You are on page 1of 93

1

UNIVERSITE LIBRE DE BRUXELLES


FACULTE DES SCIENCES
SERVICE DE CHIMIE PHYSIQUE II

DETAILED INSIGHT INTO THE


COUETTE-TAYLOR INSTABILITY

THESE PRESENTEE POUR L’OBTENTION


DU GRADE SCIENTIFIQUE DE DOCTEUR
ES SCIENCIES PHYSIQUES

Alexandre S.F. POMPOSO

1985
2

Viento que soplas y que

Las ideas traes y llevas ; Verdad

en toda región y tiempo,

te doy las gracias desde lo más

profundo de mi alma; oblaciones quiero cantarte;

en ti encuentro la querida madre

que mi espíritu eleva a las aéreas moradas

en las que no podría yo respirar sin

la maravillosa esencia, que todos esos tus perfumes ideales,

en un sopor, conjuga el vocablo sin par de Amar.


3

Je remercie en tout premier lieu l’inappréciable encouragement spirituel


exercé par la mémoire de ma mère, et malgré l’éloignement géographique, la
présence permanente dans mon cœur de mon père et de mon frère Oscar.

Ma plus sincère reconnaissance envers le Professeur Ilya Prigogine pour la


qualité et finesse avec lesquelles il m’a appris le goût de la recherche scientifique,
ainsi que pour le rôle fondamental qu’il aura joué pour toujours dans ma vie.

Aux Professeurs P. Borckmans, G. Dewel et D. Walgraef je voudrais faire


part de ma gratitude pour la patience et l’attention qu’ils ont montré en suivant
mon travail.

Je serai toujours débiteur de Monsieur le Chanoine A. Vander Perre et de


Monsieur l’Abbé A. Léonard qui m’ont guidé et encouragé face aux obstacles qui
me semblaient insurmontables.

Finalement, je voudrais remercier ma chère famille Van Enst pour sa


présence et pour ses encouragements bienveillants.

_______________
4

CHAPTER I

INTRODUCTION

The problem of the collective behaviour of the systems is an important


question in physics as well as in philosophy. It seems that the description of such
studies of the bulk takes us to consider the transition from the microscopic to the
macroscopic worlds. We already know though that this question concerns
several branches in physics such as quantum mechanics, statistical physics and
so on.

Certainly, that problem is quite interesting in all sciences (noticeably


chemistry and biology); nevertheless we will concentrate our attention on fluid
dynamics keeping in mind that this question concerns some of the fundamental
concepts of the scientific thought.

Key words like self-organization, structure formation, patterning,


nonlinearity, etc., suggest many phenomena of wide interest. In fact, some of the
fundamental aspects of these topics have already been masterly analyzed and
studied by I. Prigogine and his group, applying on a great variety of problems in
several disciplines.

There are still many open questions in physics (and in biology), however
some are truly fundamental; for instance in physics, the problems of unified
quantum field theory and the arrow of time (this is to say, the well defined
direction in which time evolves) are a key part in the foundations of physical
knowledge.

The words we mentioned above have become the current language when
one talks about non-equilibrium systems, but that does not mean that they are
completely understood concepts; on the contrary, they are growing every day in
usefulness while looking for the answers to our questions.

When one observes nature, it seems that every time one intends to induce
a change in a system, it tries to oppose a resistance to it (a sort of a Lenz’s law in
a wider form). However one observes also that once the factor or quantity
inducing the change is maintained (this is to say that we insist in changing
something), then the system has to do something; either it adapts to the new
situation (according to the surrounding conditions), or it breaks up.

This way of describing what actually happens in nature is obviously too


much simplified; though, it shows already an important feature, namely the term:
to adapt.
5

At this point we think it could be enlightening the comment of G. Nicolis


and I. Prigogine (1); the statement we just wrote above is intimately related with
the second law of thermodynamics, and therefore the ingredient of irreversibility
is somehow there. The work of how is the arrow of time included in the
description of “adapting” systems is nowadays an utmost question in physical
science, and every field being connected with the study of life.

It is clear that such a problem is difficult to be tackled, since it is far from


the equilibrium state that we will want to be able to observe its behaviour.

So, the idea of self-organization comes out as an answer of the systems to


the variations incited from outside (this not always being clearly the case, since
“outside” refers to somewhere else, but it might be inside the system as well).
This brings us to think of the experimental fact of a structure formation (or
pattern) which generally means that the topological configuration of the matter
changes (therefore not necessarily observable). Actually, this is nothing more but
an immediate consequence of what we said before.

Now, how does the system know the pattern it shall choose? This is the
main question in the so called pattern selection problem. In other words, we
could say that once the structure of a system changes, we would like to know
which are the expected values (or value) of the parameters characterizing such
pattern. The interest in searching for a solution of that situation is directly
attached to the following question: how shall we introduce the time into the
evolution of systems going to turbulence?

This is the ultimate question we would like to be able to answer; however


we must look at first the previous one, namely the problem of selection of
structure.

A lot of work has been done in this direction; the present thesis will be
actually concentrating on a special aspect of such question in hydrodynamics,
however we must say that it concerns mainly biological and chemical problems.

Intimately linked to the difficulties in solving these questions, we find the


nonlinear character of the equations describing the fundamental dynamics of the
corresponding system. This has been discussed in many books and papers, both,
in general and in particular cases; nevertheless, the difficulties rising up are
usually the same. In this work we will find a beautiful example of the extreme
sensitivity (in which concerns the behaviour of the system) with respect to the
boundary conditions. By sensitivity we mean that a small change in the proposed
solutions has always large consequences in the further development; as stated it
could seem that we refer only to a theoretical situation, however this is true also
in the experimental treatment of the problem; hence the device at the laboratory
becomes very difficult to control. The basic ideas of turbulence point in the same
6

direction, namely that two points in the configuration space of the system, being
arbitrarily close to each other, they separate exponentially at any time.

The concept of dissipative structure has played a central role in


understanding the behaviour of continuous media, especially in which concerns
their evolution towards chaotic regime. There are some experiences in laboratory
which show (by means of visualization techniques in the case of hydrodynamics)
dramatically the symmetry breakdown in the regime, for instance of a fluid, as a
quantity externally controlled varies (this is to say, the control parameter).

Examples of such cases are: liquid crystals, chemical clocks (Belousov-


Zhabotinskii), superconductors under irradiation, thermal convection (Rayleigh-
Bénard, or RB), Couette-Taylor (or simply Taylor), flame fronts, snowflakes,
biological systems (morphogenesis), etc.

The physical and biological worlds provide us with an overabundance of


systems presenting instabilities leading to higher order states. Therefore, the
understanding why order may appear spontaneously and which structures are
selected (as we already pointed it out) among a large manifold of possibilities has
become a major branch of research, both theoretically and experimentally.
Indeed, superposed to this situation, there is the fact that after a finite number of
instabilities (bifurcations) it seems the systems give up and “fall” into the chaotic
behaviour, if we keep increasing the control parameter from outside.

In spite of the remarkable amount of published works in recent years, we


shall make a distinction between the kind of advances in the experimental and
theoretical approaches. The first one acquires a global knowledge of the
properties exhibited by the systems, in the sense that it is very difficult to observe
with a great detail the microscopic events, being those the responsible (or at least
it is likely to be so) of the cooperative behaviour. On the other hand, the
theoretical strategy is another one, namely it can approach with more detail to
the “fine structure” of the configuration of the system. As a consequence, the
theoretical knowledge is in some sense narrower than the experimental one.
Certainly, it is a common goal to come with compatible answers to the different
questions, and most of the efforts are being made in such direction.

There exist many possible motivations for studying this kind of problem;
among others, we find a prime motivation as we want to study fluids going into
turbulence, so that we can see throughout their analysis, the “basic events”
leading to what seams to be chaotic; in other words, we are interested in looking
at the criteria (if they exist) these systems follow, which determine the observed
pattern. This is especially important in the threshold of instability; because of the
length and time scales separation in the dynamics undergoing such instability as
compared to other macroscopic variables, analogies with phase transitions have
often been stressed.
7

On the other hand, dissipative structures are the consequence of the


exchange of energy (and matter) under conditions far from equilibrium, as stated
by Glansdorff and Prigogine (2). It is not difficult then, to see that most of the
structures we observe in nature belong to this kind.

Let us have now a brief overview of the general situation on this subject
(Chapter II will consider in a more detailed manner the different aspects of the
Couette-Taylor system).

How and why do certain hydrodynamical systems (the atmosphere, the


oceans, the sun, etc.) loss progressively their organized character and become
turbulent? This question as it is formulated is too large to be considered as
unique; indeed, it must be seen from several points of view; so, for instance such
a question takes us immediately to the problem of the so called developed
turbulence, however we are rather motivated by the “how”, this is to say by the
way systems evolve in order to attain a turbulent regime. We will therefore stay at
the level of the analysis of the steps that some devices must follow as the control
parameter is varied. These steps are called bifurcations, since the curve of the
velocity parameter (this is some component of the velocity vector, for example)
truly bifurcates at certain values of the externally controlled quantity.

When we say “bifurcation”, we are implying several things, like a change


in the symmetry of the system to be described, and this one usually refers to the
geometry of the system (at least in which concerns fluid dynamics).

Two review works must be noticed here. The experimental aspects of


hydrodynamic instabilities have been studied during the Solvay Conference in
Brussels (1977), and drafted by E.L. Koschmieder (3). During the same
Conference, R. Graham analysed the onset of cooperative behaviour in non-
equilibrium steady states; he presented an overview of the instabilities found in
quantum optics, plasma and semiconducting devices, chemical, biophysical and
hydrodynamic systems (4).

Those works and others (all of them being a collection of examples of


instabilities) show the common features of continuous instabilities of non-
equilibrium steady states, as well as the general analytical approaches for
studying them (5)(6)(7)(8)(9).

Concerning hydrodynamic systems, the classical paper by S. Kogelman


and R.C. Diprima (11) is an excellent example of a presentation of the theoretical
problem of stability of spatially periodic supercritical flows; the work of Eckhaus
on the development and stability of periodic solutions is extended and applied to
the first bifurcation of the viscous flow between concentric rotating cylinders
(Taylor). This extension is a major step in understanding the pattern selection in
supercritical flows. An important aspect of this, is that immediately after the
bifurcation point, the system has a continuum of possible values of the wave
8

number (in the Taylor problem, for example) to choose; however the width of
this band diminishes as a consequence of sideband perturbations resonating with
the first harmonic of the fundamental mode of the periodic flow to mutually
reinforce each other, leading to an instability. For further bifurcations of the same
system see P. Chossat and G. Iooss (12).

We will be pointing out later, the role of the non-equilibrium fluctuations


in patterning; the models we start with are of the hydrodynamic type (i.e.
Navier-Stokes equations), leading to a much more general description as it
relates a greater variety of phenomena.

It has been shown to be a useful approach in the study of critical


phenomena, the potential function (Ginzburg-Landau, or Brazovskii potential)
because it furnishes the system with a more transparent physical idea, and
moreover, in terms of them stability and instability analysis can be interpreted
more easily, and finally it can allow us to extend the use of variational techniques
to non-selfadjoint problems (see local potential in (10)).

This is to say, potentials show more explicitly the dynamics of the system
(in some particular problems, it remains possible to construct “ad hoc” true
potentials, which may then be treated by the usual variational techniques; this is
the case of the trimolecular model which corresponds to the “Brusselator”
reaction scheme, as it has been demonstrated by D. Walgraef, G. Dewel and P.
Borckmans) (13).

This last point constitutes one of the motivating factors for the present
work; we need to include the local fluctuations in the description of systems,
where a coherent behaviour takes place at a macroscopic level. For the
hydrodynamic case, we will see how we can explicitly appeal the bulk
mechanism of transport (like viscosity and parameters as the Taylor or Reynolds
number) through the so called hydrodynamic equations; in which concerns the
macroscopic effects, boundary conditions play a central role in the description.

At this point let us start concentrating on the hydrodynamic systems. As


there is a huge variety of these (just as the consequence of the changes in
boundary conditions, and depending on the nature of the control parameter, if
there is only one), we shall choose only some.

The criteria for making the choice can be seen throughout the history of
instabilities in fluids; notably, we have interest in considering the systems
exhibiting the basic features of the bulk behaviour. The Rayleigh-Bénard and
Couette-Taylor instabilities, coming out from temperature and angular velocity
gradients respectively, constitute such fundamental kind of systems. There are
surely other flows presenting unstable properties and pattern formation (like the
case of Poisseuille, jet, Stokes and so on), and we should say that most of those
flows were firstly studied in engineering problems, this is to say in quite
9

practical situations. For instance, the Couette-Taylor instability was detected by


the first time while measuring the viscosity of fluids in lubrication research. But
it was later on, thank to the reasons we said before, that physics and chemistry
(mainly the first one) came to be interested. Therefore, even if still many
unsolved problems in fluid dynamics keep an important place in research, just
few of them let us foresee the basic properties of liquids, say.

The problems arising from the study of these systems can be tackled by
the detailed analysis (experimentally and theoretically) of the relations among
temperature, angular velocity, viscosity, etc.

For the Bénard problem, the geometry of the device is crucial, since the
selected pattern is highly conditioned by the boundary conditions; moreover, it is
relatively easy to vary experimentally the shape of the apparatus containing the
fluid. Research work has been devoted to such aspects, namely comparing the
patterns appearing for rectangular, cylindrical, spherical boxes; in the
bibliography we give a list of references where most of the experimental papers
include pictures of the selected structures (14)(15)(16)(17)(18)(19)(20)(21)(22).

Those patters can be classified according to the shape of the boundary of


two successive vortices; so we can have rolls, hexagons, crossed-rolls, rolls in
zigzag, concentric rings and so on. The region of stable convection rolls (in an
R − P − k space 1 ) can be seen in a paper by Busse (23) (see fig.1). This is a
complicated shape whose analytical expression is quite unlikely to be obtained
from a simple mathematical equation.

In figure 1 the thick curves represent computed stability boundaries for


several types of structures: OS = oscillatory, SV = skewed varicose, CR = cross-
roll, KN = knot, and ZZ = zigzag. The other curves represent approximate
interpolations from results of Busse and Clever (23)(24).

This solid was obtained following standard Galerkin methods. As it


should be expected the region of stability becomes smaller as the Prandtl number
(i.e. the ratio of the kinematic viscosity and the thermometric conductivity) tends
to zero. The explanation of the different instabilities appearing in the figure
would take us away from our subject, so the reader is urged to look at the
detailed explanations of references (23)(24).

Other instabilities are described on systems we could name


“sophisticated”, such as combinations of thermal and angular velocity gradients,
with conical and spherical geometries.

1
R = Rayleigh number ( gαΔT kν ) d 4
P = ν k ; k = wave number; ν = kinematic viscosity
α = coefficient of volume expansion; k = thermometric conductivity
ΔT = temperature gradient; g = acceleration of gravity.
10

Figure 1.

From a theoretical point of view the evolution of our understanding has


been different, and in some sense we can even say that slower, since as we
already mentioned it the nonlinear character of the equation forces us to look
closer into the details of the flux; indeed we want to determine the velocity field
completely; however, problems like the transition to time-dependent convection
have been shown to come out as a consequence of the momentum advection
terms in the equations of motion.

Not even going to later instabilities, one must face problems of pattern
selection (translated as wave number selection) close to the first threshold (good
examples are Siggia and Zippelius (25)(26)(27), and especially more recently M.C.
Cross (28)(29)(30)(31)(32), who looks more closely to the effects of boundary
conditions on the rolls’ structure). Again, an important aspect which might seem
clearer for the Bénard problem than for other instabilities is the scaling procedure
which is the expression of the fact that once a symmetry breakdown is done, the
11

description of the evolutions of the system has to be performed with a new time
and distance scales (this is in fact one of the ingredients of the renormalization
group theory utilized in phase transitions theory).

Unhappily, we will be obliged to use the word “scaling” also for the
simple procedure through which one turns the equations to be dimensionless;
but since it is the first meaning the only one having a relevant physical
importance, we will make no distinction in the vocabulary (if otherwise it will be
remarked).

This procedure becomes more intricate (as well as all the formalism) as
soon as we consider large boxes, where defects appear on the rolls’ pattern (26),
and competition events have place between surface effects (favouring normally
placed rolls to the sidewalls) and bulk effects (favouring straight parallel rolls).

The description of the dynamics of the defects in a convective system is


made under several approximations (26); there is no strong argument why free-
slip and rigid boundary conditions should give similar results except for very
small shifts from the critical value of the Rayleigh number (where an amplitude
expansion is valid, as we will see later). However for large Prandtl number, this
shift can be much bigger (and no physical reasons have been given so far).

In order to study the pattern selection problem, a stability analysis must


be done about the bifurcated solution, rather than stability of the basic solution
(steady state) which holds prior to the occurrence of the bifurcation. This will
show to be an important disadvantage in the Couette-Taylor system, as we
ignore the exact form of the solution from the first bifurcated state (33).

We do not want to go much deeper on the Bénard problem in this


introduction, but just to the stage where the analogies with the Taylor problem
are helpful. Such is the case in which concerns the time evolution of the quantity
we call “amplitude” (it is actually thanks to this manner of working out the
problem that it is facilitated the way of studying further bifurcations, as it could
be the slow amplitude modulation). A.C. Newell and J.A. Whitehead, Segel and
M.C. Cross have obtained, starting from the hydrodynamic equations, an
amplitude relation which takes into account both rigid and free upper and lower
boundaries. In Chapter IV we perform the equivalent calculation for the Couette-
Taylor problem, underlining the differences with Bénard (34)(35)(30).

Similarly, concerning the phase diffusion problem (see Pomeau and


Manneville, etc. (36)(37)(31)), this situation refers to variation of the wavelength
immediately after the onset, and it is shown the diffusive character of the
phenomenon. Chapter IV will develop the same study for the Taylor case. For
the Rayleigh-Bénard instability there have been some discrepancies among
several approximations about the position of the wave number selection
principle (at the marginal instability or elsewhere); however for the Taylor
12

problem we do not have yet an equivalent discussion; the stability regions for the
second bifurcation (wavy vortex flow) were only found few years ago by G. King
and H. Swinney (38), and it is quite recently that D. Walgraef (39), Hall (40) have
tried to give a theoretical support to the experimental finds.

Another way of looking at the problem of pattern selection has been


presented through the numerical simulations at the computer; notably the
stability analysis for two and three-dimensional convective arrangements (H.S.
Greenside and M.C. Cross), the pattern formation near the onset (H.S. Greenside
and W.M. Coughram) where the long waiting relaxation times are exhibited (see
also M.C. Cross and A.C. Newell) (41)(42)(43)(44).

Those direct numerical simulations were performed to investigate the


behaviour of several models and to compare them with the real fluid systems
(21). The hierarchy of models presented there is capable of looking at the effects
of rigid sidewalls and the textures of curved rolls usually encountered on the
onset of various instabilities and the subsequent evolution may then be
investigated.

It is difficult to propose models which reproduce the long wavelength


behaviour, having at the same time a qualitatively correct short wavelength
structure. In (41) we can find some of these models, describing the growth of
some convective instabilities. In which concerns (42), the simulations show that
there is a two-dimensional scalar equation that can be successfully used to study
the three-dimensional dynamics of the five Boussinesq equations close to the
onset, and they illustrate the strong influence of the boundaries of the device in
determining the spatial stationary pattern. (43) shows that because the number of
defects created during the relaxation process becomes of order L (the lateral
dimension of the system), their theory does not predict the onset (i.e. the
formation of a pattern), which remains steady on the horizontal diffusion time
scale; however their wave number selection principle is consistent with
experiment (only if the rolls stay perpendicularly oriented with respect top the
sidewalls).

Finally, work has been done (mainly in the experimental context) on


turbulent behaviour when a fluid layer is heated from below; the research has
been followed on several geometries, in small boxes (containing only two rolls),
observing phase locking 2 , intermittencies 3 , etc.: concepts which would demand a
complete explanation but would take us away from our objectives
(18)(33)(23)(45)(46)(47)(48)(49)(20).

Concerning the Taylor problem, this is to say the stability of a viscous


fluid contained between two rotating cylinders, we will not say too much in this
introduction about the developments (as we did for the Bénard problem); we will

2
This is to say, the situation when the ratio of two modes frequencies stays fixed in time.
3
It refers to the fact of the stochastic appearance of bursts of chaotic behaviour, in the velocity spectrum.
13

rather refer to the respective researches all along the present work. Nevertheless
what we can do here is to say a few words about the fascinating and motivating
aspects of this system.

It is already motivating to read the original paper (1922) by G.I. Taylor (50),
where he practically solves the problem concerning the first instability to
axisymmetric disturbances, both theoretically and experimentally. He even
describes the “appearance of vortices when they begin to break up immediately
after their formation”, which will correspond forty-three years later, to the wavy
vortices recognized as an instability of the Taylor vortex flow by Coles (51).

The first theoretical works performed in a systematic way, on the growth


of Taylor vortices in 1962 by A, Davey (52), and in the well known book on
hydrodynamic and hydromagnetic stability by S. Chandrasekhar (1961) (14).
Their calculations are a pioneering way of looking at the problem of the
additional torque required to keep the cylinders in motion at given speeds, when
the vortex motion occurs (Stuart) (53), and they also distinguish the different
predictions for the onset of the Taylor vortices according to the ratio of the gap
width (i.e. the difference of radii) and one of the radii or the mean of them; if this
ratio is very small we say that we are in the small gap approximation, and it is all
along this work that we will be dealing with.

Even if many experimental observations have been made for every


possible gap, analytically most of the work has been performed about the small
gap approach, not only in which concerns the first instability but going further to
wavy and modulated wavy vortex flows. An excellent review of the whole work
done until 1980 is given by R.C. Diprima and H.L. Swinney and this on all
instabilities up to the appearance of turbulence (54).

The case of both cylinders rotating can be considered and in fact this has
been largely studied experimentally by the group of The University of Texas al
Austin; however, in order to stay faithful to the goals we fixed at the beginning
of this introduction (this is to say, in search of the basic behaviour of the system)
we will only consider the case of the inner cylinder rotating, the outer one
staying at rest. 4

It surely is the more striking fact, especially the wavy vortex flow
behaviour, regarding the different ways to approach a given state; let us
characterize it by k / m where k is the number of Taylor vortices and m the
number of azimuthal waves; Coles (51) found the sequence 28/0, 24/5, 22/5,
22/6, 22/5, 22/4, 22/0, increasing the rotating speed very slowly, and also found
that the final state usually depended on the way he approached the
corresponding angular velocity.

4
Taylor himself showed in his original paper that the case of the outer cylinder rotating and the inner one
at rest, is always stable.
14

At first, it would seem from this that it is practically impossible to get a


complete explanation of the phenomenon; however, it must be said that,
fortunately nowadays we have access to much more refined experimental
techniques (especially on visualization of fluid motion) and the stability of those
states can be studied more systematically; as we already said before, such has
been the research performed by G. King and H.L. Swinney (38); but still the
curves they have found are a challenge for the theorist (see however the very
recent work by D. Walgraef (39)).

Already with Taylor vortices we find a problem of multiplicity of states


(see the discussion on this subject by T. Brooke Benjamin and T. Mullin (55)(56)).

For the wavy vortex flow, the first calculations were performed on a non-
axisymmetric disturbances’ scheme by E.R. Krueger et al. (1965) (57) and by A.
Davey et al. (1968) (58). They include an azimuthal wave number m , which our
analysis will matter in Chapter IV.

When it is continued to study the diffusive character of the vortices (like in


Bénard), the same analysis leads us to the domain of the phase diffusion
formalism. As it has been pointed out in (59), it is at that moment that the length
of the cylinders can play an important role; in that paper we can see how the
hydrodynamic fluctuations must be considered, and we do the same in Chapter
V; the work on the phase diffusion problem takes the scaling proposed by
Tabeling (60) (which is slightly different from the one suggested by H. Brand and
M. Cross (61), and P. Hall (40)).

From this perspective, R.C. Diprima and J. Sijbrand (62) propose a


mechanism of interaction of axisymmetric and non-axisymmetric disturbances
(some sort of competition event), which might be an open door to explore what
seems right now a complicated situation, of finding a pattern selection criterion.

There are still much more experiences that we have skipped in order to
avoid a too long presentation. Notably, the group at Eugene, Oregon (R.J.
Donnelly) has performed very accurate measurements of the velocities
concerning the flux inside the system, and during the last Conference on the
sixtieth anniversary of Taylor’s work (55), it was possible to compare all the
theoretical and experimental advances (with an international character).

The best thing to do is to keep an optimistic regard on the problem;


however this optimism should be rational, in the sense that even if it is true that
the mathematical and conceptual situation is no easy to tackle, the increasing
accuracy in spectral velocimetry techniques, and the everyday better efficiency of
computing machines should encourage us to think of a new way of looking at
the things. Moreover, there is a great variety of hydrodynamic instabilities which
might furnish additional data to be considered in the Taylor case; for instance,
Bénard cells in cylindrical containers, the Stokes instability (two plates moving
15

with respect to each other and a fluid in between), Poisseuille flow along a
varying cross section tube, and so forth, are problems which can bring some light
(because of the possible or eventual analogies) onto the different bifurcations of
the Couette-Taylor system.

Let us now draft the general plan of this work. After the present
introduction, Chapter II will review the general and basic ideas in fluid dynamics;
so, the hydrodynamic equations and some problems about boundary conditions;
a brief look on the series of instabilities all the way up to the turbulence
concludes the chapter.

Chapter III comes out with the instability analysis, and this for both
axisymmetric and non-axisymmetric disturbances. We will not present there the
full detailed calculations, but we give the expressions for the critical Taylor
number (control parameter) as it depends on the axial and azimuthal wave
numbers.

As we already said, Chapter IV considers the amplitude equation as well


as the phase diffusion calculations (devoting a section on the analytical form of
the coefficients appearing in those expressions).

And finally, Chapter V will consider the contribution of the


hydrodynamic fluctuations to the onset of the instability; equation (V.22) marks
the end of our calculations and what comes out afterwards is nothing more but a
consequence connecting with results which can be found elsewhere (see the
references at the end).
16

CHAPTER II

OVERVIEW ON THE PROBLEM

It has already been pointed out by many authors that a general theory of
turbulence has not yet been developed. One might be even tempted to ask why it
is so. In which concerns the fluids, their dynamical behaviour is completely
described by the hydrodynamic equations; nevertheless, a simple look into this
system of equations shows its non-linear nature.

The central fact of this chapter will be to pose what we call the
hydrodynamic problem, and more specifically the main features in describing a
system like Couette-Taylor’s one.

The variety of the phenomena taking place in the study of most of


hydrodynamic situations is wide, then the efforts in research have been mainly
oriented onto two different problems: the Rayleigh-Bénard and the Couette-
Taylor (or simply Taylor) instabilities. This is in order to have experimentally
reproducible results, and simplified mathematical solutions.

When we say this, we are including the work on different approaches of


people for about a century. As experimental techniques and simulation
procedures in computer have greatly evolved, the research on fluid dynamics (as
in many other branches) has become quite active, encouraging the theorists to
find analytical explanations of the proposed phenomena, starting from basic
physical principles. A short review of contributions to the Couette-Taylor
problem is done below (and it shall be completed with Chapter I).

i) The hydrodynamic problem.

As we already said before, the hydrodynamic equations (which we obtain below)


describe the dynamics of a fluid. This means that all the hydrodynamic
information contained in the average behaviour of a liquid, say, is contained as
well in the Navier-Stokes equations and the continuity equation. Let us recall
that the works of the same Taylor, Rayleigh, pioneered the fundamental
hydrodynamic stability theory. The interest of such treatments is obvious under
the lights of the research of the formal basis of fluid phenomena.

During the last three decades, experimental techniques have been


enormously developed, in which concerns the accuracy of measurements in fluid
dynamics. At the same time, research works have appeared developing further
17

the already known mathematical tools, applying them to different systems and
clarifying many aspects of such systems by comparing their results with those
directly obtained with experimental devices. We can mention here the names of
Chandrasekhar, Davey, Stuart, Diprima, Coles, Newell, Whitehead, Gollub, Swift,
Hohenberg, etc. (see references). The recent work of Greenside, Cross, Ahlers (63),
Markus (64) and others, show the up-to-date level in computer simulation in the
pattern selection problem in fluid dynamics (i.e. the problem of finding the
criteria the systems follow in order to chose a structure, when we try to push
them towards a chaotic behaviour).

As we will be seeing later, different approaches like the amplitude


equation, phase diffusion (Pomeau and Manneville), being themselves issued
from the hydrodynamic equations increase our knowledge of the situation.

The problem of obtaining the Navier-Stokes equations has been quite well
presented in many books and papers (14)(15)(9)(2); however we will make a brief
review of the main physical ideas behind these equations.

To start, let us remember that the Navier-Stokes system refers to the


evolution of the velocity vector u with the time t , as it is related with the
gradients in pressure (and temperature for problems like Bénard’s one). Since the
density variable is also in such expressions, its variation with time can be
established, and it gives the continuity equation.

We consider in general a viscous fluid; the conservation of mass can be


expressed as follows. What we actually have to take into account is the balance of
mass going in and out of a volume (which is fixed), depending on the velocity of
the fluid (being this one not necessarily the bulk velocity of it).

The mass contained in a volume V is given by

∫V
ρ dV (II.1)

where ρ is the mass density. On the other hand, the rate at which the fluid flows
out of this volume element is

− ∫ ρ ui dSi (II.2)
S

where S is the surface vector associated with the volume element (and we have
used the summation convention, where i denotes the three components of the
velocity). Therefore, the rate of change of mass (eqn. II.1) has to be equal to
equation (II.2), this is to say,

∂ t ∫ ρ dV = − ∫ ρ ui dSi
V S
18

and by Gauss’ theorem, this implies

∂ t ∫ ρ dV = − ∫ ∂ xi ( ρ ui ) dV
V V

thus

∂ t ρ + ∂ xi ( ρ ui ) = 0 (II.3)

this is the continuity equation. Notice that for an incompressible fluid (i.e.
ρ = constant in time and independent of the position), (II.3) is reduced to

∂ xi ui = 0 (II.4)

thus called, the incompressibility condition.

The expression for the evolution in time of the different components of the
velocity, can be obtained in a similar way, but this time it is the balance on the
volume element by means of the conservation of momentum. We can then write

∂ t ∫ ρ ui dV = ∫ ρ Fi dV + ∫ Pij dS j − ∫ ρ ui u j dS j (II.5)
V V

where the left-hand-side denotes the rate of change of momentum in V , and the
right-hand-side is the sum of the external forces acting on the elements of the
fluids, the normal stresses acting on S , and the momentum flowing out of V .
The stress Pij which acts on the direction x j (per unit area) is assumed to be
linearly proportional to the increase of strain,

2
Pij = 2ν eij − pδ ij − νδ ij ekk (II.6)
3

where ν is the fluid viscosity coefficient, p is the isotropic pressure, and

eij ≡
1
2
( ∂ j ui + ∂ iu j ) (II.7)

(notice that ∂ k ≡ ∂ ∂xk ). Hence, from (II.4), (II.5) and Gauss’ theorem one obtains
the more familiar form

⎡ ⎤
ρ∂ t ui + ρ u j ∂ j ui = ∂ j ⎢ν ( ∂ j ui + ∂ i u j ) − ν∂ k uk ⎥ − ∂ i p + ρ Fi
2
(II.8)
⎣ 3 ⎦
or for an incompressible fluid
19

ρ∂ t ui + ρ u j ∂ j ui = ν∇ 2ui − ∂ i p + ρ Fi (II.8’)
As we said before, the time evolution of the velocity vector u not only can
it be related with the gradients of pressure but with those of temperature as well.

Since such a situation does not concern our problem, we will be dealing
with expressions (II.3) and (II.8). However, notice that a combined problem of
Rayleigh-Bénard and Couette-Taylor can be foreseen (23); in that case the
temperature enters into the equations; we stay with the more basic Taylor system.

Going back to equations (II.3), (II.8), let us say again that their relevance is
in the fact that they have all the physical information contained in the fluid.

When we talk about a volume element of the fluid, we are thinking on the
hydrodynamic approach, this is to say a very small volume (infinitesimal), but
with associated dimensions much larger than the molecular distance. So, when a
point of the fluid moves, that does not mean that it is actually a molecule which
moves, but a big number of them, constituting though a hydrodynamic point.

It is also important to remark that in the hydrodynamic equations the


influence of the geometry of the system is contained, provided that we write
down the boundary conditions. In order to make a general statement, we can say
that in a region R of physical space, the velocity vector u is given an the
boundary of R.

Before getting into the next section, already concerning the Couette-Taylor
system, let us mention that in the way we obtained the hydrodynamic equations,
we made no special assumptions on the constancy of the coefficient of viscosity
(taking into account the equation for the temperature, we should add a comment
on other transport coefficients). Seemingly, we can say about the density that it is
essentially constant for most of the practical cases. One has to be careful though
concerning the external forces term, since this one produces accelerations larger
than those resulting from the term ρ u j ∂ j ui .

ii) The Couette-Taylor instability.

We will start the study by the stability analysis of viscous incompressible flow
between concentric rotating cylinders, as the speed of the outer one is zero.

First of all let us have a look on the experimental aspects. In figure 2 (p. 21)
we show the geometry of the system (idealized for the mathematical treatment).
It was on this kind of apparatus that Couette and Mallok were working, in order
to find the value of the viscosity for the fluid contained between the cylinders.
This, in turn was made by measuring the torque of the quartz string placed along
the cylinders’ axis. They observed that from a certain critical angular velocity of
20

the inner cylinder, the torque suddenly increased. So, they thought it was due to
the fact that the fluid was already starting to become essentially turbulent.

G.I. Taylor showed later that the fluid was not turbulent, but instead was
self-organizing in a torii structure. These vortices (as they are called) become
wavy afterwards, when we increase cylinder’s speed. It was the work of Coles
(51) who recognized by the first time this as a well defined instability.

As flow visualization techniques have been improved, more details of


these structures have been found. HL Swinney found by means of power spectra,
that passed the wavy vortex flow, a modulation frequency appeared before the
onset of weak turbulence. All this happens when the outer cylinder is not
moving (this is the only case we will consider here).

Several important properties, concerning the value of the critical velocity,


depending on the variations of the geometry (say, the width of the gap, the
height of the cylinders, the ratio of them i.e. the aspect ratio), have been
experimentally studied with different techniques by Cognet, Bouabdallah and
Ait Aider (65), and Donnelly (66)(67). They have also observed the dependence
of the amplitude of the vortices with the aspect ratio.

Concerning always the Taylor vortices, Koschmieder has photographed


the different amplitudes of the vortices depending on the way we attain the
critical value of the controlled quantity.

In a recent work, G. King (38) has determined by Doppler velocimetry


techniques the stability curves for wavy vortex flow (there are several, since
there exist different azimuthal wave numbers being selected). He has also
measured the velocity at which these waves propagate. Nowadays, there is
interest in analysing the behaviour of the vortices in devices with a more
complicated geometry (square cylinders for instance).

Theoretical work has also been done, however we should say that it is in
an earlier stage, compared with the experimental one. The several contributions
on this line will be pointed out along this work. We will present here a quick
overview of the physics of the first instability.

It was G.I. Taylor who performed both, the theoretical and experimental
work for the first time, in a systematic way (close, more or less, to the idea we
have actually) concerning the stability of a fluid contained between two rotating
cylinders, at the point about which the symmetry changed by the first time.
However, previous work on the subject (i.e. Rayleigh) on the stability of an
inviscid fluid moving in concentric layers had been performed.

The remarkable point in Rayleigh’s work is the analogy he makes (for


axisymmetric disturbances) with the stability of a fluid of variable density under
21

the action of gravity; so, the centrifugal force plays the role of the gravitational
one. This leads to the so called Rayleigh’s criterion, which establishes that for an
inviscid fluid, the square of the angular momentum increases with the distance
from the axis.

The first experimental aspects were investigated, as we already said, by


Couette (and Mallok). It was actually the seeming difference between some of
these experimental results, and Rayleigh’s predicted theoretical behaviour,
which encouraged Taylor to construct a device where both inner and outer
cylinders were able to rotate independently (remember that Couette and Mallok
were studying the effects of viscosity precisely). He calculated and measured the
critical angular velocity at which the system became unstable, getting a
remarkable agreement.

Let us describe in a simplified way, the bulk motion of the fluid, before the
appearance of the first instability. The experimental arrangement can be seen in
figure 2. We call Ω1 and Ω 2 the angular velocity of the inner and outer cylinders,
respectively.

Figure 2.
22

Following Chandrasekhar’s notation, we define (14)

Ω2 R1
μ≡ ; η≡ (II.9)
Ω1 R2

Our system has obviously a cylindrical geometry, being natural the use of
cylindrical coordinates; hence, if u , v and w denote the radial, azimuthal, and
axial components of the velocity (this one being associated to a point of the fluid
in the sense we already explained), equations (II.3), (II.8) take the form:

v2 ⎛ 2 u ⎞ ⎛ p⎞
∂ t u + ( u ⋅∇ ) u − = ν ⎜ ∇ 2u − 2 ∂θ v − 2 ⎟ − ∂ r ⎜ ⎟ (II.10)
r ⎝ r r ⎠ ⎝ρ⎠
uv ⎛ 2 v ⎞ 1 ⎛ p⎞
∂ t v + ( u ⋅∇ ) v + = ν ⎜ ∇ 2 v + 2 ∂θ u − 2 ⎟ − ∂θ ⎜ ⎟ (II.11)
r ⎝ r r ⎠ r ⎝ρ⎠
⎛ p⎞
∂ t w + ( u ⋅∇ ) w = ν∇ 2 w − ∂ z ⎜ ⎟ (II.12)
⎝ρ⎠
u 1
∂ r u + + ∂θ v + ∂ z w = 0 (II.13)
r r

where the differential operators can be written as

1 1
∇ 2 ≡ ∂ r2 + ∂ r + 2 ∂θ2 + ∂ 2z (II.14)
r r
v
u ⋅∇ ≡ u∂ r + ∂θ + w∂ z (II.15)
r

In order to be able to write the stationary solution we allow

u = w = 0 , v ≡ V (r) (II.16)

where V ( r ) is a purely radial function. From (II.10) and (II.11) we have,

V2 ⎛ p⎞ ⎛ p⎞
− = −∂ r ⎜ ⎟ = − d r ⎜ ⎟ (II.17)
r ⎝ρ⎠ ⎝ρ⎠
⎛ 1⎞
ν dr ⎜ dr + ⎟V = 0 (II.18)
⎝ r ⎠

along with the boundary conditions

(II.19)
23

therefore

B
V = Ar + (II.20)
r

such that

⎛μ ⎞
⎜ η 2 − 1⎟
A= ⎝ ⎠ ⋅η 2 Ω
1 −η 2
1
(II.21)
R (1 − μ )
2

B= 1
⋅ Ω1
1 −η 2

As we said before, we are interested in the case when Ω 2 = 0 , which


implies after (II.9) that μ = 0 , hence expressions (II.21) become

η 2 Ω1 R12Ω1
A= ; B= (II.22)
1 −η 2 1 −η 2

Equations (II.20) and (II.21) describe the so called Couette flow, and it is
about this hydrodynamic regime that we will perform the stability analysis
afterwards. However, we can already say something about the first instability,
appearing when we increase Ω1 .

We could try to explain the appearance of such instability in terms of a


competition event between two facts; first, the centrifugal force acting on each
fluid element, tries to “expel” it far away from the cylinder’s axis, and secondly,
the viscous stress of the fluid opposes a resistance to such an expansion of itself.
One could even add the effect of the geometrical configuration (curvature of the
walls).

We cannot go too far with such a hand waving argument; however it


already illustrates the idea of the possible physical mechanism, driving the
system to the first bifurcation.

Anyhow, with this kind of reasoning we cannot say practically anything


about the pattern selection problem. We will perform a systematic study in an
analytical way, and will see how the perturbation analysis leads us to a new
regime which is periodic in the axial coordinate and axisymmetric (pattern).

Besides the assumption μ = 0 , we will simplify also by making η ≈ 1 ,


which is the so called small gap approximation. As it will be clear later, this
simplification will play an important role all along the calculations.
24

Finally, let us say that since we have made μ = 0 , the natural control
parameter becomes Ω1 , leaving all the geometrical quantities fixed. We express
this quantity as

4 AΩ1
T≡ d4 (II.23)
ν2

the Taylor number, where

d ≡ R2 − R1 (II.24)

is the gap (or the interval). The number T is always positive (since it is always
true that η 2 > 0 ). In the small gap approximation R1 ≈ R2 , so ( d R1 ) is very small,
therefore the Taylor number becomes,

⎛d ⎞
T ≈ 2R2 ⎜ ⎟ (II.25)
⎝ R1 ⎠

where R is the Reynolds number, defined as

R1d
ya que el número de Reynolds R ≡ Ω1
ν
y R1 ≈ R2 i.e.
(II.26)

In our calculations we have used T as control parameter, rather than R ,


because of the adimensionality of the hydrodynamic equations in cylindrical
coordinates.

iii) Wavy vortex flow.

As we just mentioned in the last section, the periodic structure appearing in the
axial coordinate points out the first instability found when the Taylor number is
gradually increased. If we continued doing so, we would find a second
instability (very close to the Taylor vortex in the small gap problem). This time a
second critical Taylor number, T1 labels an azimuthal symmetry breakdown.

According to Swinney’s experiments, if η ≈ 0, 95 he finds T1 to be between


1, 05Tc and 1,10Tc ( Tc being the critical Taylor number for Taylor vortex flow). We
will see in the next chapter how the proximity of T1 with respect to Tc lets us
hope to perform a stability analysis for non-axisymmetric disturbances.
25

Markus (64) has performed a remarkable numerical simulation of the


aspect of those modulated vortices. However, some difficulties arise in the study
of this instability. For instance, the fact of ignoring the exact solution for the
Taylor vortex flow, prevents us from a complete understanding of the selected
pattern.

The new structure consists of the old Taylor vortex flow, but this time
with an azimuthal wave number; this is to say, the torii have now a “wavy” form
in the θ coordinate. There are several ways these waves can settle with respect to
the Taylor vortices, at the same Ω1 .

These waves are not stationary, they have associated a wave speed, which
seems to depend experimentally on the geometrical parameters of the system.
Apparently, the selected pattern depends as well on the way one approaches the
critical Taylor number.

The stability curves of G. King (38) we mentioned before, were found for
the several transitions from one pattern to another with different non-
axisymmetric wave number.

The physical mechanism responsible for the onset of this instability has
not yet been clarified; it seems however that the non-axisymmetric disturbances
producing eventually the waves, are present all the time, say from the very
beginning of the experiment, but they do not become important, mainly due to
the viscosity stress opposing to any change in the structure of the vortices.

Another interesting remark has been pointed out by the experiments of


Mobbs et al. (68), and those of Cognet et al. (65), that apparently the wavy
motion starts before the onset, not in the boundary of successive torii (as it was
thought in the first observations) but at the centre (the core) of the Taylor vortices.

It was Davey et al. (58) who considered for the first time the stability
analysis of the wave vortex flow, in a bifurcation scheme. We will look at the
non-axisymmetric disturbances in the next chapter, and will also consider the
phase diffusion problem.

iv) Further bifurcations.

From a purely theoretical point of view, we know very little about the
description of higher order bifurcations, after the wavy vortex flow, in spite of
the fact that experimentally most of the power spectra have already been
analysed all the way up to the fully developed turbulence.
26

Since a systematic discussion on this problem would take us beyond the


scope of this work, let us only say a few words about it.

The experiences performed so far are restricted to small gap devices; the
flow visualization techniques are being developed every day.

It seems now clear that after the wavy vortex flow, a modulating
frequency appears, producing the so called modulated wavy vortex flow. We do
not even know if such a regime can be visualized “macroscopically”. Some
experimenters consider that such a manifestation is probably the presence of
small eddies near the boundary separating subsequent wavy vortices. In fact, one
could take this as being the first “weak” manifestation of turbulent behaviour.
Nevertheless, it is observed that what we actually call weak turbulence (i.e. the
presence of a noisy broad band in the power spectrum) settles afterwards (while
increasing the Taylor number). It is indeed very difficult to say experimentally
when the instability is going to be considered to be established.

Figure 3 shows: (a) T Tc = 1, 049 , time independent Taylor vortex flow with
18 vortices. The flows at the upper and lower fluid surfaces are inward; (b)
T Tc = 2, 449 ; (c) T Tc = 4, 000 ; (d) T Tc = 4,848 . (b) and (c) illustrate wavy vortex
flow (with four waves around the annulus), while in (d) the waves have
disappeared (69).

(a) (b)

(c) (d)

Figure 3.
27

CHAPTER III

STABILITY ANALYSIS

In Chapter II we gave several elements in order to be able to start a


systematic study of the Couette-Taylor system, in which concerns the first
instability (Taylor vortex flow).

