astro-ph0605409/R.tex
1: \documentclass[prd,aps,floats,floatfix,eqsecnum,nofootinbib]{revtex4}
2: \usepackage{array,amsmath,amssymb,verbatim}
3: \usepackage{bm}% bold math
4: \usepackage{graphicx}% Include figure files
5: \usepackage{psfrag}% use latex text in ps figures
6: \newcommand{\avg}[1]{\langle #1 \rangle}
7: \newcommand{\Avg}[1]{\Big\langle #1 \Big\rangle}
8: \newcommand{\bds}[1]{\boldsymbol{#1}}
9: 
10: \newcommand{\E}{{\cal E}}
11: \newcommand{\hphi}{{\hat\phi}}
12: \newcommand{\dm}{\partial_\mu}
13: \newcommand{\dr}{\frac{\partial}{\partial r}}
14: \newcommand{\ddr}{\frac{\partial^2}{\partial r^2}}
15: \newcommand{\udm}{\partial^\mu}
16: \newcommand{\dk}{\frac{d^3k}{(2\pi)^3}}
17: 
18: \newcommand{\vR}{{\bds r}}
19: \newcommand{\vx}{{\bds x}}
20: \newcommand{\vn}{{\bds n}}
21: \newcommand{\vk}{{\bds k}}
22: \newcommand{\vp}{{\bds p}}
23: \newcommand{\vq}{{\bds q}}
24: \newcommand{\vs}{{\bds\sigma}}
25: \newcommand{\vt}{{\bds\tau}}
26: \def\be{\begin{equation}}
27: \def\ee{\end{equation}}
28: \newcommand{\bea}{\begin{eqnarray}}
29: \newcommand{\eea}{\end{eqnarray}}
30: \begin{document}
31: \title{Dilute and Collapsed Phases of the Self-Gravitating Gas}
32: \author{C. Destri$^{(a)}$}
33: \email{Claudio.Destri@mib.infn.it}
34: \author{H. J. de Vega$^{(b)}$}
35: \email{devega@lpthe.jussieu.fr} \affiliation{
36: $^{(a)}$ Dipartimento di Fisica G. Occhialini, Universit\`a
37: Milano-Bicocca Piazza della Scienza 3, 20126 Milano and
38: INFN, sezione di Milano, via Celoria 16, Milano Italia\\
39: $^{(b)}$ LPTHE, Universit\'e Pierre~et~Marie Curie,~~Paris VI et
40: Denis Diderot, Paris VII, Laboratoire Associ\'e au CNRS UMR 7589,
41: Tour 24, 5\`eme.~~\'etage, 4, Place
42: Jussieu, 75252 Paris, Cedex 05, France}
43: \affiliation{Observatoire de Paris, LERMA.
44: Laboratoire Associ\'e au CNRS UMR 8112.
45:  \\61, Avenue de l'Observatoire, 75014 Paris, France.}
46: \begin{abstract}
47:   The self-gravitating gas in thermal equilibrium is studied using a
48:   Newtonian potential regularized at short distances. This short distance
49:   cutoff permits us to obtain a complete description of the gas including
50:   its collapsed phase. We give a field theory description of the $N$-body
51:   regularized self-gravitating gas in the canonical ensemble.  The
52:   corresponding functional integral is dominated in the $ N \to \infty $
53:   limit by saddle points which provide a mean field description. The
54:   well-known dilute solutions (isothermal spheres) are recovered. We find
55:   new solutions which are regular in the regularized theory but become
56:   singular in the zero cutoff limit. They describe collapsed configurations
57:   where the particles are densely concentrated in a region of the size of
58:   the cutoff. These collapsed solutions provide the absolute minimum of
59:   the free energy. We find further new solutions which interpolate
60:   between the collapsed and the dilute configurations and describe
61:   tunneling processes where the gas collapses. The transition probability
62:   for such collapse processes turns out to be extremely small for large
63:   $N$.  That is, the dilute solutions are in practice stable in the regime
64:   where they are locally stable.
65: \end{abstract}
66: \date{\today}
67: \maketitle 
68: \tableofcontents
69: 
70: \section{Introduction}
71: Selfgravitating system are ubiquous in the universe. Selfgravity governs
72: the formation and evolution of the large scale structure in the universe
73: as well as the dynamics of galaxy formation,
74: stars and interstellar medium \cite{astro,ism}.
75: It is therefore of great interest to study self-gravitating fluids in
76: thermal equilibrium although most astrophysical and cosmological
77: systems are not in exact thermal equilibrium \cite{astro}-\cite{nos}. 
78: 
79: The self-gravitating gas has peculiar properties from the point of view of
80: equilibrium statistical mechanics: the most important configurations are
81: inhomogeneous and the thermodynamic functions {\bf exist} in the {\bf
82:   dilute} limit \cite{I,II,pal}
83: \begin{equation}\label{limter}
84:   N\to \infty\; ,\quad V \to \infty\; ,\quad
85:   \frac{N}{V^{1/3}} \;\mbox{fixed} \; ,
86: \end{equation}
87: where $N$ is the number of particles and $ V $ stands for the volume of the
88: box containing them. In such a limit, the internal energy, the free energy
89: and the entropy turns to be extensive. That is, they take the form of $ N $
90: times a function of intensive dimensionless variables. In the
91: case of a spherical box of radius $R$ (so that $V=4\pi R^3/3$) there is
92: only one such variable, that is the ratio of the characteristic
93: gravitational energy $G m^2 N/R$ ($m$ is the particle mass and $G$
94: is Newton's constant) and the kinetic energy of the order of the temperature
95: $ T $ of a particle in the gas \cite{I,II,pal}:
96: \begin{equation*}
97: \eta = \frac{G \, m^2 N}{R \; T}  \;.
98: \end{equation*}
99: For more complex geometries there will be, besides $\eta$ (in such cases
100: $R$ is to be identified with the linear scale of the box), other
101: dimensionless shape parameters.
102: 
103: In this paper we restrict the analysis to the simplest case of a spherical
104: box and follow the approach of refs.\cite{I,II,pal} where the starting
105: point is the partition function for non-relativistic particles interacting
106: through their gravitational attraction in thermal equilibrium. The
107: configurational part of this partition function is expressed as a
108: functional integral over a scalar field $\phi(\vx)$ [proportional to the
109: the gravitational potential] with an Euclidean action (that is minus the
110: logarithm of statistical weight) proportional to $N$. Therefore, the saddle
111: point technique (or mean field method) permits to compute the configuration
112: partition function in the large $ N $ limit.
113: 
114: As investigated in refs.\cite{viej,I,II,pal} the self-gravitating gas in the
115: canonical ensemble is in a gaseous phase for $ 0 \leq \eta < \eta_T $ where
116: $ \eta_T = 2.43450\ldots $. (Further studies on self-gravitating gases are reported
117: in ref. \cite{viej2,nos,otros,cha}). At this point the isothermal compressibility
118: diverges and the speed of sound inside the sphere becomes imaginary.
119: This clearly shows that the gas phase collapses at this point in the
120: canonical
121: ensemble as a consequence of the Jeans instability. Monte Carlo
122: simulations explicitly show that the gas collapses for $ \eta \gtrsim
123: \eta_T $ into a very dense and compact phase\cite{I}.
124: 
125: It must be recalled that the small fluctuations around the mean field
126: solution in the canonical ensemble are stable for  $ \eta < \eta_C = 2.517551
127: \ldots $\cite{II}. Namely, fluctuations in the partition function do not
128: feel the divergence of the isothermal compressibility neither
129: the fact that the  speed of sound becomes imaginary.
130: 
131: Even in the regime where the gaseous phase is locally stable
132: there are always collapsed configurations which have much
133: lower energy than the gas. In fact, the energy has its absolute minimum in
134: the collapsed phase where all particles fall to one point and this
135: configuration completely dominates the free energy, causing its divergence
136: to minus infinity. Of course, interactions other than gravitational always
137: dominate for short distances. Therefore, the properties the collapsed phase
138: depend on the short distance physics. This physics is different according
139: to the nature of the `particles' considered. They can be from galaxies till
140: molecules or atoms. Still, we can study the regime when collapse
141: happens just introducing a short distance cutoff (ultraviolet regulator)
142: to the gravitational interaction.
143: 
144: We present a complete description of the self-gravitating
145: gas, including the collapsed phase, using a Newtonian potential regularized
146: at short distances. We choose the Newton potential to be $ -G \; m^2 / A $
147: for interparticle distances $ r \leq A $ instead of Newton's law
148:  $ -G \; m^2 / r $. The cutoff $ A $ is chosen very small compared with the 
149: size of the box.
150: 
151: We give an exact mapping of the $N$-body regularized problem into a field
152: theory functional integral. This functional integral is dominated
153: by its saddle points in the $ N \to \infty $, namely by the mean field
154: equations. The mean field equations coincide for zero cutoff with
155: the Lane-Emden equation obtained from hydrostatics and assuming an ideal gas
156: equation of state.
157: 
158: In this way we reproduce the well-known isothermal sphere solutions
159: describing a self-gravitating gas in local thermal equilibrium.
160: These  diluted solutions get small corrections for small cutoff.
161: 
162: We find a host of new solutions (saddle points) from our regularized
163: mean field equations. These new solutions are singular in the limit
164: of zero cutoff and cannot be find from the standard (no cutoff)
165: Lane-Emden equation. 
166: In these new solutions, which are perfectly regular for non-zero cutoff,
167: a large number of particles concentrate near the origin.
168: There are two new types of solutions.
169: 
170: The first type are the collapsed saddle point solutions: all the
171: particles are densely concentrated in a region of size of
172: the order of the cutoff. These solutions are studied analytically
173: and provide the absolute minima of the free energy, which reads,
174: at dominant order for small cutoff $ A \ll R $:
175: \begin{equation*}
176:   F-F_0 \simeq
177:   -\frac{G \; m^2 \, N(N-1)}{2\,A} +NT\,\log\frac{R^3}{(A/2)^3}\;.
178: \end{equation*}
179: Here, $ T $ is the temperature and $ F_0 $ is the free energy of an ideal gas.
180: The first term is just the potential energy of $ N $ particles clustered in
181: a small sphere of radius $ A/2 $, where the regularized gravitational
182: interaction is the same for all $\frac12 N(N-1)$ particle pairs. The second
183: term is $ T $ times the entropy loss in the collapse.  This free energy is
184: large and negative, unbounded from below as $ A\to0 $. In particular, this
185: free energy is well below the free energy of the dilute solution.
186: 
187: The second type of saddle point solutions interpolate between the gas phase
188: and the collapsed phase, in a sort of pre--collapse.  This solution has a
189: finite action in the zero cutoff limit and describes the tunneling
190: transition where a small fraction of particles coalesce into a region of
191: the size of the short--distance regulator so that the density near the
192: origin is very large.  The tunneling probability to collapse for a dilute
193: solution turns out to be extremely small for large $N$ and small $A$. This
194: means in practice that dilute solutions which are locally stable can be
195: considered stable. As mentioned above this happens in the canonical
196: ensemble for $ 0 \leq \eta < \eta_T $ where $ \eta_T = 2.43450\ldots $.
197: 
198: The parameter $ \eta $ for the collapsed solutions can be arbitrarily large.
199: That is, the particle density does not need to be dilute $ N \sim
200: V^{\frac13} $ [see eq.(\ref{limter})] but we can have $ N \sim V $.
201: 
202: We obtain in sec. \ref{metaB} an estimate for the
203: lifetime for the dilute phase of the self-gravitating gas:
204: \begin{equation*}
205:   \tau \sim \frac1{a}\,\sqrt{\log \frac1{a}} \; 
206:   e^{\frac{9 \, N}{2\,\eta}\,a\,(\log   a)^2} \sim 
207:   \frac{R}{A} \, \sqrt{\log \frac{R}{A}} \; 
208:   \exp\Big[\frac{9\,A\,T}{2\,G\,m^2}\,
209:   \Big(\log\frac{R}A\Big)^2 \Big] \; .
210: \end{equation*}
211: One can see that the lifetime becomes infinitely long in the zero cutoff
212: limit as well as when $ N\to\infty $ at fixed cutoff
213: [recall that $ R\sim N $ in the dilute limit of eq.~(\ref{limter})].
214: 
215: In summary, we provide through mean field theory a complete statistical
216: description of the self-gravitating gas including the absolute minimum
217: of the free energy.
218: 
219: \section{Mean Field Approach to the Self-Gravitating Gas}
220: 
221: At short distances, the particle interaction for the self-gravitating gas
222: in physical situations is not gravitational. Its exact nature depends on
223: the problem under consideration (opacity limit, Van der Waals forces for
224: molecules etc.).  We shall just assume a null short--distance pair force;
225: that is, we consider the Hamiltonian
226: \begin{equation*}
227:   H_N = \sum_{j=1}^N \frac{p^2_j}{2m} - 
228: \sum_{j<k}\frac{G \, m^2}{|\vq_j-\vq_k|_A} \quad ,
229: \end{equation*}
230: where
231: \begin{equation}\label{defva}
232:   |\vq|_A = 
233:   \begin{cases}\displaystyle
234:     |\vq| \quad, & \mbox{for} \; |\vq|\ge A \\ 
235:     A \quad\,, &\mbox{for}\; |\vq| \le A \quad ,
236:   \end{cases}
237: \end{equation}
238: and $ A \ll R $ is the short--distance cut-off. 
239: 
240: It must be stressed that the results presented in this work using
241: the soft-core cut-off eq.(\ref{defva}) will be qualitatively the same
242: for other types of cut-off since $ R \gg A$.
243: 
244: The partition function in the canonical ensemble reads
245: \begin{equation}\label{fp}
246:   Z_N = \frac1{N!\,\hslash^{3N}}\int \frac{d^{3N}\!p}{(2\pi)^{3N}} 
247:   \int_{V^N} d^{3N}\!q \;e^{-\beta H_N} \; ,
248: \end{equation}
249: where $ T \equiv \beta^{-1} $ is the temperature. Computing the integrals 
250: over the momenta $\vp_j, \; (1 \leq j \leq N) $,
251: \begin{equation*}
252:   \int_{-\infty}^{+\infty}\;\frac{d^3p}{(2\pi)^3}\; \exp\Big[- \frac{\beta
253:       p^2}{2m}\Big] = \left(\frac{m}{2\pi \beta}\right)^{3/2} \; ,
254: \end{equation*}
255: yields
256: \begin{equation}\label{fpconf}
257:    Z_N = Z_N^{(0)}\,  Z_N^{\rm conf} \quad,\qquad
258:   Z_N^{\rm conf} = \frac1{V^N} \int_{V^N} d^{3N}\!q   \prod_{1\leq j < k\leq N}
259:    \exp\left(\frac{G\, m^2\, \beta}{|\vq_j-\vq_k|_A}\right) \; ,
260: \end{equation}
261: where 
262: \begin{equation*}
263:    Z_N^{(0)} =\left(\frac{2\pi m}{\beta\hbar^2}\right)^{3N/2}
264:    \,\frac{V^N}{N!} \equiv \exp\left[-\beta \,F^{(0)}\right] 
265:  \end{equation*}
266:  is the partition function of the ideal gas. Thus,
267: \begin{equation*}
268:   F=F^{(0)}-T \; \log Z_N^{\rm conf}
269: \end{equation*}
270: is the full canonical free energy.
271: 
272: We can now replace the integration over the particle coordinates with a
273: functional integration over the configurations of the gravitational field
274: $\varphi$ produced by the particles. We start from the fundamental property
275: of Gaussian functional integrals, namely
276: \begin{equation}\label{gauso}
277:   \int{\cal D}\varphi\, \exp\left\{- S[\varphi] + \int d^3x\,
278:   J(\vx)\,\varphi(\vx)\right\} = \exp\left[\frac12\int d^3x
279:   \int d^3x'\,J(\vx)\; {\cal G}(\vx,\vx')\; J(\vx')\right] \; ,
280: \end{equation}
281: where $S[\varphi]$ is the quadratic action functional
282: \begin{equation*}
283:   S[\varphi] = \frac12\int d^3x\int d^3x'\; \varphi(\vx)\,
284:   {\cal G}^{-1}(\vx,\vx')\, \varphi(\vx')
285: \end{equation*}
286: and the integration measure is assumed to be normalized such that 
287: $\int{\cal D}\varphi\,\exp\{-S[\varphi]\}=1$. Then we set 
288: \begin{equation}\label{potG}
289:   {\cal G}(\vx,\vx') = \frac{G\, \beta^{-1}}{|\vx-\vx'|_A}
290: \end{equation}
291: and we identify the source field $J(\vx)$ as $\beta m$ times the
292: microscopic particle density
293: \begin{equation}\label{densu}
294:   J(\vx) = \beta m\sum_{j=1}^N \delta^{(3)}(\vx-\vq_j) \; .
295: \end{equation}
296: Inserting eqs.~(\ref{potG}) and (\ref{densu}) into eq.~(\ref{gauso}) yields,
297: \begin{equation}\label{idu}
298:   \int{\cal D}\varphi\, \exp\Bigg\{-S[\varphi] +\beta \;  m\sum_{j=1}^N 
299:     \varphi(\vq_j)\Bigg\} = \exp\Bigg[\frac12\sum_{j,k=1}^N 
300:     \frac{G \, m^2 \, \beta}{|\vq_j-\vq_k|_A}\Bigg] =
301:     e^{\frac{Gm^2\beta\,N}{2A}} \!\!\prod_{1\leq j<k\leq N}
302:    \exp\left(\frac{G \; m^2 \; \beta}{|\vq_j-\vq_k|_A}\right) \; .
303:  \end{equation}
304: This provides a functional integral representation of the product over 
305: particle pairs in the integrand of eq.~(\ref{fpconf}). 
306: Inserting the identity eq.~(\ref{idu})
307: in eq.~(\ref{fpconf}) and exchanging the functional integration with the
308: integration over the coordinates $\vq_j$ we obtain
309: \begin{equation}
310:   Z_N^{\rm conf} =  e^{-\frac{Gm^2\beta\,N}{2A}}
311:   \int D\varphi \; \exp\left\{-S[\varphi]\right\} \;
312:   \frac1{V^N} \int_{V^N} d^{3N}\!q
313:   \; \exp\Big[\beta \,  m\sum_{j=1}^{N} \varphi({\vq}_j) \Big] \; .
314: \end{equation}
315: where
316: \begin{equation*}
317:   S[\varphi] = -\frac{\beta}{8 \, \pi \,  G} \int d^3x \; 
318:   \varphi \, \nabla^2_A\varphi
319: \end{equation*}
320: and $ \nabla^2_A=-(4\pi \, G/\beta) \; {\cal G}^{-1} $ satisfies
321: \begin{equation*}
322:   -\nabla^2_A \frac{1}{|\vx-\vx'|_A} = 4\pi\,\delta^{(3)}(\vx -\vx') \; .
323: \end{equation*}
324: Since the $ {\vq}_j $ are dummy variables, we have
325: \begin{equation*}
326:     \int_{V^N} d^{3N}\!q  \; \exp\Big[\beta \, m\sum_{l=1}^{N} 
327:     \varphi({\vq}_l) \Big]= \left[ \int_{V}  d^3q \; 
328:       e^{\beta \,  m \, \varphi({\vq})} \right]^N 
329: \end{equation*}
330: and the configuration partition function becomes
331: \begin{equation}\label{rizc}
332:   Z_N^{\rm conf} = e^{-\frac{G \, m^2 \, \beta\,N}{2 \, A}}
333:   \int{\cal D}\varphi\, e^{-S_{\rm eff}[\varphi]} \; ,
334: \end{equation}
335: where
336: \begin{equation}\label{accef}
337:   S_{\rm eff}[\varphi] = S[\varphi] + S_{\rm int}[\varphi] \quad,\qquad 
338:   S_{\rm int}[\varphi] \equiv - N\log\Big[\frac1{V} \int_V d^3x \;
339:   e^{-\beta \, m \, \varphi(\vx)}\Big]
340: \end{equation}
341: We want to stress that this field--theoretic functional formulation of the
342: self-gravitating gas is fully equivalent, for any $ N $, to the original one
343: in terms of particles. The functional integral representation of the
344: partition function presented in ref.\cite{I,II} for the canonical ensemble
345: is slightly different but equivalent to eqs.~(\ref{rizc}) and~(\ref{accef})
346: for large $ N $. Let us also observe that the nonlocal operator $ \nabla^2_A $
347: reduces to the standard Laplacian operator in the limit $A\to 0$. However,
348: this limit is not so straightforward in the functional integral above,
349: since the interaction term $ S_{\rm int} $ in the action may (and indeed
350: does) become unbounded from below in such limit.
351: 
352: We now pass to dimensionless variables by setting $ \vR=\vx/R $ and 
353: \begin{equation*}
354:   -\beta m\, \varphi(\vx) = \phi(\vR) \; .
355: \end{equation*}
356: The action for $ \phi(\vx) $ now reads
357: \begin{equation}\label{accefa}
358:    S_{\rm eff}[\varphi] = N \,s[\phi]\quad,\qquad 
359:    s[\phi] = -\frac1{8\pi\,\eta}\int d^3r\; \phi  \; \nabla^2_a \phi 
360:    - \log\left[ \int_{|\vR|\le 1} d^3r \; e^{\phi}\right] +
361:    \log\frac{4 \, \pi}3 \; ,
362: \end{equation}
363: where
364: \begin{equation}\label{defeta}
365:   \eta \equiv \frac{G \, m^2\beta\, N}{R} \quad,\qquad a\equiv\frac{A}R
366:   \quad,\qquad  -\nabla^2_a \frac{1}{|\vR-\vR'|_a} = 
367:   4\pi\,\delta^{(3)}(\vR -\vR') \; .
368: \end{equation}
369: We recall that the variable $ \eta $ is the ratio of the characteristic
370: gravitational energy $ \frac{G \,  m^2 \, N}{R} $ and the kinetic energy 
371: $\sim T=\beta^{-1} $ of a particle in the gas. For $\eta=0$ the ideal gas is
372: recovered.
373: 
374: Having factored out of the action the number of particles $N$, it is
375: natural to use the saddle points method to evaluate the functional
376: integral over $\phi$ in the limit $N\to\infty$ at fixed $\eta$, that is in
377: the dilute limit. Strictly speaking, however, the action per particle
378: $s[\phi]$ still depends on $N$ through $a=A/R\propto A/N$. Thus we
379: should better regard the saddle point method as a mean field type
380: approximation yielding the free energy $F$ to leading order in $N$,
381: as we shall now show. The stationarity condition 
382: \begin{equation*}
383:   \frac{\delta s[\phi]}{\delta \phi(\vR)} = 0 \; .
384: \end{equation*}
385: leads to the regularized self--consistent Boltzmann--Poisson 
386: equation
387: \begin{eqnarray}  \label{eq;BPLE}
388: &&  \nabla^2_a \phi + \frac{4\pi\,\eta}{Q}\; e^{\phi} = 0 \quad,\qquad 
389:   Q \equiv \int_{|\vR|\le 1} d^3r \; e^{\phi(\vR)}\quad,\qquad 
390:   {\rm for} \quad  | \vR| \le 1 \; , \quad {\rm and} \cr \cr
391: && \nabla^2_a \phi = 0 \quad,\qquad {\rm for} \quad | \vR | \ge 1  \quad.
392: \end{eqnarray}
393: In the limit $a\to0$ this becomes the standard self--consistent
394: Boltzmann--Poisson (or Lane--Emden) equation derivable from hydrostatic
395: plus the assumption of a local, ideal gas equation of state \cite{viej}.
396: 
397: We may rewrite eq.~(\ref{eq;BPLE}) in integral form as
398: \begin{equation}\label{ecinta}
399:   \phi(\vR) = \eta \int d^3r' \;
400:   \frac{\rho(\vR')}{| \vR - \vR'|_a } \;,
401: \end{equation}
402: where 
403: \begin{equation}\label{ro}
404:   \rho(\vR) = \frac{e^{\phi(\vR)}}Q \quad {\rm for}\;|\vR|\le 1 \quad , 
405:   \qquad \rho(\vR) =0 \quad {\rm for} \quad |\vR| > 1 \quad , \qquad
406:   \int d^3r \; \rho(\vR) = 1 \; ,
407: \end{equation}
408: is the normalized particle density associated to $\phi(\vR)$.
409: 
410: We may also use the identity involving the standard Laplacian $\nabla^2$,
411: \begin{equation*}
412: -\nabla^2  \frac1{|\vR - \vR '|_a } =\frac1{a^2} \; \delta(|\vR-\vR'|-a) \; ,
413: \end{equation*}
414: to cast the field equation in the integro--differential form
415: \begin{equation}\label{eq:reg0}
416:   \nabla^2\phi(\vR) = -\frac{\eta}{Q \; a^2} \int_{|\vR'|\le 1} d^3R'\;
417:   \delta(|\vR-\vR'|-a) \; e^{\phi(\vR')} \; .
418: \end{equation}
419: Notice that the laplacian in the origin is always regular for nonzero cutoff
420: \begin{equation}\label{eq:r=0}
421:    \nabla^2\phi(0) = -\frac{\eta}{Q \; a^2} \int_{|\vR'|\le 1} d^3r' \;
422:    \delta(|\vR'|-a) \; e^{\phi(\vR')} \; ,
423: \end{equation}
424: implying, together with the finiteness of $ \phi(0) $, that
425: $\nabla\phi=0$ in the origin.
426: 
427: \medskip
428: 
429: We can recast the action of the saddle point eq.(\ref{accefa}) using the 
430: regularized equation of motion (\ref{eq;BPLE}) as follows,
431: \begin{equation}\label{acci}
432: s[\phi] = \frac1{2 \; Q} \; \int_{|\vR|\le 1} d^3r \; \phi(\vR) \; 
433: e^{\phi(\vR)} - \log\frac{3 \; Q}{4 \, \pi} \quad .
434: \end{equation}
435: The free energy for large $ N $ can be written in terms of the stationary
436: action as\cite{I}
437: \begin{equation}\label{Fpv}
438:   F = F_0 + N\,\frac{G\,m^2}{2A} + N\, T\, s(\eta,a) + 
439:   {\cal O}(1) \quad ({\rm at \; fixed}\;a) \, \; ,
440: \end{equation}
441: where $ s(\eta,a) \equiv s[\phi_{s}] $, $\phi_{s}$ is a solution of
442: eq.~(\ref{eq;BPLE}), and $ F_0 $ is the free energy for the ideal gas.
443: Notice that, since $ a $ vanishes as $ N^{-1} $ as $ N\to\infty $ 
444: {\em at fixed} $ A $ {\em and} $ \eta $, we can regard $ N\,T\,s(\eta,a) $ 
445: as extensive in the particle number, that is linear in $ N $, 
446: only if it is regular as $ a \to 0 $. This holds true for 
447: stationary point which are regular at the $ a = 0 $ limit.
448: We show below that there are saddle points which become singular
449: in the $ a \to 0 $ limit and consequently the free energy is not proportional
450: to $ N $. Similarly, the subleading terms in eq.~(\ref{Fpv}) are
451: order one only for  saddle points which  are regular for $ a = 0 $.
452: What is always true is that they are indeed subleading for any given saddle
453: point.
454: 
455: \section{Spherically symmetric solutions}
456: 
457: In the case of spherically symmetry the integration over angles
458: in eq.~(\ref{eq:reg0}) can be performed explicitly, yielding the
459: one--dimensional non--linear integro--differential equation
460: \begin{equation}\label{ecdar}
461:    \ddr[r\phi(r)] = -\frac{2 \, \pi \, \eta}{Q \, a}
462:    \int_{|r-a|}^{r+a} dr'\; r' \, e^{\phi(r')} \; ,
463: \end{equation}
464: where now
465: \begin{equation} \label{qr}
466:   Q = 4 \pi \int_0^1 dr\; r^2 \, e^{\phi(r)} \; .
467: \end{equation}
468: Eq.~\ref{ecdar} is to be supplemented with the boundary conditions of
469: smooth joining (continuity of $\phi(r)$ and $\phi'(r)$ at $r=1+a$)
470: with the external monopole solution $\phi(r)=\eta/r$.
471: 
472: Integration over the angles in eq.(\ref{ecinta}) yields the non--linear
473: integral equation 
474: \begin{equation}\label{ecinar}
475:   \begin{split}
476:     \phi(r) \,=\;& \frac{4 \, \pi \; \eta}{Q} \left\{ \frac1{{\rm max}(r,a)}
477:       \int_0^{|r-a|}  dr' \; r'^2 \;e^{\phi(r')}\; + \; \theta(1-r-a)
478:       \int_{r+a}^1 dr' \; r' \;e^{\phi(r')} \right.\\
479:   &\left. -\,\frac1{2\,r} \int_{|r-a|}^{r+a}  dr'  \;r' \left[
480:       \frac{a}2 - r -r' - \frac{(r-r')^2}{2 \,a } \right] e^{\phi(r')}\;
481:     \theta(1-r')\right\} \; ,
482: \end{split}
483: \end{equation}
484: which by itself determines $ \phi(r) $ for all values of $ r $. In particular
485: eq.~(\ref{ecinar}) implies that any solution is finite in the origin, since
486: \begin{equation}\label{ecinar0}
487:   \phi(0) = \frac{4 \, \pi \; \eta}{Q} \Big[ \frac1{a}\int_0^a  dr \;
488:   r^2 \;e^{\phi(r)}\; + \;\int_a^1 dr \; r \;e^{\phi(r)} \Big] \; .
489: \end{equation}
490: Moreover, eq.~(\ref{ecdar}) or eq.~(\ref{ecinar}) imply
491: \begin{equation}\label{eq:phip0}
492:   \phi'(0) = 0 \quad\mathrm{and} \quad \phi'(r) < 0  
493:   \quad \mathrm{for} \quad r > 0\; ,
494: \end{equation}
495: so that the density is monotonically decreasing away from the origin.
496: The simplest way to prove that $\phi'(0) \le 0$ is to multiply both sides of 
497: eq.~(\ref{ecdar}) by $r$ and integrate from 0 to an arbitrary value.
498: This yields
499: \begin{equation*}
500:   \phi'(r) = -\frac{2 \, \pi \,\eta}{Q \,a\,r^2} \int_0^r dr'\; r'
501:    \int_{|r'-a|}^{r'+a} dr''\; r'' \; e^{\phi(r'')} \; ,
502: \end{equation*}
503: which has a negative definite r.h.s. which vanishes when $r\to0$ as long as
504: $a>0$. 
505: 
506: The action for spherically symmetric solution becomes from
507: eq.(\ref{acci}),
508: \begin{equation}\label{accir}
509: s[\phi] = \frac{2 \, \pi}{Q} \; \int_0^1 r^2 \;  dr \; \phi(r) \; 
510: e^{\phi(r)} - \log\frac{3 \; Q}{4 \, \pi} \quad .
511: \end{equation}
512: \section{Solutions Regular as $ A \to 0 $}
513: 
514: The saddle point equation (\ref{eq;BPLE}) admits solutions which are
515: regular as $ a\to0 $ and reproduce in this limit the solutions of the standard
516: Lane--Emden equation. We call them dilute solutions.
517: These are known since longtime in the spherically
518: symmetric case \cite{viej}. To study such solutions we can set the cutoff $
519: A $ to zero from the start. Equivalently, one can take the limit $a\to0$ in
520: eq.~(\ref{ecdar}) with the assumption that $\phi(r)$ stays finite in the
521: limit for any $r$, including the origin $r=0$, and that $\phi'(0)=0$.
522: 
523: Assuming spherical symmetry we get $ \phi(r)=\eta/r $ 
524: for $ r \ge 1 $, while for  $r \le 1$ we have
525: \begin{equation}\label{eqrad}
526:   \phi''(r) + \frac2r \; \phi'(r) + \frac{4\pi\eta}{Q} \; e^{\phi(r)}= 0
527:   \quad,\qquad \phi'(0) = 0 \; ,
528: \end{equation}
529: with the boundary conditions
530: \begin{equation}\label{eq:bc}
531:   \phi(1) = \eta \quad,\qquad \phi'(1) = -\eta \; .
532: \end{equation}
533: Clearly there exists one and only one solution for a given $Q$, so the
534: classification of all solutions is equivalent to the determination of all
535: the allowed values of $ Q $.
536: 
537: We set as usual\cite{I}
538: \begin{equation}\label{eq:phichi}
539:   \phi(r) = \phi(0)+ \chi(\lambda r) \; ,\quad \phi(0) =
540:   \log\frac{Q \,\lambda^2}{4\pi \,\eta} \;,\quad \chi(0)=0 \; .
541: \end{equation}
542: where $ \chi(z) $ must satisfy, upon inserting eq.~(\ref{eq:phichi})
543: in eq.~(\ref{eqrad}),
544: \begin{equation}   \label{eq:fund}
545:   \chi''(z) + \frac2z  \; \chi'(z) + e^{\chi(z)} = 0 
546:   \quad,\qquad  \chi'(0)=0
547: \end{equation}
548: and, from the boundary conditions eq.~(\ref{eq:bc}) at $ r=1 $,
549: \begin{equation}\label{eq:bc2}
550:   \log\frac{Q  \; \lambda^2}{4 \, \pi \, \eta} + \chi(\lambda) = \eta
551:   \quad ,\qquad \lambda \; \chi'(\lambda) = -\eta \; .
552: \end{equation}
553: These two relations fix the two parameters $Q$ and $\lambda$ as functions
554: of $\eta$. In particular, we can rewrite $\phi(r)$ as
555: \begin{equation}\label{eq:phichi2}
556:   \phi(r;\eta) = \eta -\chi(\lambda(\eta)) + \chi(\lambda(\eta) r) \;.
557: \end{equation}
558: The density profile of eq.~(\ref{ro}) reads in terms of $\chi(\lambda r)$,
559: \begin{equation}
560: \label{eq:dprof}
561:   \rho(r) = \frac{\lambda^2}{4 \, \pi \; \eta} \; e^{\chi(\lambda r)} \; ,
562: \end{equation}
563: so that the normalization factor can also be written $ Q = e^\eta/\rho(1) $.
564: 
565: Due to scale--invariant behaviour of the  gravitational interaction, eq.
566: (\ref{eq:fund}) enjoys the following scale covariance property: if
567: $ \chi(z) $ is a solution, then also $ \chi_\alpha(z) \equiv
568: \chi(z\,e^\alpha)+2 \,\alpha $ is a solution. 
569: Using $ \chi_\alpha(z) $ rather than $ \chi(z) $ is compensated by the shift 
570: $ \lambda\to\lambda \, e^{-\alpha}$. Notice that $ \phi(r) $ is indeed defined 
571: as a scale transformation of $ \chi(z) $, with the scale parameter 
572: $ \lambda $ (not uniquely) fixed by $\lambda\, \chi'(\lambda) = -\eta$. 
573: As a consequence of this scale invariance
574: all physical quantities must be invariant under the simultaneous
575: replacements $ \lambda\to\lambda\, e^{-\alpha} $ and
576: $ \chi(\lambda)\to\chi(\lambda) + 2 \, \alpha $.
577: 
578: Let us observe that, if the solutions of eq.(\ref{eqrad}) or
579: (\ref{eq:fund}) did not fulfill $\phi'(0)=0$ or $ \chi'(0)=0$, a delta
580: function at the origin would appear in the right hand side and $
581: \chi(z)\simeq -A/z $, with $ A>0 $, when $z\to0$.  This is the only
582: possible singular behaviour and corresponds to a point particle with
583: negative mass $ - A $ in the origin.  We shall not consider such unphysical
584: solutions, which are ruled out when eq.(\ref{eqrad}) is considered as the
585: $a\to0$ limit of eq.~(\ref{ecdar}), and stick to $ \chi'(0)=0 $.
586: 
587: Moreover, scale invariance allows to set also $ \chi(0)=0 $.  This choice
588: completely fixes the scale and we have no more scale invariance left.  One
589: has from the second eq.~(\ref{eq:bc2}) that $ \chi'(z)<0 $ for all positive
590: $ z $, so that the density profile is monotonically decreasing with the
591: distance.
592: 
593: Since $\chi(z)$ is monotonically decreasing, the allowed values of $ Q $
594: [that is of $ \rho(1) $, see eqs.(\ref{ro}) and (\ref{eq:bc})] 
595: are in one--to--one correspondence with to the
596: roots of the relation $ \lambda=\lambda(\eta) $ which inverts the second of
597: eqs.~(\ref{eq:bc2}). Thus there is only one dilute solution for each value
598: of $ \lambda $. However, $ \lambda=\lambda(\eta) $ is a multiple valued
599: function for a certain range of $\eta$. We plot $\eta $ vs. $ \log \lambda$
600: in Fig.~\ref{fig:eta0}.  One sees that there is a unique $\lambda $ for a
601: given $\eta $ only for $ \lambda< 3.6358865\ldots $, that is $ \eta
602: <\eta_2 \equiv 1.84273139\ldots $. For $ \eta_2<\eta<\eta_C\equiv
603: 2.517551\ldots $ ($ \eta_C $ is the absolute maximum, located at
604: $\lambda_C=8.99311\ldots $, of $ -\lambda\chi'(\lambda) $ over $
605: 0<\lambda<\infty $) the relation $ \lambda=\lambda(\eta) $ is indeed
606: multivalued and as $ \eta $ approches the value $ 2 $ there are increasingly
607: more solutions which accumulate near the purely logarithmic solution
608: 
609: \begin{figure}[ht]
610:   \centering
611:   \psfrag{loglam}{$\log\lambda$}
612:   \psfrag{etavar}{$\eta$}
613:   \psfrag{etab}{$\eta_T$}
614:   \psfrag{etac}{$\eta_C$}
615:   \psfrag{eta2}{$\eta_2$}
616:   \includegraphics[width=.75\textwidth]{eta0.eps}
617:   \caption{ $ \eta $ as a function of $ \log \lambda $ according to
618:     eq.(\ref{eq:bc2}) for the dilute solutions. Notice the maximum 
619: of $ \eta $ at $ \eta_C =
620:     2.51755\ldots $, where $c_V$ diverges. The region beyond the maximum
621:     location, at $\log \lambda_{\rm C} = 2.196459\ldots $, is locally
622:     unstable as discussed in ref.\cite{II}. $ c_P $ and $ \kappa_T $ diverge
623:     before the maximum, at the point $ \eta_T=2.43450\ldots $.}
624:   \label{fig:eta0} 
625: \end{figure}
626: 
627: \begin{equation*}
628:   \phi_\infty(r) = \lim_{\lambda\to\infty} \Big[\eta -\chi(\lambda) 
629:    + \chi(\lambda r) \Big] = 2-2\,\log r \; .
630: \end{equation*}
631: which correspond to the singular density profile $\rho_\infty(r)=(4\pi
632: r^2)^{-1}$. In other words, when $\eta$ is exactly 2 there exist an
633: infinite set of solutions corresponding to the increasing sequence of
634: values of $\lambda$ which satisfy $\lambda\chi'(\lambda) = -2$. In Fig.
635: \ref{mrho} we plot in log--log scale the density profile $\rho(r)$ [see
636: eq.~(\ref{eq:dprof})] vs. $r$ for few terms of this sequence.
637: 
638: \begin{figure}[ht]
639:   \centering
640:   \psfrag{rvar}{$r$}
641:   \psfrag{rho}{$\rho$}
642:   \psfrag{eta}{$\eta=2$}
643:   \psfrag{lam0}{$\lambda=4.071496\ldots$}
644:   \psfrag{lam1}{$\lambda=37.526798\ldots$}
645:   \psfrag{lam2}{$\lambda=428.912529\ldots$}
646:   \psfrag{lam3}{$\lambda=4532.148619\ldots$}
647:   \psfrag{laminf}{$\lambda=\infty$}
648:   \includegraphics[width=.75\textwidth]{mrho.eps}
649:   \caption{Log--log plot of the density profile for some dilute solutions
650:     at $\eta=2$. Only the smallest one at $\lambda=4.071496\ldots$ is
651:     locally stable.}
652:   \label{mrho} 
653: \end{figure}
654: 
655: We recall that, for any given $ \eta<\eta_C$, only the unique
656: dilute solution with $ \lambda<\lambda_C $ is locally stable. That is,
657: any small fluctuation increases the action. 
658: At $ \lambda=\lambda_C $ a zero--mode appears in the
659: linear fluctuation spectrum and $ c_V $, the specific heat at constant
660: volume, diverges. For larger values of $ \lambda $ negative modes show up,
661: signalling local instability. Actually the locally stable dilute solutions 
662: describe a (locally) stable dilute phase, in the thermodnamic and
663: mechanical terms, only for $ \eta<\eta_T=2.43450.. $. At $
664: \lambda=\lambda_T=6.45071\ldots $ 
665: the isothermal compressibility
666: diverges and the speed of sound inside the sphere becomes imaginary.
667: The gas phase collapses at this point in the canonical
668: ensemble as a consequence of the Jeans instability. This is confirmed by
669: Monte Carlo simulations\cite{I}. Hence, in the case $ \eta=2 $ depicted in
670: Fig. \ref{mrho}, only the solution with the smallest density contrast,
671: corresponding to the smallest value $ \lambda=4.071496\ldots $, is
672: stable. The unstable solutions that exist for $ \eta_2<\eta<\eta_T $
673: should imply that the dilute phase is metastable for this range of $ \eta $,
674: However, their lifetime is so huge [see eq.(\ref{vida2})] that these 
675: dilute solutions are in practice stable. 
676: 
677: In the stable dilute phase the particles are moderately clustered
678: around the origin with a density that monotonically decreases with $ r $.
679: One sees that the density contrast between the center and the boundary 
680: \begin{equation*}
681:   \frac{\rho(0)}{\rho(1)}= e^{-\chi(\lambda)}
682: \end{equation*}
683: grows  with  $ \lambda $ since $ \chi(\lambda) $ decreases with $ \lambda $ 
684: [see eq.(\ref{eq:bc2})] and Fig. \ref{mrho} \cite{viej,I,II}.
685: 
686: The action per particle of a dilute solution can be written entirely in
687: terms of the density at the border\cite{I}
688: \begin{equation}\label{seta}
689:   s(\eta) = 3-\eta +\log\left[ \frac{4\pi}3 \, \rho(1)\right] -
690:   4\pi\rho(1)= 3-\eta +
691:   \log\frac{\lambda^2}{3 \, \eta}+ \chi(\lambda)-
692:   \frac{\lambda^2}{\eta} \, e^{\chi(\lambda)} \; ,
693: \end{equation}
694: and is multivalued for $ \eta_2 \ldots<\eta<\eta_C\ldots $. The solution
695: with the smallest density contrast has the smallest action, in accordance
696: with the linear stability analysis. We plot $ s(\eta) $ as a function of 
697: $\log\lambda$ and $\eta$ in Fig.~\ref{fig:action0}.
698: 
699: \begin{figure}[ht]
700:   \centering  
701:   \psfrag{loglam}{$\log\lambda$}
702:   \psfrag{lamc}{$\log\lambda_C$}
703:   \psfrag{etavar}{$\eta$}
704:   \psfrag{action}{$s$}
705:   \psfrag{etac}{$\eta_C$}
706:   \includegraphics[width=.75\textwidth]{action0.eps}
707:   \caption{The action per particle $ s $ vs. $ \log\lambda $ (above) and
708: vs. $ \eta $ (below) for the dilute solution. $ s(\eta) $ is multivalued 
709: for $ \eta>\eta_2 $, with a peculiar self--similar structure, 
710: as evident from the inset.}
711:   \label{fig:action0} 
712: \end{figure}
713: 
714: All physical quantities can be expressed  for large $ N $ 
715: in terms of the function
716: \begin{equation}\label{def}
717:   f(\eta) \equiv  1 + \frac{\eta}3 \; \frac{ds}{d \eta} 
718:   =\frac{4\pi}3\, \rho(1) \;. 
719: \end{equation}
720: We have for the energy, the pressure, the isothermal
721: compressibility $\kappa_T$, the specific heats ($c_V$ and $c_P$)
722: and the speed of sound at the boundary\cite{I}
723: \begin{eqnarray}\label{resto}
724: &&\frac{E}{3 \, N \; T} = f(\eta)-\frac12 + 
725: {\cal O}\left(\frac1{N}\right) \quad , \quad
726: \frac{p \; V}{N \; T} = f(\eta) + {\cal O}\left(\frac1{N}\right) \; , \cr \cr
727: && \frac{S - S_0}{N} = -3[ 1 - f(\eta)] - s(\eta) + 
728: {\cal O}\left(\frac1{N}\right)
729: \quad , \quad \kappa_T =  \frac1{f(\eta)+\frac13 \; \eta  \; f'(\eta)} 
730: + {\cal O}\left(\frac1{N}\right) \; , \\ \cr
731: && c_V = 3 \left[  f(\eta)-\eta \; f'(\eta) -\frac12 \right]
732: + {\cal O}\left(\frac1{N}\right) \quad , \quad
733: c_P = c_V +  \frac{ \left[f(\eta)-\eta \;  f'(\eta)\right]^2}{
734: f(\eta)+\frac13 \eta \;  f'(\eta)} + {\cal O}\left(\frac1{N}\right) \; ,\cr \cr
735: &&\frac{v_s^2}{T} = \frac{ \left[f(\eta)-\eta \;  f'(\eta)\right]^2}{ 3
736: \left[f(\eta)-\eta \;  f'(\eta)-\frac12 \right]} + 
737: f(\eta)+\frac13  \;  \eta \; 
738: f'(\eta) + {\cal O}\left(\frac1{N}\right)\; . \nonumber
739: \end{eqnarray}
740: Here $ S_0 $ stands for the entropy of the ideal gas.
741: 
742: Notice that $\kappa_T$ and $c_P$, the specific heat at constant pressure,
743: both diverge before $\eta_C$, at $ \eta= \eta_T \equiv 2.43450\ldots
744: $\cite{I}. Moreover, the speed of sound squared, which stays regular at the
745: boundary when $\eta\to\eta_T$, has a simple pole there when evaluated
746: inside the isothermal sphere\cite{I}. Thus, when $\eta>\eta_T$ the speed
747: of sound is purely imaginary inside the sphere implying that small
748: fluctuations grow exponentially with time.  
749: 
750: We have here discussed the dilute solutions in the zero cutoff limit
751: ($A=0$). Including the short distance cutoff only adds small $ {\cal O}(A)
752: $ corrections to the dilute solutions.
753: 
754: \section{Solutions Singular as $ A \to 0 $.}
755: 
756: We present in this section new solutions for which the mean field $ \phi(r) $
757: develops singularities in the limit $ A\to0 $. In case of spherical symmetry,
758: when eq.(\ref{ecdar}) is the saddle point equation, the only possible 
759: singularity is localized at $ r=0 $. It is convenient then to `blow up'
760: the region near $ r=0 $ by setting [compare with eq.~(\ref{eq:phichi})]
761: \begin{equation}\label{eq:phixi}
762:   \phi(r) =  \phi(0) + \mu^2 \; \xi\left(\frac{r}{a}\right) \;, \quad 
763:   \phi(0)=\log\frac{Q\lambda^2}{4\pi\,\eta} \quad , \quad \xi(0) = 0  \; ,
764:   \quad {\rm so~that} \quad 
765:     \rho(r) = \frac{\lambda^2}{4 \pi \, \eta} \, e^{\mu^2 \; 
766:       \xi\left(\frac{r}{a}\right)}  \; .
767: \end{equation}
768: where we introduce the variable:
769: \begin{equation*}
770: \mu \equiv \lambda\,a \; .
771: \end{equation*}
772: We thus obtain for $ \xi(x) , \; x=r/a $, the following rewriting of
773: the integral equation (\ref{ecinar})
774: \begin{equation}\label{eq:varinteq}
775:   \begin{split}
776:     \xi(x) \;=\; &\frac1{{\rm max}(x,1)} \, I_2(|x-1|) - I_1(m(x))
777:     - \frac{(x-1)^2}{4x} \Big[I_1(m(x)) -I_1(|x-1|)\Big] \\ 
778:     +& \frac{x+1}{2x} \Big[I_2(m(x)) -I_2(|x-1|)\Big] - 
779:     \frac1{4x} \Big[I_3(m(x)) - I_3(|x-1|)\Big] - I_2(1) + I_1(1)\;,
780:   \end{split}
781: \end{equation}
782: where,
783: \begin{equation}\label{mx}
784:   m(x) \equiv {\rm min}\left(\frac1{a}, x+1\right) 
785: \;,\quad I_n(x) \equiv \int_0^{x} dy \; y^n \; \exp[\mu^2 \, \xi(y)] \; , 
786: \end{equation}
787: and we have used eq.~(\ref{ecinar0}) to write $ \phi(0) $ in terms of
788: $\xi(x)$ as
789: \begin{equation}\label{eq:phi0}
790:   \phi(0) = \mu^2 \left[ I_2(1) + I_1\left(\frac1{a}\right) - 
791:     I_1(1) \right] \; .
792: \end{equation}
793: Notice that all dependence on $ \lambda $ appears now through 
794: the variable $ \mu = \lambda\,a $.
795: 
796: By construction we have $ \xi'(0)=0 $ [see eq.~(\ref{eq:phip0})],
797: $ \xi(0)=0 $ and $ \xi'(x)<0 , \; \xi(x)<0 $ for all $ x>0 $. 
798: Notice that $ \eta $ does not enter the integral equation above, 
799: just as it did not
800: enter the differential equation (\ref{eq:fund}) for $ \chi $ in the setup
801: without short--distance regulator. Owing to the normalization condition
802: eq.~(\ref{qr}), $ \eta $ is {\bf computed} once a solution is known, 
803: as a function of $ \mu $ (and $ a $), through
804: \begin{equation}\label{eq:etamu}
805:   \eta = \mu^2\, a \,I_2\left(\frac1{a}\right) = 
806: \mu^2\, a \,\int_0^{\frac1{a}} dx\, x^2\, \exp[\mu^2\,\xi(x)] \;.
807: \end{equation}
808: >From eq.~(\ref{ecdar}) we may also derive the integro--differential form of 
809: the equation satisfied by $ \xi(x) $, that is 
810: \begin{equation}\label{eq:vardiff}
811:    \ddr[x \, \xi(x)] = -\frac12 \Big[I_1(x+1) -I_1(|x-1|)\Big]\;.
812: \end{equation}  
813: Eq.~(\ref{eq:varinteq}) or (\ref{eq:vardiff}) can be solved numerically to
814: high accuracy for any given $ \mu $. We plot the corresponding $\eta$ as a
815: function of $\log\lambda$ in Fig.~\ref{fig:eta1}, for better comparison
816: with its behaviour in the dilute case, Fig.~\ref{fig:eta0}.
817: 
818: \begin{figure}[ht]
819:   \centering  
820:   \psfrag{loglam}{$\log\lambda$}
821:   \psfrag{etavar}{$\eta$}
822:   \psfrag{a104}{\hspace{-0.8cm}$a=10^{-4}$}
823:   \psfrag{a105}{\hspace{-0.8cm}$a=10^{-5}$}
824:   \psfrag{a106}{\hspace{-0.8cm}$a=10^{-6}$}
825:   \includegraphics[width=.75\textwidth]{eta1.eps}
826:   \caption{$ \eta $ as a function of $ \log \lambda $ according to
827:     eq.(\ref{eq:etamu}). The black circle mark the values of $ \lambda $
828:     corresponding to the interpolating solutions depicted in 
829: fig.~\ref{fig:chifit}. Here, $ \mu^2 \equiv \lambda^2 \; 
830: a^2 = {\cal O}(\log \frac1{a}) $.}  \label{fig:eta1} 
831: \end{figure}
832: 
833: We find {\bf four} different regimes 
834: according to the value of $ \mu $ for $ a \ll 1 $.
835: \begin{itemize}
836: \item $  \mu = {\cal O}(a) $. That is, $ \lambda = {\cal O}(1) $.
837: In this regime we reproduce the curve $ \eta =\eta(\lambda) $ (Fig.
838: \ref{fig:eta0}) characteristic of the diluted phase of sec. IV plus 
839: small corrections of order $ a $. 
840: 
841: \item $ \mu  = {\cal O}(1) $. Here $ \lambda = {\cal O}(\frac1{a}) $
842: and the damped oscillation pattern around $ \eta=2 $,
843: characteristic of the dilute solutions, gets disrupted:
844: the oscillations regain larger and larger amplitude until
845: $\mu^2=\lambda^2 a^2 \sim \log \frac1{a} $.
846: 
847: \item $ \mu^2  = {\cal O}(\log \frac1{a}) $. Here, a sudden drop to very small
848: $ \eta=\eta_{\rm min}(a)\sim a $ takes place [see fig. \ref{fig:eta1}]; 
849: then $ \eta $ rises again 
850: as $ \mu \; a $, for $ \mu \; a $ large enough, and keep growing 
851: indefinitely (see  fig. \ref{fig:eta1} and below). 
852: This regime corresponds to saddles interpolating
853: between the dilute phase and a collapsed phase (see below).
854: 
855: \item $ \mu^2  = {\cal O}(\frac1{a}) $. This regime describes collapsed
856: configurations where all particles are concentrated in a region of size 
857: $ \sim a $ and provide the absolute minimum of the free energy.
858: 
859: \end{itemize}
860: 
861: Notice that new maxima appear in addition to those present
862: for $ a = 0 $ (compare with Fig. \ref{fig:eta0}). Those new maxima
863: turn to be degenerate with the old ones up to small corrections of order $ a $.
864: In addition, $ \eta $ exhibits oscillations for  $ \mu  = {\cal O}(1) $
865: which are similar to those present at $ a=0 $ and $ \eta \to 2 $
866: but reversed and faster. We shall give below a qualitative explanation for
867: this peculiar behavior.
868: 
869: The new dependence of $\eta$ on $\lambda$ described above implies that the
870: number of solutions (that is the number of distinct $ \lambda $ for any given $
871: \eta $) is finite for any $ \eta $ and that, most importantly, there appear
872: new solutions. In particular, there are two new solutions for $ \eta_{\rm
873:   min}(a)<\eta<\eta_2 $ in addition to the unique solution regular as $ a
874: \to 0 $ and there is one new solution for $ \eta > \eta_C $, the region not
875: accessible to dilute solutions. These new solutions are singular in the
876: limit $ a \to 0 $, as evident from Fig.~\ref{fig:chifit}, where we plot $
877: \phi(r) $ vs. $ r $ for few values of $ a $ at a fixed value of $\eta$ and
878: $\lambda $ placed in the sudden drop of $ \eta $ vs. $ \log\lambda $ (the
879: black circles in Fig.~\ref{fig:eta1}, scaling as $ \mu^2 = \lambda^2 a^2 
880: \sim \log \frac1{a} $).
881: 
882: \begin{figure}[ht]
883:   \centering  
884:   \psfrag{log10r}{$\log_{10}r$}
885:   \psfrag{phiofr}{$\phi(r)$}
886:   \psfrag{eta1.5014}{$\eta = 1.5014\ldots$}
887:   \psfrag{azero}{$a=0$}
888:   \psfrag{a104}{$\;a=10^{-4}$}
889:   \psfrag{a105}{$\;a=10^{-5}$}
890:   \psfrag{a106}{\hspace{-0.8cm}$a=10^{-6}$}
891:   \psfrag{undiffeq}{unregularized differential equation}
892:   \psfrag{reginteq}{$\qquad$regularized integral equation}
893:  \includegraphics[width=.75\textwidth]{chifit.eps}
894:  \caption{ Profiles of the mean field $ \phi(r) $ for $ \mu^2 = {\cal O}(\log
895:    \frac1{a}) $ (interpolating solutions), with comparison between the
896:    regularized and unregularized cases. The dashed black curve is the
897:    dilute solution (that is for $ a = 0 $) at the same value of $ \eta $,
898:    namely at the location of the magenta circle of Fig.~\ref{fig:eta1}.
899: The other curves correspond to the black circles in fig. \ref{fig:eta1}.}
900:   \label{fig:chifit} 
901: \end{figure}
902: 
903: In Fig.~\ref{fig:chifit} we also compare the solutions of the integral
904: equation (\ref{ecinar}) or (\ref{eq:varinteq}) to those of the differential
905: equation (\ref{eqrad}), with boundary conditions~(\ref{eq:bc}), in which
906: $Q$ takes the value obtained by the integral method, upon using
907: eqs.~(\ref{eq:phixi}) and~(\ref{eq:phi0}). We see that the
908: agreement is very good down to $ r $ of order $ a $, when the short--distance
909: regularization becomes effective, while the solutions of the differential
910: equation (\ref{eqrad}) eventually blow to $ -\infty$ as $ r\to0 $. This
911: agreement appears very natural from the integro--differential formulation
912: of eq.~(\ref{ecdar}), where the short--distance regulator $ a $ can play a
913: significant role only for $ r\lesssim a $.
914: 
915: 
916: \begin{figure}[ht]
917:   \centering  
918:   \psfrag{eta1.6108}{$\eta = 1.6108\ldots$}
919:   \psfrag{logr}{$\log_{10} r$}
920:   \psfrag{a1e-4}{$a = 10^{-4}$}
921:   \psfrag{logrho}{$\log \rho$}
922: \includegraphics[width=.75\textwidth]{collrho.eps}
923: \caption{Profiles of the logarithm the density $ \log \rho(r) $ for 
924:  $ \mu^2 = {\cal O}(\frac1{a}) $ (collapsed solution). For the value of
925:   $\eta$ given, we have $\lambda=6012149.6972\ldots$. This collpased
926:   profile should be compared to the dilute ones (stable and unstable)
927:   depicted in Fig.~\ref{mrho}. }
928:   \label{fig:collrho} 
929: \end{figure}
930: 
931: Hence, the qualitative behaviour of the singular solutions can be understood
932: already from the differential equation (\ref{eq:fund}), provided one drops
933: the initial conditions $ \chi(0)=\chi'(0)=0 $ characteristic of the dilute
934: solutions. That is, using the relation $ \chi(\mu\, x)=\mu^2 \, \,\xi(x) $
935: that follows from eqs. (\ref{eq:phichi}) and (\ref{eq:phixi}),
936: we can solve the integral equation (\ref{eq:varinteq}) just in the interval
937: $ 0\le x\le x_1 $, with $ x_1\gtrsim 2 $, to compute 
938: \begin{equation}\label{codin1}
939: \chi(2 \, \mu)=\mu^2\, \xi(x_1) \quad {\rm and}
940: \quad   \chi'(2 \, \mu)=\mu \, \xi'(x_1)  \; ,
941: \end{equation}
942: at the boundary $ x=x_1 $. These values for $ \chi(2 \,\mu) $ and $ \chi'(2
943: \,\mu) $, both negative by construction, can then be used as initial
944: conditions at $ z = 2 \, \mu $ to solve the differential equation
945: (\ref{eq:fund}) in place of the conditions $ \chi(0)=\chi'(0)=0 $,
946: characteristic of the dilute solutions. The choice $ x_1\gtrsim 2 $ is
947: motivated by a direct numerical analysis but can be explained by the fact
948: that $ \mu $ diverges when $ a\to0 $ in the solutions which are singular as
949: $a\to0$. This implies, on one side, that the r.h.s. of
950: eq.~(\ref{eq:vardiff}) is exponentially small in $ \mu^2 $ when $
951: x_0\gtrsim 2 $ [see also eq. (\ref{mx})]
952:  and, on the other side, that the initial conditions at $ z =
953: 2 \, \mu $ affect the function $ \chi(z) $ very strongly, causing the rapid
954: drop for $ r > 2 \, a $ observed in Fig.~\ref{fig:chifit}, as well as the
955: growth followed by the fall to $ -\infty $ as $ r \to 0 $.  It must be
956: observed, in fact, that $ \xi(x) $ has a finite limit as $ \mu \to
957: \infty $ [see below eq.(\ref{scolap})], so that to leading order and 
958: $ \mu \gg 1 $, using eq.(\ref{codin1}), 
959: $$ \chi(2 \, \mu) = {\cal O} (\mu^2) \quad {\rm and } \quad 
960:  \chi'(2 \, \mu) = {\cal O}(\mu) \; . 
961: $$
962: It follows that the function $ \chi(z) $ fulfills approximately the
963: free equation $ (z\chi)''=0 $ in the rapid drop for $z>2\,\mu$ as well as in
964: the following plateau (to the right in Fig.~\ref{fig:chifit}), that is 
965: \begin{equation}\label{eq:free}
966:   \chi(z) \simeq \frac{c_0}z + c_1 \quad, \qquad c_0 = {\cal O}(\mu^3)\;,
967:   \quad c_1 = {\cal O}(\mu^2)\;. 
968: \end{equation}
969: Where we used that $ z = \lambda \; r = \mu \; x $.
970: This behaviour is valid as long as $ \chi''(z)\lesssim e^{\chi(z)} $, that is
971: $ z\lesssim {\cal O}(\mu\, \exp[\mu^2/3]) $ since $ \chi= {\cal O}(\mu^2) $
972: and $  c_0 = {\cal O}(\mu^3) $. For larger values of $ z $ the
973: function $ \chi(z) $ tends to the unique large distance fixed point given by
974: the purely logarithmic solution $ \log(2/z^2) $ and does this through the
975: characteristic damped harmonic oscillations in $\log\lambda$\cite{viej}
976: responsible for the waving behaviour of $ \eta $ in the dilute phase 
977: [see fig. \ref{fig:eta0}]. Since
978: the crossover to the logarithmic behaviour takes place for larger and
979: larger values of $ z $ the smaller is $ a $, because $ \mu = 
980: {\cal O}\left( \sqrt{ \log \frac1{a} } \right) $ grows with $ a $ as well
981: as the crossover point $ z =  {\cal O}(\mu\, \exp[\mu^2/3]) $. 
982: When $ \chi(z) $ is almost constant,
983: the characteristic oscillations tend to have the same amplitude they have
984: in the regular solutions. Moreover, this almost constant value of
985: $ \chi(z) $, which is of order $ \mu^2 $, decreases faster in $ \mu $ than the
986: end point $ \lambda=\mu/a $ at which $ \chi(z) $ is to be evaluated to give
987: $ \eta $ as in the second of eqs.~(\ref{eq:bc2}). This explain why the
988: oscillations of $\eta$ are reversed and faster right before the sudden drop
989: in Fig.~\ref{fig:eta1}. In other words, the oscillations in the left part
990: of fig. \ref{fig:eta1} describe the approach to the limiting
991: solution $ \log(2/z^2) $ [$\eta = 2$] while one is getting {\bf off} 
992: this limiting solution in the right part of fig. \ref{fig:eta1}.
993: This explains why the oscillations on the right get reversed compared
994: with the oscillations on the left of fig. \ref{fig:eta1}.
995: 
996: In the next two subsections we will provide more details on all these
997: matters and further analytical insights.
998: 
999: \subsection{Collapsed phase}
1000: 
1001: This is the case when $ \mu\sim \eta/a $ as $ a\to0 $ and corresponds to the
1002: growing branch on the right of Fig.~\ref{fig:eta1}. Notice that we have
1003: collapsed solutions for arbitrarily large values of $ \eta $ contrary
1004: to the dilute solutions which only exist for $ \eta < \eta_C $.
1005: 
1006: Since $ \xi(x) $ is negative and monotonically decreasing for $x>0$, the
1007: quantity $ \exp[\mu \, \xi(x)] $ is exponentially small except in the
1008: interval $ 0 \le x <x_0 $ where $ \xi(x) $ would vanish in the limit. We
1009: use now this property to get an approximate but analytic singular solution.
1010: That is, the dominant contribution in the r.h.s. of eq.~(\ref{eq:varinteq})
1011: is obtained when all integrations are restricted to the interval $ 0 \le x
1012: \le x_0 $. One easily realizes that, by consistency, $ x_0=1/2 $. Then one
1013: finds
1014: \begin{equation*}
1015:   I_n(x) = \frac1{n+1} 
1016:   \begin{cases}
1017:     x^{n+1} \;,& x \le 1/2 \\ 2^{-n-1} \; , & x \ge 1/2
1018:   \end{cases}
1019: \end{equation*}
1020: and in particular using eqs.(\ref{eq:phi0}) and (\ref{eq:etamu}),
1021: \begin{equation*}
1022:   \phi(0) = \mu^2 \left[I_2(1) + I_1\left(\frac1{a}\right) - I_1(1)\right] = 
1023:   \frac{\mu^2}{24} \quad,\qquad
1024:     \eta = \mu^2\, a \,I_2\left(\frac1{a}\right) = \frac{\mu^2\,a}{24} 
1025: \quad,\qquad \phi(0) = \frac{\eta}{a} \;.
1026: \end{equation*}
1027: Finally, the leading form for $\phi(r)$ reads
1028: \begin{equation}\label{scolap}
1029:   \phi(r) = \phi(0) + \frac{24\,\eta}a\, \xi\left(\frac{r}{a}\right) = 
1030:   \begin{cases}\displaystyle
1031:     \frac{\eta}a \;, & r\le a/2 \\[0.3cm]\displaystyle
1032:     \frac{\eta}a +\frac{\eta}{2r}\left(\frac{r}a - \frac12 \right)^3 
1033:     \left(\frac{r}a - \frac52 \right) \;, & a/2 \le r \le 3a/2 \\[0.3cm]
1034:     \displaystyle \frac{\eta}r \;, &r \ge 3a/2 \;.  
1035:   \end{cases}
1036: \end{equation}
1037: Notice that $ \phi(r) $ as well as its first and second derivatives are
1038: continuous at $ r = a / 2 $ and $ r = 3 \, a / 2 $. For $r \ge 3a/2$, the
1039: effective gravitational field has the free form proper of a locally
1040: vanishing mass density.
1041: 
1042: Higher corrections can be obtained by iteration over
1043: eq.~(\ref{eq:varinteq}). That is, using the leading form of $\xi(x)$
1044: that can be read out of in eq.~(\ref{scolap}) to evaluate to
1045: next--to--leading order, as $\mu\to\infty$, the integrals $I_n(x)$ in the
1046: r.h.s. of eq.~(\ref{eq:varinteq}). This procedure can then be repeated to
1047: obtain further corrections.
1048: 
1049: Eq.~(\ref{scolap}) provides the spherically symmetric, leading order
1050: singular solution of the mean field equation (\ref{eq;BPLE}) where the
1051: particles are densely concentrated around one point in a region of size of
1052: the order of the cutoff $ A $.  In fact, to leading order in $ a $ the density 
1053: reads
1054: \begin{equation*}
1055:   \rho(r) = 
1056:   \begin{cases}\displaystyle
1057:     \frac{6}{\pi\,a^3} \;, & r\le a/2 \\[0.3cm]\displaystyle
1058:     0 \;, &r\ge a/2 \;,
1059:   \end{cases}
1060: \end{equation*}
1061: while to next--to--leading order we have
1062: \begin{equation}\label{eq:nnlo}
1063:   \rho(r) = \frac{6}{\pi\,a^3} \exp\Big[\phi(r) -\frac{\eta}a \Big] \;,
1064: \end{equation}
1065: where $\phi(r)$ has the leading order form of eq.~(\ref{scolap}). This
1066: corresponds to the situation when the gravitational attraction completely
1067: overcomes the kinetic energy and all particles fall towards the origin. Of
1068: course, such singular solution only exist mathematically for non-zero
1069: values of the cutoff $ A $.
1070: 
1071: Let us now compute the action for the solution eq.~(\ref{scolap}) from
1072: eq.~(\ref{accefa}). The regularized laplacian can be computed  from the
1073: equation of motion (\ref{eq;BPLE}) with the leading order result for 
1074: $ a\to 0 $,
1075: \begin{equation*}
1076:   \int_{|\vR|\le 1} d^3 r \; \phi({\bds r}) \; \nabla^2_a  \phi({\bds r}) =
1077:   - \frac{4 \, \pi \, \eta}{Q} \; 4 \, \pi \int_0^1 r^2 \; dr \; 
1078:   \phi(r) \, e^{\phi(r)} = - 4 \, \pi \; \eta \; \phi(0) = -\frac{4 \,
1079:     \pi \, \eta^2}{a} \; .
1080: \end{equation*}
1081: The second term is just 
1082: \begin{equation*}
1083:   \log Q = \phi(0) - \log\frac{\lambda^2}{4\pi\,\eta} =
1084:   \frac{\eta}a - \log\frac6{\pi\,a^3} \;,
1085: \end{equation*}
1086: so that all together, to leading order we obtain for the free energy
1087: \begin{equation}\label{acsing}
1088:   \begin{split}
1089:     F-F_0 =\frac{Gm^2\beta\,N}{2A} + NT\,s(\eta,a) &= \frac{Gm^2\beta\,N}{2A}
1090:     -\frac{NT\,\eta}{2\,a } + NT\,\log\frac8{a^3}\\ & =
1091:     -\frac{Gm^2N(N-1)}{2\,A} +NT\,\log\frac{R^3}{(A/2)^3}\;.
1092:   \end{split}
1093: \end{equation}
1094: The first term is just the potential energy of $ N $ particles clustered in
1095: a small sphere of radius $A/2$, where the regularized gravitational
1096: interaction is the same for all $\frac12 N(N-1)$ particle pairs. The second
1097: term is $T$ times the entropy loss in the collapse.  As one could expect,
1098: this free energy is large and negative, unbounded from below as $A\to0$. In
1099: particular, this free energy is well below the free energy of the dilute
1100: solution eq.~(\ref{seta}) and Fig.  \ref{fig:action0} for all values $ \eta
1101: > 0 $.  That is, the singular solution eq.~(\ref{scolap}) provides the {\bf
1102:   absolute} minimum for the free energy eq.~(\ref{accefa}) for any $ \eta >
1103: 0 $. The dilute solution eqs.(\ref{eq:fund}) and (\ref{eq:bc}) is only a
1104: {\bf local} minimum. In the next subsection we shall address the issue of
1105: the metastability of the dilute phase.
1106: 
1107: Let us now compute some physical quantities characterizing the collapsed
1108: phase.
1109: 
1110: The equation of state has the mean field form
1111: \begin{equation}\label{eq:eos}
1112:     \frac{P \,V}{N\,T}=\frac{4\pi}3\, \rho(1) \;, 
1113: \end{equation}
1114: in both the dilute and collapsed phase. In the former case the r.h.s. of
1115: eq.~(\ref{eq:eos}) coincides with the function $f(\eta)$. In the latter 
1116: we obtain instead, from eqs.~(\ref{eq:nnlo}) and~(\ref{scolap}), to
1117: next--to--leading order
1118: \begin{equation}\label{eq:eos2}
1119: \frac{P\,V}{N\,T}=\left(\frac2{a}\right)^3 \; 
1120: \exp\Big[\eta -\frac{\eta}a \Big]  \;.
1121: \end{equation}
1122: Thus, the pressure at the boundary is exponentially small in the collapsed
1123: phase. The complete plot of  $ PV/NT $ vs. $ \eta $ is given in
1124: Fig.~\ref{fig:ffun}. One can see how the well known \cite{viej} spiraling
1125: towards $ (\eta=2,f(\eta)=1/3) $, characteristic of the dilute solutions, is
1126: now limited to the case when $ \mu $ is of order $ a $ . When $ \mu $ reaches
1127: values of order $ 1 $ the inward spiral comes to a halt and then winds back
1128: until $ \mu \lesssim {\cal O}(\log \frac1{a}) $. For larger values of $ \mu $, 
1129: that is $ \mu\sim \eta/a $ as in the collapsed phase, $ PV/NT $ drops to values
1130: exponentially small in $ \frac1{a} $. Analogous plots are obtained for other 
1131: types of cutoff in refs. \cite{otros}.
1132: 
1133: \begin{figure}[ht]
1134:   \centering  
1135:   \psfrag{etavar}{$\eta$}
1136:   \psfrag{ffun}{$PV/NT$}
1137:   \psfrag{  a1e-4}{$\;a=10^{-4}$}
1138:   \psfrag{muoa}{$\mu={\cal O}(a)$}
1139:   \psfrag{muo1}{$\mu={\cal O}(1)$}
1140:   \psfrag{muologa}{$\mu^2={\cal O}(\log \frac1{a} )$}
1141:   \psfrag{muo1oa}{$\mu^2={\cal O}(\frac1{a})$}
1142:   \includegraphics[width=.75\textwidth]{ffun2.eps}
1143:   \caption{Complete plot of $ PV/NT $ vs. $ \eta $ when $ a=10^{-4} $ (blue
1144:     curve). When $ \mu $ is of order $ a $ the blue curve cannot be
1145:     distinguished from the corresponding curve of the case $ a=0 $, here
1146:     depicted in red (dilute solutions). 
1147: When $ \mu $ is of order $ 1 $ the blue curve ($ a>0
1148:     $) and the red curve separate. The blue curve then traces the red
1149:     spiral backwards at a distance of order $ a \; (\log a)^2 $ as long as
1150:     $ \mu^2 \lesssim {\cal O}(\log \frac1{a}) $ (interpolating solutions).
1151:  Finally the blue curve drops to zero when $ \mu^2 $ is of order $\frac1{a}$
1152: (collapsed solutions).}
1153:   \label{fig:ffun} 
1154: \end{figure}
1155: 
1156: For other physically relevant quantities we similarly find
1157: \begin{eqnarray}
1158:   c_V = \frac32 \quad , \quad c_P = \frac32 + {\cal O}(a^2)
1159:   \quad , \quad  \frac{v_s^2}{T} = \frac1{\kappa_T} = 
1160:   \frac8{a^3} \exp\Big[-\frac{\eta}a \Big]
1161:   \quad , \quad \frac{S - S_0}{N} = 3 \; \log a + {\cal O}(a^0) \; .
1162: \end{eqnarray}
1163: 
1164: \subsection{Interpolating solution: metastability}\label{metaB}
1165: 
1166: The dilute solutions studied in sec. IV are metastable because the
1167: collapsed solution provides the absolute minimum for free energy
1168: [eq.(\ref{acsing})]. The collapse transition from the dilute phase to the
1169: collapsed phase can take place through an interpolating solution in an
1170: analogous way to the bubbles in liquid-gas first order phase transitions
1171: \cite{transi}.
1172: 
1173: Such interpolating solutions are the saddle points of eq.~(\ref{eq;BPLE}),
1174: which correspond to the sudden drop of $ \eta $ as a function of
1175: $\log\lambda$ (see Fig.~\ref{fig:eta1}) located approximately at $ \lambda^2
1176: a^2\sim \log \frac1{a} $. In this narrow drop region we have 
1177: $ {\cal O}(a) \lesssim \eta < \eta_C $, so the interpolating solutions 
1178: are allowed only in this interval of $ \eta $. Examples of the interpolating 
1179: solutions are given in fig.~\ref{fig:chifit}.
1180: 
1181: According to the qualitative discussion below eq.~(\ref{eq:free}), the
1182: location of the $ \eta $ drop can be obtained approximately as the values
1183: of $ \lambda $ at which the function $ \chi(\lambda) $ has the crossover from
1184: the free solution $ \chi(\lambda)\simeq c_0/\lambda +c_1 $ to the logarithmic
1185: solution $ \log(2/\lambda^2) $. That is for $ \lambda^2 a^2 \sim \log
1186: \frac1{a} $, since $ c_0={\cal O}(\lambda^3a^3) $ and $ c_1={\cal
1187:   O}(\lambda^2a^2) $, see eq.~(\ref{eq:free}). Then the scaling as $a\to0$
1188: of $\rho(0)$, the density in the origin, can be obtained from
1189: eq.~(\ref{eq:phixi}) and the numerical analysis as
1190: \begin{equation*}
1191:   \rho(0) = \frac{\lambda^2}{4\pi\,\eta} = 
1192:   {\cal O}\big(a^{-2}\log\tfrac1{a}\big) \;.
1193: \end{equation*}
1194: because $ \eta = {\cal O}(1) $ in this regime.
1195: 
1196: On the other hand, from
1197: Fig.~\ref{fig:chifit} one sees that the value of the density for $r\gg a$
1198: and in particular the value at the border $ \rho(1) $ differ by order $a$
1199: w.r.t. the dilute phase. The equation of state (\ref{eq:eos}), which holds
1200: for any value of $ \lambda $, then implies that the pressure in the dilute
1201: and interpolating solution, for the same value of $ \eta $, differ only by
1202: order $ a $, as evident also from Fig.~\ref{fig:ffun}.
1203: 
1204: In practice, one can regard an interpolating configuration as a special
1205: spherical fluctuation of the dilute phase in which a fraction of particles
1206: of order $ a \, \log\tfrac1{a} $ is almost uniformely removed from everywhere 
1207: and concentrated in a region of size $ a $ around the origin. In fact, this
1208: result has an heuristic explanation by a simple energy--entropy argument: if
1209: we denote by $ \Delta\rho $ the variation of the density in a region of size
1210: $a$ around the origin, we then have
1211: \begin{equation}\label{eq:enent} 
1212:   \Delta s  \sim  -\frac\eta{2\,a} \,  \left(\frac{\Delta N}N\right)^2 
1213:   + 3 \, \frac{\Delta N}N \,\log\frac1{a} \quad,\qquad 
1214:   \frac{\Delta N}N \equiv \,a^3 \, \Delta\rho\ ; ,
1215: \end{equation}
1216: since the first term is the gain in potential energy per particle and the
1217: second the loss of entropy per particle due to the concentration. We may
1218: neglect the effect due to the rest of the particles if we assume that
1219: $\frac{\Delta N}N$, the fraction of particle moved around, would vanish as
1220: $a\to0$. Minimizing eq.~(\ref{eq:enent}) w.r.t. $\Delta\rho$ yields
1221: \begin{equation}\label{eq:drho}
1222:   \Delta\rho \simeq \frac3{\eta\,a ^2} \, \log\frac1{a} \quad {\rm and}
1223:  \quad \Delta s = \frac{9 \, a}{2 \, \eta} \left(\log\frac1{a}\right)^2 \;,
1224: \end{equation}
1225: which is indeed consistent with the requirement that $ a^3 \; \Delta\rho $ 
1226: vanishes as $ a\to0 $ and rather accurately approximate our numerical results.
1227: 
1228: Notice that the minimized $ \Delta s $ is to be identified with the
1229: difference, between interpolating and dilute solution, of the action per
1230: particle, so that 
1231: \begin{equation}\label{ds}
1232: \Delta s =  s_{\rm interpolating} - s_{\rm dilute} 
1233: \simeq \frac9{2\,\eta}\,a\,(\log   a)^2 \;.
1234: \end{equation}
1235: This result is in very good agreement with the numerical values
1236: obtained by solving eqs.(\ref{ecinar}) and (\ref{accir}).
1237: 
1238: According to the interpretation of the interpolating solutions as saddle
1239: points through which the dilute phase may decay to the collapsed one, 
1240: the lifetime of the dilute phase is given by
1241: \begin{equation}\label{vida}
1242:   \tau \simeq \Big(\frac{|{\rm Det}|}{{\rm Det}_0}\Big)^{1/2}
1243:   \,e^{N \Delta s} \, \;,
1244: \end{equation}
1245: where $ \rm Det $ stands for the determinant of small fluctuations around
1246: the interpolating solution and $ {\rm Det}_0 $ for that around the
1247: dilute solution. 
1248: 
1249: This ratio of determinants can be estimated as follows: in the
1250: interpolating profile there should be only one negative mode necessarily
1251: localized in a region of size $ a $ around the origin, where the density is
1252: very large, since $-\rho(r)$ plays the role of Schroedinger potential in
1253: the eigenvalue equation for the small fluctuations. Eq.~(\ref{eq:drho})
1254: then implies that $ \frac1{a^2}\, \log \frac1{a} $ provides the
1255: scale of the potential felt by this negative mode. Hence the corresponding
1256: negative eigenvalue should be of the same order, while the rest of the
1257: spectrum should be almost unaffected, since the two density profiles differ
1258: only by order $a$ for $r \gtrsim 2\,a$. Therefore
1259: \begin{equation}\label{dete}
1260:   \frac{\rm |Det|}{{\rm Det}_0}\simeq  \frac1{a^2}\, \log \frac1{a} \; .
1261: \end{equation}
1262: We obtain from eqs. (\ref{ds})-(\ref{dete}) to leading order the following
1263: lifetime for the dilute phase of the self-gravitating gas:
1264: \begin{equation}\label{vida2}
1265:   \tau \sim \frac1{a}\,\sqrt{\log \frac1{a}} \; 
1266:   e^{\frac{9 \, N}{2\,\eta}\,a\,(\log   a)^2} \sim 
1267:   \frac{R}{A} \, \sqrt{\log \frac{R}{A}} \; 
1268:   \exp\Big[\frac{9\,A\,T}{2\,G\,m^2}\,
1269:   \Big(\log\frac{R}A\Big)^2 \Big] \;.
1270: \end{equation}
1271: One can see that the lifetime becomes infinitely long in the zero cutoff
1272: limit as well as when $N\to\infty$ at fixed cutoff
1273: [recall that $ R\sim N $ in the dilute limit of eq.~(\ref{limter})].
1274: 
1275: We want to stress that the lifetime of the dilute solutions
1276: is extremely long for large $ N $ and small cutoff $ A $. 
1277: This is due to the fact
1278: that although the collapsed solution has a much lower energy than the
1279: dilute solution, in the latter there is a huge entropy barrier against the
1280: gathering, in a small domain of size $ \sim A $, of a number particles
1281: large enough to start the collapse. A qualitatively similar conclusion is
1282: reached in ref. \cite{cha}.
1283: 
1284: We shall come back for a more detailed study of this problem in a
1285: subsequent paper.
1286: 
1287: 
1288: \begin{thebibliography}{}
1289: 
1290: \bibitem{astro} J. Binney, S. Tremaine, `Galactic Dynamics',
1291: Princeton Univ. Press, Princeton, NJ,  1987. P. J. E. Peebles,
1292: `Principles of Physical Cosmology', Princeton Univ. Press, Princeton,
1293: NJ,  1993. W. C. Saslaw, `Gravitational Physics of Stellar and Galactic
1294: Systems', Cambridge Univ. Press, Cambridge 1987.
1295: 
1296: \bibitem{ism} R. B. Larson, MNRAS {\bf 194}, 809 (1981).
1297: J. M. Scalo, in `Interstellar Processes', D.J. Hollenbach and
1298: H.A. Thronson Eds., D. Reidel Pub. Co, p. 349 (1987).
1299: 
1300: \bibitem{viej} R. Emden, `Gaskugeln', Teubner, Leipzig und Berlin,
1301: 1907. S. Chandrasekhar, `An Introduction to
1302: the Study of Stellar Structure', Chicago Univ. Press, 1939.
1303: R. Ebert, Z. Astrophys. {\bf 37}, 217 (1955).
1304: W.B.Bonnor, Mon. Not. R. astr. Soc. {\bf 116}, 351 (1956).
1305: V. A. Antonov, Vest. Leningrad Univ. 7, 135 (1962).
1306: D. Lynden-Bell and R Wood, Mon. Not. R. astr. Soc. {\bf 138}, 495 (1968).
1307: 
1308: \bibitem{viej2}  E. B. Aronson, C. J. Hansen, Astrophys. J.  177, 145 (1972),
1309: B. Stahl, M. K. H. Kiessling, K. Schindler, Planet. Space Sci. 43, 271 (1995),
1310: P. Hertel, W. Thirring, Commun. Math. Phys.  24, 22  (1971).
1311: 
1312: \bibitem{I} H. J. de Vega, N. S\'anchez, Nuclear Physics
1313: {\bf B 625}, 409 (2002), astro-ph/0101568. 
1314: 
1315: \bibitem{II} H. J. de Vega, N. S\'anchez, Nuclear Physics
1316: {\bf B 625}, 460 (2002), astro-ph/0101567. 
1317: 
1318: \bibitem{pal}  H. J. de Vega, N. S\'anchez, 
1319:    `Statistical Mechanics of the self-gravitating gas: thermodynamic limit, 
1320: phase diagrams and fractal structures', astro-ph/0505561, 9th Course of the 
1321: International School of Astrophysics `Daniel Chalonge', 
1322: Palermo, Italy, 7-18 September 2002, NATO ASI,  p. 291-324 in the
1323: Proceedings, N. S\'anchez and Yu. Parijskij editors, Kluwer, 2002.
1324: 
1325: \bibitem{nos}  H. J. de Vega, J. Siebert,  Phys. Rev. {\bf E66}, 016112 (2002),
1326: Nucl. Phys. {\bf B 707}, 529 (2005) and {\bf B 726}, 464 (2005).
1327: H. J. de Vega, N. S\'anchez, Nucl. Phys. {\bf B711}, 604 (2005).
1328: 
1329: \bibitem{otros} P. H. Chavanis, Phys. Rev. E 65, 056123 (2002),
1330: P. H. Chavanis, I. Ispolatov, Phys. Rev. E  66, 036109 (2002),
1331: P. H. Chavanis, J. Sommeria, MNRAS 296, 569 (1998).
1332: E. Follana, V. Laliena, Phys. Rev. E 61, 6270 (2000).
1333: 
1334: \bibitem{transi}   Langer J S in \textit{Fluctuations,
1335: Instabilities and Phase Transitions} ,   Riste T,  Ed. page 19,
1336: Plenum N.Y. (1975). Langer J S in  \textit{Solids Far
1337: From Equilibrium} , Ed.   Godreche C, Cambridge Univ. Press
1338: (1992), page 297, and  \textit{Systems Far From Equilibrium},
1339: Garrido L, et.\ al.\ eds. \textit{Lect. Notes in Phys.} {\bf 132}
1340: Springer (1975). Gunton J D,   San Miguel M and
1341: Sahni P S  in  \textit{Phase Transitions and Critical Phenomena} ,
1342:   Domb C and   Lebowitz J J, Eds. Vol 8, Academic Press, (1983);
1343:   Langer JS, \textit{Acta Metall.} {\bf 21}, 1649 (1973).
1344: D. Boyanovsky, H. J. de Vega, D. J. Schwarz,
1345: `Phase transitions in the early and the present Universe', 
1346: hep-ph/0602002,  to appear in Ann. Rev. Nucl. Part. Sci.
1347: 
1348: \bibitem{cha} P. H. Chavanis, A\&A 432, 117 (2005).
1349: 
1350: \end{thebibliography}
1351: 
1352: \end{document}
1353: 
1354: