Article Ouafa V1

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Received: Added at production Revised: Added at production Accepted: Added at production

DOI: xxx/xxxx

ARTICLE TYPE

Convergence analysis of a discrete duality finite volume scheme for


a modified Patlak-Keller-Segel chemotaxis model

Soualhi Ouafa1 Rhoudaf Mohamed2,1 Author Three1

1
Faculty of sciences Moulay Ismail University, Abstract
Meknès, Morocco
2 In this paper, we present the discrete duality finite volume method (DDFV) applied to a model of (Patlak)
Faculty of sciences Moulay Ismail University,
Meknès, Morocco Keller-Segel modeling chemosensitive movements, this model consists of a coupled system of elliptic and
parabolic equations. Firstly, we prove the existence and uniqueness of the numerical solution to the proposed
Correspondence scheme., then we demonstrate some a priori estimate later we proved an compactness results next, we passing
Corresponding author Ouafa Soualhi, This is sample to the limit to prove the convergence of DDFV schemes, finally numerical simulations are performed to
corresponding address.
verify accuracy.
Email: ouafasoualhi@gmail.com

KEYWORDS
Present address
This is sample for present address text this is sample
chemotaxis, finite volume, numerical simulation, convergence analysis, a priori estimates.
for present address text.

1 INTRODUCTION

In 1953 and 1970, respectively, Patlak 1 , Keller and Segel 2 , were created as a classic model to describe the evolution over time of
cell density n(x; t) and of the chemical signal concentration variable S(x; t) assuming that the cells directly emit chemotaxis
which is directly diffused. Many theoretical and mathematical models have been created, but the most famous model is the
following classic system: (
∂n
∂t – div(∇n – χn∇S) = 0, on Ω × [0, T],
(1)
–div(∇S) – n + S = 0, on Ω × [0, T],
where χ > 0 is the chemotactic sensitivity function and Ω is a convex, bounded and open set of R2 and T > 0. The initial
conditions on Ω are given by:
n(x, 0) = n0 (x) ∈ L2 (Ω). (2)
Therefore, the system (1) is supplemented by the following boundary conditions on ∂Ω × [0, T]:

∇n.ν = 0, in ∂Ω × [0, T], (3)

∇S.ν = 0, in ∂Ω × [0, T], (4)


with ν is the unite vector.
Many different schemes have already been proposed for the classical Keller-Segel system, let us pose: the finite volume
schemes 3,4 , the mixed finite element method 5 , the finite element method in 6,7 and existence and uniqueness of DDFV schemes 8 .
In a recent work 8 , we proposed discrete duality finite volume schemes for the Keller-Segel model. In 8 , we focused on the
existence and uniqueness of the numerical solution and the study of numerical efficiency. The numerical experiments showed the

Abbreviations:DDFV, Discrete duality finite volume

Journal 2023;00:1–20 wileyonlinelibrary.com/journal/ © 2023 Copyright Holder Name 1


2 SOAULHI ET AL .

good convergence behavior of the schemes as well as good qualitative results. In the present article, we will now consider the
convergence analysis (when time and space steps go to 0) of a DDFV scheme for the Keller-Segel model.
The DDFV (Discrete Duality Finite Volume ) method presented by Hermeline 9 , Domelevo, Omnes 10 and Andreianov, Boyer,
Hubert 11 , it was extended to convection-diffusion 12 , nonlinear diffusion 13,14 , electro and magnetostatics 15 , miscible fluid flows
in porous media 16,17 , drift-diffusion and energy-transport models 18 , degenerate elliptic-parabolic seawater intrusion problem 19 ,
Stokes flows 20,21 . The following definition of weak solution to (1) makes sense.
Definition 1 (Weak solution). Under assumptions (2), (3) and (4), a weak solution to (1) is a couple (n, S) satisfying

n ∈ C([0, T]; L2 (Ω)) ∩ L2 (O, T; H 1 (Ω)), S ∈ L∞ (0, T; H 1 (Ω)),

and for all test function ψ ∈ C 1 (QT ),


Z Z
∇S.∇ψdxdt = (n – S)ψdxdt, (5)
QT QT
Z   Z
∂ψ
n – ∇n.∇ψ + χn∇S∇ψ dxdt + n0 ψ(0, x)dx = 0. (6)
QT ∂t Ω

With QT = Ω × (0, T).


The paper is organized as follows: In section 2 we presented the meshes of the DDFV method and the notations used in this
article, also we introduced the discrete operators (gradient, diffusion and convection operators) the discrete weak formulation, the
numerical schemes and some properties of the numerical schemes. then, in section 3 we demonstrated the maximum principle of
the DDFV solution then we establish some a priori estimates later we proved an compactness results using the Aubin-Simon
lemma after we passing to the limit for proved the convergence of DDFV scheme and finally we provide some numerical
experiments.

2 DISCRETE DUALITY FINITE VOLUME SCHEMES FOR MODIFIED KELLER-SEGEL


MODEL

2.1 Meshes and notations

Let Ω be a polygonal open bounded connected subset of Rd with d ∈ N∗ , and ∂Ω = Ω\Ω its boundary .
Following Hermeline 9 , Domelevo, Omnes 10 and Andreianov, Boyer, Hubert 11 , we consider a DDFV mesh which is a triple
T = (M, M∗ , D) described below.
Firstly, the primal mesh M is defined as the triplet (M, E, P), where:

• M is a finite family of nonempty open disjoint subset K of Ω (the control volume primal) such that Ω = ∪K∈M K, ∂M is
the set of all edges of control volumes that are included in ∂Ω. We denote by M the union M ∪ ∂M. ∂K = K\K is the
boundary of K, let mK = |K| > 0 is the measure of K and dK is the diameter of K
• E is the set of edges σ of the mesh, mσ is the measure of σ, Eint is the subset of interior edges. For all K ∈ M and σ ∈ EK
(EK is the subset of edges of K) , we denote by νK,σ the unite vector normal to σ outward to K.
• P = {(xK )K∈M ; xK ∈ K}, (xK is the barycentre of K if K ∈ M and the middle of K if K ∈ ∂M ), and DK,σ is the cone
with vertex xK and basis K.

Secondly, the dual mesh M∗ is defined as the triplet (M∗ , E ∗ , P∗ ), with:

• M∗ is a finite family of nonempty open disjoint subset K∗ of Ω (the control volume dual) such that Ω = ∪K∗ ∈M∗ K∗ . ∂M∗
is the set of all edges of control volumes that are included in ∂Ω. We denote by M∗ the union M∗ ∪ ∂M∗ . Let ∂K∗ = K∗ \K∗
be the boundary of K∗ , and mK∗ = |K∗ | > 0 the measure of K∗ and dK∗ the diameter of K∗ .
• Let E ∗ is the set of the edges σ ∗ , mσ∗ the measure of σ ∗ and Eint

the subset of interior edges of Ω. For all K∗ ∈ M∗ and
σ ∈ EK∗ (EK∗ is the subset of edges of K ) , we denote by νσ∗ ,K∗ the unite vector normal to σ ∗ outward to K∗ .
∗ ∗
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 3

• P∗ is the subset of points of Ω indexed by M∗ , we denote P∗ = {(xK∗ )K∗ ∈M∗ ; xK∗ ∈ K∗ }, we than note by DK∗ ,σ∗ the cone
with vertex xK∗ and basis K∗ .

Finally, We denote by D the sets of all diamonds D, with:

• DK = {D ∈ D/σ ∈ EK }.
• DK∗ = {D ∈ D/σ ∗ ∈ EK∗ }.
• Dint = {D ∈ D/σ ∈ Eint } with Eint = {σ ∈ E; σ ∈/ ∂Ω}.
• Dext = {D ∈ D/σ ∈ Eext } with Eint = {σ ∈ E; σ ∈/ Eint }.
• MD = {K ∈ M such that σ ∈ EK }.
• M∗D = {K∗ ∈ M∗ such that σ ∗ ∈ EK∗ }.
• mD measure of the diamond.
• For a diamond cell D recall that (xK , xK∗ , xL , xL∗ ) are the vertices of Dσ,σ∗ .
• τ the unite vector parallel to σ, oriented from K∗ to L∗ .
• τ ∗ the unite vector parallel to σ ∗ , oriented from K to L.
• αD the angle between τ and τ ∗ .
• νK,σ = –cosαD νσ∗ ,K∗ + sinαD τK,σ .
• dD the diameter of Dσ,σ∗ .

FIGURE 1 interior Diamond and exterior.

We assume that the diamonds cannot be flat: there exists a unique αT ∈]0; π2 [ such that sin(αT ) = min (αD ). We also consider
D∈D
the following property
mσ mσ ∗ mes(DK,σ )
≤ . (7)
2mD 3
The size of the mesh is h = size(T ) = max dD .
D∈D

2.2 Discrete operators and duality formula

Let define the following spaces.

• RT is a linear space of scalar fields constant on the cells of M and M∗ .

RT = {uT = ((uK )K∈M , (uK∗ )K∗ ∈M∗ ), with uK ∈ R, for all K ∈ M


and uK∗ ∈ R; for all K∗ ∈ M∗ }.

• (R2 )D is a linear space of vector fields constant on the cells of D.

(R2 )D = {ξD = (ξD )D∈D ; with ξD ∈ R2 ; for all D ∈ D}. (8)


ext
Similarly, we may define R∂M , R∂D , R∂D the spaces of scalar fields constant respectively on ∂M, D and Dext and (R2 )Dext
the space of vector field constant on Dext .
4 SOAULHI ET AL .

Now, we recall the definitions of the discrete gradient and the discrete divergence which was introduced respectively in 22 and 10 .
We also introduce some trace operators and scalar products.
Definition 2. Let

∇D :RT → (R2 )D ,
uT → ∇D uT = (∇D uT )D∈D ,

the discrete gradient, such that for all D ∈ D


(
uL∗ –uK∗
∇D uT .τK∗ ,L∗ = mσ ,
D uL –uK
∇ uT .τK,L = mσ∗ ,

equivalent to  
1 uL – uK uL∗ – uK∗
∇D uT = νσ,K + νσ∗ ,K∗ ,
sin(αD ) mσ ∗ mσ
using the propriety mD = 12 mσ mσ∗ sin(αD ), we have

1 
∇D u T =

(uL – uK )mσ νσ,K + (uL∗ – uK∗ )mσ∗ νσ∗ ,K∗ .
2mD
Then, the discrete divergence divT is defined by
Definition 3. The discrete divergence operator divT is a mapping from (R2 )D to RT defined for all ξD ∈ (R2 )D by
 ∗ ∗

divT ξD = divM ξD , 0 , divM ξD , div∂M ξD ,

such that 
M
 div (ξD ) = (divK (ξD ))K∈M ,



divM (ξD ) = (divK∗ (ξD ))K∗ ∈M∗ ,
 div∂M∗ (ξ ) = (div ∗ (ξ )) ∗


D K D K ∈∂M∗ ,

with
1 X
divK (ξD ) = mσ ξD .νσ,K for all K ∈ M,
mK
D∈DK
1 X
divK∗ (ξD ) = mσ∗ ξD .νσ∗ ,K∗ for all K∗ ∈ M∗ ,
mK ∗
D∈DK∗
 
1  X X mσ
divK∗ (ξD ) = mσ∗ ξD .νσ∗ ,K∗ + ξD .νσ,K  for all K∗ ∈ ∂M∗ .
mK ∗ 2
D∈DK∗ D∈DK∗ ∩Dext

Definition 4 (Convection term). Let divcT : (R2 )D × RT → RT the convection operator defined for all ξD ∈ (R2 )D and
vT ∈ RT by  
∗ ∗
divcT (ξD , vT ) = divcM (ξD , vT ) , 0 , divcM (ξD , vT ) , divc∂M (ξD , vT ) ,
such that 


 divcM (ξD , vT ) = (divcK (ξD , vT ))K∈M ,

divcM (ξD , vT ) = (divcK∗ (ξD , vT ))K∗ ∈M∗ ,


divc∂M (ξD , vT ) = (divcK∗ (ξD , vT ))K∗ ∈∂M∗ ,


Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 5

with
1 X
mσ (ξD .νσ,K )+ vK – (ξD .νσ,K )– vL
 
divcK (ξD , vT ) = for all K ∈ M,
mK
D∈DK
σ=K/L

1 X
mσ∗ (ξD .νσ∗ ,K∗ )+ vK∗ – (ξD .νσ∗ ,K∗ )– vL∗ for all K∗ ∈ M∗ ,
 
divcK∗ (ξD , vT ) =
mK ∗
D∈DK∗
σ ∗ =K∗ /L∗

1 X
mσ∗ (ξD .νσ∗ ,K∗ )+ vK∗ – (ξD .νσ∗ ,K∗ )– vL∗
 
divcK∗ (ξD , vT ) = (
mK ∗
D∈DK∗
σ ∗ =K∗ /L∗
X mσ 
(ξD .νσ,K )+ vK – (ξD .νσ,K )– vL for all K∗ ∈ ∂M∗ .

+
2
D∈DK∗ ∩Dext
σ=K/L

Where x+ = max(x, 0) and x– = max(0, –x).


Let us now define the scalar products < ., . >T on RT and < ., . >D on (R2 )D by
!
1 X X
< vT , uT >T = mK uK vK + mK∗ uK∗ vK∗ , for all uT , vT ∈ RT . (9)
2 ∗ ∗
K∈M K ∈M

X D
< ξD , φD >D = mD ξD .φD , for all ξD , φD ∈ R2 . (10)
D∈D

The corresponding norms are denoted by ∥.∥p,T and ∥.∥p,D for all 1 ≤ p ≤ +∞.
!1/p
1 X 1 X
∥uT ∥p,T = mK |uK |p + p
mK∗ |uK∗ | , for all uT ∈ RT and for all 1 ≤ p < +∞. (11)
2 2
K∈M K∗ ∈M∗

!1/p
X D
∥ξD ∥p,D = mD |ξD |p , for all ξD ∈ R2 and for all 1 ≤ p < +∞. (12)
D∈D
 
∥uT ∥∞,T = max max |uK |, max
∗ ∗
|uK∗ | , for all uT ∈ RT . (13)
K∈M K ∈M

∥ξD ∥∞,D = max |ξD |, for all ξD ∈ (R2 )D . (14)


D∈D

Let < ., . >∂Ω be the bilinear forme defined on RDext × R∂M by


X
< φD , v∂M >∂Ω = mσ φD vσ , for all φD ∈ RDext , for all v∂Ω = (vσ )σ∈∂M ∈ R∂M . (15)
D∈Dext

Let introduce two trace operators the first is γ T : nT ∈ RT → γ T (nT ) = (γL (nT ))L∈∂M , defined by
nK∗ + 2nL + nL∗
γL (nT ) = , for all L = [xK∗ ; xL∗ ] ∈ ∂M.
4
The second is γ D : φD ∈ (R2 )D → (φD )D∈Dext .

2.3 A penalization operator

Let us define the penalization operator using in this paper, recalled in 23 , this penalization is needed in order to ensure that the
two components of a discrete “double” function uT converge to the same limit.
6 SOAULHI ET AL .

Definition 5. Let β ∈]0, 2[. The penalization operator P T : RT → RT is defined for all uT ∈ RT , by
 ∗ ∗

P T uT = P M uT , P ∂M uT , P M uT , P ∂M uT ,
∗ ∗
with P M uT = (PK uT )K∈M , P ∂M uT = 0, P M uT = (PK∗ uT )K∗ ∈M∗ and P ∂M uT = (PK∗ uT )K∗ ∈∂M∗ such that
1 1 X
For all K ∈ M, PK uT = mK∩K∗ (uK – uK∗ ),
mK size(T )β
K∗ ∈M∗
1 1 X
For all K∗ ∈ M∗ , PK∗ uT = β
mK∩K∗ (uK∗ – uK ).
mK∗ size(T )
K∈M

Let :
1 X X 1
< P T uT , uT >T = mK∗ ∩K (uK – uK∗ )2 = ∥uh,M – uh,M∗ ∥2L2 (Ω)
2size(T )β 2size(T )β
K∗ ∈M∗ K∈M

2.4 Definition of the functional spaces for approximate solutions

In this subsection, let us define some notations extract in 16 , firstly, let define two different reconstructions based either on the
primal values or the dual values for a given vector uT = ((uK )K∈M , uK∗ )K∗ ∈M∗ ) ∈ RT on a DDFV mesh T of size h,
X X
uh,M = uK 1K and uh,M∗ = uK∗ 1K∗ .
K∈M K∗ ∈M∗

Then we associate the following approximate solution


1 1
uh = uh,M + uh,M∗ . (16)
2 2
The space of the approximate solutions is denoted by HT
 

1
 
T 1 X 1 X 
HT = u ∈ L (Ω)/ there exists uT = (uK )K∈M , (uK∗ )K∗ ∈M∗ ∈ R such that uh = uK 1K + uK∗ 1K∗ (17)
 h 2 K∈M 2 K∗ ∈M∗ 

For a function uh ∈ HT , we define its approximate gradient ∇h uh by


X
∇h uh = ∇D uT 1D .
D∈D

Then, the space of piecewise constant function on each diamond (approximate gradient) is denoted by HD with
( )
X
1 2 2 D
HD = Vh ∈ (L (Ω)) / there exists VD ∈ (R ) such that Vh = VD 1D .
D∈D

We define the spaces-time approximation HT ,δt and HD,δt as follow


1 k k
n o
HT ,δt = uh,δt ∈ L ([0, T] × Ω) such that uh,δt (t, x) = uh (x) for all t ∈ [tk–1 , tk ), with uh ∈ HT , for all 1 ≤ k ≤ N ,

1 2 k k
n o
HD,δt = Vh,δt ∈ (L ([0, T] × Ω)) such that Vh,δt (t, x) = Vh (x) for all t ∈ [tk–1 , tk ), with Vh ∈ HD , for all 1 ≤ k ≤ N .

1,p
Now, we define a discrete W –norms (1 ≤ p ≤ +∞) on HT . For all uh ∈ HT , we set
  p1
∥uh ∥1,p,T = ∥uh ∥pp,T + ∥∇D uT ∥pp,D for all 1 ≤ p < +∞,
1
∥uh ∥1,∞,T = ∥uh ∥∞,T + ∥∇D uT ∥T2 ,
∥uh ∥1,∞∗,T = ∥uh ∥1,∞,T + < P T uT , uT >T ,

∥uh ∥1,–1,T = max < vT , uT >T , for all vh ∈ HT verifing ∥vh ∥1,∞∗,T ≤ 1 ,
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 7

where the norms ∥.∥p,T and ∥.∥p,D for all 1 ≤ p ≤ +∞ have been defined by ((11) – (14)).
Then, we define the discrete L1 (0, T; W 1,p (Ω)), L∞ (0, T; W 1,∞ (Ω)) and L∞ (0, T; Lp (Ω)) – norms on HT ,δt for all 1 ≤ p < +∞.
Let uh,δt ∈ HT ,δt , we set
N–1
X
∥uh,δt ∥1;1,p,T = δt∥uk+1
h ∥1,p,T , for all 1 ≤ p < +∞,
k=0
∥uh,δt ∥∞;1,∞,T = max ∥uk+1
h ∥1,∞,T ,
k∈{0,...,N–1}
! p1
1 X k+1 p 1 X
∥uh,δt ∥∞;0,p,T = max mK |uK | + mK∗ |uk+1
K∗ |
p
, for all 1 ≤ p < +∞.
k∈{0,...,N–1} 2 2
K∈M K∗ ∈M∗

For all Vh,δt ∈ HD,δt and for all 1 ≤ p < +∞, let
! 1p
X
∥Vh,δt ∥(L∞ (0,T;Lp (Ω)))2 = max mD |Vhk+1 |p ,
k∈{0,...,N–1}
D∈D
N–1
! 1p
X X
∥Vh,δt ∥(Lp ((0,T)×Ω))2 = δt mD |Vhk+1 |p .
k=0 D∈D

2.5 The numerical scheme

A DDFV scheme for the discretisation of the problem (1) is given by the following set of equations:
For all K ∈ M and K∗ ∈ M∗ , let
Z Z
1 1
n0K = u0 (x)dx and n0K∗ = u0 (x)dx. (18)
mK K mK ∗ K ∗

At each time step k, the numerical solution will be given by (nk+1 k+1
T , ST ). Then, the scheme for (1) writes for all 0 ≤ k ≤ NT – 1
 k+1 k
nT –nT T D k+1 T k D n+1
 ∆t – div (∇ nT ) + divc (nT ∇ ST ) = 0,


–divT (∇D STk+1 ) + STk+1 = nkT . (19)

∇D nk .ν = ∇D Sk .ν = 0, for all D ∈ D .

T T ext

Whith ∇D , divT and divcT , are defined respectively by definitions 2, 3 and 4.

2.6 Proprieties of the schemes

The Discrete Duality Finite Volume methods are based on the discrete duality formula recalled in Lemma 1 and proved in 17 .
Lemma 1. For all (uT , wT ) ∈ (RT )2 , we have
  
 N
XT –1 X X N
XT –1 X X
D k+1
1  k+1
mσ∗ ∇D uk+1 k+1 



2
 δt m σ ∇ uT .νK,σ wK + δt T .νK∗ ,σ ∗ wK∗ =
k=0 K∈M σ∈EK k=0 K∗ ∈M∗ σ ∗ ∈E K∗
 N
XT –1 X N
X T –1
D k+1 D k+1 T
δt < γ D (∇D uk+1
 k+1
– δt m ∇ u .∇ w + T ).ν, γ (wT ) >∂Ω ,

D

 T K

k=0 D∈D k=0

where ν is the exterior unit normal to Ω, and < ., . >∂Ω is defin by (15).
8 SOAULHI ET AL .

Theorem 1 (Poincaré inequality). Let Ω be an open bounded connected polygonal of R2 and T a DDFV mesh of Ω. There exists
C0 > 0 depending only on Ω, such that for all uT ∈ RT , we have
Z 2
2
∥uT ∥22,T ≤ C0 ∥∇D uT ∥22,D + udx .
m(Ω) Ω

With m(Ω) is the measure of Ω.


1
R
Proof. Let mΩ (u) = m(Ω) Ω
u, then we have

∥uT ∥22,T ≤ ∥uT – mΩ (u)∥22,T + 2 (mΩ (u))2 m(Ω),

applying theorem.3.4 in 15 we get the proof.


We recall to 11 for the proof of the following lemma,
Lemma 2. Let Ω be an open bounded connected polygonal of R2 and T be a DDFV mesh of Ω. Then for all φ ∈ Cc∞ (Ω),
∇D φT is strangely converge o ∇φ in (L2 (Ω))2 , when h → 9.
Let us define two important lemmas to proving the convergence of the DDFV scheme, this lemmas 3 and 4 are respectively a
direct consequence of proposition 4.1 and proposition 4.2 in 17 .
Lemma 3. Let (Tm )m be a sequence of DDFV meshes satisfying hm = siz(Tm ) → 0 when m → ∞. We consider a sequence of
functions (wm )m with wm = whm ∈ HTm . If (∥wm ∥1,1,Tm )m is bounded, then there exists w ∈ L1 (Ω) such that, up to a sequence,

wm → w in L1 (Ω) when m → ∞.

Lemma 4. Let (Tm )m be a sequence of DDFV meshes satisfying hm = siz(Tm ) → 0 when m → ∞. We consider a sequence of
functions (wm )m with wm = whm ∈ HTm . If

wm → w in L1 (Ω) and ∥wm ∥1,–1,Tm → 0 when m → ∞,

then w = 0.
Theorem 2 ( 8 ). Let Ω be an open, bonded, connected, polygonal domain of R2 and let T be a discretization of Ω × (0, T) such
that
mσ mσ ∗ mes(DK,σ )
≤ . (20)
2mD 3
Let n0 ∈ L2 (Ω), n0 ≥ 0 in Ω. Then there exists a solution {(nk+1 k+1
T , ST ), 0 ≤ k ≤ NT – 1} to (18) and (19) satisfying,



 nkK ≥ 0 and nkK∗ ≥ 0 for all K ∈ M and K∗ ∈ M∗ , for all 0 ≤ k ≤ NT .

 !
 X X
1 k k
mK n K + mK ∗ n K ∗ =


2
K∈M K ∗ ∈M∗

 !
 X X
 1 0 0

 2

 mK n K + mK∗ nK∗ = ∥n0 ∥L1 (Ω) , for all 0 ≤ k ≤ NT .
K∈M K∗ ∈M∗

3 THE MAIN RESULTS

3.1 Maximum principle



Lemma 5.(For each fixed integer 0 ≤ k ≤ NT – 1, let nkT , STk be a sequence such that the DDFV scheme (18)(19) holds.
K ∈ M,
For all n0K ≥ 0 and n0K∗ ≥ 0, nk+1 k+1
K ≥ 0 and nK∗ ≥ 0.
K∗ ∈ M∗
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 9

(
K ∈ M,
Proof. Fixing k ∈ {0, ..., NT – 1}. Let us assume that for all , n0K ≥ 0 and n0K∗ ≥ 0. Proving that nk+1
K ≥ 0 and
K∗ ∈ M∗
nk+1
K∗ ≥ 0.
We consider 
nk+1 k+1
K = min (nL )
L∈M
nk+1 k+1
K∗ = ∗min ∗ (nL∗ ),
L ∈M
– k+1 –
Multiplying the first equation in (19) by –(nk+1
K ) and –(nK∗ ) , then we get
(
GK K K
1 + G2 + G3 = 0
∗ ∗ ∗
GK K K
1 + G2 + G3 = 0.

With
mK k+1 k  k+1 –
GK
1 =– nK – nK nK ,
δt
∗ mK ∗ –
GK nk+1 k
nk+1

1 =– K ∗ – nK ∗ K∗ ,
δt
X mσ k+1 –
GK mσ (nk+1 k+1 k+1 k+1
 
2 =– K – nL )νσ,K + mσ ∗ (uK∗ – uL∗ )νσ ∗ ,K∗ νσ,K nK ,
2mD
σ∈EK
X mσ ∗

k+1 –
GK mσ (nk+1 k+1 k+1 k+1
 
2 =– K – nL )νσ,K + mσ ∗ (nK∗ – nL∗ )νσ ∗ ,K∗ νσ ∗ ,K∗ nK∗ ,
2m D
σ ∗ ∈EK∗
X  k+1 –
GK
3 =– mσ (∇D STk+1 .νσ,K )+ nk+1 D k+1
K – (∇ ST .νσ,K ) nL
– k+1
nK ,
σ∈EK
∗ X –
GK mσ∗ (∇D STk+1 .νσ∗ ,K∗ )+ nk+1 D k+1 – k+1
nk+1

3 =– K∗ – (∇ ST .νσ ∗ ,K∗ ) nL∗ K∗ ,
σ∈EK

Firstly we have nk+1 k+1 k+1 k+1


K ≤ nL and nK∗ ≤ nL∗ , then we get

GK K
2 ≥ 0 and G2 ≥ 0. (21)

In other hands it’s easy to show that



GK K
3 ≥ 0 and G3 ≥ 0. (22)
Now, using (21) and (22) we get (
GK
1 ≤ 0,

GK
1 ≤ 0,
Then (
GK k+1 2 k+1 – k
1 = (nK ) + (nK ) nK ≤ 0,

GK k+1 2 k+1 – k
= (nK∗ ) + (nK∗ ) nK∗ ≤ 0,
– k+1 –
Finally we get (nk+1
K ) = 0 and (nK∗ ) = 0, then nT ≥ 0.

3.2 A priori estimate

In this Section, we prove some a priori estimates satisfied by the solution of the scheme (19).
N
Lemma 6. Let Ω be an open bounded connected polygonal domain of R2 and δt be a times steps such that T = δt is an integer.
Let D be a DDFV mesh of Ω. Assume ((2) – (4)) hold and that the scheme (19) has a solution (nk+1
T , S k+1
T ). Then, there exists
C1 > 0 and C2 > 0 depending only on Ω, such that

∥∇h nT ∥2(L2 ((0,T)×Ω))2 + ∥nT ∥∞,0,2,T ≤ C1 (23)

∥∇h ST ∥2(L2 ((0,T)×Ω))2 ≤ C2 (24)


10 SOAULHI ET AL .

Proof. This demonstration proceeds in two steps, firstly, we prove the inequality (23), then we prove to formula (24).
Step 1: The first equation in (19) equivalent to:


 for all K ∈ M and for all k = 0, 1, ...., N – 1:
 
nk+1 k X mσ 
K –nK
mσ (nk+1 k+1 k+1 k+1
 

  m K δt + K – nL ).νσ,K + mσ ∗ (nK∗ – nL∗ ).νσ ∗ ,K∗ .νσ,K
m
 
D
 
σ∈E


 XK

mσ (∇D Sk+1 .νσ,K )+ nk+1 D k+1


.νσ,K )– nk+1
 
+ K – (∇ S = 0,

L

 
 
 
σ∈EK
∗ ∗ (25)

  all K ∈ M andX
for for all k = 0, 1, ...., N – 1:
m∗σ 

 k+1 k
n –n
mσ (nk+1 k+1 k+1 k+1
 K ∗ K ∗

m + K – nL ).νσ,K + mσ ∗ (nK∗ – nL∗ ).νσ ∗ ,K∗ .νσ ∗ ,K∗
  ∗
K

δt

mD

 

σ∈EK∗


 X
m∗σ (∇D Sk+1 .νσ∗ ,K∗ )+ nk+1 D k+1
.νσ∗ ,K∗ )– nk+1
 
+ K∗ – (∇ S L∗ = 0.

 


 

σ ∗ ∈EK∗

We multiply the first (resp. second) equation of the DDFV scheme (25) by δtnk+1 k+1
K (resp. δtnK∗ ) and sum up over all the primal
(resp. dual) cells and the integers n. Adding together the obtained equations leads to

T1 + T2 + T3 = 0.

With
N
XT –1 X N
XT –1 X
T1 = mK (nk+1 k k+1
K – nK )nK + mK∗ (nk+1 k k+1
K∗ – nK∗ )nK∗ .
k=0 K∈M k=0 K∗ ∈M∗

 N
X T –1 X X mσ
∗ k+1
[mσ (nk+1 k+1 k+1 k+1

 T = δt K – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ∗ ,K∗ ].νσ,K nK
 2

 2mD
k=0 K∈M σ∈EK
N T –1
 X X X mσ ∗ ∗ k+1
+ δt [mσ (nk+1 k+1 k+1 k+1
K – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ∗ ,K∗ ].νσ ∗ ,K∗ nK∗ .




 2mD
k=0 K∗ ∈M∗ σ ∗ ∈EK∗

 N
X T –1 X X
mσ (∇D Sk+1 .νσ,K )+ nk+1 D k+1
.νσ,K )– nk+1
  k+1
 T = δt K – (∇ S nK
 3 L


k=0 K∈M σ∈EK
 NXT –1 X X
m∗σ (∇D Sk+1 .νσ∗ ,K∗ )+ nk+1 D k+1
.νσ∗ ,K∗ )– nk+1
 k+1
+ δt K∗ – (∇ S L∗ nK∗ .





k=0 K∗ ∈M∗ σ ∗ ∈EK∗

First of all, observe that


1 2 2
x(x – y) ≥ (x – y ); for all x, y ∈ R. (26)
2
Then (26) leads to
1 X  N 2  1 X 
(nK ) – (n0K )2 + (nNK∗ )2 – (n0K∗ )2 .

T1 ≥ (27)
2 2
K∈M K∗ ∈M∗
In other hands, we have
NT –1

X X mσ 
∗ k+1
mσ (nk+1 k+1 k+1 k+1 k+1
 
T2 =
 δt K – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ∗ ,K∗ .νσ,K (nK – nL )+
2mD


k=0 σ∈E
NT –1
 X X m∗σ  ∗ k+1
mσ (nk+1 k+1 k+1 k+1 k+1

δt K – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ∗ ,K∗ .νσ ∗ ,K∗ (nK∗ – nL∗ ).



 2mD
k=0 σ ∗ ∈E ∗

Since νσ∗ ,K∗ .τσ,K = 0 and νσ,K .τσ,K = sin(αD ), then we get:
(
2mD 1
 
nk+1 k+1
K – nL =
k+1 k+1 ∗
k+1 k+1
mσ sin(αD ) 2mD mσ (nK – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ,K
∗ ∗ .τσ,K

2mD D k+1
(28)
= mσ sin(αD ) ∇ n .τσ,K .
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 11

similarly νσ,K .τσ∗ ,K∗ = 0 and νσ∗ ,K∗ .τσ∗ ,K∗ = sin(αD ) gives:
(
2mD 1
 
nk+1 k+1
K∗ – nL∗ =
k+1 k+1 ∗
k+1 k+1
mσ∗ sin(αD ) 2mD mσ (nK – nL ).νσ,K + mσ (nK∗ – nL∗ ).νσ ,K
∗ ∗ .τσ ∗ ,K∗

2mD D k+1
(29)
= mσ∗ sin(αD ) ∇ n .τσ∗ ,K∗ .

Applying (28) and (29), we get


N T –1

X X 2mD
∇D nk+1 .νσ,K ∇D nk+1 .τσ,K

T2 =

 δt mσ
mσ sin(αD )

k=0 σ∈E
N T –1
 X X 2mD
+ δt ∇D nk+1 .νσ∗ ,K∗ ∇D nk+1 .τσ∗ ,K∗ .


 mσ ∗
 mσ∗ sin(αD )
k=0 σ ∗ ∈E ∗

We have νK,σ .τK,σ = sin(αD ) and νK∗ ,σ∗ .τK∗ ,σ∗ = sin(αD ), then
N
X T –1 X N
X T –1 X
T2 = 2 δt mD (∇D nk+1 )2 + 2 δt mD (∇D nk+1 )2 .
k=0 σ∈E k=0 σ ∗ ∈E ∗

That’s give
NT –1 NT –1
! NT –1
1X X
D k+1 2 1X X X
T2 = 4 δt mD (∇ n ) + δt mD (∇D nk+1 )2 ≥ δt∥∇D nk+1 ∥22,D . (30)
2 2 ∗ ∗
k=0 σ∈E k=0 σ ∈E k=0

Now, let demonstrate that T3 ≥ 0, for that, we have ,




 For all σ ∈ EK , and for all σ ∈ EK∗


∇D S.nσ,K = –∇D S.nσ,L

 ∇D S.n ∗ ∗ = –∇D S.n ∗ ∗ ,

σ ,K σ ,L

implies that 

 For all σ ∈ EK , and for all σ ∈ EK∗


(∇D S.nσ,K )+ = –(∇D S.nσ,L )–

 (∇D S.n ∗ ∗ )+ = –(∇D S.n ∗ ∗ )– .

σ ,K σ ,L

Then,
N T –1

X X
mσ (∇D Sk+1 .νσ,K )+ nk+1 D k+1
.νσ,K )+ nk+1
  k+1 k+1
T3 =


 δt K – (∇ S L (nK – nL )
k=0 σ∈E
 NXT –1 X
m∗σ (∇D Sk+1 .νσ∗ ,K∗ )+ nk+1 D k+1
.νσ∗ ,K∗ )+ nk+1
 k+1 k+1
+ δt K∗ – (∇ S L∗ (nK∗ – nL∗ ).




k=0 σ ∗ ∈E
That’s gives
N T –1

X X
mσ (∇D Sk+1 .νσ,K )+ nk+1 k+1
  k+1 k+1
T3 =


 δt K – nL (nK – nL )
k=0 σ∈E
N T –1
(31)
 X X
m∗σ (∇D Sk+1 .νσ∗ ,K∗ )+ nk+1 nk+1 (nk+1 nk+1

+ δt – – L∗ ) ≥0



 K∗ L∗ K∗
k=0 σ ∗ ∈E
Finally, Combining between (27), (30) and (31), one proves the energy estimate (23) follows as required.
12 SOAULHI ET AL .

Step 2:
The second equation in (19) equivalent to :


 for all K ∈ M and for all k = 0, 1, ...., N – 1:
  X mσ 

k+1 k+1 k+1 k+1


 
 mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ .νσ,K
2m

D



 σ∈EK

 
  +m Sk+1 – m nk+1 = 0,
K K K K
(32)

 for all K∗ ∈ M∗ and for all k = 0, 1, ...., N – 1:

  X m∗ 
σ k+1 k+1 k+1 k+1
 

 
 mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ .νσ ∗ ,K∗
2mD


σ∈EK∗



  k+1 k+1
+mK∗ SK ∗ – mK∗ nK∗ = 0.
 

k+1 k+1
We multiply the first (resp. second) equation of the DDFV scheme (32) by δtSK (resp. δtSK ∗ ) and summing up over all the

primal (resp. dual) cells and the integers k. Adding together the obtained equations leads to

T4 + T5 = T6 .

With 
 N
XT –1 X X mσ
 k+1 k+1 k+1 k+1 k+1
 T = δt [mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ ].νσ,K SK
 4

 2mD
k=0 K∈M σ∈EK
N T –1
 X X X mσ ∗ k+1 k+1 k+1 k+1 k+1
+ δt [mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ ].νσ ∗ ,K∗ SK∗ .




 2mD
k=0 K∗ ∈M∗ σ ∗ ∈EK∗
N
XT –1 X N
X T –1 X
k+1 k+1 k+1 k+1
T5 = δt mK SK SK + δt mK∗ SK ∗ SK∗ .

k=0 K∈M k=0 K∗ ∈M∗


N
XT –1 X N
X T –1 X
T6 = δt mK nk+1 k+1
K SK + δt mK∗ nk+1 k+1
K ∗ SK ∗ .
k=0 K∈M k=0 K∗ ∈M∗
Let
N T –1

X X mσ 
k+1 k+1 k+1 k+1 k+1 k+1
 
T4

 = δt mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ .νσ,K (SK – SL )+
2mD

k=0 σ∈E
 N
X T –1 X mσ ∗ 
k+1 k+1 k+1 k+1 k+1 k+1

δt mσ (SK – SL ).νσ,K + mσ∗ (SK ∗ – SL∗ ).νσ ∗ ,K∗ .νσ ∗ ,K∗ (SK∗ – SL∗ ).




∗ ∗
2m D
k=0 σ ∈E
Applying (28) and (29), we obtain:
NT –1

X X 2mD
∇D Sk+1 .νσ,K ∇D Sk+1 .τσ,K


 T4 = δt mσ
mσ sin(αD )


k=0 σ∈E
N T –1
 X X 2mD
+ δt ∇D Sk+1 .νσ∗ ,K∗ ∇D Sk+1 .τσ∗ ,K∗ .


 mσ ∗
 mσ∗ sin(αD )
k=0 σ ∗ ∈E ∗

Then
N
X T –1 X N
XT –1 X
D k+1 2
T4 = 2 δt mD (∇ S ) +2 δt mσ∗ mD (∇D Sk+1 )2 .
k=0 σ∈E k=0 σ ∗ ∈E ∗
That’s give
NT –1 NT –1
! N T –1
1X X 1X X X
T4 = 4 δt mD (∇D Sk+1 )2 + δt mσ∗ mD (∇D Sk+1 )2 ≥ δt∥∇D Sk+1 ∥22,D . (33)
2 2 ∗ ∗
k=0 σ∈E k=0 σ ∈E k=0

Let
N
X T –1

T5 ≥ δt∥STk+1 ∥22,T . (34)


k=0
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 13

Using this inequality x2 + y2 ≥ 2xy, for all x, y ∈ R we have


NT –1 NT –1
1X X
k+1 2 k+1 2 1X X
2 k+1 2
T6 ≤ δt mK ((nK ) + (SK ) ) + δt mK∗ ((nk+1
K∗ ) + (SK∗ ) )
2 2
k=0 K∈M k=0 K∗ ∈M∗
" N –1 N –1
#
T T
1X X k+1 2 1X X
k+1 2
≤ δt mK (nK ) + δt mK∗ (nK∗ ) +
2 2
k=0 K∈M k=0 K∗ ∈M∗
" N –1 NT –1
#
T
1X X
k+1 2 1X X
k+1 2
δt mK (SK ) + δt mK∗ (SK∗ )
2 2 ∗ ∗
k=0 K∈M k=0 K ∈M
N
XT –1 N
X T –1
2
≤ δt∥nk+1
T ∥2,T + δt∥STk+1 ∥22,T .
k=0 k=0

Applying the poincare inquality we obtain


N
X T –1 N
X T –1

T6 ≤ δtC3 ∥∇D nk+1 2


T ∥2,D
2
+ 2Nδt(mΩ (n)) m(Ω) + δt∥STk+1 ∥22,T . (35)
k=0 k=0

Finally, using (33), (34), (35) and (23) we get the proof of (24).

3.3 Convergence of the DDFV schemes

3.3.1 Compactness results

For give an a priori estimate on the discrete time derivation of the approximate density, let first define the discrete time derivation
∂t,T nh,δt ∈ Hh,δt by:
 k+1 k   k+1 k  
nK – nK nK∗ – nK∗
For a given function nh,δt ∈ HT δt ; ∂t,T nk+1h,δt = ; . (36)
δt K∈M δt K∗ ∈M∗

Proposition 1. Let Ω be an open bounded polygonal subset of R2 and let Tm be a family of discretisation of Ω × (0, T),
and (δtm )m≥1 be a sequence of time steps such that δtTm is an integer and δtm → 0 as m → +∞. The scheme (19) defines a
sequence of approximation solution (nm = nhm ,δtm , Sm = Shm ,δtm ) ∈ HTm ,δtm × HTm ,δtm , then, there exists n ∈ L∞ (0, T; H 1 (Ω)), S ∈
L∞ (0, T; L2 (Ω)), such that we have the following convergence results when m → ∞

nm → n,
 strongly in L2 (QT ),

∇hm n ⇀ ∇n, weakly in (L2 (Q ))2 ,

m T
2


 Sm ⇀ S, weakly in L (QT ),

∇ Sm ⇀ ∇S, weakly in (L2 (QT ))2 .
 hm

With QT = Ω × (0, T).


Proof. Let (nm , Sm )m>0 be a family of approximate solutions to (19), let φ ∈ HT and φK (resp. φK∗ ) the average of φ in the
primal control volume K (resp. Dual control volume K∗ ), nk+1 m := nm (., tk+1 ). Let multiplying the first equation (resp. the second
equation ) in (25) by δt2 φK ( resp. δt2 φK∗ ), and add up the two equations we get

T7 = T8 – T9 .
14 SOAULHI ET AL .

With
1
 X X

 T7 = 1
2 mK (nk+1 k
K – nK )φK + mK∗ (nk+1 k
K∗ – nK∗ )φK∗ ,

 2

 K∈M K∗ ∈M∗
δt

 X X X X
T8 = δt
mσ ∇D nk+1
T .νσ,K φK + mσ∗ ∇D nk+1
T .νσ ∗ ,K∗ φK∗ ,


2

 2
 K∈M σ∈EK K∗ ∈M∗
K∗ σ ∗ ∈E
X X
δt D k+1 + k+1 D k+1 – k+1
T9 = m ((∇ S .ν ) n – (∇ S T .νσ,K ) nL ).νσ,K φK




 2 σ T σ,K K

K∈M
Xσ∈EK X



+ δt2 mσ∗ ((∇D STk+1 .νσ∗ ,K∗ )+ nk+1 D k+1 – k+1

K∗ – (∇ ST .νσ ∗ ,K∗ ) nL∗ ).νσ ∗ ,K∗ φK∗ .

 

 

K∗ ∈M∗ σ ∗ ∈EK∗

Then, we have

δt X X δt X X
|T8 | = mσ ∇D nk+1
T .νσ,K φK + mσ∗ ∇D nk+1
T .νσ ∗ ,K∗ φK∗
2 2 ∗ ∗ ∗
K∈M σ∈EK K ∈M σ ∈EK∗
δt X δt X
≤ mσ ∇D nk+1
T .νσ,K (φK – φL ) + mσ∗ ∇D nk+1
T .νσ ∗ ,K∗ (φK∗ – φL∗ )
2 2 ∗
σ∈E σ ∈E
δt X D k+1 D δt X
≤ 2mD ∇ nT ∇ φT + 2mD ∇D nk+1 D
T ∇ φT
2 2 ∗ ∗
σ∈E σ ∈E
X
D k+1 D
≤ δt 2mD ∇ nT ∇ φT
D∈D
≤ 2δt∥nk+1
T ∥1,T ∥φT ∥∞,T ,

it is clear that ∥nk+1 k+1


T ∥1,T ≤ ∥nT ∥1,1,T and ∥φT ∥∞,T ≤ ∥φT ∥1,∞,T then we have

|T8 | ≤ 2δt∥nk+1
T ∥1,1,T ∥φT ∥1,∞,T ,

and we have ∥φT ∥1,∞,T ≤ ∥φT ∥1,∞∗T then


N
X T –1

|T8 | ≤ 2∥nh ∥1;1,1,T ∥φh ∥1,∞∗,T .


k=0

Now, let’s apply the same idea as in the demonstration of T3 > 0 (proof of lemma6), then there exists C3 > 0 such that:
 X

 |T9 | = | δt2 mσ ((∇D STk+1 .νσ,K )+ nk+1 D k+1 – k+1
K – (∇ ST .νσ,K ) nL ).νσ,K (φK – φL )



 σ∈E
X
δt
mσ∗ ((∇D STk+1 .νσ∗ ,K∗ )+ nk+1 D k+1 – k+1

+ K∗ – (∇ ST .νσ ∗ ,K∗ ) nL∗ ).νσ ∗ ,K∗ (φK∗ – φL∗ )|



 2
σ ∗X∈E



= | δt2 mσ (∇D Sk+1 .νσ,K )+ nk+1 k+1
 

K – nL (φK – φL )
 Xσ∈E
δt ∗ D k+1 + k+1 k+1
 



 + 2 m σ (∇ S .νσ ∗ ,K∗ ) n K ∗ – nL∗ (φK∗ – φL∗ )|
∗ ∈E

 σ


≤ C3 δt∥∇D nk+1 D k+1
T ∥2,D ∥∇ ST ∥2,D ∥φT ∥∞,T





≤ C3 δt∥∇D nk+1 D k+1
T ∥1,2,D ∥∇ ST ∥1,2,D ∥φT ∥1,∞,T .

Then we get
N
X T –1

|T9 | ≤ C3 ∥∇h nh ∥(L2 ((0,T)×Ω))2 ∥∇h Sh ∥(L2 ((0,T)×Ω))2 ∥φ∥1,∞∗,T .


k=0
From the last two inequalities of T8 and T9 , we deduce the existence of C4 > 0 such that:
N
XT –1

T7 ≤ C4 ∥φ∥1,∞∗,T ∥nh ∥1;1,1,T + ∥∇h nh ∥(L2 ((0,T)×Ω))2 ∥∇h Sh ∥(L2 ((0,T)×Ω))2 .




k=0
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 15

With C4 > 0 depending only on Ω. employing Holder’s inequality and the a priori estimate lemma 6, we get:
N
XT –1

T7 ≤ C4 ∥φ∥1,∞∗,T .
k=0

It gives the bound for ∥∂t nh,δt ∥1,–1,T ,


∥∂t nh,δt ∥1,–1,T ≤ c′ . (37)
In other hand, the a priori estimate (23) ensure that,
N
XT –1

δt∥nkh ∥1,1,T ≤ c (38)


k=0

With c and c′ depend on Ω. Using, (37), (38)and lemmas (3-4), with the application of theorem 3.4 recalled in 24 , we get nm
converge in L1 (0, T, L1 (Ω)) to a function n then
∞ 2 2 2 q p
nm → n weakly – ∗ in L (0, T; L (Ω)), and weakly in L (0, T; L (Ω)) and strongly in L (0, T; L (Ω)) with q < ∞ and p < 2.

Since (∇hm nm ) is bounded in L2 (QT ), there exists y ∈ L2 (QT ) such that:

∇hm nm → y weakly in (L2 (QT ))2 .

To prove that y = ∇n, let φ ∈ (Cc∞ (Ω))2 and


Z Z Z Z
hm
Gm = ∇ nm (x).φ(x)dx + nm (x)div(φ(x)dx → y(x).φ(x)dx + n(x)div(φ(x))dx when m → ∞,
Ω Ω Ω Ω

we define also φD , φσ and φσ∗ respectively the mean values of φ ovr D, σ and σ . Let φ
bD such that

φ
bD .νσK = φσ .νσK and φ
bD .νσ∗ K∗ = φσ .νσ∗ K∗ .

Firstly, φ is a smooth function, we have


1 2
bD | ≤
|φD – φ (|φD – φσ∗ | + |φD – φσ |) ≤ hm ∥∇φ∥L∞ (Ω) .
sin(αD ) sin(αD )
Then, we get
X √ 2hm
mD ∇D nTm .(φD – φ
bD ) ≤ ∥∇hm nm ∥L2 (Ω) mΩ ∥∇φ∥L∞ (Ω) ,
sin(αT )
D∈Dm
that’s implies
X
mD ∇D uTm .(φD – φ
bD ) → 0. (39)
D∈Dm
Secondly, we have
X 1 X X 1 X X
mD ∇D nTm .φ
bD = – nK mσ φ
bD .νσK – nK mσ ∗ φ
bD .νσ∗ K∗ .
2 2
D∈Dm K∈Mm D∈DK K∗ ∈M∗ D∈DK∗
m

The definition of φ
bD , the fact that φ has a compact support, and the Stokes formula leads to
Z Z Z
X 1 X 1 X
mD ∇D nTm .φ bD = – nK div(φ(x))dx – nK div(φ(x))d(x) = – nm (x)div(φ(x))dx. (40)
2 K 2 ∗ K∗ Ω
D∈Dm K∈Mm ∗ K ∈Mm

Now, Using (39) and (40) we get Gm → 0, then


Z Z
y(x).φ(x)dx → – n(x)div(φ(x))dx when m → ∞.
Ω Ω
16 SOAULHI ET AL .

That’s implies y = ∇n in the sense of distributions.


The bounded of Sm in L2 (0, T; H 1 (Ω)) implies the existence of a sequence, such that

Sm → S weakly in L2 (QT ),

∇hm Sm → z weakly in (L2 (QT ))2 ,


Again, it follows that z = ∇S in the sense of distributions.

3.3.2 Convergence of poisson and parabolic equations

Theorem 3. The nonnegative functions n and S defined in Lemma (1) satisfy the poisson and continuity equations in the sense
of (6) and (5) with Neumann condition.
Proof. Let ψ ∈ C ∞ ([0, T] × Ω), we define ψTk+1
m
associated to the discrete values
Z tk+1 Z
k 1 k+1
ψK = ψ(t, x)dxdt, for all K ∈ Mm and ψK = 0, for all K ∈ ∂M, for all k ∈ {0, ..., M – 1},
mK δt tk K
Z tk+1 Z
1
k
ψK ∗ = ψ(t, x)dxdt, for all K∗ ∈ Mm , for all k ∈ {0, ..., M – 1}.
mK∗ δt tk K ∗

We define also the corresponding function ψm and ∇m ψm .


k+1
Step 1: Since Sm is the solution of (5), let multiplying the first equation resp. second equation in (32) by δtψK resp. by
k+1 ∗ ∗
δtψK∗ and summing over K ∈ M resp. over K ∈ M and over k = 0, ..., M – 1 and adding the equations together, then we get

F1 + F2 = F3 ,

with  N
X T –1 X X
1
mσ ∇hm Smk+1 .νK,σ ψKk+1

F = – δt

1


 2

 k=0 K∈M σ∈EK

NT –1


 X X X
1
mσ∗ ∇hm Smk+1 .νK∗ ,σ∗ ψK
k+1




 – 2 δt ∗,

k=0 K∗ ∈M∗ σ ∗ ∈EK∗


 N
XT –1 X N
X T –1 X
 k+1 k+1 k+1 k+1



 F2 = δt m K S ψ
K,m K,m + δt mK∗ SK ∗ ,m ψK∗ ,m ,

 k=0 K∈M k=0 ∗
K ∈M∗

NT –1 NT –1


 X X X X
 F3 = δt mK nk+1 k+1
K,m ψKm + δt mK∗ mk+1 k+1
K∗ ,m ψK∗ ,m .



k=0 K∈M k=0 K∗ ∈M∗
Using the discrete duality formula Lemma (1), F1 rewrites
N
X T –1 X
F1 = δt mD ∇hm Smk+1 .∇hm ψmk+1 .
k=0 D∈D

We deduce Z T Z
F1 = ∇hm Sm .∇ψm .
0 Ω
Using lemma 2 we have the uniform convergence of ∇hm ψm to ∇ψ and the proposition 1 gives the weak convergence of ∇hm Sm
to ∇S in (L2 ((0, T) × Ω))2 implies that Z Z T
F1 → ∇S.∇ψ.
0 Ω
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 17

In other hand, we have


N
XT –1 X N
X T –1 X
k+1 k+1 k+1 k+1
F2 = δt mK SK,m ψK,m + δt mK∗ SK ∗ ,m ψK∗ ,m

k=0 K∈M k=0 K∗ ∈M∗


Z T Z Z T Z
1 1
= Sm,M (t, x)ψm,M (t, x)dtdx + Sm,M∗ (t, x)ψm,M∗ (t, x)dtdx
2 0 Ω 2 0 Ω

The uniform convergence of ψm,M and ψm,M∗ to ψ, the weak convergence of Sm,M and Sm,M∗ to S in L∞ (0, T; L2 (Ω)), give
Z TZ
F2 → Sψ. (41)
0 Ω

Similarly we prove that Z T Z


F3 → nψ. (42)
0 Ω
Passing to the limit in each term, we have proved (5).
step 2: Let ψ ∈ C ∞ ([0, T] × Ω), since nm is the solution of (25), we obtain

F4 + F5 + F6 = 0

with
NT –1 X NT –1 X
1X nk+1 – nk 1X nk+1∗ – nk ∗
F4 = mK K,m K,m ψm + mK∗ K ,m K ,m ψm ,
2 δt 2 ∗ ∗
δt
k=0 K∈M k=0 K ∈M
N T –1 X NT –1 X
1 X X 1X X
F5 = – mσ ∇m nm .νσ,K ψK,m
k+1
– mσ∗ ∇m nm .νσ∗ ,K∗ ψK
k+1
∗ ,m ,
2 2
k=0 K∈M σ∈EK k=0 K∗ ∈M∗ σ∈EK∗
N
X T –1 X X
mσ (∇m Sm .νσ,K )+ nk+1 m m – k+1
 k+1
F6 = K,m – (∇ S .νσ,K ) nL,m ψK,m +
k=0 K∈M σ∈EK
N
XT –1 X X
mσ∗ (∇m Sm .νσ∗ ,K∗ )+ nk+1 m m – k+1
 k+1
K∗ ,m – (∇ S .νσ ∗ ,K∗ ) nL∗ ,m ψK∗ ,m .
k=0 K∗ ∈M∗ σ ∗ ∈EK∗

Firstly we have
"
NT –1 X
#
T –1
1X ψ k+1 – ψKk , k+1 1 NX ψ k+1∗ – ψK
k
∗ ,m
mK K,m mK∗ K ,m
X
F4 = nK,m + nk+1
K∗ ,m
2 δt 2 ∗ ∗
δt
k=1 K∈M k=1 K ∈M
" #
1 X n0K,m 1 1 X n0K∗ ,m 1
+ mK ψ + mK ∗ ψK∗ ,m ,
2 δt K,m 2 ∗ ∗
δt
K∈M K ∈M

than we get
Z T Z Z
ψ(t + δt, x) – ψ(t, x)
F4 = – nm (t, x) dxdt – n0 (x)ψm (δt, x)dx.
0 Ω δt Ω

The function ψ is smooth and then we have the uniform convergence of ψ(.+δt,.)–ψ(.,.)
δt and ψm (δt, .) respectively to ∂t ψ and ψ(0, .),
and the weak convergence of nm to n in L∞ (0, T, L2 (Ω)) implies that:
Z TZ Z
F4 → – n∂t ψ – n0 ψ(0, .).
0 Ω Ω

Secondly, similarly to F1 we get Z T Z


F5 → ∇n.∇ψ.
0 Ω
18 SOAULHI ET AL .

Finally, let

F6 = F6,p + F6,d ,
k
with F6,p is the sum of a primal term, and F6,d is the sum of a dual term. Using ψK,m k
– ψL,m = mσ∗ ∇D ψTk m .τK,σ and x+ = x + x– ,
F6,p rewrites
NT –1
1X X
F6,p = δt mσ mσ∗ (∇D STk+1 .νσ,K )nkK ∇D ψTk+1 .τσ,K
2 m m
k=0 D∈D
N T –1
1 X X
+ δt mσ mσ∗ (∇D STk+1 .νσ,K )– (nkK – nkL )∇D ψTk+1 .τσ,K ,
2 m m
k=0 D∈D
RT R
Let set F6∗ = 0 Ω nm,D ∇m Sm .∇m ψm . We have nm,D weakly converge to n in L2 (0, T; L2 (Ω)) and ∇m Sm , ∇m ψm are weakly
converge respectively to ∇S and ∇ψ in (L2 (0, T; L2 (Ω)))2 , then
Z TZ
F6∗ → n∇S.∇ψ.
0 Ω

Now, let us prove that |F6∗ – F6 | tends to 0, we remark that F6∗ can be split into sum of a primal term F6,p
∗ ∗
and a dual term F6,d .
The primal term is
N
X T –1 X mD

F6,p = δt nk+1∗ (∇m STk+1 Nνσ,K )(∇m ψTk+1 .τσ,K ),
sin(θD ) D m m
k=0 D∈D
1 1
since we have ∇m STk+1
m
= m k+1
sin(θD ) (∇ STm .νσ,K ).τσ,K + m k+1
sin(θD ) (∇ STm .νσ ,K ).τσ ,K .
∗ ∗ ∗ ∗ Let

N T –1

X X mD
F6,p – F6,p = δt (∇m STk+1 .νσ,K )(∇m ψTk+1 .τσ,K )(nk+1 k+1
D,m – nK,m )
sin(θD ) m m
k=0 D∈D
NT –1
X X mD
– δt (∇m STk+1 .νσ,K )– (∇m ψTk+1 .τσ,K )(nk+1 k+1
K,m – nL,m ).
sin(θD ) m m
k=0 D∈D

Applying the relation nk+1 k+1 m k+1


K,m – nL,m = mσ ∗ ∇ nTm .τK,L and Cauchy-Schwarz inequality, we obtain

 N
X T –1 X mD
(∇m STk+1 .νσ,K )– (∇m ψTk+1 .τσ,K )(nk+1 k+1

 δt K,m – nL,m )
sin(θD ) m m
k=0 D∈D
 ≤ C√Th ∥∇Dm ψ k+1 ∥

∥∇m S ∥ ∞ m
2 ∥∇ n ∥ 2

m Tm ∞,Dm2 Tm (L (0,T;L (Ω))) Tm (L ((0,T)×Ω))2 .

The a priori estimates ((23) – (24)) and Lemma 3.5 in 11 , give


NT –1
X X mD
δt (∇m STk+1 .νσ,K )– (∇m ψTk+1 .τσ,K )(nk+1 k+1
K,m – nL,m ) ≤ Chm , (43)
sin(θD ) m m
k=0 D∈D

this term tends to 0. Using the same idea in (43) and the lemma 3.3 in 16 , we get
NT –1
X X mD
δt (∇m STk+1 .νσ,K )(∇m ψTk+1 .τσ,K )(nk+1 k+1
D,m – nL,m ) ≤ Chm . (44)
sin(θD ) m m
k=0 D∈D

Then, (43) and (44) implies that



F6,p – F6,p → 0

Similarly we prove that F6,d – F6,d → 0, finally we obtain
Z T Z
F6 → c∇S.∇ψ.
0 Ω
Convergence analysis of a discrete duality finite volume scheme for a modified Patlak-Keller-Segel chemotaxis model 19

Passing to the limit in each term, we have proved (6).

3.4 Numerical experiments

To close this section, we provide a numerical example that demonstrates the convergence properties of our schemes. For that we
consider the following Keller-segel system:

∂n
 ∂t – ∆n + div(n∇S) = f1
 in [0, T] × Ω,


–∆S + S = n + f
2 in [0, T] × Ω,
(45)


 ∇n.ν = ∇S.ν = 0 on [0, T] × ∂Ω,

n(., 0) = n0 in Ω.

With Ω = (0, 1), T = 1. We present the numerical results of the convergence of the DDFV scheme applied to a Keller-Segel
system 45 in a general mesh 2.

FIGURE 2 The mesh supported in this numerical test.

Let δt = 10–2 and the functions f1 and f2 are calculated using the following analytical solutions :
(
n(x, y, t) = tsin2 (πx)sin2 (πy)
.
S(x, y, t) = x2 (x – 1)2 y2 (y – 1)2

We present in the figure 3, the convergence error in norm L2 , L1 and L∞ for the density n.

AUTHOR CONTRIBUTIONS
ACKNOWLEDGMENTS
FINANCIAL DISCLOSURE
None reported.

CONFLICT OF INTEREST
The authors declare no potential conflict of interests.

REFERENCES
1. Patlak C. Random walk with persistence and external bias. Bull. Math. Biophys. 1953;15.
2. E. Keller E, Segel L. Initiation of slime mold aggregation viewed as an instability. J. Theor. Biol.. 1970;26.
3. Filbet F. A finite volume scheme for the Patlak-Keller-Segel chemotaxis model. Numer. Math. 2006;104.
4.
5. Marrocco A. Numerical simulation of chemotactic bacteria aggregation via mixed finite elements. Math. Mod. Numer. Anal.. 2003;4:617–630.
6. Saito N. Conservative upwind finite-element method for a simplified Keller-Segel system modelling chemotaxis. IMA J. Numer. Anal.. 2007;27:332–
365.
7. Saito N. Error analysis of a conservative finite-element approximation for the Keller-Segel model of chemotaxis. Comm. Pure Appl. Anal..
2012;11:339–364.
20 SOAULHI ET AL .

FIGURE 3 Spatial convergence in the norms L1 , L2 and L∞ compared to the size off the mesh at t = 10δt.

8. Soualhi O, Rhoudaf M. The finite volume method applied to the patlak-keller-segel chemotaxis model in a general mesh. Proceedings of the 1st
International Conference of Computer Science and Renewable Energies, ICCSRE. 2018:42-–49.
9. Hermeline F. A finite volume method for the approximation of diffusion operators on distorted meshes. J. Comput. Phys.. 2000;160(2):481–499.
10. Domelevo A, Omnes P. A finite volume method for the Laplace equation on almost arbitrary two-dimensional grids. M2AN Math. Model. Numer.
Anal.. 2005;39(6):1203–1249.
11. Andreianov B, Boyer F, F H. Discrete duality finite volume schemes for Leray-Lions type elliptic problems on general 2D-meshes. Num. Meth. for
PDEs. 2007;23:145–195.
12. Coudière Y, Manzini G. The discrete duality finite volume method for convectiondiffusion problems. SIAM J Numer Anal. 2010;47:4163—4192.
13. Coudière Y, Hubert F. A 3D discrete duality finite volume method for nonlinear elliptic equations. SIAM J Sci Comput. 2011;33:1739—1764.
14. Boyer F, Hubert F. Finite volume method for 2D linear and nonlinear elliptic problems with discontinuities. SIAM J Numer Anal. 2008;46:3032—
3070.
15. Delcourte S, Domelevo K, Omnes P. A discrete duality finite volume approach to Hodge decomposition and div-curl problems on almost arbitrary
two-dimensional meshes. SIAM J Numer Anal. 2007;45:1142—1174.
16. Chainais-Hillairet C, Krell S, Mouton A. Study of discrete duality finite volume schemes for the peaceman model. SIAM J Numer Anal.
2013;35(6):A2928–A2952.
17. Chainais-Hillairet C, Krell S, Mouton A. Convergence analysis of a DDFV scheme for a system describing miscible fluid flows in porous media.
Numerical Methods for Partial Differential Equations, Wiley. 2015;315(3):1098–2426.
18. Chainais-Hillairet C. Discrete duality finite volume schemes for two-dimensional driftdiffusion and energy-transport models. Internat J Numer
Methods Fluids. 2009;59:239–57.
19. Mandari M, Rhoudaf M, Soualhi O. Numerical resolution of a degenerate elliptic-parabolic seawater intrusion problem using discrete duality finite
volume. methods Journal of Mathematical Control Science and Applications. 2020;6(1).
20. Krell S. Stabilized DDFV Schemes for Stokes Problem with Variable Viscosity on General 2D Meshes. Numer Methods Partial Differential Eq.
2011:1666-–1706.
21. Krell S. Finite volume method for general multifluid flows governed by the interface Stokes problem. Math Models Methods Appl Sci. 2012;22(5).
22. Coudiere Y, Vila JP, Villedieu P. Convergence rate of a finite volume scheme for a two-dimensional convection-diffusion problem. M2AN Math.
Model. Numer. Anal.. 1999;33(5).
23. Andreianov B, Bendahmane M, Karlsen KH. Discrete duality finite volume schemes for doubly nonlinear degenerate hyperbolicparabolic equations.
J. Hyperbolic Differ. Equ.. 2010;7(1).
24. T. Gallouet T, Latché JC. Compactness of discrete approximate solutions to parabolic PDEs –Application to a turbulence model. Commun. Pure
Appl. Anal.. 2012;11:2371-2391.

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher’s website.

You might also like