LETTERS
Femtosecond modification of electron
localization and transfer of angular
momentum in nickel
C. STAMM1 , T. KACHEL1 , N. PONTIUS1 , R. MITZNER1,2 , T. QUAST1 , K. HOLLDACK1 , S. KHAN1 *,
C. LUPULESCU1 †, E. F. AZIZ1 , M. WIETSTRUK1 , H. A. DÜRR1 ‡ AND W. EBERHARDT1
1
BESSY GmbH, Albert-Einstein-Str. 15, 12489 Berlin, Germany
Physikalisches Institut der Universität Münster, Wilhelm-Klemm-Str. 10, 48149 Münster, Germany
*Present address: Institut für Experimentalphysik, Universität Hamburg, Luruper Chausee 145, 22761 Hamburg, Germany
†
Present address: LASIM—Laboratoire de Spectrométrie Ionique et Moléculaire Bât A. Kastler, 43, bd du 11 Novembre 1918, 69622 Lyon, Villeurbanne, France
‡
e-mail: [email protected]
2
Published online: 26 August 2007; doi:10.1038/nmat1985
The rapidly increasing information density required of modern
magnetic data storage devices raises the question of the
fundamental limits in bit size and writing speed. At present,
the magnetization reversal of a bit can occur as quickly as
200 ps (ref. 1). A fundamental limit has been explored by
using intense magnetic-field pulses of 2 ps duration leading to
a non-deterministic magnetization reversal2 . For this process,
dissipation of spin angular momentum to other degrees of
freedom on an ultrafast timescale is crucial2 . An even faster
regime down to 100 fs or below might be reached by nonthermal control of magnetization with femtosecond laser
radiation3 . Here, we show that an efficient novel channel for
angular momentum dissipation to the lattice can be opened by
femtosecond laser excitation of a ferromagnet. For the first time,
the quenching of spin angular momentum and its transfer to
the lattice with a time constant of 120 ± 70 fs is determined
unambiguously with X-ray magnetic circular dichroism. We
report the first femtosecond time-resolved X-ray absorption
spectroscopy data over an entire absorption edge, which are
consistent with an unexpected increase in valence-electron
localization during the first 120 ± 50 fs, possibly providing the
driving force behind femtosecond spin–lattice relaxation.
When energy is pumped into electronic excitations of a metal
by absorbing a femtosecond optical laser pulse it takes time
to re-establish thermal equilibrium. This timescale is ultimately
determined by energy transfer from the electronic system to the
lattice4 . If for ferromagnetic metals, laser excitation should also
lead to an ultrafast quenching of the ferromagnetic order4,5 , angular
momentum conservation dictates that a transfer of spin angular
momentum to a reservoir such as the lattice has to occur5–9 .
However, there is considerable disagreement about the timescale
for such spin–lattice relaxation. It was established early that
spin–lattice relaxation should proceed on timescales of ∼100 ps
(refs 5,8). Such values are also obtained from the damping of
magnetization precession2,6 . There is growing evidence, although
no quantitative observation, that on the femtosecond timescale the
magnetic moment is affected by laser heating4,9–11 . Even on the
femtosecond timescale the total energy and angular momentum
are conserved. It is debated whether the reduction of the magnetic
moment, which corresponds mainly to spin angular momentum,
occurs via spin–orbit coupling during coherent laser excitation12 or
via a femtosecond spin–lattice relaxation mechanism6 .
Here, we address these issues by using circularly polarized
soft X-ray pulses of 100 ± 20 fs duration to determine the
temporal evolution of spin and orbital angular momentum
in ferromagnetic Ni after optical femtosecond laser excitation.
Using X-ray magnetic circular dichroism (XMCD), we show that
the spin angular momentum is quenched on a timescale of
120 ± 70 fs. We also show that electron orbits do not act as a
reservoir for angular momentum. This demonstrates the existence
of a novel femtosecond spin–lattice relaxation channel. Timeresolved X-ray absorption spectroscopy (XAS) measurements with
linearly polarized radiation indicate that the electronic system is
characterized by a significant valence-electron localization evolving
on a timescale of 120 ± 50 fs which is identical to the observed
ultrafast demagnetization. This novel process has been ignored
in existing models of ultrafast demagnetization6,12 and is likely to
be a key ingredient for other femtosecond laser-driven solid-state
dynamics processes13–15 . We argue that it could result in an efficient
way to transfer spin angular momentum to the lattice.
The laser pump–X-ray probe experimental set-up is shown
in Fig. 1. The same femtosecond laser is used to generate
femtosecond X-ray pulses and to excite the sample providing
inherent pump–probe synchronization, ultimately limited only
by the thermal stability of the optical components used. As a
unique feature compared with the femtoslicing source at the
Advanced Light Source in Berkeley16,17 , a helical undulator enables
complete polarization control of X-ray pulses18,19 . These are used
to stroboscopically probe the electronic and magnetic state by
means of XAS and XMCD, respectively. The X-ray spot size on
the sample (500 × 100 µm2 ) is smaller than the pump laser spot
size (1.5 × 0.5 mm2 ). The sample consists of a 15 nm Ni film
evaporated in situ under ultrahigh vacuum conditions onto a
500-nm-thick Al foil of 5 × 5 mm2 lateral size. This results in an
uncontaminated, homogeneous, polycrystalline film that can be
magnetically saturated in the film plane by applying a magnetic
740
nature materials VOL 6 OCTOBER 2007 www.nature.com/naturematerials
© 2007 Nature Publishing Group
LETTERS
Variable delay
a
fs laser pulse
fs X-ray pulse
Modulator
Radiator
0.3
XAS (arb. units)
Chopper
Transmission
sample
Detector
Dipole e–
magnets
H field
Figure 1 Schematic diagram of the pump–probe set-up. A femtosecond laser
(wavelength 780 nm, repetition rate 1 kHz, pulse energy ∼2 mJ) modulates the
stored electron bunches in the modulator which subsequently generate femtosecond
X-ray pulses in the radiator. The sample is excited by part of each laser pulse (15%)
via a variable delay. Transmitted X-rays are detected by an avalanche photodiode
behind the sample. The angle between the laser and X-ray beams is 1◦ . During
time-resolved measurements a mechanical chopper in the pump beam is used to
alternately measure the X-ray absorption of the laser-excited sample and the sample
in thermal equilibrium with a repetition rate of 500 Hz.
field of 0.02 T as checked with static XMCD measurements. The
Ni film thickness is matched to the optical laser extinction length.
This assures that the total Ni thickness, as probed with femtosecond
X-rays, is nearly homogeneously pumped by femtosecond optical
laser radiation.
Figure 2a shows Ni L3 edge XAS spectra taken with femtosecond
X-ray pulses of linear polarization at a 200 fs pump–probe
time delay and without laser excitation as red and black lines,
respectively. The difference between the two spectra is plotted
in Fig. 2b. X-ray absorption for the laser-excited film is clearly
increased at the leading absorption edge. It decreases just after
the XAS peak maximum. This corresponds to a ∼130 meV shift
of the XAS features. We will give an interpretation of this novel
effect below. First, we use it to determine the temporal response
of the electronic system to laser excitation. This was measured by
keeping the photon energy fixed at the value marked by arrow A
in Fig. 2a and is shown in Fig. 3a. A rapid increase in the XAS
absorption during the first 200 fs is succeeded by a slower signal
recovery for the subsequent 1 ps time interval. This behaviour
reflects the laser-induced electron population dynamics above the
Fermi level10 . We modelled the data in Fig. 3a by an approximation
to the three-temperature model (line) for energy transfer from the
laser-heated electronic system to the lattice (see the Supplementary
Information). The sharp initial XAS absorption rise is longer than
the X-ray pulse duration of 100 ± 20 fs determined by independent
pulse length diagnostics18,19 . The determined rise-time constant
of 120 ± 50 fs is very similar to the one determined with timeresolved photoemission spectroscopy, which probes the temporal
evolution of electron–hole excitations10 . This indicates that the rise
time is mainly given by the response of the electronic system to
femtosecond laser excitation. The subsequent decay on a ∼640 fs
timescale (see the Supplementary Information) characterizes the
electron–lattice energy relaxation in Ni (refs 4,10).
Figure 3b shows time-resolved XMCD measurements for the
15 nm Ni film with the photon energy fixed at the XAS peak
maximum (marked by arrow B in Fig. 2a) where a maximum
XMCD effect is observed in static measurements. At this photon
energy and with an X-ray bandwidth of 3 eV, the XMCD signal
essentially corresponds to the integral over the L3 absorption
edge. XMCD measures only the Ni magnetic properties and
clearly exhibits a step-like reduction and thus a different temporal
evolution than the electronic signal in Fig. 3a. The timescale for
the reduction of the magnetic Ni XMCD signal was determined
by approximating the three-temperature model of energy transfer
0.2
B
0.1
A
0
848
850
852
Photon energy (eV)
854
856
848
850
852
Photon energy (eV)
854
856
b
0.02
XAS change (arb. units)
fs laser
0
–0.02
Figure 2 Femtosecond X-ray absorption spectra. a, XAS spectra obtained with
linearly polarized femtosecond X-ray pulses at normal incidence. The absorption at
the L3 edge is shown for a 15-nm-thick Ni film 200 fs after (red line) and without
(black line) laser excitation. The arrows A and B indicate energy positions where
measurements in Fig. 3 were taken. b, Difference between the spectra in a.
between electron, spin and lattice reservoirs as described in the
Supplementary Information (line in Fig. 3b). It takes 120 ± 70 fs to
quench the ferromagnetic order, which is within the experimental
uncertainty identical to the electronic XAS response in Fig. 3a.
For the data in Fig. 3, care was taken to eliminate the influence
of optical path length variations between optical pump and X-ray
probe pulses during the data acquisition time. Whereas the delay
scan in Fig. 3a required a data acquisition time of ∼3 h, the data in
Fig. 3b were accumulated for a significantly longer period. Drifts of
zero time delay were checked at regular time intervals by measuring
reference delay scans such as the one shown in Fig. 3a although for a
more limited data range, that is, with shorter data acquisition times.
In Fig. 3, only data for which the optical pump–probe path length
changes were well below 100 fs were accumulated.
In 3d transition metals, sum rules relate the integral L3 XMCD
signal to a linear combination of spin, S, and orbital, L, angular
momentum components along the magnetization direction as
S + 3/2L (see the Supplementary Information). The temporal
evolution of S + 3/2L in Fig. 3b represents the first quantitative
demonstration that S is transferred to the lattice and not to L
on a 100 fs timescale. This can be visualized for the following
scenario, considering that in thermal equilibrium (at negative time
delays in Fig. 3) L is typically only about 20% of S (see the
Supplementary Information). The ∼70% decrease of S + 3/2L
during the first picosecond is then mainly due to the reduction
of S. If a 70% reduction in S would be completely compensated
nature materials VOL 6 OCTOBER 2007 www.nature.com/naturematerials
© 2007 Nature Publishing Group
741
LETTERS
XAS change (arb. units)
a
EF
3d/4sp
0.02
0
hν
–0.5
0
0.5
1.0
2p
Time delay (ps)
b
Figure 4 Schematic diagram of electronic structure changes. Schematic
diagram of the electronic structure of the Ni 2p core levels and 3d/4sp valence
band for different valence band widths. Charge neutrality dictates that with shrinking
band width the centre of the valence band moves closer to the Fermi level, E F . This
reduces the excitation energy, hν, at the 2p XAS threshold.
Normalized XMCD
1.0
0.5
0
–0.5
0
0.5
Time delay (ps)
1.0
Figure 3 Femtosecond evolution of Ni electronic and magnetic structure.
a, Time-resolved change of the XAS intensity with linearly polarized X-rays
incident perpendicular to the sample surface versus pump–probe time delay
(symbols) measured at a photon energy corresponding to the leading L3 edge slope
(arrow A in Fig. 2). The photon energy resolution was 1.5 eV. b, Time-resolved XMCD
signal with circularly polarized X-rays incident at 60◦ relative to the sample surface
versus pump–probe time delay (symbols) measured at the L3 edge maximum
(arrow B in Fig. 2). The photon energy resolution was 3 eV. Lines are fits of the
three-temperature model to the data. The pump laser fluence was 8 mJ cm−2 . All
XAS and XMCD data shown in this figure were taken from a 15-nm-thick Ni film and
are normalized to the corresponding data taken without laser pump pulses. The error
bars denote the standard deviation of the Poisson-distributed single-photon events.
by L, the quantity S + 3/2L would actually increase by about
27% in contradiction to the measurements. At present we cannot
rule out a partial spin angular momentum transfer to L, but its
contribution would have to be below ∼10% of S to keep S + 3/2L
within the statistical uncertainty in Fig. 3b. This excludes L as
a reservoir for S. We can also exclude a significant transfer of
spin angular momentum from Ni 3d electrons to more itinerant
s- or p-like electrons on the basis of the following argument.
In Ni there is typically one sp-like valence electron per atom
(compared with about 9 3d electrons). The Ni 3d hole is
almost completely spin polarized. The sp-like electrons would,
therefore, also have to become completely spin polarized to act
as a spin angular momentum reservoir. Considering the typical
width of ∼8 eV for the occupied sp-valence bands and the Ni
sp-exchange splitting of ∼100 meV, this would require either a
significant rearrangement of energy levels or electron repopulation.
Neither effect has been observed, for example, with time-resolved
photoemission spectroscopy10 . We can conclude that the data in
Fig. 3b are only compatible with a substantial femtosecond spin
angular momentum transfer to the lattice, that is, a femtosecond
spin–lattice relaxation.
Our present understanding of the microscopic origin for the
XAS energy shift observed in Fig. 2a is shown schematically in
Fig. 4. A shrinking valence band width would shift the centre of the
valence band towards the Fermi level to retain charge neutrality.
This scenario is commonly encountered for energy-shifted surface
core levels in solids20 . It would result in a lower XAS threshold
energy as indicated in Fig. 4. For quantifying this picture, we used
a cluster model where the Ni initial state without laser excitation
is given by a coherent superposition of 35% 3d 10 , 51% 3d 9 and
14% 3d 8 valence electronic configurations (see the Supplementary
Information). The experimentally observed ∼130 meV XAS peak
shift (Fig. 2a) is reproduced for a ∼9% reduction of hybridization
between Ni 3d and ligand 4sp orbitals. This reflects increased
valence-electron localization and a decoupling of 3d and 4sp
electronic states. We interpret this as being mainly due to scattering
of itinerant electrons by laser-induced electron–hole excitations.
The temporal evolution of the XAS line shape observed here,
therefore, follows that of laser-induced valence-electron dynamics
observed above the Fermi level10 . Such electron–hole scattering
is very effective and happens essentially on timescales of <10 fs
(ref. 10). It would result in a partial loss of coherence for the manyelectron wavefunctions. This is mimicked in our model by a more
localized electronic structure.
These results have significant impact on the microscopic
understanding of femtosecond spin–lattice relaxation. So far,
femtosecond laser-induced changes in the electronic structure
(Fig. 4) have not been considered in models of femtosecond
demagnetization6,12 . Koopmans et al.6 predicted spin–lattice
relaxation on the basis of a model of impurity scattering of
spin-polarized electrons. The latter should be influenced by
electron localization. Spin–orbit coupling together with coherent
electronic excitations was invoked as a source of femtosecond
demagnetization12 , although neglecting spin–lattice relaxation.
The similar timescales observed for the femtosecond spin–lattice
relaxation and valence-electron localization observed here indicate
the relation of the two phenomena. Increased electron localization
could result in an enhanced spin–orbit coupling required for strong
coupling of spins and lattice.
742
nature materials VOL 6 OCTOBER 2007 www.nature.com/naturematerials
© 2007 Nature Publishing Group
LETTERS
In summary, we have demonstrated that the absorption of an
intense femtosecond laser pulse generates ultrafast changes in the
electronic and spin structure of metallic Ni. Following femtosecond
laser excitation we observed an increase in valence-electron
localization on a timescale of 120 ± 50 fs. This mechanism is
possibly related to the driving force for a spin–lattice relaxation.
The latter process was unambiguously established to proceed on
a 120 ± 70 fs timescale by probing the evolution of spin and
orbital angular momentum with polarized femtosecond soft X-ray
pulses. These measurements represent the first separation of spin
and orbital angular momentum on the femtosecond timescale
providing element sensitivity. They demonstrate an important
new application of the emerging femtosecond X-ray sources in
studying ultrafast dynamic processes in magnetic materials and
open the door for studies of more complex materials using
polarized X-rays21 .
Received 4 June 2007; accepted 20 July 2007; published 26 August 2007.
References
1. Gerrits, Th., van den Berg, H. A. M., Hohlfeld, J., Bär, L. & Rasing, Th. Ultrafast precessional
magnetization reversal by picosecond magnetic field pulse shaping. Nature 418, 509–512 (2002).
2. Tudosa, I. et al. The ultimate speed of magnetic switching in granular recording media. Nature 428,
831–833 (2004).
3. Kimel, A. V. et al. Ultrafast non-thermal control of magnetization by instantaneous photomagnetic
pulses. Nature 435, 655–657 (2005).
4. Beaurepaire, E., Merle, J.-C., Daunoise, A. & Bigot, J.-Y. Ultrafast spin dynamics in ferromagnetic
nickel. Phys. Rev. Lett. 76, 4250–4253 (1996).
5. Vaterlaus, A., Beutler, T., Guarisco, D., Lutz, M. & Meier, F. Spin-lattice relaxation in ferromagnets
studied by time-resolved spin-polarized photoemission. Phys. Rev. B 46, 5280–5286 (1992).
6. Koopmans, B., Ruigrok, J. J. M., Longa, F. D. & de Jonge, W. J. M. Unifying ultrafast magnetization
dynamics. Phys. Rev. Lett. 95, 267207 (2005).
7. Einstein, A. & de Haas, W. J. Experimenteller Nachweis der Ampèreschen Molekülströme. Verhandl.
Deut. Phys. Ges. 17, 152–170 (1915).
8. Hübner, W. & Bennemann, K. H. Simple theory for spin-lattice relaxation in metallic rare-earth
ferromagnets. Phys. Rev. B 53, 3422–3427 (1996).
9. Scholl, A., Baumgarten, L., Jacquemin, R. & Eberhardt, W. Ultrafast spin dynamics of ferromagnetic
thin films observed by fs spin-resolved two-photon photoemission. Phys. Rev. Lett. 79,
5146–5149 (1997).
10. Rhie, H.-S., Dürr, H. A. & Eberhardt, W. Femtosecond electron and spin dynamics in Ni/W(110)
films. Phys. Rev. Lett. 90, 247201 (2003).
11. Koopmans, B., van Kampen, M., Kohlhepp, J. T. & de Jonge, W. J. M. Ultrafast magneto-optics in
nickel: Magnetism or optics? Phys. Rev. Lett. 85, 844–847 (2000).
12. Zhang, G. P. & Hübner, W. Laser induced ultrafast demagnetization in ferromagnetic metals. Phys.
Rev. Lett. 85, 3025–3028 (2000).
13. Bonn, M. et al. Phonon-versus electron-mediated desorption and oxidation of CO on Ru(0001).
Science 285, 1042–1045 (1999).
14. Sokolowski-Tinten, K. et al. Femtosecond X-ray measurement of coherent lattice vibrations near the
Lindemann stability limit. Nature 422, 287–289 (2003).
15. Melnikov, A. et al. Coherent optical phonons and parametrically coupled magnons induced by
femtosecond laser excitation of the Gd(0001) surface. Phys. Rev. Lett. 91, 227403 (2003).
16. Schoenlein, R. W. et al. Generation of femtosecond pulses of synchrotron radiation. Science 287,
2237–2240 (2000).
17. Cavalleri, A. et al. Band-selective measurements of electron dynamics in VO2 using femtosecond
near-edge X-ray absorption. Phys. Rev. Lett. 95, 067405 (2005).
18. Holldack, K., Kachel, T., Khan, S., Mitzner, R. & Quast, T. Characterization of laser-electron
interaction at the BESSY II femtoslicing source. Phys. Rev. ST Accel. Beams 8, 040704 (2005).
19. Holldack, K., Khan, S., Mitzner, R. & Quast, T. Femtosecond terahertz radiation from femtoslicing at
BESSY. Phys. Rev. Lett. 96, 054801 (2006).
20. Johansson, B. & Mårtensson, N. Core-level binding-energy shifts for the metallic elements. Phys.
Rev. B 21, 4427–4457 (1980).
21. Huang, D. J. et al. Orbital ordering in La0.5 Sr1.5 MnO4 studied by soft X-ray linear dichroism. Phys.
Rev. Lett. 92, 087202 (2004).
Acknowledgements
We thank F.M.F. de Groot for valuable and stimulating discussions. We are indebted to the BESSY staff
for the enthusiastic help and support during the construction and commissioning of the femtoslicing
facility. Work is supported by the Bundesministerium für Bildung, Wissenschaft, Forschung and
Technologie, by the Land Berlin and by the European Union.
Correspondence and requests for materials should be addressed to H.A.D.
Supplementary Information accompanies this paper on www.nature.com/naturematerials.
Competing financial interests
The authors declare no competing financial interests.
Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/
nature materials VOL 6 OCTOBER 2007 www.nature.com/naturematerials
© 2007 Nature Publishing Group
743