalytic test results. Conversion of the substrate Xl is defined as:
image file: d0cy00390e-t4.tif
where Cx is the concentration of 2-octanol at a given time in liquid aliquots and Co is the concentration of 2-octanol at the beginning of the reaction.

Selectivity towards the corresponding ketone Sone is defined as:

image file: d0cy00390e-t5.tif
where C2-octonone is the concentration of 2-octanone at a given time in liquid aliquots, n is the stoichiometric coefficient and Cby-product is the concentration of the by-product(s) at a given time in liquid aliquots.

The reaction yield in 2-octanone Yone is given by:

Yone = X1·Sone
The yield in H2 YH2 was also evaluated as:
image file: d0cy00390e-t6.tif
where nH2 is the accumulated amount of H2 produced after a given time of reaction and ntheor is the theoretical amount of H2 possibly produced during the 2-octanol dehydrogenation reaction, based on the initial amount of alcohol.

The turnover number (TON) is calculated as:

image file: d0cy00390e-t7.tif
where nconverted substrate is the amount of 2-octanol converted into product(s) during the reaction, based on 2-octanol concentration at the beginning and end of the reaction and nsurface Co atoms is the amount of surface Co (1st layer of metal) in the catalyst sample used in the reaction. This number was evaluated taking into account that the exposed facets exhibit different amounts of Co atoms per surface unit (see the ESI, part 3, Table S1).

The established uncertainty of TON values is equal to ±25 molalcohol molsurface Co−1.

2.3.3. Catalyst pre-treatments. In an attempt to increase the catalyst activity, different pre-treatment procedures were applied before the reactions:
Thermal treatment under vacuum. Around 100 mg of a catalyst was placed into a glass cell, which subsequently was connected to the vacuum system. After reaching a high vacuum (<150 mPa), the temperature program was started (10 °C min−1, 120 °C, 12 h) to desorb the ligands from the catalyst.
Washing with ethanol of a dried sample. Around 100 mg of a dried catalyst were introduced into a centrifugation vial (5.0 mL volume) and 3.0 mL of ethanol was added. After few minutes of shaking, the sample was centrifugated (5000 rpm, 10 min) and the supernatant was collected. This washing was repeated 3 times. Then, the sample was dried for 24 h in an oven (80 °C, N2 atmosphere).
H2 in situ pre-treatment. 25 mg of a catalyst and 30.0 mL of decane were introduced into the reactor. All the reaction equipment was connected and the suspension was heated up to 145 °C. Once the target temperature was reached, the suspension was treated for 1 h with a mixture of flowing gases, 10 mL min−1 H2 + 30 mL min−1 inert gases, and for the next 1 h only with 30 mL min−1 of inert gases, to remove the remaining H2 from the reactor. After this, a solution of 2-octanol in decane (15 mL) was added into the reactor to reach a final volume of a reaction solution of a concentration of 0.95 mol L−1 of 2-octanol in decane and the reaction was started.
Additional washing during the synthesis of the catalysts. The chosen catalysts (P-R-2 and P-P) were re-synthesized and modified washing procedures before drying were applied. The P-R-2 sample was washed 6 times with 120 mL of methanol, instead of 3 times with 100 mL of ethanol. The P-P sample was washed 12 times with 120 mL of methanol, instead of 3 times with 120 mL of ethanol.
2.3.4. Post-reaction analysis. To check the cobalt leaching from the NPs to the solution during the reaction, the final reaction solutions were analyzed by ICP-OES (inductively coupled plasma-optical emission spectroscopy) using the ACTIVA Jobin Yvon apparatus. The detection limit of Co was 0.2 mg L−1.

2.4. Computational details

The Vienna Ab initio Simulation Package (VASP) computer program was used to perform periodic spin-polarized density functional theory (DFT) calculations.58 The potential and the exchange correlation were calculated with the generalized gradient approximation (GGA), using the PBE functional59 with dDsC dispersion correction.60,61 A cut-off energy of 400 eV was applied to obtain a tight convergence of the plane-wave expansion. The projector augmented wave (PAW)62,63 method was used to describe the electron–ion interactions. The SCF convergence criterium was set at 10−6 eV.

A p(3 × 3) cell for the (0001) surface and a p(4 × 4) for the (11−20) surface were considered using a four-layer slab and over 15 Å of vacuum. A dipole correction in the z direction was included. For the Brillouin zone integration, a Monkhorst–Pack mesh of 3 × 3 × 1 K-points was used.64 The two bottom layers of the slabs were kept fixed to the bulk-truncated positions (with a Co–Co interatomic distance of 2.47 Å, the experimental value being equal to 2.51 Å) while the two upper layers were allowed to relax. Adsorption and reaction processes were realized on the upper surface of the slab. The structures were allowed to relax until the forces were lower than 0.015 eV Å−1. 1.63 μB per atom was used as an initial magnetic moment value for Co and it turned to 1.58 μB and 1.78 μB for the (0001) and (11−20) surfaces, respectively. It oscillated slightly (±0.05 μB) along the reaction pathways for all the considered surfaces.

Frequencies were computed numerically within the harmonic approximation. Nudge elastic band (NEB)65,66 procedures together with the reaction path generator developed by P. Fleurat-Lessard, Opt'n Path,67 were used to determine the transition state structures. They were further optimized using the dimer method68,69 and confirmed by the presence of a single imaginary frequency whose normal mode corresponds to the reaction coordinate.

Gibbs energies are derived from the electronic energies within the perfect gas model, the rigid rotator and harmonic oscillator approximations for molecules and the lattice gas for adsorbates. In other words, for molecules in the gas phase, Gibbs free energy G is calculated as follow:

G = Eele + nkBT + ZPE − T × (St + Sr + Svib)
where Eele is the electronic energy; n = 4 for non-linear molecules and n = 3.5 for linear molecules; kB is the Boltzmann constant; T is the temperature, equal to 145 °C (418 K); ZPE is the zero-point energy; and St, Sr, Svib are the translational, rotational and vibrational entropy components, respectively.

Adsorbates are considered to lose their rotational and translational degrees of freedom and have a diffusion energy that is higher than the thermal energy. Hence, their Gibbs free energies are considered as follows:

Gslab = Eele + ZPE − T × Svib

Gads/slab = Eele + ZPE − T × Svib
where Gslab is the Gibbs free energy of a slab, Gads/slab is the Gibbs free energy of the species adsorbed on a slab and ZPE and Svib are based on the harmonic vibration of the adsorbate.

Frequencies lower than 50 cm−1 were neglected for all the entropy calculations.

The Gibbs free energy of adsorption (Gads) for a given state was calculated as the difference between the energy of a molecule adsorbed on the surface and that of a molecule in the gas phase and of the surface, depending on the case. A negative energy means a stabilizing adsorption.

For both surfaces, the presence of co-adsorbed CH3COO ligands during the reaction, used to mimic the presence of the longer chain carboxylate ligands present on NPs surfaces, was considered. On the 0001) surface, for p(3 × 3), the reaction was modeled in the presence of two ligands, which corresponds to the surface coverage of θ = 4/9 ML (0.44 ML, 6.98.10−6 mol m−2). For (11−20) facet, with p(4 × 4), the presence of three ligands on the surface was considered, which corresponds to the coverage of θ = 3/4 ML (0.75 ML, 7.26.10−6 mol m−2).

3. Results and discussion

In this work, the catalytic properties of unsupported Co NPs with different shapes were evaluated for the acceptor-less dehydrogenation of a secondary alcohol model (2-octanol) and their activity was correlated with the intrinsic properties of the samples. The particles were prepared by the polyol method, using long-chain carboxylate ligands, which are adsorbed onto the different crystal facets defining the particle morphology.29 Two main factors were considered as guiding the catalyst performance: type and percentage of exposed facets and surface coverage with ligands. On the one hand, if the reaction is shape-sensitive, then a relation between the morphology and the activity should be observed. On the other hand, the presence of ligands can lower the catalyst activity by limiting the number of accessible sites, or even deactivate the catalyst.

3.1. Preparation and characterization of the NPs

Three sets of unsupported Co catalysts with different morphologies, i.e. nanorods (R), diabolos (D) and platelets (P), were prepared.70 In the first series, Co nanoparticles of various shapes were decorated with a C12 laurate ligand (L). In the second series, NPs of different shapes were also prepared, but the carbon chain length was increased to 16 using a palmitate ligand (P). Finally, the shorter chains C7 heptanoate (H), C8 octanoate (O) and C10 decanoate (D) were used to prepare nanorods exclusively. The samples are named according to type of ligand – shape – order number (when necessary). The detailed synthesis procedures are reported in the ESI, and the characterization results are gathered in Table 1.
Table 1 Characterization of unsupported shaped Co nanoparticles obtained by the polyol synthesis method
Samplea Ligand Shape Exposed facets SSACb (m2 g−1) ΔmTGA-N2c (wt%) Surface coverage (mol m−2 × 10−6)
(0001) (%) {11−20} (%) {11−21} (%)
a Samples are named according to type of ligand – shape – order number (when necessary).b SSAC – calculated specific surface area of metal in a sample, based on TEM geometric measurements.c Weight loss according to TGA-N2 measurements, equal to the mass of organic ligands in the sample.
L-R-1 Laurate Nanorods 5 95 0 29 9.6 18
L-R-2 Laurate Nanorods 6 94 0 26 10.2 22
L-R-3 Laurate Short nanorods with large tips 35 20 45 39 28.4 51
L-R-4 Laurate Nanorods without tips 6 94 0 39 11.4 17
L-D Laurate Diabolos 38 0 62 27 10.9 23
L-P Laurate Platelets 78 22 0 36 12.0 19
P-R-1 Palmitate Long nanorods 5 95 0 27 41.2 102
P-R-2 Palmitate Short nanorods 8 92 0 35 31.9 52
P-R-3 Palmitate Very short nanorods 14 86 0 47 59.7 123
P-D Palmitate Diabolos 35 0 65 39 62.5 167
P-P Palmitate Platelets 67 33 0 28 44.4 112
H-R Heptanoate Nanorods 9.7
O-R Octanoate Nanorods 8 92 0 22 6.4 22
D-R Decanoate Nanorods 5 95 0 23 8.4 23


Representative TEM images of Co particles prepared with the laurate ligand as well as their corresponding geometrical models are given in Fig. 1 (see Fig. S1–S3 for TEM images of all the Co samples prepared in this work). The length of nanorods varies in the range of LTEM = 35–187 nm, while their diameter is in the range of dTEM = 13–26 nm. Diabolos show an inner diameter of dTEM = 17–21, outer diameter of DTEM = 25–46 nm, and length of 23–39 nm. For platelets, the diameter is equal to 56–57 nm, and their thickness is 8–12 nm. Three types of crystal facets are encountered in these particles: (0001) compact planes and open {11−20} and {11−21} facets, their relative proportions depending on the nanocrystal morphology. Typically, around 90% of the surface of Co rods correspond to {11−20} facets, the last 10% to (0001) planes, while the surface corresponding to {11−21} facets, if any, can be neglected in most cases (see Table 1). The distinction between rods and diabolos is not straightforward since diabolos can be seen as rods with extended hexagonal-based conical heads (named “tip” in the brief description in Table 1). Noticeably, the L-R-3 sample can be seen as an intermediate case, that is, considered as nanorods with large tips (see Fig. S1, ESI). Particles with a high proportion of {11−21} and (0001) facets are denoted diabolos, while the rods mainly exhibit {11−20} planes. For diabolos, two different types of facets are exposed, (0001) and {11−21}, corresponding to ca. 40% and 60% of the total surface of NPs, respectively. Platelets expose mainly close-packed (0001) facets, corresponding to ca. 70–80% of the NP surface, while the rest of the surface corresponds to {11−20} facets.


image file: d0cy00390e-f1.tif
Fig. 1 TEM images for Co NPs with different shapes, protected with laurate ligands and their respective geometrical models: (a) rods, L-R-1, (b) diabolos, L-D and (c) platelets, L-P. The scale bar stands for 200 nm. LTEM – mean length; DTEM – mean diameter of the diabolos tips and of the platelets; dTEM – mean diameter of the rods and of the central column in the diabolos; tTEM – mean thickness of the platelets.

The particles were further characterized by powder X-ray diffraction (see Fig. S5). Co rods crystallize with the pure hcp structure, while the native CoO phase can be hardly detected. It has previously been shown that this phase is generated when the particles are exposed to air; it is polycrystalline and has a thickness of about 1.5 nm.71 Diabolos and platelets crystallize also mainly with the hcp structure, but the fcc phase can also be detected, especially for the platelets.

Based on the geometric measurements from the TEM images, the specific surface areas of Co (SSAC) were calculated for each sample (for the details see the ESI, part 2.3.). The values are in the range of 22–47 m2 g−1 (see Table 1), the lowest being for O-R sample, and the highest for the P-R-3 sample. However, no tendency between the shape and the exposed surface area is visible. The specific surface area of the samples (SSABET of the metal NPs covered with ligands) was measured by N2 physisorption for the L-R-1 and L-R-2 samples. SSAs from the two measurements were in very good agreement for the first sample (SSAC and SSABET = 29 m2 g−1) but they differed for the second sample (SSAC = 26 m2 g−1 and SSABET = 32 m2 g−1), which implies that the SSA of exposed metal surfaces assessed by N2 physisorption cannot be determined confidently as it can be influenced by the presence of ligands on NP surfaces. In order to assess the exposed surface area of the metal, the geometric measurements based on TEM analysis should be more suitable.

Since carboxylate ligands are adsorbed onto the particle surface,72 TGA experiments were performed in order to determine the different amounts of organic matter for all the Co particles. The decomposition of carboxylate ligands occurs at about 300 °C and is associated with an endothermic peak (see Fig. S6, ESI). The weight loss is in the range of 6.4–62.5% depending on the nature of the Co catalyst (see Table 1). These data indicate that the particles display different thicknesses of ligand coverages. The ligand surface coverages were calculated based on the molar amount of ligands quantified by TGA and the exposed metal surface area based on TEM geometric measurements (see the ESI part, nanoparticle characterization, surface coverage). The calculated values are in the range of 17–167 × 10−6 mol m−2. Based on the area of the surface unit cell for a given facet (from the crystallographic cleavage of bulk metal) and the maximum number of ligands adsorbed (1 carboxylic ligand is considered to be adsorbed onto 2 surface Co atoms), one monolayer (ML) of ligands is equivalent to 16 × 10−6 mol m−2 on the close-packed (0001) surface, to 10 × 10−6 mol m−2 on (11−20) and to 18 × 10−6 mol m−2 on the (11−21) open-type facets. This clearly indicates that the organic protecting layer thickness corresponds to at least 1–2 ML (or 1 bilayer), and in some cases, it can reach up to 10 ML.

In brief, 14 different Co NPs were synthesized by the polyol method, using carboxylate ligands of increasing chain length (from C7 to C16). They expose three main types of facets ((0001), {11−20} and {11−21}) in various proportions depending on their shape as determined by TEM. Based on TGA analyses combined with the specific surface area derived from TEM analysis, the thickness of the organic layers used to stabilize these NPs ranges from 1 to 10 ML, depending on the samples. Their variety will allow further investigation of the influence of several factors on the catalytic activity: (i) type of capping ligand, (ii) thickness of the organic layer and (iii) structure sensitivity of the alcohol dehydrogenation reaction.

3.2. Catalytic activity of shaped Co NPs

All the decorated shaped Co nanoparticles were tested as catalysts in the 2-octanol dehydrogenation using a semi-batch reactor to shift the unfavorable reaction equilibrium towards the products. The corresponding results are gathered in Table 2. Fig. 2 illustrates typical results of a catalytic test using sample L-R-1: (a) conversion of 2-octanol, yield in H2 and yield in 2-octanone as a function of time, (b) intensity of H2 production during the course of the reaction. For all the catalytic tests, the two yields are in relatively good agreement within 6%, and the selectivity towards the production of the ketone and hydrogen is nearly quantitative, confirming the acceptor-less mechanism of the reaction. The H2 production increased gradually at the beginning of the reaction, reached a maximum after a few hours and then dropped gradually. The gradual increase at the beginning of the reaction signalized an induction period to fully activate the catalyst, while the large drop at rather low conversion (below 20%) can be related to a possible partial poisoning of the surface. A possible explanation of the induction period is the presence of a Co oxide surface layer that has to be reduced first to lead to the active catalyst. To verify this, the experiment was conducted with in situ H2 pre-treatment of the catalyst before the reaction. In this case, the H2 production started immediately and intensively, confirming the presence of a small passivation layer. This test also shows that this layer does not influence the overall performance of the catalyst (see Table 2, sample L-R-1a) and illustrates the reproducibility of the catalytic tests within the experimental error (conversion ±3%).
Table 2 Catalytic results after 24 h of reaction for 2-octanol dehydrogenation using Co shaped nanoparticles decorated with various carboxylate ligands. Reaction conditions: 2-octanol (0.95 mol L−1), decane, 145 °C, inert atm, mcatalyst = 25 mg, nalcohol[thin space (1/6-em)]:[thin space (1/6-em)]ncatalyst = 100
Sample X1 (%) Yone (%) YH2 (%) TON (molalcohol molsurface Co−1)
a Experiment performed with in situ H2 pre-treatment of catalyst. n.a. – not available due to a poor definition of the shape and distribution of the corresponding nanoparticles.
L-R-1 35 35 30 1140
L-R-1a 33 33 27 1080
L-R-2 32 32 27 1170
L-R-3 4 4 2 70
L-R-4 34 34 28 830
L-D 32 32 30 610
L-P 35 35 33 600
P-R-1 2 2 1 90
P-R-2 3 3 1 100
P-R-3 9 9 7 350
P-D 1 1 1 70
P-P 16 16 15 430
H-R 22 22 17 n.a.
O-R 23 23 19 890
D-R 25 25 23 970



image file: d0cy00390e-f2.tif
Fig. 2 Catalytic test results for 2-octanol dehydrogenation with L-R-1 catalyst. (a) Conversion of 2-octanol X1 (○), yield in hydrogen YH2 (black line), and yield in 2-octanone Yone (+) vs. time, (b) H2 production (black line) vs. time.

Within the set of 14 samples, all catalysts exhibit a quantitative selectivity toward 2-octanone, in line with what has been reported for several metal-supported heterogeneous catalysts regardless of the metal used.37,38,40,42,45,47,49,50,52–54 When the laurate was used as a ligand (first series of catalysts) the corresponding catalysts showed similar catalytic activity (conversion 32–35%) except short nanorods (L-R-3, conversion of 4%). In the second series, all Co nanoparticles protected by palmitate were poorly or not active towards 2-octanol dehydrogenation. Only platelets (P-P) and very short nanorods (P-R-3) gave 16% and 9% conversion, respectively. For the last series of catalysts, i.e. nanorods protected with carboxylates of different carbon chain length (C7, C8, C10), a similar 2-octanol conversion was observed, from 22% to 25%. This conversion is lower than the one obtained with the laurate nanorods (L-R-1, L-R-2 and L-R-4) or other shaped nanoparticles covered by laurate (L-D, L-P). In short, nanorods with laurate ligands demonstrated the best performance, giving conversions of over 10% higher than those of nanorods protected with ligands of other chain lengths.

It is worth noting that the highest conversion obtained in this work (35%) is lower than the one reported previously for Co nanorods decorated with laurate ligands and the same alcohol substrate (85%).54 This is easily explained by the use of solvent-free conditions and a higher temperature of 180 °C. To compare the performance of our catalysts with those reported in the literature is not straightforward, as the reaction conditions such as the temperature, the presence of a solvent, and metal-to-alcohol ratio can differ significantly. To partly overcome these discrepancies, the turnover number (TON) can be used to express the catalyst activity. It is easier to report TON values by considering the total amount of metal atoms since assessing the number of surface atoms can be tedious and challenging. In this case, the TON value for our best Co nanocatalyst in terms of conversion (L-R-1) is equal to 39 molalcohol moltotal Co−1 after 24 h of reaction. It remains of the same order of magnitude as that of noble Re NPs (TON of 50)53 and Co/TiO2 (78).52 Also, it is higher than that for Ag/Al2O3 (25),45 Cu/hydrotalcite (13)47 and Ru/AlO(OH) (22)37 for this peculiar substrate. To better analyse the performance of our catalysts in terms of morphology and exposed sites, the activity was expressed by TON, scaled using the number of Co exposed at the catalyst surface and measured after 24 h to limit the bias likely to be induced by the variable induction period. The TON values are ranging from 70 to 1170 molalcohol molsurface Co−1, with the lowest for L-R-3 and the highest for L-R-2 samples, respectively. With this analysis, no correlation between the catalytic activity and the morphology of samples was observed; hence the structure sensitivity was not revealed. See Fig. S7 for the graphs relating TON with the type and amount of exposed Co surfaces.

3.3. Importance of the ligands

Surface coverage with ligands is another important structural feature of the present unsupported shaped Co nanoparticles. When comparing this value (Table 1) and the catalytic activity of the nanoparticles (Table 2), it is easily noticed that a high coverage is detrimental to the catalytic performance. For instance, L-R-3 has a similar shape to L-R-1 or L-R-2, but it shows a much higher surface coverage (×3 times) and a much lower conversion in 2-octanol (4% vs. 32–35%). This is highlighted when plotting the conversion of 2-octanol as a function of the coverage (Fig. 3a) or, even better, the TON (Fig. 3b). Two groups of catalysts are evident. In the first one, the coverage is around 20 × 10−6 mol m−2, which is equivalent to 1–2 monolayers or 1 bilayer of ligands, and the conversion is above 20% and the corresponding TON is above 500 mol mol−1. In the second group, the coverage is above 50 × 10−6 mol m−2 and the catalytic activities reported are much lower (conversion below 20% and TON below 500 mol mol−1). Therefore, a too thick organic layer prevents the catalytic activity, probably by limiting the access of the substrate to the catalyst surface. This was already reported in the literature for the aerobic oxidation of benzyl alcohol with Au25 clusters supported on carbon.17 The catalysts uncoated with thiolate ligands showed higher activity than the coated ones and the catalytic activity decreased with increasing amount of thiolate on the metal surface. After the surface was fully covered, the activity was entirely suppressed. Such a behavior was also observed for Pd/Al2O3.73
image file: d0cy00390e-f3.tif
Fig. 3 (a) 2-Octanol conversion (X1, %) and (b) turnover number (TON, mol mol−1) as a function of the surface coverage. The points correspond to the Co nanoparticles (Tables 1 and 2): nanorods (□), diabolos (△), platelets (○). The color of the symbols represents the type of decorating ligand: palmitate (blue), laurate (red), decanoate (orange), octanoate (green).

In the attempts to decrease the amount of ligand of the poorly or non-active catalysts, several pre-treatments were conducted for two dried non-active catalysts (L-R-3 and P-R-2): (1) thermal treatment under vacuum, (2) washing with ethanol, (3) H2 in situ pre-treatment before the reaction (see the ESI). None of these pre-treatments improved the catalytic activity (Table 3). TGA-N2 performed after the treatments demonstrated that the first two strategies considered (thermal treatment under vacuum and washing of a dried sample with EtOH) failed at reducing the amount of ligands. The synthesis protocol was then modified in order to try to decrease the amount of ligands at the surface by washing more intensively the nanoparticles recovered after synthesis but before the drying step. This last pre-treatment (labeled #4 in Table 3) was performed on the P-R-2 and P-P samples, where the final washing was modified from 3 times with absolute ethanol to 6 times with methanol and from 3 times with absolute ethanol to 12 times with methanol, respectively. TEM was used to ensure that the shape was not modified after this treatment (Fig. S4). This more intensive washing decreased the residual amount of the palmitate ligands from 32 wt% to 12 wt% for the P-R-2 sample, and from 44 wt% to 16 wt% for the P-P sample (see Table 3), which corresponds to 1–2 monolayers of ligands on the surface of both samples. The corresponding catalytic performance improved – conversion increased from 3% (100 molalcohol molsurface Co−1 in TON) up to 19% (620 molalcohol molsurface Co−1) and from 16% (430 molalcohol molsurface Co−1) to 55% (990 molalcohol molsurface Co−1) for P-R-2 and P-P samples, respectively. This further confirms the importance of the amount of decorating ligands in controlling the catalytic activity of the shaped nanoparticles.

Table 3 Effect of different pre-treatments of Co NPs: (1) thermal treatment under vacuum, (2) washing of the dried sample with ethanol, (3) H2 in situ pre-treatment, (4) additional washing with methanol during the synthesis of the catalysts. Reaction conditions: 2-octanol (0.95 mol L−1), decane, 145 °C, 24 h, inert atm, mcatalyst = 25 mg, nalcohol[thin space (1/6-em)]:[thin space (1/6-em)]ncatalyst = 100. The amount of ligands was determined by TGA-N2 before and after the pre-treatments when possible
Sample Pre-treatment X1 (%) ΔmTGA-N2a (wt%) Surface coverage (mol mol−2 10−6) TON (molalcohol molsurface Co−1)
a Weight loss according to TGA-N2 measurements, equal to the mass of organic ligands in the sample. n.a. – not available.
L-R-3 4 28.4 51 70
1 1 30.9 62 0
P-R-2 3 31.9 52 100
1 1 32.1 53 10
2 1 30.4 49 0
3 4 n.a. n.a. 120
4 19 12.3 22 620
P-P 16 44.4 112 430
4 55 16.4 19 990


Noticeably, when the samples with a high coverage of ligands were used to perform the catalytic tests, the colour of the suspension evolved from colourless to light yellow/brown, which might be an indication of Co leaching. ICP-OES of the solutions after reaction (see Table S2) confirmed the presence of Co for all samples of the second group, while this leaching was under the detection limit in the case of an active catalyst (L-R-1). TGA analyses of the two spent catalysts showed that the amount of organic ligands protecting the samples decreased significantly after the reaction: from 32 wt% to 26 wt% for the inactive P-R-1 and from 10 wt% to 6 wt% for the active L-R-1. Even though the active catalyst was also losing some protecting ligands it can be assessed that a monolayer of ligands was still present on the metal surfaces, whereas the decrease of ligand content for the inactive catalyst was not large enough to improve its activity. To verify if the shape of the NPs was changing during the reaction, TEM analysis of two spent catalysts was performed. The corresponding images are presented in Fig. 4. The shape of L-R-1 and P-R-3 is not drastically modified. An aggregation and a bluntness of the tips were mainly observed. The major difference before and after the catalytic test is for the P-R-3 sample where a grey zone appeared (Fig. 4d, centre of the image), which we assigned to a Co(II) carboxylate phase (lamellar phase).29 It may originate from the leaching of Co NPs favoured by the excess of ligand or by the presence of some unreduced Co(II)-carboxylate precursors in the organic layers covering the Co NPs and can be related with the leached Co observed by ICP-OES of the solution.


image file: d0cy00390e-f4.tif
Fig. 4 TEM images of catalysts L-R-1 (a and b) and P-R-3 (c and d), before (a and c) and after the catalytic test (b and d). The scale bars stand for 200 nm.

3.4. Structure sensitivity

Among the active catalysts, namely those that are covered by roughly one to two monolayers or a bilayer of ligands, structure/activity relationships can be investigated by plotting the catalytic activity per exposed site (TON) either as a function of the total number of sites (Fig. 5a) or as a function of the number of a given site (Fig. 5b for (0001) facets and Fig. 5c for {11−20} and {11−21} surfaces). These plots do not include H-R since the surface area could not be established on that specific sample based on TEM. However, the P-R-2 and P-P samples that underwent extra washing before drying (pre-treatment #4, see Table 3) are included as their surface coverage is below 20 × 10−6 mol m−2.
image file: d0cy00390e-f5.tif
Fig. 5 TON (molalcohol molsurface Co−1) of the catalysts as a function of the number of surface exposed Co (10−6 mol) (a) all the surfaces are included (b) only the close-packed facet sites (0001) (c) only the opened facets sites ({11−20} and {11−21}). Each point corresponds to a catalytically active Co sample with ligand surface coverage of <20 × 10−6 mol m−2 (as found in Tables 1 and 2) for the samples decorated by laurate (red), decanoate (orange) and octanoate (green). The palmitate (blue) samples were obtained after extra washing (Table 3). The shape of the symbol corresponds to the shape of the NPs: nanorods (□), diabolos (△), platelets (○).

The TON is not constant as a function of the number of surface sites (Fig. 5a) since it ranges from 600 to 1170 molalcohol molsurface Co−1. This result suggests that since Co nanoparticles expose different types of sites in various extents, these sites must be differently active. It was then plotted as a function of the amount of exposed close-packed (0001) facets (Fig. 5b) and open-type facets (namely {11−20} and {11−21}, Fig. 5c). However, it is still not possible to observe a clear trend, even though the presence of ligands is limited to 1–2 monolayers. This is likely resulting from the complicated kinetics with variable induction periods (from a few minutes to hours) and a probable variable poisoning (with a drop in H2 production even at rather low conversion, see Fig. 2). As we already reported, DFT computations predict that the type of exposed facet can influence the catalytic activity of the metal, with the (11−20) facet being more active than the close-packed (0001) surface.54 The structure sensitivity of this reaction was also reported on Cu, where stepped Cu surfaces are found to be more active than the (111) facet for alcohol dehydrogenation both experimentally55 and computationally.56 However, in none of these reports was the presence of surface ligands taken into consideration.

3.5. Understanding the role of ligands – DFT computations

To rationalize the influence of the ligands, we determined the Gibbs free energy profiles of the alcohol dehydrogenation reaction with periodic DFT on the two mainly exposed facets, namely (11−20) and (0001), using isopropanol (iPrOH) as a model secondary alcohol. We compared the pristine surfaces with the ones decorated with model carboxylic ligands (CH3COO˙) with a similar coverage of roughly 7 × 10−6 mol m−2. More precisely, we took a coverage of 0.75 ML (7.26 × 10−6 mol m−2) in acetate ligands (labeled A) on Co(11−20) and of 0.44 ML (6.98 × 10−6 mol m−2) on Co(0001). In the following, these decorated surfaces are named 0.75A-Co(11−20) and 0.44A-Co(0001), respectively. These ligand surface concentrations are lower than the experimental ones but they were chosen to leave access to a few metal atoms for the reaction to proceed. This corresponds to a local defect in the layer of ligands that would generate a ‘catalytic pocket’ in analogy with enzymes.

Alcohol dehydrogenation is found to be slightly endergonic (0.07 eV), in close agreement with the experimental Gibbs reaction energy for isopropanol dehydrogenation (0.05 eV).74 This reaction requires the scission of two bonds, O–H and C–H, to generate acetone. Depending on the ordering of these two dissociations, two pathways can be distinguished: the alkoxy path and the hydroxyalkyl path. In the alkoxy path, the O–H bond breaking yields an alkoxy intermediate, from which the C–H bond scission takes place. The hydroxyalkyl path starts with the C–H bond dissociation, leading to the formation of a hydroxyalkyl intermediate, and continues with the O–H scission. In agreement with the literature,54,75–77 the alkoxy path is systematically preferred over the hydroxyalkyl one for the alcohol dehydrogenation on all the Co surfaces investigated by us. The corresponding Gibbs free energy profiles and schematic representations of intermediates and transition states are shown in Fig. 6 and S8 for the alkoxy and hydroxyalkyl pathways, respectively.


image file: d0cy00390e-f6.tif
Fig. 6 Gibbs free energy profiles (in eV) for iPrOH dehydrogenation via the alkoxy pathway on the Co(0001) facet (left) and on the Co(11−20) facet (right). Profiles with a solid line correspond to bare surfaces and a dashed line to surfaces decorated with CH3COO ligands. The reference energy is the isolated iPrOH and the isolated surface. When a decorated surface is considered, this surface is the optimal decorated surface at the chosen coverage. Schematic drawings represent path stages.

Focusing on the favored alkoxy path, we start comparing the predicted activities of the two bare surfaces, Co(0001) and Co(11−20), which will be used as reference later when moving to the surfaces covered with carboxylic ligands. The corresponding energy profiles for the bare surfaces are shown with solid lines in Fig. 6, towards the left and the right side, respectively. On both surfaces, the adsorption of iPrOH is not stabilized, which will result in short contact time between the alcohol and the catalyst. The O–H scission is clearly structure sensitive, with a 0.70 eV barrier on (0001) that considerably drops to 0.10 eV on (11−20). The structure sensitivity of the following C–H scission is less striking with barriers of 0.62 eV and 0.48 eV on (0001) and (11−20), respectively. On both facets, iPrO is by far the most stable intermediate (and hence the resting state of the catalytic cycle) and as expected, it is more strongly adsorbed on the open-type facet, but only by 0.12 eV. Similarly, hydrogen and acetone are adsorbed more strongly on this open surface (by 0.06 eV and 0.45 eV, respectively). Hence, their desorption is less demanding from (0001) than from (11−20). However, the eased desorption of the products is not enough to make (0001) the most catalytically active facet. With both O–H and C–H bond breakings that are facilitated, the sites exposed on the Co(11−20) facet are more efficient than the ones exposed on Co(0001). This is also clearly shown by the resulting low energy span of 1.13 eV on the open-type surface, to be compared with 1.49 eV on the compact one.§

Now, a key question arises: how does the presence of carboxylic ligands change this picture? We already said that the preference for the alkoxy path is maintained. The Gibbs free energy profiles of the alcohol dehydrogenation on the decorated facets are superimposed in dashed lines to the ones of the pristine surfaces in Fig. 6 for this alkoxy pathway. Surprisingly, even though the surface coverage with ligands is very similar, its presence globally destabilizes the reaction profile on (0001) while it affects the surface species in a contrasted manner on (11−20). Importantly, it results systematically in an energy span that is lower than for the surfaces without ligand decoration. Thanks to a stronger destabilization of the adsorption of the intermediate than of the OH scission transition state, the span falls from 1.49 eV to 1.24 eV on the (0001). With a destabilization of one of the intermediates (H) and the products, the span decreases also on the most active (11−20) facet, from 1.13 eV to 0.99 eV. It is worth noting that the reduction of the energy span is not related to lower activation energies of the elementary steps but to the (de)-stabilization of reactant and products on the catalyst surface. In short, the catalytic activity is enhanced by the presence of the surrounding ligands and is still higher on the (11−20) facet than on the (0001) facet, but with a less contrasted difference, thereby likely limiting the possibility to observe experimentally the structure sensitivity of alcohol dehydrogenation using decorated shaped nanoparticles.

Let us now analyze in more detail the impact of the ligands on a selected case. Strikingly, the adsorption of iPrOH is greater on 0.75A-Co(11−20) than on the corresponding bare surface by 0.71 eV, while it is almost not affected by the ligands on the 0.44A-Co(0001), as shown in Fig. 6. The origin can be tracked back through a decomposition into deformation and interaction terms of the variation in adsorption Gibbs free energies with and without ligands. This decomposition is schematically represented in Scheme 1 and the corresponding energies are provided in Table S4 in the ESI. On both surfaces, the interaction of the iPrOH with the catalytic pocket made up of the metal surface and the surrounding ligands is favored by the formation of a H-bond (see Fig. 7c and d), with interaction energy of −0.86 eV and −0.77 eV on 0.75A-Co(11−20) and on 0.44A-Co(0001), respectively. This stabilization is counterbalanced by an important deformation penalty (0.77 eV) for the ligands and the alcohol on the (0001) surface. This penalty can be related to the necessary change in the adsorption position of the ligands upon alcohol adsorption, moving from a parallel orientation to a less organized configuration (see Fig. 7b vs. d). On the other hand, the catalytic pocket is well adapted to adsorb the alcohol on the (11−20) with a very limited deformation cost (0.15 eV), resulting in an overall stabilization (the ligands preserve their configurations in the presence of alcohol; see Fig. 7a vs. c). The superior adaptability of the 0.75A-Co(11−20) catalytic pocket holds for all intermediates and transition states with a deformation cost that is systematically lower than for 0.44A-Co(0001). The highest deformation cost is to be paid for the C–H scission (>0.80 eV), with a limited interaction gain (<0.50 eV), resulting in a systematic destabilization of the corresponding transition state. This can be related to the increase in space required by this transition state compared with the preceding iPrO intermediate. To adapt, the carboxylate ligands even had to adjust again their adsorption positions, diffusing to another adsorption site. This analysis opens the road to improving the catalytic pocket, the choice of the ligands combined with the appropriate facet appearing as an important dimension of the catalyst design. It also points out the importance of co-adsorbates. In the case of the acceptor-less alcohol dehydrogenation reaction catalyzed by pristine Co, the alkoxy intermediate is likely to also play a critical role since it is particularly stabilized by the formation of three Co–O bonds.


image file: d0cy00390e-s1.tif
Scheme 1 Decomposition of the variation of Gibbs free energy of adsorption of iPrOH on Co induced by the presence of acetate ligands (ΔΔGads) into deformation (ΔGdef) and interaction (ΔGint). The corresponding values are reported in eV for the Co(0001) and Co(11−20) surfaces in blue and orange, respectively. Details are reported in the ESI (Table S4).

image file: d0cy00390e-f7.tif
Fig. 7 Optimized structures of (a) 0.75A-Co(11−20) surface, (b) 0.44A-Co(0001) surface, (c) iPrOH@0.75A-Co(11−20) and (d) iPrOH@0.44A-Co(0001). Main distances are provided in Å. Co is shown as cyan (light blue) balls. Acetate ligands and iPrOH are represented with sticks: C in blue, O in red, H in white.

4. Conclusion

The 2-octanol dehydrogenation was found to be a very clean process when catalyzed by decorated Co NPs, leading to the production of 2-octanone and H2 exclusively. These heterogeneous catalysts are not stabilized on a support but by an organic layer of carboxylate ligands attached to their surfaces. The ligands were used in the NP synthesis, by the polyol process, to direct their shape, and are further protecting the samples. These unconventional heterogeneous catalysts were found to be very stable under the reaction conditions with no leaching of Co into the solution and no change in shape (platelets, diabolos and rods). The only cases where leaching was observed corresponded to samples with low activity.

The catalytic performance was analyzed in regard to the intrinsic properties of shaped NPs, i.e. type and amount of exposed facets and the presence of protecting ligands. The ligand length was varied from C7 to C16 but was not influencing the activity. The thickness of the ligand layer appeared to be a much more important parameter. When a thick organic layer (roughly over 5 monolayers) was found on the NP surfaces, the catalytic activity dropped dramatically. This organic layer may contain some unreduced Co(II)-carboxylate precursors or favors the re-oxidation of Co since an important leaching of Co was also observed in the solution after the catalytic test. The extra intensive washing following the synthesis of these NPs allowed recovery of a good activity by limiting the organic layer thickness to 1–2 monolayers. This amount was systematically found on each active catalyst, whatever the shape and the ligand that were used. It appears to be sufficiently low to give access to the metallic surface for the reaction to proceed, but thick enough to prevent any agglomeration or strong shape modification during the catalytic tests.

Our computational investigations of this reaction on the two mainly exposed facets revealed that the open-type facet is expected to be much more active than the compact one. This structure sensitivity prediction, made on pristine surfaces, is not in agreement with the experimental observations where no clear correlation could be found between the catalytic activity expressed in TON and the amount of open-type facet sites despite the large range covered by our various shaped NPs. However, further computational analysis of decorated surfaces evidenced that the carboxylate ligands are limiting the predicted difference in activity between the two types of surfaces and thereby the predicted structure sensitivity. In addition, the number of accessible catalytic pockets can differ on one facet compared to the other, limiting the direct experimental comparison of the two types of surfaces. Interestingly, our DFT computations also revealed a strong participation of ligands in the reaction, in particular through H-bonding. This yields a predicted increase of the catalytic activity for a given metallic site when surrounded by carboxylate ligands, both on the compact and the open surface. The catalytic pocket generated by these ligands is particularly well adapted to the alcohol dehydrogenation on the (11−20) facet where the necessary deformations are limited. This improved understanding of the role of the ligand in the catalytic cycle opens the road to designing a bespoke catalytic pocket with an optimal choice of the ligands combined with the appropriate facet.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

The work was done as a part of the ANR TANOPOL project (ANR-15-CE07-0011-01). This work was performed using HPC resources from GENCI-CINES (Grant 2018-A0050800609) and from the PSMN Data Center, which was financially supported by the SYSPROD project and the AXELERA Pôle de Compétitivité.

Notes and references

  1. R. A. Van Santen, Acc. Chem. Res., 2009, 42, 57–66 CrossRef CAS PubMed.
  2. J. Chen, W. Fang, Q. Zhang, W. Deng and Y. Wang, Chem. – Asian J., 2014, 9, 2187–2196 CrossRef CAS PubMed.
  3. F. Pinna, Catal. Today, 1998, 41, 129–137 CrossRef CAS.
  4. Catalyst preparation: science and engineering, ed. J. R. Regalbuto, Taylor & Francis, Boca Raton, 2007 Search PubMed.
  5. M. B. Gawande, A. Goswami, F.-X. Felpin, T. Asefa, X. Huang, R. Silva, X. Zou, R. Zboril and R. S. Varma, Chem. Rev., 2016, 116, 3722–3811 CrossRef CAS PubMed.
  6. R. Xu, D. Wang, J. Zhang and Y. Li, Chem. – Asian J., 2006, 1, 888–893 CrossRef CAS PubMed.
  7. K. M. Bratlie, H. Lee, K. Komvopoulos, P. Yang and G. A. Somorjai, Nano Lett., 2007, 7, 3097–3101 CrossRef CAS PubMed.
  8. S.-H. Feng and G.-H. Li, in Modern Inorganic Synthetic Chemistry, Elsevier, 2nd edn, 2017, pp. 73–104 Search PubMed.
  9. B. Roldan Cuenya, Acc. Chem. Res., 2013, 46, 1682–1691 CrossRef CAS PubMed.
  10. D. I. Sharapa, D. E. Doronkin, F. Studt, J. Grunwaldt and S. Behrens, Adv. Mater., 2019, 31, 1807381 CrossRef PubMed.
  11. H. Lee, RSC Adv., 2014, 4, 41017–41027 RSC.
  12. L. Liu and A. Corma, Chem. Rev., 2018, 118, 4981–5079 CrossRef CAS PubMed.
  13. P. Losch, W. Huang, E. D. Goodman, C. J. Wrasman, A. Holm, A. R. Riscoe, J. A. Schwalbe and M. Cargnello, Nano Today, 2019, 24, 15–47 CrossRef CAS.
  14. T. S. Rodrigues, A. G. M. da Silva and P. H. C. Camargo, J. Mater. Chem. A, 2019, 7, 5857–5874 RSC.
  15. Z. Niu and Y. Li, Chem. Mater., 2014, 26, 72–83 CrossRef CAS.
  16. T. Chen and V. O. Rodionov, ACS Catal., 2016, 6, 4025–4033 CrossRef CAS.
  17. T. Yoskamtorn, S. Yamazoe, R. Takahata, J. Nishigaki, A. Thivasasith, J. Limtrakul and T. Tsukuda, ACS Catal., 2014, 4, 3696–3700 CrossRef CAS.
  18. J. Sá and J. W. Medlin, ChemCatChem, 2019, 11, 3355–3365 CrossRef.
  19. M. A. Ortuño and N. López, Catal. Sci. Technol., 2019, 9, 5173–5185 RSC.
  20. G. Kumar, E. Nikolla, S. Linic, J. W. Medlin and M. J. Janik, ACS Catal., 2018, 8, 3202–3208 CrossRef CAS.
  21. S. H. Pang, C. A. Schoenbaum, D. K. Schwartz and J. W. Medlin, Nat. Commun., 2013, 4, 1–6 Search PubMed.
  22. K. R. Kahsar, D. K. Schwartz and J. W. Medlin, J. Am. Chem. Soc., 2014, 136, 520–526 CrossRef CAS PubMed.
  23. D. González-Gálvez, P. Nolis, K. Philippot, B. Chaudret and P. W. N. M. van Leeuwen, ACS Catal., 2012, 2, 317–321 CrossRef.
  24. G. Kumar, C.-H. Lien, M. J. Janik and J. W. Medlin, ACS Catal., 2016, 6, 5086–5094 CrossRef CAS.
  25. M. A. Ortuño and N. López, ACS Catal., 2018, 8, 6138–6145 CrossRef.
  26. F. Fiévet, S. Ammar-Merah, R. Brayner, F. Chau, M. Giraud, F. Mammeri, J. Peron, J.-Y. Piquemal, L. Sicard and G. Viau, Chem. Soc. Rev., 2018, 47, 5187–5233 RSC.
  27. Y. Soumare, C. Garcia, T. Maurer, G. Chaboussant, F. Ott, F. Fiévet, J.-Y. Piquemal and G. Viau, Adv. Funct. Mater., 2009, 19, 1971–1977 CrossRef CAS.
  28. K. Mrad, F. Schoenstein, H. T. T. Nong, E. Anagnostopoulou, A. Viola, L. Mouton, S. Mercone, C. Ricolleau, N. Jouini, M. Abderraba, L.-M. Lacroix, G. Viau and J.-Y. Piquemal, CrystEngComm, 2017, 19, 3476–3484 RSC.
  29. K. A. Atmane, C. Michel, J.-Y. Piquemal, P. Sautet, P. Beaunier, M. Giraud, M. Sicard, S. Nowak, R. Losno and G. Viau, Nanoscale, 2014, 6, 2682–2692 RSC.
  30. R. K. Ramamoorthy, A. Viola, B. Grindi, J. Peron, C. Gatel, M. Hytch, R. Arenal, L. Sicard, M. Giraud, J.-Y. Piquemal and G. Viau, Nano Lett., 2019, 19, 9160–9169 CrossRef CAS PubMed.
  31. Q. Liu, X. Guo, Y. Li and W. Shen, Mater. Lett., 2009, 63, 1407–1409 CrossRef CAS.
  32. A. F. Dalebrook, W. Gan, M. Grasemann, S. Moret and G. Laurenczy, Chem. Commun., 2013, 49, 8735–8751 RSC.
  33. H. Liu, H.-R. Tan, E. S. Tok, S. Jaenicke and G.-K. Chuah, ChemCatChem, 2016, 8, 968–975 CrossRef CAS.
  34. K. Kon, S. M. A. Hakim Siddiki and K. Shimizu, J. Catal., 2013, 304, 63–71 CrossRef CAS.
  35. G. Nicolau, G. Tarantino and C. Hammond, ChemSusChem, 2019, 12, 4953–4961 CrossRef CAS.
  36. R. Karvembu and S. Priyarega, React. Kinet. Catal. Lett., 2006, 88, 333–338 CrossRef CAS.
  37. W.-H. Kim, I. S. Park and J. Park, Org. Lett., 2006, 8, 2543–2545 CrossRef CAS PubMed.
  38. K. Kon, W. Onodera, T. Toyao and K. Shimizu, Catal. Sci. Technol., 2016, 6, 5864–5870 RSC.
  39. Y. Sawama, K. Morita, T. Yamada, S. Nagata, Y. Yabe, Y. Monguchi and H. Sajiki, Green Chem., 2014, 16, 3439–3443 RSC.
  40. W. Fang, Q. Zhang, J. Chen, W. Deng and Y. Wang, Chem. Commun., 2010, 46, 1547–1549 RSC.
  41. W. Fang, J. Chen, Q. Zhang, W. Deng and Y. Wang, Chem. – Eur. J., 2011, 17, 1247–1256 CrossRef CAS.
  42. T. Mitsudome, Y. Mikami, H. Funai, T. Mizugaki, K. Jitsukawa and K. Kaneda, Angew. Chem., Int. Ed., 2008, 47, 138–141 CrossRef CAS PubMed.
  43. K. Shimizu and A. Satsuma, J. Jpn. Pet. Inst., 2011, 54, 347–360 CrossRef CAS.
  44. A. Bayat, M. Shakourian-Fard, N. Ehyaei and M. Mahmoodi Hashemi, RSC Adv., 2015, 5, 22503–22509 RSC.
  45. K. Shimizu, K. Sugino, K. Sawabe and A. Satsuma, Chem. – Eur. J., 2009, 15, 2341–2351 CrossRef CAS PubMed.
  46. J. M. Conesa, M. V. Morales, C. López-Olmos, I. Rodríguez-Ramos and A. Guerrero-Ruiz, Appl. Catal., A, 2019, 576, 54–64 CrossRef CAS.
  47. T. Mitsudome, Y. Mikami, K. Ebata, T. Mizugaki, K. Jitsukawa and K. Kaneda, Chem. Commun., 2008, 4804–4806 RSC.
  48. D. Damodara, R. Arundhathi and P. R. Likhar, Adv. Synth. Catal., 2014, 356, 189–198 CrossRef CAS.
  49. Y. Zhu, M. Shen, Y. Xia and M. Lu, Appl. Organomet. Chem., 2015, 29, 152–156 CrossRef CAS.
  50. K. Shimizu, K. Kon, K. Shimura and S. S. M. A. Hakim, J. Catal., 2013, 300, 242–250 CrossRef CAS.
  51. H. Chen, S. He, M. Xu, M. Wei, D. G. Evans and X. Duan, ACS Catal., 2017, 7, 2735–2743 CrossRef CAS.
  52. K. Shimizu, K. Kon, M. Seto, K. Shimura, H. Yamazaki and J. N. Kondo, Green Chem., 2013, 15, 418–424 RSC.
  53. J. Yi, J. T. Miller, D. Y. Zemlyanov, R. Zhang, P. J. Dietrich, F. H. Ribeiro, S. Suslov and M. M. Abu-Omar, Angew. Chem., Int. Ed., 2014, 53, 833–836 CrossRef CAS.
  54. A. Viola, J. Peron, K. Kazmierczak, M. Giraud, C. Michel, L. Sicard, N. Perret, P. Beaunier, M. Sicard, M. Besson and J.-Y. Piquemal, Catal. Sci. Technol., 2018, 8, 562–572 RSC.
  55. X. He, Y. Wang, X. Zhang, M. Dong, G. Wang, B. Zhang, Y. Niu, S. Yao, X. He and H. Liu, ACS Catal., 2019, 9, 2213–2221 CrossRef CAS.
  56. R. A. Hoyt, M. M. Montemore, E. C. H. Sykes and E. Kaxiras, J. Phys. Chem. C, 2018, 122, 21952–21962 CrossRef CAS.
  57. L. Lutterotti, S. Matthies and H.-R. Wenk, Newsletter of the CPD, 1999, vol. 21, pp. 14–15 Search PubMed.
  58. G. Kresse and J. Hafner, Phys. Rev. B, 1993, 47, 558–561 CrossRef CAS PubMed.
  59. J. P. Perdew, K. Burke and M. Ernzerhof, Phys. Rev. Lett., 1996, 77, 3865–3868 CrossRef CAS.
  60. S. N. Steinmann and C. Corminboeuf, J. Chem. Phys., 2011, 134, 044117 CrossRef.
  61. S. N. Steinmann and C. Corminboeuf, J. Chem. Theory Comput., 2011, 7, 3567–3577 CrossRef CAS.
  62. P. E. Blöchl, Phys. Rev. B, 1994, 50, 17953–17979 CrossRef PubMed.
  63. G. Kresse and D. Joubert, Phys. Rev. B, 1999, 59, 1758–1775 CrossRef CAS.
  64. H. J. Monkhorst and J. D. Pack, Phys. Rev. B, 1976, 13, 5188–5192 CrossRef.
  65. G. Henkelman, B. P. Uberuaga and H. Jónsson, J. Chem. Phys., 2000, 113, 9901–9904 CrossRef CAS.
  66. D. Sheppard, R. Terrell and G. Henkelman, J. Chem. Phys., 2008, 128, 134106 CrossRef PubMed.
  67. P. Fleurat-Lessard, A chemist view on reaction path determination, Available at http://pfleurat.free.fr/ReactionPath.php Search PubMed.
  68. G. Henkelman and H. Jónsson, J. Chem. Phys., 1999, 111, 7010–7022 CrossRef CAS.
  69. J. Kästner and P. Sherwood, J. Chem. Phys., 2008, 128, 014106 CrossRef PubMed.
  70. F. Fiévet, S. Ammar-Merah, R. Brayner, F. Chau, M. Giraud, F. Mammeri, J. Peron, J.-Y. Piquemal, L. Sicard and G. Viau, Chem. Soc. Rev., 2018, 47, 5187–5233 RSC.
  71. T. Maurer, F. Zighem, F. Ott, G. Chaboussant, G. André, Y. Soumare, J.-Y. Piquemal, G. Viau and C. Gatel, Phys. Rev. B, 2009, 80, 064427 CrossRef.
  72. K. Ait Atmane, F. Zighem, Y. Soumare, M. Ibrahim, R. Boubekri, T. Maurer, J. Margueritat, J.-Y. Piquemal, F. Ott, G. Chaboussant, F. Schoenstein, N. Jouini and G. Viau, J. Solid State Chem., 2013, 197, 297–303 CrossRef CAS.
  73. C. A. Schoenbaum, D. K. Schwartz and J. W. Medlin, Acc. Chem. Res., 2014, 47, 1438–1445 CrossRef CAS.
  74. E. Buckley and E. F. G. Herington, Trans. Faraday Soc., 1965, 61, 1618–1625 RSC.
  75. J. Zaffran, C. Michel, F. Delbecq and P. Sautet, Catal. Sci. Technol., 2016, 6, 6615–6624 RSC.
  76. W. Luo and A. Asthagiri, J. Phys. Chem. C, 2014, 118, 15274–15285 CrossRef CAS.
  77. J. E. Sutton and D. G. Vlachos, Ind. Eng. Chem. Res., 2015, 54, 4213–4225 CrossRef CAS.

Footnotes

Electronic supplementary information (ESI) available: Detailed synthesis protocols of cobalt nanoparticles, results of physicochemical characterization, additional data connected with catalytic activity and DFT data. See DOI: 10.1039/d0cy00390e
In our previous work,54 we focused on the influence of the type of facet on the control of the chemoselectivity (primary vs. secondary alcohol) and reported the iPrOH dehydrogenation on bare Co(0001) and Co(11−20) surfaces. We improved here the level of calculation to better describe the influence of the (polar) carboxylate ligands: increase of the vacuum size over the surface up to 15 Å, inclusion of a dipole correction in the z direction, inclusion of thermal effects (Gibbs free energies) for better comparison with experiments through the energy span model. These modifications do not change the overall picture of the activity of surfaces.
§ To apply the energy span model, the Gibbs free reaction energy should be athermic or exergonic. Here, we predict a slightly endergonic reaction that we took as athermic.

This journal is © The Royal Society of Chemistry 2020