26.12.2013 Views

Adaptivity with moving grids

Adaptivity with moving grids

Adaptivity with moving grids

SHOW MORE
SHOW LESS

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> 55<br />

smooth to avoid possible deterioration of accuracy: see, e.g., Mulholland,<br />

Huang and Sloan (1998), Wang and Shen (2005), Feng, Yu, Hu, Liu, Du<br />

and Chen (2006) and Tee and Trefethen (2006). While preliminary results<br />

are promising, much further investigation is needed to determine the full<br />

potential of these adaptive spectral methods.<br />

3.2. MMPDEs and variational methods in n dimensions<br />

3.2.1. Description of some variational-based methods<br />

In the variational and MMPDE approaches of mesh adaptation in n dimensions,<br />

briefly described in Section 2, adaptive meshes are also generated<br />

as images of a computational mesh under a coordinate transformation from<br />

the computational domain to the physical domain. Such a coordinate transformation<br />

is determined by an adaptation functional, which is commonly<br />

designed to measure the difficulty in the numerical approximation of the<br />

physical solution. The functional often involves mesh properties and employs<br />

a monitor function to control mesh quality and mesh concentration.<br />

The key to the development of variational and MMPDE methods is the formulation<br />

of the adaptation functional. Direct-use standard-error estimates<br />

are often not appropriate since they often lead to non-convex functionals<br />

in two and higher dimensions. Instead, most of the existing methods have<br />

been developed based on physical, geometric, mesh quality control, and/or<br />

other considerations.<br />

The functional can be formulated in terms of either the coordinate transformation<br />

x = F(ξ,t) or its inverse transformation ξ = F −1 (x,t). The latter<br />

has been used more commonly than the former because it is less likely to<br />

produce mesh tangling for non-convex domains (e.g., see Dvinsky (1991)).<br />

In the latter case, the adaptation functional takes the general form<br />

∫<br />

I(ξ) = G(M, ξ, ∇ξ i )dx, i =1,...,n, (3.26)<br />

Ω P<br />

where G is a continuous function of its arguments, M is the (scalar or<br />

matrix-valued) monitor function, and ∇ is the gradient operator <strong>with</strong> respect<br />

to the physical coordinate x. Once a functional has been defined,<br />

an MMPDE can be obtained as described in (2.26) as the gradient flow<br />

equation of the functional (Huang and Russell 1997b, 1999), i.e.,<br />

∂ξ i<br />

∂t = −P ∂I<br />

, i =1,...,n, (3.27)<br />

ɛ ∂ξ i<br />

where P is a positive differential operator and ɛ>0 is a parameter for<br />

adjusting the time scale of mesh movement. For the general form (3.26),<br />

this becomes<br />

∂ξ i<br />

∂t = P (<br />

∇ ·<br />

ɛ<br />

∂G<br />

∂(∇ξ i ) − ∂G<br />

∂ξ i<br />

)<br />

, i =1,...,n. (3.28)

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

Saved successfully!

Ooh no, something went wrong!