Author |
Message |
Michael Köhler (madcat)
Junior Member Username: madcat
Post Number: 3 Registered: 11-2007
| Posted on Tuesday, November 13, 2007 - 03:31 am: | |
Hi I simulate magnetic fields with FlexPDE. The permeability of some materials I import with table-files. Yesterday I saw, that the numbers of data in the table-files have a effect for the calculation-time. If I use a table with 28 data points FlexPDE need a hour to solve the problem. If I use a table-file with 1000 data points FlexPDE needs for the same problem only 15 minutes. Now my Question is, what are the optimum for data points? Is this dependent by the problem? |
Robert G. Nelson (rgnelson)
Moderator Username: rgnelson
Post Number: 992 Registered: 06-2003
| Posted on Tuesday, November 13, 2007 - 01:22 pm: | |
FlexPDE uses Newton's method to solve nonlinear systems (and, in fact, linear systems are solved with one step of Newton's method). In each iteration, the step is extrapolated along the slope of the functional to find a new trial solution. If you have too few points in your table, you may be getting "V" shapes in the data, causing Newton's method to extrapolate past the vertex and find itself hanging in the air. This would cause long iteration sequences and long running time. With more points, the real shape of the data function may be more smoothly resolved, allowing Newton's method to proceed more efficiently to the functional minimum. The optimum number of table points is therefore the number of points necessary to resolve the curvature of the data. |
|