Ciao,
I'm trying to understand the why one of my fits fails using
ROOT 3.00/06 and the TMinuit/TFitter class. I have two questions:
1. Under what conditions does MIGRAD actually determine it will
quit? (other than the call limit)
2. In the context of my situation, I'm seeing this sequence of messages
when I call TFitter::ExecuteCommand("MIGRAD",arglist,2):
//
**********
** 3 **MIGRAD 3.8e+04 1
**********
START MIGRAD MINIMIZATION. STRATEGY 1. CONVERGENCE WHEN EDM .LT.
1.00e-03
FCN=35111.1 FROM MIGRAD STATUS=INITIATE 76 CALLS 1037 TOTAL
EDM=65193.8 STRATEGY= 1 ERR MATRIX NOT POS-DEF
.
.
.
MIGRAD MINIMIZATION HAS CONVERGED.
MIGRAD WILL VERIFY CONVERGENCE AND ERROR MATRIX.
EIGENVALUES OF SECOND-DERIVATIVE MATRIX:
-5.2169e-01 9.5659e-02 1.4906e-01 1.9453e-01 2.6754e-01
2.7675e-01 3.2096e-01 3.6102e-01 4.4657e-01 4.8688e-01 5.2103e-01
7.0896e-01 7.7633e-01 8.2491e-01 8.4358e-01 9.0652e-01 9.1783e-01
9.3161e-01 9.3667e-01 9.6279e-01 9.8268e-01 9.8491e-01 9.9168e-01
1.1732e+00 1.1881e+00 1.2584e+00 1.2805e+00 1.3221e+00 1.5238e+00
1.5253e+00 1.6193e+00 1.7037e+00 1.7502e+00 1.7898e+00 1.8261e+00
1.8920e+00 2.1636e+00 2.6171e+00
MINUIT WARNING IN HESSE
==============
FCN=33792 FROM HESSE STATUS=NOT POSDEF 892 CALLS 11151 TOTAL
EDM=0.00162124 STRATEGY= 1 ERR MATRIX NOT POS-DEF
[ my parameter list, etc. ]
MIGRAD FAILS TO FIND IMPROVEMENT
MIGRAD TERMINATED WITHOUT CONVERGENCE.
FCN=33792 FROM MIGRAD STATUS=FAILED 10450 CALLS 11411 TOTAL
EDM=0.146157 STRATEGY= 1 ERR MATRIX NOT POS-DEF
//
Why does MIGRAD announce failure? I didn't ask it to call
HESSE; does it do that normally in order to "verify convergence
and error matrix"? I'm not used to MIGRAD doing that on its own,
when using MINUIT outside of ROOT. Things look fine until immediately
after the HESSE call, I get the "terminated without convergence." What
happened?
I know its hard to diagnose this without the context of my
particular fitting problem, but I'm just asking what conditions cause
the messages above. They don't appear to be consistent in my
private picture of how MIGRAD does its job...
Thanks,
TIMeyer
Stanford University
This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:50:37 MET