nlin0011040/paper.tex
1: \documentstyle[epsfig,12pt]{article}
2: %________________________________________________________________________________________
3: \textwidth 16.cm
4: \textheight 21.cm
5: \topmargin 0.2in
6: \oddsidemargin -0.35cm
7: \baselineskip 0.7cm
8: %________________________________________________________________________________________
9: % Tex definitions
10: \newcommand {\ds}   {\displaystyle} 
11: \newcommand {\be}   {\begin{equation}}
12: \newcommand {\ba}   {\begin{array}}
13: \newcommand {\bea}  {\begin{eqnarray}}
14: \newcommand {\bfi}  {\begin{figure}}
15: \newcommand {\ee}   {\end{equation}}
16: \newcommand {\ea}   {\end{array}}
17: \newcommand {\eea}  {\end{eqnarray}}
18: \newcommand {\efi}  {\end{figure}}                  
19: \newcommand {\dV}  {\delta V}
20: \newcommand {\UNIV}   {Universit\`a }
21: %________________________________________________________________________________________
22: \begin{document}
23: %________________________________________________________________________________________
24: % Titolo
25: %
26: \title{Scaling laws and intermittency in homogeneous shear flow}
27: %________________________________________________________________________________________
28: % Autori
29: \author{
30: P\@. Gualtieri
31: \thanks{Dip. Mecc. Aeron., \UNIV di Roma ``La Sapienza'',
32:         via Eudossiana 18, 00184, Roma, Italy.},
33: %
34: C\@.M\@. Casciola$^*$
35: %
36: R\@. Benzi
37: \thanks{AIPA, via Solferino 15, 00185, Roma, on leave of absence from
38:         Dip. di Fisica, \UNIV di Roma ``Tor Vergata'', Italy.}
39: %
40: G\@. Amati
41: \thanks{CASPUR, p.le A. Moro 5, 00185 Roma, Italy.},
42: %
43: \&
44: R\@. Piva$^*$.}
45: \maketitle
46: %________________________________________________________________________________________
47: \section{Abstract}
48: %__________________________________________________________________________
49: In this paper we discuss the dynamical features of intermittent fluctuations
50: in homogeneous shear flow turbulence. In this flow the energy cascade is 
51: strongly modified by the production of turbulent kinetic energy related to 
52: the presence of vortical structures induced by the shear. By using direct 
53: numerical simulations, we show that the refined Kolmogorov
54: similarity is broken and a new form of similarity is observed, in agreement 
55: to previous results obtained in turbulent boundary layers. As a consequence,
56: the intermittency of velocity fluctuations increases with respect to 
57: homogeneous and isotropic turbulence. We find here that the statistical
58: properties of the energy dissipation are practically unchanged with respect
59: to homogeneous isotropic conditions, while the increased 
60: intermittency is entirely captured in terms of the new similarity law.
61: %__________________________________________________________________________
62: \newpage
63: %__________________________________________________________________________
64: \section{Introduction}
65: %________________________________________________________________________________________
66: 
67: The issue of anomalous scaling laws in turbulence
68: has been largely addressed for the idealized case of 
69: homogeneous and isotropic turbulence once it became clear that a purely dimensional
70: power law \cite{kolm_41} can not be consistent with the intermittent nature of energy 
71: dissipation \cite{Landau}. The Kolmogorov-Obukhov refined similarity (RKSH) 
72: \cite{kolm_62} provides the key theoretical point by relating the anomalous correction 
73: of the scaling exponents of velocity structure functions directly to the statistical 
74: properties of the dissipation field. The RKSH has been subject to close 
75: scrutiny by a number of independent investigations making use of both experiments and 
76: direct numerical simulations and it may be considered a well assessed physical law. 
77: One of the early difficulty for an accurate evaluation of scaling exponents, 
78: associated to the existence of too small scaling ranges in moderate Reynolds number 
79: turbulence, has been recently overcome with the use of the Extended Self-Similarity (ESS) 
80: \cite{benzi_1}. 
81: This approach has allowed accurate estimates of the exponents 
82: from moderate Reynolds number flows  \cite{frisch} and the evaluation of the RKSH, in 
83: its extended form, from DNS data \cite{benzi_2}.
84: 
85: A more recent research interest in turbulence has been focused on the assessment of RKSH
86: in non-isotropic and non-homogeneous conditions, namely in flows characterized by strong 
87: mean shear and anisotropy. The main focus is to understand to what extent 
88: RKSH may be considered still able to capture, in such conditions, the essential features 
89: of turbulence dynamics.
90: 
91: In this context recent results for a turbulent channel flow \cite{PF}, have pointed out 
92: how the classical RKSH holds in the higher part of the logarithmic region. 
93: This result should not be surprising since the RKSH is established as a consequence of 
94: the  balance between energy transfer due to non linear interactions and dissipation. 
95: In the log-layer, where the mean shear is weak, the dynamics of the flow is 
96: characterized by the inertial energy transfer. In fact, when approaching the wall, the 
97: shear becomes larger and larger and a clear failure of the RKSH is observed \cite{PF}. 
98: This result gives a partial answer to the point raised above since it gives evidence 
99: that, in the near wall region, the homogeneous isotropic behavior is far from being 
100: recovered. 
101: 
102: At this point a question naturally arises about the existence of an alternative scaling 
103: law in such conditions. On the basis of physical considerations on the respective role 
104: of energy production and energy transfer a different Kolmogorov-like scaling law has 
105: been proposed in \cite{PF} and preliminarily checked against DNS data from the buffer 
106: region of a turbulent channel flow. The validity of this similarity law has been 
107: confirmed by a successive experimental work in a turbulent boundary layer.
108: However, certain aspects of the problem are still not completely clear.
109: Actually, the new similarity law has been conceived as a direct consequence of the 
110: presence of a strong shear, but it has been verified only in wall turbulence where, in 
111: principle, some other features such as the continuous variation of the local shear, the 
112: non uniform momentum flux across adjacent layers and the suppression of wall normal 
113: velocity fluctuations may play a quite significant role.
114: 
115: To isolate the effect of the shear we reconsider here the problem in the context of a 
116: DNS of a homogeneous shear flow in a confined box. In fact, in the rationale of dimensional 
117: analysis, this flow presents the minimum level of complexity though maintaining the 
118: essential features we want to address: thus, it is ideal to study the effect of 
119: a pure shear avoiding other concurrent effects. On the other hand, it allows to exploit 
120: homogeneity in all spatial directions so that a more complete and accurate statistical 
121: analysis can be performed, at a level of detail that could never be achieved in the very 
122: limited buffer region of a wall bounded flow.
123: 
124: As a counterpart, the homogeneous shear flow presents certain drawbacks related to the 
125: artificial nature of its confinement. To understand to what extent the present flow may 
126: serve as a prototype for shear-dominated flows, we do analyze in detail 
127: the dynamics of the coherent vortical structures observed in the numerical simulation. 
128: This is instrumental to qualify the flow, i.e. its dependence on the aspect ratio of the 
129: computational box, in view of the use we do of the flow itself for the evaluation of scaling 
130: laws and their assessment.
131: 
132: Homogeneous shear flows have been investigated in the literature in many 
133: theoretical, numerical and experimental works. Rogers and Moin \cite{rogers} and Lee 
134: {\em{et. al.}} \cite{lee} studied the topology and mutual interactions of vorticity
135: showing that many dynamical features are quite similar to 
136: those observed in the wall region of a turbulent boundary layer. A more detailed study 
137: on the same subject has been performed by Kida and Tanaka \cite{kida} who discussed   
138: the regeneration cycle of the streamwise vortices in a homogeneous shear flow.   
139:    
140: Experimentally the papers by Rose \cite{rose}, Champagne Harris and Corrsin 
141: \cite{champagne}, Tavoularis and Corrsin \cite{tavoularis_1} \cite{tavoularis_2},    
142: have investigated homogenous shear flows obtained in a wind tunnel, focusing on the local
143: isotropy of small scale fluctuations. Following the Kolmogorov theory,   
144: an important question to investigate, which becomes extremely well posed in the case of 
145: the homogeneous shear flow, concerns the recovery of small scales  local isotropy
146: in turbulent flows characterized by large scale non isotropic forcing. 
147: The same kind of question has been addressed by Saddoughi and Veeravalli 
148: \cite{saddoughi} who investigated local isotropy for the turbulent flow in a logarithmic 
149: boundary layer.    
150:    
151: The issue of local isotropy has been also addressed more recently by Pumir and Shraiman 
152: \cite{pumir_2} and by Pumir \cite{pumir_3} who performed an extensive analysis of    
153: numerical simulations of statistically stationary homogeneous shear flows. Interestingly 
154: enough, most of the statistical properties found by Pumir \cite{pumir_3} are quite close 
155: to the experimental findings of Garg and Warhaft \cite{garg} and Shen and Warhaft 
156: \cite{shen}. In the latter experiment, for the first time, an active grid was used in 
157: order to produce a homogeneous shear flow in a wind tunnel whose integral scale in the 
158: streamwise direction is rather constant.   
159:    
160: Taking full advantage of most of these results, and in order to be able to reconsider
161: the issue of local isotropy from a different perspective, as we plan to do in the near 
162: future, our interest here is mainly focused on the relationship between intermittency and the
163: anisotropy induced by the mean velocity gradient at large scale.  In the context of
164: boundary layer we have shown how the increase of intermittency may be explained in 
165: terms of the phenomenology described by the new form of scaling law. In particular,  
166: as suggested in \cite{PF}, a new  length scale $L_s  = \sqrt{\bar{\epsilon} / S^3}$ enters 
167: in the  description of the statistical properties of turbulence, where $\bar{\epsilon}$ is 
168: the mean rate of energy dissipation and $S$ the mean large scale shear. 
169: For scales smaller than $L_s$ the statistical 
170: properties of the turbulent fluctuations are similar to those observed in homogenous and 
171: isotropic flows while for scales larger than $L_s$ the refined  Kolmogorov similarity 
172: (RKSH) is broken  and intermittency increases. The phenomenological theory developed for 
173: the turbulent  boundary layers relies upon the mean shear strength $S$ as the only 
174: parameter which fixes the scale $L_s$ where a new form of RKSH should be observed.    
175: For this reason, the homogenous shear flow becomes a natural test case to understand whether 
176: or not the phenomenological theory devised for the  boundary layer is sufficiently 
177: general to be applied to any turbulent flow.
178: As we shall discuss in this paper, our numerical simulations of homogenous shear flows 
179: support rather well the new phenomenology, leading to possible important implications in 
180: our understanding of shear turbulence.   
181:    
182: The paper is organized in the following way. In section 3 we briefly describe how the 
183: numerical simulations are performed. 
184: In section 4 we discuss the regeneration cycle of the vortical 
185: structures observed in the flow while in section 5 we address the intermittency cycle 
186: which characterizes the dynamical behavior of the system. In section  6, after a short 
187: review of the phenomenological theory proposed for the turbulent boundary layer, we describe 
188: the intermittency properties of turbulent fluctuations. In section 7, we investigate the 
189: anomalous scaling of the energy  dissipation field, and, finally, in section 8 we draw our 
190: main conclusions.     
191: %________________________________________________________________________________________
192: \section{Homogeneous shear flow}
193: %________________________________________________________________________________________
194: We consider here a turbulent flow in a confined box with an imposed mean shear $S$ 
195: free from boundaries.
196: The full Navier-Stokes equations are solved (DNS), after decomposing the velocity field 
197: into mean value and fluctuation
198: %________________________________________________________________________________________
199: \be
200: \label{velo_dec}
201:  \vec v(\vec x,t)= U(y) \vec e_1 + \vec u(\vec x,t) \, ,
202: \ee
203: %________________________________________________________________________________________
204: where 
205: $\vec x \in V =[-\lambda_x,\lambda_x] \times [-\lambda_y,\lambda_y] 
206: \times [-\lambda_z,\lambda_z]$, $V$ identifies the computational box, $U(y)=S(y+\lambda_y)$ 
207: is the mean flow and $\vec e_1$ is the unit vector in the streamwise direction $x$. 
208: The mean gradient $S$ is in the normal direction $y$ while $z$ denotes the spanwise coordinate. 
209: The Navier Stokes equations are written in terms of velocity fluctuations 
210: %________________________________________________________________________________________
211: \be
212: \label{shear_eq}
213: \left\{
214: \ba{l}
215: \ds  \vec\nabla\cdot\vec u=0 \\ \\
216: \ds \frac{\partial\vec u}{\partial t}=(\vec u {\bf{\times}} \vec \zeta)
217:              - \vec \nabla (p+\frac{u^{2}}{2})
218:              +\nu \nabla^{2}\vec u -Sv \vec e_{1}
219:              -U(y) \frac{\partial \vec u}{\partial x} \, ,
220: \ea
221: \right.
222: \ee
223: %________________________________________________________________________________________
224: where $\vec \zeta$ is the vorticity, $v$ the normal velocity, $p$ the pressure and $\nu$ the 
225: kinematic viscosity.
226: To achieve efficiency and accuracy a Fourier spectral method is advisable to solve the 
227: initial value problem for eq. (\ref{shear_eq}).
228: However, due to the mean flow, the intrinsically  non-periodic term 
229: $U(y) \partial \vec u / \partial x$ appears in the equations. 
230: Luckily, by using Rogallo's technique \cite{rogallo}, this difficulty is removed by the 
231: transformation
232: %________________________________________________________________________________________
233: \be
234: \label{var_transf}
235: \left\{
236: \ba{l}
237:     \ds  \xi=x-U(y)t \\
238:     \ds  \eta=y     \\
239:     \ds  \zeta=z     \\
240:     \ds  \tau=t.
241: \ea
242: \right.
243: \ee
244: %________________________________________________________________________________________
245: In the new variables periodic boundary conditions can be enforced in all spatial directions 
246: allowing efficient pseudo-spectral methods to be used for spatial discretization. 
247: The time integration is performed using a third order low storage Runge-Kutta method 
248: \cite{solver} and the nonlinear terms are fully de-aliased by zero padding. 
249: Since the image of the computational box in physical space gets distorted,  
250: eq. (\ref{var_transf}), a re-meshing procedure is periodically applied to allow long time 
251: integrations. 
252: Using periodicity in the $\xi$ direction, the computational domain is transformed back
253: into a non skewed domain whenever the plane $y=\lambda_y$ has moved by $2\lambda_x$
254: in the streamwise direction,  i.e. every $\Delta t_r =\lambda_x / S \lambda_y$.
255: This procedure may introduce aliasing errors since the dynamics of wave vectors in 
256: physical space is time dependent \cite{townsend}, as seen from the relation between 
257: the Fourier transforms in the computational and in the physical space, 
258: %________________________________________________________________________________________
259: \be
260: \label{coef_rel}
261: \hat{u}_i(k_x,k_y,k_z,t)=\hat{\bar{u}}_i(k_{\xi},k_{\eta}-S \tau k_{\xi},k_{\zeta},\tau).
262: \ee
263: %________________________________________________________________________________________
264: To avoid aliasing errors, the spectral components with wave numbers outside the 
265: interval $[-k_{max},k_{max}]$ are set to zero at re-meshing. 
266: This filtering introduces a characteristic wave number defined as
267: %________________________________________________________________________________________
268: \be
269: \label{k_remesh}
270:   k_{r}=\frac{k_{max}}{\sqrt{1+A^2}}
271: \ee
272: %________________________________________________________________________________________
273: where $k_{max}= \pi N_y / 2 \lambda_y$ and $A=\lambda_x / \lambda_y$ is the aspect 
274: ratio of the computational domain. In order to asses the influence of the filtering,
275: the aspect ratio of the box and the resolution of the grid were varied. 
276: In all cases $k_r$ was larger than the dissipation wavenumber and no appreciable alteration
277: of the dynamics was observed.
278: 
279: A crucial point for our successive analysis is the ability of the system to reach statistical 
280: stationarity. 
281: Under this respect, our computations entirely confirm the results of Pumir and Shraiman 
282: \cite{pumir_2} and Pumir \cite{pumir_3}.
283: Actually the evolution of the turbulent kinetic energy is described by the equation
284: %________________________________________________________________________________________
285: \be
286: \label{energy_eq}
287: \frac{\partial}{\partial t} [\frac{u^2}{2}] + S[u v] = -\nu [ \zeta^2]
288: \ee
289: %________________________________________________________________________________________
290: where the square brackets denote spatial average. 
291: This equation suggests a possible balance between the production $S[u v]$ and the 
292: dissipation $\nu [\zeta^2]$.
293: This balance is not achieved for the instantaneous fields, as shown in figure 1
294: which reports the history of turbulent kinetic energy in our longest 
295: calculation. 
296: Large fluctuations in energy ($42 \%$) and enstrophy ($50 \%$) are apparent in the 
297: pseudo-cyclic behavior shown in the figure. The fluctuation level in this system
298: is quite huge, expecially when compared to that of forced homogeneous isotropic
299: turbulence, which is limited to only a few percent of the rms value ($5-6 \%$) for both kinetic
300: energy and enstrophy.
301: In fact, the pseudo-cyclic behavior corresponds to statistically stationary conditions, as 
302: follows from time averages performed over time periods much larger than the typical length 
303: of the cycle.
304: 
305: Clearly, DNS enables to achieve a good level of homogeneity, as shown in figure 2.
306: In particular both the turbulent kinetic energy and the Reynolds stresses are substantially
307: constant in the $y$ direction.  
308: 
309: In order to check the dynamics of the velocity gradients, the probability density
310: function of $\partial u / \partial x$ and of $\partial u / \partial y$ have been computed,
311: figure 3 and 4. They are often used to asses the small 
312: scale dynamics of the flow and to characterize its degree of anisotropy.
313: The computed values of skewness and flatness are in good agreement with the experimental 
314: results at $Re_{\lambda} \sim 100$ of Shen and Warhaft \cite{shen} and of Ferchichi and
315: Tavoularis \cite{ferchichi} at $Re_{\lambda}=140$.
316: %________________________________________________________________________________________
317: \section{Regeneration cycle of vortical structures}
318: %________________________________________________________________________________________
319: The energy fluctuations discussed in the previous section correspond to a regeneration 
320: cycle of the vortical structures.  Most of the previous numerical simulations have described
321: the main features of coherent structures and the vorticity statistics for the early
322: stages of development of the flow.
323: In particular, Rogers $\&$ Moin \cite{rogers} showed typical hairpin vortex structures 
324: remarkably similar to those observed in wall bounded flows.
325: Lee {\em{et. all.}} \cite{lee} discussed streamwise vortices and high and low speed streaks 
326: commenting on the similarities with the buffer region of wall bounded flows. 
327: In a more recent work Kida $\&$ Tanaka \cite{kida} proposed a regeneration mechanism 
328: for the  streamwise vortices and pointed out the role of vortex sheet instability in the
329: formation of new vortices.
330: The present analysis deals with the same issues, but it is focused on the statistical steady 
331: state regime of the flow. 
332: 
333: The phenomenology is better understood by considering a few snapshots along the history
334: of the turbulent kinetic energy and of the Reynolds stresses. 
335: In figure 5 the large energy bursts already noticed in the previous 
336: section are clearly correlated to large negative values of the Reynolds stresses.
337: The bursts are induced by large rates of energy injection from the mean flow associated with
338: the presence of streamwise vortices.
339: Figure 6, corresponding to an instant at the beginning of a stage of
340: energy growth, actually shows a large population of quasi-streamwise vortices, here
341: visualized through the discriminant of the velocity gradient \cite{chong}.
342: During the successive phase of energy growth the streamwise vortices, by interacting with 
343: the mean velocity field, give rise to instantaneous profiles characterized by the 
344: typical ramp and cliff pattern. The pdf of $\partial u / \partial y$ is a suitable tool
345: to characterize statistically the ramp and cliff structures, figure 4. 
346: The skewness of $0.82$ gives reason of intense positive events of $\partial u / \partial y$ 
347: much more probable than negative ones.
348: As for the passive scalar, the streamwise velocity recovers the mean gradient through ramps 
349: followed by cliffs \cite{pumir_1}. 
350: The ramps correspond, on average, to streamwise velocity increasing with $y$, hence
351: a resulting negative spanwise vorticity is expected, on average.
352: The pdf of the spanwise vorticity, figure 7, confirms this analysis.
353: 
354: Ramp and cliffs are associated with thin regions of (negative) spanwise vorticity, i.e.
355: vortex sheets, as shown in figure 8.
356: The sheets become unstable and eventually roll up into spanwise vortices through the classical  
357: Kelvin-Helmholtz mechanism, see figure 9 where spanwise vortices are 
358: systematically located in the regions where the roll-up process is occurring.
359: Vortex sheets and spanwise vortices are clearly responsible for the observed energy bursts
360: by producing large negative Reynolds stress events.
361: 
362: Immediately after each burst, the non-linear interactions are enhanced and the original 
363: vortex structures lose their order resulting into a randomized  vorticity field.
364: This phase, shown in figure 10, reduces the level of anisotropy of the flow. 
365: However, soon afterwards, the shear term of the vorticity equation enforces a mean orientation 
366: to the structures, see figure 11, corresponding to a minimum for the energy, 
367: just before the occurrence of the successive burst. 
368: The vortical structures are now getting more and more aligned with the streamwise direction 
369: and a new cycle starts.
370: 
371: The regeneration cycle just analyzed here for the statistical steady state phase 
372: is very similar to that described by Kida $\&$ Tanaka \cite{kida} for the early stages of 
373: evolution of the flow from isotropic initial conditions, suggesting that the operating
374: mechanisms are substantially identical.
375: %________________________________________________________________________________________
376: \section{Intermittency cycle}
377: %________________________________________________________________________________________
378: The increase of the turbulent kinetic energy due to the streamwise vortices can be 
379: understood, basically, in terms of the linear lift-up mechanism and the related 
380: transient growth.
381: A non linear mechanism is required instead to explain the saturation and the break-down 
382: of the ordered system of vortices. 
383: In spectral space, as already discussed by Pumir \cite{pumir_3}, the Fourier mode 
384: $(0,0,\pm 1)$ gives the leading contribution to the energy growth. 
385: From the linearized equation for this  mode,  
386: %________________________________________________________________________________________
387: \be
388: \label{linear_ns}
389: \left\{
390: \ba{l}
391: \ds  \frac{d \hat{u}}{dt}=-S \hat{v} - \nu \hat{u}\\ \\
392: \ds \frac{d \hat{v}}{dt}= - \nu \hat{v} ,
393: \ea
394: \right. 
395: \ee
396: %________________________________________________________________________________________
397: a growing amplitude is expected whenever 
398: $S[Re(\hat{u}) Re(\hat{v}) + Im(\hat{u}) Im(\hat{v}) ] < 0$. 
399: As soon as the energy grows appreciably, the energy drained by the interactions with the 
400: other modes in favor of the smaller scales quite rapidly saturates its amplitude.
401: Hence, the characteristic frequency of the intermittency cycle is determined by the 
402: dynamical balance between growth of the basic mode and energy transfer. 
403: 
404: In principle, if the mode $(0,0,\pm 1)$ were isolated from the others during its growth, the
405: energy due to the interaction with the mean shear would be almost entirely 
406: found in this mode. Then, the characteristic time of the system would 
407: essentially coincide with the eddy turn over time of the basic mode, hence to the spanwise 
408: size of the box $L_z$. 
409: 
410: In fact, the leading mode evolves in presence of many others. 
411: In this case the saturation time, i.e. the time of the effective activation of the nonlinear 
412: energy transfer, is related to the separation, in wavenumber space, between the leading 
413: mode and the small scales modes. 
414: The activation time becomes shorter and shorter as the distribution of energy among all 
415: Fourier modes approaches a continuous distribution. For the statistical steady state regime, 
416: in particular, the non linear energy transfer is crucial from the very beginning of the growth.
417: Actually the energy spectrum in correspondence of the large bursting phenomena, 
418: figure 12, shows initialy a growth which, though dominated by the
419: mode $(0,0,\pm 1)$, is spread all over the modes in the first decade. The evolution of the
420: spectra manifest a strong energy transfer which though occurring during the entire cycle, is 
421: clearly prevailing in the phase of energy decrease.
422: Hence the bursting frequency can not be estimated from the isolated dynamics of the 
423: sole $(0,0,\pm 1)$ mode, and, as a consequence, there is no {\sl a priori} reason for a strong 
424: dependence of the bursting period on the size of the box.
425: 
426: In view of a quantitative analysis of this issue, let $E_U$ and $E_u$ denote the kinetic 
427: energy of the basic flow and of the fluctuations,  respectively. 
428: Clearly, $E_u$ is a function of time.
429: Two quite different cases can be distinguished, namely
430: %________________________________________________________________________________________
431: \begin{itemize}
432: \item[$a)$] $E_u \geq E_U$ at $t=0$
433: \item[$b)$] $E_u \ll  E_U$ at $t=0$ \, .
434: \end{itemize}                                                
435: %________________________________________________________________________________________
436: Case b) corresponds to a problem of transition to turbulence from small, though finite, 
437: disturbances and will not be considered here in further detail. We are presently 
438: interested in case a), which is more appropriate to describe statistical stationarity.  
439: Here $t=0$ should be understood as an arbitrary instant of time along the cyclic 
440: evolution of the system.
441: 
442: Given the correlation function $C(\tau) = <E_u(t + \tau) E_u(t)>$, let us consider the 
443: correlation time, $\tau_C$, defined, e.g., as the smallest $\tau$ such that 
444: $C(\tau_C)=1/2 C(0)$.
445: Following the discussion of section 4, $\tau_C$ corresponds, roughly, to the characteristic 
446: time of production of large scale velocity$/$vorticity instability which is the primary 
447: forcing mechanism of the homogeneous shear flow.
448: If $\tau_C$ is almost independent or, at least,  weakly dependent on $L_z$, the dynamical 
449: behavior of the system may be considered as substantially unaffected by the finite size of 
450: the box. 
451: A rigorous analysis of the correlation time dependence on the spanwise size $L_z$ would 
452: require a large number of highly resolved numerical simulations far beyond the present 
453: computer capabilities. In figure 13, we show a preliminary evidence that 
454: the behavior of the kinetic energy is not too sensitive to $L_z$, by comparing the 
455: simulations performed with $L_z= 2 \pi$ and $L_z= 4 \pi$.
456: The characteristic time of the generation mechanism of large scale vorticity seems not to 
457: depend on $L_z$. These results suggest that the finite size of the box does not affect 
458: in a significant way the dynamics and the statistical properties of the homogeneous 
459: shear flow, despite the importance of the basic mode, whose length scale coincides with 
460: that of the box, in the dynamics of the flow.
461: %________________________________________________________________________________________
462: \section{Intermittency and scaling}
463: %________________________________________________________________________________________
464: In the previous sections we have analyzed the influence of the shear on the dynamics of 
465: the vortical structures. We are now ready to discuss its effect on statistical features of 
466: turbulent fluctuations, such as scaling laws and intermittency.
467: 
468: Concerning the velocity field, a quantitative description can be given in terms of the
469: possible scaling behavior of the structure functions, i.e.  the moments of longitudinal 
470: velocity increments
471: %________________________________________________________________________________________
472: \be
473: \label{strut_fun}
474: < \dV^p> = < \{ [ \vec u(\vec x + \vec r,t)- \vec u(\vec x,t)] \cdot 
475: \frac{\vec r}{r} \}^p >
476: \ee
477: %________________________________________________________________________________________
478: where angular brackets denote ensemble averaging. 
479: For homogeneous and isotropic conditions, the dimensional prediction of Kolmogorov 
480: theory (K41) provides a scaling law in terms of separation \cite{kolm_41}
481: %________________________________________________________________________________________
482: \begin{eqnarray}
483: \label{K41}
484: < \dV^p > & \propto & \bar{\epsilon}^{p/3} \, r^{p/3} \,
485: \end{eqnarray}
486: %________________________________________________________________________________________
487: where $\bar \epsilon$ denotes the mean rate of energy dissipation per unit mass.
488: For $p=3$, 
489: equation (\ref{K41}) is consistent with the exact result usually referred to as 
490: the {\em{four fifth law}},
491: %________________________________________________________________________________________
492: \begin{eqnarray}
493: \label{four_fifth}
494: < \dV^3 > & = & -\frac{4}{5} \bar{\epsilon} \, r \,.
495: \end{eqnarray}
496: %________________________________________________________________________________________
497: This  equation has been widely confirmed by a number of experimental investigations 
498: \cite{benzi_1} which, at the same time, showed that dimensional scaling does not hold for 
499: moments different from the third. In fact, a more complex dependence of the scaling exponents 
500: is found,
501: %________________________________________________________________________________________
502: \begin{eqnarray}
503: \label{EXP}
504: < \dV^p > & \propto &  r^{\zeta_p},
505: \end{eqnarray}
506: %________________________________________________________________________________________
507: with $\zeta_p$ a nonlinear convex function of $p$. This anomalous scaling related to 
508: {\em{intermittency}}, has been actively investigated over the last twenty years. 
509: 
510: %________________________________________________________________________________________
511: \subsection{Dimensionless parameters}
512: %________________________________________________________________________________________
513: To understand how intermittent fluctuations are affected by the shear, it is worthwhile 
514: to consider the typical length scales involved in the process and the relevance
515: of the shear production term over the nonlinear inertial interactions. 
516: %________________________________________________________________________________________
517: In presence of shear, velocity fluctuations arise for two main reasons, advection of
518: streamwise momentum across the mean gradient and non-linear mixing. Let us denote by 
519: $\delta U = S r$ the velocity difference at scale $r$ due to the mean flow and by 
520: $\delta u$ the fluctuation at the same scale. A rough estimate for the latter is 
521: $\delta u \propto \bar{\epsilon}^{1/3} r^{1/3}$.
522: The length scale for which turbulent fluctuations are of the same order of magnitude as
523: the velocity increments induced by the shear, uniquely identifies the {\sl shear scale} 
524: $L_s$ 
525: %________________________________________________________________________________________
526: \be
527: \label{shear_scale}
528: L_s=\sqrt{\frac{\bar{\epsilon}}{S^3}} \, ,
529: \ee
530: %________________________________________________________________________________________
531: that separates two different sub-ranges within the inertial range.
532: For $ L_0 \gg r \gg L_s$ (with $L_0$ the integral scale),  $\delta U \gg \delta u$ and
533: the dynamics and statistics of turbulence is expected to be dominated by the momentum flux 
534: due to the Reynolds stresses, i.e. by the production of kinetic energy.
535: For $\eta \ll r \ll L_s$, $\delta u \gg \delta U$ and the fluctuations induced by the mean 
536: shear are negligible with respect to those typically produced by the nonlinear term. 
537: In this case the dynamics of the flow is characterized by the energy transfer due to the same 
538: inertial interactions which are typical of isotropic turbulence.
539: 
540: In order to quantify the predominance of either mechanism over the other, we consider the
541:  dimensionless parameter, defined as the ratio of an inertial and a shear time scale,
542: %________________________________________________________________________________________
543: \be
544: \label{shear_par1}
545: S^*=\frac{S q^2}{\bar{\epsilon}}
546: \ee
547: %________________________________________________________________________________________
548: with $q^2=<u_i u_i>$. $S^*$ can be interpreted as the ratio of two length scales  
549: %________________________________________________________________________________________
550: \be
551: \label{shear_par2}
552: S^*=\left(\frac{l_d}{L_s}\right)^{2/3} \,
553: \ee
554: %________________________________________________________________________________________
555: where $l_d=q^3 / \bar{\epsilon}$. It follows that for large values of $S^*$ the dynamics 
556: of most of the inertial scales is dominated by the mean shear. 
557: 
558: The other dimensionless parameter needed for a complete description of a homogeneous
559: turbulent shear flow can be given as the ratio of the dissipative and the shear time scale,
560: %________________________________________________________________________________________
561: \be
562: \label{shear_par3}
563: S^*_c=S (\nu / \bar{\epsilon})^{1/2} \, ,
564: \ee
565: %________________________________________________________________________________________
566: or, equivalently, as the ratio of the Kolmogorov length and the shear scale
567: %________________________________________________________________________________________
568: \be
569: \label{shear_par4}
570: S^*_c=\left(\frac{\eta}{L_s}\right)^{2/3}.
571: \ee
572: %________________________________________________________________________________________
573: Since $S^*_c$ measures the separation between the shear scale and the dissipative range, 
574: a sub-range of isotropic behavior may be recovered within the inertial range when it is small.
575: 
576: Typical values of $S^*$ and $S^*_c$ are shown in figure 14, which is a
577: compilation of several independent numerical and experimental simulations.
578: The variety of behaviors observed in these simulations can easily be interpreted in terms of 
579: the location of the shear scale $L_s$, quite different from one case to the other. 
580: 
581: %________________________________________________________________________________________
582: \subsection{Intermittency}
583: %________________________________________________________________________________________
584: Let us now turn to the issue of intermittency. A quantitative measure of intermittency 
585: is provided by the flatness factor of the random variable $\dV(r)$. 
586: Figures 15 and 16 show the probability distribution function of 
587: $\dV(r)$ for 
588: two different separations both in the shear flow and  in homogeneous isotropic 
589: turbulence. 
590: For small separations in both cases the pdf is clearly non Gaussian and the flatness factor, 
591: shown in figure 17, grows as separation decreases confirming that 
592: the dimensional scaling of the structure functions is clearly violated. 
593: However, important differences are apparent.
594: The tails for the shear flow are quite higher and the pdf is more skewed towards negative 
595: values. 
596: For the larger separation, instead, the two pdfs are much more similar, though still not 
597: Gaussian. 
598: The Gaussian behavior is recovered only for $r$ comparable with the integral scale. 
599: Globally, we find a correspondence with what was found in the turbulent channel flow 
600: \cite{PF}, where at small separation the pdfs of $\dV(r)$ in the buffer and in the log-region 
601: differs substantially, with the negative tail much higher than the positive and an increased
602: flatness factor in the buffer.
603: 
604: %________________________________________________________________________________________
605: \subsection{Similarity laws}
606: %________________________________________________________________________________________
607: The issue of intermittency can be quantified by looking directly at the scaling properties of 
608: the structure functions. As well known, starting from Landau objection \cite{frisch} 
609: a revised form of similarity law has been proposed by Kolmogorov and Obhukov 
610: \cite{kolm_62} (K62): assuming that the dissipation field is a random variable, 
611: equation (\ref{K41}) is corrected to
612: %________________________________________________________________________________________
613: \begin{eqnarray}
614: \label{K62}
615: < \dV^p > & \propto & <\epsilon_r^{p/3}> \, r^{p/3} \,,
616: \end{eqnarray}
617: %________________________________________________________________________________________
618: where $<\epsilon_r^q>$ denotes the q-th moment of the local energy dissipation field 
619: $\epsilon_{loc}$ averaged over a volume of characteristic dimension $r$. 
620: Given for the dissipation field a scaling law in terms of separation,
621: %________________________________________________________________________________________
622: \begin{eqnarray}
623: \label{eps_scaling}
624: < \epsilon_r^q > & \propto & r^{\tau(q)} \,,
625: \end{eqnarray}
626: %________________________________________________________________________________________
627: from equation (\ref{K62}) the scaling exponents of the structure 
628: functions are expressed as
629: %________________________________________________________________________________________
630: \be
631: \label{zeta_p}
632: \zeta(p) = \tau(p/3) + p/3
633: \ee
634: %________________________________________________________________________________________
635: with $\tau(p/3)$ a nonlinear function of $p$ and $\tau(1)=0$.
636: Equation (\ref{zeta_p}) is consistent with the intermittent nature of the random variable 
637: $\dV(r)$, i.e. its increasing flatness with decreasing separation,
638: since the flatness factor, within the inertial range, turns out to be an unbounded function, 
639: diverging for small separations as $r^{\zeta_4-2\zeta_2}$.  
640: The intermittency is related to bursty signals characterized by localized
641: and intense fluctuations and to the presence of coherent structures and associated
642: regions of intense gradients which give rise to a highly non uniform dissipation field.  
643: 
644: Scaling laws in terms of separation  are found only for sufficiently large values 
645: of the Reynolds number i.e., so far, only in experimental facilities. 
646: 
647: An extension of the scaling range can be achieved by using the Extended Self Similarity (ESS) 
648: as proposed by Benzi et. al. \cite{benzi_1}, who introduced 
649: a generalized form of the scaling law (\ref{EXP}),
650: %________________________________________________________________________________________
651: \begin{eqnarray}
652: \label{ESS}
653: < \dV^p > & \propto &   < \dV^3 >^{\zeta_p/\zeta_3} \,.
654: \end{eqnarray}
655: %________________________________________________________________________________________
656: Using experimental data, the scaling law (\ref{ESS}) has been shown to hold even at low and 
657: moderate Reynolds number, i.e. in cases, such as those amenable to DNS,
658: where the scaling (\ref{EXP}) is not directly detectable. 
659: Most interestingly, the ratios $\zeta_p/\zeta_3$ have systematically been found independent
660: of the Reynolds number and consistent with the measurements in the high Reynolds number regime. 
661: The ESS allows a generalization of the refined Kolmogorov similarity hypothesis 
662: \cite{benzi_2}, eq. (\ref{K62}), namely
663: %________________________________________________________________________________________
664: \begin{eqnarray}
665: \label{GESS}
666: < \dV^p > & \propto & \frac{< \epsilon_r^{p/3} >} {< \epsilon >^{p/3}}
667: \, < \dV^3 >^{p/3} \,.
668: \end{eqnarray}
669: %________________________________________________________________________________________
670: This equation, which we will hereafter refer to as RKSH, has been carefully 
671: checked against experimental data for homogeneous and isotropic turbulence and confirmed also 
672: in those cases where neither $< \epsilon_r^{p/3}>$ nor the structure functions showed a 
673: scaling behavior in $r$. 
674: The two terms in eq.~(\ref{GESS}), namely $< \epsilon_r^{p/3} >$ and $< \dV^3 >$, take into 
675: account the fluctuations in the energy dissipation rate and the energy transfer through the 
676: inertial scales, respectively.
677: 
678: Scaling laws as discussed so far heavily rely on the assumption of isotropy.
679: In many turbulent flows, however, isotropy is broken by the mean shear and by the boundary 
680: conditions and a large production of turbulent kinetic energy occurs due to the interaction 
681: of the mean flow with the turbulent fluctuations.
682: We may wonder if scaling laws emerge also in such conditions, and, in case, under which
683: respects they differ from the classical ones.
684: 
685: Recent numerical investigations of wall turbulence have shown how intermittency increases
686: approaching the wall and how the scaling exponents $\zeta_p$ are not consistent with
687: equation (\ref{GESS}). Actually, in the wall region, the main dynamical process is 
688: represented by the momentum transfer occurring in the wall normal direction associated 
689: with a large production of turbulent kinetic energy via the Reynolds stresses
690: $d U /dy <u v>$. In this case a term of the form $d U /dy <\dV^2>$ is expected to
691: account for the new balance between production and dissipation in the Karman-Howarth 
692: equation. 
693: Based on this analysis a new form of similarity law has been proposed for the wall region 
694: in terms of the structure function of order two \cite{PF}
695: %________________________________________________________________________________________
696: \begin{eqnarray}
697: \label{new}
698: < \dV^p > & \propto & \frac{< \epsilon_r^{p/2} >} {< \epsilon >^{p/2}}
699: \, < \dV^2 >^{p/2} \,.
700: \end{eqnarray}
701: %________________________________________________________________________________________
702: 
703: As discussed in \cite{PF} for a turbulent channel flow, the classical RKSH, eq. (\ref{GESS}),
704: holds at the center of the channel in the bulk region where the momentum transfer 
705: due to the shear term is negligible with respect to inertial transfer. The new form of 
706: similarity law holds in the near wall region where high turbulent kinetic 
707: energy production occurs. 
708: 
709: We are presently going to check the consistency of equation (\ref{new}) in the homogeneous 
710: shear flow where, in the absence of solid walls, the mean shear $S$ is constant. 
711: In this flow the effect of the shear is isolated from other concurrent effects such as the 
712: suppression of wall normal velocity fluctuations and the non uniform momentum transfer across 
713: adjacent layers in the normal direction. Moreover the statistical analysis is simplified 
714: since homogeneity is exploited in all spatial directions.
715: 
716: In figure 18 the similarity law (\ref{GESS}), for $p=6$, is applied to DNS data 
717: for both isotropic and homogeneous shear turbulence. It holds for all the resolved 
718: scales in the isotropic case. For the shear flow, a clear failure is observed instead. 
719: In this case two different slopes may be extracted from the data, one in the 
720: dissipative region with the trivial value of $0.99$, the other, in the inertial region, 
721: with value of $0.92$.  
722: Analogous results are found for $p=9$, figure 19, with a single slope for 
723: isotropic turbulence.
724: For the shear flow, the dissipative slope is $0.98$ confirming the quality of our DNS
725: while the inertial fit gives a slope of $0.88$.
726: 
727: A similar analysis is presented in figures 20 and 21 
728: for the new similarity law,
729: eq. (\ref{new}), in the case of the homogeneous shear flow,  for $p=4$ and $p=6$. 
730: A unique slope can be effectively fitted all over the range of resolved scales with the same 
731: value of $1$ independently of the order of the moment. 
732: The failure of the classical RKSH and the validity of the new form for the shear flow
733: are still better appreciated in the compensated plot of equations (\ref{GESS}) and (\ref{new}) 
734: versus the separation $r$, figure 22 and 23.
735: 
736: As a technical detail concerning the evaluation of the RKSH, we remark
737: that $L_s=0.8$ and that the dissipation fit is extracted from the range of separations 
738: $[0, \, l_{fit}]$, with $l_{fit}=0.2$. 
739: According to our theoretical considerations, for $r \ll L_s$ we should recover the classical 
740: RKSH. 
741: However, the present value of $0.15$ for $S_c^*$ implies that when $r \ll L_s$ we also 
742: have $r \sim \eta$, hence no room is left for the establishment of the classical form 
743: of scaling below $L_s$.
744: 
745: %________________________________________________________________________________________
746: \subsection{Extended self similarity}
747: %________________________________________________________________________________________
748: Before closing the section, it is  worth commenting on the issue of ESS scaling in 
749: shear flows.
750: Actually, the use of equation (\ref{ESS}) requires some caution in the present conditions, 
751: where a superposition of different laws may be present.
752: Clearly, a blind fit through the entire available range can extract an effective slope. 
753: It is much safer, instead, to consider the local slopes in detail, to understand to what 
754: extent a power law is really observed, and, more interestingly, the range of scales over which 
755: a certain value for the exponent is found.
756: As an example, in figures 24 and 25,
757: the local slopes,  $p=6$ and $p=9$, for the shear flow are contrasted with those of  
758: homogeneous isotropic turbulence.
759: In both cases at small separation the dissipative scaling $\dV^p \propto r^p$ is 
760: achieved confirming the good quality of the data. As we move towards inertial separations
761: the local slopes of isotropic turbulence remain quite constant over almost one decade.
762: For the shear flow instead a definite deviation from constant is found at approximately 
763: $r / \eta =15$, with the shear scale falling near by ($L_s/\eta=16$). 
764: Similarly to what found in the lower part of the logarithmic region of the turbulent 
765: channel \cite{PRL}, the local slopes do not display clear ESS scaling regions, presumably 
766: due to the competition of two different statistical trends and to the lack of sufficient 
767: scale separation to distinguish between the two.
768: 
769: %________________________________________________________________________________________
770: \subsection{An estimate of the flatness}
771: %________________________________________________________________________________________
772: A different estimate of the flatness is here proposed taking into account the similarity
773: laws given by eq. (\ref{GESS}) and (\ref{new}). Coming back to the issue of 
774: intermittency, let us consider again the flatness factor of the random variable 
775: $\dV(r)$, 
776: %________________________________________________________________________________________
777: \begin{eqnarray}
778: \label{F_dv}
779: F(r) & = &  \frac{<\dV^4(r)>}{<\dV^2(r)>^2}  \, .
780: \end{eqnarray}
781: %________________________________________________________________________________________
782: According to the two forms of RKSH, eqs.~(\ref{GESS}) and (\ref{new}), the same flatness 
783: factor can be expressed in terms of moments of the dissipation field as
784: %________________________________________________________________________________________
785: \begin{eqnarray}
786: \label{F_w}
787: F_b  \, = \, \frac{<\epsilon_r^{4/3}>}{<\epsilon_r^{2/3}>^2} & \qquad \qquad &
788: F_w  \, = \, \frac{<\epsilon_r^{2}>}{<\epsilon_r>^2}  \,,
789: \end{eqnarray}
790: %________________________________________________________________________________________
791: respectively.
792: The plots in figure 26 show the curves corresponding to the 
793: three equations (\ref{F_dv}) and (\ref{F_w}). 
794: The values computed using the new RKSH, $F_w$ in eq.~(\ref{F_w}), are in excellent
795: agreement with those evaluated from the definition (\ref{F_dv}). 
796: It should be noted that the use of the wrong form of RKSH, corresponding to expression
797: $F_b$ in eq.~(\ref{F_w}), substantially underestimates the intermittency.
798: 
799: The same analysis can be performed for homogeneous and isotropic turbulence. In the plots
800: of figure 27, we show $F$, $F_b$ and $F_w$ for turbulence 
801: measured at the center of a jet far downstream ($R_\lambda \sim 800$)\cite{benzi_1}. 
802: As one can easily observe, the new form of refined similarity hypothesis fails in homogeneous 
803: isotropic conditions, as we should expect.
804: %________________________________________________________________________________________
805: 
806: The ability of the new RKSH to predict the amount of intermittency
807: in the velocity increments from the pdf of the dissipation field is, in our opinion, 
808: a strong check on the validity of the proposed form of scaling.
809: %________________________________________________________________________________________
810: \section{Statistical properties of the dissipation field}
811: %________________________________________________________________________________________
812: 
813: Given its role in both the new and the classical form of RKSH, the study of the dissipation 
814: field may give further insight in the behavior of shear turbulence. To this purpose 
815: figure 28 reports the self-scaling of the dissipation field
816: in logarithmic scale. The solid line with slope $\alpha =3.1$ shows clearly that for 
817: sufficiently large separations $r$, which  correspond to small values of the two moments, 
818: the power law $<\epsilon_r^3> \; \propto \; <\epsilon_r^2>^\alpha$ is able to fit the data.
819: The scaling is strikingly similar to that found in homogenous and isotropic turbulence, 
820: where the estimated exponent is $3$.
821: The self-scaling seems to hold for all the moments we have been able to check, and
822: the values found for the shear flow are always in agreement with those of homogeneous 
823: and isotropic turbulence, see table \ref{tabella}.
824: This analysis suggests that relations of the form 
825: %________________________________________________________________________________________
826: \begin{eqnarray}
827: \label{diss_scal}
828: <\epsilon_r^q> & \propto & <\epsilon_r^2>^{\alpha_q}
829: \end{eqnarray}
830: %________________________________________________________________________________________
831: \noindent with, more or less, fixed exponents are able to describe the statistics
832: of the fluctuations of the dissipation field either in shear or isotropic turbulence.
833: Hence, basic statistical properties of the dissipation field remain substantially unchanged
834: in the two cases.  This issue becomes even more evident when considering the second moment 
835: of the dissipation as a function of $r$. This quantity  is reported in figure 
836: 29, where the triangles and the circles, corresponding to isotropic 
837: and shear turbulence respectively, indicate a substantially identical behavior.
838: All together, these findings suggest a universal form for the pdf of the dissipation field, 
839: $p(\epsilon_r / \bar{\epsilon})$ normalized by its mean value $\bar{\epsilon}$ which is
840: obviously dependent on the flow field.
841: %________________________________________________________________________________________
842: \section{Concluding remarks \& comments}
843: %________________________________________________________________________________________
844: 
845: The direct numerical simulation of a homogeneous shear flow in a confined box has been 
846: analyzed to address the issue of scaling laws and intermittency in the simplest conceivable 
847: environment where anisotropy is prevailing. The dynamics of the vortical structures observed 
848: in the DNS is characterized by a regeneration cycle associated to a substantial fluctuation 
849: in the level of turbulent kinetic energy. The regeneration mechanism has been discussed in 
850: detail, and our analysis confirms substantially to previous descriptions \cite{kida} which, 
851: however, where restricted to the initial phase of flow development.
852: In fact, in complete agreement with \cite{pumir_2} and \cite{pumir_3}, we confirm that 
853: the confinement of the
854: flow to a periodic box generates a dynamics, which, although strongly characterized by the
855: regeneration cycle of the vortical structures, reaches statically stationary conditions. 
856: The characteristic time of the system seems to be quite insensitive to changes in the aspect 
857: ratio of the computational box,  hence we feel that the present flow may be considered as a 
858: valid prototype for the study of shear flows. 
859: 
860: We find an increased intermittency, as evaluated by the flatness of the velocity increments,
861: with respect to that observed in isotropic turbulence. In isotropic turbulence, the 
862: intermittency of the velocity increments is related to the intermittent behavior of the 
863: dissipation field via the Refined Kolmogorov Similarity Hypothesis.
864: For the shear flow, instead, we observe a clear failure of the RKSH which confirms the 
865: conclusions previously reached from data in the buffer layer of a channel flow (DNS) and 
866: successively verified with hot-wire measurements in a flat plate boundary layer.
867: 
868: To attempt a clarification of this point, we have considered the possible existence of a 
869: self-scaling behavior of the structure functions.
870: Under this respect, the extended self-similarity expressing the $p^{th}$-order 
871: structure function as a power law with exponent $\zeta(p)$ in terms of, say, the third 
872: order moment, is a well established feature of isotropic turbulence which is able
873: to characterize the amount of intermittency in terms of the scaling exponents.
874: For the homogeneous shear flow, as already observed in the lower part of the logarithmic region of 
875: turbulent channel flow, scaling exponents can be defined only for a limited region 
876: of scales ($r < L_s$) above which a competition between two different statistical behaviors 
877: makes it difficult to define any scaling at all. However, recent experimental 
878: results concerning the log-layer of a turbulent boundary layer at a higher Reynolds
879: number, seem to indicate the existence of a double scaling regime \cite{ruiz}.
880: 
881: Nonetheless, we are presently able to grasp the nature of the intermittency in the shear flow,
882: since  the failure of the RKSH is accompanied by the establishment of a new form of 
883: scaling, as proposed in \cite{PF}.
884: Originally the idea behind the new scaling law was to account for the crucial role of the 
885: shear in originating a substantial production of turbulent kinetic energy via the Reynolds 
886: stress.  
887: In principle, this mechanism, directly associated to the presence of a mean shear,
888: is independent of other details of the flow, such as the presence of nearby
889: solid boundary or the specific dynamics of the vortical structures which sustain the turbulent 
890: fluctuations. This view is confirmed by the present results which show how the new form
891: of RKSH holds not only for the buffer layer of wall bounded flows, as originally verified in 
892: \cite{PF}, but also in the present homogeneous shear flow.
893: Here the dynamics of the vortical structures, a part from a general resemblance, is in fact 
894: much different from that observed in the wall region and no wall is present to inhibit
895: wall normal motions. Nonetheless the new scaling law is well established, confirming that its
896: presence is the signature of a predominant effect of the shear in the proper range of scales.
897: 
898: Beyond the direct observation of the new scaling law, the detailed analysis presented in the
899: paper shows how the increased intermittency of the longitudinal velocity increments in shear 
900: dominated flow is, in fact, to be ascribed to the new scaling law.
901: Actually, the statistical properties of the energy dissipation field are substantially
902: similar to those we find in isotropic conditions.
903: From our data, the scaling behavior of the energy dissipation field can not be consistent 
904: both with the observed level of intermittency of the velocity structure functions and with
905: the existence of the classical RKSH.
906: On the contrary, the new RKSH is able to  consistently predict the level of intermittency in 
907: the velocity structure function in terms of the sole dissipation field.
908: This confirms that the intermittency of velocity structure functions at shear dominated scales
909: is largely affected by the instantaneous process of turbulent kinetic energy production. 
910: In other words, since the dissipation field is approximately unchanged, to compute the
911: increased intermittency of the velocity in presence of shear we have to include 
912: explicitly the production mechanism in the scaling law.
913: 
914: These comments suggest that the source of intermittency in the shear flow is in principle 
915: substantially different from that operating in isotropic turbulence.
916: To speculate a little further on this point, let us recall that, a part from the aspect ratio 
917: of the box, the confined homogeneous shear flow is defined by two dimensionless parameters, 
918: namely $S^*$ and $S_c^*$. 
919: These two parameters control the position of the shear scale $L_s$ relative to the length scale
920: of the large eddies, $l_d$, and to the Kolmogorov scale, $\eta$, respectively.
921: A large value of $S^*$ implies a substantial range of scales where the shear is crucial
922: to the dynamics. 
923: On the other hand, a small value of $S_c^*$  corresponds to an extended range of scales 
924: where the dynamics is purely inertial and where the effect of shear is largely irrelevant.
925: One should expect the classical scaling laws of homogeneous and isotropic turbulence 
926: to emerge clearly in the limit of $S_c^*$ small within the range $ L_s >> r >> \eta$.
927: A major contribution of the present paper is to confirm the existence of a new form of scaling
928: law which emerges in the limit of $S^*$ large, in the range $l_d >> r >> L_s$.
929: Ideally, when both $S^*$ and $1/S_c^*$ are large, one should expect a coexistence of the two
930: laws in the two different scaling ranges.
931: In this ideal conditions, the intermittent behavior of velocity increments is induced
932: by the energy injection associated to the turbulent kinetic energy production mechanism
933: and is originated in the shear dominated range, where the scaling is totally different from
934: the classical RKSH.
935: However, once the energy is transferred to the smaller scales where direct injection
936: due to shear is irrelevant, the classical energy cascade mechanism takes over.
937: The intermittency of the velocity increments at the shear dominated scales appears to smaller 
938: scales which belongs to the classical inertial range as large scale modulations. 
939: Along the classical line of reasoning, the large scale modulations, through the cascade, are 
940: reduced to the intermittent behavior  of the dissipation field.
941: This is tantamount to postulating a certain universality of the probability distribution of the 
942: dissipation field, once, in the spirit of K62 theory, universal values for the anomalous 
943: corrections to the exponents of the structure functions are assumed in the classical part of
944: the inertial range. 
945: These concepts are consistent with our numerical experiments with the homogeneous shear flow, 
946: even though the scale separation is not sufficient to observe both scalings simultaneously.
947: 
948: To conclude we like to comment about the different sets of data available to
949: us in the present analysis.  As figure \ref{mappa_any} shows, a large shear dominated 
950: range, i.e. $S^*$ large, is not frequently met in practice.
951: An exception is the buffer region of wall bounded flows. There, almost the entire range of
952: available scales falls above $L_s$, which for this case is also very close to the Kolmogorov 
953: length, $\eta$, since $S_c^* \simeq 1$. 
954: These are optimal conditions to observe the new RKSH as a pure law, without interference with 
955: its classical counterpart since the contiguous purely inertial range below $L_s$ does not exist
956: at all. The logarithmic region, both from DNS and experiments,  present  
957: $L_s \simeq l_d$, i.e. $S^* \simeq 1$, and $L_s >> \eta$. Here only the classical scaling 
958: can be expected. The conditions of the present confined homogeneous shear flows are somehow 
959: intermediate between the ones of the buffer and the log region respectively.
960: Here $L_s$ falls almost exactly in the middle of the available range of scales, a more complex
961: situation and by far the more interesting to discuss. If the shear dominated and the purely 
962: inertial range were large enough we might have observed the simultaneous presence of the two 
963: forms of scaling law. However this cannot be  the case with the Reynolds number we can reach 
964: by DNS. In practice, what we observe in the present case  can be described as a sort of 
965: superposition of the two regimes, with a dominating contribution from the new RKSH. 
966: %________________________________________________________________________________________
967: %%
968: %% Tabella esponenti autoscaling dissipazione
969: %%
970: \newpage
971: \begin{table}
972: \caption{Scaling exponents, $\alpha_q$, of the moments $<\epsilon_r^q>$ with respect to 
973:          $<\epsilon_r^2>$, for different, non-integer, values of $q$, 
974:          see eq. (\ref{diss_scal}).
975:          \label{tabella} }
976: \begin{tabular}{ccc}
977:  $\alpha_{q}$  &   Hom. Iso.  &      Hom. Shear      \\
978:   \hline
979: $\alpha_{2/3}$ &   -0.11      &      -0.11           \\
980: $\alpha_{4/3}$ &    0.22      &       0.22           \\
981: $\alpha_{5/3}$ &    0.55      &       0.55           \\
982: $\alpha_{6/3}$ &    1.00      &       1.00           \\
983: $\alpha_{7/3}$ &    1.56      &       1.57           \\
984: $\alpha_{8/3}$ &    2.10      &       2.25           \\
985: $\alpha_{9/3}$ &    3.00      &       3.10           \\
986: \end{tabular}
987: \end{table}
988: %________________________________________________________________________________________
989: %________________________________________________________________________________________
990: %_______________________________________________________________________________
991: \begin{thebibliography}{99}
992: %_______________________________________________________________________________
993: \bibitem{kolm_41}
994:  Kolmogorov,
995:  {\em{The local structure of turbulence in incompresiible viscous fluid for very
996:       large Reynolds number}},
997:  {Dolk. Akad. SSSR} {\bf 30}, 301, {1941}, reprinted in {Proc. R. Soc. Lond.},
998:  A {\bf434}, 15-19, {1991}.
999: %_______________________________________________________________________________
1000: \bibitem{Landau}
1001:  Landau, Lifshitz,
1002:  {\em{Fluid Mechanics}},
1003:  {Vol. 6 course of theoretical physics}, {Pergamon press}, {1975}.
1004: %_______________________________________________________________________________
1005: \bibitem{kolm_62}
1006:  Kolmogorov, 
1007:  {\em{A refinement of previous hypotheses concerning the local structure of 
1008:       turbulence in incompresiible viscous fluid for very
1009:       large Reynolds number}},
1010:  {J. Fluid. Mech.} {\bf 13}, 82-85, {1962}.
1011: %_______________________________________________________________________________
1012: \bibitem{benzi_1}
1013:  Benzi, Ciliberto, Baudet, Ruiz Chavarria,
1014:  {\em{On the scaling of three dimensional homogeneous and isotropic turbulence}},
1015: {Physica D}, {\bf 80}, 385-398, {1995}.
1016: %_______________________________________________________________________________
1017: \bibitem{frisch}
1018:  Frisch,
1019:  {\em{Turbulence}},
1020:  {Cambridge university press }, {1995}.
1021: %_______________________________________________________________________________
1022: \bibitem{benzi_2}
1023:  Benzi, Biferale, Ciliberto, Struglia, Tripicccione,
1024:  {\em{Generalized scaling in fully developed turbulence}},
1025: {Physica D}, {\bf 96}, 162-181, {1996}.
1026: %_______________________________________________________________________________
1027: \bibitem{PF}
1028:  Benzi, Amati, Casciola, Toschi,  Piva,
1029:  {\em{Intermittency and scaling laws for wall bounded turbulence}},
1030:  {Phys. Fluids}, {\bf 11}(6), 1-3, {1999}.
1031: %_______________________________________________________________________________
1032: \bibitem{rogers}
1033:  Rogers and Moin, 
1034:  {\em{The structure of vorticity in homogeneous turbulent flow}},
1035:  {J. Fluid Mech.}, {\bf 176}, 33-66, {1987}.
1036: %_______________________________________________________________________________
1037: \bibitem{lee}
1038:  Lee, Kim, and Moin,
1039:  {\em{Structure of turbulence at high shear rate}},
1040:  {J. Fluid Mech.}, {\bf 216}, 561-583, {1990}.
1041: %_______________________________________________________________________________
1042: \bibitem{kida}
1043:  Kida and Tanaka,
1044:  {\em{Dynamics of vortical structures in homogeneous shear flow}},
1045:  {J. Fluid Mech.}, {\bf 274}, 43-68, {1994}.
1046: %_______________________________________________________________________________
1047: \bibitem{rose}
1048:  Rose,
1049:  {\em{Results of an attemp to generate a homogeneous turbulent shear flow}},
1050:  {J. Fluid Mech.}, {\bf 25}, 97-120, {1966}.
1051: %_______________________________________________________________________________
1052: \bibitem{champagne}
1053:  Champagne, Harris, Corrsin, 
1054:  {\em{Experiments on nearly homogeneous turbulent shear flow}},
1055:  {J. Fluid Mech.}, {\bf 41}, 81-139, {1970}.
1056: %_______________________________________________________________________________
1057: \bibitem{tavoularis_1}
1058:  Tavoularis and Corrsin, 
1059:  {\em{Experiments in nearly homogeneous turbulent shear flow with a uniform
1060:       mean temperature gradient. Part 1}},
1061:  {J. Fluid Mech.}, {\bf 104}, 331-347, {1981}.
1062: %_______________________________________________________________________________
1063: \bibitem{tavoularis_2}
1064:  Tavoularis and Corrsin,
1065:  {\em{Experiments in nearly homogeneous turbulent shear flow with a uniform
1066:       mean temperature gradient. Part 2, the fine structure}},
1067:  {J. Fluid Mech.}, {\bf 104}, 349-367, {1981}.
1068: %_______________________________________________________________________________
1069: \bibitem{saddoughi}
1070:  Saddoughi and Veeravalli,
1071:  {\em{Local isotropy in turbulent boundary layer at high Reynolds number}},
1072:  {J. Fluid Mech.}, {\bf 268}, 333-372, {1994}.
1073: %_______________________________________________________________________________
1074: \bibitem{pumir_2}
1075:  Pumir and Sharaiman,
1076:  {\em{Persistent small scale anisotropy in homogeneous shear flow}},
1077:  {Phys. Rev. Lett.}, {\bf 75}(17), 3114-3117, {1995}.
1078: %_______________________________________________________________________________
1079: \bibitem{pumir_3}
1080:  Pumir, 
1081:  {\em{Turbulence in homogeneous shear flow}},
1082:  {Phys. Fluids}, {\bf 8}(11), 3112-3127, {1996}.
1083: %_______________________________________________________________________________
1084: \bibitem{garg}
1085:  Garg and Warhaft, 
1086:  {\em{On the small scale structure of simple shear flow}},
1087:  {Phys. Fluids}, {\bf 10}(3), 662-673, {1998}.
1088: %_______________________________________________________________________________
1089: \bibitem{shen}
1090:  Shen and Warhaft,
1091:  {\em{The anisotropy of small scale structures in high Reynolds number 
1092:       $(Re_{\lambda} \sim 1000)$ turbulent shear flow}},
1093:  {Phys. Fluids}, {\bf 12}(11), 2976-2982, {2000}.
1094: %_______________________________________________________________________________
1095: \bibitem{rogallo}
1096:  Rogallo,
1097:  {\em{Numerical experimients in homogeneos turbulence}},
1098:  {NASA T.M.}, {\bf 81315}, {1981}.
1099: %_______________________________________________________________________________
1100: \bibitem{solver}
1101:  Lundbladh, Henningson, Johansson,
1102:  {\em{An efficient spectral integration method for the solution of the 
1103:       Navier Stokes equations}},
1104:  {FFA-TN 1992-28}
1105: %_______________________________________________________________________________
1106: \bibitem{townsend}
1107:  Townsend,
1108:  {\em{The structure of turbulent shear flow}},
1109:  {Second edition, Cambridge university press }, {1976}.
1110: %_______________________________________________________________________________
1111: \bibitem{ferchichi}
1112:  Ferchichi and Tavoularis,
1113:  {\em{Reynolds number effects on the fine structure of uniformly shared 
1114:       turbulence}},
1115:  {Phys. Fluids}, {\bf 12}(11), 2942-2953, {2000}.
1116: %_______________________________________________________________________________
1117: \bibitem{chong}
1118:  Chong, Perry, Cantwell,
1119:  {\em{A general classification of three dimensional flow fields}},
1120:  {Phys. Fluids} A{\bf 2} 765-777 {1990}
1121: %_______________________________________________________________________________
1122: \bibitem{pumir_1}
1123:  Pumir,
1124:  {\em{A numerical study of the mixing of a passive scalar in three dimension in 
1125:       the presence of a mean gradient}},
1126:  {Phys. Fluids}, {\bf 6}(6), 2118-2132, {1994}.
1127: %_______________________________________________________________________________
1128: \bibitem{PRL}
1129: Toschi, Amati, Succi, Benzi,  Piva, 
1130: {\em{Intermittency and structure functions in channel flow turbulence}},
1131:  {Phys. Rev. Lett}, {\bf 82}(25), 5044-5049, {1999}.
1132: %_______________________________________________________________________________
1133: \bibitem{ruiz}
1134: Ruiz-Chavarria, Ciliberto, Baudet, Leveque,
1135:  {\em{Scaling properties of the streamwise component of velocity in a 
1136:       turbulent boundary layer}},
1137: {Physica D}, {\bf 141}, 183-198, {2000}.
1138: %_______________________________________________________________________________
1139: \end{thebibliography}
1140: %________________________________________________________________________________________
1141: %________________________________________________________________________________________
1142: \end{document}
1143: