1: \documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2: %\documentclass[twocolumn,showpacs,amsmath,amssymb]{revtex4}
3: %\documentclass[preprint,amsmath,amssymb]{revtex4}
4:
5: \usepackage{graphicx}% Include figure files
6:
7: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
8: %\documentclass[twocolumn,showpacs,amsmath,amssymb]{revtex4}
9: %\documentclass[preprint,amsmath,amssymb]{revtex4}
10:
11: %\usepackage{graphics,dcolumn,bm}
12: \usepackage{natbib}
13: \usepackage{multirow}
14: \usepackage{latexsym,amssymb}
15: \usepackage{amsmath}
16:
17: \newcommand{\Rpp}{\vec{R}_{\perp}}
18: \newcommand{\kpp}{\vec{k}_{\perp}}
19: \newcommand{\kpr}{\vec{k}_{\parallel}}
20: \newcommand{\rpr}{\vec{r}_{\parallel}}
21: \newcommand{\Pel}{\vec{P}_{el}}
22: \newcommand{\ver}{\vec{r}}
23: \newcommand{\veR}{\vec{R}}
24: \newcommand{\vek}{\vec{k}}
25:
26: \begin{document}
27:
28:
29: \title [Polarization structure]{The structure of electronic
30: polarization and its strain dependence}
31: \author{Yanpeng Yao} \author{Huaxiang Fu}
32: \affiliation{Department of
33: Physics, University of Arkansas, Fayetteville, AR 72701, USA}
34: \date{\today}
35:
36: \begin{abstract}
37: The $\phi(\kpp)\sim \kpp$ relation is called polarization structure.
38: By density functional calculations, we study the polarization structure
39: in ferroelectric perovskite PbTiO$_3$, revealing (1) the $\kpp$ point that
40: contributes most to the electronic polarization, (2) the magnitude of
41: bandwidth, and (3) subtle curvature of polarization dispersion. We also
42: investigate how polarization structure in PbTiO$_3$ is modified by
43: compressive inplane strains. The bandwidth of polarization dispersion in
44: PbTiO$_3$ is shown to exhibit an unusual decline, though the total
45: polarization is enhanced. As another outcome of this study, we
46: formulate an analytical scheme for the purpose of identifying what
47: determine the polarization structure at arbitrary $\kpp$ points
48: by means of Wannier functions. We find that $\phi(\kpp)$ is determined
49: by two competing factors: one is the overlaps between neighboring Wannier
50: functions within the plane {\it perpendicular} to the polarization direction,
51: and the other is the localization length {\it parallel} to the polarization direction.
52: Inplane strain increases the former while decreases the latter, causing
53: interesting non-monotonous effects on polarization structure.
54: Finally, polarization dispersion in another paradigm
55: ferroelectric BaTiO$_3$ is discussed and compared with that of PbTiO$_3$.
56:
57: \pacs{77.22.Ej, 77.80.-e}
58: \end{abstract}
59:
60: \maketitle
61:
62:
63: \section{Introduction}
64:
65: Electric polarization is a key quantity for computing and
66: understanding technologically-relevant effective charges, dielectric
67: and piezoelectric responses that are the derivatives of polarization
68: with respect to atomic displacement, electric field, and strain,
69: respectively.\cite{Lines} Polarization also plays an important role
70: in the methodology development of the theory dealing with finite
71: electric fields in infinite solids, by minimization of the free
72: energy $F=U-\vec{E}\cdot{\vec{P}}$~\cite{Souza,Umari,Fu_E,Dieguez}.
73: Total electric polarization consists of electronic contribution
74: ($\Pel$) and ionic component ($\vec{P}_{ion}$). Computing the latter
75: component is straightforward using point charges, while calculating
76: the electronic polarization is not. Today $\Pel$ is calculated using
77: the sophisticated modern theory of
78: polarization\cite{King-Smith,Resta}. According to the theory, $\Pel$
79: corresponds to a geometrical phase of the valence electron states,
80: \begin{equation}
81: \Pel=\frac{2e}{(2\pi)^3}\int d\kpp\phi(\kpp) \ \ ,
82: \label{EPel}
83: \end{equation}
84: where
85: \begin{equation}
86: \phi(\kpp)=i\sum_{n=1}^M\int_0^{G_{\parallel}}d\kpr\langle
87: u_{n\vek}|\frac{\partial}{\partial k_\|}|u_{n\vek}\rangle
88: \label{EPhi_1}
89: \end{equation}
90: is the Berry phase of occupied Bloch wave functions $u_{n{\vek}}$.
91: Subscripts $\parallel$ and $\perp$ mean parallel and perpendicular
92: to the polarization direction, respectively. Practically,
93: to carry out the $\Pel$ calculations,
94: the integral in Eq.(\ref{EPel}) is replaced by a weighted summation of
95: the phases at sampled discrete $\vek$-points
96: (Monkhorst-Pack scheme\cite{Monkhorst}, for example) in the 2D $\kpp$ plane,
97: namely, $\Pel=\sum_{\kpp}\omega (\kpp)\phi(\kpp)$ with weight $\sum_{\kpp}\omega (\kpp)=1$.
98: The polarization at individual $\kpp$, $\phi(\kpp)$,
99: is calculated as the phase of the determinant formed
100: by valence states at two neighboring $\vek$s on the $\kpr$ string
101: as \cite{King-Smith,Resta}
102: \begin{equation}
103: \label{EPhi_2}
104: \phi(\kpp)={\rm Im}\{\ln\prod_{j=0}^{J-1}\,
105: {\rm det}(\langle u_{k_j,m}| u_{k_{j+1},n}\rangle)\} \ \ .
106: \end{equation}
107: Defined as such, the total
108: polarization $\vec{P}=\vec{P}_{ion}+\Pel$ could be uniquely
109: determined and gauge independent up to a modula constant. In
110: Eq.(\ref{EPel}) one sees that, it is the $\phi(\kpp)$ phases
111: at different $\kpp$ that determine the electronic
112: polarization. The purpose of this work is to study the properties
113: of $\phi(\kpp)$.
114:
115:
116: The physical significance of the $\phi(\kpp)$ quantity can be
117: understood by analogy. It is well known that band structure, which
118: describes the relation between single-particle orbital energy and
119: electron wave vector $\vek$, is very useful for understanding
120: electronic, photo-excitation, and photoemission properties in
121: solids\cite{Yu}. The $\phi(\kpp)\sim \kpp$ relation may be similarly
122: termed as ``polarization structure'', or ``polarization-phase
123: structure''. Electron states in band structure can be changed by
124: photo-excitation or emission. The $\kpp$-point polarization phase
125: can be altered by electric fields, which act as a possible
126: excitation source for electrical polarization. Note that electrical
127: fields do not alter the electron wave vector ($\kpp$) perpendicular
128: to the direction of the field, and thus $\kpp$ remains a conserved
129: quantity. The field-induced variation of $\phi(\kpp)$ in fact
130: manifests the $\kpp$-dependent polarization current. As a result,
131: the relevance of polarization structure to electronic polarization
132: is like the band structure to electronic properties.
133:
134: Furthermore, understanding the $\phi(\kpp)$ quantity is of useful
135: value from both fundamental and computational points of view.
136: Fundamentally, this quantity is determined by the Bloch wave
137: functions, not in the ordinary sense of spatial distribution, but
138: through the interesting aspects of the Berry's phase of occupied
139: manifold of electron states. Studying how $\phi(\kpp)$ depends on
140: $\kpp$ may yield better understanding of electron states, as well as
141: the rather intriguing connection between these states and their
142: contributions to polarization in insulator solids. Computationally,
143: we first recognize that the $\phi(\kpp)$ phase computed from
144: Eq.(\ref{EPhi_2}) always produces a value within the principal range
145: $[0,2\pi]$. In reality, depending on the dispersion of $\phi(\kpp)$
146: as a function of $\kpp$, it is possible that the phases for
147: different $\vek$ points fall in different branches. In other words, the
148: true $\phi(\kpp)$ values may fall in the principal range for some
149: $\kpp$ points (Let us denote this set of $\kpp$ points as
150: $\kpp^{(I)}$), while falling out of the principal range for other
151: $\kpp$ (to be denoted as $\kpp^{(II)})$. We find numerically that
152: this indeed happens for real materials particularly when
153: polarization is large; a specific example is given in section II.
154: When this occurs, one must {\it not} artificially shift the phases
155: of the $\kpp^{(II)}$ into the principal range, as computers do
156: according to Eq.(\ref{EPhi_2}). Though this shift makes no
157: difference to the polarization phase of individual $\kpp$ points, it
158: will alter the total $\Pel$ polarization, yielding spurious
159: magnitude of polarization. Only when the phase of {\it every} $\kpp$
160: is shifted by a constant $2\pi$ will the total $\Pel$ polarization
161: remain equivalent. To find out which $\kpp$ may generate a phase not
162: in the principal range, one in principle should compute the whole
163: dispersion structure of polarization and then map out the
164: $\phi(\kpp)$ for all $\kpp$ points based on the assumption that the
165: $\phi(\kpp)$ phase is a continuous function of wave vector $\kpp$,
166: which makes it important to study the properties of the $\phi(\kpp)$
167: phase as a function of $\kpp$.
168:
169:
170: Despite the relevance, the dispersion structure of polarization is
171: nevertheless not completely understood. More specifically, (1) little is
172: known about what determine the $\phi(\kpp)$ phase at individual
173: $\kpp$. In Eq.(\ref{EPhi_2}), $\phi(\kpp)$ is determined by the wave
174: functions of a string of $\kpr$ points, not just a single $\vek$. As
175: a result, the answer to the question is highly non-trivial. (2) For
176: a given ferroelectric substance (say, the prototypical PbTiO$_3$),
177: it is not clear which $\kpp$ exhibits the largest polarization
178: contribution. Does the $\Gamma $ point always contribute most or
179: least? (3) We do not know if the Berry's phases at different $\kpp$s
180: share a similar value or are very different from each other, that
181: is, a problem concerning the dispersion width of the polarization
182: structure. Slightly more intriguing, one may wonder along which
183: direction the $\phi\sim \kpp$ curve shows the largest dispersion?
184: (4) Even for two commonly studied ferroelectrics, BaTiO$_3$ and
185: PbTiO$_3$, we do not know how different or similar their
186: polarization structures are.
187:
188: Recently, there is another active field in the study of
189: polarization, which concerns the use of inplane strain to tune the
190: ferroelectric polarization \cite{Choi,Haeni,Ederer,Lee}. This
191: tunability stems from the fundamental interest in the
192: strain-polarization coupling. Imposed under inplane strain
193: ferroelectrics subject to modifications of chemical bonds and/or
194: charge transfer, thereby the interaction between atoms is altered.
195: It has been known that a compressive inplane strain tends to enhance
196: the total polarization. But the amplitude of enhancement was found
197: to be highly material dependent.\cite{Ederer,Lee} Considering the
198: importance of the strain effects, one might want to know how the
199: $\phi(\kpp)$ phase from each $\kpp$ can be influenced by strain.
200: Strain effects on the polarization dispersion remain largely
201: unknown, however. It would be of interest to examine how the strain
202: may tune and modify the dispersion of polarization structure.
203: Specific questions on this aspect are: in what manner would the
204: inplane strain change the relative contributions and curvatures at
205: different $\kpp$, and how the band width of the dispersion curve is
206: to be altered.
207:
208:
209: With these questions in mind, we here study the dispersion structure
210: of the polarization in ferroelectric perovskites, as well as its
211: dependence on inplane strains. Two complementary approaches
212: (first-principles density functional calculations and analytical
213: formulations) are used. By means of analytical formulation, we aim
214: at a better understanding on what specific quantities and/or
215: interactions determine the polarization at individual $\kpp$ point.
216: Our calculations reveal some useful knowledge on the
217: polarization structure in perovskite ferroelectrics. For example,
218: the largest $\phi(\kpp)$ contribution is shown not to come from the
219: zone center, but from the zone boundary. We also find that the
220: polarization curve in PbTiO$_3$ is notably flat along the
221: $\Gamma-X_1$ direction, and exhibits, however, a strong dispersion
222: along the $\Gamma-X_2$ axis. Our theoretical analysis further
223: reveals that the flat dispersion along the $\Gamma-X_1$ direction is
224: caused by a small amount of participation from the nearest-neighbor
225: interaction between the Wannier functions. Finally, the present
226: study also demonstrates some rather interesting differences in
227: PbTiO$_3$ and BaTiO$_3$, in terms of the polarization structures as
228: well as their strain dependences.
229:
230:
231: \section{The polarization structure of lead titanate}
232:
233: We first present the density functional calculations on the
234: polarization structure in PbTiO$_3$. In its ferroelectric
235: phase PbTiO$_3$ is tetragonal ($|\textbf{a}_1|=|\textbf{a}_2|=a, |\textbf{a}_3|=c$)
236: and possesses a large spontaneous polarization. The
237: polarization is along the $c$-axis direction, perpendicular to the
238: $\kpp$ plane. Calculations are performed within the local density
239: approximation (LDA)~\cite{Hohenberg}. We use pseudopotential method
240: with mixed basis set\cite{Fu_mix}. The Troullier-Martins type of pseudopotentials
241: are employed~\cite{Troullier}. Details for generating pseudopotentials,
242: including atomic configurations, pseudo/all-electron matching radii, and
243: accuracy checking, were described elsewhere\cite{Vpseudo}. The energy
244: cutoff is 100 Ryd, which is sufficient for convergence. The calculations
245: are performed in two steps: the optimized cell structure and atomic positions
246: are first determined by minimizing the total energy and Hellmann-Feynman
247: forces, and after the structural optimization, the polarization dispersion
248: of $\phi(\kpp)$ is calculated using the modern theory of polarization.\cite{King-Smith,Resta}
249: Our LDA-calculated inplane lattice constant for unstrained PT is $a$=3.88{\AA}, with
250: $c/a=1.04$, both agreeing well with other existing calculations.
251:
252: Figure \ref{FPT}(a) shows the reduced 2D Brillouin zone that the $\kpp$
253: points sample over. The calculated $\phi$ phases at individual $\kpp$ points along
254: the $\Gamma \rightarrow X_1 \rightarrow X_2 \rightarrow \Gamma$ path are given in Fig.\ref{FPT}(b).
255: Reciprocal-space coordinates of $X_1$ and $X_2$ are $\kpp=(\pi/a,0)$
256: and $(\pi/a, \pi/a)$, respectively. The dispersion curve is rigidly shifted such
257: that the phase at $\Gamma$ is taken as the zero reference.
258:
259:
260: Before we discuss the specific results in Fig.\ref{FPT}, we need to
261: point out that the shape of this $\kpp$-dependent phase curve is
262: translation invariant. As is known, the electronic polarization
263: alone can be an arbitrary value, if the solid is uniformly translated
264: with respect to a fixed origin of coordinates.
265: Though different translations will
266: change the absolute location of the polarization-dispersion curve,
267: the shape of the curve remains unaffected, however. This can be
268: easily illustrated by analyzing the change in the $\phi(\kpp)$ phase
269: when one displaces the solid arbitrarily. Let the
270: wave function of the original system be
271: $\psi_{n\vec{k}}(\ver)=e^{i\vec{k}\cdot\ver}u_{n\vec{k}}(\ver)$,
272: where $u_{n\vec{k}}(\ver)=u_{n\vec{k}}(\ver+\veR)$. Now, we
273: displace the solid by an arbitrary vector $\vec{r_0}$ while the
274: origin of coordinates is fixed. Let us
275: denote the original system using script A and the displaced system
276: using script B, so $\vec{r}_B=\vec{r}_A+\vec{r}_0$. The wave
277: functions of the displaced system satisfy
278:
279: \begin{equation}
280: \psi_{n\vec{k}}^B(\ver_B)=\psi_{n\vec{k}}^A(\ver_A)
281: =\psi_{n\vec{k}}^A(\ver_B-\ver_0) \ \ .
282: \end{equation}
283:
284: \noindent
285: Thus we have $u_{n\vec{k}}^B(\ver_B)=
286: e^{-i\vec{k}\cdot\ver_0}u_{n\vec{k}}^A(\ver_B-\ver_0)$. Substituting this relation
287: into Eq.(\ref{EPhi_1}) or Eq.(\ref{EPhi_2}), one can obtain that
288: the $\phi(\kpp)$ of the displaced system is
289:
290: \begin{equation}
291: \phi^B(\vec{k}_\bot)=\phi^A(\vec{k}_\bot)+\ver_0\cdot\vec{G}_\| N_{band}^{occ} \ ,
292: \label{EPhi_BA}
293: \end{equation}
294: where $N_{band}^{occ}=M$ is the number of bands occupied by
295: electrons. The phase differences between the A and B systems are thus a
296: constant, independent of $\kpp$.
297:
298:
299: Several observations are ready in Fig.\ref{FPT}(b): (1) The largest
300: $\phi(\kpp)$ polarization does not come from the zone-center
301: $\Gamma$ point. Rather surprisingly, the largest
302: $\phi(\kpp)$ phase is from the $X_2$ point which lies at the far end
303: of the BZ. (2) The polarization curve is flat along the $\Gamma-X_1$
304: line, showing only a small dispersion. On the other hand, the
305: dispersion becomes very large along the $\Gamma-X_2$ direction. (3)
306: At $\kpp$ points of high symmetry (such as $\Gamma$, $X_1$, or
307: $X_2$), the curve in Fig.\ref{FPT}(b) has zero slope, similar to the
308: electron band structure. (4) The dispersion of polarization also
309: shows subtle details which could not be easily understood. For
310: example, there is a local (though not very pronounced) maximum along
311: the $\Delta _1$ line, making the $X_1$ point a local minimum in both
312: $\Gamma-X_1$ and $X_1-X_2$ directions.
313:
314: Our calculations further reveal that, despite the fact that the
315: polarization in Fig.\ref{FPT}(b) exhibits substantial $\kpp$
316: dependency, the dispersion width ($\sim $0.6) is much smaller than
317: $2\pi$. This finding is important for the following reason. As
318: described in the introduction, if the differences of the
319: $\phi(\kpp)$ phases at different $\kpp$ points are greater than
320: $2\pi$, one would encounter a difficulty in determining which branch
321: of phase a specific $\kpp$ point should be assigned. This difficulty
322: can be avoided only after the phases of all $\kpp$ points are mapped
323: out. Fortunately, the result in Fig.\ref{FPT}(b) tells us that the
324: phase contributions from different $\kpp$ points are fairly close,
325: and the differences are far less than the critical value of $2\pi$
326: that may cause the above difficulty. Nevertheless, we should point
327: out that even a small polarization dispersion as in Fig.\ref{FPT}(b)
328: may still give rise to spurious results on total polarization. To
329: illustrate this, we displace all five atoms in PbTiO$_3$ along the
330: polar {\it c}-axis by a distance $z_0$. Fig.\ref{FZo}(a) shows the
331: total (electronic +ionic) polarization, computed from the geometric
332: phase, as a function of the displacement $z_0$ (in unit of $c$).
333: Intuition tells us that the total polarization should be uniquely
334: determined and translationally invariant. However, we see in
335: Fig.\ref{FZo}(a) that unphysical discontinuity happens for some
336: $z_0$ points, and this discontinuity shows up periodically. To
337: understand what causes the discontinuity, we examine the phase
338: contributions from individual $\kpp$ (sampled according to the Monkhorst-Pack
339: scheme\cite{Monkhorst}), depicted in Fig.\ref{FZo}(b). Figure
340: \ref{FZo}(b) shows that the individual-$\kpp$ phases indeed are a
341: periodic function of $z_0$, explaining why the discontinuity in
342: Fig.\ref{FZo}(a) is periodic. Here it may be useful to comment
343: briefly on the length of the periodicity. One might think that by
344: displacing the solid by a distance of $c$ in the
345: $c$-axis direction, the $\phi(\kpp)$ phase would change by a value
346: of $2\pi$. However, the periodicity in Fig.\ref{FZo} is much smaller
347: than $c$. The explanation is simple. As a matter of fact, in real space the individual
348: $\phi(\kpp)$ has a periodicity of $\frac{1}{N_{band}^{occ}}c$
349: (instead of $c$), which for PbTiO$_3$ the periodicity is 0.0455$c$
350: because $N_{band}^{occ}=22$. This is indeed consistent with the
351: numerical calculation in PT (Fig.\ref{FZo}b). The length of
352: periodicity can be seen from Eq.(\ref{EPhi_BA}), showing
353: that, whenever $\vec{r}_0=\frac{n}{N_{band}^{occ}}\vec{R}_\|$ ($n$
354: is an arbitrary integer and $\vec{R}_\|$ is the lattice vector along
355: the $\vec{G}_\|$ direction), the $\phi^B(\kpp)$ and $\phi^A(\kpp)$
356: differ by $\phi^B(\vec{k}_\bot)=\phi^A(\vec{k}_\bot)+2\pi n$.
357: Fig.\ref{FZo}(b) also reveals the reason responsible for the
358: discontinuity of the total polarization. Spurious discontinuity
359: occurs when the $\phi(\kpp)$ phases of some (but not all) individual
360: $\kpp$ exceed $2\pi$ [Fig.\ref{FZo}(b)]. Under this situation,
361: computers incorrectly shift the phases of these $\kpp$ points back
362: to the principle range, yielding spurious total polarization.
363: According to our experience, spurious polarization often takes place
364: in two circumstances: one is for materials of very large
365: polarization, such as tetragonal BiScO$_3$, and another is when
366: atoms in the unit cell are translationally shifted. Given the
367: small bandwidth of the $\phi(\kpp)$ dispersion, it is now straightforward
368: that, by using different $\vec{r}_0$s, we can avoid the spurious
369: polarization. However for some materials, if the dispersion width
370: from different $\kpp$ points is larger than $2\pi$, one may have to
371: rely on the continuity of the $\phi(\kpp)$ phases, and map out the
372: phases of individual $\kpp$ points over the whole two-dimensional
373: $\kpp$ plane in order to find the correct phase branch.
374:
375:
376: \section{Strain dependence of polarization structure}
377:
378: An important property of ferroelectrics is that the polarization is
379: strongly dependent on strain. While strain can change the total
380: polarization, response of the polarization dispersion structure to
381: strain could also be an interesting problem. Here we investigate the
382: response of the polarization structure under inplane strain in
383: PbTiO$_3$. For each in-plane ($a$) lattice constant, the
384: out-of-plane $c$ lattice constant and atomic positions are fully
385: relaxed, by minimizing the DFT total energy. The polarization
386: structure is then determined using the optimal structure.
387:
388: Figure \ref{FPT_str} shows the phase dispersion curves for PbTiO$_3$ at
389: different inplane lattice constants. All curves are shifted so that
390: the phase at $\Gamma$ point is zero, in order to conduct direct
391: comparison. Three conclusions can be drawn from Fig.\ref{FPT_str}:
392: (1) The relative phase, $\phi (\kpp)-\phi(\Gamma )$, changes
393: drastically for $X_2$, but not so significantly for $X_1$. (2) At increasing strain,
394: (or smaller inplane $a$ constant), the bandwidth of the dispersion
395: initially changes very little when $a=3.84${\AA}, and then starts to
396: {\it decrease} upon further increasing strain to $a=3.80${\AA}. The
397: decline of the dispersion bandwidth is rather surprising, since a
398: compressive inplane strain is known to enhance the total
399: polarization in PT. The decline is also counterintuitive when one
400: considers that the decreasing inplane lattice constant makes the
401: atom-atom coupling stronger within the inplane directions, and
402: should therefore have increased the bandwidth. One possible reason
403: that may cause the decrease of the bandwidth is given in the next
404: section. As a result of the
405: declining dispersion, the polarization curve becomes notably
406: ``flat'' at small $a=3.65${\AA}. (3) The curvature of the
407: dispersion also shows subtle changes, featured by the fact that a
408: new dispersion minimum appears along the $X_2-\Gamma$ line at large
409: strain. As a consequence, the dispersion curvature [i.e., the second
410: derivative $\bigtriangledown ^2_{\kpp}P(\kpp)$] at $\Gamma $ point
411: alters its sign from being positive (at large $a$) to negative (at small $a$).
412: Furthermore, the local maximum between $\Gamma-X_1$ for unstrained
413: PT turns into a new minimum at large inplane strains. Meanwhile, the
414: $X_1$ point changes from a minimum into a saddle point, when strain
415: increases.
416:
417:
418: The calculations thus reveal that, while inplane strain has been
419: previously known to introduce interesting modifications (sometimes
420: markedly enlarged \cite{Ederer} and sometimes remarkably small
421: \cite{Lee}) to the {\it total} $c$-axis polarization, its effects on
422: the polarization dispersion at individual $\kpp$ points appear to be
423: even richer, showing that the polarization structure indeed worths
424: studying. The subtle response of the polarization structure, as
425: predicted above, indicate that there is new and rather complex
426: physics behind the results in Fig.\ref{FPT_str}. While we know that
427: the strain-induced changes in the polarization dispersion must be
428: associated with the fundamental modification of electron wave
429: functions, we also have to admit that the DFT results obtained in
430: our numerical calculations are puzzling, and an intuitive
431: understanding of the results is difficult for two reasons. First,
432: this is an early attempt to investigate the polarization structure,
433: and there is not much previous understanding in the literature.
434: Second, although Eq.(\ref{EPhi_1}) and Eq.(\ref{EPhi_2}) allow us to
435: compute precisely the polarization of individual $\kpp$, a direct
436: and more intuitive connection between $\phi(\kpp)$ and Bloch wave
437: functions is hard to capture from these equations. As a result, it
438: would be very helpful if one could find an alternative way to
439: understand the polarization structure and the computation results.
440: For instance, what determines the polarization at individual $\kpp$
441: point, and why $\phi(\kpp)$ maximizes at the $X_2$ point? In the
442: next section, we attempt a scheme which we wish to be able to offer
443: a more intuitive understanding of the polarization structure.
444:
445:
446: \section{Wannier function formulation of polarization structure}
447:
448:
449: As mentioned above, Eq.(\ref{EPhi_1}) and Eq.(\ref{EPhi_2}) give us
450: little intuitive sense on the direct $\kpp$ dependence of the Berry's phase. In
451: order to get more insight, we use Wannier functions to analyze the
452: polarization structure. Previously, Wannier functions have been
453: found very useful in analyzing real-space local
454: polarization\cite{Wu,Maxi}. Here we employ the Wannier-function
455: approach for a different purpose, namely to understand the
456: $\kpp$-dependence of the polarization structure. The Wannier functions
457: are defined as
458: \begin{equation}
459: W_n(\ver-\veR)=\frac{\sqrt{N}\Omega}{(2\pi)^3}
460: \int_{BZ}d\vec{k}e^{i\vec{k}\cdot(\ver-\veR)}u_{nk}(\ver)\ \,
461: \end{equation}
462: or
463: \begin{equation}
464: u_{nk}(\ver)=\frac{1}{\sqrt{N}}\sum_{\veR}
465: e^{-i\vec{k}\cdot(\ver-\veR)}W_n(\ver-\veR) \ \,
466: \label{EWa}
467: \end{equation}
468: where $\veR$ runs over the whole real-space lattice vectors.
469: By substituting Eq.(\ref{EWa}) into Eq.(\ref{EPhi_1}) and carrying
470: out analytically the integral over $\kpr$, it is
471: straightforward to derive, for tetragonal
472: perovskites, the polarization at individual $\kpp$ as
473: \begin{equation}
474: \phi(\kpp)=\frac{2\pi}{c}\sum_{\Rpp}\sum_{n=1}^M \int
475: \rpr W_n^*(\vec{r})W_n(\vec{r}-\Rpp)
476: e^{i\kpp \cdot \Rpp}d\vec{r} \ \,
477: \end{equation}
478: where $\rpr $ is the projection of vector $\ver$ along the
479: polarization direction, $\Rpp$ is the projection of lattice vector
480: $\veR$ onto the plane perpendicular to the polarization direction.
481: For convenience of discussion, we separate the sum over $\Rpp$ into
482: the $\Rpp=0$ term and the rests,
483: \begin{equation}
484: \phi(\kpp)=\phi_{0}+\frac{2\pi}{c}\sum_{\Rpp\neq0}\sum_{n=1}^M \int
485: \rpr W_n^*(\vec{r})W_n(\vec{r}-\Rpp)
486: e^{i \kpp \cdot \Rpp}d\vec{r}\ \,
487: \label{EWPhi}
488: \end{equation}
489: where for $\Rpp=0$,
490: $\phi_{0}=\sum_{n=1}^{M}\int(\vec{r})_{\parallel}
491: W_{n}^{*}(\vec{r})W_{n}(\vec{r})d\vec{r}$ is the phase contribution
492: from the same unit cell. Eq.(\ref{EWPhi}) is the basis for
493: understanding the polarization structure. From this equation, we
494: observe the following.
495:
496: First, it is now clear that the $\kpp$-dependent part of $\phi(\kpp)$ comes
497: only from the $\Rpp \neq 0$ terms, which correspond to the overlap of the
498: Wannier functions in neighboring cells. In other words, the $\kpp$
499: dependence of the $\phi(\kpp)$ phase results from the overlap of
500: the Wannier functions of different cells that are displaced by $\Rpp$
501: from each other within the plane that is perpendicular to the
502: direction of polarization. While the choice of the Wannier function is
503: known to be non-unique due to the gauge uncertainty, the sum of the
504: Wannier-function overlap over occupied bands is a uniquely defined
505: quality which does not depend on the gauge. It is this quantity
506: that determines the shape of the polarization structure.
507:
508: Second, Eq.(\ref{EWPhi}) explains why the bandwidth of polarization
509: dispersion is often much smaller than $2\pi$. Since only the second term
510: in this equation is $\kpp$ dependent, and since the Wannier
511: functions are generally well localized compared to the size of unit
512: cell, one expects the overlap $W^*_{n}(\ver)W_{n}(\vec{r}-\Rpp)$ to
513: be much smaller than unity for $\Rpp\neq0$. This is consistent with
514: our numerical results in Fig.\ref{FPT}, namely,
515: $\phi(\kpp )-\phi_{0} \approx 0.6 \ll 2\pi$.
516:
517: Third, since the dispersion in $\phi(\kpp)$ comes from the overlap
518: of the Wannier functions between cells of different $\Rpp$s in the
519: {\it xy}-inplane directions, it explains why the polarization
520: structure is very sensitive to inplane strain, where by changing
521: inplane lattice constant, the distances between neighboring
522: cells are effectively altered. Meanwhile, we recognize that a
523: precise understanding of how the bandwidth depends on the inplane
524: strain is not as simple as one might think. Naively one tends to
525: think that, with the decline of inplane lattice constant, the
526: dispersion is to increase, since the overlap
527: $W_{n}^{*}(\ver)W_{n}(\vec{r}-\Rpp)$ increases when $\Rpp$
528: decreases. This will lead to the widening of the polarization
529: dispersion width, which is opposite to what we found in
530: Fig.\ref{FPT_str}. This puzzling contradiction can be resolved by
531: noticing that, in addition to being dependent on the overlap
532: strength between $W_{n}(\ver)$ and $W_{n}(\vec{r}-\Rpp)$ within the
533: {\it perpendicular} plane, the dispersion width also hinges on the
534: localization length ($l^{\rm WF}_{\parallel}$) of the Wannier
535: functions along the direction {\it parallel} to the polarization, as
536: a result of the $\rpr $ operator in Eq.(\ref{EWPhi}). With the
537: increasing inplane strain, the $l^{\rm WF}_{\parallel}$ is to
538: shrink. We thus see that the bandwidth of polarization is determined
539: by the balance of two competing factors between the increasing
540: Wannier-function overlap and the decreasing $l^{\rm WF}_{\parallel}$
541: localization length. When the latter dominates, the bandwidth
542: declines as we have seen in Fig.\ref{FPT_str} from numerical
543: calculations.
544:
545:
546: \section{Curve analysis}
547:
548: With the general understanding of the polarization structure in the
549: above section, we next attempt to determine analytically the
550: polarization dispersion specifically for PbTiO$_3$, aimed to obtain
551: further insight into the important details of the polarization
552: structure. As will become clear later, our analysis in the following
553: also explains what determines the $\phi(\kpp)$ polarization at
554: special points of $\Gamma $, $X_1$ and $X_2$. We begin by defining
555: parameters
556: \begin{equation}
557: t(\Rpp)=\frac{2\pi}{c}\sum_{n=1}^M\int
558: \rpr W^*_n(\ver)W_n(\ver-\Rpp)d\ver \ ,
559: \label{EP_t}
560: \end{equation}
561: and then,
562: \begin{equation}
563: \phi(\kpp)=\sum_{\Rpp}t(\Rpp)e^{i\kpp \cdot \Rpp} \ .
564: \label{EP_TB}
565: \end{equation}
566:
567: For dielectrics of insulating nature, Wannier functions are highly
568: localized, and decay exponentially with the distance
569: \cite{Kohn_WF,Marzari}. As a result, $t(\Rpp)$ also decay quickly
570: with the increase of $|\Rpp |$, so we can adopt the tight-binding like
571: approach and consider only several $\Rpp$s that correspond to some
572: nearest neighbors (NN). We consider up to the $2^{nd}$ NNs, where
573:
574: \[\Rpp = \left\{ \begin{array}{ccc}
575: (0 & 0 )& {\rm on\ site} \\
576: (\pm a & 0 )& {\rm 1NNs} \\
577: (0 & \pm a )& {\rm 1NNs} \\
578: (\pm a & \pm a )& {\rm 2NNs} \\
579: \end{array}\right. \]
580: Taking advantage of tetragonal symmetry, we can rewrite
581: Eq.(\ref{EP_TB}) as
582: \begin{equation}
583: \begin{split}
584: \phi(\kpp)=t_0+
585: 2t_1[\cos(k_1a) + \cos(k_2a)] \\
586: + 2t_2[\cos(k_1+k_2)a + \cos(k_1-k_2)a ] \ ,
587: \end{split}
588: \end{equation}
589: where $t_i$ is the $i^{th}$ NNs contribution defined in Eq.(10),
590: and $\kpp=(k_1,k_2)$. This expression gives us a more direct sense
591: of the $\phi(\kpp)\sim \kpp$ polarization dispersion, approximated to
592: the second nearest neighbors. At special $\kpp$ points of $\Gamma $,
593: $X_1$, and $X_2$, the phases are $\phi(\Gamma )=t_0+4t_1+4t_2$,
594: $\phi(X_1)=t_0-4t_2$, and $\phi(X_2)=t_0-4t_1+4t_2$, respectively.
595: We could thus clearly see that the $t_0$ term, corresponding to
596: $\Rpp=0$, acts to rigidly shift the polarization curve as a whole.
597: Meanwhile, the phase relative to the $\Gamma$ (i.e., the dispersion)
598: is determined by the $t_1$ and $t_2$ quantities, and more specifically,
599: \begin{eqnarray}
600: \phi(X_1)-\phi(\Gamma )=&-4t_1 & -8t_2 \ , \nonumber\\
601: \phi(X_2)-\phi(\Gamma )=&-8t_1 & \ .
602: \label{Edif}
603: \end{eqnarray}
604: These equations are useful, since they tell us that (1)
605: the relative height at $X_2$ (which contributes most to the
606: polarization in PT), $\phi(X_2)-\phi(\Gamma)$, is
607: determined by $t_1$, associated with the overlap of the Wannier
608: function in the $1^{st}$ NNs. $t_1<$0 for PbTiO$_3$ in equilibrium.
609: (2) Under the assumption that $t_2$ is negligible, $\phi(X_2)-\phi(\Gamma)$
610: will be larger than $\phi(X_1)-\phi(\Gamma)$ by a factor of 2.
611:
612: Within the second nearest-neighbor approximation, one can further
613: determine analytically the dispersion along the $\Gamma \rightarrow
614: X_1 \rightarrow X_2 \rightarrow \Gamma $ line in the 2D Brillouin
615: zone as
616:
617: \begin{widetext}
618: \[ \phi(\kpp)=\left \{
619: \begin{array}{ll}
620: t_0+2t_1+(2t_1+4t_2)\cos(k_1a), & {\rm for }\ \Gamma \rightarrow X_1 \ {\rm with}\ k_2=0 \\
621: t_0-2t_1+(2t_1-4t_2)\cos(k_2a), & {\rm for }\ X_1 \rightarrow X_2 \ {\rm with }\ k_1=\pi /a \\
622: t_0+2t_2+4t_1\cos(k_1a)+2t_2\cos(2k_1a), & {\rm for }\ X_2
623: \rightarrow \Gamma \ {\rm with}\ k_1=k_2
624: \ .
625: \end{array}
626: \right. \]
627:
628: \end{widetext} The polarization structure could thus be expressed as a simple
629: combination of cosine functions.
630:
631: To examine whether the second-NN approximation is sufficient, we fit
632: the analytical results to the numerical DFT calculations to
633: determine the $t_i$ ($i=0,1,2$) parameters. Note that only
634: $\phi(\kpp)$s at three points (i.e., $\Gamma$, $X_1$ and $X_2$) are
635: fitted. The obtained $t_i$ values are given in Table \ref{Tpara}.
636: These values are then used to determine the whole dispersion curve,
637: shown in Fig.\ref{FPT}(b) for PbTiO$_3$ in equilibrium structure of
638: $a=3.88${\AA}. We could see that the analytical curve agrees well
639: with the DFT result, implying that the 2nd NN approximation works.
640: On the other hand, some fine structure of the curve (such as the
641: small local maximum along the $\Gamma-X_1$) can not be reproduced,
642: where for a better fitting, approximation beyond the 2nd NNs would
643: be necessary.
644:
645: \begin{table}
646: \caption{The fitting $t_1$ and $t_2$ parameters for PbTiO$_3$ at different lattice
647: constants. $t_0$ is not shown here since it does not affect dispersion.}
648: \label{Tpara}
649: \begin{tabular}{ccc}
650: \hline \hline
651: a({\AA}) & $t_1$ & $t_2$ \\
652: \hline
653: 3.88 &-0.072 & 0.031 \\
654: 3.84 &-0.072 & 0.032 \\
655: 3.80 &-0.064 & 0.031 \\
656: 3.72 &-0.031 & 0.023 \\
657: 3.65 &-0.010 & 0.016 \\
658: \hline\hline
659: \end{tabular}
660: \end{table}
661:
662: From Table \ref{Tpara} one can also see how the $t_i$ quantities
663: are influenced by inplane strain. $t_1$ declines substantially as
664: $a$ decreases below 3.80{\AA}, while $t_2$ shows a less dependence
665: on inplane strain. This makes
666: sense since, by varying the inplane strain, the main effect lies
667: in altering the nearest-neighbor interaction among Wannier functions.
668: For $a>3.80${\AA}, $|t_1|$ approximately equals 2$|t_2| $, confirming
669: the importance of the nearest neighbor interaction. For large strains
670: of $a<3.72${\AA}, $|t_1|$ and $|t_2|$ become comparable, for which it
671: is likely that higher orders of NNs are also needed.
672:
673:
674: \section{Comparison with Barium Titanate}
675:
676: It is of interest to compare the polarization dispersions between
677: BaTiO$_3$ (BT) and PbTiO$_3$ (PT), since these two substances have rather different tetragonality,
678: magnitude of polarization, and sizes of A-site atoms. For this
679: purpose, we have studied the polarization structure in BT, for which
680: a tetragonal symmetry is enforced so that a direct comparison with
681: PT can be made. Following the same procedure as for PT, we optimize
682: the cell structure and atomic positions of BT at different inplane
683: lattice constants, and calculate the corresponding polarization
684: structures.
685:
686: Fig.\ref{FBT} displays the polarization structure for BaTiO$_3$ at
687: different inplane lattice constants. Let us first focus on the
688: dispersion of the equilibrium BaTiO$_3$. The LDA-calculated
689: equilibrium inplane lattice constant of BT is $a=3.95${\AA}. Apart
690: from similarities to PT (e.g., $\phi $ maximizes at $X_2$), our
691: calculations reveal some interesting differences between PT and BT
692: under zero strain: (1) The BT dispersion curve has a significantly
693: smaller bandwidth ($\sim $0.42) than that of PT ($\sim $0.57). Since
694: the bandwidth is determined by the difference
695: $\phi(X_2)-\phi(\Gamma)$, i.e., by $t_1$, a smaller bandwidth
696: indicates less overlapping Wannier's functions between nearest
697: neighbors in BaTiO$_3$, which could be explained by the larger
698: inplane lattice constant $a$ for BT at equilibrium. (2) Unlike PT,
699: the polarization in BT is not small at $X_1$. This again can be
700: attributed to the large inplane lattice constant in BT, which leads
701: to a negligible contribution from the 2nd NNs, i.e., $t_2$ is small
702: in BT. Indeed, we numerically found that $t_2$ is -0.007 in BT,
703: compared to 0.031 in PT. By Eq.(\ref{Edif}), $\phi(X_1)$ is about
704: half of the $\phi(X_2)$ value if $t_2$ is small, which is indeed
705: born out in Fig.\ref{FBT}. (3)As a consequence of observation (2),
706: the dispersions of BT and PT along the $\Gamma \rightarrow X_1$ are
707: not quite similar. There is a local maximum between $\Gamma-X_1$ for
708: PT, whereas for BT, no local maximum exists and $X_1$ becomes a
709: saddle point.
710:
711: Upon strain, BaTiO$_3$ and PbTiO$_3$ exhibit sharp difference in
712: their strain dependence of dispersion bandwidth. As we saw
713: previously in Fig.\ref{FPT_str}, inplane strain causes the bandwidth
714: declining for PbTiO$_3$. However, for BaTiO$_3$, a dramatic {\it
715: enlargement} in bandwidth occurs, when $a$ decreases from 3.95{\AA}
716: to 3.85{\AA}. The bandwidth maintains a large value at
717: $a$=3.75{\AA}, after which it starts to drop. In BaTiO$_3$ the
718: polarization dispersion bandwidth thus shows an interesting
719: non-monotonous dependence on inplane strain. This characteristic
720: non-monotonous dependence strongly supports our conjecture that the
721: two competing factors determine the bandwidth, as described above
722: in Section IV.
723: When strain is small in BT, the overlapping of Wannier functions
724: located at the nearest neighboring $\Rpp$s plays a dominant role,
725: and the increasing overlap leads to a larger $|t_1|$ and thus larger
726: bandwidth. As inplane strain becomes large ($a<3.85$\AA), the
727: atom-atom interaction along the $c$-axis is considerably weakened
728: due to elongated $c$-lattice length. As a consequence, the
729: shrinking $l^{\rm WF}_{\parallel}$ localization length of Wannier
730: functions along the $\rpr$ direction takes over and becomes
731: dominant, giving rise to the declining bandwidth. This, once again,
732: reveals that the polarization dispersion contains rich information.
733: To make more quantitative comparison, we replot in Fig.\ref{Fx1x2}
734: the strain dependence of the $\phi(\kpp)$ phases at $X_1$ and $X_2$,
735: relative to the $\Gamma$ point.
736: Fig.\ref{Fx1x2} is of some useful value since it allows us to contrast the
737: $\kpp$-specific polarizations in two materials at {\it the same}
738: fixed inplane lattice constant. The difference between BT and PT is
739: thus not related in a significant sense to atom-atom distance, but
740: largely due to the overlap of respective Wannier functions. In
741: Fig.\ref{Fx1x2}, both $\phi(X_1)$ and $\phi(X_2)$ are seen to be
742: far greater in BaTiO$_3$ than in PbTiO$_3$, for a fixed $a$
743: constant. The greater values of $\phi(\kpp)$ in BT could possibly
744: originate from the fact that the Wannier functions in this material
745: is more spreading due to the larger size of Ba atom.
746:
747: From the comparison between PT and BT, we could see that the
748: polarization structure has some common features for materials with
749: similar structure, and meanwhile, some distinctions revealing the
750: identities of materials. The common features allow us to understand
751: the polarization structure in general, just as for band structure,
752: most III-V semiconductors have direct band gaps. Differences in
753: polarization structure manifest the electron wave functions and
754: interatomic interactions on microscopic scale.
755:
756:
757: \section{Summary}
758:
759: Two different approaches are employed to study the polarization
760: structure in perovskite ferroelectrics. Numerically we use the
761: density functional total-energy calculations and the modern theory
762: of polarization. Analytically we formulate a scheme to describe the $\kpp$
763: dependence of the polarization phase using Wannier functions. By
764: parameterizing the Wannier-function overlapping, we further identify
765: the quantities that determine the $\phi(\kpp)$ phases at special
766: $\kpp$ points of interest. Our specific findings are summarized in
767: the following.
768:
769: For PbTiO$_3$ at equilibrium, (i) the $\phi(\kpp)$ phase maximizes
770: at the Brillouin zone boundary of the 2D $\kpp$ plane, not the zone
771: center. (ii) The polarization structure shows little dispersion
772: along the $\Gamma-X_1$ line. However, the dispersion is large along
773: the $\Gamma-X_2$. (iii) The bandwidth of the dispersion curve is
774: far below 2$\pi$. The small dispersion considerably eases the
775: difficulty in assigning the correct branch of individual $\kpp$
776: phase, but caution still needs to be taken when the $\phi(\kpp)$
777: phase is approaching $2\pi$.
778:
779: Analytically, (iv) the expression, Eq.(\ref{EWPhi}), is given as the basis for understanding
780: the polarization structure. It also explains why the polarization
781: bandwidth is small compared to $2\pi$. (v) The polarization phase at individual
782: $\kpp$ is revealed to depend on the competition of two factors, namely the
783: overlapping strength of Wannier functions within the perpendicular $\veR_{\perp}$
784: plane and the localization length $l^{\rm WF}_{\parallel}$ of these Wannier functions.
785: (vi) Within the 2NN approximation, the $\phi(X_1)$ and $\phi(X_2)$ values in
786: ferroelectric perovskite are found to be
787: $ \phi(X_1)-\phi(\Gamma )=-4t_1 - 8t_2$, $\phi(X_2)-\phi(\Gamma )=-8t_1$.
788: If $t_2$ is negligible, the latter is 2 times of the former.
789: (vii) When PbTiO$_3$ is under compressive inplane strain, the polarization
790: bandwidth is found to decrease, whereas the total polarization increases.
791: The declining bandwidth implies that the localization length $l^{\rm WF}_{\parallel}$
792: of Wannier functions plays a dominating role in PbTiO$_3$.
793:
794: By comparing BaTiO$_3$ with PbTiO$_3$, we show (viii) the equilibrium BT exhibits a
795: smaller bandwidth of 0.42, as compared to the bandwidth of 0.57 in PT.
796: (ix) $\phi(X_1)$ in BaTiO$_3$ is not small, unlike PT. The difference comes from
797: the fact that $t_2$ is negligible in BT, leading to the result that $\phi(X_1)$
798: is about half of the value of $\phi(X_2)$. But in PT, $t_2$ can not be neglected,
799: and acts to offset the $t_1$ contribution, giving rise to smaller $\phi(X_1)$ and
800: flat dispersion along the $\Gamma-X_1$ line. (x) As BaTiO$_3$ is under increasing
801: inplane strains, its polarization bandwidth displays a characteristic non-monotonous
802: variation by first increasing dramatically and then declining. The finding
803: lends a support to the qualitative understanding that two competing factors determine the
804: $\phi(\kpp)$ phase. (xi) When BaTiO$_3$ and PbTiO$_3$ are constrained to the
805: same inplane lattice constant, the $\phi(X_1)$ and $\phi(X_2)$ are shown to be
806: significantly larger in BT than in PT, unlike the case when two materials are in
807: equilibrium.
808:
809: We conclude by pointing out that there are still many aspects of polarization structure
810: we do not yet understand. For example, we have not pursued beyond the 2nd
811: nearest neighbors to explain the local maximum between $\Gamma $
812: and $X_1$ in unstrained PT. We also do not know the physical significance
813: when $\phi(X_1)$ changes from a local minimum to a saddle point as displayed
814: in Fig.\ref{FPT_str} for PbTiO$_3$ under strains. We believe that further analysis of the
815: polarization structure could yield better knowledge on the
816: physics of dielectrics. Like band structure of solids, we hope that the polarization
817: structure can provide us a new tool of studying ferroelectric materials and
818: properties.
819:
820: This work was supported by the Office of Naval Research.
821:
822:
823:
824: \begin{references}
825: \bibitem{Lines} M.E. Lines and A.M. Glass, {\it Principles and
826: Applications of Ferroelectrics and Related Materials} (Clarendon,
827: Oxford, 1979).
828: \bibitem{Souza} I. Souza, J. Iniguez, and D. Vanderbilt, Phys. Rev. Lett.
829: {\bf 89}, 117602 (2002).
830: \bibitem{Umari} P. Umari and A. Pasquarello, Phys. Rev. Lett. {\bf 89},
831: 157602 (2002)
832: \bibitem{Fu_E} H. Fu and L. Bellaiche, Phys. Rev. Lett. {\bf
833: 91}, 057601 (2003).
834: \bibitem{Dieguez} O. Dieguez and D. Vanderbilt, Phys. Rev. Lett. {\bf 96},
835: 056401 (2006).
836: \bibitem{King-Smith} R.D. King-Smith and D. Vanderbilt, Phys. Rev. B {\bf 47}, 1651 (1993).
837: \bibitem{Resta} R. Resta, Rev. Mod. Phys. {\bf 66}, 889 (1994).
838: \bibitem{Monkhorst} H.J. Monkhorst and J.D. Pack, Phys. Rev. B {\bf 13}, 5188 (1976).
839: \bibitem{Yu} P.Y. Yu and M. Cardona, {\it Fundamentals of Semiconductors},
840: (Springer, Berlin, 2001).
841: \bibitem{Choi} K.J. Choi, M. Biegalski, Y.L. Li, A. Sharan, J. Schubert,
842: R. Uecker, P. Reiche, Y.B. Chen, X.Q. Pan, V. Gopalan, L.-Q. Chen,
843: D.G. Schlom, and C.B. Eom, Science {\bf 306}, 1005 (2004).
844: \bibitem{Haeni} J.H. Haeni, P. Irvin, W. Chang, R. Uecker, P. Reiche,
845: Y.L. Li, S. Choudhury, W. Tian, M.E. Hawley, B. Craigo, A.K. Tagantsev,
846: X.Q. Pan, S.K. Streiffer, L.Q. Chen, S.W. Kirchoefer, J. Levy, and
847: D. G. Schlom, Nature (London) {\bf 430}, 758 (2004).
848: \bibitem{Ederer} C. Ederer and N.A. Spaldin , Phys. Rev. Lett. {\bf 95},
849: 257601 (2005).
850: \bibitem{Lee} H.N. Lee, S.M. Nakhmanson, M.F. Chisholm, H. M. Christen,
851: K.M. Rabe, and D. Vanderbilt, Phys. Rev. Lett. {\bf 98}, 217602 (2007).
852: \bibitem{Hohenberg} P. Hohenberg and W. Kohn, Phys. Rev. {\bf 136}, B864 (1964);
853: W. Kohn and L.J. Sham, Phys. Rev. {\bf 140}, A1133 (1965).
854: \bibitem{Fu_mix} H. Fu and O. Gulseren, Phys. Rev. B {\bf 66},
855: 214114 (2002).
856: \bibitem{Troullier} N. Troullier and J.L. Martins, Phys. Rev. B
857: {\bf 43}, 1993 (1991).
858: \bibitem{Vpseudo} Details were given in Ref.\onlinecite{Fu_mix}
859: and in Z. Alahmed and H. Fu, Phys.
860: Rev. B {\bf 76}, 224101 (2007).
861: \bibitem{Wu} X. Wu, O. Dieguez, K.M. Rabe, and D. Vanderbilt, Phys.
862: Rev. Lett. {\bf 97}, 107602 (2006).
863: \bibitem{Maxi} M. Stengel and N.A. Spaldin, Phys. Rev. B {\bf 75}, 205121 (2007).
864: \bibitem{Kohn_WF} J. Des Cloizeaux, Phys. Rev. {\bf 135}, A698 (1964).
865: \bibitem{Marzari} N. Marzari and D. Vanderbilt, Phys. Rev. B {\bf 56}, 12847
866: (1997).
867: \end{references}
868:
869:
870:
871: \begin{figure}
872: \centering
873: \includegraphics[width=10cm]{Fig1.eps}
874: \caption{(a) The 2D Brillouin zone for the $\kpp $ plane; (b) Berry's phase
875: at different $\kpp $ points for PbTiO$_3$ at equilibrium (symbols:
876: direct calculation results; curve: analytical results). The $\phi(\kpp )$ phase
877: is in units of radian.}
878: \label{FPT}
879: \end{figure}
880:
881:
882: \begin{figure}[discontinuity]
883: \centering
884: \includegraphics[width=10cm]{Fig2.eps}
885: \caption{(Color online) (a) Total polarization in strained PbTiO$_3$ of inplane lattice
886: constant $a=3.72${\AA} as a function of the uniform displacement $z_0$ of
887: five atoms; (b) the $\phi(\kpp)$ phases at six Monhorst-Pack sampling
888: $\kpp$ points as a function of $z_0$. For each $c/N^{occ}_{band}$ change
889: in $z_0$, the $\phi(\kpp)$ phases change by $2\pi$. In (b), the $\phi(\kpp)$
890: phase curves are enlarged in the right side of the figure for $z_0$ between
891: 0.044 and 0.048.}
892: \label{FZo}
893: \end{figure}
894:
895:
896: \begin{figure}[PTstrain]
897: \centering
898: \includegraphics[width=12cm]{Fig3}
899: \caption{(Color online) The $\phi $ phases of different $\kpp$-points, for PbTiO$_3$
900: under different inplane lattice constants. Symbols are direct calculation
901: results; curves are guides for eyes. }
902: \label{FPT_str}
903: \end{figure}
904:
905:
906: \begin{figure}[BTstrain]
907: \centering
908: \includegraphics[width=12cm]{Fig4}
909: \caption{(Color online) Polarization dispersions for BaTiO$_3$ at different inplane lattice
910: constants. Symbols are direct calculation results; lines are guide for eyes. }
911: \label{FBT}
912: \end{figure}
913:
914:
915: \begin{figure}[x1x2fit]
916: \centering
917: \includegraphics[width=10cm]{Fig5}
918: \caption{Dependencies of the $\phi(\kpp)$ phases at $X_1$ point (left) and at $X_2$ point
919: (right) as a function of inplane lattice constant, for PT and BT.}
920: \label{Fx1x2}
921: \end{figure}
922: \end{document}
923: