Surface Structure Dependent Electrocatalytic Activity of Co3O4 Anchored on Graphene Sheets toward Oxygen Reduction Reaction

Catalytic activity is primarily a surface phenomenon, however, little is known about Co3O4 nanocrystals in terms of the relationship between the oxygen reduction reaction (ORR) catalytic activity and surface structure, especially when dispersed on a highly conducting support to improve the electrical conductivity and so to enhance the catalytic activity. Herein, we report a controllable synthesis of Co3O4 nanorods (NR), nanocubes (NC) and nano-octahedrons (OC) with the different exposed nanocrystalline surfaces ({110}, {100}, and {111}), uniformly anchored on graphene sheets, which has allowed us to investigate the effects of the surface structure on the ORR activity. Results show that the catalytically active sites for ORR should be the surface Co2+ ions, whereas the surface Co3+ ions catalyze CO oxidation, and the catalytic ability is closely related to the density of the catalytically active sites. These results underscore the importance of morphological control in the design of highly efficient ORR catalysts.

T he advent of nanotechnology and the ability to synthesize a marvelous panoply of nanocrystals have breathed a new life to the catalysis science [1][2][3][4][5] . The notion that catalysts are necessarily nanomaterials is rooted in the importance of surface in activating chemical bonds. Although there have been numerous reports on the catalytic activities of nanomaterials, detailed understanding of how surface structure affects catalytic performance is still lacking. There is therefore a need to systematically study the catalytic activity as a function of nanocrystalline morphology other than the size since the surface structure is tunable by varying the morphology. The prerequisite is the selective synthesis of differently shaped nanocrystal catalysts with uniform crystal surfaces, preferably dispersed on a supporting substrate.
The spinel type Co 3 O 4 , in which the Co 21 and Co 31 ions occupy the tetrahedral and octahedral sites, respectively 6 , is known to be a promising catalytic material [7][8][9][10][11][12] . It has been reported that different morphologies of Co 3 O 4 nanocrystals have a direct bearing on their catalytic activities for CO oxidation. For example, the {110} faces of Co 3 O 4 nanocrystals have a higher catalytic activity for CO oxidation than {100} and {111}, because of the more abundant catalytically active Co 31 sites on the former 12 . For the CH 4 combustion, however, the catalytic activity of the nanocrystalline surfaces was found to be in the order of {112} . {011} ? {001}, depending instead on the surface energy 13 . In the main, the catalytic activity of a given catalyst is therefore determined by the nature of adsorption/activation/desorption of the reactants and products on the catalytically active sites [12][13][14][15] .
The spinel-type Co 3 O 4 nanocrystals are also a potential alternate for the high cost Pt and its alloys to catalyze the oxygen reduction reaction (ORR), a critical reaction which underlies a battery of renewable-energy technologies such as fuel cell. To our knowledge, however, no study has been reported on the correlation between the shape and the ORR catalytic activities of Co 3 O 4 nanocrystals. Such a study requires anchoring the Co 3 O 4 nanocrystals onto a substrate, which is preferably conductive and thus can enhance the ORR activity and stabilize the catalyst system. As a relatively new class of carbon-based nanomaterials, graphene and carbon nanotube (CNT) have high electrical conductivity, large surface area, high mechanical strength, and structural flexibility, making them ideal substrates for supporting such nanocrystal catalysts. Indeed, graphene and CNT supported Co-based electro-catalysts have already been used for ORR with improved catalytic activity and stability [16][17][18][19] .
However, shape-controllable synthesis of Co 3 O 4 nanocrystals on graphene and CNT as composites is still an unmet challenge.
In this paper, we report the controllable synthesis of Co 3 O 4 nanorods, nanocubes and nano-octahedrons with difference exposed surfaces uniformly immobilized in situ on graphene sheets. This series of nanocrystals showed much enhanced ORR catalytic activity when dispersed on graphene. More significantly, the quantitative catalytic activity depends on the detailed nanocrystalline morphology and thus the surface structure of the nanocrystals, namely, {111} . {100} . {110}, pointing to the Co 21 ions as the ORR active sites.

Results
Shape-selective synthesis of Co 3 O 4 nanocrystals. Detailed procedures for the synthesis of Co 3 O 4 nanoparticles (NP), nanorods (NR), nanocubes (NC) and nano-octahedrons (OC) on the surface of reduced graphene oxides (RGO) have been given in the experimental section, and are here illustrated in Scheme 1. The crystalline phases of these nanocomposites were ascertained by XRD patterns ( Figure SI-1), with the help of the standard crystal structure of Co 3 O 4 (JCPDS 65-3103). Co 3 O 4 NP around 10 nm across were formed by thermally decomposing the precursors nucleated from the supersaturated metal bicarbonate solution accompanied by the slow release of CO 2 (Figure SI-2), as we reported previously [20][21][22] .
Presumably, through the Ostwald ripening process, the initially nucleated precursors were transformed into cobalt carbonate hydroxide (Co(CO 3 ) 0.5 OH) nanorods with a length of several hundred nanometers and a diameter of ,10 nm ( Figure SI-3) 21 . The subsequent calcination process caused a spontaneous transformation into the Co 3 O 4 NR with the overall morphology conserved. Figure 1A shows a typical low-magnification transmission electron microscopy (TEM) image of the synthesized Co 3 O 4 NR, densely distributed on the surface of RGO sheets. We further examined the crystallographic nature of the individual Co 3 O 4 NR through high resolution TEM (HRTEM) observations. Shown in Figure 1B Figure 1B, C). When the nanorod was titled to the [1][2][3][4][5][6][7][8][9][10] zone axis, the (004) side planes and the (222) crystal planes could also be clearly observed ( Figure 1D, E). On the basis of these results, the nanorod morphology can be approximately laid out as shown in Figure 1F: the nanorod assumes its axial direction along [110] and is bounded by the side planes of {001} and {1-10}. Similar Co 3 O 4 NR were prepared previously by the calcination of cobalt carbonate hydroxide nanorod precursors obtained by the precipitation of cobalt acetate and sodium carbonate in ethylene glycol, and used for low temperature catalytic CO oxidation 12 .
When H 2 O 2 was added, during the Ostwald ripening process, the cobalt carbonate hydroxide (Co(CO 3 ) 0.5 OH.0.11H 2 O) nanorods were gradually transformed into Co 3 O 4 nanocrystals, as seen from SEM images of the intermediates ( Figure SI-4). After complete transformation, well-defined Co 3 O 4 OC were uniformly dispersed on the surface of graphene sheets ( Figure 2A). To reveal the exposed surfaces, we refer to the TEM images in Figure 2. According to the HRTEM image of a single Co 3 O 4 OC particle taken along the [001] direction ( Figure 2B), the two-dimensional (2D) projection appears to be square (see the model octahedron in the bottom left inset in Figure 2B). By tilting the Co 3 O 4 OC to the [011] zone axis, the 2D projection of the corresponding TEM image ( Figure 2C) becomes diamond-like. These results are consistent with the octahedron morphology of the Co 3 O 4 nanocrystals anchored on the surface of RGO sheets with eight exposed {111} surfaces (see the model structure in Figure 2D).   Figure 3C).
The composition of the prepared cobalt oxides was investigated by XPS spectra, which are shown in Figure SI-5A. The Co 2p spectra all show a doublet consisting of a low energy band (Co 2p 3/2 at 780. 6 eV) and a high energy band (Co 2p 1/2 at 796.0 eV) for the Co 3 O 4 NP, NR, NC and OC, in agreement with the standard spectra of Co 3 O 4 23,24 . The energy difference between the peak of Co 2p 3/2 and 2p 1/2 splitting is approximately 15 eV, indicating the presence of both Co 21 /Co 31 species in the cobalt oxides samples [23][24][25] . The RGO contents in the composites were obtained from the TGA curves ( Figure SI-5B), and calculated based on the weight loss below 400uC. According to that analysis, the mass percentages of RGO are around 16   electrodes. The electrodes were interrogated by cyclic voltammetry (CV) in O 2 -and for reference, N 2 -, saturated 0.1 M KOH solutions, and the data are shown in Fig. 4A. The Co 3 O 4 NP (size ,10 nm) on the surface of RGO sheets exhibited very poor ORR activity: with an onset potential of ,20.25 V vs. Hg/HgO. The Co 3 O 4 NR/RGO hybrids with exposed surfaces dominated by {110} showed a much more positive ORR onset potential (,20.1 V), suggesting higher ORR catalytic activity than Co 3 O 4 NP/RGO. Remarkably, the Co 3 O 4 NC/RGO with the six exposed {100} surfaces and the Co 3 O 4 OC/RGO nanocomposites with the eight exposed {111} surfaces achieved even more positive onset potentials, e.g., ,20.06 V for Co 3 O 4 NC/RGO, and ,20.04 V for Co 3 O 4 OC/ RGO, closely approaching that of Pt/C, the gold standard for ORR catalysts.
The ORR kinetics of the Co 3 O 4 /RGO composites was investigated using the rotating-disk electrode (RDE) technique in O 2 -saturated 0.1 M KOH electrolyte. As can be seen from the LSV curves in Figure 4B, the ORR process is diffusion controlled when the potential is negative to 20.20 V, mixed diffusion kinetic controlled in the potential region from 20.20 to 20.10 V, and kinetic controlled in the potential range from 20.01 to 0 V. Unsupported Co 3 O 4 NP prepared by thermal decomposition of Co(CO 3 ) 1/2 OH precursors, as we reported previously 26  A series of rotating disk voltammograms of oxygen reduction are shown in Figure Figure 4C. We can see that the Co 3 O 4 OC/RGO composite electrode can catalyze the ORR via a 4 e process, in much the same way as a high-quality commercial Pt/C catalyst does, which is impressive for a non-Pt catalyst. However, the ORR electron transfer number for the Co 3 O 4 NR/RGO and Co 3 O 4 NC/ RGO composite catalysts were calculated to be ,3.5, suggesting incomplete reduction of oxygen, but still domination by the 4e process.
The ORR catalytic activity of the Co 3 O 4 /RGO hybrid catalysts can also be gleaned from the Tafel slopes at low and high overpotentials in O 2 -saturated 0.1 M KOH aqueous solution. The Tafel data are shown in Figure 4D. The E versus log (-J ) curves of the samples similarly show two Tafel slopes at low and high overpotentials, respectively, indicating a similar change in reaction mechanisms with the potential. The two slopes can be explained in term of the isotherms at two different O 2 coverages; i) the Temkin isotherm (high O 2 coverage) associated with an intermediate oxide coverage arising from ORR at low overpotential, whereby the first electron transfer step involving an adsorbed product such as OH 2 is the ratedetermining step; and ii) the Langmuir isotherm (low O 2 coverage) at high overpotential wherein significant oxide coverage ceases to exist, which is commonly the case when a two-electron transfer reaction is the rate-determining step. This is a characteristic feature of ORR on mixed valence spinel oxide, e.g., Co-based ORR catalysts 25,27 . As can be observed in Figure 4D,  Other important performance metric for an ORR catalysts include the tolerance of the commonly used fuel molecules and cycle stability, which are especially relevant for fuel cells such as direct methanol fuel cell. To examine the possible crossover effect for the catalytic performance, we measured the electrocatalytic selectivity of the Co 3 O 4 /RGO composite electrode against electro-oxidation of methanol molecules. The current density-time (J-T) chronoamperometric response profiles are shown in Figure 4E. For the commercial Pt/C electrode, a sharp decrease (,85%) in current density was observed upon methanol addition (20 vol%) into the O 2 -saturated 0.1 M KOH electrolyte. However, the amperometric responses of the Co 3 O 4 /RGO composite electrodes are strong and stable, showing a retention ratio of at least 80% after the addition of methanol. Such high selectivity of the Co 3 O 4 /RGO composite electrodes toward the ORR and the remarkably good tolerance to crossover effect can be attributed to the much lower ORR potential than required for oxidation of the fuel molecules 28 . Moreover, the Co 3 O 4 /RGO hybrid electrodes also exhibited excellent stability as measured by chronoamperometric measurements (Figure 4F). At a constant voltage of 20.40 V vs Hg/HgO, the ORR current density produced in the hybrid catalysts almost had no decay over 10000 s of continuous operation, whereas the commercial Pt/C catalyst exhibited ,22% decrease in current density. Thus, in comparison with the commercial Pt/C catalyst, our Co 3 O 4 /RGO composite electrodes are more insensitive to methanol molecules, thus more resistive to poisoning by the possible methanol crossover from the anode of a fuel cell, and are more stable under operating conditions.

Discussion
Co 3 O 4 has the normal-spinel structure Co 21 Co 2 31 O 4 , in which the Co 21 ion in the formula unit occupies the tetrahedral site, while the two Co 31 ions occupy the octahedral sites 6 , as shown in Figure 5A. Figure  of such metallic oxide particles with a fcc structure differ not only in the surface atomic density but also in the electronic structure, geometric bonding and chemical reactivity 29 . As a result, those planes have different surface energies, following the order of c{111} , c{100} , c{110}, which is closely parallel to the catalytic activities for CO and CH 4 oxidation 12,13,30,31 .
For catalyzing CO oxidation, the CO molecule interacts preferably with the surface Co 31 cations, which is the only favorable site for CO adsorption, as confirmed both theoretically 32 and experimentally 33,34 . The oxidation of the adsorbed CO then occurs by abstracting the surface oxygen that had been coordinated with the Co 31 cations. The partially reduced cobalt site, i.e., Co 21 cation with a neighboring oxygen vacancy, is re-oxidized by a gas-phase oxygen molecule back to the active Co 31 Figure 5B-D). This scenario has been proved by surface differential diffraction studies concluding that the Co 31 cations are present solely on the {110} plane 38,39 . Similarly, in our own experiment with the Co 3 O 4 NR/RGO composite catalyst, the initial transformation temperatures for CO oxidation is 60uC, considerably lower than that with Co 3 O 4 NC/RGO (100uC) and Co 3 O 4 OC/ RGO (120uC) (Figure SI-10). Although the catalytic activities of the Co 3 O 4 /RGO composites for CO oxidation are by no means optimized, our study suffices to conclude that the Co 3 O 4 NRs with the predominantly {110} exposed surfaces have higher catalytic activity for CO oxidation than the Co 3 O 4 NCs with the sole six {100} exposed surfaces and the OCs with the only eight {111} exposed surfaces, in excellent agreement with the literature reports 12,30,31 .
In sharp contrast, for ORR catalysis, the Co 3 O 4 OC enclosed by the eight {111} facets on the RGO sheets was found to exhibit the highest catalytic activity among the four Co 3 O 4 /RGO nanocomposite catalysts we have studied in the present work, followed by Co 3 O 4 NC/ RGO, and then Co 3 O 4 NR/RGO, with Co 3 O 4 NP/RGO being the least active ( Figure 4B). Surprisingly, this ORR catalytic activity order correlates very well with the surface Co 21 density order of   Figure SI-5B). The similar specific surface areas together with the similar amounts of catalysts used for the ORR testing exclude the possibility of the specific surface area being an important factor that determines the ORR catalytic activity. Thus, it is the exposed crystal planes of the Co 3 O 4 nanocrystals that play a vital role in determining the ORR catalytic activity.
Assuming that the adsorption/desorption process of O 2 on the catalytic active sites is involved in the rate-determined step of the ORR, the surface bonding of O 2 to the composite catalysts should be critical. In general, O 2 molecules with the bond length of 0.12 nm adsorbs on the catalytic active sites mainly via three modes (Griffths, Bridge, and Pauling) 40  Conceivably, when O 2 molecules are absorbed on the catalytically active sites via the Pauling mode, it is the surface Co 21 (3d 5 4s 2 ) cations rather than the surface Co 31 (3d 5 4s 1 ) cations which prefer to transfer electrons to the absorbed O 2 molecules to weaken and to assist breaking the O-O bond, meanwhile leaving themselves oxidized to Co 31 . This suggests that the surface Co 21 sites should be the catalytically active sites instead of Co 31 sites for ORR, and can naturally explain why the Co 3 O 4 nanocrystals with the predominant {111} and {110} exposed surfaces exhibited higher catalytic activity for ORR than that with the {110} exposed surfaces. Such explanation also applies to the observation that CoO exhibits better catalytic ORR performance than Co 3 O 4 16,41 . In addition, the density of Co 21 cations in {111} planes (4= ffiffi ffi 3 p a 2 ) is higher than that in {100} planes (2=a 2 ), resulting in catalytic activity of the Co 3 O 4 OC/RGO composite catalysts enclosed by the eight {111} facets than that of the Co 3 O 4 NC/ RGO composite catalyst surrounded by the six {100} exposed surfaces.
It is widely known that the RGO can serve as both a supporter for the catalyst dispersion and a conduction path for shuttling electrons involved in redox reactions. Here it is clearly the case as well, and this would indiscriminately enhance the catalytic activity of the different nanocomposite catalysts in our study. It is also possible, however, that due to the specific interaction of the different crystal faces with the RGO, the catalytic activity enhancement may be different for different nanocomposites. In particular, the interaction of the Co 21 -rich surface with the RGO may be more beneficial to the ORR catalysis. Indeed, some studies along thins line have appeared in recent years 16,17 . Nevertheless, this aspect still underlines the role of the nanocrystal surfaces in ORR catalysis and adds to source of inspiration for tuning nanocrystal morphologies for optimizing catalytic efficiency.
In sum, we have demonstrated the morphological control of Co 3 O 4 nanocrystals uniformly immobilized in situ on RGO sheets by judiciously choosing the oxidant and tuning the reaction conditions such as pH value. The resulting nanorods predominantly exposing the {110} surfaces, nanocubes surrounded by the six {100} facets, and nano-octahedrons enclosed by the eight {111} facets have allowed us to further investigate the crystal face effects on the ORR catalytic activity. We found that, while the surface Co 31 ions are the catalytically active sites for CO oxidation, and the surface Co 21 ions act as the catalytically active sites for ORR. Additionally, the density of the catalytically active sites on the surface is closely related to the catalytic activity. Accordingly, we have established that the catalytic activity for ORR of these crystalline facets decreases in the sequence of {111} . {100} ? {110}. This fundamental understanding shows that morphological control of metal oxide catalysts is a promising surface engineering strategy for the development of nanostructured catalysts in general and non-precious metal free nano-catalysts in particular for ORR in alkaline media. Synthesis of graphene oxide (GO) sheets. Graphene oxide sheets were synthesized from natural graphite by a modified Hummers method 42 . Briefly, 0.5 g of graphite (,325 mesh, Alfa Aesar) and NaNO 3 (0.5 g; Aldrich, .99%) were dispersed into concentrated H 2 SO 4 (20 mL; Fisher Scientific, 98%) with an ice bath. Under vigorous stirring, KMnO 4 (2.0 g; Riedel-de Haën, .99%) was then added gradually. After removing the ice bath, the mixture was stirred at room temperature for 24 h. As the reaction progressed, the mixture became pasty with a brownish color. Successively, 20 mL of H 2 O was slowly added to the pasty mixture while keeping the mixture in an ice bath, since the addition of water into the concentrated H 2 SO 4 medium will release a large amount of heat. After dilution with 40 mL of H 2 O, 5 mL of 30% H 2 O 2 (VMR) was added to the mixture, accompanied by bubbling and changing to brilliant yellow color. After continuously stirring for 2 h, the mixture was filtered and washed with DI water. Then, the products were dispersed in 10 wt% HCl aqueous solution, and washed with DI water again three times to remove impurity ions. Finally, the products were dispersed in DI water via ultrasonication, and then centrifuged at 7000 rpm for 1 h. The supernatant was collected as a GO aqueous solution with a concentration of ,1 mg mL 21 .

Methods
Controllable synthesis of differently shaped Co 3 O 4 nanocrystals on reduced graphene oxide (RGO) sheets. To start with, 20 mM of CoCl 2 ?6H 2 O and 40 mM of NaHCO 3 were dissolved into 90 mL of DI water and mixed with 10 mL of GO solution, and the mixture was flushed with gaseous CO 2 for 2 h forming Solution A. Solution A was stirred at room temperature for 12 h, and subsequently refluxed at 100uC for 10.0 h after adding in 0.1 mL N 2 H 4 , forming Solution B. The precipitates in Solution B were centrifuged and washed with DI water three times, and then freeze dried. Finally, the products were thermally treated at 400uC for 1 h in a N 2 atmosphere with a heating rate of 5uC/min, to form the Co 3 O 4 nanoparticles/RGO (Co 3 O 4 NP/RGO) composites. In parallel experiments, when Solution B was poured into a 70 mL capacity autoclave with Teflon liner and then hydrothermally treated at 100uC for 12 h before the thermal treatment process, the as-formed products were labeled as Co 3  respectively. X-ray photoelectron spectroscopy (XPS) was measured on a Perkin-Elmer model PHI 5600 XPS system with a resolution of 0.3-0.5 eV from a monochromated aluminum anode X-ray source with Ka radiation (1486.6 eV). The thermogravimetric analysis (TGA) was performed from 30 to 700uC on a TGA Q5000 (TA Instruments Ltd) at a heating rate of 5uC min 21 under an air flow of 25 mL min 21 . Brunauer-Emmett-Teller (BET) surface areas were measured on a Coulter SA 3100 surface area analyzer.
Catalytic measurements for oxygen reduction reaction (ORR). Electrochemical measurements were carried out by cyclic voltammetry (CV) on a CHI 660D electrochemical workstation. A conventional, three-electrode cell consisting of glassy carbon electrode (GCE) with an area of 0.125 cm 2 was used as the working electrode, Pt foil was employed as the counter electrode and Hg/HgO (1.0 M KOH) (MMO, 0.098 V vs. SHE) was used as the reference electrode. The working electrode was modified with a catalyst layer by dropping a suitable amount of catalyst ink on the GCE. The catalyst ink was prepared by ultrasonically dispersing 10 mg of the carbon supported catalysts in a 2.0 mL solution (1.9 mL of ethanol and 0.1 mL of 5 wt% Nafion solution) for 30 min to obtain a homogeneous solution. 10 mL of the dispersion was pipetted out and dropped onto a glassy carbon rotating disk electrode of 3 mm in diameter, which was then dried in air. CV experiments were conducted at room temperature in 0.1 M KOH solution saturated with nitrogen. For all of the experiments, stable voltammogram curves were recorded after scanning for 20 cycles in the potential region from 0 to 0.6 V in 0.1 M KOH solution. Polarization curves for the oxygen reduction reaction (ORR) were obtained in 0.1 M KOH solution using the rotating ring disk electrode (RRDE-3A). Before the RRDE study, the electrodes were cycled at 50 mV s 21 between 0 and 0.6 V until reproducible cyclic voltammograms were obtained. Normalized currents are given in terms of geometric weight (mA cm 22 ). The working electrode was scanned cathodically at a rate of 5 mV s 21 with varying rotating speed from 400 rpm to 2400 rpm. Koutecky-Levich plots (J 21 vs. v 21/2 ) were analyzed at various electrode potentials. The slopes of their best linear fit lines were used to calculate the number of electrons transferred (n) on the basis of the Koutecky-Levich equation: Where J is the measured current density, J k and J L are the kinetic-and diffusionlimiting current densities, v is the angular velocity, n is transferred electron number, F is the Faraday constant (96485 C mol 21 ), C o is the bulk concentration of O 2 (1. Catalytic measurements for CO oxidation. The catalytic activity toward CO oxidation was evaluated in a continuous flow reactor. In brief, the reaction gas, 5% CO in nitrogen (99.999%) (10 mL min 21 ) and air (99.999%) (40 mL min 21 ) was fed to a catalyst (22.5 mg) containing fixed-bed flow reactor made of glass with an inner diameter of 2.4 mm. Steady-state catalytic activity was measured at each chosen temperature, from room temperature to 200uC in a step of 20uC. The effluent gas was analyzed on-line by an on-stream gas chromatograph (Ramiin GC 2060) equipped with a TDX-01 column.