When we say a systematic study we mean of course the mathematical


technique, which includes the bifurcation analysis; moreover, as it should be
expected we will see that being simpler to deal with a dimensionless system of
equations, how a scaling factor is introduced, giving rise in a natural fashion to a
quantity like the Taylor number (this is by no means an obtaining method of
such parameter). Therefore, this prevents us from taking other quantities like Ω1
or the already known Reynolds number, as the parameter whose values will be
controlled.

Hence, the Taylor number (II.23) becomes the natural choice. The present
chapter is the presentation, in detail of the linear stability analysis for both
axisymmetric and non-axisymmetric perturbations. We will also see the
mathematical “sensibility” of the system of equations to the choice of the basis
functions as well as the boundary conditions, before the corresponding
calculation of the critical Taylor number.

i) Axisymmetric perturbations.

Let us consider equations (II.10), (II.11), (II.12), (II.13), which has a stationary
solution (II.20), (II.22), the Couette flow. Therefore we allow the disturbance to
the state (II.16),

δp
u , V (r ) + v , w , ≡ω (III.1)
ρ

hence

(v +V )
2
1
∂ t u + u∂ r u + ( v + V ) ∂θ u + w∂ z u −
r r (III.2)
⎛ 2 u ⎞ V2
= −∂ rω + ν ⎜ ∇ 2u − 2 ⋅ ∂θ ( v + V ) − 2 ⎟ −
⎝ r r ⎠ r
28

v u
∂ t ( v + V ) + u∂ r ( v + V ) + ∂θ ( v + V ) + w∂ z ( v + V ) + ( v + V )
r r
(III.3)
1 ⎛ 2
= − ∂θ ω + ν ⎜ ∇ 2 ( v + V ) + 2 ⋅ ∂θ u −
(v +V ) ⎞

r ⎝ r r2 ⎠
1
∂ t w + u∂ r w + ( v + V ) ∂θ w + w∂ z w = −∂ zω + ν∇ 2 w (III.4)
r
u 1
∂ r u + + ∂θ ( v + V ) + ∂ z w = 0 (III.5)
r r

where we have written all the terms explicitly, so that we can see for instance,
that the fact of adding a perturbation leaves the continuity equation (III.5)
unchanged, since (II.16) tells us precisely that V is a purely radial function.

Let us consider now the axisymmetric perturbation,

u = u ( r ) e qt cos kz , w = w ( r ) e qt sin kz
(III.6)
v = v ( r ) e qt cos kz , ω = ω ( r ) e qt cos kz

where we have called the radial part in the same way as the total variable, on
purpose in order to retain the same appearance. Knowing this we drop the r
dependence and we simply call them u, v, w, ω .

The z – dependent functions are chosen in a natural manner, this is to say,


having in mind that (as we will do later) a general procedure would demand a
systematic Fourier analysis, and at the same time that the expression (III.5) must
be satisfied; the quantity k is the axial wave number, and q is the frequency
which indeed will play the role of the parameter whose sign will predict a stable
or unstable behaviour. Substituting (III.6) into (III.2) – (III.5) one gets the system:

( q +ν k 2
−ν DD* ) u ( r ) + Dω ( r ) − v ( r ) V
2
r (III.7)
1
= −e ( cos kz ) u ( r ) Du ( r ) + kw ( r ) u ( r ) e sin kz tan kz + v 2 ( r ) e qt cos kz
qt qt

r
( q +ν k −ν DD* ) v ( r ) + u ( r ) D*V
2

(III.8)
= −e qt ( cos kz ) u ( r ) D*v ( r ) + kw ( r ) v ( r ) e qt sin kz tan kz + ν DD*V
( q +ν k 2
−ν D* D ) w ( r ) − kω ( r )
(III.9)
= −e qt ( cos kz ) ( u ( r ) D − kw ( r ) ) w ( r )
D*u ( r ) + kw ( r ) = 0 (III.10)

where the differential operators D and D* are defined as


29

1
D ≡ dr ; D* ≡ d r + (III.11)
r

along with the boundary conditions

ω =u =v=w=0 , r = R1 , R2 (III.12)

and as usual V ( r ) has the form (II.20), (II.22).

The system (III.7) – (III.10) can be reduced. Indeed, from (III.10) one gets
w ( r ) , and ω ( r ) in terms of w ( r ) from (III.9), thus

1 qt
ω (r ) = − e ( cos kz ) ⎡⎣u ( r ) D + D*u ( r ) ⎤⎦ D*u ( r )
k2
− 2 ( q + ν k 2 −ν D* D ) D*u ( r )
1
k

and if we keep going in this way, and eliminate ω ( r ) and w ( r ) , we obtain the
following system

⎛ q⎞
ν ⎜ DD* − k 2 − ⎟ v ( r ) − u ( r ) D*V
⎝ ν⎠ (III.13)
= e qt ( cos kz ) u ( r ) D*v ( r ) + v ( r ) D*u ( r ) eqt sin kz tan kz

and

ν ⎛ q⎞
⎜ DD* − k − ⎟ ( DD* − k ) u ( r ) − v ( r ) V
2 2 2
k ⎝
2
ν⎠ r
⎛1 ⎞
= e qt ( cos kz ) ⎜ v 2 ( r ) − u ( r ) Du ( r ) ⎟ − u ( r ) D*u ( r ) e qt sin kz tan kz (III.14)
⎝r ⎠
1
+ 2 e qt ( cos kz ) ⎡⎣ Du ( r ) + u ( r ) D + 2 D*u ( r ) ⎤⎦ DD*u ( r )
k

with the conditions

1
w(r ) = D*u ( r ) (III.15)
k
ω ( r ) = − 2 ( q + ν k 2 −ν D* D ) D*u ( r )
1
k
(III.16)
1
− 2 e qt ( cos kz ) ⎡⎣u ( r ) D + D*u ( r ) ⎤⎦ D*u ( r )
k

where we have taken into account that


30

D*V = 2 A

and the condition for the stationary solution

⎛ p ⎞ V2
D⎜ ⎟ =
⎝ρ⎠ r

We now have a system of only two differential equations, and the


boundary conditions are

u ( r ) = v ( r ) = D*u ( r ) = 0 , r = R1 , R2 (III.17)

these being different from those (III.12); this is not a surprising fact, since
reducing a system of differential equations alters the boundary conditions.

An important aspect comes out from this. With the conditions (III.12) it is
relatively easy to propose a function vanishing at the walls of the cylinders;
however (III.17) imposes that the first total derivative of the radial component of
the velocity be zero also at the boundaries. We will see afterwards in which way
this condition is not so easily satisfied.

In order to get an approach for the solution of (III.13), (III.14), the small
gap approximation ( η ≈ 1 ) provides a simplification. As a matter of fact this
condition has to be taken very carefully because of the sensitivity the equations
present to such a limiting case. Formally, we write this as,

1
d = R2 − R1 ( R1 + R2 ) (III.18)
2

and it is easy to check that effectively (Davey), if we define the quantity x such
that

x=r−
( R1 + R2 ) ≡ r − R0 (III.19)
2d d

then dx = dr

1 d
and D* = d r + = dx +
r xd + R1 + R2

so

lim D* = d x = D (III.20)
d →0
31

Notice that instead of x we could define the variable ξ like

r − R1
ξ≡ (III.21)
d

which will show to be useful afterwards. Therefore, in the small gap limit, we
make no difference between D* and D . Similarly, from (III.21)

B ⎧ r − R1 ⎫
A+ → Ω1 ⎨1 − ⎬ = Ω1 (1 − ξ ) (III.22)
⎩ R2 − R1 ⎭
2
r

Now is the moment to propose the change of scale (i.e. the dimensions) we
already mentioned in the last chapter. Having a look on equations (III.13) and
(III.14) it is easy to see that the following quantities would simplify the form of
the coefficients, namely

r = ξ d + R1 , ξ ∈ [ 0,1]
qd 2 (III.23)
a ≡ kd , σ=
ν

and the system becomes

ν
d2
(D 2
− a 2 − σ ) v − 2 Au
(III.24)
=e σν t d 2
( cos az ) uDv + e σν t d 2
( sin az )( tan az ) vDu

ν
a2d 2
(D 2
− a 2 − σ )( D 2 − a 2 ) u − 2vΩ1 (1 − ξ )
(III.25)
eσν t d
2
1 1
= ( cos az ) v 2 − eσν t d ( sec az ) uDu + 2 eσν t d ( cos az )( 3Du + uD ) D 2u
2 2

ξ d + R1 d ad

Looking at the coefficients of u in the second term of (III.24) one can realize
where the Taylor number T will come from (see (II.23)).

Let us make a further simplification; we will keep only the linear terms as
a first approach, and finally, let us take the case σ = 0 (marginal state). So, from
equations (III.7) – (III.10) it follows

( q +ν k 2 −ν DD* ) u + Dω − 2r vV = 0
( q +ν k 2 −ν DD* ) v + uD*V = 0 (III.26)
( q +ν k 2
−ν DD* ) w − kω = 0
D*u + kw = 0
32

and we will perform the narrow gap approximation later. Eliminating the
pressure we find,

⎛ q⎞ 2 ν ⎛ q⎞
ν ⎜ DD* − k 2 − ⎟ u + Vv + D ⎜ D* D − k 2 − ⎟ w = 0
⎝ ν⎠ r k ⎝ ν⎠

( D*V ) u −ν ⎛⎜ DD* − k 2 −
q⎞
v=0 (III.27)
⎝ ν ⎟⎠
D*u + kw = 0

and after doing the small gap approximation, and using (III.23)

d2
(D 2
− a − σ ) u + 2Ω1
2

ν
(1 − ξ ) v +
1
a
D ( D2 − a2 − σ ) w = 0
2
u − ( D2 − a2 − σ ) v = 0
d A
2 (III.28)
ν
Du + aw = 0

which can be simplified by making the transformation

2Ω1d 2
u → u (III.29)
ν

and reminding (II.23) for the Taylor number, (III.28) becomes

ν
(D 2
− a 2 − σ ) u + (1 − ξ ) v +
2aΩ1d 2
D ( D2 − a2 − σ ) w = 0

Tu + ( D − a − σ ) v = 0
2 2
(III.30)

Du + w=0
2Ω1d 2

Since the boundary conditions are that u , v , and w be zero for ξ = 0 and 1 , one
would be tempted to consider as a possible solution of (III.30),

u = C1 sin πξ
v = C2 sin πξ (III.31)
w = C3 sin πξ

however, a substitution into (III.30) shows that it is not so, unless we take rather
than the third equation of the system (III.30), its derivative. Otherwise, one finds
contradictory results. Doing so, for the marginal state σ = 0 one finds a critical
Taylor number Tc
33

2 (
π + a2 )
2 2 3
Tc = (III.32)
a

which is a very rough result compared to those obtained so far by


Chandrasekhar, Davey et al., etc. (14)(52)(58), but reveals in a quite transparent
way how rough is our assumption (III.31). In fact, not only should we satisfy the
vanishing conditions on the walls, but those of vanishing derivatives as well.

Obviously (III.31) is not the solution; so, for the sake of simplicity, one can
imagine to propose rather

u = C1 sin πξ ⎡⎣θ (ξ ) − θ (1 − ξ ) ⎤⎦
v = C2 sin πξ ⎡⎣θ (ξ ) − θ (1 − ξ ) ⎤⎦ (III.33)
w = C3 sin πξ ⎡⎣θ (ξ ) − θ (1 − ξ ) ⎤⎦

where θ (ξ ) is the Heaviside function

0 , ξ <0
θ (ξ ) =
1 , ξ >0

so that every time we find cos πξ we substitute it by cos πξ ⎡⎣θ (ξ ) − θ (1 − ξ ) ⎤⎦ ,


vanishing then also at ξ = 0 and 1 , and leading us to (III.32) again. From this we
can already see that even the linear case demands carefulness, since extremely
simplified solutions like (III.31) could lead us to the wrong conclusions, even
though they already exhibit some aspects of the solutions which are basically
right.

At this point we shall say that we are driven to do the things in this way,
because we think it is worthwhile to check the effects of choosing simple
solutions; concluding what we said above.

Physically this means that not only are the solutions like (III.31) partially
right, but they should include higher order harmonics as well. We had already
inserted this idea before, when we slightly discussed about wavy, and
modulated wavy vortex flow. It will become clearer though in the next sections
that the harmonics (axisymmetric and non-axisymmetric) are all present in the
fluid all the time, however only some of them are dominant, and hence they
show off. This does not mean that they should all be present in the allowed
solutions.
34

All this implies that the instability starts in the system before the onset
itself. It does not seem to be so for the first instability though, probably because
of the suddenness of the event, but it could be the case in further bifurcations (as
it is experimentally supported by Cognet (65)).

In a more complete context, we should consider as solutions functions


being combinations of sin aξ , cos aξ , sinh aξ , and cosh aξ (14).

We are now going to consider the calculations of stability with an


azimuthal wave number; in other words, the non-axisymmetric disturbances.

ii) Non-axisymmetric perturbations.

So far we have not taken into account the azimuthal part of the hydrodynamic
equations. In the aim of considering a more general analysis we introduce the θ -
dependent part in the proposed solutions (III.6).

Continuing from the last section, the idea of putting the θ - dependent
disturbance from the very beginning, implies that we assume that they are
present from the starting point. Notice that we do not pretend to be already
describing an eventual wavy behaviour, however we do expect that the inclusion
of m (the azimuthal wave number) in (III.6) will modify the predicted critical
Taylor number; otherwise, Tc would have a value regardless of the presence of
an azimuthal dependence.

First of all, let us consider the full hydrodynamic system of four equations
with all the non – linear and non – axisymmetric terms (II.10) – (II.13);
considering the perturbation (III.1) to the stationary Couette flow, and getting
then (III.2) – (III.5).

Consider now rather than (III.6),

u = u ( r ) eqt ei( mθ + kz ) , w = w ( r ) e qt ei( mθ + kz )


(III.34)
v = v ( r ) e qt ei( mθ + kz ) , ω = ω ( r ) eqt ei( mθ + kz )

The hydrodynamic system then becomes,

⎛ m2 q⎞ 2 m⎛ 2ν ⎞
ν ⎜ DD* − − k 2 − ⎟ u + Vv − Dω ( r ) − i ⎜ Vu + v⎟
⎝ r 2
ν⎠ r r⎝ r ⎠
(III.35)
⎧⎛ m ⎞ v2 ⎫
= ⎨⎜ i v + ikw + uD ⎟ u − ⎬ eqt ei( mθ + kz )
⎩⎝ r ⎠ r⎭
35

⎛ m2 q⎞ m⎛ 2ν ⎞
ν ⎜ DD* − 2 − k 2 − ⎟ v − uD*V − i ⎜ Vv + ω ( r ) − u ⎟
⎝ r ν⎠ r⎝ r ⎠
(III.36)
⎧⎛ m ⎞ ⎫
= ⎨⎜ i v + ikw + uD* ⎟ v ⎬ eqt ei( mθ + kz )
⎩⎝ r ⎠ ⎭

⎛ m2 q⎞ m
ν ⎜ D* D − − k 2 − ⎟ w − ikω ( r ) − i Vw
⎝ r 2
ν⎠ r
(III.37)
⎧⎛ m ⎞ ⎫
= ⎨⎜ i v + ikw − uD ⎟ w⎬ e qt e (
i mθ + kz )

⎩⎝ r ⎠ ⎭

m
D*u + i v + ikw = 0 (III.38)
r

Again, as a first step we take only the linearized terms, hence from (III.38)

1⎛ m ⎞
w= ⎜ iD*u − v ⎟ (III.39)
k⎝ r ⎠

and following about the same steps as in the last section, we can reduce (III.35) –
(III.38) to two equations in u and v (remember we dropped the r - dependence
bracket),

⎛ m2 q mV ⎞
ν ⎜ DD* − − k2 − −i ⎟ v − uD*V
⎝ r 2
ν νr ⎠
(III.40)
m ⎪⎧ ν ⎛ m2 q mV ⎞ ⎛ m ⎞ ν ⎛ m ⎞ 2ν ⎪⎫
= i ⎨ 2 ⎜ DD* − 2 − k 2 − − i ⎟⎜ D*u + i v ⎟ + 2 2 ⎜ D*u + i v ⎟ − u⎬
r ⎩⎪ k ⎝ r ν ν r ⎠⎝ r ⎠ k r ⎝ r ⎠ r ⎭⎪

ν ⎛ m2 q mV ⎞
− − k2 − −i ⎟ ( DD* − k ) u
2
2 ⎜
DD
k ⎝
*
r 2
ν νr ⎠
2 mν ⎛ m2 q mV ⎞ ⎛ v ⎞
= Vv − i 2 ⎜ DD* − 2 − k 2 − − i ⎟D (III.41)
r k ⎝ r ν ν r ⎠ ⎜⎝ r ⎟⎠
mν ⎛ B ⎞⎛ v⎞ mν
−2 2 3 ⎜
m + i ⎟ ⎜ D*u + im ⎟ − 2i 2 v
k r ⎝ ν ⎠⎝ r⎠ r

with (III.39) and

ν ⎛ m2 q mV ⎞⎛ m ⎞
ω (r ) = 2 ⎜ *
D D − − k2 − −i ⎟ ⎜ D*u + i v ⎟ (III.42)
k ⎝ r 2
ν νr ⎠⎝ r ⎠

The small gap approximation of (III.40), (III.41) comes out from (III.11),
(III.18), (III.19), (III.22), (III.23) and the fact that
36

Ω1 R2 2 Ω1
B≈ ; A≈− (III.43)
2δ 2δ

where

d
δ≡
R1

thus, we obtain

⎛ 2 d2 ⎞
⎜ D − a 2
− σ − im Ω1 (1 − ξ ) ⎟ v = u
⎝ ν ⎠
(III.44)
⎛ 2 2
⎞ 2
⎜ D − a − σ − im Ω1 (1 − ξ ) ⎟ ( D − a ) u = − a T (1 − ξ ) v
2 d 2 2

⎝ ν ⎠

and the boundary conditions u = v = Du = 0 at ξ = 0 and ξ = 1 . These equations


are equivalent to those given by Krueger (57) (this is also true when we make
ν = 0 , i.e. the inviscid flow). Similarly to the last section, one could think of a
system of three rather than two equations. Let us do it in a slightly different way
this time, and consider a matrix notation to write out the system. We do not take
dimensionless variables, and for this time only, a case where μ ≠ 0 .

If we define

u ≡ ( u , v, w )

1 ∂ ∂
∂ ≡ ∂r , ∂* ≡ ∂ + , ∂z ≡ , ∂θ ≡
r ∂z ∂θ

⎛ ∂ ⎞ ⎛ ∂* ⎞
⎜ ⎟ ⎜ ⎟
1 1
∂ ≡ ⎜ ∂θ ⎟ , ∂ * ≡ ⎜ ∂θ ⎟
⎜r ⎟ ⎜r ⎟
⎜ ∂ ⎟ ⎜ ∂ ⎟
⎝ z ⎠ ⎝ z ⎠

( )
N (u, u ) ≡ − u ⋅ ∂ u + δ u ⋅ u

⎛ 0 uθ
0 ⎞⎟
⎜ r
⎜ ⎟
δ u ≡ ⎜ − uθ r 0 0⎟ (III.45)
⎜ ⎟
⎜⎜ 0 0 0⎟

⎝ ⎠
37

hence

∂ t u = ν D0 ⋅ u − ∂ω + N ( u , u )
(III.46)
∂* ⋅ u = 0

where D0 is given below, such that

1
2Ω1 R12 ⎛ μ⎞ 2
(1 − μ )
1
β= T = 2 ⎜
1− 2 ⎟ 2
ν (1 − η ) ⎝ η ⎠
1
⎛ μ⎞ 2

⎜1 − η 2 ⎟
α=⎝ ⎠
(1 − μ ) 2
1

⎛ ⎡⎛ 1 2 2 ⎞⎤ ⎞
⎜ ⎢⎜ ∂∂* + 2 ∂θ + ∂ z ⎟ ⎥ ⎟
⎜ ⎢⎝ r ⎠⎥ ⎡ ⎛ 1 ⎞ 2 ⎤ ⎟
⎢ − β ⎜ α − 2 ⎟ − 2 ∂θ ⎥ 0
⎜⎢ β ⎛ 1 ⎞ ⎥ ⎣ ⎝ αr ⎠ r ⎦ ⎟
⎜ ⎢ + ⎜ α − 2 ⎟ ∂θ ⎥ ⎟
⎜⎣ 2 ⎝ α r ⎠ ⎦ ⎟
⎜ ⎡⎛ ⎟
1 2 2 ⎞⎤
⎜ ⎢⎜ ∂∂* + r 2 ∂θ + ∂ z ⎟ ⎥ ⎟
⎜ ⎡ 2 ⎤ ⎢⎝ ⎠⎥ ⎟
D0 ≡ ⎜ ⎢⎣ βα + r 2 ∂θ ⎥⎦ ⎢ β⎛ 1 ⎞ ⎥
0 ⎟ (III.47)
⎜ ⎢ + ⎜ α − 2 ⎟ ∂θ ⎥ ⎟
⎜ ⎣ 2⎝ αr ⎠ ⎦ ⎟
⎜ ⎟
⎜ ⎡⎛ 1 2 2 ⎞⎤ ⎟
⎜ ⎢⎜ ∂∂* + r 2 ∂θ + ∂ z ⎟ ⎥ ⎟
⎜ 0 0 ⎢⎝ ⎠⎥

⎜ ⎢ β⎛ 1 ⎞ ⎥⎟
⎜ ⎢ + ⎜ α − 2 ⎟ ∂θ ⎥ ⎟
⎝ ⎣ 2⎝ αr ⎠ ⎦ ⎠

which is the generalization to the usual way this system is presented (70). In the
small gap limit, and using the solution (III.34) this set of equations can be written
as

⎛ 2 Ω1 2 ⎞
⎜ D − a − σ − im
2
d (1 − ξ ) ⎟ u + (1 − ξ ) v
⎝ ν ⎠
(III.48)
iν ⎛ 2 Ω ⎞
+ 2 ⎜
D − a 2 − σ − im 1 d 2 (1 − ξ ) ⎟ Dw = 0
2Ω1ad ⎝ ν ⎠
⎛ Ω ⎞
Tu + ⎜ D 2 − a 2 − σ − im 1 d 2 (1 − ξ ) ⎟ v = 0 (III.49)
⎝ ν ⎠
iaν
Du + w=0 (III.50)
2Ω1d 2
38

or

D3n ⋅ u = 0 (III.51)

where

⎛ iν ⎞
⎜E (1 − ξ ) E⎟
2Ω1ad 2
⎜ ⎟
⎜T E 0 ⎟ (III.52)
⎜ ⎟
⎜D iaν ⎟
⎜ 0
⎝ 2Ω1ad 2 ⎟⎠

⎛ Ω ⎞
E ≡ ⎜ D 2 − a 2 − σ − im 1 d 2 (1 − ξ ) ⎟ (III.53)
⎝ ν ⎠

If we allow again a solution like (III.31) we find it is too rough to expect a


consistent result for Tm . However, thinking on the boundary conditions for the
first derivative, we rather propose

u = c1 sin 2 πξ
(III.54)
v = c2 sin πξ

which is a system satisfying the boundary conditions. To begin with, we do


know that (III.54) is not a solution of (III.44), but we think it is worthwhile to
substitute it into the system in order to look at the Tm coming out from this. The
result we obtain is the following,

2π 2 2
2 (
Tc − π + a 2 )(π 2 + 2a 2 )
Tm = a (III.55)

2δ ⎛ a2 ⎞ ⎛ 1 1 ⎞ π 2 a 2γ ⎤
1 + m 2 2 ⎢⎜ 2π 2 + ⎟ ⎜ − 2 ⎟ − + ( γ − 3) ⎥
a ⎣⎝ 2 ⎠ ⎝ 6 4π ⎠ 2 32 ⎦

where

3a 2 + 2π 2
γ≡ (III.56)
2a 2 + π 2

and Tc is given by (III.32). Remember that Tc is already the result coming out
from a very rough approximation, then we should not expect (III.55) to represent
an accurate result. Nevertheless, it is interesting to have a look in this expression.
Notice that m , the azimuthal wave number takes the values 0, 1, 2, … and so on.
39

Expression (III.55) is not entirely correct since, if one takes the case m = 0 ,
one gets a T0 < Tc , which would mean that because of the presence of the
azimuthal disturbance the Taylor instability appears earlier than usual, and the
same is true for all m ≠ 0 . However it is interesting to see that T is modified
anyhow due to m . The last thing to observe (and about this, we would expect it
to be true in the exact solution), is that Tm only depends on a ( = kd ) and δ
( = d R1 ), this is to say the width of the gap (we are considering infinite cylinders,
otherwise it would be the aspect ratio).

Let us go back to system (III.40), (III.41) of two equations. The azimuthal


wave number dependence is included there, and the small gap approximation
has not yet been performed. So, one thing to be done is to go as far as possible in
the calculation we outline below, without taking any limit, and thereafter take
the narrow gap.

We have considered so far the marginal case q = 0 for the stability curve.
As we are interested in an eventual periodic azimuthal behaviour, we will
consider the quantity q to be complex,

q ≡ qr + iqi (III.57)

therefore, separating the equations in real and imaginary parts. Solving this
system of equations, give the solutions for qr and qi (that we call the
eigenfrequencies). To get the purely periodic description we make qr = 0 for both
the real and the imaginary equations. The result after all this is a system of two
quadratic equations (with real coefficients).

The expressions for the eigenfrequencies are too long to be written here.
We performed the algebraic calculations with the MACSYMA 303 program (MIT)
at the University of Texas at Austin Computation Centre. However we will try to
outline the type of equations.

First of all, system (III.40), (III.41) can be written as

mν ⎛ m2 q mV ⎞ mν mν
i 2 ⎜
DD − − k2 − −i ⎟ ( D*u ) + i 2 3 D*u − 2i 2 u
rk ⎝
*
r 2
ν rν ⎠ k r r
m 2ν ⎛ m2 q mV ⎞ ⎛ v ⎞
+ ( D*V ) u − 2 ⎜ DD* − 2 − k 2 − − i ⎟⎜ ⎟ (III.58)
rk ⎝ r ν rν ⎠ ⎝ r ⎠
⎛ m2 q mV ⎞ m 2ν
−ν ⎜ DD* − 2 − k 2 − − i ⎟ v − v=0
⎝ r ν rν ⎠ k 2r 4

and
40

ν ⎛ m2 q mV ⎞ mν ⎛ B⎞
DD* − 2 − k 2 − − i ⎟ ( DD* − k ) u + 2 2 3 ⎜ m + i ⎟ D*u
2
2 ⎜
k ⎝ r ν rν ⎠ k r ⎝ ν⎠
ν ⎛ m2 q mV ⎞ ⎛ v ⎞ 2
+im 2 ⎜ DD* − 2 − k 2 − − i ⎟ D ⎜ ⎟ − Vv (III.59)
k ⎝ r ν rν ⎠ ⎝ r ⎠ r
m 2ν ⎛ B⎞ mν
+2 2 4 ⎜ m + i ⎟ v + 2i 2 v = 0
k r ⎝ ν⎠ r

but notice that

⎛v⎞ 1 1
DD* ⎜ ⎟ = d r2 v − 2 d r v
⎝r⎠ r r
1⎛ 2 1 ⎞
= ⎜ DD*v − Dv + 2 v ⎟
r⎝ r r ⎠

and therefore the operator acting on v becomes

⎡ mV ⎛ m2 ⎞ ⎤
⎢q + i +ν ⎜ − DD* + 2 + k 2 ⎟ + ⎥
⎢ r ⎝ r ⎠ ⎥
⎢ ⎥v (III.60)
m 2
⎛ mV ⎛ m 2
⎞ ⎞ m 2
ν ⎛ 1 ⎞
⎢+ ⎜q+i + ν ⎜ − DD* + 2 + k ⎟ ⎟ + 2 2 3 ⎜ D − ⎟ ⎥
2

⎢⎣ k 2 r 2 ⎝ r ⎝ r ⎠⎠ k r ⎝ r ⎠ ⎦⎥

and for u ,

⎡ m ⎛ mV ⎛ m2 ⎞⎞ ⎤
⎢ *
D V − i 2 ⎜
q + i + ν ⎜ − DD* + 2
+ k 2 ⎟ ⎟ D* + ⎥
⎢ k r⎝ r ⎝ r ⎠⎠ ⎥u (III.61)
⎢ mν mν ⎥
⎢ +i 2 3 D* − 2i 2 ⎥
⎣ k r r ⎦

Before taking the small gap limiting process let us remember that Bessel’s
equation is (71)(72)

d 2 f 1 df ⎛ n 2 ⎞
+ ⋅ + ⎜1 − ⎟ f = 0 (III.62)
dx 2 x dx ⎝ x 2 ⎠

implying that

DD* f = −τ 2 f = D 2 f (III.63)

where we have done D → D* , n = 1 , and introduced the parameter τ , which is


the distance between subsequent zeroes of the asymptotic Bessel’s functions.
Expression (III.63) suggests us to identify
41

D → iτ (III.64)

therefore the final system can be written as

⎪⎧ ⎡ mV ⎛ 2 m2 ⎤
2⎞ ⎛ m2 ⎞ m 2ν ⎛ 1 ⎞ ⎪⎫
⎨⎢ q + i + ν ⎜ τ + + k ⎟⎥ ⎜ 1 + 2 2 ⎟
+ 2 2 3 ⎜
iτ − ⎟ ⎬ v
⎪⎩ ⎣ r ⎝ r 2
⎠⎦ ⎝ k r ⎠ k r ⎝ r ⎠ ⎪⎭
(III.65)
⎧⎪ τm ⎡ mV ⎛ 2 m2 ⎞ ⎤ mν ⎛ τ ⎞ ⎫⎪
+ ⎨ D*V + 2 ⎢ q + i +ν ⎜τ + 2 + k 2 ⎟ ⎥ − 2 ⎜ 2 + 2i ⎟ ⎬ u = 0
⎩⎪ k r⎣ r ⎝ r ⎠⎦ r ⎝ k r ⎠ ⎭⎪

⎧⎪ ⎡ mV ⎛ m2 ⎞⎤ ⎛ τ 2 ⎞ mν ⎛ B ⎞ ⎫⎪
⎨⎢q + i +ν ⎜τ 2 + 2 + k 2 ⎟ ⎥ ⎜1 + 2 ⎟ + 2τ 2 3 ⎜ im − ⎟ ⎬ u
⎩⎪ ⎣ r ⎝ r ⎠⎦ ⎝ k ⎠ k r ⎝ ν ⎠ ⎭⎪
⎧τ m ⎡ mV ⎛ 2 m2 ⎤
2⎞ ⎛ i ⎞ ⎫
ν (III.66)
⎪ 2 ⎢ q + i + ⎜τ + 2 + k ⎟ ⎥ ⎜1 + ⎟ − ⎪
⎪k r ⎣
+⎨
r ⎝ r ⎠⎦ ⎝ τ r ⎠ ⎪ v = 0

⎪ 2V m 2ν ⎛ B⎞ mν ⎪
⎪− r + 2 k 2 r 4 ⎜ m + i ν ⎟ + 2i r 2 ⎪
⎩ ⎝ ⎠ ⎭

or equivalently

Au + Bv = 0
(III.66’)
Cu + Dv = 0

where the definitions of A, B, C , D come straightforward from (III.66). If we want


(III.66’) to have a solution, it must satisfy the condition of a vanishing
determinant, this is to say

⎡ ⎛ m2 ⎞ m 2ν ⎛ 1 ⎞⎤ ⎡ ⎛ τ 2 ⎞ mν ⎛ B ⎞⎤
⎢ Ω ⎜1 + 2 2 ⎟ + 2 2 3 ⎜ iτ − ⎟ ⎥ ⎢Ω ⎜ 1 + 2 ⎟ + 2τ 2 3 ⎜ im − ⎟ ⎥ +
⎣ ⎝ k r ⎠ k r ⎝ r ⎠⎦ ⎣ ⎝ k ⎠ k r ⎝ ν ⎠⎦
⎡ τm mν ⎛ τ ⎞⎤
+ ⎢ D*V + 2 Ω − 2 ⎜ 2 + 2i ⎟ ⎥ ⋅ (III.67)
⎣ k r r ⎝k r ⎠⎦
⎡ 2V mτ ⎛ i ⎞ 2m 2ν ⎛ B⎞ mν ⎤
⋅⎢ − 2 Ω ⎜1 + ⎟ − 2 4 ⎜ m + i ⎟ − 2i 2 ⎥ = 0
⎣ r k r ⎝ τr ⎠ k r ⎝ ν⎠ r ⎦

or

PΩ 2 + ∅Ω + M = 0 (III.67’)

where

mV ⎛ 2 m2 ⎞
Ω ≡ q+i +ν ⎜τ + 2 + k 2 ⎟ (III.68)
r ⎝ r ⎠
42

τ2 m2 m 2τ
P = 1+ + − i (III.69)
k2 k 2r 2 k 4r3

⎛ m 2 B mντ 2 mν τ B τν ⎞
⎜ −τ B − 2 2
− 2 2 −2 2 − − m3 2 + ⎟
2m ⎜ k r 2k r r r k r ⎟
∅= 2 2 (III.70)
k r ⎜ ⎧ m3ν mντ 3 m 2τ B mντ mντ ⎫ ⎟
⎜ +i ⎨τ mν + τ 2 2 + 2 − 2 − A + 2 3 + ⎬⎟
⎝ ⎩ k r k r k r k r r ⎭⎠

4m 4τ 2ν 2 4τ m3ν B ⎛ B⎞ m3ν
M =− + + 4 A ⎜ A + ⎟ − 4 A −
k 4r 5 k 4r 6 ⎝ r2 ⎠ k 2r 4
mντ ⎛ B⎞ m 4ν 2τ m3ν B m 2ν 2
−2 ⎜ A + ⎟ + 2 − 4 − 4 + (III.71)
k 2r 3 ⎝ r2 ⎠ k 4r 7 k 2r 6 r4
⎧ 2 mν m 4ν 2 mν ⎫
3
m 2 AB
⎪⎪ − 4τ B − 4 τ − 4 − 8 A 2 −⎪
4 5
k r 4 6
k r k r2 4
r ⎪
+i ⎨ ⎬
⎪−4 mν B + 2 m ντ B + 4 m ν + 2 m ν τ
4 4 2 2 2

⎩⎪ r 4 k 4r 7 k 2r 6 k 2r 5 ⎪⎭

This is to say, that the conditions of a vanishing determinant takes the form of a
quadratic equation in Ω , with complex coefficients, being Ω itself complex.
Before separating into the real and imaginary parts, we consider (III.57), then

⎡ ⎛ 2 m2 ⎞⎤ ⎛ mV ⎞
Ω = ⎢ qr + ν ⎜τ + 2 + k 2 ⎟ ⎥ + i ⎜ qi + ⎟ (III.72)
⎣ ⎝ r ⎠⎦ ⎝ r ⎠

2
⎡ ⎛ 2 m2 2⎞
⎤ ⎛ mV ⎞
2

Ω = ⎢ qr +ν ⎜ τ + 2 + k ⎟ ⎥ − ⎜ qi +
2
⎟ +
⎣ ⎝ r ⎠⎦ ⎝ r ⎠
(III.72’)
⎡ ⎛ m2 ⎞⎤ ⎛ mV ⎞
+2i ⎢ qr + ν ⎜τ 2 + 2 + k 2 ⎟ ⎥ ⎜ qi + ⎟
⎣ ⎝ r ⎠⎦ ⎝ r ⎠

Remember at this point that the parameter τ represents the distance


between subsequent zeroes of the Bessel function, assuming its asymptotic form,
which indeed is nothing more but a combination of sinus and cosine functions,
justifying thus the choices we have made so far. For the sake of simplicity in the
presentation, we will sketch the results only.

We call the separated equations, real and imaginary keeping in mind in


this way their respective origin. Then we make qr = 0 , so that the description
becomes purely periodic. Doing this leads us to a system of the form

α qi2 + β qi + γ = 0 (III.73)
α ' qi2 + β ' qi + γ ' = 0 (III.73’)
43

in qi ; α has three terms, β has twelve, γ has fifty four, however we obtained
the full expressions for qi1 , qi 2 , qi1 ', qi 2 ', being these the solutions to (III.73) and
(III.73’).

Remember that the only thing we have done about the small gap
approximation, is concerning the differential operator identified with Bessel’s
one (III.62) and (III.63). Indeed, as we must be careful about these approaches,
we see that (III.73) and (III.73’) have a solution only if

α' ⎛ 2
( ) 2⎞ β2
( )
1
( )
1
⎜ 2 β − 4αγ ∓ 2 β β 2
− 4αγ ⎟ + − β ± β '− 4αγ 2
+γ ' = 0 (III.74)
4α 2 ⎝ ⎠ 2α

being too intricate to be useful, so we prefer to write again (III.67’) as

∅ M
Ω2 + Ω + =0 (III.75)
P P

such that

⎛ τ 2 m2 ⎞
⎜1 + 2 + 2 2 ⎟
1 ⎝ k k r ⎠ m 2τ
= +i (III.76)
P ⎡⎛ τ 2 m 2 ⎞ 2 ⎛ m 2τ ⎞ ⎤ ⎡ τ 2 m 2 ⎞ ⎛ m 2τ ⎞ ⎤
2
4 3 ⎛
⎢⎜ 1 + 2 + 2 2 ⎟ + ⎜ 4 3 ⎟ ⎥ k r ⎢⎜ 1 + 2 + 2 2 ⎟ + ⎜ 4 3 ⎟ ⎥
⎢⎣⎝ k k r ⎠ ⎝ k r ⎠⎥
⎦ ⎢⎣⎝ k k r ⎠ ⎝ k r ⎠⎥

We have taken 1 P rather than P to be able to find a linear equation in qi


(without performing any other approximation apart from D = iτ ).

It is relatively easy to check that, making m = 0 for both the real and
imaginary equations, implies that qi = 0 as it should be. Doing this we can write

⎛τ 4 k4 ⎞ 4Ω 2 R14 ⎛ R22 ⎞
ν 2τ 2 ⎜ + 3τ 2
+ 3k 2
+ ⎟ + ⎜1 − ⎟ (III.77)
⎝k
2
τ 2 ⎠ d 2 ( R1 + R2 )2 ⎝ R02 ⎠

and this expression comes out from (III.58), (III.59) making m = 0 , and
eliminating qi , and making the determinant of the system equal to zero. As it
should be expected, we shall now include the Taylor number T into the
expressions. T is given by (II.23), which can also be written as

4 Ω12 R14
T= ⋅ ⋅ R2 4 (III.78)
ν ( R1 + R2 )
2 2 2
d

Substituting in (III.77) we get


44

R2 4 R0 2τ 2 ⎛ τ 4 k4 ⎞
T= + τ 2
+ 2
+ (III.79)
( R02 + R22 ) ⎜⎝ k 2 ⎟
3 3k
τ2 ⎠

Since the parameter τ is expected to be approximately equal to π in the


asymptotic form of the Bessel function, hence

R 4 ( 2 R2 − d ) 1 2
2

⋅ 2 (π + k 2 )
3
T= 2 (III.80)
( 4 R2 − d ) d k

Ahlers, Dominguez-Lerma, and Cannell (73) have also calculated the k -


dependence form of T , and found an expression similar to ours.

To obtain (III.80) we took the general expressions for A and B , not the
small gap approximations (III.43). To compare let us do so; the result is the
following,

4 R2 4 ( R2 − d ) 1 2
2

⋅ 2 (π + k 2 )
3
T= (III.82)
( 4 R2 − 3d ) d k

We already pointed out that these expressions are far from being exact, so
we cannot expect a Tc predicting the onset of the Taylor instability correctly. This
is true, mainly due to the coefficient appearing before the k -dependent function.
In figure 4 we show the graph associated to this one. The constant b is such that

R2 4 ( 2 R2 − d )
2

, for A, B from (II.22)


( 4 R2 − d ) d
(III.83)
b≡

4 R2 4 ( R2 − d )
2

, for A, B from (III.43)


( 4 R2 − 3d ) d

This curve represents the stability graph for axisymmetric disturbances. We


included it in this section rather than in the last one, since we found the
expressions from a series of approximations on equations containing m .

Recently, a study of the marginal stability curve and linear growth rate for
Couette-Taylor flow and Rayleigh-Bénard convection was made by Dominguez-
45

Lerma et al. (74). In figure 5 we show the points predicted by our calculation and
theirs (see table 1). 5

Following the same method one can also find an expression for Tm . The
algebraic calculation is longer though, and moreover, we are practically obliged
to perform the small gap approximation from the very beginning.

2 (
π + k2 )
1 2 3
T = b⋅
k
(see eqn. (III.83))

Figure 4.

A subtle thing comes out at that moment; making d → 0 eliminates all


the m -free terms containing T . This is not good, since a Tm depending
exclusively on m -terms leaves it indefinite for the case m = 0 . So, we keep the
term

ν2 ⎛ R2 2 ⎞
T ⎜1 − ⎟
R2 4 ⎝ R0 2 ⎠

This is concerning the real equation; for the imaginary one this problem does
not arise. At the end we have the system

5 Both curves are comparable only with reduced variables.


46

−ξ qi 2 + α T ( 1
2
) (
+ β qi + γ T + δ T
1
2
+ζ ) = 0
(III.84)
+ (α ' T + β ') q + ( γ ' T + δ ' T + ζ ') = 0
1 1
ξ ' qi 2 2
i
2

where the coefficients are easily found.

Table 1.

(k − k ) *
c kc* (T − Tc ) Tc (T − T )
c
*
Tc*
-0,70 2,7582 2,8523
-0,65 1,8919 1,9596
-0,60 1,3335 1,3838
-0,30 0,1671 0,1761
-0,10 0,0146 0,0156
-0,05 0,0035 0,0037
0,00 0,0000 0,0000
0,05 0,0032 0,0035
0,10 0,0124 0,0135
0,30 0,1009 0,1118
0,60 0,3714 0,4220
1,00 0,9992 1,1745
1,50 2,3258 2,8409
* (74)

If we define

V ⎛ m2 ⎞
ω ≡ q + im + ν ⎜τ 2 + 2 ⎟ (III.85)
R0 ⎝ R0 ⎠

then the determinant of (III.84) becomes

A ω 2 ( A + iB ) + ω ( A '+ iB ' ) + ( A ''+ iB '' ) = 0 (III.86)

such that ω is complex, and the quantities A, B, … are identified with the terms
appearing in the determinant. The solvability condition for (III.86) is

AA '+ BB '+ 2α ' ( A 2 + B2 ) = 0

such that
47

⎛ 2 m2 ⎞
α ' ≡ ν ⎜τ + 2 ⎟ (III.87)
⎝ R0 ⎠

⎛ m 2 ⎞ ⎛ τ 2 ⎞ m 2τ 2
A ≡ ⎜1 + 2 2 ⎟ ⎜1 + 2 ⎟ − 4 2 (III.88)
⎝ k R0 ⎠ ⎝ k ⎠ k R0

m 2τ
B≡− (III.89)
k 4 R03

⎛ τ 2 ⎞⎛ m 2 ⎞ 2τ m ⎛ m2 ⎞
A ' ≡ 2ν k 2 ⎜1 + 2 ⎟ ⎜ 1 + 2 2 ⎟ − 2 3 B ⎜1 + 2 2 ⎟
⎝ k ⎠⎝ k R0 ⎠ k R0 ⎝ k R0 ⎠
(III.88’)
m 2ντ 2 τ m ⎛ 2V mτν m3ν ⎞ mτ
− 4 4 + 2 ⎜ −2 − 2 2 4 ⎟ − 2 D*V
k R0 k R0 ⎝ R0 R0 k R0 ⎠ k R0

τ m 2ν ⎛
m2 ⎞ m 2ντ 3 τ m3 B m ⎛ mντ ⎞
B ' ≡ 2 2 3 ⎜1 + 2 2 ⎟ + 2 4 3 − 2 4 5 − 2 2 ⎜ D*V − 2 3 ⎟ (III.89’)
k R0 ⎝ k R0 ⎠ k R0 k R0 k R0 ⎝ k R0 ⎠

and solving we find

⎧ 2 ⎡
(R12 ( k 2 + τ 2 ) + m 2 ) ⎫
2
⎪ mτ m2 m 2τ ⎛ 1 ⎞⎤ ⎪
⎨ 2 4 ⎢τ + 2 + 2 2 2 ⎜
2 + 2 2 ⎟⎥ − 2 ⎬
⎪ k R1 ⎢

R1 R ( k + τ 2
) + m ⎝ k R1 ⎠ ⎥

R14 ⎪
=⎩ ⎭
1 1
2
Tm
⎧ m 2τ ⎞ ⎞ ⎤ ⎪⎫
⎪ mτ ⎛ d ⎞ 2 ⎡ d m2 ⎛
1
1 ⎛
⎨ 3⎜ ⎟ ⎢ − ⎜ 1 − 2 ⎜
1 − ⎟ ⎟⎟ ⎥ ⎬
⎪⎩ d ⎝ 2 R1 ⎠ ⎢

2 R 1 k 2
R 1
2

⎝ R 1
2
( k 2
+ τ 2
) + m ⎝ k 2
R1 ⎠ ⎠ ⎦⎥ ⎪⎭
(III.90)

Expression (III.90), like (III.55), (III.56) cannot predict the correct critical Taylor
number (depending on m ), since the case m = 0 is not properly defined. In this
sense we can say that (III.90) is valid for every m ≠ 0 . This last equation is more
intricate than the former one, showing that the results are extremely dependent
on the different approaches.

This is precisely what we have had in mind all along this chapter, and
this is the reason why we came back to the axisymmetric case in equation
(III.79).

The pioneering work of Davey et al. (58), on the non-axisymmetric


disturbances is for sure much more elaborated than the calculations we have
presented here. We make though two different approaches; apart some small
differences on the proposed solutions (for the radial dependence), the main
point was the stage we chose for doing the small gap approximation, noticeably,
48

before or after developing all the algebraic calculation involved in the


substitution of the disturbance.

(T − Tc )
Tc

• our calculation
reference (74)

2,5

1,5

0,5

( k − kc )
kc

-0,5 0 0,5 1 1,5

Figure 5.
49

This fact is rather surprising, since one would expect no difference in


doing so; however, having a look on the procedure, it is easy to remark that the
origin of the discrepancy of the results comes out from the small gap
approximation itself.

Indeed, when we have a look on the procedure, the small gap


approximation is a simplification by means of which we expect to get rid of (or
neglect) several terms, when substituting the proposed solutions. Thus, we can
expect that doing this, the final result will depend on the chosen moment for
performing the simplification. However, notice that such a dependence is not
defined automatically (from the algebra’s point of view), but rather as follows:
the obtained results, even if they are different in which concerns their form,
they exhibit the same global behaviour.

Therefore, in one sense (the main one), our results are suitable, in spite of
the differences existing among the coefficients.

The discussion has been principally oriented to the calculation procedure,


which is not without any physical meaning. The small gap approximation is
certainly a geometrical limit, and that is why we are interested on its
mathematical consequences. However, as we are aware of the dependence on
the geometry of the system in most of the problems of fluid mechanics (as we
pointed out in the Introduction), the physical consequences of such
approximation are imminent.

It is clear that the limit d → 0 is not attainable in the reality (as well as
infinitely cylinders). We could notice along the calculations we presented in this
chapter, that the narrow gap approach can be physically understood as a “very
small gap”. This statement is ambiguous from an experimental point of view, as
it gives no idea of the actual dimension the gap should have. However, in
reality this problem is simplified when we think that the gap must be small
with respect to both, the height and the mean radius of the cylinders. And this
is expressed by the fact that we have taken the ratio d R0 tending to zero, and
not simply d .

We will make just a comment about the statement after equations (III.73)
and (III.73’). Let us point out the usefulness of having qi1 , qi 2 , qi1 ', qi 2 ' . Notice
that (III.66’) could have been written as

( q + A) u + Bv = 0
(III.91)
Cu + ( q + D ) v = 0

from which we get the eigenfrequencies we said before. One of them is zero (the
slow frequency), and the other one is negative (the fast frequency). Substituting
these values back into (III.91) lets us define the so called slow and fast modes
50

(as Hohenberg et al. (75) do for Rayleigh-Bénard). It is the slow mode which
becomes unstable. So, time evolution equations for these modes can be written
out, and on such time scale, the dynamics is described by a time-dependent
Ginzburg-Landau equation in the Langevin form. In chapter V we will see how
this is attainable in detail.
51

CHAPTER IV

THE AMPLITUDE EQUATION

We already mentioned in the last two chapters, that expressions for the
Taylor number as a function of the wave number k (this is to say, the stability
curve), have been written out by several authors, and we pointed out the
sensitivity of such results on the approaching ways.

In chapter II we considered the problem of the evolution in time of a


parameter like the velocity vector u . This is in fact a fundamental problem in
fluid dynamics. The question arises now, regarding the time evolution called
amplitude, which is actually the modulus of the vector u . This is an interesting
question, since we must remember that we expect to describe the behaviour of
the fluid taking into account not only the axial disturbances, but those being
non-axisymmetric as well.

There are several aspects that make this time evolution of the amplitude
to be an attractive way for analysing the problem. As we said in the last chapter
we already could establish some difference between slow and fast frequencies;
we will see here how this idea can go further introducing slow and fast
variables, this is to say, in time and position. This will facilitate the application
of the method of separation of variables in a wider manner, because what seems
to be the Fourier coefficients will furnish the slow varying functions we call
amplitude. These functions, therefore, are not constants and depend on the
slow variables.

In other words, the amplitude equation isolates the slow time variation
of the velocity; and this is interesting because then, we will be able to use the
scaling procedure (in terms of a small quantity) as a systematic way for
increasing the detail in the description of the instability.

At this point we can eliminate many degrees of freedom so that the bulk
behaviour of the system is governed by few of them only. So, we want to show
explicitly how to eliminate most of the variables close to the point where linear
stability is lost.

We saw in the last chapter that the neutral stability curve has a minimum
at Tc (along with the dispersion relations), which at a finite value of k reflects
the choice of some optimal scale at which the release of potential energy
overcomes the opposing effects of viscosity. The small parameter is issued from
this scaling. We saw how a basic solution (which is not an exact one) with a
52

periodic structure can be allowed, and it has associated the usual quantities of
amplitude, frequency and wave number. The amplitude equation is the
evolution of the first of them, assuming that everywhere (position and time) the
solution is locally given by the previously determined basic structure.

The present chapter will practically deal all the time on the so called
multiscale perturbation theory. This analytical procedure can continue ad
infinitum, so we cut it at the stage we will appoint by the amplitude equation,
with explicit analytical expressions for the coefficients. This approach was
pioneered on the Rayleigh – Bénard instability by Newell and Whitehead (35);
in their analysis they took into account the temperature and therefore the form
of the equations is not exactly the same, as compared to ours.

This will let us continue to the description of the phase diffusion problem.
This one comes out as a consequence of a perturbation on the amplitude
mentioned above. The procedure that we will illustrate there is a variant of the
method followed by Pomeau and Manneville (36)(76).

i) Form of the amplitude equation.

We just made a comment about the scaling problem. In fact we must make a
distinction between the scaling in the sense of (III.23), and the one we are going
to deal with so on (see chapter I). The first one has been introduced in order to
set the hydrodynamic system of equations in a simpler manner, simplifying
then the form of the coefficients; its importance has already been pointed out,
since it gave a sense to the expression for the Taylor number (II.23), this is to say
the control parameter. However, the other one will give a sense to the
amplitude, which will be identified with the coefficient of the lowest order in a
power series development (on a suitable defined small parameter).

Therefore, such scaling procedure should include some sort of difference


between two quantities, in order to characterize amplitude. So, let us start with
the definition of the scaling factors.

Since in a certain way we are forced to utilize the current letters followed
by most of the authors, we will repeat some of them used in former sections but
having a different meaning. Hence, we define

1
τ ≡ εt , z ≡ε 2z (IV.1)

so the differential operators become


53

∂ t → ∂ t + ε∂τ
1
∂ z → ∂ z + ε 2∂ z
(IV.2)
∂ 2t → ∂ 2t + 2ε∂ t ∂τ + ε 2∂ 2τ
1
∂ 2 z → ∂ 2 z + 2ε 2 ∂ z ∂ z + ε 2 ∂ 2 z

where the small parameter ε is defined by means of the series expressions,

1 3
u = ε 2 u0 + ε u1 + ε 2 u2 +
1 3
v = ε 2 v0 + ε v1 + ε 2 v2 +
(IV.3)
1 3
w = ε w0 + ε w1 + ε w2 +
2 2

T → T0 + ε T1 +

Notice that ε is a small parameter, however there are several ways for defining
it, namely the ratio d R1 or

T ∼ T0 + ε T0 (IV.4)

In spite of their different origin these two quantities are proportional to one
another. This is experimentally supported by Cognet et al. (65). They found that
the difference between the critical Taylor numbers for the onset of wavy vortex
flow and Taylor vortex flow decreases with d ; therefore, since our T cannot be
bigger than the critical value for wavy flow, it must be true that small d (small
gap approximation) implies T very close to T0 , and so ε is a small parameter.
Hence, we can think of ε as a quantity proportional to the shift with respect to
the critical Taylor number for the onset of vortices.

Remark that we are considering first of all the axisymmetric case, so we


do not have to define yet the azimuthal variable.

Reminding (III.19) we can write out the hydrodynamic equations as


follows:

(∂ 2
x + ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) v − T (1 − x ) 2∂ 2 z v

= − ∂ 2 z ( u∂ x u + w∂ z u − Tv 2 ) − ( ∂ 2 x + ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( u∂ x v + w∂ z v )
1 1
(IV.5)
2 2
+ ∂ x ∂ z ( u∂ x w + w∂ z w )
1
2

(∂ 2
x + ∂ 2 z − ∂t ) v − u =
1
2
( u∂ x v + w∂ z v ) (IV.6)
54

∂ xu + ∂ z w = 0 (IV.7)

this being the usual system (III.2) – (III.5) reduced to three equations. It is easy
to see that substituting (IV.1) – (IV.3) into (IV.5) – (IV.7) will give a series of
equations with all sort of orders in ε . We will perform the analysis up to the
order ε 2 , because going all the way to such order will give us a lot of
information about the time evolution of the amplitude. Just for completeness in
the presentation of this work, let us explicitly write the equations as follows.

1
Order ε 2
.

(∂ 2
x + ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) v0 − T0 (1 − x ) 2∂ 2 z v0 = 0 (IV.8)

(∂ 2
x + ∂ 2 z − ∂ t ) v0 − u0 = 0 (IV.8’)

∂ x u0 + ∂ z w0 = 0 (IV.8’’)

Order ε .

{( ∂ + ∂ − ∂ )( ∂
2
x
2
z t
2
x }
+ ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) − T0 (1 − x ) 2∂ 2 z v1

+ {( 2∂ ∂ )( ∂ + ∂
z z
2
x
2
z }
− ∂ t )( 3∂ 2 x + 3∂ 2 z − ∂ t ) − T0 (1 − x ) 4∂ z ∂ z v0

= − ∂ 2 z ( u0 ∂ x u0 + w0∂ z u0 − T0 v0 2 )
1
(IV.9)
2
− ( ∂ 2 x + ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( u0∂ x v0 + w0∂ z v0 )
1
2
+ ∂ x ∂ z ( u0 ∂ x w0 + w0 ∂ z w0 )
1
2

(∂ 2
x + ∂ 2 z − ∂ t ) v1 − u1 + 2∂ z ∂ z v0 =
1
2
( u0∂ x v0 + w0∂ z v0 ) (IV.9’)

∂ xu1 + ∂ z w1 + ∂ z w0 = 0 (IV.9’’)
55

3
Order ε 2
.

{( ∂ 2
x }
+ ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) − T0 (1 − x ) 2∂ 2 z v2

{ }
+ 4∂ z ∂ z ( ∂ 2 x + ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) + 2∂ z ∂ z ( ∂ 2 x + ∂ 2 z − ∂ t ) − T0 (1 − x ) 4∂ z ∂ z v1
2

+ {2 ( ∂ 2 z − ∂τ ) ( ∂ 2 x + ∂ 2 z )( ∂ 2 x + ∂ 2 z − ∂ t ) + ( ∂ 2 x + ∂ 2 z − ∂ t ) ∂ 2 z
2 2
}
+8 ( ∂ z ∂ z ) ( ∂ 2 x + ∂ 2 z − ∂ t ) + 4 ( ∂ z ∂ z ) ( ∂ 2 x + ∂ 2 z ) − T0 (1 − x ) ∂ 2 z − T0 (1 − x ) 2∂ 2 z v0

= − ∂ 2 z ( u0 ∂ xu1 + u1∂ x u0 + w0 ∂ z u1 + w1∂ z u0 + w0∂ z u0 − 2T0 v0 v1 )


1
(IV.10)
2
−∂ z ∂ z ( u0 ∂ x u0 + w0 ∂ x u0 − T0 v0 2 )

− (
1 2
2
∂ x + ∂ 2 z − ∂ t )( ∂ 2 x + ∂ 2 z )( u0∂ x v1 + u1∂ x v0 + w0∂ z v1 + w1∂ z v0 + w0 ∂ z v0 )

−2 ( ∂ z ∂ z ) ( u0 ∂ x v0 + w0 ∂ z v0 )
2

+ ∂ x ∂ z ( u0∂ x w1 + u1∂ x w0 + w0 ∂ z w1 + w1∂ z w0 + w0∂ z w0 ) + ∂ x ∂ z ( u0∂ x w0 + w0∂ z w0 )


1 1
2 2

(∂ 2
x + ∂ 2 z − ∂ t ) v2 − u2 + 2∂ z ∂ z v1 + ( ∂ 2 z − ∂τ ) v0
(IV.10’)
=
1
2
( u0∂ x v1 + u1∂ x v0 + w0∂ z v1 + w1∂ z v0 + w0∂ z v0 )

∂ x u0 + ∂ z w2 + ∂ z w1 = 0 (IV.10’’)

The more interesting fact while having a look on these expressions is that
3
the new time variable τ does not appear until the order ε 2 , and not before.
Even if these equations are the axisymmetric version only, one can think of
many possibilities for making approximations. Let us consider, for instance the
stationary state, this is to say that no one of the functions u0 , v0 , w0 , etc. depend
3
on t , but rather on τ (at order ε 2
).

At this point let us recall the nature of slow and fast variables. We could
have said rather than taking the shift from Tc , taking the shift from kc (the
critical wave number). However, since the stability curve T ( k ) has a minimum
at kc precisely, it comes out that ε is proportional to both shifts. It is convenient
to distinguish between the length scale of the vortices (mainly fixed by the
geometry), and the length and time scales on which the amplitude (that we
define below) varies. We expect in the small gap limit that the dynamics will be
described by the slow mode dynamics; this is to say, that the perturbation
56

scheme we develop here is suggested by the presence of advection terms in the


hydrodynamic equations (this will be much clearer at the moment when we
will consider the non – axisymmetric case). The new time scale is smaller than
the one related to the critical slowing down.

For the small gap approximation, we will make x = 1 2 , as another way


of saying that we will take a constant radial coordinate, i.e. at the middle of the
gap. We propose now

{
v0 = A0 ( z ,τ ) e
ik z z
}
+ c.c. v0 x ( x ) (IV.11)

where c.c. means complex conjugate; the variables z ,τ (and y later) are called
slow (as they are scaled), and the old ones are fast. The x -dependent function
v0 x satisfies the differential equation

v0 xVI − 3k z 2 v0 x IV + 3k z 4 v0 x ''− (k z 6 − 2T0 (1 − x ) k z 2 )v0 x = 0 (IV.12)

where derivatives are total with respect to x . Consequently we find for the
other components,

{
u0 = ( v0 x ''− k z 2 v0 x ) A0 ( z ,τ ) e
ik z z
+ A0 * e
− ik z z
} (IV.13)

w0 =
i
kz
( {
v0 x '''− k z 2 v0 x ' ) A0 e z − A0 * e z
ik z − ik z
} (IV.14)

In fact what we should do is to develop these functions in Fourier series, but for
the sake of clearness, we just kept the first term of the series, then describing the
periodic behaviour at T0 . At order ε ,

{
v1 = K1 + A1e
ik z z
+ A1 * e
− ik z z
+ B1e
2 ik z z
+ B1 * e
−2 ik z z
}v
1x ( x) (IV.15)

such that

K1 ( v1xVI − T0 (1 − x ) 2v1x ) + (V1V2 − V3V2 ) A0 = 0


IV 2
(IV.16)

where K1 is complex, and

V0 ≡ v0 x ; V1 ≡ v0 x ' ; V2 ≡ v0 x ''− k z 2 v0 x (IV.17)

and the relation of A1 , B1 (and their respective complex conjugates) with A0 are
A1 ⎡( d x 2 − k z 2 ) v1x − T0 (1 − x ) 2v1x ⎤ + 2ik z ( 3V4 − 3k z 2V2 − T0V9 ) ∂ z A0 = 0
3
(IV.18)
⎢⎣ ⎥⎦
57

B1 ( v1xVI − 12k z 2 v1x IV + 48k z 4 v1x ''− 64k z 6 v1x − T0 (1 − x ) 2v1x )


⎡ 1
= ⎢ −2T0V0 2 k z 2 − (V1V2 − V3V2 ) + 4k z 2 (V1V2 − V3V2 ) ''
IV
(IV.19)
⎣ 2
−8k z 4 (V1V2 − V3V2 ) − V2V4 − (V3V3 ) '⎤⎦ A0 2

and we have defined

Vn ≡ v0 x n − k z 2 v0 x n − 2 , ∀n ≥ 2
(IV.20)
Vn N = Vn + N

therefore, u1 and w1 are

(
u1 = K1v1x ''− (V0V2 + V3V2 ) A0
2
)
+ (V2 A1 + 2ik zV0∂ z A0 ) e + (V2 A1 * −2ik zV0 ∂ z A0 *) e
ik z z − ik z z

⎡ ⎤ 2ik z (IV.21)
+ ⎢( v1x ''− 4v1x k z 2 ) B1 − (V0V2 − V3V2 ) A0 2 ⎥ e z
1
⎣ 2 ⎦
⎡ ⎤ −2ik z
+ ⎢( v1x ''− 4v1x k z 2 ) B1 * − (V0V2 − V3V2 ) A0 *2 ⎥ e z
1
⎣ 2 ⎦

w1 = ⎡(V0V2 + V3V2 ) ' A0 − K1v1x '''⎤ z


2
⎣ ⎦
⎛ i 1 ⎞ ik z
+ ⎜ V3 A1 − 2V1∂ z A0 − 2 V3∂ z A0 ⎟ e z
⎜ kz kz ⎟
⎝ ⎠
⎛ i 1 ⎞ −ik z
− ⎜ V3 A1 * +2V1∂ z A0 * + 2 V3∂ z A0 * ⎟ e z (IV.22)
⎜ kz kz ⎟
⎝ ⎠
i ⎡ ⎤ 2ik z
+ ⎢
2k z ⎣
( v1x '''− 4v1x ' k z 2 ) B1 − (V0V2 − V3V2 ) ' A0 2 ⎥ e z
1
2 ⎦
i ⎡ ⎤ −2ik z
− ⎢
2k z ⎣
( v1x '''− 4v1x ' k z 2 ) B1 − (V0V2 − V3V2 ) ' A0 *2 ⎥ e z
1
2 ⎦

Notice that equation (IV.22) has a linear dependence on z . This cannot be likely,
otherwise there would be a linear growth superposed to the periodic behaviour.
Such a term is a consequence of two facts: to propose a z -independent factor in
(IV.15), and multiply directly each term with its corresponding complex
conjugate, then eliminating all z -dependence.
58

As we have already pointed out, the slow time derivative appears by the
3
first time at the order ε 2 . We call amplitude equation the expression coming out
as the coefficient of eik z . In fact, given the proposed solutions we shall expect to
have many amplitude equations. Already at order ε we can see that the
intricacy of the expressions at higher order will be mostly due to the non -
specification of the functions depending on x only (i.e. the radial part),
Therefore, let us propose at this stage a very simple sinusoidal dependence on
x and look at higher orders; at the same time we will try to write a Fourier
series like

v0 = ∑ ( K 0 n + A0 n eik z + A0 n * e − ik z ) sin nπ x (IV.23)


n =0

Substituting (and seemingly at each order) we find “selection principles” which


are nothing more but consequences of the orthonormality satisfied by periodic
functions. Hence, (IV.23) implies at lowest order in ε ,

T0 =
1 2
k2
( π + k )
2 3
(IV.24)

and since we have projected onto the first normal mode

v0 = ( A01eik z + A01 * e− ik z ) sin π x (IV.23’)

Similarly

v1 = ∑ { K1n + A1n eik z + B1n e 2ik z + c.c.} sin nπ x (IV.25)


n=0

and every time we project onto the fundamental mode, we integrate from 0 to 1
(71)(72)(77). In this way we show that

K11 = K1n = 0 ∀n ≥ 3
(IV.26)
K12 =
1

( π 2 + k 2 ) A01
2

We remind that we make x = 1 2 everywhere (except in the argument of the


sinusoidal functions). The amplitude we mentioned before is indeed A01 ; and
i
A11 = − ( k − 1) ∂ z A01
k (IV.27)
A1n = 0 ∀n ≥ 2
59

8k 2T0
B11 = A012
⎛ ⎞
3π ⎜ 2T0 k 2 − (π 2 + 4k 2 ) ⎟
1 3
(IV.28)
⎝ 2 ⎠
B12 = 0 , Bin = 0 ∀n even

In this way we could calculate B1n for each n odd. Each one of these terms is
more and more negligible as n grows; however, they all satisfy that

B1n ∝ A012 ∀n odd (IV.28’)

From here we can write expressions for u1 and w1 , that we omit in order to keep
going to the next order. Hence, we propose

v2 = ∑ { K 2 n + A2 n eik z + B2 n e 2ik z + C2 n e3ik z + c.c.} sin nπ x (IV.29)


n=0

Again

K 21 = K 2 n = 0 ∀n ≥ 3

⎡ −2π (π 2 + k 2 ) ( A01 A11 * + A01 * A11 ) + π ( A01U11 * + A01 *U11 )


1
K 22 = 2 ⎣
(IV.30)
16π
iπ ⎤
+ik ( A01W11 * − A01 *W11 ) + (π 2 + k 2 ) ( A01 * ∂ z A01 − A01∂ z A01 *) ⎥
k ⎦

such that

U11 , U11*, A11 , A11*,W11 ,W11 * ∝ ∂ z A01

Taking into account that

+
A01 * B1,2 n −1 ∝ A01 A01 ∀n ∈
2
, (IV.31)

we can write for the axisymmetric case

⎡ 3 ( n − 1) 4n ⎤ A01 A01
2

∂τ A01 ∝ (π 2 + k 2 ) A01 +
1 2T0
⎢ − ⎥
(π + k 2 ) ⎣ ( 2n − 3) ( 2n + 1) ⎦ ( 2n − 1)
2
2 2

(IV.32)
⎧ T0 ⎫ 2
⎨( k − 1) ⎡⎣⎢3 (π + k ) − T0 ⎤⎦⎥ + (π + k ) + 6k (π + k ) − ⎬ ∂ z A01
1 2 2 3 2 2 2
+ 2 2 2 2

(π + k2 ) ⎩ 2⎭
2
2 2

The Taylor number appearing in (IV.32) is the axisymmetric one. In Davey (52)
and Davey et al. (58) we can find amplitude equations having the same form
like (IV.32), except that in the first paper the second derivative with respect to z
60

does not appear. This is just because of the proposed solutions which do not
depend on the axial coordinate, this is to say that the mean motion equations
obtained by equating terms which are independent of z give rise to the system
they deal with. Nevertheless, the form of the coefficients for the linear and cubic
terms is the same than ours. In Davey et al. (58) the comparison is
straightforward since they also deal with the whole hierarchy of equations.

We can now think of the non – axisymmetric case, therefore taking into
account the azimuthal dependence; so the hydrodynamic equations become:

(∂ x
2
+ ∂ z 2 − 2 (1 − x ) ∂ y + ϒ∂ y 2 − ∂ z ) v − u =
1
2
( u∂ x v − 2v∂ y v + w∂ z v ) (IV.33)

∂ xu − 2∂ y v + ∂ z w = 0 (IV.34)

(∂ x
2
+ ∂ z 2 − 2 (1 − x ) ∂ y + ϒ∂ y 2 − ∂ t )( ∂ x 2 + ϒ∂ y 2 + ∂ z 2 ) ( ∂ x 2 +
+∂ z 2 − 2 (1 − x ) ∂ y + ϒ∂ y 2 − ∂ t ) v − 2T (1 − x ) ∂ z 2 v =
(IV.35)
= − ⎡⎣( ∂ x 2 + ∂ z 2 − 2 (1 − x ) ∂ y + ϒ∂ y 2 − ∂ t )( ∂ x 2 + ∂ z 2 + ϒ∂ y 2 ) +
1
2
+2∂ x ∂ y ⎦⎤ ( u∂ x v − 2v∂ y v + w∂ z v ) + ∂ x ∂ z ( u∂ x w − 2v∂ y w + w∂ z w )
1
2

We follow then exactly the same method to find equation (IV.32), modifying
(IV.2) by adding

1
∂y → ∂ y + ε 2 ∂Y
1
(IV.36)
∂ y2 → ∂ y 2 + 2ε 2 ∂ y ∂Y + ε∂Y 2

in the small gap limit the azimuthal variation becomes slow in the y coordinate,
even though periodic in θ , being Y therefore the new slow azimuthal variable.
Again, for simplicity we make x = 1 2 (as small gap approximation) and we
have considered ϒ ∼ 0 .

1
At order ε 2
we propose,

n =0
{
v0 = ∑ n K 000 + n K 001eimy + n A010 eik z + n A011e (
i k z + my )
+ n A01−1e (
i k z - my )
}
+ c.c. sin nπ x (IV.37)

where the role of the indices can be easily seen.

Even though expression (IV.37) contains m , the unique real solution at


this order gives always
61

(π + k2 )
2 3

T0 = (IV.38)
k2

this implying that we shall have m = 0 at the onset of the Taylor instability.

The expressions for u0 , w0 are

u0 = ⎡⎣ − (π 2 + k 2 )( 1 A010 eik z + 1 A010 * e −ik z )

(
− (π 2 + k 2 + im ) 1 A011ei( k z + my ) + 1 A01−1 * e − i( k z -my ) ) (IV.39)

− (π 2 + k 2 − im ) ( A 1
01−1 e(
i k z - my )
+ 1 A011 * e − i( k z + my )⎦ sin π x
) ⎤

iπ 2
w0 =
k
( π + k 2 )( 1 A010 eik z − 1 A010 * e− ik z ) cos π x

+ A011ei( k z + my ) ⎡⎣ 2m sin π x − iπ (π 2 + k 2 + im ) cos π x ⎤⎦


11
k
+ A011 * e − i( k z + my ) ⎣⎡ 2m sin π x + iπ (π 2 + k 2 − im ) cos π x ⎦⎤
11
(IV.40)
k
− 1 A01−1ei( k z − my ) ⎡⎣ 2m sin π x + iπ (π 2 + k 2 − im ) cos π x ⎤⎦
1
k
− 1 A01−1 * e i( k z − my ) ⎡⎣ 2m sin π x − iπ (π 2 + k 2 + im ) cos π x ⎤⎦
1 −

The expressions at higher orders in ε are too lengthy to be written here (so we
show the expressions for u1 , v1 , w1 in appendix A), however we show the main
results concerning directly the amplitude equation.

Notice that all along these calculations, the expressions for the functions
v0 , u0 , w0 , v1 , u1 , w1 , etc. what we have written so far, are just approximate results,
since there is an infinite number of possible combinations (along the summation)
or exponents giving the same order; we have kept just the first three, as the
contribution of the remaining terms is negligible.

At order ε we find,

1
A110 = ik
(T − 3 (π
0
2
+ k2 )
2
)∂ 1
A010 (IV.41)
(π + k )
z
2 2 3

1
( π 2 + k 2 ) 1 A010
2
2
K100 = , n
K100 = 0 ∀n ≠ 2 (IV.42)

62

2 ⎡⎢(π 2 + k 2 )(π 2 + k 2 + im ) ( 2ik ∂ z − ∂Y ) + ik (π 2 + k 2 + im ) ∂ z − ikT0∂ z ⎤⎥


2

1
A111 = ⎣ ⎦ 1A
⎡(π 2 + k 2 + im )2 (π 2 + k 2 ) − (π 2 + k 2 )3 ⎤
011

⎣⎢ ⎦⎥
(IV.43)

2 ⎡⎢(π 2 + k 2 )(π 2 + k 2 − im ) ( 2ik ∂ z − ∂Y ) + ik (π 2 + k 2 − im ) ∂ z − ikT0∂ z ⎤⎥


2

1
A11−1 = ⎣ ⎦ 1A
⎡(π + k − im ) (π + k ) − (π + k ) ⎤
2 3 01−1
2 2 2 2 2 2
⎢⎣ ⎥⎦
(IV.44)

n
A120 =
⎡T 4k − ( n π
2
4k 2
2 2
+ 4k ) ⎤
2 3
{∫ sin π x sin nπ xdx} ( −T
1

0
2
0
1
A010 2 + 2 (1 − 4m 2 ) 1 A011 1 A01−1 )
⎣⎢ 0 ⎦⎥
(IV.45)

n
A121 =
4 {∫ ⎡⎣2k (im − T ) sin π x + 2π m sin 2π x + 2π m cos 2π x⎤⎦ sin nπ xdx} A
1

0
2
0
2 2 2 2
1 1
A011
⎡T 4k 2 − ( n 2π 2 + 4k 2 + im ) ( n 2π 2 + 4k 2 ) ⎤
2 010

⎢⎣ 0 ⎥⎦
(IV.46)

and doing similarly one can show that

n
A122 ∼ 1 A0112
n
A12−1 ∼ 2 1 A010 1 A01−1 (IV.47)
n
A12− 2 ∼ 1 A01−12

3
In this way one also finds v1 , u1 , and w1 . At order ε 2
(if we define 1 A010 ≡ A )

(π + k 2 ) ( −∂τ A + C0 A + C1∂ z 2 A + C2i∂ Y ∂ z A + C3∂ Y 2 A ) = NL


2
2
(IV.48)

where

6k 2 − π 2
C0 =
2
(π + k 2 )
1 2
, C2 =
k (π 2 + k 2 )
, C3 = −
1
2 (π + k 2 )2

(IV.49)
3 (π + k ) ⎛ π 2
⎡ ⎞⎤
⎞ ⎛⎜
2 2
1 ⎢ 6k 2 2 1 ⎟⎥
C1 = + − + ⎜ 2 − 2 ⎟ 1− 2 −
4 ⎢π 2 + k 2 2 2k 2 ⎜ π + (π + k ) ⎠ ⎦
2 2 ⎟⎥
2
⎝ k ⎠⎝ k 2

63

and NL denotes the contribution of the non – linear terms to the amplitude
equation which is seen to be of the form

2
NL ∼ A A (IV.49’)

then defining a proportionality constant C4 , which can be found following the


same method. The wave number is scaled by the gap width (see (III.23)).

Equation (IV.48) is an important relation for the time evolution of the


amplitude. Notice the complex coupling term of the axial and azimuthal
variation of A . We have not yet looked at the physical meaning of the
coefficients, however one can already ask why C3 < 0 . We will see in the next
sections that indeed this happening is nothing more but the consequence of the
oversimplification of the choice we have made of the orthogonal basis in the
series developments. We will see then the characteristics that such functions
must satisfy (in accordance with the boundary conditions). It is worthwhile
however, to calculate approximate numerical values for them; under the
condition

⎛1 ⎞
v0 ⎜ , y, Y , z , z ,τ ⎟ = F ( y, Y , z , z ,τ ) (IV.50)
⎝2 ⎠

which is a normalization condition in the radial coordinate, F being a function


of the remaining variables only. The values of the coefficients are thus

C0 ≈ 26,16 , C1 ≈ 1, 019
(IV.51)
C2 ≈ 0, 7182 , C3 ≈ −0, 0096

Comparing (IV.51) with the results obtained by Tabeling (60) we can easily see
that C0 is the same; C1 is practically the same; C2 is more different, and C3 is
truly far from being congruent with Tabeling’s value; however, notice that
C3 ∼ 0 , so we should not worry that much about the sign.

The usefulness of the amplitude is pointed out for the Rayleigh – Bénard
problem in the work of Newell and Whitehead (35), where demonstrated that
using a single differential equation (i.e. the amplitude equation) describing the
slow time and spatial variation of the convective roll pattern close to the onset
of convection, the equation turns out to have the form of a time – dependent
Ginzburg – Landau one.

The consequence of this is that a continuous finite bandwidth of modes


can be readily incorporated into the description of the Rayleigh – Bénard
system after the onset.
64

We cannot say that we possess the equivalent result for the Couette –
Taylor system, however we do have the corresponding amplitude equation,
and in chapter V we will see how the consideration of the fluctuations can take
us further in such direction.

We can conclude directly from here that the problems are coming out
from the slow azimuthal variable (as it contributes more and more to the
amplitude equation, the numerical values of the coefficients disagree with the
expected ones). On the basis of what we will be saying in the next sections, we
can argue that the oversimplification of the chosen basic functions has mainly
consequences on the azimuthal behaviour of the amplitude. Let us now have a
look onto the physical meaning of some coefficients.

ii) Phase diffusion.

Equation (IV.48) can be written as

∂τ A = C0 A + C1∂ z 2 A + C2i∂Y ∂ z A + C3∂Y 2 A − C4 A A


2
(IV.52)

This expression came out after the approximation we already pointed out. We
neglected the quantity ϒ which completed the Laplacian operator; in fact, since
this term was associated to the second partial derivative with respect to Y , we
expect that including it into the calculation might appropriately modify C3 ;
indeed it does; C3 becomes

3 1
C3 = ϒ− (IV.53)
2 2 (π + k 2 )
2

Even we know that our approximation for the coefficients is quite rough
(although the form of the amplitude equation is essentially correct), it is
worthwhile to see how (IV.53) imposes a geometrical limitation if we ask C3 to
be positive. If we scale the radial variable on R2 the gap must be

⎛ R2 2 ⎞
d t R2 ⎜ 2 − 1⎟
6 6
(IV.54)
⎝ R1 ⎠

This inequality tells us very little of what is going on, however it shows as we
expected, that the form of C3 (and in fact of the other coefficients as well) is
intimately related with the geometry of the system. In mathematical terms that
means the basis functions (as they are a representation of the boundary
conditions).
65

In order to get a physical picture of all this, let us have a look on the
dynamics of a phase variable we define below. This is the kind of analysis
pioneered by Pomeau and Manneville on the Rayleigh – Bénard instability
(36)(76). We do not develop here the details concerning this method (see the
references); however, we can say that quantities like a modulus quantifying the
strength of the unstable mode and a phase specifying to the position of the cells
(vortices) can directly be observed in experimental devices.

This point of view is theoretically very interesting while describing


secondary instabilities (wavy and modulated wavy vortex flows); since they are
close to Taylor’s instability threshold and involve long wavelength modulations,
we can use the analysis in the framework of the amplitude equation. This
means, slight modifications of the neutral mode which is generated by
translational invariance in the case of infinite cylinders (i.e. Taylor vortex flow).

We simply propose for the amplitude a z -dependence of the form

A (Y , z ,τ ) = A00 eiKz (IV.55)

this is to say the stationary state at every scale. We will see in the next section
how the slow variables Y and τ are related.

Substituting (IV.55) into the amplitude equation (IV.52) implies that

1
⎛ C − C1 K 2 ⎞ 2
A00 = ⎜ 0 ⎟ (IV.56)
⎝ C4 ⎠

The wave number K has a meaning in the z direction; since we are thinking in
terms of slow variables, K shall be related to the position in some sense of the
Taylor vortices. Let us consider this possibility in a more general context, by
taking rather than (IV.55),

A = eiKz eiϕ ( A00 + μ ) (IV.57)

such that

K = k − kc
(IV.58)
ϕ = ϕ (τ , z , Y ) ; μ = μ (τ , z, Y )

so effectively K is a measure of the compression (or dilation) of the vortices


along z . The slow function ϕ represents some sort of disturbance or
perturbation to the phase, and we assume its azimuthal (and time) dependence,
since it is coupled to the axial one. Similarly, to complete the description of the
66

effects of such a modification, we must add a slow function μ to the amplitude;


we expect first of all, to find a relationship between μ and ϕ .

Introducing (IV.57) into (IV.52) implies (using (IV.56) also)

1
⎛ C1 K ⎞ ⎛ C0 − C1 K 2 ⎞ 2
μ =⎜ ⎟⎜ ⎟ ∂ zϕ (IV.59)
⎝ C1 K − C0 ⎠ ⎝
2
C4 ⎠

To get (IV.59) we have assumed that all the nonlinear terms on the gradient of
ϕ (or μ ), and all those terms producing any kind of nonlinear terms, are
negligible (and this thinking that μ ∝ ∂ zϕ in advance). This is reasonable
because of the smallness of the values taken by μ and ϕ , and as we said before,
the proportionality physically consistent between these functions. In order to
eliminate ∂τ μ we assume slow steady conditions on this function (this being
acceptable if we think on wavy vortex flow).

Finally, the slow time evolution equation for ϕ is

⎧ ⎛ C0 ⎞ ⎫
⎪ ⎜ ⎟ − 3K ⎪
2

⎪ C ⎪ 2
∂τ ϕ = C1 ⎨ ⎝ 1 ⎠ ⎬ ∂ z ϕ + C3∂Y ϕ − C2 K ∂ Y ϕ
2
(IV.60)
⎪ ⎛ ⎞
− K2 ⎪
C
⎪⎩ ⎜⎝ C1 ⎟⎠
0
⎪⎭

or

∂τ ϕ + C2 K ∂Y ϕ = D ∂ z 2ϕ + D⊥ ∂Y 2ϕ ≡ ∂τ ϕ (IV.60’)

such that

⎧ ⎛ C0 ⎞ ⎫
⎪ ⎜ ⎟ − 3K ⎪
2

⎪ C ⎪
D⊥ ≡ C3 ; D ≡ C1 ⎨ ⎝ 1 ⎠ ⎬ (IV.61)
⎪ ⎛ C0 ⎞ − K 2 ⎪
⎪⎩ ⎜⎝ C1 ⎟⎠ ⎪⎭

Equation (IV.60’) tells us that we must scale again the time evolution if we want
to have a diffusion equation for the phase function ϕ . In other words, we are
saying that C2 K represents the velocity (in the direction of the mean flow) with
which the phase perturbation propagates.

It is clear now that asking C3 > 0 is actually a requirement for the


stability, since it represents a diffusion coefficient on the Y coordinate.
67

Similarly, D must be positive; this imposes conditions on K ( = k − kc ) because


we know from (IV.49) that ( C0 C1 ) > 0 ; hence

−1 ⎛ C0 ⎞ < K < C0 (IV.62)


3 2
⎜ C⎟
⎝ 1⎠ C1

In accordance with Eckhaus, and Kogelman and Diprima (11), expression (IV.62)
shows a subinterval of possible wave numbers for stable vortex flow, inside the
interval of possible Taylor vortex flows.

For completeness in the presentation, let us calculate the form of the


amplitude equation at order ε 2 . As it should be expected this means the
derivation of new terms in the amplitude relation, while keeping all the
nonlinear terms. We simply show here the results. We must keep in mind that
the amplitude equation is issued from the e terms.
ik z

So, what we actually do is the continuation of equation (IV.48); as usual,


we project on sin π x , and make

1
A210 ≡ A2 ; 1
A110 ≡ A1 ; 1
A010 ≡ A0 (IV.63)

In fact, the amplitude equation we get has three different amplitudes A2 , A1 , A0


inside. It should be possible to reduce all of them only to A0 ; however our
calculation is mainly motivated by the proposed amplitude equation appearing
in the reference (59), where several terms absent in (IV.52) seem to come out
from higher order calculations in ε .

The expression found at order ε 2 is:

⎡3 1 ⎤
ik ⎡3 (π 2 + k 2 ) − T0 ⎤ ∂ z A2 − (π 2 + k 2 ) ∂Y A2 + ⎢ (π 2 + k 2 ) + 8k 2 (π 2 + k 2 ) − T0 ⎥ ∂ z 2 A1
2 2 2

⎢⎣ ⎥⎦ ⎣2 2 ⎦

+ (π 2 + k 2 ) ⎡⎣ ϒ (π 2 + k 2 ) − 1⎤⎦ ∂Y 2 A1 − (π 2 + k 2 ) ∂τ A1 + 6ik (π 2 + k 2 ) ∂ Y ∂ z A1 + k 2T0 A1


1
2
− ⎡⎣(π + k ) ξ 0 + 1⎤⎦ (π + k ) ∂Y A0 − 2ik (π + k ) ( 3 − 2C1 ) ∂ z A0
2 2 2 2 2 2 3

⎡ ⎤
−2ik (π 2 + k 2 ) ⎢ 2ϒ − 2C3 + C2 ⎥ ∂Y 2 ∂ z A0 + 2ik ⎡⎣(π 2 + k 2 ) 2C0 − T0 ⎤⎦ ∂ z A0
1
⎣ 2k ⎦
⎡ 1 ⎤
− (π 2 + k 2 ) 2 ⎢ 2kC2 − 1 + C1 ⎥ ∂Y ∂ z 2 A0 + (π 2 + k 2 ) ( ϒ − C3 ) ∂ Y 3 A0
⎣ 2 ⎦
+ (π 2 + k 2 ) ( ϒ − C3 ) ∂Y 3 A0 − (π 2 + k 2 ) [ 4ik ∂ z A0 − ∂ Y A0 ] C4 A0 = NL
2

(IV.64)
68

The underlined terms are those proposed by the authors mentioned above,
3
which are not predicted by the ε 2 expression. The quantity ξ 0 will be defined
in a general manner in the next section, where we look at the general form of
the coefficients.

iii) The coefficients.

In this section we present what we already announced in the last two sections,
concerning the basis functions and orthogonality conditions. The starting point
is, as usual

∂ t u + T 1 2 (1 − x ) ∂ y u + {u∂ x u + v∂ y u + w∂ z u} − v 2 = −∂ x p + ( ∂ x 2 + ∂ z 2 ) u + T 1 2 (1 − x ) v
1
2
(IV.65)

∂ t v + T 1 2 (1 − x ) ∂ y v + {u∂ x v + v∂ y v + w∂ z v} = ( ∂ x 2 + ∂ z 2 ) v + T 1 2u (IV.66)

∂ t w + T 1 2 (1 − x ) ∂ y w + {u∂ x w + v∂ y w + w∂ z w} = −∂ z p + ( ∂ x 2 + ∂ z 2 ) w (IV.67)

∂ xu + ∂ y v + ∂ z w = 0 (IV.68)

where p is the pressure and we have made

1
1 ⎛ R ⎞ 2 ν
x = ( r − R1 ) , y = ⎜2 1 ⎟ θ , t= T (IV.69)
d ⎝ d ⎠ d2

In this system we can make a generalization by calling (1 − x ) a function of x ,


V ( x ) . Next, consider the series (IV.2) and (IV.3), (IV.36) 6 . Again, we perform
the same steps of the multiscale perturbation method.

Let us define the complete basis of functions

{P }jln (IV.70)

where the indices denote: j - coefficient of k ; l - coefficient of m , and n a real


number. They are such that

6 We must add a series development for the pressure, being this completely analogue to the
other ones.
69

1
∫0
Pjln Pjln ' dx ≡ δ nn '
(IV.71)
∀ j , i, l ; (i )
Pjln (i )
= Pjln ( x ) , ∀n

and from the background of former sections, we know that

i∈ ∪ {0} ; j∈ +
∪ {0} ; l∈ (IV.72)

hence (the index i denotes the i -th derivative on x )

v0 = ∑ A10( n) eik z P10 n ( x ) (IV.73)


0

such that

(∂ + ∂ z 2 ) v0 − T0V ( x ) ∂ z 2 v0 = 0
3
x
2
(IV.74)

The critical Taylor number T0 is

∑ A10 n ∫ P101 ( x ) ( Dx 2 − k 2 ) P10 n ( x ) dx


1 3
0
0
T0 = n
(IV.75)
k 2 ∑ 0 A10 n ∫ V ( x ) P10 n ( x ) P101 ( x ) dx
1

0
n

At order ε we find that the orthogonality of the amplitude set of functions fixes
the integro – differential equation that P10n must satisfy; namely,

∑ A10 n ∫ P101 ( Dx 2 − k 2 ) P10 n dx


1 3
0

(D − k 2 ) P10 n ( x ) = V ( x ) P10 n
3 0
2 n
(IV.76)
∑ A10 n ∫ V ( x ) P10 n P101dx
x 1
0
0
n

At the same time, the solvability condition takes the form

∂τ 0 A10 n + ξ 0T01 2 ∂Y 0 A10 n = 0 , ∀n (IV.77)

where

∫ V ( x ) P ( x ) ( D − k ) P ( x ) dx
1
2 2 2
101 x 10 n
ξ0 ≡ 0
(IV.78)
∫ P ( x ) ( D − k ) P ( x ) dx
1
2 2 2
0 101 x 10 n

Equations (IV.77) and (IV.78) remind us that we are dealing with a rotating
system; this is to say that as we go deeper into the approximation (to higher
70

orders in ε ) we must scale each time our variables; but these expressions tell us
precisely that it is only the azimuthal coordinate the one affected by such a
change of scale.

In other words, we must write again the equations at order ε , taking into
account this time that we are in a rotating frame of reference; so, v1 = 0 and u1
becomes

u1 = ∑ eik z P10 n ( x ) (T0 −1 2 ∂τ + V ( x ) ∂Y − 2T0 −1 2ik ∂ z ) 0 A10 n


n

⎛ ⎞⎛ ⎞
− T0 −1 ⎜ ∑ 0 A10 n eik z ( Dx 2 − k 2 ) P10 n ⎟ ⎜ ∑ 0 A10 n eik z Dx P10 n ⎟ (IV.79)
⎝ n ⎠⎝ n ⎠
⎛ ⎞⎛ ⎞
+ T0 −1 ⎜ ∑ 0 A10 n eik z Dx ( Dx 2 − k 2 ) P10 n ⎟⎜ ∑ 0 A10 n eik z P10 n ⎟
⎝ n ⎠⎝ n ⎠

and similarly for w1 from the incompressibility condition. We can then go on to


order ε 3 2 . One should expect that at his order we shall scale again the
azimuthal variable (or the time).

The final equation at this order is in terms of v2 , v1 , v0 , plus the nonlinear


part. The part on v2 disappears from the equation at the moment we project
onto the fundamental mode (because of orthogonality); the terms on v1 take
part into the solvability condition, but anyhow as we said above v1 = 0 because
of (IV.77); the derivatives with respect to z ' (and ∂ z 2 ) of v0 vanish since we
have to introduce the fact that T0 is the solution of the minimization procedure
on the stability curve, and the remaining terms constitute the linear part of the
amplitude equation. As usual we show that the nonlinear terms are
proportional to A A (by making ∂τ → −ξ 0T01 2∂Y ). Then
2

−∂τ ' A + C0 A + C1∂ z 2 A + iC2 ∂Y ∂ z A + C3∂Y 2 A = ξ 0T01 2 A + C4 A A


2

or

∂τ ' A = C0 A + C1∂ z 2 A + iC2 ∂Y ∂ z A + C3∂Y 2 A − C4 A A


2
(IV.80)

in accordance with (IV.52); and the coefficients are,

C0 ≡ T0 k 2 ϒ ∫ V ( x ) P10 n ( x ) P101 ( x ) dx
1
(IV.81)
0

C1 ≡ −12k 2 ϒ ∫ P101 ( x ) ( Dx 2 − k 2 ) P10 n ( x ) dx


1
(IV.82)
0
71

C2 ≡ −8k ϒ ∫ (T01 2V ( x ) − ξ 0 ) P101 ( x ) ( Dx 2 − k 2 ) P10 n ( x ) dx


1
(IV.83)
0

C3 ≡ ϒ ∫ (ξ 0 − T01 2V ( x ) ) P101 ( x ) ( Dx 2 − k 2 ) P10 n ( x ) dx


1 2
(IV.84)
0

where we have defined

{∫ }
−1
P101 ( x ) ( Dx 2 − k 2 ) P10 n ( x ) dx
1 2
ϒ≡ (IV.85)
0

The very first comment is to observe the similarity of these expressions


with those obtained on a less general kind of polynomials basis by Yahata (78);
notice also the similarity of C0 with the term called ε −1 by Davey (equation 6.8
in reference (52)). In the context of what we said before, we had made x = 1 2 ,
this is to say more generally

V ( x) ≡ V ∈
(IV.86)
hence ξ0 = V

So, taking V ( x ) = 1 2 we recover expressions (IV.49) for C0 and C1 . Regarding


the polynomials of the basis, and reducing them to the sinusoidal functions we
are used to, does not bring any light onto the problem. On the contrary, we find
a C3 (the phase diffusion coefficient) which is negative. We can therefore
conclude that P10 n ( x ) = sin nπ x is worse as a basis than we thought, and making
(IV.86) has direct consequences on the sign of C3 .

By comparison with Tabeling (60), and Brand and Cross (61) calculations
with our results, we can conclude that their scaling procedures are equivalent to
each other.

Finally, let us add a comment on Hall’s work about the amplitude (40).
The central point of his approach is that in (IV.5) – (IV.7) he points out, we did
not take into account a θ -dependence of the pressure (which should appear in
the equation for v ). Following essentially the same steps, Hall finds in the
axisymmetric case an amplitude which has exactly the same form like ours,
with corrections in the form of the coefficients. His remark concerning the
driving of the radial mean velocity field at order ε 3 2 , as a consequence of the
Y ’s correction at order ε and continuity equation, is quite pertinent; however
he does not talk about the behaviour, precisely due to the same aspects, of the
axial component w . The analysis he makes afterwards is hard to be compared
with those of Tabeling, Brand and Cross, and ours, since the form of the
developments in ε is different. The consequence is that he finds not one but
two amplitude equations at order ε 3 2 which are not just one amplitude, but
72

three (i.e. complex conjugate amplitudes of two pressure terms, plus the usual
one).

Hall’s assumption was inspired on the work of Stewartson and Stuart (79)
and Davey, Hocking and Stewartson (80) on the plane Poisseuille flow. The
only thing we could argue here is that both hydrodynamic problems (i.e. Taylor
and Poisseuille) possess differences concerning the mathematical treatment.
This last hydrodynamic instability had in fact the same history than Taylor’s;
this is to say, that in the first analysis to get an amplitude equation, the pressure
was eliminated from the very beginning, however it was afterwards that it was
pointed out the necessity of keeping such terms.

The conclusion is that in any case, Hall’s remark must be taken seriously;
however, our results show at least a good agreement with several theoretical
and experimental analysis as we saw above.
73

CHAPTER V

FLUCTUATIONS

So far we have been dealing with the hydrodynamic system of


differential equations, for a cylindrical geometry which is the representative one
for the Taylor problem.

It was in the last chapter that we found the amplitude equation,


underlining the relevance of the analytical form of its coefficients; while doing
so, we introduced a complete set of orthogonal polynomials, satisfying the
equation (IV.76). However, a point we have not considered in a general manner
so far is that of the boundary conditions appropriate to the Taylor problem.
While performing the analysis of the effects of the fluctuations on the
hydrodynamic equations, we will regard at the boundary conditions that must
be satisfied by the basic functions. 7

Let us draft the history associated to this problem, in order to have a


general idea of the interest that some works have devoted to the hydrodynamic
fluctuations.

As it might be expected, it is around the problems presenting instabilities,


and leading to ordered structures that much of the work of fluctuations was
concentrated. In which concerns hydrodynamics, it is quite likely that most of
the interest has been turned to the Rayleigh – Bénard problem, and this is surely
due to the fact that it presents many basic properties of fluids.

Regarding the effects of fluctuations on hydrodynamic systems, several


classical works have been done (see (10)); such is the case of Zaitsev and
Shliomis (81) who calculated the velocity correlation function using a linear
theory. Then the following question aroused: which are the effects of the
nonlinear coupling of fluctuations on the mean field results?

Starting from the basic ideas of Landau and Lifchitz (2)(82), the works of
Graham (83), Swift and Hohenberg (75) have clarified the role of the
fluctuations in the Bénard problem (see also(84)(63)(85)).

The situation for the Taylor instability is much less treated. However an
important step in this subject is given in Walgraef, Borckmans and Dewel
(59)(86), when they analyse the effects of the fluctuations in the transition to

7 Notice however that such conclusion is of a purely mathematical order and no physical
interpretation should be tempted from here. Equations (V.19) will illustrate this.
74

Taylor vortex flow in finite geometries; they propose a pseudo potential


function, from which they find the way the mean field transition is modified by
introducing the fluctuations, and this depending on the size of the system. Since
their Ginzburg – Landau function has a form suggesting the possibility to be
issued from our amplitude equation, this motivated us to have a look on the
form the amplitude of the fluctuations could take in the procedure we outlined
in the last chapter.

So, the present chapter follows the tracks of those hydrodynamic


fluctuations, along the ε -expansion method, all the way up to the order ε 3 2 . As
it should be expected, we find the usual amplitude equation with an added
term representing the fluctuations. It will be analysing this term which we will
remark in some aspects of the polynomials (basis functions), as conditions they
have to satisfy. Finally, this leads us to the expressions for the δ -correlated
relations, which in turn contain the information about the amplitude of such
fluctuations. At that moment we can think of establishing the bridge between
the work mentioned above, and ours.

Then, let us consider (IV.65) – (IV.68) with V ( x ) rather than (1 − x ) .


Taking into account the usual scaling for units, and adding up the noise 8 , this
system becomes

∂ t u + T 1 2V ( x ) ∂ y u − ( ∂ x 2 + ∂ z 2 ) u − T 1 2V ( x ) v =

− {u∂ x u + v∂ y u + w∂ z u} + v 2 − ∂ x p
1
(V.1)
2
⎧ 1
⎛ 2d ⎞ 2 ⎫
⎪ ⎪
+ ⎨∂ x S1x + ⎜ ⎟ ∂ y S1 y + ∂ z S1 z ⎬
⎪⎩ ⎝ R1 ⎠ ⎪⎭

∂ t v + T 1 2V ( x ) ∂ y v − ( ∂ x 2 + ∂ z 2 ) v − T 1 2u = − {u∂ x v + v∂ y v + w∂ z v}
⎧ 1
⎛ 2d ⎞ 2 ⎫ (V.2)
⎪ ⎪
+ ⎨∂ x S 2 x + ⎜ ⎟ ∂ S
y 2y + ∂ z 2z ⎬
S
⎪⎩ ⎝ R1 ⎠ ⎪⎭

∂ t w + T 1 2V ( x ) ∂ y w − ( ∂ x 2 + ∂ z 2 ) w = − {u∂ x w + v∂ y w + w∂ z w} − ∂ z p
⎧ 1
⎛ 2d ⎞ 2 ⎫ (V.3)
⎪ ⎪
+ ⎨∂ x S3 x + ⎜ ⎟ ∂ S
y 3y + ∂ z 3z ⎬
S
⎪⎩ ⎝ R1 ⎠ ⎪⎭

∂ xu + ∂ y v + ∂ z w = 0 (V.4)

8At the end of this chapter we will make a comment on a recent paper (87) which makes an
analysis of the method we use here (83).
75

where the fluctuating terms satisfy

Sij ( x, y, z , t ) = 0 , i = 1, 2,3 ; j = x, y , z (V.5)

Sij ( x, y, z , t ) Slm ( x ', y ', z ', t ') =


(V.5’)
2Q (δ ilδ jm + δ imδ jl ) δ ( x − x ') δ ( y − y ') δ ( z − z ') δ ( t − t ' )

Notice that the units of Sij (that we represent like ⎡⎣ Sij ⎤⎦ ) are (and we must find
the appropriate scaling for the fluctuating terms as well),

⎡⎣ Sij ⎤⎦ = ρν 2 d 2 (V.6)

where ρ ≡ density; ν ≡ kinematic viscosity; d ≡ length scale. This implies that on


a purely dimensional basis we can write the units for

1 1
2R 2 ν ⎛ d ⎞
⎡ S S ' ⎤ = ( ρν d ) ⎛⎜ 3 ⎞⎟ ⎛⎜ 1 ⎞⎟ ⎛⎜ 2 ⎞⎟ (νρ ) ⎜
2
+ 21
⎣ ⎦ ⎟
⎝ d ⎠⎝ d ⎠ ⎝ d ⎠ ⎝ 2 R1 ⎠

so, in order to make this product dimensionless we must write out

Sij ( x, y, z , t ) Slm ( x ', y ', z ', t ') =


(V.7)
2Q (δ ilδ jm + δ imδ jl ) δ ( x − x ') δ ( y − y ') δ ( z − z ') δ ( t − t ' )

such that

1 1
k T ⎛ 2R ⎞ 2
⎛ 2R ⎞ 2
Q ≡ B2 ⎜ 1 ⎟ = ⎜ 1 ⎟ QRB (V.8)
ρν d ⎝ d ⎠ ⎝ d ⎠

where d denotes the width of the gap, k B is Boltzmann’s constant, and T is the
temperature (this is the only place T will represent the temperature, otherwise
it means Taylor’s number). QRB is the equivalent of our Q for the Rayleigh –
Bénard. So, as we can see the only scaling difference comes out as a
consequence of the cylindrical geometry in Taylor’s problem.

When we perform the small gap approximation into these terms, the
noise in (V.1) – (V.3) becomes Cartesian (this is to say, with no scaling factor in
front of the derivatives). Since we want to look at the fluctuations around an
axial wave (Taylor vortex flow), we propose for the random stress tensor,

Sij ( x, y, z , t ) = eik z Sij ( x, y, z, t ) + c.c. (V.9)


76

therefore

1
⎛ 2d ⎞ 2
∂ x S1x + ⎜ (
⎟ ∂ y S1 y + ∂ z S1 z = ∂ x S1x + ikS1 z e )
ik z

⎝ R1 ⎠
1
⎛ 2d ⎞ 2
∂ x S2 x + ⎜ (
⎟ ∂ y S 2 y + ∂ z S 2 z = ∂ x S 2 x + ikS 2 z e )
ik z
(V.10)
⎝ 1⎠
R
1
⎛ 2d ⎞ 2
∂ x S3 x + ⎜ (
⎟ ∂ y S3 y + ∂ z S3 z = ∂ x S3 x + ikS3 z e
ik z
)
⎝ R1 ⎠

and the stochastic average can be written as

Sij ( x, Y , z ,τ ) Slm * ( x ', Y ', z ',τ ' ) =


(V.11)
ε 2 Q (δ ilδ jm + δ imδ jl ) δ ( x − x ') δ (Y − Y ') δ ( z − z ') δ (τ − τ ')
3

Sij ( x, Y , z ,τ ) Slm ( x, Y , z ,τ ) = 0 (V.11’)

In order to know the order in ε of Sij it is enough to know that in


general for any fluid

Q dε 5 2 (V.12)

hence, from (V.11)

Sij ∼ ε 2 (V.13)

Moreover, since ε must be of the order of Δk (i.e. the deviation with respect to
kc ), the series expansion for the velocity can be written as

u = ε 1 2 ( ε 1 2 u0 + ε u1 + ) (V.14)

or then we can keep our original scaling for the velocity, but then

Sij ∼ ε 3 2 (V.15)

Therefore, the random stress tensor will not appear before the order ε 3 2 .

Following the usual steps we get

∂τ ' A = C0 A + C1∂ z 2 A + iC2 ∂Y ∂ z A + C3∂Y 2 A − C4 A A + Γ (Y , z ,τ ')


2
(V.16)
77

which is the generalization to the fluctuating case of (IV.80). The coefficients are
already given by (IV.81) – (IV.85), and

{ ( ) (
Γ (Y , z ,τ ') ≡ ϒ ∫ k 2T01 2 ∂ x S1x + ikS1 z + ( ∂ x 2 − k 2 ) ∂ x S 2 x + ikS 2 z +
1

0
2
)
(V.17)
(
+ikT ∂ x ∂ x S3 x + ikS3 z
0
12
)} P101 ( x ) dx

This expression is very general; it will let us see some conditions that must be
satisfied by the basis { Pjln } so that we can go on in the calculation, and it is the
step we have to fulfil in order to find the expression for the “critical length” of
the system, given by Walgraef et al. (59).

Then (V.17) becomes

( )
Γ = ϒ ∫ ⎡ −k 2T01 2 S1x − k 4 S 2 x + k 2T01 2 S3 z P101 ' ( x ) − 2k 2 S 2 x P101 ''' ( x ) − S2 x P101V ( x )
1

0⎣

{( ) ( )
+ i k 3T01 2 S1 z + k 5 S2 z P101 ( x ) + −2k 3 S 2 z + kT01 2 S3 x P101 '' ( x ) (V.18)

}
+ kS2 z P101IV ( x ) ⎤ dx

To find this expression, we have assumed that

Sij ( 0, Y , z ,τ ) = Sij (1, Y , z ,τ ) = 0 , ∀i, j


P101 ( 0 ) = P101 (1) = 0
P101 ' ( 0 ) = P101 ' (1) = 0 (V.19)
P101 '' ( 0 ) = P101 '' (1) = 0
P101 ''' ( 0 ) = P101 ''' (1) = 0

These conditions could be deduced from the usual boundary conditions, except
for the last two. In fact, this should not be too surprising since the
hydrodynamic system of equations is even higher order in the derivatives.

Because of (V.11’) we know that

Γ ( x, Y , z , τ ) = Γ * ( x, Y , z , τ ) = 0
(V.20)
Γ ( x, Y , z ,τ ) = Γ ( x ', Y ', z ',τ ) = 0

and we look for

Γ * ( x, Y , z ,τ ) Γ ( x ', Y ', z ',τ ')

as correlation condition; utilizing (V.11), (V.19) we find


78

Γ * ( x, Y , z ,τ ) Γ ( x ', Y ', z ',τ ') = ε 3 2 ϒδ (Y − Y ' ) δ ( z − z ') δ (τ − τ ') Q ⋅

⋅∫ ⎡ k 6 (T0 + k 4 ) P1012 + 2k 4 (T0 − k 4 ) P101 P101 ''+ k 2 (T0 + 4k 4 ) ( P101 '' ) +


1 2
0 ⎣
+ 2k 6 P101 P101IV − 4k 4 P101 '' P101IV + k 2 ( P101IV ) + k 4 ( k 4 + 4T0 ) ( P101 ') +
2 2
(V.21)

+ 4k 6 P101 ' P101 '''+ 2k 4 P101 ' P101V + 4k 4 ( P101 ''' ) + 4k 2 P101 ''' P101V +
2

+ ( P101V ) ⎤⎥ dx
2

where Q is given by (V.8).

It is clear by now that (V.21) is an important relation, since it is necessary


for the calculation of a Fokker – Planck equation. Let us write the equation (V.21)
for the case of P101 ( x ) = sin π x (we know it only satisfies the first of the
conditions (V.19)); so,

ε3 2
Γ*Γ ' = Qϒδ (Y − Y ' ) δ ( z − z ') δ (τ − τ ' ) ⋅
2 (V.22)
⋅ ( 3π 10 + 23π 8 k 2 + 46π 6 k 4 + 54π 4 k 6 + 5π 2 k 8 + 3k 10 )

This completes the frame of presentation of results coming out from the
amplitude equation (V.16). At this point we join the usual presentation of most
of the classical texts, in order to get the Fokker – Planck expression associated to
our results. Those texts usually start from a Master equation; in other words,
from a microscopic equation (almost always describing the Brownian motion)
associated to a birth-and-death description of fluctuations (see an excellent brief
presentation in (10)), they obtain the general structure of the equation
describing the time evolution of some quantity (which is usually identified
macroscopically with the order parameter) like the equilibrium distribution
function associated with the dynamical variable we have called amplitude so
far.

From the general theory (10) we can see that the coefficients appearing in
the Fokker – Planck equation are related with the drift and diffusion coefficients
of the system; then, remark that expressions (IV.60) – (IV.61) do not include C2
(i.e. the coefficient of the coupling term in the amplitude equation) into the
expressions for the diffusion coefficients, but rather it is the “re – scaling” time
factor included in τ . Therefore, we would expect that C2 will not appear in the
Fokker – Planck equation. This is a hand waving manner of saying that the
coupling term in the amplitude equation shall be “divergence – free”.

The model equations are treated in (88), however in that reference only a
special case of the amplitude equation is shown (namely the linear case with no
derivatives with respect to the position).
79

Just for the sake of completeness, let us show here a very brief
presentation of the general ideas behind all this. 9 If the dynamical variables are
denoted by ϕi , the Langevin equation can be written as

⎛L ⎞
∂ tϕ i = − ⎜ i ⎟ ϕi + f i (V.23)
⎝ χi ⎠

where Li are the transport coefficients, χ i the static susceptibility for ϕi , and fi
is the delta – correlated random force, such that

fi ( t ) f j = 2 Liδ ijδ ( t ) (V.24)

The transport coefficients mean in our case the viscosity (probably combined
with the wave vector k ). Expression (V.23) is compatible with a Gaussian
distribution function in ϕi (i.e. at equilibrium), which can be written as

P0 = exp ( − F ) (V.25)

so, (V.23) can also be written as

∂ tϕi = − Li ∂ϕi F + f i (V.26)

It is actually P0 the static solution to the Fokker – Planck equation. If we


consider a more general case than the one proposed by (V.23), this is to say
taking into account that mode – coupling terms might exist in the equation of
motion; we denote it by a streaming velocity V (ϕ ) , then,

∂ tϕi = Vi (ϕ ) − Li ∂ϕi F + fi (V.23’)

where

Vi (ϕ ) = λ0 ∑ P0 −1∂ϕ j { P0Qij } (V.27)


j

and Qij are commutators of ϕi ; λ0 is the mode – coupling strength parameter.

The expression (V.27) is valid only if the streaming velocity satisfies the
divergence – free property (in the ϕ -space) through the so called probability
current VP0 , this is to say

9The nomenclature we use is that of (88) and so it does not correspond with the one we have
used so far.
80

∑ ∂ϕ {V (ϕ ) P } = 0
i
i i 0 (V.28)

Since a divergence – free current does not alter the distribution function,
considering a streaming velocity V in the equation of motion does not change
the equilibrium properties of (V.25) and therefore the whole idea stays (the
wave vector susceptibility does not depend on the mode – coupling strength
parameter).

This time the distribution function P ( t ) is

P (t ) = ∏δ ( a j − ϕ j (t )) (V.29)
j

where the a j ’s are the coefficients appearing in the amplitude equation (i.e. our
C1 , C2 , etc.). 10 P ( t ) is the equilibrium distribution function (related to ϕ j ) and
P ( t ) obeys the equation

∂ t P ( t ) = − ∑ ∂ a j P ( t ) ∂ tϕ j ( t )
j
(V.30)
{
= −∑ ∂ a j P ( t ) V j ( a ) − L j ∂ a j F ( a ) + f j
j
}
This result together with

fi ( t ) P ( t ) = − Li ∂ ai P ( t ) (V.31)

leads to the Fokker – Planck equation

∂t P (t ) = L P (t ) (V.32)

where

{ (
L ≡ −∑ ∂ a j V j − L j ∂ a j + ∂ a j F
j
)} (V.32’)

Let us now translate all this to our results (V.16) and (V.21). After doing
all the steps we have outlined here we get the Fokker – Planck equation:

⎣⎢ ⎣ { ( 2
⎦ }
∂ t A = ∫∫ ⎡δ A − ⎡C0 + C4 A ⎤ A − ⎡⎣C1∂ z 2 + C3∂Y 2 ⎤⎦ A +ϒQK ( k , T0 ) δ A* ) A ⎤⎦ + c.c. dθ d z

(V.33)

10 Or simply the amplitude A we have been dealing with.


81

where A is the usual amplitude, and A is an order parameter introduced to be


able to write such an expression. So, (V.33) can be written as

⎣ (
∂ t A = ( ϒQK ( k , T0 ) ) ∫∫ ⎡δ A δ A* F ({ A} ) + δ A* A ⎤
⎦ ) (V.34)

such that

⎡ 2⎤
F ≡ − ( ϒQK ( k , T0 ) )
−1 1 4
∫∫ ⎢⎣C A + A + C1 ( ∂ z A ) + C3 ( ∂Y A ) ⎥ dθ d z
2 2
0 (V.35)
2 ⎦

On the other hand, the velocities correlations may be expressed like

AA ' ≈ exp ( − z − z ' Lc ) (V.36)

where the length Lc is (see (59))

2C1ε
Lc ≡ (V.37)
ϒQK ( k ) C4

and the function K of the wave number is obtained from (V.21) (in fact the
integral term).

In terms of fast variables (V.16) becomes

ε −1∂ t A = C0 A + ε −1 ( C1∂ z 2 + iC2∂Y ∂ z + C3∂Y 2 ) A − C4 A A + ε −3 2Γ


2
(V.38)

but from (V.14) we make A → Aε −1 2 , so finally we get

∂ t A = ε C0 A + C1∂ z 2 A + iC2∂Y ∂ z A + C3∂ Y 2 A − C4 A A + Γ


2
(V.39)

At this point let us make a comment on the paper by Gardiner and Steyn
– Ross (87). Essentially, what they do is the same calculations we have already
illustrated in the former pages, but utilizing rather than the method of multiple
space scales, the stochastic adiabatic elimination. As they pointed out, their
results are exactly the same as ours (83) then showing the equivalence between
both methods.

(87) criticizes though the Langevin approach, and following the same
kind of discussion we did about Hall’s calculations, we notice that some
remarks appearing in that article are quite pertinent (notably in which concerns
the carefulness in the scaling procedure and the renormalization techniques,
mainly the overall physical interpretation).
82

In any case we have insisted enough on the extreme sensitivity of the


Couette – Taylor system to the small gap approximation (as it should be for any
nonlinear system, but in our case it makes worse because of the anisotropic
nature of it), and saw how we could not let d go to zero arbitrarily.

Therefore it seems to us that Gardiner et al. analyses a situation which is


far from being the more general one, and hence being quite difficult to say
something in general, say extended to non – isotropic systems.
83

CONCLUSIONS

At the end of each chapter we tried to give some sort of conclusive view;
so, in this last section we intend to come back to the spirit of chapters I and II,
but with the support of the experience of all the calculations and results of the
intermediate sections.

We have seen how problems arise, for instance when trying to find an
explicit expression for the critical Taylor number for the wavy vortex flow,
starting from a pseudo – solution for the first instability. This might be nothing
more but the consequence of an extreme sensibility of the Navier – Stokes
equations face to any approximation (small gap for instance). While saying that,
we have in mind the unexpected acuteness of the boundary conditions. Remark
the fact that such situation forced us to have a close look on the conditions that
the basis functions had to satisfy on the walls of the cylinders (see eqn. (V.19));
and this in relation with the effects of the fluctuations. We shall observe that
most of the quantities entering into the calculations (which characterizes the
fluid at the same time) are held constant; this attitude face to the problem is the
same adopted in all the nonlinear problems, where usually one finds too many
parameters which could be playing the role of variables.

As we said immediately after equation (III.55), it is even a remarkable


fact that starting from very rough approximations (specially after neglecting so
many terms for the sake of transparency) we could conclude that Tm is a
function of the gap d only (this is to say the geometry), for infinitely long
cylinders. This conclusion was corroborated afterwards with expression (III.79)
(being m -independent though) obtained following a different way. Finally,
comparisons like those of the table 1 should encourage us.

We have pointed out that the discussion we began with the amplitude
equation (IV.48), (IV.49) continued to section ii) with the expression (IV.80) and
its coefficients, and showing that our calculation establishes the connection
between the scaling procedure of Tabeling, and Brand and Cross. This is
important under the light of the inclusion of fluctuations as we did in chapter V,
because it was such a scaling procedure the one we utilized.

The expression (V.21), and in particular (V.22) show that due to the
geometrical characteristics of the system, the amplitude of fluctuations
appearing in the correlation condition is not simply Q , but rather Q multiplied
by a positive quantity (whose value actually depends on the assumed boundary
conditions, expressed through the behaviour of the basis functions); so that in a
figurative sense, the Q is substituted by a new Q ' , being this last one an
“amplified” version of the old one. The mechanisms for this behaviour are
84

obscure to us, and we should not try to get out from this formalism a purely
physical conclusion; nevertheless, again such a result points out on the fact that
we must be careful about the approximations performed along the calculations.
This observation seems to be supported by a quite general model on the
behaviour of fluctuations in non – isotropic systems being worked out, in an
early stage for the moment (89).

Finally, we can mention here the amazement we feel while looking at


this kind of systems. As macroscopic as they are, and as relatively accessible as
they are for experimental measurements, there are still many open doors in the
research, if we want to succeed some day in the understanding of fluids
behaviour.
85

APPENDIX A

Let us exhibit here the expressions for u1 , v1 and w1 , this is to say the
components of the velocity at the order ε :

( 2)
v1 = K100 sin 2π x
(1) ik z (1) ( i k z + my )
+ A11(1−) 1e (
i k z − my )
+ ⎡⎣ A110 e + A111e
() () − i ( k z + my )
+ A11( −) 1 * e
− i ( k z − my )
+ A110
1
* e − ik z + A111 *e
1 1
⎤ sin π x

+ ∑ ⎡⎣ A120
( n ) 2 ik z
e + A121( n ) i ( 2 k z + my )
e
n =0 (A.1)
(n) 2 i ( k z + my ) ( n) i ( 2 k z − my )
+ A122 e + A12−1e
( n ) 2i ( k z − my ) ( n)
+A 12 − 2 e + A120 * e −2ik z
( ) − i ( 2 k z + my ) ( ) −2 i ( k z + my )
+ A121 + A122
n n
*e *e
+ A12( −)1 * e
− i ( 2 k z − my )
+ A12( −) 2 * e
−2 i ( k z − my )
n n
⎤ sin nπ x

where

(1)
A110 = ik
(T − 3 (π
0
2
+ k2 )
2
) ⋅∂ A (1)
(A.2)
(π )
z 010
2 3
2
+k

( )
=0 ∀n ≠ 1 (A.2’)
n
A110 ,

( 2)
K100 =
1

( π 2 + k 2 ) A010
(1) 2
(A.3)

(n)
K100 =0 , ∀n ≠ 2 (A.3’)

(1)
A111 =
{2 ⎡( π
⎢⎣
2
+ k 2 )(π 2 + k 2 + im ) ( 2ik ∂ z − ∂Y ) + ik (π 2 + k 2 + im ) ∂ z − ikT0∂ z ⎤⎥ A011
2


(1)

(A.4)
}
⎡(π + k + im ) (π + k ) − (π + k ) ⎤
2 2 2 2 2 2 2 3

⎢⎣ ⎥⎦

A11( −) 1
1
=
{2 ⎡(π
⎢⎣
2
+ k 2 )(π 2 + k 2 − im ) ( 2ik ∂ z − ∂Y ) + ik (π 2 + k 2 − im ) ∂ z − ikT0∂ z ⎤⎥ A01
2 (1)
⎦ −1 }
⎡(π 2 + k 2 − im )2 (π 2 + k 2 ) − (π 2 + k 2 )3 ⎤
⎢⎣ ⎥⎦
(A.5)
86

( )
n
A120 =
⎡T 4k − ( n π
2
4k 2
2 2
+ 4k )
2 3⎤
{ ∫ sin π x sin nπ xdx} ( −T A
0
1
2
0
(1) 2
010 + 2 (1 − 4m 2 ) A011
() () 1
A01−1
1
)
⎢⎣ 0 ⎥⎦
(A.6)

(n)
A121 =
{ 4 ∫ ⎡⎣ 2k 2 ( im − T0 ) sin 2 π x + 2π m 2 sin 2π x +2π 2 m 2 cos 2π x ⎤⎦ sin nπ xdx A010
0
1
(1) (1)
A011 }
T0 4k 2 − ( n 2π 2 + 4k 2 + im ) ( n 2π 2 + 4k 2 )
2

(A.7)

( ) ( )2
(A.8)
n
∼ A011
1
A122

A12( −)1 ∼ 2 A010


() ()
(A.9)
n 1 1
A01−1

A12( n−) 2 ∼ A01(1)−12 (A.10)

and the other two components of the velocity,

(1) ik z (1) i ( k z + my )
u1 = ⎡⎣U110 e + U111e + U11(1−) 1ei( k z − my ) + U110
(1)
* e − ik z
(1)
+U111 * e− i( k z + my ) + U11(1−) 1 * e − i( k z − my ) ⎤⎦ sin π x

+ ∑ ⎡⎣U120
( n ) 2 ik z
e + U121( n ) i ( 2 k z + my )
e ( n ) 2i ( k z + my )
+ U122 e + U12( n−)1ei( 2 k z − my ) + U12( n−) 2 e 2i( k z − my ) + U120
(n)
* e −2ik z
n =0
(n)
+ U121 * e − i( 2 k z + my ) +U122
(n)
* e −2i( k z + my ) + U12( n−)1 * e −i( 2 k z − my ) + U12( n−) 2 * e−2i( k z − my ) ⎤⎦ sin nπ x

+ ⎡⎣U101 ' eimy + U102 ' e 2imy + U101 '* e − imy + U102 '* e −2imy ⎤⎦ ( sin 2π x + sin 2 π x )
(A.11)

where all the coefficients can be written in terms of the A ’s.

()
w1 = W110( 1 ()
sin π x + W110 ) (()
'cos π x eik z + W110
1 ()
*sin π x + W110 '*cos π x e −ik z
1
) 1

+ (W ( ) sin π x + W ( ) 'cos π x ) e ( )
+ (W ( ) *sin π x + W ( ) '*cos π x ) e (
1 1 i k z + my 1 1 − i k z + my )
111 111 111 111

+ (W ( ) sin π x + W ( ) 'cos π x ) e ( )
+ (W ( ) *sin π x + W ( ) '*cos π x ) e
1 1 i k z − my 1 1 − i ( k z − my )
11−1 11−1 11−1 11−1

{ (A.12)
2
+ ∑ ∑ ⎡⎣W1(2 j) sin nπ x + W12( j) 'cos nπ x ⎤⎦ e (
n n i 2 k z + jmy )

n = 0 j =−2

+ ⎡⎣W12( j) *sin nπ x + W12( j) '*cos nπ x ⎤⎦ e


n n − i ( 2 k z + jmy )
}
2
+ ∑ ⎡⎣U10 j ' eijmy + U10 j '* e − ijmy ⎤⎦ π ( 2 cos 2π x + sin 2π x ) z
j =1

Again, the coefficients can be written out in terms of the A ’s.


87

REFERENCES

(1) G. Nicolis, I. Prigogine; Self – organization in non – equilibrium systems,


(John Wiley & Sons, New York, 1977).

(2) L. Landau, E. Lifchitz; Mécanique des fluides, (Ed. MIR, Moscou, 1971).

(3) E.L. Koschmieder; Experimental aspects of hydrodynamic instabilities;


Proceedings of the Solvay Conference, 1978.

(4) R. Graham; Onset of cooperative behaviour in non – equilibrium steady


states; Ibid.

(5) L.E. Reichl; A modern course in Statistical Physics, (University of Texas


Press, Austin, 1980).

(6) M. Krasnov, A. Kissélev, G. Makarenko ; Recueil de problèmes sur les


équations différentielles ordinaires, (Ed. MIR, Moscou, 1981).

(7) V. Arnold ; Chapitres supplémentaires de la théorie des équations


différentielles ordinaires, (Ed. MIR, Moscou, 1980).

(8) S. Ma; Modern theory of critical phenomena, (The Benjamin Cummings


Publishing Company Inc., Massachusetts, 1976).

(9) R.D. Richtmyer; Principles of advanced Mathematical Physics; (vol. II,


Springer – Verlag , New York, 1981).

(10) P. Glansdorff, I. Prigogine; Thermodynamic theory of structure, stability,


and fluctuations, (Wiley – Interscience, London, 1971).

(11) S. Kogelman, R.C. DiPrima; Phys. of Fluids, 13, 1 January, 1970.

(12) P. Chossat, G. Iooss; Primary and secondary bifurcations in the Couette –


Taylor problem, (Pré – Publications mathématiques № 35, Université de
Nice, 1984).

(13) D. Walgraef, G. Dewel, P. Borckmans; non – equilibrium phase


transitions and chemical instabilities, (Advances in Chem. Phys., vol. 49,
Ed. Prigogine and Rice; Wiley, 1982).

(14) S. Chandrasekhar; Hydrodynamic and Hydromagnetic stability, (Dover


Publications Inc., New York, 1981).
88

(15) M. Lesieur; La turbulencia desarrollada; Mundo Científico, № 22, vol.3,


1983.

(16) C. Vidal, J.C. Roux ; Comment naît la turbulence ; Science, 1981.

(17) A.L. Robinson ; Science, vol. 221, 1983.

(18) M.G. Velarde, C. Normand; Sci. Am. Vol. 243, № 1, July 1980.

(19) G. Ahlers; Onset of convections and turbulence in a cylindrical container.

(20) P. Bergé; Rayleigh – Bénard convection in high Prandtl number fluid.

(21) F.H. Busse, J.A. Whitehead; J. Fluid Mech. (1971), vol.47, part 2, pp. 305 –
320.

(22) V. Croquette, M. Mory, F. Schosseler; Rayleigh – Bénard convective


structures in a cylindrical container; from Service de Chimie Physique du
solide et de résonance magnétique, juillet 1982.

(23) F.H. Busse ; Rep. Prog. Phys., vol. 41, 1978.

(24) R.M. Clever, F.H. Busse; J. Fluid Mech. (1974) vol. 65 part 4, pp. 625 – 645.

(25) E.D. Siggia, A. Zippelius; Phys. Rev. Lett. 47, 12, September 1981.

(26) E.D. Siggia, A. Zippelius; Phys. Rev. A, 24, 2, August 1981.

(27) A. Zippelius, E.D. Siggia; Phys. Fluids 26, 10, October 1983.

(28) M.C. Cross; Phys. Fluids 25, 6, June 1982.

(29) M.C. Cross; Phys. Rev. A, 25, 2, February 1982.

(30) M.C. Cross; Phys. Fluids 23, 9, September 1980.

(31) M.C. Cross; Phys. Rev. A, 27, 1, January 1983.

(32) M.C. Cross, D.G. Daniels, P.C. Hohenberg, E.D. Siggia; J. Fluid Mech. 127,
155, (1983).

(33) F.H. Busse; Transition to turbulence in Rayleigh – Bénard convection;


from Topics in Applied Physics; editors H. Swinney, J.P. Gollub; vol.45,
Springer – Verlag, 1981.

(34) L.A. Segel; J. Fluid Mech. (1969), vol. 38, part 1, pp. 203 – 224.
89

(35) A.C. Newell, J.A. Whitehead; J. Fluid Mech. 36, 239 (1969).

(36) Y. Pomeau, P. Manneville ; Journal de Physique – Lettres, tome 40, № 23,


décembre 1979.

(37) J.E. Wesfreid, V. Croquette; Phys. Rev. Lett. 45, 8, August 1980.

(38) G.P. King, H.L. Swinney; Phys. Rev. A 27 (1983), 1240.

(39) D. Walgraef; Drift flow and phase instabilities in a model for Taylor –
Couette systems, (Preprint).

(40) P. Hall; The evolution equations for Taylor vortices in the small gap limit;
ICASE (NASA), October 1983.

(41) H.S. Greenside, M.C. Cross; Stability analysis of two – dimensional


models of three – dimensional convection, (Preprint).

(42) H.S. Greenside, W.M. Coughram; Non – linear pattern formation near the
onset of Rayleigh – Bénard convection, (Preprint).

(43) M.C. Cross, A.C. Newell; Convection pattern in large aspect ratio systems,
(Preprint).

(44) H.S. Greenside, W.M. Coughram, N.L. Schryer; Phys. Rev. Lett. 49, 10,
September 1982.

(45) A. Schlüter, D. Lortz, F. Busse; J. Fluid Mech. (1965), vol. 23, part 1, pp.
129 – 144.

(46) P. Bergé, M. Dubois; Different routes to turbulence in high Prandtl


number convection, and in rectangular cells; influence of the aspect ratio
and of the structure.

(47) M. Dubois, P. Bergé ; J. Physique 42 (1981) pp. 167 – 174.

(48) A. Libchaber, J. Maurer ; J. Physique 41 (1980).

(49) J. Maurer, A. Libchaber ; J. Physique Lett., 40, août 1979.

(50) G.I. Taylor; Stability of a viscous liquid contained between two rotating
cylinders; Philos. Trans. R. Soc. London A 223, pp. 289 – 343 (1923).

(51) D. Coles; Transition in circular Couette flow; J. Fluid Mech. 21, pp. 385 –
425 (1965).
90

(52) A. Davey; J. Fluid Mech. 14, pp. 336 – 368 (1962).

(53) R.C. DiPrima, P.M. Eagles; The Physics of Fluids, vol. 20, № 2, February
1977.

(54) R.C. DiPrima, H.L. Swinney; Instabilities and transition in flow between
concentric rotating cylinders; from Topics in Applied Physics; editors H.L.
Swinney, J.P. Gollub; vol. 45, Springer – Verlag, 1981.

(55) National Science Foundation Workshop on the Sixtieth Anniversary


Taylor Vortex Flow; Eugene (Oregon),12 to 15 June 1983.

(56) T.B. Benjamin, T. Mullin; J. Fluid Mech. (1982), vol. 121, pp. 219 – 230.

(57) E.R. Krueger, A. Gross, R.C. DiPrima; J. Fluid Mech. (1966) vol. 24, part 3,
pp. 521 – 538.

(58) A. Davey, R.C. DiPrima, J.T. Stuart; J. Fluid Mech. (1968), vol.31, part 1,
pp. 17 – 52.

(59) D. Walgraef, P. Borckmans, G. Dewel; Phys. Rev. A 25, 5, May 1982.

(60) P. Tabeling; J. Physique – Lett. 44 (1983) L665 – L672.

(61) H. Bra, M.C. Cross, P.C. Hohenberg, S. Safran; J. Fluid Mech., vol. 110, pp.
297 – 334 (1981).

(62) R.C. DiPrima, J. Sijbrand; Interactions of Axisymmetric and non –


axisymmetric disturbances in the flow between concentric rotating
cylinders: bifurcations near multiple eigenvalues.

(63) G. Ahlers, M.C. Cross, P.C. Hohenberg, S. Safran; J. Fluid Mech. , vol. 110,
pp. 297 – 334 (1981).

(64) P.S. Marcus, (Preprint).

(65) G. Cognet, A. Bouabdallah, A. Ait Aider; Laminar turbulent transition in


Taylor – Couette flow, influence of geometrical parameters, (Preprint).

(66) K. Park, R.J. Donnelly; Phys. Rev. A 24, 4, October 1981.

(67) K. Park, G.L. Crawford, R.J. Donnelly; Phys. Rev. Lett. 47, 20, November
1981.

(68) F.R. Mobbs, S. Preston, M.S. Ozogan; An experimental investigation of


Taylor vortex waves. (Taylor vortex flow working party; Leeds (1979)).
91

(69) P.R. Fenstermacher, H.L. Swinney, J.P. Gollub; J. Fluid Mech. (1979), vol.
94, part 1, pp. 103 – 128.

(70) R. Graham, J.A. Domaradski; Phys. Rev. A 26, 3, September 1982.

(71) M. Abramowitz, I.A. Stegun; Handbook of mathematical functions,


(Dover Publishing Inc., New York, 1972).

(72) E.T. Whittaker, G.N. Watson; A modern course of Modern Analysis,


(Cambridge University Press, London, 1902).

(73) G. Ahlers, D.S. Cannell, M.A. Dominguez – Lerma; Phys. Rev. Lett. 49, 6,
August 1982.

(74) M.A. Dominguez – Lerma, G. Ahlers, D.S. Cannell; Phys. Fluids 27, 4,
April 1984.

(75) J. Swift, P.C. Hohenberg; Phys. Rev. A 15, 1, January 1977.

(76) P. Manneville ; Comptes rendus – Workshop “Common trends in particle


and condensed matter Physics” 02/27/83 – 03/11/83; Les Houches.

(77) I.S. Gradshteyn, I.M. Ryzhik; Table of integrals, series and products,
(Academic Press, New York, 1980).

(78) H. Yahata; Prog, Th. Phys. 57, 5, May 1977.

(79) K. Stewartson, J.T. Stuart; J. Fluid Mech. (1971), vol. 48, part 3, pp. 529 –
545.

(80) A. Davey, L.M. Hocking, K. Stewartson; J. Fluid Mech. (1974), vol. 63,
part 3, pp. 529 – 536.

(81) V.M. Zaitsev, M.I. Shliomis; Soviet Phys. JETP 32, 5, May 1971.

(82) L. Landau, E. Lifchitz; Physique Statistique, (Ed. MIR, Moscou, 1967).

(83) R. Graham; Phys. Rev. 10, 5, November 1974.

(84) J. Wesfreid, Y. Pomeau, M. Dubois, C. Normand, P. Bergé ; J. Physique –


Lettres, 39, 7, 7 juillet 1978.

(85) G. Ahlers, R.W. Walden ; Phys. Rev. Lett. 44, 7, February 1980.

(86) D. Walgraef, P. Borckmans, G. Dewel ; Onset of wavy Taylor vortex flow


in finite geometries, (Preprint).
92

(87) C.W. Gardiner, M.L. Steyn – Ross; Phys. Rev. A, 29, 5, May 1984.

(88) S.W. Lovesey; Condensed matter Physics, dynamic correlations, (The


Benjamin/Cummings Publishing Co. 1980).

(89) A. Garcia; private communication.


93

TABLE OF CONTENTS11

Page

CHAPTER I. INTRODUCTION 4

CHAPTER II. OVERVIEW ON THE PROBLEM 16

i) The hydrodynamic problem 16


ii) The Couette-Taylor instability 19
iii) Wavy vortex flow 24
iv) Further bifurcations 25

CHAPTER III. STABILITY ANALYSIS 27

i) Axisymmetric perturbations 27
ii) Non-axisymmetric perturbations 34

CHAPTER IV. THE AMPLITUDE EQUATION 51

i) Form of the amplitude equation 52


ii) Phase diffusion 64
iii) The coefficients 68

CHAPTER V. FLUCTUATIONS 73

CONCLUSIONS. 83

APPENDIX A. 85

REFERENCES. 87

TABLE OF CONTENTS. 93

11N.B. Since the original text of this thesis was in a printed form, the actual transcription to
digital format does not forcedly coincide with the original one, concerning the page numbers.
Otherwise we tried to respect every single aspect of the original presentation, even though we
would have actually changed some writing turns in a present day realization.

You might also like