From: Joseph Comfort (comfort@megpc1.la.asu.edu)
Date: Mon Feb 12 2007 - 00:28:38 CET
Dear Giuseppe,
Thank you for pointing out the bug regarding negative C in log(C).
That was the problem, and I did some apprpriate recoding. (It turns
out that E2 was not needed in that case; only E3.) I had copied code
from someone else's Geant4 application, so MC comparisons could be
made. I'll let them know. I guess traps have to be set explicitly
in order for such errors to be flagged.
Thanks for the note on the random number generator. I didn't know
the name of the function in Fluka. Using rand(0) was convenient
because I could check the algorithm with a stand-alone program,
and the quality was a minor issue for me here.
I initially ran the calculations on an x86_64 machine with SuSE 10.1,
kernel 2.6.16 (rev. 27), compiling the Fluka application with gcc/g77
3.3.5. I then tried a i686 machine that has SuSE 10.2, kernel
2.6.20_rc5, with the same gcc/g77. (SuSE 10.2 comes with kernel
2.6.18, but I picked up a later, unofficial version, to get around
a bug of some sort--not successfully, unfortunately.) In both cases,
gdb, versions 6.4 and 6.5, did not recognize the core dump as being
a core file. I then tried a i386 (PIII) machine on which I still
had SuSE 9.3, kernel 2.6.11 (rev. 4). The only thing I could get
from gdb, version 6.3, was the lines I put in the first message.
Only gcc3/g77 is available on SuSE 9.3. The 10.x distributions
install with gcc4/gfortran. Although the gdb is said to have
partial support for g77 in these latter cases, that apparently is
not enough.
SuSE was close to not making gcc3/g77 available at all, but several
users got a reprieve for awhile. I believe g77 is no longer being
supported, and all work is going into gfortan. It is pretty stable
now, and I have not had any problems, after making suitable adjustments
(e.g., for the Cernlib). Other distributions are also moving to
gfortran. It would be good to have a parallel gfortran version of
Fluka.
As a final note, the output I am getting from the simulation is
puzzling to me. It is just a simulation of a beamline collimation.
Although the beam is (or is intended to be) cylindrically symmetric
I am seeing a slight drift towards positive X as it goes through
the collimator. Not much, but well outside statistical errors,
I think. I had been seeing this is some other such simulations.
I'm examining the issue and if I can confirm that it is not some
bias on my part, I'll make a separate issue. In the meantime,
any ideas of things to look at would be appreciated.
Thank you,
Joe Comfort
This archive was generated by hypermail 2.1.6 : Mon Feb 12 2007 - 09:01:15 CET