No label defined (Q4214992)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 1217642
Language Label Description Also known as
default for all languages
No label defined
    English
    No label defined
    scientific article; zbMATH DE number 1217642

      Statements

      2 November 1998
      0 references
      mesh generation
      0 references
      finite element method
      0 references
      distorted elements
      0 references
      spectral condition number
      0 references
      large eigenvalues
      0 references
      iterative methods
      0 references
      conjugate gradients
      0 references
      convergence behaviour
      0 references
      pre-conditioning
      0 references
      Practical experience with a number of commercial mesh generators confirms that elements often arise that are either severely distorted, or are considerably smaller than is necessary to resolve the phenomenon being modelled. In principle, this is not a problem since as long as the maximum size of an element is appropriate, the resolution of the mesh is usually assured. The practical difficulty arises from the fact that the small length scales associated with the small, or distorted elements introduce undesirable effects in the equations which arise from the finite element discretization. After linearization, the discrete equations which arise in finite element analysis often take the form \[ Au= f\tag{1} \] or \[ M{du\over dt}= -Au+f,\quad u(0)= u^0,\tag{2} \] where \(A\) and \(M\) are symmetric, positive definite matrices, \(f\) and \(u^0\) are known vectors, and \(u\) is a vector of unknowns. The matrices \(A\) and \(M\) are usually called the stiffness and mass matrices. The linear system (1) describes a static, or steady state situation, while (2) may result from a continuous in time discretization of a transient boundary value problem. The presence of distorted elements can significantly increase the spectral condition number of \(A\) by giving rise to some large eigenvalues. If direct elimination methods are used to solve (1), this may not be too much of a concern. However, iterative methods such as conjugate gradients are often used to solve static problems, and the presence of extreme eigenvalues can seriously degrade the convergence behaviour of such methods. For transient problems, the presence of distorted elements, or elements which are unnecessarily small gives rise to large eigenvalues \(\lambda\) of \(Ax=\lambda Mx\). Depending on the time-stepping method used to solve (2), these large eigenvalues may adversely effect the stability properties of the method. Even if an unconditonally stable method is used, then the linear system which needs to be solved will have a larger spectral condition number than would otherwise be the case; and so, just as in the static case, if iterative methods are used to solve this linear system, poor performance will usually result. In this paper, we briefly describe a method for handling distorted elements which works directly with the assembled equations (1) and (2). By modifying the manner in which these are solved, the undesirable effects of distorted elements can in many cases be avoided. The method can be thought of as a simple pre-conditioning of \(A\) or \(M^{-1}A\). Alternatively, it can be viewed as being very similar to the direct mesh manipulations which would be used to manually remove distorted elements. The difference though being that we do not work directly with the mesh geometry. For the entire collection see [Zbl 0897.00034].
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references