32 SUMMER 2002

REATING MATTER from the vacuum,taking an atomic scale motion picture ofa chemical process in a time of a fewfemtoseconds (1 fs = 10-15 sec) or unravelingthe complex molecular structure of a singleprotein or virus. These are some of the newexciting experiments envisioned with a novel

radiation source, the X-ray free-electron laser (FEL). John Madey and collaborators built the first FEL in the

1970s. It is a powerful and challenging combination of parti-cle accelerator and laser physics and technology. Untilrecently FELs have been operating at infrared or near ultravi-olet wavelengths. A combination of theoretical, experimen-tal, and technological advances has made possible theirextension to the X-ray region. X rays have allowed us to seethe invisible for almost a century. With their help we havebeen making great progress in understanding the propertiesof materials and of living systems. Today the best sources ofX rays utilize synchrotron radiation from relativistic electronbeams in storage rings. The most advanced, so called thirdgeneration facilities, combine storage rings and undulatormagnets, and are used by thousands of scientists around theworld.

USEFUL AS THEY ARE these facilities havelimitations. The minimum X-ray pulse duration isabout 50 picoseconds (1 ps = 10-12 sec), and the num-

ber of photons that one can focus on a small sample islimited. Future experiments will need X-ray pulses with a

With advances

in technology,

it is now possible

to realize the

dream of a fully

coherent X-ray

laser.

X-ray Free-Electron Lasers –

by CLAUDIO PELLEGRINI & JOACHIM STÖHR

C

BEAM LINE 33

large number of photons focused on a sample as small as amolecule, squeezed in a time a thousand times shorter, tostudy the dynamics of atomic and molecular processes. X-rayfree-electron lasers can satisfy these requirements, as shownin the table.

AN FEL CONSISTS OF A LINEAR acceleratorfollowed by an undulator magnet, as shown in thefigure. The undulator has a sinusoidal magnetic field

of period λw, usually a few centimeters, and amplitude Bw,typically about 1 tesla. In this field a single electron movesalong a sinusoidal, oscillating trajectory, and radiates an elec-tromagnetic wave-train with a number of waves equal to thenumber of undulator periods, Nw. The wavelength of thisspontaneous radiation is equal to the undulator periodreduced by a relativistic contraction factor inversely propor-tional to the square of its energy. This makes it easy toshorten the wavelength by increasing the electron beamenergy, and for GeV electron beams one can produce

– Principles, Properties and Applications

ElectronAccelerator

ElectronBeam

E-MRadiation

UndulatorMagnet

Schematic representation of a SASE free-electron laser.

34 SUMMER 2002

wavelengths of about 1 Å, the size ofan atom. The radiation is within anarrow wavelength band, equal tothe inverse of the number of undu-lator periods, which for an X-ray FELcan be a few thousand. The radiationwithin this energy band is very wellcollimated, with an opening angle ofa few microradians. The number ofphotons spontaneously emitted fromone electron within this energy bandand angle is however rather low,about one photon per 100 electrons.In an FEL the electron beam has alarge number of electrons, about 109

to 1010. In this case a phenomenon ofself-organization of the electrons,known as the FEL collective insta-bility, can greatly increase the num-ber of photons per electron and thetotal X-ray intensity. This instabil-ity transforms an electron beam witha random electron position distrib-ution into one in which the electronsare clustered in groups regularly

spaced at about 1 Å, producing whatcould be called a 1-dimensional rel-ativistic electron crystal. The radi-ation from this crystal has new andexciting properties.

Consider an ensemble of elec-trons propagating through an undu-lator. The total radiation field theygenerate is the sum of the fields gen-erated by all electrons. When theelectrons enter the undulator, thereis no correlation between theirposition and the X-ray radiationwavelength. As a result the fieldsthey generate superimpose at ran-dom, with a partial cancellation, asshown at the left in the next figure.What the beam produces in this caseis called “spontaneous undulatorradiation”, and its intensity is pro-portional to the number of electrons,Ne.

If we could order the electrons sothat they were all within a fractionof the radiation wavelength, or

Third Generation SASE-Free- Short pulse SASE-Electron Free-Electron

Laser Laser

Wavelength range, nm 1-0.1 1-0.1 1-0.1

Emittance, nm-rad 2 0.03 0.03

Pulse length, ps 15-30 0.06 0.01

Average brightness 1020 1022 1021

Peak brightness 1023 1033 1033

Peak power, W 103 1010 1010

Some typical characteristics of the undulator radiation from third-generation ring-based lightsources, and free-electron laser light sources. The emittance is in nm rad; the pulse length in picoseconds; the average and peak brightness are in photons/sec/mm2/mrad2/0.1% bandwidth;the peak power in watts.

Undulator Radiation Characteristics

BEAM LINE 35

separated by a wavelength, the radi-ation fields would superimpose inphase, as shown at the right in thefigure. The intensity is then propor-tional to the number of electronssquared, a huge gain. The effect issimilar to what you would have dur-ing a large party if instead of hav-ing all the people talking casually,producing a lot of noise, you couldorganize them in a well performingchoir, all persons singing exactly thesame note in perfect tune. Theresults would probably break all theglasses and windows in the room!

While we do not know how tocontrol the electron position at theÅngstrom level, we can convenientlylet nature order the electrons for us,using the FEL collective instability.The way it works is: o Electrons propagating through the

undulator interact with the elec-tromagnetic field generated byother electrons. The interactionchanges their energy, and thechange is modulated at the X-raywavelength.

o Within the undulator the trajec-tory of electrons with larger(smaller) energy is bent less(more). As a result the electronswill bunch together within awavelength.

o The electromagnetic fields emit-ted by the bunched electronssuperimpose in phase, and thetotal field amplitude increases.Thus the electron energy changeis larger, and the bunching mech-anism is stronger.The end result is that the ampli-

tude of the electromagnetic fieldgrows exponentially. The growthrate is called the gain length and isthe most important FEL parameter.

The exponential growth saturateswhen all the electrons are wellordered, and sing perfectly in tune!

The instability can start with theaid of an external electromagneticfield, in which case the system is ahigh gain FEL amplifier, or it canstart from the random synchrotronradiation noise produced by theelectron beam at the undulatorentrance, a Self Amplified Sponta-neous Emission FEL (SASE-FEL). Aone-dimensional theory of a SASE-FEL, describing all its propertieswith a single parameter ρ, wasdeveloped in 1984 by RoldolfoBonifacio, Claudio Pellegrini andLorenzo Narducci.

In the X-ray region ρ is about10-3 or less. This means that satu-ration is reached in about 1000undulator periods, and that onethousandth of the electron beamkinetic energy is transformed intophotons. Consider as an example a1 Å X-ray SASE-FEL with 15 GeV elec-tron energy, and ρ~10-3. In this casethe number of photons per electronwithin the same frequency band andangle is about 1000, a gain of 100,000over the spontaneous radiation case.

The instability develops if theelectron beam has an energy spreadsmaller than ρ, and a radius andangular spread similar to that of the

The superposition of the fields generatedfrom many electrons; on the left is thespontaneous radiation case, on the rightthe free-electron laser case.

36 SUMMER 2002

X-ray beam. The gain must also belarge enough to overcome radiationlosses due to diffraction. This meansthat as the FEL wavelength is reducedthe electron beam must satisfy morestringent , and this requires avoid-ing damaging collective effects in theaccelerator. Operating an X-ray SASE-FEL is a balancing act between con-trolling the unwanted collectiveeffects in the accelerator, and lettingthe FEL instability develop in theundulator.

IN A WORKSHOP held atSLAC in 1992 to discuss the pos-sibility of developing advanced,

fourth generation X-ray sources,Claudio Pellegrini showed thatusing a novel electron source, thephotoinjector, developed by JohnFraser, Richard Sheffield, and EdwardGray at Los Alamos, and the newlinac technologies developed for theSLAC linear collider, it was possibleto build a 1Å X-ray SASE-FEL. A studygroup coordinated by HermanWinick developed this concept, call-ing it the Linac Coherent LightSource (LCLS). A design groupdirected by Max Cornacchia contin-ued the project development until itsapproval by the Department ofEnergy. A group at DESY in Germanyalso became interested in this workand developed its own SASE-FEL pro-ject as part of the TESLA linear col-lider project.

The first experimental observa-tions of the FEL instability were donein 1998 by a UCLA-Kurchatov group,which demonstrated exponentialgain over about 4 gain lengths, in a60 cm long undulator, at a wave-length of 16 mm. A much larger gain,3×105 at 12 µm, was obtained in the

same year, using a 2 meter long un-dulator by a UCLA-Kurchatov-LANL-SSRL group. In 2000-2001 three SASE-FELs have reached a gain largerthan 107 and saturation: LEUTL, atArgonne, at 530 and 320 nm in a 20meter long undulator; VISA, a BNL-SLAC-LLNL-UCLA collaboration, ina 4 meter long undulator, using abeam with characteristics similarto those required for LCLS, as shownin the figure; and TTF, at DESY, at 92nm, the shortest wavelength for anFEL, with a 15 meter undulator.

These results support the feasi-bility of LCLS and the TESLA X-rayFEL project. The main differencebetween these projects is the linearaccelerator, an existing room tem-perature linac for LCLS at SLAC, anda future superconducting linac forTESLA. The LCLS project is scheduledto start operation in 2008. Its char-acteristics are given in the table.

WHEN THE LCLS wasfirst discussed, an over-riding concern was the

destructive power of its X-raybeam, an extraordinary beam witha peak power of about 10 GW.Wouldn’t such a beam simply destroyeverything put in its path, fromX-ray optics to the sample itself?Traditionally X rays have been usedas a gentle probe of matter, avoid-ing radiation damage caused bythe beam itself. While some of theexcitement surrounding LCLS isbased on utilization of the highpower to create new forms of matter,there are fortunately also ways touse its X rays as a gentle probe.

X rays, like visible light, are elec-tromagnetic waves. In the case oflight the wave oscillation period is

Electron BeamElectron energy, GeV 14.3Emittance, nm rad 0.05Peak current, kA 3.4Energy spread, % 0.02Pulse duration, fs 230

UndulatorPeriod, cm 3Field, T 1.32K 3.7Gap, mm 6Total length, m 100

RadiationWavelength, nm 0.15

FEL parameter, ρ 5x10-4

Field gain length, m 11.7Bunches/sec 120

Average Brightness 4x1022

Peak brightness 1033

Peak power, GW 1010

Intensity fluctuations, % 8

LCLS Parameters

ENERGY SPREAD, PULSE LENGTH,emittance are rms values. Bright-

ness is in the same unit as the earliertable. The energy spread is the localenergy spread within 2πcooperationlengths. A correlated energy chirp of0.1 percent is also present along thebunch.

105

103

101

10–1

1 2 3z (m)

Gain Length18.7 cm

Onset ofSaturationat 3.6 m

En

ergy

(n

J)

4

Exponential growth and saturation in theVISA experiment at 830 nm, in a 4 mlong undulator.

BEAM LINE 37

about 1 fs, while for X rays it is 1000times faster. The LCLS beam has apeak power of 10 GW in a 100×100µm2 area near the undulator exit,corresponding to peak electric andmagnetic fields of 2 volts/Ång-stromand 60 tesla, respectively. Focusingthe beam by state-of-the-art meth-ods to 1×1µm2 the field strengthgrows to 200 volts/Ångstrom and6,000 tesla, strong enough to pullelectrons out of atomic shells. Forthe focused LCLS beam the magneticfields rival the exchange fields in fer-romagnets, and would affect all mag-netic interactions in matter. Thesenumbers appear to confirm our worstfears–that nothing would survive theLCLS beam.

Fortunately, the very rapid vari-ation of the fields with time savesthe day. Just as humans cannot hearvery high frequencies because oureardrums cannot follow the rapidmovement of the sound waves, mat-ter becomes insensitive to high fre-quency fields because the electroncharge or spin cannot follow them.While the most intense visible laserscan ionize atoms by the shearstrength of the electric field, con-verting a material into a plasma, thethousand times faster oscillatingfields of X rays only cause the elec-trons and spins to quiver, as illus-trated in the figure.

In addition, one can minimize theinteraction of the X rays with the

Laser (~4 eV) X-Ray FEL (~4 keV)

1 fs (femtosecond, 10–15s)

E

1 as (attosecond, 10–18s)

e– e–

H

Distortion of the electron orbit and spin direction in an atom by strong oscillatingelectric and magnetic fields for typical laser and X-ray frequencies. For lasers, thefield frequency is comparable to the Bohr orbit frequency of valence electrons, andthe spins can precess back and forth by a large angle around the strong magneticfield. For X rays the electrons and spins cannot follow the fast fields and just quiver.

38 SUMMER 2002

mirrors and other optical elements,due mostly to the photoelectriceffect, by choosing materials witha small atomic number to buildthem.

Three general classes of experi-ments have been proposed for LCLS.In the first the X-ray beam is used toprobe the sample without modify-ing it, as is done in most experimentstoday. In the second class, X rays areused to induce non-linear photo-processes or produce states of mat-ter with extremely large temperatureand pressure. In the third class theultra-fast nature of the LCLS pulse isused to obtain a snapshot of thestructure of the sample before theeffects of radiation damage set in.

The LCLS can be used for all threetypes of experiments. The unfocusedX-ray beam size at the exit of theundulator is about 100×100 µm2, thediameter of a fine needle. At a typ-ical experimental station locatedabout 200 meters from the undula-tor exit, the size expands to only200×200 µm2. This beam can befocused by state-of-the-art tech-niques to about 0.1×0.1 µm2 withmaybe a factor of 10 loss in power.This beam size is about the size of

a virus or the smallest structures thatcan be lithographically fabricatedtoday on an electronic chip. Throughvariable focusing the flux density ofthe X-ray beam can therefore betuned by a staggering factor of aboutone million.

THE X-RAY PHOTONflux typically delivered to asample in today’s synchro-

tron radiation experiments is about1012-1013 photons/sec. It typicallytakes one second to record a dif-fraction pattern of a sample. Fromthe diffraction pattern we can thenconstruct a real space image ofthe atomic positions in the sam-ple. Because of the time duration ofthe measurement only the averageposition of the vibrating atoms canbe observed. LCLS can give the samenumber of X rays obtained now in 1second in an ultra-fast pulse of about100 femtoseconds.

This time is hard to imagine. It isso short that sound would travel onlythe distance of one atom, and thefastest “thing” we know, light, wouldtravel only about 50 µm or the widthof human hair. It is also the timewith which atoms, the building

Snapshots with atomic resolution and femtosecond time intervals of a moleculardissociation will be made possible by X-ray FELs. The real space images wouldbe reconstructed from ultrashort X-ray diffraction patterns.

blocks of materials, vibrate. HenceLCLS will allow us not only to “see”the invisible, down to the size of anatom, but also to take snapshots ofthe motion of atoms and of ultra-small, so called nanoscale, objectslike molecules and clusters. This isillustrated in the figure, showingschematically how a small moleculeseparates from a bigger molecule, achemical process called dissociation.Today we can only imagine such amovie, but with LCLS we will be ableto record it and then play it back tolook at the detailed process.

The spontaneous emission ofradiation from X-ray tubes or syn-chrotron radiation sources ischaotic or incoherent by nature. Atfirst sight it is surprising that it per-mits us to observe X-ray diffractionphenomena, originating from inter-ference. Under certain conditions,incoherent sources can be a source

BEAM LINE 39

Left: Magnetic worm domain pattern of CoPt alloy with perpendicularanisotropy, recorded by transmission X-ray microscopy. Right: CoherentX-ray diffraction or speckle pattern from a 5 µm diameter region of the samesample, selected by a 5 µm circular aperture. The central rings are theFraunhofer diffraction pattern from the circular aperture which are wellseparated from the speckle pattern of the smaller magnetic domains.

of coherent X rays. It all depends onthe experimental geometry and onthe dimensions of interest in thesample! If the sample is far from thesource, the X-ray phase and ampli-tude is well defined over a certainsmall sample volume, and one canget interference from structureswithin that volume. Examples areBragg peaks or small-angle scatter-ing peaks, which reflect “structure”within a coherence volume of a fewhundred Ångstroms. To extend thestructural sensitivity to larger, saymicrometer, dimensions the X-raycoherence volume must be increasedbeyond these dimensions. If this isdone by moving the sample furtherfrom the source, the cost is a reducedintensity.

Only with the advent of highbrightness third generation synchro-tron sources have such experimentsbecome practical. An example is the

40 SUMMER 2002

coherent diffraction or “speckle” pat-tern shown on the previous page. Thepattern arises from the magneticworm domains in a CoPt alloy,shown on the left. The speckleimage on the right was recorded intransmission through a 5 µm diame-ter aperture, which defines the lat-eral X-ray coherence. The specklepattern contains the detailed infor-mation on the true structure of thesample in the illuminated area. Anychange of the magnetic structure ofthe sample would be reflected by anintensity change in the speckle pat-tern. With LCLS a speckle pattern canbe recorded in a single shot, openingthe door for femtosecond dynamicson the nanoscale. Inversion of themagnetic speckle pattern, using tech-niques such as oversampling, alsopromises to give ultrafast real spaceimages of nanostructures. Besidesmagnetic thin films, systems ofinterest include various materialsundergoing phase transitions, simpleand complex fluids or glasses, andcorrelated materials with complexcharge and spin ordering dynamics.

On a typical undulator beam lineon a third generation X-ray facility,equipped with a monochromator of0.01 percent bandpass, the X-raybeam is coherent within a microm-eter size volume, as shown in thenext figure. The number of photonsin the coherence volume, thedegeneracy parameter, scales linear-ly with the source brightness andwith the third power of the wave-length and is on average less thanone. Therefore all present day dif-fraction experiments are based onsingle photon interference effects.

In contrast, the LCLS radiationhas unprecedented coherence,

about 109 photons in the coherencevolume. The energy of coherent pho-tons can be pooled to create multi-photon excitations and carry out non-linear X-ray experiments. This is alargely unexplored area of science.

In its initial configuration theLCLS pulse duration is about 200 fem-toseconds. Even though this is hun-dred times shorter than in storagerings, it can probably be furtherreduced to about 1/10 of this value.Several schemes to achieve ultrashortpulses have been proposed. They alluse an energy-position correlation inthe electron bunch, and, as a result,in the X-ray pulse. A short X-ray pulseis then obtained by slicing out partof the bunch with a monochromator.A promising scheme using thisapproach and two undulators hasbeen studied, and could be imple-mented in LCLS.

THE EXPERIMENTS withLCLS and TESLA will cover awide range of fields, from

high-energy physics, to atomicphysics, plasma physics, chemistry,materials science, condensed matterphysics, and biology. Here we canonly mention some of the presentideas and explain why they areexpected to break truly new ground.

The very first LCLS experimentwill be aimed at understanding thefundamental interaction process ofthe high power X-ray beam withatoms, molecules and clusters. It willexplore the formation of hollowatoms, where the X rays can stripelectrons from the inside out. It willalso study multiphoton processesenabled by the large coherent inten-sity. Finally, it will investigate thedisintegration and explosion of clus-

3rd GenerationBeam Line

LCLS Source

Coherence Volume1x5x50 µm3

Coherence Volume0.1x100x100 µm3

Contains < 1 Photon Contains 109 Photons

Comparison of the coherence properties of astate-of-the-art third generation beam line withthat of an XFEL.

BEAM LINE 41

ters, yielding information on thetime scale of the damage caused bythe X-ray beam, a fundamental ques-tion for other experiments.

A second proposed experimentuses LCLS to create and investigatewarm (WDM) and hot (HDM) densestates of matter that exist in astro-nomical objects and are importantfor inertial fusion. Conventionallasers have provided limited infor-mation on these systems, becausethey cannot penetrate the high-density matter, and few theories canmake any prediction.

The study of molecular reactionsis the very heart of chemistry.With ultrafast laser spectroscopywe can catch a glimpse of electrontransfer processes in response toan optical excitation pulse. Butdespite their great success, conven-tional lasers can only study elec-tronic excitations, and cannot“see” the positions of the atomsduring the various transformationstages. This is, of course, the domainof X rays, and LCLS will permit usto take diffraction snapshots asshown in the earlier figure. Theimportance of these experimentsis underscored by the fact thatmany important discoveries inbiology and chemistry can be tracedback to the determination of a struc-ture.

Another experiment, similar tothat shown in the “speckle” figure,pushes the envelope in probingordering phenomena in hard and softcondensed matter on the importantnanometer length scale, whichcannot be seen by optical photons,over a broad range of time scales.This area is not only scientifi-cally interesting, but it constitutes

the competitive arena of future tech-nological devices.

LCLS also holds great promise forstructural biology. Today, radiationdamage is one of the main obstaclesin determining the structure of pro-teins that cannot be crystallized, likecell membrane proteins, which con-stitute nearly half of all proteins.With the peak brightness of LCLS thestructure of a virus or even a singleprotein molecule may be determinedby recording a three dimensionalarray of ultrafast diffraction patterns,each recorded in a single shot on anew sample before radiation damagesets in.

Finally, just as the application ofconventional lasers has been accom-panied and driven by the R&D onlasers themselves, LCLS will alsohave an R&D program, exploringnew ways to improve its character-istics.

THE HISTORY of andexperience with three gen-erations of synchrotron

radiation sources has taught us thatthe above experiments are at best thetip of the iceberg of scientificopportunities. It is safe to predictthat we have not yet thought of themost important experiments thateventually will be done with thisnew class of radiation sources–X-rayfree electron lasers! Quoting Antoinede Saint-Exupery (The Wisdom ofthe Sands), “The most importanttask before us may be . . . not to fore-see it, but to enable it.”