CO surface electrochemistry on Pt-nanoparticles: A selective review

Electrochimica Acta 50 (2005) 5144–5154

CO surface electrochemistry on Pt-nanoparticles: A selective review K.J.J. Mayrhofer, M. Arenz 1 , B.B. Blizanac, V. Stamenkovic, P.N. Ross, N.M. Markovic ∗ Materials Science Division, Lawrence Berkeley National Laboratory, University of California, Berkeley, CA 94720, USA Received 27 December 2004; received in revised form 24 February 2005; accepted 24 February 2005 Available online 1 July 2005

Abstract Oxidation of CO on platinum nanoparticles ranging in size from 1 to 30 nm has been studied in acid electrolytes. We found that Pt nanoparticles, characterized by transmission electron microscopy, are not perfect cubo-octahedrons and that large particles have “rougher” surfaces than small particles. The importance of “defect” sites for the catalytic properties of nanoparticles was probed by using infrared reflection absorption spectroscopy (IRAS) and rotating disk electrode. From IRAS experiments, by monitoring how the vibrational frequency of a-top CO (νCO ) as well as the concomitant development of dissolved CO2 are affected by the number of defects on Pt nanoparticles, we suggested that defects play a significant role in CO “clustering” on nanoparticles, causing CO to decrease/increase in local coverage, which results in anomalous redshift/blueshift νCO frequency deviations from the normal Stark-tuning behavior. The observed νCO deviations are accompanied by CO2 production, which increases by increasing the number of defects on the nanoparticles, i.e., 1 ≤ 2 < 5  30 nm. We suggest that the catalytic activity for CO adlayer oxidation (CO stripping) is predominantly influenced by the ability of the surface to dissociate water and to form OHad on defect sites. We demonstrate that the catalytic activity of Pt nanoparticles for CO oxidation under the condition of continuous CO supply to the surface depends on the pre-history of the electrode. If the surface is precovered by CO, the particle size has a negligible effect on CO oxidation. However, on an oxide-precovered surface CO bulk oxidation increases with decreasing particle size, i.e., with increasing oxophilicity of the particles. We found, if specific sites on the surface are active for OH adsorption, then the electrocatalytic activity for CO oxidation changes as the concentration of these sites changes with particle size. © 2005 Elsevier Ltd. All rights reserved. Keywords: Particle size effect; Infrared spectroscopy; Platinum; Nanoparticles; Carbon monoxide oxidation

1. Introduction The surface electrochemistry of CO adsorbed on transition metal surfaces has been the subject of intense theoretical and experimental work; for an overview see Refs. [1,2]. In these studies emphasis has been placed on linking the microscopic structure information concerning the CO adlayer structures on single crystal surfaces to the thermodynamics and other macroscopic electrochemical responses at electrified interfaces. Whereas information regarding macroscopic properties has come from classical electrochemical techniques [1], ∗

Corresponding author. Tel.: +1 510 495 2956; fax: +1 510 486 5530. E-mail address: [email protected] (N.M. Markovic). 1 Present address: Department of Surface Chemistry and Catalysis, University of Ulm. 0013-4686/$ – see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.electacta.2005.02.070

the chemical, physical and structural properties that occur on the atomic scale have been obtained from either a combination of in situ scanning tunneling microscopy and vibrational spectroscopy [3,4] or surface X-ray diffraction and vibrational spectroscopy [5]. Of the various systems studied, the adsorptive and catalytic properties of CO adsorbed on platinum single crystals have been of central significance because this system provided the opportunity to gain understanding that ultimately can lead to the design of new catalysts and in part for the purpose of understanding the activity pattern of Pt metal nanoparticles employed in fuel cells in the size range of few nanometers. There is no simple ideal structure that will necessarily model all the aspects of nanoparticle catalysts, particularly in the exact configuration as they are used in electrolytic cells. However, if the equilibrium shape of a nanoparticle,

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

as is the case for the fcc metal Pt, is a cubo-octahedral structure consisting of (1 1 1) and (1 0 0) facets bounded by edge atom rows that are like the top-most rows in the (1 1 0) surface, then well-characterized single-crystal surfaces may be used as reasonable models [6]. Many research groups have used this strategy to predict what the particle size effect might be, i.e., the variation of the reaction rate or selectivity with the characteristic dimension of metallic clusters at the electrified solid–liquid interface. Although this approach did provide some new insight into the relationship between the structure sensitivity [7–12] and the particle size effect [13–15], in many cases this tactic has also showed significant weaknesses. Probably the most notable example that the catalytic activities observed on single crystal surfaces cannot be used as one-to-one models for real nanoparticles has been the oxidative removal of saturated CO adlayers (the so-called CO stripping voltammetry) on Pt nanoparticles in CO-free acid electrolytes. In particular, it has been found that, although more oxophilic, smaller particles are less active than larger particles. It has been proposed that the reason for this observation is a stronger bonding of CO to the surface of smaller particles and a concomitant decrease in CO diffusion [12,14,16–18]. Interesting yet largely unexplored issues in the sizedependent physicochemical properties of Pt nanoparticles are the importance of “defect” sites, which are inherently present on Pt nanoparticles [19] and the impact of Pt oxide on the CO bulk oxidation. The latter issue is of practical importance considering that the catalyst experiences usually a continuous supply of CO to the surface. Given the catalytic richness of the Pt–CO system, the examination of the effects of defects on Pt nanoparticles pre-covered by CO as well as the nature of surface oxide on CO bulk oxidation are of interest. Such a study of Pt nanoparticles in the size range of 1–30 nm is reported here. We demonstrate how the particle size dependent number of defects in cubo-octahedral particles affects the vibrational properties of CO and the CO2 production during CO oxidation. We also consider, how the electrocatalytic activity for CO oxidation changes with the experimental conditions, i.e., excess of OHad or CO on the catalyst surface.

2. Experimental 2.1. Catalyst samples Four different Pt high surface area catalysts were used in this study. Three samples, supplied by TKK (Tokyo, Japan), were carbon supported Pt nanoparticles with mean diameters of 1–1.5, 2–3 and 5 nm, respectively (analysis by TKK). The fourth sample, consisting of a nanostructured Pt film supported on crystalline organic whiskers [20], was supplied by 3 M Company (St. Paul, MN, USA). The particle size of the latter sample was roughly estimated to be about 30 nm based on the charge required to oxidize a full monolayer of adsorbed CO in a CO-stripping experiment. The Pt loading

5145

of the catalyst was roughly 20% for the 1 and 2 nm catalysts, 50% and 91% for the 5 and 30 nm samples, respectively. In the following, the catalysts will simply be denoted as 1, 2, 5 and 30 nm catalysts. 2.2. Electrochemical measurements The catalyst preparation has been described previously [21]. In short, the catalyst was dispersed ultrasonically in ultrapure water and 20 ␮l of the suspension was pipetted onto a glassy carbon substrate (0.283 cm2 geometrical surface area) leading to a Pt loading of 14 ␮gPt /cm2 for the carbon supported catalysts (TKK 1, 2 and 5 nm) and 42 ␮gPt /cm2 for the 3 M (30 nm) sample. The dried (Ar atmosphere) catalyst film was attached to the substrate by a thin Nafion film. The thus prepared surface was then transferred to the electrochemical cell protected by a drop of ultrapure water, immersed under potential control at 0.05 V in argon-saturated solution and a cyclic voltammogram was recorded. Two different procedures were used to form a saturated CO adlayer on the Pt nanoparticles. In the first procedure, denoted hereafter as “oxide-annealing”, before CO is adsorbed at 0.05 V the catalyst was cycled in Ar (Air Products 5N5 purity) saturated solution until a well-established cyclic voltammogram was observed. A complete CO adlayer was achieved by holding at 0.05 V for 10 min in CO-saturated solution. In the second procedure, denoted hereafter as “CO-annealing” the electrode was annealed in CO saturated solution by cycling of the electrode potential between 0.05 < E < 1.0 V for 5 min, before holding at 0.05 V for 5 min. We used this experimental approach because we recently found that the “CO-annealing” pretreatment of Pt(1 1 1) may facilitate the removal of surface irregularities, which after the flame annealing preparation method, are inherently present on the Pt(1 1 1) surface. As a consequence, upon CO-annealing of the (1 1 1) surface the domain size of the p(2 × 2)–3CO structure, which is formed on Pt(1 1 1) at low potentials, is significantly improved [1]. As we demonstrate below, the catalytic and spectroscopic properties of high surface area catalysts pre-treated by these two methods are completely different. The stripping curves were recorded with a scan rate of 1 mV/s and the upper potential limit was set to 1.0 V to guarantee the complete oxidation of the CO adlayer. For the electrooxidation of CO dissolved in the electrolyte (CO bulk), before the potentiodynamic measurements were performed, the electrolyte was saturated with CO (1 atm) for 25 min while the electrode potential was held at 0.05 V. The electrochemical measurements were conducted in a standard three-compartment electrochemical cell. The reference electrode was a saturated calomel electrode (SCE) separated by an electrolytic bridge from the reference compartment. All potentials, however, are referenced to the potential of the reversible hydrogen electrode (RHE) calibrated from the hydrogen oxidation/reduction reaction measured in a rotating ring disk configuration in the same electrolyte. As counter electrode a Pt mesh was used. The electrolyte was

5146

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

prepared using pyrolytically triply distilled water and concentrated HClO4 (Aldrich, double distilled). 2.3. Infrared spectroscopy For measuring FTIR spectra on supported nanoparticles, we applied a new methodology, which has been recently described in detail in a separate paper [22]. In short, as for the electrochemical measurements the catalyst was dispersed ultrasonically in water and about 20 ␮l of the suspension were pipetted onto a heated (∼130 ◦ C) gold substrate (0.785 cm2 geometrical surface area), cooled down in an argon stream and rinsed carefully with ultrapure water to remove loosely bounded catalyst particles. No further attachment to the substrate was necessary. For the in situ FTIR measurements a Nicolet Nexus 670 spectrometer was available equipped with a liquid-N2 cooled MCT detector. All IR measurements were performed in a spectroelectrochemical glass cell designed for an external reflection mode in a thin layer configuration [23]. The cell is coupled at its bottom with a CaF2 prism beveled at 60◦ from the prism base. Prior to each experiment the solution was saturated with argon. The electrode was immersed into the electrolyte at a potential of 0.05 V and a CV recorded. For the CO stripping measurements recorded in the spectroelectrochemical cell, the above-described two different adsorption procedures for obtaining a saturated CO adlayer, i.e., oxide annealing and CO-annealing, were applied before pressing the sample onto the prism. Starting at 0.05 V the potential was scanned with a scan rate of 1 mV/s in positive direction while continually recording spectra. In order to obtain a single spectrum, four interferometer scans were coadded. The recording time was thus reduced to ca. 2.5 s per spectrum. The resolution of the spectra was 4 cm−1 and ppolarized light was used. Absorbance spectra were calculated as the ratio −log(R/R0 ) where R and R0 are the reflectance values corresponding to the sample and reference spectra, respectively. Reference spectra were recorded either at 0.95 or 0.05 V, where adsorbed CO (COad ) either is completely oxidized or well before the onset of COad oxidation, respectively. The potential in the spectroelectrochemical cell was controlled by a SCE separated by a glass frit from the main compartment. All potentials, however, are referenced to the RHE calibrated from the hydrogen evolution/oxidation reaction in a separated cell using the same electrolyte.

3. Results and discussion The interpretation of the oxidation of a CO monolayer on Pt nanoparticles in acid electrolytes has been extensively discussed in Refs. [13–15]. Under experimental conditions in which CO is adsorbed at potentials close to the RHE and oxidatively removed in the first positive sweep (recorded at the relatively fast sweep rates), they found that the position of the CO stripping peak shifts to positive potentials with decreasing the Pt particle size. Although it is mislead-

ing and fundamentally incorrect to correlate the position of the peak maxima in the CO stripping voltammetry with the catalytic activity of CO, the authors correctly proposed that the oxidative removal of CO is enhanced on larger Pt particles. Further insight into the particle size effects has been obtained from chronoamperometry, where the shape of transient currents has been modeled employing the active site model [15]. The model for perfect cubo-octahedral particle suggested restricted CO mobility at Pt nanoparticles below ca. 2 nm size and a transition towards fast diffusion when the particle size exceeds 3 nm. Unfortunately, the model did include neither the competitive adsorption of bisulfate anions (which can control the adsorption of OH [11] as well as the mobility of co-adsorbed CO [24]) nor the existence of irregularities on nanoparticles which, as for extended single crystal surfaces, may serve as the active centers for OH adsorption [25]. The proposition that the CO mobility is the determining factor in CO stripping voltammetry is consistent with FTIR findings that CO interacts stronger with the high density of edge-corner coordination in small nanoparticles than with (1 0 0) and (1 1 1) facets [13]. Very recently, using high-resolution transmission electron microscopy (TEM), infrared reflection absorption spectroscopy (IRAS) and electrochemical (EC) measurements, Arenz et al. found that platinum nanoparticles ranging in size from 1 to 30 nm are, in fact, not perfect cubo-octahedrons and that large particles have “rougher” surfaces than small particles [25]. The importance of “defect” sites for the catalytic properties of nanoparticles was probed in IRAS experiments by monitoring how the vibrational frequency of a-top CO (νCO ) as well as the concomitant development of the asymmetric O–C–O stretch of dissolved CO2 are affected by the number of defects on the Pt nanoparticles. In what follows, selected TEM/FTIR/EC results published recently from our group [25] for CO-stripping on Pt nanoparticles will be combined and compared with new FTIR and EC data for the bulk CO oxidation on the same Pt catalysts. 3.1. Surface characterization Low magnification TEM-images of the Pt catalysts supported on carbon and on crystalline organic whiskers are summarized in Fig. 1a–c and d, respectively. The images of the carbon-supported catalysts illustrate that the distribution of metal particles on the carbon support is rather uniform. Histograms of the particle size distribution (not shown), which included analysis of several different regions of the catalysts, established that the average particle size of Pt supported on carbon is ca. 1, 2 and 5 nm respectively. This was also in very good agreement with the particle size calculated from electrochemical measurements, based on the charge required to oxidize a monolayer of adsorbed CO in a CO-stripping experiment. On the contrary to the carbon-supported catalysts, the organic whiskers support is completely covered by Pt [20] (91% Pt content), and the establishment of a “particle size” from the TEM images was

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

5147

Fig. 1. Transmission electron microscopy images of the four catalysts studied. In the left top corner is the 1 nm (a), right top the 2 nm (b), left bottom the 5 nm and (c) catalysts all by TKK. The right bottom picture shows the 30 nm catalyst (d) supplied by 3 M.

not possible. However, by using the CO-stripping charge we were able to estimate the active surface area to approximately 8 m2 /gPt , which would be equivalent to particles of about 30 nm on a carbon support. The analysis of images taken with higher resolution (not shown, for details see Ref. [25]) clearly indicated that the Pt crystallites are in general not perfect cubo-octahedrons and that the particles provide a variety of low-coordination sites for adsorption and catalysis that differ from the surface processes on regular facets. As previously discussed by Sattler and Ross [19] and later confirmed by Arenz et al. [25], large particles are quite rough on the atomic scale, containing irregular sites and steps and occasionally twinned particles. On the other hand, the smallest particles show mostly regular facets and uniform edge-corner sites, as would be expected from perfect cubo-octahedrons. The increase of the relative amount of defects with particle size has some major implications on the oxidation of adsorbed CO molecules, as we will show in the upcoming sections. 3.2. Electrochemical measurements: CO monolayer oxidation in CO-free solution Assuming that CO oxidation on small particles obeys the Langmuir–Hinshelwood reaction mechanism: 2H2 O → OHad + H3 O+ + e−

(1)

COad + OHad → CO2 + H+ + e−

(2)

which was previously proposed for extended surfaces [11], the CO oxidation rate should increase in the same order as the oxophilicity of the Pt particles [25–28], e.g., 1 < 2 < 5 < 30 nm. In line with previous findings, however, Fig. 2c unambiguously shows that on oxide-annealed surfaces (results for the CO-annealed surfaces can be found in Ref. [25]) the anodic CO stripping peak for the 30 nm particles occurs at lower potentials compared to the particles with a size between 1 and 5 nm. A reason for this observation is still unclear, although it was proposed that it originates from a stronger bonding of CO to the surface of smaller particles and a concomitant decrease in CO diffusion [13,15]. Further inspection of Fig. 2c indicates that in contrast to previous reports [15] the position of the CO stripping peak in the range of 1–5 nm is almost particle size independent. Admittedly hypothetical, it is plausible to suggest that differences in the position of the CO-stripping peaks observed in Fig. 2c and in previous experiments arise primarily due to a much slower sweep rate used in our experiments. Nevertheless, we found that based on CO stripping voltammetry it was, in fact, impossible to determine a true particle size effect for CO oxidation. Because of that, the analysis of the oxidation of CO adsorbed on Pt nanoparticles has recently been discussed by means of FTIR on both “CO-annealed and oxide-annealed

5148

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

Fig. 2. Summary of the results of CO-adsorption spectra for different “oxide annealed” nanoparticle samples in a CO-stripping experiment in Ar purged solution; the upper part (a) shows the potential dependent C–O stretch frequency of atop adsorbed CO, the lower part (b) compares the normalized peak area of CO2 dissolved into the thin layer (peak position 2343 cm−1 ). The actual stripping curves (µA/cm2geo vs. ERHE ) are depicted in the inset (c), offset against each other for clarity.

surfaces [25]. For our purposes here we will focus on the results for CO oxidation on “oxide annealed” surfaces. 3.3. FTIR measurements In this section, the analysis and discussion of the FTIR results will be divided into two parts: first we describe the vibrational properties of CO adsorbed on Pt nanoparticles pretreated by the “oxide-annealing” method in CO-free solution and then we compare them to the results obtained in CO saturated solution. 3.3.1. Vibrational properties of CO monolayer in CO-free solution The surface sensitivity for the oxidative removal of adsorbed CO is probed in IRAS experiments by monitoring the potential dependence vibrational frequency of a-top bonded CO and the concomitant development of asymmetric O–C–O stretching frequency of dissolved CO2 . In contrast to standard infrared spectral acquisition tactics where the potential is increased stepwise by 50 or 100 mV increments, the spectra here were recorded continuously during a positive potential sweep of 1 mV/s. The focus in the discussion is pointed at the CO-atop band, which lies in between 2050 and 2080 cm−1 on Pt-nanoparticles [29,30], and the band of

dissolved CO2 at 2343 cm−1 , which appears once the CO is being oxidized. The less pronounced CO band around 1880 cm−1 , which can be assigned to bridge-bonded CO, is disregarded here. The summary of the results from the series of potential dependent CO adsorption spectra and the corresponding CO2 production in Fig. 2 serves to illustrate how the nanoparticle size affects the vibrational properties of COad as well as the catalytic properties of Pt, apparent by the CO2 production in Fig. 2b. The spectral data of the 5 nm particles in Fig. 2 reveals some important differences as well as some similarities with the 30 nm particles. As for larger particles, a linear positive Stark-tuning slope of ∼25 cm−1 /V is observed over the potential region where no CO2 is detected, 0.05 < E 0.35 V the slope starts to decrease but, in contrast to the 30 nm particles, it still remains positive all the way up to 0.7 V. Notice that the small change in the Stark-tuning slope is accompanied by the appearance of a relatively weak CO2 band. Although the onset potential for CO2 production is the same on the 5 and 30 nm particles, Fig. 2b reveals that at low potentials the integrated intensities for the CO2 band (ICO2 ) are much lower for the 5 nm particles than for the 30 nm particles. The confirmation that the oxidative removal of COad on 30 nm particles is faster than on the 5 nm particles is also obtained from the ignition potential region (the negative dν/dE slope and extensive production of CO2 ), which is on the 5 nm catalyst shifted by ca. 50 mV towards higher potentials. In the case of the 2 nm particles the linear Stark-tuning slope is observed up to 0.65 V. Concomitant with the development of CO2 the CO frequencies downshift substantially, the latter being indicative of the dissipation of CO islands during oxidative removal of CO by co-adsorbed OH. For the 1 nm particles we observe an anomalous behavior in the νCO –E frequency shift. Fig. 3a shows that around 0.35 V the Stark-tuning slope first slightly increases, then, while still remaining positive, at E > 0.55 V begins to decrease and finally at 0.7 V where the rapid CO oxidation occurs, a negative sign in dν/dE is observed. Interestingly, the blueshift deviations in the νCO frequency are accompanied by a small yet clearly discernable CO2 production, indicating that although the surface is less covered by CO, the interaction energy of CO with the Pt particles is weaker (!). In our recent paper, we have reported that at low overpotentials a similar blueshift in the νCO frequency can be observed at the onset of CO oxidation on Pt(1 1 1) [24]. For the Pt(1 1 1)–CO system, we suggested that the initial CO oxidation is accompanied by adsorption of anions from the supporting electrolyte. We also argued that co-adsorbed anions might lead to a compression of CO molecules, resulting in a higher local CO coverage thus yielding blueshifted a-top CO frequencies due to enhanced dipole–dipole coupling. Extending this phenomenon to CO adsorbed on Pt nanoparticles, the surprising blueshift in the νCO frequency for 1 nm particles at E > 0.3 V may be a consequence of slow but continuous CO oxidation which is accompanied by anion adsorption (either

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

Fig. 3. CO-adsorption spectra recorded in argon saturated 0.1 M HClO4 solution on 1 nm nanoparticles; the upper part summarizes the potential dependent C–O stretch frequency of atop adsorbed CO, the lower part represents the normalized peak area of CO2 dissolved into the thin layer (peak position 2343 cm−1 ).

ClO4 − or trace levels of Cl− impurities) and concomitant CO compression on (1 1 1) facets. Furthermore, the close similarity between the vibrational properties of CO adsorbed on Pt single crystals and very small Pt nanoparticles at low potentials is in line with the TEM analysis, which showed that the surface of large particles contains undistorted (1 1 1) facets in a smaller fraction compared to the surface of small particles. Notice that an U-shaped the dν/dE deviation, i.e., a positive Stark-tuning slope at E > 0.65 V for Pt(1 1 1), is not obtained on 1 nm nanoparticles, suggesting that CO frequency upshift above 0.6 V requires wider (1 1 1) terraces than one present on the 1 nm particle. From data summarized in Figs. 2 and 3 we can conclude that while the onset potential of CO oxidation is almost independent of the particle size, the rate of CO2 production is strongly dependent of the particle size, i.e., 1 ≤ 2 < 5  30 nm. We suggest that the oxidative removal of CO is mainly controlled by the number of “defects”, which in an “ideal” cubo-octahedral particle may serve as an active center for OH adsorption. It was also found that, the onset of CO2 production is accompanied either by redshift (d > 5 nm) or anomalous blueshift (d < 2 nm) deviations from the normal Stark-tuning slope. The interpretation of this “anomalous” dνCO /dE behavior is discussed in relation to similar changes observed at the Pt(1 1 1)–solution interface, i.e., in terms of

5149

Fig. 4. Summary of the CO-adsorption spectra obtained for different nanoparticles in a “CO-bulk” experiment (oxide-annealed electrode surface) in CO purged solution; the upper part (a) shows the potential dependent C–O stretch frequency of atop adsorbed CO, the lower part (b) compares the normalized peak area of CO2 dissolved into the electrolyte (peak position 2343 cm−1 ).

slow CO oxidation, co-adsorption of anions and concomitant CO compression into small islands on fairly smooth (1 1 1) facets. 3.3.2. Vibrational properties of CO monolayer in CO saturated solution As shown in Fig. 4a, by decreasing the particle size the CO band position shifts towards lower wavenumbers, i.e., from 2067 to 2062 and 2052 cm−1 for 30, 5 and 2 nm particles, respectively. As discussed in Refs. [31–33], the observed redshift can be rationalized by an interplay between the chemical shift (stronger CO adsorption on smaller nanoparticles) and the dipole-shift (enhanced CO–CO lateral interactions on larger particles). Further inspection of Fig. 4 shows that below 0.25 V the “normal” Stark-tuning behavior, with a slope of ∼25 cm−1 /V, is observed on all three catalysts in the potential region in which the CO adlayer is stable, i.e., no CO2 production in Fig. 4b. Concomitant with the development of CO2 at ca. 0.3 V, CO frequencies either downshift substantially (as in the case of the 30 nm particles) or moderately (as in the case of the 2 nm particles). If we use ICO2 or the redshift deviations from the normal Stark-tuning slope as a measure for the catalytic activity, then the CO oxidation rate increases in the order 30 > 5 > 2 nm, in the same order as in the case for the oxidation of a CO monolayer in Ar purged solution.

5150

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

Comparison between CO vibrational properties in CO free solution (Fig. 2) and CO-saturated solution (Fig. 4) reveals some differences as well as some similarities: (i) the CO band position for all catalysts in CO-saturated solution blueshifts by about 4–5 cm−1 relative to CO-free solution, consistent with the formation of more compact CO adlayer in the presence of CO (i.e., a weaker interaction of CO with the Pt-surface) which, in turn, would effect (increase) the dipole–dipole coupling within the CO-adlayer; (ii) although the Stark-tuning slope (dν/dE = 25 cm−1 /V) is identical in CO-free and in CO saturated solution, in the former electrolyte a linear dν/dE dependence is observed in a much wider potential region; (iii) the CO2 production, expressed as integrated intensities of the CO2 band as a function of the electrode potential, is much more pronounced in CO saturated solution than in argon saturated solution, which is consistent with a positive reaction order with respect to the CO partial pressure. 3.4. CO bulk oxidation: IRAS versus RDE measurements Based on the above IRAS analysis, we can conclude that the rate of CO bulk oxidation increases with decreasing oxophilicity of the nanoparticles, i.e., by increasing the particle size and the number of defects on which OH adsorption may take place. Surprisingly, we found a positive reaction order with respect to the CO partial pressure in both the pre-ignition as well as the ignition potential regions. In our previous studies of CO bulk oxidation on Pt(h k l) electrode by the RDE methodology, however, we found that the reaction order with respect to the CO partial pressure can be either positive (pre-ignition potential region) or negative (the ignition potential region), depending on the nature of CO on the platinum surface [5]. In order to resolve these apparent differences, the RDE and FTIR measurements summarized and compared in Fig. 5 will be discussed below. Fig. 5a shows that in RDE measurements, recorded on “oxide-annealed” and “CO-annealed’ surfaces, three different potential regions can be distinguished, i.e., the so-called pre-ignition potential region (0.55 < E < 0.9 V) which is followed by the ignition potential (ca. 0.9 V) and finally by the diffusion limiting currents (E > 0.9 V). The continuous supply of CO to the electrode surface, characteristic for the RDE, leads to a positive shift in the onset of CO bulk oxidation relative to CO stripping experiments, as was the case for the extended surfaces [5]. As expected, therefore, the onset for the development of both the pre-ignition potential region and the ignition potential increases by 0.2–0.3 V compared to the stripping peak depicted in Fig. 2. This shift arises from the competition between CO and OH for the same active sites. The consequence of a fast repopulation of active sites by CO is that, there is not a big difference in activity between CO oxidation on oxide-annealed versus CO-annealed 1 nm particles, although the former is more defected. In particular, the ignition potential is shifted positively on the “CO-annealed” surfaces only by ca. 15 mV, consistent with the supposition

Fig. 5. CO bulk oxidation on the 1 nm catalyst (a) with a RDE at 1600 rpm in CO saturated perchloric acid electrolyte (sweep rate: 1 mV/s, T = 298 K). Curves are shown for an oxide-annealed (dark line) as well as a CO-annealed (light line) electrode. Inset (b): magnification of the pre-ignition potential region. The FTIR results for the same catalyst in CO saturated solution are depicted in (c ) CO band frequencies and (c

) integrated CO2 peak areas, analog to Fig. 3.

that “CO-annealed” surfaces contain fewer “irregularities” than the respective surfaces that were never pretreated in CO saturated solution. Following the ignition potential, the diffusion limiting currents are clearly discernable between 0.92 and 1.0 V. On the reverse sweep almost independent on the pre-history of the electrode a sharp deactivation is observed at E < 0.8. Interestingly, in contrast to Fig. 5a, but in line with the CO stripping experiments of Fig. 2, the CO2 production in COsaturated solution is strongly affected by the methodology applied for surface preparation. Fig. 5b reveals that the oxideannealed surface is unambiguously more active than the COannealed surface, suggesting that in FTIR experiments, due to slow repopulation of surface by CO, the defect sites on the former surface are more available for the adsorption of OH. Consequently, a less defected CO-annealed surface is less active for CO oxidation, as confirmed with the integrated

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

5151

intensities for CO2 production (inset c

) and the potentialdependent redshift deviations from the normal Stark-tuning slope (inset c ). Clearly, the thin layer configuration of the FTIR cell applied in spectroelectrochemical measurements does not provide experimental conditions that are identical to the RDE measurements. In particular, in contrast to the well defined mass transport of CO molecules as well as the thickness of the diffusion layer, which is characteristic for RDE measurements, in an IRAS setup the thickness of the diffusion layer and thus the transport of CO molecules is not well defined (limited supply of CO in unstirred solution) and, thus, the conditions in a thin-layer configuration does not represent a true steady state behavior for CO adsorption/oxidation, which can be achieved in RDE measurements. It is therefore misleading to use FTIR measurements to determine the true catalytic activity of bulk CO oxidation. 3.5. Bulk CO oxidation on an oxide precovered surface We found in Fig. 5a that if the surface is completely covered by CO, the difference in catalytic activity in the pre-oxidation potential region and at the ignition potential is negligible. Interestingly, in gas-phase catalysis McCarthy et al. [34] found that under high CO partial pressure conditions (equivalent to our experimental conditions when surface is completely covered by CO) the specific rate is insensitive to the particle size but structure sensitive for low CO partial pressure. Under the latter conditions the specific rate is higher for low area (sintered) Pt. It has been suggested that the observed deactivation with decreasing particle size stems most likely from the build-up of a non-reactive oxygen species, in a manner similar to that observed by Ostermaier et al. [35] for ammonia oxidation over Pt. In electrocatalysis, it has also been found that under different experimental conditions either a non-reactive oxide (E > 1.1 V) or reactive oxide (termed as OHad ; E < 1.1 V) can be built up on the surface [11]. One example of how the nature of Pt oxide may affect the rate of CO bulk oxidation is summarized in Fig. 6 for a massive Pt electrode. Starting at 0.8 V and sweeping the potential positive (1 mV/s) a sharp raise in the CO oxidation current is followed by almost immediate deactivation, so that at ca. 1.6 V the CO oxidation current is negligible. If, however, the potential sweep was reversed at less positive potentials then small but clearly visible CO oxidation currents are still observed on the negative sweep. Notice that on the reversed sweep the diffusion-limiting-like plateau is followed by a sharp increase in the current, which is then followed by an equally fast deactivation. The shape of the I versus E curves is determined by the nature of oxide that can be formed above 1 V. Namely, the formation of more reactive oxygenated species (at less positive potential) will allow higher activity, but on the surface oxidized to a higher degree the position of the peak maximum shifts to more negative potentials. With the highest positive limit of 1.8 V the peak almost vanishes, because the surface

Fig. 6. CO bulk oxidation on CO-annealed polycrystalline Pt with a RDE at 1600 rpm in CO saturated perchloric acid electrolyte (sweep rate: 1 mV/s, T = 298 K). Polarization curves for three different positive potential limits are demonstrated.

transforms virtually direct from an oxide-blocked into a COblocked state. In order to test how the build-up of reactive oxygen affects CO oxidation on Pt nanoparticles in an electrochemical environment, polarization curves were recorded upon sweep reversal at 1.0 V, where the surface is predominantly covered by reactive oxygenated species. The results in Fig. 7 clearly show that under these experimental conditions the CO oxidation reaction is “demanding”, and the activity of Pt nanoparticles increases in opposite order to that observed in CO stripping experiments, i.e., 30 < 5< 2 < 1 nm.

Fig. 7. Comparison of the activity of CO bulk oxidation on the 1, 5 and 30 nm catalyst (CO-annealed) in the negative going sweep with a RDE at 1600 rpm in CO saturated perchloric acid electrolyte (sweep rate: 1 mV/s, T = 298 K). For clarity the positive potential sweep is shown only the 30 nm catalyst.

5152

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

Comparing the RDE and the FTIR results it appears that small particles are able to exhibit either higher or lower specific activity for CO oxidation than larger ones depending upon the reaction conditions. For example, because the isosteric heat of CO adsorption on Pt(h k l) is relative insensitive to the surface structure [11] it is reasonable to suggest that if the surface is completely covered by CO (“high pressure conditions”) the Pt–CO energetics will play negligible role in the kinetics of CO oxidation. By contrast, the number of defects on which the initial adsorption of reactive OH will occur is very important, so that the rate of oxidative CO removal in CO stripping experiments increases with the availability of those sites, i.e., 1 < 2< 5 < 30 nm. However, under experimental conditions in which the surface is predominantly covered by OHad the reversed order in activity is obtained, consistent with the fact that the heat of oxide formation increases by increasing the number of low-coordinated surface atoms. Clearly, if specific sites on the surface are active for OH adsorption then the electrocatalytic activity for CO oxidation changes as the concentration of these sites changes with the particle size, in agreement with gas-phase catalysis [36].

4. Summary and conclusion We will now complement the obtained results from electrochemistry with the various effects experienced in vacuum and gas-phase catalysis for the same reaction under comparable conditions from literature (extended Pt surfaces and high CO partial pressures), and thereby pointing out the striking similarities and differences. Unfortunately, studies of nanocatalysts in gas phase cannot be correlated in a straightforward manner to conclusions drawn from our work on carbon-supported Pt, since they are generally conducted with catalysts on oxide support, which is known to have decisive influence on the reactivity [37]. Applying two different preparation procedures, oxide- and CO-annealing, the role of surface irregularities or defects could be elucidated in CO monolayer [25] as well as in “bulk” CO oxidation. (I) We suggested that on the CO precovered surface the active sites are defects on which the dissociative adsorption of water is facilitated in acid electrolyte. It has been shown that while the onset potential of CO oxidation is almost independent of the particle size, the rate of CO2 production is strongly dependent of the particle size, i.e., 1 ≤ 2 < 5  30 nm. We suggest that the oxidative removal of CO is mainly controlled by the number of “defects”, which in an ideal cubo-octahedral particle may serve as an active center for OH adsorption. It was also found that, the onset of CO2 production is accompanied either by redshift (d > 5 nm) or anomalous blueshift (d < 2 nm) deviations from the normal Starktuning slope. The interpretation of the “anomalous” dνCO /dE behavior is discussed in relation to similar changes observed at the Pt(1 1 1)–solution interface,

i.e., in terms of slow CO oxidation, co-adsorption of anions and concomitant CO compression into small islands on fairly smooth (1 1 1) facets. Accordingly, in high-pressure studies of CO oxidation on Pt(1 1 1) in UHV [38] it was demonstrated that below a certain temperature the active CO species are those adsorbed at “non-registry” sites or defect (or distorted) Pt surface sites. Moreover, the reaction rate increases linearly with the surface concentration of those sites, similar to our qualitative results for CO-striping from Pt nanoparticles. (II) In a CO bulk oxidation experiment, i.e., when CO is continuously supplied to the surface, the effect of defect sites on the rate of CO oxidation is attenuated because readsorbing CO can effectively block the active center required for OH adsorption. As a consequence it was found that in the preignition potential region, CO bulk oxidation is almost independent of the particle size. For comparison, in UHV the effect of preadsorbed CO on O2 adsorption was described by Su et al. [38]. For Pt(1 1 1) fully covered by CO, the CO oxidation is completely suppressed below the CO desorption temperature, independent of the partial O2 pressure. A surface with such a densely packed overlayer is no longer able to dissociatively chemisorb oxygen, which is the reactive intermediate in gas phase CO oxidation, as compared to dissociatively adsorbed H2 O in acid electrolyte. However, once the CO desorption temperature is reached, referred to as the ignition temperature, more and more adsorption sites for oxygen become accessible and the oxidation rate becomes proportional to the oxygen concentration and proceeds rapidly. In analogy to the ignition temperature, at the ignition potential, where the strong Pt–OH interaction requires a small shift in thermodynamic driving force to dissociate water (or even to displace some CO from the defects), oxidation of the adsorbed CO monolayer occurs rapidly and CO bulk oxidation becomes self-sustained. The importance of defect sites is, as in the preignition region, relatively small (gain of only 15 mV). (III) In the potential region beyond the ignition potential the surface is predominantly covered by oxygenated species and the limiting step in the CO oxidation is the diffusion of dissolved CO to the electrode surface. However, at high enough potentials oxides become so strongly bonded to the Pt surface so that they can block the adsorption of CO and therefore inhibit the progress of the reaction. The surface oxide formation on Pt in an electrochemical environment, which was extensively described in electrochemistry by Conway [39], is very similar to the one in the gas-phase at elevated temperatures. For instance, Niehus and Comsa [40] demonstrated the decisive transformation of a simple chemisorbed state to an oxide state on Pt(1 1 1) during oxygen exposure above 800 K, where even subsurface oxide is formed. Concomitantly, other groups

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

[41] found a decrease in the CO oxidation activity under high temperature and oxygen partial pressure conditions independent of the reactive gas mixture, which is related to the blocking of surface sites by inactive (sub-surface) oxide. Again, instead of temperaturecontrolled formation of oxide on metal surfaces in UHV, at the solid liquid interface we can use the electrode potential to control the fractional coverage and the nature of adsorbed oxygenated species (reactive versus active). (IV) Under experimental conditions where the surface is initially covered with active OH species, which is the case in the cathodic sweep from 1.0 V, the electrocatalytic activity of supported Pt catalysts for CO oxidation follows closely the change in oxophilicity with the particle size, i.e., 30 < 5 < 2 < 1 nm. As already mentioned, the particle size effects observed in our work and on oxide-supported catalysts in gas phase are not comparable in a straightforward manner [37]. However, Ertl [42] demonstrated, by utilizing the UHV STM, that on an oxide-covered Pt(1 1 1) surface the initial CO oxidation is determined by adsorption of CO on the defect sites. The reaction further propagates on the domain boundaries of CO and oxide covered phases, instead of randomly reacting in mixed CO-oxide covered phases, i.e., the reaction rate is proportional to the domain boundaries. With the progressive consumption of Oad by the oxidation reaction, the empty sites become occupied by impinging CO molecules (in the absence of O2 in the molecular beam), which form continuously growing CO patches. If, however, oxygen would be present in the molecular beam, a competition for the extricated sites would occur, depending on the relative binding energies. The same is most likely true in an electrochemical experiment, in which the infinite source of oxygen (in acid electrolytes water) and dissolved CO will constantly compete for the same active sites. In contrast to UHV, however, in an electrochemical environment there is a third parameter in the reacting scheme, anions from the supporting electrolyte, which even among many electrochemist are erroneously ignored, for details on the anion effects on the rate of electrochemical reaction see Ref. [1]. Finally, the assortment of the presented results were intended to demonstrate the high degree of complexity of such an apparently simple reaction like the CO oxidation reaction on Pt. Varying reaction conditions can alter the macroscopic kinetics so extensively, that electrocatalytic activity sequences for the here investigated carbon-supported high surface area catalysts of either 30 > 5 > 2 > 1 nm, 30 = 5 = 2 = 1 nm or 30 < 5 < 2 < 1 nm can be established. Only the understanding of the elementary processes occurring on the atomic scale at the catalyst surface can give further insight into phenomena related to catalytic reactions. From this perspective, it is essential to always consider the variations of the

5153

catalyst surface depending on the experimental conditions, before the reactions thereon are analyzed and discussed.

Acknowledgements This work was supported by the Director, Office of Science, Office of Basic Energy Sciences, Division of Materials Sciences, U.S. Department of Energy under Contract No. DE-AC03-76SF00098. K.M. acknowledges his supervisor Prof. Christoph Fabjan at the TU Vienna and the Austrian BMBWK for a Ph.D. scholarship. M.A. acknowledges the A.v. Humboldt Foundation for a Feodor Lynen scholarship. Furthermore we would like to thank Adams & Chittenden Scientific Glass for the electrochemical cells, and T. Tomoyuki from TKK as well as Dr. R. Atanasoski from 3 M for the supply with the catalyst samples.

References [1] N.M. Markovic, P.N. Ross, Surf. Sci. Rep. 45 (2002) 121 (and references cited therein). [2] A. Wieckowski, E.R. Savinova, C.E. Vayenas (Eds.), Catalysis and Electrocatalysis at Nanoparticle Surfaces, Marcel Dekker, New York, 2003. [3] I. Villegas, M.J. Weaver, J. Chem. Phys. 101 (1994) 1648. [4] K. Yoshima, M. Song, M. Ito, Surf. Sci. 368 (1996) 389. [5] N.M. Markovic, C.A. Lucas, B.N. Grgur, P.N. Ross, J. Phys. Chem. B 103 (1999) 9616. [6] N.M. Markovic, V. Radmilovic, P.N. Ross, in: A. Wieckowski, E.R. Savinova, C.E. Vayenas (Eds.), Catalysis and Electrocatalysis at Nanoparticle Surfaces, Marcel Dekker, New York, 2003. [7] S.C. Chang, J.D. Roth, Y. Ho, M.J. Weaver, J. Electron Spectrosc. Relat. Phenom. 54–55 (1990) 1185. [8] N.P. Lebedeva, A. Rodes, J.M. Feliu, M.T.M. Koper, R.A. van Santen, J. Phys. Chem. B 106 (2002) 9863. [9] N.M. Markovic, T.J. Schmidt, B.N. Grgur, H.A. Gasteiger, P.N. Ross Jr., R.J. Behm, J. Phys. Chem. B 103 (1999) 8568. [10] C.A. Lucas, N.M. Markovic, P.N. Ross, Surf. Sci. 425 (1999) L381. [11] N.M. Markovic, in: W. Vielstich, A. Lamm, H. Gasteiger, 1 (Eds.), Handbook of Fuel Cell Technology, John Wiley & Sons Ltd., New York, 2002. [12] N.M. Markovic, B.N. Grgur, C.A. Lucas, P.N. Ross, Surf. Sci. 384 (1997) L805. [13] K.A. Friedrich, F. Henglein, U. Stimming, W. Unkauf, Colloids Surf. A—Physicochem. Eng. Aspects 134 (1998) 193. [14] O.V. Cherstiouk, P.A. Simonov, E.R. Savinova, Electrochim. Acta 48 (2003) 3851. [15] F. Maillard, N. Eikerling, O.V. Cherstiouk, S. Schreier, E. Savinova, U. Stimming, R. Soc. Chem. Faraday Discuss. (2004) 357. [16] K.A. Friedrich, F. Henglein, U. Stimming, W. Unkauf, Electrochim. Acta 45 (2000) 3283. [17] K.A. Friedrich, F. Henglein, U. Stimming, W. Unkauf, Electrochim. Acta 47 (2001) 689. [18] O.V. Cherstiouk, P.A. Simonov, V.I. Zaikovskii, E.R. Savinova, J. Electroanal. Chem. 554 (2003) 241. [19] M.L. Sattler, P.N. Ross, Ultramicroscopy 20 (1986) 21. [20] M.K. Debe, in: W. Vielstich, A. Lamm, H. Gasteiger (Eds.), Handbook of Fuel Cells—Fundamentals, Technology and Applications, John Wiley & Sons, 2002, p. 576. [21] T.J. Schmidt, H.A. Gasteiger, G.D. Stab, P.M. Urban, D.M. Kolb, R.J. Behm, J. Electrochem. Soc. 145 (1998) 2354.

5154

K.J.J. Mayrhofer et al. / Electrochimica Acta 50 (2005) 5144–5154

[22] V. Stamenkovic, M. Arenz, P.N. Ross Jr., N.M. Markovic, J. Phys. Chem. B 108 (2002) 17915. [23] T. Iwasita, F.C. Nart, Prog. Surf. Sci. 55 (1997) 271. [24] V. Stamenkovic, K.C. Chou, G.A. Somorjai, P.N. Ross Jr., N.M. Markovic, J. Phys. Chem. B 109 (2002) 678. [25] M. Arenz, K.J.J. Mayrhofer, V. Stamenkovic, B.B. Blizanac, T. Tomoyuki, P.N. Ross, N.M. Markovic, J. Am. Chem. Soc. 127 (2005) 357. [26] Y. Takasu, N. Ohashi, X.-G. Zhang, Y. Murakami, H. Minagawa, S. Sato, K. Yahikozawa, Electrochim. Acta 41 (1996) 2595. [27] S. Mukerjee, J. McBreen, J. Electroanal. Chem. 448 (1998) 163. [28] H.A. Gasteiger, S.S. Kocha, B. Sompalli, F.T. Wagner, Appl. Catal. B, in press. [29] S.C. Chang, M.J. Weaver, Surf. Sci. 238 (1990) 142. [30] C. Rice, Y. Tong, E. Oldfield, A. Wieckowski, F. Hahn, F. Gloaguen, J.M. Leger, C. Lamy, J. Phys. Chem. B 104 (2000) 5803.

[31] M.W. Severson, C. Stuhlmann, I. Villegas, M.J. Weaver, J. Chem. Phys. 103 (1995) 9832. [32] M.W. Severson, M.J. Weaver, Langmuir 14 (1998) 5603. [33] S. Park, S.A. Wasileski, M.J. Weaver, J. Phys. Chem. B 105 (2001) 9719. [34] E. McCarthy, J. Zahradnik, G.C. Kuczynski, J.J. Carberry, J. Catal. 39 (1975) 29. [35] J.J. Ostermaier, J.R. Katzer, W.H. Manogue, J. Catal. 33 (1974) 457. [36] M. Boudart, Adv. Catal. 20 (1969) 153. [37] C.R. Henry, Surf. Sci. Rep. 31 (1998) 231. [38] X. Su, P.S. Cremer, Y. Ron Shen, G.A. Somorjai, J. Am. Chem. Soc. 119 (1997) 3994. [39] B.E. Conway, in: S. Davison (Ed.), Progress in Surface Science, Pergamon Press, Fairview Park, NY, 1984, p. 1. [40] H. Niehus, G. Comsa, Surf. Sci. 93 (1980) L147. [41] J. Segner, C.T. Campbell, G. Doyen, G. Ertl, Surf. Sci. 138 (1984) 505. [42] G. Ertl, J. Mol. Catal. A 182–183 (2002) 5.