This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.



Successive finite element methods for Stokes equations

Chunjae Park Department of Mathematics, Konkuk University, Seoul 05029, Korea.  [email protected]
Abstract

This paper will suggest a new finite element method to find a P4P^{4}-velocity and a P3P^{3}-pressure solving Stokes equations. The method solves first the decoupled equation for the P4P^{4}-velocity. Then, four kinds of local P3P^{3}-pressures and one P0P^{0}-pressure will be calculated in a successive way. If we superpose them, the resulting P3P^{3}-pressure shows the optimal order of convergence same as a P3P^{3}-projection. The chief time cost of the new method is on solving two linear systems for the P4P^{4}-velocity and P0P^{0}-pressure, respectively.

1 Introduction

High order finite element methods for incompressible Stokes problems have been developed well in 2-dimensional domain and analyzed along to the inf-sup condition [5, 7, 10]. They, however, endure their large numbers of degrees of freedom. To be worse in case of Scott-Vogelius, instability appears on pressure, if the mesh bears singular vertices.

Recently, we found a so called sting function having an interesting quadrature rule. In the Scott-Vogelius finite element method, it causes the discrete Stokes problem to be singular on the presence of exactly singular vertices. Even on nearly singular vertices, the pressure solution is easy to be spoiled. To overcome the problem based on understanding the causes, we did a new error analysis in a successive way and restored the ruined pressure by simple post-process driven from it [9].

In this paper, we will suggest a new finite element method to calculate a pressure solution at low cost, utilizing the successive way in the precedented new error analysis. In our method, the characteristics of a sting function depicted in Figure 1 will also play a key role.

We will solve first the decoupled equation for velocity in the P4P^{4} divergence-free space inherited from the 𝒞1\mathcal{C}^{1}-Argyris P5P^{5} stream function space. It is simpler and smaller than the divergence-free subspace of the P4P^{4}-P3P^{3} Scott-Vogelius finite element space.

The main stage of our method is the successive 5 steps calculating four kinds of local P3P^{3}-pressures for triangles, regular vertices, nearly singular vertices and singular corners, respectively, as well as one P0P^{0}-pressure in the last step. They are successive in the sense that each step needs the calculated in the previous step.

Superposition of all the calculated reaches at the final P3P^{3}-pressure which shows the optimal order of convergence same as a P3P^{3}-projection of the continuous pressure. The chief time cost of the new method is on solving two linear systems for the P4P^{4}-velocity and P0P^{0}-pressure, respectively.

The paper is organized as follows. In the next section, the detail on finding a P4P^{4}-velocity will be offered. After short review of nearly singular vertices in Section 3, we will introduce a new basis for P3P^{3}-pressures consisting of non-sting, sting and constant functions and decompose the space of P3P^{3}-pressures in Section 4. Then, we will devote Section 5 to describing the successive 5 steps to find a P3P^{3}-pressure solution. Finally, a numerical test will be presented in the last section.

Throughout the paper, for a set S2S\subset\mathbb{R}^{2}, standard notations for Sobolev spaces are employed and L02(S)L_{0}^{2}(S) is the space of all fL2(S)f\in L^{2}(S) whose integrals over SS vanish. We will use m,S\|\star\|_{m,S}, ||m,S|\star|_{m,S} and (,)S(\cdot,\cdot)_{S} for the norm, seminorm for Hm(S)H^{m}(S) and L2(S)L^{2}(S) inner product, respectively. If S=ΩS=\Omega, it may be omitted in the indices. Denoting by PkP^{k}, the space of all polynomials on 2\mathbb{R}^{2} of order less than or equal kk, f|SPkf\big{|}_{S}\in P^{k} will mean that ff coincides with a polynomial in PkP^{k} on SS.

2 Velocity from the decoupled equation

Let Ω\Omega be a simply connected polygonal domain in 2\mathbb{R}^{2} and {𝒯h}h>0\{\mathcal{T}_{h}\}_{h>0} a regular family of triangulations of Ω\Omega. Denote by 𝒫hk(Ω)\mathcal{P}_{h}^{k}(\Omega), discrete polynomial spaces:

𝒫hk(Ω)={vhL2(Ω):vh|KPk for all triangles K𝒯h},k0.\mathcal{P}_{h}^{k}(\Omega)=\{v_{h}\in L^{2}(\Omega)\ :\ v_{h}\big{|}_{K}\in P^{k}\mbox{ for all triangles }K\in\mathcal{T}_{h}\},\quad k\geq 0.

In this paper, we will approximate a pair of velocity and pressure (𝐮,p)[H01(Ω)]2×L02(Ω)(\mathbf{u},p)\in[H_{0}^{1}(\Omega)]^{2}\times L_{0}^{2}(\Omega) which satisfies an incompressible Stokes problem:

(𝐮,𝐯)+(p,div𝐯)+(q,div𝐮)=(𝐟,𝐯) for all (𝐯,q)[H01(Ω)]2×L02(Ω),(\nabla\mathbf{u},\nabla\mathbf{v})+(p,\mathrm{div}\hskip 1.42262pt\mathbf{v})+(q,\mathrm{div}\hskip 1.42262pt\mathbf{u})=(\mathbf{f},\mathbf{v})\quad\mbox{ for all }(\mathbf{v},q)\in[H_{0}^{1}(\Omega)]^{2}\times L_{0}^{2}(\Omega), (1)

for a given body force 𝐟[L2(Ω)]2\mathbf{f}\in[L^{2}(\Omega)]^{2}.

Let Ah5A_{h}^{5} be a space of 𝒞1\mathcal{C}^{1}-Argyris triangle elements [2, 4] such that

Ah5=𝒫h5(Ω)H02(Ω),A_{h}^{5}=\mathcal{P}_{h}^{5}(\Omega)\cap H_{0}^{2}(\Omega),

where

H02(Ω)={ϕH2(Ω):ϕ,ϕx,ϕyH01(Ω)}.H_{0}^{2}(\Omega)=\{\phi\in H^{2}(\Omega)\ :\ \phi,\phi_{x},\phi_{y}\in H_{0}^{1}(\Omega)\}.

Defining a divergence-free space VhV_{h} as

Vh={(ϕh,y,ϕh,x):ϕhAh5}[𝒫h4(Ω)H01(Ω)]2,V_{h}=\{(\phi_{h,y},-\phi_{h,x})\ \ :\ \phi_{h}\in A_{h}^{5}\hskip 1.0pt\}\subset[\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega)]^{2},

we can solve 𝐮hVh\mathbf{u}_{h}\in V_{h} satisfying

(𝐮h,𝐯h)=(𝐟,𝐯h) for all 𝐯hVh.(\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})=(\mathbf{f},\mathbf{v}_{h})\quad\mbox{ for all }\mathbf{v}_{h}\in V_{h}. (2)
Theorem 2.1.

Let (𝐮,p)[H01(Ω)]2×L02(Ω)(\mathbf{u},p)\in[H_{0}^{1}(\Omega)]^{2}\times L_{0}^{2}(\Omega) satisfy (1). If 𝐮[H5(Ω)]2\mathbf{u}\in[H^{5}(\Omega)]^{2}, then

|𝐮𝐮h|1Ch4|𝐮|5.|\mathbf{u}-\mathbf{u}_{h}|_{1}\leq Ch^{4}|\mathbf{u}|_{5}. (3)
Proof.

If (𝐮,p)[H01(Ω)]2×L02(Ω)(\mathbf{u},p)\in[H_{0}^{1}(\Omega)]^{2}\times L_{0}^{2}(\Omega) satisfies (1), we have div𝐮=0\mathrm{div}\hskip 1.42262pt\mathbf{u}=0. Thus, if 𝐮[H5(Ω)]2\mathbf{u}\in[H^{5}(\Omega)]^{2}, there exists a stream function ϕH02(Ω)H6(Ω)\phi\in H_{0}^{2}(\Omega)\cap H^{6}(\Omega) such that [6]

𝐮=(ϕy,ϕx).\mathbf{u}=(\phi_{y},-\phi_{x}).

Let ΠhϕAh5\Pi_{h}\phi\in A_{h}^{5} be the projection of ϕ\phi and Πh𝐮=((Πhϕ)y,(Πhϕ)x)Vh\Pi_{h}\mathbf{u}=\left((\Pi_{h}\phi)_{y},-(\Pi_{h}\phi)_{x}\right)\in V_{h}. Then we have

|ϕΠhϕ|2Ch4|ϕ|6,|𝐮Πh𝐮|1Ch4|𝐮|5.|\phi-\Pi_{h}\phi|_{2}\leq Ch^{4}|\phi|_{6},\quad|\mathbf{u}-\Pi_{h}\mathbf{u}|_{1}\leq Ch^{4}|\mathbf{u}|_{5}. (4)

We have from (1), (2) that

(𝐮𝐮h,𝐯h)=0 for all 𝐯hVh.(\nabla\mathbf{u}-\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})=0\quad\mbox{ for all }\mathbf{v}_{h}\in V_{h}.

It is written in the form:

(Πh𝐮𝐮h,𝐯h)=(Πh𝐮h𝐮,𝐯h) for all 𝐯hVh.(\nabla\Pi_{h}\mathbf{u}-\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})=(\nabla\Pi_{h}\mathbf{u}_{h}-\nabla\mathbf{u},\nabla\mathbf{v}_{h})\quad\mbox{ for all }\mathbf{v}_{h}\in V_{h}. (5)

We establish (3) from (4), (5) with 𝐯h=Πh𝐮𝐮hVh\mathbf{v}_{h}=\Pi_{h}\mathbf{u}-\mathbf{u}_{h}\in V_{h}. ∎

3 Nearly singular vertex

A vertex 𝐕\mathbf{V} is called exactly singular if the union of all edges sharing 𝐕\mathbf{V} belongs to the union of two lines. To be precise, let K1,K2,,KJK_{1},K_{2},\cdots,K_{J} be all JJ triangles sharing 𝐕\mathbf{V} and denote by θ(Kk)\theta(K_{k}), the angle of KkK_{k} at 𝐕\mathbf{V}, k=1,2,,Jk=1,2,\cdots,J. Define

Υ(𝐕)={θ(Ki)+θ(Kj):KiKj is an edge, i,j=1,2,,J}.\Upsilon(\mathbf{V})=\{\theta(K_{i})+\theta(K_{j})\ :\ K_{i}\cap K_{j}\mbox{ is an edge, }i,j=1,2,\cdots,J\}.

Then Υ(𝐕)={π} or \Upsilon(\mathbf{V})=\{\pi\}\mbox{ or }\emptyset if and only if 𝐕\mathbf{V} is exactly singular.

Since {𝒯h}h>0\{\mathcal{T}_{h}\}_{h>0} is regular, there exists ϑ>0\vartheta>0 such that

ϑ=inf{θ:θ is an angle of a triangle K𝒯h,h>0},\vartheta=\inf\{\theta\ :\ \theta\mbox{ is an angle of a triangle }K\in\mathcal{T}_{h},h>0\},

which depends on the shape regularity parameter σ\sigma of {𝒯h}h>0\{\mathcal{T}_{h}\}_{h>0}. Set

ϑσ=min(ϑ,π/6),\vartheta_{\sigma}=\min(\vartheta,\pi/6),

then call a vertex 𝐕\mathbf{V} to be nearly singular if Υ(𝐕)=\Upsilon(\mathbf{V})=\emptyset or

|Θπ|<ϑσ for all ΘΥ(𝐕),|\Theta-\pi|<\vartheta_{\sigma}\mbox{ for all }\Theta\in\Upsilon(\mathbf{V}),

otherwise regular. From the following lemma [9], we note that each interior nearly singular vertex is isolated from others.

Lemma 3.1.

There is no interior edge connecting two nearly singular vertices.

4 Decomposition of 𝒫h3(Ω)\mathcal{P}_{h}^{3}(\Omega)

For each triangle K𝒯hK\in\mathcal{T}_{h}, define

P3(K)={qh𝒫h3(Ω):qh=0 on ΩK}.P^{3}(K)=\{q_{h}\in\mathcal{P}_{h}^{3}(\Omega):\ q_{h}=0\mbox{ on }\Omega\setminus K\}.

In the remaining of the paper, we will use the following notations:

  • CσC_{\sigma} : a generic constant which depends only the shape regularity parameter σ\sigma of {𝒯h}h>0\{\mathcal{T}_{h}\}_{h>0},

  • 𝒦(𝐕)\mathcal{K}(\mathbf{V}) : the union of all triangles in 𝒯h\mathcal{T}_{h} sharing a vertex 𝐕\mathbf{V} as in Figure 3-(b),

  • 𝒱h\mathcal{V}_{h} : the set of all vertices in 𝒯h\mathcal{T}_{h},

  • |K|,|E||K|,|E|: the area and length of a triangle KK and an edge EE, respectively.

4.1 sting function

Let 𝐕\mathbf{V} be a vertex of a triangle KK. Then there exists a unique function 𝒮𝐕KP3(K)\mathcal{S}_{\mathbf{V}K}\in P^{3}(K) satisfying the following quadrature rule:

K𝒮𝐕K(x,y)q(x,y)𝑑x𝑑y=|K|q(𝐕) for all qP3,\int_{K}\mathcal{S}_{\mathbf{V}K}(x,y)q(x,y)\ dxdy=|K|q(\mathbf{V})\quad\mbox{ for all }q\in P^{3}, (6)

since the both sides of (6) are linear functionals on P3P^{3}. We call 𝒮𝐕K\mathcal{S}_{\mathbf{V}K} a sting function of 𝐕\mathbf{V} on KK, named after the shape of its graph as in Figure 1.

Refer to caption
Figure 1: a sting function 𝒮𝐕K\mathcal{S}_{\mathbf{V}K} of 𝐕\mathbf{V} on KK.

In the reference triangle K^\widehat{K} of vertices 𝐕^1=(0,0),𝐕^2=(1,0),𝐕^3=(0,1)\widehat{\mathbf{V}}_{1}=(0,0),\widehat{\mathbf{V}}_{2}=(1,0),\widehat{\mathbf{V}}_{3}=(0,1), we have the following 3 sting functions:

𝒮𝐕^1K^=560(1xy)3630(1xy)2+180(1xy)10,𝒮𝐕^2K^=560x3630x2+180x10,𝒮𝐕^3K^=560y3630y2+180y10.\begin{array}[]{lll}\mathcal{S}_{\widehat{\mathbf{V}}_{1}\widehat{K}}&=&560(1-x-y)^{3}-630(1-x-y)^{2}+180(1-x-y)-10,\\ \mathcal{S}_{\widehat{\mathbf{V}}_{2}\widehat{K}}&=&560x^{3}-630x^{2}+180x-10,\\ \mathcal{S}_{\widehat{\mathbf{V}}_{3}\widehat{K}}&=&560y^{3}-630y^{2}+180y-10.\end{array} (7)

Since 𝒮𝐕K=𝒮𝐕^1K^F1\mathcal{S}_{\mathbf{V}K}=\mathcal{S}_{\widehat{\mathbf{V}}_{1}\widehat{K}}\circ F^{-1} for an affine transformation F:K^KF:\widehat{K}\longrightarrow K, we estimate

𝒮𝐕K0,KCσ|K|1/2.\|\mathcal{S}_{\mathbf{V}K}\|_{0,K}\leq C_{\sigma}|K|^{1/2}. (8)

4.2 non-sting function

Given triangle K𝒯hK\in\mathcal{T}_{h} of vertices 𝐕1,𝐕2,𝐕3\mathbf{V}_{1},\mathbf{V}_{2},\mathbf{V}_{3}, the following 16-point Lyness quadrature rule is exact for every polynomial qP6q\in P^{6}:

Kq(x,y)𝑑x𝑑y=|K|i=015q(𝐆i)gi,\int_{K}q(x,y)\ dxdy=|K|\sum_{i=0}^{15}q(\mathbf{G}_{i})g_{i}, (9)

where 𝐆4,𝐆5,,𝐆15\mathbf{G}_{4},\mathbf{G}_{5},\cdots,\mathbf{G}_{15} belong to K\partial K and 𝐆0\mathbf{G}_{0} is the gravity center of KK and 𝐆1,𝐆2,𝐆3\mathbf{G}_{1},\mathbf{G}_{2},\mathbf{G}_{3} are the centers of the medians, that is,

𝐆1=12𝐕1+14𝐕2+14𝐕3,𝐆2=14𝐕1+12𝐕2+14𝐕3,𝐆3=14𝐕1+14𝐕2+12𝐕3,\mathbf{G}_{1}=\frac{1}{2}\mathbf{V}_{1}+\frac{1}{4}\mathbf{V}_{2}+\frac{1}{4}\mathbf{V}_{3},\quad\mathbf{G}_{2}=\frac{1}{4}\mathbf{V}_{1}+\frac{1}{2}\mathbf{V}_{2}+\frac{1}{4}\mathbf{V}_{3},\quad\mathbf{G}_{3}=\frac{1}{4}\mathbf{V}_{1}+\frac{1}{4}\mathbf{V}_{2}+\frac{1}{2}\mathbf{V}_{3},

and g0,g1,,g15g_{0},g_{1},\cdots,g_{15} are nonzero quadrature weights [8, 9].

Lemma 4.1.

If a cubic function qP3q\in P^{3} satisfies the following 10 conditions, then q=0q=0.

q(𝐆j)=(0,0),j=0,1,2,3, and q(𝐆0)=q(𝐆1)=0.\nabla q(\mathbf{G}_{j})=(0,0),j=0,1,2,3,\quad\mbox{ and }\quad q(\mathbf{G}_{0})=q(\mathbf{G}_{1})=0. (10)
Proof.

Since 𝐆0\mathbf{G}_{0} is the gravity center of 𝐆1,𝐆2,𝐆3\mathbf{G}_{1},\mathbf{G}_{2},\mathbf{G}_{3}, it suffices to prove q=0q=0 in case

𝐆1=(0,0),𝐆2=(1,1),𝐆3=(1,1),𝐆0=(2/3,0).\mathbf{G}_{1}=(0,0),\mathbf{G}_{2}=(1,1),\mathbf{G}_{3}=(1,-1),\mathbf{G}_{0}=(2/3,0).

Let

a(t)=q(1,t),b(t)=q(2/3,t),c(t)=q(t,t),d(t)=q(t,t).a(t)=q(1,t),\ b(t)=q(2/3,t),\ c(t)=q(t,t),\ d(t)=q(t,-t).

From the condition (10), qq vanishes on the line y=0y=0 and a(±1)=0a^{\prime}(\pm 1)=0. It means

a(t)=β(t33t) for some constant β.a(t)=\beta(t^{3}-3t)\quad\mbox{ for some constant }\beta. (11)

Then since c(0)=c(0)=c(1)=0,c(1)=a(1)=2βc(0)=c^{\prime}(0)=c^{\prime}(1)=0,\ c(1)=a(1)=-2\beta, we deduce

c(t)=β(4t36t2).c(t)=\beta(4t^{3}-6t^{2}). (12)

Similarly, from d(0)=d(0)=d(1)=0,d(1)=a(1)=2βd(0)=d^{\prime}(0)=d^{\prime}(1)=0,\ d(1)=a(-1)=2\beta, we have

d(t)=β(4t36t2).d(t)=-\beta(4t^{3}-6t^{2}). (13)

We note that b′′′=qyyy=a′′′=6βb^{{}^{\prime\prime\prime}}=q_{yyy}=a^{{}^{\prime\prime\prime}}=6\beta since qP3q\in P^{3}. Thus, we can write

b(t)=βt3,b(t)=\beta t^{3}, (14)

from b(0)=b(0)=0,b(2/3)=c(2/3)=d(2/3)=b(2/3)b(0)=b^{\prime}(0)=0,\ b(2/3)=c(2/3)=-d(2/3)=-b(-2/3) by (12),(13).

Now, β=0\beta=0 comes from the following equation:

β(2/3)3=b(2/3)=c(2/3)=β(4(2/3)36(2/3)2).\beta(2/3)^{3}=b(2/3)=c(2/3)=\beta(4(2/3)^{3}-6(2/3)^{2}).

Then q=0q=0 since β=0\beta=0 in (11)-(14). ∎

From the unisolvancy in the above lemma, for each k=1,2,3k=1,2,3, we can define two functions 𝒩𝐆kK1,𝒩𝐆kK2P3(K){\mathcal{N}_{\mathbf{G}_{k}K}}^{1},{\mathcal{N}_{\mathbf{G}_{k}K}}^{2}\in P^{3}(K), called non-sting, by their following values:

𝒩𝐆kK1(𝐆0)=0,𝒩𝐆kK1(𝐆k)=0,𝒩𝐆kK1(𝐆j)=(δkj,0),j=0,1,2,3,𝒩𝐆kK2(𝐆0)=0,𝒩𝐆kK2(𝐆k)=0,𝒩𝐆kK2(𝐆j)=(0,δkj),j=0,1,2,3,\begin{array}[]{c}{\mathcal{N}_{\mathbf{G}_{k}K}}^{1}(\mathbf{G}_{0})=0,\ {\mathcal{N}_{\mathbf{G}_{k}K}}^{1}(\mathbf{G}_{k})=0,\ \nabla{\mathcal{N}_{\mathbf{G}_{k}K}}^{1}(\mathbf{G}_{j})=(\delta_{kj},0),\quad j=0,1,2,3,\\ {\mathcal{N}_{\mathbf{G}_{k}K}}^{2}(\mathbf{G}_{0})=0,\ {\mathcal{N}_{\mathbf{G}_{k}K}}^{2}(\mathbf{G}_{k})=0,\ \nabla{\mathcal{N}_{\mathbf{G}_{k}K}}^{2}(\mathbf{G}_{j})=(0,\delta_{kj}),\quad j=0,1,2,3,\end{array} (15)

where δ\delta is the Kronecker delta.

In the reference triangle K^\widehat{K} of vertices 𝐕^1=(0,0),𝐕^2=(1,0),𝐕^3=(0,1)\widehat{\mathbf{V}}_{1}=(0,0),\widehat{\mathbf{V}}_{2}=(1,0),\widehat{\mathbf{V}}_{3}=(0,1), they appear in

𝒩𝐆^1K^1=13(12+75x+45y261xy108x227y2+228x2y+180xy2+40x316y3),𝒩𝐆^1K^2=13(12+45x+75y261xy27x2108y2+180x2y+228xy216x3+40y3),𝒩𝐆^2K^1=526x28y+140xy+21x2+28y2112x2y112xy2+8x3,𝒩𝐆^2K^2=13(10+57x+51y249xy75x254y2+228x2y+180xy2+16x3+8y3),𝒩𝐆^3K^1=13(10+51x+57y249xy54x275y2+180x2y+228xy2+8x3+16y3),𝒩𝐆^3K^2=528x26y+140xy+28x2+21y2112x2y112xy2+8y3.\begin{array}[]{lll}{\mathcal{N}_{\widehat{\mathbf{G}}_{1}\widehat{K}}}^{1}&=&\frac{1}{3}(-12+75x+45y-261xy-108x^{2}-27y^{2}+228x^{2}y+180xy^{2}+40x^{3}-16y^{3}),\\ {\mathcal{N}_{\widehat{\mathbf{G}}_{1}\widehat{K}}}^{2}&=&\frac{1}{3}(-12+45x+75y-261xy-27x^{2}-108y^{2}+180x^{2}y+228xy^{2}-16x^{3}+40y^{3}),\\ {\mathcal{N}_{\widehat{\mathbf{G}}_{2}\widehat{K}}}^{1}&=&5-26x-28y+140xy+21x^{2}+28y^{2}-112x^{2}y-112xy^{2}+8x^{3},\\ {\mathcal{N}_{\widehat{\mathbf{G}}_{2}\widehat{K}}}^{2}&=&\frac{1}{3}(-10+57x+51y-249xy-75x^{2}-54y^{2}+228x^{2}y+180xy^{2}+16x^{3}+8y^{3}),\\ {\mathcal{N}_{\widehat{\mathbf{G}}_{3}\widehat{K}}}^{1}&=&\frac{1}{3}(-10+51x+57y-249xy-54x^{2}-75y^{2}+180x^{2}y+228xy^{2}+8x^{3}+16y^{3}),\\ {\mathcal{N}_{\widehat{\mathbf{G}}_{3}\widehat{K}}}^{2}&=&5-28x-26y+140xy+28x^{2}+21y^{2}-112x^{2}y-112xy^{2}+8y^{3}.\end{array}

We note that

𝒩𝐆kKi0,KCσ|K|1/2|𝒩𝐆kKi|1,KCσ|K|,k=1,2,3,i=1,2.\|{\mathcal{N}_{\mathbf{G}_{k}K}}^{i}\|_{0,K}\leq C_{\sigma}|K|^{1/2}|{\mathcal{N}_{\mathbf{G}_{k}K}}^{i}|_{1,K}\leq C_{\sigma}|K|,\quad k=1,2,3,\ i=1,2. (16)

4.3 basis for P3(K)P^{3}(K)

Given triangle K𝒯hK\in\mathcal{T}_{h} of vertices 𝐕1,𝐕2,𝐕3\mathbf{V}_{1},\mathbf{V}_{2},\mathbf{V}_{3} and centers 𝐆1,𝐆2,𝐆3\mathbf{G}_{1},\mathbf{G}_{2},\mathbf{G}_{3} of medians, define a set

={𝒮𝐕1K,𝒮𝐕2K,𝒮𝐕3K,𝒩𝐆1K1,𝒩𝐆1K2,𝒩𝐆2K1,𝒩𝐆2K2,𝒩𝐆3K1,𝒩𝐆3K2,1}P3(K).\mathcal{B}=\{\mathcal{S}_{\mathbf{V}_{1}K},\mathcal{S}_{\mathbf{V}_{2}K},\mathcal{S}_{\mathbf{V}_{3}K},{\mathcal{N}_{\mathbf{G}_{1}K}}^{1},{\mathcal{N}_{\mathbf{G}_{1}K}}^{2},{\mathcal{N}_{\mathbf{G}_{2}K}}^{1},{\mathcal{N}_{\mathbf{G}_{2}K}}^{2},{\mathcal{N}_{\mathbf{G}_{3}K}}^{1},{\mathcal{N}_{\mathbf{G}_{3}K}}^{2},1\}\subset P^{3}(K).
Lemma 4.2.

\mathcal{B} is a basis for P3(K)P^{3}(K)

Proof.

For some 10 constants α1,α2,α3,β1,β2,,β6,c\alpha_{1},\alpha_{2},\alpha_{3},\beta_{1},\beta_{2},\cdots,\beta_{6},c, assume that

qh=α1𝒮𝐕1K+α2𝒮𝐕2K+α3𝒮𝐕3K+β1𝒩𝐆1K1+β2𝒩𝐆1K2+β3𝒩𝐆2K1+β4𝒩𝐆2K2+β5𝒩𝐆3K1+β6𝒩𝐆3K2+c=0.q_{h}=\alpha_{1}\mathcal{S}_{\mathbf{V}_{1}K}+\alpha_{2}\mathcal{S}_{\mathbf{V}_{2}K}+\alpha_{3}\mathcal{S}_{\mathbf{V}_{3}K}\\ +\beta_{1}{\mathcal{N}_{\mathbf{G}_{1}K}}^{1}+\beta_{2}{\mathcal{N}_{\mathbf{G}_{1}K}}^{2}+\beta_{3}{\mathcal{N}_{\mathbf{G}_{2}K}}^{1}+\beta_{4}{\mathcal{N}_{\mathbf{G}_{2}K}}^{2}+\beta_{5}{\mathcal{N}_{\mathbf{G}_{3}K}}^{1}+\beta_{6}{\mathcal{N}_{\mathbf{G}_{3}K}}^{2}+c=0. (17)

There exists a quartic function vP4v\in P^{4} such that

v(𝐆1)=1,v(𝐆2)=v(𝐆3)=0,v vanishes on K.v(\mathbf{G}_{1})=1,\quad v(\mathbf{G}_{2})=v(\mathbf{G}_{3})=0,\quad v\mbox{ vanishes on }\partial K. (18)

Then since vxP3v_{x}\in P^{3} vanishes at 𝐕1,𝐕2,𝐕3\mathbf{V}_{1},\mathbf{V}_{2},\mathbf{V}_{3}, we expand the following from (15), (17),(18) and the quadrature rules (6), (9),

0=Kqhvx𝑑x𝑑y=Kqˇhvx𝑑x𝑑y=kqˇh,xv𝑑x𝑑y=|K|β1𝒩G1Kx1(𝐆1)v(𝐆1)g1,0=\int_{K}q_{h}\hskip 1.0ptv_{x}\ dxdy=\int_{K}\check{q}_{h}\hskip 1.0ptv_{x}\ dxdy=-\int_{k}\check{q}_{h,x}\hskip 1.0ptv\ dxdy=-|K|\beta_{1}{\mathcal{N}_{G_{1}K}}_{x}^{1}(\mathbf{G}_{1})v(\mathbf{G}_{1})g_{1}, (19)

where qˇh=qhα1𝒮𝐕1Kα2𝒮𝐕2Kα3𝒮𝐕3K.\check{q}_{h}=q_{h}-\alpha_{1}\mathcal{S}_{\mathbf{V}_{1}K}-\alpha_{2}\mathcal{S}_{\mathbf{V}_{2}K}-\alpha_{3}\mathcal{S}_{\mathbf{V}_{3}K}. Thus, β1=0\beta_{1}=0 and similarly βj=0,j=2,3,,J\beta_{j}=0,j=2,3,\cdots,J.

There also exists a quartic function wP4w\in P^{4} such that

w(𝐕1)𝐕1𝐕2=1,w(𝐕2)𝐕1𝐕2=0,𝐕1𝐕2¯w𝑑=0,w vanishes on 𝐕1𝐕3¯𝐕2𝐕3¯.\nabla w(\mathbf{V}_{1})\cdot\overrightarrow{\mathbf{V}_{1}\mathbf{V}_{2}}=1,\ \nabla w(\mathbf{V}_{2})\cdot\overrightarrow{\mathbf{V}_{1}\mathbf{V}_{2}}=0,\ \int_{\overline{\mathbf{V}_{1}\mathbf{V}_{2}}}w\ d\ell=0,\ w\mbox{ vanishes on }\overline{\mathbf{V}_{1}\mathbf{V}_{3}}\cup\overline{\mathbf{V}_{2}\mathbf{V}_{3}}. (20)

Then with z=w𝐕1𝐕2P3z=\nabla w\cdot\overrightarrow{\mathbf{V}_{1}\mathbf{V}_{2}}\in P^{3}, we get the following from (20) and the quadrature rule (6),

0=K(α1𝒮𝐕1K+α2𝒮𝐕2K+α3𝒮𝐕3K+c)z𝑑x𝑑y=α1|K|z(𝐕1).0=\int_{K}\left(\alpha_{1}\mathcal{S}_{\mathbf{V}_{1}K}+\alpha_{2}\mathcal{S}_{\mathbf{V}_{2}K}+\alpha_{3}\mathcal{S}_{\mathbf{V}_{3}K}+c\right)z\ dxdy=\alpha_{1}|K|z(\mathbf{V}_{1}).

It means α1=0\alpha_{1}=0 and similarly α2=α3=0\alpha_{2}=\alpha_{3}=0. ∎

4.4 decomposition of 𝒫h3(Ω)\mathcal{P}_{h}^{3}(\Omega)

Given triangle KK, let 𝒩h(K)\mathcal{N}_{h}(K) be a space spanned by 6 non-sting functions in P3(K)P^{3}(K), that is,

𝒩h(K)=<𝒩𝐆1K1,𝒩𝐆1K2,𝒩𝐆2K1,𝒩𝐆2K2,𝒩𝐆3K1,𝒩𝐆3K2>.\mathcal{N}_{h}(K)=<{\mathcal{N}_{\mathbf{G}_{1}K}}^{1},{\mathcal{N}_{\mathbf{G}_{1}K}}^{2},{\mathcal{N}_{\mathbf{G}_{2}K}}^{1},{\mathcal{N}_{\mathbf{G}_{2}K}}^{2},{\mathcal{N}_{\mathbf{G}_{3}K}}^{1},{\mathcal{N}_{\mathbf{G}_{3}K}}^{2}>.

Then by Lemma 4.2, we can decompose P3(K)P^{3}(K) into

P3(K)=𝒩h(K)<𝒮𝐕1K,𝒮𝐕2K,𝒮𝐕3K,1>.P^{3}(K)=\mathcal{N}_{h}(K)\bigoplus<\mathcal{S}_{\mathbf{V}_{1}K},\mathcal{S}_{\mathbf{V}_{2}K},\mathcal{S}_{\mathbf{V}_{3}K},1>. (21)

Given vertex 𝐕\mathbf{V}, let 𝒮h(𝐕)\mathcal{S}_{h}(\mathbf{V}) be a space spanned by all sting functions of 𝐕\mathbf{V}, that is,

𝒮h(𝐕)=<𝒮𝐕K1,𝒮𝐕K2,,𝒮𝐕KJ>,\mathcal{S}_{h}(\mathbf{V})=<\mathcal{S}_{\mathbf{V}K_{1}},\mathcal{S}_{\mathbf{V}K_{2}},\cdots,\mathcal{S}_{\mathbf{V}K_{J}}>,

where K1,K2,,KJK_{1},K_{2},\cdots,K_{J} are all triangles in 𝒯h\mathcal{T}_{h} sharing 𝐕\mathbf{V}.

We call qh𝐕𝒮h(𝐕)q_{h}^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}) a sting pressure of 𝐕\mathbf{V} and qhK𝒩h(K)q_{h}^{K}\in\mathcal{N}_{h}(K) a non-sting pressure of KK. Examples of their supports are depicted in Figure 2.

Refer to caption
(a) The support of qh𝐕q_{h}^{\mathbf{V}}, a sting pressure of 𝐕\mathbf{V}
qh𝐕q_{h}^{\mathbf{V}} is determined by qh𝐕|Kj(𝐕),j=1,2,3,4,5.q_{h}^{\mathbf{V}}\big{|}_{K_{j}}(\mathbf{V}),\ j=1,2,3,4,5.

Refer to caption

(b) The support of qhKq_{h}^{K}, a non-sting pressure of KK
qhKq_{h}^{K} is determined by qhK(𝐆j),j=1,2,3.\nabla q_{h}^{K}(\mathbf{G}_{j}),\ j=1,2,3.
Figure 2: characteristics of qh𝐕𝒮h(𝐕)q_{h}^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}) and qhK𝒩h(K)q_{h}^{K}\in\mathcal{N}_{h}(K)

Then from (21), we can decompose 𝒫h3(Ω)\mathcal{P}_{h}^{3}(\Omega) into

𝒫h3(Ω)=(K𝒯h𝒩h(K))(𝐕𝒱h𝒮h(𝐕))𝒫h0(Ω).\mathcal{P}_{h}^{3}(\Omega)=\left(\bigoplus_{K\in\mathcal{T}_{h}}\mathcal{N}_{h}(K)\right)\bigoplus\left(\bigoplus_{\mathbf{V}\in\mathcal{V}_{h}}\mathcal{S}_{h}(\mathbf{V})\right)\bigoplus\mathcal{P}_{h}^{0}(\Omega). (22)

5 Successive method for pressure

In this section, we will find the following discrete pressures:

phK𝒩h(K),ph𝐕𝒮h(𝐕),ph𝒞𝒫h0(Ω),p_{h}^{K}\in\mathcal{N}_{h}(K),\quad p_{h}^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}),\quad p_{h}^{\mathpzc{C}}\in\mathcal{P}_{h}^{0}(\Omega),

in a successive way consisting of the following 5 steps.

Step 1.

Calculate a non-sting pressure phK𝒩h(K)p_{h}^{K}\in\mathcal{N}_{h}(K) using 𝐮h,𝐟\mathbf{u}_{h},\mathbf{f} for each triangle K𝒯hK\in\mathcal{T}_{h} and superpose them to form

ph𝒩=phK.p_{h}^{\mathcal{N}}=\sum p_{h}^{K}.
Step 2.

Calculate a sting pressure ph𝐕r𝒮h(𝐕r)p_{h}^{\mathbf{V}_{r}}\in\mathcal{S}_{h}(\mathbf{V}_{r}) using ph𝒩,𝐮h,𝐟p_{h}^{\mathcal{N}},\mathbf{u}_{h},\mathbf{f} for each regular vertex 𝐕r\mathbf{V}_{r} and superpose them to form

ph˙=ph𝒩+ph𝐕r.\dot{p_{h}}=p_{h}^{\mathcal{N}}+\sum p_{h}^{\mathbf{V}_{r}}.
Step 3.

Calculate a sting pressure ph𝐕s𝒮h(𝐕s)p_{h}^{\mathbf{V}_{s}}\in\mathcal{S}_{h}(\mathbf{V}_{s}) using ph˙,𝐮h,𝐟\dot{p_{h}},\mathbf{u}_{h},\mathbf{f} for each nearly singular vertex 𝐕s\mathbf{V}_{s} except corners and superpose them to form

ph¨=ph˙+ph𝐕s.\ddot{p_{h}}=\dot{p_{h}}+\sum p_{h}^{\mathbf{V}_{s}}.
Step 4.

Calculate a sting pressure ph𝐕c𝒮h(𝐕c)p_{h}^{\mathbf{V}_{c}}\in\mathcal{S}_{h}(\mathbf{V}_{c}) using ph¨,𝐮h,𝐟\ddot{p_{h}},\mathbf{u}_{h},\mathbf{f} for each nearly singular corner 𝐕c\mathbf{V}_{c} and superpose them to form

ph˙˙˙=ph¨+ph𝐕c.\dddot{p_{h}}=\ddot{p_{h}}+\sum p_{h}^{\mathbf{V}_{c}}.
Step 5.

Calculate a piecewise constant pressure ph𝒞𝒫h0(Ω)p_{h}^{\mathpzc{C}}\in\mathcal{P}_{h}^{0}(\Omega) using ph˙˙˙,𝐮h,𝐟\dddot{p_{h}},\mathbf{u}_{h},\mathbf{f}.

The detail of Step ii above will be given in Section 5.ii below, i=1,2,3,4,5i=1,2,3,4,5.

If we define ph𝒫h3(Ω)p_{h}\in\mathcal{P}_{h}^{3}(\Omega) as

ph=ph˙˙˙+ph𝒞,p_{h}=\dddot{p_{h}}+p_{h}^{\mathpzc{C}}, (23)

it shows the optimal order of convergence as in the following theorem.

Theorem 5.1.

Let (𝐮,p)[H01(Ω)]2×L02(Ω)(\mathbf{u},p)\in[H_{0}^{1}(\Omega)]^{2}\times L_{0}^{2}(\Omega) satisfy (1). If (𝐮,p)[H5(Ω)]2×H4(Ω)(\mathbf{u},p)\in[H^{5}(\Omega)]^{2}\times H^{4}(\Omega), then

pph0Ch4(|𝐮|5+|p|4).\|p-p_{h}\|_{0}\leq Ch^{4}(|\mathbf{u}|_{5}+|p|_{4}). (24)

5.0 decomposition of Πhp\Pi_{h}p

Let Πhp𝒫h3(Ω)\Pi_{h}p\in\mathcal{P}_{h}^{3}(\Omega) be a Hermite interpolation of pp in the theorem 5.1 such that

Πhp(𝐕)=p(𝐕),Πhp(𝐕)=p(𝐕),Πhp(𝐆)=p(𝐆),\nabla\Pi_{h}p(\mathbf{V})=\nabla p(\mathbf{V}),\quad\Pi_{h}p(\mathbf{V})=p(\mathbf{V}),\quad\Pi_{h}p(\mathbf{G})=p(\mathbf{G}), (25)

at all vertices 𝐕\mathbf{V} and gravity centers 𝐆\mathbf{G} of triangles in 𝒯h\mathcal{T}_{h}. Then from (22), we can split Πhp𝒫h3(Ω)\Pi_{h}p\in\mathcal{P}_{h}^{3}(\Omega) into

Πhp=K𝒯h(Πhp)K+𝐕𝒱h(Πhp)𝐕+(Πhp)𝒞,\Pi_{h}p=\sum_{K\in\mathcal{T}_{h}}(\Pi_{h}p)^{K}+\sum_{\mathbf{V}\in\mathcal{V}_{h}}(\Pi_{h}p)^{\mathbf{V}}+(\Pi_{h}p)^{\mathpzc{C}}, (26)

where

(Πhp)K𝒩h(K),(Πhp)𝐕𝒮h(𝐕),(Πhp)𝒞𝒫h0(Ω).(\Pi_{h}p)^{K}\in\mathcal{N}_{h}(K),\quad(\Pi_{h}p)^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}),\quad(\Pi_{h}p)^{\mathpzc{C}}\in\mathcal{P}_{h}^{0}(\Omega). (27)

We note that Πhp\Pi_{h}p satisfies that

(Πhp,div𝐯)=(𝐟,𝐯)(𝐮,𝐯)(pΠhp,div𝐯) for all 𝐯[H01(Ω)]2.(\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{v})=(\mathbf{f},\mathbf{v})-(\nabla\mathbf{u},\nabla\mathbf{v})-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{v})\quad\mbox{ for all }\mathbf{v}\in\left[H_{0}^{1}(\Omega)\right]^{2}. (28)

We assume the following to exclude pathological meshes.

Assumption 5.1.

𝒯h\mathcal{T}_{h} has no two adjacent triangles whose union has two corner points of Ω\partial\Omega.

5.1 Step 1. non-sting pressure for triangle

Let’s fix a triangle KK of vertices 𝐕1,𝐕2,𝐕3\mathbf{V}_{1},\mathbf{V}_{2},\mathbf{V}_{3} and centers 𝐆1,𝐆2,𝐆3\mathbf{G}_{1},\mathbf{G}_{2},\mathbf{G}_{3} of medians. There exists a unique function v𝒫h4(Ω)H01(Ω)v\in\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega) such that

v(𝐆1)=1,v(𝐆2)=v(𝐆3)=0,v vanishes on ΩK.v(\mathbf{G}_{1})=1,\quad v(\mathbf{G}_{2})=v(\mathbf{G}_{3})=0,\quad v\mbox{ vanishes on }\Omega\setminus K.

Then by same argument as in (19) with a test function 𝐯h=(v,0)\mathbf{v}_{h}=(v,0), all 6 non-sting basis functions of 𝒩h(K)\mathcal{N}_{h}(K) satisfy

(𝒩GkKi,div𝐯h)={|K|𝒩G1Kx1(𝐆1)v(𝐆1)g1=|K|g1, if k=1 and i=1,0, otherwise.({\mathcal{N}_{G_{k}K}}^{i},\mathrm{div}\hskip 1.42262pt{\mathbf{v}_{h}})=\left\{\begin{array}[]{l}-|K|{\mathcal{N}_{G_{1}K}}_{x}^{1}(\mathbf{G}_{1})v(\mathbf{G}_{1})g_{1}=-|K|g_{1},\ \mbox{ if }k=1\mbox{ and }i=1,\\ 0,\quad\mbox{ otherwise}.\end{array}\right. (29)

Generally, for each k=1,2,3k=1,2,3, choose vk𝒫h4(Ω)H01(Ω)v_{k}\in\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega) such that

vk(𝐆j)=δkj,j=1,2,3,vk vanishes on ΩK,v_{k}(\mathbf{G}_{j})=\delta_{kj},j=1,2,3,\quad v_{k}\mbox{ vanishes on }\Omega\setminus K, (30)

and define

𝐯h,k1=(vk,0),𝐯h,k2=(0,vk).{\mathbf{v}_{h,k}}^{1}=(v_{k},0),\quad{\mathbf{v}_{h,k}}^{2}=(0,v_{k}). (31)

Then by same argument as in (29) with (15), (30), (31), all 6 basis functions of 𝒩h(K)\mathcal{N}_{h}(K) satisfy

(𝒩GkKi,div𝐯h,lj)=|K|gkδklδij,k,l=1,2,3,i,j=1,2.({\mathcal{N}_{G_{k}K}}^{i},\mathrm{div}\hskip 1.42262pt{\mathbf{v}_{h,l}}^{j})=-|K|g_{k}\delta_{kl}\delta_{ij},\quad k,l=1,2,3,\ i,j=1,2. (32)

Now with the aid of (32), we can find a unique phK𝒩h(K)p_{h}^{K}\in\mathcal{N}_{h}(K) such that

(phK,div𝐯h)=(𝐟,𝐯h)(𝐮h,𝐯h) for alll 𝐯h{𝐯h,ki:k=1,2,3,i=1,2}.(p_{h}^{K},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})=(\mathbf{f},\mathbf{v}_{h})-(\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})\quad\mbox{ for alll }\mathbf{v}_{h}\in\{{\mathbf{v}_{h,k}}^{i}\ :\ k=1,2,3,\ i=1,2\}. (33)

Actually, if we set phKp_{h}^{K} for 6 unknown constants α11,α12,α21,α22,α31,α32\alpha_{11},\alpha_{12},\alpha_{21},\alpha_{22},\alpha_{31},\alpha_{32} as

phK=α11𝒩𝐆1K1+α12𝒩𝐆1K2+α21𝒩𝐆2K1+α22𝒩𝐆2K2+α31𝒩𝐆3K1+α32𝒩𝐆3K2,p_{h}^{K}=\alpha_{11}{\mathcal{N}_{\mathbf{G}_{1}K}}^{1}+\alpha_{12}{\mathcal{N}_{\mathbf{G}_{1}K}}^{2}+\alpha_{21}{\mathcal{N}_{\mathbf{G}_{2}K}}^{1}+\alpha_{22}{\mathcal{N}_{\mathbf{G}_{2}K}}^{2}+\alpha_{31}{\mathcal{N}_{\mathbf{G}_{3}K}}^{1}+\alpha_{32}{\mathcal{N}_{\mathbf{G}_{3}K}}^{2},

we deduce that

αk,i=|K|1gk1((𝐟,𝐯h,ki)(𝐮h,𝐯h,ki)),k=1,2,3,i=1,2.\alpha_{k,i}=-|K|^{-1}g_{k}^{-1}\left((\mathbf{f},{\mathbf{v}_{h,k}}^{i})-(\nabla\mathbf{u}_{h},\nabla{\mathbf{v}_{h,k}}^{i})\right),\quad k=1,2,3,\ i=1,2. (34)
Lemma 5.2.
(Πhp)KphK0,KCσ(|𝐮𝐮h|1,K+pΠhp0,K).\|(\Pi_{h}p)^{K}-p_{h}^{K}\|_{0,K}\leq C_{\sigma}(|\mathbf{u}-\mathbf{u}_{h}|_{1,K}+\|p-\Pi_{h}p\|_{0,K}). (35)
Proof.

Let V={𝐯h,ki:k=1,2,3,i=1,2}V=\{{\mathbf{v}_{h,k}}^{i}\ :\ k=1,2,3,\ i=1,2\}. We can rewrite (28) for 𝐯hV\mathbf{v}_{h}\in V into

((Πhp)K,div𝐯h)=(𝐟,𝐯h)(𝐮,𝐯h)(pΠhp,div𝐯h),\left((\Pi_{h}p)^{K},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}\right)=(\mathbf{f},\mathbf{v}_{h})-(\nabla\mathbf{u},\nabla\mathbf{v}_{h})-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}), (36)

since (1,div𝐯h)=0(1,\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})=0 from (30), (31) and by quadrature rule in (6),

(𝒮𝐕kK,div𝐯h)=|K|div𝐯h(𝐕k)=0,k=1,2,3.(\mathcal{S}_{\mathbf{V}_{k}K},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})=|K|\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}(\mathbf{V}_{k})=0,\ k=1,2,3.

If we denote the error by ehK=(Πhp)KphKe_{h}^{K}=(\Pi_{h}p)^{K}-p_{h}^{K}, then from (33), (36), it satisfies

(ehK,div𝐯h)=(𝐮𝐮h,𝐯h)(pΠhp,div𝐯h) for all 𝐯hV.(e_{h}^{K},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})=-(\nabla\mathbf{u}-\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})\quad\mbox{ for all }\mathbf{v}_{h}\in V. (37)

Represent ehKe_{h}^{K} with 6 constants e11,e12,e21,e22,e31,e32e_{11},e_{12},e_{21},e_{22},e_{31},e_{32} as

ehK=e11𝒩𝐆1K1+e12𝒩𝐆1K2+e21𝒩𝐆2K1+e22𝒩𝐆2K2+e31𝒩𝐆3K1+e32𝒩𝐆3K2.e_{h}^{K}=e_{11}{\mathcal{N}_{\mathbf{G}_{1}K}}^{1}+e_{12}{\mathcal{N}_{\mathbf{G}_{1}K}}^{2}+e_{21}{\mathcal{N}_{\mathbf{G}_{2}K}}^{1}+e_{22}{\mathcal{N}_{\mathbf{G}_{2}K}}^{2}+e_{31}{\mathcal{N}_{\mathbf{G}_{3}K}}^{1}+e_{32}{\mathcal{N}_{\mathbf{G}_{3}K}}^{2}. (38)

Then by same argument inducing (34), we have

eki=|K|1gk1((𝐮𝐮h,𝐯h,ki)+(pΠhp,div𝐯h,ki)),k=1,2,3,i=1,2.e_{ki}=|K|^{-1}g_{k}^{-1}\left((\nabla\mathbf{u}-\nabla\mathbf{u}_{h},\nabla{\mathbf{v}_{h,k}}^{i})+(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt{\mathbf{v}_{h,k}}^{i})\right),\quad k=1,2,3,\ i=1,2. (39)

We note that |vk|1Cσ,k=1,2,3|v_{k}|_{1}\leq C_{\sigma},\ k=1,2,3 from (30). Thus, (35) comes from (16), (38), (39). ∎

Superpose all the calculated non-sting pressures phK,K𝒯hp_{h}^{K},K\in\mathcal{T}_{h} and define

ph𝒩=K𝒯hphK,(Πhp)𝒩=K𝒯h(Πhp)K.p_{h}^{\mathcal{N}}=\sum_{K\in\mathcal{T}_{h}}p_{h}^{K},\quad(\Pi_{h}p)^{\mathcal{N}}=\sum_{K\in\mathcal{T}_{h}}(\Pi_{h}p)^{K}. (40)

5.2 Step 2. sting pressure for regular vertex

5.2.1 test functions in two triangles

Let 𝐕\mathbf{V} be a vertex and K1,K2K_{1},K_{2} two back-to-back triangles sharing 𝐕\mathbf{V} as in Figure 3-(a). Denote by 𝐖0,𝐖1,𝐖2,𝝉\mathbf{W}_{0},\mathbf{W}_{1},\mathbf{W}_{2},\boldsymbol{\tau}, other 3 vertices and a unit tangent vector such that

𝐕𝐖0¯=K1K2,𝐖jKj{𝐕,𝐖0},j=1,2,𝝉=𝐕𝐖0|𝐕𝐖0¯|.\overline{\mathbf{V}\mathbf{W}_{0}}=K_{1}\cap K_{2},\quad\mathbf{W}_{j}\in K_{j}\setminus\{\mathbf{V},\mathbf{W}_{0}\},j=1,2,\quad\boldsymbol{\tau}=\frac{\overrightarrow{\mathbf{V}\mathbf{W}_{0}}}{|\overline{\mathbf{V}\mathbf{W}_{0}}|}.

There exists a function w𝒫h4(Ω)H01(Ω)w\in\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega) such that

w𝝉(𝐕)=1,w𝝉(𝐖0)=0,𝐕𝐖0¯w𝑑=0, the support of w is K1K2.\frac{\partial w}{\partial\boldsymbol{\tau}}(\mathbf{V})=1,\ \frac{\partial w}{\partial\boldsymbol{\tau}}(\mathbf{W}_{0})=0,\ \int_{\overline{\mathbf{V}\mathbf{W}_{0}}}w\ d\ell=0,\ \mbox{ the support of }w\mbox{ is }K_{1}\cup K_{2}. (41)

Assume K1,K2K_{1},K_{2} are counterclockwisely numbered with respect to 𝐕\mathbf{V}. Then by simple calculation, we have

w|K1(𝐕)=|𝐕𝐖0¯|2|K1|𝐕𝐖1,w|K2(𝐕)=|𝐕𝐖0¯|2|K2|𝐕𝐖2,\nabla w\big{|}_{K_{1}}(\mathbf{V})=\frac{|\overline{\mathbf{V}\mathbf{W}_{0}}|}{2|K_{1}|}\ {\overrightarrow{\mathbf{V}\mathbf{W}_{1}}}^{\perp},\quad\nabla w\big{|}_{K_{2}}(\mathbf{V})=-\frac{|\overline{\mathbf{V}\mathbf{W}_{0}}|}{2|K_{2}|}\ {\overrightarrow{\mathbf{V}\mathbf{W}_{2}}}^{\perp}, (42)

where ()(\star)^{\perp} denotes the 9090^{\circ} counterclockwise rotation of vector \star. Thus if we set a test function 𝐰h𝝃=w𝝃=(ξ1w,ξ2w)\mathbf{w}_{h}^{\boldsymbol{\xi}}=w\boldsymbol{\xi}=(\xi_{1}w,\xi_{2}w) for a vector 𝝃=(ξ1,ξ2)\boldsymbol{\xi}=(\xi_{1},\xi_{2}), it satisfies

div𝐰h𝝃|K1(𝐕)=|𝐕𝐖0¯|2|K1|𝐕𝐖1𝝃,div𝐰h𝝃|K2(𝐕)=|𝐕𝐖0¯|2|K2|𝐕𝐖2𝝃,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\xi}}\big{|}_{K_{1}}(\mathbf{V})=\frac{|\overline{\mathbf{V}\mathbf{W}_{0}}|}{2|K_{1}|}\ {\overrightarrow{\mathbf{V}\mathbf{W}_{1}}}^{\perp}\cdot\boldsymbol{\xi},\quad\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\xi}}\big{|}_{K_{2}}(\mathbf{V})=-\frac{|\overline{\mathbf{V}\mathbf{W}_{0}}|}{2|K_{2}|}\ {\overrightarrow{\mathbf{V}\mathbf{W}_{2}}}^{\perp}\cdot\boldsymbol{\xi}, (43)

and div𝐰h𝝃\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\xi}} vanishes at all other vertices.

Let qh𝒫h3(Ω)q_{h}\in\mathcal{P}_{h}^{3}(\Omega) be represented with some constants αj,βj,j=1,2,3\alpha_{j},\beta_{j},j=1,2,3 as

qh=α1𝒮𝐕K1+α2𝒮𝐖0K1+α3𝒮𝐖1K1+β1𝒮𝐕K2+β2𝒮𝐖0K2+β3𝒮𝐖2K2,q_{h}=\alpha_{1}\mathcal{S}_{\mathbf{V}K_{1}}+\alpha_{2}\mathcal{S}_{\mathbf{W}_{0}K_{1}}+\alpha_{3}\mathcal{S}_{\mathbf{W}_{1}K_{1}}+\beta_{1}\mathcal{S}_{\mathbf{V}K_{2}}+\beta_{2}\mathcal{S}_{\mathbf{W}_{0}K_{2}}+\beta_{3}\mathcal{S}_{\mathbf{W}_{2}K_{2}},

that is, it is a sum of sting pressures on K1K2K_{1}\cup K_{2}. Then, by quadrature rule of sting functions in (6), we expand from (43) that

(qh𝐕,div𝐰h𝝉)=|K1|div𝐰h𝝉|K1(𝐕)+|K2|div𝐰h𝝉|K2(𝐕)=12(α1𝐕𝐖1β1𝐕𝐖2)𝐕𝐖0,(qh𝐕,div𝐰h𝝉)=|K1|div𝐰h𝝉|K1(𝐕)+|K2|div𝐰h𝝉|K2(𝐕)=12(α1𝐕𝐖1β1𝐕𝐖2)𝐕𝐖0.\begin{array}[]{lll}(q_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}})&=&|K_{1}|\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}}\big{|}_{K_{1}}(\mathbf{V})+|K_{2}|\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}}\big{|}_{K_{2}}(\mathbf{V})=\displaystyle\frac{1}{2}\left(\alpha_{1}{\overrightarrow{\mathbf{V}\mathbf{W}_{1}}}^{\perp}-\beta_{1}{\overrightarrow{\mathbf{V}\mathbf{W}_{2}}}^{\perp}\right)\cdot\overrightarrow{\mathbf{V}\mathbf{W}_{0}},\\ (q_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}^{\perp}})&=&|K_{1}|\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}^{\perp}}\big{|}_{K_{1}}(\mathbf{V})+|K_{2}|\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}^{\perp}}\big{|}_{K_{2}}(\mathbf{V})=\displaystyle\frac{1}{2}\left(\alpha_{1}{\overrightarrow{\mathbf{V}\mathbf{W}_{1}}}^{\perp}-\beta_{1}{\overrightarrow{\mathbf{V}\mathbf{W}_{2}}}^{\perp}\right)\cdot{\overrightarrow{\mathbf{V}\mathbf{W}_{0}}}^{\perp}.\end{array}

It can be written in simpler form:

(qh𝐕,div𝐰h𝝉)=01sinθ12α1+02sinθ22β1=|K1|α1+|K2|β1,(qh𝐕,div𝐰h𝝉)=01cosθ12α102cosθ22β1,\begin{array}[]{lll}(q_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}})&=&\displaystyle\frac{\ell_{0}\ell_{1}\sin\theta_{1}}{2}\alpha_{1}+\frac{\ell_{0}\ell_{2}\sin\theta_{2}}{2}\beta_{1}=|K_{1}|\alpha_{1}+|K_{2}|\beta_{1},\\ (q_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}^{\perp}})&=&\displaystyle\frac{\ell_{0}\ell_{1}\cos\theta_{1}}{2}\alpha_{1}-\frac{\ell_{0}\ell_{2}\cos\theta_{2}}{2}\beta_{1},\end{array} (44)

where θ1,θ2\theta_{1},\theta_{2} are angles of K1,K2K_{1},K_{2} at 𝐕\mathbf{V}, respectively, and j=|𝐕𝐖j¯|,j=0,1,2\ell_{j}=|\overline{\mathbf{V}\mathbf{W}_{j}}|,j=0,1,2 as in Figure 3-(a).

Refer to caption
(a) The union of two adjacent triangles sharing 𝐕\mathbf{V}
Refer to caption
(b) 𝒦(𝐕)\mathcal{K}(\mathbf{V}), the union of all triangles sharing 𝐕\mathbf{V}
Figure 3: Unions of triangles

5.2.2 least square solution

Let’s fix an interior vertex 𝐕\mathbf{V} and K1,K2,,KJK_{1},K_{2},\cdots,K_{J} all triangles in 𝒯h\mathcal{T}_{h} sharing 𝐕\mathbf{V} as in Figure 3-(b). We assume the numbering is counterclockwise and use the index modulo JJ.

For each j=1,2,Jj=1,2,\cdots J, name a vertex 𝐕j\mathbf{V}_{j} and a unit tangent vector 𝝉j\boldsymbol{\tau}_{j} so that

𝐕𝐕j¯=KjKj+1,𝝉j=𝐕𝐕j|𝐕𝐕j¯|.\overline{\mathbf{V}\mathbf{V}_{j}}=K_{j}\cap K_{j+1},\quad\boldsymbol{\tau}_{j}=\frac{\overrightarrow{\mathbf{V}\mathbf{V}_{j}}}{|\overline{\mathbf{V}\mathbf{V}_{j}}|}. (45)

There exists a function wj𝒫h4(Ω)H01(Ω)w_{j}\in\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega) such that

wj𝝉j(𝐕)=1,wj𝝉j(𝐕j)=0,𝐕𝐕j¯wj𝑑=0, the support of wj is KjKj+1,\frac{\partial w_{j}}{\partial\boldsymbol{\tau}_{j}}(\mathbf{V})=1,\ \frac{\partial w_{j}}{\partial\boldsymbol{\tau}_{j}}(\mathbf{V}_{j})=0,\ \int_{\overline{\mathbf{V}\mathbf{V}_{j}}}w_{j}\ d\ell=0,\ \mbox{ the support of }w_{j}\mbox{ is }K_{j}\cup K_{j+1}, (46)

for each j=1,2,Jj=1,2,\cdots J. For the uniqueness of wjw_{j}, we add the following condition:

wj(𝐆)=0,wj(𝐆)=(0,0) at each gravity center 𝐆 of Kj,Kj+1.w_{j}(\mathbf{G})=0,\ \nabla w_{j}(\mathbf{G})=(0,0)\mbox{ at each gravity center }\mathbf{G}\mbox{ of }K_{j},K_{j+1}. (47)

Then we prepare 2J2J test functions in [𝒫h4(Ω)H01(Ω)]2[\mathcal{P}_{h}^{4}(\Omega)\cap H_{0}^{1}(\Omega)]^{2} as

𝐰h𝝉j=wj𝝉j,𝐰h𝝉j=wj𝝉j,j=1,2,,J.\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}=w_{j}\boldsymbol{\tau}_{j},\quad\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}=w_{j}\boldsymbol{\tau}_{j}^{\perp},\quad j=1,2,\cdots,J. (48)

Let ph𝐕𝒮h(𝐕)p_{h}^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}), a sum of sting pressures of 𝐕\mathbf{V}, be represented as

ph𝐕=α1𝒮𝐕K1+α2𝒮𝐕K2++αJ𝒮𝐕KJ,p_{h}^{\mathbf{V}}=\alpha_{1}\mathcal{S}_{\mathbf{V}K_{1}}+\alpha_{2}\mathcal{S}_{\mathbf{V}K_{2}}+\cdots+\alpha_{J}\mathcal{S}_{\mathbf{V}K_{J}}, (49)

with JJ unknown constants α1,α2,,αJ\alpha_{1},\alpha_{2},\cdots,\alpha_{J}. Consider the following system of 2J2J equations:

(ph𝐕,div𝐰h𝝉j)=(𝐟,𝐰h𝝉j)(𝐮h,𝐰h𝝉j)(ph𝒩,div𝐰h𝝉j),\displaystyle(p_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}), (50a)
(ph𝐕,div𝐰h𝝉j)=(𝐟,𝐰h𝝉j)(𝐮h,𝐰h𝝉j)(ph𝒩,div𝐰h𝝉j),\displaystyle(p_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}), (50b)

for j=1,2,,Jj=1,2,\cdots,J.

Lemma 5.3.

If 𝐕\mathbf{V} is a regular vertex, then the smallest singular value ss of the system in (50) satisfies

Cσ(|K1|+|K2|++|KJ|)sC_{\sigma}(|K_{1}|+|K_{2}|+\cdots+|K_{J}|)\leq s (51)
Proof.

Let θj\theta_{j} be an angle of KjK_{j} at 𝐕\mathbf{V} and j=|𝐕𝐕j¯|,j=1,2,,J\ell_{j}=|\overline{\mathbf{V}\mathbf{V}_{j}}|,j=1,2,\cdots,J as in Figure 3-(b). Then from (44), (46), (49), we can rewrite (50) as, for j=1,2,,Jj=1,2,\cdots,J,

j1jsinθj2αj\displaystyle\displaystyle\frac{\ell_{j-1}\ell_{j}\sin\theta_{j}}{2}\alpha_{j} +jj+1sinθj+12αj+1=(𝐟,𝐰h𝝉j)(𝐮h,𝐰h𝝉j)(ph𝒩,div𝐰h𝝉j),\displaystyle+\frac{\ell_{j}\ell_{j+1}\sin\theta_{j+1}}{2}\alpha_{j+1}=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}), (52a)
j1jcosθj2αj\displaystyle\displaystyle\frac{\ell_{j-1}\ell_{j}\cos\theta_{j}}{2}\alpha_{j} jj+1cosθj+12αj+1=(𝐟,𝐰h𝝉j)(𝐮h,𝐰h𝝉j)(ph𝒩,div𝐰h𝝉j).\displaystyle-\frac{\ell_{j}\ell_{j+1}\cos\theta_{j+1}}{2}\alpha_{j+1}=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}). (52b)

Denote βj=j1jαj/2\beta_{j}=\ell_{j-1}\ell_{j}\alpha_{j}/2 and the right hand side of (52a), (52b) by aj,bja_{j},b_{j} j=1,2,,Jj=1,2,\cdots,J, respectively. Then we rewrite (52) into

sinθjβj\displaystyle\sin\theta_{j}\beta_{j} +sinθj+1βj+1=aj,j=1,2,,J,\displaystyle+\sin\theta_{j+1}\beta_{j+1}=a_{j},\quad j=1,2,\cdots,J, (53a)
cosθjβj\displaystyle\cos\theta_{j}\beta_{j} cosθj+1βj+1=bj,j=1,2,,J.\displaystyle-\cos\theta_{j+1}\beta_{j+1}=b_{j},\quad j=1,2,\cdots,J. (53b)

If 𝐕\mathbf{V} is a regular vertex, we can assume without loss of generality,

Cσ|sin(θ1+θ2)|,C_{\sigma}\leq|\sin(\theta_{1}+\theta_{2})|, (54)

which tells the bound of the determinant of two equations in (53) for j=1j=1.

Let AJ×JA\in\mathbb{R}^{J\times J} be the matrix of the following subsystem of JJ equations:

sinθjβj\displaystyle\sin\theta_{j}\beta_{j} +sinθj+1βj+1=aj,j=1,2,,J1,\displaystyle+\sin\theta_{j+1}\beta_{j+1}=a_{j},\quad j=1,2,\cdots,J-1, (55a)
cosθ1β1\displaystyle\cos\theta_{1}\beta_{1} cosθ2β2=b1.\displaystyle-\cos\theta_{2}\beta_{2}=b_{1}. (55b)

Then from (54) and Cσsinθj,j=1,2,,JC_{\sigma}\leq\sin\theta_{j},j=1,2,\cdots,J, we deduce

|A1|2Cσ.{\left|\kern-1.07639pt\left|\kern-1.07639pt\left|A^{-1}\right|\kern-1.07639pt\right|\kern-1.07639pt\right|}_{2}\leq C_{\sigma}.

It completes the proof since the smallest singular value of AA is the reciprocal of |A1|2{\left|\kern-1.07639pt\left|\kern-1.07639pt\left|A^{-1}\right|\kern-1.07639pt\right|\kern-1.07639pt\right|}_{2} and the smallest singular value of (53) is greater than that of its subsystem (55). ∎

We note that the sting component (Πhp)𝐕𝒮h(𝐕)(\Pi_{h}p)^{\mathbf{V}}\in\mathcal{S}_{h}(\mathbf{V}) of Πhp\Pi_{h}p in (26) and wjw_{j} in (46) satisfies

(Πhp,wj,x)=((Πhp)𝒩,wj,x)+((Πhp)𝐕,wj,x),j=1,2,,J,(\Pi_{h}p,w_{j,x})=\left((\Pi_{h}p)^{\mathcal{N}},w_{j,x}\right)+\left((\Pi_{h}p)^{\mathbf{V}},w_{j,x}\right),\quad j=1,2,\cdots,J, (56)

since (1,wj,x)Kj=(1,wj,x)Kj+1=0(1,w_{j,x})_{K_{j}}=(1,w_{j,x})_{K_{j+1}}=0 and wj,xw_{j,x} vanishes at all vertices except 𝐕\mathbf{V}.

Let ph𝐕p_{h}^{\mathbf{V}} be the least square solution of the system (50). Then the error (Πhp)𝐕ph𝐕(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}} is estimated in the following lemma.

Lemma 5.4.

Let 𝐕\mathbf{V} be a regular vertex. Then, the least square solution ph𝐕p_{h}^{\mathbf{V}} of (50) satisfies

(Πhp)𝐕ph𝐕0,𝒦(𝐕)Cσ(|𝐮𝐮h|1,𝒦(𝐕)+pΠhp0,𝒦(𝐕)).\|(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}\|_{0,\mathcal{K}(\mathbf{V})}\leq C_{\sigma}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,\mathcal{K}(\mathbf{V})}+\|p-\Pi_{h}p\|_{0,\mathcal{K}(\mathbf{V})}\right). (57)
Proof.

From (28), (56) and the definition of 𝐰h𝝉j,𝐰h𝝉j\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}} in (46), (48), we have

((Πhp)𝐕,div𝐰h𝝉j)=(𝐟,𝐰h𝝉j)(𝐮,𝐰h𝝉j)((Πhp)𝒩,div𝐰h𝝉j)(pΠhp,div𝐰h𝝉j),((Πhp)𝐕,div𝐰h𝝉j)=(𝐟,𝐰h𝝉j)(𝐮,𝐰h𝝉j)((Πhp)𝒩,div𝐰h𝝉j)(pΠhp,div𝐰h𝝉j),\begin{array}[]{l}\left((\Pi_{h}p)^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}\right)=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(\nabla\mathbf{u},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-\left((\Pi_{h}p)^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}\right)-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}),\\ \left((\Pi_{h}p)^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}\right)=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-(\nabla\mathbf{u},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})-\left((\Pi_{h}p)^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}\right)-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}),\end{array}

(58)

for j=1,2,,Jj=1,2,\cdots,J.

If we denote the error by eh𝐕=(Πhp)𝐕ph𝐕e_{h}^{\mathbf{V}}=(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}, then from (50),(58), eh𝐕e_{h}^{\mathbf{V}} is the least square solution of the following system of 2J2J equations:

(eh𝐕,div𝐰h𝝉j)=((𝐮𝐮h),𝐰h𝝉j)((Πhp)𝒩ph𝒩,div𝐰h𝝉j)(pΠhp,div𝐰h𝝉j),(eh𝐕,div𝐰h𝝉j)=((𝐮𝐮h),𝐰h𝝉j)((Πhp)𝒩ph𝒩,div𝐰h𝝉j)(pΠhp,div𝐰h𝝉j),\begin{array}[]{l}(e_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})=-\left(\nabla(\mathbf{u}-\mathbf{u}_{h}),\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}\right)-\left((\Pi_{h}p)^{\mathcal{N}}-p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}\right)-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}),\\ (e_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}})=-\left(\nabla(\mathbf{u}-\mathbf{u}_{h}),\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}\right)-\left((\Pi_{h}p)^{\mathcal{N}}-p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}\right)-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}^{\perp}}),\end{array}

(59)

for j=1,2,,Jj=1,2,\cdots,J, since the least square solution of a system is that of its normal equation. Now, (57) comes from (8), (59), Lemma 5.2, 5.3 and

|wj|1Cσ(|Kj|+|Kj+1|)1/2,j=1,2,,J.|w_{j}|_{1}\leq C_{\sigma}(|K_{j}|+|K_{j+1}|)^{1/2},\quad j=1,2,\cdots,J.

Even in case of regular vertices on Ω\partial\Omega, we can copy the above argument to establish Lemma 5.3 and 5.4 with a system of 2(J1)2(J-1) equations.

Denote by ph˙\dot{p_{h}}, the superposition of all the calculated pressures up to now, that is,

ph˙=ph𝒩+𝐕hph𝐕,\dot{p_{h}}=p_{h}^{\mathcal{N}}+\sum_{\mathbf{V}\in\mathcal{R}_{h}}p_{h}^{\mathbf{V}}, (60)

where h\mathcal{R}_{h} is a set of all regular vertices.

5.3 Step 3. sting pressure for nearly singular vertex except corners

When a vertex 𝐕\mathbf{V} is exactly singular, the system (50) is underdetermined, since the determinant of (53) makes

sin(θj+θj+1)=0 for all j=1,2,,J.-\sin(\theta_{j}+\theta_{j+1})=0\quad\mbox{ for all }j=1,2,\cdots,J.

Although 𝐕\mathbf{V} is not exactly singular, the error eh𝐕e_{h}^{\mathbf{V}} in (59) goes through a tiny smallest singular value, if it is nearly singular.

To overcome the problem on nearly singular vertices, we will replace equations in (50b) with new equations utilizing jumps of ph˙\dot{p_{h}} in (60) we have already calculated.

5.3.1 boundary singular vertex not a corner

Let’s fix a boundary vertex 𝐕\mathbf{V} which is not a corner point of Ω\partial\Omega. If 𝐕\mathbf{V} is nearly singular, it is exact and has only two triangle K1,K2K_{1},K_{2} which share 𝐕\mathbf{V} as in Figure 4-(a). Denote by 𝐕0,𝐕1,𝐕2,𝝉1\mathbf{V}_{0},\mathbf{V}_{1},\mathbf{V}_{2},\boldsymbol{\tau}_{1}, other 3 vertices and a unit tangent vector such that

𝐕𝐕1¯=K1K2,𝐕2K2{𝐕,𝐕1},𝐕0K1{𝐕,𝐕1},𝝉1=𝐕𝐕1|𝐕𝐕1¯|.\overline{\mathbf{V}\mathbf{V}_{1}}=K_{1}\cap K_{2},\quad\mathbf{V}_{2}\in K_{2}\setminus\{\mathbf{V},\mathbf{V}_{1}\},\quad\mathbf{V}_{0}\in K_{1}\setminus\{\mathbf{V},\mathbf{V}_{1}\},\quad\boldsymbol{\tau}_{1}=\frac{\overrightarrow{\mathbf{V}\mathbf{V}_{1}}}{|\overline{\mathbf{V}\mathbf{V}_{1}}|}.

Define the jump of a function qhq_{h} across K1K2K_{1}\cap K_{2} at 𝐕\mathbf{V} as

𝒥12(qh)=|K1K2|3(𝝉1(qh|K1)(𝐕)𝝉1(qh|K2)(𝐕)).\mathcal{J}_{12}(q_{h})={|K_{1}\cap K_{2}|^{3}}\left(\frac{\partial}{\partial\boldsymbol{\tau}_{1}}\left(q_{h}\big{|}_{K_{1}}\right)(\mathbf{V})-\frac{\partial}{\partial\boldsymbol{\tau}_{1}}\left(q_{h}\big{|}_{K_{2}}\right)(\mathbf{V})\right). (61)

If 𝐕\mathbf{V} is nearly singular, then instead of (50b), consider the following new system of 22 equations:

(ph𝐕,div𝐰h𝝉1)\displaystyle(p_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}}) =(𝐟,𝐰h𝝉1)(𝐮h,𝐰h𝝉1)(ph𝒩,div𝐰h𝝉1),\displaystyle=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}}), (62a)
𝒥12(ph𝐕)\displaystyle\mathcal{J}_{12}(p_{h}^{\mathbf{V}}) =𝒥12(ph˙).\displaystyle=-\mathcal{J}_{12}(\dot{p_{h}}). (62b)
Lemma 5.5.

Let 𝐕\mathbf{V} be a boundary nearly singular vertex, not a corner of Ω\partial\Omega. If ph𝐕p_{h}^{\mathbf{V}} is the solution of (62), then we estimate

(Πhp)𝐕ph𝐕0,K1K2Cσ(|𝐮𝐮h|1,K1K2+pΠhp0,K1K2).\|(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}\|_{0,K_{1}\cup K_{2}}\leq C_{\sigma}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,K_{1}\cup K_{2}}+\|p-\Pi_{h}p\|_{0,K_{1}\cup K_{2}}\right). (63)
Proof.

Since Πhp\Pi_{h}p is continuous on K1K2K_{1}\cap K_{2}, we have 𝒥12(Πhp)=0\mathcal{J}_{12}(\Pi_{h}p)=0. It is written in

𝒥12((Πhp)𝐕)=𝒥12(Πhp(Πhp)𝐕).\mathcal{J}_{12}\left((\Pi_{h}p)^{\mathbf{V}}\right)=-\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}\right). (64)

Thus, if we denote the error by eh𝐕=(Πhp)𝐕ph𝐕e_{h}^{\mathbf{V}}=(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}, by same argument in regular case, we get

(eh𝐕,div𝐰h𝝉1)=((𝐮𝐮h),𝐰h𝝉1)((Πhp)𝒩ph𝒩,div𝐰h𝝉1)(pΠhp,div𝐰h𝝉1),𝒥12(eh𝐕)=𝒥12(Πhp(Πhp)𝐕ph˙).\begin{array}[]{rll}(e_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}})&=&-\left(\nabla(\mathbf{u}-\mathbf{u}_{h}),\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}}\right)-\left((\Pi_{h}p)^{\mathcal{N}}-p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}}\right)-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{1}}),\\ \mathcal{J}_{12}(e_{h}^{\mathbf{V}})&=&-\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\dot{p_{h}}\right).\end{array}

(65)

We can represent eh𝐕e_{h}^{\mathbf{V}} with 22 unknown constants e1,e2e_{1},e_{2} as

eh𝐕=e1𝒮𝐕K1+e2𝒮𝐕K2.e_{h}^{\mathbf{V}}=e_{1}\mathcal{S}_{\mathbf{V}K_{1}}+e_{2}\mathcal{S}_{\mathbf{V}K_{2}}. (66)

Let θ1,θ2\theta_{1},\theta_{2} be angles of K1,K2K_{1},K_{2} at 𝐕\mathbf{V}, respectively, and j=|𝐕𝐕j¯|,j=0,1,2\ell_{j}=|\overline{\mathbf{V}\mathbf{V}_{j}}|,j=0,1,2 as in Figure 4-(a). Then, from the property of sting functions in (7), we have

J12(eh𝐕)=60012e1+60012e2.J_{12}(e_{h}^{\mathbf{V}})=-600{\ell_{1}}^{2}e_{1}+600{\ell_{1}}^{2}e_{2}. (67)

If the first right hand side in (65) is abbreviated by aa, we can rewrite (65) into

01sinθ12e1+12sinθ22e2=a,60012e1+60012e2=𝒥12(Πhp(Πhp)𝐕ph˙).\begin{array}[]{ccccl}\displaystyle\frac{\ell_{0}\ell_{1}\sin\theta_{1}}{2}e_{1}&+&\displaystyle\frac{\ell_{1}\ell_{2}\sin\theta_{2}}{2}e_{2}&=&a,\\ -600{\ell_{1}}^{2}e_{1}&+&600{\ell_{1}}^{2}e_{2}&=&-\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\dot{p_{h}}\right).\end{array} (68)

We can repeat the same in regular case for the estimation:

|a|Cσ(|K1|+|K2|)1/2(|𝐮𝐮h|1,K1K2+pΠhp0,K1K2).|a|\leq C_{\sigma}\left(|K_{1}|+|K_{2}|\right)^{1/2}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,K_{1}\cup K_{2}}+\|p-\Pi_{h}p\|_{0,K_{1}\cup K_{2}}\right). (69)

If we have the following similar estimation for the second right hand side in (68):

|𝒥12(Πhp(Πhp)𝐕ph˙)|Cσ(|K1|+|K2|)1/2(|𝐮𝐮h|1,K1K2+pΠhp0,K1K2),|\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\dot{p_{h}}\right)|\leq C_{\sigma}\left(|K_{1}|+|K_{2}|\right)^{1/2}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,K_{1}\cup K_{2}}+\|p-\Pi_{h}p\|_{0,K_{1}\cup K_{2}}\right),

(70)

the estimation (63) will be established from (8), (66)-(70).

If 𝐕1\mathbf{V}_{1} is an interior vertex, it is regular by Lemma 3.1. Even in case of a boundary vertex 𝐕1\mathbf{V}_{1}, it is regular unless Ω=K1K2\Omega=K_{1}\cup K_{2}. We exclude that pathological case by Assumption 5.1. Thus, we conclude that 𝐕1\mathbf{V}_{1} is a regular vertex.

We also note that 𝒮𝐕0K1\mathcal{S}_{\mathbf{V}_{0}K_{1}}, 𝒮𝐕2K2\mathcal{S}_{\mathbf{V}_{2}K_{2}} are constant on K1K2K_{1}\cap K_{2} from (7). It means that they do not have any effect on 𝒥12(Πhp)\mathcal{J}_{12}(\Pi_{h}p) and 𝒥12(ph˙)\mathcal{J}_{12}(\dot{p_{h}}). It results in

𝒥12(Πhp(Πhp)𝐕)=𝒥12((Πhp)𝒩+(Πhp)𝐕1).\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}\right)=\mathcal{J}_{12}\left((\Pi_{h}p)^{\mathcal{N}}+(\Pi_{h}p)^{\mathbf{V}_{1}}\right). (71)

Similarly, whether 𝐕0,𝐕2\mathbf{V}_{0},\mathbf{V}_{2} are regular or not, we have

𝒥12(ph˙)=𝒥12(ph𝒩+ph𝐕1).\mathcal{J}_{12}(\dot{p_{h}})=\mathcal{J}_{12}(p_{h}^{\mathcal{N}}+p_{h}^{\mathbf{V}_{1}}). (72)

From (71), (72), we write

𝒥12(Πhp(Πhp)𝐕ph˙)=𝒥12((Πhp)𝒩ph𝒩)+𝒥12((Πhp)𝐕1ph𝐕1).\mathcal{J}_{12}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\dot{p_{h}}\right)=\mathcal{J}_{12}\left((\Pi_{h}p)^{\mathcal{N}}-p_{h}^{\mathcal{N}}\right)+\mathcal{J}_{12}\left((\Pi_{h}p)^{\mathbf{V}_{1}}-p_{h}^{\mathbf{V}_{1}}\right). (73)

For each function qh𝒫h3(Ω)q_{h}\in\mathcal{P}_{h}^{3}(\Omega), we can estimate

|𝒥12(qh)|13(|qh|K1(𝐕)|+|qh|K2(𝐕)|)Cσ13(|K1|+|K2|)1/2|qh|1,K1K2Cσ12|qh|1,K1K2Cσ1qh0,K1K2Cσ(|K1|+|K2|)1/2qh0,K1K2.\begin{array}[]{lll}\left|\mathcal{J}_{12}\left(q_{h}\right)\right|&\leq&{\ell_{1}}^{3}\left(\left|\left.\nabla q_{h}\right|_{K_{1}}(\mathbf{V})\right|+\left|\left.\nabla q_{h}\right|_{K_{2}}(\mathbf{V})\right|\right)\leq C_{\sigma}{\ell_{1}}^{3}\left(|K_{1}|+|K_{2}|\right)^{-1/2}\left|q_{h}\right|_{1,K_{1}\cup K_{2}}\\ &\leq&C_{\sigma}{\ell_{1}}^{2}\left|q_{h}\right|_{1,K_{1}\cup K_{2}}\leq C_{\sigma}{\ell_{1}}\left\|q_{h}\right\|_{0,K_{1}\cup K_{2}}\leq C_{\sigma}\left(|K_{1}|+|K_{2}|\right)^{1/2}\left\|q_{h}\right\|_{0,K_{1}\cup K_{2}}.\end{array} (74)

Thus we obtain (70) from (73), (74) and Lemma 5.2, 5.4. ∎

Refer to caption
(a) 𝐕Ω{corners of Ω}\mathbf{V}\in\partial\Omega\setminus\{\mbox{corners of }\partial\Omega\}
Refer to caption
(b) a corner 𝐕\mathbf{V} of Ω\partial\Omega.
Figure 4: examples of boundary singular vertices (dashed lines belong to Ω\partial\Omega)

5.3.2 interior singular vertex

Let 𝐕\mathbf{V} be an interior vertex and go back to the setting in Section 5.2.2 as in Figure 3-(b). For each j=1,2,,Jj=1,2,\cdots,J, define the jump of a function qhq_{h} across KjKj+1K_{j}\cap K_{j+1} at 𝐕\mathbf{V} as

𝒥jj+1(qh)=|KjKj+1|3(𝝉j(qh|Kj)(𝐕)𝝉j(qh|Kj+1)(𝐕)).\mathcal{J}_{j\hskip 1.0ptj+1}(q_{h})={|K_{j}\cap K_{j+1}|^{3}}\left(\frac{\partial}{\partial\boldsymbol{\tau}_{j}}\left(q_{h}\big{|}_{K_{j}}\right)(\mathbf{V})-\frac{\partial}{\partial\boldsymbol{\tau}_{j}}\left(q_{h}\big{|}_{K_{j+1}}\right)(\mathbf{V})\right). (75)

If 𝐕\mathbf{V} is nearly singular, replace (50b) with new equations using the jumps of ph˙\dot{p_{h}} in (75). That is, we consider the following system of 2J2J equations:

(ph𝐕,div𝐰h𝝉j)\displaystyle(p_{h}^{\mathbf{V}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}) =(𝐟,𝐰h𝝉j)(𝐮h,𝐰h𝝉j)(ph𝒩,div𝐰h𝝉j),\displaystyle=(\mathbf{f},\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(\nabla\mathbf{u}_{h},\nabla\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}})-(p_{h}^{\mathcal{N}},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h}^{\boldsymbol{\tau}_{j}}), (76a)
𝒥jj+1(ph𝐕)\displaystyle\mathcal{J}_{j\hskip 1.0ptj+1}(p_{h}^{\mathbf{V}}) =𝒥jj+1(ph˙),j=1,2,,J.\displaystyle=-\mathcal{J}_{j\hskip 1.0ptj+1}(\dot{p_{h}}),\quad j=1,2,\cdots,J. (76b)

Let ph𝐕p_{h}^{\mathbf{V}} be the least square solution of (76). We note that all other adjacent vertices 𝐕1,𝐕2,,𝐕J\mathbf{V}_{1},\mathbf{V}_{2},\cdots,\mathbf{V}_{J} are regular from Lemma 3.1. Thus we can repeat the arguments in the proofs of Lemma 5.3, 5.4, 5.5 to establish the following lemma.

Lemma 5.6.

Let 𝐕\mathbf{V} be an interior nearly singular vertex. If ph𝐕p_{h}^{\mathbf{V}} be the least square solution of (76), then we estimate

(Πhp)𝐕ph𝐕0,𝒦(𝐕)Cσ(|𝐮𝐮h|1,𝒦(𝐕)+pΠhp0,𝒦(𝐕)).\|(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}\|_{0,\mathcal{K}(\mathbf{V})}\leq C_{\sigma}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,\mathcal{K}(\mathbf{V})}+\|p-\Pi_{h}p\|_{0,\mathcal{K}(\mathbf{V})}\right). (77)

Denote by ph¨\ddot{p_{h}}, the superposition of all the calculated pressures up to now, that is,

ph¨=ph𝒩+𝐕𝒱hph𝐕,\ddot{p_{h}}=p_{h}^{\mathcal{N}}+\sum_{\mathbf{V}\in\mathcal{V}_{h}^{\prime}}p_{h}^{\mathbf{V}}, (78)

where 𝒱h\mathcal{V}_{h}^{\prime} is a set of all vertices except nearly singular corner points of Ω\partial\Omega.

5.4 Step 4. sting pressure for nearly singular corner

Let 𝐕\mathbf{V} be a corner point of Ω\partial\Omega which is nearly singular and K1,K2,,KJK_{1},K_{2},\cdots,K_{J}, triangles in 𝒯h\mathcal{T}_{h} sharing 𝐕\mathbf{V}. If J2J\geq 2, we can calculate the least square solution ph𝐕p_{h}^{\mathbf{V}} satisfying the system of 2(J1)2(J-1) equations as in (76) and estimate its error (Πhp)𝐕ph𝐕(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}} as in (77).

If J=1J=1, there exists a triangle KK in 𝒯h\mathcal{T}_{h} sharing two vertices 𝐕0,𝐕1\mathbf{V}_{0},\mathbf{V}_{1} with K1K_{1} as in Figure 4-(b). Denote by 𝐕2\mathbf{V}_{2}, the third vertex of KK not shared with K1K_{1}.

Define the jump of a function qhq_{h} across K1KK_{1}\cap K at 𝐕1\mathbf{V}_{1} as

𝒥(qh)=3(𝐧(qh|K1)(𝐕1)𝐧(qh|K)(𝐕1)),\mathcal{J}(q_{h})={\ell^{3}}\left(\frac{\partial}{\partial\mathbf{n}}\left(q_{h}\big{|}_{K_{1}}\right)(\mathbf{V}_{1})-\frac{\partial}{\partial\mathbf{n}}\left(q_{h}\big{|}_{K}\right)(\mathbf{V}_{1})\right), (79)

where 𝐧\mathbf{n} is a unit outward normal vector on K1KK_{1}\cap K of K1K_{1} and \ell is the distance between 𝐕\mathbf{V} and K1KK_{1}\cap K. Then we can determine ph𝐕=α𝒮𝐕K1p_{h}^{\mathbf{V}}=\alpha\mathcal{S}_{\mathbf{V}K_{1}} for some constant α\alpha satisfying

𝒥(ph𝐕)=𝒥(ph¨).\mathcal{J}(p_{h}^{\mathbf{V}})=-\mathcal{J}(\ddot{p_{h}}). (80)
Lemma 5.7.

Let 𝐕\mathbf{V} be a corner which meets only one triangle. If ph𝐕p_{h}^{\mathbf{V}} is the solution of (80), then we estimate

(Πhp)𝐕ph𝐕0,K1Cσ(|𝐮𝐮h|1,K1K+pΠhp0,K1K).\|(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}\|_{0,K_{1}}\leq C_{\sigma}\left(|\mathbf{u}-\mathbf{u}_{h}|_{1,K_{1}\cup K}+\|p-\Pi_{h}p\|_{0,K_{1}\cup K}\right). (81)
Proof.

We note that Πhp\nabla\Pi_{h}p is continuous at 𝐕1\mathbf{V}_{1}, since it is a Hermite interpolation of pp in (25). It can be written in

𝒥((Πhp)𝐕)=𝒥(Πhp(Πhp)𝐕).\mathcal{J}\left((\Pi_{h}p)^{\mathbf{V}}\right)=-\mathcal{J}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}\right). (82)

Set the error eh𝐕=(Πhp)𝐕ph𝐕=e𝒮𝐕K1e_{h}^{\mathbf{V}}=(\Pi_{h}p)^{\mathbf{V}}-p_{h}^{\mathbf{V}}=e\mathcal{S}_{\mathbf{V}K_{1}} for some constant ee, then from (80), (82), we have

𝒥(e𝒮𝐕K1)=𝒥(Πhp(Πhp)𝐕ph¨).\mathcal{J}(e\mathcal{S}_{\mathbf{V}K_{1}})=-\mathcal{J}\left(\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\ddot{p_{h}}\right). (83)

From the property of sting functions in (7), we have

𝒥(e𝒮𝐕K1)=1802e.\mathcal{J}(e\mathcal{S}_{\mathbf{V}K_{1}})=-180{\ell}^{2}e. (84)

We have already calculated ph𝐕0,ph𝐕1p_{h}^{\mathbf{V}_{0}},p_{h}^{\mathbf{V}_{1}}, ph𝐕2p_{h}^{\mathbf{V}_{2}}, since 𝐕0,𝐕1,𝐕2\mathbf{V}_{0},\mathbf{V}_{1},\mathbf{V}_{2} are not corners by Assumption 5.1. Thus the difference in (83) is written as

Πhp(Πhp)𝐕ph¨=(Πhp)𝒩ph𝒩+j=02(Πhp)𝐕jph𝐕j+(Πhp)C.\Pi_{h}p-(\Pi_{h}p)^{\mathbf{V}}-\ddot{p_{h}}=(\Pi_{h}p)^{\mathcal{N}}-p_{h}^{\mathcal{N}}+\sum_{j=0}^{2}(\Pi_{h}p)^{\mathbf{V}_{j}}-p_{h}^{\mathbf{V}_{j}}+(\Pi_{h}p)^{C}. (85)

For each function qh𝒫h3(Ω)q_{h}\in\mathcal{P}_{h}^{3}(\Omega), we can estimate the following as in (74),

|𝒥(qh)|Cσ(|K1|+|K|)1/2qh0,K1K.\left|\mathcal{J}\left(q_{h}\right)\right|\leq C_{\sigma}\left(|K_{1}|+|K|\right)^{1/2}\left\|q_{h}\right\|_{0,K_{1}\cup K}. (86)

Now we can reach at (81) with (8), (83)-(86) and Lemma 5.2, 5.4-5.6 since 𝒥((Πhp)C)=0\mathcal{J}\left((\Pi_{h}p)^{C}\right)=0. ∎

Denote by ph˙˙˙\dddot{p_{h}}, the superposition of all the calculated pressures up to now,

ph˙˙˙=ph𝒩+𝐕𝒱hph𝐕,\dddot{p_{h}}=p_{h}^{\mathcal{N}}+\sum_{\mathbf{V}\in\mathcal{V}_{h}}p_{h}^{\mathbf{V}}, (87)

where 𝒱h\mathcal{V}_{h} is a set of all vertices in 𝒯h\mathcal{T}_{h}. For notation consistency, denote again

Πhp˙˙˙=Πhp(Πhp)𝒞=(Πhp)𝒩+𝐕𝒱h(Πhp)𝐕.\dddot{\Pi_{h}p}=\Pi_{h}p-(\Pi_{h}p)^{\mathpzc{C}}=(\Pi_{h}p)^{\mathcal{N}}+\sum_{\mathbf{V}\in\mathcal{V}_{h}}(\Pi_{h}p)^{\mathbf{V}}. (88)

Then hereby, we have established the following lemma.

Lemma 5.8.
Πhp˙˙˙ph˙˙˙0Cσ(|𝐮𝐮h|1+pΠhp0).\left\|\dddot{\Pi_{h}p}-\dddot{p_{h}}\right\|_{0}\leq C_{\sigma}(|\mathbf{u}-\mathbf{u}_{h}|_{1}+\|p-\Pi_{h}p\|_{0}).

5.5 Step 5. piecewise constant pressure

Let YhY_{h} be a space of all functions in [𝒫h2(Ω)H01(Ω)]2\left[\mathcal{P}_{h}^{2}(\Omega)\cap H_{0}^{1}(\Omega)\right]^{2} vanishing at all vertices and whose value at the midpoint of each edge EE is normal to EE. Define

Xh=[𝒫h1(Ω)H01(Ω)]2Yh.X_{h}=\left[\mathcal{P}_{h}^{1}(\Omega)\cap H_{0}^{1}(\Omega)\right]^{2}\bigoplus Y_{h}.

Then, there exists a unique (𝐰h,phc)Xh×𝒫h0(Ω)L02(Ω)(\mathbf{w}_{h},p_{h}^{c})\in X_{h}\times\mathcal{P}_{h}^{0}(\Omega)\cap L_{0}^{2}(\Omega) which satisfies a following discrete Stokes problem:

(𝐰h,𝐯h)+(phc,div𝐯h)=(𝐟,𝐯h)(𝐮h,𝐯h)(ph˙˙˙,div𝐯h),(qh,div𝐰h)=0,\begin{array}[]{rll}(\nabla\mathbf{w}_{h},\nabla\mathbf{v}_{h})+(p_{h}^{c},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})&=&(\mathbf{f},\mathbf{v}_{h})-(\nabla\mathbf{u}_{h},\nabla\mathbf{v}_{h})-(\dddot{p_{h}},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}),\\ (q_{h},\mathrm{div}\hskip 1.42262pt\mathbf{w}_{h})&=&0,\end{array} (89)

for all (𝐯h,qh)Xh×𝒫h0(Ω)L02(Ω)(\mathbf{v}_{h},q_{h})\in X_{h}\times\mathcal{P}_{h}^{0}(\Omega)\cap L_{0}^{2}(\Omega), since Xh×𝒫h0(Ω)L02(Ω)X_{h}\times\mathcal{P}_{h}^{0}(\Omega)\cap L_{0}^{2}(\Omega) satisfies the inf-sup condition [1, 3].

Denote by 𝓂(𝒻)\mathpzc{m}(f), the average of a function ff over Ω\Omega, then define

ph𝒞=phc𝓂(𝓅𝒽˙˙˙).p_{h}^{\mathpzc{C}}=p_{h}^{c}-\mathpzc{m}(\dddot{p_{h}}). (90)
Lemma 5.9.
(Πhp)𝒞ph𝒞0Cσ(|𝐮𝐮h|1+pΠhp0).\|(\Pi_{h}p)^{\mathpzc{C}}-p_{h}^{\mathpzc{C}}\|_{0}\leq C_{\sigma}(|\mathbf{u}-\mathbf{u}_{h}|_{1}+\|p-\Pi_{h}p\|_{0}). (91)
Proof.

From (28), (88), we have for all 𝐯hXh\mathbf{v}_{h}\in X_{h},

((Πhp)𝒞,div𝐯h)=(𝐟,𝐯h)(𝐮,𝐯h)(pΠhp,div𝐯h)(Πhp˙˙˙,div𝐯h).\left((\Pi_{h}p)^{\mathpzc{C}},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}\right)=(\mathbf{f},\mathbf{v}_{h})-(\nabla\mathbf{u},\nabla\mathbf{v}_{h})-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})-\left(\dddot{\Pi_{h}p},\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}\right). (92)

Let’s decompose XhX_{h} as

Wh={𝐱hXh|(qh,div𝐱h)=0 for all qh𝒫h0(Ω)L02(Ω)},Wh={𝐳hXh|(𝐱h,𝐳h)=0 for all 𝐱hWh}.\begin{array}[]{lll}W_{h}&=&\{\mathbf{x}_{h}\in X_{h}\ |\ (q_{h},\mathrm{div}\hskip 1.42262pt\mathbf{x}_{h})=0\mbox{ for all }q_{h}\in\mathcal{P}_{h}^{0}(\Omega)\cap L_{0}^{2}(\Omega)\},\\ W_{h}^{\perp}&=&\{\mathbf{z}_{h}\in X_{h}\ |\ (\nabla\mathbf{x}_{h},\nabla\mathbf{z}_{h})=0\ \mbox{ for all }\mathbf{x}_{h}\in W_{h}\}.\end{array} (93)

Then, we can rewrite (89) for all 𝐳hWh\mathbf{z}_{h}\in W_{h}^{\perp},

(phc,div𝐳h)=(𝐟,𝐳h)(𝐮h,𝐳h)(ph˙˙˙,div𝐳h).(p_{h}^{c},\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h})=(\mathbf{f},\mathbf{z}_{h})-(\nabla\mathbf{u}_{h},\nabla\mathbf{z}_{h})-(\dddot{p_{h}},\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h}). (94)

Denote the error by eh𝒞=(Πhp)𝒞phce_{h}^{\mathpzc{C}}=(\Pi_{h}p)^{\mathpzc{C}}-p_{h}^{c}. Then from (92), (94), we have for all 𝐳hWh\mathbf{z}_{h}\in W_{h}^{\perp},

(eh𝒞,div𝐳h)=(𝐮𝐮h,𝐳h)(pΠhp,div𝐳h)(Πhp˙˙˙ph˙˙˙,div𝐳h).\left(e_{h}^{\mathpzc{C}},\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h}\right)=-(\nabla\mathbf{u}-\nabla\mathbf{u}_{h},\nabla\mathbf{z}_{h})-(p-\Pi_{h}p,\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h})-\left(\dddot{\Pi_{h}p}-\dddot{p_{h}},\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h}\right). (95)

Since eh𝒞𝓂(𝒽𝒞)02(Ω)e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})\in L_{0}^{2}(\Omega), there exists a nonzero 𝐯hXh\mathbf{v}_{h}\in X_{h} such that

eh𝒞𝓂(𝒽𝒞)0|𝐯𝒽|1β(𝒽𝒞𝓂(𝒽𝒞),div𝐯𝒽),\|e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})\|_{0}\ |\mathbf{v}_{h}|_{1}\leq\beta(e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}}),\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h}), (96)

from the inf-sup condition, where 0<β0<\beta is a constant, independent of 𝒯h\mathcal{T}_{h}.

If we decompose 𝐯h\mathbf{v}_{h} orthogonally into 𝐱hWh\mathbf{x}_{h}\in W_{h} and 𝐳hWh\mathbf{z}_{h}\in W_{h}^{\perp} such that

𝐯h=𝐱h+𝐳h,\mathbf{v}_{h}=\mathbf{x}_{h}+\mathbf{z}_{h}, (97)

from (95)-(97) and Lemma 5.8, we expand

eh𝒞𝓂(𝒽𝒞)0|𝐳𝒽|1𝒽𝒞𝓂(𝒽𝒞)0|𝐯𝒽|1β(𝒽𝒞𝓂(𝒽𝒞),div𝐯𝒽)=β(eh𝒞𝓂(𝒽𝒞),div𝐳𝒽)=β(𝒽𝒞,div𝐳𝒽)β𝒞σ(|𝐮𝐮𝒽|1+𝓅Π𝒽𝓅0)|𝐳𝒽|1.\|e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})\|_{0}\ |\mathbf{z}_{h}|_{1}\leq\|e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})\|_{0}\ |\mathbf{v}_{h}|_{1}\leq\beta(e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}}),\mathrm{div}\hskip 1.42262pt\mathbf{v}_{h})\\ =\beta(e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}}),\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h})=\beta(e_{h}^{\mathpzc{C}},\mathrm{div}\hskip 1.42262pt\mathbf{z}_{h})\leq\beta C_{\sigma}(|\mathbf{u}-\mathbf{u}_{h}|_{1}+\|p-\Pi_{h}p\|_{0})|\mathbf{z}_{h}|_{1}. (98)

If zh=𝟎z_{h}=\mathbf{0}, then eh𝒞𝓂(𝒽𝒞)=0e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})=0 from (96), (97). Therefore, from (98), we have

eh𝒞𝓂(𝒽𝒞)0𝒞σ(|𝐮𝐮𝒽|1+𝓅Π𝒽𝓅0).\|e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})\|_{0}\leq C_{\sigma}(|\mathbf{u}-\mathbf{u}_{h}|_{1}+\|p-\Pi_{h}p\|_{0}). (99)

Now, since 𝓂(𝒽𝒞)=𝓂((Π𝒽𝓅)𝒞)\mathpzc{m}(e_{h}^{\mathpzc{C}})=\mathpzc{m}\left((\Pi_{h}p)^{\mathpzc{C}}\right), we expand from (87), (88), (90) that

(Πhp)𝒞ph𝒞=(Πhp)𝒞(phc𝓂(𝓅𝒽˙˙˙))=eh𝒞𝓂(𝒽𝒞)+𝓂((Π𝒽𝓅)𝒞)+𝓂(𝓅𝒽˙˙˙)=eh𝒞𝓂(𝒽𝒞)+𝓂(Π𝒽𝓅)𝓂(Π𝒽𝓅˙˙˙)+𝓂(𝓅𝒽˙˙˙).\begin{array}[]{lll}(\Pi_{h}p)^{\mathpzc{C}}-p_{h}^{\mathpzc{C}}&=&(\Pi_{h}p)^{\mathpzc{C}}-\left(p_{h}^{c}-\mathpzc{m}(\dddot{p_{h}})\right)=e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})+\mathpzc{m}\left((\Pi_{h}p)^{\mathpzc{C}}\right)+\mathpzc{m}(\dddot{p_{h}})\\ &=&e_{h}^{\mathpzc{C}}-\mathpzc{m}(e_{h}^{\mathpzc{C}})+\mathpzc{m}(\Pi_{h}p)-\mathpzc{m}\left(\dddot{\Pi_{h}p}\right)+\mathpzc{m}\left(\dddot{p_{h}}\right).\end{array} (100)

For the averages in (100), we observe the following two basic inequalities:

|𝓂(Π𝒽𝓅)|=|𝓂(Π𝒽𝓅𝓅)||Ω|1/2Π𝒽𝓅𝓅0,|\mathpzc{m}(\Pi_{h}p)|=|\mathpzc{m}(\Pi_{h}p-p)|\leq|\Omega|^{-1/2}\|\Pi_{h}p-p\|_{0}, (101)
|𝓂(Π𝒽𝓅˙˙˙)𝓂(𝓅𝒽˙˙˙)|=|𝓂(Π𝒽𝓅˙˙˙𝓅𝒽˙˙˙)||Ω|1/2Πhp˙˙˙ph˙˙˙0.\left|\mathpzc{m}\left(\dddot{\Pi_{h}p}\right)-\mathpzc{m}\left(\dddot{p_{h}}\right)\right|=\left|\mathpzc{m}\left(\dddot{\Pi_{h}p}-\dddot{p_{h}}\right)\right|\leq|\Omega|^{-1/2}\|\dddot{\Pi_{h}p}-\dddot{p_{h}}\|_{0}. (102)

Therefore, (91) is established from (99)-(102) and Lemma 5.8. ∎

Now we reach at Theorem 5.1 with the definitions (23), (87), (88) and Lemma 5.8, 5.9 and Theorem 2.1.

6 Numerical test

We tested the suggested successive method in Ω=[0,1]2\Omega=[0,1]^{2} with the velocity 𝐮\mathbf{u} and pressure pp such that

𝐮=(s(x)s(y),s(x)s(y)),p=sin(4πx)eπy, where s(t)=(t2t)sin(2πt).\mathbf{u}=\left(s(x)s^{\prime}(y),-s^{\prime}(x)s(y)\right),\quad p=\sin(4\pi x)e^{\pi y},\quad\mbox{ where }s(t)=(t^{2}-t)\sin(2\pi t).

For triangulations, we first formed the meshes of uniform squares over Ω\Omega, then added one exactly singular vertex in every square. An example of 8×8×48\times 8\times 4 mesh is given in Figure 5.

We have calculated the successive pressures ph𝒩,ph˙,ph¨p_{h}^{\mathcal{N}},\dot{p_{h}},\ddot{p_{h}} and ph𝒞p_{h}^{\mathpzc{C}} along to Step 1,2,3,5 in Section 5. Since the meshes have not any singular corner, ph˙˙˙=ph¨\dddot{p_{h}}=\ddot{p_{h}}. Their examples are depicted in Figure 6 as well as ph=ph˙˙˙+ph𝒞p_{h}=\dddot{p_{h}}+p_{h}^{\mathpzc{C}}.

The error table in Table 1 shows the optimal order of convergence, expected in Theorem 2.1 and 5.1. We adopted a direct linear solver in LAPACK on solving the systems from the problem (2) and (89) to focus on testing the suggested method.

Refer to caption
Figure 5: 8×8×48\times 8\times 4 mesh bearing 8×88\times 8 interior exactly singular vertices
Refer to caption
(a) ph𝒩p_{h}^{\mathcal{N}}
Refer to caption
(b) ph˙=ph𝒩+ph𝐕r\dot{p_{h}}=p_{h}^{\mathcal{N}}+\sum p_{h}^{\mathbf{V}_{r}}
Refer to caption
(c) ph¨=ph˙+ph𝐕s,ph˙˙˙=ph¨\ddot{p_{h}}=\dot{p_{h}}+\sum p_{h}^{\mathbf{V}_{s}},\quad\dddot{p_{h}}=\ddot{p_{h}}
Refer to caption
(d) ph𝒞p_{h}^{\mathpzc{C}}
Refer to caption
(e) ph=ph˙˙˙+ph𝒞p_{h}=\dddot{p_{h}}+p_{h}^{\mathpzc{C}}
Figure 6: successive pressures calculated in 8×8×48\times 8\times 4 mesh in Figure 5
mesh |𝐮𝐮h|1|\mathbf{u}-\mathbf{u}_{h}|_{1} order pph0\|p-p_{h}\|_{0} order
4 x 4 x 4 1.1264E-2 5.8000E-2
8 x 8 x 4 6.1498E-4 4.1950 2.7012E-3 4.4244
16 x 16 x 4 3.5942E-5 4.0968 1.6760E-4 4.0105
32 x 32 x 4 2.2002E-6 4.0299 1.0454E-5 4.0029
Table 1: error table

References

  • [1] C. Bernardi and G. Raugel, Analysis of some finite elements for the Stokes problem, Mathematics of Computation, 44 (1985), 71-79, DOI: 10.2307/2007793
  • [2] S. C. Brenner and L. R. Scott, The mathematical theory of finite element methods, Springer-Verlag, New York, 2nd edition, 2002
  • [3] F. Brezzi and M. Fortin, Mixed and hybrid finite element methods, Springer-Verlag, New York, 1991
  • [4] P. G. Ciarlet, The finite element method for elliptic equations, North-Holland, Amsterdam, 1978
  • [5] R. S. Falk and M. Neilan, Stokes complexes and the construction of stable finite elements with pointwise mass conservation, SIAM journal of Numerical Analysis, 51 (2013), 1308-1326, DOI: 10.1137/120888132
  • [6] V. Girault and P. A. Raviart, Finite element methods for the Navier-Stokes equations: Theory and Algorithms, Springer-Verlag, New York, 1986
  • [7] J. Guzman and L. R. Scott, The Scott-Vogelius finite elements revisited, Mathematics of Computation, electronically published (2018), DOI: 10.1090/mcom/3346
  • [8] J. N. Lyness and D. Jespersen, Moderate degree symmetric quadrature rules for the triangle, IMA Journal of Applied Mathematics, 15 (1975), 19-32, DOI: 10.1093/imamat/15.1.19
  • [9] C. Park, Spurious pressure in Scott–Vogelius elements, Journal of Computational and Applied Mathematics, 363 (2020), 370-391, https://doi.org/10.1016/j.cam.2019.06.007
  • [10] L. R. Scott and M. Vogelius, Norm estimates for a maximal right inverse of the divergence operator in spaces of piecewise polynomials, ESIAM: M2AN, 19 (1985), 111-143, DOI: 10.1051/m2an/1985190101111