Re: [ATM] FigureXP - monte carlo analysis

["Bob May" <rmay@nethere.com>]

Systematic errors like making all of the readings several
thousandths of an inch off in one direction won't matter.  The
end error is so small that the shape of the surface will
completely wash them out.  Two thou over the hundrred inchs or so
is a real small error.
Bob May

rmay at nethere.com
http: slash /nav.to slash bobmay
http: slash /bobmay dot astronomy.net

----- Original Message -----
From: Mark Cowan <toolontop@yahoo.com>
To: <atm@atmlist.net>
Sent: Thursday, May 15, 2008 10:08 PM
Subject: Re: [ATM] FigureXP - monte carlo analysis


> Hey Mitch,
>
> The Monte Carlo analysis doesn't account for systematic bias in
testing,
> unconscious or otherwise.  If you add significant systematic
bias to a random
> distribution you still have systematic bias, thus unrecognized
errors.  In such
> cases (which for various reasons I think reflect the majority
of manual
> measurements of mirrors) there's no "guarantee" at all - just a
WAG.  Perhaps a
> slightly better informed WAG, though. ;)
>
> If you do want to use the one in FigXP and you have a lot of
data sets (the
> only way it's useful) note that the readings need to be
normalized first - ie,
> if you setup more than once, bumped the tester, or thermal
drift changed the
> baseline you need to manually adjust the numbers so that they
all look like the
> same session - otherwise its calculation is goofy.
>
> I'd like to hear other opinions on this, these are just my, uh,
random
> thoughts.
>
> Best,
> Mark
>
> --- Mitchell R <funnybone101@embarqmail.com> wrote:
>
> >
> > Yes it gives a percent probability that the values are
> > within certain limits.
> >
> > See
> > http://en.wikipedia.org/wiki/Standard_deviation
> >
> > Scroll down to see the graph is section 3.3
> > Rules for normally distributed data
> >
> > Dale Eason
> >
> >
> >
> > Hey Dale,
> >
> > Sill me! I just studied this is math earlier in the year; I
should have
> > recognized that it is evenly distributed data. So +/- 1 means
a 68.2% chance
> > your mirror is within the said numbers and +/- 2 means a
95.4% chance your
> > mirror is within the specified vales.
> >
> > So roughly 70% and 95%. Your mirror, if you select the
calculated standard
> > deviation of your readings for the Monte Carlo analysis
(calculated by
> > FigureXP if you put in two or more data sets), is almost
guaranteed to be
> > within +/-2 and there is a very good chance it is within +/-
1. So, roughly
> > speaking, if one wants confidence of better than 1/4 wave,
they should look
> > for the +/- 2 sigma results to be better then .8 Strehl.
> >
> > Thanks again everyone, very helpful in teaching the youngin'
as usual ;)
> >
> > Clear Skies,
> >
> > Mitchell
> >
> >
> >
> >
> > _______________________________________________
> > ATM mailing list http://www.atmlist.net/
> >
>
>
>
>
> _______________________________________________
> ATM mailing list http://www.atmlist.net/
>

_______________________________________________
ATM mailing list http://www.atmlist.net/