cond-mat0405312/scu.tex
1: %last correction: September 1, 2004
2: %
3: \tolerance = 10000
4: \documentclass[epj]{svjour}
5: \usepackage{amsmath,amssymb}
6: \usepackage{bm}
7: \usepackage{epsfig}
8: 
9: % set \bd to \bf or \bm
10: \newcommand{\bd}{\bm}
11: 
12: \begin{document}
13: 
14: \title{What are spin currents in Heisenberg magnets?}
15: 
16: \author{Florian Sch\"{u}tz\inst{1}\and Peter Kopietz\inst{1}\and Marcus Kollar\inst{2}}
17: 
18: \institute{Institut f\"{u}r Theoretische Physik, Universit\"{a}t
19:   Frankfurt, Robert-Mayer-Strasse 8, 60054 Frankfurt, Germany
20:   \and
21:   Theoretische Physik III, Elektronische Korrelationen und
22:   Magnetismus, Institut f\"{u}r Physik, Universit\"{a}t Augsburg,
23:   86135 Augsburg, Germany}
24: 
25: \date{May 14, 2004}
26: 
27: \abstract{ 
28:   We discuss the proper definition of the spin current operator in
29:   Heisenberg magnets subject to inhomogeneous magnetic fields.  We
30:   argue that only the component of the naive ``current operator''
31:   $J_{ij}{\bd S}_i\times{\bd S}_j$ in the plane spanned by the local
32:   order parameters $\langle{\bd S}_i\rangle$ and $\langle{\bd
33:     S}_j\rangle$ is related to real transport of magnetization.
34:   Within a mean field approximation or in the classical ground state
35:   the spin current therefore vanishes.  Thus, finite spin currents are
36:   a direct manifestation of quantum correlations in the system.
37: }
38: 
39: \PACS{{75.10.Jm}{Quantized spin models} \and
40:   {75.10.Pq}{Spin chain models} \and 
41:   {75.30.Ds}{Spin waves} \and
42:   {73.23.Ra}{Persistent currents}}
43: 
44: \maketitle
45: 
46: \section{Introduction}
47: 
48: In a recent Letter \cite{Schuetz03} and a subsequent paper
49: \cite{Schuetz04} we have calculated the persistent spin currents in
50: mesoscopic Heisenberg rings subject to inhomogeneous magnetic fields.
51: We have emphasized the close analogy between this phenomenon and the
52: well known persistent charge currents in mesoscopic metal rings
53: pierced by an Aharonov-Bohm flux.  In the ensuing discussions with
54: several colleagues we have become aware of the fact that the
55: definition of the spin current operator in Heisenberg magnets subject
56: to inhomogeneous magnetic fields is not obvious.  In this note we
57: shall attempt to clarify this point.  
58: 
59: A related problem, which will not be discussed in this work, is the
60: definition of the spin current operator in semi-conducting electronic
61: systems with strong spin-orbit interactions. Recently, Rashba
62: \cite{Rashba03} pointed out that also for this case the precise
63: meaning of the concept of a spin current is rather subtle.  In
64: particular, he emphasizes that spin currents in thermodynamic
65: equilibrium, which arise with the standard definition of the
66: spin-current operator used in the literature, are unphysical and
67: should be regarded as background currents which do not correspond to
68: real transport of magnetization.  A clear understanding of this
69: concept is essential for the highly active field of information
70: processing using spin degrees of freedom subsumed under the name of
71: spintronics \cite{Awschalom02}.
72:   
73: For itinerant systems the spin is an intrinsic property of the charge
74: carriers and is carried around with their motion. For localized spin
75: systems considered here, transport of spin is a consequence of the
76: time evolution of the magnetization. For special cases the transport
77: can be ascribed to the movement of quasi-particles as magnons or
78: spinons and again a simple physical picture emerges \cite{Meier02}.
79:   
80: In this context, it is also interesting to note that in effective
81: low-energy models for ferromagnets involving only the spin degrees of
82: freedom even the concept of the linear momentum is not well defined
83: \cite{Volovik87}. In general, the dynamical equation for the spin
84: degrees of freedom have to be supplemented by kinetic equations for
85: the underlying fermionic excitations.
86: 
87: At the heart of the ambiguities involved in defining a spin current
88: operator both for itinerant systems with spin orbit interaction as
89: well as for Heisenberg magnets in inhomogeneous fields is the fact
90: that the magnetization is not strictly conserved for these systems.
91: Still, the intuitive concept of magnetization transport should also be
92: useful for these systems and one is therefore led to define effective
93: current operators, as we will do in this note for the case of a
94: Heisenberg magnet in an inhomogeneous magnetic field.  
95: 
96: \section{Problems with the naive definition of the spin current operator}
97: 
98: Consider a general Heisenberg magnet with Hamiltonian
99: \begin{equation}
100:   \hat{H} = \frac{1}{2} \sum_{  i,j}   J_{ij} {\bd{S}}_i \cdot
101:   {\bd{S}}_j   - g \mu_{\text{B}} \sum_{i  }  {\bd{B}}_i \cdot
102:   {\bd{S}}_i
103:   \,,
104:   \label{eq:Hamiltonian}
105: \end{equation} 
106: where the sums are over all sites ${\bd{r}}_i$ of a chain with
107: periodic boundary conditions, the $J_{ij}$ are general exchange
108: couplings, and ${\bd{S}}_i$ are spin-$S$ operators normalized such
109: that ${\bd{S}}_i^2 = S ( S+1 )$.  The last term in
110: Eq.~(\ref{eq:Hamiltonian}) is the Zeeman energy associated with an
111: inhomogeneous magnetic field ${\bd{B}}_i = {\bd{B}} ( {\bd{r}}_i )$.
112: We assume that the magnetic field at each lattice site is sufficiently
113: strong to induce permanent magnetic dipole moments ${\bd{m}}_i = g
114: \mu_{\text{B}} \langle {\bd{S}}_i \rangle$, not necessarily parallel
115: to $\bd{B}_i$, where $\langle \ldots \rangle$ denotes the usual
116: thermal average.  The simplest geometry is a ferromagnetic ring in a
117: crown-shaped magnetic field, as illustrated in Fig.~\ref{fig:angle}.
118: This geometry is used in the following for illustrative purposes, but
119: our arguments are not restricted to this case.
120: \begin{figure}[tb]
121:   \begin{center}
122:     \epsfig{file=fig1.eps,width=75mm}
123:   \end{center}
124:   \vspace{-4mm}
125:   \caption{%
126:     Classical spin configuration $\hat{\bd{m}}_i$ of a
127:     nearest-neighbor ferromagnetic Heisenberg ring in a radial
128:     magnetic field ${\bd{B}}_i$.}
129:   \label{fig:angle}
130: \end{figure}
131: The Hamiltonian (\ref{eq:Hamiltonian}) implies the equation of motion
132: \begin{equation}
133:   \hbar \frac{ \partial   {\bd{S}}_i}{\partial t }    
134:   +   {\bd{h}}_i
135:   \times  {\bd{S}}_i +
136:   \sum_{j }  {J_{ij}} {\bd{S}}_i \times {\bd{S}}_j
137:   = 0
138:   \,,
139:   \label{eq:eom}
140: \end{equation}
141: where ${\bd{h}}_i = g \mu_{\text{B}} {\bd{B}}_i$.  Note that the last
142: term in Eq. (\ref{eq:eom}) can be written in the form $\sum_j
143: {\bd{I}}_{i \rightarrow j}$, with
144: \begin{equation}
145:   {\bd{I}}_{i \rightarrow j} = J_{ ij}  {\bd{S}}_i \times {\bd{S}}_j
146:   \; .
147:   \label{eq:spincur1}
148: \end{equation}
149: By analogy with the discrete lattice version of the equation of
150: continuity for charge currents, it is tempting to identify $
151: {\bd{I}}_{i \rightarrow j}$ with the operator whose expectation value
152: gives the spin current from site $i$ to site $j$.  In this work we
153: shall argue that this identification is only correct for a strong
154: {\it{homogeneous}} magnetic field, where in equilibrium the
155: expectation values $ \langle {\bd{S}}_i \rangle$ of the spins at all
156: sites are aligned along the same spatially constant direction of the
157: field.  Then ${\bd{h}}_i \times \langle {\bd{S}}_i \rangle = 0$.
158: Using the fact that equilibrium averages are time independent,
159: $\frac{d}{dt} \langle {\bd{S}}_i \rangle =0$, we conclude from the
160: equation of motion (\ref{eq:eom}) that the lattice divergence of the
161: spin current in the presence of a homogeneous magnetic field vanishes
162: \begin{equation}
163:   \sum_j \langle  {\bd{I}}_{i \rightarrow j} \rangle = 0
164:   \; .
165: \end{equation}
166: For a one-dimensional ring with nearest neighbor hopping this implies
167: for each site $i$
168: \begin{equation}
169:   \langle {\bd{I}}_{ i \rightarrow i+1}\rangle  
170:   + \langle {\bd{I}}_{ i \rightarrow i -1} \rangle 
171:   =
172:   \langle {\bd{I}}_{ i \rightarrow i+1}\rangle  
173:   - \langle {\bd{I}}_{ i-1 \rightarrow i } \rangle =0 
174:   \; ,
175: \end{equation}
176: so that the same spin current $ \langle {\bd{I}} \rangle = \langle
177: {\bd{I}}_{ i \rightarrow i+1} \rangle $ flows through each link of the
178: ring.  However, the equation of motion contains only the divergence of
179: the current, so that it does not fix the value of $\langle {\bd{I}}
180: \rangle $.  From the point of view of elementary vector analysis this
181: is a consequence of the fact that both the divergence and the curl are
182: necessary to uniquely specify a vector field. Because the equation of
183: motion contains only the divergence, circulating spin currents cannot
184: be calculated using the equation of motion.  In fact, even the
185: definition of the spin current operator in a geometry permitting
186: circulating spin currents cannot be deduced from the equation of
187: motion.  Of course, for a ring with a collinear spin configuration we
188: know that $ \langle {\bd{I}} \rangle = 0$, so that there are no
189: circulating currents.
190: 
191: The case of a non-uniform magnetic field is more interesting.  In
192: general, the spin configuration in the ground state is then also
193: inhomogeneous. For example, let us consider a radial magnetic field
194: ${\bd{B}}_i = |\bd{B}| {\bd{r}}_i / | {\bd{r}}_i |$ situated at
195: constant latitude $\vartheta_i = \vartheta$, as shown in
196: Fig.~\ref{fig:angle}.  We assume that the direction $\hat{\bd{m}}_i$
197: $=$ ${\bd{m}}_i/ | {\bd{m}}_i |$ of the magnetic moments $ {\bd{m}}_i
198: = g \mu_B \langle {\bd{S}}_i \rangle$ trace out a finite solid angle
199: $\Omega$ on the unit sphere in order-parameter space as we move once
200: around the ring.  If we consider a nearest neighbor Heisenberg
201: ferromagnet with $J_{ij} = - J ( \delta_{ i, j+1} + \delta_{ i ,
202:   j-1})$ then for $ | {\bd{h}} | \equiv g \mu_{\text{B}} |\bd{B}|
203: \gtrsim J S (2 \pi / N )^2$ the classical ground-state configuration
204: $\hat{\bd{m}}_i$ is radial as well, with a slightly different latitude
205: $\vartheta_m$ satisfying \cite{Schuetz03}
206: \begin{equation}
207:   \sin ( \vartheta_m - \vartheta ) = - ({JS}/| {\bd{h}} | ) \left[ 
208:     1 - \cos ( 2 \pi /N  ) \right] \sin ( 2 \vartheta_m )
209:   \,.
210:   \label{eq:classicalangle}
211: \end{equation} 
212: 
213: The main point of this work is that in the presence of an
214: inhomogeneous magnetic field the spin current operator is {\it{not}}
215: simply given by Eq. (\ref{eq:spincur1}).  The fact that the
216: expectation value of Eq. (\ref{eq:spincur1}) cannot be a spin current
217: is perhaps most obvious if we consider the simple case of classical
218: spins in a star-shaped magnetic field, corresponding to $\vartheta_m =
219: \vartheta = \pi /2$ in Fig. \ref{fig:angle}.  In this case Eq.
220: (\ref{eq:spincur1}) gives for a ring with evenly spaced sites at zero
221: temperature
222: \begin{equation}
223:   {\bd{I}}_{i \rightarrow j} = J_{ij} {\bd{e}}_z \sin ( 2 \pi / N )
224:   \; , 
225: \end{equation}
226: where ${\bd{e}}_z$ is a unit vector perpendicular to the plane of the
227: ring.  Note that at the classical level the statics and dynamics of a
228: Heisenberg magnet are completely decoupled. Because the classical
229: ground state does not have any intrinsic dynamics, it does not make
230: any sense to associate a spin current with it which would correspond
231: to moving magnetic moments.  Furthermore, if the classical Heisenberg
232: model is provided with Poisson bracket dynamics, the classical ground
233: state yields a stationary solution, since it minimizes the energy.
234: Clearly, such a completely stationary state cannot be used to
235: transport magnetization.  We conclude that for twisted spin
236: configurations Eq.~(\ref{eq:spincur1}) is not a physically meaningful
237: definition of the spin current operator.
238: 
239: \section{Effective spin currents with correct classical limit}
240: 
241: In order to arrive at a better definition, consider a non-equilibrium
242: situation, i.e.  start with a given density matrix at time $t=0$ and
243: let the system evolve according to the unitary dynamics generated by
244: the Hamiltonian in Eq.~(\ref{eq:Hamiltonian}). The equation of motion
245: (\ref{eq:eom}) then directly translates to a relation for the local
246: and instantaneous order parameter
247: \begin{equation}
248:   \partial_t  \langle {\bd{S}}_i \rangle_t 
249:   + {\bd{h}}_i \times \langle {\bd{S}}_i \rangle_t
250:   + \sum_j \langle {\bd I}_{i\to j}\rangle_t=0 \,.
251:   \label{eq:eomt}
252: \end{equation}
253: Here, $\langle \dots \rangle_t$ denotes an average with respect to the
254: time dependent density matrix.  It is then reasonable to demand that a
255: transport current can lead to an accumulation of magnetization, i.e. a
256: change in the magnitude of the local order parameter in time. For this
257: magnitude, we obtain the equation of motion
258: \begin{equation}
259:   \partial_t  |\langle {\bd{S}}_i \rangle_t| 
260:   + \sum_j {\hat{\bd m}}_i(t) 
261:   \cdot \langle{\bd I}_{i\to j}\rangle_t=0\,,
262: \end{equation}
263: where $\hat{\bd m}_i(t)=\langle{\bd S}_i\rangle_t/|\langle{\bd
264:   S}_i\rangle_t|$ is the time dependent direction of the order
265: parameter.  Note that only the longitudinal component of the naive
266: ``current operator'' appears in this continuity equation without
267: source terms. It is precisely this contribution which we have
268: identified as the dominant one in our spin-wave calculation in
269: \cite{Schuetz03}.  The transverse components lead to a change in the
270: direction of the local order parameter, but they are largely
271: counteracted by the magnetic field term that acts as a source and
272: generates a precession.  If one wants to discuss the electric fields
273: generated by the magnetization dynamics, one either has to take into
274: account both the current ${\bd I}_{i\to j}$ and the local precessional
275: motion, or devise a way to make the cancelation explicit by including
276: part of the ``transverse current'' in an effective magnetic field. We
277: will attempt the second route here.
278: 
279: That this is a sensible way to proceed can be appreciated by a simple
280: approximate calculation. In the classical ground state, the
281: magnetization aligns parallel to the sum of the external and the
282: exchange field.  A necessary condition for the minimum of the
283: classical energy is the invariance under small variations of the
284: directions of the magnetization.  This leads to the condition
285: \cite{Schuetz03}
286: \begin{equation}
287:   {\bd{h}}_i^{\rm eff} \times 
288:   \langle {\bd{S}}_i \rangle 
289:   =0
290:   \; \; , \; \; 
291:   {\bd{h}}_i^{\rm eff} = 
292:   {\bd{h}}_i -  \sum_j J_{ij}  \langle {\bd{S}}_j \rangle
293:   \, .
294:   \label{eq:classical}
295: \end{equation} 
296: Note that the effective magnetic field contains a part of the exchange
297: interaction, which therefore should not be included into the
298: definition of the spin current operator to avoid double counting.
299: Rewriting the exact equation of motion (\ref{eq:eom}) in terms of the
300: effective magnetic field ${\bd{h}}_i^{\rm eff}$ defined in Eq.
301: (\ref{eq:classical}), we obtain
302: \begin{equation}
303:   \hbar \frac{ \partial   {\bd{S}}_i}{\partial t }    
304:   +   {\bd{h}}_i^{\rm eff}
305:   \times  {\bd{S}}_i +
306:   \sum_{j } {\bd{I}}^{\rm eff}_{ i \rightarrow j}   = 0
307:   \,,
308:   \label{eq:eom2}
309: \end{equation}
310: where
311: \begin{equation}
312:   {\bd{I}}^{\rm eff} _{ i \rightarrow j}  = 
313:   {J_{ij}} {\bd{S}}_i \times [  {\bd{S}}_j 
314:   - \langle {\bd{S}}_j \rangle  ]
315:   \label{eq:Ispin3}
316:   \; .
317: \end{equation}
318: Obviously,
319: \begin{equation}
320:   \langle {{\bd{I}}}^{\rm eff}_{i \rightarrow j} \rangle  =
321:   J_{ ij} \left[   \langle  {\bd{S}}_i \times  {\bd{S}}_j \rangle
322:     -   \langle  {\bd{S}}_i \rangle \times   
323:     \langle  {\bd{S}}_j \rangle
324:   \right] \; ,
325:   \label{eq:spincureff}
326: \end{equation}
327: which vanishes identically in the classical ground state, or if the
328: spins are treated within the mean-field approximation, where the spin
329: correlator is factorized.  Physically, this is due to the fact that
330: within the mean-field approximation the Heisenberg exchange
331: interaction is replaced by an effective magnetic field, so that the
332: different sites are uncorrelated and there are no degrees of freedom
333: to transfer magnetization between them. In this work we discuss only
334: localized spin models, so that charge degrees are not available to
335: transfer magnetization between different sites.  
336: 
337: \section{New definition of the spin current operator}
338: 
339: The definition of ${\bd I}^{\rm eff}_{i\to j}$ in
340: Eq.~(\ref{eq:Ispin3}) has the disadvantage of not being antisymmetric
341: with respect to the exchange of the site labels, although its
342: expectation value is obviously antisymmetric.  In order to cure this
343: problem and to generalize the concept of an effective current operator
344: beyond the mean-field description, we propose the following definition,
345: \begin{equation}
346:   \tilde{\bd I}_{i\to j} = {\bd I}_{i\to j}
347:   -{\bd \gamma}_{ij}({\bd \gamma}_{ij}\cdot{\bd I}_{i\to j})\,,
348:   \label{eq:Idef}
349: \end{equation}
350: with the unit vector
351: \begin{equation}
352:   {\bd \gamma}_{ij} = \frac{{\bd m}_i\times{\bd m}_j}
353:   {|{\bd m}_i\times{\bd m}_j|}\,.
354: \end{equation}
355: Thus, we interpret only the projection of ${\bd I}_{i\to j}$ onto the
356: plane spanned by the two local order parameters ${\bd m}_i$ and ${\bd
357:   m}_j$ as a physical transport current. The contribution subtracted
358: in Eq.~(\ref{eq:Idef}) can be incorporated in an effective magnetic
359: field. More precisely, the equilibrium expectation value of the exact
360: equation of motion (\ref{eq:eom}) can be rewritten as
361: \begin{equation}
362:   {\bd h}_i^{\rm eff}\times\langle{\bd S}_i\rangle 
363:   + \sum_j \langle\tilde{\bd I}_{i\to j}\rangle =0\,,
364: \end{equation}
365: with the effective magnetic field now defined as 
366: \begin{equation}
367:   {\bd h}^{\rm eff}_i = {\bd h}_i - \sum_j 
368:   \frac{\langle{\bd S}_i\times J_{ij}{\bd S}_j\rangle\cdot{\bd \gamma}_{ij}}
369:   {[\langle{\bd S}_i\rangle
370:     \times\langle{\bd S}_j\rangle]\cdot{\bd \gamma}_{ij}}
371:   \langle{\bd S}_j\rangle\,.
372:   \label{eq:heff}
373: \end{equation}
374: This reduces to Eq.~(\ref{eq:classical}) for the classical ground
375: state or at the mean-field level, where the correlation function in
376: the numerator is factorized.  The spin current operator defined in
377: Eq.~(\ref{eq:Idef}) is manifestly antisymmetric under the exchange of
378: the labels, as it should be. It implicitly depends on the spin
379: configuration via the unit vector ${\bd \gamma}_{ij}$, so that in
380: twisted spin configurations the spin current operator is a rather
381: complicated functional of the exchange couplings.  The fact that the
382: current operator of an interacting many body system is a complicated
383: functional of the interaction is well known from the theory of
384: interacting Fermi systems \cite{Pines89}. In particular, when the
385: effective interaction does not involve densities only the construction
386: of the current operator is not straightforward \cite{Metzner98}.
387: 
388: For explicit calculations we use a representation of $\tilde{\bd
389:   I}_{i\to j}$ in terms of spin operators quantized in local reference
390: frames with the z-axes pointing along $\hat{\bd m}_i$, i.e. we
391: decompose
392: \begin{equation}
393:   {\bd S}_i = \sum_{\alpha=1,2,3} S_i^{\alpha} {\bd e}_i^{\alpha}\,,
394:   \qquad
395:   S_i^{\alpha}={\bd e}_i^{\alpha}\cdot{\bd S}_i\,,
396: \end{equation}
397: with ${\bd e}_i^3= \hat{\bd{m}}_i$. One still has a freedom in the
398: orientation of the transverse basis $\{{\bd e}_i^1,{\bd e}_i^2\}$,
399: which can elegantly be parametrized, if one uses spherical basis
400: vectors ${\bd e}_i^{\pm}={\bd e}_i^1\pm i{\bd e}_i^2$. We can then
401: write
402: \begin{equation}
403:   {\bd e}_i^+ = e^{i\omega_{i\to j}} \tilde{\bd e}_i^+\,,
404: \end{equation}
405: where $\{\tilde{\bd e}_i^1,\tilde{\bd e}_i^2\}$ is the special
406: transverse basis where $\tilde{\bd e}_i^2=\tilde{\bd e}_j^2={\bd
407:   \gamma}_{ij}$. With this notation we obtain the following expression
408: for the spin current operator
409: \begin{eqnarray}
410:   \tilde{\bd I}_{i\to j} &=& \frac{J_{ij}}{2i}
411:   \Big[
412:   S_i^-S_j^+e^{i(\omega_{i\to j}-\omega_{j\to i})}
413:   \frac{\hat{\bd m}_i+\hat{\bd m}_j}2\nonumber\\
414:   &&-S_i^-S_j^-e^{i(\omega_{i\to j}+\omega_{j\to i})}
415:   \frac{\hat{\bd m}_i-\hat{\bd m}_j}2\nonumber\\
416:   &&+S_i^{\parallel}S_j^-e^{i\omega_{i\to j}}
417:   ({\bd \gamma}_{ij}\times\hat{\bd m}_i)\nonumber\\
418:   &&-S_i^-S_j^{\parallel}e^{i\omega_{j\to i}}
419:   ({\bd \gamma}_{ij}\times\hat{\bd m}_j)
420:   - {\rm H.c.}
421:   \Big]\,,
422:   \label{eq:Idecom}
423: \end{eqnarray}
424: where $S_i^{\pm}=S_i^1\pm i S_i^2={\bd e}^{\pm}_i\cdot{\bd S}_i$ are
425: the usual ladder operators and $S_i^{\parallel}=S_i^3=\hat{\bd
426:   m}_i\cdot{\bd S}_i$.  The third and fourth terms in this expression
427: couple longitudinal and transverse degrees of freedom and therefore do
428: not contribute to leading order in a spin-wave calculation. The first
429: and second summand are dominant for ferromagnetic and
430: antiferromagnetic rings respectively and have been discussed in detail
431: in \cite{Schuetz03} and \cite{Schuetz04}. For a magnetic field that
432: varies smoothly as one moves through the system, the magnetic moments
433: on neighboring lattice sites are almost collinear, so that in both
434: cases the component of the naive ``current operator'' ${\bd I}_{i\to
435:   j}$ along the local order parameter is the one that really
436: corresponds to the transport of magnetization.  In \cite{Schuetz03} we
437: had come to the same conclusion by invoking the gauge freedom in the
438: choice of the transverse axes of quantization. Rotating the coordinate
439: frame around $\hat{\bd m}_i$ corresponds to the gauge transformation
440: \begin{equation}
441:   \omega_{i\to j} \rightarrow \omega_{i\to j} + \alpha_i\,,
442:   \qquad 
443:   S_i^{\pm} \rightarrow S_i^{\pm} e^{\pm i\alpha_i}\,.
444: \end{equation}
445: By this gauge freedom one is then let to identify the derivative of
446: the Hamiltonian with respect to the gauge field $\omega_{i\to j}$ as
447: the relevant current operator.  A comparison with
448: Eq.~(\ref{eq:Idecom}) shows that this is indeed the longitudinal
449: component of ${\bd I}_{i\to j}$.  A more general gauge invariant
450: formulation of the Heisenberg model is discussed in \cite{Chandra90}
451: (see also \cite{Kopietz91}).  In these works, an $O(3)$ gauge field
452: ${\bd{A}}_{i \rightarrow j}$ was introduced in a rather formal manner
453: to write the Heisenberg model in a gauge invariant way and to obtain
454: the spin stiffness tensor by means of differentiation with respect to
455: the gauge field \cite{Singh89}.
456: 
457: Note that the procedure adopted in this section is not restricted to
458: the isotropic Heisenberg interaction of the Hamiltonian
459: (\ref{eq:Hamiltonian}). For a general bilinear spin-spin interaction
460: of the form
461: \begin{equation}
462:   \hat{H} = \frac{1}{2} \sum_{i,j} {\bd{S}}_i \cdot {\bd J}_{ij}
463:   {\bd{S}}_j   - \sum_{i  }  {\bd{h}}_i \cdot
464:   {\bd{S}}_i
465:   \,,
466: \end{equation}
467: where ${\bd J}_{ij}$ is now a 3x3 matrix with ${\bf J}_{ij}^T={\bd
468:   J}_{ji}$, the equation of motion (\ref{eq:eom}) remains valid, if
469: the expression for ${\bd I}_{i\to j}$ is replaced by
470: \begin{equation}
471:   {\bd I}_{i\to j} = {\bd S}_i\times{\bd J}_{ij}{\bd S}_j\,.
472: \end{equation}
473: With this notation, Eqs.~(\ref{eq:Idef}-\ref{eq:heff}) still hold
474: (provided $J_{ij}$ is replaced by the matrix ${\bd J}_{ij}$) and part
475: of the naive current operator ${\bd I}_{i\to j}$ can again be absorbed
476: into the definition of an effective magnetic field.  
477: 
478: \section{Conclusion}
479: 
480: Let us emphasize again that our main point is rather simple: The
481: microscopic equation of motion (\ref{eq:eom}) contains only the
482: (lattice) divergence of the spin current operator, which is not
483: sufficient to fix its rotational part.  A certain part of the operator
484: $\sum_j J_{ij} {\bd{S}}_i \times {\bd{S}}_j$ leads to a
485: renormalization of the effective magnetic field and therefore should
486: not be included into the definition of the spin current operator, see
487: Eqs. (\ref{eq:classical}--\ref{eq:heff}).  The physical spin current,
488: which corresponds to the motion of magnetic dipoles, must be defined
489: such that a purely static twist in the ground state spin configuration
490: of a classical Heisenberg magnet is considered to be a renormalization
491: of the effective magnetic field, and does not contribute to the spin
492: current. To further substantiate our proposal for an effective current
493: operator, it would certainly be insightful to look for a more
494: microscopic derivation by starting from a model involving charge
495: degrees of freedom. It would also be instructive to explicitly
496: investigate non-equilibrium situations with time dependent
497: magnetizations.
498: 
499: We thank P. Bruno and K. Saito for discussions and for urging us to
500: clarify the proper definition of the spin current operator in magnetic
501: insulators.  This work was supported by the DFG via Forschergruppe FOR
502: 412, Project No. KO 1442/5-3.
503: 
504: \begin{thebibliography}{99}
505:   
506: \bibitem{Schuetz03} F. Sch\"{u}tz, M. Kollar, and P. Kopietz, Phys.
507:   Rev. Lett. {\bf{91}}, 017205 (2003).
508:   
509: \bibitem{Schuetz04} F. Sch\"{u}tz, M. Kollar, and P. Kopietz, Phys.
510:   Rev.  B {\bf{69}}, 035313 (2004).
511:   
512: \bibitem{Rashba03} E. I. Rashba, Phys. Rev. B {\bf{68}}, 241315
513:   (2003); cond-mat/0404723; cond-mat/0408119.
514:   
515: \bibitem{Awschalom02}%
516:   See, for example, D.~D.\ Awschalom, D.\ Loss, and N.\ Samarth
517:   (Eds.), {\it{Semiconductor Spintronics and Quantum Computation}},
518:   (Springer, Berlin, 2002).
519: 
520: \bibitem{Meier02}%
521:   F.\ Meier and D.\ Loss, Phys.\ Rev.\ Lett.\ {\bf{90}}, 167204
522:   (2003).
523: 
524: \bibitem{Volovik87} G. E. Volovik, J. Phys. C {\bf 20}, L83 (1987).  
525:   
526: \bibitem{Pines89} D. Pines and P. Nozi\`{e}res, {\it{The Theory of
527:       Quantum Liquids}}, Volume I, (Addison-Wesley Advanced Book
528:   Classics, Redwood City, 1989).
529:   
530: \bibitem{Metzner98} W. Metzner, C. Castellani, and C. Di Castro, Adv.
531:   Phys. {\bf{47}}, 317 (1998).
532: 
533: \bibitem{Chandra90} P. Chandra, P. Coleman, and A. I. Larkin, J.
534:   Phys.: Condens. Matter {\bf{2}}, 7933 (1990).
535:   
536: \bibitem{Kopietz91} P. Kopietz and G. E. Castilla, Phys. Rev. B
537:   {\bf{43}}, 11100 (1991).
538:   
539: \bibitem{Singh89} R. R. P. Singh and D. Huse, Phys. Rev. B {\bf{40}},
540:   7247 (1989).
541:   
542: \end{thebibliography}
543: 
544: 
545: \end{document}
546: 
547: