Edgar Knobloch Reprint List - Coupled Mean Flow …tardis.berkeley.edu/reprints/papers/KnMaVe_NYACADSCI974...Ann. N.Y. Acad. Sci. 974: 201–219 (2002). ©2002 New York Academy of

Ann. N.Y. Acad. Sci.

974: 201–219 (2002). ©2002 New York Academy of Sciences.

Coupled Mean Flow-Amplitude Equations for Nearly Inviscid Parametrically Driven Surface Waves

EDGAR KNOBLOCH,

a

CARLOS MARTEL,

b

AND JOSÉ M. VEGA

b

a

Department of Physics, University of California, Berkeley, California, USA

b

E.T.S.I. Aeronáuticos, Universidad Politécnica de Madrid, Madrid, Spain

A

BSTRACT

: Nearly inviscid parametrically excited surface gravity-capillarywaves in two-dimensional periodic domains of finite depth and both small andlarge aspect ratio are considered. Coupled equations describing the evolutionof the amplitudes of resonant left- and right-traveling waves and their interac-tion with a mean flow in the bulk are derived, and the conditions for theirvalidity established. In general the mean flow consists of an inviscid parttogether with a viscous streaming flow driven by a tangential stress due to anoscillating viscous boundary layer near the free surface and a tangential veloc-ity due to a bottom boundary layer. These forcing mechanisms are importanteven in the limit of vanishing viscosity, and provide boundary conditions forthe Navier–Stokes equation satisfied by the mean flow in the bulk. The stream-ing flow is responsible for several instabilities leading to pattern drift.

K

EYWORDS

: Faraday waves; streaming flow; gravity-capillary waves; parametric resonance

INTRODUCTION

Parametrically driven surface gravity-capillary waves in low viscosity fluids aretraditionally discussed using a velocity potential formulation. However, this formu-lation has a serious shortcoming in that it ignores large-scale streaming flows thatmay be driven by the time-averaged Reynolds stress in the oscillatory boundary lay-ers at the container walls, or at the free surface. We describe a systematic asymptotictechnique that includes such flows and show that these flows can interact nontriviallywith the waves that are in turn responsible for them. We show that this interactioncan lead to new types of instability of parametrically driven waves at finite ampli-tude, and discuss some of their consequences. We treat only systems with periodicboundary conditions, either in two dimensions or in a cylindrical container, and dis-cuss two cases. In the first the length of the periodic domain is large relative to thewavelength of the instability, and the mean flow contains an important inviscid con-tribution. In the second the wavelength and container length are comparable and thedynamics of the system is more sensitive to the container shape. In particular, if theshape is perturbed away from circular, the interaction between the streaming flow

Address for correspondence: Edgar Knobloch, Department of Physics, University of Cali-fornia, Berkeley, CA 94720, USA. Voice: 510-642-3395; fax: 510-643-8497.

[email protected]

202 ANNALS NEW YORK ACADEMY OF SCIENCES

and the wave amplitude is enhanced, and complex dynamics may result. These con-clusions generalize readily to containers of other shapes suggesting a new class ofexperiments.

THE FARADAY SYSTEM

Surface gravity-capillary waves or Faraday waves can be excited parametricallyby the vertical oscillation of a container.

1–3

We consider a container in the form of aright cylinder with horizontal cross-section

Σ

, filled level with the brim at

z

=

0. Inthis geometry the contact line is pinned at the lateral boundary and complicationsassociated with contact line dynamics are reduced. We nondimensionalize the gov-erning equations using the unperturbed depth

h

as unit of length and the gravity-cap-illary time [

g

/

h

+

T

/

(

ρ

h

3

)]

–1/2

as unit of time (here

g

is the gravitational acceleration,

T

is the coefficient of surface tension, and

ρ

is the density), and obtain

(1)

where

v

is the velocity,

f

is the associated vertical deflection of the free surface (con-strained by volume conservation),

Π

=

p

+

|

v

|

2

/2

+

(1 –

S

)

z

−

4

µω

2

z

cos2

ω

t

is thehydrostatic stagnation pressure,

n

is the outward unit normal to the free surface,

e

z

is the upward unit vector, and

∂Σ

denotes the boundary of the cross-section

Σ

(i.e.,the lateral walls). The real parameters

µ

>

0 and 2

ω

denote the amplitude and fre-quency of the forcing. The quantity

C

g

≡

ν

/(

gh

3

+

Th

/

ρ

)

1/2

, where

ν

is the kinematicviscosity, is a gravity-capillary number and

S

≡

T

/

(

T

+

ρ

gh

2

) is a gravity-capillarybalance parameter; these are related to the usual capillary number

C

≡

andBond number

B

≡

ρ

gh

2

/

T

by

(2)

The parameter

S

is such that 0

≤

S

≤

1 with

S

=

0 and

S

=

1 corresponding to the pure-ly gravitational limit (

T

=

0) and the purely capillary limit (

g

=

0), respectively.In this paper we consider the (nearly inviscid, nearly resonant, weakly nonlinear)

limit

(3)

where

Ω

is an inviscid eigenfrequency of the linearized problem around the flatstate. In this case the vorticity contamination of the bulk from the boundary layersat the walls and the free surface remains negligible for times that are not too long,and the flow in the bulk is correctly described by an inviscid formulation but with

t∂∂v v ∇ v×( )×– ∇Π– Cg∆v, ∇ v⋅ 0, if x y,( ) Σ, 1 z f ,< <–∈=+=

v 0 if z 1 or x y,( ) ∂Σ, f 0 if x y,( ) ∂Σ,∈=∈–= =

v n⋅t∂

∂ f ez n⋅( ), v∇ vT∇+( ) n⋅[ ] n× 0 at z f ,== =

Π v 2

2--------– 1 S–( ) f– S∇ f∇

1 f∇ 2+( )1 2/-----------------------------------⋅+

Cg v∇ vT∇+( ) n⋅[ ] n 4µω2 f 2ωt at zcos–⋅ f ,= =

ν ρ Th⁄

CgC

1 B+( )1 2/-------------------------, S 1

1 B+-------------.==

Cg << 1, ω Ω– << 1, µ << 1,

203KNOBLOCH et al.: COUPLED MEAN FLOW-AMPLITUDE EQUATIONS

boundary conditions determined by a boundary layer analysis. In general, this flowconsists of an inviscid part and a viscous part, hereafter called the streaming flow.As discussed elsewhere4,5 the streaming flow enters into the problem because thelinearized problem admits hydrodynamic (or viscous) modes,6 in addition to the usu-al surface modes. In the nearly inviscid limit the former decay more slowly than thesurface modes, and so are easily excited, forming the streaming flow. For small Cg,these modes take the form (v,Π, f) = (U,CgP,CgF)exp(Cgλt) + …, with the (real)eigenvalue λ < 0 given by

