back to list

Wedgie error -- getting closer

🔗Graham Breed <gbreed@gmail.com>

3/17/2008 7:23:59 AM

Anybody remember this from Gene?

/tuning-math/message/12188

I think he was almost right. There's a formula

E = ||W^j|| = ||W^(j-u)|| <= 3 ||W|| ||j-u|| = 3Ce

which is basically correct. If C is scalar complexity, e is TOP-RMS error. Provided the ||...|| norm is the measure of the wedgie, as I use for scalar complexity following Browne. Gene didn't like this for some reason and so took the max-abs value instead, which didn't seem to mean anything.

Anyway, I hit on a similar formula from a different logic and now I've tracked down Gene's original post. I'll try to nail it down further. That "3" is clearly related to the "5" that came up in they weighted wedgie problem I posted about before. ||W^j|| does seem to be scalar badness give or take a constant that depends on the rank and dimension of the problem.

Graham

🔗Carl Lumma <carl@lumma.org>

3/17/2008 9:22:44 AM

At 07:23 AM 3/17/2008, you wrote:
>Anybody remember this from Gene?
>
>/tuning-math/message/12188

I do remember that. But I think you got the wrong link.

-Carl

🔗Graham Breed <gbreed@gmail.com>

3/17/2008 7:11:56 PM

Carl Lumma wrote:
> At 07:23 AM 3/17/2008, you wrote:
>> Anybody remember this from Gene?
>>
>> /tuning-math/message/12188
> > I do remember that. But I think you got the wrong link.

Yes, sorry

/tuning-math/message/14189

Graham