J.

Differential Equations 244 (2008) 2909–2933

www.elsevier.com/locate/jde

Well-posedness and stability of a free boundary problem

modeling the growth of multi-layer tumors
Fujun Zhou a,∗ , Joachim Escher b , Shangbin Cui c
a Department of Mathematics, South China University of Technology, Guangzhou,
Guangdong 510640, People’s Republic of China
b Institute of Applied Mathematics, Leibniz University of Hannover, Walfengarten 1, Hannover, Germany
c Institute of Mathematics, Sun Yat-Sen University, Guangzhou, Guangdong 510275, People’s Republic of China

Received 26 January 2007; revised 28 February 2008

Abstract
In this paper we study well-posedness and stability of a free boundary problem modeling the growth of
multi-layer tumors under the action of external inhibitors. An important feature of this problem is that the
surface tension of the free boundary is taken into account. We first reduce this free boundary problem into
an evolution equation in little Hölder space and use the well-posedness theory for differential equations
in Banach spaces of parabolic type (i.e., equations which are treatable by using the analytic semi-group
theory) to prove that this free boundary problem is locally well-posed for initial data belonging to a little
Hölder space. Next we study flat solutions of this problem. We obtain all flat stationary solutions and give
a precise description of asymptotic stability of these stationary solutions under flat perturbations. Finally
we investigate asymptotic stability of flat stationary solutions under non-flat perturbations. By carefully
analyzing the spectrum of the linearized stationary problem and employing the theory of linearized stability
for differential equations in Banach spaces of parabolic type, we give a complete analysis of stability and
instability of all flat stationary solutions under small non-flat perturbations.
© 2008 Elsevier Inc. All rights reserved.

MSC: 35R35; 35B35

Keywords: Free boundary problem; Multi-layer tumors; Inhibitors; Well-posedness; Stability

* Corresponding author.
E-mail addresses: [email protected] (F. Zhou), [email protected] (J. Escher),
[email protected] (S. Cui).

0022-0396/$ – see front matter © 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jde.2008.02.038
2910 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

1. Introduction

In this paper we study well-posedness and asymptotic behavior of solutions of the following
multidimensional free boundary problem:
⎧
⎪ σ = λ1 σ + β in Ωρ(t) , t > 0,
⎪
⎪
⎪
⎪ β = λ2 β in Ωρ(t) , t > 0,
⎪
⎪
⎪
⎪ p = −μ(σ − σ̃ − ιβ) in Ωρ(t) , t > 0,
⎪
⎪ ∂σ
⎨ ∂β ∂p
= 0, = 0, = 0 on Γ0 , t > 0, (1.1)
⎪ ∂y ∂y ∂y
⎪
⎪ σ = σ̄ , β = β̄, p = γ κρ
⎪
⎪ on Γρ(t) , t > 0,
⎪
⎪ ∂p
⎪
⎪ ∂t ρ + =0
⎪
⎪
on Γρ(t) , t > 0,
⎩ ∂ν
ρ(0, ·) = ρ0 at t = 0.

Here σ = σ (t, x, y), β = β(t, x, y) and p = p(t, x, y) are unknown functions defined on the
unknown n-dimensional domain
 
Ωρ(t) := (x, y) ∈ Rn−1 × R: 0 < y < ρ(t, x), x ∈ Rn−1 ,

where ρ = ρ(t, x) is an unknown function, and  represents the Laplacian in the (x, y)-variables.
Besides, Γ0 denotes the lower boundary y = 0 of Ωρ(t) , Γρ(t) denotes the upper boundary
y = ρ(t, x) of Ωρ(t) , ν is the outward normal of the boundary Γρ(t) , i.e.,

ν = −∇x ρ(t, x), 1 , x ∈ Rn−1 , t ∈ [0, ∞),

κρ denotes the mean curvature of Γρ(t) , and λ1 , λ2 , μ, ι and γ are positive constants. The sign of
κρ is fixed on by the condition that κρ  0 at points where Γρ(t) is convex with respect to Ωρ(t) .
Note that the third equation in (1.1) is derived from Darcy’s law and mass balance equation.
Indeed, letting V be the velocity of the tumor cell movement, the Darcy’s law gives V = −∇p
and the mass balance equation yields div V = μ(σ − σ̃ − ιβ), which combined together leads to
p = −μ(σ − σ̃ − ιβ).
The above problem is a mathematical model for the growth of so-called multi-layer tumors
under the action of external inhibitors. A multi-layer tumor is a cluster of tumor cells cultivated
in laboratory by using the recently developed tissue culture technique [8,22,23,25,26,28]. It is
similar to other in vitro tumors such as the multi-cell spheroid tumor and the monolayer tumor in
biological property, but is different from them in geometric configuration. Recall that a multi-cell
spheroid is an in vitro tumor cultivated in nutrient solution and has a spherical or near-spherical
shape [2,3,5–7,9–12,14,18–21], and a monolayer is in vitro tumor cultivated on an impermeable
support membrane and consists of only one layer of tumor cells. Similar to the monolayer tumor,
a multi-layer tumor is also cultivated on an impermeable support membrane and does not have
a spherical or near-spherical shape. However, unlike the monolayer tumor, a multi-layer tumor
consists of many layers of tumor cells so that it has an observable thickness. For more details
about these phrases we refer the reader to see Refs. [13,22,23,25]. In the above model σ rep-
resents the (scaled) nutrient concentration, β represents the (scaled) inhibitor concentration, p
stands for the (scaled) internal pressure within the tumor that causes the motion of cellular mate-
rial, and σ̃ is the (scaled) threshold value for apoptosis of tumor cells. The conditions σ = σ̄ and
β = β̄ on the upper boundary Γρ(t) mean that the tumor receives constant nutrient and inhibitor
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2911

∂β ∂p
supply from the upper boundary, and the conditions ∂σ ∂y = 0, ∂y = 0 and ∂y = 0 on the lower
boundary Γ0 reflect the fact that none of nutrient, inhibitor and tumor cells can pass through the
lower boundary. The term γ κρ in the fifth line of (1.1) takes surface tension effects of the free
boundary Γρ(t) into account. Note that since we consider both avascular and vascularized tumors
simultaneously in this model, the (scaled) parameters σ̄ , β̄ and σ̃ may be negative as a result of
vascularization. Indeed, from [4] we know that, instead of the first two equations in (1.1), original
equations for unscaled σ and β are respectively as follows:

−σ = Γ1 (σ̄ − σ ) − λσ − γ1 β, −β = Γ2 (β̄ − β) − γ2 β,

where Γ1 , Γ2 , σ̄ , β̄, λ, γ1 , γ2 are constants. These equations are usable to both avascular tumors
and vascularized tumors: For avascular tumors we have Γ1 = Γ2 = 0, whereas for vascularized
tumors we have Γ1 > 0, Γ2 > 0. In both cases these equations can be rescaled into the first two
equations in (1.1), respectively. However, in the case Γ1 = Γ2 = 0 we have σ̄ > 0, β̄ > 0 and
σ̃ > 0 after rescaling, while in the case Γ1 > 0, Γ2 > 0 these parameters can become negative
after rescaling, cf. Section 1 in [5] and [11] for details.
In the previous work [8] the special inhibitor-free situation (i.e., β = 0) of the problem (1.1)
was systematically studied. Existence of non-flat stationary solutions of (1.1) was considered
in [26,28] by using the classical bifurcation theorem. The present paper aims at studying well-
posedness and asymptotic behavior of solutions of the inhibitor-present situation of the problem
(1.1). As one could see in the forthcoming sections, the inhibitor-present situation is more diffi-
cult than the inhibitor-free situation. We would like to mention that study of effects of inhibitors
to the growth of tumors is a significant topic due to its evident applications to tumor medicine.
Indeed, as was pointed out by Byrne and Chaplain in [4], analysis of mathematical models like
(1.1) can help medical doctors and researchers to assess the relative merits of different courses
of drug treatment and/or chemotherapy. As far as rigorous analysis is concerned, we refer the
reader to see [5,11] for previous work where tumors with spherical shapes are considered. Note
that the assumption that the tumor has a spherical shape renders the corresponding free boundary
problem to be of one dimension in the space variable in essence. In this paper we shall consider
non-spherical tumors, so that the free boundary problem under this study is essentially of more
than one dimension in the space variable. This determines that the problem considered in this
paper is more difficult than those investigated in Refs. [5] and [11].
Before stating our main results, let us introduce some notations. For the sake of simplicity we
impose the additional condition that ρ(t, x), σ (t, x, y), β(t, x, y) and p(t, x, y) are 2π -periodic
in every component of x. Moreover, it is not an essential restriction to consider the case n = 2,
because higher-dimensional periodic cases can be treated similarly. Thus x ∈ R and we assume
the following additional conditions:

ρ(t, x), σ (t, x, y), β(t, x, y) and p(t, x, y) are 2π-periodic in x. (1.2)

In addition, we identify 2π -periodic functions with functions over the circle S1 = R/2πZ.
Accordingly we identify the function spaces Cper (R), etc. of periodic functions on R with corre-
sponding function spaces C(S1 ), etc. on the circle S1 . Given m ∈ N+ and α ∈ (0, 1), we denote
by hm+α (S1 ) (respectively hm+α (Ω̄)) the so-called little Hölder space on S1 (respectively Ω̄),
i.e., the closure of C ∞ (S1 ) (respectively C ∞ (Ω̄)) in the usual Hölder space C m+α (S1 ) (respec-
tively C m+α (Ω̄)). Besides, C+ (S1 ) (respectively hm+α m+α 1
+ (Ω̄), h+ (S )) stands for the cone of all
2912 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

positive functions in C(S1 ) (respectively hm+α (Ω̄), hm+α (S1 )). Hereafter we shall fix α ∈ (0, 1)
and let δ ∈ (α, 1).
Given T > 0 and ρ ∈ C+ ([0, T ) × S1 ), let
 
Dρ,T := (t, x, y) t ∈ [0, T ), x ∈ S1 , 0  y  ρ(t, x) .

By a solution of (1.1) we mean a quartet (ρ, σ, β, p) such that ρ ∈ C([0, T ), h3+δ + (S )) ∩

1

C 1 ((0, T ), h1+α (S1 )), (σ, β, p) ∈ C(Dρ,T )×C(Dρ,T )×C(Dρ,T ), σ (t, ·), β(t, ·) ∈ h4+δ (Ω̄ρ(t) ),
p(t, ·) ∈ h2+δ (Ω̄ρ(t) ) for fixed t ∈ [0, T ), and (ρ, σ, β, p) satisfies (1.1) pointwise in Dρ,T .
Our first main result is as follows:

3+δ 1
Theorem 1.1. Given ρ0 ∈ h+ (S ), the problem (1.1) has a unique solution (ρ, σ, β, p) on
+
a maximal interval [0, t (ρ0 )), and the free boundary Γρ(t) depends smoothly on (t, x) ∈
(0, t + (ρ0 )) × S1 . Furthermore, if ρ0 ∈ h+
4+δ 1
(S ) then
 4+δ  1  
ρ ∈ C 0, t + (ρ0 ) , h+ S ∩ C 1 0, t + (ρ0 ) , h1+δ S1 .

The proof of this theorem will be given in Section 2. Next we study existence and numbers
of flat stationary solutions of (1.1) and investigate their asymptotic stability under flat perturba-
tions. Recall that a solution (ρ, σ, β, p) of (1.1) is called a flat solution if ρ = const and σ, β, p
are independent of the variable x. Later on we shall always assume that λ1 = λ2 ; the special case
λ1 = λ2 will not be particularly considered because it can be treated similarly with suitable mod-
ifications (cf. Remark 4.5). Our results depend on the relations between σ̃ , γ and the following
three parameters:

λ2 β̄ β̄
φ= , A1 = σ̄ − , A2 = − ιβ̄. (1.3)
λ1 λ2 − λ1 λ2 − λ1

For fixed φ = 1, the two lines A1 + φ 2 A2 = 0 and φA1 + A2 = 0 divide the A1 A2 -plane (with
the origin subtracted) into four disjoint regions Δ1 , Δ2 , Δ3 , Δ4 :
   
Δ1 := (A1 , A2 ) min φA1 + A2 , A1 + φ 2 A2  0 ,
   
Δ2 := (A1 , A2 ) max φA1 + A2 , A1 + φ 2 A2  0 ,
 
Δ3 := (A1 , A2 ) A1 + φ 2 A2 < 0 < φA1 + A2 ,
 
Δ4 := (A1 , A2 ) φA1 + A2 < 0 < A1 + φ 2 A2 . (1.4)

Our second main result is as follows:

Theorem 1.2. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(i) If (A1 , A2 ) ∈ Δ1 , then in the case A1 + A2 < σ̃ < 0 (1.1) has a unique flat stationary
solution (ρ∗ , σ∗ , β∗ , p∗ ), and in the case either σ̃  A1 + A2 or σ̃  0 (1.1) has no flat
stationary solution.
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2913

(ii) If (A1 , A2 ) ∈ Δ2 , then in the case 0 < σ̃ < A1 + A2 (1.1) has a unique flat stationary
solution (ρ∗ , σ∗ , β∗ , p∗ ), and in the case either σ̃  A1 + A2 or σ̃  0 (1.1) has no flat
stationary solution.
(iii) If (A1 , A2 ) ∈ Δ3 , then in the case either min(0, A1 + A2 ) < σ̃  max(0, A1 + A2 )
or σ̃ = A∗ (see (3.10) for its definition) (1.1) has a unique flat stationary solution
(ρ∗ , σ∗ , β∗ , p∗ ), in the case max(0, A1 + A2 ) < σ̃ < A∗ (1.1) has two flat stationary
solutions (ρ∗− , σ∗− , β∗− , p∗− ) and (ρ∗+ , σ∗+ , β∗+ , p∗+ ) (ρ∗− < ρ∗+ ), and in the case either
σ̃  min(0, A1 + A2 ) or σ̃ > A∗ (1.1) has no flat stationary solution.
(iv) If (A1 , A2 ) ∈ Δ4 , then in the case either min(0, A1 + A2 )  σ̃ < max(0, A1 + A2 )
or σ̃ = A (see (3.11) for its definition) (1.1) has a unique flat stationary solution
(ρ∗ , σ∗ , β∗ , p∗ ), in the case A < σ̃ < min(0, A1 + A2 ) (1.1) has two flat stationary
solutions (ρ∗− , σ∗− , β∗− , p∗− ) and (ρ∗+ , σ∗+ , β∗+ , p∗+ ) (ρ∗− < ρ∗+ ), and in the case either
σ̃  max(0, A1 + A2 ) or σ̃ < A (1.1) has no flat stationary solution.

Besides, given any positive constant ρ0 , let (ρ, σ, β, p) be the solution of (1.1) starting from
the initial ρ0 . Then (ρ, σ, β, p) is flat and exists globally, i.e., t + (ρ0 ) = ∞, and, as t → ∞, it
either converges to zero or a flat stationary solution, or ρ(t) tends to infinity (see Theorem 3.6 in
Section 3 for details).

The proof of the above theorem will be given in Section 3. Finally we study asymptotic sta-
bility of the flat stationary solutions ensured by the above result under non-flat perturbations.
We say that a stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is asymptotically stable if it is exponentially
stable under small h4+δ -perturbations, i.e., there are positive constants ω, ε, M such that if
ρ0 − ρ∗ 4+δ < ε then

ρ(t, ·) − ρ∗ 4+δ
+ σ (t, ·) − σ∗ 4+δ
+ β(t, ·) − β∗ 4+δ
+ p(t, ·) − p∗ 2+δ
 Me−ωt ,
t  0.

Similarly, a stationary solution is said to be unstable if it is unstable under h4+δ -perturbation.

Our last main result is as follows:

Theorem 1.3. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(i) If (A1 , A2 ) ∈ Δ1 , then in the case A1 + A2 < σ̃ < 0 the unique flat stationary solution
(ρ∗ , σ∗ , β∗ , p∗ ) is unstable.
(ii) If (A1 , A2 ) ∈ Δ2 , then in the case 0 < σ̃ < A1 + A2 there exists a threshold value γ ∗ > 0
such that the unique flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is asymptotically stable pro-
vided γ > γ ∗ , and unstable provided γ < γ ∗ .
(iii) If (A1 , A2 ) ∈ Δ3 , then in the case 0 < σ̃  A1 + A2 there exists a constant γ ∗  0 such that
the unique flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is asymptotically stable provided γ > γ ∗ ,
in the case A1 + A2 < σ̃  0 the unique flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is unstable,
in the case max(0, A1 + A2 ) < σ̃ < A∗ (ρ∗− , σ∗− , β∗− , p∗− ) is unstable and there exists a
constant γ ∗  0 such that (ρ∗+ , σ∗+ , β∗+ , p∗+ ) is asymptotically stable provided γ > γ ∗ .
(iv) If (A1 , A2 ) ∈ Δ4 , then in the case 0  σ̃ < A1 + A2 there exists a threshold value
γ ∗ > 0 such that the unique flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is asymptotically
stable provided γ > γ ∗ , and unstable provided γ < γ ∗ , in the case A1 + A2  σ̃ < 0
2914 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

the unique flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) is unstable, in the case A < σ̃ <
min(0, A1 + A2 ) (ρ∗+ , σ∗+ , β∗+ , p∗+ ) is unstable and there exists a threshold value γ ∗ > 0
such that (ρ∗− , σ∗− , β∗− , p∗− ) is asymptotically stable provided γ > γ ∗ , and unstable pro-
vided γ < γ ∗ .

The proof of Theorem 1.3 is based on careful analysis of the spectrum of the linearized sta-
tionary problem and applications of the theory of linearized stability for differential equations in
Banach spaces of the parabolic type [24], and will be given in Section 4.
Let us now discuss the results obtained above from the point of modeling. In fact, from the
above results we see that the relations between nutrient supply, inhibitor supply, apoptosis value,
tumor’s initial size and surface tension coefficient determine the final size of a tumor. A vascular-
ized tumor (σ̃ < 0) will eventually disappear, or converge to its smaller dormant (if there are two
dormant states) for large surface tension coefficient, or expand unbounded. An avascular tumor
(σ̃  0) will eventually disappear, or converge to its unique dormant state or larger dormant state
(if there are two dormant states) for large surface tension coefficient, regardless of its initial size.
The remaining of this paper is organized as follows. In the next section, we establish local
well-posedness of (1.1) and give the proof of Theorem 1.1. In Section 3, we study existence of
flat stationary solutions of (1.1) and consider asymptotic behavior of the free boundary ρ with
flat initial data. Section 4 aims at investigating the asymptotic stability of flat stationary solutions
of (1.1) under small non-flat perturbations and giving the proof of Theorem 1.3.

2. Well-posedness

In this section, we establish local well-posedness of the problem (1.1) for initial data belong-
ing to a little Hölder space.
First, we transform the problem (1.1) into a new problem on a fixed domain. Given
ρ ∈ C+2 (S1 ), we denote

 
y
θρ (x , y ) := x , for (x , y ) ∈ Ωρ .
ρ(x )

It can be easily verified that θρ is a C 2 -diffeomorphism from Ωρ /(2πZ × {0}) onto the manifold
Ω := S1 × (0, 1). Moreover, it is obvious that θρ−1 (x, y) = (x, yρ(x)) for (x, y) ∈ Ω. Let

θ∗ u := u ◦ θρ−1 and θρ∗ v := v ◦ θρ

ρ
for u ∈ C(Ω̄ρ ) for v ∈ C(Ω̄)

denote the push forward and pull back operators, respectively, induced by θ . Given ρ ∈ C+
2 (S1 )

and v ∈ C (Ω̄) we define the following operators:

2

ρ  ρ  ρ 
A(ρ)v := θ∗  θρ∗ v , B0 (ρ)v := θ∗ Υ0 ∇ θρ∗ v , n0 and B1 (ρ)v := θ∗ Υρ ∇ θρ∗ v , n1 ,

where Υ0 , Υρ stands for the trace operators on Γ0 , Γρ , respectively, and n0 = (0, −1), n1 =
(−ρx , 1) are the outward normals on Γ0 , Γρ , respectively. We also introduce the transformed
mean curvature operator
ρ  1
N (ρ) := θ∗ κρ , ρ ∈ C+
2
S .
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2915

Defining the transformed unknowns:

ρ ρ ρ
σ̂ (t) := θ∗ σ (t, ·), β̂(t) := θ∗ β(t, ·), p̂(t) := θ∗ p(t, ·), (2.1)

we see that (1.1) is transformed into the following problem on (ρ, σ̂ , β̂, p̂):
⎧
⎪
⎪ A(ρ)σ̂ = λ1 σ̂ + β̂ in (0, T ) × Ω,
⎪
⎪
⎪
⎪ A(ρ) β̂ = λ2 β̂ in (0, T ) × Ω,
⎪
⎪
⎨ A(ρ)p̂ = −μ(σ̂ − σ̃ − ιβ̂) in (0, T ) × Ω,
B0 (ρ)σ̂ = 0, B0 (ρ)β̂ = 0, B0 (ρ)p̂ = 0 on (0, T ) × Γ0 , (2.2)
⎪
⎪
⎪
⎪ Υ1 σ̂ = σ̄ , Υ1 β̂ = β̄, Υ1 p̂ = γ N (ρ) on (0, T ) × Γ1 ,
⎪
⎪
⎪
⎩ ∂t ρ + B1 (ρ)p̂ = 0
⎪ on (0, T ) × Γ1 ,
ρ(0, ·) = ρ0 on S1 ,

where Γi := S1 × {i}, i = 0, 1, and Υ1 is the trace operator on Γ1 . It is clear that, under the
transformation (2.1), the problem (2.2) subject to the periodic condition (1.2) is equivalent to the
problem (1.1). For simplicity of notation, in the rest part of this section we briefly write σ̂ , β̂, p̂
as σ , β, p, respectively. Note that this abbreviation does not produce confusion, because later on
in this section we shall only work on the problem (2.2) and shall not consider (1.1) any longer.
Given ρ ∈ C+ 2 (S1 ), elementary calculation implies that


2 
2
A(ρ) = aj k (ρ)∂j ∂k + a2 (ρ)∂2 and Bi (ρ) = bj i (ρ)Υi ∂j , i = 0, 1,
j,k=1 j =1

where

yρx 1 + y 2 ρx2 2yρx2 − yρρxx

a11 (ρ) := 1, a12 (ρ) := a21 (ρ) := − , a22 := , a2 (ρ) := ,
ρ ρ2 ρ2
1 ρx2 + 1
b10 (ρ) := 0, b20 (ρ) := , b11 (ρ) := −ρx , b21 (ρ) := .
ρ ρ

Using the fact that little Hölder spaces are Banach algebras it is not difficult to verify that
 3+α  1  
(A, Bi ) ∈ C ∞ h+ S , L hm+2+α (Ω̄), hm+α (Ω̄) × hm+1+α S1 , i = 0, 1, (2.3)

3+α 1
for m = 0, 1, cf. Lemma 2.2 in [16] and (2.2) in [8]. Besides, given ρ ∈ h+ (S ), let P be the
linear operator

 − 32 ∂ 2v
P(ρ)v := − 1 + ρx2 , (2.4)
∂x 2

so that the transformed curvature is given by N (ρ) = P(ρ)ρ. Moreover, we have that
 3+α  1   
P ∈ C ∞ h+ S , L h4+α S1 , h2+α S1 . (2.5)
2916 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

Next, we reduce (2.2) into a single equation containing the unknown ρ only. Given
3+α 1
ρ ∈ h+ (S ), by the theory of elliptic partial differential equations we know that the problem
(recall that we have abbreviated β̂ as β)
 A(ρ)β = λ β in Ω,
2
B0 (ρ)β = 0 on S1 × {0}, (2.6)
Υ1 β = β̄ on S1 × {1}

has a unique solution β ∈ h3+α (Ω̄), which we denote by Q(ρ). By (2.3) and the regularity theory
for elliptic equations we see that
 3+α  1 3+α
Q ∈ C ∞ h+ S , h (Ω̄) . (2.7)

Next we consider the boundary value problem (recall that σ is abbreviation of σ̂ )


A(ρ)σ = λ1 σ + Q(ρ) in Ω,
B0 (ρ)σ = 0 on S1 × {0}, (2.8)
Υ1 σ = σ̄ on S1 × {1},

where we replaced β with Q(ρ). It is not difficult to see that (2.8) has a solution σ := R(ρ)
satisfying
 3+α  1 3+α
R ∈ C ∞ h+ S , h (Ω̄) . (2.9)

Then we consider the following boundary value problem (recall that p is abbreviation of p̂)
⎧ 
⎨ A(ρ)p = −μ R(ρ) − σ̃ − ιQ(ρ) in Ω,
B (ρ)p = 0 on S1 × {0}, (2.10)
⎩ 0
Υ1 p = γ N (ρ) on S1 × {1},

where we replaced σ and β with R(ρ) and Q(ρ), respectively. Similarly as in [8,9,15–17,27],
3+α 1
for given ρ ∈ h+ (S ) we introduce two operators S(ρ) and T (ρ) by defining u = S(ρ)f and
v = T (ρ)g to be respectively solutions of the problems
 
A(ρ)u = f in Ω, A(ρ)v = 0 in Ω,
and
Υ1 u = 0 on S1 × {1} Υ1 v = g on S1 × {1}.

By the theory of elliptic partial differential equations we know that the solution of (2.10) is given
by

p = −μS(ρ) R(ρ) − σ̃ − ιQ(ρ) + γ T (ρ)P(ρ)ρ,

where the solution operators S and T satisfy

 3+α  1 
S ∈ C ∞ h+ S , L h1+α (Ω̄), h3+α (Ω̄) ,
 3+α  1  
T ∈ C ∞ h+ S , L h2+α S1 , h2+α (Ω̄) , (2.11)
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2917

3+α 1
respectively. Now we can introduce the following mappings for ρ ∈ h+ (S ):

Φ(ρ)ρ := γ B1 (ρ)T (ρ)P(ρ)ρ and F (ρ) := −μB1 (ρ)S(ρ) R(ρ) − σ̃ − ιQ(ρ) . (2.12)

Then the problem (2.2) is reduced to the following single equation containing ρ only

∂t ρ + Φ(ρ)ρ = F (ρ), ρ(0) = ρ0 . (2.13)

Besides, it follows from (2.3), (2.5), (2.7), (2.9) and (2.11) that
 3+α  1      1 2+α  1
Φ ∈ C ∞ h+ S , L h4+α S1 , h1+α S1 and F ∈ C ∞ h4+α
+ S ,h S . (2.14)

The above analysis also shows that Eq. (2.13) inherits a quasilinear structure.
To investigate well-posedness of (2.13) we can use the theory of abstract quasilinear evolution
equations of parabolic type developed by Amann [1]. A thorough knowledge of the linear part
Φ(ρ) is essential in order to apply this theory. For this, let E0 and E1 be Banach spaces such
that E1 is densely injected in E0 and let H(E1 , E0 ) denote the set of all A ∈ L(E1 , E0 ) such
that −A generates a strongly continuous analytic semigroup on E0 . Due to the fact that γ > 0,
as a special case of a more general result obtained in [15,16], we have
   3+α  1
Φ(ρ) ∈ H h4+α S1 , h1+α S1 for each ρ ∈ h+ S . (2.15)

We then have the following local existence, uniqueness and regularity result for (2.13).

3+δ 1
Theorem 2.1. Given ρ0 ∈ h+ (S ) and δ ∈ (α, 1), the problem (2.13) has a unique maximal
solution
 3+δ  1   
ρ ∈ C 0, t + (ρ0 ) , h+ S ∩ C 1 0, t + (ρ0 ) , h1+α S1 ∩ C ∞ 0, t + (ρ0 ) × S1 ,

where [0, t + (ρ0 )) denotes the maximal interval of existence. The map (t, ρ0 ) → ρ(t, ρ0 ) defines a
smooth semiflow on h+ 3+δ 1
(S ). If furthermore ρ0 ∈ h+ (S ), then ρ ∈ C([0, t + (ρ0 )), h+
4+δ 1 4+δ 1
(S ))∩
1 + 1+δ
C ([0, t (ρ0 )), h (S )). 1

Proof. The result on the existence of a unique maximal solution and (2.13) generating semiflow
on h3+δ (S1 ) follows from Theorem 12.1 in [1]. The fact that the solution ρ is smooth in spacial
and temporal variables is based on a bootstrapping argument in the scale hk+δ (S1 ), k ∈ N. We
refer to [16,17] for more details. 2

Returning to the original problem (1.1), we get the desired result and complete the proof of
Theorem 1.1.

3. Flat solutions

In this section we establish existence and numbers of flat stationary solutions of (1.1) and
investigate long-term behavior of transient solutions with flat initial data.
2918 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

The stationary form of (1.1) for a general stationary solution (ρs (x), σs (x, y), βs (x, y),
ps (x, y)) reads as follows
⎧
⎪ σ = λ1 σs + βs in Ωρs ,
⎪ s
⎪
⎪
⎪ βs = λ2 βs in Ωρs ,
⎪
⎪
⎪
⎪ p = −μ(σs − σ̃ − ιβs ) in Ωρs ,
⎨ ∂σ s ∂βs ∂ps
s
= 0, = 0, =0 on Γ0 , (3.1)
⎪
⎪ ∂y ∂y ∂y
⎪
⎪
⎪
⎪ σs = σ̄ , βs = β̄, ps = γ κρ on Γρs ,
⎪
⎪
⎪
⎩ ∂ps = 0 on Γρs .
∂ν

Let φ = 1 be the parameter defined in (1.3). If ρs (x) ≡ ρ∗ , with a positive constant ρ∗ , then the
equations in the first five lines of (3.1) can be solved as
⎧ √ √
⎪
⎪ cosh( λ1 y) β̄ cosh( λ2 y)
⎪
⎪ σ ∗ (y) = A 1 √ + √ ,
⎪
⎪ cosh( λ1 ρ∗ ) λ2 − λ1 cosh( λ2 ρ∗ )
⎪
⎪ √
⎨
cosh( λ2 y)
β∗ (y) = β̄ √ , (3.2)
⎪
⎪ cosh( λ2 ρ∗ )
⎪
⎪  √   √ 
⎪
⎪
⎪
⎪ μA1 cosh( λ1 y) μA2 cosh( λ2 y) 1 
⎩ p∗ (y) = 1− √ + 1− √ + μσ̃ y 2 − ρ∗2 .
λ1 cosh( λ1 ρ∗ ) λ2 cosh( λ2 ρ∗ ) 2

Substituting the expression of p∗ into the last equation in (3.1), we see that

f (η) = σ̃ . (3.3)

Here the function f is defined by

f (η) := A1 g(η) + A2 g(φη) (η > 0), (3.4)

where

tanh η 
g(η) := , η := λ1 ρ ∗ . (3.5)
η

Hence we have proved the following result:

Lemma 3.1. The problem (1.1) has flat stationary solutions if and only if Eq. (3.3) has positive
solutions. Moreover, each positive solution of (3.3) corresponds uniquely to one flat stationary
solution of (1.1).

To determine how many positive solutions that (3.3) admits, we need to study the behavior of
the function f (η) defined in (3.4) in more detail.

Lemma 3.2. The function g(η) defined in (3.5) is strictly monotone decreasing for η > 0. More-
over,
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2919

lim g(η) = 1, lim g(η) = 0,

η→0 η→∞

lim f (η) = A1 + A2 , lim f (η) = 0.

η→0 η→∞

Proof. It can be easily verified by using the explicit expressions of g and f . 2

Lemma 3.3. The function gg(φη)

(η)
is strictly monotone increasing (respectively, decreasing) for
η > 0 provided φ > 1 (respectively, 0 < φ < 1). Besides, for fixed φ > 0 we have

g (η) 1 g (η)
lim = , lim = φ2. (3.6)
η→0 g (φη) φ η→∞ g (φη)

Proof. We prove the first assertion by showing that gg(φη)

(η) is strictly monotone decreasing (re-
spectively, increasing ) for η > 0 provided φ > 1 (respectively, 0 < φ < 1). In fact, since
   
d g (φη) g (φη) φηg (φη) ηg (η)
= − ,
dη g (η) ηg (η) g (φη) g (η)

it is suffice to prove that

ηg (η)
is strictly monotone decreasing for η > 0. (3.7)
g (η)

Denoting

η cosh η − sinh η
m(η) := for η > 0,
η sinh η

we easily see that

1
g(η) = ,
ηm(η) + 1

and
ηg (η) 2η[2m (η) + ηm (η)] −2η[ηm (η) + m(η)]
= + = I (η) + II(η).
g (η) ηm (η) + m(η) ηm(η) + 1

Thus in order to prove (3.7), it is sufficient to prove that the functions I (η) and II(η) are both
strictly monotone decreasing for η > 0.
Firstly, we assert that

m(η)
is strictly monotone increasing for η > 0. (3.8)
ηm (η)

In fact, noticing

m(η) η cosh η sinh η − sinh2 η η cosh η − sinh η sinh η

= = · ,
ηm (η) sinh η − η
2 2 sinh η − η sinh η + η
2920 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

η cosh η−sinh η
by using Taylor’s expansion in η = 0 of the numerator of the derivative of sinh η−η , we
get that the function η cosh η−sinh η
sinh η−η is strictly monotone increasing for η > 0. Noticing that
η cosh η−sinh η sinh η
sinh η−η is positive, and that sinh η+η is strictly monotone increasing for η > 0, we find that
the assertion (3.8) holds true. Besides, from Lemma 2.3 in [11] we know that

ηm (η)
is strictly monotone decreasing for η > 0. (3.9)
m (η)

Observing the relation

ηm (η)
2η[2m (η) + ηm (η)] 2(2 + m (η) )
I (η) = =
ηm (η) + m(η) m(η)
1 + ηm (η)

and the fact that ηm (η) m(η)

m (η) , ηm (η) are both positive for all η > 0, we conclude from (3.8) and (3.9)
that I (η) is strictly monotone decreasing for η > 0.
On the other hand, similarly as the proof of (3.8), one can easily get that

 
2η(ηm (η) + m(η)) cosh η sinh η − η cosh2 η − η sinh2 η
II (η) = − = < 0 for all η > 0,
ηm(η) + 1 sinh2 η cosh2 η

which means that II(η) is also strictly monotone decreasing for η > 0. Thus we complete the
proof of (3.7) and get the desired assertion. The calculation of the limits (3.6) is standard. 2

Lemma 3.4. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(i) If (A1 , A2 ) ∈ Δ1 then f (η) > 0 for all η > 0.

(ii) If (A1 , A2 ) ∈ Δ2 then f (η) < 0 for all η > 0.
(iii) If (A1 , A2 ) ∈ Δ3 then there exists a unique η∗ > 0 such that f (η∗ ) = 0 and

f (η) > 0 for 0 < η < η∗ , f (η) < 0 for η > η∗ ,

so that

A∗ := f η∗ := max f (η) > 0. (3.10)
η>0

(iv) If (A1 , A2 ) ∈ Δ4 then there exists a unique η > 0 such that f (η ) = 0 and

f (η) < 0 for 0 < η < η , f (η) > 0 for η > η ,

so that

A := f η := min f (η) < 0. (3.11)
η>0
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2921

Proof. For definiteness we assume that A1 = 0. By differentiating (3.4) we get

 
g (η) φA2
f (η) = A1 g (η) + φA2 g (φη) = A1 g (φη) + .
g (φη) A1

Using Lemma 3.3 and the fact that g (η) < 0 for all η > 0 (see Lemma 3.2), the desired assertion
follows readily. 2

Under the help of Lemma 3.4, one can easily get the following result on existence and numbers
of positive solutions of (3.3):

Theorem 3.5. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(a) If (A1 , A2 ) ∈ Δ1 , then in the case A1 + A2 < σ̃ < 0 (3.3) has a unique positive solution η∗ ,
and in the case either σ̃  A1 + A2 or σ̃  0 (3.3) has no positive solution.
(b) If (A1 , A2 ) ∈ Δ2 , then in the case 0 < σ̃ < A1 + A2 (3.3) has a unique positive solution η∗ ,
and in the case either σ̃  A1 + A2 or σ̃  0 (3.3) has no positive solution.
(c) If (A1 , A2 ) ∈ Δ3 , then in the case either min(0, A1 + A2 ) < σ̃  max(0, A1 + A2 ) or σ̃ = A∗
(3.3) has a unique positive solution η∗ , in the case max(0, A1 + A2 ) < σ̃ < A∗ (3.3) has two
positive solutions η∗− , η∗+ (η∗− < η∗+ ), and in the case either σ̃  min(0, A1 + A2 ) or σ̃ > A∗
(3.3) has no positive solution.
(d) If (A1 , A2 ) ∈ Δ4 , then in the case either min(0, A1 + A2 )  σ̃ < max(0, A1 + A2 ) or σ̃ = A
(3.3) has a unique positive solution η∗ , in the case A < σ̃ < min(0, A1 + A2 ) (3.3) has two
positive solutions η∗− , η∗+ (η∗− < η∗+ ), and in the case either σ̃  max(0, A1 + A2 ) or σ̃ < A
(3.3) has no positive solution.

Then a complete description in regard to existence and numbers of flat stationary solutions of
(1.1) follows readily from Lemma 3.1 and Theorem 3.5. This proves the first part (existence of
flat stationary solutions) of Theorem 1.2.
In the following, we study transient solutions of (1.1) with flat initial data and prove the second
part (asymptotic behavior of transient solutions) of Theorem 1.2.

Theorem 3.6. Assume that φ = 1 and |A1 | + |A2 | = 0. Given a constant ρ0 ∈ (0, ∞), let
(ρ, σ, β, p) be the solution of (1.1) starting from ρ0 . Then (ρ, σ, β, p) is flat and exists glob-
ally. Moreover, we have the following asymptotic behavior of the free boundary ρ(t):

(i) If (A1 , A2 ) ∈ Δ1 then


∞, if σ̃  A1 + A2 , or A1 + A2 < σ̃ < 0 and ρ0 > ρ∗ ,
lim ρ(t) =
t→∞ 0, if σ̃  0, or A1 + A2 < σ̃ < 0 and ρ0 < ρ∗ .

(ii) If (A1 , A2 ) ∈ Δ2 then


∞, if σ̃  0,
lim ρ(t) = ρ∗ , if 0 < σ̃ < A1 + A2 ,
t→∞
0, if σ̃  A1 + A2 .

(iii) If (A1 , A2 ) ∈ Δ3 then

2922 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

(a) in the case A1 + A2 > 0,

⎧
⎪ ∞, if σ̃  0,
⎨
ρ∗ , if 0 < σ̃  A1 + A2 ,
lim ρ(t) = +
t→∞ ⎪
⎩ ∗,
ρ if A1 + A2 < σ̃ < A∗ and ρ0 > ρ∗− ,
0, if A1 + A2 < σ̃ < A∗ and ρ0 < ρ∗− , or σ̃ > A∗ ;

(b) in the case A1 + A2  0,

⎧
⎪ ∞, if σ̃  A1 + A2 , or A1 + A2 < σ̃  0 and ρ0 > ρ∗ ,
⎨ +
ρ∗ , if 0 < σ̃ < A∗ and ρ0 > ρ∗− ,
lim ρ(t) =
t→∞ ⎪ 0,
⎩ if 0 < σ̃ < A∗ and ρ0 < ρ∗− , or σ̃ > A∗ ,
or A1 + A2 < σ̃  0 and ρ0 < ρ∗ .

(iv) If (A1 , A2 ) ∈ Δ4 then

(c) in the case A1 + A2 > 0,
⎧
⎪
⎪ ∞, if σ̃ < A , or A < σ̃ < 0 and ρ0 > ρ∗+ ,
⎨
ρ∗ , if 0  σ̃ < A1 + A2 ,
lim ρ(t) = −
t→∞ ⎪
⎪ ∗,
ρ if A < σ̃ < 0 and ρ0 < ρ∗+ ,
⎩
0, if σ̃  A1 + A2 ;

(d) in the case A1 + A2  0,

⎧
⎪
⎪ ∞, if σ̃ < A , or A < σ̃ < A1 + A2 and ρ0 > ρ∗+ ,
⎨
or A1 + A2  σ̃ < 0 and ρ0 > ρ∗ ,
lim ρ(t) =
t→∞ ⎪
⎪ ρ−, if A < σ̃ < A1 + A2 and ρ0 < ρ∗+ ,
⎩ ∗
0, if σ̃  0, or A1 + A2  σ̃ < 0 and ρ0 < ρ∗ .

Proof. Let the positive constant ρ0 be given and let (ρ, σ, β, p) be the solution of (1.1) starting
from ρ0 . On the other hand, as above, we may construct an explicit flat solution in the following
way: Let ρe (t) denote the unique global solution of
 √ √ 
∂ρ(t) A1 tanh( λ1 ρ(t)) A2 tanh( λ2 ρ(t))
= μρ(t) √ + √ − σ̃ , ρ(0) = ρ0 , (3.12)
∂t λ1 ρ(t) λ2 ρ(t)

and set
√ √
cosh( λ1 y) β̄ cosh( λ2 y)
σe (y) = A1 √ + √ ,
cosh( λ1 ρe (t)) λ2 − λ1 cosh( λ2 ρe (t))
√
cosh( λ2 y)
βe (y) = β̄ √ ,
cosh( λ2 ρe (t))
 √   √ 
μA1 cosh( λ1 y) μA2 cosh( λ2 y) 1 
pe (y) = 1− √ + 1− √ + μσ̃ y 2 − ρe2 (t) .
λ1 cosh( λ1 ρe (t)) λ 2 cosh( λ2 ρe (t)) 2

Then (ρe , σe , βe , pe ) is a global flat solution of (1.1). By uniqueness of the solution of (1.1), we
get that (ρ, σ, β, p) = (ρe , σe , βe , pe ). Thus (ρ, σ, β, p) is flat and exists globally.
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2923

Besides, the asymptotic behavior of the free boundary ρ follows readily from Lemma 3.4,
(3.3)–(3.5), (3.12) and a standard Lyapunov function argument. 2

4. Asymptotic stability

In this section, we investigate asymptotic stability of flat stationary solutions of (1.1) under
perturbations in little Hölder space.
For simplicity of the statement, later on we shall use the notation (ρ∗ , σ∗ , β∗ , p∗ ) to de-
note a general flat stationary solution of (1.1). Due to the change of space variables, we see
that (ρ∗ , σ∗ (ρ∗ y), β∗ (ρ∗ y), p∗ (ρ∗ y)) forms a flat stationary solution of the transformed prob-
lem (2.2). Besides, it can be easily verified that

 2yhx 2h yhxx
∂A(ρ∗ )h v = − ∂12 v − 3 ∂22 v − ∂2 v, ∂H(ρ∗ )h = −hxx ,
ρ∗ ρ∗ ρ∗
 h  h
∂B0 (ρ∗ )h v = − 2 Υ0 ∂2 v, ∂B1 (ρ∗ )h v = −hx Υ1 ∂1 v − 2 Υ1 ∂2 v (4.1)
ρ∗ ρ∗

for h ∈ h2+δ (S1 ) and v ∈ h2+δ (Ω̄). To compute the linearization of (2.2) we put

ρ = ρ∗ + εξ(t, x), σ = σ∗ (ρ∗ y) + εΣ(t, x, y),

β = β∗ (ρ∗ y) + εΠ(t, x, y), p = p∗ (ρ∗ y) + εP (t, x, y),

where ξ , Σ, Π and P are new unknowns. Substituting these expressions into (2.2) and using
the fact that (ρ∗ , σ∗ (ρ∗ y), β∗ (ρ∗ y), p∗ (ρ∗ y)) is an equilibrium, we get the following linearized
problem:
⎧ 
⎪ A(ρ∗ )Σ = λ1 Σ + Π − ∂A(ρ∗ ) ξ, σ∗ (ρ∗ y) in (0, T ) × Ω,
⎪
⎪ 
⎪
⎪ A(ρ∗ )Π = λ2 Π − ∂A(ρ∗ ) ξ, β∗ (ρ∗ y) in (0, T ) × Ω,
⎪
⎪ A(ρ )P = −μ(Σ − ιΠ) − ∂A(ρ ) ξ, p (ρ y)
⎪
⎨ in (0, T ) × Ω,
∗ ∗ ∗ ∗
B0 (ρ∗ )Σ = 0, B0 (ρ∗ )Π = 0, B0 (ρ∗ )P = 0 on (0, T ) × S1 × {0}, (4.2)
⎪
⎪
⎪
⎪
⎪ Υ1 Σ = 0, Υ1 Π = 0, Υ1 P = γ ∂N (ρ∗ )ξ
⎪
on (0, T ) × S1 × {1},
⎪
⎪
⎩ ∂t ξ + B1 (ρ∗ )P = 0 on (0, T ) × S1 × {1},
ξ(0, ·) = ξ0 on S1 × {1}.

Note that [∂A(ρ∗ ) · ]v∗ and ∂N (ρ∗ ) are second-order differential operators.
For given ξ ∈ h4+δ (S1 ), similarly as in Section 2, solving the second equation in (4.2) and
the corresponding boundary conditions imposed on Π we get a unique solution Π ∈ h4+δ (Ω̄),
which is 2π -periodic in x. Substituting Π into the first equation in (4.2) and solving it with the
corresponding boundary conditions imposed on Σ , we get a 2π -periodic function Σ ∈ h4+δ (Ω̄).
Then substituting Σ and Π into the third equation in (4.2) and solving it with boundary condi-
tions B0 (ρ∗ )P = 0 and Υ1 P = γ ∂N (ρ∗ )ξ on (0, T ) × S1 × {1}, we get a function P ∈ h2+δ (Ω̄)
(observe that ∂H (ρ∗ )ξ belongs to h2+δ (S1 ), as pointed out above), which is also 2π -periodic
in x. We now can define a linear operator

Lξ(x) := B1 (ρ∗ )P , x ∈ S1 . (4.3)

2924 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

It can be easily verified that L ∈ L(h4+δ (S1 ), h1+δ (S1 )). Letting

Ψ (ρ) := Φ(ρ)ρ − F (ρ) for ρ ∈ h4+δ S1 , (4.4)

from (2.14) we know that Ψ ∈ C ∞ (h4+δ (S1 ), h1+δ (S1 )). The above construction also shows that
the derivative of Ψ at ρ∗ is given by L, i.e.,

DΨ (ρ∗ )ξ = Lξ. (4.5)

On the other hand, we have


DΨ (ρ∗ )ξ = Φ(ρ∗ )ξ + DΦ(ρ∗ )ξ ρ∗ − DF (ρ∗ )ξ,

where
   
DΦ(ρ∗ ) · ρ∗ , DF (ρ∗ ) ∈ L h4+δ S1 , h2+δ S1 .

Now the fact that Φ(ρ∗ ) ∈ H(h4+δ (S1 ), h1+δ (S1 )) (cf. Theorem 4.1 in [16]) and a well-known
perturbation result imply that:

Lemma 4.1. L ∈ H(h4+δ (S1 ), h1+δ (S1 )).

We shall represent the operator L as Fourier multiplication operator. In the sequel we always
employ the natural complexification in connection with spectral theory without distinguishing
this notationally. Since h4+δ (S1 ) is compactly embedded into h1+δ (S1 ), the resolvent (λI − L)−1
is a compact operator for every λ in the resolvent set of L. Therefore the spectrum of L, which
we denote by σ (L), consists of a sequence of isolated eigenvalues. Furthermore, we have the
following expressions.

Lemma 4.2. σ (L) = {λk ; k ∈ N}, where


μA1 k 2 + λ1  
λk = μσ̃ − μ(A1 + A2 ) − μσ̃ kρ∗ tanh(kρ∗ ) + √ tanh( λ1 ρ∗ ) tanh k 2 + λ1 ρ∗
λ1

μA2 k 2 + λ2  
+ √ tanh( λ2 ρ∗ ) tanh k 2 + λ2 ρ∗ + γ k 3 tanh(kρ∗ ), k = 0, 1, 2, . . . .
λ2
(4.6)

Proof. We consider Fourier expansions of ξ , Σ, Π and P :

∞

ξ(x) = a0 + (ak cos kx + bk sin kx),
k=1
∞
 
Σ(x, y) = A0 (y) + Ak (y) cos kx + Bk (y) sin kx ,
k=1
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2925

∞
 
Π(x, y) = E0 (y) + Ek (y) cos kx + Fk (y) sin kx ,
k=1
∞
 
P (x, y) = M0 (y) + Mk (y) cos kx + Nk (y) sin kx .
k=1

Substituting the expression of Π into the second equation in (4.2) and the corresponding bound-
ary conditions imposed on Π , and comparing coefficients of cos kx, sin kx for every k, we get
the following boundary value problem for Ek (y) and Fk (y) respectively:

−k 2 Ek (y) + ρ∗−2 Ek (y) = λ2 Ek (y) + ak dk (y),
Ek (0) = 0, Ek (1) = 0, for k = 0, 1, 2, . . . ,
 2
−k Fk (y) + ρ∗−2 Fk (y) = λ2 Fk (y) + bk dk (y),
Fk (0) = 0, Fk (1) = 0, for k = 1, 2, . . . ,

where
√ √
cosh( λ2 ρ∗ y)  2 y sinh( λ2 ρ∗ y)
dk (y) = 2ρ∗−1 β̄λ2 √ − β̄ λ2 k √ .
cosh( λ2 ρ∗ ) cosh( λ2 ρ∗ )

One can easily verify that solutions of these two problems are respectively given by

Ek (y) = ak Dk (y), k = 0, 1, 2, . . . ,
Fk (y) = bk Dk (y), k = 1, 2, . . . ,

where
 √
  cosh( k 2 + λ2 ρ∗ y)  y sinh( λ2 ρ∗ y)
Dk (y) = −β̄ λ2 tanh( λ2 ρ∗ )  + β̄ λ2 √ .
cosh( k 2 + λ2 ρ∗ ) cosh( λ2 ρ∗ )

Next, substituting the expressions of Π(x, y) and Σ(x, y) into the first equation in (4.2) and
the corresponding conditions imposed on Σ(x, y), we get that

−k 2 Ak (y) + ρ∗−2 Ak (y) = λ1 Ak (y) + Ek (y) + ak gk (y),
Ak (0) = 0, Ak (1) = 0, for k = 0, 1, 2, . . . ,
 2
−k Bk (y) + ρ∗−2 Bk (y) = λ1 Bk (y) + Fk (y) + bk gk (y),
Bk (0) = 0, Bk (1) = 0, for k = 1, 2, . . . ,

where
√ √
cosh( λ1 ρ∗ y) β̄ cosh( λ2 ρ∗ y)
gk (y) = 2ρ∗−1 λ1 A1 √ −1
+ 2ρ∗ λ2 √
cosh( λ1 ρ∗ ) λ2 − λ1 cosh( λ2 ρ∗ )
√ √
 y sinh( λ1 ρ∗ y)  β̄ y sinh( λ2 ρ∗ y)
− k λ1 A1
2
√ − k λ2
2
√ .
cosh( λ1 ρ∗ ) λ2 − λ1 cosh( λ2 ρ∗ )
2926 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

Solving these two problems and using the relations Ek (y) = ak Dk (y) and Fk (y) = bk Dk (y), we
get the following solutions:

Ak (y) = ak Gk (y), k = 0, 1, 2, . . . ,
Bk (y) = bk Gk (y), k = 1, 2, . . . ,

where
√ 
 y sinh( λ1 ρ∗ y)   cosh( k 2 + λ1 ρ∗ y)
Gk (y) = λ1 A1 √ − λ1 A1 tanh( λ1 ρ∗ ) 
cosh( λ1 ρ∗ ) cosh( k 2 + λ1 ρ∗ )
√ 
 β̄ y sinh( λ2 ρ∗ y)  β̄  cosh( k 2 + λ2 ρ∗ y)
+ λ2 √ − λ2 tanh( λ2 ρ∗ )  .
λ2 − λ1 cosh( λ2 ρ∗ ) λ2 − λ1 cosh( k 2 + λ2 ρ∗ )

Then substituting the expressions of Σ(x, y), Π(x, y) and P (x, y) into the third equation in
(4.2) and the conditions B0 (ρ∗ )P = 0 and Υ1 P = γ ∂H(ρ∗ )ξ on (0, T ) × S1 × {1}, we get
 
−k 2 Mk (y) + ρ∗−2 Mk (y) = −μ Ak (y) − ιEk (y) + ak hk (y),
(4.7)
Mk (0) = 0, Mk (1) = ak γ k 2 , for k = 0, 1, 2, . . . ,
 2 
−k Nk (y) + ρ∗−2 Nk (y) = −μ Bk (y) − ιFk (y) + bk hk (y),
(4.8)
Mk (0) = 0, Mk (1) = bk γ k 2 , for k = 1, 2, . . . ,

where
√ √
cosh( λ1 ρ∗ y) cosh( λ2 ρ∗ y)
hk (y) = −2μA1 ρ∗−1 √ − 2μA2 ρ∗−1 √ + 2μσ̃ ρ∗−1
cosh( λ1 ρ∗ ) cosh( λ2 ρ∗ )
√ √
k 2 μA1 y sinh( λ1 ρ∗ y) k 2 μA2 y sinh( λ2 ρ∗ y)
+ √ √ + √ √ − μσ̃ k 2 ρ∗ y 2 .
λ1 cosh( λ1 ρ∗ ) λ2 cosh( λ2 ρ∗ )

Noticing the inhomogeneous boundary conditions in (4.7) and (4.8), we split the solution of (4.7)
in the following way

Mk (y) = Ik (y) + II k (y) + III k (y),

where Ik , II k are the solutions of

 
−k 2 Ik (y) + ρ∗−2 Ik (y) = −μ Ak (y) − ιEk (y) + ak hk (y),
(4.9)
Ik (0) = 0, Ik (1) = 0, for k = 0, 1, 2, . . . ,
 2
−k II k (y) + ρ∗−2 II k (y) = ak γ k 4 y 2 − 2ak γ k 2 ρ∗2 ,
(4.10)
II k (0) = 0, II k (1) = 0, for k = 0, 1, 2, . . . ,

respectively, and

III k (y) = ak γ k 2 y 2 . (4.11)

 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2927

Solving (4.9), (4.10) and using the relations Ak (y) = ak Gk (y), Bk (y) = bk Gk (y), we get
√  √
ak μA1 tanh( λ1 ρ∗ ) cosh( k 2 + λ1 ρ∗ y) ak μA1 y sinh( λ1 ρ∗ y)
Ik (y) = √  − √ √
λ1 cosh( k 2 + λ1 ρ∗ ) λ1 cosh( λ1 ρ∗ )
√  √
ak μA2 tanh( λ2 ρ∗ ) cosh( k 2 + λ2 ρ∗ y) ak μA2 y sinh( λ2 ρ∗ y)
+ √  − √ √
λ2 cosh( k 2 + λ2 ρ∗ ) λ2 cosh( λ2 ρ∗ )
cosh(kρ∗ y)
− ak μσ̃ ρ∗ + ak μσ̃ ρ∗ y 2 , k = 0, 1, 2, . . . , (4.12)
cosh(kρ∗ )

and

cosh(kρ∗ y)
II k (y) = ak γ k 2 − ak γ k 2 y 2 , k = 0, 1, 2, . . . . (4.13)
cosh(kρ∗ )

Combining (4.11)–(4.13) we get the following solutions of (4.7) and (4.8):

Mk (y) = ak Hk (y), k = 0, 1, 2, . . . ,
Nk (y) = bk Hk (y), k = 1, 2, . . . , (4.14)

where
√  √
μA1 tanh( λ1 ρ∗ ) cosh( k 2 + λ1 ρ∗ y) μA1 y sinh( λ1 ρ∗ y) cosh(kρ∗ y)
Hk (y) = √  −√ √ − μσ̃ ρ∗
λ1 cosh( k 2 + λ1 ρ∗ ) λ1 cosh( λ1 ρ∗ ) cosh(kρ∗ )
√  √
μA2 tanh( λ2 ρ∗ ) cosh( k 2 + λ2 ρ∗ y) μA2 y sinh( λ2 ρ∗ y)
+ μσ̃ ρ∗ y 2 + √  −√ √
λ2 cosh( k 2 + λ2 ρ∗ ) λ2 cosh( λ2 ρ∗ )
cosh(kρ∗ y)
+ γ k2 .
cosh(kρ∗ )

Now substituting (4.14) into the expression of P (x, y) and using the definition of the opera-
tor L (see (4.3)), we have

∞

Lξ(x) = λ0 a0 + λk (ak cos kx + bk sin kx),
k=1

where
  √ 
μA1 k 2 + λ1   tanh( λ1 ρ∗ )
λk = √ tanh( λ1 ρ∗ ) tanh k 2 + λ1 ρ∗ − μA1 √ +1
λ1 λ1 ρ ∗
  √ 
μA2 k 2 + λ2   tanh( λ2 ρ∗ )
+ √ tanh( λ2 ρ∗ ) tanh k + λ2 ρ∗ − μA2
2 √ +1
λ2 λ2 ρ ∗
+ 2μσ̃ − μσ̃ kρ∗ tanh(kρ∗ ) + γ k 3 tanh(kρ∗ ), k = 0, 1, 2, . . . .
2928 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

Observing Eq. (3.3), i.e.,

√ √
tanh( λ1 ρ∗ ) tanh( λ2 ρ∗ )
σ̃ = A1 √ + A2 √ , (4.15)
λ1 ρ ∗ λ2 ρ ∗

a direct computation in L2 (S1 ) yields the desired assertion. This completes the proof. 2

To determine the position of each eigenvalue λk , we split it as

λk = λ0 (k) + γ k 3 tanh(kρ∗ ), k = 0, 1, 2, . . . , (4.16)

where

μA1 k 2 + λ1  
λ (k) := μσ̃ − μ(A1 + A2 ) +
0
√ tanh( λ1 ρ∗ ) tanh k 2 + λ1 ρ∗
λ1

μA2 k 2 + λ2  
+ √ tanh( λ2 ρ∗ ) tanh k 2 + λ2 ρ∗ − μσ̃ kρ∗ tanh(kρ∗ ). (4.17)
λ2

Lemma 4.3. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(i) If (A1 , A2 ) ∈ Δ1 , then in the case A1 + A2 < σ̃ < 0 there holds

λ0 (0) < 0, lim λ0 (k) > 0.

k→∞

(ii) If (A1 , A2 ) ∈ Δ2 , then in the case 0 < σ̃ < A1 + A2 there holds

λ0 (0) > 0, lim λ0 (k) < 0.

k→∞

(iii) If (A1 , A2 ) ∈ Δ3 , then

(a) in the case A1 + A2 > 0 there holds
⎧
⎪
⎪ λ0 (0) > 0, lim λ0 (k) < 0, if 0 < σ̃  A1 + A2 ,
⎪
⎨ k→∞
λ0 (0) < 0, 0
lim λ (k) > 0, if A1 + A2 < σ̃ < A∗ and ρ∗ = ρ∗− ,
⎪
⎪ k→∞
⎪
⎩ λ0 (0) > 0, lim λ0 (k) > 0, if A1 + A2 < σ̃ < A∗ and ρ∗ = ρ∗+ ;
k→∞

(b) in the case A1 + A2  0 there holds

⎧
⎪
⎪ λ0 (0) < 0, lim λ0 (k) > 0, if A1 + A2 < σ̃  0,
⎪
⎨ k→∞
λ0 (0) < 0, 0
lim λ (k) > 0, if 0 < σ̃ < A∗ and ρ∗ = ρ∗− ,
⎪
⎪ k→∞
⎪
⎩ λ0 (0) > 0, lim λ0 (k) > 0, if 0 < σ̃ < A∗ and ρ∗ = ρ∗+ .
k→∞

(iv) If (A1 , A2 ) ∈ Δ4 , then

 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2929

(c) in the case A1 + A2 > 0 there holds

⎧
⎪
⎪ λ0 (0) > 0, lim λ0 (k) < 0, if 0  σ̃ < A1 + A2 ,
⎪
⎨ k→∞
λ0 (0) > 0, 0
lim λ (k) < 0, if A < σ̃ < 0 and ρ∗ = ρ∗− ,
⎪
⎪ k→∞
⎪
⎩ λ0 (0) < 0, lim λ0 (k) < 0, if A < σ̃ < 0 and ρ∗ = ρ∗+ ;
k→∞

(d) in the case A1 + A2  0 there holds

⎧
⎪
⎪ λ0 (0) < 0, lim λ0 (k) > 0, if A1 + A2  σ̃ < 0,
⎪
⎨ k→∞
λ0 (0) > 0, lim λ0 (k) < 0, if A < σ̃ < A1 + A2 and ρ∗ = ρ∗− ,
⎪
⎪ k→∞
⎪
⎩ λ0 (0) < 0, lim λ0 (k) < 0, if A < σ̃ < A1 + A2 and ρ∗ = ρ∗+ .
k→∞

Proof. It should be observed that the conditions stated in (i)–(iv) ensure the existence of equi-
libriums of (2.13). It follows from (4.15) that
 
λ0 (0) = μσ̃ − μ(A1 + A2 ) + μA1 tanh2 ( λ1 ρ∗ ) + μA2 tanh2 ( λ2 ρ∗ )
 √ 
tanh( λ1 ρ∗ ) 
= μA1 √ + tanh ( λ1 ρ∗ ) − 1
2
λ1 ρ ∗
 √ 
tanh( λ2 ρ∗ ) 
+ μA2 √ + tanh2 ( λ2 ρ∗ ) − 1
λ2 ρ ∗
√ √ √ √ √ √
cosh( λ1 ρ∗ ) sinh( λ1 ρ∗ ) − λ1 ρ∗ cosh( λ2 ρ∗ ) sinh( λ2 ρ∗ ) − λ2 ρ∗
= μA1 √ √ + μA2 √ √
( λ1 ρ∗ ) cosh2 ( λ1 ρ∗ ) ( λ2 ρ∗ ) cosh2 ( λ2 ρ∗ )
cosh η sinh η − η cosh(φη) sinh(φη) − φη
= μA1 2
+ μA2
η cosh η φη cosh2 (φη)

= −μ A1 ηg (η) + A2 φηg (φη)
= −μηf (η), (4.18)
√
where η = λ1 ρ∗ > 0, and φ, f are defined in (1.3), (3.4), respectively. Besides, using (4.15)
again we can easily verify that

lim λ0 (k) = −μ(A1 + A2 − σ̃ ). (4.19)

k→∞

Then the desired assertion follows readily from (4.18), (4.19), Lemma 3.4 and Theorem 3.5. 2

Theorem 4.4. Assume that φ = 1 and |A1 | + |A2 | = 0. Then we have the following conclusions:

(I) Suppose that one of the following conditions is satisfied:

(a) (A1 , A2 ) ∈ Δ2 and 0 < σ̃ < A1 + A2 ,
(b) (A1 , A2 ) ∈ Δ3 , A1 + A2 > 0 and 0 < σ̃  A1 + A2 ,
(c) (A1 , A2 ) ∈ Δ4 , A1 + A2 > 0 and 0  σ̃ < A1 + A2 ,
2930 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

(d) (A1 , A2 ) ∈ Δ4 , A1 + A2 > 0, A < σ̃ < 0 and ρ∗ = ρ∗− ,

(e) (A1 , A2 ) ∈ Δ4 , A1 + A2  0, A < σ̃ < A1 + A2 and ρ∗ = ρ∗− .
Then there exists a constant γ ∗ > 0 depending only on A1 , A2 , σ̃ , μ and ρ∗ such that for
every γ > γ ∗ we have

λk  C(γ ) > 0 for all k = 0, 1, 2, . . . ,

where C(γ ) is a positive constant depending on γ . While for every 0 < γ < γ ∗ there exists
positive integer k0 = k0 (γ ) such that λk0 < 0.
(II) Suppose that one of the following conditions is satisfied:
(f) (A1 , A2 ) ∈ Δ3 , A1 + A2 > 0, A1 + A2 < σ̃ < A∗ and ρ∗ = ρ∗+ ,
(g) (A1 , A2 ) ∈ Δ3 , A1 + A2  0, 0 < σ̃ < A∗ and ρ∗ = ρ∗+ .
Then there exists a constant γ ∗  0 such that λk  C(γ ) > 0 for all k ∈ N provided γ > γ ∗ ,
where C(γ ) is a positive constant depending on γ .

Proof. Firstly, we give the proof of (I). In fact, each of the assumptions (a)–(e) and Lemma 4.3
imply that λ0 (0) > 0 and limk→∞ λ0 (k) < 0. Thus there exists a set Uk := {kj ∈ N: k1 < k2 <
k3 < · · · , j = 1, 2, 3, . . .} such that

λ0 (k)  0 ∀k ∈ Uk and λ0 (k) > 0 ∀k ∈ N \ Uk .

Obviously there holds kj  1 and kj → +∞. Using (4.16) and noticing that γ > 0, we see that

λk > 0 for all k ∈ N \ Uk . (4.20)

In the following, we study the property of λkj for kj ∈ Uk .

Denoting

M := − inf λ0 (kj )
kj ∈Uk

and noticing λ0 (k) is bounded, we can easily find that 0 < M < +∞. If γ > M
tanh ρ∗ then we have

M
λkj  −M + γ kj3 tanh(kj ρ∗ )  −M + k 3 tanh(kj ρ∗ ) > 0 for all kj ∈ Uk . (4.21)
tanh ρ∗ j

Besides, for any given kj ∈ Uk , λkj is strictly monotone increasing in γ , limγ →0 λkj = λ0 (kj )
and limγ →∞ λkj = ∞. Then there exists a constant γk∗j > 0 such that λkj > 0 if γ > γk∗j , and
λkj < 0 if 0 < γ < γk∗j . Noticing (4.21), one can easily verify that γk∗j  M
tanh ρ∗ . Denoting

 
γ ∗ := sup γk∗j ; kj ∈ Uk ,

we see that if γ > γk∗j then λkj > 0 for all kj ∈ Uk , while if 0 < γ < γk∗j then λkj < 0 for some
kj ∈ Uk . Combining this with (4.20) we get the desired assertion in (I).
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2931

Next, we consider the case (II). By Lemma 4.3 and the condition (f) or (g) we see that
λ0 (0) > 0 and limk→∞ λ0 (k) > 0. If λk > 0 for all k ∈ N then we choose γ ∗ = 0; while if λk  0
for some k ∈ N then we follow the same argument as above. This completes the proof. 2

We can now give the proof of Theorem 1.3.

Proof of Theorem 1.3. Denote

G(r) := Ψ (r + ρ∗ ) − Lr,

where Ψ and L are given by (4.4) and (4.3), respectively. It follows from the relations (2.14) and
(4.4) that G ∈ C ∞ (h+
4+δ 1
(S ), h1+δ (S1 )). Besides, from (4.5) we know that

G(0) = Ψ (ρ∗ ) = 0 and DG(0) = DΨ (ρ∗ ) − L = 0. (4.22)

Thus the problem (2.13) is equivalent to the following problem

∂t r + Lr + G(r) = 0, r(0) = r0 . (4.23)

Firstly, we give the proof for the case (A1 , A2 ) ∈ Δ2 and 0 < σ̃ < A1 + A2 . Let γ ∗ > 0 be the
threshold value given in (I) of Theorem 4.4 and assume that γ > γ ∗ . Noticing (4.22) and (4.23)
and invoking Lemma 4.1 and Theorem 4.4, we see that all the assumptions of Theorem 9.1.2
in [24] are satisfied. This means that there are positive constants ω, ε and M1 such that if r0 ∈
4+δ 1
h+ (S ) satisfying r0 4+δ < ε then the solution r of (4.23) exists globally and satisfies

r(t) 4+δ
 M1 e−ωt r0 4+δ , t  0.

Returning to the equivalent problem (2.13) this means that if ρ0 − ρ∗ 4+δ < ε then the solution
ρ of (2.13) exists globally and satisfies

ρ(t) − ρ∗ 4+δ
 M1 e−ωt ρ0 − ρ∗ 4+δ , t  0. (4.24)

From the construction we further have


β∗ = Q(ρ∗ ), σ∗ = R(ρ∗ ) and p∗ = −μS(ρ∗ ) R(ρ∗ ) − σ̃ − ιQ(ρ∗ ) + γ T (ρ∗ )P(ρ∗ )ρ∗ .

Thus the mean value theorem implies that there exists a constant C such that

β(t, ·) − β∗ 4+δ
= Q ρ(t) − Q(ρ∗ ) 4+δ
 C ρ(t) − ρ∗ 4+δ
∀t  0.

Combining this estimate with (4.24) we get

β(t, ·) − β∗ 4+δ
 Me−ωt , t  0.

The corresponding estimates for σ and p can be obtained similarly. If γ < γ ∗ then Theorem 9.1.3
in [24] and Theorem 4.4 imply that the equilibrium ρ∗ of (2.13) is unstable, which means that the
flat stationary solution (ρ∗ , σ∗ , β∗ , p∗ ) of (1.1) is also unstable. This argument can also be applied
2932 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933

to some other cases, e.g. (A1 , A2 ) ∈ Δ1 , A1 + A2 < σ̃ < 0, to prove that the corresponding
flat stationary solution is unstable, because under these conditions Lemma 4.3 shows that λ0 =
λ0 (0) < 0.
The proofs of other cases, e.g. (b)–(g) in Theorem 4.4, can be treated similarly. This completes
the proof of Theorem 1.3. 2

Remark 4.5. Throughout this paper we assumed that φ = 1 and |A1 | + |A2 | = 0. For complete-
ness we briefly discuss the special cases φ = 1 and |A1 | + |A2 | = 0, respectively.
(a) The case φ = 1, or equivalently λ1 = λ2 , can be handled by letting φ → 1 in the above
analysis. By L’Hospital’s rule we find that

β̄ηg (η)
f (η) = (σ̄ − ιβ̄)g(η) + .
2λ1

Existence and stability of flat stationary solutions of (1.1) follow by the same argument as above
applied to this limit function f .
β̄
(b) Suppose now that |A1 | + |A2 | = 0, or equivalently that σ̄ = λ2 −λ 1
= ιβ̄ (cf. (1.3)). Then
f ≡ 0, which implies that (2.13) has equilibria if and only if σ̃ = 0. Our analysis in Section 3
shows that any positive constant ρ∗ is an equilibrium of (2.13). Further, the proof of Theorem 3.6
shows that for given initial ρ0 ∈ (0, ∞), the flat solution ρ(t) of (2.13) is equivalent to ρ0 for
all t  0, which implies that ρ∗ is stable under flat perturbation. The spectrum of L consists of
λk = γ k 3 tanh(kρ∗ ), k ∈ N. Then center manifold argument (cf. Section 9.2 in [24]) shows that
ρ∗ is stable in h4+δ (S1 ) for all γ > 0, so that the corresponding flat stationary solution of (1.1)
is also stable under h4+δ -perturbation for all γ > 0.

Acknowledgments

This work on the parts of the first and the third authors is supported by China National Science
Foundation under the grant numbers 10471157 and 10771223. The first author also wishes to
acknowledge his sincere thanks to the faculty and staff of the Institute for Applied Mathematics
of the Leibniz University of Hannover for their hospitality during his visit there under the support
of the KaiSi Foundation of Sun Yat-Sen University. We would like to thank the referee very much
for the valuable suggestions.

References

[1] H. Amann, Nonhomogeneous linear and quasilinear elliptic and parabolic boundary value problems, in:
H.J. Schmeisser, H. Triebel (Eds.), Function Spaces, Differential Operators and Nonlinear Analysis, Teubner, 1993,
pp. 9–126.
[2] B. Bazaliy, A. Friedman, A free boundary problem for an elliptic parabolic system: Application to a model of tumor
growth, Comm. Partial Differential Equations 28 (3–4) (2003) 517–560.
[3] B. Bazaliy, A. Friedman, Global existence and asymptotic stability for an elliptic–parabolic free boundary problem:
An application to a model of tumor growth, Indiana Univ. Math. J. 52 (5) (2003) 1265–1304.
[4] H. Byrne, M. Chaplain, Growth of nonnecrotic tumors in the presence and absence of inhibitors, Math. Biosci. 130
(1995) 151–181.
[5] S. Cui, Analysis of a mathematical model for the growth of tumors under the action of external inhibitors, J. Math.
Biol. 44 (2002) 395–426.
[6] S. Cui, Analysis of a free boundary problem modeling tumor growth, Acta Math. Sin. (Engl. Ser.) 21 (2005) 1071–
1082.
 F. Zhou et al. / J. Differential Equations 244 (2008) 2909–2933 2933

[7] S. Cui, Well-posedness of a multidimensional free boundary problem modelling the growth of nonnecrotic tumors,
J. Funct. Anal. 245 (2007) 1–18.
[8] S. Cui, J. Escher, Well-posedness and stability of a multi-dimensional tumor growth model, submitted for publica-
tion.
[9] S. Cui, J. Escher, Asymptotic behavior of solutions of a multidimensional moving boundary problem modeling
tumor growth, Comm. Partial Differential Equations, in press.
[10] S. Cui, J. Escher, Bifurcation analysis of an elliptic free boundary problem modeling stationary growth of avascular
tumors, SIAM J. Math. Anal. 39 (2007) 210–235.
[11] S. Cui, A. Friedman, Analysis of a mathematical model of the effect of inhibitors on the growth of tumors, Math.
Biosci. 164 (2000) 103–137.
[12] S. Cui, A. Friedman, A free boundary problem for a singular system of differential equations: An application to a
model of tumor growth, Trans. Amer. Math. Soc. 355 (9) (2003) 3537–3590.
[13] D. Drasto, S. Höhme, A single-cell based model of tumor growth in vitro: Monolayers and spheroids, Max-Planck
Institute preprint series, 58, 2005.
[14] J. Escher, Classical solutions to a moving boundary problem for an elliptic–parabolic system, Interfaces Free
Bound. 6 (2004) 175–193.
[15] J. Escher, G. Simonett, Maximal regularity for a free boundary problem, NoDEA Nonlinear Differential Equations
Appl. 2 (4) (1995) 463–510.
[16] J. Escher, G. Simonett, Classical solutions for Hele–Shaw models with surface tension, Adv. Differential Equa-
tions 2 (1997) 619–642.
[17] J. Escher, G. Simonett, A center manifold analysis for the Mullins–Sekerka model, J. Differential Equations 143
(1998) 267–292.
[18] A. Friedman, B. Hu, Asymptotic stability for a free boundary problem arising in a tumor model, J. Differential
Equations 227 (2) (2006) 598–639.
[19] A. Friedman, B. Hu, Bifurcation from stability to instability for a free boundary problem arising in a tumor model,
Arch. Ration. Mech. Anal. 180 (2) (2006) 293–330.
[20] A. Friedman, F. Reitich, Analysis of a mathematical model for the growth of tumors, J. Math. Biol. 38 (1999)
262–284.
[21] A. Friedman, F. Reitich, On the existence of spatially patterned dormant malignancies in a model for the growth of
non-necrotic tumors, Math. Models Methods Appl. Sci. 11 (4) (2001) 601–625.
[22] J.B. Kim, R. Stein, M.J. O’Haxe, Three-dimensional in vitro tissue culture models for breast cancer—A review,
Breast Cancer Res. Tr. 149 (2004) 1–11.
[23] A.H. Kyle, C.T.O. Chan, A.I. Minchinton, Characterization of three-dimensional tissue cultures using electrical
impedance spectroscopy, Biophys. J. 76 (1999) 2640–2648.
[24] A. Lunardi, Analytic Semigroups and Optimal Regularity in Parabolic Problems, Birkhäuser, Basel, 1995.
[25] W. Mueller-Klieser, Three-dimensional cell cultures: From molecular mechanisms to clinical applications, Amer. J.
Cell Physiol. 273 (1997) 1109–1123.
[26] F. Zhou, S. Cui, Bifurcation for a free boundary problem modeling the growth of multi-layer tumors, Nonlinear
Anal. 68 (7) (2008) 2128–2145.
[27] F. Zhou, S. Cui, Well-posedness and stability of a multidimensional moving boundary problem modeling the growth
of tumor cord, Discrete Contin. Dyn. Syst., in press.
[28] F. Zhou, J. Escher, S. Cui, Bifurcation for a free boundary problem with surface tension modeling the growth of
multi-layer tumors, J. Math. Anal. Appl. 337 (2008) 443–457.