1: %% LyX 1.4.3 created this file. For more info, see http://www.lyx.org/.
2: %% Do not edit unless you really know what you are doing.
3: \documentclass[english,twocolumn,showpacs,preprintnumbers,amsmath,amssymb,pra]{revtex4}
4: \usepackage[T1]{fontenc}
5: \usepackage[latin1]{inputenc}
6: \usepackage{prettyref}
7: \usepackage{amsmath}
8: \usepackage{graphicx}
9: \usepackage{amssymb}
10:
11: \makeatletter
12:
13: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
14: %% A simple dot to overcome graphicx limitations
15: \newcommand{\lyxdot}{.}
16:
17:
18: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
19: %% LyX 1.4.3 created this file. For more info, see http://www.lyx.org/.
20: %% Do not edit unless you really know what you are doing.
21:
22:
23:
24:
25:
26:
27: \makeatletter
28:
29: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LyX specific LaTeX commands.
30: %% Bold symbol macro for standard LaTeX users
31:
32:
33:
34: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
35:
36: %\documentclass[a4paper,showpacs,english,aps,pra,twocolumn]{revtex4}
37:
38:
39: %\usepackage{amsmath}
40: \usepackage{bm}
41:
42:
43: \makeatletter
44:
45: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
46: %\usepackage{comment}
47: %\renewcommand{\vec}[1]{\mathbf{#1}}
48: \renewcommand{\Re}[1]{\mathrm{Re\,}}
49: \renewcommand{\Im}[1]{\mathrm{Im\,}}
50: \renewcommand{\vec}[1]{\bm{#1}}
51:
52: %%\makeatother
53:
54:
55:
56: \makeatother
57:
58: \usepackage{babel}
59: \makeatother
60: \begin{document}
61:
62: \title{Non-BCS superfluidity in trapped ultracold Fermi gases}
63:
64:
65: \author{L. M. Jensen$^{\text{1}}$, J. Kinnunen$^{\text{1,2}}$ and P. T\"{o}rm\"{a}$^{\text{1}}$}
66:
67:
68: \affiliation{$^{1}$Department of Physics, Nanoscience Center, P.O.Box 35, FIN-40014
69: University of Jyv\"{a}syl\"{a}, Finland\\
70: $^{2}$JILA and Department of Physics, University of Colorado at Boulder,
71: CO 80309-0440, USA }
72:
73:
74: \pacs{03.75.Ss, 03.75.Hh, 05.30.Jp}
75:
76: \begin{abstract}
77: Superconductivity and superfluidity of fermions require, within the
78: BCS theory, matching of the Fermi energies of the two interacting
79: Fermion species. Difference in the number densities of the two species
80: leads either to a normal state, to phase separation, or - potentially
81: - to exotic forms of superfluidity such as FFLO-state, Sarma state
82: or breached pair state. We consider ultracold Fermi gases with polarization,
83: i.e.\ spin-density imbalance. We show that, due to the gases being
84: trapped and isolated from the environment in terms of particle exchange,
85: exotic forms of superfluidity appear as a shell around the BCS-superfluid
86: core of the gas and, for large density imbalance, in the core as
87: well. We obtain these results by describing the effect of the trapping
88: potential by using the Bogoliubov-de Gennes equations. For comparison
89: to experiments, we calculate also the condensate fraction, and show
90: that, in the center of the trap, a polarized superfluid leads to
91: a small dip in the central density difference. We compare the results
92: to those given by local density approximation and find qualitatively
93: different behavior.
94: \end{abstract}
95: \maketitle
96: %\email{melwyn@phys.jyu.fi}
97:
98:
99:
100: \affiliation{$^{\text{1}}$Department of Physics, Nanoscience Center, P.O.Box
101: 35, FIN-40014 University of Jyväskylä, Finland \\
102: $^{\text{2}}$JILA and Department of Physics, University of Colorado
103: at Boulder, CO 80309-0440, USA}
104:
105:
106: \section{Introduction}
107:
108: There are several suggestions of non-BCS superfluidity for fermion
109: systems with spin-population imbalance, i.e. with the polarization
110: $P=(N_{\uparrow}-N_{\downarrow})/(N_{\uparrow}+N_{\downarrow})\neq0$
111: where $N_{\sigma}$ are the particle numbers of the (pseudo)spins.
112: The FFLO-state \cite{Fulde1964,Larkin1965}, the Sarma state \cite{Sarma1963}
113: and the breached pair (BP) state \cite{Liu2003} all appear as extremal
114: points of the mean-field energy of the system. Such superfluidity
115: is of interest e.g. in condensed matter, high-energy and nuclear physics
116: \cite{Casalbuoni2004}, but firm experimental evidence is lacking.
117: With the recently realized superfluids of alkali Fermi gases, the
118: first studies of density-imbalanced gases \cite{Zwierlein2005c,Partridge2005,Zwierlein2006a,Partridge2006c}
119: have shown that these systems offer unprecedented opportunities for
120: investigating this question. The FFLO-state is predicted to appear
121: in a narrow parameter window in several systems \cite{Sedrakian2005,Mizushima2005a,Castorina2005,Sheehy2006},
122: and the Sarma/BP state has been shown to be unstable under many conditions
123: \cite{Gubankova2003,Forbes2005}. The existence of these exotic forms
124: of superfluidity is thus an intriguingly subtle question and requires
125: careful analysis, taking into account the specific features of the
126: physical system. In this article, we demonstrate the important consequences
127: of such features in case of trapped ultracold alkali gases. First,
128: in ultracold gases, the physical system under study is isolated from
129: the environment in terms of particle exchange. This could be contrasted
130: to an electronic system where external voltage fixes the chemical
131: potential by allowing particle exchange between the system of interest
132: and the environment. Second, the gas is trapped by an inhomogeneous
133: potential (often of harmonic form). The second issue leads to phase
134: separation of superfluid and normal phases, as shown by a series of
135: experiments \cite{Zwierlein2005c,Partridge2005,Zwierlein2006a,Zwierlein2006c,Shin2006a,Partridge2006b}
136: and theoretical studies \cite{Bedaque2003,Sheehy2006,Pieri2006a,Kinnunen2006a,Yi2006a,Haque2006,Chevy2006,deSilva2006a,Chien2006b,Gubbels2006b}.
137: The finiteness and harmonic confinement of the system leads, as we
138: show in this article, to stabilization of exotic forms of superfluidity
139: in a shell surrounding a BCS-superfluid core, and eventually in the
140: core itself.
141:
142: Ultracold Fermi gases offer the possibility of realizing the BCS-BEC
143: crossover with the use of Feshbach resonances. At the resonance point,
144: the system is a strongly interacting Fermi superfluid, and on different
145: sides away from the resonance it is either a BEC of molecules or a
146: weakly interacting Fermi gas realizing a BCS superfluid. Condensation
147: of molecules, fermion pairs, pairing gap and the crossover behavior
148: have been studied by a series of experiments and vortices confirming
149: superfluidity have been created, for a review see for instance~\cite{Grimm2005}.
150: Recently, seminal experiments studying density imbalanced (nonzero
151: polarization $P$) Fermi gases were done~\cite{Zwierlein2005c,Partridge2005,Zwierlein2006a}.
152: The analysis presented in \cite{Zwierlein2005c,Shin2006a,Zwierlein2006c}
153: combined the study of vortex patterns, density profiles (in \cite{Shin2006a}
154: 3D reconstructed) and condensate fractions, and thereby showed that,
155: in the trapped system, a superfluid core (with equal densities of
156: the components $\uparrow$ and $\downarrow$) appears in the center
157: of the trap and the excess atoms of the majority component ($\uparrow$)
158: tend to be located on the edges of the trap. In the following, the
159: BCS paired superfluid (SF) implies equal local densities, whereas
160: the polarized superfluid (PS) at zero temperature denotes exotic forms
161: of superfluidity with non-zero local density difference of the $\uparrow$
162: and $\downarrow$ components. In addition, the normal state (N) shell
163: surrounding the condensate can either be partially polarized or fully
164: polarized depending on whether the minor component is present or not.
165:
166: In connection to the experiments \cite{Zwierlein2005c,Partridge2005,Zwierlein2006a},
167: several authors have analyzed the trapped, polarized Fermi gas using
168: the local density approximation (within mean-field theory) \cite{Sheehy2006,Pieri2006a,deSilva2006a,Chevy2006,Yi2006a,Haque2006,Chien2006b,Gubbels2006b,beyond}.
