High energy surface x-ray diffraction applied to model catalyst surfaces at work

Catalysts are materials that accelerate the rate of a desired chemical reaction. As such, they constitute an integral part in many applications ranging from the production of fine chemicals in chemical industry to exhaust gas treatment in vehicles. Accordingly, it is of utmost economic interest to improve catalyst efficiency and performance, which requires an understanding of the interplay between the catalyst structure, the gas phase and the catalytic activity under realistic reaction conditions at ambient pressures and elevated temperatures. In recent years efforts have been made to increasingly develop techniques that allow for investigating model catalyst samples under conditions closer to those of real technical catalysts. One of these techniques is high energy surface x-ray diffraction (HESXRD), which uses x-rays with photon energies typically in the range of 70–80 keV. HESXRD allows a fast data collection of three dimensional reciprocal space for the structure determination of model catalyst samples under operando conditions and has since been used for the investigation of an increasing number of different model catalysts. In this article we will review general considerations of HESXRD including its working principle for different model catalyst samples and the experimental equipment required. An overview over HESXRD investigations performed in recent years will be given, and the advantages of HESXRD with respect to its application to different model catalyst samples will be presented. Moreover, the combination of HESXRD with other operando techniques such as in situ mass spectrometry, planar laser-induced fluorescence and surface optical reflectance will be discussed. The article will close with an outlook on future perspectives and applications of HESXRD.


Introduction
The term 'catalysis' was coined in 1836 by the Swedish chemist Jöns Jakob Berzelius in his annual progress report in chemistry, which he wrote after having reviewed previous works on homogeneous and heterogeneous systems [1]. Therein, he ascribed the chemical decomposition of substances into their elements and their subsequent recombination into new products to a 'catalytic force' (ancient Greek: 'κατ αλυσιζ ' = 'dissolution') [2], which he concluded to emanate from materials, 'catalysts', that stay indifferent during the process.
Today, about 180 years later, it is Berzelius' 'catalytic force' that has entailed several noble prizes 6 and makes, as integral part of chemical industry and of numerous scientific disciplines, our modern every-day life tick. Hence, it is by the aid of catalysts that crude oil is transferred into processable plastics [3,4] or into the lead-free, low-sulfur gasoline that meets increasingly stringent environmental regulations [5,6]. In fact, more than 90% of all fine chemicals, such as syngas, nitric and sulphuric acid, see in their industrial production process at least once a catalyst [7]. The increasing demand for catalyst materials in fuel cells and exhaust gas control systems makes them vital for the world's energy supply and the conservation of our environment [8][9][10]. And it is the catalyst-based production of ammonia and its nitrates employed in fertilizers that guarantees the required vegetable food of two thirds of the world's steadily growing population [11].
Besides preserving the amenities of our modern world, catalysts play a crucial role in the solving of our current and future global challenges such as global warming and environmental pollution. Hence, the development of catalysts that transform greenhouse gases such as CO 2 and CH 4 into valuable products such as the sustainable fuel methanol [12,13], that facilitate the development of sustainable polymers to replace current petroleum-based plastics [14] or that even transfer plastics to fuels [15], has progressively moved into the centre of interest of catalysis research.
As a consequence, catalysis has a significant economic impact. This is fortified by the scarcity of noble metals, which is most often the main ingredient of a catalyst, as well as by the regular shutdown of industrial plants due to catalyst deactivation and renewal. Accordingly, a sustainable handling of catalyst materials and an improvement of catalyst efficiency and lifetime is of utmost significance.
To improve catalyst performance, a detailed understanding of the correlation between catalyst surface structure on the atomic scale, the gas phase and the catalytic activity is inevitable. This is however hampered by the complexity of the catalyst morphology, featuring monometallic and alloy nanoparticles on dispersed oxide supports (see inset of figure 1). Furthermore, the catalytic reactions are carried out at elevated temperatures and pressures. To bridge the pressure and materials gap-i.e. the discrepancy between (1) experimentally accessible simplified model catalysts at With the advent of high-pressure compatible in situ and operando techniques in the last three decades such as high pressure scanning tunnelling microscopy (HP-STM) [17], high pressure atomic force microscopy (HP-AFM) [17], x-ray absorption spectroscopy (XAS) [18,19], high pressure x-ray photoelectron spectroscopy (HP-XPS) [20][21][22], polarizationmodulation infrared reflectance absorption spectroscopy (PM-IRRAS) [23][24][25], summed frequency generation (SFG) [26], environmental transmission electron microscopy (E-TEM) [27], grazing incidence small angle x-ray scattering (GISAXS) [28] and surface x-ray diffraction (SXRD) [29], and with the investigation of more complicated sample systems, such as vicinal surfaces and epitaxial nanoparticle systems, the gaps are being progressively bridged.
One of these novel techniques is high energy surface x-ray diffraction (HESXRD) [30,31], which will be discussed in this review. HESXRD, that uses high photon energies of typically 70-80 keV, has in recent years become feasible thanks to advances in synchrotron technology and due to a progressive detector development. HESXRD has proven to allow a fast data acquisition with a very high resolution of reciprocal space, which is needed to catch and determine transient surface structures usually present under dynamic reaction conditions. In this review, the working principle of HESXRD will be elucidated and compared to conventional SXRD performed at lower photon energies of 10-20 keV (section 2). Thereafter, an overview over HESXRD investigations performed on model catalysts of increasing complexity will be given (section 3). Furthermore, the combination of HESXRD with other operando techniques reported so far will be summarized (section 4). This review paper will close with a summary and an outlook on future perspectives of the technique itself and its combination with other techniques.

Basic concepts of HESXRD and its experimental realization in catalysis research
Due to the continued improvement of synchrotron source performance, insertion device characteristics and detector technology, the use of synchrotron-based high photon energy x-rays (E 40 keV) has in recent years progressively increased [32]. This includes their application in a wide range of different investigation methods, which all benefit in their own way from the properties of high photon energy x-rays.
The first part of this section, subsection 2.1, will give an overview of the characteristics of high energy x-rays and how their advantages compared to conventional photon energies are exploited in various characterization techniques. Subsection 2.2 will then zoom in on the technique of HES-XRD and its measurement modes. Based on a mathematical approach, the working principle of HESXRD will be elucidated and compared to the one of conventional SXRD at lower photon energies (E = 10-20 keV) [31]. It will be explained why and to which extent HESXRD allows for probing large areas in reciprocal space with a stationary detector and sample geometry and it will be discussed how sample imperfections aid in enlarging the signal-detectability range. In this section also the monitoring limitations of HESXRD will be discussed.
Finally, subsection 2.3 will illustrate how HESXRD experiments for catalysis studies can be realized, giving an overview of typical high energy x-ray beamline set-ups and sample environments for in situ and operando investigations of model catalyst samples.

Properties of high energy x-rays and their application
With increasing photon energy the x-ray properties continuously change. A prominent property of high energy x-rays is the flattening of the Ewald sphere which comes along with a reduction of the scattering angle 2ϑ. To illustrate this, figure 2 displays scattering angles 2ϑ for selected Rh Bragg reflections (left vertical axis) as a function of the photon energy E. The right vertical axis moreover illustrates the distance of the diffracted signals with scattering angles 2ϑ on a 2D detector to the position of the direct x-ray beam, assuming a typical sample-detector distance of 1.5 m. It clearly shows that in the HESXRD regime an increased number of different diffraction signals can be probed when using a large stationary 2D detector. Such typical area detectors include the Perkin Elmer (i) or the Pilatus CdTe 2M detector (ii), whose respective sizes are indicated at the bottom right of figure 2. A potential deterioration of the image resolution that comes along with the compression of reciprocal space into a certain detector pixel size [32] may be overcome by a large enough sample-detector distance.
As the absorption of x-rays strongly decreases as a function of increasing x-ray photon energy via μ ∼ 1 E 3 (where μ refers to the x-ray absorption coefficient), another property of high energy x-rays is their lower absorption and accordingly their higher penetration depth in matter.
A reduced x-ray absorption moreover results in a reduced amount of power injected into the sample, and a reduction of electron excitation events is expected when using high photon energies [32]. This is because although the crosssection for Compton scattering, in which outer shell electrons are ejected and the sample atoms become ionized, is increasing as a function of the x-ray photon energy, the crosssection for the photoelectric effect, which may leave the sample atoms in multiply ionized states via the emission of additional Auger electrons, is decreasing at the same time.
These characteristics of high photon energy x-rays are exploited in different measurement techniques and experimental strategies. Hence, the use of large 2D detectors and the vast reciprocal space area accessible makes in powder diffraction and pair distribution methods a fast data collection possible [33,34]. The probing of a vast reciprocal space area with a very high time resolution (typically 0.5 s) is also exploited during in situ and operando powder diffraction investigations of tool material behaviour under working conditions [35] or fuel cells during electrochemical reactions [36]. The high photon energy x-rays' greater ability to penetrate matter moreover facilitates to probe either sample bulk properties, as is needed for the investigation of thicker sample slabs [37], or interfacial phenomena and deeply buried interfaces [38]. This includes both, solid/solid and liquid/solid interfaces [39][40][41] which can for instance be studied using high energy x-ray reflectivity, where the vast q-space accessible allows for extracting the microscopic density profiles across the interfacial layer. The x-rays' increased penetration depth moreover facilitates the use of special sample environments for in situ and operando studies that require thick walls of materials other than Beryllium or Kapton. Examples include novel reactor gas cells for operando catalysis studies up to 1000 bar gas pressure [42] and largevolume cells for the study of pressurized gas/solid, liquid/solid and liquid/liquid interfaces up to 100 bar pressure [43]. Their reactor walls are made up of materials such as sapphire, quartz or stainless steel.
High photon energy x-rays also allow for the advance of already existing x-ray based techniques. Thus, the use of high energy x-rays in XAS allows for recording spectra with excellent signal-to-noise ratio at the K-edges of heavy elements [44,45]. By means of high energy multi-grain diffraction information on the grain orientation of polycrystalline materials under operando conditions can be obtained [46]. It is the high photon energy x-rays that make it possible to deduce information on the local structure of solids by means of the investigation of diffuse scattering [47]. And it is thanks to the high brilliance available at 3 rd and 4 th generation synchrotrons and the improved detector technology that high energy x-rays can be used to study the sample surface structure under normal incidence of the x-rays in a low energy electron diffraction (LEED)-like manner [48], environments inaccessible with LEED.
The small scattering angles present when using high energy x-rays moreover lead to a rather fixed sample position during the measurements. This results in having more physical space around the sample environment which makes the combination with further measurement techniques possible. These include for instance diffuse reflectance infrared spectroscopy with both x-ray diffraction (XRD) or pair distribution function measurements [49][50][51][52] (for the combination of various operando techniques see also section 4).

Employing high energy x-rays in SXRD
The use of high photon energy x-rays has in recent years also led to advances in SXRD. Developed in the 1980s, SXRD was originally carried out at photon energies of typically 10-20 keV, referred to in the following as conventional SXRD [53][54][55][56][57][58]. Within the past 10 years, high photon energy x-rays of typically 70-80 keV along with large-area detectors have increasingly been applied in surface sensitive XRD [30,31]. This combination of HESXRD and 2D detectors leads to a significantly faster data acquisition thanks to the almost distortion-free tracking of diffraction patterns over a large q-range, which is essential for the tracking of transient phases under catalytic operando conditions and for the saving of costly synchrotron beamtime.
In the course of the past years especially two modes of operation of HESXRD have demonstrated to be successful, these will be discussed in section 2.2.1. Recently, we have moreover described the working principle of HESXRD for the operando investigation of single crystal and nanoparticle model catalysts at work [31], which will be summarized in section 2.2.2. This overview will also include how the calculations for the scattering angles at conventional photon energies can be transferred to the high energy regime. Moreover, in section 2.2.3, it will be discussed what determines the detectability range of reciprocal space, and how it is enhanced by sample imperfections. Finally, section 2.2.4 will present the monitoring limitations of the technique.

Measurement modes in HESXRD.
In recent years two HESXRD measurement modes have proven to be very The first mode of operation is rocking scan measurements. During these, images are taken with a 2D detector while the sample is rotated around its surface normal. Since every 2D image probed at a certain rotational angle yields a 2D image of diffraction signals in a plane that corresponds to a 'slice' of reciprocal space, a rotational scan can yield a complete 3D data set of reciprocal space. Using HESXRD its measurement typically takes 5-10 min and is accordingly much faster compared to the probing of reciprocal space using conventional SXRD at photon energies at 10-20 keV, for which the measurement of a comparable data set is on the order of hours or even days. Accordingly, if an operando catalysis experiment allows for keeping the same stable steady-state reaction condition during the duration of a rocking scan, these measurements allow for obtaining a complete HESXRD data set that can be used for the atomistic surface structure determination under the probed operando conditions. Furthermore, HESXRD rocking scan measurements are also very favourable if the sample system is not well known as diffraction signals from unknown or unexpected structures are not missed when the complete reciprocal space is being probed. A sketch of the principle of rocking scan measurements is depicted in figure 3(a). HESXRD data that were obtained using rocking scans can in this article be found in figures 5(c), 8(c) and (d), 10(b)-(d), 11(a)-(e), 13(a)-(e), 14, 16, 18(a)-(c), 35(e) and 36.
A drawback of such rocking scan measurements can be seen in the fact that changes in the HESXRD diffraction pattern during operando measurements, for instance when changing the experimental conditions, cannot temporarily be resolved. This is possible with the second mode of operation, which corresponds to static measurements using a fixed sample position. During these measurements the stationary sample is aligned to a reciprocal space plane of interested, which is monitored (typically with a temporal resolution of 0.1-0.5 s) while for instance the gas composition is being changed. This allows for the time-resolved probing of a reciprocal space plane of interest, which is especially useful if the sample system including its potential surface structures to appear during the experiment are well known. However, since not the full reciprocal space is being probed, diffraction patterns of unknown structures could be missed. The principle of such time-resolved measurements is shown in figure 3(b). HESXRD data that were Why a fast data acquisition when using these two measurement modes in HESXRD is possible will be elucidated in the following in sections 2.2.2 and 2.2.3.

Working principle of conventional SXRD and HESXRD.
In surface sensitive XRD an x-ray beam impinges on a sample surface which gives rise to a diffraction pattern probed by a detector. In catalysis research the probed samples are often model catalysts such as single crystal surfaces, vicinal surfaces, as well as epitaxial and faceted nanoparticles supported on oxide single crystals (see figure 1). Accordingly, the samples feature a certain surface and bulk crystalline periodicity with well-defined surfaces. As a consequence, the diffraction pattern typically consists of well-defined Bragg peaks from the sample bulk, as well as of crystal truncation rod (CTR) signals. CTRs are interference effects resulting from the truncation of (and/or from electron density changes within) the crystal bulk, and can accordingly be traced back to the presence of the sample's surfaces and/or interfaces. CTRs feature intensity modulations of non-zero intensity that are oriented perpendicular to the corresponding surface/interface and that interconnect the bulk Bragg peaks [53][54][55][56]. They may not only stem from the surfaces of single crystal samples, but also from the various terraces of vicinal surface samples, as well as from the facets of epitaxial nanoparticles, which give in a diffraction experiment rise to CTR-like signals that interconnect the particle Bragg peaks.
When investigating single crystal or vicinal surfaces, it is the intensity modulation along these CTRs in between the Bragg peaks that contain direct information on the atomic surface structure and which accordingly needs to be probed during the measurement. In the case of epitaxial nanoparticles the orientation and signal intensity of the CTR-like signals contain information on the quantitative particle shape and size [59]. Further information on the nanoparticle structure is encoded in the particle Bragg peak vicinities such as the Bragg peak width along different reciprocal space directions, which accordingly also need to be probed in an experiment [59,60].
The diffraction geometry for such an experiment using surface sensitive XRD is depicted in figure 4 for high (a) and conventional x-ray photon energies (b). It is the x-ray beam's wavevector −→ k i with magnitude k = 2π λ = E c , and hence the photon energy E, that defines the radius of the Ewald sphere which, as can be deduced from figure 4, scales with the photon energy. In an experiment, there is trade-off to be made between signal intensity and surface sensitivity, where the latter can be improved by using incident angles of the x-ray beam onto the sample surface that are typically smaller than the sample material's critical angle. This increases the signal-to-noise ratio for diffraction from the surface [58]. When employing a constant grazing incident angle the measurement technique is referred to as grazing incidence SXRD. In the following, and also for all experimental results discussed throughout this review, such a grazing incidence geometry has been used. The grazing incident angle is in the diffraction geometry of figure 4 referred to as α.
To measure the diffracted intensity at a distinct lattice point in reciprocal space given by c * is probed using high (a) and conventional photon energies (b). Note that the total scattering angle 2ϑ is greatly reduced for high photon energies.
To deduce the surface structure of a single crystal or vicinal surface, the diffraction intensities along CTRs need to be acquired. At conventional photon energies each reciprocal space point along a CTR can be reached by changing the detector angles γ and δ, as well as the sample rotation θ (γ, δ and θ in the laboratory frame). In the limit of α ≈ 0 the diffraction angles γ and δ with respect to the surface plane correspond to the detector angles γ = γ and δ = δ . These are connected to the sample rotation θ when following a CTR perpendicular to the surface by the following parametrization [55]: In equation (3) 2ϑ corresponds to the total diffraction angle illustrated in figure 4. Moreover, θ B in equations (1) and (2) corresponds to the in-plane Bragg angle defined by n −→ q || = 2k sin(θ B ), where −→ q || expresses the in-plane component of the total momentum transfer −→ q in the limit α ≈ 0 and γ ≈ 0, and n corresponds to an integer.
When using conventional photon energies, the resulting high Ewald sphere curvature and the large scattering angles make extended sample and detector movements necessary. This is illustrated in figure 4(b), where it is shown that for the probing of three adjacent CTRs, the detector position needs to be varied to a great extent. These large sample and detector position movements required for conventional photon energies are time-consuming. Especially during conventional CTR measurements using a point-detector, rocking scans (θ scans) need to be performed at each point along a CTR to obtain the integrated intensities at different L-values. These rocking scans contain the intensity measured while rotating the sample around its surface normal. Although the use of 2D detectors has significantly sped up the measurement time for CTRs since rocking scans are not needed to be performed at higher L-values, rocking scans are still required at lower L-values for conventional low photon energies [31].
This changes when high photon energies are employed since the Ewald sphere flattens and the scattering angles are greatly reduced, making an almost distortion-free monitoring of a large reciprocal space area with a 2D detector possible. Figure 4(a) illustrates how the intersection points of the three aforementioned adjacent CTRs with the Ewald sphere become detectable by a stationary detector in a single image when using high photon energies. Accordingly, no extended sample or detector movements are required, and even at low L-values no rocking scans need to be performed, which significantly speeds up the data acquisition. Furthermore, the high photon energies make the small angle approximation feasible, which allows for a simplification of the scattering angle parametrization as follows: Note that for both energies the same in-and out-of-plane momentum transfer range of 70 and 60 mm −1 , respectively, was covered. (c) Sum of images measured during a rotational scan on a metallic Ir(001) surface, the metallic Ir CTRs are visible as almost straight lines. Figure partly adapted from [31].
The thus obtained equations for the scattering angle parametrization at conventional ((1)-(3)) and at high photon energies ((4)-(6)) yield a direct comparison of the beam trajectories along CTRs in the (δ, γ)-plane. The trajectories are plotted in figures 5(a) and (b) as a function of the in-and outof-plane scattering angles δ and γ, where both plots cover the same in-and out-of-plane momentum transfer range of 70 and 60 mm −1 , respectively.
As can be inferred from figure 5(a), the trajectories at conventional photon energies are curved and level off at a certain γ-value. To follow the trajectory at γ-values above this levelling off, the sample rotation θ needs to be varied to a great extent. The levelling off occurs for small in-plane momentum transfers (small δ values at γ = 0) already at low out-of-plane momentum transfers. This explains why only a limited outof-plane range can be measured for small in-plane momentum transfers at conventional photon energies.
Contrary, for high photon energies, the CTR trajectories in the (δ, γ)-plane correspond to almost straight lines. This implies that almost no sample rotation θ needs to be performed to follow the trajectories. At the same time a wide out-of-plane momentum range can be accessed for small in-plane momentum transfers. To illustrate this, figure 5(c) shows the sum of 2D diffraction images measured during a rotational scan on a metallic Ir(100) single crystal surface, in which the CTR signals of the metallic surface are evidenced as almost straight lines.

Sample-and photon energy-dependent signal-
detectability range. To completely follow the CTR trajectories described above, a variable sample rotation θ is required, even if the extent of rotation is significantly reduced for higher photon energies. Especially during catalytic operando investigations, the surface processes are usually too fast to be followed by means of sample rocking scans, and time-resolved measurements of a large reciprocal space area with a stationary sample (fixed sample rotation θ) are desired.
We have shown that the probing of large reciprocal space areas with a fixed sample rotation θ and a stationary detector are feasible thanks to intrinsic sample imperfections [31]. In the case of single crystal model catalysts such imperfections correspond to finite size crystallite domains that feature angular distributions ΔΘ on the order of 0.05-0.1 • around the main direction. In the case of epitaxial nanoparticle systems it is the usually weak interaction between the particles and the single crystal oxide substrate that results in a particle angular mosaicity distribution ΔΘ around the particles aligned with respect to the substrate lattice of typically 1-3 • .
To illustrate how the in-plane mosaic spread of crystallites/nanoparticles increase the signal-detectability range, a typical scattering geometry is depicted in figures 6(a) and (b) for a single crystal and a nanoparticle sample, respectively. The in-and outgoing wave vectors −→ k span the Ewald sphere. The crystallites (a) and the particles (b) that are aligned along the main direction are depicted in orange. Their CTRs are indicated by the orange dotted line, and intersect the Ewald sphere at a certain point. This gives rise to a diffraction spot on the detector, whose position depends on the scattering angles γ, δ and the sample rotation θ. Without any mosaic spread of the crystallites/particles the diffraction pattern would be limited to this single sharp diffraction spot only. However, the crystallites/particles rotated counterclockwise by +ΔΘ (clockwise by −ΔΘ) with respect to the main direction (indicated in orange) give rise to additional CTRs indicated in blue (green) that intersect the Ewald sphere at different positions along the CTRs. Accordingly, the mosaicity distribution ΔΘ results in the fulfillment of the diffraction condition for an extended part along the CTRs, despite a fixed sample rotation θ.
Additional, but less pronounced, contributions to an enlargement of the detectable reciprocal space can be found in the finite width D of the crystallites/particles that leads to a broadening Δ To illustrate the amount of reciprocal space area that can be probed with a fixed sample and detector position, the estimated accessible reciprocal space area for a nanoparticle mosaic distribution of ΔΘ = 3.5 • is displayed in figures 7(a) and (b) as green area for high and conventional photon energies, respectively. The accessible area was calculated using the parametrization in equations (1)- (6). It is made up of half circles in the (δ, γ)-plane for which the scattering condition at a certain θ x within the ΔΘ range is fulfiled. The half circle in the (δ, γ)-plane that corresponds to the particles that grow aligned with the substrate lattice is plotted as solid black line and  runs through the particle (111) Bragg peak. Figure 7 illustrates that a much wider reciprocal space area can be probed when using high photon energies, which makes the time-resolved measurement (typical temporal resolutions: 0.1-5 s) to follow atomic-scale structural changes with a stationary sample and detector possible.
The widening of the signal-detectability range in reciprocal space discussed so far can be traced back to the in-plane mosaic distribution of crystallites and particles that are rotated around the sample surface normal. In the following, the contribution to the signal-detectability range that can be traced back to the out-of-plane mosaic distribution of the crystallites/particles will be discussed. In this case the crystallites and particles are tilted around axes that lie in the sample surface plane. This is illustrated in figure 8(a), which sketches crystalline islands/particles with different degrees of rotation around an in-plane axis, figure 8(b) shows the corresponding scattering geometry with CTRs whose varying degrees of tilt correspond to the islands' degrees of rotation. For islands whose surface is perfectly aligned with the substrate surface (island 1) a Bragg peak along the specular (00L) CTR cannot be probed since there is no intersection point with the Ewald sphere which is always pinned to the origin of reciprocal space. Contrary, for islands that feature a certain degree of out-of-plane mosaicity (islands 2-4), the tilt of the corresponding out-of-plane (00L) CTR can lead to the intersection of a diffraction spot with the Ewald sphere. This can make the diffraction spot detectable in a measurement (island 3).
As is illustrated in figure 8(b), the tilt angle α, required to make the detection of the Bragg peak possible, is given as sinα = d 2 d 1 . Therein, d 1 corresponds to the distance between the reciprocal space origin and the Bragg peak of interest on the specular rod, and d 2 that can be expressed as |k| is the length of the wavevector of the incident x-rays.
To estimate a typical value for α needed to make Bragg peaks of interest along the specular rod in an experiment visible, figure 8(c) shows as example a diffraction map measured on a highly oxidized Pd(001) single crystal. It  figure 8(c)) that is required for the oxide islands to make the (004) Bragg peak visible. The Debye Scherrer ring segment that runs in figure 8(c) through the (004) Bragg peak of the PdO(101) bulk oxide can be traced back to bulk oxide islands that feature an out-of-plane mosaic distribution, where figure 8(d) shows the intensity distribution along the ring plotted as a function of the polar angle α * . The fit to the curve reveals a full width at half maximum of 9 • which clearly shows that the tilt of the oxide islands is in this case sufficient to make the (004) Bragg peak visible.
In an experiment the influence of both, the in-and out-ofplane mosaicity, is superimposed, which results in an extended accessible reciprocal space area. To illustrate this, figure 9(a) shows as example a 2D image measured with a fixed sample position on as-prepared Pt nanoparticles supported on a MgAl 2 O 4 (001) single crystal substrate. The 2D image is centred around the particle (111) Bragg peak of the (001)-oriented particles, from which the size and shape of the Pt nanoparticles can be obtained. Due to the particle mosaicity distribution, even a much wider area becomes visible. Hence, also signals from domains of (110)-oriented particles, from domains of (111)-oriented particles, as well as from stacking faults (internal particle defect sites) are visible, yielding immediate information on the epitaxial relationships. The visibility of the particle Bragg peaks along the specular rod can be traced back to the particles' out-of-plane mosaicity. Figure 9(b) shows a 2D HESXRD image acquired on (001)-oriented Rh nanoparticles supported on MgAl 2 O 4 (001), where a zoom in on the (111) particle Bragg peak is displayed. In this case clear particle facet signals are discernible, from which the truncated octahedral particle shape can be deduced.

Monitoring limitations of HESXRD.
Despite all the advantages discussed above, the HESXRD technique also faces certain limitations. A main limitation for the measurements is that only samples can be investigated which feature structures of crystalline periodicity. This means that for instance amorphous structures cannot be studied. Moreover, in order to be able to distinguish the diffracted signals from the signal background, the crystalline structures to be probed are required to feature (1) a certain size, and also (2) a certain coverage on the sample surface. Accordingly, for the investigation of epitaxial nanoparticle samples (experiments described in section 3.3) the nanoparticles feature typically a minimum percental coverage of 30% on the sample surface. At the same time, in order to be able to distinguish nanoparticle facet signals (for instance in subsections 3.3.2 and 3.3.3), the corresponding nanoparticle facets require to consist of a minimum of around 30 surface unit cells. In order to be able to distinguish the formation of the surface oxide on a Pd(001) surface, a certain minimal surface coverage is required (see also sections 4.2 and 4.3).
A further limitation of the HESXRD technique can be seen in the fact that due to the shallow incident angles of the x-ray beam (see section 2.2.2) the sample temperature needs to be kept constant during the measurements since the sample would otherwise become immediately misaligned. Accordingly, in order to probe different reaction conditions operando, only the gas flow ratios may be varied while keeping a constant reactor pressure and a constant sample temperature. This was the general approach used in the experiments described in sections 3 and 4.
Finally, it should be noted that especially for single crystal surfaces a temporal resolution down to 0.1 s per image can be obtained. In the case of epitaxial nanoparticle samples a counting time of typically 20 s is required to probe a reasonably good diffraction image as the diffracted signals from nanoparticles are weaker.

Experimental realization
The HESXRD investigation of model catalysts under in situ or even realistic reaction conditions requires sophisticated sample environments and beamline instrumentation that will be discussed in the following.

Sample environments for catalytic in situ and operando
studies. For the in situ SXRD investigation of model catalysts during exposure to static gases, e.g. to O 2 , different UHV/HP batch chambers have been designed and used, the prototype of which was described by Bernard and coworkers in 1999 [62]. The basic concept of the chambers available nowadays is the same, but they differ in their maximum pressure and/or the maximum sample annealing temperature. All chambers comprise a total volume of around 2 l and feature a cylindrical Beryllium window that allows the entrance and exit of the x-rays. They allow for both, UHV sample preparation and x-ray measurements under atmospheric pressures. The UHV sample preparation is made possible by a pumping stage resulting in a typical base pressure in the low 10 −9 /high 10 −10 mbar range, by gas lines and leak valves and the possibility to sputter and anneal the sample. To allow for x-ray investigations under gas exposure, valves can close off the sputter gun and the pumping stage, and high pressures can be reached. The experiments reported in sections 3.3.1 and 3.3.2 have been performed using such setups.
For the operando SXRD investigation of model catalysts under steady gas flows and reaction conditions, a dedicated in situ catalysis chamber [63] has been used in the experiments described in sections 3.1, 3.2, 3.3.3 and 3.3.4. This setup also allows for a UHV sample preparation by sputtering and annealing (typical base pressure in the low 10 −9 mbar range), as well as for the SXRD investigation of the sample during catalytic reaction with pressures up to 1 bar. For the operando investigations, the sample is mounted inside the chamber's beryllium dome (11.5 ml volume) which acts as the flow reactor for the catalysis experiments. The total gas flow into the reactor and its gas composition is regulated by mass flow controllers. They are installed on each of the individual gas lines, which are led together prior to the reactor inlet to ensure a mixture of the reactant gases. The total reactor pressure is regulated by a back pressure controller installed in the reactor outlet line. The reactor volume is sealed off from the UHV part of the system except for a small gap, which makes leaking into a mass spectrometer possible. In some experiments an additional mass spectrometer was mounted to a small UHV volume connected to the outlet gas line via a leak valve. To realize the UHV sample surface preparation, the reactor volume can be opened to the UHV volume of the system. As a consequence, the chamber ports, which host surface preparation equipment such as sputter guns, point towards the sample surface and render the surface preparation possible.
In an operando catalysis experiment it is often of great interest to probe the structural changes of the catalyst at conditions that lead to catalytic light-off, i.e. conditions that result in highest catalytic activity. For the CO oxidation reaction, the main catalytic reaction discussed in this review article, the catalytic light-off can typically be found for a stoichiometric CO:O 2 gas flow ratio of 2:1, where the light-off sample temperature typically lies between 450 and 600 K for the late transition metals of Pt, Rh and Pd discussed in this article.

Beamline setup and instrumentation.
The HESXRD experiments discussed in this review were carried out at the Materials Chemistry and Materials Engineering Beamline ID15A, ESRF (experiments discussed in sections 3.3.1 and 3.3.3) and at the High Energy Materials Science Beamline P07, Petra III, DESY [64] (experiments discussed in sections 3.1, 3.2, 3.3.2 and 3.3.4). The principle beamline setup and the setups within the beamline experimental hutches are comparable. In both cases the white beam is monochromatized by a Si(111) double crystal monochromator in Laue geometry, which, at the time the experiments discussed in this article were performed, allowed for tuning the photon energy in a range between 40 and 300 keV (ID15A, energy bandwidth: (ΔE/E) = 2 ×10 −3 ) and between 50 and 200 keV (P07, (ΔE/E) = 2 ×10 −3 ), respectively. The focussing of the beam to the sample position is performed by a variable number of Al compound refractive lenses. At the time when the experiments discussed were performed, this provided beam spot sizes (vertical × horizontal) at the sample position down to 5 × 50 μm 2 (ID15A) and 2 × 30 μm 2 (P07) that ensured that a defined sample area to be illuminated by the beam.
Due to the small critical and scattering angles present when working with high photon energy x-rays (typically on the order of 0.04 • for the critical angle, 2-4 degrees for the scattering angles), a very high precision of the sample movements is required. For this reason all experiments described in this review used a specially designed six-circle sample manipulation tower in vertical geometry [38] available at various high energy beamlines, onto which the sample chamber was mounted. The sample manipulation tower provides for both, the beam incident angle α and the sample rotational angle θ, an accuracy higher than 0.0001 • . The sample's vertical position z can be controlled with an accuracy higher than 0.1 μm.
To map the diffraction signals, 2D area detectors optimized for high photon energy x-rays are used. In the experiments reported in this review the employed detectors included a charge couple device camera ( . To block the direct beam and the scattering signals from the chambers' Beryllium windows on the detector, a combination of beamstop (direct beam) and mask (scattered signals), typically made of Tantalum and/or Densimet, is used. For the investigations of single crystal surfaces, tungsten beamstops that block the intensities from bulk Bragg peaks to protect the detector, are put directly onto a Plexiglas plate in front of the detector. These beamstops give rise to shadows on the 2D detector image (see shadows in diffraction patterns in figures 5, 8, 10, 13, 15, 16, 18, 19, 29, 33 and 36).
As of today, after the refurbishment and the long shutdown of the ESRF (December 2018-August 2020), Table 1. Overview over the high energy undulator beamlines at which HESXRD measurements can be carried out and their characteristics. The beamlines listed here are equipped with a high precision diffractometer [38] that was used in all experiments discussed in this topical review. The data shown for the ESRF ID31 beamline correspond to the characteristics after the ESRF shutdown 2018-2020.

Synchrotron
Beamline HESXRD experiments under grazing incidence are no longer performed at beamline ID15A, but at the High-Energy Beamline for Buried Interface Structure and Materials Processing ID31. This is because the high precision sample manipulation tower required for grazing incidence experiments was moved from beamline ID15A to beamline ID31. At DESY the Swedish Materials Science Beamline P21 is has opened to general users since summer 2019 and provides a similar experimental setup as at beamline P07. Further high photon energy beamlines are available at various synchrotrons around the world, including the Advanced Photon Source, Spring8 and Diamond Light Source. However, the combination of the high photon energy x-rays and the high precision sample manipulation tower are today available at beamlines ID31 (ESRF), P07 and P21 (DESY). An overview over their beamline characteristics is given in table 1.

In situ and operando HESXRD studies of various model catalyst systems
Industrial catalysts are complex material systems consisting of faceted monometallic and alloy nanoparticles dispersed on branched oxides. The nanoparticles' overall concentration of only a few weight percent and their random orientation exposing different particle facets and thus different crystal planes makes accessing the particle surface structure under realistic reaction conditions for most measurement techniques very difficult [65,66]. However, a detailed atomic-scale understanding of the catalyst surface structure and its correlation to the surrounding gas phase under realistic operando conditions is required to finally improve catalyst efficiency. A theoretical strategy to overcome the complexity of real catalysts but at the same time gaining atomistic insight into the catalyst surface structure was first expressed in 1922 by the Nobel Prize winner Irving Langmuir who stated that 'in order to simplify our theoretical consideration of reactions at surfaces' we need to 'confine our attention to reactions on plane surfaces. If the principles in this case are well understood, it should then be possible to extend the theory to the case of porous bodies' [67], thereby introducing the concept of model catalysts, implying that the investigation of a single crystal surface will yield information on the physics and the chemistry of the respective catalyst nanoparticle facet.
Being a photon-in photon-out technique, HESXRD can be used as a probe even under industrially relevant conditions of 100 bar and beyond. In the following it will be described how the technique has in recent years been applied in the operando investigation of model catalysts of various complexity. First, its application for the operando structure study of single crystal and vicinal surfaces (sections 3.1 and 3.2), mimicking nanoparticle facets and edges, respectively, will be discussed using examples of the oxidation of CO into CO 2 over different Pd surfaces [30,61,[68][69][70]. Section 3.3 will then summarize the results obtained on the study of nanoparticles epitaxially grown on oxide single crystals. This includes results obtained on the in situ oxidation of Pd nanoparticles (section 3.3.1) [71] and Pd-Rh alloy nanoparticles (section 3.3.2) [72], results obtained on operando CO oxidation investigations on Pt-Rh alloy nanoparticles (section 3.3.3) [73], as well as studies on the interaction of H 2 and methylcyclohexane (MCH) with Pd nanoparticles and the resulting diffusion of hydrogen and carbon into Pd nanoparticles (section 3.3.4) [74].

Single crystal surfaces
Single crystal surfaces of different surface crystallographic orientations mimic the corresponding catalyst nanoparticle facets and are the least complex model catalyst systems. Using single crystal surfaces, essential insights into atomic scale processes at surfaces have been gained in catalysis research. These include the unravelling of the detailed molecular mechanism of ammonia synthesis from hydrogen and nitrogen over iron single crystal model catalysts [75], as well as the identification of the undercoordinated Ru atoms in the RuO 2 (110) bulk oxide as catalytically active sites in CO oxidation [76].
Using the example of CO oxidation over a Pd(001) single crystal, this section will at first illustrate how HESXRD enables a fast 3D reciprocal space mapping under steady-state conditions, which can be used for the identification of complex surface structures under industrial reaction conditions and their analysis on the atomic-scale (section 3.1.1) [30,69]. It will be demonstrated how high-resolution diffraction patterns measured over the Pd(001)-( (101) surface oxide allowed for refining its atomistic structural model, a structure which has been discussed for decades, and which has interestingly previously been found under UHV conditions by means of LEED [77]. Section 3.1.2 will show that the fast reciprocal space mapping of HESXRD provides within 10 min a complete data set for a quantitative CTR analysis, for which the measurement time in conventional SXRD is on the order of several hours [61]. Finally, section 3.1.3 will demonstrate how HESXRD was employed to determine the surface structure dynamics of a Pd(100) single crystal during CO oxidation, ranging from a metallic surface, to an ultrathin PdO surface oxide, and finally to the formation of additional epitaxial and polycrystalline PdO bulk oxide islands [68].  [30]. Reprinted with permission from AAAS.

Fast 3D reciprocal space mapping under steady-state
conditions. Section 2.2 explained in detail why the use of high photon energies along with a 2D detector allows for probing a large area of reciprocal space with a single snapshot. In the following it will be shown how the measuring of such 2D images while rotating the sample around its surface normal (rocking scans) results in a 3D data set of reciprocal space, allowing for the atomic-scale identification of complex surface structures under operando conditions. The data discussed in the following was obtained during CO oxidation over a Pd(100) single crystal [30] (note also the supplementary video material of this reference). Under the probed conditions, consisting of a sample temperature of 600 K, a total pressure of 100 mbar, and gas flows of 2 ml n min O 2 , 4 ml n min CO and 44 ml n min Ar (where 1 ml n min corresponds to 1 ml min at 0 • C at 1013 mbar), the sample was highly catalytically active, converting all CO that reached the sample surface into CO 2 . Under these conditions, the sample was concluded to be covered by a 2D ultrathin surface oxide layer, the atomic model of which is shown in figure 10(a).
To probe the atomic structure of this surface oxide, 2D images were taken while the sample was rotated around its surface normal, where the scan comprised a total rotation of 90 • with a stepsize of 0.1 • and an exposure time of 1 s. This rotation of the real space lattice corresponds to a rotation of the reciprocal lattice and its CTRs with respect to the Ewald sphere. Whenever a rod intersects the Ewald sphere, the diffraction condition is fulfilled and the corresponding signal can be detected. Accordingly, a rotational scan results in a scanning along the CTRs and a corresponding 3D data set of reciprocal space is obtained (see also description of the rocking scan measurement mode in section 2.2.1). Figure 10(b) displays a sum over the diffraction data measured during a full 90 • rotation, where for each pixel the highest intensity measured during the rotational scan is displayed. Thus, figure 10(b) represents a 2D projection of the scanned 3D area containing all the CTRs probed in the 90 • rotation. In addition to the CTRs from the Pd(001) surface running through the bulk Pd Bragg peaks, also the superstructure rods from the 2D surface oxide layer of figure 10(a) are visible. The absence of other additional Bragg peaks indicates that the sample was only covered by the surface oxide and that no bulk oxides had formed.
The obtained 3D data set can be visualized in different ways. Thus, in-plane slices can be extracted, which corresponds to a data representation as is used in LEED. Such an in-plane map extracted at L = 0.5 is shown in figure 10(c), from which with a very high spatial resolution the position of signals from the surface oxide rods can be deduced. The close-up in figure 10(d) demonstrates that the rod signals are not in the centres of the circles, as would be expected from a perfect periodicity of the Pd(001)-( (101) surface oxide structure. This observation can be traced back to the stress-induced mismatch between the surface oxide and the Pd(100) surface. The ability to deduce such subtle discrepancies in the in-plane maps demonstrates the power of the HESXRD technique as compared to LEED, with which this mismatch cannot be deduced due to resolution limitations. Moreover, while HESXRD allows for investigating the sample under realistic reaction conditions, LEED, that uses electrons as probe, requires ultra-high vacuum conditions.
The atomic structure of the Pd(001)-( (101) surface oxide has previously been discussed for decades by several authors [77,78]. In a recent publication combining SXRD, STM and LEED, we have shown that a refined atomistic model of the surface oxide can be retrieved by means of HESXRD [69].
While the HESXRD in-plane maps displayed in figures 10(c) and (d) represent only a thin slice of reciprocal space around L = 0.5, the in-plane map shown in figure 11(a) corresponds to a 2D projection of diffraction intensities for L-values between 0 and 3. When inspecting the surface oxide rod intensities in figure 11(a) it becomes clear that they are not only shifted with respect to the expected positions for an unstressed Pd(001)-( (101) surface oxide structure (compare to red circles in figure 10(d)), but that they are also split, featuring double or even triple rods. Figures 11(b) and (c) show zoom-ins of the double and the triple rods indicated by the numbers 1 and 2 in figure 11(a). Figures 11(d) and (e) contain the same double and triple rods, respectively, shown as 2D out-of-plane and 3D representations. Such diffraction features cannot be explained by a perfectly regular and squared ( To explain the observed diffraction features of figure 11, several surface oxide models were tested [69]. The correct model also needed to agree with the STM image of figure 12(a) measured on the Pd(001)-( (101) surface oxide, where the white arrows indicate stripes of distorted periodicity. Following the black line in the STM image, the bright row, corresponding to Pd atoms coordinated by two oxygen atoms, continues with a small shift downwards, even across a distortion step. Contrary, along the blue line, the bright row does not continue, which implies a change in the Pd coordination from two to four oxygen atoms across the distortion step.
The double splitting and the shifting of the diffraction rods in the HESXRD in-plane map can be explained by four domains of a distorted surface oxide unit cell, characterized by an oblique lattice with lattice parameters |a 1 | = 6.20 Å, |a 2 | = 6.02 Å, and an angle of 88 • between them (lattice constant of non-distorted cell: a = 6.151 Å). However, this structure can neither explain the triple splitting of the rods, nor the regular stripes of distorted periodicity indicated by white arrows in the STM images of the surface oxide (see figure 12(a)).
To explain these observations, an extended superstructure unit cell of the distorted cell is required, accounting for all long-range incommensurabilities. Hence, the change in the Pd coordination number can be explained by a lateral shift along the [011] substrate direction of the surface oxide unit cell atoms with respect to the substrate unit cell by 1.375 Å, corresponding to half a substrate unit cell. Such a relative shift occurs every 14 th substrate unit cell. The surface oxide lattice is incommensurate only in one direction, and can along the other direction be well described by a (5 × 5) coincidence lattice [30]. Accordingly, a good fit to the experimental observations was achieved by describing the surface oxide unit cell as a a11 = 14, a12 = −1 a21 = 0, a22 = 5 matrix with respect to the substrate lattice.
To obtain the best fit to the experimental HESXRD data in figure 11(a) using such a distorted unit cell, several variants of the internal atomic unit cell structure featuring the lateral shift were compared. Figure 12(b) shows the model of the surface oxide unit cell for which the best fit to the experimental data was obtained. Therein, the required internal lateral shift is represented by the gradual transition of the blue atoms from ideal bridge to four-fold hollow sites. Considering four unit cell domains, not only all diffraction peaks could be reproduced but also the intensity ratios between individual split diffraction rods, as can be deduced from the size and position of the crosses in figures 11(b) and (c), which represent the respective positions and intensities of the calculated diffraction signals.
The data shown and discussed in section 3.1.1 illustrate how HESXRD provides within a measurement time of typically only 10 min a complete 3D data set of reciprocal space. Not only provides HESXRD a much higher spatial resolution compared to for instance LEED, but it even allows for investigating samples under harsh industrial conditions at elevated pressures. In conventional SXRD at low photon energies the investigated sample system needs to be known well beforehand, since the time frame of a synchrotron beamtime allows only for probing selected areas of reciprocal space. Contrary, HESXRD provides a full overview, which yields the possibility for exploring unknown sample structures and minimizing the risk of missing unexpected features, making HESXRD an extremely powerful operando technique.

Quantitative CTR analysis.
One of the main applications of SXRD is the derivation of atomistic surface structure models via fitting the intensity modulation along selected CTRs [53,54]. As was discussed in section 2.2, the data collection using low photon energies as in conventional SXRD, is very time-consuming and on the order of several hours. Contrary, an HESXRD rocking scan yields within 10 min a full 3D data set of reciprocal space including all CTRs of interest (see figure 10(b)). Recently, we have demonstrated that the data quality of the thus obtained CTRs is even suitable for a complete quantitative surface structure determination yielding exact atomic positions [61]. This approach of the extraction and the quantitative analysis of the HESXRD data will be elucidated in the following for the investigation of the ultrathin surface oxide structure on Pd(001) during CO oxidation.
The surface oxide structure was formed by exposing the Pd(001) surface to a flow of 2 ml n min O 2 , 4 ml n min CO and 25 ml n min Ar at 600 K and 100 mbar total pressure. Under these conditions, the sample was highly catalytically active, transferring all CO, that reached the sample surface, into CO 2 [61]. To obtain the 3D data set of reciprocal space, the sample was rotated around the surface normal, resulting in a scanning up and down along the CTRs since the intersection point of the CTR and the Ewald sphere moves accordingly. During the rotational scan, To extract data for the quantitative CTR analysis, the yellow ROI was divided into smaller, equally sized parts, such as the small white rectangle shown in figures 13(a)-(c). The integrated intensity was then plotted as a function of the rotational angle θ as depicted in figure 13(d). Fitting these data with a Gaussian function and a linear background yielded the integrated intensities I int at the individual L-values along the CTR. From these the respective structure factors F str , needed for the CTR analysis, were deduced using I int = p · |F str | 2 · C tot , where p takes properties of the sample and the incident beam into account. C tot can be represented as C tot = C hp · C L · C rod · C d · C i and considers the characteristics of the experimental geometry (C hp : horizontal polarization factor, C L : Lorentz correction factor, C rod : rod interception correction factor, C d : solid scattering angle correction factor, C i : beam inclination correction factor) [79,80], where, thanks to the conditions and setup used in the present case, the amount of corrections could be considerably simplified. The thus obtained structure factors F str are plotted in figure 13(e) as a function of L and represent the overall shape of the Pd (20L)-CTR.
To deduce the atomistic structure of the ultrathin (101) surface oxide on the Pd(001) surface, a total of 9 inequivalent Pd CTRs and oxide superstructure rods were used. The experimentally obtained structure factors are displayed in figure 14 as black data points for selected rods. In the quantitative data analysis, these experimental data were compared to and fitted by structure factors that were calculated based on underlying structural surface oxide models using the ANA ROD software package [81]. The underlying structural models were based on single-layer Pd surface oxide structures that were reported and suggested previously [77,78,[82][83][84]. They consisted of a PdO(001) and a PdO(101) surface oxide as described in [77,84], respectively, as well as of a PdO(100) layer whose atomic structure was obtained by moving the subsurface oxygen of a PdO(101) layer to the surface. The atomic ball models of the thus obtained underlying Pd surface oxide structures put to the test are displayed at the bottom of figure 14.
The fit results for the different surface oxide models are summarized at the top of figure 14 and are represented by green dotted lines (PdO(001)), blue dashed lines (PdO(100)) and red solid lines (PdO (101)). While the PdO(001) model did not fit the data and could be excluded, both, the structural models for the PdO(100) and the PdO(101) surface oxide were in good agreement with the experimental data. This can be explained by the fact that the Pd lattice is the same for both structures while the structures differ only by the position of the O atoms. The latter are, however, only weak scatterers in XRD compared to Pd, and further experimental techniques such as highpressure x-ray photoelectron spectroscopy were needed to identify the PdO(101) structure as the correct atomic structure [78,85]  for obtaining a complete 3D data set of reciprocal space within the order of several minutes. Such a fast data acquisition is only feasible since already only one single 2D detector snapshot at a fixed sample position probes a vast area in reciprocal space, which is mainly possible thanks to intrinsic sample imperfections (see section 2.2.3). Recently, we have shown that the time-resolved monitoring of such a selected reciprocal space plane with a stationary sample allows for, with subsecond temporal resolution, the tracking of surface structural dynamics during catalytic processes [68] (see also the time-resolved measurement mode using a fixed sample position discussed in section 2.2.1). As example, the time-resolved tracking of transient oxide phases appearing over a Pd(100) single crystal during CO oxidation will be discussed in the following [68]. Different oxide phases have been reported on the Pd(001) surface before. These include the previously discussed epitaxial ( √ 5 × √ 5)R27 • surface oxide [30,61], an epitaxial PdO(101) bulk oxide [86][87][88], as well as a disordered polycrystalline PdO film [88,89]. However, the transition between these different structures has, prior to the studies demonstrated in the following, been unknown [68].
The experiment was carried out by recording 2D diffraction patterns with a temporal resolution of 0.5 s while stepwise changing the O 2 :CO partial pressure ratios from 1:4 over 1:2 to 1:1 and back, while the sample temperature and total pressure were kept constant at 600 K and 100 mbar, respectively. The absence of CO 2 production and hence of a catalytic activity of the sample is in line with the notion of a CO-poisoned surface, that inhibits catalytic activity [85]. Upon changing the O 2 :CO ratio to a 1:2 mixture, the sample became catalytically active as a tremendous increase in CO 2 production was evidenced. This increase in catalytic activity came along with the appearance of superstructure rods from the ( √ 5 × √ 5)R27 • surface oxide ( figure 15(b)). When further increasing the O 2 :CO ratio to 1:1 and thus to excess in O 2 , the catalytic activity remained at the same level, indicating that the reaction had reached the mass transfer limited regime. The diffraction pattern under these conditions, shown in figure 15(c), revealed the appearance of powder rings, characteristic of polycrystalline PdO. As the superstructure rods were still present when the powder rings appeared, a coexistence of the ( √ 5 × √ 5)R27 • surface oxide and the polycrystalline PdO could be concluded, indicating that the latter initially grows only locally. This coexistence lasted only for around 120 s, and thereafter the superstructure rods disappeared and the powder rings became more prominent (see figure 15(d)). This is in line with the disappearance of the surface oxide and the formation of a polycrystalline film covering the complete surface, where the monocrystallite mean size of the PdO was estimated to amount to 200 Å. After switching back to the reducing conditions of an O 2 :CO mixture of 1:4, the powder rings immediately disappeared again, in line with the recovery of the surface towards a highly ordered metallic state [68].
It was moreover concluded that, prior to the coexistence of the Pd surface oxide and the polycrystalline PdO (diffraction pattern of figure 15(c)), the surface had featured a coexistence of the Pd surface oxide and epitaxial PdO(101) bulk oxide islands, growing in an energetically favoured Stranski-Krastanov growth mode [68,86,90]. The latter was concluded from rotational scans measured at an O 2 :CO ratio  (101) bulk oxide Bragg peaks are rather sharp, they were likely to be missed in the stationary geometry used in the dynamic measurements of figure 15, but are expected to have been present for a limited time. The deduced surface structural changes as a function of the relative concentrations of the reactants are summarized in figure 15(g).

Vicinal surfaces
Vicinal surfaces consist of a regular array of steps and terraces. They are accordingly more complex model systems compared to the single crystals discussed earlier and represent the catalyst particle edges. Earlier studies have shown that the surface steps are essential catalytic sites for the adsorption and dissociation of molecules [91][92][93][94][95][96] and that they may undergo refaceting when exposed to gases or reaction conditions [97][98][99]. In the following it will be described how we have used HESXRD for the operando investigation of a Pd(553) vicinal surface structure during different conditions of CO oxidation [70]. It will be shown how the fast data acquisition facilitated by HESXRD allows for the monitoring of transient adsorption structures hardly accessible with other operando methods.
A ball model of the studied Pd(553) vicinal surface is shown in figure 17(a). It consists of four atom wide (111)oriented terraces interrupted by monoatomic (111)-oriented steps. Throughout the experiment the sample temperature and total reactor pressure were held constant at 600 K and 100 mbar, respectively, where CO, O 2 and Ar (carrier gas) were used. To probe the three different CO:O 2 ratios of interest (condition (i) O 2 :CO = 1:4; condition (ii) O 2 :CO = 2.1:4, condition (iii) O 2 :CO = 1:1) and to maintain the total gas flow of 50 ml n min , the CO flow was held constant at 4 ml n min , while the O 2 and Ar flows were adjusted accordingly. Figure 17(b) shows the partial pressures for CO, O 2 and the reaction product CO 2 as measured by in situ MS as a function of the three probed conditions in the transition from reducing reaction conditions to O 2 overstoichiometry (conditions i → ii → iii). While under reducing conditions (i) the CO 2 production and hence the catalytic activity was fairly low, it increased tremendously in the transition to slight O 2 excess (ii) and kept approximately the same level under high O 2 overstoichiometry (iii). This indicates that the catalytic reaction was under conditions (ii) and (iii) in the mass transfer limited regime, i.e. that the CO 2 production was limited by the rate of CO transported to the sample surface.
To investigate the Pd(553) surface structure, the HESXRD measurements included (1) rocking scans (θ scans) at fixed gas compositions (i), (ii) and (iii) (range: 31.2 • , step size: 0.2 • , exposure time: 1 s), as well as (2) timescans probing 2D images with a temporal resolution of 1 s at a fixed rocking angle θ while changing gas composition (i → ii → iii). In each case the measured partial pressures reproducibly showed the behaviour depicted in figure 17(b). Figures 18(a)-(c) show the rocking scans measured under conditions (i), (ii) and (iii), respectively, where in each case the data taken is projected onto a 2D image by displaying for each pixel the highest intensity probed during the scan. The rocking scan of condition (i) in figure 18(a) shows a diffraction pattern consisting of sharp vertical CTRs compatible with a well-ordered metallic Pd(553) surface. In line with previous findings [100,101], it is likely that the surface stays metallic under these conditions as it is CO poisoned, which also explains the low CO 2 production rate.
After switching the gas ratios to slight oxygen overstoichiometry (condition ii), which came along with the strong increase in catalytic activity into the mass transfer limited regime, the rocking scan data of figure 18(b) implies that a refaceting of the surface took place: while the vertical CTR signals of the clean Pd(553) surface vanished, new CTRs, tilted at different angles, appeared. They can be ascribed to the formation of (332), (331) and (111) crystallographic planes, as is indicated for the CTRs within the dashed area of figure 18(b). A real space model of the formed crystallographic planes with respect to the initial Pd(553) surface are indicated in   (111) planes. The parallel diffraction rods outside the dashed box in figure 18(b) can be assigned to the (332) facets only. Their relative distance in reciprocal space indicates that the (332) facets consist of 5 atoms wide, (111) oriented terraces separated by (111) monoatomic steps. The terrace width corresponds to the width of the PdO(101) surface oxide unit cell and was accordingly wide enough to accommodate the surface oxide's growth, which has in earlier studies been assigned as reason for the refaceting [97]. A real space model of the PdO(101) surface oxide on the (332) facet is illustrated in figure 18(e). The oxide's presence is moreover underlined by the fact that it has been found to be very reactive towards the oxidation of CO [61,101,102], which could explain the tremendous CO 2 production increase in the transition to condition (ii). Accordingly, a possible interpretation of these observations could be that the surface structure under condition (ii) features the coexistence of (332), (111) and (331) type facets with the (332) facets covered by the PdO(101) surface oxide.
After switching the gas ratios to high O 2 excess (condition iii), the CO 2 productivity kept its high level while drastic changes were observed in the diffraction data of the rocking scans (see figure 18(c)). Hence, a disappearance of the CTRs and an appearance of Bragg peaks and powder rings, whose  (101) is catalytically reactive towards the oxidation of CO into CO 2 [106][107].
The diffraction data discussed above were obtained by means of rocking scans under steady state conditions. In the following, 2D timescans measured at a fixed sample position (stationary θ angle) while changing the gas composition, will illustrate the power of the HESXRD technique for the timeresolved tracking of dynamic structural changes on the atomic scale, easily to be missed by many other techniques. Figure 19 shows a selection of 2D images with a temporal resolution of 1 s measured during the transition from reducing conditions to high catalytic activity under O 2 excess (transition conditions i → ii → iii). In the transition from condition (i) to (ii), the 2D images reveal the appearance of additional diffraction rods (indicated by white arrows in figures 19 (1) and (2)). These point towards the formation of 2D oxygen adsorbate structures with the same periodicity as the Pd(553) surface steps that would form prior to the surface refaceting. Figures 19(I)-(III) show three potential subsequent oxygen adsorbate structures as discussed previously in [97] that may have formed as function of increasing O 2 partial pressure. The time-resolved 2D images reveal moreover that these oxygen-induced superstructures coexisted thereafter with (332) and (111) facets (see figure 19(2)) prior to the complete surface restructuring into (332), (111) and additional (331) facets (compare to steady state condition (ii) of figure 18(b)). The data moreover show that the ultrathin 2D adsorbate structures were stable for about 30 s.
The time-resolved data measured in the transition from condition (ii) to the high O 2 excess of condition (iii) revealed that first the (331) CTRs, then all other rods disappeared (see figure 19(3)), implying a roughening of the surface, while signals from the PdO(101) bulk oxide were already present but still weak (see grey circle in figure 19(3)). The tracking of the transient features discussed above would very likely have been missed when using conventional photon energies, which reveals the power of HESXRD as an in situ and operando technique.

Epitaxial nanoparticles on single crystal oxide supports
Epitaxial metal nanoparticles on oxide supports constitute the most realistic but also the most complex model catalyst systems discussed in this topical review. An understanding of such complex systems under reaction conditions is, however, essential, as the interplay and the 'crosstalk' between neighbouring particle facets, the presence of particle edges and corners, as well as the influence of the support cannot be neglected in the search for the catalytically most active sites [108,109]. Moreover, particle sintering, i.e. the agglomeration of smaller particles into larger ones, constitutes a main reason for catalyst deactivation and process shutdown in catalyst industry [110,111]. Accordingly, it is of highest scientific and economic interest to find means to suppress it by understanding the underlying mechanisms on the atomic scale.
Nanoparticles may grow epitaxially on a single crystal substrate surface if certain nanoparticle atomic row distances are similar to the ones of the substrate. This results in the alignment of the nanoparticle lattices to that of the substrate. The model systems discussed in the following feature nanoparticles consisting of the fcc metals Pt, Rh, and Pd and their alloys, supported on MgO(001), MgAl 2 O 4 (001) or Al 2 O 3 (0001) single crystal oxide substrates [71][72][73], they are sketched in figure 20. The cubic unit cells of MgO(001) and MgAl 2 O 4 (001) result predominantly in the growth of (001)-oriented particles that feature a truncated octahedral shape with a (001)-type facet at the interface and as top facets. The hexagonal unit cell of the Al 2 O 3 (0001) substrate features the growth of mainly flatshaped (111)-oriented particles with (111)-type facets at the interface and as top facets. For the fcc metals discussed here, the facet surface energy is lowest for the (100) and (111) surfaces. Accordingly, both type of particle shapes are dominated by (001)-and (111)-type facets (see figure 20).
When investigating epitaxial nanoparticle systems by means of surface sensitive XRD, a multitude of information is encoded in the vicinity of the particle Bragg peaks. Information on the particle height, diameter and shape can be deduced from the peak width along certain directions. Information on the presence of nanoparticle facets and the particle shape can be obtained from potential particle facet rod signals. And the particle lattice parameter and thus the particle alloying can be deduced from the respective Bragg peak positions.
For the nanoparticle systems discussed in the following, the particle Bragg peaks of interest are located in high symmetry reciprocal space planes that contain also further diffraction signals of interest. These include potential signals of bulk oxides that may form, of internal particle defect sites, as well as of different nanoparticle epitaxies. Accordingly, probing these reciprocal space planes by means of HESXRD with a large 2D detector provides a majority of the information of interest for epitaxial nanoparticle systems under reaction conditions. The need to probe only one reciprocal space plane renders a stationary sample position possible, which allows for (a) time-resolved studies as well as for (b) the investigation of combinatorial nanoparticle sample systems as the ones discussed in the following. These feature the nanoparticles grown in stripes on the sample surface, where, from stripe to stripe, a gradient in nanoparticle size and/or alloy composition is present. Accordingly, a combinatorial sample allows for the composition-and size-dependent HESXRD investigation of supported nanoparticles under exactly the same experimental conditions, where one particle stripe at a time can be studied by translating the sample perpendicular to the incident beam direction (see figure 20).
In the following an overview over HESXRD investigations that used this experimental approach will be given. They include our HESXRD studies of the particle-size-dependent oxidation and oxide formation of MgO(001)-supported Pd nanoparticles (section 3.3.1) [71], and of the compositiondependent oxidation and oxide formation of Pd-Rh nanoparticles grown on MgAl 2 O 4 (001) (section 3.3.2) [72]. It will be elucidated how we used intensity modulations in the particle Bragg peak vicinities to obtain direct information on the composition-and particle shape-dependent sintering of Al 2 O 3 (0001)-supported Pt-Rh nanoparticles during CO oxidation (section 3.3.3) [73]. The last section will describe how HESXRD was used to deduce the incorporation of hydrogen and carbon into Al 2 O 3 (0001)-supported Pd nanoparticles during dehydrogenation of liquid organic hydrogen carriers (section 3.3.4) [74].

In situ oxidation of Pd nanoparticles.
It has been found that the catalytic activity of a catalyst may strongly depend on the size of its nanoparticles . Reasons for this can be found in the fact that the electronic properties may change for reduced dimensions, which may promote/suppress the catalytic activity. Furthermore, when scaling down the nanoparticle size, the number of undercoordinated metal atoms at edges and corners increases with respect to the number of metal atoms of the particle facets, which may also have an impact.
In the experiment discussed in the following, we shed light on the particle size-dependent oxidation of (001)-oriented Pd nanoparticles epitaxially grown on MgO(001) [71] (see also sample 1 in figure 20). The study dates back to the year 2008 and pioneered in the use of HESXRD for the investigation of combinatorial nanoparticle model catalyst samples. The Pd nanoparticles featured a diameter size range from 4 to 24 nm, where the different particle sizes could be individually studied by translating the sample perpendicular to the incident beam direction (see figure 20). Three particle size regimes with distinct oxidation mechanisms could be identified, which correspond to the particle diameters of D = 4-5 nm (I), D = 5-9 nm (II), and D = 9-24 nm (III).
The in situ oxidation experiment was carried out by exposing the sample at 570 K to different oxygen pressures, where for each condition 2D reciprocal space maps were measured for the selected nanoparticle sizes. Figure 21 shows an overview over the 2D diffraction patterns zoomed in onto the Pd particle (111) Bragg peak vicinities for the different sizes (diameters D = 4.8, 5.6, 11, and 24 nm). They were measured under UHV conditions (after the sample had initially been annealed to 920 K), under 0.3, as well as under 56 mbar O 2 pressure. Apart from the Pd (111) Bragg peak, the diffraction images moreover contain the (111) Bragg peak of the MgO(001) substrate, as well as PdO(101) bulk oxide Bragg peaks and PdO(101) bulk oxide Debye Scherrer rings, which emerged upon oxidation. At 0.3 mbar O 2 , the particle Pd(111) Bragg peak decreased in intensity and became broader, which indicates that the particles decreased in overall size upon oxidation. In addition, diffraction signals from the PdO(101) bulk oxide appeared. In the case of the smaller particles (D = 4.8, 5.6 nm) the signals emerged as a defined PdO(101) Bragg peak, indicating the formation of epitaxial PdO bulk oxide. In the case of the larger particles (D = 9-24 nm), the signals appeared as PdO (101) bulk oxide Debye Scherrer rings, suggesting the formation of polycristalline PdO.
At 56 mbar O 2 , the pd particle (111) Bragg peak vanished completely for the smallest particles (D = 4.8 nm), while the intensity of the PdO(101) Bragg peak increased significantly, suggesting the particles' complete oxidation to epitaxial PdO (101). For the particles with a diameter of D = 5.6 nm the Pd(111) Bragg peak did not vanish completely, but also in this case its intensity decreased dramatically for the benefit of the intensity of the Bragg peak intensity of the epitaxial PdO(101) bulk oxide. Accordingly, the particles featuring this size did not oxidize completely, but yet oxidized to a great extent into epitaxial PdO(101) bulk oxide. For the larger particles (D = 9-24 nm), the intensity of the particle Pd(111) Bragg peak did not change significantly, and no formation of a defined PdO(101) Bragg peak belonging to epitaxial bulk oxide was evidenced. Instead, the PdO Debye Scherrer rings were still present also under 56 mbar O 2 , suggesting the formation of a passivating polycristalline PdO layer on top of the Pd particles.
To obtain the heights H and diameters D of the metallic Pd particle cores, linescans were extracted from the 2D images through the particle Pd(111) Bragg peaks along the radial and along the out-of-plane direction. From these the full widths at half maximum were obtained which were used to determine the particle size. The thus deduced values for H and D are plotted in figure 22 as a function of the sample position with respect to the incident beam position, i.e. as a function of the probed particle size. The data reveals that at 0.3 mbar O 2 the metallic particles underwent an overall shrinkage, especially pronounced for the smallest and the largest particles (size regimes I and III). At 56 mbar, the metallic core of the smallest particles (regime I) disappeared completely, since all metallic Pd was oxidized into epitaxial PdO(101) bulk oxide. The medium-sized particles (regime II) underwent further shrinkage, while the large particles of regime III maintained their size, probably due to a passivating layer of polycrystalline PdO, as evidenced by the formation of the PdO Debye Scherrer rings in figure 21. Figure 23 summarizes the scenario concluded from the HESXRD data. Thus, in the first oxidation step at 0.3 mbar, the particles of all sizes underwent shrinkage, especially pronounced for size regimes I and III. While for the size regimes I and II the particles oxidized into bulk oxide PdO growing epitaxially on the MgO(001) surface [114], the particles of size regime III featured a passivating layer of polycristalline PdO. This layer acted as a kinetic barrier against further oxidation and prevented the further shrinkage of the metallic particle core in the second oxidation step at 56 mbar. Contrary, the metallic particle core of the smaller particles of regime II underwent at 56 mbar further shrinkage resulting in the formation of more epitaxial PdO (101). At this second oxidation This investigation illustrated for the first time the power of the experimental strategy using HESXRD to study combinatorial nanoparticle sample systems. Hence, the oxidation behaviour of the various nanoparticle sizes could be investigated under the exact same conditions. The large 2D diffraction pattern yielded immediate information on the type of oxide that formed and on the size of the remaining metallic particle core.

In situ oxidation of Pd-Rh alloy nanoparticles.
In the previous subsection the oxidation of monometallic Pd nanoparticles was discussed. The current subsection will yield insight into the composition-dependent oxidation of Pd-Rh alloy nanoparticles. Alloy nanoparticle catalysts are widely employed in catalytic industry and in vehicle exhaust gas treatment as they may show superior catalytic properties compared to their parent metals. Synergetic effects between the individual elements may result in an enhanced catalytic activity, an improved catalyst stability or a selectivity towards a desired reaction pathway [115][116][117][118][119]. Reasons for the improved catalytic properties can be traced back to novel electronic states at the alloy surface or to segregation or strain effects.
The data discussed in the following concerns the composition-dependent oxidation behaviour of Pd-Rh nanoparticles and was obtained by means of HESXRD employed on a combinatorial sample containing Pd-Rh alloy nanoparticles supported on MgAl 2 O 2 (001) [72] (see also sample 2 in figure 20). The data showed that HESXRD allows for tracking the formation of different bulk oxide structures on individual nanoparticle facets, and for deducing direct information on the oxidation-induced segregation and change of alloy concentration in the metallic particle core. The sample's particle stripes featured a gradient in alloy composition from pure Rh to pure Pd (alloy compositions: Rh, Rh 0.6 Pd 0.4 , Rh 0.34 Pd 0.66 , Rh 0.15 Pd 0.85 , Pd) with particle diameters ranging from 6 to 11 nm. The combinatorial approach allowed for probing one particle stripe at a time under the same condition by translating the sample perpendicular to the incident beam direction. During the in situ oxidation, the oxygen pressure was stepwise increased, where for each condition a 2D reciprocal space map was measured on the individual alloy compositions. Figure 24 shows a thus obtained overview over the 2D diffraction patterns centred around the particle (111) Bragg peak, where the probed oxygen pressures at 670 K comprised 10 −3 mbar O 2 and 0.1 mbar O 2 (see section 2.2.3 and figure 9(a) for more information on the diffraction signals contained in a 2D map measured on (001)-oriented particles in the (H = K, L)-plane).
In the first oxidation step at 10 −3 mbar O 2 pressure, the 2D maps of the Rh-rich particles (Rh, Rh 0.6 Pd 0.4 , Rh 0.34 Pd 0.66 ) were characterized by the appearance of additional Bragg peaks. They were concluded to stem from RhO 2 and spinellike Rh 3 O 4 bulk oxide phases (see explanation on the Rh bulk oxide structures below), implying an outward segregation of the Rh atoms. For the pure Pd and the Rh 0.15 Pd 0.85 particles no additional Bragg peaks appeared. Instead, the particle facet signal intensities became greatly reduced, especially for the pure Pd particles. This finding is in line with the notion of a roundening of the particle shape, as found in previous investigations [120].
It was only in the second oxidation step at 0.1 mbar O 2 that bulk oxide Bragg peaks were evidenced for the Pd-rich Rh 0.15 Pd 0.85 and the pure Pd particles. As in the case of the other Rh-containing particles, the diffraction maps of the Pdrich Rh 0.15 Pd 0.85 particles contained only Bragg peaks from the RhO 2 and the spinel-like Rh 3 O 4 bulk oxide phases. Contrary, the diffraction pattern measured on the pure Pd particles revealed the formation of epitaxial PdO bulk oxide (see discussion on the Pd bulk oxide structure below). For the Rh-rich particles (Rh, Rh 0.6 Pd 0.4 , Rh 0.34 Pd 0.66 ), the intensity of the Rh bulk oxide Bragg peaks increased even more compared to the first oxidation step. From linescan fittings along high symmetry directions through the particle Bragg peaks it was concluded that the intensity increase of the bulk oxide Bragg peaks came along with a shrinkage of the overall size of the internal metallic core, where both were found to scale with the Rh content of the particles. This indicates that a continued outward segregation of Rh atoms for the formation of even thicker Rh bulk oxides took place, which was even more enhanced for Rh-rich particles. However, in this respect also a potential influence of the nanoparticle size, which in this case scaled with the particles' Pd content, cannot be neglected.
The possibility of the HESXRD technique to probe large areas in reciprocal space allowed for identifying the growth and the orientation of the bulk oxide phases, including a phase not reported earlier, which would have been easily missed when using conventional photon energies. Thus, figure 25(a) shows the diffraction pattern measured on the Rh 0.6 Pd 0.4 particles at 10 −1 mbar O 2 , which is representative for all the investigated particles containing Rh. The positions of the Bragg reflections that appeared during the in situ oxidation can be explained by a (111)-oriented spinel-like Rh 3 O 4 bulk oxide growing perpendicular to the particle (100) and (111)    PdO bulk oxide with the Pd atoms sitting at the positions of a body centred tetragonal (bct) lattice. The diffraction pattern can be explained by two different (100)-oriented, inplane rotated PdO domains as well as, to a minor degree, PdO(001) domains. The corresponding real space unit cells are shown in figure 26(d), the resulting reciprocal space lattices in figures 26(b) and (c). Domains of a PdO(101) bulk oxide could also explain some of the observed features, but were ruled out since their presence would have resulted in additional Bragg peaks which were not observed (see figure 26(c)).
The absence of Debye Scherrer rings in all diffraction patterns, even after the second oxidation step, indicates that the bulk oxides grew either in epitaxy with the nanoparticle facets or the substrate. For the PdO bulk oxide formation it was concluded that, the PdO bulk oxide growth took mainly place in contact with the substrate, also since signals from bulk oxides forming on the particle (111)-type facets were missing. This is in line with results obtained for the previously discussed oxidation of Pd nanoparticles on MgO(001) [71]. Moreover, although the heats of formation of PdO, Rh 2 O 3 and RhO 2 are very similar [121][122][123], the oxidation of Pd took place only at the higher oxygen pressure of 10 −1 mbar, indicating an enhanced kinetic barrier for the formation of PdO bulk oxides.
In addition to the information obtained on the formed bulk oxide structures, their epitaxial relationships and the size of the metallic particle core, the HESXRD investigation also yielded immediate information on the particle alloy composition. This was obtained by tracking the positions of the maximum particle Bragg peak intensities during the oxidation experiment, from which the corresponding particle lattice parameters were deduced. The obtained data suggested a gradual depletion of Rh from the metallic core of the Rh-containing nanoparticles during oxidation, resulting in the formation of Rh oxide shells. These prevented the Pd inside the particle core from oxidation, even at the higher oxygen pressures of the second oxidation step.
This experiment revealed that the combination of HESXRD and a 2D detector for the investigation of epitaxial nanoparticle systems allows for the tracking of the structure of particular nanoparticle facets (bulk oxide formation), where thanks to the combinatorial approach different particle characteristics (here: Pd-Rh alloy composition) can be probed under the same condition. The vast area that was probed in reciprocal space contained diffraction signals that allowed for the identification of the different bulk oxide phases, including a spinel-like Rh 3 O 4 bulk oxide not reported in literature so far.

CO oxidation over Pt-Rh alloy nanoparticles.
The harsh reaction conditions in industrial catalysis, characterized by elevated temperatures and pressures, result in general in a gradual degradation of the catalyst efficiency. Such catalyst deactivation constitutes a giant problem in catalytic industry, especially against the background that the costs for catalyst renewal and process shutdown amount to tens of billions every year. One of the main reasons for catalyst deactivation is considered to be catalyst nanoparticle sintering, i.e. the agglomeration of smaller particles into fewer larger ones, resulting in a loss of effective catalyst surface area [110,111,124,125]. One promising route for the stabilization of the nanoparticles constitutes the alloying of the catalyst particle metals. However, the atomic scale processes during sintering and the beneficial effect of alloying are not yet understood [124][125][126][127][128].
In a recent publication we showed that HESXRD allowed for an in situ tracking of the alloy composition-and shapedependent vertical sintering of Pt-Rh alloy nanoparticles supported on Al 2 O 3 (0001) under realistic reaction conditions during CO oxidation [73]. Thererin we demonstrated that the information encoded in the diffraction intensities of the particle Bragg peak vicinities even allows for obtaining the quantitative particle shape and the composition-dependent restructuring during sintering processes.
The investigated combinatorial sample consisted of Al 2 O 3 (0001)-supported stripes of (111)-oriented nanoparticles of varying Pt-Rh alloy composition (see sample 3 in figure 20, compositions: Pt, Pt 0.85 Rh 0.15 , Pt 0.7 Rh 0.3 , Pt 0.5 Rh 0.5 , Rh). Initially, the particles featured, independent of their alloy composition, the same particle height of about 2 nm, but displayed a composition-dependent particle diameter, ranging from 12 nm for the pure Pt particles down to 5 nm for the pure Rh particles. Accordingly, the Pt-rich particles were characterized by a rather disc-like shape, the Rh-rich particles by a more compact shape. Thus, by using the approach of combining a combinatorial sample with HESXRD and a large 2D detector, not only the composition-, but also the particle shape-dependent sintering could be investigated. (f) Particle heights H and aspect ratios H D of the different particle alloy compositions as a function of the probed conditions (i)-(vi). The particle heights H were obtained by fitting the linescans extracted in L-direction through the particle Bragg peaks. The particle diameters D were deduced from a novel rocking scan analysis procedure described in the supporting material to [73] in combination with AFM measurements performed after the experiment. Particle shapes and sample surface coverages for selected Pt-Rh compositions before (g) and after (h) the reaction-induced sintering as obtained from fits to the Bragg peaks and x-ray reflectivity measurements. The figure was taken from [73]. 10 ml n min CO flow; condition (iv): 5 ml n min O 2 flow, 10 ml n min CO flow; condition (vi): 7 ml n min O 2 flow, 10 ml n min CO flow; the carrier gas Ar was used to add up to a total flow of 100 ml n min ). For each O 2 :CO ratio, a large 2D diffraction map was measured, which was centred around the particle (311) Bragg peak. Figure 27(a) shows zoom-ins onto these Bragg peaks plotted as a function of the alloy composition (vertical panels) and of the probed condition (horizontal panels). The MS data displayed in figure 27 show that the sample became progressively more catalytically active when switching to higher O 2 :CO ratios, as the CO 2 production increased stepwise. Figure 27(a) shows that the particle Bragg peak intensities are, especially for the Pt-rich particles, smeared out along the L-direction, and that distinct intensity fringes (called Laue oscillations) can be seen. Since the distance between these fringes is inversely proportional to the particle height, we used their respective positions on the detector as in situ probe to track the vertical particle sintering. While at the beginning of the experiment (condition (i): first vertical panel) the particles of all alloy compositions had the same height of 2 nm and accordingly the distances between the positions of the Laue oscillations were the same, the fringes of the Pt-rich particles moved progressively closer towards the particle Bragg peak in the transition to higher O 2 :CO ratios and thus to conditions of higher catalytic activity. This indicates that the pure Pt and Ptrich particles underwent a tremendous vertical sintering. Contrary, the vertical sintering was concluded to be progressively suppressed with increasing Rh content, as the movement of the fringes was increasingly inhibited for higher Rh compositions.
The data analysis moreover pioneered in demonstrating that even quantitative atomic scale information on the nanoparticle shapes is encoded in the intensity modulations and that accordingly this experimental approach can be used to track the 3D restructuring of particles during sintering processes. To this end, linescans were extracted from the 2D maps in vertical L-direction through the particle Bragg peak maxima. The thus obtained intensity-corrected linescans, represented by the open symbols in figures 28(a)-(e), were fitted using different underlying particle shape models based on the Wulff construction [128]. The particle model shapes put to the test consisted of (111) and (001) type facets and varied in the relative ratio of the (001) and (111) type facet surface area, where the number of atomic layers per particle corresponded to the particle height deduced from the fringe distance. During the fitting procedure, the particle shape models were refined to make the simulated linescans match the damping of the experimentally obtained Laue oscillations. This was obtained by varying the interlayer distances and the occupancy values of selected atomic layers at the top and at the interface of the particle models, thereby accounting for a potential roundening of the particle shape, a certain particle height distribution as well as for misfit dislocations at the substrate-particle interface. The thus obtained fits are represented by solid lines in figures 28(a)-(e). Figure 28(f ) illustrates the thus deduced particle heights H as well as the particle height-to-diameter aspect ratios H/D for the different Pt-Rh compositions as a function of the probed conditions. The data reveal a tremendous vertical sintering of the pure Pt particles with a doubling in height, which becomes progressively suppressed for higher Rh contents. Surprisingly, the particle diameters were found to stay relatively constant. Accordingly, the pure Pt and Pt-rich particles underwent during the sintering process a restructuring from flat and disclike particle shapes to more 3D and compact-shaped particles. Contrary, the Rh-rich particles maintained their original shape, which already initially was, due to the small diameter, very compact, and which proved to be resistant against sintering. Figures 28(g) and (h) illustrate the particle shapes as determined from the linescan fittings before and after the sintering process, respectively. They moreover show that a tremendous mass transport occurred on the sample surface during the sintering of the Pt-rich particles, since the increase in particle height came at the expense of the number of particles per surface area, as was deduced from complementary x-ray reflectivity measurements.
The HESXRD investigation revealed a novel sintering mechanism, a non-classical Ostwald ripening, that was not reported in literature before. The ripening is characterized by the destabilization of smaller particles by the heat released during the CO oxidation reaction. Their atoms are then used for the restructuring of particles that were initially kinetically trapped in flat, disc-like shapes, into higher, more compact shaped particles of thermodynamic stability.
This HESXRD study described above revealed how, under industrially relevant reaction conditions, the compositiondependent atomic-scale restructuring of particles can be deduced. This experimental strategy could be of interest for other operando and in situ investigations in catalysis and materials research.

Dehydrogenation of liquid organic hydrogen carriers
over Pd nanoparticles. The operando experiments reported in this topical review so far discussed the oxidation of CO into CO 2 , which is a chemical reaction considered as the fruitfly of catalysis research due to its seemingly simplicity. In a recent publication we turned to a more complex reaction, namely the dehydrogenation of methylcyclohexane (MCH) over Al 2 O 3 (0001)-supported Pd nanoparticles, where MCH represents a model for liquid organic hydrogen carriers (LOHC). LOHCs are considered an important hydrogen storage material essential for future energy supply [129]. To dehydrogenate LOHCs, a catalyst is required, where Pd is a suitable material due to its hydrogenation and dehydrogenation capabilities [130]. During the reaction processes, both hydrogen and carbon may diffuse into the Pd lattice, which may have a strong effect on the activity and selectivity of the catalyst. Especially carbon may result in a deactivation of the catalyst.
To improve the understanding of these processes we used in this experiment the sensitivity of the HESXRD technique towards the monitoring of changes in the Pd particle size and the lattice parameters caused by the potential incorporation of hydrogen and/or carbon. Figure 29 shows a sketch of the sample and the experimental setup including a HESXRD 2D map measured on the as-prepared Pd nanoparticles with a size of 3.4 nm supported on Al 2 O 3 (0001). The presence of Debye Scherrer rings in the diffraction pattern revealed that the particles were randomly oriented on top of the oxide substrate, yet, the majority of the Pd particles was concluded to still prefer to grow as (111)-oriented particles with a hexagon-on-hexagon growth at the metal-oxide interface, which was deduced from the non-uniform intensity distribution along the rings.
In the first part of the experiment the sample was exposed to different H 2 :Ar partial pressure ratios as a function of the sample temperature. The positions of the Debye Scherrer rings shifted at T = 300 K to a great extent to lower 2ϑ values, which implies that the particle lattice expanded due to the incorporation of hydrogen into the Pd nanoparticles. Accordingly, at 300 K, the Pd lattice parameter changed form 3.88 ± 0.02 Å to 3.97 ± 0.02 Å which corresponds to a maximum hydrogenation fraction of 0.37 ± 0.03. With increasing temperature, the change in position of the Debye Scherrer rings was found to be progressively suppressed. Accordingly, the hydrogenation of the particles was concluded to be more and more reduced as a function of temperature. At 500 K no measurable change in the Debye Scherrer ring positions could be monitored, which leads to the interpretation that only a negligible amount of hydrogen could be incorporated into the Pd particles under these conditions. The second part of the experiment was carried out at a constant temperature of 500 K and consisted of the exposure of the sample to a gas mixture of MCH and H 2 , followed by the exposure to a mixture of Ar and MCH. Since the sample temperature constantly comprised 500 K, potential shifts in the lattice parameter could in this part of the experiment be directly ascribed to a potential incorporation of carbon.   Figure 30(b) shows the corresponding data obtained for the mixtures of 500 mbar H 2 and 500 mbar MCH being changed to 500 mbar Ar and 500 mbar MCH. The stability of the Bragg peak positions with respect to the reference data measured under pure Ar revealed that the presence of H 2 suppressed in both cases the incorporation of carbon from MCH. However, as soon as the H 2 was replaced by Ar, the Pd lattice parameter changed due to the diffusion of carbon into the lattice. For the lower MCH concentration, we deduced a total lattice expansion of 0.03 Å or 0.7%, corresponding to a carbon fraction of 4.8 ± 1% inside the particles, as can be deduced from figure 30(c). For the higher MCH concentration, the total lattice expansion was concluded to have amounted to 0.04 Å (1.0%), indicating a larger concentration of lattice carbon, with an estimated concentration of 6.5 ± 1% (see figure 30(d)). Accordingly, higher MCH partial pressures resulted in an increased carbon incorporation into the particles, where the presence of H 2 in the gasflow at 500 K inhibited this incorporation. While the formation of Pd hydride was concluded to be easily reversible when switching back to pure Ar flow, the formation of lattice carbon was not. In addition, a gradual sintering of the Pd particles from a size of 3.4 nm to around 5.0 nm was deduced from the HESXRD data to have taken place in the course of the experiment.
This study showed that HESXRD allows for the monitoring of diffraction signal positions with a high spatial resolution, which was in this case used for the investigation of particle lattice parameter changes, and accordingly allowed for determining the incorporation of hydrogen and/or carbon into a nanoparticle system.

Combination of HESXRD with other techniques in catalysis research
The experiments discussed above illustrate how the use of high energy x-rays results in a compression of reciprocal space which, along with a large 2D detector, allows for probing a vast area of reciprocal space even if the sample is not being rotated. Having such a stationary sample environment provides several advantages, one of them being the possibility to add equipment of additional measurement techniques to the experimental set-up, thus making the combination of several operando techniques feasible.
In this section an overview over the measurement techniques that we have combined with HESXRD in operando catalysis research will be given. Apart from in situ MS, that was already used in the investigations described above, the combined techniques include moreover planar laser-induced fluorescence (PLIF) [131] and surface optical reflectance (SOR) [132]. While PLIF probes a target molecule in the gas phase just above the sample surface, SOR yields information on the optical constants of the probed sample surface and/or the sample surface morphology. How the combination of the different techniques can be realized in the experimental setup is sketched in figure 31.

In situ MS
In situ MS is the most common technique to be combined with HESXRD in catalysis research, also since it is already included in the in situ catalysis chamber set-up [63] that was used in most of the investigations discussed earlier. In MS, the constituents of the gas composition to be probed are ionized (and sometimes fragmentized), and sorted according to their mass-to-charge ratio, which is specific for a certain atom or molecule. In the in situ catalysis chamber discussed in section 2.3.1 the mass spectrometer is located in the UHV part of the chamber, and the gas to be probed is being leaked into it from the reactor via a small gap. In closed reactor systems, the mass spectrometer can be placed downstream after the reactor. It can then be mounted onto a small UHV chamber which hosts the mass spectrometer into which parts of the gas stream can be led using a leak valve.
One significant advantage of MS is that several masses (typically on the order of ten) can be monitored at the same time which ensures that no reactant, reaction product or a fragment thereof is being missed. This is especially important for more complex catalytic reactions like for instance ammonia oxidation, which features three competing reaction pathways and accordingly a higher number of reaction products [134]. Furthermore, MS provides a fairly high temporal resolution, probing all selected individual masses typically once per second. A disadvantage of MS can be seen in the fact that-depending on where the mass spectrometer is located-there is usually a time delay between the changes in the gas composition inside the reactor and the visibility of these changes in the MS data. This can make the correlation between structural changes on the sample surface as evidenced with x-rays to changes in catalytic activity difficult. Moreover, the measured gas composition corresponds to its average in the reactor, which makes it impossible to correlate distinct or local areas of the sample to a potentially higher catalytic activity. MS setups that are equipped with a capillary sniffer, which can locally probe the gas composition, exist [135,136]. However, such capillary probes are intrusive since they might affect the gasflow above the catalyst by disturbing the gasflow dynamics. The approach using a capillary probe has not yet been combined with the HESXRD technique.

Planar laser induced fluorescence (PLIF)
The main drawbacks of MS, which are (1) the delay in response time created by the distance between the sample and the mass spectrometer position as well as (2) its probing of masses being averaged over the whole reactor volume, can be overcome by planar laser induced fluorescence (PLIF) [131]. PLIF is a non-intrusive technique that allows for probing, under industrial reaction conditions, a certain constituent of the gas phase in close vicinity to the catalyst sample. In PLIF a laser beam excites a vibrational/rotational level of the molecule of interest. The excited molecule emits upon relaxation fluorescence light of a known wavelength that is detected by an infrared camera. The process is sketched in figure 32(a) for the example of the monitoring of the reaction product CO 2 during CO oxidation.
In PLIF the probing laser beam is moreover shaped into an ultrathin laser sheet with a width down to 50 μm. This laser sheet allows for probing the gas phase just above the sample surface with a very high temporal and spatial resolution (typically 100 μm and 20 μs, respectively), as is depicted in figure 32(b). Accordingly, it is possible to deduce if certain areas on the sample, e.g. a certain particle stripe on a combinatorial sample, is catalytically more active than another (see figure 32(c)).
PLIF has traditionally been used for combustion diagnostics and its use in catalysis research is only very recent [137][138][139][140]. The different gases that have been probed by PLIF in catalysis research so far include CO 2 , CO, NH 3 [131], as well as CH 3 OH by using infrared degenerate four-wave mixing [141].
In a recent experiment we combined for the first time PLIF with HESXRD for the investigation of the oxidation of CO into CO 2 over a Pd(001) model catalyst [142]. This was realized by using a reactor dome made of sapphire instead of beryllium, which allowed both, the PLIF laser and the x-rays, for accessing the sample surface. As the surface structure may be highly sensitive to the composition of the gas phase just above the sample surface, the aim of this experiment was to obtain a better understanding of the direct correlation between the gas phase and the surface structure.
During the experiment the sample position was kept stationary at a position at which the 2D detector could probe the CTRs of the Pd(001) metal surface as well as potential superstructure rods from the ( √ 5 × √ 5)R27 • surface oxide. The HES-XRD 2D diffraction maps were acquired every 0.5 s, while the PLIF data was measured every 0.1 s using a 30 μs lasershot. To match the HESXRD repetition rate, the PLIF images were averaged resulting in an image every 0.5 s. The ultrathin laser sheet probed the CO 2 molecules and hence the reaction product within a 6 mm wide laser sheet that was located about 0.3 mm above the sample surface.
The CO oxidation experiment was carried out by changing the partial pressures of 6 mbar CO and 144 mbar Ar to 24 mbar O 2 , 6 mbar CO and 120 mbar Ar, thus keeping the total pressure of 150 mbar at a sample temperature of 200 • C constant. Figures 33(a)-(d) show the HESXRD (top) and PLIF (bottom) images probed during the transition to catalytic activity. While at the beginning, under reducing conditions ( figure 33(a)), the PLIF data show no CO 2 production and the HESXRD pattern corresponds to the one of the metallic Pd(001) surface, the ignition was first visible in the PLIF data while the HESXRD pattern still indicated a metallic surface ( figure 33(b)) However, the formation of small oxide domains that were not yet visible in the diffraction at this point in time cannot be excluded. It was about 2.5 s after the ignition that first diffraction signals from the surface oxide rods were discernible ( figure 33(c)). At the same time the CO 2 production as measured by PLIF increased further into the mass transfer limited regime ( figure 33(c)), characterized by a CO 2 boundary layer that prevents CO molecules to reach the surface [143], hence resulting in a CO-diffusion limited reaction. Thereafter ( figure 33(d)), the CO 2 boundary layer was continuously observed while the surface oxide rods grew more intense, indicating a continuing growth of the surface oxide layer. In PLIF an ultra-thin laser sheet probes the gas phase above the sample surface in vertical (left) and horizontal (right) slices, in the data shown it is the CO 2 gas phase during CO oxidation in the mass transfer limited regime. (c) Sketch illustrating how the spatial resolution of PLIF may allow for identifying the catalytically most active particle size and/or alloy composition on a combinatorial sample. Figure partly adapted from [131]. (c) Superstructure rods of the ( √ 5 × √ 5)R27 • surface oxide appear in the HESXRD pattern after the sample has been active for 2.5 s, a pronounced CO 2 boundary layer is evidenced in the PLIF. (d) After additional reaction time the surface oxide superstructure rods become more intense, the PLIF image shows that the CO 2 boundary layer is still present above the surface. The data was taken from [142].  To obtain a deeper insight into the temporal correlation between the appearance of the CO 2 production and the surface oxide formation, the integrated intensities of the surface oxide superstructure rods were plotted along with the PLIF CO 2 partial pressure measured just above the sample surface in figure 34(a). The data confirm that it was about 2.5 s after the ignition that signals from the ( √ 5 × √ 5)R27 • surface oxide were monitored. In addition, the gradual increase of the surface oxide rod intensities after ignition indicated that it took about 10 s for the surface oxide islands to grow into a more or less complete surface oxide layer.
The experimental setup also contained a mass spectrometer which was connected downstream to the outlet of the reactor via a tube of 4 m length and a diameter of 6 mm. This accordingly allowed in addition for a direct temporal comparison between the PLIF and the MS data, which is shown in figure 34(b). For the onset of CO 2 detection during the ignition a significant delay in response time of 9 s for MS compared to PLIF was concluded. Moreover, while a partial CO 2 pressure of 2 mbar was concluded from the MS data in the mass transfer limited regime, the PLIF data extracted in the vicinity of the sample surface measured a partial pressure of about 4-4.5 mbar CO 2 . This significant difference is due to the complex gas flow configuration of the reactor and its volume (11.5 ml), which leads to a smearing out of the globally detected mass spectrometer signal.
The above experiment demonstrates the importance of the combination of various techniques in operando catalysis to allow for a correct interpretation of the data. It also demonstrates that HESXRD renders such combinations possible as it allows for probing a vast reciprocal space area even with a stationary sample and detector. The combination of PLIF and HESXRD provides the unique possibility to directly relate the catalytic activity and the gas phase composition in the sample vicinity to the surface structure. This is of utmost importance, since the structure can be directly affected by the local gas composition during reaction (in the example discussed above by the CO-depleted boundary layer of CO 2 in the mass transfer limited regime).

Surface optical reflectance (SOR)
The samples investigated by means of HESXRD feature typically diameters of 6-10 mm. With the width of the horizontal x-ray beam on the order of 30 μm only a small part of the sample surface is accordingly probed in HESXRD, while the MS data is averaged over the whole reactor volume. Since it has moreover been found that even the catalyst surface of a single crystal model catalyst may not always behave homogeneously, e.g. due to inhomogeneous gas distributions around the sample [144], a technique that visualizes the surface morphology/optical properties of the whole sample surface at the same time is essential.
It was shown that the technique of SOR provides such an overview over the whole sample area [132,133,144,145]. In SOR, LED light is reflected from the catalyst sample surface and detected by a camera. The reflected intensity depends on the surface morphology or the oxidation state, where in a recent investigation combining SOR with HESXRD during the in situ oxidation of a Pd(001) surface we concluded that SOR can even detect the formation of a 2-3 Å thin Pd(001)- [146]. Since it is possible with SOR to monitor the entire sample surface at a time, the combination of SOR and HESXRD allows for correlating changes in the microscopic surface structure (HESXRD) to changes in the macroscopic surface structure (SOR).
In a very recent experiment we have examined for the first time the correlation between the microscopic and macroscopic surface structure by combining and synchronizing HESXRD, SOR, PLIF and MS during CO oxidation over Pd(001) using a  figure 35. The SOR data reveal how the sample oxidation starts at the centre right and spreads within minutes over the whole sample area. The HESXRD data reveal how during the oxidation process the signal intensity from the metal CTR becomes weaker (magenta rectangle), while the intensity of the bulk oxide Bragg peak increases (blue rectangle). The CO 2 partial pressure as measured by PLIF reveals a relatively constant CO 2 production since the reaction is in the mass transfer limited regime. Reproduced from S Pfaff et al Rev. Sci. Instrum. 90 033703 (2019) [133], with the permission of AIP Publishing. sapphire reactor dome [133]. Therein, the sample was probed in the transition from a CO-rich to an O 2 -rich gas composition, keeping the total pressure and the sample temperature constant at 180 mbar and 630 K, respectively. The probed O 2 :CO gas ratios are shown in figure 35(a). Figure 36 shows for three points in time during this gradual gas composition change the data measured by SOR (a), HESXRD (b) and PLIF (c), where in the case of the HESXRD data the summed images showing the highest intensity per pixel during a 15 • rotation are displayed. Panels (b)-(e) in figure 35 show the parameters of the different measurement techniques (MS, PLIF, SOR, HES-XRD) as a function of time during the gas ratio change, where the corresponding intensities were obtained by integrating over the respective regions of interest indicated in figure 36. Since the PLIF and the SOR measurements were not affected by the rotation of the chamber, the PLIF and SOR data shown in figure 35 have a temporal resolution of 0.1 s. Contrary, the HESXRD intensities shown in figure 35 were deduced from the HESXRD images summed over the 15 • chamber rotations and have a temporal resolution of 10.3 s, corresponding to the total time of the rotational scan.
During the gas ratio changes the sample surface was found to oxidize when approaching an O 2 :CO ratio of 6:1 at t 2 . In the HESXRD diffraction pattern this was evidenced by a decrease in intensity of the CTRs of the Pd(001) surface (magenta line/rectangle in figure 35(e) and figure 36(b), respectively) and by the appearance and intensity increase of the (004) Bragg peak of the PdO(101) bulk oxide (blue line/rectangle in figure 35(e) and figure 36(b), respectively). These changes in the HESXRD pattern coincided with a decrease in the reflected intensity in SOR (see figure 35(d)). However, the data showed that the intensity decrease in the SOR data appeared at different times for different positions on the sample surface. The snapshots of the SOR data showing the complete sample surface area in figure 36(a) indicate that the oxidation spot started to grow to the top right of the sample's centre, probably due to the flow geometry of the reactor. From there it spread over the whole sample surface. The changes in the HESXRD diffraction pattern corresponded well to the changes in the SOR data measured in the centre right where the oxidation started (red line/rectangle in figure 35(d) and figure 36(a)). This was explained by the fact that the x-ray beam was mainly probing this sample area. Moreover, the comparison between the MS and the PLIF data showed that the PLIF data which was measured further away from the surface and closer to the gas outlet, corresponded well to the MS data. Contrary, the PLIF data measured in the sample surface vicinity, varied greatly in its general trend and in the measured CO 2 partial pressure as compared to the MS data. Since the changes monitored by SOR and HESXRD appeared about 1.5 s after the ignition monitored by PLIF (see figure 7 in [133]), it was concluded that the sample featured either a metallic surface during the catalytic light-off, in line with recent findings [107], or that the amount of PdO was too small to be monitored.
The experiment clearly demonstrated that high care must be taken when correlating data taken over a small part of the sample, such as in HESXRD, to global condition changes, as monitored by MS. This is because the surface may not always act homogeneously, also for instance due to local variations, especially under operando conditions which are often characterized by gas composition and temperature gradients. Accordingly, it is important to monitor the whole sample area and vicinity to obtain a complete picture of the reaction conditions and to see if changes detected in HESXRD are representative for the whole sample surface. This is for instance provided by a combination of HESXRD with PLIF and SOR. The above discussed data moreover clarify that for a profound understanding of catalytic processes the combination of various techniques during operando investigations is inevitable. HESXRD provides, thanks to its ability of probing of a large reciprocal space area even with a fixed sample position, the possibility of combining a multitude of operando methods.

Summary and perspectives
In the present review we have shown that HESXRD allows for the operando investigation of the atomic surface structure of different model catalysts at work. The systems discussed include single crystal and vicinal surfaces, but also oxidesupported epitaxial nanoparticles. Being a photon-in photonout technique, HESXRD provides the possibility to investigate samples even under high gas pressure conditions mimicking 'real world' industrial processes. Accordingly, HESXRD is a surface sensitive operando technique that allows for bridging the materials and pressure gap, thereby linking the traditional surface science approach to industrial catalysis.
In HESXRD x-ray photon energies of 60-80 keV are used, which greatly reduces the scattering angles and the curvature of the Ewald sphere that probes reciprocal space. This provides the possibility of monitoring extended reciprocal space areas with a stationary 2D detector. Moreover, intrinsic sample imperfections, such as the mosaic spread of internal sample crystallites or the angular mosaicity of supported nanoparticles, enlarge the fulfillment of the diffraction condition to extended reciprocal space ranges. This results in the visibility of even larger reciprocal space areas with a single 2D snapshot.
Especially two modes of measurement operation have proved of value so far, and we have demonstrated how they are applied to different sample systems. The first mode of operation corresponds to the fast data acquisition of 3D reciprocal space at fixed temperature and gas flow conditions by rotating the sample around its surface normal while taking 2D images with a stationary detector. Thanks to the high energy approach full 3D data sets with high spatial resolution are thereby measured within a few minutes. This stands in stark contrast to the measurement time of several hours or even days in conventional SXRD. The latter results in well-chosen fragments of reciprocal space, which however, still may miss the appearance of unexpected diffraction signals and thus the formation of unforeseen sample structures. We have shown that the complete 3D data set as obtained by HESXRD can be visualized in different ways. Thus, in-plane reciprocal space maps reveal the diffraction data in a LEED-like manner, but with essentially better spatial resolution and under reaction conditions at near ambient pressures. Accordingly, in-plane maps allow for instance the observance of double and triple split superstructure rods that in this review's example resulted in a refinement of the atomic positions for the long-range Pd surface oxide unit cell. In addition, it was demonstrated that even structure factors for CTRs and superstructure rods can be extracted from a HESXRD 3D data set which allows for a quantitative atomicscale surface structural analysis.
The second mode of operation corresponds to the acquiring of 2D images with a fixed sample position that probes a reciprocal space plane of interest while gas ratios and thus the reaction conditions are being changed. We showed that this allows for a tracking of transient structural phases with subsecond time resolution such as the formation and transition of different oxide diffraction signals into others on single crystal surfaces, or the complex refaceting of a vicinal surface. We moreover showed that the measurement mode with a stationary sample allows for the investigation of combinatorial nanoparticle samples featuring a gradient in nanoparticle size and/or alloy composition. This offers the unique possibility of probing the influence of gradients in the sample characteristics under the exact same reaction conditions. As examples we showed results obtained on the size-and alloy composition-dependent oxidation of monometallic Pd and Pd-Rh alloy nanoparticles, which allowed for the identification and the tracking of various bulk oxide structures on the nanoparticle systems under various O 2 pressures. Moreover, the composition-dependent sintering of Pt-Rh alloy nanoparticles under realistic reaction conditions for CO oxidation could be followed, where it was shown that the quantitative particle shapes and shape changes could be deduced from the HESXRD 2D maps. In addition, HESXRD has shown to allow for quantifying the incorporation of carbon and hydrogen into Pd nanoparticles, which can be deduced from the highly resolved reciprocal space positions of particle diffraction signals.
Despite the advantages and experimental possibilities of HESXRD pointed out in this review article, also disadvantages that come along with the use of high photon energies have been experienced in the last couple of years. Hence, a disadvantage can be seen in the sensitivity of the HESXRD technique to the sample alignment due to the small incident angles of the beam (typically on the order of 0.03-0.05 • ) and the small vertical beam size (typically 3 μm). This makes a change of the sample temperature during operando measurements without losing the sample alignment impossible. It has been shown that also lower photon energies at around 24 keV [107], and hence in between the conventional and the high photon energy regime, can be used to perform fast measurements using a 2D detector, which could constitute a possible trade-off between a too sensitive sample alignment and a reduced detectability range in reciprocal space. A further disadvantage of the HESXRD technique can be seen in the fact that only well-ordered crystalline structures can be probed, where the structures in addition need to feature a certain size and quantity to be detected. Furthermore, since x-rays are used as probe, the detection of materials characterized by a low atomic number Z is also in HESXRD in general impeded. This makes, as in standard SXRD, the detection of for instance molecules on the sample surface very difficult.
On the other hand, the small scattering angles in HESXRD result in greatly reduced sample movements. This along with a stationary detector leaves in general enough space around the sample reactor for the mounting of additional equipment. Hence, HESXRD allows for its combination with other operando techniques. Even more so, since the high penetration depths of the high photon energy x-rays make the use of reactor materials other than beryllium and Kapton possible, which in turn makes the sample surface accessible by additional wavelengths such as IR light or visible light used in other methods. In this review it was shown how HESXRD was for the first time combined with PLIF by using a sapphire reactor dome. The gas phase probed by PLIF could, without any time delay, be directly assigned to the changes in the sample surface structure. The combination of HESXRD with SOR in another study revealed that while HESXRD probed the local surface structure of the sample, the SOR technique yielded a macroscopic overview over sample surface morphological changes. The presented examples emphasize the importance for the combination of different techniques for an overall understanding of the correlation between catalyst surface structure, gas phase and catalytic activity, and they demonstrate that HESXRD makes such combined approaches feasible.
The future combination of HESXRD with further techniques in addition to MS, PLIF and SOR features an interesting experimental strategy that would enable a more profound understanding of catalytic processes on the atomic scale. Potential techniques to be combined include spectroscopy methods such as polarization modulation infrared absorption spectroscopy (PM-IRRAS) or surface enhanced Raman spectroscopy. These would provide complementary information on the orientation and adsorption sites of the gas molecules on the model catalyst surfaces, hence helping to unravel atomic scale processes and catalytically active sites under reaction conditions.
As far as the probing of the gas phase in the catalyst surroundings is concerned, we have demonstrated that PLIF is a very powerful technique that provides a high temporal and spatial resolution. However, PLIF is very restrictive, as it is capable of monitoring only one target molecule. Accordingly, for more complex catalytic reactions with more than one reaction product PLIF provides no overall information on the sample's catalytic activity. Therefore, the combination of HESXRD with gas phase probing techniques such as absorption spectroscopy or x-ray fluorescence are very promising. They are easier to handle and yield an overview over the whole gas composition, where the probing position can be changed via the translation of the sample. In this respect, also the use of localized MS measurements using a capillary sniffer show promise for future experiments.
The ability of high photon energy x-rays to penetrate other materials than beryllium and Kapton not only allows for optical access to the sample surface and hence for the combination of different operando methods, but also for the use of more robust reactor cell materials. A more robust cell, for instance made of thicker sapphire walls, would allow for achieving higher cell pressures. This would make the HESXRD investigation of catalytic reactions that require such extreme conditions due to low conversions, like for instance methanol synthesis, possible. This would provide the possibility to enlarge the experimental strategies discussed in this review to industrially relevant reaction conditions of pressures up to 100 bar or higher.
The HESXRD experimental approaches developed so far will moreover benefit from the x-ray beam characteristics available at 4th generation synchrotrons. The higher photon flux is expected to speed up the measurement time even more, while the reduced beam size may result in a higher spatial resolution.
Finally, it should be emphasized that this topical review reports only on the use of HESXRD in operando catalysis research, i.e. during gas-surface interactions. Thanks to the properties of high photon energy x-rays the approaches and experimental strategies discussed here can also be applied to other sample systems and operando environments. These include the investigation of solid-liquid interfaces as used for instance in operando electrochemistry [36], but also the study of solid-solid and deeply buried interfaces, as present in model systems for grain boundaries [38]. In addition, HES-XRD is well-suited for in situ growth and deposition studies of nanoparticles and thin films [147]. funded project 'Atomistic design of new catalysts' project number KAW 2015.0058.