From: Paul
Date: Fri, 20 Oct 2000 23:43:56 +0100
Subject: Re: [TSSP] Progress report 18th Oct 2000
Terry wrote: > "I" would like to see the capacitance per unit length graph and > it's bathtube curve. OK, I've added another section to pn1710 in http://www.abelian.demon.co.uk/pn1710/ Current version is now 0.2b. > Can you do more than 32 points? Yes, if necessary, but since Cint is 2D, the time taken to compute the capacitance matrix is proportional to N^2. The number 32 of cylindrical electrodes into which the coil is broken for capacitance matrix estimation is a compromise between speed and accuracy. I figured 32 would give about 3% error in resolution, therefore 1.5% error in Fres. It takes between 2 and 4 pentium-cpu-days to produce the Cint and Cext matrices from a coil description, about 4hrs on the cluster. The new solver will require only a third the cpu time, but I think I'll use that gain for a speed up rather than more resolution in order to explore lots of geometries. > I worry that the top and bottom numbers may get "averaged" > excessivly from their true spikyness. I agree. Nothing much happens to Cext along most of the length so maybe a varying segment length could be applied but this would be tricky to add to Cint. I wouldn't want to try it without some evidence that the existing lack of resolution at the ends was causing a problem. With Cint, which is a function of two points on the coil, the resolution is also lacking when the points are close together. This causes the flat top to the example slice through Cint in pn1710. Cheers, -- Paul Nicholson, Manchester, UK. --
Maintainer Paul Nicholson, paul@abelian.demon.co.uk.