1: %\documentclass[preprint2]{aastex}
2: %\documentclass[12pt,preprint]{aastex}
3: \documentclass{emulateapj}
4:
5: %\usepackage{amssymb}
6: %\usepackage{amsmath}
7: %\usepackage{graphicx}
8:
9: \newcommand{\lO} {\lambda_{\Omega}}
10: \newcommand{\led} {\lambda_{e}}
11: \newcommand{\cAi} {c_{Ai}}
12: \newcommand{\cAn} {c_{An}}
13: \newcommand{\RM} {R_{M}}
14: \newcommand{\mbfB}{\mathbf{B}}
15: \newcommand{\mbfu}{\mathbf{u}}
16: \newcommand{\mbfui}{\mathbf{u}_i}
17: \newcommand{\mbfun}{\mathbf{u}_n}
18: \newcommand{\mbfz}{\mathbf{z}}
19: \newcommand{\mbfn}{\mathbf{n}}
20: \newcommand{\mbfk}{\mathbf{k}}
21: \newcommand{\mbfx}{\mathbf{x}}
22: \newcommand{\mbfv}{\mathbf{v}}
23: \newcommand{\mbfnabla}{\mathbf{\nabla}}
24: \newcommand{\f} {\frac}
25: \newcommand{\ddx}{\partial_x}
26: \newcommand{\ddy}{\partial_y}
27: \newcommand{\ddt}{\partial_t}
28: \newcommand{\lomega}{\lambda_\Omega}
29:
30: \begin{document}
31:
32: \title{Magnetized Non-linear Thin Shell Instability: Numerical Studies in 2D}
33:
34: \author{Fabian Heitsch\altaffilmark{1}}
35: \author{Adrianne D. Slyz\altaffilmark{2,3}}
36: \author{Julien E.G. Devriendt\altaffilmark{2}}
37: \author{Lee W. Hartmann\altaffilmark{1}}
38: \author{Andreas Burkert\altaffilmark{4}}
39: \altaffiltext{1}{Dept. of Astronomy, University of Michigan, 500 Church St.,
40: Ann Arbor, MI 48109-1042, U.S.A}
41: \altaffiltext{2}{Universit\'e Claude Bernard Lyon 1,
42: CRAL, Observatoire de Lyon, 9 Avenue Charles Andr\'{e},
43: 69561 St-Genis Laval Cedex, France; CNRS, UMR 5574; ENS Lyon}
44: \altaffiltext{3}{Oxford University, Astrophysics, Denys Wilkinson Building, Keble Road,
45: Oxford, OX1 3RH, United Kingdom}
46: \altaffiltext{4}{University Observatory Munich, Scheinerstr. 1, 81679 Munich, Germany}
47: \lefthead{Heitsch et al.}
48: \righthead{Magnetized NTSI in 2D}
49:
50: \begin{abstract}
51: We revisit the analysis of the Non-linear Thin Shell Instability (NTSI)
52: numerically, including magnetic fields. The magnetic tension force is expected to
53: work against the main driver of the NTSI -- namely transverse momentum transport.
54: However, depending on the field strength and orientation,
55: the instability may grow. For fields aligned with the inflow, we find that the NTSI is suppressed
56: only when the Alfv\'en speed surpasses the (supersonic) velocities
57: generated along the collision interface. Even for fields perpendicular
58: to the inflow, which are the most effective at preventing the NTSI from developing, internal structures
59: form within the expanding slab interface, probably leading to fragmentation in the presence
60: of self-gravity or thermal instabilities. High Reynolds numbers result in local turbulence
61: within the perturbed slab, which in turn triggers reconnection and dissipation of the
62: excess magnetic flux. We find that when the magnetic field is initially aligned with the flow,
63: there exists a (weak) correlation between field strength and gas density. However, for transverse
64: fields, this correlation essentially vanishes. In light of these results, our general conclusion is
65: that instabilities are unlikely to be erased unless the magnetic energy in clouds is much larger than
66: the turbulent energy. Finally, while our study is motivated by the scenario of molecular cloud formation
67: in colliding flows, our results span a larger range of applicability, from supernovae shells to colliding
68: stellar winds.
69: \end{abstract}
70: \keywords{instabilities --- MHD --- turbulence --- methods:numerical
71: --- ISM:clouds --- ISM: magnetic fields}
72:
73: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
74: %
75: %\section{Motivation}
76: %
77: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
78: \section{Motivation}
79:
80: Shocks and shells exist abundantly in the interstellar medium (ISM).
81: Driven by supernovae, expanding HII-regions, gravitational
82: flows or wholesale cloud collisions, they not only strongly influence the ISM
83: dynamics, but also affect its chemistry.
84: However, structural analyses
85: of the ISM show spectral indices closer to the Kolmogorov
86: value of incompressible turbulence (see \citealp{2004ARA&A..42..211E} for a review).
87: Independently of the problem of how well these indices constrain the type of turbulence,
88: there are plenty of physical mechanisms to explain the
89: closeness to the Kolmogorov value, ranging from the intrinsic nature of MHD turbulence
90: (e.g. \citealp{1995ApJ...438..763G}, \citealp{2002PhRvL..89c1102B},
91: \citealp{2003MNRAS.345..325C}) to the
92: conversion of compressible to solenoidal modes (\citealp{1994ApJ...436..728F},
93: \citealp{2004ARA&A..42..211E}).
94:
95: It is the latter mechanism that motivated this study. In the absence
96: of shear flows (oblique shocks) and thermal instabilities, the
97: Non-linear Thin Shell Instability \citep{1994ApJ...428..186V} provides a natural
98: mechanism to convert compressible motions into solenoidal ones.
99: The NTSI is a rippling instability, relying on transverse momentum
100: transport due to bends in the collision interface of two opposing flows.
101: It is likely to arise in a wide range of environments, from colliding stellar winds,
102: supernova shells, colliding HI streams/clouds, to galaxy mergers.
103:
104: The NTSI has been widely studied numerically (see \citealp{2006ApJ...648.1052H} for
105: a summary of the literature), mostly focusing on the effects of
106: self-gravity and thermal instabilities.
107: Our interest in the NTSI comes from the role it plays in the evolution
108: of molecular clouds in colliding HI flows
109: \citep{2005ApJ...633L.113H,2006ApJ...643..245V,2006ApJ...648.1052H},
110: but the results have a wider applicability. With the exception
111: of \citet{1998ApJ...497..777K}, who included the magnetic pressure term in their
112: study of cloud collisions, work on the NTSI has so far neglected the
113: effect of magnetic fields. However, fields could have a deciding influence on the
114: evolution of a shock-bounded slab. Motivated by the numerical models of
115: \citet{1995ApJ...441..702V} and \citet{1995ApJ...455..536P},
116: \citet{2001ApJ...562..852H} and \citet{2004ApJ...612..921B}
117: suggested that fields could, in fact, lead to a selection effect for
118: molecular cloud formation: clouds can only form if the fields are aligned
119: with the flows assembling the gas.
120:
121: As a first step, we revisit the isothermal analysis of \citet{1994ApJ...428..186V}
122: numerically, and study the evolution of the NTSI in a
123: two-dimensional, magnetohydrodynamical environment. While the general
124: expectation (\S\ref{ss:physics}) is met in the laminar case, namely that
125: magnetic fields can efficiently damp the NTSI, the geometry of the rippled interface induces
126: non-ideal MHD effects, requiring a numerical method capable of handling
127: such effects in a stable and accurate way (\S\ref{ss:numerics}).
128: However, we find that the exact amount of dampening crucially depends on the field
129: orientation and strength (\S\ref{s:results}). This is
130: especially true in the turbulent case, where turbulent reconnection inside the
131: over-pressured slab leads to a pressure deficit, thus compressing the gas even further.
132: Finally, we show that the correlation between field strength and
133: gas density is at best weak, even in the case of fields
134: perpendicularly oriented with respect to the inflow, for which the field
135: would be expected to scale linearly with the density.
136:
137: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
138: %
139: %\section{Physics and Numerics}
140: %
141: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
142: \section{Physics and Numerics\label{s:physnum}}
143:
144: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
145: %\subsection{Physics}
146: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
147: \subsection{Physics\label{ss:physics}}
148: The growth rate of the NTSI is mostly controlled by $k\eta$, the product
149: of the wave number of the slab perturbation $k$, and the amplitude of the
150: slab's initial displacement $\eta$ (equivalently, the amplitude of the
151: collision interface's geometrical perturbation).
152: The instability is driven by lateral transport of longitudinal momentum,
153: i.e. if the inflow is parallel to the $x$ direction, and the slab is in the
154: $y$-$z$-plane, $x$-momentum is transported laterally in $y$ (and $z$),
155: collecting at the focal points of the perturbed slab. The efficiency of lateral
156: momentum transport is key to the development of the instability, since
157: it is the imbalance of ram pressure at the focal points that eventually propels
158: matter forward, driving the growth of the slab's perturbation.
159: \citet{1994ApJ...428..186V} derived a growth rate of
160: \begin{equation}
161: \omega \approx c_sk (k\eta)^{1/2},\label{e:vishniac}
162: \end{equation}
163: with the sound speed $c_s$. \citet{1996NewA....1..235B}
164: found that at constant $\eta$ and for small $k$s,
165: equation~(\ref{e:vishniac}) yields only a lower limit, while for large $k$s, the
166: analytical growth rates agree well with the numerical results.
167: The reason for this seems to lie in the efficiency of deflecting the incoming
168: flow: for small $k$s, a small fraction of the incoming flow's momentum
169: is converted to lateral motions, while a large part compresses the
170: slab (depending on the equation of state, this could lead to an increase in energy
171: losses).
172:
173: After the initial growth-phase (eq.~[\ref{e:vishniac}]), the NTSI reaches saturation
174: through two mechanisms: (i) expansion of the slab which stops the lateral momentum
175: transport by preventing the inflow from reaching the focal points, and (ii)
176: shear flow (Kelvin-Helmholtz) instabilities (KHI) in regions of the slab connecting
177: the focal points. The KHIs both generate inner structure and revive the slab's
178: expansion. Note that strong cooling (not modeled here) can also suppress
179: the NTSI via early fragmentation \citep{2003NewA....8..295H}.
180:
181: Qualitatively, we expect magnetic fields to prevent the NTSI and
182: subsequent KHI-modes from occurring. However, the detailed quantitative
183: extent of the damping should depend on the orientation
184: of the field with respect to the inflow. Indeed, fields aligned with the inflow
185: resist instabilities via the magnetic tension force, and therefore should be more efficient
186: in suppressing the NTSI when $k\eta$ is small, even though the strong pairwise
187: field reversals arising from the opposed shear velocities along the slab (see \S\ref{ss:estimate})
188: could trigger reconnection. On the other hand,
189: fields perpendicular to the inflow (but still in the 2D plane), primarily prevent
190: instabilities from growing because of the magnetic pressure term in the Lorentz
191: force, and to a lesser extent, because of
192: magnetic tension. For the sake of completeness, we mention that the third possible
193: field configuration in 2D, i.e. the one in which the magnetic field is perpendicular
194: to the flow plane, is irrelevant for this study since in that case the gas behaves
195: as a system with an adiabatic exponent of $2$ for which the NTSI cannot be excited
196: \citep{1994ApJ...428..186V}.
197:
198: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
199: %\subsection{Numerics}
200: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
201: \subsection{Numerics\label{ss:numerics}}
202: The magnetohydrodynamical scheme\footnote{We called the scheme PROTEUS, under
203: which name we will refer to it subsequently. Proteus is a lesser
204: god in Greek mythology, also known as "The Old Man of the Sea".
205: He lives in the sea off the coast of
206: Egypt and can see things in the past, presence and future, but
207: is very unwilling to share his knowledge. In order to evade questions,
208: he has the ability to change his appearance. However, if you manage
209: to catch and hold him, he will assume his true shape and answer
210: your questions.} is based on a conservative gas-kinetic flux-splitting
211: method, introduced by \citet{1999JCoPh.153..334X} and
212: \citet{2000JCoPh.165...69T} and derived from the 1st-order BGK
213: \citep{1954PhRv...94..511B} model. Representing
214: the velocity distributions as Maxwellians in each cell, fluxes
215: across cell walls are derived from the differences in the velocity moments of
216: Maxwellian distributions reconstructed at the cell walls.
217: The reconstruction is second order in space using MUSCL limiters,
218: and it allows a fast and consistent way to implement
219: viscosity and Ohmic resistivity in the form of dissipative
220: fluxes \citep{2004ApJ...603..165H} at close to zero extra computing cost, while preserving the
221: time order of PROTEUS since the dissipative terms are not simply added as source terms
222: but are part of the flux computation. This allows us to control dissipation
223: in a physical manner, without having to rely on numerical dissipation
224: to terminate the turbulent cascade at grid scale. Total energy is conserved
225: at machine-accuracy level for an adiabatic equation of state. In the isothermal
226: version which we are using in \S\ref{s:results}, the total energy equation is not evolved. PROTEUS uses a
227: 2nd order TVD Runge-Kutta time stepping \citep{1988JCP..77..439S} for the MHD equations
228: to achieve 2nd order temporal accuracy. Fluxes are updated
229: in time-unsplit fashion, i.e. flux updates for spatial directions
230: are computed using the initial conditions of the current time step.
231: In order to keep $\nabla\cdot\mbfB=0$, PROTEUS employs
232: a Hodge projection \citep{1994JSSC..15..263Z,1998ApJS..116..133B}. The code is fully
233: message passing interface (MPI) parallelized.
234:
235: With PROTEUS, one may switch between the
236: MHD-solver previously described and a purely hydrodynamical solver based on
237: the 2nd-order BGK model. The latter implementation has been
238: introduced and extensively discussed by \citet{1993JCoPh.109...53P}, \citet{1999A&AS..139..199S},
239: \citet{2006ApJ...648.1052H} and \citet{slyz2006submit}, and we therefore refer the reader interested
240: in implementation details to these papers.
241:
242: One-dimensional shock tests
243: and the Orszag-Tang vortex have already been discussed by \citet{2000JCoPh.165...69T},
244: hence in what follows, we focus on three other MHD test cases.
245: The two first ones, i.e. the propagation of a linear
246: Alfv\'{e}n wave under resistive damping (\S\ref{ss:linalfven})
247: and the current sheet evolution (\S\ref{ss:currentsheet})
248: are both meant to test the resistive fluxes, while the third one, i.e.
249: the advection of a field loop (\S\ref{ss:fieldloop}) is a geometry test.
250: A detailed study of the magnetized Kelvin-Helmholtz
251: Instability is under way \citep{palotti2006submit}.
252:
253: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
254: %\subsubsection{Propagation of a linear Alfv\'{e}n Wave}
255: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
256: \subsubsection{Propagation of a linear Alfv\'{e}n Wave\label{ss:linalfven}}
257: This one-dimensional test checks the resistive flux implementation
258: as well as the accuracy of the overall scheme.
259: A linear Alfv\'{e}n wave under weak Ohmic dissipation is damped
260: at a rate of
261: \begin{equation}
262: \omega_i = \f{1}{2}\lomega k^2,\label{e:alfven}
263: \end{equation}
264: where $\lomega$ is the Ohmic resistivity, and $k=2\pi\kappa/L$ is the
265: wave number of the Alfv\'{e}n wave, with $\kappa\in\mathbb{N}$.
266: The strongly damped case, where the decay dominates the time evolution,
267: is physically uninteresting for our application, since the Ohmic
268: resistivity is mainly used to control numerical dissipation.
269: Figure~\ref{f:alfven} shows the damping rate against Ohmic resistivity
270: $\lomega$ for $\kappa=1,2,4$ at a grid resolution of $N=64$.
271:
272: \begin{figure}
273: % \includegraphics[width=\columnwidth]{../figures/cralfvenkx.eps}
274: \includegraphics[width=\columnwidth]{f1.eps}
275: \caption{\label{f:alfven}Damping rate (eq.~[\ref{e:alfven}]) of a
276: linear Alfv\'{e}n wave against Ohmic resistivity for
277: $\kappa=1,2,4$. The resolution is $N=64$. Errors of the
278: measured damping rates are smaller than the symbol sizes.
279: Lines denote the analytical solution.}
280: \end{figure}
281:
282: From Figure~\ref{f:alfven}, it is clear that, as one diminishes the value of
283: $\lomega$, there comes a point when the numerical resistivity of the scheme
284: becomes comparable to the physical one, causing the measured damping rate to
285: flatten out and depart from the analytical solution. For $\kappa=4$ and
286: $\lomega=0.1$, the wave decays too quickly to allow a reliable measurement,
287: and the system enters the strongly damped branch of the dispersion relation.
288: However, we emphasize that, even for this high value of $\kappa$ in light of
289: the modest resolution we used, the resistivity range available to PROTEUS
290: spans three orders of magnitude.
291:
292: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
293: %\subsubsection{Current Sheet}
294: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
295: \subsubsection{Current Sheet\label{ss:currentsheet}}
296: This test is taken from \citet{1995CoPhyCom.89..127H} and the ATHENA test
297: suite\footnote{{\tt http://www.astro.princeton.edu/\\
298: $\sim$jstone/tests/field-loop/Field-loop.html}}.
299: A square domain of extent $-0.5\leq x,y\leq 0.5$ and of constant density $n_0=1$
300: and pressure $p_0$ is permeated by a magnetic field along the $y$ direction such that
301: $B_y(|x|<0.25)=\sqrt{4\pi}$, and $B_y=-\sqrt{4\pi}$ elsewhere. This results in two
302: magnetic null lines, which then are perturbed by velocities $v_x=A\sin(2\pi y)$.
303: The goal is to find the pressure $p_0$ and velocity amplitude $A$ for which the code
304: crashes. The main problem is -- especially in conservative schemes -- that the
305: resistive decay of the field leads to strong localized heating that in turn
306: generates strong magnetosonic waves. Thus, the smaller $p_0$ and/or the larger
307: $A$, the harder the test. We chose $p_0=0.1$ and $A=0.3$ fairly close to
308: the ``standard'' values quoted on the ATHENA web site$^6$, $p_0=0.05$ and $A=0.1$.
309: Here, we use an adiabatic exponent of $\gamma=5/3$ and employ the
310: conservative formulation of the scheme.
311:
312: The test turns out to be very hard for PROTEUS, because of its low
313: numerical diffusivity. Figure~\ref{f:emagtime} summarizes the test results
314: in the form of the total magnetic energy against time. The energy evolution
315: splits into two branches: one corresponding to the lower-resolution models at
316: all resistivities, and the other representing the higher-resolution models
317: at $\lomega=0$ and $\lomega=10^{-5}$. At $\lomega=10^{-4}$, the field decay has
318: converged: both resolutions give the same curve. At $N=256^2$ and $\lomega=0$,
319: the code crashes around $t=6$. All other models run up to $t=10$ and further.
320: \begin{figure}
321: % \includegraphics[width=\columnwidth]{../figures/emagtime.eps}
322: \includegraphics[width=\columnwidth]{f2.eps}
323: \caption{\label{f:emagtime}Total magnetic energy against time for the current sheet
324: test. A finite resistivity $\lomega$ helps stabilize the code.}
325: \end{figure}
326:
327: The result confirms the discussion by \citet{2000JCoPh.165...69T}, namely that while
328: the conservative gas-kinetic flux splitting method in the BGK-formalism performs well
329: for high-$\beta$ plasmas (with $\beta$ defined as the ratio of thermal over magnetic pressure),
330: it might not be the method of choice for low-$\beta$ plasmas, i.e.
331: magnetically dominated systems. Note, however, that this is mainly a consequence of the
332: scheme's conservative formulation.
333: We experimented with a non-conservative version (i.e. just evolving the internal energy
334: instead of the total energy), which was stable for lower $p_0$ and higher $A$, albeit at the cost
335: of a less accurate total energy evolution.
336:
337: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
338: %\subsubsection{Advection of a Field Loop}
339: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
340: \subsubsection{Advection of a Field Loop\label{ss:fieldloop}}
341:
342: A cylindrical current distribution (i.e. a field loop) is advected diagonally across
343: the simulation domain. We follow the implementation presented by \citet{2005JCoPh.205..509G}
344: and the ATHENA test suite$^6$,
345: based on an earlier version by \citet{1996JCoPh.128...82T}. Density and pressure are both initially
346: uniform at $n_0=1$ and $p=1$, and the fluid is described as an ideal gas with an adiabatic exponent
347: of $\gamma=5/3$. The square
348: grid ranges from $-0.5\leq x \leq 0.5$, and the loop is advected at an angle of $30$ degrees
349: with respect to the $x$-axis. Thus, two round trips in $x$ correspond to one crossing in $y$.
350: The amplitude of the field loop is set to values $10^{-3}$, $10^{-2}$ and $10^{-1}$, with an
351: initial radius of $R_0=0.3$. Figure~\ref{f:fieldloop} shows the initial current distribution
352: with the magnetic field vectors over-plotted ({\em left}), and the final current
353: distribution after two time-units measured in horizontal crossing times.
354: The overall shape is preserved, although some artifacts are visible at the upper
355: rim of the loop. These results concerning the shape are qualitatively similar
356: to those posted on the above mentioned website.
357: Note that we show the current density, since the artifacts do not show up in the
358: magnetic energy. This test uses $\lomega \equiv 0$.
359:
360: \begin{figure}
361: % \includegraphics[width=\columnwidth]{../figures/loopadvect.eps}
362: \includegraphics[width=\columnwidth]{f3c.eps}
363: \caption{\label{f:fieldloop}Current density for field loop advection test
364: (see \S\ref{ss:fieldloop}).
365: {\em Left:} Initial condition. {\em Right:} After two horizontal crossings. The grid resolution
366: is $256^2$.}
367: \end{figure}
368:
369: Figure~\ref{f:decayloop} presents a quantitative diagnostic of the behavior of the code as it tracks
370: the magnetic energy decay against the simulation time,
371: in units of horizontal crossing times. The initial energy is normalized to one.
372: Line styles denote different amplitudes of the field strength. The solid lines correspond
373: to the case given by \citet{2005JCoPh.205..509G}, the dashed and dash-dotted lines denote cases with larger
374: amplitudes. At an amplitude of $A=10^{-3}$ and a resolution of $128^2$, the magnetic
375: energy decays by $3.3$\% over two horizontal crossing times. The ATHENA website$^6$ quotes a decay
376: of $3.5$\% with a $256x148$ grid, using a Roe solver
377: and 3rd order reconstruction.
378:
379: \begin{figure}
380: % \includegraphics[width=\columnwidth]{../figures/decay_fieldloop.eps}
381: \includegraphics[width=\columnwidth]{f4.eps}
382: \caption{\label{f:decayloop}Normalized magnetic energy against time
383: (in units of horizontal crossing time), for two resolutions
384: and three field strengths. The highest amplitude leads to waves when perturbed
385: by advection. At a resolution of $128^2$, the magnetic energy has decayed by $3.3$\% after
386: two crossing times. The fit decay times $\tau$ are indicated for each model.}
387: \end{figure}
388:
389: In summary, these numerical test cases demonstrate that PROTEUS models dissipative MHD effects
390: accurately, due to a low intrinsic numerical diffusivity that compares well with that of higher-order
391: Godunov methods. Furthermore, it can advect geometrically complex magnetic field patterns properly, and
392: is well suited in its energy conserving form to model MHD flows with $\beta > 1$.
393:
394: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
395: %\subsection{Initial Conditions}
396: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
397: \subsection{Initial Conditions\label{ss:initcond}}
398:
399: To remain as close as possible to \citet{1994ApJ...428..186V}, we will use the isothermal
400: version of PROTEUS in the following. The initial conditions are similar to those
401: discussed in \citet{2005ApJ...633L.113H,2006ApJ...648.1052H}.
402: Two uniform, identical flows in
403: the $x$-$y$ computational plane initially collide head-on at a sinusoidal
404: interface with given wave number $k_y$ and amplitude $\eta$. The field is either
405: aligned or perpendicular to the inflow, but in both cases in the $x$-$y$ plane.
406: For the standard runs, we used a rectangular grid with an extent of $88$~pc
407: in $x$ and $44$~pc in $y$. Field strength as well as viscosity and resistivity
408: are varied. The grid resolution varies between $N_x\times N_y=256\times 128$ and
409: $2048\times 1024$ by factors of $2$ in linear resolution.
410: The isothermal sound speed is $c_S = 5.3$ km s$^{-1}$, the
411: Mach number of the instreaming gas is ${\cal M} = 4$, and the inflow density
412: is set to $n_0 = 1$ cm$^{-3}$. Thus, in the code unit system, the Alfv\'{e}n speed
413: in the inflow region is given by $c_A=B$, the magnetic field strength.
414:
415: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
416: %
417: %\section{Results}
418: %
419: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
420: \section{Results\label{s:results}}
421:
422: We give a rough estimate for the field strength required to prevent the
423: excitation of the NTSI in \S\ref{ss:estimate}.
424: The morphology of the instability naturally depends strongly on the field
425: orientation. We present some examples in \S\ref{ss:morphology}. Because
426: of the strong shear flows, the explicit control of dissipation is
427: crucial in reaching numerical convergence. This can be further quantified by
428: monitoring the growth rates (\S\ref{ss:growthrates}). Finally, we show that the geometry of the
429: flow and magnetic field strongly influence the field-density relation
430: (\S\ref{ss:fielddensity}).
431:
432: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
433: %\subsection{Estimate of Threshold Fieldstrength}
434: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
435: \subsection{Estimate of Threshold Field Strength\label{ss:estimate}}
436:
437: A very rough estimate of the threshold field strength preventing the
438: excitation of the NTSI can be derived by simple pressure considerations.
439: Figure~\ref{f:sketch} gives a sketch of the simplified situation.
440: Only one half of the slab in the vertical direction is shown. The slab
441: is displaced by $\eta$ in the horizontal direction around the
442: (dotted) center line. The angle
443: between the slab and the symmetry line measured at point $0$ is
444: given by $\alpha$, with $\tan\alpha\approx 2\eta k/\pi$.
445: Gas is streaming in horizontally from the left and the right
446: with velocity $\mbfu$, and the magnetic field $\mbfB$ is aligned with
447: the inflow, pointing to the right. Incoming flow with positive velocities
448: exerts a pressure at point $0$, which can be split into a normal
449: component $\overline{0D}$ and a tangential component
450: $\overline{0F}=\overline{0E}\,\sin\alpha$.
451: The tangential component corresponds to the ram pressure exerted
452: by material deflected by the slab and sliding along $\overline{0F}$.
453: To a zeroth order approximation, the component of this ram pressure
454: perpendicular to $\mbfB$ is available for bending the field lines, thus
455: \begin{equation}
456: \rho\,u^2\,\sin\alpha\,\cos\alpha \approx \mbfB^2/2.\label{e:crudeest}
457: \end{equation}
458:
459: \begin{figure}
460: % \includegraphics[width=\columnwidth]{../figures/crudeestimate.eps}
461: \includegraphics[width=\columnwidth]{f5.eps}
462: \caption{\label{f:sketch}Sketch of force geometry for estimating
463: the threshold field strength. A thick solid line marks the
464: interface of the colliding flows (the ``slab''). The magnetic
465: field is represented by horizontal
466: long short-dashed arrows, and inflow velocities by short solid ones.
467: Other (force) vectors are defined in the text.}
468: \end{figure}
469:
470: The maximum is reached for $\alpha = \pi/4$, in which case
471: $\rho u^2 \approx \mbfB^2$. For an inflow velocity of $|u| = 4 c_s$,
472: we thus require a field strength corresponding to
473: an Alfv\'{e}n speed of $c_A\approx 4 c_s$ to suppress the NTSI.
474:
475: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
476: %\subsection{Morphology}
477: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
478: \subsection{Morphology\label{ss:morphology}}
479:
480: We begin with our standard runs, in order to compare to Vishniac's
481: analysis. These are the ``laminar'' cases (\S\ref{sss:laminar}), i.e.
482: cases that do not develop turbulent substructure. The turbulent case
483: and the relevance of fixed physical dissipation scales are discussed
484: in \S\ref{sss:turbulent}.
485:
486: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
487: %\subsubsection{Laminar Case}
488: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
489: \subsubsection{Laminar Case\label{sss:laminar}}
490: The left column of Figure~\ref{f:denshdmhd} displays a model sequence in
491: resolution for the hydro runs, corresponding to Vishniac's analysis. The top
492: panel shows the initial condition (strictly, just after $t=0$),
493: The increasing overall amplitude of the slab signifies that the NTSI is clearly at work.
494: Most of the gas is collected at the focal
495: points, and by the end of the simulations, the system is close to saturation.
496: The lowest resolution run differs from all others in that it is the only one for which
497: numerical convergence has not been
498: achieved. The two highest resolution runs ($N_x=1024,2048$, lower two panels),
499: on the other hand have converged even in detail.
500: Note that the viscosity and resistivity provide fixed
501: physical dissipative scales, independent of the resolution. Thus, the model at
502: $N_x=256$ is not resolved with respect to these dissipative scales, while
503: the high-resolution models are.
504:
505: \begin{figure}
506: % \includegraphics[width=\columnwidth]{../figures/denshdmhd.eps}
507: \includegraphics[width=\columnwidth]{f6c.eps}
508: \caption{\label{f:denshdmhd}{\em Left column:} Logarithmic density maps of
509: hydrodynamical models at $t=3.8$~Myr corresponding to the end of simulation,
510: resolution increasing
511: from top to bottom ($N_x=256$ to $2048$ by factors of $2$). The two
512: highest resolutions have
513: converged even in detail. {\em Right column:} Same as left, but for the magnetic
514: models, where the field is
515: aligned with inflow and $c_A/c_s=1.0$. Again, the two highest resolutions have
516: converged.}
517: \end{figure}
518:
519: This is also true for the magnetic runs (right column of Figure~\ref{f:denshdmhd}) where
520: the field has slowed down the growth of the NTSI. The resulting slab is also more structured than in the
521: pure hydro case. High-density regions, especially thin filaments, coincide with regions
522: of field reversals (loss of magnetic pressure support). To see this, one can compare
523: the gas density to the magnetic energy (Figure~\ref{f:emag}) maps. The center column corresponds to the
524: right column of Figure~\ref{f:denshdmhd}. Not only do the field reversals lead to
525: dense structures, but magnetic waves arise. The left column of
526: Figure~\ref{f:emag} shows the same resolution sequence at half the field strength,
527: i.e. $c_A/c_s = 0.5$, while the right column stands for $c_A/c_s = 2.0$.
528: Higher field strengths not only reduce the growth rate of the NTSI, but also suppress
529: the internal turbulent structure visible for $c_A/c_s=0.5$. Numerical convergence
530: is more easily achieved with higher field strength, which is another indicator that
531: turbulence plays a minor role, i.e. we remain safely entrenched within the laminar regime.
532: This is slightly different in the weak-field case where the two highest resolution runs
533: have only mildly converged. Here, the field starts to get too weak to prevent the
534: excitation of KHI modes.
535:
536: \begin{figure*}
537: % \includegraphics[width=\textwidth]{../figures/emagstrength.eps}
538: \includegraphics[width=\textwidth]{f7c.eps}
539: \caption{\label{f:emag}Logarithm of magnetic energy at $t=3.8$~Myr. {\em Left:} Four resolutions
540: increasing from top to bottom ($N_x=256$ to $2048$ by factors of $2$) at $c_A/c_s=0.5$.
541: {\em Center:} For $c_A/c_s=1.0$. {\em Right:} For $c_A/c_s=2.0$.}
542: \end{figure*}
543:
544: A variation on the theme is shown in Figures~\ref{f:densy} and \ref{f:emagy}
545: where the field is oriented
546: vertically this time around. In a 1D geometry, the magnetic pressure would prevent the
547: gas from efficiently accumulating to form high-density regions (i.e. clouds, see \citealp{2004ApJ...612..921B}).
548: The system would behave as if the gas had an adiabatic exponent of $2$. This is the situation traced
549: out by the models shown in the right column of Figures~\ref{f:densy} and \ref{f:emagy}:
550: because of the stiffened equation of state, high (flux-)density regions expand faster. However,
551: high density is found at the focal points, hence the initial perturbation of the slab
552: is smoothed out, and the slab ends up with plane-parallel shock fronts (due to the fields,
553: there are some waves inside the slab, though). Reducing the field strength (left column) by
554: a factor of $2$ however changes the situation drastically. Although the NTSI is only weak,
555: high-density filaments start to form in the thick slab, again at the locations
556: of field reversals. Thus, a transverse field is much less of an inhibiting factor for
557: substructure or high-density region generation than what would be expected from a
558: 1D-argument. For $c_A/c_s = 1.0$, the fast magnetosonic modes have already reached the boundaries,
559: causing the geometric patterns visible in the density plot.
560:
561: \begin{figure}
562: % \includegraphics[width=\columnwidth]{../figures/densymhd.eps}
563: \includegraphics[width=\columnwidth]{f8c.eps}
564: \caption{\label{f:densy}Logarithmic density maps for models with the field oriented
565: transversally, i.e. perpendicular to the
566: inflow, at $t=3.8$~Myr. {\em Left:} Three resolutions increasing from top to bottom
567: ($N_x=256$ to $1024$ by factors of $2$) at $c_A/c_s=0.5$. {\em Right:} For $c_A/c_s=1.0$.
568: The geometric pattern at the boundaries is an artifact caused by magnetosonic
569: modes reaching the inflow boundaries.}
570: \end{figure}
571:
572: \begin{figure}
573: % \includegraphics[width=\columnwidth]{../figures/emagy.eps}
574: \includegraphics[width=\columnwidth]{f9c.eps}
575: \caption{\label{f:emagy}Logarithmic magnetic energy maps of models with
576: the field oriented transversally, i.e. perpendicular to the
577: inflow, at $t=3.8$~Myr. {\em Left:} Three resolutions increasing from top to bottom
578: ($N_x=256$ to $1024$ by factors of $2$) at $c_A/c_s=0.5$. {\em Right:} For $c_A/c_s=1.0$.}
579: \end{figure}
580:
581: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
582: %\subsubsection{Turbulent Case}
583: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
584: \subsubsection{Turbulent Case\label{sss:turbulent}}
585:
586: In the previous section, we discussed models with a fixed physical dissipative scale.
587: The purpose of this section is to demonstrate that without a fixed dissipation scale,
588: numerical convergence cannot be reached. In other words, for large Reynolds numbers,
589: the system can evolve qualitatively differently.
590:
591: Figure~\ref{f:turb} shows a sequence in resolution of models with zero physical resistivity and
592: viscosity, i.e. models for which the numerical dissipation at resolution scale will set the
593: Reynolds number. Thus, higher resolution will lead to larger Reynolds numbers.
594: For resolution reasons we chose the wave number of the interface perturbation to be
595: $k=1$ in \S\ref{sss:laminar}. Since the condition for fast growth of the instability
596: is given by $k\eta\approx 1$, this required a larger initial amplitude perturbation and an elongated box.
597: Here, we are interested in the (later) turbulent evolution of the slab, thus we start with $k=4$,
598: which allows us to reduce the initial amplitude of the perturbation by the same factor $4$
599: and therefore considerably extends the spatial range in which the slab can develop.
600:
601: At our lowest resolution ($N_x=256$), we essentially
602: get a laminar behavior: the pairwise field reversal regions are stretched along the width
603: of the slab but persist to the end of the simulation (top). At $N_x=1024$,
604: the main magnetic null regions are accompanied by secondary regions as
605: a result of additional shear flows. The slab is thinner. Increasing the resolution
606: further introduces more and more substructure in the slab, especially at
607: the ``heads'' of the slab's perturbations: here, field reversals seem to accumulate,
608: leading to additional field dissipation. Thus, with higher Reynolds numbers,
609: the slab gets more turbulent, and reconnection proceeds not only in the two
610: main magnetic null regions, but all throughout the slab in small regions.
611: Consequently, turbulent field structures inside the slab are dissipated faster, leading
612: to a deficit in magnetic pressure inside the slab, and thus to a thinner slab.
613:
614: \begin{figure}
615: % \includegraphics[width=\columnwidth]{../figures/emagturb2.eps}
616: \includegraphics[width=\columnwidth]{f10c.eps}
617: \caption{\label{f:turb}Logarithmic magnetic energy maps for models with
618: viscosity $\nu\equiv 0$ and resistivity $\lambda_\Omega=0$, i.e.
619: the dissipation scale is given by the grid resolution.
620: Resolution increases from top to bottom
621: ($256^2$ to $2048^2$ by factors of $2$). Since the period in $y$ is
622: repeated four times, we show only one quarter of the domain. $t=3.8$~Myr.}
623: \end{figure}
624:
625: While this effect is certainly interesting to note, the situation might be less
626: extreme in a truly three-dimensional system: a third field component without
627: magnetic null could give rise to sufficient magnetic pressure to prevent
628: reconnection (see also \citealp{2003ApJ...590..291H}). In this case, small-scale
629: fields entangled by the turbulence in the slab could actually lead to
630: additional pressure.
631:
632: The pressure profiles (Fig.~\ref{f:pressprof}) actually tell us a slightly more
633: complicated story than that of simple magnetic energy dissipation. Pressures were
634: averaged transversally (i.e. along the $y$-axis) and plotted against $x$, the inflow
635: direction.
636: The magnetic pressure profile stays pretty much at a constant
637: level, independent of resolution and time. What changes is the kinetic pressure,
638: which drops below its inflow value (all panels but bottom). This effect gets
639: stronger with increasing resolution. Thus, the magnetic field only acts as a
640: dissipation channel for the kinetic energy. One could even interpret the
641: kinetic pressure drop and simultaneous magnetic and internal (red lines)
642: pressure rise as an attempt of the system to achieve equipartition (Fig.~\ref{f:pressequi}).
643:
644: \begin{figure}
645: % \includegraphics[width=\columnwidth]{../figures/pressprof_fldmhdk4m10.eps}
646: \includegraphics[width=\columnwidth]{f11.eps}
647: \caption{\label{f:pressprof}Transversally averaged pressure profiles for
648: four times (in Myr), plotted in ascending time order
649: from bottom to top, against $x$-axis coordinate.
650: Shown are pressure profiles for magnetic models with $\nu\equiv 0$ and
651: $\lomega\equiv 0$ (see text).
652: Gas is streaming in from the left
653: and from the right. See bottom panel for color code and line styles.
654: Shown are the total, the kinetic, the magnetic and the internal pressures.}
655: \end{figure}
656:
657: \begin{figure}
658: % \includegraphics[width=\columnwidth]{../figures/pressequi_turb.eps}
659: \includegraphics[width=\columnwidth]{f12.eps}
660: \caption{\label{f:pressequi}Average pressure in slab against time. Color coding
661: and line style are the same as in Figure~\ref{f:pressprof}.}
662: \end{figure}
663:
664: This is not to say that the various pressure components are in equilibrium,
665: as the left panel of Figure~\ref{f:presscorrel} easily demonstrates.
666: This panel shows the correlation coefficient ${\cal C}$ for the three pairs of pressures,
667: $P_{mag}$, $P_{int}$, and $P_{kin}$, for the case without (left) and with (right)
668: a physical dissipation scale. Balance between the pressure components would show
669: up as an anti-correlation, whereas a correlation can be interpreted as pressures being in phase
670: (and thus driving waves). Decorrelated pressures indicate a mixture.
671: Kinetic and magnetic pressure decorrelate at higher resolution,
672: pointing to strong reconnection events. Internal and magnetic pressure are only slightly
673: correlated (see below), while kinetic and internal pressure anti-correlate at late
674: times, because high-density regions show more inertia. The models with a fixed
675: physical dissipation scale do not show strong resolution effects.
676: Note, however, that the ``dissipation-less'' models are farther in the dynamical
677: evolution than the ``controlled'' models, because of the different initial
678: conditions ($k_y=4$ against $k_y=1$).
679:
680: \begin{figure}
681: % \includegraphics[width=\columnwidth]{../figures/presscorrel.eps}
682: \includegraphics[width=\columnwidth]{f13c.eps}
683: \caption{\label{f:presscorrel}Correlation of pressures within slab against time.
684: Line colors
685: denote correlations, e.g. red stands for ${\cal C}(P_{mag},P_{kin})$.
686: Line styles denote resolution. {\em Left:} Models without fixed
687: physical dissipation scales. {\em Right:} Viscous and resistive scales
688: have been set.}
689: \end{figure}
690:
691: These results demonstrate that only a fixed dissipative scale can guarantee full convergence
692: of the models with resolution. Relying on numerical diffusion leads to flows with increasing
693: Reynolds numbers as the resolution increases and results that depend qualitatively and quantitatively
694: on resolution.
695:
696: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
697: %\subsection{Morphology}
698: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
699: \subsection{Growth Rates\label{ss:growthrates}}
700:
701: Figure~\ref{f:growthrates_fld} summarizes the growth of the slab's amplitude with
702: time. The hydrodynamical growth rates are consistent with the analytical predictions
703: (eq.~[\ref{e:vishniac}], solid straight black line). Saturation sets in when the
704: focal points are shut off from the inflow (see also left column of Fig.~\ref{f:denshdmhd}).
705: The two highest resolution runs have converged also in terms of growth rates (we established
706: detailed convergence in \S\ref{sss:laminar}). Lower resolutions lead to slightly smaller amplitudes
707: initially. The lowest resolution model is already resolved high enough to reproduce the laminar
708: result, but the specified physical viscosity is too small to guarantee convergence
709: with regards to turbulent substructure in the slab. Thus, when increasing to the next
710: higher resolution, the unresolved turbulence leads to less converged growth rates.
711: Note that the growth rate as given
712: in equation~(\ref{e:vishniac}) does not include the effect of turbulence generation
713: within the slab, although the possible effects of turbulence are discussed by \citet{1994ApJ...428..186V}.
714:
715: The dotted line in Figure~\ref{f:growthrates_fld} and all subsequent
716: figures of the same type denotes the growth of the amplitude of an unperturbed shock-bounded
717: slab of width $\Delta(t)$,
718: \begin{equation}
719: \Delta(t) = 2\,c_s\,t\left({{\cal M}/2+(1+({\cal M}/2)^2)^{1/2}}\right)^{-1}\label{e:unpertslab}.
720: \end{equation}
721: Thus, all models except for those with field strength $c_A/c_s=2$ show a faster growth of the slab
722: amplitude than just the shock-bounded expansion. Obviously, equation~\ref{e:crudeest}
723: only gives a rough estimate of the field strength required, and in fact underestimates
724: the effect of the field. Our crude approximation treats the slab as a solid wall,
725: allowing to split the pressure exerted by the inflow on the wall along the normal
726: and tangential directions without any losses. This is most likely unrealistic, i.e. the
727: tangential component will generally be smaller, thus requiring a smaller field
728: to balance it.
729:
730: With increasing field strength, the ordering of the curves with respect to resolution
731: is inverted: now, the lowest-resolution runs show the fastest growth. Nevertheless
732: we get convergence for the two highest resolution simulations. This inversion is a consequence
733: of the slight increase in the Reynolds number with respect to the runs at lower resolution.
734: As discussed above, higher Reynolds numbers lead to more turbulence and more
735: field reversals. Thus, the field is dissipated faster, leading to a pressure deficit
736: in the slab.
737:
738: \begin{figure}
739: % \includegraphics[width=\columnwidth]{../figures/mhdamplitudes_fldmhdk1m40_amp.eps}
740: \includegraphics[width=\columnwidth]{f14.eps}
741: \caption{\label{f:growthrates_fld}Slab amplitude against time for four models
742: at zero field strength $c_A/c_s = 0$ to highest field strength $c_A/c_s=2$.
743: Line styles stand for resolution. The straight solid black line denotes
744: the analytical solution (eq.[\ref{e:vishniac}], \citealp{1994ApJ...428..186V}).
745: The dotted line denotes the
746: expansion of a shock-bounded unperturbed slab (eq.~[\ref{e:unpertslab}]).}
747: \end{figure}
748:
749: A vertical field leads to a strongly reduced growth rate or suppresses the instability
750: completely, depending on its strength (Figure~\ref{f:growthrates_xy}). The NTSI
751: arises, however, for weaker field strengths, and the bending mode in the slab
752: persists, thus allowing material to be funneled towards the focal points.
753: In the presence of self-gravity or a strong thermal instability (e.g. that
754: provided by atomic line cooling), these density enhancements could then
755: fragment and generate further substructure, despite the initially
756: unfavorable field orientation.
757:
758: \begin{figure}
759: % \includegraphics[width=\columnwidth]{../figures/mhdamplitudes_fldmhdk1m40xy_amp.eps}
760: \includegraphics[width=\columnwidth]{f15.eps}
761: \caption{\label{f:growthrates_xy}Slab amplitude against time for models with transverse
762: fields.
763: Thin lines denote the corresponding models with fields oriented along the inflow
764: for comparison. Again, the straight solid black line denotes
765: the analytical solution (eq.~[\ref{e:vishniac}], \citealp{1994ApJ...428..186V}).
766: The dotted line denotes the
767: expansion of a shock-bounded unperturbed slab (eq.~[\ref{e:unpertslab}]).}
768: \end{figure}
769:
770: The amplitude growth corresponding to Figure~\ref{f:turb} is shown in
771: Figure~\ref{f:growthrates_k4}. With higher resolution, saturation sets in earlier in the
772: magnetic runs, and the
773: amplitudes even decrease with time, mirroring the loss of pressure inside the slab
774: due to (resistive) dissipation.
775: Note that the slab amplitude now grows faster than that of an unperturbed
776: shock-bounded slab (see dotted line in Fig.~\ref{f:growthrates_k4}) only at times $t\lesssim 1$~Myr. As
777: Figure~\ref{f:turb} already had made us suspect, saturation sets in earlier for models
778: with higher $k$.
779:
780: \begin{figure}
781: % \includegraphics[width=\columnwidth]{../figures/mhdamplitudes_fldmhdk4m10_amp.eps}
782: \includegraphics[width=\columnwidth]{f16.eps}
783: \caption{\label{f:growthrates_k4}Slab amplitude against time for models with initial perturbation
784: wave number $k=4$ and without fixed resistive scale. Higher resolution leads to
785: saturation of the growth rate at lower amplitudes. The straight solid black line denotes
786: the analytical solution (eq.~[\ref{e:vishniac}], \citealp{1994ApJ...428..186V}).
787: The dotted line denotes the
788: expansion of a shock-bounded unperturbed slab (eq.~[\ref{e:unpertslab}]).}
789: \end{figure}
790:
791: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
792: %\subsection{Field-density Relation}
793: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
794: \subsection{Field-Density Relation\label{ss:fielddensity}}
795:
796: The field-density relation $B(n)$ is often used as an observational
797: measure for the dynamical importance of magnetic fields in the interstellar
798: medium. It describes the mass-loading of field lines, and is related
799: to the mass-to-flux ratio (e.g. \citealp{1976ApJ...210..326M};
800: \citealp{1987ARA&A..25...23S})
801: quantifying the importance of magnetic fields in a gravitationally
802: dominated cloud. Microscopically, the mass loading of field lines
803: can be changed in two ways: by Ohmic diffusion and by ion-neutral (or
804: ambipolar) drift. However, the two parameters controlling the degree
805: to which the interstellar magnetic field is frozen to the gas are
806: huge. The magnetic Reynolds number Re$_M$ is the ratio of the Ohmic
807: diffusion time to the dynamical time, and is typically of order
808: $10^{15}-10^{21}$. The second parameter, the ambipolar Reynolds number
809: Re$_{AD}$, is the ratio of the ion-neutral drift time to the
810: dynamical time. This number is typically many orders of magnitude
811: less than the first one, and can approach unity in dense molecular gas.
812: These numbers
813: suggest that the magnetic field should be nearly perfectly frozen
814: to the plasma component of the gas, and generally well
815: frozen to the neutrals, except in the densest, nearly absolutely neutral regions.
816: Thus, the $B(n)$ relation is determined primarily by dynamical
817: rather than by microscopic processes. Parameterizing the relation
818: by $q\equiv d\ln B/d \ln n$, one finds $q=1$ for compression
819: perpendicular to $\mbfB$, $q=2/3$ for isotropic compression,
820: $q=1/2$ for self-gravitating, magnetically sub-critical clouds
821: \citep{1993ApJ...415..680F}, and $q=0$ for compression parallel to $\mbfB$.
822:
823: Observations of the $B(n)$ relation in molecular gas indeed show
824: that the strongest fields are associated with the densest gas
825: (e.g. \citealp{1999ApJ...520..706C}). However, in the more diffuse ISM,
826: the $B(n)$ relation is consistent with $q\approx 0$ over
827: 3 orders of magnitude in $n$ (\citealp{1986ApJ...301..339T}, \citealp{2005ApJ...624..773H}). Thus, processes
828: beyond microscopic diffusion are required to decouple field and density.
829: The possibility of accelerating the decoupling through turbulent transport
830: has been explored by \citet{2002ApJ...567..962Z} and \citet{2004ApJ...603..165H} (see also
831: \citealp{2002ApJ...578L.113K}, \citealp{2002ApJ...570..210F}, and
832: \citealp{2004ApJ...609L..83L}). Numerical evidence for a weak $B(n)$
833: relation includes \citet{1999ApJ...526..279P} and \citet{2005A&A...436..585D}.
834:
835: Figure~\ref{f:fielddensity} shows scatter plots of $\log B$ against
836: $\log n$ for four models, each measured at $t=4$~Myr. The top row
837: shows models with the field parallel to the inflow (denoted by
838: $c_{Ax}$ in the label), while the field is oriented transversally,
839: or perpendicularly to the inflow in the bottom row (denoted by
840: $c_{Ay}$). Idealized scalings are indicated by the dashed lines.
841:
842: \begin{figure*}
843: % \includegraphics[width=\textwidth]{../figures/allbrho.eps}
844: \includegraphics[width=\textwidth]{f17c.eps}
845: \caption{\label{f:fielddensity}Scatter plots of
846: magnetic field strength $\log B$ against density $\log n$ for
847: models with parameters indicated in panels. Dashed lines denote idealized
848: $B(n)$-scalings.}
849: \end{figure*}
850:
851: The most striking difference between the top and the bottom row is that
852: for the configuration where the field is parallel to the inflow, field and density seem to
853: correlate more strongly than for the configuration where the field is perpendicular to the inflow.
854: This might seem surprising at first. After all, one would expect the field
855: to be correlated least with density if the gas is compressed along the
856: field lines (top row), and to be correlated most with density when the
857: gas is compressed perpendicularly to the field lines (bottom row).
858: However, looking back at Figures~\ref{f:emag} and \ref{f:emagy} on the one hand,
859: and Figure~\ref{f:turb} on the other, we realize that the models with fields
860: parallel to the inflow generally develop turbulence, leading to field line
861: stretching. Thus, in the (denser) slab, the field generally will be stronger.
862: Why, then, is there close to no correlation observable in the bottom row
863: of Figure~\ref{f:fielddensity}? The answer is hidden in the way we set up
864: the initial conditions. To avoid the generation of strong MHD waves, we
865: kept the field uniform, despite the fact that the flow collision interface
866: is strongly perturbed (see top row of Fig.~\ref{f:denshdmhd} for the initial
867: conditions). Thus, when gas is deflected at the flanks (point ``0'' in
868: Figure~\ref{f:sketch}), it is essentially free to move along the field lines,
869: i.e. transversally, but is increasingly prevented from continuing its trip towards
870: the slab, because the magnetic pressure is increasing (note that the bulk magnetic
871: field strength is higher in the bottom row of Fig.~\ref{f:fielddensity}). Thus,
872: the density can take on any values in the slab, while the field value is given
873: by the ram pressure. A weaker field (lower right panel vs lower left panel
874: of Figure~\ref{f:fielddensity}) reduces this effect, allowing some scatter
875: in the field strength, and, indeed, checking Figure~\ref{f:emagy} for this case,
876: the slab actually shows substructure and is allowed to bend. Note that there is
877: a strict $d \ln B/d\ln n=1$ scaling for the strong field model (lower left panel),
878: which results from the initial compression of the field lines.
879:
880: While it is hard to generalize the results of our two-dimensional models, they
881: demonstrate that the $B(n)$ relation is strongly influenced by the geometry of
882: the fields and the gas flows. Even compression perpendicular to the field lines
883: can lead to a nearly complete decoupling of field and density -- as long as the
884: gas flow is given the chance to break the symmetry.
885:
886: In a sense, we expect the ``geometrical'' mechanism decoupling the field from the density
887: (lower row of Figure~\ref{f:fielddensity}) to compete with the turbulent transport
888: of the magnetic field: both act on dynamical (flow) timescales. For fields oriented
889: perpendicularly to the inflow, our models do not develop any substantial turbulence
890: (this might also be due to a too small Reynolds number). On the other hand, the models with
891: field parallel to the inflow direction do generate some turbulence.
892: For these models, $B$ and $n$ decorrelate only at higher density values, corresponding
893: to small scales, while the lower density regions show a reasonably
894: well established correlation between $B$ and $n$.
895:
896:
897: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
898: %
899: %\section{Summary}
900: %
901: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
902: \section{Summary\label{s:summary}}
903:
904: The Non-linear Thin Shell Instability (NTSI, \citealp{1994ApJ...428..186V})
905: is expected to occur in expanding shells, shocks or colliding gas streams.
906: Previous studies have addressed the evolution of the NTSI under hydrodynamical
907: conditions, including gravity and cooling. We have presented
908: a numerical study of the NTSI including magnetic fields. We have established that our numerical
909: method is well suited to tackle the problem. We have found that the effects of magnetic fields on the NTSI
910: can be summarized as follows:
911:
912: (1) Fields principally tend to weaken or even suppress the NTSI. We further distinguish
913: between two cases: (i) fields aligned with the inflow resist the transverse momentum
914: transport -- which is the main driving agent of the NTSI -- via the magnetic
915: tension force; (ii) fields perpendicular to the inflow lead to a stiffer equation
916: of state. If $c_A\approx u$, the NTSI is suppressed. However, even for
917: transverse fields, substructures can form within the slab, which can serve as
918: fragmentation seeds in the presence of thermal instabilities or self-gravity.
919:
920: (2) A fixed physical scale both for viscous and resistive dissipation is necessary
921: to reach numerical convergence. When relying on numerical dissipation at the resolution
922: scale, the Reynolds number will increase with resolution, leading to a more turbulent
923: environment and thus to results which qualitatively and quantitatively depend on resolution
924: (Figs~\ref{f:emag}, \ref{f:turb} and \ref{f:growthrates_k4}).
925:
926: (3) At larger Reynolds numbers, turbulent reconnection plays a role in the turbulent
927: dense slab generated by the NTSI. Magnetic energy is therefore dissipated at higher rates,
928: leading to a pressure deficit in the dense slab. The magnetic field acts as a dissipation
929: channel (Figs~\ref{f:pressprof} and \ref{f:pressequi}).
930:
931: (4) Although the energies (or average pressures) seem to show a tendency of the system
932: to evolve towards equipartition, pressures do not balance locally within the
933: slab (Figs.~\ref{f:pressequi} and \ref{f:presscorrel}). Correlated pressures
934: lead to waves, i.e. the slab's inner structure is highly dynamical.
935:
936: (5) The relation between field and density is, at best, weak in all models
937: (Fig.~\ref{f:fielddensity}). Models with fields parallel to
938: the inflow exhibit a stronger $B(n)$ correlation than models with fields
939: oriented perpendicularly to the inflow. The main reason for this is the
940: generation of turbulence, which leads to field line stretching and thus
941: field amplification within the denser slab. Fields oriented perpendicularly
942: to the inflow allow instreaming material to move laterally, permitting the
943: field and density to decorrelate.
944:
945: Our isothermal models only allow a limited exploration of the effect of fields
946: on colliding flows in a thermally or gravitationally unstable medium. Clearly,
947: substructure can form in the slab under most conditions, providing potential
948: seeds for thermal or gravitational instabilities. Thus, to establish
949: the role of magnetic fields for molecular cloud formation in the colliding flow
950: scenario, the thermal and gravitational effects have to be addressed.
951:
952:
953: \acknowledgements
954: We thank E.~Zweibel for a critical reading of the manuscript and for enlightening
955: discussions, and E.~Vishniac for comments on magnetic field effects in the NTSI.
956: %and we thank the anonymous referee for a critical and constructive report.
957: Computations were performed at the NCSA (AST040026) and on the local resources
958: at U of M, perfectly administered and maintained by J.~Hallum.
959: This work was supported by the University of Michigan
960: and has made use of the NASA Astrophysics Data System.
961:
962:
963: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
964: %
965: %\references
966: %
967: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
968:
969: \bibliographystyle{apj}
970: \bibliography{./references}
971:
972: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
973: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
974: \end{document}
975: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
976: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
977:
978:
979:
980:
981:
982:
983: