1: % ****** Start of file apssamp.tex ******
2: %
3: % This file is part of the APS files in the REVTeX 4 distribution.
4: % Version 4.0 of REVTeX, August 2001
5: %
6: % Copyright (c) 2001 The American Physical Society.
7: %
8: % See the REVTeX 4 README file for restrictions and more information.
9: %
10: % TeX'ing this file requires that you have AMS-LaTeX 2.0 installed
11: % as well as the rest of the prerequisites for REVTeX 4.0
12: %
13: % See the REVTeX 4 README file
14: % It also requires running BibTeX. The commands are as follows:
15: %
16: % 1) latex apssamp.tex
17: % 2) bibtex apssamp
18: % 3) latex apssamp.tex
19: % 4) latex apssamp.tex
20: %
21: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
22: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
23:
24: % Some other (several out of many) possibilities
25: %\documentclass[preprint,aps]{revtex4}
26: %\documentclass[preprint,aps,draft]{revtex4}
27: %\documentclass[prb]{revtex4}% Physical Review B
28:
29: \usepackage{graphicx}% Include figure files
30: \usepackage{dcolumn}% Align table columns on decimal point
31: \usepackage{bm}% bold math
32:
33: %\nofiles
34:
35: \begin{document}
36:
37: \preprint{APS/123-QED}
38:
39: \title{Core Collapse Via Coarse Dynamic Renormalization}
40: % Force line breaks with \\
41:
42: \author{Andras Szell}
43: %\email{szell@astro.rit.edu}
44: \author{David Merritt}
45: %\email{merritt@astro.edu}
46: \affiliation{Rochester Institute of Technology, 54 Lomb Memorial Drive,
47: Rochester, NY 14623}
48:
49: \author{Ioannis G. Kevrekidis}
50: \affiliation{Department of Chemical Engineering and Program in Applied and Computational Mathematics, Princeton University,
51: Princeton, NJ 08544}
52: %\email{yannis@princeton.edu}
53:
54: \date{\today}% It is always \today, today,
55: % but any date may be explicitly specified
56:
57: \begin{abstract}
58: In the context of the recently developed ``equation-free'' approach
59: to computer-assisted analysis of complex systems,
60: we extract the self-similar solution describing core collapse
61: of a stellar system from numerical experiments.
62: The technique allows us to side-step the core ``bounce''
63: that occurs in direct $N$-body simulations due to the small-$N$
64: correlations that develop in the late stages of collapse,
65: and hence to follow the evolution well into the self-similar regime.
66: \end{abstract}
67:
68: \pacs{Valid PACS appear here}% PACS, the Physics and Astronomy
69: % Classification Scheme.
70: %\keywords{Suggested keywords}%Use showkeys class option if keyword
71: %display desired
72: \maketitle
73:
74: \section{Introduction}
75:
76: In many areas of current research,
77: physical models are available at a fine,
78: microscopic scale, while the questions
79: of most interest concern the system
80: behavior on a coarse-grained, macroscopic level.
81: An example is the gravitational $N$-body problem
82: \cite{aarseth:03}.
83: Coarse-grained approximations exist, e. g.
84: the orbit-averaged Fokker-Planck equation
85: \cite{spitzer:87}, which approximates the full 6N
86: equations of motion by a differential operator
87: that acts on the single-particle distribution function
88: $f$.
89: But the derivation of the Fokker-Planck equation
90: from the full $N$-body equations of motion requires a
91: number of approximations.
92: The ``equation-free'' computational framework
93: \cite{kevr1,kevr2,kevr3} has recently been
94: proposed for the computer-assisted study of such
95: problems, circumventing the derivation of
96: explicit coarse-grained equations.
97: The coarse-grained behavior is analyzed directly,
98: through appropriately designed short computational
99: experiments by the fine-scale models.
100: In the case of problems where the macroscopic behavior
101: is characterized by scale invariance, dynamic renormalization
102: \cite{papa:1,papa:2,papa:3},
103: combined with equation-free computation and a template-based
104: approach \cite{rowley:05},
105: can in effect convert the self-similar problem into a
106: stationary one,
107: by working in a frame of reference that expands or
108: shrinks along with the macroscopic system observables.
109:
110: Here we apply equation-free dynamic renormalization techniques
111: to the problem of gravitational core collapse \cite{henon:61,henon:65}.
112: A star cluster or galaxy evolves due to gravitational
113: encounters between the stars, which drive the local velocity
114: distribution toward a Maxwellian.
115: Stars in the high-velocity tail of the Maxwellian
116: escape from the system; the probability of escape
117: in one relaxation time $t_r$,
118: \begin{equation}
119: t_r \equiv {0.065 v_m^3\over \rho m G^2\ln\Lambda},
120: \label{eq:tr}
121: \end{equation}
122: is roughly 1\%.
123: Here $v_m^2$ is the mean square (3D)
124: velocity of the stars, $\rho$ is the mass density,
125: $m$ is the individual stellar mass, $G$ is the gravitational
126: constant, and $\ln\Lambda\approx \ln(0.4N)$
127: is the Coulomb logarithm \cite{spitzer:87}, with
128: $N$ the number of stars in the cluster.
129: Escape occurs primarily from the high-density central
130: region, or ``core''; if the density contrast
131: between core and envelope is sufficiently
132: great, the core exhibits the negative specific heat
133: characteristic of gravitationally bound systems \cite{wood:68}
134: and contracts.
135:
136: Treatments of core collapse based on fluid
137: \cite{eggleton:80,louis:91}
138: or Fokker-Planck \cite{cohn:79,cohn:80,heggie:88} approximations
139: to the full $N$-body equations of motion
140: suggest that the late stages are self-similar,
141: \begin{subequations}
142: \begin{eqnarray}
143: \rho(r,t) &\approx& \rho_c(t)\rho_*\left[{r\over r_c(t)}\right], \\
144: \rho_c(t) &=& \rho_{c,0}\left(1-t/t_{cc}\right)^\gamma, \\
145: r_c(t) &=& r_{c,0}\left(1-t/t_{cc}\right)^\delta, \\
146: \rho_*(r) &\rightarrow& r^{-\alpha},\ \ \ r\gg r_c,
147: \end{eqnarray}
148: \end{subequations}
149: with $\rho_c$ and $r_c$ the central density and core radius
150: respectively, and $t_{cc}$ the time at which the central density
151: diverges; $t_{cc}-t$ is roughly $330$ times the relaxation time
152: of Eq. 1 if the latter is evaluated at the center of the cluster.
153:
154: These predictions are in reasonable agreement with the
155: results of direct $N$-body integrations
156: \cite{spurzem:96,makino:96,baumgardt:03}.
157: But when $N$ is finite, the number of particles in the
158: core drops to zero as collapse proceeds,
159: causing two-body and higher order correlations to
160: dominate the evolution; typically, binary stars form
161: which halt the collapse (``core bounce'').
162: In all $N$-body simulations published to date,
163: this occurs before or only shortly after the core
164: has entered the self-similar regime.
165: The rate of binary formation per unit volume due to three-body
166: interactions is \cite{hut:85}
167: \begin{equation}
168: \dot{n}_{\rm 3-b}\approx 1.2\times 10^2 {G^5m^2\rho^3\over v_m^9}.
169: \end{equation}
170: By the time that the number of stars in the core
171: has dropped to $N_c$ ($N_c\equiv \rho_cr_c^3$),
172: the total number of binaries formed is $\sim 10^3 N_c^{-2}$.
173: Since a single hard binary can halt the collapse,
174: bounce occurs when $N_c$ has dropped to $\sim 30$.
175: This limits the maximum central density that
176: can be reached in an $N$-body simulation
177: to a multiple $\sim 10^{3.5}(N/10^4)^3$ of the initial
178: density (assuming the initial state described below);
179: for $N\approx 10^4$, the achievable density contrast is
180: $\sim 10^4$, which is also roughly where
181: self-similar behavior first appears \cite{cohn:80}.
182:
183: In this paper, we exploit coarse dynamic renormalization to
184: compute representative self-similar solutions at scales
185: ``away from'' the latest stages of core collapse.
186: This allows us to avoid the finite-$N$ correlations
187: that develop at those stages and recover features of
188: the self-similar behavior predicted at the large-$N$ limit.
189: We define the ``macroscopic'' quantity of
190: interest to be the single-particle distribution function
191: $f(E)$, $E=v^2/2 + \Psi(r)$, with $\Psi(r)$ the self-consistent
192: gravitational potential (spherical symmetry is assumed throughout).
193: By dynamically renormalizing the $N$-body model after
194: each short ``burst'' of integration, we are able to
195: maintain a large ($\sim 10^3$) number of particles in the core
196: even with modest ($\sim 16K$) total numbers,
197: effectively halting binary formation and
198: allowing us to accurately extract the form of the
199: self-similar solution.
200:
201: \section{Description of the Calculations}
202: We adopted the standard Plummer model \cite{plummer:11} initial
203: conditions for this problem, with density,
204: gravitational potential, and single-particle distribution function
205: given by
206: \begin{eqnarray}
207: \rho(r) &=& {3\over 4\pi}{1\over (1+r^2)^{5/2}},\ \ \ \
208: \Psi(r) = -{1\over (1+r^2)^{1/2}}, \nonumber \\
209: f(E) &=& {24\sqrt{2}\over 7\pi^3} (-E)^{7/2}, \ \ \ \ E = v^2/2+\Psi.
210: \label{eq:plum}
211: \end{eqnarray}
212: Here and below, the gravitational constant $G$ has been set to one.
213: The $N$-body algorithm was an adaptation of S. J. Aarseth's {\tt NBODY1}
214: code \cite{aarseth:99} to the GRAPE-6 special-purpose
215: computer and included a fourth-order
216: particle advancement scheme with individual, adaptive, block time steps.
217: We first used this code to carry out a direct integration until
218: core bounce of a Plummer model with $N=16384$ particles;
219: the results (e.g. evolution of the central density, time of
220: core bounce) agreed well with published studies using different
221: $N$-body codes \cite{spurzem:96,makino:96,baumgardt:03}.
222: We used the same number of particles in the calculations described
223: below.
224:
225: Our coarse renormalization procedure consisted of short bursts
226: of simulation in a ``lift-simulate-restrict-rescale-truncate''
227: procedure, repeated in a loop until the asymptotic form of
228: the self-similar solution emerged.
229: In detail, the steps were:
230:
231: \noindent
232: 1. {\bf Lift:} Given a smooth representation $\rho(r)$ of the
233: density profile, a set of Monte-Carlo positions and velocities
234: was generated, as follows.
235: First an estimate of the gravitational potential
236: $\Psi(r)$ was computed from $\rho(r)$ via Poisson's equation.
237: The isotropic, single-particle distribution function $f(E)$
238: corresponding to this density-potential pair is given
239: uniquely by Eddington's formula \cite{eddington:16}.
240: For the purposes of generating a new Monte-Carlo set of
241: positions and velocities, the relevant function is not $f(E)$ but
242: $N(<v,r)$, the cumulative number of stars with velocities
243: less than $v$ at radius $r$.
244: Using Eddington's formula, this can be shown to be
245: \begin{eqnarray}
246: &&N(<v,r) = 1 - {1\over \rho}\int^{E}_0 d\Psi' {d\rho\over d\Psi'}
247: \times \nonumber \\
248: &&\left\{ 1 + {2\over\pi} \left[{v/\sqrt{2}\over\sqrt{E-\Psi'}} -
249: \tan^{-1}\left({v/\sqrt{2}\over\sqrt{E-\Psi'}}\right)\right]\right\}.
250: \end{eqnarray}
251:
252: \noindent
253: 2. {\bf Integrate.} The $N$-body realization was advanced in
254: time until the central density had increased by a factor of
255: $\sim 5$ compared with the initial model.
256: In the nearly-self-similar regime, this required
257: approximately 250 central relaxation times.
258:
259: \noindent
260: 3. {\bf Restrict.} An estimate of $\rho(r)$ was
261: computed from the particle coordinates ${\bf x}_i, i=1...N$ at the
262: final time step.
263: The position of the cluster center was first determined as in
264: \cite{casertano:85} and the particle radii $r_i$
265: were defined with respect to this center.
266: Each particle was then replaced by the kernel function
267: \begin{eqnarray}
268: K(r,r_i,h_i)&=&{1\over 2(2\pi)^{3/2}}\left({rr_i\over h_i^2}\right)^{-1}
269: e^{-\left(r_i^2+r^2\right)/2h_i^2} \nonumber \\
270: &\times&\sinh\left(rr_i/h_i^2\right)
271: \label{eq:kernel}
272: \end{eqnarray}
273: which is an average over the sphere of radius $r_i$ of the
274: 3D Gaussian kernel of width $h_i$.
275: The density estimate,
276: $\rho(r) = \sum_{i=1}^N h_i^{-3} K(r,r_i,h_i)$,
277: was then computed on a logarithmic radial grid.
278: The kernel width $h_i$ associated with the $i$th particle
279: was determined by first constructing
280: a pilot estimate of the density via a nearest-neighbor scheme,
281: then allowing the $h_i$ to vary as the inverse square root of this
282: pilot density \cite{silverman:87}.
283:
284: \noindent
285: 4. {\bf Rescale:} The density estimate was
286: rescaled,
287: $\rho(r) \rightarrow A\rho(Br)$,
288: according to an algorithm that left the core properties
289: unchanged in the self-similar regime.
290: The vertical normalization $A$ was fixed by comparing the values
291: of the density at $r=0$ at the start and end of the integration interval.
292: The radial scale factor $B$ was adjusted such that the two density
293: estimates had the same value at the radius where the
294: density was $1/50$ of its central value.
295:
296: \noindent
297: 5. {\bf Truncate:} In the late stages of core collapse,
298: the system develops a $\rho\sim r^{-\alpha}$, $\alpha\approx 2$
299: envelope around the shrinking core.
300: In direct $N$-body integrations, this envelope grows
301: to contain most of the total mass, and the number of stars in
302: the core drops to zero, leading to the small-$N$ effects
303: discussed above.
304: We avoided this by truncating the rescaled density
305: beyond a radius $r_1\approx 50 r_c$
306: using a function that falls smoothly to zero at a radius
307: $r_2\approx 10 r_1$.
308:
309: \noindent
310: 6. Step 1 was then repeated in a loop.
311:
312: \begin{figure}
313: \includegraphics[angle=0.,width=7.0cm]{fig1.ps}
314: \caption{\label{fig:f1} Evolution of the rescaled density.
315: Dashed line is the initial state and points are the
316: self-similar $\rho_*(r)$ derived in Ref. \cite{heggie:88}.
317: Inset shows $\alpha\equiv -d\log\rho/d\log r$ at the
318: end of the final integration; dashed line is
319: $\alpha=-2.23$.
320: }
321: \end{figure}
322:
323: Generation of the new particle coordinates from the single-particle
324: $f(E)$ in step 1 has the effect of removing any binaries that
325: may have formed in the previous integration interval, and of
326: resetting to zero any anisotropies that developed in the velocities.
327: Since the number of particles in the core after each rescaling
328: was $\sim 10^3$, the chance that even a single
329: binary would form before the next rescaling was negligible.
330: Velocity anisotropy was evaluated at the end of each interval
331: and found to be very small, and restricted to the largest radii.
332: The effects of binary formation and anisotropy growth could
333: be reduced still more by increasing $N$ and by reducing the
334: lengths of the integration intervals.
335: Newton-type fixed point algorithms that converge on the
336: self-similar solution are also possible.
337:
338: \section{Results}
339:
340: Fig. 1 shows the density profile at the
341: end of each rescaling step,
342: compared with the self-similar $\rho_*(r)$
343: computed from the orbit-averaged
344: isotropic Fokker-Planck equation \cite{heggie:88},
345: which has $\rho_\star \propto r^{-2.23}$ at large radii.
346: The renormalized density is quite close to the self-similar
347: solution after 4-5 iterations.
348:
349: \begin{figure}
350: \includegraphics[angle=0.,width=7.0cm]{fig2.ps}
351: \caption{\label{fig:f2}
352: Evolution of the dimensionless collapse rate parameter
353: $\xi\equiv (\dot\rho_c/\rho_c)/t_r$.
354: Each time interval corresponds to one ``burst'' of integration,
355: after which the model was rescaled as described in the text.
356: Dashed line shows the asymptotic (self-similar) value
357: $\xi=3.6\times 10^{-3}$
358: as computed via the isotropic orbit-averaged Fokker-Planck equation
359: \cite{cohn:80,heggie:88}.
360: }
361: \end{figure}
362:
363: In the self-similar regime, the finite changes
364: in central density and core radius during one integration
365: interval satisfy
366: $\log(\rho_{c,f}/\rho_{c,i})/\log(r_{c,f}/r_{c,i})=\gamma/\delta$;
367: in the Fokker-Planck approximation \cite{heggie:88},
368: this ratio is equal to the asymptotic density slope,
369: $\gamma/\delta=\alpha\approx -2.23$.
370: We computed this ratio from the rescaling factors
371: ($A,B$) in each integration interval and found it to be
372: $(-2.09,-2.15,-2.23,-2.24,-2.23,-2.23)$, consistent
373: with the Fokker-Planck prediction at late times,
374: and also consistent with our computed value of $\alpha$
375: (Fig. 1).
376:
377: The dimensionless collapse rate is
378: \begin{equation}
379: \xi\equiv t_{r}(0){1\over\rho_c}{d\rho_c\over dt}
380: \end{equation}
381: with $t_{r}(0)$ the value of the relaxation time (Eq. 1)
382: evaluated at $r=0$; in the self-similar regime,
383: $\xi$ should reach a constant value.
384: Fig. 2 shows the evolution of $\xi$, computed by fitting a
385: smoothing spline to $\rho_c(t)$ during each of the integration
386: intervals.
387: Isotropic Fokker-Planck treatments \cite{cohn:80,heggie:88} find
388: $\xi\approx 3.6\times 10^{-3}$ in the self-similar regime,
389: consistent within the noise with Fig. 2.
390:
391: \begin{figure}
392: \includegraphics[angle=0.,width=7.0cm]{fig3.ps}
393: \caption{\label{fig:f3}
394: Central density vs core radius in the rescaled runs
395: (points) and in a direct $N$-body integration (blue line).
396: Dashed line shows $\rho_c\propto r_c^{-2.23}$.
397: In the direct $N$-body simulation without rescaling,
398: the central density reaches a peak value then decreases
399: (``core bounce'') due to formation of binary stars.
400: }
401: \end{figure}
402:
403: Another way to measure the degree of evolution achieved
404: via the rescaled integrations is shown in Fig. 3.
405: In this plot, the cumulative effect of the rescalings
406: on $\rho_c$ and $r_c$ is included; the central
407: density has increased by a cumulative factor of $\sim 10^{4.5}$
408: by the end of the fifth rescaling,
409: and the core radius has decreased by a factor $\sim 10^2$.
410: The $\rho_c(r_c)$ relation from the $N$-body integration
411: without rescaling is also shown.
412: In the un-rescaled integration, the central density peaks
413: due to binary formation at a value $\sim 10$ times lower
414: than in the rescaled runs.
415:
416: In principle, one can extract the similarity
417: exponents $\gamma$ and $\delta$ of Eq. 2
418: (not just their ratio)
419: from the numerical models.
420: For instance,
421: if $t_1$ and $t_2$ are two
422: distinct times in the self-similar regime,
423: then
424: \begin{equation}
425: \gamma = {t_2-t_1\over {\rho_c(t_2)\over d\rho_c/dt|_{t_2}} -
426: {\rho_c(t_1)\over d\rho_c/dt|_{t_1}}}
427: \end{equation}
428: and similarly for $\delta$.
429: Another approach \cite{rowley:05} is via numerical
430: calculation of the scaling exponents that characterize
431: the effective differential operator.
432: A third approach is simply to fit Eqs. 1 to the time-dependent
433: density and core radius in the self-similar regime.
434: We had limited success with each of these methods due
435: to noise associated with the modest particle numbers.
436: The noise could be reduced by averaging the results from
437: different random realizations of the same initial conditions,
438: or by increasing $N$.
439:
440: \section{Conclusions}
441:
442: Coarse dynamic renormalization is a tool for
443: numerically extracting self-similar solutions by
444: evolving a dynamical system to a stationary state,
445: in a scaled reference frame where the self similarity
446: has been ``factored out.''
447: We have demonstrated the usefulness of the method
448: for studying gravitational core collapse.
449: In the limit of large particle numbers, core collapse
450: is characterized by a central density that increases
451: without limit, due to the combined effects of heat transfer
452: and the negative specific heat of the core.
453: In numerical simulations with finite $N$,
454: collapse is halted when the number of particles
455: in the shrinking core drops to a small enough value
456: that binaries can form.
457: By carrying out short bursts of integration of
458: appropriately rescaled initial conditions,
459: we showed that it is possible to
460: avoid these small-$N$ effects and to follow
461: collapse well into the self-similar regime.
462: Furthermore we achieved this with quite modest
463: particle numbers, $N\approx 10^4$.
464: In our approach, the degree to which core collapse
465: can be followed is essentially independent of
466: the number of particles used, while in direct $N$-body
467: simulations, the time to core bounce is determined
468: by $N$.
469:
470: We chose the macroscopic function of interest
471: to be the isotropic, single-particle distribution function
472: $f(E,t)$.
473: The same approximation is commonly made in
474: Fokker-Planck and fluid treatements of core
475: collapse
476: \cite{eggleton:80,cohn:80,heggie:88}.
477: A larger particle number would allow us to
478: relax the assumption of isotropy and extract
479: $f(E,L,t)$ with $L$ the orbital angular momentum
480: per unit mass.
481: Results so obtained could be compared with
482: solutions to the anisotropic
483: Fokker-Planck \cite{cohn:79,takahashi:95,joshi:00}
484: and fluid \cite{louis:91} equations.
485:
486:
487: \begin{acknowledgments}
488: This work was supported by grants AST-0206031, AST-0420920,
489: AST-0437519, and CTS-0205484 from the NSF,
490: grants NNG04GJ48G and NAG 5-10842 from NASA,
491: and grant HST-AR-09519.01-A from STScI.
492: \end{acknowledgments}
493:
494: \bibliography{ms}
495:
496: \end{document}
497: