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

2: \usepackage[dvips]{graphicx}

3: \usepackage{latexsym}

4: \usepackage{bm}

5: %\renewcommand{\topmargin}{.0in}

6: \begin{document}

7: \newcommand{\fig}[2]{\includegraphics[width=#1]{#2}}

8: %\newcommand{\fig}[2]{}

9: \newcommand{\pprl}{Phys. Rev. Lett. \ }

10: \newcommand{\pprb}{Phys. Rev. {B}}

11: \newcommand{\be}{\begin{equation}}

12: \newcommand{\ee}{\end{equation}}

13: \newcommand{\bea}{\begin{eqnarray}}

14: \newcommand{\eea}{\end{eqnarray}}

15: \newcommand{\nn}{\nonumber}

16: %

17: \newcommand{\la}{\langle}

18: \newcommand{\ra}{\rangle}

19: \newcommand{\dg}{\dagger}

20: \newcommand{\upa}{\uparrow}

21: \newcommand{\dna}{\downarrow}

22: 

23: 

24: \title{Electron correlation and Fermi surface topology of Na$_x$CoO$_2$}

25: 

26: \author{Sen Zhou$^1$, Meng Gao$^1$, Hong Ding$^1$, Patrick A. Lee$^2$,

27: and Ziqiang Wang$^1$}

28: \affiliation{$^1$ Department of Physics, Boston College, Chestnut Hill,

29: MA 02467}

30: \affiliation{$^2$ Department of Physics, Massachusetts Institute of Technology,

31: Cambridge, MA 02139}

32: 

33: 

34: \date{\today}

35: 

36: \begin{abstract}

37: 

38: The electronic structure of Na$_x$CoO$_2$ revealed by recent photoemission

39: experiments shows important deviations from band theory predictions.

40: The six small Fermi surface pockets predicted by LDA calculations have not

41: been observed as the associated $e_g^\prime$ band fails to cross the

42: Fermi level for a wide range of sodium doping concentration $x$. In addition,

43: significant bandwidth renormalizations of the $t_{2g}$ complex

44: have been observed. We show that these discrepancies are due to strong

45: electronic correlations by studying the multi-orbital Hubbard model

46: in the Hartree-Fock and strong-coupling Gutzwiller approximation.

47: The quasiparticle dispersion and the Fermi surface topology obtained

48: in the presence of strong local Coulomb repulsion are in good agreement

49: with experiments.

50: 

51: \typeout{polish abstract}

52: \end{abstract}

53: 

54: \pacs{71.27.+a, 71.18.+y, 74.25.Jb, 74.70.-b}

55: 

56: \maketitle

57: The cobaltates (Na$_x$CoO$_2$) are doped 3d transition metal oxides in which

58: the Co atoms form a layered hexagonal lattice structure. In contrast to

59: the high-$T_c$ cuprates, where the Cu$^{2+}$ has a 3d$^9$ configuration and

60: occupies the highest {\it single} $e_g$ ($d_{x^2-y^2}$) orbital near the

61: Fermi level, the cobaltates are {\it multi-orbital} systems where

62: the Co$^{4+}$ is in the 3d$^5$ configuration, occupying the three

63: lower $t_{2g}$ orbitals, similar to the ruthenates (Sr$_2$RuO$_4$).

64: The unexpected discovery \cite{takada03} of a superconducting phase of yet

65: unknown origin in hydrated Na$_x$CoO$_2$ around $x\sim0.3$ has generated

66: renewed interests in this material. However, such basic issues as the

67: low energy electronic structure and Fermi surface topology in the

68: cobaltates have not been well understood.

69: Band structure (LDA) calculations \cite{singh00}

70: find that the trigonal symmetry of the Co site in the triangular

71: lattice splits the three $t_{2g}$ complex into an $a_{1g}$

72: and two degenerate $e_g^\prime$ states at the zone center ($\Gamma$ point).

73: The LDA predicts a large Fermi surface (FS) associated with the $a_{1g}$

74: band enclosing the $\Gamma$ point and

75: six small FS pockets of mostly $e_g^\prime$ character near the K points

76: \cite{singh00,pickett}.

77: 

78: However, recent angle-resolved photoelectron spectroscopy (ARPES)

79: measurements on the cobaltates revealed

80: only a {\it single} hole-like FS centered around the $\Gamma$ point for

81: a wide range of Na concentration $x$ \cite{hbyang2,hasan2,hasan,hbyang1}.

82: The area enclosed

83: by the FS exhausts the Luttinger volume, which is consistent with the

84: observation that the dispersion of the $e_g^\prime$ band associated

85: with the LDA FS pockets lies below and

86: never crosses the Fermi level \cite{hbyang2}.

87: The absence of the FS pockets is unexpected and puts serious

88: constraints on several proposed theories of non-phonon mediated

89: superconductivity as well as magnetic properties based on the

90: nesting conditions of the FS pockets \cite{kuroki04,johannes04,yanase}.

91: Furthermore, the measured quasiparticle bandwidths are significantly

92: narrower than the LDA predictions \cite{hbyang2,hasan2}.

93: These fundamental discrepancies between ARPES and LDA suggest that the

94: effects of strong electronic correlations are important in the cobaltates.

95: The effects of local Coulomb repulsion $U$ has been considered

96: in the LSDA+U approach, which indeed finds the absence of the

97: small FS pockets \cite{PHZhang}. However, the latter is tied to

98: the fully polarized ferromagnetic state in the LSDA+U theory which

99: gives spin-split bands and a spin polarized FS with an area twice as large.

100: This is inconsistent with ARPES

101: and likely an artifact of the LSDA+U approximation.

102: A recent calculation based on the multi-orbital Hubbard model

103: and the dynamical mean-field theory finds that the FS pockets

104: become even larger in size than the LDA predications \cite{liebsch}.

105: 

106: The focus of the present work is to explain how strong correlations

107: drive orbital polarization and the band narrowing observed in ARPES.

108: We adopt a multi-orbital

109: Hubbard model description where the noninteracting part is determined by

110: fitting the LDA band structure. The interacting part contains both

111: the intra ($U$) and the inter-orbital ($U^\prime$) local Coulomb repulsion

112: as well as the Hund's rule coupling $J_H$.

113: First, a basis independent Hartree-Fock (HF) calculation is performed which is

114: in essence a LDA+U calculation in the paramagnetic phase. We find that

115: for $U^\prime$ much less than U, multi-orbital occupation is favored in order

116: to reduce the cost of double occupation. As a result, the HF self-energy

117: renormalizes the atomic level spacing in such a way that the size of the

118: FS pockets associated with the $e_g^\prime$ band grows. This trend is however

119: reversed when $U^\prime$ grows and becomes

120: comparable to U. In the HF theory, the size of the $e_g^\prime$ FS

121: pockets begins to shrink for $U^\prime/U > 3/5$. To correctly capture

122: the physics of strong correlation for large $U$ and $U^\prime$, we generalize

123: the Gutzwiller approximation to the case of multi-orbitals.

124: We find that in the strong-coupling regime, orbital polarization is

125: tied to Pauli-blocking, i.e. the orbital occupation dependence of the

126: Gutzwiller band renormalization factors. We obtain both band narrowing

127: and the disappearance of the FS pockets in good agreement with the

128: ARPES experiments.

129: 

130: We start with the multi-orbital tight-binding model on a two-dimensional

131: triangular lattice,

132: %

133: \begin{equation}

134: H_0=-\sum_{ij,\sigma}\sum_{\alpha \beta}t_{ij,\alpha \beta}

135: d^{\dagger}_{i\alpha \sigma}d_{j\beta \sigma}

136: -{\Delta \over 3}\sum_{i,\sigma}\sum_{\alpha \neq \beta}

137: d^{\dagger}_{i\alpha \sigma}d_{i\beta \sigma},

138: %\nonumber \\

139: %&-&\mu_f \sum_{i,\alpha,\sigma}d^{\dagger}_{i\alpha \sigma}d_{j\alpha \sigma},

140: \label{h-tb}

141: \end{equation}

142: %

143: where the operator $d^{\dagger}_{i\alpha\sigma}$ creates an electron in the

144: $\alpha$ orbital with spin $\sigma$ on the Co site and

145: $t_{ij,\alpha\beta}$ is the hopping integral between the $\alpha$

146: orbital on site $i$ and the $\beta$ orbital on site $j$.

147: The relevant valence bands near the FS consist of

148: the Co $t_{2g}=\{d_{xy},d_{yz},d_{zx}\}$ orbitals and have an electron

149: occupancy of $5+x$. The $\Delta$ in Eq.~(\ref{h-tb}) describes

150: the crystal field due to trigonal distortion that splits the $t_{2g}$

151: complex into a lower $a_{1g}$ singlet and a higher $e'_g$ doublet,

152: where $a_{1g}=(d_{xy}+d_{yz}+d_{zx})/\sqrt{3}$,

153: and $e'_g=\{(d_{zx}-d_{yz})/\sqrt{2}$, $(2d_{xy}-d_{yz}-d_{zx})/\sqrt{6}\}$.

154: For convenience, we will work in the hole-picture via a particle-hole

155: transformation $d \rightarrow \tilde d^\dagger$, in which the band

156: filling of holes is $1-x$. The structure of

157: the tight-binding Hamiltonian in k-space on the triangular lattice is

158: \begin{equation}

159: H_0=\sum_{k,\sigma,\alpha\beta}

160: {\bf K}^d_{\alpha\beta}(k)

161: \tilde d^{\dagger}_{k\alpha\sigma}\tilde d_{k\beta\sigma}

162: +{\Delta\over 3}\sum_{k,\sigma,\alpha\neq\beta}

163: \tilde d^{\dagger}_{k\alpha\sigma}\tilde d_{k\beta\sigma}.

164: %\nonumber \\

165: %&+&\mu_f\sum_{k,\alpha,\sigma}d^{\dagger}_{k\alpha\sigma}d_{k\alpha\sigma}.

166: \label{hk-tb}

167: \end{equation}

168: %

169: The hopping matrix $\bf K$ in the $t_{2g}$ basis is given by

170: %

171: \begin{equation}

172: {\bf K}^d(k)=\left ( \begin{array}{clcr}

173: \varepsilon(t,1,2,3) &\varepsilon(t',3,1,2) &\varepsilon(t',2,3,1)\\

174: \varepsilon(t',3,1,2) &\varepsilon(t,2,3,1) &\varepsilon(t',1,2,3)\\

175: \varepsilon(t',2,3,1) &\varepsilon(t',1,2,3) &\varepsilon(t,3,1,2)

176: \end{array} \right),

177: \label{htk}

178: \end{equation}

179: with $(1,2,3)=(k_1,k_2,k_3)$, $k_1=\sqrt{3}k_x/2-k_y/2$,

180: $k_2=k_y$, $k_3=-k_1-k_2$, and

181: $\varepsilon(t,1,2,3)=2t_1\cos{k_1}+2t_2(\cos{k_2}+\cos{k_3})

182: +2t_3\cos{(k_2-k_3)}+2t_4[\cos{(k_3-k_1)}+\cos{(k_1-k_2)}]

183: +2t_5\cos{(2k_1)}+2t_6[\cos{(2k_2)}+\cos{(2k_3)}]+\cdots$.

184: The $(t,t')$ denote the (intra,inter)-orbital hopping.

185: 

186: 

187: Fig.{\ref{fig1}} shows the fitting of the tight-binding dispersions

188: obtained by diagonalizing Eq.~(\ref{hk-tb}) to the LDA band structure

189: at $x$=1/3 \cite{pickett}. We note that the fit with up to

190: third-nearest-neighbor (NN) hopping or more describes the LDA bands

191: quite well. On the other hand, the

192: tight-binding model cannot reproduce completely the LDA dispersions

193: even with up to eighth-NN hopping.

194: The discrepancy is most pronounced along the M-K

195: direction where the two $e'_g$ bands cross in the tight-binding fit

196: (Fig.~1a). Similar disagreement can be traced back to the previous

197: tight-binding fits \cite{yanase,liebsch,johannes}.

198: We believe the difficulty arises from the hopping path via the O 2$s$ and 2$p$

199: orbitals. Nevertheless, the tight-binding model works

200: very well at low energies near the Fermi level.

201: The FS consists of a cylindrical sheet around the

202: $\Gamma$-point and six hole pockets near the K-points as shown in Fig.~1b-d.

203: The central FS has a dominant $a_{1g}$ character while the six FS

204: pockets are mainly of the $e_g^\prime$ character.

205: The hopping integrals obtained from the fit with up to third-NN are

206: $t$=(-44.6, -9.0, 36.2, 5.9, 57.9, 36.7)meV and

207: $t'$=(-157.8, -30.2, 37.1, 9.2, -11.9, -21.0)meV.

208: The crystal-field splitting $\Delta$ is chosen to be 0.01eV.

209: In the rest of the paper, we use these parameters for $H_0$.

210: Our results are insensitive to these values provided that

211: they provide a good fit of the LDA band structure near the FS.

212: Note that although holes are evenly distributed among the three

213: $t_{2g}$ orbitals, in the $a_{1g}$ and $e_{g}^\prime$ basis

214: (hereafter referred to as the $\{a\}$ basis), the hole occupations

215: are 0.123 ($e_g^\prime$) and 0.421 ($a_{1g}$) respectively.

216: Despite its higher orbital energy, the $a_{1g}$ hole

217: orbital has a higher occupation due to its larger bandwidth.

218: %

219: \begin{figure}

220: \begin{center}

221: %\vskip-2.8cm

222: %\hspace*{-3.8cm}

223: \fig{3.3in}{fig1.eps}

224: \vskip-2.4mm

225: \caption{Tight-binding fits to the LDA band structure at $x=1/3$.

226: (a) The fitted band dispersions with up to 3rd, 5th and 8th NN hopping.

227: The corresponding Fermi surfaces are plotted in (b), (c), and (d)

228: respectively.}

229: \label{fig1}

230: \vskip-9mm

231: \end{center}

232: \end{figure}

233: %

234: 

235: The correlation effects are described by

236: the multi-orbital Hubbard model $H=H_0+H_I$, where

237: $H_0$ is the tight-binding Hamiltonian in Eq.(\ref{hk-tb}),

238: $H_I$ represents the local Coulomb repulsion $U$ (intra-orbital) and

239: $U'$ (inter-orbital) and Hund's rule coupling $J_H$.

240: For $t_{2g}$ orbitals, $H_I$ has been shown to take the

241: form \cite{castellani}

242: \begin{eqnarray}

243: H_I&=&U\sum_{i,\alpha}{\hat n}_{i\alpha\uparrow}{\hat n}_{i\alpha\downarrow}

244: +(U'-{1\over 2}J_H)\sum_{i,\alpha>\beta}{\hat n}_{i\alpha}{\hat n}_{i\beta}

245: \label{h-int} \\

246: &-&J_H\sum_{i,\alpha\neq\beta}{\bf S}_{i\alpha}\cdot {\bf S}_{i\beta}

247: +J_H\sum_{i,\alpha\neq\beta}a^{\dagger}_{i\alpha\uparrow}

248: a^{\dagger}_{i\alpha\downarrow}a_{i\beta\downarrow}a_{i\beta\uparrow}.

249: \nonumber

250: \end{eqnarray}

251: with $U^\prime = U - 2J_H$.

252: Here ${\hat n}_{i\alpha}$ and ${\bf S}_{i\alpha}$ are the density and

253: the spin operators in the $\{a\}$ basis where

254: the tight-binding part $H_0$ is

255: %

256: \begin{equation}

257: H_0=\sum_{k,\sigma}\sum_{\alpha\beta}

258: {\bf K}^a_{\alpha\beta}(k)

259: a^{\dagger}_{k\alpha\sigma}a_{k\beta\sigma}

260: +\sum_{k,\alpha,\sigma}\Delta_{\alpha}

261: a^{\dagger}_{k\alpha\sigma}a_{k\alpha\sigma}.

262: \label{hk-tba}

263: \end{equation}

264: %

265: Here $\Delta_{\alpha}$ = $-\Delta/3$, $2\Delta/3$ for the $e'_g$ and $a_{1g}$

266: orbitals respectively. The hopping matrix

267: ${\bf K}^a(k)={\bf O}^{T}{\bf K}^d(k){\bf O}$,

268: with ${\bf O}$ the orthogonal rotation from the $t_{2g}$ to

269: the $\{a\}$ basis. $H_I$ is identical in

270: these two bases. The hierarchy of the interaction

271: strength is $U>U'>J_H\geq 0$.

272: \begin{figure}

273: \begin{center}

274: %\vskip-2.8cm

275: %\hspace*{-3.8cm}

276: \fig{2.8in}{fig2.eps}

277: \vskip-2.4mm

278: \caption{HF results for $U$ = 3.0\textit{eV} at $x=1/3$.

279: The band dispersions are shown for $\eta= 1/3, 1/2, 2/3$.

280: }

281: \label{fig2}

282: \vskip-9mm

283: \end{center}

284: \end{figure}

285: 

286: We first study the effects of interactions in the HF theory in the

287: orbital sector. In the paramagnetic phase, the interacting Hamiltonian

288: is given by,

289: \begin{eqnarray}

290: H_I^{HF} &=& \sum_{k,\sigma,\alpha}

291: ({1\over 2}Un_{\alpha}+U_{\rm eff}^\prime

292: \sum_{\beta\neq\alpha}n_{\beta})

293: a^{\dagger}_{k\alpha\sigma}a_{k\alpha\sigma}\nonumber \\

294: &-&{U\over 4}\sum_{k,\alpha}n^2_{\alpha}

295: -{U_{\rm eff}^\prime\over 2}\sum_{k,\alpha\neq\beta}

296: n_{\alpha}n_{\beta}

297: \label{hartree} \\

298: &+&(U-2U_{\rm eff}^\prime)\sum_{k,\alpha\neq\beta}

299: \bigl[n_{\alpha\beta}

300: a^{\dagger}_{k\alpha\sigma}a_{k\beta\sigma}

301: -{n^2_{\alpha\beta}\over2}\bigr],

302: \nonumber

303: \end{eqnarray}

304: where $n_{\alpha\beta}

305: =(1/N_s)\sum_{k,\sigma}n_{\alpha\beta}^\sigma(k)$,

306: $n_{\alpha\beta}^\sigma(k)=\langle a^{\dagger}_{k\alpha\sigma}a_{k\beta\sigma}

307: \rangle$, $n_{\alpha}=n_{\alpha\alpha}$, and $U_{\rm

308: eff}^\prime=U^\prime-J_H/2$.

309: In essence, this HF analysis, also discussed in Ref.~\cite{liebsch},

310: is equivalent to the LDA+U theory \cite{LDA+U}.

311: Since we are interested in the orbital dependent corrections, we have not

312: displayed in Eq.~(\ref{hartree}) the double-counting term which corrects

313: for the energy already included in the LDA, because it depends only on the

314: total density.

315: 

316: Note that the HF theory is basis independent. It is thus convenient to

317: stay in the $\{a\}$ basis where the local density matrix, and thus the

318: HF self-energies are diagonal in orbital space, i.e.

319: $n_{\alpha\ne\beta}=0$.

320: Then the average energy has contributions only from the first two

321: terms in Eq.~(\ref{hartree}). Recall that most of the holes reside in

322: the $a_{1g}$ orbitals. If $U^\prime_{\rm eff} < U/2$, it is

323: favorable to increase the population of the $e_g^\prime$ orbitals to

324: take advantage of the smaller inter-orbital repulsion $U^\prime_{\rm

325: eff}$. On the other hand for $U^\prime_{\rm eff} > U/2$, the

326: tendency is to empty the $e_g^\prime$ orbitals in favor of $a_{1g}$

327: occupation. The crucial factor of $1/2$ in front of $U$ comes

328: from the fact that due to exchange, intra-orbital repulsion operates

329: only between holes with opposite spin, whereas both spins contribute

330: to $U_{\rm eff}^\prime$.

331: 

332: We proceed to calculate the HF self-energy

333: in terms of the average ${\bar n}=(n_{a_{1g}}+2n_{e_g^\prime})/3=(1-x)/3$

334: and the difference $\delta=(n_{a_{1g}}-n_{e'_g})/3$ between the

335: hole occupation of the $a_{1g}$ and $e_g^\prime$ orbitals,

336: \begin{eqnarray}

337: \Sigma^{\text{HF}}_{e'_g} &=&{1\over2}{\bar n}U(1+4\eta)+\delta U

338: (\eta-1/2)

339: \\

340: \Sigma^{\text{HF}}_{a_{1g}}&=&{1\over2}{\bar n}U(1+4\eta)-2\delta U

341: (\eta-1/2)

342: \end{eqnarray}

343: where $\eta=U_{\rm eff}^\prime/U$ is the relative strength of the

344: inter-orbital interaction.

345: The interaction effect in the paramagnetic HF theory

346: is to simply shift the atomic levels

347: by $\Sigma^{\text{HF}}_{e'_g}$ and $\Sigma^{\text{HF}}_{a_{1g}}$

348: respectively, resulting in a renormalization of the atomic level spacing

349: $\Delta^\prime=-3\delta U(\eta-1/2)$.

350: As expected, the direction

351: of the charge transfer depends on the ratio $\eta$.

352: Since the majority of the holes resides in

353: the $a_{1g}$ orbital in the noninteracting

354: limit, $\delta>0$.

355: Thus, for $\eta<1/2$, the level splitting renormalizes upward,

356: $\Delta^\prime >0$, and interactions induce a transfer of carriers

357: from the $a_{1g}$ to the $e_{g}^\prime$ orbital. The self-consistent

358: HF results are shown in Fig.~2 at $x=1/3$ for $U=3$eV which is close

359: to the value ($3.7$ eV) estimated from the LDA

360: \cite{johannes}. For $\eta=1/3$, the size of the hole pockets

361: indeed becomes larger than that of the noninteracting/LDA ones.

362: At $\eta=1/2$, the

363: self-energy corrections are equal among the orbitals and

364: the noninteracting (LDA) band dispersions are unchanged as shown in Fig.~2.

365: When $\eta>1/2$, i.e. for $U^\prime/U>3/5$ or $J_H/U<1/5$ which is

366: reasonable for the cobaltates \cite{PHZhang}, the level splitting

367: renormalizes downward, $\Delta^\prime <0$,

368: triggering a transfer of holes from the $e'_g$ to the $a_{1g}$ orbital.

369: The six FS pockets continues to shrink as the $e_g^\prime$ band

370: sinks with increasing $\eta$ and disappears beyond a critical ratio $\eta_c$,

371: as shown in Fig.~2 for $\eta=2/3$.

372: We find that $\eta_c(U=3.0{\rm eV})\simeq 5/8$.

373: %

374: \begin{figure}

375: \begin{center}

376: %\vskip-0.8cm

377: %\hspace*{-3.8cm}

378: \fig{2.8in}{fig3.eps}

379: \vskip-2.4mm

380: \caption{The doping $x$ dependence of the level spacing and bandwidth

381: renormalization for the $a_{1g}$ and $e_g^\prime$ orbitals.}

382: \label{fig3}

383: \vskip-9mm

384: \end{center}

385: \end{figure}

386: %

387: 

388: The HF analysis shows that the disappearance of the six

389: FS pockets near the K-points is the physics of large $U$ and $U'$

390: compared with $J_H$.

391: In this case, the HF theory itself becomes unreliable. Moreover, the

392: localization tendency leading to the bandwidth reduction cannot be

393: captured by the HF theory. It is therefore instructive to study the

394: problem in the strong-coupling limit by projecting out the states of

395: double-occupation prohibited by the large on-site Coulomb repulsions.

396: This can be achieved by the Gutzwiller projection

397: $\vert \Psi(\{n_\alpha\})\rangle

398: =P_G \vert \Psi_0(\{n_\alpha\})\rangle$, where $P_G$ is the

399: projection operator that operates on

400: the noninteracting state $\vert\Psi_0(\{n_\alpha\})\rangle$

401: in a given orbital occupation scheme $\{n_\alpha\}$. It removes the

402: double-occupancy by electrons from both the same and different orbitals.

403: This variational procedure is most conveniently implemented in

404: the Gutzwiller approximation where the effect of projection is

405: taken into account by the statistical weighting factor multiplying

406: the quantum coherent state \cite{fczhang}. Specifically,

407: we approximate the hopping term by

408: \begin{equation}

409: \langle\Psi\vert a_{i\alpha\sigma}^\dagger

410: a_{j\beta\sigma}\vert \Psi\rangle

411: =g_t^{\alpha\beta} \langle \Psi_0\vert

412: a_{i\alpha\sigma}^\dagger a_{j\beta\sigma}

413: \vert \Psi_0\rangle,

414: \label{gwt}

415: \end{equation}

416: where the Gutzwiller renormalization factor $g_t$ is given by the

417: ratio of the probabilities in the hopping process in the

418: projected $\vert\Psi\rangle$ and

419: the unprojected $\vert\Psi_0\rangle$. We find,

420: \begin{equation}

421: g_t^{\alpha\beta}={x\over\sqrt{(1-n_{i\alpha\sigma})(1-n_{j\beta\sigma})}}.

422: \label{gt}

423: \end{equation}

424: It is important to note that in a multi-orbital system $g_t^{\alpha\beta}$

425: depends on the occupation of the orbitals connected by the hopping

426: integral as seen in the denominator in Eq.~(\ref{gt}). The latter

427: originates from the Pauli principle.

428: It compensates for the effects of

429: ``Pauli-blocking'' of double-occupation by electrons in the same

430: quantum states which already operate in the free fermion term on the RHS of

431: Eq.~(\ref{gwt}), while the numerator $x$ in Eq.~(\ref{gt}) 

432: describes the ``Coulomb-blocking''

433: due to the large on-site $U$ and $U^\prime$. It turns out that

434: the denominator is crucial for carrier transfer and orbital-polarization in the

435: strong-coupling limit. In the uniform paramagnetic phase, $g_t^{\alpha\beta}

436: =2x/\sqrt{(2-n_\alpha)(2-n_\beta)}$. The orbital occupations are variational

437: parameters determined by minimizing the ground state energy

438: of the Hamiltonian,

439: \begin{eqnarray}

440: H_{GW}&=&\sum_{k,\sigma,\alpha\beta}

441: g_t^{\alpha\beta}

442: {\bf K}^a_{\alpha\beta}(k)

443: a^{\dagger}_{k\alpha\sigma}a_{k\beta\sigma}

444: +\sum_{i,\alpha,\sigma}\Delta_{\alpha}

445: a^{\dagger}_{i\alpha\sigma}a_{i\alpha\sigma}

446: \nonumber \\

447: &+& \sum_{i,\alpha}\varepsilon_{i\alpha}

448: (\sum_\sigma a_{i\alpha\sigma}^\dagger a_{i\alpha\sigma}

449: -n_{i\alpha}),

450: \label{hgw}

451: \end{eqnarray}

452: where $\varepsilon_{i\alpha}$ are the Lagrange multipliers enforcing

453: the occupation $n_{i\alpha}=\sum_\sigma\langle

454: a_{i\sigma\alpha}^\dagger a_{i\sigma\alpha}

455: \rangle$. They are determined by the self-consistency equation

456: \begin{equation}

457: \varepsilon_{i\alpha}=\varepsilon_\alpha

458: ={1\over 2-n_\alpha}{1\over N_s}\sum_{k,\beta,\sigma}

459: g_t^{\alpha\beta} {\bf K}^a_{\alpha\beta}

460: \langle a^{\dagger}_{k\alpha\sigma}a_{k\beta\sigma}\rangle.

461: \label{epsilon}

462: \end{equation}

463: The RHS of this equation is the derivative of the kinetic energy of

464: band $\alpha$ with respect to $n_\alpha$, and can be understood by

465: the following argument. The transfer of a hole from band $\alpha$ to

466: $\beta$ causes the respective bandwidths to decrease and increase by

467: ${\cal O} (1/N_s)$. However, the kinetic energies of the occupied

468: states in each band are changed by order unity, and this energy

469: difference must be reflected in the equilibrium condition by an

470: energy shift $\varepsilon_\alpha$. Thus in contrast to HF theory,

471: there is no energy cost proportional to $U$. Instead,

472: both the band-narrowing and the renormalization of the level spacing 

473: $\Delta^\prime=\varepsilon_{a_{1g}}-\varepsilon_{e_g^\prime}$

474: contribute to the redistribution of holes among the orbitals.

475: In Fig.~3, we show the self-consistently determined band-narrowing factor

476: $z_\alpha=g_t^{\alpha\alpha}$ and the renormalized level spacing.

477: For orbitals with a larger hole occupation, the bandwidth reduction

478: is smaller and the renormalized band energy is lower, resulting

479: in the transfer of more holes into these bands. The combined

480: effects cause the holes to move out of the $e_g^\prime$ band into

481: the $a_{1g}$ band.

482: The calculated band dispersions and the FS topology at

483: $x=0.3, 0.5, 0.7$ are shown in Fig.{\ref{fig4}} in the strong

484: coupling theory. The corresponding non-interacting case is also shown

485: for comparison. The six hole pockets are completely absent at all

486: levels of sodium doping due to strong correlation, leaving a single hexagonal

487: Fermi surface centered around the $\Gamma$ point satisfying the

488: Luttinger theorem. This, as well as the band-narrowing due to strong

489: Coulomb repulsion, is in very good agreement with the photoemission

490: experiments \cite{hbyang2,hasan2}.

491: \begin{figure}

492: \begin{center}

493: %\vskip-0.8cm

494: %\hspace*{-3.8cm}

495: \fig{3.3in}{fig4.eps}

496: \vskip-2.4mm

497: \caption{The band dispersions and the Fermi surfaces in the strong

498: coupling limit (red solid lines) for doping $x$ = 0.3, 0.5 and 0.7.

499: The noninteracting dispersions (blue dashed lines) are also plotted

500: for comparison.}

501: \label{fig4}

502: \vskip-9mm

503: \end{center}

504: \end{figure}

505: 

506: In conclusion, we have shown that strong correlation plays

507: an important role in understanding the electronic structure of Na$_x$CoO$_2$.

508: It pushes the $e_g^\prime$ band below

509: the Fermi level, leading to an orbital polarized state with a single

510: hole-like FS. The absence of the small FS pockets, which

511: would have contributed significantly to the density of states in band

512: structure calculations, further suggests that the large mass enhancement

513: observed in the specific heat measurement \cite{chou}

514: is due to strong correlation.

515: 

516: This work is supported by DOE grants DE-FG02-99ER45747 and DE-FG02-03ER46076,

517: ACS grant 39498-AC5M, and NSF grants DMR-0072205 and DMR-0201069.

518: \begin{thebibliography}{99}

519: \bibitem{takada03}

520: K. Takada, et. al., Nature (London) {\bf422}, 53 (2003).

521: \bibitem{singh00}

522: D. J. Singh, Phys. Rev. B{\bf61}, 13397 (2000).

523: \bibitem{pickett}

524: K.-W. Lee, J. Kune\v{s}, and W. E. Pickett,

525: Phys. Rev. B{\bf70}, 045104 (2004).

526: \bibitem{hbyang2}

527: H.~B. Yang, et. al., cond-mat/0501403 (2005).

528: \bibitem{hasan2}

529: M.~Z. Hasan, et. al., cond-mat/0501530 (2005).

530: \bibitem{hasan}

531: M.~Z. Hasan, et. al., Phys. Rev. Lett. {\bf92}, 246402 (2004).

532: \bibitem{hbyang1}

533: H.~B. Yang, et. al., Phys. Rev. Lett. {\bf92}, 246403 (2004).

534: \bibitem{kuroki04}

535: K. Kuroki, Y. Tanaka, and R. Arita,

536: \pprl{\bf93}, 077001 (2004).

537: \bibitem{johannes04}

538: M.~D. Johannes,

539: I.~I. Mazin, D.~J. Singh, and D.~A. Papaconstantopoulos,

540: Phys. Rev. Lett. {\bf93}, 097005 (2004).

541: \bibitem{yanase}

542: Y. Yanase, M. Mochizuki, and M. Ogata, J. Phys. Soc. Jpn. {\bf74}, 430 (2005);

543: M. Mochizuki, Y. Yanase, and M. Ogata, cond-mat/0407094.

544: \bibitem{PHZhang}

545: P. Zhang \textit{et al}., Phys. Rev. Lett. \textbf{93}, 236402 (2004);

546: P. Zhang \textit{et al}., Phys. Rev. B{\bf70}, 085108 (2004).

547: \bibitem{liebsch}

548: H. Ishida, M. D. Johannes, and A. Liebsch, cond-mat/0412654.

549: \bibitem{johannes}

550: M.~D. Johannes, I. I. Mazin, and D. J. Singh, cond-mat/0408696.

551: \bibitem{castellani}

552: C. Castellani, C.~R. Natoli, and J. Ranninger,

553: Phys. Rev. B{\bf18}, 4945 (1978)

554: \bibitem{LDA+U}

555: A.~I. Liechtenstein, V.~I. Anisimov, and J. Zaanen,

556: Phys. Rev. B{\bf52}, R5467 (1995).

557: \bibitem{fczhang}

558: F.~C. Zhang, \textit{et al}.,

559: %C. Gros, T.~M. Rice, and H. Shiba,

560: Supercond. Sci. Technol. {\bf1}, 36 (1988).

561: \bibitem{chou}

562: F.C. Chou, et. al., Phys. Rev. Lett. \textbf{92}, 157004

563: (2004).

564: \end{thebibliography}

565: \end{document}

566: 

567: