J. Phys. Chem. 1994, 98, 11011-11014
11011
Hydrogen Adsorption on Platinum Particles Studied by 195PtNMR Y. Y. Tong and J. J. van der mink* Institut de Physique Exptrimentale, Ecole Polytechnique Ftdtrale de Lausanne, PHB-Ecublens, CH-1015 Lausanne, Switzerland Received: June 22, 1994; In Final Form: August 23, 1994@
A 4.4 wt % platinum on anatase catalyst with average particle size 1.7 nm is studied by 195PtNMR after increasing dosing with hydrogen (Wsurface Pt ratios of 0.1,0.5, and 1). Both spectral changes and variations of TI at the maximum of the spectrum are observed. The relaxation curves are not single exponential but show Korringa-type timehemperahre scaling, indicating that all platinum atoms are in a metallic environment. It is concluded that there is a very large distribution of local densities of state at the Fermi energy on different surface atoms and that the values are still lower than found previously.
Introduction Because of its decisive importance in many molecular rearrangements and catalytic reactions and of its widespread use in the determination of the dispersion of transition metal catalysts, the hydrogen-transition metal system is one of the most studied chemisorption systems.' The experimental work is often complicated by the fact that hydrogen is a rather difficult species to detect by most of the regular techniques of surface science.2 Several nuclear magnetic resonance (NMR) studies of hydrogen-covered supported platinum catalysts have been r e p ~ r t e d . ~Most - ~ of these use 'H NMR; the major observation is that the peak due to hydrogen chemisorbed on the metal shifts from -50 ppm at low equilibrium pressure to -20 ppm at high pressure. Using lg5PtNMR, Bucher et al.6*7studied a m i 0 2 catalyst with saturation coverage of hydrogen and demonstrated that certain electronic properties of the platinum particles can be deduced from the data. Specifically, they concluded that the local densities of states at the Fermi energy (LDOS) on "typical" metal surface sites (and even on subsurface and sub-subsurface sites) are considerably influenced by the hydrogen chemisorption. Such results are of some interest in certain descriptions of the chemisorption where attention is focused on the Fermi level electrons, since these can be rearranged at low energetic cost. Here we report a similar, but more detailed, 195PtNMR study that shows the increasing influence of the presence of hydrogen on the platinum surface LDOS (we use a fairly high-dispersion catalyst, so that most of the NMR signal comes from surface atoms) when the coverage is increased from 10% to 50% and to 100%.
Experimental Section The starting material to prepare our samples is from the same batch as sample 3 in ref 10. It has also been used in our work on alkali-promoted catalysts." It is 4.4 wt % Pt on anatase; the dispersion, which is the ratio (surface atoms Pt)/(total atoms Pt), was determined from electron microscopy to be 60%, and we refer to the material as Pt-60. The samples were prepared as described before,6 except that the quantity of adsorbed hydrogen was determined by a volumetric method. The coverages, in terms of (atoms H)/(surface atoms Pt), were lo%, 50%, and loo%, and the samples are labeled Pt-60-H10, Pt60-H50, and Pt-60-H100. A reference sample with a nominally @Abstractpublished in Advance ACS Abstracts, October 15, 1994.
0022-3654/94/2098-11011$04.50/0
clean surface" was also prepared and designated Pt-60-clean. The NMR equipment was the same as before. Nuclear spinlattice relaxation times were determined by a variant of the saturation recovery technique, using progressively diminishing pulse intervals inside the saturation comb and a composite pulse as its last pulse (seven nL2 pulses followed by ( ~ / 4 ) - ~ ( n / 2 ) + ~ (nL?)+,(n/4)+,). Recovered magnetization was detected by (n/ 2)-(n)-spin echo. Between 40 000 and 720 000 scans were accumulated, depending on experimental details. The relaxation curves were not single exponential (except for Pt-60-clean). At low temperature, only the faster relaxing signal was measured with precision to avoid excessively long measurement times. At high temperature the full relaxation curve can be recorded and shows clear nonexponentiality. At 110 K we also performed a TI measurement on a sample used earlier (sample LTRH of ref 8; here we will denote it Pt-36-Hsat); the data were fitted to a double exponential, constraining the amplitude ratio to be equal to that found for Pt-60-H100. The fitted results are in excellent agreement with those of Pt-60-H100. At 80 K, the 195PtNMR spectra of Pt-60-clean, Pt-60-H10, Pt-60-H50, and Pt-60-H100 (see Figure 1) were measured by the point-by-point spin-echo method. The pulse repetition rate was 33.3 s-l, which implies some saturation in the surface region of the spectrum for the hydrogen-covered samples.
Results and Discussion Our usual analysis of 195PtNMR data of catalysts is based on two sets of assumptions: one concerning the relation between NMR parameters and certain electronic properties (the local densities of state (LDOS) description6) and the other concerning a relation between geometrical site statistics derived from the particle size distribution as observed by transmission electron microscopy and the shape of the NMR spectrum (the NMR layer model7). The LDOS description assumes that variations in Knight shift ( K ) and in spin-lattice relaxation time (TI)from one nuclear spin site to another are mainly determined by variations in the intensity of electronic wave functions at the Fermi energy. (The intensity is averaged over the atomic cell containing the nucleus.) It is furthermore assumed that sand d-like local densities of state (and their associated hyperfine fields) can be distinguished. It then follows that at a single resonance frequency (fixed value of K ) one might find signals from nuclei with many different combinations of s- and d-like LDOS. Each such combination would give rise to the same K , but a different TI. Therefore, generally, the spin-lattice relaxation curves measured at a certain resonance frequency 0 1994 American Chemical Society
11012 J. Phys. Chem., Vol. 98, No. 43, 1994
Letters 1.0 0.8 0.6 0.4
0.2
1.0
d 0.8
.--a 0.6 4
i4 Q*4 0.2 I
I
L
0.01
0.1
1
10
(d) R-60-HI00
FieldFrequency (GkHz) Figure 1. 195F’t N M R spectra obtained by the point-by-point spin-echo method. Panels (a)-(d) correspond to samples Pt-60-clean, Pt-60-H10, Pt-60-H50, and Pt-60-Hl00, respectively. All spectra were recorded at 80 K and normalized to their area. The hydrogen-coveredspectra are not fully relaxed under their recording conditions (see text). should be nonexponential. In the earlier work from our laboratory6 such nonexponentiality has not been observed to within the experimental precision, and the data have been discussed in terms of a “most likely site” that dominates the NMR behavior at a given resonance frequency. The NMR layer model7 assumes that the spectrum can be decomposed into a sum of Gaussians of different widths and center frequency, each Gaussian describing the signal resulting from a certain layer of atoms (surface layer, subsurface layer, and so on). The sequence of center frequencies is taken as exponentially approaching the bulk value (a similar variation for neighboring sites to an impurity in dilute bulk alloys is rather well documented12),with a characteristic length called healing length; the widths are left as fittable parameters that mimic the diversity in sites within a layer. The actual number of atoms that make up a given layer is determined from TEM micrographs of the sample and from layer statistics for cubooctahedral particles. The integral of each Gaussian is proportional to the number of atoms in the layer. It has been supposed7 that hydrogen chemisorption does not perturb the geometrical layer statistics but that it changes the difference between surface and bulk densities of states and also the healing length. Our recent more precise TI measurements still show singleexponential decays for the signal from clean platinum surfaces and also for that from alkali salt-promoted surfaces.ll However, in the case of (partially) hydrogen-covered surfaces, we now detect nonexponential decays. (For R-36-Hsat the initial values can still be well described by the earlier reported single 7‘1.) The decay curves show very good time-temperature scaling, as required by the Korringa mechanism for spin-lattice relaxation by conduction electrons in a metal. For this relaxation mechanism, the nuclear spin-lattice relaxation rate TI-’ is proportional to temperature I“, so that T I T = C. At a given
0.01
0.1
1 7.T (s*K)
10
100
Figure 2. Time-temperature scaled spin-lattice relaxation curves for (a) Pt-60-Hl0, (b) Pt-60-H50,and (c) Pt-60-Hl00,taken at the maxima in Figure 1. The curves measured at different temperatures collapse into one, as required by the Komnga mechanism (see text). The solid lines show a two-exponential fit. The fitted parameters are collected in Table 1. The normalized time-temperature scaled curve of Pt-36Hsat recorded at 110 K is also shown in panel (c) (symbol D) . It coincides very well with the others. resonance position and temperature, we measure a series of recovered signal amplitudes Ai as a function of the relaxation interval zi. When the Ai are normalized by the fully relaxed amplitude, the Bloch equation is 1 - A i= exp(-z,/T,)
(1)
and using the Komnga relation, one has
If the relaxation curve is a sum of N different exponentials, there are N different constants C, but still when plotted as normalized Ai versus (ziT), curves taken at different temperature collapse into one, as shown in Figure 2. So, while to our precision all surface platinum atoms stay in the metallic state, we have now, in terms of the NMR layer model, several different “most likely sites” in the surface region of the spectrum. We find that, at the resonance frequency of the maximum of the surface peak, the relaxation curves can be described by a sum of two exponentials, with temperatureindependent amplitude ratios (solid lines in Figure 2; the fitted parameters are collected in Table 1). Similar behavior is found at other frequencies; for catalysts with the dispersion used here, only a small part of the signal comes from bulklike atoms. While it is clear that nuclei in at least two different environments resonate at a given frequency, it is of course impossible to demonstrate that there are not more than two environments.
Letters
TABLE 1: Final Parameters of the Two-Experimental Fir Pt-60-HlO Pt-60-H50 Pt-60-H100 fast TIT 0.060 & 0.005 0.071 & 0.005 0.112 & 0.008 fraction 0.73 0.61 0.63 slow TIT 2.73 f 0.64 7.76 f 0.95 6.94 f: 1.01 fraction 0.27 f 0.02 0.39 f:0.01 0.37 f 0.01 rms deviation 0.05 0.03 0.04 We fitted all curves of a given sample with one two-exponential function. The root-mean-square deviations of the fits are referred to the fully relaxed amplitudes at each temperature (unity on the vertical scales of Figure 2). The units of T I T are s*K. Therefore, one should be very cautious in interpreting the amplitude ratio (in Table 1) as a ratio of “site frequencies”. In terms of the layer model, we should distinguish “horizontal” and “vertical” effects of chemisorption on the LDOS. The “horizontal” effect is a change of NMR parameters of atoms in the surface layer (their number is supposed to be independent of the chemisorption), and the “vertical” effect is a change in healing length (the vertical distance it takes to find back an essentially bulk LDOS). Already in the clean-platinum slab Knight shift calculations of Weinert and Freeman13that contain exactly three well-defined different sites (the (100) surface, subsurface, and sub-subsurface layer), huge shifts between these sites are found. No such calculations are available for hydrogencovered surfaces, and usual chemisorption calculation^^^ assume regular overlayers of adsorbate, so that there is only a single type of metal surface atom for a given overlayer. In a real catalyst, different crystal faces are simultaneously exposed, and there may also be size-dependent effects, so that the adsorbate is not uniformly distributed over all initially “clean” surface sites seen by NMR. Then, some signals in the surface peak may be more strongly affected by chemisorption then others (variations in “horizontal effect”). In the sample with 100% coverage there are probably no platinum surface sites “far away” from a hydrogen chemisorption site. Nevertheless, we find nonexponential spin-lattice relaxation at the frequency of maximum signal amplitude. To arrive at an estimate for the widths and for the average values of the LDOS distribution on the surface sites, we use a procedure that is slightly different from that in ref 7. In that paper, the variation of the (single-exponential) spin-lattice relaxation across the part of the spectrum ascribed to the surface layer was used to deduce these quantities (see Figures 5 and 6 of ref 7). Here we will consider the relaxation curve measured at the frequency of the maximum in the spectrum as representative of the surface sites: we neglect the distribution of Knight shifts of the surface sites and take the range of their spinlattice relaxation times as lying between the “fast” and the “slow” values of Table 1. In earlier work from our laboratory,6 we have proposed equations that express Knight shift and relaxation rates in terms of the s- and d-like local densities of states, with parameters that give very satisfactory agreement for the NMR of bulk platinum and that have been used for consistent interpretations of the NMR data for platinum catalyst^.^,^.^^ We use these equations to obtain the LDOS (given in Table 2) corresponding to the fast and the slow relaxation values of Table 1. We found that there was no exact solution for the slow components in Pt-60-H50 and Pt-60-Hl00. This is probably related to the fact that some of the parameters (fitted to bulk platinum) are not completely appropriate here. The solution adopted was to look for the best fit while constraining the d-LDOS to be zero. (When left free, we found unphysical, small negative values.) The range between the LDOSs of the slowly and the rapidly relaxing nuclei is now taken as a measure of the width of the LDOS distribution over
J. Phys. Chem., Vol. 98, No. 43, 1994 11013 TABLE 2: Local Density of States Obtained from the Data in Table 1“ LDOS s-like
sample Pt-60-clean Pt-60-H10 Pt-60-H50 Pt-60-H100
fast
slow
4.84 4.68 4.0
0.88
0.74 0.79
d-like fast slow 11.88 10.24 8.25
2.49 0 0
average s-like d-like 3.95 11.04 3.77 9.34 3.14 6.25 2.81 5.18
Calculated according to ref 6. The units are number of states per Rydberg and per atom. The data for Pt-60-clean are from ref 11. The averages are weighted with the fast and slow fractions mentioned in Table 1. all surface sites. To obtain the average values mentioned in Table 2, the fast and slow LDOS have been weighted by the corresponding fractions of the double-exponential fits in Table 1.
As a measure of the “vertical effect” (mainly the change in healing length) of chemisorption of progressively more hydrogen, we propose to use the fraction of nuclei resonating above 1.115 G/kHz. In Figure 1, these fractions are, from top to bottom, 0.31,0.28,0.18, and 0.15. If fully relaxed spectra had been recorded (see Experimental Section), the fractions for the H-covered samples would have been smaller due to a relative increase in the contribution from the low-field surfacelike region. It is seen from Figure 1 that the “surface Gaussian” shifts with increasing hydrogen coverage from 1.100 G M z for the clean surface to 1.098 G k H z (at 10% coverage) and finally to 1.096 GkHz (for 50% coverage or more) while its half-width decreases. The usual procedure to obtain the width of the surfacelike peak is to fit its low-field side only by either half a Lorentzian15 or half a G a ~ s s i a nwe ; ~ have not undertaken such a refinement. In Figure 1, the low-field half-half-widths with respect to the peak position are, from panel (a) to (d), 0.0095, 0.0054, 0.0045, and 0.0045 G M z . The peak is already narrowed about 43% for 0.1 coverage. Our earlier analysis (based on single-exponential fits to relaxation data for Pt-36Hsat) implied that, although on a NMR shift scale the surface peak narrows after chemisorption, the associated distribution of LDOS actually widens (from about 3.3 to about 5.1 Ry-’ The present more detailed results (Table 2) show an even bigger widening. It means that on hydrogen-covered metal surface the detailed LDOS changes dramatically when moving from one “typical” environment to another. On the high-field side the half-half-widths are 0.0095,0.0072,0.0081, and 0.01 17 GkHz going from panel (a) to (d) in Figure 1. The increase of the asymmetrical broadening at the high-field side with increasing hydrogen coverage gives some support to the layer model: an increase of healing length pushes the subsurface signal further downfield, and the overlap of surface and subsurface signals becomes more important at the high-field side. Conclusion Compared to the earlier results of lg5Pt NMR on Pt-36-Hsat, the present study shows in more detail the effect of hydrogen chemisorption on the local density of states at the Fermi energy (LDOS) on a “typical” surface site. The range in LDOS values is even larger and its “typical” value still lower, than thought before. Already a 10% coverage of the surface of the 60% dispersion catalyst leads to visible changes in the spectrum, reflecting mainly an increase in healing length (the distance below the surface where a bulklike LDOS is retrieved). The “typical” surface site LDOS drops from 15.0 states Ry-’ atom-’ for the clean surface gradually to 8.0 states Ry-’ atom-’ for
11014 J. Phys. Chem., Vol. 98, No. 43, 1994 the 100% covered surface. This big change can be qualitatively compared to calculated values for the hydrogen-covered Pd (1 11) surface.l6
Acknowledgment. We thank I. Rodicio and M. Graetzel for providing the starting material for the samples. This research was supported by the Swiss National Science Foundation (Grant 21-31 127.91). References and Notes (1) For a review see: Paal, Z., Menon, P. G., Eds. Hydrogen Efeects in Catalysis: Fundamentals and Practical Applications, Marcel Dekker: New York, 1988. (2) Paal, Z.; Menon, P. G. Preface in ref 1. (3) De Menorval, L. C.; Fraissard, J. P. Chem. Rev. Lett. 1981, 77, 309.
Letters (4) Rhodes, H. E.; Wang, P.-K.; Stokes, H. T.; Slichter, C. P.; Sinfelt,
J. H. Phys. Rev. B 1982, 26, 3559. (5) Chesters, M. A.; D o h , A.; knnon, D.; Williamson, D. J.; Packer, K. J. J. Chem. Soc., Faraday Trans. 1990,86,3491 and references therein. (6) Bucher, J. P.; van der Klink, J. J. Phys. Rev. B 1988, 38, 11038. (7) Bucher, J. P.; Buttet, J.; van der Klink, J. I.; Graetzel, M. Sur$ Sci. 1989, 214, 347. (8) Feibelman, P.; Hamann, D. R. Sur$ Sci. 1985, 149, 48. (9) For a review see: Hoffmann, R. Rev. Mod. Phys. 1988, 60, 601. (10) Bucher, J. P.; Buttet, J.; van der Klink, J. J.; Graetzel, M. J. Phys. Chem. 1990, 94, 1209. (1 1) Tong, Y. Y.; Martin, G. A.; van der Klink, J. J. J. Phys.: Condens. Mafter 1994, 6, L533. (12) Inone, N.; Sugawara, T. J. Phys. SOC. Jpn. 1978, 45, 450. (13) Weinert, M.; Freeman, A. J. Phys. Rev. E 1983, 28, 6262. (14) Tomhek, D.; Sun, Z.; Louie, S. G . Phys. Rev. B 1991, 43,4699. (15) Makowka, C. D.; Slichter, C. P.; Sinfelt, J. H. Phys. Rev. B 1985, 31, 5663. (16) Louie, S . G. Phys. Rev. Left. 1979, 42, 476.