1: \documentstyle[aps,prb,epsf]{revtex}

2: %\documentstyle[twocolumn,aps,prb]{revtex}

3: 

4: %\documentstyle[aps]{revtex}

5: 

6: %\documentstyle[aps,prb,preprint]{revtex}

7: 

8: \def\batio3{BaTiO$_3$}

9: 

10: \begin{document}

11: 

12: %%%%%%  PP Comment out next two lines for preprint

13: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname

14: @twocolumnfalse\endcsname

15: 

16: \title{Devonshire-Landau free energy of BaTiO$_3$ from

17: first principles}

18: 

19: \author{Jorge \'I\~niguez$^{1}$\cite{byline}, S. Ivantchev$^2$,

20: J. M. Perez-Mato$^2$, and Alberto Garc\'{\i}a$^1$}

21: 

22: \address{$^1$ Departamento de F\'{\i}sica Aplicada II, Universidad del

23: Pa\'{\i}s Vasco, Apdo 644, 48080 Bilbao, Spain\\ $^2$ Departamento de

24: F\'{\i}sica de la Materia Condensada, Universidad del Pa\'{\i}s Vasco,

25: Apdo 644, 48080 Bilbao, Spain}

26: 

27: \maketitle

28: 

29: \begin{abstract}

30: We have studied the Devonshire-Landau potential underlying the phase

31: transition sequence of BaTiO$_3$ using the first-principles effective

32: Hamiltonian of Zhong, Vanderbilt, and Rabe {[}{\it Phys. Rev. Lett.}

33: {\bf 73}, 1861 (1994){]}, which has been very successful in

34: reproducing the phase transitions and the dielectric and piezoelectric

35: properties of this compound. The configuration space (determined by

36: the polarization {\bf P} as order parameter) was explored with the

37: help of auxiliary electric fields. We show that the typically assumed

38: form of the potential, a sixth-order expansion in {\bf P} around the

39: paraelectric cubic phase, properly accounts for the behavior of the

40: system, but we find a non-trivial temperature dependence for all the

41: coefficients in the expansion, including the quadratic one, which is

42: shown to behave non-linearly. Our results also prove that the

43: sixth-order terms in the free-energy expansion (needed to account for

44: the first-order character of the transitions and the occurrence of an

45: orthorhombic phase) emerge from an interaction model that only

46: includes terms up to fourth order.

47: \end{abstract}

48: 

49: %%PP%% Comment out these two lines for preprint...

50: \vskip1pc

51: ]

52: 

53: \narrowtext \marginparwidth 2.7in \marginparsep 0.5in

54: 

55: \section{INTRODUCTION}

56: 

57: %%% Brief summary of Landau theory

58: 

59: From a phenomenological point of view, the behavior of a system in the

60: vicinity of a phase transition can be described in the framework of

61: Landau theory.~\cite{landau} In this scheme, one begins by identifying

62: the so-called order parameter, a (in general multidimensional)

63: variable ${\bf Q}$ that characterizes the symmetry change in the

64: transition, and then constructs the Landau free-energy function

65: $F({\bf Q};T)$, with the property that the equilibrium value of ${\bf

66: Q}$ as a function of temperature is that which minimizes

67: $F$. Formally, the Landau free energy is an incomplete thermodynamic

68: potential, which in principle could be calculated as:

69: %

70: \begin{equation}

71: F({\bf Q};T) = -k_BT\ln \sum_{j/{\bf Q}}{e^{-U_j/k_BT}} \; ,

72: \label{eq:incomplete}

73: \end{equation}

74: %

75: in analogy with the procedure to obtain the standard free energy in

76: terms of the partition function. Here the sum is only over those

77: states with the given value of ${\bf Q}$. In a Landau phenomenological

78: treatment, one just assumes that $F({\bf Q};T)$ exists, and that it

79: can be represented as a simple power series in ${\bf Q}$ in the

80: vicinity of the transition:

81: %

82: % Adapt Eq. 136.7 in Statistical Physics

83: %

84: \begin{equation}

85: F({\bf Q};T) = F_0(T) + A(T)Q^2 + 

86:                         B\sum_{\beta}{f_{\beta}^{(4)}}({\bf Q})+ 

87:                         \dots \; ,

88: \label{eq:landau}

89: \end{equation}

90: %

91: where $f_{\beta}^{(j)}$ are invariants of order $j$ constructed from

92: ${\bf Q}$. The quadratic coefficient $A$ is assumed to vary linearly

93: with temperature in the form $A(T)=\alpha(T-T_0)$, in such a way that

94: $T_0$ is the transition temperature (for a second-order transition) or

95: the lower metastability limit of the upper phase (for a first-order

96: one). These assumptions are known to break down in the critical region

97: close to the transition point of continuous phase transformations, but

98: they are valid when describing the approximate behavior of the system

99: in wider temperature intervals around the transition

100: temperature. Classical Landau theory ignores any temperature variation

101: of the higher-order coefficients in the expansion. Nevertheless, it

102: has been applied quite successfully to the analysis of many materials

103: in wide temperature ranges.~\cite{radescu} Nowadays Landau theory is

104: the most common approach to study the phenomenology of any

105: symmetry-breaking structural phase transition.

106: %

107: 

108: One of the early successes of this phenomenological approach to phase

109: transitions was the description by Devonshire~\cite{devonshire} of the

110: sequence of transitions in barium titanate (\batio3), a sequence that

111: spans a temperature interval of more than 200 degrees. This material

112: exhibits at high temperatures a paraelectric cubic perovskite

113: structure and, as temperature decreases, it undergoes three successive

114: first-order transitions to ferroelectric phases with tetragonal,

115: orthorhombic, and rhombohedral symmetries. In these phases the

116: polarization ${\bf P}$ points along one of the $\langle 1,0,0\rangle$,

117: $\langle 1,1,0\rangle$, and $\langle 1,1,1\rangle$ cubic directions

118: respectively.~\cite{jona} The polarization ${\bf P}$ can be identified

119: as the (three-dimensional) order parameter, and the Devonshire-Landau

120: potential (per unit volume) written as:

121: %

122: \begin{equation}

123: \label{eq-cowley}

124: \begin{array}{lll}

125: F & = & F_0+\frac{1}{2}aP^2 + u P ^4 + v (P_x^4+P_y^4+P_z^4)\\ 

126:       & & + h_1 P^6 + h_2 (P_x^6+P_y^6+P_z^6)\\ 

127:       & & + h_3 [P_x^4(P_y ^2+P_z^2)+P_y^4(P_z^2+P_x^2)\\

128:       & & + P_z^4(P_x^2+P_y^2)] \; ,

129: \end{array}

130: \end{equation}

131: %

132: where $F_0$ is the free energy of the reference cubic phase and $P^2$ stands

133: for $P_x^2+P_y^2+P_z^2$ (the notation is taken from the review paper by

134: Cowley~\cite{cowley}).  This is the complete expansion of $F$ up to

135: sixth order, and its relatively simple form is due to the high symmetry of the

136: system. Sixth-order terms in ${\bf P}$ in the expansion are needed to account

137: for the first-order character of the transitions (to model non-equivalent

138: coexisting free-energy minima).  Moreover, it can be seen that one needs

139: anisotropic sixth-order terms in $F$ to account for the orthorhombic phase of

140: \batio3.~\cite{vander-cohen} Devonshire showed that by assuming a linear

141: temperature dependence for $a$ and suitable {\sl constant} values for the rest

142: of the coefficients it is possible to qualitatively reproduce the transition

143: sequence as well as the dielectric properties of the system.~\cite{devon-more}

144: 

145: %% Experimental facts against Devonshire's theory

146: 

147: After Devonshire's work, the use of an expansion like that of

148: Eq.~\ref{eq-cowley} has become the standard way to model the

149: thermodynamic properties of ferroelectric perovskites, even though the

150: correctness of the assumption of constant high-order coefficients was

151: questioned early on. Drougard {\it et al.\/} and Huibregtse {\it et

152: al.\/}~\cite{exp} studied BaTiO$_3$ at temperatures just above the

153: cubic to tetragonal transition and in the orthorhombic phase

154: respectively; they assumed the existence of a Devonshire-Landau

155: potential but showed that there should be a significant temperature

156: dependence of the fourth-order terms.~\cite{gonzalo} It has since been

157: shown that there is a relatively large latitude to produce

158: Devonshire-like models that lead to qualitatively sensible predictions

159: for the transition sequence, divergence of the dielectric constants,

160: etc., even though these models may not be {\sl realistic} when confronted

161: with other experimental evidence.~\cite{other-forms} On the other

162: hand, it could be questioned whether such a relatively simple

163: free-energy expansion as that in Eq.~\ref{eq-cowley} exists at

164: all. For instance, Nakamura and Kinase have worked on a different

165: approach to the problem in which a non-polynomical from of the free

166: energy is used; they also discuss the connection of their work to the

167: Devonshire theory.~\cite{nakamura}

168: 

169: In the past decade, first-principles methods based on

170: density-functional theory have demonstrated to be accurate enough to

171: reproduce and predict structural phase transitions in perovskite

172: oxides.~\cite{vanderbilt} In particular, the phase transition sequence

173: and other properties of BaTiO$_3$,~\cite{zhong,garcia}

174: PbTiO$_3$,~\cite{waghmare-rabe} and KNbO$_3$~\cite{krakauer} have been

175: studied using the so called {\sl effective Hamiltonian} approach. An

176: ab-initio effective Hamiltonian $H_{\rm eff}$ is a mechanical model

177: that includes the relevant microscopic degrees of freedom of the

178: system and is constructed on the basis of first-principles

179: calculations. This model can then be analyzed by statistical methods

180: (typically, Monte Carlo or molecular dynamics) to explore the

181: finite-temperature behavior of the system. Monte-Carlo simulations

182: with an ab-initio effective Hamiltonian for \batio3\ have been

183: extremely successful, replicating approximately the experimental

184: transition sequence~\cite{zhong} and succeeding in reproducing the

185: main features of the dielectric and piezoelectric properties of the

186: real system.~\cite{garcia}

187: 

188: A natural question to ask then is to what extent the thermal behavior

189: resulting from these ab-initio effective Hamiltonians is compatible

190: with the Devonshire-Landau framework. In this paper we show that a

191: Devonshire-Landau expansion of the form of Eq.~\ref{eq-cowley} does

192: indeed emerge from detailed Monte Carlo simulations with the ab-initio

193: effective Hamiltonian for \batio3 of Ref.~\onlinecite{zhong}, with the

194: important qualification that the coefficients in the expansion have a

195: non-trivial temperature dependence. In particular, we show that the

196: sixth-order terms in $F({\bf P};T)$ are non-zero only in the

197: temperature range in which the transitions occur. Moreover, it becomes

198: clear that these sixth-order coefficients appear as

199: products of the statistical fluctuations of a Hamiltonian that only

200: includes terms up to fourth order in the polar degrees of freedom.

201: %

202: 

203: \section{METHOD AND TECHNICAL DETAILS}

204: 

205: %%% The effective Hamiltonian for barium titanate

206: 

207: Zhong, Vanderbilt, and Rabe~\cite{zhong} constructed their effective

208: Hamiltonian for BaTiO$_3$ retaining as relevant degrees of freedom of the

209: system a local polar distortion ${\bf u}_{i}$ in each cell, the homogeneous

210: strain $\eta$, and an inhomogeneous strain represented by a second set of

211: local vectors ${\bf v}_{i}$. The $H_{\rm eff}$ has the form:

212: %

213: \begin{equation}

214: \label{eq-zhong}

215: \begin{array}{lll}

216: H_{\rm eff} & = & \sum_{i,j;\alpha,\beta} J(i,j;\alpha,\beta)

217: u_{i\alpha} u_{j\beta} \\ & & + \sum_{i;\alpha,\beta}

218: \Gamma(\alpha,\beta) u_{i\alpha}^2 u_{i\beta}^2 \\ & & +

219: \sum_{i;\alpha,\beta;l} B(i;\alpha,\beta;l) u_{i\alpha} u_{i\beta}

220: \eta_{l} \\ & & + \sum_{l,h} B(l,h) \eta_{l} \eta_{h},

221: \end{array}

222: \end{equation}

223: %

224: where, for clarity, we have not written the terms associated to the

225: inhomogeneous strain. Here $i$ and $j$ range over the cells in the

226: system, $\alpha$ and $\beta$ are cartesian indexes, and $l$ and $h$

227: refer to tensor components in Voigt notation.  For the local polar

228: modes, the $H_{\rm eff}$ contains harmonic couplings

229: $J(i,j;\alpha,\beta)$ (on-site, $i=j$, and between modes in different

230: cells up to the third nearest neighbors) that reproduce the

231: instabilities of the cubic phase of \batio3,~\cite{uns-band} and

232: fourth-order on-site terms $\Gamma(\alpha,\beta)$ that define the

233: low-symmetry minima and stabilize the system. The effect of strain is

234: included through the standard elastic energy and through the coupling

235: coefficients $B(i;\alpha,\beta;l)$, which account for the spontaneous

236: strain in the low-symmetry phases. This model is probably the simplest

237: one that captures the essential physics of the system. As mentioned in

238: the introduction, $H_{\rm eff}$ is fourth-order in the polar

239: variables.

240: 

241: %%% Simple connections between the effective Hamiltonian and

242: %%% Devonshire's theory.

243: 

244: A direct link between this $H_{\rm eff}$ and the phenomenological

245: potential of Eq.~\ref{eq-cowley} can be established in the very

246: low-temperature limit $T\rightarrow$ 0. Eq.~\ref{eq:incomplete} shows

247: that $F({\bf P};0)$ is just the energy of the lowest-lying state with

248: polarization ${\bf P}$. For \batio3 this state exhibits homogeneously

249: polarized cells with the global strain adjusted so as to minimize the

250: energy (the inhomogeneous strain is zero). It can be

251: shown~\cite{king-smith} that for a given ${\bf P}$, the homogenous

252: strain is proportional to the square of the polarization, so $F({\bf

253: P};0)$ is fourth-order in the polarization. This means that

254: $h_1=h_2=h_3=0$ in the low-temperature limit. Using the $H_{\rm eff}$

255: parameters, we obtain the rest of coefficients in $F$ at 0~K:

256: $a=-4/3$, $u=0.094$, and $v=0.051$. Here, units have been chosen so

257: that the first-octant rhombohedral ground state minimum is located at

258: ${\bf P}=(1,1,1)$ and its energy per unit volume with respect to the

259: cubic phase is $-1$. (This election fixes the value of $a$ and the

260: relation $9u+3v=1$; i.e., $F({\bf P};0)$ is determined only by one

261: coefficient.) We have plotted this 0~K free-energy map in

262: Fig.~\ref{fig-maps}h), which shows the cubic phase as a local

263: free-energy maximum, tetragonal and orthorhombic saddle points (both

264: stable in the radial direction), and the rhombohedral global minima.

265: %

266: \begin{figure}

267: \epsfxsize=\hsize\epsfbox{fig1_a.eps}

268: \epsfxsize=\hsize\epsfbox{fig1_b.eps}

269: \epsfxsize=\hsize\epsfbox{fig1_c.eps}

270: \epsfxsize=\hsize\epsfbox{fig1_d.eps}

271: \epsfxsize=\hsize\epsfbox{fig1_e.eps}

272: \epsfxsize=\hsize\epsfbox{fig1_f.eps}

273: \epsfxsize=\hsize\epsfbox{fig1_g.eps}

274: \epsfxsize=\hsize\epsfbox{fig1_h.eps}

275: \caption{Free-energy maps of BaTiO$_3$ at several temperatures. Panel

276: a): 300~K, b): 290~K, c): 250~K, d): 220~K, e): 210~K, f): 190~K, g):

277: 100~K, and h): 0~K. For each temperature we show the (010) and

278: (1$\bar{1}$0) planes of the order-parameter configuration space, with

279: contour lines depicted only in the low free-energy regions. Symbols

280: characterize the relative stability of the critical points with

281: respect to displacements within a given plane: circles for free-energy

282: minima (a bigger circle corresponds to the stable-equilibrium state),

283: squares for free-energy maxima, and triangles for saddle points

284: (``up'' (resp. ``down'') triangles for points that are maxima

285: (resp. minima) along the radial direction). Note that critical points

286: with orthorhombic symmetry may need to be represented by two different

287: symbols in the $(010)$ and $(1\bar{1}0)$ planes (see panel b)).}

288: \label{fig-maps}

289: \end{figure}

290: 

291: Now we tackle the calculation of $F$ at finite temperatures. A

292: straightforward approach to obtain information about the Landau

293: potential from the results of Monte Carlo~\cite{mc} simulations was

294: presented by Radescu {\sl et al.}~\cite{radescu} in the context of a

295: three dimensional $\Phi^4$ model.  From Eq.~\ref{eq:incomplete} it is

296: easy to show that there exists a relation between the probability

297: distribution of the order parameter ${\cal P}({\bf Q};T)$ and the

298: corresponding Landau potential $F({\bf Q};T)$:

299: %

300: \begin{equation}

301: \label{eq-radescu}

302: F({\bf Q};T) - F_{eq}(T) = -k_B T ln [{\cal P}({\bf Q};T)],

303: \end{equation}

304: %

305: where $F_{eq}(T)$ is the equilibrium free energy of the system at the

306: temperature considered. This procedure, which can be quite efficiently

307: applied to relatively simple cases such as the $\Phi^4$ model

308: (one-dimensional order parameter, a single second-order phase

309: transition), is not well suited for our problem. We have to study a

310: three-dimensional free-energy map with coexisting non-equivalent

311: minima. A typical Monte Carlo simulation gives information only about

312: one free-energy minimum (usually, the one corresponding to the

313: stable-equilibrium state of the system), with a poor sampling of the

314: order-parameter probability-distribution function in other regions. On

315: the other hand, more sophisticated sampling strategies would be very

316: demanding from the computational point of view.

317: 

318: This problem can be overcome, and $F({\bf P};T)$ computed in an

319: efficient manner, by modifying the effective Hamiltonian so as to

320: include the effect of an external electric field ${\bf

321: E}$:~\cite{garcia}

322: %

323: \begin{equation}

324: H'_{\rm eff} = H_{\rm eff} - {\bf E} \cdot Z^* \sum_i {\bf u}_i\; ,

325: \label{eq:hprime}

326: \end{equation}

327: %

328: where $Z^*$ is the Born effective charge associated to the local polar

329: distortions. As the quantity multiplying the electric field is just

330: the total dipole moment of the system, the new Landau potential is

331: simply $F'({\bf P};T,{\bf E}) = F({\bf P};T) - {\bf EP}$. An applied

332: electric field changes the location (and possibly the symmetry) of the

333: stable-equilibrium state. The new location can be determined by a MC

334: simulation of the modified effective Hamiltonian (${\bf

335: P}_{eq}=\langle {\bf P}\rangle$) and also computed

336: directly from $F'$. The idea can be mathematically expressed as:

337: %

338: \begin{equation}

339: \label{eq-def}

340: \left[\frac{\partial F'({\bf P};T,{\bf E})}{\partial {\bf P}}\right]_{{\bf

341: P}_{eq}(T,{\bf E})} = {\bf 0}\; .

342: \end{equation}

343: %

344: The left hand side of this equation depends linearly on the

345: coefficients of $F$ and on the electric field. Thus, for a given

346: temperature we can consider several electric fields, perform MC runs

347: to obtain the equilibrium values ${\bf P}_{eq}$,~\cite{tech1} and then

348: find the best solution of an overdetermined set of linear equations of

349: the form of Eq.~\ref{eq-def} to get the coefficients of the expansion

350: of $F$.

351: %

352: 

353: For each temperature, we first performed a zero-field calculation to

354: obtain a snapshot of the stable-equilibrium configuration. We then

355: used this configuration as the starting point for the runs with

356: electric field applied. In order to illustrate the kind of information

357: we were pursuing, consider $T$ such that the stable-equilibrium state

358: of the system is orthorhombic with ${\bf P}_{eq} = P_{eq} (1,1,0)$. In

359: this case, we would first use fields of the form $(E,E,0)$ to obtain

360: information about the response of the system within the orthorhombic

361: phase. Fields of the forms $(E,0,0)$, $(0,0,E)$, and $(E,E,E)$ would

362: probe the relative stability of minima with different

363: symmetries. Finally, general fields leading to $P_{x,eq} \ne P_{y,eq}

364: \ne P_{z,eq}\ne P_{x,eq}$, would explore in detail the intermediate

365: regions of the configuration space.  In our Monte Carlo simulations we

366: used a $14\times 14\times 14$ supercell (which corresponds to 13720

367: atoms) with periodic boundary conditions; we typically performed 15000

368: MC sweeps for the thermalization of the system and 35000 more to

369: calculate the averages of the polarization (these are very

370: well-converged calculation conditions, as can be checked in

371: Refs.~\onlinecite{zhong,garcia,waghmare-rabe}). For each temperature

372: we considered around one hundred different electric fields and

373: constructed a largely overdetermined system of equations for the

374: coefficients in $F$, which could then be reliably fitted.

375: 

376: \section{RESULTS AND DISCUSSION}

377: 

378: The data from our Monte Carlo simulations can indeed be represented by

379: a Landau free energy in the form of Eq.~\ref{eq-cowley} in a wide

380: temperature interval that includes all the transitions. (The

381: transition temperatures predicted by the effective Hamiltonian are

382: $T_{c1}=297$~K (cubic to tetragonal), $T_{c2}=230$~K (tetragonal to

383: orthorhombic) and $T_{c3}=200$~K (orthorhombic to rhombohedral)

384: respectively.~\cite{temps}) Our results show that the coefficients in

385: the expansion of $F$ have a significant and non-trivial temperature

386: dependence.

387: 

388: Before discussing in detail the behavior of the coefficients, it is

389: helpful to present the shape of the free energy they determine at a

390: few temperatures in the range of the phase

391: transitions. Fig.~\ref{fig-maps}a) (300~K) shows how tetragonal local

392: minima appear just above the first transition, while the absolute

393: minimum of $F$ is is still located at {\bf P}=0 in the cubic well.

394: %

395: Panels b) and c) correspond to 290~K and 250~K respectively, i.e.,

396: temperatures in the range of stability of the tetragonal phase.  We

397: see that the tetragonal wells are indeed the global free-energy

398: minima, and that the orthorhombic and rhombohedral wells nucleate in

399: the form of saddle points that are unstable with respect to a

400: tetragonal distortion (in the rhombohedral case, the saddle points are

401: unstable with respect to an orthorhombic distortion also).

402: %

403: Panels d) and e) refer to 220~K and 210~K respectively, i.e.,

404: temperatures in the orthorhombic range. The tetragonal wells become

405: metastable with respect to the orthorhombic ones, which are now the

406: free-energy global minima. At the same time, rhombohedral local minima

407: appear.

408: %

409: Finally, panels f) and g) correspond to 190~K and 100~K respectively,

410: both in the rhombohedral temperature range. Here we see that the

411: rhombohedral minima have finally become the deepest ones.

412: %

413: It is clear that the sequence converges to the free-energy map given

414: by the effective Hamiltonian itself (panel h)). For all the

415: transitions, the coexistence of different free-energy minima in the

416: figures evidences their first-order character.

417: 

418: Our method to explore the $F$ landscape has a limitation when probing

419: the vicinity of the cubic phase. After the point {\bf P}=0 becomes a

420: local free-energy maximum a few degrees below $T_{c1}$, it is no

421: longer possible to sample the region around it using auxiliary

422: electric fields: the cubic structure is already unstable and thus no

423: longer useful as a starting point for the simulations, and attempts to

424: steer a system in a tetragonal, orthorhombic, or rhombohedral well

425: towards {\bf P}=0 only succeed in landing it in the corresponding

426: symmetry-inverted domain. As the behavior of $F$ around {\bf P}=0 is

427: basically determined by the quadratic coefficient $a$ of the

428: Devonshire-Landau expansion, the value assigned to this coefficient by

429: a fit of the simulation data should be considered suspect. (This

430: problem pertains only to the region near {\bf P}=0. Information about

431: the relative stability of tetragonal, orthorhombic, and rhombohedral

432: phases is available at any temperature, since it is always possible to

433: apply fields that lead the system to each of these symmetries.)

434: %

435: \begin{figure}[b!]

436: \epsfxsize=\hsize\epsfbox{fig2.eps}

437: \caption{Temperature behavior of the quadratic coefficient ($a$) of

438: the Devonshire-Landau potential: The dashed line connects the values

439: obtained from a direct fit to MC data, the solid line shows the

440: interpolated $a(T)$, and the solid circles are the points used to fit

441: the interpolating polynomial (see text). The transition temperatures

442: are marked with vertical lines.}

443: \label{fig-quadra}

444: \end{figure}

445: %

446: As shown in Fig.~\ref{fig-quadra}, at temperatures around and above

447: $T_{c1}$ the behavior of $a$ is fairly reasonable. It is positive at

448: high temperatures, becoming negative a few degrees below $T_{c1}$, as

449: corresponds to a first-order transition.  For very low temperatures

450: $a$ also behaves well, tending smoothly to its 0~K value. This is

451: because in this region the underlying Landau potential is so simple

452: (the thermal fluctuations are relatively small and the sixth-order

453: terms almost negligible) that we can use the information obtained

454: around the low-symmetry minima to reconstruct the complete free-energy

455: map. However, in the intermediate region (from 150~K to 250~K), $a$

456: turns positive again, which would mean that the cubic phase becomes

457: metastable in a wide temperature interval. This metastability is an

458: unphysical result (explicitly ruled out within our model by zero-field

459: MC simulations starting from a cubic phase, in which we found no trace

460: of it). We thus reconsidered the fitting of our data, imposing a more

461: physical temperature evolution of $a$: we assumed that in the

462: intermediate-temperature range $a(T)$ is given by a simple

463: interpolation (Fig.~\ref{fig-quadra}) between the high-temperature

464: region, where we can reliably sample it, and the low-temperature limit

465: determined directly by the convergence to the effective Hamiltonian (a

466: third-order polynomial suffices for our purposes). With this from for

467: $a(T)$ fixed, our MC data are still well fitted to the Landau

468: potential of Eq.~\ref{eq-cowley}. For instance, at 190~K the relative

469: error obtained when $a(T)$ is freely fitted is $2.2\%$, and when

470: $a(T)$ is fixed by the interpolation the error is still very small:

471: $3.6\%$. This clearly shows that, in the intermediate-temperature

472: region, our Monte Carlo data contain little information about the

473: value of $a$.~\cite{umbrella} Higher-order terms were tried but found

474: to play no role in the fit, so they can be excluded from the

475: potential. Our final results are shown in Fig.~\ref{fig-coeffs}.

476: %

477: \begin{figure}[b!]

478: \epsfxsize=\hsize\epsfbox{fig3.eps}

479: \caption{Free-energy coefficients fitted to MC data assuming $a(T)$ is

480: given by the solid line in Fig.~\protect\ref{fig-quadra}. The

481: transition temperatures are marked with vertical lines.}

482: \label{fig-coeffs}

483: \end{figure}

484: %

485: At very low temperatures $a$, $u$, and $v$ tend to their 0~K values

486: and the sixth-order terms go to zero. On the other hand, at

487: temperatures well above the first transition ($T>350$~K) we find the

488: MC data are properly fitted to a fourth-order Landau potential, so the

489: sixth-order terms can again be excluded from the model. The sixth-order

490: terms then occur only in the temperature range in which the

491: transitions take place (as plainly shown in Fig.~\ref{fig-coeffs}). It

492: is also remarkable that the fourth-order terms exhibit a very strong

493: temperature dependence. Particularly, the anisotropic term $v$ changes

494: sign at around 90~K and takes large negative values all through the

495: intermediate-temperature range. So, apart from the influence it has,

496: together with $h_2$ and $h_3$, in determining the transition sequence,

497: we find that a negative $v$ is responsible for the first-order character of

498: the cubic to tetragonal transition.

499: 

500: In Fig.~\ref{fig-thexp} we have plotted our coefficients near the

501: cubic to tetragonal transition temperature together with the available

502: experimental results.~\cite{exp,gonzalo} (We have changed the

503: temperature scale to make the experimental $T_{c1}$ coincide with the

504: theoretical one.) The agreement is only qualitative but as good as

505: could be expected given the simplifying assumptions involved in the

506: experimental work of Ref.~\onlinecite{exp} (the coefficients are

507: restricted to be constant or to depend linearly on temperature) and

508: the fact that the differences among the two experiments are comparable

509: to those between experiment and theory.

510: 

511: Our work strongly suggests that the phase transitions of BaTiO$_3$ can

512: be described in terms of a single Landau potential $F$ with the form

513: of Eq.~\ref{eq-cowley}. Despite what is assumed in most of the

514: previous work on this problem, the quadratic parameter $a$ is found to

515: exhibit a strongly non-linear temperature dependence. This is clear

516: because we can reliably calculate $a$ at high (over $T_{c1}$) and very

517: low temperatures, and the two regions cannot be joined linearly.  We

518: show that all the high-order terms in $F$ present a very significant

519: evolution with temperature. This conclusion is essentially opposed to

520: the temperature-independent behavior that is still assumed by some

521: authors (see, for example, Ref.~\onlinecite{other-forms}). It is also

522: very remarkable that the two features of BaTiO$_3$ that require the

523: inclusion of sixth-order terms in the Landau potential, i.e., the

524: first-order character of the transitions and the occurrence of an

525: orthorhombic phase, have been reproduced using a fourth-order

526: effective Hamiltonian.~\cite{PZT} This piece of information should be

527: taken into account when constructing mechanical models to study the

528: finite-temperature behavior of similar compounds. For instance, the

529: authors of Ref.~\onlinecite{nakamura} included sixth-order terms in

530: the underlying interaction model for \batio3\ with the explicit aim to

531: account for the first-order character of the transitions; our results

532: indicate that those terms should be unnecessary.

533: %

534: \begin{figure}[b!]

535: \epsfxsize=\hsize\epsfbox{fig4.eps}

536: \caption{Comparison of our free-energy coefficients (lines) with the

537: experimental values of Refs.~\protect\onlinecite{exp} (open symbols)

538: and~\protect\onlinecite{gonzalo} (solid symbols): solid line and

539: circles for $a$, dotted line and squares for $u+v$, dashed line and

540: ``up'' triangles for $h_1+h_2$, and dash-dotted line and ``down''

541: triangles for $h_3$.  The vertical line marks the cubic-tetragonal

542: transition temperature.}

543: \label{fig-thexp}

544: \end{figure}

545: %

546: 

547: \section{SUMMARY}

548: 

549: We have studied the Devonshire-Landau potential underlying the phase

550: transition sequence of BaTiO$_3$ using the first-principles effective

551: Hamiltonian parametrized by Zhong, Vanderbilt, and Rabe. The

552: order-parameter configuration space was explored with the help of

553: auxiliary electric fields that change the location and relative

554: stability of the free-energy minima.  Our results show that the

555: typically assumed form of the potential, an expansion up to sixth

556: order in the polarization from the paraelectric cubic phase, properly

557: accounts for the behavior of the system. But, despite what is usually

558: presumed, we find a non-trivial temperature dependence for all the

559: coefficients in the expansion, including the quadratic term $a$, which

560: is shown to behave non-linearly. Our work also shows that the

561: sixth-order terms in polarization needed to explain basic features of

562: BaTiO$_3$ in a Devonshire-Landau approach (the first-order character

563: of the transitions and the occurrence of an orthorhombic phase) are

564: properly accounted for by an interaction model that only includes

565: terms up to fourth order.

566: 

567: \section{ACKNOWLEDGEMENTS}

568: 

569: This work has been supported by the Spanish CICyT grant

570: No. PB98-0244. J.~I. and S.~I. acknowledge financial support from the

571: Basque regional government and the UPV, respectively. We thank

572: Javier Junquera and Jon Saenz for their help.

573: 

574: 

575: \begin{thebibliography}{99}

576:  

577: \bibitem[*]{byline} Email address: wdbingoj@lg.ehu.es

578: 

579: \bibitem{landau} L. D. Landau and E. M. Lifshitz, {\it Statistical

580: Physics} (Pergamon Press Ltd., London, 1958).

581: 

582: \bibitem{radescu} S. Radescu, I. Etxebarria, and J. M. Perez-Mato,

583: {\it J. Phys.: Cond. Matt.} {\bf 7}, 585 (1995).

584:   

585: \bibitem{devonshire} A. F. Devonshire, {\it Phyl. Mag.} {\bf 40}, 1040

586: (1949); {\it ibid.} {\bf 42}, 1065 (1951); {\it Advances in Physics}

587: {\bf 3}, 85 (1954).

588: 

589: \bibitem{jona} See, for example: F. Jona and G. Shirane, {\it

590: Ferroelectric Crystals} (Pergamon Press, Oxford, England 1962;

591: reedition by Dover Publications, Inc., New York 1993) or M. E. Lines

592: and A. M. Glass, {\it Principles and Applications of Ferroelectric

593: Materials} (Clarendon Press., Oxford 1977).

594: 

595: \bibitem{cowley} R. A. Cowley, {\it Advances in Physics} {\bf 29}, 1

596: (1980).

597: 

598: \bibitem{vander-cohen} For a discussion of the conditions that a

599: Devonshire-Landau potential must fulfill to account for phases of a

600: given symmetry, see D. Vanderbilt and M. H. Cohen, cond-mat/0009337.

601: 

602: \bibitem{devon-more} In the second article in

603: Ref.~\onlinecite{devonshire}, Devonshire extended the theory to

604: consider the strain and gave a proper account for the elastic and

605: piezoelectric properties of the system.

606: 

607: \bibitem{exp} M. E. Drougard, R. Landauer, and D. R. Young, {\it

608: Phys. Rev.} {\bf 98}, 1010 (1955); E. J. Huibregtse and D. R. Young,

609: {\it ibib.} {\bf 98}, 1562 (1956).

610: 

611: \bibitem{gonzalo} A temperature dependence of the sixth-order terms in

612: $F$ was reported by J. A. Gonzalo and J. M. Rivera in Ferroelectrics

613: {\bf 2}, 31 (1971).

614: 

615: \bibitem{other-forms} See the discussion in Ref.~\onlinecite{cowley}

616: and that by K. Fujita and Y. Ishibashi, {\it Jpn. J. Appl. Phys.} {\bf

617: 36}, 254 (1997).

618:   

619: \bibitem{nakamura} K. Nakamura and W. Kinase, {\it J. Phys. Soc. Jap.}

620: {\bf 61}, 4596 (1992); {\it ibid.} {\bf 61}, 2114 (1992).

621: 

622: \bibitem{vanderbilt} See the short review by D. Vanderbilt in {\it

623: Current Opinions in Solid State and Materials Science} {\bf 2}, 701

624: (1997) and references therein.

625: 

626: \bibitem{zhong} W. Zhong, D. Vanderbilt, and K. M. Rabe, {\it

627: Phys. Rev. Lett.} {\bf 73}, 1861 (1994); {\it Phys. Rev.} {\bf B52},

628: 6301 (1995).

629: 

630: \bibitem{garcia} A. Garc\'{\i}a and D. Vanderbilt, {\it

631: Appl. Phys. Lett.} {\bf 72}, 2981 (1998); {\it Proceeding of the First

632: Principles Calculations for Ferroelectrics: Fifth Williamsburg

633: Workshop}, edited by R. E. Cohen (AIP, Woodbury, New York, 1998),

634: p. 53.

635: 

636: \bibitem{waghmare-rabe} U. V. Waghmare and K. M. Rabe, {\it

637: Phys. Rev.} {\bf B52}, 13236 (1995).

638: 

639: \bibitem{krakauer} H. Krakauer, R. Yu, C. Z. Wang, and C. Lasota, in

640: {\it Proceedings of the 1997 Williamsburg Workshop on Ferroelectrics},

641: edited by H. Chen, {\it Ferroelectrics} {\bf 206}, 133 (1998).

642:   

643: \bibitem{uns-band} The dispersion curves associated to the harmonic

644: effective system are essentially those of the unstable band calculated

645: by Ghosez {\sl et al.} [Ph. Ghosez, E. Cockayne, U. V. Waghmare, and

646: K. M. Rabe, {\it Phys. Rev.} {\bf B60}, 836 (1999)].

647: 

648: \bibitem{king-smith} R. D. King-Smith and D. Vanderbilt, {\it

649: Phys. Rev.} {\bf B49} 5828 (1994).

650: 

651: \bibitem{mc} Although our emphasis is on Monte Carlo simulations,

652: the discussion applies also to molecular dynamics methods.

653: 

654: \bibitem{tech1} From a MC simulation we typically obtain $\langle{\bf

655: P}\rangle$ corresponding to the stable-equilibrium state of the

656: system.  Eq.~\protect\ref{eq-def} can be used also if the system gets

657: stuck in a metastable-equilibrium state, but not if the simulation

658: alternates from one free-energy minimum to another, giving a spurious

659: $\langle{\bf P}\rangle$.

660:   

661: \bibitem{temps} The experimental transition temperatures for BaTiO$_3$ are

662:   $T_{c1} = 403$~K, $T_{c2} = 278$~K, and $T_{c3} = 183$~K.

663:   

664: \bibitem{umbrella} We tried to compute $a$ in the problematic

665: temperature interval by modifying the effective Hamiltonian to keep

666: the point ${\bf P}=0$ as a local minimum (what could be understood

667: as a kind of umbrella sampling, see M. P. Allen and D. J. Tildesley,

668: {\it Computer Simulation of Liquids} (Oxford, New York,

669: 1990)). However, this modification results in inhomogeneous

670: equilibrium states, which lie outside the Devonshire-Landau framework.

671: 

672: \bibitem{PZT} For Pb(Zr$_x$Ti$_{1-x}$)O$_3$, a similar fourth-order

673: effective Hamiltonian reproduces a monoclinic phase, and thus would

674: formally lead to eigth-order terms in the Devonshire-Landau

675: potential. See Ref.~\onlinecite{vander-cohen} and references therein.

676: 

677: \end{thebibliography}

678: 

679: \end{document}

680: