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.

<strong>Adaptivity</strong> <strong>with</strong> <strong>moving</strong> <strong>grids</strong> 17<br />

It is shown by Huang (2005a) that measures s defined in (2.6) and Q geo are<br />

mathematically equivalent.<br />

Some other quality measures can be found in Liseikin (1999, Chapter 3),<br />

Knupp (2001) and Shewchuk (2002). Again, a good adaptive method aims<br />

to control some or all of these measures of skewness, either explicitly or implicitly<br />

throughout the calculation, and we discuss this later in this section.<br />

In the case of scale-invariant meshes we shall show that, whilst the adaptation<br />

factor changes a great deal, the skewness hardly varies. More generally,<br />

it should be noted that whereas the adaptation factor often changes a great<br />

deal in a mesh, the skewness generally does not. To control terms in the<br />

error expression (2.2) arising from large solution gradients, it is generally<br />

more important to vary the adaptation factor. If this results in a locally<br />

larger value of the skewness then this can usually be tolerated.<br />

2.2.3. Mesh smoothness and regularity<br />

The smoothness or regularity of a mesh is a measure of how much the mesh<br />

elements vary over the mesh. This can be important since the accuracy and<br />

error in the numerical solution of partial differential equations generally depend<br />

upon the type of discretization, the quality of the mesh, the treatment<br />

of boundary conditions, and so on. A uniform mesh has the highest degree<br />

of regularity, which can lead to particularly low error estimates on such<br />

meshes. It is sometimes claimed that only uniform meshes have such low<br />

estimates, but in fact, as we shall see, they share this <strong>with</strong> sufficiently regular<br />

meshes. Although there is generally no simple relationship between the<br />

smoothness of the mesh and the error (see Veldman and Rinzema (1992)),<br />

for most problems and most discretization methods, abrupt variations in the<br />

mesh will cause a deterioration in the convergence rate and an increase in<br />

the error (Thompson et al. 1985), or indeed in the accuracy of the approximation<br />

of a function over the mesh. Moreover, most discrete approximations<br />

of spatial differential operators have much larger condition numbers on an<br />

abruptly varying mesh than they do on a gradually varying one, and these<br />

ill-conditioned approximations may result in stiffness in the time integration<br />

for time-dependent problems.<br />

The smoothness of a mesh can be expressed in terms of the regularity of<br />

the underlying mesh function F.<br />

Definition. Ameshτ P has degree of regularity r if F ∈ C r (Ω C ).<br />

The regularity of F can often be achieved by determining F as a solution<br />

of a PDE system or a minimizer of a functional as in variational mesh<br />

generation methods. In many cases it is possible to have strong control<br />

over the derivatives of F allowing guaranteed regularity of the mesh τ P .It<br />

should be noted that an obvious way to determine a mesh is to prescribe<br />

the Jacobian function J exactly (Brackbill and Saltzman 1982). However,

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

Saved successfully!

Ooh no, something went wrong!