Re: A detail in TMinuit MIGRAD call to FCN

From: Rene Brun (Rene.Brun@cern.ch)
Date: Fri Oct 30 1998 - 18:48:03 MET


Chris Jillings wrote:
> 
> Hello,
>     This point is a fine detail, but it may be important to me.
> >From reading Fred James'reference manual for version 94.1 of Minuit, I
> find on page 9 a list of iflag values used in the call to FCN. It has I
> flag = 1,2, and 3 (see page 9). It seems that FCN is called with iflag=1
> for startup. So I test on iflag, as in the example, and so some
> initialization there.
> 
>     But when I run TMinuit, I get the following output (having sent print
> level to 3.)
>  **********
>  **    4 **MIGRAD         500           1
>  **********
>  FIRST CALL TO USER FUNCTION AT NEW START POINT, WITH IFLAG=4.
>  START MIGRAD MINIMIZATION.  STRATEGY 1.  CONVERGENCE WHEN EDM .LT.
> 1.00e-03
>  FCN=-0.519027 FROM MIGRAD    STATUS=INITIATE       21 CALLS          22
> TOTAL
> 
> What does IFLAG=4 mean? "First call to use function" suggests to me that I
> should test on iflag==4 to do my initializations.
> 
> Thank you for any help.

The convention used by MINUIT is the following:

 - IFLAG=1 when FCN is called for the first time.

 - IFLAG=2 if you have specified the option to calculate GRAD,
      the first derivatives (optional)

 - IFLAG=3 call to FCN when the fit is finished.

 - IFLAG=4 normal calls to FCN during the minimization process

The Minuit documentation at page 9 shows an example of FCN
with tests on IFLAG=1,2 or 3. IFLAG=4 is not explicitly quoted,
but from the simple example, on can deduce that this is the normal 
call.

Rene Brun



This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:34:39 MET