(4)

The associated (scaled) free surface deflection F is calculated a posteriori from thenormal stress balance across z = 0,

(5)

subject to F = 0 on ∂Σ, and = 0. Thus, in contrast to the surface modes, thehydrodynamic modes are nonoscillatory and exhibit O(Cg) free surface deflection.Moreover, these modes decay on an O( ) time scale, in contrast to the O( )time scale of the surface modes, and hence cannot be ignored a priori in a weaklynonlinear theory.

MODERATELY LARGE DOMAINS: L >> 1

In this regime it is possible to perform a multiscale analysis of the governingequations using Cg, L–1, and µ as unrelated small parameters. The problem is sim-plest in two dimensions, where we can use a stream function formulation—that is,we write v = (–ψz,0,ψx). We focus on two well-separated scales in both space (x ∼ 1and x >> 1) and time (t ∼ 1 and t >> 1), and derive equations for small, slowly varyingamplitudes A and B of left- and right-propagating waves defined by

(6)with similar expressions for the remaining fields. The quantities f ± and f m representresonant second order terms, whereas NRT denotes nonresonant terms. The super-script m denotes terms associated with the mean flow; f m depends weakly on timebut may depend strongly on x. A systematic expansion procedure5 then leads to theequations

(7)

(8)

∇ U⋅ 0, λU P∇– ∆U if x y,( ) Σ, 1 z 0,< <–∈+==

U 0 if z 1 or x y,( ) ∂Σ,∈–==

ez U⋅ 0, ez U∇ UT∇+( )⋅[ ] ez× 0 at z 0.== =

S∆F 1 S–( )F– P– U∇ UT∇+( ) ez⋅[ ] ez⋅+( )z 0== ,

F xd ydΣ∫

Cg1– Cg

1 2/–

f eiωt Aeikx Be ikx–+( ) γ1ABe2ikx γ2e2iωt A2e2ikx B2e 2ikx–+( )+ +=

f +eiωt ikx+ f –eiωt ikx– c.c. f m NRT ,+ + + + +

At vgAx– iαAxx δ id+( )A– i α3 A 2 α4 B 2–( )A iα5µB+ +=

iα6 g z( ) ψzm⟨ ⟩ x zAd

1–

0∫ iα7 f m⟨ ⟩ xA,+ +

Bt vgBx+ iαBxx δ id+( )B– i α3 B 2 α4 A 2–( )B iα5µA+ +=

iα6 g z( ) ψzm⟨ ⟩ x zBd

1–

0∫ iα7 f m⟨ ⟩ xB,+–


(9)

The first seven terms in these equations, accounting for inertia, propagation at thegroup velocity vg, dispersion, damping, detuning, cubic nonlinearity and parametricforcing, are familiar from weakly nonlinear, nearly inviscid theories.7 These theorieslead to the expressions

where ω(k) = [(1 − S + Sk2)kσ]1/2 is the dispersion relation and σ ≡ tanhk, that arerecovered in the present formulation. In particular, the cubic coefficients coincidewith those obtained in strictly inviscid formulations.8–10 The coefficient α3 divergesat (excluded) resonant wave numbers that satisfy ω(2k) = 2ω(k). The detuning d isgiven by

N = integer,

where the last term represents the mismatch between the wavelength 2π/k selectedby the forcing frequency and the domain length L. The last two terms in Equations(7) and (8) describe the coupling to the mean flow in the bulk (be it viscous or invis-cid in origin) in terms of (a local average ⟨⋅⟩x of) the stream function ψm for this flowand the associated free surface elevation fm. The coefficients of these terms and thefunction g are given by

(10)

and are real. The new terms are, therefore, conservative, implying that at leadingorder the mean flow does not extract energy from the system. This result is consis-tent with the small steepness of the associated surface displacement and its smallspeed compared with the speed |∇ψ| due to the surface waves. The mean flow vari-ables in the bulk depend weakly on time but strongly on both x and z, and evolveaccording to the equations

(11)

with

A x L t,+( ) A x t,( ), B x L t,+( ) B x t,( ).≡≡

vg ω′ k( ), α ω″ k( )– 2⁄ , δ α1Cg1 2/ α2Cg,+= = =

α1k ω 2⁄( )1 2/

2ksinh--------------------------, α2 k2 2 1 σ2+

4 ksinh2-------------------+ ,= =

α3ωk2 1 S–( ) 9 σ2–( ) 1 σ2–( ) Sk2 7 σ2–( ) 3 σ2–( )+[ ]

4σ2 1 S–( )σ2 Sk2 3 σ2–( )–[ ]---------------------------------------------------------------------------------------------------------------------------------=

ωk2 8 1 S–( ) 5Sk2+[ ]4 1 S– Sk2+( )

------------------------------------------------------,+

α4ωk2

2--------- 1 S– Sk2+( ) 1 σ2+( )2

1 S– 4Sk2+( )σ2------------------------------------------------------- 4 1 S–( ) 7Sk2+

1 S– Sk2+--------------------------------------+ ,=

α5 ωkσ,=

d α1Cg1 2/ 2πN L 1– k–( )vg,–=

α6 kσ 2ω, α7⁄ ωk 1 σ2–( ) 2σ,⁄= =

g z( ) 2ωk 2k z 1+( )[ ]cosh k ,sinh2⁄=

Ωtm ψz

m A 2 B 2–( )g z( )+[ ]Ωxm– ψx

mΩzm+ Cg Ωxx

m Ωzzm+( ),=

Ωm ψxxm ψzz

m+ ,˙=


(12)

and

(13)

