In long-time simulations of turbulence, symmetry considerations can be disposed
off in a few lines: In absence of boundary conditions, the Navier-Stokes equations are invariant under any 3D translation, 3D rotation, and inversion through the origin. For compact invariant solutions and their instabilities the symmetries play a paramount role, both for
their classification and for understanding the dynamics connecting them. We show that close to onset of turbulence there are five inequivalent isotropy subgroups of interest.
Written in collaboration with P. Cvitanovic and J.F. Gibson, Appendix C of http://ChaosBook.org/projects/Halcrow/thesis.pdf .

© All Rights Reserved

284 views

In long-time simulations of turbulence, symmetry considerations can be disposed
off in a few lines: In absence of boundary conditions, the Navier-Stokes equations are invariant under any 3D translation, 3D rotation, and inversion through the origin. For compact invariant solutions and their instabilities the symmetries play a paramount role, both for
their classification and for understanding the dynamics connecting them. We show that close to onset of turbulence there are five inequivalent isotropy subgroups of interest.
Written in collaboration with P. Cvitanovic and J.F. Gibson, Appendix C of http://ChaosBook.org/projects/Halcrow/thesis.pdf .

© All Rights Reserved

- Which Functor is the Projective Line?
- Ex 6 Solutions
- Captivating Symmetry by RNDr Daniela Richtarikova Slovak Technical University (Slovakia)
- fields.utoronto.ca Blaise Pascal: David Hilbert's Problems.
- Maximality Methods in Local Probability
- math
- sdarfdticle
- G.S. Ezra- On the symmetry properties of non-rigid molecules: Semifactorizability of the isometric group
- Soft Neutrosophic Group
- maths
- CS and IT.pdf
- Sudoku.docx
- Galois theory notes
- Abstract Algebra - Beachy
- MTECHCSNEWSYLL2012
- Classic Lie
- mathgen-1274246079
- 2018SU B.sc Maths Syllabus
- Mathematics General Cbcs Draft Syllabus
- teach-2019

You are on page 1of 24

CONNECTIONS

1) a short one, from http://arxiv.org/abs/0808.3375 - first 9 pages

2) a longer one, from the (updated) J Halcrow thesis - remaining pages

Predrag Cvitanovic, Feb 23 2009

A Thesis

Presented to

The Academic Faculty

by

Jonathan J Halcrow

In Partial Fulﬁllment

of the Requirements for the Degree

Doctor of Philosophy in the

School of Physics

ver. 1.4, October 24 2008

the short version, extracted from http://arxiv.org/abs/0808.3375

plane Couette flow

By J. H A L C R O W, J. F. G I B S O N

A N D P. C V I T A N O V I Ć

School of Physics, Georgia Institute of Technology, Atlanta, GA 30332, USA

We survey equilibria and traveling waves of plane Couette flow in small periodic cells

at moderate Reynolds number Re, adding in the process eight new equilibrium and two

new traveling-wave solutions to the four solutions previously known. Bifurcations un-

der changes of Re and spanwise period are examined. These non-trivial flow-invariant

unstable solutions can only be computed numerically, but they are ‘exact’ in the sense

that they converge to solutions of the Navier-Stokes equations as the numerical resolution

increases. We find two complementary visualizations particularly insightful. Suitably cho-

sen sections of their 3D-physical space velocity fields are helpful in developing physical

intuition about coherent structures observed in moderate Re turbulence. Projections of

these solutions and their unstable manifolds from ∞-dimensional state space onto suit-

ably chosen 2- or 3-dimensional subspaces reveal their interrelations and the role they

play in organizing turbulence in wall-bounded shear flows.

1. Introduction

In Gibson et al. (2008) (henceforth referred to as GHC) we have utilized exact equilib-

rium solutions of the Navier-Stokes equations for plane Couette flow in order to illustrate

a visualization of moderate Re turbulent flows in an infinite-dimensional state space,

in terms of dynamically invariant, intrinsic, and representation independent coordinate

frames. The state space portraiture (figure 1) offers a visualization of numerical (or ex-

perimental) data of transitional turbulence in boundary shear flows, complementary to

3D visualizations of such flows (figure 2). Side-by-side animations of the two visualiza-

tions illustrate their complementary strengths (see Gibson (2008b) online simulations).

In these animations, 3D spatial visualization of instantaneous velocity fields helps elu-

cidate the physical processes underlying the formation of unstable coherent structures,

such as the Self-Sustained Process (SSP) theory of Waleffe (1990, 1995, 1997). Running

concurrently, the ∞-dimensional state space representation enables us to track unstable

manifolds of equilibria of the flow, the heteroclinic connections between them (Halcrow

et al. 2009), and gain new insights into the nonlinear state space geometry and dynamics

of moderate Re wall-bounded shear flows such as plane Couette flow.

Here we continue our investigation of exact equilibrium and traveling-wave solutions

of Navier-Stokes equations, this time tracking them adiabatically as functions of Re and

periodic cell size [Lx , 2, Lz ] and, in the process, uncovering new invariant solutions, and

determining new relationships between them.

The history of experimental and theoretical advances is reviewed in GHC, Sect. 2;

here we cite only the work on equilibria and traveling waves directly related to this

investigation. Nagata (1990) found the first pair of nontrivial equilibria, as well as the

first traveling wave in plane Couette flow (Nagata 1997). Waleffe (1998, 2003) computed

4 J. Halcrow, J. F. Gibson, and P. Cvitanović

approximately 100 wall units. These rolls, in turn, generate streamwise streaks of high

and low speed fluid, by convecting fluid alternately away from and towards the walls.

The streaks have streamwise instabilities whose length scale is roughly twice the roll

separation. These ‘coherent structures’ are prominent in numerical and experimental

observations (see figure 2 and Gibson (2008b) animations), and they motivate our inves-

tigation of how equilibrium and traveling-wave solutions of Navier-Stokes change with

Re and cell size.

Fluid states are characterized by their energy E = 12 kuk2 and energy dissipation rate

D = k∇ × uk2 , defined in terms of the inner product and norm

Z

1

(u, v) = dx u · v , kuk2 = (u, u) . (2.3)

V Ω

R

The rate of energy input is I = 1/(Lx Lz ) dxdz ∂u/∂y, where the integral is taken over

the upper and lower walls at y = ±1. Normalization of these quantities is set so that

I = D = 1 for laminar flow and Ė = I −D. In some cases it is convenient to consider fields

as differences from the laminar flow. We indicate such differences with hats: û = u − yx̂.

In an infinite domain and in the absence of boundary conditions, the Navier-Stokes

equations are equivariant under any 3D translation, 3D rotation, and x → −x, u → −u

inversion through the origin (Frisch 1996). In plane Couette flow, the counter-moving

walls restrict the rotation symmetry to rotation by π about the z-axis. We denote this

rotation by σx and inversion through the origin by σxz . The σxz and σx symmetries

generate a discrete dihedral group D1 × D1 = {e, σx , σz , σxz } of order 4, where

σx [u, v, w](x, y, z) = [−u, −v, w](−x, −y, z)

σz [u, v, w](x, y, z) = [u, v, −w](x, y, −z) (3.1)

σxz [u, v, w](x, y, z) = [−u, −v, −w](−x, −y, −z) .

The subscripts on the σ symmetries thus indicate which of x, z change sign. The walls also

restrict the translation symmetry to 2D in-plane translations. With periodic boundary

conditions, these translations become the SO(2)x × SO(2)z continuous two-parameter

group of streamwise-spanwise translations

τ (`x , `z )[u, v, w](x, y, z) = [u, v, w](x + `x , y, z + `z ) . (3.2)

The equations of plane Couette flow are thus equivariant under the group Γ = O(2)x ×

O(2)z = D1,x n SO(2)x × D1,z n SO(2)z , where n stands for a semi-direct product, x

subscripts indicate streamwise translations and sign changes in x, y, and z subscripts

indicate spanwise translations and sign changes in z.

The solutions of an equivariant system can satisfy all of the system’s symmetries, a

proper subgroup of them, or have no symmetry at all. For a given solution u, the subgroup

that contains all symmetries that fix u (that satisfy su = u) is called the isotropy (or

stabilizer) subgroup of u. (Hoyle 2006; Marsden & Ratiu 1999; Golubitsky & Stewart

2002; Gilmore & Letellier 2007). For example, a typical turbulent trajectory u(x, t) has

no symmetry beyond the identity, so its isotropy group is {e}. At the other extreme is

the laminar equilibrium, whose isotropy group is the full plane Couette symmetry group

Γ.

In between, the isotropy subgroup of the Nagata equilibria and most of the equilibria

Equilibria and traveling waves of plane Couette flow 5

reported here is S = {e, s1 , s2 , s3 }, where

s1 [u, v, w](x, y, z) = [u, v, −w](x + Lx /2, y, −z)

s2 [u, v, w](x, y, z) = [−u, −v, w](−x + Lx /2, −y, z + Lz /2) (3.3)

s3 [u, v, w](x, y, z) = [−u, −v, −w](−x, −y, −z + Lz /2) .

These particular combinations of flips and shifts match the symmetries of instabilities of

streamwise-constant streaky flow (Waleffe 1997, 2003) and are well suited to the wavy

streamwise streaks observable in figure 2, with suitable choice of Lx and Lz . But S is

one choice among a number of intermediate isotropy groups of Γ, and other subgroups

might also play an important role in the turbulent dynamics. In this section we provide a

partial classification of the isotropy groups of Γ, sufficient to classify all currently known

invariant solutions and to guide the search for new solutions with other symmetries. We

focus on isotropy groups involving at most half-cell shifts. The main result is that among

these, up to conjugacy in spatial translation, there are only five isotropy groups in which

we should expect to find equilibria.

A few observations will be useful in what follows. First, we note the key role played by

the inversion and rotation symmetries σx and σz (3.1) in the classification of solutions

and their isotropy groups. The equivariance of plane Couette flow under continuous

translations allows for traveling-wave solutions, i.e., solutions that are steady in a frame

moving with a constant velocity in [x, z]. In state space, traveling waves either trace

out circles or wind around tori, and these sets are both continuous-translation and time

invariant. The sign changes under σx , σz , and σxz , however, imply particular centers of

symmetry in x, z, and both x and z, respectively, and thus fix the translational phases

of fields that are fixed by these symmetries. Thus the presence of σx or σz in an isotropy

group prohibits traveling waves in x or z, and the presence of σxz prohibits any form

of traveling wave. Guided by this observation, we will seek equilibria only for isotropy

subgroups that contain the σxz inversion symmetry.

Second, the periodic boundary conditions impose discrete translation symmetries of

τ (Lx , 0) and τ (0, Lz ) on velocity fields. In addition to this full-period translation sym-

metry, a solution can also be fixed under a rational translation, such as τ (mLx /n, 0) or a

continuous translation τ (`x , 0). If a field is fixed under continuous translation, it is con-

stant along the given spatial variable. If it is fixed under rational translation τ (mLx /n, 0),

it is fixed under τ (mLx /n, 0) for m ∈ [1, n − 1] as well, provided that m and n are rela-

tively prime. For this reason the subgroups of the continuous translation SO(2)x consist

of the discrete cyclic groups Cn,x for n = 2, 3, 4, . . . together with the trivial subgroup

{e} and the full group SO(2)x itself, and similarly for z. For rational shifts `x /Lx = m/n

we simplify the notation a bit by rewriting (3.2) as

τxm/n = τ (mLx /n, 0) , τzm/n = τ (0, mLz /n) . (3.4)

Since m/n = 1/2 will loom large in what follows, we omit exponents of 1/2:

τx = τx1/2 , τz = τz1/2 , τxz = τx τz . (3.5)

If a field u is fixed under a rational shift τ (Lx /n), it is periodic on the smaller spatial

domain x ∈ [0, Lx /n]. For this reason we can exclude from our searches all equilibrium

whose isotropy subgroups contain rational translations in favor of equilibria computed on

smaller domains. However, as we need to study bifurcations into states with wavelengths

longer than the initial state, the linear stability computations need to be carried out

6 J. Halcrow, J. F. Gibson, and P. Cvitanović

in the full [Lx , 2, Lz ] cell. For example, if EQ is an equilibrium solution in the Ω1/3 =

[Lx /3, 2, Lz ] cell, we refer to the same solution repeated thrice in Ω = [Lx , 2, Lz ] as

“spanwise-tripled” or 3 × EQ. Such solution is by construction fixed under the C3,x =

1/3 2/3

{e, τx , τx } subgroup.

Third, some isotropy groups are conjugate to each other under symmetries of the full

group Γ. Subgroup S 0 is conjugate to S if there is an s ∈ Γ for which S 0 = s−1 Ss.

In spatial terms, two conjugate isotropy groups are equivalent to each other under a

coordinate transformation. A set of conjugate isotropy groups forms a conjugacy class.

It is necessary to consider only a single representative of each conjugacy class; solutions

belonging to conjugate isotropy groups can be generated by applying the symmetry

operation of the conjugacy.

In the present case conjugacies under spatial translation symmetries are particularly

important. Note that O(2) is not an abelian group, since reflections σ and translations

τ along the same axis do not commute (Harter 1993). Instead we have στ = τ −1 σ.

Rewriting this relation as στ 2 = τ −1 στ , we note that

σx τx (`x , 0) = τ −1 (`x /2, 0) σx τ (`x /2, 0) . (3.6)

The right-hand side of (3.6) is a similarity transformation that translates the origin of

coordinate system. For `x = Lx /2 we have

τx−1/4 σx τx1/4 = σx τx , (3.7)

and similarly for the spanwise shifts / reflections. Thus for each isotropy group containing

the shift-reflect σx τx symmetry, there is a simpler conjugate isotropy group in which σx τx

is replaced by σx (and similarly for σz τz and σz ). We choose as the representative of each

conjugacy class the simplest isotropy group, in which all such reductions have been made.

However, if an isotropy group contains both σx and σx τx , it cannot be simplified this

way, since the conjugacy simply interchanges the elements.

Fourth, for `x = Lx , we have τx−1 σx τx = σx , so that, in the special case of half-cell

shifts, σx and τx commute. For the same reason, σz and τz commute, so the order-16

isotropy subgroup

G = D1,x × C2,x × D1,z × C2,z ⊂ Γ (3.8)

is abelian.

We now undertake a partial classification of the lattice of isotropy subgroups of plane

Couette flow. We focus on isotropy groups involving at most half-cell shifts, with SO(2)x ×

SO(2)z translations restricted to order 4 subgroup of spanwise-streamwise translations

(3.5) of half the cell length,

T = C2,x × C2,z = {e, τx , τz , τxz } . (3.9)

All such isotropy subgroups of Γ are contained in the subgroup G (3.8). Within G, we

look for the simplest representative of each conjugacy class, as described above.

Let us first enumerate all subgroups H ⊂ G. The subgroups can be of order |H| =

{1, 2, 4, 8, 16}. A subgroup is generated by multiplication of a set of generator elements,

with the choice of generator elements unique up to a permutation of subgroup elements.

A subgroup of order |H| = 2 has only one generator, since every group element is its own

inverse. There are 15 non-identity elements in G to choose from, so there are 15 subgroups

of order 2. Subgroups of order 4 are generated by multiplication of two group elements.

There are 15 choices for the first and 14 choices for the second. However, each order-4

Equilibria and traveling waves of plane Couette flow 7

subgroup can be generated by 3 · 2 different choices of generators. For example, any two

of τx , τz , τxz in any order generate the same group T . Thus there are (15 · 14)/(3 · 2) = 35

subgroups of order 4.

Subgroups of order 8 have three generators. There are 15 choices for the first generator,

14 for the second, and 12 for the third. There are 12 choices for the third generator and

not 13, since if it were chosen to be the product of the first two generators, we would

get a subgroup of order 4. Each order-8 subgroup can be generated by 7 · 6 · 4 different

choices of three generators, so there are (15 · 14 · 12)/(7 · 6 · 4) = 15 subgroups of order

8. In summary: there is the group G itself, of order 16, 15 subgroups of order 8, 35 of

order 4, 15 of order 2, and 1 (the identity) of order 1, or 67 subgroups in all (Halcrow

2008). This is whole lot of isotropy subgroups to juggle; fortunately, the observations of

§ 3.1 show that there are only 5 distinct conjugacy classes in which we can expect to find

equilibria.

The 15 order-2 groups fall into 8 distinct conjugacy classes, under conjugacies between

σx τx and σx and σz τz and σz . These conjugacy classes are represented by the 8 isotropy

groups generated individually by the 8 generators σx , σz , σxz , σx τz , σz τx , τx , τz , and

τxz . Of these, the latter three imply periodicity on smaller domains. Of the remaining

five, σx and σx τz allow traveling waves in z, σz and σz τx allow traveling waves in x. Only

a single conjugacy class, represented by the isotropy group

{e, σxz } , (3.10)

breaks both continuous translation symmetries. Thus, of all order-2 isotropy groups, we

expect only this group to have equilibria. EQ9 , EQ10 , and EQ11 described below are

examples of equilibria with isotropy group {e, σxz }.

Of the 35 subgroups of order 4, we need to identify those that contain σxz and thus

support equilibria. We choose as the simplest representative of each conjugacy class the

isotropy group in which σxz appears in isolation. Four isotropy subgroups of order 4

are generated by picking σxz as the first generator, and σz , σz τx , σz τz , or σz τxz as the

second generator (R for reflect-rotate):

R = {e, σx , σz , σxz } = {e, σxz } × {e, σz }

Rx = {e, σx τx , σz τx , σxz } = {e, σxz } × {e, σx τx } (3.11)

Rz = {e, σx τz , σz τz , σxz } = {e, σxz } × {e, σz τz }

Rxz = {e, σx τxz , σz τxz , σxz } = {e, σxz } × {e, σz τxz } ' S .

These are the only isotropy groups of order 4 containing σxz and no isolated translation

elements. Together with {e, σxz }, these 5 isotropy subgroups represent the 5 conjugacy

classes in which expect to find equilibria.

The Rxz isotropy subgroup is particularly important, as the Nagata (1990) equilibria

belong to this conjugacy class (Waleffe 1997; Clever & Busse 1997; Waleffe 2003), as do

most of the solutions reported here. The NBC isotropy subgroup of Schmiegel (1999) and

S of Gibson et al. (2008) are conjugate to Rxz under quarter-cell coordinate transforma-

tions. In keeping with previous literature, we often represent this conjugacy class with

S = {e, s1 , s2 , s3 } = {e, σz τx , σx τxz , σxz τz } rather than the simpler conjugate group Rxz .

Schmiegel’s I isotropy group is conjugate to our Rz ; Schmiegel (1999) contains many

examples of Rz -isotropic equilibria. R-isotropic equilibria were found by Tuckerman &

Barkley (2002) for plane Couette flow in which the translation symmetries were broken

by a streamwise ribbon. We have not searched for Rx -isotropic solutions, and are not

aware of any published in the literature.

The remaining subgroups of orders 4 and 8 all involve {e, τi } factors and thus involve

8 J. Halcrow, J. F. Gibson, and P. Cvitanović

states that are periodic on half-domains. For example, the isotropy subgroup of EQ7 and

EQ8 studied below is S × {e, τxz } ' R × {e, τxz }, and thus these are doubled states of

solutions on half-domains. For the detailed count of all 67 subgroups, see Halcrow (2008).

GHC presents a method for visualizing low-dimensional projections of trajectories in the

infinite-dimensional state space of the Navier-Stokes equations. Briefly, we construct an

orthonormal basis {e1 , e2 , · · · , en } that spans a set of physically important fluid states

ûA , ûB , . . . , such as equilibrium states and their eigenvectors, and we project the evolv-

ing fluid state û(t) = u − yx̂ onto this basis using the L2 inner product (2.3). It is

convenient to use differences from laminar flow, since û forms a vector space with the

laminar equilibrium at the origin, closed under addition. This produces a low-dimensional

projection

a(t) = (a1 , a2 , · · · , an , · · · )(t) , an (t) = (û(t), en ) , (3.12)

which can be viewed in 2d planes {em , en } or in 3d perspective views {e` , em , en }. The

state-space portraits are dynamically intrinsic, since the projections are defined in terms

of intrinsic solutions of the equations of motion, and representation independent, since

the inner product (2.3) projection is independent of the numerical or experimental repre-

sentation of the fluid state data. Such bases are effective because moderate-Re turbulence

explores a small repertoire of unstable coherent structures (rolls, streaks, their mergers),

so that the trajectory a(t) does not stray far from the subspace spanned by the key

structures.

There is no a priori prescription for picking a ‘good’ set of basis fluid states, and

construction of {en } set requires some experimentation. Let the S-invariant subspace be

the flow-invariant subspace of states u that are fixed under S; this consists of all states

whose isotropy group is S or contains S as a subgroup. The plane Couette system at

hand has a total of 29 known equilibria within the S-invariant subspace four translated

copies each of EQ1 - EQ6 , two translated copies of EQ7 (which have an additional τxz

symmetry), plus the laminar equilibrium EQ0 at the origin. As shown in GHC, the

dynamics of different regions of state space can be elucidated by projections onto basis

sets constructed from combinations of equilibria and their eigenvectors.

In this paper we present global views of all invariant solutions in terms of the or-

thonormal ‘translational basis’ constructed in GHC from the four translated copies of

EQ2 :

τx τz τxz

e1 = c1 (1 + τx + τz + τxz ) ûEQ2 + + +

e2 = c2 (1 + τx − τz − τxz ) ûEQ2 + − − (3.13)

e3 = c3 (1 − τx + τz − τxz ) ûEQ2 − + −

e4 = c4 (1 − τx − τz + τxz ) ûEQ2 − − +,

where cn is a normalization constant determined by ken k = 1. The last 3 columns indicate

the symmetry of the basis vector under half-cell translations; e.g. ±1 in the τx column

implies τx ej = ±ej .

We seek equilibrium solutions to (2.1) of the form u(x, t) = uEQ (x) and traveling-wave

or relative equilibrium solutions of the form u(x, t) = uTW (x − ct) with c = (cx , 0, cz ).

Equilibria and traveling waves of plane Couette flow 21

Gibson, J. F. & Cvitanović, P. 2009 Periodic orbits of plane Couette flow. In preparation.

Gibson, J. F., Halcrow, J. & Cvitanović, P. 2008 Visualizing the geometry of state-space

in plane Couette flow. J. Fluid Mech. 611, 107–130, arXiv:0705.3957.

Gilmore, R. & Letellier, C. 2007 The Symmetry of Chaos. Oxford: Oxford Univ. Press.

Golubitsky, M. & Stewart, I. 2002 The symmetry perspective. Boston: Birkhäuser.

Halcrow, J. 2008 Geometry of turbulence: An exploration of the state-space of plane

Couette flow. PhD thesis, School of Physics, Georgia Inst. of Technology, Atlanta,

ChaosBook.org/projects/theses.html.

Halcrow, J., Gibson, J. F., Cvitanović, P. & Viswanath, D. 2009 Heteroclinic connections

in plane Couette flow 621, 365–376, arXiv:0808.1865.

Hamilton, J. M., Kim, J. & Waleffe, F. 1995 Regeneration mechanisms of near-wall turbu-

lence structures. J. Fluid Mech. 287, 317–348.

Harter, W. G. 1993 Principles of Symmetry, Dynamics, and Spectroscopy. New York: Wiley.

Hoyle, R. 2006 Pattern Formation: An Introduction to Methods. Cambridge: Cambridge Univ.

Press.

J.E. Dennis, Jr., & Schnabel, R. B. 1996 Numerical Methods for Unconstrained Optimization

and Nonlinear Equations. Philadelphia: SIAM.

Jiménez, J., Kawahara, G., Simens, M. P., Nagata, M. & Shiba, M. 2005 Characterization

of near-wall turbulence in terms of equilibrium and bursting solutions. Phys. Fluids 17,

015105.

Kim, H., Kline, S. & Reynolds, W. 1971 The production of turbulence near a smooth wall

in a turbulent boundary layer. J. Fluid Mech. 50, 133–160.

Kleiser, L. & Schuman, U. 1980 Treatment of incompressibility and boundary conditions

in 3-D numerical spectral simulations of plane channel flows. In Proc. 3rd GAMM Conf.

Numerical Methods in Fluid Mechanics (ed. E. Hirschel), pp. 165–173. GAMM, Viewweg,

Braunschweig.

Lan, Y. & Cvitanović, P. 2008 Unstable recurrent patterns in Kuramoto-Sivashinsky dynam-

ics. Phys. Rev. E 78, 026208, arXiv.org:0804.2474.

Marsden, J. E. & Ratiu, T. S. 1999 Introduction to Mechanics and Symmetry. New York,

NY: Springer-Verlag.

Nagata, M. 1990 Three-dimensional finite-amplitude solutions in plane Couette flow: bifurca-

tion from infinity. J. Fluid Mech. 217, 519–527.

Nagata, M. 1997 Three-dimensional traveling-wave solutions in plane Couette flow. Phys. Rev.

E 55 (2), 2023–2025.

Peyret, R. 2002 Spectral Methods for Incompressible Flows. Springer-Verlag.

Schmiegel, A. 1999 Transition to turbulence in linearly stable shear flows. PhD thesis, Philipps-

Universität Marburg, available on archiv.ub.uni-marburg.de/diss/z2000/0062.

Schneider, T., Gibson, J., Lagha, M., Lillo, F. D. & Eckhardt, B. 2008 Laminar-

turbulent boundary in plane Couette flow. Phys. Rev. E. 78, 037301, arXiv:0805.1015.

Skufca, J. D., Yorke, J. A. & Eckhardt, B. 2006 Edge of chaos in a parallel shear flow.

Phys. Rev. Lett. 96 (17), 174101.

Tuckerman, L. S. & Barkley, D. 2002 Symmetry breaking and turbulence in per-

turbed plane Couette flow. Theoretical and Computational Fluid Dynamics 16, 91–97,

arXiv:physics/0312051.

Viswanath, D. 2007 Recurrent motions within plane Couette turbulence. J. Fluid Mech. 580,

339–358, arXiv:physics/0604062.

Viswanath, D. 2008 The dynamics of transition to turbulence in plane Couette flow. In Math-

ematics and Computation, a Contemporary View. The Abel Symposium 2006 , Abel Sym-

posia, vol. 3. Berlin: Springer-Verlag, arXiv:physics/0701337.

Waleffe, F. 1990 Proposal for a self-sustaining mechanism in shear flows. Center for Turbulence

Research, Stanford University/NASA Ames, unpublished preprint (1990).

Waleffe, F. 1995 Hydrodynamic stability and turbulence: beyond transients to a self-sustaining

process. Stud. Applied Math. 95, 319–343.

Waleffe, F. 1997 On a self-sustaining process in shear flows. Phys. Fluids 9, 883–900.

Waleffe, F. 1998 Three-dimensional coherent states in plane shear flows. Phys. Rev. Lett. 81,

4140–4143.

22 J. Halcrow, J. F. Gibson, and P. Cvitanović

Waleffe, F. 2003 Homotopy of exact coherent structures in plane shear flows. Phys. Fluids

15, 1517–1543.

Wang, J., Gibson, J. F. & Waleffe, F. 2007 Lower branch coherent states in shear flows:

transition and control. Phys. Rev. Lett. 98 (20).

The long version

APPENDIX C

[A shorter (but more carefully proofread) version of the material covered in this appendix

can be found in ref. [30].]

In long-time simulations of turbulence, symmetry considerations can be disposed oﬀ in

a few lines: In absence of boundary conditions, the Navier-Stokes equations are equivariant

under any 3D translation, 3D rotation, and x → −x, u → −u inversion through the

origin that we shall refer to as σxz . The pair of counter-moving plane Couette ﬂow walls

restricts the translation symmetry to 2D in-plane translations. The streamwise shearing

motion restricts the 3D rotation symmetry to a single discrete symmetry operation, the

wall interchange wall-normal-streamwise rotation by π about z-axis that we shall refer

to as σx (symmetry referred to as “centrosymmetry” by Tuckerman and Barkley [84]).

Together with inversion σxz , this symmetry generates a discrete dihedral group D1,x ×D1,z =

{e, σx , σz , σxz } of order 4,

σx [u, v, w](x, y, z) = [−u, −v, w](−x, −y, z) (107)

σxz [u, v, w](x, y, z) = [−u, −v, −w](−x, −y, −z) ,

where σz is the spanwise reﬂection in z, and and σxz = σx σz . It would be more correct to

refer to the subgroup of rotations generated by σx as C2,x (or D1,x , the integers modulo

2 notation very frequently used in applied mathematics literature [36]) but we use the

isomorphic group D1,x here to distinguish this group from the Cn,x subgroups of SO(2) in

what follows. The subscripts on σx , σz , and σxz are chosen to indicate the sign changes in

x, y and z under the action of each symmetry.

With periodic boundary conditions, the two translations are restricted to the compact

SO(2)x × SO(2)z continuous two-parameter group of streamwise-spanwise translations,

SO(2) is isomorphic to the circle group S 1 , the group of rotations in plane. The equations

of plane Couette ﬂow with boundary conditions (13) are thus equivariant under the group

GPCF = O(2)x ×O(2)z = D1,x SO(2)x ×D1,z SO(2)z , where subscript x indicates stream-

wise translations and wall-normal-streamwise inversion, z indicates spanwise translations

and inversion, and stands for a semi-direct product. As far as a generic turbulent state

is concerned, the symmetry discussion can stop here.

However, for compact time-invariant solutions (equilibria, relative equilibria, periodic

solutions) and their instabilities the symmetries play a paramount role [34, 28], both for

their classiﬁcation and for understanding the dynamics connecting them, so here we shall

need a much ﬁner grained description of symmetries.

The key observation is that the G-equivariance of the equations does not imply G-

invariance of their solutions. A solution of an G-equivariant equation is in general mapped

64

into a group orbit of solutions under the action of G, that is, a set of solutions that map into

each other under the action of G. A solution may be invariant under an isotropy subgroup

(or stabilizer subgroup) H ⊂ G; that is, gu = u for g ∈ H. At one extreme, a generic

turbulent solution has the trivial isotropy subgroup Hturb = {e}. The other extreme is the

laminar solution whose isotropy group is the full GPCF . The full classiﬁcation of the lattice

of intermediate isotropy subgroups is a laborious undertaking, even for a set of symmetries

as plain as the ones considered here; fortunately, the set of dynamically important isotropy

subgroups is relatively small. We refer the reader to the immense literature on the subject

(we enjoyed reading [?] and [?] most), and give here a briefest of summaries, suﬃcient for

the classiﬁcation of the time-invariant solutions encountered so far.

Why are the equilibria for moderate Re born equipped with non-trivial isotropy groups?

The physical intuition is that while the nonlinear convective terms amplify unstable ﬂuid

motions, the dissipation damps out ﬂuid states with rapidly varying velocity ﬁelds. For

moderate Re the balance of the two smooths out the long-time solutions, and equilibria

with most symmetry tend to be the smoothest time-invariant solutions. A bit of history

illustrates this. The laminar solution is the smoothest, least dissipative solution. Nagata

discovered the EQ1 and EQ2 equilibria in 1990 by continuing a known solution from Taylor-

Couette ﬂow to plane Couette. Waleﬀe calculated the same solutions in a diﬀerent way,

noted that they are invariant under ‘shift-rotate’ and ‘shift-reﬂect’ isotropy subgroup that

we refer to below as the subgroup S, and utilized this symmetry to reduce the dimensional-

ity of his PDE discretizations. Busse and Clever [?] also noted this invariance. Schmiegel [?]

undertook an investigation of several isotropy subgroups, and in what follows we shall note

the subgroups that he identiﬁed. We started our exploration of plane Couette dynamics

around Nagata/Waleﬀe equilibria, and focused our searches for new equilibria within the

S-invariant subspace, since this symmetry restriction ﬁxes the x, z phase of solutions and

reduces the dimensionality of their unstable manifolds. In what follows we ﬁnd primar-

ily S-invariant equilibria because we initiated our guesses within that invariant subspace.

However, as the subspace is unstable, errors accrued by long numerical simulations had

suﬃcient deviation from the S-invariant subspace that the Newton search detected close-by

equilibria whose isotropy subgroup is S3 (discussed below). One would do well to look for

solutions with other symmetries by initiating searches within other invariant subspaces.

A state with no symmetry: A turbulent state space trajectory isotropy group is Hturb =

{e}. Such trajectory maps into a diﬀerent trajectory under the action of the group,

with translations generating a torus, and the inversions generating four tori of symmetry-

equivalent solutions.

Relative equilibria: Let us ﬁrst dispose of continuous symmetries. Equivariance of plane

Couette ﬂow equations under continuous translations allows for relative equilibria solutions,

i.e., solutions that are steady in a frame moving with a constant velocity in the [x, z] plane.

A generic relative equilibrium solution traces out a circle or torus in state space; this set

is both continuous translation and time invariant. The sign changes under reﬂections /

rotations σx , σz imply particular centers of symmetry and thus ﬁx the phase of a ﬁeld with

respect to x and z, respectively. Hence the presence of σx in an isotropy group rules out

traveling waves in x, and likewise for z. The isotropy subgroups with relative equilibria

solutions are the streamwise-spanwise traveling SO(2)x × SO(2)z , the streamwise traveling

SO(2)x × D1,z Cm,z , and the spanwise traveling D1,x Cm,x × SO(2)z (we explain the

emergence of cyclic subgroups Cm next).

Equilibria: Velocity ﬁelds invariant under the inversion σxz (i.e., under both σx and σz

65

ﬂips) cannot be traveling waves, so we search for equilibria in subspaces whose isotropy

group includes the inversion among its group elements.

O(2) is not an abelian group, since reﬂections σ and translations τ along the same axis

do not commute [?]: στ = τ −1 σ. Rewriting this relation as στ 2 = τ −1 στ , we note that

σz τz (0, z ) = τ −1 (0, z /2) σz τ (0, z /2) . (109)

x in x followed by a ﬂip across x = 0 is conjugate to a ﬂip across x = 0. For x = Lx /2

−1/4 1/4

we have σx τx = τx σx τx . Hence in classifying isotropy groups we can trade half-shift

and ﬂip σx τx in favor of a ﬂip σx in the shifted coordinate system. However, if both σx

and σx τx are elements of a subgroup, neither can be eliminated, as the conjugation simply

interchanges them.

Periodic boundary conditions on equilibrium solutions identify the ﬁelds on opposing

boundaries and restrict their isotropy subgroups to discrete cyclic subgroups Cm,x × Cn,x ⊂

SO(2)x × SO(2)z along the span-, stream-wise directions [?]. For rational shifts x /Lx =

m/n we simplify the notation a bit by rewriting (108) as

τx±m/n = τ (±mLx /n, 0) , τz±m/n = τ (0, ±mLz /n) , 0 < m < n, (110)

and, as m/n = 1/2 will loom large in what follows, omit 1/2 exponents:

it is the [Lx /n, 2, Lz ]-cell solution repeated n times downstream the cell. Invariance under

1/n

τx indicates that a [Lx , 2, Lz ] is redundant, in the sense that a cell of length Lx /n captures

the same state, so it quicker and easier to compute it in the small cell. However, as we

need to study bifurcations into states with wavelengths longer than the initial state, the

linear stability computations need to be carried out in the full [Lx , 2, Lz ] cell. Suppose

EQ is an equilibrium solution in the Ω1/3 = [Lx /3, 2, Lz ] cell. Then we shall refer to the

same solution repeated thrice in Ω = [Lx , 2, Lz ] as 3 × EQ. This solution is invariant under

1/3 2/3

C3,x = {e, τx , τx } subgroup.

Which subgroup is realized as an equilibrium isotropy subgroup is a physical question.

As the shearing is streamwise, the earliest bifurcations (as function of increasing Re) have

something to do with streamwise instabilities. As the structure of Navier-Stokes solutions is

a sensitive function of spanwise / streamwise / wall-normal nonlinear interactions and ratios

of cell and wall-unit scales, which solutions arise ﬁrst is a highly nontrivial issue [?, ?]. In

addition, for small cells the existence of solutions is a sensitive function of (and an artifact

of) the periodicities imposed by the cell aspect ratios [Lx , 2, Lz ].

A velocity ﬁeld constant along the streamwise or spanwise direction is always damped

by the ﬂow to the laminar solution [?, ?, ?], so equilibrium solutions have to wiggle in

both directions [?]. The more wiggles a solution has, the higher its dissipation rate.

Solutions invariant under the combinations of reﬂections (107) and half-cell shifts (111)

allow for the gentlest variation in velocity ﬁelds across the cell, so these generally have the

66

lowest dissipation rates, are the least unstable, and consequently are the ones to emerge

ﬁrst as equilibrium solutions as Re is increased. So we commence our discussion of equilib-

rium isotropy subgroups by SO(2)x × SO(2)z translations restricted to order 4 subgroup of

spanwise-streamwise translations of half the cell length,

of the full symmetry group contains all half-shift isotropy subgroups within it.

For spanwise full-cell z = Lz and half-cell z = Lz /2 shifts (109) implies

τz−1 σz τz = σz (115)

τz−1/4 σz τz1/4 = σz τ z , (116)

Let us ﬁrst enumerate the subgroups H ⊂ G. From (115) we see that the half-shifts

and reﬂections commute, so G is an abelian subgroup of O(2)x × O(2)z . Subgroups

can be of order |H| = {1, 2, 4, 8, 16}. A subgroup is generated by multiplication of a set of

generator group elements, with the choice of generator elements unique up to a permutation

of subgroup elements. A subgroup of order |H| = 2 has only one generator, since every group

element is its own inverse. There are 15 non-identity elements to choose from, so there are

15 subgroups of order 2. Subgroups of order 4 are generated by multiplication of two group

elements. There are 15 choices for the ﬁrst and 14 choices for the second. Equivalent

subgroup generators can be picked in 3 · 2 diﬀerent ways, so there are (15 · 14)/(3 · 2) = 35

subgroups of order 4. Subgroups of order 8 have three generators. There are 15 choices for

the ﬁrst, 14 for the second, and 12 for the third (If the element were the product of the

ﬁrst two, we would get a subgroup of order 4). There are 7 · 6 · 4 ways of picking equivalent

generators, i.e., there are (15 · 14 · 12)/(7 · 6 · 4) = 15 subgroups of order 8. In summary:

there is the group itself, of order 16, 15 subgroups of order 8, 35 of order 4, 15 of order

2, and 1 (the identity) of order 1, or 67 subgroups in all.This is whole lot of subgroups to

juggle; fortunately, the following observations reduce this number to 5 isotropy subgroups

that suﬃce –up to conjugacies– to describe all equilibria studied here.

(a) The isotropy subgroups for which we seek equilibrium solutions include σxz within its

elements.

(b) The right hand side of (116) is a product of two group elements, but the left hand side

is a similarity transformation that moves the coordinate frame origin by a quarter-

shift. Hence we can always trade in a single spanwise (or streamwise) shift-reﬂect

combination στ in favor of a reﬂect operation σ. (If both σj and σj τj are elements

of a subgroup, neither can be eliminated, as the similarity transformation simply

interchanges them.)

(c) By (112) any state whose isotropy subgroup includes a factor {e, τi } is a doubled state.

lack of a systematic labeling scheme more compact than listing the group elements.

67

(a) (b)

Figure 26: (a) The isotropy group of equilibrium EQ8 is Axz given in (121). According to

(112), factor {e, τxz } means that the state tiles the [Lx , Lz ] cell twice. How so? Transport

the spanwise / wall-normal section distance Lx /2 along the x direction, and then shift it

spanwise by Lz /2 along the z direction. The equilibrium solution in this half-cell, invariant

under the R isotropy subgroup (124) tiles the cell twice. (b) x-average of EQ8 velocity

ﬁeld û(y, z) is C2,z = {e, τz } (spanwise half-cell shift) invariant. The R invariance (isotropy

subgroup (124)) under discrete reﬂections is immediate. ΩW03 cell, Re = 270.

reduced to the 8 order-2 groups {e, g} where g is one of σx , σz , σxz , σx τz , σz τx , τx , τz , and

τxz . Of these, the latter three imply periodicity on smaller domains. Of the remaining ﬁve,

σx and σx τz support traveling waves in z, σz and σz τx support traveling waves in x. Only

a single subgroup of order |H| = 2, generated by σxz , supports equilibria:

Four distinct isotropy subgroups of order 4 are generated by picking σxz as the ﬁrst

generator, and σz , σz τx , σz τz or σz τxz as the second generator (R for reﬂect-rotate):

Rx = {e, σx τx , σz τx , σxz } = S3 × {e, σz τx } (118)

Rz = {e, σx τz , σz τz , σxz } = S3 × {e, σz τz } = I

Rxz = {e, σx τxz , σz τxz , σxz } = S3 × {e, σz τxz } = S .

Each of these isotropy subgroups of GP CF has three other isotropy subgroups conjugate

to it under quarter-cell shifts (116). For example, the three groups {e, σx τx , σz , σxz τx },

{e, σx , σz τz , σxz τx }, and {e, σx τx , σz τz , σxz τxz }, are conjugate to R under quarter-cell coor-

dinate shifts in x, z, and x, z respectively. Thus the isotropy subgroups given by (118)

account for 12 of the 35 order-4 subgroups of G. Note that R, Rx , Rz and Rxz are the

only inequivalent isotropy subgroups of order 4 containing σxz and no isolated translation

elements. Together with S3 these yield the total of 5 non-conjugate isotropy subgroups in

which we can expect to ﬁnd equilibria.

The order-8 subgroups of G are generated as in our combinatorial count above, by

multiplying out all distinct products of three Z2 subgroups, {e, σi τj }×{e, σk τ }×{e, σm τn },

68

i, j, k, , m, n ∈ {0, x, z}, and σ0 = τ0 = e;

{e, σx , σz , τx , σxz , σx τx , σz τx , σxz τx }

{e, σx , σz , τz , σxz , σx τz , σz τz , σxz τz }

{e, σx , τx , σx τx , σz τz , σz τxz , σxz τz , σxz τxz }

{e, σx , τx , τz , τxz , σx τx , σx τz , σx τxz }

{e, σx , τxz , σz τx , σz τz , σx τxz , σxz τx , σxz τz }

{e, σx , τz , σx τz , σz τx , σz τxz , σxz τx , σxz τxz }

{e, σxz , τxz , σx τx , σx τz , σz τx , σz τz , σxz τxz }

{e, σz , τx , σx τz , σz τx , σx τxz , σxz τz , σxz τxz }

{e, σz , τx , τz , τxz , σz τx , σz τz , σz τxz }

{e, σz , τxz , σx τx , σx τz , σz τxz , σxz τx , σxz τz }

{e, σz , τz , σx τx , σz τz , σx τxz , σxz τx , σxz τxz }

{e, τx , σxz , σx τz , σz τz , σx τxz , σz τxz , σxz τx }

{e, τx , τz , σxz , τxz , σxz τx , σxz τz , σxz τxz }

{e, τz , σxz , σx τx , σz τx , σx τxz , σz τxz , σxz τz }

(119)

They all involve factors of {e, τj } and are thus repeats of a smaller fundamental domain

half-cells. The 15 order-8 isotropy subgroups can be reduced by coordinate transformation

to the following 9 subgroups: R × {e, τx }, R × {e, τz }, R × {e, τxz }, Rx × {e, τz }, Rz ×

{e, τx }, Rx × {e, τxz } = Rz × {e, τxz }, T × {e, σx }, T × {e, σz }, and T × {e, σxz }. As each

of these subgroups contains pure translation elements, each can be reduced to a simpler

symmetry subgroup on a smaller domain. R × {e, τxz } is the isotropy subgroup of EQ7 and

EQ8 .

There are 4 subgroups of order 8 which by (112) tile the cell twice: A? is generated by

elements

Az = {e, σx , σz , σxz , τz , σx τz , σz τz , σxz τz } = R × {e, τz }

Axz = {e, σx , σz , σxz , τxz , σx τxz , σz τxz , σxz τxz } (121)

= R × {e, τxz }

A? = {e, σx , σz , σxz , τ? , σx τ? , σz τ? , σxz τ? } .

As illustrated by ﬁgure 26, Axz is the isotropy subgroup of equilibria EQ7 and EQ8 discussed

below (after a spanwise quarter-shift (116) is applied to eliminate some half-shifts). There

is one subgroup which appears with 4 translational copies, generated by σx , σz and τxz .

It contains copies of all R subgroups (118). There are 3 subgroups of order 8 which are

invariant under the translational similarity transform. These all contain T and its products

with either σx , σz , or σxz .

Finally, the laminar solution EQ0 isotropy group |H| = 16 is the full subgroup |G| = 16.

Its multiplicity is m = 1, i.e., it appears only once.

69

What about G = D1,x × Cm,x × D1,z × Cn,z subgroups of O(2)x × O(2)z for m, n > 2?

By (112), a solution in [Lx , 2, Lz ] repeated n times across the cell of size [Lx , 2, nLz ] is

also a solution, but with a larger isotropy group, invariant under Cn,z . Already for small

aspect cells studied here we encounter solutions (such as 2 × EQ1 ) with C4 symmetry,

invariant under quarter shifts τ 1/4 . They are important, as they can bifurcate into states

with wavelengths longer than the initial state.

Now that we have enumerated the ﬁve distinct isotropy subgroups, it is helpful to

introduce notation for the corresponding invariant subspaces of solutions of Navier-Stokes

equations (by an “invariant subspace” we mean that a state of the ﬂuid initiated within

such a subspace remains in it for all times). We denote invariant subspaces of solutions of

(13) as follows. Let S denote the order-4 isotropy subgroup Rxz (118), and S3 = {1, σxz }

the order-2 isotropy subgroup (117) inversion. Let U be the space of velocity ﬁelds that

satisfy the boundary and zero-divergence conditions of plane Couette ﬂow, and let US and

US3 be the S and S3 -invariant subspaces of U respectively: US = {u ∈ U | su = u, ∀s ∈ S},

and US3 = {u ∈ U | σxz u = u}. As σxz changes the sign of both x and z, US and US3 do

not admit relative equilibrium solutions.

collides with the nonlinear dynamics and ﬂuid dynamics terminology. Here we refer to a

point in the dynamical system’s d− d state space where ȧ = v(a) = 0 as an ‘equilibrium’

point, and the point in a 3D ﬂuid ﬂow where the Eulerian velocity vanishes as a ‘stagnation

point’ point. The corresponding notions for symmetries of ﬂows under continuous Lie group

transformations we usually preface by a group-theoretic qualiﬁer, such as ‘G−.’ We refer

to a ‘group orbit’ in contradistinction to ‘orbit,’ the set of state space points related by

time evolution, and to ‘G−ﬁxed point space’ in contradistinction to dynamically stationary

‘equilibrium point.’

Example: Order 8 invariant subgroups describe doubled states. The isotropy subgroup

of EQ7 and EQ8 is

itself to a binary labeling of its |G| = 16 elements:

In this form the group multiplication table is simply mod 2 addition of j ; for example,

g1011 g1100 = g0111 .

70

In the Cn -invariance case the translational label 4 in (122) is n-ary, i.e., in group

multiplication it is composed by addition mod n.

Associated with action of G are projection on 16 irreducible representations, symmet-

ric/antisymmetric under action of g1 2 3 4 :

1

P±±±± = (1 ± σx )(1 ± σz )(1 ± τx )(1 ± τz ) . (123)

24

We shall ﬁnd these projection operators handy for visualization of state space ﬂows.

There is only one isotropy group of order 2: S3 = {g0000 , g1010 }. Multiplication with 7

group elements of form g02 3 4 yields 7 isotropy subgroups of order 4

S3 × g0110 = {g0000 , g0110 , g1010 , g1100 } = Rx

S3 × g0011 = {g0000 , g0011 , g1001 , g1010 } = Rz

S3 × g0111 = {g0000 , g0111 , g1010 , g1101 } = Rxz , (124)

S3 × g0001 = {g0000 , g0001 , g1010 , g1011 } = Dz

S3 × g0101 = {g0000 , g0101 , g1010 , g1111 } = Dxz . (125)

Group elements of form g12 3 4 all appear in the above subgroups and contribute no further

order 4 subgroups.

Multiplication of R’s with group elements not already within them generates 6 isotropy

subgroups of order 8 (all names temporary, for now)

B = {g0000 , g0110 , g1010 , g1100 , g0011 , g0101 , g1001 , g1111 }

C = {g0000 , g0011 , g1001 , g1010 , g0100 , g0111 , g1101 , g1110 } (126)

Uz = {g0000 , g0010 , g1000 , g1010 , g0001 , g0011 , g1001 , g1011 }

Uxz = {g0000 , g0010 , g1000 , g1010 , g0101 , g0111 , g1101 , g1111 } (127)

Rx × g0001 = A , Rx × g0010 = Ux , Rx × g0011 = B

Rz × g0001 = Uz , Rz × g0100 = C , Rz × g0101 = B

Rxz × g0001 = A , Rxz × g0010 = Uxz , Rxz × g0011 = C (128)

Multiplication of D’s with group elements not already within them generates further sub-

groups.

71

S3 isotropy subgroup {1, σxz } is order 2. The smaller group means a bigger quotiented

state space. There are eight copies of any S3 -invariant ﬁeld, generated by the four half-cell

translations and by reﬂection about the z = 0 plane.

Let u be s3 -invariant: u = s3 u. s3 commutes with all four elements of T = {e, τx , τz , τxz }.

Thus s3 (τ u) = τ s3 u = τ u, and so τ u is s3 symmetric for all four choices of τ .

Moreover, s1 u is S3 -invariant for S3 -invariant u (and generally s1 u = τ u for τ ∈

{τx , τz , τxz }. We have u = s3 u, so s1 u = s1 s3 u = s3 (s1 u). That then gives us eight

copies: sτ u for s ∈ {e, s1 } and τ ∈ T . Trying to expand by s2 multiplication doesn’t pro-

duce any new ﬁelds, since s2 u = s1 s3 u = s1 u. So in general, each S3 -invariant ﬁeld has eight

copies in the s3 invariant subspace, unless these copies coincide due to further symmetries

(e.g. ﬁelds with additional s1 symmetry total only four copies). Figure 27 shows the eight

S3 -invariant copies of EQ11 .

Translations of half the cell length in the spanwise and/or streamwise directions com-

mute with S. These operators generate T , a discrete subgroup (113) of the continuous

translational symmetry group SO(2) × SO(2) : Take EQ2 as an example of an S3 -invariant

ﬁeld. It has four distinct copies within US , and US is a subset of the S3 -invariant subspace.

So we have an example of an S3 -invariant ﬁeld with at least four copies in the S3 -invariant

subspace.

EQ7 and EQ8 have multiplicity 2, as their isotropy subgroup is (121).

The basis elements of Gibson basis which are in the s1 and s2 subspaces are given by:

αjΦuσjkl (y) + Φvσjkl (y) + γkΦw

σjkl (y) = 0, (129)

where

⎧

⎪

⎨sin jx j<0

gj (x) = 1 j=0 (130)

⎪

⎩

cos jx j>0

and where Φu,v,w

σjkl (y) is given by Rl (y) or a multiple of Ql (y) deﬁned as

⎧

⎪ 1+y

⎪

⎨ 2 l=0

1−y

Rl (y) = 2 l=1 , (131)

⎪

⎪

⎩√ 1 (Pl−2 (y) − Pl (y)) l≥2

2(2l+1)

y

Ql (y) = Rl (y )dy .

−1

x = −x gives:

−gj (x) j < 0

gj (−x) = (132)

gj (x) j≥0

Translation by a half period is given by

gj (x) j=0

gj (x + π) = (133)

−gj (x) otherwise

72

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 27: The eight copies of EQ11 within the S3 -invariant subspace: (a) uEQ11 (b)

τx uEQ11 (c) τz uEQ11 (d) τxz uEQ11 (e) σz uEQ11 (f) τx σz uEQ11 (g) σz τz uEQ11 and (h) σz τxz uEQ11 .

The S3 invariance of each solution has been veriﬁed numerically. ΩW03 cell, Re = 400.

73

Then for the y-dependent parts of the basis:

⎧

⎪

⎪ R1 (y) l=0

⎪

⎪

⎨R (y) l=1

0

Rl (−y) = (134)

⎪

⎪ Rl (y) l > 1 and l is even

⎪

⎪

⎩−R (y) l > 1 and l is odd

l

⎧

⎪

⎪ Q (1) − Q1 (y) l = 0

⎪ 0

⎪

⎪

⎪

⎨Q1 (1) − Q0 (y) l = 1

Ql (−y) = Q2 (1) − Q2 (y) l = 2 , (135)

⎪

⎪

⎪

⎪ Ql (y) l ≥ 3 and l is odd

⎪

⎪

⎩

−Ql (y) l ≥ 3 and l is even

⎛ ⎞ ⎛ ⎞

u u

s1 v ⎠ (x, y, z)

⎝ = ⎝ v ⎠ (x + Lx /2, y, −z) (136)

w −w

⎛ ⎞ ⎛ ⎞

u −u

s2 v ⎠ (x, y, z)

⎝ = ⎝ −v ⎠ (−x + Lx /2, −y, z + Lz /2) (137)

w w

⎛ ⎞ ⎛ ⎞

u −u

s3 ⎝ v ⎠ (x, y, z) = ⎝ −v ⎠ (−x, −y, −z + Lz /2) , (138)

w −w

each element of the basis set is either symmetric or antisymmetric with respect to each

symmetry, see table 15.

Note the following identity for Legendre polynomials:

Pl+1 (y) − Pl−1 (y)

Pl (y) = (139)

2l + 1

y y

Ql (y) = Rl (y ) = Pl−2 (y ) − Pl (y ) (140)

−1 −1

Pl−1 (y) − Pl−3 (y) Pl+1 (y) − Pl−1 (y)

= − (141)

2l − 3 2l + 1

Rl+1 (y) Rl−1 (y)

= − (142)

2l + 1 2l − 3

Since (Pm , Pn ) = δm n/(2n + 1), (Rm , Rn ) = δm n(1/(2m + 1) + 1/(2m − 3)) + δm,n+2 /(2m +

1) + δm,n−2 /(2m + 1).

74

σ j k l Φuσjkl (y) Φvσjkl (y) Φwσjkl (y) s1 s2 s3

0 0 0 {2, . . . , L}e Rl (y) 0 0 S A A

0 0 0 {2, . . . , L}o Rl (y) 0 0 S S S

1 0 0 {2, . . . , L}e 0 0 Rl (y) A S A

1 0 0 {2, . . . , L}o 0 0 Rl (y) A A S

0 j 0 {3, . . . , L−1}e Rl (y) −αjQl (y) 0 A S A

0 j 0 {3, . . . , L−1}o Rl (y) −αjQl (y) 0 A A S

0 −j 0 {3, . . . , L−1}e Rl (y) −αjQl (y) 0 A A S

0 −j 0 {3, . . . , L−1}o Rl (y) −αjQl (y) 0 A S A

1 j 0 {2, . . . , L}e 0 0 Rl (y) S S S

1 j 0 {2, . . . , L}o 0 0 Rl (y) S A A

1 −j 0 {2, . . . , L}e 0 0 Rl (y) S A A

1 −j 0 {2, . . . , L}o 0 0 Rl (y) S S S

0 0 k {2, . . . , L}e Rl (y) 0 0 A A S

0 0 k {2, . . . , L}o Rl (y) 0 0 A S A

0 0 −k {2, . . . , L}e Rl (y) 0 0 S A A

0 0 −k {2, . . . , L}o Rl (y) 0 0 S S S

1 0 k {3, . . . , L−1}e 0 −γkQl (y) Rl (y) A A S

1 0 k {3, . . . , L−1}o 0 −γkQl (y) Rl (y) A S A

1 0 −k {3, . . . , L−1}e 0 −γkQl (y) Rl (y) S A A

1 0 −k {3, . . . , L−1}o 0 −γkQl (y) Rl (y) S S S

0 j k {2, . . . , L}e γkRl (y) 0 -αjRl (y) S A A

0 j k {2, . . . , L}o γkRl (y) 0 -αjRl (y) S S S

0 j −k {2, . . . , L}e γkRl (y) 0 -αjRl (y) A A S

0 j −k {2, . . . , L}o γkRl (y) 0 -αjRl (y) A S A

0 −j k {2, . . . , L}e γkRl (y) 0 -αjRl (y) S S S

0 −j k {2, . . . , L}o γkRl (y) 0 -αjRl (y) S A A

0 −j −k {2, . . . , L}e γkRl (y) 0 -αjRl (y) A S A

0 −j −k {2, . . . , L}o γkRl (y) 0 -αjRl (y) A A S

1 j k {3, . . . , L−1}e γkRl (y) −αγjkQl (y) αjRl (y) S A A

1 j k {3, . . . , L−1}o γkRl (y) −αγjkQl (y) αjRl (y) S S S

1 j −k {3, . . . , L−1}e γkRl (y) −αγjkQl (y) αjRl (y) A A S

1 j −k {3, . . . , L−1}o γkRl (y) −αγjkQl (y) αjRl (y) A S A

1 −j k {3, . . . , L−1}e γkRl (y) −αγjkQl (y) αjRl (y) S S S

1 −j k {3, . . . , L−1}o γkRl (y) −αγjkQl (y) αjRl (y) S A A

1 −j −k {3, . . . , L−1}e γkRl (y) −αγjkQl (y) αjRl (y) A S A

1 −j −k {3, . . . , L−1}o γkRl (y) −αγjkQl (y) αjRl (y) A A S

Table 15: The polynomial triplets for a complete set of linearly independent basis functions

Φσjkl up to a given Fourier discretization |j| ≤ J, |k| ≤ K and polynomial degree l ≤ L.

‘A’ indicates that a member of the basis is anti-symmetric with respect to the symmetry

operator si and ‘S’ indicates symmetric. Indices are j and k are strictly positive unless

explicitly stated otherwise. e and o indicate the even and odd members of a set, respectively.

Note we could drop one of the columns since s1 s2 = s3 , so if one is S the other two are

either S,S or A,A and if one is A then the others are S, A or A,S.

75

TABLE OF CONTENTS

SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

I INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Poincaré, Turbulence, and Hopf’s Last Hope . . . . . . . . . . . . . . . . 1

1.2 The Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 What is New in This Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 In the Beginning... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Scale Invariance and the Reynolds number . . . . . . . . . . . . . . . . . 5

2.3 Plane Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.3.1 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.3.2 On 2D Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.4 Turbulent plane Couette ﬂow . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.4.1 Reynolds Decomposition . . . . . . . . . . . . . . . . . . . . . . . 9

2.4.2 Wall Units and The “Law of the Wall” . . . . . . . . . . . . . . . 10

2.5 A Brief History of Turbulent Shear Flows . . . . . . . . . . . . . . . . . . 11

3.1 Mathematical Background . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1.1 Group Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1.2 Representation Theory . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1.3 Equivariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1.4 Irreducible Representations and Equivariance . . . . . . . . . . . . 19

3.2 Discrete Symmetry: The Lorenz Attractor . . . . . . . . . . . . . . . . . 20

3.3 Symmetries of Plane Couette Flow . . . . . . . . . . . . . . . . . . . . . . 21

IV METHODOLOGY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1 Discretization and the Method of Lines . . . . . . . . . . . . . . . . . . . 24

4.1.1 Spatial Discretization . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1.2 Temporal Discretization . . . . . . . . . . . . . . . . . . . . . . . . 25

4.2 A Survey of Methods for Integrating the Navier-Stokes Equations . . . . 26

ii

4.2.1 Primitive Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.3 Krylov Subspace Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4.3.1 Arnoldi Iteration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.3.2 GMRES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.4 Newton Solvers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.4.1 Termination of Newton Algorithms . . . . . . . . . . . . . . . . . 30

4.4.2 Line Search Algorithms . . . . . . . . . . . . . . . . . . . . . . . . 30

4.4.3 Model Trust Region Algorithms . . . . . . . . . . . . . . . . . . . 31

4.5 Finding Invariant Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.6 State Space Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

V EQUILIBRIA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Equilibria and Relative Equilibria . . . . . . . . . . . . . . . . . . . . . . 37

5.2 Bifurcations under Variation of Re . . . . . . . . . . . . . . . . . . . . . . 42

5.3 Bifurcations under Variation of Spanwise Width . . . . . . . . . . . . . . 43

VI HETEROCLINIC CONNECTIONS . . . . . . . . . . . . . . . . . . . . . . . . 46

6.1 Three Heteroclinic Connections for Re=400 . . . . . . . . . . . . . . . . . 46

6.1.1 A Heteroclinic Connection from EQ4 to EQ1 . . . . . . . . . . . . 46

6.1.2 Heteroclinic Connections from EQ3 and EQ5 to EQ1 . . . . . . . . 47

6.2 Heteroclinic Connection from EQ1 to EQ2 at Re = 225 . . . . . . . . . . 48

7.1 Turbulence as a walk about exact coherent states . . . . . . . . . . . . . . 49

7.2 Reﬁnement of Heteroclinic Connections . . . . . . . . . . . . . . . . . . . 51

7.3 For the Experimentalist . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

VIII CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

VITA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

iii

VITA

Jonathan Halcrow was born in 1981 in Port Jeﬀerson, NY, and started roaming the coun-

tryside between New York and Virginia for 14 years. Eventually he settled in Georgia and

attended North Gwinnett High School. After graduation in 1999, it was oﬀ to Georgia Tech

to get a B.S. in Math and another in Physics in 2003. Realizing that there’s no place like

home, he stayed there for graduate school.

In 2008 he got a haircut, cleaned his desk and the access corridor to it, started eat-

ing fruits and vegetables, started bicycling to ‘work,’ stopped smoking, lost weight, all in

preparation to accession to yuppiedom and long AARP years that are to follow. At some

point in all this he managed to ﬁnd the time to write a thesis.

82

- Which Functor is the Projective Line?Uploaded byflores3831_814460512
- Ex 6 SolutionsUploaded bySanjeev Shukla
- Captivating Symmetry by RNDr Daniela Richtarikova Slovak Technical University (Slovakia)Uploaded byJak Boumans
- fields.utoronto.ca Blaise Pascal: David Hilbert's Problems.Uploaded byMark Richard Hilbert(Mark Richard Rossetti)
- Maximality Methods in Local ProbabilityUploaded byHax
- mathUploaded byjojo2k
- sdarfdticleUploaded byOnur Öztırpan
- G.S. Ezra- On the symmetry properties of non-rigid molecules: Semifactorizability of the isometric groupUploaded byCvgdnnny
- Soft Neutrosophic GroupUploaded byAnonymous 0U9j6BLllB
- mathsUploaded byAnshu Kumar Gupta
- CS and IT.pdfUploaded bymm8871
- Sudoku.docxUploaded byAngy Isidro Hancco
- Galois theory notesUploaded byEric Parker
- Abstract Algebra - BeachyUploaded byIlirjan Cane
- MTECHCSNEWSYLL2012Uploaded byAlok H Tripathi
- Classic LieUploaded byAnonymous 9vFSHhJh0
- mathgen-1274246079Uploaded bysamsmith
- 2018SU B.sc Maths SyllabusUploaded bysachin81185
- Mathematics General Cbcs Draft SyllabusUploaded byHãrsh Shâw
- teach-2019Uploaded byEdward Aris Fajardo

- ChaosBook.org Chapter 10 overheads: Relativity for cyclistsUploaded bybirdtracks
- How well can one resolve the state space of a chaotic flow?Uploaded bybirdtracks
- Mimi Gross illustrated programUploaded bybirdtracks
- "Knowing when to stop: How noise frees us from determinism,"Uploaded bybirdtracks
- Nenad Belic remembers his father, comments by JackieUploaded bybirdtracks
- Jackie's letter to Elizabeth Phillips 1964Uploaded bybirdtracks
- Family Golmajer archiveUploaded bybirdtracks

- Uploaded byluisdaniel
- Arnold - Ergodic Problems of Classical MechanicsUploaded byDiana Uriza
- Quantum Spiral TheoryUploaded bySuraj Kumar
- Why Does Unsupervised Deep Learning Work? - A Perspective From Group TheoryUploaded byMichael Chiu
- Complex Surfaces and Sums of Complex Projective PlanesUploaded bysabsamtarte
- Costas Kounnas, Jan Troost and Nicolaos Toumbas- A wave-function for stringy universesUploaded byJomav23
- A family of p-adic isometries, ﬁxed points, and the number threeUploaded bydmtri
- SymHS15NotesUploaded byGerman Chiappe
- Fanghua Lin and Yisong Yang- Existence of Knots of Minimum Energy and Topological Growth Laws in the Faddeev ModelUploaded byPomac232
- 091 - The.Geometry.of.Discrete.Groups,.Beardon.A.F..(Springer.1995)(ISBN.3540907882)(T)(347s).pdfUploaded byplatinumwolf13610
- Lie groups and lie algebras for physicistsUploaded byJoel Curtis
- cis610sl7Uploaded bymaxxagain
- SpinorsUploaded byjosemeri
- Young's Inequality for ConvolutionUploaded byTiệp Minh Hoàng
- (Universitext) Elmer G. Rees (Auth.)-Notes on Geometry-Springer-Verlag Berlin Heidelberg (1983)Uploaded bylibrary gcpeshawar
- From Monopoles to Textures: A Survey of Topological Defects in Cosmological Quantum Field TheoryUploaded byDamian Sowinski
- ChainUploaded bytjny699
- 0905.2720v1 Edward Witten - Geometric Langlands from six dimensionsUploaded byJack Ignacio Nahmías
- The Hodge-Arakelov Theory of Elliptic CurvesUploaded bycharlyshaka1
- cosetsUploaded byHaq

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.