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.

including the Fisher equation,<br />

<strong>Adaptivity</strong> <strong>with</strong> <strong>moving</strong> <strong>grids</strong> 117<br />

u t = βu xx + αu(1 − u). (5.34)<br />

This equation has an exact solution <strong>with</strong> a <strong>moving</strong> front given by<br />

u = (1 + exp( √ α/6x − 5αt/6)) −2 .<br />

Whilst it was found that the wave front was well resolved, the <strong>moving</strong> mesh<br />

method appeared to give worse results than for the fixed grid, <strong>with</strong> the front<br />

<strong>moving</strong> at the wrong speed (too fast). Li et al. (1998) proposed that this<br />

error was largely due to the effects of discretizing the additional convective<br />

terms due to the <strong>moving</strong> mesh, as discussed in Section 3, which can lead<br />

both to a high truncation error and instabilities in the solution. A possible<br />

solution is to use higher-order methods or a curvature-dependent monitor<br />

function; e.g., see Qiu and Sloan (1998). However, there is an interesting<br />

alternative reason for the error. In problems such as (5.34), the speed of<br />

motion of the front is not so much determined by the shape of the front<br />

itself but by the nature of the exponential terms in the tails of the solution<br />

and how they interact <strong>with</strong> the boundaries. Similar behaviour arises in<br />

the Cahn–Hilliard equation, and also in similar systems such as the Gray–<br />

Scott equation in chemistry. Somewhat paradoxically, the solution needs to<br />

be well resolved in the boundary regions where it appears to be behaving<br />

smoothly, in order for the front speed to be accurately resolved. Thus a<br />

uniform mesh may well perform better than a <strong>moving</strong> mesh in these cases,<br />

as it will probably be placing more mesh points close to the boundary.<br />

5.4. Problems involving a change of phase and/or combustion<br />

A very rich set of examples of the use of <strong>moving</strong> mesh methods is given<br />

by Stefan-type problems, involving phase changes or combustion, and examples<br />

of these can be found in Beckett et al. (2001a), Mackenzie and<br />

Mekwi (2007a), Mackenzie and Robertson (2002), Miller, Gleyzer and Imhoff<br />

(1998), Huang and Zhan (2004), Zegeling (2005), Tan (2007) and Tan<br />

et al. (2007).<br />

As an example we consider the combustion problem described in Cao<br />

et al. (1999a) and Moore and Flaherty (1992). The mathematical model is<br />

a system of coupled nonlinear reaction–diffusion equations,<br />

∂u<br />

∂t − ∇2 u = − R αδ u eδ(1−1/T ) ,<br />

∂T<br />

∂t − 1 Le ∇2 T =<br />

R<br />

δLe u eδ(1−1/T ) ,<br />

where u and T represent the dimensionless concentration and temperature<br />

of a chemical which is undertaking a one-step reaction. We consider the<br />

J-shape solution domain shown in Figure 5.10. The initial and boundary

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

Saved successfully!

Ooh no, something went wrong!