Академический Документы
Профессиональный Документы
Культура Документы
J. M. OTTINO
Dcprrrtrnc~nt01' C l ~ r r , ~ i r , uE11qirlrc~ri11g,
l
Unir.c~r.sitj01' Mu.s.srrc~hu.sc~tt.s
CAMBRIDGE
UNIVERSITY PRESS
P U B L I S H E D B Y T H E P R E S S S Y N D I C A T E OF T H E U N I V E R S I T Y O F C A M B R I D G E
The Pitt Building, Trumpington Street, Cambridge CB2 IRP, United Kingdom
CAMBRIDGE UNIVERSITY PRESS
The Edinburgh Building, Cambridge CB2 2RU, United Kingdom
40 West 20th Street, New York, NY 1001 1-421 I , USA
I 0 Stamford Road. Oakleigh. Melbourne 3 166, Australia
A catalogue record for this book is available from the British Library
Ottino, J. M.
The kinematics of mixing : stretching, chaos, and transport 1 J. M.
Ottino.
p. cm.
Bibliography : p.
Includes index.
ISBN 0 521 36335 7. ISBN 0 52 1 36878 2 (paperback)
I. Mixing. I. Title.
TP156.M5087 1989
660.2 '84292-dc 19
88-30253
I Introduction
I I Physical picture
1.2 Scope and early works
1.3 Applications and geometrical structure
1.4 Approach
Notes
Author index
l~11crethe right hand side is the Eulerinn velocity field (a solution of the
Nal ier Stokes equations. for example) and the initial condition corresponds
the initial configuration of the line o r surface placed o r fed into the
tlolv ( S represents the location of the initial condition x = X ) . Seen in this
light. the problem can be formulated by merging the kinematical founda-
tions of fluid mechanics (Chapters 3 and 3 ) and the theory of dynamical
s!\lems (Chapters 5 ; ~ n d6). The approach adopted here is to analyze
siml'lc protypical flows to enhance intuition and to extract conclusions
of general validity. Mi.uincl i.s strc~tchinyund ,jOMiny and srrc~tchinycrnd
,ji)ldir~gis r l ~ r,fingc~rprint 01' chlros. Relatively simple flows can act as
prototypes of real problems and provide a yardstick of reasonable
expectations for the completeness of analyses of more complex flows.
Undoubtedly, I expect that such a program would facilitate the analysis
of mixing problems in chemical, mechanical, and aeronautical engineering,
physics, geophysics, oceanography, etc.
The plan of the book is the following: Chapter 1 is a visual summary to
motivate the rest of the presentation. In Chapter 2 I have highlighted,
whenever possible, the relationship between dynamical systems and
kinematics as well as the usefulness of studying fluids dynamics starting
with the concepts of motion and ,flow.2 Mixing should be embedded in a
kinematical foundation. However, I have avoided references to curvilinear
co-ordinates and differential geometry in Chapters 2 and 4, even though
it could have made the presentation of some topics more satisfying but the
entire presentation slightly uneven and considerably more lengthy. The
chapter on fluid dynamics (Chapter 3) is brief and conventional and
stresses conceptual points needed in the rest of the work. The dynamical
systems presentation (Chapters 5 and 6 ) includes a list of topics which I
have found useful in mixing studies and should not be regarded as a
balanced introduction to the subject. In this regard, the reader should
note that most of the references to dissipative systems were avoided in
spite of the rather transparent connection with fluid flows.
A few words of caution are necessary. Mixing is intimately related to
flow visualization and the material presented here indicates the price one
has to pay to understand the inner workings of deterministic unsteady
(albeit generally periodic in this work) two-dimensional flows and three-
dimensional flows in general. However, we should note that the geometrical
theory used in the analysis will not carry over when v itself is chaotic.
Though mixing is still dependent upon the kinematics, the basic theory
for analysis would be considerably different. Also, even though many of the
examples presented here pertain to what is sometimes called 'Lagrangian
turbulence', the reader might find a disconcerting absence of references
to conventional (or Eulerian) turbulence. In this regard I have decided to
let the reader establish possible connections rather than present some
feeble ones.
I give full citation to articles, books, and in a few cases, if an idea is
unpublished, conferences. When only a last name and a date is given,
particularly in the case of problems or examples, and the name does not
appear in the bibliography, it serves to indicate the source of the problem
or idea. I t is important to note also that some sections of the book can be
rcgnrded as work in progress and that complete accounts most likely will
(‘allow, expanding over the short descriptions given here; a few of the
problems, those at the level of small research papers are indicated with
an asterisk (*). In several passages I have pointed out problems that need
work. Ideally, new questions will occur to the reader.
Notes
I Even the terminology is complicated. For example. in chemical engineering the terms mixing,
agitation. and blending are common (Hyman, 1963; McCabe and Smith, 1956, Chap.
2. Section 9; Ulbrecht and Patterson, 1985). The terms mixing. advection, and
stirring appear in geophysics; e.g.. Eckart. 1948; Holloway and Kirstmannson. 1984.
Inevitably. different disciplines have created their own terminology (e.g., classical reaction
engineering, combustion, polymer processing, etc.).
2 Kinematics appears as an integral part of books in continuum mechanics but much less so
in modern fluid mechanics. There are exceptions ofcourse: Chapters V and VI of the work
of Tietjens based on the lecture notes by Prandtl contain and unusually long description
01"deformation and motion around a point (Prandtl and Tietjens, 1934).
-
/
Acknowledgments
/-
This hook grew out from a probably unintelligible course given in Santa
EL:. Argentina, in July 1985, followed by a short course given in Amherst,
Massachusetts, also in 1985. Most of the material was condensed in eight
lectl~rcsgiven at the California Institute of Technology in June 1986,
\vhcrc the bulk of the material presented here was written.
.rhc connection between stretching and folding, and mixing and chaos,
bcc;~rnc transparent after a conversation with H. Aref, then at Brown
IJni\crsity, during a visit to Providence, in September 1982. 1 would
particularly like to thank him for communicating his results regarding
the 'blinking vortex' prior to publication (see Secton 7.3), and also for
many research discussions and his friendship during these years. I would
like to thank also the many comments of P. Holmes of Cornell University,
on a rather imperfect draft of the manuscript, the comments of J . M.
Grcenc of G. A. Technologies, who provided valuable ideas regarding
syrnnictries as well as to the many comments and discussions with S.
Wiggins and A. Leonard, both at the California Institute of Technology,
during my stay at Pasadena. I am also grateful to H. Brenner of the
Massachusetts Institute of Technology, S. Whitaker of the University of
California at Davis, W. R. Schowalter of Princeton University, C . A.
Trucsdell of Johns Hopkins University, W. E. Stewart of the University
of Wisconsin, R . E. Rosensweig and Exxon Research and Engineering,
and J . E:.. Marsden, of the University of California at Berkeley, for various
comments and support. I am also particularly thankful to those who
Supplicd photographs or who permitted reproductions from previous
Pllhlications (G. M . Corcos of the University of California at Berkeley,
R . ('hevray of Columbia University. P. E. Dimotakis and L. G. Leal.
the California Institute of Technology, R. W. Metcalfe of the
University of Houston, D . P. McKenzie, of the University of Texas at
Allstin. A. E. Perry of the University of Melbourne, 1. Sobey of Oxford
IJni\crsity, and P. Wellnnder of the University of Washington).
1 a m particularly indebted to all my former and present students, but
xiv A~~know1cdgmc~nt.s
striation
thickness
stretching
@
-'
-
concentration profile
initial
condition
diffusion
distance
breakup
Scope trtltl eurljs works 3
s e sthe layers, say s, and s,, and :(s, + s,,) is called the
~ l ~ i ~ . k n e sof
slr.;t,~;o,~ ~liick~ios.s (see Ottino, Ranz, and Macosko, 1979). The amount
of intcrfi~cialarea per unit volume, interpreted as a structured continuum
prc)pHty. is called the i r l t c ' r r i ~ ~ ~ tLIrett (I,..Thus, if S designates
~ r i ~dcnsit~,,
~l
t1,c area within a volume V enclosing the point x at time t ,
S
u,.(x, t ) = lim .
v-0 v
~ ~ loflthe
~ above
c concepts require modification if the fluids are miscible
or immiscible. I f the fluids are immiscible, at some point in the mixing
process the striations o r blobs d o not remain connected and break into
snlnllcr fragments (Figure 1.1.1 (c)). At these length scales the interfaces
arc not prrs,sir.c~and instead of being convected (passively) by the flow,
they nlodify the surrounding flow, making the analysis considerably more
complicated (in this case the interfaces are termed uc.tioe, see Aref and
Tryggvuson, 1984). If the fluids are miscible we can still track material
volumes in terms of a (hypothetical) non-diffusive tracer which moves
with the mean mass velocity of the fluid o r any other suitable reference
velocity. Designated surfaces of the tracer remain connected and diffusing
species traverse them in both directions.' However, during the mixing
process, connected iso-concentration surfaces might break and cuts might
rcvcal islands rather than striations (Figure I . l . l ( h ) ) . In this case the
specification of the concentration fields of 11 - 1 species constitutes a
complete description of mixing but it is also clear that this is a n elusive goal.
Thus, it is apparent that the goal of mixing is reduction of length scales
(thinning of material volumes and dispersion throughout space, possibly
involving breakup), and in the case of miscible fluids, uniformity of
concentration.' With this as a basis, we discuss a few of the ideas used
to describe mixing and then move to examples before returning to the
problem formulation in Section 1.4.
Examplc'.s
( 1 ) The striation t h ~ c k n e s s ,s. Figure 1.1.1. is important in processes
in\,olving diffusing and reacting fluids a n d represents the distance that
the molecules must diffuse in order t o react w ~ t heach other. In simple
Gases. .s can be calculated exactly with a knowledge of the velocity
field (see Chella a n d O t t i n o (1985a) a n d examples In C h a p t e r 4 ) .
t i i ) Molten polymers are often mixed ( a n oper;ation often referred to a s
blending) to produce materials with ~ ~ n ~ properties.
que F o r example.
In the mnnuf;acture o f barrier polymer films ~t might be desirable to
Produce structllrcs with effective permeability. This requires that
Introduction
Figure 1.3.3. Mixing in a creeping flow. The ligure shows the deformation of
a material region in a journal bearing flow when it operates in a time-periodic
hishion lexpcriment from Chaiken rt rrl. (1986)), for a complete description see
Scction 7.4. (Reproduced with permission.)
12 Introduction
scales. Figure 1.3.4 (see color plates) shows an image processed two-
dimensional structure produced by mixing and preserved by quenching
-
o f two immiscible molten polymers. In this case there has been a complex
process of breakup and coalescence. As indicated earlier, the character-
Figure 1.3.6. Concentration of a turbulent round jet fluid injected into water
at Reynolds number 2,300, measured by laser induced fluorescence; the cut
is along a plane including the axis of a symmetry of the jet. (Reproduced with
permission from Dimotakis. Miake-Lye, and Papantoniou (1983).)
Approuch 13
1.4. Approach
In spite of its overwhelming diversity, fluid mixing is basically a process
ln~olvinga reduction of length scales accomplished by stretching and
folding of material lines or surfaces. In some cases the material surface
Or line in question is placed in the flow and then subseqently stretched
le.%- Figures 12.1, 1.3. I 3). in others, the surface is continuously fed into
'he flow (cg.. Figures 1.3.5 6). Thus. at the most elementary level (i.e..
without averaging but at the continuum level) rnising c.onsists of' stretching
lrr~d,fbldii~~g
of ,fluid filtrmc~nts,lrrld distribution tllrou~ghoutspuce, accom.
panied by breakup if the fluids are sufficiently different, and simultaneous
diffusion of species and energy (Figure 1.1.1 ). In the most general case,
various chemical species might be reacting. We seek an understanding of
this process in terms of simple problems which can serve as a 'window'
for more complicated situations. O u r approach is to combine the kine-
matical foundations of fluid mechanics with dynamical systems concepts,
especially chaotic dynamics. The objective throughout is to gain insight
into the working of mixing flows. The goal is not to construct detailed
models of specific problems but rather to provide prototypes for a broad
class of problems. Nevertheless. we expect that the insight gained by the
analysis will be important in practical applications such as the design of
mixing devices and understanding of mixing experiments.
Mixing is also inherently related to flow visualization. However,
contrary to popular perception the 'unprocessed' Eulerian velocity field
gives very little information about mixing and the typical ways of
visualizing a flow (streamlines. pathlines, and to a lesser degree. streaklines)
are insufficient to completely understand the process. As we shall see, our
problem begirls rather rharl ends with the specificurion of v(x, r). The solution
of
dxidr = V ( X , I )
with x = X at time t = 0. x = @,(X),which is called the flohv or motion,"
provides the starting point for our analysis. In even the simplest cases this
'solution' might be extremely hard to obtain. Actually. the impossibility of
integrating the velocity field in the conventional sense is the subject of
much of Chapters 5 and 6, where the modern notion of irlteyrability is
introduced. The kinematical foundations lie in an understanding of the
point transformation x = @,(X).
We consider the following sub-problems:
( 1 ) Within the framework of x = @,(X):mixing of a single fluid or similar
fluids.
The basic objective here is to compute the length (or area) corresponding
to a set of initial conditions. As we shall see only in a few cases can this
be done exactly and in most of these the length stretch is mild. The best
achievable mixing corresponds to exponential stretching nearly everywhere
and occurs in some regions of chaotic flows. However. under these
conditions the (exact) calculation of the length and location of lines and
areas is hopelessly complicated. As we shall see in Chapter 5 even extremely
simpliljcd ~ O R might S be inherently chaotic and a corripleie char;~cterization
is not p o ~ i b l c .F o r example. from a dynamical systems viewpoint Re shall
,,,that ifthe system possesses horseshoes we have infinitely many periodic
and with it the implication that we cannot possibly calculate
;dl of t h e m Fortunately. a s far ;IS mixing is concerned we are
intercstcd in low period events, since we want to achieve mixing quickly.
~ ~ ~ ~ ~there t h is calways l e ~the~intrinsic limitation of being unable t o
ca]culatc p ~ c i s information e (most practical problems involve stretchings
of 10' o r higher) such a s length stretch a n d location of material
surf;lccs, Note that this is true even though none o f the flows discussed
in Chapters 7 and 8 is turbulent in a n Eulerian sense. Rather, the
previous findings should be used t o establish the limits o f what might
constitute reasonable answers in more complicated flows (real turbulent
flows come immediately t o mind). The prohlcm /irr.cl is /io\c, to hcsr
~ ~ / 1 1 1 1 . 1 1 c ~ r c ~ 1 ~t/]o
; : 1 ~ t i ~ i - ~ i kt~o\vit~q
i~q, hgf:)rP/lut1dt/iiit u c o t ~ i p l ~~ ~i ~/ ~ ~ ~ r ~ ~ ~ ~ t
i,s ; l ~ l p o , ~ , ~ ; l ~ l ~ ~ .
( 2 ) Within the framework of a family of flows x, = @,,(X,), s = 1, . . .. ,1';
each of the motions is assumed t o be topological (see Section 2.3): mixing
of similar diffusing and reacting fluids.
This case corresponds t o the case of mixing of two streams. composed
ofpossibly several species that are rheologically identical, i.e., they have the
same density. viscosity. etc., a n d have n o interfacial tension. Concur-
rently with the mechanical mixing there is mass diffusion, and possibly.
chemical reaction. However. for simplicity, w e will assume that neither
the diffusion nor t h e reaction affects t h e fluid m o t i o n . ' T h i s case is
discussed in C h a p t e r 9 a n d corresponds t o the case of lumellur structures.
( 3 ) Mixing of different fluids; case in which the motions are non-
topological. i,e,. there is breakup a n d o r fusion of material elements.
In this case. the mixing of two o r more fluids leads t o breakup a n d
coalescence of material regions. This problem is complicated a n d only a
few special crlses belonging t o this category are discussed in Chapter 9.
by decon~posingthe problem into I o ~ o land qlobiil components.
An outline of the organization o f the rest o f t h e chapters is the following:
2 describes the kinematical foundations a n d Chapter 3 presents
a b r i e f n i c r v i e , o f fluid mechanics. Where;~sthe material of Chapter 2 is
' n d i s ~ e n ~ i i b l eI;~rge
. p;lrts o f Chilpter 1 were ; ~ d d e dt o provide b;ll;~ncc.
4 focuses on ;l f e ex;lmp]es~ which can be solved in detail and
ends in ;I riit11cr dcfc;ltist note t o provide a bridge for the study o f chaos.
5 iwcrcnts ;I pener.11 discussion o f dyni~mic;~l systems ; ~ n dC h i ~ p t ~ r
16 Introduction
Notes
1 Obviously, one o f the diffusing species can be temperature, as is the case of mixing
of fluids with different initial temperatures or processes involving exothermic chemical
reactions.
2 A gross, but popular, measure of the concentration variation is given by the intensity of
srqreqution, I . If ~ ( xdenotes
) the concentration at point x and ( . ) denotes a volume
average, I is defined as [(((.- ( c ) ) ~ ) ] ' / ' (Danckwerts, 1952).
3 Many ofthese references inspired additional work. Some, however, were largely ignored.
4 Subsequently, this problem took a life of its own and much research followed. See for
example Rallison (1984).
5 This idea was not widely followed and most of the mixing work in the area of drop breakup
and coalescence in complex flow fields resorts to population balances where breakup and
coalescence are taken into account in a probabilistic sense.
6 This paper contains many good ideas, however, it has remained largely ignored
by the mixing community.
7 In the past few years there has been a revival of this idea (e.g., Spalding, 1976, 1978b;
Ottino, 1982).
8 The theory also found use in two-phase mixing. An early reference is Shinnar (1961).
9 The analysis of mixing of viscous fluids has been largely confined to polymers
(Middleman, 1977; Tadmor and Gogos, 1979).A follow up paper, written in the context
of the mixing of glasses. is Cooper (1966). Even though Cooper's treatment of the
kinematics is at the same level as Mohr, Saxton, and Jepson (1957), there is
substantially more, since it deals explicitly with mass diffusion. This paper, however,
has remained largely ignored.
10 This approach works well for pre-mixed reactors with slow reactions, but it is not suited
for diffusion controlled reac!ions, such as in combustion. Curiously enough, the
participation of chemical engineers in subjects dealing with non-pre-mixed reactors has
been relatively minor and in spite of complex chemistry, the area has become largely the
domain of mechanical and aerospace engineering researchers. See the discussion (pp. 100-
102) following the paper by Danckwerts (1958). Much work followed along these
lines. For a summary, see Nauman and Buffham (1983).
I I With the possible exception of the very last example of the last chapter. we do not consider
fluid-solid systems.
12 See for example Veronis, 1973: Rhines. 1979; 1983. and the articles by Holland
('Ocean circulation models'. pp. 3 4 5 ) . Veronis ('The use of tracer in circulation
studies'. pp. 169188), and Rhines ('The dynamics of unsteady currents'.
pp. 18%318), in Goldberg er a l . , 1977.
13 nllld dynaniical aspcct"arc discussed by McKenzic, Roberts, and Weis.,
( 1974,, see MeKenfie 11983).
14 cxpcct;ltion was not confirmed in their study. It is apparent that there is a
for more cxpcr~mcnthadopting primarily a 'Lagrangian viewpoint'.
F ~ ex:lmple.
, ~ ;lver;igc cluster s i x and cluster size distribution, intcrfiIcial
per llnit ;Ij-c;~,ctc. (see Sax :und Ottino. 1985).
16 ,mpingcmcnt mixing I similar t o 'the stopped flow' mcthod devised to study the
kinctl~sof very fklst r c i l c ~ i ~An review
~. of the technique by the inventors o f t h e method
is by Roughton and <'hanee (1963). C h a p . XIV.
17 flow.ia the tcrm preferred in dynamical systems. Inorro,! the tcrm ofchoice in continuum
mcchan~cs.The re;ider should bc warned about possible confusion o f terms; the
word /lo\, i \ used often in the convcntioni~lfluid mechanical scnsc, the term f,r;.r;frc,
has :I prcclsc definition in dyn:lmical systems and ergodic theory (see Waters. 1 9 ~ 2 ,
p. 40). To avoid confusion wc will often use the term /luic] ~ > t ; . \ - , , ! ~ .
18 That is. the fluid mechanics governs the transport processes but not the (Ither
way a r o u l ~ dT. h c r c i 110 clean way of incorporirting thcaccouplings at the present timc.
Flow, trajectories, and deformation
In the first part of this chapter we record the basic kinematical foundations
of fluid mechanics, starting with the primitive concept of particle and
motion, and the classical ways of visualizing a flow. In the second part
we give the basic equations for the deformation of infinitesimal material
lines, planes, and volumes, both with respect to spatial, x, and material,
X, variables, and present equations for deformation of lines and surfaces
of finite extent.
2.1. Flow
The physical idea of flow is represented by the map or point transformation
(Arnold, 1985, Chap. 1 )
x = @,(X) with X = @,=,(X), (2.1.1)
i.e., the initial condition of particle X (a means of identifying a point in
a continuum, in this case labelled by its initial position vector) occupies
the position x at time t (see Figure 2.1 .I ). We say that X is mapped to x
after a time t.'
In continuum mechanics (2.1 .I ) is called the motion and is usually
assumed to be invertible and differentiable. In the language of dynamical
systems a mapping
@,(XI x
+ (2.1.2)
is called a Ck diffeomorphism if it is 1-1 and onto, and both a,(.)and
its inverse are k-times differentiable. If k = 0 the transformation is called
a homeomorphism. In fluid mechanics, k is usually taken equal to three
(see Truesdell, 1954; Serrin, 1959). Also, the transformation (2.1.1 ) is
required to satisfy
or alternatively,
J = det(D@,(X))
whert! D denotes the operation i, (),/ax,, i . e . , derivativeswith respect to the
configuration, in this case X. I f the Jacobian J is equal to one
flow is called isochoric.
~ h requirement
c (2.1.3) precludes two particles, X, and X,, from
occup),ing the same position x at a given time, or one particle splitting
into ~LI'O:i.e., notz-topologicul motions such as breakup or coalescence are
not allowed (Truesdell and Toupin, 1960, p. 510).
In the language of dynamical systems, the set of diffeomorphisms (2.1.1 )
for all particles X belonging to the body Vo is called the flow (i.e., a
one-parameter set of diffeomorphisms) and is represented by
r 1 - 1
- ,@,(XI; = @ , ( X i
,XI
motion
-
20 Flow, trujrc.tories ur~tltlt$)rmtrtior~
-
and it is the velocity of the particle X. The acceleration, a , is defined as
a (c'.'@,(~)/i?t')I~ = a(X, t).
Any function G (scalar, vector, tensor) can therefore be viewed in two
different ways:
G(X, t ) = Lagrangian or material
i.e., follows the motion of a particular fluid particle, or
G(x, t ) = ~ u l e r i a n % r spatial
i.e., the property of the particle X that happens to be at the spatial location
x at time t.
Thus, v(X, t ) is the Lagrangian velocity and v(x, t14 is the Eulerian
velocity. In most classical problems in fluid mechanics it is enough to
defined as
-
obtain the spatial description. The material (or Lagrangian) derivative is
DGIDt (('G/?t)1,
representing the change of property G with time while following the motion
of particle X, whereas the standard time derivative is
?G/?t = (?G/?t)lx,
and represents the change at a fixed position x. The relationship between
the two is easily obtained from the chain rule
DG/Dt = ?G/it + V-VG,
where V is defined as V = (?/?.u,)e, (see Appendix). The expression allows
the computation of the acceleration at (x, t ) without computing the motion
first.
The expression for the material time derivative of the Jacobian of the
flow is known as Euler's formula (Serrin, 1959, p. 131 ; Chadwick, 1976,
p.65).
DJIDt = J(V.v) = J tr(Vv)
and is the basis of a kinematical result known as the trtrnsport theorem.
Veloc,ity,acceleration, Layrangian and Eulerian
and since the domain V, is not a function of time, expanding the material
derivative we obtain
r: I",
G(x. r ) d r = jv![& + G(V. v)] dc.
Problem 2.2 .J
Show that if A(r) is invertible, d(det A)/dr = idet A) t r [ ( d A / d r ) ' A ' l .
Problem 2.2.2
that if I I ; / = e l [v~,]then ;iv,\= @ , ; i V , , ] , i.e., the boundary is
mapped into the hound;iry. Usua]]y, this is taken to be that the surfiice
Of a material body consists of the siime p;irticles (von Mises and
22 Flow, trajectories and defbrmution
-
the sum runs from 1 to N. Individual velocities are defined as
v, (a@ls)(x,)~at)l,= v,(x,, t )
v -x
and the average mass velocity is defined as
(P,/P)~,.
The time derivative of any function G following the motion of the species
denoted s, is given by
DG'"/Dt = aG/at + v;VG,
and the relative velocities are defined by
Us = v , - V .
2.4.3. Stveaklines
The picture at time t of the streakline passing through the point x' is the
curved formed by all the particles X which happened to pass by x' during
the time 0 < t' < t. Physically, it corresponds to the curve traced out by
a non-diffusive tracer (i.e., the particles X of the tracer move according
to x = @,(X))injected at the position x'.
Example 2.4.1
Compute the pathlines, streamlines, and streaklines corresponding to the
unsteady Eulerian velocity field
"1 = x l l ( l +
t), v 2 = 1. 8
To compute the pathlines we solve
d.u,/dt = x1/(l + t), dx2/dt = 1
with the condition x, = X,, x, = X,, at t = 0.
.Y, = X l ( l + t), s, = X, +t
and, eliminating t, we obtain
.Y1 - x 1 x 2= X1(l - X2).
24 Flow, tr~ljectories~ l n ddeformrrtion
i.e., the particles move in straight lines. The streamlines are given by the
solution of
dx,~d= s x,/(l + t), dx,/ds = 1
with the condition x, = x y , x, = x;, at s = 0, while holding r constant.
Thus, the streamline passing by x, = xy, x , = x;, is given by
x , = xy exp[s/(l + t ) ] , x, = x,0 + s,
and eliminating the parameter s,
( 1 + t ) ln(x,jxy) = x , - x,,0
which shows that the streamlines are time dependent. T o get the streakline
passing through x',, x; we first invert the particle paths at time t'
+
X I = xi/( l t'), X , = ,u; - t ' ,
which indicates that the particlt X , , X , will be found at the position x i , x;
at time t'. The place occupied by this particle at any time t is found again
from the particle path as
+ +
x , = x',(l t ) J ( l t ' ) , +
x, = .u; - t' t ,
and is interpreted as: the particle which occupied position x;, x; at time
t' will be found in position x,, x , at time t . Eliminating ['we get the locus
of the streakline passing by x',, x;:
x,x, x , ( l + x ; + t ) + x',(I + t ) = 0
-
which shows that the streaklines ure ulso ,functions of' time (plots
corresponding to this example are given by Truesdell and Toupin, 1960,
p. 333).
~.rtcrnp/d.5.1
consider the streamlines, pathlines, and streaklines in the shear flow
V, = 1 + t a n h .u,, V, = 0
subjected to a time dependent perturbation
I.', = 2ri sech 27-r.~~
tanh 27-r.~~
sin[2rc(.u1 - r)],
I,> = 2ri sech 2 n s 2 cos[2n(.uI - t)].
where tr is the amplitude of the fluctuations. This problem was analyzed
by Hnma (1962) to serve as a warning in the interpretation of flow
visuoli~ationstudies since it is a case where the streaklines are considerably
more complicated than the streamlines and pathlines. The motion of the
fluid particles is governed by the system of equations
rls,ldr = I . , ( x , ,.Y,, t), d.u2/dr = r',(.u,, x, t )
where I . , = V, + I.', and I . , = V, + r;. Examples of computed streaklines
and pathlines are shown in Figure E2.5.1 (computation by Franjione,
1987).
Exurnple 2.5.2
Calculate the possible streamlines corresponding to the class of flows
V = (Ov)' ex with V . v = 0 (this material is used repeatedly throughout the
book ).
For simplicity consider first two-dimensional flows. By continuity
-
( 9 . v = 0. the sum of the eigenvalues being equal to ~ e r o )we
, can infer
of the character of the three-dimensional flow. We consider a cut
of\. = ( v \ . )xI by settings, 0. In general, the two-dimensional version of
( V v ) ' . (Vv);, = L. has all non-zero components, but by a suitable trans-
formation R. R-L-R- '. L can be written in one of the three possible ways
(we follow Hirsch and Smale. 1974. Chap. 5. Section 4):
focus o r focus o r
center
improper improper
node node
spiral spiral
sink sourcc
tr L
simple saddle
shear
outside parabola A > 0
28 Flow, trujectories und deformution
problem 2.5.3
consider the flow c, = Gx,, c2 = KGx,, where - 1 < K < 1. Show that the
are given by x: - Kx: =constant, which corresponds to
ellipses with axes ratio ( l . ' ( ~ ( ) "if~K, < 0, and to hyperbolas forming an
angle / j = arctan(l.'K)"' between the axis of extension and x,, if K > 0
(see Figure P2.5.3). Prove that this flow is the most general representation
of a linear isochoric two-dimensional flow.' '
Problem 2.5.4
show that the pathlines corresponding to the flow v, = ux,, c, = -u(x, - bt)
represent a circular motion about a center moving with a velocity
= (hr, h,!u). Find the streamlines and particle paths corresponding to
= ( . ~ , t ,- . Y , ) Show that the streamlines are given by x,x; = const. (for
this and other examples, see Patterson, 1983).
Pvohlrm 2.5.5
Consider the one-dimensional time-periodic Eulerian velocity field, v, =
C; cos[k(.u - ct)]. Find the velocity and time averaged velocity experienced
by a material particle.
Pvohlrm 2.5.6
Given the Eulerian velocity field
v = x-L,L = const.,
obtain the Langrangian velocity field. Generalize for L = L(t).
J = det F.
-
N -= d ~ / l d ~ l .
For comparison, the volumetric change J (dv/dV) is given by
(2.7.6)
The orientations of the vectors dx and da are given by
Example 2.7.1
Consider a flow @,(X)that transforms the body V,, into V,; i.e.,
P r o b l ~ ~ 2.7.3
m
Obtain ( 2 . 7 . 4 )through (2.7.8).
Problem 2.7.4
Prnvc that DF/Dt =(vv)~-F.
Pvoblcm 2.7.5
Calculate the optimum orientation M for maximum stretching in a given
time I , in a simple shear flow I., = 7w2, L., = 0, r , = 0.
and
v,,, - n = ( D : nn)(dxl,
where n = dx/ldx(. Note that since R : nn = 0, R does not contribute to
the velocity in the direction normal to the sphere ldxl = constant and the
contribution of R is wholly tangential to the sphere (see Figure 2.8.1 ).
Problem 2.8.1
Where is v,,,.n maximum'?
Pvoblem 2.8.2
Prove that R : nn = 0.
Problem 2.8.3
Show that the vector R - n can be written as :(o x n) where w =
2(Q,,, Q,, , Q , ,) and w = V x v (ois called the vorticity vector, see Section
3.8). Present an argument to show that o represents angular rotation
with speed 10/21 in a plane perpendicular to o .
Pvoblem 2.8.4
Show that v,,, can be written as
.
v,,, = fV(dx D . dx) + i ~ xpdx.
Pvoblem 2.8.5
Show that near a surface with normal n such that v =0 (and o - n = 0 ) D
can be written as
D = (ox n)n + n(w x n)
if V.v =0 (Caswell, 1967; Huilgol, 1975).
Figure 2.8.1. Velocity field around a point
Kinemutics of deformation: rute of' struin 35
prohlrm 2.8.6
s t L l d \ the possibility of expanding the relative velocity field near a surface
il S
v = $h(x)I-. x
(/)(x)is a scalar function and I- is a matrix (Perry and Fairlie, 1974).
p l . ~ h l c n2.8.7
~
Obtain the general form of the velocity near a solid surface.
Problem 2.9.2
Starting with
D(ln q)/D/ = V-v - D: nn,
show that
D(ln q)/D/ = (V,-v,) - KnU,,,
where the velocity is written as v = v, + Nr.,,, such that I:,, is the speed
normal t o the surface, Vs is the surface gradient operator, and K, is
the mean curvature, K A = - VS . n.
,,itablc formula is
and since
( m e ( V v ) ) . mIdxl.
Noting that
we obtain.
where the h is the normal of the Frenet triad (m, tangent; h, normal; b,
binormal. See Aris, 1962, p. 40 o n ) and the curvature ti, is defined as
dml(dx(= ~ , h .
Special case
If x , = x z or if
x2
v.m/ =O
x1
S,.
v.htiL Idxi.
Problem 2.10.1
Denote by G(x, t ) a scalar function. Show that
Problem 2.10.2
Obtain the evolution of a finite area (Equation (2.10.4))starting with
D(ln q)/Dt = V. v - D:nn.
Hint: A rough derivation, without using the theory of calculus on surfaces,
can be obtained by using the identity
v x [f x v] = (Vf).v - (Vv).f + f(V.v) - v(V.f).
Another possibility is to integrate
D(ln q)lDt = (Vs.vs) - KAU,.
Problem 2.10.3
Using Equation (2.9.4), prove (2.10.3).
Problem 2.1 1 .I
Compute the rate of rotation of the maximum and minimum directions
of Stretching in the shear flow L!, = $.u,, u, = 0, 0, = 0. DO they actually
rotate?
Problem 2 .I 1.2
Find the crror in the following reasoning (given in Ottino, Ranz, and
Macosko. 1981): Since D(ln ;.)/Dt = D:mm and D(ln rl)/Dt = V.V - D: nn.
n = I. we have D(ln ;.)/Dl + D(ln q)/Dr = V .v. Hence. if V-v = 0. we
Obtain ;.I/ = 1.
Bibliography
Even though the kinematical foundation of fluid mechanics dates from
the seventeenth century, little is given nowadays in standard works in
fluid mechanics. For modern accounts the reader should consult works
in continuum mechanics. The presentation given here is based on a much
larger body of work. The most comprehensive account is given by Truesdell
and Toupin (1960, pp. 226-793). A brief and lucid account is given by
Chadwick (1976) in Continuum mechunics. A more advanced treatment
is given in Chapter I1 of Truesdell (1977). Other accessible accounts,
slanted towards solid mechanics and fluid mechanics are given by Malvern
(1969) and Aris (1962), respectively, whereas for historical references the
reader should consult Truesdell (1954). A classical work in fluid mechanics
with an unusually long discussion on the kinematics of flow around a
point is Fundamentals of hydro- and aerodynamics by Tietjens, based
on the lectures by Prandtl (see Prandtl and Tietjens, 1934). One of the
best visual demonstrations of Lagrangian and Eulerian descriptions is
given in the movie Eulerian and Layranyian descriptions in ,fluid mechanics.
One clear visual treatment of deformation around a point is given in the
movie Dqformation in continuou.~media. Both films are by J. L. Lumley.
The scripts are given in Illustrated experiments in ,fluid mechanics, produced
by the National Committee for Fluid Mechanics Films, MIT Press,
Cambridge, 1972. The 'kinematical reversibility' of creeping flows is
illustrated in the film Low Reynolds number creeping flows, by Taylor, also
in Illustrated experiments in fluid mechanics, and in the article 'An
unmixing demonstration', by Heller (1960). Both works focused on Couette
flows, which are the exception rather than the rule among all two-
dimensional flows. In most cases reversibility is impossible in practice due
to unbounded growth of initial errors.
Notes
I Throughout this work X E R3.However, most of the results of this chapter are valid
even if x E iWn The extension. however, does not carry over to vorticity since the
relation between an antisymmetric tensor and an axial vector is valid only in iW"
(Truesdell. 1954, pp. 58. 59).
2 See Arnold. 1985, p. 4. Distinguish carefully between the label of the particle. X.
and its position, x .
3 It is well documented that the association of names is incorrect but it is apparently
too late to set the record straight. Both descriptions are actually due to Euler. For
the historical account of credits see Truesdell and Toupin. 1960. p. 327 and Truesdell,
1954. p. 30.
4 In the framework of Equation (2.1.1) v(x. 0 should be interpreted as the velocity
of the particle X which happens to be at x at time I . The solution of most fluid mechanicill
problems yields the Eulerian velocity since, usually, the equations are solved after
bcing formulated in this viewpoint. Historically, the very earliest formulations of
fluid mechanics were formulated using this viewpoint, taking v(x, I ) as a primitive
quantity (for a n authoritative account. see Truesdell. 1954, p. 37). For a n alternative
dcfinition ofthe Eulerian velocity, considering it as the primitive quantity, see Problem
3 3 1.
i I'articles of different species can coexist at the same position x (see Bowen. 1976. for
~ a r i o u sextensions of this concept).
(, Rcciill that the non-autonomous case can always be transformed into a n
;~iltonomousone by defining I = .u,.
7 rhere have been claims that the veloc~tyfield of a n ~ n v i s ~ c ifluid
d in turbulent motion
t n ~ g h tnot be Lipschitz. An early Indication is glven by Onsager, 1949).
i; I o r reasons that w ~ l be l painfully apparent in Chapters 5 and 6, theexamples cannot be. In
scncral, much more complicated than this one.
0 As in Equation (2.1.1 ), we are labell~ngparticles by their initla1 positions. however.
111 this case one has to distinguish carefully between the luhrl of the particle and
it\ pl~c~rnrn atl an arbitrary time, since it can be a source of confusion.
10 It iscommon practice in continuum mechanics to usecapital letters to denote reference
\[ate and lower case letters t o denote present state (see Figure 2.1 . I ) . However, as
I \ also universal practice in dynamical systems in the case of periodic flows. we denote
\ a \ x,,. and subsequent states as x , , x,, etc.
I I 1'111sflow can be realized by means of a four-roller apparatus: the concept is due
ro (iicsckus (1962). See also Section 9.3.1 and Bentley and Leal (19X6a).
12 I'quation (2.7.1) can be interpreted as a matrix multiplication where F operates o n
,I ~ . o l u m nvector d X . We shall not make any distinction between row a n d column
icctors by means of a transpose since the meanlng is always cle;ir from the context
13 [ ' h ~ srelation gives the physical meaning t o the J a c o b ~ a nof the flow ( = d e t F ) .
I4 ['he ho-called left Cauchy-Green strain tensor is F e F ' . The tensor C ' is called
tile Piola tensor.
IS Note that a steady state orientation need not exist. For example the rate of rotation
can be constant o r periodic in time (see Chapter 4). Also, note that the steady state
orientation might be unstable.
Consevvation equations, change of' frame,
and vovticity
In this chapter we record for further use and completeness the equations
of conservation of mass and linear momentum and laws of tri~nsformation
for velocity. acceleration, velocity gradient, etc., for frame transformations
involving translation and rotation. We conclude the chapter by studying
thc equations of motion in terms of vorticity and the streamfunction.
v.1 =
I( St,,pix. tl dl..
tll
jL,
I ~ ~ 1() dr i .= jl DG
, 11 Dr d r>
dt
jb, jL,
[ ~ v d i .= pfdi.
contact Corcca
(3.2.1 )
k > r a n yregion V,, a t any time (where V, can be a n arbitrary control volume).
Pr.ohlcm 3.2.1
f'l.o\e that the Principle of Conservation of Linear M o m e n t u m implies
the continuum version of 'Newton's third law': t(x. n, 1 ) = t ( x . -n. t ) .
T h ~ implies
s that thc information a b o u t tractions a t the point x a n d a n y
with normal n is contained in the tensor T. T h e proof consists of
'.ll~fi~cc
tllrcc steps which \ye will briefly repeat here.
(i) Cuuchy's theorcm
('onsider a (small) region V, surrounding a particle X . T h e n , if L repre-
hents s o m e length scale o f V, we have:
14 C'onserlution rquutions, chunge of irutne, untl lwrtic.it.r
That is, the tractions are locally in equilibrium (see Serrin, 1959, p. 134).
(iii) The third step is to prove that [TI is indeed the matrix
rc~presentutionof a tensor T (see Appendix)
In order to prove that the T,,s are the components of a tensor T , we need
to prove that the components Ti, transform as a tensor. Since n is just a
frcc vector, which is objective,
n' = Q - n ,
and since the traction t transforms as
t' =Q.t,
then. T transforms as'
T'= Q.T.QT.
Problem 3.3.1
(Jsc Cauchy's construction in conjunction with the 'mass balance' to show
that the mass flux in the direction normal to a plane n is given by j - n .
I+licrc j is the mass flux vector. This allows the definition of the Eulcrian
\clocity as v = j/p which can now be regarded as the primitive quantity.
I-athcr than the velocity obtained by means of time differentiation of thc
'notion (Section 2 . 9 ) .
3.4. Cauchy's equation of motion
Using Gauss's theorem, Equation (3.2.2) can be written as
(3.5.1)
'I' = T~ but we note that non-symmetric tensors are indeed possible and
have highly non-trivial consequences as well as practical importance
(Rrcnner, 1984; Rosensweig, 1985).
The tensor T is normally written as
T = -pl+t
-
where p = - tr(T)/3, so that t r ( t ) 0. For a fluid at rest t 0, and
T = - P I . These conditions ident~fyalso an inviscid fluid. The simplest
-
constitutive equation involving viscosity is the incompressible Newtonian
fluid which is defined as T = 2 p D h n d where p is the shear viscosity.
Replacing T = - p l + 2 p D into (3.4.2) and assuming V p = V v = 0, we .
obtain the Navier-Stokes equation
For a symmetric T,
V . ( T - V )= ( V - T ) . v T:Vv, +
and since R is antisymmetric,
T:Vv=T:(D+R)=T:D.
RupIacing into (3.6.1) and using Gauss's theorem we obtain
(fl
r;itc'
1,.ipv2 dl. = I v , pf-v dl1 + $-v, t - v d s
of change of rate of work rate of work
- 1,. (T: D ) dl..
expansion of work
L~neticenergy of body forces of contact forccs and viscous dissipation
48 Conservation equations, change of frame, and vorticity
where
is the rate of work on dV, and within Vt. The energy equation is the first
law of thermodynamics for a continuum. If we define:
E = Jvt pr dv.
where e is internal energy per unit mass and q the heat flux vector, then
the Principle of Conservation of Energy states:
d
-(K + E) = Jv, p f v do + $,v, (T-v).n ds -
dt
Combining (3.6.2) and (3.6.3) we obtain,
i.e., the rate of generation of internal energy within the volume is due to
viscous dissipation-expansion within the volume and energy input-output
through the boundaries.
Problem 3.6.1
With the aid of the mechanical energy equation, show that any fluid of
a Newtonian incompressible fluid completely enclosed within a rigid
non-moving boundary and acted upon by a conservative body force must
approach zero velocity everywhere for long times.
Pvoblem 3.6.2
Show that T:D gives rise to p(V-v) (expansion work) and t : D (viscous
dissipation). Using the results of Section 3.7, prove that the viscous
dissipation is frame indifferent.
Problem 3.6.3
Verify that the viscous dissipation produced by a Newtonian fluid is
positive.
problem 3.6.4
fj, manipulation of T : D show that for the case of an inviscid fluid
('1' = - p l ) obeying the ideal gas law, the time integral of the rate of work,
(' ( ( / W tir)dt, is given by the popular equation j p dl.. Indicate any other
ncccssary assumptions.
3.7.1. Ohjectiritj
Denote generic scalars, vectors, and tensors, as 1; w, and S, respectively.
f . H . and S are ohjrc,titt. if they transform according to:
( i ) , l '=,f'
(ii) H'=Q.w
( i i i ) S' = Q.S.QT.
f
50 Conservution equations, change of' jrumr, untl uorticity
3.7.2. Velocity
The velocity of a particle X is the time derivative of the position vector
x. Thus, the velocity measured in F' is given by dx'ldt:
dxi/dr = dx,(r)/dt + d(Q(r).x)/dt.
Since d(QT.Q)/dt = dlldr = 0,
- + -
[dQ/dr] QT Q [dQT/dt] = 0,
and since dQT/dr = (dQ/dr)T, then
+
[dQ/dr]. QT Q .(dQ/dt)T = [dQ/dt] - + ([dQIdt].
QT QT)T
3.7.3. Accelevution
(J,ing (3.7.2) and after similar manipulations as in the previous section,
M'L' get
Problem 3.7.9
Assuming that if in a frame F we have
pa = p f + V - T
and in F' we have
p'a' = p'f' + V' T ' ,
show that
( 2 ~ ' - A+T x i . [ d A / d t - A . A I T + dcldt - C - A T )
= ( f ' - f . ~ ~ )( ~+ I ~ ) ) v ' . ( T ' - Q . T . Q ~
Note that the first term is independent of the material while the second
is dependent on the nature of the material. Present an argument to show
that T ' = Q - T . Q T Note that this reasoning assumes only that the linear
momentum equation has the same form in any frame and does not invoke
frame indifference for T (suggested by Serrin, 1977).
Pvoblem 3.7.10
Show that the equation in the moving frame F' is identical to that in
frame F if the 'new' body force, f', is:
f'= Q-f + p [ 2 A - V '+ (dAldt - A . A ) . x 1 + (dcldt - A - c ) ] .
'old force' 'extra body forces that arisedue to motion of frame'
Interpret the additional terms from a physical viewpoint. It is customary
to designate the additional terms as (Batchelor, 1967, p. 140)
2A.v' = 2w x v', Coriolis 'force'
- A - A . x l = - w x (w x x ' ) , centrifugal 'force'
( d A / d t ) . x l = ( d w l d t ) x x', Euler's acceleration.
Interpret w .
Problem 3.7.1 1
Consider the linear velocity field r , = G.x2,v2 = KG-u,. Work out, explicitly,
v', for a change of frame such that
[QI =
sin w t
[
cos tot - sin o t
cos o t 1
Obtain V'v' and verify that V'v' = Q - V v . Q T - A . Obtain D': D' and verify
that it is independent of (o. Compute Q':Qf and obtain its dependence
with tu. Investigate the possibility of selecting o ( t ) in such a way that
R':Q1= 0 .
Problem 3.7.12
Consider O(x. t ) = 0, cos[k(s, - c t ) ] .Sample tI in a trajectory x, = R cos o f ,
s, = R sin mt. Compute the time history of 0 observed following this
trajectory.
Vorticity distribution 53
Origin of vorticity
We have seen that the velocity gradient can be written as the sum of a
symmetric part (D) and an antisymmetric part (Q). Since in three-
dimensions an antisymmetric tensor can be expressed in terms of an axial
vector we obtain
VV= D - 1 x 012,
where o is the vorticity, o = V x v (see Truesdell, 1954, p. 3 and p. 58).
The vorticity is readily interpreted as twice the local angular velocity in
the fluid (see Section 2.8). Note also that
v.o=v.v xv=o.
Vortex line
A vortex line is a line everywhere tangential to the local vorticity (i.e.,
same relationship as streamlines and velocity field). Obviously, vortex
lines cannot cross for that would imply that a given fluid element has two
different rates of rotation. Similarly, physical arguments indicate that a
vortcx line cannot end somewhere in the fluid. However, they can end at
boundaries, form closed loops, and can also, as will be shown in Chapter 7,
continuously wander throughout space without ever intersecting themselves
(SCC Section 8.7).
A cortex tube is the surface formed by all the vortex lines passing through
a given closed curve C (which is assumed to be reducible). Consider the
construction of Figure 3.8.l(h) (at an arbitrary instant of time). Applying
Figure 3.8.1. ( a )Vortex line. vortex tube, and circulat~onaround a closed curve
C; ( h ) construction to prove constancy of circulation in a vortex tube.
1 1 1 ~divergence theorem we obtain:
S o - n ' da -
'top'. A ' S o e n " du
'bottom'. A"
+
S o . n du =
'lateral area' S V . o dl,= 0
hcnce r
J w . n d u = constant.
or alternatively,
Remarks
Note that the vorticity equation does not involve pressure. Note also that
o - V v = 0 in two cases: two-dimensional motion, where o is perpendicular
to v, and unidirectional motion, i.e., v = (u,, 0,O). Under these conditions
the vorticity diffuses as a passive scalar. Note that the transport coefficient
is the kinematical viscosity.
Problem 3.9.1
Verify that R = - l x 4 2 .
Problem 3.9.2
Show that if V-v = 0 and V x v = 0, then the viscous forces in a Newtonian
fluid, V2v, are zero.
Problem 3.9.3
There are cases where the intensification of vorticity due to velocity
gradients and the vorticity diffusion balance in such a way that there is
:I steady state distribution of vorticity. Consider
1 % - = x. r r = - ;%I-.
Vortex line stretching in inviscid jluid
material
line
Streurnjunction und potentiul junction 59
I = J" d'.,
c, = i7*/('y, v, = - (!*/?x,
in such a way that V-v = 0. The curve $(x, y, t = given) = constant, gives
thc instantaneous picture of the streamlines. Since v.V$ = 0, then if
it//:'?t=O, IC/ is constant following a fluid particle.'
The streamfunction and the vorticity are related by
v2* = - w .
In irrotational motions there are several simplifications. T o start with, $
satislies Laplace's equation
v2* = 0.
Also, since the flow is irrotational the velocity can be obtained from a
potential, 4 ( x , t), as
v =v4,
( V x V4 = w = 0). Also since V-v = 0, the potential function satisfies
v24= 0
<lnd the streamfunction and the potential function are related by
?*/?Y = ?l+b/?y,
?4/?y = - (?l+b/?.~),
60 Conserrution eyuutions, chunge c?j' jrume, unil rorticity
Example 3.12.1
Consider the complex potential = UI + Uii'iz which corresponds to
\it(:)
where 4 = UuZ cos I)/r and therefore v " UU'aJ'r4which indicates that
all the particles located at r. = const. have the same speed. In particular
for r = u the fluid particles slip with speed U. Consider the cylinder at
.Y = - Y_ and label a fluid particle in the as yet undisturbed flow by its
Example 3.12.2
Assume that we have an ~rtlstcrirl~.
description of a flow field
r I = r , ( . v , ,.Y,. t ) and
r, = r2(.vI..v2, 1).
How d o we calculate the instantaneous picture of the streamlines? In this
case
d$ = d r , + (i$i s 2 d) s 2 + ( ? $ l i t ) dr
(i$,i.\-,)
and
d$ = I., d . ~-- I.? +
~ d . ~ , ( i $i t d t .
Strrumfunction und potentiul junction
position of
point with
w / i ; n d e r a t + m
,'
position of
point with
cylinder
is approximately
a circle
62 Conseruution equutions, change of frume, und vorticity
S
xI.xz.1'
A list of problems (such as jets, boundary layer problems, etc.) for which
this approach works is given by Cantwell ( 1 98 1 ) . I 0
Bibliography
This chapter is a very succinct account of fluid mechanics suited only to
our rather specific needs. The interested reader should consult the original
works on which this exposition is based. The sections on conservation
equation are inspired in the classical article, 'The mathematical principles
of classical fluid mechanics', by Serrin (1959). The treatment of vorticity
follows Batchelor (1979). For a treatment of change of frame and
objectivity the reader should consult the book by Chadwick, Continuum
r,ic~c~hclriir..s
(1976). A few sections, such as Example 3.12.1. are based on
the article by Truesdell and Toupin (1960).
Notes
I We havcdroppcd the posltlon x a n d t ~ m rc. It should be understood that all qunntltles
,ire evaluated at x, r .
7 hhown in Problem 4.7.1, if ;@,';I = 0 the mlxlng is poor slnce fluid particles a r c
-
~ i ' t l i stretching.
c
oi' tlic material clement d X ( M = d X d ~ )p l, ~ ~ c catd X. a s
el, ( D In iD r ) . ( D :D l ' ' < I . (4.1.1)
I lie stretching efficiencies. el,, = c,,(X. N. r ) of the area element d A placed
,I[ N ( U = d.4 d . 4 ) . are defined in a similar \\ay. I f the motion is isochoric
\\c. define
cl,,= ( D In rl D r ) ( D : D ) ' ' < I , (4.1.2)
,irid if i t is not
e,, E ( D In L D r ) 12' ' ( D : D l '1 < I .
111 this \\ark \vc will use the first definition since \\e will assume throughout
[ l i ~ cthe flo\vs a r c isochoric. W e define the a.vj~r~lptotic efficiencies a s
lim c,,(X. N , t )
el,,, =
', I
c ~ ~ l ctheir
l corresponding asymptotic \ ; ~ l u e s are denoted as ( ( I , ) , and
(',, \,.
/
r c s p c c t i ~ c l y . Efficiencies c a n also be defined \\ith respect t o
i \ \ : ( V F ) ' ) '': thus we h a \ c
OT (I) In D I ) ( V F :( V v l ' ) ' ' = D : m m (Vv: ( V v ) ' ) ' '
f /,
denoted (e;), and (e,*), . Other efficiencies, for finite material lines and
surfaces will be used when convenient.
Frame indifference
The efficiencies e, and e, are frame indifferent, i.e., they are objective,
e, = e>,e, = eb, under the frame transformation x' = x,(t) + Q ( t ) . x . Note
that D, m, and n, are frame indifferent. Since D, m, and n transform as
Q . D.QT, Q - m , and Q - n , respectively, then D: mm, D : nn, and D: D are
also frame indifferent. Additionally, since V.v is also frame indifferent, it
follows that e, and e, are frame indifferent. Note, however, that e: and
e,* are not frame indifferent since Vv is not objective.
Physical meaning of e, and e,
For purely viscous fluids, ( D : D)'I2 is related to the viscous dissipation.
For example, for incompressible Newtonian fluids T : D = 2p(D: D). In this
case the efficiency can be thought of as the fraction of the energy dissipated
locally that is used to stretch fluid elements.
Relationships among e,, e, and eX and e,*
Since
Vv: ( V V )=~D: D - O : O
it follows that
e* = e,/(l + W:)'I2, i = j., q
where
W i = - (Q:O)/(D:D )
is the kinematical vorticity number.' Although neither the C$( i = i.,q )
nor W: are objective, the product eT(1 + W;)'I2 isobjective.
Pvohlem 4.1.3
Show that for a two-dimensional isochoric flow such that DmlDi = 0, eAm
can be written as (1/2119[1- W:]'12 (Fra~ljione,1986).
Problem 4.1.4
Consider a two-dimensional isochoric flow. Let O denote the angle
between a material filament and the maximum direction of stretching.
Show that the instantaneous value of the stretching efficiency is given by
(2'12/2)cos 20 (Franjione, 1987).
1
exp(i:i) 0 0
[F] = 0 exp(-i;t/2) 0
I 0 0 exp( - i:t/2)
cxp(2i:t ) 0
[C] =
The magnitude of D, ( D : D ) ' ', is equal to ?(3/2)' '. The long time value
'
of the efficiency, (D In i.,'Dt).'(D:D ) ' is equal to (213)' 2 0.816 (which '
is the i~pperbound for thrcc-dinicnsional flows). unless M is of the form
( 0 . M,. M,). Note that i. = i ( X . M. i ) in this case reduces to i. = i ( M , t).
This is [rile for all linear flows.
1:'suniple 4.2.2 Efficicvrcy of siriiple slrc~ar.i l o ~ c '
In this case
(I.Y, 111 = f Y : . t l . ~ ,t l r = 0. (I.\--3 111 = 0.
w'ith the samc initial condition as before,
+ XI,
: ;]
.u, = $s,t .'2 = X,, .u, = X,.
similarly, C is computed to be
[C] = ['
$1 1 + (7r)Z
+
h: exp(?r) h: exp(-$1) + 2h,h2
for K > 0, where
$ =2G(l~1)"~
The time evolution of (e,) for different values of K > 0 is given in Figure
E4.2.1 for a material plane with initial orientation N = ( I , 0). For K > 0 the
average efficiency rises rapidly to a constant value , = (2K)'l2/(1+ K )
(for $1 = 0(1),as expected) and it is relatively insensitive to the value of
K ((cJ,,), is equal to 0.666 for K = 0.5 and 0.406 for K = 0.1). This
observation is significant and indicates that the cffic.ienc.y tun he significuntly
inr-reused hy introducing u smull urnount of irrotutionulity into the ,flow.
For K = 0, corresponding to simple shear flow, (e,) decays as t ' . For
K < 0 the efficiency oscillates with a period proportional to I/GIKI"~.
Thus, flows with elongational character are efficient but the efficiency of
shear-like flows is poor unless special precautions are taken. One way to
increase the efficiency of rotational flows is by periodic reorientation of
filaments or areas to offset the tendency of the flow efficiency to decay as
I ' . Such a concept is explored in detail in Section 4.6 and forms the basis
for the high efficiencies encountered in Chapters 7 and 8.
E.vurnple 4.2.4 Efficiency in the helical annular mixer
Hcre we consider the stretching in the mixer of the Figure E4.2.2. The
flow is a combination of Couette and Poiseuille flow. For a Newtonian
fluid the azimuthal (0) and axial ( z )components of the velocity field are
~lncoupled.The velocity field is of the form
ur = 0, o0 = .f'(r), u, = Y ( r ) ,
:ind the motion is given by the solution of
drldt = 0, r dO/dt = f'(r), dzldt = y(r),
that is,
r= R, +
0 =O [f(r)/r]t, z = Z g(r)t, +
uhcre ( R , O , 2 )represents the initial position of the particle ( r , H, z ) . The
deformation tensor is computed using
c?r 1 c'r (?r
2 1 c'z ?z
c'R RiO (72
and
tiolls for largc aspect ratios, WIH > 0 (10)).In order to compute the length
,tl-~tchwe use
~ ( t =)
SI.( 0 )
(c:M M ) I / ~ ~ ~ X I ,
wt~crcthe integration (numerical, but based on an analytical expression
rot. thc velocity field) is carried out with respect to the initi~~l
configuration
of ~ h cline, L(0). For purposes of illustration, we consider two different
orlcntations: M = (0, I ) , a horizontal line spanning the cavity at .u2 = H/2,
Llnd M = ( 1, O), a vertical line at .u, = Wl2, also spanning the entire cavity
(XC Figure 7.5.2 for experimental results).
A typical result is shown in Figure E4.2.3 where [d In L(t)/dt]/(U/H),
~ h l c l ican be interpreted as the efficiency of the stretching, is shown as
a function of time for the two initial placements of the line (note that
( D : D)'I2 = O(2UIH)). As expected, for short times the stretching of the
vertical interface is more efficient since the line is placed perpendicular to
the streamlines. However, rather quickly, the efficiency becomes indepen-
dent of the initial orientation of the line and, on the average, the efficiency
decays as t l . This is a case of mixing with weak reorientation. More
complete computational results indicate the aspect ratio WIH, has a very
small effect on the (dimensionless) stretch, which is mostly confined to
changing the period of oscillation (of order WIU).
Example 4.2.6 Extensions of the cavity flow and the helical mixer flow
Several flows can be created by combining variations of cavity flows -
square, trapezoidal, with one or two walls moving in steudy motion with -
Poiseuille or axial drag flows. One such case is shown in Figure E4.2.4
where an upper plate slides diagonally at an angle 0 over a rectangular
~ a v i t y Provided
.~ that the cross-sectional flow and axial flows can be
decoupled (as in the case of a Newtonian fluid) the previous results of the
cavity flow can be extended to this situation. The mixing characteristics
of such a system are similar to those of the cavity flow; fluid particles are
;(/itnutha1 flow, since the axial flow is similar to the simple shear flow in
12stns of rnixing and involves n o reorientation.
p~.ohlem4.2.1
(';ilculate the time necessary to achieve the maxirnurn efficiency in a simple
,hcar flow as a function of M , , M,, and M,.
problem 4.2.2
['rove that the relative flow around any material particle in a linear flow
15 den tical to the flow itself.
Problem 4.2.3
('alculate the length stretch i in a Couette flow by means of the formulas
~ 1 1 cinn Chapter 2, by using both Cartesian and cylindrical co-ordinates.
P~.ohlem4.2.4
Study the stretching efficiency in vortex decay (Batchelor, 1967, p. 2 0 4 ) .
I.,, = (C/2nr)[I - exp( - r 2 / 4 0 t ) ] .
Show that the efficiency decays as t- '.
Problem 4.2.5
('onsider the velocity field
c,) = ( C r ) / ( 2 n r ? ) for r < r,
r, = Cl(2rrr) for r > r ,
Show that the striation thickness varies as
s .= rrr/[l + ( C ' t 2 / 4 n r 4 ) ] ' " .
Thus. s A r 3 for c 2 t 2 / 4 n r 4>> I and d In .s/rlt G t I . Sirni!arly, if r, represents
Ihc radial position of the layer with striation thickness s, r, moves outward
:is r, .- t 1 / 3(see Figure P4.2.1) (Ottino, 1 9 8 2 ) ) . The striation thickness
cllstribution is important in problems involving mixing with diffusion and
r~:iction(see Chapter 9).
Problem 4.2.6
Study the efficiency in the linear three-dirnensional flow
tlx/tlr = x Vv -
\\ith Vv constant. We can formally solve for the motion as
x =X . exp(rVv).
I'hc deformation gradient is
F = [exp(rVv)]' = exp(r[Vvlr)
76 Computution qj' stretching und efliciency
Figure P4.2.1. Stretching of a material line in a vortex after two and four turns
of the vortex core.
Flows with u speciul form of' Vv 77
-
dx d X - F T ,
but, if D(Vv)lDt f 0 this route does not offer any advantage over the
co~lventional approach, since in general, we cannot compute Vv(X, t )
without knowing the flow. However, for two special subcases this approach
is useful, and even though their practical utility is limited, it is convenient
to record them here.
(i) D(Vv)lDt = 0. In this case the problem can be solved in terms of the
cigenvectors of Vv.
( i i ) For slowly varying flows F(t) can be determined approximately by
perturbation e x p a n ~ i o n . ~
which generate the dual basis (gii and (gii, which we assume to be linearly
independent. (Note that the eigenvalues for both problems are the same.)
Taking the dot product of both sides of (4.3.1) with g',
(D(dx)/Dt).gi= dx. (Vv).gi
and since the gi are independent of time,
D(dx.gi)/Dt = g i -( V ~ ) ~ . d x
D(dx.gi)/Dt = /Iigi.dx
we can solve for dx.gi,
.
dx g' = (dX .gi) exp(/Iit).
Since a vector can be written in terms of the dual bases (giJ and jg,) as
dx = C (dx-gi)gi
we have
dx = C (dX-g')exp(/l',t)gi,
and in terms of the initial orientation, since M = dX/ldXI
dx = C ui e ~ p ( P ~ t ) ~ , l d X I ,
where ai = ( d ~ / l d ~ I ) Then,
. ~ ' . the length stretch is computed as
jb2= C 1 +
aidj exp[(pi pj)t]gij
where yij = gi.gj (summations on i and j carried from I to 3). The results
for area stretch are obtained in a similar manner except that the governing
equation is
D(da)lDt = da D(det F)lDt - d a . ( V ~ ) ~ .
Example 4.3.1 Linear two-dimensional flow
Consider again the flow of Problem 2.5.3. We can compute the stretching
without solving any differential equations. In this case
sp~vciulcase
When Vv is symmetric (as for example in all irrotational flows), {g;) = {g')
(i.c.. orthogonal), and the equations simplify considerably.
problem 4.3.1
Show that if D(Vv)/Dt = 0, then the length stretch is given by
).' = xx +
bibj exp[(Pi /3,)t]hij
where hij = gi.g', bi = (dX/ld~I)-g,,and where the gi are obtained from
gi. ( V V )=~pigi,
Assume D(det F)/Dt = 0.
Problem 4.3.2
show that
D In 2.
lim - = /Ima,
1-, Dt
whcre P,, is the largest positive real part of the eigenvalues of Vv. Note
that if the eigenvalues are of the form fit, D In E./Dt oscillates. Since the
eigenvalues of Vv are the same as those of ( V V ) ~note
, also that the limit
of D(ln q)/Dt is also given by p,,,.
Problem 4.3.3
Study the general conditions for which D(Vv)/Dt = 0 for steady flows.
,I
t:" :
D(dx,,)= dx,, .VV,,
+ dx,, - I . Vv,
Dl I =1
80 Computution of' stretching and ~ficiency
Exumple 4.3.2
Consider again the Hamel flow, solved earlier using the conventional
approach to compare the exact and perturbation solutions. In this case
we have
where
;I= 1K I / ( R ~ ) ~
and R 1s the initial radial co-ordinate of the particle X. Integrating v,, i.e.,
d(r2)/dt = 2K(I - d2)/r
with the initial condit~onr = R, and expanding the series, we obtain
(Rlr)' = 1 / ( 1 - ~ t *=) 1 + i:t* + (ct*)' + . . .
where t* = 27t and the small parameter c is
T ( 1 - d2)r.
E =
[v"ol=[;
f 2 ~ 4o
-:
;]
and
vv, = (~*)"VV,,.
Such an approach was used by Chella and Ottino (1985b). Figure P4.3.1
shows a comparison of the exact solution (obtained as indicated in
Problem 4.2.8) and a five-term perturbation solution for two different sets
of parameters.
Problem 4.3.4
Calculate the stretch of a filament with orientation N = ( N , , 0 ) placed at
cf, = 0 (this is the easiest case and requires no knowledge of the material
of Chapter 2).
Flows with u speciul form of V v 81
plnhlem 4.3.5
stlow that near the wall, the stretch is similar to that of a shear flow only
the case of diverging flow ( K > 1 ) . Show also that away from the walls
[he stretch is ulwuys weaker than an elongational flow.
problem 4.3.6
~ ' ~ ~ l c u lthe
a t evorticitj number for Hamel flow. Compare l o u r results with
thc l~mitingbehavior obtained in the example above. Does the vorticitj
number provide a good measure of the balance between shear and
elongation in the flow, in terms of stretching abilitb? Studj the l ~ m i cases
t
of II: + 0 and W: + 1 . Does W: = 1 mean that the flow is elongational?
Note also that W: does not dist~nguishbetween converging ( K < 0) and
dl\erglng flows ( K > 0).
Prohlrm 4.3.7
Calculate the viscous dissipation following a fluid particle in Hamel flow.
in an orthogonal system (.uk] for which the magnitudes of the natural basis
<
of (.uki, e, (see Appendix), are constant along the trajectory = <(X,t ) ,
c3
with 5(X,t = 0 ) = 0,given by 2' = X 1 , t2= X2 + t r ( X 1 ) , = X3 + tw(X1).
Thus, in an S C F the surfaces .u' = constant are material surfaces and they
slide over one another without net stretching (Huilgol, 1975, Chap. 2).
In the language of dynamical systems these surfaces are invariants of the
motion. See Figure 4.4.1.
For an SCF, H , and therefore F , can be calculated explicitly from the
spatial derivatives of v using a theorem due to Noll (1962). In this case
and with respect to the fixed orthogonal basis referred to above, the
physical components of H are given by:
Flows with a special , f i r m of F: constant stretch 83
a ithLI = (02/01), h = (e3/el), and where the eis are the magnitudes of the
i.c.. i z MI?, for y >> I , and except for the trivial case M, = 0 the stretch
gocs linearly with time, and since (D:D)'I2 is constant following a fluid
particle, the efficiency decays as t - I . This result is very important since
cxnmples of SCF are all rectilinear flow in pipes, Couette flow, helical
flow. torsional flow, flow between coaxial cones or concentric spheres, etc.
Example 4.4.1
Consider the shear flow c, c2 = 0, c3 = 0. F can be written as
= 7x2,
F=l+tH
where H has the representation
Problem 4.4.1
Show that if D(Vv),/Dt= 0, then F ( t ) is given by
F ( t ) = exp[tVv]
The flow is called ciscometric if (Vv12 = 0. Show that if (Vv13 = 0, j. z t 2
and therefore D(ln E.)/Dt z t ' for long times.
Problem 4.4.2
Consider a velocity field such that
r , = 0, C 2 = Ii.Y,, v , = cpx, + K X 2 ,
where K and cp are constants. Show that F can be written as
F ( t ) = exp[tH]
with H 3 = 0 (from Noll, 1962).
-
c J , , = e,, ) we need to determine the eigenvalues of Vv normalized by
( I ) : D ) 1 2 . However, it turns out to be more convenient to calculate the
eigenvalues of L Vv/(Vv: ( V V ) ~2, ) ' which gives e*, (which is not invariant)
and then use the relation between e*, and e, (Section 4.1) to convert the
results to an invariant form. The invariants of L are:
I,= tr L = V . V / ( V V : ( V V ) ~ )= ' O
11, = :[(tr L)' - tr L2] = - $ tr L2 = :(I - W:)/(l + w:)
111, = det L = (Vv: ( V V ) ~{ah(u ) ' ~ + b) + [y2(u + h) - j 2 a - h2h])
where
W: = (CJ' + h2 + j 2 ) / ( a 2+ ah + h2).
[See Olbricht, Rallison, and Leal (1982).] Thus, every linear three-
dimensional flow can be represented by a point on a plane with coordinates
t r ( L 2 )and det(L) and assigned a unique value of e*, o r e.. The axis tr(L2)
is related to the amount of vorticity in the flow whereas det(L) can be
thought of as related to the three-dimensionality of the flow. The bounds
Tor the region tr(L2), det(L) are:
- 1 ,< tr(L2),< 1 (4.5.1)
:I 11ci
det L
lmportunce of reorientution 87
probl em 4.5.2
Obtain the pathlines of periodic points corresponding to the vector field
dxldt = x.Vv.
L)o all the flows that have closed orbits have poor efficiency'? (Hint: Write
Ov in Jordan form and expand the exponential.)
Problem 4.5.3
Identify the region in the accessible domain such that v = const. o (such
flows are called Beltrami flows, see Section 8.7).
P~.ohlem4.5.4
('onstruct the iso-helicity curves corresponding to Figure 4.5.1
with
one possible way to increase mixing in such flows. One might start with
hyperbolic flow (see Section 2.5) and try to improve the stretching by
somehow closing the flow, i.e., by feeding the outflow into the inflow as
indicated qualitatively in Figure 4.7.1. By continuity, nearby orbits have
to close smoothly and form closed loops. The expectation might be that
since the stretching in the neighborhood of the hyperbolic flow is
exponential, then every time that a line segment passes by the hyperbolic
region the stretches are compounded in such a way that the stretching is
exponential.
However, this is illusory. As long as the flow is steady, isochoric, and
two-dimensional, we can construct a streamfunction $(x, y) such that
v . V $ = 0. This implies that a material filament is trapped between curves
of constant $. Another way of seeing this is to note that the topology of
the streamlines in two-dimensional area preserving flows is composed of
basically two building blocks: hyperbolic points (stagnation points) and
elliptic points. It is obvious that if the flow is bounded and steady the
streamlines join smoothly, and that a n initially designated material
filament can spiral between the two curves, $ , and $, very much in
-
the same way as in a Couette flow but that the length increase will grow
-
linearly for long times (see Problem 4.7.1).11 Indeed, as we shall see in
Chapter 6, the long time value of the averaged stretching function
2 = lim
r - j
{: ' D In j.(tl) dt,]
Dt'
is zero for these flows (termed 'integrable', a term defined in Chapter 6.
Loosely speaking it means that it is possible to find the ~ t r e a m f u n c t i o n ' ~ ) .
The same argument applies to the combination of several hyperbolic
points as shown in Figure 4.7.1. Apparently we face the conclusion that
all two-dimensional flows are poor mixing flows. However, this need not
be so. and as we shall see in Chapter 6 and demonstrate by means of
examples in Chapter 7, if the flows become unsteady (e.g., time-periodic)
the mixing can be excellent. In three dimensions the situation is more
complicated. If the flow is three-dimensional (Chapter 8). combinations
of inflows and outflows belonging to hyperbolic points can produce
excellent mixing even if the flow is steady.
P~*ohlem4.7.1
Consider a two-dimension:tl steady bounded flow containing a region of
closed strc:tmlincs $(.Y.J . ) . Denote by T ( $ ) the time i t takes for a fluid
particle belonging to the stre:tmlinc 4'/ to return to its original position.
Possible ways to improw mixing
outflow
/
0 - - . ,/-\ -
\
\
I \
I I
/
\
,
point inflow
\ +elliptic point
--- ------.
outflow of A joins inflow of B
. I-- \
%?( I\
_-
'--- ----
_--- i r
j j
-_-A'
94 Computution qf stretching und <ificiencl;
Bibliography
The foundations for the material discussed in this chapter can be found
in continuum mechanics works. The most comprehensive treatment is
given in Truesdell and Toupin (1960). A more accessible presentation can
be found in Truesdell (1977). The original material on S C F is from No11
( 1 962). A particularly clear discussion can be found in Coleman, Markovitz,
and Noll (1966). Major portions of this chapter can be found in Chella
and Ottino (1985b).The example of the cavity flow can be found in Chella
and Ottino (1985a).Section 4.5 is based on Khakhar and Ottino (1986a).
Notes
1 See Sections 56 -61 in Truesdell (1954).
2 This flow allows the computation of many quantities of interest. T h e flow
is a 'steady curvilineal flow' (see Section 4 . 4 ) a n d consequently all the mixing functions
can be calculated analytically.
3 For example. mixing in many polymer processing applications. See for example,
Chella and Ottino (1985a).
4 Such a flow forms the basis of a practical mixing device known a s a single screw
extruder. Fairly detailed analyses of the mixing are possible but numerical computations
;ire necessary (see Chella and Ottino. 19X5a).
5 These sections are based o n R. Chella a n d J. M . Ottino (1985b). The notation is
slightly different. Also we corrected several unfortunate typographical errors which
appear in the paper.
6 This class of flows is relevant t o loil Reynolds number floils, such a s some of those
encountered in polymer processes, geophysics. etc.
7 The results of this section necessit;~teuse of the concept of rrltrrir,t, n ~ o r i c . However,
~~.
since the concept is not used anywhere else in this work. we d o not discuss i t here
even though we lose s o m e rigor in not doing so (see Noll 1962).
X A p;~rticularlyclear and succinct discussion is given in Section 13 of C o l c m ; ~ n .
Markovitz. and Noll ( 1966).
9 For additional dctails the reader should consult Chella a n d Ottino (19X5b).
Notes 95
10 This involves changing simultaneously the stretching time t and the location r if w is
regarded as constant in the vortex flow or the time t and the shear rate G in theshear flow.
I I There is substantially more to the analogy with the Couette flow. The Couette flow
might be regarded as the prototype of an integrable flow (Chapter 6).
12 This does not imply that the streamfunction cannot be complicated. For example,
Berker (1963), mentions three special cases studied by Oseen (1930) (cf.
pp. 3 and 4). The streamfunctions studied can be interpreted as the superposition
of two mutually orthogonal logarithmic spirals giving rise to Ivl = constant. In this
case the portrait of the streamlines is complex enough to deserve the name of 'pseudo-
turbulent' indicated by Berker.
Chaos in dynamical systems
5.1. Introduction
In Chapter 3 we considered classical ways of visualizing flows using
'classical' fluid mechanical tools such as streamlines, pathlines, and
streaklines, which could be regarded as global in the sense that they give
some information about the entire flow. The same can be said about
instantaneous contour plots of viscous dissipation, vorticity, helicity, and
various other quantities. On the other hand, in Chapter 4, we considered
descriptors that have to do with stretching at local scales, e.g., specific
length stretch, area stretch, and mixing efficiency. However, as we have seen,
only in a few cases can the velocity field be integrated exactly to compute
the stretching, and in those cases where i t is indeed possible it does appear
that the rate of stretching is mild. As we shall see, for most velocity fields
with good mixing abilities, i t is often impossible to obtain the flow.
Consequently, in many cases of interest we are unable to calculate the
local quantities described in Chapter 4.
In this chapter we reconsider the central question of our analysis,
namely: What are the conditions under which a deterministic flow
K = @,(X) is able to stretch as efficiently as possible a material surface
:hroughout the space occupied by the fluid?' Even though it is not possible
.o give a complete answer to this question, in this chapter we will focus
)n some relevant aspects of the general problem, to build intuition when
Ne examine the prototypical model flows in Chapter 7 and 8 from a
lynamical systems viewpoint. Therefore, in order to provide some back-
;round, we present here some of the main elements of the theory of
lynamical systems.
D.ynumicu1 systems 97
-
autonomous systems or time-periodic systems.
I f we denote d V = d X , dX, . . . dXN as the volume of the initial
conditions, and by dc dx, d.u, . . . dx, as the volume of the initial
conditions at time t , we have, as before,
where
where V(0) denotes the volume of the initial conditions, {X}, and V(t)
denotes the volume of the initial conditions at time t , i.e., {x}= @,{X}.4
If V - f < 0 (see Equation (5.2.1)) the system is called dissipative.
Dissipative systems contract volume in phase space. In Chapter 6 we will
consider Hamiltonian systems, which conserve volume in phase space.
These systems have the structure
where H is called the Hamiltonian and is a function of the pks and q,s,
and in some cases an explicit function of time. In particular, problems
arising from Newton's law of motion can be cast into this structure.
Under fairly non-restrictive conditions the solution of the dynamical
system (5.2.1) (see Section 2.4) gives rise to a flow x = @,(X), i.e., the
initial condition X is found at x at time t . Given any point x, belonging
to the phase space iWn x iW at some time arbitrarily designated as zero, the
orbit or trajectory based at x, is given by @,(x,) for all times t . Thus,
@,,,(x,) denotes the orbit of x, for t > 0 (future times) and @,,,(x,) for
all t <O (past times). In general, we refer to the system of ordinary
differential equations (5.2.1) and its associated flow as a continuous
dynumical system.
In some instances it is not necessary (or possible) to follow the trajectory
of every initial condition continuously in time through the phase space
and it is more convenient to record their positions at specified times.
Consider for example that the position of the initial condition X = x, is
Fixetl points und periodic points 99
X I = @,.(x,)
x2 = @.l.(@,
(x,))
...
x,, = @,.[@.[.[. . . (x,)];
or in more compact notation,
xn = @y.(x,) (5.2.2)
where @;(. ) denotes the composition of n mappings a,. Alternative ways
of expressing the same concept are
X,, + @T(x,,) (5.2.3)
a~id
Xn + I =@T(x~). (5.2.4)
Obviously, such a mapping contains some information about the original
f l o w h a d in some instances it is convenient to deal with such a map
rather than with the entire flow. In general we refer to the mappings
(5.2.2)-(5.2.4) as a discrete dynamical system.
P = @,(P)
1.c.. the particle located at the position P with orbit x = @,(P) returns to
its initial position after a time T (@,(P) # P for any t < T ) , as for example
111 the orbits of circular Couette flow. For a discrete dynamical system
Xn + 1 = f(x,,),
the definitions are similar (see Smale, 1967). A point P is a ji.ued point of'
tllc mupping f( . ) if
P = f"(P)
lor ull n. Thus, a ,fi.ued point of' 11 ,flow and its corresponding mapping
are in general not the same. Generally, a fixed point of a mapping
corresponds to a periodic point of the flow (see Figurc 5.3.1 ).
100 Chaos in dynumicul systems
5.4.1. Definitions
1-ixed and periodic points can be stable or ~ n s t a b l e Consider
.~ first the
dcfinitinns of stability (Guckenheimer and Holmes, 1983, p. 4; Hirsch and
Smoie, 1974, p. 185). These definitions arc for local stability since they
~;,cuson the behavior near the points.' We discuss the definitions in terms
of fixed points of flows; the definitions for mappings are similar (see
Arnold, 1980, p. 1 15).
Iiupunov stable
The point P is a stable equilibrium of the flow if for all neighborhoods U
of P there exists a neighborhood U , of P E U such that x = @,(X)belongs
to U for all times if X belongs to U , (see Figure 5.4.l(a)).This behavior
is typified by centers; see Section 2.4.
See Figure 5.4.1 ( b ) . This behavior is typified by sinks; see Section 2.4.
The effects of the different linear mappings S on the square (0, 1 ) x (0, 1)
are shown in Figure 5.4.2 (this analysis follows Percival and Richards,
1982, Chap. 2). The parabolic case corresponds to simple shear. Note
that if the initial condition x, belongs to one of the eigenspaces, successive
mappings are given by x, = Lyx, (i = 1 , 2); therefore all the x,s remain in
the eigenspace. It follows that if /tr[A]/ > 2 one of the eigenvalues has
modulus greater than one, and x, becomes unbounded. The mapping is
said to be unstable. In general we say that the mapping is hyperbolic if
none of its eigenvalues belongs to the unit circle; therefore hyperbolic
mappings are unstable. O n the other hand if /tr[A]/ < 2 the eigenvalues
are complex conjugates and lie on the unit circle (Arnold, 1980, p. 116).
In this case the system is called cllipticx and the system is .stirhle. The
(degeneratc) case ( i ,= I ) is called ptirribolic..
5.4.3. Families of' periodic points
('onsider a fixed point P of order k. Successive iterations produce the
kimily
P-P, +P,+. ..+P, (5.4.1 )
until reaching P with the kth iterate. The stability around each point is
given by a linearized mapping as indicated earlier. If we want to relate
the zeroth and nth iterates in the family (5.4.1) we compose the linearized
mappings around each point to obtain
X, = J - X ,
P~noblem5.4.1
Prove that the eigenvalues, and consequently the stability, of all the
members of the family (5.4.1) are the same (Lichtenberg and Lieberman,
1983, p. 186).
Figure 5.5.1. Poincare section iterates: ( a )focuses o n orbits near P and shows
the intersections ~ i t ah plane X orthogonal to the orbit passing by P. A point
close to P. x , , is mapped to x 2 = @ ( x , ) . and then to x, = @ ( @ ( x , l l : ( h )
representation of n period-? point in Poincare section.
Invariant .sub.spuce.s: stable and unstable rnanifi,lds 105
In this work Poincare sections will be used in two main ways: globally,
when the system is periodic in either space o r time, or locally, near periodic
or homoclinic orbits. Both ideas will be discussed and used in Chapters 6
through 8. In the case of time-periodic systems the technique amounts
to taking stroboscopic pictures of initial conditions placed on C ( x ) at
intervals T, 2T, 3T, . . ., etc. Thus Poincare surfaces are a convenient way
of displaying in a single plot the churacter qf the solutions belonging to
t i l l possible iniriul conditions.1° As we shall see some of the most important
properties of dynamical systems are better understood with this viewpoint.
In particular, Poincare sections are of utmost importance in the case of
periodic Hamiltonian systems.
I t is clear that E", Eu. and E', are i111.rrr.irrntsets; initial conditions remain
[rapped in the set. Examples are given in Figure 5.6.1.
Similarly. for the linearization of the mapping x + G ( x ) (or x , , + , = G(x,,))
\r it11 x E R*', 5 -+ DG(P)T, around n fixed point P , the strrhlc. E:', ~ir~.sttrhlc~.
I:". and C O I I ~ O ~1;'.
, subspaces are defined in terms of the Jncobian matrix
I ) G ( P ) as:'-'
106 Chaos in dynamical sy.stems
These ideas can be extended to the non-linear system. For example, for
the system dx,rlr = f ( x ) with a flow x = @,(X), the stable, W S ( P ) , and
unstable, Wu(P), manifolds associated with the hyperbolic point P are
defined by:
W'(P) = (all X E R v such that @,(X) -t P as t -t x I
W u ( P )= :all X E R,' such that @,(X) -t P as t -t - x ) .
Similarly, for mappings we have
-
W'(P) = [all X E R' such that G " ( x ) -t P as 11 x 1
-+
of the blob when P returns to its original position whereas the hyperbolic
cycle contracts the blob in one direction and stretches i t in another. I n
this case, the unstable manifold Wu(P)corresponds to the region mapped
out by the blob when time runs forward (see Figure 5.6.3). A seemingly
minor complication in the blob experiment corresponds to the case when
the blob S comes back folded upon its initial position, as shown in Figure
5.6.4. The complication, however, is profound. This construction, which
will be studied in Section 5.8.3, is prototypical of chaotic flows.
Note that uniqueness of solutions places strong restrictions on the
behavior of stable and unstable manifolds. For example, two unstable
(or stable) manifolds belonging to two different periodic (or fixed) points
cannot intersect; also WU(P)cannot intersect with itself, similarly, Ws(P)
cannot intersect with itself. However, intersections of stable and unstable
manifolds belonging to the same or different points are permitted and
are, in fact, responsible for much of the complex behavior of flows.
Figure 5.6.4. A complication on the case of Figure 5.6.3: the blob comes back
folded (1, < r , < T ) .
110 Chuos in dynumicul systems
Problem 5.7.1
Study the structural stability of the one-dimensional equation dxldt =
C x n , where C is a constant, for n = 1 , 2, and 3. Consider a perturbation
of the form p g ( x ) . Take y ( x ) = 1 .
topologically
equivalent
topologically
non-equivalent
structurally
stable 6 + % - 6
structurally
unstable + % -6
structurally
unstable
Signutures of chuos 111
heteroclinic
connection
of the same cycle, the connection is called homoclinic (see Figure 5.8.1).
Similar definitions apply to maps. A point y is called heteroclinic (Smale,
1967; Guckenheimer and Holmes, 1983, p. 22, etc.) if it belongs simultane-
ously to both the stable and unstable manifolds of two different fixed (or
periodic) points p and q. Thus, y is heteroclinic if
Y E W U ( p ) nW"q).
The point is called homoclinic if p = q. If the point of intersection of the
manifolds at y is transversal, i.e., the manifolds intersect non-tangentially,
then y is called a transverse homoclinic point (similarly for heteroclinic).
Generally, when no confusion arises, a transverse homoclinic point is
called a homoclinic point (similarly for heteroclinic). Figure 5.8.2 shows
homoclinic and heteroclinic points for a map; only parts of the stable and
unstable manifolds are shown. Such maps can be originated by the
transverse intersection of one or two hyperbolic cycles, as shown in Figure
5.8.1.
By definition, homoclinic and heteroclinic points belong simultaneously
to two invariant sets and therefore cannot escape from them.'' One
intersection implies infinitely many. In the case of an orientation preserving
flow - as are all the systems originated from differential equations the -
area A is mapped to A' (if the area A is mapped to B the map does not
transverse
\ homoclinic point
preserve orientation). Note that the intersection of invariant manifolds, as
shown in Figure 5.8.2, is a projection onto the Poincare section: the point
belongs to the intersection at time T, 2T, etc. However, what happens at
intermediate times? Again a fluid flow visualization is useful. Figure 5.8.3
(which is a continuation of Figure 5.6.3) shows, schematically, intermediate
simultaneously to
hyperbolic
periodic point
blob stretched by WU
(lower part not shown)
blob moves;
position coincides
at t and r + T
V)
W
2
.*2
."E"
4-
view in
Poincar6
section
114 Chaos in dynumicul systems
for some t' > 0. Non-wut~drringpoints are defined to be those points which
are not wandering. For example, the points belonging to a closed orbit
are non-wandering since portions of the neighborhood intersect the initial
location. Non-wandering points do not imply periodic behavior. In fact,
it is possible to show that homoclinic points are non-wandering (and they
are clearly not periodic). Moreover, in every neighborhood of a homoclinic
point there is a periodic point (see Smale, 1967).
5.8.2. Sensitivity t o initiul conditions und Liupunov exponents
One of the manifestations of chaos most readily related to fluid mixing
is the sensitivity to initial conditions.
The flow @,(X) is said to be sensitive to initial conditions on a domain
S if for all X # X,,, with X belonging to an >:-ballaround XI, ( i t . , the set
of Xs such that IX - Xl,I < E , with E small), there exists a time, t < such
that @,(X)lies outside the >:-ballfor all X,, contained in S (see Figure 5.8.5).
A similar definition applies to maps.'"
A signature of chaotic systems is that they are characterized by a rapid
divergence of initial conditions. Usually the divergence of initial conditions
is quantified by means of numbers called Liapunov exponents which are
related to the stretching o f a filament of initial conditions. l f l d ~ represents
l
the length of a vector of initial conditions dX around X with orientation
M = dX/dXI, and its length at time t is Idxi, the Liapunov exponent ai
corresponding to a given orientation M,, is defined as (Lichtenberg and
Lieberman, 1983, p. 2621~':
since /. = I(D@,(X).M,~.
An N-dimensional flow has at the most N, in general different, Liapunov
exponents ( i t . , there are N vectors Mi, linearly independent, each of which
givcs. possibly. a different value). This is so since the eigenvalue problem
,/@,(X)
.*'
has at the most N different eigenvectors. In the cases of interest here,
V - v = 0, det(D<D,(X))
= I and
Y
X M i = 0.
< - I
i.e., the Liapunov exponent is the long time average of the specific rate
of stretching, D In ;./Dr. Similarly, the average stretching efficiency
where
e(X, M,, t ) = D:mimi/(D:D)'I2 = (D In i . / ~ t ) / ( DD)'"
:
can be interpreted as a normalized Liapunov exponent (with respect to
( D :D ) ' '). The relationship between the (maximum) Liapunov exponent
and the efficiency is not direct, unless ( D : D ) is constant over the pathlines.
In most cases of interest ( D : D)'" is a function of both X and t.
- -
D A
cut and
stretch fuse
Figure 5.8 7 continurd
fold
D' A'
0
stretch
n
C' B'
v t c h )(
backward, f - '
(b)
and H , ) are stretched by the same amount in the vertical direction by a
factor i., and contracted in the horizontal direction by a factor i.,. Thus,
the Jacobian everywhere in H,, is
since it involves a 180" rotation. The values of i., and i., are selected in
such a way that 0 < i., < and i., > 2 so that the vertical striations are
disjoint. Thus, both f' restricted to f ( S ) n S and f ' - I restricted to
f ' - ' ( S ) n S are linear; the non-linearity comes in due to the folding. I t is
clear that n-applications of the forward mapping create 2" vertical
striations occupying an area 2";"; (Figure 5.8.8 shows f 2 ( S ) ) . Since
0 < i., < the total area goes to zero as n tends to infinity. Something
similar occurs for the backward striations (area 2"E.;). Note that if the
map is area preserving i.,i., = 1 .
Figure 5.8.8 shows the 'loss' of material that does not get mixed by the
transformation (i.e., it goes somewhere else in the flow region where it
may or may not get mixed by another horseshoe map) and, in fact we
might think of the horseshoe map as a baker's transformation with loss
(Rising and Ottino, 1985). The set remaining in the original square,
denoted I,
forward backward
(a)
124 Chuos in dynumical systems
Problem 5.8.1
Verify that the Jacobian corresponding to H, has the form indicated in
the text.
Problem 5.8.2
Verify the inverses of Figure 5.8.7(b), and Figure 5.8.9
Bibliography
The origin of most of the material presented in this chapter is mathematical
but much of the motivation can be found in physics and astronomy. In
a very general sense there are two kinds of chaotic systems: dissipative
.sy.stt.ms - with one-dimensional linear mappings being the most pervasive
example - and Hu~niltoniunsyste~ns,which will be treated in Chapter 6.
Historically, the earliest studies focused on Hamiltonian systems and were
concerned with celestial mechanics; e.g., stability of the solar system. The
earliest work can be traced back to Poincare (1892, 1893, 1899); the
Notes
1 Thus, the goal is not to he able to integrate the flow exactly but rather to find the general
conditions for efficient stretching in tltly flow.
2 Recall that ifv E IW" the non-autonomous systems can beconverted into a n autonomous
system by defining .x,, + ,= r. See also Section 2.4.
3 Recall that, with the exception of vorticity, most of our results carry from IWZ to IW".
4 The relationship between the Reynolds theorcm and the Liouville theorcm is mentioned
by Synge (1960. p. 174). The d a t e of Liouville theorem is 1838; the date of Reynolds
( o r transport) theorem is 1903.
5 For example if the flow is smooth, say !i-times continuously differentiable with bounded
Jacohian, then the mapping is smooth and has a smooth inverse.
6 We restrict the discussion to autonomous systems.
7 Stability can be defined also in a global sense. but it is usually much harder to prove (see
Hirsch a n d Smale, 1974. Section 9.3.).
X Sometimes called cross-section.
9 In some cases, such a s time-periodic systems, we can select surfaces in such a way that
the timc between successive intersections is the same
10 It sould be mentioned that there is a converse construction whlch is also of importance.
A s~r\pc,t~.\iot~
allows the reconstruction of the flow starting with the m a p (Smale. 1967).
1 I A set S is called a n i ~ ~ r w i r rset
t ~ rof the flow x = @,(XIo n a mnn~foldM ( S c M ) i f @ , ( X ) € S
r time t (for our purposes we can regard a manifold as a smooth surface).
for all X ~ S , f o all
Thus, if a point belongs to a n invariant set, then when acted o n by the flow, it remains
in the set. The definition for mappings is analogous.
12 F o r additional details see Section 2.4.
13 See Section 5.4.2 for the two-dimensional case.
14 According to the stable manifold theorem near the fixed point the stable and unstable
manifolds a r e tangents to the linear eigenspaces E ' a n d Eu(for references see Smale. 1967,
also Guckenheimer and Holmes. 1983, p. 18).This theorem was known to Poincare and
Birkhoff.
15 F o r conditions o n g(xlsee Andronov, Vitt. and Khaiken (1966).T h e concept ofstructural
stability is largely due t o Andronov. A similar question is addressed in Section 6.10.
16 That is, the surface has two 'sides'.
17 Note that in many cases this theorem is of rather limited use t o us since we are generally
interested in area ( o r volume) preserving perturbations a n d not just any perturbation.
18 The implications of this invariance are extremely important when the flow is area
preserving as in the case of Hamiltonian systems (see Chapter 6).
19 Note that sensitivity to initial conditions is compatible with continuity with respect to
initial conditions
20 A similar definition. with timc t replaced by 11, applies to mappings (Lichtenberg
and Lieherman, 1983, p. 267).
21 In general the numerical calculation oft111 Liapunov exponents is complicated. Straight-
forward application of the definition produces the value of the maximum Lii~punov
exponent for almost all initial Mis (Lichtenberg and Lieberman. 1983, p. 280). See
also Greene a n d Kim. 1987.
22 The term mixing has a precise meaning in ergodic theory. T h e term has been used in
a fluid mechanical sense throughout this work (stretching plus difTusion). This is
the only section in which it is used, unavoidably, in a mathematical sense. T h e terms
'recurrence'. 'wandering', 'mixing' (measure-theoretic). and 'ergodic', are obviously
related but their relationship will not he discussed in detail here. Forexample,ergodicity
implies recurrence, but the converse is not true. Both recurrence and ergodicity
d o not imply mixing.
Notrs 129
23 Tlic relationship bctwccn Iluid mixing ant1 a hakcr's transformaton was pointctl out
in the 1050s hy Spencer and Wilcy (I951 ) hut its full mathematical implications were
apparently not rcali/cd.
23 Static mixers provide. po\sihly. the closest experiniental and practical approximation
tt) a hahcr's tranaformati~,nin the context of mixing (see Section 8.1 ). T h c c devices
arc used in the polymer processing i n d u t r y tt1 mix viscous liquids ( w e Middleman.
1977). The most popular is the Kcnics" mixer which consists of a tuhc with internal
helical s~~bdivisions. a twisted plane. of alternating right hand and left hand pitches.
Fach s~~bdivisit)n is callcd a n clement. Ideally. after each element the s t r e a m fed
into the mixer arc suhd~videdinto two a n d after 11 elements the striation thickness
d c c r c a s c a s 3 ". Expcrime~itscarried out hy the Kcnics corp~)rationshow indeed
a layered structure. I'hotographs a r e reproduced in Middlcnian's hook (op. (rr).
Some of these matters arc discussed in the context of the 'partitioned-pipe m ~ x c r '
described in Chapter 8.
25 Many other types of horseshoes a r c possible; the o n e depicted in Figure 5.8.7(h)
i.; the simplcst and most popular one.
1h In dissipative systems a system is callcd chat)tic if it possesses a .slrr1116/crrlrrrlc./or ( a n
attractor is called strange if it is a n attractor. and it is sensitive to initial conditions.
i.e.. ~t pt)sscsscs at Icast one positive Liapunov exponent). Attractors arc impossible
in Hamiltonian systems. ]..or dcfin~tionsscc Ci~lckcnhcirncrand Holmes. 1983. For
applications of strange attractors concepts in fluid mechanics. scc Landford ( 1083)
and Gl~ckcnheimcr( 1986).
17 Otlicr possihlc definitions of (temporal ) c h a o s arc to ohscrvea signal .x(r)asa function
of time a n d t o compute the power spectrum of.\-(r1. An indication ol'chaos IS a continuous
.;pcctrum. Another possibility is t o compute the correlation function. r.(r). of thc
signal \ - ( I ). i.c..
-
\rlicrc \ . ' ( I ) is the difference hctucen . \ ( I ) a n d the time average value of .x(r ). If c ( r )+ 0
as 7
n~~rncrically
/ the system IS considered chaotic. S~)mctimcs. the visual appearance of
computed Poincarb scctit)n is taken as cvidcncc of c h a m . If possible.
\\c prefer t o designate as chaotic any system uhicli satisfies any of thc conditions
( 1 ) ( 3 ) abobc. sincc they are. in principle mathematically based and anienahlc t o
~I-ooI'.If tlic\e cannot he proved we then might consider any of these alternative
dclinitions. In any case we will clcnrly specify which criterion is being used.
1 S S ~ n c c1111stechnique \*ill be uscd mostly in the context of Hamiltonian systems tlic
di\cush~oni\ I-escr\cd until Section 0.10.
7') S ~ ~ i i e t i ~ ithe
i c s\vord 'ilidc\' i \ ~ l s e din this co~itcxt.The index of It'" is the numbel-
of tlimcnsions of It"'.
.30 The situ;~tionsslioun Ilere ;il-e commtjn sincc hyperbolic I ~ n c a rvolume prcsel-bing
IIo\*.; a r c dense ant1 open a m o n g all i o l u m c prcscl-\ing flous (Smale. 1007). H ~ v . c \ c r .
it is a n unsolved prohlcm whether o r not this is true when the right hand side of the
c \ o l u t ~ o ncclu;~tion(I\ (11 \I\. I ) is c o n t r a i n e d to satisl'x the Navicr S t o h c ecluationa
-
6.1. Introduction
It seems at first contradictory to focus on Hamiltonian systems in the
context of mixing of viscous fluids. However, the connection is purely
kinematic and involves no approximations. As seen in Chapter 3, the
equations describing the trajectory of a fluid particle in an isochoric
two-dimensional velocity field are
dsl/dr = (7$/i?u2, dr2/dt = - i?$/?.u,,
where $ is the streamfunction. Such a system of equations, regardless of
the form of $ is a Hamiltonian system (Aref, 1984) and therefore it is
profitable to study fluid mixing from such a viewpoint to exploit the
substantial body of theory focusing on these systems. The system is said
to have one degree of freedom if the flow is steady, $ = $(.u,, x,), and
two if the flow is unsteady, $ = $ ( r ls,,
, t). If $ is time-periodic we say
that the system has 'one and a half' degrees of freedom. As a confirmation
of thc ideas of Section 4.7 we will prove that two-dimensional steady flows
arc poor mixing flows and that time-periodic flows are, most likely,
effective mixing flows at least in some region of space.
Hamilton's ryuutior~s 131
that a s long a s the system is Hamiltonian, the general features will appear
c \ c n in the simplest representatives of the class.
Example 6.3.1
Consider a onc-degree of freedom system, i.c.. t 1 = H ( p , 4 ) . In this casc
wc need to find only o n e E;. In this case the necessary function is H itself
since [ H , HI = 0. T h e r e f o r e all steady two-dimensional flow fields a r e
integrable. I n this case material is t r a p p e d by streamlines a n d t h e system
is a p o o r mixing flow (see P r o b l e m 4.7. I ) .
In the case of a two-dcgrccs of freedom system. H = H ( p , . pz, q , , 4 2 )
o r t 1 = H ( p . 11. 1 ) . wc need t o find t w o functions. F, a n d I.'? T h e first o n e
is casy t o find since [ [ I . H ] = 0. a n d we c a n take I.', = t l . T h e n we have
to find I., # I I H # t 1 + h. where t i and h a r c constants. such that [I.'?. 111 = 0.
Howevcr. I.', need not exist! As Helleman ( 1 9 7 7 ) points o u t , most
tlamiltoninn systems with two-degrees of freedom arc rlon-intc$qrrrhlr,i.e.,
i t is in general not possible to find F, and F2 such that dF,/dt = O and
rlk'2/dt = 0 along a trajectory.
admits only three kinds of fixed points: saddle or hyperbolic points, centers
or elliptic points, and the degenerate case of parabolic points (see Example
2.5.2, two-dimensional area preserving case). In the most general case the
unstable and stable manifolds of hyperbolic points join smoothly by a
curve of constant H. The essence of any integrable system is embodied
in Figure 6.4.1. To see why this is so, note that all integrable systems with
N-degrees of freedom can be transformed into N-uncoupled one-degree
of freedom systems, i.e., the system
Ap hyperbolic point
separatrix
I elliptic point
Phtrsr sptrce of' Hrrmiltor~itrr~
.s~~.sien~.s 135
rlH
=V-VH +i H
-
tlt it
irhich is interpreted as the rate of change of H (energy in the usual case)
I'ollowing the flow. F o r Hamiltonian systems v.VH 0, which implies
(hat v belongs t o surfaces in phase space of constant H . Furthermore, if
[I is not a n r.\-plic,it function of time H remains constant for a given inilia1
condition X.
Another way of seeing this. a n d a t the same time adding some physical
~ n e a n i n gt o the Poisson brackets, is to note that (6.5.1) c a n be written a s
,
Such a section defines a return map in the plane C, x,, = G(x,). Note
that all trajectories originating from C return to C after the period T.
Note also that area is conserved in the Poincare section.
In the case of an integrable system, the cross-section of the torus can
be regarded as an integrable twist mapping (Equation (6.3.4a,b)).An initial
trajectory starting on C wraps around the torus as indicated in Figure
6.6.1(b). The trajectory may or may not join with itself depending on the
value of toi(r). If the trajectory comes back to its initial position p in the
Poincare section after going around the torus nz-times, then p is a periodic
point of order m.
As seen in Chapter 5 the stability of fixed and periodic points
corresponds to the solution of the eigenvalue problem
where DG(p) is the Jacobian matrix of the mapping evaluated at the point
Phase space in periodic Hamiltonian ,flows 137
..
Hyperbolic ~ j ~ , ~ > l > ~ j~. .2, ~/ ., ~ = l
Elliptic 1.1 = I (i = I , 2 ) but ii # 1
Parabolic j.i= f I (i = I , 2 ) .
Figure 6.6.1. (a) Flow on a torus and correspondingmap on Poincare section; (h)
rational trajectory.
138 C ~ U O Sln Humiltonlun systems
Problem 6.6.1
Prove that area is conserved in the Poincart: section of a Hamiltonian
system.
+
< ln(1 A )
I ~ ' ( . Y ' ,, .\->)/(AU/1,'2)
rhc orbits are closed and periodic. Since the flow is integrable we can
concludc that a material line marked in the flow will stretch linearly. Also,
c\ery segment of a streakline fed into the flow will stretch linearly.
Howevcr, as we have seen in similar examples (Example 2.5.1) the
streaklines can be quite complicated. This flow is reconsidered in Chapter
8. Section 8.3.
E-rumple 6.7.2
In the simplest case, the basic equations for the motion of point vortices
in the plane form a Hamiltonian dynamical system (Batchelor, 1967,
1'. 530). Thus if we consider point vortices of strengths r i , . K ? , . . .. K,,. with
positions (s,, .
y1), (.\-?, y 2 ) , . . . (.Y,,y,,) the instantaneous value of the
\trcamfunction is
I<,
rl.\-j
=
c'w (lyi aw
Ki =--
tlt ?r; ' tlt ax,
3fjust one vortex (one-degree of freedom) and that the case of two vortices
In the plane can be integrated has been known for a long time. In this
case, for example, the distance between the two vortices remains constant,
and details can be found in Batchelor (1967, p. 530), and various other
places. However, it was only recently that the case of three vortices in the
plane was shown to be integrable, and that a system of jbur, o r more
than four, vortices in an unbounded two-dimensional region is in general
chaotic (Aref and Pomphrey, 1982; for a review see Aref, 1983).
-
a n d g are smooth. In this work we will identify f with a n Eulerian velocity
ficld and H with the streamfunction. Consider that the unperturbed system
I;: 0 ) has a hyperbolic saddle point H with a homoclinic orbit, q"'(u1, .uz)
= const. o r q"'(t), and with stable, W ' , and unstable, Wu, manifolds
surrounding a n elliptic point E as shown in Figure 6.9.1(~1). Assume also
that the interior of the homoclinic orbit q"' is filled with elliptic orbits
q"', where r is a parameter ranging from 0 to 1. Denote by T(q'"') the
period corresponding to the orbit and assume also that
,e distance in the Poincare section between the stable. W:. and the
Istable manifold, W,",01' the pertlrrhed system is given by
F ~ g u r e6.9.1. ( t i ) H o m o c l ~ n ~orb11
c before p e r t u r b a t ~ o n(h)after
: p e rttu r b a t ~ o n .
qcO),unperturbed manifolds
n hcrc [
M(to) =
1 denotes
f-+,'I H ( ~ ( O-) (,,)I,
( ~ ~ ( q ( ~-to),
-
of Equations (6.10.2a,b). Select a radius r, such that o(r,) = m/n and an
initial angle 0,. After n mappings we get
0,, = Oo + n2na(ro)= 0, + n2n(m/n) 0, (up to 271) and ro = r,,.
On the other hand, in the non-integrable case we get, in general,
+
0,, = 0, n[2na(ro)+ ph(r,, Oo)l # Oo, and rn # ro
I rational
(periodic)
Behurior neur elliptic points 145
,+llcrc 1.1 is a perturbation and h(r, 0,) is some complicated but calculable
fullction that can be obtained as a function of the perturbations ,/'and q
in the non-integrable mapping (Equation (6.10.2a,b)).In general the point
( I , , , , ; , will return neither to its initial radial position nor to its initial
illlgular position. However, if both the functions h(r. 0,) and a ( r ) have a
Tilylor series expansion in r (condition a ' # 0) we can find a point 0,. I)(,
!]car the initial O,, r,, such that
2na(p,) + ,ull(p,, 0,) = 2n1~7~n.
Thus. for this value of / I ,we d o get 0 , = Or,, bur p,, # I),,. Since the same
i~rgumentholds for any 0,) we obtain a curve p,,(O) of points that are
mapped rudiull!'. Since this happens after exactly n mappings and the
%inding number is monotonic in the neighborhood of r,,, we can envision
rhar the circle corresponding to r, does not rotate and that the neighboring
orbits (r > r,, and r < %) rotate in opposite directions, as shown in Figure
6.10.2 (see Helleman, 1980, p. 185). Also. since the map preserves area,
not all the points belonging to the carve p,(O) can move inwdrd or
ouruard; some have to move inward and some have to move outward.
Also since the mapping is continuous some points d o not move at all (i.e..
tlicy cross the circle r, or in other words, they are fixed points) and due
to periodicity and conservation of area the number of such points has to
bc even. say 2k. Examination of the flow in the neighborhood of the fixed
points shows that k points are hyperbolic and k are elliptic (see Figure
0.10.2). Thus the Poincare-Birkhoff theorem (Birkhoff, 1935) tells us under
iin 1.1-perturbationthe rational orbits break into a collection of k-hyperbolic
irrational
146 Chuos in Humiltonbt~sv.stetn.s
Figure 6.1 1.2. Sketch of formation of tendrils and whorls; (a) shows part of
the manifold structure and orbits near the elliptic point E. ( h ) shows the
,,
deformation of a line with an initial condition passing by H I I , , and E (broken
lines). In general, the stretching produced by tendrils is much more noticable
than the whorl produced by the elliptic point.
Near integrable chaotic Hamiltonian systems 151
Figure P6.11.2
\ material
line at
t=0
\ material
line at
time r
152 Chaos in Hamiltoniun systems
Bibliography
Much of the material of this chapter, especially in the first sections, makes
reference to classical mechanics and theory of differential equations and
dynamical systems. For an introduction to classical mechanics we suggest
Percival and Richards (1982). An excellent but more advanced treatment
is given by Arnold (1980). A modern encyclopedic treatment covering
quantitative dynamics and the three-body problem is given by Abraham
and Marsden (1 985). Traditional treatments are given by Goldstein (1 950)
and by Synge (1960). For the general theory of dynamical systems we
recommend the books by Hirsch and Smale (1974) and any of the books
by Arnold (Arnold. 1983; 1985) and the article by Birkhoff (1920),
which is a long but accessible account of the foundations of dynamical
systems with two-degrees of freedom written in the language of classical
mechanics.
Much of the material on chaos reviewed in this section can be found
in the recent books by Lichtenberg and Lieberman (1983), Guckenheimer
and Holmes (1983) and in the review article by Helleman (1980). The book
by Lichtenberg and Lieberman (1983) addresses primarily Hamiltonian
systems and is especially useful. Other general articles containing material
on dynamical systems and chaos are: Moser (1973) which is the main
reference for the twist theorem, and Arnold (1963) which is the main
reference for the KAM theorem. This last paper is especially recommended
in spite of its difficulty and length. A useful collection of classical papers
addressing chaos in Hamiltonian systems is presented by MacKay and
Meiss (1987). The Melnikov method is analyzed in detail by Wiggins
(1988b).
Notes
I Newton's second law for a particle of unit mass is:
d2x/dt = F
where x represents the position of the particle and F the sum of the forces acting
on the particle. Using standard techniques (see Hirsch and Smale, 1 9 7 4 ) the second
order system can be transformed into a first order system by defining q i = .Y, and
pi = d.c,,/dr. I f in addition we define the Hamiltonian H ( p , q )
H ( p . q ) =:lp12 + V ( q ) ,
which corresponds to total energy of a particle moving in a potential V . Newton's
equations can be written as
dq,/dt = c'lf/c'p,, dp,/dt = i M / ? q , ,
which are known as Hamilton's equations (Goldstein, 1 9 5 0 ) . From a more general
viewpoint, in classical mechanics it is shown that Hamilton's equations can be derived
from Lagrange's equations which in turn are an extension of Newton's equations
(see Goldstein. 1950, for a complete discussion). It should be noted that although
Notes 15:
there is no new physics involved. Hamiltonian theory often gives a much more powerful
itls~ghtinto the structure of problems in classical mechanics. Hamiltonian systems
with the hamiltonian even if the momenta and position are called reversible, i.e.,
t-+-1ifp-t-p.
1 For example, in an oscillating circuit with no resistance but containing a saturated
core, inductance can be described in terms of Hamilton's equations (see Minorsky,
l962. p. 62).
7 Note that in a steady two-dimensional isochoric velocity field the streamfunctions
;:~tisfics DGlDr = 0.
j A necessary condition is that the Jacobian d(p, q)/ci(l, 8 ) be equal to one.
5 Sincc finding the action-angle variables (often referred to as normal coordinates)
i\ ;i way of solving problems in Hamiltonian mechanics, much of the initial effort
111 thib area was focused on finding such transformations. assuminq their existence.
Thc general possible ways to accomplish this task are the subject of the Hamilton-
Jacob1 theory (Goldstein, 1950, Chaps. 8 and 9). However, as it became clear in
rcccnt ycars it is almost always impossible to find the action-angle variablescorresponding
to a glvcn Hamiltonian.
6 Note that the action-angle variable transformation does not work o n the separatrix
(see Figure 6.4.1). For example a hyperbolic system dxldl = .u, dyldr = - y cannot
hc tr;insformed into a twist map.
7 According to Peixoto's theorem these phase portraits are structurally unstable in
the class of 1111 systems (for example if one allows a small amount of dissipation:
i.e., volume contraction). Note, however, that this takes usout from the classof
Hamiltonian systems.
X This has to happen even in dissipative systems and is thecontent of the 'A-lemma'. See
Guckenheimer and Holmes, 1983, Section 5.2; Palis, 1969.
0 The '7.-lemma' implies homoclinic tangles. The implications of infinitely many
~ntcrsectionsof the stable and unstable manifolds in Hamiltonian systems were pointed
out by Poincare in 1899 (see Arnold, 1963, p. 178)
10 Meaning all, except for a set of Lebesgue measure zero.
I I Ilistinguish clearly the assumptions from the conclusions; compare with the twist
theorem.
I? The effects ofsurviving KAM surfaces on transport in Hamiltonian systems are discussed
by MacKay, Meiss, and Percival (1984). A tutorial presentation of these and other
matters is given by Salam, Marsden, and Varaiya (1983).
13 The condition C' ( s > 5) is not sharp (see Moser, 1973). This theorem was proved
by Moser in 1962 with s 2 333 and later improved by Riissman t o s 2 5; with additional
conditions it can be reduced to s 3.
14 A substantial review is given by Escande (1985). An application example is discussed
by Greene (1986).
15 Also. it should be emphasized that according to the values of the parameters, the
region ofchaos might occupy a very small region of the phase space. In most Hamiltonian
systems there is no sharp 'transition to chaos' and chaotic islands and bands surrounding
Islands of instability. sometimes too small to be detected by computational means,
may exist for any non-zero value of the perturbation. Examples are given in Chapters
7 and 8 . Sometimes the term 'transition to global chaos' is used to refer to the breakup
of thc last KAM torus.
I6 Note that the whorls can be produced in flows without vorticity and even in llows
without circulation (Jones and Aref. 1988).
Captions to color illustrations
Figure 1.3.5. Mixing below the mixing transition in a mixing layer, visualization
by laser induced fluorescence [the term 'mixing transition' refers to the onset
of small scale three-dimensional motion]. The experimental conditions corre-
spond to a speed ratio U,/U, = 0.45, and a Reynolds number based on the
local thickness of 1,750. A fluorescent dye is pre-mixed with the low speed
free stream. Laser induced fluorescence allows the measurement of the local dye
concentration, ( a ) single vortex, (h) pairing vortices. (Reproduced with
permission from Koochesfahani and Dimotakis (1986).)
Figure 7.4.5. Similar comparison to that of Figure 7.4.4 except that I),,, = 180":
( a ) Poincare section; colored points reveal that the initial conditions 'do not
mix' even for a large number of periods, e.g., see magenta points; (h)
corresponding experimental result for 10 periods (blob initially located in the
neighborhood of the period-1 hyperbolic point). in this case, the blob does
not invade the entire chaotic region.
Fig. 7.3.10
Fig. 1.3.5
Color il1u.strc1tiorl.s
Fig. 7.4.4
Fig. 7.4.6
I ('olor- illlr.stri~tior~.s
Fig. 7.4.7
Fig. 7.4.10
Fig. 7.5.8
Mixing and chaos in two-dimensional
time-perjodic flows
vortex flow are somewhat idealized and admit analytical treatment; the
-
last two examples - the journal bearing flow and the cavity flow - can
be simulated experimentally.
7.1. Introduction
This chapter consists of two parts. In the first part (Sections 7.2 and 7.3)
we study in some detail two idealized periodic mappings, which can be
regarded, in an approximate sense, as building blocks for complex velocity
fields: ( 1 ) the tendril-whorl mapping (TW), and (2) the blinking vortex
flow (BV).' Both flows can be regarded as prototypical local flow
representations of more complex flows and they are simple enough so as to
admit some detailed analytical treatment. The analysis concentrates on:
(i) the structure of periodic points and their local bifurcations, and (ii)
the global bifurcations, or interactions among the stable and unstable
manifolds belonging to hyperbolic points. It is found that these simple
systems possess a rich and complex mixing behavior. Both mappings are
chaotic according to muthemutically accepted definitions. One of the
lessons to be extracted from the complexities encountered in the analysis
of these mappings is that there are practical limits to the level of probing
and understanding of the details of mixing in complex systems.
In the second part (Sections 7.4 and 7.5) we focus on two flows which
do not admit such a detailed treatment: (3) a Stokes's time-modulated
flow in a journal bearing, where we focus on the comparison between
experiments and computations, as well as the need for various compu-
tational and analytical tools; and (4) several kinds of low Reynolds number
cavity flows in this case we compare the performance of steady flows
-
with that of periodic flows and study the evolution and bifurcations of
coherent structures. Undoubtedly, many other examples and variations
will occur to the reader.
Tendril-whorl flow 155
should be clear that even though the examples presented here are
gc)vcrnd by the theory of Chapters 5 and 6, our problem is not simply
one of adaptation and much work remains. Possibly the most important
diffcrcncc between conventional dynamical systems studies and the work
here is that we are interested in the rate at which phenomena
occur. Thercfore attention is focused on low period phenomena since we
want to mix both efficiently and quickly. There is also the matter of the
area occupied by chaotic and 'non-chaotic' or regular regions. Presently,
there are no predictions of the size of the mixed and unmixed regions,
no thorough studies regarding the stability of the results with respect
to geometrical changes, and no indepth studies of the relationship of
the flow to the morphology produced by the m i ~ i n g . ~
An important question to keep in mind throughout this chapter is:
What constitutes a complete analysis or understanding of a mixing flow?
In the first two examples we follow a program such as the one outlined in
Table 7.1. A loose definition as to what constitutes complete understunding
is the following: A system can be regarded as 'completely understood',
from a practical viewpoint, when the answer to the n + I question in a
program, such as the one sketched in the table, can be qualitatively
predicted from the previous n answers. In the first two examples, since
we have an explicit expression for the mapping, we can accomplish a
substantial portion of the program. In the case of the journal bearing flow
we have an exact solution to the Navier-Stokes equations, but the Eulerian
velocity is not nearly as convenient as having the mapping itself. However,
there are number of avenues left for analysis and we exploit many of these.
In the case of the cavity flow we do not have a solution for the velocity
field and the comparison with computations and analysis has to be of a
less direct nature.
Study of symmetries Entirely analytical Convenient to produce Can be used to Can be exploited to reduce computational
symmetric Poincare verify experimental time (e.g.. search of periodic points)
sections accuracy
Location of pvr~odicpoints
Order one. or fixed points. Only in very simple cases Order one given by conjugate lines Lowest order points most
Order two Higher order obtained by important in rapid mixing
minimization/symmetry methods
Order n
Locul unulysis
Stability of periodic points Possible in simple cases Possible (orten only method Regular islands surrounding elliptic
Biiurcation as parameters to obtain eigenvalues) points (obstruction to mixing)
are changed
PoincurJ sections Only in very Relatively easy Long time behavior of the
simple cases system Tor all initial conditions
Glohul analysrs
Stable and unstable manifolds Only possible in Can be done by placing a blob Hard, purely accidental Speed along manirolds is proportional
associated with hyperbolic points very simple cases encircling the point to magnitude of eigenvalueslperiod.
(order one, two, n ) . Degree of manirold overlapping
Interactions between manifolds controls overall rate or dispersion.
Melnikov method Possible in simple cases Relatively easy Generally valid Tor small perturbations only:
convenient iranalytical expression for
homoclinic/heteroclinic trajectory is available
Liapunov exponents Relatively easy
Stretching of material Hard for large stretchings Relatively easy
lines and blobs
Formations of horseshoes Possible in simple cases Resolution might be a problem Possible, especially in systems
but possible with suitable symmetries
the simplest case is complex enough and can be considered as the point
of departure for several generalizations (smooth variation, distribution of
time periods, etc.).
The physical motivation for this flow is that, locally, a velocity field can
be decomposed into extension and rotation (see Section 2.8). Alternatively,
according to the polar decomposition theorem, a local deformation can
be decomposed into stretching and rotation (see Section 2.7).
In the simplest case the velocity field over a single period is given by
I?, = - CX, 0, = EY, for 0 < t < Text,extensional part (7.2.1)
[', = 0, uO = - o ( r ) for Text< t < Text +
Trot,rotational part (7.2.2)
where Textdenotes the duration of the extensional component and Trot
the duration of the rotational component. Thus, the mapping consists of
vortices producing whorls which are periodically squeezed by the hyperbolic
flow leading to the formation of tendrils, and so on. The function o ( r ) is
a positive quantity that specifies the rate of rotation. Its form is fairly
arbitrary and the results are, in a qualitative sense, fairly independent of
this choice. The most important aspect is that o ( r ) has a maximum, i.e.
d(o(r))/dr= 0 for some r. Nevertheless, independently of the form of w(r),
we can integrate the velocity fields (7.2.1) and (7.2.2) over one period to
give
fext(x, Y) = (xla, a ~ ) , frot(r7 6) = (r, 8 + A$),
where a = exp(~T~,,) and A8 = -o(r)Trot/r, i.e., the point (x, y) goes to
(.K/z,ay) due to the extensional part, the point (r, 8), which corresponds
to (xla, ay), goes to (r, 8 + A8) due to the rotational part, and so on.
I
and
(
iu
[(*')sin
AO + zy cos AO ], : [(;)sin A 8 + ay cos A 8 1
After some manipulations we get
This expression is valid for all choices of AO(r).For our choice, we obtain
tr[Df(p*)] =2 +G (7.2.3)
where G is defined as
G = ( r - l / z ) t a n - ' [ ( z - 1/2)/2!(r*- I )
G = g(x)(r*- 1 ).
Thus, the character of the eigenvalues depends on the value of G. We have:
G >0 + hyperbolic,
G =0 + parabolic,
0 > G > - 4 + elliptic,
G = - 4 + parabolic,
- 4 > G + hyperbolic.
Based on the above analysis we can now study the sequence of (local)
bifurcations that take place as P is increased from zero for a fixed value
I
Figure 7.2.3. Graphical interpretation of the equation for location of period-l
points.
I.- no penod~cpoints
,-
I Z
one period~cpoint
\
material line
saddle-node
period doubling
162 Mixing and chaos in two-dimensional time-periodic Jows
the two at radial distance r* > 1 are hyperbolic, while those at r* < 1 are
initially elliptic. Thus, at P = 1, r* = 1 there is a 'saddle-node' bifurcation
(Guckenheimer and Holmes, 1983, p. 1 56).4 As P is increased further, the
inner fixed point, which is initially elliptic, moves closer to the origin and
G becomes more negative. If g(cc)> 4, for 1) large enough, G = - 4 and a
second bifurcation takes place in which the elliptic point becomes parabolic
and then hyperbolic. In addition there are two additional period-2 elliptic
points. This bifurcation is called 'flip' or 'period-doubling'.
Assuming that cc and fl are large enough to have period-2 fixed points
we have
f2(p*)= p* (7.2.4)
i.e., the following relations hold:
f(r7, 0:) = (r;, 0;)
f(r;, 0;) = (r?, 0:)
where (r:, 0:) and (r:, 0;) are the two period-2 points formed near r*, 0*
I
A material line
(by symmetry we can easily obtain the results, r*, O* + n).On substitutin
in the mapping, we obtain:
r* - r *, (cos2 O:/a2 +
a' sin2 0:)'l2
r* - *
- r2(cos2 O;/a2 + a 2 sin2 0;)'"
0; = tan- '(a2 tan 0:) + AO(r;)
0: = tan- '(a2 tan 0:) + AO(r:)
which can bc simplified to give:
tan 0: tan 0) = I/a2 +
An: = - (n/2 - 20:)
An: = - (n/2 - 20))
r) = yr?
where
where
tr[Df(x,)Df(x,)] =2 + GIG2+ 2(Gl + G,) - 2 + G' (7.2.7)
A0)(x4 - I ) ( I - r))
G2 =
z2(x: 1 ) +
At the birth of the period-' fixcd points r: =r:. r , = r 2= r so that
GI = Gz=G , and
G' = G' + 4G.
A t thc flip bifurcation. G = - 4 which implies that G ' = O so that thc
Period-' points arc parabolic at their point of formation. For slightly
64 Mixing and chaos in two-dimensional time-periodic flows
F~gure7.2.6. (a) Rotat~onalstrength for transition of the periodic point from
elliptic to hyperbolic, a,-,,
as a function of the strength of the extensional
flow, a, for per~od-land period-2 polnts. (Reproduced with permission from
Khakhar, Rising, and Ottino (1986) ) ( h ) Magn~ficat~on of region in (u). The
broken line corresponds to u = 10.
5.0 -
(a)
4 period- l
4.0 - period-2
.c
TZ
3.0 -
2.0 -
1.o
0 5 .o 10.0 15.0 20.0
Q
- - - - ---
1.2
-7
0 1.1
c?.
1 .o
8.O 12.0 16.0
(Y
Tmtlril-whorl flow 165
121-guvalues of /j we find that (;' < 0 and the period-2 points become
cIliptic. As /j is increased furthcr, depending on the value of x, a second
flip may occur in which the period-2 points become hypcrbolic and two
pc1-~od-4elliptic points are formed for each period-2 point. Let us now
cc,~~sidcr a few aspects of the problem which are discussed in more detail
in Khakhar, Rising, and Ottino (1986). Some aspects of the bifurcation
behavior are summarized in Figure 7.2.6, where we have plotted flip
bifurcation values ( i t . , the value of /j at which the elliptic point becomes
hqpcrbolic) versus x for the period-l and period-2 elliptic points. Figure
7.2.6(h) shows a magnified region with the broken line corresponding to
2 = I0 (as is apparent from this figure the period-1 elliptic points do not
exhibit a period doubling cascade of bifurcations). A similar study, much
harder from an algebraic viewpoint, can be carried out for period-3
periodic points, etc.
Figure 7.2.7. Schematic view of mixing zone. Pi and P , are the outer period-]
points, P; and P2 are the inner period-1 points The large arrows show the
direction of transport along the stable and unstable manifolds.
I.ipurc 7.2.8. Manifolds associated with the central hyperbolic fixed point and
period-l points for various values of a and 1.(u) z = 1.500 and P = 1.075. (h)
r = 1.500 and /{ = 1.099. Note the change between figures ( ( I ) and (h; in (u)
P , forms a homoclinic connection; in ( h ) the central point forms a homoclinic
connection. (Reproduced with permission from Khakhar, Rising, and Ottino
(IOX6).)
Mixing and chaos in two-dimensional time-periodic flows
Figure 7.2.9. Same as Figure 7.2.8, but at larger values of r . In ((1) r = 5.000
and /j = 1.082. (h) r = 5.000 and /j = 1.595. Elliptic islands are expected
surrounding P; and P, in case ((1). In ( h ) the islands have been reduced in
size (note change of scale). (Reproduced with permission From Khakhar.
Ris~ng,and Ottino (19861.)
Tmdril-whorl flow 169
and
-
f,,,(r, 0 ) = (r[(cos2 O/r2)+ r 2 sin2 OIL", t a n 1 ( r 2tan 0))
polar coordinates. Show that fr,,,(r,0) (r, O + AO) in polar coordinates,
Pr~hlem7.2.2
Prmc that the point at r = 0 is hyperbolic
Figure 7.2.10. Horseshoe formation in the tendril-whorl flow. In ( a ) the
elongational flow acts first, followed by a twist clockwise. In ( h ) the tw~stacts
lirst, counter-clockwise, followed by the elongational flow. (c) Actual example
corresponding t o r = 10.000 and = 2.180.
, .
\
\
I ./ out of
rectangle
twist, ciockwise
' I I \,
/
/ \
\
\
\. A' goes to A"
I
twist, counter-clockwise I i
I
stretching,~,,< 0
backward
Blinking vortex jlow 171
problem 7.2.3*
Inject particles in one of the conduits and study the distribution of
times in the 'chaotic' region. Interpret in terms of the behavior
of manifolds.
p~.ohlem7.2.4
consider the flow o, = Gx,, o2 = KGx,. Can a function K = K(t) be
specified such that this system will give horseshoes?
T O""
72 Mixing and chaos in two-dimensional time-periodic flows
The velocity field due to a single point vortex at the origin is given by:
or = 0, o,, = T/2nr
where T is the strength of the vortex. The mapping consists of two parts;
each of the form
G(r, 8) = (r, 8 + A8)
where r is measured with respect to the center of the vortex and where
the angle A8 is given by A0 = TT/2nr2. Placing two vortices at distances
( - a , 0) and (a, 0) in a cartesian co-ordinate system (Figure 7.3.1), the
complete mapping, in dimensionless form, is given by:
fi(x, y) = (ti+ (x - t i )cos A8 - y sin AH, (x - ti)sin A8 + y cos A8)
where ti denotes the position of the vortex i (i = A , B), A0 = p/r2, with
p = TT/2na2,and r = ((x- ti)2+ y2)"2 (distances are made dimensionless
with respect to u so that the vortices A and B are placed at ti = +_ 1). In
the following analysis we assume that the vortex at ti=,= 1 is switched
on first and that rotation is counter-clockwise.
In this case perturbations are introduced by varying the period of the
flow, i.e., by starting from the integrable case, p = 0, and increasing the
value of p. As we shall see there are several important differences with
the TW mapping: (i) in the BV both flows are weak'' (i.e., the length of
filaments increases linearly in time) instead of having a sequence of strong
and weak flows as in the TW; and (ii) the BV flow is bounded, i.e., material
is not attracted into the flow from large distances; and (iii) the flow is a
one-parameter system rather than two as for the TW map."
islands appears only for long times (0(104)iterations) since they are of
high period.
With this as a basis, we explore some of the above behavior by focusing
on aspects of local and global bifurcations. The analysis is augmented
by computations of Liapunov exponents, 'irreversibility', and mixing
efficiencies. In the last part of the analysis we use the mapping to compute
the evolution of the intensity of segregation which might be used to mimic
mixing times. Several other possible numerical experiments are suggested
and many others might occur to the reader.
To find the period-I points, first we map the x-axis by vortex A c-lockwise
und with hulf' tile trnyle. Then we do the same thing with vortex B but
i.e., one line is the mirror image of the other about
c~ourtrer-c~loc~kwisr,
J = 0. The points at the intersection of the mapped lines, c~or!jugurelines,
period-] points
Blinking vortex flow 177
;ire the location of the fixed points (Figure 7.3.5(u)). The reader should
vcrjfy that this is indeed the case."
Figures 7.3.5(h)-(d) show the location of the period-l periodic points
for different values of the flow strength p . The periodic points are labelled
by integers n , , 11, which are the number of complete rotations of right
and left vortices respectively. At low values of ,u (Figure 7.3.5(u)) there
;ire many periodic points, all of which lie above the x-axis. However, as ,u is
increased (Figure 7.3.5(h))all but three intersections disappear, and finally,
;it high strengths (Figure 7.3.5(c))new periodic points are created, some of
w h ~ c hlie below the x-axis.
question to ask next is which of the periodic points are hyperbolic,
which are elliptic, etc. This entails the computation of the Jacobian of the
mapping and an analysis of the eigenvalues at each periodic point. As with
the T W mapping, we can write the trace of the Jacobian as
ji-
j.> 0 (for additional details, see Khakhar, 1986). At low flow strengths,
0 ', the point is hyperbolic. As / i is increased the point becomes
parabolic at 11 2 15 and then elliptic (this kind of bifurcation is characterized
by ;in cxchange of stability and the birth of two fixed points of the same
period: the bifurcation is called 'pitchfork', Guckenheimer and Holmes,
19x3. p. 156). An analysis for the remaining period-l points yields the
follo\ving conclusions: If the graphs of the segments ni and ni are just
tangent to one another, at the point of formation the periodic point is
llli[i;illy p:irabolic, and on decreasing 11, splits into two period-l periodic
Pc"n[s. the one closer to the origin is hyperbolic, the other elliptic (see
G~lckcnhcimerand Holmes, 1983, p. 146).
\\ mapping clockwise
mapping counter-clockwise
with A812 (a)
with A812
Blinking vortex ,flow
180 Mixing and chaos in two-dimensional time-periodic flows
x, = f(x,- 1 )
and
mi = D f ( x i ~ l ) ~ m i ~ l / i . , ~ ,
where i i - ,is the length stretch in the i - l to i cycle. The total length
stretch, i.'"'. from i = I to i = 11 is
I
returns to A
182 Mi.uing trnrl chtros in two-rli~~~c~n,siot~rrl
tir~c~-pc~riotlic
,flo~s
' ,.
,,..+-tZ\.,
. .-........ +..--*.-.., +,.... -- --.. .....--.,...... ....-.-
o--* -. . .'
Blinking vortex jlow 185
Figure 7.3.9. Effect of regular and chaotic flows on a blob, (a) initial condition,
( h ) blob after mixing with p =0.8 and 30 cycles, and ( c ) blob after mixing
with p = I and 24 cycles. (Reproduced with permission from Khakhar, Rising,
and Ottino (1986).)
Blinking vortex flow 187
where C is the concentration of points in the pixel and the angular brackets
represent a volume average. Since I is reminiscent of the intensity of
segregation (Danckwerts, 1952) we will use this name. Figure 7.3.12 shows
an example of such a computation where the value of the intensity of
segregation is plotted as a function of pN. Even though the results of the
computations are specific to the grid size used and the initial location of
the blob. in most cases studied there is a rapid decrease of the intensity
of segregation. I t is possible however, for the particles to be 'demixed' at
short times. The curve corresponding to 11 = 0.5, which shows an apparently
anonlalous behavior, can be explained in terms of the restrictions imposed
the degree of overlapping of man~folds.Recall that in this case during
the first stage of mixing the particles are dispersed in the region near the
"Orticcs ;111dI decliys quickly. In the second stage, however, the dispersion
Of the ~;~l.ticlrs is controlled by transport through the cantori. and 1
slowly. as marc and more particles leak through the gaps of the
1.egio11s.At higher flow strengths (1, > 0.5) the restriction is hardly
F ~ g u r e7.3.1 1 . Obstruction t o dispersion of blob d u e t o poor communication
of m a n ~ f o l d sa t 11 = 0.5 after 25 cycles of the flow. Placement a s in ( a )a n d ((,)
results in dispersion indicated in ( h ) a n d (d). respectively (see Figure 7.3.10
a n d compare with 7.3.2. (Reproduced with permission from Khakhar. Rising.
a n d Ottino (1986).)
, .
...:..
) .
. . .. , . 1.
,
I
...
. . .. , . . . ' : . . , .
I
, , :, '. ..' . ~
... . . . .
( I
. .
.. . .. . . . . . : . . . . 1- I . (
- * . . , . . . . .. , .. . .. . .
. . . ,
. .
. .. .
-. ,,. <..;,:..,
. I .
.
. :... . .
.
. ,
,,.,.-
. . . . . . . .
,:
.
I(.
, , .
,; , . ;.. '. ' ' I .
.
., .. .~ . ,. ..,.,
.... .:. . ;
.
:. ;.:;,;- ; .,, : I '. .;-.
. . . . . , . .. ., .. ..., ., . .
.:
... . . .. '. . . ' . .. . . .. ... . . ... .: ... ... ..\..., .(. . . / /
. ~ . .. ..
_ I j
. . ,,. .
.i.
'
, ... , .., .. .
.;
, "
.' , . .?., ,,;- ,;.I,! *.<; , . ' ,.
.. . : . . : . ,
.,)
I . ,
, . : . ; : . -. ,.,.;..
. ' i
' . .. . . _
. . .,. ... . .. ' , .
:
. ..
.
. ., .
, .,;,.
:. . . . - . ,,, .
,..,:..,,
,
.,,,.
., , ,. .
I .
, , '.. $..'
. .
... ,...: ' . .
.. , .
., .. .
. , .
:, .. .: .
, .
.. , .
,
,
,
.
190 Mixing und chuos in two-dimensionul time-periodic flows
Example 7.3.1
Rather than perturbing the integrable system of Figure 7.3.2(a)by turning
vortices on and off in a discontinuous manner, consider the situation of
Figure E7.3.1 where the system of two co-rotating vortices is perturbed
by a linear flow
4.0
Blinking rortex flow 19 1
-4resolutionI-
base flow
perturbation
192 Mixing and chuos in two-dimensional time-periodic flows
+ + +
If we denote (x - I ) 2 y2 = Dl, (x 1 )' y2 = Dl, then DID, = 1 on the
homoclinic orbit t,h0 associated with the central hyperbolic point. The
evolution o f a point (r, 0 )belonging to the homoclinic orbit isgoverned by
drldt = - r sin 20, dOldt = r2 - cos 20,
which can be integrated to give
0 = tan- '(tanh t), r = (2/cosh t)'12,
or, in x,y co-ordinates
.u = 2''' cosh tlcosh 2t, y = 2'" sinh tlcosh 2t
These are the parametric equations of the homoclinic orbit *O. In order
to set-up the Melnikov function, we need to evaluate f = (f,, f 2 ) on
$O[f($O(t)]means the velocity of a point belonging to 4'). In this case this
happens to be
, f l = - 2112
sinh t(l 2 cosh2 r )+
cosh2 2r
1. - 2112 cosh t(l - 2 sinh2 t )
. 2 -
cosh2 2t
We need now to account for the effect of the perturbation 8, = Gx2,
g2 = GK.yI on the homoclinic orbit, i.e., (,flu2- f 2 y l ) evaluated on $O.
sinh r cosh t ( l + 2 cosh2 t )
f 1 q 2= GKf1.y = -2GK
coshqt
sinh t cosh t ( l - 2 sinh2 r )
f i g , = Gf,y = 2G
cosh" 2r
Consider now perturbations of the type G ( t ) = Go(', + hhl(r)) when
11, ( t )= c,, exp(ico,r). In this case the Melnikov's function is
MI ( t o ) = -Gob exp(icot,,)
2(K + I ) sin 2r
x J+'
- il
exp(itot)
cosh3 2t
+ (K - I ) sinh 2r
cosh2 2r 1
Mixing in u journul beuring flow 193
problem 7.3.1
Interpret p h y 4 ~ a l l ythe parameter p
prohlern 7.3.2
Examine geometrically the flow in the neighborhood of A and B in Figure
7.3.6. Is the flow hyperbolic'?
Problem 7.3.3*
Consider that a point u,, y, has not returned to its original location after
n forward and 11 backward iterations if (x, - x,)' + (y,, - 1:'. Study
the fraction of points that d o not return for different precisions and different
value4 of i .
Prohlem 7.3.4*
Study numerically the assumption of random reorientation used in the
examples of Section 4.5. Introduce a blob of infinitesimal vectors and
compute the orientation distribution function as a function of time.
Prohlem 7.3.5*
AS a simple model for the decrease in striation thickness in the BV flow
assume that s = .so exp(-a(/i)n) where a ( p ) is given by Figure 7.3.7(u) and
% is or the order of 2tr. Assume that the intensity of segregation levels off
when s = c 5 l K where iiis the pixel size and K is a number of order 10'-lo2.
Calculate the values estimated in this way with the ones obtained in Figure
7.3.12. What is the best fit of K ? Develop a more sophisticated model.
Figure 7.4.1. Flow region in journal bearing flow and definition of parameters.
f i P denotes the approximate location of the period-l hyperbolic point.
Mixing in u journul bearing jlow 195
if both S r < < 1 and Re << 1 the flow can be regarded as quasi-static, and
time enters only as a parameter. From Re<<1 we get the condition
( V ; L ) < <v , , ~ ~and
, from Sr<<1 we get o<< V / L . Combining the two
, , ,+
Figure 7.4.3. Angular velocity histories: R,,(t), R,,,,(t). As long as the motion5
of the inner and outer cylinders d o not overlap and the displacements arc
kept the same, histories such a s ( a ) and ( h ) are indistinguishable and produce
%M~:
identical results (e.g.. identical Poincare sections), Note also that the inlt~al
angular displacement is one-half of the total displacement. This results In
symmetric Poincare sections.'"
c0
-
h
L.
-
h
i-----T-----i time time
However, this kind of agreement is the exception nither than the rule.
nit h ally, Poincari. sections do not convey a sense of the structure existent
within the chaotic region and give no indication of the mixing rate. Figure
7.4.5 (see color plates) shows a similar comparison to that of Figure 7.4.4.
In this case however, after the same number of periods, the blob invades
only a small part of the chaotic region. A more careful analysis of the
poincare sections reveals part of the reason. Initial conditions, identified
by color, 'do not mix' due to poor communication of manifolds of low
order hyperbolic points, even for a large number of periods (see Section
7.3). However, color-coding does not solve all the problems and Figure
7.4.6 (see color plates) shows an example where this is indeed the case. The
figure shows two Poincare sections, identical in all respects except that in
Figure 7.4.6(u) the angular displacement, O,,,,, is 165 , and in Figure
7.4.6(b) it is 166'. A naive interpretation of these results, especially after
the discussion on tendrils' and whorls of Section 6.1 1, might suggest that
the mixing structure in both systems should be very different since ( h )
should produce whorls within whorls whereas ( t r ) should produce just one
large whorl (not taking into account the smallest islands invisible in the
photographs). Furthermore, if this were the case the results would also
indicate that extreme care should be exercised in experimental work since
an error of 0.6% could determine the outcome of the result. However. a
deeper analysis reveals that this is not so and these variations are unlikely
to be of any consequence in the experiments. Indeed it can be shown that
the rate of rotation within the large island is nearly a solid body rotation
and 'whorls' are not noticeable even for a large number of iterations. The
rate of rotation in the smaller islands is even slower and for all practical
Purposes they can be ignored in the stretching of lines placed in this region.
In most cases the motion within the regular holes does not produce
significant stretching (see experiments on reversibility in Section 7.5). It
is apparent that a thorough understanding of this system requires a
detailed analysis of rate processes. A first step towards this goal is the
location of manifolds associated with the hyperbolic points of various
Orders. In order to accomplish this we first need to locate the periodic
Points of the system.
An important consequence of the symmetry of Poincare sections (Figure
's4.3) is that the search for periodic points is o n r c l i r ~ ~ n ~ r i orather
t ~ t ~ l than
two-dimensional. Figure 7.4.7 (see color plates) shows the location of
Periodic points up to order 8 for the case of a discontinuous velocity
along with the corresponding Poincare ~ e c t i o n . ~The ' character
Of the points can be determined by examining the eigenvalues of the
linearized flow; the crosses represent hyperbolic points whereas the crl-clcs
represent elliptic points. Note that, as expected, the elliptic point5 kill
within the islands, and there is a heavy concentration of points near the
inner cylinder since the motion in the immediate neighborhood of the
cylinders has (nearly) circular streamlines and therefore, points of
orders are possible.
The rate of spreading of a passive tracer is controlled by the 1111\tnbl~
manifolds of the hyperbolic points and is roughly proportional to the
value of the eigenvalues and inversely proportional to the period of the
point. Figure 7.4.8(u) shows part of the manifolds associated with perrod-I
points after five periods of the flow. Figure 7.4.8(h) shows the manrfolds
associated with four period-4 hyperbolic points and sixteen periods. Note
that the spreading is significantly less and that the period-l manifolds act
as a templurp for the structure of the system. A blob placed in the
neighborhood of the point is 'captured' by the low-period manifolds.
Indeed, Figure 7.4.5(h)(see color plates) can be regarded as an experimental
manifestation of this effect.
All the previous results, including those of the examples of Sections 7.2
and 7.4, were for the case of a discontinuous velocity history of the
boundaries. However, it might be argued that such a history is hard to
achieve under laboratory conditions and that smoother histories might
result in different phenomena; that is, the relevant question seems to be:
How different are the results if the histories are different'? Figure 7.4.9
shows several limit cases protluc,itly rltr s~lrnrwtrll displricet~~etzt
pcr. pc~r.iorl.
The main result insofar as Poincare sections are concerned is that the
results are remarkably similar and that the most important aspect of the
flow prescription is the period rather than its shape40 (see comparison
in Figure 7.4.10, see color plates). Similar results hold for other systems
such as the blinking vortex of Section 7.3.
The previous discussion suggests that most of the stretching, and
therefore mixing, occurs in the neighborhood of the unstable manifolds
of low period points. In order to investigate the regions of significant
stretching we adopt the following procedure: We integrate the equation
DFIDt = ( V V ) ~F. with F(X, 0) = 1
for various initial conditions X and compute the stretching of a vector
with initial orientation M as
i.
= (F.M(.
Problem 7.4.1
Discuss the possibility of using the method of conjugate lines of Sect~on
----
1.-igurc 7.5.1. f l o w region. The dimensions of thc cavity are W = 10.3 cm ancl
I 1 = 6.2 cm.
n n
glycerine
ters
plexiglass
tank
alum
Figum 7.5.2. Stretching of a line placed vertically. a s shown in b).in v a r i o ~
steady flows ( b x d ) .The fluid is glyarim and the t r a a r is a fIUOXuamt dye.
The expcrimental.conditions arc the following: Re = 1.0 and the vetocitk~of
the top and bottom walls are ,o., =, o = 1.58 cm/s. In (b) both walls move in
the same direction, in (c) both walls move in opposite directions, and (d) only
the top wall moves. The total time of the experiment is 5 min.
204 M i x i n g und chaos in two-dimmsionul time-periodic flows
1,'igurcs7.5.3(t1), ( h ) , ((.I, (ti) and returns to (u) at the end of the period.
wc ,xgard the period T as the governing parameter.
~'igurcs7.5.4 show the result of placing a blob as initial condition when
s?stcm operates in a time-periodic mode. It is clear that the periodic
,pcr.;~tionimproves the mixing (otherwise the blob would remain trapped
1,'igurc 7.5.4. I!volution of u blob of initial conditions originallq plucctl at th,
position \- -: 2.2 cm. !. = 3.1 cm (diitnictcr = 0.5 cm). Visualization is providcc.
by a fluorcsccnt dye dissoltcd in glycerine. excited by long-wuvc ~~ltritviole
light of 365 nm. The top and hottoni walls move according to the prcscriptior
of equation (7.5.1) with C ' = 2.60 cni s (Kc, = 1.7 and 0.05 < St. < 0.091. Th.
pictl~rcs;ire ti~kcni ~ the
t insti~ntwhen the top wall and hottoni wit11 arc moving
at m a s i m ~ ~and
n i minimum (tcrol spccd. rcspcctivcly. The pictures corrcspont
to time period5 7': (111 15 s (a second blob. near the ccntcr of the cavity wa:.
added in this cube). (1)) 20 a. ( 1 . ) 25 s. and ( 1 1 ) 30 s: and to wall displuccnicn--
(the distance travelled by the walls in one period). ( a ) 1612cm. ( h ) 752c.
I',) 951 cm. and ( t i ) 1452 cni. Note the symmetry in Figure (t0.
b! ,,,,~ll'ininpstrcanilincs). I t appears that. as compared with the journal
hcal-lllg 110%'. the cavity flow !nixes better and fewer periods arc necessary
to ;rchicvc significant mixing. However, this system is a bit more suscep-
tible to experimental error due to corners and the flexibility of the belts.
Another significant difference is that there are no fixed hyperbolic points
208 Mixing tlntl chtros in two-rlit71et1,siontrItimr-periodic t1ow.s
in the velocity field and that we cannot properly speak of small pertur-
bations with respect to steady flow.
An important difference with the previous sections is that in this case
there is no analytical solution for the velocity field. However, computations
based on a discretization of the velocity field provide only a crude picture
of the flow (e.g., exact location of periodic points, manifolds) and we are
therefore left with experiments as the main investigative tool. Therefore,
the experimental observations are limited to macroscopic structures, such
as islands, produced by mixing, and the question is how d o these structures
behave under change of the governing parameter (in this case T ) . Also,
any phenomena which are relatively long term (asymptotic)are not directly
observable, unless they form structures which are more o r less clearly
delineated from the start. As we have seen in the previous section, the
objects which can be detected are those with macroscopic spatial extent
and low periods.
The comparison between analysis, computations, and actual experiments
is not trivial and is often misunderstood. In theory there are always three
systems; a t ~ ~ t r t h r n ~ t i t i system,
ct~l a cotnputer sitnultrted system, and an
e.uperinlc~nta1system. However, it is naive to expect that all the information
extracted from experiments is easily obtainable from computations, and
ric8r w r s a . For example, asymptotic phenomena and sets with measure
zero are mathematical objects, and clearly a set with measure zero is not
observable by means of computations. The requirements for computational
observability are also different from those of experimental observability.
In general an object will be observable in computer experiments if it is
robust, i.e., relatively insensitive to round-off error, and if it occurs with
high probability, i.e., if its presence can be ascertained regardless of the
discretization choice (pixel size in the computer screen). O n the other
hand, an experimental system is fundamentally analog so that non-robust,
low probability events can be seen, though perhaps not easily repeated.35
. 2
Consider now a few experiments comparing the discontinuous and sin
histories (Equations 7.5.1 (a,b)). The comparisons reinforce the need for
careful experimental interpretation and point out some of the pitfalls of
improper and incomplete comparison. Figure 7.5.4 shows the state of the
system operating for various values of T under the prescription of Equation
7.5.I(a,b) (i.e. sin". The initial condition in all the experiments is a small
blob of t r a ~ e r . "In~ this case the system evolves from no holes to a rather
large hole for the highest value of T. Note that the large scale structures
of all cases are similar. By analogy with the previous section we might
conjecture that this is bec:iuse the initial stretching and folding is
cj(,mill;~tcdby low period hyperbolic points a n d their manifolds. T h e
additional folds which occur a s t h e n u m b e r of periods is increased a r e
indcc.d contained within the already striated region a n d very quickly the
l I I I ~ I c Lfolding
~ serves as a template for further stretching, a n d the structure
,,I,~ ~ shown ~,lc in the photographs is very similar t o the o n e observed for
l(,llg (illlcs. T h e holes ( a n d also the large folds) form 'coherent regions'
,\lllch c;Ln be followed in space a n d time: they a r e translated a n d deformed
b) tllc Ilow but conserve identity a n d d o not disappear u p o n further
,,1lk11lg.In terms of qualitative appearance. there seems t o be a n o p t i m u m
mi\.lllg at T = 20 s. Beyond that, a large regular region grows with
illcre;~s~ng 7:
r h c mixing structures produced by the discontinuous protocol are
s u b ~ t : ~ ~ i t i adifferent
lly (Figure 7.5.5). A large hole is present at the lowest
c.aluc of T used in the experiments a n d the results of experiments placing
blobs inside a n d outside the island offer a nice demonstration of the speed
of mixing within the islands a n d in chaotic regions. In Figure 7.5.5(u) the
blob was placed inside the island; in ( h ) the blob was placed outside the
island: it is clear that the interior of the island does not communicate
uith the rest of the flow. An increase in the value of T decreases the size
of thc island and the best mixing is obtained a t approximately T = 58 s .
It thus appears that the t w o prescriptions, the discontinuous protocol a n d
the sin' protocol, give opposite results, a n d that the conclusions of the
prcLious section, robustness of Poincare sections (Figure 7.4.10) a n d s o
on. cannot possibly hold, since in o n e case we get hole-opening with
increasing T whereas in the other we get hole-closing with increasing T.
Houever. this is only partially correct. Indeed if the pictures a r e compared
at cjc~rrtrltlispltrc~rmentsper period, there is a region (values of displacements
In both flows) which produce similar results. Figure 7.5.6 shows a
comparison a t equal displacements for both prescriptions. G o o d agree-
ment. in terms of similarity of large scale structures, is seen, especially a t
Ion \;dues o f T. However. both systems behave quite differently a t large
c.alues o f T.
T h e most important consequences of the coherence present in these
P1ctllrcs is that careful observations (video recording) allow the detection
of pcrlodic points. a n d that the evolution of the system through a series
of changes in the parameter T gives information regarding the bifurcations
Occurring in the system. Figure 7.5.7 (see color plates) shows a n example
ofabifurcation which would be hard t o capture in computer simulations. It
" l ~ \ \ . s the behavior of the island boundary at the time of island collapse.
result is a consequence of the b r e a k u p of the last K A M tori (rotation
210 Mixing trncl c-htios in two-ilirnm.siontrI time-prriotlic ,flows
Figure 7.5.5. Similar to I.'igure 7.5.4 ( K r = 1.2. 0.07 < Sr < 0.1 l ) except that
now the motion of the upper itnd lower walls is discontinuous: the top wall
moves from lcft to right for half a period and then stops. then the bottom
wall moves from right to lcft for half a period and then stops. and so on.
There is a five sccond pause between each half period to rcducc inertial cffects.
The pictures are taken at the end of the desired period. The pcriods T arc:
( 1 1 ) 34 s. ( h ) 34 s. ( 1 . ) 48 s. and ( t l ) 58 s: the corresponding displacements arc:
( ( I ) 1290 em. ( h ) 650 cm. ( 1 . ) 760 cm. and ( t l ) 1 100 em. Note that the structure
shown in ( u ) fits within the 'holes' left in ( h ) and that relatively little
stretching is observed in the interior of the island.
Mixing in cuoity jlows 21 1
(a 3)
discontinuous
Mixing in cavity flows 213
and their moons (elliptic points) returning to their initial locations after the
periods characterizing the points.
wc have seen that the Smale horseshoe map involves the stretching
and folding of a blob onto itself. In the context of mixing in periodic
two-dimensional flows such a map has a clear physical significance: A
fine subdivision of the blob in the region initially occupied by the
blob. Obviously i t is desirable that the flow forms many horseshoes,
possibly interacting, in such a way that they influence a large region of
the flow. From the point of view of mixing it is desirable that they be of
low (number of transformations needed to produce the horseshoe)
since we want to achieve mixing as quickly as possible, and also since the
domain of a high period horseshoe is usually small. The investigation of
horseshoes involves the superposition of forward and backward transfor-
mations with the initial location of the blob. The construction is similar
to thc one given in Sections 7.2 and 7.3. There are several properties
which must be verified in order to deduce the presence of a horseshoe
(Moser, 1973). All the conditions must be carefully examined. For
example, if the placement of the blob is not exact (and it actually never
is) the striations will be partially filled with the tracer. In such a case
the system might appear to violate some of the conditions (Chien, Rising,
and Ottino, 1986).
Chaos magnifies errors and one might wonder to what extent it is
possible to unscramble pictures such as those of Figures 7.5.4 and 7.5.5.
We anticipate that the configurations in Figures 7.5.2(~1)-(c), which are
integrable, can be reversed within experimental error, and there have been
experiments showing that this is indeed possible in creeping flows (see
Bibliography, Chapter 2). However, we have already seen that this might
be impossible, even numerically, if the flow is chaotic (Section 7.3.4).
Figure 7.5.9 (see color plates) shows what happens to initial conditions
placed in the interior of an island and in a chaotic region (see Figure
7.S.4(tl)).Figure 7.5.9(h)shows the result of forward mappings and Figure
7.5.9((')shows the results of backward mappings. I t is apparent that the
line pl;lced in the hole undergoes relatively little stretching" and returns
its initial location. However, the line placed in the chaotic region
( ~ ' ~ ~ u m a bnear
l y a horseshoe) does not return to its initial location; every
'Irne the line passes by the horseshoe it is stretched exponentially and the
experimental errors are magnified. Apparently, two periods are enough
to
significantly the errors of this particular experimental system.
. w e have emphasized that the degree of smoothness achieved in these
'lctures would be hard to achieve in straightforward numeric,I' 1 simulations
bee Examples and 9.2.4). Consider now the experimental determination
2 14 Mixing und cthrios in two-rlimensionul time-periodic) ,flows
Bibliography
The first two sections are tin abbreviated version of the presentation given
in Khakhar, Rising. and O t t i n o (1986). T h e or~ginillpaper treating the
Bihliogruphy 215
,,thcr hand. in the steady state flow the growth is linear in time.
sin2 flow
0-
-8 m I I
+steady
1
flow
100 200
time (s)
2 16 Mixing untl chuos in two-rlimen.sionu1 time-periodic ,flows
blinking vortex system is Aref (1984): the first treatment of the tendril-
whorl map is given in the paper by Khakhar, Rising, and Ottino (1986).
Preliminary results for both systems were given in various presentations:
The horseshoe construction of the tendril whorl system was given in
Rising and Ottino (1985, abstract only): several results regarding efficiency
of stretching were given in Khakhar and Ottino (1985, abstract only).
The material of Section 7.4, journal bearing flow, is based on the
presentation of Swanson and Ottino (1985, abstract only). A substantially
morecomplete presentation isgiveninSwansonandOttino( 1990).Thefirst
study of the journal bearing flow is the computational study of Aref
and Balachandar ( 1 986). Several nice experimental results are presented
by Chaiken cJt (11. (1986), however, most studies involve a very large
number of periods (order 10' and higher). Section 7.5, cavity flow, is
based on experimental results by Leong (Ph.D. thesis, University of
Massachusetts, Amherst); the original paper is by Chien, Rising, and
Ottino (1986). The relationship between analysis, computations, and
experiments is discussed by Ottino rt 01. (1988) in terms of the journal
bearing flow and the cavity flow. A related topic is dispersion in
Rayleigh Benard flows: even though there is a number of recent papers
in this area, the treatments of mixing are rather scarce (e.g., Solomon and
Gollub, 1988). A paper addressing mixing in the context of geophysics is
given in Hoffman and McKenrie (1985).
Notes
I Note that d u e to widespread usage the word 'flow' IS used in two d~ffcrcntways
throughout thls work: the first one in the mathcm;~ticnlsense consldcrcd in Chapter 6,
the second in the accepted fluid mechanical way. as for example, in the terms 'flow
field' o r 'fluid flow'.
2 F o r example. fractal d i m e n s ~ o n sin the chaotic regions. c o r r c l a t ~ o nfunct~ons.etc
3 These terms were defined by Berry 1.1 (11. (1979) In the context of quantum maps.
In practice. whorls arc hard t o obtain in most chaotic flows (see Section 7.4)
4 In a saddle-node hifurcation for area preserving systems. as the parameter of the
system increases the flow goes from n o fixcd points to two fixed points. one hyperbolic
and one elliptic.
5 This kind of limitation is common to moht of the computationi~lstildics described here.
6 Notice that there is efficient mixing in a rcglon where the flow is i~pproximatelylineaf
due to the ~ntcractionw ~ t hthe ol~tsldcflow. T h ~ should s be useful in understanding
mixing at the nl~croscnlcof turhulcncc. ~ h c r cthe f l o ~ arc. s at some scale. linear
and Ianiini~r.
7 Statements like this must alwi~ysbe qualified by saylllg ' w ~ t h i nthe rcsolut~onof
the grnphlca o l ~ t p l ~ctc.'.
t.
X A g a ~ n .scvcl-al possible gcllcrali,;~t~onaarc posbihlc such distribution of pcrlods, et'
0 However. a n important diffcrcncc is that. near the origln. the cakltj flow hchavior
1s I,,, :r wlicrcas it1 the hlinking vortex I.,, :I r .
A, \\c have seen In Section 4.6 this 1s a n indication that this flow might havc a n
O 1 , ~ ~ ~ l ; ~ loperating period for which the efficiency 1s maximized.
I I ,lllolllcr diffcrcncc is that D: D diverges at the center of the vortices and the integral
,,('1). I) i\ ilnboundcd. By contrast. the energy d~ssipationin the T W mapping is
hchi~vcd.Nevertheless. the efficiency in the BV 1s well behaved.
13 A ,1111ilarconstruction can he used t o find the periodic points in the T W map. and
l l l , ~ ~generally.
-c of generalized bllnking flows (Rising and Ottino. 1985).
I 3 .rile ss:itlcr m ~ g h gtun t a n appreciation of the complexi:y ofchaotic systems by considering
rl1.11 ;I \iniilar t~nalysisapplies t o period-11 points, where 11 can be a very large number!
14 .l,lll\ idea was used in Rising and O t t ~ n o(1985).
15 .rile ~ - c ; ~ dshould er notice that in this case. since we d o havc a n expllcit mapplng.
r l l ~calculat~onis free of numerical errors which might arise d u e to renormalization
,,('r~-:~~cctories (see Lichtcnberg a n d Lieberman. 1983%p. 2x0).
16 Tllcrc arc still some questions regarding the computation of long tlme averages d u e
10 rhc finite precision nature of the computations. However, when the motion is
~ \ L I ~ I ' I C I C I I chaotlc'
II~ and numerical prec~sionhigh (see Figure 7.3.10) the time averages
;ll.c c\lxctcd t o be reliable.
17 N ~ r cthat the accurate computation of length is not trivial: we can label points
at~tli l ~ ~ r o d u cadditionale ones if they become too separated and join nearby points.
s:~! I and i + I . by short segments to approximate the length. However. a s the number
of cvclcs increases we need to a d d polnts at a n exponential rate (see Example 9.2.3)
I X Thc.\c numbers are only approximate. Regular reglons a n d chaotic regions are intricately
c o ~ i ~ ~ e c tlcf d; ~.point gets 'trapped'in a regular region it remains there for a longtime.
I9 Althoi~phthe polnts arc elliptic at the point of formation they are hyperbolic for
~ h craligc of 11-value5 considered here
20 This is one of the reasons why 11 is convenient t o construct Poincare sections using
part~clcslabelled with a color representing the initial condition.
71 Thl\ suggests the need for transport studies in Hamiltonian systems. possibly using
rcnorlnitlizntion methods. a s indicated earlier. o r in terms of thedynamics of manifolds
Iscc I.1g. S.X.3.).
22 111 rhc nilmcrical experiments of Section 7.2.6 we can consider the line separating
two difkrent matcrlnls. Then if A is the area of the globally chaotic region and L(O)
the in~tiallength of the line. we can write A = L(O)s,, = L(/).\(/).where .\ is a distance
of the order of magnitude between the vortices. Thus. s ( t ) 2 I,(/) '.
23 It i h in~crcstingt o point out that J. W. Gibbs used a fluid mixing analogy, a regular
flo~ :lclually. to explain the origin of irreversibility in statistical mechanics (Gibbs.
1948). His argument was re-interpreted by Welander (1955) and is closely related
to the content of the previous sections. Although the argument is not entirely rigorous
it brings u p s o m e important issues a n d it is convenient t o repeat it here.
('("l\idcr home two-dimensional region with area H where we have placed a blob
"f 11011-diffusive tracer with density 111=d,,l:drr). We consider that p(x. 1 ) = I if
bclOll@ t o the tracer region and p ( x . I ) = 0 anywhere else. The total mass of the
trilccr i l l K and the integral of the square density are both equal. i.c..
L i . d i 1 - M. 1 /I2 dil = M
"here .A1 is less than H. Given these choices. no matter how erective the mixing.
both thcsc quulitities remain equal ( t h e particles of the tracer always carry the same
1
is
-. I 1) I). However. what happens if the resolution of the integration
tlii111 the striation thickness:' (see Figure 7.3.13). At this scale the density is
(il\ ,TI,, (5.4. nlld the square density. = (i5ttt;ciA)'. Therefore the integral of
2 18 Mixing trnd chaos in two-tlimm.siont11 tinlr-periodic ,flows
but since iinl,'iiA cannot be equal t o one everywhere (since hf < K ) we conclude that
r
-
(x. y ) (-x. - y ) . The symmetries of composition of Hows can be deduced by using
-+
these rulcs For example. the How Fz, BIT,: 1.e.. a series of top and bottom wall
motion\, has symmetry with respect to the x-axis. F,, = S, F ~ , ' s whereas , the flow
G,,= H,, TI B,,; i.e.. a series of top and bottom wall motlons. but where observations
are made half-way through the bottom wall displacement. has y-symmetry instead.
Thi\ means that the Islands ~nthe flow are located in p a r s across the y-axis or on
the axis ~tsclf Note that a How can have more than one symmetry. For example.
the F,, flow can be written also as F,, = S * Fz; 'S* where the symmetry S * 1s given
S * S,T, = T l - l S , .The fixed line of S* 1s not obvious; ~tcorresponds to the mapping
of the xis wlth T , , ' ; that is. the inverse of the top wall flow wlth half the
displacement. In t h ~ case s the symmetry line is a curve (these ideas are developed
In the ;~rtlclcsby J . G Franjlone. C W . Leong. and J . M . O t t ~ n o'Symmetries . with~n
chaos' ;I route to effcctlvc mixing'. Phys. Fluids A . 1. 1772-83. 1989 and C. W.
Leon?. and J . M Ottino. 'Experiments on mixing due to chaotic advectlon in a cavity'.
,' F l l i l ( j Ihlech.. 209. 46.V99. 1989).
Mixing and chaos in three-dimensional
and open flows
8.1. Introduction
S o far we have studied mixing in two-dimensional time-perlodic flows a n d
have seen that considerable theoretical guidance exists for this case. \2'ith
the exception of the tendril-whorl flow. in which fluid entered a n d left
through conduits. the systems were closed. W e now move t o the casc 01'
three-dimensional a n d o p e n flows. Seheral cases a r e discussed. T h e chaptcr
starts by analyzing mixing under creeping flow conditions in two continuous
mixing systems. In the first example the partitioned-pipe mixer the
helocity field is periodic in space: in the second example the eccentric
helical annular mixer - the velocity field is time-periodic. These two flc)\+s
a r e analyzed in some detail since they have many potential applications.
As we shall see, intuition based o n the two-dimensional case can bc
somewhat misleading in the understanding of three-dimensional flous in
general. T h e next example, discussed in less detail. includes inertial effcith
a n d corresponds t o mixing in a channel with sinusoidal walls. 1.11~.
treatment of this case, largely based o n the work by Sobey (1985). is
computational in nature a n d relatihely little analytical work is possible.
T h e next examples a r e of relevance t o turbulent flou,s. T h e first one of
this class is a perturbation o f the Kelhin's cat eyes flow, which is important.
as a first approximation. t o mixing in shear layers. T h e next exampic
considers the class of admissible flows obtained by constructing solutions
of the Navier Stokes equations by expanding the velocity field in a T q l o r
series expansion near solid walls. This example is of relevance to p e r t ~ ~ r b c d
flows near walls and has obvious r e l e ~ a n c eto turbulence. Finally, the la.;[
e x i ~ m p l e originally
. d u e t o H e n o n (1966). is the analysis of mixing in a n
inviscid fluid by Dombre rr (11. (1986)and our presentation is largely based
on their analysis.
where R is the radius of the pipe, and the average axial velocity (u,)is
8 - n2 C7p
(U,) = - R2
4nZp ?z
where p is the viscosity, and p the pressure.' The cross-sectional
component is a two-dimensional flow in a semi-circular cavity, and is
given as a solution of
v4* = 0
where
Figure 8.2.1. Schematic view of the partitioned-pipe mixer; the length of the
plate is L.
Figure 8.2.2. Kenics" static mixer, idealized view of the system. Two streams,
A and B. enter segregated. After the first twist (clockwise looking from the
entrance) each stream is halved and joined by half of the other stream in the
second twist (counterclockwise); the process is repeated periodically. The
actual system is more complicated due to flows in the r , 0 plane, produced by
the twisting of the planes, and developing flows.
with the velocity field given by
$= Z <m(rbrn(())
m= 1
where {a,) is a set of known trial functions which satisfy the boundary
conditions, and (5,) is a set of unknown functions to be found by solving
the differential equations, obtained by minimizing the residual over rhe
interior as in the Kantorovich-Galerkin method (Kantorovichand Krylov,
1964). The weighting functions in this case are the trial functions
themselves, and are taken to be
The one-term solution for the stream function for the cross-sectional flow is
where v = ( 1 1/3)'12- 1.
The streamlines for the different values of the normalized stream
function $* = ll//[4vRR/3v] are shown in Figure 8.2.3. The motion is
obtained by integrating the following set of equations which represent the
approximate velocity field
dr
u = - - = /rl( I - rtJ)
sin 20
dt
1t is clear from this procedure that though the particle path is continuous,
its derivatives are not, and this results in infinite stresses. However,
including the developing flows would significantly complicate the problem
(a flow involving smooth trajectories is analyzed in Section 8.3).
Figure 8.2.5. Poincare sections for the partitioned-pipe mixer for various values
of the mixing strength P : ( a ) /I = 2.0, ( h )/I = 4. ( c , ) /I = 8, and ( d )/I = 10, Initial
positions for the different trajectories were placed on an (r-11) grid: eight
r-values uniformly spaced between 0 and I . and five 11-values uniformly spaced
between 45 and 135 . Each point was mapped for 300 iterations. (Reproduced
with permission from Khakhar. Franjione, and Ottino (1987).)
Mixing in the partitioned-pipe mixer
Figure 8.2.5 cotiri~~lrurl
Figure 8.2.5 corrtirrrred
232 Mi.uinq trntl chtros in three-tlinrensiontrl trntl open flows
In some sense, the exit age distribution gives an indication of the inhomo-
geneity of the cross-sectional mixing: fluid streams in some tubes emerge
faster than others, and there is no mixing between the streams in the
different tubes. This is confirmed by a contour plot of iso-exit times and
it is possible to identify the regions responsible for the two peaks. At higher
values of the mixing strength, /? = 8 (Figure 8.2.9(/?)), the second peak
disappears, in spite of the rather non-uniform cross-sectional mixing
reflected in the Poincare section and the contour plot of iso-exit times.
In this case, some particles of the peak belong to the regular island, others
to the 'chaotic' region outside. In factjhe lowest residence time corresponds
Figure 8.2.8. (a) Poincare section for the partitioned-pipe mixer with plug
flow, (b) Poincare section for the partitioned-pipe mixer for counter-rotating
cross-sectional flows in adjacent elements. The initial positions of the trajectories
are the same as in the Poincare sections of Figure 8.2.5. In both cases, \\ = 2.
(Reproduced with permission from Khakhar, Franjione. and Ottino (1987).)
(a) 0 = 2.0
t l ~ cri1</iotl olrtsir/ii t/lcl i.s/rrr~r/.s.In this case. the peak is more spread out
and is shifted towards higher residence times. The results here show
some of the limitations of exit age distributions for capturing the complex
structure of the mixing in continuous flows. It is clear that residence
time distributions alone are insufficient for an analysis of the mixing.
0 10 20 30 40 50
time
(a I )
242 Mixing and chrros in three-tlimensiontrl und open Jows
Problem 8.2.1 *
Consider a mixer of field dimensions (L, R ) . What is the optimal value of
1,' to maximize the mixing efficiency'? Attempt a n analysis.
Figure 8.3.1. Schematic of the eccentric helical annular mixer. The length of
the mixer is L. In the case of the ligure the cross flow corresponds to 7.4.2(c).
Mixing in the eccentric heliccil utztzul~rmixer 247
work is necessary. I t is significant that even though the axial flow always
moves forward, the streaklines can 'go backward' since they can wander
into regions of low axial velocity, whereas other parts of the streaklines
can bulge forward. Figure 8.3.3(u) shows an instantaneous snapshot of a
lateral projection of a streakline injected in a chaotic region, as well as
a view from the end of the mixer. It is significant to notice that in the
*end view' the streaklines crosses itself at various points, and that these
streaklines would be the ones observed for a system with plug flow, or
equivalently the journal bearing flow (see also Example 8.5.1). Figure
8.3.3(b) shows a similar computation but now the streakline is injected
in a regular region. (The situation however is considerably more complex;
the 'regular regions' are not static but move in a time periodic fashion;
thus the dye injection apparatus can inject material in both regular and
chaotic regions depending upon the injection time.)
Various other diagnostic tests used for the partitioned-pipe mixer in t l l c
previous section are valid also for the EHAM. In particular, it is possihlc
to inject a dense grid of passive particles or vectors with various
orientations and to compute contours of constant age, stretching, efficicncv.
and so forth. Multiple peaks in the exit distribution are possible in t h i s
mixer. Figure 8.3.4 shows an instantaneous picture of the axial concen-
tration distribution created by the injection of a square array of particles
at the entrance of the mixer (the concentration has been averaged with
respect to the cross section). I t thus appears that the peaks present in the
partitioned-pipe mixer are not due to any artificiality present in the miser
and that similar behavior is indeed possible in other systems. Contours of
constant stretch reveal that in this case, as opposed to the partitioned-pipe
0 5 10 15 20
axial distance
M i x i n g ~itlrlrli.sper.sion in u ,furrower1 chuntlel 25 1
middle of it. The same flow was studied by Ralph (1986) for Reynolds
up to 300 and a range of Strouhal numbers of to 1.
As we have seen in previous examplcs (notably, Section 7.4) the
present only part of the picture insofar as mixing is concerned.
Nevertheless, a superficial analysis indicates that the flow will be able to
well since it consists of two interacting co-rotating vortices as the
blinking vortex flow (Section 7.3) and the cavity flows (Section 7.5). In a
rough sense we might interpret Sr as 1/p, if the circulation of the vortex
is taken to be UII.Also in the symmetric case we expect the mixing to be
rather poor since it is obvious that the centerline is an invariant surface
ofthe flow, and prevents interaction between the regions .u, > Oand .u, < 0.
As we have seen in previous examples, two-dimensional time-periodic
flows are far richer in structure than what might be revealed by a cursory
analysis and an in-depth analysis requires an entire arsenal of tools. It is
apparent that this flow contains periodic points and is capable of
homoclinic and heteroclinic behavior. However, an analysis on the level
of that presented for the TW and BV flows is tremendously complicated
and only a few aspects of the program of Table 7.1 can be completed.
Here we present the results of Sobey for macroscopic dispersion of tracer
Note that, in general, I., and I.? arc periodic but the period need not bc
equal to w - I . Also, the identification of periodic points of the velocity
field is far from trivial.
Figures 8.4.2 4 show the dispersion, and initial condition, of a plug of
3,000 passive particles for Sr = 0.07 and KC,= 10, 75, and 100, t r f r c v . o ~ ~
of' rlic. ,/low. Qui~litatively,it appears that there is a n optimum /<,,
(,j.(,/(j
for a given S r . Figures 8.4.5 6 show similar processes occurring in a
chi1nncl with n phase angle (1) = i~ and equal amplitudes and another case
corresponding to one wall with amplitudc k' = 0. A measure of the ;i\ial
dispersion is given by the variance of the axial particle positions ( Y , I,,
i = I . . . .. N.
where ( x , ) is given by
Problem 8.4.1
Consider that the particles are not material particles but rather actual
tracer particles. Discuss the effect of the particle size on the process, i.e.,
account for Brownian diffusion and/or inertial effects.
F(w)=
Sr. sin(.xl)l,,,,,,,,, sin(wr) dl.
The function F(w) is plotted in Figure 8.5.2. Since M(t,) represents the
distance between the perturbed manifolds, we expect that an extreme in
F(w) should maximize the 'extent of chaos'. In this particular case, the
optimal frequency, (11, appears to be in the neighborhood of 0.3. I t is
important to check whether or not such a prediction is substantiated by
Figure 8.4.5. Similar initial conditions as Figures 8.4.2-4, but in a channel with
a phase angle 4 = n and equal amplitudes. Dispersion after one cycle of the
flow. The frames correspond to: ( 1 1 ) (01 = 0. ( h ) wt = 0.25, ((,) (ot = 0.5, ( d )
wr = 0.6, ( r ) cut = 0.75, and ( / ) (ot = l .O. (Reproduced with permission from
Sobey, 1985.)
other techniques, such as Poincare sections. A Poincare section for tlli,
problem consists of recording intersections of initially designated Iluid
particles with the r; x i plane every 2n time units. Also, taking advantage
of the spatial periodicity of the flow in the xi-direction, we record tllc
intersections making .xi mod 2n. Visual inspection of the results shown 111
Figure 8.5.3 indicates that, indeed, chaos is maximized for (,) 2 0.3
(substantially higher values of the frequency, ( 1 ) = 4, indicate a reduction
in the degree of chaos and mixing).
Since streaklines would be the tool of choice in experimental studies. it is
important toexamine how they look in the perturbed case. Experimentall!.
the streaklines would be injected with respect to the fixed frame, but
c~mputationallyit is possible also to investigate injection in a frame moving
with the cat eyes (Figure 8.5.4). A comparison between the streaklines
and the corresponding Poincare section reveals that the streakline injected
within the chaotic region undergoes significant stretching and folding
whereas the streaklines injected within the 'cat eyes' and outside 'cat eyes'
undergo little stretching. Figure 8.5.5 shows the stable and unstable
manifolds associated with the hyperbolic points.
The streakline injected with respect to the fixed frame reveals also
significant stretching and folding (Figure 8.5.6). However, in this case, as
opposed to the Hama flow of Example 2.5.1, the streaklines serve as a
good indicator of the position of the vortices (there are none in the Hama
flow). It is important to notice that given the time-periodicity of the flow,
it is possible to simplify substantially the problem of streakline tracking
Figure8.5.3. Poincaresections for the perturbed cat eyes flow ( A = 0.8. i:= 0.1 ).
The values of cv are as follows: ( u )0. I , ( h )0.2, (c) 0.3, (d)0.4. ( r )0.5, and ( f ) 0.6.
The maps consist of eight initial conditions and 100 iterations.
(a) (b)
M i x i n g in the K e l v i n cut eyes Jow
with solution
x, = ( . u , ) cos
~ t - ( . Y ~ sin ) ~ t + [to/(l -to2)](costot - cos t )
x 2 = ( . Y , ) ~sin t +
( . Y , ) , cos t + [1/(1 - w2)](sin tot - t o sin t ) ,
E = 0.1,
-
Figure 8.5.5. Stable and unstable manifolds corresponding to the case A 0.8,
o = 0.3, for various injection times (the 'most chaotic system', see
Figure 8.5.3(c)).
Figure 8.5.6. Snapshot of streaklines injected in a fixed frame in the cat eyes
flow. The time of injection corresponds to two periods (spanning two cat
'eyes'). The position of the vortices is indicated by crosses. The parameters
are: A = 0 . 8 , ~ = 0 . l ,tu=O.l
Flow near wulls 26 1
where the variables associated with the (n + 1 ) order tensor Aijk,,, are of
with n = (1 + b + c, where u , b, c are integers or
the form (X,)"(.Y,)~(.K,)'
zero (c = 0 for a two-dimensional flow).'' It is clear that the tensors Aijk,,,
are symmetric in all indices but the first one due to the fact that the order
of differentiation is immaterial. These tensors constitute the unknowns
and are to be found by forcing the series to satisfy the continuity and
Navier-Stokes equations as well as the boundary conditions of the
problem in question. The coefficients may have time dependence or not,
depending on whether the flow is steady or time varying (due to some
external perturbation).
The number of independent coefficients, N,, generated by an Nth order
expansion of a three-dimensional flow field is given by
Table 8.6.1
Two-dimensional Three-dimension:il
-
A few tabulated values, taken from the work of Perry and Chong (1986)are
given in Table 8.6.1. An examination of Table 8.6.1 shows that, in general,
several boundary conditions should be specified in order to obtain the
numerical value of the remaining coefficients. For example, for n = 5 We
need a number of conditions to obtain the unspecified 42 - (15 + 6) = 11
coefficients. Boundary conditions such as no-slip at the wall and a
specification of the surface vorticity are commonly used and generate a
Flow rlrtrr wcills 263
Since the vorticity changes sign at &.u, there are two separation point,,
We also specify the angle of separation and reattachment of the streamlines
O . The relationship between the angle at the point of separation, measurcd
in the clockwise direction from the .:all, and the derivatives of pressure
and vorticity evaluated at u2 = 0 x1 = f I (separation point), is given by
tan 0 = 3p(i(o,/?.~,)/(?p/?xl),
which specifies a relationship between the angle, the separation points,
and the gradient of pressure. However, these conditions are not sufficient
to specify all the coefficients. In order to close the equations the location
of a critical point (for example an elliptic point) also must be specified.
In this case we can specify the location of an elliptic point; the second
critical point (i.e.,the hyperbolic point in Figure 8.6.I ( u ) )arises as a result
of solving the problem with the boundary conditions as stated above.
Thus, in this problem, a fifth order expansion, the total number of
unknowns is 42, and according to Table 8.6.1, the number of independent
relations between coefficients provided by the Navier--Stokes equation is
6, whereas the number of relations provided by the continuity equation
is 15. Therefore the number of coefficients to be determined is 2 1 . I t can
be shown that for a two-dimensional flow the no-slip condition provides
2(N + 1 ) relations, the specification of the surface vorticity N relations,
and the angle relations and the location of the elliptic point (x;, x;), two
conditions each. For N = 5, this provides an additional 21 relations, from
which all coefficients can be determined."
In this particular example the length scale L is selected as x,, and thc
time scale T as ~ / [ K ( x , ) ~ ]The
. resultant velocity field is given by
+
v1 = A12x2+ A12,x: + 3All12x:x2 3Al122xlx: A,,,,x: +
+
+6All122.~:.~: A12222.~:+ I O A l l l 1 2 2 ~ ~ . ~ :
+ 5Ai 12222x1.~;+ A 1 2 2 2 2 2 ~5 2
q = -3All12xlx:- A l 1 2 2 x ~ - 4 A l l 1 2 2 x l x1~ - 0 ~2 3 ~ ~ ~
where the coefficients, and relationships between coefficients, is given by
A,,= -1
A i l , , = 113
A 1 2 2 2 = -213
A l l 1 2 2 =(3-A1222)/6
= I / ( x ; ) ~ + 2 / ( 3 ~ ; )+ (1/IO)RAll22x; -A122/(x;12
A111122=-(3/10)A1122
A112222 = -~1122/(xS)~
A 1 2 2 2 2 2 = -(lllO)RA, 1 2 2
and where the values of A , ,, and A , , , are given by the solution of the
system
Problem 8.6.3
Show that the equations describing the time evolution of the coefficients
anc a r e
Ai22 * ^1222 volume contracting.
Problem 8.6.4
Show that the angle of separation and attachment depends only on the
viscous terms, and that the non-linear and time dependent terms are
identically zero for all expansion orders.
all originate from the same initial condition); all the isolated point,
originate also from one initial condition (this behavior was called
'semi-ergodic' by Henon). The behavior of A = B = C = I is similar except
that the 'semi-crgodic' region occupies more space.
The same flow was analyzed, in a more complete study, by Dombre cJt (11.
(1986). I t is easy to check that the velocity field has eight fixed points
corresponding to
- C sin u 2 = B cos s, = + [(B' + C' - AL)/2]'
-B sin u , = A cos .u, = + [ ( A 2 + B' - C2)/2]'/'
- A sin u , = C cos .u2 = + [(C' + A' - B')/2]'i'.
The condition that x,, .uz, and .u,, be real is that a triangle be formed
with sides A ' , B 2 , and C ' . The condition dxldt = 0 implies o = 0 and
therefore at the fixed point Vv is symmetric (since Q = 0 ) . I t follows that
the eigenvalues of Vv are real and the flow is hyperbolic (the sum of thc
eigenvalues is zero since V . v = 0 ) . Furthermore they are all different except
Figure X 7 1 . Poincark section in the ABC flow for the case A = 3", B = 2'".
C'= I.
in the case A = B = C = 0. (For C' = 0 the equations are integrable and
the streamlines belong to the surfaces B sin .Y, + A cos .Y, = const.).
The stable and unstable manifolds of the fixed points (either two-
dimensional surfaces or lines) have either to join smoothly or intersect
transversally and complex possibilities of the type depicted in Section 5.10
are possible. Such a scenario is indicated by Dombre et trl.
Extensive numerical simulations indicate that the system displays
chaos." I f the flow were a truly effective mixing flow every filament of
length i, placed in any orientation whatever in the chaotic regions will
increase its length exponentially. However, even though this is suggested
by the numerical calculations, i t is not true in general. Given the character
of the flow we know that if the filament of length j., is placed initially
coinciding with a vortex line with vorticity wo its length, when it moves
to a region of vorticity o is given by
i = (1o1/1o0l)i.,.
However, o is bounded (in this case its value is exactly equal to v ) and
we have to conclude that the stretching i. cunnot incrrclse esponentiully as
is expected in the chaotic region of a two-dimensional time-periodic flow
no matter what the initial orientation. Note however, that the stretching
history can be extremely complicated, and that filaments placed in any
position, most likely, will be stretched exponentially.
Bibliography
Most of the work reported in this chapter is currently in progress and
the presentation should be regarded as a sketch of the current state of
the field and as an outline of possibilities. Some of the original material
on the partitioned-pipe mixer appeared in the Ph.D. thesis by Khakhar
(1986). A more thorough presentation appears in Khakhar, Franjione,
and Ottino (1987). T h e work is continued and considerably extended
in the Ph.D. thesis of Franjione. The treatment of Section 8.4 follows
closely the presentation by Franjione and Ottino (1987, abstract only).
The material in Section 8.4 follows entirely the article by Sobey (1985).
Additional material can be found in Ralph (1986). The treatment of
the Kelvin cat eyes flows (Section 8.5) is based on the Ph.D. thesis
of Danielson. The treatment of flows near walls is based on the work
by Perry et al. (see references in Section 8.6) and recent work by Daniel-
Son. T h e material in Section 8.7 is based on the short article by Henon
(1966) and Dombre et a[. (1986) with some interjections of our own.
A relevant paper for the analysis of three-dimensional flows is Crawford
and Omohundro (1984) since it e m p h a s i ~ e sdynamical aspects which
are not visible in a PoincarC map.
Notes
(a)
motion
concentration profile
along broken line
maximum direction
of contraction
The simplest theories of diffusion arc bnscd on Fick's law, i.c.. rllc
relative diffi~sionvelocities are given by
u,=~ - (0~7Ills VW,~. (9.1.3)
Since us transforms as ul, = Q - u , and V transforms as V' = Q - V we concludc
that the diffusion equation has the same form in any moving framc 1;'
provided that the velocity field is referred to the frame F'. An identical
argument applies to the energy equation.' In particular we can apply
Equation (9.I . I ) to a frame attached to a material particle X, such t hot.
at small distances from X,,, in frame F', we have
Example 9.1.1
The equation
where the D,,s are the diagonal components of D. which in general are
functions of time. but are considered constant here. A heat spot is released
at the origin, and according to the value of the determinant of D, it is
converted into a filament (det D > 0 ) or into a flat disk (det D < 0). We
take tr D = 0 and consider D l , 3 DZ23 D33; i.e., D l , > 0 and D,, < 0. The
temperature evolves as
and
T,,,(t) = A / L I , u ~ L I ~ ,
where A is a constant determined by the initial conditions. Thus, the
isotherms are ellipsoids, and for long times ( D l ,t >> I ) , T,,,(t) decays
exponentially in time. Batchelor (1959)considered a similar problem, but
with the initial condition
T = A, sin(l-x )
where I = (I,, 12, I,) is a constant vector and x = (s,,r 2 , s,).In this case
the solution is of the form
T(x, t ) = A(t) sin(m(t).x)
where m ( t ) is giver, by
dm/dt = -D-m,
and therefore (assuming 1, # 0 ) we obtain
m2= + tn: + ni: + 15 exp(-2D3,t), as t -+ rd
and
D ~ ~ ) x),
T-+ A , , ( ( Y + ~ / ~sin(m(t) ast+m
which shows that the normal to surfaces of constant T (i.e., with normal
V T ) approaches the direction of the greatest rate of contraction. Note
however, that this local result is valid only for linear velocity fields.'"
Example 9.1 -2
. O n many occasions it is sufficient to characterize the mixing in average
or structured terms, e.g., profile of striation thickness across a
278 Epilogue
and
Similarly, we define pn(X,t ) as the mass of tracer per unit area of the
surface at the location X and with orientation n. We have
and
Problem 9.1.1
Show that material planes remain flat in the flow field v,,, =x-(VV)~.
Problem 9.1.2
Confirm the invariance of the convective-diffusion equation.
Problem 9.1.3
Show that the distance 6(t) between two small, nearby, parallel planes
with normal n varies as a function of time as (Ottino, Ranz, and Macosko-
1979)
I DS(t)
= D: nn.
6(t) Df
Trunsport crt striation thickne.ss .scule.s 279
'warped time'
Figure 9.1.2. U'i-premixed reactor. The reactants A and B enter the system and
are mixed by complex motions. The microflow element SXmexits the system
after a time t,,,.
where rl is the area stretch associated with the microflow clement .' ;ll,d
r,. is a characteristic time. The usual choices are I,. = I,,,, (diffusional I I I ~ , ~ )
and t,. = r, (reaction time). A dimensionless distance is defined as
f = s/s(t), spirc-c1S L - I I ~~L ~I I S Part~ tho i ~ t ~ t r ~ ~ t i r ~I ' L~I t~cI I~Oof'
o ~l/,c,
rs
(r(t)) = :1; r ( t l )dtr, tinw trcleragrd oul~reof the stretc.hing ,firrlc.r ion
and striation ~hickness based scale <, the apparent reaction rate is
multiplied by a factor 11-q~) due to the local stretching."
The usual initial conditions correspond to complete segregation of the
streams: the physical situation might correspond to streams being fed to
the reactor (in this case the interface is a streaksurface) or two segregated
volumes initially placed in a batch reactor. As initial condition we establish
that the streams are unmixed. i.e.,
T or 3 0, c?C,/?<= 0, for = 0, I . <
where Hi(() is a unit step function discontinuous at ( = 4, where 4 is the
volumetric fraction of the one-stream (one,of the streams is arbitrarily
designated as 'one'). l5 T h e boundary conditions, in the periodic case, are
obtained from symmetry considerations:
7 or r 2 0, (?c,/ay= 0, for ( = 0, I .
Consider now the value of the initial striation thickness, so. In
establishing this value we face two competing effects: We might argue
that if so is too large, and curvature effects are important, the lamellar
structure assumption is probably not quite valid and consequently the
model a poor approximation. O n the other hand, if we wait longer and
select a smaller so, we might miss some of the diffusion-reaction process
and not be able to estimate the initial conditions. However, if so is large,
as for example during the first stages of mixing in a shear flow (see for
example, the Hama flow of Example 2.5.1 ), the interfaces are essentially
non-interacting and therefore it really does not matter that the lamellar
structure has not been established.
Problem 9.1.5
Eliminate the convective term (v = x . D ) in the equation
This leaves several choices for the selection of A([) (Chella, 1984).
9.1.2. Regimes
Equations (9.1.6a and b) form a system of coupled, non-linear partial
differential equations, which can be solved, in general, only by numerical
Table 9.1
means. The limit cases. 'fast'or 'slow' reactions, are more tractable than the
general case. The different regimes of reaction are shown. schematically.
in Table 9.1 . I .
It is evident that the presence of at least three major time scales produces
two dimensionless ratios (Damkohler numbers) which characterize the
processes occurring at small scales:
DtrjL'= tR/tb 1 6
--
Dull tD/tR.
Note that the characterization of a reaction as fast or slow depends, in
gneral, on t,, t,, and t, (see Figure 9.1.3).
Slow reactions
A reaction is called 'slow' if t,/tR << 1 during the course of the reaction,
except possibly during the first stages of mixing (this is a conservative
criterion since t,,, is defined with respect to the initial segregation - the
loosely defined so). Actually, the diffusion time-scale varies in time, since
the striation thickness is decreasing in time. Consider a simple example
corresponding to exponential thinning. Initially, the diffusional time-scale,
t , = s ( t ) ' / ~ ,might be larger than the reaction time-scale, i.e., s(O)'/D > t,,
but after some time, we might achieve t, z t,, since s(t) decays as
s(0) exp(-xt), and quite rapidly afterwards we might be in a regime such
that t, << t,. From this time onwards, intrinsic kinetics controls and local
mixing has no influence on the course of the reaction. Under these
conditions the mass balance for species-s reduces to:
dC,7/dt= /j,,R,. (9.1.7)
This limit corresponds to residence time theory (see Chapter 1).
case corresponding to
'fast reaction'
case corresponding
0
L
time (age)
/to
- 'slow reaction'
284 Epilogue
Vtry f i s t reactions
A reaction is called 'very fast' ~f t,)/t, >> 1 during the course of the reactloll,
In such a case the reacting species cannot co-ex~stand the reactlng 1011,
reduces to a plane. The rate of diffusion to this plane controls the ovcl-;~ll
reaction rate, and we lose the influence of the intrinsic kinetics. ' I t l c
controlling factor is the rate of stretching and the location of the reaction
Lone, which in the simplest case does not move and acts as a m a t ~ r i ; ~ l
surface." In addition to the initial and boundary cond~tions,the f l i i \ of
reactants to the plane must satisfy the overall reaction stoichiometry. Hcrc
the variable T, with I, selected as I,,,, contains all the effects of the local
convect~ve mixing (through the stretching function r(X. t ) ) and tllc
problem reduces to
Thus, the solution of (9.1.8) for a particular reaction scheme allows thc
superpos~tionof any stretching history (cho~ceof r(X. I ) ) . This is thc b a x
idea behind the use of fast reactions as tracers of the fluid motion.lx
Exumple 9.1.3
Figure E9.1.3 represents the effect of the local flow on a series-parallel
reaction (A + B -t P), with a specific rate of reaction k2, R + B -t S. ~ i i t h
a speclfic rate of reaction k,). The system is governed by the equatiollr
premixed feed
/
/ elongational flow /
2 4 6 8 10
time (s)
286 Epilogue
Problem 9.1.7
As a crude indicator, assume that reactions are diffusion controlled 11
tI,/tR > lo4 and kinetically controlled if !,It, < 1 0 ' . Calculate the lengt/,
scales necessary for these two regimes for the following second order
reactions: HCI + HONa. k , 2 lo", C O , + H O N a , A , -, 1.3 x lo-'.
+
H C O O C H , H O N a , k2 = 4.7 x lo1( k 2in l/mol. s and at 30°C).
ABABABABABABABABABAB
x =0 x= 1
290 Epilogue
Use this result to prove that, at the point x, and at any instant t,
9.2.2. Illustrations
The following illustrations serve to reinforce some of the previous concepts
and exemplify the limits of current computational efforts and the ability
to resolve striation thickness scales.
Example 9.2.2
Figure E9.2.2 shows what happens during the wrapping of an interface
due to the first stages of instability in a shear flow in the direction .Y. for
times such that the layer is constrained to be two-dimensional (Corcos
and Sherman, 1984). The initial velocity and concentration profiles are
u, = U erf[n"2y/(26)], p = - 0.5Ap[(nPr)1'2y/(2(5)1
i.e., the streamwise velocity changes from + U to - U and the scalar from
+Ap to -Ap. The value of 6 is such that 6 = U[((7uX/dy),,,]- ' : the
Reynolds number is defined as Re = U6/v and the Schmidt number is \','D.
The time-scale is made dimensionless with respect to d/U where d is the
length scale of the initial disturbance.
Figure E9.2.2(a) shows the wrapping of the interface along with the
instantaneous picture of the streamlines, and iso-concentration curvcs of
Cotnplications and illustrations 29 1
Figure E9.2.2(a).Wrapping of the interface between two fluids due to the first
stages of instability in a shear flow: ( a l ) location of the interface at
dimensionless times 0.5, l .O, 1.5, and 2.0, with a = (2nii)ld = 0.43. Re = 100:
((12)iso-concentration curves of a passive scalar with Sc = I . a = 0.4, Re = 100,
at times 1.5 and 2.0. (Reproduced with permission from Corcos and Sherman
(1984).)
292 Epilogue
the diffusive passive scalar. Note that the stretch of the interface is fiiil-lv
small (O(10))and that the lamellar structure assumption is a reasonable
one except near the center of the 'cat eye'" (something more complica[cd
occurs during vortex pairing). The details of the stretching history of [Ilc
elements are highly complicated. With the exception of elements near the
hyperbolic points, all other points experience a quasi-periodic s t r e t c h ~ n ~
history where a ( X , t ) becomes negative. Corcos and Sherman (1984) noted.
as might have been expected, that the stretching history is highly sensitiic
to the initial conditions near the hyperbolic point.
Example 9.2.3
As we have seen in Section 9.1.3, the accurate computation of material
interfaces is important in problems involving fast reactions producing t h i n
reaction zones. The problem also has relevance to computations such 2 s
those described above and the direct computations given below. Order
of magnitude calculations indicate formidable problems due to stori~gc
requirements. Computationally, the material interface is composed of a
large number of points connected by small linear segments. However, duc
to the flow the interface stretches and folds, and if the distance between
points becomes too large, the linear segments connecting adjacent polnts
could cross, which is physically inadmissible. In order to maintain
consistent resolution, the numbel of particles should be increased in such
a way as to be able to resolve the striation thickness accurately. Obviously,
such a calculation is a precondition for the resolution of the concentration
order one. The simplest possible case, which is a lower bound for nio1.c:
complicated situations, corresponds to a two-dimensional (deterministic,
time-periodic) mapping in a 'globally chaotic' regime where the length of
a line increases, on the average, as Lo exp(crk),where k refers to the number
of mappings and a is a suitable Liapunov exponent (in the case of
integration of the Eulerian velocity field, we have estimated that tllc
number of operations for each dimension, F, is of the order 10 to 100,
and that each period requires 10' time steps with t of order 3). An estimate
of the number of floating point operations per mapping, taking into account
the loss of precision due to growth of errors, for an accuracy of O.OI'!;,.
yields the results shown in Table 9.2 (Franjione and Ottino, 1987b). I t
should be noted that the bulk of the computational effort is not due
to growth of errors. For example for the case of a striation reduction
of relaxing the accuracy to 1% and 5 % , reduces the storage
requirements by only 1.5 and 2 Gigabytes respectively.
Example 9.2.4
There have been a few direct simulations incorporating chemical reactions.
Figure E9.2.2(h). Tracking of the interface between two fluids being mixed.
The distance between material points is 6, the striation thickness is s. Material
points are used to represent the blob itself or the interface between the blob
and the surrounding fluid.
Complictrtions trntl i1lu.strtrtio11.s 295
The article by Riley et ul. (1986) presents one of the most complete. They
considered a binary chemical reaction with no heat release (i.e., the
reaction does not alter the fluid mechanics) in a temporully growing mixing
layer, which is simpler to analyze, and contains some of the same physics
as a spatiully developing mixing layer (i.e., their geometry is similar to that
in Figure E9.2.2(u)). The initial condition was also qualitatively similar to
that of Corcos and Sherman (1984). Riley, Metcalfe, and Orzag (1986)
used a pseudo spectral method using 323 and 643 grids; typical runs
involved between 500 and I000 time steps and consumed one to two hours
of Cray 1 computer time. The Schmidt number was kept of order one so
that the resolution requirements for both velocity and concentration were
roughly the same, the Reynolds number based on the Taylor microscale
was approximately 50, and the diffusivities of all species equal. Figure
E9.2.2(c) shows snapshots of iso-vorticity plots and iso-concentration
contours, the iso-concentration contours being simpler and more lamellar-
like (compare with the cat eyes flow). Note that the limit of resolution is
set by the computational grid. Striations created below this scale will be
averaged out and the lamellar structure, if present, will be lost. If the
reactions are slow the loss of resolution is not serious. However, depending
on time-scales, the reaction might be diffusion controlled at these length
scales and in this case the resolution problems become serious (see previous
example). Similar treatments have included compressibility, for low Mach
numbers, and also heat release (McMurtry et al., 1986).
Problem 9.2.2
A possible way of connecting the results for microflow elements with
Eulerian measurements is by computing the spectrum of concentration of
a product or a passive scalar. Lundgren (1985) used an axisymmetric
microflow element to obtain the product spectrum corresponding to small
296 Epilogue
each of which consisted of 12g3 points and in each point they recorded
the value of o,, rlo,/dx,, and Vc). They found that 23% of the points
analyzed were such that /j < 0 and the eigenvalues were ordered approxi-
mately as IfiI x ($4 and H x 1 .2IYI (a sphere becomes a filament). In the
remaining 77% of the cases fi > 0 and /j' -,ly1/3 and H x (3/4)Iyl (a sphere
becomes a disk). Moreover, in this case Vc was found to be nearly parallel
to the maximum direction of compression of the flow ( 7 ) . Attempt a
rationalization of these findings in terms of the two-dimensional results of
Sections 7.2 (tendril-whorl) and 7.3 (blinking vortex). Note that if the
efficiency decreases, the flow becomes less extensional. Compare the results
with the classical findings of Batchelor and T o w n ~ e n d Investigate
.~~ also
K < 1, however, Ca,, increased with p, for p > I . For the flow with the
most vorticity considered, there was a limiting value ( p z 14.7, based on
theory), beyond which breakup was not possible. For high and intermediate
viscosity ratios (p > 0.05),the data agreed well with the small deformation
theory of Barthks-Biesel and Acrivos (1973), and for low viscosity ratios
( p < 0.01) with an empirically modified version of the result of Hinch and
Acrivos (1979) for hyperbolic extensional flow. In a related study, Stone,
Bentley, and Leal (1986) focused on the interfacial driven motion which
Figure 9.3.1. ( a )Drop being deformed in a flow; the flow in the neighborhood
of the drop is Lax (the center of coordinates corresponds to the center of mass
of the drop). (b) Deformation measures for nearly spherical drops, D =
( I - b)l(l+ b ) , and elongated drops, [la.
near spherical
drop
(viscous)
(b)
pointed drop
(non-viscous)
302 Epilogue
occurs after the flow is stopped abruptly, for various flows with K = I .().
0.8, 0.6, 0.4, and 0.2 and viscosity ratios in the range 0.01 < p < 12. The!
found that for modest extensions d r o p breakup does not occur with the
flow but may occur as the result of a new mechanism, which they called
'end pinching'. In this mechanism the ends of the elongated d r o p become
spherical, forming a dumbbell-like shape; the ends then proceed to pinch
off leaving a cylindrical thread of fluid which can again repeat the proces\
(1972. 1973) studied the same problem, within a somewhat more rigoro~1.s
framework, and apparently without knowledge of Taylor's work. and
found that there could be several possible steady-state solutions for the
drop shape. but based on physical grounds he selected the one in which
the drop was stretched the least as the only realistic solution (this was
also the solution found by Taylor (1964)). Several years later. Acrivos
and Lo (1978) obtained the same conclusion by means of a stability
analysis. Hinch and Acrivos (1980) studied the case of a hyperbolic
two-dimensional flow as perturbation of the axisymmetric case. They
found that even though the cross-section was not circular, the area in the
axisymmetric and two-dimensional cascs were approximately the same
(Cu, = 0 . 1 4 5 p 1 "). Thus, it appears that only the area and not the shape
is important in the breakup process and suggests the possibility of using
the approximation of circular cross-sectional area for other flows.
The analysis corresponding to simple shcar flow is substantially more
complicated than the case of axisymmetric flow. Hinch and Acrivos (1980)
sti~diedthe case in which the d r o p is almost sligned with the streamlines
in a shear flow by assuming that the d r o p cross section remains circular
due to the slenderness. They found that the d r o p axis bends slightly and
that the elongated droplet attains a sigmoidal shape. but that the deviation
from a straight line was small for most cases of interest. A steady-state
analysis of the d r o p shape indicated that steady shapes were possible for
all values of Crr. However, a time dependent simulation indicated that the
drops were unstable beyond Crr, = 0 . 0 5 4 1 p 2 ". In numerical calculations
involvingjump in the strain rate they found that drops 'broke' (interpreted
as failure of the numerical scheme) at the lower strain rate.'"
Extensions of these analyses yield models which can be incorporated in
the context of the flows of Chapters 7 and 8. For example, for the case
p << 1 and Ca >> I , the dynamics of a nearly axisymmetric drop with pointed
ends, characterized by a n orientation m ( m = 1 ) and a length I(t), is given
by (Khakhar and Ottino, 1 9 8 6 ~ )
+
D In I(t)/Dt = L: mm - (a/2(5)'l'/~,u)[(/(t)/rr)'!~/(l 0.8~(l(t)/~)~)]
(9.3.5a)
DmlDr = ( G D + R)' em - ( G D : mm)m (9.3.5b)
where G(t) = ( I + 1 2 . 5 ~ ~ I ( r ) ~-) /2.5u3/l(t)".
(l The underlined term in
-
(9.3.5a) acts as a resistance to the deformation [contrast with (9.3.la,b)].
A very long drop, (I(t)/u)+ r,, G I , rotates and stretches as a passive
element since the resistance to stretching is unable to counterbalance the
effect of the outer flow. Note also that since G > 1 the droplet 'feels' a
flow which is slightly more extensional than the actual flow.
Other researchers have sought a unification of results and developed a
format which is useful for both low and high viscosity drops. Olbricht,
306 Epilog uc.
Example 9.3.1
In the context of the model of Equation (9.3.5a,b) a drop is said to brcak
when it undergoes infinite extension and surface tension forces are unable
to balance the viscous stresses. Consider breakup in flows with L:mm
constant in time (for example, an orthogonal stagnation flow with thc
drop axis initially coincident with the maximum direction of stretching).
Rearranging Equation (9.3.5a), and defining a characteristic length as
tr/pl'-', we obtain
L: mrn/(D: D)"' = efficiency = (1/2(5)'i2E)[(1,"2/(l+ 0.81;')]
where I, denotes the steady-state length and E = p'lhCu. A graphical
interpretation of the roots I, is given in Figure E9.3.1. The horizontal line
represents the asymptotic value of the efficiency (i.e., corresponding to
DrnlDt = 0), which in three-dimensions is (2/3)'12, and the value of the
resistance is a function of the drop length for various values of the
dimensionless strain rate E. For E < E , there are two steady states: one
stable and the other unstable. For E > E,. there are no steady states
and the drop extends indefinitely (see Khakhar and Ottino, 1 9 8 6 ~ ) .
Pvohlem 9.3.1
Estimate the order of magnitude of the additional body forces that arise
due to the translation and rotation of the moving frame. How significant
Lire they to the analysis based on Equation (9.3.3)?
Problem 9.3.2
Stone. Hentley, and Leal (1986) investigatcd the time evolution of the
d r o p Icngth, L(t)ltr, for conditions slightly above the critical capillary
number in the flow field of Figure P2.5.3. Thcy plotted their results using
a dimensionless time, ( ; K 1 ' r . They found that curves of L(r)/u versus
G K ' ' r for a wide range of K frill approximately on the same curve. Justify.
Problem 9.3.3
Cast the results of Hinch and Acrivos (1980) and Huckmaster (1973) in
the form of Equations (9.3.5a.b).
11
resistance to stretching
b
stable unstable drop length
308 Epilogue
Reynolds number flow, the other is high Reynolds flow there seems to
be agreement that the explanation of the phenomena in the case of'
polymers is due to flow-induced conformational changes and that much
benefit can be obtained by focusing o n the behavior of a single micro-
structure in a linear, possibly time dependent. flow. Macromolecular
stretching can modify the surrounding flow: in turbulence the smullc\r
eddies might be suppressed; in porous media the hydrodynamic interactionh
might result in non-Newtonian effects.
Even though considerable experimental and theoretical work has bccn
done in both areas, the suggestion here is that microstructures in chaotic
flows might have a bearing o n these kinds of problems. However. since
o u r primary focus is o n a few c o m m o n aspects of both problems, our
review of the literature is rather superficial." A review of experimental
studies of flow induced conformation studies is given by Leal (1984). M o h ~
of the work focusing o n conformational changes has been o n well
controlled linear flows. for example, opposed jets, two-roll mill, four-roll
mill. six-roll mill (see. e.g., Berry and MacKley, 1977); only a few studit\
have focused o n spatially periodic flows (Nollert and Olbricht, 19851.
There should be n o question that such flows contain some of the r e l e ~ a n ~
aspects of both problems (i.e., porous media, turbulent flows) and thar
the conceptual and experimental advantages are significant. T h e flows arc
relatively simple to characterize completely and measurement of conform;^-
tional changes by light scattering are r e l a t ~ ~ e lsimple
y t o perform ('I\
opposed t o measurements in a n actual turbulent flow o r the interior of a
real porous media). What is missing, however is the unsretrtij nature of
the flow experienced by microstructures in actual turbulent flows ancl
topologically disordered porous media. O u r thesis here is that there i \
some opportunity for a n unexplored mid-ground.
O n e such possible mid-ground is t o focus o n the behavior of micro-
structures in two-dimensional chaotic flows (either experimentnlly or
theoretically). Carefully designed low Reynolds number chaotic flo\\\
provide several key ingredients missing in the studies listed above. Let L I \
consider a few
( i ) Chaotic flows provide unsteady (chaotic) Lagrangian histories. The
flows are capable of generating regions where the stretching is mild
and time-periodic and chaotic regions where. o l e r time. the stretching
of material elements is nearly exponential (efficient flows)." In the
chaotic regions the microstructure is neLer in steady extension: the
maximum directions of stretching constantly change.
( i i ) T h e unsteadiness of the maximum directions of stretching, prcscnl
in turbulent flows, is also present in chaotic two-dimensional flows,
such as the journal bearing flow (Chapter 8); the randomness
associated with porous media, usually ascribed to the porous media
itself, is present in the partitioned-pipe mixer, which is spatially
periodic, or in the eccentric helical annular mixer, which is time
periodic.
(iii) Chaotic flows allow the investigation of spatial co-operative effects
(e.g., percolation of chaotic regions, percolation of significantly
stretched regions, see Figure 7.4.1 1 , color plates).
(iv) Practical mixing problems start from an unmixed state; for example,
polymer is injected into a pipe line and the drag reducing effects take
place during mixing of the polymer with the flow. This effect cannot
be incorporated into the conventional picture; however, chaotic flows
can mimic this effect.
The central question is what happens to microstructures, such as those
described by Equation (9.3.6), when placed in a chaotic flow (and also,
but this is a harder question since it involves hydrodynamic interactions,
what happens to the ,flow as a result of the interaction between the
microstructure and the flow). The simplest studies can proceed under the
assumption that it is possible to decouple the global and local aspects of
the problem (i.e., there is no hydrodynamic interactions). Even with this
idealization, several questions come to mind. For example, where is the
region of significantly stretched elements'? The answer to this question
might have relevance in several co-operative phenomena. Besides stretching,
several other phenomena can be incorporated at the microstructural level;
breakup and coagulation/coalescence came to mind. A whole range of
phenomena can be anticipated; for example it seems obvious that the
process can generate fractal-like structures as obtained in diffusion limited
aggregation (see, e.g., Witten and Sander, 1981, 1983; Meakin, 1985).
Example 9.4.1 *
Consider stretching and breakup of pointed drops (with volume (4/3)7-rtr3)
in the flow described in Section 8.2 (partitioned-pipe mixer). The objective
is to determine if it is possible to predict (qualitatively) the behavior of
the droplets based on the results obtained for the stretching of (passive)
material filaments.
In an admittedly oversimplified model (Franjione and Ottino, 1987,
Unpublished), we assume that the drops are characterized by a vector p
1 = lpl have a viscosity ratio p , and obey an evolution equation of the type
(9.3.5), i.e.
DplDt = p - ( V v ) * - (((3 - 1 ) [ D : P P / ( P - P )+~ . f ) p
where
(Vv)* = GD + R,
is the velocity gradient experienced by the droplet a s it moves in the no\\.
and G a n d ,I' are functions defined a s
G = [ I + 12.5(t1.1)" [ I - 2.5(ti 1)".
a / f ) + 0.8p(a/1)3].
f = [ ~ / 2 ( 5 ) " ~ ~ , a ] ( '12/[1
Further, we will arbitrarily assume that the drops break when / , t i > 10.
T h e droplets, somewhat stretched, are placed a t the entrance of tllc
partitioned-pipe mixer and arranged in a square lattice. T h e initial
orientation is purely radial and the initial number of droplets is 3.14 x I oJ.
We will consider two operating conditions, P = 2 a n d /) = 8. T h e valucs
of the parameters for the results presented below are p = 3 x and
efficient regions are the large regular islands where no breakup occurs
all. The results for P = 8 are no more encouraging. Figure E9.4.1 ( h )~ h o \ \ , ~
that breakup occurs near the walls and in two large central regions.
whereas Figure 8.2.10(ul) and (02) although roughly consistent u.itll
each other do not predict such a pattern. It thus appears that these results
could hardly have been predicted based on the maps of stretching of
Section 8.3. Obviously, extensive studies are needed in order to make
qualitative predictions of breakup in this and other complex flows.
Example 9.4.2*
The chaotic flows of Chapters 7 and 8 can be used to study variouh
aggregation processes. Possibly the simplest example, which can bc
cluster order
(b)
Figure E9.4.2 continued
10,000
time
(c)
the kinetics of cluster formation can be described by simple laws, such as
Smoluchowski's mean field kinetics (even though there is no diffusion), and
the dependency of the kinetic constant upon parameters such as p and 6,
can be predicted based on simple models. Figure 9.4.2(c) shows the type
of agreement observed between computations and a kinetic model for
clusters of order one (single particles), two (dimers), and the total number
of particles.
Bibliography
The content of this chapter is more indicative of future possibilities than
of the current state of the art. The common thread is the role played by
local and global flows on diffusion and reaction processes and in stretching,
breakup, and aggregation processes. For example, the fluid mechanical
path is not very important in conversion and selectivity of simple reactions
but appears to be very important in complex reactions such as polymeriz-
ations; small variations in the coefficients of a dynamical equation appear
to determine whether a microstructure breaks or not.
There is a substantial amount of literature on the foundations of both
topics and the review presented here is a rather biased account suited to our
needs. The sections on diffusion and reaction rely on transport across
stretching interfaces. The main ideas and a brief historical review can be
found in Ottino (1982). The importance of stretching of material surfaces
is addressed in many works; one of the earliest is Batchelor (1952) and
various other articles referred to in the chapter. In the context of chemical
engineering, the idea of a local flow was used by Fisher (1968) and by
Chan and Scriven (1970); in mechanical engineering, the idea has been
used in the context of combustion, e.g., Spalding (1978a). The trans-
formations of Section 9.1 can be found in Ranz (1979); Ottino (1982);
and Chella and Ottino (1984) (many of the examples are taken from this
work). The basic ideas of the lamellar viewpoint are discussed by Ranz
('1985).
The review of stretching and breakup does not d o justice to the
state of the art and the reader should consult the reviews by Rallison
(1984); and Acrivos (1983). The most complete summary of experimental
work is given by Grace (1982). Probably the most complete experimental
studies are those from Leal et al. A partial listing includes Bentley and
Leal (1986a); and Stone, Bentley, and Leal (1986). Complete details are
given by Bentley (1985).
Epilogue
Notes
I These terms wcre coined by Arcf and Tryggvason (1984).
3 This viewpoint was suggested by Khakar. Chclln. and Ottino (19x4).
3 This is a good approximation if the distribution o f s is narrow. In chaotic flows
the distribution can be very broad, especially if there are large regular regions.
4 See Example 9.2.3.
5 If the structure is nearly axisymmetric, as the whorl of Problem 4.3.5, we have.
approximately.
where T is the temperature, r,- is an average thermal diffusivity, and r,. is a gcncratic
term that includes energy generation due to chemical reaction, viscous dissipation.
and volume change (see Bird, Stewart, and Lightfoot, 1960, Chap. 18). We d o not
consider multi-component effects.
8 This equation is valid for a material region S,", of size d where a linear velocity
field approximation is valid. For turbulent flow thisscale can be taken as the Kolmogoro
scale, (~'/O~I:)~:'. where E is the viscous dissipation per unit volume. Note, however.
that there can be significant stretching and folding below these scales as seen in thc
examples of Sections 7.1.
9 For example, we know that if the flow is a steady curvilineal flow, the stretching
history s( decays as I/(; in chaotic flows, on the average, s(, is nearly constant, etc.
10 A brief list of other works using similar equations are: Saffman, 1963; Tennekes
and Lumley. 1980; Foister and van de Ven. 1980; and Batchelor. 1979.
l l The application of u,. to mixing was suggested by Corrsin (1954).
12 See Ottino and Macosko (1980). and references therein, and Townsend's hot spot
model, Example 9.1.1.
13 This delinition of 'warped time' coincides with that of Ranz (1979) iff,. = t,,,,.
14 The choice of the scaling time t,. is also important in improving the accuracy of
numerical solutions (it should be selected in such a way that each term (e.g.. i 2 C ,i v
in Equations (9.1.6a.b) is of the order one. The coellicient (e.g.. r,/r,,,) then reflects
the importance of the term). Thus, 1,. is chosen to be r, for 'slow' reactions and it,,,
for 'hst' reactions.
15 Also, note that the streams themselves can be partially premixed and that everything
can be generalized to the case of three or more streams.
I6 Damkohler (19361 defined what is now called the first Damkiihler number, Du,. as
(length/velocity )/characteristic reaction time which can be interpreted ns the riitio
of the characteristic Eulcrian time t o a characteristic reaction time. O u r Du~'.'isdefined
in an analogous fashion. the superscript ( L )refers to the use ofa Lagrangian time-scale.
Notes 3 17
-
17 For examplc. for the reaction A + IJB P. the conditions b r a stationary reacting
'
,,, s,,) = I . ( I J C ,,#, c,,,)(D, D,)' = I.
zone are: (I)(.,,,, s.\~,,(,
18 The case of List reactions has been considered in the context of combustion and
has reccivcd considerable attention in applicd mathematics (see for cxample Kapila.
1983): the LISC of fist reactions as tracers is discussed in Ottino (1981 ).
19 An extreme case is given by polymerizations; see Ex;imple 9.2.1. It is possible also
to imagine situations where imperfect mixing can lead to 'thermal exlosions' due
to imbalance between heat generation and dissipation due to thermal conduction. The
presence ofa thermal explosion depends strongly upon the Lewis number (Ottino, 1982).
20 This topic dates from the work of Danckwerts (19581 and Zweitering (1959). See
references in Chapter I. A book devoted to these issues is Nauman and Buffham (1983).
21 See Section 8.5.
22 Batchelor and Townsend ( 1956)obtained.for homogeneous turbulence. r = 0.43(1:/~0~,'.
'.
/I = o.12(c//O1 ;.= -0.55(1:,'~0' '.
23 The tensor L(X. t ) plays the same role as thestretching function, r(X, 1 ) in Section 9.1.
24 This formulation is largely due to Taylor (see Chapter I ).
25 As indicated in our presentation in Chapter I . Taylor (1934) invented the two basic
experimental configurations; to mimic a two-dimensional hyperbolic flow hc invented
the four-roller apparatus, to mimic shear flows he constructed a parallel flow apparatus.
26 A few references discussing various aspects of experimental results. prior to the work
of Bentley (1985). are the following: Rumscheidt and Mason (1961). repeated some
of the same experiments of Taylor (1934); their rcsults matched the theory for small
deformations. Karam and Bellingr (1968). carried out experiments in simple shear
flow using a Couette apparatus over a relatively wide range of p and obtained a
graph of Cu, versus p which had a minimum for p in the range 0.2 to I. They also
reported a viscosity ratio of p s 4 beyond which breakup was not possible in shear
flow. A few years later an exhaustive experimental study was reported by Grace
(1971, published 1982). In the first half of his paper Grace concentrated on shear
and hyperbolic flows (as Taylor had done), but on a much wider range of viscosity
ratios ( I O - q o 950). whereas in the second half of the paper he concentrated on
the applicability ofthe results to static mixers. In addition to Cu,, the time for breakup
and the length and deformation at breakup were also recorded. He also investigatcd
the breakup under conditions of super-critical strain (Ctr > Cu,). Grace also found
a maximum viscosity ratio, slightly lower than Karam and Bellinger's, p s 3.5, beyond
which drops could not be broken in shear flow. At low viscosity ratios he found
Ca, s p-n.s"or simple shear flow and Ca, s p O . l hfor two-dimensional extensional
flow. For drops of higher viscosity ratio, the data are more sparse and indicate an
increase of Ca, with p . There have been relatively few studies focusing on the effect
of the history of the imposed flow and the dynamics of the drop. One early study
is by Torza, Cox, and Mason (1972). They found that the rate of increase
of the strain rate (dS/(/t) had a significant impact on breakup. Though data are only
reported for relatively low values ofdS/dr. they found that the critical strain rate
was a function ofdS/dt, especially at high dS/dt.
27 See Bentley and Leal ( 1986a)for details of the apparatus and Bentley and Leal ( 1986b)
for experimental results.
28 It is significant that the breakup mechanism wus rtor due to capillary wave instabilities
as studied by various researchers (e.g., Mikami, Cox, and Mason, 1975; Khakhar
and Ottino, 1987). A long thread is not equivalent to an infinite cylinder and end
effects cannot be neglected under most conditions of interest. Obviously, a significant
difference between an infinite and a linite cylinder is that an infinite thread is an
equilibrium shape whereas a finite cylinder is not; in the case of an infinite cylinder
the only question is whether o r not it is stable (Stone and Leal, 1989). However,
if drops are significantly stretched, Na>>l@, as they are in chaotic flows,
the prevalent mode of breakup is by capillary waves.
Epilog ur
20 Other results for time dependent flows indicate the severity of the flu~dmechanic;~l
path. For example. Khakhnr and Ottino (1987) considered the number of f r a g n i c n ~ ~
N produced ny bursting of an ittlittirc, thread by means of a model based o ~ gi r o ~ l h
of capillary w;~vcs.The flow field around the thread was taken to be
I.: = 1(1)I. I., = - :E(I)~.
Two cases were compared; one with constant 1. the other with l ( r ) = S/( l + Sr ) where
S is a constant. The number of fragments goes as Cii2~"forconstant 1 and as Cii'
for the u s e 1(r)= S, ( I + SI ) (where Crr is the capillary number. Cii = ~r,Sil.o. and
il the initial diameter of the thread). This implies that under identical conditions
n flow with dccaying efficiency will produce significantly fewer fragments than onc
with constant elliciency. These results appear to have interesting implications in thc
context ofexperimental studiesof breakup in chaotic flows,such as those ofsections 7.4
and 7.5. Although the experiments are just at the beginning and it is early to draw
lirm conclusions, it does appear that there is significantly more stretching and
breakup within the chaotic regions as compared with the regular regions, where thc
elliciency of the flow is poor.
30 Choices corresponding to two-dimensional extensional flow and simple shear flow
are given by Khakhar and Ottino (1986b).
31 This simple picture implicitly considers the possibility of drag reduction withour
any wall eNects. Such a viewpoint was advocated by de Gennes (1986).
32 Fornn introduction to the literatureofdrag reduction see Lumley (1973). Virk (1975).
Rerman (1978). and many of the articles in Vol. 24(5). J . 1!/' Rhrology. 1980.
33 Recent studies involving experiments in the journal bearing flow of Section 7.4 show
that droplets in chaotic regions can stretch by several orders of magnitude and break
primarily by capillary wave instabilities whereas drops placed in large islands, such
as those of Figure 7.4.5(a), hardly deform at all (Tjahjadi, 1989).
34 Recent theoretical analyses and computations involving fast bimolecular reactions
+
( A B- P) in systems with initially distributed striations of the reactants - such as
Figure E9.2.1- have shown scaling behavior. Owing to the unevenness in the striation
distribution the reaction planes move and the average striation thickness increases
in time. The remarkable result is that, regardless of the initial distribution, the system
evolves in such a way as to produce a self-similar time-dependent striation thickness
distribution (F. J . Muzzio and J . M. Ottino, 'Evolution of a lamellar system with
diffusion and reaction: a scaling approach', Phys. Rev. Lerr., 63,47-50, 1989).
Cartesian vectors and tensors
I Sum
(1) Associative u+(v+w)=(u+v)+w
(2) Commutative u+v=v+u
(3) Existence of zero (0) u+O=u
(4) Existence of negative u+(-u)=O.
11 Multiplication
( I ) Associative dpu)=( 4 0 ~
(2) Unit multipication lu=u
(3) Distributive +
( a + P)u = r u flu
(4) Distributive a(u + v) = au + av.
Properties I + 11 form a vector space.
Operations
Co-ordinate system, basis, dual basis
From here on the discussion is restricted to three-dimensional space. In
general, a point x in space is located by its co-ordinates, x = x ( s l . r'. .s.').
For example, in cylindrical coordinates, x is specified by r, (I, z . The nrirlr~.ol
hllsi.s jei) is defined as
ei = ix/isi
-
We require the je,) to be linearly independent, i.e., xiei = 0 implies th~tt
xi 0." Consider a vector v located at x, i.e., v = v(x). The components
defined with respect to the natural basis,
r l ( x )= e, v(x)
are called the conrrc~ruriantc8omponent.s.The dual basis, ( e i ) , is defincd
such that
e i - e J= 1 if i = j
e i - e j = 0 if i # j
The components defined with respect to the dual husis
ui(x) = ei. v(x)
are called the corclritrnr componrnts. A coordinate system is callcd
orrhogpnul if the natural basis is such that
e i . e j # O if i = j
e i . e j = 0 if i # j
For an orthogonal co-ordinate system the vectors
e < i , - ei/leil = ei/c)i
Figure A . I
322 Appendix: C u r t e s i u ~rectors
~ trr~dterlsors
Problem
Verify that bjrijks ijjsSkr
l : ~ , , ,-
~ ~ ~ ~=
Problem
Using the results given above prove
u x (V x w)=v(u-w)-w(u.v).
This identity provides a means for proving the Cauchy Schwarz inequalit!
The are the components of v with respect to { e , ) , and the ri are the
tii
Remark: If the basis is not orthogonal we need the dual basis and v is
expressed as
v = ( V .el)ei= (v.ei)el.
and
v +u -
Addition and subtraction and multiplication by scalar in component form
(ri+ ui)ei
XV =( ~ 1 1 : ~ ) ~ ~ .
.v - J
Scalar product
~I . e .c..e . = uI.t.J.e..
I
eJ. = u.r .h.. = u.r
I
- u .c.
~ ' i - 11'
Representation theorem
Any L S F has the representation f ( x ) = f - x , where the vector f is unique
and independent of x.
( i ) Uniqueness: Assume f and f' such that f - x =f'.x = f(x). Then
f - x - f'.x = 0 and using property III(2), (f - f'). x = 0. Since x is arbitrary,
using property III(5) we obtain f = f'.
(ii) Existence: f(x) = f(xiei) and using linearity f ( x ) = xi f(ei) = x.e,f(e,) =
f(ei)ei. x, so f = f(ei)ei.
Problem
Show that ,/(.)=I 1, where the bars denote magnitude, is not a LSF.
Linear. vector .frrnctions (1- V F )
I>cfinition: Mapping t ( . ) such that V + V (vectors into hectors)
+
t(x T y ) = f ( ~ ) f(y)
t(xx) = xt(x).
Exutnplc 1
The multiplication of a vector by scalar is a LVF.
Exanlple 2
I>cfinc (a b).u -a(b.u), where the dot can be interpreted as (a b) actin!
on u. Note that the result is parallel to a. The entity (a b) is ailled a d!ad.
Notc also that if the order of a and b is reversed the result of the operation
is parallel to b. (b a).u = b(a.u). Similarly, u.(a b) =- (u.a)b.
A note uhout notution
The dyad (a b) is often denoted (a 8b). Customarily. within this notation
the product of a vector and a tensor. here denoted as T-v. is denoted a,
.l'v. The contraction (or inner product) of two tensors or a tensor a n d
dyad. here designated as T: S or T: nn, is designated as T - S and Tan S n .
respectively. The main motivation for this notation, which the reader i \
urged to master. is that the inner product of both vectors and tensors is
designated with the same symbol.
Representation theorem
Any I.VF has the representation
+
t(u)= ( e l t l ) - u (e,t,).u +
(e,t,).u
+ +
= ( e l t l ezt, e,t,).u
= (eiti).u= e,(ti.u)
whcrc the ti arc unique.
( i ) L'rlicluc~rlc~s.~:
Consider two representations
t(u) = e,(t,. u) = ei(t;.u).
Taking the inner product with ei,
ei.t(u) = ( t i - u )= (ti-u)
and by the same reasoning as with LSF, ti = tj.
( i i ) E.~I'sIL~II(.~:
We write
t(u) = e,(t,-u)= ti(u) e,
VL'CIO~ component5 basis
It is enough to prove that t j ( u )is a LSF. Write
t(u) = t,(u)e,
t ( v ) = ti(v)ei.
Appendix: Cartesian vectors and tensors
Then
+ +
t(u) t(v) = ti(u)ei ti(v)ei= [ti(u) ri(v)]ei. +
Also since t( ) is a LVF,
t(u + v) = ti(u + v)ei,
then
ti(u) + t,(v) = ti@+ v).
Also
t(ru) = (eiti).ru = ei(ti.ru)= eiti(ru)= r(eiti).u = rei(ti.u) = eiati(u)
then
ri(ru)= rti(u).
Tensors
A LVF is the most general linear transformation of vectors into vectors.
Symbolically, t(u) is written as
t(u) = T - U ,
+
where T = e,t, e2t2+ e3t3.This sum of dyadics is called a tensor. (More
precisely, a second rank tensor. Unless stated explicitly, whenever we use
the word tensor we mean a second rank tensor.) Similarly, we can define
linear transformations between tensors, and so on.
Components of a tensor
Any of the ti, for example, t,, has component representation
t , = t l Jej, where t , means the jth component of t,
[Remark: Note convention here. We could just as well have written tjl
to mean the same thing.]
Thus, in general,
t 1. = r.*I J J
and
T = e.t.
1 1
= e.t..e.
I I J J
= t.e.e.
I J L J '
Usually, the components rij are denoted with the same letter as the tensor,
1.e..
T = TiJeiej.
As with vectors, the array of Tij,usually displayed in matrix form, is one
of the infinitely many possible representations of the tensor T. Thus, we
can write
T = T.l .e.e.
I I J
= Tr .ere'.
1 I I J
in any basis.
Unit tensor ( I )
Defined such that I mu = u for u. The components are
[llij = e i . ( l . e j )= ei.ej= Sij.
Thus the component representation is
Product by scalar
Defined such that 2T.u - T-xu.
Addition (and subtraction)
+ -
Defined such that IT S).u T-u S . U .
Inverse tensor
+
Givcn T , there exists 'r-'. such that if det T #0, then T - T - = 1. '
T - . '- T = 1. If det 1' # 0, T is callcd non-singular. I t can be shown that
det 'l' docs not depend on the basis and that 'r-I is unique. The rule of
computation of T ' is similar to that used in matrices.
Appendix: Cartesian Gectors and tensors
Transpose
Vector function t T ( . )such that tr(u).v= t(v).u.
Theorem
tT( ) is a LVF: Note
tT(u+v).w= t(w).(u+ v ) = t(w).u + t(w).v
= tT(u).w+ tT(v).w= (tT(u)+ tT(v)).w.
Since w is arbitrary,
+
tT(u v ) = tT(u)+ uT(v).
Also
tT(u).rw= atT(u).w= rt(w).u = t(w).ru = tT(ru).w
and since w is arbitrary,
rtT(u)= tT(ru).
Note: The transpose of T is denoted TT.Thus, we write (TT.v).u= (T-u).v.
Component representation of transpose
[TTIij= ei. (TT.ej)= ( ~ ~ . e =~ (T.ei).ej
) - e ~ = Tji.
Note: If T = Tijeiej,then TT= Tjieiej.If T = TT, T is called symmetric. If
T = -TT, T is called antisymmctric or skew.
Problem
Show that T = (TT)T.
Problem
Show that any tensor T can be decomposed, uniquely. into symmetric
and antisymmetric parts:
T, = (T + TT)/2,symmetric.
TA= (T - TT)/2,antisymmetric.
Component representation of a dyad
= akbl(ei.ek)(el.ej)
= u,h,di,dlj = uih,dlj= aibj.
Note:
(a b)' = (b a).
Useful relations inuolcing unit vectors
( 1 ) eiej-ek= ei(ej.ek)= eidjk
(2) eiej.eke,= ei(ej.e,)el = eiSkjel
(3) eieJ x ek= e,[ej x ek]
(4) e, x eJek= Ce, x ejle,
(5) eiej:eke,= (ej.ek)(ei.el) = 6jkiii,.
328 Appendix: Cartesian vectors and tensors
Warning: definition (5) is not universal. For example, some authors define
eiej:ekel= (ej.el)(ei.ek)
rather than (ej.e,)(ei.el).
Remarks: ck6jk= (i.
Tk16jk
= Tjl
bjk= d k j .
uv: wz = U i V j W j Z i .
Problem
Show that if T = TT and S = -ST then T : S = 0. Thus if ':' is interpreted
as an inner product this means that the projection is zero.
Product of two tensors
T .S = TiJeiej.
&,ekel= TijSklbjkeiel
= TikSkleiel, which is a tensor.
Note: This corresponds to usual matrix multiplication ( [ 3 x 3][3 x 31).
Table A. 1
In general,
e!1 = a [J eJ .
ei = aie'.
:/I.
J J'
CQI = [:!:!
a3 a3 a3
Q is an orthogonal matrix. Consider now some of the properties of Q.
Note that
e i . ei = d,,.
Replacing the transformation we obtain
Similarly,
01 02 a31
Note that a x b-c = Volume of prism = 6,
b2
CI c2 c3I
Then,
e; xe;.e;= 1 .
Since
e; = a i e , + u$e, + a:e,
eI, = u i e , + uie, + u:e3
then detjurj = 1 .
330 Appendix: Cartesian vectors and tensors
Problem
Prove that uiu; = 6,,.
Problem
Prove that detla;] = 1.
Since
es = aze:
we write
L' ase' = c'e'
s r r r r
(csa: - c:)e: = 0.
Since all the components of 0 are zero,
L"r = r IS) or in matrix form, [v'] = [Q][v].
Similarly,
0, = ci'u' or in matrix form, [v] = [QT][v'].
r r7
Further properties of Q
Since v transforms as [v'] = [Q][v] and the length of v is invariant, we
obtain
= Ca'lCa'l= CQICalCQICaI = CQTICQICalCal = Cal [a]= la21
then,
[QTICQI= C11
or
CQTl= CQ- 'I.
Also,
det([QT][Q]) = det[l] = 1
det[QT] det[Q] = (det[Q])2 = 1
det[Q] = k 1.
Q is called proper orthogonal if det[Q] = 1 .+
Problem
Prove that [Q] preserves angles between vectors.
Appendix: Cartesian uectors und tensors 33 1
Tensors
Consider two vectors v and u and a tensor T such that u = T . v . The
question is: What happens when T is seen in terms of {e;) and (e,)'? We
know that:
U: =aJu.
I J
and
Uj = Tjqt',
Then
u: = uiT.14 6 4
we get
T i , = aja4,Tj,
Problem
Derive the above result using matrix manipulations. Consider
and
332 Appendix: Cartesian t'ecrors and tensors
Trace
Definition: Mapping, tr( ), 8 + R, such that
tr(T + S) = tr(T) + tr(S)
tr(rT) = r tr(T)
(i.e., linear), and
tr(a b) = a-b.
Calculation: tr(T) = tr(Ti,eiej)= Tj tr(eiej)= 7;.j6ij= Ti.
Properties:
tr(ST)= tr(S)
tr(S.T) = tr(T-S).
Magnitude of a tensor
IT/ = /[tr(T-TT)]Ii2I.
The Cauchy-Schwarz inequality, for tensors, reads
IT: SI < IT1 ISI.
Problem
Show that IT1 = I(T:TT)1121.
(See earlier comments about notation.)
Problem
Show that in 1=
1 1 if In1 = 1
Problem
Show that T: TT2 0,
Problem
Prove that tr(A.B.C) = tr(B.C.A). Use this result to prove that the trace
is invariant.
Problem
Prove uv: wz = uw: vz = (u.z)(v.w).
Appendix: Curtesiun rectors und tensors 333
Determinant
The determinant is calculated using the representation of T in any basis.
The result is independent of the basis (without proof).
7.11 TI, 7-13,
det T = 7i1 T,, TZ3.
T3 1 T32 T.33
P~soblem
Prove that dct(A I) = [det(A)]- I.
Problem
Prove that det(T') = det(T).
I,.. II,.. and Ill, are called the principal invariants of the tensor. As we
have sccn I,.. 11-,.and Ill-,. arc independent of the choice of basis.
Chyley-Humilton theorem (without proof)
Any tensor T satisfies its own characteristic equation, i.e.,
1- .3 - l l ~ r 2 + l l , ' l ~ - I l I , l = O .
Note: Besides I,, 11,. and 111, the other invariants commonly encountered
in the literature are the moments
I; = tr('f), II', = tr('f2), 111; = tr(-r3).
Representution theorem for symmetric tensor (without proof)
Any isotropic scalar function of a symmetric tensor argument can be
expressed as a function of the invariants of the argument:
,f(T)= (](IT? 111, 111,)
or. alternatively,
,/IT) = gC(l',
, I I.;.. 11 I.;.).
Example
Show that if T is symmetric, the eigenvectors belonging to different
eigenvalues are orthogonal.
Ilcnotc n,, ii and nj. il as two sets of eigenvcctors eigenvalues such
that ii# /.j. then
,.
I . n ., = ;..n.
I ! and T.nj = ;.,nj
then, cross-multiplying.
(T.ni).nj= iini.nj and (T-nj).ni= ijnj-ni.
Using the delinition of transpose. since ?' is symmetric.
and
. .
(1.;- /.j)nisnj = 0.
Since. by assumption. ii# i j then ni and nj are orthogonal.
Exumple
Obtain the rcprcscntation of the above tensor with respect to its eigen-
vectors. Construct
T..= ( ' r - nI . ) . nI . = ; . . n . . n i = / ...I (.)11'
I I .
So the matrix representation is
i , 0 0
0 0 i,
Problem
Using the Cayley Hamilton theorem prove that 'T' = Q - T - Q ' and T have
the same invariants.
Problem
Using the Cayley Hamilton theorem prove that T". II > 3, can be written
in terms of 1, '1'. and T'.
Problem
Show that if T is invariant, then Tnis invariant.
Problem
Show that T and TT have the same cigenvalucs.
Prohlcm
Show that if 'l'.n =in. thcn T ' an = i In.
Problem
Show that if T = 'f.''and the 7; are real then the eigenvalues of T are
real.
Problem
Continuing the problem above, construct the matrix [L] such that its
columns, n,, are the eigenvectors of T, i.e.,
rl.1= rnln2n,l
where the ni are written as ( 3 x 1) vectors. Show that [LT][I'][I.] is
diagonal.
Problem
Prove that T and T ' = Q . T . V T have the same eigenvalues and
eigenvectors.
Problem
Show that if W = - = 0 is an eigenvalue.
W ' thcn i.
Problem
Show that thc maximum (or minimum) of D : nn, with D given and In1 = I ,
is given by the solution of the eigenvalue problem D . n = Ln. Identify i.
with a 1.agrange multiplier.
Denoting ldxl = ds
V f is a vector.
therefore
Appendix: Carteslun rectors unrl tensors
and
component form.
If V . v = 0, v is called solenoidal.
If V x v = 0, v is called irrotational.
Other operations, such as V x T, can be defined similarly.
Problem
Prove the following results. Both j'and g are scalar functions, v and u
are vector functions.
V(f9) = fV9 + 9Vf
V(f+g)=Vf+Vg
v~(u+v)=v.u+v~v
V.(fv) = (Vf).v + f(V-v)
v .
(fv) = f ( V V )+ .
~ (Vf) v
Problem
Prove the following results
+
V*(vw) = v - v w ~(V'V)
v x f ~ = v f ~ v + f ( v ~ ~ )
V2(V.v)= v . v 2 v
V.(fl)=Vf
V . ( V V )= ~ V(V.v)
V(a-x) = a, where a is a constant vector and x = xiei
VxVf=O
V-VXVEO
vv:1 =v . v
) -x/1xI3 where x = xiei.
~ ( l / l x l=
Vectors
Appendix: Curtesiun oectors N I I ~tensors
Tensors
Tensors
Notes
1 Excellent references for this material in the context of continuum mechanics and fluid
mechanics, respectively. are Chadwick (1976)and Aris (1962).Portions of this appendix
are based on the appendix of Coleman, Markovitz, and Noll (1966).
2 As we shall see this definition includes both 'vectors' and 'tensors'.
3 Unless explicitly stated we use Einstein's summation convention: rcpcatcd indices are
summed from 1 to 3, i.e.. riei = riel + r,e, + r3e3.For example. q j r j r imeans the
double sum 11 Tijrjriwith both indims running from I to 3.
4 This applies to components of vectors and tensors. Throughout this work, unless
explicitly stated otherwise, we will deal with orthonormal bases.
5 As is usual in fluid mechanics we do not distinguish between column and row
vectors to simplify the notation; also, the distinction it is always clear by the
context of the operation.
Figure A.Z.
n
6 'I'he re~idcrrnighr find i r in\trucIr\c r r ) conlp'irc t l i i h \\irh [lie c\pari\iorl of a f~rnclrori
in terms of base functions, a, in Fourier herich for example. and to identify the terms
(v'e,) with the Fouriercoelficienta.
7 In the Iltcraturc one~on~e~imcscncouritcrs opcr,~lioli\siich as ~ v . V lO. n e intcrprclarion
15 lo regard V. by irhelf. as a vecror. We prefcr. ho~vcver10 regard ir ,is a n opcraror
C'hell;~. R.. and J. hl. O t t ~ n o119x4) ('onicrs~on itnd aelecti\it! modific;~tiorls due to
rnl\ing i n unpre~nixedre;ictors. C'II~III. 1:11,q. Sci.. 39. 551 67.
('liella. K..,ind J. M .Orrino (1085:l) IFIuid nlechd11ic.so f m ~ k ~ 111
n gi t sinplcscre\\ ex1rudi.r.
/II(/. \:ILI. . 24. I70 80.
('IICIII. 1 LIII~I~II~I
C'hcll;i, R.. and J. M .Ottino (I9XX5h) Stretching i n some classes of fluitl niotion.; and
asy~nproticmixing elliciencies as measure o l f l o ~clu~sification. 41~11.Ktrt. .Llcc./l. ..llr~rl..
90. IS 47.
('ooper. A . R . (1066) Diffusive mixlng i n continuous laminar flow systems. Clrr~,~. Enq. St,i..
21. 1095 1106.
('orcou. (;. 'M.. and 1'. S.Sherman (19114) I'he niixlrlg layer: detcrmlr~istic~nodelro f
;I ti~rhuleritflow. f'arl I . Introducr~oriand thc two-diriicn.;~orl;II flow. J. I ~ l i t i i .M,*111..
l
193. 70 0 5 .
('orr\in. S. (19541A nieasure of the .crc.l of .I Ilornogeneoua r;~ntlt)riisurf:~~.c in space.
~1r11rr..l/l/l/. lldr/l.. 12. 404 8
('orrsin. S. ( 1057) Si~npletheor> of ~ i ~dc:il~/cd
i t ~ ~ r b u l c niI\er,
r~t A/('/I/: J.. 3, 329 30.
('orrii~l.S. ( 1072)Simple proofoflluid line grouth In st;itlon;lr! homogeneou.; turbulence.
/'/I\\. 1:/11rd\. 15. I370 72.
C'orr\in. S.. :~ntlh l . K;~r\rciti 1969)1.luiil linc groiith in gr~ct-gencr;~ted so tropic t~~rbulcncc.
.I. k ~ l ~.~l rl ~~ l~ 39.
~ ~ 87
/ ~ 06.
..
Cox. R . <i (1969)The ilcforrn:~tiono i ,I drop in ;I gcncr;il time-dependent Iluid flou.
.I. l'/lli</ll<,l~/l,. 37. 601 23,
('ra~rford.J . I).. and S. Omohundro (IOX41 On the global structllre of per~oddoubling
flous. I'l~~.\ic.tr.13I). 101 X O
I)anlk(ihler. ( i . (1030) Firilliisse tier Stri)riiung. Diffus~on~ l n ddc\ Wiirrneuberganges
:~ufJIG I.e~sturlgi o n Re;~htionsiifen.I . Allpernei~icGes~clitspunktefiir die ilbertragung
cirleb chemls~.hcnPro/essei au5 dern Kleinen Ins Grosse. Z.E1cArroc.lic~r11.. 42. 846 62.
Di~richucrth.P. V (1952) 'I'he definition and measurelnent of solne characteristics of
rniulure\. /tp/ll. S C I .Kcr.. A3. 279 96.
I)a~lchuerts.P. V . ( I1)53l Contlnuou\ llou \!>terns-di\tnbutiun of residence time\. ('/I(,III
1>11</..Sl.l.. 2. I 13.
D;inch\rcrth. P. V . i 1958) Ihe effect of ~ncomplctenilxlng on homogeneous rc;ictions.
CIICIII.t.~rq.S(,i,,8. 03 9,
I>anielson.~I.J.. I'h.I).~lhcsib. Dept. ofChemical Ikginccr~ng.University of Massachusetts.
Amherbt. in progress.
tlc<;cnrics. P. G ( I9Xh)Towards o sci~lingtheor) of drag reduction. P/~!..~ic~tr. 140.4. 9 75.
De\aney. K. L. (1986) ,411 ir~rrorl~r(.rio~~ ro c./~rrr~ric tlvt~trrilic~trl
\!,.\rc~i~rs. Menlo k i r k :
Benjamin ('u~nmings.
1)iniotakis. P. F.. R . C. M~;ihc-l.ye.and L). A . I'npnntoniou (IOX3)Structure;indd~namics
of turbulent round lets. I'lii.\. I.'Iirie/\. 26. 31x5 9-7.
Dohcrty. M . F.. ; ~ n dJ . V . Oltino (19881('ha05 in dcterrninistic sh\tcrn\: strange ,ittr;ictors.
turbulence. and applications in chemical engineering. C'lic~~ii. t ~ i q S(,i..
, 43. 139-8.1.
1)ornbre. 1 . .. I;. Frisch, J . M . <irccne, M. tlinon. A . Mehr. and A. M . Soward ( 19x6)
Ch;iotic streamlines in the AH(' lloirs. .I. I.l~ri(l11~c.11..167. 353 91
Duffing. ( i . (1924) Hcitrag f u r Theoric dcr Flussighcitsbcwegunp /wisehen Zapfcn
llnd I.ager. Z. oti~qc,1r..2./trrh. .tlrc.h. ( % / t . M2.1). 4. 297 314.
Cchart. C. (1048) A n arlalysis of ttic stirring ;tnd mixing proccs\ in incornprcss~blefluids.
.I. .31[1ri11eK1~5..V I I . 265 75.
1Jsc:indc. 1). 1:. (1985) Stochasticit! in cl;i\>ic;il Il,~rniltoniansystems: uniccrsal aspects.
1'hj.j. Kc/~orr.\.121. 165 261.
Feigenhaum. M .J . (I9XO) Lniicrsal bchavlor in non-linear sysrctns. Lo.\ A ~ ~ ~ I S~.iet~t,c,. IIIJ>
1. 4.27.
I..eingold. M.. L. P. Kadanoff. and 0. Piro (19x8) l'assive scalars, thrcc-dimensional
volume prcscrving maps and chaos. J . Srur. Phxs.. 50. 529-65.
IFields. S. D..;~ndJ . M.Ottino (IVX7a)Effect ofscgreg;~tionon [he course ofpolymerization.\.
.AIC/lE .I.. 33. 959 75.
IFiclds. S . D.. ;ind J . M . Ottino (1987b) Effect of stretching path on the course or
polymcri7ations: applic;itions to ide;~lii.ed unpremixed rctictors. ('/~(JII~. E I I ~S1,i..
. 42,
467 77.
ticldh. S. D.. and J . M . Ottino (IC)X7c)Effect of striiition thiekncsh distribution on the
coursc or an unpremixed polymcrii.;~tion. (.horn. E ~ l y S1.i.. . 42. 459 65.
pol,mcri/:~tion h> Incnn, of :I neu rnlcrobtagc reactor. 1'011111t*r.27. (13 72.
I-'iri1:iyson. 1%.A . ( 1072) \ I ~ ~ I I I11I1J ~1\ti!/111tvl
/ I.c,.\~~/~Io/.\
111111I ~1riori01rl11
/~ri~t~,iplo,\.
Xeu
York : A c ; ~ d c ~ n iI're\
c
Fisher. I). A. ( IOhXI A modcl for first rc;rction$ in t u r b u l e n t l mixed fluids. M.Sci. I'hesis.
I>cpt. of Chc~nicalEnp~~lcerinp. l;nr\cr\it! of Minnesota. Minneapolis.
I--oister. R . T.. and T . <;. M . \;in cle Ven IIUXO) I>iffi~sionof Brownian particles in shear
MacKay. R. S.. J. I). Mciss. and I. C. Percival (19S4) Transport in Harniltonian systems.
Physica. 13». 55 81.
MacKay. R. S..ancl. J. D. Meiss (1987) Humiltonian dynamical systems: a reprint selection.
Philadelphia: Taylor and Francis.
Malvern. L. F:. (1969) Introduction ю the mechanics of a continuous medium. Fnglewood
Cliffs. N.J.: Prentice-Hall.
Marble. F. Li., and J. F. Broad well (1977) The coherent flame model for turbulent chemical
reactions. Project SQL II). lech. Rep. 7КИ-У-/Ч..
May. R. M. (1976) Models with very complicated dynamics. Suture. 261. 459 67.
McOibc. W. L.. and J. C. Smith (1956) L'nit operations of chemical enaineerinq. New
York: McGraw-Hill. 2nd edition.
MeKen/ie. D. P. (1983) The earth's mantle. Scientific American. 249, 66 78.
McKen/.ie. D. P.. J. M. Roberts, and N. (). Weiss (1974) Convection in the earth's
mantle: towards a numerical simulation. J. fluid Mech.. 62. 465 538.
McMurtry. P. Л.. W.-H. Jou. J. J.. Riley. and R. W. Metcalfe (1986) Direct numerical
simulations of a reacting mixing layer with chemical heat release. А1ЛЛ Journal. 24.
962 70.
Meakin. P. (1985) Computer simulation of growth and aggregation processes, in On arowth
and form: fractal and non-fractal patterns in physics. H. I'. Stanley and N. Ostrovvsky.
eds.. pp. 111 135. Dordrecht: Nijhoff.
Melnikov. V. K. (1963) On the stability of the center for time periodic perturbations.
Trans. Moscow Math. Soc. 12. 1 57.
Middleman. S, (1977) Fundamentals of polymer processina. New York: McGraw-Hill.
Mikami. Г.. R. Cox. and R. G. Mason (1985) Breakup of extending liquid threads.
Int. J. Multiphase Flow. 2. 113-18.
Milne-Thomson, 1.. M. (1955) Theoretical hydrodynamics. New York: Macmillan. 3rd
edition.
Minorsky. N. (1962) Sonlincar oscillations. Princeton: Van Nostrand.
Moffat. H. K. (1969) The degree of knottedness of tangled vortex lines. J. Fluid Mech..
35. 117 129.
Mohr. W. D., R. L. Saxton. and С. Н. Jepson (1957) Mixing in laminar flow systems.
I mi. Fnq. Chem.. 49. 1855 7.
Moser. J. (1962) On invariant curves of area-preserving mappings of an annulus. Saihr.
Akad. Wiss. (iottinaen Math. Phys. Kl. II. I 20.
Moser. J. (1973) Stable and random motion in dynamical systems. Princeton: Princeton
Lnivcrsit} Press.
Muzzio, F. J. and J. M. Ottino (1988) Coagulation in chaotic flows, Phxs. Rev. Л. 38.
2516-24.
Naumann. F. В.. and В. Л. Buffham (183) Mi\ina in continuous flows systems. New
York: Wiley.
Noll. W. (1962) Motions with constant stretch history. Arch. Rat. Mech. Anal.. 11.97 105.
Nollert. M. I)., and W. L. Olbricht (1985) Macromoleeular deformation in periodic
extensional flows. Rheol. Acta. 24. 3 14.
Olbricht. W. 1... J. M. Rallison. and L. (j. Leal (1982) Strong flow criteria based on
micros!ructlire deformation. J. non-\cwt. Fluid Mech.. 10. 291 318.
Onsagcr. L. (1949) Statistical hydrodynamics. Suoro Cimento. VI ser. IX. 279 87.
Oseen. C. W. (1931) Vehr. des .?. int. Knnar. fur techn. Mechanik [Stockholm 1930j
Bd. I. pp. 3 22. Stockholm.
0111110. J. X I . (1081 1 l..fficicncy of mixing f r ~ ) md d t ; ~on fils1 react~onsi n ~nulti-jetreactorb
and \tirrcd tanks. ..11('1rL J.. 27. I 8 4 01.
Ortino. J. .M (19x7) 1)cscription of mixing u ~ t hd~ffusionand reacton 111 terms of the
concept o f m;~tcrialsurfaces. J. t lrtid .Llz(,l~..114. 83 103.
Ottino. J. M.. ant1 C. W. Macosko (1980) A n efficiency parameter for hatch mlxlng
of viscous fluids. ('hrr11. I-II~. S(.i.. 35. 1454 7.
Otrino. J. M.. W. E. Kiln/. and C. W . 51acosko (1979) A lamellar model for an;ilh\is
of I ~ q u ~ d - l i q u miring.
id ('~IC~III. I<~ey.Si-i.. 34. 877 00.
Ottino. J. 51.. W. I;. Ran/. ;lnd C. W . M;icosko ( I W I ) A fr;~me\borh for the mr.chanic;~l
mixing o f fluids. .4/(-/1l:' .I.. 27. 505 77.
Ottino. J. M.. C. W. Leong. H . Rising. and P. D.Swanson (1988) Morphological structures
produced b y mixing i n chaotic flows, h'uturr. 333, 419-25.
Patterson. A K. (1983) t1 /ir\r (,olrr\lJi r ~/l~rcilr ~ ~ ~ ~ c ~ l r i r r<'ambridge:
ric~\. C;~mbridgcLnivcrsit)
l'rcss
I';~lih. J. 119691 O n Morse Smalc dynitm~calaystcms. 7r1polo(lr. 8. 385 405.
I'an. t'..and A. A c r ~ r o s(1967)Stead) flous i n rectangular cav~ties,J. I- 11ccd .tf(,ch.. 28.643 55.
Peixoto. M. M. (1962) Structural stability o n two-dimensional manifolds, Topology.
2. 101 20.
Pcrci\ul. I..and I > . Richards (1987) 111rroilrrc~rio11 10 tl~.rcu~~ri(~\.('amhridge: ('amhrldge
I'ni\crsit! Press.
Pcrr). A. E.. and M . S. ('hong (I9861 A \cries-expa~isionstudy o f the Savrer-Stohcs
equ;~tion\ with app11c;uions to three-dimens~onalseparation patterns. .I. I.'lucd .\lri./i..
173, 2907 2 3 .
I ' c r r ~ .A. L:.. and M S. (-hong ( 1987) A de\cr~ptionofeddqing motionsand flow patterns
using cr~tlc;~l-point concepts. AIII. Kvc . I--lrritl Mec.11.. 19. 115 55.
Pcrr! . A. I-:.. and B. I). f';~irlic ( 1074) Critical polnts i n flou patterns. Atll-. ill Grol)h!-\..
18B. 299 315.
t'oinc;~rC. II.(1892. lX03. 1x99) /,I,\ ~ ~ r c ~ t l r IIOI~I~L~~IC~~ o i / ~ ~ \ (I(, /u I I ~ ~ ~ ~ I3~
(.L;/L,Y~C,. ~~II
LOIS..
I'i~rib: ~ i : ~ u t h ~ c r - V ~ I l ~ r r ~ .
Pr;lndtl. 1.. (10-751I k r i c h t uhcr Untcrsuchungen /ur irusgebildcten -1-urbulcn~.Z. .AII!~L,I~.
If(ir11. .\f1~1,.(Z,.l ,Ll 11 J.5. 136.- 139.
I'randtl. I. .and 0 . C;. T~et.jens(1934) t~u~ci/irrric~~~~trl\ (!/'111'rlro-cold ierro-~~~c,(.h(r~l~l..\
(Fnginccring
Socictics Mvnogr;~ph\). I \ t etin. X.Y.: McGr;i\\.-tiill (repr~ntedby Dobcr. 1957).
R:~llison.J. M . (1984) The deformation of small viscou\ drops and hubblcs i n shear
flo\rs. ~11e1i.K ~ Y I .I c e c i l .21(,(11.. 16. 45 66.
Ralph. M. F . (1986) Oscillatory flows i n wavy-w;~llcd tuhcs. J. Fllrid :Wr(.h.. 168. 515-40.
R;rn~. W. E. (1979) Applicirt~onof a stretch model t o m ~ r i n g diffusion. . and rc;iction
i n laminar ant1 turbulent flo\v\. ...lIC'11t .I.. 25. 41 --7.
Ranz. W. E. (1985) I-luid mechanical mixing-lamellar description, Ch. 1. pp. 1-78 In
.Miuin<l qf liqu~dshy ~t~echrrnir,trl u~qir(rtior~. J. J. L'lbrccht and Ci. I(. Pattcrson. eds. Y c w
York: G o r d o n and Breach.
Reichl. L. 1:. ( I V X O I ..I ~r~oclerrr (.ollr\cl ill .\rieri\ei(-~tl
plr,i\ic,.\. Au\tin: University ofTexas
Press
Khinci. 1'. R. 1 I9771 The dynarnics o f unbteady currents. i n 7'he Scvr. C'olrrrr~u6 , Murittc,
~ ~ ~ ~ ~ ( l e 1:.
l l i rI).r q (ioldberg.
. I. X . MCC;IVC. J. J. O'Bricn. and J. H. Stcele. eds.. New
York: Wil~.!
Rhll~c\.P. H. ( 11)791 (icortrophic lurhulcncc. Ali11. K r r . I.11riJ .Vlrch.. 11. 401-41.
Rhinc\. 1'. H. (11)X3) Vort~cit) d!~i;irilic\ of tlic (iee;i~~ic pc~icr;iI ~ I ~ C U ~ ; IIIIII. I I ~KPI..
II.
111riiI \ l o ~ / i . .18. 433 07.
Klle!. J . J.. R. W. Mctc;ilfc. and S .I 0.r t . i ~(10X61 1)irect \irrl~~latiiin\ ofchern~c;~ll~
rccrctlng turbulcnt niix~nglci>er>./ > I I I , \t. 111ti/\.29. 406 2 2 .
Rising H . (1089) Applications of chaos and dynamical systems approaches to mixing in
fluid. P h . D . 'rhesi5. I k p t of M~ithemnticsand Statistics. 1;niversity of Ma,sachusetts.
Amherst.
Ki\ing. 11.. ,rnd J M . Oitino (1085) L,'\c of horscshtrc furict~on\In n u ~ d~ n ~ t i n\!\tern\. p
U~tii.:IIII. /'/I!.\. SIC .. 30. 1600 (abstract oiil) ).
Robert\. f-'. A. (I9X5)T-ITcct\ ofa pcriodicd~>turbancet)11 \tructure and ~lil'iingin I U ~ ~ L I I C I I I
\hear I;rer> and \va!ies. PIi.1) . I h c ~ \ (;r.~duatc
. A e r o n . ~ u f ~ c aL;rboriitor~c\.
l ('~liforni;~
In.;titute of Tcchilolopx.
Rom-Kedar. V.. A . 1.eonard. and S. Wiggins (1990) A n an:rlytic;rl study oftransport.
mixing. a n d chaos in a n unsteady vortical flow. J. f'luitl Mecll. 214.337-94.
Ko\cn\\reig. K. I:. (198.5) t e r r o l ~ ~ t l r o t / ~ t ~ r ~Kewt ~ ~ iYork:
i ~ . \ . C'arnbr~dgelin~versit) Prc\s.
Ro\hho. A. (10761 Structure of turbulent \hear flows: a new looh. .Al'ltl J . . 14. 1349 57.
Roughton. I.. J . W . ant1 H. ('hirnce ( 19631 in /(,I/~tiiyirc~\ I)/ ,ir!l[itiic. c llrt~ii\rri.Vol. X.
A. Wc~>sbergcr.cd.. Ch'lp. XIV. NCH York . I~iterscience.
Rumscheidt. 1'. D.. and S. <i.Miison (1061) ['article motions in sheared su~pc~i\iorrz.
X11. D e f o r m a t ~ o n~ r l dhur\t of fluid drops in \hear and hyperbolic no\\\. .I. C'rtll~~rl
Sr.i.. 16. 238 61.
Russmann. H. (1970) c b e r invariante Kurvcn difkrcn7icrbarcr Abhildungen clnes Kreisrinpe\.
.\[I( /ir. ,,lh(rr/. H.r\\. Gijrri~iqc~i ,5101~1,P/IJ,\..Kl. 67 105,
Rxu. 11.-W.. H.-K. ('hiing. and I).-I. Lee (I9861 Crccping flows in recfangul;ir c:r\~tie\
uith tr;rn\l;rting top and buttom ~ a l l s nu~ncrical: study and flow \ i s u a l i / a t i o ~ ~fi'or[,iti~
.
.I. (./I~,III. l:r~i/.. 32, I77 85.
Saffman. 1'. <;. (19631 O n the fine-scale structure of Lector field, con\ccted h a t u r h u l e ~ ~ t
fluld. J. Fliritl . \ l l ~ ~ l r16.
. , 545 72.
Saliim. I-'. 41. A , . J. I?. Marsden. iind f'. f'. V a r a i ~ a(1983) Chaos and Arnold diffuhion
In dknarnical s)stem\. 1l.t.t. f i t i t i , \ , oti Girl iiii\ tititi S \ \ r ~ ~ t i<:AS-30.
i\, 697 708.
S'IX. J . F. (1085) Iranhport of small molecule> In pollmcr blend\: transport-morpI~oIog>
relationship>. 1'h.D. Thcais. Dcpt. o f C h c m ~ c a lEngineering. L!ni\crsi~qofMassachusct1s.
Amherst.
Sax. J . IS.. a n d J . M. Ottlrlo (1985) Influence of morphology on the transport properllcs
of polyhutirdiene p o l y s t ~ r e n eblends: experimental results. PoI!.ti~rt.. 26. 1073- 80.
Schlichting, 11. (1955) Ho~rtit/tir! Itr,,rr r l i c o r ~ .K c u Yorh: Mc<iraw-Hill.
Schoualter. W . R . (10791 .'llctllutiic.\ of tioti-.Yc~i\~/o~~itrr~ /lirril.\. Oxford: Pcrg:imon I'roh.
Schuster. II. (;. (1984) D(~it~rttiit~i\ric c./iuo\: rrti r t i i n i r i ~ r ~ ~ ~Wcinhcim:
n~ti. Phjsik-Verlag.
Scrrin. J . (1050) '1 he mathematical principles of classical fluid mechanics. in tl[rt~rlh~rt~/l
t1c.r f'/~i,.\ih,Vlll I . S. I.lijggc. cd.. pp. 125 263. Berlin: Springer-Vcrlag.
Serrin. J. (1977) Lecture5 for 8.430. Mathematical theory of fluid dynamics. 1:niverslt)
of Minnesota. Minneapolis (unpublished).
S e v ~ c k E.. 41.. P. A. Monson, a n d J. M . Ottino (19x8) Morphology a n d transport using
the lsing lartlce as a morphological dcscript~on.Chrni. ktly. Sci.. to appear.
Shirlnar. R. (1961 1 O n the heha\iour of Ilquid dispersions In mixing vessels. .I. I-'lttitl
. l f r ~ h . .10. 259 75.
Slnalc. S. (1967) I)iffcrcntiahle dqnairlicitl syslcm.;. Hitll. AIII. !lrrrli. Sol .. 73. 747 817.
Sliyder, W . l'..arid < i . A . Golclsie~n(1065) Arl anitlysl of fully dcvelopcd flow in ;ill
ccce~itric~ I I I I I ~ I ~.11('/11,
U~. .I.. 6-12 467.
S o h e . I . J . (I9S.i) I)ispcr\~onc a i ~ \ ~h!d \cp;~rationduring o\c~llator>flo\r. through
:I rurrc>\\ed cl1;1111ie1. (./ICJIII.b.~tq.Sc 1,. 40. 2120 34.
Solomon. T . H. ant1 J. 1'. (iolluh 119XX)Chaotic particle trun\port in time dependent
Kayleigh-Hhnard convection. Pl~j,.\.KIT. :l. to appear.
Sp;~l(li~ip. I). I*. I 19771 The l:S('IHO thtor! of turbi~let~t c o ~ ~ i b u s t i o1111pc,ri[rl
r~. ('o11~,!11,
.\li,<.ll.1;lt</,I)l~/lr.Re,/>.,so.t i 1.s 76 I.?.
Sp;~lcling.I). H. (107X;1)The inflilence\ of laminar tr;lnsport ; ~ n dchcn~icalkinetic5 ~ I I
the Ilrnc-nie;~rire:ictio~irate 111 :I l ~ ~ r b ~ ~l1;11iic. l e n tin I'ro(,. 1 ?!/I SVIII/I. I/III.)OII (~~III/III\~~O
pp. 431 40. The ombu bus ti or^ In\titute.
Spalding. I). 13. 11078b1<'liemic;~l reaction5 iri ti~rhulentfluitls. in I'l~t~.\ico-('lrc~~trrctrl
I I i i l r i ~ c l ~ ~ ~ \r:r ~IJrr~[..
~ r i c I.~r.rc11 OOrh hirrl~ilir~. ('or~/c~-r~lc.c,.pp. 321 28. I.ontlon: Adtance
l ' ~ ~ h l i c ; ~ t ~ ohnlsa. .
Sperlccr. K S.. and K. M . Wile! 11951 ) The mixing of \ c r \iscou\ liqultls. . I . ('olloicl.
5c.i.. 6. 133 45.
Stone, 11.. B. J . Hentlcy, and L. <;. Leal (1986) A n experimental study ol'transicnt
effects in the breakup of viscous drops. J. Fluid M e c l ~.. 173. 13 1-58.
Stone. f 1 . A.. and L. ( i . Ideal (1989) Relaxation and breakupof an initialy extended
drop in an otherwise quiescent fluid. J. I.71cid Mrcll. 198.39')-427.
Stuart. .I.'1.. (1967) O n finite arr~pl~tude oscillation in laminar mixing I;lycrh. J . Flrrid
.\lc~e~h.. 29. 417 40.
Swanson, P. I). and J . M. Ottino (1990) A comparative. compuntalional and cxperimcntal
study of chaotic mixing of viscous fluids. J. Rttici Mccll. 213. 277-249.
Swanson. P. D.. and .I. \I.Ottino (198.5)('hilotic mixing of tiscous liquids b e t ~ e e ~ i
ccccntrlc cylinders. Mrrll. 4111. \'IIJ\. Sot... 30. 1702 (abstract orily).
S ~ i r ~ n e !H . . L. (1985) Ohscr\ations of complex d n a r n i c s arid ch;ios. in F ~ r r ~ t l i r r ~ ~ ~ ~ ~ ~ r
/JVO/J/~,I~I.\ I I I .\r[rri\ri(~trl 1'1. li. (i. I). Cohen. ed.. pp. 253 80. Amsterdam:
~irr,c./~rr~ric.\
lilseiicr (the a p p e n d i of thi5 paper was published in 1083 in IJl~j.\ii~ir. 7D. 3 15).
Synge. J . I.. ( IYhO) ('lassical dxnamic5. in t1~111dhrrc.L tier I'l~,r.\iL.Ill I . S. I-liiggc. ed..
pp. I 225. Berlin: Spririgcr-Verlag.
Tadmor. Z.. and ('. <;. <;egos (1070) P r ~ ~ ~ c . i p01l rp01?.111ur ~\ proc.c,\vi~lll.Kcu Yorh: Wiley
Intcr~cic~ice.
Ttlnner. K. I. (19761 A test ptirticlc ;ippro;ich to flo\r classification for ii\coelastic fluids.
Al('11t; J.. 22. 910 14.
'Fanner. K. I.. and K . K. lluilgol 119751 O n a elassilication scheme for flou fields. K/rc,ol.
A(,rlr. 14. 959 -62.
Taylor. G. I. (I9341 The formation of emulsion.; in defirl;~blcfields of flow. I'roc.. Ko!.
Soc., Lond.. A146. 501-23.
Ta>lor. G . 1. 119641 <'onic;d free surf;~ces; ~ n dflu~ciin1crf;lces. in I'ro~.. llrlr I~rr.C O ~ I O .
/lpl~l.:Glcc.l~..pp. 790 6. Munich.
Tennckes. I I.. and J. I.. Lumle! ( 1 0 8 0 ) .A fir.\/ co[cr\cJill rtohtrlr,rrc,e. ('iimbridge. Mass. :
M1.T l'ress. 6th printing.
Thirring. W. ( lc)7X) A c.ordr\c, it1 tirur/~c~~n[iti~~~rl p/~~..\ic\. I : (~lci.~,~iccrI~ / J I I ~ I I I I.\j~.\roiir\.
~~,LI/
New York. Springer-Vcrlag.
I'orra. S.. R . ('ox. :rnd S. (;. Mason (1072) l'article motions in shc;~rcdsuspensions.
XXVII. Tronhlenl and steady deformation and burst of liquid drops. J. ('~lloitl. S1.i..
38. 305 41 I .
l'ownscnd. A . A . 11051 1 .She dlflusioll or heat spota in ihotropic turbulcricc. i.'rrJc'. KOJ.
S o . . 1 . t 1 1 4209. 418 30.
'rritton. D. J. (1977)I'l~y.\i~,trl \l~riili/ytrior~r~
3 . Kcu Yorh: Van Kostrand Re~nhold(rcprintcd
10x21.
Truesdell. C. A. (1954) '1'11~~ klt~rtilclri~~s
I!/ r~~rric,iry.
Bloomington: 1ndi;uiu Univers~tyPress.
trues dell.(^. A. ( 1977) (1 /;r,vr cotirv<,in ri~riot~ul c,ot~ri~~tit~tt~
t~~c,(,l~clt~ic,.\.
Purr 1: F'~it~di~ri~~,trr~r/
t,otli.oprs.New York: Academic I'ress.
Truesdell, <'. A,. and K. l'oupin ( I 960) The cl;rssic;rl field theories. in I l ~ r t ~ d h ider
i~~l~
Pl~ysik.111 1. S. Fliigpe. ed.. pp. 226 703. Rerlln: Springer-Verlag.
Ulbrecht. J. J.. and G . K. Patterson. eds. ( 1 985) Mixing i~/'liyuid,sh~rr~c~~I~i~n~culu~i~~~
New York: <;ordon and 13re;ich.
van Ilyke. M . (1982) At1 trlh~rti~ of fllritl rrlorior~.Stanford: The Parabolic Press.
Veronis. G. (1973) Large scale ocean circulation. ..lilr.. Appl. .\lrc.l1.. 13. 1 -92.
Veronis. G. (1977) The use of tracer in circulation studies. in 7'11~SPU. ).i)lunlr 6 . .Wtrrirtc.
F.. D. (ioldberg. I. N.McCave. J. J. 0'Rrien.and J. H. Steele.eds.. pp. 169-88.
tr~r~tlt.llit~q.
New York: Wiley.
Virk. P. S. (1975) Drag reduction fundamentals. /IIC'kE J.. 21. 625 56.
von Mises, R.. and K . 0 . Friedrichs (1971) Fluit1 tnrc.11ut1ic.s.New York. Springer-Verl;~g.
Wallerstein. G . (1988) Mixing in stars, Scirtlc,r, 240, 1743. 50.
Walters. P. (1982) An irltrodl~c,riot~ ro rr(qodic, rhrory. hew York. Springer-Verlag.
Wannier. G . H . (1950) A contribution to the hydrodynamics of lubrication. Qirurr. .4ppl.
Murk.. VIII. 1 - 32.
Welander. P. (1955) Studies on the general development of motion in a two-dimensional.
ideal fluid. 7i~llrt.s,7, 141 56.
Wickert. 1'. D., <:.W.. Macosko. and W. E. Kanz (1987) Small-scale mixing phcnonien;t
during reaction injection moulding. Polytnrr. 28. l I05 10.
Wiggins, S. (1988a) On the detection and dynamical consequences of orbits homoclinic
to hyperbolic periodic orbits and normally hyperbolic invariant tori in a class oiardin;rr>
diMerential equations. SIAM J . App1::Vurh.. ro appear
Wiggins, S. (1988b) Glohul hifurcutiot~.~ und chuos-ut~ulyticulmrlliods. New York:
Springer-Vcrlag.
Witten. T . A.. and L. M. Sander (1981) Iliffusion-limited aggregation. a kinetic critical
phenomenon. P11j.s. Rer.. l.r,rr.. 47. 1400 3.
Witten. 'I'. A,, and I.. M. Sander (1983) Diffusion-limited aggregation. Ph!,s. Her... B27.
5686 97.
Zweilcring. I'h. N. (1959) The degree of m~xingin continuous flow systems. Chetn. k t ~ y .
Sci.. 11. 1 15.
Author index
Kote: A subscript following a page number indicates an endnote; for example 16,. indicatcs
the second footnote of page 16. An italic number indicates a figure or table.