Surface science at the PEARL beamline of the Swiss Light Source

The Photo-Emission and Atomic Resolution Laboratory is a new soft X-ray beamline at the Swiss Light Source for the study of surface structure using photoelectron diffraction and scanning tunneling microscopy.


Scientific case
Surface science is an active, interdisciplinary field with applications in chemistry and physics such as heterogeneous catalysis, energy conversion, semiconductor and molecular electronics, spintronics and quantum magnetism. In these applications, chemical bonding, electronic charge transfer and magnetic interactions at interfaces play an important role, and many of the effects are intimately coupled to the atomic structure at the interface. Therefore, knowing the detailed structure is essential for the understanding of the underlying physics, and for the development and testing of theoretical calculations.
The Photo-Emission and Atomic Resolution Laboratory (PEARL) is a new soft X-ray beamline and surface science laboratory at the Swiss Light Source (SLS). It has been designed by a consortium of Swiss research groups active in surface science for the study of local atomic geometry at the surface of a wide range of novel organic and inorganic systems.
Owing to the stringent requirements of ultra-high vacuum (UHV) in surface science experiments many of the mentioned samples have to be prepared in situ without breaking UHV conditions between preparation and measurement. It is therefore crucial to have standard analytical tools available which allow for a quick assessment of the sample quality before long-running measurements are made. Furthermore, it is important that complementary analytical methods can be applied on the same sample if the correlation between atomic structure and electronic or magnetic properties is investigated. At synchrotron-based facilities these requirements are particularly challenging since the focus on instrument development is often laid on one specific technique. At PEARL, the aspect of integrating a reasonably complete surface science laboratory at a synchrotron facility was the major design goal.

Technical case
In surface science, charged particles, mostly electrons, are typically used as probes in a variety of experimental methods because they are particularly surface sensitive. At PEARL, the main analytical methods are soft X-ray photoelectron spectroscopy (XPS), photoelectron diffraction (XPD/PhD) as well as scanning tunneling microscopy (STM) and spectroscopy (STS). Since the instrumentation for XPD/PhD is mostly the same as for angle-resolved photoelectron spectroscopy (ARPES) in general, the facility also offers the spectroscopy of core-levels, valence bands (with limited resolution), Auger decays and resonant excitations. In the following, we will discuss mainly photoelectron diffraction and scanning tunneling microscopy.
Owing to their ease of use, scanning probe techniques have evolved to be popular methods for real-space imaging of surface structure and other surface properties at various length scales down to atomic resolution. However, vertical distance and angles between atoms are not easily accessible, and the contributions of the atomic and the electronic structure to a measured contrast are sometimes difficult to distinguish. Photoelectron diffraction is a technique to measure local atomic structure by exploiting the wave nature of photoelectrons and its diffractive properties at atomic potentials. Diffraction features appear as a variation of the photoelectron intensity as a function of emission angle and electron energy. By selecting a particular core-level photoemission peak, the technique is chemically selective, and due to a short mean free path it is sensitive to the topmost atomic layers. Photoelectron diffraction in angle-scanned mode (XPD) (Fadley, 1984;Osterwalder et al., 1995;Fadley, 2010) is suitable for measuring bonding angles and the orientation of small molecules with respect to the substrate (Fasel et al., 1996); while in energy scanned mode, where the acronym PhD is more commonly used, it is sensitive to the distance between neighboring atoms, for example, between adsorbed molecules and the substrate (Woodruff, 2007).
Energy-scanned PhD requires a photon source with smoothly tunable energy in the soft X-ray range. Tunable photon energy is also beneficial in many other cases. In photoelectron spectroscopy (XPS), the photoionization yield and the probing depth depend on photon energy, and the surface/bulk ratio can be adjusted. By tuning the photon energy to a particular X-ray absorption line, the photoemission cross section can be resonantly enhanced (Treier et al., 2009), or a particular symmetry of the photoelectron wavefunction can be selected (Morscher et al., 2011;Matsui et al., 2015). Using the core-hole clock method with Raman-active Auger decays, ultrafast delocalization dynamics in the conduction band can be studied (Fö hlisch et al., 2005;Jaouen et al., 2015). In angle-scanned XPD the ratio between forward and backward scattering can be tuned. Fig. 1 illustrates the basic measurement principle of XPD in the hypothetical example of two atoms. Let us assume a nitrogen atom is bonded to a nickel atom below, and the connecting line of the two atoms is tilted by 15 with respect to the reference axis (the sample normal in an actual system). This diatomic cluster is a reduced version of the h-BN/Ni system to be discussed in the Scientific highlights section below. At a fixed kinetic energy of 215 eV, the intensity of the N 1s XPS peak exhibits the angular distribution shown in panel (a), as calculated using the EDAC electron diffraction code (García de Abajo et al., 2001). The angular distribution of the intensity is displayed in a stereographic projection: the polar coordinates of the unit hemisphere ð; 'Þ are mapped to planar Cartesian coordinates according to ðx; yÞ = 2 tanð=2Þðcos '; sin 'Þ. The wavefunction of the photoelectron emanating from the emitter atom is scattered at the neighboring atom, and the direct and scattered waves interfere in the detector. The results are characteristic, circular interference fringes around the nearest-neighbor direction. Such features have been used in experiments to locate the adsorption site of atoms on metal surfaces, for example, in the case of O/Rh(111) (Wider et al., 1998).
If we cut the angular distribution of the photoelectron intensity along the 90 azimuth and expand the calculation in kinetic energy, we obtain the pattern shown in Fig. 1(c). The interference fringes appear again centered on the angle of the connecting line between the atoms. The frequency of the fringes is given by the distance between the two atoms, and the energy-dependent wavelength of the photoelectron. In the plot (note the logarithmic gray scale) it is obvious that the diffraction features typically correspond to a small variation of intensity on a strong but slowly varying background. Theoretically, this is due to the angular dependence of the photoemission cross section and the scattering factors. In the experiment, additional factors such as photon flux, sample orientation and detection efficiency may contribute, and are often difficult to separate from the diffraction signal. The relevant diffraction features are extracted by calculating the modulation function where I 0 is a smooth function from a non-parametric fit of the data, such as a cubic spline or locally weighted linear regression (Woodruff, 2007). The modulation function of the present data set near = 15 is shown in Fig. 1(d). Experiment and theory can be compared quantitatively by calculating the Pendry R-factor of the modulation functions (Woodruff, 2007), Conventionally, PhD scans were measured as one-dimensional line scans similar to the profile in Fig. 1(d). However, modern angle-dispersive analysers make it easy to measure multiple directions in parallel as in Fig. 1(c). This has two advantages over line scans. First, the precise emission angle can be determined from the same dataset as the distance between emitter and scatterer. Second, if the data contain diffraction features from multiple scattering configurations, correspondingly, more structural parameters can be determined at once.

Technical setup 2.1. Beamline optics
The specifications for the X-ray optics are based on the scientific and technical case described above. The details of the optical design have been discussed in a previous article (Oberta et al., 2011). Essentially, the beamline covers the photon energy range from 60 to 2000 eV. It is optimized for high photon flux in the range between 500 and 1000 eV where most photoelectron diffraction measurements of the lighter elements take place. Higher photon energies give access to resonant excitation of 4f levels in rare earths, and the low end allows for basic spectroscopy of the valence region. The key figures are summarized in Table 1.
The beamline is installed at a 1.4 T bending magnet which delivers a smooth photon spectrum with a critical energy of 5 keV. The main polarization mode of the bending magnet is linear horizontal. By tilting the trajectory of the stored electron beam, the polarization can be switched to elliptical. The optical layout is based on a plane-grating monochromator (Petersen et al., 1995) operating in non-collimated light and negative diffraction order, as shown schematically in Fig. 2. This scheme provides a good compromise between high photon flux and high energy resolution. In the optimum energy range, it allows to distinguish chemically shifted core levels or spin multiplets of the order of 0.1 to 0.5 eV. For systems where high resolution is not needed, it is possible to trade resolution for flux. Compared with other soft X-ray beamlines at the SLS operating in collimated light (Strocov et al., 2010;Piamonteze et al., 2012), the number of reflecting surfaces is reduced by one to save photon flux. As a drawback, the fixed-focus condition is set by design and cannot be modified during operation, giving the user less control over the harmonics in the spectrum. Using two selectable diffraction gratings (600 and 1200 lines mm À1 ), photon energy is smoothly tunable in two overlapping energy ranges (60-1100 and 200-2000 eV, respectively).    Although the signal-to-noise ratio in photoelectron diffraction can benefit strongly from high photon flux, many samples, particularly organic molecules, are susceptible to radiation damage if the flux is too high. To mitigate the problem, it is not sufficient to reduce the photon flux because that reduction would have to be compensated by increased exposure time. Rather, high electron yield with high conver-sion and detection efficiency is required. For the one part this is achieved with tunable photon energy, as the photoionization cross section can vary by orders of magnitude over the energy range of soft X-rays. For the other part, the photon flux can be spread over a larger area of the sample, thereby reducing the flux density, while the entire illuminated sample area is seen by the detector. At PEARL, the refocusing mirror unit can be switched to produce either a focused or a defocused beam on the sample.

Experimental station
The experimental station is divided into three sub-systems ( Fig. 3): one (attached to the beamline) for the photoemission measurements, one for scanning tunneling microscopy and one for surface preparation. All processes and measurements take place in UHV at a base pressure below 2 Â 10 À10 mbar. The sub-systems are connected by a reliable in situ sample transfer system. Though the system operates at room temperature, the transfer time between measurement positions is short enough to prevent cold samples (below 100 K initially) from heating up above 200 K. Samples and organic powders for evaporation are introduced from ambient or a UHV suitcase via a fast-entry lock. The sample preparation system provides standard surface science techniques for preparation (ion bombardment, annealing by radiative heating) and characterization [low energy electron diffraction (LEED), Auger electron spectroscopy, residual gas analysis]. A sixfold array of molecular beam evaporators with in situ exchangeable quartz crucibles for materials which sublimate below 900 K is available, as well as gated ports for user-supplied evaporators. A hightemperature annealing stage (1500 K) is under construction. For full specifications, see Table 2.
The low-temperature STM (Omicron Nanotechnology GmbH, Table 3) provides real-space sample characterization down to atomic resolution. Standard topography mode allows for quick assessment of the surface quality and reference to measurements at the user's home laboratory, dI=dV spectroscopy and mapping can be used to measure the local density of states near the Fermi level. Thanks to careful damping inside Conceptual rendering of the endstation. The three substations for angleresolved photoelectron spectroscopy and diffraction (XPS/XPD, green), scanning tunneling microscopy (LT-STM, blue) and surface preparation (red) are connected to a central, rotary sample transfer under UHV. The synchrotron radiation (SR) enters the XPS/XPD station along the path marked by an arrow. The drawing does not accurately represent the installation status of auxiliary devices.  and outside the chamber, the STM has proven insusceptible to vibrations and acoustic noise from the synchrotron environment.
The photoemission station is designed as a state-of-the-art ARPES facility with a 'Carving 2.0' six-axis manipulator designed by PSI and Amsterdam University, and a Scienta EW4000 hemispherical electron analyser with two-dimensional detection. The specifications are summarized in Table 4. The measurement geometry is illustrated in Fig. 4(a). The entrance lens stack of the analyser is at a fixed angle = 60 with respect to the incoming synchrotron light. The entrance slit of the analyser is oriented vertically (parallel to the main axis of rotation). In this orientation, the symmetry of the differential photoemission cross section with respect to the light polarization allows for a homogeneous illumination of the detector.
The primary rotation axis is the polar rotation about the z axis. The secondary rotation axes are the tilt about the y 0 axis, and the azimuthal rotation ' about the surface normal n of the sample. The new version 2.0 of the Carving manipulator features an improved bearing concept of the primary rotation to reduce the sphere of confusion: three-dimensional mechanical test measurements after assembly show that, under polar rotation , the sample moves by less than 25 mm in the scattering ðxyÞ plane and less than 65 mm along the z axis. For the secondary rotation axes and ', the displacement is less than 25 mm. Such high mechanical precision is essential for angle-scanned measurements due to the small beam size and the small focal depth of the analyser, either of which is of the order of 100 mm. To take advantage of the high precision, however, the sample must be mounted with the same precision on the sample plate so that the surface is aligned with the rotation center of the manipulator. Usually, this requires a precise optical survey of the shape of the specimen and the manufacturing of a sample holder that is tailored to the specific shape.
The sample can be cooled down to 35 K using liquid helium (LHe). The actual sample temperature was confirmed by adsorption and desorption of argon on a Cu(111) surface, compared with literature values of the desorption temperature (Berthold et al., 2004;Meyer et al., 2008). Due to the particular design of the Carving manipulator, thermal contraction of the primary axis due to cryogenic cooling is negligible.
The EW4000 electron analyser contains a two-dimensional multi-channel plate detector where one axis corresponds to the kinetic energy of the electron and the other axis to the emission angle . The nominal acceptance angle of this detector is 60 . In practice, transmission and matrix element effects limit the useful range to about 50 . Combining the manipulator and detector angles, photoelectron counts are collected as a function of the four angles , , ' and . However, in presentation graphs, angle-scanned photoelectron diffraction data are typically displayed in the spherical coordinate system ð s ; ' s Þ of the sample as in Fig. 1(a). Assuming that the angular dependence of the matrix element can be neglected (e.g. by normalization), instrument coordinates are mapped to sample coordinates by applying a series of rotations to the Cartesian vector k = ðcos ; 0; sin Þ which marks the detection angle in the laboratory frame of reference (Greif et al., 2014). In the sample frame, the emission vector k s becomes, thus where R À1 x , R À1 y and R À1 z denote the inverse rotation matrices about the coordinate axes x, y and z, respectively. k s can then be mapped to spherical coordinates in the canonical way taking the x axis as the surface normal. Fig. 4(b) shows the lines accepted by the analyser for a number of manipulator  Table 4 Specifications of the photoelectron spectroscopy system.    positions and how they map to the sample frame according to equation (3). On the unit hemisphere, each of the lines corresponds to an arc of a great circle. In the stereographic projection, it appears curved with a -dependent curvature. A full hemispherical diffractogram can be measured by combined ð; 'Þ scans with a step size of 1 or 2 for , and between 15 and 50 for ' in typically 6 to 24 h, depending on the signal and desired amount of oversampling.

Photon flux
The photon flux is measured by a calibrated silicon diode after the refocusing mirror, Fig. 5. The two laminar diffraction gratings cover an energy range from 60 eV up to 2000 eV with an overlapping region between 100 and 1000 eV. The 600 lines mm À1 grating is optimized for high photon flux, whereas the 1200 lines mm À1 grating is required for photon energies above 1000 eV, or for better energy resolution below 1000 eV. The dashed and solid lines mark the practical lower and upper limits which can be set by the front-end aperture, respectively, at a typical exit slit aperture of 100 mm. The upper limit is given by the physical size of the focusing mirror. The results are summarized in Table 5.

Energy resolution
The energy resolution in photoelectron spectroscopy is limited by the beamline optics and the electron analyser. In this section, we first demonstrate the ultimate resolution of the optics by measuring gas phase X-ray absorption spectra of nitrogen. Second, we discuss the energy resolution of the complete system derived from photoelectron spectra under typical measurement conditions. In most practical cases, it is necessary to find a compromise between energy resolution and count rate by opening the apertures of the beamline and the analyser.
Total ion yield gas phase spectra are measured in a gas cell installed after the exit slit of the monochromator. The apertures are set at the lowest practical values of 1 mm Â 1 mm for the front-end, corresponding to an acceptance angle of (120 mrad) 2 , and 50 mm for the exit slit. The measured N 2 1s-* spectra are plotted in Fig. 6. The ratio of total yield between the first valley at 400.8 eV and the third peak at 401.2 eV is a sensitive measure of the overall energy resolution where lower values indicate better energy resolution (Chen & Sette, 1989). The advantage of the valley-to-peak (v/p) ratio over curve fitting is that it is independent of the calibration of the energy scale and less susceptible to correlations between fit parameters. For a quantitative measure, the spectrum is modeled with the sum of seven Voigt profiles and v/p is compared with the measurement. The basic parameters for the model spectrum, the natural line width of 113 meV FWHM and peak positions, are taken from the literature (Kato et al., 2007). The resulting values for the resolving power are 5550 and 6860 for the 600 lines mm À1 and 1200 lines mm À1 gratings, respectively, very close to the corresponding values 5500 and 7000 from the design calculations (Oberta et al., 2011). Benchmark values are summarized in Table 5. Detailed results and an additional discussion of curve fits are given in the supporting information.
To check the energy resolution of the complete system we measure the width of the Fermi edge of a polycrystalline gold  Table 5 Summary of measured performance values of the beamline optics.
Maximum photon flux is measured at h = 800 eV, energy resolution at 400 eV. G 600 (G 1200) denotes the 600 lines mm À1 (1200 lines mm À1 ) grating. See text and supporting information for details Property Unit G 600 G 1200 Maximum photon flux photons s À1 1.7 Â 10 11 5.2 Â 10 10 Photon flux (small aperture) photons s À1 5.6 Â 10 9 3.7 Â 10 9 Resolving power (XAS) 5550 6860 Energy resolution (XAS) meV 72 58 Energy resolution (XPS) meV 103 98 Figure 6 (a) Total yield absorption spectra of nitrogen measured with monochromatic light from the 600 lines mm À1 grating (G 600) and the 1200 lines mm À1 grating (G 1200), at an exit slit setting of 50 mm. Solid lines are least-squares fits of Voigt functions as described in the text, dashed lines show the decomposed lowest-energy peak. The intensity ratio between the first valley and third peak (v/p) is indicated. (b) Calculated energy resolution of the beamline optics at the aperture settings of the nitrogen spectra. Experimental values from nitrogen XAS are marked.

Figure 5
Photon flux after the refocusing mirror measured on a calibrated silicon photodiode for the 600 and 1200 lines mm À1 diffraction gratings. The solid and broken lines were measured for maximum (720 mrad Â 1200 mrad) and minimum (120 mrad Â 120 mrad) front-end aperture, respectively. The exit slit size is 100 mm.
sample with XPS. The spectra in Fig. 7 are taken at essentially the same beamline settings as the nitrogen absorption spectra except that the front-end aperture is widened to (240 mrad) 2 to increase the count rate. The effect of the wider aperture on the energy resolution is less than 5% as confirmed in separate XAS measurements of nitrogen. The width of the Fermi edge contains two components, the intrinsic thermal broadening of the electron distribution in the material and the instrumental broadening by the analyser and the beamline. To first order, these effects add up quadratically as discussed in the supporting information . To reduce the first contribution as much as possible, we cool the sample to 40 K. The spectra can be fit with a Fermi-Dirac distribution at T = (341 AE 40) K and T = (325 AE 62) K, which amounts to a total instrumental broadening of 103 and 98 meV, respectively. Given the resolution of the X-ray optics discussed before, the analyser resolution is estimated to be (76 AE 17) meV at the selected entrance slit (0.2 mm) and pass energy (50 eV). Though the resolution of the analyser could be improved by lowering the pass energy, the low count rate due to the very low photoemission cross section of the valence band in the soft X-ray regime did not allow so as the acquisition of each displayed spectrum took about 12 h. On the other hand, the spectrum of an intense peak such as the 4f 7=2 core level resolving the surface core-level shift as in Fig. 7(b) can be acquired with the same high-resolution settings in less than 10 min. These measurements demonstrate that the beamline is capable of resolving chemical shifts of core-levels of the order of 100 meV. However, at very high resolution and for low cross-section transitions, the count rate is limited.

Spot size
The spot size on the sample is a result of the size of the electron beam in the bending magnet, the optical magnification, aberrations and manufacturing tolerances. These effects sum up to a theoretical minimum spot size of 170 mm Â 73 mm on the sample as predicted by ray-tracing calculations (Oberta et al., 2011). Experimentally, the beam profile is measured on a scintillator plate at the nominal focus position of the refocusing mirror, cf. Fig. 8. The results for the small spot geometry, panel (a), agree very well with the calculations. The minimum FWHM spot size observed is 190 mm Â 70 mm at a photon energy of 1000 eV and a vertical exit slit aperture of 100 mm. The small spot is almost independent of the front-end aperture and the photon energy. Its horizontal width increases slightly towards lower photon energy (230 mm at h = 400 eV).
The large spot setting of the refocusing mirror is designed to produce a convergent beam with an image distance of 5.7 m in the meridional (horizontal) plane, and a divergent beam with an image distance of À28.8 m in the sagittal plane. The observed spot size depends significantly on the front-end acceptance, varying from 180 mm Â 160 mm at the smallest aperture (not shown) to 1.1 mm Â 1.3 mm at the maximum aperture [panel (b)].
Since the electron optics of the EW4000 analyser is optimized for a small spot of 100 mm, it is interesting to check the effect of the spot size on the angle and energy resolution of the analyser. For this, we measure the Shockley surface state on Cu(111) at a photon energy of 70 eV [Figs. 8(c)

and 8(d)].
High-resolution measurements of this system are available in the literature (Reinert et al., 2001). It is obvious that in the large spot configuration the angle distribution is broader than     in the small spot configuration where the peak width approaches the nominal angle resolution of the analyser which is limited by the entrance mesh of the wide-angle lens. The influence of the spot size on the energy resolution is not obvious. In either case, the line width at the apex of the dispersion curve is 59 meV, limited by the instrumental energy resolution used in these measurements. We also find that the electron count rate is about 26% lower for the large spot while the total yield (71 pA) does not change. Because this loss of electron counts has to be compensated by longer integration time, the advantage of the large spot (longer protection against radiation damage due to lower flux density), is reduced from the original ratio of beam size to about a factor 80.

Scientific highlights 4.1. Measuring adsorbate-substrate distance in boron nitride
As a scientific example, we show angle-and energy-scanned photoelectron diffraction of hexagonal boron nitride (h-BN) on a Ni (111) surface. h-BN is a well known atomically thin insulating layer that can be used to chemically and electronically decouple molecular adsorbates from the underlying metal (Muntwiler et al., 2005). The atomic structure of h-BN/ Ni(111) has been studied by low-energy electron diffraction, angle-resolved photoelectron diffraction, scanning tunneling microscopy, and density functional theory in the past (Gamou et al., 1997;Auwä rter et al., 1999;Muntwiler et al., 2001;Grad et al., 2003). h-BN forms a commensurate 1 Â 1 overlayer with a nitrogen atom at the top site and a boron atom at the f.c.c. hollow site, cf. Figs. 9(a) and 9(b). The layer is slightly corrugated due to a lattice mismatch and different bonding of nitrogen and boron to the substrate. The distance between the h-BN layer and the substrate, d A-S , was studied by LEED previously and reported as 2.22 Å (Gamou et al., 1997). A later angle-resolved XPD study reported a different value of 1.95 Å (Muntwiler et al., 2001). Here, we look for an independent result using energy-scanned PhD in the backscattering configuration.
Angle-resolved XPD has the advantage that directions of atomic bonds can often be identified rather easily in a stereographic mapping of the photoelectron intensity without the need for a calculation. Such a map also helps to find the correct manipulator position with respect to specific diffraction features or bond directions for subsequent spectroscopy or PhD measurements. The diffraction pattern of the N 1s peak of h-BN in Fig. 9(c) is assembled from XPS spectra measured at 2148 angular settings according to the procedure described in the supporting information. The polar angle dependence of the data is removed by normalization (see below), and a threefold average is applied according to the symmetry of the substrate. The diffraction pattern shows notable rings at > 60 that are centered on the nitrogenboron (N-B) and the nitrogen-nitrogen (N-N) nearestneighbor directions. In contrast to earlier published data (Auwä rter et al., 1999), the pattern in Fig. 9(c) is sixfold symmetric due to the presence of two domains rotated by 180 with respect to each other. It is known that, in addition to the most stable N-top, B-f.c.c. adsorption configuration, an N-top, B-h.c.p. configuration with a slightly lower binding energy can grow depending on a subtle difference in the quality of the substrate Grad et al., 2003). The Ni crystal used in the present experiment was newly procured, and had not undergone the same number of cleaning steps as the one in the previous studies. Panel (d) shows the corresponding simulation using the EDAC multiple-scattering code (García de Abajo et al., 2001). The cluster in Figs. 9(a) and 9(b), showing only the most stable structure, is based on the optimized structural parameters discussed below. Qualitatively, measurement and calculation show the same diffraction features. However, a shift of features at higher polar angles indicates that the refraction at the surface may not be accurately described in the model.
To determine the adsorbate-substrate distance d A-S between the N and top-layer Ni atoms, the PhD intensity modulation of the N 1s peak is measured as a function of electron energy and simulated numerically using the EDAC code. In the simulations, seven structural and non-structural parameters are optimized using a particle swarm global search algorithm (Duncan et al., 2012) which minimizes the Pendry R-factor, equation (2). The optimized parameters are the adsorbate-substrate distance d A-S , the corrugation of h-BN, the possibly relaxed distance between the top two nickel  layers, the size of the cluster, the position of the refractive surface above the top layer, and the amplitude of the modulation function.
The measured modulation function normalized according to equation (1) is shown in Fig. 10, panels (a) and (c). Panel (a) shows the full two-dimensional dataset ðE kin ; Þ while panel (c) shows a line profile ðE kin Þ integrated over À2.5 < < 2.5 . The corresponding simulations of the optimized model structure are shown in panels (b) and (c). Panel (d) shows the R-factor results from over 10000 calculated configurations as a function of d A-S , the main parameter of interest. It shows a strong dependence on the adsorption parameter where the minimum R = 0.36 designates the best-fit value, and the width of the distribution can be used to estimate the uncertainty according to Booth et al. (1997). The result is d A-S = (2.11 AE 0.02) Å .
Since the raw data of Fig. 10(a) were measured with a twodimensional detector, we have carried out the optimization procedure for the two-dimensional and one-dimensional datasets separately. As can be seen in panels (a) and (b), the agreement between the calculation and the experiment is not reached in every detail, and the absolute values of the R-factor are correspondingly large. Nevertheless, the locations of the minima of d A-S are compatible in both cases. The advantage of a two-dimensional dataset is that multiple angles are measured at the same time. In the present case, this allows for a (coarse) optimization of other parameters such as the relaxed distance between the top two nickel layers (1.99 AE 0.04) Å , which is not possible from the normal emission measurement alone because the back-scattering directions are off-normal.
We conclude this section with a brief discussion of the normalization procedure applied to the angle-scanned data in Fig. 9(c). Processing of angle-scanned XPD data from a twodimensional electron analyser is more complex than from conventional channeltron-based detectors because the measured angle distribution is modified by additional physical and instrumental effects (Greif et al., 2014). Such effects include the angular dependence of the differential photoionization cross section, the cross section of the illuminated and the analysed volume, as well as angular inhomogeneities of the electron lens and the detector (transmission function). Fig. 11(a) shows the detector image of the N 1s peak at the normal emission setting of the manipulator. The distribution of photoelectrons has a pronounced dependence on the polar emission angle that is extremely sensitive to the distance of the sample from the entrance lens of the analyser. Only after measuring the data presented here we found that the transmission function could be flattened significantly by more careful alignment of the beam, the sample and the focal point   of the analyser using a reference sample with well defined emission angles. However, a normalization step is still necessary in any case because the transmission curve is never perfectly flat and because of the polar dependence of the photoemission matrix element.
The normalization procedure is demonstrated on an excerpt from the raw data of the XPD measurement. A more detailed description is given in the supporting information. Fig. 11(b) shows the photoelectron intensity for a single polar scan trace at ' = 47 after peak integration. The image contains diffraction features on top of the slowly varying, non-structural and distribution. The normalization function Nð; Þ is calculated by averaging Ið; ; 'Þ over all measured ' settings and subsequent smoothing. Panel (c) shows that the diffraction features are washed out after averaging. This is most easily obtained if the azimuthal scan steps do not coincide with the symmetry of the sample. The normalization function is smoothed in and using a locally weighted regression (LOESS) algorithm (Cleveland et al., 1992) with a smoothing factor large enough so that the smooth distribution varies slower than the diffraction features. By dividing I=N we finally obtain the distribution shown in panel (d). The features that are not related to diffraction have been successfully removed. The normalized distribution is finally mapped to the stereographic representation in Fig. 9(c).

Quantum well states in a metal-organic network
Metal-coordinated organic networks provide one possible route to integrate and connect molecular electronic devices with the help of self-assembly (Barth et al., 2005). In these networks, the spatial extent and the energetic alignment of the electronic states at the interface can be tuned by a judicious choice of molecular building blocks (Scheybal et al., 2009;Seufert et al., 2013;Wang et al., 2013). The mutual interaction of electronic states of adsorbate and substrate is, however, complex and poses a challenge to current numerical methods for theoretical predictions. Experimentally, the properties of occupied electronic states in ordered systems, including their degree of localization, are probed efficiently in ARPES (Lingle Jr et al., 1994;Lobo-Checa et al., 2009;Puschnig et al., 2011). STM and STS probe the local density of states directly, and are able to detect unoccupied states. Both technical features are helpful in the case of a two-dimensional metalorganic network of 9,10-dicyano-anthracene (DCA) molecules (Zhang et al., 2014). Grown by molecular beam deposition on a clean Cu(111) substrate at room temperature, this network exhibits a long-range periodic 8 Â 8 porous superstructure as can be seen in the STM image in Fig. 12(a). The detailed topography image in panel (b), measured after attaching a single DCA molecule to the STM tip, shows the arrangement of the molecules and the threefold coordination of the cyano groups with Cu adatoms with submolecular resolution.
dI=dV spectra of the clean and DCA covered Cu(111) surface are shown in Fig. 12(c). The kink in the clean spectrum marks the onset of the Shockley surface state at 0.43 eV below the Fermi level. In strong contrast, the spectra of the DCA network, probed at different sites in the unit mesh, show distinct peaks of unoccupied states. Based on the site dependence of their amplitude we assign the peak at +0.8 eV to the molecular lattice, and the peak at +0.14 eV to a surface quantum well state (QWS) inside the pore (Zhang et al., 2014). The confined spatial distribution of the QWS peak becomes evident in a constant-height dI=dV map at +0.14 V bias in panel (d).
Quantum well states are a result of the confinement of a dispersive state, in this case the free electron-like Shockley surface state of Cu(111), in a potential well imposed by an atomic structure of lower dimensionality (Crommie et al., 1993;Bü rgi et al., 1998;Baumberger et al., 2002;Seufert et al., 2013). The confinement can be treated in the same way as the quantum mechanical particle in a box. The states inside the pore have to fulfill both the quadratic dispersion relation of the surface state, EðkÞ = h -2 k 2 =2m ? , and the boundary conditions of the quantum well which allow only a discrete series of states. The allowable wavevectors k n are given essentially by the reciprocal area of the quantum well (Li et al., 1998;Kaufman et al., 1999). Since larger pores are present in small concentration at domain boundaries of the DCA network, states with different wavevectors can be probed with STS as illustrated in Fig. 13. Panel (a) shows that the first-order peak in the largest pore A appears at a lower energy than the corresponding peak in the smaller pores B and C. In addition to the first-order peak measured in all pores, second-order  peaks are observed in the larger pores A and B. Using the estimated effective area of the quantum wells [blue boundaries in panel (b)], the dispersion of the QWS is plotted in panel (c). We notice that, with respect to the unperturbed surface state, the dispersion of the QWS is shifted by 80 meV, and the effective mass is slightly (but not significantly) increased. We interpret the shift as a result of the overlap of the wavefunction with the finite confining barrier imposed by the molecular network (Zhang et al., 2014). This way, QWS can be used as a sensitive probe of the potential landscape in molecular adsorbate systems.

Circular dichroism in photoelectron diffraction
A bending magnet produces a superposition of linearly and circularly polarized synchrotron radiation. In the deflection plane, the light is linearly polarized, whereas the light emitted out of the plane contains a significant fraction of circularly polarized light. At the PEARL beamline, the trajectory of the stored electron beam inside the bending magnet can be tilted to extract partially polarized light in the same way as introduced earlier at the PolLux beamline X07DA (Raabe et al., 2008;Dunn et al., 2004).
Circularly polarized radiation is often used to study ordered magnetic moments in atomic systems due to X-ray magnetic circular dichroism. Furthermore, circular dichroism in the angular distribution of photoelectrons has also been observed for non-magnetic systems. In Fig. 14, we demonstrate the transfer of angular momentum from the circularly polarized photon to the emitted photoelectron, which gives rise to a parallax shift of the forward-focusing peak in the angular distribution of the photoelectron intensity (Daimon, 2001). The curves show the integrated area of the Cu 3p 3=2 photoelectron peak as a function of the azimuthal rotation angle ' (cf. Fig. 4) at three specific polar angles . In the scan at = 35 , the peak and shoulder pattern at ' = 120 is attributed to the forward-focusing of the photoelectron along the [110] direction (the nearest-neighbor direction in the Cu f.c.c. crystal). While the photoelectrons excited by linearly polarized photons are detected exactly at 120 , the photoelectrons which were excited by a circularly polarized photon deviate from the straight path and give rise to the shoulders at either side of the [110] direction, depending on the helicity of the photon. The effect is also seen in the [100] direction which corresponds to the second-nearest neighbor direction. Since they depend on the distance between the emitting and the scattering atom, such forward-focusing parallax shifts can be used as a means to measure interatomic distance (Daimon, 2001). If the atomic geometry is known, the variation of the forward focusing peak intensity can reveal site-specific local electronic and magnetic information (Matsui et al., 2008(Matsui et al., , 2015.

Summary
The performance measurements and the scientific examples show that the PEARL beamline of the Swiss Light Source is equipped for a wide range of surface science problems which can benefit from a combination of complementary experimental techniques. In particular, atomic structure can be studied with both local and space-averaging techniques. PEARL is one of very few synchrotron beamlines world-wide that are dedicated to photoelectron diffraction in angle-and energy-scanned modes. Because the instrumentation is mostly the same as for most photoemission spectroscopy methods, the  (c) Energy dispersion of the quantum well states. Red dots are deduced from the peak position and pore size measured in a series of STS and STM measurements. Error bars indicate the uncertainty of determining the size of the pore. The fine dashed curve is a parabolic fit through the data points. The broken curve is the dispersion of the Shockley surface state on bare Cu(111) for reference.

Figure 14
Angle-scanned photoelectron diffraction of the Cu 3p 3=2 photoelectron peak of a Cu (111) surface measured with left-(CÀ) and right-handed (C+) circularly polarized light at h = 679.5 eV and E kin = 600 eV. Three pairs of azimuthal distribution curves are measured at polar angles of 35 , 55 and 70 , cutting across the [110], the [100] and an arbitrary crystal direction, respectively. The curves are symmetrized for a slight difference in the degree of polarization at the two beamline settings. Sparse markers are drawn at every tenth data point.
beamline also supports the spectroscopy of core levels, resonant excitations, Auger modes or valence bands. The high resolution and stable imaging over several hours demonstrate the successful implementation of a low-temperature scanning tunneling microscope at a synchrotron facility. PEARL is open to users from the surface science community. Proposals are accepted semi-annually during the regular calls of the Swiss Light Source.