Applied Mathematics Letters

Error analysis of nonconforming finite element method for the incompressible Navier-
Stokes problem with low regularity
--Manuscript Draft--

Manuscript Number: AML-D-24-01295

Article Type: Regular Article

Keywords: Navier-Stokes equation; Error estimates; Nonconforming finite element; Low

regularity.

Abstract: The conventional errors estimates of the nonconforming finite elements for the
Navier-Stokes (NS) equation require a high regularity assumption on the exact
solution. In this paper, we derive an error analysis for the nonconforming Crouzeix-
Raviart finite element approximated to the stationary incompressible NS problem with
low regularity $(\mathbf{u}, p) \in \mathbf{H}^{1+s}(\Omega) \times H^{s}(\Omega), s
\in (0,1]$. The optimal error estimates both in a broken $H^1$-norm for the velocity
and in an $L^2$-norm for the pressure are derived under such a regularity assumption,
which are supplemental to the existing error estimates of nonconforming elements for
the NS equations with lower regularity $(\mathbf{u}, p) \in \mathbf{H}^{1+s}(\Omega)
\times H^{s}(\Omega), s \in (0,\frac{1}{2}]$.

Finally, the theoretical result is verified through numerical experiment.

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation
Manuscript Click here to view linked References

Error analysis of nonconforming finite element method for the

incompressible Navier-Stokes problem with low regularity

Xiaolong Wanga , Mingxia Lia,∗, Shipeng Maob

a School of Science, China University of Geosciences (Beijing), Beijing 100083, China
b LSEC, Academy of Mathematics and System Sciences, Chinese Academy of Sciences, School of Mathematical Science,
University of Chinese Academy of Sciences, Beijing, 100190, China

Abstract
The conventional errors estimates of the nonconforming finite elements for the Navier-Stokes (NS) equa-
tion require a high regularity assumption on the exact solution. In this paper, we derive an error analysis
for the nonconforming Crouzeix-Raviart finite element approximated to the stationary incompressible NS
problem with low regularity (u, p) ∈ H1+s (Ω) × H s (Ω), s ∈ (0, 1]. The optimal error estimates both in a
broken H 1 -norm for the velocity and in an L2 -norm for the pressure are derived under such a regularity
assumption, which are supplemental to the existing error estimates of nonconforming elements for the NS
equations with lower regularity (u, p) ∈ H1+s (Ω)×H s (Ω), s ∈ (0, 21 ]. Finally, the theoretical result is verified
through numerical experiment.
Keywords: Navier-Stokes equation; Error estimates; Nonconforming finite element; Low regularity.

1. Introduction

The incompressible NS problem has been extensively studied by using nonconforming finite elements
approximation, see [5, 7, 11, 18] and the references therein. In particular, the classical Crouzeix-Raviart
nonconforming finite elements [4, 7] are known for being pointwise divergence-free within each element and
continuous at edge midpoints. It seems to be quite attractive for discretizing the NS problem since they
possess favorable stability properties and several advantages concerning parallelization and fast iterative
solvers.
For the error estimates of nonconforming finite element methods for the incompressible NS equation, the
traditional approach commonly employs the techniques based on the Second Strang’s Lemma [3, 10], which
considers both the approximation and consistency errors. The consistent errors either rely on integration
by parts or are transformed into some jump flux terms along the element edges [3, 6], which results in
the requirement regularity of (u, p) ∈ H1+s (Ω) × H s (Ω), s > 12 . However, The regularity of the solution
is deteriorated by the presence of interior or boundary layers, shock fronts and complex geometries with
corners. Due to the highly nonlinear structure and complex geometries in practice, the incompressible NS
problem often admit low regularity solutions, which does not satisfies the assumption on the conventional
consistency error estimates. For the NS problem under low regularity (u, p) ∈ H1+s (Ω) × H s (Ω), s ∈ (0, 21 ],
both the normal derivative of the velocity u and the pressure p are not well-defined on the edge of the mesh,
which makes the traditional consistency error estimation techniques no longer applicable.
In recent years, starting from the initial works [12] and [17], some important progress have been made to
handle the consistency error for various models with nonconforming finite elements in case of low regularity.
The methods for handling error estimates for problems with low regularity have been extended to various

∗ Corresponding author.
Email addresses: wxl@email.cugb.edu.cn (Xiaolong Wang), limx@lsec.cc.ac.cn (Mingxia Li), maosp@lsec.cc.ac.cn
(Shipeng Mao)

Preprint submitted to Elsevier July 22, 2024

other situations, the readers are referred to [8, 14] for Crouzeix–Raviart type and mixed type approximation
to the second order problems, [9, 16] for nonconforming Morley type elements for fourth order problems, and
[13] for the nonconforming virtual element methods. Concerning the Stokes equation, which is the problem
nonconforming finite element methods truly advantageous applied, we refer to [1, 2, 15, 19] for various type
numerical methods with low regularity. However, all the works listed above only considered linear problems,
none of them have studied the error estimates of nonconforming finite element methods for the NS equation
with low regularity.
In this paper, we provide an error analysis for the nonconforming Crouzeix-Raviart finite element ap-
plied to the NS problem with low regularity. The main technique is to develop a proper high order quasi-
interpolation operator and a special treatment to the nonlinear convective term. We prove that the consis-
tency error is determined by its associated approximation error and a higher order term (h.o.t.), which will
become negligible when f is a piecewise smooth function. Therefore, the total error of the nonconforming
finite element method is primarily controlled by the approximation error, which aligns with the conforming
finite element methods. As a result, the error estimates are obtained in energy norms under the regularity
assumption(u, p) ∈ H1+s (Ω) × H s (Ω) with any s ∈ (0, 1], which is a supplement to the prior error estimates
of nonconforming elements for the NS problem with lower regularity (u, p) ∈ H1+s (Ω) × H s (Ω), 0 < s ≤ 12 .
The paper is organized as follows. In Section 2, we present the mathematical framework for the steady
NS equation and introduce the nonconforming Crouzeix-Raviart finite element space. We also propose a
special quasi-interpolation operator and its key properties. In Section 3, we derive a new error estimate of
the nonconforming finite element methods for NS equation in the energy norm. Some numerical tests are
performed to verify the theoretical results in Section 4.

2. Nonconforming finite element methods and the quasi-interpolation operator

2.1. Governing equations and the variational formulation

Let Ω be a bounded, open subset in R2 , with Lipschitz boundary ∂Ω. We use the standard notation for
the Sobolev spaces. The norm and seminorm of H s (Ω) are represented by ∥ · ∥s,Ω and | · |s,Ω , respectively.
The inner product in L2 (Ω) is denoted as (·, ·)Ω .
We consider the stationary NS problem: find the velocity u and the pressure p such that

−ν∆u + (u · ∇)u + ∇p =f
 in Ω,
∇ · u =0 in Ω, (2.1)

u =0 on ∂Ω,


where f denotes a given external force, and ν > 0 is the viscosity.

Then the variational formulation of problem (2.1) is (cf. [10, 5]): find (u, p) ∈ V × Q such that
(
a(u, v) + as (u; u, v) + b(v, p) = (f , v) ∀v ∈ V,
(2.2)
b(u, q) = 0 ∀q ∈ Q,

where V = H10 (Ω)2 , Q = L20 (Ω), a(u, v) = ν(∇u, ∇v), b(v, p) = −(∇ · v, p), and the trilinear form
1
as (u; v, w) = [a1 (u; v, w) − a1 (u; w, v)] , a1 (u; v, w) = ((u · ∇)v, w) .
2
2.2. The discrete problem and the quasi-interpolation operator construction
Let Th be a regular triangulation (cf. [3]) of Ω, with each element T being an open triangle in 2D, the
size of T is denoted by hT , h = maxT ∈Th hT . All the edges in 2D and vertexes of the mesh are denoted
by Eh and Vh . Let Eh = Eh′ ∪ Eh∂ with Eh′ := {E ∈ Eh ; E ∩ ∂Ω = ∅} , Eh∂ := {E ∈ Eh ; E ⊂ ∂Ω}. For any
E ∈ Eh′ , there exists T1 , T2 ∈ Th such that E = T1 ∩ T2 . We define the jumps and averages of function v

2
on E by [v]E = v T1 E − v T2 E , {v}E = 12 v T1 E + v T2 E , where v Ti = v|T i , i = 1, 2. Let Vh denote the

nonconforming finite element space over Th , which is defined as

 Z 
Vh = vh ∈ L2 (Ω), vh |T ∈ P1 (T ), ∀T ∈ Th , [vh ]E ds = 0, ∀E ∈ Eh .
E

Then there exists an interpolation operator Πh such that

∥v − Πh v∥h ≤ Chs |v|1+s,Ω , ∀v ∈ H1+s (Ω), 0 < s ≤ 1. (2.3)
The pressure will be approximated by the piecewise-constant functions, which is given by
Qh := {qh ∈ Q, qh |T ∈ P0 (T ), ∀T ∈ Th } .
We introduce the bilinear ah (·, ·), bh (·, ·) and trilinear a1,h (·, ·) forms as
X X
ah (uh , vh ) = ν (∇uh , ∇vh )T , bh (vh , ph ) = − (∇ · vh , ph )T ,
T ∈Th T ∈Th
X
a1,h (uh ; vh , wh ) = ((uh · ∇)vh , wh )T ,
T ∈Th

1
as,h (uh ; vh , wh ) = [a1,h (uh ; vh , wh ) − a1,h (uh ; wh , vh )] . (2.4)
2
For any uh , vh , wh ∈ Vh , integration by parts on each element gives
X X
a1,h (uh ; vh , wh ) = −a1,h (uh ; wh , vh ) − ((∇ · uh )vh , wh )T + ((uh · n)vh , wh )E , (2.5)
T ∈Th E∈Eh

and hence
1 X 1 X
a1,h (uh ; vh , wh ) = as,h (uh ; vh , wh ) − ((∇ · uh )vh , wh )T + ((uh · n) vh , wh )E . (2.6)
2 2
T ∈Th E∈Eh

! 21
|21,T
P
With the discrete norm defined as ∥ · ∥h = |· , we know from [5] that
T ∈Th

a1,h (uh ; vh , wh ) ≤ C∥uh ∥h ∥vh ∥h ∥wh ∥h , as,h (uh ; vh , wh ) ≤ C∥uh ∥h ∥vh ∥h ∥wh ∥h . (2.7)
The nonconforming mixed FE solution of (2.2) is defined by: find a pair (uh , ph ) ∈ Vh × Qh such that
(
ah (uh , vh ) + as,h (uh ; uh , vh ) + bh (vh , ph ) = (f , vh ) ∀vh ∈ Vh ,
(2.8)
bh (uh , qh ) = 0 ∀qh ∈ Qh .

Following the second Strang’s lemma [3, 10], we have

 
Eh (u, p, wh )
∥u − uh ∥h + ∥p − ph ∥0,Ω ≤ C inf ∥u − vh ∥h + inf ∥p − qh ∥0,Ω + sup , (2.9)
vh ∈Vh qh ∈Qh wh ∈Vh ∥wh ∥h

where Eh (u, p, wh ) = ah (u, wh ) + as,h (u; u, wh ) + bh (wh , p) − (f , wh ) is the usual consistency error.
Given a fixed integer of k ≥ 4, let Xkh be the associated kth-order vector conforming Pk finite element
space. We will construct a proper enriching operator by define the value of Πkh vh at the vertexes of elements.
For any a ∈ Vh , we set ωa as the union of all the elements sharing as one of their vertexes and N (a) as the
corresponding cardinality. Then, for any vh ∈ Vh , we set
(
0, if a ∈ ∂Ω,
k
Πh vh (a) = P
T ⊂ωa vh (a)|T
(2.10)
N (a) , else.
3
The degrees of freedom on the edges and on the elements can be defined by means of
Z Z
Πkh vh qds = {vh }E qds, ∀q ∈ Pk−2 (E) ∀E ∈ Eh , (2.11)
E E
Z Z
Πkh vh qdx = vh qdx, ∀q ∈ Pk−3 (T ) ∀T ∈ Th . (2.12)
T T

According to [10, 17], the interpolation operator Πkh is well-posed and satisfies
2 2
vh − Πkh vh 0,T
+ h2T Πkh vh 1,T
≤ Ch2T |vh |21,ωT , ∀T ∈ Th , (2.13)

where ωT is the patch of elements adjacent to the element T .

3. New energy norm error estimates

In this section, we will prove the following theorem, which is the main result of this paper.

Theorem 3.1. Let (u, p)and (uh , ph ) be the solutions of (2.2) and (2.8), respectively, we have
 ! 21 
 X 
∥u − uh ∥h + ∥p − ph ∥0,Ω ≤ C inf ∥u − vh ∥h + inf ∥p − qh ∥0,Ω + h2T inf ∥f − q∥20,T
vh ∈Vh qh ∈Qh q∈Pk−3 (T ) 
T ∈Th
(3.1)

Proof. Firstly, for any wh ∈ Vh , Πkh wh ∈ V, by the first equation of (2.2), we have

ah u, Πkh wh + as,h u; u, Πkh wh + bh Πkh wh , p − f , Πkh wh = 0.

(3.2)

Next, by the variational formulation (2.8) and (3.2), for any vh ∈ Vh and qh ∈ Qh , we have

Eh (u, p, wh ) =ah (u, wh ) + as,h (u; u, wh ) + bh (wh , p) − (f , wh )

=ah u, wh − Πkh wh + as,h u; u, wh − Πkh wh + bh wh − Πkh wh , p − f , wh − Πkh wh

=ah u − vh , wh − Πkh wh + as,h u − vh ; u, wh − Πkh wh + as,h vh ; u − vh , wh − Πkh wh

+ bh wh − Πkh wh , p − qh − f , wh − Πkh wh + ah vh , wh − Πkh wh

+ as,h vh ; vh , wh − Πkh wh + bh wh − Πkh wh , qh .

(3.3)
We first consider the term ah vh , wh − Πkh wh . An application of Green’s formula, noticing that the integral
of wh along the interior edge is continuous and ∂v ∂n E is a constant vector function, we have
h

X Z
ah vh , wh − Πkh wh = ∇vh ∇ wh − Πkh wh dx

T ∈Th T
X Z
div (∇vh ) wh − Πkh wh dx


=−
T ∈Th T
X Z  ∂vh      (3.4)
∂vh
wh − Πkh wh E + wh − Πkh wh E ds
 
+
∂n E ∂n E
E∈Eh E
X Z X Z  ∂vh  
div (∇vh ) wh − Πkh wh dx + wh − Πkh wh


=− E
ds.
T E ∂n E
T ∈Th E∈Eh

Since k ⩾ 4, and noticing that div (∇vh )|T = 0, by the property (2.11), we have

ah vh , wh − Πkh wh = 0.


(3.5)
4
We can also check that
X Z X Z
bh wh − Πkh wh , qh = div wh − Πkh wh qh dx = − wh − Πkh wh ∇qh dx

T ∈Th T T ∈Th T
X Z
wh − Πkh wh · n wh − Πkh wh · n E ds




+ [qh ]E E
+ {qh }E
E∈Eh E (3.6)
X Z X Z
wh − Πkh wh ∇qh dx + [qh ]E wh − Πkh wh · nds



=−
T ∈Th T E∈Eh E

= 0.

Due to (2.4), we have

1
as,h vh ; vh , wh − Πkh wh = a1,h (vh ; vh , wh − Πkh wh ) − a1,h (vh ; wh − Πkh wh , vh ) .

2
Noticing that (vh · ∇)vh |T ∈ P1 (T ), if we choose k ≥ 4, by the property (2.12), there holds
X Z
a1,h (vh ; vh , wh − Πkh wh ) = (vh · ∇)vh (wh − Πkh wh )dx = 0. (3.7)
T ∈Th T

By (2.5), it can be derived that

a1,h (vh ; wh − Πkh wh , vh ) = − a1,h (vh ; vh , wh − Πkh wh )

1 X 1 X
((∇ · vh )vh , wh − Πkh wh )T + (vh · n) vh , wh − Πkh wh E .


−
2 2
T ∈Th E∈Eh

Since (∇ · vh )vh |T ∈ P1 (T ), (vh · n) vh |E ∈ P2 (E), by (2.11), (2.12), we can derive that

X X Z
((∇ · vh )vh , wh − Πkh wh )T = (∇ · vh )vh (wh − Πkh wh )dx = 0, (3.8)
T ∈Th T ∈Th T

X X Z
Πkh wh E (vh · n)vh (wh − Πkh wh )ds = 0,


(vh · n) vh , wh − = (3.9)
E∈Eh E∈Eh E

which implies that

a1,h (vh ; wh − Πkh wh , vh ) = 0. (3.10)
Then by (3.7) and (3.10), we have

as,h vh ; vh , wh − Πkh wh = 0.


(3.11)


Concerning the term f , wh − Πkh wh , by the properties (2.12) and (2.13), for any q ∈ Pk−3 (T ), we derive
X Z X Z
f , wh − Πkh wh = f wh − Πkh wh dx = (f − q) wh − Πkh wh dx

T ∈Th T T ∈Th T
! 21 ! 12
X X 2
≤C h2T ∥f − q∥20,T h−2
T wh − Πkh wh 0,T (3.12)
T ∈Th T ∈Th
! 21
X
≤C h2T ∥f − q∥20,T ∥wh ∥h .
T ∈Th

5
The left four terms can be estimated easily by an application of Cauchy-Schwarz inequality, (2.7) and (2.13),

ah u − vh , wh − Πkh wh ≤ C ∥u − vh ∥h ∥wh ∥h ,


(3.13)

bh wh − Πkh wh , p − qh ≤ C ∥p − qh ∥0,Ω ∥wh ∥h ,

(3.14)

as,h u − vh ; u, wh − Πkh wh ≤ C ∥u − vh ∥h ∥wh ∥h ,

(3.15)
as,h vh ; u − vh , wh − Πkh wh ≤ C ∥u − vh ∥h ∥wh ∥h .


(3.16)
Then, the proof is finished by a combination of (3.2)-(3.16). □
Let Ih be the L2 projection operator from Q to Qh , we have

∥p − Ih p∥0,Ω ≤ Chγ ∥p∥γ,Ω . (3.17)

By taking vh = Πh u, qh = Ih p in (3.1), by (3.1), (3.17) and (2.3) we can get the following theorem.
Theorem 3.2. Let (u, p) and (uh , ph ) be the solutions of (2.2) and (2.8), respectively. Suppose that u ∈
H1+γ
0 (Ω) and p ∈ H γ (Ω) with 0 < γ ≤ 1, then we have

∥u − uh ∥h + ∥p − ph ∥0,Ω ≤ Chγ (∥u∥1+γ,Ω + ∥p∥γ,Ω + h.o.t. ) . (3.18)

4. Numerical experiments

In this section, we present some numerical examples to confirm our theoretical analysis. For simplicity,
we consider the stationary NS equations with Ω = (0, 1) × (0, 1) with the exact velocity u = (u1 , u2 )T and
pressure p are
1 θ 1 θ
u = (−r 2 cos( ), r 2 sin( )), p = r(cos(θ) + sin(θ)).
2 2
2
Here, r and θ are polar coordinates in R . It can be checked that the singular velocity solution is
1
in H 1+ 2 −ϵ with arbitrarily small ϵ > 0. Since the velocity solution is not smooth, the highest possible
convergence order of the velocity in energy norm for the the nonconforming finite element method is only
1
2 . This is confirmed by the data in the Table 1 and the strong evidence presented here serves to validate
our newly proposed theory for the error estimation with low regularity, as stated in Theorem 3.2.

Table 1: The errors and convergence orders for the singular solution.

m×n 2×2 4×4 8×8 16 × 16 32 × 32 64 × 64 128 × 128

∥u − uh ∥h 0.309003 0.221667 0.144669 0.0941586 0.0628737 0.0429878 0.0298468
Convergence order 0.48 0.62 0.62 0.58 0.55 0.53
∥p − ph ∥0,Ω 0.276308 0.150102 0.0776451 0.0451564 0.0291859 0.0198208 0.0137138
Convergence order 0.88 0.95 0.78 0.63 0.55 0.53

References
[1] S. Badia, R. Codina, T. Gudi, J. Guzmán, Error analysis of discontinuous Galerkin methods for the Stokes problem under
minimal regularity. IMA J. Numer. Anal. 34(2), 800–819 (2013).
[2] A. Baier-Reinio, S. Rhebergen, G. N. Wells, Analysis of pressure-robust embedded-hybridized discontinuous Galerkin
methods for the stokes problem under minimal regularity, Journal of Scientific Computing 92 (2) (2022) 51.
[3] S. C. Brenner, The mathematical theory of finite element methods, Springer, 2008.
[4] S. C. Brenner, Forty years of the Crouzeix-Raviart element, Numerical Methods for Partial Differential Equations 31 (2)
(2015) 367–396.
[5] Z. Cai, J. Douglas, X. Ye, A stable nonconforming quadrilateral finite element method for the stationary stokes and
navier–stokes equations, Calcolo 36 (1999) 215–232.
[6] P. G. Ciarlet, The finite element method for elliptic problems, SIAM, 2002.

6
 [7] M. Crouzeix and P. A. Raviart, Conforming and nonconforming finite element methods for solving the stationary Stokes
equations, RAIRO Anal. Numér. 7 (1973), 33-76.
[8] A. Ern, J.-L. Guermond, Quasi-optimal nonconforming approximation of elliptic pdes with contrasted coefficients and
h1+r , r > 0 regularity, Foundations of Computational Mathematics 22 (5) (2022) 1273–1308.
[9] D. Gallistl, Morley finite element method for the eigenvalues of the biharmonic operator, IMA J. Numer. Anal., 35 (2015),
1779–1811.
[10] V. Girault and P. A. Raviart, Finite Element Methods for Navier-Stokes Equations, Theorem and Algorithms, Springer-
Verlag, Berlin, 1986.
[11] Z. Ge, M. Feng and Y. N. He, A stabilized nonconforming finite element method based on multiscale enrichment for the
stationary Navier–Stokes equations, Applied Mathematics and Computation, 202 (2008),700-707.
[12] T. Gudi, A new error analysis for discontinuous finite element methods for linear elliptic problems, Mathematics of
Computation 79 (272) (2010) 2169–2189.
[13] J. Huang, Y. Yu, A medius error analysis for nonconforming virtual element methods for Poisson and biharmonic equations,
J. Comput. Appl. Math. 386 (2021), Paper No. 113229, 20 pp.
[14] M. Li, S. Mao, A new a priori error analysis of nonconforming and mixed finite element methods, Applied Mathematics
Letters, 26 (1) (2013) 32–37.
[15] M. Li, S. Mao, S. Zhang, New error estimates of nonconforming mixed finite element methods for the stokes problem,
Mathematical Methods in the Applied Sciences 37 (7) (2014) 937–951.
[16] M. Li, X. Guan, S. Mao, New error estimates of the Morley element for the plate bending problems., J. Comput. Appl.
Math 263(2014), 405–416.
[17] S. Mao, Z. Shi, On the error bounds of nonconforming finite elements, Science China Mathematics, 53 (2010) 2917–2926.
[18] D.Y. Shi and J. Ren, Nonconforming mixed finite element approximation to the stationary Navier-Stokes equations on
anisotropic meshes, Nonlinear Analysis. Theory, Methods & Applications, 71(2009), 3842–3852.
[19] T. Wang, J. K. Zhao, Error analysis of nonconforming virtual element method for Stokes problem with low regularity,
Taiwanese J. Math. 27 (2023), no. 4, 719–735.

7
LaTeX Source Files

Click here to access/download

LaTeX Source Files
Navier-Stokes problem with low regularity6.tex