1: %\documentclass[preprint,showtitle,amsmath,amssymb]{revtex4}
2: \documentclass[aps,twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
3: %\documentclass[pre,preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
4: \usepackage{graphicx}
5: \usepackage{dcolumn}
6: \usepackage{color}
7: %\bibliographystyle{abbrv}
8:
9: \def\NOTE#1{{\textcolor{red}{\bf [#1]}}} % note
10: \def\ADD#1{{\textcolor{blue}{#1}}} % question
11:
12: \newcommand{\beq}{\begin{equation}}
13: \newcommand{\eeq}{\end{equation}}
14: %
15: \topmargin -50pt
16:
17: \begin{document}
18:
19: \title{\ADD{}Energy and enstrophy dissipation in steady state 2-d turbulence}
20:
21: \author{Alexandros Alexakis}
22: \affiliation{National Center for Atmospheric Research\\
23: P.O. Box 3000, Boulder, CO 80307-3000, U.S.A. }
24: \author{Charles R. Doering}
25: \affiliation{Department of Mathematics and Michigan Center for Theoretical Physics\\
26: University of Michigan, Ann Arbor, MI 48109-1109, U.S.A.
27: }
28:
29:
30: \date{\today}
31: %%%%%%%%%%
32:
33:
34: \begin{abstract}
35: Upper bounds on the bulk energy dissipation rate $\epsilon$ and enstrophy dissipation rate
36: $\chi$ are derived for the statistical steady state of body forced two
37: dimensional turbulence in a periodic domain.
38: For a broad class of externally imposed body forces it is shown that
39: $\epsilon \le k_{f} U^3 Re^{-1/2}\left(C_1+C_2 Re^{-1}\right)^{1/2}$ and
40: $\chi \le k_{f}^{3}U^3 \left(C_1+C_2 Re^{-1}\right)$
41: where $U$ is the root-mean-square velocity, $k_f$ is a wavenumber (inverse length scale)
42: related with the forcing function, and $Re = U /\nu k_f$.
43: The positive coefficients $C_1$ and $C_2$ are uniform in the the kinematic viscosity $\nu$,
44: the amplitude of the driving force, and the system size.
45: We compare these results with previously obtained bounds for
46: body forces involving only a single length scale, or for
47: velocity dependent a constant-energy-flux forces acting at finite wavenumbers.
48: %we establish even shaper estimates of the form
49: %$\epsilon \lesssim k_{f}U^3 Re^{-1}$ and
50: %$\chi \lesssim k_{f}^{3}U^3Re^{-1}$.
51: Implications of our results are discussed.
52: \end{abstract}
53: %\pacs{.}
54: \maketitle
55:
56:
57: \section{Introduction}
58:
59: The study of two dimensional turbulence was originally justified as a simplified version of
60: the more complex three dimensional turbulence, but it has come to be regarded as an interesting
61: research field in its own right with deep connections to geophysical and astrophysical problems
62: such as as strongly rotating stratified flows \cite{Tabeling02}.
63: A large number of experimental methods have been devised to constrain flows in two dimensions
64: (e.g. soap films) allowing some theories theories to be tested in the lab \cite{Kellay02}.
65: Direct numerical simulations are far easier than the three dimensional case, and this has enabled
66: researchers to investigate two dimensional turbulence computationally at much higher
67: Reynolds numbers \cite{Pouquet75,Boffeta00,Tran04,Chen03,Danilov05,Dmitruk05}.
68: As a result, there are more simulation data to test theories of two dimensional turbulence.
69: Nevertheless many fundamental questions remain open;
70: see \cite{Tabeling02} for a recent review.
71:
72: The inviscid conservation of enstrophy as well as energy in two dimensions results in two
73: cascading quadratic invariants that make the phenomenology of two dimensional turbulence
74: somewhat more complex than three dimensional turbulence
75: and not derivable from simple dimensional arguments.
76: Theoretical studies of turbulence usually employ a statistical description and often
77: involve assumptions about homogeneity isotropy and the nature of the interactions.
78: Based on physical arguments, Kraichnan \cite{Kraichnan67}, Leith \cite{Leith68} and
79: Batchelor \cite{Batchelor69} conjectured that there is a dual cascade in two dimensional
80: turbulence:
81: energy flows to the larger scales while enstrophy moves to the small scales
82: (when the system is driven at some intermediate scale).
83: Kraichnan-Leith-Batchelor (KLB) theory assumes isotropy and homogeneity in the limit of
84: infinite domain size, and in the zero viscosity limit predicts a $k^{-5/3}$ energy spectrum
85: for the large scales and a $k^{-3}$ energy spectrum for the small scales.
86: The assumptions of the KLB theory, as well as the power index and the universality of the two
87: energy spectra has been questioned in the literature \cite{Kraichnan71,Saffman71,Sulem71,
88: Bowman96,Tran02,Gkioulekas05a,Gkioulekas05b}.
89:
90: In this paper we derive some simple rigorous bounds for the long time averaged bulk energy
91: and enstrophy dissipation rates for two dimensional statistically stationary flows
92: sustained by a variety of driving forces. The study of physically relevant rigorous
93: bounds on the energy dissipation rate, i.e.,
94: the power consumption of turbulence, for a class of boundary-driven flows can be traced
95: back to the seminal work of Howard \cite{Howard63},
96: and in recent years an alternative approach \cite{Doering92} has renewed interest in those
97: kinds of problems, providing direct connections to experiments in some cases.
98: Bounds for the energy dissipation of steady body forced flows---more convenient for
99: theoretical and numerical investigations---in three dimensions
100: have been derived by Foias \cite{Foias97} and others
101: \cite{Doering02,DoeringES03,Doering05,Petrov05}.
102: Not unexpectedly, Foias et al. \cite{Foias02} also derived a bound for the
103: enstrophy dissipation rate in the statistically stationary states of
104: two-dimensional turbulence driven by a restricted class of forces.
105: Bounds for the energy and enstrophy dissipation in two dimensional flow driven by
106: a monochromatic forcing were derived in \cite{Constantin94} and \cite{Tran02}.
107: The case of temporally white noise forcing was studied by Eyink \cite{Eyink96}.
108: More recently Tran {\it el at} \cite{Tran05,Tran06} derived bounds of the
109: enstrophy dissipation for freely decaying two dimensional turbulence
110: in terms of the initial ideal invariants of the flow.
111: (See also \cite{Eyink01} for the treatment of the same problem in terms of
112: the inviscid Euler Equations).
113: Finally, we mention that bounds on the dimension
114: of attractor of the 2D Navier-Stokes have been derived in
115: \cite{Constantin85, Constantin88, Doering91, Doering98, Tran04b}
116: and more recently by \cite{Gibbon06}.
117:
118: The results for the energy and enstrophy dissipation of forced flows derived in this paper
119: apply to a more general type of forcing than the single scale forcing
120: that was considered in \cite{Constantin94,Tran02}.
121: We also consider forces that are smoothly varying in time, unlike the white noise forcing
122: investigated in \cite{Eyink96}, and we are particularly interested in the behavior of
123: the long times averaged dissipation rates in the vanishing viscosity limit.
124: Because the viscosity is a dimensional quantity we must specify what we mean by
125: ``small'' viscosity.
126: To be precise, we measure the magnitude of the viscosity in terms of the Reynolds
127: number in the statistically steady state,
128: \beq
129: Re = \frac{U}{k_{f}\nu}
130: \label{Re}
131: \eeq
132: where $U$ is the root mean square velocity and $k_{f}$ is a natural wavenumber
133: (inverse length scale) in the driving force.
134: The dissipation rates are also dimensional quantities, so we measure them in terms of the
135: inviscid scales determined by $U$ and $k_{f}$.
136: That is, we estimate
137: \beq
138: \beta = \frac{\epsilon }{k_{f}U^{3}}
139: \quad \quad \text{and} \quad \quad
140: \gamma = \frac{\chi}{k_{f}^{3}U^{3}}
141: \eeq
142: in terms of $Re$ and focus on the $Re \rightarrow \infty$ limit holding other parameters
143: (such as the large scale eddy turnover time in the most general case) fixed.
144: For a broad class of external driving we find that
145: \beq
146: \beta \lesssim Re^{-1/2}
147: \quad \quad \text{and} \quad \quad
148: \gamma \lesssim Re^{0},
149: \label{gsum}
150: \eeq
151: consistent with an enstrophy cascade of sorts.
152:
153: However, for special cases of forcing such as ``ultra narrow band'' monochromatic
154: (i.e., involving on a single length scale, albeit with a broad range of time dependence)
155: forcing, or for a fixed energy flux forcing popular for
156: direct numerical simulations, a stronger bound holds:
157: \beq
158: \beta \lesssim Re^{-1}
159: \quad \quad \text{and} \quad \quad
160: \gamma \lesssim Re^{-1}.
161: \label{msum}
162: \eeq
163: This kind of $Re^{-1}$ scaling suggests ``laminar'' flows where the energy is
164: concentrated at or above the smallest length scale of the forcing.
165: This sort of scaling has been shown before in the literature \cite{Constantin94,Tran02}
166: for monochromatic forcing and for white noise in time forcing \cite{Eyink96}.
167:
168: In every case the bounds derived here are strictly less than those
169: available --or expected-- for three dimensional turbulence.
170: The upper bounds (\ref{gsum}) on the energy and enstrophy dissipation for two dimensional
171: flows derived here are in a sense a consequence of combining the approaches in
172: \cite{Foias02} and \cite{Eyink96,Tran02} applied to a class of forcing
173: functions concentrated in a finite range of length scales.
174: Even though some steps in our analysis have been taken before, in order to make this paper
175: self-contained the complete (but nevertheless short) proofs will be presented here.
176:
177: The rest of this paper is organized as follows.
178: In the next section II we introduce the problem and basic definitions, and perform the analysis
179: leading to (\ref{gsum}) for the simplest case of time independent body forces.
180: Section III generalizes the analysis to include a broad class of time-dependent forces.
181: In section IV we briefly review the results for
182: time-dependent but single-length scale forcing
183: and ``fixed-flux" forces in order to establish the stronger results in (\ref{msum}).
184: The concluding section V contains a brief discussion of the results and their implications.
185:
186: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
187:
188: \section{Time-independent forcing}
189:
190: Consider a two dimensional periodic domain $[0,L]^2$, i.e., $\mathbb{T}^2_L$, filled with an
191: incompressible fluid of unit density evolving according to the Navier-Stokes equation:
192: \beq
193: \partial_t {\bf u} +{\bf u \cdot \nabla u} =-\nabla p +\nu \nabla^2{\bf u} +{\bf f}
194: \label{NS}
195: \eeq
196: where ${\bf u} = {\bf \hat{i}} u_x(x,y,t) + {\bf \hat{j}} u_y(x,y,t)$ is the
197: incompressible (divergence-free) velocity field,
198: $p(x,y,t)$ is the pressure, $\nu$ is the viscosity, and
199: ${\bf f(x,y)}= {\bf \hat{i}} f_x(x,y) + {\bf \hat{j}} f_y(x,y)$ is a smooth, mean zero,
200: divergence-free body force with characteristic length scale $\sim k_{f}$
201: (defined precisely below).
202: The scalar vorticity $\omega = \partial_x u_y -\partial_y u_x$ satisfies
203: \beq
204: \partial_t \omega +{\bf u \cdot \nabla}\omega = \nu \nabla^2 \omega +\phi
205: \label{VE}
206: \eeq
207: where $\phi = {\bf \hat{k} \cdot}\nabla \times {\bf f} = \partial_x f_y -\partial_y f_x$.
208:
209: The Reynolds number is defined in (\ref{Re}) where
210: \beq
211: U \equiv \langle |{\bf u}|^{2} \rangle^{1/2}
212: \eeq
213: is the root-mean-square velocity with $\langle \cdot \rangle$
214: representing the space-time average
215: \beq
216: \langle {\bf g} \rangle = \lim_{T\to\infty} \frac{1}{T}\int_0^T
217: \left(\frac{1}{L^2}\int_{\mathbb{T}^2_L} {\bf g}(x,y,t) \, d^{2}x \right) \, dt.
218: \eeq
219: (The limit in the time average is assumed to exist for all the quantities of interest.)
220: The forcing length scale associated with the wavenumber $k_{f}$ is defined by
221: \beq
222: k_f^{2} \equiv \frac{\|\nabla^2 {\bf f}\|} {\|{\bf f}\|}
223: \eeq
224: where $\| \cdot \|$ is the $L_2$ norm on $\mathbb{T}^2_L$.
225: It is apparent that we are restricting ourselves to sufficiently smooth forcing functions.
226:
227: The time and space averaged energy dissipation rate is
228: \beq
229: \epsilon \equiv \nu \left \langle | {\bf \nabla u}|^2 \right \rangle
230: = \nu \left \langle \omega ^2 \right \rangle,
231: \label{energy_diss}
232: \eeq
233: the second expression resulting from integrations by parts utilizing incompressibility.
234: %A conventional non-dimensional measure of the energy dissipation, called the
235: %{\it friction factor} in appropriate contexts, is
236: %\beq
237: %\beta = \frac{\epsilon} {k_{f}U^3}.
238: %\eeq
239: The bulk averaged enstrophy dissipation rate is
240: \beq
241: \chi \equiv \nu \langle |\nabla \omega| ^2 \rangle
242: = \nu \langle | {\bf \nabla^{2} u}|^2 \rangle.
243: \label{enstrophy_diss}
244: \eeq
245: %and we define it's dimensionless counterpart
246: %\beq
247: %\gamma = \frac{\chi} {k_{f}^{3}U^3}.
248: %\eeq
249:
250: We think of $\beta = \epsilon/k_{f}U^3$ and $\gamma = \chi/k_{f}^{3}U^3$ as functions
251: of $Re$ and the functional form or ``shape'' of the forcing, but not explicitly on its amplitude
252: \beq
253: F = \frac{\|{\bf f}\|}{L}
254: \eeq
255: except indirectly through its influence on $U$.
256:
257: We are considering the Reynolds number to be the ``control parameter'' even though
258: it is defined in terms of the emergent quantity $U$.
259: Strictly speaking the flow is determined by the structure and amplitude of the body force
260: (and the initial data) so the Grashof number such as $Gr = F/k_{f}^{3}\nu^2$
261: should naturally be used as the relevant dimensionless control parameter
262: indicating the intensity of the driving and the resulting flow.
263: Indeed, while we can always realize any given value of $Gr$, it is not at all evident that every
264: particular value of $Re$ can be achieved.
265: Nevertheless, in order to express the results in terms of quantities
266: familiar to the theory of homogeneous isotropic
267: turbulence we will (without loss of rigor) express the bounds in terms of $Re$.
268:
269: Poincare's inequality applied to (\ref{energy_diss}) and (\ref{enstrophy_diss})
270: immediately yield the lower estimates
271: \beq
272: \epsilon \ge \nu \frac{4\pi^{2}}{L^{2}} U^{2}
273: \quad \quad \text{and} \quad \quad
274: \chi \ge \nu \frac{16\pi^{4}}{L^{4}} U^{2}
275: \eeq
276: so that
277: \beq
278: \beta \ge 4 \pi^{2} \alpha^{2} Re^{-1}
279: \quad \quad \text{and} \quad \quad
280: \gamma \ge 16\pi^{4} \alpha^{4} Re^{-1}
281: \eeq
282: where $\alpha=(k_{f}L)^{-1} \le (2\pi)^{-1}$ is the ratio of the forcing to domain length scales.
283: If $\beta$ and $\gamma$ scale both as $\sim Re^{-1}$
284: then we say that the flow exhibits laminar behavior
285: because the energy is then necessarily concentrated at
286: relatively long length scales determined by the
287: prefactor, rather than over a broad range of scales that increases as $Re \rightarrow \infty$.
288:
289: On the other hand if $\beta \sim Re^{0}$, the scaling expected in three dimensional turbulence,
290: the flow exhibits finite (residual) dissipation
291: in the limit of zero viscosity indicating the
292: presence of an active and effective energy cascade to small scales.
293: It was recently shown \cite{Doering02, Doering05} that $\beta \le c Re^{0}$ for the
294: vanishing viscosity limit of three dimensional versions of the systems under consideration here
295: and in section III, where the coefficient $c$ is uniform in $\nu$, $L$, and $F$.
296: There is, however, no known {\it a priori} enstrophy dissipation rate bound for the three dimensional
297: turbulence; this is related to the outstanding open question of the regularity of
298: solution for the three dimensional Navier-Stokes equations.
299: As the results of this paper suggest quantitatively, the dissipation rates of
300: two dimensional turbulence falls somewhere between laminar
301: scalings and the rates for three dimensional turbulence.
302:
303: To prove the two dimensional bounds we first take the inner product of the vorticity equation
304: (\ref{VE}) with $\omega$ and average to obtain the enstrophy production-dissipation balance
305: \beq
306: \chi = \left\langle \omega \phi \right\rangle
307: \label{EPDB}
308: \eeq
309: where the time derivative term drops out when we take the long time average.
310: Integrating by parts to move the $\hat{\bf k}\cdot\nabla\times$ from $\omega$
311: onto $\phi$ and the Cauchy-Schwarz inequality, we easily obtain
312: \beq
313: \chi \le k_{f}^{2}UF.
314: \label{VBI}
315: \eeq
316:
317: For the second step, consider a smooth incompressible vector field ${\bf v}(x,y)$.
318: Take the inner product of $\bf v$ with the Navier-Stokes equation, integrate by parts
319: and average to obtain
320: \begin{eqnarray}
321: \frac{1}{L^{2}} \int_{\mathbb{T}^2_L} {\bf v\cdot f } d^{2}x =
322: - \left\langle {\bf u} \cdot (\nabla{\bf v}) \cdot{\bf u} +
323: \nu {\bf u} \cdot \nabla^2{\bf v} \right\rangle.
324: \end{eqnarray}
325: Using the Cauchy-Schwarz and H\"older's inequality (as in \cite{Doering02}) we deduce
326: \begin{equation}
327: F \times \frac{1}{L^2\|{\bf f}\|} \int_{\mathbb{T}^2_L} {\bf v\cdot f } \, d^{2}x \le
328: U^2 \| \nabla {\bf v}\|_\infty +
329: \nu \frac{U}{L} \|\nabla^2{\bf v}\|
330: \label{F1a}
331: \end{equation}
332: where $\|\cdot\|_\infty$ is the $L_\infty$ norm on $\mathbb{T}^2_L$.
333: In order for the inequality to be non-trivial we need to restrict $\bf v$ so that
334: $\int_{\mathbb{T}^2_L} {\bf v\cdot f} \, d^{2}x >0$.
335: This is easy to arrange.
336: For example the choice ${\bf v = f}/F$ will satisfy this condition
337: if ${\bf f}$ is sufficiently smooth
338: that the right hand side of (\ref{F1a}) is finite.
339: If it is not so smooth, then for instance we can take
340: ${\bf v} \sim K \circ {\bf f}$ where $K(x,y,x',y')$
341: is a (positive) smoothing kernel.
342: In any case we can choose ${\bf v}$ appropriately and use (\ref{F1a})
343: to eliminate $F$ in (\ref{VBI})
344: so that:
345: \beq
346: \chi \le U^3k_{f}^{3} \, \left(C_{1} +\frac{C_{2}}{Re}\right)
347: \quad \quad \Rightarrow \quad \quad \gamma \le \left(C_{1} +\frac{C_{2}}{Re}\right)
348: \label{Xbound}
349: \eeq
350: where the dimensionless coefficients $C_{1}$ and $C_{2}$
351: are independent of $k_f$ and $L$, depending only on the functional ``shape" of $\bf v$
352: (and thus also on the shape of $\bf f$) but not on its amplitude F or the viscosity $\nu$.
353: Explicitly they are
354: \begin{equation}
355: C_{1} = \frac {\| \nabla_l {\bf v}\|_\infty}{\langle {\bf v\cdot f}/F \rangle}
356: \quad \quad \mathrm{and} \quad \quad
357: C_{2} = \frac{\langle |\nabla_{l}^2{\bf v} |\rangle^{1/2}}{\langle {\bf v\cdot f}/F \rangle}
358: \end{equation}
359: where $\nabla_{l}$ is the gradient with respect to the non-dimensional coordinate $k_{f}{\bf x}$.
360: An upper bound for the enstrophy dissipation rate like that in (\ref{Xbound})
361: was first derived in \cite{Foias02}.
362: Note that for strictly band-limited forces, i.e., if the Fourier transform of the force is supported on
363: wavenumbers with $|{\bf k}| \in (k_{min}, k_{max})$ with $0 < k_{min} < k_{max} < \infty$,
364: then the coefficients $C_{1}$ and $C_{2}$ are bounded by
365: pure numbers depending only on $k_{max}/k_{min}$.
366:
367: For the final step of the proof we use integrations by parts and the Cauchy-Schwarz inequality
368: to see that
369: \beq
370: \langle{\omega^{2}}\rangle^{2} =
371: \left\langle {\bf u \cdot \nabla} \times (\hat{{\bf k}} \omega) \right\rangle^{2}
372: \le \langle |{\bf u}|^2\rangle \langle |{\bf\nabla \omega}|^2 \rangle .
373: \label{trickI}
374: \eeq
375: %Eyink \cite{Eyink96} used precisely this fact in the context of white-in-time noise forcing.
376: Combining (\ref{trickI}) with (\ref{Xbound}) we deduce
377: \beq
378: \langle{\omega^{2}}\rangle^{2} \le
379: \frac{k_{f} U^5}{\nu} \left(C_{1} +\frac{C_{2}}{Re} \right),
380: \eeq
381: and in terms of the energy dissipation rate this is the announced result
382: \beq
383: \beta \le Re^{-1/2} \left(C_{1} +\frac{C_{2}}{Re} \right)^{1/2}.
384: \label{bound}
385: \eeq
386:
387: \section{Time-dependent forces}
388:
389: Now consider the Navier-Stokes equation (\ref{NS}) where the time dependent body force
390: ${\bf f}(x,y,t)$ is smooth and incompressible with characteristic
391: length scale $\sim k_{f}^{-1}$ given by
392: \beq
393: k_f^{4} \equiv \frac{\langle |\nabla^2 {\bf f}|^2\rangle}{\langle |{\bf f}|^2 \rangle},
394: \eeq
395: and time scale $\Omega_f^{-1}$ defined by
396: \beq
397: \Omega_f^{2} \equiv \frac{\langle |\partial_t {\bf f}|^2 \rangle}
398: {\langle |{\bf f}|^2 \rangle}.
399: \eeq
400: We define
401: \beq
402: \tau = \frac{\Omega_f}{k_{f}U},
403: \eeq
404: the ratio of the ``eddy turnover'' time $(k_{f}U)^{-1}$ to the forcing time scale $\Omega_f^{-1}$.
405: In this time-dependent setting the amplitude $F$ of the force is
406: \beq
407: F = \langle |{\bf f}|^{2}\rangle^{1/2}.
408: \eeq
409:
410: As before, the space and time average of $\omega$ times the vorticity equation (\ref{VE})
411: yields the enstrophy balance equation (\ref{EPDB}), and integration by parts
412: and the Cauchy-Schwarz inequality implies
413: \beq
414: \chi\le k_{f}^{2}UF.
415: \label{VB}
416: \eeq
417:
418: For the second step here we introduce a smooth incompressible vector field
419: ${\bf v}(x,y,t)$ and take space and time average of the inner product of
420: with the Navier-Stokes equation to obtain
421: %\beq
422: %\langle {\bf v\cdot f}\rangle=
423: %-\frac{\langle{\bf u \partial_{t} v}\rangle}{\tau}
424: %-\frac{\langle {\bf u (u \cdot \nabla) v} \rangle}{l_f}
425: %-\frac{\nu \langle {\bf w} \nabla^2{\bf v}\rangle}{l_f^2}
426: %\eeq
427: \begin{eqnarray}
428: \left\langle {\bf v \cdot f} \right\rangle =
429: - \left\langle {\bf u} \cdot \partial_{t} {\bf v} +
430: {\bf u \cdot (\nabla v) \cdot u} + \nu {\bf u} \cdot \nabla^2{\bf v} \right\rangle.
431: \end{eqnarray}
432: Cauchy-Schwarz and H\"older's inequalities then imply
433: \begin{eqnarray}
434: F \frac{ \langle{\bf v \cdot f} \rangle}{\langle |{\bf f}|^{2} \rangle^{1/2}} &\le&
435: U\langle |{\bf \partial_t v}|^2 \rangle^{1/2} \ + \
436: U^2 \sup_t \| \nabla {\bf v}\|_\infty \nonumber \\
437: &+&
438: \nu U \langle |\nabla^2{\bf v}|^2 \rangle^{1/2}.
439: \label{F1}
440: \end{eqnarray}
441: Now we need to be able to choose $\bf v$ satisfying $\langle{\bf v \cdot f} \rangle > 0$
442: such that all the coefficients on the right hand side are all finite.
443: Our ability to do this depends on details of ${\bf f}(x,y,t)$.
444:
445: For example if ${\bf f}$ is sufficiently smooth in space and appropriately uniformly
446: bounded in time then we can choose ${\bf v} \sim {\bf f}$.
447: We could also choose ${\bf v}$ as an appropriately filtered version of
448: ${\bf f}$ to cover more general cases.
449: For the purposes of this study and to display the results in the clearest
450: (if not the sharpest or most general form)
451: we will simply presume that ${\bf f}$ is sufficiently regular
452: that we can take ${\bf v} = {\bf f}$.
453: In that case (\ref{F1}) becomes
454: \beq
455: F \le \Omega_{f}U +
456: U^2 \, \frac{\sup_t \| \nabla {\bf f}\|_\infty}{F} +
457: \nu k_{f}^2U.
458: \label{F2}
459: \eeq
460: Then using this to eliminate $F$ from (\ref{VB}) we have
461: \beq
462: \chi \le k_{f}^{3}U^3 \left( \tau + C_{3} +\frac{1}{Re} \right)
463: \ \ \Rightarrow \ \ \gamma \le
464: \left( \tau + C_{3} +\frac{1}{Re} \right)
465: \label{VboundT}
466: \eeq
467: where the coefficient $C_{3}$ is
468: \beq
469: C_{3} = \frac{\sup_t \| \nabla_{l} {\bf f}\|_\infty}{F}
470: \eeq
471: with $\nabla_{l}$ denoting the gradient with respect to the non-dimensional coordinate
472: $k_{f}{\bf x}$.
473: The dimensionless number $C_{3}$ is independent of the scales of $F$, $k_f$, $L$, etc.,
474: depending only on the ``shape" of $\bf f$.
475: For example if $\bf f$ is quasi-periodic with $N$ frequencies and involves only wavenumbers
476: ${\bf k}$ with $0 < k_{min} < |{\bf k}| < k_{max} < \infty$,
477: then $C_{3}$ is bounded by $\sqrt{N}$ times a function of $k_{max}/k_{min}$.
478:
479: The final step again uses the inequality
480: \beq
481: \epsilon^{2} = \nu^{2} \langle{\omega^2}\rangle^{2} =
482: \nu^{2}\langle {\bf u} \cdot \nabla \times (\hat{{\bf k}} \omega) \rangle^{2}
483: \le \nu U^{2} \chi
484: \label{trick2}
485: \eeq
486: and it then follows immediately from (\ref{VboundT}) that
487: \beq
488: \beta \le Re^{-1/2} \left(\tau + C_{3} +\frac{1}{Re} \right)^{1/2}.
489: \label{boundT}
490: \eeq
491: Note in this case $\tau$ depends on $U$ and features of the forcing through
492: $k_{f}$ and $\Omega_{f}$, but {\it not} on $\nu$.
493:
494:
495: \section{Monochromatic and constant flux forces}
496:
497: An even sharper scaling bound on the energy and enstrophy dissipation rates can be derived when
498: the driving is monochromatic in space, whether it is steady or time dependent
499: \cite{Constantin94,Tran02}.
500: Suppose the body force involves only a single length scale, i.e.,
501: \beq
502: -\nabla^{2} {\bf f} =k_{f}^{2} {\bf f}.
503: \eeq
504: This does not preclude complex time-dependence for ${\bf f}(x,y,t)$, just that it
505: involves only spatial modes with wavenumbers ${\bf k}$ with $|{\bf k}| = k_{f}$.
506: Then the enstrophy production-dissipation balance (\ref{EPDB}) implies
507: \beq
508: \chi = \langle \omega \phi \rangle =
509: \langle {\bf u} \cdot (-\nabla^{2}{\bf f}) \rangle =
510: k_{f}^{2}\langle {\bf u} \cdot {\bf f} \rangle = k_{f}^{2} \epsilon.
511: \label{EPDBmono}
512: \eeq
513: Combining this with (\ref{trick2}), we observe that
514: \beq
515: \epsilon^{2} \le \nu U^{2} \chi = \nu k_{f}^{2}U^{2}\epsilon
516: %\label{trick3}
517: \eeq
518: so that
519: \beq
520: \epsilon \le \nu k_{f}^{2} U^{2}
521: \quad \quad \text{and} \quad \quad
522: \chi \le \nu k_{f}^{4} U^{2}
523: \label{trick3}
524: \eeq
525: implying that both $\beta$ and $\gamma$ are bounded by $Re^{-1}$.
526: Note that this kind of monochromatic forcing is a special case that
527: has been shown in the literature for some cases
528: to lead to a laminar flow that never looses stability
529: \cite{Marchioro86}.
530:
531: An other type of forcing that results in this scaling is
532: \beq
533: {\bf f}(x,y,t) = \epsilon \frac{{\cal P}{\bf u}}{L^{-2}\|{\cal P}{\bf u} \|^{2}}
534: \eeq
535: where ${\cal P}$ is the projector onto spatial modes of wavenumber
536: ${\bf k}$ with $|{\bf k}| \in [k_{min}, k_{max}]$,
537: and the coefficient $\epsilon$ is now the control parameter.
538: This type of forcing is often applied in
539: numerical simulations of homogeneous isotropic turbulence.
540: With this forcing in the Navier-Stokes equations constitutes an autonomous dynamical system
541: with kinetic energy injected at a constant rate $\epsilon$ at wavenumbers with
542: $|{\bf k}| \in [k_{min}, k_{max}]$.
543: The rms speed $U$ (i.e., the Reynolds number) and the enstrophy dissipation rate $\chi$
544: are then emergent quantities, determined by the imposed energy flux $\epsilon$.
545: The mean power balance for solutions is still
546: \beq
547: \nu \langle |\nabla {\bf u}|^{2} \rangle =
548: \nu \langle \omega^{2} \rangle = \epsilon,
549: \eeq
550: and the enstrophy production-dissipation balance reads
551: \beq
552: \chi = \nu \langle |\nabla {\omega}|^2 \rangle
553: = \epsilon
554: \left\langle \frac{\|\nabla {\cal P}{\bf u}\|^{2}}{\|{\cal P}{\bf u} \|^{2}}
555: \right\rangle.
556: \label{EPDBff}
557: \eeq
558: Because the the forcing only involves wavenumbers in $[k_{min}, k_{max}]$
559: with positive energy injection at each wavenumber, at each instant of time
560: \beq
561: k_{min}^{2} \|{\cal P}{\bf u} \|^{2} \le \|\nabla {\cal P}{\bf u}\|^{2}
562: \le k_{max}^{2} \|{\cal P}{\bf u} \|^{2}.
563: \eeq
564: Then (\ref{EPDBff}) implies that
565: \beq
566: k_{min}^{2} \epsilon \le \chi \le k_{max}^{2} \epsilon.
567: \label{con}
568: \eeq
569: Using this with (\ref{trick2}) we see that
570: \beq
571: \epsilon^{2} \le \nu U^{2} \chi \le \nu U^{2} k_{max}^{2} \epsilon,
572: \eeq
573: and we conclude
574: \beq
575: \epsilon \le \nu k_{max}^{2} U^{2}
576: \quad \quad \text{and} \quad \quad
577: \chi \le \nu k_{max}^{4}U^{2}.
578: \eeq
579: Hence also in this case both $\beta$ and $\gamma$ are bounded $\sim Re^{-1}$.
580:
581: Note that in both these derivations a condition like (\ref{con}) or the
582: stronger condition (\ref{EPDBmono}) was used.
583: It is an open question whether such a condition holds for more general
584: and more ``realistic'' forcing functions.
585: The results (\ref{VboundT}) and (\ref{boundT}) give restrictions
586: on the energy and enstrophy dissipation rate for a broader
587: class of driving, but it is natural to wonder how
588: broad of a class of forcing functions
589: would actually result in the
590: $Re^{-1}$ scalings
591: in the vanishing
592: viscosity
593: limit.
594:
595: \section{Discussion}
596:
597: These quantitative bounds show that for two dimensional
598: turbulence sustained by forces as described
599: in the previous sections, there is no residual dissipation in the
600: vanishing viscosity limit defined
601: by $Re \rightarrow \infty$ at fixed $U$, $L$, $k_{f}$ and $\Omega_{f}$.
602: To be precise, $\epsilon$ vanishes at least as fast as $Re^{-1/2}$ in this limit.
603: This confirms that there is no forward energy cascade
604: in the steady state in the inviscid limit.
605: On the other hand the residual enstrophy dissipation allowed by (\ref{VboundT}) in this limit
606: does not rule out a forward enstrophy cascade.
607: This combination, $\epsilon \rightarrow 0$ with $\chi = {\cal O}(1)$ in the inviscid limit, is
608: consistent with the dual-cascade picture of two-dimensional turbulence developed by
609: Kraichnan \cite{Kraichnan67}, Leith \cite{Leith68} and Batchelor \cite{Batchelor69}.
610: Note however that the absence of forward energy cascade is not necessarily true for any
611: finite value of $Re$.
612: The $\beta\sim Re^{-1/2}$ scaling allowed by the bound is less severe than what a laminar flow
613: (or a flow with only inverse cascade of energy) would predict, and as a result
614: (\ref{bound}) does not exclude the presence of some direct subdominant cascade
615: of energy when the Reynolds number is finite as suggested by \cite{Gkioulekas05a}.
616:
617: On the other hand the direct cascade of enstrophy is necessarily absent
618: for some forcing functions (see \cite{Constantin94,Eyink96,Tran02}).
619: When the forcing acts at a single scale or constant power is injected in a
620: finite band of wavenumbers, both $\epsilon$ and $\chi$ vanish $\sim Re^{-1}$.
621: This suggests an essentially laminar behavior for these flows: if the energy spectrum
622: follows a power law $E(k)\sim k^{-\alpha}$ for large wavenumbers then the exponent
623: must be $\alpha \le -5$ for $\chi$ to vanish in the vanishing viscosity limit.
624: These results have been interpreted as absence of enstrophy cascade in finite domains.
625: However, both of these results rely on the condition (\ref{con})
626: which is not guaranteed for a general forcing functions.
627: Whether (\ref{con}) might hold for more general forcing functions is an open question;
628: the results (\ref{VboundT}) and (\ref{boundT}) give the restrictions on the
629: energy and enstrophy dissipation rate for a general forcing.
630: Note that the $\beta\sim Re^{-1/2}$ does not impose significant restriction on the
631: energy spectrum given the bound on $\chi$.
632: These considerations suggest that it is possible therefore that in two dimensional
633: turbulence the steady state energy spectrum depends on the type of forcing used,
634: even within the class of relatively narrow-band driving.
635: High resolution numerical simulations with forcing that
636: does not necessarily satisfy (\ref{con})
637: would be useful at this point to resolve this issue.
638:
639: We conclude by noting that an interesting question that follows from these results is that of
640: the $Re$-scaling of the energy dissipation in systems that almost have two dimensional
641: behavior like strongly rotating, strongly stratified or conducting
642: fluids in the presence of a strong magnetic field.
643: For example, is there a critical value of the rotation
644: such that the scaling of the energy dissipation
645: rate with the Reynolds number transitions from $\epsilon \sim Re^0$
646: to $\epsilon \sim Re^{-1/2}$?
647: These questions remain for future studies.
648:
649: \begin{acknowledgements}
650:
651: The authors thank J.D. Gibbon and M.S. Jolly for helpful discussions and comments.
652: CRD was supported in part by NSF grants PHY-0244859 and PHY-0555324
653: and an Alexander von Humboldt Research Award.
654: AA acknowledges support from the National Center for Atmospheric Research.
655: NCAR is supported by the National Science Foundation.
656: \end{acknowledgements}
657:
658: %\bibliographystyle{plain}
659: \bibliographystyle{unsrt}
660: \bibliography{ms}
661:
662:
663: \end{document}
664: