The FERMI@Elettra free-electron-laser source for coherent x-ray physics: photon properties, beam transport system and applications

FERMI@Elettra comprises two free electron lasers (FELs) that will generate short pulses (τ∼25–200 fs) of highly coherent radiation in the XUV and soft x-ray region. The use of external laser seeding together with a harmonic upshift scheme to obtain short wavelengths will give FERMI@Elettra the capability of producing high-quality, longitudinally coherent photon pulses. This capability, together with the possibilities of temporal synchronization to external lasers and control of the output photon polarization, will open up new experimental opportunities that are not possible with currently available FELs. Here, we report on the predicted radiation coherence properties and important configuration details of the photon beam transport system. We discuss the several experimental stations that will be available during initial operations in 2011, and we give a scientific perspective on possible experiments that can exploit the critical parameters of this new light source.


Introduction
The dynamics and temporal evolution of matter down to sub-femtosecond time scales and atomic space scales are at the base of all chemical, physical and biological processes [1]. Consequently, the past couple of decades have seen an impressive quest for tools capable of temporally resolving ultrafast processes related to the dynamics of the electronic interaction with other electrons or phonons, photons, magnons and polarons [2]. The characteristic time scales span from as short as attoseconds in the case of electronic processes at the core levels in atoms, to a few femtoseconds for the electron dynamics at the valence band, and to as large as a few picoseconds in the case of processes that involve interactions with heavy particles, such as phonons [2].
The drive towards shorter time scales has been accompanied by a push to probe the structure and chemistry of transient events at their natural nanometer and sub-nanometer length scales [3]. In this respect, coherent light sources have enabled imaging with diffraction-limited spatial resolution and can be a powerful tool to resolve the structure, chemistry and energetics of single functional units. Thanks to a new generation of fully coherent laser sources with short pulses and high peak brightness, it has now become possible to obtain simultaneously both high temporal resolution and high spatial resolution [4]. However, conventional lasers emit radiation only in a limited wavelength range and their use is, in general, restricted to optical and spectroscopic techniques in the infrared (IR), visible and near-ultraviolet range, thus excluding all measurements that require photon energy higher than a few eV. The desire to extend this range to the XUV and harder energies dates back to the time of the first functioning lasers and includes many attempts to generate coherent x-ray pulses, starting with ultrashort pulses from IR lasers [5]. More recently, interest has turned to short-wavelength, free electron lasers (FELs) [6], which can produce light pulses with peak brilliance up to ten orders of magnitude higher than the pulses generated in the present third generation synchrotron light sources and with photon energies spanning from the vacuum ultraviolet to the hard x-ray, i.e. from about 10 eV (120 nm) to 10 keV (0.12 nm). 3 FEL sources can operate in several ways. To date, most of the existing and planned veryshort-wavelength FELs (e.g. FLASH, LCLS, SCSS, XFEL and SPARX) have employed the self amplification of spontaneous emission (SASE) [7] mode of operation. While it is possible to get extremely high brilliance, the temporal structure of a SASE output pulse is normally composed of a series of micro-pulses that individually have random phase and highly fluctuating peak intensity and time duration. For SASE devices, synchronization to external sources is normally limited by the temporal jitter of the accelerator. This jitter can be tens of femtoseconds or greater, especially for accelerators based upon non-superconducting cavities. As an alternative, shortwavelength FEL sources based upon 'seeding' techniques, in which the FEL pulse is initiated by a coherent radiation pulse generated by a conventional laser, can produce output pulses with a well-defined temporal shape and intensity stability [8,9] that permits relatively straightforward synchronization with external pump or probe lasers. In principle, seeded FELs can obtain output radiation bandwidths close to the Fourier transform limit. In this paper, we describe a new light source currently under construction, FERMI@Elettra [10], comprising two FELs that will use a combination of coherent seeding and harmonic upshifting [11,12] to provide coherent emission over a photon energy range of 12-300 eV (and up to 900 eV at the third harmonic). We note that, as proof of the attractiveness of this approach, other groups, such as the s-FLASH project at DESY [13], are also actively investigating short-wavelength seeded FEL sources.
The remainder of this paper is organized as follows. Section 2 summarizes the basic physics principles of the FERMI FEL sources and gives a characteristic sample of the numerical simulation predictions for the expected output radiation coherence properties. Section 3 presents the FERMI photon beam transport and diagnostic systems. We conclude with section 4, which gives an overview of the principal instruments that will make it possible to reach the scientific goals enabled by FERMI's intense, highly coherent radiation pulses in the XUV and soft x-ray region. These include temporally synchronized pump-probe experiments and others involving third harmonic radiation with photon energies as high as 900 eV.

Basic physics principle and configuration
The two FERMI [10] free-electron lasers are based upon the principle of harmonic upshifting [11,12] of an initial, coherent radiation 'seed' signal within a single pass, FEL amplifier configuration employing multiple undulators. FERMI's FEL-1 will cover the lowenergy photon spectral range (20-100 nm, i.e. 12-60 eV) using a single-stage harmonic upshift, while FEL-2 uses two upshift stages to reach output wavelengths (fundamental) as short as 4 nm (i.e. 300 eV). Each upshift stage begins with a relatively short magnetic undulator (the 'modulator'), in which a coherent radiation input signal imprints a relatively strong, coherent energy modulation on the electron beam, whose magnitude is much greater than the initial uncorrelated energy spread σ E . For FERMI's FEL-1 and the first stage of FEL-2, the input signal is provided by a wavelength-tunable, high-power, external UV laser (see table 1). Following the modulator is a relatively weak chromatic dispersion section (R 56 ∼ 40 µm or less) that converts the energy modulation into a coherent density modulation with strong harmonic overtones. Finally, each upshift stage culminates with a 'radiator' undulator whose normalized rms magnetic strength a w is tuned such that FEL resonance occurs at a radiation wavelength where λ w is the undulator period and γ is the electron relativistic Lorentz factor. In both radiators of FEL-1 and the first stage of FEL-2, we plan to use h in the range 2-12. The second stage of FEL-2, whose initial input radiation signal is the output from the first stage radiator, will likely be limited by σ E to a maximum h 5. FERMI will employ the 'fresh bunch' approach [14] for FEL-2, thus requiring a delay chicane between the two stages. This choice, together with the currently expected accelerator pulse timing jitter levels and e-beam pulse length, will likely limit useful seed and output pulse durations to 200 fs or less. The radiators of both FEL-1 and the second stage of FEL-2 are many exponential gain lengths long and are expected to lead to FEL power saturation, as in the classic high-gain harmonic generation scheme described by Yu [12]. By contrast, the first stage radiator of FEL-2 is significantly shorter (∼2 gain lengths or less) because the needed input power for the second stage modulator is much less than that corresponding to saturation. Consequently, there is a 2× or less increase in microbunching from the entrance to the exit, and the resulting radiation is essentially coherent spontaneous emission of a prebunched electron beam.

Predicted coherence properties of the FERMI FEL-2 harmonic cascade
Much of the interest in harmonic cascade FELs over the past decade has stemmed from their promise of producing pulses with a much higher degree of temporal coherence than is normally possible from SASE FELs. While SASE output can have nearly full transverse coherence, the longitudinal coherence is generally limited to a length l c ∼ (L G /λ w ) × λ R , where the term in parentheses is of the order of 50-1000 for reasonably high gain devices. At 4 nm output wavelength, l c /c 13 fs. This limitation arises from the physics of the SASE process and not from imperfections of the electron beam (e.g. energy chirp and current ripples) or the magnetic undulator. By contrast, a harmonic cascade FEL, presuming a perfect external laser seed and an electron beam without any macroscopic variations, can in principle produce a nearly transformlimited output pulse with coherence lengths exceeding 100 fs.
To illustrate the differences in the coherence properties between SASE and that possible with FERMI's FEL-2 operating as a seeded cascade, we have done a series of numerical simulations with the GINGER simulation code [15] using the seed and electron beam 5 parameters listed in table 1. The comparison has been done for an output wavelength of 4.2 nm (i.e. near the K-edge of C). This is close to the shortest wavelength of operation for FERMI FEL-2 and will thus be fairly sensitive to non-ideal electron beam properties. There are three different simulation cases. The first involves a SASE configuration with an ideal, time-steady e-beam and a single undulator; the second employs a two-stage harmonic cascade (208 nm → 21 nm → 4.2 nm), again using an ideal e-beam, and the third is identical to the second but uses macroparticles from a 'start-to-end' (S2E) tracking code simulation of the FERMI electron beam, beginning at the injector and ending at the undulator entrance. The S2E simulation includes the effects of the longitudinal space-charge microbunching instability [16], which leads to temporally localized fluctuations in beam energy and current. However, the noise level from which the instability grows was not initialized according to the Poisson statistics and thus the fluctuation levels may be significantly overestimated. ) has a very smooth output pulse with an FWHM of ∼80 fs, slightly smaller than that of the input seed. The double cascade case with a non-ideal, S2E electron beam (figure 1(c)) has a P(t) profile that is much less smooth, displaying oscillations at the ±25% level.
Near-field power spectra (figures 2(a)-(c)) show a similar range of differences. The SASE case is composed of multiple spikes in the range of ±0.0075 nm, the ideal e-beam cascade case has a single peak with an FWHM of 1.0 × 10 −3 nm and the non-ideal, S2E e-beam cascade case has a dominant spike of similar width, but an equal amount of power is contained in a halo approximately three times larger. The reader should note that the exact details of radiation output in the SASE and S2E cases are sensitive to the initial microbunching distributions and will thus vary from shot to shot.
To investigate the output radiation properties in more detail, we have calculated the Wigner transform W of the far field at the on-axis position (i.e. θ = 0). W effectively measures the local phase space density of the radiation and is defined as The integration of W over t gives the on-axis power spectrum (times a constant factor), while the integration over ω gives the instantaneous intensity (also times a constant factor). Figure 3(a) shows the false color image of the Wigner transform of the SASE output case. There is a chaotic structure in both time and frequency with many individual coherent 'hot spots', each encompassing ∼5 × 10 −3 fs nm in area. The disordered structure is indicative of random phase jumps between different intensity spikes. By contrast, the ideal beam, external laser seededcascade case (figure 3(b)) consists of one large and well-defined coherent region spanning ∼100 × 10 −3 fs nm. The smoothness of the region is indicative of a smoothly varying phase. There is a noticeable linear chirp in wavelength with magnitude ∼ −1.5 × 10 −4 nm fs −1 . This chirp is a natural feature of FEL radiation from short pulses and tends to increase in magnitude as one approaches power saturation. The S2E cascade case has a Wigner transform (figure 3(c)) that is far less ordered than the ideal beam case, albeit far smoother and more confined than the SASE case. Here, the effective coherence time is reduced from the ∼20 fs value of the ideal e-beam case to ∼5 fs. The underlying reason for the apparent breakup from a single region in the ideal beam case to ∼5 topologically distinct regions in the S2E case is a relatively rapid (but ordered) phase variation, which arises from the interplay of ∼3 fs period energy oscillations on the electron beam (due to microbunching instability growth upstream in the accelerator) with the strong chromatic dispersion region in the first harmonic cascade stage. As mentioned above, we have limited confidence that the initial noise level (i.e. in the injector region) for these energy oscillations was correctly modeled by the tracking code simulations, whose output produced Predicted, near-field radiation power spectra for the three cases of figure 1. The power levels depend upon the effective bin width and the temporal output window of the individual simulations and thus quantitatively cannot be compared directly.
the input particle distributions for the FEL simulations. If the initial noise level is a factor of two too high and/or if the first stage chromatic dispersion section strength were to be reduced by a similar factor (e.g. by using a stronger external laser), the output phase oscillation amplitude would decrease by at least two and the number of distinct regions could shrink to two or possibly even one. Consequently, while it will certainly be difficult to get output radiation pulses approaching the near perfection shown in figure 3(b), we believe that there are a number of 'tuning knobs' for FERMI's FEL-2 that will produce output better than the S2E case shown in figure 3(c) (and far superior to the SASE case of figure 3(a)). The transverse coherence for each of the cases presented here is quite good. The fraction of the power that is contained in a 'best-fit' TEM 00 mode is above 90% and the effective emission point in z (i.e. best fit waist location) is typically ∼5 m before the saturation point   in the undulator. The transverse coherence of the seeded S2E case is noticeably worse (albeit of still relatively good quality) than those found for the ideal e-beam SASE and seeded harmonic cascade cases. This difference may be due to a refractive guiding effect caused by the local energy chirp on the electron beam. Altogether, we have reasonably good confidence that the transverse coherence of the FERMI FEL-2 will be adequate for nearly all proposed experiments at wavelengths λ 4 nm and will not require spatial filtering.

Third harmonic emission
We currently believe that the FERMI linear RF accelerator in its present configuration will have output electron energies limited to 1.5 GeV. For the particular choice of undulator wavelength in the final radiator (i.e. 35 mm), output power and photon number levels for fundamental wavelengths below ∼4 nm will drop precipitously because the strong increase in exponential gain length will prevent saturation by the end of the second radiator. Consequently, in order to reach photon energies in the 300-1000 eV range corresponding to L-edges of magnetically active elements (e.g. Fe and Co), we plan to utilize third harmonic emission from FEL-2. When the final radiator is operated with linear or elliptical polarization, coherent emission at third and other odd harmonics occurs naturally with power levels approaching 0.1-0.5% that of the fundamental.
For the case of pure circular (helical) polarization, for aligned electron beams, there is negligible harmonic emission directly along the undulator axis. Coherent emission off axis is strongly (but not fully) suppressed by destructive interference effects. In order to produce reasonable power levels of higher harmonic, circularly polarized emission on axis, we are currently examining a number of options possible with the Apple-2 undulator topology of the FERMI radiators. Most easily, we can operate the undulators with a small (a few per cent level) amount of elliptical polarization. Experiments at the BESSY synchrotron light source [17] have shown that this configuration leads to relatively high levels of circularly polarized emission on axis, at both the fundamental and the odd harmonics, with small contamination (<10%) by linearly polarized emission. A second possibility is the so-called cross-polarized mode [18], where undulator segments alternate between horizontal and vertical linear polarizations (with appropriate phase shifts in between-such phase shifters exist and are needed by the FERMI radiators for other reasons). This can produce quite strong circularly polarized radiation at both the fundamental and the odd harmonics, with predicted purity levels above 90% [19]. Finally, the magnetic field topology of the radiator undulator can be modified to have a large third harmonic component [20]. Such a modification can strongly enhance the coupling between the electron motion and third harmonic emission in all polarization modes without significantly affecting the emission at the fundamental. In a different context, similar topology modifications have been used to suppress harmonic emission [21].
We note that one need not make any of these suggested changes for all the radiator undulator segments. In general, since the FEL power exponentiates with undulator length in the final radiator, just changing the last couple of undulator segments should be adequate. Nearterm experiments on the ELETTRA storage ring FEL and FERMI's FEL-1 will allow us to investigate and optimize various options for enhanced, circularly polarized harmonic emission. In terms of predicted coherence properties, our simulation studies and some externally seeded FEL experiments [8] have shown that the coherence length of the harmonic emission is reduced from that of the fundamental. There are at least two possible effects that can lead to this.
(i) Seeding with non-flat-top shapes (e.g. parabolic or Gaussian profiles) in a high gain situation often leads to a reduced pulse width for the overall power at higher harmonics. (ii) If there are small eikonal phase variations on the fundamental of amplitude ± , these will be increased by a factor h at harmonic number h [22]. The variations, once they approach a level of ∼ π, limit the coherence length to values below that of the full radiation pulse. Conversely, to ensure that the longitudinal coherence length at a higher harmonic h does not drop much when compared to the fundamental requires at least a reasonably flat temporal power profile and eikonal phase variations of amplitude π/ h. Similar phenomena can be expected for the transverse coherence properties. For both SASE and seeded configurations, one expects a drop-off with radius in both the harmonic microbunching fraction and the corresponding harmonic emission strength. Thus the effective waist size r h of the harmonic emission will be less than that found for the fundamental. This reduction appears to have been detected in coherent optical transition radiation measurements at the LEUTL SASE experiment at Argonne [23]. The sensitivity to phase variations also applies transversely and may decrease the overall coherent power fraction contained in the TEM 00 mode. Since the natural e-beam transverse shape is close to a Gaussian, there may be no simple way to prevent the effective waist size from decreasing between the fundamental and higher harmonics. On the other hand, since the Rayleigh range at harmonic h scales directly with the factor (hr 2 h ), the reduction in waist size tends to equalize the Rayleigh ranges. This might help with downstream experimental design. Figure 4 summarizes the predicted output levels from FERMI's FEL-1 and FEL-2 as a function of final wavelength. The chosen electron beam parameters are those of table 1 but with the current reduced to 750 A, the transverse normalized emittance increased to 1.0 mm mrad and the incoherent energy spread increased to 500 keV (to take into account the effects of the energy modulation by the input radiation seed). The solid and dashed lines correspond to the predicted radiation energy for an output pulse of 40 fs duration that has reached saturation according to the Xie empirical formula [24]. Predictions at wavelengths below 3 nm correspond to the third harmonic emission in linear polarization. While there is non-negligible emission at the fifth harmonic, typically it is less than 1 part in 10 4 of the fundamental.

The photon beam transport system
The photon transport system of the FERMI project is divided into two distinct parts: the first one common to all the beamlines, called Photon Analysis Delivery and Reduction System (PADReS), and a second one consisting of individual beamlines.
PADReS will be installed between the undulators and the beamlines (see figure 5). It has the dual purpose of delivering the radiation emitted by the two FELs to each experimental station and characterizing online the pulse-by-pulse radiation. With the analysis system one can determine the absolute intensity of each pulse (i.e. photons/pulse), the relative spatial position and angular tilt of the photon beam, and the pulse-resolved spectral distribution. Moreover, it will be possible to control the absolute intensity delivered to the beamlines via a 6 m-long gas attenuation chamber with a maximum attenuation factor of 10 4 . The information will be collected pulse by pulse, then made available for the user in real time and stored for future data reduction. A system of plane mirrors will deflect the radiation of both FELs (only one operates at a given time) to each of the three currently approved beamlines: DIPROI, dedicated to diffraction and projection imaging, LDM, dedicated to the study of the diluted system, and EIS, dedicated to elastic and inelastic scattering.  PADReS comprises the following components: a shutter, a beam-defining aperture, a beam position monitor, an intensity monitor and a differentially pumped gas absorption cell. After the gas absorption cell, the system is symmetrically repeated, with a second differential pumping system, a second intensity monitor, a second beam position monitor and then the first mirror. Inside the safety hutch, a system of three mirrors-two for FEL-1 and one for FEL-2-delivers the radiation to the online photon energy spectrometer. This analyzes, shot by shot, the energy spectrum of the emitted radiation. Starting from this point, the light can be deflected to the TIMER part of the EIS beamline (at whose beginning a system to measure the coherence of the beam will be installed) or can go straight to the LDM, DIPROI or TIMEX-EIS beamlines that share a number of components.
The TIMER-EIS beamline will work without a monochromator, while its use is optional on the TIMEX-EIS, LDM and DIPROI beamlines. The latter can work without a monochromator because the emitted photon number at the fundamental wavelength exceeds by a factor of 100 or more that emitted at higher harmonics. Consequently, the spectral purity of the nonmonochromatized radiation beam is good enough to perform most experiments without having to suffer the additional efficiency loss associated with a monochromator. However, there will be a problem in selecting individual higher harmonics for experiments that need higher photon energy. These include pump and probe experiments in which one pumps the system at the fundamental wavelength and then probes it with the third harmonic. Our solution, although useful for only a limited number of wavelengths, is to use multilayer reflective surfaces. The multilayers are optimized to work at one particular wavelength while simultaneously having the smallest possible efficiency at a wavelength three times larger. In this way, for instance, a Co/C multilayer has 60 times more efficiency at 6.66 nm than at 20 nm. After two mirrors, a reduction of more than 3000 is obtained; after four mirrors, a factor of 10 7 can be reached. Depending on the degree of spectral purity needed, one can use wavefront and time-preserving multilayer mirrors instead of a complicated and costly grating-based monochromator to select the proper harmonic.
The roll in, roll out monochromator for the TIMEX-EIS, LDM and DIPROI beamlines is based upon a constant included angle scheme. The details of the calculations made for its design are described elsewhere [25]. Here we wish just to mention the reasons for such a choice and a few details. The request from users was to have a relatively low resolution, to be able to select one harmonic and to suppress the background associated with undulator spontaneous emission and the beam dump bending magnet radiation. Meanwhile, one should try to maximize the fluence, reducing or completely avoiding (if possible) time elongation of the output pulses. The solution adopted is a fixed angle monochromator (174 • ) involving three interchangeable gratings that will cover wavelengths from 80 down to 1 nm with a single movement (a rotation). Longer wavelengths will be used in zero order (non-dispersive).
A removable collimating mirror before the gratings system and a focusing mirror just after the gratings will guarantee a stigmatic focus at a fixed exit slit. If these two mirrors are removed, the radiation beam travels unchanged to a subsequent delay line. The delay line will split the beam into two parts. One part can be delayed relative to the other by up to 30 ps by using grazing incidence optics. A larger delay requires the introduction of a multilayer. With this system, one can also perform two-color experiments, with or without delay. The delay can be controlled with a minimum step of 0.3 fs. To guarantee the stability of the direction of the beam, a closed-loop system (using a quadrant photodiode) is coupled to piezo correctors that act directly upon the mirrors that generate the time delay.
After the delay line, the beam can be directed to LDM, DIPROI or TIMEX-EIS beamlines. The TIMEX-EIS branch consists of plane deflecting optics and a fixed focus elliptical mirror. Conversely, the LDM and DIPROI will consist of two custom-made active mirrors mounted in a Kirkpatrick-Baez configuration. The use of active optics providing shape control is necessary to compensate for the movement of the exit slit as well as the difference in distance between the source positions of FEL-1 and FEL-2 (15 m or more). Another reason for these optics is the need to be able to change the spot dimension in the experimental chamber as desired. With these mirrors we will be able to go from a perfect unfocused beam (several mm across) down to a micron-sized spot. Moreover, these mirrors are also wavefront preserving optics. This is made possible by several actuators mounted at the back of the mirror surface. They not only preserve the wavefront and coherence of the incoming beam (by correcting the mid-frequency shape error of the mirror) but also compensate for deformation of the wavefront due to the previous optics.
With such mirrors, we expect to have a spot size in the experimental chamber of the order of 2 × 3 µm 2 , together with a very high fluence (i.e. power per unit area). Table 2 gives a fluence estimate for the DIPROI beamline, taking into account the geometrical acceptance, the mirror reflectivity and, of course, the spot dimension. It is clear that very high fluences, above 10 17 W cm −2 , are expected at the longest wavelengths, but even at shorter wavelengths (e.g. 5 nm) the fluence is higher than 10 16 W cm −2 .

Experimental stations
The results of the simulations presented above clearly show the superior performance of a seeded FEL in terms of the pulse temporal structure, coherence and photon polarization. In order to reach the full potential of FERMI@Elettra, we prepare for time-resolved experiments based on resonant coherent diffraction imaging, elastic and inelastic scattering, photon and electron spectroscopy and transient grating (TG) spectroscopy, which will monitor transient states and nonlinear material responses at mesoscopic and nano-scales, exploiting selection rules as well. The planned and under construction experimental stations, briefly described below, will give access to dynamic phenomena such as excitation lifetimes, phase separation and nucleation, ultrafast magnetization, complex rearrangements of constituents in cells, and multi-photon single and multiple ionization.

Coherent diffraction imaging
The extraordinary opportunity for single-shot coherent diffraction (lens-less) imaging (CDI) [26], which has become an excellent probe for the transient sample structure evolving after an excitation pulse from a pump laser [27] or using different delay schemes [28], will be fully exploited at FERMI. The photon-energy tunability of an FEL adds chemical imaging via resonant coherent diffraction at the atomic absorption edges and, when combined with the variable circular or linear polarization available from FERMI, extends the information to spin and orbital momentum. FERMI's DIPROI beamline and end-station is designed to meet the requirements for performing numerous types of static and dynamic coherent imaging. It includes a split-delay correlation system and focusing optics to image single small objects that also add the option for complementary projection imaging. The possible measurement modes are as follows: • Single-shot CDI for probing the specimen structure with diffraction-limited resolution.
• Single-shot dynamic CDI using a back reflective mirror for probing non-repetitive phenomena (e.g. radiation-induced damage) on fs scales.
• Time-resolved CDI for probing transient nanoscale dynamics on fs to ns time scales using split FEL pulses with adjustable delay or a short-pulse optical laser pump.
Stroboscopic imaging with fs time resolution can be used to explore ultrafast dynamics at nanometer length scales, such as fracture, phase fluctuations, motion in soft matter, changes in various forms of magnetic or electronic segregations, and copolymer assemblies. By splitting the pulse and simultaneously hitting the object from two directions, it is possible to explore stereo three-dimensional imaging as well.

Elastic and inelastic scattering
The elastic and inelastic scattering (EIS) experimental end-station will be dedicated to two different research projects: (i) TIME-Resolved spectroscopy of mesoscopic dynamics in condensed matter (TIMER) and (ii) ultrafast TIme-resolved studies of Matter under EXtreme and metastable conditions (TIMEX). For TIMER, the photon energy and brilliance of the FEL radiation will be used for TG formation in the sample with a nanometer scale spatial period [29]. This will be achieved by using two identical pulses impinging at the same time on the sample to create a standing wave that imposes a transient density modulation in the sample [30]. A third, delayed pulse at the third harmonic is then scattered by the TG. The scattering amplitude is related to the collective dynamics present in the system and can serve as a time-dependent monitor. The aim of TIMER is to experimentally access the mesoscopic range of dynamics that at present cannot be investigated by any alternative experimental technique. This ability would be extremely interesting because it could solve several open scientific problems regarding the physics of systems without translational invariance [31]. We stress the fact that the possibility of creating and probing TGs with spatial periods in the nanometer range will also be extremely useful for the study of surfaces and interfaces, with potential applications in the study of thin films and nanostructured materials [32].
TIMEX will exploit the unique intensity, energy domain and time structure of the FEL radiation to probe metastable and/or excited matter under extreme conditions. In particular, the energy and intensity of the FERMI FEL radiation beam are suitable for efficient, ultrafast heating of most bulk-like dense samples. The main idea is to use the FEL beam within a pump-probe scheme for time-resolved studies of the optical and soft x-ray properties of matter, providing direct information regarding surfaces and bulk under extreme conditions. Such experiments in the 0.1-10 ps range are relevant to a variety of physical and chemical phenomena, including high-pressure and high-temperature phase transitions, applied material studies, the understanding of chemical reaction and catalysis paths, planetary interiors, inertial confinement fusion, various forms of plasma production [33] in which energy is rapidly deposited into a solid, and non-equilibrium and metastable states of matter [34].

Low-density matter
The high brilliance of FELs is an ideal match for experiments involving low-density matter (LDM). The LDM beamline has been designed for experiments with supersonic (atomic, molecular and cluster) beams. Supersonic beams provide a relatively intense ensemble of noninteracting atoms and molecules, usually at a low and well-characterized temperature [35]. Experiments on atoms and molecules will explore nonlinear multiple-ionization processes, whose interpretation may call for the development of novel theoretical approaches. For molecules, in addition, energy redistribution processes subsequent to excitation, fragmentation in particular, can be studied from any well-defined initial state that can be prepared with the pump laser. The close-to-ideal pulse structure (time as well as energy profile) will greatly expand the range of experiments and will facilitate their interpretation as compared to a SASE source, such as FLASH [36]. Clusters will be studied both for providing complementary information to supported samples, such as those investigated in TIMEX and DIPROI experiments (note, in particular, the fact that TIMEX experiments imply isochoric heating, as opposed to isobaric in bare clusters) and for their intrinsic properties.
Of particular interest is that many materials that are non-magnetic in bulk form become magnetic in cluster form but the mechanisms are largely unknown [37]. Also, ultrafast changes in magnetization can be caused by slight changes to lattice parameters [38], usually induced with a femtosecond time scale laser pulse. The LDM beamline is ideal for performing this kind of experiment, especially because of its ability to take advantage of the full control of the FEL pulse polarization. The LDM beamline has a further unique feature, namely the ability to produce beams of ultracold superfluid He nanodroplets (T = 0.4 K). These will be used both to study superfluidity at the nanoscale and as a versatile substrate to assemble and cool heterogeneous molecular complexes that can then be studied via both one-color and pump-probe experiments (see, e.g., [39] for a review of experiments with conventional lasers).

Conclusions
In this paper, we have discussed various aspects of the FERMI@Elettra FEL design, including its expected output coherence properties, the radiation beam transport system and some possible experiments. Based upon the results of the numerical simulations reported here, the external seeded, harmonic upshift approach has clear advantages over a simple SASE design with regard to longitudinal coherence. The photon transport system was designed to take maximum advantage of the source characteristics, preserving polarization, wavefront and coherence, and with the aim of maximizing the fluence in the whole expected photon energy range. This was accomplished without precluding a possible upgrade to shorter wavelengths.
FERMI will begin user operation in 2011 and, given its expected high transverse and longitudinal coherence together with variable polarization, should open up the possibility of performing unique experiments at photon energies up to 900 eV. These will allow exploration of the structure and transient states of condensed, soft and low-density matter using a wide variety of diffraction, scattering and spectroscopy techniques and all the temporal correlation modes.