Chemin J.-Y., Desjardins B., Gallagher I., Grenier E. Mathematical geophysics

Подождите немного. Документ загружается.

Review of physical phenomena 9

of the boundary layer), and by u =0atx

= 0 (Dirichlet boundary condi-

tion). Solving that diﬀerential equation is then straightforward and provides the

classical expression for the tangential velocity,

tan

(t, x



Id −exp



−

√

2εν





−

√

2εν



∞

(t, x

)

where R(ζ) denotes the rotation of angle ζ. This boundary layer is called the

Ekman layer, or the Ekman spiral taking into account the shape of u

tan

(t, x

Now to enforce incompressibility we must have ∂

= −∂

− ∂

, which

leads, together with u

= 0 for x

=0,to

(t, x



εν

curl

∞

(t, x



−

√

2εν



with f(ζ)=−1/2e

−ζ

(sin ζ + cos ζ). As a consequence, u

does not go to 0 as x

tends to inﬁnity: if the velocity in the interior of the domain is not constant,

a small amount of the ﬂuid, of order

√

εν, enters the domain (or the boundary

layer, depending on the sign of the two-dimensional curl). This phenomenon is

called Ekman suction, and u

(t, x

, ∞) is called the Ekman suction velocity or

Ekman transpiration velocity. This velocity is responsible for global circulation

in the whole domain, of order

√

εν, but not limited to the boundary layer, a ﬁrst

three-dimensional eﬀect in the interior of the domain.

This small velocity has a very important eﬀect in the energy balance. Namely,

let us compute the energy dissipation in the Ekman layer, that is let us eval-

uate the order of magnitude of ν



|∇u|

in the layer. The gradient |∇u| is of

order λ

−1

∼

√

εν

−1

, therefore ν



|∇u|

scales as

√

εν

−2

√

εν ∼



the last term

√

εν corresponding to the volume of integration. In the most inter-

esting case ν ∼ ε, the dissipation of energy per unit of time is of order one:

despite its smallness, the Ekman layer dissipates a signiﬁcant amount of energy,

and cannot be neglected in the energy balance. In other words, the Ekman layer

damps the interior motion, like an order-one friction term. This phenomenon

is called Ekman pumping. Note that if ν remains constant while ε goesto0,

then the dissipation is inﬁnite! In fact the ﬂuid is immediately stopped by its

boundary layers. This remains true if ν ≫ ε when ε tends to 0. On the contrary,

if ν ≪ ε then Ekman pumping is negligible.

Let us compute the Ekman pumping precisely . The dissipation in one layer

equals





|∂

+ |∂



dx where |∂

u| =

∞

√

εν

exp



−

√

2εν



10 General introduction

Hence the dissipation equals

∞

εν

√

εν

√

= |u

∞



2ε

As there are two layers (one for each boundary), the total dissipation equals

√

2β|u

∞

where β = ν/ε. Hence Ekman pumping is a linear dissipation, and

adds the linear term

√

2βu in the Euler or Navier–Stokes equations. The limit

equation on the x

independent ﬁeld is



∂

u + u ·∇u +

√

2βu + ∇p =0

div u =0.

As will be clear in the rigorous derivation of Ekman layers, this friction term is

deeply connected to the Ekman suction velocity. The role of this vertical velocity

is to move the ﬂuid from the interior to the boundary layers, where it slows down

because of the large vertical viscosity. The slow ﬂuid then goes back to the interior

of the domain again with the help of the Ekman suction velocity. This can be

observed very simply, by making a cup of tea turn rapidly with a spoon [67].

Everyday experience shows that once one stops turning the spoon, the tea stops

moving within a dozen seconds. The ﬁrst idea is that the dissipation is due to

viscosity. Let us make a short computation of the associated order of magnitudes:

the typical size of the cup is L = 5 cm, the kinematic viscosity of water is of

the order of 10

−2

·s

−1

, and the rotation is, say Ω = 4π s

−1

. The diﬀusion

time-scale is therefore T

diﬀ

def

= L

/ν ∼ 40 minutes, whereas the suction time

scale is T

suct

def

= L/

√

νΩ = 14 seconds. Tea can be considered as a rotating ﬂuid

and the main eﬀect is in fact Ekman suction which brings tea from the interior

of the cup to boundaries where dissipation is high. The time-scale is then the

time needed for a particle of ﬂuid to cross the cup and reach a boundary. This

is the right time-scale.

Now in the general case Poincar´e waves propagate in the medium. They also

violate in general the Dirichlet boundary condition, and have their own boundary

layers, which are very similar to Ekman boundary layers, except for their slightly

diﬀerent size and their dependence on the frequency. Again those boundary layers

dissipate energy and damp the waves like a linear friction term (which depends on

the frequency and the wavenumber). The detailed computations of the boundary

layers of Poincar´e waves is the object of Part III, Chapter 7.

To summarize, rotating ﬂuids in T

×[0, 1] or R

×]0, 1[ consist of:

• a limit two-dimensional, divergence-free ﬂow;

• horizontal boundary layers at x

= 0 and x

= 1 to match the interior ﬂow

with Dirichlet boundary condition (Ekman layers);

• Poincar´e waves, with high time frequency;

• horizontal boundary layers at x

= 0 and x

= 1 to match Poincar´e waves

with Dirichlet boundary conditions.

Review of physical phenomena 11

Moreover, the limit ﬂow is damped by Ekman pumping and satisﬁes a damped

Euler (or Navier–Stokes) equation. Poincar´e waves are also damped by Ekman

pumping, and in the periodic case satisfy a quadratic damped equation. In the

case R

×[0, 1], Poincar´e waves go to inﬁnity very fast (with speed ε

−1

) and go

to 0 locally in time and space (for t>0). All this will be detailed in Part III,

Chapter 7.

When, instead of e

, the direction of rotation r is not perpendicular to

the boundaries, the boundary layer is still an Ekman layer of size



εν/|r · e

provided r ·e

does not vanish. The situation is, however, diﬀerent if r ·e

=0:

the vertical layers are very diﬀerent and more diﬃcult to analyze. We shall discuss

the layers in a general three-dimensional domain in the last chapter of this book.

Stability of Ekman layers

A general problem in boundary layer theory is to know whether the layer remains

laminar or becomes turbulent, that is, to know whether the characteristics of the

ﬂow vary over lengths of size

√

εν in x

and of size 1 in x

in times of order 1,

or whether small scales also appear in the horizontal directions, or whether the

ﬂow evolves in small time, of order say

√

εν. The answer is not straightforward,

since the velocity, being of order 1 in the layer, a particle may cross it in time

scales of order

√

εν if by chance its velocity is not parallel to the boundary. Hence

naturally, the system could evolve in a signiﬁcant manner in times of order

√

εν.

This would in fact create tangential structures of typical size

√

εν. In other words

there is a priori no reason why the ﬂow in the boundary layer would remain so

anisotropic (the x

-direction being the only direction of high variation), as the

transport term has a natural destabilizing eﬀect.

On the other hand, we can think that in the boundary layer the viscosity is

so important that it suﬃces to stabilize everything and to cancel motions in the

vertical direction.

The answer lies in between these two limit cases, and depends on the ratio

between inertial forces and viscous forces. Let us deﬁne the Reynolds number Re

of the boundary layer as the typical ratio between inertial forces and viscous

forces in the boundary layer. The inertial forces are of order U

−1

where U is

the typical velocity in the layer (|u

∞

| in our case), and the viscous forces are of

order νUλ

−2

. Therefore we can deﬁne Reynolds number by

Re = |u

∞

= |u

∞



If the Reynolds is small, then viscous forces prevail and the ﬂow is expected to be

stable. On the contrary if the Reynolds is large, inertial forces are important and

the ﬂow may be unstable. This is a classical phenomenon in ﬂuid mechanics, a fea-

ture common to many diﬀerent physical cases (boundary layers of viscous ﬂows,

rotating ﬂows, magnetohydrodynamics (MHD)): there exists a critical Reynolds

number Re

such that the ﬂow is stable and the boundary layer remains stable

12 General introduction

provided Re<Re

, and such that the ﬂow is unstable and the boundary becomes

more complicated or even turbulent if Re>Re

. In the case of Ekman layers,

the critical Reynolds number can be computed, and equals approximately 54.

Therefore if Re<Re

∼ 54, the ﬂow of a highly rotating ﬂuid between two

plates splits into an interior two-dimensional ﬂow and a boundary layer ﬂow,

completely laminar.

If Re>Re

, on the contrary, the boundary layer is no longer laminar. If the

Reynolds is not too high (say Re < 120) rolls appear, of size

√

νε×

√

νε×1 and

move with a velocity of order 1. At high Reynolds number (say Re > 120), the

rolls themselves become unstable and are destroyed. The ﬂow near the boundary

is fully chaotic and turbulent. An interesting question is to know whether this

turbulent layer remains near the boundary or goes into the interior of the domain

and destabilizes it. In this latter case, the whole ﬂow would be turbulent (and

of course the Taylor–Proudman theorem would no longer be true).

This leads to a restriction of the size of the limit solution to get convergence.

Note that if the viscosity is anisotropic and if ν

/ν

tends to +∞, all those

problems disappear since the critical Reynolds number is inﬁnite: Ekman layers

are always stable. This is not surprising since anisotropic viscosity accounts for

turbulent behavior. This only says that the model is strong enough and that no

other phenomenon (instability of the boundary layer) appears in this turbulent

model.

Let us end this introduction with the proof that at small Reynolds num-

bers Re<Re

for some Re

<Re

, Ekman layers are linearly stable, by using

energy estimates. Let u

be a pure Ekman layer given by













Id −exp



−

√

2εν





−

√

2εν



∞

(t, x

)



εν

curl

∞

(t, x



√

2εν



and let us linearize the Navier–Stokes Coriolis equations (NSC

) around u

. This

yields, denoting the perturbation w,

(LNSC)











∂

w + u

·∇w + w ·∇u

∧ w

− ν∆w + ∇p =0

div w =0

|∂Ω

=0.

Let us estimate the L

norm of the perturbation w. We have, using the

divergence-free condition and integrating over Ω,



|w|

dx + ν



|∇w|

dx +



w · (w ·∇u

) dx =0.

References 13

The main term of



w · (w ·∇u

) dx turns out to be



w · ∂

dx.As

∂

(·,x

)

∞

(Ω

)

≤u

∞



∞

(Ω

)

√

εν

exp



−

√

2εν



the following estimate holds:

def





w · ∂



≤

u

∞



∞

(Ω

)

√

εν









∂

w(t, x

,z)dz





∂

(t, x

,z)dz



−x

√

2εν



≤

u

∞



∞

(Ω

)

√

εν

∂

w

∂





−z/

√

2εν

≤ C

√

ενu

∞



∞

(Ω

)

∂

u

∂



where C

is a numerical constant. This last term can be absorbed by the viscous

term ν



|∇u|

dx provided C

u

∞



∞

(Ω

)

√

εν ≤ ν, namely provided

Re = u

∞



∞

(Ω

)



≤ Re

Hence (LNSC) is stable provided Re

<Re

. Of course Re

<Re

and the com-

putation of C

leads to Re

∼ 4. There is therefore a large gap between Re

and Re

. Note that this proof uses little of the precise description of u

since it

only uses the exponential decay of ∂

with respect to x

√

2εν. It is therefore

not surprising that the result is not optimal at all. The method, however, can be

used in a wide range of situations, and we refer to [52] for other applications.

References

After each part of this book we will give speciﬁc references related to the topics

treated in that part. Here we wish to present to the reader a (far from exhaustive)

selection of classical physical monographs on geophysical ﬂuids.

• G.K. Batchelor: An introduction to ﬂuid dynamics [9]

• P. Bougeault and R. Sadourny: Dynamique de l’atmosph`ereetdel’oc´ean [11]

• B. Cushman-Roisin: Introduction to geophysical ﬂuid dynamics [40]

• P.G. Drazin, W.H. Reid: Hydrodynamic stability [52]

• M. Ghil and S. Childress: Topics in Geophysical Fluid Dynamics: Atmo-

spheric Dynamics, Dynamo Theory, and Climate Dynamics [63]

• A.E. Gill: Atmosphere–ocean dynamics [65]

• H.P. Greenspan: The theory of rotating ﬂuids [67]

14 General Introduction

• H. Lamb: Hydrodynamics [85]

• R. Lewandowski. Analyse Math´ematique et Oc´eanographie [88]

• C.C. Lin: The Theory of Hydrodynamic Stability [90]

• A. Majda: Introduction to PDEs and Waves for the Atmosphere and

Ocean [94]

• J. Pedlosky: Geophysical Fluid dynamics [103]

• H. Schlichting: Boundary layer theory [111].

PART II

On the Navier–Stokes equations

This part is devoted to the mathematical study of the Navier–Stokes equations

for incompressible homogeneous ﬂuids evolving in a domain (i.e. an open connec-

ted subset) Ω of R

, where d = 2 or 3 denotes the space dimension. For instance,

the space domain may be either the whole space R

, or a domain of R

,ora

three-dimensional unbounded domain such as R

×]0, 1[, which will be of par-

ticular interest in the framework of geophysical ﬂows. Denoting by u ∈ R

the

velocity ﬁeld, p ∈ R the pressure ﬁeld, and ν>0 the kinematic viscosity, the

Cauchy problem can be written as follows:

(NS

)











∂

u + u ·∇u − ν∆u + ∇p = f

div u =0

|t=0

= u

where f is a given bulk force. Note that the density ρ has been chosen equal to

1 for simplicity. It corresponds to the case of the non-dimensionalized Navier–

Stokes equations, in which the viscosity is expressed as the inverse Reynolds

number. The system (NS

) is supplemented with the no-slip boundary condi-

tions u

|∂Ω

= 0. Multiplying (NS

)byu and integrating over Ω formally yields

the well-known energy equality



Ω

|u(t, x)|

dx + ν



Ω

|∇u(t

′

,x)|

dxdt

′



Ω

(x)|

dx +



Ω

f · u(t

′

,x) dxdt

′

It follows that the natural regularity assumptions for the data are u

square

integrable and divergence-free, while f is a vector ﬁeld the components of which

should be in L

loc

; H

−1

(Ω)) where H

−1

(Ω) denotes the set of functions which

are the sum of an L

function and the divergence of an L

vector ﬁeld. This

implies the introduction of a little bit of functional analysis. In particular, the

basic properties of the Stokes operator and Sobolev embeddings are recalled in

Chapter 1, followed by the proof of Leray’s global existence theorem in Chapter 2.

The question of uniqueness and stability is addressed in Chapter 3: the stability of

Leray solutions is proved in two dimensions, and the existence of stable solutions

is proved in three dimensions for bounded domains and for domains without

boundary, namely R

or T

Some elements of functional analysis

Before introducing the concept of Leray’s weak solutions to the incompress-

ible Navier–Stokes equations, classical deﬁnitions of Sobolev spaces are required.

In particular, when it comes to the analysis of the Stokes operator, suitable

functional spaces of incompressible vector ﬁelds have to be deﬁned. Several

issues regarding the associated dual spaces, embedding properties, and the

mathematical way of considering the pressure ﬁeld are also discussed.

1.1 Function spaces

Let us ﬁrst recall the deﬁnition of some functional spaces that we shall use

throughout this book. In the framework of weak solutions of the Navier–

Stokes equations, incompressible vector ﬁelds with ﬁnite viscous dissipation and

the no-slip property on the boundary are considered. Such H

-type spaces of

incompressible vector ﬁelds, and the corresponding dual spaces, are important

ingredients in the analysis of the Stokes operator.

Deﬁnition 1.1 Let Ω be a domain of R

(d =2or 3). The space D(Ω)

is deﬁned as the space of smooth functions compactly supported in the

domain Ω, and D

′

(Ω) as the space of distributions on Ω.

The space H

(Ω) denotes the space of L

functions f on Ω such that ∇f

also belongs to L

(Ω). The Hilbertian norm is deﬁned by

f

(Ω)

def

= f 

(Ω)

+ ∇f

(Ω)

The space H

(Ω) is deﬁned as the closure of D(Ω) for the H

(Ω) norm, and

the space H

−1

(Ω) as the dual space of H

(Ω) for the D

′

×Dduality and we

state

f

−1

(Ω)

def

= sup

ϕ∈H

(Ω)

ϕ

(Ω)

≤1

f,ϕ.

18 Some elements of functional analysis

Let us recall the Poincar´e inequality. When Ω is a bounded domain and f is

in D(Ω), writing

f(x

′



−∞

∂

f(y

′

) dx

and using the Cauchy–Schwarz inequality, we get

|f(x

′

≤ C

Ω



∞

−∞

|∇f(y

′

By integration, it follows that

f

(Ω)

≤ C

Ω

∇f

(Ω)

and by density this also holds for all functions f ∈ H

(Ω). This means that the

norm ·

deﬁned by

f

def

= ∇f

is equivalent to the ·

norm if the domain is bounded.

As we shall be working with divergence-free vector ﬁelds, we have to adapt

the above deﬁnition to that setting.

Deﬁnition 1.2 Let Ω be a domain of R

(d =2or 3). We shall denote

by V(Ω) the space of vector ﬁelds, the components of which belong to H

(Ω),

and by V

(Ω) the space of divergence-free vector ﬁelds in V(Ω). The closure

of V

(Ω) in L

(Ω) will be denoted H(Ω). Finally, we shall denote by V

′

(Ω)

the space of vector ﬁelds with H

−1

(Ω) components.

If E is a subspace of V(Ω), the polar space E

◦

is the space of vector

ﬁelds f in V

′

(Ω) such that for all v ∈ E,

f,v

def



j=1

f



−1

×H

=0.

If F is a subspace of V

′

(Ω), the polar space F

⋆

is the space of vector ﬁelds v

in V(Ω) such that for all f ∈ F, f,v =0.

To make the notation lighter, we shall omit mention of Ω when no confusion is

likely.

Remark The space V

′

(Ω) is a Hilbert space, endowed with the following norm

f

′

(Ω)

= sup

v∈V(Ω)

f,v

= u

(Ω)