Received 13 July 2005; received in revised form 15 September 2005; accepted 22 September 2005 Available online 8 November 2005

Applied Catalysis B: Environmental 63 (2006) 137 149 www.elsevier.com/locate/apcatb The methanol oxidation reaction on platinum alloys with the first row transition metals The case of Pt Co and Ni alloy electrocatalysts for DMFCs: A short review Ermete Antolini a, *, Jose R.C. Salgado b,1, Ernesto R. Gonzalez b a Scuola di Scienza dei Materiali, Via 25 Aprile 22, 16016 Cogoleto, Genova, Italy b Instituto de Química de São Carlos, USP, C.P. 780, São Carlos, SP 13560-970, Brazil Received 13 July 2005; received in revised form 15 September 2005; accepted 22 September 2005 Available online 8 November 2005 Abstract In recent years there has been much activity in examining Pt alloys with first row transition metals as catalysts materials for DMFCs. In this work, the electrochemical oxidation of methanol on Pt Co and Ni alloy electrocatalysts is reviewed. The effect of the transition metal on the electrocatalytic activity of Pt Co and Ni for the methanol oxidation reaction (MOR) has been investigated both in half-cell and in direct methanol fuel cells. Conflicting results regarding the effect of the presence of Co(Ni) on the MOR are examined and the primary importance of the amount of non-precious metal in the catalyst is remarked. For low base metal contents, an enhancement of the onset potential for the MOR with increasing Co(Ni) amount in the catalyst is observed, whereas for high contents of the base metal, a drop of the MOR onset potential with increasing Co(Ni) is found. As well as the base metal content, an important role on the MOR activity of these catalysts has to be ascribed to the degree of alloying. # 2005 Elsevier B.V. All rights reserved. Keywords: Methanol oxidation; Platinum alloy catalysts; Nickel; Cobalt; Direct methanol fuel cell 1. Introduction The use of methanol as energy carrier and its direct electrochemical oxidation in direct methanol fuel cells (DMFCs) represents an important challenge for the polymer electrolyte fuel cell technology, since the complete system would be simpler without a reformer and reactant treatment steps. The use of methanol as fuel has several advantages in comparison to hydrogen: it is a cheap liquid fuel, easily handled, transported, and stored, and with a high theoretical energy density [1 3]. Althougha lot ofprogresshas beenmadeinthedevelopmentof DMFC, its performance is still limited by the poor kinetics of the anode reaction[3 5] and the crossover of methanol from the anode to the cathode side through the proton exchange membrane [6 8]. Methanol oxidation is a slow reaction that requires active multiple sites for the adsorption of methanol and the sites that * Corresponding author. Tel.: +39 0109162880; fax: +39 0109182368. 1 Present address: Instituto de Química, UnB, C.P. 4478, Brasilia, DF 70919-970, Brazil. can donate OH species for desorption of the adsorbed methanol residues [9]. Methanol oxidation has been extensively investigated since the early 1970 s with two main topics: identification of the reaction intermediates, poisoning species and products, and modification of Pt surface in order to achieve higher activity at lower potentials and better resistance to poisoning. The results have been reviewed by several authors [10 12]. The main reaction product is CO 2 [13], although significant amounts of formaldehyde [14,15], formic acid [13] and methyl formate [15,16] were also detected. Most studies conclude that the reaction can proceed according to multiple mechanisms. However, it is widely accepted that the most significant reactions are the adsorption of methanol and the oxidation of CO, according to this simplified reaction mechanism: CH 3 OH!ðCH 3 OHÞ ads (1) ðch 3 OHÞ ads!ðcoþ ads þ 4H þ þ 4e (2) ðcoþ ads þ H 2 O! CO 2 þ 2H þ þ 2e (3) 0926-3373/$ see front matter # 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.apcatb.2005.09.014

138 E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 Platinum is the most active metal for dissociative adsorption of methanol, but, as it is well-known, at room or moderate temperatures it is readily poisoned by carbon monoxide, a by product of methanol oxidation. To date, the remedy has been to use binary or ternary eletrocatalysts based on platinum, all containing ruthenium as the activity promoting component [17 22]. According to the bifunctional mechanism [23,24], the CO-poisoned platinum is regenerated via a surface reaction between CO- and O-type species associated with ruthenium to yield CO 2. According to the ligand model [12,23,25], instead, the change in Pt electronic properties induced by the presence of Ru rends Pt atoms more susceptible for OH adsorption [23] or even for dissociative adsorption of methanol [12]. But also when Pt Ru is used as anode electrocatalyst the power density of a DMFC is about a factor of 10 lower than that of a proton exchange membrane fuel cell operated on hydrogen if the same Pt loading is used. Therefore, a number of Ru-alternative elements, showing a co-catalytic activity for the anodic oxidation of methanol, if used either as platinum alloys or as adsorbate layers on platinum, have been investigated [26 33]. The problem of methanol crossover in DMFCs has been extensively studied [6 8,34,35]: methanol adsorbs on Pt sites in the cathode for the direct reaction between methanol and oxygen. The mixed potential, which results from the oxygen reduction reaction and the methanol oxidation occurring simultaneously, reduces the cell voltage, generates additional water and increases the required oxygen stoichiometric ratio. This problem could be solved either by using electrolytes with lower methanol permeability or by developing new cathode electrocatalysts with both higher methanol-tolerance and higher activity for the oxygen reduction reaction (ORR) than Pt. Higher methanoltolerance is reported in the literature for non-noble metal electrocatalysts based on chalcogenides [35 38] and macrocycles of transition metals [39,40]. These electrocatalysts have shown nearly the same activity for the ORR in the absence as well as in the presence of methanol. However in methanol-free electrolytes, these materials did not reach the catalytic activity of dispersed platinum. Developing a sufficiently selective and active electrocatalyst for the DMFC cathode remains one of the key tasks for further progress of this technology. The current direction is to test the activity for the oxygen reduction reaction in the presence of methanol of some Pt alloys with the first row transition metals which present a higher activity for the ORR than platinum in low temperature fuel cells operated on hydrogen, and use them as DMFC cathode electrocatalysts [41 45]. The improvement in the ORR electrocatalysis has been ascribed to different factors such as changes in the Pt Pt interatomic distance [46] and the surface area [47]. But the behaviour of binary alloys with respect to electrocatalysis can be better understood in terms of the electronic ligand effect and/or the geometric ensemble effect. To rationalise these effects it is necessary to know precisely the local concentration and arrangement of both components at the very surface (in contact with the reactants), and also in the sublayers which influence electronically the outer atoms [48]. The electronic effect of elements present in the sublayers is illustrated on PtNi (1 1 1) and Pt 3 Fe (1 1 1), which present a quasi-complete Pt surface layer (with more or less Ni or Fe in the sublayers) and strong modifications of their chemisorptive properties and electrocatalytic performances [48]. This behaviour was attributed to the electronic effect of intermetallic bonding of the alloying component-rich second layer with the top-most Pt atoms. The electrocatalytic behaviour of Pt alloys with increasing contents of the second element can be explained by the model of Toda et al. [49], based on an increase of d-electron vacancies of the thin Pt surface layer caused by the underlying alloy. The ensemble effects where the dilution of the active component with the catalytically inert metal changes the distribution of active sites, open different reaction pathways [50]. The dissociative chemisorption of methanol requires the existence of several adjacent Pt ensembles [51,52] and the presence of atoms of the second metal around Pt active sites could block methanol adsorption on Pt sites due to the dilution effect. Consequently, methanol oxidation on the binarycomponent electrocatalyst is suppressed. On the other hand, oxygen adsorption, which usually can be regarded as dissociative chemisorption, requires only two adjacent sites and is not affected by the presence of the second metal. Pt Ni and Co alloy catalysts have been proposed both as methanol-tolerant cathode material and anode material with improved MOR for DMFCs. The choice of Co and Ni to modify Pt electrocatalyst to improve the MOR is due to the lowering of the electronic binding energy in Pt by alloying with these metals, promoting the C H cleavage reaction at low potential. Moreover, the presence of cobalt or nickel oxides provides an oxygen source for CO oxidation at lower potentials. On the other hand, a higher methanol-tolerance is expected on Pt Co and Ni alloy catalysts than on Pt, ascribed to the dilution effect of Pt, hindering the methanol adsorption. Furthermore, these alloys present an improved activity for the oxygen reduction than Pt alone. On the basis of this discrepancy, we will attempt to outline the electrochemical activity for the MOR of Pt Ni and Co alloy catalysts. 2. Structural characterization of Pt Co and Ni alloys In the composition range from 0 to 50 at.% Co(Ni), Pt and Co(Ni) form a substitutional continuous solid solution and two ordered phases [53 56]. The dependence of the lattice parameter of the Pt Co and Ni bulk alloys on the alloy composition is reported in Fig. 1. In the region of 75 at.% Pt there are face-centered cubic (fcc) superlattices Pt 3 Co and Pt 3 Ni of the Cu 3 Au (LI 2 ) type. Regular termination of the bulk LI 2 structure normal to the three major zone axes produces a variety of surface compositions, from the pure Pt ((2 0 0) and (2 2 0) planes), 25 at.% Co(Ni) ((1 1 1) plane) to 50 at.% Co(Ni) ((1 0 0) and (1 1 0) planes) [57]. To better understand the relationship between the surface composition and the catalytic activity, it is very important to determine if surface segregation, i.e. enrichment of one element at the surface relative to the bulk, takes place during the preparation of these alloy catalysts. The details of segregation are still not completely understood, especially in the case of segregation in nanoparticles in which the characteristics may differ from those of the bulk. This is not surprising, considering

E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 139 Fig. 1. Dependence of the Pt Co and Ni lattice parameters [53 56] on the Co(Ni) content in the alloy. that nanoclusters represent a finite quantity of material, so there is no infinite source/sink of constituent atoms and hence material balance constraints become important [58]. Conflicting results regarding the surface segregation on Pt Co and Ni alloys are reported in literature. XPS data by Shukla et al. [45] indicated some surface enrichment of base metal in Pt Co/C (atomic ratio 0.84 versus 0.72 in the bulk) and Pt Ni/C (0.86 versus 0.64) prepared by alloying at high-temperature. Paulus et al. [59] found 70 at.% Pt on the surface of the Pt 3 Ni particles and 58 at.% Pt on the surface of the Pt 3 Co particles for the Pt 3 Ni(Co) catalyst by E-TEK. The value for Pt 3 Ni is very close to the bulk composition of 75 at.% Pt and indicates that no segregation has taken place, whereas in case of the Pt 3 Co a slight segregation of Co to the surface is observed. For the Pt Ni (1:1) catalyst by E-TEK, however, they found only about 20 at.% Pt on the surface Pt and for the Pt Co (1:1) catalyst by E-TEK 35 at.% Pt on the surface, indicating Ni(Co) segregation to the surface. Park et al. [60] observed enrichment of Pt at the surface of Pt Ni nanoparticles relative to the bulk. For instance, Pt Ni (1:1) had 53.4 at.% of Pt and 46.6 at.% Ni. It is known that, given a similar size, the metal having the lower heat of sublimation tends to surface segregate in binary alloys. The heats of vaporization of Pt and Ni are 509.6 and 370.3 kj/mol, respectively [61]. Therefore, an enrichment of Ni at the surface is expected. However, a strong surface enrichment in Pt was found by low-energy ion scattering (LEIS) in Pt Ni alloys [62,63]. This shows that the thermodynamic explanation fails to predict the enrichment behaviours. Further, Mukerjee and Moran-Lopez [64] used the electronic theory of d-band density of states of pure components to demonstrate that surface enrichment of Pt in Pt Ni should occur. Finally, to produce a different surface composition, Stamenkovic et al. [58] either annealed at 727 8C or mildly sputtered with a 0.5 kev beam of Ar + ions clean Pt 3 Co and Pt 3 Ni samples. The surface composition of alloy samples was determined by LEIS spectroscopy. The LEIS spectra taken after mild sputtering unambiguously showed that Ni(Co) are present in the outermost layer of the clean sputtered surface. Surface composition was estimated to be 75 at.% of Pt and 25 at.% of Ni(Co), equal to the bulk concentration of Pt 3 Co and Pt 3 Ni alloys. Conversely, The LEIS data showed that the first layer of a clean annealed Pt 3 Ni(Co) surface contains only Pt atoms, implying that the Pt-skin structure can also be created on a polycrystalline Pt 3 Ni(Co) alloy. The actual Pt Co and Ni alloy catalysts, particularly the carbon-supported catalysts, are formed by alloyed and nonalloyed Co(Ni) species. The degree of alloying depends on the preparation method of the catalyst. X-ray photoelectron spectroscopy (XPS) analysis on commercial carbon-supported Pt 3 Co and Pt 3 Ni electrocatalysts indicated the presence of PtO, CoO and NiO on the surface [65]. In the same way, XPS analysis on unsupported Pt Ni alloy nanoparticles indicated the presence of metallic Ni, NiO, Ni(OH) 2 and NiOOH [60]. Moreover, XPS data suggested that the amount of platinum oxide content in the carbon-supported Pt Co alloy electrocatalyst is lower than that in Pt and Pt Ni [45]. Finally, X-ray adsorption near-edge structure (XANES) analysis revealed that Pt 3 Co and Pt 3 Ni possess higher Pt d-band vacancies per atom relative to Pt [66]. 3. Preparation of Pt Co and Ni alloy catalysts Generally, the starting materials used in the preparation of unsupported Pt Co(Ni) alloys are Pt and Co(Ni) metals. Pure Pt and Co(Ni) metals can be alloyed by melting in an arc furnace under an inert atmosphere (Pt Ni) [67] or by sputtering (Pt Co and Ni) [49]. Unsupported nanosized Pt Ni catalysts were synthesized at room temperature by Park et al. [60] using a conventional reduction method of Pt and Ni precursors (H 2 PtCl 6 and NiCl 2 ) with NaBH 4. XRD data suggested a good alloy formation. The size of the alloy nanoparticles was approximately 3 4 nm. Zhang et al. [68] obtained Pt Co nanoparticles by a twomicroemulsion technique. The microemulsion system was composed of Triton X-100 as the surfactant, propanol-2 as a co-surfactant, and either the Pt Co precursor solution or a hydrazine solution dispersed in a continuous oil phase of cyclohexane. Platinum cobalt nanoparticles were formed upon contact between the precursor containing microemulsion droplets and the hydrazine containing microemulsion droplets. For all Pt Co compositions, a narrow distribution of the particle size centered around 2 nm was observed. Martz et al. [69] prepared different Pt/M/Pc and Pt/M/ complex catalysts (with M = Co, Ni, and Pc = phthalocyanine) in the composition Pt:M = 80:20 by an impregnation method. A commercially available platinum catalyst was impregnated with solutions of cobalt phthalocyanine (CoPc) and nickel phthalocyanine tetrasulphonic salt (NiPc). After the reaction, part of the catalyst was heat treated at 700 8C under a nitrogen atmosphere. The resulting catalysts were structurally and electrochemically characterized before (Pt/M/Pc) and after heat treatment (Pt/M/Complex). The Pt/M/Pc had an average particle size of about 3 nm, while the average size after heat treatment increased to about 7 nm. For that regarding carbon-supported alloys, the method of preparation of Pt Co/C and Pt Ni/C commonly used consists of the formation of carbon-supported platinum followed by the

140 E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 deposition of the second metal on Pt/C and alloying at hightemperatures. This thermal treatment at high-temperatures gives rise to an undesired metal particle growth, by sintering of platinum particles [70]. Using this method, Beard and Ross [71] prepared Pt Co/C catalysts in the atomic ratio 3:1 starting from commercial Pt/C in two ways. One way (series A) consisted in the preparation of an acidic (ph 2) Co(OH) 2 solution, followed by Pt/C addition into this solution. In the other way (series B), Pt/C was added into a basic (ph 11) solution of the cobalt precursor. Thermal treatments at 700, 900 and 1200 8C under inert atmosphere were performed on each catalyst. Following thermal treatment in series A the lattice parameter decreased with increasing heating temperature, indicative of alloy formation. In series B the lattice parameter decreased after heating, but to a lesser extent than in series A. The particle size for series A at each thermal treatment temperature was larger than the corresponding size in series B. The final particle size of the series A material treated at 1200 8C (12 nm) was about four times larger than that of the starting Pt catalyst. Shukla et al. [45] prepared Pt Co/C and Pt Ni/C with a Pt:Co(Ni) atomic ratio 1:1 nominal composition starting from 16 wt.% Pt/C, dispersed in distilled water. The ph of the solution was raised to 8 with dilute ammonium hydroxide. The required amount of Co[(NO 3 )] 2 or Ni[(NO 3 )] 2 salt solution was added to this solution. This was followed by the addition of dilute HCl until a ph of 5.5 was attained. The resulting powder was heat-treated at 900 8C in a nitrogen atmosphere for 1 h. Min et al. [72] prepared carbon-supported Pt Co and Pt Ni alloy catalysts starting from commercial Pt/C (10%) catalyst. Appropriate amounts of CoCl 2 and NiCl 2 solutions were added to Pt/C. The atomic ratio of Pt to Co(Ni) was 3:1. These catalysts were subjected to thermal treatment at 700, 900 or 1100 8C ina reducing atmosphere. XRD measurements indicated a decrease of lattice parameter, i.e. an increase in the degree of alloying, with increasing heating temperature. The particle size, obtained from both XRD and TEM measurements, increased with increasing thermal treatment temperature. Oliveira Neto et al. [73] prepared Pt Co/C with various Pt:Co atomic ratios in the range 9:1 1:9 by the following procedure: CoSO 4 was dissolved in a methanol/water solution containing a small amount of NH 4 OH. A commercial 20% Pt/C catalyst was added and the suspension was thermally treated at 1000 8C in a reducing atmosphere. The amount of CoSO 4 and Pt/C in the mixture were those corresponding to the desired final composition of Pt Co/C. Cyclic voltammetry was used to evaluate the platinum active area which decreased with increasing Co contents in the samples following an exponentially decay. This behaviour was interpreted as due to the covering of the active Pt sites by cobalt. Using the same method as described by Shukla et al. [45], Salgado et al. [74] prepared carbon-supported Pt Co alloy catalysts with Pt:Co atomic ratios 90:10, 85:15, 80:20 and 75:25. The degree of alloying increased with increasing Co content in the catalyst. Conversely, the metal particle size decreased with increasing Co content in the catalyst. Finally, Sirk et al. [75] synthesized carbon-supported Pt Co by mixing a Co oxide sol precursor with Pt/C, followed by heat treatment at 700 or 900 8C. Recently, methods to synthesize carbon-supported Pt Co and Ni catalysts at low temperature, to avoid metal particle sintering, have been developed. Xiong et al. [76] prepared Pt Co(Ni) alloy catalysts on a high surface area carbon support by reducing a mixture of chloroplatinic acid and the respective metal salt solution with sodium formate in aqueous medium. Typically, the reduction reaction was carried out at 70 8C. In the case of Co, the reduction was also carried out by adding first a few drops of sodium borohydride followed by further reduction with sodium formate. The particle size was 3.6 and 4.5 nm, without and with sodium borohydride, respectively. Xiong and Manthiram [77] synthesised a highly dispersed Pt Co alloy catalyst on a carbon support in the nominal Pt:Co atomic ratio 80:20 by the microemulsion method, using sodium bis(2- ethylhexyl)sulphosuccinate as the surfactant, heptane as the oil phase and NaBH 4 as the reducing agent. The synthesis occurred at room temperature. By XRD analysis the samples prepared by the microemulsion method showed broad reflections compared to those obtained by the high-temperature route, indicating a smaller particle size for the former. The reflections of the Pt Co samples shifted to higher angles compared to that of Pt, indicating a contraction of the lattice and alloy formation. However, the shift was more significant for the samples prepared by the high-temperature route compared to those prepared by the microemulsion method, suggesting a greater extend of alloy formation in the former case. Deivaraj et al. [78] synthesised carbon-supported Pt Ni by hydrazine reduction of Pt and Ni precursors under different conditions, namely by heating at 60 8C, by prolonged reaction (12 h) at room temperature and by microwave-assisted reduction. The particle size of Pt Ni prepared by microwave-assisted reduction was the lowest, in the range 2.9 5.6 nm, while the particle size of Pt Ni prepared by thermal treatment at 60 8C and by prolonged reaction at room temperature were in the ranges 12.5 50 and 13 25 nm, respectively. Yang et al. [79] used the carbonyl chemical route to prepare carbon-supported Pt Ni. Pt and Ni carbonyl complexes were synthesized simultaneously using methanol as solvent through the reaction of Pt and Ni salts with CO at about 55 8C for 24 h. After the synthesis of Pt Ni carbonyl complexes, Vulcan XC-72 carbon was added to the mixture under a N 2 gas flow and stirred for more than 6 h at about 55 8C. Subsequently, the solvent was removed and the catalyst powder was subjected to heat treatment at different temperatures under nitrogen and hydrogen, respectively. The alloying temperature under hydrogen ranged from 200 to 500 8C. According to the authors, the nearly linear relationship between the lattice parameter and the EDX composition again attests that Ni is completely alloyed with Pt. Furthermore, the metal particle size decreases with increasing the content of nonprecious metal in the alloy. Finally, carbon-supported Pt Co [80,81] and Pt Ni [82] alloy electrocatalysts were prepared by impregnating high surface area carbon with Pt and Co(Ni) precursors, followed by reduction of the precursors with NaBH 4 at room temperature. The metal particle size was in the range 3.8 4.8 nm. It has to be remarked that, independently of the EDX composition, the actual composition of the alloy was around 92:8. As a consequence, for low Co(Ni) content

E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 141 (10 15 at.%) a high degree of alloying was attained, while the degree of alloying was low for the catalyst with high content (30 at.%) of the non-precious metal. 4. Oxygen reduction reaction and stability of Pt Co and Ni in PAFC and PEMFC environment The search for catalysts for the oxygen reduction reaction (ORR) that are more active, less expensive and with greater stability than Pt has resulted in the development of Pt alloys. It has been reported that alloying platinum with transition metals enhances the electrocatalytic activity for the ORR. This enhancement has been ascribed to different factors such as geometric factors (decrease of the Pt Pt bond distance) [83], dissolution of the more oxidisable alloying component [84], change in surface structure [71] or electronic factors (increase of Pt d-electron vacancy) [49]. Considering the use of Pt Co and Ni as methanol resistant cathode materials in low temperature fuel cells, the ORR activity of these catalysts will be briefly discussed. Mukerjee and Srinivasan [66] investigated the electrocatalysis of the ORR on five carbon-supported binary Pt alloys (PtCr/C, PtMn/C, PtFe/C, PtCo/C and PtNi/C) in proton exchange membrane fuel cells (PEMFC). All five binary alloy catalyst showed a two three folds activity enhancement in terms of the electrode kinetic parameters obtained from halfcell data, as compared to that on Pt. According to the authors, the enhanced ORR activity by the alloys was rationalised on the basis of the interplay between the electronic and geometric factors on one hand and their effect on the chemisorption behaviour of OH species from the electrolyte. Toda et al. [49] studied the ORR activity in perchloric acid solution of bulk Pt alloys with Ni, Co and Fe at room temperature. Maximum activity was observed at ca. 30, 40 and 50% content of Ni, Co and Fe, respectively, observing 10, 15 and 20 times larger kinetic current densities than that on pure Pt. By X-ray photoelectron spectroscopy (XPS) measurements they found that Ni, Co or Fe disappeared from all the alloy surface layers and the active surfaces were covered by a Pt-skin of a few monolayers. The authors proposed the modification of the electronic structure of the Pt-skin layer originating from that of the bulk alloys. More recently, the temperature dependence of the ORR activity on the same bulk alloy catalysts in 0.1 HClO 4 solution in the temperature range 20 90 8C was investigated by the same research group [85]. They found that from 20 to 50 8C the apparent rate constants k app for the ORR on Pt M electrodes were 2.4 4 times larger than that on a pure Pt electrode. The k app values at the alloy electrodes decreased by elevating the temperature above 60 8C, and settled to almost the same values observed on the Pt electrode. Stamenkovic et al. [58,86] studied the intrinsic catalytic activity of Pt 3 Ni and Pt 3 Co bulk alloy catalysts for the ORR with particular emphasis on the description of alloy surface preparation. They demonstrated that the ability to make a controlled and well-characterized arrangement of two elements in the electrode surface region is essential to interpreting the kinetic results. They observed that in 0.1 mol L 1 HClO 4 at 60 8C, the Pt-skin structure is more active than both pure Pt and Pt 3 Co, suggesting that a uniform monatomic layer of Pt surface atoms, with Pt depletion and Co enrichment in the second layer, has unique catalytic properties. According to the authors, these results show that the kinetics of the ORR is dependent not only on the nature of alloying component (Pt < Pt 3 Ni < Pt 3 Co) but also on the exact arrangement of the alloying element in the surface region (Pt bulk < Pt 3 Co < Pt-skin on Pt 3 Co). They proposed, in agreement with Toda et al. [49], that the catalytic improvement on the Pt-skin is caused by electronically modified Pt atoms on top of the Co-enriched layer. The enhancement of the catalytic activity for the ORR on Pt 3 Ni and Pt 3 Co alloy surface was ascribed to the inhibition of Pt OH ad formation on Pt sites surrounded by oxide -covered Ni and Co atoms. In a study on the ORR activity of carbon-supported Pt Co alloy with Pt:Co atomic ratio 55:45 under phosphoric acid fuel cell (PAFC) conditions, Watanabe et al. [87] observed higher activity on the alloys than on Pt. They found that the ordered Pt Co structure presents 1.35 times higher mass activity compared to the disordered alloy. Moreover, they demonstrated that both Pt and Co dissolve out from small-size alloy particle and Pt redeposits on the surface of large-size ones in hot H 3 PO 4. The observed decay in the performance of the alloy catalysts was then explained by the leaching of the alloying non-precious metal to the electrolyte. The alloy with a disordered crystallite structure, which is more corrosion-resistant than an ordered one, maintains higher electrocatalytic activity for a longer time. Regarding the stability of Pt Co alloy catalysts in PAFC conditions, it has to be pointed out that Beard and Ross [71],as previously reported, found an opposite result. Xiong and Manthiram [88] investigated the electrocatalytic activity of carbon-supported Pt Co in PEMFCs in a wide range of compositions (27 77 at.%). They found that alloys with ordered Pt 3 Co or PtCo structures have higher ORR activity than Pt or disordered Pt Co alloys. The same authors studied the effect of atomic ordering on the ORR activity of carbonsupported Pt M (M = Fe, Co, Ni and Cu, Pt:M 80:20 wt.%, ca. 55:45 at.%) [77]. Evaluation of the Pt M alloy catalysts for oxygen reduction in proton exchange membrane fuel cells indicates that the alloys with the ordered structures have higher catalytic activity with lower polarization losses than Pt and the disordered Pt M alloys. According to the authors, the enhanced catalytic activity is explained on the basis of optimal structural and electronic features, like the number of Pt and M nearest neighbors, d-electron density in Pt, atomic configuration on the surface, and Pt Pt distance. Recently many studies were performed on carbon-supported Pt Co electrocatalysts in a wide range of Pt:Co compositions prepared with different methods. Salgado et al. [74] investigated in PEMFC carbon-supported Pt Co alloys prepared by alloying at 900 8C, with Co atomic ratio 10, 15, 20 and 25 at.%. Pt 75 Co 25 /C showed the best kinetic parameters for the ORR, ascribed to the optimal Pt Pt bond distance. Xiong et al. [76] investigated carbon-supported Pt M (M = Fe, Co, Ni and Cu) synthesized at low temperature by reduction with sodium formate, in H 2 SO 4 solutions and in PEMFCs. The Pt M alloy catalysts showed improved catalytic activity for the ORR in

142 E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 comparison to Pt. Among the various alloy catalysts investigated, the Pt Co catalysts presented the best performance, with the maximum catalytic activity for a Pt:Co atomic ratio around 1:7. Paulus et al. [59,89] investigated the oxygen reduction kinetics on carbon-supported Pt Ni and Pt Co alloy catalysts in the atomic ratio Pt:M 3:1 and 1:1 using the thin film RDE method in 0.1 mol L 1 HClO 4 in the temperature range between room temperature and 60 8C. Kinetic analysis revealed a small activity enhancement (per Pt surface atom) of ca. 1.5 for the 25 at.% Ni and Co catalysts, and a more significant factor of 2 3 for the 50 at.% Co in comparison to pure Pt. The 50 at.% Ni catalyst was less active than Pt and unstable at oxygen electrode potentials at 60 8C. Yang et al. investigated the effect of the composition on ORR activity of Pt Ni [79] alloy catalysts prepared by a Ptcarbonyl route. The maximum activity of the Pt-based catalysts was found with ca. 30 40 at.% Ni content in the alloys, corresponding to Pt Pt mean interatomic distances of ca. 0.2704 0.2724 nm. Thus, the authors concluded that the high activity of these catalysts for the ORR comes from the favorable Pt Pt mean interatomic distance caused by nickel alloying and the disordered surface structures induced by the particle size. As previously reported, there is evidence of dissolution of the transition metal from the Pt alloy in hot H 3 PO 4. However, the operating environment of the polymer electrolyte fuel cells is not nearly as severe as in phosphoric acid fuel cells then a better stability of these alloy catalysts in the PEMFC environment would be expected. Mukerjee and Srinivasan [65] investigated durability and stability of carbon-supported Pt 3 Cr, Pt 3 Co and Pt 3 Ni alloy catalysts in PEMFCs. The lifetime studies on these catalysts under PEMFC operational conditions showed only negligible losses in performance over periods of 400 1200 h. In this time range a high stability of the ratio between the amount of the alloying component and the amount of Pt in the catalyst was observed. As previously reported, by XPS measurements, Toda et al. [49] found that most of the Ni, Co or Fe easily disappeared from all the Pt alloy surface layers, probably by dissolution, by submitting the surface to an anodic potential of 1.1 V, even in diluted acid solution. However, the alloy compositions determined with EDX analysis did not show apparent differences before and after the electrochemical experiments. Also, it was observed negligible differences in the XRD patterns before and after electrochemical tests. These results indicate that the loss of the base metal only occurs within few monolayers of the alloy surface. The modification of the electronic structure of this Pt layer with respect to that of the bulk alloys gives rise to an enhancement of the ORR. Colon-Mercado et al. [90] evaluated the catalytic, corrosion and sintering properties of commercial Pt/C and Pt 3 Ni/C catalysts using an accelerated durability test. The degree of alloying of the Pt 3 Ni catalyst was not indicated. They found that the total amount of Ni dissolved depends on the applied potential, and increases from 8.3 to 12% when the potential is increased from 0.4 to 0.9 V versus the standard hydrogen electrode. A strong correlation between the amount of Ni dissolved and the oxygen reduction activity of the catalyst was observed. Moreover, the carbon-supported Pt 3 Ni alloy showed better resistance to sintering than a pure platinum catalyst. According to the authors, the mobility of platinum on a carbon surface is hindered when Ni is present; thus, the sintering effect of platinum atoms is suppressed. On the other hand, Park et al. [60] observed no dissolution of Ni in the bulk Pt Ni (1:1) alloy nanoparticle catalyst in 2.0 mol L 1 CH 3 OH + 0.5 mol L 1 H 2 SO 4 in the potential range 0 1.6 V versus NHE. Although some dissolution of Ni could take place, the amount dissolved from the Pt lattice was apparently very small. According to the authors, this indeed implies that the metallic state of nickel is either passivated by Ni hydroxides or exists as a stable phase within the platinum lattice. Salgado et al. [74] evaluated the stability of the Pt 75 Co 25 /C catalyst following 24 h of PEMFC operation. The Pt:Co atomic ratio increased from the nominal composition to 82:18. On the basis to XRD analysis, the amount of cobalt lost was ascribed to theloss ofnon-alloyedcobalt.abetter stability ofthecell with the cobalt-containing catalyst upon several cycles between 0.05 and 0.78 V versus RHE than that of the cell with Pt/C was observed. Yu et al. [91] evaluated the durability of Pt Co cathode catalysts in a dynamic fuel cell environment with continuous water fluxing on the cathode. The results indicated that cobalt dissolution neither detrimentally reduces the cell voltage nor dramatically affects the membrane conductance. The overall performance loss of the PtCo/C membrane electrode assemblies (MEAs) was less than that of the Pt/C MEA. Gasteiger et al. [92] proposed a pre-leaching of the alloy to minimize the contamination of the membrane electrode assembly (MEA) during operation owing to Co dissolution. They tested leached and unleached catalysts in small 50 cm 2 single cells under oxygen to evaluate catalyst activity. A multiply leached Pt Co/C catalyst shows the highest activity (with a gain of about 25 mvover Pt/C) over the entire range of current densities as compared to Pt/C under identical conditions. Finally, Bonakdarpour et al. [93] studied the dissolution of Fe and Ni from Pt 1 x M x (M = Fe, Ni) catalyst under simulated operating conditions of PEMFCs. Electron microprobe measurements showed that transition metals are removed from all compositions during acid treatment, but that the amount of metal removed increases with x, acid strength and temperature. For low M content (x < 0.6) the dissolved transition metals originated from the surface, while for x > 0.6 the transition metals dissolved also from the bulk. XPS results indicated complete removal of surface Ni(Fe) after acid treatment at 80 8C for all compositions. 5. The methanol oxidation reaction on Pt, Pt Ni and Co electrocatalysts 5.1. Improved activity for the MOR on Pt Ni and Co electrocatalysts Pt Ni, Co and other transition metal alloys were investigated by Page et al. [30] as low cost alternative catalysts for the direct oxidation of methanol and compared them with Pt

E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 143 Fig. 2. Cyclic voltammetries at 75 8C in 1 mol L 1 CH 3 OH of the carbon membrane electrodes with Pt, Pt Ru, Co and Ni. A saturated calomel electrode (SCE) was used as the reference electrode. Reprinted from Ref. [30] with permission from Elsevier. and Pt Ru using cyclic voltammetry. Commercial carbonsupported Pt Ni and Co in the atomic ratio 1:1 were used. The alloy catalyst Pt Co/C was found to be a better catalyst for methanol oxidation in acid solution compared with Pt and other transition metal alloys. Fig. 2 shows the cyclic voltammetries at 75 8C in 1 mol L 1 CH 3 OH of the carbon membrane electrodes with Pt, Pt Ru, Co and Ni. The highest oxidation current was obtained with the Pt Co electrocatalyst. Compared with Pt Ru/ C, Pt Co/C is less costly and has better electrochemical performance. The onset potential of methanol oxidation in 0.5 mol L 1 H 2 SO 4 +1molL 1 CH 3 OH at 25, 50 and 75 8C was evaluated and found to shift negatively at high-temperatures. The apparent activation energy for CH 3 OH ads formation was overcome at lower potentials as the temperature was increased. The different potentials for the onset of the CH 3 OH oxidation on carbon-supported catalysts are given in Table 1.At 75 8C (DMFC operation temperature) the onset potentials for the MOR on Pt Co and Ni were lower than that on pure Pt. Chi et al. [94] prepared nanoparticles of different atomic ratios of Pt Co and measured the peak currents in cyclic voltammetry of formic acid oxidation (as previously reported formic acid may be a reaction product of the oxidation of methanol). The maximum activity of Pt Co catalysts was about one order of magnitude higher than that of pure Pt nanoparticles. The optimum Pt:Co atomic ratio was between 1:1.1 and 1:3.5. The presence of Co appears to significantly enhance the electro-oxidation of formic acid. Zhang et al. [68,95] prepared unsupported Pt Co nanoparticles using a water-in-oil reverse microemulsion of water/ Triton X-100/propan-2-ol/cycloexane with hydrazine solution as the reducing agent. Electrodes with different Pt Co compositions were tested for methanol oxidation in an alkaline electrolyte. From the cyclic voltammograms, the electrodes with low Co content (Pt:Co < 1:1) outperformed those with high Co content (Pt:Co > 1:1), while all Pt/Co electrodes better performed than the pure Pt electrode. The authors also made a comparison of steady-state currents obtained by current step experiments. Fig. 3 shows the chronopotentiograms at 20 ma cm 2 for methanol oxidation at room temperature on electrodes with different Pt:Co ratios. In agreement with the transient CV results, the 1:0.5 atomic ratio of Pt:Co alloy shows the best performance with the highest catalytic activity of all the composition investigated. The order of activity for Pt:Co with different compositions is 1:0.5 > 1:0.75 > 1:0.25 > 1:3 > 1:2 > 1:1 > 1:0. According to the authors, in addition to the possible enhancement of formaldehyde oxidation by cobalt, the alloying of Co atoms to Pt lowers the electronic binding energy in Pt and favours the C H cleavage reaction at low potential. Moreover, the presence of cobalt oxides provides an oxygen source for CO oxidation at lower potentials. The two considerations combined determine the optimum Pt:Co ratio to be about 1:0.5. Zeng and Lee [96] prepared carbon-supported Pt and Pt Co catalysts by reduction of metal precursors with NaBH 4. Electrochemical measurements by cyclic voltammetry and chronoamperometry demonstrated consistently high catalytic activity and improved resistance to carbon monoxide for the Pt Co catalysts, particularly for that prepared in unbuffered solution (smaller particle size). Park et al. [60] studied the electro-oxidation of methanol in sulphuric acid solution using unsupported Pt, Pt Ni (1:1 and 3:1), Table 1 Potential for the onset of CH 3 OH oxidation at various temperatures on carbonsupported alloy catalysts [30] Catalyst Onset potential at 25 8C (mv vs. RHE) Onset potential at 50 8C (mv vs. RHE) Pt Co (1:1) 395 345 270 Pt Ni (1:1) 370 335 280 Pt Ru (1:1) 300 280 250 Pt 385 345 320 Onset potential at 75 8C (mv vs. RHE) Fig. 3. Chronopotentiograms of methanol oxidation at 20 ma cm 2 in 1 mol L 1 CH 3 OH in 1 mol L 1 KOH at room temperature using carbon papers with nanoparticles of different ratios of Pt to Co and the same Pt loading, 0.495 mg cm 2. (a) Pt; (b) Pt Co 1:3; (c) Pt Co 1:2; (d) Pt Co 1:1; (e) Pt Co 1:0.25; (f) Pt Co 1:0.75; (g) Pt Co 1:0.5. Reprinted from Ref. [68] with permission from Elsevier.

144 E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 and Pt Ru (1:1) alloy nanoparticle catalysts. The methanol oxidation current measured on the Pt Ni based catalysts in 2.0 mol L 1 CH 3 OH + 0.5 mol L 1 H 2 SO 4 at room temperature exceeded that obtained with pure Pt. The comparison of the onset potentials for methanol oxidation on Pt Ni electrocatalysts (320 mv for Pt:Ni atomic ratio = 3:1, and 290 for Pt:Ni = 1:1) and on Pt (350 mv) indicated that the Pt Ni nanoparticles show relatively good electrocatalytic activity. Using Pt alloy nanoparticles, Park et al. [60] measured plots of oxidation current versus time (chronoamperometry, CA) in 2.0 mol L 1 CH 3 OH + 0.5 mol L 1 H 2 SO 4 at 0.42 V, for 3600 s. For each catalyst, the decay in the methanol oxidation was different; for instance, pure platinum nanoparticles required 10 min to reach 70% of the initial current and the oxidation current is reduced steeply. After 1 h, the current decreased below 40% of the initial value. In contrast, Pt Ni (1:1) and Pt Ru (1:1) supported higher currents, and it may be concluded that they have higher activity than pure Pt. After 1 h, the order of surface activity for the methanol oxidation was Pt Ni (1:1) > Pt Ru (1:1) > Pt. By combining voltammetry and CA, the authors concluded that Pt Ni (1:1) represent the best alternative candidate for the DMFC anodecatalysts with respecttopt Ru, evenifithastoberemarked that, from the results depicted in the second paper of this series [97], is difficult to asses such improvement. In the Pt Ni alloy nanoparticles, the Ni species included metallic Ni,NiO, Ni(OH) 2, andniooh,andtheratiobetweenthethreeoxideswassimilarfor the different Pt Ni alloys. XPS Pt4f peak values for Pt Ni and Pt Ru alloy nanoparticles were compared to the value obtained from pure Pt. The peaks were shifted from 0.09 for Pt Ru to 0.35 and 0.36 evat Pt Ni; that is, they moved toward the lower Pt4f binding energy. The binding energy shift for Pt in the Pt Ni nanoparticles was interpreted to result from the modification of the electronic structure of platinum by electron transfer from Ni to Pt. According to the authors, the electron transfer may contribute to the enhanced CO oxidation (CO generated from methanol oxidation), that is, to the CO tolerance on the Ni-containing composites, in comparison to pure Pt samples. However, the competing effect in the Ni enhancement may be due to the surface redox activity of Ni oxides toward the CO. Mathiyarasu et al. [98] investigated the electrocatalytic activity of electrodeposited Pt Ni alloy layers on an inert substrate electrode for methanol oxidation reaction. By solidstate polarization measurements in 0.5 mol L 1 CH 3 OH/ 0.5 mol L 1 H 2 SO 4 solutions they observed that the onset of the electro-oxidation shifts to less anodic potential values, while also exhibiting current enhancements up to about 15 times the currents obtained for the pure Pt electrodeposit. A critical composition of Pt 92 Ni 8 was found to exhibit the maximum electrocatalytic activity, beyond which the activity drops. According to the authors, while the promotion of the electro-oxidation is understood to be largely due to the alloy catalyst, surface redox species of Ni oxide formed during the electro-oxidation process may also contribute to the oxygenation of CO ads, thereby enhancing the oxidation current. Park et al. [60,97] also investigated the effect of Ni insertion on PtRu catalysts in the methanol oxidation. The activity of Pt Ru Ni in the atomic ratio 5:4:1 for the MOR was compared with that of Pt, Pt Ru and Ni. The onset potential for methanol oxidation was in the order Pt Ru Ni (5:4:1) < Pt Ru (1:1) < Pt Ni (1:1) < Pt Ni (3:1) < pure Pt. Pt Ru Ni had a larger current density, a larger turnover number and a smaller activation energy for methanol oxidation than Pt Ru (1:1). Polarization and power density data in single DMFC tests were in good agreement with the voltammetry and chronoamperometry data, for which Pt Ru Ni showed a higher catalytic activity than Pt Ru (1:1). According to the authors, one way to interpret this result it is that the shift of d electron density from Ni to Pt would reduce the Pt CO bond energy. Furthermore, Ni (hydro)oxides on the Pt Ru Ni nanoparticles could promote methanol oxidation via a surface redox process. 5.2. No effect of Co(Ni) presence on the MOR activity on Pt Co and Ni electrocatalysts Goikovic [99] investigated the electrochemical oxidation of methanol on a Pt 3 Co bulk alloy in acid solutions. Contrary to the previous results, she found that cobalt does not show a promoting effect on the rate of methanol oxidation on the Pt 3 Co bulk alloy with respect to a pure Pt surface. Drillet et al. [67] prepared an unsupported Pt 70 Ni 30 catalyst by melting together Pt and Ni pellets in a vacuum arc and studied the methanol oxidation and the electrochemical oxygen reduction reaction at Pt and Pt 70 Ni 30 in 1 mol L 1 H 2 SO 4 /0.5 mol L 1 CH 3 OH. By cyclic voltammetry they found no significant difference in the methanol oxidation on Pt and Pt 70 Ni 30, particularly regarding the onset potential for methanol oxidation. On the other hand, by means of a rotating disc electrode they found that in a methanol containing electrolyte solution the onset potential for oxygen reduction at Pt Ni is shifted to more positive potentials and the alloy catalyst has an 11 times higher limiting current density for oxygen reduction than Pt. Thus, they concluded that Pt Ni as cathode catalyst should have a higher methanol-tolerance for fuel cell applications. 5.3. Decreased activity for the MOR on Pt Ni and Co electrocatalysts Salgado et al. [100] found that the onset potential for methanol oxidation at room temperature on Pt Co/C electrocatalysts with Pt:Co atomic ratio 85:15 and 75:25 is shifted to more positive potentials than Pt. According to the authors, the carbon-supported Pt Co/C alloy electrocatalysts possess enhanced oxygen reduction activity compared to Pt/C in the presence of methanol in a sulphuric acid electrolyte. The higher methanol-tolerance of Co-containing catalysts with respect to that of Pt alone can be clearly seen in Fig. 4, where the potentials at 0.1 ma cm 2 ðe 0:1mAcm 2Þ are plotted against methanol concentration. The decrease of E 0:1mAcm 2 on the Pt/C electrocatalyst with increasing methanol concentration is much higher than that on the alloys, showing that the Pt Co/C electrocatalysts have a better tolerance to the presence of methanol than Pt/C in sulphuric acid solution. Antolini et al. [101] prepared carbon-supported Pt 70 Ni 30 by NaBH 4 reduction of the precursors and investigated the activity

E. Antolini et al. / Applied Catalysis B: Environmental 63 (2006) 137 149 145 Fig. 4. Dependence of the potential at 0.1 ma cm 2 on methanol concentration during O 2 reduction in 0.5 mol L 1 H 2 SO 4 for carbon-supported Pt and Pt Co electrocatalysts. Reprinted from Ref. [100] with permission from Elsevier. for the methanol oxidation and the oxygen reduction reactions in sulphuric acid. They found that the current densities for the methanol oxidation reaction on the Pt Ni/C alloy electrocatalyst were lower than that on the Pt/C electrocatalyst and the onset potential for methanol oxidation at room temperature on the Pt Ni/C (440 mv) shifted to more positive potentials as compared to Pt/C (375 mv), indicating that the alloy electrocatalyst is less active for methanol oxidation than the Pt/C electrocatalyst. The experimental results regarding the ORR in H 2 SO 4 solution in the presence of methanol are summarized in Fig. 5, where the mixed potential is plotted versus the methanol concentration at 0.05 and 0.1 ma cm 2 (specific activity). The polynomial regression for the Pt and Pt Ni data was the following: Pt : E ¼ E 0 73 ½CH 3 OHŠ þ6:8 ½CH 3 OHŠ 2 (4) Pt Ni : E ¼ E 0 44 ½CH 3 OHŠ þ6:6 ½CH 3 OHŠ 2 (5) In a first approximation up to 2 mol L 1 CH 3 OH the dependence of E on [CH 3 OH] is linear for both Pt and Pt Ni, and de/ d[ch 3 OH] is about twice that for Pt ð 60 mv mol CH3 OH 1 LÞ than for PtNi ð 31 mv mol CH3 OH 1 LÞ both at 0.05 and 0.1 ma cm 2. Then, it seems that Pt Ni is more methanoltolerant than pure Pt. Inuikai and Itaya [102] prepared bimetallic Pt Ni catalysts by vapour deposition of alternate layers of Pt and Ni on Pt(1 1 1) at room temperature. The methanol oxidation reaction was carried out in an electrochemical chamber. The onset potential and that of the maximum current for the MOR were shifted to higher potentials. No Ni was detected on the surface by Auger electron spectroscopy (AES). Then, in agreement to Toda et al. [49], the electrochemical behaviour of Pt Ni could be ascribed to the modification of the electronic structure of Pt atoms on top of the Ni layer. Yang et al. [103] prepared carbon-supported Pt Ni alloy catalysts with 40 wt.% total metal loading via the carbonyl complex route, and studied the activity for the MOR and for ORR in the presence of methanol in H 2 SO 4. By linear sweep voltammetry measurements, as can be seen in Fig. 6, they found that the methanol oxidation current densities on Pt Ni alloy catalysts are lower than that on a Pt/C catalyst and that the methanol oxidation peaks on Pt Ni alloy catalysts shift slightly to more positive potentials as compared to the Pt/C catalyst, indicating that the oxidation of methanol on the alloy catalysts is less active than that on a Pt/C catalyst. They observed significantly enhanced electrocatalytic activities for ORR in methanol-containing electrolyte than pure Pt. Among the cathode catalysts used, a maximum activity was found with a Pt:Ni atomic ratio of 2:1. According to the authors, the high methanol-tolerance of Pt Ni alloy catalysts during oxygen reduction could be ascribed to a lowered activity for methanol oxidation, originated from the composition effect and the disordered structure of the alloy catalysts. Fig. 5. Electrode potential vs. methanol concentration at 0.05 and 0.1 ma cm 2 (current expressed as specific activity) for Pt/C and Pt 70 Ni 30 /C electrocatalysts. Circles: Pt/C; triangles: Pt 70 Ni 30 /C. Solid symbols: j = 0.05 ma cm 2 ; open symbols: j = 0.1 ma cm 2. Reprinted from Ref. [101] with permission from Elsevier. Fig. 6. Linear sweep voltammetries of methanol oxidation on nanosized Pt/C and Pt Ni alloy catalysts in nitrogen saturated 0.5 mol L 1 H 2 SO 4 +0.5molL 1 CH 3 OH solution at a scan rate of 5 mv s 1 and a rotation speed of 2000 rpm. Solid line: Pt/C; dashed line: Pt 2 Ni/C; dotted line: Pt 3 Ni 2 /C; dashed dotted line: PtNi/C. Reprinted from Ref. [103] with permission from Elsevier.