1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %
3: % History: Version received from Luca Mon Oct 25 21:08:42 MEST 1999
4: % Version submitted to chao-dyn
5: %
6: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
7:
8: \documentstyle[12pt,epsfig,citesort]{article}
9: \textwidth 400pt
10: \oddsidemargin 8mm
11: \evensidemargin 8mm
12: \baselineskip 5mm
13: \textheight 580pt
14: \parindent=0pt
15:
16: \renewcommand{\baselinestretch}{1.}
17: \newcommand{\be}{\begin{equation}}
18: \newcommand{\ee}{\end{equation}}
19: \newcommand{\bdis}{\begin{displaymath}}
20: \newcommand{\edis}{\end{displaymath}}
21: \newcommand{\bx}{{\bf x}}
22: \newcommand{\bR}{{\bf R}}
23: \newcommand{\la}{\left\langle}
24: \newcommand{\ra}{\right\rangle}
25: \newcommand{\lp}{\left(}
26: \newcommand{\rp}{\right)}
27: \newcommand{\lab}{\left|}
28: \newcommand{\rab}{\right|}
29:
30:
31: \title{Statistics of pressure and of pressure-velocity
32: correlations in isotropic turbulence.}
33: \author{L. Biferale$^{1}$, P. Gualtieri$^2$ and F. Toschi$^{3}$}
34: \begin{document}
35: %\baselineskip 8ex
36: \maketitle
37: \centerline{$^1$Dipartimento di Fisica and INFM, Universit\`{a}
38: di Tor Vergata,}
39: \centerline{ Via della Ricerca Scientifica 1, I-00133 Roma, Italy.}
40: \centerline{$^2$Dip. di Meccanica e Aereonautica, Universit\'a di Roma ``La Sapienza'',}
41: \centerline{Via Eudossiana 18, 00184, Roma, Italy.}
42: \centerline{$^3$University of Twente, Department of Applied Physics and }
43: \centerline{J.M. Burgerscentrum for Fluid Dynamics,}
44: \centerline{P.O. Box 217, 7500 AE, Enschede, The Netherlands}
45: \centerline{and INFM, Unit\'a di Tor Vergata,}
46: \centerline{Via della Ricerca Scientifica 1, I-00133 Roma, Italy.}
47: \medskip
48:
49: \begin{abstract}
50: Some pressure and pressure-velocity correlations
51: in a direct numerical simulations of a three-dimensional
52: turbulent flow at moderate
53: Reynolds numbers have been analyzed.
54: We have identified a set
55: of pressure-velocity correlations which posseses a good scaling
56: behaviour. Such a class of pressure-velocity correlations
57: are determined by looking at the energy-balance across any sub-volume
58: of the flow. \\
59: According to our analysis,
60: pressure scaling is determined by the dimensional
61: assumption that pressure behaves as a ``velocity squared'', unless
62: finite-Reynolds effects are overwhelming.
63: The SO(3) decompositions of pressure
64: structure functions has also been applied
65: in order to investigate anisotropic effects on the pressure scaling.
66: \end{abstract}
67:
68: \section{Introduction}
69: Scaling in turbulent flows is one of the most challenging open issue in
70: fluid-dynamics \cite{frisch}.
71: Typical problems concern both the understanding of
72: the ideal case of isotropic and homogeneous turbulence in the limit of high
73: Reynolds numbers \cite{iso1,iso2} or more realistic and applied
74: situations with anisotropic and inhomogeneous statistics (for recent
75: examples see \cite{prl1,prl2,itamar1}).
76: In 1941, Kolmogorov, used a clever application of dimensional analysis
77: to predict that the scaling of velocity increments in the inertial range
78: should have a power law
79: behaviour depending only on the averaged energy dissipation
80: in the flow, $\epsilon$. Namely, for structure function of order $p$, i.e.
81: the p-th moment of a velocity difference across a distance $R$ we have:
82: \begin{equation}
83: S_q(R) = \la\lp v(x+R)-v(x)\rp^q\ra \sim \epsilon^{q/3}\,R^{q/3}
84: \label{sf}
85: \end{equation}
86: with $\eta\ll r\ll L_0$ where $\eta$ is the dissipative scale and $L_0$ is
87: the typical
88: external scale where forcing acts. Let us notice that in (\ref{sf}) we have
89: explicitly
90: neglected any tensorial structure in the velocity field such as to stress the
91: typical
92: dimensional character of the Kolmogorov theory, i.e.
93: the scaling properties
94: must be the same
95: for any observable which has the same physical dimension and which is built
96: in terms
97: of local field increments, $\delta_r v(x) =
98: v(x+r) -v(x)$.\\
99: Kolmogorov theory, as previously summarized, is quantitatively wrong.
100: Experiments and numerical simulations show a quantitative disagreement with
101: the dimensional prediction $p/3$ for the scaling exponents.
102: For example, the longitudinal velocity structure functions:
103: \begin{equation}
104: S^v_q\lp r\rp = \la \lab \lp v_i\lp\bx+\bR \rp-v_i\lp\bx\rp\rp\hat{R}_i\rab^q\ra
105: \sim R^{\zeta_v(p)}
106: \label{sf1}
107: \end{equation}
108: show a power law behaviour
109: with a set of exponents $\zeta_v(p)$ non linear in $p$. \\
110: The failure of the dimensional estimate $p/3$
111: goes under the name of anomalous scaling. \\
112: Many problems naturally arises as a consequence of the failure of the main
113: Kolmogorov prediction. The main open problem is to find an analytical
114: way to calculate from first principle the anomalous exponents, a
115: problem which is still out of control but for the
116: case of anomalous exponents characterizing the statistics of passive
117: quantities advected by gaussian velocity fields
118: \cite{kraich,gk,vergassola,dada,bbw,bif_magnetico}. \\
119: Another interesting question, opened by the failure of Kolmogorov
120: dimensional prediction, consists in the possibility
121: that local observable with the same physical dimensions but with
122: different tensorial structures have different scaling properties.
123: About this point, there are some experimental and numerical
124: evidences regarding different possible anomalous
125: behaviour
126: of longitudinal and transversal structure functions \cite{sreene}
127: even in isotropic turbulence. Somehow related to this issue is also
128: the apparent different anomalous scaling between the coarse grained
129: averages of dissipative quantities like
130: enstrophy and energy dissipation \cite{chen1}. On the other hand,
131: on the basis of a SO(3) decomposition of velocity correlation functions,
132: other authors \cite{alp} claim that the
133: supposed different scaling of quantities like transversal and longitudinal
134: structure functions can only be due to spurious sub-leading
135: non-isotropic effects, i.e. in isotropic high-Reynolds numbers all components
136: of the same tensorial observable should have the same -maybe anomalous-
137: scaling behaviour. A first numerical support to this claim
138: has been presented in the analysis of a channel flow simulation
139: in \cite{prl1}. \\ Even more complex is the situation when multi-point
140: pressure correlations is
141: involved \cite{iso1,hb97,press2}. Dimensionally speaking pressure is
142: just a velocity squared, and Kolmogorov-like argument can be easily
143: generalized
144: to the case of pressure structure functions, $F_q(r)$. Indeed, a simple
145: application of dimensional analysis leads to \cite{yaglom}:
146: \begin{equation}
147: F_q(r) \equiv \la\lab P(x)-P(x+r)\rab^q\ra \sim \epsilon^{2q} r^{2q}
148: \label{press}
149: \end{equation}
150: where as usual, all distances are supposed to belong to the inertial range
151: of scales. Of course, intermittency will also affect pressure scaling. By
152: following the straightforward hypothesis that pressure can be treated as a
153: velocity squared one would be tempted to
154: assign the same intermittency exponents of the velocity field to the
155: pressure scaling, i.e. to replace (\ref{press}) with
156: \begin{equation}
157: F_q(r) \sim r^{\zeta_v(2q)}
158: \label{press1}
159: \end{equation}
160: This prediction is just a simple consequences of the assumptions that all
161: velocity correlations have the same scaling behaviour supposed that all
162: distances involved are in the inertial range and that the statistics is
163: locally isotropic. Such a dimensional ansatz has been
164: questioned on the basis of a phenomenological argument in \cite{iso1},
165: some numerical
166: support to the latter argument
167: have been recently presented in \cite{press2}. \\
168: In this paper we will mainly present some numerical evidences that indeed
169: the dimensional ansatz (\ref{press1}) is correct. It is well known that this
170: must be the case at least for
171: for $q=2$ in (\ref{press1}). In this case, there exist
172: an exact relation \cite{hill} which connect the scaling of the second order
173: pressure structure function with a linear integral combination of
174: fourth-order velocity structure functions. The problem is if the
175: exact result can be simply extrapolated to other
176: pressure-dependent observable and, in the case,
177: how strong finite-Reynolds effects can be.
178: Indeed, one may argue that
179: pressure feels
180: strongly non-local effects, being just the inversion of the Poisson problem
181: $\Delta P = -\partial_i \partial_j v_i v_j$, and therefore the assumptions of
182: independence from large scales and/or from boundary conditions
183: may not be satisfied even at very high Reynolds numbers.
184: Indeed, to our knowledge,
185: neither experimental studies nor numerical simulations have ever been able
186: to make a firm quantitative statement about pressure scaling properties
187: \cite{nc98,hb97,press2,lesieur}. In this paper we
188: show that it is possible
189: to find a set of velocity-pressure observable which have indeed a quite good
190: scaling behaviour in agreement with the dimensional ansatz (\ref{press1})
191: also at moderate Reynolds numbers.\\
192: Scaling in turbulence is particularly
193: difficult to test in both experiments and numerical simulations. Experiments
194: reaches high Reynolds numbers by paying the price to have
195: a very limited set of information on the whole velocity fields,
196: typically only a long
197: time series of one velocity components in a few spatial points.
198: Moreover, in most cases, there
199: is not a precise control of the degree of isotropy and homogeneity in the
200: flow. On the other hand, numerical simulations have a
201: perfectly controlled lay out,
202: the velocity field is exactly known at any point,
203: but the maximum reachable Reynolds number is still
204: order of magnitude smaller then in typical experiments \cite{chen1024}.
205: \\ Nevertheless, numerical
206: simulations, if exploited in a clever
207: way, are the only tool where complex measurements can be performed. Therefore,
208: questions like the dependency of scaling properties from the tensorial
209: nature of the observable
210: can, up to know, be investigated only in numerical data base. \\
211: In this paper, we present a detailed analysis of pressure scaling and
212: pressure-velocity correlations scaling in a set of moderate Reynolds number
213: simulations. \\
214: Starting from the analysis of the energy transfer in real space
215: we propose a set of
216: pressure-velocity
217: observable which show better scaling properties then the usual
218: pressure structure functions. We presents {\it quantitative }
219: evidences that indeed, while pressure structure functions are strongly
220: affected from Reynolds numbers effects, the pressure-velocity correlation
221: functions we investigated have a fairly good scaling behaviour,
222: even at modest Reynolds numbers, in agreement
223: with the hypothesis that pressure "behaves" like a velocity squared. In
224: order to understand whether the bad scaling behaviour detected in the
225: pure-pressure structure functions is due to spurious anisotropic sub-leading
226: effects we also present some
227: results on the SO(3) decomposition of the pressure field. \\ The paper is
228: organized as follows. In section 2 we summarize the known analitycal result
229: which connect the second order pressure structure function to the integral
230: linear combination of fourth-order velocity correlations and the experimental
231: and numerical attempts to test the relation. In section 3 we introduce the
232: set of pressure-velocity correlations which
233: should have a better scaling properties on the basis of a simple argument
234: based on the
235: energy transfer of Navier-Stokes equations in the real space.
236: In section 4
237: we present the analysis of our numerical data base. In section 5
238: we briefly comment on the analysis of non isotropic fluctuations. Conclusions
239: follow in section 6.
240:
241: \section{Pressure structure functions}
242: Under the assumptions of local isotropy, local homogeneity, incompressibility,
243: and by use of Navier-Stokes equation, one can relate the second
244: order pressure structure functions, $F_2(r)$, to some fourth-order
245: velocity structure functions \cite{hill}. Namely:
246: \begin{eqnarray}
247: F_2(r) \equiv -\frac{1}{3} D_{1111}(r) + \frac{4}{3} r^2 \int_r^{\infty}
248: y^{-3}\left[D_{1111}(y)+D_{\beta\beta\beta\beta}(y)-6
249: D_{11\gamma\gamma}\right] dy \nonumber \\
250: + \frac{4}{3} \int_r^{\infty}
251: y^{-1}\left[D_{\beta\beta\beta\beta}(y)-3
252: D_{11\gamma\gamma}\right] dy
253: \label{s4p2}
254: \end{eqnarray}
255: where the fourth-order structure function is
256: $$D_{ijkl}(\vec{r})\equiv \la(u_i-u_i')(u_j-u_j')(u_k-u_k')(u_l-u_l')\ra$$
257: and where for simplicity we have used primed variables to express velocities at the position $\vec{x'}=\vec{x}+\vec{r}$ and where $i,j,k,l$
258: is $1$ if the velocity component is parallel to the separation
259: vector, $\vec{r}$, and $2,3$ otherwise. Subscripts $\beta,\gamma$
260: denote either $2$ or $3$. Of course, (\ref{s4p2}) implies that
261: whenever the fourth-order structure functions entering in the above
262: expressions are all dominated by the inertial-range intermittent
263: scaling behaviour, $D_{i,j,k,l}(r) \sim r^{\zeta_v(4)}$, then
264: also the second order pressure structure functions should scale with the
265: exponent $\zeta_p(2) = \zeta_v(4)$. Relation (\ref{s4p2}) have been carefully
266: tested in numerical simulations without any appreciable deviations
267: \cite{hb97,nc98}. Nevertheless, the overall scaling behaviour
268: of the pressure structure function is very poor.
269: Similarly, the analysis
270: of experimental data \cite{nc98} does not show any power law behaviour
271: for the pressure structure functions even if the Reynolds number
272: was extremely high
273: ($Re_{\lambda} \sim 10000$). In the latter case, authors tried to explain
274: the difference between pressure scaling quality and velocity
275: scaling quality by invoking a possible different scaling
276: for the different velocity correlations entering in the RHS
277: of (\ref{s4p2}), leading
278: to the final prediction that pressure structure functions is made in terms
279: of different power law contributions with slightly different exponents.
280: The resulting superposition of power laws would be the responsible
281: of the poor observed scaling behaviour. This statement would anyhow
282: contradict the theoretical prediction made in terms of the SO(3) decomposition
283: which forbids different component of the same tensorial observable to
284: scale differently in a isotropic ensemble. \\
285: Another interesting remark consists in the strong cancellation among
286: the different contribution of (\ref{s4p2}) observed in numerical simulations
287: \cite{hb97,nc98}: the LHS of (\ref{s4p2}) is more then an order of magnitude
288: smaller than the single different contributions entering in the RHS.
289: One, cannot exclude apriori the possibility that there exist
290: an almost perfect cancellations of all leading scaling
291: terms of all contribution appearing in the RHS of (\ref{s4p2}),
292: even if such a perfect cancellation would call for some unknown physical
293: interpretation.\\ More probably, the cancellation is not perfect but strong
294: enough to hide completely the pressure scaling at the available experimental
295: and numerical Reynolds numbers. \\ On the other hand,
296: the possibility that pressure-increments behave as
297: velocity-increments, $\delta P \sim \delta v$,
298: instead than as a velocity-increment squared has been recently proposed
299: \cite{press2}.
300: This would violate
301: the exact results previously reported and therefore cannot be correct
302: unless strong anisotropic effects are present at all scales. \\
303: In order to better assess the pressure
304: statistical properties we present in the following,
305: some results for pressure-dependent observable. This observable
306: do not posses the strong cancellation properties showed by
307: structure function.
308:
309: \section{Pressure-velocity correlation}
310:
311: Let us start by looking at the energy balance inside any volume $V$
312: of the flow. From the Navier-Stokes eqs we obviously have:
313: \begin{equation}
314: \partial_t E_V +
315: \int_V v_i(\bx)v_j(\bx)\partial_jv_i(\bx) d\bx +
316: \int_V v_i(\bx)\partial_i P(\bx) d\bx = \nu \int_V v_i(\bx)\Delta v_i(\bx)
317: d\bx
318: \label{en_ba}
319: \end{equation}
320: where with $E_V= 1/2 \int_V d \bx v_i v_i$
321: we denote the total energy in the sub-volume $V$.
322: %Let us now average over many realization without necessarly averaging
323: %over the whole space. We also assume stationarity and we neglect
324: %the time derivative in what follows.
325: Let us notice that the two terms in the LHS of (\ref{en_ba}) can be obviously written as the fluxes across the boundaries of $V$ by using Gauss theorem:
326: \begin{equation}
327: \int_V v_i(\bx)v_j(\bx)\partial_jv_i(\bx) d\bx \equiv \int_{\Omega_V}
328: d\Sigma n_j v_j(\bx) v^2(\bx) \equiv \Phi_{\Omega_V} ({\bf v} v^2)
329: \end{equation}
330: \begin{equation}
331: \int_V v_i(\bx)\partial_i P(\bx) d\bx \equiv \int_{\Omega_V}
332: d\Sigma n_i v_i(\bx) P(\bx) \equiv \Phi_{\Omega_V}({\bf v} P)
333: \end{equation}
334: where with $n_i$ we denote the versor perpendicular to the
335: infinitesimal surface on the boundaries
336: of $V$. Rewritten in this way, relation (\ref{en_ba}) is just a
337: simple restatement of the conservation of energy:
338: the total energy change inside a volume is given by the flux across
339: the volume surface and by the energy dissipation
340: inside the volume. Let us now use this simple fact in order to
341: extract some useful guess about scaling properties of velocity-pressure
342: correlations. Let us consider a very particular class of volume $V$, i.e.
343: a cylinder with an infinitesimal squared basis of surface $\epsilon$ and with
344: a finite axis in the direction of $\bf{R}$. In the limit when the
345: basis becomes smaller and smaller, the flux across the lateral
346: sides goes to zero because contributions from two opposite walls
347: are equal but with different sign. The only contributions to the
348: total flux come from the two infinitesimal basis and can be
349: written as
350: \begin{equation}
351: \Phi_{\Omega_V}({\bf v} v^2) = \epsilon( v_i(\bx)v^2(\bx)-
352: v_i(\bx +\bR)v^2(\bx+\bR))\hat{R}_i
353: \end{equation}
354: for the flux involving the velocity correlation and as
355: \begin{equation}
356: \Phi_{\Omega_V}({\bf v} P ) = \epsilon( v_i(\bx)P(\bx)-
357: v_i(\bx +\bR)P(\bx+\bR))R_i
358: \end{equation}
359: for the flux involving pressure-velocity correlations.
360: In both cases we have exploited
361: the fact that the two infinitesimal basis are
362: centered in $\bx$ and in $\bx +\bR$, i.e. their versor is
363: oriented along
364: $\hat{R}$. Similarly the two volume integral giving the time
365: variation of the total energy and the energy dissipation becomes two
366: linear integral times the infinitesimal basis area $\epsilon$:
367: $\partial_t (\epsilon \int_{\bR} ds v^2(\bx))$ and
368: $(\epsilon \int_{\bR} ds v_i(\bx)\Delta v_i(\bx))$ respectively,
369: where with $ds$ we parametrized the segment
370: going from $\bx$ to $\bx+\bR$. \\Let us now
371: assume that all the four
372: observable entering in the energy balance have the same statistical
373: behaviour.
374: This is somehow a ``local Kolmogorov refined hypothesis'': we link the
375: scaling of the local energy dissipation to the scaling of
376: some particular third order velocity correlation and to the
377: scaling of a velocity-pressure correlation. The claim is therefore
378: that the particular structure functions emerging from our flux
379: analysis should scale exactly like the coarse grained energy dissipation,
380: i.e. should have an anomalous scaling like the usual
381: longitudinal structure functions which satisfy
382: the original Kolmogorov Refined Hypothesis. In the next section
383: we present some numerical data in support of this claim.
384:
385: \section{Numerical Analysis}
386: The data set we are going to analyze has been obtained from
387: a direct numerical integration of NS eqs using a pseudo-spectral method
388: with dealiasing on a grid of $128^3$ points. The forcing was
389: implemented isotropically on all wavevectors with $|k|<1$ such as to
390: enforce the $k^{-5/3}$ spectrum at small wavectors \cite{forcing}.
391: We have analyzed about
392: 100 configurations stored each eddy-turn over time. The
393: simulation has a $Re_{\lambda}=70$.\\
394: Let us denote with:
395: \begin{eqnarray}
396: S_{q}^{vv^2}(R)& = &\la\lab\la v_i(\bx)v^2(\bx)-
397: v_i(\bx +\bR)v^2(\bx+\bR)\ra\hat{R}_i\rab^{q/3}\ra \nonumber \\
398: S_{q}^{vP}(R)& = &\la\lab\lp v_i(\bx)P(\bx)-
399: v_i(\bx +\bR)P(\bx +\bR)\rp\hat{R}_i\rab^{q/3}\ra
400: \label{sf3}
401: \end{eqnarray}
402: the two different structure functions which can be made in terms of the
403: two flux quantities defined in the previous section.
404: Let us notice that in (\ref{sf3}) both
405: $ S_{q}^{vv^2}(R)$ and $S_{q}^{vP}(R) $ have been defined as the $q/3$ power
406: of the original fluxes such as to have the same dimensions of $S^v_{q}(R)$.\\
407: Let us start by showing in Fig.1
408: the strong cancellation effects present in the pressure
409: structure functions $F_{q}(R)$ with respect
410: to the velocity longitudinal structure functions with the
411: same physical dimensions $S_{2q}^v(R)$. In Fig. 1 we show
412: the log-log plot of $F_1(R)$ and of $S_2(R)$,
413: as one can see the overall amplitude
414: of pressure fluctuations is about
415: one order of magnitude smaller than the velocity fluctuations. This
416: is just to confirm that pressure by itself is a much weaker signal than
417: the usual velocity correlations.\\ As one can see in Fig. 1
418: the scaling is quite poor, as one can expect in any DNS. As usual, in order
419: to extract quantitative statement about scaling exponents one has
420: to exploit the Extended Self Similarity (ESS) property enjoyed
421: by homogeneous and isotropic turbulent flows \cite{ess}. ESS consists
422: in looking for relative scaling of two different observable. Usually,
423: one takes two structure functions of two different orders, i.e.
424: in the case of longitudinal structure functions
425: $S_{q}^v(R) \sim \left[S_{q'}^v(R)\right]^{\zeta_v(q)/\zeta_v(q')}$. \\
426: Let us now define the same relative scaling for the two generalized
427: structure functions defined in (\ref{sf3}).
428: \begin{equation}
429: S_{q}^{vv^2}(R) \sim \left[S_{q'}^{vv^2}(R)\right]^{\frac{\zeta_{vv^2}(q)}
430: {\zeta_{vv^2}(q')}}, \;\; S_{q}^{vP}(R)
431: \sim \left[S_{q'}^{vP}(R)\right]^{\frac{\zeta_{vP}(q)}{\zeta_{vP}(q')}}
432: \label{ess_gen}
433: \end{equation}
434: In Figs. 2,3,4 we show the ESS plot for the two generalized
435: structure functions and for the usual longitudinal structure functions
436: respectively, with $q=1,\,q'=2$.
437: As one can see the all ESS plots showed a
438: scaling behaviour consistent with the usual homogeneous and isotropic
439: high Reynolds value which give for the relative exponents:
440: $\zeta_v(2)/\zeta_v(1) = 1.92 \pm 0.02$ \cite{ess}. %Previous \cite{exponents}
441: Similar agreements are found
442: for higher order moments (not showed). \\ The scaling
443: ansatz assumed for the generalized structure functions seems therefore
444: quite well satisfied. These findings support the fact that pressure
445: does not behave abnormally as far as its ``dimensional'' scaling
446: properties are concerned. Indeed, pressure-velocity correlations
447: behaves exactly like velocity-velocity correlations once pressure
448: is counted as a ``velocity squared''. Nevertheless,
449: The only realistic way to perform a quantitative statement about
450: scaling exponents is to study the logarithmic local slopes of
451: (\ref{ess_gen}). Only when logarithmic local slopes show a fairly
452: constant behaviour one can really speak about scaling. In Fig. 5
453: we show the logarithmic local slopes of (\ref{ess_gen}) for $q=1,\,q'=2$
454: together with the corresponding quantities measure for the
455: longitudinal velocity structure functions. In order to show that
456: the hypothesis that pressure-increments behave as a linear
457: velocity-increment, $\delta P \sim \delta v$, is definitely
458: ruled out by our data we also show in Fig. 5 the logarithmic local slope
459: of the ESS applied to pressure structure-functions for $q=1,q'=2$.
460: As one can see, while the three slopes measured on the flux structure
461: functions and on the longitudinal structure functions agree perfectly
462: with the high-Reynolds numbers measurements, the pure-pressure
463: structure functions is definitely much poorer. In Fig. 6 we plot
464: the same of Fig. 5 but for a different choice of moments, $q=2,\;q'=4$,
465: for both fluxes and longitudinal structure functions and with
466: $q=1,q'=2$ for the pure pressure structure functions. In this way
467: we are comparing quantities with exactly the same dimensional properties.
468: Again, while the flux-made structure functions, $S^{vP}_q(R)$, $S^{vv^2}_q(R)$
469: and the longitudinal structure
470: functions, $S^v_q(R)$ have the same local slope the pure-pressure
471: result obtained on the ESS of $F_q(R)$ shows a poorer and
472: different scaling. \\ A few comments are now in order. On one hand, we see
473: from Figs. 5 and 6 that the simple local-refined Kolmogorov hypothesis
474: derived in the previous section is correct, i.e. fluxes (\ref{sf3}) have
475: the same scaling properties of the usual longitudinal structure
476: function in homogeneous and isotropic turbulence, confirming
477: that these observable with the same physical dimensions and built
478: in terms of local field increments scale in the same way.
479: On the other hand, (see Fig. 6)
480: pure-pressure structure functions seem to violate the previous statement
481: despite of the fact that in this case there even exist an
482: exact result (\ref{s4p2}) supporting it.
483: Why pure-pressure correlations show this strong
484: deviation from the straightforward dimensional estimate? One possible
485: explanation is connected to the -possible- lack of isotropy in the statistics.
486: Any isotropically forced DNS is affected by possible
487: non-isotropic fluctuations induced by the
488: discretization of the numerical grid. In the following section
489: we have analyzed non-isotropic effects on both velocity and pressure
490: fluctuations.
491: \section{Anisotropic effects}
492: The exact relation which connect the second order
493: pressure structure function with a linear integral combination
494: of fourth order velocity structure function (\ref{s4p2}) is correct
495: only in the isotropic and homogeneous case. In order to test the
496: degree of isotropy of our simulation we have proceeded in a systematic
497: decomposition in terms of the irreducible representations of the SO(3)
498: symmetry group \cite{alp,prl1,itamar1}. The SO(3)
499: decomposition is particularly simple to apply to scalar observable,
500: i.e. observable whit all vectorial indexes contracted, like
501: pressure structure functions or longitudinal structure functions
502: on the kind analyzed in this work. In these case, the SO(3) decomposition
503: is nothing but a decomposition in spherical harmonics. For example, the
504: longitudinal structure functions, $S_q^v(\bR)
505: =< (v_i(\bx+\bR)-v_i(\bx))\hat{R}_i)^q> $,
506: can be decomposed as:
507: \begin{equation}
508: S_q^v(\bR) \equiv \sum_{jm} S_q^{jm}(R) Y_{jm}(\hat{\bR})
509: \label{so3}
510: \end{equation}
511: where now, we have explicitly considered the possibility that the undecomposed
512: structure functions depend on the whole vector $\bR$ and not only
513: on its magnitude as in the previous sections when isotropy was assumed.
514: The coefficient of the decomposition, $S_q^{jm}(R)$ depend only on the
515: magnitude of $R$ and on the two ``quantum'' numbers $j,m$ which labels
516: the properties under rotations of the $Y_{jm}$ eigenfunction.
517: Obviously, in the case of perfect isotropy we would have only one
518: projection alive, i.e. the projection on $Y_{00}$. The SO(3) decomposition
519: here summarized as been already used in some experimental and numerical
520: data analysis to properly disentangle the anisotropic effects
521: from the isotropic
522: ones \cite{alp,prl1,itamar1,bif_magnetico}. In our case, the relative
523: amplitudes of $S_q^{00}(|R|)$ with respect to the anisotropic fluctuations
524: $S_q^{jm}(|R|)$ with $j>0, -j \leq m \leq j$, gives a direct
525: quantitative estimate of the degree of anisotropic fluctuations
526: for any scale $|R|$. \\
527: In Fig. 7 we show the log-log plot of the undecomposed second order
528: longitudinal structure functions and of its projection on the fully-isotropic
529: eigenfunction $Y_{00}$. As it is possible to see, despite of the
530: isotropic forcing used in the simulation, the finite-size effects introduced
531: by our computational grid are quite important at large scales: the
532: projection, $S_2^{00}(|R|)$, shows a definitely better scaling than the
533: undecomposed structure function already as a functions of the
534: real separation $R$, i.e. without using ESS. This dramatic effect was already
535: observed in a similar application to the decomposition of velocity
536: fluctuations inside a channel \cite{prl1}. Fig. 7 definitely show that
537: the SO(3) decomposition can help in cleaning scaling properties
538: also in ``quasi-isotropic'' simulations. \\
539: On the other hand, the situation is quite different when the same decomposition
540: is applied to the pressure structure functions:
541: \begin{equation}
542: F_q^v(\bR) \equiv \sum_{jm} F_q^{jm}(R) Y_{jm}(\hat{\bR})
543: \label{Pso3}
544: \end{equation}
545: Let us notice that pressure is a quasi-isotropic observable
546: also for strong anisotropic velocity configuration. Indeed, being the
547: solutions of a Poisson problem, pressure is always an average of
548: velocity fluctuations on all spatial directions. This simple
549: considerations is perfectly verified on our numerical simulation. In Fig. 8
550: we show the undecomposed second-order pressure structure function
551: together with its projection on the fully isotropic harmonics, $F_q^{00}(R)$.
552: One can hardly detect any differences, suggesting tha anisotropic
553: fluctuations cannot be the responsible for the poor
554: scaling observed in the previous sections for the pure-pressure
555: structure functions. \\ The only possibility to reconcile the
556: exact result (\ref{s4p2}) with the poor scaling agreement
557: between pressure structure functions and velocity structure functions
558: is, in our opinion, to invoke strong Reynolds effects. Indeed, the resolution
559: of the Poisson problem certainly introduces strong non-local
560: effects on the statistics. Non-locality may also translate in strong
561: long-range effects in the Fourier space as far as the importance
562: of boundary conditions and forcing on the inertial range properties are
563: concerned. If this is correct, there are not reason to expect good
564: scaling properties for pure-pressure correlations, unless Reynolds number
565: is high enough to recover also in laboratory experiments
566: an almost-``infinite'' inertial range extension.
567: \section{Conclusions}
568: We have analyzed some pressure and pressure-velocity correlations
569: in a Direct Numerical Simulations at moderate Reynolds numbers.
570: We have derived on the basis of a simple analysis of energy transfer
571: properties across any sub-volume in the real space what we call
572: a ``local''-Refined Kolmogorov Hypothesis. We have identified a set
573: of pressure-velocity correlations which should have a good scaling
574: behaviour because connected via the local-RKH to the scaling of the
575: energy dissipation coarse grained on inertial range scales. \\We have
576: showed that our scaling hypothesis is well verified, while pure pressure
577: correlations feels strong Reynolds effects. According to our analysis
578: pressure scaling is perfectly determined by the dimensional
579: assumption that pressure behaves as a ``velocity squared'', unless the
580: finite-Reynolds effects are overwhelming. We do not find any sign
581: which could support the fact that pressure differences
582: behave as velocity differences as proposed in \cite{press2}.\\
583: We have also applied the SO(3) decompositions to the pressure
584: structure functions in order to show that poor scaling properties
585: showed by pure-pressure structure functions are not connected
586: to anisotropic fluctuations. \\
587: We acknowledge some help by I. Mazzitelli in the SO(3) analysis. LB
588: and FT have been partially supported by INFM (PRA-TURBO) and by the EU
589: contract FMRX CT98-0175,
590:
591:
592:
593: \begin{thebibliography}{99}
594:
595: \bibitem{frisch} U. Frisch, ``Turbulence: the Legacy of A.N. Kolmogorov
596: (Cambridge University Press, Cambridge, UK, 1995).
597:
598: \bibitem{iso1}M. Nelkin, ``Universality and scaling in fully developed
599: turbulence'', Advances in Physics, {\bf 43}, 143 (1994).
600:
601: \bibitem{iso2}K.R. Sreenivasan and R.A. Antonia,
602: ``The phenomenology of small-scale turbulence''
603: Annu. Rev. Fluid Mech. {\bf 29}
604: 435 (1997).
605:
606: \bibitem{prl1} I. Arad, L. Biferale, I. Mazzitelli and I. Procaccia,
607: `` Disentangling scaling properties in anisotropic and
608: inhomogeneous Turbulence''
609: Phys. Rev. Lett. {\bf 82} 5040 (1999).
610:
611: \bibitem{prl2} F. Toschi, G. Amati, S. Succi, R. Piva, R. Benzi,\\
612: ``Intermittency and structure functions in channel flow turbulence'',\\
613: Phys. Rev. Lett. {\bf 82} (1999) 5044.
614:
615: \bibitem{itamar1}I. Arad, B. Dhruva, S. Kurien, V.S.L'vov, I. Procaccia
616: and K.R. Sreenivasan, ``Extraction of anisotropic contributions in
617: turbulent flows'' Phys. Rev. Lett. {\bf 81} 5330 (1998).
618:
619: \bibitem{kraich} R.H. Kraichnan, ``Anomalous scaling of a
620: randomly advected passive scalar'' Phys. Rev. Lett. {\bf 72} 1016 (1994)
621:
622: \bibitem{gk} K. Gawedski and A. Kupiainen, ``Anomalous scaling of the passive scalar'' Phys. Rev. Lett. {\bf 75} 3834-3837 (1995).
623:
624: \bibitem{vergassola} M. Vergassola ``Anomalous scaling for passively advected magnetic fields'', Phys. Rev. E {\bf 53} R3021-R3024 (1996).
625:
626: \bibitem{dada} A. Lanotte and A. Mazzino ``Anisotropic nonperturbative zero modes for passively advected magnetic fields'' Phys. Rev. E {\bf 60} R3483-R3486 (1999).
627:
628:
629: \bibitem{bbw} R. Benzi, L. Biferale and A. Wirth
630: ``Analytic calculation of anomalous scaling in random
631: shell models of passive
632: scalars''
633: Phys. Rev. Lett. {\bf 78} 4926 (1997).
634:
635: \bibitem{bif_magnetico} I. Arad, L. Biferale and I. Procaccia,
636: ``Nonperturbative Spectrum of Anomalous Scaling Exponents
637: in the Anisotropic Sectors of Passively Advected Magnetic Fields ``
638: Phis. Rev. E (1999) submitted.
639:
640: \bibitem{sreene} B. Dhruva, Y. Tsuji and K.R. Sreenivasan
641: ``Transverse structure functions in high-Reynolds-number turbulence''
642: Phys. Rev. E {\bf 56} R4928 (1997).
643:
644: \bibitem{chen1} S. Chen, K.R. Sreenivasan, M. Nelkin, and N. Cao ``Refined
645: similarity hypothesis for transverse structure functions in fluid turbulence''
646: Phys. Rev. Lett. {\bf 79} 2253 (1997).
647:
648: \bibitem{alp} I. Arad, V.S. L'vov and I. Procaccia, ``Correlation functions in isotropic and anisotropic turbulence: the role of the symmetry group'' Phys. Rev. E {\bf 59} 6753 (1999).
649:
650: \bibitem{nc98} M. Nelkin and S. Chen, ``The scaling of pressure in isotropic
651: turbulence'', Phys. Fluids {\bf 10} 2119 1998.
652:
653: \bibitem{hb97} R.G. Hill and O. N. Boratav ``Pressure statistics for
654: locally isotropic turbulence'' Phys. Rev. E {\bf 56} R2363 1997.
655:
656: \bibitem{press2} N. Cao, S. Chen and G.D. Doolen ``Statistics and structures
657: of pressure in isotropic turbulence'' Phys. Fluids {\bf 11} 2235 (1999).
658:
659: \bibitem{yaglom} A.S. Monim and A.M. Yaglom ``Statistical Fluid Mechanics''
660: Vol. 2 M.I.T Press Cambridge (1975).
661:
662: \bibitem{hill}R. J. Hill and J. M. Wilczak ``Pressure structure functions
663: and spectra for locally isotropic turbulence'' J. Fluid Mech. {\bf 296}
664: 247-269, 1995.
665:
666: \bibitem{vedula} P. Vedula and P.K. Yeung ``Similarity scaling
667: of accelaration and pressure statistics in numerical
668: simulations of isotropic turbulence'' Phys. Fluids {\bf 11} 1208 (1999).
669:
670: \bibitem{duady}S. Douday, Y. Couder and M.E. Brachet ``Direct
671: observation of the intermittency of
672: intense vorticity filaments in turbulence''
673: Phys. Rev. Lett. {\bf 67} 983 (1991).
674:
675: \bibitem{forcing} M.R. Overholt and B.Pope,
676: ``A deterministic forcing scheme for direct numerical
677: simulation of turbulence'', Computers \& Fluids {\bf 27} 1 1998
678:
679: \bibitem{lesieur} M. Lesieur ``Turbulence in Fluids'',(third revised
680: edition, Kluwer Academic Publisher 1997).
681:
682: \bibitem{chen1024} S.Y. Chen and X. Shan ``High resolution
683: turbulent simulations using the connectin machine-2'' Compt. Phys. {\bf 6} 643
684: (1992).
685:
686: \bibitem{ess} R. Benzi, S. Ciliberto, R. Tripiccione, C. Baudet, F. Massaioli
687: and S. Succi ``Extended Self Similarity in turbulent flows'' Phys. Rev. E
688: {\bf 48} R29 (1993).
689:
690: %\bibitem{exponents} a ref where one can finds exponents measured at high
691: %reynolds numebrs, maybe one of the ref of roberto on ESS
692:
693: \end{thebibliography}
694:
695: \newpage
696: %%%%%%%%%%%%%%%%%%%%%%%%
697: \begin{figure}
698: %\narrowtext
699: \epsfxsize=16truecm
700: \epsfysize=10truecm
701: \epsfbox{fig1.eps}
702: \caption{Log-log plot of $F_1(R)$ $+$, and of $S_2(R)$, $\times$. Notice
703: that the pressure structure function is about one order of magnitude
704: smaller than the velocity structure function at large scale. }
705: \end{figure}
706: %%%%%%%%%%%%%%%%%%%%%%%%
707: %%%%%%%%%%%%%%%%%%%%%%%%
708: \begin{figure}
709: %\narrowtext
710: \epsfxsize=16truecm
711: \epsfysize=10truecm
712: \epsfbox{fig2.eps}
713: \caption{ESS log-log plot of $S^{vv^2}_2(R)$ vs $S^{vv^2}_1(R)$,
714: superimposed
715: is the straight line with the isotropic and homogeneous high-Reynolds slope
716: $\zeta_2^{v}/\zeta_1^v = 1.92$.}
717: \end{figure}
718: %%%%%%%%%%%%%%%%%%%%%%%%
719: %%%%%%%%%%%%%%%%%%%%%%%%
720: \begin{figure}
721: %\narrowtext
722: \epsfxsize=16truecm
723: \epsfysize=10truecm
724: \epsfbox{fig3.eps}
725: \caption{The same as in Fig. 2 but for $S^{vP}_2(R)$ vs $S^{vP}_1(R)$.}
726: \end{figure}
727: %%%%%%%%%%%%%%%%%%%%%%%%
728: %%%%%%%%%%%%%%%%%%%%%%%%
729: \begin{figure}
730: %\narrowtext
731: \epsfxsize=16truecm
732: \epsfysize=10truecm
733: \epsfbox{fig4.eps}
734: \caption{The same as in Fig. 2 but for the longitudinal
735: structure function, $S^v_2(R)$ vs $S^v_1(R)$.}
736: \end{figure}
737: %%%%%%%%%%%%%%%%%%%%%%%%
738:
739: %%%%%%%%%%%%%%%%%%%%%%%%
740: \begin{figure}
741: %\narrowtext
742: \epsfxsize=16truecm
743: \epsfysize=10truecm
744: \epsfbox{fig5.eps}
745: \caption{Logarithmic local slope of: ($\ast$)
746: $S^{vv^2}_q(R)$ vs $S^{vv^2}_{q'}(R)$; ($\Box$)
747: $S^{vP}_q(R)$ vs $S^{vP}_{q'}(R)$; ($\times$) $S^{v}_q(R)$ vs $S^{v}_{q'}(R)$,
748: for $q=2$ and $q'=1$.
749: Notice that all the above values are in perfect agreement with
750: the high-Reynolds number value $1.92$ (straight line), while the
751: logarithmic local slope for the pure-pressure structure
752: functions $F_2(R)$ vs $F_1(R)$, ($+$) is different.
753: The error bars are estimated by looking at the fluctuations over the first
754: half and the second half of the whole set of configurations.}
755: \end{figure}
756: %%%%%%%%%%%%%%%%%%%%%%%%
757: %%%%%%%%%%%%%%%%%%%%%%%%
758: \begin{figure}
759: \epsfxsize=16truecm
760: \epsfysize=10truecm
761: \epsfbox{fig6.eps}
762: \caption{Logarithmic local slopes of: ($\Box$)
763: $S^{vv^2}_q(R)$ vs $S^{vv^2}_{q'}(R)$; ($\ast$)
764: $S^{vP}_q(R)$ vs $S^{vP}_{q'}(R)$; ($\times$) $S^{v}_q(R)$ vs $S^{v}_{q'}(R)$,
765: for $q=4$ and $q'=2$.
766: Notice that, as in Fig. 5, the flux-based structure
767: functions have the same scaling behaviour
768: of the longitudinal structure function in agreement with the
769: high-Reynolds regime (straight line). Here we present a comparison with
770: the Pressure structure function with the same physical dimensions
771: of the flux-based structure functions, i.e. $F_2(R)$ vs $F_1(r)$, ($+$),
772: still the pure-pressure
773: structure function seems to have a different local slope.}
774: \end{figure}
775: %%%%%%%%%%%%%%%%%%%%%%%%
776:
777: %%%%%%%%%%%%%%%%%%%%%%%%
778: \begin{figure}
779: \epsfxsize=16truecm
780: \epsfysize=10truecm
781: \epsfbox{fig7.eps}
782: \caption{Undecomposed second order velocity structure functions
783: $S_2^v(\bR)$ measured on the plane $x-y$, ($\times$);
784: and the projection, $S_2^{00}(R)$ ($+$),
785: on the isotropic eigenfunction. The straight
786: line has the high-Reynolds slope $\zeta_2^v=0.7$. Notice that already
787: in the $R$-space, the SO(3)
788: decomposition improve the overall scaling behaviour. The two curves have been
789: shifted along the $y$ axis for the sake of presentation.}
790: \end{figure}
791: %%%%%%%%%%%%%%%%%%%%%%%%
792:
793:
794: %%%%%%%%%%%%%%%%%%%%%%%%
795: \begin{figure}
796: %\narrowtext
797: \epsfxsize=16truecm
798: \epsfysize=10truecm
799: \epsfbox{fig8.eps}
800: \caption{Undecomposed second order pressure structure functions
801: $F_2({\bf R})$ measured on the plane $x-y$, ($\times$);
802: and the projection, $F_2^{00}(R)$ ($+$),
803: on the isotropic eigenfunction.
804: Notice that, at difference from Fig. 7, here
805: the decomposed and undecomposed structure functions are almost identical,
806: indicating a high degree of isotropization in the pressure statistics.}
807: \end{figure}
808: %%%%%%%%%%%%%%%%%%%%%%%%
809:
810: \end{document}
811:
812: