1: %\documentclass[aps,pra,preprint,groupedaddress,showpacs]{revtex4}
2: \documentclass[aps,pra,twocolumn,groupaddress,showpacs,floatfix]{revtex4}
3: \usepackage{graphicx}
4: %\usepackage{graphics}
5: %({[
6:
7: \begin{document}
8:
9: \title{Phase diagram of harmonically confined one-dimensional fermions
10: with attractive and repulsive interactions}
11:
12: \author{V. L. Campo Jr.}
13: \author{K. Capelle}
14: \email{capelle@if.sc.usp.br}
15: \affiliation{Departamento de F\'{\i}sica e Inform\'atica,
16: Instituto de F\'{\i}sica de S\~ao Carlos,
17: Universidade de S\~ao Paulo,
18: Caixa Postal 369, 13560-970 S\~ao Carlos, SP, Brazil}
19: \date{\today}
20:
21: \begin{abstract}
22: We construct the complete $U$-$\mu$ phase diagram for harmonically confined
23: ultracold fermionic atoms with repulsive and attractive interactions.
24: ($\mu$ is the chemical potential and $U$ the interaction strength.)
25: Our approach is based on density-functional theory, and employs analytical
26: expressions for the kinetic and correlation energy functionals.
27: For repulsive interactions our calculations confirm previous
28: numerical studies, and complement these by closed expressions for
29: all phase boundaries and characteristic lines of the phase diagram, and by
30: providing an explanation and solution for difficulties encountered in earlier
31: density-functional work on the same system. For attractive interactions we
32: propose a new and accurate interpolation for the correlation energy, and
33: use it to extend the phase diagram to $U<0$.
34: \end{abstract}
35:
36: \pacs{03.75.Ss, 71.15.Mb, 71.10.Pm, 71.30.+h}
37: % 03.75.Ss Degenerate Fermi gases
38: % 71.15.Mb Density functional theory
39: % 71.10.Pm Fermions in reduced dimensions
40: % 71.30.+h Metal-insulator transitions and other electronic transitions
41:
42: \maketitle
43:
44: \newcommand{\be}{\begin{equation}}
45: \newcommand{\ee}{\end{equation}}
46: \newcommand{\bea}{\begin{eqnarray}}
47: \newcommand{\eea}{\end{eqnarray}}
48: \newcommand{\bi}{\bibitem}
49:
50: \renewcommand{\r}{({\bf r})}
51: \newcommand{\rp}{({\bf r'})}
52:
53: \newcommand{\ua}{\uparrow}
54: \newcommand{\da}{\downarrow}
55: \newcommand{\la}{\langle}
56: \newcommand{\ra}{\rangle}
57: \newcommand{\dg}{\dagger}
58:
59: The field of strongly correlated low-dimensional fermions has a long history
60: and multifarious applications in condensed-matter physics.
61: More recent experimental progress in the field of confined ultracold atoms
62: \cite{atoms1,atoms2,atoms3,atoms4} has further widened the scope of this
63: already very rich field by introducing, e.g., an unprecedented degree of
64: control over the confining potential, and the possibility to continuously
65: tune the particle-particle interaction from repulsive to attractive --- which
66: is not easy for cold atoms, but unconceivable for electrons. After much
67: progress on bosonic atoms and their condensation, more recently fermionic
68: atoms have come under intense study.
69:
70: In the present paper we focus on one particular model for optically confined
71: fermionic atoms, the one-dimensional Hubbard model. The parameters
72: characterizing this model and spanning its phase diagram are the on-site
73: interaction $U$ \cite{footnote0} and the filling factor $n=N/L$, where $N$
74: is the number of particles and $L$ the number of lattice sites. For systems
75: in which $n$ is spatially varying, such as in the presence of a confining
76: potential, it is more useful to specify the chemical
77: potential $\mu$ instead, which is a constant throughout the system. Hence,
78: we are interested in the phase diagram of the one-dimensional Hubbard model
79: in $U$-$\mu$ space. Our aim is to map out this phase diagram, identify the
80: physics governing each possible phase, and to provide analytical expressions
81: for all phase boundaries.
82:
83: The $U>0$, i.e., repulsive, part of this phase diagram has recently been
84: studied also by Liu et al. \cite{drummond}, who used a half-numerical
85: half-analytical local-density approximation (LDA).
86: Rigol et al. \cite{rigol} obtained essentially the same
87: $U>0$ phase diagram from fully numerical Quantum Monte Carlo simulations.
88: An interesting feature that emerged in these works is the appearance of mixed
89: Mott-insulator Luttinger-liquid phases, characterized by spatially
90: separated incompressible and compressible regions in the optical trap,
91: and plateaus in the density profiles coinciding with the incompressible regions.
92: Xianlong et al. \cite{polini} have recently studied the case of negative
93: $U$, i.e., attractive interactions, where the physics is dominated by
94: Luther-Emery-type pairing correlations and the spontaneous formation of
95: atomic density waves. However, the Kohn-Sham-type density-functional
96: calculations of Ref. \cite{polini} do not allow one to map out the entire
97: phase diagram, because the self-consistency cycle does not converge in the
98: mixed metal-insulator phases at $U>0$.
99:
100: In the present paper we use an analytical local-density approximation both
101: for the kinetic energy (in the spirit of Thomas-Fermi theory) and for the
102: correlation energy (in the spirit of the Bethe-Ansatz LDA proposed in
103: Ref.~\cite{balda}) to obtain the $U$-$\mu$ phase diagram. For repulsive
104: interactions our results coincide with those obtained numerically
105: in Refs.~\cite{drummond,rigol}. However, as our approach is
106: entirely analytical, we obtain closed expressions for the various phase
107: boundaries, which previously had to be extracted numerically. Moreover, our
108: present work also covers the case of attractive interactions, $U<0$, thus
109: mapping out the entire physically relevant phase diagram.
110:
111: To make our paper self-contained we start by briefly describing and
112: comparing the various approximation schemes employed. Density-functional
113: theory (DFT) \cite{kohnrmp,dftbook} shows that the total energy of a many-body
114: system in an arbitrary potential can be written
115: \be
116: E[n]= T_s[n] + E_H[n] + V[n] + E_{xc}[n],
117: \ee
118: where $T_s$ is the kinetic energy of noninteracting particles, $E_H$ is
119: the Hartree energy (i.e., the mean-field approximation to the interaction
120: energy), $V$ is the potential energy (arising from the spatial distribution of
121: the nuclei in solid-state and molecular applications of DFT, and from the
122: optical confining potential for atoms in optical traps), and $E_{xc}$ is the
123: exchange-correlation energy \cite{footnote1}. The local-density approximation
124: to any energy component $F$ is defined by $F^{LDA}=\int dr f(n)|_{n \to n\r}$
125: in the continuum, and $F^{LDA}= \sum_i f(n)|_{n \to n_i}$ on a lattice,
126: where $f$ denotes the per-volume or per-site value of $F$ in a spatially
127: uniform (homogeneous) system.
128:
129: Although DFT is mostly applied to {\it ab initio} electronic-structure
130: calculations in solid-state physics and quantum chemistry
131: \cite{dftbook,kohnrmp}, its formal framework is sufficiently general to permit
132: application also to wide classes of model Hamiltonians \cite{valtercondmat},
133: and, in particular, to the Hubbard model \cite{balda,gs,gsn}. A Bethe-Ansatz
134: based LDA functional for the
135: correlation energy of the one-dimensional Hubbard model was proposed
136: in \cite{balda} and applied to Mott insulating phases \cite{mottepl} and
137: superlattice structures \cite{superl}. These works employed a
138: parametrization of the per-site ground-state energy of the spatially
139: uniform Hubbard model, $e(n,U)$, which was constructed \cite{balda} to recover
140: exactly known results at $U=0$, $U\to \infty$ and $n=1$, and to be close
141: to data obtained from numerical solution of the Bethe-Ansatz (BA)
142: \cite{liebwu} equations inbetween \cite{footnote2}. For $ 0 \leq n \leq 1$
143: \be
144: e(n,U) =
145: -\frac{2\beta(U)}{\pi}\sin\left(\frac{\pi n}{\beta(U)}\right)
146: \label{lsoc1}
147: \ee
148: and for $ 1 \leq n \leq 2$
149: \be
150: e(n,U)=
151: (n-1)U -\frac{2\beta(U)}{\pi}\sin\left(\frac{\pi(2-n)}{\beta(U)}\right)
152: \label{lsoc2}
153: \ee
154: Here $\beta(U)$ is obtained from
155: \be
156: -\frac{\beta}{\pi}\sin\left(\frac{\pi}{\beta}\right) = -2\int_0^\infty
157: \:\frac{J_0(x)J_1(x)}{x(1 + e^{Ux/2})}\:dx =: -2 I(U),
158: \ee
159: and $J_m$ is the $m$'th order Bessel function. The difference between the
160: $n<1$ and the $n>1$ branch is responsible for a derivative discontinuity of
161: $E_c$ and the opening of a Mott gap at $n=1$ \cite{liebwu,mottepl}.
162:
163: To perform analytical calculations for attractive interactions, our first
164: task is to extend this construction to $U<0$.
165: The construction of a parametrization of the per-site energy
166: is not unique, but a convenient and accurate choice is
167: \be
168: e(n,U) = \frac{Un}{2} - 4I(|U|)\sin\left(\frac{\pi n}{2}\right).
169: \label{uneg}
170: \ee
171: This expression by construction satisfies the following exactly known
172: properties of $e(n,U)$ for negative $U$:
173: (i) $e(n=0,U) = 0$,
174: (ii) $e(n=1,U) = -4\,I(|U|) + U/2$,
175: (iii) $e(n,U=0) = -(4/\pi)\sin(\pi n/2)$,
176: (iv) $e(n,U \to -\infty) \to Un/2$,
177: (v) $\partial e(n,U)/\partial n|_{n=1} = U/2$, and
178: (vi) $e(2-n,U) + U(n-1) = e(n,U)$.
179: Note that unlike (\ref{lsoc1}) and (\ref{lsoc2}), the $U<0$ expression
180: (\ref{uneg}) is particle-hole symmetric [property (iv)], and defined on only
181: one branch.
182:
183: The Kohn-Sham approach treats $T_s$ exactly, by means of single-particle
184: orbitals, and approximates $E_{c}$, e.g., by the BA-LDA of Ref.~\cite{balda}.
185: \bea
186: E^{LDA}_{KS}[n,U] = T_s[n] + V[n] + E_H[n,U] + \sum_i e_c(n_i,U),
187: \label{kslda}
188: \eea
189: where for a spin-unpolarized system $E_H[n,U]=(U/4)\sum_i^L n_i^2$, and
190: $V[n] =\sum_i n_i V_i$ is the potential energy in the confining potential
191: $V_i$. $t_s(n)=e(n,U=0)$ is the per-site kinetic energy of noninteracting
192: particles, and the difference $e(n,U)-e_H(n,U)-t_s(n) =: e_c(n,U)$ is
193: by definition the correlation energy.
194:
195: The total-energy LDA (TLDA), employed below, uses an LDA both
196: for the kinetic and the correlation energy, so that
197: \bea
198: E^{TLDA}[n] = \sum_i t_s(n_i) + V[n] + E_H[n] + \sum_i e_c(n_i)
199: \\
200: = \sum_i e(n_i,U) + V[n],
201: \label{tlda}
202: \eea
203: where the second equality follows because the Hartree energy is a local
204: functional for an on-site interaction. Note that the TLDA approach is
205: different from the Thomas-Fermi approximation (TFA), which also employs an
206: LDA for $T_s$, but takes $E_c\equiv 0$.
207:
208: We now proceed to characterize the phase diagram. To this end we employ the
209: TLDA, which is more accurate than the TFA, and unlike KS procedures can yield
210: closed analytical results. The density profile is obtained from minimizing
211: the energy functional under the constraint of fixed total particle number.
212: Hence
213: \be
214: \frac{\delta(E-\mu N)}{\delta n_i} =
215: \frac{\partial e(n_i,U)}{\partial n_i} + V_i - \mu = 0.
216: \ee
217: From the above parametrizations (\ref{lsoc1}), (\ref{lsoc2}) and (\ref{uneg})
218: of $e(n,U)$ we obtain the Euler equations
219: \bea
220: -2\cos\left(\frac{\pi n_i}{\beta}\right) + V_i- \mu = 0 \hspace*{0.4cm}
221: 0 \leq n_i < 1 \label{euler1}\\
222: U + 2\cos\left(\frac{\pi (2-n_i)}{\beta}\right) + V_i- \mu = 0\ \hspace*{0.4cm}
223: 1 < n_i \leq 2 \label{euler2}
224: \eea
225: for $U>0$, and
226: \be
227: \frac{U}{2} - 2\pi I(|U|)\cos\left(\frac{\pi n}{2}\right) + V_i - \mu = 0
228: \label{euler3}
229: \ee
230: for $U<0$. Note that at $n_i=1$ the expression $e(n,U)$ for $U>0$ is
231: not differentiable and no Euler equation is obtained. The three Euler
232: equations (\ref{euler1}) - (\ref{euler3}) have a complex solution space
233: with a rich structure, which we now proceed to analyse.
234:
235: \begin{figure}
236: \centering
237: \includegraphics[height=80mm,width=65mm,angle=-90]{tflda_fig1.ps}
238: \caption {\label{fig1} Classification of density shapes across the
239: $i>0$ half of a harmonic trap with $V_i = 0.004 i^2$, for $U=4>U^*$.
240: The four dividing lines have the form $V_i + A$, where, from top to
241: bottom, $A=U+2$, $A=U+2 \cos(\pi/\beta)$, $A=- 2 \cos(\pi/\beta)$
242: and $A=-2$. The horizontal line at $\mu =4$ is the example discussed
243: in the main text.}
244: \end{figure}
245:
246: Let us first consider $U>0$. In the low-density region far from the trap
247: center the solution of Eq.~(\ref{euler1}) is
248: $n_i = \frac{\beta}{\pi}\arccos\left(\frac{V_i-\mu}{2}\right)$.
249: However, this solution only belongs to the interval $[0,1]$ if
250: \be
251: V_i - 2 \le \mu \le V_i- 2\cos\left(\pi/\beta\right).
252: \label{cond1}
253: \ee
254: Closer to the center of the trap $n_i>1$, and the relevant branch of the
255: solution is (\ref{euler2}), leading to
256: $n_i = 2-(\beta/\pi)\arccos((\mu - V_i-U)/2)$, provided that
257: \be
258: V_i + U + 2\cos\left(\pi/\beta\right) \le \mu \le V_i + U + 2.
259: \label{cond2}
260: \ee
261: Note that the conditions (\ref{cond1}) and (\ref{cond2}), which guarantee
262: that the solution found does indeed pertain to the density interval it is
263: obtained from, do not necessarily match continuously. In particular, if
264: \be
265: U+2\cos\left(\pi/\beta\right) > -2\cos\left(\pi/\beta\right)
266: \label{ustar}
267: \ee
268: there are values of $\mu$ for which
269: neither of the two branches of the Euler equation has a solution. The actual
270: density in this situation is then the one for which the Euler equations
271: are not defined, i.e., $n_i=1$. This possibility occurs only for $U>U^*$,
272: where $U^*=1.7349...$ is the lowest value of $U$ satisfying (\ref{ustar}).
273:
274: \begin{figure}
275: \centering
276: \includegraphics[height=80mm,width=65mm,angle=-90]{tflda_fig2.ps}
277: \caption {\label{fig2} Typical density profiles for $U=4$ and a
278: confining potential $V_i=0.004 i^2$, for five different values of the
279: total fermion number, ilustrating the five basic types of density profiles
280: obtained for repulsive interactions.}
281: \end{figure}
282:
283: In Fig.~\ref{fig1} we plot the four limiting expressions defining the
284: boundaries of Eqs.~(\ref{cond1}) and (\ref{cond2}) for $U> U^*$. The chemical
285: potential is constant throughout the trap, and thus corresponds to horizontal
286: lines crossing the various boundaries. As an illustration, consider a system
287: with $\mu =4$ (horizontal line in Fig.~\ref{fig1}).
288: At the trap center ($i=0$) one first finds a high-density
289: solution with $1 < n_i < 2$. Moving outward, one crosses the left boundary
290: of (\ref{cond2}), encountering a region in which neither Euler equation
291: has a solution and the density is fixed at $n_i=1$. Further out a solution
292: of (\ref{cond1}) becomes possible, and for $i>39$ even the low-density
293: Euler equation ceases to have a solution, implying $n_i=0$.
294: Density profiles for $U=4$ and different fillings are shown in Fig.~\ref{fig2}.
295: Each of the data sets in Fig.~\ref{fig2} illustrates one of the five phases
296: of the $U>0$ part of the phase diagram, Fig.~\ref{fig3}. These are the same
297: five phases also obtained numerically in Refs.~\cite{drummond,rigol}.
298:
299: The present TLDA approach localizes the sites with $n_i=1$ by the conditions
300: that $0<n<2$ and $\mu$ such that neither Euler equation has a solution.
301: On the other hand, in a Kohn-Sham approach \cite{polini,balda,gs,gsn} a
302: self-consistent solution is obtained iteratively, yielding in each iteration
303: step a density that is defined simultaneously at all sites. If for particular
304: sites there is no value of $n_i$ corresponding to a solution of the Euler
305: equations, the Kohn-Sham self-consistency cycle does not find any solution
306: at all and does not converge. This is precisely what was observed in trying
307: to perform calculations of the type of Refs.~\cite{balda,polini} in a region
308: with plateaus. The TLDA procedure, which allows to obtain and characterize
309: solutions site by site, has no such problem.
310:
311: The negative $U$ analysis is performed in the same way, by
312: starting from Eq.~(\ref{euler3}). Since there is only one branch, the
313: criterium for a valid solution is
314: \be
315: \frac{U}{2} + V_i - 2\pi I(|U|) \le \mu \le \frac{U}{2} + V_i + 2\pi I(|U|).
316: \ee
317: For $\mu$ inside this interval the solution of (\ref{euler3}) is
318: $n_i = (2/\pi)\arccos\left((V_i + U/2 - \mu)/(2\pi I(|U|))\right)$.
319: For $\mu < \frac{U}{2} + V_i - 2\pi I(|U|)$ the solution
320: is $n_i=0$ and for $\mu > \frac{U}{2} + V_i + 2\pi I(|U|)$ it is $n_i=2$.
321: Typical density profiles in this region (not shown) are of the same type as
322: for $U>0$, with possible plateaus at $n=0$ and $n=2$. The absence of any
323: possible plateaus with $n=1$ is due to
324: the absence of a derivative discontinuity of $E_c$. The atomic density waves
325: found from Kohn-Sham calculations in Ref.~\cite{polini} are not reproduced by
326: the TLDA, showing that observation of these oscillations (just as Friedel
327: oscillations) require an exact (orbital) treatment of the single-body kinetic
328: energy.
329:
330: \begin{figure}
331: \centering
332: \includegraphics[height=80mm,width=65mm,angle=-90]{tflda_fig3.ps}
333: \caption {\label{fig3} The $U$-$\mu$ phase diagram. See main text for details.}
334: \end{figure}
335:
336: The expresions for the phase boundaries obtained within TLDA can be
337: used to analytically construct the full $U$-$\mu$ phase diagram, shown in
338: Fig.~\ref{fig3}. For a harmonic confining potential of the form $V_i=k i^2$
339: region II (characterized by a central plateau with $n=1$, i.e., a local
340: Mott-insulator-like state) is bounded by
341: $k - 2\cos(\pi/\beta) < \mu < U + 2\cos(\pi/\beta)$.
342: Region III (two lateral plateaus with $n=1$) is bounded by
343: $U+2\cos(\pi/\beta)<\mu<k + U/2 + (U/4 + \cos(\pi/\beta))^2/k$,
344: and region V (a central band-insulator-like plateau with $n=2$) by
345: $\mu > U + 2 + k$.
346: Region IV is the set of points satisfying simultaneously the criteria for
347: region III and V, i.e., has lateral plateaus with $n=1$ and a central plateau
348: with $n=2$.
349: Region I (purely metallic without any plateaus, except for, possibly, sites
350: with $n=0$ at the wings of the trap) is the set of points with $\mu>-2$
351: that do not belong to any other $U>0$ region.
352: In the TFA ($E_c\equiv 0$) the $U>0$ phases II, III and IV disappear, and the
353: dividing line between phase I and V becomes $\mu = U + 2 + k$. Phases II, III
354: and IV thus owe their existence to correlation.
355: Region VI is the $U<0$ extension of region V and characterized by a
356: plateau at $n=2$ in the trap center. It is bounded from below by
357: $\mu=k+U/2+2\pi I(|U|)$.
358: Region VII, finally, is the $U<0$ extension of region I and characterized
359: by the absence of any plateaus except for, possibly, sites with $n=0$ at the
360: wings of the trap.
361:
362: Special values of $U$ and $\mu$ can be used to further classify the
363: possibilities: For $\mu<\mu^*$, where $\mu^*(U>0)=-2$ and
364: $\mu^*(U<0)= U/2-2 \pi I(|U|)$, the only solution is $n_i=0$ at all sites,
365: i.e., an empty system (hatched region in the phase diagram).
366: For $0<U<U^*$ the intervals permitting solutions of (\ref{cond1}) and
367: (\ref{cond2}) overlap, and no $n=1$ plateaus can appear, leaving only
368: phases I, V and VI. (In this situation, at isolated sites, an $n>1$
369: and an $n<1$ solution can coexist.) For $U^*<U<U^{**}(k)$, where $U^{**}(k)$
370: is the solution of $(U/4 + \cos(\pi/\beta))^2=k(U/2+2)$, there cannot be a
371: phase IV (i.e, plateaus at $n=1$ {\em and} $n=2$) and instead there appears a
372: reentrant metallic phase of type I above phase III.
373:
374: In summary, we have constructed the complete $U$-$\mu$ phase diagram of
375: one-dimensional
376: interacting harmonically confined fermions. For $U>0$ we confirm earlier
377: numerical results, but complement them by providing analytical expressions
378: for the phase boundaries. The TLDA approach also provides an explanation
379: why Kohn-Sham calculations do not work in the metal Mott-insulator
380: phase-separated state (with plateaus at $n=1$) and provides an alternative way
381: for obtaining density profiles in this region. For $U<0$ we propose, in
382: Eq.~(\ref{uneg}), a simple analytical parametrization of the Bethe-Ansatz
383: solution, which we use, within TLDA, to extend the phase diagram to attractive
384: interactions.
385:
386: {\bf Acknowledgments}\\
387: This work was sup\-por\-ted by FAPESP and CNPq. We thank Mariana Odashima
388: for useful discussions on the Bethe-Ansatz parametrization for $U<0$ and
389: Marco Polini for useful discussions on the physics at $U<0$.
390:
391: \begin{thebibliography}{99}
392: \bi{atoms1} W. Ketterle, Rev. Mod. Phys. {\bf 74}, 1131 (2002).
393: \bi{atoms2} C. A. Regal, M. Greiner and D. S. Jin, Phys. Rev. Lett. {\bf 92},
394: 040403 (2004).
395: \bi{atoms3} S. L. Cornish, N. R. Claussen, J. L. Roberts, E. A. Cornell and
396: C. E. Wieman, Phys. Rev. Lett. {\bf 85}, 1795 (2000).
397: \bi{atoms4} J. Weiner, V. S. Bagnato, S. Zilio and P. S. Julienne,
398: Rev. Mod. Phys. {\bf 71}, 1 (1999).
399: \bi{footnote0} Below all energies ($U$, $\mu$, $V$, etc) are given in
400: multiples of the hopping parameter $t$.
401: \bi{drummond} X.-J. Liu, P. D. Drummond and H. Hu, Phys. Rev. Lett. {\bf 94},
402: 136406 (2005).
403: \bi{rigol} M. Rigol, A. Muramatsu, G. G. Batrouni and R. T. Scalettar,
404: Phys. Rev. Lett. {\bf 91}, 130403 (2003).
405: \bi{polini} G. Xianlong, M. Polini, M. P. Tosi, V. L. Campo, Jr. and
406: K. Capelle, cond-mat/0506570.
407: \bi{balda} N.~A.~Lima, M.~F.~Silva, L.~N.~Oliveira and K.~Capelle,
408: Phys. Rev. Lett. {\bf 90}, 146402 (2003).
409: \bi{kohnrmp} W.~Kohn, Rev. Mod. Phys. {\bf 71}, 1253 (1999).
410: \bi{dftbook} R.~M.~Dreizler and E.~K.~U.~Gross,
411: {\em Density Functional Theory} (Springer, Berlin, 1990).
412: \bi{footnote1} For the Hubbard model the constraint
413: of allowing double occupation only with opposite spins zeros the exchange
414: energy. In all calculations below $E_x$ is thus identically zero, and only
415: $E_c$ needs to be approximated.
416: \bi{valtercondmat} V.~L.~Libero and K.~Capelle, cond-mat/0506206.
417: \bi{gs} O.~Gunnarsson and K.~Sch\"onhammer, Phys. Rev. Lett. {\bf 56},
418: 1968 (1986).
419: \bi{gsn} K.~Sch\"onhammer, O.~Gunnarsson and R.~M.~Noack, Phys. Rev. B
420: {\bf 52}, 2504 (1995).
421: \bi{mottepl} N. A. Lima, L. N. Oliveira and K. Capelle,
422: Europhys. Lett. {\bf 60}, 601 (2002).
423: \bi{superl} M.~F.~Silva, N.~A.~Lima, A.~L.~Malvezzi and K.~Capelle,
424: Phys. Rev. B {\bf 71}, 125130 (2005).
425: \bi{liebwu} E.~H.~Lieb and F.~Y.~Wu, Phys. Rev. Lett. {\bf 20}, 1445 (1968).
426: \bi{footnote2} A BA-LDA for $E_c$ that is directly based on these numerical
427: data, bypassing the use of interpolations, was used recently to study, by
428: means of Kohn-Sham calculations, the Luther-Emery correlations and atomic
429: density waves appearing for $U<0$ \cite{polini}. A similar approach was used
430: earlier \cite{gsn} for the $U>0$ ionic Hubbard model. Numerical BA-based LDA
431: functionals for the total energy have been used, e.g., in Ref.~\cite{drummond}
432: for the Hubbard model and Ref.~\cite{grabert} for point-contact interactions.
433: \bi{grabert} L. Kecke, H. Grabert and W. H\"ausler, Phys. Rev. Lett. {\bf 94},
434: 176802 (2005).
435: \end{thebibliography}
436:
437: %)}]
438: \end{document}
439: