Homogenization is a mathematical tool that allows changing the scale in problems containing several characteristic scales. Typical examples of its utilization are finding effective models for composite materials, in optimal shape design, etc. Another important example, which we are interested in, is the fluid mechanics of the flow through porous medium.
In porous medium, there are at least two length scales: microscopic scale and macroscopic scale. The partial differential equations describing a physical phenomenon are posed at the microscopic level whereas only macroscopic quantities are of interest for the engineers or the physicists. Therefore, effective or homogenized equations should be derived from the microscopic ones by an asymptotic analysis. To this end, it is convenient to assume that the porous medium has a periodic structure.
A number of known laws from the dynamics of fluids in porous media were derived using homogenization. The most well-known example is Darcy's law, being the homogenized equation for one-phase flow through a rigid porous medium. Its formal derivation by two-scale expansion goes back to the classical paper by Sanchez- Palencia [1], Keller [2] and the classical book Bensoussan [3]. It was rigorously derived by using oscillating functions by Tartar [4]. In other cases of periodic porous media, we refer the readers to the papers by Allaire [5-8] and Mikelic [9, 10]. Other works can be seen in [11-13] and the references therein. Besides the Darcy law, Brinkman [14] introduced a new set of equations, which is called the Brinkman law, an intermediate law between the Darcy and Stokes equations. The so-called Brinkman law is obtained from the Stokes equations by adding to the momentum equation a term proportional to the velocity (see [7]).
Inspired by the work from Feireisl [11], we consider the asymptotic behavior of a compressible fluid in a periodic medium. Before stating the system, let us recall the domain we consider. A porous medium is defined as the periodic repetition of an elementary cell of size $ \varepsilon $ (we assume that $ \frac{1}{\varepsilon} $ to be an integral) in a bounded domain $ \Omega $ of $ R^n $ with $ n = 2,3 $. The solid part of the porous medium is also taken of size $ \varepsilon $. The domain $ \Omega_\varepsilon $ is then defined as the intersection of $ \Omega $ with the fluid part. We consider the density dependent fluid governed by the full compressible Navier-Stokes equations. So, we have the following equations
where $ u_\varepsilon, \rho_\varepsilon, \mathbb{T}_\varepsilon $ are the unknown quantities velocity, density and temperature. $ p_\varepsilon = p_\varepsilon(\rho_\varepsilon,\mathbb{T}_\varepsilon) $ is the pressure, $ e_\varepsilon = e_\varepsilon(\rho_\varepsilon,\mathbb{T}_\varepsilon) $ is the internal energy, $ f $ (the exterior force) is given on $ \Omega\times(0,T) $. We assume that $ f $ is smooth enough. $ T\in (0,\infty) $ is fixed; $ \kappa = \kappa(\mathbb{T}_\varepsilon), \mu = \mu(\mathbb{T}_\varepsilon) $ are positive for $ \mathbb{T}_\varepsilon\geq0 $ and $ \kappa,\mu\in W^{1,\infty}([0,\infty)) $.
We also assume that $ u_\varepsilon $ satisfies
and in order to fix ideas we impose Neumann boundary conditions on $ \mathbb{T}_\varepsilon $ namely
where $ n $, as usual, the unit outward normal to $ \partial\Omega_\varepsilon $.
In this paper, we assume that the initial conditions
are bounded in $ L^\infty(\Omega_\varepsilon) $.
In this paper, we also assume that the transport coefficients $ \mu(\mathbb{T}_{\varepsilon}) $ and $ \kappa(\mathbb{T}_{\varepsilon}) $ satisfying the following conditions
for all $ \mathbb{T}_{\varepsilon}\geq 0 $. $ \underline{\mu},\underline{\kappa}, \overline{\kappa} $ are positive constants.
Let us recall that the equation with temperature in (1.1), it is equivalent (at least formally) to
For simplicity, in this paper, we consider the models in astrophysics and the state equation for the pressure $ p_\varepsilon $ and the internal energy $ e_\varepsilon $ satisfying the Joule's law (see [15])
where $ \gamma\geq n $ for $ n = 2,3 $, $ \frac{3}{2}<\beta<\gamma $, $ a,b,c $ are positive constants. Then the specific entropy reads,
We assume that the initial condition
Let us also recall that, at least formally, the following identity holds
Our aim here is to investigate the asymptotic behaviors of $ \rho_\varepsilon,\; u_\varepsilon $ and $ \mathbb{T}_{\varepsilon} $ as $ \varepsilon\rightarrow 0 $ under the assumptions mentioned above. The main difficulty in this paper is how to pass the limit in the momentum and energy equations. To overcome this obstacle, we have to regularize the system both in time and in space before we can pass the limit. In this paper, we exert the conditions on the entropy to get the estimates. Moreover, we rigorously proved that the low boundary of $ \gamma $ would be $ n $ when passing the limit to the convection term. At the limit process, we fall back on the two-scale convergence method to obtain the homogenized model. Those are quite different from [11].
Let $ \Omega $ be an open bounded subset of $ R^n $ with $ n = 2 $ or $ 3 $ and defined $ \mathcal{Y} = [0,1]^n $ to be the unit open cube of $ R^n $. Let $ \mathcal{Y}_s $ be a closed smooth subset of $ \mathcal{Y} $ with a strictly positive measure. The fluid part is then defined by $ \mathcal{Y}_f = \mathcal{Y}-\mathcal{Y}_s $. Let $ \theta = |\mathcal{Y}_f| $. The constant $ \theta $ is called the porosity of the porous medium. We assume that $ 0<\theta<1. $
Repeating the domain $ \mathcal{Y}_f $ by $ \mathcal{Y} $-periodicity, we get the whole fluid domain $ D_f $, we can write it as
Then the solid part is defined by $ D_s = R^n-D_f $. It is easy to see that $ D_f $ is a connected domain, while $ D_s $ is formed by separated smooth subsets. In the sequel, we denote for all $ k\in Z^n $, $ \mathcal{Y}^k = \mathcal{Y}+k $ and then $ \mathcal{Y}_f^k = \mathcal{Y}_f+k $. For all $ \varepsilon $, we define the domain $ \Omega_\varepsilon $ as the intersection of $ \Omega $ with the fluid domain scaled by $ \varepsilon $, namely, $ \Omega_\varepsilon = \Omega\cap \varepsilon D_f $. To get a smooth connected domain, we will not remove the solid part of the cells which intersect with the boundary of $ \Omega $. Now, the fluid domain can be also defined by
Throughout this paper, we denote $ L^p(0,T;L^q(X)) $, the time-space Lebesgue spaces, where $ X $ would be $ \Omega $ or $ \Omega_\varepsilon $; $ W^{s,p}(X) $ is the classical Sobolev space with all functions, whose all derivatives up to order $ s $ belong to $ L^p $ and $ H^s(X) = W^{s,2}(X) $; $ W_0^{1,p}(X) $ is the subset of $ W^{1,p}(X) $ with trace 0 on $ X $. We also denote $ W^{-s,p'}(X) $, the dual space of $ W_0^{s,p}(X) $, where $ p' $ is the conjugate exponent of $ p $; $ C $ is a constant that may differ from one place to another. Throughout this paper, we use $ ||\cdot||_X $ to denote the modules for all vectors and matrixes if there is no confusion.
Due to the presence of the holes, the domain $ \Omega_\varepsilon $ depends on $ \varepsilon $ and hence to study the convergence of $ \{u_\varepsilon,\rho_\varepsilon,p_\varepsilon\} $, we have to extend the functions defined in $ \Omega_\varepsilon $ to the whole domain. This can be done in two different possible ways.
Definition 1.1 For any fixed $ \varepsilon\in L^1(\Omega_\varepsilon) $, we define
the null extension and
the mean value extension.
The relation between the weak limits of both types of extensions is given by the following lemma (see [13]).
Lemma 1.1 For all $ \omega_\varepsilon\in L^1(\Omega_\varepsilon) $, the following two assertions are equivalent
1. $ \widehat{\omega}_\varepsilon\rightharpoonup \omega $ in $ L^1(\Omega) $; 2. $ \widetilde{\omega}_\varepsilon\rightharpoonup \theta\omega $ in $ L^1(\Omega) $.
A very important property of the porous media is a variant of the Poincare's inequality. Due to the presence of the holes in $ \Omega_\varepsilon $, the Poincare's inequality reads in [12].
Lemma 1.2 Let $ 1\leq p,q<\infty $ and $ u\in W^{1,p}_0(\Omega_\varepsilon) $, then
where $ C $ depends only on $ \mathcal{Y}_f $ and $ p,q $ satisfies (1) $ 1\leq p <n, p\leq q\leq p^* = \frac{np}{n-p} $; (2) $ p = n, p\leq q<\infty $; (3) $ p>n, p\leq q\leq\infty $.
We also need the restriction operator constructed by Tartar (see [4]).
Lemma 1.3 There exists an operator $ \mathcal{R}_\varepsilon $ with the following properties
1. $ \mathcal{R}_\varepsilon $ is a bounded linear operator on $ W^{1,p}_0(\Omega) $ ranging in $ W^{1,p}_0(\Omega_\varepsilon),\; \; p\geq 2 $;
2. $ \mathcal{R}_\varepsilon[\varphi] = \varphi|_{\Omega_\varepsilon} $ provide $ \varphi = 0 $ in $ \Omega-\Omega_\varepsilon $;
3. $ \rm{div}_x\varphi = 0 $ in $ \Omega $ implies $ \rm{div}_x\mathcal{R}_\varepsilon[\varphi] = 0 $ in $ \Omega_\varepsilon $;
4. $ ||\mathcal{R}_\varepsilon[\varphi]||_{L^p(\Omega_\varepsilon)}+\varepsilon||\nabla\mathcal{R}_\varepsilon[\varphi]||_{L^p(\Omega_\varepsilon)} \leq C(||\varphi||_{L^p(\Omega)}+\varepsilon||\nabla\varphi||_{L^p(\Omega)}) $.
In addition, we can find the restriction operator $ \mathcal{R}_\varepsilon $ satisfies a compatibility relation with the extension operator introduces in Definition 1.1, namely,
Lemma 1.4 (Bogovskii operator) There exists a linear operator $ \mathcal{B}_\varepsilon $ with the properties: if $ f\in L^p(\Omega_\varepsilon) $, then $ \phi = \mathcal{B}_\varepsilon[f] $ such that
Moreover, the following estimates is satisfied
In addition, if $ f = \rm{div}g $ with $ g\in L^q(\Omega_\varepsilon),\; g\cdot n|_{\partial\Omega_\varepsilon} = 0 $, then
There are many ways to construct $ \mathcal{B}_\varepsilon $. An explicit formula was proposed by Bogovskii [16] on Lipschitz domains. Some properties of $ \mathcal{B}_\varepsilon $ were discussed by Galdi [17]. In the domain with porous medium, the relevant estimates were obtained by Masmoudi [13].
Finally, we define the permeability matrix $ \mathcal{A} $. For $ 1\leq i\leq n $, let $ (\omega_i,\pi_i)\in H^1(\mathcal{Y}_f)\times L^2(\mathcal{Y}_f)/R $ be the unique solution of the following system
where $ e_i $ is the standard basis of $ R^n $. Set $ \omega_i^\varepsilon = \omega_i(\frac{x}{\varepsilon}),\; \pi_i^\varepsilon = \pi_i(\frac{x}{\varepsilon}) $. Then we get the cell problem
Lemma 1.5 Let $ \omega_i^\varepsilon, \pi_i^\varepsilon $ be the solution to the cell problem and be extended to zero outside $ \Omega_\varepsilon $. Then the following estimates hold
for any $ 1\leq q\leq +\infty $, $ C $ only depends on $ q $ and $ \mathcal{Y}_f $.
The construction of $ \omega_i^\varepsilon $ and the properties of $ \omega_i^\varepsilon,\; \pi_i^\varepsilon $ stated in Lemma 1.5 can be found in Tartar [4] or Masmoudi [13]. Let us define
It is easy to see that $ \mathcal{A} $ is a symmetric positive defined matrix. The form of the permeability matrix has different form if $ \mathcal{Y}_s $ has different forms. For more information about $ \mathcal{A} $, we refer the interested readers to Allaire [7] for detail.
Now we introduce the definition of weak solution to the systems (1.1)–(1.4)
Definition 1.2 We shall say that a trio $ \{u_\varepsilon,\rho_\varepsilon,p_\varepsilon\} $ is a weak solution of (1.1), supplemented with the boundary and initial conditions (1.2)–(1.4) if only if
1. $ \rho_\varepsilon\in L^\infty(0,T;L^\gamma(\Omega_\varepsilon)),\; \; \gamma\geq n $ and $ u_\varepsilon \in L^2(0,T;H^1_ 0(\Omega_\varepsilon)) $, and the integral identity
holds for any test function $ \varphi \in C_0^\infty([0,T)\times\overline{\Omega}_\varepsilon) $.
2. $ p_\varepsilon \in L^q(\Omega_\varepsilon\times (0,T)) $ for some $ q>1 $, and
3. $ \mathbb{T}_\varepsilon \in L^2(0,T; L^6(\Omega_\varepsilon)) \cap L^2(0,T; H^1(\Omega_\varepsilon)) $, $ \mathbb{T}_\varepsilon >0 $ a.a in $ \Omega_\varepsilon\times (0,T) $ and the integral identity
holds for any $ \varphi\in C_0^\infty([0,T)\times\bar{\Omega}_\varepsilon) $.
With all the preparation above, we are now in the position to state our main result in this paper.
Theorem 1.1 Let $ \{u_\varepsilon, \rho_\varepsilon, \mathbb{T}_\varepsilon\}_{\varepsilon>0} $ be a family of weak solutions to system (1.1). We assume that $ \gamma\geq n $ for $ n = 2,3 $ and
Then, there exist three functions $ u, \rho, \Xi $ such that
where $ \Xi = \Xi(t) $ is a spatially homogeneous function, $ \rho_(x,0) = \rho_0(x) $. Moveover, $ \{u, \rho, \Xi\} $ satisfies the following homogenized system
where $ p,e $ are given by
and $ \mathcal{A} $ is the so-called permeability matrix. The specific homogenized entropy $ s $ related to the homogenized pressure $ p $ and the homogenized inner energy $ e $ through Gibbs' equation
where $ s $ is given by $ s = 3c\sqrt{\Xi}-\frac{b}{\beta-1}\rho^{\beta-1}. $
In this section, we collect all available bounds on the family $ \{u_\varepsilon, \rho_\varepsilon, \mathbb{T}_\varepsilon\} $. Let us begin with the basic estimates
In this subsection, we obtain some estimates for the solutions to system (1.1) which are independent of $ \varepsilon $. First, from the conservation of mass, we have $ \rho_\varepsilon\in L^\infty(0,T;L^1(\Omega_\varepsilon)) $. We set
Next, integrating (1.7) over $ \Omega_\varepsilon\times(0,t) $ for any $ t\in [0,T] $, we have
We then deduce the uniform bound of $ \rho_\varepsilon |s_\varepsilon| $ in $ L^\infty(0,T;L^1(\Omega_\varepsilon)) $ and $ \rho_\varepsilon $ is uniformly bounded in $ L^\infty(0,T;L^\gamma(\Omega_\varepsilon)) $. By Lemma 1.2 and the special $ s_\varepsilon $, $ e_\varepsilon $, we also have
In this subsection, we want to deduce the uniform bounds on $ \mathbb{T}_\varepsilon $. Note that we only have the bounds of $ \nabla\mathbb{T}_\varepsilon $ and $ \rho_\varepsilon \mathbb{T}_\varepsilon^{\frac{3}{2}} $. The estimate on $ \mathbb{T}_\varepsilon $ itself isn't included. To fill this gap, we fall back on the Ne\u{c}as' lemma [18] and the Sobolev embedding theorem in the porous medium
Lemma 2.1 Let $ \Omega $ be a bounded Lipschitz domain in $ R^2 $ or $ R^3 $. Let $ M,K $ be two positive real numbers and $ \rho $ a non-negative function such that
Then there exists a constant $ C = C(M,K) $ such that
for any $ \omega\in L^2(\Omega) $.
Proof see [11].
Following the idea in [11] and [12], we prove the Sobolev embedding theorem in the porous medium has the form.
Lemma 2.2 Let $ v\in W^{1,p}(\Omega_\varepsilon) $. Then we have
with $ C $ independent of $ \varepsilon $, $ 1<p\leq q \leq \frac{np}{n-p} $ for $ n\geq 3 $ and $ 1<p\leq q <\infty $ for $ n = 2 $.
Proof Obviously, it is enough to show
By the definition of the module of $ L^p $ and Lemma 1.2, Lemma 1.3, we have
where $ \frac{1}{q}+\frac{1}{q'} = 1 $, $ \frac{1}{p}+\frac{1}{p'} = 1 $.
By Lemma 1.3 and (2.2), we obtain
which implies that
Next, we write
In accordance with Lemma 2.1 and (2.2)–(2.4), we have
We deduce that $ ||\mathbb{T}_\varepsilon||_{L^2(\Omega_\varepsilon\times (0,T))}\leq C $. By Lemma 2.2, we obtain
with $ C $ independent of $ \varepsilon $.
We get the uniform bound on $ \rho_\varepsilon $ in $ L^\infty(0,T;L^\gamma(\Omega_\varepsilon)) $. This is not enough when we consider the strong convergence of the density. In this part, we want to obtain some more delicate estimates. To this end, we choose the test function $ v_\varepsilon = \psi(t)\mathcal{B}_\varepsilon(\rho_\varepsilon- \frac{1}{|\Omega_\varepsilon|} \int_{\Omega_\varepsilon}\rho_\varepsilon dx) $, where $ \psi(t)\in C_0^\infty(0,T),\; \psi(t)>0 $, $ \mathcal{B}_\varepsilon $ denotes Bogovskii's operator introduced in Lemma 1.3.
For any $ t\in (0,T) $, we have
Multiplying the second equation of (1.1) by $ v_\varepsilon $, we get (we drop the $ dxdt $)
The first term is the most technical and requires some spatial regularization of $ v_\varepsilon $ (see [13]). Let us explain how the difficulty related to $ I_1 $ can be solved. The estimation of rest terms are the same as [13]. Set $ \chi(x)\in C_0^\infty(R^n) $ such that $ \chi(x)\geq0 $, $ \int_{R^n}\chi(x) dx = 0 $. For all $ \delta\in(0,1), $ we denote $ \chi_\delta(x) = \frac{1}{\delta^n}\chi(\frac{x}{\delta}) $. Next, we denote $ \widetilde{\rho}_{\varepsilon,\delta} = \widetilde{\rho}_{\varepsilon}\ast\chi_\delta(x) $, where $ \widetilde{\rho}_{\varepsilon} $ is defined in Definition 1.1. Then we have the following relation
where $ r_{\varepsilon,\delta} $ is nothing but a commutator, $ r_{\varepsilon,\delta}\rightarrow 0 $ in $ L^2(0,T;L^{\frac{2\gamma}{\gamma+2}}(\Omega)) $ as $ \delta\rightarrow 0 $. Now we take $ v_{\varepsilon,\delta} = \psi(t) \mathcal{B}_\varepsilon(\widetilde{\rho}_{\varepsilon,\delta}-\frac{1}{|\Omega_\varepsilon|} \int_{\Omega_\varepsilon}\widetilde{\rho}_{\varepsilon,\delta} dx) $ in stead of $ v_{\varepsilon} $. Taking the time derivative of $ t $ of $ v_{\varepsilon,\delta} $, we have
It is easy to check that
where $ 3+n(\frac{1}{p}-\frac{1}{2})\geq 2 $ for $ n = 3 $ and $ 3+n(\frac{1}{p}-\frac{1}{2})>2 $ for $ n = 2 $.
Next, after a straightforward manipulation we have
By Lemma 1.2 and Lemma 1.3, we deduce that
To $ I_{112} $, we have
where $ L^2_T(\Omega_\varepsilon) = L^2(0,T;L^{2}(\Omega_\varepsilon)) $. Then we let $ \delta\rightarrow0 $ and deduce that $ \int_0^T \int_{\Omega_\varepsilon}\varepsilon^2 \rho_\varepsilon u_\varepsilon\cdot \psi(t)I_{112} $ also tends to 0.
From above, we get the estimate
With those estimations, we can also obtain $ \widehat{p_\varepsilon} $ converges strongly to $ p $ in $ L^2(0,T;L^s(\Omega)) $ for some $ s>1 $, see [12].
We divide three steps to finish the proof.
Step 1 Passing the limit in the continuum equation. As in [13], we can prove that
Multiplying above equation by $ \varphi \in C_0^\infty(\Omega\times(0,T]) $ and integrating over $ \Omega\times(0,T) $, we have
Passing the limit and using the strong convergence of $ \widehat{\rho}_\varepsilon $ and $ \widehat{\rho}_{0,\varepsilon} $, we obtain
We then recover the homogenized equation and the initial condition
Step 2 Passing the limit in the momentum equations.
To pass the limit, we have to regularize the second equation of (1.1) both $ x $ and $ t $. To this end, we set $ \chi_1(x)\in C_0^\infty(R^n) $ such that $ \chi_1(x)\geq 0 $, $ \int_{R^n}\chi_1(x) dx = 1 $. Let $ \eta_1\in (0,1) $ and set $ \chi_{1\eta_1}(x) = \frac{1}{\eta_1^n}\chi_1(\frac{x}{\eta_1}) $. We also set $ \chi_2(t)\in C_0^\infty(R^+) $ such that $ \chi_2(t)\geq 0 $, $ \int_{R^+}\chi_2(t) dt = 1 $. Let $ \eta_2\in (0,1) $ and set $ \chi_{2\eta_2}(t) = \frac{1}{\eta_2}\chi_2(\frac{t}{\eta_2}) $. Take $ T_m(\widehat{\rho_\varepsilon})\ast\chi_{1\eta_1}\ast\chi_{2\eta_2}\omega_n^\varepsilon \varphi $ as the test function, where $ T_m $ is the truncated function by integer $ m $ and $ T_m(\widehat{\rho_\varepsilon})\ast\chi_{1\eta_1}\ast\chi_{2\eta_2} $ is prolonged by zero outside $ \Omega\times(0,T) $, $ \omega_n^\varepsilon $ be the solution of the cell problem and $ \varphi \in C_0^\infty(\Omega\times(0,T)) $. We have
To pass the limit in $ I_1 $ and $ I_2 $, we can easily deduce that $ I_1\rightarrow 0, $ $ I_2\rightarrow 0 $ as $ \varepsilon\rightarrow 0. $ Now, we can pass the limit in $ I_3 $ and $ I_4 $ since $ \widehat{\rho}_\varepsilon $ converges strongly,
and
Finally, we consider the limit on the left-hand side. We write it as
On the other hand,
We also have
To $ L_{11} $, by virtue of the cell problem, we have
We continue discussing the convergence of those three terms.
Finally, passing the limit with $ \varepsilon\rightarrow 0 $, the limit of $ L_{111} $ is given by
Let $ \eta_1,\eta_2\rightarrow 0 $ and $ m\rightarrow +\infty $ and by using the arbitrary property of $ \varphi $, (3.4), (3.5) and (3.6) lead to the homogenized system of the momentum. That is
Step 3 Passing the limit in the energy equation.
Integrating the energy equation in $ \Omega\times(0,t) $ and passing the limit, we have
Then the following Gibbs' equation holds
Now we finish Theorem 1.2.