Journal of Computational Mathematics, Vol.26, No.6, 2008, 816–824.

A MIXED FINITE ELEMENT METHOD ON A STAGGERED

MESH FOR NAVIER-STOKES EQUATIONS*

Houde Han and Ming Yan

Department of Mathematics, University of Science and Technology of China, Hefei 230026, China
Email: hhan@math.tsinghua.edu.cn, yanm@math.ucla.edu

Abstract
In this paper, we introduce a mixed finite element method on a staggered mesh for the
numerical solution of the steady state Navier-Stokes equations in which the two components
of the velocity and the pressure are defined on three different meshes. This method is
a conforming quadrilateral Q1 × Q1 − P0 element approximation for the Navier-Stokes
equations. First-order error estimates are obtained for both the velocity and the pressure.
Numerical examples are presented to illustrate the effectiveness of the proposed method.
Mathematics subject classification: 35Q30, 74G15, 74S05.
Key words: Mixed finite element method, Staggered mesh, Navier-Stokes equations, Error
estimate.

1. Introduction
It is well known that the simplest conforming low-order elements like the P1 − P0 (linear
velocity vector, constant pressure) triangular element and Q1 − P0 (bilinear velocity vector,
constant pressure) quadrilateral element are not stable when applied to the Navier-Stokes (NS)
equations [6]. Therefore, some special treatments are needed in order to keep the schemes stable.
During the last two decades, there has been a rapid development in practical stabilization
technique for the P1 − P0 element and the Q1 − P0 element for solving the NS equations
[1, 7, 8, 9, 11]. In [3], an economical finite element scheme is proposed to construct three finite-
dimensional subspaces for the two velocity components and the pressure. In [2], a mixed finite
element scheme for the Stokes equations is investigated. In this paper, we extend the idea in [3]
to construct a mixed finite element scheme for the NS equations, which is more efficient than
the scheme given in [3] as the degree of freedom is reduced. The optimal error estimate of this
scheme is obtained.
The outline of the paper is as follows. In the next section, we give a formulation of the
mixed finite element method for the Navier-Stokes equations. In Section 3, the error estimates
will be provided. In Section 4, two numerical examples will be considered. Finally, we end the
paper with a short concluding section.

2. A Mixed Finite Element Formulation for the NS Equations

We consider the following boundary value problem of the Navier-Stokes equations:

 −ν∆u + (u · ∇)u + ∇p = f , in Ω,
div u = 0, in Ω, (2.1)
u = 0, on ∂Ω,


* Received January 18, 2007 / Revised version received December 4, 2007 / Accepted January 3, 2008 /
A Mixed Finite Element Method on a Staggered Mesh for Navier-Stokes Equations 817

q aa q aa q aa q aa q q q q q q a
a
a
a
a
a
a
a
qaqaqaqaq q q q q q a a a a
qaqaqaqaq q q q q q a a a a
qaqaqaqaq q q q q q
(a) (b) (c)
a a a a
Fig. 2.1. Quadrangulations: (a) Jh , (b) Jh1 , (c) Jh2 .

where Ω ⊂ R2 is a rectangular domain, ν is the viscosity, u = (u1 , u2 )T represents the velocity

vector, p is the pressure, and f = (f1 , f2 )T is the given body force. Let H n (Ω) and H01 (Ω) denote
the standard Sobolev spaces with the norm k · kn,Ω and k · k1,Ω respectively. Furthermore, let
 Z 
V≡ H01 (Ω) × H01 (Ω), 2
M ≡ q : q ∈ L (Ω) and qdx = 0 .
Ω

Then the boundary value problem (2.1) is reduced to the following equivalent variational prob-
lem [3]:

 Find u ∈ V and p ∈ M, such that
a(u, v) + a1 (u; u, v) + b(v, p) = (f , v) ∀v ∈ V, (2.2)
b(u, q) = 0 ∀q ∈ M,


where
Z
a(u, v) = ν ∇u · ∇vdx,
Ω
2 Z
1 X  ∂u
i ∂vi 
a1 (w; u, v) = wj vi − ui dx,
2 i,j=1 Ω ∂xj ∂xj
Z Z
b(v, q) = − qdiv vdx, (f , v) = f · vdx.
Ω Ω

For simplicity we assume that the domain Ω is a unit square, but the finite element method
discussed below can be easily generalized to include the case that the domain Ω is rectangular.
Let N be a given integer and h = 1/N . We shall construct the finite-dimensional subspaces of
V and M by introducing three different quadrangulations Jh , Jh1 , Jh2 of Ω. First we divide Ω
into equal squares
n o
Ti,j = (x1 , x2 ) : (x1 )i−1 ≤ x1 ≤ (x1 )i , (x2 )j−1 ≤ x2 ≤ (x2 )j , i, j = 1, · · · , N,

where (x1 )i = ih and (x2 )j = jh. The corresponding quadrangulation is denoted by Jh . Then
for all Ti,j ∈ Jh we connect all the midpoints of the vertical sides of Ti,j by straight line
segments if the midpoints have a distance h, and extend the resulting mesh to the boundary
Γ. Then Ω is divided into squares and rectangles, and the corresponding quadrangulation is
denoted by Jh1 . Similarly, for all Ti,j ∈ Jh we connect all the midpoints of the horizontal sides
of Ti,j by straight line segments if the midpoints have a distance h, and extend the resulting
mesh to the boundary Γ. Then we obtained the third quadrangulation of Ω, which is denoted
by Jh2 (see Fig. 2.1).
818 H.D. HAN AND M. YAN

Corresponding to the quadrangulation Jh , let

 Z 
Mh := qh : qh |T = constant ∀T ∈ Jh and qh dx = 0 ,
Ω

Mh is a subspace of M . Furthermore, using the quadrangulation Jh1 and Jh2 , we construct two
subspaces of H01 (Ω). Set
n o
Sh1 = vh ∈ C (0) (Ω) : vh |T 1 ∈ Q1 (T 1 ) ∀T 1 ∈ Jh1 , and vh |Γ = 0 ,
n o
Sh2 = vh ∈ C (0) (Ω) : vh |T 2 ∈ Q1 (T 2 ) ∀T 2 ∈ Jh2 , and vh |Γ = 0 ,

where Q1 denotes the space of all polynomials of degree not exceeding one with respect to each
of the two variables x1 and x2 . Let

Vh = Sh1 × Sh2 ;

obviously, Vh is a subspace of V. Using the subspaces Vh and Mh instead of V and M

in the variational problem (2.2), we obtain the discrete problem which is the finite element
approximation of the nonlinear variational problem (2.2):

 Find uh ∈ Vh and ph ∈ Mh , such that
a(uh , vh ) + a1 (uh ; uh , vh ) + b(vh , ph ) = (f , vh ) ∀vh ∈ Vh , (2.3)
b(uh , qh ) = 0 ∀qh ∈ Mh .


3. Well-Posedness of Problem (2.3) and Optimal Error Estimate

In order to attain the error estimate of the finite element approximation to the nonlinear
variational problem (2.2), we first introduce some notations. Let
|a1 (w; u, v)|
N= sup ,
u,v,w∈V |u|1,Ω |v|1,Ω |w|1,Ω
|a1 (wh ; uh , vh )|
Nh = sup ,
uh ,vh ,wh ∈Vh |uh |1,Ω |vh |1,Ω |wh |1,Ω
(f , v) (f , vh )
kf k∗ = sup , kf k∗h = sup ,
v∈V |v| 1,Ω vh ∈Vh |vh |1,Ω
G(u; v, w) = a1 (u; u, w) − a1 (v; v, w),

where | · |1,Ω is the semi-norm of H 1 (Ω). Obviously, the following inequalities hold

Nh ≤ N, kf k∗h ≤ kf k∗ . (3.1)

For the quadrangulation Jh , we divided the edges of all squares into two sets. The first set
contains all vertical edges and is denoted by LV . The second set contains all horizontal edges
and is denoted by LH . We define the operator Π : V → Vh by Πu = (Π1h u1 , Π2h u2 )T ∈ Sh1 × Sh2
satisfying:
Z Z Z Z
Π1h u1 ds = u1 ds ∀l ∈ LV , Π2h u2 ds = u2 ds ∀l ∈ LH .
l l l l

For the problem (2.3), we have the following result. Its proof is omitted as it is basically
the same as in continuous problem (see [4] Chapter IV).
A Mixed Finite Element Method on a Staggered Mesh for Navier-Stokes Equations 819

Theorem 3.1. Problem (2.3) has at least one solution (uh , ph ) ∈ Vh × Mh . Moreover, the
solution is unique if the following condition holds

Nh kf k∗h /ν 2 < 1.

Theorem 3.2. Under the assumption that

N kf k∗ /ν 2 < 1 − δ, (3.2)

where 0 < δ < 1 is a constant, the finite element solution of problem (2.3) (uh , ph ) ∈ Vh × Mh
satisfies:
  
|G(u; Πu, wh )|

 |u − uh |1,Ω ≤ C |u − Πu|1,Ω + inf kp − qh k0,Ω + sup ,



 qh ∈Mh wh ∈Vh |wh |1,Ω
 
kp − ph k0,Ω ≤ C |u − Πu|1,Ω + inf kp − qh k0,Ω (3.3)

 qh ∈Mh 
|G(u; Πu, wh )| |G(u; uh , wh )|


+ sup + sup .


|wh |1,Ω |wh |1,Ω

wh ∈Vh wh ∈Vh

Before the proof we recall some results from [2], which only give the error estimate for the
Stokes equations, using the same finite element formulations as in this paper.

Lemma 3.1. (i) For any u ∈ V, we have

Z
qh div (u − Πu)dx = 0 ∀qh ∈ Mh . (3.4)
Ω

(ii) There exist two constants C1 and C2 independent of h, such that

inf ku − vh k1,Ω ≤ ku − Πuk1,Ω ≤ C1 h|u|2,Ω ,

vh ∈Vh
(3.5)
inf kp − qh k0,Ω ≤ C2 h|p|1,Ω .
qh ∈Mh

(iii) There is a constant B > 0 such that

|b(v, q)| ≤ Bkvk1,Ω kqk0,Ω . (3.6)

(iv) (Ladyzhenskaya-Babus̆ka-Brezzi condition) There exists a constant β0 > 0 independent

of h such that
b(vh , qh )
sup ≥ β0 kqh k0,Ω ∀qh ∈ Mh . (3.7)
vh ∈Vh kvh k1,Ω
(v) There exists a constant C3 independent of h such that

kΠuk1,Ω ≤ C3 kuk1,Ω ∀u ∈ V.

Proof of Theorem 3.2. By inequality (3.2), we know that

Nh kf k∗h /ν 2 ≤ N kf k∗ /ν 2 ≤ 1 − δ. (3.8)

It follows from Theorem 3.1 that problem (2.3) has a unique solution (uh , ph ) ∈ Vh × Mh .
Setting vh = uh in the first equation in (2.3), and observing that a1 (uh ; uh , uh ) = 0 and
b(uh , ph ) = 0, we obtain

(f , uh ) = a(uh , uh ) + a1 (uh ; uh , uh ) + b(uh , ph ) = a(uh , uh ).

820 H.D. HAN AND M. YAN

Consequently,

ν|uh |21,Ω = a(uh , uh ) = (f , uh ) ≤ |f |∗h |uh |1,Ω ,

which yields

kf k∗h
|uh |1,Ω ≤ . (3.9)
ν
Let zh = uh − Πu and

Φ = a(uh , zh ) + a1 (uh ; uh , zh ) − a(Πu, zh ) − a1 (Πu; Πu, zh ).

Then we have

Φ = a(uh , zh ) + a1 (zh ; uh , zh ) + a1 (Πu; uh , zh ) − a(Πu, zh ) − a1 (Πu; Πu, zh )

= a(zh , zh ) + a1 (zh ; uh , zh ) + a1 (Πu; zh , zh )
= a(zh , zh ) + a1 (zh ; uh , zh ) ≥ ν|zh |21,Ω − Nh |uh |1,Ω |zh |21,Ω
 Nh kf k∗h   N kf k∗ 
≥ ν− |zh |21,Ω ≥ ν − |zh |21,Ω ≥ νδ|zh |21,Ω ,
ν ν
which gives that
1
|zh |21,Ω ≤ |Φ|. (3.10)
νδ
On the other hand, we know that

Φ = (f , zh ) − b(zh , ph ) − a(Πu, zh ) − a1 (Πu; Πu, zh )

= a(u, zh ) + a1 (u; u, zh ) + b(zh , p) − b(zh , ph ) − a(Πu, zh ) − a1 (Πu; Πu, zh )
= a(u − Πu, zh ) + b(zh , p − ph ) + G(u; Πu, zh )
= a(u − Πu, zh ) + b(zh , p − qh ) + b(uh − Πu, qh − ph ) + G(u; Πu, zh )
= a(u − Πu, zh ) + b(zh , p − qh ) + b(uh − u, qh − ph ) + G(u; Πu, zh )
= a(u − Πu, zh ) + b(zh , p − qh ) + G(u; Πu, zh ) ∀qh ∈ Mh .

Furthermore, we obtain, for any qh ∈ Mh ,

 
|G(u; Πu, wh )|
|Φ| ≤ ν|u − Πu|1,Ω + Bkp − qh k0,Ω + sup |zh |1,Ω . (3.11)
wh ∈Vh |wh |1,Ω

Combining (3.10) and (3.11), we have

 
1 |G(u; Πu, wh )|
|zh |1,Ω ≤ ν|u − Πu|1,Ω + B inf kp − qh k0,Ω + sup .
νδ qh ∈Mh wh ∈Vh |wh |1,Ω

Using the triangle inequality yields

|u − uh |1,Ω ≤ |u − Πu|1,Ω + |Πu − uh |1,Ω

 
|G(u; Πu, wh )|
≤ C |u − Πu|1,Ω + inf kp − qh k0,Ω + sup , (3.12)
qh ∈Mh wh ∈Vh |wh |1,Ω
A Mixed Finite Element Method on a Staggered Mesh for Navier-Stokes Equations 821

where C is a constant dependent only on B, δ and ν. To estimate error kp − ph k0,Ω , consider

b(vh , ph − qh ) = b(vh , ph − p) + b(vh , p − qh )

= a(u − uh , vh ) + G(u; uh , vh ) + b(vh , p − qh ),

for any vh ∈ Vh and qh ∈ Mh . By (iv) of Lemma 3.1, we obtain, for any qh ∈ Mh ,

1 |b(vh , ph − qh )|
kph − qh k0,Ω ≤ sup
β0 vh ∈Vh |vh |1,Ω
 
1 |G(u; uh , vh )|
≤ ν|u − uh |1,Ω + Bkp − qh k0,Ω + sup . (3.13)
β0 vh ∈Vh |vh |1,Ω
Combining inequalities (3.12), (3.13) and using the triangle inequality lead to the conclusion
(3.3). 
Now we estimate G(u; Πu, vh ) and G(u; uh , vh ). We have

Lemma 3.2. Suppose (u, p) is the solution of problem (2.2) and u ∈ V ∩ (H 2 (Ω))2 , p ∈ M ∩
H 1 (Ω). Then there exists a constant C independent of h and (u, p), such that
|G(u; Πu, vh )|
sup ≤ Chkuk1,Ω |u|2,Ω , (3.14)
vh ∈Vh |vh |1,Ω
|G(u; uh , vh )| 
sup ≤ C|u − uh |1,Ω kf k∗ + |u|1,Ω . (3.15)
vh ∈Vh |vh |1,Ω
Proof. Observe that

|G(u; Πu, vh )| = |a1 (u; u, vh ) − a1 (Πu; Πu, vh )|

= |a1 (u − Πu; u, vh ) + a1 (Πu; u, vh ) − a1 (Πu; Πu, vh )|
= |a1 (u − Πu; u, vh ) + a1 (Πu; u − Πu, vh )|

≤ N |u − Πu|1,Ω |u|1,Ω + |Πu|1,Ω |vh |1,Ω
≤ Ch|u|2,Ω kuk1,Ω |vh |1,Ω ,

where the last inequality used (ii) and (v) of Lemma 3.1. This ends the proof of (3.14). Similarly,
we have

|G(u; uh , vh )| = |a1 (u; u, vh ) − a1 (uh ; uh , vh )|

= |a1 (u − uh ; u, vh ) + a1 (uh ; u, vh ) − a1 (uh ; uh , vh )|
= |a1 (u − uh ; u, vh ) + a1 (uh ; u − uh , vh )|

≤ N |u − uh |1,Ω |u|1,Ω + |uh |1,Ω |vh |1,Ω

≤ C|u − uh |1,Ω |u|1,Ω + kf k∗ |vh |1,Ω ,

where we used (3.9) in the last step. The inequality (3.15) is then proved. 
Finally, an application of Theorem 3.2 and Lemma 3.2 yields the following error estimates.

Theorem 3.3. Suppose that the condition (3.2) holds and that the solution of problem (2.2)
satisfies u ∈ V ∩ (H 2 (Ω))2 , p ∈ M ∩ H 1 (Ω). Then the following error estimates hold:

|u − uh |1,Ω ≤ Ch |u|2,Ω + |p|1,Ω + kuk1,Ω |u|2,Ω , (3.16)
 ∗


kp − ph k0,Ω ≤ Ch |u|2,Ω + |p|1,Ω + kuk1,Ω |u|2,Ω 1 + |u|1,Ω + kf k . (3.17)
822 H.D. HAN AND M. YAN

Proof. The inequality (3.16) follows from (ii) of Lemma 3.1, Theorem 3.2 and Lemma 3.2.
For the inequality (3.17), we have

kp − ph k0,Ω
 
|G(u; Πu, wh )| |G(u; uh , wh )|
≤ C ku − Πuk1,Ω + inf kp − qh k0,Ω + sup sup
qh ∈Mh wh ∈Vh kwh k1,Ω wh ∈Vh kwh k1,Ω
n

o
≤ C h |u|2,Ω + |p|1,Ω + kuk1,Ω |u|2,Ω |u − uh |1,Ω kf k∗ + |u|1,Ω
 

≤ Ch |u|2,Ω + |p|1,Ω + kuk1,Ω |u|2,Ω 1 + |u|1,Ω + kf k∗ .

This completes the proof of the theorem. 

4. Numerical Examples
In this section, we present two numerical examples to show the performance of the FEM
described above: one has a known analytical solution that allows a study of the convergence
rate and another one is a benchmark problem.

Example 4.1. Let Ω = (0, 1) × (0, 1) and L be the number of partitions for the interval (0, 1),
i.e., h = 1/L. We choose the velocity vector u(x) = (u1 (x), u2 (x))T and the pressure p as
follows:  2
 u1 (x1 , x2 ) = sin πx1 sin πx2 cos πx2 ,
u (x , x ) = − sin2 πx2 sin πx1 cos πx1 , (4.1)
 2 1 2
p(x1 , x2 ) = − cos 2πx1 cos 2πx2 /16.
It is straightforward to check that (u, p) is the exact solution of problem (2.1) with a body force
f = (f1 , f2 )T , where
1
f1 (x1 , x2 ) = π cos 2πx2 sin 2πx1 − 2π 2 cos2 πx1 cos πx2 sin πx2
8
+ 6π 2 cos πx2 sin2 πx1 sin πx2 + π cos πx1 cos2 πx2 sin3 πx1 sin2 πx2
+ π cos πx1 sin3 πx1 sin4 πx2 , (4.2)
1
f2 (x1 , x2 ) = π cos 2πx1 sin 2πx2 + 2π 2 cos πx1 cos2 πx2 sin πx1
8
− 6π 2 cos πx1 sin πx1 sin2 πx2 + π cos2 πx1 cos πx2 sin2 πx1 sin3 πx2
+ π cos πx2 sin4 πx1 sin3 πx2 . (4.3)

Table 4.1: Numerical error and rate for Example 4.1.

Meshes H 1 error for u rate H 0 error for p rate
N =4 5.39 × 10−1 1.47 × 10−1
N =8 3.13 × 10−1 0.78 5.47 × 10−2 1.42
N = 16 1.64 × 10−1 0.93 1.65 × 10−2 1.73
N = 32 8.46 × 10−2 0.95 4.57 × 10−3 1.85

Fig. 4.1 shows the pressure and velocity field given by (4.1). Table 4.1 gives the numerical
errors and convergence rates obtained on successively refined meshes. These results agree with
the optimal theoretical convergence rates except that the convergence rate for p is great than
1.
A Mixed Finite Element Method on a Staggered Mesh for Navier-Stokes Equations 823
1

0.8

0.6

y
0.05

0
0.4

p
-0.05
0 0

0.2 0.2
0.2
0.4 0.4
Y 0.6 0.6 X

0.8 0.8

1 1 0
0.2 0.4 0.6 0.8
(a) (b) x

Fig. 4.1. The solution (4.1). (a) the pressure field and (b) the velocity field.

Example 4.2. The second example is the driven cavity problem, in which the incompressible
fluid is enclosed in a square box, with an imposed velocity of unity in the horizontal direction
on the top boundary, and a no slip condition on the remaining walls.
This problem has been widely used for validating incompressible fluid dynamic algorithms,
in spite of the singularities at two of its corners. We will compare our results to those obtained
by Ghia et al. [5] and Kaya and Riviere [10].
1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

y

y
y

0.4 0.4 0.4

0.2 0.2 0.2

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

Fig. 4.2. Velocity field for Example 4.2. From left to right: Re = 1, 100 and 400.

We consider the flow for different Reynolds numbers on a fixed mesh with h = 1/20 or
h = 1/40. For low Reynolds numbers (Re = 1) the flow has only one vortex located above
the center. When the Reynolds number increases to Re = 100, the flow pattern starts to form
reverse circulation cells in two lower corners. It is found that the results for Re = 1, 100 and
400 are in good agreement with the solutions presented in [5] and [10].

5. Conclusion
In this paper, we proposed a mixed finite element method on a staggered mesh for the
numerical solution of the Navier-Stokes equations. The method is a conforming quadrilateral
Q1 × Q1 − P0 element approximation on three different meshes. The optimal error estimates of
the numerical solution are given, and two numerical examples are used to show the effectiveness
and feasibility of the given method.
824 H.D. HAN AND M. YAN

Acknowledgements. This work is supported by National Nature Science Foundation of China

(No. 10471073).

References
[1] J.M. Boland and R.A. Nicolaides, On the stability of bilinear-constant velocity-presssure finite
elements, Numer. Math., 44 (1984), 219-222.
[2] H. Han and X. Wu, A new mixed finite element formulation and the MAC method for the Stokes
equations, SIAM J. Numer. Anal., 35 (1998), 560-571.
[3] H. Han, An economical finite element scheme for Navier-Stokes equations, J. Comput. Math., 5
(1987), 135-143.
[4] V. Girault and P.A. Raviart, Finite Element Approximation of the Navier-Stokes Equations,
Springer-Verlag, 1981.
[5] U. Ghia, K.N. Ghia and C.T. Shin, High-resolutions for incompressible flows using the Navier-
Stokes equations and a multigrid method, J. Comput. Phys., 48 (1982), 387-411.
[6] R.L. Sani, P.M. Gresho, R.L. Lee and D.F. Griffiths, The cause and cure of the sprious pres-
sures generated by certain finite element method solutions of the incompressible Navier-Stokes
eaquations. Parts 1 and 2, Int. J. Numer. Meth. Fl., 1 (1981), 17-43; 171-206.
[7] L.P. Franca and T.J.R. Hughes, Two classes of mixed finite element methods, Comput. Method.
Appl. Math., 69 (1988), 89-129.
[8] W. Heinrichs, Least-squares spectral collocation for the Stokes problem, J. Sci. Comput., 21
(2004), 81-90.
[9] Y. He, A. Wang and L. Mei, Stabilized finite-element method for the stationary Navier-Stokes
equations, J. Eng. Math., 51 (2005), 367-380.
[10] S. Kaya and B. Riviere, A two-grid stabilization method for solving the steady-state Navier-Stokes
equations, Numer. Meth. Part. Diff. Eq., 22 (2006), 728-743.
[11] A. Masud and R.A. Khurram, A multiscale finite element method for the incompressible Navier-
Stokes equations, Comput. Method. Appl., 195 (2006), 1750-1777.