0804.2805/OETD.tex
1: \documentclass[12pt,tightenlines,showkeys]{revtex4}
2: \pdfoutput=1
3: 
4: \usepackage{graphicx}
5: \usepackage{amsmath}
6: \usepackage{amssymb}
7: \usepackage{bm}
8: \usepackage{color}
9: \usepackage[loose]{subfigure}
10: 
11: %\graphicspath{{figs/}{figs_hi/}}
12: \graphicspath{{figs/}{figs_lo/}}
13: 
14: \pagestyle{plain}
15: 
16: %%%%%TEXT START%%%%%
17: \begin{document}
18: 
19: \title{Mixing effectiveness depends on the source-sink structure:
20:   Simulation results}
21: 
22: \author{Takahide Okabe}
23: \affiliation{Department of Physics and Institute for Fusion Studies,
24:   The University of Texas at Austin, Austin, TX 78712-0264, USA}
25: 
26: \author{Bruno Eckhardt}
27: \affiliation{Fachbereich Physik Philipps-Universtit\"at, D-35032
28:   Marburg, Germany}
29: 
30: \author{Jean-Luc Thiffeault}        
31: \affiliation{Department of Mathematics, University of Wisconsin,
32:   Madison, WI 53706-1388, USA}
33: 
34: \author{Charles R. Doering}
35: \affiliation{Departments of Mathematics and Physics, University of
36:   Michigan, Ann Arbor, MI 48109-1043, USA}
37: 
38: \date{\today}
39: 
40: \begin{abstract}
41:   The mixing effectiveness, i.e., the enhancement of molecular diffusion,
42:   of a flow can be quantified in terms of the suppression of concentration
43:   variance of a passive scalar sustained by steady sources and sinks.
44:   The mixing enhancement defined this way is the ratio of the RMS  
45:   fluctuations of the scalar mixed by molecular diffusion alone to
46:   the (statistically steady-state) RMS fluctuations of the scalar density
47:   in the presence of stirring.  This measure of the
48:   effectiveness of the stirring is naturally related to the
49:   enhancement factor of the equivalent eddy diffusivity over molecular
50:   diffusion, and depends on the P\'eclet number. 
51:   It was recently noted that the maximum possible
52:   mixing enhancement at a given P\'eclet number depends as well on the
53:   structure of the sources and sinks.  That is, the mixing efficiency, the
54:   effective diffusivity, or the eddy diffusion of a flow generally depends
55:   on the sources and sinks of whatever is being stirred.  
56:   Here we present the results of particle-based simulations quantitatively
57:   confirming the source-sink dependence of
58:   the mixing enhancement as a function of P\'eclet number for a model
59:   flow.
60: \end{abstract}
61: 
62: \keywords{
63:     stirring \& mixing;
64:     transport processes;
65:     stochastic particle dynamics;
66:     Brownian motion;
67:     turbulence}
68: 
69: \maketitle
70: 
71: \section{Introduction}
72: 
73: Mixing by fluid flows is a ubiquitous natural phenomenon that plays a
74: central role in many of the applied sciences and engineering.  A
75: geophysical example is the mixing of aerosols (e.g., $\text{CO}_2$
76: supplied by a volcano, say, or by human activity) in the atmosphere.
77: Aerosols are dispersed by molecular diffusion on the smallest scales
78: but are more effectively spread globally by atmospheric flows.  The
79: density --- and density fluctuations --- of some aerosols influence
80: the albedo of the earth and thus have a significant environmental
81: impact.  Hence it is important to understand fundamental properties of
82: dispersion, mixing, and the reduction of concentration fluctuations
83: by stirring flow fields.
84: 
85: Various aspects of mixing have been the
86: focus of many review articles~\cite{Ottino1990,Majda1999,Warhaft2000,%
87: Shraiman2000,Sawford2001,Falkovich2001,Aref2002,Wiggins2004}.
88: At the most basic level, the mixing of a passive scalar can be modeled
89: by an advection-diffusion equation for the scalar concentration field
90: with a specified stirring flow field.  In this work we will focus on
91: problems where fluctuations in the scalar field are generated and
92: sustained by temporally steady but spatially inhomogeneous sources.
93: The question of interest here is this: for a given source
94: distribution, how well can a specified stirring flow mix the scalar
95: field?  
96: 
97: Mixing effectiveness can be measured by the scalar variance over the domain.
98: A well-mixed scalar field will have a more uniform density with
99: relatively ``small'' variance while increased fluctuations in the scalar density
100: will be reflected in a ``large'' variance.  We put quotes around the
101: quantifiers small and large because the variance is a dimensional
102: quantity that needs an appropriate dimensional point of reference 
103: from which it is being measured.
104: 
105: Several years ago Thiffeault {\it et al} \cite{Thiffeault2004} introduced a
106: notion of ``mixing enhancement'' for a velocity field stirring a steadily
107: sustained scalar by comparing the bulk (space-time) averaged density
108: variance with and without advecting flow.  Mixing is accomplished by 
109: molecular diffusion alone in the absence of stirring, which can be
110: quite effective on small scales but is not generally so good at
111: breaking up and dispersing large-scale fluctuations quickly.  
112: Stirring can greatly enhance the transport of the scalar from regions
113: of excess density to depleted regions, suppressing the variance far
114: below its diffusion-only value.  The magnitude of this variance
115: suppression by the stirring --- the ratio of the variance without
116: stirring to the variance in the presence of stirring --- is a
117: dimensionless quantity that provides a sensible gauge of the mixing
118: effectiveness of the flow.  Different advection fields will have different
119: mixing efficiencies stirring scalars supplied by different sources.  It is
120: then of obvious interest both to determine theoretical limits on
121: mixing enhancements for various source configurations and to explore
122: whether those limits may be approached---or even perhaps achieved----for 
123: particular flows.
124: 
125: There have been many studies of stirring and mixing of a scalar with
126: fluctuations sustained by spatially inhomogeneous sources and sinks.  Some of the
127: earliest are by Townsend~\cite{Townsend1951,Townsend1954}, who was
128: concerned with the effect of turbulence and molecular diffusion on a
129: heated filament.  He found that the spatial localization of the source
130: enhanced the role of molecular diffusivity.  Durbin~\cite{Durbin1980}
131: and Drummond~\cite{Drummond1982} introduced stochastic particle models
132: to turbulence modeling, and these allowed more detailed studies of the
133: effect of the source on diffusion.  Sawford and
134: Hunt~\cite{Sawford1986} pointed out that small sources lead to a
135: dependence of the variance on molecular diffusivity.  These models
136: were further refined by~\cite{Thomson1990,Borgas1994,Sawford2001}.
137: Chertkov {\it et
138:   al}~\cite{Chertkov1995,Chertkov1995b,Chertkov1997,Chertkov1997b,%
139:   Chertkov1998} and Balkovsky \& Fouxon~\cite{Balkovsky1999}
140:   addressed the case of a random, statistically-steady source.
141: 
142: In this paper we study the enhancement of mixing by an advection field
143: using a particle-based computational scheme that is easy to implement
144: and applicable to a variety of source distributions.
145: The idea is to develop a method
146: that accurately simulates advection and diffusion of large numbers of
147: particles supplied by a steady source, and to measure density
148: fluctuations by ``binning'' the particles to produce
149: an approximation of the hydrodynamic concentration field.  Unlike a
150: numerical PDE code, a particle code does not prefer specific forms of
151: the flow or the source (PDE methods generally work best with 
152: very smooth fields).  
153: There is, however, no free lunch: the accuracy of the
154: particle code is ultimately limited by the finite number of particles
155: that can be tracked.  The limitation to finite numbers of particles
156: inevitably introduces statistical errors due to discrete fluctuations
157: in the local density and systematic errors in the variance
158: measurements due to binning.  But these problems are tractable, and as
159: we will show, the method proves to be quantitatively accurate and 
160: computationally efficient for some applications.
161: 
162: \section{Theoretical background}
163: 
164: In this section we review basic facts about the mixing enhancement
165: problem as formulated by Thiffeault, Doering \& Gibbon {\it et al}
166: \cite{Thiffeault2004} and developed by Plasting \& Young \cite{Plasting2006},
167: Doering \& Thiffeault \cite{DoeringThiffeault2006}, Shaw {\it et al} \cite{Shaw2007}, and
168: Thiffeault and Pavliotis \cite{Thiffeault2008}.  The dynamics is given
169: by the advection-diffusion equation for the concentration of a passive
170: scalar $\rho(t, \bm{x})$ with time-independent but spatially
171: inhomogeneous source field $S(\bm{x})$:
172: \begin{equation}
173: \frac{\partial\rho}{\partial t}+\bm{u}\cdot\nabla\rho=\kappa \Delta \rho + S(\bm{x}),
174: \label{ADE1}
175: \end{equation}
176: where $\kappa$ is the molecular diffusivity and $\bm{u}(t,
177: \bm{x})$ is a specified advection field that satisfies (at each
178: instant of time) the incompressibility condition
179: \begin{equation}
180: \nabla \cdot \bm{u} = 0.
181: \end{equation}
182: For simplicity, the domain is the $d$-torus, i.e., $[0, L]^d$ with
183: periodic boundary conditions.  
184: We limit attention to stirring
185: fields that satisfy the properties of statistical homogeneity and
186: isotropy in space defined by
187: \begin{equation}
188: \overline{u_{i}(\cdot, \bm{x})} = 0,\qquad
189: \overline{u_{i}(\cdot, \bm{x})u_{j}(\cdot, \bm{x})} = \frac{U^{2}}{d}\delta_{ij}
190: \end{equation}
191: where the overbar denotes time-averaging and $U$ is the root mean
192: square speed of the velocity field, a natural indicator of the
193: intensity of the stirring.  
194: These are statistical properties of
195: homogeneous isotropic turbulence on the torus,
196: but they are also shared by many other kinds of flows.
197: 
198: We are interested in fluctuations in the concentration $\rho$ so the
199: spatially averaged background density is irrelevant.  It is easy to
200: see from \eqref{ADE1} that the spatial average of $\rho$ grows
201: linearly with time at the rate given by the spatial average of $S$.
202: Hence we change variables to spatially mean-zero quantities
203: \begin{equation}
204:  \theta (t, \bm{x}) =  \rho(t, \bm{x}) - \frac{1}{L^d}\int \mathrm{d}^{d}x'\, \rho(t, \bm{x}')
205: \end{equation} 
206: and
207: \begin{equation}
208: s(\bm{x}) = S(\bm{x}) - \frac{1}{L^d}\int \mathrm{d}^{d}x'\, S(\bm{x}')
209: \end{equation} 
210: that satisfy
211: \begin{equation}
212: \frac{\partial\theta}{\partial t}+\bm{u}\cdot\nabla\theta=\kappa \Delta \theta + s(\bm{x}).
213: \label{ADE2}
214: \end{equation}
215: (We must also supply initial conditions for $\rho$ and/or $\theta$ but
216: they play no role in the long-time steady statistics that we are
217: interested in.)
218: 
219: The ``mixedness'' of the scalar may be characterized by, among other
220: quantities, the long-time averaged variance of $\rho$, proportional to
221: the long-time averaged $L^2$ norm of $\theta$,
222: \begin{equation}
223: \langle \theta^2 \rangle:=\lim_{T \to \infty}\frac{1}{T} \int_{0}^{T}\mathrm{d}t \
224: \frac{1}{L^d}\int \mathrm{d}^dx\, \theta^2(t, \bm{x})
225: \end{equation}
226: The smaller $\langle \theta^2 \rangle$ is, the more uniform the
227: distribution.  The ``mixing enhancement'' of a stirring field is
228: naturally measured by comparing the scalar variance to the variance
229: with the same source but in the absence of stirring.  To be precise,
230: we compare $\langle \theta^2 \rangle$ to $\langle \theta_{0}^{\ 2}
231: \rangle$ where $\theta_{0}$ is the solution to
232: \begin{equation}
233: \frac{\partial\theta_{0}}{\partial t}=\kappa \Delta \theta_{0} + s(\bm{x})
234: \label{ds}
235: \end{equation}
236: (with, say, the same initial data although these will not affect the
237: long-time averaged fluctuations).  
238: The dimensionless {\it mixing enhancement factor} is then defined as
239: \begin{equation}
240: {\cal E}_{0}:=\sqrt{\frac{\langle \theta_{0}^{\,2} \rangle}{\langle \theta^2 \rangle}}.
241: \end{equation} 
242: This quantity carries the subscript $0$ because we can also define
243: {\it multiscale mixing enhancements} \cite{DoeringThiffeault2006,Shaw2007} by weighting
244: large/small wavenumber components of the scalar fluctuations:
245: \begin{equation}
246: {\cal E}_{p}:=\sqrt{\frac{\langle |\nabla^{p}\theta_{0}|^{2} \rangle}{ \langle|\nabla^{p}\theta|^{2} \rangle}}\,,
247: \qquad p=-1,0,1.
248: \end{equation}
249: As discussed in Doering \& Thiffeault \cite{DoeringThiffeault2006}, Shaw {\it et al}
250: \cite{Shaw2007} and Shaw \cite{ShawGFD2005}, ${\cal E}_{\pm1}$ provide a gauge of the
251: mixing enhancement of the flow as measured by scalar fluctuations on
252: relatively small and large length scales, respectively. 
253: We refer to them as enhancement factors because if one were to define an
254: effective, eddy, or equivalent diffusivity $\kappa_{e,p}$ as the value of a molecular
255: diffusion necessary to produce the same value of $\langle |\nabla^{p}\theta|^{2} \rangle$
256: with stirring, then $\kappa_{e,p}=\kappa {\cal E}_{p}$.
257:  In this paper, however, we will focus exclusively on ${\cal E}_{0}$, the mixing enhancement at
258: ``moderate'' length scales.
259: 
260: There is a theoretical upper bound on ${\cal E}_{0}$ valid for any
261: statistically stationary homogeneous and isotropic stirring field
262: \cite{DoeringThiffeault2006, ShawGFD2005, Shaw2007}:
263: \begin{equation}
264: {\cal E}_{0} \leq \sqrt{\frac{\sum_{\bm{k} \neq \bm{0}}|\hat{s}(\bm{k})|^{2}/k^{4}}{\sum_{\bm{k} \neq \bm{0}}|\hat{s}(\bm{k})|^{2}/(k^{4}+\frac{{\rm Pe}^{2}}{L^{2}d}k^{2})}}
265: \label{bound}
266: \end{equation} 
267: where $\hat{s}(\bm{k})$ are the Fourier coefficients of the source
268: and the P\'eclet number,
269: \begin{equation}
270: {\rm Pe}:={UL}/{\kappa},
271: \end{equation}
272: is a dimensionless measure of the intensity of the stirring.
273: Generally, we anticipate that ${\cal E}_{0}$ is an increasing function
274: of Pe and the estimate in \eqref{bound} guarantees that 
275: ${\cal E}_{0}(\text{Pe}) \lesssim \text{Pe}$ as $\text{Pe} \rightarrow
276: \infty$, the ``classical'' scaling necessary if there is to be any residual variance
277: suppression in the singular vanishing diffusion limit.
278: That is, if ${\cal E}_{0}(\text{Pe}) \sim \text{Pe}$ then
279: $\kappa_{e,0}$ has a nonzero limit as $\kappa \rightarrow 0$
280: with all other parameters held fixed.
281: It is natural to refer to any of the possible sub-classical scalings as ``anomalous''.
282: 
283: The upper limit to the mixing enhancement in \eqref{bound} depends on the
284: stirring field only through $U$ via Pe, but it depends on all the
285: details of the source distribution.  As studied in depth in references
286: \cite{DoeringThiffeault2006,Shaw2007,ShawGFD2005}, the structure of the scalar source can have a
287: profound effect on the high Pe scaling of ${\cal E}_{0}$, notably for sources with small scales.
288: It is physically meaningful to consider measure-valued source-sink distributions, like delta-functions,
289: with arbitrarily small scales.  
290: It is precisely this source size dependence of ${\cal E}_{0}(\text{Pe})$ that
291: motivates the development of a computational method that can handle
292: singular source distributions.
293: 
294: In this study, for computational simplicity and efficiency, we utilize the ``random sine flow'' as the stirring field.
295: In the two-dimensional case this is defined for all time by
296: \begin{equation}
297: {\bm{u}}(t,\bm{x})= \begin{cases}
298: w\sin \left({2 \pi y}/{L}+\phi\right)\hat{\textbf{\i}}\,, \qquad &nT< t \le nT+\tfrac12T;\\
299: w\sin \left({2 \pi x}/{L}+\phi'\right)\hat{\textbf{\j}}\,, \qquad
300: & nT+\tfrac12T < t \le (n+1)T,
301: \end{cases}
302: \end{equation}
303: where $T$ is the period, $n=0, 1, 2,\dots$, and $\phi$ and $\phi'$
304: are random phases chosen independently and uniformly on $[0,2\pi)$ in
305: each half cycle, which assures the homogeneity of the flow field. In
306: this case, $w=\sqrt{2}U$. 
307: In the three-dimensional case, we employ
308: \begin{equation}
309: \bm{u}(t, \bm{x})=
310: \begin{cases}
311: w\left[\sin \alpha \sin\left(2\pi y/L+\phi_{2}\right)+\cos \alpha \sin\left(2 \pi z/L+\phi_{3}\right) \right]\hat{\textbf{\i}}\,,\  &nT < t \leq nT+ \tfrac13T; \nonumber \\
312: w\left[\sin \alpha \sin\left(2 \pi z/L+\psi_{3}\right)+\cos \alpha \sin\left(2 \pi x/L+\phi_{1}\right) \right]\hat{\textbf{\j}}\,,\ &nT+\tfrac13T < t \leq nT+\tfrac23T; \nonumber \\
313: w\left[\sin \alpha \sin\left(2 \pi x/L+\psi_{1}\right)+\cos \alpha \sin\left(2 \pi y/L+\psi_{2}\right) \right]\hat{\textbf{k}}\,,\ &nT+\tfrac23T < t \leq (n+1)T, \nonumber 
314: \end{cases}
315: \end{equation}
316: where again $w=\sqrt{2}U$, $n=0,1,2,\dots$, and $\alpha$,
317: $\phi_{1,2,3}$ and $\psi_{1,2,3}$ are uniform random numbers in $[0,
318: 2\pi)$ chosen independently every $T/3$.  The angle~$\alpha$ randomizes
319: the shear direction to guarantee isotropy of the flow.
320: 
321: 
322: \section{Numerical method}
323: 
324: In a particle code for solving the advection-diffusion equation, the
325: concentration field $\rho$ is represented by a distribution of
326: particles.  Particles are introduced by generating random locations
327: using the properly normalized source $S(\bm{x})$ as a probability
328: distribution function, then they are transported by advection and
329: diffusion.  The particle density, $\rho(t, \bm{x})$, is
330: measured by covering the domain with bins counting the number of
331: particles per bin.
332: 
333: A discrete particle method is employed because it can easily deal with small-scale
334: sources such as $\delta$ functions.  
335: It is also straightforward to implement with any advection field.  
336: The downside of a particle method is that it necessarily involves two kinds of errors: 
337: the number density of particles calculated by dividing the domain into bins is
338: only resolved down to the size of the bins, and the
339: measurement of $\rho$ always includes statistical errors due to
340: the use of finite numbers of particles.
341: 
342: \subsection{Time evolution}
343: 
344: At each time step the system is evolved by advection, diffusion, the source, and sinks.  
345: An advection-only equation would be solved by moving particles along characteristics, 
346: and a diffusion-only equation would be solved by adding independent Gaussian noises
347: to each coordinate of each particle.  
348: With both advection and diffusion we need to solve a stochastic differential equation to 
349: determine the proper displacement of the particles during a time step.  
350: The stochastic differential equation is
351: \begin{equation}
352: d \bm{X}=\bm{u}(t, \bm{X})dt+\sqrt{2\kappa}\ d\bm{W}
353: \label{eq:sde}
354: \end{equation}
355: where $\bm{W}(t)$ is a standard vector-valued Wiener process. 
356: 
357: In order to solve \eqref{eq:sde}, we will consider cases where the displacement
358: due to the noise in a subinterval of length $T/d$ (where $d$ is the dimension) 
359: is much smaller than the wavelength of the random sine flow.  
360: This condition is realized better and better as Pe increases.
361: Then, during each subinterval, the drift field $\bm{u}(t,
362: \bm{X})$ experienced by each particle can be approximated by a
363: steady flow with a linear shear.  
364: In 2D, for the first
365: half of the period for a particle starting at $(x_{0}, y_{0})=(X(t=0),
366: Y(t=0))$ we approximate \eqref{eq:sde} by
367: \begin{align}
368: \begin{split}
369: dX&=w\sin\left({2\pi y_{0}}/{L}+\phi\right)dt+w\cos\left({2 \pi y_{0}}/{L}+\phi\right)\dfrac{2 \pi}{L}(Y-y_{0})dt+\sqrt{2\kappa}\,dW_{1},\\
370: dY&=\sqrt{2\kappa}\,dW_{2},
371: \end{split}
372: \end{align}
373: and for the second half of the period, starting from $(x'_0,y'_0)=(X(t=T/2), Y(t=T/2))$,
374: \begin{align}
375: \begin{split}
376: dX&=\sqrt{2\kappa}\,dW_{1}, \\
377: dY&=w\sin\left({2\pi x'_{0}}/{L}+\phi '\right)dt+w\cos\left({2 \pi x'_{0}}/{L}+\phi '\right)\dfrac{2 \pi}{L}(X-x'_{0})dt+\sqrt{2\kappa}\,dW_{2}\,.
378: \end{split}
379: \end{align}
380: Therefore, during the first half period we evolve the position of a
381: particle through a time interval $\Delta t$ (where $\Delta t \le T/2$
382: need {\it not} be small) by the map
383: \begin{align}
384: \begin{split}
385: x_{0} &\rightarrow x_{0}+w\sin\left({2\pi y_{0}}/{L}+\phi\right)\Delta t+ R_{1}, \\
386: y_{0} &\rightarrow y_{0}+R_{2},
387: \end{split}
388: \end{align}
389: where $R_{1}$ and $R_{2}$ satisfy 
390: \begin{align}
391: \begin{split}
392: dR_{1}&= S_{2}R_{2}dt+\sqrt{2 \kappa}dW_{1} \qquad \left(S_{2}:=2\pi w
393:   L^{-1}\cos\left({2 \pi y_{0}}/{L}+\phi\right)\right), \\
394: dR_{2}&=\sqrt{2\kappa}dW_{2}.
395: \end{split}
396: \end{align}
397: The variance-covariance matrix of $R_{1}$ and $R_{2}$ is
398: \begin{equation}
399:  \left(
400: \begin{array}{cc}
401: \bm{E}(R_{1}^{\ 2})& \bm{E}(R_{1}R_{2})  \\
402: \bm{E}(R_{2}R_{1})& \bm{E}(R_{2}^{\ 2})  \\
403: \end{array}
404: \right)
405: =
406:  \left(
407: \begin{array}{cc}
408: \frac{2}{3}S_{2}^{\ 2}\kappa t^3+2\kappa t& S_{2}\kappa t^2\\
409: S_{2} \kappa t^2&  2\kappa t \\
410: \end{array}
411: \right),
412: \label{eq:varcovar}
413: \end{equation}
414: which is realized by 
415: \begin{align}
416: R_{1}&=\sqrt{\tfrac{1}{6}S_{2}^{\ 2}\kappa t^3+2\kappa t}\times N_{1}+\sqrt{\tfrac{1}{2}S_{2}^{\ 2}\kappa t^3} \times N_{2},\\
417: R_{2}&=\sqrt{2\kappa t}\times N_{2},
418: \end{align}
419: where $N_1$ and $N_2$ are independent $N(0,1)$ random variables
420: (normally distributed with mean~$0$ and standard deviation~$1$).  The
421: matrix~\eqref{eq:varcovar} describes the evolution of a passive scalar
422: field in a shear flow~\cite{Taylor1954,Aris1956,Rhines1983}.
423: 
424: Therefore the time evolution map during the first half period is
425: \begin{subequations}
426: \begin{align}
427: x_0 &\rightarrow x_0+w\sin\left({2\pi y_{0}}/{L}+\phi\right)\,\Delta t+
428: \sqrt{\tfrac{1}{6}S_2^{\ 2} \kappa (\Delta t)^3 + 2 \kappa \Delta t} \, N_1 +
429:  \sqrt{\tfrac{1}{2} S_2^{\ 2} \kappa (\Delta t)^3} \, N_2\,, \label{eq:evolution1}\\
430: y_0 &\rightarrow y_0+\sqrt{2 \kappa \Delta t} \, N_2\,.  \label{eq:evolution2} 
431: \end{align}%
432: \end{subequations}%
433: A similar map is employed during the second half of the period.  These
434: stochastic maps include the shear --- in the approximation that the
435: shear remains constant for each particle during each half cycle ---
436: that causes a ``distortion'' of a Gaussian cloud of particles; see Fig.~\ref{fig:gauss}.
437: 
438: \begin{figure}
439: \begin{center}
440: \subfigure[]{
441:   \includegraphics[height=4.15cm, angle=270]{gauss}
442:   \label{fig:gauss1}
443: }\hspace{3em}%
444: \subfigure[]{
445:   \includegraphics[height=5cm, angle=270]{gauss2}
446:   \label{fig:gauss2}
447: }
448: \end{center}
449: \caption{(a) A circular Gaussian distribution of particles transported
450:   and sheared into (b) an elliptical Gaussian cloud.}
451: \label{fig:gauss}
452: \end{figure}
453: 
454: 
455: The same calculations apply for the three-dimensional case: for the
456: first subinterval, the time-evolution map is
457: \begin{subequations}
458: \begin{align}
459: x_0 &\rightarrow x_0+w\left[\sin \alpha \sin\left({2\pi y_0}/{L}+\phi_{2}\right)+\cos \alpha \sin\left({2 \pi z_0}/{L}+\phi_{3}\right) \right]\nonumber\\
460: &\phantom{\rightarrow x_0}+\sqrt{\tfrac{1}{6}(S_{2}^{\ 2}+S_{3}^{\ 2})\kappa (\Delta t)^{3}+2\kappa\Delta t } \, N_{1}
461: +S_{2}\sqrt{\tfrac12\kappa}(\Delta t) ^{\frac{3}{2}} \, N_{2}
462: +S_{3}\sqrt{\tfrac12\kappa}(\Delta t) ^{\frac{3}{2}} \, N_{3},\\
463: y_0 &\rightarrow y_0+\sqrt{2 \kappa \Delta t} \, N_{2},\\
464: z_0 &\rightarrow z_0+\sqrt{2 \kappa \Delta t} \, N_{3},
465: \end{align}
466: \end{subequations}
467: where $S_{2}= 2\pi w L^{-1} \sin\alpha
468: \cos({2\pi y_{0}}/{L}+\phi_{2})$, $S_{3}= 2\pi w L^{-1} \cos\alpha
469: \cos({2\pi z_{0}}/{L}+\phi_{3})$, and $N_1, N_2$ and $N_{3}$ are
470: independent $N(0,1)$ random variables. The maps for the other
471: subintervals can be obtained by cyclic permutation of the coordinates.
472: 
473: The steady scalar source is realized by introducing a new particle one
474: by one using normalized $S(\bm{x})$ as a probability distribution
475: function.  Numerically, such a probability distribution function can
476: be realized by mapping uniform random numbers over $[0,1]$ with the
477: inverse of the cumulative probability distribution function in
478: question.
479: 
480: New particles are added constantly so the total number of particles
481: continues to increase, which slows down the computation.  To cope with
482: increasing particles, we implement a particle subtraction scheme.
483: Particles eventually get well mixed and ``older'' particles do not
484: contribute to the value of the hydrodynamic variance.  There is no
485: added value in keeping track of particles that have been in the mix
486: for a very long time, and we can simply remove them from the system
487: after a sufficiently long time.  It is very important to keep track of
488: the ``age'' of each particle, however, and to only remove sufficiently
489: old well-mixed particles.  (For example, if a random fraction of
490: particles is removed at regular time intervals, then the simulation
491: becomes one of a system particles with a random finite lifetime,
492: described by an advection-diffusion equation with an additional
493: density decay term.)
494: 
495: In order to determine how old particles must be in order to safely
496: remove them without affecting the hydrodynamic variance, prior to a
497: full simulation run a test is performed as follows.  Starting from an
498: initial set of $N_{i}$ particles located in space according to the
499: source distribution, the flow and diffusion are allowed to act and the
500: variance of the number of particles per bin, which decays with time,
501: is monitored.  The number $N_{i}$ is of the order of the number of
502: particles that are introduced in the full simulation during, say, an
503: interval of length $T$ characteristic of the random sine flow.  The
504: variance does not decay all the way to zero, however, but rather to
505: the variance expected when $N_{i}$ particles are randomly distributed
506: among the bins.  The time when the variance achieves this
507: random-distribution variance, measured beforehand for a given flow and
508: diffusion strength, is then the required ``aging'' time before
509: particles can be safely removed in the full simulation with the steady
510: source.  Such a trial run is performed for each flow, diffusion
511: strength, source distribution and particle number because this
512: ``mixing time'' depends on all these factors.  Futher details of the
513: criteria for removing old particles and extensive tests and benchmark
514: trials may be found in Ref.~\cite{OkabeGFD2006}.
515: 
516: 
517: \subsection{Variance calculation and background noise}
518: 
519: The variance $\langle \theta^2\rangle$ is measured by monitoring the
520: fluctuations in the number of particles per bin, and time-averaging.
521: In $d$ dimensions the domain is divided into $l^d$ bins and the code
522: calculates $\langle n^2 \rangle$, where $n$ is the number of particles
523: in a bin.
524: Then $\langle \theta^2\rangle$ is initially approximated by
525: \begin{equation}
526:   \langle n^2 \rangle - \langle n \rangle^2
527:   = \left(L/l\right)^{2d} \langle \theta^2 \rangle.
528:   \label{measure}
529: \end{equation}
530: We say ``initially'' because the expression above includes both the
531: hydrodynamic fluctuations of interest {\it and} discreteness
532: fluctuations resulting solely from the fact that each bin contains a
533: finite number of particles.
534: 
535: The subtraction scheme eliminates the ``well-mixed'' particles that do
536: not contribute to the value of the hydrodynamic variance.  But even if
537: the system were completely mixed so that theoretically, $\langle
538: \theta^2 \rangle = 0$, the measured variance $\langle n^2 \rangle -
539: \langle n \rangle^2$ would be (very close to, for small bins) $\langle
540: n \rangle$, which is on the order of $N/l^d$, where $N$ is the total
541: number of particles in the domain.  This follows from the fact that
542: $\theta(t, \bm{x})$ is represented in this particle method by only a
543: finite number of particles in each finite size bin.  That is,
544: $\langle \theta^2 \rangle$ as defined by (\ref{measure}) is nonzero
545: even when the particles are uniformly distributed: then the bulk
546: variance includes fluctuations as if $N$ particles were randomly
547: thrown in $l^d$ bins.  The helpful fact is that the bulk variance
548: contribution from these {\it background fluctuations} due to finite
549: numbers of particles in the bins does not depend on (i.e., is
550: uncorrelated with) the hydrodynamic density variation from bin to bin.
551: The total contribution to the variance is the sum of the ``extra''
552: variance in each bin which is linear in the (mean) number of particles
553: in each bin.  Hence the sum of the variances is $\sim N$ and the bulk
554: variance contribution from the background fluctuations,
555: $Nl^{d}/L^{2d}$, can simply be subtracted from the initial estimate
556: for $\langle \theta^2 \rangle$ in (\ref{measure}).  The net result is
557: our measured value of the hydrodynamic variance.
558: 
559: In addition to the inevitable fluctuations due to discreteness,
560: density variations are observed only down to the length scales $\sim L/l$
561: because of the binning density, which is another source of error in
562: this procedure.
563: We use $l \ge 100$, which tests and benchmark studies indicate
564: is sufficient for the examples studies here~\cite{OkabeGFD2006}.
565: 
566: The variance is calculated once for each subinterval, and the instant
567: when it is calculated is determined randomly in order to obtain an
568: unbiased time average.  Thus each subinterval is divided into two
569: parts, before and after variance calculation, and the particle
570: transport and source processes are appropriately adapted.
571: The final measured quantities are long time averages that are 
572: observed to be converged to within the error indicated on the plots below.
573: 
574: \section{Results}
575: 
576: In order to investigate the effect of source-sink scales on maximal
577: and actual mixing enhancements, we performed a series of
578: simulations for square-shaped sources of various sizes $a < L$ as
579: illustrated in Fig.~\ref{fig:sqfig}.
580: 
581: \begin{figure}
582: \begin{center}
583: \subfigure[\ {$a=L/2$}]{
584:   \includegraphics[height=5cm, angle=270]{sqfig2}
585:   \label{fig:sqfig2}
586: }\hspace{3em}%
587: \subfigure[\ {$a=L/10$}]{
588:   \includegraphics[height=5cm, angle=270]{sqfig10}
589:   \label{fig:sqfig10}
590: }
591: \caption{Square-shaped source of two different sizes; particles shown
592: sampled from the uniform source distribution over the squares.}
593: \label{fig:sqfig}
594: \end{center}
595: \end{figure}
596: 
597: Fig~\ref{fig:2Dbounds} shows the upper bounds on ${\cal E}_{0}$ for
598: square sources and a $\delta$-function source in 2D computed from (\ref{bound}).
599: The upper bound for any finite-size source is asymptotically $\sim $ Pe, but for the
600: $\delta$-function source it is $\sim {\rm Pe}/\ln {\rm Pe}$ in the large Pe limit.  
601: In 3D, the distinction between cubic sources and a $\delta$-function source is more
602: apparent as shown in Fig~\ref{fig:3Dbounds}: the upper bound for a
603: $\delta$-function source behaves $\sim \sqrt{{\rm Pe}}$ in 3D.
604: We stress that these mixing enhancement bounds apply for {\it any} statistically
605: homogeneous and isotropic flows stirring sources with these shapes.
606: 
607: \begin{figure}
608: \begin{center}
609: \subfigure[\ {$d=2$}]{
610:   \includegraphics[height=6cm]{fig3}
611:   \label{fig:2Dbounds}
612: }\hspace{2em}%
613: \subfigure[\ {$d=3$}]{
614:   \includegraphics[height=6cm]{fig4}
615:   \label{fig:3Dbounds}
616: }
617: \caption{The theoretical upper bounds for (a) square sources with
618:   sizes $a=L/2, L/10, L/50$, and a $\delta$-function source; (b) cubic
619:   sources with sizes $a=L/5, L/50, L/500$, and a $\delta$-function
620:   source (from top to bottom).}
621: %\label{fig:bounds}
622: \end{center}
623: \end{figure}
624: 
625: Simulation results for the random sine flow, shown in Fig.
626: \ref{fig:2DMixing} for 2D and Fig. \ref{fig:3DMixing} for 3D
627: qualitatively confirm the behavior of the enhancements suggested by the
628: upper limits.  
629: As the source size shrinks, the measured mixing enhancement gets smaller
630: in a way that is remarkably similar to the bounds.
631: In these simulations Pe is varied by decreasing $\kappa$ at a
632: fixed values of $L, U$ and $T$.  Other values of $T$ and other
633: (shorter) wavelengths of the stirring flow were also checked,
634: producing similar plots.  These 2D simulation results have recently been
635: confirmed quantitatively by a PDE computation~\cite{Chertock2008}.
636: 
637: \begin{figure}
638: \begin{center}
639: \subfigure[\ {$d=2$}]{
640:   \includegraphics[height=6cm]{fig5}
641:   \label{fig:2DMixing}
642: }\hspace{2em}%
643: \subfigure[\ {$d=3$}]{
644:   \includegraphics[height=6cm]{fig6}
645:   \label{fig:3DMixing}
646: }
647: \caption{Measured mixing enhancements for (a) square sources with sizes
648:   $a=L/2, L/10, L/50$ and a $\delta$-function source, (b) cubic
649:   sources with sizes $a=L/5, L/50, L/500$ and a $\delta$-function
650:   source (from top to bottom). }
651: %\label{fig:bounds}
652: \end{center}
653: \end{figure}
654: 
655: The simulations also show that the upper estimates can give the
656: correct {\it quantitative} behavior of ${\cal E}_{0}$ as a function of
657: Pe.  Indeed, in Fig \ref{fig:3DDelta} we plot the upper bound on
658: ${\cal E}_{0}$ for the $\delta$-function source in 3D and the measured
659: enhancement from the simulations.  The upper bound, which scales
660: anomalously $\sim \sqrt{{\rm Pe}}$, is an excellent predictor of the
661: data.  From this we conclude that the random sine flow is an
662: ``almost-optimal'' mixer (among statistically homogeneous and isotropic
663: flows) for this source-sink distribution.
664: 
665: \begin{figure}
666: \begin{center}
667: %\includegraphics[height=5cm,angle=270]{Fig/3Dbounds.eps}
668: \includegraphics[height=6cm]{fig7}
669: \caption{Mixing enhancement for a $\delta$-function source.  The solid
670:   line is the upper bound for any SHI flow and the data are mixing
671:   enhancements for the random sine flow measured in the discrete
672:   particle simulations.}
673: \label{fig:3DDelta}
674: \end{center}
675: \end{figure}
676: 
677: \section{Summary and conclusions}
678: 
679: We have devised an accurate and computationally efficient particle
680: method to study hydrodynamic variance suppression by a mixing flow.
681: Rigorous upper bounds for the mixing enhancement ${\cal E}_{0}$, i.e.,
682: the effective diffusion enhancement factor, were compared to measured
683: enhancements for the simple random sine flow.  A key prediction of the
684: analysis in Refs.~\cite{DoeringThiffeault2006,Shaw2007} is that the source-sink shape is a
685: determining factor in the mixing enhancement of any flow.  The simulation
686: results reported here show that the upper estimates give the correct
687: qualitative picture as regards the Pe and source-shape dependence of
688: ${\cal E}_{0}$.
689: 
690: Future work should focus on investigating enhancements of other stirring
691: flows.  No attempt has been made here to find a more efficient
692: stirring flow (or, indeed, the {\it most} efficient flow, if there
693: is one) whose enhancement approaches more closely (or perhaps even
694: saturates) the upper bound.  It is remarkable that the simple
695: random sine flow appears to saturate the upper bound scaling
696: ${\cal  E}_{0} \sim \sqrt{{\rm Pe}}$ in 3D.
697: 
698: Stirring with appropriate turbulent solutions to the
699: Navier--Stokes equation is also of significant interest.
700: The central question here is, is statistically homogeneous and isotropic
701: turbulence generically an efficient mixer?
702: The answer may depend on the source-sink distribution.
703: 
704: \section*{Acknowledgements}
705: 
706: The authors thank Jai Sukhatme for helpful comments on the paper.
707: This work was supported in part by US National Science Foundation
708: through awards PHY-0555324 and DMS-0553487, by the Geophysical Fluid
709: Dynamics Program at Woods Hole Oceanographic Institution, and by the
710: Alexander von Humboldt Foundation.
711: 
712: %\bibliographystyle{jlt}
713: %\bibliography{journals_abbrev,articles,chertok}
714: 
715: \begin{thebibliography}{32}
716: \newcommand{\enquote}[1]{`#1'}
717: \providecommand{\natexlab}[1]{#1}
718: \providecommand{\url}[1]{\texttt{#1}}
719: \providecommand{\urlprefix}{URL }
720: \providecommand{\bibinfo}[2]{#2}
721: \providecommand{\eprint}[2][]{\url{#2}}
722: 
723: \bibitem[{Ottino(1990)}]{Ottino1990}
724: \bibinfo{author}{J.~M. Ottino}, \enquote{\bibinfo{title}{Mixing, chaotic
725:   advection, and turbulence},} \emph{\bibinfo{journal}{Annu. Rev. Fluid Mech.}}
726:   \textbf{\bibinfo{volume}{22}}, \bibinfo{pages}{207--253}
727:   (\bibinfo{year}{1990}).
728: 
729: \bibitem[{Majda and Kramer(1999)}]{Majda1999}
730: \bibinfo{author}{A.~J. Majda} and \bibinfo{author}{P.~R. Kramer},
731:   \enquote{\bibinfo{title}{Simplified models for turbulent diffusion: {T}heory,
732:   numerical modelling and physical phenomena},} \emph{\bibinfo{journal}{Physics
733:   Reports}} \textbf{\bibinfo{volume}{314}}~(\bibinfo{number}{4-5}),
734:   \bibinfo{pages}{237--574} (\bibinfo{year}{1999}).
735: 
736: \bibitem[{Warhaft(2000)}]{Warhaft2000}
737: \bibinfo{author}{Z.~Warhaft}, \enquote{\bibinfo{title}{Passive scalars in
738:   turbulent flows},} \emph{\bibinfo{journal}{Annu. Rev. Fluid Mech.}}
739:   \textbf{\bibinfo{volume}{32}}, \bibinfo{pages}{203--240}
740:   (\bibinfo{year}{2000}).
741: 
742: \bibitem[{Shraiman and Siggia(2000)}]{Shraiman2000}
743: \bibinfo{author}{B.~I. Shraiman} and \bibinfo{author}{E.~D. Siggia},
744:   \enquote{\bibinfo{title}{Scalar turbulence},}
745:   \emph{\bibinfo{journal}{Nature}} \textbf{\bibinfo{volume}{405}},
746:   \bibinfo{pages}{639--646} (\bibinfo{year}{2000}).
747: 
748: \bibitem[{Sawford(2001)}]{Sawford2001}
749: \bibinfo{author}{B.~L. Sawford}, \enquote{\bibinfo{title}{Turbulent relative
750:   dispersion},} \emph{\bibinfo{journal}{Annu. Rev. Fluid Mech.}}
751:   \textbf{\bibinfo{volume}{33}}, \bibinfo{pages}{289--317}
752:   (\bibinfo{year}{2001}).
753: 
754: \bibitem[{Falkovich \emph{et~al.}(2001)Falkovich, Gaw\c{e}dzki, and
755:   Vergassola}]{Falkovich2001}
756: \bibinfo{author}{G.~Falkovich}, \bibinfo{author}{K.~Gaw\c{e}dzki}, and
757:   \bibinfo{author}{M.~Vergassola}, \enquote{\bibinfo{title}{Particles and
758:   fields in turbulence},} \emph{\bibinfo{journal}{Rev. Mod. Phys.}}
759:   \textbf{\bibinfo{volume}{73}}~(\bibinfo{number}{4}),
760:   \bibinfo{pages}{913--975} (\bibinfo{year}{2001}).
761: 
762: \bibitem[{Aref(2002)}]{Aref2002}
763: \bibinfo{author}{H.~Aref}, \enquote{\bibinfo{title}{The development of chaotic
764:   advection},} \emph{\bibinfo{journal}{Phys. Fluids}}
765:   \textbf{\bibinfo{volume}{14}}~(\bibinfo{number}{4}),
766:   \bibinfo{pages}{1315--1325} (\bibinfo{year}{2002}).
767: 
768: \bibitem[{Wiggins and Ottino(2004)}]{Wiggins2004}
769: \bibinfo{author}{S.~Wiggins} and \bibinfo{author}{J.~M. Ottino},
770:   \enquote{\bibinfo{title}{Foundations of chaotic mixing},}
771:   \emph{\bibinfo{journal}{Phil. Trans. R. Soc. Lond. A}}
772:   \textbf{\bibinfo{volume}{362}}, \bibinfo{pages}{937--970}
773:   (\bibinfo{year}{2004}).
774: 
775: \bibitem[{Thiffeault \emph{et~al.}(2004)Thiffeault, Doering, and
776:   Gibbon}]{Thiffeault2004}
777: \bibinfo{author}{J.-L. Thiffeault}, \bibinfo{author}{C.~R. Doering}, and
778:   \bibinfo{author}{J.~D. Gibbon}, \enquote{\bibinfo{title}{A bound on mixing
779:   efficiency for the advection--diffusion equation},}
780:   \emph{\bibinfo{journal}{J. Fluid Mech.}} \textbf{\bibinfo{volume}{521}},
781:   \bibinfo{pages}{105--114} (\bibinfo{year}{2004}).
782: 
783: \bibitem[{Townsend(1951)}]{Townsend1951}
784: \bibinfo{author}{A.~A. Townsend}, \enquote{\bibinfo{title}{The diffusion of
785:   heat spots in isotropic turbulence},} \emph{\bibinfo{journal}{Proc. R. Soc.
786:   Lond. A}} \textbf{\bibinfo{volume}{209}}~(\bibinfo{number}{1098}),
787:   \bibinfo{pages}{418--430} (\bibinfo{year}{1951}).
788: 
789: \bibitem[{Townsend(1954)}]{Townsend1954}
790: \bibinfo{author}{A.~A. Townsend}, \enquote{\bibinfo{title}{The diffusion behind
791:   a line source in homogeneous turbulence},} \emph{\bibinfo{journal}{Proc. R.
792:   Soc. Lond. A}} \textbf{\bibinfo{volume}{224}}~(\bibinfo{number}{1159}),
793:   \bibinfo{pages}{487--512} (\bibinfo{year}{1954}).
794: 
795: \bibitem[{Durbin(1980)}]{Durbin1980}
796: \bibinfo{author}{P.~A. Durbin}, \enquote{\bibinfo{title}{A stochastic model of
797:   two-particle dispersion and concentration fluctuations in homogeneous
798:   turbulence},} \emph{\bibinfo{journal}{J. Fluid Mech.}}
799:   \textbf{\bibinfo{volume}{100}}, \bibinfo{pages}{279--302}
800:   (\bibinfo{year}{1980}).
801: 
802: \bibitem[{Drummond(1982)}]{Drummond1982}
803: \bibinfo{author}{I.~T. Drummond}, \enquote{\bibinfo{title}{Path-integral
804:   methods for turbulent diffusion},} \emph{\bibinfo{journal}{J. Fluid Mech.}}
805:   \textbf{\bibinfo{volume}{123}}, \bibinfo{pages}{59--68}
806:   (\bibinfo{year}{1982}).
807: 
808: \bibitem[{Sawford and Hunt(1986)}]{Sawford1986}
809: \bibinfo{author}{B.~L. Sawford} and \bibinfo{author}{J.~C.~R. Hunt},
810:   \enquote{\bibinfo{title}{Effects of turbulence structure, molecular diffusion
811:   and source size on scalar fluctuations in homogeneous turbulence},}
812:   \emph{\bibinfo{journal}{J. Fluid Mech.}} \textbf{\bibinfo{volume}{165}},
813:   \bibinfo{pages}{373--400} (\bibinfo{year}{1986}).
814: 
815: \bibitem[{Thomson(1990)}]{Thomson1990}
816: \bibinfo{author}{D.~J. Thomson}, \enquote{\bibinfo{title}{A stochastic model
817:   for the motion of particle pairs in isotropic high-{R}eynolds number
818:   turbulence, and its application to the problem of concentration variance},}
819:   \emph{\bibinfo{journal}{J. Fluid Mech.}} \textbf{\bibinfo{volume}{210}},
820:   \bibinfo{pages}{113--153} (\bibinfo{year}{1990}).
821: 
822: \bibitem[{Borgas and Sawford(1994)}]{Borgas1994}
823: \bibinfo{author}{M.~S. Borgas} and \bibinfo{author}{B.~L. Sawford},
824:   \enquote{\bibinfo{title}{A family of stochastic models for two-particle
825:   dispersion in isotropic homogeneous stationary turbulence},}
826:   \emph{\bibinfo{journal}{J. Fluid Mech.}} \textbf{\bibinfo{volume}{279}},
827:   \bibinfo{pages}{69--99} (\bibinfo{year}{1994}).
828: 
829: \bibitem[{Chertkov \emph{et~al.}(1995{\natexlab{a}})Chertkov, Falkovich,
830:   Kolokolov, and Lebedev}]{Chertkov1995}
831: \bibinfo{author}{M.~Chertkov}, \bibinfo{author}{G.~Falkovich},
832:   \bibinfo{author}{I.~Kolokolov}, and \bibinfo{author}{V.~Lebedev},
833:   \enquote{\bibinfo{title}{Statistics of a passive scalar advected by a
834:   large-scale two-dimensional velocity field: Analytic solution},}
835:   \emph{\bibinfo{journal}{Phys. Rev. E}}
836:   \textbf{\bibinfo{volume}{51}}~(\bibinfo{number}{6}),
837:   \bibinfo{pages}{5609--5627} (\bibinfo{year}{1995}{\natexlab{a}}).
838: 
839: \bibitem[{Chertkov \emph{et~al.}(1995{\natexlab{b}})Chertkov, Falkovich,
840:   Kolokolov, and Lebedev}]{Chertkov1995b}
841: \bibinfo{author}{M.~Chertkov}, \bibinfo{author}{G.~Falkovich},
842:   \bibinfo{author}{I.~Kolokolov}, and \bibinfo{author}{V.~Lebedev},
843:   \enquote{\bibinfo{title}{Normal and anomalous scaling of the fourth-order
844:   correlation function of a randomly advected passive scalar},}
845:   \emph{\bibinfo{journal}{Phys. Rev. E}}
846:   \textbf{\bibinfo{volume}{52}}~(\bibinfo{number}{5}),
847:   \bibinfo{pages}{4924--4941} (\bibinfo{year}{1995}{\natexlab{b}}).
848: 
849: \bibitem[{Chertkov \emph{et~al.}(1997)Chertkov, Kolokolov, and
850:   Vergassola}]{Chertkov1997}
851: \bibinfo{author}{M.~Chertkov}, \bibinfo{author}{I.~Kolokolov}, and
852:   \bibinfo{author}{M.~Vergassola}, \enquote{\bibinfo{title}{Inverse cascade and
853:   intermittency of passive scalar in one-dimensional smooth flow},}
854:   \emph{\bibinfo{journal}{Phys. Rev. E}}
855:   \textbf{\bibinfo{volume}{56}}~(\bibinfo{number}{5}),
856:   \bibinfo{pages}{5483--5499} (\bibinfo{year}{1997}).
857: 
858: \bibitem[{Chertkov(1997)}]{Chertkov1997b}
859: \bibinfo{author}{M.~Chertkov}, \enquote{\bibinfo{title}{Instanton for random
860:   advection},} \emph{\bibinfo{journal}{Phys. Rev. E}}
861:   \textbf{\bibinfo{volume}{55}}~(\bibinfo{number}{3}),
862:   \bibinfo{pages}{2722--2735} (\bibinfo{year}{1997}).
863: 
864: \bibitem[{Chertkov \emph{et~al.}(1998)Chertkov, Falkovich, and
865:   Kolokolov}]{Chertkov1998}
866: \bibinfo{author}{M.~Chertkov}, \bibinfo{author}{G.~Falkovich}, and
867:   \bibinfo{author}{I.~Kolokolov}, \enquote{\bibinfo{title}{Intermittent
868:   dissipation of a passive scalar in turbulence},}
869:   \emph{\bibinfo{journal}{Phys. Rev. Lett.}}
870:   \textbf{\bibinfo{volume}{80}}~(\bibinfo{number}{10}),
871:   \bibinfo{pages}{2121--2124} (\bibinfo{year}{1998}).
872: 
873: \bibitem[{Balkovsky and Fouxon(1999)}]{Balkovsky1999}
874: \bibinfo{author}{E.~Balkovsky} and \bibinfo{author}{A.~Fouxon},
875:   \enquote{\bibinfo{title}{Universal long-time properties of {L}agrangian
876:   statistics in the {B}atchelor regime and their application to the passive
877:   scalar problem},} \emph{\bibinfo{journal}{Phys. Rev. E}}
878:   \textbf{\bibinfo{volume}{60}}~(\bibinfo{number}{4}),
879:   \bibinfo{pages}{4164--4174} (\bibinfo{year}{1999}).
880: 
881: \bibitem[{Plasting and Young(2006)}]{Plasting2006}
882: \bibinfo{author}{S.~Plasting} and \bibinfo{author}{W.~R. Young},
883:   \enquote{\bibinfo{title}{A bound on scalar variance for the
884:   advection--diffusion equation},} \emph{\bibinfo{journal}{J. Fluid Mech.}}
885:   \textbf{\bibinfo{volume}{552}}, \bibinfo{pages}{289--298}
886:   (\bibinfo{year}{2006}).
887: 
888: \bibitem[{Doering and Thiffeault(2006)}]{DoeringThiffeault2006}
889: \bibinfo{author}{C.~R. Doering} and \bibinfo{author}{J.-L. Thiffeault},
890:   \enquote{\bibinfo{title}{Multiscale mixing efficiencies for steady sources},}
891:   \emph{\bibinfo{journal}{Phys. Rev. E}}
892:   \textbf{\bibinfo{volume}{74}}~(\bibinfo{number}{2}),
893:   \bibinfo{pages}{025301(R)} (\bibinfo{year}{2006}).
894: 
895: \bibitem[{Shaw \emph{et~al.}(2007)Shaw, Thiffeault, and Doering}]{Shaw2007}
896: \bibinfo{author}{T.~A. Shaw}, \bibinfo{author}{J.-L. Thiffeault}, and
897:   \bibinfo{author}{C.~R. Doering}, \enquote{\bibinfo{title}{Stirring up
898:   trouble: {M}ulti-scale mixing measures for steady scalar sources},}
899:   \emph{\bibinfo{journal}{Physica D}}
900:   \textbf{\bibinfo{volume}{231}}~(\bibinfo{number}{2}),
901:   \bibinfo{pages}{143--164} (\bibinfo{year}{2007}).
902: 
903: \bibitem[{Thiffeault and Pavliotis(2008)}]{Thiffeault2008}
904: \bibinfo{author}{J.-L. Thiffeault} and \bibinfo{author}{G.~A. Pavliotis},
905:   \enquote{\bibinfo{title}{Optimizing the source distribution in fluid
906:   mixing},} \emph{\bibinfo{journal}{Physica D}}  (\bibinfo{year}{2008}),
907:   \bibinfo{note}{in press},
908:   \eprint{http://dx.doi.org/10.1016/j.physd.2007.11.013}.
909: 
910: \bibitem[{Shaw(2005)}]{ShawGFD2005}
911: \bibinfo{author}{T.~A. Shaw}, \enquote{\bibinfo{title}{Bounds on multiscale
912:   mixing efficiencies},} in \emph{\bibinfo{booktitle}{Proceedings of the 2005
913:   Summer Program in Geophysical Fluid Dynamics}}, \bibinfo{pages}{291}
914:   (\bibinfo{publisher}{Woods Hole Oceanographic Institute},
915:   \bibinfo{address}{Woods Hole, MA}, \bibinfo{year}{2005}),
916:   \eprint{http://www.whoi.edu/cms/files/Shaw\_21281.pdf}.
917: 
918: \bibitem[{Taylor(1954)}]{Taylor1954}
919: \bibinfo{author}{G.~I. Taylor}, \enquote{\bibinfo{title}{The dispersion of
920:   matter in turbulent flow through a pipe},} \emph{\bibinfo{journal}{Proc. R.
921:   Soc. Lond. A}} \textbf{\bibinfo{volume}{223}}~(\bibinfo{number}{1155}),
922:   \bibinfo{pages}{446--468} (\bibinfo{year}{1954}).
923: 
924: \bibitem[{Aris(1956)}]{Aris1956}
925: \bibinfo{author}{R.~Aris}, \enquote{\bibinfo{title}{On the dispersion of a
926:   solute in a fluid flowing through a tube},} \emph{\bibinfo{journal}{Proc. R.
927:   Soc. Lond. A}} \textbf{\bibinfo{volume}{235}}~(\bibinfo{number}{1200}),
928:   \bibinfo{pages}{67--77} (\bibinfo{year}{1956}).
929: 
930: \bibitem[{Rhines and Young(1983)}]{Rhines1983}
931: \bibinfo{author}{P.~B. Rhines} and \bibinfo{author}{W.~R. Young},
932:   \enquote{\bibinfo{title}{How rapidly is a passive scalar mixed within closed
933:   streamlines?}} \emph{\bibinfo{journal}{J. Fluid Mech.}}
934:   \textbf{\bibinfo{volume}{133}}, \bibinfo{pages}{133--145}
935:   (\bibinfo{year}{1983}).
936: 
937: \bibitem[{Okabe(2006)}]{OkabeGFD2006}
938: \bibinfo{author}{T.~Okabe}, \enquote{\bibinfo{title}{A particle-simulation
939:   method to study mixing efficiencies},} in
940:   \emph{\bibinfo{booktitle}{Proceedings of the 2006 Summer Program in
941:   Geophysical Fluid Dynamics}} (\bibinfo{publisher}{Woods Hole Oceanographic
942:   Institute}, \bibinfo{address}{Woods Hole, MA}, \bibinfo{year}{2006}),
943:   \eprint{http://www.whoi.edu/cms/files/Taka\_21241.pdf}.
944: 
945: \bibitem[{Chertock \emph{et~al.}(2008)Chertock, Kashdan, Kurganov, and
946:   Doering}]{Chertock2008}
947: \bibinfo{author}{A.~Chertock}, \bibinfo{author}{E.~Kashdan},
948:   \bibinfo{author}{A.~Kurganov}, and \bibinfo{author}{C.~Doering},
949:   \enquote{\bibinfo{title}{A fast explicit operator splitting method for
950:   passive scalar advection by a random stirring field},}
951:   (\bibinfo{year}{2008}), \bibinfo{note}{in preparation}.
952: 
953: \end{thebibliography}
954: 
955: \end{document}
956: