E ﬀ ect of catalyst carbon supports on the oxygen reduction reaction in alkaline media: a comparative study

Some carbon materials commonly used as electronically-conductive supports for catalysts in fuel cell research are Carbon Black (CB), multi-walled carbon nanotubes (MWCNT), Graphene Oxide (GO) and reduced graphene oxide (rGO). Here we present a comparative study into the relative e ﬀ ects of each of these on the performance towards the oxygen reduction reaction (ORR) in alkaline media. For the purposes of comparing the supports, a simple Pt catalyst is used and the performance is evaluated via Koutecky – Levich analysis and direct measurement of peroxide by rotating ring-disk electrode (RRDE) to determine the number of electrons ( n ) transferred in the ORR. It is found that Pt/CB follows a quasi 4-electron mechanism due to that the ORR takes place mainly on the active Pt particles, whereas Pt/ MWCNT, Pt/GO and Pt/rGO exhibit a mixed behaviour between the two proposed mechanisms due to the higher activity of the graphene-derived supports towards the peroxide formation compared to CB. The e ﬀ ect of the oxide groups of GO and the metal impurities of MWCNT on the catalytic performance is also studied.


Introduction
Carbon has been widely used as material for catalyst support for several reasons, such as its abundance, potential low cost, availability in a variety of forms with different properties, excellent electronic conductivity, high surface area and relatively inert surface. 1 Among the different carbon supports carbon black (CB) is the most widespread and the most common commercial CB support is Vulcan XC-72, consisting of quasi-spherical carbon particles with diameters < 50 nm and polycrystalline structures of several layers with an interplanar separation of 0.35-0.38 nm, reaching a high surface area of around 250 m 2 g À1 . 2 Countless studies investigating CBsupported catalysts for the oxygen reduction reaction (ORR) in alkaline media exist, including many concerning platinum group metal (PGM) catalysts namely Pt/C, 3,4 PdO/C, 5 Pd-Ag/C 6 and even a complete integrated study investigating all the PGM (excluding Os) supported on CB: Pt/C, Pd/C, Rh/C, Ru/C and Ir/ C, showing Pd/C and Pt/C the best performance in this experiment. 7 CB has been also tested as carbon support in non-PGM catalysts, such as CoFeN/C, 8 Au/C, 9 Ag/C, [9][10][11] Co-Ppy/C, 12 FePc/ PANI/C 13 or MnOOH/C. 14 The behaviour of CB as a catalyst support depends strongly on its specic characteristics, as well as on the application of different preparatory treatments. For example, oxidation of CB provides surface groups on the carbon support that facilitate the deposition of Pt. 4 In another study, an impregnation/ reduction method using the weak reducing agent Na 2 S 2 O 4 followed by a thermal treatment at 700 C in Ar produced growth and then crystallisation of Pt particles on the carbon structure. 15 Another investigation reported that Vulcan XC-72 treated with citric acid induced the deposition of smaller Pt particles. 16 Regarding the physical properties, the specic surface area of the carbon support has been proven to be a crucial factor in the catalytic activity, 17 whereas the porosity of the CB support has also been shown to play a role in the activity of a Pt-Ru catalyst towards methanol oxidation. 18 Another carbon material increasingly nding application as a catalyst support are carbon nanotubes in either single-walled (SWCNT) or the most common multi-walled (MWCNT) forms. The latter have been shown to display excellent electrocatalytic activity towards ORR in alkaline media. 19 MWCNTs have been used as carbon supports for a variety of metal catalysts in a number of studies, including the demonstration that the ORR on Ag nanoparticles deposited on MWCNT showed high electrocatalytic activity towards the peroxide reduction. 20 Yu and Manthiram proved that MnNiCoO 4 nanoparticles deposited on N-doped MWCNTs have better catalytic selectivity towards ORR than Pt in alkaline media. 21 Similarly, experiments to evaluate the performance of a PdNi/MWCNT catalyst towards the methanol oxidation in alkaline media suggest that the electrocatalytic activity of this system is higher than Pd/MWCNT due to a decrease on the reaction overpotential. 22 Moreover, the catalytic performance of MWCNT can be modied by chemical functionalization of the nanotubes. Dumitru and co-workers demonstrated that the ORR in alkaline media proceeds by a two-electron mechanism when MWCNT are functionalized with citric acid, whereas it undergoes a four-electron mechanism when the nanotubes are treated with peroxymonosulfuric acid, due to an increase in the number of C-O groups. 23 Recent developments on catalyst supports have been made on graphene derivatives, driven by its unique properties such as very high electronic and thermal conductivity, extremely high surface area (theoretically 2630 m 2 g À1 ) and high mechanical strength. 24 Although graphene has been proposed as catalyst support itself, 25 it cannot be produced for large-scale by micromechanical exfoliation from graphite. 26 The present study focuses on two graphene derivatives: graphene oxide (GO) and reduced-graphene oxide (rGO), which can be easily obtained from abundant graphite crystal using well-known oxidation methods, 27 followed by a reduction in the case of rGO. 28 The main characteristic of GO with respect to graphene is the loss of electronic conductivity due to the partial transition from sp 2 to sp 3 carbon atoms. 29 Several techniques have been reported in the literature to characterise GO, such as X-ray photoelectron spectroscopy (XPS), 26 Fourier-transform infrared spectroscopy (FTIR), 30 X-ray diffraction (XRD), thermogravimetric analysis (TGA) 27 and cyclic voltammetry analysis of colloidal suspensions of GO. 31 The reduction of GO would recover part of the characteristic conductivity of graphene by partially restoring the pnetwork. 32 Moreover, the elimination of oxide groups would create vacancies along the sp 2 carbon structure which could favour the deposition and dispersion of the catalyst particles. 33 GO reduction can be achieved via thermal, electrochemical or chemical methods. 34 Of the latter, hydrazine is a selective reagent to reduce GO, 28 but has the disadvantages of toxicity and the aggregation of rGO layers leading to the increasing use of L-ascorbic acid as a more benign reductant. 35 To the best of our knowledge no integral studies comparing CB, MWCNT, GO and rGO as catalyst carbon supports for the ORR in alkaline media have been carried out to date. In this study we will rst characterise the fabricated catalyst-supports via TGA, X-ray uorescence (XRF), transmission electron microscopy (TEM), FTIR and XPS. Then, the different catalysts will be investigated in ex situ rotating-ring disk electrode (RRDE) experiments by deposition onto a glassy carbon (GC) electrode to study the different kinetic effects of the four carbon supports. AC impedance measurements have been also performed.

Preparation of the catalysts and reduction of GO
The synthesis of Pt/C catalysts by a chemical reduction process with sodium borohydride (NaBH 4 -CRP) followed the procedure of Fang and co-workers. 36 195.2 mg of carbon support were deposited into 200 mL of ultrapure water. Aer 30 minutes of sonication, 5 mL of 0.05 M K 2 PtCl 6 aqueous solution (121.5 mg of K 2 PtCl 6 in 5 mL ultrapure water) was added and stirred at 500 rpm for 3 hours. Next, KOH was added to adjust the pH at 8 and then 5 mL of an aqueous solution containing 94.6 mg of NaBH 4 (the amount was calculated to provide a NaBH 4 : Pt molar ratio of 10 : 1) was added to the solution. Aer stirring at 500 rpm overnight the solution was ltered, thoroughly washing the catalyst with ultrapure water. Finally, the catalyst was le to dry in the oven at 80 C overnight and then weighed.
Carbon black was Ensaco Conductive Carbon Black provided by TIMCAL Graphite & Carbon (a member of IMERYS). MWCNT was provided by Nanostructured & Amorphous Materials INC (Los Alamos, NM 87544, USA), having the next properties: purity: 95%, length: 10-20 mm, diameter: 20-40 nm, SSA > 360 m 2 g À1 . The preparation of rGO was partially based on the procedure described by Zhang and co-workers. 37 For this, 300 mg of commercial GO (Nanoinnova, ref.: GO.M.10.29-1) were deposited into 300 mL of ultrapure water together with 3 g of Lascorbic acid at room temperature (ca. 298 K). The mixture was le to stir overnight at 500 rpm and then ltered 3 times, washing with copious amount of ultrapure water between ltrations. The catalyst was then dried in the oven at 80 C overnight and nally weighed to check the amount of rGO obtained.

Characterisation methods
TGA was carried out using a TG 209 F1 Libra thermo-gravimetric analyser (TA Instruments). The samples were placed in an alumina crucible and then heated from 25 to 900 C with a heating rate of 10 C min À1 under a ow of air at 50 mL min À1 . The different catalysts were also imaged by a TEM using a JEOL 2100F at 200 kV, equipped with a FEG source and a Gatan Orius CCD camera in order to obtain images with 0.025 nm pixel resolution. XRF was performed using a Niton XL3T TYPE 980 portable XRF analyser + Helium GOLDD for light element analysis. An X-ray tube with a 50 kV Ag anode was used for precious metal spectra evaluation. Catalysts based on GO and rGO carbon supports were characterised by FTIR using a Perki-nElmer FT-IR Spectrum 100. XPS were obtained at the National EPSRC XPS Users' Service (NEXUS) at Newcastle University using a Thermo Scientic K-Alpha XPS instrument with a monochromatic Al Ka X-ray source.
For the ex situ characterisation of the catalysts rotating ringdisk voltammetry was performed using a Metrohm AutoLAB PGSTAT128N potentiostat in a Faraday cage. The reference electrode was an Ag/AgCl (3 mol dm À3 ) electrode (E 0 ¼ +0.210 V vs. SHE) and the counter electrode was a Pt mesh. The RRDE (E7R9 Series, Pine Instruments Inc., USA) was formed by a GC disk (5.61 mm diameter) and a Pt ring with an area of 0.1866 cm 2 , with a collection efficiency of 37%. Prior to each experiment the RRDE was thoroughly polished with consecutive alumina slurries of 1, 0.3 and 0.05 mm and then sonicated to remove any impurities. All the experiments were carried out in 0.1 M KOH solution made with ultrapure water (resistivity $ 18.2 MU cm, milli-Q Millipore), except for electrochemical surface area (ECSA) characterisation carried out in 0.1 M HClO 4 solution.
Several catalyst inks were prepared with the different catalysts. The ink was made by mixing 4 mg of the required catalyst with 0.8 mL of isopropyl alcohol (IPA), 3.18 mL of ultrapure water and 0.02 mL of the commercial I2 ionomer solution (Acta S.p.a.) to act as a binder, obtaining an ink with 1 mg mL À1 of catalyst concentration and 0.02% of I2 ionomer. This mixture was then sonicated for 1 hour and aer this a 5 mL aliquot was pipetted onto the GC disk. This amount was calculated from the optimal loading of 20 mg cm À2 , since for higher loadings the thickness of catalyst lm does not satisfy the mass transport characteristics of RRDE. 38 Aer the ink deposition the droplet was le to dry at room temperature for 30 minutes at 700 rpm as described in literature 39 in order to get an uniform layer. In the case of ECSA determination, the ink was formed by 5 mg of the catalyst, 1.6 mL of ultrapure water, 0.4 mL of IPA and 20 mL of 5% Naon solution (Fluka 117-Naon Solution in a mixture of water and aliphatic alcohols), with a nal concentration of 2.5 mg catalyst per mL (1 hour sonication). 10 mL of this ink were pipetted onto a GC electrode (5 mm diameter) and let dry for 30 min at 700 rpm.
The RRDE was then immersed in the alkaline solution and cycled between À1.2 and +1.0 V at 100 mV s À1 until a stable response was observed. The solution was then thoroughly purged with N 2 (BOC gases, O 2 free, 99.998% purity) and voltammograms were measured to determine the capacitive current and ohmic resistance of the solution. The solution was then bubbled with O 2 (BOC gases, N5, 99.999% purity) for 45 minutes until saturation. Linear sweep voltammograms (LSV) were then recorded at 25 mV s À1 between À0.4 and +1.0 V at the rotation speeds from 400 to 2400 rpm. The Pt ring voltage was xed at +0.47 V to ensure complete HO 2 À decomposition. 40 For the ECSA experiment, the solution was purged with N 2 for 30 minutes and cycled between À0.2 V and +0.8 V at 20 mV s À1 . The measurements were carried out at 293 AE 2 K. AC impedance measurements were also taken in potentiostatic mode at 0 V vs. the same Ag/AgCl (3 mol dm À3 ) ref. electrode in 0.1 M KOH solution purged with N 2 , with the frequency varying from 100 to 0.01 kHz. The volume of the catalyst solution deposited onto the GC electrode was in this case 6 mL for all the samples, with the rest of the procedure remaining the same as previously explained. The resistance of the layers were not found to depend on the dropcast volume (and hence thickness) in the range from 3 to 12 mL.

Characterisation of catalyst-supports
TGA analysis was conducted to conrm the actual amount of Pt deposited onto the carbon supports (see Fig. 1). CB and MWCNT display a similar pattern with no signicant amount of water and gradual combustion of carbon. For the CB the onset of carbon oxidation (T onset ) is 684 C compared to MWCNT where T onset is 484 C. The residual mass for MWCNT before Pt deposition is 4.5% due to impurities remaining from the fabrication process of nanotubes as reported elsewhere. 41 The behaviour of GO and rGO is different due to the presence of oxide groups. In GO there is an initial weight loss of 14.9% up to 150 C, due to evaporation of water trapped in the GO bulk, followed by a decrease of 27.1% up to 260 C due to decarboxylation processes. At this point a distortion appears in the graph due to "micro-explosive" decompositions previously reported in literature, 42 with nal combustion of carbon starting at 470 C. A similar pattern is observed in rGO, with differences in the degree of decarboxylation being due to the lower level of oxide species compared to GO. Table 1 shows the amount of Pt in the different catalysts calculated from the values of residual mass and the analysis by XRF of the metal residue observed in Fig. 1. The Pt loading can be considered the same for Pt/CB, Pt/MWCNT and Pt/rGO with an average value of (14.9 AE 1.1)%, whereas it is meaningfully higher for Pt/GO with a value of 19.1%, probably due to the effect of the oxide groups during the Pt deposition. Fig. 2 shows the TEM images of the different catalysts together with their corresponding particle size distribution. It is remarkable the homogeneity of the particle size in the Pt/CB catalyst, with 80% of the particles being in the range between 3-4 nm and a mean particle diameter (d p ) of 3.7 nm. On the contrary, Pt/MWCNT shows very high dispersion of particle size with particles up to 7 nm of diameter, with a value of d p of 4.5 nm. This observation highlights the importance of functionalization of nanotubes, for example by a chemical modication with H 2 SO 4 and/or HNO 3 , to create nucleation sites which favour a more homogenous Pt distribution. 43 The TEM images for Pt/GO and Pt/rGO show a homogeneous particle size in both catalysts with identical value of d p : 3.8 nm. Nevertheless, it can be observed a higher particle agglomeration effect in the case of Pt/GO with respect to Pt/rGO. This may be related to the vacancies created during the reduction of GO with L-ascorbic acid when the oxide groups leave the graphene structure. These have been reported to create binding sites to anchor the Pt precursor ions which are subsequently reduced to Pt particles, 44 reducing by this way the particle agglomeration. This good dispersion of Pt particles on rGO structures has been already reported in previous investigations. [45][46][47] As we have seen, the Pt particle diameters are in the range between 2 and 7 nm for all the catalysts. According to some investigations, this range is beyond the particle size that can inuence the catalytic behaviour, 48 especially regarding the mass activity 49 and the ORR performance. 50 FTIR spectra were also taken to evaluate the inuence of the reductive treatments on the oxide groups present in the GOderived carbon supports ( Fig. 3A and B). First, the attenuation of peaks in the characteristic spectra of oxygen-containing moieties (Table 2) in the case of rGO, Pt/GO and Pt/rGO indicates the extent of reduction of the support that has occurred. The peak at 1614 cm À1 in GO supports Montes-Morán and coworkers, 51 who assign this peak to cyclic ethers or C]C stretching related to carbon rings decorated with phenolic groups rather than to aromatic ring vibrations. Dékány and coworkers 52 discovered that a drying process modies the intensity of this peak, suggesting that it could be related to water present in the GO structure. Another study independently conrms that the peak in this region is related to vibrations of adsorbed water molecules. 53 In our case, this could explain why this peak does not appear in the samples that were subjected to a thorough dry procedure as part of the processes for GO reduction and/or Pt deposition.
XPS was used to quantify which groups were reduced during the procedure. According to the literature the different carbon groups contributing to the C 1s peak, namely oxygen-free C-C/ C]C bonds, hydroxyl C-OH (including phenols), epoxy C-O-C, carbonyl pC]O and carboxyl HO-C]O, were assigned to the corresponding binding energies 284.49, 285.86, 286.55, 287.54 and 288.94 eV, respectively. 54 It has been reported that epoxy and hydroxyl groups are normally found on the basal plane of the GO layer, while carboxyls are present on the edges and ketones can be found on both basal plane and edges. 55,56 Fig. 4 shows that the proportion of C-C/C]C bonds in the rGOcontaining samples increases with respect to GO due to the effect of L-ascorbic acid used in the GO reduction treatment. NaBH 4 used in the treatment for Pt deposition also increases the proportion of C-C/C]C bonds in Pt/GO with respect to GO due to collateral reduction of oxide groups, although in a lesser extent than L-ascorbic acid (see Table 3). With regard to the oxygen-carbon bonds, the epoxy groups are the main contributors to the oxide species in GO, followed by carbonyls. Nevertheless, the epoxy groups were the most susceptible to be reduced in presence of both L-ascorbic acid and sodium borohydride. It should be also mentioned that the peak at 293.3 eV present in Pt/GO and Pt/rGO is due to the presence of potassium species (K 2p 3/2 ) remaining from the K 2 PtCl 6 precursor used in the Pt deposition. 57

Electrochemical characterisation
In this section the behaviour of the catalyst-supports towards the ORR in alkaline media is studied. The ORR takes place in alkaline media according to the overall, 4-electron reaction (1): 58 However, the reaction can also proceed stepwise, via the following 2-electron processes: 59 followed by a further reduction of the peroxide generated: or a disproportionation: As it can be seen, the alternative 2-step pathway produces peroxide intermediates. This is a problem in alkaline media, as the lower working potential of the ORR electrode in this environment favours the desorption of HO 2 À , whereas in acidic conditions the high working potential avoids the peroxide desorption thus facilitating the further reduction to OH À . 60 The ORR takes place mainly through HO 2 À formation at graphite and most carbon surfaces, whereas it follows the 4-electron pathway in Pt catalysts. 58 As a result, in catalytic systems formed by Pt particles deposited on carbon supports it is expected to nd a mixed behaviour between the two proposed mechanisms, depending on the behaviour of the carbon support. In order to determine which mechanism is predominant the peroxide intermediate production should be monitored. This can be done with a RRDE, setting the ring potential at the peroxide diffusion-limiting value. This methodology has been successfully proven in former work carried out by Damjanovic and coworkers 61 and then developed by subsequent investigations. [62][63][64] In order to compare the ORR activity of each catalyst, voltammograms were measured at a voltage scan rate of 25 mV s À1 over the range from +0.4 to À1.0 V (vs. Ag/AgCl 3 M ref. electrode) in oxygen-saturated 0.1 M KOH solution using a RRDE modied with the catalyst, as described in Section 2.2. Fig. 5A shows the voltammograms of the different catalysts at a rotation speed of 1600 rpm. If we consider the ORR activity in terms of more positive onset potential it can be seen that this parameter is in good correlation with the values of ECSA calculated (Table 4). It is observed that the ECSA of Pt/GO and Pt/rGO is very low, which is attributed to the presence of oxide groups affecting negatively to the catalytic activity of the Pt particles, in spite of the good particle size distribution displayed in Fig. 2. A similar pattern is shown in terms of halfwave potential, except for the case of Pt/MWCNT and Pt/rGO which swap their positions with respect to the onset values. The behaviour of Pt/MWCNT at more negative potentials could be inuenced by the presence of metal impurities in its structure, as we will analyse later.
To investigate the dominant ORR mechanism the number of electrons n involved in the electrochemical reaction was determined via Koutecky-Levich (KL) analysis (eqn (5)): 65 where I is the measured current, i k is the kinetic current, i l is the diffusion-limited current, n is the number of electrons involved in the reaction, F is the Faraday constant, A is the electrode surface area (0.196 cm 2 ), D 0 is the diffusion coefficient (1.93 Â 10 À5 cm 2 s À1 ), 63 n is the kinematic viscosity of the solution (0.01009 cm 2 s À1 ), 63 C is the concentration of O 2 (saturated) in 0.1 M KOH alkaline solution (1.26 Â 10 À6 mol cm À3 at 25 C) 66 and u is the rotation rate of the electrode (in rad s À1 ). The KL plot for each catalyst at À0.9 V is displayed in Fig. 5B. The values of n calculated from the slope are 3.5 for Pt/ CB, 2.5 for Pt/MWCNT, 2.4 for Pt/GO and 2.5 for Pt/rGO. The same procedure was applied to other currents measured at different potentials in the range À0.9 to À0.4 V and shown in Fig. 5C. It can be seen that there is a slight decrease in the value of n as the potential becomes more negative for Pt/CB, Pt/ rGO and Pt/GO (in this case until À0.7 V only), whereas it slightly increases as the potential is more negative for Pt/ MWCNT. According to these results, the ORR seems to proceed by a direct 4-electron mechanism for Pt/CB while there is a mixed behaviour between the two proposed mechanisms for the rest of the catalysts.
To gain more information on the mechanism, a RRDE was used to directly measure the rate of peroxide formation. Here the ORR takes place at the GC disk (where the different catalysts were deposited) and the concentric Pt ring detects the peroxide produced. In Fig. 6 it is shown the RRDE plots for each catalyst, showing the measured disk current I D at the bottom and the measured ring current I R at the top.
From these graphs it can be calculated the number of electrons transferred in the electrochemical reaction (n) from the values of I R and I D using the eqn (6): 64  n ¼ where n is the number of electrons transferred, I D the current measured at the GC disk (I D (H 2 O) + I D (H 2 O 2 )), I R the current measured at the Pt ring (related to the oxidation of H 2 O 2 ) and N the collection efficiency which is a design parameter provided by the RRDE manufacturer (N ¼ 0.37). Fig. 7A shows the values of n calculated by this method versus the potential applied at 1600 rpm.
If these values are compared to those obtained by the KL equation it can be seen that the RRDE method provides higher values of n. For Pt/CB the ORR involved 3.8 electrons which is independent of the potential applied in the range between À0.4 and À1.0 V. This stability of n in that potential range was also shown by Pt/rGO with a constant value of 3.2. The value of n varied in the cases of Pt/GO and Pt/MWCNT, indicating that the ORR was more sensitive to changes in the potential, especially for the later. These values of n are closely related to the selectivity of the reaction which can be quantied by the fraction of peroxide intermediate produced in the reaction (X H 2 O 2 ) using the eqn (7): 38 The values of X H 2 O 2 versus the potential applied at 1600 rpm are plotted in Fig. 7B. The peroxide intermediate production is around 10% for Pt/CB, 35% for Pt/MWCNT, 40% for Pt/rGO and 55% for Pt/GO, which is in accordance with the catalytic behaviour showed by each catalyst in previous measurements.
The variation of the values of n and X H 2 O 2 with the applied voltage is signicant in the case of Pt/MWCNT. There could be two different explanations for this. First, this could be attributed to metal impurities remaining in the MWCNTs from the fabrication process. 67,68 In the present investigation, XRF analysis has been carried out in order to determine the nature of these metal impurities, 69 showing that Fe is by far the main contributor with 11 130 AE 90 ppm, followed in a lesser extent by Cd (270 AE 10 ppm), Ni (200 AE 20 ppm), In (130 AE 10 ppm), Zn  proposed that in some types of carbon nanotubes the number of electrons transferred during the ORR is function of the potential applied, increasing the value of n at more negative potentials. 74 Nevertheless, Pumera and Wang demonstrated that the catalytic activity of these supposedly "metal-free" materials is in fact due to metal impurities remaining from the process of fabrication. 73 We can also observe that Pt/GO undergoes a considerably lower catalytic activity in spite of its relatively higher Pt loading (around 19%) compared to the rest of the catalysts (around 15%). The reason of this poor performance of the Pt/GO catalyst should be obviously related to the intrinsic characteristics of the carbon support. Especially remarkable are the differences of catalytic activity in Pt/GO and Pt/rGO taking into account that they were synthesised from the same GO carbon support, but subjected to a previous reductive method in the case of rGO. XPS studies (see Fig. 4C and D) demonstrated that the oxide groups had been removed in a larger extent in Pt/rGO with respect to Pt/GO. Although it is commonly accepted that the electrical conductivity of GO can be improved by partially restoring the sp 2 structure through oxide group elimination (especially epoxy and hydroxyl species located on the basal plane), 28 this is not our case. AC impedance measurements carried out as described in Experimental section 2.2, show that    respectively, which conrms that the conductivity played just a secondary role on the catalytic activity observed. Considering the number of electrons involved in the ORR, the higher value of n calculated for Pt/CB suggests that the catalytic activity of the Pt dominates the ORR in Pt/CB, since it is well-known that the oxygen reduction follows a 4-electron mechanism on Pt surfaces in both alkaline and acid media, while the 2-electron mechanism is preferred on carbon surfaces. 58,76 For example, Mao et al.
reported that the ORR takes place in MWCNT by the 2-electron mechanism in alkaline media, 19 while Papakonstantinou et al. concluded that electrochemically reduced graphene oxide follows a more predominant 2e À pathway. 75 Likewise, the lower value of n for Pt on MWCNT, GO and rGO reveals a greater contribution of the carbon-mediated pathway and thus reects that these forms of carbon are more active towards the ORR than CB. In conclusion, in the supports where the carbon conguration is derived from the graphene structure (MWCNT, GO and rGO), the ORR seems to take place on both carbon and Pt particles, thus showing a mixed behaviour between the two proposed mechanisms.

Conclusions
An integral study on the behaviour towards the oxygen reduction of catalysts with different carbon supports has been carried out in alkaline media. TGA analysis demonstrated that the reduction of potassium hexachloroplatinate with sodium borohydride is an efficient method for Pt deposition, whereas XPS analysis proved that reduction of GO with L-ascorbic acid is an effective method to eliminate the characteristic oxide species present in GO, especially epoxy groups. It is also concluded that the Pt/CB catalyst performs a quasi 4-electron pathway for the oxygen reduction, whereas Pt/MWCNT, Pt/GO and Pt/rGO underwent a mixed behaviour between the 1-step and the 2step proposed mechanisms. Especially Pt/GO yielded a poorer performance compared to Pt/rGO, presumably due to the presence of carbonyl and carboxyl species on the edges hindering the oxygen reduction reaction. These results were obtained by indirect calculation using the Koutecky-Levich equation and conrmed by direct determination of peroxide intermediate formation with RRDE measurements.