Also, ψm(x + L,z, t) ≡ ψm(x,z, t), fm(x + L, t) ≡ fm(x, t), subject to = 0.Here, β1 = 2ω/σ, β2 = 8ωk2/σ, β3 = (1− σ2)ω2/σ2, β4 = 3(1− σ2)ωk/σ2. Thus, themean flow is forced by the surface waves in two ways. The right sides of the bound-ary conditions (12a) and (12c) provide a normal forcing mechanism; this mechanismis the only one present in strictly inviscid theory9,11 and does not appear unless theaspect ratio is large. The right sides of the boundary conditions (12b) and (13a)describe two shear forcing mechanisms, a tangential stress at the free surface12 anda tangential velocity at the bottom wall.13 Note that neither of these forcing termsvanishes in the limit of small viscosity (i.e., as Cg → 0). The shear nature of theseforcing terms leads us to retain the viscous term in (11a) even when Cg is quite small.In fact, when Cg is very small, the effective Reynolds number of the mean flow isquite large. Thus, the mean flow itself generates additional boundary layers near thetop and bottom of the container, and these must be thicker than the original boundarylayers for the validity of the analysis. This puts an additional restriction on the valid-ity of the equations.5 There is a third, less effective but inviscid, volumetric forcingmechanism associated with the second term in the vorticity equation (11a) that lookslike a horizontal force (|A|2 − |B|2)g(z)Ωm and is sometimes called the vortex force.Although this term vanishes in the absence of mean flow, it can change the stabilityproperties of the flow and enhance or limit the effect of the remaining forcing terms.

In the following we refer to Equations (7)–(9) and (11)–(13) as the general cou-pled amplitude–mean-flow (GCAMF) equations. These equations differ from theexact equations forming the starting point for the analysis in three essential simplif-ications: the fast oscillations associated with the surface waves have been filteredout, the effect of the thin primary viscous boundary layers is replaced by effectiveboundary conditions on the flow in the bulk—viz. (12b) and (13a)—and the surfaceboundary conditions are applied at the unperturbed location of the free surface—viz.z = 0. Thus, only the much broader (secondary) boundary layers associated with the(slowly varying) streaming flow need to be resolved in any numerical simulation.

GRAVITY-CAPILLARY WAVES IN MODERATELY LARGE ASPECT RATIO CONTAINERS

The GCAMF equations describe small amplitude slowly varying wavetrainswhenever the parameters Cg, L–1, and µ are small, but otherwise unrelated to oneanother. Any relation between them will therefore lead to further simplification. Toderive such simplified equations we consider the distinguished limit (see the shadedregion in FIGURE 1A).

ψxm f t

m– β1 B 2 A 2–( )x, ψzzm β2 A 2 B 2–( ),= =

1 S–( ) f xm S f xxx

m– ψztm– Cg ψzzz

m 3ψxxzm+( )+ β3 A 2 B 2+( )x at z 0=–=

ψzm β4 iABe2ikx c.c. B 2 A 2–+ +[ ],–=

Ωzm xd

0

L∫ ψm 0 at z 1.–= = =

f m x t,( ) xd0L∫


(14)

with k = O(1) and |lnCg| = O(1). The simplified equations will then be formally validfor 1 << L << if k ∼ 1. These are derived under the assumption 1 − S ∼ 1 usinga multiple scale method with x and t as fast variables and

(15)

as slow variables. In terms of these variables the local horizontal average ⟨⋅⟩x

becomes an average over the fast variable x. Note that assumption (14) imposes animplicit relation between L and Cg. When 1 − S ∼ 1 the nearly inviscid and viscousmean flows can be clearly distinguished from one another and the viscous mean flowcan be identified by taking appropriate averages of the entire mean flow over theintermediate time scale τ, that is, the mean flow variables ψm, Ωm, and fm take theform

δL2

α--------- ∆ 1, dL2

α---------∼ D 1, µL2

α--------- M 1,∼≡∼= =

Cg1 2/–

ζ xL---,≡ τ t

L---, T t

L2------≡ ≡

FIGURE 1. (A) Sketch of the primary resonance tongue in the scaling regime of inter-est. (B) Bifurcation diagrams for the resulting standing waves near onset.


(16)

with integrals over τ of , , , Ωi, and f i required to be bounded as τ → ∞.Thus, the nearly inviscid mean flow is purely oscillatory on the time scale τ. Sinceits frequency is of the order of L–1 (see Eq. (15)), which is large compared with Cg,the inertial term for this flow is large in comparison with the viscous terms (seeEq. (11)), except in two secondary boundary layers, of thickness of the order of(CgL)1/2 (<< 1), attached to the bottom plate and the free surface. Note that, asrequired for the consistency of the analysis, these boundary layers are much thickerthan the primary boundary layers associated with the surface waves (see FIGURE 2),which provide the boundary conditions (12) and (13) for the mean flow. Moreover,the width of these secondary boundary layers remains small as τ → ∞ and (to leadingorder) the vorticity of this nearly inviscid mean flow remains confined to theseboundary layers. Note that without the requirement that the inviscid flow be purelyoscillatory, the vorticity would diffuse out of these boundary layers and affect thestructure of the whole nearly inviscid solution even at leading order. In fact, vorticitydoes diffuse (and is convected) away from the boundary layers, but this vorticitytransport is included in the viscous mean flow. The vorticity associated with thenearly inviscid mean flow is at most of the order of

(17)

in the upper and lower secondary boundary layers, respectively; the jump in theassociated stream function ψi across each boundary layer is O(CgL) times smaller,and affects only higher order terms; as a consequence the secondary boundary layers

ψm x z ζ τ T, , , ,( ) ψv x z ζ T, , ,( ) ψi x z ζ τ T, , , ,( ),+=

Ωm x z ζ τ T, , , ,( ) Ωv x z ζ T, , ,( ) Ωi x z ζ τ T, , , ,( ),+=

f m x ζ τ T, , ,( ) f v x ζ T, ,( ) f i x ζ τ T, , ,( ),+=ψx

i ψζi ψz

i

A 2 B 2– and A 2 B 2+( ) CgL( ) 1 2/–

FIGURE 2. Sketch of the primary and secondary boundary layers, indicating theirwidths in comparison with the layer depth.


can be completely ignored and no additional contributions to the boundary condi-tions on the nearly inviscid flow need be included in (12) and (13). Outside of theseboundary layers, the complex amplitudes and the flow variables associated with thenearly inviscid mean flow are expanded as

(18)

Substitution of (14)–(16) and (18) into (7)–(13) leads to the following:

(1) From (11)–(13), at leading order,

together with = 0. Thus,

(19)

At second order, the boundary conditions (12a) and (12c) yield

at z = 0. Since the right hand sides of these two equations are independent of the fastvariable x and both and must be bounded in x, it follows that

(20)

where

(21)

is the phase velocity of long wavelength surface gravity waves. Equations (20) mustbe integrated with the following additional conditions

(22)

and the requirements that integrals over τ of and remain bounded as τ → ∞,with = 0.

(2) The leading order contributions to Equations (7) and (8) yield

Thus,(23)

where ξ and η are the characteristic variables(24)

Moreover, according to (9),

A B,( ) L 1– A0 B0,( ) L 2– A1 B1,( ) … ,+ +=

ψv Ωv,( ) L 2– φ0v W0

v,( ) … , f v+ L 3– F0v … ,+= =

ψi f i,( ) L 2– φ0i F0

i,( ) L 3– φ1i F1

i,( ) … ,+ +=

Ωi L 3– W0i … .+=

φ0xxi φ0zz

i+ 0 in 1– z 0,< <=

φ0i 0 at z 1, φ0x

i– 0 at z 0,= = = =

F0xi

φ0i z 1+( )Φ0

i ζ τ T, ,( ), F0i F0

i ζ τ T, ,( )˙ .= =

φ1xi x 0 ζ τ T, , , ,( ) F0τ

i Φ0ζi– β1 B0

2 A02–( )ζ+=

1 S–( )F1xi SF1xxx

i– Φ0τi 1 S–( )F0ζ

i– β3 A02 B0

2+( )ζ–=

φ1i F1

i

Φ0ζi F0τ

i– β1 B02 A0

2–( )ζ,=

Φ0τi vp

2 F0ζi– β3 A0

2 B02+( )ζ,=

vp 1 S–( )1 2/=

Φ0i ζ 1+ τ T, ,( ) Φ0

i ζ τ T, ,( ), F0i ζ 1+ τ T, ,( ) F0

i ζ τ T, ,( ),≡≡

Φ0ζi F0

i

F0i ζd

01∫

A0τ vgA0ζ– B0τ vgB0ζ+ 0.= =

A0 A0 ξ T,( ), B0 B0 η T,( ),= =

ξ ζ vgτ, η+ ζ vgτ.–= =


(25)

Substitution of these expressions into (20) followed by integration of the resultingequations yields

(26)

(27)

where ⟨⋅⟩ζ denotes the mean value in the slow spatial variable ζ, that is,

(28)

and the functions F± are such that

(29)

The particular solution of (26) and (27) yields the usual inviscid mean flow includedin nearly inviscid theories;9 the averaged terms are a consequence of volumeconservation9 and the requirement that the nearly inviscid mean flow has a zeromean on the time scale τ; the latter condition is never imposed in strictly inviscidtheories but is essential in the limit we are considering, as explained above. To avoidthe breakdown of the solution (26) and (27) at vp = vg we assume that

(30)

The functions F± remain undetermined at this stage. In fact, they are not neededbelow because the evolution of both the viscous mean flow and of the complexamplitudes is decoupled from these functions. However, at next order one finds thatF± remain constant on the time scale T, but decay exponentially due to viscouseffects (resulting from viscous dissipation in the secondary boundary layer attachedto the bottom plate) on the time scale t ∼ (L/Cg)1/2.

(3) The evolution equations for A0 and B0 on the time scale T are readily obtainedfrom Equations (7)–(9), invoking (14), (16), (26), (27), (29), and eliminating secularterms (i.e., requiring |A1| and |B1| to be bounded on the time scale τ):

(31)

subject to (25). Here ξ and η are the comoving variables defined in (24), and ⟨⋅⟩x,⟨⋅⟩ζ, ⟨⋅⟩ξ, and ⟨⋅⟩η denote mean values over the variables x, ζ, ξ, and η, respectively.Note that ζ averages over functions of A0 are equivalent to ξ averages, whereas thoseover functions of B0 are equivalent to η averages.

The real coefficient α8 is given by

A0 ξ 1+ T,( ) A0 ξ T,( ), B0 η 1+ T,( ) B0 η T,( ).≡≡

Φ0i

β1vp2 β3vg+

vg2 vp

2–------------------------------ A0

2 B02– A0

2 B02–⟨ ⟩ζ–[ ]=

vp F+ ζ vpτ+ T,( ) F– ζ vpτ– T,( )–[ ],+

F0i

β1vg β3+

vg2 vp

2–----------------------- A0

2 B02 A0

2 B02+⟨ ⟩ζ–+[ ]=

F+ ζ vpτ+ T,( ) F– ζ vpτ– T,( ),+ +

G⟨ ⟩ζ G ζ,d0

1∫=

F± ζ 1 vpτ±+ T,( ) F± ζ vpτ± T,( ), F±⟨ ⟩ζ≡ 0.=

vp vg– 1.∼

A0T iα A0ξξ ∆ iD+( ) A0– i α3 α8+( ) A02 α8 A0

2⟨ ⟩ ξ– α4 B02⟨ ⟩ η–[ ] A0+=

iα5M B0⟨ ⟩ η iα6 g z( ) φ0zv⟨ ⟩ x⟨ ⟩ ζ z A0,d

1–

0∫+ +

B0T iαB0ηη ∆ iD+( )B0– i α3 α8+( ) B02 α8 B0

2⟨ ⟩ η– α4 A02⟨ ⟩ ξ–[ ]B0+=

iα5M A0⟨ ⟩ ξ iα6 g z( ) φ0zv⟨ ⟩ x⟨ ⟩ ζ zB0,d

1–

0∫–+


(32)

Equations (31) are independent of F± because of the second condition in (29).When ∆ > 0 Equations (31) can be used to show that

as

and the result used to simplify the equations for the viscous mean flow in the longtime limit:

(33)

(34)

(35)

(36)

where

(37)

is the effective Reynolds number associated with this flow.Several remarks about these equations and boundary conditions are now in order.

1. The viscous mean flow is driven by the short gravity-capillary waves through theinhomogeneous term in the boundary condition (35c). Since depends onboth ζ and T (unless either A0 or B0 is spatially uniform) the boundary conditionimplies that (and hence ) depends on both the fast and slow horizontal spa-tial variables x and ζ. This dependence cannot be obtained in closed form, and onemust, therefore, resort to numerical computations for realistically large values of L.2. Higher order oscillatory terms omitted from the boundary condition (35c) oscil-late on the intermediate time scale τ, and hence generate secondary boundarylayers. However, the contributions from these boundary layers are all subdominantand have no effect on the streaming flow at leading order. Moreover, the free-surface deflection accompanying the viscous mean flow is also small, f v ∼ L–3 (seeEq. (18)), and so plays no role in the evolution of this flow, as expected of a flowinvolving the excitation of viscous modes.3. The dominant forcing of the viscous mean flow comes from the lower boundary.This forcing vanishes exponentially when k >> 1 leaving only a narrow rangeof wave numbers within which such a mean flow is forced when δ = O(Cg).5 Thus,in most cases in which a viscous mean flow is present, one may assume thatδ = O(Cg

1/2). Note, however, that in fully three-dimensional situations14 in whichlateral walls are included a viscous mean flow will be present even when k >> 1because the forcing of the mean flow in the oscillatory boundary layers along thesewalls remains.4. According to the scaling (14) and the definition of δ the effective Reynoldsnumber Re is large, and ranges from logarithmically large values if k ∼ |lnCg| toO(Cg

–1/2) if k ∼ 1. However, even in the latter limit we must retain the viscous terms

α8

α6 2ω σ⁄( ) β1vp2 β3vg+( ) α7 β1vg β3+( )+

vg2 vp

2–--------------------------------------------------------------------------------------------------------.=

A02 B0

2–⟨ ⟩ τ A02⟨ ⟩ξ B0

2⟨ ⟩η– 0→= T ∞→

W0Tv φ0z

v W0xv– φ0x

v W0zv+ Re 1– W0xx

v W0zzv+( ),=

φ0xxv φ0zz

v+ W0v in 1 z 0,< <–=

φ0xv φ0zz

v 0 at z 0,= = =

W0zv⟨ ⟩ x⟨ ⟩ζ φ0

v 0, φ0zv β4 i A0B0⟨ ⟩ τe2ikx c.c.+[ ]– at z 1,–= = = =

φ0v x L+ ζ 1+ z T, , ,( ) φ0

v x ζ z T, , ,( ),≡

Re CgL2( ) 1–=

A0B0⟨ ⟩ τ

φ0v W0

v


in (33a) in order to account for the boundary conditions (34b) and (35c). Of course,if Re >> 1 vorticity diffusion is likely to be confined to thin layers, but the structureand location of all these layers cannot be anticipated in any obvious way, and onemust again rely on numerical computations.

5. Note that the change of variables

(38)

where

(39)

reduces Equations (31) to the much simpler form

(40)

from which the mean flow is absent. This decoupling is a special property of theregime defined by Equation (14), but is not unique to it, as discussed further below.The resulting equations provide perhaps the simplest description of the Faradaysystem at large aspect ratio, and it is for this reason that they have been extensivelystudied.15 FIGURE 3A shows a typical bifurcation diagram obtained from (40) whenD = 0; FIGURE 3B shows the corresponding results obtained for standing waves, thatis, imposing the requirement that = (= , say) identically. FIGURE 1Bsketches the consequence of including a small but nonzero detuning. These diagramsshow the L2 norms of , , or (after transients have died out) at successiveintersections of a trajectory with the hypersurfaces

and

respectively. A finite number of points therefore indicates a steady or a periodicsolution, whereas scattered points indicate a chaotic trajectory. The origin of suchamplitude chaos is discussed elsewhere.16 In both cases the primary instability isto spatially uniform standing waves (shown by a dashed line when unstable),followed by a supercritical secondary bifurcation to a pattern of spatially non-uniform standing waves, and then a Hopf bifurcation. However, subsequent bifurca-tions differ as instabilities that break the equality = set in, producing stablewaves that are not standing. As discussed elsewhere15 instabilities that break theequality = are also responsible for introducing a type of drift into the dynam-ics, but the origin of this drift is entirely different from that discussed below, beinga drift in the temporal phase of the oscillations in contrast to a drift in the spatialphase. The latter drift is described by the decoupled phase-mean flow equations,(33)–(36) and (39) that can exhibit complex dynamics in their own right (see below);this drift can be present regardless of whether = or not.

A0 A0e ikθ– , B0 B0eikθ,= =

θ′ T( ) α6k 1– g z( ) φ0zv⟨ ⟩ x⟨ ⟩ζ z,d

1–

0∫–=

A0T iα A0ξξ ∆ iD+( ) A0–=

i α3 α8+( ) A02 α4 α8+( ) B0

2⟨ ⟩ η–[ ] A0 iα5M B0⟨ ⟩ η,+ +

B0T iαB0ηη ∆ iD+( )B0–=

i α3 α8+( ) B02 α4 α8+( ) A0

2⟨ ⟩ ξ–[ ]B0 iα5M A0⟨ ⟩ ξ,+ +

A0 ξ 1+ T,( ) A0 ξ T,( ), B0 η 1+ T,( ) B0 η T,( ),≡≡

A0 B0 C0

A0 B0 C0

A0 L22 B0 L2

2+ µ A0⟨ ⟩ B0⟨ ⟩ c.c.+( ),= C0 L22 1

2---µ C0⟨ ⟩2

c.c.+( ),=

A0 B0

A0 B0

A0 B0


FIGURE 3. Bifurcation diagrams for Equations (40) showing successive maximaof the L2 norm as a function of µ ≡ α5M when ∆ = 1, D = 0, α = 0.1, α3 + α8 = 1.5, andα4 + α8 = 0.5. After Martel et al.15


SMALL DOMAINS: L ~ 1

The theory described above simplifies substantially in small domains, because ofthe absence of the slow spatial scale ζ and the advection time scale τ. Although thebasic set up of the calculation is now quite different, the results are (almost) identicalto those just described. This time17

(41)

with A and B spatially constant and the coefficients given by expressions that areidentical to those in (7) and (8). In these equations t denotes a slow time whose mag-nitude is determined by the damping δ > 0 and the detuning d, both assumed to beof the same order as the forcing amplitude µ; in the long time limit |A|2 = |B|2 = R2.It follows that the mean flow (uv(x,z, t),wv(x,z, t)) is now entirely viscous in origin,and obeys a two-dimensional Navier–Stokes equation of the form (33). If we absorbthe standing wave amplitude R (and some other constants) in the definition of theReynolds number

(42)

this equation is to be solved subject to the boundary conditions

(43)

Using the dimensional values of the amplitude of the waves Ad, frequency ωd, wavenumber kd, kinematic viscosity ν, and container depth h, the Reynolds number (42)of the streaming flow can also be written as

Because the structure of Equations (41) is identical to that of Equations (31) thechange of variables (38) leads to a decoupling of the amplitudes from the spatialphase θ, which now satisfies the equation

(44)

FIGURE 4 summarizes the solutions of this coupled phase-streaming flow problem asa function of the wave amplitude R, showing the maxima and minima in the driftspeed θt as a function of the Reynolds number (42) for k = 2.37 and L ≡ 2π/k = 2.65.The primary solution that sets in at µ = µc consists of stable standing waves with|A|2 = |B|2 = R2, defined up to a constant spatial phase θ. The streamlines of the asso-ciated streaming flow are shown in FIGURE 5A for Re = 260, and take the form of anarray of reflection-symmetric counter-rotating eddies with spatial period L/2. Suchreflection-symmetric states do not drift: θt ≡ 0. At Re = Re1 ≈ 270 this reflection-symmetric state loses stability at a (secondary) symmetry-breaking Hopf bifurcationthat produces direction-reversing waves.18 In this state the standing waves drift peri-odically, first to the left and then to the right, with no net displacement, but the

At δ id+( )– A i α3 A 2 α4 B 2–( ) A iα5µB iα6L 1– g z( )uv xd zA ,d0

L∫1–

0∫–+ +=

Bt δ id+( )B– i α3 B 2 α4 A 2–( )B iα5µ A iα6L 1– g z( )uv xd zB ,d0

L∫1–

0∫+ + +=

Re 2β4R2 Cg⁄ ,≡

uv 2k x θ–( )[ ]sin– , wv 0 at z 1,–= = =

uzv 0, wv 0 at z 0.= = =

Reωd Ad

2

ν--------------

6kdh

kdh( )sinh2---------------------------.=

θt

α6

kL------ g z( )uv x z t, ,( ) xd z.d

0

L∫1–

0∫=


FIGURE 4. Bifurcation diagram for the coupled phase–streaming flow equationsshowing the drift speed dθ/dt as a function of the Reynolds number (42) for k = 2.37(L = 2.65). The secondary Hopf and parity-breaking bifurcations occur at Re ≈ 270 andRe ≈ 800, respectively.

FIGURE 5. Streamlines of the streaming flow when (A) Re = 260, (B) Re = 325, and(C) Re = 850. Vertical lines indicate the location of the nodes of the standing wave.


spatial period L/2 of the standing waves is preserved. This bifurcation is supercriticaland the direction-reversing waves are, therefore, stable.

The onset of this instability is insensitive to the domain length L once this is largeenough. These oscillations lose stability at Re ≈ 291.5 to a new family of stabledirection-reversing waves, this time with spatial period L. These new solutions inturn become unstable at Re ≈ 466 where a subcritical bifurcation takes the system toa new branch of steadily drifting solutions. These solutions drift either to the left orto the right, and are stable in the interval 323 ≈ Re3 < Re < Re2 ≈ 800. In FIGURE 4only the waves drifting to the right (θt > 0) are shown; the corresponding streamingflow streamlines are shown (in the comoving frame) in FIGURE 5B. The bifurcationat Re = Re2 is a parity-breaking steady state bifurcation; at this bifurcation the driftceases, and a reflection-symmetric streaming flow is restored, but now with spatialperiod L (FIG. 5C). This flow is stable, but in FIGURE 4 is projected on top of theunstable original symmetric state with period L/2. For larger L (e.g., L = 4π/k) sev-eral different stable states coexist, some of which lack reflection symmetry, andexhibit complex time-dependence.17 Note that neither type of drift is associated withthe loss of reflection-symmetry of the waves themselves, at least at leading order;instead they are due to a symmetry-breaking instability of the associated streamingflow.

EXCITATION OF STREAMING FLOWS IN NEARLY CIRCULAR DOMAINS

The results just described are characteristic of domains with periodic boundaryconditions, such as circular domains. In such domains the amplitudes decouple fromthe spatial phase and the streaming flow only causes drift motion in an otherwisestanding wave pattern. However, the instabilities found set in only at sufficientlylarge Reynolds number, necessitating a numerical solution of the coupled phase-streaming flow problem. There is a simple way, however, of pushing these instabili-ties to much smaller amplitudes and, hence, into a regime where analytic progress ispossible. This occurs as soon as the translation (rotation) invariance of the containeris broken (eg., by deforming it slightly), thereby forcing the wave amplitudes to cou-ple to the spatial phase. In systems undergoing a symmetry-breaking primary Hopfbifurcation (but with no parametric forcing) this type of forced symmetry-breakingis known to lead to rich dynamics,19–21 and the present system is no exception.To this end we note that the dominant conservative term that breaks the symmetry(A,B) → (e–iφA,eiφB) arising from rotation invariance of the system takes the formiΛ(B,A), where Λ is a real coefficient; the addition of this term to the equationsbreaks rotational invariance, but preserves a plane of reflection symmetry. In the fol-lowing we assume that Λ ∼ δ ∼ d ∼ µ, and project the streaming flow onto the firsthydrodynamic mode. This assumption requires that the effective Reynolds number(42) of the streaming flow be sufficiently small that the nonlinear terms in (33) maybe neglected. In the following we denote the (real) amplitude of this mode by v1. Weobtain22

(45)v1′ τ( ) ε v1– B 2 A 2–+( ),=


(46)

(47)

where ε = −λ1Re–1 > 0 and λ1 < 0 is the first hydrodynamic eigenvalue. These equa-tions illustrate well the interplay between breaking of rotation invariance (Λ ≠ 0) andthe excitation of streaming flow (γ ≠ 0). When Λ = γ = 0 they reduce to the exact, butdecoupled, amplitude equations for A and B that follow from Equations (41) onusing (38).

Equations (45)–(47) have solutions in the form of reflection-symmetric standingwaves (A,B) ≡ (C,C), v1 = 0, where

(48)

this state sets in at µ = µc, where

(49)

The amplitude R ≡ |C| increases monotonically for µ > µc provided (d − Λ)/(α3 − α4)≤ 0; if (d − Λ)/(α3 − α4) > 0 the branch bifurcates subcritically at µ = µc before turn-ing around towards larger µ at a secondary saddle-node bifurcation.

To determine the linear stability of these states we replace A, B, and v1 by C +X± + and Z + c.c., respectively, and linearize. The resulting equationshave even eigenmodes (X+ = X−, Y+ = Y−, Z = 0) and odd eigenmodes (X+ = −X−,Y+ = −Y−, Z ≠ 0). The former preserve the reflection symmetry of C and are charac-terized by the dispersion relation

(50)

Such instabilities are, therefore, always nonoscillatory and correspond either to theprimary bifurcation at µ = µc or to the secondary saddle-node bifurcation at

(51)

In contrast, the symmetry-breaking perturbations satisfy a cubic dispersion relation,

(52)

with a steady state bifurcation when

(53)

and a Hopf bifurcation when

(54)

The oscillations that result are invariant under reflection followed by evolution intime by half the oscillation period 2π/|λI|, where

(55)

and are the analogues of the direction-reversing waves discussed in the precedingsection. Such symmetric oscillations can, therefore, be present only if

At δ– id– i α3 A 2 α4 B 2–( )+[ ]A iα5µB iΛB iγ v1A,–+ +=

Bt δ– id– i α3 B 2 α4 A 2–( )+[ ]B iα5µA iΛA iγ v1B,+ + +=

δ2 d Λ– α3 α4–( ) C 2–[ ]2+ α52µ2, α3 α4 0;≠–=

µcδ2 d Λ–( )2+[ ]1 2/

α5---------------------------------------------.≡

eλt Y ± eλt eλt

λ δ+( )2 d Λ– 2 α3 α4–( ) C 2–[ ]2+ δ2 d Λ–( )2.+=

C 2 d Λ–α3 α4–------------------.=

λ2 2δλ 4dΛ 8Λ γελ ε+------------ α3+ C 2–+ + 0,=

C 2 d2 α3 γ+( )-----------------------, if dΛ 0,≠=

C 2 4Λd 2δε ε2+ +4Λ 2α3 δ 1– εγ–( )------------------------------------------ 0.>=

λI2 ε2– 4δ 1– εγΛ C 2 0,>–=


(56)and so cannot occur near µ = µc without both the streaming flow and the breaking ofthe rotation invariance of the system. This is a consequence of the low Reynoldsnumber assumption used to model the streaming flow by a single mode. Recall thatFIGure 4 shows that if Re is sufficiently large a symmetry-breaking Hopf bifurcationcan destabilize the standing waves even if Λ = 0.

From Equations (51), (53), and (54) we also find conditions for codimension-twodegeneracies:1. A Takens–Bogdanov bifurcation, corresponding to the coalescence of the sym-metry-breaking Hopf and steady bifurcations, occurs if (56) holds and

(57)

2. An interaction between the saddle-node and symmetry-breaking steady bifurca-tion occurs when

(58)

3. An interaction between the saddle-node and symmetry-breaking Hopf bifurca-tion, with one zero plus two nonzero imaginary eigenvalues, occurs when (55)holds and

(59)

The first of these bifurcations contains within its unfolding periodic oscillationsthat are symmetric in the sense defined above. The second also contains periodicoscillations but this time the oscillations are asymmetric. Finally, the third bifurca-tion contains quasiperiodic solutions23 that are symmetric on average; these corre-spond to three-frequency states in the Faraday system. Chaotic dynamics are presentnear global bifurcations with which the quasiperiodic solutions terminate.24,25

Closely related results apply to containers subjected to horizontal vibration aswell.26

CONCLUDING REMARKS

In this paper we have reexamined the theory of parametrically excited surfacegravity-capillary waves in nearly inviscid liquids; the corresponding microgravityresults are obtained on setting the Bond number B to zero (Cg = C, S = 1). We focusedon brimful containers with a pinned contact line and restricted attention to two-dimensional systems with periodic boundary conditions. In systems of this type theprimary instability is always to a pattern of standing waves—this is a very generalresult.27 In the present case, we have found that these standing waves may losestability at finite amplitude to two different types of instabilities that break the reflec-tion symmetry of these waves: a Hopf bifurcation producing standing wavesthat drift back and forth, and a parity-breaking bifurcation that produces waves thatdrift with a constant speed in one direction or other. We have shown that these insta-bilities involve the excitation of a viscous mean flow we called streaming flow. Thisflow is driven in the boundary layers at the container walls or at the free surface bytime-averaged Reynolds stresses produced by the waves, and in turn couples to the

γ Λ 0,<

γ α3+( )ε 2δ 1– dγΛ+ 0;=

d α3 α4 2γ+ +( ) 2Λ α3 γ+( );=

4δΛ 2δε ε2+ +4Λ 2α3 δ 1– εγ–( )------------------------------------------ d Λ–

α3 α4–------------------.=


amplitudes of the waves responsible for it. This coupling arises already at third orderindicating that it is in general inconsistent to ignore the streaming flow while retain-ing other cubic terms. We have shown that for both extended (L >> 1) and small(L ∼ 1) domains the general coupled amplitude-mean flow (GCAMF) equationsdecouple into equations for the wave amplitudes, and a set of equations for thespatial phase of the waves coupled to the Navier–Stokes equation for the streamingflow. In an extended domain the dynamics in the decoupled amplitude equationscan be complex, and could break, at large enough forcing, the instantaneous equality|A0| = |B0| characteristic of standing waves (FIG. 3); in a small domain this is not pos-sible and standing waves persist for all time. However, in both cases the coupling tothe streaming flow could destabilize the spatial phase of the standing waves resultingin complex drift motion. Our calculations suggest that the uniformly drifting stand-ing waves observed in some experiments14 could form as a result of a parity-break-ing bifurcation of the type described here. We have presented a concrete example ofthe two types of drift instability that can occur, and described briefly the interactionbetween the excitation of the streaming flow and forced breaking of the translationinvariance of the (annular) container that suggests a variety of new experiments onthe Faraday system. These conclusions readily generalize to containers of othershapes, such as square containers perturbed to rectangular ones.22

ACKNOWLEDGMENTS

We are grateful to María Higuera and Elena Martín for assistance in obtaining theresults reported here, and to the NASA Microgravity Program for support undergrant NAG3-2152.

REFERENCES

1. MILES, J. & D. HENDERSON. 1990. Parametrically forced surface waves. Annu. Rev.Fluid Mech. 22: 143–165.

2. FAUVE, S. 1995. Parametric instabilities. In Dynamics of Nonlinear and DisorderedSystems. G. Martínez Mekler & T.H. Seligman, Eds.: 67–115. World Scientific.

3. KUDROLLI, A. & J.P. GOLLUB. 1997. Patterns and spatio-temporal chaos in parametri-cally forced surface waves: a systematic survey at large aspect ratio. Physica D 97:133–154.

4. MARTEL, C. & E. KNOBLOCH. 1997. Damping of nearly inviscid water waves. Phys.Rev. E 56: 5544–5548.

5. VEGA, J.M., E. KNOBLOCH & C. MARTEL. 2001. Nearly inviscid Faraday waves inannular containers of moderately large aspect ratio. Physica D 154: 313–336.

6. LAMB, H. 1932. Hydrodynamics. Cambridge University Press.7. EZERSKII, A.B., M.I. RABINOVICH, V.P. REUTOV & I.M. STAROBINETS. 1986. Spatiotem-

poral chaos in the parametric excitation of a capillary ripple. Sov. Phys. JETP 64:1228–1236.

8. MILES, J. 1993. On Faraday waves. J. Fluid Mech. 248: 671–683.9. PIERCE, R.D. & E. KNOBLOCH. 1994. On the modulational stability of traveling and

standing water waves. Phys. Fluids 6: 1177–1190.10. HANSEN, P.L. & P. ALSTROM. 1997. Perturbation theory of parametrically driven capil-

lary waves at low viscosity. J. Fluid Mech. 351: 301–344.11. DAVEY, A. & K. STEWARTSON. 1974. On three-dimensional packets of surface waves.

Proc. R. Soc. London, Ser. A 338: 101–110.


12. LONGUET-HIGGINS, M.S. 1953. Mass transport in water waves. Phil. Trans. R. Soc. Ser.A 245: 535–581.

13. SCHLICHTING, H. 1932. Berechnung ebener periodischer Grenzschichtströmungen.Phys. Z. 33: 327–335.

14. DOUADY, S., S. FAUVE & O. THUAL. 1989. Oscillatory phase modulation of parametri-cally forced surface waves. Europhys. Lett. 10: 309–315.

15. MARTEL, C., E. KNOBLOCH & J.M. VEGA. 2000. Dynamics of counterpropagatingwaves in parametrically forced systems. Physica D 137: 94–123.

16. HIGUERA, M., J. PORTER & E. KNOBLOCH. 2002. Heteroclinic dynamics in the nonlocalparametrically driven Schrödinger equation. Physica D. 162: 155–187.

17. MARTÍN, E., C. MARTEL & J.M. VEGA. 2002. Drift instability of standing Faradaywaves. J. Fluid Mech. In press.

18. LANDSBERG, A.S. & E. KNOBLOCH. 1991. Direction-reversing traveling waves. Phys.Lett. A 159: 17–20.

19. DANGELMAYR, G. & E. KNOBLOCH. 1991. Hopf bifurcation with broken circular sym-metry. Nonlinearity 4: 399–427.

20. DANGELMAYR, G., E. KNOBLOCH & M. WEGELIN. 1991. Dynamics of travelling wavesin finite containers. Europhys. Lett. 16: 723–729.

21. HIRSCHBERG, P. & E. KNOBLOCH. 1996. Complex dynamics in the Hopf bifurcationwith broken translation symmetry. Physica D 90: 56–78.

22. HIGUERA, M., J.M. VEGA & E. KNOBLOCH. 2002. Coupled amplitude-streamingflow equations for nearly inviscid Faraday waves in small aspect ratio containers. J.Nonlinear Sci. In press.

23. GUCKENHEIMER, J. & P. HOLMES. 1983. Nonlinear Oscillations, Dynamical Systemsand Bifurcations of Vector Fields. Springer-Verlag.

24. LANGFORD, W.F. 1983. A review of interactions of Hopf and steady-state bifurcations.In Nonlinear Dynamics and Turbulence. G.I. Barenblatt, G. Iooss & D.D. Joseph,Eds.: 215–237. Pitman.

25. KIRK, V. 1993. Merging of resonance tongues. Physica D 66: 267–281.26. HIGUERA, M., J.A. NICOLÁS & J.M. VEGA. 2002. Weakly nonlinear oscillations of non-

axisymmetric capillary bridges at small viscosity. Phys. Fluids. In press.27. RIECKE, H., J.D. CRAWFORD & E. KNOBLOCH. 1988. Time-modulated oscillatory con-

vection. Phys. Rev. Lett. 61: 1942–1945.

Documents

Edgar Knobloch Reprint List - Coupled Mean Flow …tardis.berkeley.edu/reprints/papers/KnMaVe_NYACADSCI974...Ann. N.Y. Acad. Sci. 974: 201–219 (2002). ©2002 New York Academy of