COMPUTING ERROR BOUNDS WHEN THE TRUNCATION ERROR AND PROPAGATED ERROR ARE OF THE SAME ORDER.
Abstract
In a previous report, a program for computing bounds for the error in a numerical solution of a system of linear differential equations is given. Experimenting with this program led to two observations: (1) In some cases a much smaller integration step size than expected would be required for several steps at the start. (2) The bound was unduly large relative to the actual error for the integration steps at the start of the program. These features, while undesirable, are not critical, in general. The reason for this behavior is discussed in this report. A program is given which can overcome these undesirable features of the program. Also given is a complete account of the equations on which this program is based. Included in this report are instructions for the use of this program and the results of a number of sample problems. These results illustrate the reasons for this program and they enable the user to check his interpretation of the instructions for the use of the program. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1970
- Accession Number
- AD0709877
Entities
People
- Charles L. Keller
- Thomas E. Reeves
Organizations
- Air Force Research Laboratory