1: %Revisions to ms downloaded from website following referee's report.
2: %Ellen's revised text set off by "%%egz
3: %ver 05 (F -> E)
4: % * questions, comments in bold face headed by ``fh:''
5: % * v_{th} -> c_s (``thermal speed'' i assume is ``sound speed'')
6: %
7: % * had to rename the models, bc the labels didn't make any sense.
8: % see table 1. The new DA is the old LF, and the old DA is now DR. Sorry ...
9: %
10: %
11: %\documentclass[preprint2]{aastex}
12: \documentclass[12pt,preprint]{aastex}
13: %\documentclass{emulateapj}
14:
15: %\usepackage{amssymb}
16: %\usepackage{amsmath}
17: %\usepackage{graphicx}
18:
19: \newcommand{\kapGP}{\kappa_f}
20: \newcommand{\kapP}{\kappa_P}
21: \newcommand{\kGP} {k_f}
22: \newcommand{\vGP} {v_f}
23: \newcommand{\kP} {k_P}
24: \newcommand{\lAD} {\lambda_{AD}}
25: \newcommand{\lO} {\lambda_{\Omega}}
26: \newcommand{\led} {\lambda_{e}}
27: \newcommand{\lnum} {\lambda_{num}}
28: \newcommand{\tAD} {\tau_{AD}}
29: \newcommand{\tO} {\tau_{\Omega}}
30: \newcommand{\tGP} {\tau_f}
31: \newcommand{\cAi} {c_{Ai}}
32: \newcommand{\cAn} {c_{An}}
33: \newcommand{\RM} {R_{M}}
34: \newcommand{\RAD} {R_{AD}}
35: \newcommand{\mbfB}{\mathbf{B}}
36: \newcommand{\mbfE}{\mathbf{E}}
37: \newcommand{\mbfj}{\mathbf{j}}
38: \newcommand{\mbfu}{\mathbf{u}}
39: \newcommand{\mbfui}{\mathbf{u}_i}
40: \newcommand{\mbfun}{\mathbf{u}_n}
41: \newcommand{\mbfuD}{\mathbf{u}_D}
42: \newcommand{\mbfuGP}{\mathbf{u}_{GP}}
43: \newcommand{\mbfv}{\mathbf{v}}
44: \newcommand{\mbfz}{\mathbf{z}}
45: \newcommand{\mbfn}{\mathbf{n}}
46: \newcommand{\mbfk}{\mathbf{k}}
47: \newcommand{\mbfx}{\mathbf{x}}
48: \newcommand{\mbfy}{\mathbf{y}}
49: \newcommand{\nuin}{\nu_{in}}
50: \newcommand{\mbfnabla}{\mathbf{\nabla}}
51: \newcommand{\f} {\frac}
52: \newcommand{\ddx}{\frac{\partial}{\partial x}}
53: \newcommand{\ddy}{\frac{\partial}{\partial y}}
54: \newcommand{\ddt}{\frac{\partial}{\partial t}}
55: \newcommand{\mq}{\langle q\rangle}
56: \newcommand{\dq}{\delta q}
57: \begin{document}
58:
59: \title{Fast Dynamos in Weakly Ionized Gases}
60:
61: \author{Ellen G. Zweibel\altaffilmark{1,2}}
62: \author{Fabian Heitsch\altaffilmark{3}}
63: \altaffiltext{1}{Depts. of Astronomy \& Physics, U. Wisconsin-Madison, 475 N Charter St, Madison,
64: WI 53706, U.S.A.}
65: \altaffiltext{2}{Center for Magnetic Self-Organization in Laboratory \& Astrophysical Plasmas}
66: \altaffiltext{3}{Dept. of Astronomy, U. Michigan, 500 Church St, Ann Arbor, MI 48109-1042, U.S.A.}
67:
68: \lefthead{Zweibel \& Heitsch}
69: \righthead{Fast Dynamos}
70:
71: \begin{abstract}
72: The turnover of interstellar gas on $\sim 10^9$yr timescales argues for the
73: continuous operation of a galactic dynamo. The conductivity of interstellar
74: gas is so high that the dynamo must be ``fast" - i.e. the magnetic field must
75: be amplified at a rate nearly independent of the magnetic diffusivity. Yet,
76: all the fast dynamos so far known - and all direct numerical simulations of
77: interstellar dynamos - yield magnetic power spectra that peak at the resistive
78: scale, while galactic magnetic fields have substantial power on large scales.
79: In this paper we show that in weakly ionized gas the limiting scale may be the
80: ion-neutral decoupling scale, which although still small is many orders of magnitude
81: larger than the resistive scale.
82: \end{abstract}
83: \keywords{dynamos--- MHD --- methods:numerical --- ISM:magnetic fields}
84:
85: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
86: \section{Introduction}\label{s:introduction}
87:
88: Despite many theoretical and observational advances, our understanding of galactic magnetic fields
89: is still incomplete. Although there is evidence for magnetic fields in young galaxies (Kulsrud \& Zweibel 2008), it is likely that
90: dynamo processes still operate
91: continuously in galaxies today. Perhaps the most compelling argument for ongoing
92: dynamo activity is that the turnover time of interstellar gas due to loss and replacement
93: of material - $\sim$ 10$^9$ yr in the case of the Milky Way - is
94: much less than the ages of galaxies. As interstellar gas is added - whether by
95: infall from the IGM or through stellar mass loss - its magnetic field must be brought to the galactic value
96: so as to maintain the
97: field in a steady state.
98:
99: In contrast to the primordial situation at early times, when a magnetic field had to be
100: built up from nearly zero, it is likely that all the material
101: on which galactic dynamos now act is at least somewhat magnetized. Under these conditions, the
102: primary functions of the dynamo are to increase magnetic energy to the observed value, to
103: generate and maintain a
104: component of the field which is coherent over at least several kpc, as observed, and to transport magnetic flux into
105: the gas which has newly arrived, which especially if it is intergalactic in origin or was shed by stars, may be
106: under-magnetized. These processes are also necessary in theories such as that of Rees (1987), in which galactic magnetic fields
107: are seeded by random fields injected by many small scale sources.
108:
109: Dynamos are considered ``fast" if the rate at which the field is amplified tends
110: to a finite limit as the magnetic diffusivity $\eta$ approaches zero
111: (Childress \& Gilbert 1995). It may seem paradoxical that magnetic diffusivity,
112: which removes energy from the magnetic field, is required for dynamos at all. But
113: without diffusion the magnetic topology is fixed, and this places strong
114: constraints on magnetic field amplification (Moffat 1978). The Ohmic diffusion
115: time in galaxies is so much longer than any dynamical time that it seems galactic
116: dynamos must be fast.
117:
118: Fast dynamo theory has followed two approaches. One involves computation and
119: analysis of the action of prescribed flows on magnetic fields
120: under nearly non-diffusive conditions (e.g. Galloway \& Proctor 1992, Ott 1998, Gilbert 2002, Couvoisier, Hughes, \&
121: Tobias 2006). These so-called kinematic studies
122: have demonstrated the role of chaotic flow in magnetic field amplification and
123: elucidated important relationships between the topological properties of the
124: flows and their dynamo properties (Klapper \& Young 1995). The modification of some of these flows by
125: magnetic forces has also been considered, and has provided some insight into
126: how dynamos might saturate (Cattaneo, Hughes, \& Kim 1996,
127: Tanner \& Hughes 2003, Cameron \& Galloway 2005).
128:
129: The second approach to fast dynamos involves analysis and direct numerical simulations of driven
130: magnetohydrodynamic (MHD) turbulence, again with the lowest magnetic diffusivity
131: $\lambda$ possible. In particular, $\lambda$ is chosen to be less than the viscous
132: diffusivity $\nu$ (i.e. the large Prandtl number case). These models are
133: self-consistent in the sense that magnetic forces are fully included in the
134: dynamics, and theory and simulation are in good agreement on how the field evolves (Schekochihin et al. 2004).
135:
136: Both approaches yield magnetic fields which are
137: dominated by structure on the resistive scale, which is far smaller than any other
138: characteristic scale in the interstellar medium. This follows from the
139: fact that in the absence of magnetic diffusion, any divergence free flow which
140: amplifies the field lengthens the field lines in the same proportion. Therefore,
141: as the field is amplified it becomes folded or tangled. The large random field is in stark contrast
142: to the observed structure of galactic magnetic fields,
143: which display considerable long range order, and thus presents a serious challenge
144: to dynamo theory. In particular,
145: the magnetic field is amplified on scales far below
146: the minimum velocity scale,
147: whether this scale is prescribed (the kinematic case)
148: or created by viscous effects (the full turbulent simulation).
149:
150: Most of the mass, and a considerable part of the volume of the interstellar
151: medium, is weakly ionized, with the plasma and neutral fluids coupled by
152: collisions and by ionization and recombination. On long lengthscales
153: and timescales the medium behaves like a single conducting fluid, but on short
154: lengthscales and timescales, the two fluids decouple. For present interstellar
155: medium parameters, the magnetic field is in approximate energy equipartition with the
156: bulk fluid and dominates the plasma component. The question therefore arises whether
157: a dynamo in a weakly ionized medium could be quenched on scales below the
158: plasma-neutral decoupling scale while continuing to amplify the field on larger
159: scales. Based on analysis and computations presented in this paper, the answer
160: seems to be that it can.
161:
162: %%egz
163: This is far from the first study of the effects of partial ionization on dynamos. There have been a number of studies of large
164: scale dynamos in weakly ionized gases which show that the nonlinearity introduced by ion-neutral drifts has important
165: consequences for mean field theory (Zweibel 1988, Proctor \& Zweibel 1992) and for the general problem of how large scale fields
166: are amplified in a medium with small scale turbulence (Kulsrud \&
167: Anderson 1992, Subramanian 1997, 1999, Brandenburg \& Subramanian 2000). These
168: studies all used the strong coupling approximation (Shu 1983), according to which the plasma drift relative to the
169: neutrals is found by balancing the Lorentz force by ion-neutral drag. This approximation breaks down at small scales.
170: In this work, we treat
171: the plasma and neutrals as two frictionally coupled fluids, which is more appropriate at the small scales, and also more general.
172:
173: In order to see how ion-neutral friction might affect the dynamo, consider the equation for magnetic energy, which in a domain $V$
174: with periodic or infinitely distant boundaries can be written as
175: \begin{equation}\label{e:adenergy}
176: \f{\partial}{\partial t}\int_V\f{B^2}{8\pi}d^3x = -\int_V\left[\mbfv_i\cdot\f{(\mbfnabla\times\mbfB)\times\mbfB}{4\pi} +
177: \lambda\f{\vert\mbfnabla\times\mbfB\vert^2}{4\pi}\right]d^3x,
178: \end{equation}
179: where $\mbfv_i$ is the plasma velocity and $\lambda$ is the Ohmic diffusivity.
180: The first term in the integral on the right hand side represents the work down by the field on the plasma; the second term is
181: resistive dissipation.
182:
183: The plasma velocity can always be expressed in terms of the neutral velocity $\mbfv_n$ and the drift
184: velocity $\mbfv_D\equiv \mbfv_i-\mbfv_n$; $\mbfv_i=\mbfv_v+\mbfv_D$. If the ionization fraction is low, $\mbfv_n$, in turn, is
185: very nearly the center of mass velocity $\mbfv$. In the strong coupling approximation, $\mbfv_D$ is
186: \begin{equation}\label{e:sca}
187: \mbfv_D=\f{(\mbfnabla\times\mbfB)\times\mbfB}{4\pi\gamma\rho_i\rho_n},
188: \end{equation}
189: where $\gamma\equiv\left<\sigma v\right>_{in}/(m_i+m_n)$
190: is the ion-neutral friction coefficient.
191: Replacing $\mbfv_i$ by $\mbfv+\mbfv_D$ in equation~(\ref{e:adenergy}) and using equation~(\ref{e:sca}), we see that when the strong
192: coupling approximation holds, the magnetic energy equation is
193: \begin{equation}\label{e:adenergy2}
194: \f{\partial}{\partial t}\int_V\f{B^2}{8\pi}d^3x = -\int_V\left[\mbfv\cdot\f{(\mbfnabla\times\mbfB)\times\mbfB}{4\pi} +
195: \rho_i\rho_n\gamma v_D^2 +\lambda\f{\vert\mbfnabla\times\mbfB\vert^2}{4\pi}\right]d^3x.
196: \end{equation}
197: Equation (\ref{e:adenergy2}) shows, as expected, that ion-neutral friction is a dissipative effect. The work term, however, can
198: have either sign, and it is possible that in the presence of ambipolar drift, the magnetic field and flow are modified so as
199: to increase the rate at which energy flows to the field. Equation (\ref{e:adenergy}), on the
200: other hand, is valid whether the strong coupling
201: approximation holds or not, is closer to the magnetic energy equation for a plasma,
202: and does not attempt to split the inductive and dissipative effects.
203: %%egz
204:
205: In \S{\ref{s:setup}} we introduce the basic setup. In \S{\ref{s:results}} we show
206: how plasma-neutral friction can allow the field to saturate at the small scales
207: while still being amplified at the large scales. Section {\ref{s:summary}} is a summary.
208:
209: \section{Formulation}\label{s:setup}
210: \subsection{Important timescales and lengthscales}
211:
212: In a turbulent cascade, the velocity at scale $l$, $v_l$, typically decreases with decreasing $l$, but slowly enough that the
213: turnover time $\tau_l\equiv l/v_l$, decreases with $l$ as well. For example, in Kolmogorov turbulence, or
214: in the magnetohydrodynamic
215: turbulence model of Sridhar \& Goldreich (1994) and
216: Goldreich \& Sridhar (1995, 1997), $v_l$ is related to $l$ and the scale $l_d$ at which the turbulence is
217: driven by $v_l\sim v_d(l/l_d)^{1/3}$. The turnover time $\tau_l$ is then $(l_d/v_d)(l/l_d)^{2/3}$. The cascade terminates at the scale
218: $l_K$ at which the dissipation rate becomes faster than the turnover rate.
219:
220: In diffuse, weakly ionized interstellar gas, $\tau_K$, the turnover time at $l_K$, is comparable to the neutral-ion collision time
221: $\tau_{ni}$, and much longer than the ion-neutral collision time $\tau_{in}$. Only for motions on timescales less than $\tau_{in}$ is
222: friction with the neutrals unimportant for plasma dynamics. Resistive effects are typically unimportant for motions with turnover times
223: $\tau_{in}$, but become important on much smaller scales. The ion viscous scale is probably terminates the ion flow on scales much larger
224: than the resistive scale (the suppression of viscous stresses perpendicular to $\mbfB$ even for a weak magnetic field makes the actual
225: value difficult to assess),
226: and the
227:
228: The computational resources available to us preclude modeling all these scales in a realistic
229: manner. Our main focus is on separating the
230: ion-neutral decoupling scale and the resistive scale. As a consequence, instead of
231: assigning the neutral motions a third, still
232: larger scale, we allow the neutral flow to take place on
233: the ion-neutral decoupling scale. Furthermore, we rely entirely on numerical viscosity, which allows the
234: flow in the ions to extend below the resistive scale. In reality, ion viscosity probably terminates the ion flow on scales much larger
235: than the resistive scale (the suppression of viscous stresses perpendicular to $\mbfB$ even for a weak magnetic field makes the actual
236: value difficult to assess), but again we are unable to achieve true scale separation (or implement anisotropic viscosity).
237:
238: The consequences of these
239: aspects of our calculations are discussed in the
240: next two sections.
241:
242: \subsection{Case study}
243: We assume that the neutral velocity $\mbfv_n$ is of a form which would give fast dynamo action
244: if it were the plasma velocity $\mbfv_i$. On timescales much longer than the ion-neutral collision time $(\rho_n\gamma_{in})^{-1}$,
245: the plasma should move with the neutrals - $\mbfv_i\sim\mbfv_n$ - and the magnetic field should
246: grow at the fast dynamo rate, at least while Lorentz forces are unimportant. On short lengthscales,
247: inertial and Lorentz forces compete with ion-neutral friction to drive $\mbfv_i$ away from
248: $\mbfv_n$. We investigate the dynamo properties of the resulting plasma flow. The setup is somewhat similar to the two-fluid study of
249: Kim (1997), although that work focused on diffusion of a large scale field, and the computations were two dimensional and incompressible.
250:
251: We simplify the problem by assuming the ionization and recombination timescales are much
252: shorter than the advective timescales, so that ionization equilibrium holds. This assumption
253: %%egz
254: allows us to bypass the plasma continuity equation, but it
255: is unrealistic in the diffuse gas
256: which is the primary application for this study [although it is much better for molecular gas (see Table 3 of Heitsch \& Zweibel
257: 2003a)].
258: %%egz
259: The neutral velocity we choose is divergence free, so by taking the neutral
260: density $\rho_n$ to be initially uniform, we force it, and the plasma density $\rho_p\sim\rho_i$,
261: to remain so. Therefore, there are no thermal pressure gradient forces in the problem. Under these
262: conditions, the momentum equation for the plasma is
263: \begin{equation}\label{e:ionm1}
264: \rho_i\left(\frac{\partial}{\partial t}+\mbfv_i\cdot\mbfnabla\right)\mbfv_i=
265: \f{\left(\mbfnabla\times\mbfB\right)\times\mbfB}{4\pi}
266: -\gamma\rho_i\rho_n\left(\mbfv_i-\mbfv_{n}\right),
267: \end{equation}
268: where
269: $\mbfv_n$ is to be prescribed.
270: Equation~(\ref{e:ionm1}) is to be solved together with the magnetic induction equation
271: \begin{equation}\label{e:induction}
272: \f{\partial\mbfB}{\partial t}=\mbfnabla\times(\mbfv_i\times\mbfB)+\lambda\nabla^2\mbfB.
273: \end{equation}
274:
275: For $\mbfv_n$ we choose the 2.5D
276: flow shown by Galloway \& Proctor (1992) to be a fast dynamo. This flow, which we will
277: refer to as $\mbfv_{GP}$, can be written as
278: \begin{equation}\label{e:gpflow}
279: \mbfv_n=\hat{x}2\pi k_{GP}\psi+\hat{y}\frac{\partial\psi}{\partial z}-\hat{z}\frac{\partial\psi}{\partial y},
280: \end{equation}
281: where the stream function $\psi$ is
282: \begin{eqnarray}\label{e:psi}
283: \psi =\sqrt{\frac{3}{2}}\frac{V_0}{2\pi k_{GP}}\Big(& &\sin{[2\pi( k_{GP}z+\epsilon\sin{2\pi\omega t})]}\nonumber\\
284: &+&\cos{[2\pi(k_{GP}y+\epsilon\cos{2\pi\omega t})]}\Big),
285: \end{eqnarray}
286: and $\epsilon$ is a constant. This flow has periodic, cellular structure on a single spatial scale,
287: $k^{-1}$, and temporal structure at frequency
288: $\omega$ and all of its harmonics. This can readily be seen by rewriting equation~(\ref{e:psi}) in the form
289: \begin{eqnarray}\label{e:psiseries}
290: \psi =\sum_{n=0}^{\infty}\psi_n
291: \Big(& &\sin{[2\pi (k_{GP}z+n\omega t)]} \nonumber\\
292: &+&\cos{[2\pi (k_{GP}y+n(\omega t + \frac{1}{4})]}\Big),
293: \end{eqnarray}
294: where
295: \begin{equation}\label{e:psin}
296: \psi_n\equiv \sqrt{\frac{3}{2}}\frac{V_0}{2\pi k_{GP}}J_n(2\pi\epsilon)
297: \end{equation}
298: and the $J_n$ are ordinary Bessell functions of the first kind. However, bearing in mind that
299: $J_n(x)\sim (x/2)^n/n!$ for $x<n$, and that $x=\pi/2$ in our problem, we see that only the first
300: few harmonics of the series are important in the flow. The cell pattern drifts back and
301: forth with temporal frequency $\omega$ and amplitude $\epsilon$. The flow has chaotic regions
302: along the cell edges. In this paper, we choose units of length and time such that
303: $k_{GP}= V_0 = \omega \equiv 1$.
304: The eddy turnover time and the period over
305: which the pattern oscillates are then of the same order. We set $\epsilon\equiv 0.25$, near the value
306: at which chaos is maximized (Brummell, Cattaneo, \& Tobias (2001)). The ranges of $y$ and $z$ are
307: $-1\le y\le 1$, $-1\le z\le 1$, so only one GP cell fits into the domain. We choose $\lambda$
308: such that $\lambda k_{GP}^2$, the resistive decay rate at the GP scale, is between $10^{-2}$ and
309: $10^{-4}$, and the friction coefficient, $\gamma_{in}\rho_n$, is between $0.03$ and $0.60$. We set
310: $\rho_i=1$. The magnetic field is initialized by choosing a vector potential which is a Fourier
311: series in modes of the computational domain with random amplitudes and random
312: phases selected from uniform distributions.
313:
314: The numerical scheme, Proteus, is based on the conservative gas-kinetic flux splitting method
315: introduced by Xu (1999) and Tang \& Xu (2000), and is second order in time and space.
316: Resistivity is included through dissipative fluxes (Heitsch et al. 2004, 2007), but
317: viscosity is not (there is a small amount of numerical viscosity). The frictional force
318: is implemented through a drag term (Heitsch et al. 2004).
319:
320: \section{Results}\label{s:results}
321:
322: The complete set of models we considered is summarized in Table 1.
323: We first verified that our computational scheme recovers the previously known
324: dynamo behavior of the GP flow, and assessed numerical convergence.
325: The left panel of Figure~\ref{f:allki} shows the
326: magnetic energy against time for three resolutions and five diffusivities.
327: Once the short wavelength magnetic power present in the random
328: initial conditions has
329: decayed, the magnetic energy grows at an exponential rate. The growth
330: rates are plotted in the right panel of Figure~\ref{f:allki}. With our choice of units, $R_m=\lambda^{-1}$.
331: While some flattening of the curves is apparent, the growth rate has evidently not
332: converged to the $R_m\rightarrow\infty$ limit. Nevertheless, the evident lack of convergence
333: at the largest $R_m$ and highest numerical resolution practical for us shows that it would not
334: be meaningful to set $\lambda$ to any value smaller than what we used here. This series of models
335: will be referred to as KI (for model keys, see Table~\ref{t:models}). All the models discussed subsequently have $R_m=10^{4}$.
336:
337: \clearpage
338: \begin{figure*}
339: \includegraphics[width=\textwidth]{./f1.eps}
340: \caption{\label{f:allki}{\em Left:} Natural logarithm of the normalized magnetic energy against
341: time for all kinematic models (series KI), at linear resolutions of $N=64$ (thin lines),
342: $128$ (medium lines), and $256$ (thick lines). The higher $\lambda$, the stronger the
343: initial decay of the (noise-like) seed field. {\em Right:} (Exponential) growth rates
344: derived from fits to the curves in the left panel, for all resolutions and magnetic Reynolds numbers.}
345: \end{figure*}
346: \clearpage
347:
348: \begin{deluxetable}{l|cc}
349: \tablewidth{0pt}
350: \tablecaption{Key to model names\label{t:models}}
351: \tablehead{\colhead{Name}&\colhead{Mnemonic}
352: &\colhead{Physics}}
353: \startdata
354: KI & Kinematic, Ions & single-fluid, GP flow in ions\\
355: KN & Kinematic, Neutrals & two-fluid, GP flow in neutrals\\
356: DL & Dynamic, Lorentz & two-fluid, GP flow in neutrals, Lorentz force\\
357: DR & Dynamic, Reynolds & two-fluid, GP flow in neutrals, Reynolds Stress\\
358: DA & Dynamic, All & DL combined with DR
359: \enddata
360: \end{deluxetable}
361: \clearpage
362:
363: We next set $\mbfv_n=\mbfv_{GP}$ and solved equation~(\ref{e:ionm1}) with the Lorentz force and
364: Reynolds stress terms dropped. These models, which we denote by KN, demonstrate the interplay
365: between inertia and friction in determining the ion flow. Using equation~(\ref{e:psiseries}),
366: equation~(\ref{e:ionm1}) can then be reduced to a series of equations for the $j$th Fourier harmonic of $\mbfv_i$
367: \begin{eqnarray}\label{e:vidot}
368: \left(\f{\partial}{\partial t}+\rho_n\gamma_{in}\right)\mbfv_{ij}
369: &=&\rho_n\gamma_{in}\psi_j\left(\hat\mbfx+\hat\mbfy\f{\partial}{\partial z}-
370: \hat\mbfz\f{\partial}{\partial y}\right)\nonumber \\
371: &\times& \Big(\sin{[2\pi( k_{GP}z+j\omega t)]}\nonumber\\
372: &+&\cos{[2\pi(k_{GP}y+j[\omega t+\frac{1}{4}])]}\Big),
373: \end{eqnarray}
374: which can be solved analytically. For the
375: parameters chosen here, $\rho_n\gamma_{in}\ll j\omega$ for $j>0$, and $\mbfv_{ij}$ is nearly $\pi/2$ out of phase with
376: the frictional forcing term, and reduced in amplitude by a factor of
377: $\rho_n\gamma_{in}/j\omega$. Thus, inertial effects force the plasma
378: away from the GP flow (this is an artifact of the overlap in our models between the turnover rate of the neutral eddies and the ion-neutral
379: collision rate; in the interstellar medium the latter is much larger).
380: However, even though the plasma flow departs significantly from the GP flow, it does amplify
381: the magnetic field. This is shown in Figure~\ref{f:alltimes}.
382: The amplification rate decreases with decreasing $\rho_n\gamma_{in}$. This is caused
383: primarily by the reduced amplitude of the ion flow, but also by its structure, which as $\rho_n\gamma_{in}$ decreases departs more and
384: more from $\mbfv_{GP}$.
385: %%egz
386: The dissipative nature of ion-neutral friction may also be playing a role here; see equation~(\ref{e:adenergy2}) - although equation~(\ref{e:sca}) is not completely accurate in this case.
387: %%egz
388:
389: It can be seen from equation
390: (\ref{e:induction}) that magnetic fluctuations at wave number $k$ interacting
391: with flow at the GP wave number $k_{GP}$ are coupled to fields
392: at wave number $k\pm k_{GP}$. This sets off a magnetic cascade in wavenumber, and therefore, both
393: the GP flow and the modified flow present in the KI models drive magnetic field at all spatial scales.
394: \clearpage
395: \begin{figure*}
396: \includegraphics[width=\textwidth]{./f2.eps}
397: \caption{\label{f:alltimes}Natural logarithm of the normalized magnetic energy against
398: time for models at linear resolution $N=128$. See Table~\ref{t:models} for
399: keys to the model names. Back-reaction via Lorentz-forces let the field amplification
400: saturate (models KN vs DL), while the inclusion of the Reynolds stress suppresses the
401: dynamo (models KN vs DR and DA). Note that there is no difference in the energy evolution
402: between models DR and DA because the magnetic field simply stays too weak for the
403: Lorentz force to be of importance. The dashed horizontal line denotes energy equipartition
404: between magnetic energy and the nominal kinetic energy in the GP flow.}
405: \end{figure*}
406: \clearpage
407:
408: %%egz
409: The Reynolds stress $\mbfv_i\cdot\mbfnabla\mbfv_i$,
410: because of its nonlinearity, drives higher spatial harmonics of the GP wave number. In our parameter regime, it is of
411: the same order as the inertial term, at least if $v_i$ is of the same magnitude as $v_{GP}$.
412: Because in our calculation the resistivity exceeds
413: the (numerical) viscosity, the ion flow extends to smaller scales than the magnetic field.
414: It is known that dynamo activity is often
415: suppressed in these so-called low magnetic Prandtl number situations unless the
416: magnetic Reynolds number is large (Boldyrev
417: \& Cattaneo 2004, Schekochihin et al 2007). Because the ion density remains constant, the compressibility of the medium
418: is exaggerated; suppression of the dynamo has also been associated with compressibility effects (Haugen, Brandenburg, \& Mee,
419: 2004). For both these reasons, it is perhaps unsurprising that including the Reynolds stress obliterates the dynamo; see
420: the DR models in Figure~\ref{f:alltimes}.
421: %%egz
422:
423:
424: We also considered the effect of dropping the Reynolds stress term but including the Lorentz
425: force (models DL). Because it is nonlinear in $B$, and because magnetic power is driven at all scales, the
426: Lorentz force too drives higher spatial harmonics in the flow.
427: The feedback from the magnetic
428: field eventually saturates the dynamo (Fig.~\ref{f:alltimes}). At the time of saturation,
429: the volume integrated plasma kinetic energy is
430: about 100 times larger than the magnetic energy. However, magnetic forces along the cell
431: walls, where the field is concentrated (see
432: Figure \ref{f:midplanes}), are locally strong, and inhibit the exponentially fast
433: stretching needed for further amplification. This is
434: consistent with the results of Cattaneo, Hughes, \& Kim (1996), who showed that magnetic forces
435: suppress chaos - characterized by exponential stretching - in the Galloway-Proctor flow even when
436: the field is too weak to modify the flow kinetic energy.
437:
438: Models with the most complete physics, namely both Reynolds stress and Lorentz forces in addition
439: to inertia and ion-neutral friction, fail to be dynamos (Fig.~\ref{f:alltimes}; models DA). This is to be expected,
440: since the initial magnetic field is too weak to affect the dynamics, and nonlinear
441: advection alone suppresses the dynamo.
442:
443: It is revealing to look at the structure of the magnetic field as well as its growth rate.
444: Figure~\ref{f:midplanes}
445: shows the logarithm of the magnetic energy density
446: %fh
447: from three views (along the $x$, $y$, and
448: $z$ axes) for the KI, KN, DL, DR, and DA models, at $t=20$. The top set of panels represent the
449: canonical kinematic GP dynamo (KI). The dark lines, where the magnetic field is very small,
450: demonstrate that it is folded on scales much smaller than the flow scale,
451: as is characteristic of dynamos with large $P_m$.
452: The second row of
453: panels shows the field when the plasma flow is determined by friction and inertia, but not Lorentz
454: forces or Reynolds stresses (KN). It is weaker than in models KI, but still spatially intermittent
455: and folded. If these folds were true magnetic nulls, they could be sites of fast magnetic
456: reconnection, but even a small amount of magnetic shear is enough to quench this process
457: (Heitsch \& Zweibel 2003a,b).
458:
459: \clearpage
460: \begin{figure*}
461: \begin{center}
462: \includegraphics[width=0.7\textwidth]{./f3.eps}
463: \end{center}
464: \caption{\label{f:midplanes}Logarithm of magnetic energy density at midplane, in planes
465: $(y,z)$ (left), $(x,z)$ (center) and $(x,y)$ (right). Models from top to bottom:
466: KI, KN, DL, DR and DA. For model names, see Table~\ref{t:models}.}
467: \end{figure*}
468: \clearpage
469:
470: More dramatic changes come with the inclusion of the Lorentz force, but not the Reynolds
471: stress (third row of Figure 3, models DL). Although Figure~\ref{f:alltimes} shows that the magnetic field is
472: still some way from saturation, and the magnetic energy is indistinguishable from that of the
473: kinematic models,the spatial intermittency and amount of folding are markedly reduced.
474:
475: The final two rows, models DR and DA, show how the small scale flow generated by nonlinear
476: advection affects the magnetic field. The ordered orientation associated with the strong
477: shear in the GP flow is essentially gone. Since there is little dynamo action, magnetic
478: feedback is unimportant, and the magnetic fields in the two cases are virtually identical.
479: Although there has been little amplification of the field as a whole, there are thin
480: filaments where it is locally strong.
481: %[Are these structures are associated with strong
482: %compression, or with strong vorticity?].
483: %fh: yes, they are. midplane plots similar to energy plots show negative divergence and large
484: % vorticities.
485: These structures are associated with strong compression (negative divergence in the ion
486: velocity) and with strong vorticity.
487:
488: Magnetic energy spectra for the KI, KN, and DL models at a linear resolution of $N=256$ are shown in
489: Figure~\ref{f:spectra}. Although there is a
490: slight reduction of power at the smallest scales it is not as important as the overall diminished
491: amplitude and it is clear that power spectra without phase information give an incomplete picture of
492: the structure of the magnetic field.
493:
494: \clearpage
495: \begin{figure}
496: \begin{center}
497: \includegraphics[width=\columnwidth]{./f4.eps}
498: \end{center}
499: \caption{\label{f:spectra}Power spectra of the magnetic energy for models KI, KN and DL at
500: a linear resolution of $N=256$, and at system time $t=16$, i.e. at a point when
501: the energy evolution of models KN and DL has separated due to saturation
502: (see Fig.~\ref{f:alltimes}).}
503: \end{figure}
504: \clearpage
505:
506: \section{
507: %%egz
508: Discussion and
509: %%egz
510: Summary}\label{s:summary}
511:
512: Galactic magnetic fields show considerable long range order, while computations
513: of small scale turbulent dynamos in highly conducting plasmas have shown instead that magnetic energy is concentrated at the
514: resistive scale. This suggests that there must be some feature of the interstellar dynamo which suppresses growth of the small
515: scale field,
516: %%egz
517: and that there exist mechanisms for transferring energy to scales larger than the turbulent injection scale.
518: %%egz
519:
520: In this paper we investigated the role of ion-neutral friction in eliminating the smallest scales. We
521: solved a simple model problem in which the neutral flow was known to be a dynamo, and was coupled to the plasma
522: through frictional drag. The plasma flow was determined by friction, inertia, and Lorentz forces. Ionization equilibrium
523: was assumed.
524: We used a spatially constant Ohmic resistivity, and we relied on viscous dissipation
525: at the resolution scale.
526: %%egz
527:
528: This study is, as far as we know, the first dynamo model in which the plasma flow is calculated from the full equation of motion,
529: rather than assuming the strong coupling approximation (Shu 1983). The strong coupling approximation breaks down at small scales,
530: and leads to an equation for magnetic energy growth which differs somewhat from the standard equation in a plasma [compare equations~(\ref{e:adenergy}) and (\ref{e:adenergy2})]. However, our
531: models differ from the interstellar medium in several significant respects, due to the prohibitively large computational resources
532: required for a realistic simulation. In the models, the
533: neutral forcing occurs on
534: timescales comparable to the ion-neutral collision time, whereas in the diffuse,
535: weakly ionized interstellar medium the shortest timescale in neutral turbulence
536: is probably 2-3 orders of magnitude longer. This introduces an artificially large difference between $\mbfv_i$ and $\mbfv_n$. By taking
537: the magnetic diffusivity larger than the viscous diffusivity (small magnetic Prandtl number $P_m$), we reverse the interstellar ordering
538: ($P_m\gg 1$). By assuming ionization equilibrium, an isothermal equation of state,
539: and a uniform density of neutrals, we preclude the possibility of plasma pressure gradients. And finally, the resistive scale is only about
540: two orders of magnitude less than the dynamical scale, instead of 5-6, as in the ISM.
541:
542: %%egz
543: Our work does not address the growth of the field at the largest scales. Whether ambipolar diffusion acting on small scale
544: turbulence can drive a large scale field, either by quasilinear (Zweibel 1988, Proctor \& Zweibel 1992) or nonlinear
545: (Subrahmanian 1997, 1999, Brandenburg \& Subrahmanian 2000) effects cannot be probed by our study due to the large range of
546: scales that would be required, and also, perhaps, because the neutral velocity is already a dynamo.
547: %%egz
548:
549: Our main results are as follows. First, friction successfully imprints a dynamo flow on the ions, despite the importance of inertial terms
550: [models KN; Fig. (\ref{f:alltimes})]. That is, computing $\mbfv_i$ from equation~(\ref{e:vidot}) leads to a flow which amplifies a weak magnetic field. Second,
551: adding nonlinear advection ($\mbfv_i\cdot\mbfnabla\mbfv_i$) to equation~(\ref{e:vidot}) results in an ion flow which does {\textbf{not}}
552: amplify the magnetic field, except in thin regions with large vorticity or negative divergence [models DR; Fig. (\ref{f:midplanes})].
553: This may be a manifestation of the difficulty of driving a dynamo when $P_m\ll 1$,
554: %%egz
555: or in a highly compressible medium. It would not necessarily carry over
556: to the diffuse ISM, in which $P_m$ is large{\footnote{The viscosity tensor in interstellar plasma is highly anisotropic due to the
557: large ratio of the ion gyrofrequency to the Coulomb collision rate. Therefore,
558: viscous transport across the magnetic field is strongly suppressed (Braginskii 1965), making the actual value of $P_m$ somewhat
559: unclear.}} and the recombination time is longer than the eddy turnover time.
560: %%egz
561: Third, when the Lorentz force is added, but the Reynolds stress is dropped, the dynamo saturates when the magnetic energy
562: is about two orders of magnitude less than the plasma kinetic energy [models DL; Fig. (\ref{f:alltimes})].
563: According to conventional wisdom, the magnetic energy
564: saturates only close to equipartition; the difference here
565: is due to the spatial intermittency of the magnetic field, which allows Lorentz forces to become strong just where the
566: field is being amplified. Finally, adding the Lorentz force suppresses small scale structure (Fig. \ref{f:midplanes}).
567:
568: The saturation reported here for models without nonlinear advection (models DL) has some
569: features in common with the study by Tanner \& Hughes (2003), although these authors
570: considered fully ionized systems. They studied a superposition of the GP flow (which they
571: call the CP, or Circularly Polarized flow) and a steady flow known to be a slow dynamo,
572: restricted the dynamics by omitting the Reynolds stress term and averaged the Lorentz force
573: over what in our case would be the $x$ direction. They found that when the GP flow dominates,
574: saturation occurs through suppression of exponential field line stretching by Lorentz forces,
575: rather than by enhanced dissipation brought about by an increase in small scale structure.
576:
577:
578: Taken at face value, our computations suggest that in weakly ionized gases, the efficiency of
579: dynamos is reduced at scales below the neutral-ion coupling length. The
580: main point is that saturation of the field
581: at small scales is determined by the kinetic energy
582: %fh
583: in the plasma at these scales rather than in the bulk medium.
584: The inertia of the neutrals contributes, and presumably increases the saturation level, only
585: for motions with turnover time $\tau_l$ greater than the neutral-ion collision time $\tau_{ni}\equiv (\rho_i\gamma_{in})^{-1}$. If we take
586: $n_i\sim 2\times 10^{-3}$ cm$^{-3}$, representative of both cold and warm neutral interstellar gas, then $\tau_{ni}\sim 3\times 10^{11}$s.
587: Taking the expressions for
588: Kolmogorov turbulence given in \S 2.1, and assuming the driving scale and velocity are 10 pc and 10 km s$^{-1}$,
589: respectively, we find $\tau_l\sim
590: \tau_{ni}$ at $l\sim 10^{-2}$pc. The equipartition magnetic field strength at this scale relative to the ions is only $\sim .02\mu G$, or about
591: 1\% of the Galactic value. Thus, saturation of the dynamo below the neutral-ion decoupling scale could prevent amplification of the field
592: at very small scales.
593:
594: At early times, when the magnetic field might have been several orders of magnitude or more weaker
595: than it is today, the decoupling effect would have been less important. Furthermore, the
596: interstellar medium was probably almost fully ionized until metallic coolants had mixed into it.
597: Thus, the mechanisms addressed here are probably more relevant to the maintenance of galactic
598: magnetic fields than to their ultimate origin.
599:
600: \acknowledgements
601: We are happy to acknowledge supports by NSF Grants AST-0507367 and
602: PHY-0215581 to the University of Wisconsin. Parts of the computations were performed at
603: the National Center for Supercomputing Applications under projects DAC AST-040026 and
604: MRAC AST-060031. We are grateful to an anonymous referee for useful suggestions.
605:
606:
607:
608: \begin{thebibliography}{}
609: \bibitem[Boldyrev \& Cattaneo(2004)]{517}
610: Boldyrev, S. \& Cattaneo, F. 2004, \prl, 92, 144501
611: \bibitem[Brandenburg \& Subramanian(2000)]{}
612: Brandenburg, A. \& Subramanian, K. 2000, \aa, 361, L33
613: \bibitem[Brummell, Cattaneo \& Tobias(2001)]{BCT2001}
614: Brummell, N.~H., Cattaneo, F., Tobias, S.~M.\ 2001,
615: {\em Fl. Dyn. Res.}, 28, 237
616: \bibitem[Cameron \& Galloway(2005)]{522}
617: Cameron, R. \& Galloway, D.\ 2005, \mnras, 358, 1025
618: \bibitem[Cattaneo, Hughes, \& Kim(1996)]{524}
619: Cattaneo, F., Hughes, D.W., \& Kim, E-J. 1996, \prl, 76, 2057
620: \bibitem[Childress \& Gilbert(1995)]{526}
621: Childress, F. \& Gilbert, A.D. 1995, {\em Stretch, Twist, Fold: The Fast Dynamo}, Springer
622: \bibitem[Courvoisier, Hughes, \& Tobias(2006)]{528}
623: Couvoisier, A., Hughes, D.W. \& Tobias, S.M. 2006, PRL, 96, 034503
624: \bibitem[Galloway \& Proctor(1992)]{GAP1992}
625: Galloway, D.~J. \& Proctor, M.~R.~E.\ 1992, \nat, 356, 691
626: \bibitem[Gilbert(2002)]{532}
627: Gilbert, A.D. 2002, Phys. Rev. D, 166, 167
628: \bibitem[Goldreich \& Sridhar(1995)]{534}
629: Goldreich, P. \& Sridhar, S. 1995, \apj, 438, 763
630: \bibitem[Goldreich \& Sridhar(1997)]{536}
631: Goldreich, P. \& Sridhar, S. 1997, \apj, 485, 680
632: \bibitem[Haugen, Brandenburg, \&Mee(2004)]{}
633: Haugen, N.E., Brandenburg, A., \& Mee, A.J. 2004, \mnras, 353, 947
634: \bibitem[Heitsch et al.(2007)]{538}
635: Heitsch, F., Slyz, A.D., Devriendt, J.E.G., Hartmann, L.W., Burkert, A.\ 2007, ApJ, 665, 445
636: \bibitem[Heitsch \& Zweibel(2003a)]{540}
637: Heitsch, F. \& Zweibel, E.G.\ 2003, \apj, 583, 229
638: \bibitem[Heitsch \& Zweibel(2003b)]{542}
639: Heitsch, F. \& Zweibel, E.G.\ 2003, \apj, 590, 291
640: \bibitem[Heitsch et al.(2004)]{544}
641: Heitsch, F., Zweibel, E.G., Slyz, A.D., Devriendt, J.E.G.\ 2004, \apj, 603, 165
642: \bibitem[Kim(1997)]{546}
643: Kim, E-J. 1997, \apj, 477, 183
644: \bibitem[Klapper \& Young(1995)]{548}
645: Klapper, I.M. \& Young, L.S. 1995, Comm Math Phys 173, 623
646: \bibitem[Kulsrud \& Anderson(1992)]{}
647: Kulsrud, R.M. \& Anderson, S.W. 1992, \apj, 396, 606
648: \bibitem[Kulsrud \& Zweibel(2008)]{550}
649: Kulsrud, R.M. \& Zweibel, E.G.\ 2008, astro-ph/0707.2783
650: \bibitem[Moffatt(1978)]{MOF1978}
651: Moffatt, H.K. \ 1978, in {\em Magnetic field generation in electrically conducting
652: fluids}, Cambridge Univ. Press, Cambridge
653: \bibitem[Ott(1998)]{555}
654: Ott, E. 1998, Phys. Pl. 5, 1636
655: \bibitem[Proctor \& Zweibel (1992)]{}
656: Proctor, M.R.E. \& Zweibel, E.G. 1992, GAFD, 64, 145
657: \bibitem[Schekochihin et al.(2004)]{557}
658: Schekochihin, A.A., Cowley, S.C., Taylor, S.F., Maron, J.L., \& McWilliams, J.C. 2004, \apj, 612, 276
659: \bibitem[Schekochihin et al.(2007)]{559}
660: Schekochihin, A.A., Isakov, A.B., Cowley, S.C., McWilliams, J.C., Proctor, M.R.E., \& Yousef, T.A. 2007, New J Phys, 9, 300
661: \bibitem[Shu(1983)]{}
662: Shu, F.H. 1983, \apj, 273, 202
663: \bibitem[Sridhar \& Goldreich(1994)]{561}
664: Sridhar, S. \& Goldreich, P. 1994, \apj, 432, 612
665: \bibitem[Subramanian(1997)]{}
666: Subramanian, K. 1997, arXiv:astro-ph/9708216
667: \bibitem[Subramanian(1999)]{}
668: Subramanian, K. 1999, PRL, 83, 2957
669: \bibitem[Tanner \& Hughes(2003)]{563}
670: Tanner, S.E.M. \& Hughes, D.W.\ 2003, \apj, 586, 685
671: \bibitem[Tang \& Xu(2000)]{TAX2000}
672: Tang, H.~Z. \& Xu, K.\ 2000, J. Comp. Phys., 165, 69
673: \bibitem[Xu(1999)]{XUK1999}
674: Xu, K.\ 1999, J. Comp. Phys., 153, 334
675: \bibitem[Zweibel (1988)]{}
676: Zweibel, E.G. 1988, \apj, 329, 384
677: \end{thebibliography}
678:
679:
680:
681: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
682: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
683: \end{document}
684: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
685: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
686:
687:
688:
689:
690:
691:
692: