1: \documentclass[12pt]{iopart}
2: % Uncomment next line if AMS fonts required
3: %\usepackage{iopams}
4: \usepackage{epsf}
5: \begin{document}
6:
7: \title{Rashba spin-orbit coupling and spin
8: precession in carbon nanotubes}
9:
10: \author{A De Martino and R Egger\footnote[3]{To whom
11: correspondence should be addressed (egger@thphy.uni-duesseldorf.de)} }
12:
13: \address{Institut f{\"u}r Theoretische Physik,
14: Heinrich-Heine-Universit{\"a}t, D-40225 D{\"u}sseldorf}
15:
16: \begin{abstract}
17: The Rashba spin-orbit coupling in carbon nanotubes and its
18: effect on spin-dependent transport properties are
19: analyzed theoretically.
20: We focus on clean non-interacting nanotubes with tunable
21: number of subbands $N$.
22: The peculiar band structure is shown to allow in principle for
23: Datta-Das oscillatory behavior in the tunneling magnetoresistance as a
24: function of gate voltage, despite the presence of multiple bands.
25: We discuss the conditions for observing Datta-Das oscillations in
26: carbon nanotubes.
27: \end{abstract}
28:
29: \pacs{72.25.-b, 72.80.Rj, 73.63.Fg}
30:
31: % Uncomment for Submitted to journal title message
32: \submitto{\JPCM}
33:
34: % Comment out if separate title page not required
35: \maketitle
36:
37: \section{Introduction}
38:
39: Spintronics in molecular conductors is a field attracting more
40: and more attention, both from fundamental physics as well as
41: from application-oriented material science \cite{zutic}.
42: Here the quantum-mechanical electronic spin is the central
43: object controlling transport properties. For a conductor sandwiched
44: between ferromagnetic leads, a different resistance can be observed
45: depending on the relative orientation of the lead magnetizations.
46: Quite often, the resistance is larger in the antiparallel configuration
47: than in the parallel one, but sometimes also the reverse situation can
48: be observed. It is useful to define the tunnel magnetoresistance
49: (TMR), $\rho_t= (R_{AP}-R_P)/R_P$, as the relative difference between
50: the corresponding resistances.
51:
52: A particularly interesting material in that context is provided
53: by carbon nanotubes (CNTs), see Refs.~\cite{forro,ando05} for
54: general reviews.
55: Quite a number of experimental studies concerning spin transport
56: through individual multi- (MWNT) or single-walled (SWNT) nanotubes
57: contacted by ferromagnetic leads have been reported over the past few years
58: \cite{alphenaar,schneider,kim,chakra,hoffer,kontos,sahoo}.
59: In particular, the experiments of the Basel group
60: \cite{kontos,sahoo} use
61: thin-film PdNi alloys as ferromagnetic leads in order
62: to contact either SWNTs or MWNTs,
63: where the shape anisotropy and the geometry of the setup
64: allow for the study of the spin-dependence of electrical transport. These
65: experiments have revealed
66: {\sl oscillatory} behavior of the TMR as a function of the external
67: gate voltage. Similar oscillations were predicted as a consequence
68: of the gate-voltage-tunable Rashba spin-orbit (SO) interaction
69: \cite{rashba,bychkov}
70: in a classic paper by Datta and Das some time ago \cite{datta}.
71: Since Datta-Das oscillations have still not been observed experimentally
72: so far, a thorough theoretical investigation of this effect
73: in nanotubes
74: is called for and provided here.
75: Unfortunately, from our analysis below, we find that the weakness of
76: SO couplings in nanotubes excludes an
77: interpretation of these data in terms of the Datta-Das effect
78: -- they can, however, be explained
79: in terms of quantum interference effects
80: \cite{sahoo}. Nevertheless, we show that the
81: presence of multiple bands in CNTs is not detrimental,
82: and under certain circumstances, the effect may be sufficiently enhanced
83: to be observable, e.g., by a tuning of the number of bands via external
84: gates along the lines of Ref.~\cite{strunk}.
85: In the original Datta-Das proposal \cite{datta}, subband mixing was
86: ignored so that different channels just add up coherently, but subband
87: mixing has later been argued to spoil the effect \cite{mireles,valin}. In CNTs,
88: the special band structure requires a careful re-examination of
89: the Datta-Das idea in this context, and we shall show that the
90: arguments of Refs.~\cite{mireles,valin} do not necessarily apply here.
91:
92: Recent theoretical studies of spin-dependent transport in CNTs
93: have mainly focused on the single-channel limit,
94: taking into account electron-electron interactions within
95: the framework of the Luttinger liquid theory
96: \cite{balents,ando,ourprl,jpcm,hausler2}
97: (see also \cite{hausler1,zulicke,gritsev}
98: for related discussions on interacting quantum wires
99: with Rashba SO coupling).
100: Here we confine ourselves to the noninteracting problem in
101: order to not overly complicate the analysis, but study
102: the many-band case and details of the band structure.
103: Interactions can be taken into account %straightforwardly
104: within the Luttinger liquid approach at a later stage,
105: and may enhance the effect of SO
106: couplings \cite{hausler1,chenraikh}.
107: We shall also neglect disorder
108: effects. Mean free paths in high-quality SWNTs typically
109: exceed $1 \mu$m, while in MWNTs this may be a more severe
110: approximation for some samples. However, high-quality MWNTs with
111: ultra-long mean free paths have also been reported recently
112: \cite{urbina}.
113:
114: The structure of this paper is as follows. In Sec.~\ref{sec2} we derive
115: the Rashba spin-orbit hamiltonian from microscopic considerations.
116: The resulting tight-binding SO
117: hamiltonian will be studied at low
118: energy scales in Sec.~\ref{sec3}, where we derive its
119: continuum form. In Sec.~\ref{sec4}, the consequences
120: with regard to Datta-Das oscillations in the TMR are analyzed.
121: We shall always consider the zero-temperature limit, and (in
122: most of the paper) put $\hbar=1$.
123:
124:
125: \section{Rashba spin-orbit coupling in nanotubes}\label{sec2}
126:
127: We start by noting that transport effectively proceeds through the outermost
128: shell of a MWNT only, such that we can take a single-shell model even
129: when dealing with a MWNT.
130: Experimentally and theoretically, it is understood
131: that such a model works very well in good-quality MWNTs \cite{forro},
132: essentially because only the outermost shell is electrically
133: contacted and tunneling between different shells is largely
134: suppressed \cite{bachtold,christian}.
135: Naturally, a single-shell description is also appropriate for SWNTs,
136: where we assume a sufficiently large radius $R$ such that occupation
137: of multiple subbands can be possible. (For a MWNT, $R$ denotes the
138: radius of the outermost shell.)
139: Depending on the electrochemical potential $\mu$ (doping level),
140: we then have to deal with $N$ spin-degenerate bands.
141: We assume full quantum coherence (no dephasing), so that the usual
142: Landauer-B\"uttiker approach applies, and
143: exclude external magnetic fields or electric field inhomogeneities, say,
144: due to the electrodes. We proceed to derive the Rashba
145: SO interaction, $H_{so}$, for this problem.
146: Notice that this is different from the intrinsic
147: atomic SO interaction discussed in Refs.~\cite{ando,chico}.
148: In particular, the SO coupling in Refs.~\cite{ando,chico}
149: vanishes in the limit of large radius, which is not the case
150: for the Rashba SO coupling we discuss below. Though Ando's SO
151: coupling \cite{ando} could straightforwardly be included in our analysis,
152: being gate-voltage independent it could not change our conclusions
153: relative to the gate-voltage dependent oscillations in the
154: magnetoresistance and is neglected in what follows.
155:
156:
157:
158: We first define a fixed reference frame
159: ${\cal S}= \{ \hat Y, \hat Z, \hat X \}$, with unit vector
160: $\hat X$ pointing in the axis direction and $\hat Z$ perpendicular to
161: the substrate on which the CNT is supposed to be located.
162: Next we introduce a second, local
163: reference frame ${\cal S}_i = \{ \hat \rho_i, \hat t_i,\hat X\} $
164: relative to %anchored at
165: each lattice site
166: $\vec R_i$ on the tube surface, where $\hat \rho_i$
167: and $\hat t_i$ are unit vectors along the local normal and tangential
168: (around the circumference) directions at $\vec R_i$, respectively.
169: Using polar coordinates in the plane transverse to the tube axis,
170: the relation between ${\cal S}$ and ${\cal S}_i$ is given by
171: \begin{equation}
172: \hat \rho_i = \cos \varphi_i \hat Y + \sin \varphi_i \hat Z , \quad
173: \hat t_i = -\sin \varphi_i \hat Y + \cos \varphi_i \hat Z .
174: \end{equation}
175: The position vector of a given carbon atom can then be written as
176: $\vec R_i = R\hat \rho_i + X_i \hat X$.
177: For later convenience, we introduce also
178: another reference frame.
179: For each pair of sites $\vec R_i$ and $\vec R_j$, we define
180: \begin{equation} \label{rij}
181: \vec R_{ij} =\vec R_i -\vec R_j \equiv X_{ij} \hat X +
182: \vec \rho_{ij},
183: \end{equation}
184: and denote the direction perpendicular to
185: $\hat \rho_{ij}$ and $\hat X$ as $\hat \rho^\perp_{ij}$. Then
186: $\{\hat \rho^\perp_{ij}, \hat \rho_{ij},\hat X \}$ constitutes a new local
187: frame ${\cal S}_{ij}$, and one has
188: \begin{eqnarray}
189: \hat \rho^\perp_{ij} &=& \cos [(\varphi_i+\varphi_j)/2] \, \hat Y +
190: \sin [(\varphi_i+\varphi_j)/2] \, \hat Z ,\\ \nonumber
191: \hat \rho_{ij} &=& -\sin [(\varphi_i+\varphi_j)/2] \, \hat Y +
192: \cos [(\varphi_i+\varphi_j)/2] \, \hat Z .
193: \end{eqnarray}
194: The $2p_z$ orbital at position $\vec R_i$ can then be represented as
195: \begin{equation}\label{2pz}
196: \chi_i(\vec r- \vec R_i) = \alpha
197: (\vec r -\vec R_i)\cdot \hat \rho_i
198: e^{-\beta |\vec r -\vec R_i|} ,
199: \end{equation}
200: where $4\alpha =(2\pi a_0^5)^{-1/2}$,
201: $\beta =(2a_0)^{-1}$, $a_0=\hbar^2/me^2=0.53 $\AA~is the Bohr radius,
202: and $m$ is the electron's mass.
203: We introduce an index $i$ on the orbital in order
204: to keep track of the atom at which it is centered.
205: The wavefunction (\ref{2pz}) is expected to be highly accurate
206: for not too small $R$, where hybridization with the $sp^2$
207: orbitals is negligible.
208:
209: At large distances from the tube, external gates
210: generally produce an electric field perpendicular
211: to the tube axis and the substrate.
212: As it has been shown in detail in previous works \cite{louie,saslow},
213: polarization effects of the CNT itself
214: due to a transverse field result in
215: a reduction of the externally applied
216: field described by
217: \[
218: E_0 = \frac{1}{1+2\alpha_{0yy}/R^2} E_{ext},
219: \]
220: where $\alpha_{0yy}$ is the unscreened transverse static
221: polarizability.
222: Since $\alpha_{0yy}$ is approximately proportional to $R^2$,
223: the factor in front of $E_{ext}$
224: practically equals a constant,
225: $\approx 0.2$ \cite{louie}.
226: Then, assuming homogeneity, the electric field due to
227: the gate can be written as
228: \begin{equation}\label{e0}
229: \vec {E}= E_0 \hat Z,
230: \end{equation}
231: which in turn produces the (first-quantized)
232: Rashba spin-orbit interaction \cite{rashba,bychkov}. With
233: standard Pauli matrices $\vec \sigma$ acting in spin space,
234: \begin{equation}\label{rashbac}
235: H_{so}= \frac{e\hbar}{4m^2c^2} \vec E \cdot ( {\vec\sigma} \times \vec p ) .
236: \end{equation}
237:
238: We proceed to derive the second-quantized spin-orbit hamiltonian
239: within the tight-binding approximation. For that purpose, we
240: need the matrix element of the momentum operator between two
241: $2p_z$ orbitals $\vec p_{ij} = \langle \chi_i |\vec p | \chi_j \rangle $,
242: from which we get
243: the following form for the SO hamiltonian:
244: \begin{equation}\label{so1}
245: H_{so} = g\sum_{ij} c_i^\dagger \left[
246: ( \vec \sigma \times \vec p_{ij} ) \cdot \hat Z \right]
247: c_j ,
248: \end{equation}
249: where the fermionic operator $c_{i\sigma}$
250: destroys an electron with spin $\sigma=\uparrow,\downarrow$
251: in the $2p_z$ orbital centered at $\vec R_i$, and
252: $g =E_0/4m^2c^2$.
253: For calculational convenience,
254: the matrix element $\vec p_{ij}$ can be written as
255: $g \vec p_{ij} = i(\vec v_{ij}+\vec u_{ij})$,
256: where the spin-orbit vectors $\vec v_{ij}$ and $\vec u_{ij}$
257: are defined as
258: \begin{eqnarray}
259: \label{sovector1}
260: \vec v_{ij} &=& -g \alpha
261: %\frac{-\hbar E_0 \alpha}{4m^2c^2}
262: \int d^3\vec r \; \chi_i(\vec r -\vec R_{i})
263: \hat \rho_j e^{-\beta |\vec r -\vec R_j|} ,
264: \\ \label{sovector2}
265: \vec u_{ij} &=& g \beta
266: %\frac{\hbar E_0 \beta}{4m^2c^2}
267: \int d^3\vec r \; \chi_i(\vec r -\vec R_{i})
268: \frac{\vec r -\vec R_j}{|\vec r -\vec R_j |} \chi_j (\vec r- \vec R_j) ,
269: \end{eqnarray}
270: Note that the modulus
271: of $\vec v_{ij}$ and $\vec u_{ij}$
272: has dimension of energy, and
273: their sum (but not necessarily each term separately)
274: is antisymmetric under exchange of $i$ and $j$.
275:
276: We first observe that the spin-orbit vectors connecting
277: a site with itself clearly vanish, since
278: $\langle \chi_i |\vec p| \chi_i \rangle =0$.
279: Let us then discuss spin-orbit vectors connecting different sites.
280: Since the orbitals (\ref{2pz}) decay exponentially, it is sufficient to
281: consider only the case of nearest neighbors. We start with $\vec v_{ij}$.
282: Shifting $\vec r \rightarrow \vec s + \vec R_i$ in Eq.~(\ref{sovector1})
283: and using Eq.~(\ref{2pz}), we obtain
284: \[
285: \vec v_{ij} = - g \alpha^2
286: \hat \rho_j
287: \int d^3\vec s \; (\vec s\cdot \hat \rho_i ) e^{-\beta s}
288: e^{-\beta |\vec s +\vec R_{ij}|} .
289: \]
290: Using $\vec s= s_\parallel \hat R_{ij} +
291: \vec s_\perp $, we then
292: rewrite the above integral as
293: \[
294: \int d^3\vec s \; (s_\parallel \hat R_{ij} +
295: \vec s_\perp ) \cdot \hat \rho_i \, e^{-\beta s}
296: e^{-\beta \sqrt{(s_\parallel + d)^2 + s_\perp^2}} ,
297: \]
298: where we use $|\vec R_{ij}|=d$, with the nearest-neighbor distance
299: among carbon atoms in graphene $d=1.42$~\AA.
300: Note that $\beta d= 1.34$.
301: The second term in the brackets is odd in $\vec s_\perp$ and thus vanishes,
302: and we obtain
303: \begin{equation}
304: \vec v_{ij} = - g \alpha^2 \hat \rho_j \frac{2R}{d} \sin^2
305: (\frac{\varphi_i-\varphi_j}{2}) d^4 \gamma_0 ,
306: \end{equation}
307: where we have used
308: $\hat R_{ij} \cdot \hat \rho_i = \frac{2R}{d}
309: \sin^2(\frac{\varphi_i-\varphi_j}{2} )$
310: and the dimensionless numerical factor $\gamma_0$ :
311: \[
312: \gamma_0 = \int dx\,dy\,dz \, x\,e^{-\beta d \sqrt{x^2+y^2+z^2}}
313: e^{-\beta d \sqrt{(x+d)^2+y^2+z^2}} .
314: \]
315: For $\vec v_{ji}$, we find
316: \[
317: \vec v_{ji} = g \alpha^2 \hat \rho_i \frac{2R}{d} \sin^2
318: (\frac{\varphi_i-\varphi_j}{2}) d^4 \gamma_0 .
319: \]
320: Notice that, up to higher orders in $d/R$, the unit
321: vectors $\hat \rho_{i,j}$
322: can be replaced by $\hat \rho^{\perp}_{ij}$, which makes clear that
323: $\vec v_{ij}$ is normal to the tube surface.
324: Now $|\sin[(\varphi_i-\varphi_j)/2]|$ varies between zero
325: (when the two sites are aligned in the axis direction) and
326: $d/2R\ll 1$ (when the two sites are aligned in the circumferential
327: direction).
328: Thus, to zeroth order in $d/R$, $\vec v_{ij}$ vanishes:
329: it is a pure curvature effect, peculiar of nanotubes,
330: which does not exist in graphene. In practice, $\vec v_{ij}$
331: is tiny and certainly subleading to $\vec u_{ij}$, which turns
332: out to be of order $(d/R)^0$. We shall therefore neglect it
333: in what follows.
334:
335:
336: Let us now turn to $\vec u_{ij}$. We shift
337: $\vec r \to \vec s +(\vec R_i+\vec R_j)/2$ in Eq.~(\ref{sovector2}),
338: and rewrite $\vec u_{ij}$ as the sum of two terms:
339: \begin{eqnarray}\label{int1}
340: \vec u^{(1)}_{ij} &=& g\beta \int
341: d^3\vec s \, \chi_i(\vec s -\vec R_{ij}/2) \,\chi_j(\vec s + \vec R_{ij}/2)
342: \frac{\vec s}{|\vec s + \vec R_{ij}/2|} , \\
343: \label{int2}
344: \vec u^{(2)}_{ij} &=& \frac{g \beta}{2}
345: %\frac{\hbar \beta E_0}{8 m^2 c^2}
346: \vec R_{ij} \int
347: d^3\vec s \, \chi_i(\vec s -\vec R_{ij}/2) \, \chi_j(\vec s + \vec R_{ij}/2)
348: \frac{1}{|\vec s + \vec R_{ij}/2|} .
349: \end{eqnarray}
350: Writing again $\vec s= s_\parallel \hat R_{ij} + \vec s_\perp$,
351: the computation of the above integrals leads,
352: to the lowest non-vanishing order in $d/R$, to the following
353: expressions:
354: \begin{eqnarray}\label{int3}
355: \vec u^{(1)}_{ij} &=& g\beta \alpha^2 \vec R_{ij}
356: d^4 \gamma_1 \equiv u_1 \vec R_{ij} ,
357: \\
358: \label{int4}
359: \vec u^{(2)}_{ij} &=& \frac{g \beta}{2} \alpha^2
360: \vec R_{ij} d^4 \gamma_2 \equiv u_2 \vec R_{ij} ,
361: \end{eqnarray}
362: with the dimensionless numerical factors
363: \begin{eqnarray} \label{gamma1}
364: \gamma_1 & = & \int dx\, dy\,dz
365: \frac{x z^2 e^{-\beta d\sqrt{(x-1/2)^2+y^2+z^2}}
366: e^{-\beta d\sqrt{(x+1/2)^2+y^2+z^2}}}
367: {\sqrt{(x+1/2)^2+y^2+z^2}} \\ \nonumber &\simeq&-0.0375,
368: \end{eqnarray}
369: and
370: \begin{eqnarray} \label{gamma2}
371: \gamma_2 &=& \int dx\, dy\,dz \frac{ z^2 e^{-\beta d \sqrt{(x-1/2)^2+y^2+z^2}}
372: e^{-\beta d\sqrt{(x+1/2)^2+y^2+z^2}}} {\sqrt{(x+1/2)^2+y^2+z^2}} \\
373: &\simeq & 0.3748. \nonumber
374: \end{eqnarray}
375: To lowest order in $d/R$, it does not make a difference
376: whether we take the tangent unit vector
377: at $\vec R_j$, $\vec R_i$, or at $(\vec R_i+\vec R_j)/2$.
378: Hence we may write
379: $\hat \rho_{ij} \to \hat e_\varphi$, where $\hat e_\varphi$
380: is the unit tangent vector at $(\vec R_i +R_j)/2$.
381: We then get SO couplings along the axial and along the
382: circumferential direction,
383: \begin{equation}\label{su}
384: \vec u_{ij} = u \left[ (\vec R_{ij} \cdot \hat X) \hat X +
385: (\vec R_{ij} \cdot \hat e_\varphi) \hat e_\varphi \right] ,
386: \end{equation}
387: with $u=u_1+u_2$. Note that
388: we have neglected a tiny component
389: of $\vec R_{ij}$ normal to the tube surface.
390: The above discussion then results in the tight-binding
391: hamiltonian $H = H_0 + H_{so}$, where
392: \[
393: H_0 = - t \sum_{\vec r, a} c^\dagger_{B,\vec r+\vec \delta_a }
394: c^{}_{A,\vec r} + h.c.,
395: \]
396: with $t\approx 2.7$~eV \cite{ando05}. Here the $\vec r$ denote all
397: sublattice-A tight-binding sites of the lattice. Furthermore,
398: the $\vec \delta_{a=1,2,3}$ are vectors connecting
399: $\vec r$ with the three nearest-neighbor sites which are all located
400: on sublattice B \cite{ando05}.
401: Since we consider the limit $d/R\ll 1$,
402: the $\vec \delta_a$ at each site effectively lie in the
403: tangent plane to the tube surface at that site.
404: The Rashba spin-orbit hamiltonian then reads
405: \begin{equation}
406: \label{Hso}
407: H_{so} = i u \sum_{\vec r, a} c^\dagger_{B,\vec r+\vec \delta_a }
408: \Bigl[ \left(
409: \vec \sigma \times
410: [ (\vec \delta_a \cdot \hat X) \hat X +
411: (\vec \delta_a \cdot \hat e_\varphi )
412: \hat e_\varphi ] \right)
413: \cdot \hat Z \Bigr]
414: c^{}_{A,\vec r} +h.c.
415: \nonumber
416: \end{equation}
417:
418: \section{Continuum limit}\label{sec3}
419:
420: Since we are interested in the low-energy long-wavelength properties,
421: we now expand the electron operator around
422: the Fermi points $K,K'$ in terms of Bloch waves \cite{ando05},
423: \begin{equation}
424: \frac{c_{p\vec r}}{\sqrt{S}} = e^{i\vec K \cdot \vec r} F_{1p}(\vec r) +
425: e^{- i\vec K \cdot \vec r} F_{2p}(\vec r) ,
426: \end{equation}
427: where $S= \sqrt{3}a^2/2$ is the area of the unit cell, $a=\sqrt{3} d$,
428: and $p=A/B$ is the sublattice index. The $F_{\alpha p}$ are
429: slowly varying electron field operators, and we choose the Fermi points at
430: $\vec K = (4\pi/3a,0)$ and $\vec K' = -\vec K$ \cite{ando05}. We
431: then expand $F(\vec r + \vec \delta ) \simeq F(\vec r)
432: + \vec \delta \cdot \nabla F(\vec r)$ and use the bond vectors
433: \begin{equation} \label{delta}
434: \vec \delta_1 = \frac{a}{\sqrt{3}}(0,-1), \quad\quad
435: \vec \delta_2 = \frac{a}{2}(1, 1/\sqrt{3}), \quad\quad
436: \vec \delta_3 = \frac{a}{2}(-1,1/\sqrt{3}).
437: \end{equation}
438: These vectors are given in a fixed reference frame for a 2D graphene
439: sheet, and
440: we then must perform a rotation to longitudinal
441: and circumferential directions via the chiral angle. This rotation
442: results in fixed phases that can be absorbed in the definition of
443: $F_{\alpha p}$ and do not appear in final results. This is of course
444: expected from the $U(1)$ symmetry emerging at low energies in the dispersion
445: relation of graphene \cite{ando05}.
446: After some algebra, the usual Dirac hamiltonian for the kinetic term follows,
447: \begin{equation}
448: H_0 = v \int \! d^2\vec r \; F^\dagger \left[
449: (T_0\otimes\tau_2\otimes\sigma_0)(-i\partial_x) +
450: (T_3\otimes \tau_1\otimes\sigma_0)(-i\partial_y)
451: \right] F ,
452: \end{equation}
453: where $v=\sqrt{3}at/2\simeq 8\times 10^5$~m/sec is the Fermi velocity,
454: and $x,y$ are longitudinal and circumferential coordinates,
455: respectively, with $0<y\leq 2\pi R$. Finally,
456: $T_i$ and $\tau_i$ are also Pauli matrices that now act in
457: the space of Fermi ($K,K'$) points and sublattice space ($A,B$), respectively.
458: For $i=0$, these are defined as $2\times 2$ unit matrices.
459:
460: The low-energy limit of the SO term (\ref{Hso}) can be obtained in the
461: following way. First we observe that
462: \[
463: \left[ (\vec \delta_a \cdot \hat X)
464: \hat X +
465: (\vec \delta_a \cdot \hat e_\varphi ) \hat e_\varphi \right] \times \hat Z =
466: - (\vec \delta_a\cdot \hat X) \hat Y
467: - \sin (y/R) (\vec \delta_a \cdot \hat e_\varphi ) \hat X .
468: \]
469: Here the only approximation is the assumption that the bond vectors lie in
470: the plane tangent to the nanotube surface at $\vec r$.
471: Second, by using the bond vectors (\ref{delta}) and taking into account
472: the chiral angle $\eta$ between the fixed direction on the
473: graphite sheet and the circumferential direction on the nanotube,
474: one obtains
475: \begin{eqnarray}
476: \sum_a c^\dagger_{\vec r + \vec \delta_a}
477: \left( \vec \delta_a \cdot \hat X \right)
478: c_{\vec r}
479: & \approx & \frac{-3d}{2} \left(
480: F^\dagger_{B1} e^{-i\eta}
481: F_{A1} +
482: F^\dagger_{B2} e^{i\eta}
483: F_{A2} \right) , \nonumber \\
484: \sum_a c^\dagger_{\vec r + \vec \delta_a}
485: \left( \vec \delta_a \cdot \hat e_\varphi \right)
486: c_{\vec r}
487: & \approx & \frac{-3d}{2}
488: \left( i F^\dagger_{B1} e^{-i\eta} F_{A1}
489: -i F^\dagger_{B2} e^{i\eta} F_{A2} \right) . \nonumber
490: \end{eqnarray}
491: Notice that we take into account exactly the relative orientation
492: of the bond vectors with respect to the directions $\hat X$ and
493: $\hat e_\varphi$ for a generic nanotube, which is encoded in the
494: chiral angle $\eta$.
495: The constant phases $e^{\pm i \eta}$ can be absorbed by appropriately
496: redefining the operators as $F_{A2}\rightarrow e^{-i\eta } F_{A2}$ and
497: $F_{B1}\rightarrow e^{-i\eta } F_{B1}$,
498: and the final result can be written down in the form
499: \begin{equation}\label{hsso}
500: H_{so} = \int \! d^2\vec r \; F^\dagger \left[
501: u_\parallel T_0\otimes\tau_1\otimes \sigma_2
502: + u_\perp \sin(y/R) T_3\otimes\tau_2 \otimes \sigma_1 \right] F ,
503: \end{equation}
504: with $u_{\parallel}=u_\perp = 3 d u/2$.
505: For the sake of generality, we continue to use different
506: coupling constants $u_\perp$ and $u_\parallel$.
507: It is worthwhile to mention that the leading term for the
508: Rashba spin-orbit coupling in a CNT, Eq.~(\ref{hsso}),
509: does not depend on longitudinal momentum.
510: This is due to the peculiar band structure of graphene with its isolated
511: Fermi ($K$) points. In the above derivation, we also find terms
512: that are linear in momentum, i.e., contain spatial
513: derivatives of the electron operators.
514: Such terms only produce
515: %highly irrelevant (in the renormalization
516: %group sense) at low energies and
517: tiny renormalizations of the velocities and
518: will be neglected here.
519: The second term in Eq.~(\ref{hsso}) allows for spin flips and
520: mixes transverse subbands.
521:
522: {}From now on, for simplicity, we consider just a single
523: Fermi point, say, $K$. After the global $SU(2)$ rotation
524: $\sigma_1\to \sigma_2 \to \sigma_3$ in spin space,
525: we get in compact notation
526: \begin{eqnarray} \label{unperham}
527: {\cal H}_0 &=& v \left[ -i \tau_1 \partial_y - i \tau_2 \partial_x \right] .
528: \\ \label{soham}
529: {\cal H}_{so} &=&
530: u_\parallel \tau_1 \sigma_3 + u_\perp \sin(y/R) \tau_2 \sigma_2 .
531: \end{eqnarray}
532: Note that the exact spectrum of ${\cal H}_0 + {\cal H}_{so}$ with $u_\perp=0$
533: can be obtained straightforwardly. In general, however,
534: due to the smallness of the SO coupling (see below), it is enough to treat
535: ${\cal H}_{so}$ perturbatively.
536: The following detailed derivation is then necessary
537: to correctly evaluate the effect of the SO coupling, and moreover
538: it is interesting and important for the generalization to the
539: interacting case, and for the analysis of features involving the
540: electron wavefunction (as for instance electron-phonon interactions).
541:
542: The eigenvalues of ${\cal H}_0$ are given by
543: \begin{equation} \label{eigzero}
544: \epsilon_{an\sigma}(q) = av \sqrt{k^2_\perp(n) +q^2}\equiv a\epsilon_n(q) ,
545: \end{equation}
546: where $k_\perp(n) = (n+n_0)/R$ denotes the transverse momentum,
547: $q$ the longitudinal one, $a=\pm$ labels the conduction/valence band,
548: and $\sigma=\pm$ the spin.
549: Here $n_0=0$ for intrinsically metallic shells, but
550: generally it can be taken as $0\leq n_0 \leq 1/2$ to take into account
551: chirality gaps or orbital magnetic fields along $\hat X$.
552: The transverse subbands are labeled by integer values
553: $n=0,1,2,\ldots,{\cal N}-1$, where
554: ${\cal N}= 2(N^2+M^2+NM)/{\rm gcd}(2M+N,2N+M)$ for ($N,M$) tubes
555: \cite{ando05}.
556: ${\cal N}$ is typically much larger than the actual number
557: $N=[k_F R]$ of occupied subbands, where we define $k_F=\mu/v$ with
558: the doping level $\mu$ that we assume positive here.
559: The velocity $v_n$ for electrons in subband $n$ at the Fermi level
560: (in the absence of $H_{so}$)
561: and the corresponding Fermi momentum $q_n$ are then given by
562: \begin{equation}\label{vnqn}
563: v_n=v\sqrt{1-[(n+n_0)/(k_F R)]^2},\quad
564: q_n=k_F v_n/v.
565: \end{equation}
566:
567: The eigenvalues (\ref{eigzero}) are spin-independent
568: and thus doubly degenerate.
569: The corresponding eigenstates are denoted $| n q a \sigma \rangle$,
570: where $|n\rangle$ and $|q \rangle$ are respectively plane waves
571: in circumferential and longitudinal direction.
572: In coordinate representation they read
573: \begin{equation}
574: \psi_{nqa\sigma}(x,y)\equiv \langle x,y| n q a \sigma \rangle =
575: \frac{e^{ik_\perp(n)y}}{\sqrt{2\pi R}} e^{iqx} \xi_{na}(q) \otimes
576: \chi_\sigma ,
577: \end{equation}
578: with the bispinor (in sublattice space)
579: \begin{equation}\label{bispinor2}
580: \xi_{n, a=\pm}(q) = \frac{1}{\sqrt{2}} \left(\begin{array}{c}
581: e^{i \theta_n(q)/2} \\
582: \pm e^{- i\theta_n(q)/2} \end{array} \right) , \quad
583: e^{i \theta_n(q)} = \frac{v (k_\perp (n) - i q ) }{ \epsilon_n(q)} .
584: \end{equation}
585: A different, and here more convenient basis is given by the sublattice
586: states $| n q p \sigma \rangle$. Their coordinate representation is
587: \begin{equation}
588: \psi_{nqp\sigma}(x,y) =
589: \frac{e^{ik_\perp(n)y}}{\sqrt{2\pi R}} e^{iqx} \xi_p \otimes \chi_\sigma ,
590: \end{equation}
591: where $p=A,B$ and
592: \[
593: \xi_A = \left( \begin{array}{c} 1 \\ 0 \end{array} \right) , \quad
594: \xi_B = \left( \begin{array}{c} 0 \\ 1 \end{array} \right) .
595: \]
596: Their usefulness stems from the fact that the $| n q p \sigma \rangle$
597: can be factorized as
598: \begin{equation}\label{factor}
599: | n q p \sigma \rangle = |n\rangle |q \rangle \otimes | p \sigma \rangle,
600: \quad | p \sigma \rangle = \xi_p \otimes \chi_\sigma,
601: \end{equation}
602: where $|p \sigma \rangle$ is independent of $n$ and $q$.
603: Using this basis,
604: we can expand the field operator $F(\vec r)$ on the tube surface as
605: \begin{equation}
606: F(\vec r) = \sum_{n,p,\sigma}
607: \int \frac {dq}{2\pi} \psi_{nqp\sigma}(x,y) c_{np\sigma}(q)
608: = \sum_{n} F_n(x) \langle y| n \rangle,
609: \end{equation}
610: where the operator $c_{np\sigma}(q)$ destroys an electron in
611: the state $|nqp\sigma \rangle$, and we introduce the 1D field operators
612: $F_n(x)$. Alternatively, using the basis of eigenstates of $H_0$,
613: $F(\vec r)$ can be expanded as
614: \begin{equation}
615: F(\vec r) = \sum_{n,a,\sigma}
616: \int \frac {dq}{2\pi} \psi_{nqa\sigma}(x,y) c_{na\sigma}(q) ,
617: \end{equation}
618: where the operators $c_{na\sigma}(q)$ destroy conduction ($a=+$)
619: or valence ($a=-$) electrons with spin $\sigma$ in subband $n$.
620: Notice that in what follows the spin index is left implicit.
621: The relation between the operators $c_{na}$ and $c_{np}$ is
622: easily found to be
623: \begin{equation}
624: \left( \begin{array}{c} c_{n+}(q) \\ c_{n-}(q) \end{array} \right)=
625: \frac{1}{\sqrt{2}} \left( \begin{array}{cc} e^{- i\theta_n(q)/2} &
626: e^{i\theta_n(q)/2}\\ e^{-i\theta_n(q)/2} & -e^{i\theta_n(q)/2}
627: \end{array} \right)
628: \left( \begin{array}{c} c_{nA}(q) \\ c_{nB}(q) \end{array}\right) .
629: \end{equation}
630:
631: We now proceed by treating
632: the spin-orbit hamiltonian using perturbation theory.
633: %, which is
634: %justified due to the smallness of the couplings.
635: First, we diagonalize $H_0-\mu N$ for a fixed transverse subband $n$,
636: \begin{eqnarray*}
637: H^{(n)}_0 - \mu N^{(n)} &=& v \int dx \,
638: F^\dagger_n [k_\perp(n) \tau_1 + (-i\partial_x) \tau_2 - \mu ] F_n \\
639: &=& \sum_{a=\pm} \int \frac{dq}{2\pi} [a\epsilon_n(q) - \mu] c^\dagger_{na}
640: c^{}_{na}.
641: \end{eqnarray*}
642: Next we expand around the Fermi points $\pm q_n$ defined
643: in Eq.~(\ref{vnqn}), which
644: introduces right- and left-movers, $r=\pm=R/L$, as the relevant
645: low-energy degrees of freedom. For small deviations $k$ from $\pm q_n$,
646: Taylor expansion yields $\epsilon_n(\pm q_n +k) \simeq \mu \pm v_n k$,
647: where $v_n$ is given in Eq.~(\ref{vnqn}).
648: Since we assumed $\mu>0$, we may now restrict ourselves to the
649: conduction band, $a=+$. For the hamiltonian, we then obtain
650: \begin{eqnarray*}
651: H^{(n)}_0 -\mu N^{(n)} & = &
652: \sum_{r=\pm} v_n \int \frac{dk}{2\pi} ( r k) c^\dagger_{nr}(k)
653: c^{}_{nr}(k)\\ &=& \sum_{r=\pm} v_n \int dx \; \psi^\dagger_{nr} (-ir\partial_x)
654: \psi^{}_{nr} ,
655: \end{eqnarray*}
656: where $c_{nr}(k)\equiv c_{n+}(rq_n+k)$ and
657: $\psi_{nr}(x) = \int \frac{dk}{2\pi} e^{ikx} c_{nr}(k)$.
658: This introduces $R/L$-moving 1D fermion operators for each subband $n$
659: (and spin $\sigma$).
660: The relation of these 1D fermions with the original
661: operator $F_n(x)$ is given by
662: \begin{eqnarray}\label{fn}
663: F_n(x) &=& e^{iq_n x} \int \frac{dk}{2\pi} \frac{e^{ikx}}{\sqrt{2}}
664: \left(\begin{array}{c} e^{i\theta_n(q_n)/2} \\ e^{-i\theta_n(q_n)/2}
665: \end{array} \right)
666: c_{nR}(k) \nonumber\\
667: & + & e^{- iq_n x} \int \frac{dk}{2\pi} \frac{e^{ikx}}{\sqrt{2}}
668: \left(\begin{array}{c} e^{-i\theta_n(q_n)/2} \\ e^{i\theta_n(q_n)/2}
669: \end{array} \right) c_{nL}(k) .
670: \end{eqnarray}
671: Notice that, while in general the unitary
672: transformation from sublattice space to the conduction/valence band
673: description depends on longitudinal momentum, in the continuum limit,
674: one can use the transformation directly at the Fermi momenta.
675: This is consistent with the neglect of band curvature effects
676: implicit in the linearization
677: of the dispersion relation, which
678: is unproblematic away from van Hove singularities associated with
679: the onset of new subbands \cite{hugle}. At these points, the concept
680: of $R/L$-movers breaks down, and some of our conclusions below may change.
681:
682: Next we express the Rashba hamiltonian (\ref{soham})
683: in terms of $R/L$-movers.
684: The first term results in
685: \begin{equation}\label{rasb}
686: H^{\parallel}_{so} = \frac{ u_\parallel v}{\mu} \sum_{nr}
687: k_\perp (n)
688: \int \frac{dk}{2\pi}
689: c^\dagger_{nr}(k) \sigma_3 c^{}_{nr}(k) .
690: \end{equation}
691: The presence of the factor $k_\perp(n)$ results from
692: a careful treatment of the phases in Eq.~(\ref{fn}).
693: In Eq.~(\ref{rasb}) we omit an additional term
694: mixing right- and left-movers. This term contains a
695: rapidly oscillating factor $e^{\pm 2 i q_nx}$ and
696: therefore is strongly suppressed by momentum conservation.
697: The second term in Eq.~(\ref{soham})
698: again contains the oscillating phase factor
699: $e^{\pm i(q_n \pm q_{n+1})x}$, which leads to a drastic suppression of
700: $H^\perp_{so}$ at low energies and long wavelengths.
701: Of course, this argument relies
702: in an essential way on the smallness of the coupling $u_\perp$,
703: as one expands around the hamiltonian $H_0$.
704: We conclude that away from van Hove singularities,
705: the only important Rashba term is given by $H^\parallel_{so}$
706: in Eq.~(\ref{rasb}). This term has the appearance of a static
707: homogeneous but {\sl channel-dependent magnetic field}.
708:
709: \section{Oscillatory TMR effects in nanotubes}\label{sec4}
710:
711: In this section we will analyze the consequences of our findings
712: regarding spin-orbit couplings in CNTs,
713: see Eqs.~(\ref{rasb}), % and (\ref{local2}),
714: for the observability of spin precession effects encoded in the
715: Datta-Das oscillations of the TMR. Based on
716: our expressions, it is possible to estimate the
717: order of magnitude of this effect.
718:
719: For a concrete estimate, let us put $E_0=0.2eV_G/(\kappa D)$,
720: where $D$ is the gate-tube distance, $V_G$ the gate voltage,
721: and $\kappa$ denotes the dielectric constant of the substrate.
722: For a given channel $n$, the Rashba-induced energy splitting is then
723: easily estimated as
724: \[
725: \frac{\Delta E_n}{eV_G}=
726: (\gamma_1 + \gamma_2) \frac{0.6dv}{\mu} \frac{|n+n_0|}{R}
727: \frac{\alpha^2 \beta d^4\lambda_c^2}{4\kappa D},
728: \]
729: where $\lambda_c=\hbar/mc= 3.86\times 10^{-13}$~m is the Compton length.
730: Plugging in the definition of $\alpha,\beta$, we get
731: \begin{equation}
732: \label{splitting}
733: \frac{\Delta E_n}{eV_G} =\frac{0.6(\gamma_1+\gamma_2)}{256\pi \kappa} (d/a_0)^5
734: \frac{\lambda_c^2}{D a_0} \frac{ |n+n_0| }{k_F R}.
735: \end{equation}
736: Bands with small $n$ are only weakly split, and hence do
737: not contribute to oscillatory TMR behavior.
738: This argument suggests that Datta-Das oscillations in
739: principle could survive in a CNT, even when there are
740: many channels. The major
741: contribution will come just from the few bands with the largest $n$.
742:
743: To estimate the accumulated phase difference due to the different
744: precession length of the two split eigenstates,
745: let us put $(n+n_0)/(k_F R)\to 1$, which represents the dominant contribution,
746: and set $\kappa=1$. Then Eq.~(\ref{splitting})
747: gives as order-of-magnitude estimate
748: \begin{equation}
749: \Delta E/(eV_G) \approx 2 \times 10^{-6} \ a_0/D.
750: \end{equation}
751: Even when assuming a very close-by gate,
752: this gives only a tiny splitting, in retrospect justifying perturbation
753: theory. This splitting now translates into a
754: momentum splitting $\Delta k_n= \Delta E/v_n$,
755: and hence into a precession phase mismatch along the CNT of length
756: $L$ \cite{datta}. For the $n$th band, this phase difference is
757: \begin{equation}\label{phase}
758: \Delta \phi_n= \Delta k_n L\approx 2 \times 10^{-6}
759: \frac{L}{D} \frac{eV_G}{\hbar v_n/a_0}.
760: \end{equation}
761: This phase difference should be of order $2\pi$ to allow for
762: the observation of Datta-Das oscillatory TMR effects \cite{datta}.
763:
764: Away from a van Hove singularity, Eq.~(\ref{phase}) predicts
765: that oscillations appear on a gate voltage scale of the order
766: of $10^6$ to $10^7$~V for $L\approx D$, which would make Datta-Das
767: oscillations unobservable.
768: This argument also shows that this interpretation can be ruled
769: out for the parameters relevant for the Basel experiment \cite{sahoo}.
770: {}From Eq.~(\ref{phase}),
771: we can then suggest several ways to improve the situation.
772: First, one should use very long CNTs, while at the same time keeping
773: the gate very close, and second, an enhancement can be
774: expected close to van Hove singularities. Of course, very close
775: to a van Hove singularity, some of our arguments above break down,
776: but the general tendency can nevertheless be read off from Eq.~(\ref{phase}).
777: Furthermore, electron-electron interactions
778: can also enhance spin-orbit effects \cite{hausler1,chenraikh}.
779:
780: To conclude, we have presented a detailed microscopic derivation
781: of Rashba spin-orbit coupling in carbon nanotubes. It turns out that
782: the Rashba SO coupling is small,
783: and therefore the prospects for observing spin-precession effects
784: like Datta-Das oscillations in the tunneling magnetoresistance
785: are not too favorable. However, for very long CNTs, close-by gates,
786: and in the vicinity of a van Hove singularity, the requirements
787: for observability of these effects could be met in practice.
788:
789:
790: \ack
791: We thank T. Kontos and C. Sch\"onenberger for motivating this study.
792: Support by the DFG (Gerhard-Hess program) and by the EU (DIENOW network)
793: is acknowledged.
794:
795: \section*{References}
796:
797: \begin{thebibliography}{99}
798:
799: \bibitem{zutic} Zutic I, Fabian J and Das Darma S 2004
800: {\it Rev. Mod. Phys.} {\bf 76} 323
801:
802: \bibitem{forro}
803: Forr{\'o} L and Sch\"onenberger C 2001
804: {\it Topics in Appl. Phys.} {\bf 80} 1
805:
806: \bibitem{ando05} Ando T 2005
807: {\it J. Phys. Soc. Jpn.} {\bf 74} 777
808:
809: \bibitem{alphenaar} Tsukagoshi K, Alphenaar B W and Ago H 1999
810: {\it Nature} {\bf 401} 572
811:
812: \bibitem{schneider} Zhao B, M\"onch I, Vinzelberg H, M\"uhl T
813: and Schneider C M 2002
814: {\it Appl. Phys. Lett.} {\bf 80} 3144
815:
816: \bibitem{kim}
817: Kim J R, So H M, Kim J J and Kim J 2002
818: {\sl Phys. Rev B} {\bf 66} 233401
819:
820: \bibitem{chakra}
821: Chakraborty S, Walsh K M, Alphenaar B W, Liu L and
822: Tsukagoshi K 2003
823: {\it Appl. Phys. Lett.} {\bf 83} 1008
824:
825: \bibitem{hoffer}
826: Hoffer X, Klinke C, Bonard B M, Gravier L and Wegrowe J E 2004
827: {\it Europhys. Lett.} {\bf 67} 103
828:
829: \bibitem{kontos}
830: Sahoo S, Kontos T, Sch\"onenberger C and S\"urgers S 2005
831: {\it Appl. Phys. Lett.} {\bf 86} 112109
832:
833: \bibitem{sahoo} Sahoo S, Kontos T, Furer J, Hoffmann C and
834: Sch\"onenberger C 2005 (unpublished preprint)
835:
836: \bibitem{rashba}
837: Rashba E I 1960
838: {\it Fiz. Tverd. Tela} {\bf 2} 1224
839: [{\it Sov. Phys. Solid State} {\bf 2} 1109]
840:
841: \bibitem{bychkov}
842: Bychkov Y and Rashba E I 1984
843: {\it J. Phys. C} {\bf 17} 6039
844:
845: \bibitem{datta} Datta S and Das B 1990
846: {\it Appl. Phys. Lett.} {\bf 56} 665
847:
848: \bibitem{strunk}
849: Stojetz B, Miko C, Forr{\'o} L and Strunk C 2005
850: {\it Phys. Rev. Lett.} {\bf 94} 186802
851:
852: \bibitem{mireles} Mireles F and Kirczenow G 2001
853: {\it Phys. Rev. B} {\bf 64} 024426
854:
855: \bibitem{valin}
856: Valin-Rodriguez M, Puente A and Serra L 2003
857: {\it Eur. Phys. J. B} {\bf 34} 359
858:
859: \bibitem{balents} Balents L and Egger R 2001
860: {\it Phys. Rev. B} {\bf 64} 035310
861:
862: \bibitem{ando}
863: Ando T 2000
864: {\it J. Phys. Soc. Jpn.} {\bf 69} 1757
865:
866: \bibitem{ourprl} De Martino A, Egger R, Hallberg K and Balseiro C A 2002
867: {\it Phys. Rev. Lett.} {\bf 88} 206402
868:
869: \bibitem{jpcm} De Martino A, Egger R, Murphy-Armando F and
870: Hallberg K 2004
871: {\it J. Phys.: Cond. Matt.} {\bf 16}, S1437
872:
873: \bibitem{hausler2}
874: H\"ausler W 2004
875: {\it Phys. Rev. B} {\bf 70}, 115313
876:
877: \bibitem{hausler1}H\"ausler W 2001,
878: {\it Phys. Rev. B} {\bf 63} 121310
879:
880: \bibitem{zulicke} Governale M and Z\"ulicke U 2002
881: {\it Phys. Rev. B} {\bf 66} 073311
882:
883: \bibitem{gritsev}
884: Gritsev V, Japaridze G, Pletyukhov M and Baeriswyl D 2005
885: {\it Phys. Rev. Lett.} {\bf 94} 137207
886:
887: \bibitem{chenraikh} Chen G H and Raikh M E 1999
888: {\it Phys. Rev. B } {\bf 60} 4826
889:
890: \bibitem{urbina}
891: Urbina A, Echeverr{\'i}a I, P{\'e}rez-Garrido A, D{\'i}az-S{\'a}nchez
892: A and Abell{\'a}n J 2003
893: {\it Phys. Rev. Lett.} {\bf 90} 106603
894:
895: \bibitem{bachtold} Bachtold A, Strunk C, Salvetat J P, Bonard J M, Forr{\'o}
896: L, Nussbaumer T and Sch\"onenberger C 1999
897: {\it Nature} {\bf 397} 673
898:
899: \bibitem{christian}
900: Sch\"onenberger C, Bachtold A, Strunk C, Salvetat J P and
901: Forr{\'o} L 1999
902: {\it Appl. Phys. A} {\bf 69} 283
903:
904: \bibitem{chico}
905: Chico L, L\'opez M P and Mu\~noz M C 2004
906: {\it Phys. Rev. Lett. \bf 93} 176402
907:
908:
909: \bibitem{louie} Benedict L X, Louie S G and Cohen M L 1995
910: {\it Phys. Rev. B} {\bf 52} 8541
911:
912: \bibitem{saslow} Kr\v{c}mar M, Saslow W and Zangwill A 2003
913: {\it J. Appl. Phys.} {\bf 93} 3495
914:
915: \bibitem{hugle} H\"ugle S and Egger R 2002
916: {\it Phys. Rev. B} {\bf 66} 193311
917:
918: \end{thebibliography}
919:
920: \end{document}
921:
922:
923: