Multiscale velocity correlation in turbulence : experiments,
numerical simulations, synthetic signals
Citation for published version (APA):
Benzi, R., Biferale, L., Ruiz-Chavarria, G., Ciliberto, S., & Toschi, F. (1999). Multiscale velocity correlation in
turbulence : experiments, numerical simulations, synthetic signals. Physics of Fluids, 11(8), 2215-2224.
https://doi.org/10.1063/1.870083
DOI:
10.1063/1.870083
Document status and date:
Published: 01/01/1999
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
Download date: 16. Jun. 2020
PHYSICS OF FLUIDS
VOLUME 11, NUMBER 8
AUGUST 1999
Multiscale velocity correlation in turbulence: Experiments, numerical
simulations, synthetic signals
R. Benzi
AIPA, Via Po 14, 00100 Roma, Italy
L. Biferale
Dipartimento di Fisica, Università di Tor Vergata, Via della Ricerca Scientifica 1, I-00133 Roma, Italy
and INFM-Unitá di Tor Vergata, Roma, Italy
G. Ruiz-Chavarria
Departamento de Fisica, Facultad de Ciencias, UNAM 04510 Mexico D.F., Mexico
S. Ciliberto
Laboratoire de Physique, URA 1325, Ecole Normale Supérieure de Lyon, 46 Allée d’Italie,
69364 Lyon, France
F. Toschi
INFM-Unitá di Tor Vergata, Roma, Italy and Dipartimento di Fisica, Università di Pisa,
Piazza Torricelli 2, I-56126 Pisa, Italy
~Received 23 September 1998; accepted 3 May 1999!
Multiscale correlation functions in high Reynolds number experimental turbulence, numerical
simulations, and synthetic signals are investigated. Fusion Rules predictions as they arise from
multiplicative, almost uncorrelated, random processes for the energy cascade are tested. Leading
and subleading contribution, in the inertial range, can be explained as arising from a multiplicative
random process for the energy transfer mechanisms. Two different predictions for correlations
involving dissipative observable are also briefly discussed. © 1999 American Institute of Physics.
@S1070-6631~99!04208-7#
I. INTRODUCTION
In order to better characterize the transfer mechanism, it
is natural to look also at correlations among velocity fluctuations at different scales and at different times. The prototype
of such a class of correlations is
Understanding the statistical properties of intermittency
is one of the most challenging open problem of threedimensional fully developed turbulence.
Intermittency in the inertial range is usually analyzed by
means of the statistical properties of velocity differences,
d r v (x)5 v (x1r)2 v (x). In the following, being mainly interested in one-dimensional cuts of experimental, synthetic,
and numerical signals, we will disregard all vectorial dependencies in the velocity fields.
In the last 20 years,1 overwhelming experimental and
theoretical works focused on structure functions, S p (r)
5 ^ ( d r v (x)) p & . A wide agreement exists on the fact that
structure functions show a scaling behavior in the limit of
very high Reynolds numbers, i.e., in the presence of a large
separation between integral and dissipative scales, L/r d
→`,
S p~ r ! ;
SD
r
L
C p,q ~ r,R; t ! 5 ^ ~ d r v~ x,t !! p ~ d R v~ x,t1 t !! q & ,
where, hereafter we will always assume the obvious notation, r,R.
Unfortunately, the nontrivial time dependency of correlations such as ~2! is completely hidden, in a Eulerian
reference-frame, by the sweeping of small scales by large
scales. The ‘‘positive’’ side of sweeping is connected to the
Taylor hypothesis, i.e., to the possibility of identifying single
point measurements at a time delay t with single time measurements at separation scale r; t V̄, where V̄ is the large
scale sweeping velocity. The ‘‘negative’’ side of sweeping is
connected to the fact that the inertial time scales are always
subdominant with respect to the sweeping time. This implies
that in order to measure the temporal properties of the inertial range energy-transfer it is necessary to abandon the usual
Eulerian reference-frame and to move in a Lagrangian or
quasi-Lagrangian reference-frame where sweeping effects
are absent.2,3,5,7,9 Of course, from the experimental point of
view, it is much harder to measure the velocity field in a
Lagrangian frame than in the usual laboratory referenceframe. To our knowledge, the only results about multi-time
velocity correlations are purely theoretical5,9 or numerical.4,5
For this reason, in the following, we will limit to an experi-
z~ p !
.
~1!
The velocity fluctuations are anomalous in the sense that
z (p) exponents do not follow the celebrated dimensional
prediction made by Kolmogorov, z (p)5p/3. In fact, z (p)
are observed to be a nonlinear function of p, which is the
most important signature of the intermittent transfer of fluctuations from large to small scales.
1070-6631/99/11(8)/2215/10/$15.00
~2!
2215
© 1999 American Institute of Physics
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
2216
Benzi et al.
Phys. Fluids, Vol. 11, No. 8, August 1999
mental and theoretical analysis of simultaneous -single timemultiscale correlation functions only. Recently, some theoretical work6–9 and an exploratory experimental
investigation10 have been devoted to the behavior of singletime multiscale velocity correlations ~MSVC!,
F p,q ~ r,R ! [ ^ ~ v~ x1r ! 2 v~ x !! p ~ v~ x1R ! 2 v~ x !! q &
[ ^ ~ d r v~ x !! p ~ d R v~ x !! q & ,
~3!
with r d ,r,R,L. When the smallest among the two scales
r goes beyond the dissipative scales, r d , new properties of
the correlation functions ~3! may arise due to the nontrivial
physics of the dissipative cutoff.1,11 From now on, we will
mostly concentrate on correlation functions with both r and
R in the inertial range, only in the last section we will address the important point concerned with the crossover of the
dissipative scale. Moreover, in order to simplify our discussion, we will confine our analysis to the case of longitudinal
velocity differences.
The main purpose of this paper is to review and to extend a recent experimental and theoretical analysis of multiscale correlations ~3!.16 In particular, we present a systematic
analysis of multiscale correlation functions in different experimental setup, we also perform a critical comparison with
the same observable measured in synthetic turbulent signals
defined in terms of purely multiplicative random processes.
The comparison with the synthetic signals will allow us
to conclude that multiscale correlation functions are in quantitative agreement, with the prediction one obtains by using a
pure uncorrelated multiplicative process for the energy cascade, as long as both separations r,R are in the inertial range.
As for the case when one of the two separations is already in
the dissipative range we will critically review the two most
important different predictions one can obtain imposing the
dissipative cut-off using either multifractals15 or the GESS
phenomenology.11 Unfortunately, the actual state-of-the-art
experimental dissipative-scales data does not allow to clearly
distinguish among the two predictions.12,13
The paper is organized as follows. In Sec. II we briefly
review the ansatz that one simply obtains for MSVC ~3! by
using a multiplicative random process for the inertial-range
energy cascade. In Sec. III we discuss subleading corrections
induced by geometrical constraints which necessarily affects
any MSVC for finite separation of scales. These geometrical
constraints introduce subleading power laws behavior which
may strongly interfere with the leading multiplicative predictions for finite separation of scales r/R;O(1). In Sec. III
we also present our experimental data-analysis and the comparison with the synthetic multifractal field. In Sec. IV we
briefly address the problem of dissipative correlation functions. Conclusions follow in Sec. V.
II. CASCADE PROCESSES
Stochastic cascade processes are simple and well known
useful tools to describe the leading phenomenology of the
intermittent energy transfer in the inertial range. Both
anomalous scaling exponents and viscous effects1,11 can be
reproduced by choosing a suitable random process for the
multiplier, W(r,R), which connects velocity fluctuations at
two different scales, R.r.
The main idea turns around the hypothesis that small
scale statistics is fully determined by a cascade process conditioned to some large scale configuration,
d r v~ x ! 5W ~ r,R ! • d R v~ x ! ,
~4!
where, requiring homogeneity along the cascade process, the
random function W should depend only on the ratio r/R and
it is not-positive defined. In the hypothesis of negligible correlations among multipliers we obtain
F p,q ~ r,R ! 5C p,q
;
K F S D G LK F S D G L
W
r
R
p
S p~ r !
S
~ R !.
S p ~ R ! p1q
W
R
L
p1q
~5!
This expression was for the first time proposed in Ref. 6 and
later examined in more details in Ref. 7, where it was named
‘‘fusion rules.’’ In the same article the authors proved that
the fusion rule prediction gives the leading behavior of ~3!
when r/R→0 as long as some hypothesis of scaling invariance and of universality of scaling exponents in Navier–
Stokes equations hold. The name fusion-rules refers, probably, to the fact that thanks to the—supposed—uncorrelated
nature of the cascade process every multiscale correlation
can be written in terms of single scale correlations, i.e., structure functions.
Let us notice that, beside any rigorous claim, expression
~5! is also the zeroth order prediction starting from any multiplicative uncorrelated random cascade satisfying S p (r)
5 ^ @ W(r/R) # p & S p (R). Let us also stress that the fusion rules
prediction as stated in ~5! does not necessarily require any
scaling property of the underlying structure functions, a fact
which suggests that the validity of the statement should be
almost Reynolds independent.
In this paper we want to address three main questions; ~i!
whether the prediction ~5! gives the correct leading behavior
in the limit of large scales separation: r/R;0; ~ii! if this is
the case, what one can say about subleading behavior for
separation r/R;O(1); ~iii! what happens to those observable for which the ‘‘multiplicative prediction;’’ ~5! is incorrect because of symmetry reasons.
The last item comes from the observation that for correlation like
F 1,q ~ r,R ! 5 ^ ~ d r v !~ d R v ! q & ,
~6!
the multiplicative prediction gives
F 1,q ~ r,R ! 5
S 1~ r !
•S
~ R !.
S 1 ~ R ! 11q
Such a prediction is wrong because, if homogeneity can be
assumed, S 1 (r)[0 for all scales r. In this case prediction ~5!
does not represent the leading contribution.
In the following we propose a systematic investigation
of ~3! in high Reynolds number experiments,17,18 numerical
simulation,19 and synthetic signals.14 The main purpose consists in probing whether multiscale correlation functions may
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 11, No. 8, August 1999
Multiscale velocity correlation in turbulence . . .
show new dynamical properties ~if any! which are not taken
into account by the standard simple multiplicative models for
the energy transfer.
III. DATA ANALYSIS
In this section we present our main contributions by discussing the three items listed in the previous section and by
presenting a detailed data-analysis in experiments at different
Reynolds numbers, in numerical simulations and in synthetic
signals.
Experimental data sets come mainly from two different
laboratories. We have analyzed data obtained in a wind tunnel ~Modane! with Rel52000, the integral scale was L
;20 m and the dissipative scale was r d 50.31 mm. The second data set comes from a recirculating wind tunnel ~ENS de
Lyon! with a working section 3 m long and 50350 cm2 cross
section. Rel involved in experiments were 400 ~wake behind
a cilinder! and 800 ~jet turbulence!. Integral scales were 0.1
m and 0.125 m, respectively, whereas the dissipative scales
were 0.15 mm and 0.1 mm.
Synthetic signals are built in terms of a Wavelet decomposition with coefficients defined by a pure uncorrelated random multiplicative process.14 In the following, the comparison between the synthetic field and the experimental data
will play a central role in our discussion. Therefore, in the
Appendix, we briefly recall how a multiaffine field may be
synthesized—and analyzed—in terms of a wavelet representation.
In the Appendix we prove that such a signal shows the
fusion rules prediction ~5! and therefore it will turn out to be
an useful tool for testing how much deviations from ~5!,
observed in experiments or numerical simulations, are due to
important dynamical effects or only to unavoidable geometrical corrections. Let us proceed with a simple but basic
observation.
Notice that for any one-dimensional string of number
~such as the typical outcome of laboratory experiments in
turbulence! the multiscale correlations ~3! feel unavoidable
strong geometrical constraints. In particular, for any MSVC,
with two velocities at the same spatial point v (x) and the
two other velocities in a collinear geometry at spatial locations v (x1r) and v (x1R), like those analyzed in the following, we will always write down what we like to call the
‘‘ward-identities’’ ~WI!,
S p ~ R2r ! [ ^ @~ v~ x1R ! 2 v~ x !! 2 ~ v~ x1r ! 2 v~ x !!# p &
~7!
5
(
k50,p
b ~ k,p !~ 21 ! k F k,p2k ~ r,R ! ,
~8!
where b(k,p)5 p!/ @ k!( p2k)! # . For example, for p52 we
have
2F 1,1~ r,R ! [S 2 ~ r ! 1S 2 ~ R ! 2S 2 ~ R2r !
;
FS D
r
R
z~ 2 !
1O
S DG
r
R
•S 2 ~ R ! ,
~9!
where the latter expression has been obtained by expanding
S 2 (R2r) in the limit r/R→0.
2217
The ‘‘ward-identities’’ will turn out to be useful for understanding subleading predictions to the multiplicative cascade process. One may argue that in a geometrical setup
different from the one specified in ~3! the same kind of constraint will appear with eventually different weights among
different terms.
The most important result one must extract from ~8! is
that the multiscale correlation functions, as stated in ~3!, may
not be a perfect scaling functions even in the limit of very
high Reynolds number. Indeed, the WI tell us that MSVC
with different order of velocity moments must be connected
unavoidably one with the other, which would be in contrast
with the assumption that each MSVC should be determined
by a single power law behavior.
The main result presented in this work is that all multiscale correlations functions are well reproduced in their leading term, r/R→0, by a simple uncorrelated random cascade
~5! and that their subleading contribution, r/R;O(1), are
fully captured by the geometrical constrained previously discussed, namely the ‘‘ward-identities.’’
The recipe for calculating multiscale correlations is the
following: First, apply the multiplicative guess for the leading contribution and look for geometrical constraints in order
to find out sub-leading terms. Second, in all cases where the
leading multiplicative contribution vanishes because of underlying symmetries, look directly for the geometrical constraints and find out what is the leading contribution applying
the multiplicative random approximation to all, nonvanishing, terms in the WI.
A. Fusion rules: Even moments
Let us check the fusion rules prediction ~5! for even
moments p,q52,4,... . In order to better highlight the scaling
properties we will often use in the following, F̃ p,q (r,R), the
MSVC compensated with the fusion-rule prediction,
F̃ p,q ~ r,R ! 5
F p,q ~ r,R ! •S p ~ R !
.
S p ~ r ! •S p1q ~ R !
~10!
In order to compare experiments with different Reynolds
numbers we may use as independent variable in our plot the
quantity, x(R)[(R2r)/(L2r), where with L we intend the
integral scale of each different experiment. In this way, by
fixing the small scale r55 h and by changing r<R<L for
each set of data we have a variation of 0<x(R)<1. In Fig.
1 we have checked the large scale dependency by plotting
the compensated MSVC as a function of x[(R2r)/(L2r)
at fixed, r, for p52, q52 and different Reynolds numbers
~experimental and numerical!. Expression ~5! predicts the
existence of a plateau ~independent of R! at all scales r<R
<L, where the leading multiplicative description is correct.
From Fig. 1 one can see that experiments with low Re
numbers show a slightly poor plateau. In particular the direct
numerical simulation ~DNS! with Rel;40 does not show
any plateau. The absence of a plateau is connected to the
overwhelming geometrical effects present at such low Reynolds numbers ~see below!. For this reason, in the following
figures we will show only experimental data from Modane
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
2218
Phys. Fluids, Vol. 11, No. 8, August 1999
Benzi et al.
FIG. 1. Compensated MSVC F̃ 2,2(r,R) at fixed r and changing x(R)5(R
2r)/(L2r) for different experiments and numerical simulation: ~3! direct
numerical simulation (Rel540), ~1! jet (Rel5800), ~*! modane (Rel
52000), ~h! wake (Rel5400). Typical statistics are 100 eddy turn over
times for the DNS, and 50 eddy turn over times for the experimental signals.
wind tunnel, which have the highest Re number we can access.
In Fig. 2~a! we plot structure functions of order 2 and 4
for the highest Reynolds number data we had. As it is possible to see, a quantitative good agreement with the accepted
high-Reynolds number intermittent scaling is detectable for
almost two decades. On the other hand, in Fig. 2~b! we plot
the multiscale correlation function F 2,2(r,R) and F 2,4(r,R)
as a function of R at fixed r. As it is possible to see the
scaling is not as clear as was for the single-scale structure
function and indeed it is impossible to extract any quantitative measurements about scaling exponents even at such as
high Reynolds numbers (Rel52000). As already stated
above, in order to better appreciate the violation to the
fusion-rule scaling we plot in Fig. 2~c! the compensated correlation functions for two different set of moments. In the
limit of large separation R→L at fixed r, we indeed see a
tendency toward a plateau. On the other hand, there are clear
deviations for r/R;O(1). The same behavior is seen in Fig.
3 for the same compensated quantities fixing the large scale
R and by changing the small scale r. Such deviations show a
very slow decay as a function of the scale separation. The
decaying is so slow that a clear plateau is seen only for the
largest Reynolds number available. The question whether the
observed finite-size corrections have an important physical
origin or not is therefore of primary importance.
In order to understand the physical meaning of the observed deviations to the fusion rules ~5!, we compare, in Fig.
4, the experimental data against the equivalent quantities
measured by using the synthetic signal.
We notice an almost perfect superposition of the two
data sets, indicating that the deviations observed in real data
can hardly be considered a ‘‘dynamical effect.’’ For dynamical effect we mean a correlation of the energy cascade which
must be incorporated in the statistical properties of the multiplicative process.
Using the WI plus the multiplicative ansatz for the leading behavior of all correlation functions for p54 we quickly
read that the leading contribution to F 2,2 is O(r z (2) )
•O(R z (4)2 z (2) ), while subleading terms scale as O(r z (4) ),
FIG. 2. ~a! Log–log plot of the second order, S 2 (R), and fourth order,
S 4 (R), structure functions with superimposed the straight line expected for
the corresponding high-Reynolds number intermittent scalings. x-scale is in
unity of 16 Kolmogorov scales, y-units are arbitrary. ~b! Log–log plot of the
two-scale correlation function F 2,2(r,R) ~1! and F 2,4(r,R) ~3! at fixed r
and at changing R. The small scale r is fixed to be 16 Kolmogorov scales.
Units are as in ~a!. ~c! Compensated MSVC F̃ p,q (r,R) at fixed r and changing the large scale R for p52, q52 ~1! and p52, q54 ~3!. Units are as
in ~b!.
and as O(r z (3) )•O(R z (4)2 z (3) ). This superposition of power
laws is responsible for the slowly-decaying correlations in
Figs. 1–4.
In Table I we summarize the leading and subleading
contributions that may be inferred from the WI for the standard MSVC with p52, q52. Similar tables can be constructed for any other even MSVC.
The result so far obtained, i.e., that both the experimental data and the synthetic signal show the same quantitative
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 11, No. 8, August 1999
Multiscale velocity correlation in turbulence . . .
2219
TABLE I. Leading ~first column! and subleading ~second column! contribution to the different multiscale velocity correlations entering in the WI
written for p54. Notice that all the leading behaviors have been obtained by
using the multiplicative ansatz ~when applicable!. The subleading behaviors
are consistent with the constraints imposed by the WI.
Leading
z (4)
Subleading
S 4 (R2r)
R
F 3,1(r,R)
r z (3) R z (4)2 z (3)
r z~4!
R
R
r z~4!
R
R
S 4 (R2r),F 3,1(r,R)
r z (2) R z (4)2 z (2)
rz~3!Rz~4!2z~3!,
F 1,3(r,R)
F 2,2(r,R)
FIG. 3. Experimental compensated MSVC F p,q (r,R)/S p1q (R)•S p (r) at
fixed R and changing the small scale 1/r for p52, q52 ~1! and p52, q
54 ~3!.
behavior, is a strong indication that multiscale correlation
functions, at least for even order of the moments, i.e., in all
cases where the signal is not affected by cancellation problems, are in good agreement with the random multiplicative
model for the energy transfer.
An even stronger proof of this statement comes from the
analysis of multiscale correlations in terms of the coefficients
obtained by a Wavelet analysis of the experimental signal
~see Appendix!. The Wavelet coefficient a j,k may be seen as
the representative of a velocity fluctuation at scale r52 2 j
and centered in one of the k51,2,...,2 j spatial point chosen
equispaced in the original total length of the signal.
With this interpretation in mind, we may think at the
Wavelet coefficients as the ideal observable which minimize
the geometrical constraints and therefore as the ideal cases
where one can test the idea that behind the multiplicative
process there are only geometrical constraints. In other
words, in terms of the coefficients obtained by a wavelet
analysis of the experimental signal, the multiscale correlation
function should show the fusion rules prediction for a range
of scales much wider than for the velocity increments, i.e.,
geometrical constraints, which introduce subleading powerlaws decaying, should be minimized. Of course the degree of
elimination of geometrical constraints may depend on the
FIG. 4. Comparison between experimental and synthetic compensated
MSVC, F̃ p,q (r,R) at fixed r and changing the large scale R for p52, q
52: ~1! synthetic and ~3! experimental. For p52, q54: ~*! synthetic and
~h! experimental.
r z~4!
R
R
kind of analyzing wavelets used. For instance by analyzing
the synthetic signal with the same kind of wavelets we used
to build it we have a ‘‘perfect’’ elimination of all geometrical constraints and we find again the pure uncorrelated multiplicative process. On the other hand, had we used a different wavelet basis we would not have obtained in output
exactly the same coefficients we put in input. Nevertheless,
the physical intuition leads us to the claim that—a part from
pathological choices–using wavelets should always clean the
spurious effects described by the ward identities ~8!. An experimental proof supporting this intuition is shown in Fig. 5.
In Fig. 5 we show the equivalent of F 2,2(r,R) built in terms
of the Wavelet coefficients,
2j
2 j8
F wav
! 5 ^ u a j,k u 2 u a j 8 ,k 8 u 2 & .
2,2 ~ r52 ,R52
~11!
In the figure we plot, as in the previous figures, the compensated correlation, obtained from the wavelet coefficients, at
fixed small scale and at changing the large scale. In Fig. 6
the same quantities are plotted at changing the small scale.
As it is evident, the finite scale-separation effects visible in
the standard MSVC have here disappeared; the plateau is
reached immediately after, say, one fragmentation step.
B. A case where fusion rules fail
For multiscale correlations where the direct application
of the random-cascade prediction is useless ~because of the
FIG. 5. Comparison between real space ~1! and wavelet analysis ~3! of the
experimental data set from Modane. Compensated F̃ 2,2 is shown for fixed r
at varying R.
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
2220
Benzi et al.
Phys. Fluids, Vol. 11, No. 8, August 1999
FIG. 6. Comparison between real space ~1! and wavelet analysis ~3! of the
experimental data set from Modane. Compensated F̃ 2,2 is shown for fixed R
at varying r.
translation symmetry!, like F 1,q (r,R), we suppose that the
main leading contribution is simply due to the geometrical
constraints. In other words, we say that as soon as the main
leading effect induced by the presence of a multiplicative
random energy transfer is depleted because of symmetry reasons, the subleading contributions induced by the geometry
becomes the leading contributions.
In order to give a prediction for such class of MSVC we
therefore use the WI applying the multiplicative prediction to
all terms, except the F 1,q . One obtains the expansion
FSD
F 1,q ~ r,R ! ; O
r
R
z~ 2 !
1¯1O
1O
SD
r
R
z~ 3 !
SD G
r
R
1O
SD
r
R
z~ 4 !
z ~ q11 !
•S q11 ~ R ! ,
~12!
which coincides when q51 with the exact result ~9! using
z (3)51.
In Fig. 7 we show the experimentally measured F 1,2 and
the fit that we obtain by keeping only the first two terms of
the compensated expansion in ~12!, i.e., F 1,2(r,R)/S 3 (R)
5(a(r/R) z (2) 1b(r/R) z (3) ). The fit has been performed by
imposing the value for the scaling exponents z~2!, z~3! measured on the structure functions, i.e., only the coefficients in
front of the power laws, a,b have been fitted. As one can
FIG. 7. Experimental F 1,2(r,R) at fixed r516r d and at varying R. The
integral scale L;13104 r d . Let us remark that the observed change of sign
in the correlation implies the presence of at least two power laws. The
continuous line is the fit in the region r,R,L obtained by using only the
first two terms in ~12!.
FIG. 8. Comparison between compensated F̃ 3,1 odd MSVC with absolute
values ~1! and without ~3!. Data are shown for fixed r at varying R. It is
evident that the odd MSVC with absolute values has the same behavior of
even MSVC, while the one without absolute value does not follow the same
behavior.
notice, the fit works perfectly in the inertial range. Let us
remark that the correlation changes sign in the middle of the
inertial range, which is a clear indication that a single powerlaw fit ~neglecting subleading terms! would completely miss
the correct behavior.
Next we consider the WI for p53. Due to the fact that
S 3 (r);r in the inertial range, one can easily show that the
WI enforces F 12;F 21 . Therefore, we can safely state that
also correlation functions of the form F p,1 feel nontrivial
dependency from the large scale R, proving that the prediction given in Ref. 8 using isotropy arguments is wrong.
C. Fusion rules: Odd moments
For the most general MSVC involving odd moments of
velocity increments, F p,q (r,R) with p,q53,5,7,..., the situation is slightly more confused. The problem stems from the
fact that the fusion rules contribution to this correlation feels
indeed the skewed part of the process which is order of magnitudes less important than the even part. For example, the
multiplicative contribution to the correlation F 3,1(r,R)
would be S 3 (r)/S 3 (R)•S 4 (R) which is different form zero
only due to the fact that the process for the longitudinal
velocity correlation is skewed.
The weakness of the signal from the multiplicative contributions makes these class of correlation functions very
hard to analyze from the point of view of scaling. Here, the
geometrical constraints may well be more important, in a
large range of scale separation, than the fusion rules prediction. For example in Fig. 8 we plot the standard MSVC for
p53, q51 and the same correlations but with moduli of
velocity increments, such as to get rid, in the second case, of
cancellation effects. As it is evident, the two correlation have
a very different amplitude as soon as the scale separation
becomes important and it is hard to say whether the MSVC
without moduli follow the fusion rules prediction for large
scale separation or not. On the other hand, the correlation
with absolute values does follow the multiplicative prediction reaching a plateau after the usual finite size transient as
the ordinary even-MSVC.
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 11, No. 8, August 1999
Multiscale velocity correlation in turbulence . . .
A high statistics and high Reynolds number investigation of such a class of correlation may well be of some interest in order to elucidate whether the odd part of velocity
increments follows the same physics of the even part or not.
1 d 2S 4~ R !
5 lim r 22 3F 2,2~ r,R ! 2 r 2
1O ~ r 3 !
2
4
dR
r→0
In this section we discuss the application of fusion rules
in the dissipative range. We will be mainly interested in the
following two quantities:
A n ~ R ! 5 ^ D v~ x ! • d v R ~ x ! n & ,
~13!
B p,n ~ R ! 5 ^ T ~ x ! p • d v R ~ x ! n & ,
~14!
where D v (x) is the Laplacian computed at the point x,
d v R (x)5 v (x1R)2 v (x) and T(x) is the velocity gradient
computed at x. In order to simplify the discussion we restrict
to the one-dimensional case, namely, the Laplacian and the
gradient are computed in one dimension and velocity differences are longitudinal. Moreover we restrict our analysis to
the cases of n odd and n1p even. Our findings will anyway
be valid in the most general case. The scaling properties of
A n and B p,n have been investigated in Refs. 7, 12, and 16.
We start by considering the scaling properties of A n . By
its definition we have
r→0
KS
v~ x1r ! 1 v~ x2r ! 22 v~ x !
r2
• ~ v~ x1R ! 2 v~ x !! n
D
L
5 lim r 22 ~ F 1,n ~ r,R ! 1F 1,n ~ 2r,R !! .
~15!
r→0
In order to understand how Eq. ~15! works, we compute the
easiest quantities, i.e., A 1 and A 3 . By using Eq. ~9!, we
obtain
r 22
~ 2S 2 ~ R ! 12S 2 ~ r ! 2S 2 ~ R2r ! 2S 2 ~ R1r !!
r→0 2
~16!
A 1 ~ R ! 5 lim
5 lim r 22
r→0
S
1 d 2S 2~ R !
S 2~ r ! 2 r 2
1O ~ r 3 !
2
dR 2
D
1 d 2S 2~ R !
.
5 ^~ ] xv ! & 2
2 dR 2
2
~17!
~18!
r→0
~20!
S 2~ r !
.
r2
r→0
24F 3,1~ r,R ! 24F 3,1~ 2r,R !
2S 4 ~ R2r ! 2S 4 ~ R1r !!
~21!
r→0
At this point it is quite easy to find the most general expression for A n , which is
A n ~ R ! 5nB 2,n21 ~ R ! 2
1 d 2 S n11 ~ R !
.
n11
dR 2
~22!
Equation ~22! is an exact results which is independent on any
physical assumption on the fusion rules. The last term on the
r.h.s. of ~22! becomes small for R in the inertial range. On
the other hand, for small value of R, i.e., for R→0, the last
term of the r.h.s. of ~22! cannot be neglected. In particular,
for R→0, an explicit computation, either using ~22! or ~13!,
gives @after cancellations of leading terms in the r.h.s. of
~22!#,
A n ~ R ! .O ~ R n11 ! .
~23!
In order to complete our computation for A n , we need an
estimate for B 2,n . There are in principle two ways to compute B p,n ; the first one using the multiscaling approach,15
the second one using the GESS theory discussed in Ref. 11;
We first analyze the case of multiscaling. In this case,
one can use the approach of mutliplicative processes with
multiscaling viscous cutoff.15 Namely, for the correlation
B 2,n (R)5 ^ ( ] x v ) 2 ( d R v (x)) n & one obtains
K
B 2,n ~ R ! ; ~ d R v~ x !! n
S DL
d rdv
rd
2
~24!
,
where r d is the dissipative scale. In the multifractal interpretation we assume d r d v 5(r d /R) h • d R v with probability
P h (r d ,R)5(r d /R) 32D(h) . Following Ref. 15 we have
d r d v •r d ;
SD
rd
R
h
d R v •r d ; n .
~25!
Inserting the last expression in the definition of B 2,n (R), we
finally have
E
dm~ h !
S
n
~19!
The computation of A 3 is similar and we find, using Eq. ~8!,
A 3 ~ R ! 5 lim r 22 41 ~ 2S 4 ~ R ! 12S 4 ~ r ! 112F 2,2~ r,R !
B 2,2~ R ! 5 lim r 22 F 2,2~ r,R ! .
B 2,n ~ R ! ;
In Eq. ~18!, we have used the relation
^ ~ ] x v ! 2 & 5 lim
1 d 2S 4~ R !
.
53B 2,2~ R ! 2 r 2
4
dR 2
D
In Eq. ~20! we used the definition of B 2,2 , namely,
IV. DISSIPATIVE PHYSICS
A n ~ R ! 5 lim
S
2221
•
R• d R v
~ d R v ! n12
R2
D
@ 2 ~ h21 ! 132D ~ h !# / ~ 11h !
;
S n13 ~ R !
,
n •S 3 ~ R !
~26!
where we have used the fact that the multifractal process is
such that n ^ ( ] x v ) 2 & →O(1) in the limit n →0. Expression
~26! coincides with the prediction given in Ref. 8. The above
computation,
are
easily
generalized
for
any
^ ( ] x v ) p ( d R v (x)) q & . By using ~26! and ~22! we finally obtain
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
2222
Benzi et al.
Phys. Fluids, Vol. 11, No. 8, August 1999
A n ~ R ! 5nC n
S n12 ~ R !
1 d 2 S n11 ~ R !
.
2
n S 3 ~ R ! n11
dR 2
~27!
Let us note that, for R→0 Eq. ~27! predicts that A n (R)
;O(R n21 ) which violates Eq. ~23!.
We now compute A n (R) by using the GESS approach
discussed in Ref. 11. In this case the computation of B 2,n can
be easily done by noting that, within the GESS approach, the
fluctuations of the dissipation scale are confined in the range
where d R v ;R. This implies that, for what concerns the scaling properties of B 2,n (R), the effect of a fluctuating dissipation scale can be disregarded. Following Ref. 11, after a long
but straightforward computation, we obtain
B 2,n ~ R ! 5D n
^ T 2 & S n12 ~ R !
S 2~ R !
~28!
,
where T is the velocity gradient and D n is a constant. Equation ~28! can be easily understood by noting that, within the
GESS approach, F 2,n (r,R); @ S 2 (r)•S n12 (R) # /S 2 (R) for
any values of r and R, i.e., also in the limit r→0. Using ~28!
we finally obtain
A n ~ R ! 5nD n
^ T 2 & S n11 ~ R !
S 2~ R !
2
1 d 2 S n11 ~ R !
.
n11
dR 2
~29!
For R→0, using the estimate S n (R); ^ T n & R n 1O(R n12 ),
and the fact that D n 51 for R50, we can easily show that
Eq. ~29! satisfies the constrain ~23!.
From an experimental point of view, it is extremely difficult to distinguish between the two predictions ~27! or ~29!.
We note that the experimental and numerical analysis discussed in Refs. 12 and 13, has been done neglecting the
second term on the r.h.s. of ~22!. Also, the experimental
analysis performed in Ref. 11 seems to indicate that multiscaling effects are not observed in real turbulence. At any
rate, no definitive conclusions can be drawn from existing
experimental data. A third, different parameterization of viscous effects has also been presented in Ref. 22.
V. CONCLUSIONS
Let us summarize what is the framework we have found
until now.
Whenever the simple scaling ansatz based on the uncorrelated multiplicative process is not prevented by symmetry
arguments, the multiscale correlations are in good
asymptotic agreement with the fusion rules prediction even if
strong corrections due to subleading terms are seen for
small-scale separation r/R;O(1). Subleading terms are
strongly connected to the WI previously discussed, i.e., to
geometrical constraints. In the other cases @i.e., F 1,q (r,R)#
the geometry fully determines both leading and subleading
scaling. All these findings, led us to the conclusions that
multiscale correlations functions measured in turbulence are
fully consistent with a multiplicative, almost uncorrelated,
random process. Nevertheless, the strong and slowlydecaying subleading corrections to the naive multiplicative
fusion rules predictions are the main obstacle for any attempts to attack analytically the equation of motion for structure functions; in that case, multiscale correlations at almost
coinciding scales are certainly the dominant contributions in
the nonlinear part of the equations.8 Indeed, as shown in an
analytical calculation for a dynamical toy model of random
passive-scalar advection,20 fusion rules are violated at small
scale-separation and the violations are relevant for correctly
evaluating the exact behavior of structure functions at all
scales.
Finally, let us remark that the standard multiplicative
process cannot be the end of the story, i.e., the dynamics is
certainly more complex than what here summarized. For example, one cannot exclude that also subleading ~with respect
to the multiplicative ansatz! dynamical processes are acting
in the energy transfer from large to small scales. These dynamical corrections must be either negligible with respect to
the geometrical constraints or, at best, of the same order. As
already discussed, a better understanding of this small deviations is necessary in order to improve our analytic control of
intermittent deviations in Navier–Stokes equations. Let us
cite, for example, a previous attempt, made by some of the
authors, to close the equation of motion in a dynamical
model of turbulence by using a simple multiplicative
process.28 In that case, one was able to find a satisfactory
qualitative agreement with the numerical simulations and at
the same time one could also prove that the closure was not
exact due to the presence of small out of control, deviations
from the multiplicative ansatz. Unfortunately, these deviations, as shown in this paper, are hardy detectable experimentally. Also, as shown in Sec. III C, the odd correlation
functions are not fully understood; higher Reynolds number
experiments, with higher statistics, are needed.
The question connected to the transfer properties of
quantities with different physical dimension from the energy,
say the helicity, may reveal different physical mechanisms.25
What happens for all those multiscale correlation functions
which feel a nontrivial helicity dependency for nonparity invariant flows is in this framework an open question.
In the past, similar analysis in a class of multiplicative
models for the energy dissipation have been done.26 Also in
that case, the multipliers connecting coarse-grained energy
dissipation over two overlapping intervals shows some weak
correlations among scales. This correlation can also be understood in terms of unavoidable geometrical effects due to
the overlapping nature of the intervals where the coarsegrained energy dissipation is defined.
For what concerns fusion rules involving velocity gradients or Laplacian and velocity differences, we observe that
there are controversial arguments leading to different predictions. It is difficult to distinguish which predictions is really
observed in real turbulence, because experimental data at
large Reynolds number do no resolve the far dissipative
range with enough accuracy.
Finally, let us mention that other possible candidates to
investigate the previous problems are shell models for turbulence, where geometrical constraints do not affect the energy
cascade mechanism.
ACKNOWLEDGMENTS
We acknowledge useful discussions with A. L. Fairhall,
V. L’vov, and I. Procaccia. M. Pasqui is kindly acknowl-
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
Phys. Fluids, Vol. 11, No. 8, August 1999
Multiscale velocity correlation in turbulence . . .
edged for his help in the analysis of the synthetic signal. We
are indebted to Y. Gagne for having allowed us the access to
the experimental data. L.B. and F.T. have been supported by
INFM ~PRA TURBO!. G.R.C. and S.C. acknowledge support by ECOS comitee and CONACYT under Project No.
M96-E03.
APPENDIX: SYNTHETIC SIGNALS
We build up a one-dimensional synthetic signal according to a random multiplicative process defined in a dyadic
hierarchical structure as originally introduced in Ref. 14 ~for
a review and references see also Ref. 21!.
Let us consider a wavelet decomposition of the function
f (x),
`
f~ x !5
(
j,k50
a j,k c j,k ~ x ! ,
~A1!
where c j,k (x)52 j/2c (2 j x2k) and c (x) is any wavelet with
zero mean. The above decomposition defines the signal as a
dyadic superposition of basic fluctuations with different
characteristic widths ~controlled by the index j! and centered
in different spatial points ~controlled by the index k!. For
functions defined on N52 n points in the interval @0, 1# the
sums in ~A1! are restricted from zero to n21 for the index j
and from zero to 2 j 21 for k.
In Ref. 14 it has been shown that the statistical behavior
of signal increments,
^ u d f ~ r ! u p & 5 ^ u f ~ x1r ! 2 f ~ x ! u p & ;r z ~ p !
is controlled by the coefficients a j,k . By defining the a coefficients in terms of a multiplicative random process on the
dyadic tree it is possible to give an explicit expression for the
scaling exponents z (p). For example, it is possible to recover the standard anomalous scaling by defining the a’s tree
in term of the realizations of a random variable h with a
probability distribution P( h ),
a 0,05const.
a 1,05 h 1,0a 0,0 ;
a 1,15 h 1,1a 0,0 ;
a 2,05 h 2,0a 1,0 ;
a 2,15 h 2,1a 1,0 ;
a 2,25 h 2,2a 1,1 ;
a 2,35 h 2,3a 1,1 ;
~A2!
and so on. Let us note that in the previous multiplicative
process different scales are characterized by different values
of the index j, i.e., r j 52 2 j . If the h j,k are independent identically distributed random variables it is straightforward to
realize that a j,k are random variables with moments given by
2log2 ~ h p !
^ u a j,k u p & 5r j
5r zj ~ p ! ,
~A3!
where the ‘‘mother eddy’’ a 0,0 has been chosen to be equal
to one. In ~A3! with ¯ we intend averaging over the P( h )
distribution. In Ref. 14 it has been shown that also the signal
f (x) has the same anomalous scaling of ~A3!.
The same arguments used in order to prove that the field
f (x) has an anomalous scaling can be invoked to show also
that the fusion-rules prediction ~5! are satisfied -at least for
large scale separation-.
2223
On the other hand, it is trivial matter to realize that the
above signal will show exactly, and for any separation of
scale, the fusion-rules prediction if expressed for the wavelet
coefficients a j,k . For example, let us consider two wavelet
coefficients at different scales r j ,r j 8 and let us chose the k
indices such that the two coefficients refer to two spatially
overlapping wavelets, then it is trivial to realize that, due to
the multiplicative nature of the wavelet coefficients, we have
^ u a j,k u p u a j 8 ,k 8 u q & [
S D
rj
r j8
z~ p !
z ~ p1q !
r j8
~A4!
which shows that the fusion rules prediction is satisfied exactly for any separation of scales as long as the two fluctuations are chosen with overlapping distances. In the case the
two distances are not overlapping, deviations from the fusion
rules prediction are certainly seen in the synthetic field due
to the dyadic ~ultrametric! nature of the underlying structure.
The question whether such deviations may be seen also in
the experimental data is an interesting point which is outside
the scope of this paper ~see, for similar problems, Refs. 23,
24!.
Other synthetic signals27 can be built either starting by a
multiplicative process from the energy dissipation, and using
the refined Kolmogorov hypothesis in order to have a signal
for the velocity, or by using ‘‘wavelet-like’’ tent functions.
Our signal, focusing directly on the velocity fields increments, without passing from the energy dissipation, and being infinitely differentiable, at difference from the tentfunction signal, looks more appropriate for our purposes.
1
U. Frisch, Turbulence: The Legacy of A. N. Kolmogorov ~Cambridge University Press, Cambridge, 1995!.
2
R. H. Kraichnan, ‘‘Lagrangian-history closure approximation for turbulence,’’ Phys. Fluids 8, 575 ~1965!.
3
V. I. Belinicher and V. L’vov, ‘‘A scale invariant theory of fully developed hydrodynamic turbulence,’’ Sov. Phys. JETP 66, 303 ~1987!.
4
C. Meneveau and T. S. Lund, ‘‘On the Lagrangian nature of the energy
cascade,’’ Phys. Fluids 6, 8 ~1994!.
5
L. Biferale, G. Boffetta, A. Celani, and F. Toschi, ‘‘Multitime multiscale
correlation functions in turbulence and in turbulent models,’’ Physica D
127, 187 ~1999!.
6
G. Eyink, ‘‘Lagrangian field theory, multifractals, and universal scaling in
turbulence,’’ Phys. Lett. A 172, 355 ~1993!.
7
V. S. L’vov and I. Procaccia, ‘‘Fusion rules in turbulent systems with flux
equilibrium,’’ Phys. Rev. Lett. 76, 2898 ~1996!.
8
V. S. L’vov and I. Procaccia, ‘‘Toward a nonperturbative theory of hydrodynamic turbulence: Fusion rules, exact bridge relations, and anomalous
viscous scaling functions,’’ Phys. Rev. E 54, 6268 ~1996!.
9
V. L’vov, E. Podivilov, and I. Procaccia, ‘‘Temporal multiscaling in hydrodynamic turbulence,’’ Phys. Rev. E 55, 7030 ~1997!.
10
A. L. Fairhall, B. Druva, V. S. L’vov, I. Procaccia, and K. S. Sreenivasan,
‘‘Fusion rules in Navier–Stokes turbulence: First experimental tests,’’
Phys. Rev. Lett. 79, 3174 ~1997!.
11
R. Benzi, L. Biferale, S. Ciliberto, M. V. Struglia, and R. Tripiccione, ‘‘A
generalized scaling of fully developed turbulence,’’ Physica D 96, 162
~1996!.
12
A. L. Fairhall, V. S. L’vov, and I. Procaccia, ‘‘Dissipative scaling functions in Navier–Stokes turbulence: Experimental tests,’’ chao-dyn/
9709034 ~unpublished!.
13
A. Celani and D. Biskamp, ‘‘Bridge relations in Navier–Stokes turbulence,’’ chao-dyn/9806027 ~unpublished!.
14
R. Benzi, L. Biferale, A. Crisanti, G. Paladin, M. Vergassola, and A.
Vulpiani, ‘‘A random process for the construction of a multiaffine field,’’
Physica D 65, 352 ~1993!.
15
U. Frisch and M. Vergassola, ‘‘A prediction of the multifractal model: The
intermediate dissipation range,’’ Europhys. Lett. 14, 439 ~1991!.
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions
2224
16
Benzi et al.
Phys. Fluids, Vol. 11, No. 8, August 1999
R. Benzi, L. Biferale, and F. Toschi, ‘‘Multiscale correlation functions in
turbulence,’’ Phys. Rev. Lett. 80, 3244 ~1998!.
17
H. Kahaleras, Y. Malecot, Y. Gagne, and B. Castaing, ‘‘Intermittency and
Reynolds number,’’ Phys. Fluids 10, 910 ~1998!.
18
R. Benzi, S. Ciliberto, C. Baudet, and G. Ruiz-Chavarria, ‘‘On the scaling
of three-dimensional homogeneous and isotropic turbulence,’’ Physica D
80, 385 ~1994!.
19
R. Benzi, F. Toschi, and R. Tripiccione, ‘‘On the heat transfer in
Rayleigh–Bénard systems,’’ J. Stat. Phys. 93, 901 ~1998!.
20
R. Benzi, L. Biferale, and A. Wirth, ‘‘Analytic calculation of anomalous
scaling in random shell models for a passive scalar,’’ Phys. Rev. Lett. 78,
4926 ~1997!.
21
G. Stolovitzky and K. R. Sreenivasan, ‘‘Kolmogorov’s refined similarity
hypothesis for turbulence and general stochastic processes,’’ Rev. Mod.
Phys. 66, 229 ~1994!.
22
G. Stolovitzky, K. R. Sreenivasan, and A. Juneja, ‘‘Scaling functions and
scaling exponents in turbulence,’’ Phys. Rev. E 48, R3217 ~1993!.
R. Benzi, L. Biferale, and E. Trovatore, ‘‘Ultrametric structure of multiscale energy correlations in turbulent models,’’ Phys. Rev. Lett. 79, 1670
~1997!.
24
J. O’Neil and C. Menevau, ‘‘Spatial correlations in turbulence: Predictions
from multifractal formalism and comparison with experiments,’’ Phys.
Fluids A 5, 158 ~1993!.
25
L. Biferale, D. Pierotti, and F. Toschi, ‘‘Helicity transfer in dynamical
models for turbulence,’’ Phys. Rev. E 57, R2515 ~1998!.
26
K. R. Sreenivasan and G. Stolovitzky, ‘‘Turbulent cascades,’’ J. Stat.
Phys. 78, 311 ~1995!.
27
A. Juneja, D. P. Lathrop, K. R. Sreenivasan, and G. Stolovitzky, ‘‘Synthetic turbulence,’’ Phys. Rev. E 49, 5179 ~1994!.
28
R. Benzi, L. Biferale, and G. Parisi, ‘‘On intermittency in a cascade model
for turbulence,’’ Physica D 65, 163 ~1993!.
23
Downloaded 07 Aug 2011 to 131.155.2.66. Redistribution subject to AIP license or copyright; see http://pof.aip.org/about/rights_and_permissions