1: \documentstyle[pre,aps,twocolumn]{revtex}
2: %\documentstyle[prl,aps,preprint]{revtex}
3: \input epsf
4:
5:
6: \begin{document}
7:
8: \title{Passive scalar intermittency in low temperature helium flows}
9:
10: \author{F. Moisy, H. Willaime, J.S. Andersen$^{*}$ and P. Tabeling}
11:
12: \address{Laboratoire de Physique Statistique, \'Ecole Normale
13: Sup\'erieure, 24 rue Lhomond, 75231 Paris Cedex 05 (France),\\
14: $^*$Centre for Chaos and Turbulence Studies, Niels Bohr Institute,
15: Blegdamsvej 17-19, DK-2100 K{\o}benhavn {\O} (Denmark).}
16:
17: \date{\today}
18: \maketitle
19:
20:
21: \begin{abstract}
22: We report new measurements of turbulent mixing of temperature
23: fluctuations in a low temperature helium gas experiment, spanning a
24: range of microscale Reynolds number, $R_{\lambda}$, from 100 to 650.
25: The exponents $\xi_{n}$ of the temperature structure functions
26: $\langle |\theta(x+r)-\theta(x)|^n \rangle \sim r^{\xi_{n}}$ are shown
27: to saturate to $\xi_{\infty} \simeq 1.45 \pm 0.1$ for the highest
28: orders, $n \sim 10$. This saturation is a signature of statistics
29: dominated by front-like structures, the cliffs. Statistics of the
30: cliff characteristics are performed, particularly their width are
31: shown to scale as the Kolmogorov length scale.
32: \end{abstract}
33:
34: \pacs{47.27.Jv, 47.27.Qb}
35: \narrowtext
36:
37: The strong intermittency of a passive scalar field advected by
38: a turbulent flow has recently received considerable
39: attention~\cite{Shraiman99,Warhaft2000}. Two facets of this
40: intermittency are the persistence of small scale
41: anisotropy~\cite{Mestayer76,Gibson77} and the anomalous scaling of the
42: structure functions \mbox{$\langle |\theta(x+r)-\theta(x)|^n
43: \rangle$}~\cite{Antonia84,Kraichnan94}. It is well established that
44: the persistence of scalar gradient skewness arises from the
45: ramp-and-cliff structures~\cite{Mestayer76,Gibson77}, {\it i.e} high
46: scalar jumps separated by well mixed regions. The observed
47: preferential alignment of cliffs with the large scale
48: gradient~\cite{Holzer94,Pumir94} apparently prevents a universal
49: description of the odd order statistics, reflecting the asymetry of
50: scalar fluctuations. However, the genericity of cliffs in ``scalar
51: turbulence'' raises the issue of their influence on high-order scaling
52: of even moments, and of the possible universality of the suspected
53: saturation of high order exponents~\cite{Yakhot97,Celani2000,caro}.
54: Laboratory experiments able to cover a wide range of Reynolds numbers
55: in well controlled conditions appear to be crucial to address this
56: point. We report in this Letter new measurements of turbulent mixing
57: of a scalar field, namely temperature fluctuations, performed in a low
58: temperature helium gas experiment. Such measurements of temperature
59: fluctuations, as a passively advected scalar field, are performed for
60: the first time in low temperature helium gas~\cite{DonnellySR},
61: opening new and encouraging perspectives in investigation of turbulent
62: mixing in high Reynolds number flows.
63:
64: The set-up we use is the same as the one described in
65: Ref.~\cite{Zocchi94}, with an additional apparatus to induce
66: temperature fluctuations. The flow takes place in a cylindrical
67: vessel and is driven by two rotating disks, as sketched in
68: Fig.~\ref{fig:dessinmanip}. The disks, 20~cm in diameter and spaced
69: 13.1~cm apart, are mounted with 6 radial blades. The cell is filled
70: with helium gas, held at a controlled pressure. Temperature
71: measurements are performed with a specially designed cold-wire
72: thermometer in the thermal wake of a heated grid. The grid is made of
73: a double nichrome wire net, stretched across a $4.0 \times 4.5$~cm
74: frame. The wire diameter is 250~$\mu$m, and the mesh size, $M$, is
75: 2.0~mm.
76:
77: %%%%%%%%%%%%%%%%%
78: \begin{figure} [b]
79: \epsfxsize=8cm
80: \centerline{ \epsffile{FIG01.eps}}
81: \vspace{3mm}
82: \caption{(a)~Sketch of the experiment. (b)~Upper view, in
83: corotating mode. (1)~Disks, (2)~thermometer, (3)~heated grid.}
84: \label{fig:dessinmanip}
85: \end{figure}
86: %%%%%%%%%%%%%%%%%
87:
88: The thermometer is located 20 mesh sizes downstream from the grid, at
89: a distance of 2.2~cm from the wall. It has been designed as the
90: velocity probes described in Ref.~\cite{Zocchi94}. It consists in a
91: thin carbon fiber, 7~$\mu$m in diameter, stretched across a 2~mm rigid
92: frame. A metallic deposition of gold and silver of about
93: 2000~\AA~thick covers the whole fiber, except for a 7~$\mu$m long
94: central area. Temperature fluctuations are measured from resistance
95: fluctuations, by the way of a constant current Wheatstone bridge. The
96: tuning of the current has received considerable attention, in order to
97: ensure an high enough signal, without suffering from velocity
98: contamination. In practice, the optimal value of the current is
99: chosen in order to maximize the flatness factor of the temperature
100: derivative. Temperature resolution is estimated, from the
101: high-frequency white noise level, to around 100~$\mu$K.
102: The spatial resolution is limited by the probe size, which
103: is at least twice smaller than the Kolmogorov scale in the present
104: experiments. The frequency response is limited by the constant
105: current amplifier to 10~kHz, a value comfortably high to resolve the
106: highest frequency of the temperature fluctuations.
107:
108: Since the same probe, with different electronic devices, is used as
109: both thermometer and anemometer, we are able to perform temperature
110: and velocity measurements at the same point in the same flow
111: conditions. We have measured the various turbulence characteristics,
112: such as the microscale Reynolds number $R_{\lambda}$ and the
113: Kolmogorov scale $\eta$, for the same disks rotation rates and
114: kinematic viscosity $\nu$, allowing us to characterize each
115: temperature data set. We define here $R_{\lambda} = u' \lambda / \nu$
116: the Reynolds number based on the Taylor scale $\lambda=u'(\nu /
117: \epsilon)^{1/2}$, where $u'$ is the rms of the velocity fluctuations,
118: $\epsilon$ the mean dissipation rate and $\eta=(\nu^{3}/\epsilon)^{1/4}$.
119:
120: The disks rotation can be set in two modes, co- and counter-rotating,
121: hereafter denoted by COR and CTR. In the corotating mode, the two
122: disks are rotating in the same direction at the same speed, so that
123: the flow can be thought as a solid-like rotation. The fluctuation
124: rate $u' / \langle U \rangle$ is found to lie around 10~\% in this
125: case, a value close to the one obtained in a previous experiment in a
126: similar set-up~\cite{Willaime99}. In the counter-rotating mode, the
127: turbulence intensity is much higher, but the speed ratio is tuned so
128: that a strong mean advection remains at the height of the grid and the
129: thermometer. This ensures a low enough fluctuation rate, between 13
130: and 25~\%, allowing use of the Taylor hypothesis to convert temporal
131: fluctuations into spatial ones. We have checked that the fluctuation
132: rate is the same with and without the grid, meaning that the grid
133: contribution to the turbulence can be considered as negligible. The
134: only influence of the grid is injection of heated sheets in an already
135: turbulent flow. Thus thermal turbulence originates in the mixing of
136: these hot layers behind each wire, over a typical distance of $M
137: \langle U \rangle /u' = $1--2~cm downstream.
138:
139: The grid temperature has to be high enough to induce thermal
140: fluctuations giving rise to an acceptable signal to noise ratio of at
141: least 50~dB. The noise level of 100~$\mu$K requires a rms signal of
142: typically $\theta' = \langle \theta^2 \rangle^{1/2} \simeq 40$~mK at
143: the probe location, which is around 4~\% of the grid overheat. Since
144: we are working with a closed flow, the heat transfer from the cell to
145: the cryostat has to be balanced by the power supplied to the grid, in
146: order to ensure thermal stationarity. This constraint limits the
147: maximum duration of an experiment (up to 5 hours of continuous run),
148: and the range of velocity and fluid density; in practice, values of
149: $R_{\lambda}$ in the range 100--300 in the corotating case, and
150: 200--650 in the counter-rotating case, are obtained. Although much
151: lower than the highest $R_{\lambda}$ achievable in our set-up (up to
152: 5000~\cite{Zocchi94}), we hope that future noise improvement of the
153: constant current amplifier will increase the upper bound well beyond
154: 650.
155:
156: %%%%%%%%%%%%%%%%%
157: \begin{table} [b]
158: \begin{tabular}{llllll}
159: File no.& $10^4 \times \kappa$ & $R_{\lambda}$ & $\langle U \rangle $ &
160: $\theta'$ & $10^{-6} \,N^{*}$\\
161: & (cm$^{2}$/s) & & (cm/s) & (mK) & \\
162: \hline
163: 1 COR & 75 & 105 & 16.3 & 45.5 & 0.84 \\ %39
164: 2 COR & 17 & 280 & 27.4 & 56.8 & 14.7 \\ %78
165: 3 CTR & 17 & 360 & 28.2 & 96.0 & 30.6 \\ %147
166: 4 CTR & 17 & 650 & 34.4 & 69.0 & 85.7 %141
167: \end{tabular}
168: \caption{Typical experimental parameters. $\kappa = \nu / \mbox{Pr}$
169: is the thermal diffusivity, $\nu$ the kinematic viscosity and
170: $\mbox{Pr} \simeq 0.8$. The sample size $N^*$ is expressed as $N
171: \langle U \rangle / (2 \pi f_{s} \eta)$, where $N$ is the number of
172: data points of the sample and $f_{s}$ the sampling rate.}
173: \label{tab:tempfiles}
174: \end{table}
175: %%%%%%%%%%%%%%%%%
176:
177: The temperature signal is filtered, then sampled and recorded on a 16~bit
178: acquisition board (ITC-18 from InstruTech), at a sampling rate between
179: 1 and 10~kHz, a value at least twice the low-pass filter frequency.
180: Special attention has been paid to the low noise level of the whole
181: amplifying and sampling channel. In order to achieve correct
182: convergence of higher-order statistics, the sample sizes are of the
183: order of $10^7 - 10^8$ Kolmogorov time scales long.
184: Table~\ref{tab:tempfiles} summarizes the characteristics of the
185: typical data sets used here.
186:
187: In order to characterize the large scale of the temperature and
188: velocity fluctuations, quantities of interest are the integral scales
189: $\Lambda_{\theta}$ and $\Lambda_u$, defined from the autocorrelation
190: functions of temperature $\theta$ and longitudinal velocity $u$. For
191: our whole data set, we measure $\Lambda_{\theta} = 7.0 \pm 1.5$~mm and
192: $\Lambda_u = 9.6 \pm 1.0$~mm, with no noticeable $R_{\lambda}$
193: dependence~\cite{Willaime99}. This two values are rather close,
194: indicating that kinetic energy and temperature variance are injected
195: at the same large scale. The Prandtl number of the gas being close to
196: unity, we are on the case where velocity and scalar fluctuations take
197: place on the same range of scales.
198:
199: %%%%%%%%%%%%%%%%%
200: \begin{figure} [b]
201: \epsfxsize=8.6cm
202: \centerline{ \epsffile{FIG02.eps}}
203: \caption{Probability density functions of the normalized temperature
204: increments, for $R_{\lambda}=650$ (file \#4 of
205: Tab.~\ref{tab:tempfiles}). From the inner to the outer pdf, $r/\eta =
206: 10^4$, 600, 30 and 3.}
207: \label{fig:pdf141}
208: \end{figure}
209: %%%%%%%%%%%%%%%%%
210:
211: The small scale intermittency of temperature fluctuations can be
212: characterized through their distributions at different scales.
213: Figure~\ref{fig:pdf141} shows the probability density functions (pdf)
214: of temperature increments $\Delta \theta(r) = \theta(x+r)-\theta(x)$
215: for 4 different separations $r$, ranging from $r/\eta$ = 10$^4$ (large
216: scale), 600, 30 (typical inertial scales) down to 3 (close to the
217: dissipative scale). Each distribution has been normalized by its
218: standard deviation $\sigma = \langle \Delta \theta(r)^2
219: \rangle^{1/2}$. These pdf are strongly not self-similar, reflecting
220: strong intermittency effects. The width of the tails is remarkable,
221: showing scalar jumps of amplitude up to 40 times the standard
222: deviation at the smallest scale. We can note the slight asymmetry of
223: the distributions, linked to the well known property of small scale
224: persistence of anisotropy~\cite{Warhaft2000}.
225:
226: %%%%%%%%%%%%%%%%%
227: \begin{figure} [b]
228: \epsfxsize=7cm
229: \centerline{ \epsffile{FIG03.eps}}
230: \caption{Temperature structure function exponents, $\xi_{n}$,
231: normalized by $\xi_{2}$. $\bigcirc$:~$R_{\lambda}=280$ in
232: COR mode, $\triangle$:~$R_{\lambda}=650$ in CTR mode. The dashed line
233: indicates the Corrsin-Obukhov scaling $n/2$.}
234: \label{fig:xnx2}
235: \end{figure}
236: %%%%%%%%%%%%%%%%%
237:
238: %%%%%%%%%%%%%%%%%
239: \begin{figure} [b]
240: \epsfxsize=8.6cm
241: \centerline{ \epsffile{FIG04.eps}}
242: \caption{Ratio of two pdfs, at different inertial scales, $r_{1}=45 \eta$
243: and $r_{2}=246 \eta$ ($R_{\lambda}=280$). The long dashed curve is a
244: gaussian fit, and the horizontal dashed lines indicate the constant
245: ratio of the pdf tails. Vertical arrows, at $\Delta \theta \simeq 4
246: \theta'$, indicate the dominant contribution to the 8th order moment.}
247: \label{fig:pdfratio}
248: \end{figure}
249: %%%%%%%%%%%%%%%%%
250:
251: The evolution of these pdfs is characterized by the structure
252: functions, defined as \mbox{$S_{n} (r) = \langle |\theta(x+r)-\theta(x)|^n
253: \rangle$}, where angular brackets denote space (time) average. They
254: are found to follow power law in terms of the scale $r$, namely
255: $S_{n} (r) \sim r^{\xi_{n}}$. The measured structure
256: function exponents $\xi_{n}$, divided by $\xi_2$, are shown in
257: Fig.~\ref{fig:xnx2}, in the COR ($R_{\lambda}=280$) and CTR
258: ($R_{\lambda}=650$) cases. The exponents are defined by plotting the
259: compensated structure functions $r^{-\xi_{n}} S_{n} (r)$ and
260: tuning the value of the exponent to obtain a well defined plateau for
261: inertial separations. This procedure allows to estimate the error bar
262: for each order. Although the second order exponent presents some
263: scatter (with a systematic increase from 0.45 to 0.65 with increasing
264: $R_{\lambda}$), the normalization of the higher order exponents by
265: $\xi_{2}$ provides an excellent collapse for the different
266: $R_{\lambda}$. It is remarkable that, for comparable sample sizes,
267: the highest available order is much lower at higher Reynolds number:
268: the fluctuations becomes much more intermittent and the convergence
269: becomes poorer. For $R_{\lambda}=650$ we have to restrict to $n \leq
270: 8$, whereas $n=12$ can be achieved for $R_{\lambda}=280$. The two
271: sets of exponents are consistent within error bars for $n \leq 8$,
272: meaning that the large scale properties of the two flow configurations
273: do not affect these inertial range statistics.
274:
275:
276: The values of $\xi_n / \xi_{2}$ are found to strongly depart from the
277: linear law $n/2$, and the gap increases with the order, a usual
278: signature of inertial range intermittency of the passive
279: scalar\cite{Warhaft2000}. Furthermore, the exponents are found
280: to increase extremely slowly with the order, strongly suggesting a
281: saturation, for $n$ around 10, at a value
282: \begin{equation}
283: \xi_{\infty} = (2.3 \pm 0.1) \xi_{2} \simeq 1.45 \pm 0.1
284: \label{eq:xiinf}
285: \end{equation}
286: (where $\xi_2$ is considered equal to 2/3). This observation only
287: relies on the COR data set, but we have noted that no deviation
288: appears between the COR and CTR data for $n \leq 8$. A saturation of
289: the structure function exponents is a signature of statistics
290: dominated by shock-like structures, the thermal cliffs. A consequence
291: of this saturation is a constant ratio of the far tails of the pdfs
292: for inertial range separations. The ratio of two pdfs, for
293: separations $r_{1}$ and $r_{2}$ well into the inertial range, is
294: displayed in Fig.~\ref{fig:pdfratio}. For temperature increments
295: $|\Delta \theta| > 4 \theta'$, this ratio tends towards a constant.
296: It must be noted that the $|\Delta \theta| \simeq 4 \theta'$ parts of
297: this distribution give the dominant contribution to the 8th order
298: moment (see the vertical arrows), {\it i.e} they maximize the
299: integrant $|\Delta \theta|^8 p(\Delta \theta)$. This additional test
300: confirms the observed saturation of high order exponents.
301:
302: %%%%%%%%%%%%%%%%%
303: \begin{figure} [b] \epsfxsize=8.6cm
304: \centerline{\epsffile{FIG05.eps}}
305: \caption{Histograms of cliff width (a) and amplitude (b) for values of
306: the threshold $s=3$ ($\bigcirc$), 7 ($\Box$) and 11 ($\diamond$). The
307: dashed curves on the figure (b) are log-normal fits.}
308: \label{fig:pdf_cliffs}
309: \end{figure}
310: %%%%%%%%%%%%%%%%%
311:
312: The observed trend towards a saturation of high order exponents
313: motivates a detailed analysis of the cliffs. We have performed
314: statistics of the cliff characteristics, defined as the strongest
315: temperature gradients, singled out from time series of temperature
316: fluctuations. We define the cliffs from a simple threshold on the
317: temperature derivative,
318: \begin{equation}
319: |\partial \theta / \partial x | > s \, \langle (\partial \theta /
320: \partial x)^{2} \rangle^{1/2},
321: \label{eq:thres}
322: \end{equation}
323: where $s$ is a non dimensional constant. Spatial derivatives are
324: obtained from temporal ones using the Taylor hypothesis, and the
325: gradients are estimated from finite difference over the smallest
326: resolved separation. We define the cliff amplitude $\delta \theta$ as
327: the difference between the two extrema surrounding the gradient
328: satisfying~(\ref{eq:thres}), and the cliff width $\Delta$ such that
329: the temperature derivative takes values exceeding 90~\% of its local
330: maximum. Histograms of amplitudes and widths are shown in
331: Fig.~\ref{fig:pdf_cliffs}, for three values of the threshold $s$, at a
332: Reynolds number $R_{\lambda} = 280$.
333:
334: We first note that both amplitude and width histograms can be well
335: fitted by log-normal distributions (given by a parabola in log-log
336: coordinates). The mean amplitude of the cliffs is of order $\theta'$,
337: and is found to increase with the threshold $s$. On the other hand,
338: the cliff width remains constant for different thresholds. This means
339: that the strength of the gradients depends mainly on the amplitude of
340: the scalar jump, and not on its width.
341:
342: %%%%%%%%%%%%%%%%%
343: \begin{figure} [b]
344: \epsfxsize=8.6cm
345: \centerline{ \epsffile{FIG06.eps}}
346: \vspace{2mm}
347: \caption{Mean cliff width divided by the Kolmogorov length scale, as a
348: function of $R_{\lambda}$. $\bigcirc$:~COR,
349: $\triangle$:~CTR. The dashed line is the average, $\langle
350: \Delta \rangle / \eta \simeq 13$.}
351: \label{fig:cliffwidth}
352: \end{figure}
353: %%%%%%%%%%%%%%%%%
354:
355: A central issue concerning the cliffs is the Reynolds number
356: dependence of their width. Figure~\ref{fig:cliffwidth} shows the mean
357: cliff width $\langle \Delta \rangle$, divided by the Kolmogorov scale
358: $\eta$, for $R_{\lambda}$ ranging from 100 to 650. This plot extends a
359: preliminary study~\cite{moisyETC} performed only on the COR mode. We can
360: see a well defined plateau,
361: \begin{equation}
362: \langle \Delta \rangle = (13 \pm 3) \eta.
363: \end{equation}
364: The scatter is moderate, and probably reflects the difficulty to resolve
365: properly the smallest scales. However, this plot confirms that
366: the mean cliff width follows a $R_{\lambda}^{-3/2}$ scaling. This law
367: holds for values of the threshold $s$ from 3 up to 40 in the case of
368: the highest $R_{\lambda}$ (see Fig.~\ref{fig:pdf141}). We can note
369: that the data from the two configurations overlap on the central
370: region $R_{\lambda} = 200-300$, suggesting that this small scale
371: characteristics is not affected by the large scale properties of the
372: flow.
373:
374:
375: To summarize, we have performed for the first time measurements of
376: turbulent mixing of temperature, considered as a passive scalar field,
377: in a low temperature helium experiment. The study of inertial range
378: statistics of the temperature increments gives evidence of a
379: saturation of the high order exponents, to a value $\xi_{\infty}
380: \simeq 1.45 \pm 0.1$. This observation reveals that inertial range
381: statistics are dominated by the cliffs, concentrating large scalar
382: jumps over small distance~\cite{Yakhot97,Celani2000,caro}. The cliff
383: widths are shown to scale as the Kolmogorov length scale, in the
384: whole range of observed $R_{\lambda}$ (100--650), suggesting that the
385: strongest cliffs remain concentrated over the smallest scale of the
386: flow (see also Ref.~\cite{Nieuwstadt}). This observation may have
387: important consequences for processes such as reactive mixing or
388: combustion, where the reaction rate is enhanced where concentration
389: gradients are strong. Further insight into the cliffs contribution to
390: the saturation of the high order structure function exponents needs a
391: detailed study of their morphology and spatial distribution.
392: Preliminary results~\cite{moisyETC}, based on the waiting time between
393: cliffs from the temperature time series, strongly suggests
394: self-similar clustering for inertial separations.
395:
396:
397: The authors thank B.~Shraiman, V.~Hakim, M.~Vergassola, P.~Castiglione
398: and M.C.~Jullien for fruitful discussions. This work has been
399: supported by Ecole Normale Sup\'erieure, CNRS, the Universities Paris
400: 6 and Paris 7, and the European Commission's TMR programme, Contract
401: no.~ERBFMRXCT980175 ``Intermittency''.
402:
403:
404: \begin{references}
405:
406: \bibitem{Shraiman99} B.I.~Shraiman and E.~Siggia, Nature {\bf 405},
407: 639 (2000).
408:
409: \bibitem{Warhaft2000} Z.~Warhaft, Annu. Rev. Fluid. Mech. {\bf
410: 32}, 203 (2000).
411:
412: \bibitem{Mestayer76} P.G.~Mestayer, C.H.~Gibson, F.M.~Coantic, and
413: A.S.~Patel, Phys. Fluid. {\bf 19} (9), 1279 (1976).
414:
415: \bibitem{Gibson77} C.H.~Gibson, C.A.~Friehe and S.O.~McConnell, Phys.
416: Fluid. {\bf 20}, S156 (1977).
417:
418: \bibitem{Antonia84} R.A.~Antonia, E.~Hopfinger, Y.~Gagne and
419: F.~Anselmet, Phys. Rev. A {\bf 30}, 2704 (1984).
420:
421: \bibitem{Kraichnan94} R.H.~Kraichnan, Phys. Rev. Lett. {\bf 72} (7),
422: 1016 (1994).
423:
424: \bibitem{Holzer94} M.~Holzer and E.~Siggia, Phys. Fluid {\bf 6} (5),
425: 1820 (1994).
426:
427: \bibitem{Pumir94} A.~Pumir, Phys. Fluid. {\bf 6} (6), 2118 (1994).
428: Phys. Fluid. {\bf 6} (12), 3974 (1994).
429:
430: \bibitem{Yakhot97} V.~Yakhot, Phys. Rev. E {\bf 55},329 (1997).
431:
432: \bibitem{Celani2000} A.~Celani, A.~Lanotte, A.~Mazzino and
433: M.~Vergassola, Phys. Rev. Lett.Ê{\bf 84}, 2385Ê(2000)
434:
435: \bibitem{caro} M.C.~Jullien, P.~Castiglione and P.~Tabeling, subm. to
436: Phys. Rev. Lett.
437:
438: \bibitem{DonnellySR} {\sl Flow at ultra-high Reynolds and Rayleigh
439: numbers, a status report}, edited by R.J.~Donnelly and
440: K.R.~Sreenivasan (Springer-Verlag, New York, 1998).
441:
442: \bibitem{Zocchi94} G.~Zocchi, P.~Tabeling, J.~Maurer and H.~Willaime,
443: Phys. Rev. E {\bf 50} (5), 3693 (1994). F.~Moisy, P.~Tabeling and
444: H.~Willaime, Phys. Rev. Lett. {\bf 82}, 3994 (1999).
445:
446: \bibitem{Willaime99} H.~Willaime, J.~Maurer, F.~Moisy and P.~Tabeling,
447: to appear in Eur. Phys. J. B (2000).
448:
449: \bibitem{Nieuwstadt} F.T.M.~Nieuwstadt and G.~Brethouwer, {\it
450: Advances in Turbulence VIII}, edited by C.~Dopazo (CIMNE, Barcelona,
451: 2000), 133.
452:
453: \bibitem{moisyETC} F.~Moisy, H.~Willaime, J.S.~Andersen and
454: P.~Tabeling, {\it Advances in Turbulence VIII}, edited by C.~Dopazo
455: (CIMNE, Barcelona, 2000), 835.
456:
457: \end{references}
458:
459: \end{document}