Mathematical Theory of Nonlinear Single-Phase Poroelasticity

Journal of Nonlinear Science (2023) 33:44
https://doi.org/10.1007/s00332-023-09896-z
Mathematical Theory of Nonlinear Single-Phase

Poroelasticity
C. J. van Duijn1,2 · Andro Mikelić3
Received: 28 October 2020 / Accepted: 11 February 2023 / Published online: 10 March 2023
© The Author(s) 2023
Abstract
In this paper, we study the equations of nonlinear poroelasticity derived from mixture
theory. They describe the quasi-static mechanical behavior of a fluid saturated porous
medium. The nonlinearity arises from the compressibility of the fluid and from the
dependence of porosity and permeability on the divergence of the displacement. We
point some limitations of the model. In our approach, we discretize the quasi-static
formulation in time and first consider the corresponding incremental problem. For this,
we prove existence of a solution using Brézis’ theory of pseudo-monotone operators.
Generalizing Biot’s free energy to the nonlinear setting, we construct a Lyapunov
functional, yielding global stability. This allows us to construct bounds that are uniform
with respect to the time step. In the case when dissipative interface effects between
the fluid and the solid are taken into account, we consider the continuous time case in
the limit when the time step tends to zero. This yields existence of a weak free energy
solution.
Communicated by Alain Goriely.
Andro Mikelić passed away on 28 November 2020.
A.M. was partially supported by Darcy Center Eindhoven-Utrecht, the Netherlands, by the project
UPGEO ANR-19-CU05-032 of the French National Research Agency (ANR) and by the LABEX
MILYON (ANR-10-LABX-0070) of Université de Lyon, within the program “Investissements d’Avenir”
(ANR-11-IDEX-0007) operated by the French National Research Agency (ANR).
B C. J. van Duijn
C.J.v.Duijn@TUE.nl
Andro Mikelić
mikelic@univ-lyon1.fr
1 Department of Mechanical Engineering, Darcy Center, Eindhoven University of Technology,

Eindhoven, The Netherlands
2 Department of Earth Sciences, Utrecht University, Utrecht, The Netherlands
3 Univ Lyon, Université Claude Bernard Lyon 1, CNRS UMR 5208, Institut Camille Jordan, 43 blvd.
du 11 novembre 1918, F-69622 Villeurbanne cedex, France
123
44 Page 2 of 41 Journal of Nonlinear Science (2023) 33:44
Keywords Quasi-static nonlinear poroelasticity · Free energy · Incremental

problem · Pseudo-monotonicity · Continuous time limit
1 Introduction
The elastic quasi-static deformation of a fluid saturated porous medium received much
attention in the civil engineering literature because of its relevance to many problems
of practical interest. In the framework of consolidation in soil mechanics, these prob-
lems relate to the physical loading of soil layers or the effect of soil subsidence due
to groundwater withdrawal for drinking water supply or industrial and agricultural
purposes. Examples and underlying theories are given in the well-known works of
Coussy (2004), Lewis and Schrefler (1998) and Verruijt (2015). They build on the
classical theory of Terzaghi (1951) and the pioneering approach of Biot (1962) and
Tolstoy (1992).
Recently, other examples of elastic deformation of porous media arise in the context
of industrial and biomedical applications, such as paper printing (Bosco et al. 2015),
bone regeneration (Cowin 1999; Cardoso et al. 2013), blood flow (Prosi et al. 2005;
Čanić et al. 2006) and car filters (Marciniak-Czochra and Mikelić 2015; Mikelić and
Tambača 2016).
In its simplest form, assuming both the fluid and the porous material (grains) to
be incompressible and assuming the porous medium to be homogeneous and linearly
elastic with small strains, the mathematical formulation reads (see Bear and Bachmat
1990; Verruijt 2015 or van Duijn et al. 2019):

K
div ∂t u + div (ρ f g − ∇ p) = q (1)
ηf
and
− div σ = F, (2)
where
σ = Ge(u) − α pI, (3)
with
G E = 2μE + λTr(E)I, for symmetric matrices E. (4)
In these equations, u [m] denotes skeleton displacement, K [m2 ] intrinsic permeability

(a symmetric positive-definite rank−2 tensor), η f [Pa s] fluid viscosity, p [Pa] fluid
pressure and q [ 1/s ] sources/sinks. Further, σ [Pa] is the total stress, F a given body
force (generally linked to gravitational effects), G the symmetric, positive-definite,
rank-4 Gassmann tensor, e(u) the linearized strain tensor and α ∈ (0, 1] Biot’s effective
123
Journal of Nonlinear Science (2023) 33:44 Page 3 of 41 44
stress parameter. Finally, μ [Pa] and λ [Pa] are Lamé’s parameters. Using for G the
specific form (4), i.e., Hooke’s law, assumes that the skeleton is mechanically isotropic.
The linear quasi-static Biot system, as well as its dynamical analog, was also derived
by means of a multiscale approach, where the starting point is the linear fluid–structure
interaction at the pore level. We refer to the monographs Sanchez-Palencia (1980) and
Mei and Vernescu (2010) for derivations using two-scale expansions and to Mikelić and
Wheeler (2012) for a rigorous mathematical derivation by means of homogenization.
The derivations using multiscale analysis confirm Biot’s models in the linear setting.
Hence, from different points of view system (1)–(4) is well accepted.
In the engineering literature, one writes α = 1 − K /K g , where K is the drained
bulk modulus of the porous skeleton and K g the bulk modulus of the grains. Since it
is assumed that K g = +∞, we will set α = 1 in (3).
From a mathematical perspective, Eqs. (1)–(4) received much attention. Here, we
mention the pioneering paper by Auriault and Sanchez-Palencia (1977) and the work
of Ženíšek (1984), who were the first to demonstrate existence and uniqueness. More
recent studies include Showalter (2000), Owczarek (2010) and Marciniak-Czochra
and Mikelić (2015). Later, Cao et al. (2013) considered a nonlinear extension of (1),
by replacing the permeability tensor K by the product Kk(div u). The function k(·) is
a relative permeability depending on the volumetric strain div u. From (1), we notice
that the overall mixture of two incompressible phases is not incompressible itself.
Though system (1)–(4) is linear, its mathematical complexity lies in the fact that it
is of quasi-static nature. In particular (2)–(4) allow to control the size of the volumetric
strain only through the size of the data. Some authors circumvent this by introducing
a time dependence in (2)–(4) as well. For instance, Bociu et al. (2016) replace u in (3)
by u + δ∂t u, where δ ≥ 0 is a visco-elastic parameter. Their study allows δ = 0, hence
it includes the true quasi-static case as well. A different regularization was proposed
by Murad and Cushman (1996) who replaced (3) and (4) by
σ = 2μe(u) + (λdiv u + λ∗ div ∂t u − α p)I, (5)
with λ∗ > 0. This form arises in the non-equilibrium theory, where the fluid pressure
and the solid pressure differ by λ∗ div ∂t u.
In this paper, we propose to study the quasi-static formulation in which we replace
Eq. (1) by the nonlinear fluid phase mass balance based on the mixture theory of
Bedford and Drumheller (1978) and Bedford and Drumheller (1983), see, e.g., Rutqvist
et al. (2001) and Lewis and Schrefler (1998):
n∂t ρ + ρdiv ∂t u + div j = Q, (6)
where j denotes the Darcy mass flux
Kk(n)ρ
j= (ρg − ∇ p). (7)
ηf
Here, n denotes porosity, ρ = ρ f [kg/m3 ] fluid density, k relative permeability and Q

[kg/m3 s] sources/sinks.
123
In Eqs. (6)–(7), the porosity n is a given function of the volumetric strain: i.e.,
n = n(div u). (8)
An explicit expression for (8) is derived from the Lagrangian solid mass balance
equation. This is shown in Sect. 2. Through (8), the relative permeability depends on
div u.
Since n is the volume fraction of voids in the porous medium, it should satisfy the
natural bounds
0 < n < 1. (9)
However, in Sect. 2 we show by means of a counter example that the porosity can
attain negative—and thus physically unrealistic—values. Therefore, the bounds in (9)
are a major concern in the mathematical model.
To close system (2)–(4), (6)–(7), we introduce a constitutive relation for the fluid
density in terms of the pressure. Assuming weak compressibility, we write
ρ = ρ( p) = ρ0 (1 + β( p − p0 )). (10)
Further, we propose an explicit expression for the relative permeability in terms of the
porosity
k = k(n). (11)
In (10), ρ0 and p0 are reference values for, respectively, density and pressure and
β [Pa−1 ] is the fluid compressibility coefficient. The relative permeability in (11)
satisfies

k ∈ C 1 [0, 1],
(12)
k(0) > 0 and k > 0 in ([0, 1).
A well-known example is the Kozeny–Carman formula, see for instance Bear and
Bachmat (1990),
n3
k(n) = k0 (k0 > 0), (13)
(1 − n)2
in a realistic porosity interval, bounded away from n = 0 and n = 1. Thus, taking k

such that (12) holds and
⎧
⎪
⎪ k0 n 3∗
⎪
⎪ ≥ , for n ≤ n ∗ ,
⎪
⎨ 2 (13 − n ∗ )
2
⎪
n
k(n) = k0 , for 0 < n ∗ < n < n ∗ < 1, (14)
⎪
⎪ (1 − n)2
⎪
⎪
⎪
⎪ (n ∗ )3
⎩ ≤ 2k0 , for n ≥ n ∗ ,
(1 − n ∗ )2
123
for appropriately chosen 0 < n ∗ < n ∗ < 1, gives a relative permeability satisfying
(13) in the interval [n ∗ , n ∗ ].
We notice that Eq. (6), coupled with (2) and (4), is nonlinear due to the relation
k = k(n) and the products involving time derivatives. Assuming constant fluid phase
density in the poroelastic mixture is therefore an important simplification. This is
studied in Cao et al. (2013) and Bociu et al. (2016).
In studying system (2)–(4), (6)–(11), a crucial role is played by its free energy.
The idea is to generalize Biot’s free energy (Biot 1962), which is quadratic in strain
and fluid density, to the nonlinear poroelastic setting. This free energy serves as a
Lyapunov functional. This approach is linked to general entropy methods for PDEs.
For a detailed survey covering various fields of applications, we refer to Evans (2004)
and to the recent book by Jüngel (2016). An interesting application of the entropy
method is discussed in Mikelić (2010), Cao and Pop (2016) and Milišić (2018), where
the authors consider dynamic capillary pressure effects in two-phase porous media
flow.
This paper is organized as follows. In Sect. 2, we present details of the model
formulation. The starting point is the mass balance for the fluid and the solid phase.
The latter implies an explicit expression for (8). Introducing a lower bound for the
porosity, we modify the fluid mass balance so that a Lyapunov functional can be
constructed for the modified system. This modification is such that the fluid equation
reduces to its original form in the physical range of the fluid density ρ and solid
volumetric strain E. Section 2 is concluded by a weak formulation of the modified
system.
In Sect. 3, we consider, for the relaxation parameter λ∗ ≥ 0, the incremental version
of the modified system. Using Brézis’ theory of pseudo-monotone operators, existence
is demonstrated. Applying the Lyapunov functional yields global (in time) estimates.
Next, in Sect. 4, we use these estimates to solve the time-continuous problem when
λ∗ > 0. In both Sects. 3 and 4, we borrow ideas from Roubiček (2005). Finally, in
Sect. 5 we present a discussion and conclusions.
2 Problem Formulation
In a number of steps, we construct in this section the equations that serve as starting
point for the analysis. The general setting of the problem is as follows:
Let Ω ⊂ Rm (m=2,3) denote a smooth bounded domain, occupied by a linear
elastic skeleton. The skeleton material (grains) is assumed incompressible: i.e., the
bulk modulus of the grains is infinitely large. The voids in the porous structure are
completely filled with a slightly compressible fluid, in the sense that the fluid pressure
p and density ρ are related by (10).
2.1 Balance Equations
For given ξ ∈ Ω, let x(ξ, t) denote the location of a solid particle at time t > 0, that
started at x(ξ, 0) = ξ,. Then the skeleton velocity vs is given by vs = ∂t x|ξ .
123
Restricting themselves to small displacements u (within the elastic regime),

Rutqvist et al. (2001) and Lewis and Schrefler (1998) argue that in the mass bal-
D
ance equation for the fluid and solid, the material derivative = ∂t + vs · ∇ can be
Dt
replaced by the partial derivative ∂t . This is made explicit by a scaling argument in
van Duijn et al. (2019). The resulting Lagrangian form of the mass balances reads:
n∂t ρ + ρdiv vs + div j = Q (fluid phase) (15)
and
∂t (1 − n) + (1 − n)div vs = 0 (solid phase), (16)
where j is mass flux (7).

Within the same approximation, one may write
div vs = ∂t div u.
Using this in (15) and (16) gives
n∂t ρ + ρ∂t div u + div j = Q (17)
and
∂t (1 − n) + (1 − n)div ∂t u = 0. (18)
Integrating (18) in time from t = 0, say, to t > 0, we have
1 − n = (1 − n 0 )e−div (u − U0 ) for t > 0.. (19)
Here, U0 is the initial displacement and n 0 the initial porosity. With n 0 ∈ (0, 1) in Ω,
expression (19) ensures
n < 1 in Ω for all t > 0. (20)
To avoid technical complications, we restrict ourselves to n 0 = constant in Ω.

For small displacements u − U0 , expression (19) is approximated by
n = n 0 + (1 − n 0 )div (u − U0 ). (21)
Remark 1 Frequently, the linear form (21) is used for values of div u in a neighborhood
of div U0 : i.e., in practical circumstances (21) is applied when E∗ < div (u−U0 ) < E ∗ ,
where E∗ < 0 < E ∗ are appropriately chosen.
123
Throughout the paper, we redefine
u := u − U0 , (22)
where U0 ∈ H01 (Ω)m ∩ H 2 (Ω)m is the initial displacement. Redefining accordingly
F := F + div (Ge(U0 )), (23)
we obtain for the fluid pressure p and the skeleton displacement u the system:
Kk(n)ρ
n∂t ρ + ρ div ∂t u + div (ρg − ∇ p) = Q, (24)
ηf
− div (Ge(u) − pI) = F, (25)
where
ρ = ρ( p) = ρ0 (1 + β( p − p0 )), (26)
n = n(div u) = 1 − (1 − n )e−div u
0 (27)
≈ n 0 + (1 − n 0 )div u (small strains). (28)
Remark 2 Concerning the initial displacement U0 , we note that only div U0 , the initial
volumetric strain, is used. However, when discussing the free energy, one needs in
addition that U0 is such that the corresponding elastic energy is finite. For simplicity,
we suppose U0 ∈ H 2 (Ω)m .
In the next sections, we will develop the mathematical theory for system (24)–(28).
The issue of negative porosity in (27) (or, for that matter, a porosity exceeding one
in approximation (28)), is discussed next.
2.2 Negative Porosity
We consider a simplified version of the linear problem (1)–(4) and show that div u can
attain values for which the porosity from (27)–(28) becomes negative.
For simplicity, we give the construction in R2 .
Let Ω = (0, L)2 for some L > 0. We suppose, as in the rest of this paper, that div
u|t=0 = 0. Further we set F = 0 in (25). Using (4) in (25) gives
− div 2μe(u) + (λdiv u − p)I = 0 in Ω., (29)
Proceeding as in Verruijt (2015), when he discusses the Mandel problem, we take the
divergence of (29) to obtain
Δ (2μ + λ)div u − p = 0 in Ω. (30)
123
Fig. 1 Sketch of level set of porosity n at some t > 0. The region where n > 0 shrinks with increasing time
and disappears after a finite time T p > 0
Hence the function
H = (2μ + λ)div u − p
is harmonic in Ω.
The idea is to prescribe boundary conditions for Eqs. (24) and (25) so that H |∂Ω is
given. For instance, if we set along the four edges, see Fig. 1,
⎧
⎪
⎪ {x1 = 0} : u 2 = 0, σ11 = Σ 1,0 and p = 0;
⎨
{x1 = L} : u 2 = 0, σ11 = Σ 1,L and p = 0;
(31)
⎪
⎪ {x = 0} : u 1 = 0, σ22 = Σ 2,0 and p = 0;
⎩ 2
{x2 = L} : u 1 = 0, σ22 = Σ 2,L and p = 0,
and use
∂u 1
σ11 = 2μ + λdiv u − p,
∂ x1
we have
∂u 1
Σ 1,0 = (2μ + λ) at {x1 = 0},
∂ x1
implying
H = Σ 1,0 at {x1 = 0}.
123
Repeating this along the other edges gives
H |∂Ω = Σ b ,
where Σ b denotes the given value of σ along the edges.

Then, we have
Proposition 1 Let E = divu denote the volumetric stress and let n(E) be given by (27).
Suppose there exists a constant Σ > 0 such that Σ b ≤ −Σ. Then, for Σ sufficiently
large, there exists a T p = T p (Σ) > 0 such that
n(E(x, t)) < 0 for t > T p and x ∈ Ω. (32)
Proof Note that the sign of Σ b implies compression of the medium. Restricting our-
selves to the linear case (1) in a homogeneous and isotropic porous medium in which
sources/sinks and gravity are absent, we have
K
∂t div u − Δp = 0 in Ω, t > 0. (33)
ηf
Since
Δp = (2μ + λ)Δ(div u)
and
Σb Σ
div u|∂Ω = ≤− ,
2μ + λ 2μ + λ
we have for E = div u the problem

⎧ (2μ + λ)K
⎪
⎪ ∂t E = ΔE in Ω, t > 0;
⎪
⎨ ηf
Σ (34)
⎪
⎪ E|∂Ω ≤ − for t > 0;
⎪
⎩ 2μ + λ
E|{t=0} = 0 in Ω.
By the strong maximum principle, E < E in Ω and for t > 0, where E is the solution
of problem (34) with E = −Σ/(2μ + λ) on ∂Ω. Writing E as a Fourier series, one
observes that
Σ
E(x, t) → − as t → +∞,
2μ + λ
uniformly in x ∈ Ω.
Thus, if
(1 − n 0 )eΣ/(2μ+λ) > 1,
123
or
1
Σ > (2μ + λ) ln ,
1 − n0
the result is immediate.

This example shows that there is a problem with the model. A modification is
needed to prevent the porosity (27), or (28), to become negative. Of course, one could
argue that this is outside the scope of the model or outside the range of practical
applications, since linear elasticity and small strains are supposed. However, since it
is not clear how to ensure that indeed small displacements/strains are guaranteed, one
needs to impose a porosity modification to prevent negative values.
2.3 Modification of Balance Equations
In a number of steps, we modify Eq. (24) so that it becomes well-posed in a

mathematical sense and reduces to its original form in the physical range of the
unknowns.
First, to satisfy the natural bounds (9), we replace the porosity approximation (28)
by a smooth increasing function n : R → R such that
⎧
⎪
⎪ lim n(E) = δ0 > 0,
⎨ E →−∞
n(E) = n 0 + (1 − n 0 )E, for E∗ ≤ E ≤ E ∗ ; (35)
⎪
⎪
⎩ lim n(E) = 1.
E →+∞
Here, E∗ and E ∗ are practical values chosen such that −n 0 /(1 − n 0 ) < E∗ < 0 <
E ∗ < 1 and δ0 = n(E∗ )/2, see Fig. 2for a sketch.
This construction ensures that the modified porosity n(E) remains in the physical
range (0, 1) and coincides with the linear approximation in the interval (E∗ , E ∗ ).
Realistic porosity measurements are always done away from the bounds n = 0 and
n = 1, see, e.g., Bear and Bachmat (1990).
We choose to study Eq. (24) with the fluid density as primary unknown. Hence, we
need to express the pressure p in terms of ρ. Using (26), we have explicitly
ρ − ρ0
p = p(ρ) := p0 + . (36)
βρ0
When considering (24), one clearly has in mind that ρ takes values near the reference
ρ0 . However, the mathematical nature of the equations does not guarantee this behavior.
Hence, a second modification is needed, now for ρ in the second and third term of
the left-hand side of (24). Disregarding gravity, we replace (24) by the modified fluid
mass balance equation
n(E)∂t ρ + d(ρ)∂t E − div k(E)D(ρ)K∇ρ = Q, (37)
123
Fig. 2 Sketch of porosity cutoff n(E), with n 0 = 0.5, E∗ = −0.5, E ∗ = 0.5 and δ0 = 0.125
where n(E) is given by (35) and k(E) = k(n(E)). Further, d, D : R → R are chosen
such that
⎫
d(ρ) = ρ, ⎬
ρ for |ρ − ρ0 | ≤ ρ0 − ρ∗ , (38)
D(ρ) = ,
η f βρ0 ⎭
where ρ∗ ∈ (0, ρ0 ) is a small constant. Outside this range we take for d and D
extensions that suit the mathematical analysis. We clarify this at a later point in this
section.
Remark 3 The composite function k(E) = k(n(E)) satisfies: k ∈ C 1 (R) ∩ L ∞ (R) and
k > k(δ0 ) > 0, k > 0 in R.
The balance of forces (25) is modified by adding the regularizing term λ∗ ∂t E, as in
expression (5). This gives
− div Ge(u) + (λ∗ ∂t E − p)I = F, (39)
where E = div u and where λ∗ ≥ 0.

We consider system (37), (39) in the set
Q T = {(x, t) : x ∈ Ω, 0 < t < T },
where T > 0 is arbitrarily chosen. To avoid technical complications, we take ∂Ω ∈ C 1

throughout the rest of this paper.
As initial conditions, we have
E|t=0 = 0 and ρ|t=0 = ρ 0 in Ω, (40)
123
where ρ 0 : Ω → (0, +∞) is taken near the reference value ρ0 . Along the boundary,
we prescribe
u|∂Ω = 0, ∇ρ · ν|∂Ω = 0, for 0 < t ≤ T . (41)
where ν is the outward unit normal at ∂Ω.
2.4 Lyapunov Functional
In this section, we derive an expression for the free energy which acts as a Lyapunov
functional for system (37), (39). This generalizes the free energy introduced originally
by Biot (1962).
Let {u, ρ} be a smooth solution of Eqs. (37), (39) that satisfies conditions (40) and
(41). Further, let g : R → R be a smooth, strictly increasing and globally Lipschitz
function satisfying g(ρ0 ) = 0.
We first multiply equation (39) by ∂t u and integrate the result in Ω. This gives

1 d d
Ge(u) : e(u) dx + λ∗ (∂t E)2 dx − F · u dx
2 dt dt Ω
Ω Ω
− p(ρ)∂t E dx = − ∂t F · u dx. (42)

Ω Ω
Next, we multiply (37) by g(ρ) and integrate the result in Ω. This results in

n(E)g(ρ)∂t ρ dx + d(ρ)g(ρ)∂t E dx − k(E)D(ρ)g (ρ)K∇ρ · ∇ρ dx
Ω Ω Ω

= Qg(ρ) dx. (43)
Ω
With
ρ
G(ρ) = g(z) dz, (44)
ρ0
the first term in (43) can be written as

n(E)∂t G(ρ) dx = ∂t n(E)G(ρ) dx − n (E)G(ρ)∂t E dx. (45)
Ω Ω Ω
Note that G is a nonnegative, convex function with G(ρ0 ) = 0.

We substitute (45) back into (43). Adding the resulting expression and (42) yields

d 1 ∗
Ge(u) : e(u) + n(E)G(ρ) − F · u dx + λ (∂t E)2 dx
dt Ω 2 Ω

+ k(E)D(ρ)g (ρ)K∇ρ · ∇ρ dx
Ω
123

+ d(ρ)g(ρ) − n (E)G(ρ) − p(ρ) ∂t E dx = Qg(ρ) dx − ∂t F · u dx.
Ω Ω Ω
(46)
Before considering the general nonlinear case described by this expression, we first
show its implication for the simplified linear setting where we have
1
n(E) = n 0 , d(ρ) = ρ0 , k(E) = 1 and D = .
ηfβ
Then,

d(ρ)g(ρ) − n (E)G(ρ) − p(ρ) ∂t E dx (47)
Ω
in expression (46) simplifies to

ρ0 g(ρ) − p(ρ) ∂t E dx. (48)
Ω
Since

∂t E dx = 0,
Ω
expression (48) vanishes if g(ρ) is chosen such that
ρ0 g(ρ) − p(ρ) = constant = − p0 .
This gives
ρ − ρ0
g(ρ) =
βρ02
and
(ρ − ρ0 )2
G(ρ) = .
2βρ02
Using these expressions in (46) yields

d 1 n0 ∗
Ge(u) : e(u) + (ρ − ρ0 ) − F · u dx + λ
2
(∂t E)2 dx
dt Ω 2 βρ02 Ω

1
+ K∇ρ · ∇ρ dx = Qg(ρ) dx − ∂t F · u dx. (49)
Ω η f β ρ0
2 2
Ω Ω
123
Hence,

1 n0
L(u, ρ) = Ge(u) : e(u) + (ρ − ρ0 ) − F · u dx
2
(50)
Ω 2 2βρ02
acts as a Lyapunov functional for the linear form of system (37), (39). The first term
denotes the elastic energy of the skeleton, the second term the compression energy of
the fluid, and the third term the work done by the force F.
Expression (50) coincides with Biot’s original free energy expression from Biot
(1962).
Next, we return to the nonlinear case (46). As a first step, we restrict ourselves to
the physical range of the porosity. Then, integral (47) becomes

d(ρ)g(ρ) − (1 − n 0 )G(ρ) − p(ρ) ∂t E dx. (51)
Ω
This integral vanishes if g(ρ) is chosen such that
d(ρ)g(ρ) − (1 − n 0 )G(ρ) − p(ρ) = − p0 (52)
Differentiating the expression yields a first-order equation for g. Thus for (51) to
vanish, g should satisfy the initial value problem
⎧
⎨ 1
d(ρ)g (ρ) + (d (ρ) − (1 − n 0 ))g = , for ρ ∈ R;
ρ0 β (53)
⎩ g(ρ ) = 0.
0
We first consider this problem in the interval |ρ −ρ0 | < ρ := ρ0 −ρ∗ where d(ρ) = ρ.
Then, (53) reduces to
⎧
⎨ 1
ρg + n 0 g = ,
βρ0 (54)
⎩ g(ρ ) = 0.
0
Direct integration results in

n 0
1 ρ0
g(ρ) = 1− . (55)
βn 0 ρ0 ρ
A second integration yields (Fig. 3)

ρ
1
G(ρ) = g(ξ ) dξ = (1 − n 0 )ρ − ρ0n 0 ρ 1−n 0 + n 0 ρ0 . (56)
ρ0 βn 0 (1 − n 0 )ρ0
123
Fig. 3 Sketch of the free energy βG(ρ/ρ0 ). The linear case is in blue. The nonlinear case, see (56) and
(59) with n 0 = 1/3 and ρ∗ /ρ0 = 0.01, is in black
When |ρ − ρ0 | > ρ, the function d(ρ) has not yet been defined. We do this by first
extending g(ρ) for |ρ − ρ0 | > ρ and then by solving d(ρ) from (52): i.e.,
(1 − n 0 )G(ρ) + p(ρ) − p0
d(ρ) = . (57)
g(ρ)
Clearly, (55) cannot be used for ρ ≤ 0. Instead, we extend (55) in a linear C 1 -manner
for |ρ − ρ0 | > ρ. With ρ̃ = ρ0 + ρ = 2ρ0 − ρ∗ , we set
⎧ n 0
⎪1 ρ0 ρ − ρ∗ ρ0 n 0
⎪
⎨ 1− + for ρ < ρ∗ ,
βn 0 ρ0 ρ ρ∗ ρ
g(ρ) = ∗ n 0 ∗n 0 (58)
⎪
⎪ 1 ρ0 ρ − ρ̃ ρ0
⎩ 1− + for ρ > ρ̃,
βn 0 ρ0 ρ̃ ρ̃ ρ̃
yielding
⎧ n 0
⎪ ρ − ρ∗ ρ0 ρ − ρ∗ ρ0 n 0
⎪
⎨ G(ρ∗ ) + 1− + for ρ < ρ∗ ;
βn 0 ρ0 ρ∗ 2ρ∗ ρ∗
G(ρ) = (59)
⎪
⎪ ρ − ρ̃ ρ0 n 0 ρ − ρ̃ ρ0 n 0
⎩ G(ρ̃) + 1− + for ρ > ρ̃.
βn 0 ρ0 ρ̃ 2ρ̃ ρ̃
Substituting expressions (58) and (59) in (57) yields the desired extension for d(ρ)
when |ρ − ρ0 | > ρ. Thus,

ρ for |ρ − ρ0 | ≤ ρ,
d(ρ) = (60)
(57)with g and G given by(58)and(59) for |ρ − ρ0 | > ρ.
123
Hence, the triple {g(ρ), G(ρ), d(ρ)} constructed above satisfies (52). For this choice,
the integral (51) drops from expression (46). So far, we considered for the porosity
the linear approximation n(E) = n 0 + (1 − n 0 )E. To deal with the full cutoff (35), we
introduce a second modification. The starting point is (47). This integral vanishes if
d(ρ)g(ρ) − n (E)G(ρ) = p(ρ) − p0 . (61)
Keeping g as in (55), (58) and G as in (56), (59), we now modify d(ρ), calling it
D(ρ, E), such that
n (E) p(ρ) − p0
D(ρ, E) = G(ρ) + . (62)
g(ρ) g(ρ)
Using (57) in this expression gives
G(ρ)
D(ρ, E) = d(ρ) + (n (E) − (1 − n 0 )) . (63)
g(ρ)
Clearly, for |ρ − ρ0 | < ρ and E∗ < E < E ∗ , this expression reduces to
D(ρ, E) = ρ.
Finally, we use in the Darcy mass flux term j from Eq. (37)
⎧
ρ̃, for ρ ≥ ρ̃;
1 ⎨
D(ρ) = ρ, for ρ∗ < ρ < ρ̃; (64)
η f ρ0 β ⎩ ρ , for ρ ≤ ρ .
∗ ∗
Thus, in the end we consider the “second” modified fluid mass balance equation
n(div u)∂t ρ + D(ρ, div u) div ∂t u = div k(n(div u))D(ρ)K∇ρ + Q. (65)
System (39), (65) serves as starting point of the analysis. The function D(ρ, E) in (65)
generalizes the fluid density. It is chosen so that

1
J (u, ρ) = Ge(u) : e(u) dx + n(div u)G(ρ) dx − F · u dx (66)
2 Ω Ω Ω
acts as a Lyapunov functional for the system. The function G : R → R satisfies

G(ρ0 ) = 0, G(ρ) > 0 if ρ = ρ0 and G is strictly convex, with quadratic behavior for
large values of |ρ|. It is explicitly given by (56) and (59).
123
2.5 Summary of Equations and Weak Formulation
The problem describing the nonlinear poroelastic behavior of a fluid saturated porous
medium is to find the displacement u : Q T → Rm and the fluid density ρ : Q T → R
satisfying
(i) the balance equations
n(E)∂t ρ + D(ρ, E)∂t E = div k(E)D(ρ)K∇ρ + Q, (67)
− div Ge(u) + λ∗ ∂t EI − p(ρ)I = F, (68)
in Q T = (0, T ) × Ω and
(ii) the initial-boundary conditions (40)–(41).
The coefficients in Eqs. (67)–(68) were introduced in this section. Specifically,
n(E) and k(E) satisfy (35) and Remark 3,
D(ρ, E), D(ρ) and p(ρ) are given by (62), (64) and (36),
and λ∗ ≥ 0.
We recast this classical formulation in the following weak form.
Definition 1 We call a triple (u, E, ρ) ∈ L ∞ (0, T ; H 1 (Ω)m )× L ∞ (0, T ; Hloc

1 (Ω))×
L 2 (0, T ; H 1 (Ω))∩L ∞ (0, T ; L 2 (Ω)) , ∂t E ∈ L 2 (Q T )∩L ∞ (0, T ; Hloc

1 (Ω)) a weak
free energy solution if (i)

T
− ρn(E)∂t Φ dxdt − n 0 ρ 0 (x)Φ(x, 0) dx
0 Ω Ω
T
+ ∂t E D(ρ, E) − ρn (E) Φ dxdt
0 Ω
T
+ k(E)D(ρ)K∇ρ · ∇Φ dxdt
0 Ω
T
= QΦ dxdt, ∀Φ ∈ H 1 (Q T ), Φ|t=T = 0; (69)
0 Ω
(ii)
E = div u;
(iii)

∗
Ge(u) : e(ξ) dx + λ ∂t E div ξ dx − p(ρ)div ξ dx
Ω Ω Ω

= F · ξ dx, ∀ξ ∈ H01 (Ω)3 and for almost all t ∈ (0, T ]; (70)
Ω
123
(iv)
E|t=0 = 0 in Ω. (71)
(v) For every t1 , t2 ∈ [0, T ], t1 < t2 ,

1
Ge(u(t2 )) : e(u(t2 )) + n(E(t2 ))G(ρ(t2 )) − F(t2 ) · u(t2 ) dx
Ω 2
t2
+ λ∗ (∂t E)2 + k(E)D(ρ)g (ρ)K∇ρ · ∇ρ − Qg(ρ) + ∂t F · u dxdt
t Ω
1
1
≤ Ge(u(t1 )) : e(u(t1 )) + n(E(t1 ))G(ρ(t1 )) − F(t1 ) · u(t1 ) dx, (72)
Ω 2
where g(ρ) and G(ρ) are given, respectively, by (55), (58) and (56), (59).
Here, ρ 0 ∈ L 2 (Ω), Q ∈ C([0, T ]; L 2 (Ω)) and F ∈ H 1 (0, T ; L 2 (Ω)m ).
In Definition 1, we explicitly incorporate energy inequality (72). When dealing with

classical solutions, Eqs. (67)–(68) imply the energy balance (see (46), (47) and (66))

∂t J (u, ρ) + λ∗ (∂t E)2 dx + k(E)D(ρ)g (ρ)K∇ρ · ∇ρ dx
Ω Ω

= Qg(ρ) dx − ∂t F · u dx. (73)
Ω Ω
However, in the weak formulation (69)–(70) we cannot use Φ = g(ρ) and ξ = ∂t u,

due to lack of smoothness. Therefore, (v) has to be added explicitly. Hence, we consider
only those weak solutions satisfying additionally (72). Therefore, they are called weak
free energy solutions.
In a number of steps, we prove existence of weak solutions when λ∗ > 0. We
achieve this by first considering the incremental formulation. In this approximation,
which is clearly relevant when treating the problem numerically, we obtain existence
results which hold for all λ∗ ≥ 0.
3 Existence of a Solution to the Incremental Problem
In this section, we study the time discretized form of (67), (68).

In doing so we use the function g = g(ρ), defined by (55) and (58) as the primary
unknown. This is allowed since g : R → R is smooth and strictly increasing. The
switch to g is done for mathematical convenience, because it allows us to obtain Lya-
punov functional estimates in a straightforward way. We start with some definitions.
Let
p(g) := p(ρ(g)) and D(g) := D(ρ(g))ρ (g). (74)
123
Further, since
ρ(z) z
G(ρ(z)) = g(ξ ) dξ = ζρ (ζ ) dζ, z ∈ R, (75)
ρ0 0
let
g ⎫
G(g) := ζρ (ζ ) dζ ⎪
⎪
⎪
⎬
0
and, from (62), (76)
n (E ) p(g) − p0 ⎪⎪
⎪
D(g, E ) = G(g) + . ⎭
g g
Note that the first term in D(g, E) is bounded with respect to E and grows linearly in
g for large |g|. The second (pressure) term is bounded with respect to g since
ρ(g) − ρ0
p(g) − p0 = p(ρ(g)) − p0 = .
βρ0
Using these definitions in (67) and (68), we find in terms of g
n(E)∂t ρ(g) + D(g, E)∂t E = div k(E)D(g)K∇g + Q, (77)
− div Ge(u) + λ∗ ∂t EI − p(g)I = F, (78)
in Q T .
Next, we turn to the time discretized form of Eqs. (77) and (78).
Let τ ∈ (0, 1) denote the time discretization step and N ∈ N a large integer such that
N τ = T . At each discrete time t j = jτ , with j = 0, 1, . . . , N , we set
F j (x) = F(x, jτ ), Q j (x) = Q(x, jτ ), x ∈ Ω.
Let u j−1 and g j−1 denote, respectively, the displacement and transformed density at
t j−1 for some j ∈ {1, 2, . . . , N }: i.e.,
u j−1 (x) = u(x, t j−1 ), g j−1 (x) = g(x, t j−1 ), x ∈ Ω.
Then, u and g at time t j are obtained as solutions of the incremental problem (writing
U = u j−1 , Ξ = g j−1 and V = H01 (Ω)m × H 1 (Ω)):
Problem (PD): Given (U, Ξ ) ∈ V , find (u, g) ∈ V such that

n(div U) u−U
(ρ(g) − ρ(Ξ ))ψ dx + Dτ (g, div u, div U) div ψ dx
Ω τ Ω τ

+ k(div u)D(g)K∇g · ∇ψ dx = Q j ψ dx, ∀ψ ∈ H 1 (Ω); (79)
Ω Ω
123

λ∗
Ge(u) : e(ξ) dx + div (u − U) div ξ dx − p(g)div ξ dx
Ω τ Ω Ω

= F j · ξ dx, ∀ξ ∈ H01 (Ω)m . (80)
Ω
The coefficient Dτ in Eq. (79) is given by
n( div u) − n( div U) G(g) p(g) − p0

Dτ (g, div u, div U) = + . (81)
div u − div U g g
This expression results from D(g, E) in (76), when the derivative n (E) is replaced
n( div u) − n( div U)
by the finite difference . The specific choice of (81) appears
div u − div U
convenient in the estimates concerning the time-discrete Lyapunov functional.
Using the weak topology of the space H01 (Ω)m × H 1 (Ω), serious difficulties arise
with the coefficients n, Dτ and k depending on div u. To remedy this, we introduce
a mollifier Υε , where ε is a small positive parameter (see, e.g., Roubiček 2005, page
203), and replace div u in the nonlinearities by the convolution div uΥε = u∇Υε .
Using this substitution one can treat nonlinear coefficients containing div u as lower
order terms in the equations. This allows us to use the theory of pseudo-monotone
operators.
Applying this convolution, the regularized form of Problem (PD) reads:
Problem (PD)ε : Given (U, Ξ ) ∈ V , find (uε , gε ) ∈ V such that, with Eε = uε ∇Υε ,

n(div U) n(Eε ) − n(div U) p(gε ) − p0
(ρ(gε ) − ρ(Ξ ))ψ dx + G(gε ) +
Ω τ Ω τ gε τ gε
div (uε − U)ψ dx

+ k(Eε )D(gε )K∇gε · ∇ψ dx = Q j ψ dx, ∀ψ ∈ H 1 (Ω), (82)
Ω Ω

λ∗
Ge(uε ) : e(ξ) dx + div (uε − U) div ξ dx − p(gε )div ξ dx
Ω τ Ω Ω

= F j · ξ dx, ∀ξ ∈ H01 (Ω)m . (83)
Ω
Note that the denominator τ gε in (82) originates from the time step τ in the discretiza-
tion and the term g in the denominator of (76). We have the following existence result
Proposition 2 Let ε > 0 be a small positive constant. Under the assumptions of

Definition 1, Problem (PD)ε admits at least one solution (uε , gε ) ∈ V .
Proof We start by introducing a nonlinear operator A, defined on V and with values

in its dual V . It results from adding (82) and (83). We write the resulting relation,
with (u, g) ∈ V , as
A(u, g) = b, (84)
123
where

1 λ∗
A(u, g), (ξ, ψ) := G e(u) : e(ξ) dx + 2 div (u − U) div ξ dx
τ Ω τ Ω

p(g)
− div ξ dx + k(u∇Υε )D(g)K∇g · ∇ψ dx
τ
Ω Ω
n(div U)
+ (ρ(g) − ρ(Ξ ))ψ dx
Ω τ

n(u∇Υε ) − n(div U) p(g) − p0
+ G(g) + div (u − U)ψ dx, ∀(ξ, ψ) ∈ V .
Ω τg τg
(85)
and

b, (ξ, ψ) := F · ξ dx +
j
Q j ψ dx, ∀(ξ, ψ) ∈ V . (86)
Ω Ω
The idea is to show that A is a perturbed monotone operator: i.e., A is monotone in

its principal part containing derivatives of u and g. To be precise, we show that A
is pseudo-monotone and coercive. This allows to apply Brézis’ theorem to (84) (see
Chapter 2 in monographs Lions 1969; Roubiček 2005 or Chapters 26 and 27 in Zeidler
(1990)) to conclude existence for Problem (PD)ε .
For the comfort of the reader, we recall that an operator A : V → V is pseudo-
monotone if and only if A is bounded and
∀(v, h) ∈ V ,
{ur , gr }{u, g} weakly in V ,
lim supA(ur , gr ), (ur , gr ) − (u, g) ≤ 0, ⇒ A(u, g), (u, g) − (v, h) ≤
r →+∞ lim inf A(ur , gr ), (ur , gr ) − (v, h),
r →+∞
(87)
The boundedness of A is immediate. To show (87), we follow Chapter 2 from

Roubiček (2005) or Chapter 17 from Schweizer (2018) and rewrite A in a form having a
principal part containing partial derivatives of u (in e(u) and div u) and ∇g, and a lower
order part containing u and g. Specifically, we introduce the operator B : V × V → V
by

1 λ∗
B (w, ), (u, g) , (ξ, ψ) = G e(u) : e(ξ) dx + 2 div (u − U) div ξ dx
τ Ω τ Ω

p() n(w∇Υε ) − n(div U) p() − p0
− div ξ dx + G() + div (u − U)ψ dx
Ω τ Ω τ τ

n(div U)
+ (ρ() − ρ(Ξ ))ψ dx + k(w∇Υε )D()K∇g · ∇ψ dx, ∀(ξ, ψ) ∈ V .
Ω τ Ω
(88)
We observe that B (u, g), (u, g) = A(u, g). The introduction of B is useful because
it reflects the monotonicity of the principal part of A(u, g). This is a direct consequence
123
of
B (w, ), (u1 , g1 ) − B (w, ), (u2 , g2 ) , (u1 , g1 ) − (u2 , g2 ) ≥ 0, (89)
with equality if and only if u1 = u2 and g1 = g2 . Inequality (89) is checked by a short

computation in (88).
To show (87) we consider a sequence {ur , gr } ⊂ V such that
(ur , gr )(u, g) weakly in V and lim supA(ur , gr ), (ur , gr ) − (u, g) ≤ 0.

r →+∞
(90)
As in Roubiček (2005) we set (uδ , gδ ) = (1 − δ)(u, g) + δ(v, h), where δ ∈ [0, 1]

and (v, h) ∈ V . Using the monotonicity from (89), we obtain
δA(ur , gr ), (u, g) − (v, h) ≥ −A(ur , gr ), (ur , gr ) − (u, g)

+B (ur , gr ), (uδ , gδ ) , (ur , gr ) − (u, g) + δB (ur , gr ), (uδ , gδ ) ,
(u, g) − (v, h). (91)
The sequence (ur , gr ) is bounded in V and there exists a subsequence which strongly
converges in L 5 (Ω)m and (a.e.) in Ω, to (u, g). Hence, it suffices to pass to the limit
along this subsequence. In (91), the terms containing the operator B are fixed with
respect to the gradients. Hence,
lim B (ur , gr ), (uδ , gδ ) , (ur , gr ) − (u, g) = 0, (92)

r →+∞
and
lim B (ur , gr ), (v, h) , (ξ, ψ) = B (u, g), (v, h) , (ξ, ψ), (93)
r →+∞
for any (ξ, ψ) ∈ V . With these results, we are in a position to pass to the limit
r → +∞ in inequality (91). It yields
δ lim inf A(ur , gr ), (u, g) − (v, h) ≥ − lim supA(ur , gr ), (ur , gr ) − (u, g)
r →+∞ r →+∞
+δB (u, g), (uδ , gδ ) , (u, g) − (v, h). (94)
By the pseudo-monotonicity hypothesis (90), inequality (94) implies
lim inf A(ur , gr ), (u, g) − (v, h) ≥ B (u, g), (u, g) , (u, g) − (v, h)
r →+∞
= A(u, g), (u, g) − (v, h), ∀(v, h) ∈ V . (95)
We use this inequality to conclude
lim inf A(ur , gr ), (ur , gr ) − (v, h)

r →+∞
123
≥ lim inf A(ur , gr ), (ur , gr ) − (u, g)

r →+∞
+ lim inf A(ur , gr ), (u, g) − (v, h)
r →+∞
= lim inf B (ur , gr ), (u, g) , (ur , gr ) − (u, g)
r →+∞
=0 by (92)
+ lim inf B (ur , gr ), (ur , gr ) − B (ur , gr ), (u, g) , (ur , gr ) − (u, g)

r →+∞
≥0 by (89)
+ lim inf A(ur , gr ), (u, g) − (v, h) ≥ lim inf A(ur , gr ), (u, g) − (v, h)
r →+∞ r →+∞
≥ A(u, g), (u, g) − (v, h), ∀{v, h} ∈ V .
This completes the proof of the pseudo-monotonicity.

It remains to prove coercivity. We evaluate directly the term
A(u, g), (u − U, g).
Taking ξ = u − U and ψ = g in (85), the cross terms involving the product p(g)
div (u − U) cancel and the term p0 div (u − U)/τ drops out after integration. What
remains is
A(u, g), (u − U, g)

1 λ∗
= Ge(u) : e(u − U) dx + 2 div (u − U)2 dx
τ Ω τ Ω

n(div U)
+ (ρ(g) − ρ(Ξ ))g dx
Ω τ

n(u∇Υε ) − n(div U)
+ G(g) dx + Kk(u∇Υε )D(g)|∇g|2 dx.
Ω τ Ω
(96)
The third and fourth terms in the right-hand side need special attention.
Since ρ = ρ(g) is a C 1 monotonically increasing function, we have the elementary
inequality
x
x(ρ(x) − ρ(y)) ≥ ζρ (ζ ) dζ, ∀x, y ∈ R. (97)
y
Using this inequality and the expression for G (see (76)) in these terms gives
g
n(div U)(ρ(g) − ρ(Ξ ))g + (n(u∇Υε ) − n(div U)) ζρ (ζ ) dζ
0
g Ξ
≥ n(u∇Υε ) ζρ (ζ ) dζ − n(div U) ζρ (ζ ) dζ. (98)
0 0
123
Applying Korn’s inequality, see Theorem 1.33 from Roubiček (2005), and inserting
inequality (98) into equality (96) yields
C1
A(u, g), (u − U, g) ≥ ||u||2H 1 (Ω)m + C2 ||∇g||2L 2 (Ω)m
τ 0
g
C3 C4
− + ( ζρ (ζ ) dζ ) dx , (99)
τ τ Ω 0

≈Cg 2 for large |g|
where Ci , i = 1, . . . , 4 are positive constants. This proves the coercivity.

Having established pseudo-monotonicity and coercivity of the operator A, we are
in position to apply Brézis’ theorem. This concludes the assertion of the proposition.

Theorem 1 Problem (PD) admits at least one solution (u, g) ∈ V .
Proof For each ε > 0, let (uε , gε ) be a solution of Problem (PDε ) as obtained in
Proposition 2. From the coercivity part of the proof of Proposition 2 and Eq. (84), it
follows that
||uε || H 1 (Ω)m + ||gε || H 1 (Ω) ≤ C, (100)

0
where C is independent of ε. Estimate (100) yields weak compactness in H 1 . However,

this is not enough to prove that uε ∇Υε converges strongly in L 2 and (a.e.) on Ω as
ε → 0. The remedy is to consider the momentum Eq. (83), which gives us improved
regularity through the elasticity term. Since p(gε ) is bounded in H 1 (Ω), uniformly
with respect to ε, we conclude that
||uε || H 2 (Ω)m ≤ C, (101)
where C does not depend on ε. Using estimates (100)–(101), there is a subsequence

(uε , gε ), denoted by the same subscript, and a pair (u, g) ∈ (H01 (Ω)m ∩ H 2 (Ω)m ) ×
H 1 (Ω) such that
uε → u strongly in H01 (Ω)m , (102)

div uε → div u strongly in L 2 (Ω) and (a.e) on Ω, (103)
gε g weakly in H 1 (Ω), (104)
gε → g strongly in L 2 (Ω) and (a.e) on Ω, (105)
as ε → 0. The convergence properties allow to pass to the limit in system (82)–

(83). Hence, the pair (u, g) satisfies the equations of Problem (PD), which proves the
theorem.

To complete the study of the incremental problem, we need to estimate the behavior
of solutions after at least O(1/τ ) times steps. Here, we use the discrete version of
Lyapunov functional (66).
123
In Problem (PD), where the discrete time step τ enters as parameter, one find
after one step (u1 , g 1 ) from the initial values (div u, ρ)|t=0 = (0, ρ 0 ). The idea is to
repeat this procedure for an arbitrary number of steps. If M ∈ N, M ≤ N = T /τ ,
then (u M , g M ) denotes the time discretized approximation of the original quasi-static
equation, at t = t M = Mτ .
The corresponding Lyapunov functional at t = t M reads

1
JM = Ge(u M ) : e(u M ) − F M · u M + n(div u M )G(g M ) dx. (106)
Ω 2
It satisfies
Theorem 2 For each M ∈ N, M ≤ N = T /τ, we have
M
2
div (u j − u j−1 )
J M
+τ λ∗
τ
j=1 Ω

F j − F j−1
+ · u j−1 + k(div u j )D(g j )K∇g j · ∇g j − Q j g j dx ≤ J 0 .
τ
(107)
Here,

J 0 = n0 G(g 0 ) dx, g 0 = g(ρ 0 ).
Ω
Proof At time t = t j , with j = 1, . . . , N , the equations in Problem (PD) read

λ∗
Ge(u j ) : e(ξ) dx + div (u j − u j−1 ) div ξ dx − p(g j )div ξ dx
Ω τ Ω Ω

= F j · ξ dx, ∀ξ ∈ H01 (Ω)m , (108)
Ω

n(div u j−1 ) n(div u j ) − n(div u j−1 )
(ρ(g j ) − ρ(g j−1 )) + G(g j ) ψ dx
Ω τ τgj

p(g j ) − p0
+ div (u j
− u j−1
)ψ dx + k(div u j )D(g j )K∇g j · ∇ψ dx
Ω τgj Ω

= Q j ψ dx, ∀ψ ∈ H 1 (Ω).
Ω
(109)
Note that in Eq. (109) we have used explicitly the form of Dτ from (81). Next, we
take ξ = (u j − u j−1 )/τ in (108) and ψ = g j in (109). The resulting two equalities
are added and summed up with respect to j up from j = 1 to j = M. Using the
observations
(i) cross terms containing pressure cancel;
123
(ii)

M
1
Ge(u j ) : e(u j − u j−1 ) ≥ Ge(u M ) : e(u M ) − Ge(u0 ) : e(u0 ) ;
2
j=1
(iii)
M

n(div u j−1 )g j (ρ(g j ) − ρ(g j−1 )) + (n(div u j ) − n(div u j−1 ))G(g j )
j=1
≥ n(div u M )G(g M ) − n(div u0 )G(g 0 ),
where (97) is used;

(iv)

M
M−1
F · (u − u
j j j−1
)=F M
·u M
−F ·u −
0 0
(F j+1 − F j ) · u j ,
j=1 j=0
one finds inequality (107). The reduced expression for J 0 results from u|t=0 = 0.

Having established existence for the discrete Problem (PD) in Theorem 1 and a
Lyapunov estimate in Theorem 2, we are now in a position to obtain estimates that are
uniform in the time step τ .
Proposition 3 There exists a constant C > 0 such that
||u M ||2H 1 (Ω)m + ||g M ||2L 2 (Ω) ≤ C, (110)

and
M
div (u j − u j−1 ) 2
τ λ∗ + |∇g j |2 dx ≤ C, (111)
Ω τ
j=1
for all M and τ such that 1 ≤ M ≤ N = T /τ , with τ sufficiently small.
Proof Combining expression (106) for J M and inequality (107) yields for any 1 ≤
M≤N

1
Ge(u M ) : e(u M ) dx + n(div u M )G(g M ) dx
2 Ω Ω

M
≤ F M · u M dx + J 0 + τ Q j g j dx
Ω j=1
123
M
F j − F j−1 δ 1
+τ · u j−1 dx ≤ ||u M ||2L 2 (Ω)m + ||F M ||2L 2 (Ω)m + J 0
τ 2 2δ
j=1 Ω
τ j 2 τ j−1 2
M M
+ ||g || L 2 (Ω) + ||u || L 2 (Ω)m
2 2
j=1 j=1
τ
M
τ F j − F j−1 2
M
+ ||Q j ||2L 2 (Ω) + || || L 2 (Ω)m .
2 2 τ
j=1 j=1
By the assumptions on Q and F, the last two terms are uniformly bounded with respect
to τ and M. We estimate the left-hand side from below by applying Korn’s inequality
to the first term and the quadratic growth of G to the second term. Then for δ and τ
sufficiently small, we obtain for the combination
U j = ||u j ||2H 1 (Ω)m + ||g j ||2L 2 (Ω) , j = 0, . . . , M,
the inequality

M−1
U M ≤ C1 + C2 τ Uj,
j=0
where C1 and C2 do not dependent on τ and M. Next, we apply the discrete Gronwall
inequality1 , see footnote, to find
U M ≤ C1 eC2 (M−1)τ < C1 eC2 T for all 1 ≤ M ≤ N .
The second estimate follows directly from Theorem 2.

However, to pass to the limit τ → 0 in the nonlinearities, one needs more information
on the behavior of the ratios { div (u j − u j−1 )/τ } and {(g j − g j−1 )/τ }. In fact, we
must establish relative compactness of the sequences {div u j } and {g j }.
We start with a local H 1 -estimate for E j = div u j .
Lemma 1 Let ϕ ∈ C0∞ (Ω) and τ > 0 sufficiently small. Then, there exists a constant
C = C(ϕ) such that

N
λ∗
τ ||ϕE j ||2H 1 (Ω) + max ||ϕE M ||2H 1 (Ω) ≤ C. (112)
2μ + λ 1≤M≤N
j=1
Proof Let
div (u j − u j−1 )
L j = (2μ + λ)E j − p(g j ) + λ∗ , j = 1, . . . M. (113)
τ
1 Discrete version of Gronwall’s lemma: Let {U } and {w } be nonnegative sequences satisfying U ≤
n n n

A + M−1 M−1
j=0 U j w j . Then for all n, Un ≤ A exp{ j=0 w j }.
123
Inequality (110) implies
M

τ ||u || H 1 (Ω)m + ||g || L 2 (Ω) ≤ τ MC ≤ T C.
j 2 j 2
j=1
Combined with (111) this gives for L j

M
τ ||L j ||2L 2 (Ω) ≤ C. (114)
j=1
As in the counterexample for negative porosity, we take the divergence of the time-
discrete momentum equation. This yields
− ΔL j = div F j in Ω. (115)
In general, however, there are no boundary conditions for L j available. Here, we

must rely on local estimates to obtain (112). Let us first write the equation for ϕ L j ∈
H 2 (Ω) ∩ H01 (Ω):
Δ(ϕ L j ) = −ϕdiv F j + 2∇ϕ · ∇ L j + L j Δϕ.
Its weak form reads

∇(ϕ L j )∇ζ dx = div F j ϕζ dx + L j (2∇ϕ · ∇ζ
Ω Ω Ω
+ζ Δϕ) dx, ∀ζ ∈ H01 (Ω). (116)
Taking ζ = ϕ L j results in

|∇(ϕ L )| dx = −
j 2
F · ∇(ϕ L )ϕ dx −
j
F j · ∇ϕϕ L j dx
j
Ω Ω Ω

+ (L j )2 ϕΔϕ dx − 2L j ∇ϕ · ∇(ϕ L j ) dx.
Ω Ω
With C = C(ϕ) denoting a generic constant depending on ϕ, we have
||ϕ L j ||2H 1 (Ω) ≤ C ||F j ||2L 2 (Ω) + ||L j ||2L 2 (Ω) , (117)
for 1 ≤ j ≤ M ≤ N . Combining this inequality with (114) gives

M
M
τ ||ϕ L j ||2H 1 (Ω) ≤ C(τ ||F j ||2L 2 (Ω)m + 1) ≤ C. (118)
j=1 j=1
123
Next, we multiply expression (113) by τ ϕ and write it as
τ (2μ + λ)ϕE j + λ∗ ϕ(E j − E j−1 ) = τ L j ϕ + τ p(g j )ϕ ∈ H 1 (Ω).
Taking the H 1 -inner product of this expression with ϕE j gives
(2μ + λ)τ ||ϕE j ||2H 1 (Ω) + λ∗ (ϕ(E j − E j−1 ), ϕE j ) H 1 (Ω)

= τ (ϕ(L j + p(g j )), ϕE j ) H 1 (Ω)
or
τ
(2μ + λ) ||ϕE j ||2H 1 (Ω) + λ∗ (ϕ(E j − E j−1 ), ϕE j ) H 1 (Ω)
2
τ
≤ ||ϕ(L j + p(g j ))||2H 1 (Ω) . (119)
2(2μ + λ)
Using the identity

M
(a M )2 (a 0 )2 1 j
M
a (a − a
j j j−1
)= − + (a − a j−1 )2 ,
2 2 2
j=1 j=1
when summing up (119) gives

M
λ∗
τ ||ϕE j ||2H 1 (Ω) + ||ϕE M ||2H 1 (Ω)
2μ + λ
j=1
τ
M
≤+ ||ϕ(L j + p(g j ))||2H 1 (Ω) .
(2μ + λ)2
j=1
Combining this inequality with (111) and (118), results in the estimate of the lemma.

We conclude this section with an estimate for (ρ(g j )−ρ(g j−1 ))/τ . However, since
in Eqs. (67) or (69) the (discrete) time derivative is multiplied by n(E), we look for an
estimate for
N j = n(E j )ρ(g j ). (120)
With the results of Proposition 3 and Lemma 1, the space-time compactness of N will
imply the same property of g.
We summarize our findings in the next proposition
Proposition 4 For given τ > 0 and j = 1, . . . , N , let (uτ (t j ), gτ (t j )) ∈ V denote a
solution of Problem (PD). Then, we have
max ||uτ (t j )|| H 1 (Ω)m + ||gτ (t j )|| L 2 (Ω) ≤ C, (121)

1≤ j≤N
123
N

div (uτ (t j ) − uτ (t j−1 )) 2
τ λ∗ + |∇gτ (t j )|2 dx ≤ C, (122)
Ω τ
j=1

N
τ ||ϕ div uτ (t j )||2H 1 (Ω) + λ∗ max ||ϕ div uτ (t j )||2H 1 (Ω) ≤ C(ϕ), (123)
1≤ j≤N
j=1

N
N j − N j−1 2
τ || || H −2 (Ω) + ||ϕN j ||2H 1 (Ω) ≤ C(ϕ), (124)
τ
j=1
where
N j = n(div uτ (t j ))ρ(gτ (t j ))
and where ϕ ∈ C0∞ (Ω).
Proof We only need to prove estimate (124). Rewriting Eq. (79), we have

N j − N j−1 n(E j−1 )(ρ(g j ) − ρ(g j−1 ))
ψ dx = ψ dx
Ω τ Ω τ

(n(E j ) − n(E j−1 ))ρ(g j )
+ ψdx
Ω τ

n(E j ) − n(E j−1 )
= ρ(g j ) − G(g j ) ψdx
Ω τ

(E j − E j−1 )( p(g j ) − p0 )
− ψ dx
Ω τgj

+ Q j ψ dx − k(E j )D(g j )K∇g j ∇ψ dx, for ψ ∈ H02 (Ω).
Ω Ω
Recalling that for m ≤ 3, H 2 (Ω) ⊂ L ∞ (Ω), we have
N j − N j−1 2
|| || H −2 (Ω)
τ
(E j − E j−1 ) 2
≤ C || || L 2 (Ω) ||g j ||2L 2 (Ω) + ||Q j ||2L 2 (Ω) + ||∇g j ||2L 2 (Ω)m
τ
and the full estimate reads

N
N j − N j−1 2
τ || || H −2 (Ω)
τ
j=1

N
E j − E j−1 ) 2
N
≤C max ||g j ||2L 2 (Ω) τ || || L 2 (Ω) + 1 + τ ||∇g j ||2L 2 (Ω)m ≤ C
1≤ j≤N τ
j=1 j=1
(125)
123
The local estimate for the space derivatives is given by

N
τ ||∇(ϕN j )||2L 3/2 (Ω)
j=1

N
N
≤ C ( max ||∇(ϕE j )||2L 2 (Ω) )τ ||g j ||2L 6 (Ω) + C + τ ||∇g j ||2L 2 (Ω)m ≤ C.
1≤ j≤N
j=1 j=1
(126)
This results in estimate (124).

4 Existence for Continuous Time Problem with ∗ > 0
In Proposition 4, where the time step τ enters as a parameter, one finds

{(uτ (t j ), gτ (t j ))} j=1,...,N from the “initial value” div u(0) = 0 and g(0) = g 0 . Here
N = O(1/τ ) and g 0 = g(ρ 0 ). This procedure yields a time discretized approximation
of the original quasi-static equations.
In this section, we investigate the limit τ 0. Here a crucial role is played by the
parameter λ∗ , which is needed to control the behavior in time of E = div u.
Using the discrete solution (uτ (t j ), gτ (t j )), we construct two approximations that
hold for all 0 ≤ t ≤ T . The first is the piecewise constant approximation
(uτ (t), g τ (t)) = (uτ (t j ), gτ (t j )) for jτ ≤ t < ( j + 1)τ. (127)
The second is the Rothe interpolant, which is the piecewise linear time-continuous
approximation
t t
(ũτ (t), g̃τ (t)) = j + 1 −(uτ (t j ), gτ (t j )) + − j (uτ (t j+1 ), gτ (t j+1 )),
τ τ
for jτ ≤ t ≤ ( j + 1)τ. (128)
In (127) and (128), the index j runs from j = 0 to j = N − 1.

Applying Proposition 4 yields for both approximations, with ∈ {− ,∼ },
max ||uτ (t)||2H 1 (Ω)m + ||gτ (t)||2L 2 (Ω) dt ≤ C, (129)

0≤t≤T
T
|∇g τ (t)|2 dxdt ≤ C, (130)
0 Ω
T
||ϕ Eτ (t)||2H 1 (Ω) dt ≤ C, (131)
0
λ max ||ϕEτ (t)||2H 1 (Ω) ≤ C,
∗
(132)
0≤t≤T
T
||ϕNτ (t)||2W 1,3/2 (Ω) ≤ C, (133)
0
123

where Eτ = div uτ , N τ = n(E τ )ρ(g τ ) and Ñτ (t) = ( j +1−t/τ )N j +(t/τ − j)N j+1 .
Further, we have
N j+1 − N j E j+1 − E j
∂t Ñτ = and ∂t Ẽτ = ,
τ τ
for t j ≤ t ≤ t j+1 and j = 0, . . . , N − 1.
Hence, by (122)
T
λ∗ |∂t Ẽτ (t)|2 dxdt ≤ C (134)
0 Ω
and
T
||∂t Ñτ (t)||2H −2 (Ω) dt ≤ C. (135)
0
In what follows, we rely heavily on the material and theory collected in Roubiček
(2005, Chapters 7 and 8). Since the piecewise constant approximation (uτ (t), g τ (t))
is discontinuous in time, its time derivative is only a measure. To deal with this, we
introduce the space M(0, T ; L 2 (Ω)) of regular Borel measures in [0, T ] with values
in L 2 (Ω), which is the dual space of C([0, T ]; L 2 (Ω)). With δ(t j ) denoting the Dirac
measure concentrated in t j , we have

N
||∂t E τ ||M(0,T ;L 2 (Ω)) = || (E j − E j−1 )δ(t j )||M(0,T ;L 2 (Ω))
j=1

N
E j − E j−1
=τ || || L 2 (Ω) = ||∂t Ẽτ || L 1 (0,T ;L 2 (Ω))
τ
j=1
√
≤ T ||∂t Ẽτ || L 2 (0,T ;L 2 (Ω)) ≤ C. (136)
Analogously
||∂t N τ ||M(0,T ;H −2 (Ω)) ≤ C, (137)
where M(0, T ; H −2 (Ω)) is the dual space of C([0, T ]; H02 (Ω)).

For the convergence of the time-continuous approximation (128), we use estimates
(129)–(135) and the well-known weak and weak∗ compactness theorems. The result
is that there exists a quadruple {ũ, g̃, Ẽ, Ñ } such that along a subsequence τ 0 we
have
ũτ → ũ weak∗ in L ∞ (0, T ; H01 (Ω)m ), (138)

g̃τ g̃ weakly in L (0, T ; H (Ω)), 2 1
(139)
Ẽτ Ẽ weakly in L 2 (0, T ; H 1 (ω)), (140)
123
∂t Ẽτ ∂t Ẽ weakly in L 2 (0, T ; L 2 (Ω)), (141)

Ñτ Ñ weakly in L (0, T ; W 2
(ω)), 1,3/2
(142)
∂t Ñτ ∂t Ñ weakly in L 2 (0, T ; H −2 (Ω)). (143)
Concerning the convergence of (uτ , g τ ), we use estimates (129)–(133), now combined

with (136)–(137). Moreover, applying (Roubiček 2005, Corollary 7.9), we use that
the spaces
dz
W 1,2,M (0, T ; H 1 (ω), L 2 (ω)) = {z ∈ L 2 (0, T ; H 1 (ω)) | ∈ M(0, T ; L 2 (ω))}
dt
and W 1,2,M (0, T ; W 1,3/2 (ω), H −2 (ω)) are compactly embedded in

L (0, T ; L 2 (ω)), for any smooth bounded subset ω of Ω. The result is that there
2
exists (u, g, , E, N ) such that along a subsequence τ 0 one has the same conver-
gence as in (138)–(140) and (142). The convergence in (141) and (143) is now replaced
by weak−∗ convergence in M(0, T ; L 2 (Ω)) for ∂t E τ and in M(0, T ; H −2 (Ω)) for
∂t N τ .
Furthermore, the estimates allow us to conclude
E τ → E strongly in L 2 ((0, T ) × ω) and (a.e) on (0, T ) × ω, (144)

N τ → N strongly in L 2 ((0, T ) × ω) and (a.e) on (0, T ) × ω. (145)
As a consequence,
Nτ N
ρ(g τ ) = → (146)
n(E τ ) n(E)
and
Nτ N
g τ = ρ −1 → ρ −1 = g. (147)
n(E τ ) n(E)
strongly in L 2 ((0, T ) × ω) and a.e. on (0, T ) × ω. This in turn implies

ρ(g τ ) → ρ(g) strongly in L 2 ((0, T ) × ω) and (a.e) on (0, T ) × ω;
(148)
D(g τ ) → D(g) strongly in L 2 ((0, T ) × ω) and (a.e) on (0, T ) × ω.
Inherited from E τ = div uτ , the convergence properties imply
E = div u a.e. in (0, T ) × Ω. (149)
123
As in (Roubiček 2005, pages 224–226), one shows that ũ = u and g̃ = g. Then, (149)
implies that E = Ẽ. Alternatively, this follows from estimate (134) which gives
N −1
T τ3
||E τ (t) − Ẽτ (t)||2L 2 (Ω) dt = ||E j − E j+1 ||2L 2 (Ω)
0 2
j=1
= Cτ ||∂t Ẽτ ||2L 2 (0,T ;L 2 (Ω)) = Cτ 2 .

2
(150)
Similarly,
T
||N τ (t) − Ñτ (t)||2H −2 (Ω) dt = Cτ 2 , (151)
0
which yields N = Ñ .
From this point on, we denote the limit, as τ 0, by the quadruple (u, g, E, N ),
where
E = div u and N = n(E)ρ(g).
We are now in a position to prove the main existence result for a weak solution of the
time-continuous case.
Theorem 3 Let λ∗ > 0. Then, there exists at least one weak free energy solution
(u, E, ρ) satisfying Definition 1.
Proof In the proof, we use approximations (127) and (128), and their convergence
properties.
Let τ > 0, sufficiently small, and let t ∈ (τ, T ). Then, t j ≤ t < t j+1 for some
j ∈ {1, . . . , N − 1} and uτ (t) = u j and g τ (t) = g j .
We first consider the momentum balance Eq. (80).
The starting point is Problem (PD). Using Eq. (108), we have for any ξ ∈ H01 (Ω)m

Ge(uτ ) : e(ξ) dx = Ge(u j ) : e(ξ) dx
Ω Ω

λ∗
=− (E j − E j−1 ) div ξ dx + p(g j )div ξ dx + F j · ξ dx
τ Ω
Ω Ω
∗
= −λ ∂t Ẽτ (t − τ ) div ξ dx + p(g τ )div ξ dx + Fτ · ξ dx. (152)
Ω Ω Ω
Here, we introduced
Fτ (t) = F(t j ) = F j for t j ≤ t < t j+1 and j = 0, . . . , N − 1.
Multiplying Eq. (152) by α ∈ C0∞ (0, T ) and integrating the result over (τ, T ), yields
T T
{ Ge(uτ ) : e(ξ) dx}α(t) dt + λ∗ { ∂t Ẽτ (t − τ ) div ξ dx}α(t) dt
τ Ω τ Ω
123
T T
+ { p(g τ )div ξ dx}α(t) dt = { Fτ · ξ dx}α(t) dt. (153)
τ Ω τ Ω
Next, we send τ 0 along the appropriate subsequence to have convergence of the

terms containing u, E and F. What remains is the pressure term. We recall that p(g)
is the composite function ( p ◦ ρ)(g), where p(ρ) is given by (36)and ρ(g) is defined
through (53) and (55). Since g τ → g strongly in L 2 ((0, T ) × ω), see (147), we have
similarly
p(g τ ) = ( p ◦ ρ)(g τ ) → ( p ◦ ρ)(g) = ( p ◦ ρ)(g) = p(g)

strongly in L 2 ((0, T ) × ω) and a.e. in (0, T ) × ω.
This concludes the first part of the proof.

Next, we tackle the mass balance Eq. (69).
We first put Eq. (77) in the form
∂t N − ∂t n(E)ρ(g) + D(g, E)∂t E − div k(E)D(g)K∇g = Q
and apply the discretization of Problem (PD). Similarly to (153) this gives for any
ψ ∈ C0∞ (Ω) and α ∈ C ∞ [0, T ]

T G(g τ ) p(g τ ) − p0
∂t Ñτ (t − τ ) − ∂t ν̃τ (t − τ ) ρ(g τ ) − + ∂t Ẽτ (t − τ )
τ Ω gτ gτ
ψ(x)α(t) dxdt +
T T
k(E τ )D(g τ )K∇g τ · ∇ψ(x)α(t) dxdt = Q τ ψ(x)α(t) dxdt,
τ Ω τ Ω
(154)
where

t t
ν̃τ (t) = j +1− n(E j ) + − j n(E j+1 ),
τ τ
and
Q τ (t) = Q(t j ) = Q j
for jτ ≤ t < ( j + 1)τ. (155)
The boundedness of n implies
||∂t ν̃τ || L 2 ((0,T )×Ω) ≤ C (156)
and inherited from (131)
||ν̃τ || L 2 (0,T ;H 1 (ω) ≤ C. (157)
123
Hence,
∂t ν̃τ ∂t n(E) weakly in L 2 ((0, T ) × Ω) (158)
and
ν̃τ → n(E) strongly in L 2 ((0, T ) × ω) and a.e. in (0, T ) × ω. (159)
We are now in position to pass to the limit τ 0 in (154) and obtain

T
∂t (n(E)ρ(g)), ψ H −2 (Ω),H 2 (Ω) α(t) dt
0
0

T G(g) p(g) − p0
− ∂t n(E) ρ(g) − + ∂t E ψ(x)α(t) dxdt
0 Ω g g
T T
+ k(E)D(g)K∇g · ∇ψ(x)α(t) dxdt = Qψ(x)α(t) dxdt,
(160)
0 Ω 0 Ω
or

∂t n(E)ρ(g) − ∂t n(E)ρ(g) + D(ρ, E)∂t E − div k(E)D(g)K∇g
= Q in D ((0, T ) × Ω), (161)
It remains to check the initial and boundary conditions and the energy inequality (72).
First, we notice that (140)–(141) imply
Ẽτ E weakly in W 1,2,2 (0, T ; H 1 (ω), L 2 (ω)), (162)
where W 1,2,2 (0, T ; H 1 (ω), L 2 (ω))={z ∈ L 2 (0, T ; H 1 (ω))|∂t z ∈ L 2 (0, T ; L 2 (ω))}.

In this space, the trace in time E → E(0) is a weakly continuous map from W 1,2,2 (0, T ;
H 1 (ω), L 2 (ω)) to L 2 (ω). Hence,
Ẽτ (0)E(0) weakly in L 2 (ω), (163)
where E(0) = div u0 = 0.

Next, using (142), (143) and (145), we conclude that
Ñτ (0)N (0) weakly in H −2 (ω), (164)
which justifies the initial condition for N . Since N = n(E)ρ(g), we have

simultaneously the initial conditions for the density ρ and for g.
We still miss the flux boundary condition for the mass balance equation (67). The
starting point is again Eq. (154), now with ψ ∈ H 1 (Ω) and α(T ) = 0. Since
T T
d
∂t Ñτ (x, t − τ )ψ(x)α(t) dxdt = − Ñτ (x, t − τ )ψ(x) α(t) dxdt
τ Ω τ Ω dt
123
T
d
− Ñτ (x, 0)ψ(x)α(τ ) dx → − Ñ (x, t)ψ(x) α(t) dxdt
Ω Ω dt
0
−α(0) n(0)ρ 0 (x)ψ(x) dx

Ω
and since the strong convergence (144)–(145), together with the weak convergence
(140) and (142), implies the same for Ω, we may pass to the limit τ 0 and conclude
that
T
− ρ(g)n(E)∂t Φ(x, t) dxdt − n(0)ρ 0 (x)Φ(x, 0) dx
0 Ω Ω
T
+ ∂t E D(ρ(g), E)
0 Ω
T
−ρ(g)n (E) Φ(x, t) dxdt + k(E)D(g)K∇g · ∇x Φ(x, t) dxdt
0 Ω
T
= QΦ(x, t) dxdt, ∀Φ ∈ H 1 (Ω × (0, T )), with Φ|t=T = 0.
0 Ω
To show inequality (72), we follow again (Roubiček 2005, pp. 223–226). Starting
point is inequality (107), which we write for any K ∈ N, K ≤ N − 1, and for any
time step τ as
K +1
E − EK 2
J K +1− +τ
JK λ∗ + k(E K +1 )D(g K +1 )K∇g K +1 · ∇g K +1
Ω τ
(165)
F K +1 − F K K +1 +1
+ ·u − Q K g K dx ≤ 0.
τ
With the notation from (127) and (128), we rewrite inequality (165):

d ˜
λ∗ ∂t Ẽτ (t) + k E τ (t + τ ) D ḡτ (t + τ ) K∇ ḡτ (t + τ ) · ∇ ḡτ (t + τ )
2
Jτ (t) +
dt Ω
+ ∂t F̃τ (t) · u(t) − Q τ (t + τ )ḡτ (t + τ ) dx ≤ 0,
for K τ ≤ t ≤ (K + 1)τ and 0 ≤ K ≤ N − 1.

Integrating this inequality from t = t1 to t = t2 , with 0 ≤ t1 ≤ t2 ≤ T , using
the convergence results from (138)–(145) and the weak lower semi-continuity of the
gradient ∇ ḡτ in L 2 (Q T ), yields the desired result.

5 Discussion and Conclusion
In this paper, we study a model that describes the quasi-static mechanical behavior
of a fluid saturated porous medium. In it simplest (linear) form, it is described by
123
Eqs. (1)–(4), where (1) results from the fluid phase mass balance in the case that the
fluid is incompressible.
We follow Rutqvist et al. (2001) and Lewis and Schrefler (1998) and propose a fluid
mass balance that is based on the mixture theory of Bedford and Drumheller (1978)
and Bedford and Drumheller (1983). This yields Eq. (6) and the resulting nonlinear
system is given by (2)–(4) and (6). Note that the time derivative of the fluid density ρ
appears in (6), since the fluid is assumed weakly compressible, see expression (10).
Models where the fluid density is constant (see Bociu et al. 2016; Cao et al. 2013)
do not contain this source term. Moreover, the porosity n and the deformation of the
medium are related through (8). An expression for this relation is derived from the
solid phase mass balance. It is given by (19) or, when the deformation is small, by
approximation (21).
It is shown by means of a counterexample that the porosity may admit non-physical,
i.e., negative, values. This is made precise in Proposition 1. To obtain a well-posed
mathematical problem, the porosity is modified according to cutoff (35). This cutoff
is chosen such that it reduces to the correct expression in the physical range. Outside
this range, it remains positive. Likewise, a cutoff for the density is introduced through
expressions (60) and (64).
The momentum balance Eq. (2)–(4) is modified as well. Following Murad and
Cushman (1996), we add the term
λ∗ div ∂t u (λ∗ ≥ 0) (166)
to the expression for the total stress. This results in expression (5). Murad and Cushman
give a thermodynamically based derivation of the equation in which (166) appears as
the difference between the fluid and solid pressures. Having λ∗ > 0, (166) acts as a
time regularization of the volumetric stress for our quasi-static problem.
An important role in the analysis of the equations is played by the free energy of
the system. This free energy acts as a Lyapunov functional. It is given by (66), which
generalizes Biot’s original expression developed for the linear case (Biot 1962). In the
case that the deformation and fluid density are in the physical range, the free energy
simplifies to, see also (56),

1
J (u, ρ) = Ge(u) : e(u) − F · u
Ω 2

n(div u)
+ (1 − n 0 )ρ − ρ0n 0 ρ 1−n 0 + n 0 ρ0 dx. (167)
β0 n 0 (1 − n 0 )ρ0
We introduce a weak formulation and prove existence of a solution in a number of

steps. Discretizing in time, we first consider the incremental equations. Using Brézis’
fundamental theorem for pseudo-monotone operators, see for instance Lions (1969)
and Roubiček (2005), we obtain existence for the corresponding incremental problem.
The result holds for any λ∗ ≥ 0. Moreover, using the free energy, estimates that are
global in time are derived. These (stability) estimates are crucial when considering
the time-continuous, quasi-static, formulation for which we prove existence at the
expense of having λ∗ > 0. The free energy implies global stability of the solution.
123
The quasi-static case with λ∗ = 0 fails to have the time estimate from inequality (111).
Therefore, we are not able to extend the results to the quasi-static case. Furthermore,
the convergence results are based on compactness arguments. Hence, also uniqueness
remains an open problem. Clearly, we have existence (globally in time) for the time-
discrete quasi-static case, which is relevant for numerical purposes.
We note that only in the proof of the local H 1 (Ω)− estimates for div u, we use the
fact that the Gassmann tensor has the specific form of Hooke’s law (4). In the incremen-
tal problem we could have replaced G by a general rank-4, symmetric, positive-definite
Gassmann tensor.
Some particular cases of system (24)–(25) were studied before. An interesting
example is the consolidation with an irrotational composite flow rate, when the system
reduces to a scalar pseudo-parabolic PDE. For details see Holland and Showalter
(2018).
We notice also that the model studied in this paper was extensively used by Schrefler
et al., see Schrefler et al. (1990) and Lewis and Schrefler (1998) and references therein.
It is broadly accepted in the computational poromechanics community. A review of
different numerical methods and software is given in Rutqvist et al. (2001) and Minkoff
et al. (2003).
Acknowledgements The first author expresses his gratitude to Tea Mikelić for retrieving the missing files
from her father’s computer. Further, he wants to thank the colleagues from Hasselt University, Sorin Pop
and Koondanibha Mitra, for their assistance with the revision.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give
appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence,
and indicate if changes were made. The images or other third party material in this article are included
in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If
material is not included in the article’s Creative Commons licence and your intended use is not permitted
by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
References
Auriault, J.-L., Sanchez-Palencia, E.: Étude du compartement macroscopique d’un milieu poreux.saturé
deformable. J. Mécanique 16, 575–603 (1977)
Bear, J., Bachmat, Y.: Introduction to Modeling of Transport Phenomena in Porous Media. Springer Science
and Business Media, New York (1990)
Bedford, A., Drumheller, D.S.: A variational theory of immiscible mixtures. Arch. Ration. Mech. Anal. 68,
37–51 (1978)
Bedford, A., Drumheller, D.S.: Theories of immiscible and structured mixtures. Int. J. Eng. Sci. 21, 863–960
(1983)
Biot, M.A.: Mechanics of deformation and acoustic propagation in porous media. J. Appl. Phys. 33, 1482
(1962)
Bociu, L., Guidoboni, G., Sacco, R., Webster, J.T.: Analysis of nonlinear poro-elastic and poro-visco-elastic
models. Arch. Ration. Mech. Anal. 222, 1445–1519 (2016)
Bosco, E., Peerlings, R.H., Geers, M.G.: Predicting hygro-elastic properties of paper sheets based on an
idealized model of the underlying fibrous network. Int. J. Solids Struct. 56, 43–52 (2015)
Čanić, S., Hartley, C.J., Rosenstrauch, D., Tambača, J., Guidoboni, G., Mikelić, A.: Blood flow in compliant
arteries: an effective viscoelastic reduced model, numerics and experimental validation. Ann. Biomed.
Eng. 34, 575–592 (2006)
123
Cao, X., Pop, I.S.: Degenerate two-phase porous media flow model with dynamic capillarity. J. Differ. Equ.
260, 2418–2456 (2016)
Cao, Y., Chen, S., Meir, A.J.: Analysis and numerical approximations of equations of nonlinear
poroelasticity. Discrete Contin. Dyn. Syst. Ser. 18, 1253–1273 (2013)
Cardoso, L., Fritton, S.P., Gailani, G., Benalla, M., Cowin, S.C.: Advances in assessment of bone porosity,
permeability, and interstitial fluid flow. J. Biomech. 46, 253–265 (2013)
Coussy, O.: Poromechanics. Wiley (2004)
Cowin, S.C.: Bone poroelasticity. J. Biomech. 32, 218–238 (1999)
Evans, L.C.: A survey of entropy methods for partial differential equations. Bull. Am. Math. Soc. 41,
409–438 (2004)
Holland, E., Showalter, R.E.: Poro-visco-elastic compaction in sedimentary basins. SIAM J. Math. Anal.
50, 2295–2316 (2018)
Jüngel, A.: Entropy Methods for Diffusive Partial Differential Equations. BCAM Springer Briefs. Springer
(2016)
Lewis, R.W., Schrefler, B.A.: The Finite Element Method in the Static and Dynamic Deformation and
Consolidation of Porous Media. Wiley (1998)
Lions, J.L.: Quelques méthodes de résolution des problémes aux limites non linéaires. Dunod, Paris (1969)
Marciniak-Czochra, A., Mikelić, A.: A rigorous derivation of the equations for the clamped Biot–Kirchhoff–
Love poroelastic plate. Arch. Ration. Mech. Anal. 215, 1035–1062 (2015)
Mei, C.C., Vernescu, B.: Homogenization Methods for Multiscale Mechanics. World scientific (2010)
Mikelić, A.: A global existence result for the equations describing unsaturated flow in porous media with
dynamic capillary pressure. J. Differ. Equ. 248, 1561–1577 (2010)
Mikelić, A., Tambača, J.: Derivation of a poroelastic flexural shell model. Multiscale Model. Simul. 14,
364–397 (2016)
Mikelić, A., Wheeler, M.F.: On the interface law between a deformable porous medium containing a viscous
fluid and an elastic body. Math. Models Methods Appl. Sci. 22, 1250031 (2012)
Milišić, J.P.: The unsaturated flow in porous media with dynamic capillary pressure. J. Differ. Equ. 264,
5629–5658 (2018)
Minkoff, S.E., Stone, C.M., Bryant, S., Peszynska, M., Wheeler, M.F.: Coupled fluid flow and geomechanical
deformation modeling. J. Pet. Sci. Eng. 38, 37–56 (2003)
Murad, M.A., Cushman, J.H.: Multiscale flow and deformation in hydrophilic swelling porous media. Int.
J. Eng. Sci. 34, 313–338 (1996)
Owczarek, S.: A Galerkin method for Biot consolidation model. Math. Mech. Solids 15, 42–56 (2010)
Prosi, M., Zunino, P., Perktold, K., Quarteroni, A.: Mathematical and numerical models for transfer of
low-density lipoproteins through the arterial walls: a new methodology for the model set up with
applications to the study of disturbed lumenal flow. J. Biomech. 38, 903–917 (2005)
Roubiček, T.: Nonlinear Partial Differential Equations with Applications. Springer (2005)
Rutqvist, J., Börgesson, L., Chijimatsu, M., Kobayashi, A., Jing, L., Nguyen, T.S., Noorishad, J., Tsang, C.F.:
Thermohydromechanics of partially saturated geological media: governing equations and formulation
of four finite element models. Int. J. Rock Mech. Min. Sci. 38, 105–127 (2001)
Sanchez-Palencia, E.: Non-homogeneous Media and Vibration Theory, Springer Lecture Notes in Physics,
vol. 129. Springer (1980)
Schrefler, B.A., Simoni, L., Xikui, L., Zienkiewicz, O.C.: Mechanics of partially saturated porous media.
In: Desai, D.S., Gioda, G. (eds.) Numerical Methods and Constitutive Modelling in Geomechanics,
Courses and Lectures CISM no. 311, pp. 169–211. Springer, New York (1990)
Schweizer, B.: Partielle Differentialgleichungen. Springer, Berlin (2018)
Showalter, R.E.: Diffusion in poro-elastic media. J. Math. Anal. Appl. 251, 310–340 (2000)
Terzaghi, K.: Theoretical Soil Mechanics. Chapman and Hall Limited, London (1951)
Tolstoy, I. (ed.): Acoustics, Elasticity, and Thermodynamics of Porous Media. Twenty-One Papers by M.A.
Biot. Acoustical Society of America, New York (1992)
van Duijn, C.J., Mikelić, A., Wick, T.: A monolithic phase-field model of a fluid-driven fracture in a nonlinear
poroelastic medium. Math. Mech. Solids 24, 1530–1555 (2019)
Verruijt, A.: Theory and problems of poroelasticity (2015). https://geotechpedia.com
Zeidler, E.: Nonlinear Functional Analysis and Its Applications II/B Nonlinear Monotone Operators.
Springer, New York (1990)
Ženíšek, A.: The existence and uniqueness theorem in Biot’s consolidation theory. Appl. Math. 29, 194–211
(1984)
123
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
123

Mathematical Theory of Nonlinear Single-Phase Poroelasticity

Uploaded by

Copyright:

Available Formats

Mathematical Theory of Nonlinear Single-Phase Poroelasticity

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mathematical Theory of Nonlinear Single-Phase Poroelasticity

Uploaded by

Copyright:

Available Formats

Journal of Nonlinear Science (2023) 33:44

Mathematical Theory of Nonlinear Single-Phase

C. J. van Duijn1,2 · Andro Mikelić3

Communicated by Alain Goriely.

Andro Mikelić passed away on 28 November 2020.

1 Department of Mechanical Engineering, Darcy Center, Eindhoven University of Technology,

Keywords Quasi-static nonlinear poroelasticity · Free energy · Incremental

σ = Ge(u) − α pI, (3)

G E = 2μE + λTr(E)I, for symmetric matrices E. (4)

In these equations, u [m] denotes skeleton displacement, K [m2 ] intrinsic permeability

σ = 2μe(u) + (λdiv u + λ∗ div ∂t u − α p)I, (5)

n∂t ρ + ρdiv ∂t u + div j = Q, (6)

where j denotes the Darcy mass flux

Here, n denotes porosity, ρ = ρ f [kg/m3 ] fluid density, k relative permeability and Q

n = n(div u). (8)

0 < n < 1. (9)

in a realistic porosity interval, bounded away from n = 0 and n = 1. Thus, taking k

2.1 Balance Equations

Restricting themselves to small displacements u (within the elastic regime),

n∂t ρ + ρdiv vs + div j = Q (fluid phase) (15)

∂t (1 − n) + (1 − n)div vs = 0 (solid phase), (16)

where j is mass flux (7).

Using this in (15) and (16) gives

n∂t ρ + ρ∂t div u + div j = Q (17)

Integrating (18) in time from t = 0, say, to t > 0, we have

1 − n = (1 − n 0 )e−div (u − U0 ) for t > 0.. (19)

n < 1 in Ω for all t > 0. (20)

To avoid technical complications, we restrict ourselves to n 0 = constant in Ω.

Throughout the paper, we redefine

where U0 ∈ H01 (Ω)m ∩ H 2 (Ω)m is the initial displacement. Redefining accordingly

F := F + div (Ge(U0 )), (23)

2.2 Negative Porosity

− div 2μe(u) + (λdiv u − p)I = 0 in Ω., (29)

Δ (2μ + λ)div u − p = 0 in Ω. (30)

Hence the function

H = Σ 1,0 at {x1 = 0}.

Repeating this along the other edges gives

where Σ b denotes the given value of σ along the edges.

n(E(x, t)) < 0 for t > T p and x ∈ Ω. (32)

we have for E = div u the problem

the result is immediate. 

2.3 Modification of Balance Equations

In a number of steps, we modify Eq. (24) so that it becomes well-posed in a

n(E)∂t ρ + d(ρ)∂t E − div k(E)D(ρ)K∇ρ = Q, (37)

− div Ge(u) + (λ∗ ∂t E − p)I = F, (39)

where E = div u and where λ∗ ≥ 0.

Q T = {(x, t) : x ∈ Ω, 0 < t < T },

where T > 0 is arbitrarily chosen. To avoid technical complications, we take ∂Ω ∈ C 1

E|t=0 = 0 and ρ|t=0 = ρ 0 in Ω, (40)

u|∂Ω = 0, ∇ρ · ν|∂Ω = 0, for 0 < t ≤ T . (41)

where ν is the outward unit normal at ∂Ω.

2.4 Lyapunov Functional

− p(ρ)∂t E dx = − ∂t F · u dx. (42)

the first term in (43) can be written as

Note that G is a nonnegative, convex function with G(ρ0 ) = 0.

in expression (46) simplifies to

expression (48) vanishes if g(ρ) is chosen such that

ρ0 g(ρ) − p(ρ) = constant = − p0 .

Using these expressions in (46) yields

the result is immediate.

B (w, ), (u1 , g1 ) − B (w, ), (u2 , g2 ) , (u1 , g1 ) − (u2 , g2 ) ≥ 0, (89)

(ur , gr )(u, g) weakly in V and lim supA(ur , gr ), (ur , gr ) − (u, g) ≤ 0.

The second estimate follows directly from Theorem 2.