1: %\documentclass[aps,prl,preprint,superscriptaddress,showpacs]{revtex4}
2: \documentclass[aps,prl,twocolumn,superscriptaddress,showpacs]{revtex4}
3:
4: \usepackage{amsmath}
5: \usepackage{amsfonts}
6: \usepackage{amssymb}
7: \usepackage{graphicx}
8: \usepackage[bf]{subfigure}
9:
10:
11: \newcommand{\R}{$\mathbb{R}$}
12: \newcommand{\UD}{1D}
13: \newcommand{\DD}{2D}
14: \newcommand{\TD}{3D}
15: \newcommand{\FD}{4D}
16: \newcommand{\CWT}{CWT}
17: \newcommand{\WTMM}{WTMM}
18: \newcommand{\WTMMM}{WTMMM}
19: \newcommand{\WT}{WT}
20: % bold typo in math mode
21: \newcommand{\gra}[1]{{\mathbf #1}}
22: \newcommand{\ron}[1]{{\mathcal #1}}
23: \newcommand{\bpsi}{\boldsymbol{\psi}}
24: % bold b
25: \newcommand{\gb}{\gra{b}}
26: \newcommand{\Mpsi}{\ron{M}_{\bpsi}}
27: % T psi (bold).
28: \newcommand{\Tpsi}{{\mathbf T}_{\bpsi}}
29: \newcommand{\Tpsii}{{\mathbf T}_{\bpsi_i}}
30: \newcommand{\Tpsirho}{{\mathbf T}_{\bpsi,\rho}}
31: %
32: \newcommand{\carreBlanc}{\protect\scalebox{0.6}{\ensuremath{\square}}}
33: \newcommand{\rondBlanc}{\ensuremath{\circ}}
34: \newcommand{\carreNoir}{\protect\scalebox{0.6}{\ensuremath{\blacksquare}}}
35: \newcommand{\rondNoir}{\ensuremath{\bullet}}
36: \newcommand{\conv}{\ensuremath{*}}
37: \newcommand{\ud}{\mathrm{d}}
38: \newcommand{\triangleNoir}{\protect\scalebox{0.9}{\ensuremath{\blacktriangle}}}
39: \newcommand{\triangleBlanc}{\protect\scalebox{0.9}{\ensuremath{\triangle}}}
40:
41:
42: \begin{document}
43:
44: \title{Generalizing the wavelet-based multifractal formalism to vector-valued random fields: application to turbulent velocity and vorticity 3D numerical data}
45:
46:
47: \author{Pierre Kestener}
48: \affiliation{CEA-Saclay, DSM/DAPNIA/SEDI, 91191 Gif-sur-Yvette, France}
49: \affiliation{Laboratoire de Physique, Ecole Normale Sup\'erieure de Lyon, 46 all\'ee d'Italie, 69364 Lyon c\'edex 07, France}
50: \author{Alain Arneodo}
51: \affiliation{Laboratoire de Physique, Ecole Normale Sup\'erieure de Lyon, 46 all\'ee d'Italie, 69364 Lyon c\'edex 07, France}
52:
53: \date{\today}
54:
55: \begin{abstract}
56: We use singular value decomposition techniques to generalize the
57: wavelet transform modulus maxima method to the multifractal
58: analysis of vector-valued random fields. The method is calibrated on
59: synthetic multifractal 2D vector measures and monofractal 3D
60: fractional Brownian vector fields. We report the results of some
61: application to the velocity and vorticity fields issued from 3D
62: isotropic turbulence simulations. This study reveals the existence of
63: an intimate relationship between the singularity spectra of these two
64: vector fields which are found significantly more intermittent than
65: previously estimated from longitudinal and transverse velocity
66: increment statistics.
67: \end{abstract}
68:
69: \pacs{47.53.+n, 02.50.Fz, 05.40-a, 47.27.Gs}
70:
71: \maketitle
72:
73: The multifractal formalism was introduced in the context of
74: fully-developed turbulence data analysis and modeling to account for
75: the experimental observation of some deviation to Kolmogorov theory
76: (K41) of homogenous and isotropic turbulence~\cite{bFri95}.
77: The predictions of various multiplicative cascade models, including
78: the weighted curdling (binomial) model proposed by
79: Mandelbrot~\cite{aMan74b}, were tested using box-counting (BC) estimates of
80: the so-called $f(\alpha)$ singularity spectrum of the dissipation
81: field~\cite{aMen91}. Alternatively, the intermittent nature of the
82: velocity fluctuations were investigated via the computation of the
83: $D(h)$ singularity spectrum using the structure function (SF)
84: method~\cite{aPar85}.
85: Unfortunately, both types of studies suffered from severe
86: insufficiencies. On the one hand, they were mostly limited by one
87: point probe measurements to the analysis of one (longitudinal)
88: velocity component and to some 1D surrogate
89: approximation of the dissipation~\cite{aAur92}.
90: On the other hand, both the BC and SF
91: methodologies have intrinsic limitations and fail to fully
92: characterize the corresponding singularity spectrum since only the
93: strongest singularities are a priori amenable to these
94: techniques~\cite{aMuz91aMuz94aArn95}.
95: In the early nineties, a wavelet-based statistical approach was
96: proposed as a unified multifractal description of singular measures
97: and multi-affine functions~\cite{aMuz91aMuz94aArn95}.
98: Applications of the so-called {\it wavelet transform modulus maxima}
99: (WTMM) method have already provided insight into a wide variety of
100: problems, e.g., fully developed turbulence, econophysics, meteorology,
101: physiology and DNA sequences~\cite{bArn95,bDis02}.
102: Later on, the WTMM method was generalized to 2D for multifractal
103: analysis of rough surfaces~\cite{aArnDec00b}, with very promising
104: results in the context of the geophysical study of the intermittent
105: nature of satellite images of the cloud
106: structure~\cite{aArn99caArnDec00c,bArn02} and the medical assist in
107: the diagnosis in digitized mammograms~\cite{bArn02,akes01}.
108: Recently the WTMM method has been further extended to 3D analysis
109: and applied to dissipation and enstrophy 3D numerical data issue from
110: isotropic turbulence direct numerical simulations
111: (DNS)~\cite{aKes03_prl,tKes03}.
112: Thus far, the multifractal description has been
113: mainly devoted to scalar measures and functions. In the spirit of a
114: preliminary theoretical study of self-similar vector-valued measures
115: by Falconer and O'Neil~\cite{aFal96}, our objective here is to
116: generalize the WTMM method to vector-valued random fields with the
117: specific goal to achieve a comparative 3D vectorial multifractal
118: analysis of DNS velocity and vorticity fields.
119:
120: Let us note $\mathbf{V}(\mathbf{x}=(x_1,x_2,..,x_d))$, a
121: vector field with
122: square integrable scalar components $V_j(\mathbf{x})$, $j=1,2,..,d$.
123: Along the line of the 3D WTMM
124: method~\cite{aKes03_prl,tKes03}, let us define $d$ wavelets
125: $\psi_i(\mathbf{x}) = \partial \phi(\mathbf{x})/\partial x_i$
126: for $i=1,2,..,d$ respectively, where $\phi(\mathbf{x})$ is a
127: scalar smoothing function well localized around $|\mathbf{x}|=0$.
128: The wavelet transform (WT) of $\mathbf{V}$ at point $\mathbf{b}$ and
129: scale $a$ is the following tensor~\cite{tKes03}:
130: \begin{equation}
131: {\mathbb T}_{\bpsi} [\mathbf{V}] ({\mathbf b},a) =
132: \begin{pmatrix}
133: T_{\psi_1}[V_1] & T_{\psi_1}[V_2] & ... & T_{\psi_1}[V_d]\\
134: T_{\psi_2}[V_1] & T_{\psi_2}[V_2] & ... & T_{\psi_2}[V_d]\\
135: \vdots & \vdots & & \vdots\\
136: T_{\psi_d}[V_1] & T_{\psi_d}[V_2] & ... & T_{\psi_d}[V_d]\\
137: \end{pmatrix},
138: \label{eq1}
139: \end{equation}
140: where
141: \begin{equation}
142: T_{\psi_i}[V_j](\mathbf{b},a) = a^{-d} \int d^d\mathbf{r}\; \psi_i
143: \bigl( a^{-1} (\mathbf{r} - \mathbf{b}) \bigr) V_j(\mathbf{r}).
144: \label{eq2}
145: \end{equation}
146: In order to characterize the local H\"older
147: regularity of $\mathbf{V}$,
148: one needs to find
149: the direction that locally corresponds to the maximum amplitude
150: variation of $\mathbf{V}$. This can be obtained from the {\it singular
151: value decomposition} (SVD)~\cite{bGol89} of the matrix $(T_{\psi_i}[V_j])$
152: (Eq.~(\ref{eq1})):
153: \begin{equation}
154: {\mathbb T}_{\bpsi} [\mathbf{V}] = \mathbb{G} \Sigma \mathbb{H}^T\, ,
155: \label{eq3}
156: \end{equation}
157: where $\mathbb{G}$ and $\mathbb{H}$ are orthogonal matrices
158: ($\mathbb{G}^T\mathbb{G}=\mathbb{H}^T\mathbb{H}=\mathbb{I}_d$) and
159: $\Sigma=diag (\sigma_1,\sigma_2,..,\sigma_d)$ with $\sigma_i \geq 0$,
160: for $1\leq i\leq d$.
161: The columns of $\mathbb{G}$ and $\mathbb{H}$ are referred to as the
162: left and right singular vectors, and the singular
163: values of ${\mathbb T}_{\bpsi} [\mathbf{V}]$
164: are the non-negative square roots $\sigma_i$ of the
165: $d$ eigenvalues of ${\mathbb T}_{\bpsi} [\mathbf{V}]^T{\mathbb
166: T}_{\bpsi} [\mathbf{V}]$.
167: Note that this decomposition is unique, up to some permutation
168: of the $\sigma_i$'s.
169: The direction of the largest amplitude variation of $\mathbf{V}$, at
170: point $\mathbf{b}$ and scale $a$, is
171: thus given by the eigenvector $\mathbf{G}_{\rho}(\mathbf{b},a)$
172: associated to the spectral radius $\rho(\mathbf{b},a)=\max_j
173: \sigma_j(\mathbf{b},a)$. One is thus led to the analysis of the vector
174: field ${\mathbf T}_{\bpsi,\rho} [\mathbf{V}](\mathbf{b},a) =
175: \rho(\mathbf{b},a) \mathbf{G}_{\rho}(\mathbf{b},a)$.
176: Following the WTMM analysis of scalar
177: fields~\cite{aArnDec00b,aKes03_prl,tKes03}, let us define, at a given
178: scale $a$, the WTMM as the position $\mathbf{b}$ where the modulus
179: $\Mpsi [\mathbf{V}](\gb,a) = |\Tpsirho [\mathbf{V}](\gb,a)| =
180: \rho(\gb,a)$ is locally maximum along the direction of
181: $\mathbf{G}_{\rho}(\mathbf{b},a)$.
182: These WTMM lie on connected $(d-1)$ hypersurfaces called {\it maxima
183: hypersurfaces} (see Figs \ref{fig2}b and \ref{fig2}c). In theory, at
184: each scale $a$, one only needs to record the position of the local
185: maxima of $\Mpsi$ (WTMMM) along the maxima hypersurfaces together with
186: the value of $\Mpsi[\mathbf{V}]$ and the direction of
187: $\mathbf{G}_{\rho}$. These WTMMM are disposed along connected curves
188: across scales called {\it maxima
189: lines}
190: living in a $(d+1)$ space
191: $(\mathbf{x},a)$. The WT skeleton is then defined as the set of
192: maxima lines that converge to the $(x_1,x_2,..,x_d)$ hyperplane in the
193: limit $a\rightarrow 0^+$ (see Fig.~\ref{fig2}d). The local H\"older
194: regularity of $\mathbf{V}$ is estimated from the power-law behavior
195: $\Mpsi[\mathbf{V}] \bigl( \ron{L}_{{\mathbf r_0}}(a) \bigr) \sim
196: a^{h(\gra{r}_0)}$ along the maxima line $\ron{L}_{{\mathbf r_0}}(a)$
197: pointing to the point ${\mathbf r_0}$ in the limit $a\rightarrow 0^+$,
198: provided the H\"older exponent $h({\mathbf r_0})$ be smaller than the
199: number $n_{\bpsi}$ of zero moments of the analyzing wavelet
200: $\bpsi$~\cite{remark}. As for scalar
201: fields~\cite{aMuz91aMuz94aArn95,aArnDec00b,aKes03_prl}, the
202: tensorial WTMM method consists in defining the partition functions:
203: \begin{equation}
204: {\mathcal Z}(q,a)=\sum_{{\mathcal L}\in {\mathcal L}(a)} \left (
205: \ron{M}_{\bpsi}[\mathbf{V}]({\mathbf r},a)\right)^q\; \sim \;
206: a^{\tau(q)}\, ,
207: \label{eq4}
208: \end{equation}
209: where $q \in \mathbb{R}$ and ${\mathcal L}(a)$ is the set of maxima
210: lines that exist at scale $a$ in the WT skeleton. Then by Legendre
211: transforming $\tau(q)$, one gets the singularity spectrum $D(h)=\min_q
212: (qh-\tau(q))$, defined as the Hausdorff dimension of the set
213: of points $\mathbf{r}$ where $h(\mathbf{r})=h$. Alternatively, one can
214: compute the mean quantities:
215: \begin{equation}
216: \begin{aligned}
217: h(q,a)=&
218: \sum_{{\mathcal L}\in{\mathcal L}(a)} \ln \left|
219: \ron{M}_{\bpsi}[\mathbf{V}]({\mathbf r},a) \right| \; W_{\bpsi}[\mathbf{V}](q,{\mathcal L}, a)\; ,
220: \raisebox{-0.5cm}{(5)}
221: \label{eq56}
222: \\
223: D(q,a)=&
224: \sum_{{\mathcal L}\in{\mathcal L}(a)} W_{\bpsi}[\mathbf{V}](q,{\mathcal L}, a) \; \ln
225: \bigl( W_{\bpsi}[\mathbf{V}](q,{\mathcal L}, a) \bigr) \; ,
226: \notag
227: \end{aligned}
228: \end{equation}
229: where
230: $W_{\bpsi}[\mathbf{V}](q,{\mathcal L}, a)=\bigl(\ron{M}_{\bpsi}[\mathbf{V}]({\mathbf r},a) \bigr)^q/{\mathcal Z}(q,a)$
231: is a Boltzmann weight computed from the WT skeleton. From the scaling
232: behavior of these quantities, one can extract $h(q) =\lim_{a\rightarrow 0^+} h(q,a)/\ln a$ and $D(q) =\lim_{a\rightarrow 0^+} D(q,a)/\ln a$
233: and therefore the $D(h)$ spectrum.
234:
235: As a test application of this extension of the WTMM method to the
236: vector situation, let us consider the self-similar 2D vector measures
237: supported by the unit square defined in Ref.~\cite{aFal96}.
238: As sketched in Fig.~\ref{fig1},
239: \begin{figure}
240: \centering
241: \subfigure[]{\includegraphics[scale=0.23]{fig1a}}
242: \subfigure[]{\includegraphics[scale=0.23]{fig1b}}
243: \subfigure[]{\includegraphics[scale=0.23]{fig1c}}
244: \caption{First construction steps of a
245: singular vector-valued measure supported by the unit square. The
246: norm of the four similitude $S_i$ are $p_1=p_4=1/2$, $p_2=2$ and
247: $p_3=1$~\cite{aFal96}.}
248: \label{fig1}
249: \end{figure}
250: from
251: step $n$ to step $n+1$, each square is divided into 4 identical
252: sub-squares and for each of these sub-squares, one defines a
253: similitude $S_i$ that transforms the vector $\mathbf{V}^{(n)}$ at step
254: $n$ into the vector $\mathbf{V}_i^{(n+1)}$ for the sub-square $i$ at
255: step $n+1$.
256: The
257: $\sigma$-additivity property of positive scalar measures is now
258: replaced by
259: the vectorial
260: additivity condition $\mathbf{V}^{(n)}=\sum_{i=1}^4
261: \mathbf{V}_i^{(n+1)}$. A straightforward calculation yields the
262: following analytical expression for the partition function scaling
263: exponents $\tau(q)$ (Eq.~(\ref{eq4})):
264: \addtocounter{equation}{1}
265: \begin{equation}
266: \tau(q) = -\log_2(p_1^q+p_2^q+p_3^q+p_4^q) -q,
267: \label{eq7}
268: \end{equation}
269: where $p_i$ ($i=1$ to 4), are the norms of the similitudes $S_i$. Note
270: that this formula is identical to the theoretical
271: spectrum of a non-conservative scalar multinomial measure distributed
272: on the unit square with the weights
273: $p_i$~\cite{aKes03_prl,tKes03}.
274: Indeed, if the construction process in Fig.~\ref{fig1} is
275: conservative from a vectorial point of view, it does not conserve the
276: norm of the measure: $\sum_{i=1}^4 p_i=4>1$.
277: From Legendre
278: transforming Eq.~(\ref{eq7}), one gets a $D(h)$ singularity spectrum
279: with a characteristic multifractal single-humped shape (see Fig.~\ref{fig3}d)
280: supported by the interval
281: $[h_{\min},h_{\max}]=[-1-\log_2(\max_i p_i),-1-\log_2(\min_i p_i)]$
282: and whose maximum $D_F=-\tau(0)=2$ is the signature that the considered
283: vector-valued measure
284: is almost everywhere singular on the unit square.
285:
286: In Fig.~\ref{fig2}
287: \begin{figure}
288: \begin{minipage}[c]{.45\linewidth}
289: \includegraphics[width=4.2cm]{fig2a}
290: \end{minipage}
291: \hfill
292: \begin{minipage}[c]{.45\linewidth}
293: \includegraphics[width=4.2cm]{fig2b}
294: \end{minipage}
295: %
296: \begin{minipage}[c]{.45\linewidth}
297: \includegraphics[width=4.2cm]{fig2c}
298: \end{minipage}
299: \hfill
300: \begin{minipage}[c]{.45\linewidth}
301: \includegraphics[width=4.2cm]{fig2d}
302: \end{minipage}
303: \caption{
304: \DD{} \WT{} analysis of the 2D vector-valued self-similar
305: measure shown in Fig.~\ref{fig1} but with systematic random
306: permutation of the $S_i$.
307: $\bpsi$ is a first-order
308: analyzing wavelet ($\phi(\gra{r})$ is the Gaussian).
309: (a) 32 grey-scale coding of the central $(128)^2$ portion of the
310: original $(1024)^2$ field.
311: In (b) $a=2^2\sigma_W$ and (c) $a=2^3\sigma_W$, are shown the
312: maxima chains; from the local maxima (\WTMMM{}) of $\Mpsi$
313: along these chains ($\carreNoir$) originates a
314: black arrow whose length is proportional to $\Mpsi$ and direction
315: is along $\Tpsirho[\mathbf{V}]$.
316: (d) \WT{} skeleton obtained by
317: linking the \WTMMM{} across scales. $\sigma_W=7$ (pixels) is
318: the characteristic size of $\bpsi$ at the smallest resolved scale.
319: \label{fig2}
320: }
321: \end{figure}
322: are illustrated the main steps of our tensorial WT methodology when
323: applied to 16 $(1024)^2$ realizations of a random generalization of the
324: vectorial multiplicative construction process described in
325: Fig.~\ref{fig1}. Focusing on the central $(128)^2$ sub-square, we show
326: the singular vector-valued measure (Fig.~\ref{fig2}a) and the
327: corresponding WTMM chains
328: computed with a first order analyzing wavelet at two different scales
329: (Figs~\ref{fig2}b and \ref{fig2}c).
330: On these maxima chains, the black dots correspond to the
331: location of the WTMMM at these scales. The size of the
332: arrows that originate from each black dot is
333: proportional to the spectral radius $\rho(\mathbf{b},a)$ and its
334: direction is along the
335: eigenvector $\mathbf{G}_{\rho}(\mathbf{b},a)$.
336: When linking these WTMMM across scales, one gets the set of maxima lines
337: shown in Fig.~\ref{fig2}d as defining the WT skeleton.
338: In Fig.~\ref{fig3}
339: \begin{figure}
340: \centering
341: \includegraphics[scale=0.48,angle=-90]{fig3}
342: \caption{
343: Multifractal analysis of the 2D vector-valued random
344: measure field
345: using the \DD{} tensorial \WTMM{} method
346: ($\carreBlanc$) and BC techniques ($\carreNoir$).
347: (a) $\log_2 {\cal Z}(q,a)$ vs $\log_2 a$;
348: (b) $h(q,a)$ vs $\log_2 a$;
349: the solid lines correspond
350: to linear regression fits over
351: $\sigma_W \lesssim a \lesssim 2^{4}\sigma_W$.
352: (c) $\tau(q)$ vs $q$; the solid line
353: corresponds to the theoretical prediction (Eq.~(\ref{eq7})).
354: (d) $D(h)$ vs $h$;
355: the solid line is the Legendre transform of Eq.~(\ref{eq7}).
356: \label{fig3}
357: }
358: \end{figure}
359: are reported the results of the computation of the multifractal
360: spectra (annealed averaging).
361: As shown in Fig.~\ref{fig3}a, ${\mathcal Z}(q,a)$
362: (Eq.~(\ref{eq4})) display nice scaling behavior over four octaves (when
363: plotted versus $a$ in a logarithmic representation), for $q\in ]-2,4[$
364: for which statistical convergence turns out to be achieved. A linear
365: regression fit of the data yields the nonlinear $\tau(q)$ spectrum
366: shown in Fig.~\ref{fig3}c, in remarkable agreement with the theoretical
367: spectrum (Eq.~(\ref{eq7})). This multifractal diagnosis is confirmed
368: in Fig.~\ref{fig3}b where the slope of $h(q,a)$ (Eq.~(\ref{eq56}))
369: versus $\log_2 a$, clearly depends on $q$. From the estimate of $h(q)$
370: and $D(q)$ (Eq.~(\ref{eq56})), one gets the
371: single-humped $D(h)$ curve shown in Fig.~\ref{fig3}d which matches
372: perfectly the theoretical $D(h)$ spectrum.
373: In Fig.~\ref{fig3}, we have reported
374: for comparison, the results obtained when using a box-counting (BC)
375: algorithm adapted to the multifractal analysis of singular
376: vector-valued measures~\cite{tKes03,aFal96,remark2}. There is no doubt
377: that BC provides much poorer results, especially concerning the
378: estimates of $\tau(q)$, $h(q)$ and $D(q)$ for
379: negative $q$ values. This deficiency mainly results
380: from the fact that the vectorial resultant may be very small whereas
381: the norms of the vector measures in the sub-boxes are
382: not small at all. The results reported in Fig.~\ref{fig3}
383: bring the demonstration that our tensorial WTMM methodology paves the
384: way from multifractal analysis of singular scalar measures to singular
385: vector measures.
386:
387: In Fig.~\ref{fig4}
388: \begin{figure}
389: \centering
390: \includegraphics[scale=0.48,angle=-90]{fig4}
391: \caption{
392: Multifractal analysis of L\'ev\^eque DNS velocity
393: ($\rondNoir$) and vorticity ($\rondBlanc$)
394: fields ($d=3$, 18 snapshots)
395: using the tensorial \TD{} \WTMM{} method;
396: the symbols ($\carreNoir$) correspond to a similar analysis of
397: vector-valued fractional Brownian motions, $\mathbf{B}^{H=1/3}$.
398: (a) $\log_2 {\cal Z}(q,a)$ vs $\log_2 a$;
399: (b) $h_{\boldsymbol{\omega}}(q,a)$ vs $\log_2 a$ and
400: $h_{\mathbf{v}}(q,a)-\log_2a$ vs $\log_2 a$;
401: the solid and dashed lines correspond
402: to linear regression fits over
403: $2^{1.5}\sigma_W \lesssim a \lesssim 2^{3.9}\sigma_W$.
404: (c) $\tau_{\mathbf{v}}(q)$,
405: $\tau_{\boldsymbol{\omega}}(q)$ and
406: $\tau_{\mathbf{B}^{1/3}}(q)$ vs $q$;
407: (d) $D_{\mathbf{v}}(h+1)$,
408: $D_{\boldsymbol{\omega}}(h)$ vs $h$;
409: the dashed lines correspond to log-normal regression fits with
410: the parameter values
411: $C_2^{\mathbf{v}}=0.049$ and
412: $C_2^{\boldsymbol{\omega}}=0.055$; the dotted line is
413: the experimental singularity spectrum
414: ($C_2^{\delta v_{\!/\!\!/}}=0.025$) for 1D
415: longitudinal velocity increments~\cite{aArn98aMal00aDelArn01}.
416: \label{fig4}
417: }
418: \end{figure}
419: are reported the results of the application of our tensorial WTMM
420: method to isotropic turbulence DNS data obtained by L\'ev\^eque. This
421: comparative 3D multifractal analysis of the velocity ($\mathbf{v}$)
422: and vorticity ($\boldsymbol{\omega}$) fields corresponds to some
423: averaging over 18 snapshots of $(256)^3$ DNS run at
424: $R_\lambda=140$. As shown in Figs.~\ref{fig4}a and \ref{fig4}b, both
425: the $\mathcal{Z}(q,a)$ and $h(q,a)$ partition functions display
426: rather nice scaling properties for $q=-4$ to 6, except at small
427: scales ($a\lesssim 2^{1.5}\sigma_W$) where some curvature is observed
428: in the log-log plots likely induced by dissipation
429: effects~\cite{bFri95,aArn98aMal00aDelArn01}.
430: Linear regression fit of the data
431: (Fig.~\ref{fig4}a) in the range $2^{1.5}\sigma_W \leq a \leq
432: 2^{3.9}\sigma_W$ yields the nonlinear $\tau_{\mathbf{v}}(q)$ and
433: $\tau_{\boldsymbol{\omega}}(q)$ spectra shown in Fig.~\ref{fig4}c, the
434: hallmark of multifractality. For the vorticity field,
435: $\tau_{\boldsymbol{\omega}}(q)$ is a decreasing function similar to
436: the one obtained for the singular vector-valued measure in
437: Fig.~\ref{fig3}c; hence $h(q)$($=\partial \tau(q)/\partial q$)$<0$ and
438: the support of the $D(h)$ singularity spectrum expands over negative
439: $h$ values as shown in Fig.~\ref{fig4}d. In contrast
440: $\tau_{\mathbf{v}}(q)$ is an increasing function
441: which implies that $h(q)>0$ as the signature that $\mathbf{v}$ is a
442: continuous function. Let us point out that the so-obtained
443: $\tau_{\mathbf{v}}(q)$ curve significantly departs from the linear
444: behavior obtained for 18 $(256)^3$ realizations of vector-valued
445: fractional Brownian motions $\mathbf{B}^{1/3}$ of index $H=1/3$, in
446: good agreement with the theoretical spectrum
447: $\tau_{\mathbf{B}^{1/3}}(q)=q/3-3$. But even more remarkable, the
448: results reported in Fig.~\ref{fig4}b for $h(q,a)$
449: suggest, up to statistical uncertainty, the validity of
450: the relationship
451: $h_{\boldsymbol{\omega}}(q)=h_{\mathbf{v}}(q)-1$. Actually, as shown
452: in Fig.~\ref{fig4}d,
453: $D_{\boldsymbol{\omega}}(h)$ and $D_{\mathbf{v}}(h)$ curves are likely to
454: coincide after translating the later by one unit on the left.
455: This is to our knowledge the first numerical evidence that the
456: singularity spectra of $\mathbf{v}$ and $\boldsymbol{\omega}$ might be
457: so intimately related: $D_{\mathbf{v}}(h+1)=D_{\boldsymbol{\omega}}(h)$
458: (a result that could have been guessed intuitively by noticing that
459: $\boldsymbol{\omega}=\boldsymbol{\nabla}\wedge\mathbf{v}$ involves
460: first order derivatives only).
461: Finally, let us note that, for both fields, the $\tau(q)$ and $D(h)$
462: data are quite well fitted by log-normal parabolic
463: spectra~\cite{aArn98aMal00aDelArn01}:
464: \begin{equation}
465: \begin{aligned}
466: \tau(q)&=-C_0-C_1q-C_2q^2/2\, ,\\
467: D(h)&=\,C_0-(h+C_1)^2/2C_2\,.
468: \label{eq8}
469: \end{aligned}
470: \end{equation}
471: Both fields are found singular almost everywhere:
472: $C_0^{\mathbf{v}}=-\tau_{\mathbf{v}}(q=0)=D_{\mathbf{v}}(q=0)=3.02 \pm
473: 0.02$ and
474: $C_0^{\boldsymbol{\omega}}=3.01 \pm 0.02$.
475: The most frequent H\"older exponent $h(q=0)=-C_1$ (corresponding to
476: the maximum of $D(h)$) takes the value $-C_1^{\mathbf{v}}\simeq
477: -C_1^{\boldsymbol{\omega}}+1=0.34\pm0.02$. Indeed, this estimate is
478: much closer to the K41 prediction $h=1/3$~\cite{bFri95} than previous
479: experimental measurements ($h = 0.39\pm0.02$) based on the
480: analysis of longitudinal velocity
481: fluctuations~\cite{aArn98aMal00aDelArn01}.
482: Consistent estimates are obtained for $C_2$ (that characterizes the width of
483: $D(h)$):
484: $C_2^{\mathbf{v}}=0.049\pm0.003$ and
485: $C_2^{\boldsymbol{\omega}}=0.055\pm0.004$.
486: Note that these values are much larger than the experimental estimate
487: $C_2=0.025\pm0.003$ derived for 1D longitudinal velocity
488: increment statistics~\cite{aArn98aMal00aDelArn01}.
489: Actually they are comparable to the value
490: $C_2=0.040$ extracted from experimental transverse
491: velocity increments~[19b].
492:
493: To conclude, we have generalized the WTMM method to vector-valued
494: random fields. Preliminary applications to DNS turbulence data have
495: revealed the existence of an intimate relationship between the
496: velocity and vorticity 3D statistics that turn out to be significantly
497: more intermittent than previously estimated from
498: 1D longitudinal velocity increments statistics.
499: This new methodology looks very promising to many extents. Thanks to
500: the SVD, one can focus on fluctuations that
501: are locally confined in 2D ($\min_i \sigma_i=0$) or in 1D (the two
502: smallest $\sigma_i$ are zero) and then simultaneously proceed to a
503: multifractal and structural analysis of turbulent flows.
504: The investigation along this line of vorticity sheets and vorticity
505: filaments in DNS is in current progress.
506: We are very grateful to E. L\'ev\^eque for allowing us to have access
507: to his DNS data and to the CNRS under GDR turbulence.
508:
509:
510: \begin{thebibliography}{10}
511:
512: \bibitem{bFri95}
513: U. {Frisch}, {\em Turbulence} (Cambridge Univ. Press, Cambridge, 1995).
514:
515: \bibitem{aMan74b}
516: B.~B. {Mandelbrot}, J. Fluid Mech. {\bf 62}, 331 (1974).
517:
518: \bibitem{aMen91}
519: C. {Meneveau} and K.~R. {Sreenivasan}, J. Fluid Mech. {\bf 224}, 429 (1991).
520:
521: \bibitem{aPar85}
522: G. {Parisi} and U. {Frisch}, in {\em Turbulence and Predictability in
523: Geophysical Fluid Dynamics and Climate Dynamics},
524: edited by M. {Ghil} {\it et~al.} (North-Holland, Amsterdam,
525: 1985), p.\ 84.
526:
527: \bibitem{aAur92}
528: E. {Aurell} {\it et~al.}, J. Fluid Mech. {\bf 238}, 467 (1992).
529:
530: % reference 7
531: \bibitem{aMuz91aMuz94aArn95}
532: J.~F. {Muzy}, E. {Bacry}, and A. {Arneodo}, Phys. Rev. E {\bf 47}, 875
533: (1993); Int. J. of Bifurcation and Chaos {\bf 4}, 245 (1994); A. {Arneodo},
534: E. {Bacry}, and J.~F. {Muzy}, Physica A {\bf 213}, 232 (1995).
535:
536: \bibitem{bArn95}
537: A. {Arneodo} {\it et~al.}, {\em Ondelettes, Multifractales et Turbulences : de
538: l'ADN aux croissances cristallines} (Diderot Editeur, Art et Sciences, Paris,
539: 1995).
540:
541: \bibitem{bDis02}
542: {\em The Science of Disasters : climate disruptions, heart attacks and market
543: crashes}, edited by A. {Bunde}, J. {Kropp}, and H. {Schellnhuber} (Springer
544: Verlag, Berlin, 2002).
545:
546: \bibitem{aArnDec00b}
547: A. {Arneodo}, N. {Decoster} and S.~G. {Roux}, Eur. Phys. J. B {\bf
548: 15}, 567 (2000); N. {Decoster}, S.~G. {Roux}, and A. {Arneodo},
549: Eur. Phys. J. B {\bf 15}, 739 (2000).
550:
551: \bibitem{aArn99caArnDec00c}
552: A. {Arneodo}, N. {Decoster}, and S.~G. {Roux}, Phys. Rev. Lett. {\bf 83}, 1255
553: (1999). S.~G. {Roux}, A. {Arneodo}, and N. {Decoster},
554: Eur. Phys. J. B {\bf 15}, 765 (2000).
555:
556: \bibitem{bArn02}
557: A. Arneodo {\it et~al.}, Advances in Imaging and
558: Electron Physics, {\bf 126}, 1 (2003).
559:
560: \bibitem{akes01}
561: P. {Kestener} {\it et~al.}, Image Anal.
562: Stereol. {\bf 20}, 169 (2001).
563:
564: \bibitem{aKes03_prl}
565: P. Kestener and A. Arneodo, Phys. Rev. Lett. {\bf 91}, 194501 (2003).
566:
567: \bibitem{tKes03}
568: P. {Kestener}, Ph.D. thesis, University of Bordeaux I, 2003.
569:
570: \bibitem{aFal96}
571: K.~J. {Falconer} and T.~C. {O'Neil}, Proc. R. Soc. Lond. A {\bf 452}, 1433
572: (1996).
573:
574: \bibitem{bGol89}
575: G.~H. {Golub} and C.~V. {Loan}, {\em Matrix Computations}, 2nd ed. (John
576: Hopkins University Press, Baltimore, 1989).
577:
578: \bibitem{remark}
579: Note that if $h_i(\mathbf{r}_0)$ are the H\"older exponents
580: of the $d$ scalar components $V_i(\mathbf{r}_0)$ of $\mathbf{V}$, then
581: $h(\mathbf{r}_0)=\min_i h_i(\mathbf{r}_0)$.
582:
583: \bibitem{remark2}
584: We recall the relationship $h=\alpha-d$, between the H\"older exponent
585: $h$ and the singularity exponent $\alpha$ generally obtained by BC
586: techniques ($D(\alpha-d)=f(\alpha)$).
587:
588: \bibitem{aArn98aMal00aDelArn01}
589: (a) A. {Arneodo}, S. {Manneville}, and J.~F. {Muzy}, Eur. Phys. J. B
590: {\bf 1}, 129
591: (1998); (b) Y. {Mal\'ecot} {\it et~al.}, Eur. Phys. J. B {\bf 16},
592: 549 (2000);
593: (c) J. {Delour}, J.~F. {Muzy}, and A. {Arneodo}, Eur. Phys. J. B {\bf 23}, 243
594: (2001).
595:
596: \end{thebibliography}
597:
598: \end{document}
599: %
600:
601: