1: \documentclass[floatfix,aps,prb,twocolumn,superscriptaddress,showpacs]{revtex4}
2: %\documentclass[aps,prl,preprint,groupedaddress,showpacs]{revtex4}
3:
4: \usepackage{graphicx}% Include figure files
5:
6:
7: \begin{document}
8: \title{Spin Diffusion in Double-Exchange Manganites}
9: \author{A. L. Chernyshev}
10: \affiliation{Department of Physics and Astronomy, University of
11: California, Irvine CA 92697}
12: \affiliation{Condensed Matter Sciences Division, Oak Ridge National Laboratory,
13: P.O. Box 2008, Oak Ridge, Tennessee 37831}
14: \author{R. S. Fishman}
15: \affiliation{Condensed Matter Sciences Division, Oak Ridge National Laboratory,
16: P.O. Box 2008, Oak Ridge, Tennessee 37831}
17:
18: \date{\today}
19:
20: \begin{abstract}
21: The theoretical study of spin diffusion in double-exchange magnets by
22: means of dynamical mean-field
23: theory is presented. We demonstrate that the spin-diffusion coefficient
24: becomes independent of the Hund's coupling $J_H$ in the range
25: of parameters $J_H S \gg W\gg T$, $W$ being the bandwidth, relevant to
26: colossal magnetoresistive manganites in the metallic part of
27: their phase diagram. Our study reveals a close
28: correspondence as well as some counterintuitive differences
29: between the results on Bethe and hypercubic lattices. Our
30: results are in accord with neutron scattering data and with
31: previous theoretical work for high temperatures.
32: \end{abstract}
33: \pacs{72.10.-d, 72.15.Lh, 75.30.Vh, 75.40Gb}
34:
35: \maketitle
36:
37: Spin diffusion dominates the low-$\omega$, low-${\bf k}$ excitation
38: spectrum of a magnet in its paramagnetic state and contains important
39: information about the spin dynamics. In the past, spin diffusion
40: was studied intensively in Heisenberg systems and, recently,
41: has been investigated both theoretically and experimentally for
42: strongly-correlated {\it itinerant} magnets.\cite{Kopietz,Bonca,neutrons}
43: The current growth of interest in spintronics requires understanding
44: how local spins relax through their interactions mediated by
45: itinerant charge carriers rather than through their direct interactions
46: with each other.\cite{spintronics}
47: Among such
48: systems are the colossal magnetoresistive (CMR)
49: manganites which are ferromagnetic metals in a
50: large part of their phase diagram.\cite{Dagotto}
51: Recent systematic neutron scattering experiments on the ferromagnetic CMR
52: materials revealed a peak centered at $\omega=0$
53: associated with the spin diffusion. This peak was also seen
54: below the ordering temperature indicating electronic
55: inhomogeneity with regions having lower $T_c$'s.
56:
57: In this Letter we present a comprehensive, self-consistent,
58: microscopic calculation of spin diffusion applying dynamical mean-field
59: theory (DMFT) to the double-exchange (DE)
60: model.
61: We demonstrate that the spin-diffusion coefficient ${\cal D}_s$
62: is related to the local single-particle Green's
63: function and can be evaluated as a function of doping and temperature.
64: Our results
65: agree quantitatively with neutron scattering data on manganites
66: for a range of doping concentrations.
67: Thus, our
68: approach creates a framework for the self-consistent study of
69: diffusive spin dynamics in many real materials, including magnetically
70: doped semiconductors.
71:
72: Following the general hydrodynamic arguments of Ref.~\onlinecite{hydro},
73: we write the generalized susceptibility of a
74: paramagnet for low energies and long wavelengths as
75: \begin{eqnarray}
76: \label{chi}
77: \chi({\bf q},\omega)\simeq \chi({\bf q})\
78: \frac{{\cal D}_s q^2}{-i\omega+{\cal D}_s q^2}
79: \ ,
80: \end{eqnarray}
81: which through the fluctuation-dissipation theorem gives the
82: neutron-scattering dynamical structure factor
83: ${\cal S}({\bf q},\omega)
84: \simeq 2 [n_B(\omega)+1]\mbox{Im}\chi({\bf q},\omega)$,
85: where $\chi({\bf q})$ is the static susceptibility at wave-vector
86: ${\bf q}$ and $n_B(\omega)=[e^{\omega/T}-1]^{-1}$. We take
87: $\hbar=k_B=1$ throughout this paper.
88: Further, the generalized susceptibility can be
89: related to the spin current-current correlation function using the
90: dispersion relations\cite{Maleev} and the continuity equation
91: $\partial S^\alpha({\bf r},t)/\partial t=-\nabla_i \,
92: j^\alpha_i({\bf r},t)$:
93: \begin{eqnarray}
94: \label{chi1}
95: \chi({\bf q},\omega)=-\frac{q^2a^2}{\omega^2}\biggl[
96: \Pi
97: ({\bf q},\omega)-\Pi({\bf q},0)\biggr] \ ,
98: \end{eqnarray}
99: where $\Pi_{ij}^{\alpha\beta}({\bf q},\omega)=-i\int dt e^{i\omega t}
100: \theta(t)\langle[j^{\alpha\dag}_i({\bf q},t), j^\beta_j({\bf q},0)]
101: \rangle$
102: is the retarded current-current correlation function\cite{Mahan},
103: $j^\alpha_i({\bf q},t)$ is the $i$th component of the spin
104: current for the $\alpha$-spin projection,
105: $i=1\dots d$, $d$ is the dimensionality,
106: and $a$ is the lattice constant. We use the isotropy
107: of the spins above $T_c$ and assume the isotropy of real
108: space to suppress the indices in
109: $\Pi_{ij}^{\alpha\beta}({\bf q},\omega)
110: =\Pi({\bf q},\omega)\delta_{\alpha\beta}\delta_{ij}$.
111: Combining Eqs. (\ref{chi}) and (\ref{chi1}) in the ${\bf q},\omega
112: \rightarrow 0$ limit, we write
113: the Einstein relation between the spin-diffusion coefficient and the spin
114: conductivity $\sigma_s$ (which in general is distinct from the
115: particle conductivity) as
116: \begin{eqnarray}
117: \label{sigma}
118: {\cal D}_s\chi=\sigma_s=-a^2\lim_{\omega \rightarrow 0}\frac{\mbox{\rm Im}
119: [\Pi(0,\omega)]}{\omega} \ ,
120: \end{eqnarray}
121: where $\chi=\chi({\bf q}=0)$. These expressions are general and do
122: not depend on the microscopic model.
123:
124: We now consider the DE model with Hamiltonian
125: \begin{eqnarray}
126: \label{H}
127: {\cal H}=-t \sum_{\langle ij
128: \rangle\sigma}\left(c^\dag_{i\sigma}c_{j\sigma}
129: +\mbox{H.c.}\right)-2J_H\sum_i {\bf S}_{i}\cdot {\bf s}_{i} \ ,
130: \end{eqnarray}
131: where $t$ is the nearest-neighbor kinetic energy, $J_H$ is the Hund's
132: coupling
133: between the local Mn$^{3+}$ $S=3/2$ spin and the electronic spin
134: ${\bf s}=c^\dag_{\gamma}
135: {\mbox{\boldmath$\hat{\sigma}$}}_{\gamma\delta}
136: c_{\delta}/2$, and
137: ${\mbox{\boldmath$\hat{\sigma}$}}$ are the Pauli matrices. To describe
138: the multitude of phases in manganites requires that the orbital,
139: phonon, or Jahn-Teller terms be included in the above model.\cite{Millis}
140: However, the
141: {\it magnetic} properties of these materials in the metallic part of
142: their phase diagram, such as the magnetic
143: excitation spectrum and the ferromagnetic
144: transition temperature,\cite{Furukawa,Golosov,Bishop}
145: are {\it quantitatively} well described by the model in
146: Eq. (\ref{H}). Therefore, such a model must
147: be also capable of describing spin diffusion in these
148: systems.\cite{remark2} Regardless of modifications of the
149: DE model needed to describe a particular real system,
150: elucidating the dynamic properties of this basic model of
151: strongly-correlated itinerant magnets
152: is an important task on its own.
153:
154: Within the DE model there is no direct
155: interaction between the local spins. Thus, the total on-site
156: spin ${\bf S}^{tot}_l={\bf S}_l+{\bf s}_{l}$
157: commutes with the exchange part of Eq. (\ref{H})
158: and the spin current can be expressed in terms of electronic
159: operators only:\cite{RF1} $j^\alpha_i({\bf q})=\sum_{\bf k}v^i_{\bf k}
160: c^\dag_{{\bf k},\gamma}
161: \hat{\sigma}^\alpha_{\gamma\delta}c_{{\bf k}-{\bf
162: q},\delta}/2$,
163: where $v^i_{\bf k}={\nabla^i}
164: \varepsilon_{\bf k}$ and
165: $\varepsilon_{\bf k}=-2t\sum_{i=1}^d \cos k_i a$.
166: This is simply another way of saying that electrons mediate the
167: magnetic relaxation processes in an itinerant system.
168:
169: The physical
170: situation relevant to manganites corresponds to strong Hund's
171: coupling $J_H S \gg W$. Since the
172: characteristic relaxation time for the electronic spin is short
173: and the
174: spin relaxation is essentially local, perturbative
175: approaches to the spin diffusion\cite{Zou} are inapplicable.
176: Therefore, we employ
177: DMFT, which takes into account the local dynamics in
178: strongly-correlated systems and has been successfully applied to a
179: number of problems.\cite{DMFT_review,DMFT_caution}
180: Using DMFT also simplifies our problem significantly because
181: the higher-order diagrams in the current-current correlation function of
182: Eq. (\ref{sigma}), often referred to as vertex corrections, are
183: identically zero within this approach.\cite{DMFT_review}
184:
185: Thus, to evaluate the spin-diffusion coefficient we
186: apply the standard Matsubara formalism to Eq. (\ref{sigma})
187: using the above definition of the spin current:
188: \begin{eqnarray}
189: \label{Ds1}
190: \frac{{\cal D}_s\chi}{a^2}= \frac{\pi}{2} \sum_{\bf k} (v^i_{\bf k})^2
191: \int_{-\infty}^{\infty}
192: d\nu A_{\bf k}(\nu)^2 \left(-\frac{\partial n(x)}{\partial x}\bigg|_{x=\nu}
193: \right) \ ,
194: \end{eqnarray}
195: where ${\bf k}$ and $\nu$ are the internal momentum and frequency of
196: the ``bubble'' diagram, respectively. Here $v^i_{\bf k}=2t\sin k_i a$,
197: $A_{\bf k}(\nu)=-(1/\pi)\mbox{Im}
198: G_{\bf k}(\nu)$ is the electronic spectral function,
199: $n(\nu)=[e^{(\nu-\mu)/T}+1]^{-1}$ is the Fermi
200: function, and $\mu$ is the chemical potential.
201:
202: The DMFT imposes a special form of the Green's function in
203: which the self-energy is ${\bf k}$-independent and
204: is defined from a self-consistency condition
205: specified below. Within the DMFT parameters of the
206: model are rescaled such that $\bar{t}=t\sqrt{z}$ is
207: finite as the dimensionality $d\rightarrow\infty$, where $z=2d$
208: is the number of nearest neighbors.\cite{DMFT_review}
209: In the following $\sqrt{2}\bar{t}$ is set to unity.
210: Generally, the $d=\infty$ limit
211: is well defined for Bethe and hypercubic lattice geometries.
212: While the semicircular
213: electronic density of states (DOS) of the Bethe lattice is
214: convenient for calculations, the Bethe lattice itself lacks the
215: translational invariance and inversion symmetry
216: implicitly used to obtain Eq. (\ref{Ds1}).
217: Thus, it is important to determine whether
218: the results on the Bethe lattice are equivalent to the results on
219: the hypercubic lattice, which is
220: free from such deficiencies.
221: Since many problems have been studied using the Bethe
222: lattice,\cite{DMFT_review}
223: this comparison will have an even broader significance for the DMFT
224: in general.
225:
226: We briefly sketch here the DMFT equations for the
227: DE model.\cite{Furukawa,Auslender} Since the self-energy
228: is local, one can change $G_{\bf k}(E)\Rightarrow
229: G_\varepsilon(E)=[E-\varepsilon-\Sigma(E)]^{-1}$, and
230: $\sum_{\bf k}\Rightarrow\int d\varepsilon\rho_0(\varepsilon)$, where
231: $\varepsilon=\varepsilon_{\bf k}$, and
232: $\rho^B_0(\varepsilon)=\sqrt{2-\varepsilon^2}/\pi$ and
233: $\rho^H_0(\varepsilon)=\exp(-\varepsilon^2)/\sqrt{\pi}$ are the bare DOS's
234: for Bethe and hypercubic lattices, respectively.
235: The properties of the system are obtained from the local Green's function:
236: \begin{eqnarray}
237: \label{G_loc1}
238: g(E)= \int
239: d\varepsilon\frac{\rho_0(\varepsilon)}{E-\varepsilon-\Sigma(E)}
240: \ ,
241: \end{eqnarray}
242: where the self-energy is defined from $\Sigma(E)=g_0^{-1}(E)-
243: g^{-1}(E)$, reminiscent of the Dyson equation, where
244: $g_0^{-1}(E)$ is the ``Weiss'' function containing the
245: dynamic influence of the environment on a given local site. The
246: solution of the single-site problem provides a relation between
247: $g_0(E)$ and $\Sigma(E)$.\cite{Furukawa,DMFT_review}
248: In the paramagnetic state and in the quasiclassical limit $S\gg 1$,
249: such a relation is
250: particularly simple:\cite{Furukawa} $\Sigma(E)=(J_HS)^2g_0(E)$, which
251: yields
252: \begin{eqnarray}
253: \label{G_loc2}
254: g(E)= \frac{\Sigma(E)}{(J_HS)^2-\Sigma(E)^2}
255: \ .
256: \end{eqnarray}
257: Together with Eq. (\ref{G_loc1}), this gives a
258: self-consistent condition for $\Sigma(E)$ or $g(E)$.
259: %-----------------------------------------------
260: \begin{figure}[b]
261: \includegraphics[angle=270,width=8cm,clip=true]{MIT_paper}
262: \caption{Interacting DOS $N(\omega)$ for several values of $J_HS$ for
263: the Bethe (solid lines) and hypercubic (dashed lines) lattices.}
264: \label{fig_1}
265: \end{figure}
266: %-----------------------------------------------
267:
268: We now compare the Bethe and
269: hypercubic solution for the single-particle
270: properties. Fig. \ref{fig_1} shows the evolution
271: of the interacting DOS $N(\omega)=-(1/\pi)\mbox{Im}g(\omega)$
272: for several values of $J_HS$.
273: The band splits as $J_H$ increases and
274: both lattice geometries exhibit the same qualitative behavior.
275: Since
276: the hypercubic DOS is expected to have
277: in-gap states, one may ask whether the
278: metal-insulator transition is well defined for a half-filled band.
279: We find that the band splitting in the Bethe and hypercubic lattices
280: happens at the same critical value
281: $(J_HS)_c=1/\sqrt{2}$. At the transition, the imaginary part of the
282: self-energy at $\omega=0$ vanishes and the real part diverges.
283: As a result,
284: $N(\omega=0)$ is exactly zero for $J_H>J_H^c$ and the
285: metal-insulator transition for the half-filled band in the hypercubic
286: geometry is well defined.\cite{Anokhin}
287: At small energies,
288: $N(\omega)\propto e^{-(J_HS)^4/\omega^2}$ vanishes quite
289: abruptly.\cite{remark3} In contrast, the ``outer'' tails of the upper and
290: lower bands behave very similar to the ``bare'' Gaussian form. Note
291: that for $J_HS\agt 1$, the form of $N(\omega)$ for
292: each subband is rather insensitive to the further increase of $J_H$.
293:
294: We now rewrite Eq. (\ref{Ds1}) for the spin-diffusion coefficient
295: within the DMFT:
296: \begin{eqnarray}
297: \label{Ds3}
298: \frac{{\cal D}_s\chi}{a^2}= \frac{1}{2z}
299: \int_{-\infty}^{\infty}
300: d\nu \left(-\frac{\partial n(x)}{\partial x}\bigg|_{x=\nu} \right)
301: \left[\frac{1}{b}-\frac{\partial}{\partial b}\right] \hat{A}(\nu)
302: \ ,
303: \end{eqnarray}
304: where $b=-\mbox{Im}\Sigma(\nu)$ and
305: $\hat{A}(\nu)=\int d\varepsilon\hat{\rho}_0^\alpha (\varepsilon)
306: A_\varepsilon(\nu)$
307: with $\hat{\rho}^{\alpha}_0(\varepsilon)$ being the ``current''
308: DOS defined from the transformation $\sum_{\bf k} \sin^2 k_i a\Rightarrow
309: \int d\varepsilon \hat{\rho}_0(\varepsilon)$.\cite{Chardopathy} For
310: the hypercubic lattice
311: $\hat{\rho}^H_0(\varepsilon)=\rho^H_0(\varepsilon)/2$ while for the
312: Bethe lattice
313: $\hat{\rho}^B_0(\varepsilon)=(2-\varepsilon^2)\rho^B_0(\varepsilon)/3$.
314: To obtain Eq. (\ref{Ds3}) we used the relation
315: $A_{\varepsilon}(\nu)^2=\left[1/b-\partial/\partial b\right]
316: A_{\varepsilon}(\nu)/2\pi$.
317: Since the spin conductivity is proportional to the correlation
318: function of two spin currents, each scaling as $t\sim 1/\sqrt{z}$, the
319: prefactor in Eq. (\ref{Ds3}) contains $1/z$.
320: This means that $d{\cal D}_s\chi$ is finite as $d\rightarrow\infty$,
321: similar to the particle conductivity.\cite{Metzner}
322:
323: In the manganites, the Curie temperature $T_c$ is much smaller than
324: either the
325: Hund's coupling or the bandwidth, in agreement with DMFT
326: and Monte Carlo calculations for the DE
327: model.\cite{Furukawa,Furukawa1}
328: Therefore, all
329: realistic temperatures are much smaller than the bandwidth $T\ll W$ and
330: the derivative of the Fermi-function in the integrand of Eq. (\ref{Ds3})
331: should be replaced by a $\delta$-function at the chemical
332: potential.
333: Then, combining Eq. (\ref{Ds3})
334: with the specific form of the DOS's for the Bethe and hypercubic
335: lattices, one arrives at:
336: \begin{eqnarray}
337: \label{Ds4}
338: \frac{{\cal D}_s\chi z}{a^2}=
339: \left\{
340: \begin{array}{c}
341: \frac{1}{6\pi}\left(2-\frac{g^{\prime\prime}}{b}
342: (2-2b^2-f^2)- f g^\prime
343: \right)^B_{\nu=\mu} \ , \\
344: \frac{1}{4\pi} \left(2-\frac{g^{\prime\prime}}{b}
345: (1-2b^2)-2fg^\prime\right)^H_{\nu=\mu} \ , \ \ \ \ \
346: \end{array}
347: \right.
348: \end{eqnarray}
349: where $g^\prime=\mbox{Re}g(\nu)$,
350: $g^{\prime\prime}=\mbox{Im}g(\nu)$, and
351: $f=\nu-\mbox{Re}\Sigma(\nu)$. Thus,
352: the spin-diffusion coefficient is expressed through
353: the local electronic Green's function and self-energy only.
354: Fig. \ref{fig_2} presents ${\cal D}_s\chi$ as a function of
355: the electronic concentration $n$ for
356: several $J_HS$. As $n$
357: varies from 0 to 1, the chemical potential sweeps
358: from $\omega=-\infty$ to $\omega=0$ in Fig. \ref{fig_1}.
359: The results are very similar in both geometries and become
360: independent of $J_HS$ as $J_HS\rightarrow \infty$ with the maximum
361: located at $n=0.5$.
362: In the limit $J_HS\gg~1$, Eq. (\ref{Ds4}) yields a numerical value for
363: this maximum in ${\cal D}_s \chi z/a^2$:
364: $5/6\pi(=0.265)$ and $0.292$ for the Bethe and hypercubic case,
365: respectively.
366: This demonstrates a close quantitative correspondence between
367: the results in the Bethe and hypercubic lattices, which justifies the
368: use of the former despite the concerns outlined earlier. An interesting feature
369: appears in the results for the hypercubic lattice as $n\rightarrow 0$.
370: Instead of vanishing, the spin-diffusion coefficient tends to a finite
371: limit ${\cal D}_s \chi z/a^2
372: =1/4\pi (J_HS)^2$, shown by circles in Fig. \ref{fig_2}.\cite{Remark4}
373:
374: In the high-temperature limit $T\gg J_HS\gg 1$ Eq. (\ref{Ds3})
375: yields the result ${\cal D}_s\chi\sim 1/T$ obtained previously using
376: a Tchebycheff bounds (TB) formalism.\cite{RF2}
377: Numerically, $({\cal D}_s^{DMFT}/{\cal D}_s^{TB})_{Bethe}=
378: 40\sqrt{2}/9\pi^{3/2}=1.13$
379: agree very closely as well.
380: %-----------------------------------------------
381: \begin{figure}[t]
382: \includegraphics[angle=270,width=8cm,clip=true]{Fig_2}
383: \caption{${\cal D}_s\chi z/a^2$ versus $n$
384: for several $J_HS$ for the Bethe (solid lines) and hypercubic (dashed
385: lines) lattices.}
386: \label{fig_2}
387: \end{figure}
388: %-----------------------------------------------
389:
390: It is interesting to analyze our results in the context of CMR
391: materials. The superexchange (SE) interaction
392: is often discussed\cite{Dagotto} as necessary to correct the DE model.
393: Within DMFT, the SE coupling $J^{SE}$ must scale as $1/z$. Combining
394: this with the result of Ref. \onlinecite{Kopietz} one finds that
395: ${\cal D}_s^{SE}\propto 1/z^{3/2}$ is suppressed in comparison with
396: the DE result ${\cal D}_s\propto 1/z$.
397: This also serves as a demonstration that the DE model cannot be simply
398: reduced to an effective Heisenberg model.
399: In a real $d=3$
400: material the role of SE is further reduced by the
401: smallness of $J^{SE}_{ij}\sim t^2/J_HS$ in comparison with the kinetic
402: energy $\sim xt$. Therefore, the DE must dominate the
403: spin diffusion and one can expect our results to be valid not
404: only for the metallic part of the CMR phase diagram ($x=0.22..0.5$ for
405: La$_{1-x}$Ca$_x$O$_3$, $n=1-x$ in Fig.~\ref{fig_2}) but also
406: for the ferromagnetic insulating phase ($x\leq 0.22$).\cite{Remark5}
407: Of course, in the limit $x\rightarrow 0$ DE will
408: diminish and the SE will dominate. Also, the critical scaling in
409: the mixed phase $0 < x < 0.12$ (with a mixture of antiferromagnetic
410: and ferromagnetic orders) can be strongly modified.\cite{Kiselev}
411: These cases require separate consideration. At $x\geq 0.5$
412: charge ordering prevents the DE mechanism from being operative.
413:
414: A systematic neutron scattering study of the metallic
415: manganites recently focused on spin diffusion.\cite{neutrons}
416: The width of the observed peak in ${\cal S}({\bf q},\omega)$ centered
417: at $\omega=0$ scales as $\Lambda q^2$, where
418: $\Lambda=2{\cal D}_s$. Experimental results taken at $T^{exp}(x)\simeq
419: 1.1 T_c(x)$, where $T_c(x)$ is the Curie temperature for a given hole
420: concentration $x$, give $\Lambda=15-30$ meV\AA$^2$. To compare our
421: results to experiments we use $z=6$ and $a=3.87$ \AA.
422: But keep in mind
423: that spin diffusion persists below the transition point and
424: that, at least for the experimentally accessible wave-vectors, the
425: correlation length saturates at about 20 \AA. This
426: implies that local magnetic correlations are suppressed by
427: electronic inhomogeneities, which are probably associated with the
428: local charge ordering.\cite{neutrons}
429: So for simplicity, we take the susceptibility in the Curie form
430: $\chi^{exp}=S^\prime (S^\prime +1)/3T^*$, where $S^\prime(x)=S+(1-x)/2$ is
431: an average on-site spin and $T^*=T^{exp}(x)$ is known from
432: Ref.~\onlinecite{neutrons}. Fig. \ref{fig_3}
433: shows the theoretical data for ${\cal D}_s\chi/a^2$ for $J_HS\gg 1$ from
434: Eq. (\ref{Ds4}) as a function of $x$ together with
435: experimental ${\cal D}^{exp}_s\chi^{exp}/a^2$. We note here, that the
436: theoretical curves in Fig. \ref{fig_3} are virtually independent of
437: the actual value of $J_HS$ for $J_HS\agt 1$.
438: This figure
439: demonstrates a remarkable agreement between the experimental and
440: theoretical results,
441: which contains no fitting parameters.
442: Further improvement of the agreement can be
443: sought, for example, from taking into account the second $e_g$
444: band which would effectively reduce $J_H/W$.\cite{remark2}
445: %-----------------------------------------------
446: \begin{figure}[t]
447: \includegraphics[angle=0,width=8cm,clip=true]{Fig_3b}
448: \caption{${\cal D}_s\chi/a^2$ as a function of $x$.
449: Experimental values of ${\cal D}_s$ are from Fig. 5 of
450: Ref.~\onlinecite{neutrons}, $\chi^{exp}$ is described in the text,
451: theoretical results are from Eq. (\ref{Ds4}) for
452: $J_HS\gg 1$.}
453: \label{fig_3}
454: \end{figure}
455: %-----------------------------------------------
456:
457: Since various other experiments also
458: indicate the presence of local inhomogeneities in CMR systems,\cite{Adams}
459: we propose the following analysis.
460: As ${\bf q}$ decreases, long-range magnetic correlations must dominate
461: $\chi({\bf q})$ and the magnetic correlation length must eventually
462: exceed the size of the local polaronic distortions.
463: So at a fixed temperature close to $T_c$, $\chi^{exp}({\bf
464: q})$ will increase and there will be a systematic {\it decrease}
465: in the observed value of ${\cal D}^{exp}_s \sim 1/\chi$.
466: This set of measurements would provide further information about
467: magnetic correlations within the inhomogeneities in CMR systems.
468:
469: In conclusion, we have presented a self-consistent study of the
470: spin-diffusion in the double-exchange magnets within the framework of
471: DMFT. This non-perturbative approach allows us
472: to calculate the spin-diffusion coefficient at any temperature down to
473: a transition point. A good agreement with the experiments in the
474: ferromagnetic CMR manganites and earlier work is
475: found. Altogether, this provides a new insight into the dynamics of
476: strongly-correlated itinerant magnets.
477:
478: We would like to acknowledge valuable discussions with
479: A.~H.~Castro~Neto, P.~Dai, J.~Fernandez-Baca, M. Jarrell, M.~N.~Kiselev,
480: V.~Perebeinos, and A. G. Yashenkin. This research was supported by
481: ORNL, managed by UT-Battelle, LLC, for the U.S. DOE under
482: contract DE-AC05-00OR22725.
483:
484: %-------------------------------------------------------
485: \begin{thebibliography}{}
486:
487: \bibitem{Kopietz} P. Kopietz, Phys. Rev. B {\bf 57}, 7829 (1998);
488: Mod. Phys. Lett. B {\bf 7}, 1747 (1993).
489:
490: \bibitem{Bonca} J. Bonca and J. Jaklic, Phys. Rev. B {\bf 51}, 16083
491: (1995).
492:
493: \bibitem{neutrons} P. Dai {\it et al.}, Phys. Rev. B {\bf 64},
494: 224429 (2001).
495:
496: \bibitem{spintronics} M. E. Flatt\'e and J. M. Byers, Phys. Rev. Lett.
497: {\bf 84}, 4220 (2000).
498:
499: \bibitem{Dagotto} E. Dagotto {\it et al.}, Phys. Rep. {\bf 344}, 1
500: (2001) and {\it Physics of Manganites},
501: edited by T. A. Kaplan and S. D. Mahanti
502: (Kluwer Academic and Plenum, New York, 1998).
503:
504: \bibitem{hydro} D. Forster, {\it Hydrodynamic Fluctuations, Broken
505: Symmetry, and Correlation Functions}, (Benjamin, Reading, 1975);
506: B. I. Halperin and P. C. Hohenberg, Phys. Rev. {\bf 177}, 952 (1969).
507:
508: \bibitem{Maleev} S. V. Maleev, Zh. Eksp. Teor. Fiz. {\bf 65},
509: 1237 (1973) [Sov. Phys. JETP {\bf 38}, 613 (1974)];
510: Yu. A. Izyumov and Yu. N. Skryabin, in {\it Statistical Mechanics of
511: Magnetically Ordered Systems}, (Plenum, New York 1988).
512:
513: \bibitem{Mahan} G. D. Mahan, {\it Many-Particle Physics},
514: Third Edition (Kluwer Academic and Plenum, New York, 2000).
515:
516: \bibitem{Millis} A. J. Millis {\it et al.}, Phys. Rev. Lett. {\bf 77},
517: 175 (1996).
518:
519: \bibitem{Furukawa} N. Furukawa, J. Phys. Soc. Jpn. {\bf 64}, 2754 (1995);
520: {\bf 65}, 1174 (1996).
521:
522: \bibitem{Golosov} D. I. Golosov, Phys. Rev. Lett. {\bf 84}, 3974 (2000).
523:
524: \bibitem{Bishop} J. Zang {\it et al.},
525: J. Phys.: Condens. Matter {\bf 9}, L157 (1997).
526:
527: \bibitem{remark2} A more realistic model with two
528: $e_g$ bands becomes effectively equivalent to our single-band
529: DE model Eq. (\ref{H}) if the
530: Jahn-Teller splitting or Hubbard repulsion are included.
531: Thus, the two bands with splitting
532: should only change our results quantitatively.
533: Our theory can be extended to the Jahn-Teller
534: case, but it transforms to a different problem
535: for the case of the Hubbard interaction.
536:
537: \bibitem{RF1} R. S. Fishman, Phys. Rev. B {\bf 62}, R3600 (2000).
538:
539: \bibitem{Zou} L.-J. Zou {\it et al.}, J. Appl. Phys. {\bf 87}, 5499 (2000).
540:
541: \bibitem{DMFT_review} A. Georges {\it et al.}, Rev. Mod. Phys. {\bf
542: 68}, 13 (1996).
543:
544: \bibitem{DMFT_caution} If the nano-inhomogeneities are truly
545: important for CMR\cite{Dagotto} the DMFT
546: will eventually fail to reproduce the
547: percolative processes for transport in the CMR regime.
548:
549: \bibitem{Auslender} M. Auslender and E. Kogan, cond-mat/0102469.
550:
551: \bibitem{Anokhin} Also see A. O. Anokhin {\it et al.}, J. Phys.:
552: Condens. Matter {\bf 3}, 1475 (1991).
553:
554: \bibitem{remark3} Similar behavior has been found in the Hubbard model:
555: A. Georges and W. Krauth, Phys. Rev. B {\bf 48}, 7167 (1993).
556:
557: \bibitem{Chardopathy} A. Chattopadhyay {\it et al.},
558: Phys. Rev. B {\bf 61}, 10738 (2000).
559:
560: \bibitem{Metzner} W. Metzner {\it et al.}, Phys. Rev. B {\bf 45}, 2237 (1992).
561:
562: \bibitem{Furukawa1} Y. Motome and N. Furukawa, J. Phys. Soc. Jpn. {\bf
563: 69}, 3785 (2000).
564:
565: \bibitem{Remark4}
566: In our problem, diffusion originates from
567: correlations and the states in the remote tails ($|\mu|\gg J_H S$)
568: are weakly correlated with diverging relaxation time:
569: $\tau\propto e^{\mu^2}$.
570: The smallness of the DOS cancels this divergence giving a
571: finite value of ${\cal D}_s\chi\propto \tau N(\mu)$
572: as $n\rightarrow 0$.
573: In this limit the dynamical
574: spin-diffusion coefficient\cite{RF2} ${\cal
575: D}(\omega)\chi=\sigma_s(\omega)\propto 1/(\omega^2\tau^2+4)$ has
576: negligible width and weight given by $1/\tau$.
577:
578: \bibitem{RF2} R. S. Fishman, J. Phys.: Condens. Matter {\bf 12}, L575
579: (2000); {\it ibid.} {\bf 14}, 1337 (2002).
580:
581: \bibitem{Remark5} In fact, the experimental data on ${\cal D}_s$ are very
582: similar for the insulating and metallic compounds.\cite{neutrons}
583:
584: \bibitem{Kiselev} K. A. Kikoin and M. N. Kiselev,
585: Zh. Eksp. Teor. Fiz. {\bf 112},
586: 1816 (1997) [Sov. Phys. JETP {\bf 85}, 994 (1997)].
587:
588: \bibitem{Adams} C. P. Adams {\it et al.}, Phys. Rev. Lett.
589: {\bf 85}, 3954 (2000).
590:
591: \end{thebibliography}
592:
593:
594: \end{document}
595:
596:
597:
598:
599:
600:
601:
602:
603: