HAL Id: hal-00680686
https://hal.archives-ouvertes.fr/hal-00680686
Preprint submitted on 19 Mar 2012
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Hybrid finite volume scheme for two-phase flow in porous media
Konstantin Brenner
To cite this version:
Konstantin Brenner. Hybrid finite volume scheme for two-phase flow in porous media. 2011. �hal-
00680686�
Hybrid finite volume scheme for two-phase flow in porous media
Konstantin Brenner
∗March 6, 2012
Abstract We apply a finite volume method on general meshes for the discretization of an incompressible and immiscible two-phase flow in porous media. The problem is con- sidered in the global pressure formulation. Mathematically, it amounts to solve an elliptic equation for the global pressure, with an anisotropic and heterogeneous permeability ten- sor coupled to a parabolic degenerate convection-diffusion equation for a saturation, again with the same permeability tensor. Extending ideas which we had previously developed for the numerical solution of a degenerate parabolic convection-reaction-diffusion equation we discretize the diffusion terms by means of a hybrid finite volume scheme, while we use a Godunov scheme for the non monotone convection flux. We prove the convergence of the numerical scheme in arbitrary space dimension and we present results of a number of numerical tests in space dimension two.
This work was supported by the GdR MoMaS (PACEN/CNRS, ANDRA, BRGM, CEA, EdF, IRSN), France.
1 Introduction
The two-phase flow in porous media is an important problem arising in many engineering and scientific context as e.g the secondary oil recovery, the basin modeling and the con- taminated sites remediation [9], [7], [13]. In this paper we consider the simplified dead-oil model, that is to say, we assume that there are only two incompressible and immiscible fluids. The problem is described by the mass conservation of each phase together with Darcy-Muskat law
ω∂
ts
o− ∇ ·
K kr
o(s
o) µ
o( ∇ p
o− ρ
og)
= k
o(1.1)
ω∂
ts
w− ∇ ·
K kr
w(s
w) µ
w( ∇ p
w− ρ
wg)
= k
w(1.2)
where ω is the porosity, K absolute permeability, s
wstands for the wetting phase (water) saturation and s
ofor the non-wetting phase (oil) saturation. The relative permeabilities kr
o(s
o) and kr
w(s
w) model the effects of coexistence of both phases in porous medium;
ρ
αand µ
adenote the densities and the viscosity of phase α ∈ { o, w } , g is the gravity vector. It is also assumed that the porous medium is saturated and that two phases are immiscible
s
o+ s
w= 1. (1.3)
∗Laboratoire de Math´ematiques, Universit´e de Paris-Sud 11, F-91405 Orsay Cedex, France
The phase pressures are related by a capillary pressure law
p
o+ p
w= π(s). (1.4)
In view of (1.3) we can take as unknown only one saturation, for instance s = s
o. We denote by λ
o(s) = kr
o(s)
µ
oand λ
w(s) = kr
w(1 − s) µ
wthe relative mobilities, which are such that λ
o(0) = λ
w(1) = 0. In this paper we transform the system (1.1)-(1.4) into the so-called global pressure or fractional flow formulation. The global pressure p which is defined by
p = p
w+ Z
s0
λ
o(a)
λ
o(a) + λ
w(a) π(a)da
was first introduced in [5] and [13]. We also define the fractional flow f , total mobility λ and capillary diffusion ϕ
λ(s) = λ
o(s) + λ
w(s), f (s) = λ
o(s)
λ(s) , ϕ(s) = Z
s0
λ
w(τ )f (τ)π
′(τ )dτ.
The system (1.1)-(1.4) is equivalent to
−∇ · K (λ(s) ∇ p − ξ(s)g) = k
w+ k
oin Ω × (0, T ), (1.5) q = − K (λ(s) ∇ p − ξ(s)g) in Ω × (0, T ), (1.6) ω ∂s
∂t + ∇ · (qf (s) + γ(s)Kg) − ∇ · (K ∇ ϕ(s)) = k
oin Ω × (0, T ), (1.7) where
γ(s) = (ρ
o− ρ
w) λ
o(s)λ
w(s)
λ
o(s) + λ
w(s) and ξ(s) = (λ
o(s)ρ
o+ λ
w(s)ρ
w) .
The usual assumption λ
o+ λ
w≥ λ > 0 implies that the first equation is uniformly elliptic in p, whereas the second one is parabolic degenerate with respect to the saturation s.
The system (1.5)-(1.7) has a remarkable mathematical structure, which permits to obtain a number of energy estimates. Many numerical methods were proposed for solving two- phase problem, such as finite element methods (see e.g. [21], [20], [13], [6], [14], [15], [16], [17]), discontinuous Galerkin (see e.g. [8], [19]) and finite volume methods (see e.g.
[1], [23], [29], [25], [2], [30], [4]). The convergence results for the different finite elements schemes were obtained in [6], [14] and [17]. For finite volume schemes convergence was shown in [1], [29] and [25], where the initial formulation (1.1)-(1.4) was considered. We also would like to mention some interesting convergence results which were obtained in the case of miscible displacement [31], [33], [12].
The heterogeneity and anisotropy of porous media is a numerical challenge even when studying the elliptic problem derived from Darcy’s law for one-phase problem. Many schemes was proposed and analyzed in last decades for its discretization. See [27] and [24]
for more references and for the detailed description and comparison of those numerical
methods. In this paper we propose an implicit fully coupled Hybrid finite volume scheme
for the system (1.5)-(1.7). The discretization of the diffusion terms is based upon a hy-
brid finite volume method [22], which allows the tensor K to be anisotropic and highly
variable in space. Remark that heterogeneity also affects the first order terms K(x)gξ( · )
and qf ( · ) + K(x)gγ( · ), which may be discontinuous with respect to the space variable
x; in that case they require a suitable treatment see e.g. [35], [34], [26]. We apply the
Godunov scheme proposed in [28], which seems to be a natural choice, since the hybrid
(interface) unknowns are used.
Assumptions on the data: ( H
1) ϕ ∈ C( R ), ϕ(0) = 0, is a strictly increasing piece- wise continuously differentiable Lipschitz continuous function with a Lipschitz constant L
ϕ. We assume that the function ϕ
−1is H¨older continuous, namely that there exists H
ϕ> 0 and α ∈ (0, 1] such that | s
1− s
2| ≤ H
ϕ| ϕ(s
1) − ϕ(s
2) |
α. It is also such that ϕ(s) = s for all s < 0 and ϕ(s) − ϕ(1) = s − 1 for all s > 1;
( H
2) The functions λ, ξ, γ, f ∈ C([0, 1]) are Lipschitz continuous; for f = λ, ξ, γ, f we denote by L
fthe corresponding Lipschitz constant.
( H
2a) λ is such that 0 < λ ≤ λ(s);
( H
2b) ξ and γ are convex functions, moreover γ is such that γ(0) = γ(1) = 0; g is a constant vector from R
d;
( H
2c) f is a nondecreasing function and it satisfies f(0) = 0, f (1) = 1;
( H
3) The functions λ, ξ, γ, f are constant outside of (0, 1), i.e. for f = λ, ξ, γ, f we assume that f(s) = f(0) for all s < 0 and f(s) = f(1) for all s > 1.
( H
5) s
0∈ L
∞(Ω); ω ∈ L
∞(Ω) and such that 0 < ω ≤ ω(x) ≤ ω for a.e. in x ∈ Ω;
( H
6) k
o, k
w∈ L
∞(0, T ; L
2(Ω)) and such that k
o+ k
w≥ 0 a.e. in Q
T.
Remark 1.1 The physical range of values for the saturation is [0, 1], however we extend the definition of all nonlinear functions in (1.5)-(1.7) outside of [0, 1], since the numerical scheme which we study does not preserve neither the maximum principle, nor the posi- tivity of s (nor the bound s ≤ 1). It is worth noting that one can replace ( H
3) by the assumption that the functions λ, ξ, γ, f are in L
∞( R ). In turn the assumption ( H
1) can also weakened, namely the capillary diffusion ϕ can be extended by an arbitrary strictly increasing function, such that ϕ
−1is Lipschitz continuous on R \ (0, 1).
Assumptions on the geometry:
( H
4a) Ω is a polyhedral open bounded connected subset of R
d, with d ∈ N
∗, and ∂Ω = Ω \ Ω its boundary.
( H
4b) K is a piecewise constant function from Ω to M
d( R ), where M
d( R ) denotes the set of real d × d matrices. More precisely we assume that there exist a finite family (Ω
i)
i∈{1,...,I}of open connected polyhedral in R
d, such that Ω = S
i∈{1,...,I}
Ω
i, Ω
iT
Ω
j= ∅ if i 6 = j and such that K(x) |
Ωi= K
i∈ M
d( R ). By Γ
i,jwe denote the interface between the sub-domains i, j, Γ
i,j= Ω
iT
Ω
j. We suppose that there exist two positive constants K and K such that for all i ∈ { 1, . . . , I } the eigenvalues of the symmetric positive definite K
iare included in [K, K].
Remark 1.2 For the sake of simplicity we have assumed that the heterogeneity of the medium is only expressed through the x-dependence of the absolute permeability tensor K = K(x). However it is very simple to extend the analysis to the case where λ, ξ, γ and f depend on the rock type. On the other hand if we suppose that ϕ is discontinuous in space, this may lead to significant difficulties. The analysis of the case where the capillary pressure field (and also ϕ) is discontinuous was carried out in [11] and [10]. It is also worth noting that the partitioning of Ω introduced in H
4bis only used in order to provide a control on the gravity terms (see Remark 3.2) below). In the case that the gravity effects are neglected, one can consider a fully heterogenous permeability field K.
Let T > 0, we consider the system (1.5)-(1.7) in the domain Q
T= Ω × (0, T ) together with the initial condition
s( · , 0) = s
0in Ω (1.8)
and homogeneous Dirichlet boundary conditions
p = 0 and s = 0 on ∂Ω × (0, T ). (1.9)
We now present the definition of a weak solution of the problem (1.5)-(1.9).
Definition 1.1 (Weak solution) A function pair (s, p) is a weak solution of the problem (1.5)-(1.6) if
(i) s ∈ L
∞(0, T ; L
2(Ω));
(ii) ϕ(s) ∈ L
2(0, T ; H
01(Ω));
(iii) p ∈ L
∞(0, T ; H
01(Ω));
(iv) for all ψ, χ ∈ L
2(0, T ; H
01(Ω)) with ϕ
t∈ L
∞(Q
T), ϕ( · , T ) = 0, s and p satisfy the integral equalities
− Z
T0
Z
Ω
sψ
tdxdt − Z
Ω
s
0ψ( · , 0) dx − Z
T0
Z
Ω
K(f(s)q + γ(s)g) · ∇ ψ dxdt +
Z
T 0Z
Ω
K ∇ ϕ(s) · ∇ ψ dxdt = Z
T0
Z
Ω
k
oψ dxdt and
− Z
T0
Z
Ω
q · ∇ χ dxdt = Z
T0
Z
Ω
(k
w+ k
o)χ dxdt, where q is given by (1.6).
In Section 2 we present the finite volume scheme and some technical lemmas. In Section 3 we provide the a priori estimates and we prove an existence of a discrete solution. We prove the estimates on space and time translates of a discrete solution in Section 4, those estimates allow to establish a strong convergence property for a subsequence of discrete saturation. The convergence result is shown in Section 5. Finally in Section 6 we present a number of numerical results obtained on different two-dimensional meshes.
2 The finite volume scheme
2.1 The main definitions
In order to describe the numerical scheme we introduce below some notations related to the space and time discretization, which follows [22].
Definition 2.1 (Discretization of Ω) Let Ω be a polyhedral open bounded connected subset of R
d, with d ∈ N
∗, ∂Ω = Ω \ Ω its boundary, and (Ω
i)
i∈{1,...,I}it’s partition in the sense of ( H
4). A discretization of Ω, denoted by D , is defined as the triplet D = ( M , E , P ), where:
1. M is a finite family of non empty connex open disjoint subsets of Ω (the ”control volumes”) such that Ω = S
K∈M
K. For any K ∈ M , let ∂K = K \ K be the boundary of K; we define m(K) > 0 as the measure of K and h
Kas the diameter of K. We also assume that the mesh resolve the structure of the medium, i.e. for all K ∈ M there exist i ∈ { 1, . . . , I } such that K ⊂ Ω
i.
2. E is a finite family of disjoint subsets of Ω (the ”edges” of the mesh), such that, for all σ ∈ E , σ is a non empty open subset of a hyperplane of R
d, whose (d − 1)-dimensional measure m(σ) is strictly positive. We also assume that, for all K ∈ M , there exists a subset E
Kof E such that ∂K = S
σ∈EK
σ. For each σ ∈ E , we set M
σ= { K ∈ M| σ ∈ E
K} .
We then assume that, for all σ ∈ E , either M
σhas exactly one element and then σ ∈ ∂Ω
(the set of these interfaces called boundary interfaces, is denoted by E
ext) or M
σhas exactly two elements (the set of these interfaces called interior interfaces, is denoted by E
int). For all σ ∈ E , we denote by x
σthe barycenter of σ. For all K ∈ M and σ ∈ E
K, we denote by n
K,σthe outward normal unit vector.
3. P is a family of points of Ω indexed by M , denoted by P = (x
K)
K∈M, such that for all K ∈ M , x
K∈ K ; moreover K is assumed to be x
K-star-shaped, which means that for all x ∈ K, there holds [x
K, x] ∈ K. Denoting by d
K,σthe Euclidean distance between x
Kand the hyperplane containing σ, one assumes that d
K,σ> 0. We denote by D
K,σthe cone of vertex x
Kand basis σ.
Next we introduce some extra notations related to the mesh. The size of the discretization D is defined by
h
D= sup
K∈M
diam(K); (2.1)
moreover we define
θ
D= max( max
σ∈Eint,{K,L}=Mσ
d
K,σd
L,σ, max
K∈Mσ,σ∈EK
h
Kd
K,σ). (2.2)
Imposing a uniform bound on θ
Dforces the mesh to be sufficiently regular. Next, we define several discrete spaces, which are going to be used in the sequel.
Definition 2.2 (The hybrid space X
D(Ω)) Let D = ( M , E , P ) be a discretization of Ω. We define
X
D= { ((v
K)
K∈M, (v
σ)
σ∈E), v
K∈ R , v
σ∈ R } ,
X
D,0= { v ∈ X
Dsuch that (v
σ)
σ∈Eext= 0 } . (2.3) The space X
Dis equipped with the semi-norm | · |
XDdefined by
| v |
2XD= X
K∈M
| v |
2XD,K, where | v |
2XD,K= X
σ∈EK
m(σ) d
K,σ(v
σ− v
K)
2for all v ∈ X
D. (2.4) Note that | · |
XDis a norm on the space X
D,0.
Moreover, for each function ψ = ψ(x) regular enough we define its projection P
Dψ ∈ X
Don the space X
Din following way
(P
Dψ)
K= ψ(x
K) for all K ∈ M , (P
Dψ)
σ= ψ (x
σ) for all σ ∈ E .
Definition 2.3 (The discrete flux space Q
D(Ω)) Let D = ( M , E , P ) be a discretiza- tion of Ω. We define
Q
D= { (q
K,σ)
K∈M,σ∈EK, q
K,σ∈ R } . (2.5) Next we introduce the time discretization.
Definition 2.4 (Time discretization) We divide the time interval (0, T ) into N equal time steps of length δt = T /N , where δt is the uniform time step defined by δt = t
n− t
n−1. Taking into account the time discretization leads us to define of the following discrete spaces
X
D,δt= X
DN= { (v
n)
n∈{1,...,N}, v
n∈ X
D} and
X
D,δt,0= X
D,0N= { (v
n)
n∈{1,...,N}, v
n∈ X
D,0} ; moreover we define the following semi-norm on X
D,δt| v |
2XD,δt=
N
X
n=1
δt | v |
2XD. (2.6)
2.2 The numerical scheme
2.2.1 The discrete problem
In this section we present the fully implicit finite volume scheme for the problem (1.5)- (1.9). Let us introduce the discrete saturation (s
nK)
K∈M, (s
nσ)
σ∈En∈{1,...,N}
∈ X
D,δtand the discrete global pressure (p
nK)
K∈M, (p
nσ)
σ∈En∈{1,...,N}
∈ X
D,δt, which are the main discrete unknowns. Moreover let f denote λ, ξ, γ or f we introduce the following notation f
nK= f(s
nK) and f
nσ= f(s
nσ) for all K ∈ M , σ ∈ E and n ∈ { 1, . . . , N } . Let k
i,Kndenote the mean value of the source term k
i(x, t) over a cell K × (t
n−1, t
n), i.e.
k
ni,K= 1 m(K )δt
Z
tntn−1
Z
K
k
i(x, t) dxdt with i ∈ { w, n } . (2.7) We denote the porous volume of the element K by ω(K),
ω(K) = Z
K
ω(x) dx.
Next, let Q
nK,σbe an approximation of the total flux through the interface σ Q
nK,σ≈ 1
δt Z
tntn−1
Z
σ
K(λ(s) ∇ p − ξ(s)g) · n
K,σdνdt (2.8) and let F
K,σnbe an approximation of the non-wetting phase flux
F
K,σn≈ 1 δt
Z
tntn−1
Z
σ
(qf (s) + γ (s)Kg − K ∇ ϕ(s)) · n
K,σdνdt. (2.9) The numerical fluxes Q
nK,σand F
K,σnhave to be constructed as functions of the discrete unknowns. Using the notations (2.7) and (2.8) we discretize the equation (1.5) by
X
σ∈EK
Q
nK,σ= m(K)(k
nw,K+ k
no,K) for all K ∈ M . (2.10) We also prescribe the continuity of the fluxes
Q
nK,σ+ Q
nL,σ= 0 for all σ ∈ E
intwith { K, L } = M
σ. (2.11) Similarly, the equation (1.7) is discretized by
ω(K) s
nK− s
n−1Kδt + X
σ∈EK
F
K,σn= m(K)k
o,Knfor all K ∈ M , (2.12) and
F
K,σn+ F
L,σn= 0 for all σ ∈ E
intwith { K, L } = M
σ. (2.13) The discrete equations (2.16)-(2.13) have to be prescribed at each time step n ∈ { 1, . . . , N } . We prescribe the initial and the boundary conditions for the numerical scheme by setting
s
0K= 1 m(K)
Z
K
s
0(x) dx for all K ∈ M (2.14) and
s
nσ= p
nσ= 0 for all σ ∈ E
ext. (2.15)
Remark that opposite to the classical two-point flux approximation, the discrete fluxes
Q
nK,σand F
K,σn(which still remain to be constructed) are not a priori continuous across
the element’s interfaces, so that the continuity is prescribed in the scheme by (2.11) and
(2.13).
2.2.2 The discrete weak formulation
Following the ideas of [22] we write the scheme in the variational form.
For each n ∈ { 1, . . . , N } find s
n∈ X
D,0and p
n∈ X
D,0such that for all v
n, w
n∈ X
D,0: X
K∈M
X
σ∈EK
(v
Kn− v
σn)Q
nK,σ= X
K∈M
m(K)v
Kn(k
nw,K+ k
no,K), (2.16) X
K∈M
ω(K )w
Kns
nK− s
n−1Kδt + X
K∈M
X
σ∈EK
(w
nK− w
nσ)F
K,σn= X
K∈M
m(K )w
Knk
no,K, (2.17) s
0K= 1
m(K) Z
K
s
0(x) dx. (2.18) In order to complete the scheme we have to define the numerical fluxes Q
nK,σand F
K,σn. Let K
Kdenote the mean value of K(x) over a cell K,
K
K= 1 m(K)
Z
K
K(x) dx (2.19)
and let
g
K,σ= m(σ)K
Kg · n
K,σ. (2.20) Note that g
K,σsatisfies
X
σ∈EK
g
K,σ= 0 for all K ∈ M , (2.21)
but not necessarily
g
K,σ+ g
L,σ= 0 with { K, L } = M
σ.
The above equality remains true for the interfaces which are ”interior” with respect to some sub-domain Ω
i, that is to say
g
K,σ+ g
L,σ= 0 with { K, L } = M
σfor all σ 6∈ Γ
i,j, (2.22) for any i, j. Next, we define Q
nK,σand F
K,σnby
Q
nK,σ= λ
nKF
K,σ(p
n) + G (ξ( · )g
K,σ; s
nK, s
nσ), (2.23) F
K,σn= G (Q
nK,σf( · ) + γ( · )g
K,σ; s
nK, s
nσ) + F
K,σ(ϕ(s
n)), (2.24) where λ
nK= λ
nKfor all K ∈ M and n ∈ { 1, . . . , N } , in general. The terms F
K,σ( · ) correspond to the diffusive fluxes which are discretized using the SUSHI scheme (Section 2.2.4). The terms G ( · ) stand for the discretization of the convective fluxes, using the Godunov scheme (Section 2.2.3 below, see also [28, p. 3]).
2.2.3 The Godunov scheme and the convection term
Let a, b ∈ R and f ∈ L
∞( R ) we define the Godunov flux by G (f; a, b) =
min
s∈[a,b]f (s) if a ≤ b,
max
s∈[b,a]f(s) if b ≤ a. (2.25)
Since it is often useful to write the discrete flux in more explicit form we define S (f ; a, b) =
argmin
s∈[a,b]f (s) if a ≤ b,
argmax
s∈[b,a]f(s) if b ≤ a,
and we introduce the following notations
ξ
K,σn= ξ( S (ξ( · )g
K,σ; s
nK, s
nσ)),
f
K,σn= f( S (Q
nK,σf( · ) + γ( · )g
K,σ; s
nK, s
nσ)), γ
K,σn= γ( S (Q
nK,σf ( · ) + γ( · )g
K,σ; s
nK, s
nσ)).
(2.26)
Using the notations (2.26) we can write the discrete fluxes in the form
Q
nK,σ= λ
nKF
K,σ(p
n) + ξ
K,σng
K,σ(2.27) and
F
K,σn= Q
nK,σf
K,σn+ γ
K,σng
K,σ+ F
K,σ(ϕ(s
n)). (2.28) 2.2.4 The discrete gradient and the diffusion term
In this section we recall a construction of the discrete gradient and of the numerical flux F
K,σ( · ) proposed in [22]. Let u ∈ X
D, for all K ∈ M and σ ∈ E
Kwe define
∇
K,σu = ∇
Ku + R
K,σu · n
K,σ, (2.29) where
∇
Ku = 1 m(K)
X
σ∈EK
m(σ)(u
σ− u
K)n
K,σ(2.30) and
R
K,σu =
√ d d
K,σ(u
σ− u
K− ∇
Ku · (x
σ− x
K)). (2.31) Note that the stabilizing term R
K,σis a second order error term, which vanishes for piecewise linear functions. We define the discrete gradient ∇
Du as the piecewise constant function equal to ∇
K,σu in the cone D
K,σwith vertex x
Kand basis σ
∇
Du(x) |
x∈DK,σ= ∇
K,σu.
Let u = (u
n)
n∈{1,...,N}∈ X
D,δt, taking into account the time discretization, we define the discrete gradient ∇
D,δtu(x, t) by
∇
D,δtu(x, t) |
t∈(tn−1,tn]= ∇
Du
n(x), (2.32) for all x ∈ Ω and all n ∈ { 1, . . . , N } . For an arbitrary u ∈ X
Dthe numerical flux F
K,σ(u) can be defined through the following discrete integration by parts formula
X
σ∈EK
(v
K− v
σ) F
K,σ(u) = X
σ∈EK
m(D
K,σ)K
K∇
K,σu · ∇
K,σv for all v ∈ X
D, (2.33) which in particular implies that
X
K∈M
X
σ∈EK
(v
K− v
σ) F
K,σ(u) = Z
Ω
K ∇
Du · ∇
Dv dx for all v ∈ X
D. (2.34)
The explicit form of F
K,σcan be obtained by setting v
K− v
σ= 1 and v
K− v
σ′= 0 for
all σ
′6 = σ. We refer to [22] for more details on construction of F
K,σand its practical
implementation. Next we state without proof three results from [22].
Lemma 2.1 (Strong consistency) Let D be a discretization of Ω in sense of Definition 2.1, moreover let θ ≥ θ
Dbe given. Then for all ψ ∈ C
2(Ω), there exist a positive constant C only depending on d, θ and ϕ such that
|∇
KP
Dψ − ∇ ψ(x) | ≤ Ch for all x ∈ K (2.35) and also
k∇
DP
Dψ − ∇ ψ k
(L∞(Ω))d≤ Ch. (2.36) The following lemma, which is the slightly modified version of [22, Lemma 4.1] shows the equivalence between the semi-norm in X
Dand the L
2-norm of the discrete gradient.
Lemma 2.2 Let D be a discretization of Ω and let θ > θ
Dbe given. Then there exists m > 0 and M > 0 only depending on θ and d such that
m | v |
XD,K≤ k∇
Dv k
L2(K)≤ M | v |
XD,Kfor all K ∈ M for all v ∈ X
D.
Lemma 2.3 (Discrete Poincar´ e inequality) There exists a positive constant C, in- dependent of the mesh size h
Dsuch that
k Π
D,δtu k
L2(Ω)≤ C | u |
XDfor all u ∈ X
D. (2.37)
Proof: The result follows from Lemma 5.3 of [22].
The direct consequence of (2.33) and Lemma 2.2 is the lemma below.
Lemma 2.4 Let u, v be an arbitrary couple from X
D, then there exist two positive con- stants C
1, C
2independent of the mesh size such that
X
σ∈EK
(v
K− v
σ) F
K,σ(u)
≤ C
1| u |
XD,K| v |
XD,K(2.38)
and X
σ∈EK
(u
K− u
σ) F
K,σ(u) ≥ C
2| u |
2XD,K(2.39) Lemma 2.4 implies the following useful technical result.
Lemma 2.5 Let D = ( M , E , P ) be a discretization of Ω, let q ∈ Q
Dand u ∈ X
D. Then
X
K∈M
X
σ∈EK
q
K,σF
K,σ(u)
≤ C
1| u |
XDX
K∈M
X
σ∈EK
m(σ) d
K,σq
K,σ2!
12. (2.40)
Proof: Let K be an element of M , setting v
K= 0 and v
σ= f
K,σin (2.38) we obtain
X
σ∈EK
q
K,σF
K,σ(u)
≤ | u |
XD,KX
σ∈EK
m(σ) d
K,σq
K,σ2!
12. Proceeding the same way for all K ∈ M we get
X
K∈M
X
σ∈EK
q
K,σF
K,σ(u)
≤ X
K∈M
X
σ∈EK
q
K,σF
K,σ(u)
≤ C
1X
K∈M
| u |
XD,KX
σ∈EK
m(σ) d
K,σq
K,σ2!
12.
We use the Cauchy-Schwarz inequality to complete the proof.
Lemma 2.6 Let D = ( M , E , P ) be a discretization of Ω, let v ∈ X
Dthen there exists a positive constant C independent of the mesh size such that for all n ∈ { 1 . . . N }
X
K∈M
X
σ∈EK
(v
K− v
σ) G (Q
nK,σf ( · ) + γ( · )g
K,σ; s
nK, s
nσ)
≤ C | v |
XD( | p |
XD+ 1).
Proof: In view of (2.24), (2.27) and (2.28) we have that X
K∈M
X
σ∈EK
(v
K− v
σ) G (Q
nK,σf ( · ) + γ( · )g
K,σ; s
nK, s
nσ) = T
1+ T
2+ T
3, where
T
1= X
K∈M
X
σ∈EK
(v
K− v
σ)γ
K,σng
K,σT
2= X
K∈M
X
σ∈EK
(v
K− v
σ)ξ
K,σnf
K,σng
K,σT
3= X
K∈M
X
σ∈EK
(v
K− v
σ)λ
nKF
K,σ(p
n)f
K,σnIn view of (2.20)
T
1= X
K∈M
X
σ∈EK
m(σ)(v
K− v
σ)γ
K,σnK
Kg · n
K,σRemark that for all K ∈ M and σ ∈ E
Kone has m(σ)d
K,σ= m(D
K,σ)d, where d is the space dimension. Using Cauchy-Schwarz inequality
(T
1)
2≤ X
K∈M
X
σ∈EK
m(σ) d
K,σ(v
K− v
σ)
2X
K∈M
X
σ∈EK
m(σ)d
K,σ(γ
K,σn)
2| g |
2K
2≤ K
2| g |
2m(Ω)d k γ k
2L∞((0,1))| v |
2XD.
(2.41)
In the same way
(T
2)
2≤ K
2| g |
2m(Ω)d k ξ k
2L∞((0,1))k f k
2L∞((0,1))| v |
2XD. (2.42) Applying Lemma 2.5 to the term T
3we obtain
(T
3)
2≤ | p |
2XDX
K∈M
X
σ∈EK
m(σ) d
K,σλ
nK(v
K− v
σ)f
K,σn 2≤ k λ k
2L∞((0,1))k f k
2L∞((0,1))| p |
2XD| v |
2XD.
(2.43)
Gathering (2.41)-(2.43) we complete the proof.
Finally we present the technical lemma which is used in the proof of the a priori estimates.
Lemma 2.7 Let ϕ(s) satisfying the hypothesis ( H
1) and let the function Φ be defined by Φ(s) =
Z
s 0ϕ(τ )dτ. (2.44)
Then,
1 2L
ϕ(ϕ(s))
2≤ Φ(s) ≤ L
ϕ2 s
2. (2.45)
Proof: The function ϕ is invertible, then setting ξ = ϕ(τ) in (2.44) gives Φ(s) =
Z
ϕ(s) 0ξdξ
ϕ
′(ϕ
−1(ξ)) ≥ 1 L
ϕZ
ϕ(s) 0ξdξ = 1 2L
ϕ(ϕ(s))
2. On the other hand
Φ(s) ≤ L
ϕZ
s 0τ dτ = L
ϕ2 s
2.
3 A priory estimates and existence of discrete solu- tion
3.1 A priori estimates
Definition 3.1 (Approximate solution) Let D be a discretization of Ω, N ∈ N
∗and δt = T /N > 0. We say that the sequence (s
D,δt, p
D,δt) = (s
n, p
n)
n∈{1,...,N}∈ (X
D,δt,0) is an approximate solution of the problem (1.5)-(1.9) if for all n ∈ { 1, . . . , N } , (s
n, p
n) satisfies (2.16)-(2.18). We also denote by s
D,δtand p
D,δtthe function pair defined by
s
D,δt(x, 0) = s
0Kfor all x ∈ K, (3.1)
s
D,δt(x, t) = s
nKfor all (x, t) ∈ K × (t
n−1, t
n], (3.2) s
D,δt(x, t) = s
nσfor all (x, t) ∈ σ × (t
n−1, t
n] (3.3) and
p
D,δt(x, t) = p
nKfor all (x, t) ∈ K × (t
n−1, t
n], (3.4) p
D,δt(x, t) = p
nσfor all (x, t) ∈ σ × (t
n−1, t
n]. (3.5) Remark 3.1 In general the capillary diffusion ϕ(s
D,δt) is more regular that s
D,δtand it is easer to work with the function ϕ
D,δt, thus the element ϕ
D,δt= ϕ(s
D,δt) of X
D,δt,0can be considered as a set of primary discret unknowns. We also define the function ϕ
D,δt(x, t)
ϕ
D,δt(x, t) = ϕ(s
D,δt(x, t)) for all (x, t) ∈ Q
T. (3.6) Theorem 3.1 (A priori estimate) Let (s
D,δt, p
D,δt) be a solution of the discrete prob- lem (2.16)-(2.18). Then there exists a positive constant C independent of h
Dand δt such that
k∇
D,δtp
D,δtk
L∞(0,T;L2(Ω))+ k ϕ
D,δtk
L∞(0,T;L2(Ω))+ k∇
D,δtϕ
D,δtk
L2(QT)≤ C. (3.7) Proof: Pressure equation. In order to obtain the estimate on the first term in (3.7) we use p
nas a test element in the pressure equation (2.16), which implies
X
K∈M
X
σ∈EK
(p
nK− p
nσ)(λ
nKF
K,σ(p
n) + ξ
K,σng
K,σ) = X
K∈M
m(K)p
nK(k
w,Kn+ k
no,K). (3.8) Let us first estimate the term T
ξ= P
K∈M
P
σ∈EK
(p
nK− p
nσ)ξ
nK,σg
K,σ. In view of (2.20) we deduce from Cauchy-Schwarz inequality that
(T
ξ)
2= ( X
K∈M
X
σ∈EK
m(σ)(p
nK− p
nσ)n
K,σ· ξ
K,σnK
Kg)
2≤ X
K∈M
X
σ∈EK
m(σ) d
K,σ(p
nK− p
nσ)
2X
K∈M
X
σ∈EK
m(σ)d
K,σ( | K
Kg | ξ
K,σn)
2.
Then, in view of the assumptions ( H
2), ( H
3) and ( H
4b), and also thanks to (2.4) and Lemma 2.2 we deduce that
| T
ξ| ≤ dm(Ω)K | g |k ξ k
L∞((0,1))k∇
D,δtp( · , t
n) k
L2(Ω). (3.9) In view of (2.7) and (3.4) the right-hand side of (3.8) can be written as
X
K∈M
m(K)p
nK(k
nw,K+ k
no,K) = X
K∈M
1 δt
Z
tntn−1
Z
Ω
p
D,δt(x, t)(k
w(x, t) + k
o(x, t)) dxdt Applying Cauchy-Schwarz inequality and the discrete Poincar´e inequality (Lemma 2.3) we obtain
X
K∈M
m(K)p
nK(k
nw,K+ k
no,K) ≤ k p
D,δt( · , t
n) k
L2(Ω)k k
w+ k
ok
L∞(0,T;L2(Ω))≤ C k∇
D,δtp
D,δt( · , t
n) k
L2(Ω)k k
w+ k
ok
L∞(0,T;L2(Ω)). so that in view of ( H
6),
X
K∈M
m(K)p
nK(k
w,Kn+ k
o,Kn) ≤ C k∇
D,δtp
D,δt( · , t
n) k
L2(Ω). (3.10) Gathering (2.39), (3.8), (3.9), (3.10) we obtain
λ k∇
D,δtp
D,δt( · , t
n) k
2L2(Ω)≤ X
K∈M
X
σ∈EK
(p
nK− p
nσ)λ
nKF
K,σ(p
n) ≤ C k∇
D,δtp
D,δt( · , t
n) k
L2(Ω), with some positive C; the proof of the estimate on the first term of (3.7) is complete.
The estimate on the other terms of (3.7) can be obtained by setting w
n= ϕ(s
n) in the saturation equation (2.17) and summing over n ∈ { 1, . . . , m } , which yields
m
X
n=1
X
K∈M
ω(K)ϕ
nK(s
nK− s
n−1K) + δt
m
X
n=1
X
K∈M
X
σ∈EK
(ϕ
nK− ϕ
nσ)F
K,σ(ϕ
n)
=
m
X
n=1
X
K∈M
δtm(K)ϕ
nKk
no,K. We define the terms
T
t=
m
X
n=1
X
K∈M
ω(K )ϕ
nK(s
nK− s
n−1K),
T
C=
m
X
n=1
X
K∈M
X
σ∈EK
δt(ϕ
nK− ϕ
nσ) G (Q
nK,σf ( · ) + γ( · )g
K,σ; s
nK, s
nσ).
and
T
D=
m
X
n=1
X
K∈M
X
σ∈EK
δt(ϕ
nK− ϕ
nσ) F
K,σ(ϕ(s
n)).
Accumulation term. Remark that using the notations of Lemma 2.7 Φ(s
nK) − Φ(s
n−1K) = ϕ
nK(s
nK− s
n−1K) +
Z
snK snK−1(ϕ(τ) − ϕ
nK)dτ. (3.11)
Since the function ϕ is increasing, the second term in the right-hand side of (3.11) is negative, which implies that
T
t≥
m
X
n=1
X
K∈M
ω(K )(Φ(s
nK) − Φ(s
n−1K)) = X
K∈M
ω(K)Φ(s
mK) − X
K∈M
ω(K)Φ(s
0K)).
It follows from (2.45) and the assumption ( H
5) that T
t≥ 1
2L
ϕX
K∈M
ω(K)(ϕ
mK)
2− L
ϕ2 X
K∈M
ω(K)(s
0K)
2≥ ω
2L
ϕk ϕ
D,δt( · , t
m) k
2L2(Ω)− ωL
ϕ2 k s
D,δt( · , 0) k
2L2(Ω). Convection term. In this subsection we use the simplified notation
G
K,σn(a, b) = G Q
nK,σf ( · ) + γ( · )g
K,σ; a, b
for all a, b ∈ R . (3.12) For all K ∈ M , σ ∈ E
Kand s ∈ R we define the function
G
nK,σ(s) = Z
s0
G
K,σn(τ, τ )ϕ
′(τ)dτ, (3.13) which is such that
G
nK,σ(s
nK) − G
nK,σ(s
nσ) = (ϕ
K− ϕ
σ) G
K,σn(s
K, s
σ) +
Z
sKsσ
G
K,σn(τ, τ ) − G
K,σn(s
K, s
σ)
ϕ
′(τ)dτ. (3.14) In view of (2.25) the second term on the right-hand side of (3.14) is negative, which implies the following estimate
T
C≥
m
X
n=1
X
K∈M
X
σ∈EK
δt G
nK,σ(s
nK) − G
nK,σ(s
nσ) .
In view of (3.12), (3.13) and (2.25) we have that G
nK,σ(s
nK) − G
nK,σ(s
nσ) = Q
nK,σZ
snK snσf(τ )ϕ
′(τ )dτ + g
K,σZ
snK snσγ (τ )ϕ
′(τ )dτ.
It follows from (2.16) and the hypotheses ( H
1), ( H
2c) that X
K∈M
X
σ∈EK
Q
nK,σZ
snKsnσ
f(τ )ϕ
′(τ )dτ = X
K∈M
m(K )(k
w,Kn+ k
o,Kn) Z
snK0
f (τ)ϕ
′(τ)dτ ≥ 0, where we have set
v
Kn= Z
snK0
f (τ)ϕ
′(τ)dτ and v
nσ= Z
snσ0
f(τ )ϕ
′(τ)dτ.
Therefore we have that T
C≥
N
X
n=1
δt X
K∈M
X
σ∈EK
g
K,σZ
snKsnσ
γ(τ )ϕ
′(τ )dτ
≥
N
X
n=1
δt X
K∈M
Z
snK 0γ(τ)ϕ
′(τ)dτ X
σ∈EK
g
K,σ−
N
X
n=1
δt X
K∈M
X
σ∈EK