1: \documentclass[apj]{emulateapj}
2: \journalinfo{{\rm Journal reference:
3: {\sc The Astrophysical Journal,} 635:1126--1135, 2005 December 20}}
4: \slugcomment{Received by ApJ 2004 May 28, accepted 2005 July 30,
5: published 2005 December 20}
6:
7: \usepackage{mathptmx}
8:
9: %
10: %Vectors
11: %
12: \newcommand{\vB}{{\bf B}}
13: \newcommand{\vv}{{\bf v}}
14: \newcommand{\bk}{{\bf k}}
15: \newcommand{\br}{{\bf r}}
16: \newcommand{\bx}{{\bf x}}
17:
18: %
19: % Other symbols
20: %
21: \newcommand{\del}{{\partial}}
22: \newcommand{\rar}{{\rightarrow}}
23: \newcommand{\eps}{{\epsilon}}
24: \newcommand{\mdot}{{\dot{M}}}
25: \newcommand{\qq}{{\frac{3}{2}}}
26: \newcommand{\lfrac}[2]{{{#1}/{#2}}}
27: \newcommand{\rhomean}{{\bar{\rho}}}
28: \newcommand{\EK}{{E_K}}
29: \newcommand{\seed}{{\rm Seed}}
30: \newcommand{\LAD}{{L_{\rm{AD}}}}
31: \newcommand{\tni}{{t_{ni}}}
32: \newcommand{\nion}{{n_i}}
33: %
34: \newcommand{\cs}{{c_s}}
35: \newcommand{\ts}{{t_s}}
36: \newcommand{\tg}{{t_g}}
37: \newcommand{\Ls}{{L_s}}
38: %
39: \newcommand{\nHtwo}{{n_{{\rm H}_2}}}
40: \newcommand{\vA}{{v_{\rm{A}}}}
41: \newcommand{\rhoS}{{\rho_{\rm{S}}}}
42: \newcommand{\phiS}{{\phi_{\rm{S}}}}
43: \newcommand{\MS}{{M_{\rm{S}}}}
44: \newcommand{\nJ}{{n_{\rm{J}}}}
45: \newcommand{\LJ}{{L_{\rm{J}}}}
46: \newcommand{\MJ}{{M_{\rm{J}}}}
47: \newcommand{\NT}{{N_{\rm{T}}}}
48: \newcommand{\rhoT}{{\rho_{\rm{T}}}}
49: \newcommand{\tT}{{t_{\rm{T}}}}
50: \newcommand{\tten}{{t_{10}}}
51:
52: %
53: % Units of measurement
54: %
55: \newcommand{\um}{{\,\mu\rm m}}
56: \newcommand{\cm}{{\rm\,cm}}
57: \newcommand{\km}{{\rm\,km}}
58: \newcommand{\au}{{\rm\,AU}}
59: \newcommand{\pc}{{\rm\,pc}}
60: \newcommand{\kpc}{{\rm\,kpc}}
61: \newcommand{\mpc}{{\rm\,Mpc}}
62: \newcommand{\yr}{{\rm\,yr}}
63: \newcommand{\Myr}{{\rm\,Myr}}
64: \newcommand{\gm}{{\rm\,g}}
65: \newcommand{\kms}{{\rm\,km\,s^{-1}}}
66: \newcommand{\erg}{{\rm\,erg}}
67: \newcommand{\ev}{{\rm\,eV}}
68: \newcommand{\msun}{{\rm\,M_\odot}}
69: \newcommand{\lsun}{{\rm\,L_\odot}}
70: \newcommand{\rsun}{{\rm\,R_\odot}}
71: \newcommand{\K}{{\rm\,K}}
72: \newcommand{\second}{{\rm\,s}}
73:
74: %
75: \begin{document}
76:
77: \title{Nonlinear Criterion for the Stability of Molecular Clouds}
78:
79: \author{Ruben Krasnopolsky and Charles F. Gammie}
80:
81: \affil{Center for Theoretical Astrophysics,
82: University of Illinois at Urbana-Champaign, Loomis Laboratory of Physics,\\
83: 1110 West Green Street, Urbana, IL 61801}
84:
85: \shortauthors{{\sc Krasnopolsky and Gammie}}
86: \shorttitle{{\sc Nonlinear Stability of Molecular Clouds}}
87:
88: \begin{abstract}
89:
90: Dynamically significant magnetic fields are routinely observed in
91: molecular clouds, with mass-to-flux ratio $\lambda \equiv (2\pi\sqrt{G})
92: \Sigma/B \sim 1$ (here $\Sigma$ is the total column density and $B$ is
93: the field strength). It is widely believed that ``subcritical'' clouds with
94: $\lambda < 1$ cannot collapse, based on virial arguments by
95: Mestel and Spitzer and a linear
96: stability analysis by Nakano and Nakamura. Here we confirm, using high
97: resolution numerical models that begin with a strongly supersonic
98: velocity dispersion, that this criterion is a fully nonlinear stability
99: condition. All the high-resolution models with $\lambda \le 0.95$ form
100: ``Spitzer sheets'' but
101: collapse no further. All models with $\lambda \ge 1.02$ collapse to the
102: maximum numerically resolvable density. We also investigate other
103: factors determining the collapse time for supercritical models. We show
104: that there is a strong stochastic element in the collapse time: models
105: that differ only in details of their initial conditions can have
106: collapse times that vary by as much as a factor of 3.
107: The collapse time cannot be determined from just the velocity dispersion;
108: it depends also on its distribution. Finally, we discuss the
109: astrophysical implications of our results.
110:
111: \end{abstract}
112:
113: \keywords{star formation}
114:
115: \section{Introduction}
116:
117: Molecular clouds evolve under the influence of self-gravity so as to
118: condense part of their mass into dense cores and, ultimately, stars.
119: The presence of magnetic fields can prevent or delay condensation. The
120: possibility was first studied by \cite{ms56}, who noted that the magnetic
121: energy and the gravitational energy scale in exactly the same way with
122: the radius $R$ of the cloud ($\propto 1/R$) if flux freezing obtains.
123: They argued that there was therefore a critical mass below which a cloud
124: threaded by a particular field strength would be unable to collapse.
125:
126: A more precise but less general argument was advanced by \cite{nn78},
127: who studied the linear theory of a self-gravitating, isothermal,
128: equilibrium sheet of plasma threaded by a perpendicular magnetic field.
129: They found that magnetic fields stabilize the sheet against
130: gravitational collapse if the mass-to-flux ratio is smaller than
131: $1/2\pi\sqrt{G}$.
132:
133: These results motivate the definition of a dimensionless mass-to-flux
134: ratio,
135: \begin{equation}
136: \lambda \equiv 2\pi\sqrt{G} {\Sigma\over{B}},
137: \end{equation}
138: where $\Sigma$ is the column density of the sheet, and $B$ is the
139: magnetic field strength. The exact coefficient used to define $\lambda$
140: depends somewhat on the geometry of the collapse. Here we have chosen
141: the coefficient most relevant to the magnetic field geometry adopted in
142: this paper, tending to produce thin sheets, in agreement with the
143: expectations for magnetically supported clouds. Clouds with $\lambda >
144: 1$ are termed {\it supercritical}, and clouds with $\lambda < 1$ are
145: termed {\it subcritical}.
146:
147: Both the \citeauthor{ms56} and the \citeauthor{nn78} models consider
148: exact equilibria. Molecular clouds are far from equilibrium, however,
149: with near-virial, highly supersonic velocity dispersion. These internal
150: velocities must arise from strong turbulence.\footnote{The most
151: plausible alternative to turbulence, some type of weakly dissipative
152: ordered flow, does not emerge naturally in any relevant numerical
153: experiments that we are aware of. The mode-mode coupling is always
154: strong. Even circularly polarized Alfv\'en waves, which are exact
155: solutions to the compressible equations of motion, suffer from a
156: parametric instability with a dynamical decay rate \citep{sag69,gol78}.}
157: Turbulence might change the stability properties of the cloud, either by
158: compressing a $\lambda < 1$ flow until it collapses, or by providing
159: turbulent support to a cloud with $\lambda > 1$.
160:
161: Many works have suggested that turbulence could provide support to
162: star-forming clouds. \citet{cf53a} included turbulent support in their
163: model for interstellar gaseous structures. \citet{ms56} pointed out
164: that turbulence tends to decay, and that turbulence of amplitude large
165: enough to support a cloud against self-gravity would decay especially
166: quickly, although allowing the possibility that a strong magnetic field
167: might perhaps allow longer lived turbulence. The supersonic linewidths
168: observed in molecular clouds were attributed to radial motions inside
169: the cloud instead of turbulence by \cite{gk74}. \citet{zp74} argued
170: that if this interpretation were true for all clouds where such
171: fluctuations are observed, the star formation rate would be too large by
172: at least one order of magnitude. \citet{am75} then suggested that the
173: observed velocity fluctuations are due to hydromagnetic waves. By the
174: late 1980s, this idea was widely accepted \citep[e.g.,][]{sal87}. In the
175: late 1990s, however, a succession of numerical experiments
176: \citep{ml98,sog98,go96} strongly suggested that the damping time of
177: turbulence in
178: magnetized molecular clouds is close to the dynamical time. If one
179: accepts this, then turbulent pressure can be effective in supporting
180: self-gravitating clouds only if it is constantly replenished, in which
181: case the support is perhaps more readily identified with the
182: stirring mechanism than with the turbulence itself.
183:
184: Other work has tended to emphasize the role of turbulence in initiating
185: gravitational collapse \citep[e.g.,][]{mk04}. Regions with a convergent
186: velocity field will naturally tend to collapse sooner than regions with
187: divergent velocity fields. It seems highly likely that some parts of
188: molecular clouds have strongly convergent velocity fields; is this ever
189: enough to overcome the stabilizing effects of the magnetic field?
190: Can a subcritical cloud be induced to collapse by squeezing, or can a
191: supercritical cloud be prevented from collapsing by the introduction of
192: turbulence? The purpose of this paper is to investigate these questions
193: using a simple series of numerical experiments.
194:
195: The plan of the paper is as follows. In \S 2 we describe the
196: experimental design, our numerical methods, and the diffusion
197: characteristics of our code (based on the ZEUS algorithm).
198: In \S 3 we describe results, including a ``fiducial'' run,
199: and the influence of physical and numerical parameters on the outcome.
200: \S 4 summarizes and discusses astrophysical implications.
201:
202: \section{Description of numerical experiments}
203: \label{description}
204:
205: We will consider the simplest possible system that can manifest
206: sub/supercritical behavior: a two-dimensional, periodic box containing a
207: magnetized, self-gravitating, isothermal gas. Since we are interested in
208: studying the effects of turbulence, we will introduce a velocity field
209: in the initial conditions with statistical properties similar to those
210: found in interstellar clouds. We will then allow the system to evolve
211: for many dynamical times, or until it ``collapses.''
212:
213: Specifically, we consider a square domain in the $x-y$ plane of size $L
214: \times L$. The $z$ direction points out of this plane; no quantity depends
215: on $z$. The initial fluid density is $\rhomean$, and the sound speed,
216: which is constant in space and time, is $\cs$. The initial field is
217: ${\bf B} = B_x {\hat {\bf x}}$, where $B_x$ is constant. The strength
218: of the field can be characterized by $\lambda = 2 \pi \sqrt{G}
219: \rhomean L/B_x$. We set the initial value of $\left< B_z \right>=0$,
220: because otherwise in this $z$-independent geometry,
221: asymptotically there is no collapse.
222:
223: The initial velocity field is a Gaussian random field with zero
224: divergence, constructed as in \cite{ogs99}. The initial velocity field
225: has a power spectrum $\left<v_k^2\right>\propto k^{-3}$ for
226: $2\pi/L<k<8\,(2\pi/L)$ and ${\bf k} \cdot {\bf v}_k = 0$. This power
227: spectrum is consistent, in 2D, with Larson's Law $v_\lambda \sim
228: \lambda^{1/2}$, which is equivalent to an energy spectrum $E_k \sim
229: k^{-2}$ (in 3D Larson's Law
230: implies $\left<v_k^2\right>\propto k^{-4}$).
231: The velocity is normalized so that the kinetic energy $\EK$ matches the
232: desired value, and $1/3$ of the kinetic energy is in motions
233: perpendicular to the $x - y$ plane of the simulation.
234:
235: \subsection{Spitzer sheets and simulation units}
236:
237: In our experiments we will frequently find that matter flows along
238: magnetic field lines to form sheets normal to the field. These sheets
239: are given coherence by the self-gravity of the medium. \cite{spitz42}
240: was the first to consider the problem of the vertical structure of an
241: infinite, self-gravitating, isothermal sheet. Spitzer's solution turns
242: out to be highly useful in understanding the evolution of our
243: simulations. Spitzer found that the equilibrium density profile of a
244: sheet of surface density $\Sigma$ is $\rho(z) = (\Sigma/2 H) {\rm
245: sech}^2(z/H)$ where $H = \cs^2/(\pi G \Sigma)$. The corresponding
246: gravitational potential and field are $\phiS = 2 \cs^2
247: \log(\cosh(z/H))$ and $g = -2\pi G \Sigma \tanh(z/H)$.
248:
249: For any given Spitzer sheet formed during our simulation the surface
250: density parameter $\Sigma$ corresponds to $\rhomean \Ls$, where $\Ls$ is
251: the extent of the region along the fieldlines that a given sheet has
252: collected its mass from. At late times we can assume that $\Ls\approx
253: L$: most of the mass originally distributed along the fieldlines will be
254: collected into the given sheets. This is true for the most massive
255: sheets in the supercritical simulations, and it is seen even more
256: clearly in the subcritical simulations, where at late times in the
257: simulation a single large scale, stable sheet incorporates most of the
258: mass of the system. In the following we will assume $\Ls\approx L$ and
259: $\Sigma=L\rhomean$ for the Spitzer sheets we are largely interested in
260: --- those that have collected most mass.
261:
262: We nondimensionalize our models by setting $L = 1, \rhomean = 1$, and
263: $\cs = 1$. The simulation time unit is therefore $L/\cs$, the sound
264: crossing time. In these units, Newton's gravitational constant equals
265: $\pi\nJ^2$, the Jeans length is $1/\nJ$, the peak density $\rhoS$ of
266: an equilibrium Spitzer sheet of $\Ls=L$ is $(\pi\nJ)^2/2$, and its
267: half-thickness $H$ equals $(\pi\nJ)^{-2}$. Notice that, because of the
268: periodic boundary conditions the Spitzer sheets are slightly distorted,
269: but as long as $H \ll 1$, the Spitzer solution will be approximately
270: correct.
271:
272: Most of the simulations presented in this paper have $\nJ=3$. This implies
273: that the semithickness of the Spitzer sheet is $0.011$, and if we are to
274: resolve this with at least four grid zones we need $N > 360$, where the
275: resolution of our uniform grid is $N^2$. This rather stringent
276: resolution requirement explains why we have chosen to study the problem
277: in 2D rather than 3D.
278:
279: This Spitzer-sheet model is especially useful when $\lambda\lesssim 1$
280: inside a largely ordered magnetic field, able to channel the flow into sheets,
281: which later might become unstable and collapse, through accretion, collision,
282: and merger.
283: This sheet model, however, is not useful where $\lambda\gg 1$ and
284: collapse is unconstrained by the field.
285:
286: Simulation units can be converted to dimensional values by assuming for
287: illustrative purposes a typical density $\nHtwo=10^2\cm^{-3}$,
288: and a typical temperature $T=10\K$. Then the sound speed $\cs=0.19\kms$
289: fixes the unit of speed, and from the Jeans length
290: $\LJ=\cs(\pi/G\rhomean)^{1/2}=1.9\pc$ we obtain the unit of length
291: $L=\nJ\LJ=5.7\pc$ for our standard value $\nJ=3$; a Spitzer sheet would
292: then have a peak density of $\nHtwo=4.4\times10^3\cm^{-3}$, and
293: $H=0.064\pc$. The unit of mass is given by $\rhomean L^3=\rhomean
294: \nJ^3\LJ^3 =1.3\times 10^3\msun$. The unit of time is the sound
295: crossing time $\ts=L/\cs \approx 30 \Myr$. A characteristic
296: gravitational contraction time is $\tg=\LJ/\cs\approx 10\Myr$; free-fall
297: collapse times are on the order of $0.3 \tg\approx 3\Myr$, depending on
298: the geometry of the collapse.
299:
300: The sound-crossing time of a Spitzer sheet is $\sim H/\cs\approx
301: 0.3\Myr$. From the Spitzer sheet parameters $\rhoS$ and $H$ it is
302: possible to define a characteristic ``Spitzer'' mass $\MS = (\pi H)^2
303: \Sigma = \cs^4/(G^2 \Sigma) = (\pi^2\nJ^4)^{-1}$ in dimensionless
304: units, with $\Sigma= L\rhomean$. The factor of $\pi^2$ is designed to
305: capture the mass inside half a wavelength of the shortest unstable mode
306: of the sheet \citep{led51,ee78}. For parameters typical of a molecular
307: cloud, $\MS = 1.6\msun$, which is a suggestive result. This may be
308: compared with the thermal Jeans mass $\MJ = \rhomean \LJ^3
309: =\lfrac{\pi^{3/2} \cs^3}{G^{3/2} \rhomean^{1/2}} =\nJ^{-3}$ in
310: dimensionless units, with a typical value of $\MJ= 49\left(\lfrac{T}{10
311: \K}\right)^{3/2} \left(\lfrac{\nHtwo}{10 \cm^{-3}}\right)^{-1/2}
312: \msun$.
313:
314: \subsection{Ambipolar diffusion}\label{AD}
315: Our simulation utilizes the ideal MHD equations, which have some
316: well-known limitations.
317: Ambipolar diffusion is expected to become relevant
318: \citep{kp69} at lengthscales
319: smaller than the damping length for Alfv\'en waves,
320: $\LAD\sim \vA \tni$, where $\vA=B/\sqrt{4\pi\rho}$,
321: and $\tni=1/(K \nion)$, with $K\approx 1.9\times 10^{-9}\cm^3\second^{-1}$
322: \citep{drd83}.
323: The number density of ions $\nion$ may depend strongly on environmental
324: factors, such as the UV illumination and its attenuation by
325: the cloud material \citep{cm95,mk89}; it also depends on chemical properties,
326: such as the metal abundance in the gas phase.
327: For our fiducial mean density $\rhomean$, a representative
328: value could be $\nion \sim 2\times 10^{-4}\cm^{-3}$, largely limited
329: by an assumed metal abundance $x_M\approx 10^{-6}$.
330: For the typical peak density of a Spitzer sheet $\rhoS$ in our
331: conditions, $\nion \sim 6\times 10^{-4}\cm^{-3}$
332: for cosmic-ray dominated ionization, and about ten times larger for regions
333: of the cloud that are moderately well UV-illuminated.
334:
335: The ambipolar diffusion lengthscale $\LAD$ can now be compared with
336: the size $L$ of the computational volume, giving
337: an estimate of the scale at which ideal MHD stops
338: being a complete dynamical description of the flow.
339: For our mean density profile, we find that this lengthscale is
340: $\LAD\sim L/35$, much larger than our typical grid spacing $L/512$,
341: and comparable to our typical sheet thickness $2H=L/45$.
342: However, ideal MHD is still a good description of the most important portions
343: of this study; the regions where mass is collected to form dense sheets.
344: For $\rho=\rhoS$, the lengthscales are $\LAD\sim L/700$ for a UV-dark region,
345: and $\LAD\sim L/7000$ for the more illuminated case. The decrease of
346: $\vA$ with density has contributed to this effect, together with the
347: larger $\nion$.
348:
349: We keep in mind, however, that densities and ionization rates vary
350: widely inside clouds, and so the ambipolar diffusion lengthscales
351: and timescales may vary widely.
352: Turbulent conditions inside the flow are also expected to increase the
353: importance of ambipolar diffusion, even at larger lengthscales, especially
354: near sharp velocity and magnetic gradients.
355:
356: \subsection{Stopping criterion}
357:
358: We must fix some criterion for stopping the numerical integration; when
359: the density in any zone equals or exceeds the largest allowed by the
360: Truelove numerical stability condition \citep{truelove97} the run is
361: terminated and classified as having collapsed. The Truelove condition
362: requires that the local Jeans length be resolved by some
363: algorithm-dependent number $\NT$ of grid
364: zones, typically about $4$. This requirement sets a maximum resolvable
365: density of $\rhoT=(N/\nJ\NT)^2 = 1820 (N/512)^2$ for $\nJ=3$
366: on our uniform
367: grid of $N^2$ zones. We have found that further integration of
368: Truelove-unstable models results in large local fluctuations in the
369: density, which can produce ``explosions'' that corrupt the entire
370: computational domain. Runs that reach $t = 2$ without violating the
371: Truelove condition are classified as stable.
372:
373: We have experimented with other collapse detection schemes, because the
374: Truelove criterion has the deficiency that it is resolution dependent.
375: In the Tables described below we report not only the time $\tT$ at which
376: the Truelove condition is violated, but also the time $\tten$ when
377: $1\%$ of the mass exceeds $10$ times the Spitzer density $\rhoS$.
378: These times are typically close to each other, and both can be
379: considered as measures of the onset of gravitational instability. The
380: time $\tten$ has the advantage of not depending explicitly on numerical
381: resolution, but it can be fooled into producing misleadingly short
382: collapse times by strong density fluctuations, particularly when the
383: turbulent kinetic energy is large.
384:
385: \subsection{Numerical methods and tests}
386:
387: Our simulations are run on a fixed 2D Cartesian grid, using the ZEUS
388: algorithm \citep{sn92a,sn92b} as implemented for instance in
389: \citet{ogs99}. ZEUS is a numerical algorithm to evolve ideal
390: (non-resistive, non-viscous) non-relativistic MHD flows. It is
391: operator-split, representing the fields on a (possibly moving) Eulerian
392: staggered mesh. The magnetic field evolution uses constrained transport
393: \citep{eh} which guarantees that $\nabla\cdot\vB=0$ to machine
394: precision, combined with the method of characteristics \citep{hs}, which
395: ensures accurate propagation of Alfv\'en waves. ZEUS is explicit in
396: time, and so the timestep $\Delta t$ is limited by the Courant
397: conditions. In our problem, usually the most stringent has been
398: $\Delta t < \Delta x/\vA$, where $\vA=B/\sqrt{4\pi\rho}$
399: can take very large values in density-depleted regions.
400: A numerical density floor, $\rho_{\rm floor}$, has been set to limit
401: density depletion, preventing $\Delta t$ from becoming too small;
402: we have directly tested that this tiny non-conservation of mass
403: by the code does not alter the simulation results regarding collapse
404: in any way.
405: The Poisson equation, needed to describe self-gravity, is solved by
406: Fourier transform methods, using the FFTW code \citep{fftw}.
407:
408: Any Eulerian scheme will cause some diffusion of the magnetic field with
409: respect to the mass. It is crucial for our experiment that this
410: nonconservation of $\lambda$ be as small as possible. The numerical
411: diffusivity of ZEUS is difficult to estimate because, unlike a physical
412: resistivity, it is flow dependent. An empirical approach is therefore
413: required.
414:
415: We have studied conservation of $\lambda$ using two distinct methods.
416: In the first method we initialize a non--self-gravitating box using the
417: same initial data as in our main experiments, as described above. We
418: evolve the computation to $t = 0.5$ and then damp the velocity field
419: exponentially (with timescale $t_{\rm damp} = 0.05$) until $t = 10$. If
420: there were no $\lambda$ diffusion the box would return to a uniform
421: density, uniform field state. Diffusion changes $\lambda$, so the final
422: state consists of a unidirectional magnetic field with density and field
423: strength varying only perpendicular to the field, from which $\lambda$
424: can be easily measured.
425:
426: In the second method we initialize a non--self-gravitating box using the
427: same initial data as in our main experiments, but we evolve the
428: computation only to $t = 0.5$. We then sample 80 field lines chosen to
429: lie at equal intervals of the vertical component of the vector potential
430: (equivalent to lines equally spaced in magnetic flux). We then
431: integrate $\rho/\left(B_x^2 + B_y^2\right)^{1/2}$
432: along the field line in the $x - y$ plane,\footnote{This is equivalent
433: to integrating $\rho/\left(B_x^2 + B_y^2 + B_z^2\right)^{1/2}$ along
434: the 3D fieldline.}
435: using linear interpolation to determine $\rho$ and ${\bf B}$ at each
436: position, which immediately yields $\lambda$.
437:
438: These two methods give nearly identical results. We therefore adopt the
439: second method exclusively, since it can be used to probe existing
440: numerical data without any additional, expensive evolution.
441:
442: One possible figure of merit for the diffusion in $\lambda$ is
443: $\sigma_\lambda/\lambda_0$, where $\sigma_\lambda$ is the dispersion in
444: sampled values of $\lambda$ at the final instant of the simulation,
445: and $\lambda_0$ is the nominal initial value (which we call simply
446: $\lambda$ outside of this subsection).
447: Tables \ref{table:lambdavsN} and \ref{table:lambdavst}
448: show $\sigma_\lambda/\lambda_0$ as a function of
449: resolution and of time during a single simulation, respectively. The
450: run shown in Table~\ref{table:lambdavst}
451: has a resolution of $512^2$. The key points here
452: are that $\sigma_\lambda/\lambda_0$ decreases as resolution increases,
453: and that in every case $\sigma_\lambda/\lambda_0$ is about $10\%$ or
454: less, which suggests that we should be able to measure the critical
455: value of $\lambda$ to similar accuracy.
456:
457: Evidently $\sigma_\lambda/\lambda_0$ is converging, but as $\approx
458: N^{-1/2}$ rather than the expected $N^{-1}$. This may be because of the
459: existence of unresolved regions in the flow where most of the diffusion
460: occurs, or it may be the result of irreducible ``turbulent'' diffusion
461: that is present independent of the magnitude of the effective numerical
462: diffusion.
463: The numerical diffusion is correlated with the amplitude of turbulence.
464: According to Table~\ref{table:lambdavst}, much of the diffusion occurs
465: very early in the run, when the rms velocity is large.
466:
467: \begin{deluxetable}{rllll}[b]
468: \tablecaption{
469: Mass-to-flux diffusion in ZEUS,
470: as a function of numerical resolution
471: \label{table:lambdavsN}}
472: \tablehead{
473: \colhead{$N$} &
474: \colhead{$t$} &
475: \colhead{$\sigma_\lambda/\lambda_0$} &
476: \colhead{$\lambda_{\rm max}/\lambda_0$} &
477: \colhead{$\lambda_{\rm min}/\lambda_0$} }
478: \startdata
479: 400 & 0.1 & 9.0 \% & 1.26 & 0.78 \\ %
480: 512 & 0.1 & 8.6 \% & 1.31 & 0.78 \\ %
481: 1024 & 0.1 & 6.1 \% & 1.30 & 0.87 \\ %
482: 2048 & 0.1 & 4.4 \% & 1.10 & 0.86 \\ %
483: \tableline
484: 512 & 0.2 & 8.3 \% & 1.20 & 0.79 \\ %
485: 2048 & 0.2 & 4.7 \% & 1.12 & 0.85 %
486: \enddata
487: \end{deluxetable}
488:
489: \begin{deluxetable}{lrllr}
490: \tablecaption{
491: Mass-to-flux diffusion in ZEUS, as a function of time
492: \label{table:lambdavst}}
493: \tablehead{
494: \colhead{$t$} &
495: \colhead{$\sigma_\lambda/\lambda_0$} &
496: \colhead{$\lambda_{\rm max}/\lambda_0$} &
497: \colhead{$\lambda_{\rm min}/\lambda_0$} &
498: \colhead{$\EK(t)$}}
499: \startdata
500: 0 & 0\phantom{.0}\%
501: & 1 & 1 & 50 \\ %
502: 0.01 & 0.2\% & 1.01 & 0.99 & 37 \\ %
503: 0.02 & 0.9\% & 1.02 & 0.97 & 25 \\ %
504: 0.03 & 2.1\% & 1.05 & 0.92 & 18 \\ %
505: 0.04 & 3.6\% & 1.10 & 0.90 & 15 \\ %
506: 0.05 & 6.4\% & 1.18 & 0.84 & 16 \\ %
507: 0.06 & 8.0\% & 1.32 & 0.74 & 19 \\ %
508: 0.1 & 8.6\% & 1.31 & 0.78 & 17 \\ %
509: 0.2 & 8.3\% & 1.20 & 0.79 & 11 \\ %
510: 0.3 & 9.6\% & 1.20 & 0.74 & 8.5 \\ %
511: 0.34 & 11.1\% & 1.21 & 0.71 & 5.5 %
512: \enddata
513: \end{deluxetable}
514:
515: Another possible figure of merit is the total variation in $\lambda$,
516: that is, $|\lambda_{\rm max} - \lambda_{\rm min}|/(2 \lambda_0)$,
517: measuring the possible existence of localized diffusion events in
518: addition to the overall diffusivity of the code.
519: We find in Tables~\ref{table:lambdavsN} and \ref{table:lambdavst}
520: that this quantity is typically of the size $\sim 2.4 \sigma_\lambda$,
521: expected in the mean for the half-range of a sample of 80 elements randomly
522: taken from a Gaussian distribution.
523: However, the distribution of values of $\lambda$ might not always be Gaussian,
524: because localized numerical diffusion could be important for some fieldlines.
525: In our tables, this may be happening when
526: $\lambda_{\rm max}$ reaches values as large as $1.3\lambda_0$,
527: even at the relatively high resolution of $N=1024$.
528: The larger total variation of $\lambda$ observed in those
529: simulations suggests a possible risk of masking the
530: subcritical nature of some models.\footnote{As a possible example of
531: this effect, in Table~\ref{table:resolution}
532: one model collapses despite having $\lambda_0=0.9$;
533: also the very low resolution
534: models with $N=128$ that collapse for $\lambda_0=0.744$.}
535: However, we also find that the total variation of $\lambda$
536: starts to drop at the even higher resolution of $N=2048$;
537: numerical resolution seems apparently able to reduce also this
538: more local measure of the diffusivity of the numerical code.
539:
540: It is worth noting that ambipolar diffusion in nature is likely strong
541: enough to dominate the diffusion measured here. For our nominal cloud
542: parameters, we have seen in \S 2.2 that the damping length of Alfv\'en
543: waves is of the same order as the expected thickness of an equilibrium sheet.
544: Thus our models may misrepresent the situation in nature by tying the
545: fluid too closely to the magnetic field. The combination of
546: ambipolar diffusion and turbulence (which can drive sharp features for
547: the ambipolar diffusion to act on) may be a potent driver of variations
548: in mass-to-flux ratio in Galactic molecular clouds.
549:
550: \section{Results}
551: \label{results}
552:
553:
554: \subsection{Fiducial run} % Name of this run: sc132 (also sc136)
555:
556: As a guide to the dynamics of our numerical experiments, we will first
557: describe a ``fiducial'' run, whose behavior is in a sense typical of the
558: other experiments. This run is supercritical, with $\lambda=1.5$,
559: initial $\EK=50$ (equivalent to an rms Mach number of $10$), $\nJ=3$,
560: and $N=512$. The panels of Figures \ref{fig:fiducial}
561: and \ref{fig:fiducial_lines} show how condensation
562: proceeds. At first, small density concentrations form due to both ram
563: pressure associated with the supersonic velocity fluctuations in the
564: initial conditions and fluctuations in the magnetic pressure. These
565: later coalesce into larger clumps, typically oriented perpendicular to
566: the magnetic field.\footnote{Clumps do not tend to orient
567: perpendicular to the field in our three dimensional models \citep{glso03}.
568: Those runs had a resolution of $256^3$, however, and the
569: supercritical runs did not have $\lambda$ as close to $1$ as the models
570: considered here.} We have seen that these clumps develop into fully
571: stable Spitzer sheets in the simulations of subcritical clouds; here
572: they can be considered also as approximate Spitzer sheets, which later
573: come unstable, as the peak density of these sheets grows. This density
574: growth takes place when the clumps merge or collide, and when matter
575: accretes from outside the sheet.
576: The largest density of these clumps increases as shown in
577: Fig.~\ref{fig:fiducial_max_density}; slowly at the beginning,
578: but very steeply close to the end of the run.
579: The energies, on the other hand, vary smoothly in time
580: (Fig.~\ref{fig:fiducial_energies}), and are not a good predictor
581: of the time required for instability.
582:
583: At a time $\tten=0.325$, the fraction of matter denser than 10 times
584: the nominal Spitzer density includes more than 1\% of the mass
585: $\rho_{1\%}(t)>10\rhoS$. Not long afterwards, at a time $\tT=0.341$
586: (soon after the last panel in Fig.~\ref{fig:fiducial}) the peak density
587: of the simulation box exceeds the Truelove limit, forcing an end to the
588: run. We conclude that the initial state of this fiducial run represents
589: an unstable cloud, able to produce dense cores. These two times are
590: much larger than the linear e-folding time found in \citet{nakano88}
591: ($\approx 0.035$ for $\lambda=1.5$); collecting matter into the unstable
592: structures takes a longer time than the instability process, linear or
593: nonlinear, and dominates the total time necessary to achieve
594: instability in the mildly supercritical clouds. Using our nominal
595: conversion factors from simulation to physical units, $\tten = 9.8\Myr$.
596:
597: \subsection{The nonlinear stability criterion}
598: \label{stabilitycriterion}
599:
600: To discover how precisely the criticality condition was obeyed in the
601: numerical experiments, we considered a series of runs with $\EK = 50$
602: and $\EK = 10$ while gradually varying $\lambda$.
603: Table~\ref{table:stability} lists the
604: collapse times for each of these runs. Evidently the criticality
605: condition is very nearly obeyed in the numerical evolutions, and there
606: is no evidence of collapse induced by compression. Indeed, given the
607: diffusion of $\lambda$ measured in \S 2,
608: it is remarkable (from a
609: numerical standpoint) that we are able to reproduce the condition so
610: accurately. Some sense of the ``error bars'' can be obtained by
611: noticing that the $\lambda = 1$ model with $\EK = 50$ does collapse,
612: while the $\lambda = 1.02$ model with $\EK = 10$ does not. This
613: suggests that the Nakano \& Nakamura condition is the true nonlinear
614: stability condition.
615:
616: Tables~\ref{table:turbulence} and \ref{table:resolution} show a clear
617: trend to make the simulations shorter lived as $\lambda$ increases.
618: This trend is expected from the already observed
619: stability criterion.\footnote{The most strongly supercritical
620: models ($\lambda\gtrsim 10$) collapse very quickly, in around one
621: free-fall time.} Models with $\lambda \approx 1$ can be quite long
622: lived; the $\lambda = 1$, $\EK = 50$ model persists until $t = 0.756$,
623: or about $23\Myr$ for our nominal cloud parameters.
624:
625: The run with $\lambda=1.05$ and $\EK=10$ has been done twice, with
626: different values of the numerical density floor; the collapse times
627: are identical up to reasonable precision. This allows us to trust the
628: runs using the larger density floor, which are much more convenient
629: because it allows a larger timestep, largely
630: controlled by the maximum value of the Alfv\'en speed
631: $B/\sqrt{4\pi\rho}$ on the grid. From here on, we will not report the
632: values of this purely numerical parameter in these simulations.
633:
634: \subsection{Influence of the turbulence energy and distribution}
635:
636: We have also investigated the effect of the amplitude and structure of
637: the initial velocity field. Table~\ref{table:turbulence}
638: shows the results from a series of runs with
639: $\nJ = 3$ and $N = 512$.
640: The column marked ``Seed'' is the seed used to
641: initiate the random number generator used to generate the initial
642: velocity field. Runs with the same seed but
643: different initial kinetic energies have velocity fields that are
644: linearly proportional to each other.
645:
646: The first series of runs with $\rm{seed}=1$ show a monotonic increase in the
647: lifetime of the cloud with kinetic energy, consistent with results
648: reported elsewhere \citep{go96, ogs99}.
649: The effect is weak at low energies but more pronounced once $\EK > 50$.
650:
651: An even larger effect is obtained by changing the structure of the
652: initial velocity field, i.e.\ by changing the seed. We find that models
653: that differ only in the initial seed can have collapse times that vary
654: by up to a factor of $3$.
655:
656: To further explore this effect, we performed a series of runs, choosing
657: sixty different values of the random seed used to set up the shape of
658: the initial velocity distribution. In this series, we have fixed
659: $\lambda=1.5$, $\EK=50$, $N=512$, and $\nJ$=3. The results
660: can be seen in Figure~\ref{fig:distribution}, showing a wide distribution
661: of collapse times. The total range of this sample goes from $\tT=0.228$ to
662: $\tT=0.667$, equal to $\sim 7$ to $20\Myr$ for the typical cloud
663: parameters used in \S 2. The mean time is $\langle \tT
664: \rangle=0.368$ ($\sim 11\Myr$); the median is located at $\tT=0.355$,
665: and the peak near $\tT=0.3$, showing a moderate asymmetry. This
666: asymmetry is more pronounced in the tails: $\tT<0.2$ is not observed in
667: the sample; while a few values of $\tT>0.5$ (at a similar distance from
668: the median but in the opposite direction) are present in the
669: distribution, corresponding to clouds lasting between $\sim 15$ and
670: $20\Myr$ before collapse, much longer than the mean lifetime value.
671:
672: This stochastic variation in cloud lifetime doubtless has a counterpart
673: in nature. The origin of this variability is clear: almost all velocity
674: variations occur at the largest scales, and are driven by just a few
675: Fourier modes. If these modes happen to have the right amplitude and
676: phase then collapse is hastened. If they are unfavorable, then collapse
677: can be delayed by as much as $12\Myr$ for our nominal cloud parameters.
678:
679: \begin{deluxetable}{lllll}
680: \tablecaption{
681: Models with $\lambda$ close to 1
682: \label{table:stability}}
683: \tablehead{
684: \colhead{$\lambda$} & \colhead{$\EK$} &
685: \colhead{$\rho_{\rm floor}$} &
686: \colhead{$\tten$} & \colhead{$\tT$} } % name
687: \startdata
688: 1.1 & 50 & $10^{-6}$ & 0.304 & 0.376 \\ % sc145
689: 1.05 & 50 & $10^{-4}$ & 0.387 & 0.500 \\ % sc211
690: 1.0 & 50 & $10^{-4}$ & 0.497 & 0.756 \\ % sc212
691: 0.95 & 50 & $10^{-4}$ & $>2$ & $>2$ \\ % sc215
692: 0.9 & 50 & $10^{-4}$ & $>2$ & $>2$ \\ % sc214
693: \tableline
694: 1.1 & 10 & $10^{-6}$ & 0.314 & 0.367 \\ % sc111
695: 1.05 & 10 & $10^{-6}$ & 0.702 & 0.741 \\ % sc146
696: 1.05 & 10 & $10^{-4}$ & 0.702 & 0.741 \\ % sc210
697: 1.02 & 10 & $10^{-4}$ & $>2$ & $>2$ \\ % sc213
698: 1.0 & 10 & $10^{-4}$ & $>2$ & $>2$ \\ % sc149
699: 1.0 & 10 & $10^{-6}$ & $>2$ & $>2$ \\ % sc112
700: 0.9 & 10 & $10^{-6}$ & $>2$ & $>2$ % sc113
701: \enddata
702: \tablecomments{
703: Parameters kept fixed in these runs: $\nJ=3$, $N=512$,
704: random $\rm{seed} = 2$.}
705: \end{deluxetable}
706:
707: \begin{deluxetable}{lrcll}
708: \tablecaption{
709: Supercritical and subcritical models
710: \label{table:turbulence}}
711: \tablehead{
712: \colhead{$\lambda$} & \colhead{$\EK$} &
713: \colhead{$\seed$} &
714: \colhead{$\tten$} & \colhead{$\tT$} } % name
715: \startdata
716: 1.5 & 100 & 1 & 0.601 & 0.615 \\ % sc129
717: 1.5 & 70 & 1 & 0.521 & 0.533 \\ % sc128
718: 1.5 & 50 & 1 & 0.325 & 0.341 \\ % sc132, sc136
719: 1.5 & 20 & 1 & 0.260 & 0.270 \\ % sc237
720: 1.5 & 10 & 1 & 0.250 & 0.257 \\ % sc109
721: 1.5 & 1 & 1 & 0.259 & 0.269 \\ % sc123
722: \cline{2-4}
723: 1.5 & 50 & 2 & 0.126 & 0.211 \\ % sc231
724: 1.5 & 20 & 2 & 0.193 & 0.200 \\ % sc240
725: 1.5 & 10 & 2 & 0.233 & 0.244 \\ % sc110
726: \cline{2-4}
727: 1.5 & 50 & 3 & 0.510 & 0.537 \\ % sc130
728: 1.5 & 20 & 3 & 0.346 & 0.360 \\ % sc238
729: 1.5 & 10 & 3 & 0.299 & 0.310 \\ % sc124
730: 1.5 & 1 & 3 & 0.314 & 0.324 \\ % sc117
731: \tableline
732: 1.2 & 100 & 1 & 0.715 & 0.744 \\ % sc242
733: 1.2 & 50 & 1 & 0.376 & 0.400 \\ % sc102
734: \tableline
735: 1.1 & 100 & 1 & 0.787 & 0.910 \\ % sc232
736: 1.1 & 50 & 1 & 0.377 & 0.534 \\ % sc104
737: \cline{2-4}
738: 1.1 & 100 & 2 & 0.314 & 0.372 \\ % sc243
739: 1.1 & 50 & 2 & 0.304 & 0.376 \\ % sc145
740: 1.1 & 10 & 2 & 0.314 & 0.367 \\ % sc111
741: \cline{2-4}
742: 1.1 & 50 & 3 & 0.735 & 0.746 \\ % sc148
743: \tableline
744: 0.8 & 100 & 2 & 0.216 & $>2$ \\ % sc235
745: 0.8 & 10 & 2 & $>2$ & $>2$ \\ % sc114
746: \cline{2-4}
747: 0.8 & 10 & 3 & 0.664 & $>2$ \\ % sc115
748: 0.8 & 1 & 3 & $>2$ & $>2$ % sc116
749: \enddata
750: \tablecomments{
751: Parameters kept fixed in these runs: $\nJ=3$, $N=512$.
752: }
753: \end{deluxetable}
754:
755: \begin{deluxetable}{lrcrll}
756: \tablecaption{Supercritical and subcritical models
757: run at different numerical resolutions
758: \label{table:resolution}}
759: \tablehead{
760: \colhead{$\lambda$} & \colhead{$\EK$} &
761: \colhead{$\seed$} & \colhead{$N$} &
762: \colhead{$\tten$} & \colhead{$\tT$} } % name
763: \startdata
764: $\infty$ & 50 & 1 & 512 & 0.068 & 0.077 \\ % sc138, sc137
765: $\infty$ & 50 & 1 & 1024 & 0.071 & 0.083 \\ % sc139
766: \tableline
767: 1000 & 50 & 1 & 512 & 0.068 & 0.077 \\ % sc140
768: \tableline
769: 10 & 50 & 1 & 512 & 0.112 & 0.124 \\ % sc141
770: \tableline
771: 1.5 & 50 & 1 & 256 & 0.283 & 0.283 \\ % sc96, sc135
772: 1.5 & 50 & 1 & 400 & 0.305 & 0.312 \\ % sc127
773: 1.5 & 50 & 1 & 512 & 0.325 & 0.341 \\ % sc132, sc136
774: 1.5 & 50 & 1 & 1024 & 0.409 & 0.434 \\ % sc120
775: 1.5 & 50 & 1 & 1536 & 0.429 & 0.453 \\ % sc133
776: 1.5 & 50 & 1 & 2048 & 0.445 & 0.481 \\ % sc134
777: \cline{2-5}
778: 1.5 & 20 & 1 & 512 & 0.260 & 0.270 \\ % sc237
779: 1.5 & 20 & 1 & 1024 & 0.263 & 0.283 \\ % sc121
780: \cline{2-5}
781: 1.5 & 10 & 1 & 512 & 0.250 & 0.257 \\ % sc109
782: 1.5 & 10 & 1 & 1024 & 0.253 & 0.273 \\ % sc122
783: \tableline
784: 1.2 & 50 & 1 & 256 & 0.319 & 0.319 \\ % sc101
785: 1.2 & 50 & 1 & 512 & 0.376 & 0.400 \\ % sc102
786: 1.2 & 50 & 1 & 1024 & 0.504 & 0.523 \\ % sc118
787: \tableline
788: 1.1 & 100 & 2 & 512 & 0.314 & 0.372 \\ % sc243
789: 1.1 & 100 & 2 & 1024 & 0.337 & 0.738 \\ % sc247
790: \tableline
791: 0.9 & 100 & 2 & 512 & 0.079 & 0.351 \\ % sc234
792: 0.9 & 100 & 2 & 576 & 0.079 & $>2$ \\ % sc252
793: \cline{2-5}
794: 0.9 & 100 & 1 & 512 & $>2$ & $>2$ \\ % sc249
795: \tableline
796: 0.85 & 100 & 2 & 512 & 0.075 & $>2$ \\ % sc236
797: \tableline
798: 0.8 & 100 & 2 & 512 & 0.216 & $>2$ \\ % sc235
799: \cline{2-5}
800: 0.8 & 50 & 1 & 256 & $>2$ & $>2$ \\ % sc97
801: \tableline
802: 0.7 & 100 & 2 & 512 & 0.435 & $>2$ \\ % sc241
803: \tableline
804: 0.6 & 50 & 1 & 256 & $>2$ & $>2$ % sc48
805: \enddata
806: \tablecomments{
807: Parameter kept fixed in these runs: $\nJ=3$.
808: }
809: \end{deluxetable}
810:
811: \begin{deluxetable}{lrcrll}
812: \tablecaption{Models with different values of $\nJ$.
813: \label{table:nj}}
814: \tablehead{
815: \colhead{$\nJ$} & \colhead{$\lambda$} &
816: \colhead{$\seed$} & \colhead{$N$} & \colhead{$\tT$} } % name
817: \startdata
818: 2.5 & 1.2 & 1 & 512 & 0.87 \\ % sc261
819: & 1.2 & 2 & 256 & 0.44 \\ % sc273
820: & 1.2 & 3 & 256 & 0.83 \\ % sc271
821: \cline{2-5}
822: & 1.1 & 1 & 256 & 0.75 \\ % sc262
823: & 1.1 & 2 & 256 & 0.48 \\ % sc266
824: & 1.1 & 3 & 256 & 0.82 \\ % sc265
825: & 1.1 & 3 & 512 & 1.21 \\ % sc257
826: & 1.1 & 4 & 256 & 0.49 \\ % sc269
827: \cline{2-5}
828: & 0.9 & 3 & 512 & $>2$ \\ % sc258
829: \tableline
830: 2.0 & 1.2 & 1 & 256 & 1.77 \\ % sc263
831: & 1.2 & 1 & 512 & 1.87 \\ % sc260
832: & 1.2 & 2 & 256 & 1.00 \\ % sc272
833: & 1.2 & 3 & 256 & 1.50 \\ % sc270
834: \cline{2-5}
835: & 1.15& 2 & 256 & 1.33 \\ % sc275
836: & 1.15& 3 & 256 & $>2$ \\ % sc276
837: \cline{2-5}
838: & 1.1 & 1 & 256 & $>2$ \\ % sc267
839: & 1.1 & 1 & 512 & $>2$ \\ % sc259
840: & 1.1 & 2 & 256 & 9.77 \\ % sc264, sc274
841: & 1.1 & 2 & 512 & $>2$ \\ % sc256
842: & 1.1 & 3 & 256 & $>2$ \\ % sc255
843: & 1.1 & 3 & 512 & $>2$ \\ % sc253
844: & 1.1 & 4 & 256 & $>2$ \\ % sc268
845: \cline{2-5}
846: & 0.9 & 3 & 512 & $>2$ % sc254
847: \enddata
848: \tablecomments{
849: Parameter kept fixed in these runs: $\EK=50$.
850: }
851: \end{deluxetable}
852:
853: \subsection{Influence of numerical parameters}
854:
855: The influence of the density floor $\rho_{\rm floor}$ has already been
856: shown in \S 3.2 to be fully negligible, provided
857: this floor is not unreasonably large.
858:
859: Numerical resolution, on the other hand, can be quite relevant. Any
860: serious simulation of condensation in a nearly critical cloud
861: must be able to resolve the
862: half-thickness $H$ of a Spitzer sheet, requiring at the very minimum
863: $N>1/H =(\pi\nJ)^2 $. For our fiducial choice of $\nJ$=3, this requires a
864: minimum of $N > 89$, and more reasonably $N>200$; any simulation run at
865: smaller resolution would not be exploring the most basic physics of mass
866: condensation. However, this requirement does not seem to
867: take care of all the effects of numerical resolution.
868:
869: Table~\ref{table:resolution} lists simulations where we have
870: varied the number $N$ of active zones in the grid on each direction.
871: There is a clear tendency for the more resolved simulations to delay
872: $\tT$. This was in part expected, as the Truelove limit density
873: $\rhoT=(N/\nJ\NT)^2$ depends steeply on $N$. This is not, however,
874: the main reason for the observed trend. Peak densities grow very quickly
875: in the neighborhood of the condensation time, almost nullifying in most
876: cases the influence of the exact magnitude of $\rhoT$ on the value of
877: $\tT$. The quantity $\tten$ is expected to be less directly
878: dependent on resolution, having a definition where $N$ does not appear;
879: it still shows some dependence on resolution, closely correlated to the
880: dependence shown by $\tT$, and probably due to details of the dynamics
881: being more revealed at higher resolutions and to reduced numerical
882: diffusion.
883:
884: Most of our simulations were performed at $N=512$.
885: For comparison purposes, a simulation run with the same initial
886: conditions as the fiducial run, but with $N=2048$, has $\tT=0.445$ and
887: $\tten=0.481$, noticeably larger, but not enough to change the
888: qualitative conclusions.
889:
890: One of the subcritical models with $\lambda=0.9$ and $N=512$
891: (Table~\ref{table:resolution})
892: is remarkable because the run presents an unexpected collapse
893: at $\tT=0.351$.
894: However, the same table shows that either changing the numerical seed,
895: or a moderate increase in numerical resolution is enough to suppress
896: this unusual behavior; also a small change in $\lambda$
897: can suppress this apparently purely artificial collapse.
898:
899: We have also run a few comparison simulations at $N=128$. This resolution
900: is insufficient to represent equilibrium Spitzer sheets,
901: and indeed the models collapse even for $\lambda=0.744$. This is due
902: to the excessively low resolution: it is just enough to accommodate one sheet
903: semithickness per grid zone, and, through the
904: Truelove stability criterion, it allows only a narrow density range,
905: limited by $\rhoT=(N/\nJ\NT)^2=114\rhomean=2.6\rhoS$,
906: insufficient to accommodate
907: an eventual moderately large oscillation in the density $\rhoS$
908: of the equilibrium sheets formed in subcritical simulations.
909:
910: In another series of tests, we started the simulations from an
911: equilibrium sheet, and let it evolve in the presence
912: of a very small perturbation. For the values $\lambda=1.5$
913: and $\lambda=2$, the perturbation grows
914: linearly with $e$-folding times equal to $0.036\pm 0.001$ and
915: $0.027\pm 0.001$, quite comparable to the values predicted
916: by the linear theory, $0.035$ and $0.025$, known from
917: \citet{nakano88} to a precision of $5$\%.
918: For values of $\lambda<1$, the simulations have remained stable
919: up to a time $t\approx 4$ during a few linear test runs
920: performed at $\lambda=0.9$ at various resolutions.
921:
922: \subsection{Influence of $\nJ$}
923: We have performed a small set of simulations exploring the
924: influence of $\nJ$ on the instability criterion found (Table~\ref{table:nj}).
925: For $\nJ=3$, we had found instability for $\lambda>1.05$;
926: for $\nJ=2.5$, the numerical requirement is no more stringent than
927: $\lambda>1.1$;
928: however, for $\nJ=2.0$, the numerical requirement for instability
929: becomes $\lambda>1.15$.
930: A weak dependence of the numerical criterion on $\nJ$ had already
931: been predicted by \citet{nakano88} in the linear regime. Linearly
932: unstable modes have a minimum critical wavelength; if we require
933: that this wavelength must fit inside the computational box size $L$,
934: we find that the instability criterion will be approximately\footnote{
935: Following \citet{nakano88}, certain
936: integrals involving generalized Riemann zeta functions have been
937: replaced by simpler expressions.
938: These approximations are excellent
939: inside our range of interest $\nJ\geq 2$.}
940: \begin{equation}\label{eq:nj}
941: \lambda> \left[1 - 2/(\pi\nJ^2)\right]^{-1}\ ,
942: \end{equation}
943: which for small values of $\nJ$ can be more stringent than the
944: infinite disk value $\lambda>1$.
945: The results in Table~\ref{table:nj} are consistent with Eq.~\ref{eq:nj}.
946: \citet{nakano88} presents for the case of a finite disk
947: a still more stringent criterion for linear instability,
948: $\lambda> \left[1 - 4/(\pi\nJ^2)\right]^{-1}$,
949: based on the assumption that {\em two}
950: critical wavelengths should fit inside the computational box.
951: Our slightly larger unstable range might be related to the geometric
952: difference between a finite disk and our periodic boundary conditions.
953: Spitzer sheets pull in magnetic field lines
954: during their formation and contraction (Fig.~\ref{fig:fiducial_lines});
955: this may also allow collapse at smaller wavelengths than expected in a
956: purely linear theory.
957:
958: \section{Conclusions}
959:
960: Our simulations confirm that the single most important element in
961: determining the long term gravitational stability of turbulent
962: magnetized clouds is indeed the mass-to-flux ratio, dividing
963: supercritical from subcritical clouds.
964: The relevant coefficient is that corresponding to a sheet geometry,
965: as derived by \citet{nn78}.
966:
967: Turbulent energy has comparatively little influence on the presence or
968: absence of stability, up to Mach numbers $\sim 10$.
969: Subcritical clouds will develop density concentrations due to
970: this turbulence, but under an ideal MHD regime, the consequent
971: increase in magnetic pressure prevents further collapse.
972: However, total turbulent energy has some influence on the lifetime of
973: supercritical clouds, especially as the Mach number becomes large
974: enough (of the order of $\sim 7$ in these simulations).
975:
976: More interesting is the fact that turbulence introduces a stochastic
977: element. The collapse time cannot be predicted with certainty from
978: physical parameters such as the mass and field in the cloud, and the
979: typical energy of the turbulence motions, because the random
980: distributions of velocity and density can change the lifetime by some
981: factor, seen to be of the order of 3 in one large sample.
982: The resulting distribution of lifetimes has an asymmetric tail of
983: unusually long-lived clouds. We suggest that the existence of such a
984: tail may introduce a bias in the observed samples of star-forming clouds.
985: Most star formation will take place in the more frequent,
986: shorter lived clouds, while observations of clouds will tend to focus
987: on the fewer longer lived ones.
988:
989: We have seen that the numerical resolution requirements needed to
990: study cloud collapse are very stringent, and we expect they will be even
991: more stringent in 3D.
992: There is a necessity of resolving the possible equilibrium structures,
993: such as the Spitzer sheets, which we have seen fully formed in the
994: subcritical clouds, and partially formed during the run-up to
995: instability of the mildly supercritical ones.
996: The thickness of these sheets scale with the number $\nJ$ of
997: Jeans lengths as $\nJ^{-2}$.
998: Accommodating a large number $\nJ$ of Jeans lengths
999: inside the computational volume will therefore be numerically challenging.
1000: Increasing $\nJ$ by only a factor of 2 requires increasing the space
1001: resolution by a factor of 4.
1002: Unless adaptive mesh refinement (AMR) is used, this requires
1003: increasing the simulation runtime by factors on the order of $64=4^3$
1004: in 2D, and $256=4^4$ in 3D.
1005: We anticipate that AMR will be used in many of the
1006: successful simulations of core formation in the future.
1007:
1008:
1009: Numerical stability, through the Truelove condition, sets a maximum
1010: density that can be accommodated at a given spatial resolution.
1011: Shocks in strongly turbulent flows have large compression ratios,
1012: sometimes requiring increasing resolution in order to distinguish a
1013: transient density increase due to a shock from an authentically
1014: unstable accumulation of mass able to form a collapsed object.
1015:
1016: We have seen that artificially enforcing numerical density floors,
1017: even relatively large ones, on the order of $10^{-4}$ times the
1018: background density, had almost no influence in the evolution of the
1019: collapse.
1020: This result is again not surprising, because wide regions of small
1021: density have little influence on the dense, self-gravitating regions
1022: that undergo collapse. Density floors can significantly speed up ideal MHD
1023: simulations, whose Courant timestep is often limited by large
1024: Alfv\'en speeds $B/\sqrt{4\pi\rho}$ in the least dense regions.
1025:
1026: This work is limited due to the periodic boundary conditions.
1027: We believe this may have favored the collection of clumps
1028: into larger clumps until the instability can take
1029: place. Some simulations occasionally show fast-moving clumps
1030: flowing past each other, and later merging
1031: once one of them returns through the other side of the periodic
1032: computational volume. The periodic boundary conditions
1033: make it plausible that sooner or later, most of the mass in a given
1034: fieldline will collect into a single clump, which then
1035: can undergo instability if its mass is even slightly supercritical.
1036: In real clouds with ordered magnetic fields, clumps inside the same
1037: fieldline but moving in opposite directions are not expected to merge;
1038: however, it is improbable this will apply to all of the fieldlines and
1039: so we expect that the instability will still take place in a similar form,
1040: albeit with an additional stochastic factor in the cloud lifetime.
1041:
1042: Two-dimensionality is also a limitation of this work.
1043: It has strongly limited the topological possibilities for the fieldlines;
1044: it is conceivable that the consequent limitations in motion have favored
1045: the collection of mass into massive sheets and other structures.
1046: Observations \citep[e.g.,][]{gbmm90, crutcher04}, and
1047: 3D simulations and studies \citep[e.g.,][]{basu00, glso03}
1048: indeed indicate that sheets aligned perpendicular to the magnetic field
1049: are not always the preferred possibility for the long term development
1050: of clouds. More variety of clump shapes is expected in a 3D study.
1051: The larger variety in motions allowed by a 3D magnetic field
1052: is expected to enhance the already observed stochastic effects, and
1053: perhaps might also delay mass collection into potentially unstable structures.
1054: However, even in 3D, the simulations performed by \citet{osg01}
1055: suggest that the stability criterion will still be dominated by the
1056: mass-to-flux ratio.
1057:
1058: In some of our models, artificial numerical diffusion has
1059: turned an initially uniform mass-to-flux ratio $\lambda$
1060: into a non-uniform distribution, sometimes with striking
1061: effects on the numerical stability.
1062: While this has a numerical origin, non-uniform
1063: distributions of mass-to-flux are also expected on astrophysical grounds.
1064: For instance, turbulence provides structures and shocks
1065: with small lengthscales and strong magnetic gradients,
1066: conditions favorable to a localized, efficient ambipolar diffusion,
1067: which can redistribute mass and magnetic flux independently.
1068: Cloud collisions can also merge together portions of gas
1069: having different masses and magnetic fields.
1070: We plan to study directly the physical effect of a
1071: non-uniform mass-to-flux ratio in our future work.
1072:
1073: \acknowledgments
1074:
1075: This work was supported by NASA grant NAG 5-9180. We thank Jon McKinney,
1076: Eve Ostriker, Zhi-Yun Li, and Chris Matzner for comments.
1077:
1078: \begin{thebibliography}{}
1079:
1080: \bibitem[Arons \& Max(1975)]{am75}
1081: Arons, J., \& Max, C. E. 1975,
1082: \apj, 196, L77
1083:
1084: \bibitem[Basu(2000)]{basu00}
1085: Basu, S. 2000,
1086: \apj, 540, L103
1087:
1088: \bibitem[Chandrasekhar \& Fermi(1953)]{cf53a}
1089: Chandrasekhar, S., \& Fermi, E. 1953,
1090: \apj, 118, 113
1091:
1092: \bibitem[Ciolek \& Mouschovias(1995)]{cm95}
1093: Ciolek, G. E., \& Mouschovias, T. Ch.\ 1995,
1094: \apj, 454, 194
1095:
1096: \bibitem[Crutcher(2004)]{crutcher04}
1097: Crutcher, R. M. 2004,
1098: in The Magnetized Interstellar Medium,
1099: eds.\ B. Uyaniker, W. Reich \& R. Wielebinski,
1100: (Copernicus GmbH: Katlenburg-Lindau), 123
1101:
1102: \bibitem[Draine, Roberge, \& Dalgarno(1983)]{drd83}
1103: Draine, B. T., Roberge, W. G., \& Dalgarno, A. 1983,
1104: \apj, 264, 485
1105:
1106: \bibitem[Elmegreen \& Elmegreen(1978)]{ee78}
1107: Elmegreen, B. G., \& Elmegreen, D. M. 1978,
1108: \apj, 220, 1051
1109:
1110: \bibitem[Evans \& Hawley(1988)]{eh}
1111: Evans, C. R., \& Hawley, J. F. 1988,
1112: \apj, 332, 659
1113:
1114: \bibitem[Frigo \& Johnson(2005)]{fftw}
1115: Frigo, M., \& Johnson, S. G. 2005,
1116: Proc.\ IEEE, 93, 216
1117:
1118: \bibitem[Gammie et al.(2003)]{glso03}
1119: Gammie, C. F., Lin, Y.-T., Stone, J. M., \& Ostriker, E. C. 2003,
1120: \apj, 592, 203
1121:
1122: \bibitem[Gammie \& Ostriker(1996)]{go96}
1123: Gammie, C. F., \& Ostriker, E. C. 1996,
1124: \apj, 466, 814
1125:
1126: \bibitem[Goldreich \& Kwan(1974)]{gk74}
1127: Goldreich, P., \& Kwan, J. 1974,
1128: \apj, 189, 441
1129:
1130: \bibitem[Goldstein(1978)]{gol78}
1131: Goldstein, M. L. 1978,
1132: \apj, 219, 700
1133:
1134: \bibitem[Goodman et al.(1990)]{gbmm90}
1135: Goodman, A. A., Bastien, P., Myers, P. C., \& M{\'e}nard, F. 1990,
1136: \apj, 359, 363
1137:
1138: \bibitem[Hawley \& Stone(1995)]{hs}
1139: Hawley, J. F., \& Stone, J. M. 1995,
1140: Comput.\ Phys.\ Commun., 89, 127
1141:
1142: \bibitem[Kulsrud \& Pearce(1969)]{kp69}
1143: Kulsrud, R., \& Pearce, W. P. 1969,
1144: \apj, 156, 445
1145:
1146: \bibitem[Ledoux(1951)]{led51}
1147: Ledoux, P. 1951,
1148: Ann.\ d'Astrophys.\ 14, 438
1149:
1150: \bibitem[Mac Low et al.(1998)]{ml98}
1151: Mac Low, M.-M., Klessen, R. S., Burkert, A., \& Smith, M. D. 1998,
1152: \prl, 80, 2754
1153:
1154: \bibitem[Mac Low \& Klessen(2004)]{mk04}
1155: Mac Low, M.-M., \& Klessen, R. S., 2004
1156: Rev.\ Mod.\ Phys., 76, 125
1157:
1158: \bibitem[McKee(1989)]{mk89}
1159: McKee, C. F. 1989,
1160: \apj, 345, 782
1161:
1162: \bibitem[Mestel \& Spitzer(1956)]{ms56}
1163: Mestel, L., \& Spitzer, L., Jr.\ 1956,
1164: \mnras, 116, 503
1165:
1166: \bibitem[Nakano \& Nakamura(1978)]{nn78}
1167: Nakano, T., \& Nakamura, T. 1978,
1168: \pasj, 30, 671
1169:
1170: \bibitem[Nakano(1988)]{nakano88}
1171: Nakano, T. 1988,
1172: \pasj, 40, 593
1173:
1174: \bibitem[Ostriker, Gammie, \& Stone(1999)]{ogs99}
1175: Ostriker, E. C., Gammie, C. F., \& Stone, J. M. 1999,
1176: \apj, 513, 259
1177:
1178: \bibitem[Ostriker, Stone, \& Gammie(2001)]{osg01}
1179: Ostriker, E. C., Stone, J. M., \& Gammie, C. F. 2001,
1180: \apj, 546, 980
1181:
1182: \bibitem[Sagdeev \& Galeev(1969)]{sag69}
1183: Sagdeev, R. Z., \& Galeev, A. A. 1969,
1184: Nonlinear Plasma Theory (New York: W. A. Benjamin)
1185:
1186: \bibitem[Shu, Adams, \& Lizano(1987)]{sal87}
1187: Shu, F. H., Adams, F. C., \& Lizano, S.\ 1987,
1188: \araa, 25, 23
1189:
1190: \bibitem[Spitzer(1942)]{spitz42}
1191: Spitzer, L., Jr.\ 1942,
1192: \apj, 95, 329
1193:
1194: \bibitem[Stone \& Norman(1992a)]{sn92a}
1195: Stone, J. M., \& Norman, M. L. 1992,
1196: \apjs, 80, 753
1197:
1198: \bibitem[Stone \& Norman(1992b)]{sn92b}
1199: Stone, J. M., \& Norman, M. L. 1992,
1200: \apjs, 80, 791
1201:
1202: \bibitem[Stone, Ostriker, \& Gammie(1998)]{sog98}
1203: Stone, J. M., Ostriker, E. C., \& Gammie, C. F. 1998,
1204: \apjl, 508, L99
1205:
1206: \bibitem[Truelove et al.(1997)]{truelove97}
1207: Truelove, J. K., Klein, R. I., McKee, C. F.,
1208: Holliman, J. H., II,
1209: Howell, L. H.,
1210: \&
1211: Greenough, J. A.
1212: 1997,
1213: \apj, 489, L179
1214:
1215: \bibitem[Zuckerman \& Palmer(1974)]{zp74}
1216: Zuckerman, B., \& Palmer, P. 1974,
1217: \araa, 12, 279
1218:
1219: \end{thebibliography}
1220:
1221: {\vskip 15ex
1222: \plotone{f1.eps}}
1223: {\figcaption[f1.eps]{\label{fig:fiducial}
1224: Colormaps of $\rho$ in the fiducial run. Snapshots at times $t=0.01$,
1225: $0.02$, $0.04$, $0.06$, $0.12$, $0.18$, $0.24$, $0.30$, and $0.34$.
1226: The logarithmic
1227: colorscale goes from dark blue (saturating on black) to deep red,
1228: corresponding to densities going from $0.01\rhoS = 0.4441\rhomean$ to
1229: $10\rhoS=444.1\rhomean$, where $\rhoS$ is the peak density of an
1230: equilibrium Spitzer sheet for the given parameters.
1231: }}
1232:
1233: \plotone{f2.eps}
1234: {\figcaption[f2.eps]{\label{fig:fiducial_lines}
1235: Field lines in the fiducial run. The figure shows snapshots at times $t
1236: = 0.01$, $0.02$, $0.04$, $0.06$, $0.12$, $0.18$, $0.24$, $0.30$, and $0.34$.
1237: }}
1238:
1239: \plotone{f3.eps}
1240: {\figcaption[f3.eps]{\label{fig:fiducial_max_density}
1241: Increase of the maximum mass density $\rho_{\max}$ with time
1242: in the fiducial run. The run finishes when $\rho_{\max}=1820$,
1243: the Truelove value, at a time $t=0.341$, slightly beyond the plotted
1244: region.
1245: Some of the transient peaks shown here could have provoked a numerical
1246: instability at a resolution smaller than the value $N=512$
1247: adopted for the fiducial run.
1248: }}
1249:
1250: \plotone{f4.eps}
1251: {\figcaption{\label{fig:fiducial_energies}
1252: Variations in time of the turbulent kinetic energy $\EK$, the total
1253: magnetic energy $E_B$, and (minus) the gravitational energy
1254: $-E_G$.
1255: }}
1256:
1257: {\quad\\ \vskip 15ex
1258: \plotone{f5.eps}}
1259: {\figcaption{\label{fig:distribution}
1260: Frequency of the different values of $\tT$. Sixty simulations have been
1261: run with the parameters $\lambda=1.5$, $\EK=50$, $N=512$, and different
1262: random seeds. Each of these runs has reported a value of $\tT$. This
1263: plot was constructed by summing 60 Gaussian profiles (with
1264: $\sigma=0.03 \ts=0.9\Myr$) centered at each of these $\tT$ values.
1265: Collapse times range from $\sim 7$ to $20\Myr$.
1266: }}
1267:
1268:
1269: \end{document}
1270: