1: \documentstyle[aps,multicol,epsfig]{revtex}
2: \newcommand{\bm}[1]{\mbox{\boldmath $#1$}}
3: \begin{document}
4: %\draft
5: %\input epsf
6:
7: \title{Lagrangian Statistics and Temporal Intermittency in a
8: Shell Model of Turbulence}
9:
10: \author{G. Boffetta, F. De Lillo and S. Musacchio}
11: \address{Dipartimento di Fisica Generale and INFM, Universit\`a di Torino \\
12: Via Pietro Giuria 1, 10125 Torino, Italy}
13: \address{CNR-ISAC - Sezione di Torino \\
14: Corso Fiume 4, 10133 Torino, Italy}
15: \date{\today}
16:
17: \maketitle
18:
19: \begin{abstract}
20: We study the statistics of single particle Lagrangian velocity in
21: a shell model of turbulence.
22: We show that the small scale velocity fluctuations are
23: intermittent, with scaling exponents connected to the
24: Eulerian structure function scaling exponents.
25: The observed reduced scaling range is interpreted as a
26: manifestation of the intermediate dissipative range, as it
27: disappears in a Gaussian model of turbulence.
28: \end{abstract}
29:
30: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
31: \begin{multicols}{2}
32: %\narrowtext
33:
34: In the recent years there has been a great improvement in the
35: experimental investigation of turbulence from a Lagrangian
36: point of view \cite{VSB98,OM00,LVCAB01,MMMP01}.
37: In the Lagrangian approach the
38: flow is described by the (Lagrangian) velocity ${\bm v}({\bm x}_0,t)$
39: of a fluid particle initially at position ${\bm x}(0)={\bm x}_0$.
40: This is the natural description for studying transport and
41: mixing of neutrally advected substances in turbulent flows.
42:
43: One of the simplest statistical quantities one can be interested in is
44: single particle velocity increments
45: $\delta {\bm v}(t)={\bm v}(t)-{\bm v}(0)$ (where, assuming
46: statistical homogeneity,
47: we have dropped the dependence on ${\bm x}_0$) for which
48: dimensional analysis in fully developed turbulence predicts
49: \cite{MY75}
50: \begin{equation}
51: \langle \delta v_{i}(t) \delta v_{j}(t) \rangle =
52: C_{0} \varepsilon t \delta_{ij}
53: \label{eq:1}
54: \end{equation}
55: where $\varepsilon$ is the mean energy dissipation and $C_{0}$ is
56: a numerical constant. The remarkable coincidence that the
57: variance of $\delta {\bm v}(t)$ grows linearly with time
58: is the physical basis on which stochastic models of particle
59: dispersion are based. It is important to recall that the
60: ``diffusive'' nature of (\ref{eq:1}) is purely incidental:
61: it is a direct consequence of Kolmogorov scaling in the inertial
62: range of turbulence and is not directly related to a diffusive
63: process.
64: Let us recall briefly the argument leading
65: to the scaling in (\ref{eq:1}). We can think at the velocity
66: $v(t)$ advecting the Lagrangian trajectory as the superposition
67: of the different velocity contributions coming from the turbulent eddies
68: (which also move with the same velocity of the Lagrangian trajectory).
69: After a time $t$ the components associated to the smaller
70: (and faster) eddies, below a certain scale $\ell$ are decorrelated
71: and thus at the leading order one has
72: $\delta v(t) \simeq \delta v(\ell)$. Within Kolmogorov scaling,
73: the velocity fluctuation at scale $\ell$ is given by
74: $\delta v(\ell) \sim V (\ell/L)^{1/3}$ where $V$
75: represents the typical velocity at the largest scale $L$.
76: The correlation time of $\delta v(\ell)$ scales as
77: $\tau(\ell) \sim \tau_0 (\ell/L)^{2/3}$ and thus one obtains
78: the scaling in (\ref{eq:1}).
79:
80: Equation (\ref{eq:1}) can be generalized to higher order moments
81: with the introduction of a set of temporal scaling exponent $\xi(p)$
82: \begin{equation}
83: \langle \delta v(t)^{p} \rangle \sim V^p (t/\tau_0)^{\xi(p)}
84: \label{eq:2}
85: \end{equation}
86: The dimensional estimation sketched above gives the prediction $\xi(p)=p/2$
87: but one expects corrections to dimensional scaling in presence of
88: intermittency.
89:
90: A generalization of (\ref{eq:1}) which takes into
91: account intermittency corrections can be easily developed within
92: the multifractal model of turbulence \cite{PF85,Frisch95}.
93: The dimensional argument is repeated for the local
94: scaling exponent $h$, giving $\delta v(t) \sim V (t/\tau_0)^{h/(1-h)}$.
95: Integrating over the $h$ distribution one ends with the prediction
96: \begin{equation}
97: \langle \delta v(t)^{p} \rangle \sim V^p
98: \int dh \left({t \over \tau_0} \right)^{[ph-D(h)+3]/(1-h)}
99: \label{eq:3}
100: \end{equation}
101: where, in the limit $t/\tau_0 \to 0$, the integral can be estimated
102: by a steepest descent argument giving finally
103: \begin{equation}
104: \xi(p)=\min_{h}\left[{ph-D(h)+3 \over 1-h} \right]
105: \label{eq:4}
106: \end{equation}
107: The fractal dimension $D(h)$ is related to the Eulerian structure function
108: scaling exponents $\zeta(q)$ by the Legendre transform
109: \cite{Frisch95} $\zeta(q)=\min_{h}[qh-D(h)+3]$.
110: The standard inequality in the multifractal model (following from
111: the exact result $\zeta(3)=1$), $D(h)\le 3h+2$ implies for (\ref{eq:4})
112: that even in presence of intermittency $\xi(2)=1$. This is a direct
113: consequence of the fact that energy dissipation enters in (\ref{eq:1})
114: at the first power.
115: Our expression for scaling exponents (\ref{eq:4}) recovers in a more
116: compact form the prediction obtained on the basis of an ``ergodic
117: hypothesis'' of the statistics of energy dissipation \cite{Borgas93}.
118:
119: Recent experimental results \cite{MMMP01} have shown that indeed
120: Lagrangian velocity fluctuations are intermittent and display
121: anomalous scaling exponents, as predicted by the above argument.
122: Despite the relative high Reynolds number of the experiments, a
123: true temporal scaling range is not observed. Thus the estimation
124: of the scaling exponents in (\ref{eq:2}) can be done only
125: relatively to a reference moment (the so-called ESS procedure
126: \cite{BCTBMS93}).
127:
128: In this work we use a dynamical toy model of turbulence for
129: investigating scaling (\ref{eq:2}) and prediction (\ref{eq:4})
130: at very high Reynolds numbers. The model is a popular shell model
131: of turbulence \cite{BJPV98} in which the velocity fluctuations
132: of the eddies at the scale $\ell_{n}=L 2^{-n}$ are represented by a
133: single variable $u_{n}$ ($n=1,2,...,N$).
134: Only local interaction among shells
135: are represented and therefore no sweeping effects are present.
136: In this sense shell models are dynamical models of velocity
137: fluctuations in a Lagrangian framework, and have been already
138: used for the study of turbulent dispersion \cite{BCCV99}.
139: The equation of motion for the complex shell variable $u_n$ is \cite{sabra}
140: \begin{eqnarray}
141: \frac{d u_n}{dt}&=&i k_n \left( u_{n+2}u_{n+1}^*
142: - {\delta \over 2} u_{n+1}u_{n-1}^*+
143: {1-\delta \over 4} u_{n-1}u_{n-2} \right) \nonumber \\
144: &&-\nu k_n^2 u_n +f_n
145: \label{eq:5}
146: \end{eqnarray}
147: where $k_n=\ell_n^{-1}$ and $f_n$ is a deterministic forcing
148: acting on the first two shells only.
149: Shell model (\ref{eq:5}) is characterized by a chaotic dynamics
150: with a statistically steady state with a constant flux of energy from large
151: to small scales.
152: The fluctuations generate by chaoticity induce a breaking of the
153: global scaling invariance and corrections to the Kolmogorov exponents
154: for the structure functions close to the experimental values \cite{BJPV98}.
155:
156: Lagrangian velocity in the shell model framework can be represented
157: as the superposition of the contributions of all the different eddies.
158: Let us thus define
159: \begin{equation}
160: v(t) \equiv \sum_{n=1}^{N} Re(u_n)
161: \label{eq:6}
162: \end{equation}
163: where we have taken, rather arbitrarily, only the real part of the
164: shell variables with unit coefficient. From the definition of the
165: shell model, there is not a precise recipe for reconstructing the
166: Lagrangian velocity.
167: More in general, one could think of a representation in which
168: shell variables are multiplied by an appropriate wavelet functions.
169: Of course numerical prefactor, such as $C_0$ in (\ref{eq:1}) will
170: depend on the wavelet basis. Nevertheless one expects that different
171: choices should not affect Lagrangian scaling exponents $\xi(p)$
172: which are determined by the dynamical properties of the model.
173:
174: Previous studies of multi-time correlations in shell models of
175: turbulence have demonstrated the existence of a set of correlation
176: times compatible with the multifractal picture of the turbulent
177: cascade \cite{BBCT99}. This is an indication that, as we will see,
178: Lagrangian velocity defined as (\ref{eq:6}) will be affected by intermittency.
179:
180: Very long and accurate numerical simulations of the shell model (\ref{eq:5})
181: with $N=24$ shells and $\delta=1/2$ have been performed.
182: The energy is injected at a constant flux $\varepsilon=0.01$ is the
183: first $2$ shells and is removed at the smallest shells by viscosity
184: $\nu=10^{-7}$. With these parameters, our simulations correspond to
185: a Reynolds number $Re\simeq 10^{8}$.
186: For each realization, Lagrangian structure functions
187: are computed from the Lagrangian velocity (\ref{eq:6}) up to the
188: large-scale time $\tau_{0}$. Average is then taken over $10^{5}$
189: independent realizations.
190:
191: In Fig.~\ref{fig1} we plot the set of numerically determined
192: Eulerian structure function scaling exponent $\zeta(q)$ together with
193: the fractal dimension $D(h)$ reconstructed by means of the Legendre
194: transform. We observe strong intermittency in velocity statistics
195: with scaling exponent which deviates from Kolmogorov prediction.
196: The scaling exponents are not universal with respect to the
197: particular shell model. Model (\ref{eq:5}) gives a set of exponents
198: which are a little more intermittent than, but not far from, the
199: experimentally observed exponents \cite{Frisch95}:
200: $\zeta(2)\simeq 0.72$, $\zeta(4)\simeq 1.25$, $\zeta(6)\simeq 1.71$.
201: We thus expect that the values of $\xi(p)$ obtained from (\ref{eq:4})
202: using the $D(h)$ of Fig.~\ref{fig1} will be directly comparable with real
203: experimental data.
204:
205: Figure \ref{fig2} shows the second-order Lagrangian structure
206: function (\ref{eq:1}) as a function of time. The linear behavior
207: is evident (see the inset) even if a long crossover from
208: the ballistic scaling at short time
209: $\langle \delta v(t) ^2 \rangle \sim t^2$ is present.
210: In spite of the very high Reynolds numbers achievable in the shell model,
211: the extension of the temporal scaling (\ref{eq:2}) is still moderate.
212: For a comparison with the available
213: experimental data, in the inset we also plot the result obtained
214: from two simulations at lower resolution, with $Re \simeq 2 \times 10^{6}$
215: and $Re \simeq 10^{5}$. In the latter case almost no scaling
216: range is observable.
217: Despite these limitations, we will see that high $Re$ simulations
218: allow the determination of the Lagrangian scaling exponents with
219: good accuracy.
220:
221: The long crossover in Fig.~\ref{fig2} can be understood in terms
222: of intermediate dissipative range as a consequence of the fluctuating
223: dissipative scale \cite{Frisch95,FV91}. The smallest time at which
224: one can expect scaling (\ref{eq:1}) is the Kolmogorov time
225: $\tau_{\eta} \sim \tau_{0} Re^{-(1-h)/(1+h)}$ which fluctuates with
226: $h$. A demonstration of the effects induced by intermittency
227: is given by considering a non-intermittent Gaussian model.
228:
229: Setting $f_{n}=\nu=0$, (\ref{eq:5}) becomes a conservative system
230: with two conserved quantitities which depends on the value of the
231: $\delta$ \cite{BJPV98}. In statistical stationary condition,
232: the model shows equipartition of the conserved quantities among the
233: shell, in agreement with statistical mechanics prediction
234: \cite{ABCFPV94}.
235: For $\delta=1+2^{-2/3}$ the equipartition state leads at small scales
236: to Kolmogorov scaling $\langle |u_{n}|^2 \rangle \sim \ell_{n}^{2/3}$
237: with Gaussian statistics. In Fig.~\ref{fig3} we plot the second-order
238: Lagrangian structure function (\ref{eq:1}) for the Gaussian model.
239: Both the ballistic and the diffusive scaling is clearly observable
240: and the crossover is strongly reduced with respect to Fig.~\ref{fig2}.
241:
242: In Fig.~\ref{fig4} we plot the probability density functions
243: of $\delta v(t)$ computed at different $t$ in the linear
244: scaling range of Fig.~\ref{fig2} rescaled with their variances.
245: The form of the pdf varies continuously from almost Gaussian
246: at large time ($t \sim \tau_0$) to the development of
247: stretched exponential tails at short times, similar to
248: what observed in laboratory experiments \cite{MMMP01}.
249: Flatness $F$ grows from Gaussian value $F=3$ up to
250: $F \simeq 20$ at smallest times.
251: This is an indication of Lagrangian intermittency,
252: in the sense that the Lagrangian
253: statistics cannot be described in term of a single scaling
254: exponent.
255:
256: In Fig.~\ref{fig5} we plot the set of Lagrangian scaling exponents
257: $\xi(p)$ obtained from a direct fit of temporal structure functions.
258: The nonlinear behavior in $p$ confirms the presence of Lagrangian
259: intermittency already observed from the pdf.
260: We present the result for moments up to $p=6$ which approximatively
261: corresponds, from (\ref{eq:4}),
262: to Eulerian structure function of order $q=8$. In this sense
263: temporal structure functions are more intermittent.
264: Figure~\ref{fig5} shows that the agreement with the multifractal
265: prediction (\ref{eq:4}) is very good up to the moment achievable
266: with our statistics.
267: What is even more remarkable is that our prediction is very close
268: to experimentally determined exponents. For example
269: we find $\xi(3)\simeq 1.31$, $\xi(4)\simeq 1.58$, $\xi(5)\simeq 1.85$,
270: while the experimental data give \cite{MMMP01} $\xi_{exp}(3)=1.34 \pm 0.02$,
271: $\xi_{exp}(4)=1.56 \pm 0.06$ and $\xi_{exp}(5)=1.8 \pm 0.2$.
272:
273: In this work we have investigated the statistical properties of
274: single particle Lagrangian velocity in fully developed turbulence.
275: A prediction for intermittent scaling exponents of Lagrangian structure
276: function is given within the multifractal framework.
277: Very high Reynolds number simulations in shell model confirm
278: the multifractal prediction, even if rather small scaling ranges
279: are observed. In particular, our simulations show that at
280: experimental Reynolds number presently available almost no
281: scaling is observable. The reduction of the scaling range in
282: Lagrangian statistics is interpreted as an effect of the intermediate
283: dissipative range. A Gaussian, non-intermittent version of the shell
284: model confirms this interpretation.
285: An important consequence of our findings is that usual models
286: of particle dispersion, based on stochastic model \cite{Thomson87},
287: are incorrect in a fundamental sense, and one should take into
288: account the modifications due to intermittency.
289:
290: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
291: This work was supported by MIUR-Cofin2001 contract 2001023848.
292: We acknowledge the allocation of computer resources from INFM
293: ``Progetto Calcolo Parallelo''.
294: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
295: \begin{references}
296:
297: \bibitem{VSB98}
298: G.A. Voth, K. Satyanarayan and E. Bodenschatz,
299: Phys. Fluids {\bf 10}, 2268 (1998).
300:
301: \bibitem{OM00}
302: S. Ott and J. Mann J. Fluid Mech. {\bf 422}, 207 (2000).
303:
304: \bibitem{LVCAB01}
305: A. La Porta, G.A. Voth, A.M. Crawford, J. Alexander and
306: E. Bodenschatz, Nature {\bf 409}, 1017 (2001).
307:
308: \bibitem{MMMP01}
309: N. Mordant, P. Metz, O. Michel and J.F. Pinton,
310: Phys. Rev. Lett., {\bf 87}, 214501 (2001).
311:
312: \bibitem{MY75}
313: A.S. Monin and A.M. Yaglom, {\it Statistical Fluid Mechanics},
314: MIT Press, Cambridge, MA (1975).
315:
316: \bibitem{PF85}
317: G. Parisi and U. Frisch,
318: in {\it Turbulence and Predictability in Geophysical Fluid
319: Dynamics, Proceed. Intern. School of Physics `E. Fermi', 1983},
320: eds. M. Ghil, R. Benzi and G. Parisi, North-Holland, Amsterdam.
321:
322: \bibitem{Frisch95}
323: U. Frisch, {\it Turbulence}, Cambridge Univ. Press, Cambridge, UK (1995).
324:
325: \bibitem{Borgas93}
326: M.S. Borgas,
327: Phil. Trans. R. Soc. Lond. A, {\bf 342}, 379 (1993).
328:
329: \bibitem{BCTBMS93} R. Benzi, S. Ciliberto, R. Tripiccione, C. Baudet, F.
330: Massaioli and S. Succi,
331: Phys. Rev. E {\bf 48}, R29 (1993).
332:
333: \bibitem{BJPV98}
334: T. Bohr, M. Jensen, G. Paladin and A. Vulpiani,
335: {\it Dynamical Systems Approach to Turbulence},
336: Cambridge Univ. Press, Cambridge, UK (1998).
337:
338: \bibitem{BCCV99}
339: G. Boffetta, A. Celani, A. Crisanti and A. Vulpiani,
340: Phys. Rev. E, {\bf 60}, 6734 (1999)
341:
342: \bibitem{sabra}
343: V.S. L'vov, E. Podivilov, A. Pomyalov, I. Procaccia and
344: D. Vandembroucq,
345: Phys. Rev. E, {\bf 58}, 1811 (1998).
346:
347: \bibitem{BBCT99}
348: L. Biferale, G. Boffetta, A. Celani and F. Toschi,
349: Physica D {\bf 127}, 187 (1999).
350:
351: \bibitem{FV91}
352: U. Frisch and M. Vergassola,
353: Europhys. Lett. {\bf 14}, 439 (1991).
354:
355: \bibitem{ABCFPV94}
356: E. Aurell, G. Boffetta, A. Crisanti, P. Frick, G. Paladin and A. Vulpiani,
357: Phys. Rev. E, {\bf 50}, 4705 (1994).
358:
359: \bibitem{Thomson87}
360: D.J. Thomson,
361: J. Fluid Mech. {\bf 180}, 529 (1987).
362:
363: \end{references}
364:
365: \newpage
366: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
367: \begin{figure}[htb]
368: \epsfxsize=8.5cm
369: \centerline{\epsfbox{fig1.ps}}
370: \narrowtext
371: \caption{Shell model Eulerian structure function scaling exponents
372: $\zeta(q)$ determined from over the statistics of $10^5$
373: independent configurations. In the inset we plot the
374: codimension $3-D(h)$ as determined from numerically solving
375: the Legendre transform.}
376: \label{fig1}
377: \end{figure}
378: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
379:
380: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
381: \begin{figure}[htb]
382: \epsfxsize=8.5cm
383: \centerline{\epsfbox{fig2.ps}}
384: \narrowtext
385: \caption{Second-order Lagrangian structure function
386: $\langle \delta v(t) ^2 \rangle$ as a function of time
387: delay $t$ for the simulation at $Re=10^{8}$.
388: Continuous line is the ballistic behavior
389: $t^2$ at short time. Dashed line represents the
390: linear growth (\ref{eq:1}).
391: Inset: $\langle \delta v(t) ^2 \rangle$
392: compensated with the dimensional prediction $\varepsilon t$
393: at $Re=10^{8}$ (continuous line), $Re=2\times 10^{6}$ (dashed line)
394: and $Re=10^{5}$ (dotted line).}
395: \label{fig2}
396: \vspace{2cm}
397: \end{figure}
398: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
399:
400: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
401: \begin{figure}[htb]
402: \epsfxsize=8.5cm
403: \centerline{\epsfbox{fig3.ps}}
404: \narrowtext
405: \caption{Second-order Lagrangian structure function
406: $\langle \delta v(t) ^2 \rangle$ as a function of time
407: delay $t$ for the equilibrium Gaussian model.
408: Continuous line is the ballistic behavior
409: $t^2$ at short time. Dashed line represents the
410: linear growth (\ref{eq:1}).}
411: \label{fig3}
412: \end{figure}
413: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
414:
415: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
416: \begin{figure}[htb]
417: \epsfxsize=8.5cm
418: \centerline{\epsfbox{fig4.ps}}
419: \narrowtext
420: \caption{Probability density functions of velocity differences
421: $\delta v(t)$ normalized with the variance at time lags
422: $t/\tau_0=0.002 (\Box), 0.01 (*), 0.06 (\times), 0.35 (+)$.
423: Continuous line represents a Gaussian. Inset: flatness
424: $F=\langle \delta v(t)^4 \rangle / \langle \delta v(t)^2 \rangle^2$
425: as function of time and Gaussian value $F=3$ (dashed line).}
426: \label{fig4}
427: \end{figure}
428: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
429:
430: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
431: \begin{figure}[htb]
432: \epsfxsize=8.5cm
433: \centerline{\epsfbox{fig5.ps}}
434: \narrowtext
435: \caption{Lagrangian structure function scaling exponents
436: $\xi(p)$ numerically determined by a best fit on (\ref{eq:2}).
437: The line represents the multifractal prediction (\ref{eq:4})
438: with $D(h)$ obtained from Fig.~\ref{fig1}.}
439: \label{fig5}
440: \end{figure}
441: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
442:
443:
444: \end{multicols}
445: \end{document}
446:
447:
448: