26.12.2013 Views

Adaptivity with moving grids

Adaptivity with moving grids

Adaptivity with moving grids

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.

14 C. J. Budd, W. Huang and R. D. Russell<br />

2.2. Static mesh properties: skewness, regularity and smoothness<br />

We consider first some immediate properties of the mesh τ P which are related<br />

to the function F. Broadly speaking these divide into local and global<br />

properties. The global properties relate to the isotropy of the mesh, orthogonality<br />

issues and the behaviour close to boundaries. We discuss these<br />

presently.<br />

The local properties relate to the size and shape of the elements of τ P .<br />

If τ C is divided into logical regular rectangular or triangular elements τC e ,<br />

then these are mapped to elements τP e in τ P . These elements will then be<br />

distorted rectangles/triangles, possibly <strong>with</strong> small angles at the vertices.<br />

Locally we can characterize each such element by the size h e of the largest<br />

side and (in two dimensions) the radius ρ e of the largest inscribed circle. If a<br />

partial differential equation is discretized over τ P using (say) a finite element<br />

method in two dimensions, then the error in the solution has contributions<br />

from the size and the shape of the elements, as well as from the derivatives<br />

of the solution itself. For example, if the solution u is interpolated over<br />

τp e by a piecewise linear interpolant Π(u), then the following apriorierror<br />

estimates are standard (Johnson 1987):<br />

max |Π(u) − u| ≤2h 2 τp<br />

e e max<br />

∂ 2 ∣<br />

u ∣∣∣<br />

∣ ,<br />

∂x i ∂x j max<br />

∂ ( ) ∣ ∣ Π(u) − u ∣∣ ≤ 6 h2 e<br />

max<br />

∂ 2 ∣<br />

u ∣∣∣<br />

∂x i ρ e<br />

∣ .<br />

∂x i ∂x j<br />

τ e p<br />

(2.2)<br />

An adapted mesh will usually aim to control the size and the shape of each<br />

element so that, for any particular solution u, the overall error is controlled.<br />

Thus, for example, if the second derivatives of u are large over τ e p , then the<br />

error (2.2) can be controlled locally by taking h e to be small, and ensuring<br />

that h e /ρ e remains bounded. We discuss these issues in detail later in this<br />

section. (See Cao (2005, 2007a) and Chen, Sun and Xu (2007) for a very<br />

complete analysis of this problem.)<br />

Both the size and the shape of the mesh elements can be described in<br />

terms of the local properties of F , and in particular its Jacobian J, given by<br />

The following is immediate.<br />

J = ∂F<br />

∂ξ . (2.3)<br />

Condition 2.1. For the map to be locally well-posed we require that J<br />

should be both bounded and invertible at all points in Ω C .

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

Saved successfully!

Ooh no, something went wrong!