Rene et al.,
I'm having a problem with TMinuit. I have a fit with 5 parameters which work. Then I added a 6th parameter and it fails when I try to fit that parameter.
I found one problem in that I for got to change new TMinuit(5) to new TMinuit(6), but with that fixed the same kind of problem occures.
In the debugger I see that the parameter (which is a background level) when I'm fitting only for it looks ok for the 1st couple of calls, but then minuit inputs par[5]=-20 which makes my PDF negative and my logLike an NaN .. so the fit fails.
Here are 1st three calls.
Breakpoint 2, vvfit::funcToMin (this=0x9de8ef0, npar=@0xbffe0324,
gin=0x9dfd008, f=@0xbffe03d0, par=0x9dfcae8, flag=4) at vvfit.cc:196
196 f=-logLike;
$4 = 0.10114172087954518
$5 = -132589.34477643354
(gdb) c
Continuing.
Breakpoint 2, vvfit::funcToMin (this=0x9de8ef0, npar=@0xbffe0324,
gin=0x9dfd008, f=@0xbffe03c8, par=0x9dfcae8, flag=4) at vvfit.cc:196
196 f=-logLike;
$6 = 0.098858279120454834
$7 = -130241.44136714273
(gdb) c
Continuing.
Breakpoint 2, vvfit::funcToMin (this=0x9de8ef0, npar=@0xbffe0194,
gin=0x9dfd008, f=@0xbffe0268, par=0x9dfcae8, flag=4) at vvfit.cc:196
196 f=-logLike;
$8 = -21.289116679404454
$9 = nan(0x8000000000000)
(gdb)
I don't have debug sysmbols for minuit so I can't see what's genrating this crazy step. The step size a specified in mnparm was small --- 0.01.
I'm a little supprised that 1st call is for 0.10011417 when starting value of fit is exactly 0.1. The values of this parameter are sensible when I fit for other parameters and indeed I get the correct answer for other parameters only when I've set this one to the correct value (0.1).
Doubteless I'm still doing something stupid like TMinuit(5), but I can't see what it is?
Thanks for any help/adivce you can give.
-Art S.
A.E. Snyder, Group EC \!c*p?/ SLAC Mail Stop #95 ((. .)) Box 4349 | Stanford, Ca, USA, 94309 '\|/` e-mail:snyder_at_slac.stanford.edu o phone:650-926-2701 _ http://www.slac.stanford.edu/~snyder BaBar FAX:650-926-2657 CollaborationReceived on Tue Dec 05 2006 - 01:51:10 MET
This archive was generated by hypermail 2.2.0 : Mon Jan 01 2007 - 16:32:02 MET