XUV double-pulses with femtosecond to 650 ps separation from a multilayer-mirror-based split-and-delay unit at FLASH

In order to probe the complex dynamics of light–matter interaction on all relevant time scales, a multilayer-mirror-based long-range split-and-delay unit was installed as a part of the CAMP end-station at the FLASH free-electron laser in Hamburg.

Of fundamental interest are the FEL-induced dynamics. One important aspect is ultrafast radiation damage, which limits the achievable resolution of CDI (Quiney & Nugent, 2011;Ziaja et al., 2012; see also Garman & Weik, 2017, and references therein). More general, the high power density provides a controlled way to prepare extreme plasma states and to study them with unprecedented resolution Gorkhover et al., 2012;Schroedter et al., 2014;Vinko et al., 2015;Ferguson et al., 2016). The complex and intertwined dynamics of electrons and ions proceed on various timescales from the femtosecond to the nanosecond regime, as, for example, indicated by studies on atomic clusters (Ferguson et al., 2016;Gorkhover et al., 2016;Schü tte et al., 2015;Flü ckiger et al., 2016). In pump-probe experiments using an optical laser, time delays up to the nanosecond regime are conceptually easier to realize. So far, existing short-wavelength split-and-delay units (SDUs) only reach pulse separations of up to a few picoseconds (Roling & Zacharias, 2016).
The CAMP instrument (Strü der et al., 2010), which has been very successful in the first few years at the X-ray FEL LCLS , was recently installed as a permanent end-station at beamline BL1 (Erk et al., 2018) of the FLASH FEL in Hamburg (Ackermann et al., 2007;Feldhaus, 2010). In order to investigate the complex FEL-induced dynamics on femtosecond to nanosecond time-scales with a second FEL pulse, the split-and-delay unit DESC was designed and integrated into the BL1 beamline in front of the CAMP end-station. With a maximum time-delay of 654 ps (maximum negative time-delay: 29 ps), this SDU enables XUV-pump-XUV-probe experiments in a, so far, unexplored time range. At the same time a (sub-)femtosecond precision also allows for the investigation of faster processes. This paper is organized as follows. The basic concept of the split-and-delay unit is described in x2. x3 covers the key components of the setup and their implementation in the CAMP end-station at FLASH. In x4, wavelength tuning and alignment procedures required after an exchange of the mirrors are explained. x5 introduces procedures for establishing the temporal overlap ('time zero') in time-resolved experiments and to correct for misalignment of the spatial overlap of pump-and probe-foci when the delay is changed. As an example, time-resolved ion spectra of atomic clusters are used to illustrate the optimization procedure for long timedelays.

Basic design
Recently, the FLASH beamline BL1 was upgraded and equipped with a permanent end-station, comprising the experimental chamber CAMP (CFEL-ASG Multi Purpose chamber) (Strü der et al., 2010; Erk et al., 2018), a microfocusing Kirkpatrick-Baez (KB) mirror-pair (Erk et al., 2018), and the split-and-delay unit DESC, which is presented in this paper.
The design of DESC was motivated by the need to cover a time-delay from femtoseconds to hundreds of picoseconds. In the XUV and soft X-ray regime, grazing-incidence angles are usually chosen for SDUs that yield a high reflectivity for a wide range of photon energies Wö stmann et al., 2013;Roling et al., 2017); for a comprehensive review of short-wavelength SDUs, see Roling & Zacharias (2016). Grazing-incidence geometries require very long mirrors, and the achievable time-delays are commonly restricted to the order of 10 ps due to their lever arms of a few meters (Roling & Zacharias, 2016). Hence, to reach the required path difference on the order of 20 cm, multilayer mirrors and an almost normal-incidence angle were chosen.
On the other hand, a (sub-)femtosecond precision was desired, in order to enable experiments in a special operation mode, where the multi-spike pulse structure of the SASE-FEL FLASH (Self-Amplified Spontaneous Emission) (Dü sterer et al., 2014) is reduced to a single spike of below 10 fs duration . In addition, a better time resolution than given by the pulse duration was demonstrated with longer SASE pulses by making use of their spike-limited coherence length (Jiang et al., 2010b;Meyer et al., 2012;Fung et al., 2016).
To account for both short and long delay ranges, the movable (delay-)mirror of DESC is mounted on a combination of two linear stages for nanometer precision and decimeter range, respectively.
As a trade-off for reaching long delays, the use of multilayer mirrors restricts experiments to a single photon energy, and the reflectivity of a single multilayer mirror is usually lower than that of a grazing-incidence mirror. However, with the currently installed highly reflective multilayer mirrors optimized for the wavelength of 13.5 nm (mirror characteristics further described in x3.5) and by using a two-reflection beam path, a total transmission of 41% is achieved (the reflectivity of multilayer mirrors optimized for other wavelengths may be lower). Combined with the KB focusing optics of the BL1 beamline (Erk et al., 2018), the achievable peak intensity in the interaction region is above 5 Â 10 14 W cm À2 for a single DESC arm and thus allows for the study of multiphoton processes (Sorokin et al., 2007).

Beam path
The incident FEL wavefront is split by two mirrors and reflected onto a third mirror, which guides both beams downstream to the CAMP instrument (see Fig. 1). A difference in optical path length, and, thus, a time-delay between the two pulses can be achieved by moving one of the splitmirrors along the axis of the incident FEL-beam. This nearnormal-incidence geometry based on multilayer mirrors allows for a compact beam path design.
A reasonable high transmission is achieved with DESC since only two reflections are needed; usually four or more reflections have to be used for grazing-incidence SDUs beamlines 1518 Mario Sauppe et al. Multilayer-mirror-based split-and-delay unit at FLASH (Roling & Zacharias, 2016). We note that another long-delay approach via multilayer mirrors is based on four 45 reflections (Roling & Zacharias, 2016).
DESC is placed before the KB focusing optics and is activated by moving the split-mirrors into the FEL-beam [motorized split-and-delay unit in Fig. 2(a)]. The mirrors M1 and M2 split the incident wavefront into two horizontally separated parts. Split-ratios other than 50% /50% [as shown in Fig. 2(c)] can be adjusted flexibly by changing the lateral position of the split-and-delay setup. Both beam parts are reflected under almost normal incidence on mirror M3, which directs the beams into the experimental chamber with a parallel offset of $ 15 mm with respect to the incident FEL-beam [offset 1, center to center, see Fig. 2(c)]. This parallel offset is compensated by moving the chamber for the KB optics and the CAMP chamber accordingly; both are installed on rails and on a fully motorized frame, respectively.
With this approach of wavefront splitting, a small percentage of photons is lost in the gap between M1 and M2 since a certain distance of the order of 200 mm is needed to avoid a collision when reaching time-zero. Note that changing the time-delay is accompanied by an offset of the probe beam with respect to the pump beam [offset 2, see Fig. 2(c), compare also the beam path in Fig. 1]. In other words, the large gap between pump-and probe-beam up to 1 mm, observed at longer delays on the Ce:YAG screen [cf. Fig. 2(c)], is not associated with further photon loss.
In order to transport the split-beam into the CAMP interaction region without clipping the split-beam by the geometric acceptance of the KB optics of $ 10 mm, its full diameter (pump-beam and probe-beam, plus gap) cannot exceed this value. This is realized by the very small design angle of = 0.7 between incoming and reflected beam.

Time-delay
The time-delay between pump-and probe-pulse is achieved by moving the mirror M2 parallel to the incident FELbeam axis via the combination of a short-range and a long-range linear stage, as further detailed in x3.4 and Table 2. The time-delay is calculated via the formula with Áz = z M2 À z M1 being the distance between the positions of the mirrors M1 and M2 along the z-axis, = 0.7 being the full angle between incident FELbeam and reflected beam (cf. Fig. 1) and c the speed of light in a vacuum. Temporal overlap by means of zero time-delay (t 0 ) occurs when both mirrors are at the same position along the z-axis, i.e. at z M1 = z M2 [cf. Fig. 2(a)]. As explained in x5.1, in a recent timeresolved experiment, the temporal overlap of pump-and probe-pulses was identified at a position of the long-range stage of z M2;long ðt 0 Þ = 4.38 mm (full range  (1) Multilayer mirror M2 which can be moved with a (3) short-range or (4) long-range linear stage with respect to the (2) fixed multilayer mirror M1. (5) Fluorescent Ce:YAG screen, which can be introduced into the beam path for monitoring the beam(s). (b and c) Camera view of the FEL-beam(s) on the Ce:YAG screen: (b) intensity profile without SDU, (c) intensity profile at active SDU and 50% / 50% split-ratio at the full delay. Note that a lateral offset of the split-beam with respect to the incoming FEL results from the z-shape of the DESC beam path (offset 1). In addition, the finite incident angle leads to a time-delay-dependent gap in between pump-and probe-beam (offset 2). The ellipsoidal shape of the beam profiles is caused by a 45 angle of incidence on the screen. The ring structures overlaying the intensity profiles originate from beamline-apertures in the FEL tunnel. The bright oblique lines indicate the edges of the screen (edge length = 25 mm).

Figure 1
Schematic beam path of DESC (top view). For better visibility, the angle between the incoming and reflected beam is exaggerated (in reality = 0.7 and the distance between mirrors M1 and M3 is 1166 mm). DESC is activated by jointly moving M1 and M2 (gray box) into the incident FELbeam (along the x-axis), which is then split into two parts. Mirror M1 is fixed along the z-axis, M2 is installed on two linear stages (short-and long-range) for delaying one part of the beam (yellow) with respect to the other (green). The red dashed lines illustrate the optical path difference. Both beams are reflected in the downstream direction (towards the KB optics and experimental chamber) by a third multilayer mirror M3, with a parallel (x-)offset (indicated by the dashed-dotted arrow) with respect to the incident direct FEL-beam (SDU not active). All mirrors can be adjusted in pitch and tilt (cf. x3.4).
102.49 mm), and of the short-range stage at z M2;short ðt 0 Þ = 125.65 mm [full range 250 mm; note that a position of z M2;short ðt 0 Þ close to mid-range was chosen on purpose]. Thus Áz is calculated by The exact values given here have to be checked and/or reestablished for each experiment. A detailed description for this procedure is given in x5.1.

Vacuum apparatus
As shown in Fig. 3, the SDU is placed in a vacuum apparatus consisting of three interconnected chambers (D1, D2, D3). From upstream to downstream, the incident FEL-beam first passes chamber D1 housing M3 and chamber D2, which contains an optical beamsplitter for the initial alignment procedure (cf. x4.1). Finally, in chamber D3, the FEL-beam is split and delayed by the wavefront split-mirrors M1 and M2, and sent back upstream.
Chamber D3 is based on a DN 400 CF tube, providing sufficient space for comfortable installing and maintaining the interior as well as for exchanging mirrors M1/2 through DN 400 CF door openings. Also, the third mirror M3 in D1 can be easily accessed through DN 150 CF flanges. The whole apparatus measures 1.8 m along the z-direction (beam axis) and 0.5 m in the x-direction, thus being relatively compact considering the achievable long time-delays in the shortwavelength regime.
The SDU is operated at a base pressure of 10 À9 mbar, sufficient for a low contamination of the mirrors and consequently a long lifetime at high reflectivity. For this reason, all installed materials and motors were chosen to be ultra-highvacuum compatible. The chamber is opened (e.g. for mirror change) using a mobile flow box in order to keep dust contamination of the setup as small as possible. Pumping down to the 10 À6 mbar range is achieved with a turbomolecular pump, typically after $ 36 h. Then, two ion getter pumps (410 l s À1 overall) are started, and the turbomolecular pump is switched off and disconnected, isolated by a metal sealed valve. Ion pumps were chosen to avoid mechanical vibrations and for their almost maintenance-free operation. Reaching the 10 À9 mbar range takes 20 to 30 days mainly affected by (1) Granite stone, (2) breadboard on top of granite stone, (3) optical setup for beamline laser, (4) bellow at incident FEL-beam, (5) bellow compensating elongation during bake-out, (6) rail-guided bellow required for lateral offset split-beam in active SDU, (7) ion getter pumps, (8) breadboards on a three-point bearing (on top of chamber breadboards), (9) linear stepper motor for SDU activation, (10) guide rails supporting aluminium board on the top, (11, 12) piezo-driven small linear stages, (13) stepper-motor-controlled long-range delay stage (shown at full delay), (14) piezo-driven short-range delay stage, (15) optical beamsplitter for alignment with the beamline laser, (16) set of filters for attenuation of pump-and probe-beam with a camera for alignment, (17) Ce:YAG and diffusing screen (for beamline laser) with a camera for beam monitoring. out-gassing of motors and cables. Bake-out of the apparatus can accelerate the pumping process, reaching the 10 À9 mbar range after $ 10 days (maximum 80 C, limited by in-vacuum motors). A flexible bellow between D1 and D2 accounts for elongation of the apparatus during the bake-out process with respect to the granite base (cf. x3.3). Temporal and spatial overlap might be altered by the bake-out and have to be rechecked (cf. x5.1, x5.2).
It is important to note that pumping down to 1 mbar and venting to 1 bar has to be done very slowly (within at least 1 h) to avoid destruction of the installed ultrathin filter foils (cf. following x3.2).

Beam monitoring and filters
Precise control of the beam path is essential for an accurate alignment. For monitoring the direct beam as well as the double-pulses, a fluorescent Ce:YAG screen (stacked on a diffusion screen for the optical beamline laser) is fixed on a motorized z-manipulator and can be introduced into the path of the exiting beams [cf. Fig. 2 and (17) in Fig. 3]. For initial alignment purposes with the optical beamline laser, a glass beamsplitter may also be introduced into the beam path halfway (chamber D2) between M1/2 and M3 [cf. Fig. 3, (15)], as further detailed in x4.1.
A set of zirconium filter foils [ Fig. 3, (16)] offers the possibility to vary the intensity ratios of the pump-and probepulses at a geometrical split-ratio of 50% /50% [as in Fig. 2(c)]. Currently, Zr filters with a thickness of 300 nm are installed on a motorized xyz-manipulator, attenuating the beam at 13.5 nm wavelength down to one-third (Henke et al., 1993). In addition, Zr filters with a thickness of 650 nm [10% transmission (Henke et al., 1993)] are available but have to replace the 300 nm foils. The filter holder also includes blockers for separately blocking the pump-or probe-beam.
In case of an exchange of the multilayer mirrors for other wavelengths (cf. x4.4), the filters can be adapted as well. Transmission curves can be calculated online using the CXRO X-ray database, based on Henke et al. (1993).

Beamline integration
DESC is placed upstream of the CAMP experiment and the KB focusing optics. The required space of the existing beamline-infrastructure (cooling water, gas lines, cable tray) prevented a direct mounting of the SDU to the floor. Hence, a bridging aluminium frame was chosen, supporting a 650 kg granite stone, that reduces vibrations from the surrounding experimental hall, with the chambers fixed on top.
As described in x2.2, the active mode of DESC is accompanied by the parallel offset of the split-beam in the horizontal plane [offset 1 (x-axis), cf. Figs. 1 and 2(c)]. To allow for lateral displacement of all downstream beamline components [i.e. connecting parts to KB optics, chamber with KB optics, CAMP chamber (Erk et al., 2018)], a rail-guided DN 63 CF bellow is mounted at the exit of DESC.
A separation of the vacuum system by two gate valves, directly upstream and downstream of DESC, enables inde-pendent vacuum operation of the different beamline sections, e.g. for mirror exchange.

In-vacuum motors
The whole SDU is controlled by eight in-vacuum motors, which can be classified into two categories (see Table 1): (a) those needed for the operation of DESC, i.e. before or during each experiment, and (b) those necessary for alignment procedures, e.g. after an exchange of mirrors.
DESC is activated by moving mirrors M1 and M2 horizontally into the incident FEL-beam [cf. Fig. 2(a)] with a stepper-motor-controlled linear stage. For this purpose, the split-and-delay setup in chamber D3 is situated on a railguided aluminium support plate. The beam split-ratio can be monitored on the fluorescent Ce:YAG screen at the exit of DESC (cf. Fig. 2).
Time-delays between pump-and probe-beam are adjusted by a short-range and a long-range linear stage. The shortrange delay stage is mounted on top of the long-range delay stage. Both stages move mirror M2 along the incident FELbeam. The piezo-operated short-range delay stage allows a bidirectional movement in 14 attosecond (as) time-steps with a repeatability of 14 as. However, fluctuations of roughly AE 200 as (encoder read-out) due to mechanical vibrations of the SDU limit the absolute temporal resolution. While the range of the short-range delay stage is AE 830 fs, the long-range delay stage allows up to 654 ps time-delay with a step size of 7 fs, a uni-directional repeatability of 2 fs and a bi-directional repeatability of 7 fs.
The short-range delay stage uses an encoder to determine the position, while the position for the long-range delay stage is calculated by the steps of the motor. The main specifications of the delay stages are listed in Table 2.
Inevitably, adjusting time-delay induces a loss of spatial overlap of the foci due to the pitch and yaw movement of the delay stages connected to their finite mechanical precision, as further discussed in x5.2. For quick realignment, mirror M2 was equipped with a fast adjustable closed-loop mirror mount. Here, a piezo-driven tip/tilt platform was chosen, allowing for 0.05 mrad resolution and 0.15 mrad repeatability. Fluctuations  Table 1 Overview of the in-vacuum motors for (a) DESC operation and for (b) DESC alignment procedures after mirror exchange.
See text for a detailed explanation and Fig. 3 ) is minimized with the help of a small linear piezodriven stage, which moves M2. Because of the small distance between mirror M3 and the incident FEL-beam, M3 is mounted on a small linear piezo-driven stage along the x-axis. If larger FEL diameters than usual were required in inactive SDU mode, this stage would allow moving M3 out of the beam.

Mirrors
Photon energies at FLASH1 may be tuned from 24 eV to 295 eV in the fundamental (Schreiber & Faatz, 2015), a range that is fully covered by suitable multilayer materials (Bourassin-Bouchet et al., 2015). Currently, a set of flat multilayer mirrors with a design wavelength of 13.5 nm is installed in the SDU. The multilayers consist of 50 periods of molybdenum/silicon on a fused silica substrate (surface flatness /20), with the thickness of every single period of 6.900 AE 0.005 nm (Feigl et al., 2006). A maximum reflectivity of 64% at a wavelength of 13.5 nm ($ 91.8 eV) for each mirror is achieved. The transmission through DESC, after passing M1 (or M2) and M3, is calculated via convolution of the particular reflectivities, to be 41%, as visualized in Fig. 4(a). Measured transmissions have been observed to be lower sometimes (cf. x4.2), maybe depending on FEL tuning conditions.
The mirrors are exchangeable for other desired wavelengths; replace-ment mirrors would have to be user provided. However, the costs are comparatively low because of the small mirror dimensions, resulting from the near-normal-incidence geometry. The mirror sizes are specified in Fig. 4(b).
The facing planes of M1 and M2 were polished for sharp edges, resulting in an unimpaired splitting of the incident wavefront. Since the distance between the edges of M1 and M2 is very small (of the order of 200 mm), the inner plane of M2 was beveled in order to avoid a collision when M2 passes M1 towards negative time-delays.

Alignment procedures and wavelength matching 4.1. Initial alignment
During the initial setup of DESC, after adjustment of the chambers on the granite stone, all mirrors and motors were leveled with a three-point bearing on their breadboards and fixed afterwards. A rough beam alignment on beamline laser was realized using large-angle mirror mounts. Note that since the range of the tip/tilt platform for M2 is limited to AE 1 mrad, M2 is additionally installed in a manual mirror mount with a large angular range, which can be adjusted when the chamber is opened.
Theoretically, an ideal beam alignment of DESC is reached when the incident angles on the KB mirrors with and without DESC are identical. Then, the focus quality in the CAMP chamber is optimal. Such an ideal alignment can be achieved via two steps. First, a parallel path of the outgoing split beams with respect to the direct FEL-beam is established using the optical beamline alignment laser, which is discussed below in this section. Second, the slight change of the angles of the KB chamber itself, induced by its lateral motion, is compensated via a direct optimization of the FEL-beam angles by tuning  (a) Reflectivity versus photon wavelength of the currently installed multilayer mirrors. Blue: manufacturer data (measured by PTB@BESSY II); yellow and green: total reflectivity of pump-beam path and probe-beam path, calculated by M1 * M3 and M2 * M3, respectively. Here a reflectivity of 41% is reached at 13.5 nm. At a geometrical split-ratio of 50% / 50%, each beam transmits $ 20% of the originally incident photons. (b) Geometrical dimensions of the mirrors. Table 2 Main specifications of the short-range and long-range linear delay stages.
All times Át are calculated with equations (1) and (2)  To achieve the first step, i.e. a parallel alignment of the outgoing split beams with respect to the incident beam, an outof-vacuum optical setup is used, as depicted in Fig. 3, (2, 3). An optical beamsplitter mounted on a linear manipulator in chamber D2 is moved into the beam path of the beamline laser [ Fig. 3, (15)]. With an incident angle of 45 and a 30% / 70% intensity splitting ratio (reflected/transmitted), both the incident beam and the split-beam are coupled out side-by-side through a viewport towards the out-of-vacuum breadboard. There, the setup essentially consists of a lens and a camera within the focal plane of the lens. By spatially overlapping all three focal spots (incident beam, split beams), a parallel path of the split-beam relative to the incident beam is ensured.

Wavelength tuning
An exact wavelength tuning of FLASH prior to any experiment using DESC is crucial, since a deviation of only 0.3 nm from the ideal wavelength would cause a drastic transmission loss [cf. Fig. 4(a)]. Matching the wavelength to the mirrors' reflectivity may be performed by an absolute measurement with the plane-grating monochromator at the PG beamlines (Martins et al., 2006;Gerasimova et al., 2011), or by tuning directly on an optimized transmission through DESC, as briefly described in the following.
The intensity profiles of the direct beam and the transmitted beam can be observed on the same Ce:YAG screen mounted on the beam monitor at the exit of DESC (cf. x3.2). The ratio of the transmitted beam intensity and the direct beam intensity is determined [see, for example, different intensities on the Ce:YAG in Figs. 2(b) and 2(c)] and tuned for maximum values. As tested in several experiments, tuning on the highest intensity ratios can be used to match the wavelength to the reflectivity window of the multilayer mirrors of DESC.

Focus optimization
In addition to a residual misalignment of the beams through DESC after the steps described in x4.1, the lateral shifting of the chamber with the KB optics into active SDU mode also slightly changes the angles of incidence compared with the position of inactive SDU mode, caused by inevitable pitch and yaw errors of its rails. However, having the mirror angles optimized once, we found the change in pitch and yaw angles of the KB optics between the modes 'DESC inactive' and 'DESC active' to be reproducible when the KB chamber was moved between precise limit stoppers. Therefore, one can automatically correct these angular offsets when switching between modes.
The focal waist of the KB optics is extremely sensitive to the incident angles of the beam. Hence, we use a power-densitydependent signal from the focal region in the center of the CAMP chamber for an exact alignment of the SDU mirrors in terms of smallest foci. The focus quality can, for example, be characterized by nonlinear ionization processes, e.g. of xenon gas, and optimizing the creation of highest ion charge states (cf. Erk et al., 2018).
To rule out any double-pulse effects during focus optimization of DESC, pump-and probe-beams are aligned separately. Because the delay mirror M2 is mounted on a fast and reproducible closed-loop mirror mount (cf . Table 1), the probe-beam path is aligned first by reflecting 100% of the incident FEL-beam on M2 and adjusting M2 and M3 iteratively. Subsequently, the pump-beam path can be easily aligned. The full FEL-beam is now reflected by M1 and the M2 alignment is retraced via digital markings on the DESC beam monitor [Fig. 3, (17)] and a fluorescence screen in the focus of CAMP. As a final step, the pitch and tilt angles of M1 are finetuned, e.g. by optimizing on high xenon charge states.
For a quantitative comparison of the KB focus qualities with and without DESC, beamline filter foils can be used to attenuate the direct beam (DESC inactive) in order to match the reduced transmission through the SDU. Fig. 5 illustrates the outcome of the focus optimization procedure in a commissioning beam time in summer 2016 after mechanical changes in DESC. Here, the FEL was tuned to 100 mJ pulse energy at 13.5 nm wavelength [single bunch, not including beamline transmission, cf. Erk et al. (2018)], and xenon gas with a partial pressure of $ 1 Â 10 À6 mbar was introduced into the CAMP chamber. In active DESC mode, a maximum xenon charge state q = 12 was found (full beam on probe-arm). The measurement with reduced transmission in inactive DESC mode [$ 37%, using a 200 nm niobium foil (Henke et al., 1993)] resulted in the same charge state, indicating a good alignment of DESC.

Mirror exchange
All mirrors are accessible through large flanges (cf. x3.1). This allows the installation of multilayer mirrors optimized for wavelengths other than 13.5 nm, if required in future experiments. As mentioned, to keep the beamline and the interior of DESC dust free, a portable flow box should be used. By using the alignment of the installed mirrors, the successive exchange of one mirror after the other will allow for a quick alignment restoration. For instance, the focus positions on a screen in the CAMP chamber provide a long lever and thereby a precise marking. The alignment should be checked and may be optimized after all mechanical changes in DESC, and procedures to establish temporal and spatial overlap have to be repeated (cf. x5).

Temporal overlap
Precise knowledge of the temporal relation between pumpand probe-pulses is necessary for any time-resolved experiments. Thus, establishing a 'zero delay' time reference is essential. In particular after any mechanical change in the DESC setup such as the exchange of mirrors, reestablishing of the temporal overlap between pump-and probe-pulses is beamlines required. In addition, it is recommended to confirm 'time zero' as a check-up procedure before every experiment.
To this aim, the two split beams can be overlapped on a fluorescence screen, providing access to a spatial interference pattern in case of temporal overlap (Mitzner et al., 2008). As the procedure does not rely on a focused beam, it is not necessary to correct for pitch and yaw motion of the mirror M2 (cf. x5.2) while scanning time-delay for temporal overlap. However, considering the femtosecond pulse duration and the particularly short longitudinal coherence time of a SASE FEL such as FLASH (Mitzner et al., 2008;Schlotter et al., 2010;Roling et al., 2011), the determination of temporal overlap starting from an arbitrary position can become a tedious and time-consuming procedure. As the fringe pattern is only visible within the delay time interval corresponding to the FEL pulse coherence time, typically in the sub-10 fs range, a small step size within a fairly large scanning window is required. When found, maximizing the fringe pattern visibility ensures that the two partial pulses arrive in the common focal region in a sub-10 fs interval, which is typically shorter than the FEL pulse duration.
In a recent experiment, zero time-delay at 13.5 nm wavelength was identified at a short-range delay stage position of z M2;short ðt 0 Þ = 125.65 mm while the long-range delay stage was at position z M2;long ðt 0 Þ = 4.38 mm. An interference pattern, shown in Fig. 6, was clearly visible on a screen $ 80 cm behind the focus of CAMP, confirming temporal overlap, after it had been initially determined by ion signal over the course of the respective experiment. The pattern was observable within a position range of the short-range delay stage of less than 1.5 mm, corresponding to a sub-10 fs longitudinal coherence length of the FLASH FEL, being in good agreement with a previous finding at this wavelength (Roling et al., 2011).
Note that under consideration of the limited bi-directional repeatability of AE 1 mm of the long-range delay stage (cf. Table 2), the value of z M2;short ðt 0 Þ may vary if the long-range stage is moved.

Spatial overlap of pump-and probefoci
Pump-probe experiments usually require high intensities, and thus the use of small foci. When changing time-delay, the spatial overlap of the two foci of the pump-and probe-beam is often lost due to the pitch and yaw movements of the delay stages resulting from finite fabrication precision. Caused by the lateral offset 2 of the probe-beam [cf. Fig. 2(c)], the virtual FEL source point of the diverging FEL beam is also laterally offset (Campi et al., 2016). Taking the magnification of the KB optics into account, this translates into a lateral offset of the probe-focus of up to 10 mm at maximum time-delay. Therefore, keeping the spatial overlap while changing the time-delay is an important, though challenging, task.
Generally, all correction motions in DESC are carried out by the pitching and tilting mirror M2 using the closed-loop actuator. The displacement of the angles from the two delay stages is not fully reproducible and changes systematically in succession after mechanical changes (e.g. exchanging mirrors). Thus, look-up tables with correction values for pitch and yaw Interference pattern at maximum visibility of partially overlapping pumpand probe-beams on a fluorescent screen $ 80 cm after the focus in CAMP. The fringes appear at temporal overlap (zero time-delay) and disappear at a few femtoseconds time-delay, consistent with the coherence time of the FLASH FEL at 13.5 nm (see text). Note that in order to overlap both beams on the screen, focal overlap had to be abandoned during this measurement.

Figure 5
Ion spectra from xenon gas, zoomed in on high charge states. Comparison for (i) DESC inactive, (ii) attenuated intensities at inactive DESC with filter foil, imitating the transmission of DESC and (iii) DESC active. The time-of-flight measurements were averaged and converted to massover-charge ratio. Without DESC, the highest observed charge state was reduced from Xe 14+ at full beam to Xe 12+ using a 200 nm niobium foil (transmission $ 37%). After iteratively optimizing all mirrors of DESC for highest charge states, 12-fold ionized xenon was also observed, indicating a good alignment of DESC. We note that the total lower ion yield at (iii) is induced by slightly changed positions of the retractable time-of-flight spectrometer, which is not relevant for the appearance of the high charge states. Besides Xe q+ , N + at m/q = 14 from residual gas can also be observed in the 'DESC inactive' measurements, which was present only at the beginning of the experiment.
angles are measured on a regular basis and can be used as a first-order correction. Nevertheless, the values have to be checked and adapted for each experiment.
At the CAMP end-station, several tools are available for creating look-up tables and fine-tuning of the overlap. A rough overlap can be established by direct monitoring of the foci with the help of a Ce:YAG screen in combination with a microscope and placing the probe-beam on top of the pumpbeam. It should be ensured that the Ce:YAG screen is in the focal plane by scanning it along the beam axis. However, this method is limited by the resolution of the microscope and saturation of the fluorescent signal.
Alternatively, a velocity map imaging (VMI) spectrometer (Erk et al., 2018), operated in spatial imaging mode and in time-of-flight mode, is a convenient tool for gas-phase experiments to overlap both foci in all three dimensions with micrometer precision (Johnsson et al., 2010). Using proper voltage settings and a gas-flooded chamber, the beam profiles can be spatially imaged, allowing for a direct observation of the focus waist and enabling a comfortable way to overlap the foci along the x-axis. By matching the flight times of ions, it is possible to correct for offsets in height (y-axis, cf. the coordinate system in Fig. 3).
The spatial overlap achieved via Ce:YAG screen or VMI spectrometer should be cross-checked and further fine-tuned via an overlap-dependent signal from the focus, e.g. on ions. For short timedelays, a delay scan around temporal overlap on a highly charged gas ion signal, giving access to high-order intensity autocorrelation functions of the radiation field, should clarify the spatial overlap (Mitzner et al., 2009;Moshammer et al., 2011;Rompotis et al., 2017). For long time-delays in the picosecond range, different mechanisms could be used. As an example, we describe an approach using gas-phase clusters in the following section.

Overlap optimization on ion spectra from rare-gas clusters
Optimizing the overlap of pump-and probe-foci in the CAMP chamber at long delays requires an overlap-dependent signal that is still present at delays of several hundreds of picoseconds. Dynamics up to the picosecond and nanosecond regime have been reported in ion spectra from atomic clusters (Krikunova et al., 2012;Schü tte et al., 2015;Flü ckiger et al., 2016). For the commissioning of DESC and first user experiments, xenon clusters were utilized for overlap optimization, as the overlap-dependent signatures are clearly visible, very sensitive to ideal overlap, and even increase in strength for long delays.
As an example, in Fig. 7 time-of-flight ion spectra of xenon clusters with $ 24 nm radius are shown. They were measured during the course of the first user experiment. The ion spectra were recorded using a look-up table for readjusting the mirror M2 that had been created several hours before under different cluster generation conditions, suggesting a satisfying reproducibility on the scale of one beam-time shift. The correction values had been determined by iteratively optimizing pitch and tilt of M2 by tuning on maximum yield of the higher charged xenon ions for all displayed delays.
The left and right columns of Fig. 7 display the spectra of the pump-or probe-pulse alone, which do not change with the applied delay. In the center column the ion spectra of doublepulse excited clusters are given, showing a striking increase of higher charge states of xenon with a longer delay between the two pulses.
The physics underlying these dynamics is well understood . While the same total amount of energy impinges on the clusters at all time-delays, the recombination following the ionization is strongly suppressed for XUV double-pulses with long delays. The reason for this suppressed recombination lies in the expansion of the cluster nanoplasma after the excitation by the first pulse .  Time-of-flight ion spectra from xenon clusters with an average size of 24 nm in radius, averaged over 200 single FEL shots. The central column presents pump-probe ion spectra from À33 ps to 650 ps (the maximum delay values are not fully identical to Table 2 due to a different mirror holder setup at that time). The signal from higher charge states increases with time-delay, suitable for optimizing the spatial overlap at long time-delays. The left and right columns show corresponding spectra of pump-or probe-pulse only, each obtained with a relatively low power density of about 2.0 Â Generally, the expansion leads to a cooling of the electrons as they transfer their kinetic energy to the ions of the cluster. This cooling effect is decreased for the electrons created by the second pulse which, subsequently, are less likely to recombine.

Summary and outlook
The multilayer-mirror-based split-and-delay unit DESC was set up as a part of the beamline BL1 at the CAMP end-station at the free-electron laser FLASH. By selecting a single wavelength with multilayer mirrors, a beam path design became feasible which allows up to 654 ps time-delay. Furthermore, (sub-)femtosecond steps enable experiments within the coherence time of FLASH. The mirrors are exchangeable for operating DESC at different fixed wavelengths.
The presented SDU creates intense XUV double-pulses reaching into a, so far, unexplored time regime. The parameters of DESC offer the possibility to disentangle fast and slower dynamics such as excitation, ionization, charge transfer and structural dynamics in highly excited non-equilibrium states of matter within a single experimental approach. For example, over the course of the experiment investigating xenon clusters (Fig. 7), also the light-induced expansion of single xenon clusters up to 650 ps was followed via coherent diffractive imaging as a new avenue to study XUV-induced structural changes.
In conclusion, DESC provides novel possibilities for time-resolved experiments using intense XUV pulses with characteristics complementary to other XUV split-anddelay units. It is commissioned and operational for experiments at the CAMP end-station of the FLASH free-electron laser.