See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/45877161

Stable Crank-Nicolson Discretisation for Incompressible MiscibleDisplacement

Problems of Low Regularity

Conference Paper · October 2009

DOI: 10.1007/978-3-642-11795-4_50 · Source: arXiv

CITATIONS READS

6 195

2 authors, including:

Max Jensen
University College London
45 PUBLICATIONS 516 CITATIONS

SEE PROFILE

All content following this page was uploaded by Max Jensen on 23 May 2014.

The user has requested enhancement of the downloaded file.

 Stable Crank-Nicolson Discretisation for
Incompressible Miscible Displacement Problems
of Low Regularity
arXiv:0910.1516v1 [math.NA] 8 Oct 2009

Max Jensen∗, Rüdiger Müller†

October 8, 2009

Abstract
In this article we study the numerical approximation of incompressible miscible displace-
ment problems with a linearised Crank-Nicolson time discretisation, combined with a mixed
finite element and discontinuous Galerkin method. At the heart of the analysis is the proof
of convergence under low regularity requirements. Numerical experiments demonstrate that
the proposed method exhibits second-order convergence for smooth and robustness for rough
problems.

1 Introduction and Initial Boundary Value Problem

Mathematical models which describe the miscible displacement of fluids are of particular economical
relevance in the recovery of oil in underground reservoirs by fluids which mix with oil. They also
play a significant role in CO2 stratification.
This publication extends the analysis of [1], which studies the discretisation of miscible displacement
under low regularity. Unlike to [1] which is based on a first-order implicit Euler time-step (leading
to a nonlinear system of equations in each time step), here we examine the discretisation in time
by a linearised second-order Crank-Nicolson scheme. Crucially, the new, more efficient method
inherits stability under low regularity. Like in [1], the concentration equation is approximated with a
discontinuous Galerkin method, while Darcy’s law and the incompressibility condition is formulated
as a mixed method. High-order time-stepping for miscible displacement under low regularity has
recently also been addressed in [4], however, with a continuous Galerkin discretisation in space and
discontinuous Galerkin in time. We refer for an outline of the general literature to [1, 2, 3, 4].
Definition 1 (Weak Formulation). A triple (u, p, c) in


L∞ (0, T ; HN (div; Ω)) × L∞ (0, T ; L20 (Ω)) × L2 (0, T ; H 1 (Ω)) ∩ H 1 (0, T ; H 2 (Ω)∗ )

is called weak solution of the incompressible miscible flow problem if

∗
Mathematical Sciences, University of Durham, England, [email protected]
†
WIAS, Berlin, Germany, [email protected]

1
 (W1) for t ∈ (0, T ), v ∈ HN (div; Ω) and q ∈ L20 (Ω)




µ(c) K−1 u, v − p, div v = ρ(c) g, v



q, div u = q I − q P , q .

(W2) for all w ∈ D(0, T ; H 2 (Ω))

Z T





− φ c, ∂t w + D(u)∇c, ∇w + u · ∇c, w + q I c, w − ĉq I , w dt = 0.
0

(W3) c(0, ·) = c0 in H 2 (Ω)∗ .

For the data qualification we refer to condition (A1)–(A8) in [1] and for the physical interpretation
of the system to [1, 2, 3]. We point out that D growths proportionally with u:

d◦ (1 + |u|)|ξ|2 ≤ ξ T D(u, x) ξ ≤ d◦ (1 + |u|)|ξ|2 , u, ξ ∈ Rd , x ∈ Ω.

Thus D is in general unbounded on Lipschitz domains Ω and in the presence of discontinuous

coefficients, which are permitted in this paper.

2 The Finite Element Method

We compactly recall the definition of the finite element spaces from [1]. Let 0 = t0 < t1 < . . . <

tM = T be a partition of the time interval [0, T ]. Let kj := tj − tj−1 and dt aj := kj−1 aj − aj−1 .
We consider meshes T of Ω with elements K and set hK := diam(K). We denote by Ss (T) the
space of elementwise polynomial functions of total or partial degree s. For wh ∈ Ss (T) the function
∇h wh is defined through (∇h wh )|K = ∇(wh |K ). The sets of interior and boundary faces are
EΩ (T) and E∂Ω (T). We set E(T) = EΩ (T) ∪ E∂Ω (T) and assign to each E ∈ E(T) its diameter
hE . We denote jump and the average operators by [·] and {·}. The concentration c is discretised
at time j on the mesh Tcj or simply by T j . The approximation space for the variable c at time
step j is denoted by Sjc . Often we abbreviate Ej := E(Tcj ), EjΩ := EΩ (Tcj ), Ej∂Ω := E∂Ω (Tcj ). We
denote the Raviart-Thomas space of order ℓ by RTℓ (Tuj ). The approximation spaces of u and p
are Sju := RTℓ (Tuj ) ∩ HN (div; Ω) and Sjp := Sℓ (Tuj ) ∩ L20 (Ω). We frequently use the global mesh
size and time step hj := maxK∈Tcj ∪Tuj hK , e h := max0≤j≤M hj , e k := max0≤j≤M kj as well as to
Q j QM j QM j
Su = M j=1 Su , Sp = j=1 Sp , Sc = j=0 Sc . In addition we impose conditions (M1)–(M5) of [1]
which are on shape-regularity, boundedness of the polynomial degree, control kvh kL4 . kvh kH 1 and
the structure of hanging nodes.
To deal with discontinuous coefficients and the time derivative, we substitute D by

Dh : L2 (Ω)d → Ss (Tc , Rd×d ), v 7→ ΠT ◦ D(v, ·)

where the ΠT are projections such that kΠT DkK . kDkK . Given quantities aj , aj−1 and aj−2 at
times tj , tj−1 , tj−2 , we denote aj = 12 aj + 12 aj−1 and ă = 23 aj−1 − 12 aj−2 .

2
The diffusion term of the concentration equation is discretised by the symmetric interior penalty
discontinuous Galerkin method: Given ch , wh ∈ Sjc , uh ∈ Sju , we set



Bd (ch , wh ; uh ) := Djh (uh )∇h ch , ∇h wh − [ch ], {Djh (uh ) ∇h wh } Ej
Ω
j
2


− [wh ], {Dh (uh ) ∇h ch } Ej + σ [ch ], [wh ] Ej
Ω Ω

where σ is chosen sufficiently large to ensure coercivity of Bd , cf. [1]. The convection, injection and
production terms are represented by




Bcq (ch , wh ;uh ) := /2 uh ∇h ch , wh − uh ch , ∇h wh + (q I + q P )ch , wh
1
(1)
X


+ c+ +
h , (uh · nK )+ [wh ]K ∂K\∂Ω − (uh · nK )− [ch ]K , wh ∂K\∂Ω ,
K∈Tj

where (uh · n)+ := max{uh · n, 0} and (uh · n)− := min{uh · n, 0}. We set B = Bd + Bcq .
Algorithm (AdG ). Choose cjh ∈ Sjc for j = 0, 1. Given cjh , find (ujh , pjh ) ∈ Sju × Sjp such that




µ(cjh ) K−1 uh ,
vh − ph , div vh = ρ(cjh ) g, vh ,

(2)
qh , div uh = (q I − q P )j , qh .

For 2 ≤ j ≤ M find cjh ∈ Sjc such that, for all wh ∈ Sjc ,

j j

φ dt cjh , wh + B(ch j , wh ; ŭjh ) = ĉ q I , wh (3)

and solve (2) to obtain (ujh , pjh ) ∈ Sju × Sjp .

The algorithm only requires the solution of a linear system in each time step. The iterate c1h can
be computed with an implicit Euler method and fine time steps. The use of extrapolated values
such as ŭjh is classical, e.g. see [5, p. 218].

3 Unconditional Well-posedness, Boundedness and Convergence

Given chj−1 and chj−2 , there exists a solution cjh ∈ Sju of (3) because the bilinear form B is positive
t−t t −t
ch (t, ·) := kj−1
definite. For t ∈ [tj−1 , tj ], let e j
cjh + jkj chj−1 . Then ∂t e
ch (t, ·) = dt cjh (·). We interpret
elements of Su , Sp and Sc as time-dependent functions with stepwise constant values. Let




|ch |2ŭh := Dh (ŭh )∇h ch , ∇h ch + σ 2 [ch ], [ch ] Ej + |ŭh · nEj | [ch ], [ch ] Ej .
Ω Ω

Theorem 1. Let ρ◦ = kρk∞ . There exists a constant C > 0 such that

kŭjh k + kdiv ŭjh k + kp̆jh k . kρ◦ gk + kq̆ I − q̆ P k (4)

holds for all j = 2, 3 . . . , M . Equally we have

Z tj Z tj
1/2 j 2 2 1/2 1 2 i
1/2 i 2
kφ ch k + |ch |ŭj dt ≤ kφ ch k + k qI ĉ k dt (5)
t1 h t1

for all j = 2, 3 . . . , M .

3
Proof. The stability of uj−1 , uj−2 , pj−1 , pj−2 follows from a classical inf-sup argument. This implies
stability of ŭj and p̆j . We choose wh = ch i in (3) to verify that

i i

dt kφ1/2 cih k2 + |ch i |2ŭi + k(q I + q P )1/2 ch i k2 ≤ 2 φ dt cih , ch i + 2B(ch i , ch i ; ŭih ) = 2 ĉ q I , ch i .
h

The Cauchy-Schwarz inequality, multiplication by ki and summation over i give

j
X j
X i
1/2 i 2
kφ1/2 cjh k2 + ki |ch i |2ŭi ≤ kφ1/2 c1h k2 + ki k q I ĉ k
h
i=2 i=2

for all j = 2, 3, . . . , M .

For simplicity the next theorem is stated assuming meshes are not adapted in time. For the
extension to changing meshes consult [1]. However, observe that that the discretisation with the
implicit Euler method gives additional stability in ki kφ1/2 dt cih k2 , which allows to change meshes
more rapidly.

ch belongs to L2 (t1 , T ; H 2 (Ω)∗ ) and

Theorem 2. The time derivative ∂t e

k∂t e
ch kL2 (t1 ,T ;H 2(Ω)∗ ) = kdt ch kL2 (t1 ,T ;H 2 (Ω)∗ ) . 1,

independently of the mesh size and time step.

Proof. Let wh ∈ Sjc . We recall from [1]

1
Bd (cjh , wh ; ŭjh ) . (1 + kŭjh k /2 ) |cjh |Tj (k∇h wh kL4 (Ω) + kwh kL4 (Ω) + kσ[wh ]kEj ),
Ω
1
Bcq (cjh , wh ; ŭjh ) . (1 + kŭjh k /2 ) |cjh |Tj (k∇h wh k + kwh kL4 (Ω) + kσ[wh ]kEj ),
Ω
1
kσ[wh ]k2Ej . (1 + kŭjh k) e
h /2 kwk2H 2 (Ω) .
Ω

With L2 -orthogonality and

Z T Z T
j
j j

φ dt ch , w dt = ĉ q I , wh − B(ch j , wh ; ŭjh ) dt
t1 t1
Z T 1/
. (1 + kŭjh k)(1 + kŭjh kH(div;Ω)
2
) |cjh |uj kwkH 2 (Ω) dt
h
0
. kwkL2 (0,T ;H 2 (Ω))

one completes the proof.

Theorem 3. Let (ui , pi , ci )i∈N be a sequence of numerical solutions with (e hi , e

ki ) → 0 as i → ∞.
2 1 1 2 ∗
Then there exists c ∈ L (0, T ; H (Ω)) ∩ H (0, T ; H (Ω) ) such that, after passing to a subsequence,
ci → c in L2 (ΩT ), ∂t e
ci ⇀ ∂t c in L2 (0, T ; H 2 (Ω)∗ ) and ∇ci ⇀ ∇c in L2 (0, T ; H −1 (Ω)). If c0i , c1i → c0
2 ∗
in H (Ω) then c satisfies (W3).

4
The proof is, up to the treatment of the initial conditions, exactly as in [1]. It is based on the
Aubin-Lions theorem and the embedding

Ss (Ti ) ֒→ [BV(Ω) ∩ L4 (Ω), L4 (Ω)]1/2 ֒→ L2 (Ω),

where [·, ·]θ denotes the complex method of interpolation.

Theorem 4. Let (ui , pi , ci )i∈N be numerical solutions with (e

hi , e
ki ) → 0 and ci → c in L2 (ΩT ) as
i → ∞. There exists u ∈ L (0, T ; HN (div; Ω)) and p ∈ L (0, T ; L20 (Ω)) such that, after passing to
∞ ∞

a subsequence, ui → u in HN (div; Ω) and pi → p in L20 (Ω) as (e hi , e

ki ) → 0. Furthermore, (u, p, c)
satisfies (W1).

Proof. Use Strang’s lemma, for details see [1].

We interpret ŭi as piecewise constant function in time, attaining in (tj−1 , tj ] the value 32 u(tj−1 ) −
1 j−2 ).
2 u(t

Theorem 5. Let (ui , pi , ci )i∈N be a sequence of numerical solutions with (e hi , e

ki ) → 0 as i → ∞
∞ 2 1 1 2 ∗
and let u ∈ L (0, T ; HN (div ; Ω)) and c ∈ L (0, T ; H (Ω)) ∩ H (0, T ; H (Ω) ) be a limit of (ui , ci )i
in the sense of Theorems 3 and 4. Then (u, c) satisfies (W2).

Proof. Let v ∈ D(0, T ; C ∞ (Ω)) and vi (t) ∈ Sjc an approximation to v(t) in (tj−1 , tj ]. Using the
strong convergence of (∇h vi )i in L∞ (ΩT )d and the weak convergence of the lifted gradient of ci in
L2 (ΩT )d , we find
Z T Z T




∇c, D(u)∇v dt = lim ∇h ci , Dh (ŭi )∇h vi − [ci ], {Dh (ŭi )∇h vi } E dt.
t1 i→∞ t1 Ω

As in [1] it follows that Bd (ci , vi ; ŭi ) coincides in the limit with ∇c, D(u)∇v . One can also
conclude by adapting [1] that
Z T Z T



u · ∇c, v + q I c, v dt = lim Bcq (ci , vi ; ŭi ) dt.
t1 i→∞ t1

One arrives at
Z T





− φ c, ∂t v + D(u)∇c, ∇v + u · ∇c, v + q I c, v − ĉq I , v dt
t1
ZT
j j

= lim φ dt cjh , wh + B(ch j , wh ; ŭjh ) − ĉ q I , wh dt = 0.
i→∞ t1

Hence (W2) is satisfied for v ∈ D(0, T ; C ∞ (Ω)). The extension to D(0, T ; H 2 (Ω)) follows from
boundedness and density of smooth functions.

5
 qI

IK = k2 Id IK = k1 Id

IK = k1 Id

qP

Figure 1: Example 1: Left: computational domain; right: absolute value |uh | of the Darcy velocity
at t = 1.0 before any interaction between the concentration front and the corner singularity.

1 1
1 1
0.8 0.8
0.8 0.8
0.6 0.6

0.4 0.6 0.4 0.6

0.2 0.4 0.2 0.4

0 0
0.2 0.2
0 0
0.2 0.2
0.4 0.4
0.6 0 0.6 0
0.8 0.8
1 1

Figure 2: Snapshots of ch at t = 1.5 and 2.0, computed with the Crank-Nicolson scheme.

6
4 Numerical Experiments

The numerical experiments are carried out in two space dimensions with the lowest-order method
on a mesh which consists of shape-regular triangles without hanging nodes and which is not changed
over time. The diffusion–dispersion tensor takes the form

D(u, x) = φ(x) (dm Id + |u| dℓ E(u) + |u| dt (Id − E(u))) . (6)

1 1 1

0.5 0.5 0.5

0 0 0
1 1 1

0.5 0.5 0.5

0 0 0

−0.5 1 −0.5 1 −0.5 1

0 0 0
−1 −1 −1 −1 −1 −1

Figure 3: Example 2: Snapshots of the concentration cref at t = 0.25, 1.0 and 3.0.

Numerical Example 1 (Singular Velocities). To examine the effect of a singular velocity field
caused by a discontinuous permeability distribution and a re-entrant corner we employ the L-
shaped domain Ω and K with k1 = 0.1 and k2 = 10−6 as depicted in Figure 1. The injection
and production wells are located at (1, 1) and (0, 0), respectively. The porous medium is almost
impenetrable in the upper left quarter, forcing a high fluid velocity at the reentrant corner where
the nearly impenetrable barrier is thinnest. This leads to a singularity |u| ∼ r −α , where r is the
distance to the reentrant corner and α ≈ 1, cf. [1]. Figure 2 shows the concentration when the
front passes the corner and at a later time. The solution ch contains steep fronts but shows only
the localised oscillations that are characteristic for dG methods.

Numerical Example 2 (Convergence rates). Convergence rates are determined by comparing

the numerical solution ch to a reference solution cref that is computed with high accuracy on a one
dimensional grid. More precisely, we set φ = 1, ĉ = 1, K = 1 and g = 0 and choose Ω to be the
ball B(0, 1) ⊂ R2 . Using polar coordinates (r, ϕ), we choose q I = 4 (1 − r)6 and q P = 47 r 6 . Then
the Darcy velocity only changes in the radial direction and is determined by an ODE, which
has
r 6 5 4 3 2
the nonnegative exact solution u(r) = 7 3 r − 24 r + 70 r − 112 r + 105 r − 56 r + 14 . Conse-
quently, the concentration equation reduces to a linear parabolic equation in one space dimension.
Figure 3 shows snapshots of the solution cref with dm = 1.0 × 10−5 , dℓ = 4.0 × 10−4 and Figure
4 shows that L2 error of implicit Euler method is of order O(h2 + k) whereas the Crank-Nicolson
reaches the order O(h2 + k2 ).

7
 2
Euler implicit, k ∼ h
Crank Nicolson, k ∼ h
2
−2 ∼h
10

|| ch −cref ||L2(Ω)
−3
10

−4
10

−2 −1
10 10
h

Figure 4: Error kch − cref kL2 (Ω) of the implicit Euler method the Crank-Nicolson method at time
t = 1.

References

[1] S. Bartels, M. Jensen, R. Müller, Discontinuous Galerkin finite element convergence

for incompressible miscible displacement problems of low regularity, to appear in SIAM Jour-
nal on Numerical Analysis, submitted December 2007 (also preprint HU-Berlin 2008 No. 2,
www.mathematik.hu-berlin.de/publ/pre/2008/P-08-02.pdf).

[2] Z. Chen, Reservoir simulation (Mathematical techniques in oil recovery), SIAM , 2007.

[3] X. Feng, Recent developments on modeling and analysis of flow of miscible fluids in porous
media, Fluid flow and transport in porous media, Contemp. Math. 295:229–24, 2002.

[4] B. Rivière, N. Walkington, Convergence of a Discontinuous Galerkin Method for

the Miscible Displacement Equations Under Minimal Regularity, preprint May 2009,
(http://www.math.cmu.edu/∼noelw/Noelw/Papers/RiWa09.pdf).

[5] V. Thomée, Galerkin finite element methods for parabolic problems, Springer Series in Com-
putational Mathematics 25, 1997.

View publication stats