Steady compressible Oseen flow with slip boundary conditions

# Steady compressible Oseen flow with slip boundary conditions

Tomasz Piasecki

Mathematical Institute, Polish Academy of Sciences

e-mail: T.Piasecki@impan.gov.pl

Abstract

We prove the existence of solution in a class to steady compressible Oseen system with slip boundary conditions in a two dimensional, convex domain with the boundary of class . The method is to regularize a weak solution obtained via the Galerkin method. The problem of regularization is reduced to a problem of solvability of a certain transport equation by application of the Helmholtz decomposition. The method works under additional assumption on the geometry of the boundary.

MSC: 35Q10, 76N10
Keywords: Compressible Navier-Stokes flow, Slip boundary conditions

## 1 Introduction

In this paper we consider a system of Stokes-type equations describing steady flow of a barotropic, compressible fluid in a two dimensional, convex domain with - boundary, supplied with inhomogeneous slip boundary conditions with nonnegative friction coefficient. The system can be considered as a linearization of a Navier-Stokes system for compressible fluid around a constant flow , thus we will call it compressible Oseen system. The slip boundary conditions involving friction enable to describe the interactions between the fluid and the boundary of the domain. It also turns out that they allow to extract some information on the vorticity of the velocity, that can be used to show that the velocity has higher regularity. Such approach has been applied in  and  to incompressible flows. In this paper we follow these ideas, modifying them in a way that they can be applied to the compressible system. A significant feature of this system is its elliptic-hyperbolic character: the momentum equation is elliptic in the velocity, while the continuity equation is hyperbolic in the density. Therefore we can prescribe the values of the density only on the part of the boundary where the flow enters the domain and a singularity appears in the points where the inflow and outflow parts of the boundary meets.

We show existence of a solution . A method we apply is to regularize a weak solution obtained via the Galerkin method. Analysing the vorticity of the velocity we can show that the density is in fact solution to a certain transport equation, obtained via elimination of the velocity from the continuity equation. The problem of regularization is thus reduced to a problem of solvability of a transport equation. The values of the density are prescribed on the part of the boundary where the flow enters the domain, and the density can be found as a solution to the transport equation via method of characteristics, thus singularities appear in points where the inflow and outflow parts of the boundary coincides. We show that the solvability of this transport equation is relied with the geometry of the boundary near the singularity points, thus we can define classes of domains where our method of regularization can or cannot be applied.

Since similar difficulties resulting from the mixed character of the problem appear in the analysis of steady compressible Navier-Stokes system, it is likely that the results of this paper will turn out useful in future analysis of the nonlinear problem. Now let us formulate the problem more precisely.

 ⎧⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎨⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎩∂x1u−μΔu−(ν+μ)∇divu+γ∇w=FinΩ,divu+∂x1w=GinΩ,n⋅2μD(u)⋅τ+f( u⋅τ)=Bon∂Ω,n⋅u=0on∂Ω,w=0onΓin, (1.1)

where is a bounded, convex domain in with a boundary of class . is the velocity of the fluid and is the density. denotes outward unit normal to . We assume that , and are given functions. and are viscosity constants satisfying and is a friction coefficient (note that if then the conditions (1.1) reduce to a homogeneous Dirichlet condition). The system (1.6) can be considered as a linearization of a steady compressible Navier-Stokes system around a constant flow . More precisely, the perturbed flow satisfies inhomogeneous boundary conditions and , but if we assume that and are regular enough we can reduce the problem to homogeneous boundary conditions (1.1). Thus we distiguish the inflow and outflow parts of the boundary as the parts where the perturbed flow enters and leaves the domain:

 Γin={x:n1(x)<0},Γout={x:n1(x)>0}.

Let us also denote . We assume that consist of two points: and (see Fig. 1). Due to the convexity of we can define functions and for in the following way:

 (x1–––(x2),x2)∈Γin,(¯¯¯¯¯x1(x2),x2)∈Γout

Around and is given as a -function of . We will denote these functions by and respectively (Fig. 2) For convenience we will denote
The main result of this paper is

###### Theorem 1.

Assume that and is large enough. Assume further that the boundary near the singularity points satisfies the following condition

 ∃1

Then the system (1.6) has a unique solution and

 ||u||H2(Ω)+||w||H1(Ω)≤C(DATA). (1.3)

The geometric condition (1.2) may look strange since it is formulated in a general form, but it has a clear meaning. Namely, the boundary near the singularity points can not be too flat, more precisely, our method works if the boundary is less flat than a graph of a function around zero for some . We also show (lemma 13 (b)) that the method does not work if the boundary behaves like or is more flat. The limit case if the boundary is more flat that the graph of for all , but less flat than . An example of such a function is . In lemma 14 we show that our method doesn’t work in such case. The proof of theorem 1 is divided into several steps. In section 2 we show existence of a weak solution in a class using the Galerkin method (theorem 3). To obtain a weak solution it is enough to assume that , and no further constraint on the geometry of is required. The constraint (1.2) arises when we want to show that the weak solution belongs to class , and we also need . The issue of regularity of the weak solution is treated in section 3. First we prove that the vorticity of the velocity belongs to (lemma 7). Such approach has been applied to incompressible Navier-Stokes equations in  and . In the incompressible case we can next solve a div-rot system to show higher regularity of the velocity, but in the compressible case we have to extract some information on the density. The idea is to use the Helmholtz decomposition in , that means express the velocity as a sum of a divergence-free vector function and a gradient. The standard theory of elliptic equations enables us to show that the divergence-free part belongs to , and in order to show higher regularity of the gradient part it is enough to show that . In lemma 10 we show that , thus we have to show that . The method is to show that the density is a solution to the transport equation

 ¯γw+wx1=H∈H1(Ω). (1.4)

Thus the problem of regularization of the weak solution is reduced a problem of solvability of the transport equation (1.4). The boundary condition (1.6) prescribes the values of the density on the inflow part of the boundary and (1.4) can be solved via method of characteristics, thus a singularity appears in the points and , which we will call the singularity points. It turns out that we can solve the equation (1.4) provided that the singularity is not too strong, what is reflected in the constraint (1.2). We will finish the introductory part removing inhomogeneity on the boundary. Let us construct a function satisfying

 n⋅2μD(u)⋅τ+f( u⋅τ)|Γ=Bandn⋅u|Γ=0, (1.5)

such that . Then a pair , where , satisfies

 ⎧⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎨⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪⎩∂x1~u−μΔ~u−(ν+μ)∇div~u+γ∇w=~FinΩ,div~u+∂x1w=~GinΩ,n⋅2μD(~u)⋅τ+f( ~u⋅τ)=0onΓ,n⋅~u=0onΓ,w=0onΓin, (1.6)

where

 {~F=F+μΔu0+(ν+μ)∇divu0−∂x1u0∈L2(Ω)~G=G−divu0∈H1(Ω). (1.7)

Obviously we have

 ||~F||L2(Ω)≤C(||F||L2(Ω)+||B||L2(Γ))and||~G||H1(Ω)≤C(||G||H1(Ω)+||B||L2(Γ)),

thus from now on we can work with the system (1.6) denoting , , and .

## 2 Weak solution

In order to define a weak solution to the system (1.6) consider a space

 V0={v∈C∞(Ω):v⋅n|Γ=0,n⋅2μD(v)⋅τ+f(v⋅τ)|Γ=0}

and equipped with the norm . Consider also a space

 W={η∈L2(Ω):ηx1∈L2(Ω)andη|Γin=0}

with the norm .

Now we want to introduce a weak formulation of (1.6). First, observe that for regular enough we have

 ∫Ω(−μΔu−(ν+μ)∇divu)⋅vdx=∫Ω2μD(u):∇v+νdivudivvdx− ∫Γn⋅[2μD(u)]⋅vdσ−∫Γn⋅[ν(divu)Id]⋅vdσ, (2.1)

where for .
Thus taking in (1.6) and multiplying it by a function we get

 ∫Ω{v⋅∂x1u+2μD(u):∇v+νdivudivv−γwdivv}dx+∫Γf(u⋅τ)(v⋅τ)dσ= =∫ΩF⋅vdx. (2.2)

Multiplying (1.6) by a regular function we get

 ∫Qη[divu+wx1]dx=∫QGηdx. (2.3)

The above considerations leads to a natural definition of a weak solution to the system (1.6).

###### Definition 1.

By a weak solution to the system (1.6) we mean a couple satisfying (2) - (2.3) for each .

We want to show existence of a weak solution using the Galerkin method. In order to show existence of solutions to approximate problems in section 2.1 we apply well known result (lemma 1). This result automatically gives uniform boundedness of the sequence of approximate solutions, what enables us to show convergence of approximate solutions to the weak solution in section 2.2.

### 2.1 Approximate solutions

In order to construct a Galerkin approximation of a weak solution let us introduce an orthonormal basis of : and finite dimensional spaces: . We will search for a sequence of approximations to the velocity in the form

 uN=N∑i=1cNiϕi. (2.4)

Let us denote . Taking , and where

 wN(x1,x2)=∫x1x1–––(G−divuN)(s,x2)ds

in (2) we get

 ∑icNi∫Ω∂x1ϕi⋅ϕkdx+2μ∑icNi∫ΩD(ϕi):∇ϕk +ν∑icNi∫Ωdivϕi⋅divϕkdx−γ∑i∫Ω{∫x1x1–––(G−∑icNidivϕi)(s,x2)ds}divϕkdx +f∑icNi∫Γ(ϕi⋅τ)(ϕk⋅τ)dσ=∫ΩF⋅ϕkdx. (2.5)

For we obtain a system of equations on coefficients . If a function of a form (2.4) satisfies the equations (2.1) for , it means that a pair satisfies (2)-(2.3) for each . We will call such a pair an approximate solution to (2) - (2.3).

The system (2.1), is rather complicated thus in order to solve it we will use the following well known result (see for example ):

###### Lemma 1.

Let be a finitely dimensional Hilbert space and let be a continuous operator satisfying

 ∃M>0:(P(ξ),ξ)>0for||ξ||=M (2.6)

Then

In order to apply lemma 1 we will need some auxiliary results in spaces and .

###### Lemma 2.

(Poincare inequality in )

 ∀v∈V:||u||L2(Ω)≤C(Ω)||∇u||L2(Ω). (2.7)
###### Proof.

Assume that (2.12) doesn’t hold. Then such that . Without loss of generality we can assume , thus

 ||∇vk||L2(Ω)→0. (2.8)

Clearly is a bounded sequence in and thus thanks to boundedness of the compact embedding theorem implies that it contains a subsequence that is a Cauchy sequence in . But (2.8) implies that is also a Cauchy sequence in . Thus is a Cauchy sequence in , hence for some . Obviously and , thus is constant almost everywhere. But also , and since is a bounded set with regular boundary, the unit normal takes all the values from the unit sphere on . Therefore

 v∗a.e.≡const(v∗⋅n)|Γ=0}⇒v∗a.e.≡0,

Now we will use the Poincare inequality to show that in a following modification of the Korn inequality holds:

###### Lemma 3.

Assume that is large enough. Then for :

 ∫Q2μD2(u)+∫Γf(u⋅τ)2dσ≥C∥u∥2H1. (2.9)
###### Proof.

The proof is based on a proof of a different version of the Korn inequality in . We have

 2∫ΩD2(u)=2∑i,j=1[(uixj)2+uixjujxi]=||∇u||2L2(Ω)+∫Ωk∑i,j=1uixjujxidx= =||∇u||2L2(Ω)+∫Ωk∑i,j=1uixiujxjdx−∫Γk∑i,j=1uiujxjnidσ−∫Γk∑i,j=1uiujnjxidσ. (2.10)

The second term of the r.h.s is equal to and the third term vanishes since , thus from (2.1) we get

 2μ∫ΩD2(u)≥μ||∇u||2L2(Ω)−μ∫Γk∑i,j=1uiujnjxidσ, (2.11)

but we have and thus using the Poincare inequality (2.7) we get

 ∫Ω2D2(u)+f(u⋅τ)2≥C(Ω,μ)||u||H1(Ω)+[f−C(Ω,μ)]||u||L2(Γ)

and the last term will be positive provided that is large enough. ∎

The last inequality we need is the Poincare inequality in .

###### Lemma 4.

(Poincare inequality in )

 ∀η∈W:||η||L2(Ω)≤diam(Ω)||ηx1||L2(Ω). (2.12)
###### Proof.

The proof is straightforward using density of smooth functions in and the Jensen inequality. ∎

The following theorem gives a solution to the system (2.1)

###### Theorem 2.

For and there exists a solution to the system (2.1), . The function satisfies

 ||uN||H1(Ω)≤C(DATA). (2.13)
###### Proof.

In order to apply Lemma 1 we have to define an appropriate operator
. For convenience let us define :

 BN(ξN,vN)=∫ΩvN∂x1ξN+2μ∫ΩD(ξN):∇vN+ν∫ΩdivξNdivvN
 −γ∫Ω{∫x1x1–––(G−divξN)(s,x2)ds}divvNdx+f∫Γ(ξN⋅τ)(vN⋅τ)dσ−∫ΩF⋅vNdx.

Now (2.1) can be rewritten as and thus it is natural to define

 PN(ξN)=∑iBN(ξN,ϕk)ϕkforξN∈VN. (2.14)

We have to verify the assumptions of Lemma 1. Obviously and it is a continuous operator. For we have

 (PN(ξN),ξN)=(N∑k=1BN(ξN,ϕk)ϕk,N∑i=1aNiϕi)= =N∑k=1{BN(ξN,ϕk)N∑i=1aNi(ϕi,ϕk)}=N∑k=1BN(ξN,ϕk)aNk=BN(ξN,ξN). (2.15)

Using the definition of we can rewrite (2.1) as

 (PN(ξN),ξN)=2μ∫ΩD2(ξN)dx+ν∫Ωdiv2ξNdxI1+∫ΩξN∂x1ξNdx+∫Γf(ξN⋅τ)2dσI2
 −γ∫Ω{∫x1x1–––(G−divξN)(s,x2)ds}divξNdxI3−∫ΩF⋅ξNdx.

Using the Korn inequality (2.9) we get for large enough. Now let us denote

 ηN(x1,x2)=∫x1x1–––(G−divξN)(s,x2)ds.

Then and we have

 I3=−γ∫ΩηNdivξNdx=∫ΩηNηNx1−∫ΩGηNdx≥
 ≥∫ΩGηNdx≥−C||G||L2(Ω)||ηN||L2(Ω)≥−C||G||L2(Ω)(||G||L2(Ω)+||ξN||H1(Ω)).

Combining these bounds we get

 (PN(ξN),ξN)≥C(μ,Ω)||ξN||2H1(Ω)−(||F||L2(Ω)+||G||L2(Ω))||ξ||H1(Ω)−||G||2L2(Ω).

Thus there exists such that and applying lemma 1 we conclude that But since is a basis of , the definition of (2.14) yields

 PN(ξ∗)=0⟺BN(ξ∗,ϕk)=0,k=1…N,

thus is a solution to (2.1) ∎

### 2.2 Existence of weak solution

Now we show that the sequence constructed in previous section converges to the weak solution of our problem.

###### Theorem 3.

Assume that and is large enough. Then there exists a weak solution to (1.6) satisfying the estimate

 ||u||V+||w||W≤C(DATA). (2.16)
###### Proof.

The estimate (2.13) together with (2.12) gives

 ||uN||H1(Ω)+||wN||L2(Ω)+||wNx1||L2(Ω)≤C(DATA). (2.17)

Since the sequence is bounded in , there exists a subsequence and a function such that . Now let us denote for simplicity . It is bounded in , thus there exists a subsequence for some function . Now we need to show that , but this is quite obvious. We have

 ∀v∈L2:−∫ΩwNkvx1=∫ΩwNkx1v→∫Ωζvand∫ΩwNkvx1→∫Ωwvx1,

thus .

It is a bit more complicated to show the existence of . The estimate (2.17) gives boundedness in of the sequences , , and boundedness in of . Thus up to a subsequence

 divuNL2⇀ξ,∂x1uNL2⇀α,D(uN)L2⇀βanduN⋅τL2(Γ)⇀δ

for some and some . On the other hand, since the sequence is bounded in , the compactness theorem yields up to a subsequence for some . We want to show that in fact and that satisfies (2) - (2.3). But we have :

 −∫Ωu∂x1ϕ←−∫Ωun∂x1ϕ=∫Ωϕ∂x1un→∫Ωαϕ

thus . Similarily we can verify that

 ⎧⎪⎨⎪⎩ξ=divuβ=D(u)δ=u⋅τ|Γ. (2.18)

Thus , and the pair satisfies (2) - (2.3) . The density of in implies that it also satisfies (2) - (2.3) . Thus indeed is a weak solution. The estimate (2.16) is obtained in a standard way taking and in (2) - (2.3) and then applying the Korn inequality (2.9) and the Poincare inequality in (2.12). ∎

## 3 Regularity

In this section we will show that the weak solution belongs to a class . The idea of the proof has been outlined in the introduction. We start with showing that if is a weak solution then .

Since on this level we have only weak solutions, we have to work with the weak formulation (2) - (2.3). Consider a special class of test functions:

 V1={v∈V:∇⊥ϕ:ϕ∈H2(Ω),v⋅n|Γ=0,ϕ|Γ=0}

where . Note that on we have .
Let us denote .
Since for , thus for (2) takes the form

 ∫Ωα∂x1ϕdx+2μ∫ΩD(u):∇vdx=∫ΩF⋅∇⊥ϕdx−∫Γf(u⋅τ)∂ϕ∂ndσ (3.1)
###### Lemma 5.

For we have

 ∫Ω2μD(u):∇vdx=−μ∫ΩαΔϕdx+∫Γ2(μχ−f)(u⋅τ)∂ϕ∂ndσ (3.2)

where and denotes the curvature of .

To prove lemma 5 we will use following auxiliary result, proved in :

###### Lemma 6.

For we have

 rotu|Γ=(2χ−fμ)(u⋅τ), (3.3)

where is the curvature of .

Proof of lemma 5. Due to density of in it is enough to proove (3.2) for
. For such functions we have (we omit the subscript ):

 ∫Ω2μD(u):∇vdx=−∫Ω2μdivD(u)⋅vdx+∫Γn⋅2μD(u)⋅vdσ

Since we have , and using the definition of we can write

 ∫Ω2μD(u):∇vdx=−∫Ωμ(Δu+∇divu)⋅∇⊥ϕdx−∫Γf(u⋅τ)∂ϕ∂ndσ. (3.4)

Integration by parts yields

 ∫Ω∇divu⋅∇⊥ϕdx=∫Γdivu∂ϕ∂τdσ=0 (3.5)

and

 ∫ΩΔu⋅∇⊥ϕ=∫ΩϕΔrotudx+∫ΓϕΔu⋅τdσ=0 =−∫Ω∇ϕ⋅∇rotu+∫Γϕ∂∂nrotu=0=∫ΩrotuΔϕ−∫Γrotu∂ϕ∂ndσ (3.6)

Substituting (3.5) and (3) into (3.4) we get

 ∫Ω2μD(u):∇vdx=−μ∫ΩrotuΔϕdx+μ∫Γrotu∂ϕ∂ndσ−∫Γf(u⋅τ)dσ, (3.7)

and application of (3.3) to the boundary term yields (3.2)

With lemma 5 (3.1) takes the form

 ∫Ωα∂x1ϕdx−∫QαΔϕdx=∫ΩF⋅∇⊥ϕdx−∫Γ