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.

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

the computational space. Unlike the MFE methods, this partial differential<br />

equation is well-defined for all time. In a similar manner to the optimal<br />

mesh methods, we need only solve a scalar equation; however, unlike the<br />

optimal mesh methods this equation is linear in the physical coordinates.<br />

Observe, further, that unlike certain of the MMPDE-based and variationalbased<br />

methods described in Section 3, this system automatically deals <strong>with</strong><br />

the mesh location on the boundaries. Having determined φ from this equation,<br />

the mesh velocity v can be determined directly, and the mesh then<br />

found from integrating the equation x t = v <strong>with</strong> respect to time using, for<br />

example, an SDIRK method. This time integration to give x has the disadvantage<br />

of possibly introducing mesh tangling, and of mesh points <strong>moving</strong><br />

out of the domain during the course of the integration, if an appropriately<br />

coarse discretization is used. This method is described in Cao et al. (2002)<br />

An alternative formulation, also described in Cao et al. (2002), considers<br />

a variational formulation, where it is shown that the solution of (4.3) is also<br />

the minimizer of the functional<br />

I[v] = 1 2<br />

∫Ω P<br />

(<br />

|∇ · (Mv)+M t | 2 +<br />

( M<br />

w<br />

) 2<br />

|∇ × w(v − u)| 2 )<br />

dx. (4.7)<br />

This equation can be discretized using a finite element method <strong>with</strong> basis<br />

functions defined over the mesh in the physical space, and v found using<br />

a simple Galerkin method. The mesh points can then also be found from<br />

v through a Galerkin calculation. Details of these calculations are given in<br />

Baines et al. (2005, 2006), where the GCL method is coupled to an ALE (arbitrary<br />

Lagrangian–Eulerian) method for discretizing the underlying PDE.<br />

In the usual implementation of the GCL method the weight function<br />

w =1<br />

is taken. This gives an irrotational mesh (in the physical coordinates) when<br />

the background velocity u = 0. In many implementations of GCL, u =0.<br />

However, for certain applications, such as in computational fluid dynamics,<br />

the background velocity can be taken to be the flow velocity.<br />

The appeal of the GCL methods is that to find φ (and hence v) we need<br />

only solve a scalar linear elliptic partial differential equation. This seems<br />

to have the advantage over the optimal transport methods, which require<br />

the solution of a nonlinear equation. However, like other velocity-based<br />

methods it has the potential disadvantages of having problems <strong>with</strong> mesh<br />

tangling and mesh skewness as the mesh points follow the <strong>moving</strong> features<br />

in a monitor function. Furthermore, we require the solution of the mesh<br />

equations in the physical domain rather than the computational domain.<br />

This loses some of the speed advantages gained (by, for example, the use of<br />

spectral methods) when solving the mesh equations on a very uniform mesh<br />

in the computational domain.

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

Saved successfully!

Ooh no, something went wrong!