Operando QEXAFS studies of Ni₂P during thiophene hydrodesulfurization: direct observation of Ni—S bond formation under reaction conditions

Structural changes in Ni₂P/MCM-41 were followed by quick extended X-ray absorption fine structure (QEXAFS) and were directly related to changes in X-ray absorption near-edge structure (XANES) which had been used earlier for the study of the active catalyst phase. An equation is proposed to correct the transient QEXAFS spectra up to second-order in time to remove spectral distortions induced by structural changes occurring during measurements. A good correlation between the corrected QEXAFS and the XANES spectral changes was found, giving support to the conclusions derived from the XANES in the previous work, namely that the formation of a Ni—S bond in a surface NiPS phase is involved in the active site for the hydrodesulfurization reaction.

1. Introduction

Hydrodesulfurization (HDS) catalysts, which reduce sulfur content in transportation fuels, are important in the refining industry. Transition metal phosphides are a new class of HDS catalysts with promising activity (Dhandapani et al., 1998). Among these, Ni₂P has been found to show the highest activity for HDS reactions, especially for refractory compounds such as 4,6-dimethyldibenzothiophene, with activity higher than conventional catalysts like CoMoS or NiMoS on an active site basis (Oyama, 2003). It is a challenging issue to elucidate the active site and reaction mechanism of the highly active Ni₂P. Elemental analysis has shown that a substantial amount of sulfur accumulated in the catalysts in the course of reactions (Oyama et al., 2002; Kim et al., 2005; Sawhill et al., 2005). In situ X-ray absorption fine structure (XAFS) is a powerful technique for characterizing catalyst structure during reactions (Bart, 1986; Bart & Vlaic, 1987; Iwasawa, 1996); XAFS is composed of two parts: XANES (X-ray absorption near-edge structure) and EXAFS (extended X-ray absorption fine structure). Kawai et al. found a clear oscillation in the difference EXAFS spectra between a sample of Ni₂P/SiO₂ in hydrogen before reaction and the same sample during HDS reaction that could be associated with the formation of a Ni—S bond of length 0.230 nm (Kawai et al., 2006, 2007, 2008). They also found a feature at 8333.3 eV in the XANES region that decreased in intensity in going to steady-state reaction. They attributed the change in the XANES spectrum also to the formation of a Ni—S bond. In order to better elucidate the relationship between the Ni—S bond and the reaction mechanism, time-resolved XAFS studies together with transient product analysis can be a powerful combination. In the past we have carried out such measurements with in situ simultaneous time-resolved XANES measurements of the supported Ni₂P, focusing on the changes in the XANES signal, together with in situ Fourier transform infrared spectroscopy (FTIR) as well as product analysis (Bando et al., 2009; Oyama et al., 2009). We found that Ni—S bond formation occurred before high activity was observed for the HDS reaction. We concluded that a surface phosphosulfide (NiPS) was the active structure of the Ni₂P catalyst for the HDS reaction (Bando et al., 2009). But the formation of the Ni—S bond was based on the assumption that the XANES peak change was due precisely to the Ni—S bond. XANES gives strong signal intensity but is not easily interpreted, while EXAFS gives much weaker signals, but can be analyzed based on single scattering theory to provide an easily interpreted local structure. Also XANES covers a narrow energy region of a few tens of electronvolts, while the EXAFS spectra cover a wide energy region of 600–1000 eV and require a long scanning time. There are two methods for acquiring time-resolved EXAFS (Newton et al., 2002). One method is dispersive EXAFS (DEXAFS), which provides the whole EXAFS spectra at once using a bent crystal and a position-sensitive detector (Matsushita & Phizackerley, 1981; Pascarelli & Mathon, 2010). Although the acquisition time is potentially fast, it requires high sample homogeneity (Newton et al., 2002). The other method is quick-EXAFS (QEXAFS), which is achieved by the fast scanning of the monochromator (Frahm, 1989; Dent, 2002; Oudenhuijzen et al., 2002; Okumura et al., 2007; Bando et al., 2012). Since QEXAFS can use standard measurement equipment, it provides reliable data and is more tolerant of sample inhomogeneity. In this work we conducted QEXAFS measurements during the HDS reaction of thiophene. Problems with QEXAFS are low signal-to-noise ratio and the distortion of the EXAFS spectrum for the case that a structural change occurs in the time scale of a scan (Nomura, 2011). Since Ni EXAFS requires a measurement range of almost 800 eV and a high signal-to-noise ratio, each QEXAFS spectrum needs about 10 s for acquisition and the structure may change during the scan because the reaction time scale is a few minutes.

In this paper we propose a method of correction of transient EXAFS data up to second-order in time. We have compared the time dependence of in situ QEXAFS and XANES of Ni₂P/MCM-41 during the thiophene HDS reaction to find a direct correlation for the active site of Ni—S and the XANES changes. The good correlation between the measurements demonstrated that the Ni—S signal was not associated with deactivation or with adsorbed reaction intermediates but was related to the active site, supporting the formation of an active NiPS phase.

2. Experimental

2.1. Sample preparation

An MCM-41-supported Ni₂P catalyst was synthesized according to a previous report (Oyama et al., 2004). The sample was reduced by temperature-programmed reduction in a flow of H₂ and was then passivated in a flow of 0.5% O₂/He for 4 h to prevent uncontrolled oxidation of the freshly produced phosphide. The loading of Ni was 1.16 mmol Ni g_support^-1. The sample was pressed into a disc and set at the center of an in situ cell (Bando et al., 2012). The catalyst was activated in the cell under H₂ flow (100 ml min⁻¹) at 803 K for 2 h. Then the sample was cooled to HDS reaction temperatures in H₂ flow. The reaction was started by introduction of the reactant gas composed of thiophene (0.1 vol%), He (1.9 vol%) and H₂ (98 vol%) at a total flow rate of 100 ml min⁻¹.

2.2. In situ QEXAFS experiments

In situ QEXAFS measurements were mainly carried out at beamline 9C of the Photon Factory, the Institute of Materials Structure Science, High Energy Accelerator Research Organization. The ring energy and current were 2.5 GeV and 430 mA, respectively. The ring was operated in a top-up mode. A Si(111) double-crystal monochromator was rotated continuously to measure a whole EXAFS spectrum of 8900 eV in 10 s. The data were processed using REX2000 (Rigaku) (Asakura, 1996). Curve-fitting analysis was performed using phase shift and amplitude functions derived from FEFF8 (University of Washington) (Zabinsky et al., 1995; Ankudinov et al., 1998, 2002).

3. Results and discussion

3.1. Correction of QEXAFS during the scan

The observed QEXAFS spectrum, can be expressed as shown in equation (1)

where are, respectively, the EXAFS oscillations of the catalyst before reaction, oscillations due to Ni—S appearing under the steady-state conditions, and oscillations of other components present only under transient conditions. Note that has a time dependence while , , are independent of time. Our previous work (Oyama et al., 2004; Kawai et al., 2006, 2008; Bando et al., 2009) indicated that Ni—S bond formation occurs with the Ni₂P framework unchanged so that c₁(t) is set constant and (1) can be reduced to

The difference spectrum is described as

Under steady-state conditions, c₂ (t = steady state) = 1, c₃ (t = steady state) = 0. Thus the difference spectrum is expressed as (k,t = steady state) = . In the transient region 0 c₂(t) 1, c₃(t) 0.

If the time, τ, is measured from the origin of each nth run, t_n, then t = , and χ(k,t) is expanded by τ at t_n,

where (t) = dc₂(t)/dt, (t) = dc₃(t)/dt, (t) = d²c₂(t)/dt², (t) = d²c₃(t)/dt².

In order to obtain the first and second derivatives, differences of sequential spectra are taken,

where the third and higher derivatives are neglected. δt is the time difference between two sequential scans: δt = t_n+1 − t_n.

Here a virtual difference spectrum δχ(k, t_n) is defined between two sequential spectra at time t_n and is approximated using (5) and (6),

Consequently the difference δχ(k, t_n) = χ(k, t_n+1) − χ(k, t_n) can be expressed as

3.2. Difference spectrum of QEXAFS in the transient state

Fig. 1 shows the difference spectra of EXAFS, Δχ(k, t = steady state) = χ₂(k), i.e. the difference in spectra of the sample before reaction and under steady-state HDS conditions (reaction temperature = 559 K). The figure also shows the change in XANES spectra. The difference EXAFS spectrum yields a single sinusoidal curve as shown in Fig. 1(b). Curve-fitting analysis revealed the formation of Ni—S bonds with bond distance = 0.230 ± 0.003 nm and coordination number = 0.10 ± 0.02, as reported in previous work (Kawai et al., 2006). There also appeared a shoulder at 8333.3 eV in the XANES spectrum before reaction as shown in Fig. 1(c). This feature decreased by 10% under steady-state conditions as shown by the dotted line. This small change was reproducible and could be used to follow changes in reaction by monitoring the intensity change in the XANES region as in previous work (Oyama et al., 2009; Bando et al., 2009, 2012).

Figure 1 (a) XAFS oscillations before reaction (solid black line) and under steady-state reaction (broken red line). (b) Difference spectrum between the XAFS oscillations. (c) Expanded figure around the edge region before reaction (solid black line) and under steady-state reaction (broken red line).

Fig. 2 shows the δχ(k, t_n) data and the Fourier transforms. The data can be divided into four regions. In region I (n = 1, 2) δχ(k, t_n) gradually increases, in region II (n = 3–5) it is relatively constant, in region III (n = 6–8) it decreases, and in the steady-state region IV (n = 9–11) it shows no signal. The δχ(k, t_n) in region II (n = 3, 4, 5) can be described with only the first term in (8) because δχ(k, t_n) ≅ δχ(k, t_n–1). The δχ(k, t_n) in region II has a similar shape to χ₂(k) within the limit of error, indicating a negligibly small contribution of χ₃(k).

Figure 2 δχ(k, tn) and their Fourier transforms. The difference spectrum between the spectra before reaction and under steady state [Δχ(k)] and its Fourier transform are given at the top.

Consequently using (6) and (7),

In regions I and III, δχ(k, t_n) showed little difference in oscillation from χ₂(k), indicating that the contribution of χ₃(k) was minimal. The χ₃(k) signal is expected to be due to the adsorption of thiophene. The lack of detection of a thiophene adsorbate in XANES could be due to thiophene being adsorbed very weakly or thiophene reacting very rapidly. It can be safely concluded that the main component in δχ(k, t_n) is proportional to χ₂(k) and to the number of Ni—S bonds.

The signal χ(k, t_n) can be reconstructed using δχ(k, t_i), the EXAFS oscillation, at time t_n as follows,

Fig. 3 shows the time dependence of [ + ] together with the intensity change in the XANES at 8333.3 eV. The two series of data points agree very well with each other. They also agree at different temperatures as shown in Fig. 1S of the supplementary material.¹ The occurrence of the same time-dependence indicates that the changes in the XANES and EXAFS signals have the same origin, i.e. the formation of Ni—S bonds. Combining the previous results of simultaneous XANES and products analysis (Bando et al., 2009, 2012), the present paper demonstrates that the assumption that the XANES change is due to Ni—S formation is valid. Hence the Ni—S bond at 0.230 nm is involved in the active phase. The reaction activity is accelerated after the formation of sufficient Ni—S bonds, indicating that a Ni—S assemble is necessary for producing high activity (Bando et al., 2009).

Figure 3 Time dependence of (open red circles) and XANES peak intensity (filled black squares).

QEXAFS has many advantages compared with DEXAFS, although the latter is a much faster measurement: (i) QEXAFS is less affected by the inhomogeneity of the sample; (ii) QEXAFS covers a wide range with good energy resolution and sufficiently high speed; (iii) the sample and beamline set-up can be shared with normal EXAFS; (iv) QEXAFS can be used with other in situ detection techniques such as FTIR.

The difference spectrum method for sequential transient data using (8) will help to enhance the time resolution of QEXAFS with little distortion.

Footnotes

¹Supplementary data for this paper are available from the IUCr electronic archives (Reference: HI5626 ). Services for accessing these data are described at the back of the journal.