18.03.2014 Views

2 DGM for elliptic problems

2 DGM for elliptic problems

2 DGM for elliptic problems

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

lem:AP<br />

16 2 Elliptic <strong>problems</strong><br />

By ( 2.31), Ltwoproj π K,p is the operator of L 2 (K)-projection. This implies that Π hp<br />

is the L 2 (Ω)-projection on S hp : if ϕ ∈ L 2 (Ω), then<br />

∫<br />

∫<br />

Π hp ϕ ∈ S hp , (Π hp ϕ)v dx = ϕv dx ∀v ∈ S hp . (2.33) LtwoprOm<br />

Lemma 2.7. Let assumption (A1) be valid. Then<br />

Ω<br />

Ω<br />

|Π hp v − v| Hq (Ω,T h ) ≤ C Ah µ−q |v| Hµ (Ω,T h ), v ∈ H s (Ω, T h ), (2.34) eq:AP<br />

where µ = min(p + 1,s), 0 ≤ q ≤ s and C A is the constant from ( 2.30).<br />

eq:ap<br />

Proof. Using ( 2.32), eq:Pi_h definition of a seminorm in a broken Sobolev space and<br />

the approximation properties ( 2.30), eq:ap we obtain ( 2.34). eq:AP<br />

⊓⊔<br />

sec:2.2<br />

2.3 <strong>DGM</strong> based on a primal <strong>for</strong>mulation<br />

In this section we shall describe and analyze the <strong>DGM</strong> <strong>for</strong> the solution of<br />

problem ( 2.1) A0.1 – ( 2.3) A0.2b based on a primal <strong>for</strong>mulation. The approximate solution<br />

will be sought in the space S hp ⊂ H 1 (Ω, T ). However, the weak <strong>for</strong>mulation<br />

( 2.5) A0.3 given in Section 2.1 sec:2.0 is not suitable <strong>for</strong> the derivation of the <strong>DGM</strong>, because<br />

( 2.5) A0.3 does not make sense <strong>for</strong> u ∈ H 1 (Ω, T ) ⊄ H 1 (Ω). There<strong>for</strong>e, we shall<br />

introduce a “weak <strong>for</strong>m of ( 2.1) A0.1 – ( 2.3) A0.2b in the sence of broken Sobolev spaces”.<br />

Let as assume that u is a sufficiently regular solution of ( 2.1) A0.1 – ( 2.3). A0.2b We<br />

shall consider the so-called strong solution, i.e. a solution u ∈ H 2 (Ω). We<br />

proceed in the following way: We multiply ( 2.1) A0.1 by a function v ∈ H 2 (Ω, T h )<br />

integrate over K ∈ T and use Green’s theorem. Summing over all K ∈ T h , we<br />

obtain the identity<br />

∑<br />

∇u · ∇v dx − ∑<br />

∫<br />

(n K · ∇u)v dS = f v dx, (2.35) A2.4<br />

K∈T h<br />

∫K<br />

K∈T h<br />

∫∂K<br />

where n K denotes the unit outer normal to ∂K. The surface integrals over<br />

∂K make sense due to the regularity of u. We split them according to the<br />

type of faces Γ that <strong>for</strong>m the boundary of the element K ∈ T h :<br />

∑<br />

K∈T h<br />

∫∂K<br />

= ∑<br />

Γ ∈F D h<br />

+ ∑<br />

∫<br />

Γ ∈F I h<br />

Γ<br />

∫<br />

(n K · ∇u)v dS (2.36) A2.5<br />

(n Γ · ∇u)v dS + ∑<br />

Γ<br />

(<br />

n Γ ·<br />

Γ ∈F N h<br />

(∇u| (L)<br />

Γ<br />

)v|(L) Γ<br />

∫<br />

Γ<br />

Ω<br />

(n Γ · ∇u)v dS<br />

− (∇u|(R) Γ<br />

)v|(R) Γ<br />

)<br />

dS.<br />

(There is a sign “−” in the last integral, since n Γ is the unit outer normal to<br />

but the unit inner normal to ∂K (R) sec:2.1.1<br />

Γ<br />

, see Section 2.2.1 or Figure 2.2.<br />

fig:normals<br />

∂K (L)<br />

Γ

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!