No label defined (Q3140796)
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 440639
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 440639 |
Statements
28 November 1993
0 references
Laplace equation
0 references
consistent sparse factorization
0 references
finite element
0 references
test problems
0 references
Laplacian operators
0 references
0.8938704
0 references
0 references
0.87128854
0 references
0.86830044
0 references
0.86759114
0 references
0.86721134
0 references
0.8670988
0 references
The nine-point equations characteristic of finite element discretization over a rectangular grid are accurate to \(O(h^ 2)\). Thus, when neighbors are expanded in Taylor series about a central point truncated at \(O(h^ 2)\) the partial differential equation is recovered. The new method described in detail here uses quadratic rather than linear interpolation so that the local accuracy remains \(O(h^ 2)\). The difference equations of the factored matrix \(B\) are thus ``consistent'' with \(A\) in the sense that local truncation error is the same. This motivated the name ``consistent sparse factorization'' (CSF). One might jump to the somewhat optimistic conclusion that solution of the CSF equations is as accurate as that of the standard equations and that CSF essentially is a means for efficient direct solution. Unfortunately, this is not the case. Whereas in the alternative sparse factorizations it can be shown that the coefficients are bounded, this is not so for CSF. The coefficient growth in CSF is a manifestation of loss in accuracy in representation of the higher modes of \(A\). Fortunately, only the inverse of \(B\) is used in the iteration. Thus, coefficient growth does not lead to a corresponding growth in the error but rather to negligible reduction of high-mode error components. Numerical studies reported by \textit{A. J. Baker, W. P. Norohna} and \textit{E. L. Wachspress} [Appl. Math. Lett. 2, No. 3, 215-220 (1989; Zbl 0705.65022)] demonstrated how relatively few CSF iterations without acceleration sufficed to solve a few sample test problems. These problems did not require reduction of high-mode error components. Hence, the new method performed extremely well when compared with commonly used methods. Additional studies have now been made on problems with significant higher-mode effects. These studies exposed shortcomings of CSF which where then resolved by appending a simple iteration to reduce high-mode error components. All studies were performed with Laplacian operators over rectangular grids. Both uniform and highly graded grids were examined. For the entire collection see [Zbl 0778.00018].
0 references