Nonlinear Analysis: Real World Applications ( ) –

Contents lists available at SciVerse ScienceDirect

Nonlinear Analysis: Real World Applications

journal homepage: www.elsevier.com/locate/nonrwa

Sulfate attack in sewer pipes: Derivation of a concrete corrosion model

via two-scale convergence
Tasnim Fatima a,∗ , Adrian Muntean b
a
Centre for Analysis, Scientific computing and Applications (CASA), Department of Mathematics and Computer Science, Technical University Eindhoven,
Eindhoven 5600, The Netherlands
b
Centre for Analysis, Scientific computing and Applications (CASA), Institute for Complex Molecular Systems (ICMS), Department of Mathematics and Computer
Science, Technical University Eindhoven, Eindhoven, The Netherlands

article info abstract

Keywords: We explore the homogenization limit and rigorously derive upscaled equations for a
Sulfate corrosion of concrete
microscopic reaction–diffusion system modeling sulfate corrosion in sewer pipes made of
Periodic homogenization
concrete. The system, defined in a periodically-perforated domain, is semi-linear, partially
Semi-linear partially dissipative system
Two-scale convergence dissipative and weakly coupled via a non-linear ordinary differential equation posed
Periodic unfolding method on the solid–water interface at the pore level. First, we show the well-posedness of
Multiscale system the microscopic model. We then apply homogenization techniques based on two-scale
convergence for a uniformly periodic domain and derive upscaled equations together
with explicit formulas for the effective diffusion coefficients and reaction constants. We
use a boundary unfolding method to pass to the homogenization limit in the non-linear
ordinary differential equation. Finally, we give the strong formulation of the upscaled
system.
© 2012 Elsevier Ltd. All rights reserved.

1. Introduction

This paper treats the periodic homogenization of a semi-linear reaction–diffusion system coupled with a nonlinear
ordinary differential equation arising in the modeling of the sulfuric acid attack in sewer pipes made of concrete. The
concrete corrosion situation we are dealing with here strongly influences the durability of cement-based materials especially
in hot environments leading to spalling of concrete and macroscopic fractures of sewer pipes. It is financially important to
have a good estimate on the moment in time when such pipe systems need to be replaced, for instance, at the level of a
city like Los Angeles, USA. To get good such practical estimates, one needs on one side good and easy-to-use macroscopic
corrosion models to be used for a numerical forecast of corrosion, while on the other side one needs to ensure the reliability of
the averaged models by allowing them to incorporate a certain amount of microstructure information. The relevant question
is: How much of this oscillatory-type information is needed to get a sufficiently accurate description of the heterogeneous medium?
Due to the complexity of possible shapes of the microstructure, averaging cement-based materials is far more difficult than
averaging metallic composites, for instance, with rigorously defined well-packed structure. In this paper, we imagine our
concrete piece to be made of a periodically-distributed microstructure. Based on this assumption, we provide here a rigorous
justification of the formal asymptotic expansion performed by us (in [1]) for this reaction–diffusion scenario. Note that
in [1] upscaled models are derived for a more general situation involving a locally-periodic distribution of perforations.1

∗ Corresponding author. Tel.: +31 402472082; fax: +31 402442489.

E-mail addresses: t.fatima@tue.nl (T. Fatima), a.muntean@tue.nl (A. Muntean).
1 The word ‘‘perforation’’ is seen here as a synonym for ‘‘pore’’ or ‘‘microstructure ’’.

1468-1218/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.nonrwa.2012.01.019
2 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

Fig. 1. Left: cross-section of a sewer pipe and pointing out domain of interest. Middle: periodic approximation of the periodic rectangular 3-dimensional
domain. Right: reference pore configuration.

Locally periodic geometries refer to a special case of x-dependent microstructures, where, inherently, the outer normals to
(microscopic) inner interfaces depend on both slow variable, say x, and fast variable, say y.
In the current framework, we combine two-scale convergence concepts with the periodic unfolding of ε -dependent in-
terfaces to pass to the homogenization limit (i.e. to ε → 0, where ε is a small parameter linked to the relative size of
the perforation) for the uniformly periodic case. Here, the outer normals to the inner interfaces depend only on the spatial
fast variable y. For more details on the mathematical modeling of sulfate corrosion of concrete, we refer the reader to [2,3]
(a moving-boundary approach: numerics and formal matched asymptotics), [4] (a two-scale reaction–diffusion system mod-
eling sulfate corrosion), as well as to [5], where a nonlinear Henry-law type transmission condition (modeling H2 S transfer
across all air–water interfaces present in this sulfatation problem) is analyzed. Modeling and mathematics background on
periodic homogenization can be found in e.g., [6–9], see also [10] (H-measures in the parabolic context), while a few rel-
evant (remotely resembling) worked-out examples of this averaging methodology are explained, for instance, in [11–16].
It is worth noting that, since it deals with the homogenization of a linear Henry-law setting, the paper [13] is related to
our approach. The major novelty here compared to [13] is that we now need to pass to the limit in a non-dissipative ob-
ject, namely a nonlinear ordinary differential equation (ode). The ode is describing the sulfatation reaction at the inner
water–solid interface—place where corrosion localizes. This aspect makes a rigorous averaging challenging. For instance,
compactness-type methods do not work in the case when the nonlinear ode is posed on ε -dependent surfaces. We circum-
vent this issue by ‘‘boundary unfolding’’ the ode. Thus we fix, as independent of ε , the reaction interface similarly as in [17],
and only then we pass to the limit using the monotonic structure of the production rate by reaction. Alternatively, one could
use varifolds (cf. e.g. [18]), since this seems to be the natural framework for the rigorous passage to the limit when both
the surface measure and the oscillating sequences depend on ε . However, we find the boundary unfolding technique easier
to adapt to our scenario than the varifolds. To close the circle of discussion (at least for the periodic case), we apply in [19]
periodic unfolding techniques to obtain corrector estimates for a reduced sulfate-corrosion problem.
Note that here we approach the corrosion problem deterministically. However, we have reasons to expect that the
uniform periodicity assumption can be relaxed by assuming instead a Birkhoff-type ergodicity of the microstructure shapes
and positions, and hence, the natural averaging context seems to be the one offered by random fields; see ch. 1, sect. 6
in [20], ch. 8 and 9 in [21], or [22]. But, methodologically, how big is the overlap between homogenizing deterministically
locally-periodic distributions of microstructures compared to working in the random fields context? We will treat these
and related aspects elsewhere. We expect the working techniques used in [23] to apply to our problem posed in a locally
periodic setting.
The paper is organized as follows. We start off in Section 2 (and continue in Section 3) with the analysis of the microscopic
model and derive ε -independent estimates. In Section 4, we extend the solution to the whole domain and give convergence
results obtained by passage to the limit ε → 0. Section 5 contains the main result of the paper: the set of the upscaled
two-scale Eqs. (61)–(65).

2. The microscopic model

In this section, we describe the geometry of our array of periodic microstructures and briefly indicate the most aggressive
chemical reaction mechanism typically active in sewer pipes. Finally, we list the set of microscopic equations.

2.1. Basic geometry

Fig. 1 (left) shows a cross-section of a sewer pipe hosting corrosion. We assume that the geometry of the porous medium
in question consists of a system of pores periodically distributed inside the three-dimensional cube Ω := [a, b]3 with
a, b ∈ R and b > a. The exterior boundary of Ω consists of two disjoint, sufficiently smooth parts: Γ N – the Neumann
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 3

boundary and Γ D – the Dirichlet boundary. The reference pore, say Y := [0, 1]3 , has three pairwise disjoint connected
domains Y s , Y w and Y a with smooth boundaries Γ sw and Γ wa , as shown in Fig. 1 (right). Moreover, Y := Ȳ s ∪ Ȳ w ∪ Ȳ a .
Let ε be a sufficiently small scaling factor denoting the ratio between the characteristic length of the pore Y and the
characteristic length of the domain Ω . Let χ w and χ a be the characteristic functions of the sets Y w and Y a , respectively. The
shifted set Ykw is defined by

Ykw := Y w + Σj3=0 kj ej for k := (k1 , k2 , k3 ) ∈ Z3 ,

where ej is the jth unit vector. The union of all shifted subsets of Ykw multiplied by ε (and confined within Ω ) defines the
perforated domain Ω ε , namely
Ω ε := ∪k∈Z3 {ε Ykw | ε Ykw ⊂ Ω }.
Similarly, Ω1ε , Γεsw , and Γεwa denote the union of the shifted subsets (of Ω ) Yka , Γksw , and Γkwa scaled by ε . Furthermore,
Y w∗ = ∪{Ykw , k ∈ Z3 }, Y a∗ = ∪{Yka , k ∈ Z3 }, Γ wa∗ = ∪{Γkwa , k ∈ Z3 } and Γ sw∗ = ∪{Γksw , k ∈ Z3 }.
Since usually concrete in sewer pipes is not completely dry, we decide to take into account a partially saturated porous
material.2 We assume that every pore has three distinct non-overlapping connected parts: a solid part, the water film which
separates the air layer from the solid part, and, finally, the air layer bounding the water film and filling the space of Y as
shown in Fig. 1 (right). All constituent parts of the pore connect neighboring pores to one another. All internal (water–air
and solid–water) interfaces are sufficiently smooth and do not touch each other. These geometrical restrictions are needed
not only to give a meaning to functions defined across interfaces, but also to introduce the concept of extension as given;
for instance, in [24,25]. Furthermore, there are no solid–air interfaces.
We refer to all geometrical restrictions mentioned within this subsection as Assumption (G).

2.2. Description of the chemistry

There are many variants of severe attack to concrete in sewer pipes, we focus here on the most aggressive one—the
sulfuric acid attack. The situation can be described briefly as follows. The anaerobic bacteria in the flowing waste water
release hydrogen sulfide gas (H2 S) within the air space of the pipe. These bacteria are especially active in hot environments.
From the air space inside the pipe, H2 S(g )3 enters the pores of the concrete matrix where it diffuses and then dissolves in the
pore water. The aerobic bacteria catalyze some of the H2 S into sulfuric acid H2 SO4 . H2 S molecules can move between air-filled
part and water-filled part, the water–air interfaces [26]. We model this microscopic interfacial transfer via Henry’s law [27],
(see the boundary conditions at Γεwa in (3) and (4)). We refer reader for some examples where microscopic interfacial transfer
to be modeled by Henry’s law to [13,28,29]. A remotely resembling scenario pointing at phase-change in low porosity porous
media is reported in [30]. H2 SO4 being an aggressive acid reacts with the solid matrix4 at the solid–water interface, which
is made up of cement, sand, and aggregate, and produces gypsum (i.e. CaSO4 · 2H2 O). Here we restrict our attention to a
minimal set of chemical reactions mechanisms as suggested in [2], namely.

10H+ + SO− 4 + org. matter −→ H2 S(aq) + 4H2 O + oxid. matter
2


H2 S(aq) + 2O2 −→ 2H+ + SO− 2

4 (1)
 H2 S (aq)
H 2 S (g )
4 + CaCO3 −→ CaSO4 · 2H2 O + HCO3 .

2H2 O + H+ + SO− 2 −


We assume that reactions (1) do not interfere with the mechanics of the solid part of the pores. This is a rather strong
assumption since it is known that (1) can actually produce local ruptures of the solid matrix [31]. Since we do not take into
account all chemical species intervening in the corrosion process, we focus on the most relevant ones. The model equations
will not be mass-conservative. Also, the moisture model we take into account is rather rudimentary. More complex models
describing the behavior of water in concrete could be considered instead; see e.g. [32,33]. For more details on the involved
cement chemistry and connections to acid corrosion, we refer the reader to [34] (for a nice enumeration of the involved
physicochemical mechanisms), [31] (standard textbook on cement chemistry), as well as to [35–38] and references cited
therein. For a mathematical approach of a similar theme related to the conservation and restoration of historical monuments,
we refer to the work by Natalini and co-workers (cf. e.g. [39]).

2.3. Setting of the equations

The unknowns of the microscopic model are uε1 , uε2 , uε3 , uε4 , uε5 denoting the mass concentration of H2 SO4 (aq), H2 S(aq),
H2 S(g ), moisture and gypsum, respectively. Here uε10 , uε20 , uε30 , uε40 and uε50 represent the initial data for H2 SO4 (aq), H2 S(aq),

2 The solid, water and air parts correspond to Y s , Y w and Y a , respectively.

3 H S(g ) and H S(aq) refer to gaseous, and aqueous H S, respectively.
2 2 2
4 The solid matrix is assumed here to consist of precipitated CaCO only. This assumption can be removed in the favor of a more complex cement
3
chemistry.
4 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

H2 S(g ), moisture and gypsum, respectively.

uD3 : ΓD × (0, T ) −→ R+ —exterior concentration (Dirichlet data) of H2 S(g )
dε1 : Ω −→ R the diffusion coefficient of H2 SO4 (aq)
dε2 : Ω −→ R the diffusion coefficient of H2 S(aq)
dε3 : Ω −→ R the diffusion coefficient of H2 S(g )
dε4 : Ω −→ R the diffusion coefficient of moisture
aε : Γεwa −→ R+ the absorption factor of H2 S air to water
bε : Γεwa −→ R+ the desorption factor of H2 S air to water
kεi : Ω −→ R+ , i ∈ {1, 2} bulk reaction ‘‘constant’’
kε3 : Γεwa −→ R+ surface reaction ‘‘constant ’’.
All functions defined in Ω , Γεwa and Γεsw are taken to be Y -periodic, i.e. dεi (x) = di ( εx ), kεi (x) = ki ( εx ), aε (x) = a( εx ), bε (x) =
b( εx ) are defined on Y w∗ , Y a∗ , Γ wa∗ , and on Γ sw∗ , respectively.
We consider the following system of mass-balance equations defined at the pore level. The mass-balance equation for
H2 SO4 is

∂t uε1 + div(−dε1 ∇ uε1 ) = −kε1 uε1 + kε2 uε2 , x ∈ Ωε, t ∈ (0, T )

ε ε ε
−n · d1 ∇ u1 = 0, x ∈ Γε ,
wa
t ∈ (0, T )
−nε · dε1 ∇ uε1 = 0, x ∈ Γ N ∩ ∂Ωε, t ∈ (0, T )
(2)
−n · d1 ∇ u1 = εη (u1 , uε5 ),
ε ε ε ε ε
x ∈ Γεsw t ∈ (0, T )
uε1 = 0, x ∈ Γ D ∩ ∂Ωε, t ∈ (0, T )
ε
u1 (x, 0) = u10 (x), x ∈ Ω ε , t = 0.
The mass-balance equation for H2 S(aq) is given by

∂t uε2 + div(−dε2 ∇ uε2 ) = kε1 uε1 − kε2 uε2 , x ∈ Ωε, t ∈ (0, T ),

ε ε ε ε ε ε ε
−n · d2 ∇ u2 = ε(a (x)u3 − b (x)u2 ), x ∈ Γεwa , t ∈ (0, T )
−nε · dε2 ∇ uε2 = 0, x ∈ Γεsw , t ∈ (0, T )
(3)
−nε · dε2 ∇ uε2 = 0, x ∈ Γ N ∩ ∂Ωε, t ∈ (0, T )
uε2 = 0, x ∈ Γ D ∩ ∂Ωε, t ∈ (0, T )
ε ε
u2 (x, 0) = u20 (x), x ∈ Ω , t = 0.
The mass-balance equation for H2 S(g ) reads

∂t uε3 + div(−dε3 ∇ uε3 ) = 0, x ∈ Ω1ε , t ∈ (0, T )

ε ε ε ε
−n · d3 ∇ u3 = 0, x∈Γ N
∩ ∂ Ω1 , t ∈ (0, T )
ε ε
u3 (x, t ) = (x, t ), x ∈ Γ ∩ ∂ Ω1 , t ∈ (0, T )
uD3 D (4)
−n · d3 ∇ u3 = −ε(aε (x)uε3 − bε (x)uε2 ), x ∈ Γεwa ,
ε ε ε
t ∈ (0, T )
uε3 (x, 0) = u30 (x), x ∈ Ω1ε , t = 0.
The mass-balance equation for moisture follows

∂t uε4 + div(−dε4 ∇ uε4 ) = kε1 uε1 , x ∈ Ωε, t ∈ (0, T )

ε ε ε ε
−n · d4 ∇ u4 = 0, x ∈ ∂Ω , t ∈ (0, T ) (5)
uε4 (x, 0) = u40 (x), x ∈ Ω ε , t = 0.
The mass-balance equation for the gypsum produced at the water–solid interface is

∂t uε5 = ηε (uε1 , uε5 ), x ∈ Γεsw , t ∈ (0, T )

(6)
uε5 (x, 0) = u50 (x), x ∈ Γεsw .

3. Weak formulation and basic results

We begin this section with a list of notations and function spaces. Then we indicate our working assumptions and give
the weak formulation of the microscopic problem; we bring reader’s attention to the well-posedness of the system (2)–(6).
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 5

3.1. Notations and function spaces

ϕ + and ϕ − will point out the positive and respectively the negative part of the function ϕ . We denote by C#∞ (Y ), H#1 (Y ),
and H#1 (Y )/R, the space of infinitely differentiable functions in Rn that are Y -periodic, the completion of C#∞ (Y ) with respect
to H 1 -norm, and the respective quotient space. Furthermore, HΓ1 D (Ω ) := {u ∈ H 1 (Ω )|u = 0 on Γ D }. The Sobolev space
H β (Ω ) as a completion of C ∞ (Ω ) is a Hilbert space equipped with the norm
 21
|ϕ(x) − ϕ(y)|2
 
∥ϕ∥H β (Ω ) := ∥ϕ∥H [β] (Ω ) + n+2(β−[β])
dxdy
Ω Ω |x − y|

and (cf. Theorem 7.57 in [40]) the embedding H β (Ω ) ↩→ L2 (Ω ) is continuous. Since we deal here with an evolution problem,
we use standard Bochner spaces like L2 (0, T ; H 1 (Ω )), L2 (0, T ; L2 (Ω )), L2 (0, T ; HΓ1 D (Ω )), and L2 ((0, T ) × Ω ; H#1 (Y )/R). In
the analysis of the microscopic model, we employ frequently the following trace inequality for ε -dependent hypersurfaces
Γεwa . For ϕε ∈ H 1 (Ω ε ), there exists a constant C , which is independent of ε , such that

ε∥ϕε ∥2L2 (Γε ) ≤ C (∥ϕε ∥2L2 (Ω ε ) + ε 2 ∥∇ϕε ∥2L2 (Ω ε ) ), (7)

where C is a constant independent of ε . The proof of (7) is given in Lemma 3 of [28]. For a function ϕ ε ∈ H β (Ω ε ) with
β ∈ ( 12 , 1), the inequality (7) refines into

|ϕ ε (x) − ϕ ε (y)|2
   
2β
ε∥ϕ ∥ 2
ε L2 (Γε ) ≤C |ϕε | dx + ε
2
dxdy . (8)
Ωε Ωε Ωε |x − y|n+2β
For proof of (8), see [17].
The quantities like Q , a, b, kj , j ∈ {1, 2, 3} with superscript ∞ are the maximum of the Q (uε5 ), aε , bε , kεj , j ∈ {1, 2, 3}
while a, b, kj , j ∈ {1, 2} denote the minimum of the respective quantities.

3.2. Assumptions on the data and parameters

We consider the following restriction on the data and parameters:

(A1) di ∈ L∞ (Y )3×3 , (di (x)ξ , ξ ) ≥ di0 |ξ |2 for di0 > 0 and every ξ ∈ R3 , y ∈ Y , i ∈ {1, 2, 3, 4}.
(A2) η is measurable w.r.t. t and x and η(α, β) = k3 R(α)Q (β), where R is a sub-linear and locally Lipschitz function with
Lipschitz constant cR , while Q is bounded and monotonically increasing such that
positive, if α ≥ 0, positive, if β < βmax ,
 
R(α) = Q (β) =
0, otherwise 0, otherwise,

where βmax > 0 represents the maximum amount of gypsum that can (locally) be produced.
sw
(A3) ui0 ∈ H 1 (Ω ) ∩ L∞
+ (Ω ), i ∈ {1, 4}, uj0 ∈ H (Ω ) ∩ L+ (Ω ), j ∈ {2, 3}, u50 ∈ L+ (Γ ).
2 ∞ ∞

(A4) The mass transfer functions at the boundary a, b ∈ L (Γ ), a, b > 0 are assumed to satisfy b(y)M2 = a(y)M3 for a.e.
∞ wa
k∞
a.e. in Ω ε and M5 ≥ k∞ sw
3 cR Q M1 a.e. on Γε . The constants M1 , M2 , M3 and
M2 k1
y ∈ Γ wa . Furthermore, 1
k2
= M1
= k∞
∞
2
M5 mentioned here are defined in (9).
(A5) uD3 ∈ H 2 (0, T ; H 1 (Ω1ε )) ∩ L∞ ε
+ ((0, T ) × Ω1 ).
sw
+ (Γ ) and kj ∈ L+ (Ȳ ) for any j ∈ {1, 2}.
(A6) k3 ∈ L∞ ∞

We also define the following constants

Mj := ∥uj0 ∥L∞ (Ω ) j ∈ {1, 4},
M2 := max{∥u20 ∥L∞ (Ω ) , ∥u30 ∥L∞ (Ω ) },
M3 := max{M2 , ∥u30 ∥L∞ (Ω ) , ∥uD3 ∥L∞ (Γ D ) },
M5 := max{∥u50 ∥L∞ (Γ sw ) , βmax }. (9)
They will play later on the role of essential supremum bounds on the active concentrations.

Remark 1. Assumption (A4) is a technical one. This can be completely removed if one deals e.g. with classical solutions
to our PDE-ode system, or it can be replaced with different ones if other auxiliary test functions are chosen to prove the
positivity of concentrations in the presence of Henry-type exchange production terms. This restriction is needed to ensure
the basic bounds mentioned in Lemma 4. In spite of the fact that (A4) artificially restricts the choice of k1 and k2 (see the link
between (9) and (A4)), we do not aim here to refine any further (A4) and stay focus on investigating the homogenization
limit.
6 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

3.3. Weak formulation of the microscopic model. Basic results

Definition 2. Assume (A1)–(A6). We call the vector uε = (uε1 , uε2 , uε3 , uε4 , uε5 ), a weak solution to (2)–(6) if uεj ∈ L2 (0, T ;
H 1 (Ω ε )), ∂t uεj ∈ L2 ((0, T ) × Ω ε ), j ∈ {1, 2, 4}, uε3 ∈ uD3 + L2 (0, T ; HΓ1 D (Ω1ε )), ∂t uε3 ∈ ∂t uD3 + L2 (0, T ; L2 (Ω1ε )), uε5 ∈
H 1 (0, T ; L2 (Γεsw )) such that the following identities hold
 T 
 ε
 T 
∂t u1 ϕ1 + dε1 ∇ uε1 ∇ϕ1 + kε1 uε1 ϕ1 − kε2 uε2 ϕ1 dxdτ = −ε ηε ϕ1 dσx dτ ,

(10)
0 Ωε 0 Γεsw
 T 
 ε
 T 
∂t u2 ϕ2 + dε2 ∇ uε2 ∇ϕ2 + kε1 uε1 ϕ2 − kε2 uε2 ϕ2 dxdτ = ε (aε uε3 − bε uε2 )ϕ2 dσx dτ

(11)
0 Ωε 0 Γεwa
 T 
 ε
 T 
∂t u3 ϕ3 + dε3 ∇ uε3 ∇ϕ3 dxdτ = −ε (aε uε3 − bε uε2 )ϕ3 dσx dτ ,

(12)
0 Ω1ε 0 Γεwa
 T 
 ε
 T 
∂t u4 ϕ4 + dε4 ∇ uε4 ∇ϕ4 dxdτ = kε1 uε1 ϕ4 dxdτ ,

(13)
0 Ωε 0 Ωε
 T   T 
ε
∂t u5 ϕ5 dσx dτ = ηε ϕ1 dσx dτ , (14)
0 Γεsw 0 Γεsw

for all ϕj ∈ L2 (0, T ; H 1 (Ω ε )), j ∈ {1, 2, 4}, ϕ3 ∈ L2 (0, T ; HΓ1 D (Ω1ε )) and ϕ5 ∈ L2 ((0, T ) × Γεsw ).

Lemma 3 (Energy Estimates). Assume (A1)–(A6) , then the weak solution to the microscopic problem (2)–(6) satisfies the
following apriori estimates

∥uεj ∥L2 (0,T ;L2 (Ω ε )) + ∥∇ uεj ∥L2 (0,T ;L2 (Ω ε )) ≤ C ,

∥uε3 ∥L2 (0,T ;L2 (Ω ε )) + ∥∇ uε3 ∥L2 (0,T ;L2 (Ω ε )) ≤ C ,
1 1
√ ε √
ε∥u5 ∥L∞ ((0,T )×Γεsw ) + ε∥∂t uε5 ∥L2 ((0,T )×Γεsw ) ≤ C for j ∈ {1, 2, 4}.

Proof. We test (10) with ϕ1 = uε1 to get

 t  t  t  t
∂t |uε1 |2 dxdτ + 2d10 |∇ uε1 |2 dxdτ ≤ k∞
2 uε1 uε2 dxdτ − ε ηε uε1 dσx dτ ,
sw
0 Ωε 0 Ωε 0 Ωε 0 Γε
 t  t
≤C (|uε1 |2 + |uε2 |2 )dxdτ + ε C |uε1 |2 dσx dτ , (15)
0 Ωε 0 Γεsw

ε
where k∞2 = sup(0,T )×Ω ε |k2 |. Here we have used the structure of η and the properties of R and Q . After applying the trace
inequality to the last term on r.h.s of (15), we get
 t  t  t
ε 2 ε 2
∂t |u1 | dxdτ + (2d10 − C ε ) 2
|∇ u1 | dxdτ ≤ C (|uε1 |2 + |uε2 |2 )dxdτ . (16)
0 Ωε 0 Ωε 0 Ωε

Taking ϕ2 = uε2 in (11), we get

 t  t  t  t
∂t |uε2 |2 dxdτ + 2d20 |∇ uε2 |2 dxdτ ≤ 2k∞
1 (|uε1 |2 + |uε2 |2 )dxdτ + εC uε3 uε2 dσx dτ
0 Ωε 0 Ωε 0 Ωε 0 Γεwa
 t
+ εC |uε2 |2 dσx dτ .
0 Γεwa

Application of (7) leads to

 t  t  t
ε 2 ε 2
∂t |u2 | dxdτ + (2d20 − C ε ) 2
|∇ u2 | dxdτ ≤ C (|uε1 |2 + |uε2 |2 )dxdτ
0 Ωε 0 Ωε 0 Ωε
 t
+C (|uε3 |2 + ε 2 |∇ uε2 |2 )dxdt . (17)
0 Ω1ε
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 7

We choose ϕ3 = uε3 as a test function in (12) to calculate

 t  t  t
∂t |uε3 |2 dxdτ + (2d30 − C ε 2 ) |∇ uε3 |2 dxdτ ≤ C |uε3 |2 dxdτ
0 Ω1ε 0 Ω1ε 0 Ω1ε
 t
+C (|uε2 |2 + ε 2 |∇ uε2 |2 )dxdτ . (18)
0 Ωε

Setting ϕ4 = uε4 in (13), we are led to

 t  t
∂t |uε4 |2 dxdτ +2d40 |∇ uε4 |2 dxdτ ≤ C (|uε1 |2 + |uε4 |2 )dxdτ .
t 
0 Ωε
(19)
0 Ωε 0 Ωε

Putting together (16)–(19), we obtain

 t  t
∂t |uε1 |2 + |uε2 |2 + |uε4 |2 dxdτ + ∂t |uε3 |2 dxdτ
 
0 Ωε 0 Ω1ε
 t  t
+ (2d20 − C ε 2 ) |∇ uε2 |2 dxdτ + (2d10 − C ε 2 ) |∇ uε1 |2 dxdτ
0 Ωε 0 Ωε
 t  t
ε 2
× (2d30 − C ε )
2
|∇ u3 | dxdτ + 2d40 |∇ uε4 |2 dxdτ
0 Ω1ε 0 Ωε
 t  t
≤C (|uε1 |2 + |uε2 |2 + |uε4 |2 )dxdτ + C |uε3 |2 dxdτ . (20)
0 Ωε 0 Ωε
1

Choosing ε small enough and applying Gronwall’s inequality, we have for j ∈ {1, 2, 4}

∥uεj ∥L∞ (0,T ;L2 (Ω ε )) ≤ C , ∥uε3 ∥L∞ (0,T ;L2 (Ω ε )) ≤ C , (21)

1

∥∇ uεj ∥L2 (0,T ;L2 (Ω ε )) ≤ C , ∥∇ uε3 ∥L2 (0,T ;L2 (Ω ε )) ≤ C . (22)

1

We set as a test function ϕ5 = uε5 in (14)

ε
 t  t
∂t |uε5 |2 dσx dτ ≤ ε C uε1 uε5 dσx dτ ,
2 0 Γεsw 0 Γεsw
 t  t
ε ∂t |uε5 |2 dσx dτ ≤ εC (|uε1 |2 + |uε5 |2 )dσx dτ .
0 Γεsw 0 Γεsw

Applying Gronwall’s inequality together with (7), we obtain

  t 
ε |uε5 (t )|2 dσx ≤ C (|uε1 |2 + ε 2 |∇ uε1 |2 )dxdτ + C |uε5 (0)|2 dσx .
Γεsw 0 Ωε Γ1ε

Hence by (21) and (22), ε∥uε5 ∥L∞ ((0,T )×Γεsw ) ≤ C . We take ϕ5 = ∂t uε5 in (14) as a test function
 t  t
ε 2
ε |∂t u5 | dσx dτ ≤ εC uε1 ∂t uε5 dσx dτ ,
0 Γεsw 0 Γεsw
 t  t
ε(1 − C δ) |∂t uε5 |2 dσx dτ ≤ C (|uε1 |2 + ε 2 |∇ uε1 |2 )dxdτ .
0 Γεsw 0 Ωε

For convenient δ and by (21) and (22), we have

ε∥∂t uε5 ∥L2 ((0,T )×Γεsw ) ≤ C .  (23)

Lemma 4 (Positivity and L∞ -Bounds on Concentrations). Assume (A1)–(A6) , and let t ∈ [0, T ] be arbitrarily chosen. Then the
following estimates hold.

(i) uεi (t ) ≥ 0, i ∈ {1, 2, 4} a.e. in Ω ε , uε3 (t ) ≥ 0 a.e. Ω1ε and uε5 (t ) ≥ 0 a.e. on Γεws .
(ii) There exist constants such that uεi (t ) ≤ Mi , i ∈ {1, 2}, uε4 (t ) ≤ (t + 1)M4 a.e. in Ω ε , uε3 (t ) ≤ M3 a.e. in Ω1ε and uε5 (t ) ≤ M5
a.e. on Γεws .
8 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

Proof. We omit to show the proof details for (i). We only mention that Assumption (A4) (on the Henry term) is relevant at
this point. Let us focus now on getting the L∞ -bounds on concentrations.
(ii). We consider the test function
(ϕ1 , ϕ2 , ϕ3 , ϕ4 ) = ((uε1 − M1 )+ , (uε2 − M2 )+ , (uε3 − M3 )+ , (uε4 − (t + 1)M4 )+ ),
where the constants Mi are defined in (9). By (10), we obtain:
 t  t
1
∂t |(uε1 − M1 )+ |2 dxdt + d10 |∇(uε1 − M1 )+ |2 dxdt
2 0 Ωε 0 Ω ε
 t  t  t
ε
≤ −(k1 M1 − k2 M2 )
∞
(u1 − M1 ) dxdt + C
+
|(uε2 − M2 )+ |2 dxdt − ε ηε (uε1 − M1 )+ dσx dt .
0 Ωε 0 Ωε 0 Γεsw

Using (A4), we get the estimate

 t  t
∂t |(uε1 − M1 )+ |2 dxdt ≤ C (|(uε1 − M1 )+ |2 + |(uε2 − M2 )+ |2 )dxdt . (24)
0 Ωε 0 Ωε

(11) in combination with (A4) gives that

 t  t
ε
∂t |(u2 − M2 ) | dxdt + (2d20 − C ε )
+ 2 2
|∇(uε2 − M2 )+ |2 dxdt
0 Ωε 0 Ωε
 t  t
ε ε
≤C (|(u1 − M1 ) | + |(u2 − M2 ) | )dxdt +
+ 2 + 2
(|(uε3 − M3 )+ |2 + ε 2 |∇(uε3 − M3 )+ |2 )dxdt . (25)
0 Ωε 0 Ω1ε

By (12) and (A4), we obtain

 t  t
ε
∂t |(u3 − M3 ) | dxdt + (2d30 − C ε )
+ 2 2
|∇(uε3 − M3 )+ |2 dxdt
0 Ω1ε 0 Ω1ε
 t  t
≤C |(uε3 − M3 )+ |2 dxdt + C (|uε2 − M2 )+ |2 +|∇(uε2 − M2 )+ |2 dxdt . (26)
0 Ωε
1
0 Ωε

By (13), we get
 t  t
∂t |(uε4 − (t + 1)M4 )+ |2 dxdt + 2d40 |∇(uε4 − (t + 1)M4 )+ |2 dxdt
0 Ωε 0 Ωε
 t
≤C (|(uε1 − M1 )+ |2 + |(uε4 − (t + 1)M4 )+ |2 )dxdt . (27)
0 Ωε

Adding up (24)–(27), we get

 t
∂t |(uε1 − M1 )+ |2 + |(uε2 − M2 )+ |2 + |(uε4 − (t + 1)M4 )+ |2
 
0 Ωε
 t  t
ε
+ ∂t |(u3 − M3 ) | dxdt + (2d20 − C ε )
+ 2 2
|∇(uε2 − M2 )+ |2 dxdt
0 Ω1ε 0 Ωε
 t  t
+ (2d30 − C ε 2 ) |∇(uε3 − M3 )+ |2 dxdt ≤ C |(uε3 − M3 )+ |2 dxdt
0 Ω1ε 0 Ω1ε
 t
+C (|(uε1 − M1 )+ |2 + |(uε2 − M2 )+ |2 + |(uε4 − (t + 1)M4 )+ |2 )dxdt .
0 Ωε

Choosing ε small enough, then Gronwall’s inequality yields the following estimate uεj (t ) ≤ Mj , j ∈ {1, 2} a.e. in Ω ε , uε3 (t ) ≤
M3 , a.e. in Ω1ε uε4 ≤ (t + 1)M4 a.e. in Ω ε for all t ∈ (0, T ). Let us now point out the fact that the bound M1 for uε1 also holds
on Γεsw ; see Claim 5 for this basic fact. 

Claim 5. If z ∈ H 1 (Ω ) ∩ L∞ (Ω ), then z ∈ L∞ (∂ Ω ).
Proof of Claim. Let z ∈ H 1 (Ω ) ∩ L∞ (Ω ). Since the set of the restrictions to Ω of functions C0∞ (Rn ) is dense in H 1 (Ω ),
we consider a sequence of smooth functions { fn } ⊂ C0∞ (Ω ) such that fn → z in H 1 (Ω ) and ∥fn ∥L∞ (Ω ) ≤ ∥z ∥L∞ (Ω ) . Trace
theorem gives fn → z in L2 (∂ Ω ). So, there exists a subsequence { fni } ⊂ { fn } converging pointwise, i.e., fni (x) → z (x) for a.e
x ∈ ∂ Ω . Therefore, ∥fni (x)∥ ≤ ∥z ∥L∞ (Ω ) and thus, ∥z ∥L∞ (∂ Ω ) ≤ ∥z ∥L∞ (Ω ) .
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 9

By Lemma 3 and Claim 5, we see that uε1 is bounded on the interface Γεsw . Now testing (14) with (uε5 − (t + 1)M5 )+ and
using the properties of R, Q , we derive
 
 T   T 
1 ε ε
∂t |(u5 − (t + 1)M5 ) | + M5 (u5 − (t + 1)M5 )
+ 2 +
dσx dτ ≤ C M1 (uε5 − (t + 1)M5 )+ dσx dτ ,
0 Γεsw 2 0 Γεsw
 T   T 
∂t |(uε5 − (t + 1)M5 )+ |2 dσx dτ ≤ −(M5 − CM1 ) (uε5 − (t + 1)M5 )+ dσx dτ ,
0 Γεsw 0 Γεsw
ε sw
3 cR Q . Using (A4) and Gronwall’s inequality we get u5 ≤ M5 a.e. in (0, T ) × Γε .
where C := k∞ ∞


Proposition 6 (Uniqueness). Assume (A1)–(A6) . Then there exists at most one weak solution in the sense of Definition 2.
j,ε j,ε j,ε j,ε j,ε
Proof. We assume that uj,ε = (u1 , u2 , u3 , u4 , u5 ), j ∈ {1, 2} are two distinct weak solutions in the sense of Definition 2.
1,ε
We set uεi := ui − u2i ,ε for all i ∈ {1, 2, 3, 4}. First, we deal with (14). We obtain
 t  t
(∂t u15,ε − ∂t u25,ε )ϕ5 dσx dτ =
 1,ε 1,ε 1,ε
η (u1 , u5 ) − η2,ε (u21,ε , u25,ε ) ϕ5 dσx dτ .

(28)
0 Γεsw 0 Γεsw
1,ε 2,ε
Testing (28) with u5 − u5 and making use of structure of η
 t  t
∂t |u15,ε − u25,ε |2 dσx dτ ≤ C
 1,ε
|u5 − u25,ε |2 + |u11,ε − u21,ε |2 dσx dτ .

0 Γεsw 0 Γεsw

Gronwall’s inequality implies

  t
|uε5 (t )|2 dσx ≤ C |uε1 |2 dσx dτ for a.e. t ∈ (0, T ). (29)
Γεsw 0 Γεsw

We calculate
 t  t  t  t
1
∂t |uε1 |2 dxdτ + d10 |∇ uε1 |2 dxdτ ≤ −k1 |uε1 |2 dxdτ + k∞
2 uε1 uε2 dxdτ
2 0 Ωε 0 Ωε 0 Ωε 0 Ωε
  t
−ε (η1,ε − η2,ε )uε1 dσx dτ ,
sw
0 Γε

where k1 := inf(0,T )×Ω ε |kε1 |. We can write

 t  t  t
∂t |uε1 |2 dxdτ + 2d10 |∇ uε1 |2 dxdτ + 2k1 |uε1 |2 dxdτ
0 Ωε 0 Ωε 0 Ωε
 t  t  t
≤C (|uε1 |2 + |uε2 |2 )dxdτ + ε C |uε5 |2 dσx dτ + ε C |uε1 |2 dσx dτ . (30)
0 Ωε 0 Γεsw 0 Γεsw

Now, inserting (29) in (30) yields

 t  t  t
∂t |uε1 |2 dxdt + 2d10 |∇ uε1 |2 dxdτ + 2k1 |uε1 |2 dxdτ
0 Ωε 0 Ωε 0 Ωε
 t  t  t τ 
ε 2 ε 2 ε 2
≤C (|u1 | + |u2 | )dxdτ + C ε |u1 | dσx dτ + ε C |uε1 |2 dσx dsdτ . (31)
0 Ωε 0 Γεsw 0 0 Γεsw

We estimate the last two terms in (31) to obtain the following inequality
 t  t  t
∂t |uε1 |2 dxdt + (2d10 − C ε 2 ) |∇ uε1 |2 dxdτ + 2k1 |uε1 |2 dxdτ
0 Ωε 0 Ωε 0 Ωε
 t  t τ 
≤C (|uε1 |2 + |uε2 |2 ) + C (|uε1 |2 + ε 2 |∇ uε1 |2 )dxdsdτ . (32)
0 Ωε 0 0 Ωε
Following the same line of arguments as before, we obtain from (11) that
 t  t
∂t |uε2 |2 dxdt + (2d20 − C ε 2 ) |∇ uε2 |2 dxdτ
0 Ωε 0 Ωε
 t  t
≤C (|uε1 |2 + uε2 |2 )dxdτ + C (|uε3 |2 +ε 2 |∇ uε3 |2 )dxdτ ,
0 Ωε 0 Ω1ε
10 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

while from (12), we deduce

 t  t
∂t |uε3 |2 dxdt + (2d30 − C ε 2 ) |∇ uε3 |2 dxdτ
0 Ω1ε 0 Ω1ε
 t  t
≤C |uε3 |2 dxdτ + C (|uε2 |2 + ε 2 |∇ uε2 |2 )dxdτ . (33)
0 Ω1ε 0 Ωε

Proceeding similarly, (13) yields

 t  t  t
∂t |uε4 |2 dxdτ + 2d40 |∇ uε4 |2 dxdτ ≤ C (|uε1 |2 + |uε4 |2 )dxdτ . (34)
0 Ωε 0 Ωε 0 Ωε

Putting together (32)–(34) and re-arranging the terms, we get

 t  t
ε 2 ε 2 ε 2
∂t |u1 | + |u2 | + |u4 | dxdτ + (2d10 − C ε ) |∇ uε1 |2 dxdτ 2
 
0 Ωε 0 Ωε
 t  t
+ (2d20 − C ε 2 ) |∇ uε2 |2 dxdτ + ∂t |uε3 |2 + (2d30 − C ε 2 )|∇ uε3 |2 dxdτ
 
0 Ωε 0 Ω1ε
 t  t
+ 2d40 |∇ uε4 |2 dxdτ + 2k1 |uε1 |2 dxdτ
0 Ωε 0 Ωε
 t  t  t τ 
≤C |uε3 |2 dxdτ + C (|uε1 |2 + uε2 |2 +|uε4 |2 )dxdτ + C (|uε1 |2 + ε 2 |∇ uε1 |2 )dxdτ ds.
0 Ω1ε 0 Ωε 0 0 Ωε

Let us choose ε such that the above inequality does not violate. Applying Gronwall’s inequality with k1 > 0, taking supre-
mum along t ∈ [0, T ], we obtain the following estimate
  T  
|uε1 |2 + |uε2 |2 + |uε4 |2 dx + C |∇ uε1 |2 dxdτ + |uε3 |2 dx ≤ 0.
 
Ωε 0 Ωε Ωε
1

1,ε 2,ε
Hence, we conclude that ui = , i ∈ {1, 2, 4} a.a. t ∈ (0, T ) in Ω ε and u13,ε = u23,ε a.a. t ∈ (0, T ) in Ω1ε . Consequently,
ui
1,ε 1,ε
(29) gives u5 = u5 a.a. t ∈ (0, T ) on Γεsw . 

Theorem 7 (Global Existence). Assume (A1)–(A6) . Then there exists at least a global-in-time weak solution in the sense
of Definition 2.

Proof. The proof is based on the Galerkin argument. The proof is rather standard, and since here we wish to focus on the
passage to the limit ε → 0, we omit it. Roughly speaking, we now have enough compactness to pass to the limit in the
Galerkin-projected PDEs, while in the ode we pass to the limit using the monotonicity of η. 

Lemma 8 (Additional a Priori Estimates). Assume (A1)–(A6) . The following ε -independent bounds hold:

∥∇∂t uε2 ∥L2 (0,T ;L2 (Ω ε )) + ∥∇∂t uε3 ∥L2 (0,T ;L2 (Ω ε )) ≤ C (35)
1
ε ε
∥∂t uj ∥L2 (0,T ;L2 (Ω ε )) + ∥∂t u3 ∥L2 (0,T ;L2 (Ω ε )) ≤ C , (36)
1

for j ∈ {1, 2, 4} and C a generic constant independent of ε .

Proof. Now, we focus on obtaining ε -independent estimates on the time derivative of the concentrations. First, we choose
ϕ1 = ∂t uε1 . We get
 t  t  t  t
∂t uε1 ∂t uε1 dxdτ + dε1 ∇ uε1 ∇∂t uε1 dxdτ = − kε1 uε1 ∂t uε1 dxdτ + kε2 uε2 ∂t uε1 dxdτ
0 Ω ε 0 Ω ε 0 Ω ε 0 Ω ε
 t  t
ε ε
−ε η ∂t u1 dσx dτ (1 − C δ) |∂t uε1 |2 dxdτ
0 Γεsw 0 Ωε
 
+ d10 |∇ uε1 |2 dx ≤ d10 |∇ uε1 (0)|2 dx
Ωε Ωε
 t  t
+C (|uε1 |2 + |uε2 |2 )dxdτ − ε ηε ∂t uε1 dσx dτ . (37)
0 Ωε 0 Γεsw
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 11

The last term in (37) can be estimated as follows:

  t    t 
ε ηε ∂t uε1 dσx dτ  = ε kε3 R(uε1 )Q (uε5 )∂t uε1 dσx dτ 
   
Γεsw Γ sw

0 0
   ε  ε 
 t u1
ε ε
= ε k ∂ R(α)dα Q (u5 ) dσx dτ

 0 Γεsw 3 t 0
 t  ε  
u1 
ε ε ε
−ε k3 R(α)dα∂uε5 Q (u5 ) ∂t u5 dσx dτ  ,

0 Γεsw 0 
   t
≤ εC |uε1 (t )|2 dσx + ε C |uε1 (0)|2 dσx + ε C |uε1 |2 |∂t uε5 |dσx dτ .
Γεsw Γεsw 0 Γεsw

By Claim 5, uε1 is bounded on Γεsw

 t    
ε  ηε ∂t uε1 dσx dτ  ≤ C + C sup[0,T ] (|uε1 (t )|2 + ε 2 |∇ uε1 (t )|2 )dx + C (|uε1 (0)|2 + ε 2 |∇ uε1 (0)|2 )dx
 
0 Γεsw Ωε Ωε
  t
ε 2 ε 2
+ C sup[0,T ] (|u1 | + ε |∇ u1 | )dx + ε C
2
|∂t uε5 |2 dσx dτ . (38)
Ωε 0 Γεsw

Combine (38) with (37) and then choose δ > 0 conveniently. Using Lemma 3, and taking supremum over the time variable,
we get
 T  
|∂t uε1 |2 dxdτ + d10 sup[0,T ] |∇ uε1 |2 dx ≤ C .
0 Ωε Ωε

Testing (11) with ϕ2 = ∂t uε2 gives

 t    t
d20 d20 C
|∂t uε2 |2 dxdτ + |∇ uε2 |2 dx ≤ |∇ uε2 (0)|2 dx + (|uε |2 + |uε1 |2 + |∇ uε2 |2 )dxdτ
0 Ωε 2 Ωε 2 Ωε δ 0 Ωε 2
 t  t
C
+ Cδ |∂t uε2 |2 dxdτ + (|uε |2 + |∇ uε3 |2 )dxdτ
0 Ωε δ 0 Ω1ε 3
 t
+ Cδ (|∂t uε2 |2 + ε 2 |∇∂t uε2 |2 )dxdτ .
0 Ωε

Choosing δ > 0 small enough, we are led to

 t   t 
ε 2 ε 2
|∂t u2 | dxdτ ≤ C 1 + ε 2
|∇∂t u2 | dxdτ . (39)
0 Ωε 0 Ωε

We consider now the Dirichlet data uD3 to be extended in whole of Ω . Testing now (12) with ϕ3 = ∂t (uε3 − uD3 ) leads to
 t 
d30
|∂t uε3 |2 dxdτ + |∇ uε3 |2 dxdτ
0 Ωε 2 Ωε
1 1
  t  t
d30 1 d30
≤ |∇ uε3 (0)|2 dxdτ + (|∂t uε3 |2 + |∂t uD3 |2 )dxdτ + (|∇ uε3 |2 + |∇∂t uD3 |2 )dxdτ
2 Ω1ε 2 0 Ω1ε 2 0 Ω1ε
 t  t
 ε2
+ εC |u3 | + |uε2 |2 dσx dτ + ε C |∂t uε3 |2 + |∂t uD3 |2 dσx dτ .
  
0 Γεwa 0 Γεwa

Using (7), (A6), (21) and (22), we end up with

 t   t 
ε 2 ε 2
|∂t u3 | dxdτ ≤ C 1 + ε 2
|∇∂t u3 | dxdτ . (40)
0 Ω1ε 0 Ω1ε

From (13), we get

 t
|∂t uε4 |2 dxdτ ≤ C . (41)
0 Ωε
12 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

In order to estimate (39) and (40), we proceed first with differentiating the PDE in (3) with respect to time and then testing
the result with ∂t uε2 . Consequently, we derive
  t   t
1 ε 2 ε 2 1 ε
|∂t u2 | dx + d20 |∇∂t u2 | dx ≤ |∂t u2 (0)| dx + C 2
(|∂t uε1 |2 + |∂t uε2 |2 )dxdτ
2 Ωε 0 Ωε 2 Ωε 0 Ωε
 t
+ εC (|∂t uε2 |2 + |∂t uε3 |2 )dσx dτ .
0 Γεwa

Using (7), it yields

  t
1 ε 2
|∂t u2 | dx + (d20 − C ε ) 2
|∇∂t u2 ε |2 dxdτ
2 Ωε 0 Ωε
  t  t  t
1
≤ |∂t uε2 (0)|2 dx + C |∂t uε1 |2 xdτ + C (|∂t uε3 |2 + ε 2 |∇∂t uε3 |2 ) + C |∂t uε2 |2 . (42)
2 Ωε 0 Ωε 0 Ω1ε 0 Ωε

Differentiating now the PDE in (4) with respect to time and then testing the result with
∂t (uε3 − uD3 ), we get
  t
1
|∂t uε3 |2 dx + d30 |∇∂t u3 ε |2 dxdτ
2 Ω1ε 0 Ω1ε
  t  t
1
≤ |∂t uε3 (0)|2 dx + dε3 ∇∂t u3 ε ∇∂t u3 D dxdτ + ∂tt uε3 ∂t u3 D dxdτ
2 Ω1ε 0 Ω1ε 0 Ω1ε
 t
+ εC (|∂t u2 ε |2 + |∂t u3 ε |2 + |∂t u3 D |2 )dσx dτ . (43)
0 Γεwa

The term with second time derivative can be estimated by

 t  t  t  t
∂tt uε3 ∂t u3 D dxdτ = ∂t uε3 ∂t u3 D dxdτ − ∂t uε3 (0)∂t u3 D (0)dxdτ − ∂tt u3 D ∂t uε3 dxdτ .
0 Ω1ε 0 Ω1ε 0 Ω1ε 0 Ω1ε

Using (7) to deal with the boundary terms and (A5), we obtain
  t  t
1 ε 2 ε 2
|∂t u3 | dx + (d3 − C ε ) 2
|∇∂t u3 | dxdτ ≤ C0 + C |∂t u3 ε |2 dxdτ
2 Ω1ε 0 Ω1ε 0 Ω1ε
 t
+C (|∂t u2 ε |2 + ε 2 |∇∂t u2 ε |2 )dxdτ , (44)
0 Ωε
where C0 contains bounded terms. The boundary data is smooth enough and regularity assumptions in u20 and u30 imply
that ∥∂t uε2 (0)∥L2 (Ω ε ) and ∥∂t uε3 (0)∥L2 (Ω ε ) can be estimated by H 2 -norm of the corresponding initial data. Adding (42) and
1
(44), and re-arranging the terms gives
 t  t
(d2 − C ε 2 ) |∇∂t u2 ε |2 dxdτ + (d3 − C ε 2 ) |∇∂t u3 ε |2 dxdτ ≤ C .
0 Ωε 0 Ω1ε

Choosing ε small and taking the supremum over the time interval in question, we get
 T   T 
|∇∂t u2 ε |2 dxdτ + |∇∂t u3 ε |2 dxdτ ≤ C . (45)
0 Ωε 0 Ω1ε

Inserting (45) in (39) and (40) yields the required result. 

4. Passage to ε → 0. Known results

Our main interest lies in the passing to the homogenization limit. We do this by following a three-steps procedure. In
Step 1, we rely on the standard extension5 results from [25] to extend all active concentrations uεℓ (ℓ ∈ {1, . . . , 4}) to Ω .
As Step 2, we use the two-scale convergence concept and the corresponding two-scale compactness result developed by
Nguetseng and Allaire to pass to the limit in the (weak form of the) PDEs. In Step 2, we unfold the ode for uε5 to replace the
oscillating boundary with a fixed one, say Γ ; see Section 5.1. The last two steps will be detailed in the next section; here we
focus more on the concept of two-scale convergence and available compactness properties.

5 Since we deal here with an oscillating system posed in a perforated domain, the natural next step is to extend all concentrations to the whole Ω .
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 13

4.1. Two-scale convergence and compactness. Basic convergences

Since we deal with periodic arrangements of microstructures, the most natural type of convergence is the one presented
in the next definition.

Definition 9 (Two-Scale Convergence; Cf. [41,42]). Let {uε } be a sequence of functions in L2 ((0, T ) × Ω ) (Ω being an open
set of RN ) where ε being a sequence of strictly positive numbers tends to zero. {uε } is said to two-scale converge to a unique
function u0 (t , x, y) ∈ L2 ((0, T ) × Ω × Y ) if and only if for any ψ ∈ C0∞ ((0, T ) × Ω , C#∞ (Y )), we have
 T   
ε
 x
lim u ψ t , x, dxdt = u0 (t , x, y)ψ(t , x, y)dydxdt . (46)
ε→0 0 Ω ε Ω Y

2
We denote (46) by uε ⇀ u0 .

Theorem 10 (Compactness; Cf. [41,42]).

(i) From each bounded sequence {uε } in L2 ((0, T )× Ω ), one can extract a subsequence which two-scale converges to u0 (t , x, y) ∈
L2 ((0, T ) × Ω × Y ).
(ii) Let {uε } be a bounded sequence in H 1 ((0, T ) × Ω ), then there exists ũ ∈ L2 ((0, T ) × Ω ; H#1 (Y )/R) such that up to a
2
subsequence {uε } two-scale converges to u0 (t , x) ∈ L2 ((0, T ) × Ω ) and ∇ uε ⇀ ∇x u0 + ∇y ũ.

Definition 11 (Two-Scale Convergence for ε -Periodic Hypersurfaces [29]). A sequence of functions {uε } in L2 ((0, T ) × Γε ) is
said to two-scale converge to a limit u0 ∈ L2 ((0, T ) × Ω × Γ ) if and only if for any ψ ∈ C0∞ ((0, T ) × Ω , C#∞ (Γ )) we have
 T   
 x
lim ε uε ψ t , x , dσx dt = u0 (t , x, y)ψ(t , x, y)dσy dxdt .
ε→0 0 Γε ε Ω Γ

Theorem 12. (i) From each bounded sequence {uε } ∈ L2 ((0, T ) × Γε ), one can extract a subsequence uε which two-scale
converges to a function u0 ∈ L2 ((0, T ) × Ω × Γ ).
(ii) If a sequence of functions {uε } is bounded in L∞ ((0, T ) × Γε ), then uε two-scale converges to a function u0 ∈ L∞ ((0, T ) ×
Ω × Γ ).
Proof. For proof of (i), see [29] and the one for (ii), see [17]. 

Estimates in Lemmas 3 and 8 lead to the following convergence results:

Lemma 13. Assume (G) together with (A1)–(A6) . Then it holds for i ∈ {1, 2, 3, 4}
(a) uεi ⇀ ui weakly in L2 (0, T ; H 1 (Ω )),
∗
(b) uεi ⇀ ui weakly in L∞ ((0, T ) × Ω ),
(c) ∂t uεi ⇀ ∂t ui weakly in L2 ((0, T ) × Ω ),
√ ε
(d) uεi → ui strongly in L2 (0, T ; H β (Ω )) for 1
2
< β < 1, also ε∥ui − ui ∥L2 ((0,T )×Γε ) → 0 as ε → 0,
2 2
(e) uεi ⇀ ui , ∇ uεi ⇀ ∇x ui + ∇y ui1 , ui1 ∈ L2 ((0, T ) × Ω ; H#1 (Y )/R),
2
(f) uε5 ⇀ u5 , and u5 ∈ L∞ ((0, T ) × Ω × Γ sw ),
2
(g) ∂t uε5 ⇀ ∂t u5 , and ∂t u5 ∈ L2 ((0, T ) × Ω × Γ sw ).

Proof. (a) and (b) are obtained as a direct consequence of the fact that uεi is bounded in L2 (0, T ; H 1 (Ω )) ∩ L∞ ((0, T ) × Ω );
up to a subsequence (still denoted by uεi ) uεi converges weakly to ui in L2 (0, T ; H 1 (Ω )) ∩ L∞ ((0, T ) × Ω ). A similar argument
gives (c). To get (d), we use the compact embedding H β (Ω ) ↩→ H β (Ω ), for β ∈ ( 12 , 1) and 0 < β < β ′ ≤ 1 (since Ω has
′

Lipschitz boundary). We have W := {ui ∈ L2 (0, T ; H 1 (Ω )) and ∂t ui ∈ L2 ((0, T ) × Ω ) for all i ∈ {1, 2, 3, 4}}. For a fixed
ε, W is compactly embedded in L2 (0, T ; H β (Ω )) by the Lions–Aubin Lemma; cf. e.g. [43]. Using the trace inequality (8)
√ ε
ε∥ui − ui ∥L2 ((0,T )×Γε ) ≤ C ∥uεi − ui ∥L2 (0,T ;H β (Ω ε ))
≤ C ∥uεi − ui ∥L2 (0,T ;H β (Ω ))
where ∥uεi − ui ∥L2 (0,T ;H β (Ω )) → 0 as ε → 0. To investigate (e)–(g), we use the notion of two-scale convergence as indicated
2
in Definitions 9 and 11. Since uεi are bounded in L2 (0, T ; H 1 (Ω )), up to a subsequence uεi ⇀ ui in L2 ((0, T ) × Ω ), and
2
∇ uεi ⇀ ∇x ui + ∇y ũi , ũi ∈ L2 ((0, T ) × Ω ; H#1 (Y )/R). By Theorem 12, uε5 in L∞ ((0, T ) × Γεsw ) converges two-scale to
u5 L∞ ((0, T ) × Ω × Γ ) and ∂t uε5 converges two-scale to ∂t u5 in L2 ((0, T ) × Ω × Γ ). 
14 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

4.2. Cell problems

In order to be able to formulate the limit (upscaled) equations in a compact manner, we define two classes of cell problems
very much in the spirit of [11]. One class of problems refers to the water-filled parts of the pore, while the second class refers
to the air-filled part of the pores.

Definition 14 (Cell Problems). The cell problems for the water-filled part are given by
3

in Y w ,

−∇ · (d ( y)∇ χ ) = ∂yk dℓ ki (y)

ℓ

 y y i

k=1
(47)
∂χi 3
on Γ sw ∪ Γ wa ,

−dℓ (y) dℓ ki (y)nk

=


∂n k=1

for all i ∈ {1, 2, 3}, ℓ ∈ {1, 2, 4}, χi are Y -periodic in y. The following solvability condition is trivially satisfied
3 
 3 

∂yk dℓ ki (y)dy = dℓ ki (y)nk dγy . (48)
k=1 Yw k=1 ∂Y w

The cell problems for the air-filled part are given by

3


−∇ · (d (y)∇ ς ) = ∂yk d3ki (y) in Y a ,



 y 3 y i
k=1
(49)
∂ςi 3

−d3 (y) d3ki (y)nk on Γ

 wa
=


∂n k=1

for all i ∈ {1, 2, 3}, ςi are Y -periodic in y. The following solvability condition is trivially fulfilled
3 
 3 

∂yk dℓ ki (y)dy = dℓ ki (y)nk dγy . (50)
k=1 Ya k=1 ∂Y a

Note that standard theory of linear elliptic problems with periodic boundary conditions [7] ensures the solvability of the
above families of cell problems.

5. Derivation of the two-scale limit equations

Theorem 15. Assume (A1)–(A6) and (G). The sequences of the solutions of the weak formulation (10)–(14) converges to the
weak solution ui , i ∈ {1, 2, 3, 4} as ε → 0 such that uj ∈ H 1 (0, T ; L2 (Ω )) ∩ L2 (0, T ; H 1 (Ω )), j ∈ {1, 2, 4}, u3 ∈ uD3 +
L2 (0, T ; HΓ1 D (Ω )), ∂t u3 ∈ ∂t uD3 + L2 (0, T ; L2 (Ω )) and u5 ∈ H 1 (0, T ; L2 (Ω × Γ sw )). The weak formulation of the two-scale
limit equations for i ∈ {1, 2, 3, 4} is given by
 T   T 
(∂t ui (t , x) − Fi (u))φi (t , x)dxdt + Di ∇ ui (t , x)∇φi dxdt = 0, (51)
0 Ω 0 Ω

where

1
F1 (u) := −k̃1 u1 (t , x) + k̃2 u2 (t , x) − k3 (y)R(u1 (t , x))Q (u5 (t , x, y))dσy ,
|Y w | Γ sw
|Y a |
F2 (u) := k̃1 u1 (t , x) − k̃2 u2 (t , x) + w ãu3 (t , x) − b̃u2 (t , x),
|Y |
|Y w |
F3 (u) := −ãu3 (t , x) + b̃u2 (t , x), F4 (u) := k̃1 u1 (t , x),
|Y a |
3 
1 
(Dℓ )jk := w ((dℓ )jk + (dℓ )ik ∂yi χj )dy, for ℓ ∈ {1, 2, 4}
|Y | i=1 Y w
3  
1  1
(D3 )jk := a ((d3 )jk + (d3 )ℓk ∂yℓ ςj )dy, b̃ := b(y)dσy
|Y | ℓ=1 Y a |Y w | Γ wa
 
1 1
ã := a ( y )d σy , k̃m := km (y)dy, m ∈ {1, 2},
|Y a | Γ wa |Y w | Y w
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 15

with the initial values ui (0, x) = ui0 (x) for x ∈ Ω and χj , ςj being solutions of the cell problems defined in Definition 14.
Furthermore, we have
 T 
∂t u5 (t , x, y) − k3 (y)R(u1 (t , x))Q (u5 (t , x, y)) φ5 (t , x, y)dtdxdσy = 0,
 
(52)
0 Ω ×Γ sw

with u5 (0, x, y) = u50 (x, y) for x ∈ Ω , y ∈ Γ sw . Also ϕj ∈ L2 (0, T ; H 1 (Ω )), j ∈ {1, 2, 4}, ϕ3 ∈ L2 (0, T ; HΓ1 D (Ω )) and
ϕ5 ∈ L2 ((0, T ) × Ω × Γ sw ).
Proof. We apply two-scale convergence techniques together with Lemma 13 to get macroscopic equations. We take test
functions incorporating the following oscillating behavior ϕi (t , x) = φi (t , x) + εψi (t , x, εx ), φj ∈ C ∞ ((0, T ) × Ω ), ψj ∈
C ∞ ((0, T ) × Ω , ; C#∞ (Y )), i ∈ {1, 2, 3, 4}, j ∈ {1, 2, 4} and φ3 ∈ C0∞ ((0, T ) × Ω ), ψ2 ∈ C0∞ ((0, T ) × Ω , ; C#∞ (Y )). Applying
two-scale convergence yields for i ∈ {1, 2, 4}
 T   T  
|Y w | ∂t ui φi (t , x)dxdt + di (y)(∇x ui (t , x)
0 Ω 0 Ω Yw
 T 
+ ∇y ũi (t , x, y))(∇x φi (t , x) + ∇y ψi (t , x, y))dydxdt = Fi (u)φi (t , x)dxdt , (53)
0 Ω

and
 T   T  
|Y | a
∂t u3 φ3 (t , x)dxdt + d3 (y)(∇x u3 (t , x)
0 Ω 0 Ω Ya
 T 
+ ∇y ũ3 (t , x, y))(∇x φ3 (t , x) + ∇y ψ3 (t , x, y))dydxdt = F3 (u)φ3 (t , x)dxdt , (54)
0 Ω

where
 T   T   x 
F1 (u)φ1 (t , x)dxdt = lim −kε1 uε1 + kε2 uε2
φ1 (t , x) + εψ1 t , x,

dxdt
0 Ω ε→0 0 Ωε ε
 T   x 
− lim ε ηε (R(uε1 ), Q (uε5 )) φ1 (t , x) + εψ1 t , x, dσx dt .
ε→0 0 Γεsw ε
Using Lemma 13 and (14), we have
 T   T  
F1 (u)φ1 (t , x)dxdt = −k1 (y)u1 (t , x) + k2 (y)u2 (t , x) φ1 (t , x)dydxdt
 
0 Ω 0 Ω  Yw
T    x 
− lim ε ∂t uε5 φ1 (t , x) + εψ1 t , x, dσx dt ,
ε→0 0 Γεsw ε
 T  T
F1 (u)φ1 (t , x)dxdt = |Y w | −k̃1 u1 (t , x) + k̃2 u2 (t , x) φ1 (t , x)dxdt
 
0 Ω  T 0 Ω 
− ∂t u5 φ1 (t , x)dσy dxdt .
0 Ω Γ sw
 T   T 
  x 
F2 (u)φ2 (t , x)dxdt = lim kε1 uε1 − kε2 uε2
φ2 (t , x) + εψ2 t , x,

dxdt
0 Ω ε→0 0 ε
Ω T  ε
 ε ε   x 
+ lim ε a u3 − bε uε2 φ2 (t , x) + εψ2 t , x, dσx dt .

ε→0 0 Γεwa ε
Passage to the limits ε → 0 gives
 T   T 
F2 (u)φ2 (t , x)dxdt = |Y w | k̃1 u1 (t , x) − k̃2 u2 (t , x) φ2 (t , x)dxdt
 
0 Ω 0 Ω
 T  
a(y)u3 (t , x) − b(y)u2 (t , x) φ2 (t , x)dxdydt ,
 
+
0 Ω Γ wa
 T 
w
−k̃1 u1 (t , x) + k̃2 u2 (t , x) φ2 (t , x)dxdt
 
= |Y |
0 Ω
 T 
|Y a |ãu3 (t , x) − |Y w |b̃u2 (t , x) φ2 (t , x)dxdt .
 
+
0 Ω
16 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

We also obtain
 T   T 
F3 (u)φ3 (t , x)dxdt = − |Y a |ãu3 (t , x) − |Y w |b̃u2 (t , x) φ3 (t , x)dxdt ,
 
0 Ω 0 Ω
 T   T 
w
F4 (u)φ4 (t , x)dxdt = |Y | k̃1 u1 (t , x)φ4 (t , x)dxdt .
0 Ω 0 Ω

We set φi = 0, i ∈ {1, 2, 4} in (53) to calculate the expression of the unknown function ũ1 and obtain
 T  
di (y)(∇x ui (t , x) + ∇y ũi (t , x, y))∇y ψi (t , x, y)dydxdt = 0, for all ψi .
0 Ω Yw

Since ũi depends linearly on ∇x ui , it can be defined as

3

ũi := ∂xj ui χj
j =1

where the function χj are the unique solutions of the cell problems defined in Definition 14. Similarly, we have ũ3 :=
∂xj u3 ςj where ςj are the unique solutions of the cell problems defined in Definition 14. Setting ψi = 0 in (53), we get
3
j =1
 
 T   3 3

dijk (y) ∂xk ui (t , x) + ∂yk χm ∂xm ui (t , x) ∂xj φi (t , x)dydxdt
0 Ω Y w j,k=1 m=1
 T 3
 
= |Y w | (Di )jk ∂xk ui (t , x)∂xj φi (t , x)dxdt .
0 Ω j,k=1

Hence, the coefficients (entering the effective diffusion tensor in water) are given by
3 
1 
(Di )jk := ((di )jk + (di )ℓk ∂yℓ χj )dy, for i ∈ {1, 2, 4}.
|Y w | ℓ=1 Yw

Similarly, we obtain the following effective diffusion coefficient of H2 S(g )

3 
1 
(D3 )jk := ((d3 )jk + (d3 )ℓk ∂yℓ ςj )dy. 
|Y a | ℓ=1 Ya

5.1. Passing to the limit ε → 0 in (14)

It is not yet possible to pass to the limit ε → 0 with the convergence results stated in Lemma 13. To overcome this
difficulty, we use the notion of periodic unfolding. It is worth mentioning that there is an intimate link between the two-scale
convergence and weak convergence of the unfolded sequences; see [17,44]. The key idea is to obtain strong convergence for
the periodic unfolding of uε5 instead of getting strong convergence for uε5 .

Definition 16. For ε > 0, the boundary unfolding of a measurable function ϕ posed on oscillating surface Γε is defined by
Tbε ϕ(x, y) = ϕ(ε y + ε k), y ∈ Γ, x ∈ Ω

where k := [ ε ] denotes the unique integer combination Σj3=1 kj ej of the periods such that x − [ εx ] belongs to Y . Note that the
x

oscillation due to the perforations are shifted into the second variable y which belongs to fixed surface Γ .

Lemma 17. If ϕε converges two-scale to ϕ and Tbε ϕε converges weakly to ϕ ∗ in L2 ((0, T ) × Ω ; L2# (Γ )), then ϕ = ϕ ∗ a.e. in
(0, T ) × Ω × Γ .
Proof. The proof details for this statement can be found in Lemma 4.6 of [17]. 

Lemma 18. If ϕ ∈ L2 ((0, T ) × Γ ε ), then the following identity holds

1 √
∥Tbε ϕ∥L2 ((0,T )×Ω ×Γ ) = ε∥ϕ∥L2 ((0,T )×Γ ε ) .
|Y |

Proof. See [17,45,46] for the proof details. 

 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 17

Lemma 19. If ϕ ∈ L2 (Ω ), then Tbε ϕ → ϕ as ε → 0 strongly in L2 (Ω × Γ ).

Proof. See e.g. [46,45] for the details of the proof. 

Lemma 20. If ϕ ∈ L2 (Γ ), then Tbε ϕ → ϕ as ε → 0 strongly in L2 (Ω × Γ ).

Proof. Proof goes on the same line as the proof of Lemma 19; see also Proposition 2.11(f) in [47].
Using the boundary unfolding operator Tbϵ , we unfold the ode (14). Changing the variable, x = ε y + ε k (for x ∈ Γεsw ) to
the fixed domain (0, T ) × Ω × Γ sw , we have

∂t Tbϵ uϵ5 (t , x, y) = ηε (Tbϵ uϵ1 (t , x, y), Tbϵ uϵ5 (t , x, y)). (55)

ε ε sw
In the remainder of this section, we prove that Tb u5 converges strongly to u5 in L ((0, T ) × Ω × Γ ). From the two-scale 2

convergence of uϵ5 , we obtain weak convergence of T ϵ uϵ5 to u5 in L2 ((0, T )× Ω ; L2# (Γ sw )). We start with showing that {Tbε uε5 }
is a Cauchy sequence in L2 ((0, T ) × Ω × Γ sw ). We choose m, n ∈ N with n > m arbitrary. Writing down (55) for the two
different choices of ε (i.e. εi = εn and εi = εm ), we obtain after subtracting the corresponding equations that

∂t |Tbϵn uϵ5n − Tbϵm uϵ5m |2 dσy dx
Ω ×Γ sw

= [k3 (y)R(Tbϵn uϵ1n )Q (Tbϵn uϵ5n ) − k3 (y)R(Tbϵm uϵ1m )Q (Tbϵm uϵ5m )) × (Tbϵn uϵ5n − Tbϵm uϵ5m )dσy dx,
Ω ×Γ sw
 
ϵn ϵn ϵm ϵm 2
≤C |Tb u5 − Tb u5 | dσy dx + C |Tbϵn uϵ1n − Tbϵm uϵ1m |2 dσy dx. (56)
Ω ×Γ sw Ω ×Γ sw
ϵ ϵ
To get (56), we have used the uniform boundedness of Tb n u1n and properties of R and Q . Since uε1 converges strongly to u1
in L2 ((0, T ) × Γεsw ) by Lemma 13 (d), we get by Lemma 18
 
|Tbϵ uϵ1 − Tbϵ u1 |2 dσy dx = ε |uϵ1 − u1 |2 dσx ≤ C ε. (57)
Ω ×Γ sw Γεsw

Since u1 is constant w.r.t. y, we have that Tbϵ u1 → u1 strongly in L2 ((0, T ) × Ω × Γ sw ) as ε → 0. Combining (57) and the
strong convergence of Tbϵ u1 to u1 we get
  
ϵn ϵn ϵm ϵm 2 ϵn
|Tb u1 − Tb u1 | dσy dx ≤ εn |u1 − u1 | dσx + εm 2
|uϵ1m − u1 |2 dσx
Ω ×Γ sw Γεsw Γεsw

|Tbϵn u1 − u1 |2 + |Tbϵm u1 − u1 |2 dσy dx
 
+
Ω ×Γ sw
≤ C (εn + εm ),
while (56) becomes
 
ϵn ϵn ϵm ϵm 2
∂t |Tb u5 − Tb u5 | dσy dx ≤ C |Tbϵn uϵ5n − Tbϵm uϵ5m |2 dσy dx + C (εn + εm ).
Ω ×Γ s w Ω ×Γ sw

Applying Gronwall’s inequality, we obtain:

 
ϵn ϵn ϵm ϵm
|Tb u5 (t ) − Tb u5 (t )| dσy dx ≤ 2
|Tbϵn uϵ5n (0) − Tbϵm uϵ5m (0)|2 dσy dx + C (εn + εm )
Ω ×Γ sw Ω ×Γ s w

≤ |Tbϵn u50 − u50 |2 dσy dx
Ω ×Γ s w

+ |Tbϵm u50 − u50 |2 dσy dx + C (εn + εm ).
Ω ×Γ s w

Using Lemma 20, we get

∥Tbϵn uϵ5n − Tbϵm uϵ5m ∥L2 ((0,T )×Ω ×Γ sw ) −→ 0 as ϵn , ϵm → 0. (58)

ϵ ϵ
By (58), {Tb u5 } is a Cauchy sequence. Now, we take the two-scale limit in the ode (55) to get
 T 
lim ε ∂t Tbϵ uε5 φ5 (t , x, y)dxdσy dτ
ε→0 0 Ω ×Γ s w
 T 
= lim ε ηε (Tbϵ uε1 , Tbϵ uε5 )φ5 (t , x, y)dxdσy dτ . (59)
ε→0 0 Ω ×Γ sw
18 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) –

Consequently, we have
 T   T   x
∂t u5 φ5 (t , x, y)dxdσy dt = lim ε Tbϵ kε3 R(Tbϵ uε1 )Q (uε5 )φ5 t , x, dxdσy dt ,
0 Ω ×Γ sw ε→0 0 Ω ×Γ s w ε
 T   x
= lim ε k3 (y)R(Tbϵ uε1 )Q (u5 )φ5 t , x, dxdσy dt
ε→0 0 Ω ×Γ s w ε
 T   x
+ lim ε k3 (y)R(Tbϵ uε1 )(Q (Tbϵ uε5 ) − Q (u5 ))φ5 t , x, dxdσy dt .
ε→0 0 Ω ×Γ s w ε
(60)

By (A2) and the strong convergence of uε1 , the first term on the right hand side of (60) converges two-scale to
 T  
k3 (t , y)R(u1 )Q (u5 )φ5 (t , x, y)dσy dxdt ,
0 Ω Γ sw

while the second integral of (60)

 T   x
ε k3 (y)R(Tbϵ uε1 )(Q (Tbϵ uε5 ) − Q (u5 ))φ5 t , x, dxdσy dt
0 Ω ×Γ sw ε
 T   12
x 2
 
≤ε k3 (y)R(Tbϵ uε1 )φ5 t , x,  dxdσy dt

0 Ω ×Γ sw ε
 T   12
ϵ ε
× |Q (Tb u5 ) − Q (u5 )| dxdσy dt
2
→ 0 as ε → 0.
0 Ω ×Γ sw

At this point, we have used again (A2) in combination with the strong convergence of Tbϵ uε5 . So, as a result of passing to the
limit ε → 0 in (59), we get (52).
Since our positivity and L∞ -bounds are independent of the choice of ε , these properties hold true also for weak solutions
to the limit problem (51)–(52). Having these bounds available, the proof of the uniqueness of solutions follows very much
in the same spirit as for the microscopic problem. 

Lemma 21 (Strong Formulation of the Two-Scale Limit Equations). Assume the hypothesis of Lemma 13 to hold. Then the strong
formulation of the two-scale limit equations (for all t ∈ (0, T )) reads

1
∂t u1 (t , x) + ∇ · (−D1 ∇ u1 (t , x)) = −k̃1 u1 (t , x) + k̃2 u2 (t , x) − k3 (y)R(u1 (t , x))Q (u5 (t , x, y))dσy , x∈Ω
|Y | Γ sw (61)
n · (−D1 ∇ u1 (t , x)) = 0, x ∈ Γ D
∪Γ ,
N

u1 (0, x) = u10 (x), x ∈ Ω̄ ,

|Y a |
∂t u2 (t , x) + ∇ · (−D2 ∇ u2 (t , x)) = k̃1 u1 (t , x) − k̃2 u2 (t , x) + ãu3 (t , x) − b̃u2 (t , x), x ∈ Ω,
|Y w | (62)
n · (−D2 ∇ u2 (t , x)) = 0, x ∈ Γ D
∪Γ ,
N

u2 (0, x) = u20 (x), x ∈ Ω̄ ,

|Y w |
∂t u3 (t , x) + ∇ · (−D3 ∇ u3 (t , x)) = −ãu3 (t , x) + b̃u2 (t , x), x ∈ Ω,
|Y a |
u3 (0, x) = u30 (x), x ∈ Ω̄ , (63)
u3 (t , x) = uD3 (x), x ∈ Γ D ,
n · (−D3 ∇ u3 (t , x)) = 0, x ∈ Γ N ,

∂t u4 (t , x) + ∇ · (−D4 ∇ u4 (t , x)) = k̃1 u1 (t , x), x ∈ Ω,

u4 (0, x) = u40 (x), x ∈ Ω̄ , (64)
n · (−D4 ∇ u4 (t , x)) = 0, x ∈ Γ D ∪ Γ N ,
∂t u5 (t , x, y) = k3 (y)R(u1 (t , x))Q (u5 (t , x, y)), x ∈ Ω , y ∈ Γ sw ,
(65)
u5 (0, x, y) = u50 (x, y) x ∈ Ω̄ , y ∈ Γ sw ,

where Di , i ∈ {1, 2, 3, 4}, ã, b̃ and k̃j , j ∈ {1, 2} are defined in Theorem 15.
 T. Fatima, A. Muntean / Nonlinear Analysis: Real World Applications ( ) – 19

Acknowledgments

The proof idea of Claim 5 is due to T. Aiki (Gifu, Japan). We acknowledge fruitful discussions on this subject with
M. Ptashnyk (Dundee, UK) and M. A. Peletier (TU Eindhoven, NL). Finally, we thank the two reviewers for their critical
comments.

References

[1] T. Fatima, N. Arab, E.P. Zemskov, A. Muntean, Homogenization of a reaction–diffusion system modeling sulfate corrosion in locally-periodic perforated
domains, J. Eng. Math. 69 (2–3) (2010) 261–276.
[2] M. Böhm, F. Jahani, J. Devinny, G. Rosen, A moving-boundary system modeling corrosion of sewer pipes, Appl. Math. Comput. 92 (1998) 247–269.
[3] C.V. Nikolopoulos, A mushy region in concrete corrosion, Appl. Math. Modelling 34 (2010) 4012–4030.
[4] V. Chalupecký, T. Fatima, A. Muntean, Numerical study of a fast micro–macro mass transfer limit: the case of sulfate attack in sewer pipes, J. Math.
Ind. 2 (2010B-7) 171–181.
[5] A. Muntean, M. Neuss-Radu, A multiscale Galerkin approach for a class of nonlinear coupled reaction–diffusion systems in complex media, J. Math.
Anal. Appl. 371 (2) (2010) 705–718.
[6] A. Bensoussan, J.L. Lions, G. Papanicolau, Asymptotic Analysis for Periodic Structures, North-Holland, Amsterdam, 1978.
[7] D. Cioranescu, P. Donato, An Introduction to Homogenization, Oxford University Press, New York, 1999.
[8] L.E. Persson, L. Persson, N. Svanstedt, J. Wyller, The Homogenization Method, Chartwell Bratt, Lund, 1993.
[9] A. Muntean, V. Chalupecky, Homogenization Method and Multiscale Modeling, in: Lecture Notes of the Institte for Mathematics for Industry, vol. 34,
Kyushu University, Fukuoka, Japan, 2011.
[10] N. Antonic, M. Lazar, Parabolic variant of H-measures in homogenisation of a model problem based on Navier–Stokes equations, Nonlinear Anal. RWA
11 (2010) 4500–4512.
[11] U. Hornung, Homogenization and Porous Media, Springer, NY, 1997.
[12] A.G. Belyaev, A.L. Pyatnitski, G.A. Chechkin, Asymptotic behaviour of a solution to a boundary value problem in a perforated domain with oscillating
boundary, Sib. Math. J. 39 (4) (1998) 621–644.
[13] M.A. Peter, M. Böhm, Different choices of scaling in homogenization of diffusion and interfacial exchange in a porous medium, Math. Methods Appl.
Sci. 31 (2008) 1257–1282.
[14] S.A. Meier, Two-scale models for reactive transport and evolving microstructures, Ph.D. Thesis, University of Bremen, Bremen, Germany, 2008.
[15] S.A. Meier, A. Muntean, A two-scale reaction–diffusion system with micro-cell reaction concentrated on a free boundary, C. R. Méc. 336 (6) (2009)
481–486.
[16] T. van Noorden, Crystal precipitation and dissolution in a porous medium: effective equations and numerical experiments, Multiscale Model. Simul.
7 (3) (2009) 1220–1236.
[17] A. Marciniak-Czochra, M. Ptashnyk, Derivation of a macroscopic receptor-based model using homogenization techniques, SIAM J. Math. Anal. 40 (1)
(2008) 215–237.
[18] J.E. Hutchinson, Second fundamental form for varifolds and the existence of surfaces minimising curvature, Indiana Univ. Math. J. 35 (1) (1986) 45–71.
[19] T. Fatima, A. Muntean, M. Ptashnyk, Unfolding-based corrector estimates for a reaction–diffusion system predicting concrete corrosion,
arXiv:1106.3233v1.
[20] G. Chechkin, A.L. Piatnitski, A.S. Shamaev, Homogenization Methods and Applications, in: Translations of Mathematical Monographs, vol. 234, AMS,
Providence, Rhode Island, USA, 2007.
[21] V. Jikov, S. Kozlov, O. Oleinik, Homogenization of Differential Operators and Integral Functionals, Springer Verlag, 1994.
[22] A. Bourgeat, A. Mikelic, A. Piatnitski, Stochastic two-scale convergence in the mean and applications, J. Reine Angew. Math. 456 (1994) 19–51.
[23] T. van Noorden, A. Muntean, Homogenization of a locally-periodic medium with areas of low and high diffusivity, European J. Appl. Math. 22 (5) (2011)
493–516.
[24] D. Cioranescu, J.S.-J. Paulin, Homogenization in open sets with holes, J. Math. Anal. Appl. 71 (1979) 590–607.
[25] E. Acerbi, V.C. Piat, G.D. Maso, D. Percivale, An extension theorem from connected sets, and homogenization in general periodic domains, Nonlinear
Anal. TMA 18 (5) (1992) 481–496.
[26] P.W. Balls, P.S. Liss, Exchange of H2 S between water and air, Atmos. Environ. 17 (4) (1983) 735–742.
[27] P.V. Danckwerts, Gas–Liquid Reactions, McGraw-Hill Book Co., 1970.
[28] U. Hornung, W. Jäger, Diffusion, convection, adsorption and reaction of chemical in porous media, J. Differential Equations 92 (1991) 199–225.
[29] M. Neuss-Radu, Some extensions of two-scale convergence, C. R. Acad. Sci. Paris Sér. I Math. 332 (1996) 899–904.
[30] A. Farina, J. Bodin, T. Clopeau, A. Fasano, L. Meacci, A. Mikelic, Isothermal water flows in low porosity porous media in presence of vapor–liquid phase
change, Nonlinear Anal. RWA (in press) doi:10.1016/j.nonrwa.2011.11.02.
[31] H.F.W. Taylor, Cement Chemistry, Academic Press, London, 1990.
[32] M. Beneš, J. Zeman, Some properties of strong solutions to nonlinear heat and moisture transport in multi-layer porous structures, Nonlinear Anal.
RWA 13 (2011) 1562–1580.
[33] E. Aulisa, A. Ibragimov, M. Toda, Geometric framework for modeling nonlinear flows in porous media, and its applications in engineering, Nonlinear
Anal. RWA 11 (2010) 1734–1751.
[34] R.E. Beddoe, H.W. Dorner, Modelling acid attack on concrete: part 1. The essential mechanisms, Cem. Concr. Res. 12 (2005) 2333–2339.
[35] L. Franke (Ed.), Simulation of Time Dependent Degradation of Porous Materials: Final Report on Priority Program 1122, Cuvillier Verlag, Göttingen,
2009, pp. 275–292.
[36] R. Tixier, B. Mobasher, M. Asce, Modeling of damage in cement-based materials subjected to external sulfate attack. I: formulation, J. Mater. Civ. Eng.
15 (2003) 305–313.
[37] F. Girardi, W. Vaona, R.D. Maggio, Resistance of different types of concretes to cyclic sulfuric acid and sodium sulfate attack, Cem. Concr. Res. 32 (2010)
595–602.
[38] Y. Zhang, Q.C. Wang, L. Jia, L.M. Huo, A similarity model in concrete attack by sulfate erosion, Adv. Mater. Res. 374–377 (2012) 2379–2383.
[39] D. Agreba-Driolett, F. Diele, R. Natalini, A mathematical model for the SO2 aggression to calcium carbonate stones: numerical approximation and
asymptotic analysis, SIAM J. Appl. Math. 64 (5) (2004) 1636–1667.
[40] A. Kufner, O. John, S. Fucik, Function Spaces, Nordhoff Publ. and Czechoslovak Academy of Sciences Prague, 1977.
[41] G. Allaire, Homogenization and two-scale convergence, SIAM J. Math. Anal. 23 (6) (1992) 1482–1518.
[42] G. Nguestseng, A general convergence result for a functional related to the theory of homogenization, SIAM J. Math. Anal. 20 (1989) 608–623.
[43] J.L. Lions, Quelques Méthodes de Résolution des Problemes aux Limites Nonlinéaires, Dunod, Paris, 1969.
[44] D. Cioranescu, A. Damlamian, G. Griso, Periodic unfolding and homogenization, SIAM J. Math. Anal. 40 (4) (2008) 1585–1620.
[45] D. Cioranescu, P. Donato, R. Zaki, Asymptotic behavior of elliptic problems in perforated domains with nonlinear boundary conditions, Asymptot.
Anal. 53 (2007) 209–235.
[46] D. Cioranescu, P. Donato, R. Zaki, Periodic unfolding and Robin problems in perforated domains, C. R. Acad. Sci. Paris, Ser. I (342) (2006) 469–474.
[47] A. Damlamian, Homogenization of oscillating boundaries, Discrete Contin. Dyn. Syst. 23 (1, 2) (2009) 197–219.