nlin0412024/SQG.tex
1: \documentclass{jfm}
2: %\documentclass[referee]{jfm}
3: \usepackage{graphicx,bm}
4: \def\dt{{\rm d}t}
5: \def\d{{\rm d}}
6: \def\u{{\bm u}}
7: \def\f{{\bm f}}
8: \def\dk{{\rm d}k}
9: \def\dx{{\rm d}{\bm x}}
10: \def\x{{\bm x}}
11: \def\k{{\bm k}}
12: \def\etal{{\it et al. }}
13: 
14: \title[Energy spectra in SQG turbulence]
15: {Large-scale energy spectra in surface quasi-geostrophic turbulence}
16: \author[C.V. Tran and J.C. Bowman]{C\ls H\ls U\ls O\ls N\ls G\ns V.\ns 
17: T\ls R\ls A\ls N\thanks{Present address: Mathematics Institute,
18: University of Warwick, Coventry CV4 7AL, UK}\ns \and J\ls O\ls H\ls 
19: N\ns C.\ns B\ls O\ls W\ls M\ls A\ls N\ns}
20: \affiliation{Department of Mathematical and Statistical Sciences,\\
21: University of Alberta, Edmonton, Alberta, Canada, T6G 2G1}
22: 
23: \date{12 May 2004 and in revised form 05 November 2004}
24: 
25: \begin{document}
26: 
27: \maketitle
28: 
29: \begin{abstract}
30: The large-scale energy spectrum in two-dimensional turbulence governed 
31: by the surface quasi-geostrophic (SQG) equation 
32: $$\partial_t(-\Delta)^{1/2}\psi+J(\psi,(-\Delta)^{1/2}\psi)
33: =\mu\Delta\psi+f$$
34: is studied. The nonlinear transfer of this system conserves the two 
35: quadratic quantities $\Psi_1=\langle[(-\Delta)^{1/4}\psi]^2\rangle/2$ 
36: and $\Psi_2=\langle[(-\Delta)^{1/2}\psi]^2\rangle/2$ (kinetic energy), 
37: where $\langle\cdot\rangle$ denotes a spatial average. The energy 
38: density $\Psi_2$ is bounded and its spectrum $\Psi_2(k)$ is shallower 
39: than $k^{-1}$ in the inverse-transfer range. For bounded turbulence, 
40: $\Psi_2(k)$ in the low-wavenumber region can be bounded by $Ck$ where 
41: $C$ is a constant independent of $k$ but dependent on the domain size. 
42: Results from numerical simulations confirming the theoretical 
43: predictions are presented. 
44: \end{abstract}
45: 
46: \section{Introduction}
47: 
48: The dynamics of a three-dimensional stratified rapidly rotating fluid 
49: is characterized by the geostrophic balance between the Coriolis force 
50: and pressure gradient. The nonlinear dynamics governed by the first-order
51: departure from this linear balance is known as quasi-geostrophic
52: dynamics and is inherently three-dimensional. The theory of 
53: quasi-geostrophy is interesting and the research performed on this subject 
54: constitutes a rich literature (see, for example, Charney 1948, 1971; 
55: Rhines 1979; Pedlosky 1987). This theory renders a variety of 
56: two-dimensional models that are appealing for their relative simplicity 
57: and yet sufficiently sophisticated to capture the underlying dynamics 
58: of geophysical fluids. One such model, the so-called surface 
59: quasi-geostrophic (SQG) equation, is the subject of the present study.
60: 
61: Quasi-geostrophic flows can be described in terms of the 
62: geostrophic streamfunction $\psi(\x,t)$. The 
63: vertical dimension $z$ is usually taken to be semi-infinite and the 
64: horizontal extent may be either bounded or unbounded. Normally, decay 
65: conditions are imposed as $z\rightarrow\infty$. At the flat surface 
66: boundary $z=0$, the vertical gradient of $\psi(\x,t)$ matches the 
67: temperature field $T(\x,t)$, 
68: i.e. $T(\x,t)|_{z=0}=\partial_z\psi(\x,t)|_{z=0}$. 
69: For flows with zero potential vorticity, this surface temperature 
70: field can be identified with $(-\Delta)^{1/2}\psi$, where $\Delta$ is 
71: the (horizontal) two-dimensional Laplacian. Here, the operator 
72: $(-\Delta)^{1/2}$ is defined by 
73: $(-\Delta)^{1/2}\widehat\psi(\k)=k\widehat\psi(\k)$, where $k=|\k|$ is 
74: the wavenumber and $\widehat\psi(\k)$ is the Fourier transform of 
75: $\psi(\x)$. The conservation equation 
76: governing the advection of the temperature $(-\Delta)^{1/2}\psi$ by 
77: the surface flow is (Blumen 1978; Pedlosky 1987; 
78: Pierrehumbert, Held \& Swanson 1994; Held \etal 1995) 
79: \begin{eqnarray}
80: \label{Tadvection}
81: \partial_t(-\Delta)^{1/2}\psi+J(\psi,(-\Delta)^{1/2}\psi)&=&0,
82: \end{eqnarray}
83: where $J(\varphi,\phi)=\partial_x\varphi\partial_y\phi
84: -\partial_y\varphi\partial_x\phi$.
85: This equation is known as the SQG equation.
86: 
87: In this paper a forced-dissipative version of (\ref{Tadvection}) 
88: is studied. A dissipative term of the form 
89: $\mu\Delta\psi$, where $\mu>0$, which results from Ekman pumping 
90: at the surface, is considered (Constantin 2002; Tran 2004). Since 
91: $(-\Delta)^{1/2}\psi$ is the advected quantity, this physical 
92: dissipation mechanism corresponds to the (hypoviscous) dissipation 
93: operator $\mu(-\Delta)^{1/2}$. The dissipation coefficient $\mu$ has 
94: the dimension of velocity and is not vanishingly small in the 
95: atmospheric context (Constantin 2002). The system is assumed 
96: to be driven by a forcing $f$, for which the spectral support is 
97: confined to wavenumbers $k\ge s>0$ (in bounded turbulence, wavenumber
98: zero is replaced by the minimum wavenumber). 
99: Thus, the forced-dissipative SQG equation can be written as 
100: \begin{eqnarray}
101: \label{governing}
102: \partial_t(-\Delta)^{1/2}\psi+J(\psi,(-\Delta)^{1/2}\psi)
103: &=&\mu\Delta\psi+f.
104: \end{eqnarray}
105: It is customary in the classical theory of turbulence to consider a
106: doubly periodic domain of size $L$; the unbounded case is obtained 
107: {\it via} the limit $L\rightarrow\infty$.
108: 
109: The Jacobian operator $J(\cdot,\cdot)$ admits the identities
110: \begin{eqnarray}
111: \label{id}
112: \langle\chi J(\varphi,\phi)\rangle=-\langle\varphi J(\chi,\phi)\rangle
113: =-\langle\phi J(\varphi,\chi)\rangle,
114: \end{eqnarray}
115: where $\langle\cdot\rangle$ denotes the spatial average. As a consequence, 
116: the nonlinear term in (\ref{governing}) obeys the conservation laws
117: \begin{eqnarray}
118: \label{conservation}
119: \langle\psi J(\psi,(-\Delta)^{1/2}\psi)\rangle=
120: \langle(-\Delta)^{1/2}\psi J(\psi,(-\Delta)^{1/2}\psi)\rangle=0.
121: \end{eqnarray}
122: It follows that the two quadratic quantities $\Psi_\theta=\langle
123: |(-\Delta)^{\theta/4}\psi|^2\rangle/2=\int\Psi_\theta(k)\,\dk$,
124: where $\theta=1,2$, are conserved by nonlinear transfer. Here, 
125: $\Psi_\theta(k)$ is defined by $\Psi_\theta(k)=k^\theta\Psi(k)$, 
126: $\Psi(k)$ is the power density of $\psi$ associated with wavenumber 
127: $k$ and $\theta$ is a real number. Note that $\Psi_2(k)$ is the kinetic 
128: energy spectrum and $\Psi_2$ is the kinetic energy density.
129: 
130: The simultaneous conservation of two quadratic quantities by advective 
131: nonlinearities is a common feature in incompressible fluid systems in 
132: two dimensions. Some familiar systems in this category are the 
133: Charney--Hasegawa--Mima equation (Hasegawa \& Mima 1978; Hasegawa, 
134: Maclennan \& Kodama 1979) and the class of $\alpha$ turbulence equations 
135: (Pierrehumbert \etal 1994), which includes both the Navier--Stokes
136: and the SQG equations. These conservation laws, together with the 
137: scale-selectivity of the dissipation and unboundedness of the domain, 
138: are the building block of the classical dual-cascade theory 
139: (Fj{\o}rtoft 1953; Kraichnan 1967, 1971; Leith 1968; Batchelor 1969). 
140: This theory, when applied to the present case, implies that $\Psi_1$ 
141: cascades to low wavenumbers (inverse cascade) and $\Psi_2$ cascades 
142: to high wavenumbers (direct cascade). For some recent discussion on 
143: the possibility of a dual cascade in various two-dimensional systems, 
144: including the Navier--Stokes and SQG equations, see \cite{TS02}, 
145: Tran \& Bowman (2003b,2004) and \cite{T04}. The inverse
146: cascade toward wavenumber $k=0$ would eventually evade viscous 
147: dissipation altogether because the spectral dissipation rate vanishes 
148: as $k\rightarrow0$. Hence, according to the classical picture, 
149: $\Psi_1$ necessarily grows unbounded, by a steady growth rate 
150: $\d\Psi_1/\dt>0$, as $t\rightarrow\infty$.
151: Strictly speaking, one may have to address the possibility of a 
152: dissipated inverse cascade, 
153: i.e. one for which the dissipation of $\Psi_1$ occurs at scales much 
154: larger than the forcing scale and for which $\d\Psi_1/\dt$ has a zero time
155: mean. Such a cascade is not a plausible scenario (and is not the 
156: traditional undissipated inverse cascade) in fluid systems, dissipated 
157: by a single viscous operator, where the viscous dissipation rate diminishes 
158: toward the large scales. A discussion of this issue can be found 
159: in \cite{T04}.
160:   
161: In this study, upper bounds are derived for the time averages of the 
162: kinetic energy density $\Psi_2$ and of the large-scale spectrum 
163: $\Psi_2(k)$. These bounds are derived from the governing equation, 
164: involving simple but rigorous estimates. The bound on $\Psi_2$ is valid 
165: in both unbounded and bounded cases, and a straightforward consequence
166: of this bound is a bound on the energy spectrum, which also applies to 
167: both unbounded and bounded turbulence. Another bound on the large-scale 
168: energy spectrum is derived by estimating the nonlinear triple-product
169: term representing the inverse transfer of $\Psi_1$. This result applies 
170: to bounded turbulence since upper bounds for the triple-product term 
171: are inherently domain-size dependent. The difficulties of extending this 
172: result to the unbounded case are discussed. Some numerical results 
173: confirming the theoretical predictions are presented.
174: 
175: \section{Bounded dynamical quantities}
176: 
177: A notable feature of unbounded incompressible fluid turbulence in two
178: dimensions is the appearance of infinite quadratic quantities (per unit
179: area): namely, the kinetic energy density $\Psi_2$ for Navier--Stokes
180: turbulence and $\Psi_1$ for the SQG case. According to the classical theory
181: (applied to the SQG case), a (steady) injection of $\Psi_1$, applied around
182: some finite wavenumber $s$, cascades to ever-larger scales, leading to an
183: unbounded growth of $\Psi_1$ (this is presumably the case for the general
184: quadratic invariant $\Psi_\alpha$ in the so-called $\alpha$ turbulence;
185: {\it cf.} Tran 2004).  In other words, if the classical inverse cascade is
186: realizable, unbounded incompressible fluid turbulence in two dimensions
187: constitutes an ill-posed problem, in the sense that a key quadratic
188: invariant becomes infinite. Of course, there still exist finite quadratic
189: quantities, in particular the dissipation agent for each quadratic
190: invariant. This section is concerned with these quantities.
191: 
192: On multiplying (\ref{governing}) by $\psi$ and $(-\Delta)^{1/2}\psi$ 
193: and taking the spatial average of the resulting equations, noting from 
194: the conservation laws that the nonlinear terms identically vanish, one 
195: obtains evolution equations for $\Psi_1$ and $\Psi_2$,
196: \begin{eqnarray}
197: \label{Psi1evolution}
198: \frac{\d}{\dt}\Psi_1&=&-2\mu\Psi_2+\langle f\psi\rangle,\\
199: \label{Psi2evolution}
200: \frac{\d}{\dt}\Psi_2&=&-2\mu\Psi_3
201: +\langle f(-\Delta)^{1/2}\psi\rangle.
202: \end{eqnarray}
203: Using the Cauchy--Schwarz and geometric--arithmetic mean inequalities, one
204: obtains upper bounds on the injection terms in (\ref{Psi1evolution}) and
205: (\ref{Psi2evolution}):
206: \begin{eqnarray}
207: \label{forcebounds}
208: \langle f\psi\rangle&\le&\langle|(-\Delta)^{1/2}\psi|^2\rangle^{1/2}
209: \langle|(-\Delta)^{-1/2}f|^2\rangle^{1/2}
210: \le\mu\Psi_2+\mu^{-1}F_{-2},\nonumber\\
211: \langle f(-\Delta)^{1/2}\psi\rangle &\le&
212: \langle|(-\Delta)^{3/4}\psi|^2\rangle^{1/2}
213: \langle|(-\Delta)^{-1/4}f|^2\rangle^{1/2}
214: \le\mu\Psi_3+\mu^{-1}F_{-1},
215: \end{eqnarray}
216: where the `integration by parts' rule $\langle(-\Delta)^\theta\phi
217: \chi\rangle=\langle(-\Delta)^{\theta'}\phi(-\Delta)^{\theta''}\chi\rangle$,
218: for $\theta=\theta'+\theta''$, has been used and $F_\theta=\langle
219: |(-\Delta)^{\theta/4}f|^2\rangle/2$.  
220: Substituting (\ref{forcebounds}) in (\ref{Psi1evolution}) and 
221: (\ref{Psi2evolution}) yields
222: \begin{eqnarray}
223: \label{evolbound1}
224: \frac{\d}{\dt}\Psi_1&\le&-\mu\Psi_2+\mu^{-1}F_{-2},\\
225: \label{evolbound2}
226: \frac{\d}{\dt}\Psi_2&\le&-\mu\Psi_3+\mu^{-1}F_{-1}.
227: \end{eqnarray}
228: To avoid unnecessary complications, zero initial conditions are assumed,
229: so that for $T>0$ the time means $\langle\d\Psi_1/\dt\rangle_t=
230: \Psi_1(T)/T$ and $\langle\d\Psi_2/\dt\rangle_t$ are non-negative. 
231: %The time-asymptotic mean $\left\langle \d\Psi_1/\d t
232: %\right\rangle_t=\lim_{T\rightarrow \infty} \frac{1}{T}\int_0^T \d\Psi_1/\d t
233: % \ge 0$ (whenever it exists) since
234: %$\int_0^T \d\Psi_1/\d t\ge-\Psi_1(0)$ for any $T$. Similarly, $\left\langle
235: %\d\Psi_2/\d t \right\rangle_t \ge 0$.
236: One can then deduce upper bounds on the time means 
237: $\langle\Psi_2\rangle_t$ and $\langle\Psi_3\rangle_t$, which are valid 
238: regardless of whether or not $\Psi_1$ remains finite in the limit 
239: $t\rightarrow\infty$:
240: \begin{eqnarray}
241: \label{averagebound1}
242: \langle\Psi_2\rangle_t &\le& \mu^{-2}\langle F_{-2}\rangle_t,\\
243: \label{averagebound2}
244: \langle\Psi_3\rangle_t &\le& \mu^{-2}\langle F_{-1}\rangle_t.
245: \end{eqnarray}
246: 
247: For $\theta\in(2,3)$, $\langle\Psi_\theta\rangle_t$ is also bounded.
248: Indeed, from the H\"{o}lder inequalities 
249: $\Psi_\theta\le\Psi_2^{3-\theta}\Psi_3^{\theta-2}$ ({\it cf.} Tran 2004) and 
250: $\langle\Psi_2^{3-\theta}\Psi_3^{\theta-2}\rangle_t\le
251: \langle\Psi_2\rangle_t^{3-\theta}\langle\Psi_3\rangle_t^{\theta-2}$,
252: one can deduce from (\ref{averagebound1}) and (\ref{averagebound2}) that
253: \begin{eqnarray}
254: \langle\Psi_\theta\rangle_t &\le& 
255: \langle\Psi_2\rangle_t^{3-\theta}\langle\Psi_3\rangle_t^{\theta-2}
256: \le \mu^{-2}
257: \langle F_{-2}\rangle_t^{3-\theta}\langle F_{-1}\rangle_t^{\theta-2}.
258: \end{eqnarray}
259: This result implies that for $\theta\in(2,3)$, 
260: $\langle\Psi_\theta\rangle_t$ is bounded, provided that both 
261: $\langle F_{-1}\rangle_t$ and $\langle F_{-2}\rangle_t$ are bounded. 
262: This condition is assured if $s>0$ and $F_0$ is bounded, a condition 
263: normally required of the forcing, because $F_{-2}\le F_{-1}/s\le F_0/s^2$. 
264: One may even consider a class of forcing for which $F_0=\infty$ and
265: $F_{-2}\le F_{-1}/s<\infty$. 
266: 
267: Upper bounds of the above type on dynamical quantities are rather 
268: trivial for bounded turbulence. However, they are important in the 
269: unbounded case, for two reasons. First, the scale-selective viscous 
270: dissipation allows for the possibility of unbounded growth of certain 
271: quadratic quantities toward the low wavenumbers. Hence, rigorous
272: bounds on dynamical quantities are not as abundant as in the bounded
273: case. Second, analytic studies of the nonlinear triple-product transfer
274: function are difficult in unbounded domains. In the absence of pointwise 
275: estimates for the spectrum, these bounds are particularly useful for 
276: qualitative estimates of the large-scale distribution of energy. For 
277: example, \cite{T04} uses inequality (\ref{averagebound1}) to argue that 
278: the energy spectrum $\Psi_2(k)$ should be shallower than $k^{-1}$,
279: as $k\rightarrow0$. 
280: 
281: \section{Large-scale energy spectrum}
282: 
283: In this section, it is shown that the physical laws of SQG dynamics 
284: admit only large-scale energy spectra shallower than $k^{-1}$. This 
285: result is due in part to the fact that the simultaneous conservation
286: of $\Psi_1$ and $\Psi_2$ allows virtually no kinetic energy to get
287: transferred toward the low wavenumbers, so that only large-scale kinetic 
288: energy spectra shallower than $k^{-1}$ are possible.
289: 
290: \subsection{Shell-averaged energy spectrum}
291: For a given wavenumber $r$, let us denote by $S=S(r)$ the wavenumber shell
292: between $k=r/2$ and $k=3r/2$, i.e. $S(r)=\{\k : r/2 \le k \le 3r/2\}$.
293: The shell-averaged energy spectrum $\overline\Psi_2(r)$ over $S(r)$ is
294: defined by
295: \begin{eqnarray}
296: \label{spectrum}
297: \overline\Psi_2(r)&=&\frac{1}{r}\int_{r/2}^{3r/2}\Psi_2(k)\,\dk.
298: \end{eqnarray}
299: In the present case of a doubly periodic domain of size $L$, the Fourier 
300: representation of the stream function is 
301: $\psi(\x)=\sum_{\k}\exp\{i\k\cdot\x\}\widehat\psi(\k)$, where 
302: $\k=2\pi L^{-1}(k_x,k_y)$ with $k_x$ and $k_y$ being integers not 
303: simultaneously zero. Let $\psi(S)$ denote the component of $\psi$ 
304: spectrally supported by $S$, i.e. $\psi(S)=\sum_{\k\in S}\exp
305: \{i\k\cdot\x\}\widehat\psi(\k)$. One has 
306: \begin{eqnarray}
307: \label{ineq}
308: \sup_{\x}|\nabla\psi(S)| &\le& \sum_{\k\in S}k|\widehat\psi(\k)| 
309: \le \left(\sum_{\k\in S}1\sum_{\k\in S}k^2|\widehat\psi(\k)|^2\right)^{1/2}
310: \le cLr\Psi_2^{1/2}(S),
311: \end{eqnarray}
312: where the Cauchy--Schwarz inequality is used, the sum $\sum_{\k\in S}1=(cLr)^2$
313: is the number of wavevectors in $S$, $c$ is an absolute constant of 
314: order unity and $\Psi_2(S)$ is the contribution to the kinetic energy 
315: from $S$.
316: 
317: \subsection{Upper bounds for the energy spectrum}
318: A simple upper bound for $\overline\Psi_2(k)$, which is applicable to both the
319: unbounded and bounded cases, can be derived from (\ref{averagebound1}).
320: In fact, it follows from (\ref{averagebound1}) and (\ref{spectrum}) that
321: \begin{eqnarray}
322: \label{spectbound1}
323: \langle\overline\Psi_2(k)\rangle_t&=&\frac{1}{k}\int_{k/2}^{3k/2}
324: \langle\Psi_2(\kappa)\rangle_t\,\d\kappa
325: \le \mu^{-2}\langle F_{-2}\rangle_tk^{-1}.
326: \end{eqnarray}
327: This bound is supposed to apply to $k$ in the inverse-transfer region.
328: For $k$ in the direct-transfer region, (\ref{averagebound2}) yields
329: \begin{eqnarray}
330: \label{spectbound2}
331: \langle\overline\Psi_2(k)\rangle_t&=&\frac{1}{k}\int_{k/2}^{3k/2}
332: \langle\Psi_2(\kappa)\rangle_t\,\d\kappa
333: \le \frac{2}{k^2}\int_{k/2}^{3k/2}
334: \langle\Psi_3(\kappa)\rangle_t\,\d\kappa
335: \le 2\mu^{-2}\langle F_{-1}\rangle_tk^{-2}.
336: \end{eqnarray}
337: 
338: The upper bound (\ref{spectbound1}) suggests that dimensional analysis
339: arguments, which predict a large-scale $k^{-1}$ energy spectrum, are not well 
340: justified. If a persistent inverse cascade of $\Psi_1$ exists 
341: ($\d \Psi_1/\dt > 0$), then the energy $\Psi_2$ ought to acquire a value
342: such that $\Psi_2<\mu^{-2}F_{-2}$. In the unbounded case, the large-scale
343: energy spectrum then needs to be strictly shallower than $k^{-1}$, to
344: ensure that the dissipation of $\Psi_1$ does not increase without bound as
345: the inverse cascade proceeds toward $k=0$.
346: On the other hand, if no inverse cascade of $\Psi_1$ exists, then 
347: a $k^{-1}$ energy spectrum with limited extent is possible. If viscous 
348: dissipation mechanisms with degrees higher than that of the 
349: natural dissipation are considered, then the upper bounds derived above 
350: are not valid. Nevertheless, diminishing energy transfer towards the lowest
351: wavenumbers appears to be consistent only with spectra shallower than $k^{-1}$
352: (for low-wavenumber convergence of the energy integral).
353: The numerical results reported in~\S\,4 are well suited to this expectation.
354: 
355: An upper bound for the large-scale energy spectrum, based on the 
356: nonlinear transfer term, can be derived for the bounded case. This 
357: analysis employs elementary but rigorous estimates of the 
358: triple-product term. For $3k/2<s$, the evolution of $\Psi_1(S(k))$ 
359: is governed by
360: \begin{eqnarray}
361: \frac{\d}{\dt}\Psi_1(S)&=&-\langle\psi(S)
362: J(\psi,(-\Delta)^{1/2}\psi)\rangle-2\mu\Psi_2(S) \nonumber\\
363: &=&\langle(-\Delta)^{1/2}\psi J(\psi,\psi(S))\rangle-2\mu\Psi_2(S)\nonumber\\
364: &\le&\langle|(-\Delta)^{1/2}\psi||\nabla\psi||\nabla\psi(S)|\rangle
365: -2\mu\Psi_2(S)\nonumber\\
366: &\le&\sup_{\x}|\nabla\psi(S)|\langle|(-\Delta)^{1/2}\psi||\nabla\psi|\rangle
367: -2\mu\Psi_2(S)\nonumber\\
368: &\le&2cLk\Psi_2^{1/2}(S)\Psi_2-2\mu\Psi_2(S)\nonumber\\
369: &\le&c^2\mu^{-1}L^2k^2\Psi_2^2-\mu\Psi_2(S)\nonumber\\
370: &=&c^2\mu^{-1}L^2k^2\Psi_2^2-\mu k\overline\Psi_2(k),
371: \end{eqnarray}
372: where the second equality is a consequence of (\ref{id}) and the second last
373: and last inequalities follow from (\ref{ineq}) and the geometric--arithmetic
374: mean inequality, respectively.
375: It follows that
376: \begin{eqnarray}
377: \label{spectbound3}
378: \langle\overline\Psi_2(k)\rangle_t&\le&c^2\mu^{-2}L^2k\langle\Psi_2^2\rangle_t.
379: \end{eqnarray}
380: A notable feature of (\ref{spectbound3}) is its dependence on the 
381: fluid domain size. The presence of $L$ in this upper bound is natural: 
382: the upper bound $\sup_{\x}|\nabla\psi(S)|$, which is associated with 
383: the fluid velocity at scales $\approx k^{-1}$, is inherently
384: domain-size dependent. There are no known analytic estimates that allow
385: one to derive an upper bound on the nonlinear transfer function
386: $\langle\psi(S)J(\psi,(-\Delta)^{1/2}\psi)\rangle$ in 
387: terms of `intensive quantities' only. This difficulty arises not only
388: in the present estimate but also in other analytic estimates 
389: of the transfer function. In other words, the nonlinear triple-product 
390: term is intrinsically domain-size dependent. This problem considerably 
391: limits our ability to assess the nonlinear transfer in unbounded systems. 
392: Finally, it is worth mentioning that although the upper bound 
393: (\ref{spectbound3}) has a linear dependence on $k$, it may be more 
394: excessive than the bound $\mu^{-2}\langle F_{-2}\rangle_t k^{-1}$ 
395: derived earlier (even for very low wavenumbers). The reason is that 
396: $L^2k\ge k^{-1}$ and the prefactor $c^2\langle\Psi_2^2\rangle_t$ 
397: may not be as optimal as $\langle F_{-2}\rangle_t$.
398: 
399: \section{Numerical results}
400: 
401: This section reports results from numerical simulations that illustrate 
402: the realization of large-scale spectra shallower than $k^{-1}$.  
403: Equation (\ref{governing}) is simulated in a doubly periodic square of 
404: side $2\pi$, where the forcing $\widehat{f}(\bm k)$ is nonzero only for 
405: those wavevectors $\bm k$ having magnitudes lying in the interval 
406: $K=[59,61]$:
407: \begin{eqnarray}
408: \label{forcing}
409: \widehat{f}(\bm k)&=&\frac{\epsilon}{N}\frac{\widehat{\psi}(\bm{k})}
410: {2\Psi_1(k)}.
411: \end{eqnarray} 
412: Here $\epsilon=1$ is the constant energy injection rate and~$N$ is the 
413: number of distinct wavenumbers in~$K$. The (constant) injection rate of 
414: $\Psi_1$ is $\epsilon/s\approx1/60$, where $1/s\approx1/60$ is the mean 
415: of $k^{-1}$ over $K$. This type of forcing was used by \cite{Shepherd87}, 
416: \cite{T04} and \cite{TB04} in numerical simulations of a large-scale 
417: zonal jet on the so-called beta-plane and of Navier--Stokes turbulence.  
418: The attractive aspect of (\ref{forcing}), as noted in \cite{Shepherd87}, 
419: is that it is steady. Dealiased $683^2$ and $1365^2$ pseudospectral 
420: simulations ($1024^2$ and $2048^2$ total modes) were performed. 
421: Three dissipative forms were considered: $2.5\times10^{-2}\Delta\psi$, 
422: $-4\times10^{-4}(-\Delta)^{3/2}\psi$, and 
423: $-6\times10^{-6}\Delta^2\psi+\mu\Delta\psi$ for several values of $\mu$.
424: The first case represents the natural dissipation of the SQG dynamics
425: due to Ekman pumping, as mentioned earlier. The second case represents
426: thermal (molecular) diffusion since $(-\Delta)^{1/2}\psi$ is equivalent 
427: to the fluid temperature. The third case---the mixed hyperviscous/Ekman 
428: dissipation form---is considered in order to demonstrate that even slight 
429: amounts of Ekman damping will inhibit the formation of an inverse cascade.
430: Unlike \cite{Smith02}, the case of mechanical friction
431: [$\propto(-\Delta)^{1/2}\psi$] was not considered. The higher resolution 
432: was used for the first (natural dissipation) case and the lower resolution 
433: was used for the second and third cases. All simulations were initialized 
434: with the spectrum $\Psi_2(k)=10^{-5}\pi k/(60^2+k^2)$. 
435: 
436: Figure~\ref{sqg2} shows the time-averaged steady-state kinetic energy
437: spectrum for the case of the natural dissipation term
438: $2.5\times10^{-2}\Delta\psi$. The dissipation agents of $\Psi_1$ and
439: $\Psi_2$ are, respectively, $\Psi_2$ (energy) and $\Psi_3$. 
440: The value of the energy, $0.3333$, implies that 
441: the dissipation of $\Psi_1$, averaged in the same period, is $0.01666$. 
442: This amounts to virtually all of the injection rate $1/60$. Hence, there
443: exists no inverse cascade of $\Psi_1$ to the large scales and both
444: $\Psi_1$ and $\Psi_2$ are steady. The small-scale energy
445: spectrum scales as $k^{-3.5}$, so that the spectrum $\Psi_3(k)$ scales as
446: $k^{-2.5}$. This scaling means that the energy dissipation occurs mainly
447: around the forcing region and is consistent with the bound
448: (\ref{spectbound2}). 
449: 
450: Unlike Navier--Stokes turbulence, for which the inverse energy cascade
451: is robust and can be simulated at relatively low resolution, it was
452: noticed that no choice for the value of $\mu$ at the present resolution
453: could be used to simulate an inverse cascade of $\Psi_1$. It is not known
454: whether an inverse cascade of $\Psi_1$ is realizable at higher resolutions,
455: using a smaller value of $\mu$. Nevertheless, this observation 
456: suggests that $\Psi_1$ is `reluctant' to cascade to the large scales, 
457: as compared with the more robust inverse energy cascade in 
458: Navier--Stokes turbulence.
459: 
460: \begin{figure}
461: \centerline{\includegraphics{sqg2}}
462: \caption{The time-averaged steady-state energy spectrum $\Psi_2(k)$ {\it vs.}\ 
463: $k$ for the dissipation term $2.5\times10^{-2}\Delta\psi$.}
464: \label{sqg2}
465: \end{figure}
466: 
467: Figure~\ref{sqgviscous1} shows the kinetic energy spectrum averaged
468: between $t=37.3$ and $t=38.7$, for a lower viscous degree.
469: The dissipation agents of $\Psi_1$
470: and $\Psi_2$ are, respectively, $\Psi_3$ and $\Psi_4$ (enstrophy). 
471: The value of $\Psi_3$ is $20$, implying that the dissipation of $\Psi_1$
472: is $1.6\times10^{-2}$. This amounts to about $96\%$ of the injection rate
473: $1/60$. The inverse cascade then carries only a few percent of the
474: injection of $\Psi_1$ to the large scales.
475: 
476: The small-scale energy spectrum scales as $k^{-4.5}$, so that the 
477: enstrophy spectrum $\Psi_4(k)$ scales as $k^{-2.5}$. 
478: Most of the energy dissipation occurs around 
479: the forcing region, consistent with a `weak' inverse cascade
480: (one that does not carry virtually all of the injection of $\Psi_1$ toward
481: $k=0$; {\it cf.} Tran and Bowman 2004, Tran 2004). No direct cascade is
482: possible for bounded turbulence in equilibrium or for unbounded turbulence
483: in the presence of a weak inverse cascade. 
484: \begin{figure}
485: \centerline{\includegraphics{sqgviscous1}}
486: \caption{The quasisteady energy spectrum $\Psi_2(k)$ {\it vs.}\ $k$ averaged
487: between $t=37.3$ and $t=38.7$ for the dissipation term
488: $-4\times10^{-4}(-\Delta)^{3/2}\psi$.}
489: \label{sqgviscous1}
490: \end{figure}
491: 
492: Similarly, Figure~\ref{sqgviscous2mixed} shows the kinetic energy spectrum 
493: averaged between $t=15.7$ and $t=16.5$ for the mixed dissipation
494: $-6\times10^{-6}\Delta^2\psi+\mu\Delta\psi$, using three different values of
495: $\mu$. When $\mu=0$, the dissipation agents of $\Psi_1$ and $\Psi_2$ are,
496: respectively, $\Psi_4$ (enstrophy) and $\Psi_5$.
497: The value of the enstrophy, $1208$, implies that the dissipation of $\Psi_1$ is
498: $1.45\times10^{-2}$, amounting to about $87\%$ of the injection rate $1/60$.
499: The small-scale energy 
500: spectrum scales as $k^{-5}$, so that $\Psi_5(k)$ scales as $k^{-2}$. 
501: Again, this scaling means that most of the energy dissipation occurs around
502: the forcing region and that the inverse cascade is weak.
503: We note that as $\mu$ is increased, the inverse cascade becomes
504: increasingly weak. We emphasize this behaviour by plotting 
505: in Fig.~\ref{invstrengthvnuL} the inverse cascade {\it strength}
506: $r=1-2s(\mu\Psi_2+6\times10^{-6}\Psi_4)/\epsilon$ for six different values
507: of $\mu$.
508: 
509: \begin{figure}
510: \centerline{\includegraphics{sqgviscous2mixed}}
511: \caption{The quasisteady energy spectrum $\Psi_2(k)$ {\it vs.}\ $k$ averaged
512: between $t=15.67$ and $t=16.52$ for the dissipation term
513: $-6\times10^{-6}\Delta^2\psi+\mu\Delta\psi$, using three different values of
514: $\mu$.
515: } 
516: \label{sqgviscous2mixed}
517: \end{figure}
518: 
519: \begin{figure}
520: \centerline{\includegraphics{invstrengthvnuL}}
521: \caption{The decay of the inverse cascade strength $r$ for
522: the dissipation term $-6\times10^{-6}\Delta^2\psi+\mu\Delta\psi$ as $\mu$
523: is increased.} 
524: \label{invstrengthvnuL}
525: \end{figure}
526: 
527: Unlike Navier--Stokes turbulence, for which the enstrophy acquires its
528: near-equilibrium value once a discernible inverse-transfer range
529: has formed, the energy in SQG turbulence can remain significantly less 
530: than its equilibrium value until a very wide inverse-transfer range 
531: has developed. For example, for a one-decade Navier--Stokes
532: inverse-transfer range (achievable in numerical simulations),
533: the enstrophy acquires 95\% of its projected equilibrium value (calculated
534: with a $k^{-5/3}$ energy spectrum extrapolated to $k=0$).
535: On the other hand, for a one-decade SQG inverse-transfer range, the energy
536: acquires only 66\% of its projected equilibrium value (calculated with a
537: $k^{-0.7}$ energy spectrum extrapolated to $k=0$, as realized in the 
538: present simulations; {\it cf.} the $\mu=0$ case of
539: Figure~\ref{sqgviscous2mixed}). This means
540: that one needs a considerably wider inverse-transfer region for SQG turbulence 
541: than for Navier--Stokes turbulence, in order to approach a 
542: quasi-steady state. This problem is in addition to the resolution 
543: limitations at the small scales for both cases.  
544: 
545: Due to the steep spectrum in the inverse-transfer region, the energy 
546: in the $\mu=0$ case of Figure~\ref{sqgviscous2mixed} has not acquired a
547: value considerably close to its equilibrium value. This means that the
548: system is still well within the transient phase, However, the dissipation
549: of $\Psi_1$ (proportional to the enstrophy) cannot grow considerably (without
550: significant change to the existing  spectrum), because of the high degree
551: of viscosity, which makes the dissipation of $\Psi_1$ relatively
552: insensitive to growth of the large-scale energy. 
553: 
554: \section{Conclusion and discussion}
555: 
556: In this paper, the kinetic energy density of SQG turbulence and its 
557: large-scale spectrum have been studied. For the unbounded case, upper 
558: bounds are derived for the time means of the kinetic energy density 
559: and of the large-scale energy spectrum, averaged over a narrow window 
560: of wavenumbers. Another result is an upper bound on the the time mean 
561: of the large-scale energy spectrum, which is derived for 
562: the bounded case. Numerical results confirming the predicted slopes 
563: of the large-scale energy spectrum are presented and discussed. 
564: 
565: An important feature in SQG turbulence that gives rise to the rigorous 
566: upper bound on the time mean of the kinetic energy density in the 
567: unbounded case is that the kinetic energy is the dissipation agent of 
568: the inverse-cascading candidate $\Psi_1$. This fact is due to the 
569: hypoviscous nature of the dissipation operator $(-\Delta)^{1/2}$, a 
570: natural physical dissipation mechanism of SQG dynamics (Ohkitani 1997; 
571: Constantin 2002; Tran 2004). If $(-\Delta)^{1/2}$ is replaced by an 
572: operator of the form $(-\Delta)^{\eta}$, where $\eta>1/2$, then the 
573: simple analysis of Section 2 fails to show that the time mean of the 
574: energy density $\langle\Psi_2\rangle_t$ is bounded, although it may 
575: remain so for low degrees of viscosity $\eta$. The reason is that  
576: the amount of energy getting transferred to wavenumbers lower than a 
577: given wavenumber $k$ decreases at least as rapidly as $k$, 
578: so that the spectral dissipation rate $\propto k^{2\eta}$, a consequence
579: of the dissipation operator $(-\Delta)^{\eta}$, may be  
580: sufficiently strong to balance the diminishing inverse energy transfer
581: and keep the energy from growing unbounded. 
582: 
583: Numerical simulations of SQG turbulence were performed, using the 
584: natural dissipation operator $(-\Delta)^{1/2}$ and two viscous 
585: operators $\Delta$ and $(-\Delta)^{3/2}$. The results show 
586: large-scale energy spectra shallower than $k^{-1}$, consistent with the
587: theoretical prediction.
588:  
589: There have been attempts to explain, within the context of SQG 
590: turbulence (Constantin 2002; Tung \& Orlando 2003), the kinetic energy 
591: spectra observed in the laboratory experiment of \cite{Baroud02} and 
592: in the atmosphere. In the former case, the 
593: turbulence in a rotating tank is driven at a sufficiently small
594: scale to allow for a wide inverse-transferring range. A $k^{-2}$
595: spectrum extending over nearly two wavenumber decades lower than
596: the forcing wavenumber is observed. In the latter case, a 
597: $k^{-5/3}$ spectrum is observed in the mesoscales (see Frisch 
598: 1995 and Tung \& Orlando 2003 and references therein), which
599: correspond to wavenumbers higher (lower) than the forcing 
600: wavenumber if the energy released from baroclinic instability 
601: (thunderstorms) is considered to be the driving force. The $-2$ power-law
602: scaling observed in \cite{Baroud02} for the wavenumber range lower than the
603: forcing wavenumber is excessively steeper than the permissible scalings
604: derived in this work. The $-5/3$ slope in the atmosphere is either steeper
605: (if considered to be on the wavenumber range lower than the forcing 
606: wavenumber) or shallower (if considered to be on the wavenumber range 
607: higher than the forcing wavenumber) than the permissible slopes.
608: According to the present analysis, these data cannot be attributed to
609: SQG turbulence. 
610: 
611: \begin{acknowledgments}
612: We would like to thank two anonymous referees for their comments,
613: which were helpful in improving this manuscript. This work was funded by a
614: Pacific Institute for the Mathematical Sciences Postdoctoral Fellowship, an
615: Alexander von Humboldt Research Fellowship, and the Natural Sciences and
616: Engineering Research Council of Canada.
617: \end{acknowledgments}
618:  
619: \begin{thebibliography}{10}
620: 
621: \bibitem[Baroud (2002)]{Baroud02} 
622:      \textsc{Baroud, C.N., Plapp, B.B, She, Z.-S. \& Swinney H.L.} 2002
623:      {Anomalous self-similarity in a turbulent rapidly rotating fluid.}
624:      \textit{Phys. Rev. Lett.} \textbf{88}, 114501.
625: 
626: \bibitem[Batchelor (1969)]{Batchelor69} 
627:      \textsc{Batchelor, G.K.} 1969 
628:      {Computation of the energy spectrum in homogeneous, 
629:      two-dimensional turbulence.} 
630:      \textit{Phys. Fluids} \textbf{12}, II 233--239.
631: 
632: \bibitem[Blumen (1978)]{Blumen78} 
633:      \textsc{Blumen, W.} 1978
634:      {Uniform potential vorticity flow, Part I: 
635:      Theory of wave interactions and two-dimensional turbulence.}
636:      \textit{J. Atmos. Sci.} \textbf{35}, 774--783.
637: 
638: \bibitem[Charney (1948)]{Charney48} 
639:      \textsc{Charney, J.G.} 1948
640:      {On the scale of atmospheric motions.} 
641:      \textit{Geofys. Publ.} \textbf{17}, 3--17.
642: 
643: \bibitem[Charney (1971)]{Charney71} 
644:      \textsc{Charney, J.G.} 1971   
645:      {Geostrophic turbulence.} 
646:      \textit{J. Atmos. Sci.} \textbf{28}, 1087--1095.
647: 
648: \bibitem[Constantin (2002)]{Constantin02} 
649:      \textsc{Constantin, P.} 2002
650:      {Energy spectrum of quasigeostrophic turbulence.} 
651:      \textit{Phys. Rev. Lett.} \textbf{89}, 184501.
652: 
653: \bibitem[Constantin (1994a)]{Constantin94a} 
654:      \textsc{Constantin, P., Majda, A.J. \& Tabak, E.G.} 1994
655:      {Singular front formation in a model for quasigeostrophic flow.} 
656:      \textit{Phys. Fluids} \textbf{6}, 9--11.
657: \bibitem[Constantin (1994b)]{Constantin94b}   
658:      \textsc{Constantin, P., Majda, A.J. \& Tabak, E.G.} 1994
659:      {Formation of strong fronts in the 2-D quasigeostrophic thermal 
660:      active scalar.}
661:      \textit{Nonlinearity} \textbf{7}, 1495--1533. 
662: 
663: \bibitem[Fj{\o}rtoft (1953)]{Fjortoft53}
664:      \textsc{Fj{\o}rtoft, R.} 1953
665:      {On the changes in the spectral distribution of kinetic energy
666:      for twodimensional, nondivergent flow.} 
667:      \textit{Tellus} \textbf{5}, 225--230.
668: 
669: \bibitem[Frisch (1995)]{Frisch95} 
670:      \textsc{Frisch, U.} 1995
671:      \textit{Turbulence: The Legacy of A.~N. Kolmogorov.}
672:      {Cambridge University Press, Cambridge.}
673: 
674: \bibitem[Hasegawa (1978)]{Hasegawa78} 
675:      \textsc{Hasegawa, A. \& Mima, K.} 1978
676:      {Pseudo-three-dimensional turbulence in magnetized nonuniform plasma.} 
677:      \textit{Phys. Fluids} \textbf{21}, 87--92.
678:               
679: \bibitem[Hasegawa (1979)]{Hasegawa79} 
680:      \textsc{Hasegawa, A., Maclennan, C.G. \& Kodama, Y.} 1979
681:      {Nonlinear behavior and turbulence spectra of drift waves 
682:      and Rossby waves.} 
683:      \textit{Phys. Fluids} \textbf{22}, 2122--2129.
684: 
685: \bibitem[Held \etal (1995)]{Held95} 
686:      \textsc{Held, I.M., Pierrehumbert, R.T., Garner, S.T. \& Swanson, K.L.}
687:      1995
688:      {Surface quasi-geostrophic dynamics.} 
689:      \textit{J. Fluid Mech.} \textbf{282}, 1--20.
690: 
691: \bibitem[Kraichnan (1967)]{Kraichnan67}
692:      \textsc{Kraichnan, R.H.} 1967 
693:      {Inertial ranges in two-dimensional turbulence.} 
694:      \textit{Phys.~Fluids} \textbf{10}, 1417--1423. 
695: 
696: \bibitem[Kraichnan (1971)]{Kraichnan71}
697:       \textsc{Kraichnan, R.H.} 1971 
698:       {Inertial-range transfer in two- and three-dimensional turbulence.}
699:       \textit{J.~Fluid Mech.} \textbf{47}, 525--535.
700: 
701: \bibitem[Leith (1968)]{Leith68}
702:       \textsc{Leith, C.E.} 1968 
703:       {Diffusion approximation for two-dimensional turbulence.}
704:       \textit{Phys.~Fluids} \textbf{11}, 671--673.
705: 
706: \bibitem[Ohkitani (1997)]{Ohkitani97} 
707:       \textsc{Ohkitani, K. \& Yamada, M.} 1997
708:       {Inviscid and inviscid-limit behavior of a surface 
709:       quasigeostrophic flow.} 
710:       \textit{Phys. Fluid} \textbf{9}, 876--882.
711: 
712: \bibitem[Pedlosky (1987)]{Pedlosky87} 
713:       \textsc{Pedlosky, J.} 1987
714:       \textit{Geophysical Fluid Dynamics.}
715:       {2nd Edition, Springer, New York, 1987.}
716: 
717: \bibitem[Pierrehumbert, Held \& Swanson (1994)]{Pierrehumbert94} 
718:       \textsc{Pierrehumbert, R.T., Held, I.M., Swanson, K.L.} 1994
719:       {Spectra of local and nonlocal two-dimensional turbulence.} 
720:       \textit{Chaos Solitons Fract.} \textbf{4}, 1111--1116.
721: 
722: \bibitem[Rhines (1979)]{Rhines79} 
723:       \textsc{Rhines, P.B.} 1979
724:       {Geostrophic turbulence.}
725:       \textit{Ann. Rev. Fluid Mech.} \textbf{11}, 401--441.
726: 
727: \bibitem[Shepherd (1987)]{Shepherd87} 
728:       \textsc{Shepherd, T.G.} 1987
729:       {Rossby waves and two-dimensional turbulence in a 
730:       large-scale zonal jet.} 
731:       \textit{J. Fluid Mech.} \textbf{183}, 467--509.
732: 
733: \bibitem[Smith \etal (2002)]{Smith02} 
734:       \textsc{Smith, K.S., Boccaletti, G., Henning, C.C., Marinov,  
735:       I., Tam, C.Y., Held, I.M. \& Vallis, G.K.} 2002
736:       {Turbulent diffusion in the geostrophic inverse cascade.} 
737:       \textit{J. Fluid Mech.} \textbf{469}, 13--48.
738: 
739: \bibitem[Tran (2004)]{T04} 
740:       \textsc{Tran, C.V.} 2004
741:       {Nonlinear transfer and spectral distribution of energy in 
742:       $\alpha$ turbulence.}
743:       \textit {Physica D} \textbf{191}, 137--155.
744: 
745: \bibitem[Tran \& Bowman (2004)]{TB04} 
746:       \textsc{Tran, C.V. \& Bowman, J.C.} 2004
747:       {Robustness of the inverse cascade in two-dimensional turbulence.}
748:       \textit{Phys. Rev. E} \textbf{69}, 036303.
749: 
750: \bibitem[Tran \& Bowman (2003a)]{TB03a} 
751:        \textsc{Tran, C.V. \& Bowman, J.C.} 2003a
752:        {On the dual cascade in two-dimensional turbulence.}
753:        \textit{Physica D} \textbf{176}, 242--255. 
754: 
755: \bibitem[Tran \& Bowman (2003b)]{TB03b} 
756:        \textsc{Tran, C.V. \& Bowman, J.C.} 2003b
757:        {Energy budgets in Charney--Hasegawa--Mima and surface 
758:        quasigeostrophic turbulence.}
759:        \textit{Phys. Rev. E} \textbf{68}, 036304. 
760: 
761: \bibitem[Tran \& Shepherd (2002)]{TS02}
762:       \textsc{Tran, C.V. \& Shepherd, T.G.} 2002 
763:       {Constraints on the spectral distribution of energy and enstrophy 
764:       dissipation in forced two-dimensional turbulence.} 
765:       \textit{Physica D} \textbf{165}, 199--212. 
766: 
767: \bibitem[Tung \& Orlando (2003)]{Tung03} 
768:       \textsc{Tung, K.K. \& Orlando, W.W.} 2003
769:       {On the differences between 2D and QG turbulence.} 
770:       \textit{Discrete Contin. Dyn. Syst. Ser. B} \textbf{3}, 145--162.
771: 
772: \end{thebibliography}
773: 
774: %\bibliography{ref}
775: 
776: \end{document}
777: 
778: 
779: