Re: ATM Plop Mirror cell optimisation

["A&S Hincks" <hincks@bigpond.net.au>]

Hello David

Essentially, unless I misunderstand badly (and believe me I frequently
misunderstand :-)) I was testing what you describe below although only as an
end to making sure that I got the most accurate result possible.

I increased the 'resolution/fineness' of the mesh with the aim of checking
that the error decreased/tended towards to a single value. I plotted the
results on a graph (which I can't post here) and hoped I could work out the
value of each support radii by visually looking at the value that the curve
tended to.

What surprised me was that as the resolution of the mesh increased (ie 20,
21, 22.....n) the values of the support radii did not necessarily tend to
any particular value.

For an example of what I am talking (hopefully not babbling) about,
referring to the (very poorly formatted) data table in my original post, for
support radii 4 (ie. r4), where n-mesh rings = 20, r4 = 0.757141. However
where n-mesh rings = 40 and 41, r4 = 0.738209 and 0.743596 respectively.

What I am trying to illustrate (again poorly) is that if I merely ran an
anlalysis at n-mesh = 20 and 40, I would have thought that 0.738209 is
closer to the optimal result. But the value n-mesh rings = 41 @ 0.743596 is
a proportionately a rather large incremental jump back in the other
direction.

Thanks

Andrew



----- Original Message -----
From: "david lewis" <lewis@eecg.toronto.edu>
To: <atm@shore.net>; <hincks@bigpond.net.au>
Sent: Thursday, February 07, 2002 12:01 PM
Subject: Re: ATM Plop Mirror cell optimisation


>
>
> This is an interesting study that I've wanted to do but never bothered.
>
> What is interesting is if you take the result with your smallest grid,
> which is 20, and rerun Plop NOT doing optimization on the largest
> grid, to see what the difference in error is.
> This will tell you how much extra error you're getting by not using the
finest
> grid possible.
>
> David Lewis
>