169: In these works, the problem was solved at each spatial point, applying
170: a local chemical potential, and the stability of the solutions at
171: that point was determined by their grand potential energies. This
172: leads, at zero temperature, to the exclusion of Sarma/BP state which
173: is known to be a maximum point of the grand potential for fixed chemical
174: potentials. This gives the following qualitative picture: a superfluid
175: (SF) core, with equal densities, appears at the center of the trap,
176: and is surrounded by a normal state (N) shell (only on the BEC side
177: of the Feshbach resonance a coexistence of molecular BEC and free
178: atoms was seen, as obviously expected since molecule formation does
179: not require matching Fermi energies). This is, however, a rather approximative
180: description of the system because LDA assumes a smooth variation of
181: the density. Since the studies \cite{Sheehy2006,deSilva2006a,Chevy2006,Yi2006a,Haque2006,Chien2006b}
182: imply a step in the density at the interface of the SF core and the
183: N shell one can question the validity of LDA at that boundary, as
184: also remarked by many of the authors of \cite{Sheehy2006,deSilva2006a,Chevy2006,Yi2006a,Haque2006,Chien2006b}
185: and recent beyond-LDA studies \cite{Partridge2006b,Imambekov2006,deSilva2006b}.
186: %
187: \begin{figure}
188: \begin{centering}\includegraphics[width=7cm]{fig-1new}\par\end{centering}
189:
190:
191: \caption{(color online) The gap $\Delta(r)$ and density difference $\delta n=n_{\uparrow}(r)$
192: -- $n_{\downarrow}(r)$ profiles for the LDA and BdG calculations.
193: \label{fig:comp-zero-T}}
194: \end{figure}
195:
196:
197: The existence of a PS for trapped, strongly interacting gases was
198: indicated already in our earlier study \cite{Kinnunen2006a}, where
199: the treatment of the trapping potential by Bogoliubov-de Gennes (BdG)
200: equations revealed oscillations of the order parameter in an area
201: located between the SF core and the N shell. Such oscillations resemble
202: the nonuniform order parameter associated with the FFLO state, therefore
203: we refer to these oscillations as \char`\"{}FFLO-type state\char`\"{}.
204: In this article, we show that the BdG analysis predicts such polarized
205: superfluid in trapped Fermi gases not only as a shell effect but,
206: for large polarizations, as a feature that extends through the whole
207: system. We confirm that superfluidity and finite local density difference
208: indeed co-exist in the center of the trap by calculating the condensate
209: fraction, central gap and the core polarization. In previous works
210: using BdG for trapped gases \cite{Castorina2005,Mizushima2005a,Kinnunen2006a,Machida2006}
211: such an analysis, confirming the existence of a polarized superfluid
212: in the center, was not performed. Moreover, the works \cite{Castorina2005,Mizushima2005a,Machida2006}
213: considered the weak interaction limit whereas we have extended the
214: BdG calculation to the unitarity regime and actually consider the
215: whole crossover from the BCS to BEC side. We analyze the dependence
216: of the oscillations on the system size (particle number) and discuss
217: the connection to phenomena occurring in superconductor - ferromagnet
218: interfaces. For comparison, we also perform calculations using LDA.
219: The LDA analysis predicts a polarized superfluid only at finite temperature,
220: not at zero temperature like the BdG calculation. Therefore our results
221: show that, in trapped Fermi gases, LDA has to be applied with care;
222: beyond LDA approaches are not only needed for describing the interfaces
223: of the system properly but, in some cases, features not predicted
224: by LDA may become dominant.
225:
226: %
227: \begin{figure}
228: \begin{centering}\includegraphics[width=7cm]{fig-1new-2}\par\end{centering}
229:
230:
231: \caption{(color online) The density profiles of the $n_{\uparrow}(r)$ for
232: BdG (solid) and LDA (dashed), and $n_{\downarrow}(r)$ for BdG (dot-dashed)
233: and LDA (dotted). \label{fig:dens-comp-zero-T}}
234: \end{figure}
235: This article is structured as follows. In section II. we present a
236: review of the Bogoliubov - de Gennes (BdG) approach for describing
237: pairing at the mean field level. We discuss the use of local density
238: approximation, and the expansion in harmonic oscillator states as
239: special cases of this general scheme. In the rest of the paper, we
240: use the following terminology: BdG refers to the case where harmonic
241: oscillator state basis has been used, and LDA to the use of local
242: density approximation. The results are presented in section II. The
243: appearance of the FFLO-type oscillations is discussed in subsection
244: II.A and the case of large polarizations, when such oscillations span
245: the whole system, is discussed in section II.B. We also analyze the
246: dependence of these features on the interaction strength through the
247: BCS-BEC crossover (section II.A.1.) and on the system size (atom number)
248: (section II.A.2). The condensate fraction is calculated in section
249: II.C. and the contributions of different harmonic oscillator states
250: to the pairing are discussed in section II.D. Conclusions and discussion
251: are presented in section III. %
252: \begin{figure}
253: \begin{centering}\includegraphics[width=7cm]{fig-1\lyxdot beta-1}\par\end{centering}
254:
255:
256: \caption{(color online) The gap and polarization profiles at slightly elevated
257: temperature $T/T_{\mathrm{F}}=0.05$ for the LDA calculations and
258: at low polarization. In contrast to the zero temperature case the
259: thermal excitation in the BCS core starts to show up as a tail of
260: the density difference into the core. \label{fig:comp-finite-T}}
261: \end{figure}
262:
263:
264:
265: \section{Bogoliubov - de Gennes approach}
266:
267: In order to properly account for the inhomogeneity due to the presence
268: of the trap potential we will in the following use the Bogoliubov-de
269: Gennes equations \cite{deGennes1999}, which in a more general context
270: are also known as the selfconsistent Hartree-Fock-Bogoliubov equations.
271: The imbalanced two-component Fermi gas is described by the grand canonical
272: Hamiltonian \begin{eqnarray}
273: H & = & \sum_{\sigma}\int d^{3}\vec{r}\ \hat{\Psi}_{\sigma}^{\dagger}(\vec{r})\left(-\frac{\hbar^{2}\nabla^{2}}{2m}-\mu_{\sigma}+V(\vec{r})\right)\hat{\Psi}_{\sigma}(\vec{r})\nonumber \\
274: & + & \int d^{3}\vec{r}d^{3}\vec{r}^{\prime}\ \hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})U(\vec{r}-\vec{r}^{\prime})\hat{\Psi}_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r}),\end{eqnarray}
275: where $\hat{\Psi}_{\sigma}(\vec{r}),\hat{\Psi}_{\sigma}^{\dagger}(\vec{r})$
276: are the real space annihilation and creation operators for an atom
277: with spin $\sigma$ at position $\vec{r}$, $\mu_{\sigma}$ is the
278: chemical potential for the components $\sigma,$ $V(\vec{r})=\frac{1}{2}m\omega^{2}r^{2}$
279: is the external (isotropic and harmonic) trapping potential, and $U(\vec{r})=U\delta(\vec{r})$
280: is the interatomic atom-atom contact interaction potential. We apply
281: the contact potential interaction and the mean-field approximation
282: \begin{eqnarray}
283: U\hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r}) & \approx & -\frac{\left|\Delta(r)\right|^{2}}{U}-Un_{\uparrow}(\vec{r})n_{\downarrow}(\vec{r})\nonumber \\
284: & & \hspace{-3.5cm}+\, Un_{\uparrow}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}(\vec{r})+Un_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r})\nonumber \\
285: & & \hspace{-3.5cm}+\,\Delta(\vec{r})\hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})+\Delta^{\ast}(\vec{r})\hat{\Psi}_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r}),\end{eqnarray}
286: where $\Delta(\vec{r})=\langle\hat{\Psi}_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r})\rangle$
287: and $n_{\sigma}(\vec{r})=\langle\hat{\Psi}_{\sigma}^{\dagger}(\vec{r})\hat{\Psi}_{\sigma}(\vec{r})\rangle$.
288: The first two terms are simply numbers, so they are neglected. This
289: gives the following mean-field Hamiltonian\begin{eqnarray}
290: H_{\mathrm{MF}} & = & \sum_{\sigma}\int d^{3}\vec{r}\ \hat{\Psi}_{\sigma}^{\dagger}(\vec{r})\left(-\frac{\hbar^{2}\nabla^{2}}{2m}-\mu_{\sigma}+V(\vec{r})\right)\hat{\Psi}_{\sigma}(\vec{r})\nonumber \\
291: & + & \sum_{\sigma}\int d^{3}\vec{r}\ Un_{\sigma}(\vec{r})\hat{\Psi}_{\bar{\sigma}}^{\dagger}(\vec{r})\hat{\Psi}_{\bar{\sigma}}(\vec{r})\nonumber \\
292: & + & \int d^{3}\vec{r}\left(\Delta(\vec{r})\hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})+\ \mathrm{H.c.}\right),\label{eq:MFH}\end{eqnarray}
293: with $\bar{\sigma}$ being the other component of $\sigma.$ The second
294: line is the mean-field Hartree corrections which represents an asymmetric
295: shift to the chemical potential $\mu_{\sigma}$ of the gas of $\sigma$
296: atoms which is proportional to the density of the other component
297: $n_{\bar{\sigma}}.$ In the Bogoliubov-de Gennes approach one expands
298: the field operators on an appropriate basis set, dictated by the symmetries
299: of the non-interacting part of the Hamiltonian and usually characterized
300: by a set $\{\eta\}$ of good quantum numbers, in order to diagonalize
301: the Hamiltonian as in the uniform case. In the balanced non-uniform
302: case and in the absence of superflow the generalized BCS pairing
303: is in general between atoms in time-reversed states. The problem can
304: now be solved by introducing the generalized canonical Bogoliubov-Valatin
305: transformation to the new fermionic operators $\alpha_{\eta},\alpha_{\eta}^{\dagger}$
306: which amounts to the expansion \[
307: \hat{\Psi}_{\sigma}(\vec{r})=\sum_{\eta}u_{\eta}(\vec{r})\hat{\alpha}_{\eta\sigma}-s_{\sigma}v_{\eta}^{\ast}(\vec{r})\hat{\alpha}_{\bar{\eta}\bar{\sigma}}^{\dagger},\]
308: where the overhead bar designates the quantum numbers of the time-reversed
309: state for $\eta$ and the other hyperfine spin state for $\sigma,$
310: and with $s_{\uparrow}=1,s_{\downarrow}=-1.$ We note that in the
311: analogous case of electronic pairing in superconductors $\bar{\sigma}$
312: denote the time-reversed spin part of the wavefunction. From the requirement
313: that the new operators $\alpha_{\eta},\alpha_{\eta}^{\dagger}$ diagonalize
314: the Hamiltonian \prettyref{eq:MFH} one derives the matrix Bogoliubov-de
315: Gennes equation \cite{deGennes1999} \begin{equation}
316: (\hat{H}_{0}\hat{\tau}_{3}-\hat{\Delta}\hat{\tau}_{1})\vec{\varphi}_{\eta}=E_{\eta}\vec{\varphi}_{\eta}(\vec{r}),\label{eq:bdg-eqn}\end{equation}
317: for the spinor $\vec{\varphi}(\vec{r})\equiv(u_{\eta}(\vec{r}),v_{\eta}(\vec{r}))$,
318: where $\eta$ denotes a set of appropriate quantum numbers and $u_{\eta},v_{\eta}$
319: are therefore to be regarded as subspinors, $\hat{H}_{0}$ is the
320: non-interacting diagonal part of the Hamiltonian, potentially including
321: a trapping potential and Hartree shifts to the chemical potentials,
322: $\hat{\Delta}$ is the pairing field part of the Hamiltonian, $E_{\eta}$
323: is the eigenenergy. The products with the Pauli matrices $\hat{\tau}_{i}$
324: on the left hand side of Eq. \prettyref{eq:bdg-eqn} are to be understood
325: as a direct products. The selfconsistent aspect of the method is due
326: to the fact that the chemical potentials, mean-field Hartree and pairing
327: fields are to be selfconsistently determined through the gap and number
328: equations. The details of the BdG calculation for the harmonic trap
329: eigenstates are presented in Section \prettyref{sub:Harmonic-trap-}.
330: Next we turn to the discussion of the local density approximation.
331:
332:
333:
334: \subsection{Local density approximation}
335:
336: We first consider the local density approximation which is assumed
337: to be valid for sufficiently large condensates. The starting point
338: for the LDA calculation is to solve the problem in the uniform case.
339: The translational invariance of a uniform superfluid implies that
340: the plane wave states $\hat{\Psi}(\vec{r)}=\mathcal{V}^{-1/2}\sum_{\vec{k}}e^{i\vec{k}\cdot\vec{r}}a_{\vec{k}}$
341: can be used to diagonalize the Hamiltonian. The main assumption of
342: the LDA is that the system is locally homogeneous and therefore we
343: initially consider the Hamiltonian $H$ in \prettyref{eq:uni-bcs-hamiltonian}
344: for a uniform system (i.e. $V(r)=0$) with contact interactions, which
345: in the momentum representation reads \begin{equation}
346: H=\sum_{\vec{k},\sigma}\varepsilon_{\vec{k}}a_{\vec{k}\sigma}^{\dagger}a_{\vec{k}\sigma}+\frac{U}{\mathcal{V}}\sum_{\vec{k},\vec{k}^{\prime}}a_{\vec{k}\uparrow}^{\dagger}a_{\vec{k}\downarrow}^{\dagger}a_{-\vec{k}^{\prime}\downarrow}a_{\vec{k}^{\prime}\uparrow},\label{eq:uni-bcs-hamiltonian}\end{equation}
347: where $a_{\vec{k}\sigma}^{\dagger},a_{\vec{k}\sigma}$ are the creation
348: and annihilation operators for free atoms with momentum $\hbar\vec{k}$,
349: and the kinetic energy $\varepsilon_{\vec{k}}=\hbar^{2}k^{2}/(2m)$.
350: The chemical potentials $\mu_{\sigma}$ are introduced as Lagrange
351: multipliers for the particle numbers $N_{\sigma}$. We define the
352: average chemical potential $\mu=(\mu_{\uparrow}+\mu_{\downarrow})/2$
353: and the imbalance potential $\delta\mu=(\mu_{\uparrow}-\mu_{\downarrow})/2$,
354: such that $\mu_{\sigma}=\mu+s_{\sigma}\delta\mu,$ with $s_{\sigma}$
355: defined as above. Within the mean field approximation the Hamiltonian
356: can be diagonalized by the standard Bogoliubov-Valatin transformation.
357: In the uniform case the order parameter is the usual $\Delta=\langle a_{\vec{k}\uparrow}a_{-\vec{k}\downarrow}\rangle$
358: which satisfies the gap equation. Within the semiclassical approximation
359: the gap $\Delta(r)$ is assumed to depend weakly on the radial position
360: $r=|\vec{r}|$ and, as is commonly done, we introduce the local chemical
361: potential $\mu_{\sigma}\to\mu_{\sigma}(r)=\mu_{\sigma}-V(r),$ where
362: $V(r)$ is the trapping potential. The quasi-particle dispersions
363: depend on the gap and therefore on the radial position through the
364: local chemical potentials and the gap profile. In the most general
365: case the quasi-particle dispersion relation becomes \[
366: E_{\vec{k}\sigma}(\vec{r})=\frac{1}{2}\left(\xi_{\vec{k}\sigma}(r)-\xi_{\vec{k}\bar{\sigma}}(r)\right)+E_{\vec{k}}(r),\]
367: where $E_{\vec{k}}(r)=[\xi_{\vec{k}}(r)+\Delta^{2}(r)]^{1/2},$ with
368: $\xi_{\vec{k}}(r)=\left(\xi_{\vec{k}\sigma}(r)+\xi_{\vec{k}\bar{\sigma}}(r)\right)/2,$
369: and $\xi_{\vec{k}\sigma}(r)=\hbar^{2}k^{2}/(2m)-\mu_{\sigma}(r).$
370: As we will assume pairing between atoms with equal mass we get $E_{\vec{k}\sigma}=-s_{\sigma}\delta\mu+E_{\vec{k}},$
371: and therefore ${2E}_{\vec{k}}=E_{\vec{k}\sigma}+E_{\vec{k}\bar{\sigma}},$
372: and $\xi_{\vec{k}}=\varepsilon_{\vec{k}}-\mu(r)$ with $\mu,\delta\mu$
373: defined above. Within LDA, both the gap and the local chemical potentials
374: depend on the radial position and therefore the system may in general
375: be locally polarized. We define an averaged Fermi energy scale $E_{\mathrm{F}}$
376: from the total atom number $N=[E_{\mathrm{F}}/(\hbar\omega)]^{3}/3$,
377: for a balanced non-interacting Fermi gas in a harmonic trap potential
378: $V(r)=m\omega^{2}r^{2}/2$, where $\omega$ is the trap frequency.
379: The characteristic scale of the size of the cloud is the Thomas-Fermi
380: radius is $R_{\mathrm{TF}}=[2E_{\mathrm{F}}/(m\omega^{2})]^{1/2}$.
381: Within LDA the local gap equation at position $f$ reads \begin{equation}
382: 1=\frac{U_{\ast}}{\mathcal{V}}\sum_{\vec{k}}\left[\frac{1}{2\varepsilon_{\vec{k}}}-\frac{1-n_{\mathrm{F}}(E_{\vec{k}\uparrow}(r))-n_{\mathrm{F}}(E_{\vec{k}\downarrow}(r))}{2E_{\vec{k}}(r)}\right],\label{eq:gap}\end{equation}
383: with $U_{\ast}=4\pi\hbar^{2}a_{s}/m$ being the usual regularized
384: effective interaction strength which replaces the bare interaction
385: $U,$ and $a_{s}$ is the $s$-wave scattering length, and where we
386: have regularized the ultraviolet divergence appearing momentum integrals
387: arising from unphysical properties of the contact interaction \cite{Fetter1971}.
388: The gap equation is an implicit equation for the gap profile $\Delta(\vec{r})$
389: \cite{Perali2003a} which for an isotropic trap is only a function
390: of the radial position $r=|\vec{r}|$. The number equation and the
391: density profiles are determined from the thermodynamic relation \begin{eqnarray*}
392: N_{\sigma} & = & -\frac{\partial\Omega}{\partial\mu_{\sigma}}=\int d^{3}\vec{r}\ n_{\sigma}(r),\end{eqnarray*}
393: where the radial distribution for the $\sigma$ component is \[
394: n_{\sigma}(r)=\frac{1}{\mathcal{V}}\sum_{\vec{k}}\left[u_{\vec{k}}^{2}n_{\mathrm{F}}(E_{\vec{k}\sigma}(r))+v_{\vec{k}}^{2}n_{\mathrm{F}}(-E_{\vec{k}\bar{\sigma}}(r))\right],\]
395: with $\bar{\sigma}$ denotes the other component of $\sigma.$ The
396: polarization $P$ is \begin{equation}
397: P=\frac{N_{\uparrow}-N_{\downarrow}}{N_{\uparrow}+N_{\downarrow}}=\frac{1}{N}\int d^{3}r\ \delta n(r),\end{equation}
398: where $\delta n(r)=n_{\uparrow}(r)-n_{\downarrow}(r).$ The equations
399: for the polarization $P$ and for the minority component $N_{\downarrow}/N=(1-P)/2$
400: are numerically solved by iteration together with the gap profile
401: as follows. We first make an initial guess for the values of the chemical
402: potentials $\mu$ and $\delta\mu$ at fixed coupling $k_{\mathrm{F}}a_{s}.$
403: The $r$ dependent gap $\Delta(r)$, minor density $n_{\downarrow}(r)$
404: and polarization density $\delta n(r)$ are then discretized on a
405: radial grid of sufficient resolution. From the initial values of $\mu$
406: and $\delta\mu$ and at fixed coupling the gap profile $\Delta(r)$
407: is calculated by solving the gap equation \prettyref{eq:gap} as a
408: root problem. The gap profile is then subsequently used to calculate
409: the densities $n_{\downarrow}(r)$ and $\delta n(r)$ as functions
410: of $\mu$ and $\delta\mu.$ The minor number and and polarization
411: equations are then solved as a multidimensional root problem for the
412: average chemical potential $\mu$ and the bare depairing width $\delta\mu.$
413: The system of equations are iterated until the gap profile and the
414: chemical potentials are sufficiently converged and then finally $n_{\uparrow}(r)=\delta n(r)+n_{\downarrow}(r).$
415: The present method applied here is well suited for considering effects
416: of the Hartree terms, but we have not included those in the results
417: presented here.
418:
419:
420: \subsection{Harmonic trap eigenstates \label{sub:Harmonic-trap-}}
421:
422: The Bogoliubov-de Gennes (BdG) approach allows treating the effect
423: of the harmonic trapping potential exactly. The approach has been
424: used for polarized Fermi gases \cite{Castorina2005,Mizushima2005a,Kinnunen2006a,Machida2006}
425: because it avoids some of the problems of a local density approximation
426: approach. We have used it in our previous work \cite{Kinnunen2006a}
427: and give in this section a detailed description of the calculations
428: on which the results in \cite{Kinnunen2006a} and in this article
429: are based. One particularly important advantage of this approach is
430: the proper description of interface effects in a phase separated gas,
431: following from the nonlocal nature of the solution and the wavefunctions.
432: Eventually, in the limit of a large number of atoms, the LDA and BdG
433: solutions are expected to become the same.
434:
435: The BdG approach used here is a generalization, to the imbalanced
436: densities case, of the calculation outlined in Ref. \cite{Ohashi2005a}
437: which was for an unpolarized Fermi gas. For the imbalanced case the
438: generalization amounts to introducing different chemical potentials
439: $\mu_{\downarrow}$ and $\mu_{\uparrow}$ for the two species of fermionic
440: atoms. We now expand the wavefunctions $\hat{\Psi}_{\sigma}(\vec{r})$
441: in the basis of the 3D harmonic oscillator eigenstates \begin{equation}
442: \hat{\Psi}_{\sigma}(\vec{r})=\sum_{nlm}R_{nl}(r)Y_{lm}(\hat{r})\hat{a}_{nlm\sigma},\end{equation}
443: where the quantum numbers $\{\eta\}\equiv\{ n,l,m\}$ are the radial
444: quantum number $n$ counting the nodes in the radial function, $l$
445: is the orbital angular momentum and $m$ is the projected angular
446: momentum onto the axis of quantization. Here $Y_{lm}(\hat{r})$ are
447: the spherical harmonics with $\hat{r}=(\theta,\varphi)$ and the radial
448: part of the wavefunction is \begin{equation}
449: R_{nl}(r)=\sqrt{2}\left(m\omega\right)^{3/4}\sqrt{\frac{n!}{\left(n+l+1/2\right)!}}e^{-\bar{r}^{2}/2}\bar{r}^{l}L_{n}^{l+1/2}\left(\bar{r}^{l}\right),\end{equation}
450: where $\bar{r}=r/a_{\mathrm{osc}}$ with the harmonic oscillator length
451: being $a_{\mathrm{osc}}=[\hbar/(m\omega)]^{1/2},$ and $L_{n}^{l+1/2}(\bar{r}^{2})$
452: is an associated Laguerre polynomial. The characteristic scale of
453: the cloud is given by the Thomas-Fermi radius $R_{\mathrm{TF}}^{2}=2E_{\mathrm{F}}/(m\omega^{2})$
454: and therefore $R_{\mathrm{TF}}=(24N)^{1/6}a_{\mathrm{osc}}$. The
455: spherical symmetry allows doing the angular integrations, and getting
456: rid of the $m$-quantum numbers, thereby making the $l$-states $(2l+1)$-fold
457: degenerate. The expansion yields the following Hamiltonian \begin{eqnarray}
458: H_{\mathrm{MF}} & = & \sum_{n,l,\sigma}(2l+1)\left(\varepsilon_{nl}-\mu_{\sigma}\right)a_{nl\sigma}^{\dagger}a_{nl\sigma}\nonumber \\
459: & + & U\sum_{n,n^{\prime},l,\sigma}J_{nn^{\prime}\bar{\sigma}}^{l}a_{nl\sigma}^{\dagger}a_{n^{\prime}l\sigma}\nonumber \\
460: & + & \sum_{n,n^{\prime},l}F_{nn^{\prime}}^{l}a_{nl\uparrow}^{\dagger}a_{n^{\prime}l\downarrow}^{\dagger}+\mathrm{H.c}.\end{eqnarray}
461: Here the single particle energies are $\varepsilon_{nl}=\hbar\omega\left(2n+l+3/2\right)$,
462: and the Hartree interaction is described by the elements \[
463: J_{nn^{\prime}\sigma}^{l}=\int_{0}^{\infty}dr\ r^{2}R_{nl}(r)n_{\sigma}(r)R_{n^{\prime}l}(r),\]
464: and the pairing field is described by \[
465: F_{nn^{\prime}}^{l}=\int_{0}^{\infty}dr\ r^{2}R_{nl}(r)\Delta(r)R_{n^{\prime}l}(r).\]
466: We note that the $\sigma$ dependence of the Hartree term is due to
467: the population imbalance which implies that the corrections are different
468: for the two components. The density of $\sigma$ atoms is \begin{equation}
469: n_{\sigma}(r)=\sum_{n,n^{\prime},l}\frac{2l+1}{4\pi}R_{nl}(r)R_{n^{\prime}l}(r)\langle a_{nl\sigma}^{\dagger}a_{n^{\prime}l\sigma}\rangle,\end{equation}
470: and the order parameter is \begin{equation}
471: \Delta(r)=\tilde{U}(r)\sum_{n,n^{\prime},l}\frac{2l+1}{4\pi}R_{nl}(r)R_{n^{\prime}l}(r)\langle a_{nl\uparrow}^{\dagger}a_{n^{\prime}l\downarrow}^{\dagger}\rangle.\end{equation}
472: The additional factor $(2l+1)$ comes from the degeneracy of the $l$-states,
473: obtained by the summation over the $m$ states. The renormalized interaction
474: strength $\tilde{U}(r)$ will be described later.
475:
476: Truncating the Hilbert space by keeping only the states with single-particle
477: energies $\varepsilon_{nl}\leq E_{\mathrm{c}}$ allows writing the
478: Hamiltonian in a matrix form that can be diagonalized. Noticing that
479: the Hamiltonian commutes for different $(l)$-quantum numbers makes
480: it possible to diagonalize each $(l)$-matrix separately, simplifying
481: the problem significantly. For a given $(l)$-quantum numbers the
482: Hamiltonian reads \begin{equation}
483: H_{\mathrm{MF}}^{(l)}=\left(\begin{array}{c}
484: a_{0l\uparrow}^{\dagger}\\
485: \ldots\\
486: a_{N_{l}l\uparrow}^{\dagger}\\
487: a_{0l\downarrow}\\
488: \ldots\\
489: a_{N_{l}l\downarrow}\end{array}\right)^{T}M^{l}\left(\begin{array}{c}
490: a_{0l\uparrow}\\
491: \ldots\\
492: a_{N_{l}l\uparrow}\\
493: a_{0l\downarrow}^{\dagger}\\
494: \ldots\\
495: a_{N_{l}l\downarrow}^{\dagger}\end{array}\right),\end{equation}
496: where $N_{l}=[E/(\hbar\omega)-l-3/2]/2$ is the $l$-specific cutoff
497: (yielding the correct energy cutoff) and $M^{l}$ is a $2(N_{l}+1)\times2(N_{l}+1)$-dimensional
498: orthogonal matrix. Notice that the $\downarrow$ states have been
499: turned into holes by switching the order of $a_{\downarrow}^{\dagger}$
500: and $a_{\downarrow}$ operators as suggested by the ordinary Bogoliubov
501: transformation. The matrix $M^{l}$ is now \begin{widetext} \begin{equation}
502: M^{l}=\left(\begin{array}{cccccc}
503: \varepsilon_{0l}-\mu_{\uparrow}+UJ_{00\downarrow}^{l} & \ldots & UJ_{{0N}_{l}\downarrow}^{l} & F_{00}^{l} & \ldots & F_{{0N}_{l}}^{l}\\
504: \ldots & \ldots & \ldots & \ldots & \ldots & \ldots\\
505: UJ_{N_{l}0\downarrow}^{l} & \ldots & \varepsilon_{N_{l}l}-\mu_{\uparrow}+UJ_{N_{l}N_{l}\downarrow}^{l} & F_{N_{l}0}^{l} & \ldots & F_{N_{l}N_{l}}^{l}\\
506: F_{00}^{l} & \ldots & F_{N_{l}0}^{l} & -\varepsilon_{0l}+\mu_{\downarrow}-UJ_{00\uparrow}^{l} & \ldots & -UJ_{N_{l}0\uparrow}^{l}\\
507: \ldots & \ldots & \ldots & \ldots & \ldots & \ldots\\
508: F_{0N_{l}}^{l} & \ldots & F_{N_{l}N_{l}}^{l} & -UJ_{0N_{l}\uparrow}^{l} & \ldots & -\varepsilon_{N_{l}l}+\mu_{\downarrow}-UJ_{N_{l}N_{l}\uparrow}^{l}\end{array}\right).\end{equation}
509: \end{widetext} Diagonalizing this Hamiltonian corresponds to the
510: Bogoliubov transformation, which yields the eigenenergies $E_{jl}$
511: and the corresponding quasiparticle eigenstates ($2(N_{l}+1)$-dimensional
512: vectors) $W_{jn}^{l}$. The indices $j,n$ both run from $0$ to
513: ${2N}_{l}+1.$ In the basis of these quasiparticle states, the Hamiltonian
514: is \begin{equation}
515: H_{MF}^{(l)}=\sum_{j=0}^{2N_{l}+1}E_{jl}\alpha_{jl}^{\dagger}\alpha_{jl}.\end{equation}
516: In this basis, the density of atoms in $\uparrow$ state is \begin{eqnarray}
517: n_{\uparrow}(r) & = & \sum_{l}\frac{2l+1}{4\pi}\sum_{j}\sum_{n,n^{\prime}=0}^{2N_{l}+1}R_{nl}(r)R_{n^{\prime}l}(r)\nonumber \\
518: & \times & W_{jn}^{l}W_{jn'}^{l}n_{\mathrm{F}}(E_{jl}),\end{eqnarray}
519: where the Fermi distribution $n_{\mathrm{F}}(E)=1/(1+e^{E/k_{\mathrm{B}}T})$.
520: Likewise, the density of atoms in $\downarrow$ state is \begin{equation}
521: \begin{split}n_{\downarrow}(r)=\sum_{l}\frac{2l+1}{4\pi} & \sum_{j}\sum_{n,n^{\prime}=0}^{2N_{l}+1}R_{nl}(r)R_{n^{\prime}l}(r)\\
522: & \times W_{j,n+N_{l}+1}^{l}W_{j,n^{\prime}+N_{l}+1}^{l}n_{\mathrm{F}}(-E_{jl}),\end{split}
523: \end{equation}
524: where the sign in the Fermi function is changed because of the $\downarrow$
525: component (the last $N_{l}+1$ components) of the eigenstates correspond
526: to holes as compared to the particles in the $\uparrow$ components
527: (the first $N_{l}+1$ components). The order parameter is given by\begin{eqnarray}
528: \Delta(r) & = & \sum_{l}\frac{2l+1}{4\pi}\sum_{j}\sum_{n,n^{\prime}=0}^{{2N}_{l}+1}R_{nl}(r)R_{n^{\prime}l}(r)\nonumber \\
529: & \times & W_{jn}^{l}W_{j{,n}^{\prime}+N_{l}+1}^{l}\left(1+2n_{\mathrm{F}}(E_{jl})\right).\label{eq:finalgap}\end{eqnarray}
530: The total number of atoms in the two components can be obtained by
531: integrating over the densities. However, numerically it is faster
532: to calculate them directly from the eigenstates using the equations
533: \begin{equation}
534: N_{\uparrow}=\sum_{l}(2l+1)\sum_{j}\sum_{n=0}^{2N_{l}+1}W_{jn}^{l}W_{jn}^{l}n_{\mathrm{F}}(E_{jl})\label{eq:finalnumberup}\end{equation}
535: and \begin{equation}
536: N_{\downarrow}=\sum_{l}(2l+1)\sum_{j}\sum_{n=0}^{2N_{l}+1}W_{j,n+N_{l}+1}^{l}W_{j,n+N_{l}+1}^{l}n_{\mathrm{F}}(-E_{jl}).\label{eq:finalnumberdown}\end{equation}
537: The gap and the number equations (\ref{eq:finalgap},\ref{eq:finalnumberup},\ref{eq:finalnumberdown})
538: are solved iteratively. We have made calculations where the Hartree
539: fields are included and compared them to the case where Hartree fields
540: are neglected by setting $J_{nn^{\prime}\sigma}^{l}=0$ for all $n,n^{\prime},l$
541: . Note that close to the Feshbach resonance, the Hartree fields become
542: formally infinite. In this extreme limit, we have limited the Hartree
543: interaction strength $U$ to $\left|\left(k_{\mathrm{F}}a_{s}\right)^{-1}\right|\leq\beta$,
544: where $\beta\approx0.5$. When comparing the results with and without
545: Hartree fields, there is no difference in the qualitative features
546: such as the order parameter oscillations and over-all shape of the
547: gap and density profiles, the Hartree fields cause only minor corrections
548: to gap and density profiles (effectively compressing the gas slightly).
549: Numerically, neglecting the Hartree fields gives a tremendous speedup
550: in the numerical solution because it decouples the density and gap
551: profiles. In the results presented in this article, the Hartree fields
552: are neglected. We have made the calculations at zero temperature and
553: present here only results at $T=0$. We have checked that the BdG
554: results do not change by using a finite but very small temperature
555: $T=0.001\ T_{\mathrm{F}}$.
556:
557: As the density and gap profiles are decoupled, it is straightforward
558: to solve the gap equation for given chemical potentials. However,
559: since we want to keep the number of atoms fixed, the total procedure
560: will require optimizing also the chemical potentials, so that the
561: number equations are satisfied. The subsequent iteration procedure
562: can be performed in several different ways. We have found that a very
563: efficient procedure is to solve the chemical potentials (by solving
564: the number equations (\ref{eq:finalnumberup},\ref{eq:finalnumberdown}))
565: for each trial gap profile $\Delta(r)$. The trial gap profile $\Delta(r)$
566: is then used for solving the new trial gap profile $\Delta^{\prime}(r)$
567: using the gap equation (\ref{eq:finalgap}) with the new obtained
568: chemical potential. The chemical potentials therefore keep changing
569: between the iteration steps of the gap profile. On the other hand,
570: the numbers of atoms stay fixed in the iteration process. %
571: \begin{figure}
572: \begin{centering}\includegraphics[width=7cm]{densityprofilesP500} \par\end{centering}
573:
574:
575: \caption{(color online) Gap and density profiles on the BCS side of the resonance
576: $(k_{\mathrm{F}}a_{s})^{-1}=-0.5$ for $P=0.50.$ \label{fig:densityprofilesP500}}
577: \end{figure}
578: The initial guesses for the profiles needed for the iteration procedure
579: are obtained by using the chemical potentials and the gap profile
580: obtained for a slightly lower polarization and, eventually, by the
581: solution obtained for unpolarized gas. Solution of the profiles for
582: a given polarization $P$ requires therefore solving the profiles
583: for all lower polarizations. The validity of the final solution $\Delta(r),\mu_{\uparrow},\mu_{\downarrow}$
584: has been checked for several values of polarization and interaction
585: strength by perturbing the final solution and using the perturbed
586: profiles $\Delta'(r),\mu_{\uparrow},\mu_{\downarrow}$ as initial
587: guesses for the profiles. Usually the disturbed profiles have converged
588: either into the final profiles $\Delta(r)$ or into the normal state
589: $\Delta(r)\equiv0$. Only in the regime where our calculations predict
590: the superfluid core polarization have our solutions sometimes strayed
591: into a superfluid solution with unpolarized core. In such cases, the
592: final solution has been picked by choosing the solution with the lower
593: total energy, given by \begin{equation}
594: \begin{split}E=\sum_{l}^{2N_{l}+1}(2l+1) & \left[\sum_{n}\left(\varepsilon_{nl}-\mu_{\downarrow}\right)+\sum_{j}E_{jl}n_{\mathrm{F}}(E_{jl})\right]\\
595: & -\frac{1}{\tilde{U}(r)}\int dr\ r^{2}|\Delta(r)|^{2},\end{split}
596: \label{eq:bdg-free-energy}\end{equation}
597: where the first (constant) term comes from switching to use the hole
598: states in the $\downarrow$ component (and this is countered by having
599: half of the eigenenergies $E_{jl}$ negative). The last term follows
600: from the mean-field term $\langle\hat{\Psi}_{\uparrow}^{\dagger}(\vec{r})\hat{\Psi}_{\downarrow}^{\dagger}(\vec{r})\rangle\langle\hat{\Psi}_{\downarrow}(\vec{r})\hat{\Psi}_{\uparrow}(\vec{r})\rangle$
601: which was dropped from the mean-field Hamiltonian because it is a
602: number, not operator.
603:
604: The summation in the gap equation (\ref{eq:finalgap}) is divergent
605: without the energy cutoff $E_{\mathrm{c}}.$ This is a well known
606: phenomenon, following from the incapability of the contact interaction
607: potential to describe properly high energy behavior. Several different
608: regularization schemes have been proposed \cite{Bruun1999a,Bulgac2002,Grasso2003},
609: and here we apply the one suggested in Ref. \cite{Grasso2003}. This
610: implies the following form for the renormalized coupling \[
611: \frac{1}{\tilde{U}(r)}=\frac{1}{U}+\frac{1}{2\pi^{2}}\left(\frac{k_{\mathrm{F}}^{0}(r)}{2}\ln\left(\frac{k_{c}(r)+k_{\mathrm{F}}^{0}(0)}{k_{c}(r)-k_{\mathrm{F}}^{0}(0)}\right)-k_{c}(r)\right),\]
612: where the momentum cutoff $k_{c}(r)=(2N_{c}+3-r^{2})^{1/2}$ and the
613: local Fermi momentum $k_{\mathrm{F}}^{0}(r)=(\mu_{\uparrow}+\mu_{\downarrow}-r^{2})^{1/2}$
614: for the imbalanced case compared to Eq. (14) and Eq. (18) of Ref.
615: \cite{Grasso2003}, respectively. On the BCS side of the resonance
616: for $\left(k_{\mathrm{F}}a_{s}\right)^{-1}<0$ we have used as the
617: cutoff $E_{\mathrm{c}}=200\ \hbar\omega$ and on the BEC side for
618: $\left(k_{\mathrm{F}}a_{s}\right)^{-1}>0$ the cutoff $E_{\mathrm{c}}=240\ \hbar\omega$
619: was used. We have tested the remaining cutoff dependence by using
620: the cutoff $300\ \hbar\omega.$ Depending on the number of atoms,
621: the cutoff dependence in the gap profiles was at most $2\%.$
622:
623:
624:
625:
626: \section{Results }
627:
628: The results show two features that we will discuss in detail below:
629: 1) For small and intermediate polarizations, FFLO-type oscillations
630: in the superfluid-normal interface at the edge of the trap, 2) for
631: large polarization, a polarized superfluid that extends through the
632: whole trapped gas. We study the behavior of these features, especially
633: 1), throughout the BCS-BEC crossover and when the system size, i.e.
634: the atom number, is varied. We also calculate the condensate fraction
635: to make a connection to experiments. Finally, we analyze which harmonic
636: oscillator states are involved in pairing for different polarizations
637: and discuss the connection and differences to FFLO-state in a homogeneous
638: space. %
639: \begin{figure}
640: \begin{centering}\includegraphics[width=7cm]{resprofiles} \par\end{centering}
641:
642:
643: \caption{(color online) Typical profiles at resonance $(k_{\mathrm{F}}a_{s})^{-1}=0.0$
644: for $P=0.9.$\label{fig:res-profiles} }
645: \end{figure}
646:
647:
648:
649:
650:
651: \subsection{FFLO-type oscillations}
652:
653: Typical density and gap profiles at $T=0$, for small polarization,
654: are shown in Figs. \ref{fig:comp-zero-T} and \ref{fig:dens-comp-zero-T},
655: both for LDA and BdG. Comparison of the profiles gives the following
656: general picture: BdG predicts a) SF (equal densities) core, b) PS
657: (FFLO-like oscillations) shell, c) normal state shell of the majority
658: component $N_{\uparrow}.$ LDA predicts a) SF (equal densities) core,
659: b) normal state shell; the absence of the PS shell in LDA is reflected
660: in the discontinuity of the density and gap profiles at the SF-N
661: phase boundary. At finite temperatures, also LDA shows a polarized
662: shell and the boundary becomes continuous, only showing a kink (Fig.
663: \ref{fig:comp-finite-T}). As shown by Figs. \ref{fig:comp-zero-T}-\ref{fig:comp-finite-T},
664: for small polarization the PS given by BdG calculations is a narrow
665: shell and can be understood as a boundary effect. However, in the
666: following we show that, for large polarization, the FFLO-features
667: extend to the center of the trap as well.
668:
669: A seen from Fig. \ref{fig:comp-zero-T} and Fig. \ref{fig:comp-finite-T}
670: the general results from LDA and BdG calculation agree fairly well
671: and the overall agreement for the gap and density profiles becomes
672: better for increasing particle number. The incapability of LDA to
673: correctly describe the short scale behavior (explicitly excluded by
674: LDA) leads to the unphysical appearance of a discontinous order parameter
675: solution and can be viewed as a breakdown of LDA near the interface
676: much in the same way as LDA generally breaks near the edge of a balanced
677: condensate. The breakdown of LDA is however more pronounced in the
678: imbalanced case due to presence of the normal shell surrounding the
679: condensate which leads to separation of the size of the condensate
680: and the size of the cloud. %
681: \begin{figure}
682: \begin{centering}\includegraphics[width=7cm]{BECprofiles} \par\end{centering}
683:
684:
685: \caption{(color online) Gap and density profiles on the BEC side of the resonance
686: $(k_{\mathrm{F}}a_{s})^{-1}=0.5$ for P=0.95. \label{fig:bec-profiles}}
687: \end{figure}
688:
689:
690:
691: \subsubsection{Dependence on the interaction strength -- BCS-BEC crossover}
692:
693: We present here results for three different cases: (1) BCS side of
694: the crossover, $(k_{\mathrm{F}}a_{s})^{-1}=-0.50$, (2) unitarity,
695: that is, $(k_{\mathrm{F}}a_{s})^{-1}=0$, (3) BEC side $(k_{\mathrm{F}}a_{s})^{-1}=0.50$.
696: We focus on the behavior at strong interactions since the cases (1)-(3)
697: can all be considered to be at the unitarity regime (if it is defined
698: $|(k_{\mathrm{F}}a_{s})^{-1}|<1$ ) although representing different
699: sides of the crossover. We do not consider the extreme BCS limit of
700: weak interactions because the features and trends observed in the
701: unitarity limit are also expected to appear at weaker coupling.
702:
703: For attractive interactions, $(k_{\mathrm{F}}a)^{-1}<0$, the typical
704: density and order parameter profiles looks as shown in Fig. \ref{fig:densityprofilesP500}
705: for polarization $P=0.50$. In agreement with earlier studies using
706: the same approach \cite{Castorina2005,Mizushima2005a,Kinnunen2006a,Machida2006},
707: the solution reveals an unpolarized BCS-type region at the center
708: of the trap and a polarized shell with oscillating order parameter.
709: The oscillations rapidly dampen when the density of the minority component
710: drops.
711:
712: We have studied the presence of the oscillations around the unitarity
713: region, and noticed that the critical polarization for the appearance
714: of the order parameter oscillations increases with stronger interactions.
715: At unitarity ($(k_{\mathrm{F}}a)^{-1}=0.0$), the order parameter
716: oscillations did not appear until polarization $P=0.70$, compared
717: to $P=0.10$ at $(k_{\mathrm{F}}a)^{-1}=-0.50$. The disappearance
718: of the oscillations follows from the enhanced stability of the BCS-type
719: pairing at stronger interactions, making the increased distortion
720: of the minority component density profile favorable over the reduced
721: order parameter due to the polarization. The result at unitarity is
722: shown in Fig. \ref{fig:res-profiles}.
723:
724: On the BEC side, the oscillations do not appear. We have made calculations
725: for several parameters on the BEC side of the resonance and have not
726: found any FFLO-type oscillations in the order parameter on the BEC
727: side. This is consistent with the observation discussed above that
728: the critical polarization for the emergence of the nodes in the order
729: parameter increases with increasing interaction strength. The same
730: qualitative results on the BCS-BEC crossover were recently obtained
731: in \cite{Mizushima2007} using a hybrid BdG-LDA scheme in a cylindrically
732: symmetric geometry. Typical density and gap profiles on the BEC side
733: are shown in Fig. \ref{fig:bec-profiles}. %
734: \begin{figure}
735: \begin{centering}\includegraphics[width=7cm]{gapsP740} \par\end{centering}
736:
737:
738: \caption{(color online) Gap profiles at $(k_{\mathrm{F}}a_{s})^{-1}=-0.5$
739: for several values $N=9000,18000,100000$ of the total number of particles
740: at a high and fixed polarization $P=0.74.$ For increasing particle
741: number the transition from an almost constant value of the gap in
742: the BCS core to the oscillating gap at the edge becomes sharper. For
743: increasing particle number the gap FFLO-like oscillations at the edge
744: the cloud display shorter wavelength and a slight increase in amplitude.
745: \label{fig:gapsP740}}
746: \end{figure}
747:
748:
749:
750: \subsubsection{Dependence on the system size - atom number N}
751:
752: %
753: \begin{figure}
754: \begin{centering}\includegraphics[width=7cm]{expfitfortail} \par\end{centering}
755:
756:
757: \caption{(color online) Exponential fits for the envelopes of the order parameter
758: oscillations for 9000 and 100000 atoms. \label{fig:expfitfortail}}
759: \end{figure}
760: We have studied the dependence of the results on the number of atoms
761: by performing the BdG calculations for the total atom numbers 9000,
762: 10000, 18000, 20000 and 100000. Note that this means substantial increase
763: in the system size compared to our earlier work \cite{Kinnunen2006a}
764: where the atom number 10000 was used.
765:
766: The order parameter oscillations for different numbers of atoms are
767: shown in Fig. \ref{fig:gapsP740}. In Fermi units (i.e. in the cloud
768: scale $R_{\mathrm{TF}}$), the wavelength of the oscillations becomes
769: shorter with the increased atom number but, on the other hand, more
770: nodes appear. The scaling of all these factors is complicated, but
771: insight can be obtained by fitting an exponential function $\sim e^{-x/\xi}$
772: to the envelope of the order parameter oscillations as shown in Fig.
773: \ref{fig:expfitfortail}. The exponential decay gives a very good
774: description for the damping of the oscillations with increasing distance
775: from the trap center $r$. The penetration depths $\xi$ obtained
776: from the fitting indeed slowly decrease in Fermi units when the atom
777: number $N$ is increased, the scaling being roughly $N^{1/6}$. Since
778: $R_{\mathrm{TF}}\propto N^{1/6}$, one may anticipate that the decrease
779: of the penetration depth only occurs in relative scale, not absolute.
780: Indeed, in a microscopic length scale (harmonic oscillator units $a_{\mathrm{osc}}$)
781: the penetration depth remains constant, being $0.54\ a_{\mathrm{osc}}$
782: for 100000 atoms and $0.56\ a_{\mathrm{osc}}$ for 9000 atoms. These
783: observations confirm that the order parameter oscillations are not
784: a finite size effect but rather an interface effect for the superfluid-normal
785: interface. From the shape of the order parameter profile it can be
786: seen that that the interface formed becomes sharper for increasing
787: particle number and we therefore suggest that the interface forming
788: at the edge is an analog of the proximity effects appearing in superfluid-normal
789: and superfluid-ferromagnetic junctions. In the latter case it has
790: been shown that the interplay between BCS superconductivity and ferromagnetic
791: order (in our case the polarization $\delta n$) gives rise to an
792: oscillating order parameter which decays anomalously slow (over several
793: oscillations) on the ferromagnet side and with a characteristic length
794: scale that is independent of the properties of the superconductor
795: \cite{Demler1997,Kontos2001,Schaeybroeck2007}. An interesting question
796: is the dependence of the penetration depth on the interaction strength
797: and polarization. However, we have not been able to pursue this
798: question in more detail here and leave it for future work.
799:
800: %
801: \begin{figure}
802: \includegraphics[width=7cm]{densityprofilesP762}
803:
804:
805: \caption{Density and order parameter profiles at $P=0.762$ and $(k_{\mathrm{F}}a_{s})^{-1}=-0.50$
806: for $N=9000$ show the superfluid with core polarization, together with
807: a small dip in the density difference at the center of the trap. }
808:
809: \label{fig:densityprofilesP762}
810: \end{figure}
811: In this context, we believe a few remarks on the work \cite{Liu2007}
812: are in order. It is argued in \cite{Liu2007}, based on a hybrid BdG-LDA
813: scheme (no analysis of the penetration depths as presented above was
814: done in \cite{Liu2007}), that the oscillations of the order parameter
815: vanish for sufficiently large particle number and are thus interpreted
816: as a finite size effect. It is argued that as the cutoff energy $E_{c}$
817: introduced to separate the BdG and the LDA scales in the hybrid scheme
818: goes to zero, LDA should be recovered. This is of course self-evident
819: due to the construction of the hybrid algorithm but reducing the cutoff
820: energy without concern may lead to missing some features of the system.
821: The fact that a sharp interface with increasingly microscopic features
822: (oscillations with short period on the scale of the cloud size) builds
823: up for increasing number of particles requires $E_{c}$ to be increased
824: significantly in order to resolve the short length scale features.
825: Consistent with our results, Fig. 4 in \cite{Liu2007} shows that
826: the order parameter oscillates with a shorter period, and at the same
827: time the amplitude of the order parameter slowly increases, for increasing
828: particle number. The recent results for a hybrid BdG-LDA scheme \cite{Mizushima2007}
829: and the scaling analysis therein agree well with the results given
830: by our full BdG calculations.
831:
832:
833: \subsection{Polarized superfluid for large imbalances}
834:
835: When the polarization $P$ exceeds some (large) critical value, the
836: FFLO-type oscillations discussed above reach the center of the trap.
837: At this point the gas becomes polarized also at the center of the
838: trap and the order parameter drops to roughly half of its unpolarized
839: BCS value. This realizes a superfluid with finite polarization throughout
840: the system. Fig. \ref{fig:densityprofilesP762} shows the density
841: and order parameter profiles for $P=0.762$. For such a core polarized
842: system, the concept of an interface effect is intriguing as there
843: is no BCS-type superfluid core present.
844:
845: The density difference for such a polarized superfluid shows an interesting
846: feature: a small dip in the center of the trap (i.e. the density difference
847: is smaller than in the surrounding area, but still non-zero). At zero
848: temperature, this feature only appears in our BdG calculations whereas
849: LDA, which fails to predict a polarized superfluid at $T=0$, does
850: not lead to such a dip in the density difference. In contrast, LDA
851: calculations at finite temperature produce such a feature in connection
852: with the finite temperature BP phase. Recent Monte Carlo studies of
853: the trapped Fermi gas have shown that, for the strongly interacting
854: normal state, the density difference increases monotonously towards
855: the center of the trap \cite{Lobo2006}. Therefore one may argue that
856: the dip is associated with a polarized superfluid: at $T=0$ it is
857: FFLO-type, at temperatures that are finite but clearly below the critical
858: temperature it is either FFLO or BP. At higher temperatures, pseudogap
859: effects may contribute to such features as well \cite{Chien2006b}.
860: Such a dip can be seen in the experimental results of \cite{Shin2006a}.
861:
862: The interaction strength dependence of such a polarized superfluid
863: with oscillating order parameter is similar to what was already discussed
864: above. Since there are no order parameter oscillations on the BEC
865: side, we have also not seen any core polarized superfluid either.
866: Of course, there does exist a different kind of core polarization
867: on the BEC side: coexistence of a molecular condensate and free excess
868: fermions, but that is not associated with oscillating order parameter.
869:
870: The parameter window for the polarized superfluid shrinks with increasing
871: atom number $N$. For $9000$ atoms the window is $0.746<P<0.784,$
872: whereas for $18000$ atoms it is $0.746<P<0.774.$ Since the convergence
873: of the calculations near this critical window is slow, we have not
874: been able to determine the corresponding window for $100000$ and
875: have not systematically analyzed how the window scales with particle
876: number. For the order parameter oscillations, we have shown in section
877: III.A.2. that their absolute length scale stays unchanged for large
878: particle numbers. Based on the present data, we cannot conclude with
879: similar confidence whether the parameter window for the existence
880: of the core polarized superfluid becomes negligible or not for large
881: condensates. However, we would like to emphasize that the core polarized
882: superfluid is not due to a trivial finite size effect, i.e. not originating
883: from having discrete oscillator states in the system description.
884: As seen in Refs. \cite{Grasso2003,Machida2006}, such finite size
885: effects manifest as a narrow dip in the density and order parameter
886: profiles at the center of the trap. However, these effects vanish
887: when the number of atoms increases or when the interaction strength
888: is increased. Because of the stronger interactions, we do not see
889: these finite size effects even at $9000$ atoms. Note that such a
890: dip originating from finite size effects is completely different from
891: the dip in the density difference discussed above as a signature of
892: the core polarized superfluid.
893:
894:
895: \subsection{Condensate fraction}
896:
897: %
898: \begin{figure}
899: \includegraphics[width=7cm]{condfracN9000}
900:
901:
902: \caption{The condensate fraction, density difference, and the order parameter
903: as a function of polarization $P$ obtained for $9000$ atoms. The
904: interaction strength is $(k_{\mathrm{F}}a)^{-1}=-0.50$.}
905:
906: \label{fig:condfracN9000}
907: \end{figure}
908: To make connection to experiments \cite{Shin2006a} where condensate
909: fractions are measured, we calculate here the condensate fraction
910: for an imbalanced gas. In the case of a balanced Fermi gas the condensate
911: fraction is defined in \cite{Campbell1997,Salasnich2005} and can
912: be viewed as a measure of the number of condensed pairs which in the
913: extreme BEC limit and at zero temperature is just $N/2\equiv N_{\sigma}.$
914: For the imbalanced gas, the corresponding number of molecules in
915: the asymptotic BEC limit is the number of minority atoms $N_{\downarrow}.$
916: It is therefore natural to consider the following normalized condensate
917: fraction \begin{equation}
918: N_{0}/N_{\downarrow}=\frac{1}{N_{\downarrow}}\int d^{3}\vec{r}_{1}d^{3}\vec{r}_{2}\left|\langle\Psi_{\uparrow}(\vec{r}_{1})\Psi_{\downarrow}(\vec{r}_{2})\rangle\right|^{2},\label{eq:imbalance-cf}\end{equation}
919: for the condensate fraction in the case of $N_{\downarrow}<N_{\uparrow}.$
920: Figs. \ref{fig:condfracN9000} and \ref{fig:condfracN18000} show
921: the density difference, the order parameter at the center of the trap,
922: and the imbalanced condensate fraction from Eq. \prettyref{eq:imbalance-cf}
923: as function of the polarization for $(k_{\mathrm{F}}a_{s})^{-1}=-0.50$.
924: The critical polarization is roughly $P_{c}=0.78$ and the transition
925: is reflected in a rapid increase in the density difference in the
926: center of the trap. Also the order parameter in the center of the
927: trap drops rapidly. However, as the figures show, there exists a narrow
928: but finite polarization region (in this case for $0.75<P<0.78$) where
929: both the density difference and the gap have non-negligible values
930: in the center of the trap. All data points are for fully converged
931: order parameter and density profiles, satisfying the gap equation
932: with precision of $10^{-6}$. In addition, several points in the plot
933: have been calculated with different initial conditions for the iteration.
934:
935: The figures are in good qualitative agreement with the experimental
936: condensate fractions \cite{Shin2006a}, showing the sudden onset of
937: the density difference at the center of the trap when the condensate
938: fraction drops to zero. The core polarized superfluid manifests itself
939: as a weak revival of the condensate fraction as shown in Fig. \ref{fig:condfracN9000detail}.
940: In other words, a finite condensate fraction co-exists with a finite
941: central density difference. Although the qualitative agreement with
942: the experimental results in \cite{Shin2006a} is good, higher experimental
943: accuracy as well as extending our calculations to finite temperatures
944: is needed for quantitative comparison, especially regarding the small
945: window for the core polarized superfluid. Since the condensate fraction
946: is small in the interesting transition region, better signatures of
947: the polarized FFLO-type superfluid may be provided by the central
948: gap, and by the dip in the central density difference as discussed
949: above, provided that temperature dependence is also carefully investigated.
950: %
951: \begin{figure}
952: \includegraphics[width=7cm]{condfracN18000}
953:
954:
955: \caption{The condensate fraction, density difference, and the order parameter
956: as a function of polarization $P$ obtained for $18000$ atoms. The
957: interaction strength is $(k_{\mathrm{F}}a)^{-1}=-0.50$.}
958:
959: \label{fig:condfracN18000}
960: \end{figure}
961:
962:
963:
964: \subsection{Contribution of different harmonic oscillator states in pairing}
965:
966: We studied the origin of the oscillations in the BdG order parameter
967: by considering which trap shells (quantum number $n$) are involved
968: in pairing in the center of the trap, i.e.\ for quantum number $l=0$.
969: As shown in Fig. \ref{fig:pairingatcenter}, for zero polarization
970: $P=0$, the pairing involves 0-2 neighboring shells, i.e.\ is peaked
971: at $n-n'=0$, with considerable weight until $n-n'=2$ due to strong
972: interactions. For small polarization $P=0.1$ the peak is shifted,
973: but the weight is still very much on the same $n-n'$ as for $P=0$,
974: which results to only minor modification of the order parameter profile.
975: However, for large polarization $P\simeq0.75$, when the oscillations
976: of the order parameter appear also in the center of the trap, the
977: pairing is peaked at nonzero $n-n'\sim11$. Therefore, for large polarizations
978: the state clearly resembles the FFLO state where pairing is predominantly
979: between momentum states $k-k'$ determined by the mismatch of the
980: Fermi surfaces (in our example $P\simeq0.75$ means a mismatch of
981: the Fermi surfaces of about $n-n'\sim11$). We have found that this
982: behavior is not due to small particle number, in contrast, it becomes
983: more clear for larger particle numbers.
984:
985:
986: \section{Conclusions and discussion}
987:
988: %
989: \begin{figure}
990: \includegraphics[width=7cm]{condfracN9000detail}
991:
992:
993: \caption{The core polarized superfluid region for $9000$ atoms $(k_{\mathrm{F}}a_{s})^{-1}=-0.50.$}
994:
995: \label{fig:condfracN9000detail}
996: \end{figure}
997: In summary, we have shown that FFLO-type oscillations appear in a
998: trapped polarized Fermi-gas and, for large polarization, extend to
999: the center of the trap thereby realizing a non-BCS superfluid at zero
1000: temperature. These results were obtained by BdG calculations using
1001: harmonic oscillator eigenstates and particle numbers up to 100000.
1002: We have also made a comparison to the results given by local density
1003: approximation.
1004:
1005: The FFLO-type oscillations of the order parameter appear on the BCS
1006: side of the BCS-BEC crossover. When the interaction strength is increased,
1007: the polarization required for the existence of the oscillations grows.
1008: On the BEC side, no oscillating order parameter was found. The scaling
1009: analysis for the penetration depth of the oscillations shows that
1010: their characteristic length scale stays constant when the particle
1011: number is increased. Therefore the characteristic length scale relative
1012: to the atom cloud scale (given by $R_{\mathrm{TF}}$) decreases very
1013: slowly, $\propto N^{1/6}$. The features presented here should thus
1014: be observable even for condensate sizes corresponding to present-day
1015: experiments. RF-spectroscopy was proposed in our earlier work \cite{Kinnunen2006a}
1016: for detecting the gapless excitations related to the nodes of the
1017: order parameter as a signature of the FFLO-type oscillations. The
1018: first experiments using RF-spectroscopy in investigating imbalanced
1019: gases were recently done \cite{Schunck2007}. Moreover, the dip in
1020: the central density difference could provide a signature of the core
1021: polarized superfluid, as well as the simultaneous measurement of the
1022: gap and the density difference in the center of the trap.
1023:
1024: The results presented here have an interesting connection to superconductor-ferromagnet
1025: interface effects. For future work, it is fascinating to think about
1026: the freedom that the ultracold gases offer in terms of designable
1027: trapping geometries and other parameters: one should be able to systematically
1028: study this kind of effects from the limit of having large superfluid
1029: and polarized normal state (``ferromagnet'') regions and a small interface,
1030: to the limit where the interface, superfluid, and normal state are
1031: all of comparable size and novel mesoscopic phenomena may be found.
1032: %
1033: \begin{figure}
1034: \begin{centering}\includegraphics[width=7cm]{pairingatcenternew}\par\end{centering}
1035:
1036:
1037: \caption{(color online) The gap at the center of the trap (the part of the
1038: gap for the quantum number $l=0$ ) as function of the difference
1039: in radial quantum number $\Delta n$ for different values of the polarization
1040: and for $N=18000.$ For increasing polarization the mainly intra-shell
1041: and nearest-neighbor-shell pairing indicated by the central peak (at
1042: $P=0$ ) diminishes and a secondary peak appears for large $\Delta n$
1043: which indicate the increasing importance of inter-shell pairing between
1044: shell states having an energy separation of order determined by the
1045: mismatch of Fermi energies. \label{fig:pairingatcenter} }
1046: \end{figure}
1047:
1048:
1049: \begin{acknowledgments}
1050: We thank W. Ketterle and M. Zwierlein for useful discussions. This
1051: work was supported by Academy of Finland (project numbers 213362,
1052: 106299, 205470) and conducted as part of a EURYI scheme award. See
1053: www.esf.org/euryi. J.K. acknowledges the support of the Department
1054: of Energy, Office of Basic Energy Sciences via the Chemical Sciences,
1055: Geosciences, and Biosciences Division.
1056: \end{acknowledgments}
1057: \bibliographystyle{apsrev} \bibliographystyle{apsrev} \bibliographystyle{apsrev}
1058: \bibliography{pra}
1059:
1060: \end{document}
1061: