1: %\documentclass[preprint,floatfix,amsmath,amssymb]{revtex4}
2: \documentclass[twocolumn,floatfix,amsmath,amssymb]{revtex4}
3: %\documentclass[pre,preprint,showpacs,floatfix,amsmath,amssymb]{revtex4}
4: \usepackage{graphicx}
5: \usepackage{dcolumn}
6: \usepackage{color}
7:
8: \def\NOTE#1{{\textcolor{red}{\bf [#1]}}} % note
9: \def\ADD#1{{\textcolor{blue}{#1}}} % question
10: \def\AD#1{{\textcolor{magenta}{#1}}} % plug a value, a ref, ...
11: \def\DEL#1{{\textcolor{green}{ #1}}} % suggested deletion in text
12: \def\BB#1{{\textcolor{blue}{\bf #1}}}
13:
14: % table definitions
15: \def\hfq{\hfill\quad}
16: \def\cc#1{\hfq#1\hfq}
17: \def\tvi{\vrule height 12pt depth 5pt width 0pt}
18: \def\traithorizontal{\noalign{\hrule}}
19: \def\tv{\tvi\vrule}
20: %
21: \newcommand{\be}{\begin{equation}}
22: \newcommand{\ee}{\end{equation}}
23: \newcommand{\uh}{\hat{u}}
24: \def\eg{{\it e.g.}\ }
25: \def\etal{{\it et al.}}
26: \def\ie{{\it i.e.}\ }
27: \def\lhs{{\it l.h.s.}\ }
28: \def\op{{\it op. cit.}\ }
29: \def\resp{{\it resp.}\ }
30: \def\rhs{{\it r.h.s.}\ }
31: \def\rms{{\it r.m.s.}\ }
32: \def\viz{{\it viz.}\ }
33: \def\vs{{\it vs.}\ }
34: \def\al{Alfv\'en\ }
35: \def\els{Els\"asser variables\ }
36: \def\kol{Kolmogorov\ }
37: \def\nse{Navier-Stokes equations\ }
38: \def\u{{\bf u}} \def\v{{\bf v}} \def\x{{\bf x}}
39: \def\dv{\delta {\bf v}}
40: %
41: \newcommand{\curlv} {\nabla \times {\bf v}}
42: \newcommand{\ba}{\mathbf{a}} \newcommand{\bb}{\mathbf{b}}
43: \newcommand{\bA}{\mathbf{A}} \newcommand{\bB}{\mathbf{B}}
44: \newcommand{\Asz}{A_{s_z}}
45: \newcommand{\alp}{\alpha} \newcommand{\alpm}{\alpha^{-1}}
46: \newcommand{\alpmm}{\alpha^{-1}_m} \newcommand{\alpmv}{\alpha^{-1}_v}
47: \newcommand{\bc}{\mathbf{c}} \newcommand{\bd}{\mathbf{d}}
48: \newcommand{\bj}{\mathbf{j}} \newcommand{\bk}{\mathbf{k}}
49: \newcommand{\bom}{\mbox{\boldmath $\omega$}}
50: \newcommand{\bomp}{\mbox{\boldmath $\omega^+$} }
51: \newcommand{\bomm}{\mbox{\boldmath $\omega^-$} }
52: \newcommand{\bompm}{\mbox{\boldmath $\omega^{\pm}$} }
53: % bold face
54: \newcommand{\br}{\mathbf{r}}
55: \newcommand{\bu}{\mathbf{u}}
56: \newcommand{\bv}{\mathbf{v}}
57: \newcommand{\bw}{\mathbf{w}}
58: \newcommand{\bAs}{\mathbf{A_s}}
59: \newcommand{\bjs}{\mathbf{j_s}}
60: \newcommand{\bus}{\mathbf{u_s}}
61: \newcommand{\bBs}{\mathbf{B_s}}
62: \newcommand{\boms}{\mbox{\boldmath $\omega_s$}}
63: \newcommand{\bx}{\mathbf{x}} \newcommand{\bxp}{\mathbf{x^{\prime}}}
64: \newcommand{\bzp}{\mathbf{z^{+}}} \newcommand{\bzm}{\mathbf{z^{-}}}
65: \newcommand{\bzpm}{\mathbf{z^{\pm}}} \newcommand{\bzmp}{\mathbf{z^{\mp}}}
66: \newcommand{\ca}{{\rm a}} \newcommand{\caa}{{\rm aa}}
67: \newcommand{\cab}{{\rm a} b} \newcommand{\vba}{vb{\rm a}}
68: \newcommand{\K}{{\cal K}} \newcommand{\ud}{{\langle{u}^2 \rangle}}
69: \newcommand{\li}{\ell_{I}} \newcommand{\Rla}{R_{\lambda}}
70: \newcommand{\up}{{{\bf u}({\bf x})}} \newcommand{\R}{{\cal R}}
71: \newcommand{\dr}{{\partial_r}} \newcommand{\dt}{{\partial_t}}
72: \newcommand{\vg}{{{\bf v(x)} \cdot \nabla}}
73: %
74: %\topmargin -3pt
75:
76: \begin{document}
77: \title{Large scale flow effects, energy transfer, and self-similarity
78: on turbulence}
79:
80: \author{P.D. Mininni, A. Alexakis, and A. Pouquet}
81: \affiliation{NCAR, P.O. Box 3000, Boulder, Colorado 80307-3000, U.S.A.}
82: \date{\today}
83:
84: \begin{abstract}
85: The effect of large scales on the statistics and dynamics of
86: turbulent fluctuations is studied using data from high resolution
87: direct numerical simulations. Three different kinds of forcing, and
88: spatial resolutions ranging from $256^3$ to $1024^3$, are being used.
89: The study is carried out by investigating the nonlinear triadic
90: interactions in Fourier space, transfer functions, structure functions,
91: and probability density functions. Our results show that the large scale
92: flow plays an important role in the development and the statistical
93: properties of the small scale turbulence. The role of helicity is
94: also investigated. We discuss the link between these findings and
95: intermittency, deviations from universality, and possible
96: origins of the bottleneck effect. Finally, we briefly describe the
97: consequences of our results for the subgrid modeling of turbulent flows.
98: \end{abstract}
99:
100: \pacs{47.27.ek; 47.27.Ak; 47.27.Jv; 47.27.Gs}
101: \maketitle
102:
103: \section{Introduction}
104:
105: The nature of the interactions in a turbulent flow has been a long
106: standing problem. Most investigations are focused on statistically
107: homogeneous and isotropic turbulence. However, in most cases in nature
108: this assumption is not necessarily true. Instead, one usually finds
109: turbulence to be embedded in a large scale flow. In these cases,
110: turbulence originates from some instability (e.g. convection) and is
111: not uniformly distributed in space and time. To what extend the flow
112: is approaching a statistically homogeneous and isotropic state as
113: smaller and smaller scales are developed is still an open problem.
114:
115: Due to insufficient computational resources (e.g in astrophysics,
116: weather and climate prediction, for tokamaks, and in industrial
117: applications) the computational efforts to study turbulent flows at
118: these high Reynolds numbers are compelled to resort to modeling of the
119: smaller scale turbulent fluctuations for which numerical resolution is
120: not available. However, in order to accurately
121: model the small scales a good understanding of the impact of the large
122: scale flow in the small scales is needed, in addition to the knowledge
123: of what (if any) is the feedback of the small scale turbulent flow
124: on the large scale structures. This latter process is usually modeled
125: by an effective viscosity, or other forms of subgrid modeling of the
126: Reynolds stress tensor. These two types of interactions (from large to small
127: and from small to large scales, or ``downscaling'' and ``upscaling'')
128: are often considered to take place only through a local cascade of energy,
129: although there is evidence that
130: nonlocal interactions between widely separated scales are also relevant.
131:
132: Studies of local and nonlocal interactions have been done previously
133: in direct numerical simulations (DNS), although at moderate spatial
134: resolutions and Reynolds numbers \cite{Domaradzki88,Domaradzki90,Kerr90,
135: Yeung91,Okhitani92,Zhou93,Zhou93b,Brasseur94,Yeung95,Zhou96,Kishida99}.
136: While some of these studies supported the existence of nonlocal
137: interactions, it was argued by later studies that
138: their presence was associated with moderate values of the
139: Reynolds number in the simulations, or that it was linked to the precise
140: definitions used for transfer functions and the interpretation of
141: the results. In some cases \cite{Okhitani92,Zhou93,Zhou93b} large
142: eddy simulations (LES) were also used to extend the range of Reynolds
143: numbers studied, although if nonlocal interactions are present the impact
144: of the subgrid model on the transfer is unclear and is a point of study
145: in itself. Some of these studies
146: also considered the effect of anisotropic or coherent large scale forcing
147: (see e.g. Refs. \cite{Yeung91,Yeung95,Zhou96}).
148: Evidence of nonlocal interactions between large and small scales has
149: been found also in experiments \cite{Wiltse93,Wiltse98,Carlier01}.
150: Observations of the persistence of anisotropy at small scales in
151: experiments at large Reynolds numbers \cite{Sreenivasan97,Shen00}, in
152: the atmosphere \cite{Stewart69}, and in numerical simulations
153: (see e.g. Ref. \cite{Biferale01a} and references therein) also suggest
154: the existence of a direct coupling between disparate scales. The presence
155: of nonlocal interactions has also been associated in numerical studies
156: with departures from universality and the development of intermittency
157: \cite{Laval01}. Recently, a new study at high resolution
158: \cite{Alexakis05b} presented a detailed analysis of nonlocal interactions
159: using DNS for large Reynolds numbers, although for a particular and
160: coherent forcing function.
161:
162: In this work we examine three different kinds of flows in DNS in
163: triple periodic boxes, generated by different body forces. We explore
164: external forcings with and without helicity, and forcings with either infinite
165: or short correlation times. The energy injection scale is varied, as
166: well as the viscosity and spatial resolution of the runs, to explore a
167: wide range of Reynolds numbers and configurations. In all cases, we
168: observe the presence of nonlocal interactions coexisting with a local
169: direct cascade of energy. The intensity of the nonlocal interactions
170: depends on the transfer function studied. For a particular (highly
171: symmetric) forcing, we also observe a correlation between regions of
172: large scale shear and strong small scale gradients, supporting previous
173: studies that linked intermittent effects with interactions with the
174: large scale flow.
175:
176: \section{Setup and theory}
177:
178: \subsection{Equations, code, and simulations}
179:
180: For an incompressible fluid with constant mass density, the Navier-Stokes
181: equations are:
182: %
183: \begin{equation}
184: \partial_t {\bf u} + {\bf u}\cdot \nabla {\bf u} = - \nabla p
185: + \nu \nabla^2 {\bf u} +{\bf f} ,
186: \label{eq:momentum}
187: \end{equation}
188: %
189: \begin{equation}
190: \nabla \cdot {\bf u} =0,
191: \label{eq:incompressible}
192: \end{equation}
193: %
194: where ${\bf u}$ is the velocity field, $p$ is the pressure divided by
195: the mass density, and $\nu$ is the kinematic viscosity. Here, ${\bf f}$
196: is an external force that drives the turbulence. The mode with the largest
197: wavevector in the Fourier transform of ${\bf f}$ is going to be denoted
198: as $k_F$ and we are going to refer to $2 \pi k_F^{-1}$ as the forcing
199: scale. We also define the viscous dissipation wavenumber as
200: $k_\nu=(\epsilon/\nu^3)^{1/4}$, where $\epsilon$ is the energy injection
201: rate (as a result, the Kolmogorov scale is $\eta = 2\pi/k_\nu$). A large
202: separation between the two scales ($k_F^{-1} \gg k_\nu^{-1}$) is required
203: for the flow to reach a turbulent state.
204:
205: In the absence of external forcing and viscosity, the Navier-Stokes
206: equations in three dimensions have two ideal invariants: the energy
207: \begin{equation}
208: E = \frac{1}{2} \int{u^2 d {\bf x}^3} \, ,
209: \end{equation}
210: and the helicity
211: \begin{equation}
212: H = \frac{1}{2} \int{{\bf u} \cdot \nabla \times {\bf u} \, d{\bf x}^3} \, .
213: \end{equation}
214:
215: All the results discussed in the following sections result from analysis
216: of data from direct numerical simulations of the Navier-Stokes equations.
217: We solve Eqs. (\ref{eq:momentum}) and (\ref{eq:incompressible}) using a
218: parallel pseudospectral code in a three dimensional domain of size $2\pi$
219: with periodic boundary conditions \cite{Gomez05a,Gomez05b}. The pressure
220: is obtained by taking the divergence of Eq. (\ref{eq:momentum}), using the
221: incompressibility condition (\ref{eq:incompressible}), and solving the
222: resulting Poisson equation. The equations are evolved in time using a
223: second order Runge-Kutta method. The code uses the $2/3$-rule for
224: dealiasing, and as a result the maximum wavenumber is $k_{max} = N/3$
225: where $N$ is the number of grid points in each direction. All simulations
226: presented are well resolved, in the sense that the dissipation wavenumber
227: $k_\nu$ is smaller than the maximum wavenumber $k_{max}$ at all times.
228:
229: The Reynolds number is defined as $R_e = UL/\nu$, where $U$ is the
230: r.m.s. velocity and $L$ is the integral lengthscale of the flow
231: %
232: \begin{equation}
233: L = 2\pi \frac{\int{E(k) k^{-1} dk}}{\int{E(k) dk}},
234: \label{eq:integral}
235: \end{equation}
236: %
237: where $E(k)$ is the energy spectrum. The large scale turnover time
238: can then be defined as $T=U/L$. We can also introduce the Taylor
239: based Reynolds number $R_\lambda = U\lambda/\nu$, where the Taylor
240: lengthscale $\lambda$ is given by
241: %
242: \begin{equation}
243: \lambda = 2\pi \left(\frac{\int{E(k) dk}}{\int{E(k) k^2 dk}}\right)^{1/2}.
244: \label{eq:taylor}
245: \end{equation}
246:
247: Several simulations were done with different resolutions (from $N=256$ to
248: $1024$) and kinematic viscosities. Table \ref{table:runs} shows the
249: parameters for all the runs. The rms velocity in the steady state of all
250: the runs is $U\approx 1$. To asses the effect of different large scale
251: stirring forces, we used three expressions for the external force
252: ${\bf f}$. The first expression corresponds to a Taylor-Green (TG) flow
253: \cite{Taylor37}
254: {\setlength\arraycolsep{2pt}
255: \begin{eqnarray}
256: {\bf f}_{\rm TG} &=& f_0 \left[ \sin(k_F x) \cos(k_F y)
257: \cos(k_F z) \hat{x} - \right. {} \nonumber \\
258: && {} \left. - \cos(k_F x) \sin(k_F y)
259: \cos(k_F z) \hat{y} \right] ,
260: \label{eq:TG}
261: \end{eqnarray}}
262: where $f_0$ is the force amplitude. This expression is not a solution of
263: the Euler's equations, and as a result small scales are generated fast
264: when the fluid is stirred with this forcing. The resulting flow has no net
265: helicity, although regions with strong positive and negative helicity
266: develop.
267:
268: In order to study directly the effect of helicity and its transfer
269: between different scales,
270: we also did simulations using the Arn'old-Childress-Beltrami (ABC) forcing
271: {\setlength\arraycolsep{2pt}
272: \begin{eqnarray}
273: {\bf f}_{\rm ABC} &=& f_0 \left\{ \left[B \cos(k_F y) +
274: C \sin(k_F z) \right] \hat{x} + \right. {} \nonumber \\
275: && {} + \left[A \sin(k_F x) + C \cos(k_F z) \right] \hat{y} +
276: {} \nonumber \\
277: && {} + \left. \left[A \cos(k_F x) + B \sin(k_F y) \right]
278: \hat{z} \right\},
279: \label{eq:ABC}
280: \end{eqnarray}}
281: with $A=0.9$, $B=1$, and $C=1.1$. The ABC flow is an eigenfunction of
282: the curl and an exact solution of the Euler equations. When the flow
283: is forced using this expression, turbulence develops after an
284: instability sets in \cite{Podvigina94}. Unlike TG forcing, ABC forcing
285: injects net helicity into the flow.
286:
287: The amplitude and phase of these two forcings is kept constant during the
288: simulations, and as a result the external force has an infinite correlation
289: time. It is a common practice in studies of isotropic and homogeneous
290: turbulence to force in Fourier space injecting energy in all modes
291: in a Fourier shell and changing the phase of each mode with a short
292: correlation time. To compare with the results of TG and ABC forcing, we
293: also implemented a random forcing
294: \begin{equation}
295: {\bf f}_{\rm RND} = f_0 \sum_{|{\bf k}|=k_F} i {\bf k} \times
296: \hat{\bf x} \, e^{i ({\bf k} \cdot {\bf x} + \phi_{\bf k})}
297: \label{eq:random}
298: \end{equation}
299: where the phase $\phi_{\bf k}$ of each mode with wavevector ${\bf k}$ was
300: changed randomly with a correlation time $\tau_c$ that was taken to be
301: $\tau_c=0.1 T$.
302:
303: \begin{table}
304: \caption{\label{table:runs}Parameters used in the simulations. $N$ is
305: the linear grid resolution, ${\bf f}$ the forcing [either Taylor
306: Green (TG), Beltrami (ABC) or random (RND)], $k_F$
307: the forcing wavenumber, $\nu$ the kinematic viscosity, $R_e$
308: the Reynolds number, and $R_\lambda$ the Taylor based Reynolds
309: number.}
310: \begin{ruledtabular}
311: \begin{tabular}{ccccccc}
312: Run & $N$ & ${\bf f}$ & $k_F$ & $\nu$ & $R_e$ & $R_\lambda$ \\
313: \hline
314: I & 256 & TG & 2 &$2\times 10^{-3}$ & 675 & 300 \\
315: II & 512 & TG & 2 &$1.5\times 10^{-3}$& 875 & 350 \\
316: III & 1024 & TG & 2 &$3\times 10^{-4}$ & 3950 & 800 \\
317: IV & 256 & ABC & 10 &$2.5\times 10^{-3}$& 275 & 230 \\
318: V & 256 & ABC & 3 &$2\times 10^{-3}$ & 820 & 360 \\
319: VI & 512 & ABC & 3 &$6.2\times 10^{-4}$& 2520 & 670 \\
320: VII & 1024 & ABC & 3 &$2.5\times 10^{-4}$& 6200 & 1100 \\
321: VIII& 256 & RND & 1 &$1.5\times 10^{-3}$& 2030 & 650 \\
322: \end{tabular}
323: \end{ruledtabular}
324: \end{table}
325:
326: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
327: \subsection{Scale interactions and transfer}
328: To investigate the interactions between different scales we split the
329: velocity field into spherical shells in Fourier space of unit width,
330: \ie $\bv=\sum_K {\bf v}_K$ where ${\bf v}_K$ is a filtered velocity
331: field such that only the Fourier modes in the shell $K\le|k|<K+1$
332: (from now on called shell $K$) are kept.
333: From equation
334: (\ref{eq:momentum}), the rate of energy transfer $T_3(K,P,Q)$ (a
335: third-order correlator) from energy in shell $Q$ to energy in shell $K$
336: due to the interaction with the velocity field in shell $P$ is defined
337: as usual \cite{Kraichnan71,Lesieur,Alexakis05,Alexakis05b} as:
338: \begin{equation}
339: T_3(K,P,Q) = -\int \bv_K \cdot (\bv_P \cdot \nabla) \bv_Q d{\bf x}^3 \ .
340: \label{trans_eq3}
341: \end{equation}
342: Note that this term does not give information about the energy the
343: shell $P$ receives or gives to the shells $K$ and $Q$.
344: The computation of $T_3$ for the three shells $K$, $P$, and $Q$
345: from 1 up to a wavenumber $K_{max}$ requires $\sim K_{max}^3 N^3$
346: operations and is therefore demanding in computer resources. For example,
347: in the $1024^3$ simulations, to compute $T_3(K,P,Q)$ for all wavenumbers
348: up to $K_{max}=80$, it takes as much computing time as
349: evolving the hydrodynamic code for two turnover times.
350:
351: If we sum over the middle wave number $P$ we obtain the total energy
352: transfer $T_2(K,Q)$ from shell $Q$ to shell $K$:
353: %
354: \begin{equation}
355: T_2(K,Q) = \sum_P T_3(K,P,Q) = -\int \bv_K \cdot (\bv \cdot \nabla)
356: \bv_Q d{\bf x}^3 \ .
357: \label{trans_eq2}
358: \end{equation}
359: Positive transfer implies that energy is transfered from shell $Q$ to $K$,
360: and negative from $K$ to $Q$; thus, both $T_3$ and $T_2$ are antisymmetric
361: in their $(K,Q)$ arguments (see Ref. \cite{Alexakis05}). $T_2(K,Q)$ gives
362: information on the shell-to-shell energy transfer between $K$ and $Q$,
363: but not about the amplitude of the triadic interactions
364: themselves.
365:
366: If we further sum over the wave number $Q$ we obtain the transfer
367: \begin{equation}
368: T_1(K) = \sum_Q T_2(K,Q) = -\int \bv_K \cdot (\bv \cdot \nabla) \bv d{\bf x}^3
369: \label{trans_eq1}
370: \end{equation}
371: that expresses the rate the shell $K$ receives energy from the velocity
372: field (all shells).
373:
374: The energy flux is reobtained from these transfer functions as
375: \begin{equation}
376: \Pi(k) = -\sum_{K=0}^k T_1(K)=-\int \bv_K^< (\bv\cdot \nabla )\bv_K d{\bf x}^3
377: \label{eq:flux}
378: \end{equation}
379: where the notation
380: \begin{equation}
381: \bv_K^<=\sum_{K'=0}^K \bv_{K'} \,\,\, \mathrm{and} \,\,\,
382: \bv_K^>=\sum_{K'=K+1}^\infty \bv_{K'}
383: \end{equation}
384: is used (see Ref. \cite{Frisch}).
385:
386: We can further define the flux of energy at some given scale due to
387: the interactions with some other scale as
388: \begin{equation}
389: \Pi_P(k) = -\sum_{K=0}^k \sum_Q T_3(K,P,Q)=
390: -\int \bv_K^< \cdot (\bv_P \cdot \nabla) \bv^> d{\bf x}^3 \,\, ,
391: \end{equation}
392: which expresses the flux of energy at the scale $K^{-1}$ due to the
393: interactions with the scale $P^{-1}$
394: or in other words the energy flux at the scale $K^{-1}$ if the velocity
395: field was advected just by the velocity field at scale $P^{-1}$.
396: As will be discussed later, the
397: question of locality depends on which of the different transfer
398: functions or flux is under investigation.
399:
400: It is worth noticing that through the study of the three functions
401: $T_1$, $T_2$, $T_3$, and the fluxes $\Pi_P$, we can measure the nature
402: and intensity of the interactions directly,
403: and therefore we avoid
404: introducing a non-locality parameter as was done e.g. in Refs.
405: \cite{Zhou93,Zhou93b}.
406:
407: The total energy balance equation for a given shell is written as
408: \begin{equation}
409: \partial_t E(K) = T_1(K)+ \nu {\mathcal D(K)} + {\mathcal F}(K)
410: \end{equation}
411: where we have also introduced the dissipation function
412: \begin{equation}
413: \nu {\mathcal D}(K) \equiv \nu \int |{\bf \nabla u}_K|^2 d{\bf x}^3 ,
414: \end{equation}
415: and the energy injection rate to the velocity field through the forcing term
416: \begin{equation}
417: {\mathcal F}(K) \equiv \int { \bf f} \cdot {\bf u}_K \, d{\bf x}^3 \,.
418: \end{equation}
419:
420: Finally, we will also investigate the transfer of helicity
421: among different scales. Taking the inner product of the
422: Navier-Stokes equation (\ref{eq:momentum}) with the vorticity at a scale $K^{-1}$
423: ($\bw_K=\nabla \times \bu_K$) and adding the inner product of the velocity at
424: the same scale $K^{-1}$ with the $curl$ of (\ref{eq:momentum})
425: and space averaging we obtain:
426: \begin{eqnarray}
427: \partial_tH(K)=&\sum_Q \int \bw_K \cdot (\bu \times \bw_Q) d{\bf x}^3+
428: \nonumber \\
429: &\int \bw_K \cdot F d{\bf x}^3 +\nu \int \bw_K \cdot \nabla \times
430: \bw_K d{\bf x}^3.
431: \end{eqnarray}
432: Each term of the sum in the first term in the $r.h.s$
433: of the equation above can be written as:
434: \begin{equation}
435: T_H(K,Q) = \int \bw_K \cdot (\bu \times \bw_Q) d{\bf x}^3 \ .
436: \label{trans_eqh}
437: \end{equation}
438: Note that the anti-symmetric property $T_H(K,Q)=-T_H(Q,K)$ holds for
439: $T_H(K,Q)$ \ie the rate the shell $K$ is gaining helicity from the
440: interaction with the field $\bw_Q$ and $\bu$ is equal to the rate the
441: shell $Q$ is loosing helicity through the same interaction. This allows
442: us to interpret $T_H(K,Q)$ as the transfer of helicity from the
443: scale $Q^{-1}$ to the scale $K^{-1}$ (see \cite{Alexakis06} for the
444: equivalent definition of the magnetic helicity transfer).
445:
446: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
447: \section{\label{sec:runIII}Influence of the large scale flow on turbulence}
448:
449: We discuss in this section properties of Run III. First we describe the
450: geometry of the resulting flow and we present general statistical results,
451: such as probability density functions and structure functions. Then, we
452: analyze in detail the scale interactions using the formulation discussed
453: in the previous section. Preliminary results for this flow were presented
454: in \cite{Alexakis05b}. The results discussed here will be used as a
455: reference to compare with the rest of the simulations in the following
456: sections.
457:
458: %%%%%%%%%%
459: \subsection{Statistical properties, structure functions, and scaling exponents}
460:
461: \begin{figure}
462: \centerline{\includegraphics[width=8.3cm]{fig1}}
463: \caption{(Color online) Rendering of enstrophy density in a
464: $1024 \times 64 \times 1024$
465: slice of Run III. Bands of strong (1,2,3,4) and weak (A,B,C,D)
466: shear in the external Taylor-Green force are indicated. The bars
467: at the bottom show respectively the integral, Taylor, and
468: dissipation scales.}
469: \label{fig:render}
470: \end{figure}
471:
472: Run III is a $1024^3$ simulation using TG forcing. After reaching a
473: turbulent steady state, the simulation was continued for 10 turnover
474: times. The expression of the forcing has several spatial symmetries, and
475: some of these symmetries are recovered in the flow in a statistical
476: sense. Figure \ref{fig:render} shows a rendering of enstrophy density in a
477: thin slice of $1024 \times 64 \times 1024$ grid points. Thin and elongated
478: structures (vortex tubes) can be identified. As observed in previous
479: studies, these structures are characterized by one large lengthscale
480: in one direction (ranging from the Taylor to the integral scale) and
481: a short lengthscale (close to the viscous dissipation scale) in the other
482: two directions.
483:
484: \begin{figure}
485: \centerline{\includegraphics[width=7cm]{fig2}}
486: \caption{(Color online) Zoom on a region with large enstrophy density
487: showing field lines inside
488: (above) and in its surrounding (below) in Run III. Velocity field
489: lines are helical inside and in the vicinity of the tube. Note the
490: merging of two tubes in the south west corner. The side of the region
491: shown is approximately one tenth of the side of the total
492: domain.}
493: \label{fig:tube}
494: \end{figure}
495:
496: Since $k_F=2$, there are four planes in the $z$ direction where the
497: external force is zero (see Eq. \ref{eq:TG}) and the shear is maximum.
498: In the steady state turbulent flow, these planes can be easily recognized
499: since four bands where the vorticity is stronger are formed around them.
500: Regions with less (or weaker) vortex tubes separate these planes,
501: regions centered around the planes where the large scale shear has
502: a minimum. Together, these two sets of regions
503: form a large scale pattern that is observed to persist for long times,
504: giving four ``quiet'' stripes and four stripes of ``stronger" turbulence
505: (the boundary between these regions is not well defined and fluctuates
506: in space and time, see Fig. \ref{fig:render}). Considering the persistence
507: of these statistical
508: symmetries of the flow, the velocity increments for this Run will be
509: computed for displacements {\it only} in the $x$-$y$ plane.
510:
511: \begin{figure}
512: \centerline{\includegraphics[width=8.3cm]{fig3}}
513: \caption{Energy spectrum compensated by $k^{-5/3}$ for Run III. The
514: inset shows the energy flux.}
515: \label{fig:spectrum_runIII}
516: \end{figure}
517:
518: When individual vortex tubes are studied, it is seen that the
519: flow inside and surrounding the vortex tube is helical, as noted before by several authors
520: \cite{Tsinober83,Moffatt85,Moffatt86,Levich87,Farge01}. Figure
521: \ref{fig:tube} shows a rendering of a small region in the domain
522: (approximately one tenth of the box). A region of large enstrophy
523: density is indicated by dark (blue) colors, and field lines inside
524: and in the vicinity of the enstrophy containing region are shown.
525: In this region and its surroundings, field lines are helical, while
526: the flow far from the region is not. This feature has been verified
527: for a large set of vortex tubes in the complete domain. It has been
528: claimed in the past \cite{Moffatt92,Tsinober} that the development of these
529: helical structures in a turbulent flow can lead to the depletion of
530: nonlinearity and a quenching of local interactions. We will come back
531: to this issue in the following sections.
532:
533: \begin{figure}
534: \centerline{\includegraphics[width=8.3cm]{fig4}}
535: \caption{Probability density functions of transverse velocity increments
536: for Run III, for increments $l=20 \eta$ ($\cdots$), $l=10 \eta$
537: ($- - -$), $l=4 \eta$ ($-\cdot-$), $l=2 \eta$ ($-\cdots-$), and
538: $l=\eta$ (solid line), where $\eta$ is the Kolmogorov dissipation
539: scale.}
540: \label{fig:pdf_runIII}
541: \end{figure}
542:
543: In spite of the presence of the large scale pattern, many features often
544: associated with isotropic and homogeneous turbulence can be observed
545: in this simulation. Figure \ref{fig:spectrum_runIII} shows the
546: angle-averaged Fourier energy spectrum and energy flux. An inertial
547: range with constant energy flux is observed, together with a
548: Kolmogorov-like scaling and a bottleneck as the dissipative range is
549: reached. When probability density functions (pdfs) of transverse
550: velocity increments
551: \begin{equation}
552: \delta v_\perp ({\bf x},l) = \hat{\bf r} \times \left[ {\bf v}({\bf x} +
553: l\hat{\bf r}) - {\bf v}({\bf x}) \right],
554: \label{eq:incrementperp}
555: \end{equation}
556: are computed in the whole domain (Fig. \ref{fig:pdf_runIII}), we observe
557: distributions close to Gaussian for large increments, and the development
558: of exponential and stretched exponential tails as the increment $l$ is
559: decreased.
560:
561: \begin{figure}
562: \centerline{\includegraphics[width=8.3cm]{fig5}}
563: \caption{Probability density functions of transverse velocity increments
564: with $l=10 \eta$ for Run III. The solid line corresponds to the
565: whole domain, dashed lines to regions 1 to 4 (strong shear),
566: and dotted lines to regions A to D (weak shear; see Fig.
567: \ref{fig:render}). Notice the weak but systematic differences:
568: the four dashed lines (some of them overlapping) have slower decaying
569: tails and the dotted lines have faster decaying tails.}
570: \label{fig:pdf10eta}
571: \end{figure}
572:
573: \begin{figure}
574: \centerline{\includegraphics[width=8.3cm]{fig6}}
575: \caption{Probability density functions of transverse velocity increments
576: with $l=2 \eta$ for Run III. Same labels as in Fig.
577: \ref{fig:pdf10eta}. Note that the differences between quiet
578: and strong regions appear even more systematic at this scale than in Fig.
579: \ref{fig:pdf10eta}.}
580: \label{fig:pdf2eta}
581: \end{figure}
582:
583: Figures \ref{fig:pdf10eta} and \ref{fig:pdf2eta} show pdfs of transverse
584: velocity increments with $l=10 \eta$ and $2 \eta$ respectively, but
585: discriminating between bands of strong and weak shear (as indicated in Fig.
586: \ref{fig:render}). Bands with strong shear (1,2,3, and 4) have
587: slightly but systematically stronger
588: tails (i.e. a larger probability of strong gradients) than regions A,B,C,
589: and D. Note this behavior is true for each individual region, and the
590: difference increases as the increment $l$ is decreased. This confirms
591: statistically what can be inferred from Fig. \ref{fig:render}: regions
592: of strong shear have a larger density of vortex tubes and stronger
593: gradients.
594:
595: \begin{figure}
596: \centerline{\includegraphics[width=8.3cm]{fig7}}
597: \caption{First order structure function $S_1(l)$ for Run III. The solid
598: line corresponds to the whole domain, dashed lines to regions
599: 1,2,3, and 4, and dotted lines to regions A,B,C, and D. The
600: Kolmogorov scaling and the linear slope corresponding
601: to a smooth flow are shown as a reference. The arrow indicates
602: the Kolmogorov scale $\eta$. Notice again the systematic
603: differences between the two regimes identified in Fig. 1.}
604: \label{fig:S1}
605: \end{figure}
606:
607: \begin{figure}
608: \centerline{\includegraphics[width=8.3cm]{fig8}}
609: \caption{Second order structure function $S_2(l)$ for Run III. Labels
610: are as in Fig. \ref{fig:S1}.}
611: \label{fig:S2}
612: \end{figure}
613:
614: We can also compute longitudinal velocity increments
615: \begin{equation}
616: \delta v_\parallel ({\bf x},l) = \hat{\bf r} \cdot \left[ {\bf v}({\bf x} +
617: l \hat{\bf r}) - {\bf v}({\bf x}) \right] .
618: \label{eq:incrementpar}
619: \end{equation}
620: Figures \ref{fig:S1} and \ref{fig:S2} show respectively the second and
621: third order longitudinal structure functions, where the structure function
622: of order $p$ is defined as
623: \begin{equation}
624: S_p (l) = \left< \delta v_\parallel^p ({\bf x},l) \right> .
625: \label{eq:structure}
626: \end{equation}
627: At scales smaller than the dissipation scale the field is smooth and
628: the structure function of order $p$ scales as $l^p$. Kolmogorov's 1941
629: theory of turbulence predicts in the inertial range a scaling
630: $S_p (l) \sim l^{p/3}$, although corrections due to intermittency are
631: known and in general $S_p (l) \sim l^{\zeta_p}$, where $\zeta_p\not= p/3$ are
632: the scaling exponents.
633:
634: As with pdfs of velocity increments, a clear trend separating regions of
635: strong and weak shear is observed. In the range of scales corresponding to
636: the inertial range, the four regions with strong shear (regions 1,2,3 and 4)
637: show a larger slope than the four regions with weak shear. The slope of the
638: structure function computed in the whole box lies between these two
639: values. Considering the results from both pdfs and structure functions,
640: a correlation is observed between small scale gradients and large scale
641: shear. Note that a correlation between stronger tails in the pdfs
642: of velocity increments and vortex tubes has already been observed in
643: \cite{Levi01,Farge01}. Here, a correlation between these quantities with
644: large scale shear is further observed, in agreement with Ref. \cite{Laval01}
645: that suggests intermittency is related with interactions with the large
646: scale flow.
647:
648:
649: \subsection{Interactions, energy transfer, and flux}
650:
651: \begin{figure}
652: \centerline{\includegraphics[width=8.3cm]{fig9}}
653: \caption{(Color online) Above: amplitude of triadic interactions
654: as given by $T_3(K,P,Q)$
655: at $Q=40$ for Run III. Solid lines correspond to positive contour
656: levels, and dotted lines to negative values. The dark (red) shaded
657: region indicates shells in $P$ close to the forcing shell, while
658: the light (orange) shaded region indicates shells in $P$ outside
659: the local octave. Below: total shell-to-shell energy transfer
660: $T_2(K,Q)$, transfer $T_2^{\textrm LS}(K,Q)$ due to interactions
661: with $P$ close to the forcing shell (dashed line), and difference
662: between these two transfers (dash-dotted line).}
663: \label{fig:triads}
664: \end{figure}
665:
666: Figure \ref{fig:triads} shows the functions $T_3(K,P,Q)$ and $T_2(K,Q)$
667: evaluated at $Q=40$ in the turbulent steady state of Run III. When triadic
668: interactions between shells are studied, the strongest interactions are
669: with the shell $P=3$, where the large scale forcing is. Local
670: interactions between Fourier shells ($K \sim P \sim Q$) are two orders
671: of magnitude smaller than nonlocal interactions with $P \sim k_F$.
672: As a result, when $T_3(K,P,Q)$ is summed over all $P$ to obtain the
673: shell-to-shell energy transfer $T_2(K,Q)$, the energy is observed to
674: be transfered to small scales locally between shells, but with a step
675: proportional to $k_F$. In $T_2(K,Q)$, the negative peak at
676: $K \sim Q-k_F$ means energy is transfered from this shell to the shell
677: $Q$, while the positive peak at $K \sim Q+k_F$ indicates energy is
678: transfered to this shell from the shell $Q$.
679:
680: The function $T_3(K,P,Q)$ was studied in the steady state of this run
681: also for $Q=10$ and $Q=20$, obtaining the same quantitative results: a
682: dominance (when compared in amplitude) of triadic interactions with the
683: large scale flow over local triadic interactions.
684: The $T_2(K,Q)$ function for all values of $Q$ between $10$ and $80$ also
685: peaks at $K\approx Q \pm k_F$ (see Ref. \cite{Alexakis05b}). These peaks
686: at fixed values of $K-Q$ in $T_2(K,Q)$ are just the signature of the
687: strong triadic interactions with the large scale forcing. If we compute
688: the shell-to-shell transfer between shells $K$ and $Q$ due only to
689: interactions with the large scale flow
690: \begin{equation}
691: T_2^{\textrm LS} (K,Q) = \sum_{P=0}^6 T_3(K,P,Q) \, ,
692: \label{eq:T2LS}
693: \end{equation}
694: we obtain most of the shell-to-shell transfer (see Fig. \ref{fig:triads}).
695: The remaining transfer $T_2(K,Q)-T_2^{\textrm LS}(K,Q)$ still peaks at
696: larger wavenumbers, although it still does not peak at $K \sim Q = 40$.
697: This is due to nonlocal interactions outside the octave
698: band not related with the large scale forcing (light shaded region in
699: Fig. \ref{fig:triads}). Note that the definition of the range summed over
700: $P$ in Eq. (\ref{eq:T2LS}), or of octave bands (here defined as
701: $Q/2 \lesssim K \lesssim 2Q$) is somewhat arbitrary.
702:
703: \begin{figure}
704: \centerline{\includegraphics[width=8.3cm]{fig10}}
705: \caption{Total energy flux $\Pi(k)$ in Run III (solid line), flux
706: $\Pi^{\textrm LS}(k)$ due to interactions with the large scale
707: flow (dashed line), and nonlocal flux $\Pi^{\textrm NL}(k)$.}
708: \label{fig:fluxes_runIII}
709: \end{figure}
710:
711: As previously mentioned, while individual triadic interactions as described
712: by $T_3(K,P,Q)$ are dominantly nonlocal, the shell-to-shell transfer
713: $T_2(K,P,Q)$ describes a local transfer of energy although through
714: interactions with the large scale flow. Clearly when we sum over shells,
715: the larger number of modes in the small scales start dominating over the
716: large scale modes. Summing further over $K$ and $Q$, from Eq.
717: (\ref{eq:flux}) we can obtain the total energy flux. In analogy with Eq.
718: (\ref{eq:T2LS}) we can also define the flux due to interactions with the
719: large scale flow (dark shaded region in Fig. \ref{fig:triads})
720: \begin{equation}
721: \Pi^{\textrm LS}(k) = \sum_{P=0}^{6} \Pi_P(k) \, ,
722: \end{equation}
723: and the nonlocal flux due to interactions outside an octave band (light
724: shaded regions in Fig. \ref{fig:triads})
725: \begin{equation}
726: \Pi^{\textrm NL}(k) = \sum_{P=7}^{k/2} \Pi_P(k) \, ,
727: \end{equation}
728: where $\Pi_P(k)$ is defined in Eq. (15).
729:
730: Figure \ref{fig:fluxes_runIII} shows these three fluxes in the steady
731: state of Run III. The large scale flow is only responsible for a small
732: (but not insignificant) fraction of the total energy flux
733: ($\Pi^{\textrm LS} \simeq 0.2 \Pi$).
734: Furthermore although the amplitude of the nonlocal flux
735: $\Pi^{\textrm NL}(k)$ depends on the definition of an octave band, it is
736: remarkable that it peaks at wavenumbers close to the peak of the
737: bottleneck in the energy spectrum (see Fig. \ref{fig:spectrum_runIII}).
738: This gives a direct confirmation that the bottleneck is due to the
739: depletion of the energy transfer due to local interactions with
740: $K \sim P \sim Q$. These interactions are inhibited because of the
741: presence of a numerical and/or viscous cutoff in wavenumber, and
742: nonlocal triads become dominant
743: \cite{Herring82,Falkovich94,Lohse95,Martinez97}. Indeed, in Ref.
744: \cite{Herring82} it was shown using the eddy-damped quasinormal Markovian
745: (EDQNM) approximation that the bottleneck disappears if nonlocal
746: interactions are excluded from the computation. Note that the spurious
747: drop of the fluxes $\Pi^{\textrm LS}$ and $\Pi^{\textrm NL}$ at $k=80$
748: in Fig. \ref{fig:fluxes_runIII} is because the functions $T_3(Q,P,Q)$
749: and $T_2(K,Q)$ were only computed up to this wavenumber.
750:
751: Therefore there seems to be a hierarchy concerning the importance of
752: nonlocality when one investigates the transfer functions.
753: At the most basic level when one investigates the transfer function $T_3$
754: the interactions with the large scale flow are the strongest ones.
755: When averaged over the middle wave number the resulting energy transfer $T_2$
756: becomes local but not self-similar with a maximum at $K-Q\sim k_F$. When
757: averaged further to obtain the energy flux, the non local interactions with
758: the large scale flow are only responsible for 20\% of the flux.
759:
760: %%%%%%%
761: \section{\label{sec:Reynolds}Scaling with Reynolds}
762:
763: In this section we discuss results from runs I, II, and III.
764: The three runs are forced using Eq. (\ref{eq:TG}), and the only parameters
765: changed between the runs are the spatial resolution and the kinematic
766: viscosity. This investigation allows us to study how the results about
767: the non-local interactions change as we increase the Reynolds number.
768: Qualitatively the functions $T_3(K,P,Q)$ and $T_2(K,Q)$ are similar for
769: all three runs: the strongest interactions [$T_3(K,P,Q)$] are with
770: the large scale flow and the energy transfer [$T_2(K,Q)$] is local with
771: the two peaks at $K-Q \sim k_F$.
772: For this reason we focus here on the study of the local
773: and nonlocal flux of energy as the Reynolds number is changed.
774:
775: \begin{figure}
776: \centerline{\includegraphics[width=8.3cm]{fig11}}
777: \caption{Scale variation of the flux ratio $\Pi^{\textrm LS}/\Pi$ for
778: increasing Reynolds numbers; Runs I (dash-dot line),
779: II (dash line), and III (solid line). Note the presence of a wavenumber range
780: where $\Pi^{\textrm LS}/\Pi$ becomes constant only for the
781: highest resolution run.}
782: \label{fig:fluxratio}
783: \end{figure}
784:
785: In Figure \ref{fig:fluxratio} we show the ratio of the flux
786: $\Pi^{\textrm LS}(k)$ to the total flux $\Pi(k)$. As discussed in the
787: previous section, the flux due to interactions with the large flow gets
788: diminished as we sum over more and more modes. As a result, in the
789: inertial range of Run III interactions with the large scale flow are
790: only responsible for $\sim 20\%$ of the energy flux. In the smaller
791: Reynolds number runs (I and II), the amount of flux due to interactions
792: with the large scale flow is larger. This implies that as we increase
793: the Reynolds number the fraction of the energy flux in the inertial
794: range due to interactions with the large scale flow decreases.
795: However, note that in Run III a region in the inertial range where
796: the ratio $\Pi^{\textrm LS}/\Pi$ is approximately constant is observed.
797: This region with constant ratio $\Pi^{\textrm LS}/\Pi$ is not present
798: in the simulations with lower Reynolds number.
799: Therefore DNS with even higher Reynolds number than what is accomplished
800: here are needed
801: to determine how does this ratio scale with the Reynolds number.
802:
803: \section{Effects of different forcing functions}
804:
805: \subsection{Forcing expression and correlation time}
806:
807: \begin{figure}
808: \centerline{\includegraphics[width=8.3cm]{fig12}}
809: \caption{Compensated energy spectrum $k^{5/3} E(k)$ (solid line) and
810: helicity spectrum $k^{5/3} H(k)/k_F$ (dashed line) in the
811: steady state of Run VII, where $k_F$ is the forcing wavenumber.}
812: \label{fig:spectrum_runVII}
813: \end{figure}
814:
815: Nonlocal interactions discussed in the previous section were observed
816: for runs with large scale non-helical forcing with infinite correlation
817: time. It is of interest to know how much of these results translate to
818: other kind of forcings. In this section we compare results for runs
819: III, VII and VIII. Run VII is a $1024^3$ simulation with constant helical
820: forcing, and as runs I-III, it also displays a well defined large scale
821: flow (for a description of the ABC flow, see e.g. Ref. \cite{Childress}).
822: In Run VIII the phases of the forcing function are changed randomly with
823: a short correlation time. Although the resolution in this run is smaller
824: ($256^3$), we will focus here on the effect of the forcing, since a
825: systematic study of changing the Reynolds number for fixed forcing
826: was presented in the previous section.
827:
828: Figure \ref{fig:spectrum_runVII} shows the energy and helicity
829: spectra compensated by a Kolmogorov law in the turbulent steady state of Run VII.
830: As proposed in Ref. \cite{Brissaud73} and
831: observed in several simulations
832: \cite{Andre77,Borue97,Chen03,Chen03b,Gomez04} the spectrum of helicity
833: follows a Kolmogorov law and is proportional to $k_F E(k)$. The transfer
834: of helicity will be discussed in Section \ref{sec:helicity}.
835:
836: \begin{figure}
837: \centerline{\includegraphics[width=8.3cm]{fig13}}
838: \caption{Scaling exponents of velocity structure functions $\zeta_p$
839: as a function of order $p$
840: in Runs III (TG forcing) and VII (ABC forcing). The Kolmogorov
841: K41 scaling \cite{Kolmogorov41} is indicated, as well as the She-Leveque (SL)
842: prediction \cite{She94}. Note the difference between TG and ABC.}
843: \label{fig:scaling}
844: \end{figure}
845:
846: Figure \ref{fig:scaling}
847: shows the scaling exponents $\zeta_p$ of the longitudinal structure
848: functions, computed for runs III and VII. The extended self-similarity
849: (ESS) hypothesis \cite{Benzi93,Benzi93b} was used to compute the anomalous
850: exponents, which show similar behavior for both runs. However,
851: the resulting exponents for $p>3$ from Run VII (ABC forcing) are slightly
852: smaller than the exponents from Run III (TG forcing); for example,
853: in Run III $\zeta_4 = 1.2737 \pm 0.0005$ and
854: $\zeta_8 = 2.136 \pm 0.006$, while in Run VII $\zeta_4 = 1.2647 \pm 0.0005$
855: and $\zeta_8 = 2.026 \pm 0.008$.
856: Note that the difference
857: between the scaling exponents for TG and ABC at order $p=4$, though small, is
858: nevertheless more than order of magnitude larger than the error arising from
859: measuring them by virtue of the ESS hypothesis which leads to negligible
860: errors.
861:
862: It should be noted that it is unclear whether
863: this observed difference is due to
864: problems linked with using the ESS methodology itself, or if it indicates
865: a departure from universality between
866: these two flows because of loss of homogeneity, as exemplified in Fig. 1.
867: Since the two forcing functions studied here are both to some degree
868: anisotropic and inhomogeneous, subleading contributions due to departures
869: from full symmetry could be responsible for
870: this discrepancy. In this context, a decomposition of the structure
871: functions into their isotropic and
872: anisotropic components \cite{Biferale01a,Biferale01b,Biferale02} could
873: help to asses the degree of universality of each component.
874: These points will necessitate further study and better resolved flows.
875:
876: \begin{figure}
877: \centerline{\includegraphics[width=8.3cm]{fig14}}
878: \caption{Shell-to-shell energy transfer function $T_2(K,Q)$ for Run VII.
879: The superimposed curves correspond to different values of $Q$,
880: with $Q\in [10,80]$.}
881: \label{fig:T2_runVII}
882: \end{figure}
883:
884: The shell-to-shell energy transfer $T_2(K,Q)$ for Run VII is shown in Fig.
885: \ref{fig:T2_runVII}. As for non-helical forcing (see Sec. \ref{sec:runIII}
886: and Ref. \cite{Alexakis05})
887: the shell-to-shell transfer function peaks at $K \approx Q \pm k_F$,
888: indicating the transfer of energy is local but mediated by nonlocal
889: interactions with the large scales. If we sum over $K$ and $Q$ to
890: obtain the energy flux, we reobtain the results discussed in Sec.
891: \ref{sec:Reynolds} for Run III: most of the energy flux is local,
892: although in the inertial range $\sim 20\%$ of the total flux is due
893: to interactions with the large scale forcing.
894:
895: The transfer function $T_2(K,Q)$ was also computed in the turbulent
896: steady state of Run VIII (see Fig. \ref{fig:T2_scale}.a). In this run
897: the phases of the external force are changed randomly with a short
898: correlation time. It is noteworthy that even in this case with
899: isotropic and random forcing, evidence is found of nonlocal
900: interactions with the large forcing scale: the $T_2(K,Q)$
901: function peaks at $K \approx Q \pm k_F$ for all wavenumbers $Q$ studied
902: (note that in this run $k_F=1$). Fig. \ref{fig:T2_scale} is discussed further below.
903:
904: \subsection{Forcing scale}
905:
906: \begin{figure}
907: \centerline{\includegraphics[width=8.3cm]{fig15}}
908: \caption{Shell-to-shell energy transfer function $T_2(K,Q)$ for runs with
909: different forcing wavenumbers $k_F$ as indicated by the vertical
910: lines; Run VIII with $k_F=1$ (a), Run V with $k_F=3$ (b), and
911: Run IV with $k_F=10$ (c). In each panel, the several curves
912: correspond to different values of $Q$ in the inertial and
913: dissipative ranges. Note that for each run, the peak of energy
914: transfer is centered close to $k_F$. }
915: \label{fig:T2_scale}
916: \end{figure}
917:
918: Through all this work we have shown transfer functions indicating that the
919: energy is locally transfered between scales but with a fixed step that
920: we associated with the forcing scale. In this subsection we compare
921: the transfer function $T_2(K,Q)$ in simulations forced at different
922: wavenumbers.
923:
924: Figure \ref{fig:T2_scale} shows $T_2(K,Q)$ for several
925: values of $Q$ computed in the steady state of runs VIII ($k_F=1$),
926: IV ($k_F=2$), and V ($k_F=10$). A clear correlation is observed
927: between the position of the peaks ($|K-Q|$) in the shell-to-shell
928: transfer function, and the forcing wavenumber $k_F$. In the last
929: case, the peaks at $|K-Q|$ slightly smaller than $k_F$ could be
930: associated with low Reynolds number effects.
931:
932: \section{\label{sec:helicity}Helicity transfer}
933:
934: The transfer of helicity is readily studied in Run VII, since the
935: external forcing injects maximum helicity and all scales are dominated
936: by the same sign of helicity. We will make no attempt here to separate
937: the different signs of helicity in the simulations (see e.g. Refs.
938: \cite{Waleffe91,Chen03,Chen03b}).
939:
940: \begin{figure}
941: \centerline{\includegraphics[width=8.3cm]{fig16}}
942: \caption{The spectrum of relative helicity $H(k)/kE(k)$
943: compensated by $k^{-1}$ for run VII. Note that the region where the
944: spectrum is flat indicates a power law of $k^{-1}$ in the relative
945: helicity. At large wavenumber, a $k^{1/2}$ slope is shown only
946: as a reference. The increase at large $k$ is indicative of an
947: excess of relative helicity in the small scales.}
948: \label{fig:relative}
949: \end{figure}
950:
951: Before discussing the transfer of helicity, it is of interest to study
952: spectral properties of helical flows. As previously discussed, the spectrum
953: of helicity follows an approximate $k^{-5/3}$ law (see Fig.
954: \ref{fig:spectrum_runVII}). As a result, the relative helicity $H(k)/E(k)k$
955: follows in the inertial range a $k^{-1}$ slope
956: \cite{Brissaud73,Andre77,Lesieur} (i.e. small scales are less helical
957: than large scales).
958: It was predicted in Refs. \cite{Ditlevsen01,Ditlevsen01b}
959: that the dissipation scale of helicity should be larger than the energy
960: dissipation scale, giving as a result a drop in the spectrum of relative
961: helicity faster than $k^{-1}$ for small scales. This argument would be
962: in agreement with the idea that small scales slowly recover the mirror
963: symmetry broken by the injection of helicity in the large scales. However,
964: previous simulations \cite{Gomez04} and this high resolution run both suggest that
965: there is an excess of relative helicity in the small scales, when compared
966: with the $k^{-1}$ drop. Note that this slower than predicted recovery of
967: symmetries in the small scales is also in agreement with the slower than
968: expected recovery of isotropy observed in experiments
969: \cite{Sreenivasan97,Shen00}.
970:
971: Figure \ref{fig:relative} shows $H(k)/E(k)$, i.e. the spectrum of relative
972: helicity $R(k)=H(k)/kE(k)$ compensated by $k^{-1}$.
973: A scaling of $k^{-1}$ for the relative helicity $R(k)$ thus corresponds
974: to a flat spectrum in Fig. 16,
975: as observed through the inertial range up to $k\sim 20$. However, at small
976: scales the compensated spectrum of relative helicity grows,
977: possibly as $k^{1/2}$ or steeper, indicating that
978: the spectrum of helicity $H(k)$ at small scales is dropping slower than
979: the spectrum of energy $E(k)$ (see also Fig. \ref{fig:spectrum_runVII}).
980: This can be associated with the presence of vortex tubes, that are
981: known to be helical (see Fig. \ref{fig:tube}).
982:
983: \begin{figure}
984: \centerline{\includegraphics[width=8.3cm]{fig17}}
985: \caption{Energy spectrum compensated by $k^{-5/3}$ for Runs V (dashed
986: line), VI (dotted line), and VII (solid line). All runs are for
987: ABC forcing at different Reynolds numbers. Note the wavenumber
988: on the x-axis has been divided by the dissipation wavenumber to
989: make all dissipation ranges coincide. A slope of $1/3$ close to
990: the bottleneck (corresponding to a power law of $k^{-4/3}$ in
991: the energy spectrum) is shown as a reference.}
992: \label{fig:bottleneck}
993: \end{figure}
994:
995: The effect of helicity, and indirectly the effect of nonlocal interactions
996: which can be modeled for example
997: through the introduction of a second time-scale in the problem in order to
998: distinguish between the eddy turn-over time and a helical characteristic
999: time, has also been used to explain the development of the bottleneck
1000: effect that occurs at the onset of the dissipative range, see e.g.
1001: \cite{Kurien04}. Ref. \cite{Kurien04} predicts a $k^{-4/3}$ energy
1002: spectrum for the bottleneck, that we found compatible with our spectra
1003: for runs V-VII (Fig. \ref{fig:bottleneck}). We also observe a $k^{-4/3}$
1004: range close to the bottleneck in the Taylor-Green flow that has no net
1005: helicity. Note that the argument in Ref. \cite{Kurien04} is based
1006: on the presence of non zero helicity locally. The transition from the
1007: inertial range to the bottleneck seems to be dependent on the Reynolds
1008: number. It is worth noting that while at resolutions of $256^3$ mostly
1009: a bottleneck is observed, in the $1024^3$ runs a short Kolmogorov-like
1010: scaling is found before the bottleneck takes place. The origin of the
1011: bottleneck based on the dominance of nonlocal interactions in the
1012: dissipative range has also been investigated
1013: \cite{Herring82,Falkovich94,Lohse95,Martinez97} as we discussed in
1014: Sec. \ref{sec:runIII}, and is independent of the presence of helicity.
1015: These arguments are not mutually exclusive, as local in space generation
1016: of helicity in the small scales can also quench local interactions between
1017: eddies of comparable sizes.
1018: However, the prediction in Ref. \cite{Falkovich94} for the spectral
1019: shape of the bottleneck is in disagreement with the spectra obtained
1020: in all simulations here. These points deserve further study at higher
1021: resolution if one is to be able to distinguish between the different
1022: ranges that may be occurring.
1023:
1024: \begin{figure}
1025: \centerline{\includegraphics[width=8.3cm]{fig18}}
1026: \caption{Shell-to-shell helicity transfer $T_H(K,Q)$ in Run VII, for
1027: all values of $Q$ from $10$ to $80$.}
1028: \label{fig:T2H}
1029: \end{figure}
1030:
1031: Finally, we discuss the shell-to-shell helicity transfer $T_H(K,Q)$
1032: in Run VII (Fig. \ref{fig:T2H}). As for the transfer of energy,
1033: the helical transfer function peaks at $K \approx Q \pm k_F$, although this
1034: transfer is noisier. The helicity cascades directly to smaller scales
1035: as the energy, confirming previous studies
1036: \cite{Borue97,Chen03,Chen03b,Gomez04}.
1037: The helicity is not a positive definite quantity, and as a result both
1038: signs of the transfer can indicate a direct cascade depending on the
1039: sign of the helicity itself. Moreover, as we study smaller scales and the
1040: relative helicity decreases, both signs have more predominance
1041: increasing the noise observed in $T_H(K,Q)$.
1042:
1043: \section{Conclusions}
1044:
1045: We start by summarizing our results. The energy transfer, triadic
1046: interactions, and statistical flow properties were studied in
1047: several high resolution DNS with periodic boundaries. The spatial
1048: resolution ranged from $256^3$ to $1024^3$. The Reynolds number based
1049: on the integral scale spanned values from $R_e \approx 275 $ to $6200$,
1050: while the Taylor-based Reynolds number varied between
1051: $R_\lambda \approx 230$ and $1100$. Three forcing functions were used,
1052: two coherent functions with zero and maximum net helicity respectively,
1053: and a random forcing with a short correlation time. The energy injection
1054: scale was also varied to study its impact on the energy transfer.
1055:
1056: Most statistical studies were done for Run III, a high resolution
1057: simulation with $R_\lambda \approx 800$ and TG forcing. The forcing and
1058: the resulting flow have spatial symmetries that allow us to identify
1059: planes with strong and weak large scale shear easily. While in the
1060: whole domain the standard results were reobtained (e.g. exponential
1061: and stretched exponential tails in the pdfs of velocity increments,
1062: and anomalous scaling of the structure functions), when studying
1063: individual regions a correlation between large scale shear and small
1064: scale gradients was found. In regions of strong shear, the tails of
1065: the pdfs of velocity increments are stronger than in regions of weak
1066: shear, even at scales as small as twice the dissipation scale. Also,
1067: structure functions show a slightly larger slope in regions of strong
1068: large scale shear, i.e. a larger departure from a Kolmogorov $p/3$ self-similar scaling.
1069:
1070: A correlation between stronger tails in the pdfs of velocity
1071: increments and the presence of vortex tubes has already been observed
1072: in different decompositions of the flow \cite{Levi01,Farge01}. Here,
1073: we observed a correlation between these quantities and the large scale
1074: shear. The correlation was observed to be persistent even at scales as
1075: small as the dissipation scale. Small differences were also observed
1076: in the anomalous scaling of the structure functions for TG and ABC
1077: forcing. However, it is unclear whether this is related with
1078: non-universal effects associated with interactions with the large scale
1079: forcing, or with the use of the ESS hypothesis. For individual vortex
1080: tubes in several flows, we verified that the flow inside and surrounding
1081: the vortex tube is helical, as found before
1082: \cite{Tsinober83,Moffatt85,Moffatt86,Levich87,Farge01}. Note that the
1083: development of helical structures in a turbulent flow can lead to
1084: the depletion on nonlinearity and a quenching of local interactions
1085: \cite{Moffatt92,Tsinober}.
1086:
1087: Concerning the energy transfer and triadic interactions, we confirmed
1088: for several forcing functions that the cascade of energy is local between
1089: Fourier shells, although it is strongly mediated by individual triadic
1090: interactions which are nonlocal in nature. As a result, the energy cascades
1091: from one shell to the next with a fixed step proportional to the forcing
1092: wavenumber $k_F$. This effect was observed even in simulations using
1093: random forcing with a correlation time one order of magnitude smaller
1094: than the large scale turnover time. No qualitative differences have been
1095: observed as the Reynolds number was changed.
1096:
1097: However, the degree of nonlocality observed depends on the quantity
1098: studied. A hierarchy is found in the relative amplitude of nonlocal
1099: effects in triadic interactions, the shell-to-shell energy transfer,
1100: and the energy flux. Triadic interactions are dominated by interactions
1101: with the large scale flow. As more modes are summed to define the
1102: shell-to-shell transfer and the flux, the larger population of small
1103: scale modes starts dominating, and in the $1024^3$ simulations the
1104: large scale flow is responsible for only $\sim 20\%$ of the total
1105: flux.
1106:
1107: An increase of the relative contribution of nonlocal interactions
1108: (with the large scale flow and with modes outside the octave band studied)
1109: in the total flux is observed as the dissipative range is reached. This
1110: result is in good agreement with claims that the bottleneck effect is due
1111: to the quenching of local interactions at the end of the inertial range
1112: because of the presence of a cut-off in wavenumbers
1113: \cite{Herring82,Lohse95,Martinez97}.
1114:
1115: It is worth noting that the local energy cascade through nonlocal
1116: interactions, or non negligible nonlocal interactions, has been observed
1117: in the past in simulations at lower Reynolds numbers (see e.g. Refs.
1118: \cite{Domaradzki90,Okhitani92,Yeung95,Zhou96}). Our results confirm
1119: the presence of interactions between disparate scales in a turbulent flow
1120: at much larger Reynolds numbers, and for a variety of forcing functions
1121: and forcing scales. The results discussed here also shed some light on
1122: the controversy in the literature about the relevance of the nonlocal
1123: interactions. It has been claimed that interactions are local if
1124: a different measure for the locality is introduced \cite{Zhou93,Zhou93b},
1125: or if wavelets or a binning of Fourier space in octaves is used
1126: \cite{Kishida99} (note that wavelets naturally introduce a binning in
1127: octaves of the spectral space). The hierarchy found in the different
1128: transfer functions is the reason for this apparent inconsistency
1129: between previous results. As more modes are summed (e.g. to define
1130: octaves in spectral space, or to define our partial fluxes $\Pi_P$),
1131: the small scales overcome the triadic interactions with the large scale
1132: flow, and local interactions give the largest contribution. This is also
1133: in agreement with recent theoretical results about the locality of the
1134: energy flux \cite{Eyink05}, or the locality of the shell-to-shell energy
1135: transfer in Fourier octave bands \cite{Eyink94,Verma05}.
1136:
1137: The fact that in simulations with $R_\lambda \sim 1000$ most of
1138: the flux is due to local interactions, does not preclude however the
1139: existence of strong nonlocal interactions with the large scale flow at
1140: the triadic interaction level, and it should be kept in mind that even
1141: at large values of $R_\lambda$ these interactions are responsible for
1142: a non-negligible fraction of the total flux. The presence of nonlocal
1143: interactions are a deviation from the standard hypothesis often
1144: associated with Kolmogorov (K41) theory \cite{Kolmogorov41},
1145: and can possibly explain departures from self-similar
1146: models of the higher order structure functions and
1147: controversial results observed in experiments, such as
1148: the slower than predicted recovery of isotropy in the small scales
1149: \cite{Shen00}.
1150:
1151: Similar results were obtained for the helicity transfer. The
1152: injection of net helicity in the large scales breaks down the mirror
1153: symmetry in the flow. While confirming the direct cascade of helicity
1154: \cite{Brissaud73,Andre77,Borue97,Chen03,Chen03b,Gomez04},
1155: we also found a slower than
1156: expected recovery of the symmetries in the small scales, with an
1157: excess of relative helicity at small scales compared with the $k^{-1}$
1158: expected drop. As in the energy cascade, the cascade of helicity takes
1159: place in fixed steps proportional to the forcing scale, indicating
1160: strong nonlocal triadic interactions.
1161:
1162: Nonlocal interactions can also be responsible for observed
1163: departures from universality. In \cite{Alexakis05b} it was shown with
1164: a simple dimensional argument how nonlocal interactions can still
1165: be consistent with a $k^{-5/3}$ energy spectrum. Vortex tube stretching
1166: by the large scale flow plays a significant role, and the argument
1167: has points in common with multifractal models of intermittency, such
1168: as the $\beta$ model (see e.g. \cite{Frisch}). As previously mentioned,
1169: in this work we showed evidence of a positive correlation of large scale
1170: shear and small scale gradients, both in pdfs of velocity increments and
1171: in structure functions. Ref. \cite{Laval01} showed that the anomalous
1172: scaling of the structure functions is reduced when nonlocal interactions
1173: with the large scale flow are artificially suppressed in a simulation.
1174: Both results suggest that departures from K41 theory can be associated with
1175: the imprint of the large scale forcing in turbulence.
1176:
1177: The modern language used in turbulence is related to a great extent
1178: to the K41 theory developed for the isotropic and homogeneous case, with
1179: the assumption that such fundamental symmetries of the equations would
1180: be recovered in the small scales even when broken (e.g. through external
1181: forcing) in the large scales. To apply this theory to real turbulent
1182: flows, the idea that small
1183: scales restore isotropy and homogeneity is based strongly on the assumption
1184: of local interactions between scales. The persistence of
1185: anisotropies and other deviations from K41 observed in experiments and
1186: simulations have been associated in this work with the presence of strong
1187: non-local triadic interactions. At this point, the reader could ask
1188: how much of the edifice of homogeneous and isotropic turbulence remains.
1189:
1190: One of the assumptions in K41 theory is that the properties of the
1191: inertial range are universal. This is directly related to the assumption
1192: of local interactions. Since eddies in the inertial range only interact
1193: with eddies of similar size, as the energy cascades through the inertial
1194: range a self-similar solution is obtained. The fraction of the flux due to
1195: non-local interactions, found to be $\sim 20\%$ here, can
1196: be interpreted as a subleading contribution to the local flux.
1197: Deviations from K41 theory are well documented and several theories have been
1198: proposed to explain them (see e.g.
1199: \cite{Kolmogorov62,She94,Frisch,Lesieur,Tsinober}).
1200: They have also been derived theoretically for the passive scalar
1201: in the context of the so-called Kraichnan model \cite{Kraichnan68}.
1202: Several of these corrections were associated with viscous effects for
1203: turbulence at finite Reynolds number. What our results show is that
1204: these deviations can come as well from interactions with the large
1205: scale flow. In this context, simulations at higher resolution could help
1206: to study the scaling of the non-local flux with the Reynolds number.
1207:
1208: It is known that in the Kraichnan model only prefactors of the scaling
1209: laws depend on the large-scale forcing (i.e. the exponents are universal:
1210: independent of the forcing). However, in the case of the Navier-Stokes
1211: equations, this has not been proved yet since the hierarchy of equations
1212: for the $n$-point correlation functions is not closed. Then, inertial
1213: range solutions with exponents that depend on the forcing
1214: may exist.
1215:
1216: In the case of realistic turbulent flows as encountered in astrophysics
1217: and geophysics, the presence of strong non-local
1218: interactions can lead to the persistence of anisotropies in the small
1219: scales. In this case, the anisotropic effects question the applicability
1220: of the theory of isotropic and homogeneous turbulence to real flows (at
1221: least at Reynolds numbers comparable to the ones studied in this work).
1222: The properties of the large scale flow can be more important than
1223: expected to shape the small scales. A systematic study of the isotropic
1224: and anisotropic contributions to the scaling
1225: \cite{Biferale01a,Biferale01b,Biferale02} can be a first step to recognize
1226: universal (and non-universal) features in these cases.
1227:
1228: Finally, as previously indicated in \cite{Alexakis05b}, the existence of
1229: non-negligible nonlocal interactions in a variety of turbulent flows
1230: gives support to models involving as an essential agent of the nonlinear
1231: energy transfer the distortion of turbulent eddies by a large-scale
1232: flow (e.g. as in rapid distortion theory and its variants
1233: \cite{Laval01,Dubrulle04}, or as in the Lagrangian averaged Navier-Stokes
1234: equations \cite{Holm98,Chen98} where the turbulent flow interacts with a
1235: smooth velocity field). We believe that the understanding of scale
1236: interactions
1237: in turbulence can lead to the development of a new generation of subgrid
1238: models, beyond the usual hypothesis of locality done in most Large Eddy
1239: Simulations.
1240:
1241: \begin{acknowledgments}
1242: The authors would like to express their gratitude to J.R. Herring for
1243: valuable discussions and his careful reading of the manuscript. The
1244: authors also acknowledge discussions with L. Biferale and S. Kurien.
1245: Computer time was provided by NCAR and by the National Science Foundation
1246: Terascale Computing System at the Pittsburgh Supercomputing Center. The
1247: NSF grant CMG-0327888 at NCAR supported this work in part and is gratefully
1248: acknowledged. Three-dimensional visualizations of the flows were done
1249: using VAPoR, a software for interactive visualization and analysis of
1250: terascale datasets \cite{vapor}. The authors are grateful to A. Norton
1251: and J. Clyne (SCD/CISL) for help with the visualizations.
1252: \end{acknowledgments}
1253:
1254: \bibliography{ms}
1255:
1256: \end{document}
1257: