By examining Pt-core–Ru-shell nanocatalysts of different compositions for the electro-oxidation of methanol, a volcano activity response is revealed according to Ru coverage. This activity profile can be accounted for by a bifunctional mechanism of spilling over the hydroxy species from Ru to Pt in close proximity with supplemental electronic and structural promotions. At high surface coverage of Ru on Pt, it is revealed that a new ‘direct’ pathway of Ru terraces on Pt sites in close vicinity can provide synergetic catalysis. Pt sites activate the methoxy surface species, which migrate to the Ru terrace to react with its surface oxygenates, from water dissociation, for accelerated CO2 formation through a ‘reversed’ spillover mechanism. This non-CO electro-oxidation route to CO2 on a Ru surface requires a lower potential to take place than the corresponding process on a Pt surface.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Development of direct methanol fuel cells (DMFCs) is currently an area of huge interest, as they are regarded to be more efficient power converters than combustion engines and batteries. They also produce much lower pollutant emissions.1 Platinum is the most active catalyst component in the DMFCs capable of breaking carbon–hydrogen bonds at mild conditions in methanol electro-oxidation.1 The mechanism of the reaction on platinum surfaces has been extensively studied, but the elementary steps are still far from clear. It is generally believed that the electro-oxidation takes place by a dual pathway mechanism with direct and indirect pathways.1–5 In the direct pathway, methanol is oxidized on the Pt surface to CO2 via a formate species, but currently there is little detail on this route and catalytic site for its formation. For the indirect pathway, methanol is stepwise dehydrogenated to COads that is strongly bound to the Pt surface. Only at high potentials can adsorbed hydroxyl from water activation assist its oxidation to CO2.1–5 By incorporating ruthenium into platinum, alloy catalysts are found to be the most active and tolerant to CO poisoning at lower potentials, owing to introduced beneficial electronic and geometric effects.5 Several groups have investigated the origin of the Ru enhancement by using bimetallic Pt–Ru electrodes,6 including PtRu alloy,6 Ru-decorated Pt electrodes, and Ru-decorated Pt nanoparticles.7 Much research effort worldwide has been spent studying the electrocatalytic activity of Ru-decorated Pt surfaces for the indirect CO electro-oxidation.6 Present knowledge concerning direct non-CO (formate) pathways is very limited. The direct pathway characterized at lower potentials, if selectively operated, would mean a delivery of higher current per methanol molecule oxidized.
Herein we have systemically synthesized, tested, and characterized a range of Pt-core–Ru-shell nanocatalysts of increasing Ru coverage, as compared to PtRu alloy, as a way of obtaining an understanding of both direct and indirect oxidation mechanisms on these Pt/Ru surfaces. We observe two direct pathways, one on the Pt terrace and the other on the Ru terrace, both of which can be promoted at the interface of Pt and Ru surface ensembles. Particularly, the latter pathway is not known, but can take place from 0.2 V vs SCE (0.4 Vpeak max.; SCE=saturated calomel electrode) on Ru terrace, 0.1 V lower than the direct pathway on the Pt terrace from 0.3 V (0.5 Vpeak max). Thus, a study of reaction mechanisms, nature of species and metal sites for the two direct methanol oxidation pathways, is presented.
Results and Discussion
Both platinum and ruthenium monometallic seed particles were first produced by using the chemical methods described in the Experimental Section. Similarly, a range of core–shell bimetallic nanoparticles was then synthesized through the initial generation of a Pt seed (core) from chemical reduction of a Pt precursor followed by subsequent coating with a Ru shell by a second reduction. The core–shell particles are described as Pt@Ru with a variation in the Pt/Ru ratio used in the synthesis. The TEM images in Figure 1 display typical colloid nanoparticles with a narrow size distribution (sd=0.8). The average size of the particles is progressively increased as higher Ru contents are used (Scheme 1). The growth of the Ru shell layers appears to be non-uniform with island-like decoration.
The X-ray powder diffraction (XRD) patterns of monometallic Ru and Pt and a range of Pt@Ru core–shell particles of different compositions were collected. Shown in Figure 2 A are the diffraction patterns of pure platinum and ruthenium on carbon. In the platinum spectrum, the major peaks around 39.9, 46.5, 67.9° can be assigned to the Pt (111), (200), and (220) planes, respectively. From the ruthenium spectrum, the main peak around 43° is from the Ru (200) plane. Platinum is known to form a fcc lattice, whereas ruthenium forms a hcp lattice. By using the Scherrer equation, the particle size can be calculated from the broadness of the peaks, it was found that the typical platinum particles are about 8.0±0.5 nm. XRD of alloy nanoparticles of PtRu 1:3; PtRu 1:1 and PtRu 3:1 on carbon were also collected (Figure 2 A). It is noted a significant right shift in the PtRu 1:1 and PtRu 1:3 spectra compared to pure Pt, which is attributed to the shortening in the PtPt bond distance with increasing the alloyed Ru, confirmed by our powder XRD. The calculated degree of alloying from the subtle change in lattice constant is significant in the case of PtRu alloys (typically 0.3150 derived from the 1:1 PtRu alloy). As Ru is added to the alloy, the Pt lattice undergoes compressive strain, shortening the PtPt bonds. This shortening of bond distances can cause a electronic effects, owing to lattice strain (d orbitals overlap to a larger degree6). In sharp contrast, the degree of alloying in the Pt-core–Ru-shell nanoparticles over a range of compositions is small, although the shift is in the same direction (Table 1), with a very small peak shift compared to pure Pt (Figure 2 B). Pt@Ru nanoparticles showed almost identical patterns to the parent Pt particles with the diffraction peaks only slightly shifted (Figure 2), whereas the synthesized PtRu alloys showed a more pronounced peak shift. There was no peak corresponding to Ru in the core–shell particles (Ru was expected to be reduced), indicative of the absence of long range crystallinity, as in form of thin coating.6 However, at the highest Ru content (Pt@Ru 1:3) one could see the high background baseline around the region of the (200) plane of Ru (43°), indicative of a degree of long-range ordering of Ru shells. The data for the calculated lattice constant, average size from peak (220) broadening, and the degree of alloying (atomic fraction of Ru) from lattice expansion, according to Vegard’s Law,8 of these core–shell particles are summarized in Table 1. The progressive increase in their sizes is significant with a low degree of alloying (0.0885 as compared to 0.3150 derived from the 1:1 PtRu alloy particles of comparable size), clearly suggesting their core–shell structure.
Table 1. Derived lattice constants, particle sizes, and degrees of alloying from XRD patterns of Pt-core–Ru-shell particles.
2 θ [°]
X-ray photoelectron spectroscopy (XPS) was carried out on these samples. For platinum the 4f peak was focused on for analysis because it has a high sensitivity factor of 4.4, and it is in an area that does not overlap with any other species expected in the sample. For ruthenium the 3p peak was used because even though the 3d peak has a higher sensitivity factor, and is therefore easier to detect, it overlaps with the very strong carbon 1s peak. Deconvolution and analysis of the 3d peak can lead to more inaccuracy in the results than using the weaker 3p peak. The 3p3/2 peak, which has a sensitivity factor of 1.3, was used because the 3p1/2 peak is less intense. The loading of the particles on to the carbon support cannot be determined from XPS because the support has a larger particle size than the sampling depth of the x-rays, so only the surface is detected. Shown in Figure 3 is a typical deconvoluted XPS spectra of Pt@Ru 3:1 Pt 4f region and Ru 3p region. Analysis of Table 2 reveals that Pt@Ru 3:1 consisted of a high proportion of Ru oxide species (Ru is sensitive to reoxidation), presumably formed during post-treatments of the samples for the removal of the PVP stabilizer. Thus, analysis of metallic Pt of the Pt 4f region only is conducted. Shown in Table 2 is the relationship between typical Pt 4f5/2 binding energy and Ru content of the core–shell nanocatalysts measured by XPS, for which the Pt 4f5/2 binding energy increases progressively with increasing Ru content, as those observed in PtRu alloys by Rigsby, et al.6 However, our PtRu 1:1 alloy nanoparticles show the highest degree of peak shift (Table 2). The shortening of the PtPt distance, as indicated in lattice strain (XRD), could lead to a subsequent downshift in the center of gravity of the d-band, with respect to the Fermi level, in order to maintain constant filling (Pt d-band is more than half filled) and to conserve energy.15 This can account for the core-level Pt 4f binding-energy shift. Also, a degree of charge transfer may also contribute to the overall binding-energy shift.
Table 2. XPS data for PtRu samples.
Pt/Ru 1:1 alloy
The nature of the surfaces of these core–shell particles was probed by CO chemisorption. The adsorption of CO (1 % CO in helium) on the surface of colloidal Pt, Ru, and PtRu core–shell nanoparticles were investigated by using ATR-IR spectroscopy. The spectra were acquired by using a Nicolet 6700 ATR-IR spectrometer with a liquid-nitrogen cooled MCT detector. A small drop of the test sample was placed on a smart golden gate-ZeSe (zinc selenide)/diamond crystal surface and evaporated at room temperature after its exposure to CO atmosphere. Analysis of Figure 4 reveals that pure Ru particles gave three CO adsorption peaks, namely, linear (2060 cm−1), multicarbonyl (1980 cm−1), and bridging mode (1930 cm−1). The Pt particles gave mainly linear CO adsorption (2040 cm−1) and a small degree of bridging CO (1850 cm−1), but no multicarbonyl species.9 The CO adsorption mode is known to depend critically on the type of metal, surface coverage, and nature of metal sites. In general, the surface of large crystallites comprising terrace sites will give rise to bridging CO mode. The open terraces or steps of the particle gave linear mode CO, whereas highly coordination-unsaturated sites on vertices, defects, kinks, or adatoms on a corrugated surface prevalently gave the multicarbonyl mode.10 Notably, the lowest level of Ru coating (Pt/Ru=3:1) created a surface with clear multicarbonyl species on Ru, with the linear CO mode shifted towards pure Ru. This shifted and broadened CO signal may contain different populations of adsorbed CO species on Pt and Ru and at their interface, as discussed by Maillard et al.,7 but the present resolution was unable to differentiate them. There was no bridging mode of Ru indicative of isolated or very small islands of Ru deposited on Pt rich surfaces (Ru/Pt surface ratio calculated from XPS; Table 2). For higher Ru content the surfaces became more Ru-like in character with the decreasing linear to multicarbonyl ratio (increasing Ru/Pt surface ratio; Table 2). The bridging mode on Ru became observable, whereas the same mode on Pt diminished. It is apparent that on Pt@Ru at a ratio above 1:2, the surface was basically Ru-rich with almost identical distribution of the three CO adsorption species. The TEM image of Pt@Ru 1:3 (Figure 1C) shows that the Pt was indeed embedded in Ru deposition. In sharp contrast, the PtRu 1:1 alloy showed a different CO adsorption spectrum to Pt@Ru 1:1. Although the linear and multi-carbonyl peaks are retained, both the bridging species of the two metals are clearly suppressed. XPS analysis showed that it contained a Pt rich surface (Table 2) similar to those reported by Rigsby et al.6 This absence of bridging CO implies that a good degree of Pt and Ru atomic mixing on the alloy surface (3:1) did not allow the formation of a large terrace of either metal.
Electro-oxidation of methanol over the core–shell catalysts
The catalytic activities for methanol electro-oxidation of the Pt@Ru with variation in Ru content as compared to pure Pt, Ru, and PtRu 1:1 alloy, were evaluated using cyclic voltammetry. Shown in Figure 5 is the characteristic methanol oxidation peak reaching a peak maximum at around 0.8 V over Pt on carbon (forward scans).1 Adsorption of OH from the dissociation of water (H2O+Pt↔PtOHad+H++e−) occurs above 0.3 V vs SCE over rough Pt surfaces.2 The formed PtOH is believed to oxidize the strongly adsorbed CO species to CO2 by the indirect mechanism (via methoxy, formyl, formate, etc) at this potential. Interestingly, there are clear progressive negative shifts in potential and enhancement in current density of this oxidation peak when Ru is coated on the Pt (from 3:1 to 2:1). The substantial attenuation in the current density over the Pt@Ru 1:1 is related to the decrease in the active Pt surface area, owing to its coverage by the extensive Ru islands (Ru rich surface). Further deposition of Ru leads to total encapsulation (TEM image and CO chemisorption), rendering the 1:2 and 1:3 samples to resemble the pure Ru surface. This volcano behavior can be clearly accounted for by the bifunctional mechanism of spilling over the hydroxyl species from Ru to Pt in close proximity, supplemented with electronic and structural promotions (lattice strain and charge transfer).6, 7 The peak with an onset potential from 0.2 V and peaked at 0.4 V over the Ru/C clearly demonstrates pure Ru surface can also catalyze oxidation of methanol with water, but the small peak size suggests a poor methanol activation by Ru surface despite the fact that the formation of RuOH is more facile than PtOH.11 It is very interesting to note a similar small shoulder peak at 0.5 Vmax over the Pt/C. These low-potential peaks of the two monometallic elements appear to be strongly associated with terrace sites. The shouldered CV peak was also previously noted by Herrero et al. using a Pt(111) electrode.12 PtRu 1:1 alloy gave marginally the highest current density peak as compared to Pt@Ru 3:1 and 2:1. However, there was no shoulder peak at 0.5 V for this alloy nor other alloy compositions tested (they also showed the absence of extensive terrace sites from the CO chemisorption). In contrast, the 0.5 V shoulder peak area was enhanced by 2.6 times in the Pt@Ru 3:1 sample in methanol oxidation with its Pt-rich surface containing extended Pt terrace sites (see the Supporting Information).
To demonstrate that the shoulder peak arises from methanol oxidation, cyclic voltammetry of the Pt@Ru 3:1 nanocatalyst, pre-adsorbed with CO, was examined. The experiment was first run between 0 V and 1.2 V at 0.05 V s−1 in 0.5 M H2SO4 until a stable CV profile was collected (about 10 scans), as a pretreatment. Pure carbon monoxide gas was then bubbled through the 0.5 M H2SO4 for 1000 s with the sample held at −0.15 V vs SCE. The sample was then transferred to a fresh 0.5 M H2SO4 solution, the sample was held at −0.15 V for 60 seconds then the cyclic voltammetry was run between −0.15 and 1.2 V with a scan rate of 0.05 V s−1. The first scan clearly shows the oxidation of CO (with the onset voltage similar to the main peak in methanol oxidation) and there is no peak at this position in the next scan so all adsorbed CO molecules are oxidized (Figure 6). This CO stripping experiment shows no shoulder peak at 0.5 V, implying that the shoulder peak in methanol oxidation is clearly derived from methanol adsorption.
Nature of the shoulder peaks
Pt is well known to activate methanol through adsorbed formate over an oxygen covered (111) terrace surface.13 Thus, this shoulder peak is attributed to the direct oxidation of adsorbed formate to CO2, which is more facile than indirect pathways through CO adsorption. The adsorption geometry of symmetrical formate species on the flat surface is consistent with the fact that the terrace sites were involved. Similar formate adsorption on Ru terrace over a pure Ru surface can be envisaged. In the case of Pt/C a peak maximum at 0.5 V was noted and for Ru/C a peak maximum at 0.4 V was recorded. It is not yet known how many Pt atoms or Ru atoms are required to form the extended “ensemble” for this direct pathway,[6.7] but it is apparent that the core–shell samples instead of the alloy samples are able to provide the appropriate surface ensembles for this route. Interestingly, analysis of Table 3 reveals that further covering the Pt particle with Ru can also significantly enhance the shoulder peak intensity at around 0.4 V (direct oxidation of formate on Ru terrace) as the surfaces are Ru rich (2:1 and 1:1, containing extended Ru terraces; Figure 4).
Table 3. Gaussian fitted CV (forward scan data).
Charge (from peak area) [mC cm−2]
48.0 (ca. 0.5 V)
124.0 (ca. 0.5 V)
48.2 (ca. 0.4 V)
44.0 (ca. 0.4 V)
28.0 (ca. 0.V)
124.0 (0.78 V)
120.0 (0.74 V)
126.0 (0.71 V)
70.4 V (0.65 V)
Adsorbed formate has been previously identified as a key surface species in methanol electro-oxidation.2, 4, 13 As a result, the formate concentration formed over Pt@Ru 1:1 sample at different potentials was monitored by ATR-IR. Thus, a series of Pt@Ru 1:1 samples (5 μL of Pt@Ru 1:1 on carbon as ink was coated onto the standard glassy working 3 mm i.d. electrode) were placed in a typical testing methanol/sulfuric acid solution for 60 seconds each at 0, 0.2, 0.5, 0.8, and 1.0 V vs SCE, respectively. The quenched samples were monitored by ATR-IR. Comparison of these spectra is made in Figure 7 and Table 4.
Table 4. Identification of the IR peaks of the Pt@Ru 1:1 samples quenched at various potentials.
Type of vibration
OCO stretching; carbonate (main), bicarbonate, formyl and asymmetric formate regions
Characteristic region of symmetric OCO stretching of formate
CO stretching; carbonate, bicarbonate, formyl and formate regions
It is clearly evident that Pt@Ru 1:1 started to generate a surface symmetrical ‘formate’ species (characteristic symmetric mode at 1340–1410 cm−1 region from direct pathway) from 0.2 V re SCE and depleted above 0.5 V (Figures 7 and 8). This contrasted with the other observed surface species possibly involved in the non-direct pathway (i.e., adsorbed formyl, carbonate, bicarbonate, and asymmetric formate at 1195 cm−1 and 1550–1700 cm−1), which depleted only at above 0.8 V. These observations matched with the general profile of the current density of the shoulder peak in the CV spectrum in methanol oxidation. It is known that the IR absorption spectra values of formate on Ru and Pt surfaces are similar to each other. The broad nature of the ‘formate’ peak in our samples precludes its clear assignment to the type of metal involved. However, the observed 1340–1410 cm−1 value in the Pt@Ru 1:1 sample matched more closely to 1382 cm−1 reported by Nakamura as the symmetric formate on Ru (0001) surface4 than the symmetric formate species on Pt (111) surface of 1320 cm−1 and 1328 cm−1 reported by Lin et al14 and Chen et al2 independently. Thus, the dramatic increase in this symmetric formate mode at 0.2 V clearly suggests that this is the on-set voltage for the direct oxidation of formate on Ru terrace. This confirms that the direct ‘formate’ oxidation by spillover mechanism is initiated at 0.2 V.
The use of Ru for methanol electro-oxidation is generally unexplored, as Ru is inert for methanol activation.14 However, this lower potential shoulder peak associated with Ru terraces can be promoted 1.5–2 times that of pure Ru, when Ru rich surfaces contain Ru terraces adjacent to the Pt sites. It is, therefore, evident from this work that there are at least two ‘direct’ non-CO pathways; one is on Pt and the other on Ru terraces. We propose that the surface spillover process on PtRu ensemble not only involves the transfer of oxygenates from Ru to Pt in one direction (forward spillover), but also methoxy or derivative from Pt to RuOH in the opposite direction (reversed spillover) in accordance with our data (Scheme 2).
To summarize, the major research worldwide on DMFCs is being placed in optimizing Pt-based catalysts with an additive such as Ru or Sn to provide oxygenated surface species by dissociating water at the sites at lower potentials. This leads to the accelerated CO2 formation from the indirect pathway through surface CO and the direct pathway through formate on Pt flat surface. Although Pt is primarily required for methanol activation, the use of an extended Pt terrace may not be a prerequisite. This work using core–shell modification reveals a new ‘direct’ pathway on Ru terraces based on a ‘reversed’ spillover mechanism from Pt sites. We hope this study can stimulate the design and development of new types of electrocatalysts for DMFCs with lower Pt content by creating active material interfaces, as such a fundamental possibility is clearly evident.
Synthesis of the nanoparticles
In a 50 mL round-bottom flask RuCl3 (0.033 g, 1.346×10−4 mol) and polyvinylpyrrolidone (PVP, 0.120 g, 1.080 mmol) were dissolved in diethylene glycol (DEG, 30 mL). The solution was then heated to 160 °C from room temperature at 6 °C min−1 in an oil bath under nitrogen with vigorous stirring. The color of the solution changed from dark red to light yellow and finally to dark brown. After 2.5 h, a transparent dark-brown solution of colloidal Ru nanoparticles was obtained. The Ru nanoparticles were cooled, precipitated in acetone, washed extensively with the same solvent and dried. Pt nanoparticles were synthesized in the same way as Ru except that a Pt precursor was used. Pt@Ru 1:1 nanoparticles were produced by using preformed Pt nanoparticles in colloid and treated with RuCl3 (0.033 g), PVP (0.012 g), and DEG (30 mL) in the two reduction steps. The Pt@Ru nanoparticles were precipitated in acetone, washed with the solvent, and dried. Core–shell nanoparticles with a range of Pt/Ru ratios were also synthesized accordingly. Similarly, PtRu alloys of different ratios were prepared in a single reduction step, but at a higher ramping rate of 20 °C min−1 (co-reduction). Supported nanoparticles were produced by using nanoparticles in colloidal solution from any of the above synthesis techniques. The required amount, to result in 20 % loading of the nanoparticles by weight, of pretreated Vulcan XC-72R carbon from Cabot was dispersed in ethanol (50 mL) by sonication: The carbon pretreatment was carried out by refluxing in 60 % nitric acid for 3 h, filtered, washed with distilled water, and air dried. The nanoparticle colloid was then added dropwise to the carbon support while being sonicated. The deposited carbon was collected from filtration, washed with distilled water and ethanol, and dried. Once supported, the PVP stabilizer was removed from the nanoparticles by heating the sample to 300 °C under nitrogen and held for 30 min in a thermogravimetric analyzer. Electrochemical measurements were performed by using a standard three-electrode set-up equipped with an Invium Compactstat potentiostat. The working electrode was a 3 mm2 glassy carbon electrode to which 1 μL of catalyst ink was applied, together with a platinum counter electrode and a saturated calomel electrode (SCE) reference electrode which created a potential of 0.241 V compared to standard hydrogen electrode, were used. The electrolyte solutions were a 0.5 M methanol/0.5 M H2SO4 and a 0.5 M H2SO4 solution. MiliQ ultrapure water, 99.5 % H2SO4 and analytical grade methanol were used for their preparation. The catalyst ink was prepared by mixing supported nanoparticles (5 mg) with prior removal of PVP, in ethanol (200 μL) and nafion (50 μL); this was then sonicated for a minimum of 15 min until the sample was well dispersed. The 3 mm2 glassy carbon working electrode with an area of 0.0706 cm2 was polished by 0.05 μm alumina before measurement. The catalyst ink (1 μL) was transferred by pipette onto the clean electrode and dried. The electrode was then pretreated, in 0.5 M H2SO4. In the pretreatment the electrode was cycled between −0.2 and 1.0 V at a 0.05 V s−1 sweep rate until the current was stable. Cyclic voltammetry was performed at room temperature to determine the activity of the catalyst. Using the three-electrode arrangement in a 0.5 M methanol/0.5 M H2SO4 solution the potential was cycled between −0.2 and 1.2 V (vs the SCE) with a scan rate of 0.05 V s−1. This was repeated for 20 cycles or until no change was observed. The samples with and without carbon support were extensively characterized by XRD, XPS, TEM, CO chemisorption, ATR-IR spectroscopy, and CV.
The ability of the catalysts to oxidize methanol was tested using cyclic voltammetry (CV), Figure 9 A is an example of a cyclic voltammogram, without methanol (the pretreatment), in 0.5 M H2SO4 with a 0.05 V s−1 scan rate. Figure 9 B shows example CVs with methanol in the presence of catalyst; two peaks can be seen in the forward scan at 0.5 V and 0.8 V; these are owed to the oxidation of methanol. From Figure 9 B it can also be seen that a steady state develops after repeated scans. The shoulder peak at 0.5 V is from the direct pathway for methanol oxidation on terrace sites whereas the peak at 0.8 V is from the oxidation of adsorbed species produced by the partial oxidation of methanol from the indirect pathway. The activity of the catalyst is measured by peak current density, the current per unit area of the electrode, of these peaks the higher the peaks, the more active the catalyst. As there were contributions to the forward CV scans from both the direct and indirect pathways over the Pt-based nanocatalysts, the Gaussian fitting showed clearly a deconvoluted lower potential shoulder peak together with the typical indirect CO oxidation peak. Although there might be a mixed contribution from the two direct pathways at 0.4 V and 0.5 V to the main peak at 0.6–0.8 V in some core–shell samples (especially the 2:1 sample), the peak deconvolution was only performed based on two mixed peaks for simplicity and accuracy reasons. The results are summarized in Table 3 and all forward CV scans are shown in the Supporting Information.
The authors wish to thank Johnson Matthey for providing samples and advice, Dr G. Q. Li for performing TEM analysis and Dr K. M. K. Yu for technical support on this work.