Arbitrary Lagrangian-Eulerian discontinuous Galerkin

method for conservation laws: analysis and application

in one dimension
Christian Klingenberg, Gero Schn
ucke, Yinhua Xia

Abstract
In this paper, we develop and analyze an arbitrary Lagrangian-Eulerian discontinuous Galerkin (ALE-DG) method with a time-dependent approximation space for one
dimensional conservation laws, which satisfies the geometric conservation law. For the
semi-discrete ALE-DG method, when applied to nonlinear scalar conservation laws, a
cell entropy inequality, L2 stability and error estimates are proven. More precisely, we
prove the sub-optimal (k + 12 ) convergence for monotone fluxes, and optimal (k + 1)
convergence for an upwind flux, when a piecewise P k polynomial approximation space
is used. For the fully-discrete ALE-DG method, the geometric conservation law and
the local maximum principle are proven. Moreover we state conditions for slope limiters, which ensure total variation stability of the method. Numerical examples show
the capability of the method.

Key Words: Arbitrary Lagrangian-Eulerian discontinuous Galerkin method,

hyperbolic conservation laws, geometric conservation law, cell entropy inequality, error estimates, maximum principle, slope limiter conditions.

University of W
urzburg, Emil-Fischer-Str. 30, 97074 W
urzburg, Email: klingenberg@mathematik.uni-

wuerzburg.de

University

of

W
urzburg,

Emil-Fischer-Str.

31,

97074

W
urzburg,

Email:

gero.schnuecke@mathematik.uni-wuerzburg.de

Corresponding author. School of Mathematical Sciences, University of Science and Technology of

China, Hefei, Anhui 230026, P.R. China. Email: yhxia@ustc.edu.cn. Research supported by NSFC grant
No.11371342, No. 11471306.

Introduction

Grid deformation methods are unavoidable in many applications in fluid dynamics. For
instance, this kind of methods are used for aeroelastic analysis of wings in engineering (c.f.
Robinson et. al. [26]) or to describe stella-formation and galaxies in astrophysics (c.f. Keres
et. al. [17]). In this paper a grid deformation method based on a discontinuous Galerkin
(DG) discretization will be presented. To describe and analyze the method we consider the
following simple model problem:
t u + x f (u) = 0,

in (0, T ],

u(x, 0) = u0 (x),

x ,

(1.1a)
(1.1b)

with periodic boundary conditions. The set is an open interval in R, the initial data
u0 is considered to be periodic or compactly supported and f is a sufficiently smooth flux
function.
In order to describe the method, we assume that the grid points are explicitly given
for the upcoming time level, based on some grid moving methodology. Then the cells of
the partitions for the current and next time level can be connected by local affine linear
mappings. In the finite volume context a technique using a local affine mapping was used
by Fazio and LeVeque [11]. The mappings yield time depending test functions for the DG
discretization. Moreover the grid is static if the linear mappings are constant. In this case
the motion of a fluid is described by the Eulerian description of motion. On the other hand, it
is described by the Lagrangian description, if the linear mappings describing approximately
the motion of the particles in a fluid. Thus our method belongs to the class of arbitrary
Lagrangian-Eulerian (ALE) methods (c.f. Donea et. al. [8]). Thence we call our method
arbitrary Lagrangian-Eulerian discontinuous Galerkin (ALE-DG) method.
The Runge-Kutta DG method in the context of static grids has been developed and
analyzed by Cockburn, Shu et. al. in a series of papers (c.f. [4, 5, 6] and the review article
[7]). ALE-DG methods for equations with compressible viscous flows have been developed by
Lomtev et. al. [22], Persson et. al. [24] and Nguyen [23]. In their papers the focus lies on the
implementation and performance of the methods in aeroelastic applications. However there
are also some theoretical aspects about ALE methods in the literature. This discussions are
mostly about the significance of the geometric conservation law (GCL) for ALE methods.
This law governs the geometric parameters of a grid deformation method in such the way that
the method preserves constant states. The terminology GCL was introduced by Lombard
and Thomas in [21]. It is well known that there is a lack of stability in a grid deformation
2

method, if there is no GCL satisfied. For instance in [10] Grandmont, Guillard and Farhat
have proven that for monotone ALE methods the GCL is a necessary and sufficient condition
to obtain the local maximum principle for the method. Moreover in [13] Guillard and Farhat
have proven that the GCL is a necessary condition to ensure that the time discretization of
the method is high order accurate for ALE-finite volume methods. Further in [20] Lesoinne
and Farhat have analyzed the relevance and implementation of geometric conservation laws
for different ALE methods. They have shown that the GCL is not trivially satisfied for
ALE-finite element methods with a Runge-Kutta time discretization. Thus in particular for
ALE-DG methods, it is important to pay attention to the GCL. We are able to prove that
our ALE-DG method preserves constant states for any Runge-Kutta method. Therefore our
method satisfies the GCL.
It is well known that solutions of hyperbolic conservation laws are in general discontinuous, even if the initial data is chosen smooth. In high order methods for hyperbolic
conservation laws are discontinuities the cause of numerical artifacts like spurious oscillations. Without taming this artifacts a numerical method will become unstable. A possible
way to stabilize DG methods has been introduced by Cockburn and Shu in [3, 6]. They constructed slope limiters in such the way that the method stays high order accurate and the cell
average values of the DG solution become total variation stable. By following Cockburn and
Shus approach we obtain conditions for slope limiters, which stabilize our ALE-DG method.
Further in numerical test examples we show that our conditions are reasonable. Discontinuities are not the only source of instabilities in a numerical method. It is necessary that
the method preserves bounds. In general it is not easy to prove that a high order method
preserves bounds, even for methods on static grids. In [31] X. Zhang and Shu developed
a limiter for static grids, which ensures that the revised solution of a high order method
preserves bounds. We prove that this limiter works for our ALE-DG method too. Moreover
for scalar conservation laws we obtain the local maximum principle like X. Zhang and C.-W.
Shu for high order methods on static grids.
Another peculiarity of hyperbolic conservation laws is that weak solutions are in general
not unique. More precisely it is not clear that a weak solution is a physically relevant
solution. A weak solution has to satisfy an entropy inequality to be a physically relevant or
an entropy solution. For scalar conservation laws it is well known that there exists an unique
entropy solution (c.f. Kruzkov [18]). We prove that our semi-discrete ALE-DG method
satisfies a discrete version of the square entropy inequality for scalar conservation laws.
Thus in particular the method is L2 stable. Additionally we prove for smooth solutions of

scalar conservation laws the sub-optimal (k + 12 ) convergence for the semi-discrete ALE-DG
method with monotone numerical fluxes and the optimal (k + 1) convergence for the method
with an upwind numerical flux, if a piecewise P k polynomial approximation space is used.
For DG methods on static grids there are already many results in the literature about the a
priori error for smooth solutions of hyperbolic conservation laws. In the following we will list
a few results. The first a priori error estimate for a DG method has been proven by LeSaint
and Raviart [19]. In [16] Johnson and Pitkranta have proven that for linear conservation laws


the discontinuous Galerkin a priori error behaves as O hk+1 and in [25] Peterson has proven
that the result of Johnson and Pitkaranta is the optimal a priori error for any DG method for
hyperbolic conservation laws. Further nonlinear scalar conservation laws and symmetrizable
systems have been considered by Zhang and Shu in [28], [29] and [30]. They have proven for
DG methods with a second and third order total variationdiminishing (TVD)
Runge-Kutta

1

time discretization that the a priori error behaves as O hk+ 2 + (4t) , = 2, 3, in the


general case and O hk+1 + (4t) , = 2, 3, by applying an upwind numerical flux.

The organization of the paper is as follows. In Section 2 we develop our ALE-DG method
in one dimension. First we develop the semi-discrete ALE-DG scheme and prove the cell
entropy inequality as well as the L2 stability. Afterwards the error estimates are proven
for the method with monotone numerical fluxes and an upwind numerical flux. Then in
Section 2.4 we discuss the fully-discrete ALE-DG method. The geometric conservation law
and the local maximum principle are proven. Conditions for the slope limiter are derived
too. Section 3 contains numerical results for linear and nonlinear problems to demonstrate
the accuracy and capabilities of the method. Finally some concluding remarks are drawn in
Section 4.

The arbitrary Lagrangian-Eulerian

discontinuous Galerkin method

In this section we develop and analyze an arbitrary Lagrangian-Eulerian discontinuous

Galerkin (ALE-DG) method for solving conservation laws.

2.1

The semi-discrete ALE-DG discretization

In order to describe the method, wen needoto take the motion of then grid ointo account. We
N
N
assume that there are given points xnj 1
at time level tn and xn+1
at tn+1 , such
1
j
2

j=1

j=1

that
=

N h
[
j=1

xnj 1 , xnj+ 1
2
2

N h
i
[
n+1
and =
xn+1
.
,
x
j 1
j+ 1

j=1

Next we make the assumption that the points xnj 1 and xn+1
are connected by rays
j 1
2

xj 1 (t) = xnj 1 + j 1 (t tn ) ,
2

where

for all t [tn , tn+1 ] ,

xn+1
xnj 1
j 1

j 1 :=

tn+1 tn

(2.1)

(2.2)

The quantity j 1 describes the speed of motion in which the point xnj 1 moves to xn+1
.
j 21
2
h 2
i
The rays (2.1) provide for all t [tn , tn+1 ] time-dependent cells Kj (t) := xj 1 (t) , xj+ 1 (t) .
2

The length of a time-dependent cell is denoted by

4j (t) = xj+ 1 (t) xj 1 (t).
2

At this point it is useful to introduce some assumptions:

(1): For all j = 1, , N and t [tn , tn+1 ] holds
4j (t) = (j+ 1 j 1 )(t tn ) + 4j (tn ) > 0.
2

(2): There exists a constant C0 , such that

max
(x,t)[0,T ]

| (x, t)| C0 ,

where denotes the grid velocity field, which is in every cell Kj (t) given by
(x, t) = j+ 1
2

x xj 1 (t)
2

4j (t)

+ j 1
2

xj+ 1 (t) x
2

4j (t)

(2.3)

Next we define the length of the largest time-dependent cell as h(t) := max 4j (t). Moreover
1jN

for every time point the maximal cell length will be denoted by
h := max h (t) .

(2.4)

t[0,T ]

In addition we assume that the mesh is regular. Thus there exists a constant > 0 independent of h, such that
4j (t) h,

j = 1, , N.
5

(2.5)

The condition (1) of the grid velocity guarantees that the time-dependent cells Kj (t) are
well defined. Thus for any t [tn , tn+1 ] the time-dependent cells Kj (t) can be connected
with a reference cell [1, 1] by the mapping
j : [1, 1] Kj (t) ,

j (, t) =

4j (t)
( + 1) + xj 1 (t) .
2
2

(2.6)

The mapping yields a characterization of the grid velocity

t j (, t) = (j (, t) , t)

for all

(, t) [1, 1] [tn , tn+1 ] .

(2.7)

By the mapping a finite dimensional test function space can be defined




Vh := vh L2 () | vh j (, t) P k [1, 1] , t [tn , tn+1 ] and j = 1, , N , (2.8)
where P k ([1, 1]) denotes the space of polynomials in [1, 1] of degree at most k. The space
Vh contains discontinuous functions. Hence for a function vh Vh , we denote the left as well
as right limit, the cell average and the jump in a point xj 1 (t) as follows
2

vh,j
x
1 = vh
j 1
2

+
vh,j
1
2

{{vh }}j 1 :=
2

= vh

x+
j 12




(t) , t := lim vh xj 1 (t) , t ,
0




1
(t) , t := lim vh xj (t) + , t ,
0


1 +

vh,j 1 + vh,j
1
2
2
2

and [[vh ]]j 1 := vh,j

.
1 v
h,j 1
2

In addition for all v, w L2 (Kj (t)) we denote by (v, w)Kj (t) :=

R
Kj(t)

vw dx the L2 (Kj (t))

inner product. The following transport equation will be essential for the upcoming.
Lemma 2.1. Let u W1, (0, T ; H1 ()). Then for all vh Vh holds the transport equation





d
u, vh
= t u, vh
+ x (u) , vh
.
dt
Kj (t)
Kj (t)
Kj (t)

(2.9)

Proof. Let 0 (), , k () be a basis of the polynomial space P k ([1, 1]). Then in Kj (t)
the functions


2 x xj 1 (t)
2
1 ,
b` (x, t) :=
4j (t)

x Kj (t)

(2.10)

represent a basis of the test function space Vh . Furthermore a straight calculation provides




t b` (x, t) + (x, t) x b` (x, t) = 0, x Kj (t) .
(2.11)
6

Let u W1, (0, T ; H1 ()) and vh Vh . By (2.10) the function vh can be written as follows
vh (x, t) =

k
X

vh` b` (x, t) ,

x Kj (t) ,

`=0

where vh` R. Next by the identity (2.11) follows

x Kj (t) .

t (vh (x, t)) + (x, t) x (vh (x, t)) = 0,

(2.12)

Therefore by Reynolds transport theorem and (2.12) follows

d
(u, vh )Kj (t) = (t (uvh ), 1)Kj (t) + (x (uvh ), 1)Kj (t)
dt
= (t u, vh )Kj (t) + (x (u), vh )Kj (t) .

Finally with all these ingredients we can start to described the semi-discrete ALE-DG
method. The description of the ALE-DG method for the time interval [tn , tn+1 ] and the
cell Kj (t) starts by multiply the equation (1.1a) with an test function vh Vh . Next we
integrate the result over the cell Kj (t) and apply the transport equation (2.9). Then by
evaluating the resulting equation with an integration by parts we obtain



d

1
1
(uh , vh )Kj (t) = (g (, uh ) , x vh )Kj (t) g j+ , uh xj+ (t) , t vh,j+
1
2
2
2
dt



+
+ g j 1 , uh xj 1 (t) , t vh,j 1 ,
2

where g (, uh ) := f (uh ) uh and uh Vh is an unknown approximation to the solution

u of (1.1), which we try to determine by the ALE-DG method. Since uh is discontinuous in
the cell interface points xj+ 1 (t), we replace the flux g(j+ 1 , uh (xj+ 1 (t) , t)) by a numerical
2
2
2

+
flux gb j+ 1 , u
,
u
,
which
is
a
single
valued
function
defined
in the cell interface
h,j+ 1
h,j+ 1
2

points and depends on the values of the approximate

solution uh from

 both sides of the cell

+
interfaces. In general the numerical flux gb j+ 1 (t), uh,j+ 1 , uh,j+ 1 should be chosen as a
2

monotone numerical flux, which satisfies:

(b
g 1): Consistency: For any smooth function holds the identity gb (, u, u) = g (, u).
(b
g 2): Monotonicity: The numerical flux function gb (, , ) is increasing in the second argument and decreasing in the third argument.

(b
g 3): Lipschitz continuity: For all (a1 , b1 ) , (a2 , b2 ) R2 holds the inequality
+
|b
g (, a1 , b1 ) gb (, a2 , b2 )| L
g
b |a1 a2 | + Lg
b |b1 b2 | ,
+
where the Lipschitz constants L
g
b and Lg
b are independent of h.

Finally the semi-discrete ALE-DG method can be written as follows: Find a function
uh Vh , such that


d

+
(uh , vh )Kj (t) = (g (, uh ) , x vh )Kj (t) gb j+ 1 , u
vh,j+
,
u
1
h,j+ 12
h,j+ 12
2
2
dt


+

+
+ gb j 1 , uh,j 1 , uh,j 1 vh,j 1 .
2

(2.13)

for all vh Vh and j = 1, , N .

2.2

A cell entropy inequality and L2 stability

Weak solutions of equation (1.1) are in general not unique. The physically relevant unique
entropy solution can be found by the following entropy inequality
t (u) + x F (u) 0, in (0, T ]

(2.14)

in sense of distribution. The entropy : R R in (2.14) can be any convex function (c.f. Di
Perna [9]), if the flux function in (1.1) is convex. For a flux function f C 1 (R) the entropy
inequality (2.14) has to be true for all convex functions or has to be the so called Kruzkov
Ru 0
entropy (c.f. Kruzkov [18]). Further the entropy flux is given by F (u) :=
(v) f 0 (v) dv.
By integrating the entropy inequality (2.14) over the cell Kj (t) and applying the transport
equation (2.9) with vh = 1 we obtain

where uj 1





d
0
( (u) , 1)Kj (t) + F uj+ 1 j+ 1 uj+ 1
2
2
2
dt



F uj 1 j 1 uj 1 ,
(2.15)
2
2
2


:= u xj 1 (t) , t . In the following we show that in every cell Kj (t) the ALE-DG
2

method satisfies an inequality, which is for smooth functions consistent to (2.15). Thus we
have a cell entropy inequality for the ALE-DG method, like for DG methods on static grids
(c.f. Jiang and Shu [15]).
Proposition 2.2. The solution uh of the semi-discrete ALE-DG method given by (2.13)
satisfies the following cell entropy inequality


d
+
0
( (uh ) , 1)Kj (t) + H j+ 1 , u
,
u
h,j+ 21
h,j+ 12
2
dt

H j 1 , u
, u+
,
h,j 1
h,j 1
2

(2.16)

where (u) :=

u2
2

is the square entropy and

H , u , u

:=

f (v) dv +

2
+ gb , u , u+ u .
u
2

Proof. By applying the transport equation (2.9) the equation (2.13) can be written for all
test functions vh Vh as follows
0 = (t uh , vh )Kj (t) + (x (uh vh ) , 1)Kj (t) (f (uh ) , x vh )Kj (t)





+
+
+
1
+ gb j+ 1 , u
,
u
v
vh,j
,
u

g
b

,
u
1.
j
h,j+ 1
h,j 1
h,j+ 1
h,j+ 1
h,j 1
2

(2.17)

For equation (2.17) we can choose vh = uh as test function. Then by Reynolds transport
theorem follows



1
1d
(uh , uh )Kj (t) +
x u2h , 1 (f (uh ) , x uh )Kj (t)
2dt
2 



+
.
+ gb j+ 1 , uh,j+ 1 , uh,j+ 1 uh,j+ 1 gb j 1 , uh,j 1 , uh,j 1 u+
h,j 1

0=

(2.18)

Next we define the quantities

Z

f (v) dv

G (, u) :=

2
u
2

and






H , u , u+ := G , u + gb , u , u+ .
Then the equation (2.18) can be rewritten as follows


1d
+
(uh , uh )Kj (t) + H j+ 1 , u
,
u
h,j+ 12
h,j+ 12
2
2 dt


, u+
+ j 1 ,
H j 1 , u
h,j 1
h,j 1

0=

where







+
1,u
1,u

g
b

,
u
[[uh ]]j 1 .
j 1 := G j 1 , u+
1
1
1
1
j
j
h,j
h,j
h,j
h,j
2

The function G (, u) is differentiable in the second argument. Thus by the mean value
, u+
], such that
theorem exists a [u
h,j 1
h,j 1
2






G , u+

G
,
u
= g (, ) [[uh ]]j 1 .
1
1
h,j
h,j
2

Therefore by the properties (b

g 1) and (b
g 2) of the numerical flux gb (, , ) follows j 1 0.
2

We would like to mention that for static grids, which means = 0, we obtain the same
cell entropy inequality as in [15]. In addition the cell entropy inequality (2.16) implies the
L2 stability of the semi-discrete ALE-DG method.
Corollary 2.3. The solution uh of the semi-discrete ALE-DG method given by (2.13) satisfies for all t [0, T ] the inequality
kuh (t)kL2 () kuh (0)kL2 () .

2.3

A priori error estimates

In this section we present a priori error estimates for the ALE-DG method for smooth
solutions of (1.1). Thereby we follow the approach of Zhang and Shu (c.f. [28], [29] and [30])
and use Taylor expansion on the flux function g (, u) := f (u) u as well as an a priori
assumption given by
max ku uh kL () C1 h,

(2.19)

t[0,T ]

where the constant C1 is independent of uh and h. Furthermore we need to ensure that the
flux function f (u) and its derivatives are bounded. Since we consider scalar conservation laws
(1.1), the maximum principle guarantees that the flux function f (u) itself and up to third
order derivatives are bounded. In order to be able to evaluate the numerical flux function
gb (, , ) with Taylor expansion we proceed again as Zhang and Shu (c.f. [28]) and apply a
quantity b
a (b
g ; u) to measure the difference between the numerical flux function gb (, u , u+ )
and the flux g (, u). The quantity is for any piecewise smooth function v L2 () defined
by

[[v]]1 (g (, {{v}}) gb (, v , v + )) , if [[v]] 6= 0,

b
a (b
g ; v) :=
|g 0 (, {{v}})| ,
if [[v]] = 0.

(2.20)

This quantity was introduced by Harten in [14]. Moreover Zhang and Shu (c.f. [28]) have
proven the following lemma for the quantity above.
Lemma 2.4. Suppose the numerical flux function gb has the properties (b
g 1) - (b
g 3). Then for
any piecewise smooth function v L2 () the quantity
b
a (b
g ; v) given by (2.20) is non negative

and bounded by the constant C2 :=

1
2

+
L
g
b + Lg
b . In addition it holds the inequality

1 0
|g (, {{v}})| b
a (b
g ; v) + C3 |[[v]]| ,
2
where the constant C3 only depends on the maximum of |f 00 |.
10

2.3.1

Projections, interpolation properties and inverse inequalities

First of all we present two projections. The L2 projection Ph (u) of a function u L2 ()

into Vh is defined in such the way that for all vh Vh holds
(Ph (u) , vh )Kj (t) = (u, vh )Kj (t) .

(2.21)

In addition, if k 1, we can define Gauss-Radau projections Ph (u) of a function u L2 ()

into Vh in such the way that for all vh Vh with the property vh (j (, t)) Pk1 [1, 1]
holds
Ph (u) , vh

Kj (t)

= (u, vh )Kj (t)

(2.22a)

and




+
(t)
=
u
x
(t)
,
Ph+ (u) x+
j 1
j 1
2






Ph (u) x
(t)
=
u
x
(t)
.
j+ 1
j+ 1
2

(2.22b)

The L2 projection property (2.21) yields the following lemma.

Lemma 2.5. Let u L2 (), Ph (u) be the L2 projection of u and vh Vh . Suppose in Kj (t)
the function vh can be written as follows
vh (x, t) :=

k
X

vh` (t) b` (x, t) ,

`=0

where vh0 , ..., vhk H1 (0, T ) and b0 , ..., bk are given by (2.10). Then holds


u Ph (u) , t vh
= 0.

(2.22)

Kj (t)

Proof. By (2.3) and (2.6) in Kj (t) the grid velocity can be rewritten as
(j (, t), t) =


1
(1 )j 1 + (1 + )j+ 1 ,
2
2
2

[1, 1] .

Hence x vh Vh and therefore t vh Vh , since t b` = x b` , ` = 0, ..., k, by (2.11).

Thus by (2.21) yields (2.22).
In addition we apply the upcoming auxiliary lemma.
Lemma 2.6. Let u W1, (0, T ; H1 ()) and Qh be either Ph , Ph or Ph+ . Then holds












t Qh u + x Qh u = Qh t u + Qh x u .

11

(2.23)

Proof. In order to prove (2.23) we will apply Legendre polynomials. Each Legendre polynomial L` , ` = 0, , k, is an `th degree polynomial and can be expressed by Rodrigues
formula (c.f. Abramowitz and Stegun [1]). In addition the Legendre polynomials satisfying
(L` , L`0 )[1,1] =

2
``0 ,
2` + 1

L` (1) = (1)`

and L` (1) = 1.

Further the Legendre polynomials are an orthogonal basis of the space P k ([1, 1]). Therefore
the funktions




2 x xj 1 (t)
2
b` (x, t) := L`
L
1 ,
4j (t)

x Kj (t) ,

(2.24)

represent an orthogonal basis of the test function space Vh in Kj (t). Hence in Kj (t) the L2
projection of a function u L2 () can be written as follows
Ph (u) :=

k
X


b` (x, t) ,
c` (u, t) L

c` (u, t) :=

`=0

2` + 1
4j (t)




b` (x, t)
u, L

.
Kj (t)

Similar in Kj (t) the Gauss-Radau projections of a function u L2 () can be written as

follows
Ph

(u) :=

k
X

b` (x, t) ,
r` (u, t) L

`=0

where the coefficients are given by




2` + 1  b

v, L` (x, t)
,
r` (v, t) :=
4j (t)
Kj (t)

` = 0, , k 1

and
rk+

(u, t) := (1) u

x+
j 12

k1
 X
(t) , t
(1)k+` r`+ (u, t) ,
`=0

k1
 X
rk (u, t) := u x
(t)
,
t

r` (u, t) .
j+ 1

`=0

Now we can start to prove (2.23) Let u W1, (0, T ; H1 ()). A straight calculation
provides

 

 

d  

1
u xj+ 1 (t) , t = t u x
(t)
,
t
+

u
x
(t)
,
t
.
j+ 2 x
j+ 12
j+ 12
2
dt
The space derivative of the gird velocity (2.3) is given by
x ( (x, t)) =

j+ 1 j 1
2

4j (t)
12

40j (t)
=
,
4j (t)

x Kj (t) .

(2.25)

Therefore we obtain by the transport equation (2.9) and (2.11)




d 2` + 1  b 
2` + 1 
b
u, L`
=
t u + x (u) , L`
dt 4j (t)
4j (t)
Kj (t)
Kj (t)




1
1

2` + 1
j+ 2
j 2
b`
u, L

4j (t)
4j (t)
Kj (t)


2` + 1
b`
=
t u + x u, L
.
4j (t)
Kj (t)

(2.26)

Thus by (2.25) and (2.26) the time derivatives of the coefficients of the projections Ph (u)
and Ph (u) are given by
t (c` (u, t)) = c` (t (u) , t) + c` (x (u) , t)

(2.27)

t r` (u, t) = r` (t (u) , t) + r` (x (u) , t) .

(2.28)

and

Let Qh (u) be either Ph (u) or Ph (u). Then Qh (u) can be written as follows in Kj (t)
Qh (u) =

k
X

b` (x, t) ,
q` (u, t) L

`=0

where the coefficients q` (u, t) are c` (u, t) or r` (u, t). Hence we obtain by (2.27), (2.28) and
(2.10)
t (Qh (u)) =

k
X

b` (x, t)
t (q` (u, t)) L

`=0

k
X

q` (u, t) (x, t) x


b
L` (x, t)

`=0

= Qn,h (t (u)) + Qh (x (u)) x (Qh (u)) .

Further we will apply the following interpolation properties. For an arbitrary fixed function u Hk+1 () there exists constants C4 and C5 , which are independent of h, such that
2

ku Qh (u)k2L2 () C4 xk+1 u L2 () h2k+2 ,

(2.29)

u Qh (u) 2 2 C5 xk+1 u 2 2 h2k+1 ,
L ()
L ()

(2.30)

and

where we have applied the norm

kuk2L2 ()

N  
2  
2 
X





+
:=
u xj 1 (t) , t  + u xj 1 (t) , t  .
n,j=1

13

Moreover we will apply for all vh Vh the following inverse and trace inequality
h2 kx (vh )k2L2 () C6 kvh k2L2 () ,

h kvh k2L2 () C7 kvh k2L2 () ,

(2.31)

where the constants C6 and C7 are independent of h and vh . These inequalities can be proven
similar to well known results of basic approximation theory (c.f. Ciarlet [2]).
2.3.2

A suboptimal error estimate by using monotone fluxes

In this section we state an a priori error estimate for the semi-discrete ALE-DG method with
a general monotone numerical flux.


Theorem 2.7. Let u W1, 0, T ; Hk+1 () be the exact solution of equation (1.1), f
C 2 (R) and uh be the solution of the semi-discrete ALE-DG method (2.13) with a monotone
numerical flux gb. The initial data for the method is the L2 projection of the function u0 and
the grid velocity satisfies the conditions (1) as well as (2). Then there exists a constant
C independent of uh and h, such that there holds the error estimate
1

max keh kL2 () Chk+ 2 ,

t[0,T ]

where eh = u uh and h is given by (2.4).

Proof. We define the quantities
h := u Ph (u)

and h := uh Ph (u) .

(2.32)

Then the error function can be written as follows

eh := u uh = h h .

(2.33)

The exact solution u and the approximation solution uh satisfy the equation (2.13) and its
equivalent form (2.17). Hence equation (2.17) supplies the following error equation
0 = (t eh , vh )Kj (t) + (x (eh vh ) , 1)Kj (t) (f (u) f (uh ), x vh )Kj (t)





+
1
1
+ g j+ 1 , uj+ 1 vh,j+

,
u
1
j 2
j 2 vh,j 1
2
2
2
2




+

+
+
1,u
gb j+ 1 , u
v
b

,
u
vh,j
1,u
1
1 + g
1
1
1.
j
h,j+
h,j+
h,j+
h,j
h,j
2

(2.34)

Taylor expansion on the flux function f (uh ) up to second order provides the identity
1
f (uh ) = f (u) f 0 (u) eh + f 00 () (eh )2 ,
2
14

(2.35)

where is a value between u and uh . Since we assumed that the exact solution u of (1.1) is
sufficiently smooth, it holds for each cell interface point xj 1 (t) the equation
2

[[uh ]]j 1 = [[eh ]]j 1 .

2

(2.36)

Thus by applying (2.20) we obtain in each cell boundary point xj 1 (t)

2





+
g j 1 , uj 1 gb j 1 , u
1
1,u
h,j 2
h,j 2
2
2
2




=g j 1 , uj 1 g j 1 , {{uh }}j 1 + b
a (b
g ; uh )j 1 [[uh ]]j 1
2
2
2
2
2
2


2


1
{{eh }}j 1 b
a (b
g ; uh )j 1 [[eh ]]j 1 ,
=g 0 j 1 , uj 1 {{eh }}j 1 f 00 j 1
2
2
2
2
2
2
2
2

(2.37)

where j 1 is a value between uj 1 and {{uh }}j 1 .

2

Next we apply h as test function and sum the error equation (2.34). Then we obtain
by (2.35) and (2.37)
0=

N
X

(t eh , h )Kj (t) +

j=1

N
X

(x (eh h ) , 1)Kj (t)

j=1

(f 0 (u) eh , x h )Kj (t)

j=1

N
X


X 


1 X 00
f () (eh )2 , x h Kj (t)
g 0 j 1 , uj 1 {{eh }}j 1 [[h ]]j 1
2
2
2
2
2 j=1
j=1
N


2
X
1 X 00 
{{eh }}j 1 [[h ]]j 1 +
b
a (b
g ; uh )j 1 [[eh ]]j 1 [[h ]]j 1 .
+
f j 1
2
2
2
2
2
2
2 j=1
j=1

(2.38)

By applying the transport equation (2.9) and the properties (2.21) and (2.22) of the L2
projection the equation (2.38) can be rewritten as follows
1d
kh k2L2 () = a1 (h , h ) + a2 (eh , h ) + a3 (, h , h ) ,
2 dt
where
a1 (h , h ) :=

N
X

(f 0 (u) h , x h )Kj (t) ,

j=1
N

1 X 00
a2 (eh , h ) :=
f () (eh )2 , x h Kj (t)
2 j=1
N


2
1 X 00 
+
f j 1
{{eh }}j 1 [[h ]]j 1
2
2
2
2 j=1

15

(2.39)

and
N

a3 (, h , h ) :=

1X 0
f (u) , x (h )2 Kj (t)
2 j=1

N



1X

x (h )2 , 1 Kj (t)
2 j=1

N
X



g 0 j 1 , uj 1 {{eh }}j 1 [[h ]]j 1
2

j=1

N
X

b
a (b
g ; uh )j 1 [[eh ]]j 1 [[h ]]j 1 .
2

j=1

In the following we will estimate the quantities a1 (h , h ), a2 (eh , h ) and a3 (, h , h ).

Let us denoted the average value of the exact solution u of (1.1) in Kj (t) by
Z
1
uj (t) :=
u (x, t) dx.
4j (t) Kj (t)

(2.40)

The exact solution u of (1.1) satisfies a maximum principle. Hence u [m, M ] and we obtain
by the mean value theorem
max |f 0 (u (x, t)) f 0 (uj (t))| C1? 4j (t) C1? h,

(2.41)

xKj (t)

where C1? := max |f 00 (v)| and m as well as M are bounds of u0 (x) in . Therefore by the
mvM

property (2.21) of the L2 projection, Youngs inequality, the interpolation property (2.29),
the inverse inequality (2.31) and (2.41) follows
a1 (h , h ) =

N
X
j=1
C1?

((f 0 (uj (t)) f 0 (u)) h , x h )Kj (t)

C1? 2
h kx (h )k2L2 ()
2
2
C1? C4 2k+2
C1? C6
2

h
kukL (0,T ;Hk+1 ()) +
kh k2L2 () .
2
2

kh k2L2 () +

(2.42)

Next Youngs inequality, the a priori assumption (2.19), the interpolation property (2.29)
and the inverse inequality (2.31) provide
N
N


1 X 00
C ? C1 X
f () (eh )2 , x h Kj (t) 1 h
(|eh | , |x h |)Kj (t)
2 j=1
2
j=1


?
?
C C1
C C1
kh k2L2 () + kh k2L2 () + 1 h2 kx h k2L2 ()
1
2
4
C1? C1 C4 2k+2
C1? C1
2

h
kukL (0,T ;Hk+1 ()) +
(2 + C6 ) kh k2L2 () .
2
4

16

(2.43)

In a similar way Youngs inequality, the a priori assumption (2.19), the interpolation property
(2.30) and the trace inequality (2.31) yields
N
N



2
1 X 00 
C1? C1 X 


f j 1
{{eh }}j 1 [[h ]]j 1
h
{{eh }}j 1  [[h ]]j 1 

2
2
2
2
2
2 j=1
2
j=1


2 
2  3



 + 
 
 + 2  2
?
?
C1 C1 h h,j 1  + h,j 1  + C1 C1 h h,j 1  + h,j 1 
2
2
2
2
2
3
C1 C1 C5 h2k+2 kuk2L (0,T ;Hk+1 ()) + C1? C1 C7 kh k2L2 () .
2

(2.44)

Thus by (2.43) and (2.44) follows



2k+2

a2 (eh , h ) CI h

kh k2L2 ()

(2.45)

where the constant CI is independent of uh , h and t [0, T ].

For all piecewise continuous functions v, w L2 () holds in the discontinuities
[[vw]] = {{v}}[[w]] + [[v]]{{w}}.

(2.46)

N
N
X


1X
2

x (h ) , 1 Kj (t) =
j 1 {{h }}j 1 [[h ]]j 1 .
2
2
2
2 j=1
j=1

(2.47)

This identity supplies

Further by an integration by parts and (2.46) follows

N


X 


1X 0
f 0 uj 1 {{h }}j 1 [[h ]]j 1 ,
f (u) , x ((h ))2 Kj (t) C2? kh k2L2 ()
2
2
2
2 j=1
j=1

(2.48)

where C2? := max max |x (f 0 (u (x, t)))|.

t[0,T ] x

Hence by Youngs inequality, (2.47) and (2.48) follows

a3 (, h , h ) =

N
X

j 1 , uj 1 {{h }}j 1 [[h ]]j 1 +

2

j=1

N
X

b
a (b
g ; uh )j 1 [[eh ]]j 1 [[h ]]j 1
2

j=1

N  
N  
2 1 X
2
 
 
X
 0




 0

g j 12 , uj 21  {{h }}j 21  +
g j 12 , uj 12  [[h ]]j 12 
4
j=1
j=1

N
X
1
j=1

b
a (b
g ; uh )j 1
2

N

2 X

2
1




b
a (b
g ; uh )j 1 [[h ]]j 1  .
[[h ]]j 12 
2
2
2
j=1

The condition (2) of the grid velocity supplies

 

 0

g j 1 , uj 1  C0? + max | (x, t)| C0? + C0 ,
2

17

(2.49)

where C0? := max |f 0 (v)|. Thereby we obtain by Youngs inequality, lemma 2.4 and the
mvM

interpolation property (2.30)

N  
N
2 X
2

 
X
1
 0




1
1
1
1
b
a (b
g ; uh )j 1 [[h ]]j 
g j 2 , uj 2  {{h }}j 2  +
2
2
2
j=1
j=1
 ?
X
N 



C0 + C0
 + 2  2

+ C2
h,j 1  + h,j 1 
2
2
2
j=1
 ?

C0 + C 0

+ C2 C5 h2k+1 kuk2L (0,T ;Hk+1 ()) .

2

(2.50)

Furthermore we obtain by the mean value theorem and the a priori assumption (2.19)
 


  



 0
  0

0
0
1
1
1
1
1
1
g

,
u

,
{
{u
}
}

f
{
{u
}
}
=
f
u

 
 C1 C1? h.
h j
h j
j 2
j 2
j 2
j 2
2
2
Thus lemma 2.4, the a priori assumption (2.19) and the trace inequality (2.31) provide

N   
2
 1
X
1 0



1
1
1
g

b
a
(b
g
;
u
)
[[
]]
,
u

1

 h j 2 
h j
j 2
j 2 
2
4
2
j=1
N
2


 
1 X  0 



f uj 12 f 0 {{uh }}j 21  [[h ]]j 12 

4 j=1

N   
2
 1
X
1 0



1
1
1
+
g

b
a
(b
g
;
u
)
,
{
{u
}
}
[[
]]


 h j 2 
h j 1
h j
j 2
2
2
4
2
j=1
N 
N 
2  C X

2
C1 C1? X 




3

h
[[h ]]j 12  +
[[eh ]]j 21  [[h ]]j 12 
4
2 j=1
j=1
 ?

C1
C1
+ 2C3 C7 kh k2L2 () .
2

(2.51)

Finally by (2.49), (2.50) and (2.51) follows



a3 (, h , h ) CII h2k+1 + kh k2L2 () ,

(2.52)

where the constant CII is independent of uh , h and t [0, T ].

Next by plugging the estimates (2.42), (2.45) and (2.52) in the equation (2.39) we obtain


1d
2
2
2k+1
kh kL2 () CIII h
+ kh kL2 () ,
(2.53)
2 dt
where the constant CIII is independent of uh , h and t [0, T ]. Hence by Gronwalls inequality
and the identity uh (0) = Ph (u0 ) it follows for all t [0, T ]
kh k2L2 () e2CIII T h2k+1 .
18

(2.54)

Thus for all t [0, T ] the error function eh can be estimated as follows
keh kL2 () kh kL2 () + kh kL2 () CIV h2k+1 ,
where the constant CIV is independent of uh and h.
2.3.3

An optimal error estimate by using an upwind numerical flux

In order to achieve the optimal a priori error estimate for the ALE-DG method, we need the
following extra assumption for the grid velocity:
(3): There exists a constant C8 independent of h such that
max
(x,t)[0,T ]

|x ( (x, t))| C8 .

Further we assume that g 0 (, v) 0. Then we can apply an upwind numerical flux function
given by


gb

j 12

, u
, u+
h,j 21
h,j 12

:= g

j 12

, u
h,j 12

(t) ,

j = 1, , N.

(2.55)

This ingredient provides the following a priori error estimate.

Theorem 2.8. Let u W1, 0, T ; Hk+2 () be the exact solution of equation (1.1). Assume
f C 2 (R) and the grid velocity satisfies the conditions (1), (2) as well as (3). Further
the condition g 0 (, v) 0 is satisfied. Let uh be the solution of the semi-discrete ALE-DG
method (2.13) with the upwind flux (2.55). The initial data for the method is the GaussRadau projection Ph of u0 . Then there exists a constant C independent of uh and h, such
that there holds the error estimate
max keh kL2 () Chk+1 ,

t[0,T ]

where eh := u uh and h is given by (2.4).

Proof. First we define the quantities
h := u Ph (u)

and h := uh Ph (u)

(2.56)

as in the proof of theorem 2.7. Then the ALE-DG scheme (2.17) yields the following error
equation
0 = (t eh , vh )Kj (t) + (x (eh vh ) , 1)Kj (t) (f (u) f (uh ), x vh )Kj (t)





+
1
1
+ g j+ 1 , uj+ 1 vh,j+

,
u
1
j 2
j 2 vh,j 1
2
2
2


 2


+
g j+ 1 , uh,j+ 1 vh,j+ 1 + g j 1 , uh,j 1 vh,j
1.
2

19

(2.57)

By Taylor expansion on the flux function g (, ) up to second order follows





g j 1 , uj 1 g j 1 , u
h,j 12
2
2
2


2
1 00   
1
e
=g 0 j 1 , uj 1 e

,
h,j 2
h,j 12
j 21
2
2
2

(2.58)

is a value between uj 1 and u

. Then by (2.35), (2.58) and summation over
where
h,j 1
j 1
2

j the error equation can be rewritten as follows

1d
kh k2L2 () = b1 (eh , h ) + b2 (, h , h ) + b3 (, h , h ) ,
2 dt

(2.59)

where
N

1 X 00
b1 (eh , h ) =
f () (eh )2 , x h Kj (t)
2 j=1
N

1 X 00    2
f j 1
eh,j 1 [[h ]]j 1 ,
2
2
2
2 j=1

N
N



1X 0
1X
2


b2 (, h , h ) =
f (u) , x (h ) Kj (t)
x (h )2 , 1 Kj (t)
2 j=1
2 j=1

N
X

j 21

,u

j 12

[[h ]]j 1 ,
h,j 1
2

j=1

and
b3 (, h , h ) =

N
X

(t h , h )Kj (t) +

j=1

N
X

N
X

(x (h h ) , 1)Kj (t)

j=1

(f 0 (u) h , x h )Kj (t) .

j=1

The quantity b1 (eh , h ) can be evaluated similar to the quantity a2 (eh , h ) in the proof
of theorem 2.7. Thus it exists a constant CI independent of uh , h and t [0, T ] such that


b1 (eh , h ) CI h2k+2 + kh k2L2 () .
(2.60)
For all piecewise continuous functions v L2 () holds in the discontinuities


1
{{v}} v [[v]] = [[v]]2 .
2

20

Hence by an integration by parts, (2.46) and (2.47) follows

N

1X
b2 (, h , h ) =
x f 0 (u) , (h )2 Kj (t)
2 j=1
N
2 1

1X 0


g j 1 , uj 1 [[h ]]j 1  C2? kh k2L2 () ,

2
2
2
2 j=1
2

where as in the proof of theorem 2.7 the constant C2? is given by

max
(x,t)[0,T ]

(2.61)

 



0

f
(u
(x,
t))
 x
.

Next an integration by parts and the identity (2.23) yields

b3 (, h , h ) =

N
X

t u

Ph

(t u) , h

Kj (t)

j=1

N
X

N
X

x u Ph (x (u)) , h

Kj (t)

j=1

x u

Ph

N
X




(u) , h Kj (t)
(g 0 (, u) h , x h )Kj (t) .

j=1

(2.62)

j=1

Thus by Youngs inequality as well as the interpolation property (2.29) and similar arguments
as in the proof of theorem 2.7 follows


2k+2

b3 (, h , h ) CII h

kh k2L2 ()

(2.63)

where the constant CII is independent of uh , h and t [0, T ].

Next by plugging the estimates (2.60), (2.61) and (2.63) in the equation (2.59) follows


1d
kh k2L2 () CIII h2k+2 + kh k2L2 () ,
2 dt

(2.64)

where the constant CIII is independent of uh , h and t [0, T ]. The final steps in the proof
of theorem 2.8 are exactly the same as in the proof of theorem 2.7.
Remark 2.1. If we assume g 0 (, v) 0, the result in theorem 2.8 holds also true, but for
that we have to apply the numerical flux function




+
+
1,u
gb j 1 , u
:=
g

,
1,u
1
1
j
h,j
h,j
h,j
2

j = 1, , N

and the Gauss-Radau projection Ph+ instead of Ph .

2.4

The fully discrete ALE-DG method

In this section we consider and analyze the time discretization of the ALE-DG method.

21

2.4.1

The geometric conservation law

The geometric conservation law (GCL) governs the geometric parameters of a grid deformation method in such the way that the method preserves constant states. In other words, if we
consider the equation (1.1) with the initial condition uh (x, 0) 1 for all (x, t) [0, T ],
the approximate solution given by the ALE-DG method has to be uh (x, t) 1 too.
By plugging uh (x, t) 1 Vh into the semi-discrete ALE-DG scheme (2.13) follows for
all vh Vh

d
(1, vh )Kj (t) = (x , vh )Kj (t) , j = 1, , N.
(2.65)
dt
This equation is the geometric conservation law (GCL) condition for the ALE-DG method.
Certainly the equation (2.65) is a special case of the transport equation (2.9) and thus
satisfied.
However the situation is slightly different after the ALE-DG method has been discretized
in time. A discrete version of the GCL is the discrete geometric conservation law (dGCL).
In general it is not clear that the discrete geometric conservation law (dGCL) holds true
whenever the GCL is satisfied. If there is no dGCL satisfied for a method, the method will
not preserve constant states. This leads to a lack of stability and accuracy. The relationship
between the dGCL of a grid deformation method and the stability as well as order of the
time discretization of the method is well known (c.f. Grandmont, Guillard and Farhat [10]
or Farhat and Geuzaine [13]). Fortunately the forward Euler time discretization of the ALEDG method satisfies a dGCL. This can be realized as follows. By applying the mapping
(2.6) we rewrite the semi-discrete GCL condition (2.65) as



d
1, vh J
= x , vh J
,
dt
[1,1]
[1,1]
where J(t) =

j (t)
2

is the determinant of the Jacobian matrix. Notice that J(t) =

(2.66)
j (t)
2

is

only time dependent and the definition of the mapping yields

j+ 21 j 12
x (x, t) =
.
2


Thus x (x, t) is also merely time dependent and the dGCL condition becomes
 j+ 1 j 1
d
2
2
J(t) =
.
dt
2
Therefore, since J(t) is linearly dependent on time t, the dGCL can be easily satisfied for
any first order or high order single step time discretization method, e.g. the forward Euler
method, or total variation diminishing (TVD) Runge-Kutta methods also known as strong
stability preserving (SSP) Runge-Kutta methods (c.f. Gottlieb and Shu [12]).
22

Proposition 2.9. The fully discrete ALE-DG method (2.13) with the approximation space
(2.8) satisfies the discrete geometric conservation law for any first order time discretization
method or high order single step method in which the stage order is equal or higher than first
order.
2.4.2

The local maximum principle

In this section we state the local maximum principle for the ALE-DG method with the
Lax-Friedrichs flux. The Lax-Friedrichs flux is given by

+




g
,
u
+
g
,
u
+
h
h
+
gb , u

uh u
h , uh :=
h ,
2
2

(2.67)

where
:= max {|uh g ( (x, t) , uh )|} .

(2.68)

The Lax-Friedrichs flux can be split up in an increasing function

1

gb+ , u
g , u
h + uh
h :=
2

(2.69)

1
+
g , u+
gb , u+
h uh
h :=
2

(2.70)

and a decreasing function

Further for all x, y Kj (t) holds

gb ( (x, t) , a, b) gb ( (y, t) , a, b) = ( (x, t) (y, t))

a+b
.
2

The average value of the ALE-DG solution uh in the cell Kj (t) will be denoted by
Z
1
uj (t) :=
uh (x, t) dx
4j (t) Kj (t)

(2.71)

(2.72)

and the forward as well as backward differential operators of the cell average value will be
denoted by
4+ uj := uj+1 uj

and 4 uj := uj uj1 .

(2.73)

In order to rewrite the average value of the ALE-DG solution, we apply the p-point GaussLobatto quadrature rule in the reference cell [1, 1], where we choose p to be the smallest
integer satisfying p 3 k, if a piecewise P k polynomial approximation space is used. We
denote the quadrature points by
1 = 1 < 2 < < p = 1,
23

and the corresponding weights by , = 1, , p. Notice that

p
P

=1

un,+
:= uh (j (1, tn ) , tn ) := un,1
h ,
h,j 1
2

= 1. Next we define

un,
:= uh (j (1, tn ) , tn ) := un,p
h
h,j+ 1

(2.74)

and for all = 2, , p 1

uh j j , tn , tn := un,
h .
Since the parameter p is chosen in such the way that the Gauss-Lobatto quadrature rule is
exact for polynomials of degree k, we obtain
Z
p
X
1 1
n,
n
u .
uj =
uh (j (, tn ) , tn ) d =
2 1
2 h
=1

(2.75)

In the following we consider the forward Euler time discretization of the weak formulation
(2.13). By proposition 2.9 holds the geometric conservation law


n
1
1
4n+1

4
=
4t

,
j
j+
j
j
2

(2.76)

where 4nj = 4j (tn ). Thus by plugging the test function vh = 1 into the forward Euler time
discretization of the discrete weak formulation (2.13) and applying the identities (2.69),
(2.70), (2.71) as well as (2.76) follows



4t  
n,+
n,+
n
1,u
1,u

g
b

un+1
=
u

g
b

1
1

j
j+ 2
j+ 2
j
h,j+ 2
h,j 2
4n+1
j



4t  
n,
1
n+1 gb+ j 1 , un,
,
u

g
b

+
j 2
h,j+ 12
h,j 12
2
4j


4t 
1  n,
n,+
n
n+1 j+ 1 j 1
uj
u
.
1 + u
h,j 12
2
2
2 h,j+ 2
4j

(2.77)

Finally we are able to state the following lemma.

Lemma 2.10. Let uh be the solution of the forward Euler time discretization of the ALEDG method (2.13) with the Lax-Friedrichs flux (2.67). For all j = 1, , N are all values
n,p
+
n
u
(tn ) , un,1
h , , uh , uh,j+ 1 (tn ) and uj in the interval [m, M ] and the grid velocity satish,j 1
2

fies the conditions (1), (2) as well as (3). Further the quantity h :=

max h (t) (0, 1)

t[tn ,tn+1 ]

and the CFL condition

4t

min

1p


C8

(2.78)

min + 1 + 8

1p

is satisfied, where the parameter is given by (2.68), the parameter comes from the mesh
regularity property (2.5) and the constant C8 comes from the condition (3) of the grid
velocity. Then for all j = 1, , N is un+1
in the interval [m, M ].
j
24

Proof. First we define the following quantities





n,+
n,1

g
b
b
g
,u
,u

1
h
j+ 1
j+ 1

2 h,j+ 2
2
, if un,+ 1 =

6 un,1
n,1
n,+
h
h,j+ 2
uh
u
1
h,j+ 2
Cj :=

0,
if un,+
= un,1
h
h,j+ 1
2

and

Dj :=





n,

b
g
,u
g
b+ j 1 ,un,p

+
1
h
j 1

2
2 h,j 2

n,+
un,p
h u

1
h,j 2

n,
, if un,p
h 6= uh,j 1
2

0,

n,
if un,p
h = uh,j 1 .
2

It should be noted that Cj 0 and Dj 0 since gb




1
1
j+ , is a decreasing and gb+ j ,
2

is an increasing function. Further by the mean value theorem follows Cj + Dj 2. Next

we define for all a Rp+2 the function
H (a0 , , ap+1 )
1
:=
2
+

p
2



!

4t
1
2
j+ 1 j 1
1 n+1
a1
1
+ Cj
2
2
1
1
4j


!

4t
2
1
j+ 1 j 1
1 n+1
+ Dj
1
ap
2
2
p
p
4j

4t
(Cj ap+1 + Dj a0 )
4n+1
j
!
p1

X

4t 
+
1 n+1 j+ 1 j 1
a .
2
2
2
4
j
=2

Then by applying (2.75) the scheme (2.77) can be written as follows



n,
n,1
n,p
n,+
un+1
=
H
u
,
u
,

,
u
,
u
.
j
h
h
h,j 1
h,j+ 1

(2.79)

The mean value theorem and the condition (3) of the grid velocity provide




j+ 1 j 1  max |x ( (x, tn ))| 4nj C8 h.

(2.80)

xKj (tn )

Thus by applying the CFL number (2.78) follows

H (a0 , , ap+1 ) 0 for all a Rp+2 and = 0, , p + 1.
Further H (a, , a) = a for all a R, since

Pp

=1 2

= 1. Therefore is (2.79) a monotone

scheme in conservation form. This completes the proof.

25

We have seen that the dGCL (2.76) is an important ingredient to prove the local maximum
principle for the ALE-DG method. In fact Grandmont, Guillard and Farhat [10] have proven
that a monotone finite volume ALE method satisfies the local maximum principle if and only
if the method satisfies a dGCL. Finally we apply the maximum-principle-satisfying limiter
in [31] to ensure the local maximum principle for the ALE-DG method.
2.4.3

Total variation stability

We will analyze the forward Euler time discretization of the ALE-DG method for stability
in the sense of the following seminorm
|unh |TVM :=

N
X


4+ unj  ,
j=1

where unh is the ALE-DG solution at time level tn . Thus we are interested in the total
variation stability in the cell average values of an ALE-DG solution. As in the section before
we consider the ALE-DG method merely for the Lax-Friedrichs flux (2.67). In order to
obtain the total variation stability property in the average values, we follow the discussion
in [3]. Therefore we apply for all v, w R the notation
(v, w) := sign (v) sign (w) .
By subtracting the equation (2.77) for j from the equation (2.77) for j + 1 and summation
over j we obtain the following equation
 n+1 
u 
h

TVM

|unh |TVM + + = 0.

The quantity in (2.81) is given by

:=

N  



X


n,+
n,
n,+
n+1
n
n
p unj+1 , un,
,
u

p
u
,
u
,
u

4
u
,
4
u
3
1
1
1
+
+
j
j
j
h,j+
h,j+
h,j+
h,j
j=1

N



X


4t  
n,
n,
1,u
1,u
+
g
b

g
b

4 unj , 4+ un+1
1
1
+
+
j 2
j 2
j
n+1
h,j+
h,j
2
2
4j
j=1
N



X


4t  
n,+
n,+
1
1

b j+ , uh,j+ 1 gb j+ , uh,j 1
4+ unj , 4 un+1
,
j
n+1 g
2
2
2
2
4j
j=1

where for all piecewise continuous functions v, w L2 ()







4t
4t
p v, w , w+ := v n+1 gb+ j+ 1 , w + n+1 gb j+ 1 , w+ .
2
2
4j+1
4j
26

(2.81)

The other quantity in equation (2.81) results from the grid velocity. It is given by
N





1 X 4t
n+1
n
1
3
:=
a

j+1,
j+ 2 4+ uj , 4+ uj
j+ 2
n+1
2 j=1 4j+1
N





1 X 4t
n+1
n
1
1
u
,
4
u
b

4
+
+
j,
+
j
j+
j
j
2
2
2 j=1 4n+1
j
N


X
1 4t 
n,
n
1
1
+
u

j cj,
j 2
j+ 2
n+1
h,j+ 12
2
4
j
j=1
N


X
1 4t 
n,+
n
j+ 1 j 1
uj uh,j 1 dj, ,
+
2
2
2
2 4n+1
j
j=1

where
aj, :=

unj

n,+
uh,j
1
2

bj, :=

,
cj,+ := 4+ unj , 4 un+1
j

un,
h,j+ 12

unj

dj,+ := 4 unj , 4+ un+1
,
j



if j+ 1 j 1 0 and
2

aj,+ := un,
unj ,
h,j+ 1

bj,+ := unj un,+

,
h,j 12




:= 4 unj , 4+ un+1
, dj, := 4+ unj , 4 un+1
,
j
j
2

cj,


if j+ 1 j 1
2

0. Hence the ALE-DG method is total variation diminishing stable in

the average values, if we can ensure that + 0. In fact the sum is positive, if the
ALE-DG solution satisfies the following conditions:


sign 4+ unj = sign (rj, sj, ) ,




n,

u
,
sign 4 unj = sign un,
h,j+ 12
h,j 12




n,+
sign 4+ unj = sign un,+
,
1 u
1
h,j+
h,j
2

where

 1 4t 

n,+
3
1
,
u
+

rj, := p unj+1 , un,

j+ 2
j+ 2 aj+1,
h,j+ 23
h,j+ 12
2 4n+1
j+1
and

 1 4t 

n,+
1
1
sj, := p unj , un,
,
u

j+ 2
j 2 bj, .
h,j+ 21
h,j 12
2 4n+1
j
27

(2.82)
(2.83)
(2.84)

In addition


(2.85)

sign 4 unj = sign unj un,+
,
h,j 1

(2.86)





n
sign 4 unj = sign un,

u
j
h,j+ 12




n,+
n
n
sign 4+ uj = sign uj uh,j 1 ,

(2.87)

sign

4+ unj

= sign

un,
h,j+ 12

unj

if j+ 1 j 1 0 and
2

(2.88)

if j+ 1 j 1 0. In general the ALE-DG solution does not satisfy the conditions above.
2

Therefore the solution has to be revised by a post processing procedure. Cockburn and Shu
(c.f. [3], [6] and [7]) have developed TVD limiters for Runge-Kutta DG methods in such the
way that the by the limiter revised solution u
eh satisfies for all j = 1, , N the conditions


n,
n,
n
n
n
u
eh,j+ 1 = m uh,j+ 1 uj , 4 uj , 4+ uj + unj
(2.89)
2

and


n,+
n
n
n
n
u
en,+
=
u

m
u

u
,
4
u
,
4
u
j
+ j ,
j
j
h,j 1
h,j 1
2

(2.90)

where the function m () is the minmod function for all a Rs given by

min , if = sgn (1 ) = = sign (s ) ,

1 s
m (1 , , s ) :=

0,
else.
Further they proved that this kind of limiters provide the total variation diminishing stability
in the average values of the method. The following result indicates that the TVD limiters
can be applied for the forward Euler time discretization of the ALE-DG method too.
Proposition 2.11. Let u0 BV () L1 () and u
eh be the solution of the forward Euler
time discretization of the ALE-DG method revised by a conservative TVD limiter, such that
u
eh satisfies (2.89) and (2.90). The initial data for the method is the L2 projection of the
function u0 , the grid velocity satisfies the conditions (1), (2) as well as (3), the quantity
h :=

max h (t) (0, 1) and the CFL condition

t[tn ,tn+1 ]

4t
1

h
C8 + 4
is satisfied. Then for all n = 0, ..., K
|unh |TVM |u0 |BV() .
28

(2.91)

Proof. First we will prove that the sum + in equation (2.81) becomes non-negative, if we
plug the solution of the ALE-DG method, revised by a minmod limiter, in (2.81). Therefore
we have to show that u
eh satisfies the conditions (2.82) - (2.88). Since u
eh satisfies the equations
(2.89) and (2.90), are the conditions (2.85) - (2.88) obviously fulfilled. Moreover we obtain


n,
n,
n,
n
n
n
u
en,
u
+
4
u

u
e
u

u
e
=
u
e

j
j
j1 .
h,j+ 1
h,j 1
h,j 1
h,j+ 1
2

en,
Is 4 unj = 0, the equations (2.89) and (2.90) implicate that u
e
un,
= 0. Is 4 unj 6= 0
h,j+ 1
h,j 1
2

follows by the definition of the minmod function and (2.89) as well as (2.90)

n,
n,
n
n
u
eh,j
u
e

u
1
j1
j
h,j+ 12
2
2.
0 1
+
n
n
4 u j
4 uj

(2.92)

Hence (2.83) is satisfied, since

u
en,
u
en,
= 1
h,j+ 1
h,j 1
2

unj1
u
en,
h,j 1

4 unj

unj
u
en,
h,j+ 1
2

4 unj

4 unj .

Is 4+ unj 6= 0, follows

0 1

unj+1 u
en,+
h,j+ 1

4+ unj

unj u
en,+
h,j 1
2

4+ unj

2.

(2.93)

Thus the condition (2.84) can be proven in a similar way.

It remains to prove that the condition (2.82) is satisfied. First of all we have to mention that
rj, sj, =0, if 4+ unj = 0. Therefore in the following we assume that 4+ unj 6= 0. Since u
eh
satisfies the equations (2.89) as well as (2.90), follows by the mean value theorem, the mesh
regularity property (2.5), (2.92) and (2.93)



 
n,
n,
 gb 1 , u

1
e

g
b

,
u
e
3
1
+
+
j+ 2
j+ 2
4t 
h,j+ 2
h,j+ 2 
 24t
n
n+1 

4+ uj
h
4j+1 

and




 
n,+
 gb 1 , un,+

4t  j+ 2 h,j+ 12 gb j+ 12 , uh,j 12  24t

 h .
4+ unj
4n+1


j

In addition the equations (2.89) and (2.90) yields





 aj+1, 
 bj, 




 4+ un  1 and  4+ un  1.
j
j
29

Thus, since the grid velocity satisfies the condition (3) and h (0, 1), follows by the mesh
regularity property (2.5) as well as (2.80)


a
1 4t 
C8 4t
j+1, 
1
3
n
n+1  j+ 2 j+ 2
2 4j+1
4+ uj
2h
and

1 4t
2 4n+1
j



 b

 C8 4t
j,
 1 1

.
j+
j
n

2
2
4+ uj 
2h

Therefore the CFL condition (2.91) provides

 






n,
n,
4+ unj  4t gb+ 1 , u
1,u
e

g
b

e

3
1
+
j+ 2
j+ 2
n+1
h,j+
h,j+
2
2
4j+1



4t  

n,+
1
+ n+1 gb j+ 1 , un,+

g
b

,
u
1
1 

j+ 2
h,j+
h,j
2
2
2
4j
 1 4t 



1 4t 



3
1
1
1
+
n+1  j+ 2 j+ 2 aj+1,  +
n+1  j+ 2 j 2 bj,  .
2 4j+1
2 4j

(2.94)

Hence the condition (2.82) is satisfied, since for all a, b R with |a| > |b| follows sign (a) =
sign (a b). Under consideration that the TVD limiter is a conservative limiter, the function
uh evaluated in the semi-norm ||TVM has the same value as the revised function u
eh evaluated
in the same semi-norm. Therefore we obtain
 
|unh |TVM u0h TVM
by applying successive the inequality, resulting from the equation (2.81). Since u0h is the L2
projection of the function u0 BV (), we obtain the result.
To maintain the high order accuracy at local extrema, a TVB limiter has been introduced
(c.f. [27], [3], [6] and [7]). The TVB limiter based on the modified minmod function

m (1 , , s ) :=

1 ,

f 4n
if |1 | M
j

2

(2.95)

m ( , , ) , else.
1
s
f have been discussed in [6]. Further Cockburn and Shu
Selection options for the parameter M
have proven that the TVB limiter provides TVB stability and does not affect the accuracy
of the method. We have a similar result for the forward Euler time discretization of the
ALE-DG method with the TVB limiter.
Proposition 2.12. Let uh be the solution of the forward Euler time discretization of the
ALE-DG method and u
eh be the solution of the method revised by a conservative TVB limiter, such that u
eh satisfies (2.89) and (2.90) for the function (2.95) instead of the minmod
30

function. Suppose for smooth solutions of (1.1) is uh a (k+1)-th order accurate approximation. Then under the same assumption as in proposition 2.11 holds for all n = 0, , K
f || T,
|unh |TVM |u0 |BV() + (4C8 + 16) M
where || denotes the Lebesgue measure of the set and tK = T . Moreover for smooth
solutions of (1.1) is u
eh a (k+1)-th order accurate approximation.
This result can be proven by similar arguments as in [27] and the last proof. Therefore
we omit a proof in this paper.

Numerical experiments

In this section we display the performance of the ALE-DG scheme. We adopt TVD RungeKutta methods (c.f. Gottlieb and Shu [12]) for the time discretization, which are convex
combinations of the forward Euler method. Thus by an adequate adjustment of the CFL
condition, the results for the forward Euler discretization can be extended to TVD RungeKutta methods.
Example 3.1 (Burgers equation).
We solve the Burgers equation with periodic boundary condition:

 
t u + x u2 = 0, x [0, 1]
2
u(x, 0) =

1
4

+ 12 sin((2x 1)).

The exact solution is smooth at T = 0.1 and has a well developed shock at T = 0.4. Here
we choose the time step small enough to demonstrate the spatial error only. To maintain
f = 20.
the stability, the TVB limiter is used with the parameter M
In Table 3.1 we compare the convergence history of the ALE-DG method by using piecewise P 2 and P 3 polynomial elements with different cell numbers N at T = 0.1 on the static
uniform grid and the moving grid xj+ 1 (tn ) = xj+ 1 (0) + 0.4 sin(tn )(xj+ 1 (0) 1)xj+ 1 (0) re2

spectively. The moving grid starts from an uniform grid initially. And we use uSh and uM
h
denote the numerical solutions on the static and moving grid respectively. It can be seen
that numerically the optimal convergence order can be obtained for both grids. Notice that
the ALE-DG method on a static grid is the original DG method in [3, 6]. In Table 3.2 we
show the convergence of the ALE-DG scheme for both grids when the shock is developed.
With the help of the TVB limiter, the ALE-DG scheme is uniformly high order in regions of
31

u uSh

P3

u uSh

u uM
h

u uM
h

L norm

order

L2 norm

order

L norm

order

L2 norm

order

10

4.34E-03

9.10E-04

4.74E-03

9.87E-04

20

7.53E-04

2.53

1.25E-04

2.86

8.10E-04

2.55

1.28E-04

2.95

40

1.14E-04

2.72

1.70E-05

2.88

1.25E-04

2.70

1.72E-05

2.90

80

1.60E-05

2.83

2.28E-06

2.90

1.76E-05

2.83

2.32E-06

2.89

160

2.13E-06

2.91

3.00E-07

2.93

2.36E-06

2.90

3.08E-08

2.91

10

5.55E-04

7.46E-05

5.10E-04

7.47E-05

20

4.16E-05

3.74

5.21E-06

3.84

3.58E-05

3.83

5.09E-06

3.88

40

3.12E-06

3.74

3.66E-07

3.83

2.71E-06

3.72

3.51E-07

3.86

80

2.11E-07

3.89

2.49E-08

3.88

1.83E-07

3.89

2.43E-08

3.85

160

1.37E-08

3.94

1.66E-09

3.91

1.19E-08

3.94

1.64E-09

3.89

Table 3.1: Erros at time T = 0.1 for Burgers equation.

smoothness. Moreover in figure 3.1 we compare the exact and the ALE-DG solutions with
N = 80 and k = 4 at time T = 0.4. It is shown that shocks are captured in a few elements
without production of spurious oscillations.
In Table 3.3 the convergence history of the ALE-DG method with different polynomial
degree k is displayed on the same static and moving grids with the cell number N = 40 at
time T = 0.1 and T = 0.4. We can see that the ALE-DG method maintains the spectral
convergence property of the DG method. This indicates the efficiency of the ALE-DG method
using polynomials of higher degree.
Example 3.2 (Eulers equations).
We consider Eulers equations of gas dynamics for a polytropic gas


t u + x f (u) = 0, x [0, 1],
u = (, m, E)T ,

f (u) = vu + (0, p, pv)T ,

with
1
p = ( 1)(E v 2 ), m = v,
2
where = 1.4 is used in the following computation. Two sets of initial conditions are
considered. One is a smooth function (plain wave)
(, v, p) = (1 + 0.5 sin(2(x t)), 1, 1),
32

u uSh

u uSh

u uM
h

u uM
h

L error

order

L2 error

order

L error

order

L2 error

order

10

5.81E-03

1.11E-03

1.72E-02

2.45E-03

20

1.75E-04

5.05

3.81E-05

4.86

8.61E-04

4.32

8.47E-05

4.85

40

2.41E-05

2.86

4.05E-06

3.23

3.26E-05

4.72

4.12E-06

4.36

80

3.33E-06

2.86

4.59E-07

3.14

4.58E-06

2.83

4.90E-07

3.07

160

4.37E-07

2.93

5.41E-08

3.08

6.09E-07

2.91

5.57E-08

3.14

10

2.26E-03

4.27E-04

5.39E-03

8.85E-04

20

9.99E-06

7.82

1.51E-06

8.14

1.83E-04

4.88

1.81E-05

5.61

40

8.40E-07

3.57

8.83E-08

4.10

1.25E-06

7.19

9.66E-08

7.55

80

6.19E-08

3.76

5.37E-09

4.04

9.52E-08

3.71

5.95E-09

4.02

160

4.17E-09

3.89

3.29E-10

4.03

6.49E-09

3.87

3.63E-10

4.03

Table 3.2: Errors in smooth regions = {x : |x shock| 0.1} at time T = 0.4 for Burgers
equation.

T =0.1

T =0.4

u uSh

u uM
h

u uSh

u uM
h

1.89E-03

1.77E-03

7.21E-04

7.25E-04

1.14E-04

1.25E-04

2.41E-05

3.25E-05

3.12E-06

2.71E-06

8.40E-07

1.25E-06

1.21E-07

1.44E-07

3.33E-08

5.47E-08

4.25E-09

3.40E-09

1.30E-09

2.40E-09

1.53E-10

1.97E-10

5.36E-11

1.09E-10

3.98E-12

3.55E-12

2.42E-12

5.07E-12

1.38E-13

1.52E-13

1.45E-13

3.50E-13

7.88E-15

4.14E-14

1.61E-14

7.70E-14

Table 3.3: L errors at time T = 0.1 and T = 0.4 in a smooth region for Burgers equation
with N = 40.

33

0.8

uSh
uexact

0.6

0.4

0.2

-0.2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.7

0.8

0.9

0.8

uM
h
uexact

0.6

0.4

0.2

-0.2

0.1

0.2

0.3

0.4

0.5

0.6

Figure 3.1: Comparison of the exact and the ALE-DG solutions uSh (top, on the static grid)
and uM
h (bottom, on the moving grid) with N = 80, k = 4 at time T = 0.4 .
34

 Sh

P3

Sh

M
h

M
h

L norm

order

L2 norm

order

L norm

order

L2 norm

order

10

2.63E-03

9.95E-04

5.14E-03

1.48E-03

20

3.87E-04

2.77

1.42E-04

2.78

7.88E-04

2.70

2.20E-04

2.75

40

5.10E-05

2.92

1.87E-05

2.93

1.06E-04

2.89

2.94E-05

2.91

80

6.46E-06

2.98

2.38E-06

2.98

1.36E-05

2.96

3.75E-06

2.97

160

8.08E-07

3.00

2.98E-07

2.99

1.71E-06

2.99

4.71E-07

2.99

10

7.23E-05

1.92E-05

1.91E-04

3.60E-05

20

4.40E-06

4.04

1.07E-06

4.16

1.27E-05

3.90

1.97E-06

4.19

40

2.74E-07

4.01

6.65E-08

4.01

8.07E-07

3.98

1.15E-07

4.10

80

1.71E-08

4.00

4.14E-09

4.00

5.10E-08

3.98

6.99E-09

4.04

160

1.07E-09

4.00

2.59E-10

4.00

3.20E-09

3.99

4.30E-10

4.02

Table 3.4: Erros of density at time T = 1.2 for Eulers equations.

with periodic boundary condition. The other is a modified Sod shock tube problem (Riemann
problem) with left and right state
(L , vL , pL ) = (1, 0.75, 1),

(R , vR , pR ) = (0.125, 0, 1).

In Table 3.4 the convergence history of the density given by the ALE-DG method with
piecewise P 2 and P 3 polynomial elements is displayed at time T = 1.2, where Sh is the ALEDG solution on the static uniform grid and M
h is the ALE-DG solution on the same moving
grid as in the last test. We can see that the optimal convergence order can be obtained
numerically for both grids. In figure 3.2 we compare the exact and the ALE-DG solutions
uSh and uM
h on static and moving grids with N = 200 and k = 4 at time T = 0.2. The
f = 20. It is shown that both solutions converge to the entropy
TVB limiter is used with M
solution and the performance is similar.
In these numerical experiments we do not consider the methodology of how to move the
grid, but the scenario when the grid are chosen at two adjacent time levels. These tests show
that the ALE-DG method maintains the properties of the DG method for static grids, such
as uniformly high order accuracy and shock capturing. Furthermore figure 3.5 show that the
ALE-DG method satisfies the geometric conservation law numerically as we proved.

35

Sh

M
h

exact

exact

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.4

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.4

vhS

vhM

1.2

1.2
vexact

vexact

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

pSh

pM
h

pexact

pexact

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0
0

0.1

Figure 3.2:

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Comparison of the exact and the ALE-DG solutions uSh (left column, on the

static grid) and uM

h (right column, on the moving grid) for the modified Sod shock tube
f = 20.
problem with N = 200, k = 4 at time T = 0.2 and M

36

u uM
h

Table 3.5:

M
h

P2

P3

P2

P3

10

4.44E-15

9.77E-15

4.44E-15

5.77E-15

20

9.99E-15

1.24E-14

5.77E-15

9.66E-15

40

1.24E-14

2.51E-14

9.55E-15

1.78E-14

80

2.22E-14

1.89E-14

1.77E-14

2.45E-14

160 2.80E-14

3.62E-14

3.24E-14

3.30E-14

L errors at time T = 1.2 for Burgers equation and Eulers equations with

constant solutions u = 1 and (, v, p) = (1, 1, 1) .

Conclusions

In this paper we developed a geometric conservation law satisfying ALE-DG method on

moving grids with the time-dependent approximation space.We began the paper with theory,
by proving a cell entropy inequality and L2 stability. We also gave error estimates for the
ALE-DG method with monotone numerical fluxes and an upwind flux separately. For the
fully discrete scheme, the geometric conservation law and the local maximum principle have
been proven. Moreover, for shock capturing, conditions for TVD/TVB limiter have been
established. Numerically, it has been shown that our ALE-DG method is uniformly high
order accurate and shock capturing. In this paper we have merely considered how to develop
the ALE-DG scheme after the grid are chosen at two adjacent time levels. In a future work
we will consider the methodology of how to move the grid efficiently and combine it with our
ALE-DG method. In preparation is also the generalized of the method to multidimensional
problems.

References
[1] M. Abramowitz and I. Stegun, Handbook of Mathematical Functions, National Bureau
of Standards, Applied Mathematics Series 55, Washington, D.C., 1964.
[2] P.G. Ciarlet, The Finite Element Method for Elliptic Problems, SIAM Classic in Applied
Mathematics 40, SIAM Publications, Philadelphia, 2002.
[3] B. Cockburn, Discontinuous Galerkin method for convection-dominated Problems. In
High-Order Methods for Computational Physics, T.J. Barth and H. Deconinck (Eds.),
37

Lecture Notes in Computational Science and Engineering 9 (1999), Springer-Verlag,

Berlin, 69-224.
[4] B. Cockburn, S. Hou and C.-W. Shu, The Runge-Kutta local projection discontinuous
Galerkin finite element method for conservation laws IV: the multidimensional case,
Math. Comp., 54 (1990), 545-581.
[5] B. Cockburn, S.-Y. Lin and C.-W. Shu, TVB Runge-Kutta local projection discontinuous Galerkin finite element method for conservation laws III: one dimensional systems,
J. Comput. Phys., 84 (1989), 90-113.
[6] B. Cockburn and C.-W. Shu, TVB Runge-Kutta local projection discontinuous Galerkin
finite element method for conservation laws II: general framework, Math. Comp., 52
(1989), 411- 435.
[7] B. Cockburn and C.-W. Shu, Runge-Kutta discontinuous Galerkin methods for
convection-dominated problems, J. Sci. Comput., 16 (2001), 173-261.
[8] J. Donea, A. Huerta, J.-P. Ponthot and A. Rodrguez-Ferran, Arbitrary LagrangianEulerian Methods. In Encyclopedia of Computational Mechanics, Erwin Stein, Rene De
Borst and Thomas J.R. Hughes (Eds.), Volume 1: Fundamentals, Wiley, 2004.
[9] R. J. DiPerna, Convergence of approximate solutions to conservation laws, Arch. Rational Mech. Anal., 82 (1983), 27-70.
[10] C. Farhat, P. Geuzaine and C. Grandmont, The Discrete Geometric Conservation Law
and the Nonlinear Stability of ALE Schemes for the Solution of Flow Problems on
Moving Grids, J. Comp. Phys., 174 (2001), 669-694.
[11] R. Fazio and R. J. LeVeque. Moving-mesh methods for one-dimensional hyperbolic
problems using CLAWPACK. Computers & Mathematics with Applications, 45 (2003),
273-298.
[12] S. Gottlieb and C.-W. Shu, Total variation diminishing Runge-Kutta schemes, Math.
Comp., 67 (1998), 73 - 85.
[13] H. Guillard and C. Farhat, On the significance of the geometric conservation law for
flow computations on moving meshes, Comp. Meth. Appl. Mech. Eng., 190 (2000),
1467-1482.
38

[14] A. Harten, High resolution schemes for hyperbolic conservation laws, J. Comp. Phys.,
49 (1983), 357-393.
[15] G.-S. Jiang and C.-W. Shu, On a cell entropy inequality for discontinuous Galerkin
methods, Math. Comp., 62 (1994), 531-538.
[16] C. Johnson and J. Pitkranta, An analysis of the discontinuous Galerkin method for a
scalar hyperbolic equation, Math. Comp., 46 (1986), 1-26.
[17] D. Keres, M. Vogelsberger, D. Sijacki, V. Springel and L. Hernquist, Moving-mesh
cosmology: characteristics of galaxies and haloes, Oxford J. Sci. & Math. MNRAS 425
(2012), 2027-2048.
[18] S. N. Kruzkov, First order quasi-linear equations in several independent variables, Math.
USSR Sb., 10 (1970), 217-243.
[19] P. LeSaint and P. A. Raviart, On a finite element method for solving the neutron
transport equation. In Mathematical Aspects of Finite Elements in Partial Differential
Equations, C. de Boor (Eds.), Academic Press, New York 1974, 89-145.
[20] M. Lesoinne and C. Farhat, Geometric conservation laws for flow problems with moving boundaries and deformable meshes, and their impact on aeroelastic computations,
Comp. Meth. in Appl. Mech. and Eng., 134 (1996), 71-90.
[21] C. K. Lombard and P. D. Thomas, Geometric conservation law and its application to
flow computations on moving grids, AIAA J., 17 (1979), 1030-1037.
[22] I. Lomtev, R. M. Kirby, G. E. Karniadakis, A Discontinuous Galerkin ALE Method for
Compressible Viscous Flows in Moving Domains, J. Comp. Phys., 155 (1999), 128-159.
[23] V.-T. Nguyen, An arbitrary Lagrangian-Eulerian discontinuous Galerkin method for
simulations of flows over variable geometries, J. of Fluids and Struc., (26) (2010), 312329.
[24] P.-O. Persson, J. Bonet, J. Peraire, Discontinuous Galerkin Solution of the NavierStokes Equations on Deformable Domains, Comp. Meth. in Appl. Mech. and Eng., 198
(2009), 1585-1595.
[25] T. E. Peterson, A note on the convergence of the discontinuous Galerkin method for a
scalar hyperbolic equation, SIAM J. Numer. Anal., 28, (1991), 133-140.
39

[26] B. A. Robinson, H. T. Y. Yang and J. T. Batina, Aeroelastic analysis of wings using

the Euler equations with a deforming mesh, J. of Aircraft, 28 (1991), 781-788.
[27] C.-W. Shu, TVB uniformly high-order schemes for conservation laws, Math. Comp., 49
(1987), 105-121.
[28] Q. Zhang and C.-W. Shu, Error estimates to smooth solutions of Runge-Kutta discontinuous Galerkin methods for scalar conservation laws, SIAM J. Numer. Anal., 42
(2004), 641-666.
[29] Q. Zhang and C.-W. Shu, Error Estimates to Smooth Solutions of Runge-Kutta Discontinuous Galerkin Method for Symmetrizable Systems of Conservation laws, SIAM J.
Numer. Anal., 44 (2006), 1703-1720.
[30] Q. Zhang and C.-W. Shu, Stability Analysis and A Priori Error Estimates of the Third
Order Explicit Runge-Kutta Discontinuous Galerkin Method for Scalar Conservation
laws, SIAM J. Numer. Anal., 48 (2010), 1038-1063.
[31] X. Zhang and C.-W. Shu, On maximum-principle-satisfying high order schemes for
scalar conservation laws, J. Comp. Phys., 229 (2010), 3091-3120.

40