Introduction
(1)
The value of c2 is minimized with respect to c. This
object is not to find the minimum value of c2. The object is to find the fA that
most closely approximates the function underlying fi. The data generated in ArtDat\Welcome.htm has a Gaussian distribution
about the underlying value of fA with a standard deviation given by
(2)
The data is random, but the error reported is exact, that is
it uses fA as in (2). There are derivations that indicate that
almost all small signals have a Poisson distribution with errors proportional
to the size of fA. Poisson/Welcome.htm so the value of berr
is most naturally set to be such that berrfA is the
number of statistically independent counts in each channel.
This leads to four problems.
- If
equation (2)
is inserted into equation (1) which is then
minimied with respect to c, there is a term from the denominator
that in rare cases can dominate the minimization Loophole.doc. When the object is to find the best
estimator in a Poisson distribution, equation (1)
is not the correct expression to minimize [Poisson\Fitting
with Poisson random variables.doc], but repeatedly minimizing (1)
without including the variation in (2)
usually does lead to the correct minimum. – For an example that does not
quite find the exact minimum see FermiFit\Welcome.htm. The best procedure is
a) Start with a value of fA in (2)
and find all e(xi)
values.
b) Minimize (1)
holding {e}
fixed to find a better estimate of fA.
c) Use
the better fA in step a)
This will not completely minimize c2,
but it will find the fA desired.
- Zero
counts are poorly represented by this distribution and even 1 count does
not have a Gaussian distribution. A “solution” to this is to make aerr
equal to 1 – This is the strategy used in ..\robfit\Welcome.htm. This makes the terms in c2
à
fA2 in the region where there are zero counts
leading to a total c2 < N by
approximately the number of channels with zero counts.
- A peak
with a few million counts so dominates the value of c2
that changes in this peak that may not even be visible when plotted far
outweigh the changes of interest at very low count rates stanfit3.doc#Practical_Curve_Fitting. The solution to this is to make cerr
~ 0.01 for “graphical” accuracy even though the underlying statistical
errors are actually Poisson. This
has the same effect as the “solution” above in that counts near large
peaks will have values of (fi-fA(xi))/(Ö(berrfA(xi))+cerrfA(xi))
smaller than 1 due to the second term in the denominator.
- Small
values ei
in (1)
produce local minima for which a single feature has to be preserved while
others are optimized. It is much
easier numerically to minimize c2 with large
values of. A series of runs with
successively smaller values of aerr, berr, and cerr
is much less likely to be stalled by local minima.
Folders
01/31/2008 07:03a
<DIR> ArtDat\Welcome.htm – Considers the generation of
artificial data based on a standard direction file.
02/25/2008 07:45a
<DIR> Poisson/Welcome.htm
Poisson – Details about the Poisson distribution
Documents
02/27/2008 10:35
AM 35,840 FindingTheErrorConstants.doc
02/04/2008 07:25
AM 79,872 Expectation values.doc
02/04/2008 04:32
PM 199,168 SimErr.doc
02/04/2008 04:39
PM 132,608 Copy of
Estimating the error per point.doc
02/08/2008 08:09
AM 224,256 Estimating the error per
point.doc
02/25/2008 05:12 PM 30,208 Loophole.doc