Introduction
Poroelasticity describes the interaction between fluid flow and solid deformation in porous media and plays an important role in geomechanics, subsurface engineering, and biological tissue modeling. The classical Biot model couples mechanical equilibrium with fluid mass conservation through the dependence of stress on pore pressure and the contribution of volumetric deformation to fluid storage [1 , 2 , 3 ] . In heterogeneous porous media [4 ] , the elastic, hydraulic, and coupling coefficients may vary over multiple spatial scales. Resolving these variations directly can be computationally expensive, particularly when strong material contrasts and complex microstructures are present. Meanwhile, the coupling between fluid flow and solid deformation complicates both the construction of effective models and the analysis of their well-posedness and stability. This motivates the development of macroscopic models that retain the essential effects of fine-scale heterogeneity while reducing computational complexity.
To address the multiscale nature of heterogeneous media, various multiscale numerical methods have been developed. Multiscale finite element methods construct local basis functions by solving fine-scale problems on coarse-grid blocks, thereby incorporating unresolved heterogeneities into a reduced coarse-grid approximation. The generalized multiscale finite element method (GMsFEM) further employs snapshot spaces and local spectral problems to systematically select the dominant modes, and has been applied to linear and nonlinear poroelasticity problems [5 , 6 , 7 ] . The constraint energy minimizing generalized multiscale finite element method (CEM-GMsFEM) further constructs basis functions through constrained energy minimization on oversampled regions [8 ] . It has also been applied to poroelasticity problems with high-contrast coefficients [9 ] . Although these approaches can substantially reduce the computational cost of resolving fine-scale heterogeneity, their direct output is typically a discrete coarse-grid system rather than a continuous macroscopic model expressed in terms of explicit governing laws.
Multicontinuum modeling offers an alternative description for heterogeneous porous media in which a single macroscopic pressure variable may be insufficient to represent distinct fluid responses. A representative example is the multiple-network poroelastic theory (MPET), which extends the classical Biot model to a deformable porous skeleton permeated by several interacting fluid networks [10 , 11 ] . Building on double-porosity and multiple-porosity descriptions, MPET provides a flexible framework for representing coupled flow and deformation in complex porous systems and has been applied to biological tissues, particularly the human brain [12 , 13 ] . However, the predictive capability of these models depends on the specification of effective storage, transfer, and poromechanical coupling coefficients. Determining these macroscopic coefficients, especially cross-porosity storage coefficients, can be challenging because experimental data capable of quantifying them are often limited [13 , 14 ] . This motivates systematic approaches that relate macroscopic multicontinuum parameters to the underlying microscopic properties.
The multicontinuum homogenization method provides a systematic framework for deriving such macroscopic multicontinuum models from local microscopic problems. Its central idea is to introduce several macroscopic variables, or continua, at each coarse-scale location and represent the fine-scale solution through continuum-dependent basis functions obtained from constrained local problems. The constraints are associated with prescribed continuum averages, while the local problems are posed on representative volume elements and, in particular, on oversampled regions to reduce the influence of artificial boundary conditions [15 ] . This framework has been further developed to treat general systems, different types of constraints, and mixed or nonsymmetric formulations [16 , 17 , 18 ] . For poroelasticity, Ammosov et al. [14 ] constructed coupled displacement-pressure cell problems, derived generalized multicontinuum models, and numerically investigated both full and simplified models. Building on these developments, a systematic analysis of the full macroscopic system is needed to characterize the role of generalized storage and coupling terms in its variational structure, well-posedness, and energy stability.
Finite element methods for Biot poroelasticity have been studied extensively. Phillips and Wheeler developed mixed and continuous Galerkin formulations for the classical Biot system and established convergence and error estimates for continuous- and discrete-in-time approximations [19 , 20 ] . For multiple-network poroelasticity, Lee et al. [11 ] proposed and analyzed a mixed finite element formulation, including energy estimates and a priori error estimates that are robust in nearly incompressible and low-storage regimes. These works provide important analytical foundations for the numerical treatment of coupled poromechanical systems. The full multicontinuum model considered here, however, contains additional generalized pressure-storage terms and gradient-dependent coupling terms generated by first-order multiscale corrections. Its analysis therefore requires suitable boundedness and coercivity conditions on the assembled bilinear forms, together with a careful treatment of the reciprocity relations between the coupling operators. In particular, the cancellation of reciprocal coupling terms in the continuous energy estimate must be preserved by the discretization to establish corresponding stability and error bounds.
The remainder of the paper is structured as follows. Section 2 introduces the microscopic Biot problem and the variational formulation of the full multicontinuum model, and describes its fully discrete finite element scheme. Section 3 proves the existence, energy stability of the continuous model, and the well-posedness and stability of the discrete scheme. Section 4 derives a priori error estimates. Section 5 reports numerical experiments. Finally, Section 6 summarizes the main findings.
Throughout the paper we use C C C , with or without subscript, to denote a positive constant, which could have different values at different appearances.
Preliminaries. Multicontinuum models
We derive the multicontinuum macroscopic equations directly from the variational formulation of the Biot model using the multicontinuum homogenization method.
Biot model and variational formulation
Let Ω ⊂ R d \Omega\subset\mathbb{R}^{d} Ω ⊂ R d be a bounded domain and let J = ( 0 , T ] J=(0,T] J = ( 0 , T ] denote the time interval, where T > 0 T>0 T > 0 is the final time. We consider the quasi-static Biot system
− ∇ ⋅ σ ( u ) + α ∇ p = f , in Ω × J , c 0 ∂ t p + α ∇ ⋅ ∂ t u − ∇ ⋅ ( k ∇ p ) = g , in Ω × J . (1) \begin{aligned}
-\nabla\cdot\sigma(u) + \alpha\nabla p &= f, && \text{in } \Omega\times J,\\
c_0 \partial_t p + \alpha\nabla\cdot\partial_t u - \nabla\cdot(k \nabla p) &= g, && \text{in } \Omega\times J. \tag*{(1)}
\end{aligned} − ∇ ⋅ σ ( u ) + α ∇ p c 0 ∂ t p + α ∇ ⋅ ∂ t u − ∇ ⋅ ( k ∇ p ) = f , = g , in Ω × J , in Ω × J . (1)
Here, u u u denotes the displacement, p p p is the pore pressure, and σ ( u ) = 2 μ ε ( u ) + λ ( ∇ ⋅ u ) I \sigma(u)=2\mu\varepsilon(u)+\lambda(\nabla\cdot u)I σ ( u ) = 2 μ ε ( u ) + λ ( ∇ ⋅ u ) I denotes the stress tensor, ε ( u ) = 1 2 ( ∇ u + ∇ u T ) \varepsilon(u)=\frac{1}{2}(\nabla u+\nabla u^{T}) ε ( u ) = 2 1 ( ∇ u + ∇ u T ) denotes the strain tensor, α \alpha α is the Biot-Willis coefficient, c 0 c_0 c 0 is the constrained storage coefficient, k k k is the permeability tensor, and f f f and g g g denote the prescribed body force and fluid source, respectively. For simplicity, we impose homogeneous Dirichlet boundary conditions,
u = 0 , p = 0 , on ∂ Ω × J , u=0,\qquad p=0,\qquad\text{on } \partial\Omega\times J, u = 0 , p = 0 , on ∂ Ω × J ,
together with the initial condition
p ( x , 0 ) = p 0 ( x ) , x ∈ Ω . p(x,0)=p_0(x),\qquad x\in\Omega. p ( x , 0 ) = p 0 ( x ) , x ∈ Ω.
Define
W u = [ H 0 1 ( Ω ) ] d , W p = H 0 1 ( Ω ) . W_u=[H_0^1(\Omega)]^d,\qquad W_p=H_0^1(\Omega). W u = [ H 0 1 ( Ω ) ] d , W p = H 0 1 ( Ω ) .
The weak formulation reads: find ( u , p ) ∈ W u × W p (u,p)\in W_u\times W_p ( u , p ) ∈ W u × W p such that, for every t ∈ J t\in J t ∈ J ,
a u ( u , v ) − b ( p , v ) = l u ( v ) , v ∈ W u , c ( ∂ t p , q ) + b ( q , ∂ t u ) + a p ( p , q ) = l p ( q ) , q ∈ W p . (2) \begin{aligned}
a_u(u,v)-b(p,v)&=l_u(v), && v\in W_u,\\
c(\partial_t p,q)+b(q,\partial_t u)+a_p(p,q)&=l_p(q), && q\in W_p. \tag*{(2)}
\end{aligned} a u ( u , v ) − b ( p , v ) c ( ∂ t p , q ) + b ( q , ∂ t u ) + a p ( p , q ) = l u ( v ) , = l p ( q ) , v ∈ W u , q ∈ W p . (2)
where
a u ( u , v ) = ∫ Ω σ ( u ) : ε ( v ) , b ( p , v ) = ∫ Ω α p ∇ ⋅ v , a_u(u,v)=\int_{\Omega}\sigma(u):\varepsilon(v), \qquad b(p,v)=\int_{\Omega}\alpha p\nabla\cdot v, a u ( u , v ) = ∫ Ω σ ( u ) : ε ( v ) , b ( p , v ) = ∫ Ω α p ∇ ⋅ v ,
a p ( p , q ) = ∫ Ω k ∇ p ⋅ ∇ q , c ( p , q ) = ∫ Ω c 0 p q , a_p(p,q)=\int_{\Omega}k\nabla p\cdot\nabla q, \qquad c(p,q)=\int_{\Omega}c_0pq, a p ( p , q ) = ∫ Ω k ∇ p ⋅ ∇ q , c ( p , q ) = ∫ Ω c 0 pq ,
l u ( v ) = ∫ Ω f v , l p ( q ) = ∫ Ω g q . l_u(v)=\int_{\Omega}fv, \qquad l_p(q)=\int_{\Omega}gq. l u ( v ) = ∫ Ω f v , l p ( q ) = ∫ Ω g q .
Multicontinuum approximation
This multicontinuum formulation follows [14 ] ; however, we split the expansion for displacement and pressure fields. Let N c N_c N c denote the number of continua. We introduce the multicontinuum approximations
u H = ∑ i = 1 N c ( N i u U i + M i u ⋅ ∇ U i ) , (3) u_H=\sum_{i=1}^{N_c}\left(N_i^u U_i+M_i^u\cdot\nabla U_i\right), \tag*{(3)} u H = i = 1 ∑ N c ( N i u U i + M i u ⋅ ∇ U i ) , (3)
p H = ∑ i = 1 N c ( N i p P i + M i p ⋅ ∇ P i ) . (4) p_H=\sum_{i=1}^{N_c}\left(N_i^p P_i+M_i^p\cdot\nabla P_i\right). \tag*{(4)} p H = i = 1 ∑ N c ( N i p P i + M i p ⋅ ∇ P i ) . (4)
where U i U_i U i and P i P_i P i are the macroscopic displacement and pressure variables associated with the i i i th continuum. The functions N i u N_i^u N i u , M i u M_i^u M i u , N i p N_i^p N i p , and M i p M_i^p M i p denote the corresponding multicontinuum basis functions. These indices are suppressed for readability, and all products below denote the corresponding tensor contractions. In [14 ] , the multicontinuum expansion for the displacement has pressure terms and vice versa. The resulting macroscopic equations are the same. Since the goal of this paper is to analyze macroscopic equations, for simplicity, we consider split multicontinuum expansion (3) and (4) . The test functions are selected from the same approximation spaces:
v H = ∑ j = 1 N c ( N j u V j + M j u ⋅ ∇ V j ) , (5) v_H=\sum_{j=1}^{N_c}\left(N_j^u V_j+M_j^u\cdot\nabla V_j\right), \tag*{(5)} v H = j = 1 ∑ N c ( N j u V j + M j u ⋅ ∇ V j ) , (5)
q H = ∑ j = 1 N c ( N j p Q j + M j p ⋅ ∇ Q j ) . (6) q_H=\sum_{j=1}^{N_c}\left(N_j^p Q_j+M_j^p\cdot\nabla Q_j\right). \tag*{(6)} q H = j = 1 ∑ N c ( N j p Q j + M j p ⋅ ∇ Q j ) . (6)
The construction of the effective coefficients is performed locally on each Representative Volume Element (RVE) region R ω R_\omega R ω . Within R ω R_\omega R ω , the macroscopic variables and their first-order spatial derivatives are regarded as prescribed parameters, while the microscopic differential operators act on the corresponding local basis functions. Thus, the resulting expressions should be interpreted as a local first-order approximation of the multicontinuum representation, rather than as an exact substitution of spatially varying macroscopic fields into the microscopic differential operators. For a microscopic bilinear form a ( ⋅ , ⋅ ) a(\cdot,\cdot) a ( ⋅ , ⋅ ) , we denote its normalized restriction to R ω R_\omega R ω by
a R ω ( ϕ , ψ ) : = 1 ∣ R ω ∣ a ( ϕ , ψ ; R ω ) . a^{R_\omega}(\phi,\psi):=\frac{1}{\lvert R_\omega\rvert}a(\phi,\psi;R_\omega). a R ω ( ϕ , ψ ) := ∣ R ω ∣ 1 a ( ϕ , ψ ; R ω ) .
where a ( ϕ , ψ ; R ω ) a(\phi,\psi;R_\omega) a ( ϕ , ψ ; R ω ) denotes the contribution of the region R ω R_\omega R ω to the microscopic bilinear form. The other microscopic bilinear forms are defined similarly. The effective coefficients are then computed locally on each RVE region and subsequently interpreted as spatially varying macroscopic coefficient fields.
Effective operators
Substitution of the multicontinuum trial and test functions (3)-(6) into (2) yields a macroscopic system for ( U , P ) (U,P) ( U , P ) . We first give the effective coefficient blocks generated by the microscopic bilinear forms. The effective elastic bilinear form is
a u ( u H , v H ) = ∑ i , j = 1 N c ∫ Ω [ A i j u u ∇ U i ⋅ ∇ V j + B i j u u ∇ U i ⋅ V j + B ˉ i j u u U i ⋅ ∇ V j + C i j u u U i ⋅ V j ] d x a_u(u_H,v_H)=\sum_{i,j=1}^{N_c}\int_{\Omega}\left[A_{ij}^{uu}\nabla U_i\cdot\nabla V_j+B_{ij}^{uu}\nabla U_i\cdot V_j+\bar{B}_{ij}^{uu}U_i\cdot\nabla V_j+C_{ij}^{uu}U_i\cdot V_j\right]\,dx a u ( u H , v H ) = i , j = 1 ∑ N c ∫ Ω [ A ij uu ∇ U i ⋅ ∇ V j + B ij uu ∇ U i ⋅ V j + B ˉ ij uu U i ⋅ ∇ V j + C ij uu U i ⋅ V j ] d x
where
A i j u u = a u R ω ( M i u , M j u ) , B i j u u = a u R ω ( M i u , N j u ) , A_{ij}^{uu}=a_u^{R^\omega}(M_i^u,M_j^u),\quad B_{ij}^{uu}=a_u^{R^\omega}(M_i^u,N_j^u), A ij uu = a u R ω ( M i u , M j u ) , B ij uu = a u R ω ( M i u , N j u ) ,
B ˉ i j u u = a u R ω ( N i u , M j u ) , C i j u u = a u R ω ( N i u , N j u ) . \bar{B}_{ij}^{uu}=a_u^{R^\omega}(N_i^u,M_j^u),\quad C_{ij}^{uu}=a_u^{R^\omega}(N_i^u,N_j^u). B ˉ ij uu = a u R ω ( N i u , M j u ) , C ij uu = a u R ω ( N i u , N j u ) .
Evaluations involving tensor-valued basis functions are understood componentwise. Similarly, the effective diffusion form is
a p ( p H , q H ) = ∑ i , j = 1 N c ∫ Ω [ A i j p p ∇ P i ⋅ ∇ Q j + B i j p p ∇ P i ⋅ Q j + B ˉ i j p p P i ⋅ ∇ Q j + C i j p p P i Q j ] d x a_p(p_H,q_H)=\sum_{i,j=1}^{N_c}\int_{\Omega}\left[A_{ij}^{pp}\nabla P_i\cdot\nabla Q_j+B_{ij}^{pp}\nabla P_i\cdot Q_j+\bar{B}_{ij}^{pp}P_i\cdot\nabla Q_j+C_{ij}^{pp}P_iQ_j\right]\,dx a p ( p H , q H ) = i , j = 1 ∑ N c ∫ Ω [ A ij pp ∇ P i ⋅ ∇ Q j + B ij pp ∇ P i ⋅ Q j + B ˉ ij pp P i ⋅ ∇ Q j + C ij pp P i Q j ] d x
with
A i j p p = a p R ω ( M i p , M j p ) , B i j p p = a p R ω ( M i p , N j p ) , A_{ij}^{pp}=a_p^{R^\omega}(M_i^p,M_j^p),\quad B_{ij}^{pp}=a_p^{R^\omega}(M_i^p,N_j^p), A ij pp = a p R ω ( M i p , M j p ) , B ij pp = a p R ω ( M i p , N j p ) ,
B ˉ i j p p = a p R ω ( N i p , M j p ) , C i j p p = a p R ω ( N i p , N j p ) . \bar{B}_{ij}^{pp}=a_p^{R^\omega}(N_i^p,M_j^p),\quad C_{ij}^{pp}=a_p^{R^\omega}(N_i^p,N_j^p). B ˉ ij pp = a p R ω ( N i p , M j p ) , C ij pp = a p R ω ( N i p , N j p ) .
The effective storage form is
c ( p H , t , q H ) = ∑ i , j = 1 N c ∫ Ω [ H i j p p ∂ t P i Q j + G i j p p ⋅ ∇ ∂ t P i Q j + G ˉ i j p p ∂ t P i ⋅ ∇ Q j + D i j p p ∇ ∂ t P i ⋅ ∇ Q j ] d x c(p_{H,t},q_H)=\sum_{i,j=1}^{N_c}\int_{\Omega}\left[H_{ij}^{pp}\partial_tP_iQ_j+G_{ij}^{pp}\cdot\nabla\partial_tP_iQ_j+\bar{G}_{ij}^{pp}\partial_tP_i\cdot\nabla Q_j+D_{ij}^{pp}\nabla\partial_tP_i\cdot\nabla Q_j\right]\,dx c ( p H , t , q H ) = i , j = 1 ∑ N c ∫ Ω [ H ij pp ∂ t P i Q j + G ij pp ⋅ ∇ ∂ t P i Q j + G ˉ ij pp ∂ t P i ⋅ ∇ Q j + D ij pp ∇ ∂ t P i ⋅ ∇ Q j ] d x
where
H i j p p = c R ω ( N i p , N j p ) , G i j p p = c R ω ( M i p , N j p ) , H_{ij}^{pp}=c^{R^\omega}(N_i^p,N_j^p),\quad G_{ij}^{pp}=c^{R^\omega}(M_i^p,N_j^p), H ij pp = c R ω ( N i p , N j p ) , G ij pp = c R ω ( M i p , N j p ) ,
G ˉ i j p p = c R ω ( N i p , M j p ) , D i j p p = c R ω ( M i p , M j p ) . \bar{G}_{ij}^{pp}=c^{R^\omega}(N_i^p,M_j^p),\quad D_{ij}^{pp}=c^{R^\omega}(M_i^p,M_j^p). G ˉ ij pp = c R ω ( N i p , M j p ) , D ij pp = c R ω ( M i p , M j p ) .
The coupling form is
b ( p H , v H ) = ∑ i , j = 1 N c ∫ Ω [ A i j p u ∇ P i ⋅ ∇ V j + B i j p u ∇ P i ⋅ V j + B ˉ i j p u P i ⋅ ∇ V j + C i j p u P i ⋅ V j ] d x b(p_H,v_H)=\sum_{i,j=1}^{N_c}\int_{\Omega}\left[A_{ij}^{pu}\nabla P_i\cdot\nabla V_j+B_{ij}^{pu}\nabla P_i\cdot V_j+\bar{B}_{ij}^{pu}P_i\cdot\nabla V_j+C_{ij}^{pu}P_i\cdot V_j\right]\,dx b ( p H , v H ) = i , j = 1 ∑ N c ∫ Ω [ A ij p u ∇ P i ⋅ ∇ V j + B ij p u ∇ P i ⋅ V j + B ˉ ij p u P i ⋅ ∇ V j + C ij p u P i ⋅ V j ] d x
with
A i j p u = b R ω ( M i p , M j u ) , B i j p u = b R ω ( M i p , N j u ) , A_{ij}^{pu}=b^{R^\omega}(M_i^p,M_j^u),\quad B_{ij}^{pu}=b^{R^\omega}(M_i^p,N_j^u), A ij p u = b R ω ( M i p , M j u ) , B ij p u = b R ω ( M i p , N j u ) ,
B ˉ i j p u = b R ω ( N i p , M j u ) , C i j p u = b R ω ( N i p , N j u ) . \bar{B}_{ij}^{pu}=b^{R^\omega}(N_i^p,M_j^u),\quad C_{ij}^{pu}=b^{R^\omega}(N_i^p,N_j^u). B ˉ ij p u = b R ω ( N i p , M j u ) , C ij p u = b R ω ( N i p , N j u ) .
The gradient-gradient coefficient A i j p u A^{pu}_{ij} A ij p u is included because it generally arises from the local projection. Using the same coupling form in the pressure equation gives
b ( q H , u H , t ) = ∑ i , j = 1 N c ∫ Ω [ A i j u p ∇ ∂ t U i ⋅ ∇ Q j + G i j u p ∇ ∂ t U i Q j + G ˉ i j u p ∂ t U i ⋅ ∇ Q j + E i j u p ∂ t U i Q j ] d x . b(q_H,u_{H,t})=\sum_{i,j=1}^{N_c}\int_{\Omega}\left[A^{up}_{ij}\nabla\partial_tU_i\cdot\nabla Q_j+G^{up}_{ij}\nabla\partial_tU_iQ_j+\bar{G}^{up}_{ij}\partial_tU_i\cdot\nabla Q_j+E^{up}_{ij}\partial_tU_iQ_j\right]\,dx. b ( q H , u H , t ) = i , j = 1 ∑ N c ∫ Ω [ A ij u p ∇ ∂ t U i ⋅ ∇ Q j + G ij u p ∇ ∂ t U i Q j + G ˉ ij u p ∂ t U i ⋅ ∇ Q j + E ij u p ∂ t U i Q j ] d x .
where
A i j u p = b R ω ( M j p , M i u ) , G i j u p = b R ω ( N j p , M i u ) , A^{up}_{ij}=b^{R_\omega}(M^p_j,M^u_i),\qquad G^{up}_{ij}=b^{R_\omega}(N^p_j,M^u_i), A ij u p = b R ω ( M j p , M i u ) , G ij u p = b R ω ( N j p , M i u ) ,
G ˉ i j u p = b R ω ( M j p , N i u ) , E i j u p = b R ω ( N j p , N i u ) . \bar{G}^{up}_{ij}=b^{R_\omega}(M^p_j,N^u_i),\qquad E^{up}_{ij}=b^{R_\omega}(N^p_j,N^u_i). G ˉ ij u p = b R ω ( M j p , N i u ) , E ij u p = b R ω ( N j p , N i u ) .
Similarly, the source coefficients are
F j u = l R ω ( N j u ) + l R ω ( M j u ) , F j p = l R ω ( N j p ) + l R ω ( M j p ) . F^u_j=l^{R_\omega}(N^u_j)+l^{R_\omega}(M^u_j),\qquad F^p_j=l^{R_\omega}(N^p_j)+l^{R_\omega}(M^p_j). F j u = l R ω ( N j u ) + l R ω ( M j u ) , F j p = l R ω ( N j p ) + l R ω ( M j p ) .
The effective coefficients inherit the symmetry and reciprocity properties of the underlying microscopic bilinear forms. In particular, the elastic and diffusion operators satisfy
A u u = ( A u u ) T , C u u = ( C u u ) T , B ˉ u u = ( B u u ) T , A^{uu}=(A^{uu})^T,\qquad C^{uu}=(C^{uu})^T,\qquad\bar{B}^{uu}=(B^{uu})^T, A uu = ( A uu ) T , C uu = ( C uu ) T , B ˉ uu = ( B uu ) T ,
A p p = ( A p p ) T , C p p = ( C p p ) T , B ˉ p p = ( B p p ) T . A^{pp}=(A^{pp})^T,\qquad C^{pp}=(C^{pp})^T,\qquad\bar{B}^{pp}=(B^{pp})^T. A pp = ( A pp ) T , C pp = ( C pp ) T , B ˉ pp = ( B pp ) T .
Similarly, the storage coefficients satisfy
H p p = ( H p p ) T , D p p = ( D p p ) T , G ˉ p p = ( G p p ) T . H^{pp}=(H^{pp})^T,\qquad D^{pp}=(D^{pp})^T,\qquad\bar{G}^{pp}=(G^{pp})^T. H pp = ( H pp ) T , D pp = ( D pp ) T , G ˉ pp = ( G pp ) T .
The coupling coefficients obey the reciprocity relations
A u p = ( A p u ) T , G u p = ( B ˉ p u ) T , A^{up}=(A^{pu})^T,\qquad G^{up}=(\bar{B}^{pu})^T, A u p = ( A p u ) T , G u p = ( B ˉ p u ) T ,
G ˉ u p = ( B p u ) T , E u p = ( C p u ) T . \bar{G}^{up}=(B^{pu})^T,\qquad E^{up}=(C^{pu})^T. G ˉ u p = ( B p u ) T , E u p = ( C p u ) T .
Here, the transpose exchanges the corresponding trial and test component indices. These identities follow directly from the symmetry of a u R ω a^{R_\omega}_u a u R ω , a p R ω a^{R_\omega}_p a p R ω , and c R ω c^{R_\omega} c R ω , together with the fact that the two sets of coupling coefficients are generated from the same microscopic coupling bilinear form b R ω b^{R_\omega} b R ω .
Final multicontinuum equations
Using the effective coefficients in Section 2.3, we can obtain the multicontinuum equations
− ∇ ⋅ ( A u u ∇ U ) + B u u ∇ U − ∇ ⋅ ( B ˉ u u U ) + C u u U + ∇ ⋅ ( A p u ∇ P ) − B p u ∇ P + ∇ ⋅ ( B ˉ p u P ) − C p u P = F u , H p p ∂ t P + G p p ∇ ∂ t P − ∇ ⋅ ( G ˉ p p ∂ t P ) − ∇ ⋅ ( D p p ∇ ∂ t P ) − ∇ ⋅ ( A u p ∇ ∂ t U ) + G u p ∇ ∂ t U − ∇ ⋅ ( G ˉ u p ∂ t U ) + E u p ∂ t U − ∇ ⋅ ( A p p ∇ P ) + B p p ∇ P − ∇ ⋅ ( B ˉ p p P ) + C p p P = F p . (7) \begin{aligned}
&-\nabla\cdot(A^{uu}\nabla U)+B^{uu}\nabla U-\nabla\cdot(\bar{B}^{uu}U)+C^{uu}U \\
&\quad+\nabla\cdot(A^{pu}\nabla P)-B^{pu}\nabla P+\nabla\cdot(\bar{B}^{pu}P)-C^{pu}P=F_u,\\
&H^{pp}\partial_tP+G^{pp}\nabla\partial_tP-\nabla\cdot(\bar{G}^{pp}\partial_tP)-\nabla\cdot(D^{pp}\nabla\partial_tP)\\
&\quad-\nabla\cdot(A^{up}\nabla\partial_tU)+G^{up}\nabla\partial_tU-\nabla\cdot(\bar{G}^{up}\partial_tU)+E^{up}\partial_tU\\
&\quad-\nabla\cdot(A^{pp}\nabla P)+B^{pp}\nabla P-\nabla\cdot(\bar{B}^{pp}P)+C^{pp}P=F_p. \tag*{(7)}
\end{aligned} − ∇ ⋅ ( A uu ∇ U ) + B uu ∇ U − ∇ ⋅ ( B ˉ uu U ) + C uu U + ∇ ⋅ ( A p u ∇ P ) − B p u ∇ P + ∇ ⋅ ( B ˉ p u P ) − C p u P = F u , H pp ∂ t P + G pp ∇ ∂ t P − ∇ ⋅ ( G ˉ pp ∂ t P ) − ∇ ⋅ ( D pp ∇ ∂ t P ) − ∇ ⋅ ( A u p ∇ ∂ t U ) + G u p ∇ ∂ t U − ∇ ⋅ ( G ˉ u p ∂ t U ) + E u p ∂ t U − ∇ ⋅ ( A pp ∇ P ) + B pp ∇ P − ∇ ⋅ ( B ˉ pp P ) + C pp P = F p . (7)
where U = ( U 1 , … , U N c ) T U=(U_1,\ldots,U_{N_c})^T U = ( U 1 , … , U N c ) T , P = ( P 1 , … , P N c ) T P=(P_1,\ldots,P_{N_c})^T P = ( P 1 , … , P N c ) T .
We impose
U = 0 , P = 0 on ∂ Ω × J , P ( x , 0 ) = P 0 ( x ) , U=0,\qquad P=0\quad\text{on }\partial\Omega\times J,\qquad P(x,0)=P_0(x), U = 0 , P = 0 on ∂ Ω × J , P ( x , 0 ) = P 0 ( x ) ,
and some initial conditions.
Based on the numerical results in [14 ] , we can neglect some terms,
A p u = A u p = 0 , G p p = G ‾ p p = 0 , D p p = 0. A^{pu}=A^{up}=0,\qquad G^{pp}=\overline{G}^{pp}=0,\qquad D^{pp}=0. A p u = A u p = 0 , G pp = G pp = 0 , D pp = 0.
These are explicit modeling approximations and the corresponding coefficients need not vanish in the full local projection. The multicontinuum equations (7) can be written as
− ∇ ⋅ ( A u u ∇ U ) + B u u ∇ U − ∇ ⋅ ( B ‾ u u U ) + C u u U − B p u ∇ P + ∇ ⋅ ( B ‾ p u P ) − C p u P = F u , H p p ∂ t P + G u p ∇ ∂ t U − ∇ ⋅ ( G ‾ u p ∂ t U ) + E u p ∂ t U − ∇ ⋅ ( A p p ∇ P ) + B p p ∇ P − ∇ ⋅ ( B ‾ p p P ) + C p p P = F p . (8) \begin{aligned}
-\nabla\cdot\left(A^{uu}\nabla U\right)+B^{uu}\nabla U-\nabla\cdot\left(\overline{B}^{uu}U\right)+C^{uu}U
-B^{pu}\nabla P+\nabla\cdot\left(\overline{B}^{pu}P\right)-C^{pu}P&=F_u,\\
H^{pp}\partial_tP+G^{up}\nabla\partial_tU-\nabla\cdot\left(\overline{G}^{up}\partial_tU\right)+E^{up}\partial_tU\\
-\nabla\cdot\left(A^{pp}\nabla P\right)+B^{pp}\nabla P-\nabla\cdot\left(\overline{B}^{pp}P\right)+C^{pp}P&=F_p. \tag*{(8)}
\end{aligned} − ∇ ⋅ ( A uu ∇ U ) + B uu ∇ U − ∇ ⋅ ( B uu U ) + C uu U − B p u ∇ P + ∇ ⋅ ( B p u P ) − C p u P H pp ∂ t P + G u p ∇ ∂ t U − ∇ ⋅ ( G u p ∂ t U ) + E u p ∂ t U − ∇ ⋅ ( A pp ∇ P ) + B pp ∇ P − ∇ ⋅ ( B pp P ) + C pp P = F u , = F p . (8)
Assumption 1. We present the Assumption as follows.
(A1) All tensor-valued coefficients are essentially bounded:
A u u , B u u , B ‾ u u , C u u , A p u , B p u , B ‾ p u , C p u , A u p , G u p , G ‾ u p , E u p , A^{uu}, B^{uu}, \overline{B}^{uu}, C^{uu}, A^{pu}, B^{pu}, \overline{B}^{pu}, C^{pu}, A^{up}, G^{up}, \overline{G}^{up}, E^{up}, A uu , B uu , B uu , C uu , A p u , B p u , B p u , C p u , A u p , G u p , G u p , E u p , A p p , B p p , B ‾ p p , C p p , G p p , G ‾ p p , D p p , H p p ∈ L ∞ ( Ω ) . A^{pp}, B^{pp}, \overline{B}^{pp}, C^{pp}, G^{pp}, \overline{G}^{pp}, D^{pp}, H^{pp}\in L^\infty(\Omega). A pp , B pp , B pp , C pp , G pp , G pp , D pp , H pp ∈ L ∞ ( Ω ) . Moreover, the effective coefficients satisfy, almost everywhere in Ω \Omega Ω ,
A u u = ( A u u ) T , C u u = ( C u u ) T , B ‾ u u = ( B u u ) T , A p p = ( A p p ) T , C p p = ( C p p ) T , B ‾ p p = ( B p p ) T , H p p = ( H p p ) T , D p p = ( D p p ) T , G ‾ p p = ( G p p ) T , A u p = ( A p u ) T , G u p = ( B ‾ p u ) T , G ‾ u p = ( B p u ) T , E u p = ( C p u ) T . \begin{aligned}
A^{uu}&=(A^{uu})^T,& C^{uu}&=(C^{uu})^T,& \overline{B}^{uu}&=(B^{uu})^T,\\
A^{pp}&=(A^{pp})^T,& C^{pp}&=(C^{pp})^T,& \overline{B}^{pp}&=(B^{pp})^T,\\
H^{pp}&=(H^{pp})^T,& D^{pp}&=(D^{pp})^T,& \overline{G}^{pp}&=(G^{pp})^T,\\
A^{up}&=(A^{pu})^T,& G^{up}&=(\overline{B}^{pu})^T,& \overline{G}^{up}&=(B^{pu})^T,& E^{up}&=(C^{pu})^T.
\end{aligned} A uu A pp H pp A u p = ( A uu ) T , = ( A pp ) T , = ( H pp ) T , = ( A p u ) T , C uu C pp D pp G u p = ( C uu ) T , = ( C pp ) T , = ( D pp ) T , = ( B p u ) T , B uu B pp G pp G u p = ( B uu ) T , = ( B pp ) T , = ( G pp ) T , = ( B p u ) T , E u p = ( C p u ) T . The transpose exchanges the complete trial and test index groups, including the continuum, component, and gradient-direction indices, whenever applicable.
All effective coefficients are assumed to be independent of time.
(A2) The tensors A u u ( x ) A^{uu}(x) A uu ( x ) , A p p ( x ) A^{pp}(x) A pp ( x ) , H p p ( x ) H^{pp}(x) H pp ( x ) , C u u ( x ) C^{uu}(x) C uu ( x ) , C p p ( x ) C^{pp}(x) C pp ( x ) , and D p p ( x ) D^{pp}(x) D pp ( x ) are uniformly positive definite a.e. in Ω \Omega Ω . In particular, there exist constants α u u , α p p , α H , α c u , α c p , α D > 0 \alpha_{uu},\alpha_{pp},\alpha_H,\alpha_{cu},\alpha_{cp},\alpha_D>0 α uu , α pp , α H , α c u , α c p , α D > 0 such that
ξ T A u u ( x ) ξ ≥ α u u ∣ ξ ∣ 2 , ∀ ξ ∈ R d 2 N c , \xi^T A^{uu}(x)\xi\ge\alpha_{uu}|\xi|^2,\qquad\forall\xi\in\mathbb{R}^{d^2N_c}, ξ T A uu ( x ) ξ ≥ α uu ∣ ξ ∣ 2 , ∀ ξ ∈ R d 2 N c , η T A p p ( x ) η ≥ α p p ∣ η ∣ 2 , ∀ η ∈ R d N c , \eta^T A^{pp}(x)\eta\ge\alpha_{pp}|\eta|^2,\qquad\forall\eta\in\mathbb{R}^{dN_c}, η T A pp ( x ) η ≥ α pp ∣ η ∣ 2 , ∀ η ∈ R d N c , ζ T H p p ( x ) ζ ≥ α H ∣ ζ ∣ 2 , ∀ ζ ∈ R N c , \zeta^T H^{pp}(x)\zeta\ge\alpha_H|\zeta|^2,\qquad\forall\zeta\in\mathbb{R}^{N_c}, ζ T H pp ( x ) ζ ≥ α H ∣ ζ ∣ 2 , ∀ ζ ∈ R N c , v T C u u ( x ) v ≥ α c u ∣ v ∣ 2 , ∀ v ∈ R d N c , v^T C^{uu}(x)v\ge\alpha_{cu}|v|^2,\qquad\forall v\in\mathbb{R}^{dN_c}, v T C uu ( x ) v ≥ α c u ∣ v ∣ 2 , ∀ v ∈ R d N c , ζ T C p p ( x ) ζ ≥ α c p ∣ ζ ∣ 2 , ∀ ζ ∈ R N c , \zeta^T C^{pp}(x)\zeta\ge\alpha_{cp}|\zeta|^2,\qquad\forall\zeta\in\mathbb{R}^{N_c}, ζ T C pp ( x ) ζ ≥ α c p ∣ ζ ∣ 2 , ∀ ζ ∈ R N c , η T D p p ( x ) η ≥ α D ∣ η ∣ 2 , ∀ η ∈ R d N c . \eta^T D^{pp}(x)\eta\ge\alpha_D|\eta|^2,\qquad\forall\eta\in\mathbb{R}^{dN_c}. η T D pp ( x ) η ≥ α D ∣ η ∣ 2 , ∀ η ∈ R d N c . (A3) There exist constants β u u , β p p , β g p ≥ 0 \beta_{uu},\beta_{pp},\beta_{gp}\ge0 β uu , β pp , β g p ≥ 0 such that, for all U ∈ V u U\in V_u U ∈ V u and P ∈ V p P\in V_p P ∈ V p ,
2 ∣ ( B u u ∇ U , U ) ∣ ≤ β u u ( ∥ ∇ U ∥ 2 + ∥ U ∥ 2 ) , 2\left|\left(B^{uu}\nabla U,U\right)\right|\le\beta_{uu}\left(\|\nabla U\|^2+\|U\|^2\right), 2 ∣ ( B uu ∇ U , U ) ∣ ≤ β uu ( ∥∇ U ∥ 2 + ∥ U ∥ 2 ) , 2 ∣ ( B p p ∇ P , P ) ∣ ≤ β p p ( ∥ ∇ P ∥ 2 + ∥ P ∥ 2 ) , 2\left|\left(B^{pp}\nabla P,P\right)\right|\le\beta_{pp}\left(\|\nabla P\|^2+\|P\|^2\right), 2 ∣ ( B pp ∇ P , P ) ∣ ≤ β pp ( ∥∇ P ∥ 2 + ∥ P ∥ 2 ) , 2 ∣ ( G p p ∇ P , P ) ∣ ≤ β g p ( ∥ ∇ P ∥ 2 + ∥ P ∥ 2 ) , 2\left|\left(G^{pp}\nabla P,P\right)\right|\le\beta_{gp}\left(\|\nabla P\|^2+\|P\|^2\right), 2 ∣ ( G pp ∇ P , P ) ∣ ≤ β g p ( ∥∇ P ∥ 2 + ∥ P ∥ 2 ) , where
β u u < min { α u u , α c u } , β p p < min { α p p , α c p } , β g p < min { α H , α D } . \beta_{uu}<\min\{\alpha_{uu},\alpha_{cu}\},\qquad\beta_{pp}<\min\{\alpha_{pp},\alpha_{cp}\},\qquad\beta_{gp}<\min\{\alpha_H,\alpha_D\}. β uu < min { α uu , α c u } , β pp < min { α pp , α c p } , β g p < min { α H , α D } .
(A4) The source terms satisfy
F u ∈ L ∞ ( 0 , T ; L 2 ) d N c , ∂ t F u ∈ L 2 ( 0 , T ; L 2 ) d N c , F p ∈ L 2 ( 0 , T ; L 2 ) , P 0 ∈ V p . F_u \in L^\infty(0,T;L^2)^{dN_c}, \qquad\partial_t F_u \in L^2(0,T;L^2)^{dN_c}, \qquad F_p \in L^2(0,T;L^2), \qquad P_0 \in V_p. F u ∈ L ∞ ( 0 , T ; L 2 ) d N c , ∂ t F u ∈ L 2 ( 0 , T ; L 2 ) d N c , F p ∈ L 2 ( 0 , T ; L 2 ) , P 0 ∈ V p .
In particular, these conditions imply
F u ∈ H 1 ( 0 , T ; V u ′ ) , F p ∈ L 2 ( 0 , T ; V p ′ ) . F_u \in H^1(0,T;V_u'), \qquad F_p \in L^2(0,T;V_p'). F u ∈ H 1 ( 0 , T ; V u ′ ) , F p ∈ L 2 ( 0 , T ; V p ′ ) .
(A5) We assume that the exact solution satisfies
U ∈ H 1 ( 0 , T ; H k + 1 ( Ω ) ) d N c ∩ H 2 ( 0 , T ; V u ) , U \in H^1(0,T;H^{k+1}(\Omega))^{dN_c} \cap H^2(0,T;V_u), U ∈ H 1 ( 0 , T ; H k + 1 ( Ω ) ) d N c ∩ H 2 ( 0 , T ; V u ) ,
P ∈ H 1 ( 0 , T ; H k + 1 ( Ω ) ) N c ∩ H 2 ( 0 , T ; V p ) . P \in H^1(0,T;H^{k+1}(\Omega))^{N_c} \cap H^2(0,T;V_p). P ∈ H 1 ( 0 , T ; H k + 1 ( Ω ) ) N c ∩ H 2 ( 0 , T ; V p ) .
Remark 1. For the simplified model (8), we set
A p u = A u p = 0 , G p p = G ‾ p p = 0 , D p p = 0. A^{pu}=A^{up}=0, \qquad G^{pp}=\overline{G}^{pp}=0, \qquad D^{pp}=0. A p u = A u p = 0 , G pp = G pp = 0 , D pp = 0.
The conditions on these coefficients in (A1) are then automatic. The positive definiteness requirement for D p p D^{pp} D pp in (A2) , and the estimate involving G p p G^{pp} G pp together with the restriction on β g p \beta_{gp} β g p in (A3) , are removed. All remaining assumptions are retained.
Weak formulation
In this subsection, we present the weak formulation of the multicontinuum model (7).
Define
V u = [ H 0 1 ( Ω ) ] d N c , V p = [ H 0 1 ( Ω ) ] N c , V = V u × V p , V_u=[H_0^1(\Omega)]^{dN_c}, \qquad V_p=[H_0^1(\Omega)]^{N_c}, \qquad V=V_u\times V_p, V u = [ H 0 1 ( Ω ) ] d N c , V p = [ H 0 1 ( Ω ) ] N c , V = V u × V p ,
∥ U ∥ V u 2 = ∥ ∇ U ∥ L 2 2 + ∥ U ∥ L 2 2 , ∥ P ∥ V p 2 = ∥ ∇ P ∥ L 2 2 + ∥ P ∥ L 2 2 . \lVert U\rVert_{V_u}^2=\lVert\nabla U\rVert_{L^2}^2+\lVert U\rVert_{L^2}^2, \qquad\lVert P\rVert_{V_p}^2=\lVert\nabla P\rVert_{L^2}^2+\lVert P\rVert_{L^2}^2. ∥ U ∥ V u 2 = ∥ ∇ U ∥ L 2 2 + ∥ U ∥ L 2 2 , ∥ P ∥ V p 2 = ∥ ∇ P ∥ L 2 2 + ∥ P ∥ L 2 2 .
For subsequent analysis, define the bilinear forms associated with the full model (7) by
A u ( U , V ) = ( A u u ∇ U , ∇ V ) + ( B u u ∇ U , V ) + ( B ‾ u u U , ∇ V ) + ( C u u U , V ) , A p ( P , Q ) = ( A p p ∇ P , ∇ Q ) + ( B p p ∇ P , Q ) + ( B ‾ p p P , ∇ Q ) + ( C p p P , Q ) , B u p ( P , V ) = ( A p u ∇ P , ∇ V ) + ( B p u ∇ P , V ) + ( B ‾ p u P , ∇ V ) + ( C p u P , V ) , B p u ( Q , U ) = ( A u p ∇ U , ∇ Q ) + ( G u p ∇ U , Q ) + ( G ‾ u p U , ∇ Q ) + ( E u p U , Q ) , M p ( P , Q ) = ( H p p P , Q ) + ( G p p ∇ P , Q ) + ( G ‾ p p P , ∇ Q ) + ( D p p ∇ P , ∇ Q ) . \begin{aligned}
A_u(U,V) &= (A^{uu}\nabla U,\nabla V)+(B^{uu}\nabla U,V)+(\overline{B}^{uu}U,\nabla V)+(C^{uu}U,V),\\
A_p(P,Q) &= (A^{pp}\nabla P,\nabla Q)+(B^{pp}\nabla P,Q)+(\overline{B}^{pp}P,\nabla Q)+(C^{pp}P,Q),\\
B_{up}(P,V) &= (A^{pu}\nabla P,\nabla V)+(B^{pu}\nabla P,V)+(\overline{B}^{pu}P,\nabla V)+(C^{pu}P,V),\\
B_{pu}(Q,U) &= (A^{up}\nabla U,\nabla Q)+(G^{up}\nabla U,Q)+(\overline{G}^{up}U,\nabla Q)+(E^{up}U,Q),\\
M_p(P,Q) &= (H^{pp}P,Q)+(G^{pp}\nabla P,Q)+(\overline{G}^{pp}P,\nabla Q)+(D^{pp}\nabla P,\nabla Q).
\end{aligned} A u ( U , V ) A p ( P , Q ) B u p ( P , V ) B p u ( Q , U ) M p ( P , Q ) = ( A uu ∇ U , ∇ V ) + ( B uu ∇ U , V ) + ( B uu U , ∇ V ) + ( C uu U , V ) , = ( A pp ∇ P , ∇ Q ) + ( B pp ∇ P , Q ) + ( B pp P , ∇ Q ) + ( C pp P , Q ) , = ( A p u ∇ P , ∇ V ) + ( B p u ∇ P , V ) + ( B p u P , ∇ V ) + ( C p u P , V ) , = ( A u p ∇ U , ∇ Q ) + ( G u p ∇ U , Q ) + ( G u p U , ∇ Q ) + ( E u p U , Q ) , = ( H pp P , Q ) + ( G pp ∇ P , Q ) + ( G pp P , ∇ Q ) + ( D pp ∇ P , ∇ Q ) .
By the symmetry and reciprocity relations in Assumption 1, we have
B u p ( P , V ) = B p u ( V , P ) , P ∈ V p , V ∈ V u . B_{up}(P,V)=B_{pu}(V,P), \qquad P\in V_p,\quad V\in V_u. B u p ( P , V ) = B p u ( V , P ) , P ∈ V p , V ∈ V u .
Therefore, for notational convenience, define B ( P , V ) : = B u p ( P , V ) = B p u ( P , V ) B(P,V):=B_{up}(P,V)=B_{pu}(P,V) B ( P , V ) := B u p ( P , V ) = B p u ( P , V ) .
The weak formulation of (7) is: find ( U , P ) ∈ V (U,P)\in V ( U , P ) ∈ V with the required time regularity such that, for all ( V , Q ) ∈ V (V,Q)\in V ( V , Q ) ∈ V and almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) ,
A u ( U , V ) − B ( P , V ) = ( F u , V ) , M p ( ∂ t P , Q ) + B ( Q , ∂ t U ) + A p ( P , Q ) = ( F p , Q ) . (9) \begin{aligned}
A_u(U,V)-B(P,V)&=(F_u,V),\\
M_p(\partial_tP,Q)+B(Q,\partial_tU)+A_p(P,Q)&=(F_p,Q). \tag*{(9)}
\end{aligned} A u ( U , V ) − B ( P , V ) M p ( ∂ t P , Q ) + B ( Q , ∂ t U ) + A p ( P , Q ) = ( F u , V ) , = ( F p , Q ) . (9)
The initial displacement U 0 ∈ V u U_0\in V_u U 0 ∈ V u is defined by
A u ( U 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) , V ∈ V u . (10) A_u(U_0,V)-B(P_0,V)=(F_u(0),V), \qquad V\in V_u. \tag*{(10)} A u ( U 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) , V ∈ V u . (10)
Equivalently, in fully expanded form,
( A u u ∇ U , ∇ V ) + ( B u u ∇ U , V ) + ( B ˉ u u U , ∇ V ) + ( C u u U , V ) − ( A p u ∇ P , ∇ V ) − ( B p u ∇ P , V ) − ( B ˉ p u P , ∇ V ) − ( C p u P , V ) = ( F u , V ) , ( H p p ∂ t P , Q ) + ( G p p ∇ ∂ t P , Q ) + ( G ˉ p p ∂ t P , ∇ Q ) + ( D p p ∇ ∂ t P , ∇ Q ) + ( A u p ∇ ∂ t U , ∇ Q ) + ( G u p ∇ ∂ t U , Q ) + ( G ˉ u p ∂ t U , ∇ Q ) + ( E u p ∂ t U , Q ) + ( A p p ∇ P , ∇ Q ) + ( B p p ∇ P , Q ) + ( B ˉ p p P , ∇ Q ) + ( C p p P , Q ) = ( F p , Q ) . \begin{aligned}
&(A^{uu}\nabla U,\nabla V)+(B^{uu}\nabla U,V)+(\bar{B}^{uu}U,\nabla V)+(C^{uu}U,V)\\
&\quad-(A^{pu}\nabla P,\nabla V)-(B^{pu}\nabla P,V)-(\bar{B}^{pu}P,\nabla V)-(C^{pu}P,V)=(F_u,V),\\
&(H^{pp}\partial_tP,Q)+(G^{pp}\nabla\partial_tP,Q)+(\bar{G}^{pp}\partial_tP,\nabla Q)+(D^{pp}\nabla\partial_tP,\nabla Q)\\
&\quad+(A^{up}\nabla\partial_tU,\nabla Q)+(G^{up}\nabla\partial_tU,Q)+(\bar{G}^{up}\partial_tU,\nabla Q)+(E^{up}\partial_tU,Q)\\
&\quad+(A^{pp}\nabla P,\nabla Q)+(B^{pp}\nabla P,Q)+(\bar{B}^{pp}P,\nabla Q)+(C^{pp}P,Q)=(F_p,Q).
\end{aligned} ( A uu ∇ U , ∇ V ) + ( B uu ∇ U , V ) + ( B ˉ uu U , ∇ V ) + ( C uu U , V ) − ( A p u ∇ P , ∇ V ) − ( B p u ∇ P , V ) − ( B ˉ p u P , ∇ V ) − ( C p u P , V ) = ( F u , V ) , ( H pp ∂ t P , Q ) + ( G pp ∇ ∂ t P , Q ) + ( G ˉ pp ∂ t P , ∇ Q ) + ( D pp ∇ ∂ t P , ∇ Q ) + ( A u p ∇ ∂ t U , ∇ Q ) + ( G u p ∇ ∂ t U , Q ) + ( G ˉ u p ∂ t U , ∇ Q ) + ( E u p ∂ t U , Q ) + ( A pp ∇ P , ∇ Q ) + ( B pp ∇ P , Q ) + ( B ˉ pp P , ∇ Q ) + ( C pp P , Q ) = ( F p , Q ) .
Fully discrete scheme
We assume a uniform time partition t n = n τ t_n=n\tau t n = n τ , n = 0 , … , N n=0,\ldots,N n = 0 , … , N , τ = T / N \tau=T/N τ = T / N . For any function z z z , set
D t z n + 1 = z n + 1 − z n τ , D_tz^{n+1}=\frac{z^{n+1}-z^n}{\tau}, D t z n + 1 = τ z n + 1 − z n ,
where z n = z ( t n ) z^n=z(t_n) z n = z ( t n ) .
Let V u , h ⊂ V u V_{u,h}\subset V_u V u , h ⊂ V u and V p , h ⊂ V p V_{p,h}\subset V_p V p , h ⊂ V p be conforming degree-k k k Lagrange finite element spaces, with k ≥ 1 k\ge1 k ≥ 1 .
For the fully discrete scheme, we use the backward Euler method for time discretization. Given ( U h n , P h n ) (U_h^n,P_h^n) ( U h n , P h n ) , find ( U h n + 1 , P h n + 1 ) ∈ V u , h × V p , h (U_h^{n+1},P_h^{n+1})\in V_{u,h}\times V_{p,h} ( U h n + 1 , P h n + 1 ) ∈ V u , h × V p , h such that, for all ( V h , Q h ) ∈ V u , h × V p , h (V_h,Q_h)\in V_{u,h}\times V_{p,h} ( V h , Q h ) ∈ V u , h × V p , h ,
A u ( U h n + 1 , V h ) − B ( P h n + 1 , V h ) = ( F u n + 1 , V h ) , \mathcal{A}_u(U_h^{n+1},V_h)-\mathcal{B}(P_h^{n+1},V_h)=(F_u^{n+1},V_h), A u ( U h n + 1 , V h ) − B ( P h n + 1 , V h ) = ( F u n + 1 , V h ) ,
M p ( D t P h n + 1 , Q h ) + B ( Q h , D t U h n + 1 ) + A p ( P h n + 1 , Q h ) = ( F p n + 1 , Q h ) . (11) \mathcal{M}_p(D_tP_h^{n+1},Q_h)+\mathcal{B}(Q_h,D_tU_h^{n+1})+\mathcal{A}_p(P_h^{n+1},Q_h)=(F_p^{n+1},Q_h). \tag*{(11)} M p ( D t P h n + 1 , Q h ) + B ( Q h , D t U h n + 1 ) + A p ( P h n + 1 , Q h ) = ( F p n + 1 , Q h ) . (11)
Equivalently, in fully expanded form,
( A u u ∇ U h n + 1 , ∇ V h ) + ( B u u ∇ U h n + 1 , V h ) + ( B ˉ u u U h n + 1 , ∇ V h ) + ( C u u U h n + 1 , V h ) − ( A p u ∇ P h n + 1 , ∇ V h ) − ( B p u ∇ P h n + 1 , V h ) − ( B ˉ p u P h n + 1 , ∇ V h ) − ( C p u P h n + 1 , V h ) = ( F u n + 1 , V h ) , \begin{aligned}
&(A^{uu}\nabla U_h^{n+1},\nabla V_h)+(B^{uu}\nabla U_h^{n+1},V_h)+(\bar{B}^{uu}U_h^{n+1},\nabla V_h)+(C^{uu}U_h^{n+1},V_h)\\
&\quad-(A^{pu}\nabla P_h^{n+1},\nabla V_h)-(B^{pu}\nabla P_h^{n+1},V_h)-(\bar{B}^{pu}P_h^{n+1},\nabla V_h)-(C^{pu}P_h^{n+1},V_h)\\
&\quad=(F_u^{n+1},V_h),
\end{aligned} ( A uu ∇ U h n + 1 , ∇ V h ) + ( B uu ∇ U h n + 1 , V h ) + ( B ˉ uu U h n + 1 , ∇ V h ) + ( C uu U h n + 1 , V h ) − ( A p u ∇ P h n + 1 , ∇ V h ) − ( B p u ∇ P h n + 1 , V h ) − ( B ˉ p u P h n + 1 , ∇ V h ) − ( C p u P h n + 1 , V h ) = ( F u n + 1 , V h ) ,
( H p p D t P h n + 1 , Q h ) + ( G p p ∇ D t P h n + 1 , Q h ) + ( G ˉ p p D t P h n + 1 , ∇ Q h ) + ( D p p ∇ D t P h n + 1 , ∇ Q h ) + ( A u p ∇ D t U h n + 1 , ∇ Q h ) + ( G u p ∇ D t U h n + 1 , Q h ) + ( G ˉ u p D t U h n + 1 , ∇ Q h ) + ( E u p D t U h n + 1 , Q h ) + ( A p p ∇ P h n + 1 , ∇ Q h ) + ( B p p ∇ P h n + 1 , Q h ) + ( B ˉ p p P h n + 1 , ∇ Q h ) + ( C p p P h n + 1 , Q h ) = ( F p n + 1 , Q h ) . \begin{aligned}
&(H^{pp}D_tP_h^{n+1},Q_h)+(G^{pp}\nabla D_tP_h^{n+1},Q_h)+(\bar{G}^{pp}D_tP_h^{n+1},\nabla Q_h)+(D^{pp}\nabla D_tP_h^{n+1},\nabla Q_h)\\
&\quad+(A^{up}\nabla D_tU_h^{n+1},\nabla Q_h)+(G^{up}\nabla D_tU_h^{n+1},Q_h)+(\bar{G}^{up}D_tU_h^{n+1},\nabla Q_h)+(E^{up}D_tU_h^{n+1},Q_h)\\
&\quad+(A^{pp}\nabla P_h^{n+1},\nabla Q_h)+(B^{pp}\nabla P_h^{n+1},Q_h)+(\bar{B}^{pp}P_h^{n+1},\nabla Q_h)+(C^{pp}P_h^{n+1},Q_h)\\
&\quad=(F_p^{n+1},Q_h).
\end{aligned} ( H pp D t P h n + 1 , Q h ) + ( G pp ∇ D t P h n + 1 , Q h ) + ( G ˉ pp D t P h n + 1 , ∇ Q h ) + ( D pp ∇ D t P h n + 1 , ∇ Q h ) + ( A u p ∇ D t U h n + 1 , ∇ Q h ) + ( G u p ∇ D t U h n + 1 , Q h ) + ( G ˉ u p D t U h n + 1 , ∇ Q h ) + ( E u p D t U h n + 1 , Q h ) + ( A pp ∇ P h n + 1 , ∇ Q h ) + ( B pp ∇ P h n + 1 , Q h ) + ( B ˉ pp P h n + 1 , ∇ Q h ) + ( C pp P h n + 1 , Q h ) = ( F p n + 1 , Q h ) .
The initial pressure is taken as
P h 0 = Π h P 0 , P_h^0=\Pi_hP_0, P h 0 = Π h P 0 ,
where Π h \Pi_h Π h is the L 2 L^2 L 2 projection onto V p , h V_{p,h} V p , h , and U h 0 U_h^0 U h 0 is defined by
A u ( U h 0 , V h ) − B ( P h 0 , V h ) = ( F u 0 , V h ) , V h ∈ V u , h . (12) \mathcal{A}_u(U_h^0,V_h)-\mathcal{B}(P_h^0,V_h)=(F_u^0,V_h),\qquad V_h\in V_{u,h}. \tag*{(12)} A u ( U h 0 , V h ) − B ( P h 0 , V h ) = ( F u 0 , V h ) , V h ∈ V u , h . (12)
Well-posedness and stability
In this section, we establish the well-posedness and stability of the weak formulation (9) and the fully discrete scheme (11) , separately.
Lemma 1. Suppose Assumption 1 (A1)-(A3) hold. There exist constants c u , c p > 0 c_u,c_p>0 c u , c p > 0 , independent of the discretization parameters, such that
A u ( U , U ) ≥ c u ∥ U ∥ V u 2 , U ∈ V u , (13) \mathcal{A}_u(U,U)\ge c_u\lVert U\rVert_{V_u}^2,\qquad U\in V_u, \tag*{(13)} A u ( U , U ) ≥ c u ∥ U ∥ V u 2 , U ∈ V u , (13) A p ( P , P ) ≥ c p ∥ P ∥ V p 2 , P ∈ V p , (14) \mathcal{A}_p(P,P)\ge c_p\lVert P\rVert_{V_p}^2,\qquad P\in V_p, \tag*{(14)} A p ( P , P ) ≥ c p ∥ P ∥ V p 2 , P ∈ V p , (14) M p ( P , P ) ≥ c m ∥ P ∥ V p 2 , P ∈ V p . (15) \mathcal{M}_p(P,P)\ge c_m\lVert P\rVert_{V_p}^2,\qquad P\in V_p. \tag*{(15)} M p ( P , P ) ≥ c m ∥ P ∥ V p 2 , P ∈ V p . (15)
Moreover,
A u ( U h n + 1 , D t U h n + 1 ) ≥ c u 2 D t ∥ U h n + 1 ∥ V u 2 , U h n + 1 ∈ V u , h , (16) \mathcal{A}_u(U_h^{n+1},D_tU_h^{n+1}) \ge\frac{c_u}{2}D_t\|U_h^{n+1}\|_{V_u}^2,\qquad U_h^{n+1}\in V_{u,h}, \tag*{(16)} A u ( U h n + 1 , D t U h n + 1 ) ≥ 2 c u D t ∥ U h n + 1 ∥ V u 2 , U h n + 1 ∈ V u , h , (16)
M p ( D t P h n + 1 , P h n + 1 ) ≥ c m 2 D t ∥ P h n + 1 ∥ V p 2 , P h n + 1 ∈ V p , h , (17) \mathcal{M}_p(D_tP_h^{n+1},P_h^{n+1}) \ge\frac{c_m}{2}D_t\|P_h^{n+1}\|_{V_p}^2,\qquad P_h^{n+1}\in V_{p,h}, \tag*{(17)} M p ( D t P h n + 1 , P h n + 1 ) ≥ 2 c m D t ∥ P h n + 1 ∥ V p 2 , P h n + 1 ∈ V p , h , (17)
where
c u = min { α u u , α c u } − β u u > 0 , c_u=\min\{\alpha_{uu},\alpha_{cu}\}-\beta_{uu}>0, c u = min { α uu , α c u } − β uu > 0 ,
c p = min { α p p , α c p } − β p p > 0 , c_p=\min\{\alpha_{pp},\alpha_{cp}\}-\beta_{pp}>0, c p = min { α pp , α c p } − β pp > 0 ,
c m = min { α H , α D } − β g p > 0. c_m=\min\{\alpha_H,\alpha_D\}-\beta_{gp}>0. c m = min { α H , α D } − β g p > 0.
Proof. By the symmetry relations in (A1),
( B ‾ u u ∇ U , ∇ U ) = ( B u u ∇ U , U ) , ( B ‾ p p ∇ P , ∇ P ) = ( B p p ∇ P , P ) , ( G ‾ p p ∇ P , ∇ P ) = ( G p p ∇ P , P ) . (\overline{B}^{uu}\nabla U,\nabla U)=(B^{uu}\nabla U,U),\qquad
(\overline{B}^{pp}\nabla P,\nabla P)=(B^{pp}\nabla P,P),\qquad
(\overline{G}^{pp}\nabla P,\nabla P)=(G^{pp}\nabla P,P). ( B uu ∇ U , ∇ U ) = ( B uu ∇ U , U ) , ( B pp ∇ P , ∇ P ) = ( B pp ∇ P , P ) , ( G pp ∇ P , ∇ P ) = ( G pp ∇ P , P ) .
By the definition of A u \mathcal{A}_u A u and the relation B ‾ u u = ( B u u ) T \overline{B}^{uu}=(B^{uu})^{\mathsf{T}} B uu = ( B uu ) T , we have
A u ( U , U ) = ( A u u ∇ U , ∇ U ) + 2 ( B u u ∇ U , U ) + ( C u u U , U ) . \mathcal{A}_u(U,U)=(A^{uu}\nabla U,\nabla U)+2(B^{uu}\nabla U,U)+(C^{uu}U,U). A u ( U , U ) = ( A uu ∇ U , ∇ U ) + 2 ( B uu ∇ U , U ) + ( C uu U , U ) .
Consequently, (A2)-(A3) imply
( A u u ∇ U , ∇ U ) ≥ α u u ∥ ∇ U ∥ 2 , ( C u u U , U ) ≥ α c u ∥ U ∥ 2 , (A^{uu}\nabla U,\nabla U)\ge\alpha_{uu}\|\nabla U\|^2,\qquad
(C^{uu}U,U)\ge\alpha_{cu}\|U\|^2, ( A uu ∇ U , ∇ U ) ≥ α uu ∥∇ U ∥ 2 , ( C uu U , U ) ≥ α c u ∥ U ∥ 2 ,
∣ 2 ( B u u ∇ U , U ) ∣ ≥ − b u u ∥ ∇ U ∥ 2 − b u u ∥ U ∥ 2 . \lvert2(B^{uu}\nabla U,U)\rvert\ge-b_{uu}\|\nabla U\|^2-b_{uu}\|U\|^2. ∣ 2 ( B uu ∇ U , U )∣ ≥ − b uu ∥∇ U ∥ 2 − b uu ∥ U ∥ 2 .
Combining these estimates gives
A u ( U , U ) ≥ ( α u u − b u u ) ∥ ∇ U ∥ 2 + ( α c u − b u u ) ∥ U ∥ 2 ≥ c u ( ∥ ∇ U ∥ 2 + ∥ U ∥ 2 ) ≥ c u ∥ U ∥ V u 2 , \begin{aligned}
\mathcal{A}_u(U,U)\ge(\alpha_{uu}-b_{uu})\|\nabla U\|^2+(\alpha_{cu}-b_{uu})\|U\|^2 \\
&\ge c_u\left(\|\nabla U\|^2+\|U\|^2\right)\ge c_u\|U\|_{V_u}^2,
\end{aligned} A u ( U , U ) ≥ ( α uu − b uu ) ∥∇ U ∥ 2 + ( α c u − b uu ) ∥ U ∥ 2 ≥ c u ( ∥∇ U ∥ 2 + ∥ U ∥ 2 ) ≥ c u ∥ U ∥ V u 2 ,
where c u = min { α u u , α c u } − β u u > 0 c_u=\min\{\alpha_{uu},\alpha_{cu}\}-\beta_{uu}>0 c u = min { α uu , α c u } − β uu > 0 .
Besides, we have
A u ( U h n + 1 , D t U h n + 1 ) = ( A u u ∇ U h n + 1 , ∇ D t U h n + 1 ) + ( B u u ∇ U h n + 1 , D t U h n + 1 ) + ( B ‾ u u U h n + 1 , ∇ D t U h n + 1 ) + ( C u u U h n + 1 , D t U h n + 1 ) = 1 2 D t [ ( A u u ∇ U h n + 1 , ∇ U h n + 1 ) + ( C u u U h n + 1 , U h n + 1 ) ] + D t ( B u u ∇ U h n + 1 , U h n + 1 ) ≥ c u 2 D t ∥ U h n + 1 ∥ V u 2 . \begin{aligned}
\mathcal{A}_u(U_h^{n+1},D_tU_h^{n+1})
={}&(A^{uu}\nabla U_h^{n+1},\nabla D_tU_h^{n+1})+(B^{uu}\nabla U_h^{n+1},D_tU_h^{n+1})\\
&+(\overline{B}^{uu}U_h^{n+1},\nabla D_tU_h^{n+1})+(C^{uu}U_h^{n+1},D_tU_h^{n+1})\\
={}&\frac{1}{2}D_t\left[(A^{uu}\nabla U_h^{n+1},\nabla U_h^{n+1})+(C^{uu}U_h^{n+1},U_h^{n+1})\right]\\
&+D_t(B^{uu}\nabla U_h^{n+1},U_h^{n+1})\\
\ge{}&\frac{c_u}{2}D_t\|U_h^{n+1}\|_{V_u}^2.
\end{aligned} A u ( U h n + 1 , D t U h n + 1 ) = = ≥ ( A uu ∇ U h n + 1 , ∇ D t U h n + 1 ) + ( B uu ∇ U h n + 1 , D t U h n + 1 ) + ( B uu U h n + 1 , ∇ D t U h n + 1 ) + ( C uu U h n + 1 , D t U h n + 1 ) 2 1 D t [ ( A uu ∇ U h n + 1 , ∇ U h n + 1 ) + ( C uu U h n + 1 , U h n + 1 ) ] + D t ( B uu ∇ U h n + 1 , U h n + 1 ) 2 c u D t ∥ U h n + 1 ∥ V u 2 .
The proof for A p \mathcal{A}_p A p , M p \mathcal{M}_p M p are similar. This completes the proof.
Next, we establish the existence and uniqueness of the weak solution to the full multicontinuum model (7).
Theorem 1. Suppose Assumption 1 (A1)-(A4) hold. Then the weak problem (9) has a unique solution ( U , P ) (U,P) ( U , P ) such that
U ∈ H 1 ( 0 , T ; V u ) , P ∈ H 1 ( 0 , T ; V p ) . U\in H^1(0,T;V_u),\qquad P\in H^1(0,T;V_p). U ∈ H 1 ( 0 , T ; V u ) , P ∈ H 1 ( 0 , T ; V p ) .
Proof. Define the linear operator T : V p → V u \mathcal{T}: V_p \to V_u T : V p → V u by
A u ( T R , V ) = B ( R , V ) , V ∈ V u . (18) \mathcal{A}_u(\mathcal{T}R,V)=\mathcal{B}(R,V),\qquad V\in V_u. \tag*{(18)} A u ( T R , V ) = B ( R , V ) , V ∈ V u . (18)
By the continuity and coercivity of A u \mathcal{A}_u A u , and ∣ B ( R , V ) ∣ ≤ C B ∥ R ∥ V p ∥ V ∥ V u |\mathcal{B}(R,V)|\le C_B\|R\|_{V_p}\|V\|_{V_u} ∣ B ( R , V ) ∣ ≤ C B ∥ R ∥ V p ∥ V ∥ V u , the Lax–Milgram theorem implies that T \mathcal{T} T is well defined and bounded. In particular,
∥ T R ∥ V u ≤ C ∥ R ∥ V p . \|\mathcal{T}R\|_{V_u}\le C\|R\|_{V_p}. ∥ T R ∥ V u ≤ C ∥ R ∥ V p .
Next, for almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) , let w ( t ) ∈ V u w(t)\in V_u w ( t ) ∈ V u solve
A u ( w ( t ) , V ) = ( F u ( t ) , V ) , V ∈ V u . (19) \mathcal{A}_u(w(t),V)=(F_u(t),V),\qquad V\in V_u. \tag*{(19)} A u ( w ( t ) , V ) = ( F u ( t ) , V ) , V ∈ V u . (19)
The Lax–Milgram theorem gives a unique w ( t ) w(t) w ( t ) . By Assumptions (A4), we have F u ∈ H 1 ( 0 , T ; V u ′ ) F_u\in H^1(0,T;V'_u) F u ∈ H 1 ( 0 , T ; V u ′ ) , and since the effective coefficients are independent of time, the solution operator associated with A u \mathcal{A}_u A u is time independent and bounded, consequently,
w ∈ H 1 ( 0 , T ; V u ) . w\in H^1(0,T;V_u). w ∈ H 1 ( 0 , T ; V u ) .
The first equation of (9) therefore yields
A u ( U − w , V ) = B ( P , V ) ∀ V ∈ V u , \mathcal{A}_u(U-w,V)=\mathcal{B}(P,V)\qquad\forall V\in V_u, A u ( U − w , V ) = B ( P , V ) ∀ V ∈ V u ,
which, by the definition of T \mathcal{T} T , is equivalent to
U = w + T P . (20) U=w+\mathcal{T}P. \tag*{(20)} U = w + T P . (20)
In particular, ∂ t U = ∂ t w + T ∂ t P \partial_tU=\partial_tw+\mathcal{T}\partial_tP ∂ t U = ∂ t w + T ∂ t P .
For R , Q ∈ V p R,Q\in V_p R , Q ∈ V p , define
S ( R , Q ) : = M p ( R , Q ) + A u ( T R , T Q ) . (21) S(R,Q):=\mathcal{M}_p(R,Q)+\mathcal{A}_u(\mathcal{T}R,\mathcal{T}Q). \tag*{(21)} S ( R , Q ) := M p ( R , Q ) + A u ( T R , T Q ) . (21)
By Assumption (A1), the forms A u \mathcal{A}_u A u and M p \mathcal{M}_p M p are symmetric. Hence S S S is symmetric. Moreover, the continuity of M p \mathcal{M}_p M p , A u \mathcal{A}_u A u , and T \mathcal{T} T implies that S S S is continuous on V p × V p V_p\times V_p V p × V p . By the coercivity of M p \mathcal{M}_p M p ,
S ( Q , Q ) = M p ( Q , Q ) + A u ( T Q , T Q ) ≥ M p ( Q , Q ) ≥ c m ∥ Q ∥ V p 2 . \begin{aligned}
S(Q,Q)&=\mathcal{M}_p(Q,Q)+\mathcal{A}_u(\mathcal{T}Q,\mathcal{T}Q)\\
&\ge\mathcal{M}_p(Q,Q)\ge c_m\|Q\|_{V_p}^{2}.
\end{aligned} S ( Q , Q ) = M p ( Q , Q ) + A u ( T Q , T Q ) ≥ M p ( Q , Q ) ≥ c m ∥ Q ∥ V p 2 .
Thus, S S S is coercive on V p V_p V p and defines an equivalent inner product on this space.
By the definition of T \mathcal{T} T and the symmetry of A u \mathcal{A}_u A u ,
B ( Q , T ∂ t P ) = A u ( T Q , T ∂ t P ) = A u ( T ∂ t P , T Q ) . \mathcal{B}(Q,\mathcal{T}\partial_tP)=\mathcal{A}_u(\mathcal{T}Q,\mathcal{T}\partial_tP)=\mathcal{A}_u(\mathcal{T}\partial_tP,\mathcal{T}Q). B ( Q , T ∂ t P ) = A u ( T Q , T ∂ t P ) = A u ( T ∂ t P , T Q ) .
Substituting (20) into the second equation of (9) gives the reduced pressure equation
S ( ∂ t P , Q ) + A p ( P , Q ) = ( F p , Q ) − B ( Q , ∂ t w ) , Q ∈ V p . (22) S(\partial_tP,Q)+\mathcal{A}_p(P,Q)=(F_p,Q)-\mathcal{B}(Q,\partial_tw),\qquad Q\in V_p. \tag*{(22)} S ( ∂ t P , Q ) + A p ( P , Q ) = ( F p , Q ) − B ( Q , ∂ t w ) , Q ∈ V p . (22)
Let S : V p → V p ′ S:V_p\to V'_p S : V p → V p ′ and A p : V p → V p ′ \mathcal{A}_p:V_p\to V'_p A p : V p → V p ′ be the operators induced by S S S and A p \mathcal{A}_p A p , respectively; that is,
⟨ S R , Q ⟩ = S ( R , Q ) , ⟨ A p R , Q ⟩ = A p ( R , Q ) . \langle SR,Q\rangle=S(R,Q),\qquad\langle\mathcal{A}_pR,Q\rangle=\mathcal{A}_p(R,Q). ⟨ S R , Q ⟩ = S ( R , Q ) , ⟨ A p R , Q ⟩ = A p ( R , Q ) .
The continuity and coercivity of S S S imply, by the Lax–Milgram theorem, that S S S is an isomorphism from V p V_p V p onto V p ′ V'_p V p ′ . Define ℓ ∈ L 2 ( 0 , T ; V p ′ ) \ell\in L^2(0,T;V_p') ℓ ∈ L 2 ( 0 , T ; V p ′ ) by
⟨ ℓ ( t ) , Q ⟩ : = ( F p ( t ) , Q ) − B ( Q , ∂ t w ( t ) ) , Q ∈ V p . \langle\ell(t),Q\rangle:= (F_p(t),Q)-\mathcal{B}(Q,\partial_t w(t)), \qquad Q\in V_p. ⟨ ℓ ( t ) , Q ⟩ := ( F p ( t ) , Q ) − B ( Q , ∂ t w ( t )) , Q ∈ V p .
Indeed, the continuity of B \mathcal{B} B , together with F p ∈ L 2 ( 0 , T ; V p ′ ) F_p\in L^2(0,T;V_p') F p ∈ L 2 ( 0 , T ; V p ′ ) and ∂ t w ∈ L 2 ( 0 , T ; V u ) \partial_t w\in L^2(0,T;V_u) ∂ t w ∈ L 2 ( 0 , T ; V u ) , gives ℓ ∈ L 2 ( 0 , T ; V p ′ ) \ell\in L^2(0,T;V_p') ℓ ∈ L 2 ( 0 , T ; V p ′ ) .
In operator form, (22) becomes
∂ t P + S − 1 A p P = S − 1 ℓ , P ( 0 ) = P 0 . (23) \partial_tP+\mathbf{S}^{-1}A_pP=\mathbf{S}^{-1}\ell,\qquad P(0)=P_0. \tag*{(23)} ∂ t P + S − 1 A p P = S − 1 ℓ , P ( 0 ) = P 0 . (23)
Since A p : V p → V p ′ A_p:V_p\to V_p' A p : V p → V p ′ is bounded and S − 1 : V p ′ → V p \mathbf{S}^{-1}:V_p'\to V_p S − 1 : V p ′ → V p is bounded, S − 1 A p \mathbf{S}^{-1}A_p S − 1 A p is a bounded linear operator on V p V_p V p . Furthermore, S − 1 ℓ ∈ L 2 ( 0 , T ; V p ) \mathbf{S}^{-1}\ell\in L^2(0,T;V_p) S − 1 ℓ ∈ L 2 ( 0 , T ; V p ) . The standard existence and uniqueness result for linear evolution equations with bounded operators (equivalently, the variation-of-constants formula) therefore yields a unique solution
P ∈ H 1 ( 0 , T ; V p ) P\in H^1(0,T;V_p) P ∈ H 1 ( 0 , T ; V p )
with the prescribed initial value P ( 0 ) = P 0 P(0)=P_0 P ( 0 ) = P 0 .
Define U U U by (20). Since w ∈ H 1 ( 0 , T ; V u ) w\in H^1(0,T;V_u) w ∈ H 1 ( 0 , T ; V u ) , T \mathcal{T} T is bounded, and P ∈ H 1 ( 0 , T ; V p ) P\in H^1(0,T;V_p) P ∈ H 1 ( 0 , T ; V p ) , it follows that
U ∈ H 1 ( 0 , T ; V u ) . U\in H^1(0,T;V_u). U ∈ H 1 ( 0 , T ; V u ) .
The construction of P P P and U U U shows that both equations in (9) are satisfied for almost every t ∈ ( 0 , T ) t\in(0,T) t ∈ ( 0 , T ) .
It remains to verify the initial equilibrium condition. Since w ∈ H 1 ( 0 , T ; V u ) w\in H^1(0,T;V_u) w ∈ H 1 ( 0 , T ; V u ) and P ∈ H 1 ( 0 , T ; V p ) P\in H^1(0,T;V_p) P ∈ H 1 ( 0 , T ; V p ) , both have well-defined traces at t = 0 t=0 t = 0 . Using P ( 0 ) = P 0 P(0)=P_0 P ( 0 ) = P 0 , the definition of w w w , and that of T \mathcal{T} T , we obtain, for every V ∈ V u V\in V_u V ∈ V u ,
A u ( U ( 0 ) , V ) − B ( P 0 , V ) = A u ( w ( 0 ) + T P 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) + B ( P 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) . \begin{aligned}
A_u(U(0),V)-\mathcal{B}(P_0,V)
&=A_u(w(0)+\mathcal{T}P_0,V)-\mathcal{B}(P_0,V)\\
&=(F_u(0),V)+\mathcal{B}(P_0,V)-\mathcal{B}(P_0,V)\\
&=(F_u(0),V).
\end{aligned} A u ( U ( 0 ) , V ) − B ( P 0 , V ) = A u ( w ( 0 ) + T P 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) + B ( P 0 , V ) − B ( P 0 , V ) = ( F u ( 0 ) , V ) .
Thus, the initial equilibrium condition (10) holds.
Finally, the reduced evolution equation (23) has a unique solution for the prescribed initial value. Hence the pressure P P P is unique, and the representation U = w + T P U=w+\mathcal{T}P U = w + T P then implies the uniqueness of U U U . This completes the proof.
Then we give the energy identity and stability of the weak solution to the full multicontinuum model (7).
Theorem 2. Suppose Assumption 1 (A1)-(A4) hold. The solution of the weak problem (9) satisfies
sup 0 ≤ t ≤ T ( ∥ U ( t ) ∥ V u 2 + ∥ P ( t ) ∥ V p 2 ) + ∫ 0 T ∥ P ( t ) ∥ V p 2 d t ≤ C ( ∥ U 0 ∥ V u 2 + ∥ P 0 ∥ V p 2 + ∥ F u ∥ L ∞ ( 0 , T ; L 2 ) 2 + ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 + ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 ) . (24) \sup_{0\leq t\leq T}\left(\lVert U(t)\rVert_{V_u}^2+\lVert P(t)\rVert_{V_p}^2\right)+\int_0^T\lVert P(t)\rVert_{V_p}^2\,\mathrm{d}t
\leq C\left(\lVert U_0\rVert_{V_u}^2+\lVert P_0\rVert_{V_p}^2+\lVert F_u\rVert_{L^\infty(0,T;L^2)}^2+\lVert\partial_tF_u\rVert_{L^2(0,T;L^2)}^2+\lVert F_p\rVert_{L^2(0,T;L^2)}^2\right). \tag*{(24)} 0 ≤ t ≤ T sup ( ∥ U ( t ) ∥ V u 2 + ∥ P ( t ) ∥ V p 2 ) + ∫ 0 T ∥ P ( t ) ∥ V p 2 d t ≤ C ( ∥ U 0 ∥ V u 2 + ∥ P 0 ∥ V p 2 + ∥ F u ∥ L ∞ ( 0 , T ; L 2 ) 2 + ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 + ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 ) . (24)
Proof. Take V = ∂ t U V=\partial_tU V = ∂ t U and Q = P Q=P Q = P in (9). The coupling contributions cancel:
− B ( P , ∂ t U ) + B ( P , ∂ t U ) = 0. -\mathcal{B}(P,\partial_tU)+\mathcal{B}(P,\partial_tU)=0. − B ( P , ∂ t U ) + B ( P , ∂ t U ) = 0.
Since the forms A u A_u A u and M p M_p M p are symmetric and independent of time, we get
A u ( U , ∂ t U ) = 1 2 d d t A u ( U , U ) , A_u(U,\partial_tU)=\frac{1}{2}\frac{\mathrm{d}}{\mathrm{d}t}A_u(U,U), A u ( U , ∂ t U ) = 2 1 d t d A u ( U , U ) ,
M p ( ∂ t P , P ) = 1 2 d d t M p ( P , P ) . \mathcal{M}_p(\partial_t P,P)=\frac{1}{2}\frac{d}{dt}\mathcal{M}_p(P,P). M p ( ∂ t P , P ) = 2 1 d t d M p ( P , P ) .
Define the energy by
E ( t ) = 1 2 A u ( U ( t ) , U ( t ) ) + 1 2 M p ( P ( t ) , P ( t ) ) . (25) \mathcal{E}(t)=\frac{1}{2}\mathcal{A}_u(U(t),U(t))+\frac{1}{2}\mathcal{M}_p(P(t),P(t)). \tag*{(25)} E ( t ) = 2 1 A u ( U ( t ) , U ( t )) + 2 1 M p ( P ( t ) , P ( t )) . (25)
Then we have
d d t E ( t ) + A p ( P ( t ) , P ( t ) ) = ( F u ( t ) , ∂ t U ( t ) ) + ( F p ( t ) , P ( t ) ) (26) \frac{d}{dt}\mathcal{E}(t)+\mathcal{A}_p(P(t),P(t))=(F_u(t),\partial_t U(t))+(F_p(t),P(t)) \tag*{(26)} d t d E ( t ) + A p ( P ( t ) , P ( t )) = ( F u ( t ) , ∂ t U ( t )) + ( F p ( t ) , P ( t )) (26)
Integrating in time and integrating the displacement load by parts yields
E ( t ) + ∫ 0 t A p ( P , P ) d s = E ( 0 ) + ( F u ( t ) , U ( t ) ) − ( F u ( 0 ) , U 0 ) − ∫ 0 t ( ∂ t F u , U ) d s + ∫ 0 t ( F p , P ) d s . (27) \mathcal{E}(t)+\int_0^t\mathcal{A}_p(P,P)\,ds=\mathcal{E}(0)+(F_u(t),U(t))-(F_u(0),U_0)-\int_0^t(\partial_tF_u,U)\,ds+\int_0^t(F_p,P)\,ds. \tag*{(27)} E ( t ) + ∫ 0 t A p ( P , P ) d s = E ( 0 ) + ( F u ( t ) , U ( t )) − ( F u ( 0 ) , U 0 ) − ∫ 0 t ( ∂ t F u , U ) d s + ∫ 0 t ( F p , P ) d s . (27)
By Cauchy-Schwarz’s inequality and Young’s inequality,
∣ ( F u ( t ) , U ( t ) ) ∣ ≤ ε ∥ U ( t ) ∥ L 2 2 + C ε ∥ F u ( t ) ∥ L 2 2 , |(F_u(t),U(t))|\leq\varepsilon\|U(t)\|_{L^2}^2+C_\varepsilon\|F_u(t)\|_{L^2}^2, ∣ ( F u ( t ) , U ( t )) ∣ ≤ ε ∥ U ( t ) ∥ L 2 2 + C ε ∥ F u ( t ) ∥ L 2 2 ,
∣ ( F u ( 0 ) , U 0 ) ∣ ≤ C ( ∥ F u ( 0 ) ∥ L 2 2 + ∥ U 0 ∥ 2 ) , |(F_u(0),U_0)|\leq C\left(\|F_u(0)\|_{L^2}^2+\|U_0\|^2\right), ∣ ( F u ( 0 ) , U 0 ) ∣ ≤ C ( ∥ F u ( 0 ) ∥ L 2 2 + ∥ U 0 ∥ 2 ) ,
∫ 0 t ∣ ( ∂ t F u , U ) ∣ d s ≤ ε ∥ U ∥ L 2 ( 0 , T ; L 2 ) 2 + C ε ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 , \int_0^t|(\partial_tF_u,U)|\,ds\leq\varepsilon\|U\|_{L^2(0,T;L^2)}^2+C_\varepsilon\|\partial_tF_u\|_{L^2(0,T;L^2)}^2, ∫ 0 t ∣ ( ∂ t F u , U ) ∣ d s ≤ ε ∥ U ∥ L 2 ( 0 , T ; L 2 ) 2 + C ε ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 ,
∫ 0 t ∣ ( F p , P ) ∣ d s ≤ ε ∥ P ∥ L 2 ( 0 , T ; L 2 ) 2 + C ε ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 . \int_0^t|(F_p,P)|\,ds\leq\varepsilon\|P\|_{L^2(0,T;L^2)}^2+C_\varepsilon\|F_p\|_{L^2(0,T;L^2)}^2. ∫ 0 t ∣ ( F p , P ) ∣ d s ≤ ε ∥ P ∥ L 2 ( 0 , T ; L 2 ) 2 + C ε ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 .
Continuity and coercivity imply
c E ( ∥ U ∥ V u 2 + ∥ P ∥ V p 2 ) ≤ E ≤ C E ( ∥ U ∥ V u 2 + ∥ P ∥ V p 2 ) , c_E\left(\|U\|_{V_u}^2+\|P\|_{V_p}^2\right)\leq\mathcal{E}\leq C_E\left(\|U\|_{V_u}^2+\|P\|_{V_p}^2\right), c E ( ∥ U ∥ V u 2 + ∥ P ∥ V p 2 ) ≤ E ≤ C E ( ∥ U ∥ V u 2 + ∥ P ∥ V p 2 ) ,
where c E , C E > 0 c_E,C_E>0 c E , C E > 0 .
Insert these estimates into (27), take the supremum over t t t , and choose ε \varepsilon ε sufficiently small to absorb the terms involving sup 0 ≤ t ≤ T E ( t ) \sup_{0\leq t\leq T}\mathcal{E}(t) sup 0 ≤ t ≤ T E ( t ) . Returning to (27) at t = T t=T t = T then bounds the dissipation integral. The proof is complete.
Finally, we establish the well-posedness and stability of the fully discrete scheme (11).
Theorem 3. Suppose Assumption 1 (A1)-(A4) hold. The fully discrete scheme (11) has a unique solution ( U h n + 1 , P h n + 1 ) ∈ V u , h × V p , h (U_h^{n+1},P_h^{n+1})\in V_{u,h}\times V_{p,h} ( U h n + 1 , P h n + 1 ) ∈ V u , h × V p , h . Moreover, the solution satisfies a discrete stability estimate of the form
max 0 ≤ n ≤ N ( ∥ U h n ∥ V u 2 + ∥ P h n ∥ V p 2 ) + τ ∑ n = 0 N − 1 ∥ P h n + 1 ∥ V p 2 ≤ C [ ∥ U h 0 ∥ V u 2 + ∥ P h 0 ∥ V p 2 + ∥ F u ∥ L ∞ ( 0 , T ; L 2 ) 2 + ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 + ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 ] . (28) \max_{0\leq n\leq N}\left(\|U_h^n\|_{V_u}^2+\|P_h^n\|_{V_p}^2\right)+\tau\sum_{n=0}^{N-1}\|P_h^{n+1}\|_{V_p}^2
\leq C\left[\|U_h^0\|_{V_u}^2+\|P_h^0\|_{V_p}^2+\|F_u\|_{L^\infty(0,T;L^2)}^2+\|\partial_tF_u\|_{L^2(0,T;L^2)}^2+\|F_p\|_{L^2(0,T;L^2)}^2\right]. \tag*{(28)} 0 ≤ n ≤ N max ( ∥ U h n ∥ V u 2 + ∥ P h n ∥ V p 2 ) + τ n = 0 ∑ N − 1 ∥ P h n + 1 ∥ V p 2 ≤ C [ ∥ U h 0 ∥ V u 2 + ∥ P h 0 ∥ V p 2 + ∥ F u ∥ L ∞ ( 0 , T ; L 2 ) 2 + ∥ ∂ t F u ∥ L 2 ( 0 , T ; L 2 ) 2 + ∥ F p ∥ L 2 ( 0 , T ; L 2 ) 2 ] . (28) where C C C is independent of h h h and τ \tau τ .
Proof. First, we prove the well-posedness of the fully discrete scheme (11).
Consider the difference ( U ^ , P ^ ) (\widehat{U},\widehat{P}) ( U , P ) of two solutions at the same time step, with identical previous values and loads. Then
A u ( U ^ , V h ) − B ( P ^ , V h ) = 0 , \mathcal{A}_u(\widehat{U},V_h)-\mathcal{B}(\widehat{P},V_h)=0, A u ( U , V h ) − B ( P , V h ) = 0 ,
e n d s m i d = 1 {ends_mid=1} e n d s m i d = 1
after multiplying the pressure equation by τ \tau τ ,
M p ( P ^ , Q h ) + B ( Q h , U ^ ) + τ A p ( P ^ , Q h ) = 0. \mathcal{M}_p(\widehat{P},Q_h)+\mathcal{B}(Q_h,\widehat{U})+\tau\mathcal{A}_p(\widehat{P},Q_h)=0. M p ( P , Q h ) + B ( Q h , U ) + τ A p ( P , Q h ) = 0.
Take V h = U ^ V_h=\widehat{U} V h = U and Q h = P ^ Q_h=\widehat{P} Q h = P and add the two equations, we get
A u ( U ^ , U ^ ) + M p ( P ^ , P ^ ) + τ A p ( P ^ , P ^ ) = 0 , \mathcal{A}_u(\widehat{U},\widehat{U})+\mathcal{M}_p(\widehat{P},\widehat{P})+\tau\mathcal{A}_p(\widehat{P},\widehat{P})=0, A u ( U , U ) + M p ( P , P ) + τ A p ( P , P ) = 0 ,
which implies U ^ = 0 \widehat{U}=0 U = 0 and P ^ = 0 \widehat{P}=0 P = 0 . Since the algebraic system is square and finite dimensional, uniqueness implies existence.
Next, we prove the energy estimate of the fully discrete scheme (11).
Take V h = D t U h n + 1 V_h=D_tU_h^{n+1} V h = D t U h n + 1 , Q h = P h n + 1 Q_h=P_h^{n+1} Q h = P h n + 1 in (11), and add the two equations. The coupling terms cancel exactly.
By Lemma 1, we have
A u ( U h n + 1 , D t U h n + 1 ) ≥ c u 2 D t ∥ U h n + 1 ∥ V u 2 , M p ( D t P h n + 1 , P h n + 1 ) ≥ c m 2 ∥ P h n + 1 ∥ V p 2 . \mathcal{A}_u(U_h^{n+1},D_tU_h^{n+1})\geq\frac{c_u}{2}D_t\lVert U_h^{n+1}\rVert_{V_u}^2,\qquad
\mathcal{M}_p(D_tP_h^{n+1},P_h^{n+1})\geq\frac{c_m}{2}\lVert P_h^{n+1}\rVert_{V_p}^2. A u ( U h n + 1 , D t U h n + 1 ) ≥ 2 c u D t ∥ U h n + 1 ∥ V u 2 , M p ( D t P h n + 1 , P h n + 1 ) ≥ 2 c m ∥ P h n + 1 ∥ V p 2 .
Then, the system (11) gives
D t ∥ U h n + 1 ∥ V u 2 + D t ∥ P h n + 1 ∥ V p 2 + ∥ P h n + 1 ∥ V p 2 ≤ ( F u n + 1 , D t U h n + 1 ) + ( F p n + 1 , P h n + 1 ) . (29) D_t\lVert U_h^{n+1}\rVert_{V_u}^2+D_t\lVert P_h^{n+1}\rVert_{V_p}^2+\lVert P_h^{n+1}\rVert_{V_p}^2\leq(F_u^{n+1},D_tU_h^{n+1})+(F_p^{n+1},P_h^{n+1}). \tag*{(29)} D t ∥ U h n + 1 ∥ V u 2 + D t ∥ P h n + 1 ∥ V p 2 + ∥ P h n + 1 ∥ V p 2 ≤ ( F u n + 1 , D t U h n + 1 ) + ( F p n + 1 , P h n + 1 ) . (29)
Sum (29) from n = 0 n=0 n = 0 to N − 1 N-1 N − 1 , we get
∥ U h N ∥ V u 2 + ∥ P h N ∥ V p 2 + ∑ n = 0 N − 1 ∥ P h n + 1 ∥ V p 2 ≤ ∑ n = 0 N − 1 τ ( F u n + 1 , D t U h n + 1 ) + ∑ n = 0 N − 1 τ ( F p n + 1 , P h n + 1 ) . (30) \lVert U_h^N\rVert_{V_u}^2+\lVert P_h^N\rVert_{V_p}^2+\sum_{n=0}^{N-1}\lVert P_h^{n+1}\rVert_{V_p}^2\leq\sum_{n=0}^{N-1}\tau(F_u^{n+1},D_tU_h^{n+1})+\sum_{n=0}^{N-1}\tau(F_p^{n+1},P_h^{n+1}). \tag*{(30)} ∥ U h N ∥ V u 2 + ∥ P h N ∥ V p 2 + n = 0 ∑ N − 1 ∥ P h n + 1 ∥ V p 2 ≤ n = 0 ∑ N − 1 τ ( F u n + 1 , D t U h n + 1 ) + n = 0 ∑ N − 1 τ ( F p n + 1 , P h n + 1 ) . (30)
By Cauchy-Schwarz and Young’s inequalities,
∑ n = 0 N − 1 τ ( F u n + 1 , D t U h n + 1 ) = ( F u N , U h N ) − ( F u 0 , U h 0 ) − ∑ n = 0 N − 1 τ ( D t F u n + 1 , U h n ) ≤ ε max 0 ≤ n ≤ N ∥ U h n ∥ V u 2 + C ε [ ∥ U h 0 ∥ V u 2 + max 0 ≤ n ≤ N ∥ F u n ∥ L 2 2 + ∑ n = 0 N − 1 τ ∥ D t F u n + 1 ∥ L 2 2 ] ≤ ε max 0 ≤ n ≤ N ∥ U h n ∥ L 2 2 + C ε [ ∥ U h 0 ∥ L 2 2 + max 0 ≤ n ≤ N ∥ F u n ∥ L 2 2 + ∫ 0 T ∥ ∂ t F u ∥ L 2 2 d t ] . \begin{aligned}
\sum_{n=0}^{N-1}\tau(F_u^{n+1},D_tU_h^{n+1})=(F_u^N,U_h^N)-(F_u^0,U_h^0)-\sum_{n=0}^{N-1}\tau(D_tF_u^{n+1},U_h^n) \\
&\leq\varepsilon\max_{0\leq n\leq N}\lVert U_h^n\rVert_{V_u}^2+C_\varepsilon\left[\lVert U_h^0\rVert_{V_u}^2+\max_{0\leq n\leq N}\lVert F_u^n\rVert_{L^2}^2+\sum_{n=0}^{N-1}\tau\lVert D_tF_u^{n+1}\rVert_{L^2}^2\right] \\
&\leq\varepsilon\max_{0\leq n\leq N}\lVert U_h^n\rVert_{L^2}^2+C_\varepsilon\left[\lVert U_h^0\rVert_{L^2}^2+\max_{0\leq n\leq N}\lVert F_u^n\rVert_{L^2}^2+\int_0^T\lVert\partial_tF_u\rVert_{L^2}^2\,dt\right].
\end{aligned} n = 0 ∑ N − 1 τ ( F u n + 1 , D t U h n + 1 ) = ( F u N , U h N ) − ( F u 0 , U h 0 ) − n = 0 ∑ N − 1 τ ( D t F u n + 1 , U h n ) ≤ ε 0 ≤ n ≤ N max ∥ U h n ∥ V u 2 + C ε [ ∥ U h 0 ∥ V u 2 + 0 ≤ n ≤ N max ∥ F u n ∥ L 2 2 + n = 0 ∑ N − 1 τ ∥ D t F u n + 1 ∥ L 2 2 ] ≤ ε 0 ≤ n ≤ N max ∥ U h n ∥ L 2 2 + C ε [ ∥ U h 0 ∥ L 2 2 + 0 ≤ n ≤ N max ∥ F u n ∥ L 2 2 + ∫ 0 T ∥ ∂ t F u ∥ L 2 2 d t ] .
For the pressure,
∑ n = 0 N − 1 τ ∣ ( F p n + 1 , P h n + 1 ) ∣ ≤ ε ∑ n = 0 N − 1 τ ∥ P h n + 1 ∥ V p 2 + C ε τ ∥ F p n + 1 ∥ L 2 2 . \sum_{n=0}^{N-1}\tau\left|(F_p^{n+1},P_h^{n+1})\right|\leq\varepsilon\sum_{n=0}^{N-1}\tau\lVert P_h^{n+1}\rVert_{V_p}^2+C_\varepsilon\tau\lVert F_p^{n+1}\rVert_{L^2}^2. n = 0 ∑ N − 1 τ ( F p n + 1 , P h n + 1 ) ≤ ε n = 0 ∑ N − 1 τ ∥ P h n + 1 ∥ V p 2 + C ε τ ∥ F p n + 1 ∥ L 2 2 .
By choosing ε \varepsilon ε sufficiently small, we can absorb the terms involving ∥ U h n ∥ V u \lVert U_h^n\rVert_{V_u} ∥ U h n ∥ V u and ∥ P h n + 1 ∥ V p \lVert P_h^{n+1}\rVert_{V_p} ∥ P h n + 1 ∥ V p into the left-hand side of the inequality (30), yielding the desired stability estimate (28).
Error analysis for the fully discrete full model
Finite element projections
In this subsection, we define the finite element projections and establish their approximation properties.
For U ∈ V u U \in V_u U ∈ V u , define I h U ∈ V u , h I_hU \in V_{u,h} I h U ∈ V u , h by
A u ( I h U , V h ) = A u ( U , V h ) , V h ∈ V u , h . (31) A_u(I_hU,V_h)=A_u(U,V_h), \qquad V_h \in V_{u,h}. \tag*{(31)} A u ( I h U , V h ) = A u ( U , V h ) , V h ∈ V u , h . (31)
For P ∈ H p P \in H_p P ∈ H p , define Π h P ∈ V p , h \Pi_hP \in V_{p,h} Π h P ∈ V p , h by
( Π h P − P , Q h ) = 0 , Q h ∈ V p , h . (32) (\Pi_hP-P,Q_h)=0, \qquad Q_h \in V_{p,h}. \tag*{(32)} ( Π h P − P , Q h ) = 0 , Q h ∈ V p , h . (32)
Lemma 2 (Projection estimates [21 ] ). For
U ∈ H k + 1 ( Ω ) d N c ∩ V u , P ∈ H k + 1 ( Ω ) N c ∩ V p , U \in H^{k+1}(\Omega)^{dN_c} \cap V_u, \qquad P \in H^{k+1}(\Omega)^{N_c} \cap V_p, U ∈ H k + 1 ( Ω ) d N c ∩ V u , P ∈ H k + 1 ( Ω ) N c ∩ V p , the projections satisfy
∥ U − I h U ∥ V u ≤ C h k ∥ U ∥ H k + 1 , (33) \lVert U-I_hU\rVert_{V_u} \leq Ch^k\lVert U\rVert_{H^{k+1}}, \tag*{(33)} ∥ U − I h U ∥ V u ≤ C h k ∥ U ∥ H k + 1 , (33) and
∥ P − Π h P ∥ L 2 + h ∥ P − Π h P ∥ V p ≤ C h k + 1 ∥ P ∥ H k + 1 . (34) \lVert P-\Pi_hP\rVert_{L^2}+h\lVert P-\Pi_hP\rVert_{V_p} \leq Ch^{k+1}\lVert P\rVert_{H^{k+1}}. \tag*{(34)} ∥ P − Π h P ∥ L 2 + h ∥ P − Π h P ∥ V p ≤ C h k + 1 ∥ P ∥ H k + 1 . (34)
Proof. The estimate for I h I_h I h follows from the continuity and coercivity of A u A_u A u , Cea’s lemma, and standard finite element approximation.
The L 2 L^2 L 2 estimate for Π h \Pi_h Π h follows from its best-approximation property. On a shape-regular, quasi-uniform mesh family, the L 2 L^2 L 2 projection is uniformly stable in H 1 H^1 H 1 . Combining this stability with a standard conforming interpolant gives the H 1 H^1 H 1 estimate. □
A priori error estimate
Theorem 4. Suppose Assumption 1 holds. Let ( U h n , P h n ) (U_h^n,P_h^n) ( U h n , P h n ) be the solution of (11), ( U , P ) (U,P) ( U , P ) be the solution of (9). Then
max 0 ≤ n ≤ N ( ∥ U n − U h n ∥ V u 2 + ∥ P n − P h n ∥ V p 2 ) + τ ∑ n = 0 N − 1 ∥ e p n + 1 ∥ V p 2 ≤ C ( h 2 k + τ 2 ) , (35) \max_{0\leq n\leq N}\left(\lVert U^n-U_h^n\rVert_{V_u}^2+\lVert P^n-P_h^n\rVert_{V_p}^2\right)+\tau\sum_{n=0}^{N-1}\lVert e_p^{n+1}\rVert_{V_p}^2 \leq C(h^{2k}+\tau^2), \tag*{(35)} 0 ≤ n ≤ N max ( ∥ U n − U h n ∥ V u 2 + ∥ P n − P h n ∥ V p 2 ) + τ n = 0 ∑ N − 1 ∥ e p n + 1 ∥ V p 2 ≤ C ( h 2 k + τ 2 ) , (35) where the constant C C C is independent of h h h and τ \tau τ .
Proof. Set
e u n = η u n + θ u n , e p n = η p n + θ p n , e_u^n=\eta_u^n+\theta_u^n,\qquad e_p^n=\eta_p^n+\theta_p^n, e u n = η u n + θ u n , e p n = η p n + θ p n ,
η u n = U ( t n ) − I h U ( t n ) , θ u n = I h U ( t n ) − U h n , \eta_u^n=U(t_n)-I_hU(t_n),\qquad\theta_u^n=I_hU(t_n)-U_h^n, η u n = U ( t n ) − I h U ( t n ) , θ u n = I h U ( t n ) − U h n ,
η p n = P ( t n ) − Π h P ( t n ) , θ p n = Π h P ( t n ) − P h n . \eta_p^n=P(t_n)-\Pi_hP(t_n),\qquad\theta_p^n=\Pi_hP(t_n)-P_h^n. η p n = P ( t n ) − Π h P ( t n ) , θ p n = Π h P ( t n ) − P h n .
Define
ρ u n + 1 = ∂ t U ( t n + 1 ) − D t U ( t n + 1 ) , \rho_u^{n+1}=\partial_tU(t_{n+1})-D_tU(t_{n+1}), ρ u n + 1 = ∂ t U ( t n + 1 ) − D t U ( t n + 1 ) ,
ρ p n + 1 = ∂ t P ( t n + 1 ) − D t P ( t n + 1 ) . \rho_p^{n+1}=\partial_tP(t_{n+1})-D_tP(t_{n+1}). ρ p n + 1 = ∂ t P ( t n + 1 ) − D t P ( t n + 1 ) .
Consequently,
τ ∑ n = 0 N − 1 ∥ ρ u n + 1 ∥ V u 2 ≤ C τ 2 ∥ ∂ t t U ∥ L 2 ( 0 , T ; V u ) 2 , \tau\sum_{n=0}^{N-1}\lVert\rho_u^{n+1}\rVert_{V_u}^2\leq C\tau^2\lVert\partial_{tt}U\rVert_{L^2(0,T;V_u)}^2, τ n = 0 ∑ N − 1 ∥ ρ u n + 1 ∥ V u 2 ≤ C τ 2 ∥ ∂ tt U ∥ L 2 ( 0 , T ; V u ) 2 ,
τ ∑ n = 0 N − 1 ∥ ρ p n + 1 ∥ V p 2 ≤ C τ 2 ∥ ∂ t t P ∥ L 2 ( 0 , T ; V p ) 2 . (36) \tau\sum_{n=0}^{N-1}\lVert\rho_p^{n+1}\rVert_{V_p}^2\leq C\tau^2\lVert\partial_{tt}P\rVert_{L^2(0,T;V_p)}^2. \tag*{(36)} τ n = 0 ∑ N − 1 ∥ ρ p n + 1 ∥ V p 2 ≤ C τ 2 ∥ ∂ tt P ∥ L 2 ( 0 , T ; V p ) 2 . (36)
Subtracting the discrete equations from (11) the continuous equations (9) at t n + 1 t_{n+1} t n + 1 gives
A u ( e u n + 1 , V h ) − B ( e p n + 1 , V h ) = 0 , (37) A_u(e_u^{n+1},V_h)-B(e_p^{n+1},V_h)=0, \tag*{(37)} A u ( e u n + 1 , V h ) − B ( e p n + 1 , V h ) = 0 , (37)
M p ( D t e p n + 1 , Q h ) + B ( Q h , D t e u n + 1 ) + A p ( e p n + 1 , Q h ) = − M p ( ρ p n + 1 , Q h ) − B ( Q h , ρ u n + 1 ) . (38) \mathcal{M}_p(D_t e_p^{n+1},Q_h)+B(Q_h,D_t e_u^{n+1})+A_p(e_p^{n+1},Q_h)=-\mathcal{M}_p(\rho_p^{n+1},Q_h)-B(Q_h,\rho_u^{n+1}). \tag*{(38)} M p ( D t e p n + 1 , Q h ) + B ( Q h , D t e u n + 1 ) + A p ( e p n + 1 , Q h ) = − M p ( ρ p n + 1 , Q h ) − B ( Q h , ρ u n + 1 ) . (38)
Using the error decomposition in (37) –(38) , and the projection projection A u ( η u n , V h ) = 0 A_u(\eta_u^n,V_h)=0 A u ( η u n , V h ) = 0 , we obtain
A u ( θ u n + 1 , V h ) − B ( θ p n + 1 , V h ) = ( R u n + 1 , V h ) , A_u(\theta_u^{n+1},V_h)-B(\theta_p^{n+1},V_h)=(R_u^{n+1},V_h), A u ( θ u n + 1 , V h ) − B ( θ p n + 1 , V h ) = ( R u n + 1 , V h ) ,
M p ( D t θ p n + 1 , Q h ) + B ( Q h , D t θ u n + 1 ) + A p ( θ p n + 1 , Q h ) = ( R p n + 1 , Q h ) . (39) \mathcal{M}_p(D_t\theta_p^{n+1},Q_h)+B(Q_h,D_t\theta_u^{n+1})+A_p(\theta_p^{n+1},Q_h)=(R_p^{n+1},Q_h). \tag*{(39)} M p ( D t θ p n + 1 , Q h ) + B ( Q h , D t θ u n + 1 ) + A p ( θ p n + 1 , Q h ) = ( R p n + 1 , Q h ) . (39)
where
( R u n , V h ) = B ( η p n , V h ) , (R_u^n,V_h)=B(\eta_p^n,V_h), ( R u n , V h ) = B ( η p n , V h ) ,
( R p n + 1 , Q h ) = − M p ( D t η p n + 1 , Q h ) − B ( Q h , D t η u n + 1 ) − A p ( η p n + 1 , Q h ) − M p ( ρ p n + 1 , Q h ) − B ( Q h , ρ u n + 1 ) . (R_p^{n+1},Q_h)=-\mathcal{M}_p(D_t\eta_p^{n+1},Q_h)-B(Q_h,D_t\eta_u^{n+1})-A_p(\eta_p^{n+1},Q_h)
-\mathcal{M}_p(\rho_p^{n+1},Q_h)-B(Q_h,\rho_u^{n+1}). ( R p n + 1 , Q h ) = − M p ( D t η p n + 1 , Q h ) − B ( Q h , D t η u n + 1 ) − A p ( η p n + 1 , Q h ) − M p ( ρ p n + 1 , Q h ) − B ( Q h , ρ u n + 1 ) .
The error system (39) has the same structure as the full discrete scheme. Applying Theorem 3 with loads R u n R_u^n R u n and R p n + 1 R_p^{n+1} R p n + 1 yields
max 0 ≤ n ≤ N ( ∥ θ u n ∥ V u 2 + ∥ θ p n ∥ V p 2 ) + τ ∑ n = 0 N − 1 ∥ θ p n + 1 ∥ V p 2 ≤ C [ ∥ θ u 0 ∥ V u 2 + ∥ θ p 0 ∥ V p 2 + max 0 ≤ n ≤ N ∥ R u n ∥ L 2 2 + τ ∑ n = 0 N − 1 ∥ D t R u n + 1 ∥ L 2 2 + τ ∑ n = 0 N − 1 ∥ R p n + 1 ∥ L 2 2 ] . (40) \max_{0\le n\le N}\left(\|\theta_u^n\|_{V_u}^2+\|\theta_p^n\|_{V_p}^2\right)+\tau\sum_{n=0}^{N-1}\|\theta_p^{n+1}\|_{V_p}^2
\le C\left[\|\theta_u^0\|_{V_u}^2+\|\theta_p^0\|_{V_p}^2+\max_{0\le n\le N}\|R_u^n\|_{L^2}^2+\tau\sum_{n=0}^{N-1}\|D_tR_u^{n+1}\|_{L^2}^2+\tau\sum_{n=0}^{N-1}\|R_p^{n+1}\|_{L^2}^2\right]. \tag*{(40)} 0 ≤ n ≤ N max ( ∥ θ u n ∥ V u 2 + ∥ θ p n ∥ V p 2 ) + τ n = 0 ∑ N − 1 ∥ θ p n + 1 ∥ V p 2 ≤ C [ ∥ θ u 0 ∥ V u 2 + ∥ θ p 0 ∥ V p 2 + 0 ≤ n ≤ N max ∥ R u n ∥ L 2 2 + τ n = 0 ∑ N − 1 ∥ D t R u n + 1 ∥ L 2 2 + τ n = 0 ∑ N − 1 ∥ R p n + 1 ∥ L 2 2 ] . (40)
Next, we bound the right-term of (40) . Continuity of the coupling form gives
∥ R u n ∥ L 2 ≤ C ∥ η p n ∥ V p , \|R_u^n\|_{L^2}\le C\|\eta_p^n\|_{V_p}, ∥ R u n ∥ L 2 ≤ C ∥ η p n ∥ V p ,
and, because the coefficients are independent of time,
∥ D t R u n + 1 ∥ L 2 ≤ C ∥ D t η p n + 1 ∥ V p . \|D_tR_u^{n+1}\|_{L^2}\le C\|D_t\eta_p^{n+1}\|_{V_p}. ∥ D t R u n + 1 ∥ L 2 ≤ C ∥ D t η p n + 1 ∥ V p .
Moreover,
∥ R p n + 1 ∥ L 2 ≤ C ( ∥ D t η p n + 1 ∥ V p + ∥ D t η u n + 1 ∥ V u + ∥ η p n + 1 ∥ V p + ∥ ρ p n + 1 ∥ V p + ∥ ρ u n + 1 ∥ V u ) . \|R_p^{n+1}\|_{L^2}\le C\left(\|D_t\eta_p^{n+1}\|_{V_p}+\|D_t\eta_u^{n+1}\|_{V_u}+\|\eta_p^{n+1}\|_{V_p}+\|\rho_p^{n+1}\|_{V_p}+\|\rho_u^{n+1}\|_{V_u}\right). ∥ R p n + 1 ∥ L 2 ≤ C ( ∥ D t η p n + 1 ∥ V p + ∥ D t η u n + 1 ∥ V u + ∥ η p n + 1 ∥ V p + ∥ ρ p n + 1 ∥ V p + ∥ ρ u n + 1 ∥ V u ) .
Since the projections are independent of time,
D t η u n + 1 = 1 τ ∫ t n t n + 1 ( I − I h ) ∂ t U ( s ) d s , D_t\eta_u^{n+1}=\frac{1}{\tau}\int_{t_n}^{t_{n+1}}(I-I_h)\partial_tU(s)\,ds, D t η u n + 1 = τ 1 ∫ t n t n + 1 ( I − I h ) ∂ t U ( s ) d s ,
D t η p n + 1 = 1 τ ∫ t n t n + 1 ( I − Π h ) ∂ t P ( s ) d s . D_t\eta_p^{n+1}=\frac{1}{\tau}\int_{t_n}^{t_{n+1}}(I-\Pi_h)\partial_tP(s)\,ds. D t η p n + 1 = τ 1 ∫ t n t n + 1 ( I − Π h ) ∂ t P ( s ) d s .
Hence
τ ∑ n = 0 N − 1 ∥ D t η u n + 1 ∥ V u 2 ≤ C h 2 k ∥ ∂ t U ∥ L 2 ( 0 , T ; H k + 1 ) 2 , \tau\sum_{n=0}^{N-1}\|D_t\eta_u^{n+1}\|_{V_u}^2\le Ch^{2k}\|\partial_tU\|_{L^2(0,T;H^{k+1})}^2, τ n = 0 ∑ N − 1 ∥ D t η u n + 1 ∥ V u 2 ≤ C h 2 k ∥ ∂ t U ∥ L 2 ( 0 , T ; H k + 1 ) 2 ,
τ ∑ n = 0 N − 1 ∥ D t η p n + 1 ∥ V p 2 ≤ C h 2 k ∥ ∂ t P ∥ L 2 ( 0 , T ; H k + 1 ) 2 . (41) \tau\sum_{n=0}^{N-1}\|D_t\eta_p^{n+1}\|_{V_p}^2\le Ch^{2k}\|\partial_tP\|_{L^2(0,T;H^{k+1})}^2. \tag*{(41)} τ n = 0 ∑ N − 1 ∥ D t η p n + 1 ∥ V p 2 ≤ C h 2 k ∥ ∂ t P ∥ L 2 ( 0 , T ; H k + 1 ) 2 . (41)
By Lemma 2,
∥ η u n ∥ V u ≤ C h k ∥ U ( t n ) ∥ H k + 1 , ∥ η p n ∥ V p ≤ C h k ∥ P ( t n ) ∥ H k + 1 . \|\eta_u^n\|_{V_u}\le Ch^k\|U(t_n)\|_{H^{k+1}},\qquad\|\eta_p^n\|_{V_p}\le Ch^k\|P(t_n)\|_{H^{k+1}}. ∥ η u n ∥ V u ≤ C h k ∥ U ( t n ) ∥ H k + 1 , ∥ η p n ∥ V p ≤ C h k ∥ P ( t n ) ∥ H k + 1 .
Combining these bounds with (36) yields
max n ∥ R u n ∥ V u 2 + τ ∑ n = 0 N − 1 ∥ D t R u n + 1 ∥ V u 2 + τ ∑ n = 0 N − 1 ∥ R p n + 1 ∥ V p 2 ≤ C h 2 k ( ∥ U ∥ H 1 ( 0 , T ; H k + 1 ) 2 + ∥ P ∥ H 1 ( 0 , T ; H k + 1 ) 2 ) + C τ 2 ( ∥ ∂ t t U ∥ L 2 ( 0 , T ; V u ) 2 + ∥ ∂ t t P ∥ L 2 ( 0 , T ; V p ) 2 ) . (42) \max_n \lVert R_u^n\rVert_{V_u}^2+\tau\sum_{n=0}^{N-1}\lVert D_tR_u^{n+1}\rVert_{V_u}^2+\tau\sum_{n=0}^{N-1}\lVert R_p^{n+1}\rVert_{V_p}^2
\leq Ch^{2k}\left(\lVert U\rVert_{H^1(0,T;H^{k+1})}^2+\lVert P\rVert_{H^1(0,T;H^{k+1})}^2\right)
+\ C\tau^2\left(\lVert\partial_{tt}U\rVert_{L^2(0,T;V_u)}^2+\lVert\partial_{tt}P\rVert_{L^2(0,T;V_p)}^2\right). \tag*{(42)} n max ∥ R u n ∥ V u 2 + τ n = 0 ∑ N − 1 ∥ D t R u n + 1 ∥ V u 2 + τ n = 0 ∑ N − 1 ∥ R p n + 1 ∥ V p 2 ≤ C h 2 k ( ∥ U ∥ H 1 ( 0 , T ; H k + 1 ) 2 + ∥ P ∥ H 1 ( 0 , T ; H k + 1 ) 2 ) + C τ 2 ( ∥ ∂ tt U ∥ L 2 ( 0 , T ; V u ) 2 + ∥ ∂ tt P ∥ L 2 ( 0 , T ; V p ) 2 ) . (42)
Since P h 0 = Π h P 0 P_h^0=\Pi_hP_0 P h 0 = Π h P 0 , we have θ p 0 = 0 \theta_p^0=0 θ p 0 = 0 . Subtracting the continuous and discrete equilibrium equations at t = 0 t=0 t = 0 and using the elliptic projection property gives
A u ( θ u 0 , V h ) = B ( η p 0 , V h ) . \mathcal{A}_u(\theta_u^0,V_h)=\mathcal{B}(\eta_p^0,V_h). A u ( θ u 0 , V h ) = B ( η p 0 , V h ) .
Taking V h = θ u 0 V_h=\theta_u^0 V h = θ u 0 and using coercivity yields
∥ θ u 0 ∥ V u ≤ C ∥ η p 0 ∥ V p ≤ C h k ∥ P 0 ∥ H k + 1 ≤ C h k ∥ P ∥ H 1 ( 0 , T ; H k + 1 ) . \lVert\theta_u^0\rVert_{V_u}\leq C\lVert\eta_p^0\rVert_{V_p}\leq Ch^k\lVert P_0\rVert_{H^{k+1}}\leq Ch^k\lVert P\rVert_{H^1(0,T;H^{k+1})}. ∥ θ u 0 ∥ V u ≤ C ∥ η p 0 ∥ V p ≤ C h k ∥ P 0 ∥ H k + 1 ≤ C h k ∥ P ∥ H 1 ( 0 , T ; H k + 1 ) .
Insert the initial error bounds and (42) into (40) . Then combine the resulting estimate with
e u n = η u n + θ u n , e p n = η p n + θ p n , e_u^n=\eta_u^n+\theta_u^n,\qquad e_p^n=\eta_p^n+\theta_p^n, e u n = η u n + θ u n , e p n = η p n + θ p n ,
the projection estimates embedding. This proves (35) .
Numerical results
Having established the well-posedness, stability, and convergence in § 3 and § 4 , we now present illustrative numerical results. The main goal of this section is to go beyond the theoretical results and to demonstrate the accuracy of the multicontinuum model for complex multiscale media, where the exact solution is taken from the microscale. For this purpose, we solve the fine-scale Biot model (1) on a fine grid that resolves all heterogeneities and regard its averaged solution as exact. Specifically, the fine-scale pressure and displacements are averaged over each continuum in every coarse block and compared with the corresponding macroscopic variables P i P_i P i and U i U_i U i in the relative L 2 L^2 L 2 norm. Thus, the obtained errors reflect not only the discretization error but also the modeling error of the multicontinuum approach.
We consider the computational domain Ω = ( 0 , 1 ) × ( 0 , 1 ) \Omega=(0,1)\times(0,1) Ω = ( 0 , 1 ) × ( 0 , 1 ) occupied by a high-contrast heterogeneous medium. The domain is divided into two non-overlapping subregions, Ω = Ω 1 ∪ Ω 2 \Omega=\Omega_1\cup\Omega_2 Ω = Ω 1 ∪ Ω 2 , which define the two continua (N c = 2 N_c=2 N c = 2 ). In our experiments, we consider both periodic and non-periodic microstructures presented in Figure 1 , where Ω 1 \Omega_1 Ω 1 is blue and Ω 2 \Omega_2 Ω 2 is red.
Figure 1. Microstructures (Ω 1 \Omega_1 Ω 1 is blue, Ω 2 \Omega_2 Ω 2 is red). Left: periodic. Right: non-periodic.
For our numerical test, we set the following heterogeneous Lamé parameters λ \lambda λ [Pa] and μ \mu μ [Pa] and the permeability coefficient κ \kappa κ [m 2 \mathrm{m}^2 m 2 / (Pa s)]
λ = { 10 5 , x ∈ Ω 1 , 10 9 , x ∈ Ω 2 , μ = { 10 5 , x ∈ Ω 1 , 10 9 , x ∈ Ω 2 , κ = { 10 − 10 , x ∈ Ω 1 , 10 − 6 , x ∈ Ω 2 . (43) \lambda=
\begin{cases}
10^5, & x \in\Omega_1,\\
10^9, & x \in\Omega_2
\end{cases}
,\qquad
\mu=
\begin{cases}
10^5, & x \in\Omega_1,\\
10^9, & x \in\Omega_2
\end{cases}
,\qquad
\kappa=
\begin{cases}
10^{-10}, & x \in\Omega_1,\\
10^{-6}, & x \in\Omega_2
\end{cases}
. \tag*{(43)} λ = { 1 0 5 , 1 0 9 , x ∈ Ω 1 , x ∈ Ω 2 , μ = { 1 0 5 , 1 0 9 , x ∈ Ω 1 , x ∈ Ω 2 , κ = { 1 0 − 10 , 1 0 − 6 , x ∈ Ω 1 , x ∈ Ω 2 . (43)
We impose the boundary conditions
u 1 = 0 , ( σ ⋅ ν ) 2 = 0 , x ∈ Γ R , u 2 = 0 , ( σ ⋅ ν ) 1 = 0 , x ∈ Γ B , σ ⋅ ν = 0 , x ∈ Γ L ∪ Γ T , − κ ∂ p ∂ ν = 0 , x ∈ ∂ Ω . (44) \begin{aligned}
u_1 &= 0,\qquad(\sigma\cdot\nu)_2 = 0,\qquad x \in\Gamma_R,\\
u_2 &= 0,\qquad(\sigma\cdot\nu)_1 = 0,\qquad x \in\Gamma_B,\\
\sigma\cdot\nu&= 0,\qquad x \in\Gamma_L \cup\Gamma_T,\\
-\kappa\frac{\partial p}{\partial\nu} &= 0,\qquad x \in\partial\Omega. \tag*{(44)}
\end{aligned} u 1 u 2 σ ⋅ ν − κ ∂ ν ∂ p = 0 , ( σ ⋅ ν ) 2 = 0 , x ∈ Γ R , = 0 , ( σ ⋅ ν ) 1 = 0 , x ∈ Γ B , = 0 , x ∈ Γ L ∪ Γ T , = 0 , x ∈ ∂ Ω. (44)
where ν \nu ν is a unit outward normal vector, and Γ L \Gamma_L Γ L , Γ R \Gamma_R Γ R , Γ B \Gamma_B Γ B , and Γ T \Gamma_T Γ T denote the left, right, bottom, and top boundaries, respectively, such that ∂ Ω = Γ L ∪ Γ R ∪ Γ B ∪ Γ T \partial\Omega= \Gamma_L \cup\Gamma_R \cup\Gamma_B \cup\Gamma_T ∂ Ω = Γ L ∪ Γ R ∪ Γ B ∪ Γ T . Although the analysis in § 3 and § 4 was presented for homogeneous Dirichlet conditions for simplicity, similar arguments apply to these boundary conditions with minor modifications.
The body force f = ( f 1 , f 2 ) \boldsymbol{f} = (f_1, f_2) f = ( f 1 , f 2 ) [N / m 3 \mathrm{N}/\mathrm{m}^3 N / m 3 ] and the fluid source g g g [s − 1 \mathrm{s}^{-1} s − 1 ] are given by
f 1 = − 10 7 sin ( 2 π x 1 ) sin ( π x 2 ) , f 2 = 10 7 sin ( π x 1 ) sin ( π x 2 ) , g = exp ( − 40 [ ( x 1 − 0.5 ) 2 + ( x 2 − 0.5 ) 2 ] ) . (45) \begin{aligned}
f_1 &= -10^7 \sin(2\pi x_1)\sin(\pi x_2),\qquad
f_2 = 10^7 \sin(\pi x_1)\sin(\pi x_2),\\
g &= \exp\left(-40\left[(x_1-0.5)^2+(x_2-0.5)^2\right]\right). \tag*{(45)}
\end{aligned} f 1 g = − 1 0 7 sin ( 2 π x 1 ) sin ( π x 2 ) , f 2 = 1 0 7 sin ( π x 1 ) sin ( π x 2 ) , = exp ( − 40 [ ( x 1 − 0.5 ) 2 + ( x 2 − 0.5 ) 2 ] ) . (45)
The final time is T = 5 T = 5 T = 5 s, and the initial pressure is p 0 = 10 6 p_0 = 10^6 p 0 = 1 0 6 Pa. The initial displacement is u 0 = ( 0 , 0 ) u_0 = (0, 0) u 0 = ( 0 , 0 ) m.
For the periodic microstructure, we employ uniform fine grid consisting of 160801 vertices and 320000 triangular cells. We construct an unstructured fine grid consisting of 85085 vertices and 169172 triangular cells for the non-periodic microstructure. The fine-scale Biot model (1) is solved using linear finite elements. For the multicontinuum model, we use two uniform coarse grids: 10 × 10 10 \times10 10 × 10 (121 vertices and 100 square cells) and 20 × 20 20 \times20 20 × 20 (441 vertices and 400 square cells). Each coarse block is taken as the RVE, and the oversampled region is obtained by extending the RVE by l l l layers of coarse blocks, with l = 5 l = 5 l = 5 for the 10 × 10 10 \times10 10 × 10 grid and l = 6 l = 6 l = 6 for the 20 × 20 20 \times20 20 × 20 grid. The macroscopic equations are discretized using bilinear finite elements on the coarse grid. In time, both the fine-scale and multicontinuum problems are solved with the backward Euler method using 50 uniform time steps. The numerical implementation is based on the FEniCS package [22 ] , and the fine grid is generated using Gmsh [23 ] .
The coarse-grid results are based on the multicontinuum model (8) . It should be noted that we solve split constraint cell problems in the oversampled RVE for the periodic microstructure. For the non-periodic microstructure, we employ coupled approach by solving coupled cell problems (following [14 ] ).
To compare the multicontinuum and reference solutions, we use the following relative L 2 L^2 L 2 errors for pressure and displacements at the final time
e p ( i ) = ∑ K ∣ 1 ∣ K ∣ ∫ K P i d x − 1 ∣ K ∩ Ω i ∣ ∫ K ∩ Ω i p d x ∣ 2 ∑ K ∣ 1 ∣ K ∩ Ω i ∣ ∫ K ∩ Ω i p d x ∣ 2 , e_p^{(i)} = \sqrt{\frac{\sum_K \left|\frac{1}{|K|}\int_K P_i\,\mathrm{d}x-\frac{1}{|K\cap\Omega_i|}\int_{K\cap\Omega_i}p\,\mathrm{d}x\right|^2}{\sum_K\left|\frac{1}{|K\cap\Omega_i|}\int_{K\cap\Omega_i}p\,\mathrm{d}x\right|^2}}, e p ( i ) = ∑ K ∣ K ∩ Ω i ∣ 1 ∫ K ∩ Ω i p d x 2 ∑ K ∣ K ∣ 1 ∫ K P i d x − ∣ K ∩ Ω i ∣ 1 ∫ K ∩ Ω i p d x 2 ,
e u ( i ) = ∑ K ∣ 1 ∣ K ∣ ∫ K U i d x − 1 ∣ K ∩ Ω i ∣ ∫ K ∩ Ω i u d x ∣ 2 ∑ K ∣ 1 ∣ K ∩ Ω i ∣ ∫ K ∩ Ω i u d x ∣ 2 . e_u^{(i)} = \sqrt{\frac{\sum_K \left|\frac{1}{|K|}\int_K U_i\,\mathrm{d}x-\frac{1}{|K\cap\Omega_i|}\int_{K\cap\Omega_i}u\,\mathrm{d}x\right|^2}{\sum_K\left|\frac{1}{|K\cap\Omega_i|}\int_{K\cap\Omega_i}u\,\mathrm{d}x\right|^2}}. e u ( i ) = ∑ K ∣ K ∩ Ω i ∣ 1 ∫ K ∩ Ω i u d x 2 ∑ K ∣ K ∣ 1 ∫ K U i d x − ∣ K ∩ Ω i ∣ 1 ∫ K ∩ Ω i u d x 2 .
where i = 1 , 2 i=1,2 i = 1 , 2 denotes the continuum number, K K K is the coarse block taken as the RVE, and ( u , p ) (u,p) ( u , p ) is the fine-scale reference solution.
Figure 2 and Figure 3 show the reference solutions at the final time for the periodic and non-periodic microstructures, respectively. In both cases, the fine-grid fields exhibit small-scale oscillations induced by the heterogeneous microstructure. The pressure builds up around the source located at the center of the domain and remains elevated in the low-permeability subregion Ω 1 \Omega_1 Ω 1 . The horizontal displacement changes sign across the domain, indicating that the medium is pulled apart in the x 1 x_1 x 1 direction, whereas the vertical displacement is positive and peaks in the upper central part, which reflects the upward deformation. On the coarse grid 20 × 20 20 \times20 20 × 20 , the continuum averages retain these features. The averaged pressure is more widely distributed in Ω 2 \Omega_2 Ω 2 , where the permeability is higher, and is more concentrated and larger near the source in Ω 1 \Omega_1 Ω 1 . Both averaged displacement components have larger magnitudes in the softer subregion Ω 1 \Omega_1 Ω 1 . The non-periodic microstructure yields larger pressure and displacement values than the periodic one, but the overall behavior of the solutions is similar.
Figure 2. (a) Fine-grid solution. (b) Average solution in Ω 1 \Omega_1 Ω 1 on the coarse grid 20 × 20 20 \times20 20 × 20 . (c) Average solution in Ω 2 \Omega_2 Ω 2 on the coarse grid 20 × 20 20 \times20 20 × 20 . Distributions of reference pressure [Pa] and displacements [m] in x 1 x_1 x 1 and x 2 x_2 x 2 directions (from left to right) at the final time. Periodic microstructure.
(a) Fine-grid solution
(b) Average solution in Ω 1 \Omega_1 Ω 1 on the coarse grid 20 × 20
(c) Average solution in Ω 2 \Omega_2 Ω 2 on the coarse grid 20 × 20
Figure 3. Distributions of reference pressure [Pa] and displacements [m] in x 1 x_1 x 1 and x 2 x_2 x 2 directions (from left to right) at the final time. Non-periodic microstructure.
Figure 2 b–Figure 2 c and Figure 3b –Figure 3c present the reference solutions averaged over each continuum on the coarse grid 20 × 20 20 \times20 20 × 20 . The averaged fields preserve the main features of the fine-grid solutions while removing the small-scale oscillations. The averaged pressure in Ω 1 \Omega_1 Ω 1 is concentrated near the source and attains larger values, whereas in Ω 2 \Omega_2 Ω 2 it is more widely distributed owing to the higher permeability. For both displacement components, the averages in the softer subregion Ω 1 \Omega_1 Ω 1 have larger magnitudes than those in Ω 2 \Omega_2 Ω 2 . Thus, the two continua exhibit distinct responses, which motivates the use of separate macroscopic variables for each of them.
Table 1 presents the relative L 2 L^2 L 2 errors at the final time for both microstructures and coarse grids. In all cases, the errors are small: below 0.3 % 0.3\% 0.3% for the pressure and below 5 % 5\% 5% for the displacements. Moreover, the errors decrease as the coarse grid is refined. For the non-periodic microstructure, the displacement errors are larger and decrease more slowly than in the periodic case. Overall, the results demonstrate that the multicontinuum model provides an accurate approximation of the averaged microscale solution for both periodic and non-periodic media.
Table 1. Relative L 2 L_2 L 2 errors for different microstructures and coarse grids.
In summary, the numerical results show that the multicontinuum model accurately reproduces the averaged microscale solution for both periodic and non-periodic high-contrast media, with errors
that decrease as the coarse grid is refined. These results complement the theoretical analysis by demonstrating the accuracy of the model with respect to the underlying microscale problem.
Conclusion
In this work, we developed and analyzed a macroscopic multicontinuum model for Biot poroelasticity. Starting from the weak formulation of the classical Biot system, we constructed continuum-dependent Galerkin approximation spaces enriched with first-order multiscale corrections. By projecting the microscopic variational formulation onto these spaces, we systematically derived the corresponding macroscopic multicontinuum equations and their effective coefficients. In particular, the resulting model retains the generalized pressure storage and coupling terms arising from the first-order multiscale approximation, providing a consistent variational framework for describing interactions among multiple continua and the deformable solid matrix.
We investigated the mathematical properties of the full macroscopic multicontinuum system. Under suitable assumptions on the effective coefficients, we established the existence and uniqueness of weak solutions and derived an energy stability estimate. By exploiting the properties of the effective coupling operators, we obtained an energy structure in which the coupling contributions cancel exactly. Based on this structure, we developed a fully discrete Galerkin finite element scheme using the backward Euler method for temporal discretization. The well-posedness and stability of the fully discrete scheme were established, and a priori error estimates were derived through suitable projection operators and discrete energy arguments. Numerical experiments for periodic and non-periodic high-contrast media demonstrate the accuracy of the multicontinuum model with respect to the averaged fine-scale solution.