1: \tolerance = 10000
2: \documentclass[prb,eqsecnum,twocolumn,floatfix,showpacs]{revtex4}
3: %\usepackage{showkeys}
4: \usepackage{graphicx}% Include figure files
5:
6: % Borrowed from FeynMP to include labels on Feynman diagrams
7: \def\fmfL(#1,#2,#3)#4{\put(#1,#2){\makebox(0,0)[#3]{#4}}}
8:
9: \hyphenation{ferm-ion-ic Fesh-bach Lut-ting-er}
10:
11: % Definitions:
12: \newcommand{\Ham}{\mathcal{H}}
13: \newcommand{\Act}{\mathcal{S}}
14:
15: \newcommand{\Mean}[1]{\left\langle#1\right\rangle}
16: \newcommand{\sub}[1]{_{\mathrm{#1}}}
17: \newcommand{\Dim}[1]{\dim [#1]}
18: \newcommand{\units}[1]{\ensuremath{\,\mathrm{#1}}}
19:
20: \newcommand{\be}{\begin{equation}}
21: \newcommand{\ee}{\end{equation}}
22: \newcommand{\bea}{\begin{eqnarray}}
23: \newcommand{\eea}{\end{eqnarray}}
24: \newcommand{\punc}[1]{\;\mathrm{#1}}
25:
26: \newcommand{\viceversa}{{\it vice versa}}
27: \newcommand{\ie}{{\it i.e.}}
28:
29: \newcommand{\mathsmall}{\textstyle}
30: \newcommand{\smallfrac}[2]{{\mathsmall\frac{#1}{#2}}}
31: \newcommand{\half}{\smallfrac{1}{2}}
32:
33: \newcommand{\Order}[1]{\mathcal{O}\left(#1\right)}
34: \newcommand{\parder}[2]{\frac{\partial #1}{\partial #2}}
35: \newcommand{\parderat}[3]{{\left(\frac{\partial #1}{\partial #2}\right)}_{#3}}
36:
37: \newcommand{\Gt}{\widetilde{G}} % Effective propagators for b and \psi
38: \newcommand{\Xit}{\widetilde{\Xi}}
39: \newcommand{\xit}{\tilde{\xi}}
40: \newcommand{\DeltaLDA}{\widetilde{\Delta}}
41: \newcommand{\lamt}{\tilde{\lambda}}
42: \newcommand{\LW}{\Phi\sub{LW}}
43: % End of definitions
44:
45: \begin{document}
46: \setlength{\unitlength}{1mm}
47:
48: \title{Depletion of the Bose-Einstein condensate in Bose-Fermi mixtures}
49:
50: \author{Stephen Powell}
51: \author{Subir Sachdev}
52: \affiliation{Department of Physics, Yale University, P.O. Box
53: 208120, New Haven, CT 06520-8120}
54: \author{Hans Peter B\"uchler}
55: \affiliation{Institut f\"ur Theoretische Physik, Universit\"at Innsbruck,
56: Technikerstra\ss e 25, A-6020 Innsbruck, Austria}
57:
58: \begin{abstract}
59: We describe the properties of a mixture of fermionic and bosonic
60: atoms, as they are tuned across a Feshbach resonance associated
61: with a fermionic molecular state. Provided the number of fermionic
62: atoms exceeds the number of bosonic atoms, we argue that there is
63: a critical detuning at which the Bose-Einstein condensate (BEC) is
64: completely depleted. The phases on either side of this quantum
65: phase transition can also be distinguished by the distinct
66: Luttinger constraints on their Fermi surfaces. In both phases, the
67: total volume enclosed by all Fermi surfaces is constrained by the
68: total number of fermions. However, in the phase without the BEC,
69: which has two Fermi surfaces, there is a {\em second\/} Luttinger
70: constraint: the volume enclosed by one of the Fermi surfaces is
71: constrained by the total number of {\em bosons}, so that the
72: volumes enclosed by the two Fermi surfaces are separately
73: conserved. The phase with the BEC may have one or two Fermi
74: surfaces, but only their total volume is conserved. We obtain the
75: phase diagram as a function of atomic parameters and temperature,
76: and describe critical fluctuations in the vicinity of all
77: transitions. We make quantitative predictions valid for the case
78: of a narrow Feshbach resonance, but we expect the qualitative
79: features we describe to be more generally applicable.
80: As an aside, we point out intriguing connections between the BEC
81: depletion transition and the transition to the fractionalized Fermi
82: liquid in Kondo lattice models.
83: \end{abstract}
84: \pacs{03.75.Hh, 71.35.Lk, 51.30.+i, 64.60.-i}
85:
86: \vspace{3cm}
87:
88:
89: \maketitle
90: \section{Introduction}
91: \label{sec:intro}
92:
93: The Feshbach resonance has emerged as a powerful tool in studying
94: ultracold atoms in regimes of strong interactions. For two
95: isolated atoms scattering off each other, the Feshbach resonance
96: is a singularity in their scattering length due to the coupling
97: of the atomic states to a molecular bound state.
98: \cite{stwalley1976,tiesinga1993} The singularity (at $\nu = 0$)
99: occurs as a function of the detuning $\nu$, which is a measure of
100: the energy difference between the atomic and molecular states. The
101: value of $\nu$ can be varied by an applied magnetic field, and
102: this effectively allows one to tune the strength of the atomic
103: interactions.
104:
105: For systems in the thermodynamic limit, with a finite density of
106: atoms, there is no singularity at $\nu=0$. Nevertheless, the
107: vicinity of $\nu=0$ is a regime of interesting many body effects.
108: For a Feshbach resonance between two identical fermionic atoms,
109: the many body ground state changes from a Bose-Einstein condensate
110: (BEC) of molecules ($\nu \ll 0$) to a Bardeen-Cooper-Schrieffer
111: (BCS) superfluid descended from a Fermi gas of atoms ($\nu \gg
112: 0$). It is important to note that there is no true fundamental
113: distinction between the BEC and BCS states here, and so the two
114: limits are connected by a smooth crossover. Recent experiments
115: \cite{jochim2003,greiner2003,zwierlein2003} on $^{6}$Li and
116: $^{40}$K atoms have succeeded in observing the BEC of molecules.
117:
118: The consequences of the two-body Feshbach resonance are very
119: different for other atomic statistics. For a Feshbach resonance
120: between two identical bosonic atoms, it has been argued recently
121: \cite{leo,ss} that there is indeed a sharp singularity, \ie\ a
122: quantum phase transition, in the many body system as a
123: function of $\nu$. This singularity is not precisely at $\nu=0$,
124: but is shifted away from it; it is not directly a reflection of
125: the singularity in the scattering length of two isolated atoms,
126: but is a new many body effect. Here, the two limiting states are a
127: BEC of molecules ($\nu \ll 0$) and a BEC of atoms ($\nu \gg 0$).
128: Unlike the fermionic case above, these two states cannot be
129: connected smoothly to each other. The fundamental distinction
130: between these states becomes apparent upon examining the quantum
131: numbers of the vortices in the condensate: the quantum of
132: circulation differs by a factor of 2 in the two limits, being
133: determined respectively by the mass of a molecule or of an atom.
134:
135: In the present paper, we will consider the remaining case of a
136: mixture of two distinct types of atoms, one fermionic and the
137: other bosonic. Mixtures of fermionic $^{6}$Li and bosonic $^{7}$Li
138: atoms were studied by Truscott {\em et al.} \cite{truscott} and
139: Schreck {\em et al.},\cite{schreck} and they succeeded in
140: achieving simultaneous quantum degeneracy in both species of
141: atoms. Recently, Feshbach resonances have been observed between
142: bosonic $^{23}$Na and fermionic $^{6}$Li atoms by Stan {\em et
143: al.}, \cite{stan} and between bosonic $^{87}$Rb and fermionic
144: $^{40}$K atoms by Inouye {\em et al.} \cite{inouye} So the time
145: is clearly appropriate to examine the many body properties of such
146: mixed Bose and Fermi gases across a Feshbach resonance, in which
147: the fermionic molecule of the two unlike atoms can also reach
148: quantum degeneracy.
149:
150: Our primary result is that such a mixture of fermionic and bosonic
151: atoms also has a quantum phase transition. Again this transition
152: is a many body effect, and does not occur precisely at $\nu=0$. We
153: will map out the phase diagram as functions of $\nu$, temperature
154: ($T$), and the densities of the atoms (see Figs.~\ref{Graph0},
155: \ref{Phases}, \ref{ZeroTphases}, \ref{ZeroTmu}) and also describe
156: the strong consequences of thermal and quantum fluctuations in
157: the vicinity of the phase transition. It is also possible that
158: the mixture phase separates: this will be studied in the Appendix,
159: where we determine the region of instability to phase separation
160: (see Fig.~\ref{ZeroTstability}).
161:
162: As in the Fermi-Fermi and Bose-Bose cases, the existence of a
163: quantum phase transition for the Bose-Fermi case can be easily
164: understood by characterizing the two limiting cases. For $\nu \gg
165: 0$, there is a BEC of the bosonic atoms and a Fermi surface of the
166: fermionic atoms. In contrast, for $\nu \ll 0$, there is a Fermi
167: surface of the fermionic molecules. If the number of the fermionic
168: ($N_f$) and bosonic ($N_b$) atoms are unequal to each other, for $
169: \nu \ll 0$, there will also be some residual atoms which are not
170: in molecules forming their own ground state: for $N_f > N_b$ the
171: extra fermions will form a separate Fermi surface of atoms, while
172: for $N_b > N_f$, the extra bosons will form an atomic BEC. We note
173: that for $N_f > N_b$, scanning the detuning $\nu$ takes us between
174: limits with and without an atomic BEC. Consequently there must be
175: a critical detuning at which the atomic BEC is completely
176: depleted, and all the bosonic atoms have been absorbed into
177: molecules.
178:
179: A novel feature of this quantum phase transition is that it can be
180: entirely characterized in terms of the Luttinger constraints on
181: the Fermi surfaces.
182: \renewcommand{\labelenumi}{({\it\theenumi})}
183: \renewcommand{\theenumi}{\roman{enumi}}
184: \begin{enumerate}
185: \item Consider, first, the phase without the BEC with $\nu
186: \ll 0$. Here, there are 2 Fermi surfaces, one with Fermi
187: surface excitations which are primarily the fermionic atoms, while
188: the other has Fermi surface excitations which are primarily the
189: fermionic molecules. We establish in Section~\ref{luttinger} that
190: this phase obeys {\em two} Luttinger theorems: the atomic Fermi
191: surface encloses a volume associated with precisely $N_f-N_b$
192: states, while the molecular Fermi surface encloses precisely $N_b$
193: states.
194: \item Now consider the phase with the BEC. When the BEC is
195: small, this phase retains 2 Fermi surfaces, one primarily atomic
196: and the other primarily molecular. However, now the volumes
197: enclosed by these Fermi surfaces are not separately conserved;
198: only the total volume enclosed by both Fermi surfaces is required
199: to contain $N_f$ states. Eventually, for $\nu \gg 0$, the
200: molecular Fermi surface disappears entirely, and only a single
201: Fermi surface with $N_f$ states remains. The disappearance of the
202: molecular Fermi surface (in the presence of a BEC) is a second
203: quantum transition whose character we will also discuss briefly in
204: Section~\ref{EffActBoson}.
205: \end{enumerate}
206:
207: This paper will determine the value of the critical $\nu$ for the
208: BEC depletion transition, and describe critical fluctuations in
209: its vicinity. At $T=0$, we will find in Section~\ref{EffActBoson}
210: that this critical point is generically in the universality class
211: \cite{fwgf,book} of the density-driven superfluid-insulator
212: transition with dynamic exponent $z=2$. There is also an
213: interesting quantum multi-critical point for $N_f = 2 N_b$ at
214: which the BEC depletion quantum transition has a different
215: character: this we will also describe. At $T>0$, the BEC depletion
216: transition is in the universality class of the
217: $\lambda$-transition of $^4$He, and so will display similar
218: critical singularities: a peak in the specific heat, and anomalies
219: in transport coefficients.
220:
221: Our quantitative results are determined within a mean-field
222: picture, whose applicability is restricted to the case of a
223: `narrow' Feshbach resonance, where the relevant coupling is
224: sufficiently weak. We nonetheless expect our results to be
225: at least qualitatively applicable to the (experimentally more
226: common) `wide' Feshbach resonance.
227:
228: We also find an additional $T=0$ quantum phase transition
229: involving the disappearance of the molecular Fermi surface. As
230: shown in Section~\ref{EffActBoson}, this is described by a $z=2$
231: critical theory of free fermions.
232:
233: While our work was in progress, we learnt of the work of Yabu {\em
234: et al.} \cite{Yabu} who addressed some related issues, but only in
235: the limit of infinitesimal coupling between the atomic and
236: molecular degrees of freedom. We will note their limiting results
237: in Section~\ref{Zeroglimit}.
238:
239: We now outline the contents of the body of the paper.
240:
241: First, in Section \ref{BasicDefinitions}, we define the model
242: Hamiltonian that will be used throughout the rest of the paper. In
243: Section \ref{Zeroglimit}, we consider the limit of vanishing
244: coupling, where a purely classical analysis can be used.
245: \cite{Yabu}
246:
247: Section \ref{MFforMu} finds the phase structure for finite coupling,
248: treating quantum-mechanical effects using a mean-field approach. In Section
249: \ref{luttinger}, we describe our results regarding Luttinger's theorem for
250: the system. In Section \ref{EffActBoson}, the mean-field result of
251: Section \ref{MFforMu} is reproduced using a field-theoretical
252: approach, which further allows us to characterize the critical
253: properties of the transition.
254:
255: In Sections \ref{GaussianCorrections} and \ref{LowDensityApproximation}, two
256: corrections are calculated to the mean-field theory, which can be used to
257: determine the validity of this approximation. In Section
258: \ref{GaussianCorrections}, the two-loop corrections to the free energy are
259: found, while in Section \ref{LowDensityApproximation}, higher orders in the
260: coupling are included, within a low-density approximation.
261:
262: In the Appendix, we consider the stability of the system against separation
263: into two regions with differing densities. It is shown that the system is indeed
264: stable for a broad range of parameters.
265:
266: \section{Basic definitions}
267: \label{BasicDefinitions}
268:
269: The system consists of bosonic atoms $b$ and fermionic atoms $f$ which combine
270: to form fermionic molecules $\psi$. The energy, relative to the chemical
271: potential $\mu$, is for the atoms
272: \be
273: \label{Definexif}
274: \xi^f_k=\epsilon^f_k-\mu^f=\frac{k^2}{2m^f} - \mu^f
275: \ee
276: \be
277: \xi^b_k=\epsilon^b_k-\mu^b=\frac{k^2}{2m^b} - \mu^b
278: \ee
279: and for the molecule
280: \be
281: \label{Definexipsi}
282: \xi^\psi_k=\epsilon^\psi_k-\mu^\psi=\frac{k^2}{2m^\psi} - \mu^\psi + \nu\punc{,}
283: \ee
284: including the detuning $\nu$. The masses obey $m^\psi=m^f+m^b$ and, because of
285: the interaction, the chemical potentials are related by $\mu^\psi=\mu^f+\mu^b$.
286:
287: The grand Hamiltonian is
288: \bea
289: \Ham &=& \int_k (\xi^f_k f_k^\dag f_k
290: + \xi^b_k b_k^\dag b_k + \xi^\psi_k \psi_k^\dag \psi_k)
291: \nonumber\\
292: &&-\: g\int_{k,k'}(\psi_{k+k'}^\dag f_k b_{k'} + b_{k'}^\dag f_k^\dag \psi_{k+k'})
293: \nonumber\\
294: &&+ \lambda\int_{k,k',\ell} b_{k+\ell}^\dag b_{k'-\ell}^\dag b_{k'}\, b_k\punc{,}
295: \label{Hamiltonian}
296: \eea
297: where $\int_k$ denotes $\int d^3 k\,/(2\pi)^3$.
298:
299: We assume throughout that the fermion spin is
300: polarized along some direction, so that both $f$ and $\psi$ are treated as
301: spinless. The fourth term (in $g$) causes the bosonic and fermionic atoms to
302: couple and form molecules, while the final term (in $\lambda$) is an
303: interaction between pairs of bosons. We omit the interaction between fermions
304: because the exclusion principle forbids $s$-wave scattering between identical
305: fermions and we assume that the interaction between $f$ and $\psi$ will be
306: less important than the coupling $g$.
307:
308: Taking the dimensions of momentum and energy to be unity, $\Dim{k} = \Dim{E} =
309: 1$, we have $\Dim{\psi} = -\frac{3}{2}$ and the same for the operators $b$ and
310: $f$. (Throughout, we shall measure temperature, energy and frequency in the same
311: units, so that $\hbar = k\sub{B} = 1$.) The coupling constants
312: have dimensions $\Dim{g}=-\frac{1}{2}$ and $\Dim{\lambda}=-2$.
313:
314: At temperature $T = 1/\beta\neq 0$, we have six dimensionless parameters.
315: First let $N_b$ be the total density of bosonic atoms, including those
316: bound in molecules, and let $N_f$ be the same for fermionic atoms. (We
317: consider a unit volume, so that density is synonymous with number.) In the
318: absence of any fermions, the bosons would condense at a temperature
319: \be
320: T_0 = \frac{2\pi}{m^b}{\left[\frac{N_b}{\zeta(\smallfrac{3}{2})}\right]}
321: ^\frac{2}{3}\punc{.}
322: \ee
323: We can take as dimensionless parameters $T/T_0 = \beta_0/\beta$, $N_f/N_b$,
324: $m^f/m^b$, $\nu/T_0$, $\gamma^2/T_0$ and $\lambda^2 (m^b)^3 T_0$, where
325: \be
326: \label{Definegamma}
327: \gamma = \frac{g^2}{8\pi} {\left(\frac{2m^f m^b}{m^\psi}\right)}^{3/2}\punc{.}
328: \ee
329:
330: In what follows, it will not usually be necessary to take account of the
331: coupling between bosons given by the final term of (\ref{Hamiltonian}).
332: Except within the condensed phase, which will be treated in Section
333: \ref{MFforMu}, the only effect of $\lambda$ is a renormalization of the boson
334: mass, which we assume has already been incorporated into the definition of
335: $m^b$.
336:
337: \subsection{Physical units}
338:
339: In order to relate these parameters to experimental values, we may choose a unit
340: of volume of $10^{-15}\units{cm^3}$, which gives the unit of momentum as roughly
341: $10^{-27}\units{kg\cdot m/s}$. Taking the unit of mass to be $6\units{amu}$,
342: corresponding to a lithium-6 atom, the unit of energy is roughly $7\times 10^{-
343: 10}\units{eV}$ or $8\units{\mu K}$.
344:
345: For a Feshbach resonance, we assume the expression \cite{Duine}
346: \be
347: g = \sqrt{\frac{2\pi a\sub{bg}\,\Delta B \Delta\mu}{m}}\punc{,}
348: \ee
349: where $a\sub{bg}$ is the background scattering length, $\Delta B$ is the width
350: of the resonance and $\Delta\mu$ is the difference in magnetic moments.
351: Using the observed background scattering length between lithium-6 and -7 of
352: $a\sub{bg} = 2.0\units{nm}$,\cite{schreck} we may estimate the coupling constant.
353: Taking, for instance $\Delta B = 1\units{G}$, $\Delta \mu = \mu\sub{B}$, the
354: Bohr magneton, we find $g \simeq 1$ in our units. For a boson density
355: $N_b = 10^{15}\units{cm^{-3}}$ and mass $m_b = m_f = 6\units{amu}$, the value
356: $g = 1$ gives a dimensionless coupling of $\gamma^2/T_0 = 5\times 10^{-4}$.
357:
358: While the width of the resonance used here, $\Delta B = 1\units{G}$, is
359: sufficiently large that $\Delta\mu \Delta B \gtrsim T_0$, it is nonetheless
360: somewhat smaller than typical experimental values. For our purposes, a more
361: relevant measure of the resonance `width' is the
362: lifetime of the molecule state in the vacuum (for $\nu > 0$). This is calculated
363: in Section \ref{LowDensityApproximation}, where we show that it is determined by
364: the constant $\gamma$. Since the relevant energies are on the order of $T_0$,
365: the condition for a narrow resonance is that $\gamma^2/T_0 \ll 1$. For the
366: numerical results throughout this paper, we will always remain in this narrow
367: limit, which is analytically more accessible. As noted above, we expect our
368: results to be at least qualitatively applicable even for the wider Feshbach
369: resonances observed experimentally.
370:
371: Following Ref.~\onlinecite{ss}, we take
372: \be
373: \lambda = \frac{2\pi}{m^b}a_{bb}\punc{,}
374: \ee
375: where for $a_{bb}$, the scattering length for the boson-boson interaction,
376: we use $a_{bb} = 0.27 \units{nm}$,\cite{schreck} giving
377: \be
378: \lambda^2 (m^b)^3 T_0 = 2\times 10^{-3}\punc{.}
379: \ee
380:
381: The detuning $\nu$ appearing in the molecular
382: dispersion relation (\ref{Definexipsi}) is given by \cite{Duine}
383: \be
384: \nu = \Delta\mu(B - B_0)\punc{,}
385: \ee
386: where $B_0$ is the magnetic field at resonance and $B$ is the applied field.
387:
388: \section{The limit $g\rightarrow 0$}
389: \label{Zeroglimit}
390:
391: The case of vanishing coupling, which can be addressed with a classical
392: approach, has been considered by Yabu {\em et al.} \cite{Yabu} (The results
393: presented in this section produce Fig.~3 of Ref.~\onlinecite{Yabu}, which corresponds
394: to our Fig.~\ref{ZeroTphases}, below.)
395:
396: For simplicity, we restrict the analysis to zero temperature, but similar
397: arguments can be made for nonzero temperatures. We call the two Fermi energies
398: $\epsilon^f_0$ and $\epsilon^\psi_0$, and the corresponding wavenumbers $k^f_0$
399: and $k^\psi_0$. At zero temperature, all bosons are at $\epsilon^b = 0$ and
400: fermionic atoms or molecules must be added at their respective Fermi levels.
401:
402: The atomic Fermi surface (FS) vanishes when all the fermionic atoms are contained
403: in molecules, so that
404: \be
405: k_0^\psi = (6\pi^2\,N_f)^{1/3}\punc{.}
406: \ee
407: (The number of states within a unit sphere in momentum space is $1/6\pi^2$.)
408: For this arrangement to be favorable energetically, the molecular Fermi energy,
409: $\epsilon_0^\psi$, must remain below the lowest atomic energy level. The boundary
410: of the phase without an atomic FS is therefore where
411: \be
412: \frac{1}{2m^\psi}(N_f)^{2/3} + \frac{\nu}{(6\pi^2)^{2/3}} = 0\punc{.}
413: \ee
414: Similarly, the molecular FS vanishes at the point when
415: \be
416: \frac{1}{2m^f}(N_f)^{2/3} - \frac{\nu}{(6\pi^2)^{2/3}} = 0\punc{.}
417: \ee
418: The atomic (molecular) FS is therefore only absent for negative (positive)
419: detuning $\nu$.
420:
421: To find the boundary of the phase with a BEC, we must consider the depletion of
422: the condensate. Bosons will take fermions and form molecules as long as their
423: final energy is lower, \ie\ $\epsilon^\psi_0 < \epsilon^f_0 + 0$.
424: The phase boundary is therefore where $\epsilon^f_0 = \epsilon^\psi_0$, which gives
425: \be
426: \frac{1}{2m^f}{\left({N_f - N_b}\right)}^{2/3} -
427: \frac{1}{2m^\psi}{\left({N_b}\right)}^{2/3}
428: = \frac{\nu}{(6\pi^2)^{2/3}}\punc{,}
429: \ee
430: where the wavenumbers have been determined from $N_b$ and $N_f$, using the fact
431: that there is no condensate.
432:
433: It should be noted that, in this limit, the
434: coupling to fermionic atoms reduces the tendency of the bosons to condense.
435: (The same is true at nonzero temperature.)
436:
437: \section{Mean-field theory}
438: \label{MFforMu}
439:
440: It is possible to go beyond the classical analysis used for vanishing coupling,
441: by using mean-field theory. We will present here two parallel developments, in
442: this section and Section \ref{EffActBoson}, respectively. The first is based on
443: single-particle quantum mechanics, using the mixing between the fermionic
444: dispersion relations caused by the presence of a BEC. The
445: second uses a field-theoretic approach and considers perturbative corrections to
446: the bosonic propagator. The former has the advantage of giving a somewhat
447: clearer physical picture and leading more directly to thermodynamic results
448: (such as the question of phase separation, considered in the Appendix), while
449: the latter leads naturally to higher-order corrections.
450:
451: In the remainder of this section, we present the quantum-mechanical approach,
452: starting from the Hamiltonian (\ref{Hamiltonian}). First, in Section
453: \ref{MFHamiltonian}, we make a mean-field approximation and diagonalize the new
454: Hamiltonian. We then find the condition that a BEC should
455: be energetically favorable, within this approximation.
456:
457: Since the Hamiltonian is defined in the grand canonical ensemble, we must then
458: relate the chemical potentials to the particle numbers, in Section
459: \ref{MFFixedN}. Within the mean-field approximation, it is sufficient to find
460: this relation to order zero in the coupling, neglecting two-loop corrections to
461: the free energy.\cite{ohashi:130402} Later, in Section
462: \ref{GaussianCorrections}, we determine the higher-order corrections.
463:
464: In Section \ref{ZeroT}, we restrict our attention to the case of zero temperature,
465: where transitions occur between states with different numbers of Fermi surfaces. We
466: identify the positions of these transitions and present the full phase diagram
467: for $T = 0$.
468:
469: \subsection{Mean-field Hamiltonian}
470: \label{MFHamiltonian}
471:
472: Replacing the boson field $b_k$ in (\ref{Hamiltonian}) by a real constant
473: $\varphi$ gives
474: \bea
475: \Ham\sub{mf} &=& \int\!\!\frac{d^3 k}{(2\pi)^3}\!\left[ \xi^f_k f_k^\dag f_k
476: + \xi^\psi_k \psi_k^\dag \psi_k
477: - g\varphi(\psi_k^\dag f_k + f_k^\dag \psi_k) \right]\nonumber\\
478: &&- \:\mu^b \varphi^2 + \lambda \varphi^4\punc{,}
479: \label{HamBefore}
480: \eea
481: which can be diagonalized to
482: \be
483: \label{HamAfter}
484: \Ham\sub{mf} = \int\frac{d^3 k}{(2\pi)^3} \left( \xi^F_k F_k^\dag F_k
485: + \xi^\Psi_k \Psi_k^\dag \Psi_k \right) - \mu^b \varphi^2
486: + \lambda \varphi^4\punc{.}
487: \ee
488: The dispersion relations for the mixed fermions $F$, $\Psi$ are
489: \be
490: \xi^{F,\Psi}_k = \half\left( \xi^f_k + \xi^\psi_k \right) \pm
491: \half\sqrt{{\left(\xi^f_k-\xi^\psi_k\right)}^2 + 4g^2\varphi^2}
492: \label{MixedDispersion}\punc{,}
493: \ee
494: with the choice that $\xi^F_k \ge \xi^\Psi_k$ for all $k$.
495:
496: Since the mixing will cause the dispersion relations to separate, the total
497: energy of the fermions is lowered by nonzero $\varphi$. This quantum-mechanical
498: effect, in contrast to the purely classical effect described in Section
499: \ref{Zeroglimit}, therefore acts to favor condensation.
500:
501: We must analyze the energetics to determine the point at which a condensate
502: becomes favorable. The grand free energy $\Phi$ is minimized at temperature
503: $1/\beta$ by a Fermi-Dirac distribution of each fermionic species $x$ ($x \in
504: \{F,\Psi\}$). Ignoring the thermal distribution of bosons, which does not depend
505: on $\varphi$, the total free energy is
506: \be
507: \label{FreeEnergy}
508: \Phi(\varphi) \;\;=\;\; - \mu^b\,\varphi^2\;\; + \;\;\lambda\,\varphi^4\;\;
509: + \sum_{x \in \{F,\Psi\}}
510: R^x(\varphi) \punc{,}
511: \ee
512: where
513: \be
514: \label{DefineR}
515: R^x(\varphi) = -\frac{1}{\beta}\int\frac{d^3 k}{(2\pi)^3}\,\ln{\left(1+e^{-
516: \beta \xi^x_k}\right)}\punc{.}
517: \ee
518:
519: The phase transition to a state with nonzero $\varphi$ occurs when the quadratic
520: coefficient changes sign, \ie\ when
521: \be
522: \Delta \equiv \frac{1}{2}{\left.\frac{d^2 \Phi}{d\varphi^2}\right|}_{\varphi=0}
523: = 0\punc{.}\ee
524: Specifically, for negative $\Delta$, nonzero $\varphi$ is energetically favored,
525: so that the condensed phase is stable. Using (\ref{MixedDispersion}),
526: (\ref{FreeEnergy}) and (\ref{DefineR}), we find
527: \be
528: \label{DiscriminantIntegral}
529: \Delta = -\mu^b + g^2\int\frac{d^3 k}{(2\pi)^3}\,\frac{n\sub{F}(\xi^f_k)-
530: n\sub{F}(\xi^\psi_k)}{\xi^f_k-\xi^\psi_k}\punc{,}
531: \ee
532: where $n\sub{F}$ is the Fermi-Dirac distribution function. The integral equation
533: $\Delta = 0$ may be solved numerically.
534:
535: \subsection{Particle numbers}
536: \label{MFFixedN}
537:
538: Since experiments are necessarily performed at fixed particle number, the
539: expressions for the numbers in terms of the chemical potentials must be found.
540: Particles of the species $b$, $f$ and $\psi$ are not independently
541: conserved, so the relevant quantities are $N_f$ and $N_b$, the total numbers of
542: fermionic and bosonic atoms, respectively (including those contained in molecules).
543:
544: As mentioned above, it is sufficient within mean-field theory to determine these
545: numbers to order zero in the coupling. Since the species $F$ and $\Psi$ each
546: contain one atomic fermion, we have
547: \be
548: N_f = \int\frac{d^3 k}{(2\pi)^3}\,\left[n\sub{F}(\xi^\Psi_k) +
549: n\sub{F}(\xi^F_k)\right]\punc{.}
550: \ee
551: The number of bosons is (with $n\sub{B}$ the Bose-Einstein distribution
552: function)
553: \bea
554: N_b = \varphi^2 + \int\frac{d^3 k}{(2\pi)^3}\,\big[n\sub{B}(\xi^b_k) &+&
555: n\sub{F}(\xi^\Psi_k)\cos^2\theta_k\\
556: &+&n\sub{F}(\xi^F_k)\sin^2\theta_k\big]\punc{,}\nonumber
557: \eea
558: where the first term represents the condensate, the first term in the integrand
559: is the thermal distribution of the bosons, and $\theta_k$ is the mixing
560: angle.\footnote{Explicitly, $\theta_k$ is the angle parametrizing the unitary
561: transformation from the fermions $f$ and $\psi$ in (\ref{HamBefore}) to the
562: fermions $F$ and $\Psi$ in (\ref{HamAfter}).}
563:
564: When $\varphi = 0$, such as along the boundary to the phase without
565: a BEC, the expression for the number of bosons simplifies to
566: \be
567: N_b = \int\frac{d^3 k}{(2\pi)^3}\,\left[n\sub{B}(\xi^b_k) +
568: n\sub{F}(\xi^\psi_k)\right]\punc{.}
569: \ee
570: To locate this phase boundary for fixed particle numbers, we must find the values
571: of $\mu^f$ and $\mu^b$ which give the required numbers and also satisfy $\Delta
572: = 0$. (Of course, a third parameter must be tuned to its critical value to
573: satisfy these three conditions simultaneously.)
574:
575: \begin{figure}
576: \resizebox{\columnwidth}{!}{
577: \includegraphics{tnu0.ftx.eps}
578: }
579: \caption{\label{Graph0}Phase boundary with detuning $\nu$ and temperature $T$,
580: for fixed particle numbers $N_f/N_b = 1.11$. The dashed line has
581: vanishing coupling and has been found with a purely classical analysis. The
582: solid line has dimensionless coupling
583: $\gamma^2/T_0 = 2.5\times 10^{-4}$, and has been determined using the mean-field
584: theory of Section \ref{MFforMu}. For both, the condensed phase is on the
585: left-hand side (for lower $T$) and labeled by $\Mean{b}\neq 0$.}
586: \end{figure}
587: Results from such a procedure are displayed in Fig.~\ref{Graph0}, which shows
588: the boundary for $N_f/N_b = 1.11$ as a function of the detuning $\nu$
589: and temperature $T = 1/\beta$. The masses of the atoms are equal, $m^f = m^b$, and
590: the solid line has dimensionless coupling $\gamma^2/T_0 = 2.5\times 10^{-4}$. For
591: comparison, the case of vanishing
592: coupling, treated in Section \ref{Zeroglimit}, is also shown with a dashed line.
593: Both curves reach the value $T = T_0$, as in the case of free bosons, for
594: $\nu\rightarrow\infty$, when molecules cannot be formed.
595:
596: In Fig.~\ref{Phases} the same phase boundary is shown on a graph of fermion number
597: versus detuning, for three different temperature values. The solid line is at zero
598: temperature, $T = 0$, while the two dashed lines have nonzero temperatures. The
599: coupling is $\gamma^2/T_0 = 2.5\times 10^{-4}$ and the masses are equal, $m^f = m^b$.
600: As expected, Bose condensation is favored by lower temperatures, as in the case of
601: an isolated Bose gas.
602: \begin{figure}
603: \resizebox{\columnwidth}{!}{
604: \includegraphics{phases.ftx.eps}
605: }
606: \caption{\label{Phases}Phase boundary with fermion number $N_f$ and detuning $\nu$,
607: for three different temperatures. The coupling is $\gamma^2/T_0 = 2.5\times 10^{-4}$
608: and the masses are equal, $m^f = m^b$. The two phases are labeled as in
609: Fig.~\ref{Graph0}, with the condensed phase favored for higher detuning, lower
610: fermion number and lower temperature.}
611: \end{figure}
612:
613: It remains to be shown that the system is stable against separation into regions
614: with different densities. It is shown in the Appendix that it is indeed stable
615: for a large range of parameter values.
616:
617: \subsection{Zero-temperature phases}
618: \label{ZeroT}
619:
620: At $T = 0$, the Fermi-Dirac distribution function is replaced by a unit step
621: and all bosons occupy the lowest-energy state. As noted by Yabu {\em
622: et al.}, \cite{Yabu}
623: the phase diagram can be further divided into a
624: region with two Fermi surfaces and a region with a single Fermi surface. (We ignore
625: the trivial case without any Fermi surfaces, which requires $N_f = 0$.)
626:
627: Except when the atomic numbers precisely match, $N_f = N_b$, the case of a single
628: surface can only occur when there is a BEC. In this case, $\varphi$, the
629: expectation value of $b$, is given by the minimum of the free energy $\Phi$ given
630: in (\ref{FreeEnergy}), so that we must solve
631: \be
632: \label{EqCondensate}
633: - 2\mu^b\,\varphi\;\; + \;\;4\lambda\,\varphi^3\;\;
634: + \sum_{x \in \{F,\Psi\}}
635: \frac{dR^x}{d\varphi} \;=\; 0
636: \ee
637: (excluding the root $\varphi = 0$).
638:
639: Following the choice that $\xi^F_k\ge\xi^\Psi_k$ in (\ref{MixedDispersion}), the
640: second Fermi surface disappears when $\xi^F_{k=0} = 0$, making the Fermi
641: wavenumber for $F$ fermions vanish. For this to be the case, we require $\mu^f > 0$,
642: $\mu^\psi > \nu$ and
643: \be
644: g\varphi = \sqrt{\mu^f(\mu^\psi - \nu)}\punc{,}
645: \ee
646: which should be solved simultaneously with (\ref{EqCondensate}).
647:
648: \begin{figure}
649: \resizebox{\columnwidth}{!}{
650: \includegraphics{ZeroTphases.ftx.eps}
651: }
652: \caption{\label{ZeroTphases}The phase diagram at $T = 0$ with dimensionless
653: couplings (a) $\gamma^2/T_0 = 10^{-6}$, (b) $\gamma^2/T_0 = 2.5\times 10^{-4}$ and
654: (c) $\gamma^2/T_0 = 2.0\times 10^{-2}$. The atomic masses have been taken to be equal,
655: $m^f = m^b$ and the coupling between bosons is given by
656: $\lambda^2 (m^b)^3 T_0 = 2\times 10^{-3}$.
657: The three distinct phases have, respectively, no Bose-Einstein
658: condensate and two Fermi surfaces (labeled `2 FS, no BEC'), a condensate and two
659: Fermi surfaces (`2 FS + BEC'), and a condensate and a single Fermi surface
660: (`1 FS + BEC'). The dotted line indicates the fermion number at which
661: Fig.~\ref{ZeroTmass} is plotted.}
662: \end{figure}
663: These expressions, along with the results in Section \ref{MFFixedN} for the particle
664: numbers, allow the complete zero-temperature phase diagram to be plotted. In
665: Fig.~\ref{ZeroTphases}, the phase boundaries are shown on a graph of fermion number
666: against detuning, for equal atomic masses $m^f = m^b$.
667: The three sets of boundaries have couplings (a) $\gamma^2/T_0 = 10^{-6}$, (b)
668: $\gamma^2/T_0 = 2.5\times 10^{-4}$ and (c) $\gamma^2/T_0 = 2.0\times 10^{-2}$.
669: (Since the dimensionless coupling depends on the fourth
670: power of the coupling $g$ appearing in the Hamiltonian, these large changes in
671: $\gamma^2/T_0$ in fact correspond to changes in $g$ of only factors of $4$ and $3$
672: respectively. All of these coupling values are within the narrow resonance regime.)
673: Throughout, we take $\lambda^2 (m^b)^3 T_0 = 2\times 10^{-3}$.
674:
675: The boundaries divide the diagram into three regions, depending on the presence of a
676: condensate and the number of Fermi surfaces. In the region labeled `2 FS, no BEC',
677: the discriminant $\Delta$ is positive, so there is no BEC and
678: two Fermi surfaces. In the region labeled `2 FS + BEC', $\Delta$ is negative and
679: there is a condensate, as well as two Fermi surfaces. The lowermost region of the
680: diagram, labeled `1 FS + BEC', has a condensate and only a single Fermi surface.
681:
682: In the limit of vanishing coupling (as in Ref.~\onlinecite{Yabu} and Section
683: \ref{Zeroglimit}), the boundary between the regions
684: with one and two Fermi surfaces extends down to the line $N_f = 0$. The region with
685: a single Fermi surface is then divided into two, with the left-hand side having
686: a Fermi surface of molecules and the right-hand side a Fermi surface of atoms.
687: Including the quantum-mechanical effects, these two regions are no longer distinct,
688: with the single Fermi surface crossing over from having a molecular character on
689: one side (lower $\nu$) to having an atomic character on the other (higher $\nu$).
690:
691: This crossover is illustrated in Fig.~\ref{ZeroTmass}, where the effective mass
692: $m^\star$ at the Fermi surface is plotted. The fermion number is set at
693: $N_f = 0.1N_b$ and the coupling is $\gamma^2/T_0 = 2.5\times 10^{-4}$, so that the
694: system is within the phase with a single Fermi surface (of $\Psi$ fermions). The
695: effective mass is defined as
696: \be
697: m^\star = {\left({\left.\frac{d^2\xi^\Psi_k}{dk^2}\right|}_{k_0^\Psi}\right)}^{-1}\punc{.}
698: \ee
699: For $\nu\ll 0$, the Fermi surface has an essentially molecular character and
700: $m^\star\simeq m^\psi$, while for $\nu\gg 0$, it is atom-like, with
701: $m^\star\simeq m^f$.
702: \begin{figure}
703: \resizebox{\columnwidth}{!}{
704: \includegraphics{ZeroTmass.ftx.eps}
705: }
706: \caption{\label{ZeroTmass}The effective mass $m^\star$ at the Fermi surface,
707: with fermion number $N_f = 0.1 N_b$, coupling $\gamma^2/T_0 = 2.5\times 10^{-4}$
708: and equal atomic masses $m^f = m^b$. As can be seen from the dotted line in
709: Fig.~\ref{ZeroTphases}, these parameters give a phase with a single
710: Fermi surface. This surface changes from having a molecular character, with
711: $m^\star\simeq m^\psi$, to having an atomic character, $m^\star\simeq m^f$.}
712: \end{figure}
713:
714: In Fig.~\ref{ZeroTk}, the Fermi wavenumbers of the two fermionic species are plotted,
715: for coupling $\gamma^2/T_0 = 2.5\times 10^{-4}$ and two different fermion numbers,
716: $N_f = \frac{3}{2} N_b$ (solid lines) and $N_f = \frac{1}{2} N_b$ (dashed lines).
717: In both the phase without a condensate (solid lines for $\nu/T_0 < 0.25$) and the
718: phase with a single Fermi surface (solid lines for $\nu/T_0 > 2.9$, dashed lines
719: for $\nu/T_0 < -0.65$ and $\nu/T_0 > 1.3$) the wavenumbers are constant, due to the
720: fixed particle numbers. Only in the phase with two Fermi surfaces and a BEC
721: do the Fermi wavevectors change with detuning.
722: (At the fermion number used in Fig.~\ref{ZeroTmass}, the system stays in the phase
723: with a single Fermi surface throughout and $k_0^\Psi = k_f$, $k_0^F = 0$ for all
724: detunings.)
725: \begin{figure}
726: \resizebox{\columnwidth}{!}{
727: \includegraphics{ZeroTk.ftx.eps}
728: }
729: \caption{\label{ZeroTk}The Fermi wavenumbers for the two mixed species of fermions,
730: $\Psi$ and $F$, with coupling $\gamma^2/T_0 = 2.5\times 10^{-4}$ and equal atomic
731: masses $m^f = m^b$. The solid lines have fermion number $N_f = \frac{3}{2} N_b$,
732: while the dashed lines have $N_f = \frac{1}{2} N_b$. As can be seen in
733: Fig.~\ref{ZeroTphases}, the solid line goes between all three phases (at
734: $\nu/T_0\simeq 0.25$ and $\nu/T_0\simeq 2.9$), while the dashed line goes from
735: the phase with a single Fermi surface to that having two and back again (at
736: $\nu/T_0\simeq -0.65$ and $\nu/T_0\simeq 1.3$). The wavenumbers are measured
737: in units of $k_f$, the Fermi wavenumber for free fermions with number $N_f$.}
738: \end{figure}
739:
740: We now turn our attention to the line dividing the phases 2 FS, no BEC and
741: 1 FS + BEC in Fig.~\ref{ZeroTphases}. This boundary is horizontal and starts
742: at the point where the three phases meet; in Section \ref{Multi}, we will prove
743: that this is at exactly $N_b = N_f$. At this transition, two changes
744: occur: both the second Fermi surface vanishes and the BEC
745: appears, as the line is crossed from above. Physically, this results from the fact
746: that molecules are highly energetically favored in this region, so that as many
747: molecules as possible are formed, and the residual atoms form their ground state.
748: For $N_f > N_b$, these atoms are fermionic and form a Fermi surface, while for
749: $N_f < N_b$, they are bosonic and form a condensate. Precisely at $N_f = N_b$,
750: there are no residual atoms, so that there is no condensate and only a molecular
751: Fermi surface.
752:
753: This situation is illustrated by Fig.~\ref{ZeroTmu}, which shows
754: the same phase diagram as Fig.~\ref{ZeroTphases}, but with the
755: chemical potential for the fermionic atoms, $\mu^f$, on the
756: vertical axis. Throughout the plot $\mu^\psi - \nu$, and hence the
757: Fermi wavenumber for the molecules, $k_0^\psi$, is held fixed. In
758: the region where $\mu^f > 0$, the essential features are
759: unchanged, with the same three phases as shown in
760: Fig.~\ref{ZeroTphases}. The boundary between the phases 2 FS, no
761: BEC and 1 FS + BEC, however, is seen to extend into an entire
762: phase, labeled `1 FS, no BEC'. In this region, there is no
763: condensate and $\mu^f$ is negative, so that there is only one
764: Fermi surface, of molecules. This entire phase therefore has $N_f
765: = N_b$ and collapses onto a single line in Fig.~\ref{ZeroTphases}.
766: Moreover, because $k_0^\psi$ is constant, $N_f$ and $N_b$ are both
767: fixed in this phase. The situation in the shaded region of
768: Fig.~\ref{ZeroTmu} resembles that in the Mott insulator lobes in
769: the phase diagram of the boson Hubbard model (see
770: Ref.~\onlinecite{fwgf} and Chapter 10 of Ref.~\onlinecite{book});
771: at fixed $\mu^\psi$, the density of particles is insensitive to
772: the variation in the chemical potential $\mu^f$.
773: \begin{figure}
774: \resizebox{\columnwidth}{!}{
775: \includegraphics{ZeroTmu.ftx.eps}
776: }
777: \caption{\label{ZeroTmu}The phase diagram with the fermion
778: chemical potential $\mu^f$ plotted on the vertical axis and the
779: detuning $\nu$ on the horizontal axis. Both have been scaled to
780: $\mu^\psi - \nu$, which is held fixed throughout the plot. The
781: boundary between 2 FS, no BEC and 1 FS + BEC in
782: Fig.~\ref{ZeroTphases} expands into a new phase, labeled `1 FS, no
783: BEC', within which there are only molecules, whose density is
784: constant (both $N_f$ and $N_b$ remain fixed in the shaded region).
785: The atomic masses are equal, $m^f = m^b$, and the couplings are
786: $\gamma^2/T_0 = 2.5\times 10^{-4}$ and $\lambda^2 (m^b)^3 T_0 =
787: 2\times 10^{-3}$. (Since the boson density is not fixed in this
788: plot, the value of $T_0$ appropriate to the phase 1 FS, no BEC has
789: been used to define the dimensionless couplings.)}
790: \end{figure}
791:
792: \section{Luttinger's theorem}
793: \label{luttinger}
794:
795: All the ground states in our phase diagram in
796: Fig.~\ref{ZeroTphases} contain Fermi surfaces. In
797: Fig.~\ref{ZeroTk} we presented the evolution of the Fermi
798: wavevectors of these Fermi surfaces in our mean-field calculation.
799: In the present section we will discuss general constraints that
800: must be satisfied by these Fermi wavevectors which are valid to
801: all orders in the interactions. (Throughout this section, we shall
802: be concerned only with $T=0$.)
803:
804: We will base our arguments upon the existence of the Luttinger-Ward
805: functional \cite{luttingerward} $\LW$, satisfying
806: \be
807: \label{EqDefineLW}
808: \Sigma = {\left.\frac{\delta \LW[G']}{\delta G'}\right|}_{G'=G}\punc{,}
809: \ee
810: where $G'$ is a dummy variable, $G$ is the actual full (thermal) Green function
811: and $\Sigma$ is the full self-energy. (Throughout this section, we will
812: mostly be concerned with the full Green functions, which we shall denote
813: with the symbol $G$. When we make reference to the free Green function,
814: this will be denoted $G_0$.)
815:
816: Following Ref.~\onlinecite{potthoff},
817: it possible to construct the Luttinger-Ward functional non-perturbatively,
818: starting from the
819: partition function $Z$ of the system. It can be shown straightforwardly
820: that, treating the Green function as a matrix in its momentum (and frequency)
821: indices, any unitary transformation of the free Green function,
822: $G_0 \rightarrow UG_0 U^{-1}$, that leaves $Z$ invariant also
823: leaves $\LW$ invariant.
824:
825: A standard proof of Luttinger's theorem \cite{agd}
826: for a system of interacting fermions makes use of the invariance of $Z$ under
827: a shift in the frequency appearing in the free propagator,
828: $\omega\rightarrow \omega + \alpha$. In our case $\LW$ is a functional
829: of the three Green functions, one for each species, and $Z$ is invariant under
830: a simultaneous shift in two of the three frequencies, \ie
831: \bea
832: &\LW[G^\psi(i\nu_1),G^f(i\nu_2),G^b(i\omega)]&\nonumber\\
833: =&\LW[G^\psi(i\nu_1),G^f(i\nu_2-i\alpha),G^b(i\omega+i\alpha)]&\nonumber\\
834: =&\LW[G^\psi(i\nu_1+i\beta),G^f(i\nu_2),G^b(i\omega+i\beta)]&
835: \label{EqInvariance}
836: \eea
837: for any $\alpha$ and $\beta$.
838:
839: To proceed further, it is useful to set $\mu^b = \mu^\psi - \mu^f$ and
840: consider derivatives of the grand energy with respect to $\mu^f$
841: and $\mu^\psi$. The derivative with respect to $\mu^f$ yields
842: \be
843: \Mean{f^\dagger f} - \Mean{b^\dagger b} = N_f - N_b\punc{.}
844: \label{const1}
845: \ee
846: Each term on the left-hand side can be rewritten in terms of the full
847: Green functions, giving
848: \be
849: \label{EqNumbersFromGreenFunc}
850: N_f - N_b = -\int\frac{d^3k\,d\omega}{(2\pi)^4}e^{i\omega 0^+}
851: \left[G^f_k(i\omega) + G^b_k(i\omega)\right]\punc{.}
852: \ee
853: (The change of sign of the $f$ term results from the anticommutation
854: of fermion operators.)
855:
856: From now on the manipulations are standard.\cite{agd} We make use of the identity
857: \be
858: G(i\omega) = iG(i\omega)\parder{}{\omega}\Sigma(i\omega)
859: - \parder{}{\omega}\ln{G(i\omega)}\punc{,}
860: \ee
861: which results from the Dyson equation. The first equation of
862: (\ref{EqInvariance}) gives, together with (\ref{EqDefineLW}),
863: \be
864: \int\frac{d^3k\,d\omega}{(2\pi)^4}\left[
865: \Sigma^b_k(i\omega)\parder{}{\omega}G^b_k(i\omega) +
866: \Sigma^f_k(i\omega)\parder{}{\omega}G^f_k(i\omega)
867: \right] = 0\punc{.}
868: \ee
869: Combining these two with (\ref{EqNumbersFromGreenFunc}) and integrating by
870: parts gives
871: \be
872: N_f - N_b = i\int\frac{d^3k\,d\omega}{(2\pi)^4}e^{i\omega 0^+}
873: \parder{}{\omega}\left[\ln G^f_k(i\omega) + \ln G^b_k(i\omega)\right]
874: \ee
875: (with the boundary terms vanishing because $G(i\omega)\sim 1/|\omega|$ for
876: $|\omega|\rightarrow\infty$).
877:
878: The integral over $\omega$ can be treated as a contour integration and
879: closed above, due to the factor $e^{i\omega 0^+}$. Changing the integration
880: variable to $z = i\omega$ replaces this by an integral surrounding the left
881: half-plane. Since both of the full Green functions $G^{b,f}_k(z)$ have all
882: their non-analyticities and zeroes on the line of real $z$, we can write this as
883: \begin{widetext}
884: \be
885: N_f - N_b = i\int\frac{d^3k}{(2\pi)^3}\int_{-\infty}^{0}\frac{dz}{2\pi}
886: \parder{}{z}\left[\ln G^f_k(z+i0^+) + \ln G^b_k(z+i0^+)
887: - \ln G^f_k(z+i0^-) - \ln G^b_k(z+i0^-)\right]\punc{.}
888: \ee
889: \end{widetext}
890: The integral of $z$ can be performed trivially to give
891: \be
892: N_f - N_b = i\int\frac{d^3k}{(2\pi)^4}
893: \left[\ln \frac{G^f_k(i0^+)}{G^f_k(i0^-)}
894: +\ln \frac{G^b_k(i0^+)}{G^b_k(i0^-)}
895: \right]\punc{.}
896: \ee
897:
898: Using the analyticity properties of the Green functions, this gives
899: \be
900: \label{EqStepFunctions}
901: N_f - N_b = \int\frac{d^3k}{(2\pi)^3}\left[\Theta(-\xi^f_k+\Sigma'^f_k)
902: +\Theta(-\xi^b_k+\Sigma'^b_k)\right]\punc{,}
903: \ee
904: where $\Theta$ is the unit step function and $\Sigma'$ is the real part of the
905: self-energy evaluated for $\omega = 0$.
906:
907: First, we consider the phase with no BEC. Here there are
908: necessarily two Fermi surfaces, and, as we will now show, the
909: volumes of the two Fermi surfaces are separately constrained, independently
910: of the interactions.
911:
912: By definition, the absence of a BEC requires that there be no bosonic
913: quasiparticle excitations at or above the chemical potential, so that the second
914: term in the brackets in (\ref{EqStepFunctions}) vanishes. (Note that this does
915: not imply that $\Mean{b^\dagger b} = 0$, which is not the case beyond mean-field
916: order.) This leaves the statement of Luttinger's theorem for this case:
917: \be
918: \label{EqLutt1}
919: N_f - N_b = \int\frac{d^3k}{(2\pi)^3}\Theta(-\xi^f_k+\Sigma'^f_k)\punc{.}
920: \ee
921: The right-hand side of this expression is interpreted as the (reciprocal-space)
922: volume of the atomic Fermi surface.
923:
924: A similar result follows by taking the derivative of the grand energy
925: with respect to $\mu^\psi$, which gives
926: \be
927: \Mean{\psi^\dagger \psi} + \Mean{b^\dagger b} = N_b\punc{.}
928: \label{const2}
929: \ee
930: Going through the same manipulations as above leads to
931: \be
932: \label{EqStepFunctions2}
933: N_b = \int\frac{d^3k}{(2\pi)^3}\left[\Theta(-\xi^\psi_k+\Sigma'^\psi_k)
934: -\Theta(-\xi^b_k+\Sigma'^b_k)\right]\punc{,}
935: \ee
936: corresponding to (\ref{EqStepFunctions}). Since we are in the phase with no BEC,
937: this gives
938: \be
939: \label{EqLutt2}
940: N_b = \int\frac{d^3k}{(2\pi)^3}\Theta(-\xi^\psi_k+\Sigma'^\psi_k)\punc{.}
941: \ee
942:
943: We have therefore proved that there are two statements of Luttinger's theorem
944: in the phase with two Fermi surfaces and no BEC. One, (\ref{EqLutt1}), states that
945: the volume of the atomic Fermi surface is fixed by the difference in the numbers
946: of atomic fermions and bosons, while the other, (\ref{EqLutt2}), states that the
947: volume of the molecular Fermi surface is fixed by the total number of bosonic atoms.
948:
949: Now let us extend our considerations to the phases with a BEC. In
950: Fig.~\ref{ZeroTphases} we observe that these phases may have
951: either one or two Fermi surfaces. Here we show that Luttinger's
952: theorem only demands that the total volume enclosed within {\em
953: both} Fermi surfaces is determined by $N_f$; the volumes of the two
954: Fermi surfaces (if present) are not constrained separately.
955:
956: In the presence of a BEC, it is no longer the case that the second
957: term in the brackets vanishes in (\ref{EqStepFunctions}) and (\ref{EqStepFunctions2}).
958: Instead, if we add these two equations, we arrive at
959: \be
960: \label{EqLutt3}
961: N_f = \int\frac{d^3k}{(2\pi)^3}\Theta(-\xi^f_k+\Sigma'^f_k)
962: + \int\frac{d^3k}{(2\pi)^3}\Theta(-\xi^\psi_k+\Sigma'^\psi_k)\punc{.}
963: \ee
964: The two terms in this expression give the volumes of the two Fermi surfaces.
965: We see that their sum is constrained to equal the number of fermionic atoms.
966:
967: \subsection{Multicritical point}
968: \label{Multi}
969:
970: A simple application of our statements of Luttinger's theorem can
971: be used to show that the multicritical point, where the three
972: phases meet in Fig.~\ref{ZeroTphases} (and where four phases meet
973: in Fig.~\ref{ZeroTmu}), occurs at precisely $N_b = N_f$.
974:
975: Firstly, according to (\ref{EqLutt1}), the volume of the atomic
976: Fermi surface is given by $N_f - N_b$, as long as there is no BEC.
977: This is therefore the case on the line dividing the phases with
978: and without condensates, since the condensate vanishes as this
979: line is approached from below. Secondly, the line that divides the
980: regions with one and two Fermi surfaces is the point where the
981: atomic Fermi surface vanishes. Where the two lines meet, we see
982: both that (\ref{EqLutt1}) is satisfied and that its right-hand
983: side vanishes. We therefore have $N_b = N_f$.
984:
985: \section{Quantum phase transitions}
986: \label{EffActBoson}
987:
988: We now present an alternative analysis using the language of field
989: theory. In Section \ref{MF2}, we reproduce the result that
990: (\ref{DiscriminantIntegral}) determines the presence of the
991: condensate. Then, in Section \ref{SecBosonPropagator}, we
992: determine the boson propagator near the BEC depletion transition.
993: In Section \ref{critical}, we describe the critical field theories
994: for the various transitions.
995:
996: The dimensionless Euclidean action corresponding to the Hamiltonian
997: (\ref{Hamiltonian}) is
998: \bea
999: \Act &= &\frac{1}{\beta}\sum_q \bar{f}_q \Xi^f_q f_q + \frac{1}{\beta}\sum_p
1000: \bar{b}_p \Xi^b_p b_p + \frac{1}{\beta}\sum_q \bar{\psi}_q
1001: \Xi^\psi_q\psi_q\nonumber\\
1002: && -\: \frac{g}{\beta^2}\sum_{p,q} \left(\bar{\psi}_q f_{q-p} b_p +
1003: \bar{b}_p\bar{f}_{q-p}\psi_p\right)\punc{.}
1004: \label{eqAction}
1005: \eea
1006: The symbol $p$ stands for $k$ and $\omega_n$, and likewise $q$ for $\ell$ and
1007: $\nu_m$, where the Matsubara frequencies $\omega_n$ ($\nu_m$) are even (odd).
1008: The summations over $p$ ($q$) represent sums over $\omega_n$ ($\nu_m$) and
1009: integrals over the momentum $k$ ($\ell$). We have also defined
1010: \be
1011: \Xi_p = {\left(G_p\right)}^{-1} = -i\omega_n + \xi_k\punc{,}
1012: \ee
1013: the inverse of the free Green function, $G_p$, and similarly $\Xi_q$. (In this
1014: section and the following, we will use the symbol $G$ to denote the free Green
1015: function, contrary to the notation of Section \ref{luttinger}.)
1016:
1017: We have omitted from the action the coupling term between pairs of bosons,
1018: since we will be interested in the region near the phase transition, where
1019: this term is not important.
1020:
1021: Integrating out both of the fermions and expanding the resulting coupling term
1022: to quadratic order in $b$ and $\bar{b}$, we find that the effective action for
1023: the bosons is
1024: \be
1025: \label{QuadAction}
1026: \Act^{(2)}\sub{eff}[b,\bar{b}] = \frac{1}{\beta}\sum_p\bar{b}_p\Xi^b_p b_p +
1027: \frac{g^2}{\beta^2}\sum_{p,q}G^f_q G^\psi_{q+p} \bar{b}_{p} b_{p}\punc{.}
1028: \ee
1029:
1030: \subsection{Mean-field approximation}
1031: \label{MF2}
1032:
1033: By replacing $b$ by a real constant $\varphi$, we should arrive at the results
1034: of Section \ref{MFforMu}. In this approximation, we have
1035: \be
1036: \Act^{(2)}\sub{eff}[b,\bar{b}] = -\mu^b \varphi^2 +
1037: \frac{g^2}{\beta}\varphi^2\sum_{q}G^f_q G^\psi_q\punc{,}
1038: \ee
1039: so that the coefficient is
1040: \be
1041: \label{DeltaInFieldTheory}
1042: \Delta = -\mu^b + g^2\int\frac{d^3
1043: \ell}{(2\pi)^3}\frac{1}{\beta}\sum_{\nu_m}G^f_\ell(i\nu_m)\,
1044: G^\psi_\ell(i\nu_m)\label{MeanFieldProp}\punc{.}
1045: \ee
1046: The phase transition will occur when the coefficient $\Delta$ vanishes.
1047:
1048: We can represent (\ref{DeltaInFieldTheory}) by
1049: \be
1050: \parbox{35mm}{
1051: \begin{picture}(35,20)
1052: \put(0,0){\includegraphics{fmfPaper1.ps}}
1053: \input{fmfPaper.t1}
1054: \end{picture}
1055: }\punc{,}
1056: \label{FirstGraph}
1057: \ee
1058: which appears as a self-energy diagram in the boson propagator, drawn as a
1059: dashed line. (The two solid lines represent fermion propagators.)
1060:
1061: The Matsubara sum can be performed by replacing it by a contour integration,
1062: giving
1063: \be
1064: \Delta = -\mu^b + g^2\int\frac{d^3 \ell}{(2\pi)^3}\,\frac{n\sub{F}(\xi^f_\ell)-
1065: n\sub{F}(\xi^\psi_\ell)}{\xi^f_\ell-\xi^\psi_\ell}\punc{,}
1066: \ee
1067: in agreement with (\ref{DiscriminantIntegral}).
1068:
1069: \subsection{Boson propagator}
1070: \label{SecBosonPropagator}
1071:
1072: By retaining the frequency dependence of the boson field, but again keeping
1073: only terms quadratic in $b$ and $\bar{b}$, we can determine the form of the
1074: long-wavelength, low-frequency excitations.
1075:
1076: The effective boson propagator is, from (\ref{QuadAction}), the reciprocal of
1077: \bea
1078: \Xit^b_k(i\omega_n)&\equiv&\Xit^b_p\;\equiv\;\Xi^b_p +
1079: \frac{g^2}{\beta}\sum_q G^f_{q} G^\psi_{q+p}\label{EffBoseDisp}\\
1080: &=&-i\omega_n+\xi^b_k + g^2\int\frac{d^3
1081: \ell}{(2\pi)^3}\times\nonumber\\
1082: &&\frac{1}{\beta}\sum_{\nu_m}\frac{1}{-
1083: i\nu_m+\xi^f_{\ell}}\frac{1}{-
1084: i(\nu_m+\omega_n)+\xi^\psi_{\ell+k}}\nonumber\punc{,}
1085: \eea
1086: which replaces (\ref{MeanFieldProp}). For $k=0$, this gives
1087: \be
1088: \Xit^b_0(i\omega_n)=-i\omega_n-\mu^b+g^2\int\frac{d^3
1089: \ell}{(2\pi)^3}\,\frac{n\sub{F}(\xi^f_\ell)-n\sub{F}(\xi^\psi_\ell)}{\xi^f_\ell-
1090: \xi^\psi_\ell+i\omega_n}
1091: \punc{,}
1092: \ee
1093: where the result
1094: \be
1095: n\sub{F}(a - i\omega_n) = n\sub{F}(a)\punc{,}
1096: \ee
1097: for $\omega_n$ a boson Matsubara frequency, has been used. For small $\omega_n$,
1098: we can expand to give
1099: \bea
1100: \Xit^b_0(i\omega_n) &\simeq& \Delta - i\omega_n\left[ 1 - g^2\int\frac{d^3
1101: \ell}{(2\pi)^3}\frac{n\sub{F}(\xi^f_\ell)-n\sub{F}(\xi^\psi_\ell)}{{(\xi^f_\ell
1102: - \xi^\psi_\ell)}^2} \right]\nonumber\\
1103: &&-\;\omega_n^2 g^2\int\frac{d^3 \ell}{(2\pi)^3}\frac{n\sub{F}(\xi^f_\ell)-
1104: n\sub{F}(\xi^\psi_\ell)}{{(\xi^f_\ell - \xi^\psi_\ell)}^3}
1105: \punc{.}
1106: \eea
1107: (Note that, as required, the coefficient of $\omega_n^2$ is in fact positive.)
1108:
1109: The effective boson propagator (for $k=0$) is then
1110: \be
1111: \label{Gtildeb}
1112: \Gt^b_0(i\omega_n)=\frac{\mathcal{Z}}{-i\omega_n + \xit^b_0(\omega_n)}\punc{,}
1113: \ee
1114: with
1115: \be
1116: \mathcal{Z} = {\left[ 1 - g^2\int\frac{d^3
1117: \ell}{(2\pi)^3}\frac{n\sub{F}(\xi^f_\ell)-n\sub{F}(\xi^\psi_\ell)}{{(\xi^f_\ell
1118: - \xi^\psi_\ell)}^2}\right]}^{-1}\punc{,}
1119: \ee
1120: and
1121: \be
1122: \xit^b_0(\omega_n)=\mathcal{Z}\left[\Delta - \omega_n^2 g^2\int\frac{d^3
1123: \ell}{(2\pi)^3}\frac{n\sub{F}(\xi^f_\ell)-n\sub{F}(\xi^\psi_\ell)}{{(\xi^f_\ell
1124: - \xi^\psi_\ell)}^3}\right]\punc{.}
1125: \ee
1126:
1127: The integrals in the expressions for both $\mathcal{Z}$ and
1128: $\xit^b_0(\omega_n)$ diverge at zero temperature if $N_f = 2 N_b$
1129: so that the two Fermi wavenumbers coincide. For any other
1130: parameters the integrals are finite, and the effective propagator
1131: has the form (\ref{Gtildeb}). As we discuss in the following
1132: subsection, this distinction leads to different field theories for
1133: the BEC depletion quantum transition for these cases.
1134:
1135: \subsection{Critical field theories}
1136: \label{critical}
1137:
1138: First, we consider the BEC depletion quantum transition. This is
1139: the transition between the 2 FS + BEC phase and the 2 FS, no BEC
1140: phase in Fig.~\ref{ZeroTphases}. The same theory also applies to
1141: the transition between the 1 FS + BEC phase and the 1 FS, no BEC
1142: phase in Fig.~\ref{ZeroTmu}. The low momentum modes of the $b$
1143: boson field clearly constitute an order parameter for this
1144: transition. The effective action for the renormalized $b$ field
1145: near the critical point can be derived by integrating out the
1146: fermionic degrees of freedom, as already outlined in
1147: Section~\ref{SecBosonPropagator}.
1148:
1149: For $N_f \neq 2 N_b$, the integration of the fermionic excitations
1150: is entirely free of infrared singularities: the differences in the
1151: two Fermi wavevectors implies that there are no low momentum
1152: fermionic particle-hole excitations at low energies. The resulting
1153: action for $b$ contains only terms which are analytic in frequency
1154: and momentum, and has the following familiar form:
1155: \begin{eqnarray}
1156: \mathcal{S}_c [b] &=& \int d^3 r \int d \tau \Biggl[ b^\dagger
1157: \frac{\partial b}{\partial \tau} - \frac{1}{2\tilde{m}^b} b^\dagger \nabla^2
1158: b + s |b|^2 \nonumber \\ &~&~~~~~~~~~~~~~~~~~~~~~~~~~+ \frac{u}{2}
1159: |b|^4 \Biggr] \label{scb}
1160: \end{eqnarray}
1161: Note that the $b$ field has been rescaled by a factor $\sqrt{\mathcal{Z}}$
1162: from the $b$ field in Section~\ref{SecBosonPropagator} and that its
1163: mass has been replaced by the renormalized mass $\tilde{m}^b$.
1164: The action $\mathcal{S}_c
1165: [b]$ describes a quantum phase transition with dynamic exponent
1166: $z=2$ driven by tuning the coupling $s$. This transition has been
1167: discussed previously in Ref.~\onlinecite{fwgf} and in Chapter 11
1168: of Ref.~\onlinecite{book}. The upper critical dimension is $d=2$,
1169: above which the quartic coupling $u$ is formally irrelevant.
1170: Nevertheless, the coupling $u$ is important for the $T>0$
1171: crossovers in the vicinity of the quantum critical point: these
1172: are as presented in Ref.~\onlinecite{book}.
1173:
1174: For $N_f = 2N_b$, there are low energy fermionic particle-hole
1175: excitations at zero momentum, and so the above procedure has to be
1176: reconsidered. Now there are non-analytic terms in the effective action
1177: for $b$, but these have a structure similar to that found by Hertz
1178: \cite{hertz} for the onset of ferromagnetism in a Fermi liquid.
1179: Evaluating (\ref{EffBoseDisp}) for this case following Hertz, we now
1180: find the effective action
1181: \begin{eqnarray}
1182: \mathcal{S}_c [b] &=& \int d^3 k \int d \omega |b (k, \omega)|^2
1183: \left[ k^2 + c \frac{|\omega|}{k} \right] \nonumber \\
1184: &+& \int d^3 r \int d \tau \Biggl[ s |b|^2 + \frac{u}{2} |b|^4
1185: \Biggr]. \label{scHertz}
1186: \end{eqnarray}
1187: The bare $-i \omega$ term in the boson propagator is not included
1188: above because it is less relevant than the non-analytic term
1189: generated from the Fermi surface excitations. The critical
1190: properties of the $z=3$ critical theory in Eq.~(\ref{scHertz})
1191: have been described earlier by Hertz, and the $T>0$ crossovers by
1192: Millis \cite{millis} (see also Chapter 12 of
1193: Ref.~\onlinecite{book}).
1194:
1195: Next, we consider the critical theory of the 2 FS + BEC to 1 FS +
1196: BEC transition in Fig.~\ref{ZeroTphases}. The same theory also
1197: applies to the 2 FS, no BEC to 1 FS, no BEC transition in
1198: Fig.~\ref{ZeroTmu}. Here, a Fermi surface disappears as its Fermi
1199: wavevector vanishes. The critical theory is then the $z=2$ dilute
1200: Fermi gas theory already discussed in Chapter 11 of
1201: Ref.~\onlinecite{book}. All interactions are irrelevant for the
1202: critical properties, and the quantum-critical crossovers are
1203: merely those of a free Fermi gas.
1204:
1205: Finally, consider the multi-critical point, noted in
1206: Section~\ref{Multi}, where all phases in Fig.~\ref{ZeroTphases}
1207: and Fig.~\ref{ZeroTmu} meet. Here, both the $b$ bosons and the $f$
1208: fermions are critical. The critical theory is merely the direct
1209: sum of the $z=2$ dilute Bose and Fermi theories mentioned above.
1210: All interactions between the critical $f$ and $b$ modes are
1211: formally irrelevant in three spatial dimensions.
1212:
1213: \section{Gaussian corrections}
1214: \label{GaussianCorrections}
1215:
1216: In order to test the validity of the approximations made, we shall calculate two
1217: different corrections to the mean-field results of the preceding sections.
1218: First, in this section, we find the corrections to the grand free energy $\Phi$
1219: to order $g^2$. These will result in corrections to the expressions found in
1220: Section \ref{MFFixedN} relating the chemical potentials to the particle numbers.
1221: Subsequently, in Section \ref{LowDensityApproximation}, we will find a new
1222: expression for $\Delta$ by replacing the mean-field theory with a
1223: low-density approximation.
1224:
1225: In the remainder of this section, we will use $\Xit^b$ from (\ref{EffBoseDisp})
1226: to determine the corrections to the grand free energy $\Phi$. We will show that
1227: these are negligible, provided that the dimensionless coupling $\gamma^2/T_0$
1228: is sufficiently small. We therefore require a narrow Feshbach resonance for the
1229: results to be quantitatively accurate.
1230:
1231: By integrating the
1232: effective action over the boson field, we arrive at an expression for the
1233: partition function including Gaussian corrections,
1234: \be
1235: Z^{(2)} = \frac{(\det \Xi^f) (\det \Xi^\psi)}{(\det \Xit^b)}\punc{.}
1236: \label{StageinF2}
1237: \ee
1238:
1239: The grand free energy is then given by
1240: \be
1241: \Phi = \Phi_0^f + \Phi_0^\psi + \frac{1}{\beta}\sum_p \ln \left(\Xi^b_p +
1242: \frac{g^2}{\beta}\sum_q G^f_q G^\psi_{q+p}\right)\punc{,}
1243: \ee
1244: using (\ref{EffBoseDisp}), where
1245: \be
1246: \Phi^x_0 = \pm\frac{1}{\beta}\sum_q \ln \Xi^x_q
1247: \ee
1248: is the grand free energy for the species $x$ in the absence of coupling. [The
1249: plus (minus) sign applies to bosons (fermions).]
1250:
1251: Taking a factor of $\Xi^b_p$ out of the logarithm gives $\Phi^b_0$, so that the
1252: correction to $\Phi$ is
1253: \be
1254: \Phi - \Phi_0 = \frac{1}{\beta}\sum_p \ln \left(1 + \frac{g^2}{\beta}\sum_q
1255: G^f_q G^\psi_{q+p} G^b_p\right)\punc{.}
1256: \ee
1257: This is the full expression for the Gaussian corrections; to estimate the size
1258: of these corrections, we will calculate the result to order $g^2$. Dropping terms
1259: of higher order gives a correction to the free energy $\Phi$ of
1260: \be
1261: \Delta\Phi = \frac{g^2}{\beta^2}\sum_{p,q}G^f_q G^\psi_{q+p} G^b_p\punc{,}
1262: \ee
1263: which can be represented diagrammatically as
1264: \be
1265: \Delta \Phi \quad = \quad
1266: \parbox{25mm}{
1267: \begin{picture}(25,25)
1268: \put(0,0){\includegraphics{fmfPaper2.ps}}
1269: \input{fmfPaper.t2}
1270: \end{picture}
1271: }\qquad\punc{.}
1272: \ee
1273:
1274: \begin{widetext}
1275: Reinstating explicit momentum integrals and frequency sums, we have
1276: \be
1277: \Delta\Phi=\frac{g^2}{\beta^2}\sum_{\omega_n,\nu_m}\int\frac{d^3
1278: k}{(2\pi)^3}\frac{d^3 \ell}{(2\pi)^3}\:G^b(\omega_n,k)\,G^f(\nu_m,\ell)
1279: \,G^\psi(\nu_m+\omega_n,\ell+k)
1280: \ee
1281: Both Matsubara sums can be performed using contour integration, to give
1282: \be
1283: \Delta \Phi = g^2\,\int\frac{d^3 k}{(2\pi)^3} \int\frac{d^3 \ell}{(2\pi)^3}
1284: \frac{\left[n\sub{F}(\xi^f_{\ell-k})-
1285: n\sub{F}(\xi^\psi_\ell)\right]\left[n\sub{B}(\xi^b_k)-n\sub{B}(\xi^\psi_{\ell} -
1286: \xi^f_{\ell-k})\right]}{\xi^b_k+\xi^f_{\ell-k}-\xi^\psi_{\ell}}\punc{,}
1287: \ee
1288: \end{widetext}
1289: after a change of variables, $\ell\rightarrow\ell-k$.
1290:
1291: \subsection{Renormalization of the detuning}
1292: \label{Renormalize}
1293:
1294: As it stands, the integral over $k$ is in fact divergent. As $|k|$ tends to
1295: infinity (with $|\ell|$ finite), the second Bose factor, $n\sub{B}(\xi^\psi_\ell
1296: - \xi^f_{\ell-k})$, tends to $-1$. In the first bracket,
1297: $n\sub{F}(\xi^\psi_\ell)$ remains finite, so the integrand tends to $\sim 1/k^2$
1298: and the integral over $k$ is linearly divergent.
1299:
1300: This divergence can be understood by considering the self-energy diagram
1301: \be
1302: \label{BubbleDiagram1}
1303: \parbox{35mm}{
1304: \begin{picture}(35,20)
1305: \put(0,0){\includegraphics{fmfPaper3.ps}}
1306: \input{fmfPaper.t3}
1307: \end{picture}
1308: }
1309: \ee
1310: which gives a correction to the detuning $\nu$ linear in the cut-off momentum,
1311: \be
1312: \nu = \nu_0 - g^2 \int\frac{d^3 k}{(2\pi)^3} \frac{2 m^f
1313: m^b}{m^\psi}\frac{1}{k^2}\punc{,}
1314: \ee
1315: where $\nu_0$ is the `bare' detuning that appears explicitly in the
1316: action.
1317:
1318: We use this expression to write $\nu_0$, which appears within $\Xi^\psi$ in
1319: (\ref{StageinF2}), in terms of $\nu$, and then keep terms only of order $g^2$.
1320: The renormalized expression for $\Delta\Phi$ is then given by
1321: \begin{widetext}
1322: \be
1323: \Delta\Phi = g^2\,\int\frac{d^3 k}{(2\pi)^3} \int\frac{d^3 \ell}{(2\pi)^3}
1324: \left\{\frac{\left[n\sub{F}(\xi^f_{\ell-k})-
1325: n\sub{F}(\xi^\psi_\ell)\right]\left[n\sub{B}(\xi^b_k)-n\sub{B}(\xi^\psi_{\ell} -
1326: \xi^f_{\ell-k})\right]}{\xi^b_k+\xi^f_{\ell-k}-\xi^\psi_{\ell}}
1327: +\; \frac{2 m^f m^b}{m^\psi}\frac{n\sub{F}(\xi^\psi_\ell)}{k^2}\right\}\punc{,}
1328: \ee
1329: where the dispersion relation $\xi^\psi$ now involves the renormalized
1330: (physical) detuning $\nu$. (We have retained the same symbols for the new,
1331: renormalized quantities.)
1332:
1333: This expression can be simplified somewhat by performing a further change of
1334: variable, taking $k \rightarrow k + (m^b/m^\psi)\ell$, and also making use of
1335: the result
1336: \be
1337: n\sub{F}(x)n\sub{B}(y-x)+n\sub{F}(y)n\sub{B}(x-y) = -
1338: n\sub{F}(x)n\sub{F}(y)\punc{.}
1339: \ee
1340: We have finally
1341: \be
1342: \Delta \Phi\;= g^2\,\int\frac{d^3 k}{(2\pi)^3} \int\frac{d^3 \ell}{(2\pi)^3}
1343: \frac{1}{\smallfrac{m^\psi}{2m^f m^b}k^2-\nu}
1344: \left[ n\sub{F}(\xi^f)n\sub{B}(\xi^b)-
1345: n\sub{F}(\xi^\psi)n\sub{B}(\xi^b)+n\sub{F}(\xi^f)n\sub{F}(\xi^\psi)-
1346: n\sub{F}(\xi^\psi)\frac{2m^f m^b}{m^\psi}\frac{\nu}{k^2} \right]\punc{,}
1347: \label{FinalDeltaF}
1348: \ee
1349: \end{widetext}
1350: in which the energies $\xi^\psi$, $\xi^f$ and $\xi^b$ should be evaluated at the
1351: following momenta:
1352: \bea
1353: \xi^\psi &\equiv& \xi^\psi(\ell)\\
1354: \xi^f &\equiv& \xi^f(k-\smallfrac{m^f}{m^\psi}\ell)\\
1355: \xi^b &\equiv& \xi^b(k+\smallfrac{m^b}{m^\psi}\ell)\punc{.}
1356: \eea
1357: Note that there is no singularity in the integral over $k$ in
1358: (\ref{FinalDeltaF}), since the numerator also vanishes at the point where
1359: \be
1360: |k|=\sqrt{\nu\frac{2m^f m^b}{m^\psi}}
1361: \ee
1362: (for $\nu > 0$).
1363:
1364: The expression for $\Delta\Phi$ must be differentiated with respect to the
1365: chemical potentials to give the correction to the number of each species of
1366: particle. The resulting integral can then be performed numerically.
1367:
1368: \begin{figure}
1369: \resizebox{\columnwidth}{!}{
1370: \includegraphics{tnua.ftx.eps}
1371: }
1372: \caption{\label{Graph1}Two-loop corrections to particle numbers, shown as a
1373: fraction of the total particle numbers to lowest order. The masses of the
1374: particles and their total numbers are the same as in Fig.~\ref{Graph0},
1375: $N_f/N_b = 1.11$. The dimensionless coupling is
1376: $\gamma^2/T_0 = 2.5 \times 10^{-4}$.
1377: At each temperature, the corrections have been evaluated
1378: taking the detuning at its critical value, from Fig.~\ref{Graph0}.}
1379: \end{figure}
1380: The results of such a calculation are shown in Fig.~\ref{Graph1}, where we plot
1381: the corrections to the particle numbers. These have been divided by the total
1382: numbers evaluated using the results of Section
1383: \ref{MFFixedN}. The atomic masses have been taken to be equal, as in
1384: Fig.~\ref{Graph0}, and, at each temperature value, the detuning takes on its
1385: critical value.
1386:
1387: We have taken for the dimensionless coupling
1388: $\gamma^2/T_0 = 2.5 \times 10^{-4}$, as in Fig.~\ref{Graph0}, so that this
1389: corresponds to a narrow Feshbach resonance. Since the correction
1390: everywhere is less than $5\%$ of the total numbers, we expect that the
1391: mean-field results provide a good approximation for this case. The
1392: magnitude of the correction scales with $g^2\propto \gamma$, so that the
1393: quantitative predictions become less reliable for a broader resonance.
1394:
1395: \section{Low-density approximation}
1396: \label{LowDensityApproximation}
1397:
1398: In this section, we develop a low-density approximation involving all orders in
1399: the coupling $g$.
1400:
1401: \subsection{Diagrammatic description}
1402:
1403: In the mean-field approximation, we have included in the boson self-energy such
1404: diagrams as
1405: \be
1406: \parbox{35mm}{
1407: \begin{picture}(35,20)
1408: \put(0,0){\includegraphics{fmfPaper4.ps}}
1409: \input{fmfPaper.t4}
1410: \end{picture}
1411: }
1412: \ee
1413: whose amplitude is proportional to the density of atoms in the system. (A
1414: fermionic atom must be present in the system initially, in order to couple with
1415: the boson.) Diagrams involving further loops with bosonic or fermionic atoms,
1416: such as
1417: \be
1418: \parbox{35mm}{
1419: \begin{picture}(35,35)
1420: \put(0,0){\includegraphics{fmfPaper5.ps}}
1421: \input{fmfPaper.t5}
1422: \end{picture}
1423: }
1424: \ee
1425: are proportional to higher powers of the density.
1426:
1427: There is, however, a correction to the molecular propagator of order zero in
1428: density coming from the process shown in (\ref{BubbleDiagram1}), which can take
1429: place in the vacuum. To make a consistent low-density approximation, we must
1430: therefore correct the molecular propagator using a Dyson equation
1431: \be
1432: \label{Dyson}
1433: \parbox{15mm}{
1434: \begin{picture}(15,15)
1435: \put(0,0){\includegraphics{fmfPaper6.ps}}
1436: \input{fmfPaper.t6}
1437: \end{picture}
1438: }
1439: \quad=\quad
1440: \parbox{15mm}{
1441: \begin{picture}(15,15)
1442: \put(0,0){\includegraphics{fmfPaper7.ps}}
1443: \input{fmfPaper.t7}
1444: \end{picture}
1445: }\quad+\quad
1446: \parbox{35mm}{
1447: \begin{picture}(35,20)
1448: \put(0,0){\includegraphics{fmfPaper8.ps}}
1449: \input{fmfPaper.t8}
1450: \end{picture}
1451: }
1452: \ee
1453: where only the term in the bubble diagram (\ref{BubbleDiagram1}) of order zero
1454: in density is to be included.
1455:
1456: The boson self-energy diagram
1457: \be
1458: \parbox{35mm}{
1459: \begin{picture}(35,20)
1460: \put(0,0){\includegraphics{fmfPaper9.ps}}
1461: \input{fmfPaper.t8}
1462: \end{picture}
1463: }
1464: \ee
1465: should now be used to give a low-density approximation for the phase transition.
1466:
1467: \subsection{Calculations}
1468: \label{SectionCalculations}
1469:
1470: The Dyson equation (\ref{Dyson}) gives the relation between the reciprocals of
1471: the bare and full Green functions
1472: \be
1473: \Xit^\psi_q = \Xi^\psi_q - \frac{g^2}{\beta} \sum_p G^f_{q-p}G^b_p\punc{;}
1474: \ee
1475: compare (\ref{EffBoseDisp}). [The sign difference results from the fermion loop
1476: in (\ref{FirstGraph}).]
1477:
1478: Including only corrections to order zero in density, by dropping the Bose-
1479: Einstein and Fermi-Dirac factors, gives
1480: \be
1481: \Xit^\psi_q = \Xi^\psi_q - g^2 \int \frac{d^3 k}{(2\pi)^3} \left(
1482: \frac{1}{\xi^b_k + \xi^f_{\ell-k}-i\nu_m}-\frac{2m^f m^b}{m^\psi
1483: k^2}\right)\punc{,}
1484: \ee
1485: where $q \equiv (\ell, \nu_m)$. (The second term in the parentheses comes from
1486: renormalizing the detuning $\nu$, as in Section \ref{Renormalize} above.) The
1487: integral can be performed analytically, to give
1488: \be
1489: \Xit^\psi_q = \Xi^\psi_q + 2\gamma\sqrt{\xi^\psi_\ell-\nu-i\nu_m}\punc{,}
1490: \ee
1491: where $\gamma$, defined in (\ref{Definegamma}), has been used.
1492:
1493: This function can be continued to one that is analytic everywhere except along
1494: the real axis, by replacing $i\nu_m$ by $z$. In terms of $z$, the full Green
1495: function is
1496: \be
1497: \Gt^\psi_\ell(z) = \frac{-1}{z-\xi^\psi_\ell - 2\gamma \sqrt{\xi^\psi_\ell - \nu
1498: - z}}\punc{.}
1499: \ee
1500: Along the real axis, the square root has a branch cut for $z > \xi^\psi_\ell -
1501: \nu$ which corresponds to the continuum of free-atom excitations. For $\nu < 0$,
1502: $\Gt^\psi$ has a single pole at the real value
1503: \be
1504: z_0 = \xi^\psi_\ell - 2\gamma^2\left(1 - \sqrt{1 -
1505: \smallfrac{\nu}{\gamma^2}}\right)\punc{,}
1506: \ee
1507: corresponding to the renormalized molecule. For $\nu > 0$, there are no poles,
1508: since the molecule has a finite lifetime, decaying into two atoms.
1509:
1510: \subsection{Spectral representation}
1511:
1512: These analytical properties are best summarized using the spectral
1513: representation for $\Gt^\psi$,
1514: \be
1515: \Gt^\psi_\ell(z) = \int_{-\infty}^{\infty}dx \,\frac{\rho^\psi_\ell(x)}{x -
1516: z}\punc{.}
1517: \ee
1518: The `spectral density' is given by
1519: \begin{widetext}
1520: \be
1521: \rho^\psi_\ell(x) = \Theta(x-\xi^\psi_\ell + \nu) \:
1522: \frac{2\gamma}{\pi}\,\frac{\sqrt{x-\xi^\psi_\ell + \nu}}{{\left(x-
1523: \xi^\psi_\ell\right)}^2 + 4\gamma^2 \left(x-\xi^\psi_\ell +
1524: \nu\right)}
1525: + \;\Theta(-\nu) \frac{\sqrt{1 - \smallfrac{\nu}{\gamma^2}} - 1}{\sqrt{1 -
1526: \smallfrac{\nu}{\gamma^2}}}\:\delta\!\left(x-\xi^\psi_\ell-
1527: 2\gamma^2\left(\sqrt{1 - \smallfrac{\nu}{\gamma^2}} - 1\right)\right)\punc{,}
1528: \label{SpectralDensity}
1529: \ee
1530: \end{widetext}
1531: where $\Theta$ is the unit step function and $\delta$ is the Dirac delta
1532: function.
1533:
1534: \subsubsection{Weak-coupling limit}
1535:
1536: It should be noted that, in the limit of small coupling, $\gamma\rightarrow 0$,
1537: the first term of (\ref{SpectralDensity}) involves the Lorentzian representation
1538: of the Dirac delta function,
1539: \be
1540: \lim_{\varepsilon\rightarrow 0} \frac{\varepsilon}{\pi(t^2+\varepsilon^2)} =
1541: \delta(t)\punc{.}
1542: \ee
1543: For $\gamma$ small enough, the first term of (\ref{SpectralDensity}) has weight
1544: only near $x = \xi^\psi_\ell$, where $x-\xi^\psi_\ell+\nu$ can be replaced by
1545: $\nu$. The limit of vanishing $\gamma$ is therefore given by
1546: \be
1547: \lim_{\gamma\rightarrow 0} \frac{2\gamma}{\pi}\,\frac{\sqrt{x-\xi^\psi_\ell +
1548: \nu}}{{\left(x-\xi^\psi_\ell\right)}^2 + 4\gamma^2 \left(x-\xi^\psi_\ell +
1549: \nu\right)} = \delta(x - \xi^\psi_\ell)\punc{,}
1550: \ee
1551: so that the spectral density becomes, in this limit,
1552: \bea
1553: \rho^\psi_\ell(x) &\rightarrow& \Theta(x-\xi^\psi_\ell +
1554: \nu)\delta(x - \xi^\psi_\ell) + \Theta(-\nu)\delta(x-\xi^\psi_\ell)\nonumber\\
1555: &=& \delta(x-\xi^\psi_\ell)
1556: \punc{,}
1557: \eea
1558: which is precisely the result for the bare molecule, used in Section
1559: \ref{EffActBoson}.
1560:
1561: \subsection{Effective action for bosons}
1562:
1563: Using the modified spectral density for the molecule, (\ref{SpectralDensity}),
1564: we can compute the quadratic term in the effective action for the boson field.
1565: Following Section \ref{MF2}, the modified coefficient is
1566: \be
1567: \DeltaLDA = -\mu^b + g^2\int\frac{d^3 \ell}{(2\pi)^3}
1568: \frac{1}{\beta}\sum_{\nu_m} G^f_\ell(i\nu_m)\,\Gt^\psi_\ell(i\nu_m)\punc{.}
1569: \ee
1570: We can now use the spectral representation for both Green functions and then
1571: perform the Matsubara sum. Since the spectral density for the $f$ atom has the
1572: form
1573: \be
1574: \rho^f_\ell(x) = \delta(x - \xi^f_\ell)\punc{,}
1575: \ee
1576: the low-density approximation to the discriminant is given by
1577: \be
1578: \label{LDADelta}
1579: \DeltaLDA = -\mu^b + g^2\int\frac{d^3 \ell}{(2\pi)^3} \,\int_{-
1580: \infty}^{\infty}dx\,\rho^\psi_\ell(x)\frac{n\sub{F}(x)-n\sub{F}(\xi^f_\ell)}{x-
1581: \xi^f_\ell}\punc{,}
1582: \ee
1583: with $\rho^\psi_\ell(x)$ given by (\ref{SpectralDensity}), above. This integral
1584: can be performed numerically.
1585:
1586: Since $\DeltaLDA$ provides a correction to $\Delta$, we must find some measure
1587: by which to determine the significance of this correction. A comparison with
1588: $\Delta$ is obviously not possible, since this vanishes everywhere along the
1589: mean-field curve in Fig.~\ref{Graph0}. Instead, we shall find a lowest-order
1590: correction to the critical detuning by evaluating
1591: \be
1592: \label{Definedeltanu}
1593: \delta\nu \equiv \left(\DeltaLDA -
1594: \Delta\right)\Big{/}\parderat{\Delta}{\nu}{\beta,N_f,N_b}\punc{.}
1595: \ee
1596: (Finding the curve $\DeltaLDA = 0$ exactly would require a much larger
1597: computational effort, since $\DeltaLDA$ takes considerably more time to evaluate
1598: than $\Delta$. Instead, $\delta\nu$ provides the first step of a solution of
1599: $\DeltaLDA = 0$ by Newton's iterative method.)
1600:
1601: The partial derivative on the right-hand side of (\ref{Definedeltanu}) is given
1602: by
1603: \begin{widetext}
1604: \be
1605: \parderat{\Delta}{\nu}{N_f,N_b} = \quad\parderat{\Delta}{\nu}{\mu^f,\mu^b} +
1606: \parderat{\Delta}{\mu^f}{\mu^b,\nu}\parderat{\mu^f}{\nu}{N_f,N_b}
1607: + \parderat{\Delta}{\mu^b}{\mu^f,\nu}\parderat{\mu^b}{\nu}{N_f,N_b}\punc{,}
1608: \ee
1609: \end{widetext}
1610: where all the derivatives are to be taken at constant $\beta$. To lowest order
1611: in the coupling, this can be replaced by
1612: \be
1613: \parderat{\Delta}{\nu}{N_f,N_b} = \quad\parderat{\mu^b}{\nu}{N_f,N_b} +
1614: \Order{g^2}\punc{,}
1615: \ee
1616: using (\ref{DiscriminantIntegral}). The quantity on the right-hand side is the
1617: increase in the boson chemical potential needed to compensate an increase in the
1618: detuning, and keep the particle numbers unchanged. This can be evaluated using
1619: the expressions given in Section \ref{MFFixedN}.
1620:
1621: The results of such a calculation are shown in Fig.~\ref{Graph2}. The solid line
1622: is the phase boundary shown in Fig.~\ref{Graph0}, using the same parameters.
1623: The dashed line is the same curve with the quantity $\delta\nu$ from
1624: (\ref{Definedeltanu}) added. (For clarity, we have multiplied $\delta\nu$ by a
1625: factor of $3$.)
1626: \begin{figure}
1627: \resizebox{\columnwidth}{!}{
1628: \includegraphics{tnub.ftx.eps}
1629: }
1630: \caption{\label{Graph2}The phase boundary, without (solid line) and with (dashed
1631: line) the correction to the detuning of Section \ref{LowDensityApproximation}.
1632: The solid line is the phase boundary as in Fig.~\ref{Graph0}, using the same
1633: parameters (and dimensionless coupling $\gamma^2/T_0 = 2.5 \times 10^{-4}$). The
1634: dashed line includes the correction $\delta\nu$, from
1635: (\ref{Definedeltanu}), found using a low-density approximation. For clarity,
1636: this correction has been exaggerated by a factor of $3$. Since the correction is
1637: negligible for these parameters, the mean-field results of Sections
1638: \ref{MFforMu} and \ref{EffActBoson} are sufficient.}
1639: \end{figure}
1640: The small magnitude of the correction suggests that the results presented in
1641: Sections \ref{MFforMu} and \ref{EffActBoson} are valid, for the parameters
1642: chosen. In this case, it is therefore unnecessary to use a low-density
1643: approximation; the mean-field result is sufficient.
1644:
1645: \section{Conclusions}
1646: \label{sec:conc}
1647:
1648: From a theoretical perspective, the main contribution of this
1649: paper is a description of a quantum phase transition with an
1650: intimate connection to the Luttinger theorem. On one side of the
1651: transition (see Fig.~\ref{ZeroTphases}), in the 2 FS, no BEC phase,
1652: there are 2 Fermi surfaces with a separate Luttinger theorem for
1653: each Fermi surface. Remarkably, one Fermi surface is constrained
1654: by the total number of {\em bosons\/} $N_b$, while the other is
1655: controlled by $N_f-N_b$, where $N_f$ is total number of fermions.
1656: These two Luttinger theorems are consequences of two number
1657: operators that commute with the Hamiltonian: the operator
1658: $f^\dagger f - b^\dagger b$ in Eq.~(\ref{const1}), and the
1659: operator $f^\dagger f + \psi^\dagger \psi$ in Eq.~(\ref{const2}).
1660: On the other side of the quantum critical point is the 2 FS + BEC
1661: phase. Here a Bose-Einstein condensate is present, and the
1662: condensate effectively thwarts one of the Luttinger constraints.
1663: The single remaining Luttinger theorem demands only that the total
1664: volume enclosed by both Fermi surfaces is constrained by $N_f$. We
1665: presented a theory for this transition, along with phase diagrams
1666: as a function of system parameters.
1667:
1668: It is intriguing to note a connection between the above quantum
1669: phase transition and a seemingly disconnected, recent analysis of
1670: a quantum phase transition in a Kondo lattice model of the heavy
1671: fermion compounds.\cite{ssv} This was a model of electrons
1672: occupying localized $f$ orbitals (the $f_\sigma$ electrons, where
1673: $\sigma$ is a spin index) interacting the itinerant electrons in
1674: the conduction band (the $c_\sigma$ electrons). A boson, $b$, was
1675: introduced to represent the hybridization between the orbitals.
1676: The connection between the Kondo lattice model and the model of
1677: the present paper now becomes clear once we identify the
1678: $c_\sigma$ electrons with the molecular fermionic state $\psi$,
1679: and the $f_\sigma$ electrons with the $f$ fermions. The Kondo
1680: lattice model also has two number constraints analogous to
1681: Eq.~(\ref{const1}) and Eq.~(\ref{const2}), with one crucial
1682: difference: the first constraint is local rather than global, and
1683: applies separately on each lattice site. With this mapping, the
1684: heavy Fermi liquid FL state of Ref.~\onlinecite{ssv} can be
1685: identified with the 2 FS + BEC and the 1 FS + BEC phases. Further,
1686: the FL* phase of Ref.~\onlinecite{ssv} is the analog of the
1687: present 2 FS, no BEC phase. The FL* phase also has two Luttinger
1688: theorems, one fixing the volume of the conduction band Fermi
1689: surface of electronic quasiparticles, and the other the volume of
1690: the `spinon' Fermi surface. The presence of a local rather than a global
1691: constraint implies that there is an additional gauge force that
1692: affects the spinon Fermi surface and the quantum critical
1693: fluctuations of the Kondo lattice. Such gauge forces are absent in
1694: our present considerations of Bose-Fermi mixtures, but, apart from
1695: this absence, there is a remarkable similarity to the FL-FL*
1696: transition in Kondo lattice models.
1697:
1698: On the experimental front, an obvious signature of the quantum
1699: phase transition in the Bose-Fermi mixtures is in the evolution of
1700: the Bose-Einstein condensate. It would be interesting to scan the
1701: detuning and look for the disappearance of the condensate fraction
1702: at the lowest temperatures. The corresponding
1703: ``superfluid-normal'' transition should also survive at $T>0$,
1704: where its signatures are similar to the $\lambda$ transition in
1705: $^4$He.
1706:
1707: A more dramatic, but experimentally less accessible, signature of
1708: the transition lies in the values of the Fermi wavevectors, as
1709: sketched in Fig.~\ref{ZeroTk}. Measuring the Fermi wavevectors
1710: would allow detection of a Fermi surface constrained by the number
1711: of bosons, and its eventual evolution across the transition to a
1712: Fermi surface constrained by the total number of fermions.
1713:
1714: Finally, it should be noted that we have not addressed here the
1715: alternative of a paired state of fermions. Since we have dealt
1716: with spin-polarized fermions, $s$-wave pairing between the atoms
1717: is excluded, but $p$-wave pairing remains a possibility.
1718: \cite{gurarie,ho} There is
1719: also the more novel possibility for pairing between the fermionic
1720: atoms $f$ and the molecules $\psi$, which could be favorable when
1721: the two Fermi wavenumbers are approximately equal. This would then
1722: lead to condensation of a composite boson comprised of two fermionic
1723: atoms and one bosonic atom. We intend to investigate this
1724: possibility further in future work.
1725:
1726: \acknowledgments We thank R.~Hulet and W.~V.~Liu
1727: for valuable discussions. This research was supported by the
1728: National Science Foundation under grant DMR-0098226, and under
1729: grant DMR-0210790. S.S. was also supported by the John Simon
1730: Guggenheim Memorial Foundation.
1731:
1732: \appendix*
1733:
1734: \section{Stability against phase separation}
1735: \label{PhaseSeparation}
1736:
1737: This section uses the mean-field results of Section \ref{MFforMu}. The
1738: temperature will be taken as zero throughout.
1739:
1740: \subsection{The compressibility matrix}
1741:
1742: To establish the stability of the system against separation into two coexisting
1743: fluids, we evaluate the compressibility matrix, defined by
1744: \be
1745: \label{Kprimedefined}
1746: K_{\alpha\beta}' = -\frac{\partial^2
1747: \Phi}{\partial\mu^\alpha\partial\mu^\beta}\punc{,}
1748: \ee
1749: for $\alpha,\beta\in\{f,b\}$.
1750:
1751: We now define the (canonical) free energy $F(N_f,N_b)$ by a Legendre
1752: transformation,
1753: \be
1754: F(N_f,N_b) = \Phi(\mu^f,\mu^\psi) + \mu^f N_f + \mu^b N_b\punc{,}
1755: \ee
1756: where $N_f$ and $N_b$ are the total number of Fermi and Bose atoms,
1757: respectively. (Note that the full fermion and boson numbers, which are conserved
1758: by the Hamiltonian, are used.) The compressibility matrix $K'$ is then the
1759: inverse of the Hessian of $F$, so that complete stability against phase
1760: separation requires that $K'$ be positive semidefinite.
1761:
1762: It is in fact easier to work with the matrix $K_{\alpha\beta}$, given by the
1763: same expression (\ref{Kprimedefined}), but with $\alpha,\beta\in\{f,\psi\}$.
1764: This amounts to a simple (but not orthogonal) change of basis; it is sufficient
1765: (and necessary) for $K'$ to be positive semidefinite that $K$ be the same.
1766:
1767: We begin with (\ref{FreeEnergy}) and (\ref{DefineR}) and use (\ref{EqCondensate})
1768: to determine the implicit dependence of $\varphi$ on the chemical potentials.
1769: We must then take second derivatives with respect to the two chemical potentials
1770: to find the compressibility matrix. In the presence of a condensate, this leads
1771: to an expression
1772: \be
1773: K_{\alpha\beta} = K_{\alpha\beta}^{(0)} + \frac{r_\alpha r_\beta}{\lamt}\punc{,}
1774: \ee
1775: where $K_{\alpha\beta}^{(0)}$ is the matrix of second derivatives, evaluated at
1776: fixed $\varphi$ and $r_\alpha$ is a function whose form will not concern us here.
1777:
1778: The denominator of the second term is
1779: \be
1780: \label{resultantinteraction}
1781: \lamt = \lambda + g^4\int_{k_0^F}^{k_0^\Psi}\frac{dn(k)}{W_k^3}
1782: - \half g^4 \left(z^F + z^\Psi\right)\punc{,}
1783: \ee
1784: where
1785: \be
1786: \label{Definez}
1787: z^x = {\left.\frac{m^\psi m^f dn/dk}{k\left(\xi^f_k +
1788: \xi^\psi_k\right)\left(m^f\xi^f_k +
1789: m^\psi\xi^\psi_k\right)}\right|}_{k=k_0^x}
1790: \ee
1791: and
1792: \be
1793: W_k = \sqrt{{\left(\xi^f_k-\xi^\psi_k\right)}^2 + 4g^2\varphi^2}\punc{.}
1794: \ee
1795: When $\lamt$ goes through zero, the determinant of $K$ diverges, so that the
1796: Hessian of $F$ becomes singular, signifying that one of its eigenvalues vanishes.
1797: This marks the onset of instability; we conclude that stability requires that
1798: $\lamt > 0$.
1799:
1800: When there is no condensate, \ie\ in the phase labeled `2 FS, no BEC' in
1801: Fig.~\ref{ZeroTphases}, it is found that the system is always stable.
1802:
1803: \subsection{Physical interpretation}
1804:
1805: The obvious physical interpretation of $\lamt$ is that it represents the resultant
1806: interaction between the bosons, coming partly from the explicit term $\lambda$ in
1807: the Hamiltonian (\ref{Hamiltonian}) and partly from the interaction induced by
1808: coupling to the fermions. This induced interaction can alternatively be found
1809: directly by continuing the expansion (\ref{QuadAction}) to fourth order in $b$
1810: and $\bar{b}$.
1811:
1812: A resultant interaction of the form (\ref{resultantinteraction}) is familiar
1813: from the case where the molecular degrees of freedom are not included explicitly
1814: in the Hamiltonian.\cite{Buechler,Viverit} This corresponds to our model for
1815: $\nu\gg 0$, when only virtual molecules are formed and the coupling
1816: term $\psi^\dagger f b$ in the Hamiltonian (\ref{Hamiltonian}) can be replaced by
1817: a boson-fermion scattering of the form $f^\dagger b^\dagger b f$. The induced
1818: interaction then comes from the diagram
1819: \be
1820: \parbox{35mm}{
1821: \begin{picture}(35,20)
1822: \put(0,0){\includegraphics{appendix1.ps}}
1823: \input{appendix.t1}
1824: \end{picture}
1825: }
1826: \punc{,}
1827: \label{Fish}
1828: \ee
1829: which gives a term proportional to the density of states at the Fermi surface (at
1830: $T = 0$).
1831:
1832: In this case, the induced interaction is always attractive, as can be shown by a
1833: simple physical argument. For experimentally accessible parameters, however, it
1834: is not strong enough to overcome the intrinsic repulsion between the bosons, so
1835: that the phase is stable.\cite{Viverit} In our notation, the boson-fermion
1836: scattering is suppressed by a factor of $1/\nu$, so that the induced interaction
1837: falls off as $1/\nu^2$. For $\nu\ll 0$, a similar picture is obtained, with the
1838: atomic and molecular fermions exchanging r\^oles.
1839:
1840: In the case of intermediate $\nu$, the induced interaction is no longer so heavily
1841: suppressed, but it is also no longer the case that it is always attractive. The
1842: physical picture is clarified in this case by rewriting the action in
1843: (\ref{eqAction}) in terms of the fermions $F$ and $\Psi$ introduced in Section
1844: \ref{MFHamiltonian}. These fermions are defined so that there is no coupling term
1845: in the action linear in $\varphi = \Mean{b}$; instead, the lowest order interactions
1846: have the form $\bar{F}\,\varphi^2\,F$ and $\bar{F}\,\varphi^4\,F$,
1847: and the same for $\Psi$. The former reproduces exactly the diagram (\ref{Fish})
1848: above, with $f$ replaced by $F$ and $\Psi$: physically this is a boson-fermion
1849: scattering inducing an attractive interaction between the bosons, as described
1850: above. This accounts for the final term in (\ref{resultantinteraction}). Note that
1851: the exclusion principle requires the momenta of the two fermion lines to be exactly
1852: at the Fermi surface, leading to $z^x$ being evaluated at $k_0^x$.
1853:
1854: The term $\bar{F}\,\varphi^4\,F$ produces the diagram
1855: \be
1856: \parbox{25mm}{
1857: \begin{picture}(25,25)
1858: \put(0,0){\includegraphics{appendix2.ps}}
1859: \input{appendix.t2}
1860: \end{picture}
1861: }
1862: \label{Spider}
1863: \ee
1864: which also represents an induced boson-boson interaction and accounts for the
1865: integral in (\ref{resultantinteraction}). Since $W_k \ge 0$, it is always repulsive
1866: and represents the fact that the fermion energy is lowered by a uniform
1867: distribution of bosons.
1868:
1869: \subsection{Results}
1870:
1871: The sign of the resultant interaction $\lamt$ must be calculated numerically to
1872: determine whether the system is indeed stable. Using the parameters from
1873: Fig.~\ref{ZeroTphases}, stability is found everywhere within the plot for cases
1874: (a) and (b). In case (c), where the coupling $g$ is larger relative to $\lambda$,
1875: there is a region of the diagram where the phase is not stable; this is shown
1876: in Fig.~\ref{ZeroTstability}.
1877: \begin{figure}
1878: \resizebox{\columnwidth}{!}{
1879: \includegraphics{ZeroTstability.ftx.eps}
1880: }
1881: \caption{\label{ZeroTstability}The phase diagram at $T = 0$, as in
1882: Fig.~\ref{ZeroTphases}, with couplings $\gamma^2/T_0 = 2.0\times 10^{-2}$ and
1883: $\lambda^2 (m^b)^3 T_0 = 2\times 10^{-3}$.
1884: The other parameters, and the labels for the three phases, are the same as
1885: in Fig.~\ref{ZeroTphases}. The region where the phase is unstable, as determined
1886: in the Appendix, is indicated.}
1887: \end{figure}
1888:
1889: For large $|\nu|$ the attractive coupling from the diagram (\ref{Fish}) is
1890: suppressed by a factor $1/\nu^2$ as described above, so that the system becomes
1891: stable. (The region for large negative $\nu$ is not visible on this plot.) For
1892: intermediate values of $|\nu|$, the induced coupling becomes larger than the
1893: intrinsic coupling, $\lambda$, and it is the competition between the two diagrams
1894: (\ref{Fish}) and (\ref{Spider}) that determines the stability.
1895:
1896: Stability is therefore favored by a higher $N_f/N_b$, since this increases
1897: $k_0^\Psi$ and hence the phase space for the diagram (\ref{Spider}). The other
1898: diagram, (\ref{Fish}), increases more slowly with $k_0^\Psi$ since the internal
1899: fermion lines are restricted to be at the Fermi surface. For intermediate $|\nu|$
1900: and very small $N_f/N_b$, on the order of $10^{-3}$, the instrinsic interaction once
1901: more dominates the induced and the system is stable. This region is too small to
1902: be seen in Fig.~\ref{ZeroTstability}.
1903:
1904: An analysis similar to that carried out in Ref.~\onlinecite{Viverit} could be
1905: performed to determine the stabilities of the alternative, mixed phases. It should
1906: be noted, however, that, as can be seen in Fig.~\ref{ZeroTstability}, the
1907: boundaries between the three phases are not disturbed
1908: at the parameters we have considered.
1909:
1910: Furthermore, the analysis above shows that increasing the coupling
1911: $g$ (or equivalently $\gamma^2/T_0$) beyond the value used in
1912: Fig.~\ref{ZeroTstability} would increase the value of $|\nu|$ required
1913: for stability at small $N_f/N_b$ (\ie\ extend the unstable region
1914: to larger $|\nu|$), but would not decrease the stability at
1915: intermediate $|\nu|$. This follows from the fact that the latter
1916: is determined by the competition between the two diagrams
1917: (\ref{Fish}) and (\ref{Spider}), whose relative magnitude does not
1918: depend on $g$. We therefore expect that, for a broad Feshbach resonance,
1919: there remains a large region of stability for intermediate values of
1920: $|\nu|$, similar to that in Fig.~\ref{ZeroTstability}.
1921:
1922: \begin{thebibliography}{99}
1923:
1924: \bibitem{stwalley1976} W.\ C.\ Stwalley, Phys.\ Rev.\ Lett.\ {\bf 37}, 1628 (1976).
1925:
1926: \bibitem{tiesinga1993} E.\ Tiesinga, B.\ J.\ Verhaar, and H.\ T.\ C.\ Stoof,
1927: Phys.\ Rev.\ A {\bf 47}, 4114 (1993).
1928:
1929: \bibitem{jochim2003} S.\ Jochim, M.\ Bartenstein, A.\ Altmeyer, G.\ Hendl, S.\ Riedl,
1930: C.\ Chin, J.\ Hecker Denschlag, and R.\ Grimm,
1931: Science {\bf 302}, 2101 (2003).
1932:
1933: \bibitem{greiner2003} M.\ Greiner, C.\ A.\ Regal, and D.\ S.\ Jin, Nature (London) {\bf 426}, 537
1934: (2003).
1935:
1936: \bibitem{zwierlein2003} M.\ W.\ Zwierlein, C.\ A.\ Stan, C.\ H.\ Schunck,
1937: S.\ M.\ F.\ Raupach, S.\ Gupta, Z.\ Hadzibabic, and W.\ Ketterle,
1938: Phys.\ Rev.\ Lett.\ {\bf 91}, 250401 (2003).
1939:
1940: \bibitem{leo} L.\ Radzihovsky, J.\ Park, and P.\ B.\ Weichman, Phys.\ Rev.\ Lett.\ {\bf 92}, 160402 (2004).
1941:
1942: \bibitem{ss} M.\ W.\ J.\ Romans, R.\ A.\ Duine, S.\ Sachdev, and H.\ T.\ C.\ Stoof,
1943: Phys.\ Rev.\ Lett.\ {\bf 93}, 020405 (2004).
1944:
1945: \bibitem{truscott} A.\ G.\ Truscott, K.\ E.\ Strecker, W.\ I.\ McAlexander,
1946: G.\ B.\ Partridge, and R.\ G.\ Hulet, Science {\bf 291}, 2570 (2001).
1947:
1948: \bibitem{schreck} F.\ Schreck, L.\ Khaykovich, K.\ L.\ Corwin, G.\ Ferrari, T.\ Bourdel,
1949: J.\ Cubizolles, and C.\ Salomon, Phys.\ Rev.\ Lett.\ {\bf 87}, 080403
1950: (2001).
1951:
1952: \bibitem{stan} C.\ A.\ Stan, M.\ W.\ Zwierlein, C.\ H.\ Schunck, S.\ M.\ F.\ Raupach, and
1953: W.\ Ketterle, Phys.\ Rev.\ Lett.\ {\bf 93}, 143001 (2004).
1954:
1955: \bibitem{inouye} S.\ Inouye, J.\ Goldwin, M.\ L.\ Olsen, C.\ Ticknor, J.\ L.\ Bohn, and
1956: D.\ S.\ Jin, Phys.\ Rev.\ Lett.\ {\bf 93}, 183201 (2004).
1957:
1958: \bibitem{fwgf} M.\ P.\ A.\ Fisher, P.\ B.\ Weichman, G.\ Grinstein, and
1959: D.\ S.\ Fisher, Phys.\ Rev.\ B {\bf 40}, 546 (1989).
1960:
1961: \bibitem{book} S.\ Sachdev, {\em Quantum Phase Transitions},
1962: Cambridge University Press, Cambridge (1999).
1963:
1964: \bibitem{Yabu}
1965: H.\ Yabu, Y.\ Takayama, and T.\ Suzuki, Physica B {\bf 329-333}, 25
1966: (2003).
1967:
1968: \bibitem{Duine}
1969: R.\ A.\ Duine and H.\ T.\ C.\ Stoof, Phys.\ Rep.\ {\bf 396}, 115 (2004).
1970:
1971: \bibitem{ohashi:130402}
1972: Y.\ Ohashi and A.\ Griffin, Phys.\ Rev.\ Lett.\ {\bf 89}, 130402
1973: (2002).
1974:
1975: \bibitem{luttingerward} J.\ M.\ Luttinger and J.\ C.\ Ward,
1976: Physical Review {\bf 118}, 1417 (1960).
1977:
1978: \bibitem{potthoff} M.\ Potthoff, cond-mat/0406671 (2004).
1979:
1980: \bibitem{agd} A.\ A.\ Abrikosov, L.\ P.\ Gorkov, and I.\ E.\ Dzyaloshinski,
1981: {\em Methods of Quantum Field Theory in Statistical Physics}, Dover Publications Inc.,
1982: New York (1975).
1983:
1984: \bibitem{hertz} J.\ A.\ Hertz, Phys.\ Rev.\ B {\bf 14}, 1165 (1976).
1985:
1986: \bibitem{millis} A.\ J.\ Millis, Phys.\ Rev.\ B {\bf 48}, 7183 (1993).
1987:
1988: \bibitem{ssv} T.\ Senthil, S.\ Sachdev, and M.\ Vojta, Phys.\ Rev.\ Lett.\ {\bf 90},
1989: 216403 (2003); T.\ Senthil, M.\ Vojta, and
1990: S.\ Sachdev, Phys.\ Rev.\ B {\bf 69}, 035111 (2004).
1991:
1992: \bibitem{gurarie} V.\ Gurarie, L.\ Radzihovksy, and A.\ V.\ Andreev, cond-mat/0410620 (2005).
1993:
1994: \bibitem{ho} T.-L.\ Ho and R.\ B.\ Diener, cond-mat/0408468 (2004).
1995:
1996: \bibitem{Viverit} L.\ Viverit, C.\ J.\ Pethick, and H.\ Smith, Phys.\ Rev.\ A {\bf 61},
1997: 053605 (2000).
1998:
1999: \bibitem{Buechler} H.\ P.\ B\"uchler and G.\ Blatter, Phys.\ Rev.\ A {\bf 69},
2000: 063603 (2004).
2001:
2002: \end{thebibliography}
2003:
2004: \end{document}
2005: