nlin0512039/main.tex
1: %\documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2: \documentclass[showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
3: \usepackage{graphicx}% Include figure files
4: \usepackage{dcolumn}% Align table columns on decimal point
5: \usepackage{bm}% bold math
6: 
7: \voffset=0.5in
8: \textheight=9in
9: 
10: \newcommand{\pdf}{\textsc{PDF}}
11: \begin{document}
12: 
13: %**Spelling check:Condensable or Condensible?
14: 
15: \title{Statistical Equilibria of Uniformly Forced Advection Condensation}
16: \author
17: {Jai Sukhatme$^1$
18:  and Raymond T. Pierrehumbert$^2$\\
19:  $^1$ Mathematics Department, University of Wisconsin-Madison, Madison, WI 53706 \\
20:  $^2$ Department of Geophysical Sciences,
21:   University of Chicago, Chicago,  IL 60637 }
22: 
23: \date{\today}
24: 
25: \begin{abstract}
26: We examine the state of statistical equilibrium attained by a uniformly
27: forced condensable substance subjected to advection in a periodic
28: domain. In particular, we examine the 
29: probability density function (\pdf{}) of the condensable substance
30: in the limit of rapid condensation. The constraints imposed by this limit are pointed out and are shown to result in
31: a \pdf{} --- whenever the advecting velocity field admits a diffusive representation --- that features a
32: peak at small values,
33: decays exponentially and terminates in a
34: rapid "roll-off" near saturation. Possible physical implications of this 
35: feature as compared to a \pdf{} which continues to decay slowly are pointed out. A
36: set of simple numerical exercises which employ lattice maps for purposes of advection are 
37: performed to test these features.
38: Despite the simplicity of the model, the derived \pdf{} is seen to compare
39: favourably with \pdf{}'s constructed from isentropic specific humidity 
40: data. 
41: Further, structure functions associated with the condensable field are seen to scale anomalously with 
42: near saturation of the scaling exponents for high moments
43: --- a feature which agrees with studies of high resolution aircraft data. 
44: \end{abstract}
45: 
46: \pacs{PACS number 47.52.+j, 05.45.-a}
47: \maketitle
48: 
49: \section{Introduction}
50: 
51: {\it Advection-Diffusion-Condensation} (ADC) is a variant of the familiar passive advection-diffusion problem
52: \cite{Falk}, wherein apart from molecular diffusion a tracer is subject to an additional sink associated 
53: with the process of condensation. 
54: A study of the interplay of these processes is motivated by the need to understand the large 
55: scale distribution of water vapor in the troposphere \cite{Sherwood},\cite{SH},\cite{Ray-grl}. 
56: Indeed, the prominent role played by
57: water vapor in various problems related the Earth's climate \cite{Ray-Nature} --- for example, via the water 
58: vapor feedback 
59: in the greenhouse effect \cite{HS} --- makes this an important issue. 
60: In this context, ADC serves as an idealized problem for 
61: water vapor in the troposphere \cite{Ray-2}.
62: The mixing ratio $q(\vec{x},t)$ of a scalar tracer subject to advection, diffusion and condensation is
63: governed by
64: 
65: \begin{equation}
66: \frac{\partial q}{\partial t} + (\vec{u} \cdot \nabla)q = \eta \nabla^2 q + S(q,q_s) + F
67: \label{1}
68: \end{equation}
69: here $\vec{u}$ and $F$ are the advecting velocity field 
70: and the forcing respectively and $\eta$ is the diffusivity of the condensable substance.
71: In the present work we shall assume the domain to be spatially bi-periodic and restrict our attention to the
72: nondiffusive limit, $\eta=0$.
73: The problem thus consists of the advection of a passive tracer supplied by the source $F$, and removed by a sink
74: associated with condensation.
75: This sink 
76: is represented as
77: 
78: \begin{eqnarray}
79: S = -\frac{1}{\tau}[q - q_s(\vec{x})]  \quad \textrm{if} \quad q>q_s \nonumber \\
80: = 0 \quad \textrm{if} \quad q \le q_s
81: \label{2}
82: \end{eqnarray}
83: where $q_s(\vec{x})$ is the saturation mixing ratio --- a prescribed function. 
84: Hence with $\eta=0$, the ADC problem is closely related to the interaction of smooth advection with linear damping \cite{Chertkov1}. 
85: The ADC problem also bears a similarity recent studies of "active" processes when coupled with 
86: fluid advection, such as 
87: the evolution of chemically \cite{Neufeld} or 
88: biologically \cite{Abr} 
89: active substances and the issue of phase separation in immiscible fluids \cite{Bert}. \\ 
90: 
91: Our aim is to examine the
92: state of statistical equilibrium attained by (\ref{1}) and (\ref{2})
93: in the limit $\tau \rightarrow 0$ or that of {\it rapid condensation}. 
94: This limit implies that
95: upon advection if $q(\vec{x},t) > q_s(\vec{x})$, then the parcel's mixing ratio is instantaneously reset
96: to the saturation mixing ratio at that location. Of course, two immediately
97: apparent opposing long time limits of
98: (\ref{1}) and (\ref{2}) are : (i) Uniform forcing with $u=0 
99: \Rightarrow q(\vec{x},t) \rightarrow q_s(\vec{x})$ and (ii) A sufficiently mixing flow with
100: $F=0 \Rightarrow q(\vec{x},t) \rightarrow \min_{\forall \vec{x}} [q_s(\vec{x})]$. Hence, to
101: achieve a non-trivial state of statistical equilibrium we require both $\vec{u},F \neq 0$. \\
102: 
103: In the absence of diabatic effects, parcel motion in the midlatitude troposphere 
104: is restricted to two-dimensional surfaces of constant potential temperature, i.e. to 
105: isentropic surfaces \cite{Hoskins}.
106: Our objective is to understand the probability density function (\pdf{}) of the water vapor mixing ratio
107: along these midlatitude isentropic surfaces \cite{Ray-4}. 
108: On average the thermal structure of the midlatitude troposphere is such that as one
109: progresses polewards, the temperature
110: along isentropic surfaces decreases in a fairly linear manner. 
111: Given the sensitivity of the saturation mixing ratio (via the Clausius-Clapeyron  
112: relation) to the temperature, following \cite{Ray-2}
113: we take $q_s$ to vary exponentially with $y$. As we are in a periodic 
114: domain,
115: $q_s(\vec{x})=q_s(y)=\exp(-\alpha |y-\frac{L}{2}|)$ ($0 \le y \le L$) where $\alpha > 0$ and 
116: increasing $\alpha$ yields progressively steeper profiles which fall 
117: off symetrically from $y=\frac{L}{2}$. This yields an idealized model problem which retains
118: the essence of the much more complex atmospheric problem that motivates our work. Even though
119: the real atmosphere does not conform exactly to the idealizations, progress can be made through
120: a detailed solution of this model problem.\\
121: 
122: \section{The PDF in the limit of rapid condensation}
123: 
124: \subsection{General formulation and boundary conditions} 
125: 
126: Before we proceed to the \pdf{} in the general case,
127: consider the limit alluded to in the Introduction. Specifically,
128: $\vec{u}=0$ with constant forcing. On one hand this case is quite straightforward but on the other it
129: exhibits some pathologies inherent in dealing with the \pdf{}. From (\ref{1}) at a location $x_0,y_0$ for $t>
130: t_1=q_s(x_0,y_0)/F$ we have
131: 
132: \begin{equation}
133: q(x_0,y_0,t) = q_s(x_0,y_0) + F \tau [ 1 - \exp(\frac{-(t-t_1)}{\tau}) ] 
134: \label{e1}
135: \end{equation}
136: As (\ref{e1}) shows, in general the domain will become supersaturated. But for rapid condensation, as claimed,
137: $q(x_0,y_0,t) \rightarrow q_s(x_0,y_0) ~\forall (x_0,y_0)$. Note that the limit
138: of saturation is approached from above. Physically this implies that as $\tau \rightarrow 0$, we simultaneously have 
139: $q(x,y,t) \rightarrow q_s(x,y)$ such that the condensation sink balances the forcing. \\
140: 
141: To estimate the \pdf{}
142: (denoted by $P=P(x,y,q,t)$), we decompose the stationary
143: solution as $P(x,y,q)=P_1(x,y) F_1(q|x,y)$ where $F_1(q|x,y)$ is the conditional \pdf{} 
144: of $q$ conditioned on $(x,y)$. From the preceeding discussion, $F_1(q|x,y)=\delta[q - q_s(x,y)]$. As $\vec{u}=0$, 
145: for an unbiased estimate it is natural to take $P_1(x,y)$ to be uniform resulting 
146: in $P(x,y,q)=\delta[q - q_s(x,y)]$. In essence 
147: we end up with a $\delta$ function supported on the fixed point of the dynamical system (\ref{1}) with 
148: $\vec{u}=0$ \cite{Cycle}. In this particular case, the system does not have any other invariant \pdf{} and
149: the $\delta$ function \pdf{} is meaningful (as will be seen later), but in general to avoid such complications we restrict
150: our attention to functions that are reasonably smooth. \\
151: 
152: Proceeding to the general case 
153: \footnote{We are grateful to the referee who pointed out problems in our original formulation of the
154: \pdf{} equation. Indeed, this section would be much weaker if not for his comments.}, the 
155: Liouville or transport equation satisfied by the \pdf{} is
156: 
157: \begin{eqnarray}
158: \frac{\partial P(x,y,q,t)}{\partial t} + \frac{\partial (u_iP)}{\partial x_i} +
159: \frac{\partial[(F+S)P]}{\partial q} = 0 \nonumber \\
160: \textrm{with}~\int_{\vec{x}} {\int_{0}}^{\infty} P(x,y,q,t) ~d\vec{x}~dq = 1 ~ \forall t
161: \label{3a}
162: \end{eqnarray}
163: as in general, i.e. outside the limit of rapid condensation, $0 \le q(x,y,t) \le \infty$ and apart from periodicity in $\vec{x}$
164: we require $P(x,y,\infty,t) = 0$. \\
165: 
166: In $(x,y,q)$ space the above can be viewed as resulting from a 
167: {\it probability current} $\vec{U} P$ where $\vec{U} = [u,v, (F+S) ]$. An immediate consequence  
168: of rapid condensation is a restriction on $q$ i.e. $q(x,y,t) \le q_s(x,y)$. Till now our choice of forcing has been 
169: completely arbitrary, at this stage we restrict ourselves to a forcing which is positive definite and 
170: also satisfies $\frac{\partial F}{\partial q}=0$.
171: This implies, whatever the initial 
172: condition on $q$ is, after a finite time $\min{(q)}=\min{(q_s)}$. Hence, $\min{(q_s)} \le q(x,y,t) \le q_s(x,y)$ and
173: using (\ref{2}), (\ref{3a}) reduces to
174: 
175: \begin{eqnarray}
176: \frac{\partial P(x,y,q,t)}{\partial t} + \frac{\partial (u_iP)}{\partial x_i} +
177: \frac{\partial(FP)}{\partial q} = 0 \nonumber \\
178: \textrm{with}~\int_{\vec{x}} {\int_{\min{(q_s)}}}^{q_s(x,y)} P(x,y,q,t) ~d\vec{x}~dq = 1 ~ \forall t
179: \label{3b}
180: \end{eqnarray}
181: and, as noted, apart from the restriction on the forcing function, we require $P$ to be relatively smooth. 
182: Also, note that now the domain is periodic in $(x,y)$ and is bounded by the surfaces $q=\min{(q_s)}$ and
183: $q = q_s(x,y)$. Further, periodicity and the imposition of rapid condensation imply $[\vec{U}P]|_{\partial \bf{D}} = 0$
184: and in essence (\ref{3b}) represents the evolution of a \pdf{} via an incompressible flow within a bounded
185: domain with the normalization constraint implicity providing the required boundary condition.
186: Interestingly, similar integral
187: formulations for the
188: PDF arise in studies of spiking neurons (see for example Fusi \& Mattia \cite{Fusi} and the references
189: therein). \\
190: 
191: \subsection{Solutions for present saturation profile}
192: 
193: In the present case, as the saturation mixing ratio is purely a function of $y$, if the forcing is also taken to be 
194: of the form $F=F(y)$ (in fact we will focus on the uniformly forced case) ---
195: it is reasonable to look for solutions which
196: are independent of $x$. Of course, it is the realization of statistical equilibrium
197: without the presence of
198: gradient fields (as $\eta =0$) that makes
199: progress possible in the present case \footnote{A similar absence of gradient 
200: fields in equilibrium has been indirectly exploited in
201: the study of advection with linear damping \cite{Chertkov1}.}. This circumvents the need 
202: to estimate conditional expectations
203: of the scalar dissipation (or diffusion) which complicate advection-diffusion problems \cite{P},\cite{Dopazo-rev},
204: \cite{jai1}. \\ 
205: 
206: \subsubsection{The uniform case}
207: 
208: Inspecting (\ref{3b}) it is evident that without any further assumptions, as per the usual situation involving 
209: the evolution of a \pdf{} via an incompressible flow, a valid stationary solution is that of a uniform distribution in $(y,q)$. 
210: To obtain the $\pdf{}$ of $q$ from this uniform distribution we estimate
211: the probability that $q<Q$ as a function of $Q$
212: (i.e. $Q$ represents the sample space variable corresponding
213: to $q$).
214: Utilizing the rapid condensation normalization constraint,
215: $P(q,y)$ has to be integrated with respect to the
216: saturation curve $q_s(y)$. Therefore
217: 
218: \begin{equation}
219: Pr(q<Q) = \frac{I_1+I_2}{I_t}
220: \label{7a}
221: \end{equation}
222: Here $I_1,I_2$ correspond to the regions
223: $y \le Y$ and $y > Y$ where $Y=Z(Q)$ ($Z$ being the inverse of $q_s(y)$). In the present case, as $q_s$ is symmetric
224: about $y=\frac{L}{2}$ we need only consider
225: half of the domain ($L/2 \le y \le L$)
226: with $q_s=\exp(-\alpha y)$, i.e. $Z(Q) = -\log(Q)/\alpha$. Specifically,
227: 
228: \begin{eqnarray}
229: I_1 = {\int_{L/2}}^{Z(Q)} {\int_{Q_2}}^{Q}~ P(q,y) ~dqdy \nonumber \\
230: I_2 = {\int_{Z(Q)}}^{L} {\int_{Q_2}}^{q_s(y)}~ P(q,y) ~dqdy
231: \label{7b}
232: \end{eqnarray}
233: where $Q_2=\min(q_s)$. $I_t$ in (\ref{7a}) is the same as $I_2$ but with $L/2$ as the lower limit of integration in the outer
234: integral, which ofcourse is nothing but the normalization constraint in (\ref{3b}). 
235: Hence,
236: 
237: \begin{equation}
238: \pdf{}(Q) = \frac{d (I_1+I_2)}{dQ}  
239: \label{11aa}
240: \end{equation}
241: Substituting $P(y,q)=$ const. in (\ref{7b}) and (\ref{11aa}) yields $\pdf{}(Q) \sim Z(Q)$ i.e. 
242: $\pdf{}(Q) \sim \log(Q^{\frac{-1}{\alpha}})$ which can be seen in the upper panel of Fig. (\ref{fignew1}). 
243: In essence, for a uniform distribution the constraint of rapid condensation forces the \pdf{} to reflect the 
244: saturation profile. \\
245: 
246: \subsubsection{The eddy-diffusion case}
247: 
248: We now assume that the the effect of the fluctuating velocity in the Liouville equation admits a diffusive representation
249:  --- say by an eddy diffusivity $\kappa_e$.  This would be the case if the parcel trajectories consisted of
250: independent Brownian motion, for example. 
251: Admittedly, this is a fairly severe assumption in that mixing by multiple scale 
252: velocity fields rarely follows a simple diffusive prescription \cite{MK}. But from 
253: a tropospheric viewpoint, it is known that the meridional ($y$ - direction) Lagrangian eddy-velocity
254: correlation function decays quite rapidly (on the order of a few days) \cite{jai3} --- hence, in this
255: context the assumption
256: of an eddy diffusivity may not completely unreasonable.  Note that the eddy diffusivity  we refer to here is
257: an eddy diffusivity applied to probability evolution in $(y,q,t)$ space.  This is not the same as characterizing
258: mixing by an eddy diffusivity to an evolution equation for a coarse-grained $q$ in $(y,t)$ space.  Indeed,
259: in \cite{Ray-2} it is shown that the Brownian model yields different coarse-grained $q$ statistics than the
260: mean-field diffusivity model; it is suggested further that the Brownian motion model constitutes a minimal model
261: for investigation of the interplay of transport processes with a nonlinear sink term such as condensation. In that
262: sense, the Brownian model is worth of study in and of itself. \\
263: 
264: With the eddy-diffusivity (\ref{3b}) yields 
265: 
266: \begin{equation}
267: \frac{\partial P}{\partial t} -\kappa_e \frac{\partial^2 P}{\partial y^2} + F(y) \frac{\partial P}{\partial q} = 0
268: \label{5}
269: \end{equation}
270: For uniform forcing $F(y)=\delta$ (a constant), as $P > 0$ and is periodic in $y$, by inspection the ansatz 
271: %substituting a separable form i.e. $P(y,q,t)=A(q)~B(y)~C(t) + C_0$ where $C_0$ is a constant, in (\ref{5}) gives
272: 
273: %\begin{equation}
274: %\frac{d \log(A(q))}{dq} = \frac{1}{\delta} \frac{d \log(C(t))}{dt} = \frac{\kappa_e}{\delta} \frac{1}{B(y)} \frac{d^2 B}{dy^2} = \pm k^2 
275: %\label{6}
276: %\end{equation}
277: %Choosing the separation constant to 
278: %be $-k^2$ as $P(q,y,t)$ is periodic in $y$, at long times (though before achieving uniformity = $C_0$) the mode with 
279: %smallest decay rate remains and
280: 
281: \begin{equation}
282: P(q,y) \sim \exp(-k^2 q)[ \cos(\lambda y)+  \sin(\lambda y)]  ~;~ \textrm{where} ~\lambda=\frac{1}{4},k^2=\lambda ^2 \frac{\kappa_e}{\delta}
283: \label{7}
284: \end{equation}
285: satisfies a stationary form of (\ref{5}). Indeed, for a quantitative estimate we would need to fix the constants that 
286: arise in (\ref{7}) via the normalization constraint. 
287: But for a qualititative estimate substituting $P(q,y)$ in (\ref{7b}) and setting $dG/dy=[ \cos(\lambda y)+ \sin(\lambda y)]$ for 
288: notational simplicity, yields
289: 
290: %\begin{equation}
291: %I_1 = \frac{[-\exp(-k^2 Q) + \exp(-k^2 Q_2)]}{k^2} ~[G(Z(Q)) - G(0)]
292: %\label{8}
293: %\end{equation}
294: %Similarly,
295: %
296: %\begin{equation}
297: %I_2 = \frac{\exp(-k^2 Q_2) }{k^2} [G(L) - G(Z(Q)) ] - \frac{X}{k^2}
298: %\label{9}
299: %\end{equation}
300: %where
301: %
302: %\begin{equation}
303: %X = {\int_{Z(Q)}}^{L} ~\frac{dG(y)}{dy}~ \exp[-k^2 q_s(y)] ~dy
304: %\label{10}
305: %\end{equation}
306: %Hence, 
307: \begin{equation}
308: \pdf{}(Q) \sim ~\exp(-k^2Q) ~[G(Z(Q)) - G(0)] 
309: \label{11}
310: \end{equation}
311: Substituting for $G(y)$
312: 
313: \begin{equation}
314: %\pdf{}(Q) \sim \exp(-k^2Q)  ~\{ \sin[Z(Q)] -  \cos[ Z(Q)] + 1 \}
315: \pdf{}(Q) \sim \frac{ \exp(-k^2Q) }{\lambda} ~\{ \sin[\lambda Z(Q)] - \cos[\lambda Z(Q)] + 1 \}
316: \label{11a}
317: \end{equation}
318: and finally substituting for $Z(Q)$  
319: 
320: \begin{equation}
321: %\pdf{}(Q) \sim \exp(-k^2Q) ~\{\sin[\log(Q^{\frac{-1}{\alpha}})] - 
322: %\cos[\log(Q^{\frac{-1}{\alpha}})] + 1 \}
323: \pdf{}(Q) \sim ~ \frac{ \exp(-k^2Q) }{\lambda} ~\{\sin[\log(Q^{\frac{-\lambda}{\alpha}})] -
324: \cos[\log(Q^{\frac{-\lambda}{\alpha}})] + 1 \}
325: \label{11b}
326: \end{equation}
327: The dependence of the \pdf{} on $\frac{\kappa_e}{\delta}$ is shown in 
328: the lower panel of 
329: Fig. (\ref{fignew1}).
330: In all cases, 
331: the \pdf{} has a peak for small $q$ 
332: decreases exponentially for intermediate
333: values of $q$ (the slope increases with $\frac{\kappa_e}{\delta}$) and then rolls-off as $q \rightarrow \max{(q_s)}$. \\
334: 
335: Comparing this with the uniform case (upper panel of Fig. (\ref{fignew1})), it is evident that the 
336: form of the \pdf{} 
337: is still controlled by the saturation mixing ratio profile. But the details, such as the
338: slope of the \pdf{} for intermediate values of $Q$, are dependent on the eddy-diffusivity and 
339: forcing. Also, it is 
340: worth noting that the \pdf{}  
341: is in marked contrast to what one would obtain for a fully saturated domain. 
342: For
343: example in
344: the saturated case we had $P(q,y)=\delta(q-q_s(y))$, this yields
345: $\pdf{}(Q_{\textrm{sat}})
346: \sim - dF(Q)/dQ$ which implies a power law ($Q^{-1}$, with no "roll-off" for high $Q$) when $q_s=\exp(-\alpha y)$. 
347: At first sight the roll-off in (\ref{11b}) appears to be an obvious result of enforcing rapid 
348: condensation, but note that the saturated case considered above also conforms with the rapid condensation 
349: limit yet it yields a very different \pdf{}. This is illustrated more clearly by choosing $q_s(y)$ to be a 
350: linearly decreasing function of $y$, now the saturated \pdf{} is uniform, whereas 
351: (\ref{11a}) again yields a slowly decaying \pdf{} with a smooth roll-off at large $q$. \\
352: 
353: With regards to water vapor, given the logarithmic dependence of the infrared cooling on the water vapor mixing ratio,
354: it follows that fluctuations in the mixing ratio actually increase infrared cooling \cite{Ray-2}. 
355: Hence a slowly decaying \pdf{}, by increasing the
356: probablity of encountering a large fluctuation (as compared to a normal \pdf{}), increases the efficiency of radiative cooling to
357: space. In this regard, the rapid roll-off of the tail of the \pdf{} could play a strong role in that it 
358: sharply decreases the possibility of very 
359: large fluctuations --- for example, in comparison to a advection-diffusion model of water vapor distribution where one 
360: would encounter
361: exponential tails under homogeneous forcing \cite{Chert} --- 
362: and would imply a reduction in the infrared
363: cooling or in other words an increase in the temperature of the surface to maintain energy balance. 
364: Quantifying this effect by using a full-fledged atmospheric radiation code is a project we hope to pursue in the near future. \\
365: 
366: \section{Numerical Simulations}
367: 
368: A primary assumption in our derivation was the use of an "eddy diffusivity" to represent the effect of 
369: an advecting velocity field. As mentioned, in most flows of interest the velocity field is expected to be quite 
370: coherent and might not conform to a diffusive representation. To test the stringency of this assumption
371: we numerically 
372: simulate the ADC system by employing a lattice map for 
373: purposes of smooth large scale advective mixing \cite{Ray},\cite{jai2}. The velocity field is
374: 
375: \begin{eqnarray}
376: u(x,y,t) = f(t)~A_1 ~ \textrm{sin}(B_1 y + p_n) \quad  \nonumber \\
377: v(x,y,t) = (1-f(t))~A_2 ~ \textrm{sin}(B_2 x + q_n)  \nonumber \\
378: \label{1m}
379: \end{eqnarray}
380: where $A_1=0.75,A_2=0.5,B_1=B_2=2.5$, $f(t)$ is 1 for $nT \le t < (n+1)T/2$ and 0 for $(n+1)T/2 \le t < (n+1)T$. 
381: $p_n , q_n$ ($\in [0,2\pi]$) are random numbers selected
382: at the beginning of each iteration, i.e., for each period $T$. Advection is implemented via sequential 
383: integer shifts in the $x$ and $y$ directions on a square lattice (see \cite{Ray} for details).
384: Apart from its numerical efficiency the lattice map has the advantage of
385: preserving moments, i.e. it does not introduce spurious diffusion into the problem. \\
386: %Opposing initial conditions, $q(x,y,0)=\min{(q_s)}$ and $q(x,y,0)=q_s(y)$, are used 
387: %to start the simulations with a uniform forcing applied at every iteration. 
388: %Fig. (\ref{fig2}) shows the ratio of the total substance being condensed out at 
389: %every step to the forcing and the spatial average of the condensable field as a function of the 
390: %iteration. As is seen both initial contitions settle into the same state of equilibrium. \\
391: 
392: To test the features of the \pdf{} seen in Fig. (\ref{fignew1}), we set $\delta=\min{(q_s)}/\Delta t$. 
393: Note that 
394: as the advective map has no time scale, we set $\Delta t= 1$ to match an iteration of the mapping.
395: Numerically this amounts to incrementing the mixing ratio at every site on the lattice by 
396: $\min{(q_s)}$ at every iteration. Now as $\min{(q_s)}$ is a function of $\alpha$, $\delta$ decreases with increasing $\alpha$.
397: Interestingly, we observe that even though the maximum attainable mixing ratio is
398: $\max{(q_s)}$, the system achieves equilibrium for much smaller maxima in $q$ field. 
399: A snapshot of the equilibrium condensable field for $\alpha=1,1.25$ and $1.5$ is shown in 
400: Fig. (\ref{fignew22}) --- note the increase in "graininess" of the field with $\alpha$. 
401: The \pdf{}'s for these cases can be seen in 
402: Fig. (\ref{fignew3}) 
403: --- the plotted \pdf{}'s
404: are averages over the last couple of iterations of the map.
405: In spite of the non-local nature of the mixing protocol the shapes of
406: the \pdf{}'s display the features anticipated from the eddy-diffusive case, in particular we see the decrease in 
407: the slope of the \pdf{} with $\delta$ (the advective map is unchanged so $\kappa_e$ is the same in all the simulations)
408: also the characteristic roll-off induced via the form of the saturation profile in conjunction with the 
409: limit of rapid condensation is evident. \\
410: 
411: Another measure of quantifying intermittency in a field is to examine its structure functions, defined as 
412: $S_n(|\vec{r}|) = < |q(\vec{x}+\vec{r})-q(\vec{x})|^n >$ (in the present situation $<\cdot>$ denotes a spatial average) 
413: higher moments (i.e. larger $n$) of the structure functions are most sensitive to 
414: the "roughest" regions of the 
415: field. 
416: For the condensable field, we observe that $S_n(r) \sim r^{\zeta_n}$ for $l_1 \le r \le l_2$ where $l_1 \rightarrow 0$ as 
417: we are dealing with a non-diffusive problem and $l_2$ is an outer 
418: scale which is smaller than the 
419: size of the domain. More to the point, the exponents $\zeta_n$ are anomalous in 
420: the sense $\zeta_n < n \zeta_1$ for $n>1$. In fact, we observe near saturation of the scaling exponents (i.e. $\zeta_n$ tends to 
421: a constant) for large $n$. Both the scaling of the structure functions and the extracted scaling exponents are 
422: shown in Fig. (\ref{fig_s}). 
423: The anomalous behaviour is physically anticipated as smooth advection by itself leads to the formation of 
424: sharp fronts in finite time --- hence 
425: one has a field composed of smooth regions
426: interrupted by step like discontinuities. Indeed the combination of these two structures yields a $\zeta_n$ profile which
427: increases linearly for $n<=1$ and then saturates to a constant (i.e. extreme intermittency) for $n>1$ \cite{Aurell}. 
428: As the additional presence of condensation 
429: does not involve any
430: smoothening of the field, we do not expect it to be able to mollify the discontinuities created via advection.
431: But rapid condensation provides a bound for the magnitude of the jump across the 
432: advective discontinuity, i.e. $|q(x+r)-q(x)| \le 1-\exp(-\alpha L/2)$, hence we expect a weak dependence of 
433: $\zeta_n$ on $\alpha$. 
434: Indeed, this qualitative reasoning is bourne out in the lower panel Fig. (\ref{fig_s}).
435: In a similar vein, it has been shown 
436: (under further assumptions regarding the nature of the velocity field) 
437: that an interplay of advection with linear damping produces severe anomalous scaling \cite{Chertkov1}.
438: Interestingly, such anomalous behaviour --- with saturation of $\zeta_n$ for large $n$ --- 
439: has been observed in an analysis of 
440: specific humidity fluctuations from high resolution aircraft data in the troposphere \cite{cho}. \\
441: 
442: %Further, as is evident from (\ref{11b}) for a fixed flow (i.e. fixed $\kappa_e$) the mean of the
443: %condensable field is a function of the forcing strength. In fact, as is seen in Fig. (\ref{fig2a}) 
444: %--- which shows the numerically computed mean as well as the mean using the first mode of (\ref{11b}) ---
445: %the mean of the condensable field increases with 
446: %the forcing strength until $\delta = \max_{\forall \vec{x}} [q_s(\vec{x})]/\Delta t$ when the entire
447: %domain becomes saturated. \\
448: 
449: \section{Conclusions and Atmospheric Data}
450: 
451: We have examined the state of equilibrium 
452: achieved via the interplay of smooth advection and condensation.
453: Exploiting the attainment of equilibrium without the presence of gradient fields allows
454: us to make progress on the equation governing the \pdf{} of the condensable substance. 
455: In particular, the limit of rapid condensation admits a simplified Liouville equation 
456: with an integral normalization constraint on the \pdf{}. Assuming a straightforward 
457: uniform solution to the governing equation shows the \pdf{} to be tied to the particular
458: saturation profile under consideration. Relaxing this uniformity and assuming an 
459: eddy-diffusive nature for the advective velocity fields shows the form of the \pdf{} to still
460: be controlled by the saturation profile but details, such as the rate of decay, to be 
461: sensetive to the eddy-diffusivity and strength of the forcing. Indeed, numerical simulations employing a 
462: lattice map for advective purposes reproduce these features.
463: Further, scaling exponents extracted from 
464: structure functions of the condensable field show anomalous behaviour. Physically, 
465: the anomalous scaling is 
466: anticipated given the tendency of undiffused smooth
467: advection to create sharp fronts. In fact, the near saturation of $\zeta_n$ for large $n$ is a reflection of these
468: fronts providing the leading contribution to the higher order structure functions. \\
469: 
470: As mentioned in the Introduction, our motivation is to understand the distribution of water vapor 
471: along midlatitude isentropic surfaces. With this in mind we construct \pdf{}'s of the 
472: midlatitudinal specific humidity
473: field along the 300 K
474: isentrope using data from the ECMWF reanalysis (ERA40) project. As is seen in Fig. (\ref{fig4}),
475: despite the simplifying assumptions made in our derivation,
476: the \pdf{}'s from data can be approximated by the eddy-diffusive \pdf{}. In fact, in
477: Fig. (\ref{fig4}) we've also plotted the \pdf{} that results from assuming uniformity (for the same 
478: $\alpha$)
479: and as is seen even though the primary shape of the data \pdf{} follows from the saturation profile in
480: conjunction with rapid condensation, it is the enhanced slope via $\frac{\kappa_e}{\delta}$ that captures
481: the decay of the \pdf{}.
482: Note that for large $q$ the \pdf{}'s from 
483: data deviate significantly from the eddy-diffusive estimate. 
484: As we have considered the nondiffusive limit a possible 
485: source of this discrepancy might lie in the homogenization induced by diffusion. Secondly, as 
486: the source of water vapor lies in the tropics a boundary forcing might be more appropriate for
487: the actual atmospheric problem.
488: Even so, these moderately encouraging results lead us to conjecture that
489: the ADC model --- in the limit of rapid condensation with the proper saturation profile --- driven by idealized velocity fields
490: might be of use in predicting the statistical properties of the large scale distribution of water vapor
491: in the midlatitude troposphere. \\
492: 
493: \acknowledgments
494: 
495: We thank Prof. W.R. Young (Scripps Institute, UCSD) for his suggestions 
496: which led to a clearer formulation of the problem.
497: J.S. would also like to acknowledge helpful conversations with Dr. A. Alexakis (NCAR). The comments of the
498: anonymous referee are gratefully acknowledged, in particular they distinctly improved the formulation of the \pdf{} equation. We also thank the referee for pointing out the resemblance to the
499: spiking neuron problem.
500: Much of this work was carried out while the first author was at the National Center for Atmospheric Research
501: which is sponsored by the National Science Foundation.  The second author's contribution
502: to this work was sponsored by the National Science Foundation under grant ATM-0123999.
503: 
504: 
505: \begin{thebibliography}{99}
506: 
507: \bibitem{Falk} G. Falkovich, K. Gawedzki and M. Vergassola, 
508: Rev. Mod. Physics, {\bf 73}, 913 (2001).
509: 
510: \bibitem{Sherwood} S.C. Sherwood,
511: J. Climate, {\bf 9}, 2919 (1996).
512: 
513: \bibitem{SH} E.P. Salathe and D.L. Hartmann,
514: J. Climate, {\bf 10}, 2533 (1997).
515: 
516: \bibitem{Ray-grl} R.T. Pierrehumbert, Geophys. Res. Lett. {\bf 25}, 151 (1998).
517: 
518: \bibitem{Ray-Nature} R.T. Pierrehumbert, Nature, {\bf 419}, 191 (2002).
519: 
520: \bibitem{HS} I. Held and B. Soden, 
521: Annu. Rev. Energy Environ. {\bf 25}, 441 (2000).
522: 
523: \bibitem{Ray-2} R.T. Pierrehumbert, H. Brogniez and R. Roca; 
524: in {\it The General Circulation of the Atmosphere}, edited by T. Schneider and A. Sobel (Princeton
525: University Press, 2005) (to appear)
526: 
527: \bibitem{Chertkov1} M. Chertkov, Phys. of Fluids,
528: {\bf 10}, 3017 (1998).
529: 
530: \bibitem{Neufeld} Z. Neufeld, C. Lopez and P.H. Haynes, Phys. Rev. Lett. {\bf 82}, 2606 (1999).
531: 
532: \bibitem{Abr} E.R. Abraham, Nature, {\bf 391}, 577 (1998).
533: 
534: \bibitem{Bert} L. Berthier, J-L. Barrat and J. Kurchan, 
535: Phys. Rev. Lett. {\bf 86}, 2014 (2001).
536: 
537: \bibitem{Hoskins} B. Hoskins, 
538: Tellus, {\bf 43A}, 27 (1991).
539: 
540: \bibitem{Ray-4} H. Yang and R.T. Pierrehumbert, 
541: J. Atmos. Sci. {\bf 51}, 3437 (1994).
542: 
543: \bibitem{Cycle} P. Cvitanovi\'c, R. Artuso, R. Mainieri, G. Tanner and G. Vattay,
544: {\em Chaos: Classical and Quantum},
545: {\tt ChaosBook.org} Niels Bohr Institute, Copenhagen (2005).
546: 
547: \bibitem{Fusi} S. Fusi and M. Mattia,
548: Neural Computation {\bf 11}, 633 (1999).
549: 
550: \bibitem{P} S.B. Pope, Prog. Energy Combust. Sci. {\bf 11}, 119 (1985). 
551: 
552: \bibitem{Dopazo-rev} C. Dopazo, L. Valino and 
553: N. Fueyo, Int. Journal of Mod. Physics B, {\bf 11}, 2975 (1997); 
554: 
555: \bibitem{jai1} J. Sukhatme, Phys. Rev. E,
556: {\bf 69}, 056302 (2004).
557: 
558: %\bibitem{Gaspard} P. Gaspard, G. Nocolis, A. Provata and S. Tasaki, Phys. Rev. E, {\bf 51}, 74 (1995). 
559: 
560: \bibitem{MK} A.J. Majda and P.R. Kramer, Physics Reports, {\bf 314}, 238 (1999). 
561: 
562: \bibitem{jai3} J. Sukhatme, J. Atmos. Sci. {\bf 62}, 3831 (2005).
563: 
564: %\bibitem{Risken} H. Risken, {\em The Fokker-Planck Equation : Methods of Solutions and Applications}, Springer
565: %Series in Synergetics, 2nd ed. 1996.
566: 
567: \bibitem{Chert} M. Chertkov, G. Falkovich, I. Kolokolov and V. Lebedev, Phys. Rev. E, {\bf 51}, 5609 (1995).
568: 
569: \bibitem{Ray} R.T. Pierrehumbert, 
570: Chaos, {\bf 10}, 1, 61 (2000). 
571: 
572: \bibitem{jai2} J. Sukhatme and R.T. Pierrehumbert, Phys. Rev. E, {\bf 66}, 056302 (2002).
573: 
574: \bibitem{Aurell} E. Aurell E, U. Frisch, J. Lusko and M. Vergassola, J. Fluid. Mech., {\bf 238}, 467 (1992).
575: 
576: \bibitem{cho} J. Cho, R. Newell and G. Sachse, Geophys. Res. Lett. {\bf 27}, 377 (2000).
577: 
578: 
579: \end{thebibliography}
580: 
581: \clearpage
582: 
583: \begin{figure}
584: \includegraphics[width=7.5cm,height=9.5cm]{figure1}% Here is how to import EPS art
585: \caption{\label{fignew1} Upper Panel : The \pdf{} for the uniform case showing the dependence on $\alpha$. Lower Panel : The \pdf{} 
586: in (\ref{11b}) as a function of $\frac{\kappa_e}{\delta}$ with $\alpha=3$.}
587: \end{figure}
588: 
589: \begin{figure}
590: \includegraphics[width=7.5cm,height=9cm]{figure2}% Here is how to import EPS art
591: \caption{\label{fignew22} A snapshot of the condensable field under the action of the 
592: map specified in (\ref{1m}). 
593: The snapshots are after the \pdf{} has settled into an invariant shape. The upper,middle and lower panels 
594: correspond to
595: $\alpha=1,1.25 $ and $1.5$ respectively with $\delta=\min{(q_s)}$.}
596: \end{figure}
597: 
598: \begin{figure}
599: \includegraphics[width=7.5cm,height=7.5cm]{figure3}% Here is how to import EPS art
600: \caption{\label{fignew3} The \pdf{}'s of the condensable field for $\alpha=1,1.25 $ and $1.5$.
601: The curves have been shifted for clarity. Note, as per Fig. (\ref{fignew1}), as $\alpha$ increases 
602: $\delta$ decreases resulting in an increase in the slope 
603: (as $\kappa_e$ is the same in every case) of the \pdf{}. Also notice the 
604: increase in the sharpness of the peak with $\alpha$.}
605: \end{figure}
606: 
607: \begin{figure}
608: \includegraphics[width=7.5cm,height=7.5cm]{figure4}% Here is how to import EPS art
609: \caption{\label{fig_s} Upper Panel : Plots of $\log(S_q(r))$ Vs. $\log(r)$ for the first six
610: moments of the equilibrium condensable field with $\alpha =2$. 
611: Lower panel : Scaling exponents extracted from above (for $\alpha=2$) and similar plots (not shown) for 
612: $\alpha=1,1.5$.} 
613: \end{figure}
614: 
615: \begin{figure}
616: \includegraphics[width=7.5cm,height=7.5cm]{figure5}% Here is how to import EPS art
617: \caption{\label{fig4} The normalized specific humidity \pdf{}'s in the midlatitudes (between $30^{\circ}$
618: and $60^{\circ}$ in both hemispheres) along the 
619: 300 K isentropic surface
620: for Jan 97-01 from ECMWF data. The solid line with dots is a plot of (\ref{11b}) (shifted vertically by a constant) 
621: with $\frac{\kappa_e}{\delta} = 45,
622: \alpha =3$ and for comparison the dash-dash line is a plot of the \pdf{} resulting from uniformity, again with $\alpha=3$.}
623: \end{figure}
624: 
625: 
626: \end{document}
627: