Studying the Formation of Ni3C from CO and ... - ACS Publications

Jan 20, 2009 - Paul Scherrer Institut, General Energy Research Department, LEM, CH-5232 Villigen PSI, Switzerland, and Swiss Federal Institute of Tech...
0 downloads 8 Views 2MB Size
Download PDF
J. Phys. Chem. C 2009, 113, 2443–2451

2443

Studying the Formation of Ni3C from CO and Metallic Ni at T ) 265 °C in Situ Using Ni K-Edge X-ray Absorption Spectroscopy Rudolf P. W. J. Struis,*,†,‡ Dimitris Bachelin,† Christian Ludwig,†,‡ and Alexander Wokaun† Paul Scherrer Institut, General Energy Research Department, LEM, CH-5232 Villigen PSI, Switzerland, and Swiss Federal Institute of Technology at Lausanne, School of Architecture, CiVil and EnVironmental Engineering, ISTE, CH-1015 Lausanne, Switzerland ReceiVed: October 23, 2008; ReVised Manuscript ReceiVed: December 16, 2008

Metallic Ni nanopowder (Ni0) was monitored during 23 h of carburization (2CO + 3Ni0 f Ni3C + CO2, T ) 265 °C) using Ni K-edge X-ray absorption spectroscopy. X-ray diffraction analysis made afterward at room temperature revealed 28 ( 3% Ni3C among 72% unreacted Ni0. The χ(k) data recorded during carburization showed small changes at low k indicative of carbon backscattering. The identification of carbon was possible with wavelet transform analysis after eliminating the integral contribution from the unreacted Ni0 phase using experimental χ(k) data collected during methanation (CO + 3H2 f CH4 + H2O) at T ) 265 °C. The Fourier-transformed χ(k) data recorded during carburization revealed destructive interference between signals from Ni atoms in slightly different (Ni0, Ni3C) environments. The interference effect mainly lowered the peak amplitude of the first two Ni-Ni coordination shells compared to metallic Ni at T ) 265 °C and it propagated very slowly with increasing carburization run time. In simulation of the amplitude lowering of the first Ni-Ni peak by destructive interference as a function of the carburization run time, it followed that the carbon atoms migrate into the Ni0 particle lattice according to the diffusion-induced grain boundary motion advocated in the literature. 1. Introduction Ni-catalyzed methanation of synthesis gas was investigated intensively in the seventies of the last century. At that time coal gasification was considered as the resource to increase the security of energy supply. Nowadays, sustainability and CO2 neutrality are motivations to produce methane from synthesis gas derived from biomass, such as wood, straw, etc., and Nibased catalysts are not only used in the methanation of wood gasifier product gas1-3 but also in the reforming of methane and other hydrocarbons and in the direct conversion of biomass to methane in supercritical water.4 Biomass gasifiers typically produce a H2 and CO rich synthesis gas but also a variety of potential contaminants like tars, alkalis, sulfur, etc.5 Although all impurities may affect the catalyst performance during operation, adsorption of sulfur and carbon on the catalyst surface are major reasons for their deactivation.6,7 A great deal of fundamental information has accumulated over time on carbon poisoning of Ni-based catalysts under different conditions of industrial interest, but a detailed understanding of the poisoning mechanism is still missing due to the complexity of the poisoning problem and the lack of careful fundamental studies under reaction relevant conditions.6 Carbon-specific information related with the carburization reaction is hard to come by when analyzing, e.g., Ni/Al2O3 methanation catalysts with surfacesensitive techniques (X-ray photoelectron spectroscopy (XPS), IR, diffuse reflectance infrared Fourier transorm spectroscopy (DRIFTS))8 due to the low amounts of Ni3C being formed during methanation among other carbonaceous species (coke, graphite, whiskers) at both (nickel, carrier) surfaces.9 Only very few Ni K-edge XAS studies10 dealt with the nickel carbide * Towhomcorrespondenceshouldbeaddressed.E-mail:[email protected]. † Paul Scherrer Institut. ‡ Swiss Federal Institute of Technology at Lausanne.

formation in the past but without addressing the near-edge (XANES) or extended fine structures (EXAFS) explicitly. Therefore, we studied the carburization reaction (2CO + 3Ni0fNi3C + CO2, T ) 265 °C) using metallic nickel nanopowder (Ni0) and, for purpose of comparison, the Ni0catalyzed methanation reaction (CO+3H2f CH4+H2O; T ) 240, 265, 300 °C) in situ through the “eyes” of the Ni atoms using synchrotron-based X-ray absorption spectroscopy (XAS). The study aims to show that most of the interpretational uncertainties expected from the fact that the bulk structure of Ni0 and Ni3C are very alike (but not identical) can be overcome by using a modified version of the difference file method, by employing complementary tools like wavelet transform analysis and by exploiting the destructive interference effect between oscillatory signals from Ni atoms in slightly different environments. 2. Experimental Section 2.1. Materials. To counterfeit the fact that the recorded X-ray spectra comprise contributions from all (surface and bulk) Ni atoms, whereas both (methanation and carburization) reactions take place along active surface sites only, each experiment started from NiO nanopowder (Ø ≈ 10-20 nm) in view of its large surface-to-volume ratio. The nanopowder was purchased from Aldrich (specifications: purity ) 99.8%; average Ø ≈ 10-20 nm; Brunauer-Emmett-Teller (BET) surface area ∼50-80 cm2/g; bulk density ) 0.51 g/mL). As pure gases we used bottled H2 and CO from Alpha Gaz exhibiting a total impurity less than 3 and 30 ppm, respectively, and class-2 quality Ar from Air Liquide. As reference compound for XAS and XRD, we made pure Ni3C at PSI following the procedure of Bahr and Bahr11 by carburizing completely reduced NiO nanopowder with CO for 240 h at T ) 265 °C in a tubular reactor. A part of the Ni3C sample was diluted by a factor of 80 in weight with BN using an agate mortar and pestle under

10.1021/jp809409c CCC: $40.75  2009 American Chemical Society Published on Web 01/20/2009

2444 J. Phys. Chem. C, Vol. 113, No. 6, 2009

Struis et al.

Figure 1. (A) In situ XAS cell used for the recording of Ni K-edge fluorescence spectra under methanation and carburization reaction conditions, respectively. The outline of the ∼2.4 cm3 large interior reaction volume had been exaggerated in the figure for purpose of visibility.

inert gas to obtain an edge jump in the X-ray transmission spectrum of about one decade. The powder mixture was put in a Plexiglas holder, then, sealed with Kapton tape and stored/ transported under inert gas prior to use. For metallic Ni we used a transmission spectrum of Ni foil, which was kindly put at our disposal by Jan-Dierk Grunwaldt (Danish Technical University, Lyngby, Denmark). 2.2. X-ray Powder Diffraction (XRD). The compositions of the pure Ni3C sample and that of the Ni nanopowder sample resulting after the in situ monitored carburization reaction were analyzed at room temperature using XRD. The XRD measurements were carried out with a diffractometer (Panalytical X‘Pert MPD, Serie DY636) at values of 2θ between 20 and 90° using Fe kR1 as the energy source (λ ) 1.936 Å). The angle was varied in steps of 0.05° every 5 s. We used the freeware software POWDERCELL version 2.4 (http://www.bam.de)12 for analyzing the X-ray powder patterns on the basis of crystal structure data reported in the literature. 2.3. In Situ Reaction Cell for XAS. The reaction cell used for the in situ XAS experiments is shown in Figure 1A. The reactor wall is made out of a Ni-Cr-Ti alloy (material number 1.4980), and it comprises an interior reaction volume of ∼2.4 cm3. The sample holder is a cylinder made from the same alloy (Figure 1B). One end of the cylinder was cut off under an angle of 45° and the inner part of the cut surface (Ø ≈ 1 cm) was milled out (0.5 mm) to prevent the sample under study from falling out. The sample holder is inserted into the reaction cell through an unused X-ray window entrance positioned opposite from the fluorescence detecting window (thus at the back side of Figure 1A). The center of the cut surface is positioned in the middle of the interior reaction volume with the cut surface making an angle of 45° with respect to the incoming and outgoing fluorescence X-rays and 45° to the vertical direction. The reaction cell is heated by 4 heating coils, and the temperature within the interior reaction volume was measured with a thermocouple positioned close to the sample holder. Gas flows (using calibrated mass flow controllers) and temperature were controlled by PC software. Interior reaction cell, sample holder, thermocouple, and the gas inlet and outlet were coated with 20 µm gold to prevent the feed gas to react with cell parts other than the sample under study and to minimize X-rays

absorption contributions from the Ni (and other elements) contained in the alloy material. The X-ray windows consisted of 0.3 mm thick BN slices. 2.4. In Situ XAS Experiments. NiO nanopowder was reduced in the in situ reaction cell with 20 mLn/min of H2 at T ) 300 °C for 1 h. The desired reaction temperature was installed thereafter under 30 mLn/min Ar, followed by the replacement of Ar with the reaction gas. With the carburization experiment, the temperature was T ) 265 °C, and the reaction gas consisted of pure 30 mLn/min CO. The run lasted about 30 h in total and resulted in 36 Ni spectra (each taking 50 min). The experiment was interrupted several times at different time periods by replacing CO by Ar while maintaining the temperature at T ) 265 °C and continuing the recording the Ni K-edge spectra. Thus from the total of 36 Ni spectra, 8 spectra were recorded under Ar. The experiment was also interrupted once after 20 h by replacing CO with Ar followed by the cooling down of the reactor to T ) 30 °C. After recording the Ni spectrum twice, the temperature was reinstalled at T ) 265 °C and Ar was replaced again by the CO flow. A third spectrum under Ar at T ) 30 °C was recorded after discontinuation of the carburization experiment. For purpose of comparison, we also recorded Ni spectra under methane forming conditions (H2/CO ) 3; total flow ) 8 mLn/min) at different temperatures (T ) 240, 265, 300 °C). Each methanation expriment started with the reduction of fresh NiO nanopowder. Ni K-edge spectra were recorded during each run for about 5-6 h and in between also under Ar for ∼2 h. After discontinuation of each methanation experiment, a final spectrum was recorded under Ar at T ) 30 °C. For purpose of safety, all in situ experiments and handling of the gas bottles were monitored continuously by a H2 and CO sensoring device. 2.5. XAS Data Collection and Reduction. Ni K-edge XAS spectra were collected at the Swiss Norwegian Beam Line (SNBL) at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France. It is a bending magnet beamline equipped with a Si(111) channel cut crystal monochromator. The monochromator angle was calibrated by assigning the energy of 8333 eV to the first inflection point with the K-edge absorption spectrum of metallic Ni. Harmonic rejection was performed with a set of double bounce Cr mirrors. The spectrum with the pure

Formation of Ni3C from CO and Metallic Ni

J. Phys. Chem. C, Vol. 113, No. 6, 2009 2445

Figure 2. (A) Background-corrected diffractogram with homemade, pure Ni3C and (B) with the carburized Ni sample resulting after the in situ XAS campaign (“Grenoble sample”). In part A, the vertical lines drawn below the 2Θ axis indicate calculated peak positions for Ni3C as described by Nagakura.22 (CPS ) counts per second.)

Ni3C (reference sample) was collected at room temperature in transmission mode (ionization chambers) and several scans were averaged to improve the signal-to-noise ratio. The measurements with the in situ reaction cell were done in the fluorescence detection mode using a 13-element germanium detector from Canberra. The detector was followed by XIA DSP 2X Digital electronics to perform energy analysis and to build up histograms. The detector-to-sample distance was adjusted such that the incoming count rate stayed below 80,000 counts. An energy window (ROI) was selected around the Ni kR1 fluorescence line. The counts in the ROI were dead time corrected. Comparison of the fluorescence spectra with metallic Ni (from reduced NiO nanopowder) at T ) 30 °C under Ar with the transmission spectrum of the Ni foil revealed that the fluorescence spectra suffered from self-absorption (SA) effects moderately, which is typical with thick samples. We could correct the fluorescence spectra for latter effects successfully by using the computational correction proposed by Booth and Bridges.13 A major plus of the approach is that it offers a nearly exact treatment of the EXAFS oscillations in case that the sample is free from pinholes and not too thick (“thin sample regime”). Assuming an overall sample thickness of 6 µm provided a very satisfactorily agreement between the SA-corrected fluorescence spectra and the metallic Ni foil transmission data. Examples are presented in section 3.2.2. Data reduction, correction, and fitting were performed with the ATHENA software package (version 0.8.049)14 following standard procedures.15 The energy was converted to photoelectron wavevector, k [Å-1], by assigning the origin E0 to the first inflection point of the absorption edge. Often, the resulting χ(k) function is multiplied with k (or a power of k) to compensate for the dampening of the EXAFS amplitude with increasing k. The χ(k) (or kpowerχ(k)) function was Fouriertransformed, typically, between k ≈ 2-12 Å-1 to obtain the radial structure function (RSF). RSF fits were performed in real space across the first coordination shell using theoretical scattering paths calculated with the FEFF code version 8.4 (“FEFF84”)16,17 on the basis of crystallographic data for Ni3C and metallic Ni, respectively. First coordination shells were fitted in terms of the amplitude reduction factor (S02), the interatomic distance (R), the Debye-Waller factor (σ2), and the zero-energy shift parameter (∆E0), while fixing the coordination number (CN) to the crystal data value. The precision in the fitted parameters were provided by the ATHENA software. 2.6. Wavelet Transform (WT) Analysis. WT analysis of EXAFS spectra was used to complement the Fourier transform. Put in short terms, WT analysis allows distinguishing between coordination shell atoms located at the same distance R but yielding backscattering contributions at different k intervals when markedly different atom numbers are involved. Examples

and description of the method can be found with the literature.18,19 AGU-Vallen Wavelet is a freeware software tool (http:// www.vallen.de/wavelet). It calculates the WT of oscillating data, here χ(k), over a user-defined k interval and displays the results in 2- and 3-dimensional plots. It had been developed for acoustic emission analysis, but it is equally well suited for EXAFS analysis because the time-frequency (tω) regime applicable with acoustic signals and the wavevector-distance (kr) regime relevant with EXAFS analysis are linked by linear transformations tfk and 2πωf2r. Consequently, the equidistant step assumed between successive data points must be defined in a multiple of the microsecond time scale displayed with the program (e.g., with χ(k) data and k step ) 5 × 10-2 Å-1, the equivalent step to be defined with the program equals 5 × 10-8) and the phase-uncorrected interatomic distance displayed on a kHz frequency scale should be multiplied by π/1000 (e.g., 700 kHzf2.20 Å). The mother wavelet used by the software is a Gabor wavelet based on the Gaussian function.20 Adjustable parameters with the program are the size of the wavelet window and the frequency resolution interval to calculate the WT. 3. Results and Discussion 3.1. XRD. Figure 2A shows the diffractogram of the longterm carburized Ni nanopowder sample (CO/240 h/T ) 265 °C) that had been synthesized at PSI and that of the carburized Ni sample recorded at PSI one week after the in situ XAS campaign at ESRF in Grenoble (“Grenoble sample”; Figure 2B). With both diffractograms, the almost linearly progressing background was subtracted by using a second order polynomial functionality in 2Θ. The purity of the long-term carburized Ni sample is very high, because the metallic Ni peak at 2Θ ) 66.7° is hardly visible. There is also no indication for the presence of oxidized nickel or that of crystalline graphite (which is a decomposition product of metastable Ni3C),21 because the NiO peaks at 2Θ ) 47.3 and 55.2° and that of graphite at 2Θ ) 33.6° are absent. Therefore, this sample is hereafter referred to as “pure Ni3C”. We used the Scherrer formula to estimate the average crystallite dimension with the pure Ni3C sample on the basis of peak widths. Without taking instrumental broadening effects into account, we derived 42 ( 2 nm large particle diameters, suggesting some sintering of the NiO nanoparticles used at start (Ø ) 10-20 nm). Comparing the experimental with calculated peak positions pertaining to the hexagonal cell description of Ni and C atoms with Ni3C by Nakagura22 revealed that the experimental peak positions were off by about 0.3° (at low 2Θ) up to 0.45° (at high 2Θ). By use of the program POWDERCELL,12 we derived ∼0.6% larger unit cell lengths (a, c) than those reported by Nagakura (a ) b ) 4.553 Å; c ) 12.92 Å).22 In line with Jacobson and Westgren23 and Yue et

2446 J. Phys. Chem. C, Vol. 113, No. 6, 2009

Struis et al.

Figure 3. (A) Background-corrected, normalized Ni K-edge transmission spectra of Ni foil and pure Ni3C as a function of the X-ray energy, E. (B) Fourier transform of k3χ(k) data derived from the spectra shown under (A) using the interval ∆k ) 2.9-10 Å-1 and a Hanning apodization function.

TABLE 1: First-Shell Fit Results (S02, R, σ2, ∆E0) with Fourier-Transformed k3χ(k) Data from Ni Foil and Ni3C, Respectively, Together with Coordination Numbers (CN) and Interatomic Distances (R) Reported in Crystallographic Studies compound

Ni0

first-shell(s) FT k range [Å-1] fitted R range [Å] window function S02 R [Å] σ2 [Å-2] ∆E0 [eV] reduced χ2 crystal data reference CN R [Å]

Ni-Ni 3.3-14.0 1.8-2.5 Hanning14 0.81 ( 0.04 2.485 ( 0.003 0.0060 ( 0.0003 8.1 ( 0.7 32.5 25 12 2.49

al.,24 also we did not observe any “super-lattice” reflections, which, according to Nagakura,22 reflect the regular arrangement of the C atoms with Ni3C. The absence of these reflections may be attributed to the different techniques used, i.e., electron diffraction by Nagakura, which is more sensitive to lighter atoms, and X-ray diffraction by Jacobson and Westgren, Yue et al., and with the present study. But, it could also mean that our pure Ni3C sample is less perfect due to lattice distortions, stacking faults, dislocations, local strains, etc. The supposition made here would also provide a more plausible explanation for the ∼0.6% larger unit cell lengths with our Ni3C sample than that of thermal expansion, because the expansion relations reported by Nakagura22 do not predict such unit cell enlargements for temperatures below T ≈ 460 °C. With pure Ni3C, the experimental peak areas vary in the ratio 18.9/23.9/100/ 14.1 when going in Figure 2A from left to right (100 ) largest peak area). The ratio is close to the calculated one (19.9/25.3/ 100/17.0) using the lattice description of Nagakura and cell lengths derived here. Focusing on the carburized Ni sample resulting after the XAS campaign (Figure 2B), it appears to be a mixture of Ni3C and metallic Ni only. From the integrated areas (4.50/6.39/4.35) under the Ni3C peaks that do not coincide with those of Ni0 and the experimental values for the corresponding peaks with the pure Ni3C sample (18.9/23.9/14.1), we inferred that the Grenoble sample comprises 28 ( 3% Ni3C among 72% unreacted Ni0. This estimate was confirmed by CO2 evolution rate peaks appearing around typical Ni3C decomposition temperatures (T ≈ 380-420 °C)21 during temperature-programmed oxidation (TPO) experiments (not shown here). From TPO and Raman experiments, it also appeared that the pure Ni3C (but not the Grenoble sample) may also contain some weight percent

Ni3C Ni-C 2.1-13.0 1.3-2.8 Hanning14 0.75 ( 0.06 (common value) 1.88 ( 0.01 0.0042 ( 0.0014 12.8 ( 0.9 (common value) 24.3 22 2 1.86

Ni-Ni

2.657 ( 0.005 0.0119 ( 0.0007

12 2.63

of highly disordered (thus XRD-invisible), graphitic-type carbon (T∼780 °C with TPO),6 the presence of which however does not affect the other observations and conclusions presented in this paper. 3.2. EXAFS. 3.2.1. Interpretation of the Ni3C and Metallic Ni Reference Spectra. For a better appreciation of the in situ XAS results, the transmission reference spectra with Ni foil (Ni0) and pure Ni3C at room temperature (Figure 3) may be discussed first. With both spectra, E0 was close to the Fermi level of elemental Ni (EF ) 8333 eV). From Figure 3A it can be seen that the oscillary part in the so-called extended EXAFS region differs entirely between the two reference samples. The Fourier transform of the k3-weighted χ(k) data between ∆k ) 2.9-10 Å-1 (Figure 3B), with Ni3C, clearly revealed the first Ni and C shells around the X-ray absorbing Ni atoms and, with Ni foil, the first Ni shell. The fitted first-shell distances (Table 1) are well in line with the superlattice structure of Ni3C22 and that of face-centered cubic (fcc) metallic Ni,25 respectively. On the average, the fitted S02 value equaled 0.78 ( 0.04, and it compares well with the value of S02 ) 0.77 ( 0.03 reported for Ni in the literature.26 Also noteworthy is that the fitted value for σ2(Ni-Ni) with Ni0 at room temperature (0.006 Å-2) matches well with the literature27 and that σ2(Ni-Ni) with Ni3C at room temperature is about two times larger (∼0.012 Å-2). 3.2.2. Self-Absorption Correction of the Fluorescence Spectra. The Ni K-edge fluorescence spectra recorded in situ during the carburization and the methanation reaction showed moderate signs of self-absorption (SA), which we corrected using the approach proposed by Booth and Bridges.13 The adequateness of the correction can be illustrated by comparing the fluorescence spectra which were recorded under Ar at T ) 30 °C after the

Formation of Ni3C from CO and Metallic Ni

J. Phys. Chem. C, Vol. 113, No. 6, 2009 2447

Figure 4. (A) Compilation of k3χ(k) data and their Fourier transforms for SA-corrected “Ni0 at T ) 30°” fluorescence and SA-unaffected Ni foil transmission data at room temperature. (B) SA-corrected fluorescence data for “Ni0 at T ) 30°C” and “active Ni0 at T ) 265°C”, respectively.

discontinuation of the three methanation runs (section 2.4) with the SA-unaffected transmission spectrum of the Ni foil collected at room temperature. With methanation it is reasonable to expect that no bulk Ni3C phase is being formed when the reaction of the carbon atoms at the active Ni surface sites with H2 is sufficiently fast. And even when some bulk Ni3C is actually being formed with our Ni0 nanopowder over technically relevant methanation run times (months), it will likely remain undetected after the short times (∼6 h) used in the present study. Whatever the case, the Ni samples probed thereafter at T ) 30 °C under Ar will appear free from Ni3C. The proposition applied well in that the SA-corrected fluorescence spectra proofed to be identical within detection limits with the Ni0 foil transmission spectrum. The SA-corrected spectra were therefore merged into a single “Ni0 at T)30°C” file. We also could not find any differences between the spectra recorded during methanation at T ) 265 °C and that of the freshly reduced NiO nanopowder recorded under Ar at T ) 265 °C at the start. The SA-corrected methanation spectra recorded at T ) 265 °C were therefore merged into a single data file to raise its signal-to-noise ratio in further analysis, and this file is hereafter referred to as “active Ni0 at T ) 265 °C”. Also the fluorescence spectra recorded during methanation at T ) 240 °C and those at T ) 300 °C showed no signs for the presence of species other than that of the metallic Ni0 phase. Figure 4 shows a compilation of k3χ(k) data and their Fourier transforms pertaining to the SA-corrected “Ni0 at T)30°” and “active Ni0 at T ) 265°C” files, together with those of the Ni foil at room temperature for purpose of comparison (as transmission spectra are not affected by SA effects).15 Details and fitted first-shell parameters results are given in Table 2. The fitted parameters with the SA-corrected “Ni0 at T ) 30 °C” fluorescence data agree well with those derived with the Ni foil transmission data (Table 1) except that the estimated uncertainties for the fitted parameters are smaller with the transmission data and that the self-correction could be finetuned (with respect to the sample thickness) to make the fitted amplitude reduction factor (S02) matching somewhat (10%) better with the literature value of S02(Ni) ) 0.77 ( 0.03.26 This, however, is not a prerequisite with the present study. Also

TABLE 2: First-Shell Fit Results (S02, R,σ2, ∆E0) with SA-Corrected, Fourier-Transformed k3χ(k) Data from “Ni0 at T ) 30 °C” and “Active Ni0 at T ) 265 °C”, Respectively, with CN ) 12 compound

Ni0 at T ) 30 °C

active Ni0 at T ) 265 °C

first-shell FT k range [Å-1] fitted R range [Å] window function S02 R [Å] σ2 [Å-2] ∆E0 [eV] reduced χ2

Ni-Ni 2.1-12.4 1.6-2.6 Hanning14 0.71 ( 0.10 2.47 ( 0.01 0.0050 ( 0.0009 6.9 ( 1.7 13.8

Ni-Ni 2.0-12.5 1.6-2.6 Hanning14 0.69 ( 0.11 2.48 ( 0.01 0.0093 ( 0.0013 7.6 ( 1.9 8.6

noteworthy is that the Debye-Waller factor increases by a factor of 2 when raising the temperature from T ) 30 °C to T ) 265 °C. This trend is well in line with the literature on metallic Ni showing that σ2 equals about 0.005 Å-2 at room temperature and that it increases up to about 0.011 Å-2 at T ≈ 250 °C.27 3.2.3. WT Analysis of the Reference Spectra. In this study, the usual EXAFS interpretation is complemented by wavelet analysis of the experimental χ(k) data to distinguish better between backscattering contributions from C and Ni atoms in the Ni3C phase formed during the carburization reaction. For a significantly large difference between the atom numbers, here Z(Ni) ) 28 and Z(C) ) 6, this should be possible, because the shape and position of the backscattering amplitude function is atom number specific.15 This is illustrated in Figure 5, showing backscattering amplitude functions for Ni and C at the top and experimental χ(k) data for Ni foil and pure Ni3C at the bottom. The backscattering amplitude functions were calculated using the FEFF84 code and assuming σ2 ) 0.005 Å-2 for ease of comparison with the experimental χ(k) data. Note that the backscattering amplitude maxima are located between k ≈ 2-4 Å-1 for C and at ∼6 Å-1 for Ni. The potential of the WT analysis is illustrated here with experimental χ(k) data of metallic Ni and that of pure Ni3C, which were both analyzed between ∆k ) 2-13 Å-1. The WT converts χ(k) into a series of wavelet coefficients, each of which

2448 J. Phys. Chem. C, Vol. 113, No. 6, 2009

Figure 5. (A) Theoretical backscattering amplitude function of carbon atoms (solid line) and nickel atoms (dashed line) as a function of wavevector, k. (B) Experimental χ(k) data for Ni foil and Ni3C.

Figure 6. Wavelet transform analysis results with experimental χ(k) data of (A, B) metallic Ni and (C, D) pure Ni3C at room temperature, shown in 2- and 3-D plots. The height of the 3-D surface plot represents the magnitude of the wavelet transform coefficients. The shading from dark to light gray indicates changes in the WT coefficient from 100% (maximum value) to 50%.

representing the amplitude of the wavelet function at a particular location within the data array under the selected wavelet window size. Focusing on metallic Ni, Figure 6A shows that the firstshell backscattering Ni are located at a phase-uncorrected distance of R + ∆R ≈ 2.1 Å and that they exhibit highest WT coefficient values at kmax ≈ 5.6Å-1, which is well in line with the predicted (calculated) amplitude function for Ni (Figure 5A). From the 3D plot (Figure 6B), it appears that the higher distanced Ni shells are not visible, likely because the contribution from the first-shell Ni atoms is dominant. With pure Ni3C, parts C and D of Figure 6 reveal the presence of carbons at R + ∆R ≈ 1.3 Å and at ∼2.8 Å by kmax ≈ 3.1-3.3 Å-1, together with first-shell Ni atoms at R + ∆R ≈ 2.3 Å by kmax ≈ 5.0 Å-1. The shading from dark to light gray indicates changes in the WT coefficient values from 100% (maximum) to 50%. The shading with the second shell carbons appears darker than with the first-shell carbons mainly because the second shell is superimposed on top of the first-shell Ni contribution. Finally, we note that the phase-uncorrected shell distances, R + ∆R, found with the WT analysis compare reasonably well with those of the Fourier-transformed k3χ(k) data shown in Figure 3B.

Struis et al. 3.2.4. Ni K-Edge Spectra Recorded in situ at T ) 265 °C. The carburization experiment at T ) 265 °C was varied a few times on purpose by replacing CO by Ar while continuing recording the Ni K-edge spectra. (The CO flow was reinstalled thereafter.) This was done to see whether the spectra recorded under CO also comprised contributions from weakly absorbed CO molecules or those of loosely bound intermediates at the Ni nanoparticle surfaces. But, after correction and normalization, the spectra recorded under CO and those under Ar were identical. Starting from the beginning of the carburization run, each six succeeding spectra (including those recorded under Ar) were merged into a single data file to raise the signal-to-noise ratio in further analyses. From Figure 7A it can be seen that the k-weighted χ(k) data with the in total seven sequentially averaged spectra are virtually identical among each other and that they differ only slightly with the averaged “active Ni at T ) 265°C” data (recorded during methanation), except for small but marked differences between k ≈ 2-6 Å-1, which may be seen indicative of carbon backscattering contributions (Figure 5A) with the carburized Ni sample. However, the corresponding Fourier transforms (Figure 7B) do not provide direct clues to the presence of carbon (R(Ni-C) ≈ 1.9 Å) nor that of nickel in a carbide-induced reconstructed Ni surface (R(Ni-Ni) ≈ 2.58 Å)28,29 or in a bulk Ni3C phase (R(Ni-Ni) ≈ 2.62 Å).22,23 The question whether carbon atoms are present with the Ni K-edge spectra recorded in situ during carburization is addressed first in the next section. 3.2.5. Modified “Difference File Method”. The “difference file method” can be used to extract contributions of heavy scatterers in the vicinity of light elements. A detailed description of this method and the underlying theory can be found in a previous work.30 The method may be briefly described as the fitting of the dominant contribution(s) to the EXAFS data in R space and then subtracting it from the data to make the remaining weaker contribution(s) visible. But, eliminating the dominant first-shell Ni-Ni path with the Fourier transformed EXAFS data recorded during carburization (Figure 7B) was not satisfactory, because the intensity and the shape of the Ni-C contribution at low R remained indefinable. This failure is likely because (1) the complexity of the first Ni-Ni coordination shell comprising contributions from different (Ni0, Ni3C) environments, (2) the presence of noise and possibly not-well-removed glitches with the spectra, and (3) the deformation of the Ni-C peak by truncation effects.15 To bypass these dilemmas, the seven sequentially averaged χ(k) files shown in Figure 7B were merged into a single “Ni during carburization at T ) 265 °C” data file to raise the signal-to-noise ratio and the difference file method was adapted for use in k space as follows: We did not fit any part of the χ(k) data, but we subtracted experimental χ(k) data of active Ni0 at T ) 265 °C (“χactive”, Figure 4B), which one may reasonably assume to reflect the integral (and not just a specific) contribution from the unreacted Ni0 phase during carburization at T ) 265 °C. XRD analysis (section 3.1) showed that the in situ recorded Ni sample had not been carburized to a full extent. With the overall-averaged “Ni during carburization at T ) 265 °C” χ(k) data file (“χcarb”), the apparent conversion degree was determined by calculating the sum of square differentials, SUM(F) ) Σ[(χcarb(k) - F(χactive(k))]2 over that part of k space being free from electron transitions (∆k ≈ 2.4-13 Å-1; see Supporting Information) as a function of fraction F between (0 e F e 1). It showed that SUM(F) passes through a minimum (0.0054) for F ) 0.76, whereas SUM(F) equalled 0.100 and 0.015 for F ) 0 and F ) 1, respectively. The “difference spectrum” resulting with F ) 0.76 (Figure 8A)

Formation of Ni3C from CO and Metallic Ni

J. Phys. Chem. C, Vol. 113, No. 6, 2009 2449

Figure 7. (A) kχ(k) from averaged Ni K-edge spectra recorded at T ) 265 °C during carburization and methanation (“active Ni”), respectively and (B) their Fourier transforms (∆k ) 2.1-12 Å-1; Hanning apodization function). With part B, the asterisk marks the averaged result calculated with the first six recorded spectra, and the R values were phase corrected by the ATHENA software.

Figure 8. (A) Experimental and calculated χ(k) data of the difference spectrum and (B) the fitted paths (see text). (C, D) WT analysis with the experimental χ(k) data shown under part A in 2- and 3-D plots The shading from dark to light gray indicates changes in the WT coefficient from 100% (maximum value) to 50%.

shows a beating pattern that could be fitted fairly well in k space (Figure 8B) in terms of one Ni-C contribution located at low distance (R ≈ 2 Å ((20%); CN ≈ 0.81; σ2 ≈ 0.0054 Å-2; ∆E0 ) 5.4 eV) and a minor Ni-Ni contribution at higher distance (R ≈ 3.5 Å ((20%); CN ≈ 2.0; σ2 ≈ 0.0013 Å-2; ∆E0 ) 5.4 eV) for S02 ) 0.77. The WT analysis of the “difference spectrum” (parts C and D of Figure 8) clearly reveals the presence of carbon atoms with a phase-uncorrected distance of about R + ∆R ≈ 1.3 Å by kmax ≈ 3.9 Å-1 and that of Ni remnants between R + ∆R ≈ 2.2-2.8 Å by kmax ≈ 6.5-7 Å-1. 3.2.6. DestructiWe Interference. Coming back at the apparent absence of a Ni-Ni peak (or shoulder) at R ) 2.62 Å from Ni3C with the Ni sample during carburization, it is striking that the first-shell Ni-Ni peak amplitude with each of the seven single “active Ni at T ) 265 °C” spectra recorded during methanation matches well with that of metallic Ni recorded after reduction of NiO with H2 at the start of the methanation run (Figure 9), whereas the “Ni during carburization at T ) 265 °C” spectra reveal significantly lower first-shell Ni-Ni peak amplitudes already within the first six recorded spectra and which attain ∼15% lower values with increasing carburization run time.

Figure 9. First-shell Ni-Ni peak amplitude derived from Ni K-edge spectra recorded during methanation at T ) 265 °C (b) and carburization at T ) 265 °C (O, 0), respectively, as a function of the file record number. Run number ) 0 denotes the result with the reduced NiO sample recorded at the start of the methanation run. (b,O) denote results from single and (0) that of sequentially averaged spectra. The trendline through the carburization results was drawn as an aid to the eyes.

The discrepancy in the first-shell Ni-Ni peak amplitude was also evidenced at T ) 30 °C between pure Ni0 and the Ni sample recorded directly after the discontinuation of the carburization run. Therefore, the explanation must be found in the destructive interference between the Ni-Ni backscattering functions at the said two Ni-Ni distances (2.49 Å with Ni0 and 2.62 Å with Ni3C). Examples of destructive interference are reported in the literature with other transition metals in much more complicated compounds31,32 but as far as we know of, not for Ni except for a very short remark by Takenaka et al.,10 who studied structural changes in a Ni/SiO2 catalyst during methane decomposition. The supposition of destructive interference is illustrated here by means of theoretical Ni-Ni single-pair backscattering functions, χSS(k), calculated with FEFF84. Figure 10A shows χSS(k) for metallic Ni (assuming R ) 2.496 Å, CN ) 12, σ2 ) 0.005 Å-2), Ni3C (R ) 2.636 Å, CN ) 12, σ2 ) 0.012 Å-2) and as an example that of the 50% mixed case. The shell parameters had been chosen in view of the fit results with the reference compounds (Table 1). Figure 10B shows Fourier transforms of k-weighted χSS(k) data (∆k ≈ 2.7-13 Å-1; Bessel apodization function) with focus on the first Ni-Ni peak with pure Ni0 (amplitude)100%), pure Ni3C, and that of selected mixtures.

2450 J. Phys. Chem. C, Vol. 113, No. 6, 2009

Struis et al.

Figure 10. (A) Simulated backscattering function, χSS(k), pertaining to the first Ni coordination shell with metallic Ni and pure Ni3C, respectively, together with their average (“50% mixed case”). (B) Fourier transform of k-weighted χSS(k) data pertaining to simulated [Ni0:Ni3C] mixtures (solid line, dotted line) and experimentally derived first-shell Ni-Ni peak amplitudes (3) as a function of the phase-uncorrected distance, R + R [Å].

From Figure 10B it is seen that the amplitude of the first Ni-Ni peak drops gradually from 100% with pure Ni0 (see mix ratio [Ni0:Ni3C])12:0) to about 36% with the 75% Ni3Ccontaining mixture (mix ratio ) 3:9), while the phase-uncorrected distance (R + ∆R) remains close to that of fcc metallic Ni. The peak amplitude recovers slightly with the higher Ni3Ccontaining mixtures and approaches 43% with the pure Ni3C case (mix ratio)0:12), while R + ∆R now gradually shifts toward the 0.12 Å higher value anticipated with bulk Ni3C. By comparison of experimental with simulated peak amplitudes, from Figure 10B, it appears that the amplitude drop evidenced near to the end of the carburization run (see lowest 3 symbol) resembles that of a calculated Ni0:Ni3C mixture with a Ni3C amount lying halfway between 8% (mix ratio 11:1) and 25% (mix ratio 9:3), whereas from XRD (section 3.1) we inferred that it should actually be 28 ( 3%. Therefore, a more detailed interpretation seems in place here. The drop in the peak amplitude evidenced with the first spectrum recorded during carburization (see second highest 3 symbol in Figure 10B) likely reflects interference effects induced by C atoms residing at Ni surface sites.28,29 The amplitude drop evidenced thereafter would then adhere to interference effects induced by the C atoms that had diffused within the Ni particles. Taking the XRD estimate of 28% Ni3C among 72% unreacted Ni0 for granted and assuming that the Ni3C phase grows systematically from the Ni particle surface inward, then the experimental peak amplitude should have dropped gradually from 100% at start to slightly below that of the simulated 25% Ni3C case (mix ratio ) 9:3) at the end of the carburization run. From Figure 10 B it is clear that this is not the case. The interference effect is determined largely by the amount of Ni atoms residing in different environments and the difference in the Ni-Ni distances involved. The apparent lower Ni3C amount indicated by the destructive interference effect may thus also mean that not all of the C atoms that had migrated into the Ni lattice reside in structurally pure, bulklike Ni3C domains or (by the same token) that the enlargement in the Ni-Ni distance invoked by the C atoms is actually smaller than expected from pure bulk Ni3C. Some support for this proposition can be found in the lattice imperfections evidenced with XRD, rendering slightly larger unit cell lengths for Ni3C (section 3.1). From literature it is known that the carbon diffusion along nickel grain boundaries is orders of magnitude faster than through the lattice.33,34 Parthasarathy and Shewmon34 inferred that numerous new grains may nucleate when the solute approaches its saturation value (∼0.4 wt % C in Ni at T ) 700 °C),35 allowing for “diffusion-

induced grain boundary motion”. Compared to bulk Ni0, the enlargement in the Ni-Ni distance invoked by C (or CO) adsorbed at Ni surfaces is actually smaller (∼0.052 Å)36 than with bulk Ni3C (∼0.13 Å).22 Therefore not surprising, simulating the peak amplitude for a mixture of 25% carbidic surface like Ni with R(Ni-Ni) ≈ 2.55 Å and 75% bulk Ni0 (R ) 2.49 Å) (see “refined case” in Figure 10B) is in excellent agreement with the experimental amplitudes evidenced toward the end of the carburization run. Destructive interference also affected the second peak located between 3-4 Å (Figure 7B) but not the higher coordination shells, suggesting that the structural changes invoked by C is limited to a few Ni atom layers only. In turn, this would speak more for a diffusion of C atoms into the Ni0 particles along grain surfaces than through its lattice. Put in a broader context, the insights gained here on destructive interference effects resulting from carbon atoms attached to (or dissolved within) Ni0 domains may, e.g., also apply with coke deposition, which, as reported recently for the steam gasification of cedar wood with Pt/Ni/CeO2/Al2O3 catalysts,37 apparently also led to lower Ni-Ni coordination numbers compared to the freshly reduced and the regenerated catalyst. 4. Summary The carburization reaction of CO with metallic Ni (Ni0) at T ) 265 °C leads to the formation of nickel carbide (Ni3C). However, the Ni K-edge spectra recorded in situ during 23 h of carburization (leading to 28 ( 3% Ni3C among 72% unreacted Ni0) only revealed small amplitude changes in the χ(k) data shortly after the white line when compared with methanation-active (Ni0) spectra recorded at the same temperature. In contrast with pure Ni3C, no C nor Ni peak features associated with the formed Ni3C phase were discernible with the usually derived radial structure functions, except for ∼10% lower peak amplitude with the first Ni-Ni peak and ∼30% lower peak area under the second Ni-Ni shell at distances typical with Ni0. The presence of light atoms was evidenced more clearly after subtracting the integral (and not a specific) contribution of the unreacted Ni0 phase from the χ(k) data collected during carburization. The identification of carbon was made possible by WT analysis. It was demonstrated that the first-shell Ni-Ni peak amplitude reduction likely stems from destructive interference of Ni atoms in slightly different (Ni0, Ni3C) environments. It was inferred that the amplitude reduction would adhere well to a diffusion induced grain boundary motion of the carbons, a process which had been postulated for the

Formation of Ni3C from CO and Metallic Ni migration of C into the Ni0 lattice, but, as far as we know, had never been evidenced by means of synchrotron Ni K-edge XAS. Acknowledgment. The experimental assistance of the staff of the Swiss-Norwegian Beam Lines at the ESRF, and in particular that of Wouter van Beek, is gratefully acknowledged. Beat Zehnder and team from Sitec AG (Maur, CH) are thanked for constructing the in situ reaction cell, the Techno team (PSI) for implementing the reaction cell periphery and software for steering and controlling gas flows, heating coils, and reactor temperature, and Jeroen van Bokhoven (ETHZ, Zu¨rich, CH) for helpful discussions. Supporting Information Available: Experimental and simulated Ni K-edge XANES with Ni0 and Ni3C. This information is available for free of charge via the Internet at http://pubs.acs.org. References and Notes (1) Seemann, M. C.; Biollaz, S. M. A.; Schaub, M.; Aichernig, C.; Rauch, R.; Hofbauer, H.; Koch, R. Methanation of biosyngas and simultaneous low-temperature reforming: First results of long duration tests at the FICFB gasifier in Gu¨ssing. In Proceedings of the 14th European Biomass Conference & Exhibition; Paris, 2005. (2) Schildhauer, T. J.; Seemann, M.; Biollaz, S. M. A.; Stucki, S.; Ulrich, D.; Rauch, R. In Successful demonstration of long term catalyst stability in the methane from wood process; Scientific Report PSI, Paul Scherrer Institut: Villigen, Switzerland, 2008. (3) Seemann, M. C.; Schildhauer, T. J.; Biollaz, S. M. A.; Stucki, S.; Wokaun, A. Appl. Catal. A: Gen. 2006, 313, 14–21. (4) Stucki, S.; Biollaz, S. M. A.; Vogel, F. In Vom Holz zum Methan, Fachtagung “RegeneratiVe Kraftstoffe”; Zentrum fu¨r Sonnenenergie und Wasserstoff-Forschung: Stuttgart, Germany, 2003; pp 210-217. (5) Trimm, D. L. Catal. ReV. Sci. Eng. 1997, 16, 155–189. (6) Bartholomew, C. H. Catal. ReV. Sci. Eng. 1982, 24, 67–112. (7) Bartholomew, C. H.; Agrawal, P. K.; Katzer, J. R. AdV. Catal. 1982, 31, 135–242. (8) Zaera, F. Prog. Surf. Sci. 2001, 69, 1–98. (9) (a) Czekaj, I.; Piazzesi, G.; Kro¨cher, O.; Wokaun, A. Surf. Sci. 2006, 600, 5158–5167. (b) Czekaj, I.; Loviat, F.; Raimondi, F.; Wambach, J.; Biollaz, S.; Wokaun, A Appl. Catal., A 2007, 329, 68–78. (10) Takenaka, S.; Ogihara, H.; Otsuka, K. J. Catal. 2002, 208, 54–63. (11) Bahr, H. A.; Bahr, Th. Ber. Chem. Ges. 1928, 61, 2177–2183. (12) (a) Kraus, W.; Nolze, G. PowderCell for Windows V2.4; Federal Institute for Materials Research and Testing: Berlin, Germany. (b) Kraus,

J. Phys. Chem. C, Vol. 113, No. 6, 2009 2451 W.; Nolze, G. Powder Diff. 1998, 13, 256. (c) Kraus, W.; Nolze, G. CPD Newslett. 1998, 20, 27. (13) Booth, C. H.; Bridges, F. Phys. Scr. 2005, T115, 202–204. (14) Ravel, B.; Newville, M. J. Synchrotron Radiat. 2005, 12, 537– 541. (15) Prins, R.; Koningberger, D. C. In X-ray Absorption: Principles, Applications, Techniques of EXAFS, SEXAFS and XANES; Koningberger, D. C., Prins, R., Eds.; John Wiley & Sons: New York, 1988. (16) Rehr, J. J.; Ankudinov, A. L. Coord. Chem. ReV. 2005, 249, 131– 140. (17) Ankudinov, A. L.; Ravel, B.; Rehr, J. J.; Conradson, S. D. Phys. ReV. B 1998, 58, 7565–7576. (18) Funke, H.; Scheinost, A. C.; Chukalina, M. Phys. ReV. B 2005, 71, 094110. (19) Mun˜oz, M.; Argoul, P.; Farges, F. Am. Mineral. 2003, 88, 694– 700. (20) Suzuki, H.; Kinjo, T.; Hayashi, Y.; Takemoto, M.; Ono, K. J. Acoust. Em. 1996, 14, 69–84. (21) Hofer, L. J. E.; Cohn, E. M.; Peebles, W. C J. Phys. Colloid Chem. 1950, 54, 1161–1169. (22) (a) Nagakura, S. J. Phys. Soc. Jpn. 1957, 12, 482–494. (b) Nagakura, S. J. Phys. Soc. Jpn. 1958, 13, 1005–1014. (23) Jacobson, B.; Westgren, A. Z. Phys. Chem. B 1933, 20, 361–367. (24) Yue, L.; Sabiryanov, R.; Kirkpatrick, E. M.; Leslie-Pelecky, D. L. Phys. ReV. B 2000, 62, 8969–8975. (25) Wyckoff, R. W. G. Cryst. Struct. 1963, 1, 7–10. (26) Mansour, A. N.; Melendres, C. A. J. Phys. Chem. A 1998, 102, 65–81. (27) Okube, M.; Yoshiasa, A. J. Synchrotron Radiat. 2001, 8, 937– 939. (28) Klink, C.; Stensgaard, I.; Besenbacher, F.; Lægsgaard, E. Surf. Sci. 1995, 3423, 250–260. (29) Nakano, H.; Nakamura, J. Surf. Sci. 2001, 482-485, 341–345. (30) (a) Duivenvoorden, F. B. M.; Koningsberger, D. C.; Uh, Y. S.; Gates, B. C. J. Am. Chem. Soc. 1986, 108, 6254. (b) Vaarkamp, M.; Modica, F. S.; Miller, J. T.; Koningsberger, D. C. J. Catal. 1993, 144, 611. (c) Vaarkamp, M.; Dring, I.; Oldman, R. J.; Stern, E. A.; Koningsberger, D. C. Phys. ReV. B 1994, 50, 7872. (31) (a) Schlegel, M. L.; Manceau, A.; Hazemann, J.-L.; Charlet, L. Am. J. Sci. 2001, 301, 798–830. (b) Schlegel, M. L.; Manceau, A. Geochim. Cosmochim. Acta 2006, 70, 901–917. (32) Manceau, A.; Lanson, B.; Drits, V. A. Geochim. Cosmochim. Acta 2002, 66, 2639–2663. (33) Massaro, T. A.; Eugene, E. E. J. Appl. Phys. 1971, 42, 5534–5539. (34) Parthasarathy, T. A.; Shewmon, P. G. Scripta Metall. 1983, 17, 943–946. (35) Lander, J. J.; Kern, H. E.; Kern, A. L. J. Appl. Phys. 1952, 32, 1305–1309. (36) Peters, K. F. Phys. ReV. Lett. 2001, 86, 5325–5328. (37) Nakamura,K.Appl.Catal.B:Environ.2008,doi:10.1016/j.apcatb.2008.07.016.

JP809409C