1: \documentstyle[prl,aps,multicol,epsfig,array]{revtex}
2:
3: \begin{document}
4: \draft
5:
6: \title{Spontaneous Branching of Anode-Directed Streamers
7: between Planar Electrodes}
8: \author{Manuel Array\'as$^{1,2}$\cite{Madrid},
9: Ute Ebert$^1$ and Willem Hundsdorfer$^1$}
10: \address{
11: $^1$CWI, P.O.Box 94079, 1090 GB Amsterdam, The Netherlands,
12: }
13: \address{
14: $^2$Instituut--Lorentz, Universiteit Leiden, P.O.Box 9506, 2300 RA Leiden,
15: The Netherlands }
16: \date{revised version from March 10, 2002 of paper submitted to PRL on Nov. 16, 2001}
17: \maketitle
18:
19: \begin{abstract}
20: Non-ionized media subject to strong fields can become locally
21: ionized by penetration of finger-shaped streamers.
22: We study negative streamers between planar electrodes in a simple
23: deterministic continuum approximation. We observe that
24: for sufficiently large fields, the streamer tip can split.
25: This happens close to the limit of ``ideal conductivity''.
26: Qualitatively the tip splitting is due to a Laplacian instability
27: quite like in viscous fingering. For future quantitative
28: analytical progress, our stability analysis of planar fronts
29: identifies the screening length as a regularization mechanism.
30: \end{abstract}
31:
32: %\pacs{47.54.+r, 52.80.Mg, 51.50.+v, 05.45.-a}
33:
34: \begin{multicols}{2}
35:
36: Streamers commonly appear in dielectric breakdown
37: when a sufficiently high voltage is suddenly applied
38: to a medium with low or vanishing conductivity.
39: They consist of extending fingers of ionized matter and
40: are ubiquitous in nature and technology \cite{Rai,Eddie}.
41: The degree of ionization inside a streamer is low, hence
42: thermal or convection effects are negligible. However,
43: streamers are nonlinear phenomena due
44: to the space charges inside the ionized body that modify
45: the externally applied electric field.
46: While in many applications, streamers by a strongly non-uniform
47: background electric field are forced to propagate towards
48: the cathode through complex mixtures of gases
49: \cite{Eddie,Kuli,DBM}, we here investigate the basic phenomenon
50: of the primary anode-directed streamer in a simple non-attaching
51: and non-ionized gas and in a uniform background field as
52: in the pioneering experiments of Raether
53: \cite{Raether}. In previous theoretical work, it is implicitly
54: assumed that streamers in a uniform background
55: field propagate in a stationary manner \cite{Firsov,Dya,Raizer2}.
56: This view seems to be supported by previous simulations \cite{DW,Vit}.
57:
58: In this paper we present the first numerical evidence that anode directed
59: (or negative) streamers do branch even in a uniform background field
60: and without initial background ionization in the minimal fully deterministic
61: ``fluid model'' \cite{Rai,Firsov,Dya,Raizer2,DW,Vit},
62: if the field is sufficiently strong. We argue that this happens
63: when the streamer approaches what we suggest to call the Lozansky--Firsov
64: limit of ``ideal conductivity'' \cite{Firsov}.
65: The streamer then can be understood as an interfacial pattern with
66: a Laplacian instability \cite{Ute}, qualitatively similar
67: to other Laplacian growth problems \cite{Ivantsov}.
68: For future quantitative analytical progress, we identify the
69: electric screening length as a relevant regularization mechanism.
70: Our finding casts doubts on the existence of a stationary mode of
71: streamer propagation with a fixed head radius.
72:
73: \end{multicols}
74:
75: \begin{figure}
76: \label{fig1}
77: \begin{center}
78: \epsfig{figure=fig11.eps,width=0.2\linewidth}
79: %0.222 .217 .21\linewidth}
80: \epsfig{figure=fig12.eps,width=0.18\linewidth}
81: %0.2 .195 .19\linewidth}
82: \epsfig{figure=fig13.eps,width=0.18\linewidth}
83: %0.2\linewidth}
84: \epsfig{figure=fig14.eps,width=0.18\linewidth}
85: %0.2\linewidth}
86: \caption[]{Evolution of spontaneous branching of anode directed streamers
87: in a strong homogeneous background field at times $t=300$, 365, 420 and 450.
88: Model, initial and boundary conditions are discussed in the text.
89: The planar cathode is located at $z=0$ and the planar anode at $z=2000$
90: (shown is $0\le z\le 1400$). The radial coordinate extends from the origin
91: up to $r=2000$ (shown is $0\le r\le 600$). The thin lines denote
92: levels of equal electron density $\sigma$ with increments of 0.1 or 0.2
93: as indicated by the labels. The thick lines denote the higher electron
94: density levels 1., 2., 3., 4., 5.\ and 6. These high densities appear
95: only at the last time step $t=450$ in the core of the new branches.}
96: \end{center}
97: \end{figure}
98:
99: \begin{multicols}{2}
100:
101: We investigate the {\it minimal streamer model}, i.e., a ``fluid
102: approximation'' with local field-dependent impact ionization reaction
103: in a non-attaching gas like argon or nitrogen
104: \cite{Rai,Firsov,Dya,Raizer2,DW,Vit,Ute}.
105: In detail, the dynamics is as follows: \\
106: ({\it i}) an impact ionization reaction in local field approximation:
107: free electrons and positive ions are generated by impact of
108: accelerated electrons on neutral molecules
109: $\partial_\tau n_e+\nabla_{\bf R}\cdot{\bf j}_e=
110: \partial_\tau n_i+\nabla_{\bf R}\cdot{\bf j}_i=
111: |\mu_e{\cal E}n_e|\;\alpha_0\;\alpha(|{\cal E}|/E_0)$;
112: $n_{e,i}$ and ${\bf j}_{e,i}$ are particle densities or currents
113: of electrons or ions, respectively, and ${\cal E}$ is the electric field;
114: in all numerical work, we use the Townsend approximation
115: $\alpha_0\;\alpha(|{\cal E}|/E_0)=\alpha_0\;\exp(-E_0/|{\cal E}|)$
116: with parameters $\alpha_0$ and $E_0$ for the effective cross-section.
117: \\
118: ({\it ii}) drift and diffusion of the charged particles in the local
119: electric field ${\bf j}_e=-\mu_e{\cal E}n_e-D_e\nabla_{\bf R}n_e$,
120: where in anode-directed streamers the mobility of the ions actually
121: can be neglected because it is more than two orders of magnitude
122: smaller than the mobility $\mu_e$ of the electrons, so ${\bf j}_i=0$,
123: \\
124: ({\it iii}) the modification of the externally applied electric field
125: through the space charges of the particles according to the Poisson equation
126: $\nabla_{\bf R}\cdot{\cal E}=e(n_i-n_e)/\epsilon_0$.
127: It is this coupling between space charges and electric field
128: which makes the problem nonlinear.
129:
130: The natural units of the model are given by the ionization length
131: $R_0=\alpha_0^{-1}$,
132: the characteristic impact ionization field $E_0$, and
133: the electron mobility $\mu_e$ determining the velocity $v_0=\mu_e E_0$
134: and the time scale $\tau_0=R_0/v_0$.
135: Hence we introduce the dimensionless coordinates \cite{Ute}
136: ${\bf r}={\bf R}/R_0$ and $t=\tau/\tau_0$, the dimensionless field
137: ${\bf E}={\bf {\cal E}}/E_0$, the dimensionless electron
138: and ion particle densities $\sigma=n_e/n_0$ and $\rho=n_i/n_0$
139: with $n_0=\varepsilon_0 E_0/(e R_0)$,
140: and the dimensionless diffusion constant $D=D_e/(R_0v_0)$.
141: After this rescaling, the model has the form:
142: \begin{eqnarray}
143: \label{1}
144: \partial_t\;\sigma \;-\;
145: \nabla\cdot\left(\sigma\;{\bf E} + D\;\nabla\sigma\right)
146: &=& \sigma \; f(|{\bf E}|)~,
147: \\
148: \label{2}
149: \partial_t\;\rho \;
150: &=& \sigma \; f(|{\bf E}|)~,
151: \\
152: \label{3}
153: \rho - \sigma &=& \nabla\cdot{\bf E}~~,~~{\bf E}=-\nabla \Phi~,
154: \\
155: \label{5}
156: f(|{\bf E}|)=|{\bf E}|\;\alpha(|{\bf E}|)~
157: &\Big(=&|{\bf E}|\;e^{-1/|{\bf E}|}~\mbox{in sim.}\Big)~.
158: \end{eqnarray}
159:
160: In the simulations presented here, a planar cathode is
161: located at $z=0$ and a planar anode at $z=2000$. The stationary
162: potential difference between the electrodes $\Delta\Phi=1000$
163: corresponds to a uniform background field ${\bf E} = -0.5 \;{\bf e}_z$
164: in the $z$ direction. For nitrogen under normal conditions with
165: effective parameters as in \cite{DW,Vit}, this corresponds to
166: an electrode separation of $\approx$ 5 mm and a potential
167: difference of $\approx$ 50 kV. The unit of time $\tau_0$
168: is $\approx3$ ps, and the unit of field $E_0$ is $\approx 200$ kV/cm.
169: We used $D=0.1$ which is appropriate for nitrogen, and we assumed
170: cylindrical symmetry of the streamer. The radial coordinate
171: extends from the origin up to $r=2000$ to avoid lateral
172: boundary effects on the field configuration.
173: As initial condition, we used an electrically neutral
174: Gaussian ionization seed on the cathode
175: \begin{equation}
176: \label{6}
177: \sigma(r,z,t=0)=\rho(r,z,t=0)
178: =10^{-6}\; e^{-(z^2+r^2)/100^2}.
179: \end{equation}
180: The parameters of our numerical experiment are essentially
181: the same as in the earlier simulations of Vitello {\it et al.}
182: \cite{Vit}, except that our background electric field is twice
183: as high; the earlier work had 25 kV applied over a gap of 5 mm.
184: This corresponded to a dimensionless background field of 0.25,
185: and branching was not observed.
186:
187: In Fig.~1 we show the electron density levels at four time
188: steps of the evolution in the higher background field of 0.5.
189: We observe that at time $t=420$, the streamer develops
190: instabilities at the tip. At time $t=450$, these instabilities
191: have grown out into separate fingers. Because of the imposed
192: cylindrical geometry, the further evolution after branching
193: ceases to be physical. On the other hand, the main effect
194: of the unphysical symmetry constraint is to suppress all linear
195: instability modes that are not cylindrically symmetric. Hence in
196: a fully 3D system, the instability will develop even earlier
197: than here.
198:
199: Further simulations show:
200: $(a)$ branching does not occur in a system of the same size
201: in the lower background field of 0.25, in agreement with \cite{Vit}.
202: $(b)$ Branching is not due to the proximity of the anode,
203: since in a system with twice the electrode separation
204: (with the anode at $z=4000$) and with twice the potential difference
205: ($\Delta\Phi=2000$) --- so with the same background field ---,
206: the streamer branches in about the same way after about the same time
207: and travel distance.
208: $(c)$ The phenomenon is not specific to the particular initial condition (5).
209: $(d)$ Branching does somewhat depend on the numerical
210: discretization. A wider numerical mesh leads to a higher effective
211: noise level; and the branching then is triggered somewhat earlier.
212: $(e)$ Occasionally,
213: we observe a different tip splitting mode. In Fig.~1 at time
214: $t=450$, the finger on the axis develops the strongest with
215: $\sigma$ exceeding 6, while in the fingers off the axis,
216: $\sigma$ stays below 3. In the other branching mode,
217: the first finger off the axis outruns the finger on the axis.
218:
219: Before we discuss the physical nature of the instability, we explain
220: our numerical approach: we used uniform space-time grids with
221: a spatial mesh of $1000 \times 1000$.
222: The spatial discretization is based on local mass balances.
223: The diffusive fluxes are approximated in standard fashion with
224: second order accuracy. For the convective fluxes a third order upwind-biased
225: formula was chosen to reduce the numerical oscillations that are common
226: with second order central fluxes. Such oscillations can be completely
227: avoided, e.g., by flux-limiting, but preliminary tests showed
228: that the upwind-biased formula already gives sufficient numerical
229: monotonicity and is much faster. Time stepping is based on an explicit
230: linear 2-step method, where at each time step the Poisson equation is
231: solved by the {\sc FISHPACK} routine. References for these procedures
232: can be found in \cite{Wess}.
233:
234: \end{multicols}
235:
236: \begin{figure}
237: \label{fig2}
238: \begin{center}
239: \epsfig{figure=fig21.eps,width=0.40\linewidth}
240: \epsfig{figure=fig22.eps,width=0.41\linewidth}
241: \caption[]{A zoom into the head of the streamer from Fig.\ 1
242: at the first two time steps. The aspect ratio is equal and
243: the axis scaling identical at both times. The thin lines
244: are the levels of equal electron density as in Fig.\ 1.
245: The thick lines are electrical equipotential lines in steps
246: of $\Delta\Phi=12$.}
247: \end{center}
248: \end{figure}
249:
250: \begin{multicols}{2}
251:
252: To understand now why and at which stage the streamer develops
253: a tip splitting instability, in Fig.\ 2, we zoom into the streamer
254: head. Shown are the first two time steps from Fig.~1 with
255: the electron density levels again as thin lines, and additionally
256: with the equipotential lines as thick lines.
257: One observes that during the temporal evolution prior to branching,
258: both the curvature and the thickness of the ionization front decrease.
259: So the width of the front becomes much smaller than its radius of
260: curvature, and an interface approximation becomes increasingly justified.
261: The electric field inside the streamer head also decreases, so that
262: the ionization front more and more coincides with an equipotential surface.
263: In summary, the ionization front evolves towards a weakly curved and
264: almost equipotential moving ionization boundary. At the same time,
265: the electric field immediately ahead of the streamer increases.
266:
267: We argue now that a transient stage of an approximately equipotential
268: and weakly curved ionization boundary leads to tip splitting. Conversely,
269: we argue that tip splitting in the lower background field of 0.25 is
270: not observed within the presently and previously \cite{Vit} investigated
271: gap lengths because the transient stage of Fig.\ 2 is not reached
272: before the streamer reaches the anode.
273:
274: In fact, the streamer in Fig.\ 2 approaches the limit of ``ideal
275: conductivity'': the conducting body has $\Phi\approx {\rm const.}$,
276: while in the non-ionized
277: region $\nabla^2\Phi=0$ due to the absence of space charges.
278: The boundary between the two regions moves approximately with the
279: drift velocity $v_f\propto\nabla\Phi$ or with the diffusion
280: corrected velocity $v_f\propto\nabla\Phi\;
281: \left(1+2\sqrt{D\;\alpha(|\nabla\Phi|)/|\nabla\Phi|}\right)$
282: \cite{Ute}. Our simulations are the first numerical
283: evidence that the ``ideal conductivity'' limit can be approached
284: within our model.
285:
286: This limit of ideally conducting streamers in an electric field
287: that becomes uniform far ahead of the front was studied by Lozansky
288: and Firsov
289: \cite{Firsov}. They realized that uniformly propagating paraboloids
290: of arbitrary radius of curvature are solutions of this problem.
291: They did not realize that these paraboloids are mathematically equivalent
292: to the Ivantsov paraboloids \cite{Ivantsov} of dendritic growth found
293: earlier.
294: The uniformly propagating Ivantsov paraboloids in the early 80'ies were
295: identified as dynamically unstable. This is generally the case for
296: such so-called Laplacian growth problems without a regularization mechanism.
297: Since ideally conducting streamers also pose such
298: a Laplacian growth problem \cite{Ute}, the dynamical instability
299: of the structure shown in Fig.\ 2 can be expected,
300: and it actually occurs as can be seen in Fig.\ 1. This
301: explains qualitatively why tip splitting occurs.
302:
303: For a quantitative analysis, a system specific regularization
304: mechanism has to be found \cite{Ute,Ivantsov}.
305: Its identification is intricate because negative streamer fronts are
306: so-called pulled fronts whose dynamics is dominated by the leading edge
307: rather than the nonlinear interior of the front \cite{pulled}. Therefore
308: standard methods like the pertubative derivation
309: of a moving boundary approximation for the model (\ref{1})--(\ref{5})
310: does not work \cite{MBA}. (Pulling also implies
311: that standard numerical methods with adaptive grids are inefficient.)
312: However, the ionization front has two intrinsic length scales,
313: a diffusion length and an electric screening length. We therefore
314: explore the approximation of $D=0$. It is smooth for the velocity of
315: negative fronts \cite{Ute} and eliminates the leading edge,
316: and hence suppresses the pulled nature of the front.
317: Rather the front becomes a shock front for the electron
318: density, while the intrinsic length scale of the electric screening
319: layer behind the shock remains.
320:
321: As a first step to understand the short wave length regularization
322: of perturbations due to this screening length,
323: we have investigated the transversal instability modes
324: of a planar ionization front in the limit $D=0$ in a field
325: that approaches the uniform limit ${\bf E}=-E_\infty\;{\bf e}_z$
326: far ahead of the front. The planar unperturbed front propagates
327: with velocity $v=E_\infty$, which equals the drift velocity of the
328: electrons precisely at the shock front. The implicit analytical
329: front solution can be found in \cite{Ute}.
330: In a comoving frame $\xi=z-vt$, we denote it by
331: $\big(\;\sigma_0(\xi), \rho_0(\xi), \Phi_0(\xi)\;\big)$.
332: The Fourier components $\big(\;\tilde\sigma_k, \tilde\rho_k,
333: \tilde\Phi_k\;\big)$ of a transversal linear perturbation are
334: defined through
335: \begin{equation}
336: \sigma=\sigma_0(\xi)+\int dk\;\tilde\sigma_k(\xi)\;e^{ikx+st}+\ldots
337: ~~~\mbox{etc.}
338: \end{equation}
339: For the derivation of the boundary conditions on the shock front,
340: it is more convenient to write a single Fourier component as
341: $\sigma=\sigma_0\left(\xi-e^{ikx+st}\right)+\sigma_k(\xi)\;e^{ikx+st}+\ldots$.
342: With this ansatz and the auxiliary field $\psi_k=\partial_\xi\phi_k$,
343: the Fourier components solve the inhomogeneous equation
344: \begin{eqnarray}
345: \label{Matr1}
346: &&\partial_\xi \left(
347: \begin{array}{c}
348: \sigma_k \\
349: \rho_k \\
350: \psi_k \\
351: \phi_k
352: \end{array}\right)={\bf M}_{s,k}\cdot
353: \left(\begin{array}{c}\sigma_k\\ \rho_k\\ \psi_k\\ \phi_k\end{array}\right)
354: - \left(\begin{array}{c}s\;\partial_\xi\sigma_0/(v+E_0)\\
355: s\;\partial_\xi\rho_0/v\\ E_0k^2\\0
356: \end{array}\right) ,
357: \\
358: &&~~~ \nonumber \\
359: \label{Matr2}
360: &&{\bf M}_{s,k}=\left(\begin{array}{cccc}
361: \frac{s+2\sigma_0-f(E_0)-\rho_0}{v+E_0}&
362: \frac{-\sigma_0}{v+E_0}&
363: \frac{\partial_\xi\sigma_0-\sigma_0 f'(E_0)}{v+E_0}
364: &0 \\
365: -f(E_0)/v & s/v & -\sigma_0 f'(E_0)/v & 0 \\
366: 1 & -1 & 0 & k^2 \\
367: 0 & 0 & 1 & 0
368: \end{array}\right) .
369: \nonumber
370: \end{eqnarray}
371: The boundary conditions at the shock $\xi=0$ can be obtained from the
372: analytical solution in the non-ionized area, and from the boundedness
373: of the charge densities:
374: \begin{equation}
375: \label{Init}
376: \left(\begin{array}{c}\sigma_k\\ \rho_k\\ \psi_k\\ \phi_k\end{array}\right)
377: \stackrel{\xi\uparrow0}{\longrightarrow}
378: \left(\begin{array}{c}f'(v)/(1+s/f(v))\\ 0\\ 1\\
379: (vk-s)/(sk)\end{array}\right) .
380: \end{equation}
381: The other boundary conditions are obtained by imposing that at
382: $\xi\to-\infty$ the electric field decays and the densities become constant.
383:
384: These equations together with the boundary conditions define an
385: eigenvalue problem for $s=s(k,v)$ with $v=E_\infty$. It can be solved
386: numerically by shooting from $\xi=0$ towards $-\infty$. In agreement
387: with analytical limits --- details will be given elsewhere ---, we find
388: \begin{eqnarray}
389: \label{sk}
390: s(k)&=&\left\{\begin{array}{ll}
391: |E_\infty| \;k &~~\mbox{ for }k\ll\alpha(|E_\infty|)/2\\
392: |E_\infty|\;\alpha(|E_\infty|)/2 &~~\mbox{ for }k\gg\alpha(|E_\infty|)/2
393: \end{array}\right. .
394: \end{eqnarray}
395: This means that the electric screening length $1/\alpha(|E_\infty|)$
396: does regularize the instability of short wave length
397: perturbations from linear growth in $k$ to the saturation
398: value $s(k)=|E_\infty|\;\alpha(E_\infty)/2$. A small
399: positive growth rate remains, but the analytical derivation of
400: (\ref{sk}) hints to the unconventional possibility that sufficiently
401: curved fronts actually are stable to short wave length perturbations.
402: This question is presently under investigation. If true, it would
403: identify a most unstable wave length determining the width of the
404: fingers that emerge after tip splitting.
405:
406: In conclusion, we have presented numerical evidence that
407: anode-directed streamers in a sufficiently strong, but uniform field
408: can branch spontaneously even in a fully deterministic fluid model.
409: We have argued that this happens when the streamer approaches
410: the limit of ideal conductivity. We have established a qualitative
411: mathematical analogy with tip splitting of viscous fingers through
412: the concept of Laplacian growth, and we have analytically demonstrated
413: that the electric screening length leads to an unconventional
414: regularization. This opens the way to future quantitative analytical
415: progress.
416:
417: {\bf Acknowledgement:}
418: M.A.\ was supported by the EU-network ``Patterns, Noise, and Chaos''
419: and U.E.\ partially by the Dutch Science Foundation NWO.
420:
421: \begin{references}
422: \bibitem[*]{Madrid} New address: Universidad Rey Juan Carlos,
423: Escuela Superior de Ciencias Exp. y Tecnologia,
424: c. Tulipan s/n, 28933 Mostoles, Madrid, Spain
425: \bibitem{Rai} Y.P.\ Raizer, {\em Gas Discharge Physics}
426: (Springer, Berlin 1991).
427: \bibitem{Eddie} E.M. van Veldhuizen (ed.):
428: {\it Electrical discharges for environmental purposes:
429: fundamentals and applications}
430: (NOVA Science Publishers, New York 1999).
431: \bibitem{Kuli} A.A. Kulikovsky, J. Phys. D: Appl. Phys.
432: {\bf 33}, 1514 (2000); and Phys. Rev. E {\bf 57}, 7066 (1998).
433: \bibitem{DBM} L. Niemeyer, L. Pietronero,
434: H.J. Wiesmann, Phys. Rev. Lett. {\bf 52}, 1033 (1984);
435: A.D.O. Bawagan, Chem Phys. Lett. {\bf 281} 325 (1997).
436: \bibitem{Raether} H. Raether: Z. Phys. {\bf 112}, 464 (1939) (in German).
437: \bibitem{Firsov} E.D. Lozansky and O.B. Firsov, J. Phys. D:
438: Appl. Phys. {\bf 6}, 976 (1973).
439: \bibitem{Dya} M.I. D'yakonov, V.Y. Kachorovskii:
440: Sov. Phys. JETP {\bf 67}, 1049 (1988); and
441: Sov. Phys. JETP {\bf 68}, 1070 (1989).
442: \bibitem{Raizer2} E.M. Bazelyan, Yu.P. Raizer,
443: {\it Spark Discharges} (CRS Press, New York, 1998);
444: Yu.P. Raizer, A.N. Simakov, Plasma Phys. Rep.
445: {\bf 24}, 700 (1998).
446: \bibitem{DW} S.K.\ Dhali and P.F.\ Williams, Phys.\ Rev.\ A {\bf 31},
447: 1219 (1985) and J.\ Appl.\ Phys.\ {\bf 62}, 4696 (1987).
448: \bibitem{Vit} P.A.\ Vitello, B.M.\ Penetrante, and J.N.\ Bardsley,
449: Phys.\ Rev.\ E {\bf 49}, 5574 (1994).
450: \bibitem{Ute} U. Ebert, W. van Saarloos and C. Caroli,
451: Phys. Rev. Lett. {\bf 77}, 4178 (1996); and
452: Phys. Rev. E {\bf 55}, 1530 (1997).
453: \bibitem{Ivantsov} For a review and a collection of original articles,
454: see P. Pelc\'e: \emph{Dynamics of Curved Fronts}
455: (Academic Press, San Diego 1988).
456: \bibitem{Wess} P. Wesseling, {\it Principles of Computational
457: Fluid Dynamics}, Springer Series in Comp. Math. {\bf 29} (Berlin 2001).
458: \bibitem{pulled} U.~Ebert and W.~van Saarloos, Phys. Rev. Lett. {\bf 80},
459: 1650 (1998); and Physica D {\bf 146}, 1--99 (2000).
460: \bibitem{MBA} U.~Ebert, W.~van Saarloos, Phys.\ Rep.\ {\bf 337}, 139 (2000).
461: \end{references}
462:
463: \end{multicols}
464:
465: \end{document}
466:
467:
468: