Strange behaviour in fitting a sum of gaussians

Hi everyone!

I have some crude behaviour in performing a fit under different versions of root (5.10 Mac binary release and a version 4.something on a Linux-Machine), while the same code gives me completely different results of fit propability (some 20-40% on the Linux 4.xx and exactly 0% on Mac 5.10)

So my main question is, if there are any known bugs or other differences between the versions that I’ve overseen.

Sorry that I am not able to be more specific on the version and exact sourcecode, but unfortunately I don’t have access to the Linux box any longer and since that time changed my sourcecode in order to get at least some propability of fit - until now without any success on my Mac - and forgot to keep a backup of the original source.

You’ll find a sample datafile and the actual sourcecode I am using attached.
(Since I’m a newbee to C++ and ROOT do not wonder if some things are far from being elegant)

Thanks in advance and happy Easter,

fitterHappier.c (2.66 KB)
s2.txt (8.26 KB)

Thanks for reporting this problem. Now fixed in the CVS head.
The problem happens when the Likelihood option is specified.
The computation of the chisquare equivalent was not taking bin errors correctly in this case.



You guys are fast!