1: \documentclass{emulateapj}

2: %\documentclass[12pt,preprint]{aastex}

3: %\usepackage{amsmath}

4: 

5: \newcommand{\Msun}{\ensuremath{M_\odot}}

6: %\newcommand{\boldsymbol}{\boldmath}

7: \newcommand{\boldsymbol}[1]{\mbox{\boldmath{$#1$}}}

8: 

9: \shorttitle{Evolution of Magnetic Turbulence in Shocks} 

10: \shortauthors{Chang, Spitkovsky, \& Arons}

11: 

12: \begin{document}

13: 

14: \title{Long Term Evolution of Magnetic Turbulence in Relativistic Collisionless Shocks: Electron-Positron Plasmas} 

15: 

16: \author{Philip Chang\altaffilmark{1,2,3}, Anatoly Spitkovsky\altaffilmark{4}, and Jonathan Arons\altaffilmark{1,2,5,6}}

17: 

18: \altaffiltext{1} {Department of Astronomy, Campbell Hall, University

19:   of California, Berkeley, CA 94720; pchang@astro.berkeley.edu,

20:  arons@astro.berkeley.edu}

21: \altaffiltext{2} {Theoretical Astrophysics Center} 

22: \altaffiltext{3} {Miller Institute for Basic Research}

23: \altaffiltext{4} {Department of Astrophysical Sciences, Peyton Hall,

24: Princeton University, Princeton, NJ: anatoly@astro.princeton.edu}

25: \altaffiltext{5} {Department of Physics, LeConte Hall, University

26:   of California, Berkeley, CA 94720}

27: \altaffiltext{6}{Kavli Institute for Particle Astrophysics and Cosmology, Stanford

28: University}

29: 

30: \begin{abstract}

31:   We study the long term evolution of magnetic fields generated by a

32:   collisionless relativistic $e^+e^-$ shock which is initially

33:   unmagnetized.  Our 2D particle-in-cell numerical simulations show

34:   that downstream of such a Weibel-mediated shock, particle

35:   distributions are close to isotropic, relativistic Maxwellians, and

36:   the magnetic turbulence is highly intermittent spatially. The

37:   non-propagating magnetic fields in the turbulence form relatively isolated regions

38:   with transverse dimension $\sim 10-20$ skin depths. These structures

39:   decay in amplitude, with little sign of downstream merging.  The

40:   fields start with magnetic energy density $\sim (0.1-0.2)$ of the

41:   upstream kinetic energy within the shock transition, but rapid

42:   downstream decay drives the fields to much smaller values, below

43:   $10^{-3}$ of equipartition after $\sim 10^3$ skin depths.

44: 

45: %  At late times in the simulations, which last for

46: %  $\sim 4000 \omega_p^{-1}$ and follow the motions of $15 \time 10^9$

47: %  particles, where $\omega_p$ is the relativistic plasma frequency

48: %  based on upstream parameters, the total magnetic energy, averaged

49: %  over the transverse dimension, declines in proportion to $t^{-1}$.

50: %  Noise due to the finite number of particles limits following this

51: %  decay to magnetic energy densities greater than $0.001 nkT$.

52:   

53:   In an attempt to construct a theory that follows field decay to

54:   these smaller values, we explore the hypothesis that the observed

55:   damping is a variant of Landau damping in an unmagnetized plasma.

56:   The model is based on the small value of the downstream magnetic

57:   energy density, which

58:   suggests that particle orbits are only weakly perturbed from

59:   straight line motion, if the turbulence is homogeneous.  Using

60:   linear kinetic theory applied to electromagnetic fields in an

61:   isotropic, relativistic Maxwellian plasma, we find a simple analytic

62:   form for the damping rates, $\gamma_k$, in two and three dimensions

63:   for small amplitude, subluminous electromagnetic fields.  We find

64:   that magnetic energy does damp due to phase mixing of current

65:   carrying particles as $(\omega_p t)^{-q}$ with $q \sim 1$. This

66:   overall decay compares well to that found in simulations, since it

67:   depends primarily on the longest wavelength modes, $kc/\omega_p \ll

68:   1$.  However, the theory predicts overly rapid damping of short

69:   wavelength modes.  We speculate that magnetic trapping of a

70:   substantial fraction of the particles within the highly spatially

71:   intermittent downstream magnetic structures may be the origin of

72:   this discrepancy.  In addition, trapping may form the basis for

73:   MHD-like behavior, permitting a small fraction of the initial

74:   magnetic energy to persist for times much greater than have been

75:   followed in the simulations.

76: 

77:   We briefly speculate on other physical processes, which depend on

78:   the presence of suprathermal particles, that may cause the

79:   generation of longer wavelength magnetic fields that create a

80:   magnetized plasma ($kr_{Larmor} \ll 1$), in which the damping is not

81:   as fast.  However, absent such additional physical processes, we

82:   conclude that initially unmagnetized relativistic shocks in

83:   electron-positron plasmas are unable to form persistent downstream

84:   magnetic fields.  These results put interesting constraints on

85:   synchrotron models for the prompt and afterglow emission from GRBs.

86:   We also comment on the relevance of these results for relativistic

87:   shocks in electron-ion plasmas.

88: 

89: % Because

90: %  the downstream magnetic turbulence actually is highly intermittent

91: %  in space, magnetic trapping of a substantial fraction of the

92: %  particles within the magnetic structures occurs and may be the

93: %  origin of the discrepancy between the theory and the late time

94: %  damping observed numerically.  We speculate that trapping may allow

95: %  persistence of a small fraction of the initial magnetic energy to

96: %  times much greater than have been followed in the simulations, since

97: %  permanently magnetically trapped particles are the basis for MHD

98: %  behavior.

99: 

100:   

101: %  Because the downstream magnetic turbulence actually is highly

102: %  intermittent in space, magnetic trapping of a substantial fraction

103: %  of the particles within the magnetic structures occurs and may be

104: %  the origin of the discrepancy between the theory and the late time

105: %  damping observed numerically. We exhibit results from the

106: %  simulations which demonstrate that magnetic trapping does occur in

107: %  the downstream medium, since even though the average magnetic energy

108: %  density is small compared to the thermal energy density, the actual

109: %  fields in the widely space filaments are strong enough to deflect

110: %  and trap a large fraction of the particles that encounter the

111: %  filaments. We speculate that trapping may allow persistence of a

112: %  small fraction of the initial magnetic energy to times much greater

113: %  than have been followed in the simulations, since permanently

114: %  magnetically trapped particles are the basis for MHD behavior.

115: 

116: \end{abstract}

117: 

118: \keywords{shock waves -- turbulence -- gamma ray: bursts -- plasmas}

119: 

120: \section{Introduction}

121: 

122: The prompt emission and afterglows of gamma-ray bursts (GRBs) may be

123: manifestations of ultrarelativistic shock waves, propagating in media

124: where the large scale upstream magnetization is too weak to affect the

125: shock structure, and too weak, if simply compressed by the shock, to

126: provide the magnetization inferred from synchrotron models of the

127: burst emission (see Piran 2005ab and references therein).  The weakly

128: magnetized outflows in the rotational equators of rotation powered

129: pulsars (Coroniti 1990) may also be sites of essentially unmagnetized

130: shock waves terminating the relativistic winds in a region which

131: occupies a finite latitude band with respect to the rotational equator

132: of the underlying neutron star.  Yet the radiation from these systems

133: has been modeled as being due to synchrotron radiation, which requires

134: the presence of a magnetic field strong enough to deflect particles

135: through a substantial fraction of their Larmor orbits. At the very

136: least, this requires magnetic structures with amplitude, $\delta B$,

137: whose characteristic dimensions, $R_B$, are comparable to or larger

138: than $mc^2 /e \delta B$, for all

139: particle energies inferred to contribute to the observed radiation.

140: 

141: Unmagnetized anisotropic plasmas spontaneously generate small-scale

142: magnetic fields via the Weibel instability (Weibel 1959).  Shocks have

143: strong plasma anisotropy in the transition layer separating the

144: upstream and downstream media, as well as in the foreshock where

145: downstream particles escape into the upstream, which provides the free

146: energy for generating magnetic fields with spatial scales on the order

147: of the plasma skin depth, $c/\omega_p$, where $\omega_p$ is the plasma

148: frequency.

149: %These magnetic fields provide the anomalous viscosity, which

150: %mediates a shock, through particles' elastic scattering from the

151: %non-propagating (in the shock frame) magnetic fluctuations (Moiseev

152: %and Sagdeev 1963).  Anisotropic flow energy is converted into

153: %isotropic themal energy, as measured in the frame of the downstream

154: %plasma.

155: Medvedev and Loeb (1999) and Gruzinov and Waxman (1999) argued that the relativistic form of this

156: instability (Yoon \& Davidson 1987) could macroscopically magnetize

157: the downsteam plasma where the GRB radiation arises (Larmor radii

158: small compared to flow scale).  Through numerical and analytic

159: studies, they and various authors (Silva {\it et al.} 2003, Medvedev

160: {\it et al.} 2005) have argued that the instability forms filaments of

161: electric current and $B$ field.  These filaments merge and cause

162: magnetic energy to cascade from the initial microscopic scale $\sim

163: c/\omega_p$ to larger scales.  Thus, the filamented plasma becomes

164: magnetized with a $B$ field hypothesized to survive throughout a large

165: fraction of the shocked medium.

166: 

167: Such inverse cascades have been observed in 2D and 3D particle-in-cell

168: simulations in the {\it foreshock} region, the part of the shock

169: structure where the upstream and downstream media interpenetrate and

170: the stream filamentation modes grow from noise with most incoming

171: particles still undeflected from the upstream flow.  

172: %In the foreshock

173: %region, the currents appear to self-organize into filaments which

174: %merge and generate ever larger scale magnetic structures until the

175: %plasma becomes magnetized.  

176: Simulations show that current filaments merge and grow in amplitude

177: until they reach the magnetic trapping limit, where the filament

178: currents and their magnetic fields become comparable to the Alfven

179: limit (Davidson {\it et al.} 1972; Kato 2005; also see Milosavljevic,

180: Nakar, \& Spitkovsky 2006; Milosavljevic \& Nakar 2006a). At that

181: point the particles' orbits on the filament boundaries become chaotic,

182: the filaments disorganize, and scattering from the disorganized

183: magnetic fluctuations halts the streaming of the bulk of the plasma,

184: isotropizing and thermalizing the flow, all within a layer tens to

185: hundreds of skin depths thick (Spitkovsky 2005), in accord with Kato

186: (2005)'s model for shock formation. The magnetic energy is as high as

187: 10-20\% of the bulk plasma flow energy {\it within this scattering

188:   layer}, where the density jump between upstream and downstream

189: occurs.

190: 

191: %This picture of merging current filaments

192: %has grown into a paradigm for Gamma Ray Burst physics, with regular

193: %invocation of this physical process as the explanation of the magnetic

194: %fields required in synchrotron models (Piran 2005ab, Katz et al 2006).

195: 

196: These initially small-scale B-fields must survive for tens of

197: thousands to millions of inverse plasma periods to serve as the source

198: of the magnetization invoked in synchrotron models of GRB emission

199: (Gruzinov \& Waxman 1999; Piran 2005ab; Katz, Keshet, \& Waxman 2007). Long-lived fields are also required in Diffusive Fermi Acceleration (DFA)

200: models used to explain the appearance of the nonthermal particle

201: spectra observed through their synchrotron and inverse Compton

202: emission in GRBs, and in pulsar wind nebulae (PWNs) and other sites of

203: nonthermal photon emission in relativistic flows.  However, present

204: simulations (Kazimura {\it et al.} 1998; Silva {\it et al.} 2003; Frederiksen  {\it et al.}

205: 2004; Hededal {\it et al.} 2005; Nishikawa {\it et al.}  2003, 2005)

206: have not followed the flow through the shock transition layer into the

207: downstream region because they employ periodic boundary conditions,

208: are too small, or both.  The question of the structure and long-term survival of 

209: the B-fields has remained open (see for instance, Gruzinov \& Waxman 1999; Gruzinov 2001ab; Medvedev 

210: {\it  et al.} 2005).

211: 

212: In this paper, we show via linear theory that the phase mixing between

213: individual particles and organized macroscopic currents implies

214: rapid decay of the magnetic energy in the downstream medium.  We begin

215: by briefly describing the essential features of the plasma produced by

216: the shock via numerical simulations in \S \ref{sec:simulation}.  These

217: simulations show the downstream magnetic structures are

218: non-propagating in the frame of the downstream medium, and are intermittent spatially, organized into clumps

219: and filaments of magnetic field with typical diameter $\sim 10-20$ skin depths, immersed into a highly isotropic plasma.  

220: %(oriented across the flow direction in the 2D results exhibited here;

221: %3D results, which form large scale loops and arcs, will be discussed

222: %elsewhere) 

223: In this downstream isotropic Maxwellian particle distribution, we

224: calculate the damping rates from Vlasov linear response theory,

225: assuming the particles are unmagnetized (Larmor radii $\gg$ magnetic

226: clump diameter), recovering a result due to Mikhailovskii (1979) in \S

227: \ref{sec:theory} and Appendix A.  Taking snapshots from the

228: simulations as initial conditions, we calculate the decay of the

229: magnetic field as a function of time and position, and compare these

230: theoretical calculations with simulations.  We find that the theory

231: does reasonably well in estimating the decay rate of the total

232: magnetic energy as $t^{-q}$ with $q\sim 1$.  However, it overestimates

233: the damping rate of shorter wavelength modes.  We speculate in

234: \S\ref{sec:trap} that this discrepancy may result from the magnetic

235: trapping of a large fraction of the particles, which suppresses the

236: disorganizing effect of phase mixing on the currents. We also discuss

237: the effects of an inverse cascade on the persistence of magnetic

238: fields, though we find little evidence for its relevance from

239: numerical simulations.  We summarize our results and draw some

240: implications for Fermi acceleration and for the magnetization required

241: for synchrotron emission in models of GRBs and other systems in

242: \S\ref{sec:discussion}.

243: 

244: 

245: %mediate the

246: %acceleration of a small number of particles to large $\gamma$'s.

247: 

248: 

249: 

250: 

251: %These small scale fields have been invoked as  In order to do so, these 

252: 

253: 

254: 

255: 

256: 

257: %In

258: %this scenario, these particle radiate through synchrotron emission in the

259: %magnetic fields created at the shock interface.

260: 

261: %For this magnetic energy to survive with little or no decline of the

262: %magnetic energy for the scale of the post shock flow that gives rise

263: %to the observed radiation ($10^6 - 10^{10}$ skin depths, Piran 2005)

264: %requires the now disorganized magnetic structures to continue

265: %merging to ever larger length scales.  If this is the case, the GRB

266: %magnetization problem would be solved.  Instead, simulations to date

267: %have shown that these fields are strongly damped {\it downstream},

268: %although because of residual electromagnetic noise introduced by the

269: %finite (although large) number ($10^{8.5} - 10^{10}$)of macroparticles

270: %employed, the asymptotic state of the magnetic energy remains

271: %inconclusive.

272: 

273: 

274: \section{Simulation Results}\label{sec:simulation}

275: 

276: Spitkovsky (2005) and Spitkovsky and Arons (in prep) describe a series

277: of 2D and 3D simulations of relativistic shock waves in $e^+e^-$

278: plasmas, for varying values of the upstream magnetic field $B_1$,

279: including $B_1 = 0$.  These are Particle-in-Cell (PIC) simulations

280: (Birdsall and Langdon 1991), using the code {\it TRISTAN-MP},

281: developed by one of us (AS).  It is a heavily modified descendant of

282: the publicly available code {\it TRISTAN} (Buneman 1993). We simulate shocks

283: by injecting cold relativistic plasma particles at one end of a large

284: domain and allowing the particles to reflect off a fixed conducting

285: wall at the other end of the box.  In this paper, we present 2D

286: calculations utilizing boxes as large as 50,000 x 2048 cells with up

287: to $1.35\times 10^{10}$ particles which allows us to fully resolve

288: shock formation. Although we track all three velocity and field

289: components, due to the two-dimensional symmetry only particle

290: velocities in the plane of the simulation are non-zero for initially

291: cold flow, and only the out-of-plane component of the magnetic field is

292: excited by in-plane currents. In-plane electrostatic fields are also

293: included.

294: 

295: The interaction of the reflected pair plasma with the incoming stream

296: forms a shock, which propagates toward the plasma injection surface.

297: In the simulations, one plasma skin depth spans 10 cells based on the upstream parameters,

298: i.e., $\lambda_{s1} = c/\omega_{p1} = \sqrt{m_\pm c^2 \gamma_1 /4\pi

299:   e^2 n_{1}}$, where $n_{1} $ is the upstream total density of

300: electrons and positrons (as measured in the frame of the simulation),

301: $\omega_{p1}$ is the upstream plasma frequency, and $\gamma_1 m_\pm

302: c^2$ is the upstream flow energy/particle.  The time step is $\Delta t

303: = 1/20 \omega_{p1}$.  In the upstream skin depth units, the largest

304: boxes are then 5000 x 205 $c/\omega_{p1}$. The longest simulation was

305: evolved for $5300 \omega_{p1}^{-1}$.  We have checked, by using boxes

306: of increasing transverse dimension, that the periodic boundary

307: conditions used on the transverse coordinate do not affect the scale

308: of the magnetic structures formed.  In the results exhibited below, we

309: use $64$ particles per cell ($32$ per species) in order to suppress

310: particle noise, which enables us to follow the dynamics of small

311: amplitude fields. We have performed convergence studies, varying the

312: number $N$ of particles per cell from $4$ to $64$, confirming that the

313: noise level decreases as $1/\sqrt{N}$, while the gross qualities of

314: the shock structure remain the same.  Figure \ref{fig:shock} shows the

315: snapshots of density and magnetic energy from a typical 2D simulation.\footnote{Note that we have defined the magnetic energy fraction $\epsilon_B\equiv B^2/4 \pi

316:   \gamma_1 n_1 m c^2$ in terms of

317:   the {\it upstream} kinetic energy density, instead of the downstream

318:   thermal energy density as is a common practice in the GRB community

319:   (c.f., Piran 1999).}  Coordinates are in units of the upstream skin

320: depth, $c/\omega_{p1}$. In the simulation shown, the upstream flow

321: moves to the left with $\gamma_1 = 15$.

322: 

323: \begin{figure*}

324: \plotone{fig1.subsamp.ps}

325: \caption{Snapshot of a region from a large 2D relativistic shock

326:   simulation. Incoming $\gamma=15$ flow is moving to the left,

327:   while the shock moves to the right at $\approx 0.5c$. The postshock

328:   plasma is stationary in the frame of the simulation.   

329: a) Density structure in the simulation plane

330:   showing the plasma density enhancements in the foreshock region that

331:   steadily grow up to the shock transition region, where the density

332:   becomes homogeneous. Density is normalized in units of the plasma

333:   density far upstream. The density

334:   jump, $\langle n_2 \rangle / \langle n_1 \rangle = 3.13$, shown is

335:   exactly what is predicted by the hydrodynamic jump conditions for

336:   $\gamma_1 = 15$ in a 2D gas. b) Magnetic energy, normalized in terms of upstream energy of the

337:   incoming flow: $\epsilon_B=B^2/4 \pi

338:   \gamma_1 n_1 m c^2$. 

339: The upstream magnetic filaments, which can be visualized as

340: sheets coming out of the page, that are

341: formed by the Weibel instability reach a peak just before the

342: shock.  These filaments become clumps of magnetic field, which can be

343: visualized as cross-sections of loops that are transverse to the page in

344: the downstream region, where they slowly decay away.  Note that

345: the B-field, which is organized into upstream filaments

346: or downstream clumps, always points in or out of the page.

347:   A power law

348:   scaling, $\epsilon_B^{1/4}$, was applied to stretch the color table

349:   to show weak field regions; this is reflected in the colorbar.  c)

350:   Plasma density averaged in the transverse direction as a function of

351:   the distance along the flow.  d) Magnetic energy density averaged in

352:   the transverse direction, as a function of distance along the flow.}

353: \label{fig:shock}

354: \end{figure*}

355: 

356: Our simulations are large enough to permit the complete development of

357: the shock and show the main features of contemporary collisionless

358: shock simulations even in reduced dimensionality.  For instance, we

359: see the factor of $\approx 3.13$ increase in density between the upstream and

360: the downstream (Fig.  \ref{fig:shock}c), which is the expected compression

361: factor for this 2D plasma when the plasma properties are measured in

362: the rest frame of the downstream plasma (Gallant {\it et al.}  1992;

363: Spitkovsky and Arons, in prep).  Current filaments (geometrically,

364: these are out-of-plane sheets) show up as the enhancement in plasma

365: density and magnetic energy density in the foreshock (Fig.

366: \ref{fig:shock}ab). The scale of the filaments grows towards the shock

367: through merging.  The shock is located where the density filaments

368: completely merge and are replaced by a quasi-homogeneous medium. The

369: subject of the saturation of the Weibel instability will be explored

370: in greater detail by Spitkovsky and Arons (in prep).

371: 

372: %occurs when

373: %the

374: %currents reach a critical strength set by the Alfven current limit (Kato 2005; 

375: %Davidson {\it et al} 1972),

376: 

377: %fields

378: 

379: 

380: %The free particle streaming in the filaments ends when the transverse

381: %magnetic fields in the filaments grow large enough to deflect

382: %particles by 90 degrees, a condition known as magnetic trapping, which

383: %has been long identified as the saturation mechanism for the

384: %non-relativistic Weibel instability (Davidson {\it et al} 1972).

385: %Magnetic trapping is equivalent to the Alfven critical current

386: %condition for saturation when the densities of the two

387: %counter-streaming beams of particles are the same. 

388: 

389: At the shock transition layer, the filaments disorganize and become

390: clumps of magnetic energy (in 2D the only appreciable magnetic

391: component is the out of plane $B_z$).  Note that these magnetic clumps

392: lose intensity the further downstream they are from the shock (Figure

393: \ref{fig:shock}b).  The magnetic energy peaks before the density

394: completes its rise (as we see in comparing c and d in Fig.

395: \ref{fig:shock}), i.e., the instability saturates at the Alfven

396: critical current before the shock fully develops.  We also note that

397: the particle distribution function changes from an anisotropic

398: free-streaming population to an isotropic (in the downstream rest

399: frame) thermal population.  In the downstream medium, we find that the

400: difference between the perpendicular and parallel momentum is

401: extremely small, $<1\%$.

402: 

403: As Figures \ref{fig:shock} b and d suggest, even though locally in the

404: shock magnetic energy can reach close to equipartition ($epsilon_B \approx 15\%$ when

405: averaged over the transverse dimension), the magnetic fields decay in

406: the downstream region of the shock.  To study this in further detail,

407: we present several cross sections of the plasma at times separated by

408: 450 $\omega_{p1}^{-1}$ in Figure \ref{fig:shock later}. Since our

409: simulation is performed in the downstream frame (frame of

410: the reflecting wall), we see the shock propagate through the box at

411: $\approx 0.5 c$ (value appropriate for 2D relativistic gas).

412: 

413: Downstream from the shock, the magnetic clumps weaken as a function of time

414: and appear to be non-propagating.  In Figure \ref{fig:shock later}, we see

415: this in panels a-d and in the black through blue curves in panel e,

416: which show magnetic energy averaged over the transverse dimension of

417: the simulation at times from panels a-d.  Although the separation

418: between prominent clumps that have larger field strength seems to

419: increase with time, this is due to the faster disappearance of

420: small-scale clumps, presumably due to decay, while the location or

421: size of strong clumps does not significantly evolve. There is little

422: evidence for clump merging far from the shock. In Figure \ref{fig:shock later}

423: we also plot a vertical line showing the location where we will

424: decompose the magnetic fields into Fourier modes later in \S3, in

425: order to study the comparison of the theory of B-field decay with the

426: numerical experiments.

427: 

428: \begin{figure*}

429: \plotone{fig2.subsamp.ps}

430: \caption{ Cross-sectional views of the magnetic energy density for the

431:   same region as Figure \ref{fig:shock}, but at different times

432:   separated by 450 $\omega_p^{-1}$ (a-d). The time of Figure

433:   \ref{fig:shock} corresponds to panel b) in this plot. In panel e) we

434:   plot the B-field energy, averaged over y-axis as a function of

435:   distance along the flow direction with one curve for each of the top panels: (a)

436:   black, (b) red, (c) green, (d) blue.  We also plot a vertical line

437:   at $x = 840 c/\omega_p$ to define the region over which we perform a

438:   modal analysis in \S\ref{sec:theory} and Figure

439:   \ref{fig:fftsingle}.}

440: \label{fig:shock later}

441: \end{figure*}

442: 

443: %There is little evidence

444: %of merging of the downstream magnetic structures, which now take the

445: %form of quasi-cylindrical $\theta$ pinches, with B organized in

446: %filaments directed out of the plane of the figure. 

447: 

448: Finally, we mention that 3D shock simulations also show similar

449: behavior (including the lack of substantial downstream merging of magnetic

450: structures) for the shorter times that can be followed with present 3D

451: simulations.  Topologically, the magnetic clumps in 2D become large

452: looping structures in 3D. If our 2D plane is thought of as a slice

453: through a 3D simulation, the 3D loops connect field emerging

454: from one clump and returning to another. The orientation of 3D loops

455: would be mostly perpendicular to the original direction of the flow.

456: The particle distribution function is also an isotropic, relativistic

457: Maxwellian in the downstream region.

458: 

459: %3D simulations show similar behavior (including lack of substantial

460: %downstream merging of magnetic structures), for the rather shorter

461: %times that can be followed. The filaments with straight magnetic

462: %fields form large looping structures with radii of curvature large

463: %compared to filament radius, as is illustrated in figure

464: %\ref{fig:threeDB}.

465: %

466: %\begin{figure}[H]

467: %\plotone{bfield.png}

468: %\caption{Visualization of the magnetic energy density in a 3D simulation, showing the upstream filaments along the flow to the right and the downstream loops, decaying 

469: %with distance from the shock front.  }

470: %\label{fig:threeDB}

471: %\end{figure}

472: 

473: 

474: \section{Downstream Evolution of Magnetic Turbulence}\label{sec:theory}

475: 

476: The simulations summarized in \S\ref{sec:simulation} suggest that

477: magnetic turbulence decays in the isotropic post-shock plasma.  We now

478: attempt to understand this theoretically with the goal of finding a model that

479: allows extrapolation of the magnetic evolution beyond the length of time that can be 

480: studied in direct shock simulations.  The simulations show that the downstream 

481: plasma is isotropic and the downstream particle distribution

482: function is well described by a relativistic Maxwellian. For the

483: purposes of this section, we {\it assume} the downstream field

484: amplitudes are so small that magnetic trapping is unimportant for

485: almost all particles; therefore, their orbits are almost straight

486: lines.  We will revisit this assumption in \S\ref{sec:trap}.

487: 

488: We begin by deriving the linear plasma response for electromagnetic

489: fluctuations in an isotropic relativistic plasma with small field

490: fluctuations (Mikhailovskii 1979).  The Vlasov equation for each species is

491: \begin{equation}\label{eq:linearize vlasov}

492:   \frac {\partial f_s}{\partial t} + {\boldsymbol v} \cdot {\boldsymbol \nabla} f_s + 

493:   \frac {q_s}{m_e} \left({\boldsymbol E} + \frac {{\boldsymbol v}} {c} \times {\boldsymbol B} \right)

494:   \cdot{\boldsymbol \nabla}_p f_s = 0.

495: \end{equation}

496: Here, $s$ is the species label, $+$ for positrons and $-$ for electrons,

497: with $q_{\pm} = \pm e$.  We linearize this equation with $f_s

498: \rightarrow f_{0s} + \delta f_s$, ${\boldsymbol B} \rightarrow \delta

499: {\boldsymbol B}$ and ${\boldsymbol E} \rightarrow \delta {\boldsymbol

500:   E}$, with the initial conditions downstream of the shock that tell

501: us that $\delta E$ and $\delta B$ are small in the sense that $(\delta

502: E^2 + \delta B^2)/8\pi nT \ll 1$.  Then $\delta f_s /f_{0s}$ is also

503: small.  The linearized Vlasov equation is

504: \begin{equation}\label{eq:vlasov}

505:   \frac {\partial\delta f_s}{\partial t} + {\boldsymbol v} \cdot {\boldsymbol \nabla}\delta f_s  

506:   + \frac {q_s}{m_e} \left(\delta {\boldsymbol E} + \frac {{\boldsymbol v}} {c} \times \delta {\boldsymbol B} \right)\cdot{\boldsymbol \nabla}_p f_{0s} = 0.

507:  \end{equation}

508: The plasma couples to the field through the Maxwell equations

509: \begin{eqnarray}\label{eq:maxwell}

510: {\boldsymbol \nabla} \times \delta {\boldsymbol E} &=& -\frac 1 c \frac {\partial \delta {\boldsymbol B}} {\partial t},\\

511: {\boldsymbol \nabla} \times \delta {\boldsymbol B} &=& \frac {4\pi} {c} \delta {\boldsymbol j} + \frac{1}{ c} \frac {\partial \delta {\boldsymbol E}} {\partial t},\label{eq:maxwell2}

512: \end{eqnarray}

513: where $\delta {\boldsymbol j} = \sum_s q_s\int {\boldsymbol v} \delta f_s d^3 p$ is the current density.

514: 

515: We orient coordinates so that $\delta {\boldsymbol E}$ is along $x$, the wave vector lies

516: along $y$, and $\delta {\boldsymbol B}$ is along $z$.  Assuming a wave solution with

517: amplitudes $\propto \exp\left(-i\omega t + i k y\right)$, equation

518: (\ref{eq:vlasov}) becomes

519: \begin{equation}

520: i \left(kv_y - \omega\right)\delta f_s + \frac {q_s}{m_e} \left(\delta E + 

521:   \frac {v_y} c \delta B\right)\frac{\partial f_{0s}}{\partial p_x} - 

522: \frac {q_s}{m_e} \frac {v_x} {c} \delta B \frac {\partial f_{0s}}{\partial p_y} = 0.  

523: \end{equation}

524: Using equation (\ref{eq:maxwell}) to eliminate $\delta E$ yields

525: \begin{equation}\label{eq:deltaf}

526:   \delta f_s = i \frac {q_s \delta B}{m_ekc} \left(\frac {\partial f_{0s}} {\partial p_x} 

527:     + \frac {kv_x}{\omega - kv_y} \frac {\partial f_{0s}}{\partial p_y} \right).

528: \end{equation}

529: Using (\ref{eq:maxwell2}) yields 

530: \begin{equation}\label{eq:deltaB}

531: i\left(\omega^2 - k^2 c^2\right)\delta B = -4\pi kc \delta j.

532: \end{equation}

533: Using (\ref{eq:deltaf}) in the definition of $\delta j$ yields

534: \begin{equation}

535: \delta j = -i \chi\omega \delta E

536: \label{eq:j2E}

537: \end{equation}

538: where $\chi$ is the plasma susceptibility,

539: \begin{equation}

540:   4 \pi \chi    =   \sum_s \frac {\omega_{p,{\rm NR},s}^2}{\omega^2} 

541:   \int v_x \left(\frac {\partial} {\partial p_x} + \frac {kv_x}{\omega - kv_y} 

542:     \frac {\partial}{\partial p_y}\right)\frac {f_{0s}}{n_s} d^3p.

543: \label{eq:susceptibility}

544: \end{equation}

545: %and the

546: %dispersion relation

547: %\begin{equation}\label{eq:dispersion}

548: %  \frac {k^2 c^2}{\omega^2} - 1 - \frac {\omega_p^2}{4\pi\omega^2} 

549: %  \int v_y \left(\frac {d} {dp_y} + \frac {kv_y}{\omega - kv_x} 

550: %    \frac {d}{dp_x}\right)F_0 d^3p = 0,

551: %\end{equation} 

552: Here $n_s$ is the number density of the electrons ($s = -$) or

553: positrons ($s=+$) in the downstream plasma, $\omega_{p,{\rm NR},s} =

554: \sqrt{4\pi q_s^2 n_s/m_e}$ is the non-relativistic plasma frequency.

555: We will assume by charge neutrality that in the downstream plasma

556: $f_{0+} = f_{0-}$ and $n_+ = n_-$ so that the sum over equal mass

557: species is trivial.

558: 

559: The dispersion relation for normal modes in the plasma plus electromagnetic 

560: field is

561: \begin{equation}\label{eq:general_dispersion}

562: \frac{k^2c^2}{\omega^2} - (1 + 4 \pi \chi) = 0.

563: \end{equation}

564: However, we emphasize that the linear relation between the currents and

565: the electric field (eq.[\ref{eq:j2E}]) through the susceptibility (\ref{eq:susceptibility}) applies to all 

566: (small amplitude) fluctuations, not only to normal modes.

567: 

568: 

569: %Since $F_0$ is isotropic, we perform the angular integration to find

570: %\begin{equation}\label{eq:worked out dispersion}

571: %  \frac {k^2 c^2}{\omega^2} - 1 - \frac {\omega_p^2}{2\omega^2} \int v

572: %  \left[\left(\frac {\omega} {kv}\right)^2 - \frac{\omega}{kv}\left(1

573: %  - \left(\frac {\omega} {kv}\right)^2\right)\left\{\frac 1 4

574: %  \log\left(\frac {1 + \omega/kv}{1 - \omega/kv}\right)^2 - i\frac {\pi}

575: %  2 \theta\left(\left(\frac {\omega}{kv}\right)^2 - 1\right)\right\}\right]

576: %  \frac {dF_0}{dp} p^2dp = 0

577: %\end{equation}

578: 

579: We evaluate equation (\ref{eq:susceptibility}) for distribution

580: functions that are isotropic in two and three dimensions in Appendix

581: A.  For a two-dimensional isotropic distribution function, we write

582: $f_{0s}({\boldsymbol p}) = f(p_{2d})g(p_z)$, where $p_{2d} =

583: \sqrt{p_x^2 + p_y^2}$ and perform the integral over $p_z$ such that

584: $\int g(p_z) dp_z = 1$.  For a three-dimensional isotropic

585: distribution function, we write $f_{0s}({\boldsymbol p}) =

586: f_{0s}(p_{3d})$, where $p_{3d} = \sqrt{p_x^2 + p_y^2 + p_z^2}$. In the

587: three-dimensional case, we confirm Mikhailovskii's (1979) result for a

588: relativistic isotropic plasma.  As we show explicitly in equation

589: (\ref{eq:worked out suscept}), subluminal waves ($\omega_r < kc$) are

590: damped, where $\omega_r = \Re(\omega)$.  The basic physics is that of

591: Landau damping (Stix 1992) or more precisely, phase mixing.

592: %The standard picture of Landau damping is that there is net power

593: %transfer between a Fourier mode and a particle when the Doppler

594: %shifted frequency vanishes, i.e., $\omega_r - {\boldsymbol k} \cdot

595: %{\boldsymbol v} = \omega_r - k v \cos \phi = 0$.  

596: For a subluminous wave in an isotropic relativistic plasma with

597: $\omega_r /k < c$, there always exists some angle $\phi$ between the

598: wave vector $k$ and a particle's momentum where a particle moving with

599: speed $c\beta$ is in phase with the field component, i.e.,

600: $\beta\cos\phi = \omega_r/kc$.

601: 

602: When the phase velocity of the field fluctuations is very small

603: compared to the thermal speeds of the particles (which includes the

604: zero wave velocity case, $\Re (\omega) = 0$), Landau damping takes on

605: the character of simple phase-mixing.  The currents, which support the

606: field fluctuations, are composed of particles.  These current carrying

607: particles have random motions which carry them out of the magnetic

608: structure, which the currents initially support.  As a result, these

609: currents and their magnetic field fluctuations are disorganized (or

610: damped) on the transit times of the particles across the field

611: fluctations.  It should be emphasized that the formal theory includes

612: both this phase mixing limit of the damping process and the limit in

613: which the phase 4-speeds $\beta_p /\sqrt{1-\beta_p^2}, \; (\beta_p

614: \equiv \Re \omega /kc),$ of the fields' Fourier components are high

615: compared to the particles' random 4-speeds $\beta /\sqrt{1 -\beta^2}$.

616: The applicable limit depends on whether the wave phase 4-velocity is

617: large or small compared to the mean 4-velocity of the particles in the

618: distribution function.  Hammett, Dorland \& Perkins (1992) present a

619: simple picture of this process in the case of non-relativistic

620: electrostatic waves.  Arons, Norman \& Max (1977) show how phase

621: mixing works for electromagnetic waves in a relativistic Maxwellian

622: plasma.  In any case, the end result is that there is a net power flow

623: from fields to particles, {\it i.e.} the fields decay, which we will

624: demonstrate explicitly below.

625: % Hence as long as population of particle are monotonically decreasing with larger momentum ($dF/dp < 0$) then wave energy is lost.  An inverted population ($dF/dp > 0$) may add energy to the wave, but such a study is beyond the scope of this analysis.  

626: 

627: We now study the action of such a plasma on an arbitary field of

628: initial B-field pertubations as generated by the Weibel

629: mediated shock.  We evaluate equation (\ref{eq:susceptibility})

630: numerically for two and three dimensions setting $\omega_r = 0$,

631: because of the non-propagating nature of the magnetic clumps, which we

632: infer from visual inspection of the simulations.  In the limit $k \ll

633: \omega_p/c$, we find:

634: \begin{equation}\label{eq:chi_limit}

635: 4\pi\chi \approx \left\{\begin{array}{ll} 

636: i \frac {\omega_{p}^2}{|k|c\omega} & \textrm{2D} \\

637: i \frac {\pi} 4 \frac {\omega_{p}^2}{|k|c\omega} & \textrm{3D}

638: \end{array}\right. . 

639: \end{equation}

640: Note that the 2D and 3D results only vary by a numerical factor.

641: Hence, long wavelength modes have the same qualitative behavior in two

642: and three dimensions.

643: %As we show in Appendix A, this simple form recovers the

644: %exact numerical solution for $\omega_r = 0$.  Hence, we will use

645: %equation (\ref{eq:chi_limit}) in the rest of the paper.

646: 

647: The plasma susceptibility, which we calculate in Appendix A (see also

648: eq.[\ref{eq:chi_limit}]), can be utilized to calculate the evolution,

649: i.e., the damping or growth, of an initial field of fluctuations.  Appendix B

650: contains this calculation in more detail; the result is

651: \begin{equation}\label{eq:damping}

652: \frac {d|\delta B_k|^2}{dt} = -2\gamma_k |\delta B_k|^2,

653: \end{equation}

654: where $\gamma_k = \left(kc\right)^2\omega^{-1} \Im\left(4\pi\chi\right)^{-1}

655: $ (see also eq.[\ref{eq:app_general gamma}]).  The

656: asymptotic forms of $\chi$ from equation (\ref{eq:chi_limit}) for 3D

657: and 2D gives (see also eq.[\ref{eq:app_gamma}])

658: \begin{equation}\label{eq:gamma}

659: \gamma_{k} = \left\{\begin{array}{ll} \frac {|kc|^3} {\omega_{p}^2} & \textrm{2D} 

660: \\ \frac {4} {\pi} \frac{|kc|^3} {\omega_{p}^2}& \textrm{3D}

661: \end{array}\right. .

662: \end{equation}

663: 

664: 

665: %Linear theory predicts that short wavelengths are

666: %strongly damped, but long wavelengths persist for longer.  

667: We now compare the expectations from linear response theory to the

668: numerical simulations.  We begin by taking the Fourier transform of

669: $\delta B$ from our 2D numerical simulations from a downstream region

670: where the shock is fully developed, {\it i.e.}, behind the shock

671: front at $ x = x_0$,

672: \begin{equation}\label{eq:fourier transform}

673:   \delta B(x_0,y) = \frac 1 {\sqrt{2\pi}} \int dk_y \delta B_{k_y}(x_0)

674:   \exp\left(ik_y y\right)

675: \end{equation}

676: We evolve $\delta B_{k_y}$ in accordance to equation

677: (\ref{eq:damping}) using the asymptotic forms in equation

678: (\ref{eq:gamma}) and compare our analytically evolved spectra with the

679: numerical simulation at later times.  In Figure \ref{fig:fftsingle},

680: we take the initial data from a region at $x_0=840 c/\omega_p$, which

681: we marked with a line in Figure \ref{fig:shock later}.  To accumulate

682: sufficient statistics, we average the fields over $14\, c/\omega_p$ in

683: the flow direction.  We evolve these spectra for 450 (red), 900

684: (green), and 1350 $\omega_p^{-1}$ (blue) using equation

685: (\ref{eq:gamma}) and compare this to snapshots taken from our

686: numerical simulations at these times. Theory and simulation agree at

687: very low wavenumber ($k_yc/\omega_p \lesssim 0.2$).  However, theory

688: overpredicts the cutoff in power at larger k.  The discrepancy

689: suggests that linear theory is insufficient to describe the nature of

690: downstream magnetic turbulence and that additional physics is needed

691: (see \S\ref{sec:trap}).

692: 

693: \begin{figure}

694: \plotone{f3.eps}

695: \caption{ Spectral evolution of magnetic field from the slice at $840 c/\omega_p$ from Fig. \ref{fig:shock later}. Initial field spectrum (black solid line) is plotted after $450\,\omega_p^{-1}$ (red),

696: $900\,\omega_p^{-1}$ (green), and $1350\,\omega_p^{-1}$ (blue) based on simulation data. Dashed curves represent analytic evolution of the initial field and overpredict decay of short-wavelength structures. 

697: %Comparison of analytic evolution (dashed line) with the 2D

698: %simulation (solid lines) at $450\,\omega_p^{-1}$ (red),

699: %$900\,\omega_p^{-1}$ (green), and $1350\,\omega_p^{-1}$ (blue) of an

700: %initial field taken from a region behind the shock.  The spectral decomposition of the initial

701: %field (black solid line) is also shown.

702: \label{fig:fftsingle} }

703: \end{figure}

704: 

705: The lack of power in short wavelength modes suggests that the total

706: B-field is determined by long wavelength modes.  We now use equation

707: (\ref{eq:damping}) to find a simple decay law for the total B-field:

708: \begin{equation}

709: \label{eq:fourier evolution}

710:   \delta B(x_0,y,t) = \frac 1 {\sqrt{2\pi}} \int dk_y \delta B_{k_y}(x_0)

711:   \exp\left(ik_y y - \gamma_k t\right),

712: \end{equation}

713: where $\delta B_{k_y}(x)$ is defined in equation (\ref{eq:fourier

714:   transform}).  Inserting equation (\ref{eq:gamma}) into (\ref{eq:fourier evolution}) and

715: integrating,  with $\delta B_{k_y} =a k_y^p$, where $p$ is the

716: long wavelength spectral index and $a$ normalizes the amplitude, we find

717: \begin{eqnarray}

718:   \delta B(x_0,y,t) &=&  \frac 1 {\sqrt{2\pi}} a \int_{k_0}^\infty dk_y k_y^p

719:   \exp\left(-\alpha \omega_{p}t \left(\frac {k_yc} 

720:       {\omega_{p}}\right)^3\right) \\ 

721:   &\approx& \frac{1}{3 \sqrt{2\pi}} a \left( \frac{\omega_{p}}{c} \right)^{p+1}

722:      \Gamma\left(\frac {p+1}{3} \right)  

723:   \left(\frac{1}{ \alpha \omega_{p} t } \right)^{(p+1)/3}

724: \label{eq:decay-power}

725: \end{eqnarray}

726: where we have taken $y=0$ without loss of generality, $\alpha = 4/\pi$

727: in 3D and $\alpha = 1$ (2D), $k_0 $ is small such that $\omega_p t

728: (k_0 c/\omega_p)^3 \approx 0$, and $\Gamma$ is the gamma function.

729: Note that $\omega_p$ is for the downstream plasma frequency.  However, this

730: detail is irrelevant in terms of determining the power law as a

731: function of $t$.

732: % - in the simulations, where time is

733: %measured in units of $\omega_{p1}^{-1}$ and length in units of

734: %$c/\omega_{p1}$, the transverse size is $200 c/\omega_{p1}$ in the 2D

735: %models.  

736: Setting $k_0$ to be small is a safe approximation in the simulations,

737: and of course is excellent in much larger astrophysical systems.

738: Thus if the initial spatial spectrum is a power law in wave number

739: $|\delta B_k|^2 \propto k^{2p}$, then the theory predicts $\delta B^2

740: \propto t^{-2(p+1)/3}$.

741: 

742: %Magnetic field energy appears initially to follow this trend in our

743: %numerical simulations as shown in .  

744: We study the region immediately following the peak of magnetic energy

745: in the shock front (where it reaches $\delta B^2 = B_{\rm max}^2$) at

746: $x_{\rm peak}$ and plot its value as a function of position in the

747: postshock region.  For a shock moving at constant velocity, we have

748: $x_{\rm peak} - x \propto t$.  Hence $\delta B^2 \propto (x_{\rm peak}

749: - x)^{-2(p+1)/3}$.  Our numerical simulations are extremely suggestive

750: that the magnetic energy density follows the $p = 0, \; \epsilon_B =

751: \delta B^2/8\pi \propto t^{-2/3}$ decay expected for an initially flat

752: magnetic spectrum at early times, then steepens to a $t^{-1}$ decay at

753: later times as shown in Figure \ref{fig:eb}, until the noise finally

754: swamps the signal.\footnote{Gruzinov (2001a) performed shock

755:   simulations (initiated by a collision between two $e^\pm$ plasmas),

756:   which show decay of the magnetic energy averaged over the dimension

757:   across the flow similar to the early phases of the decay we report

758:   here.}  

759: This $t^{-1}$ decay rate would imply an initial spatial

760: index of $p=1/2$ at low wavenumber, though our analysis of additional

761: simulations with a large transverse spatial scale suggest $p=0$. The 

762: difference in the index of the decay law expected from the theory and measured from 

763: the simulations may be due to magnetic trapping (see \S\ref{sec:trap}).

764: Gruzinov (2001b) offered an alternative

765: explanation of $t^{-1}$ decay of the magnetic energy density observed

766: in 2D simulations of Weibel instability in counterpropagating pair

767: plasmas.  In that simulation the beams were moving perpendicular to

768: the simulation plane\footnote{Being orthogonal to the direction of

769:   motion these simulations did not form a shock and retained some

770:   counterstreaming at late times. In contrast, our simulations lose

771:   the counterstreaming once the shock forms.}.  In order to explain

772: the field decay, Gruzinov (2001b) had to assume that the magnetic

773: structures increase their size at the Alfven velocity. In our shock

774: simulations, the downstream magnetic structures do not significantly

775: expand (\S2), hence the similarity of decay law between the two

776: simulations is likely a coincidence. The picture of filament expansion

777: and merging of Gruzinov (2001b) is likely valid very close to the

778: shock, but is not observed in the longer evolution of the downstream

779: plasma.

780: %However, his argument assumed that the expanding

781: %magnetic structures, which he observed in his colliding beam

782: %simulations,\footnote{Gruzinov (2001b) simulates the evolution of

783: %  magnetic field during Weibel instability driven by collision of two

784: %  infinite plasma streams.  The plane of the 2D simulation is

785: %  perpendicular to the direction of motion of the streams. As such,

786: %  this simulation does not form a shock.  In contrast, our simulation

787: %  is in the plane along the motion of the streams, and allows density

788: %  compression. }  increase their size at the Alfven velocity.  As we

789: %have pointed out in \S2, these magnetic structures neither propragate

790: %nor significantly expand.  Hence, Gruzinov's reasoning is not

791: %applicable to our results.

792: %We

793: %discuss Gruzinov's (2001) alternative explanation of $t^{-1} $ decay

794: %of the magnetic energy density, based on conservation of the vector

795: %potential (which is a much more severe requirement than conservation

796: %of canonical momentum), in Appendix \ref{sec:canonical momentum}.

797: 

798: \begin{figure}

799: \plotone{f4.ps}

800: \caption{Magnetic energy density (in units of upstream kinetic energy) as a function of position downstream of the shock. A broken power law proportional $(x-x_{\rm

801:     peak})^{-2/3}$ fits well at early times, but a $(x-x_{\rm

802:     peak})^{-1}$ power law fits better at later times.  }

803: \label{fig:eb}

804: \end{figure}

805: 

806: 

807: 

808: %However, the 2D spectra are rather noisy, since they are 1D spectra over a single

809: %transverse dimension.  Spectra from 3D simulations are rather better behaved,

810: %since they are taken over a 2D y-z plane, as shown in Figure \ref{fig:fftpeak}

811: %These spectra  reveal an interesting departure of the theory from the experiments,

812: %as shown in Figure \ref{fig:fftpeak}

813: %\begin{figure}[H]

814: %  \epsscale{1.0} \plotone{manyBkwithfitbw.png}

815: %  \caption{Decay of the magnetic fluctuation power spectra  from a 3D simulation

816: %  run with $512 \times 512 \times 6000$ cells, and 4 particles per cell initially.

817: %   The four curves show the spectra with time increasing from top to bottom. Each curve 

818: %   is separated by $\Delta t = 182/\omega_{p1} $.  The fitted straight lines are

819: %   $| \delta B_k|^2 \propto \exp[- g \omega_{p1}t (kc/\omega_{p1}) ]$ - the decay

820: %   is better fit with a damping rate linear rather than cubic in $k$. In addition, these

821: %   3D spectra show substantial magnetic merging at early times, shown by the move of

822: %   the magnetic energy peak to smaller wave number. This merging dies away as the

823: %   decaying magnetic clumps become more spatially isolated. }

824: %\label{fig:fftpeak}

825: %\end{figure}

826: \begin{figure*}

827: \plotone{fig5.subsamp.ps}

828: \caption{Segments of particle orbits from the high resolution 2D simulation overplotted on a snapshot of magnetic energy downstream of the shock. Typical sizes of magnetic clumps are

829:   10-20 $c/\omega_p$. The shock is near the right boundary of this slice, moving to the right. The large angle deflections in particle orbits suggest the Larmor

830:   radii of many particles in this slice %$\sim 300 c/\omega_{p1}$

831: are of the same order as the sizes of the magnetic clumps. 

832: \label{fig:trapped}}

833: \end{figure*}

834: 

835: \section{Magnetic Trapping}\label{sec:trap}

836: 

837: Our simulations and theory suggest power law $t^{-q}$ with

838: $q \sim 1$ temporal decay of the total magnetic energy density in

839: rough agreement with each other.  However, linear theory and numerical

840: simulations disagree on the decay rate for short wavelengths, which

841: hinders a determination of the ultimate fate of the fields on times

842: longer than several thousand plasma periods. This discrepancy may

843: arise from the nonlinear effects of magnetic trapping.  Particles in

844: the magnetic fields do not follow straight line trajectories that are

845: weakly perturbed, but are partially trapped and strongly deflected.

846: We illustrate this point from following test particles' orbits in

847: numerical simulations in Figure \ref{fig:trapped}.  The test particles

848: suffer large deflections from straight-line orbits as they encounter

849: magnetic clumps.  Therefore, the Larmor radii of many of the test

850: particles are of the same order of the sizes of these clumps or

851: smaller.  Indeed closer inspection of some of the test particle orbits

852: that are embedded inside the clumps suggests that they are completely

853: trapped.

854: 

855: 

856: 

857: 

858: %To hammer this point further, we plot the distribution of Larmor radii

859: %in the whole downstream region in Figure \ref{fig:larmor}.  Since the

860: %rough size scale of these magnetic clumps are 50-100 cells, a

861: %substantial number of particles are trapped in these clumps.

862: %\begin{figure}

863: %\plotone{LarmorRadiiDistrLinear.png}

864: %\caption{Distribution of downstream particles' formal Larmor radii

865: %  $m_\pm c^2 \gamma/eB $. Since the scale of the magnetic structure is

866: %  typically 50-100 cells, a substantial fraction of particles are likely trapped. 

867: %  \label{fig:larmor}}

868: %\end{figure}

869: 

870: These strong departures from weakly perturbed particle dynamics may be

871: the cause of the decreased damping at large wavenumber found in the

872: simulations, compared to the predictions of unmagnetized plasma theory.

873: Magnetic trapping may also modify the decay of magnetic fields at

874: small wavenumber, leading perhaps to a different decay law instead of 

875: expected $t^{2/3}$ decay law that we find from our simple linear theory 

876: (eq.[\ref{eq:decay-power}] with $p=0$).

877: The temporary (and permanent, for some particles)

878: binding of particles to the spatially intermittent magnetic fields

879: reduces the effect of rapid phase mixing central to the unmagnetized

880: plasma damping theory.  Magnetic trapping already plays a critical

881: role in the saturation of the initial Weibel instability at the shock

882: transition region (Kato 2005; Davidson {\it et al.}  1972). Our

883: analysis suggests that trapping also plays an important role

884: downstream even though the {\it average} magnetic amplitudes are

885: greatly reduced from the shock transition region - the essential point

886: is that within the isolated filaments, the magnetic pressure is not

887: small.  The problem of the damping of isolated magnetic structures in

888: a plasma with partial magnetization illustrated in Figure

889: \ref{fig:trapped} will be the subject of a separate investigation.

890: 

891: 

892: \section{Discussion}\label{sec:discussion}

893: 

894: We have studied the downstream evolution of magnetic turbulence in the

895: context of a collisionless $e^+e^-$ shock both analytically and

896: numerically.  Our large scale 2D simulations show the formation of

897: filaments in the foreshock region which merge and grow

898: until they reach the shock transition region.  Past the shock

899: transition region, these filaments break up into magnetic clumps in a

900: quasi-homogenous medium where the background particle distribution

901: function is an isotropic Maxwellian.  In such a background, we showed

902: that magnetic energy will decay like $t^{-q}$ with $q\sim

903: 1$ due to untrapped particle phase mixing, which is broadly consistent

904: with our numerical simulations.  In detail, the theoretical decay

905: rates depend more strongly on wavelength than is seen in the numerical

906: experiments.  We suggested that magnetic trapping may play an

907: important role in resolving this discrepancy.  Trapping can lead to

908: MHD-like behavior and possible long time persistence of some of the

909: magnetic energy in spatially intermittent, more strongly magnetized

910: subregions. This effect is a subject of further study.

911: 

912: % In addition, a

913: %sufficiently fast inverse cascade may reduce magnetic decay.  However,

914: %our simulations show no evidence for such a mechanism in the

915: %downstream medium.

916: 

917: Magnetic field decay may be alleviated via an inverse cascade from

918: small scales to large scales.  An inverse cascade via current filament

919: merging operates strongly in the {\it foreshock} region (Silva {\it et

920:   al.}  2003; Spitkovsky 2005).  However, it is unclear if this

921: picture can be applied to the downstream region.  The magnetic

922: filaments so prevalent in the upstream region are gone and are

923: replaced by isolated magnetic clumps (loops), a result common to both 2D and

924: 3D simulations.  Filament merging does occur, but is confined to the

925: foreshock, where the filaments exist.  We see little evidence of

926: merging magnetic clumps.  This is reinforced by Katz {\it et al.} (2007), who

927: suggest via self-similar arguments that the inverse cascade does not

928: operate downstream of the shock.  Hence, the magnetic energy should damp

929: away in the downstream region.

930: 

931: We have not studied in detail how these magnetic clumps are confined

932: in the downstream plasma.  However, Fujita et al. (2006) have found

933: these same magnetic clumps in simulation of the non-relativistic

934: Weibel instability that is appropriate for galaxy clusters.  In this

935: case, they argue that pressure of the external medium confines these

936: magnetic clumps.  The strong perturbation from straight line motion,

937: which we studied in \S\ref{sec:trap} for test particles, suggest that

938: the magnetic clumps in our simulation are also sensitive to momentum

939: transfer between particles and itself.  That is, the magnetic clumps

940: in our simulations are also confined by external pressure.

941: 

942: Whatever the final fate of the spatially intermittent fields, our

943: study of the nature of $e^+e^-$ shocks suggests that the belief in the

944: persistence of Weibel generated magnetic fields with strengths

945: comparable to those appearing within the shock transition is overly

946: optimistic.  Magnetic field energies tend to decline rapidly after

947: about a few hundred plasma skin depths. Only the very long term

948: evolution is still open to some question, i.e., does $\langle

949: \epsilon_B \rangle$ settle at some value smaller than $10^{-3}$, or

950: does the spatially intermittent magnetic field decline to zero?

951: 

952: The rapid decay suggested by our analysis puts severe constraints on

953: the synchrotron emission mechanisms for GRBs.  The width of the

954: emitting region is $\sim 10^9 c/\omega_p$ (Piran 2005b), which is much

955: larger than the region over which we expect magnetic fields to

956: persist.  However, decaying magnetic fields may not be inconsistent

957: with GRB observations.  Pe'er \& Zhang (2006) suggest that the prompt

958: emission from the internal shock may be more consistent with a

959: decaying magnetic field component rather than a persistent field

960: component.  A field that persists over a scale of $10^4-10^5$ plasma

961: skin depths fits the spectra better at low-energies than a

962: field that persists over the entire thickness of the shell ($10^9$

963: plasma skin depths).  Small magnetized regions may also be important

964: in the context of the afterglow (Rossi \& Rees 2003).

965: 

966: In young pulsar wind nebulae, post

967: shock magnetic fields averaged over the whole latitudinal extent of

968: the observed emission tori have energy densities of a few

969: percent of the post shock plasma energy density.  It is possible that

970: weaker magnetic fields exist near the midplane of the equatorial flow, a

971: region of particular interest to the conversion of flow energy into

972: the observed nonthermally emitting spectra of $e^\pm$.  If so, the

973: Weibel mediated shock dynamics studied here may be of relevance to

974: these systems' behavior.

975: 

976: It is also possible that weak systematic upstream magnetic fields are

977: of essential importance, and that shocks in completely unmagnetized

978: plasmas are an oversimplification.  Suprathermal particle generated at

979: the relativistic shock front may alter the basic physics of the

980: collisionless shock, if a mean field is present.  Milosavljevic and

981: Nakar (2006b) argue accelerated particles streaming into the {\it

982:   upstream} magnetized medium can drive long wavelength, magnetized

983: turbulence with $\delta B /B \gg1$, which might persist into the

984: downstream and provide the magnetization required in phenomenological

985: models of GRB and PWN emission.  Then the shock mediated by Weibel

986: turbulence becomes a subshock within a much larger extended structure,

987: responsible only for thermalizing the bulk of the flow and injecting

988: the particles that are Fermi accelerated in the turbulence generated

989: by high energy particle streaming.  This is a relativistic version of

990: Bell's (1978) (also see Bell 2004, 2005) picture of particle

991: acceleration in non-relativistic shocks.

992: 

993: Finally, we briefly mention the extension of this theory to

994: electron-ion plasmas.  Let us presume initially that the ions and

995: electrons are isotropic but remain decoupled, i.e., a two-temperature

996: relativistic plasma. Equation (\ref{eq:chi_limit}) becomes

997: \begin{equation}\label{eq:i-e chi_limit}

998: \chi \approx i\frac {\pi} 4 \frac {\omega_{p,i}^2 + \omega_{p,e}^2}{|k|c\omega},

999: \end{equation}

1000: where $\omega_{p, i}^2 = 4\pi n_i e^2/\gamma_i m_i$ is the plasma

1001: frequency of the ions, $\gamma_i$ is the Lorentz factor associated

1002: with the ion temperature, $\omega_{p, e}^2 = 4\pi n_e e^2/\gamma_e

1003: m_e$ is the plasma frequency of the electrons, and $\gamma_e$ is the

1004: Lorentz factor associated with the electron temperature.  Depending on

1005: the relative values of the electron temperature and ion temperature,

1006: one term may dominate.  However, initial large-scale simulations of

1007: ion-electron collisionless shocks suggest that both reach roughly

1008: equipartition with each other (Spitkovsky 2007), thereby reproducing

1009: the physics of the $e^{\pm}$ shock.  In this case, the relativistic

1010: electrons and ions contribute equally to the decay rate because

1011: thermal equipartition prevails, i.e., $m_e \gamma_e = m_i

1012: \gamma_i$. Thus, the electron-ion plasma has the same dynamics as the

1013: electron-positron plasma.

1014: 

1015: \acknowledgements

1016: 

1017: We thank S. Cowley, D. Kocelski, M. Milosalavjic, A. Pe'er and E.

1018: Quataert for useful discussions. P.C. and J.A. thank the Institute for

1019: Advanced Study for its hospitality; P.C. also thanks the Canadian

1020: Institute for Theoretical Astrophysics for similar hospitality.  P.C.

1021: is supported by the Miller Institute for Basic Research. J.A. has

1022: benefited from the support of NSF grant AST-0507813, NASA grant

1023: NNG06GI08G, and DOE grant DE-FC02-06ER41453, all at UC Berkeley; by

1024: the Department of Energy contract to the Stanford Linear Accelerator

1025: Center no. DE-AC3-76SF00515; and by the taxpayers of California.  A.S. is pleased to acknowledge that the simulations reported on

1026: in this paper were substantially performed at the TIGRESS high

1027: performance computer

1028: center at Princeton University which is jointly supported by the Princeton

1029: Institute for Computational Science and Engineering and the Princeton

1030: University Office of

1031: Information Technology.

1032: \appendix 

1033: \section{Susceptibility in Two and Three Dimensions}\label{sec:susceptibility}

1034: 

1035: In this section, we solve the susceptibility (eq.[\ref{eq:susceptibility}]) for two and three dimensional plasma.  The 3D case

1036: (\S\ref{sec:3dresult}) has been previously solved by Mikhailovskii

1037: (1979) and we reproduce his result here for completeness.  In

1038: addition, we present the 2D case (\S\ref{sec:2dresult}), which

1039: allows for a comparison to the two-dimensional simulations reported in

1040: this paper.

1041: 

1042: \subsection{Three Dimensions}\label{sec:3dresult}

1043: 

1044: After summing over the electrons and positrons, equation (\ref{eq:susceptibility}) is

1045: \begin{equation}\label{eq:susceptibility 3d}

1046:   4 \pi \chi    =   \frac {\omega_{p,{\rm NR}}^2}{\omega^2} 

1047:   \int v_x \left(\frac {d} {dp_x} + \frac {kv_x}{\omega - kv_y} 

1048:     \frac {d}{dp_y}\right)\frac {f_0}{n} d^3p,

1049: \end{equation}

1050: where $n$ is the number density of electrons and positions.  Since

1051: $f_0$ is isotropic and therefore independent of angle, we orient the

1052: spherical integral in a {\it non-standard manner} so that the pole

1053: points along the k-vector, i.e., the y-axis.  We find

1054: $\boldsymbol{\hat{p}}\cdot \boldsymbol{\hat{y}} = \cos\theta$ and

1055: $\boldsymbol{\hat{p}}\cdot \boldsymbol{\hat{x}} = \sin\theta\sin\phi$.

1056: So equation (\ref{eq:susceptibility 3d}) becomes

1057: \begin{equation}

1058:   4 \pi \chi    =   \frac {\omega_{p,{\rm NR}}^2}{\omega^2n} 

1059:   \int v\sin^2\theta\sin^2\phi \left(1 + \frac {\cos\theta}{\omega/kv - \cos\theta}\right)\frac {df_0} {dp} p^2dpd\Omega,

1060: \end{equation}

1061: where $d\Omega = \sin\theta d\theta d\phi$.  Performing the integral

1062: over $\phi$ and making the substitution $\zeta = \cos\theta$, we find

1063: \begin{equation}

1064:   4 \pi \chi   =   \frac {\pi\omega_{p,{\rm NR}}^2}{\omega^2n} 

1065:   \int v\left(1-\zeta^2\right) \left(1 + \frac {\zeta}{\omega/kv - \zeta}\right)\frac {df_0} {dp} p^2dpd\zeta.

1066: \end{equation}

1067: Integrating over $\zeta$ from -1 to 1, we find

1068: \begin{eqnarray}

1069:   4 \pi \chi  &  = & \frac {2\pi\omega_{p,{\rm NR}}^2}{\omega^2n}  \int  p^2dp \frac {df_0}{dp}  v{\Bigg \{}  \left(\frac

1070:   {\omega} {kv}\right)^2 - \frac{\omega}{kv}\left[1 - \left(\frac{\omega} {kv}\right)^2\right] 

1071:   \nonumber \\

1072:  &  \hspace*{0cm}  & \hspace{1cm} \times \left[ \frac{1}{ 2} \log \left(-\frac

1073:   {1 + \omega/kv}{1 - \omega/kv}\right)\right] {\Bigg \}}.

1074:    \label{eq:worked out suscept1}

1075: \end{eqnarray}

1076: Note that the logarithmic function will give an imaginary part when

1077: $\omega_r/kc\beta < 1$. This implies that waves whose phase velocity,

1078: $\omega_r/k$, is small compared to the thermal speed of the background

1079: particles, $c\beta$, will be damped (also see \S\ref{sec:theory}). We

1080: pull this imaginary component out of the equation, which makes this damping

1081: more explicit, to find:

1082: \begin{eqnarray}

1083:   4 \pi \chi  &  = & \frac {2\pi\omega_{p,{\rm NR}}^2}{\omega^2n}  \int  p^2dp \frac {df_0}{dp}  v{\Bigg \{}  \left(\frac

1084:   {\omega} {kv}\right)^2 - \frac{\omega}{kv}\left[1 - \left(\frac{\omega} {kv}\right)^2\right] 

1085:   \nonumber \\

1086:  &  \hspace*{0cm}  & \hspace{1cm} \times \left[ \frac{1}{ 4} \log \left(\frac

1087:   {1 + \omega/kv}{1 - \omega/kv}\right)^2 - 

1088:    i\frac {\pi}{2}

1089:   \Theta\left(k^2 v^2 - \omega_r^2 \right)\right] {\Bigg \}}

1090:   , \label{eq:worked out suscept}

1091: \end{eqnarray}

1092: where $\Theta$ is the unit step function (i.e., $\Theta(z) = 1$ for

1093: $\Re(z) \geq 0$ and $\Theta(z) = 0$ for $\Re(z) < 0$).

1094: Equation (\ref{eq:worked out suscept}) is precisely Mikhailovskii

1095: (1979)'s result.  

1096: We now apply a three dimensional relativistic Maxwellian $f_0 \propto

1097: \exp\left(-{E}/kT\right)$, which is appropriately normalized, $\int

1098: d^3p f_0 = n$, and solve equation (\ref{eq:worked out suscept})

1099: numerically.  For the ultrarelativistic case $v\approx c$, we find a

1100: simple form in the limit $\omega_r \ll kc$:

1101: \begin{equation}\label{eq:3dchi}

1102: 4\pi\chi \approx i\frac {\pi} 4 \frac {\omega_{p}^2}{|k|c\omega}. 

1103: \end{equation}

1104: 

1105: \subsection{Two Dimensions}\label{sec:2dresult}

1106: 

1107: Starting from equation (\ref{eq:susceptibility 3d}), we assume $f_0 =

1108: f(p_{2d})g(p_z)$, where $p_{2d} = \sqrt{p_x^2 + p_y^2}$ and perform

1109: the integral over $p_z$.  The resulting two dimensional analogue of

1110: equation (\ref{eq:susceptibility 3d}) is 

1111: \begin{equation}\label{eq:susceptibility 2d}

1112:   4 \pi \chi =  \frac {\omega_{p,{\rm NR}}^2}{\omega^2} 

1113:   \int v_x \left(\frac {d} {dp_x} + \frac {kv_x}{\omega - kv_y} 

1114:     \frac {d}{dp_y}\right)\frac {f} n d^2p,

1115: \end{equation}

1116: where we have dropped the subscript ``$2d$'' from $p$.  Defining $p_x =

1117: p\cos\theta$, $p_y = p\sin\theta$, and similarly for $v_x$ and $v_y$,

1118: we find:

1119: \begin{equation}

1120:   4\pi\chi= \frac {\omega_{p,{\rm NR}}^2}{\omega^2 n}

1121:   \int pdp \int_0^{2\pi} d\theta v\cos^2\theta \left(1 + \frac

1122:   {\sin\theta}{(\omega/kv) - \sin\theta} \right)\frac{df}{dp}.

1123: \end{equation}

1124: We may transform the $\theta$ integral from $0$ to $2\pi$ to a contour

1125: integral over the unit circle by making the appropriate substitutions

1126: (Carrier, Krook, \& Pearson 1983)

1127: \begin{eqnarray}

1128: \cos\theta &\rightarrow& \frac 1 2\left(z + z^{-1}\right), 

1129: \sin\theta \rightarrow \frac 1 {2i}\left(z - z^{-1}\right) \\

1130: d\theta &\rightarrow& \frac {dz} {iz}, 

1131: \int_0^{2\pi} \rightarrow \int_{\Gamma},

1132: \end{eqnarray}

1133: where $\Gamma$ is the unit circle.  After a bit of algebra, we find

1134: \begin{equation}

1135:   4\pi\chi= \frac {\omega_{p,{\rm NR}}^2}{\omega^2n}

1136:   \int v \frac{df}{dp} pdp \int_{\Gamma} \frac {-i} 4 dz 

1137:   \left(z^2 + 2 + z^{-2}\right)\frac

1138:   {(2i\omega/kv)}{(2i\omega/kv)z - z^2 + 1}.

1139: \end{equation}

1140: The singular points in this equation are $z_{\pm} = (i\omega/kv) \mp

1141: \sqrt{1 -(\omega/kv)^2}$.  We perform the contour integral by noting

1142: that only the $z_-$ root contributes for $(\omega/kv)^2 < 1$:

1143: \begin{equation}\label{eq:2-d susceptibility}

1144:   4\pi\chi = -i\frac {2\pi\omega_{p,{\rm NR}}^2}{\omega k n}

1145:   \int \sqrt{1 - \left(\frac{\omega}{kv}\right)^2} \frac{df}{dp} pdp.

1146: \end{equation}

1147: We apply a two dimensional relativistic Maxwellian $f \propto

1148: \exp\left(-{E}/kT\right)$, where $f$ is appropriately normalized, i.e.

1149: $\int f d^2p = n$.  Expanding to lowest order in $\omega/kv$, we find:

1150: \begin{equation}\label{eq:2-d result}

1151: 4\pi\chi \approx i\frac {\omega_{p}^2}{|k|c\omega}.

1152: \end{equation}

1153: 

1154: 

1155: \section{Decay of an Initial Field of Fluctuations}\label{sec:fluctuation dissipation theorem}

1156: 

1157: The plasma susceptibilities (eq.[\ref{eq:chi_limit}]) define the

1158: linear response of the plasma.  These susceptibilities are complex and

1159: hence the electric permittivity $\epsilon = 1 + 4\pi\chi$ is also

1160: complex.  The implications of Poynting's theorem for the propagation 

1161: of small amplitude waves and fluctuations in such a medium have been

1162: described in many texts and monographs (e.g., Bekefi 1966, Melrose and

1163: McPhedran 1991, Stix 1992).  For completeness, we give a brief discussion

1164: of the theory behind expression (\ref{eq:damping}).

1165: 

1166: We begin with Poynting's theorem:

1167: \begin{equation}

1168: \frac{\partial}{\partial t} \left(\frac{\delta B^2 + \delta E^2}{8\pi}

1169:      \right) + {\boldsymbol \nabla}\cdot{\boldsymbol S} = - {\delta

1170:      \boldsymbol j} \cdot \delta {\boldsymbol E},

1171: \end{equation}

1172: where ${\boldsymbol S} = (c/4\pi){\boldsymbol E}\times{\boldsymbol B}$

1173: is the Poynting vector.  We assume field energy is distributed

1174: uniformly in our uniform plasma, so ${\boldsymbol

1175: \nabla}\cdot{\boldsymbol S} = 0$, i.e., there is no transport of

1176: energy spatially. In addition, for nonpropagating Weibel modes, $\delta

1177: E \ll \delta B$.  Thus, we find a simplified expression:

1178: \begin{equation}

1179: \frac{\partial}{\partial t} \left(\frac{\delta B^2}{8\pi} \right) = -

1180:      {\delta \boldsymbol j} \cdot \delta {\boldsymbol E}.

1181: \end{equation}

1182: Writing the fields with truncated amplitudes

1183: \begin{equation}  \delta B_{TV} (  {\boldsymbol r}, t)  = \left\{

1184:      \begin{array}{rl}

1185:     \delta B, \; t \in (-T/2, T/2), \;  |{\boldsymbol r}|   \in  V, \\

1186:     0, \; t \not\in (-T/2, T/2), \;  |{\boldsymbol r}|  \not\in  V,   

1187:    \end{array} 

1188:    \right.  

1189: \end{equation}

1190: with $\delta B_{TV} = 0$ when the coordinates are outside of the

1191: volume, $V$, and time is outside of the interval $(-T/2, T/2)$. Thus,

1192: we can define space-time averages of the fields while still expressing

1193: them in terms of convergent Fourier transforms.  We write the Fourier

1194: transforms as

1195: \begin{equation}

1196:  \delta j_{k\omega} = \int d^3k \int_{-\infty}^{\infty}d\omega\,\delta j_{TV} 

1197: \exp\left(i{\boldsymbol k}\cdot{\boldsymbol r} - i\omega t\right).

1198: \end{equation}

1199: We express the field energy density and the $\delta j \cdot \delta E$

1200: work in terms of the Fourier amplitudes and then average over time and

1201: space. Taking $T$ and $V$ to $\infty$, we integrate over the resulting

1202: $\delta$-functions to find

1203:  \begin{equation}

1204: \frac{\partial}{\partial t} \left \langle \frac{\delta B^2}{8\pi} \right \rangle  = 

1205:         \int d^3k\int d\omega\frac 1 2 

1206:        \langle \delta j_{k\omega } \delta E_{k\omega}^* + c.c. \rangle,

1207: \end{equation}

1208: where $\langle \rangle$ represents the space time average over the

1209: fluctuation wavelengths and variability times and $c.c.$ is the

1210: complex conjugate.  We apply the same Fourier transform to $\delta B$

1211: and apply the same average over time and space.  We use the linear

1212: response (eq.[\ref{eq:j2E}]) $\delta E_{k\omega} = (i / \chi \omega)

1213: \delta j_{k\omega}$ to find

1214: \begin{equation}

1215: \frac 1 {8\pi}\frac{\partial \langle |\delta B_{k\omega}|^2 \rangle}{\partial t}  

1216: = -\Im\left(\omega \chi\right)^{-1} \langle |\delta j_{k\omega} |^2 \rangle. 

1217: \end{equation}

1218: Now using $\delta j_{k\omega} = ikc \delta B_{k\omega}/4\pi$,

1219: we find

1220: \begin{equation}\label{eq:app_damping}

1221: \frac{\partial  \langle |\delta B_{k\omega}|^2 \rangle }{\partial t} 

1222:   = -2 \gamma_{k\omega} \langle |\delta B_{k\omega} |^2 \rangle,

1223: \end{equation}

1224: where $\gamma_{k\omega}$ is the damping rate or

1225: \begin{equation}\label{eq:app_general gamma}

1226:   \gamma_{k\omega} = \frac {\left(kc\right)^2}{\omega} \Im\left( 4\pi\chi\right)^{-1}.    

1227: \end{equation}

1228: Equation (\ref{eq:app_damping}) is a specific form of the

1229: fluctuation-dissipation theorem (Thompson and Hubbard 1960) from which

1230: the same result can be derived.

1231:  

1232: Equations (\ref{eq:app_damping}) and (\ref{eq:app_general gamma})

1233: define the evolution of magnetic energy in terms of the linear

1234: response.  For notational simplicity we reduce $k\omega \rightarrow k$

1235: in the rest of the text.  We obtain simple forms for $\gamma_k$, using

1236: the asymptotic forms of $4\pi\chi$ from equation (\ref{eq:chi_limit}) for

1237: 3D and 2D. We find for $kc \ll \omega_p$:

1238: \begin{equation}\label{eq:app_gamma}

1239: \gamma_{k} = \left\{\begin{array}{ll} \frac {|kc|^3} {\omega_{p}^2} & \textrm{2D} 

1240: \\ \frac {4} {\pi} \frac{|kc|^3} {\omega_{p}^2}& \textrm{3D}

1241: \end{array}\right. .

1242: \end{equation}

1243: The cubic dependence on wavenumber in equation (\ref{eq:app_gamma})

1244: suggests short wavelength modes are very strongly damped, while longer

1245: wavelength modes can survive much longer.  

1246: %In addition, the 2D and 3D

1247: %damping rates are the same up to a overall multiplicative constant of

1248: %order unity.

1249: 

1250: Taking the general form of $4\pi\chi$ for 2D and 3D from equation

1251: (\ref{eq:worked out suscept}) and (\ref{eq:2-d susceptibility}), we

1252: numerically compute the damping rate from equation

1253: (\ref{eq:app_general gamma}) for Weibel modes where $\omega_r = 0$, because of the

1254: non-propagating nature of the downstream magnetic clumps. We

1255: show the results in Figure \ref{fig:damping rates}. Also

1256: plotted are the simple forms for $\gamma_k$ from equation

1257: (\ref{eq:app_gamma}).  Though, the asymptotic forms are only valid for

1258: $kc/\omega_{p} \ll 1$, the numerical and asymptotic results are in

1259: excellent agreement throughout.  Thus for simplicity, we use equation

1260: (\ref{eq:app_gamma}) (also eq.[\ref{eq:gamma}] in the main body) for

1261: the damping rates.

1262: 

1263: \begin{figure}[H]

1264: \plotone{f6.eps}

1265: \caption{Damping rates of magnetic field as a function of $kc/\omega_{p}$.  We plot

1266:   the damping rates numerically computed from equation

1267:   (\ref{eq:worked out suscept}) and (\ref{eq:2-d susceptibility}) for 3D

1268:   (thick solid line) and 2D (thin solid line) respectively.  Also

1269:   overplotted are the analytic forms of these expressions from

1270:   equation (\ref{eq:gamma}) for 2D (thin dashed line) and 3D (thick

1271:   dashed line).}

1272: \label{fig:damping rates}

1273: \end{figure} 

1274: 

1275: %We find excellent agreement between the numerical evaluation of

1276: %$\gamma_k$ for $\omega_r = 0$ and equation (\ref{eq:app_gamma}) (see

1277: %Appendix A), so we use equation (\ref{eq:app_gamma}) in the rest of this

1278: %paper.

1279:  

1280: \begin{references}

1281: 

1282: \reference{}

1283: Arons, J., Norman, C. A., \& Max, C. E. 1977, Phys. Fluids, 20, 1302

1284: 

1285: %\reference{}

1286: %Barnes, A. 1968, \apj, 154, 751

1287: 

1288: \reference{}

1289: Bekefi, G. 1966, ``Radiation Processes in Plasmas'' (John Wiley: New York)

1290: 

1291: \reference{}

1292: Bell, A. R. 1978, \mnras, 182, 147

1293: 

1294: \reference{}

1295: Bell, A. R. 2004, \mnras, 353, 550

1296: 

1297: \reference{}

1298: Bell, A. R. 2005, \mnras, 358, 181

1299: 

1300: \reference{}

1301: Birdsall, C. K. \& Langdon, A. B. 1991, ``Plasma Physics via Computer Simulations'' (McGraw-Hill: New York)

1302: 

1303: \reference{}

1304: Buneman, O. 1993 in ``Computer Space Plasma Physics'', Terra Scientific, Tokyo, 67

1305: 

1306: \reference{}

1307: Carrier, G. F., Krook, M., \& Pearson, C. E. 1983, ``Functions of a Complex Variable : Theory and Technique'' (Hod Books: Ithaca)

1308: 

1309: \reference{}

1310: Coroniti, F.~V. 1990, \apj, 349, 538

1311: 

1312: \reference{}

1313: Davidson, R. C., Hammer, D. A., Haber, I., \& Wagner, C. E. 1972, Phys. Fluids, 15, 317

1314: 

1315: \reference{}

1316: Frederiksen, J. T., Hededal, C. B., Haugb\`{o}lle, T., \& Nordlund, �Å.

1317: 2004, \apj, 608, L13

1318: 

1319: \reference{}

1320: Gallant, Y. A., Hoshino, M., Langdon, A. B., Arons, J., \& Max, C. E. 1992, \apj, 391, 73

1321: 

1322: \reference{}

1323: Gruzinov, A. 2001a, submitted to \apj, astro-ph/0111321

1324: 

1325: \reference{}

1326: Gruzinov, A. 2001b, \apj, 563, L15

1327: 

1328: \reference{}

1329: Gruzinov, A. \& Waxman E. 1999, \apj, 511, 852

1330: 

1331: \reference{}

1332: Hammett, G. W., Dorland, W., \& Perkins, F. W. 1992, Phys. Fluids B, 4, 2052

1333: 

1334: \reference{}

1335: Hededal, C. B., Trier Frederiksen, J., Haugboelle, T., \& Nordlund,

1336: A. 2005, Neutrinos and Explosive Events in the Universe, Proceedings of

1337: the 14th Course of the International School of Cosmic Rays

1338: Astrophysics, a NATO Advanced Study Institute, held in Erice, Italy,

1339: 2-13 July 2004. Edited by Maurice M. Shapiro, Stanev Todor, and John

1340: P. Wefel.

1341: 

1342: \reference{}

1343: Jackson, J.~D. 1999, ``Classical Electrodynamics'' (John Wiley \& Sons: New York)

1344: 

1345: \reference{}

1346: Kato, T. N. 2005, Phys. Plasmas, 12, 80705

1347: 

1348: \reference{}

1349: Katz, B., Keshet, U., \& Waxman, E. 2007, \apj, 655, 375

1350: 

1351: \reference{}

1352: Medvedev, M.~V. \& Loeb, A. 1999, \apj, 526, 697

1353: 

1354: \reference{}

1355: Medvedev, M. V., Fiore, M., Fonseca, R. A., Silva,

1356: L. O.; Mori, W. B. 2005, \apj, 618, L75

1357: 

1358: \reference{}

1359: Melrose, D.B., and McPhedran, R.C. 1991, ``Electromagnetic Processes in

1360:  Dispersive Media'' (Cambridge: Cambridge University Press)

1361: 

1362: \reference{}

1363: Milosavljevic, M., Nakar, E., \& Spitkovsky, A. 2006, \apj, 637, 765

1364: 

1365: \reference{}

1366: Milosavljevic, M. \& Nakar, E. 2006a, \apj, 641, 978

1367: 

1368: \reference{}

1369: Milosavljevic, M. \& Nakar, E. 2006b, \apj, 651, 979

1370: 

1371: %\reference{}

1372: %Moiseev, S.~S., \& Sagdeev, R.~Z. 1963, J. Nucl. Energy C, 5, 43

1373: 

1374: \reference{}

1375: Mikhailovskii, A. B. 1979, Plasma Phys., 22, 133

1376: 

1377: \reference{}

1378: Nishikawa, K.-I., Hardee, P., Richardson, G., Preece, R., Sol, H., \&

1379: Fishman, G. J. 2003, \apj, 595, 555

1380: 

1381: \reference{}

1382: Nishikawa, K.-I., Hardee, P., Richardson, G., Preece, R., Sol, H., \&

1383: Fishman, G. J. 2005, \apj, 622, 927

1384: 

1385: \reference{}

1386: Pe'er, A. \& Zhang, B. 2006, \apj, 653, 454 

1387: 

1388: \reference{}

1389: Piran, T. 2005a, Rev. of Modern Phys., 76, 1143

1390: 

1391: \reference{}

1392: Piran, T. 2005b, Magnetic Fields in the Universe, Angra dos Reis,

1393: Brazil, Nov. 29-Dec 3, 2004, Ed. E. de Gouveia del Pino, AIP

1394: Conference Proceedings, v784 (New York:AIP), p164

1395: 

1396: \reference{}

1397: Rossi, E. \& Rees, M. J. 2003, \mnras, 339, 881

1398: 

1399: \reference{}

1400: Silva, L. O., Fonseca, R. A., Tonge, J. W., Dawson, J. M., Mori,

1401: W. B., \& Medvedev, M. V. 2003, \apj, 596, 121

1402: 

1403: \reference{}

1404: Spitkovsky, A. 2005, AIP Conf. Proc, 801, 345; astro-ph/0603211

1405: 

1406: \reference{}

1407: Spitkovsky, A. 2007, submitted to \apjl; arXiv:0706.3126

1408: 

1409: \reference{}

1410: Stix, T.H. 1992, ``Waves in Plasmas'' (American Institute of Physics: New York)

1411: 

1412: \reference{}

1413: Thompson, W. B. \& Hubbard, J. 1960, Rev. of Modern Phys., 32, 714

1414: 

1415: \reference{}

1416: Weibel, E.~S. 1959, \prl, 2, 83

1417: 

1418: %\reference{}

1419: %Wright, T. P. \& Hadley, G. R. 1975, \pra, 12, 686

1420: 

1421: \reference{}

1422: Yoon, P.~H., \& Davidson, R.~C. 1987, \pra, 35, 2718

1423: 

1424: \end{references}

1425:  

1426: %\clearpage

1427: 

1428: 

1429: 

1430: 

1431: \end{document}