1: \documentclass[twocolumn,showpacs,amsmath,amssymb]{revtex4}

2: %\documentclass[preprint, showpacs,amsmath,amssymb]{revtex4}

3: \usepackage{graphicx}

4: \usepackage{bm}

5: \begin{document}

6: \title{Ferromagnetism from localized deep impurities in magnetic semiconductors.}

7: \author{Victor Barzykin} 

8: \affiliation{Department of Physics and Astronomy, University of Tennessee,

9: Knoxville, TN  37996-1200}

10: \begin{abstract}

11: We propose that localized defects in magnetic semiconductors  

12: act as deep impurities and can be described by the 

13: Anderson model. Within this model, hybridization of 

14: d-orbitals and p-orbitals gives rise to a non-RKKY indirect exchange mechanism, 

15: when the localized d-electrons are exchanged through both conduction 

16: and valence bands. For semiconductors with indirect band gap the non-RKKY part of exchange integral is

17: antiferromagnetic, which suppresses ferromagnetism.

18: In case of direct band gap, this exchange mechanism can, under certain conditions, lead to 

19: enhancement of ferromagnetism.

20: The indirect exchange intergral is much stronger than RKKY, and can be sufficiently long range. 

21: Thus, a potentially new class of high-temperature magnetic semiconductors emerges, where

22: doped carriers are not necessary to mediate ferromagnetism. 

23: Curie temperatures in such magnetic semiconductors are determined mostly by the interaction between 

24: localized impurities, \textit{not Zener mechanism}. This effect could also

25: be responsible for unusually high Curie temperatures in some magnetic semiconductors with

26: direct band gap, such as Ga$_{1-x}$Mn$_x$As.

27: \end{abstract}

28: \vspace{0.15cm}

29: 

30: \pacs{75.50.Pp, 72.80.Ey, 75.30Hx, 75.50Dd}

31: \maketitle

32:  

33: \section{Introduction}

34: 

35: The advantage of ferromagnetic semiconductors (FS) as a source of 

36: spin-polarized carriers is that they can be easily integrated into semiconductor devices\cite{Prinz, Wolf}.

37: When discovered, ferromagnetism at room temperatures with full polarization of itinerant carriers will be

38: a major breakthrough in semiconductor electronics. Most theoretical

39: and experimental efforts have been concentrated on III-V, group IV, and II-VI-based diluted magnetic semiconductors (DMS).

40: These semiconductors are alloys in which some atoms are randomly replaced by magnetic atoms, such as Mn$^{2+}$.

41: 

42: Ferromagnetism in diluted magnetic semiconductors (DMS) is thought to be well understood in terms of the 

43: so-called p-d exchange model,

44: which was first considered  

45: over 50 years ago\cite{Zener1,Zener2,AG} (See also Ref.\cite{Korenblit, Furdyna, Dietll} for a review). 

46: At concentrations of impurities above the Mott limit, i.e., as soon as carriers become delocalized,

47: conventional model of FS is fairly simple.

48: According to, for example, Ref.\cite{AG}, the interaction between

49: charge carriers in a semiconductor and spin-S impurities can be written as:

50: \begin{eqnarray}

51: \label{int}

52: U &=& - \int \bm{s}(\bm{r}) \sum_i \bm{S}_i J^{pd}(\bm{r} - \bm{R}_i) d^3 \bm{r} \\ 

53:   &=& - J^{pd}_{\bm{q}=0} \bm{s} \sum_i \bm{S}_i, \nonumber

54: \end{eqnarray}

55: where $\bm{S}_i$ and $\bm{R}_i$ are the spin and the position of an i-th atom of magnetic impurity, 

56: $J^{pd}_{\bm{q}=0} = \int J(\bm{r}) d^3 \bm{r}$, and we have used the fact that magnetic impurities are randomly distributed

57: in the sample. A simple analysis\cite{AG} then shows that, in the presence of one type of charge carriers, Curie temperature,

58: $T_c$, is proportional to concentration of magnetic impurities, $N_i$, and the square of the strength of the exchange 

59: interaction, $a$:

60: \begin{equation}

61: T_c = {n_i S (S+1) (J^{pd}_{\bm{q}=0})^2 \chi_0 \over 12 \mu_0^2}

62: \label{Curie}

63: \end{equation}

64: Here $\mu_0$ is the magnetic moment of charge carriers, $\chi_0$ is the Pauli term in the spin susceptibility

65: in the absence of impurities. At sufficiently large carrier densities, 

66: the spin polarization of the charge carriers at $T=0$ is given by:

67: \begin{equation} 

68: s = {n_i J^{pd}_{\bm{q}=0} S \chi_0 \over 4 \mu_0^2}

69: \end{equation} 

70: In general, for any carrier density, the spin polarization is given by Ze\`eman-split Fermi surface,

71: with the difference in chemical potentials for ``up'' and ``down'' spins given by:

72: \begin{equation}

73: \mu_{\uparrow} - \mu_{\downarrow} = J^{pd}_{\bm{q}=0} n_i S

74: \end{equation}

75: The spin polarization at any filling is then very easily calculated from this equation. For example

76: when the carrier density 

77: \begin{equation}

78: n_e \le n_c = \frac{(2 J^{pd}_{\bm{q}=0} n_i S m^*)^{3/2}}{6 \pi^2}\, ,

79: \end{equation}

80: the carriers will be fully polarized at $T=0$. When $n_e > n_c$, they are no longer fully polarized.

81: The polarization of carriers is then determined by a parametric equation 

82: (with $\mu$ as a parameter):  

83: \begin{equation}

84: s = \frac{1}{2}\, (n_{e \uparrow} - n_{e \downarrow}),

85: \end{equation}

86: where   

87: \begin{equation}

88: n_{e \uparrow, \downarrow} = 

89: \frac{(2 m^* \mu \pm  J^{pd}_{\bm{q}=0} n_i S m^*)^{3/2}}{6 \pi^2}\,,

90: \end{equation}

91: and $n_e = n_{e \downarrow} + n_{e \uparrow}$.

92: 

93: Despite the simplicity of the basic concept of Zener ferromagnetism, calculations of $T_c$ for real materials 

94: become rather involved, and depend crucially on details of the

95: band structure, the $p-d$ exchange matrix, or direct antiferromagnetic exchange between $Mn^{2+}$ spins.

96: Most theoretical and experimental efforts on magnetic semiconductors have been concentrated on  

97: finding a new DMS-based material that would have a ferromagnetic transition above room temperature, and would be possible to

98: incorporate in thin film form with mainstream semiconductor device materials.    

99: There are theoretical predictions for T$_c$'s above room temperatures in several classes of these materials\cite{dietl1,dietl2}.  

100: However, experiments indicate that the growth of Curie temperature with concentration of magnetic impurities

101: saturates at 5-10\% Mn doping in most ferromagnetic  semiconductors; it may even start to decrease at higher

102: concentrations of Mn. Since, according to the mean field p-d model, the Curie temperature Eq.(\ref{Curie}) 

103: grows linearly or faster (if the carrier concentration changes) with the growth of Mn concentration, it is

104: important to understand what limits the growth of $T_c$, and how this can be avoided. 

105: 

106: Experimental  effort in FS concentrated mostly on In$_{1-x}$Mn$_x$As\cite{Munekata} ($T_c \sim 35K$) 

107: and Ga$_{1-x}$Mn$_x$As\cite{deBoeck,Ohno,Ohno1,Ohno2} ($T_c \sim 110K$). 

108: While room temperature ferromagnetism in FS remains a theoretical possibility, the highest Curie temperature reported in

109: the homogeneous sample with Mn concentration around 5\% is 116K for Mn$_x$Ge$_{1-x}$ \cite{Park}. 

110: The saturation $T_c$ growth with increased Mn concentration is usually 

111: ascribed to increased disorder. Disorder effectively introduces an exponential cutoff for the RKKY interaction and 

112: reduces its range\cite{Ohno}. However, as we discuss in section VIII, weak disorder does not change Curie temperature. 

113: Another possible reason is that

114: the strength of direct antiferromagnetic exchange grows as the average distance between impurities becomes shorter,

115: which, in turn, lowers the Curie temperature. 

116: Room temperature ferromagnetism has been observed in a number of compounds with \textit{large} concentration of $Mn$

117: impurities (see Ref. \cite{Wolf} for a review), or, due to phase separation and formation of nanoclusters of these compounds (

118: such as Mn$_{11}$Ge$_8$ in Ref.\cite{Park}).

119: However, large carrier concentration in these compounds limits the degree of spin polarization that is necessary

120: for device applications.

121: 

122:    Some theoretical studies\cite{Millis} claimed that the reason Curie temperature can't get higher above 

123: certain concentration of Mn is that the $p-d$ model

124: cannot be treated in the mean field approxination. However, since the RKKY interaction has a large radius, 

125: mean field treatment in absence of strong disorder should be justified\cite{Larkin}.

126: 

127: 

128: In what follows, we adopt the idea that the reason for the discrepancy between mean field theory and experiment lies

129: in the presence of additional interactions in the effective spin Hamiltonian of the problem, which are not present

130: in the conventional formulation of the $p-d$ model. To derive effective spin Hamiltonian, including additional interactions, 

131: we start from a more general Anderson model of deep magnetic impurities in FS.

132: We show that, in case of a deep Anderson impurity in a semiconductor,

133: an additional long-range indirect exchange interaction appears in the effective Hamiltonian, which, if 

134: antiferromagnetic,  severely limits $T_c$-s in these materials. On the other hand, for direct gap semiconductors, this  

135: interaction could change sign and become ferromagnetic, if the magnetic impurity lies deep enough. 

136: We derive conditions under which this becomes possible. As a result, ferromagnetic correlations would become

137: enhanced, not reduced, as the concentration of magnetic impurities grows, and high-temperature ferromagnetism could be 

138: possible even without carriers. 

139: The indirect exchange between two deep impurities, whether ferro- or antiferromagnetic, is stronger than the RKKY interaction, 

140: and thus could produce large Curie temperatures. For example, it could provide a possible explanation of high-temperature 

141: weak ferromagnetism\cite{Young} in La$_x$Ca$_{1-x}$B$_6$ or recently discovered 

142: CaB$_2$C$_2$\cite{Akimitsu} (CaB$_2$C$_2$ has $T_c=770K$ and $M=10^{-4} \mu_B$).

143: Experiment\cite{Akimitsu,Fisk} indicates that these materails are direct band semiconductors with a relatively

144: small band gap, and that impurities play a major role in establishing the new high-temperature

145: ferromagnetic state. However, other reports\cite{Sato} claim that high-temperature ferromagnetism in these 

146: materials is not a bulk effect. Rather, it is related to clustering

147: or new boron phases with Fe or Ni magnetic impurities. Understanding

148: the mechanism of weak ferromagnetism in these materials could lead to a discovery of more members of this class. 

149: In either case, as we report in this paper, there is a theoretical

150: possibility of a new type of FS with direct band gap, where ferromagnetism is not carrier-driven. 

151: 

152: This paper is organized as follows. In Section II we discuss the general procedure of the derivation of the effective

153: low-energy Hamiltonian for Anderson impurities. In section III we discover that, due to strong hybridization,

154: an Anderson impurity is no longer a local center; its localized wave function acquires a finite range, 

155: which is directly related to the corresponding "Bohr" radius of a charged impurity. Sections IV and V are devoted to 

156: the derivation of the effective exchange Hamiltonian for magnetic impurities. In Section VI we

157: explore the consequences of the large range of interaction for magnetism, such as a high Curie temperature in case of a direct band

158: gap if magnetic impurities are dense enough. In section VII we consider a minor modification of the effective Hamiltonian in 

159: case of higher spin (such as $S=5/2$ for Mn), and the application of these ideas to Ga$_{1-x}$Mn$_x$As. 

160: In section VIII the influence of weak disorder and interactions on Curie temperature is briefly discussed.

161: Section XI provides a summary and conclusions.

162: 

163: \section{The effective Hamiltonian}

164: 

165:    We start by considering a simple model of FS. In III-V systems, such as Ga$_{1-x}$Mn$_x$As, 

166: it is well established that the Mn ions substitute for Ga, and contribute itinerant holes to the GaAs  valence band. Experimentally,

167: the hole density is typically a small fraction (15\% or so) of the Mn concentration, perhaps due to strong localization of carriers

168: on Mn and other defects, so Ga$_{1-x}$Mn$_x$As can be considered partially compensated. The Mn ion has half-filled d-shell, which

169: acts like a spin-5/2 local moment. The Anderson model, which is more general 

170: than the $p-d$ model, should completely account for all the physics of FS. It is well known that, when spins are well localized,

171: the single-impurity Anderson Hamiltonian is reduced to the $p-d$ Hamiltonian by the Schrieffer-Wolff transformation\cite{SW}.

172: For many impurities, this may no longer be the case. Let us start by considering a single-orbital Anderson Hamiltonian:

173: 

174: \begin{equation}

175: H = H_0 + H_V,

176: \label{theH}

177: \end{equation}

178: where

179: \begin{eqnarray}

180: H_0 &=& \sum_{\bm{p} \sigma i} \epsilon_i(p) a^{\dagger}_{i \bm{p} \sigma} a_{i \bm{p} \sigma} + \\

181: &+& \sum_n \left[ \epsilon_0 \sum_{\sigma}  d^{\dagger}_{n \sigma} d_{n \sigma} + 

182: U d^{\dagger}_{n \uparrow} d^{\dagger}_{n \downarrow} d_{n \downarrow} d_{n \uparrow} \right]. \nonumber

183: \end{eqnarray}

184:    

185: Here the first sum ($\bm{p}$) is taken over the reciprocal space, the second sum ($n$) over real space impurity sites.

186: $U$ is the on-site Coulomb repulsion term. Typically, $U$ is very large ($\sim 5 eV$), and can be taken to be infinite.

187: Here $\epsilon_i(p)$ are the energy band spectra of conduction and valence bands ($i=1,2$).

188: 

189: The hybridization term in the model Hamiltonian, $H_V$, accounts for the $p-d$ hybridization between impurity sites and 

190: conduction and valence bands:

191: 

192: \begin{equation}

193: H_V = \frac{1}{N^{1/2}} \sum_{\bm{p} n \sigma i} V_i \left\{ a^{\dagger}_{\bm{p} \sigma i} d_{n \sigma} e^{-i \bm{p} \cdot \bm{R}_n}

194: + h.c.\right\}

195: \end{equation}

196: 

197: This model is a reasonable generalization of the $p-d$ exchange Hamiltonian, usually considered in the literature. 

198: Because of large on-site Coulomb

199: repulsion, the $d$-levels are half-filled. While we consider the case of a single d-orbital, a generalization 

200: to $S=5/2$ Mn ion is straightforward (see section VII below). 

201: 

202: The Anderson Hamiltonian Eq.(\ref{theH}) describes a very complicated problem.

203: However, under the $U=\infty$ constraint, it can be reduced to the problem of Heisenberg spins 

204: (in case of a single d-orbital, spin-$1/2$).

205: The low-energy effective 

206: Hamiltonian is equivalent to Eq.(\ref{theH}) in the limit $k_B T \ll \Delta_i$, where $\Delta_i$ is the energy difference

207: between the impurity $d$-level and the top of the valence band, or the energy difference between the impurity $d$-level and the 

208: bottom  of conduction band. Since typical gap values in semiconductors are of the order of $1 eV$, this is usually a valid assumption.

209: 

210: \begin{figure}

211: \includegraphics[width=3in]{fig1.eps}

212: %\includegraphics[width=3in]{fig1.pdf}

213: \caption{The second-order contribution to the effective Hamiltonian, shown as a time-ordered process.}

214: \label{fig2ord}

215: \end{figure}

216: 

217: Following Refs. \cite{GS1,BG1}, the effective spin Hamiltonian can be derived by expanding the S-matrix

218: (or, at finite temperatures, the partition function) in $V$-s, and re-expressing various time-ordered processes

219: in terms of spin operators. Then, these processes are collected back under exponent, to obtain the effective

220: Hamiltonian. This method allows one to obtain 

221: consistently the interaction between spins and carriers (electrons or holes) in conduction and valence bands, and 

222: carrier-carrier interaction. In a way, the concept is similar to perturbative renormalization group, since 

223: we arrive at a low energy effective Hamiltonian by integrating out higher-energy states. Treating hybridization term in Eq.(\ref{theH}) 

224: as a perturbation, we can rewrite the partition function as 

225: 

226: \begin{equation}

227: Z = Tr[\exp(-\beta H_0) S(\beta)],

228: \end{equation}

229: where

230: \begin{widetext}

231: \begin{equation}

232: S(\beta) = T \exp\left( - \int_0^{\beta} H_V(\tau) d \tau \right) =

233: T \exp(- \sum_{n \sigma i} \int_0^{\beta} d \tau V_i \left\{ \Psi^{\dagger}_{n \sigma i}(\tau) d_{n \sigma}(\tau) 

234: + d^{\dagger}_{n \sigma}(\tau) \Psi_{n \sigma i}(\tau)\right\}

235: \label{smatr}

236: \end{equation}

237: \end{widetext}

238: 

239: The problem of finding the effective Hamiltonian is then to reduce these expressions to the form:

240: 

241: \begin{equation}

242: Z = Tr(\exp[-\beta H_{eff}]),

243: \end{equation}

244: using the $k_B T \ll \Delta_i$ condition, i.e., the fact that the local levels are almost always occupied, and that transitions to

245: conduction and valence bands are absent at low temperatures. Various terms in the effective Hamiltonian can then

246: be associated with certain time-ordered virtual processes.

247: For example, the first non-zero contribution to $H_{eff}$ is from the second order term in the expansion of 

248: Eq.(\ref{smatr}) in $V_i$:

249: \begin{widetext}

250: \begin{equation}

251: S_2(\beta) = \sum_{n \sigma \sigma'} V_1^2 \int_0^{\beta} d \tau_1

252: \int_{\tau_1}^{\beta} d \tau_2 \Psi_{n \sigma 1}(\tau_2) \Psi^{\dagger}_{n \sigma' 1}(\tau_1)

253: d^{\dagger}_{n \sigma}(\tau_2) d_{n \sigma'} (\tau_1).

254: \label{2order}

255: \end{equation}

256: \end{widetext}

257: The order of operators $d^{\dagger}_{n \sigma}(\tau_2) d_{n \sigma'} (\tau_1)$ ($\tau_2 > \tau_1$) is fixed

258: by the assumption of strong Coulomb repulsion on the $n$-th center. Note that, in the second order, because

259: all centers are filled, only one band contributes to the effective Hamiltonian. The "filled" band does not contribute,

260: because of the Pauli principle (this is not the case at a finite U). The perturbative time-ordered process corresponding 

261: to the term in Eq.(\ref{2order}) is shown in Fig.\ref{fig2ord}.

262: 

263: We may rewrite $\Psi_{n \sigma 1}(\tau_2) \Psi^{\dagger}_{n \sigma' 1}(\tau_1)$ in the interaction representation

264: as:

265: \begin{widetext}

266: \begin{equation}

267: \Psi_{n \sigma 1}(\tau_2) \Psi^{\dagger}_{n \sigma' 1}(\tau_1) = 

268: \{\Psi_{n \sigma 1}(\tau_2) \Psi^{\dagger}_{n \sigma' 1}(\tau_1)\}_+ - 

269: \Psi^{\dagger}_{n \sigma' 1}(\tau_1) \Psi_{n \sigma 1}(\tau_2). 

270: \label{commut}

271: \end{equation}

272: \end{widetext}

273: The first term in Eq.(\ref{commut}) is a c-number. In the limit $\Delta_1 \gg k_B T$ it is

274: possible to put $\tau_1 \simeq \tau_2$ in the second term in Eq.(\ref{commut}). Thus, for

275: the second-order contribution in the effective Hamiltonian we may write:

276: \begin{widetext}

277: \begin{equation}

278: H^{(2)}_{eff} = - \sum_{\bm{p}} \frac{V_1^2}{\epsilon_{1p} - \epsilon_0}\, 

279: + \frac{V_1^2}{\Delta_1}\, \sum_{n \sigma \sigma'} \Psi^{\dagger}_{n \sigma' 1} \Psi_{n \sigma 1}

280: \left( \frac{1}{2}\, \delta_{\sigma \sigma'} + \bm{S}_n \bm{\sigma}_{\sigma' \sigma} \right),

281: \label{2oref}

282: \end{equation}

283: \end{widetext}

284: where $\bm{S}_n$ is the localized spin of the n-th impurity. The first term corresponds to the renormalization of 

285: the energy of the localized level, while the second term involving spins of localized impurity

286: and carriers is nothing but the ordinary $H_{pd}$, the p-d model discussed in section I.

287: 

288: The next order terms in the effective Hamiltonian are fourth order in V-s. There are sums

289: over two local centers, $m$ and $n$ in $S_4(\beta)$. Also, there are contributions to $S_4(\beta)$,

290: which we will denote $S'_4(\beta)$, that are already accounted for in the effective 

291: Hamiltonian Eq.(\ref{2oref}):

292: \begin{equation}

293: S'_4(\beta) = \frac{1}{2}\, \int \int d \tau_1 d \tau_2 T\{H_{pd}(\tau_1) H_{pd}(\tau_2)\}.

294: \label{subtr}

295: \end{equation} 

296: These contributions need to be subtracted from $S_4(\beta)$, to get the fourth-order (in $V_i$)

297: contributions in the effective Hamiltonian. The most important contribution 

298: is the effective exchange interaction between localized spins:

299: \begin{equation}

300: H^{(4)}_{ex} = - \sum_{n \neq m} J(\bm{R}_n - \bm{R}_m) \bm{S}_n \bm{S}_m

301: \label{exchangeint}

302: \end{equation}

303: This exchange interaction is the result of 2 time-ordered processes shown in Figs. \ref{fig4orda}, \ref{fig4ordb}.

304: One is the superexchange (Fig. \ref{fig4orda}), which is a result of the localized spins exchanged

305: through the empty conduction band. The other process is the Bloembergen-Rowland term\cite{BR} (Fig. \ref{fig4ordb}),

306: an exchange process through both conduction and valence bands. The form of these contributions will be 

307: discussed in greater detail in sections IV and V below.

308: 

309: In addition, other interesting contributions arise as a result of fourth-order processes, such as p-d scattering by

310: spins on two centers,

311: \begin{widetext}

312: \begin{equation}

313: H^{(4)}_{pd2c} = \frac{V_1^2}{\Delta_1^2} \sum_{n \neq m, \sigma \sigma'} t(\bm{R}_n - \bm{R}_m)

314: \Psi^{\dagger}_{m \sigma' 1} \Psi_{n \sigma 1} [\bm{S}_n \bm{S}_m \delta_{\sigma' \sigma} + 

315: i ([\bm{S}_m \bm{S}_n] \hat{\bm{\sigma}}_{\sigma' \sigma})], 

316: \end{equation}

317: \end{widetext}

318: nontrivial local contribution,

319: \begin{widetext}

320: \begin{equation}

321: H^{(4)}_{local} = - \frac{V_1^4}{\Delta_1^3} \sum_{n \sigma \rho \rho'}

322: \Psi^{\dagger}_{n \sigma 1} \Psi^{\dagger}_{n \rho' 1} \left[\frac{1}{2}\, \delta_{\rho' \rho}  +

323: (\hat{\bm{\sigma}}_{\rho' \rho} \bm{S}_n)\right] \Psi_{n \rho 1} \Psi_{n \sigma 1},

324: \end{equation}

325: \end{widetext}

326: 

327: \begin{figure}

328: \includegraphics[width=3in]{fig2a.eps}

329: %\includegraphics[width=3in]{fig2a.pdf}

330: \caption{The fourth-order antiferromagnetic superexchange contribution to the effective exchange interaction between two localized 

331: impurities.}

332: \label{fig4orda}

333: \end{figure}

334: 

335: \begin{figure}

336: \includegraphics[width=3in]{fig2b.eps}

337: %\includegraphics[width=3in]{fig2b.pdf}

338: \caption{The fourth-order Bloembergen-Rowland term in the effective exchange interaction.} 

339: \label{fig4ordb}

340: \end{figure}

341: 

342: and corrections to the energy of the local level and ground state energy, which are dropped.

343: These nontrivial terms, however, are higher order in carrier density, which is small in magnetic

344: semiconductors. Thus, they don't play any significant role in magnetism, and, hence, will not be discussed in

345: any detail. We refer the reader to references \cite{GS1, GS2, BG1}, where these terms were discussed in detail

346: in connection with the 3-band model of cuprate superconductors.

347: 

348: \section{Renormalization of local level due to hybridization.}

349: 

350: To clarify the new physics of an Anderson impurity, in comparison with 

351: $p-d$ magnetic impurity, let us first consider a single-level problem ($\bm{R}_n = 0$),

352: interacting only with the conduction band (i.e., we take $V_2=0$ for simplicity).

353: The Hamiltonian is defined on the manifold of wavefunctions $\{\Phi_0, \Psi_{1 \bm{p_1}},

354: \Psi_{1 \bm{p_2}}, ... \}$, which correspond to the

355: exact solution for a center in crystalline lattice in the absence of hybridization. These

356: wavefunctions form a complete orthogonal basis for the single-particle problem. We may now

357: introduce the hybridization as a perturbation into the Schr\"odinger equation. In the matrix representation,

358: the single-level Hamiltonian takes a very simple form:

359: \begin{equation}

360: H = \left\{ \begin{array}{cccc}

361: \epsilon_0 & v_1 & v_1 & \cdots \\

362: v_1 & \epsilon_{1 \bm{p}_1} & 0 & \cdots \\

363: v_1 & 0 & \epsilon_{1 \bm{p}_2} & \cdots \\

364: \vdots & \vdots & \vdots & \ddots 

365: \end{array} \right\}

366: \end{equation}

367: Here $v_1 \equiv V_1/N^{1/2}$. This Hamiltonian can be 

368: easily diagonalized, by looking for a solution for the ground state wave function as a linear

369: combination of all single-particle states:

370: \begin{equation}

371: \Psi = \chi_0 \Phi_0 + \sum_{\bm{p}} \{\chi^{(1)}_{\bm{p}} \Psi_{1 \bm{p}}\},

372: \end{equation}

373: where the coefficients obey the following set of equations:

374: \begin{eqnarray}

375: \label{spectrum}

376: (\epsilon_0 - \epsilon) \chi_0 + v_1 \sum_{\bm{p}} \chi_{\bm{p}} &=& 0 \\

377:  v_1 \chi_0 + (\epsilon_{1 \bm{p}} - \epsilon) \chi_{\bm{p}}&=& 0 \nonumber

378: \end{eqnarray}

379: 

380: \begin{figure}

381: \includegraphics[width=3in]{fig3.eps}

382: %\includegraphics[width=3in]{fig3.pdf}

383: \caption{Localized centers acquire a finite radius in the Anderson model. This radius can be as large

384: as 2-3 lattice spaces in magnetic semiconductors. An overlap of wavefunctions from two localized level

385: leads to the new physics, which is not captured by the traditional $p-d$ model.}

386: \label{fig3}

387: \end{figure}

388: 

389: Solving Eq.(\ref{spectrum}) gives the new position of the localized level:

390: \begin{equation}

391: \epsilon = \epsilon_0 - \frac{V_1^2}{N}\, \sum_{\bm{p}} \frac{1}{\epsilon_{1 \bm{p}} - \epsilon}

392: \label{level}

393: \end{equation}

394: Thus, in presence of hybridization the position of the impurity level is shifted. If the depth

395: of the level $\Delta_1 \ll D_1$, where $D_1$ is the bandwidth of the conduction band (which

396: is usually a valid assumption for semiconductors, where the band gap is much smaller than the

397: corresponding bandwidths of conduction and valence bands), the sum in Eq.(\ref{level}) can

398: be performed numerically. The correction to the energy level,

399: \begin{equation}

400: \epsilon - \epsilon_0 \sim \frac{V_1^2}{D_1},

401: \end{equation}

402: turns out to be small for small enough

403: hybridization parameter $V_1 \ll \sqrt{D_1 \Delta_1}$. It is responsible for a

404: slight downward shift of the energy level. Such corrections will thus be dropped

405: in our further discussion.

406: 

407: The wavefunction of the localized level, according to Eq.(\ref{spectrum}),

408: acquires an admixture to its decay in the form:

409: \begin{equation}

410: \delta \Phi_0(r) \sim V_1 m_1 a^2 \frac{a}{r}\, e^{-r \sqrt{2 m_1 \Delta_1}}.

411: \label{wavef}

412: \end{equation}

413: Here $a$ is the lattice constant (we assume a simple cubic lattice), $m_1$ is the effective mass 

414: of carriers in the empty band (for hole-doped materials, the valence band is the ``empty'' band, 

415: since it is empty of holes), $\Delta_1$ is the energy difference between impurity levels and the bottom 

416: of the empty band. 

417: Thus, a new length scale enters the problem,

418: \begin{equation}

419: R_0 = (2 m_1 \Delta_1)^{-1} \sim a \sqrt{D_1/\Delta_1},

420: \label{scale}

421: \end{equation}

422: which, while small compared to the average distance between doped carriers, may

423: considerably exceed interatomic distances (the large parameter is the square root of

424: the ratio of the bandwidth to the energy gap in a semiconductor), and become comparable to the average

425: distance between magnetic impurities. The overlap of the localized wave functions

426: gives rise to a new contribution to the exchange integral, which will be derived

427: below. The picture of finite-radius localized magnetic impurities with overlapping wavefunctions is

428: shown in Fig.\ref{fig3}.

429: 

430: Existence of a large (compared to lattice spacing) radius for deep impurities was first noticed by

431: Keldysh\cite{Keldysh}, who analyzed deep charged impurities in ordinary semiconductors.

432: In his case, however, this scale corresponded to ordinary Bohr radius for a deep impurity

433: ( modified by the fact that in that case one had to consider Dirac Hamiltonian for the

434: $k-p$ model of semiconductors). In our case, the large length scale comes from

435: the hybridization of the localized impurity level with conduction and valence bands.

436: We will also see below in Section VI that effective length scale, at which direct interaction between impurities

437: starts to matter, grows logarithmically with decreasing  temperature in the framework of percolation theory or

438: virial expansion. Thus, at low temperatures, such as Curie temperature, the effective range of interaction

439: between two impurities becomes even larger.

440: 

441: \section{The exchange Hamiltonian.}

442: 

443: \begin{figure}

444: \includegraphics[width=3in]{fig4.eps}

445: %\includegraphics[width=3in]{fig4.pdf}

446: \caption{Positions of conduction and valence bands.}

447: \label{bandposition}

448: \end{figure}

449: 

450: The problem of magnetism in the Anderson model can be reduced to the spin Hamiltonian

451: Eq.(\ref{exchangeint}), described by the processes shown in Figs \ref{fig4orda}, \ref{fig4ordb}. In case

452: when the empty band (the "empty" band can also be the valence band, in case of hole doping)

453: gets partially filled by carriers (electrons or holes), the time-ordered

454: processes shown in Figs \ref{fig4orda}, \ref{fig4ordb}

455: include all possible contributions - the familiar RKKY exchange interaction, the superexchange,

456: and the Bloembergen-Rowland interaction. In presence of carriers, the superexchange gets modified

457: by the RKKY exchange term. We shall see below, that the Bloembergen-Rowland interaction  also gets

458: modified in presence of dopants. This modification is not accounted for by the

459: RKKY interaction. As we have discussed in the previous section, localized impurities in the

460: Anderson model acquire a large radius (compared with the lattice constant, $a$). Thus, exchange

461: interaction between localized spins is, in general, long range. This would allow

462: one to use mean field to describe the ferromagnetic ordering, if the concentration of localized spins

463: is high enough. On the other hand, if the concentration of magnetic impurities is low, one can

464: use percolation theory or virial expansion. 

465: 

466: Let us first consider exchange integrals given by the processes shown in

467: Figs \ref{fig4orda}, \ref{fig4ordb} for a semiconductor with conduction and valence

468: bands separated by some general reciprocal space wavevector $\bm{Q}$, as shown in Fig.\ref{bandposition}.

469: When $\bm{Q}=0$, it is a direct band gap semiconductor.

470: Surprisingly, all exchange processes can be written in a relatively compact way:

471: \begin{equation}

472: J(R) = \sum_{i,j} J_{ij}(R),

473: \label{exchch}

474: \end{equation}

475: where

476: \begin{widetext}

477: \begin{equation}

478: J_{ij} = 2 \frac{\left|V_i\right|^2 \left|V_j\right|^2 a^6}{(2 \pi)^6}\,

479: \int d^3 \bm{p} d^3 \bm{q} \frac{[1 - n(\epsilon_{\bm{p} i} - \mu)] e^{i \bm{q} \bm{R}}}{

480: (\epsilon_{\bm{p} i} - \epsilon_{\bm{p}+\bm{q} j})(\epsilon_{\bm{p} i} - \epsilon_0)^2}.

481: \label{exch}

482: \end{equation}

483: \end{widetext}

484: 

485: \begin{figure}

486: \includegraphics[width=3in]{fig5.eps}

487: %\includegraphics[width=3in]{fig5.pdf}

488: \caption{Superexchange contribution to exchange integral in q-space; modification by RKKY in presence of carriers

489: is shown by a dotted line.}

490: \label{superexchange}

491: \end{figure}

492: \begin{figure}

493: \includegraphics[width=3in]{fig6.eps}

494: %\includegraphics[width=3in]{fig6.pdf}

495: \caption{Bloembergen-Rowland contribution to exchange integral in q-space; modification in presence of carriers

496: is shown by a dotted line.}

497: \label{BlR}

498: \end{figure}

499: 

500: Here ${i,j}$ are the band indices for conduction and valence bands, $\mu$ is the chemical potential,

501: $n(\epsilon - \mu)$ is the Fermi-Dirac distribution function. The empty band corresponds to $i=1$,

502: while the filled band has the band index $i=2$. We can see now, that for $T \ll \Delta_0$, where $\Delta_0$ is the band gap,

503: the contributions $J_{21}$ and $J_{22}$ are absent. $J_{11}$ is the superexchange, which corresponds to 

504: the process shown in Fig. \ref{fig4orda}. In presence of carriers, this process, and the corresponding

505: expression, also includes the RKKY contribution. When carriers are present, the 

506: effective Hamiltonian can be written in terms of impurity spins only, with carriers "integrated out".

507: In this case, impurity spins interact with the exchange Hamiltonian Eqs.(\ref{exchch}),(\ref{exch}), which

508: includes RKKY. If, however, we wish to retain carriers, and the carrier-impurity p-d exchange interaction,

509: as we described in section II, then the RKKY part will be the part $S'(\beta)$ in the 4-th order

510: (see Eq.(\ref{subtr})), which has to be subtracted, and the indirect exchange interaction will be

511: described by Eq.(\ref{exch}) at zero doping, i.e., with the factor $[1 - n(\epsilon_{\bm{p} i} - \mu)]$ in

512: Eq.(\ref{exch}) replaced by $1$. These two descriptions are only equivalent when the chemical

513: potential for doped carriers, $\mu \ll \Delta_1$. Basically, when the doped carriers are retained, we "integrate out" electrons

514: and holes in the original Anderson Hamiltonian up to the scale $\mu$. When we consider the Hamiltonian

515: for localized spins only, we integrate out all carriers. 

516: It makes sense to define the RKKY interaction

517: for the Anderson model as the difference between the doped and undoped cases in Eqs.(\ref{exchch}),(\ref{exch}).

518: When $\mu \sim \Delta_1$, the Anderson model \textit{cannot} be reduced to the $p-d$ model, except near the Fermi surface. 

519: The Anderson model expression for the RKKY interaction in real space  

520: will then \textit{differ} from the corresponding $p-d$ model expression

521: at \textit{short} distances, which are important for ferromagnetism. We won't consider

522: this subtle point any further, since we will always assume the carrier density in FS to be small,

523: i.e., $\mu \ll \Delta_1$.  In this section and the next section we will deal with an

524: effective Hamiltonian for spins only, i.e., when all carriers are integrated out. This procedure

525: for the Anderson model is always justified (although, as we commented above, a reduction to the $p-d$ model is \textit{not}).

526: $J_{12}$ is the Bloembergen-Rowland interaction, which

527: also gets somewhat modified by doped carriers. Note that the expression for the exchange integral Eq.(\ref{exch}),

528: obtained from the Anderson model, differs from the original expression of Bloembergen and Rowland \cite{BR};

529: it includes the term $(\epsilon_{\bm{p} 1} - \epsilon_0)^2$ in the denominator, which makes the 

530: exchange integral in the Anderson model long range. Both contributions to exchange integral can be evaluated 

531: numerically and analytically; they are shown schematically in Figs \ref{superexchange} and \ref{BlR}. 

532: 

533: We see that the superexchange is, in general, antiferromagnetic, while the Bloembergen-Rowland 

534: contribution favors a spin density wave with a wavevector $\bm{Q}$, separating conduction and valence bands.

535: Both exchange processes are, in general, of the same order, and their relative strength is determined by

536: the corresponding hybridization parameters $V_1$ and $V_2$. The superexchange integral in the $\bm{q}$ - space,

537: in the absence of carriers, can be easily found analytically:

538: \begin{equation}

539: J_{11}(\bm{q}) = - \frac{2 V_1^4 a^3 m_1^3}{\pi \sqrt{2 m_1 \Delta_1}}\, \frac{1}{q^2 + 8 m_1 \Delta_1}

540: \label{superex}

541: \end{equation}

542: In presence of carriers, $J_{11}(\bm{q})$ gets modifed by the RKKY interaction:

543: \begin{widetext}

544: \begin{equation}

545: \delta J_{11}(\bm{q}) \equiv J_{RKKY}(\bm{q}) =  \frac{V_1^4 a^3 m_1 p_F}{(2 \pi)^2 \Delta_1^2} \,

546: \left\{ 1 + \left(\frac{p_F}{q}\, - \frac{q}{4 p_F}\, \right) \ln 

547: \left(\frac{1 + \frac{q}{2 p_F}\,}{1 - \frac{q}{2 p_F}\,}\right)\right\},

548: \end{equation}

549: \end{widetext}

550: where $p_F = \sqrt{2 m_1 \mu}$ is the Fermi momentum for doped carriers. The superexchange and RKKY contributions

551: can be easily rewritten in real space:

552: \begin{equation}

553: J_{11}(R) = J_{se}(R) + J_{RKKY}(R),

554: \label{j11}

555: \end{equation}

556: with

557: \begin{equation}

558: J_{se}(R) = - \frac{V_1^4 a^6 m_1^3}{2 \pi^2 \sqrt{2 m_1 \Delta_1} R}\, \exp\{-\sqrt{8 m_1 \Delta_1} R\},

559: \label{sesese}

560: \end{equation}

561: and a standard expression for the RKKY interaction, taking into account that $J_{pd} \equiv 2 V_1^2/\Delta_1$

562: (note that in Eq.(\ref{exchangeint}) we sum over each pair of impurities twice):

563: \begin{equation}

564: J_{RKKY}(R) = - J_{RKKY} F(2 p_F R),

565: \end{equation}

566: where

567: \begin{equation}

568: F(x) = \frac{\cos(x)}{x^3}\, - \frac{\sin{x}}{x^4},

569: \end{equation} 

570: and

571: \begin{equation}

572: J_{RKKY} = \frac{V_1^4 a^6 m_1 p_F^4}{\pi^3 \Delta_1^2}.

573: \end{equation}

574: We define the RKKY interaction in the Anderson model formulation as the difference between $J_{11}(R)$

575: in doped and undoped cases, given by Eq.(\ref{exch}). As we emphasized above, since it depends explicitly 

576: on impurity energy level,  when $\Delta_1 \sim \mu$, this expression differs from the usual RKKY form. 

577: Its asymptotic behavior at large distances, certainly, does not change.

578: In what follows we assume that the carrier concentration is always low, $\mu \ll \Delta_i$, so that

579: $J_{RKKY}(R)$ is given by the standard expression. 

580: Note that if $J_{pd}$ and the level position are known, the exchange integral Eq.(\ref{j11}) contains

581: no free parameters:

582: \begin{widetext}

583: \begin{equation}

584: J_{11}(R) =  - J_{RKKY} 

585: \left(F(2 p_F R) + \frac{\pi (m_1 \Delta_1)^{3/2}}{2 \sqrt{2} p_F^4 R}\, e^{-2 \sqrt{2 m_1 \Delta_1} R} \right).

586: \label{11ex}

587: \end{equation}

588: \end{widetext}

589: Similar to superexchange, the Bloembergen-Rowland interaction can also be written in two parts. The first

590: part is from the empty and filled bands, while the second part takes doped carriers into account:

591: \begin{equation}

592: J_{12}(\bm{q}) = J_{BR}(\bm{q}) + J'_{12}(\bm{q}).

593: \label{12ex}

594: \end{equation}

595: The first contribution in Eq.(\ref{12ex}) favors a spin density wave with a wavevector $\bm{Q}$, separating

596: conduction and valence bands, and has a rather cumbersome form:

597: \begin{widetext}

598: \begin{equation}

599: J_{BR}(\bm{q}) =  \frac{a^3 V_1^2 V_2^2 m_1^2 m_2}{\pi \sqrt{2 m_1 \Delta_1}}\,

600: \left\{\frac{m_1 \Delta_1 + m_2 \Delta_1 + \frac{q^2}{2}\, + m_2 \Delta_0 - \sqrt{2 (m_1 + m_2) \Delta_1}

601: \sqrt{2 m_2 \Delta_0 + q^2 \frac{m_2}{m_1 + m_2}\,}}{\Delta_1^2(m_1+m_2)^2 + \Delta_1 (q^2 (m_1-m_2) -

602: 2 m_2 (m_1+m_2) \Delta_0) + \left(\frac{q^2}{2}\,+ m_2 \Delta_0 \right)^2}\right\}. 

603: \label{BRl}

604: \end{equation}

605: \end{widetext}

606: Here and below in this section the notation will be slightly different; $\bm{q}$ is now the \textit{difference} 

607: from $\bm{Q}$ (the wavevector separating the bottoms of the two bands), i.e., we

608: change our notation in the following way: $\bm{q - Q} \longrightarrow \bm{q}$. 

609: In general, the Bloembergen-Rowland contribution Eq.(\ref{BRl}) is quite cumbersome, which makes it impossible to

610: find an explicit analytical expression

611: in real space, except in some limiting cases. However, the leading asymptotic behavior of $J_{BR}(R)$ at large distances 

612: can be derived. If the deep impurity

613: level lies inside the gap, $\Delta_1 < \Delta_0$, we find an exponential decay of exchange correlations:

614: \begin{widetext}

615: \begin{equation}

616: J_{BR}(\bm{R}) \simeq \frac{a^6 V_1^2 V_2^2 m_1 m_2}{2 \sqrt{2} \pi^2 R}\,\left(\sqrt{\frac{m_1}{\Delta_1}\,} -

617: \sqrt{\frac{m_2}{\Delta_0 - \Delta_1}\,} \right) \cos{(\bm{Q} \cdot \bm{R})} e^{-R/R_1},

618: \label{undBR}

619: \end{equation}

620: \end{widetext}

621: where the range of the integral is:

622: \begin{equation}

623: R_1 = \frac{1}{\sqrt{2 m_1 \Delta_1} + \sqrt{2 m_2 (\Delta_0 - \Delta_1)}}.

624: \label{r1}

625: \end{equation}

626: On the other hand, when the deep impurity level enters the filled band, $\Delta_1 > \Delta_0$,  the Bloembergen-Rowland 

627: contribution has an oscillating decaying asymptotic:

628: \begin{widetext}

629: \begin{equation}

630: J_{BR}(\bm{R}) = \frac{a^6 V_1^2 V_2^2 m_1 m_2}{ \sqrt{2} \pi^2 R}\, \left(\sqrt{\frac{m_1}{2 \Delta_1}\,} \cos\frac{R}{R_{c1}}\, +

631: \sqrt{\frac{m_2}{2 (\Delta_1-\Delta_0)}\,} \sin\frac{R}{R_{c1}}\,\right) \cos{(\bm{Q} \cdot \bm{R})} e^{- R/R_{c0}},

632: \end{equation}

633: \end{widetext}

634: where

635: \begin{equation}

636: R_{c1} = \frac{1}{\sqrt{2 m_1 (\Delta_1 - \Delta_0)}}, \ \ \ R_{c0} = \frac{1}{\sqrt{2 m_1 \Delta_1}}\, .

637: \label{rc1rc0}

638: \end{equation}

639: 

640: The contribution from doped carriers to $J_{12}$ can be derived as well:

641: \begin{equation}

642: J'_{12}(\bm{q}) \simeq - \frac{ a^3 V_1^2 V_2^2 p_F^3}{3 \pi^2 \Delta_1^2 (\Delta_0 + \frac{q^2}{2 m_2}\,)}\,

643: \end{equation} 

644: In real space, this contribution becomes:

645: \begin{equation}

646: J'_{12}(\bm{R}) \simeq - \frac{p_F^3 a^6 V_1^2 V_2^2 m_2}{6 \pi^3 \Delta_1^2 R} \cos{(\bm{Q} \cdot \bm{R})} e^{-\sqrt{8 m_2 \Delta_0} R}

647: \label{dopBR}

648: \end{equation}

649: 

650: We see that if $\bm{Q}=0$, i.e., when our FS has a direct band gap, the Bloembergen-Rowland mechanism gives

651: a large ferromagnetic short-range contribution to the exchange integral.

652: At high enough concentration of impurities, ferromagnetic properties are then determimed by the value of exchange

653: integral at $\bm{q}=0$. This value can be easily written down (it is only valid for a direct band gap FS):

654: 

655: \begin{widetext}

656: \begin{equation}

657: J(\bm{q}=0) = J_{11}(\bm{q}=0) + J_{12}(\bm{q}=0)

658:  = \frac{2 a^3 V_1^2 m_1^2}{ \pi \sqrt{2 m_1 \Delta_1}}\, \left(- \frac{V_1^2}{8 \Delta_1}\, 

659: + \frac{V_2^2 m_2}{ (\sqrt{2 (m_1 + m_2) \Delta_1} + \sqrt{ 2 m_2 \Delta_0})^2}\, \right) 

660: \label{qzero}

661: \end{equation}

662: \end{widetext}

663: 

664: In this section we have derived explicitly the interaction between two magnetic impurities in a semiconductor

665: with one conduction and one valence band. Typically, the band structure of semiconductors is more complex

666: than that (for example, GaAs has light and heavy hole bands). 

667: This situation is considered in the next section.

668: 

669: 

670: \section{What if a semiconductor has more than two bands?}

671: 

672: When a semiconductor has many bands, the analysis is just as straightforward as in case

673: of two bands considered in the previous section. The exchange integral between two impurities

674: is still given by Eqs(\ref{exchch}),(\ref{exch}), where the sum now goes over all pairs of band indices. The case when

675: $i=j$ is an ``empty'' band index corresponds to a superexchange contribution

676: (Fig.\ref{fig4orda}), considered in detail in the previous section; $J_{ij} = 0$ when $i$ is a ``filled''

677: band index, because of the Pauli principle; $J_{ij}$ is the Bloembergen-Rowland contribution (Fig.\ref{fig4ordb}), 

678: when $i$ is an ``empty''

679: band index, and  $j$ is a ``filled'' band index. The contribution of a new type appears when one has

680: two or more ``empty'' bands in a semiconductor. Then there will be a superexchange contribution of the type shown in

681: Fig.\ref{fig4orda}, where now the carriers are exchanged through two different ``empty'' bands - the process (1) in

682: Fig.\ref{fig4orda} puts a carrier from one magnetic impurity into the first ``empty'' band, while the process (2)

683: puts a carrier from the other magnetic impurity into the second ``empty'' band. Naturally,

684: there are two such contributions, $J_{ij}$ and $J_{ji}$, where $i \neq j$ indices correspond to two

685: different ``empty'' bands. The exchange Hamiltonian in any FS

686: is a sum of all pairwise contributions listed above. 

687: 

688: Let us now consider the new two-band superexchange contribution in detail.

689: In the most general case, 

690: the bottom of the second ``empty'' band is shifted from the bottom of the first ``empty'' band by a 

691: wavevector $\bm{Q}$. The bottoms of the two bands also lie at two different positive

692: energies $\Delta_1$ and $\Delta_2$ relative to impurity level, unless there is a symmetry-related 

693: degeneracy, and the carriers in the first and second bands have different 

694: effective masses $m_1$ and $m_2$. 

695: 

696:   Let us first consider the simplified case when the bands 1 and 2 are identical (i.e., $\Delta_1 = \Delta_2$ and

697: $m_1 = m_2$), but separated by a  wavevector $\bm{Q}$. Then we don't have to do the calculation, since  

698: $\epsilon_{\bm{p}2} = \epsilon_{\bm{p-Q} 1}$, and it can

699: be easily seen from Eq.(\ref{exch}), that

700: \begin{equation}

701: J_{12}(\bm{q}) = J_{11}(\bm{q-Q}),

702: \end{equation}

703: and

704: \begin{equation}

705: J_{12}(R) = J_{11}(R) e^{i \bm{Q} \bm{R}}.

706: \end{equation}

707: Summing all contributions from 2 identical empty bands, since $J_{11}(R) = J_{22}(R)$,

708: we get:

709: \begin{equation}

710: J_{2bse} = 2 J_{11}(R) [1 + \cos{(\bm{Q} \cdot \bm{R})}].

711: \label{2bse}

712: \end{equation}

713: This also includes RKKY-type contribution for two identical bands at finite doping, described by:

714: \begin{equation}

715: J_{12 RKKY} + J_{21 RKKY}  = 2 J_{RKKY}(R) \cos{(\bm{Q} \cdot \bm{R})},

716: \label{2bse1}

717: \end{equation}

718: and the ordinary RKKY interaction from both bands.

719: It is rather obvious from Eq.(\ref{2bse1}), that at a finite density of carriers there are ordinary one-band RKKY contributions from

720: both bands. The two-band RKKY contribution oscilates much more rapidly in space, 

721: as $\cos(\bm{Q}{R}))$, unless $\bm{Q = 0}$. In the latter case, the two-band RKKY contribution just enhances

722: the contributions from the two separate bands. In the case when $\bm{Q}=0$, the $12$ and $21$ exchange integrals take

723: the same form as shown in Fig.\ref{superexchange}. For $\bm{Q \neq 0}$, these contributions have the

724: same form in $\bm{q}$-space as for $\bm{Q}=0$, centered at the wavevector $\bm{q}=\bm{Q}$.

725: 

726:   Now that the form of new two-band contributions to the exchange integral has become clear from the simplified case, 

727: let us consider a more 

728: general case of two completely different ``empty'' bands. The bottom of the second band is shifted from

729: the bottom of the first band by the wavevector  $\bm{Q}$. Of course, $12$ and $21$ contributions to $J(\bm{q})$

730: will still be of the form shown in Fig.\ref{superexchange}, with the bottom at $\bm{q} = \bm{Q}$, but the corresponding 

731: expressions become more combersome. Let us assume that $\Delta_1 < \Delta_2$, i.e. the bottom of the first band

732: lies below the bottom of the second band, and the first band gets filled

733: by carriers first. We consider three separate cases, (1) both bands are empty; (2) the first band gets

734: filled by carriers, but the second band is empty, and (3) both bands get partially filled by carriers.

735: When $\Delta_1 = \Delta_2$, i.e., the two bands are symmetry-related, we only have cases (1) and (3). 

736: As before, we assume that the carrier

737: concentration is very small, i.e., the Fermi energy for doped carriers is much smaller than any other 

738: energy scale in the problem, except temperature.

739: 

740: \vglue 0.5cm

741: 

742: \noindent

743: \textit{(1) Both bands are empty.}

744: 

745: \vglue 0.5cm

746: 

747:   Let us now consider the exchange interaction arising from two empty bands. The integral 

748: in Eq.(\ref{exch}) can be easily calculated:

749: \begin{widetext}

750: \begin{equation}

751: J_{12+21}(\bm{q}) = - \frac{\sqrt{2} V_1^2 V_2^2 m_1 m_2 a^3}{(\bm{q}-\bm{Q})^2 + 

752: (\sqrt{2 m_1 \Delta_1} + \sqrt{2 m_2 \Delta_2})^2}\, \left(\sqrt{\frac{m_1}{\Delta_1}\,} 

753: +\sqrt{\frac{m_2}{\Delta_2}\,}\right).

754: \end{equation}

755: \end{widetext}

756: In the coordinate space, it takes the following form:

757: \begin{equation}

758: J_{12+21}(\bm{R}) = - V_{12} \cos{(\bm{Q}\cdot\bm{R})} \frac{1}{R} \exp{( - \frac{R}{R_{12}}\, )},

759: \label{2eb}

760: \end{equation}

761: where the range of this interaction is given by:

762: \begin{equation}

763: R_{12} = \frac{1}{\sqrt{2 m_1 \Delta_1} + \sqrt{2 m_2 \Delta_2}}, 

764: \end{equation}

765: and

766: \begin{equation}

767: V_{12} = \frac{V_1^2 V_2^2 a^6 m_1 m_2}{2 \sqrt{2} \pi^2}\, \left(\sqrt{\frac{m_1}{\Delta_1}\,} + 

768:  \sqrt{\frac{m_2}{\Delta_2}\,} \right)

769: \end{equation}

770: 

771: \vglue 0.5cm

772: 

773: \noindent

774: \textit{(2) One band gets filled.}

775: 

776: \vglue 0.5cm

777: 

778: Let us assume in this section that band one gets filled first, i.e., $\Delta_1 < \Delta_2$.

779: We also assume that the chemical potential (counted from the bottom of band one), $\mu_1 \ll \Delta_2 - \Delta_1$,

780: i.e., the number of carriers is low. Of course, the main part of the exchange inegral is still given by Eq.(\ref{2eb}).

781: However, at finite doping there are corrections. Obviously, since only band one gets carriers, $\delta J_{21} = 0$.

782: At small filling, RKKY-like correction can be easily calculated:

783: \begin{equation}

784: \delta J_{12}(q) \simeq \frac{2 V_1^2 V_2^2 a^3 P_{F1}^3 m_2}{3 \pi^2 \Delta_1^2 (2 m_2 [\Delta_2 - \Delta_1] + q^2)}.

785:  \end{equation} 

786: Here $q$ denotes the difference from $\bm{Q}$, the wavevector separating the bottoms of the two bands.

787: In coordinate space it can be written as:

788: \begin{equation}

789: \delta J_{12}(\bm{R}) \simeq

790: \frac{V_1^2 V_2^2 a^6 p_{F1}^3 m_2}{6 \pi^3 \Delta_1^2 R} e^{- R/R_g} \cos{\bm{Q} \cdot \bm{R}},

791: \end{equation} 

792: where

793: \begin{equation}

794: R_g = \frac{1}{\sqrt{2 m_2 (\Delta_2 - \Delta_1)}}\,

795: \end{equation}

796: 

797: \vglue 0.5cm

798: 

799: \noindent

800: \textit{(3) Both bands get filled.} 

801: 

802: \vglue 0.5cm

803: 

804: When $\Delta_1 = \Delta_2 = \Delta$ by symmetry, both bands get filled by carriers simultaneously.

805: The result of this is an RKKY-like correction to exchange integral  Eq.(\ref{2eb}), which is given

806: by Eq.(\ref{2bse1}) for two identical bands. If the bands have different masses, the interaction

807: is RKKY-like long range, but the general expression is quite messy. Here we give the its value at 

808: $\bm{q}=\bm{Q}$, which is important for ferromagnetism, when $\bm{Q} = 0$:

809: 

810: \begin{equation}

811:  \delta J_{12+21}(\bm{q}=\bm{Q}) = \frac{2 V_1^2 V_2^2 a^3 \sqrt{2 \mu} m_1 m_2}{\pi^2 \Delta^2 (\sqrt{m_1} + \sqrt{m_2})}\,

812: \end{equation}

813:  

814: To summarize this section, the exchange interaction in any magnetic semiconductor is composed of pairwise contributions:

815: the superexchange contribution through one ``empty'' band, the Bloembergen-Rowland contribution from one ``empty'' and one filled band,

816: and a pairwise contribution from two different ``empty'' bands. The last contribution was considered in this section. 

817: It includes two-band long-range RKKY-like interaction at finite doping, which is usually not taken into account in

818: the $p-d$ Hamiltonian. Note that when $\bm{Q} \neq 0$, the two-band RKKY interaction will favor spin glass order,

819: \textit{not} ferromagnetism, since it oscillates very rapidly in real space.

820: 

821: \section{The mean field approximation, virial expansion, and percolation.}

822: 

823:    Now that we have obtained all terms in the exchange interaction Eq.(\ref{exch}), we can proceed to

824: calculate magnetic properties of ferromagnetic semiconductors. In general, the RKKY exchange interaction

825: is always long-range. The shorter-range contributions may or may not be treated in mean field, depending

826: on the ratio of their range to the average inter-impurity distance. We assume here, for simplicity,

827: that one of indirect exchange processes considered in the previous two sections dominates the short-range

828: physics. In case of FS with indirect band gap, the Bloembergen-Rowland exchange contribution favors 

829: antiferromagnetism with a lattice wavevector $\bm{Q}$. It oscillates very rapidly in real space. 

830: Since impurities are distributed randomly,  it would favor a spin glass ordering.

831:  The contributions at $\bm{Q}=0$, which influence ferromagnetism, are the RKKY

832: and the superexchange. From Eq.(\ref{11ex}) one can easily see that, when impurity concentration

833: $n_i \gg (m_1 \Delta_1)^{3/2}$, the superexchange contribution totally suppresses ferromagnetism 

834: brought about by the

835: RKKY exchange interaction. On the other hand, when $n_i \ll (m_1 \Delta_1)^{3/2}$, the superexchange

836: at an average distance between impurities is

837: suppressed, and the RKKY interaction gives rise to ferromagnetism, as in the ordinary $pd$-model. Corrections

838: due to superexchange can then be calculated using the virial expansion approach\cite{BA}. We can

839: rewrite the dominant exchange contribution from Eq.(\ref{11ex}) in the following form:

840: \begin{equation}

841: J(R) = - J_{RKKY} F(2 p_F R) - V_0 \frac{R_0}{R}\, e^{-R/R_0},

842: \label{generic}

843: \end{equation}

844: where $V_0 = \pi J_{RKKY}/(64 p_F^4 R_0^4)$, $R_0 = 1/\sqrt{8 m_1 \Delta_1}$, in case when it is given by one empty band only.

845: If contributions from more than one ``empty'' band are important, $V_0$ and $J_{RKKY}$ are, in general, unrelated to each

846: other. The reason for this is that the largest short range contribution typically comes from the lightest bands, 

847: which would produce the longest range indirect exchange (if all hybridization parameters

848: are of the same order). On the other hand, the dominant

849: RKKY contribution is the one from the heaviest band, since the carriers in that band have the largest Fermi wavevector; there

850: will also be additional RKKY contributions from many bands, which were analyzed in the previous section.

851: In what follows we consider the simplest case, when only one heavy band is relevant for RKKY. 

852: 

853: In case of a FS with a direct band gap, the Bloembergen-Rowland mechanism gives rise to ferromagnetism, and

854: we can either have a ferromagnetic or an antiferromagnetic short-range exchange contribution, depending

855: on the relative strength of corresponding exchange processes. If the Boembergen-Rowland term

856: dominates the physics at short distances, the impurity spin Hamiltonian becomes:

857: \begin{equation}

858: J(R) = - J_{RKKY} F(2 p_F R) + V_{BR} \frac{R_1}{R}\, e^{-R/R_1},

859: \end{equation}

860: In general, the Bloembergen-Rowland term Eq.(\ref{12ex}) also has an antiferromagnetic contribution from

861: doped carriers. It then can be rewritten in the following form:

862: \begin{equation}

863: J_{12}(R) = V_{BR} \frac{R_1}{R} e^{-R/R_1} - V_{dop} \frac{R_g}{R_0} e^{-R/R_g},

864: \end{equation}

865: where $R_g = 1/\sqrt{8 m_2 \Delta_0}$, $V_{BR}$ and $V_{dop}$ are given by the coefficients

866: in front of exponents in Eq.(\ref{undBR}) and Eq.(\ref{dopBR}). However, since

867: \begin{equation}

868: V_{dop}/V_{BR} \sim \frac{p_F^3}{(m_1 \Delta_1)^{3/2}} \ll 1,

869: \end{equation}

870: while $R_g$ is still quite small, the contribution of doped carriers to $T_c$ through the Bloembergen-Rowland mechanism 

871: can be neglected. Thus, if one exchange process dominates the physics are short distances, the short-range 

872: exchange contribution has the same form, and can only differ in sign.

873: In what follows, we consider the general form of exchange integral Eq.(\ref{generic}), assuming that the exchange

874: constant $V_0$, its sign, and its range $R_0$ are those of the dominant indirect exchange contribution.

875: Note again that in our definition of exchange integral we sum over impurities twice.

876: We will also consider the case when the short-range part in Eq.(\ref{generic}) is rapidly oscillating,

877: \begin{equation} 

878:  J_{SR} = - V_0 \frac{R_0}{R}\, e^{-R/R_0} \cos{(\bm{Q}\cdot\bm{R})}.

879: \label{Osc}

880: \end{equation}

881: As we have seen above, this happens when  two bands separated by wavevector $\bm{Q}$ contribute

882: the most to the short-range exchange interaction. $V_0$ has a negative sign for the Bloembergen-Rowland

883: contribution, and a positive sign for the superexchange.

884: When more than one process is important for short-range physics,

885: the problem can always be solved numerically for a given set of parameters.

886: Here we obtain an analytical solution in several limiting cases.

887: 

888: \vglue 0.5cm

889: 

890: \noindent

891: \textit{(1) Dilute system with almost no carriers.}

892: 

893: \vglue 0.5cm

894: 

895: In the absence of carriers, the type of order and $T_c$ is determined by the short-range part of 

896: interaction. If ferromagnetic Bloembergen-Rowland interaction dominates at short distances, Curie temperature

897: is approximately given by the ferromagnetic interaction taken at the average distance between impurities

898: \cite{Korenblit}:

899: \begin{equation}

900: T_c \simeq  2.3 V_0 S^2 R_0 n_i^{1/3} \, e^{- \frac{0.87}{R_0 n_i^{1/3}}} \ \ V_0 > 0,

901: \label{perc}

902: \end{equation}

903: which is valid when $n_i \ll 1/R_0^3$. A comparison with Eq.(\ref{Curie}) shows that this

904: gives the following condition on the number of carriers:

905: \begin{eqnarray}

906: \label{cond}

907: p_F &=& (3 \pi^3 n_e)^{1/3} \ll p_{0F} \\

908: p_{0F}&=&\frac{27.6 V_0 \pi^2 S R_0}{(J^{pd}_{\bm{q}=0})^2 (S+1) n_i^{2/3}}\, e^{-0.87/(R_0 n_i^{1/3})} \nonumber

909: \end{eqnarray} 

910: When the short-range interaction is antiferromagnetic, it favors spin glass order, and 

911: ferromagnetism is absent when there are no or almost no carriers.

912: The condition on the number of carriers for ferromagnetism to be absent is then given by:

913: \begin{equation}

914: p_F \ll p_{0SG}=\frac{27.6 |V_0| \pi^2 R_0}{(J^{pd}_{\bm{q}=0})^2 n_i^{2/3}}\, e^{-0.87/(R_0 n_i^{1/3})}

915: \label{cond1}

916: \end{equation}

917: Finally, for a rapidly oscillating short-range part Eq.(\ref{Osc}) this condition can be rewritten as:

918: \begin{equation}

919: p_F \ll p_{0SG}=\frac{27.6 |V_0| \pi^2 R_0}{(S+1)(J^{pd}_{\bm{q}=0})^2 n_i^{2/3}}\, e^{-0.87/(R_0 n_i^{1/3})}

920: \label{cond2}

921: \end{equation}

922: 

923: For the Anderson Hamiltonian, $J^{pd}_{q=0} = 2 V_1^2 a^3/\Delta_1$. Note that if the ratio $n_e/n_i$ is fixed,

924: $p_F$ grows much slower with $n_i$ than $p_0$. Thus, conditions in  Eqs.(\ref{cond}),(\ref{cond1}),(\ref{cond2})

925: are likely to be satisfied at some finite concentration of magnetic impurities, $n_c \ll  n_i \ll 1/R_0^3$.

926: For example, for antiferromagnetic or oscillating interactions, Eqs (\ref{cond1}),(\ref{cond2}) will define the concentation 

927: of impurities above which carrier-driven ferromagnetism disappears. 

928: 

929: \vglue 0.5cm

930: 

931: \noindent

932: \textit{(2) Dilute system with carriers.}

933: 

934: \vglue 0.5cm

935: 

936: When the carrier concentration is large, i.e., $p_F \gg p_0$ in Eqs.(\ref{cond}),(\ref{cond1}),(\ref{cond2}),

937: but the concentration of magnetic impurities is still small, $n_i \ll 1/R_0^3$, short-range interactions 

938: between magnetic impurities will result in a correction to $T_c$, which

939: can be calculated by virial expansion (see, for example, Ref.\cite{BA}). Following Ref.\cite{AG},

940: since the range of RKKY interaction is large, we can represent $p-d$ interaction between carriers 

941: and magnetic impurities by a mean field Zener Hamiltonian:

942: \begin{equation}

943: \hat{H}_{MF} =  - J^{pd}_{\bm{q}=0} \bm{s} \sum_i \bm{S}_i, 

944: \label{zener}

945: \end{equation}

946: where $\bm{s}$ is the density of ordered spin of the carriers, which is assumed to be constant

947: in space. In addition, there is a relatively short-range exchange interaction between impurity spins,

948: given by the processes described in Sections IV and V: 

949: \begin{equation}

950: \hat{H}_{exch} = - \sum_{ij} J(\bm{R}_i - \bm{R}_j) \bm{S}_i \bm{S}_j.

951: \label{exx}

952: \end{equation}

953: The short-range exchange integral does not include the carrier contribution, since it is already 

954: accounted for in the mean field Hamiltonian.

955: In the Zener model Eq.(\ref{zener}) , Curie temperature Eq.(\ref{Curie}) can be found by minimizing the free energy 

956: density of the system of carriers and spins\cite{AG},

957: \begin{equation}

958: F = F_e + F_i,

959: \end{equation}

960: with respect to $\bm{s}$, and finding when the solution at small $s$ first appears. Here

961: \begin{equation}

962: F_e= \frac{(2 \mu_B s)^2}{2 \chi_0},

963: \end{equation}

964: and $F_i$ is given by the usual Zeeman term,

965: \begin{equation}

966: F_i = - n_i T \ln{\frac{\sinh{[J^{pd}_{\bm{q}=0} s (S + 1/2)/T]}}{\sinh{[J^{pd}_{\bm{q}=0} s/(2 T)]}}\,} 

967: \end{equation}

968: 

969: To find the virial correction to Eq.(\ref{Curie}) from Eq.(\ref{exx}), we need to include into $F$ the

970: contribution from two magnetic impurities, when they are close enough:

971: \begin{equation}

972: F_{2i} = \frac{1}{2}\, n_i^2 \int d^3 \bm{R} [F_{2i}(\bm{R}) - 2 F_i (\bm{R})],

973: \end{equation}

974: and calculate it from the Hamiltonian

975: \begin{equation}

976: \hat{H} = \hat{H}_{MF} + \hat{H}_{exch}.

977: \end{equation}

978: This can be done, since the Zeeman Hamiltonian for two spins, interacting via direct

979: exchange interaction, can easily be solved. The integral over this solution, however,

980: can only be taken with logarithmic accuracy. The contribution from two impurities is important

981: when the distance between them is

982: \begin{equation}

983: R \le R_0 \ln{\frac{|V_0|/T}{\ln{(|V_0|/T)}}}.

984: \end{equation}

985: Finding $F_{2i}$, and repeating the minimization over $s$, we obtain, as expected, that a for ferromagnetic exchange

986: interaction ($V_0 > 0$) Curie temperature is enhanced:

987: \begin{equation}

988: \frac{\delta T_c[F]}{T_c} \simeq \frac{4 \pi S}{3 (S + 1)}\, n_i R_0^3 \ln^3{\frac{|V_0|/T_c}{\ln{(|V_0|/T_c)}}}.

989: \label{ferr}

990: \end{equation}

991: For an antiferromagnetic exchange interaction ($V_0 < 0$), the Curie temperature is reduced:

992: \begin{equation}

993: \frac{\delta T_c[A]}{T_c} \simeq - \frac{4 \pi}{3}\, n_i R_0^3 \ln^3{\frac{|V_0|/T_c}{\ln{(|V_0|/T_c)}}}.

994: \label{aferr}

995: \end{equation}

996: A rapidly oscillating exchange interaction Eq(\ref{Osc}) gives:

997: \begin{equation}

998: \frac{\delta T_c[O]}{T_c} \simeq - \frac{4 \pi}{3 (S+1)}\, n_i R_0^3 \ln^3{\frac{|V_0|/T_c}{\ln{(|V_0|/T_c)}}},

999: \label{oscill}

1000: \end{equation}

1001: which is independent of the sign of $V_0$.

1002: \vglue 0.5cm

1003: 

1004: \noindent

1005: \textit{(3) ``Dense'' system.}

1006: 

1007: \vglue 0.5cm

1008: 

1009: The most interesting situation is the case of a ``dense'' system of magnetic impurities,

1010: when  $n_i \gg (m_1 \Delta_1)^{3/2}$. Since the short-range part of the interaction could

1011: have a range much larger than the lattice spacing, it need not be really dense.  

1012: This requirement can be rewritten as $n_i \gg (\Delta_1 / D)^{3/2}$. In a dense system,

1013: the wave functions of the neighboring impurities overlap strongly, and the main exchange contribution

1014: arises as a result of this overlap, not the RKKY interaction through free carriers. Then

1015: ferromagnetism arises even when no carriers are present, if the Bloembergen-Rowland exchange process 

1016: dominates the physics at short distances ($V_0>0$):

1017: \begin{equation}

1018: T_c = \frac{2 S (S+1) n_i}{3}\, J_{\bm{q}=0} = \frac{8 \pi S (S+1) n_i}{3} V_0 R_0^3.

1019: \end{equation}

1020: Note that $T_c$ does not depend on the carrier

1021: concentration, and should be much higher than that resulting from the RKKY

1022: interaction. 

1023: 

1024: The Bloembergen-Rowland exchange integral may be weaker than superexchange. In that case,

1025: ferromagnetism in a ``dense'' system is suppressed, and spin glass order is favored. This is always 

1026: the case for indirect band gap 

1027: semiconductors, where the short-range ferromagnetic exchange is absent.

1028:  

1029: The Bloembergen-Rowland mechanism in case of direct band gap 

1030: necessarily leads to an increase of maximum $T_c$ as a

1031: function of concentration of magnetic impurities $n_i$. 

1032: 

1033: \section{Application to GaAs:Mn.}

1034: 

1035: Application of the Anderson model to a real system, such as Ga$_{1-x}$Mn$_x$As is somewhat more 

1036: involved than the model that we considered above, since Mn ion is in 3d$^5$ configuration with a 

1037: spin $S=5/2$. This configuration has 5 d-orbitals. For symmetry reasons, 

1038: there may be more than just one conduction or valence band,

1039: which is the case for GaMnAs. Then, as we discussed above in Section V, one has to take into account all pairwise

1040: contributions to the exchange integral, Eqs.(\ref{exchch}),(\ref{exch}). In particular, 

1041: there may be unusual contributions to RKKY,

1042: such as those considered in Section V.

1043: In general, one has also to sum over all orbitals in the Anderson Hamiltonian, not just spins,

1044: and take the Hund's rule, spin orbit, and crystal field splitting into account. Different orbitals 

1045: may have different $V-s$ with conduction and

1046: valence bands. The result of this treatment, however, produces the same exchange integrals $J(R)$; for

1047: example, in case when spin-orbit and crystal field splitting is neglected, the result will still be

1048: given by the exchange integral Eqs.(\ref{exchch}),(\ref{exch}), with 

1049: $|V_1|^2$ is replaced by $\sum_m |V_{1m}|^2$, and similarly for $|V_2|^2$ ,where $V_{1m}$ is

1050: $V_1$ for $m$-th orbital. The relation between corrections to energy levels and exchange integrals

1051: is different in this more realistic case. However, the integrals involved are the same, and, as

1052: we have seen above in section III, corrections to energy levels are rather small. Taking Hund's rule into 

1053: account results in replacing $S=1/2$ operator

1054: in equations of the previous section by $\bm{S}/(2S)$, with $S = 5/2$. Thus, this leads to

1055: the same results as in the previous section, with somewhat redefined $V-s$. The relationship

1056: between the energy shift and $J_{pd}$ will change, but the

1057: actual form of $J(R)$ is determined by the energy spectrum only. Instead of $V_{\alpha m}$, we may introduce 

1058: \begin{equation}

1059: |V_{\alpha}|^2 \equiv \frac{\sum_m |V_{\alpha m}|^2}{4 S^2}\, ,

1060: \end{equation}

1061: and use the form of exchange integrals that we have obtained in the previous sections (with $S=5/2$).

1062: Equivalently, this would mean that the expression for $J(\bm{R})$ in terms of $J_{pd}$-s (or $J_{RKKY}$) for conduction 

1063: and valence bands and level position will stay the same as in the previous sections.

1064:  

1065: For a $p$-type semiconductor, such as Ga$_{1-x}$Mn$_x$As, we can adopt an inverted picture, 

1066: where the ``empty'' band is now the valence band (empty of holes), while the filled band is the conduction band. 

1067: The Anderson Hamiltonian is then easily rewritten in terms of holes. There are two types of holes in GaAs - 

1068: the heavy hole and the light hole. Since their masses are very different ($0.081 m_e$ and $0.51 m_e$),

1069: the main RKKY exchange contribution is produced by the heavy hole. On the other hand, the superexchange contribution

1070: from the light hole band has a much larger range, and thus could potentially be more important than the superexchange contribution

1071: from the heavy hole band, or the mixed superexchange contribution. However, as we have seen in previous sections, most short range

1072: contributions for a direct band gap semiconductor take the form of the second term in  Eq.(\ref{generic}), although

1073: there may be some variations. The problem is that the amplitude of superexchange $V_0 \propto V^4/D^3 \propto m^3 $, where $D$ is

1074: the band width! So, for light bands the effective range of the interaction is large, but the payback is that the amplitude turns 

1075: out to be small.  An easy estimate for the light hole band in GaAs shows that for $V$-s of the order of $1 eV$ this band plays

1076: no role in ferromagnetism. The hole mass in the split-off band

1077: is $0.15 m_e$. This band could also play an important role in the superexchange interaction, although, once again, the

1078: amplitude for realistic parameters turns out to be extremely small. The electron mass is $m_{ge} = 0.063 m_e$

1079: for the main $\Gamma$-valley. The masses and gaps for $L$- and $X$- valleys are much larger, so we don't expect them

1080: to play much role. Thus, we arrive at a simplified picture, where only the

1081: heavy hole band and $\Gamma$-valley electrons are relevant.

1082: The CFR $Mn$ $d^6/d^5$ level,

1083: which is important for our analysis, is in the conduction band, $\Delta_1 =1.5 eV$ above the top of the valence band. 

1084: We can see from Eqs (\ref{rc1rc0}),(\ref{sesese}) that, for this particular level position, the ferromagnetic  Bloembergen-Rowland 

1085: interaction has the range $R_0 \equiv R_{BR} \simeq \hbar/\sqrt{2 m_{hh} \Delta_1}$, while the range of the superexchange is

1086: $R_{se} \simeq R_{BR}/2$. The amplitude of the Bloembergen-Rowland term, $V_{BR}/V_{se} \simeq V_2^2 m_{ge}/(2 V_1^2 m_{hh})$,

1087: could become comparable or exceed the amplitude of heavy hole superexchange.

1088: Note that, in case of strong short-range ferromagnetic

1089: interaction, it would be energetically favorable for $Mn$ impurities to form ferromagnetic clusters.

1090: This, in turn, would reduce the Curie temperature. Clustering of $Mn$ impurities would make magnetic properties

1091: of this material crucially dependent on sample preparation. 

1092: On the other hand, antiferromagnetic short-range interactions should be stronger at shorter distances, 

1093: which would potentially lead to an exchange integral (in the absence

1094: of carriers), which changes its sign as a function of the distance between impurities. In general, for the particular

1095: situation when the $Mn$ level is almost at the bottom of conduction band,  the range of antiferromagnetic superexchange 

1096: is approximately $R_0/2$, and we may represent the total short-range exchange integral in the following form:

1097: \begin{equation}

1098: J(R) \simeq V_0 \frac{R_0}{R}\, [- \alpha \exp{(- 2 R/R_0)} + \exp{(- R/R_0)}].

1099: \label{GaAsex}

1100: \end{equation}

1101: Here $V_0 = V_{BR} > 0$, while $\alpha \simeq m_{hh} V_1^2/(m_{ge} V_2^2)$ is the ratio of superexchange and Bloembergen-Rowland 

1102: amplitudes (up to a factor of 2), which depends on the hybridization of  

1103: impurity d-level with the valence band $(V_2)$, of impurity d-level with the heavy hole band, $V_{hh}$, and the corresponding effective

1104: masses. When $\alpha > 1$ (which is likely the case here, since $m_{hh} \gg m_{ge}$), the exchange becomes 

1105: antiferromagnetic at short distances for $R < R_0 \ln{\alpha}$. The virial correction to $T_c$

1106: for such exchange integral (assuming $T_c$ is determined mostly by the $p-d$ interaction of magnetic impurities and heavy holes)

1107: is then given by: 

1108: \begin{widetext}

1109: \begin{equation}

1110: \frac{\delta T_c}{T_c} \simeq \frac{4 \pi S}{3 (S + 1)}\, n_i R_0^3 (\ln^3{\frac{V_0/T_c}{\ln{(V_0/T_c)}}} - [2 + (1/S)]\ln^3{\alpha}),

1111: \end{equation}

1112: \label{deltatc}

1113: \end{widetext}

1114: and could change sign as well, at some large doping level. If carriers are not present, this exchange integral alone, for $\alpha > 1$,

1115: would lead to a saturation or decrease of $T_c$ at large doping. On the other hand, when $\alpha < 1$, ferromagnetism

1116: gets significantly enhanced at short distances.

1117: 

1118: In general, the interplay between various short range contributions leads to a rather complicated physics

1119: at short distances. While a detailed calculation requires precise knowledge of all hybridization parameters

1120: from the quantum chemistry, we can estimate the $Mn$ concentration at which the short distance

1121: physics becomes important by requiring $\delta T_c / T_c \sim 0.5$ in Eq.(\ref{deltatc})

1122: We take the estimate of $J_{pd} \sim 150 eV \AA^3$ and $T_c \sim 110K$ from Ref.\cite{Ohno},  $\Delta_1 \sim 1.5 eV$, 

1123: and assume that $\alpha \simeq 1$. Then $V_0 = J_{pd}^2 m_{hh}^3/(8 \pi^2)$, and we get:

1124: \begin{equation}

1125: x_i \sim \left(\frac{a}{2 R_0 \ln{(V_0/T_c)}}\, \right)^3 \sim 8\%

1126: \end{equation} 

1127:  

1128: 

1129: \section{Effects of disorder and interactions.}

1130: 

1131: In this section we consider rather briefly effects of disorder and interactions. Since the superexchange and the Bloembergen-Rowland

1132: exchange interaction are governed by high-energy virtual processes, they are independent of disorder. The RKKY interaction,

1133: however, gets modified. This modification was first considered by de Gennes\cite{deGennes}, who argued that, since the

1134: RKKY interaction at large distances is dominated by the $2 p_F$ Kohn anomaly wavevector, the vertex corrections are

1135: not essential for the averages over disorder. 

1136: The long-distance power law in the RKKY interaction then gets multiplied by an exponential factor $\exp{(- 2 R/l)}$,

1137: where $l$ is the scattering length. These effects were indeed taken into account by Ohno et al.\cite{Ohno} in their

1138: original paper. Abrahams \textit{et al.}\cite{elihu}, however, have shown, that this is not the whole story, 

1139: since disorder introduces instead

1140: a \textit{distribution} of $J(R)$ at large distances. We note here that the long-distance behavior

1141: of the RKKY interaction is not

1142: essential for ferromagnetism. The Curie temperature is determined by RKKY exchange at \textit{short} distances,

1143: or $J_{RKKY}(\bm{q}=0)$. Of course, the vertex corrections are essential for the calculation of the RKKY loop diagram at

1144: $\bm{q} = 0$. Summing all ladder diagrams ($0$ order in $1/p_F l$),  shown in Fig. \ref{fig7}, leads to a diffuson contribution, 

1145: \begin{equation}

1146: \Pi(\bm{q}, \omega_n) = - \frac{\nu D \bm{q}^2}{|\omega_n| +  D \bm{q}^2}\, ,

1147: \end{equation}

1148: which significantly modifies frequency dependence of RKKY at $\bm{q} = 0$. Here $D$ is the diffusion coefficient.  

1149: 

1150: \begin{figure}

1151: \includegraphics[width=3in]{fig7.eps}

1152: %\includegraphics[width=3in]{fig7.pdf}

1153: \caption{Impurity corrections to Curie temperature.}

1154: \label{fig7}

1155: \end{figure}

1156: 

1157: However, the static ($\omega = 0$), not dynamic, part of the diagram in Fig. \ref{fig7}  

1158: determines $T_c$, and it is not changed at all. 

1159: Thus, to the leading order in $1/p_F l$ disorder does not modify the Curie temperature. 

1160: The interactions, if not too strong, can also be included as the standard Fermi-liquid corrections to $\chi_0$ in Eq.(\ref{Curie}).

1161: Weak localization corrections (the Cooperon diagrams), however, should modify $T_c$. They can also be included in

1162: the same way as the standard weak localization corrections to spin susceptibility (see, for example, Ref.\cite{lee}), 

1163: and Eq.(\ref{Curie}) should still be valid. 

1164: 

1165: Finally, we note that strong exchange interaction $J^{pd}$ could bind holes at $Mn$ sites, forming a shallow

1166: (or deep) complex magnetic impurity. This effect would

1167: reduce the hole concentration and the number of free $Mn$ spins, and thus lead to a reduction Curie temperature. 

1168: The interactions between

1169: these complex magnetic impurities would be determined by the overlap of the corresponding wavefunctions.  

1170: 

1171: \section{Conclusions.}

1172: 

1173:    We have investigated the model of magnetic semiconductors in which magnetic impurities are treated in the

1174: framework of the Anderson model. We have shown that the effective Hamiltonian of this model is more rich than

1175: the usual p-d model considered in the literature. Effectively, in the Anderson model, wave functions of localized

1176: impurities develop a "tail", which could be long range. When the concentration of magnetic impurities is large

1177: enough, the overlap of wavefunctions on two different sites leads to a very strong exchange interaction, which 

1178: is important, and could dominate the physics at Mn concentrations as low as 5\%.  There are two contributions

1179: to this exchange interaction - superexchange, in which localized electrons are exchanged trough only one type of 

1180: bands (either conduction or valence bands),   and the Bloumbergen-Rowland term, when the exchange is through both

1181: conduction and valence bands.  We have found that, in case of direct band gap, the

1182: Bloemberger-Rowland exchange is ferromagnetic. This could lead to a dramatic enhancement of Curie temperatures in

1183: certain magnetic semiconductors with a direct band gap, such as GaMnAs.

1184:  One other important consequence of the

1185: indirect exchange is that, if it is ferromagnetic and reasonably long range, doped carriers are not necessary

1186: to mediate ferromagnetism. This leads potentially to a new class of high-temperature magnetic semiconductors,

1187: with high Curie temperatures determined entirely by the interaction between localized impurities,  

1188: \textit{not Zener mechanism}.  This emphasizes the effort to search

1189: for new materials, where ferromagnetism is not carrier-driven (for example, driven by the Bloembergen-Rowland mechanism).

1190: Another important consequence of the Anderson model is that, if there are more than

1191: one type of carriers (for example, light and heavy holes in GaAs), the long-range RKKY interaction becomes rather 

1192: complicated, since it involves a ``mixed'' contribution. We have also found that, at large doping,

1193: the RKKY interaction for the Anderson model and the $p-d$ model is  \textit{different} at short distances.

1194: The effective exchange interaction in the $U=\infty$ Anderson model for any FS is given by Eqs.(\ref{exchangeint})

1195: (\ref{exchch}),(\ref{exch}). A numerical solution of this effective Hamiltonian for a given set of parameters (determined

1196: from quantum chemistry) would give the answer for $T_c(n_i)$ in the most general case.

1197: 

1198: We have briefly considered  effects of disorder and interactions. We have shown that when De Gennes\cite{deGennes}

1199: approximation $p_F l \gg 1$ is applicable, disorder does not modify Curie temperature. Localization effects, however,

1200: do modify Curie temperature, although their effect could be reduced to removal carriers. 

1201: Finally, in this paper we have not considered the effects of mixing of conduction and valence bands (such as kp). These effects 

1202: should also be included in the full description.

1203: We should mention that application of the Anderson model to GaAs:Mn was also considered in Ref. \cite{Kikoin},

1204: although the limits of the Anderson model and their conclusions are different from ours.

1205: 

1206: I would like to thank L. P. Gor'kov, Z. Fisk, and G. Kotliar for many useful discussions,

1207: and H. Weitering for sharing his experimental observations. 

1208: I am also very greatful to Misha Zhitomirsky for pointing out Ref.\cite{Kikoin}.

1209:  

1210: This work was supported by the University of Tennessee. I also acknowledge, with  gratitute, the 

1211: input on this work provided by the participants of the Aspen 2003 winter conference on "Complex Quantum Order",

1212: and Aspen Summer 2003 workshop on "Competing Orders and Quantum Criticality in Correlated Electrons, Bosons, and Spin Systems".

1213: 

1214: 

1215: 

1216:  

1217: \begin{thebibliography}{}

1218: \bibitem{Prinz} G.A. Prinz, 

1219: Science, v.\textbf{282}, 1660 (1998).

1220: \bibitem{Wolf} S.A. Wolf, D.D. Awschalom, R.A. Buhrman, J.M. Daughton, S. von Moln\'ar, M.L. Roukes,

1221: A. Y. Chtchelkanova, and D.M. Treger, 

1222: Science, v.\textbf{294}, 1488 (2001). 

1223: \bibitem{Zener1}C. Zener,  

1224: Phys. Rev. \textbf{81}, 440 (1951). 

1225: \bibitem{Zener2}C. Zener, 

1226: Phys. Rev. \textbf{83}, 299 (1951).

1227: \bibitem{AG} A. A. Abrikosov, L. P. Gor'kov, 

1228: Zh. Exp. and Theor. Phys., \textbf{43}, 2230 (1962).

1229: \bibitem{Korenblit} I. Ya. Korenblit and E. F. Shender,

1230: Sov. Phys. Usp., v. \textbf{21}(10), 832 (1978) [Usp. Fiz. Nauk \textbf{126}, 233 (1978).]

1231: \bibitem{Furdyna} J. K. Furdyna and J. Kossut, 

1232: ``Diluted Magnetic Semiconductors'', vol. \textbf{25} of ``Semiconductor and Semimetals'' 

1233: (Academic Press, New York, 1988).

1234: \bibitem{Dietll} T. Dietl, 

1235: ``Diluted Magnetic Semiconductors'', vol. \textbf{3B} of ``Handbook of Semiconductors'' 

1236: (North-Holland, New York, 1994).

1237: \bibitem{dietl1} T. Dietl, H. Ohno, F. Matsukura, J.Cibert, D.Ferrand,

1238: Science, \textbf{287}, 1019

1239: \bibitem{dietl2}

1240: T. Dietl, H. Ohno, and F. Matsukura,

1241: Phys. Rev B \textbf{63}, 195205.

1242: \bibitem{Munekata} H. Munekata, H. Ohno, S. von Molnar, A. Segm\"uller, L. L. Chang, and L. Esaki,

1243: Phys. Rev. Lett. \textbf{63}, 1849 (1989).

1244: \bibitem{Ohno} F. Matsukura, H. Ohno, A. Shen, and Y. Sugawara,

1245: Phys. Rev. B \textbf{57}, R2037 (1998).

1246: \bibitem{deBoeck} J. De Boeck, R. Oesterholt, A. Van Esch, H. Bender, C. Bruynseraede, C. Van Hoof, and G. Borghs,

1247: Appl. Phys. Lett. \textbf{68}, 2744 (1996).

1248: \bibitem{Ohno1} H. Ohno, A. Shen, F. Matsukura, A. Oiwa, A. Endo, S. Katsumoto, and Y. Iye,

1249: Appl. Phys. Lett. \textbf{69}, 363 (1996).

1250: \bibitem{Ohno2}  H. Ohno,

1251: Science \textbf{281}, 951 (1998).

1252: \bibitem{Park} Y. D. Park, A. T. Hanbicki, S. C. Erwin, C. S. Hellberg, J. M. Sullivan, J. E. Mattson, 

1253: T. F. Ambrose, A. Wilson, G. Spanos, B. T. Jonker,

1254: Science \textbf{295}, 651 (2002).

1255: \bibitem{Millis} A. Chattopadhyay, S. Das Sarma, A. J. Millis,

1256: Phys. Rev. Lett. \textbf{87}, 227202 (2001).

1257: \bibitem{Larkin} V. G. Vaks, A. I. Larkin, S. A. Pikin,

1258: ZhETF \textbf{53}, 281 (1967); ZhETF \textbf{53}, 1089 (1967).

1259: \bibitem{Young} D.P. Young, D. Hall, M.E. Torelli, Z. Fisk, J.L. Sarrao, J.D. Thompson, H.R. Ott, S.B. Oseroff, 

1260: R.G. Goodrich, R. Zysler, 

1261: Nature, \textbf{397}(6718), 412 (1999). 

1262: \bibitem{Akimitsu} J. Akimitsu, K. Takenawa, K. Suzuki, H. Harima , Y. Kuramoto,

1263: Science, \textbf{293} (5532), 1125  (2001).

1264: \bibitem{Fisk} Z. Fisk, H. R. Ott, V. Barzykin, and L. P. Gor'kov,

1265: Physica B \textbf{312-313}, 808 (2002).

1266: \bibitem{Sato} K. Matsubayashi, M. Maki, T. Tsuzuki, T. Nishioka, N. K. Sato,

1267: Nature, \textbf{420}, 143 (2002).

1268: \bibitem{SW} J. R. Schrieffer, P. A. Wolff, Phys. Rev. \textbf{149}, 491 (1966).

1269: \bibitem{BG1} V. Barzykin, L. P. Gor'kov,

1270: Phys. Rev. B \textbf{46}, 3059 (1992)

1271: \bibitem{GS1} L. P. Gor'kov, A. V. Sokol,

1272: J. Phys. France \textbf{50}, 2823 (1989);  L. P. Gor'kov, A. V. Sokol,

1273: Pis'ma v ZheTF \textbf{48}, 505 (1988).

1274: \bibitem{BR} N. Bloembergen, T. J. Rowland,

1275: Phys. Rev \textbf{97}, 1679 (1955).

1276: \bibitem{GS2} L. P. Gor'kov, A. V. Sokol, Physica C \textbf{159}, 329 (1989).

1277: \bibitem{Keldysh} L. V. Keldysh,

1278: Zh. Exp.Theor.Phys., \textbf{45}, 364 (1963).

1279: \bibitem{BA} V. Barzykin, I. Affleck,

1280: Phys. Rev. B \textbf{61}, 6170 (2000).

1281: \bibitem{deGennes} P. J. de Gennes, J. Phys. Radium \textbf{23}, 630 (1962).

1282: \bibitem{elihu} A. Jagannathan, E. Abrahams, and M. J. Stephen,

1283: Phys. Rev. B \textbf{37}, 436  (1988).

1284: \bibitem{lee} P. A. Lee, T. V. Ramakrishnan, - Rev. Mod. Phys.\textbf{57}, 287 (1885).

1285: \bibitem{Kikoin}  P. M. Krstaji\'c, V. A. Ivanov, F. M. Peeters,

1286: V. Fleurov, and K. Kikoin, Europhys. Lett., \textbf{61}, 235 (2003).

1287: \end{thebibliography}

1288: \end{document}

1289: