- Email: [email protected]
Sample preparation methods for femtosecond electron diffraction experiments
Ultramicroscopy 127 (2013) 9–13
Contents lists available at SciVerse ScienceDirect
Ultramicroscopy journal homepage: www.elsevier.com/locate/ultramic
Sample preparation methods for femtosecond electron diffraction experiments Maximilian Eichberger a, Marina Krumova b, Helmuth Berger d, Jure Demsar a,c,n a
Physics Department and Center of Applied Photonics and Zukunftskolleg, University of Konstanz, D-78464, Germany Department of Chemistry, University of Konstanz, D-78464, Germany c Complex Matter Department, Jozef Stefan Institute, SI-1000 Ljubljana, Slovenia d Physics Department, EPFL CH-1015 Lausanne, Switzerland b
a r t i c l e i n f o
abstract
Available online 2 August 2012
Exploring the dynamics of Charge density wave system 1T-TaS2 via femtosecond electron diffraction demonstrated the power of this technique for studying ultrafast structural phenomena in strongly correlated electron materials [1]. The results revealed first direct information on the order parameter dynamics of Charge Density Waves as well as on their photo-induced phase transition. A prerequisite to perform such experiments on modern quantum materials is the availability of laterally large ( 100 mm) and sufficiently thin ( o 100 nm) single crystalline samples. Different approaches to reach these specifications have been tried out and their effect on sample integrity has been investigated. Finally, using an ultra-microtome, we were able to prepare 30 nm free standing single crystalline films of 1T-TaS2 with lateral dimensions of 200 mm 200 mm. We have characterized these films with different techniques for their stoichiometric and crystalline integrity, ensuring no measurable alternation of sample properties. The application of this sample thinning technique is expected to find its use in further structural dynamics studies, as well as in optical time-resolved studies where homogeneous excitation profile and/or data in transmission geometry may be required. & 2012 Elsevier B.V. All rights reserved.
1. Introduction In microscopy a growing effort is currently being put into extending the field of interest to the temporal dimension of objects under study [2]. In solid state physics, where one is interested in the motion of electrons, ions and their mutual interplay (the ˚ ˚ characteristic length scale being Angstrom to sub - Angstrom), the relevant timescales are in the range of femtoseconds (fs) to picoseconds (ps), i.e. 10 15–10 12 s. Getting insights into the dynamics on the fundamental time and length scales requires the use of diffraction techniques, since even for the current state of the art electron microscopes the atomic resolution poses a challenge. Therefore electron and X-ray diffraction are the methods of choice. To achieve the required high temporal resolution, one relies on femtosecond laser systems as the common starting point for the generation of ultra-short electron or X-ray pulses. Since camera systems for electrons as well as X-ray diffractometers are not available with such a high shutter speed, a stroboscopic scheme is employed, in literature commonly referred to as the pump–probe
n Corresponding author at: University of Konstanz, Physics Department and Center of Applied Photonics and Zukunftskolleg, 78464 Konstanz, Germany. E-mail address: [email protected] (J. Demsar).
0304-3991/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ultramic.2012.07.005
technique. This method however restricts the effects under study to be light sensitive, since they are being triggered by a laser pulse. A model system to study strong electron–lattice correlations are the Charge density wave (CDW) compounds. CDWs are common to many systems which have reduced dimensionality and exhibit metallic conductivity at elevated temperatures [3]. In CDWs, a modulation of both the conduction electron density and the lattice constants appears, the latter being referred to as a periodic lattice distortion (PLD); both are individually observable by Scanning Tunneling Microscopy and diffraction techniques [4,5]. Several pump–probe studies have been carried out on the quasi twodimensional CDW system 1T-TaS2, however, solely probing the electronic subsystem by measuring the change in reflectivity or in photoelectron yield [6,7]. Yet, only in the femtosecond electron diffraction (FED) experiment described in Ref. [1], insights into the dynamics of the lattice part of the CDW order parameter, as well as the phase transition between different CDW phases could be gained. One of the main challenges on the way to perform this experiment has been the preparation of suitably dimensioned samples. Since Coulomb explosion of dense electron packets quickly blows up their temporal duration, one way to achieve high temporal resolution is to keep the propagation distance to the sample as short as possible [8]. This limits the electron optics in front of the sample to as few components as possible, typically
10
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
just one collimating electron lens [8]. Consequently the lateral dimensions of the electron beam are on the order of 100 mm, as also the laser spot size on the photocathode is typically on the order of several tens of micrometers. This measure, together with the low mean free path of 30–50 keV electrons in matter, sets the sample requirements to 100 nm in longitudinal and 100 mm in transversal direction. In contrast to conventional structure determination techniques like transmission electron microscopy (TEM) or X-ray diffraction, the outlined geometrical constraints are completely new. While TEM studies require sample thickness of o100 nm, the lateral sizes can be on the order of several 100 nm, since the electron beam can be collimated accordingly. In X-ray diffraction, bulk crystals with several millimeters in lateral size must be used to get significant scattered intensity. In order to meet the above mentioned sample requirements we have undertaken reactive ion etching, ion milling and focused ion beam techniques for preparing suitably thin single crystalline samples. All of the above methods turned out to introduce dopants at a significant level. Finally we met the required sample dimensions by using an ultra-microtome. Given the fact that the technique can be applied to numerous solid state materials ranging from simple metals to conventional and unconventional superconductors, this preparation scheme opens up a pathway for systematic studies of femtosecond structural dynamics in advanced quantum solids.
2. Motivation 2.1. Femtosecond electron diffraction technique For tabletop femtosecond electron diffraction (FED) commonly a pulsed femtosecond laser system is used, where one part of the laser beam is split off and converted to free electrons via a photoactivated cathode [9,10]. The schematic is shown in Fig. 1. For the FED experiment on 1T-TaS2 [1] an amplified Ti:sapphire laser delivered 180 fs pulses with a central wavelength of 775 nm at a repetition rate of 1 kHz. Passing through a beam splitter each pulse was divided into two, where the first served for sample excitation. The second part of the beam was fed to a Non-collinear Optical Parametric Amplifier (NOPA), yielding 20 fs pulses at a central wavelength of l ¼510 nm. Subsequent focusing on a backilluminated 10 nm thin gold photocathode resulted in the emission of photoelectrons via two-photon absorption process. The photo-electrons are accelerated by a voltage of 50 kV, guiding the electron bunch through a collimating magnetic lens onto the sample. To determine the electron current a Faraday-cup was used, yielding 4000 electrons per pulse. The electron pulse duration was determined to be 200 fs (FWHM), using an electron/laser-pulse cross-correlation method based on the ponderomotive scattering [11]. The overall instrumental time resolution was 250 fs (FWHM). At the sample position the FWHM of the pump and the electron beam were 350 mm and 150 mm, respectively, thus ensuring homogeneous excitation of the probed region. For imaging the diffraction pattern, the electrons scattered from the sample were multiplied by a pair of chevron-configured micro-channel plates (MCP) providing maximum gain up to 106. The MCP output is directed onto a phosphor screen which is photographed by a 1 megapixel, thermoelectrically cooled, backside illuminated CCD camera. In this experiment the sample was thermally connected to a cold finger cooled by liquid nitrogen. In this way a stable temperature of 200 K was achieved. In order to study the structural dynamics, the time delay between the photo-excitation and the electron probe pulse was varied by
Fig. 1. Schematics of the femtosecond electron diffraction setup: A laser system delivers femtosecond laser pulses, one part of the laser power being used to drive the Non-collinear Optical Parametric Amplifier (NOPA). The NOPA pulses are centered around a central wavelength of l ¼ 510 nm and create, via the twophoton absorption, free electrons in the electron gun. By changing the relative time delay between the two pulses at the sample, snapshots of the reciprocal lattice after photoexcitation are taken (bottom left). As the system maps all the in-plane components of the crystal, information on the superstructure, inelastic background and the host crystal structure can be extracted. The top plot shows the time evolution of the relative intensity changes of the superlattice (blue solid circles), Bragg (red diamonds) and inelastic background (gray open circles) intensities following the perturbation of the sample with an optical pulse of 2.4 mJ/cm2. The initial suppression of the CDW intensity and simultaneous intensity gain of the Bragg reflections happen on timescales of several hundred femtoseconds. The decrease of the Bragg transient as well as the increase in the inelastic background is indicative of the onset of the Debye Waller effect, i.e. excitation of qa 0 phonons. The CDW intensity is also affected by the lattice thermalization and its minimum is delayed by 300 fs with respect to the Bragg maximum. Recovery of the CDW transient happens within a characteristic timescale of 4.5 ps (for more details see text and [1]. For interpretation of the references to colour in this figure legend , the reader is referred to the web version of this article).
100–500 fs steps using a mechanical delay stage. More details about the setup can be found in Ref. [12]. 2.2. What can we learn from FED experiments on charge density wave systems? Various reports on anomalies in the physical properties of 1T-TaS2 [13] have attracted considerable interest over the years, discovering a whole repertory of correlated electron phenomena. Amongst them, Mott-insulating behavior [14], superconductivity under pressure [15] and the formation of charge density waves with different degrees of commensurability [16,17] have been reported. There are still many open questions to be answered regarding this compound, especially concerning the competition between Mott physics and CDW order, as well as the nature of various CDW phases, since recent reports challenge the classical Fermi nesting picture [18–20]. The FED study presented in [1] was carried out at a sample temperature of 200 K, meaning that the crystal was in the nearly commensurate CDW state with a maximum amplitude of the PLD of 0.1 A˚ [17]. The main results are depicted in Fig. 1 displaying the time evolution of the intensity of the Bragg peaks of the host lattice (red solid diamonds), superstructure peaks (blue solid circles) and the inelastic background (gray open circles); these traces are obtained following photo-excitation of the free standing film with a 400 nm pulse at the excitation fluence
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
of 2.4 mJ/cm2. The intensity of the superstructure peaks was suppressed by 35% within a few hundred fs. The Bragg intensity, on the other hand, showed a simultaneous increase by nearly 20%. Both effects can be shown to naturally follow the expected behavior of an electronically driven suppression of the PLD in a Peierls system. In the presence of CDWs the associated PLD acts like a static Debye Waller effect, taking the scattering intensity from the Bragg peaks of the underlying lattice. The laser pulse exciting the electronic system results in smearing out of the conduction electron density modulation. Since the interatomic potential is thereby changed towards that of the high temperature incommensurate CDW phase (present in the temperature range between 350 and 550 K), the collective motion of ions towards this less displaced phase is launched. A detailed analysis showed that the timescale for CDW melting is on the order of one half of the totally symmetric A1g phonon period ( 440 fs) which is commonly attributed to the amplitude mode of the system. Apart from the ability to track the CDW melting process, the fact that in FED the scattering intensity over the entire Brillouin zone is recorded at the same time enables further insights into the subsequent relaxation processes. As time evolves the intensity of the Bragg peaks starts to decrease rapidly, hand in hand with an increase of the inelastic background. The fitted time constant of 350 fs can be unambiguously attributed to the rapid energy transfer from the electronic system to qa0 phonons (standard Debye–Waller mechanism). Comparison with the optical data, taken at similar conditions, indeed reveals a rapid recovery of the electronic density modulation, which was in the framework of time dependent Ginzburg–Landau analysis attributed to an overdamped collective mode of the electronic system [21]. The rapid energy transfer to qa0 phonons gives rise to the recovery of the initially photo-induced enhancement of the Bragg intensity and further suppresses the intensity of the superstructure peaks, in a similar way as a standard Debye–Waller effect. After this process the system is not in equilibrium yet, as Bragg and superlattice peak intensities show further recovery with a characteristic time constant of 4.5 ps. While the exact nature of the latter process is still under debate, the temperature dependence of its timescale (obtained by optical time-resolved techniques) suggests it could be related to anharmonic decay of optical modes.
3. Sample preparation: results and discussion We have used the following approach to grow single crystalline 1T-TaS2 samples. In the first step polycrystalline TaS2 samples were synthesized by directly reacting sulfur (Alfa Products, 99.9995%) and tantalum wire (j ¼0.02 inch, Material Research Corp., 99.95%). The stoichiometrically matched amounts of tantalum and sulfur were placed in a quartz tube. An excess amount of sulfur helped in stabilizing the formation of the 1T poly-type. The tubes were flame-sealed after being evacuated to 10 6 mbar. Afterwards, the temperature was slowly increased at a rate of 50 1C per day until the final temperature of 950 1C was reached. The tubes were held at 950 1C for 10 days. After opening the tubes, the obtained crystals were crushed and reheated to 950 1C to ensure complete homogenization. Only 4 days later the tubes were quenched to 0 1C. Using X-ray diffraction, we confirmed the established 1T structure. From this compound, single crystals were grown by iodine vapor transport from 950 1C to 900 1C in closed quartz tubes (20 mm in diameter and 200 mm long). Excess sulfur (1–2 mg/cm3) was present during the preparation and quenching. Golden-colored metallic crystals of up to 1 cm in diameter were obtained in most cases. To meet the sample requirements for FED experiment different sample thinning schemes were tested. First efforts of cleaving the
11
sample with a scotch tape and other adhesives have proven to produce cracks in the sample and thus did not allow for large enough lateral sample sizes in a free standing configuration of the required thickness. An alternative approach to get free standing films was to use the focused ion beam column of a scanning electron microscope (Zeiss CrossBeam 1540XB). It turned out that tantalum disulfide is very resistive to ion milling with gallium ions and the thinning of a 100 mm surface had taken more than 10 h of operation. Moreover, already short milling times resulted in gallium implantation, leading to a doping level of 5%, as determined by the energy dispersive X-ray spectroscopy (EDX). As various studies on the intercalation of different dopants into the van der Waals-gap of 1T-TaS2 have reported quite profound effects on the physics of the system [14,22,23], uncontrolled doping has to be avoided by any means. Another approach in gaining laterally large but thin enough films of tantalum disulfide was reactive ion etching (SENTECH SI 220) with SF6 and CF4. While this method provided quite large and thin enough films (see panel (a) of Fig. 2), stopping the etching process at the desired specimen thickness turned out to be a tedious task. However, analyzing the etched specimens by EDX spectroscopy (see panels (b) and (d) in Fig. 2) proved that there was carbon implantation of up to 15%. Together with the fact that the film thickness was inhomogeneous, such high doping levels eliminated further attempts of getting thin samples along this route. In comparison to untreated and freshly cleaved samples the profound doping levels are observed (see panel (c) of Fig. 2). The least modifications to the crystal happened when we used an ultra-microtome (Leica EM FC6 Microtome). Using this method we obtained free standing single crystalline films shown in Fig. 3, panels (a)–(c). Here, no doping or intercalation was detected and the maximum lateral film dimensions reached up to 200 mm, whilst the thickness was only 30 nm. The thickness of the film was determined by adjusting the feed rate of the microtome, and was later checked by measuring the light absorbance in combination with known bulk optical constants. The profound two dimensional character of the material facilitates the cutting process, which should rather be looked upon as a cleaving process. The main challenge in microtoming single crystals is to find the right crystal orientation and to align the knife parallel to it. This proves to be an even more difficult task, as the tantalum atoms readily react with the diamond knife, thus not leaving many attempts for finding the right cutting alignment. Already after a few cuts, the used part of the diamond knife blunted and due to the resulting roughness of the knife’s edge, broken stripes were cut instead of continuous films. This effect can already be seen in the transmission electron microscope (TEM) image in Fig. 3, panel (c), where the edge of the film is not sharp, but rather scraggy. Fig. 3, panels (a) and (b), shows light microscope images of several 30 nm films on a 40 mm mesh sized transmission electron microscope grid. Panel (c) depicts a sample in the TEM on a 100 mm mesh sized grid. On this sample the temperature dependent diffraction images in panels (d)–(f) have been recorded. They are in full agreement with previously published TEM data [24] and show three different CDW phases. At T¼363 K, the sample is in the incommensurate CDW phase, reaching the nearly commensurate CDW phase upon cooling below 350 K. Here the super-modulation peaks are easily seen as six weak satellite reflections around every Bragg peak. At T¼ 170 K, the sample is in the commensurate CDW phase, with further increase in intensity of the satellite reflections. The free standing 1T-TaS2 samples prepared in this way have been used both in FED experiments as well as in the optical timeresolved experiments. While the technique is essential for the
12
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
Fig. 2. Reactive ion etching of 1T-TaS2: Panel (a) shows a transmission light microscope image of the thin film produced by reactive ion etching. The thickness inhomogeneity of the top semitransparent slab is apparent and analyzing the stoichiometry of the film via energy dispersive x-ray (EDX) spectroscopy (panel (b)) indicates doping with oxygen and carbon. The EDX spectra on pristine and reactive ion etched samples are shown in panels (c) and (d), respectively.
Fig. 3. Free standing single crystalline 1T-TaS2 films: Panels (a) and (b) show 30 nm thin slices of 1T-TaS2 under the light microscope in reflection and transmission geometry, respectively. Panel (c) depicts 30 nm thin slices on a 100 mm sized mesh in the low magnification setting of a 120 kV transmission electron microscope. Same films already show signs of the blunted diamond knife, as they are striped. Panels (d)–(f) show the diffraction pattern of the central film shown in panel (c) at different temperatures, corresponding to different CDW phases [24] — see text for more details.
FED, it should also be extremely beneficial for time resolved optical techniques. The achieved sample thicknesses provide an advantage since a homogeneous optical excitation profile can be ensured as the optical penetration length is on the order of 100 nm. By further increasing the lateral size it should be possible to use free standing samples to perform studies of the complex conductivity dynamics down to the terahertz frequencies.
4. Conclusion In conclusion, we have demonstrated a way to produce 200 mm 200 mm 30 nm free standing single crystalline samples without affecting the phase transition temperatures, which are governed by the delicate balance between the electronic and lattice degrees of freedom in this intricate system. As opposed to
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
milling techniques, the use of an ultra-microtome did not introduce any dopants at a detectable level. Preliminary efforts in microtoming other Charge density wave compounds like 2H-TaS2 have proven successful and are opening the way for systematic studies of CDWs using FED.
Acknowledgments We would like to acknowledge Matthias Hagner for his help by various sample thinning techniques and Humphrey Morhenn for his help with EDX analysis. Moreover, we would like to acknowledge Dwayne Miller, German Sciaini, Gustavo Moriena, Hanjo Sch¨afer and Markus Beyer for an extremely fruitful collaboration on FED on 1T-TaS2. The research was supported by the Sofja Kovalevskaja Award of the Alexander von Humboldt Foundation, Center for Applied Photonics and Zukunftskolleg at the University of Konstanz. M.E. acknowledges financial support through Stiftung der Deutschen Wirtschaft. H.B. acknowledges financial support from the Swiss NSF and by the NCCR MaNEP. References [1] M. Eichberger, H. Schafer, M. Krumova, M. Beyer, J. Demsar, H. Berger, G. Moriena, G. Sciaini, R.J.D. Miller, Snapshots of cooperative atomic motions in the optical suppression of charge density waves, Nature 468 (2010) 799–802. [2] W.E. King, G.H. Campbell, A. Frank, B. Reed, J.F. Schmerge, B.J. Siwick, B.C. Stuart, P.M. Weber, Ultrafast electron microscopy in materials science, biology, and chemistry, Journal of Applied Physics 97 (2005) 111101. [3] G. Gruner, The dynamics of charge-density waves, Physica Scripta 32 (1985) 11–25. [4] B. Burk, R.E. Thomson, A. Zettl, J. Clarke, Charge-density-wave domains in 1T-Tas2 observed by satellite structure in scanning-tunneling-microscopy images, Physical Review Letters 66 (1991) 3040–3043. [5] G.P.E.M. Vanbakel, J.T.M. Dehosson, Various regimes of charge-density waves in layered compounds, Physical Review B 46 (1992) 2001–2007. [6] J. Demsar, L. Forro, H. Berger, D. Mihailovic, Femtosecond snapshots of gapforming charge-density-wave correlations in quasi-two-dimensional dichalcogenides 1T-TaS2 and 2H-TaSe2, Physical Review B 66 (2002) 041101(R). [7] L. Perfetti, P.A. Loukakos, M. Lisowski, U. Bovensiepen, H. Berger, S. Biermann, P.S. Cornaglia, A. Georges, M. Wolf, Time evolution of the electronic structure of 1T-TaS2 through the insulator–metal transition, Physical Review Letters 97 (2006) 067402. [8] B.J. Siwick, J.R. Dwyer, R.E. Jordan, R.J.D. Miller, Ultrafast electron optics: propagation dynamics of femtosecond electron packets, Journal of Applied Physics 92 (2002) 1643–1648.
13
[9] T. van Oudheusden, E.F. de Jong, S.B. van der Geer, W.P.E.M.O. Root, O.J. Luiten, B.J. Siwick, Electron source concept for single-shot sub-100 fs electron diffraction in the 100 keV range, Journal of Applied Physics 102 (2007) 093501. [10] T. van Oudheusden, P.L.E.M. Pasmans, S.B. van der Geer, M.J. de Loos, M.J. van der Wiel, O.J. Luiten, Compression of subrelativistic space-charge-dominated electron bunches for single-shot femtosecond electron diffraction, Physical Review Letters 105 (2010) 264801. [11] C.T. Hebeisen, G. Sciaini, M. Harb, R. Ernstorfer, T. Dartigalongue, S.G. Kruglik, R.J.D. Miller, Grating enhanced ponderomotive scattering for visualization and full characterization of femtosecond electron pulses, Optics Express 16 (2008) 3334–3341. [12] M. Harb, W. Peng, G. Sciaini, C.T. Hebeisen, R. Ernstorfer, M.A. Eriksson, M.G. Lagally, S.G. Kruglik, R.J.D. Miller, Excitation of longitudinal and transverse coherent acoustic phonons in nanometer free-standing films of (001) Si, Physical Review B 79 (2009) 094301. [13] A.H. Thompson, F.R. Gamble, J.F. Revelli, Transitions between semiconducting and metallic phases in 1T-Tas2, Solid State Communications 9 (1971) 981–985. [14] P. Fazekas, E. Tosatti, Electrical, structural and magnetic-properties of pure and doped 1T-Tas2, Philosophical Magazine B 39 (1979) 229–244. [15] B. Sipos, A.F. Kusmartseva, A. Akrap, H. Berger, L. Forro, E. Tutis, From Mott state to superconductivity in 1T-TaS2, Nature Materials 7 (2008) 960–965. [16] J.A. Wilson, F.J. Disalvo, S. Mahajan, Charge-density waves and superlattices in metallic layered transition-metal dichalcogenides, Advances in Physics 24 (1975) 117–201. [17] A. Spijkerman, J.L. deBoer, A. Meetsma, G.A. Wiegers, S. vanSmaalen, X-ray crystal-structure refinement of the nearly commensurate phase of 1T-TaS2 in (3þ 2)-dimensional superspace, Physical Review B 56 (1997) 13757–13767. [18] F. Clerc, C. Battaglia, H. Cercellier, C. Monney, H. Berger, L. Despont, M.G. Garnier, P. Aebi, Fermi surface of layered compounds and bulk charge density wave systems, Journal of Physics—Condensed Matter 19 (2007) 355002. [19] N. Dean, J.C. Petersen, D. Fausti, R.I. Tobey, S. Kaiser, L.V. Gasparov, H. Berger, A. Cavalleri, Polaronic conductivity in the photoinduced phase of 1T-TaS(2), Physical Review Letters 106 (2011) 016401. [20] S. Hellmann, M. Beye, C. Sohrt, T. Rohwer, F. Sorgenfrei, H. Redlin, M. Kallane, M. Marczynski-Buhlow, F. Hennies, M. Bauer, A. Fohlisch, L. Kipp, W. Wurth, K. Rossnagel, Ultrafast melting of a charge-density wave in the Mott insulator 1T-TaS(2), Physical Review Letters 105 (2010) 187401. [21] H. Schafer, V.V. Kabanov, M. Beyer, K. Biljakovic, J. Demsar, Disentanglement of the electronic and lattice parts of the order parameter in a 1D charge density wave system probed by femtosecond spectroscopy, Physical Review Letters 105 (2010) 066402. [22] L. Diebolt, W. Glaunsinger, M. McKelvy, Intercalation and deintercalation processes for ammoniated 1T-TaS2, Journal of Solid State Chemistry 145 (1999) 336–344. [23] A. Prodan, V. Marinkovic, F.W. Boswell, J.C. Bennett, M. Remskar, Chargedensity waves in some Nb and Ta chalcogenides, Journal of Alloys and Compounds 219 (1995) 69–72. [24] T. Ishiguro, H. Sato, Electron-microscopy of phase-transformations in 1T-Tas2, Physical Review B 44 (1991) 2046–2060.
Contents lists available at SciVerse ScienceDirect
Ultramicroscopy journal homepage: www.elsevier.com/locate/ultramic
Sample preparation methods for femtosecond electron diffraction experiments Maximilian Eichberger a, Marina Krumova b, Helmuth Berger d, Jure Demsar a,c,n a
Physics Department and Center of Applied Photonics and Zukunftskolleg, University of Konstanz, D-78464, Germany Department of Chemistry, University of Konstanz, D-78464, Germany c Complex Matter Department, Jozef Stefan Institute, SI-1000 Ljubljana, Slovenia d Physics Department, EPFL CH-1015 Lausanne, Switzerland b
a r t i c l e i n f o
abstract
Available online 2 August 2012
Exploring the dynamics of Charge density wave system 1T-TaS2 via femtosecond electron diffraction demonstrated the power of this technique for studying ultrafast structural phenomena in strongly correlated electron materials [1]. The results revealed first direct information on the order parameter dynamics of Charge Density Waves as well as on their photo-induced phase transition. A prerequisite to perform such experiments on modern quantum materials is the availability of laterally large ( 100 mm) and sufficiently thin ( o 100 nm) single crystalline samples. Different approaches to reach these specifications have been tried out and their effect on sample integrity has been investigated. Finally, using an ultra-microtome, we were able to prepare 30 nm free standing single crystalline films of 1T-TaS2 with lateral dimensions of 200 mm 200 mm. We have characterized these films with different techniques for their stoichiometric and crystalline integrity, ensuring no measurable alternation of sample properties. The application of this sample thinning technique is expected to find its use in further structural dynamics studies, as well as in optical time-resolved studies where homogeneous excitation profile and/or data in transmission geometry may be required. & 2012 Elsevier B.V. All rights reserved.
1. Introduction In microscopy a growing effort is currently being put into extending the field of interest to the temporal dimension of objects under study [2]. In solid state physics, where one is interested in the motion of electrons, ions and their mutual interplay (the ˚ ˚ characteristic length scale being Angstrom to sub - Angstrom), the relevant timescales are in the range of femtoseconds (fs) to picoseconds (ps), i.e. 10 15–10 12 s. Getting insights into the dynamics on the fundamental time and length scales requires the use of diffraction techniques, since even for the current state of the art electron microscopes the atomic resolution poses a challenge. Therefore electron and X-ray diffraction are the methods of choice. To achieve the required high temporal resolution, one relies on femtosecond laser systems as the common starting point for the generation of ultra-short electron or X-ray pulses. Since camera systems for electrons as well as X-ray diffractometers are not available with such a high shutter speed, a stroboscopic scheme is employed, in literature commonly referred to as the pump–probe
n Corresponding author at: University of Konstanz, Physics Department and Center of Applied Photonics and Zukunftskolleg, 78464 Konstanz, Germany. E-mail address: [email protected] (J. Demsar).
0304-3991/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ultramic.2012.07.005
technique. This method however restricts the effects under study to be light sensitive, since they are being triggered by a laser pulse. A model system to study strong electron–lattice correlations are the Charge density wave (CDW) compounds. CDWs are common to many systems which have reduced dimensionality and exhibit metallic conductivity at elevated temperatures [3]. In CDWs, a modulation of both the conduction electron density and the lattice constants appears, the latter being referred to as a periodic lattice distortion (PLD); both are individually observable by Scanning Tunneling Microscopy and diffraction techniques [4,5]. Several pump–probe studies have been carried out on the quasi twodimensional CDW system 1T-TaS2, however, solely probing the electronic subsystem by measuring the change in reflectivity or in photoelectron yield [6,7]. Yet, only in the femtosecond electron diffraction (FED) experiment described in Ref. [1], insights into the dynamics of the lattice part of the CDW order parameter, as well as the phase transition between different CDW phases could be gained. One of the main challenges on the way to perform this experiment has been the preparation of suitably dimensioned samples. Since Coulomb explosion of dense electron packets quickly blows up their temporal duration, one way to achieve high temporal resolution is to keep the propagation distance to the sample as short as possible [8]. This limits the electron optics in front of the sample to as few components as possible, typically
10
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
just one collimating electron lens [8]. Consequently the lateral dimensions of the electron beam are on the order of 100 mm, as also the laser spot size on the photocathode is typically on the order of several tens of micrometers. This measure, together with the low mean free path of 30–50 keV electrons in matter, sets the sample requirements to 100 nm in longitudinal and 100 mm in transversal direction. In contrast to conventional structure determination techniques like transmission electron microscopy (TEM) or X-ray diffraction, the outlined geometrical constraints are completely new. While TEM studies require sample thickness of o100 nm, the lateral sizes can be on the order of several 100 nm, since the electron beam can be collimated accordingly. In X-ray diffraction, bulk crystals with several millimeters in lateral size must be used to get significant scattered intensity. In order to meet the above mentioned sample requirements we have undertaken reactive ion etching, ion milling and focused ion beam techniques for preparing suitably thin single crystalline samples. All of the above methods turned out to introduce dopants at a significant level. Finally we met the required sample dimensions by using an ultra-microtome. Given the fact that the technique can be applied to numerous solid state materials ranging from simple metals to conventional and unconventional superconductors, this preparation scheme opens up a pathway for systematic studies of femtosecond structural dynamics in advanced quantum solids.
2. Motivation 2.1. Femtosecond electron diffraction technique For tabletop femtosecond electron diffraction (FED) commonly a pulsed femtosecond laser system is used, where one part of the laser beam is split off and converted to free electrons via a photoactivated cathode [9,10]. The schematic is shown in Fig. 1. For the FED experiment on 1T-TaS2 [1] an amplified Ti:sapphire laser delivered 180 fs pulses with a central wavelength of 775 nm at a repetition rate of 1 kHz. Passing through a beam splitter each pulse was divided into two, where the first served for sample excitation. The second part of the beam was fed to a Non-collinear Optical Parametric Amplifier (NOPA), yielding 20 fs pulses at a central wavelength of l ¼510 nm. Subsequent focusing on a backilluminated 10 nm thin gold photocathode resulted in the emission of photoelectrons via two-photon absorption process. The photo-electrons are accelerated by a voltage of 50 kV, guiding the electron bunch through a collimating magnetic lens onto the sample. To determine the electron current a Faraday-cup was used, yielding 4000 electrons per pulse. The electron pulse duration was determined to be 200 fs (FWHM), using an electron/laser-pulse cross-correlation method based on the ponderomotive scattering [11]. The overall instrumental time resolution was 250 fs (FWHM). At the sample position the FWHM of the pump and the electron beam were 350 mm and 150 mm, respectively, thus ensuring homogeneous excitation of the probed region. For imaging the diffraction pattern, the electrons scattered from the sample were multiplied by a pair of chevron-configured micro-channel plates (MCP) providing maximum gain up to 106. The MCP output is directed onto a phosphor screen which is photographed by a 1 megapixel, thermoelectrically cooled, backside illuminated CCD camera. In this experiment the sample was thermally connected to a cold finger cooled by liquid nitrogen. In this way a stable temperature of 200 K was achieved. In order to study the structural dynamics, the time delay between the photo-excitation and the electron probe pulse was varied by
Fig. 1. Schematics of the femtosecond electron diffraction setup: A laser system delivers femtosecond laser pulses, one part of the laser power being used to drive the Non-collinear Optical Parametric Amplifier (NOPA). The NOPA pulses are centered around a central wavelength of l ¼ 510 nm and create, via the twophoton absorption, free electrons in the electron gun. By changing the relative time delay between the two pulses at the sample, snapshots of the reciprocal lattice after photoexcitation are taken (bottom left). As the system maps all the in-plane components of the crystal, information on the superstructure, inelastic background and the host crystal structure can be extracted. The top plot shows the time evolution of the relative intensity changes of the superlattice (blue solid circles), Bragg (red diamonds) and inelastic background (gray open circles) intensities following the perturbation of the sample with an optical pulse of 2.4 mJ/cm2. The initial suppression of the CDW intensity and simultaneous intensity gain of the Bragg reflections happen on timescales of several hundred femtoseconds. The decrease of the Bragg transient as well as the increase in the inelastic background is indicative of the onset of the Debye Waller effect, i.e. excitation of qa 0 phonons. The CDW intensity is also affected by the lattice thermalization and its minimum is delayed by 300 fs with respect to the Bragg maximum. Recovery of the CDW transient happens within a characteristic timescale of 4.5 ps (for more details see text and [1]. For interpretation of the references to colour in this figure legend , the reader is referred to the web version of this article).
100–500 fs steps using a mechanical delay stage. More details about the setup can be found in Ref. [12]. 2.2. What can we learn from FED experiments on charge density wave systems? Various reports on anomalies in the physical properties of 1T-TaS2 [13] have attracted considerable interest over the years, discovering a whole repertory of correlated electron phenomena. Amongst them, Mott-insulating behavior [14], superconductivity under pressure [15] and the formation of charge density waves with different degrees of commensurability [16,17] have been reported. There are still many open questions to be answered regarding this compound, especially concerning the competition between Mott physics and CDW order, as well as the nature of various CDW phases, since recent reports challenge the classical Fermi nesting picture [18–20]. The FED study presented in [1] was carried out at a sample temperature of 200 K, meaning that the crystal was in the nearly commensurate CDW state with a maximum amplitude of the PLD of 0.1 A˚ [17]. The main results are depicted in Fig. 1 displaying the time evolution of the intensity of the Bragg peaks of the host lattice (red solid diamonds), superstructure peaks (blue solid circles) and the inelastic background (gray open circles); these traces are obtained following photo-excitation of the free standing film with a 400 nm pulse at the excitation fluence
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
of 2.4 mJ/cm2. The intensity of the superstructure peaks was suppressed by 35% within a few hundred fs. The Bragg intensity, on the other hand, showed a simultaneous increase by nearly 20%. Both effects can be shown to naturally follow the expected behavior of an electronically driven suppression of the PLD in a Peierls system. In the presence of CDWs the associated PLD acts like a static Debye Waller effect, taking the scattering intensity from the Bragg peaks of the underlying lattice. The laser pulse exciting the electronic system results in smearing out of the conduction electron density modulation. Since the interatomic potential is thereby changed towards that of the high temperature incommensurate CDW phase (present in the temperature range between 350 and 550 K), the collective motion of ions towards this less displaced phase is launched. A detailed analysis showed that the timescale for CDW melting is on the order of one half of the totally symmetric A1g phonon period ( 440 fs) which is commonly attributed to the amplitude mode of the system. Apart from the ability to track the CDW melting process, the fact that in FED the scattering intensity over the entire Brillouin zone is recorded at the same time enables further insights into the subsequent relaxation processes. As time evolves the intensity of the Bragg peaks starts to decrease rapidly, hand in hand with an increase of the inelastic background. The fitted time constant of 350 fs can be unambiguously attributed to the rapid energy transfer from the electronic system to qa0 phonons (standard Debye–Waller mechanism). Comparison with the optical data, taken at similar conditions, indeed reveals a rapid recovery of the electronic density modulation, which was in the framework of time dependent Ginzburg–Landau analysis attributed to an overdamped collective mode of the electronic system [21]. The rapid energy transfer to qa0 phonons gives rise to the recovery of the initially photo-induced enhancement of the Bragg intensity and further suppresses the intensity of the superstructure peaks, in a similar way as a standard Debye–Waller effect. After this process the system is not in equilibrium yet, as Bragg and superlattice peak intensities show further recovery with a characteristic time constant of 4.5 ps. While the exact nature of the latter process is still under debate, the temperature dependence of its timescale (obtained by optical time-resolved techniques) suggests it could be related to anharmonic decay of optical modes.
3. Sample preparation: results and discussion We have used the following approach to grow single crystalline 1T-TaS2 samples. In the first step polycrystalline TaS2 samples were synthesized by directly reacting sulfur (Alfa Products, 99.9995%) and tantalum wire (j ¼0.02 inch, Material Research Corp., 99.95%). The stoichiometrically matched amounts of tantalum and sulfur were placed in a quartz tube. An excess amount of sulfur helped in stabilizing the formation of the 1T poly-type. The tubes were flame-sealed after being evacuated to 10 6 mbar. Afterwards, the temperature was slowly increased at a rate of 50 1C per day until the final temperature of 950 1C was reached. The tubes were held at 950 1C for 10 days. After opening the tubes, the obtained crystals were crushed and reheated to 950 1C to ensure complete homogenization. Only 4 days later the tubes were quenched to 0 1C. Using X-ray diffraction, we confirmed the established 1T structure. From this compound, single crystals were grown by iodine vapor transport from 950 1C to 900 1C in closed quartz tubes (20 mm in diameter and 200 mm long). Excess sulfur (1–2 mg/cm3) was present during the preparation and quenching. Golden-colored metallic crystals of up to 1 cm in diameter were obtained in most cases. To meet the sample requirements for FED experiment different sample thinning schemes were tested. First efforts of cleaving the
11
sample with a scotch tape and other adhesives have proven to produce cracks in the sample and thus did not allow for large enough lateral sample sizes in a free standing configuration of the required thickness. An alternative approach to get free standing films was to use the focused ion beam column of a scanning electron microscope (Zeiss CrossBeam 1540XB). It turned out that tantalum disulfide is very resistive to ion milling with gallium ions and the thinning of a 100 mm surface had taken more than 10 h of operation. Moreover, already short milling times resulted in gallium implantation, leading to a doping level of 5%, as determined by the energy dispersive X-ray spectroscopy (EDX). As various studies on the intercalation of different dopants into the van der Waals-gap of 1T-TaS2 have reported quite profound effects on the physics of the system [14,22,23], uncontrolled doping has to be avoided by any means. Another approach in gaining laterally large but thin enough films of tantalum disulfide was reactive ion etching (SENTECH SI 220) with SF6 and CF4. While this method provided quite large and thin enough films (see panel (a) of Fig. 2), stopping the etching process at the desired specimen thickness turned out to be a tedious task. However, analyzing the etched specimens by EDX spectroscopy (see panels (b) and (d) in Fig. 2) proved that there was carbon implantation of up to 15%. Together with the fact that the film thickness was inhomogeneous, such high doping levels eliminated further attempts of getting thin samples along this route. In comparison to untreated and freshly cleaved samples the profound doping levels are observed (see panel (c) of Fig. 2). The least modifications to the crystal happened when we used an ultra-microtome (Leica EM FC6 Microtome). Using this method we obtained free standing single crystalline films shown in Fig. 3, panels (a)–(c). Here, no doping or intercalation was detected and the maximum lateral film dimensions reached up to 200 mm, whilst the thickness was only 30 nm. The thickness of the film was determined by adjusting the feed rate of the microtome, and was later checked by measuring the light absorbance in combination with known bulk optical constants. The profound two dimensional character of the material facilitates the cutting process, which should rather be looked upon as a cleaving process. The main challenge in microtoming single crystals is to find the right crystal orientation and to align the knife parallel to it. This proves to be an even more difficult task, as the tantalum atoms readily react with the diamond knife, thus not leaving many attempts for finding the right cutting alignment. Already after a few cuts, the used part of the diamond knife blunted and due to the resulting roughness of the knife’s edge, broken stripes were cut instead of continuous films. This effect can already be seen in the transmission electron microscope (TEM) image in Fig. 3, panel (c), where the edge of the film is not sharp, but rather scraggy. Fig. 3, panels (a) and (b), shows light microscope images of several 30 nm films on a 40 mm mesh sized transmission electron microscope grid. Panel (c) depicts a sample in the TEM on a 100 mm mesh sized grid. On this sample the temperature dependent diffraction images in panels (d)–(f) have been recorded. They are in full agreement with previously published TEM data [24] and show three different CDW phases. At T¼363 K, the sample is in the incommensurate CDW phase, reaching the nearly commensurate CDW phase upon cooling below 350 K. Here the super-modulation peaks are easily seen as six weak satellite reflections around every Bragg peak. At T¼ 170 K, the sample is in the commensurate CDW phase, with further increase in intensity of the satellite reflections. The free standing 1T-TaS2 samples prepared in this way have been used both in FED experiments as well as in the optical timeresolved experiments. While the technique is essential for the
12
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
Fig. 2. Reactive ion etching of 1T-TaS2: Panel (a) shows a transmission light microscope image of the thin film produced by reactive ion etching. The thickness inhomogeneity of the top semitransparent slab is apparent and analyzing the stoichiometry of the film via energy dispersive x-ray (EDX) spectroscopy (panel (b)) indicates doping with oxygen and carbon. The EDX spectra on pristine and reactive ion etched samples are shown in panels (c) and (d), respectively.
Fig. 3. Free standing single crystalline 1T-TaS2 films: Panels (a) and (b) show 30 nm thin slices of 1T-TaS2 under the light microscope in reflection and transmission geometry, respectively. Panel (c) depicts 30 nm thin slices on a 100 mm sized mesh in the low magnification setting of a 120 kV transmission electron microscope. Same films already show signs of the blunted diamond knife, as they are striped. Panels (d)–(f) show the diffraction pattern of the central film shown in panel (c) at different temperatures, corresponding to different CDW phases [24] — see text for more details.
FED, it should also be extremely beneficial for time resolved optical techniques. The achieved sample thicknesses provide an advantage since a homogeneous optical excitation profile can be ensured as the optical penetration length is on the order of 100 nm. By further increasing the lateral size it should be possible to use free standing samples to perform studies of the complex conductivity dynamics down to the terahertz frequencies.
4. Conclusion In conclusion, we have demonstrated a way to produce 200 mm 200 mm 30 nm free standing single crystalline samples without affecting the phase transition temperatures, which are governed by the delicate balance between the electronic and lattice degrees of freedom in this intricate system. As opposed to
M. Eichberger et al. / Ultramicroscopy 127 (2013) 9–13
milling techniques, the use of an ultra-microtome did not introduce any dopants at a detectable level. Preliminary efforts in microtoming other Charge density wave compounds like 2H-TaS2 have proven successful and are opening the way for systematic studies of CDWs using FED.
Acknowledgments We would like to acknowledge Matthias Hagner for his help by various sample thinning techniques and Humphrey Morhenn for his help with EDX analysis. Moreover, we would like to acknowledge Dwayne Miller, German Sciaini, Gustavo Moriena, Hanjo Sch¨afer and Markus Beyer for an extremely fruitful collaboration on FED on 1T-TaS2. The research was supported by the Sofja Kovalevskaja Award of the Alexander von Humboldt Foundation, Center for Applied Photonics and Zukunftskolleg at the University of Konstanz. M.E. acknowledges financial support through Stiftung der Deutschen Wirtschaft. H.B. acknowledges financial support from the Swiss NSF and by the NCCR MaNEP. References [1] M. Eichberger, H. Schafer, M. Krumova, M. Beyer, J. Demsar, H. Berger, G. Moriena, G. Sciaini, R.J.D. Miller, Snapshots of cooperative atomic motions in the optical suppression of charge density waves, Nature 468 (2010) 799–802. [2] W.E. King, G.H. Campbell, A. Frank, B. Reed, J.F. Schmerge, B.J. Siwick, B.C. Stuart, P.M. Weber, Ultrafast electron microscopy in materials science, biology, and chemistry, Journal of Applied Physics 97 (2005) 111101. [3] G. Gruner, The dynamics of charge-density waves, Physica Scripta 32 (1985) 11–25. [4] B. Burk, R.E. Thomson, A. Zettl, J. Clarke, Charge-density-wave domains in 1T-Tas2 observed by satellite structure in scanning-tunneling-microscopy images, Physical Review Letters 66 (1991) 3040–3043. [5] G.P.E.M. Vanbakel, J.T.M. Dehosson, Various regimes of charge-density waves in layered compounds, Physical Review B 46 (1992) 2001–2007. [6] J. Demsar, L. Forro, H. Berger, D. Mihailovic, Femtosecond snapshots of gapforming charge-density-wave correlations in quasi-two-dimensional dichalcogenides 1T-TaS2 and 2H-TaSe2, Physical Review B 66 (2002) 041101(R). [7] L. Perfetti, P.A. Loukakos, M. Lisowski, U. Bovensiepen, H. Berger, S. Biermann, P.S. Cornaglia, A. Georges, M. Wolf, Time evolution of the electronic structure of 1T-TaS2 through the insulator–metal transition, Physical Review Letters 97 (2006) 067402. [8] B.J. Siwick, J.R. Dwyer, R.E. Jordan, R.J.D. Miller, Ultrafast electron optics: propagation dynamics of femtosecond electron packets, Journal of Applied Physics 92 (2002) 1643–1648.
13
[9] T. van Oudheusden, E.F. de Jong, S.B. van der Geer, W.P.E.M.O. Root, O.J. Luiten, B.J. Siwick, Electron source concept for single-shot sub-100 fs electron diffraction in the 100 keV range, Journal of Applied Physics 102 (2007) 093501. [10] T. van Oudheusden, P.L.E.M. Pasmans, S.B. van der Geer, M.J. de Loos, M.J. van der Wiel, O.J. Luiten, Compression of subrelativistic space-charge-dominated electron bunches for single-shot femtosecond electron diffraction, Physical Review Letters 105 (2010) 264801. [11] C.T. Hebeisen, G. Sciaini, M. Harb, R. Ernstorfer, T. Dartigalongue, S.G. Kruglik, R.J.D. Miller, Grating enhanced ponderomotive scattering for visualization and full characterization of femtosecond electron pulses, Optics Express 16 (2008) 3334–3341. [12] M. Harb, W. Peng, G. Sciaini, C.T. Hebeisen, R. Ernstorfer, M.A. Eriksson, M.G. Lagally, S.G. Kruglik, R.J.D. Miller, Excitation of longitudinal and transverse coherent acoustic phonons in nanometer free-standing films of (001) Si, Physical Review B 79 (2009) 094301. [13] A.H. Thompson, F.R. Gamble, J.F. Revelli, Transitions between semiconducting and metallic phases in 1T-Tas2, Solid State Communications 9 (1971) 981–985. [14] P. Fazekas, E. Tosatti, Electrical, structural and magnetic-properties of pure and doped 1T-Tas2, Philosophical Magazine B 39 (1979) 229–244. [15] B. Sipos, A.F. Kusmartseva, A. Akrap, H. Berger, L. Forro, E. Tutis, From Mott state to superconductivity in 1T-TaS2, Nature Materials 7 (2008) 960–965. [16] J.A. Wilson, F.J. Disalvo, S. Mahajan, Charge-density waves and superlattices in metallic layered transition-metal dichalcogenides, Advances in Physics 24 (1975) 117–201. [17] A. Spijkerman, J.L. deBoer, A. Meetsma, G.A. Wiegers, S. vanSmaalen, X-ray crystal-structure refinement of the nearly commensurate phase of 1T-TaS2 in (3þ 2)-dimensional superspace, Physical Review B 56 (1997) 13757–13767. [18] F. Clerc, C. Battaglia, H. Cercellier, C. Monney, H. Berger, L. Despont, M.G. Garnier, P. Aebi, Fermi surface of layered compounds and bulk charge density wave systems, Journal of Physics—Condensed Matter 19 (2007) 355002. [19] N. Dean, J.C. Petersen, D. Fausti, R.I. Tobey, S. Kaiser, L.V. Gasparov, H. Berger, A. Cavalleri, Polaronic conductivity in the photoinduced phase of 1T-TaS(2), Physical Review Letters 106 (2011) 016401. [20] S. Hellmann, M. Beye, C. Sohrt, T. Rohwer, F. Sorgenfrei, H. Redlin, M. Kallane, M. Marczynski-Buhlow, F. Hennies, M. Bauer, A. Fohlisch, L. Kipp, W. Wurth, K. Rossnagel, Ultrafast melting of a charge-density wave in the Mott insulator 1T-TaS(2), Physical Review Letters 105 (2010) 187401. [21] H. Schafer, V.V. Kabanov, M. Beyer, K. Biljakovic, J. Demsar, Disentanglement of the electronic and lattice parts of the order parameter in a 1D charge density wave system probed by femtosecond spectroscopy, Physical Review Letters 105 (2010) 066402. [22] L. Diebolt, W. Glaunsinger, M. McKelvy, Intercalation and deintercalation processes for ammoniated 1T-TaS2, Journal of Solid State Chemistry 145 (1999) 336–344. [23] A. Prodan, V. Marinkovic, F.W. Boswell, J.C. Bennett, M. Remskar, Chargedensity waves in some Nb and Ta chalcogenides, Journal of Alloys and Compounds 219 (1995) 69–72. [24] T. Ishiguro, H. Sato, Electron-microscopy of phase-transformations in 1T-Tas2, Physical Review B 44 (1991) 2046–2060.