Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
LAGRANGEGALERKIN METHOD FOR
CONVECTIONDIFFUSION PROBLEMS. PART I: TIME
DISCRETIZATION
∗
MARTA BEN
´
ITEZ
†
AND ALFREDO BERM
´
UDEZ
†
Abstract. We propose and analyze a second order pure Lagrangian method for variable co
eﬃcient convection(possibly degenerate) diﬀusion equations with mixed DirichletRobin boundary
conditions. First, the method is rigorously introduced for exact and approximate characteristics.
Next, l
∞
(H
1
) stability is proved and l
∞
(H
1
) error estimates of order O(∆t
2
) are obtained. More
over, l
∞
(L
2
) stability and l
∞
(L
2
) error estimates of order O(∆t
2
) with constants bounded in the
hyperbolic limit are shown. For the particular case of Dirichlet boundary conditions, diﬀusion tensor
A = ǫI and righthand side f = 0, the l
∞
(H
1
) stability estimate is independent of ǫ. In a second
part of this work, the pure Lagrangian scheme will be combined with Galerkin discretization using
ﬁnite elements spaces and numerical examples will be presented.
Key words. convectiondiﬀusion equation, pure Lagrangian method, characteristics method,
stability, error estimates, second order schemes
AMS subject classiﬁcations. 65M12, 65M15, 65M25, 65M60
1. Introduction. The main goal of the present paper is to introduce and ana
lyze a second order pure Lagrangian method for the numerical solution of convection
diﬀusion problems with possibly degenerate diﬀusion. Computing the solutions of
these problems, especially in the convection dominated case, is an important and
challenging problem that requires development of reliable and accurate numerical
methods.
Linear convectiondiﬀusion equations model a variety of important problems from
diﬀerent ﬁelds of engineering and applied sciences, such as thermodynamics, ﬂuid me
chanics, and ﬁnance (see for instance [20]). In many cases the diﬀusive term is much
smaller than the convective one, giving rise to the socalled convection dominated
problems (see [17]). Furthermore, in some cases the diﬀusive term becomes degener
ate, as in some ﬁnancial models (see, for instance, [26]).
This paper concerns the numerical solution of convectiondiﬀusion problems with
degenerate diﬀusion. For this kind of problems, methods of characteristics for time
discretization are extensively used (see the review paper [17]). These methods are
based on time discretization of the material time derivative and were introduced in
the beginning of the eighties of the last century combined with ﬁnitediﬀerences or
ﬁnite elements for space discretization. When these methods are applied to the formu
lation of the problem in Lagrangian coordinates (respectively, Eulerian coordinates)
they are called pure Lagrangian methods (respectively, semiLagrangian methods).
The characteristics method has been mathematically analyzed and applied to diﬀer
ent problems by several authors, primarily the semiLagrangian methods. In par
ticular, the (classical) semiLagrangian method is ﬁrst order accurate in time. It
has been applied to time dependent convectiondiﬀusion equations combined with ﬁ
∗
This work was supported by Xunta de Galicia under research project INCITE09 207 047 PR, and
by Ministerio de Ciencia e Innovaci´ on (Spain) under research projects Consolider MATHEMATICA
CSD200600032 and MTM200802483.
†
Dep. de Matem´ atica Aplicada, Universidade de Santiago de Compostela, Campus Vida, 15782
Santiago de Compostela, Spain (marta.benitez@usc.es, alfredo.bermudez@usc.es). The ﬁrst au
thor was supported by Ministerio de Educaci´ on.
1
2 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
nite elements ([16], [21]), ﬁnite diﬀerences ([16]), etc. Its adaptation to steady state
convectiondiﬀusion equations has been developed in [8] and, more recently, the com
bination of the classical ﬁrst order scheme with disconuous Galerkin methods has
been used to solve ﬁrstorder hyperbolic equations in [3], [2] and [4]. Higher or
der characteristics methods can be obtained by using higher order schemes for the
discretization of the material time derivative. In [22] multistep LagrangeGalerkin
methods for convectiondiﬀusion problems are analyzed. In [11] and [12] multistep
methods for approximating the material time derivative, combined with either mixed
ﬁnite element or spectral methods, are studied to solve incompressible NavierStokes
equations. Stability is proved and optimal error estimates for the fully discretized
problem are obtained. In [25] a second order characteristics method for solving con
stant coeﬃcient convectiondiﬀusion equations with Dirichlet boundary conditions is
studied. The CrankNicholson discretization has been used to approximate the ma
terial time derivative. For a divergencefree velocity ﬁeld vanishing on the boundary
and a smooth enough solution, stability and error estimates are stated (see also [9]
and [10] for further analysis). In [15] semiLagrangian and pure Lagrangian meth
ods are proposed and analyzed for convectiondiﬀusion equation. Error estimates for
a Galerkin discretization of a pure Lagrangian formulation and for a discontinuous
Galerkin discretization of a semiLagrangian formulation are obtained. The estimates
are written in terms of the projections constructed in [13] and [14].
In the present paper, a pure Lagrangian formulation is used for a more general prob
lem. Speciﬁcally, we consider a (possibly degenerate) variable coeﬃcient diﬀusive term
instead of the simpler Laplacian, general mixed DirichletRobin boundary conditions
and a time dependent domain. Moreover, we analyze a scheme with approximate
characteristic curves.
The mathematical formalism of continuum mechanics (see for instance [18]) is
used to introduce the schemes and to analyze the error. In most cases the exact
characteristics curves cannot be determined analytically, so our analysis include, as a
novelty with respect to [15], the case where the characteristics curves are approximated
using a second order RungeKutta scheme. A proof of l
∞
(L
2
) stability inequality is
developed which can be appropriately used to obtain l
∞
(L
2
) error estimates of order
O(∆t
2
) between the solutions of the time semidiscretized problem and the contin
uous one; these estimates are uniform in the hyperbolic limit. More precisely, let
´
φ
m
= {φ
n
m
}
N
n=0
and
φ
m,∆t
= {φ
n
m,∆t
}
N
n=0
denote, respectively, the exact solution of
the continuous problem in Lagrangian coordinates (see §3), and the discrete solution
of the pure Lagrangian method proposed and analyzed in this paper (see §4). We
prove (Corollary 4.12 and Theorem 4.27) the following inequalities:
¸
¸
¸
¸
¸
¸
φ
m,∆t
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
Λ
4
¸
¸
¸
¸
¸
¸
BS[∇φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
S[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤ J
1
_
φ
0
m,∆t

Ω
+
¸
¸
¸
¸
¸
¸
f ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
g ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
_
,
(1.1)
and

φ
m
−φ
m,∆t

l
∞
(L
2
(Ω))
+
_
Λ
4
¸
¸
¸
¸
¸
¸
BS [∇φ
m
−∇φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
S [φ
m
−φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤ J
2
∆t
2
_
φ
m

C
3
(L
2
(Ω))
(1.2)
+∇φ
m

C
2
(H
1
(Ω))
+∇φ
m
· m
C
2
(L
2
(Γ
R
))
+φ
m

C
2
(L
2
(Γ
R
))
HIGHER ORDER PURE LAGRANGIAN METHOD 3
+f
m

C
2
(L
2
(Ω))
+f
C
1
(T
δ
)
+g
m

C
2
(L
2
(Γ
R
))
+ g
C
1
(T
δ
Γ
R
)
_
,
where Ψ
m
:= Ψ ◦ X
e
for a spatial ﬁeld Ψ, being X
e
the motion, B is the matrix
B =
_
I
n1
Θ
Θ Θ
_
, where I
n1
is the n
1
×n
1
identity matrix,
¯
S[ψ] := {ψ
n+1
+ψ
n
}
N−1
n=0
for a sequence
´
ψ = {ψ}
N
n=0
and
X
RK
= {X
n
RK
}
N
n=0
is a second order RungeKutta
approximation of X
e
. The diﬀusion tensor has the form A =
_
A
n1
Θ
Θ Θ
_
and Λ is
a uniform lower bound for the eigenvalues of A
n1
. Here, J
1
does not depend on the
diﬀusion tensor and J
2
is bounded in the hyperbolic limit. Moreover, for the particular
case of Dirichlet boundary conditions, diﬀusion tensor A = ǫB and righthand side
f = 0, the l
∞
(H
1
) stability estimate is independent of ǫ (see Remark 4.9).
Similar stability and error estimates of order O(∆t
2
) are proved in the l
∞
(H
1
)
norm. In particular, we prove (Corollary 4.16 and Theorem 4.28) the inequalities,
¸
¸
¸
¸
¸
¸
R
∆t
[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
Λ
2
¸
¸
¸
¸
¸
¸
B∇φ
m,∆t
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
φ
m,∆t
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
≤ J
3
_
_
Λ
2
¸
¸
¸
¸
B∇φ
0
m,∆t
¸
¸
¸
¸
Ω
+
¸
¸
¸
¸
φ
0
m,∆t
¸
¸
¸
¸
Γ
R
+
¸
¸
¸
¸
¸
¸
f ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
g ◦ X
RK
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
+
¸
¸
¸
¸
¸
¸
R
∆t
[g ◦ X
RK
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
_
,
(1.3)
and
¸
¸
¸
¸
¸
¸
R
∆t
[φ
m
−φ
m,∆t
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
Λ
2
¸
¸
¸
¸
¸
¸
B(∇φ
m
−∇φ
m,∆t
)
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
φ
m
−φ
m,∆t
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
≤ J
4
∆t
2
_
φ
m

C
3
(L
2
(Ω))
+∇φ
m

C
2
(H
1
(Ω))
+
∇φ
m
· m
C
3
(L
2
(Γ
R
))
+φ
m

C
3
(L
2
(Γ
R
))
+f
m

C
2
(L
2
(Ω))
+f
C
1
(T
δ
)
+g
m

C
3
(L
2
(Γ
R
))
+g
C
2
(T
δ
Γ
R
)
_
,
(1.4)
where
R
∆t
[ψ] :=
_
ψ
n+1
− ψ
n
∆t
_
N−1
n=0
for a sequence
´
ψ = {ψ}
N
n=0
. Here, J
3
and J
4
depend on the diﬀusion tensor; however for the particular case of diﬀusion tensor of
the form A = ǫB, J
3
does not depend on it and J
4
is bounded in the hyperbolic limit.
To prove these estimates we assume that the exact solution and data of the
problem are smooth, and ∆t is suﬃciently small.
The paper is organized as follows. In Section 2 the convectiondiﬀusion Cauchy
problem is stated in a time dependent bounded domain and notations concerning
motions and functional spaces are introduced. In Section 3, the strong formulation
of the convectiondiﬀusion Cauchy problem is written in Lagrangian coordinates and
the standard associated weak problem is obtained. In Section 4, a second order
time discretization scheme is proposed for both exact and second order approximate
characteristics. Next, under suitable hypotheses on the data, the l
∞
(L
2
) and l
∞
(H
1
)
stability results are proved for small enough time step. Finally, assuming greater
regularity on the data, l
∞
(L
2
) and l
∞
(H
1
) error estimates of order O(∆t
2
) for the
solution of the time discretized problem are derived. In a second part of this work
(see [7]), a fully discretized pure LagrangeGalerkin scheme by using ﬁnite elements
in space will be analyzed and numerical results will be presented.
4 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
2. Statement of the problem and functional spaces. Let Ω be a bounded
domain in R
d
(d = 2, 3) with Lipschitz boundary Γ divided into two parts: Γ =
Γ
D
∪ Γ
R
, with Γ
D
∩ Γ
R
= ∅. Let T be a positive constant and X
e
: Ω×[0, T] −→R
d
be a motion in the sense of Gurtin [18]. In particular, X
e
∈ C
3
(Ω × [0, T]) and for
each ﬁxed t ∈ [0, T], X
e
(·, t) is a onetoone function satisfying
det F(p, t) > 0 ∀p ∈ Ω, (2.1)
being F(·, t) the Jacobian matrix of the deformation X
e
(·, t). We call Ω
t
= X
e
(Ω, t),
Γ
t
= X
e
(Γ, t), Γ
D
t
= X
e
(Γ
D
, t) y Γ
R
t
= X
e
(Γ
R
, t), for t ∈ [0, T]. We assume that
Ω
0
= Ω. Let us introduce the trajectory of the motion
T := {(x, t) : x ∈ Ω
t
, t ∈ [0, T]}, (2.2)
and the set
O :=
_
t∈[0,T]
Ω
t
. (2.3)
For each t, X
e
(·, t) is a onetoone mapping from Ω onto Ω
t
; hence it has an inverse
P(·, t) : Ω
t
−→ Ω, (2.4)
such that
X
e
(P(x, t), t) = x, P(X
e
(p, t), t) = p ∀(x, t) ∈ T ∀(p, t) ∈ Ω ×[0, T]. (2.5)
The mapping
P : T −→ Ω,
so deﬁned is called the reference map of motion X
e
and P ∈ C
3
(T ) (see [18] pp.
65 − 66). Let us recall that the spatial description of the velocity v : T −→ R
d
is
deﬁned by
v(x, t) :=
˙
X
e
(P(x, t), t) ∀(x, t) ∈ T . (2.6)
We denote by L the gradient of v with respect to the space variables.
Let us consider the following initialboundary value problem.
(SP) STRONG PROBLEM. Find a function φ : T −→R such that
ρ(x)
∂φ
∂t
(x, t) + ρ(x)v(x, t) · gradφ(x, t) − div (A(x) grad φ(x, t)) = f(x, t), (2.7)
for x ∈ Ω
t
and t ∈ (0, T), subject to the boundary conditions
φ(·, t) = φ
D
(·, t) on Γ
D
t
, (2.8)
αφ(·, t) + A(·) grad φ(·, t) · n(·, t) = g(·, t) on Γ
R
t
, (2.9)
for t ∈ (0, T), and the initial condition
φ(x, 0) = φ
0
(x) in Ω. (2.10)
HIGHER ORDER PURE LAGRANGIAN METHOD 5
In the above equations, A : O −→ Sym denotes the diﬀusion tensor ﬁeld, where
Sym is the space of symmetric tensors in the ddimensional space, ρ : O −→ R,
f : T −→ R, φ
0
: Ω −→ R, φ
D
(·, t) : Γ
D
t
−→ R and g(·, t) : Γ
R
t
−→ R, t ∈ (0, T), are
given scalar functions, and n(·, t) is the outward unit normal vector to Γ
t
.
In the following A denotes a bounded domain in R
d
. Let us introduce the Lebesque
spaces L
r
(A) and the Sobolev spaces W
m,r
(A) with the usual norms  · 
r,A
and
 · 
m,r,A
, respectively, for r = 1, 2, . . . , ∞ and m an integer. For the particular case
r = 2, we endow space L
2
(A) with the usual inner product ·, ·
A
, which induces a
norm to be denoted by  · 
A
(see [1] for details). Moreover, we denote by H
1
Γ
D
(Ω)
the closed subspace of H
1
(Ω) deﬁned by
H
1
Γ
D(Ω) :=
_
ϕ ∈ H
1
(Ω), ϕ
Γ
D ≡ 0
_
. (2.11)
For a Banach function space X and an integer m, space C
m
([0, T], X) will be abbre
viated as C
m
(X) and endowed with norm
ϕ
C
m
(X)
:= max
t∈[0,T]
_
max
j=0,...,m
ϕ
(j)
(t)
X
_
.
In the above deﬁnitions, ϕ
(j)
denotes the jth derivative of ϕ with respect to time.
Finally, vectorvalued function spaces will be distinguished by bold fonts, namely
L
r
(A), W
m,r
(A) and H
m
(A), and tensorvalued function spaces will be denoted by
L
r
(A), W
m,r
(A) and H
m
(A). For the particular case m = 1 and r = ∞, we consider
the vectorvalued space W
1,∞
(A) equipped with the following equivalent norm to the
usual one
w
1,∞,A
:= max {w
∞,A
,  div w
∞,A
, ∇w
∞,A
} , (2.12)
being
∇w
∞,A
:= ess sup
x∈A
∇w(x)
2
,
(2.13)
where  · 
2
denotes the tensor norm subordinate to the euclidean norm in R
d
.
Remark 2.1. For the sake of clarity in the notation, in expressions involving
gradients and time derivatives we use the following notation (see, for instance, [18]):
1. We denote by p the material points in Ω, and by x the spatial points in Ω
t
with t > 0.
2. A material ﬁeld is a mapping with domain Ω × [0, T] and a spatial ﬁeld is a
mapping with domain T .
3. We deﬁne the material description Ψ
m
of a spatial ﬁeld Ψ by
Ψ
m
(p, t) = Ψ(X
e
(p, t), t). (2.14)
Similar deﬁnition is used for functions, Ψ, deﬁned in a subset of T or of O.
4. If ϕ is a smooth material ﬁeld, we denote by ∇ϕ (respectively, by Div ϕ) the
gradient (respectively, the divergence) with respect to the ﬁrst argument, and by
˙
ψ
the partial derivative with respect to the second argument (time).
5. If ψ is a smooth spatial ﬁeld, we denote by gradψ (respectively, div ψ) the
gradient (respectively, the divergence) with respect to the ﬁrst argument, and by ψ
′
the partial derivative with respect to the second argument (time).
6 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
3. Weak formulation. We are going to develop some formal computations in
order to write a weak formulation of problem (SP) in Lagrangian coordinates p. First,
by using the chain rule, we have
˙
φ
m
(p, t) = φ
′
(X
e
(p, t), t) + gradφ(X
e
(p, t), t) · v(X
e
(p, t), t). (3.1)
Next, by evaluating equation (2.7) at point x = X
e
(p, t) and then using (3.1), we
obtain
ρ
m
(p, t)
˙
φ
m
(p, t) − [ div (Agradφ)]
m
(p, t) = f
m
(p, t), (3.2)
for (p, t) ∈ Ω × (0, T). Note that in (3.2) there are derivatives with respect to the
Eulerian variable x. In order to obtain a strong formulation of problem (SP) in
Lagrangian coordinates we introduce the change of variable x = X
e
(p, t). By using
the chain rule we get (see [6])
[ div (Agradφ)]
m
= Div
_
F
−1
A
m
F
−T
∇φ
m
det F
¸
1
det F
.
Then, φ
m
satisﬁes
ρ
m
˙
φ
m
det F − Div
_
F
−1
A
m
F
−T
∇φ
m
det F
¸
= f
m
det F. (3.3)
Throughout this article, we use the notation
¯
A
m
(p, t) := F
−1
(p, t)A
m
(p, t)F
−T
(p, t) det F(p, t) ∀(p, t) ∈ Ω ×[0, T],
¯ m(p, t) := F
−T
(p, t)m(p) det F(p, t) ∀(p, t) ∈ Γ ×[0, T],
where m is the outward unit normal vector to Γ. By using the chain rule and noting
that
n(X
e
(p, t), t) =
F
−T
(p, t)m(p)
F
−T
(p, t)m(p)
(p, t) ∈ Γ ×(0, T),
we get
A(x) grad φ(x, t) · n(x, t) = F
−1
(p, t)A
m
(p, t)F
−T
(p, t)∇φ
m
(p, t) ·
m(p)
F
−T
(p, t)m(p)
,
for (p, t) ∈ Γ ×(0, T) and x = X
e
(p, t). Thus, from (2.8)(2.10) and (3.3), we deduce
the following pure Lagrangian formulation of the initialboundary value problem (SP):
(LSP) LAGRANGIAN STRONG PROBLEM. Find a function φ
m
: Ω ×
[0, T] −→R such that
ρ
m
(p, t)
˙
φ
m
(p, t) det F(p, t) − Div
_
¯
A
m
(p, t)∇φ
m
(p, t)
_
= f
m
(p, t) det F(p, t), (3.4)
for (p, t) ∈ Ω ×(0, T), subject to the boundary conditions
φ
m
(p, t) = φ
D
(X
e
(p, t), t) on Γ
D
×(0, T), (3.5)
α¯ m(p, t)φ
m
(p, t) +
¯
A
m
(p, t)∇φ
m
(p, t) · m(p) = ¯ m(p, t)g(X
e
(p, t), t) on Γ
R
×(0, T),
(3.6)
HIGHER ORDER PURE LAGRANGIAN METHOD 7
and the initial condition
φ
m
(p, 0) = φ
0
(p) in Ω. (3.7)
We consider the standard weak formulation associated with this pure Lagrangian
strong problem:
_
Ω
ρ
m
(p, t)
˙
φ
m
(p, t)ψ(p) det F(p, t) dp +
_
Ω
¯
A
m
(p, t)∇φ
m
(p, t) · ∇ψ(p) dp
+α
_
Γ
R
¯ m(p, t)φ
m
(p, t)ψ(p) dA
p
=
_
Ω
f
m
(p, t)ψ(p) det F(p, t) dp
+
_
Γ
R
¯ m(p, t)g
m
(p, t)ψ(p) dA
p
,
(3.8)
∀ψ ∈ H
1
Γ
D
(Ω) and t ∈ (0, T). These are formal computations, i.e., we have assumed
appropriate regularity on the involved data and solution.
4. Time discretization. In this section we introduce a second order scheme for
time semidiscretization of (3.8). We consider the general case where the diﬀusion
tensor depends on the space variable and can degenerate, and the velocity ﬁeld is
not divergencefree. Moreover, mixed DirichletRobin boundary conditions are also
allowed instead of merely Dirichlet ones.
In the ﬁrst part, we propose a time semidiscretization of (3.8) assuming that the
characteristic curves are exactly computed. Next, we propose a secondorder Runge
Kutta scheme to approximate them and obtain some properties. Finally, stability and
error estimates are rigorously stated.
4.1. Second order semidiscretized scheme with exact characteristic curves.
We introduce the number of time steps, N, the time step ∆t = T/N, and the mesh
points t
n
= n∆t for n = 0, 1/2, 1, . . . , N. Throughout this work, we use the notation
ψ
n
(y) := ψ(y, t
n
) for a function ψ(y, t).
The semidiscretization scheme we are going to study is a CrankNicholsonlike scheme.
It arises from approximating the material time derivative at t = t
n+
1
2
, for 0 ≤ n ≤
N−1, by a centered formula and using a second order interpolation formula involving
values at t = t
n
and t = t
n+1
to approximate the rest of the terms at the same time
t = t
n+
1
2
. Thus, from (3.8), we have
1
2
_
Ω
_
ρ
n+1
m
(p) det F
n+1
(p) + ρ
n
m
(p) det F
n
(p)
_ φ
n+1
m,△t
(p) −φ
n
m,△t
(p)
∆t
ψ(p) dp
+
1
4
_
Ω
_
¯
A
n+1
m
(p) +
¯
A
n
m
(p)
__
∇φ
n+1
m,△t
(p) +∇φ
n
m,△t
(p)
_
· ∇ψ(p) dp
+
α
4
_
Γ
R
_
¯ m
n+1
(p) + ¯ m
n
(p)
_
_
φ
n+1
m,△t
(p) + φ
n
m,△t
(p)
_
ψ(p) dA
p
=
1
2
_
Ω
_
det F
n+1
(p)f
n+1
m
(p) + det F
n
(p)f
n
m
(p)
_
ψ(p) dp
+
1
2
_
Γ
R
_
¯ m
n+1
(p)g
n+1
m
(p) + ¯ m
n
(p)g
n
m
(p)
_
ψ(p) dA
p
.
(4.1)
Remark 4.1. In Section 4.4 we will prove that the approximations involved in
scheme (4.1) are O(∆t
2
) at point (p, t
n+
1
2
). Moreover, this order does not change
if we replace the exact characteristic curves and gradients F by accurate enough
approximations.
8 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
4.2. Second order semidiscretized scheme with approximate charac
teristic curves. In most cases, the analytical expression for motion X
e
is unknown;
instead, we know the velocity ﬁeld v. Let us assume that X
e
(p, 0) = p ∀p ∈ Ω. Then,
the motion X
e
, assuming it exists, is the solution to the initialvalue problem
˙
X
e
(p, t) = v
m
(p, t) X
e
(p, 0) = p. (4.2)
Since the characteristics X
e
(p, t
n
) cannot be exactly tracked in general, we propose
the following second order RungeKutta scheme to approximate X
n
e
, n ∈ {0, . . . , N}.
For n = 0:
X
0
RK
(p) := p ∀p ∈ Ω, (4.3)
and for 0 ≤ n ≤ N −1 we deﬁne by recurrence,
X
n+1
RK
(p) := X
n
RK
(p) +△tv
n+
1
2
(Y
n
(p)) ∀p ∈ Ω, (4.4)
being
Y
n
(p) := X
n
RK
(p) +
△t
2
v
n
(X
n
RK
(p)). (4.5)
A similar notation to the one in §2 is used for the Jacobian tensor of X
n
RK
, namely,
F
0
RK
(p) = I, (4.6)
and for 0 ≤ n ≤ N −1,
F
n+1
RK
(p) = F
n
RK
(p) + ∆tL
n+
1
2
(Y
n
(p))
_
I +
∆t
2
L
n
(X
n
RK
(p))
_
F
n
RK
(p). (4.7)
Now, we state some lemmas concerning properties of the approximate characteristics
X
n
RK
. For this purpose, we require the time step to be upper bounded and the
following assumption:
Hypothesis 1. There exists a parameter δ > 0, such that the velocity ﬁeld v is
deﬁned in
T
δ
:=
_
t∈[0,T]
Ω
δ
t
×{t}, (4.8)
being
Ω
δ
t
:=
_
x∈Ωt
B(x, δ). (4.9)
Moreover v ∈ C
1
(T
δ
).
Lemma 4.1. Under Hypothesis 1, there exists a parameter η > 0 such that if
∆t < η then X
n
RK
(p) is deﬁned ∀p ∈ Ω and ∀n ∈ {0, . . . , N}, and the following
inclusion holds
X
n
RK
(Ω) ⊂ Ω
δ
tn
.
HIGHER ORDER PURE LAGRANGIAN METHOD 9
Proof. The result can be easily proved by applying Taylor expansion to X
e
in the
time variable and using the regularity of v.
Lemma 4.2. Under Hypothesis 1, if ∆t < η then
F
n
RK

∞,Ω
≤ e
T(L
∞,T
δ
+C∆t)
∀ n ∈ {0, . . . , N}. (4.10)
Here constant C depends on v.
Proof. Inequality (4.10) follows by applying norms to (4.7), using the initial
condition (4.6) and applying the discrete Gronwall inequality.
Lemma 4.3. Under Hypothesis 1 if ∆t < min{η, 1/(2L
∞,T
δ )}, then
(F
n
RK
)
−1

∞,Ω
≤ e
T(L
∞,T
δ
+C∆t)
∀ n ∈ {0, . . . , N} (4.11)
and
(F
n+1
RK
)
−1
(p) = (F
n
RK
)
−1
(p)
_
I −∆tL
n+
1
2
(Y
n
(p)) + O(∆t
2
)
_
, (4.12)
being the term O(∆t
2
) depending on v and p ∈ Ω, and 0 ≤ n ≤ N −1.
Proof. Firstly, we can write
F
n+1
RK
(p) = M
n
RK
(p)F
n
RK
(p), (4.13)
with
M
n
RK
(p) := I + ∆tL
n+
1
2
(Y
n
(p))
_
I +
∆t
2
L
n
(X
n
RK
(p))
_
. (4.14)
Now, by applying norms to (4.14) we have that I −M
n
RK

∞,Ω
< 1. Thus, M
n
RK
(p)
is invertible for 0 ≤ n ≤ N −1 and then, by induction, we deduce that F
n+1
RK
(p) is in
vertible too, with (F
n+1
RK
)
−1
(p) = (F
n
RK
)
−1
(p)(M
n
RK
)
−1
(p). Moreover, (M
n
RK
)
−1
(p) =
∞
j=0
(I −M
n
RK
(p))
j
so (4.12) follows. The proof of (4.11) is analogous to the one of
the previous lemma.
The following corollaries can be easily proved (see [6] for further details).
Corollary 4.4. Under the assumptions of Lemma 4.2, we have
 det F
n
RK

∞,Ω
≤ e
T( div v
∞,T
δ
+C(v)∆t)
, (4.15)
det F
n
RK
(p) > 0 if ∆t < K, (4.16)
with K depending on v and 0 ≤ n ≤ N. Moreover, ∀p ∈ Ω det F
n+1
RK
(p) satisﬁes
det F
n+1
RK
(p) = det F
n
RK
(p)
_
1 + ∆t div v
n+
1
2
(Y
n
(p)) + O(∆t
2
)
_
, (4.17)
being O(∆t
2
) depending on v and 0 ≤ n ≤ N −1.
Corollary 4.5. Under the assumptions of Lemma 4.3, we have
det (F
n+1
RK
)
−1
(p) = det (F
n
RK
)
−1
(p)
_
1 −∆t div v
n+
1
2
(Y
n
(p)) + O(∆t
2
)
_
, (4.18)
∀p ∈ Ω, ∀n ∈ {0, . . . , N − 1}, with O(∆t
2
) depending on v. Moreover, ∀n ∈
{0, . . . , N}, we have
 det (F
n
RK
)
−1

∞,Ω
≤ e
T( div v
∞,T
δ
+C(v)∆t)
. (4.19)
10 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Lemma 4.6. Under Hypothesis 1 if ∆t < min{η, 1/(2L
∞,T
δ ), K}, where K is
the constant appearing in Corollary 4.4, then, ∀p ∈ Ω and ∀n ∈ {0, . . . , N}, we have
¯ c
1
≤ det F
n
RK
(p) ≤
C
1
, ¯ c
2
≤ (F
n
RK
)
−T
(p)u ≤
C
2
, (4.20)
being ¯ c
j
> 0,
C
j
> 0, j = 1, 2, constants depending on v and T, and u ∈ R
d
with
u = 1.
Proof. The result follows from expressions (4.10), (4.11), (4.15), (4.16) and (4.19),
and by using the following equality
1 = u =
¸
¸
(F
n
RK
)
T
(p)(F
n
RK
)
−T
(p)u
¸
¸
∀u ∈ R
d
, u = 1. (4.21)
Now, we consider a motion satisfying the following assumption:
Hypothesis 2. The motion X
e
satisﬁes
Ω
t
= Ω X
e
(p, t) = p ∀p ∈ Γ ∀t ∈ [0, T].
Remark 4.2. Notice that, if the motion veriﬁes Hypothesis 2 then
Γ
t
= Γ, v(x, t) = 0 ∀x ∈ Γ ∀t ∈ [0, T].
Under Hypothesis 2, Lemma 4.1 can be improved.
Lemma 4.7. Let us assume Hypothesis 2. If ∆t < min{K, 1/(2L
∞,T
)}, then,
X
n
RK
(p) is deﬁned ∀p ∈ Ω and ∀n ∈ {0, . . . , N}, and X
n
RK
(Ω) = Ω.
Proof. See Proposition 1 in [25].
In order to introduce approximations to the characteristic curves and gradient tensors
in scheme (4.1), some additional assumptions are required.
Firstly, we introduce a set containing X
n
RK
(Ω), for every 0 ≤ n ≤ N, namely
O
δ
:=
_
t∈[0,T]
Ω
δ
t
. (4.22)
Moreover, we deﬁne
T
δ
Γ
R :=
_
t∈[0,T]
G
δ
t
×{t}, (4.23)
being
G
δ
t
=
_
x∈Γ
R
t
B(x, δ). (4.24)
Hypothesis 3. Function ρ is deﬁned in O
δ
and belongs to W
1,∞
(O
δ
), being O
δ
the set deﬁned in (4.22). Moreover,
0 < γ ≤ ρ(x) a.e. x ∈ O
δ
. (4.25)
Let us denote ρ
1,∞
= ρ
1,∞,O
δ .
HIGHER ORDER PURE LAGRANGIAN METHOD 11
Hypothesis 4. The diﬀusion tensor, A, is deﬁned in O
δ
and belongs to W
1,∞
(O
δ
).
Moreover, A is symmetric and has the following form:
A =
_
A
n1
Θ
Θ Θ
_
, (4.26)
with A
n1
being a positive deﬁnite symmetric n
1
× n
1
tensor (n
1
≥ 1) and Θ an
appropriate zero tensor. Besides, there exists a strictly positive constant, Λ, which is
a uniform lower bound for the eigenvalues of A
n1
.
Remark 4.3. Notice that the diﬀusion tensor can be degenerate in some applica
tions. This is the case, for instance, in some ﬁnancial models where, nevertheless, the
diﬀusion tensor satisﬁes Hypothesis 4.
Hypothesis 5. Function f is deﬁned in T
δ
and it is continuous with respect to the
time variable, in space L
2
.
Hypothesis 6. Function g is deﬁned in T
δ
Γ
R
and it is continuous with respect to
the time variable, in space H
1
. Besides, coeﬃcient α in boundary condition (2.9) is
strictly positive.
Let us deﬁne the following sequences of functions of p.
¯
A
n
RK
:= (F
n
RK
)
−1
A◦ X
n
RK
(F
n
RK
)
−T
det F
n
RK
,
¯ m
n
RK
= (F
n
RK
)
−T
m det F
n
RK
,
for 0 ≤ n ≤ N. Since usually the characteristic curves cannot be exactly computed,
we replace in (4.1) the exact characteristic curves and gradient tensors by accurate
enough approximations,
1
2
_
Ω
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_ φ
n+1
m,∆t
−φ
n
m,∆t
∆t
ψ dp
+
1
4
_
Ω
_
¯
A
n+1
RK
+
¯
A
n
RK
__
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
· ∇ψ dp
+
α
4
_
Γ
R
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
ψ dA
p
=
1
2
_
Ω
_
det F
n+1
RK
f
n+1
◦ X
n+1
RK
+ det F
n
RK
f
n
◦ X
n
RK
_
ψ dp
+
1
2
_
Γ
R
_
¯ m
n+1
RK
g
n+1
◦ X
n+1
RK
+ ¯ m
n
RK
g
n
◦ X
n
RK
_
ψdA
p
.
(4.27)
For these computations we have made the assumptions of Lemma 4.3, and Hypothesis
3, 4, 5 and 6.
Notice that we have used a lowest order characteristics approximation formula pre
serving second order time accuracy.
Let us introduce L
n+
1
2
∆t
[φ] ∈ (H
1
(Ω))
′
and F
n+
1
2
∆t
∈ (H
1
(Ω))
′
deﬁned by
_
L
n+
1
2
∆t
[φ], ψ
_
:=
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
2
φ
n+1
−φ
n
∆t
, ψ
_
Ω
+
_
_
¯
A
n+1
RK
+
¯
A
n
RK
_
2
_
∇φ
n+1
+∇φ
n
_
2
, ∇ψ
_
Ω
+ α
_
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
2
_
φ
n+1
+ φ
n
_
2
, ψ
_
Γ
R
,
12 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
_
F
n+
1
2
∆t
, ψ
_
:=
_
det F
n+1
RK
f
n+1
◦ X
n+1
RK
+ det F
n
RK
f
n
◦ X
n
RK
2
, ψ
_
Ω
+
_
¯ m
n+1
RK
g
n+1
◦ X
n+1
RK
+ ¯ m
n
RK
g
n
◦ X
n
RK
2
, ψ
_
Γ
R
,
for φ ∈ C
0
(H
1
(Ω)) and ψ ∈ H
1
(Ω).
Remark 4.4. Regarding the deﬁnitions of L
n+
1
2
∆t
[φ] and F
n+
1
2
∆t
, only the values of
function φ at discrete time steps {t
n
}
N
n=0
are required. Thus, the above deﬁnitions
can also be stated for a sequence of functions
´
φ = {φ
n
}
N
n=0
∈ [H
1
(Ω)]
N+1
.
Then the semidiscretized time scheme can be written as follows:
_
Given φ
0
m,∆t
, ﬁnd
φ
m,∆t
= {φ
n
m,∆t
}
N
n=1
∈
_
H
1
Γ
D
(Ω)
¸
N
such that
_
L
n+
1
2
∆t
[
φ
m,∆t
], ψ
_
=
_
F
n+
1
2
∆t
, ψ
_
∀ ψ ∈ H
1
Γ
D(Ω) for n = 0, . . . , N −1.
(4.28)
Remark 4.5. The stability and convergence properties to be studied in the next
sections still remain valid if we replace the approximation of characteristics appearing
in scheme (4.28) by higher order ones or by the exact value.
4.3. Stability of the semidiscretized scheme. In order to prove stability
estimates for problem (4.28), the assumptions considered in the previous section are
required.
Firstly, we notice that, as a consequence of Hypothesis 4, there exists a unique positive
deﬁnite symmetric n
1
× n
1
tensor ﬁeld, C
n1
, such that A
n1
= (C
n1
)
2
. Let us denote
by C the symmetric and positive semideﬁnite d ×d tensor deﬁned by
C =
_
C
n1
Θ
Θ Θ
_
. (4.29)
Notice that A = C
2
and C ∈ W
1,∞
(O
δ
). Let us denote by G the matrix with
coeﬃcients G
ij
=  gradC
ij
, 1 ≤ i, j ≤ d. At this point, let us introduce the constant
c
A
= max{G
2
∞, O
δ
, C
2
∞, O
δ
}, (4.30)
and the sequence of tensor ﬁelds
¯
C
n
RK
:= C ◦ X
n
RK
(F
n
RK
)
−T
_
det F
n
RK
∀n ∈ {0, . . . , N}.
Next, let us denote by B the d ×d tensor
B =
_
I
n1
Θ
Θ Θ
_
, (4.31)
where I
n1
is the n
1
×n
1
identity matrix. Clearly, under Hypothesis 4 we have
ΛBw
2
Ω
≤ Aw, w
Ω
∀w ∈ R
d
. (4.32)
Let us introduce the sequence of tensor ﬁelds
¯
B
n
RK
:= B(F
n
RK
)
−T
_
det F
n
RK
∀n ∈ {0, . . . , N}.
As far as the velocity ﬁeld is deﬁned in T
δ
(see Hypothesis 1), we can introduce the
following assumption:
HIGHER ORDER PURE LAGRANGIAN METHOD 13
Hypothesis 7. The velocity ﬁeld satisﬁes
(I −B)L(x, t)B = 0 ∀(x, t) ∈ T
δ
. (4.33)
Remark 4.6. Hypothesis 7 is equivalent to having a velocity ﬁeld v whose d −n
1
last components depend only on the last d −n
1
variables.
Remark 4.7. For any d×d tensor E of the form given in (4.26) it is easy to check
that
EH
T
w
1
, w
2
= EH
T
Bw
1
, Bw
2
,
for any d × d tensor H satisfying (I − B)HB = 0, and vectors w
1
, w
2
∈ R
d
. This
equality will be used below without explicitly stated. Moreover, under Hypothesis 7,
if ∆t < min{η, 1/(2L
∞,T
δ )} it is easy to prove that
B(F
n
RK
)
−T
(p)w ≥ DBw,
for p ∈ Ω, w ∈ R
d
, n = 0, . . . , N, and D depending on v and T.
Now, it is convenient to notice that Hypothesis 4 also covers the nondegenerate case.
This hypothesis is usual in ultraparabolic equations (see, for instance, [24]), which
represent a wide class of degenerate diﬀusion equations arising from many applications
(see, for instance, [5]). Furthermore, as stated in [19], ultraparabolic problems either
have C
∞
solutions or can be reduced to nondegenerate problems posed in a lower
spatial dimension. This is an important point, as the stability and error estimates
will be obtained under regularity assumptions on the solution.
In what follows, c
v
denotes the positive constant
c
v
:= max
t∈[0,T]
v(·, t)
1,∞,Ω
δ
t
, (4.34)
where  · 
1,∞,Ω
δ
t
is the norm given in (2.12). Moreover, C
v
(respectively, J and D)
will denote a generic positive constant, related to the norm of the velocity ﬁeld v
(respectively, to the rest of the data of the problem), not necessarily the same at each
occurrence.
Corresponding to the semidiscretized scheme, we have to deal with sequences of
functions
´
ψ = {ψ
n
}
N
n=0
. Thus, we will consider the spaces of sequences l
∞
(L
2
(Ω))
and l
2
(L
2
(Ω)) equipped with their respective usual norms:
¸
¸
¸
¸
¸
¸
´
ψ
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
:= max
0≤n≤N
ψ
n

Ω
,
¸
¸
¸
¸
¸
¸
´
ψ
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
:=
¸
¸
¸
_
∆t
N
n=0
ψ
n

2
Ω
. (4.35)
Similar deﬁnitions are considered for functional spaces l
∞
(L
2
(Γ
R
)) and l
2
(L
2
(Γ
R
))
associated with the Robin boundary condition and for vectorvalued function spaces
l
∞
(L
2
(Ω)) and l
2
(L
2
(Ω)). Moreover, let us introduce the notations
¯
S[ψ] := {ψ
n+1
+ ψ
n
}
N−1
n=0
,
R
∆t
[ψ] :=
_
ψ
n+1
−ψ
n
∆t
_
N−1
n=0
.
We denote by
f ◦ X
RK
and by
g ◦ X
RK
the following sequences of functions
f ◦ X
RK
:= {f
n
◦ X
n
RK
}
N
n=0
,
g ◦ X
RK
:= {g
n
◦ X
n
RK
}
N
n=0
.
14 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Before establishing some technical lemmas, let us recall the Young’s inequality
ab ≤
1
2
_
ǫa
2
+
1
ǫ
b
2
_
, (4.36)
for a, b ∈ R and ǫ > 0, which will be extensively used in what follows.
Lemma 4.8. Let us assume Hypotheses 1, 3 and 4. Let {φ
n
m,∆t
}
N
n=1
be the solution
of (4.28). Then, there exist a positive constant c(v, T, δ) such that, for ∆t < c, we
have
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
+ φ
n
m,∆t
_
≥
1
∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n+1
RK
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
−
1
∆t
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n
RK
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
1
4
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
_
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_¸
¸
¸
¸
¸
¸
2
Ω
+
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
_
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_¸
¸
¸
¸
¸
¸
2
Ω
(4.37)
+
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
−´cγ
_
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
¸
¸
_
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
_
,
where ´c = ρ
1,∞
(c
v
+ C
v
∆t)/γ and n ∈ {0, . . . , N −1}.
Proof. First, we decompose
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
+ φ
n
m,∆t
_
= I
1
+ I
2
+ I
3
, with
I
1
=
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
2
φ
n+1
m,∆t
−φ
n
m,∆t
∆t
, φ
n+1
m,∆t
+ φ
n
m,∆t
_
Ω
,
I
2
=
1
4
__
¯
A
n+1
RK
+
¯
A
n
RK
__
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
, ∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
Ω
,
I
3
=
α
4
_
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
, φ
n+1
m,∆t
+ φ
n
m,∆t
_
Γ
R
.
Let K be the constant appearing in Corollary 4.4. If ∆t < K, we ﬁrst have
I
1
=
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
2
φ
n+1
m,∆t
−φ
n
m,∆t
∆t
, φ
n+1
m,∆t
+ φ
n
m,∆t
_
Ω
=
1
2∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n+1
RK
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
−
1
2∆t
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n
RK
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
1
2∆t
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n
RK
det F
n
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
2∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n+1
RK
det F
n+1
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
,
(4.38)
where we have used Hypothesis 3. Next, we introduce the function Y
n
RK
(p, ·) :
[t
n
, t
n+1
] −→ Ω
δ
tn
, deﬁned by Y
n
RK
(p, s) := X
n
RK
(p) − (t
n
− s)v
n+
1
2
(Y
n
(p)), which
satisﬁes Y
n
RK
(p, t
n
) = X
n
RK
(p) and Y
n
RK
(p, t
n+1
) = X
n+1
RK
(p). If ∆t is small enough,
it is easy to prove that Y
n
RK
(p, ·) ⊂ Ω
δ
tn
. By hypothesis, ρ is a diﬀerentiable function,
then by Barrow’s rule and the chain rule, the following identity holds:
ρ(X
n
RK
(p)) = ρ(X
n+1
RK
(p)) −ζ
n
(p) for a.e. p ∈ Ω, (4.39)
HIGHER ORDER PURE LAGRANGIAN METHOD 15
where
ζ
n
(p) :=
_
tn+1
tn
gradρ(Y
n
RK
(p, s)) · v
n+
1
2
(Y
n
(p)) ds for a.e. p ∈ Ω, (4.40)
veriﬁes ζ
n
(p) ≤ ρ
1,∞
c
v
∆t. Then, by using (4.17), (4.18) and (4.39) in (4.38), we get
I
1
≥
1
∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n+1
RK
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
−
1
∆t
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n
RK
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
(4.41)
− ρ
1,∞
(c
v
+ C
v
∆t)
_
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
¸
¸
_
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
_
.
For I
2
we use the fact that A = C
2
being C a symmetric tensor ﬁeld. We obtain,
I
2
:=
1
4
__
¯
A
n+1
RK
+
¯
A
n
RK
__
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
, ∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
Ω
(4.42)
=
1
4
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
_
∇φ
n+1
m,∆t
+ ∇φ
n
m,∆t
_¸
¸
¸
¸
¸
¸
2
Ω
+
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
_
∇φ
n
m,∆t
+∇φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
2
Ω
.
For I
3
we have
I
3
=
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
. (4.43)
Then, by summing up (4.41), (4.42) and (4.43) we get inequality (4.37).
Lemma 4.9. Let us assume Hypotheses 1, 3, 4 and 7. Let {φ
n
m,∆t
}
N
n=1
be the
solution of (4.28) and α > 0 be the constant appearing in the Robin boundary condition
(2.9). Then, there exist a positive constant c(v, T, δ) such that, for ∆t < c, we have
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
− φ
n
m,∆t
_
≥
1
2∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
_
φ
n+1
m,∆t
−φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
1
2
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
2
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
(4.44)
+
α
2
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
−
α
2
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
−´c∆tΛ
_
¸
¸
¸
¸
¸
¸
¯
B
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
¸
¸
¯
B
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
_
−´c∆tα
_
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
+
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
_
,
where ´c = max {c
A
C
v
/Λ, C
v
} and n ∈ {0, . . . , N −1}.
Proof. First, we decompose
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
−φ
n
m,∆t
_
= I
1
+ I
2
+ I
3
, with
I
1
=
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
2
φ
n+1
m,∆t
−φ
n
m,∆t
∆t
, φ
n+1
m,∆t
−φ
n
m,∆t
_
Ω
,
I
2
=
1
4
__
¯
A
n+1
RK
+
¯
A
n
RK
__
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_
, ∇φ
n+1
m,∆t
−∇φ
n
m,∆t
_
Ω
,
I
3
=
α
4
_
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
, φ
n+1
m,∆t
−φ
n
m,∆t
_
Γ
R
.
16 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
For I
1
, we use Hypothesis 3 to get
I
1
=
1
2∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
_
φ
n+1
m,∆t
−φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
,
(4.45)
where we have assumed that ∆t < K, being K the constant appearing in Corollary
4.4. For I
2
we ﬁrst have
I
2
=
1
4
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
(4.46)
+
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
4
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
.
Then we use Corollary 4.5, Hypotheses 4 and 7, and equality (4.7) to get
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
≥
1
4
¸
¸
¸
¸
¸
¸
¸
¸
C ◦ X
n
RK
(F
n+1
RK
)
−T
∇φ
n+1
m,∆t
_
det F
n+1
RK
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
(4.47)
−c
A
C
v
∆t
¸
¸
¸
¸
¸
¸
¯
B
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
.
Moreover, since A
n1
is symmetric and positive deﬁnite, C
n1
=
_
A
n1
is a diﬀerentiable
tensor ﬁeld. Then by Barrow’s rule and the chain rule, the following identity holds,
C(X
n+1
RK
(p)) = C(X
n
RK
(p)) + D
n
(p) for a.e. p ∈ Ω, (4.48)
where we have denoted by D
n
the d ×d symmetric tensor ﬁeld deﬁned by
D
n
ij
(p) :=
_
tn+1
tn
gradC
ij
(Y
n
RK
(p, s)) · v
n+
1
2
(Y
n
(p)) ds, (4.49)
being Y
n
RK
the mapping deﬁned in the proof of Lemma 4.8. Notice that D is of the
form given in (4.29) and veriﬁes D
n

∞,Ω
≤ c
v
√
c
A
∆t. Then, from the previous
properties, we have
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
≥
1
4
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−c
A
C
v
∆t
¸
¸
¸
¸
¸
¸
¯
B
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
. (4.50)
Similarly, we obtain the estimate
−
1
4

¯
C
n+1
RK
∇φ
n
m,∆t

2
Ω
≥ −
1
4

¯
C
n
RK
∇φ
n
m,∆t

2
Ω
−c
A
C
v
∆t
¯
B
n
RK
∇φ
n
m,∆t

2
Ω
. (4.51)
Thus, by introducing (4.50) and (4.51) in equality (4.46) we obtain the following
inequality:
I
2
≥
1
2
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
2
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
(4.52)
− c
A
C
v
∆t
¸
¸
¸
¸
¸
¸
¯
B
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−c
A
C
v
∆t
¸
¸
¸
¸
¸
¸
¯
B
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
.
For I
3
we ﬁrst have
I
3
=
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
−
α
4
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
(4.53)
+
α
4
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
−
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
.
HIGHER ORDER PURE LAGRANGIAN METHOD 17
Next, by applying Corollaries 4.4, 4.5, Lemma 4.3 and equality (4.7) we obtain
I
3
≥
α
2
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
−
α
2
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
(4.54)
− C
v
α∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
+
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
_
.
Then, by summing up (4.45), (4.52) and (4.54), inequality (4.44) follows.
Now, in order to get error estimates we need to prove stability inequalities for more
general righthand sides, namely for the problem,
_
Given φ
0
m,∆t
, ﬁnd
φ
m,∆t
= {φ
n
m,∆t
}
N
n=1
∈
_
H
1
Γ
D
(Ω)
¸
N
such that
_
L
n+
1
2
∆t
[
φ
m,∆t
], ψ
_
=
_
H
n+
1
2
∆t
, ψ
_
∀ψ ∈ H
1
Γ
D(Ω) for n = 0, . . . , N −1,
(4.55)
with
_
H
n+
1
2
∆t
, ψ
_
=
¸
S
n+1
, ψ
_
Ω
+
¸
G
n+1
, ψ
_
Γ
R
.
Hypothesis 8.
´
S = {S
n
}
N
n=1
∈ [L
2
(Ω)]
N
and
´
G = {G
n
}
N
n=1
∈ [L
2
(Γ
R
)]
N
.
Lemma 4.10. Let us assume Hypotheses 1 and 8. Let us suppose α > 0 and
∆t < min{η, 1/(2L
∞,T
δ ), K}, being η and K the constants appearing, respectively,
in Lemma 4.1 and in Corollary 4.4. Then, we have
S
n+1
, ψ + ϕ
Ω
+G
n+1
, ψ + ϕ
Γ
R ≤ c
s
S
n+1

2
Ω
+
1
2
_
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
ψ
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
¸
¸
_
det F
n
RK
ϕ
¸
¸
¸
¸
¸
¸
2
Ω
_
+
4c
g
α
G
n+1

2
Γ
R
+
α
32
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
+ ¯ m
n
RK
(ϕ + ψ)
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
,
(4.56)
∀ϕ, ψ ∈ H
1
(Ω), with c
s
= 1/ ¯ c
1
and c
g
= 1/( ¯ c
1
¯ c
2
), where ¯ c
1
and ¯ c
2
are the constants
appearing in Lemma 4.6.
Proof. The estimate follows directly by applying the CauchySchwarz inequality
to the lefthand side of (4.56), and using inequality (4.36) and Lemma 4.6.
Theorem 4.11. Let us assume Hypotheses 1, 3, 4 and 8. Let
φ
m,∆t
be the solu
tion of (4.55) subject to the initial value φ
0
m,∆t
∈ H
1
Γ
D
(Ω) and α > 0 be the constant
appearing in the Robin boundary condition (2.9). Then there exist two positive con
stants J and D, which are independent of the diﬀusion tensor, such that if ∆t < D
then
√
γ
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
RK
φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
Λ
4
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
S[∇φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
α
8
¸
¸
¸
¸
¸
¸
¸
¸
_
S [ ¯ m
RK
]S[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤ J
_
√
γφ
0
m,∆t

Ω
+
´
S
l
2
(L
2
(Ω))
+
´
G
l
2
(L
2
(Γ
R
))
_
.
(4.57)
where
´
S = {S
n
}
N
n=1
,
´
G = {G
n
}
N
n=1
.
18 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Proof. Sequence
φ
m,∆t
= {φ
n
m,∆t
}
N
n=0
satisﬁes
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
+ φ
n
m,∆t
_
=
_
H
n+
1
2
∆t
, φ
n+1
m,∆t
+ φ
n
m,∆t
_
. We can use Lemma 4.8 to obtain a lower bound of this
expression, and Lemma 4.10 for ψ = φ
n+1
m,∆t
and ϕ = φ
n
m,∆t
to obtain an upper bound.
By jointly considering both estimates, we get
1
∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n+1
RK
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
−
1
∆t
¸
¸
¸
¸
¸
¸
_
ρ ◦ X
n
RK
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
1
4
¸
¸
¸
¸
¸
¸
¯
C
n
RK
_
∇φ
n+1
m,∆t
+∇φ
n
m,∆t
_¸
¸
¸
¸
¸
¸
2
Ω
+
α
8
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
+ ¯ m
n
RK
_
φ
n+1
m,∆t
+ φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
≤ c
s
S
n+1

2
Ω
+
4c
g
α
G
n+1

2
Γ
R
+´cγ
_
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
_
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
2
Ω
_
,
(4.58)
where ´c = max {1/γ, 2ρ
1,∞
(c
v
+ C
v
∆t)/γ}. Let us introduce the notation
θ
1
n
:= γ
¸
¸
¸
¸
¸
¸
_
det F
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
,
θ
2
n
:=
Λ
4
n−1
s=0
∆t
¸
¸
¸
¸
¸
¸
¯
B
s
RK
_
∇φ
s+1
m,∆t
+∇φ
s
m,∆t
_¸
¸
¸
¸
¸
¸
2
Ω
,
θ
n
:=
α
8
n−1
s=0
∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
s+1
RK
+ ¯ m
s
RK
_
φ
s+1
m,∆t
+ φ
s
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
.
Now, for a ﬁxed integer q ≥ 1, let us sum (4.58) multiplied by ∆t from n = 0 to
n = q −1. Then, with the above notation we have
(1 −´c∆t)θ
1
q
+ θ
2
q
+ θ
q
≤ 2´c∆t
q−1
n=0
θ
1
n
+ β
_
θ
1
0
+
´
S
2
l
2
(L
2
(Ω))
+
´
G
2
l
2
(L
2
(Γ
R
))
_
,
where we have used Hypotheses 3 and 4. In the above equation β denotes a positive
constant and ´c = max {1/γ, 2ρ
1,∞
(c
v
+ C
v
∆t)/γ}. For ∆t small enough, we can
apply the discrete Gronwall inequality (see, for instance, [23]) and take the maximun
in q ∈ {1, . . . , N}. Then, estimate (4.57) follows.
Corollary 4.12. Let us assume Hypotheses 1, 3, 4, 5 and 6. Let
φ
m,∆t
be the
solution of (4.28) subject to the initial value φ
0
m,∆t
∈ H
1
Γ
D
(Ω). Then, there exist two
positive constants J and D, independent of the diﬀusion tensor and such that, for
∆t < D, we have
√
γ
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
RK
φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
Λ
4
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
S[∇φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
α
8
¸
¸
¸
¸
¸
¸
¸
¸
_
S [ ¯ m
RK
]S[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤ J
_
√
γφ
0
m,∆t

Ω
+
¸
¸
¸
¸
¸
¸
det F
RK
f ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
¯ m
RK
g ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
_
.
(4.59)
HIGHER ORDER PURE LAGRANGIAN METHOD 19
Proof. The result follows directly by replacing
S
n+1
with 1/2
_
det F
n+1
RK
f
n+1
◦ X
n+1
RK
+ det F
n
RK
f
n
◦ X
n
RK
_
and G
n+1
with 1/2
_
¯ m
n+1
RK
g
n+1
◦ X
n+1
RK
+ ¯ m
n
RK
g
n
◦ X
n
RK
_
in (4.57).
Lemma 4.13. Let us assume Hypotheses 1 and 8. Let ∆t < min{η, K}, being
η and K the constants appearing in Lemma 4.1 and in Corollary 4.4, respectively.
Then, we have
¸
S
n+1
, ψ −ϕ
_
Ω
≤
2c
s
∆t
γ
S
n+1

2
Ω
+
γ
16∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
+ det F
n
RK
(ψ −ϕ)
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
,
(4.60)
∀ϕ, ψ ∈ L
2
(Ω), where c
s
is the constant appearing in Lemma 4.10.
Proof. The result easily follows by applying the CauchySchwarz inequality, in
equality (4.36) with ǫ = 8∆t/γ and Lemma 4.6.
Lemma 4.14. Let us assume Hypotheses 1 and 8. Suppose that α > 0 and
∆t < min{η, 1/(2L
∞,T
δ ), K}. Then, for any sequence {ψ
n
}
N
n=0
∈ [L
2
(Γ
R
)]
N+1
and any q ∈ {1, . . . , N}, the following inequality holds:
¸
¸
¸
¸
¸
q−1
n=0
G
n+1
, ψ
n+1
−ψ
n
Γ
R
¸
¸
¸
¸
¸
≤
4c
g
α
G
q

2
Γ
R +
α
16

_
¯ m
q
RK
ψ
q

2
Γ
R +
1
2α
G
1

2
Γ
R
+
α
2
ψ
0

2
Γ
R +
∆tc
g
2α
q−1
n=1
¸
¸
¸
¸
¸
¸
¸
¸
G
n+1
−G
n
∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
+
∆tα
2
q−1
n=1

_
¯ m
n
RK
ψ
n

2
Γ
R.
(4.61)
Proof. The result follows from the equality
q−1
n=0
G
n+1
, ψ
n+1
−ψ
n
Γ
R = G
q
, ψ
q
Γ
R −G
1
, ψ
0
Γ
R (4.62)
−∆t
q−1
n=1
_
G
n+1
−G
n
∆t
, ψ
n
_
Γ
R
.
Indeed, the three terms on the righthand side can be bounded by using the Cauchy
Schwarz inequality, inequality (4.36) and Lemma 4.6.
Theorem 4.15. Let us assume Hypotheses 1, 3, 4, 7 and 8, and let
φ
m,∆t
be
the solution of (4.55) subject to the initial value φ
0
m,∆t
∈ H
1
Γ
D
(Ω). Let α > 0 be
the constant appearing in the Robin boundary condition (2.9). Then, there exist two
positive constants J(v, c
A
/Λ, T) and D(δ, v, T, c
A
/Λ) such that if ∆t < D then
_
γ
4
¸
¸
¸
¸
¸
¸
¸
¸
_
S[ det F
RK
]R
∆t
[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
Λ
2
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
∇φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
RK
φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
≤ J
_
_
Λ
2
¸
¸
¸
¸
B∇φ
0
m,∆t
¸
¸
¸
¸
Ω
+
_
α
4
¸
¸
¸
¸
φ
0
m,∆t
¸
¸
¸
¸
Γ
R
+
´
S
l
2
(L
2
(Ω))
+
´
G
l
∞
(L
2
(Γ
R
))
+
¸
¸
¸
¸
¸
¸
R
∆t
[G]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
_
.
(4.63)
20 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Proof. Sequence
φ
m,∆t
= {φ
n
m,∆t
}
N
n=0
satisﬁes
_
L
n+
1
2
∆t
[
φ
m,∆t
], φ
n+1
m,∆t
−φ
n
m,∆t
_
=
_
H
n+
1
2
∆t
, φ
n+1
m,∆t
−φ
n
m,∆t
_
. Then, we use Lemma 4.9 and Lemma 4.13 for ψ = φ
n+1
m,∆t
and ϕ = φ
n
m,∆t
to obtain, respectively, a lower and an upper bound for this expression.
By jointly considering both estimates, we get
1
2∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
_
φ
n+1
m,∆t
−φ
n
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
1
2
¸
¸
¸
¸
¸
¸
¯
C
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
−
1
2
¸
¸
¸
¸
¸
¸
¯
C
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
α
2
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
−
α
2
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
≤ ´c∆tΛ
_
¸
¸
¸
¸
¸
¸
¯
B
n+1
RK
∇φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
+
¸
¸
¸
¸
¸
¸
¯
B
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
_
+´c∆tα
_
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
n+1
RK
φ
n+1
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
2
Γ
R
+
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
_
+
2c
s
∆t
γ
S
n+1

2
Ω
+
γ
16∆t
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
n+1
RK
+ det F
n
RK
(φ
n+1
m,∆t
−φ
n
m,∆t
)
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
+
_
G
n+1
, φ
n+1
m,∆t
−φ
n
m,∆t
_
Γ
R
,
(4.64)
with ´c = max {c
A
C
v
/Λ, C
v
}. For n = 0, . . . , N, let us introduce the notations
θ
1
n
:=
γ
4∆t
n−1
s=0
¸
¸
¸
¸
¸
¸
¸
¸
_
det F
s+1
RK
+ det F
s
RK
_
φ
s+1
m,∆t
−φ
s
m,∆t
_
¸
¸
¸
¸
¸
¸
¸
¸
2
Ω
,
θ
2
n
:=
Λ
2
¸
¸
¸
¸
¸
¸
¯
B
n
RK
∇φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Ω
, θ
n
:=
α
4
¸
¸
¸
¸
¸
¸
_
¯ m
n
RK
φ
n
m,∆t
¸
¸
¸
¸
¸
¸
2
Γ
R
.
Now, for a ﬁxed q ≥ 1, let us sum (4.64) from n = 0 to n = q − 1. With the above
notation and by using Lemma 4.14 for
´
ψ =
φ
m,∆t
, we get
θ
1
q
+ (1 −2´c∆t)θ
2
q
+ (1 −4´c∆t)θ
q
≤ 4´c∆t
q−1
n=0
θ
2
n
+ 10´c∆t
q−1
n=0
θ
n
+β
_
θ
2
0
+ θ
0
+
´
S
2
l
2
(L
2
(Ω))
+
´
G
2
l
∞
(L
2
(Γ
R
))
+
¸
¸
¸
¸
¸
¸
R
∆t
[G]
¸
¸
¸
¸
¸
¸
2
l
2
(L
2
(Γ
R
))
_
, (4.65)
where we have used Hypotheses 3 and 4. In the above equation ´c = max {c
A
C
v
/Λ, C
v
}
and β denotes a positive constant. For ∆t small enough, we can apply the discrete
Gronwall inequality (see, for instance, [23]) and take the maximun in q ∈ {1, . . . , N}.
Thus, estimate (4.63) follows.
Corollary 4.16. Let us assume Hypotheses 1, 3, 4, 5, 6 and 7, and let
φ
m,∆t
be the solution of (4.28) subject to the initial value φ
0
m,∆t
∈ H
1
Γ
D
(Ω). Then there exist
HIGHER ORDER PURE LAGRANGIAN METHOD 21
two positive constants J(v, c
A
/Λ, T) and D(δ, v, T, c
A
/Λ) such that if ∆t < D then
_
γ
4
¸
¸
¸
¸
¸
¸
¸
¸
_
S[ det F
RK
]R
∆t
[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
Λ
2
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
∇φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
RK
φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
≤ J
_
_
Λ
2
¸
¸
¸
¸
B∇φ
0
m,∆t
¸
¸
¸
¸
Ω
+
_
α
4
¸
¸
¸
¸
φ
0
m,∆t
¸
¸
¸
¸
Γ
R
+
¸
¸
¸
¸
¸
¸
det F
RK
f ◦ X
RK
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
¸
¸
¸
¸
¸
¸
¯ m
RK
g ◦ X
RK
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
+
¸
¸
¸
¸
¸
¸
R
∆t
[ ¯ m
RK
g ◦ X
RK
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
_
.
(4.66)
Proof. The result follows directly by replacing
S
n+1
with 1/2
_
det F
n+1
RK
f
n+1
◦ X
n+1
RK
+ det F
n
RK
f
n
◦ X
n
RK
_
and G
n+1
with 1/2
_
¯ m
n+1
RK
g ◦ X
n+1
RK
+ ¯ m
n
RK
g ◦ X
n
RK
_
in (4.63).
Remark 4.8. Notice that, constants J and D appearing in Theorem 4.15 and
Corollary 4.16 depend on the diﬀusion tensor, more precisely they depend on fraction
c
A
Λ
. In most cases this fraction is bounded in the hyperbolic limit.
Remark 4.9. In the particular case of Dirichlet boundary conditions (Γ
D
≡ Γ),
diﬀusion tensor of the form A = ǫB and f = 0, the previous corollary can be improved.
Speciﬁcally, by using analogous procedures to the ones in the previous corollary we
can obtain the following l
∞
(H
1
) stability result with constants (J and D) independent
of the diﬀusion constant ǫ
_
γ
2
¸
¸
¸
¸
¸
¸
¸
¸
_
S[ det F
RK
]R
∆t
[φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
1
2
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
∇φ
m,∆t
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
≤ J(1 +
√
ǫ)
_
1
2
¸
¸
¸
¸
B∇φ
0
m,∆t
¸
¸
¸
¸
Ω
.
(4.67)
for ∆t < D.
4.4. Error estimate for the semidiscretized scheme. The aim of the present
section is to estimate the diﬀerence between the discrete solution of (4.28),
φ
m,∆t
:=
{φ
n
m,∆t
}
N
n=0
, and the exact solution of the continuous problem,
´
φ
m
:= {φ
n
m
}
N
n=0
. Ac
cording to (3.8) for t
n+
1
2
, with 0 ≤ n ≤ N −1, the latter solves the problem
_
L
n+
1
2
[
´
φ
m
], ψ
_
=
_
F
n+
1
2
, ψ
_
∀ψ ∈ H
1
Γ
D(Ω), (4.68)
where L
n+
1
2
[
´
φ
m
] ∈ (H
1
(Ω))
′
and F
n+
1
2
∈ (H
1
(Ω))
′
are deﬁned by
_
L
n+
1
2
[
´
φ
m
], ψ
_
:=
_
ρ ◦ X
n+
1
2
e
det F
n+
1
2
_
˙
φ
m
_
n+
1
2
, ψ
_
Ω
+
_
¯
A
n+
1
2
m
∇φ
n+
1
2
m
, ∇ψ
_
Ω
+ α
_
¯ m
n+
1
2
φ
n+
1
2
m
, ψ
_
Γ
R
,
_
F
n+
1
2
, ψ
_
:=
_
det F
n+
1
2
f
n+
1
2
◦ X
n+
1
2
e
, ψ
_
Ω
+
_
¯ m
n+
1
2
g
n+
1
2
◦ X
n+
1
2
e
, ψ
_
Γ
R
,
22 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
∀ψ ∈ H
1
(Ω).
The error estimate in the l
∞
(L
2
(Ω))norm, to be stated in Theorem 4.27, is proved
by means of Theorem 4.11 and the forthcoming Lemmas 4.25 and 4.26. On the other
hand, the error estimate for the gradient in the l
∞
(L
2
(Ω))norm, to be stated in
Theorem 4.28, is proved by means of Theorem 4.15 and the forthcoming Lemmas
4.25 and 4.26. Before doing this we give some results with sketched proofs (see [6] for
further details). Some auxiliary mappings will be introduced. They will be denoted
by ξ, u and Ψ depending on whether they are scalar, vector or tensor mappings,
respectively. Moreover, if v is smooth enough, it is easy to prove that F, F
−1
,
det F and their partial derivatives, as well as the ones of (F
n
RK
)
−1
and det F
n
RK
can
be bounded by constants depending only on v and T. These estimates and the ones
obtained in §4.2 for F
n
RK
, (F
n
RK
)
−1
and det F
n
RK
will be used below without explicitly
stated (see [6] for further details).
Lemma 4.17. Let us assume Hypotheses 1 and 3. Let us suppose that v ∈
C
2
(T
δ
), X
e
∈ C
4
(Ω ×[0, T]), ∆t < η, ϕ ∈ C
3
(L
2
(Ω)) and ρ
m
∈ C
2
(L
∞
(Ω)). Let us
deﬁne the function ξ
n+
1
2
: Ω −→R, for n ∈ {0, . . . , N −1}, by
ξ
n+
1
2
(p) := ρ ◦ X
n+
1
2
e
(p) det F
n+
1
2
(p) ˙ ϕ
n+
1
2
(p)
−
1
2
_
ρ ◦ X
n+1
RK
(p) det F
n+1
RK
(p) + ρ ◦ X
n
RK
(p) det F
n
RK
(p)
_
ϕ
n+1
(p) −ϕ
n
(p)
∆t
,
for a.e. p ∈ Ω. Then ξ
n+
1
2
∈ L
2
(Ω) and ξ
n+
1
2

Ω
≤ C(T, v, ρ)∆t
2
ϕ
C
3
(L
2
(Ω))
,
n = 0, . . . , N −1.
Proof. The result follows by using Taylor expansions and noting that if X
e
∈
C
3
(Ω × [0, T]) and v ∈ C
1
(T
δ
) then X
n
e
(p) −X
n
RK
(p) ≤ C(v, T)∆t
2
, and if X
e
∈
C
4
(Ω ×[0, T]) and v ∈ C
2
(T
δ
) then  det F
n
(p) − det F
n
RK
(p) ≤ C(v, T)∆t
2
.
Lemma 4.18. Let us assume that A
m
∈ C
2
(L
∞
(Ω)). Let w ∈ C
2
(L
2
(Ω)) be a
given mapping and u
n+
1
2
: Ω −→R
d
, for n ∈ {0, . . . , N −1}, be deﬁned by
u
n+
1
2
(p) :=
¯
A
n+
1
2
m
(p)w
n+
1
2
(p) −
_
¯
A
n+1
m
(p) +
¯
A
n
m
(p)
2
_
_
w
n+1
(p) +w
n
(p)
2
_
,
for a.e. p ∈ Ω. Then, u
n+
1
2
∈ L
2
(Ω) and u
n+
1
2

Ω
≤ C(T, v, A)∆t
2
w
C
2
(L
2
(Ω))
,
n ∈ {0, . . . , N − 1}. Moreover, if X
e
∈ C
4
(Ω × [0, T]), A
m
∈ C
2
(W
1,∞
(Ω)) and w ∈
C
2
(H
1
(Ω)) then u
n+
1
2
∈ H
1
(Ω) and  Div u
n+
1
2

Ω
≤ C(T, v, A)∆t
2
w
C
2
(H
1
(Ω))
,
n ∈ {0, . . . , N −1}.
Proof. The result follows by writing Taylor expansions in the time variable for w
and the tensor ﬁeld
¯
A
m
(p, s) := det F(p, s)F
−1
(p, s)A◦ X
e
(p, s)F
−T
(p, s), s ∈ [0, T].
Lemma 4.19. Let us assume Hypotheses 1 and 4. Let us suppose that v ∈
C
2
(T
δ
), X
e
∈ C
4
(Ω × [0, T]) and ∆t < min{η, 1/(2L
∞,T
δ )}, being η the constant
appearing in Lemma 4.1. Let w ∈ L
2
(Ω) be a given function and u
n
: Ω −→ R
d
be
deﬁned by
u
n
(p) :=
¯
A
n
m
(p)w(p) −
¯
A
n
RK
(p)w(p), 0 ≤ n ≤ N. (4.69)
Then, u
n
∈ L
2
(Ω) and u
n

Ω
≤ C(T, v, A)∆t
2
w
Ω
. Moreover, if v ∈ C
3
(T
δ
),
X
e
∈ C
5
(Ω × [0, T]), A ∈ W
2,∞
(O
δ
) and w ∈ H
1
(Ω), then u
n
∈ H
1
(Ω) and
HIGHER ORDER PURE LAGRANGIAN METHOD 23
 Div u
n

Ω
≤ C(T, v, A)∆t
2
w
1,2,Ω
.
Proof. The result follows by applying Taylor expansions, noting that if X
e
∈
C
4
(Ω × [0, T]) and v ∈ C
2
(T
δ
) then X
n
e
− X
n
RK

1,∞,Ω
≤ C(v, T)∆t
2
, (F
n
)
−T
−
(F
n
RK
)
−T

∞,Ω
≤ C(v, T)∆t
2
and  det F
n
− det F
n
RK

∞,Ω
≤ C(v, T)∆t
2
, and if
X
e
∈ C
5
(Ω ×[0, T]) and v ∈ C
3
(T
δ
) then (F
n
)
−T
−(F
n
RK
)
−T

1,∞,Ω
≤ C(v, T)∆t
2
and  det F
n
− det F
n
RK

1,∞,Ω
≤ C(v, T)∆t
2
.
Lemma 4.20. Let ϕ ∈ C
2
(L
2
(Γ
R
)) be a given mapping and ξ
n+
1
2
1
: Γ
R
−→ R,
ξ
n+
1
2
2
: Γ
R
−→R be deﬁned by
ξ
n+
1
2
1
(p) := ¯ m
n+
1
2
(p)ϕ
n+
1
2
(p) −
_
¯ m
n+1
(p) + ¯ m
n
(p)
2
__
ϕ
n+1
(p) + ϕ
n+1
(p)
2
_
,
ξ
n+
1
2
2
(p) := ¯ m
n+
1
2
(p)ϕ
n+
1
2
(p) −
_
¯ m
n+1
(p)ϕ
n+1
(p) + ¯ m
n
(p)ϕ
n
(p)
2
_
.
Then ξ
n+
1
2
1
, ξ
n+
1
2
2
∈ L
2
(Γ
R
) and
ξ
n+
1
2
1

Γ
R ≤
∆t
2
8
 ¯ mϕ
C
2
(L
2
(Γ
R
))
≤ C(T, v)∆t
2
ϕ
C
2
(L
2
(Γ
R
))
,
ξ
n+
1
2
2

Γ
R ≤ C(T, v)∆t
2
ϕ
C
2
(L
2
(Γ
R
))
.
Proof. The result follows by using Taylor expansions in the time variable.
Lemma 4.21. Let us assume Hypothesis 1. Let us suppose that v ∈ C
2
(T
δ
),
X
e
∈ C
4
(Ω×[0, T]) and ∆t < min{η, 1/(2L
∞,T
δ )}, being η the constant appearing
in Lemma 4.1. Let ϕ ∈ L
2
(Γ
R
) be a given function and ξ
n
: Γ
R
−→R be deﬁned by
ξ
n
(p) := ¯ m
n
(p)ϕ(p) − ¯ m
n
RK
(p)ϕ(p), 0 ≤ n ≤ N. (4.70)
Then ξ
n
∈ L
2
(Γ
R
) and ξ
n

Γ
R ≤ C(T, v)∆t
2
ϕ
Γ
R.
Proof. The result follows noting that  det F
n
(p) − det F
n
RK
(p) ≤ C(v, T)∆t
2
and
¸
¸
¸
¸
(F
n
)
−T
(p) −(F
n
RK
)
−T
(p)
¸
¸
¸
¸
2
≤ C(v, T)∆t
2
.
Lemma 4.22. Let us assume Hypotheses 1. Let us suppose v ∈ C
2
(T
δ
), X
e
∈
C
4
(Ω × [0, T]) and ∆t < min{η, 1/(2L
∞,T
δ )}, being η the constant appearing in
Lemma 4.1. Let ϕ ∈ H
1
(G
δ
tn
) be a given function, being G
δ
tn
the set deﬁned in (4.24),
and let ξ
n
: Γ
R
−→R be deﬁned by
ξ
n
(p) := ¯ m
n
(p)ϕ(X
n
e
(p)) − ¯ m
n
RK
(p)ϕ(X
n
RK
(p)), 0 ≤ n ≤ N. (4.71)
Then ξ
n
∈ L
2
(Γ
R
) and ξ
n

Γ
R ≤ C(T, v)∆t
2
ϕ
1,2,G
δ
tn
.
Proof. The result follows by applying Taylor expansions, noting that X
n
e
(p) −
X
n
RK
(p) ≤ C(v, T)∆t
2
, (F
n
)
−T
(p) − (F
n
RK
)
−T
(p) ≤ C(v, T)∆t
2
and  det F
n
(p) −
det F
n
RK
(p) ≤ C(v, T)∆t
2
.
Lemma 4.23. Let ϕ ∈ C
2
(L
2
(Ω)) be a given function and ξ
n+
1
2
: Ω −→ R, for
n ∈ {0, . . . , N −1}, be deﬁned by
ξ
n+
1
2
(p) := det F
n+
1
2
(p)ϕ
n+
1
2
(p) −
det F
n+1
(p)ϕ
n+1
(p) + det F
n
(p)ϕ
n
(p)
2
.
24 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Then, ξ
n+
1
2
∈ L
2
(Ω) and
¸
¸
¸
¸
¸
¸ξ
n+
1
2
¸
¸
¸
¸
¸
¸
Ω
≤
∆t
2
8
 det Fϕ
C
2
(L
2
(Ω))
≤ C(T, v)∆t
2
ϕ
C
2
(L
2
(Ω))
.
Proof. The result follows by applying Taylor expansions.
Lemma 4.24. Let us assume Hypothesis 1. Let us suppose that v ∈ C
2
(T
δ
),
X
e
∈ C
4
(Ω × [0, T]) and ∆t < η, being η the constant appearing in Lemma 4.1.
Let ϕ ∈ H
1
(Ω
δ
tn
) be a given function, being Ω
δ
tn
the set deﬁned in (4.9), and let
ξ
n
: Ω −→R be deﬁned by
ξ
n
(p) := det F
n
(p)ϕ(X
n
e
(p)) − det F
n
RK
(p)ϕ(X
n
RK
(p)), 0 ≤ n ≤ N.
Then ξ
n
∈ L
2
(Ω) and ξ
n

Ω
≤ C(T, v)∆t
2
ϕ
1,2,Ω
δ
tn
.
Proof. The result follows by using Taylor expansions, noting that X
n
e
(p) −
X
n
RK
(p) ≤ C(v, T)∆t
2
and  det F
n
(p) − det F
n
RK
(p) ≤ C(v, T)∆t
2
.
Lemma 4.25. Assume Hypotheses 1, 3 and 4 hold. Moreover, suppose that
X
e
∈ C
5
(Ω ×[0, T]) and that the coeﬃcients of problem (2.7)(2.10) satisfy,
v ∈ C
3
(T
δ
), ρ
m
∈ C
2
(L
∞
(Ω)), A ∈ W
2,∞
(O
δ
), A
m
∈ C
2
(W
1,∞
(Ω)).
Let the solution of (4.68) satisfy,
φ
m
∈ C
3
(L
2
(Ω)), ∇φ
m
∈ C
2
(H
1
(Ω)), φ
m

Γ
R ∈ C
2
(L
2
(Γ
R
)).
Finally, assume that ∆t < min{η, 1/(2L
∞,T
δ )}. Then, for each 0 ≤ n ≤ N − 1,
there exist two functions ξ
n+
1
2
LΩ
: Ω −→R and ξ
n+
1
2
LΓ
: Γ
R
−→R, such that
__
L
n+
1
2
−L
n+
1
2
∆t
_
[
´
φ
m
], ψ
_
=
_
ξ
n+
1
2
LΩ
, ψ
_
Ω
+
_
ξ
n+
1
2
LΓ
, ψ
_
Γ
R
, (4.72)
∀ψ ∈ H
1
Γ
D
(Ω). Moreover, ξ
n+
1
2
LΩ
∈ L
2
(Ω), ξ
n+
1
2
LΓ
∈ L
2
(Γ
R
) and the following estimates
hold:
¸
¸
¸
¸
¸
¸ξ
n+
1
2
LΩ
¸
¸
¸
¸
¸
¸
Ω
≤ ∆t
2
C(T, v, ρ, A)
_
φ
m

C
3
(L
2
(Ω))
+∇φ
m

C
2
(H
1
(Ω))
_
,
(4.73) ¸
¸
¸
¸
¸
¸ξ
n+
1
2
LΓ
¸
¸
¸
¸
¸
¸
Γ
R
≤ ∆t
2
C(T, v, A)
_
∇φ
m
· m
C
2
(L
2
(Γ
R
))
+ αφ
m

C
2
(L
2
(Γ
R
))
_
,
where α > 0 appears in (2.9).
Proof. The lefthand side of (4.72) is equal to I
1
+ I
2
+ I
3
, with
I
1
=
_
ρ ◦ X
n+
1
2
e
det F
n+
1
2
_
˙
φ
m
_
n+
1
2
, ψ
_
Ω
−
_
1
2
_
ρ ◦ X
n+1
RK
det F
n+1
RK
+ ρ ◦ X
n
RK
det F
n
RK
_
φ
n+1
m
−φ
n
m
∆t
, ψ
_
Ω
,
I
2
=
_
¯
A
n+
1
2
m
∇φ
n+
1
2
m
−
_
¯
A
n+1
RK
+
¯
A
n
RK
2
_
_
∇φ
n+1
m
+∇φ
n
m
2
_
, ∇ψ
_
Ω
I
3
= α
_
¯ m
n+
1
2
φ
n+
1
2
m
−
_
¯ m
n+1
RK
+ ¯ m
n
RK
2
__
φ
n+1
m
+ φ
n
m
2
_
, ψ
_
Γ
R
.
HIGHER ORDER PURE LAGRANGIAN METHOD 25
The bound for I
1
directly follows from Lema 4.17 for ϕ = φ
m
, so we can deﬁne a
function ξ
n+
1
2
I1
∈ L
2
(Ω) such that
I
1
=
_
ξ
n+
1
2
I1
, ψ
_
Ω
, with
¸
¸
¸
¸
¸
¸ξ
n+
1
2
I1
¸
¸
¸
¸
¸
¸
Ω
≤ C(T, v, ρ)∆t
2
φ
m

C
3
(L
2
(Ω))
. (4.74)
In order to estimate I
2
we apply Lemmas 4.18 and 4.19 for w = ∇φ
m
and w =
∇φ
n+1
m
+ ∇φ
n
m
, respectively, so I
2
=
_
u
n+
1
2
I2
, ∇ψ
_
Ω
, with u
n+
1
2
I2
∈ H
1
(Ω). Then, by
using a Green’s formula, we deduce
I
2
=
_
u
n+
1
2
I2
· m, ψ
_
Γ
R
−
_
Div u
n+
1
2
I2
, ψ
_
Ω
,
where, the involved functions are bounded as follows:
¸
¸
¸
¸
¸
¸u
n+
1
2
I2
· m
¸
¸
¸
¸
¸
¸
Γ
R
≤ C(T, v, A)∆t
2
∇φ
m
· m
C
2
(L
2
(Γ
R
))
,
(4.75) ¸
¸
¸
¸
¸
¸ Div u
n+
1
2
I2
¸
¸
¸
¸
¸
¸
Ω
≤ C(T, v, A)∆t
2
∇φ
m

C
2
(H
1
(Ω))
.
The estimate for I
3
follows by applying Lemmas 4.20 and 4.21 for ϕ = αφ
m

Γ
R and
ϕ = α(φ
n+1
m
+ φ
n
m
)
Γ
R, respectively:
I
3
=
_
ξ
n+
1
2
I3
, ψ
_
Γ
R
with
¸
¸
¸
¸
¸
¸ξ
n+
1
2
I3
¸
¸
¸
¸
¸
¸
Γ
R
≤ C(T, v)α∆t
2
φ
m

C
2
(L
2
(Γ
R
))
. (4.76)
Finally, partial results (4.74), (4.75) and (4.76) imply (4.72).
Lemma 4.26. Assume Hypothesis 1, and v ∈ C
2
(T
δ
), X
e
∈ C
4
(Ω × [0, T])
and ∆t < min{η, 1/(2L
∞,T
δ )}, being η the constant appearing in Lemma 4.1. Let
f
m
∈ C
2
(L
2
(Ω)), f ∈ C
1
(T
δ
), g
m
∈ C
2
(L
2
(Γ
R
)), g ∈ C
1
(T
δ
Γ
R
). Then, for each
n ∈ {0, . . . , N −1}, there exist ξ
n+
1
2
f
: Ω −→R and ξ
n+
1
2
g
: Γ
R
−→R, satisfying
__
F
n+
1
2
− F
n+
1
2
∆t
_
, ψ
_
=
_
ξ
n+
1
2
f
, ψ
_
Ω
+
_
ξ
n+
1
2
g
, ψ
_
Γ
R
∀ψ ∈ H
1
(Ω). (4.77)
Moreover, ξ
n+
1
2
f
∈ L
2
(Ω) and ξ
g
∈ L
2
(Γ
R
) and the following estimates hold:
¸
¸
¸
¸
¸
¸ξ
n+
1
2
f
¸
¸
¸
¸
¸
¸
Ω
≤ ∆t
2
C(T, v, T
δ
)
_
 det Ff
m

C
2
(L
2
(Ω))
+f
C
1
(T
δ
)
_
,
(4.78) ¸
¸
¸
¸
¸
¸ξ
n+
1
2
g
¸
¸
¸
¸
¸
¸
Γ
R
≤ ∆t
2
C(T, v, T
δ
Γ
R)
_
 ¯ mg
m

C
2
(L
2
(Γ
R
))
+g
C
1
(T
δ
Γ
R
)
_
.
Proof. The proof follows from Lemmas 4.20, 4.22, 4.23 and 4.24.
Hypothesis 9. Functions appearing in problem (2.7)(2.10) satisfy:
• ρ
m
∈ C
2
(L
∞
(Ω)), A ∈ W
2,∞
(O
δ
), A
m
∈ C
2
(W
1,∞
(Ω)),
• v ∈ C
3
(T
δ
),
• f
m
∈ C
2
(L
2
(Ω)), f ∈ C
1
(T
δ
), g
m
∈ C
2
(L
2
(Γ
R
)), g ∈ C
1
(T
δ
Γ
R
) and α > 0.
Hypothesis 10. Functions appearing in problem (2.7)(2.10) satisfy:
• ρ
m
∈ C
2
(L
∞
(Ω)), A ∈ W
2,∞
(O
δ
), A
m
∈ C
3
(W
1,∞
(Ω)),
• v ∈ C
3
(T
δ
),
• f
m
∈ C
2
(L
2
(Ω)), f ∈ C
1
(T
δ
), g
m
∈ C
3
(L
2
(Γ
R
)), g ∈ C
2
(T
δ
Γ
R
) and α > 0.
26 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
Lemmas in this section hold under Hypotheses 1, 3 and 4 and the previous ones.
Theorem 4.27. Assume Hypotheses 1, 3, 4, 5, 6, 7 and 9, and X
e
∈ C
5
(Ω ×
[0, T]). Let
φ
m
∈ C
3
(L
2
(Ω)), ∇φ
m
∈ C
2
(H
1
(Ω)), φ
m

Γ
R ∈ C
2
(L
2
(Γ
R
)),
be the solution of (4.68) and let
φ
m,∆t
be the solution of (4.28) subject to the initial
value φ
0
m,∆t
= φ
0
m
= φ
0
∈ H
1
(Ω). Then, there exist two positive constants J and D,
the latter being independent of the diﬀusion tensor, such that, if ∆t < D we have
√
γ
_
det F
RK
(φ
m
−φ
m,∆t
)
l
∞
(L
2
(Ω))
+
_
Λ
4
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
S [∇φ
m
−∇φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
α
8
¸
¸
¸
¸
¸
¸
¸
¸
_
S [ ¯ m
RK
]S [φ
m
−φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤ J ∆t
2
_
φ
m

C
3
(L
2
(Ω))
+∇φ
m

C
2
(H
1
(Ω))
+∇φ
m
· m
C
2
(L
2
(Γ
R
))
+φ
m

C
2
(L
2
(Γ
R
))
+ det Ff
m

C
2
(L
2
(Ω))
+f
C
1
(T
δ
)
+ ¯ mg
m

C
2
(L
2
(Γ
R
))
+g
C
1
(T
δ
Γ
R
)
_
.
(4.79)
Proof. We denote by e
m,∆t
the diﬀerence between the continuous and the discrete
solution, that is, e
m,∆t
=
_
φ
n
m
−φ
n
m,∆t
_
N
n=0
. Then, by using (4.28) and (4.68) we have
_
L
n+
1
2
∆t
[ e
m,∆t
], ψ
_
=
__
L
n+
1
2
∆t
−L
n+
1
2
_
[
´
φ
m
], ψ
_
+
_
F
n+
1
2
−F
n+
1
2
∆t
, ψ
_
, (4.80)
∀ψ ∈ H
1
Γ
D
(Ω) and 0 ≤ n ≤ N −1. Then, as a consequence of Lemmas 4.25 and 4.26,
we deduce
_
L
n+
1
2
∆t
[ e
m,∆t
], ψ
_
=
_
ξ
n+
1
2
f
−ξ
n+
1
2
LΩ
, ψ
_
Ω
+
_
ξ
n+
1
2
g
−ξ
n+
1
2
LΓ
, ψ
_
Γ
R
, (4.81)
∀ψ ∈ H
1
Γ
D
(Ω). Now the result follows by applying Theorem 4.11 to (4.81), noting
that e
0
m,∆t
= 0 and using the upper bounds for ξ
LΩ
, ξ
f
, ξ
LΓ
and ξ
g
given in Lemmas
4.25 and 4.26.
Remark 4.10. Notice that constant J appearing in the previous theorem is
bounded in the limit when the diﬀusion tensor vanishes. In particular, Theorem
4.27 is also valid when A ≡ 0.
Theorem 4.28. Let us assume Hypotheses 1, 3, 4, 5, 6, 7 and 10, and X
e
∈
C
5
(Ω ×[0, T]). Let φ
m
with
φ
m
∈ C
3
(L
2
(Ω)), ∇φ
m
∈ C
3
(H
1
(Ω)), φ
m

Γ
R ∈ C
3
(L
2
(Γ
R
)),
be the solution of (4.68) and
φ
m,∆t
be the solution of (4.28) subject to the initial value
φ
0
m,∆t
= φ
0
m
= φ
0
∈ H
1
(Ω). Then, there exist two positive constants J and D such
HIGHER ORDER PURE LAGRANGIAN METHOD 27
that, for ∆t < D we have
_
γ
4
¸
¸
¸
¸
¸
¸
¸
¸
_
S[ det F
RK
]R
∆t
[φ
m
−φ
m,∆t
]
¸
¸
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Ω))
+
_
Λ
2
¸
¸
¸
¸
¸
¸
¸
¸
¯
B
RK
(∇φ
m
−∇φ
m,∆t
)
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Ω))
+
_
α
4
¸
¸
¸
¸
¸
¸
¸
¸
_
¯ m
RK
(φ
m
−φ
m,∆t
)
¸
¸
¸
¸
¸
¸
¸
¸
l
∞
(L
2
(Γ
R
))
≤ J ∆t
2
_
φ
m

C
3
(L
2
(Ω))
+∇φ
m

C
2
(H
1
(Ω))
+∇φ
m
· m
C
3
(L
2
(Γ
R
))
+φ
m

C
3
(L
2
(Γ
R
))
+ det Ff
m

C
2
(L
2
(Ω))
+f
C
1
(T
δ
)
+ ¯ mg
m

C
3
(L
2
(Γ
R
))
+g
C
2
(T
δ
Γ
R
)
_
.
(4.82)
Proof. It is analogous to the one of the previous theorem, but using Theorem 4.15
instead of Theorem 4.11 and noting that
¸
¸
¸
¸
¸
¸
R
∆t
[ξ
LΓ
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
+
¸
¸
¸
¸
¸
¸
R
∆t
[ξ
g
]
¸
¸
¸
¸
¸
¸
l
2
(L
2
(Γ
R
))
≤
¯
C∆t
2
_
∇φ
m
· m
C
3
(L
2
(Γ
R
))
+φ
m

C
3
(L
2
(Γ
R
))
+ ¯ mg
m

C
3
(L
2
(Γ
R
))
+g
C
2
(T
δ
Γ
R
)
_
.
This estimate follows by using Taylor expansions and
¸
¸
_
X
n+1
e
(p) −X
n+1
RK
(p)
_
−(X
n
e
(p) −X
n
RK
(p))
¸
¸
≤
¯
C∆t
3
,
¸
¸
_
(F
n+1
)
−1
(p) −(F
n+1
RK
)
−1
(p)
_
−
_
(F
n
)
−1
(p) −(F
n
RK
)
−1
(p)
_¸
¸
≤
¯
C∆t
3
,
¸
¸
_
det F
n+1
(p) − det F
n+1
RK
(p)
_
−( det F
n
(p) − det F
n
RK
(p))
¸
¸
≤
¯
C∆t
3
.
Remark 4.11. In the particular case of diﬀusion tensor of the form A = ǫB with
ǫ > 0, constants J and D appearing in the previous theorem are bounded as ǫ → 0.
Remark 4.12. Notice that, from the obtained estimates and by using a change
of variable, we can deduce similar ones in Eulerian coordinates (see [6] for further
details).
5. Conclusions. We have performed the numerical analysis of a secondorder
pure Lagrangian method for convectiondiﬀusion equations with degenerate diﬀusion
tensor and nondivergencefree velocity ﬁelds. Moreover, we have considered general
DirichletRobin boundary conditions. The method has been introduced and analyzed
by using the formalism of continuum mechanics. Although our analysis considers any
velocity ﬁeld and use approximate characteristic curves, second order error estimates
have been obtained when smooth enough data and solutions are available. In the
second part of this paper ([7]), we analyze a fully discretized pure LagrangeGalerkin
scheme and present numerical examples showing the predicted behavior (see also [6]).
REFERENCES
[1] R.A. Adams. Sobolev Spaces. Academic Press, New York, 1975.
[2] J. Baranger, D. Esslaoui, and A. Machmoum. Error estimate for convection problem with
characteristics method. Numer. Algorithms, 21 (1999):49–56. Numerical methods for par
tial diﬀerential equations (Marrakech, 1998).
28 M. BEN
´
ITEZ AND A. BERM
´
UDEZ
[3] J. Baranger and A. Machmoum. Une norme “naturelle” pour la m´ethode des caract´eristiques
en ´el´ements ﬁnis discontinus: cas 1D. RAIRO Mod´el. Math. Anal. Num´er., 30:549–574,
1996.
[4] J. Baranger and A. Machmoum. A “natural” norm for the method of characteristics using
discontinuous ﬁnite elements: 2D and 3D case. Math. Model. Numer. Anal., 33:1223–1240,
1999.
[5] E. Barucci, S. Pilodoro, and V. Vespri. Some results on partial diﬀerential equations and Asian
options. Math. Models Methods Appl. Sci., 11:475–497, 2001.
[6] M. Ben´ıtez. M´etodos num´ericos para problemas de convecci´ on difusi´ on. Aplicaci´ on a la con
vecci´ on natural, Ph.D. Thesis, Universidade de Santiago de Compostela. 2009.
[7] M. Ben´ıtez and A. Berm´ udez. Numerical Analysis of a secondorder pure LagrangeGalerkin
method for convectiondiﬀusion problems. Part II: fully discretized scheme and numerical
results. Submitted.
[8] A. Berm´ udez and J. Durany. La m´ethode des charact´eristiques pour les probl`emes de
convectiondiﬀusion stationnaires. Math. Model. Numer. Anal., 21:7–26, 1987.
[9] A. Berm´ udez, M.R. Nogueiras, and C. V´ azquez. Numerical solution of (degenerated)
convectiondiﬀusionreaction problems with higher order characteristics/ﬁnite elements.
Part I: Time discretization. SIAM. J. Numer. Anal., 44:1829–1853, 2006.
[10] A. Berm´ udez, M.R. Nogueiras, and C. V´ azquez. Numerical solution of (degenerated)
convectiondiﬀusionreaction problems with higher order characteristics/ﬁnite elements.
Part II: Fully Discretized Scheme and Quadrature Formulas. SIAM. J. Numer. Anal.,
44:1854–1876, 2006.
[11] K. Boukir, Y. Maday, and B. M´etivet. A highorder characteristics method for the incompress
ible NavierStokes equations. Comput. Methods Appl. Mech. Engrg., 116:211–218, 1994.
ICOSAHOM’92 (Montpellier, 1992).
[12] K. Boukir, Y. Maday, B. M´etivet, and E. Razaﬁndrakoto. A highorder characteristics/ﬁnite
element method for the incompressible NavierStokes equations. Internat. J. Numer. Meth
ods Fluids, 25:1421–1454, 1997.
[13] K. Chrysaﬁnos and N. J. Walkington. Error estimates for discontinuous Galerkin approxima
tions of implicit parabolic equations. SIAM J. Numer. Anal., 43:2478–2499 (electronic),
2006.
[14] K. Chrysaﬁnos and N. J. Walkington. Error estimates for the discontinuous Galerkin methods
for parabolic equations. SIAM J. Numer. Anal., 44:349–366 (electronic), 2006.
[15] K. Chrysaﬁnos and N. J. Walkington. Lagrangian and moving mesh methods for the convection
diﬀusion equation. M2AN Math. Model. Numer. Anal., 42:25–55, 2008.
[16] J. Douglas, Jr., and T.F. Russell. Numerical methods for convectiondominated diﬀusion prob
lems based on combining the method of characteristics with ﬁnite element or ﬁnite diﬀer
ence procedures. SIAM J. Numer. Anal., 19:871–885, 1982.
[17] R.E. Ewing and H. Wang. A summary of numerical methods for timedependent advection
dominated partial diﬀerential equations. J. Comput. Appl. Math., 128:423–445, 2001.
[18] M.E. Gurtin. An Introduction to Continuum Mechanics. Mathematics in Science and Engi
neering, 158, Academic Press, San Diego, 1981.
[19] L. H¨ ormander. Hypoelliptic second order diﬀerential equations. Acta Math., 119:147–171, 1967.
[20] K. W. Morton. Numerical solution of convectiondiﬀusion problems. Appl. Math. Comput., 12,
Chapman Hall, London, 1996.
[21] O. Pironneau. On the TransportDiﬀusion Algorithm and Its Applications to the NavierStokes
Equations. Numer. Math., 38:309–332, 1982.
[22] O. Pironneau, J. Liou, and T. Tezduyar. CharacteristicGalerkin and Galerkin/leastsquares
spacetime formulations for the advectiondiﬀusion equations with timedependent do
mains. Comput. Methods Appl. Mech. Engrg., 100:117–141, 1992.
[23] A. Quarteroni and A. Valli. Numerical approximation of partial diﬀerential equations. Springer
Series in Computational Mathematics, SpringerVerlag, 23, Berlin, 1994.
[24] M. A. Ragusa. On weak solutions of ultraparabolic equations. Nonlinear Anal., 47:503–511,
2001.
[25] H. Rui and M. Tabata. A second order characteristic ﬁnite element scheme for convection
diﬀusion problems. Numer. Math., 92:161–177, 2002.
[26] P. Wilmot, J. Dewynne, and S. Howison. Option pricing. mathematical models and computa
tion. Oxford Financial Press, Oxford, 1993.
This action might not be possible to undo. Are you sure you want to continue?