Adaptivity with moving grids
Adaptivity with moving grids
Adaptivity with moving grids
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Adaptivity</strong> <strong>with</strong> <strong>moving</strong> <strong>grids</strong> 43<br />
where high resolution is required, and half where it is not. Such meshes<br />
are called 50:50 meshes (see Budd et al. (2005), Huang (2001a), Huang<br />
et al. (2002)) and a regularization of M to ensure that such meshes arise in<br />
practice has been proposed by Beckett and Mackenzie (2000). To show how<br />
such problems arise in a calculation in n dimensions, suppose that we have<br />
a scalar monitor function M for which ∫ Ω P<br />
M dx = θ. Consider now the<br />
situation in which there are two subsets A and B of Ω P ,<strong>with</strong>Ω P = A ∪ B<br />
so that the monitor function is designed to concentrate points in a small<br />
region A (so that A may be the support of a singularity or of a front). The<br />
preimage A ′ = F −1 (A) ⊂ Ω C represents those points in the computational<br />
domain which are mapped to A, <strong>with</strong> a similar set B ′ . Suppose now that<br />
|A ′ | and |B ′ | are the areas of these sets in Ω C , <strong>with</strong> respective areas |A| and<br />
|B| in Ω P . These areas measure the proportion of mesh points allocated<br />
to the corresponding sets A and B. Note that in most applications of an<br />
adaptive method, where mesh points have to be concentrated into a small<br />
region we would expect that<br />
|A ′ | = O(1), |A| = o(1), |B ′ | = O(1), |B| ≈|Ω P |. (2.71)<br />
Problems arise <strong>with</strong> mesh regularity and solution resolution away from<br />
the set A if |B ′ |≪|A ′ |. It follows immediately from the equidistribution<br />
principle that if θ = ∫ Ω P ≡A∪B<br />
M dx then<br />
∫<br />
Λ= |A′ |<br />
|B ′ | = ∫ A M dx<br />
B M dx = θ<br />
− 1.<br />
∫B<br />
M dx<br />
Furthermore,<br />
∫<br />
|A ′ A<br />
| =<br />
M dx .<br />
θ<br />
It follows from the conditions on A ′ and A in (2.71) that over the set A we<br />
have M ≫ θ. However, if the monitor function is so constructed such that<br />
over the set B we have M ≪ θ and hence ∫ B<br />
M dx ≪ θ (so that the integral<br />
of M is concentrated in A), then Λ will be very large and the mesh will lose<br />
regularity. Indeed, we will have |A ′ |≈1. Exactly such problems arose in<br />
some of the blow-up calculations reported in Budd et al. (1996). In such<br />
cases we must replace M by the regularized function introduced by Beckett<br />
and Mackenzie (2000) and given by<br />
ˆM = M +<br />
θ<br />
|Ω P | . (2.72)<br />
Observe that over the set A we have ˆM ≈ M, andoverB we have ˆM ≈<br />
θ/|Ω P |, and trivially<br />
∫<br />
ˆM dx =2θ.<br />
Ω P