sPHENIX discussion of software

[pinkenburg <pinkenburg AT bnl.gov>]

Hi folks,

so far we had a bit of a mess when it came to seeding our random number 
generators with the default being more or less that one had to set a 
random seed in the macro to avoid getting identical results.

I went through our G4 interface and our single particle generators and 
replaced the various seedings by something more manageable. By default 
each random number generator (including the one used in G4) gets its 
seed now from /dev/urandom (I know this is not available everywhere - 
once we switch to c++11, there is a std::random_device or take something 
from boost if this is a really urgent issue). That code is in a single 
place (offline/framework/phool/PHRandonSeed.cc) so it's easy to adapt. 
If one sets the RANDOMSEED flag its value will be used as the seed for 
all random number generators.

That change turned out to be a bit more invasive for the single particle 
generators than I initially thought. I removed the various random number 
generators (TRandom3, rand()) from the implementations and added a gsl 
random number generator (Mersenne Twister - just like TRandom3) to the 
particle generator base class (PHG4ParticleGeneratorBase). Just try to 
figure out how TRandom3 sets its seed from the source code and you'll 
see why we shouldn't use that one - it also seem to behave differently 
when used under CINT vs compiled code. Interestingly while the uniform 
distributions are absolutely identical using the same seed (it's the 
same algorithm after all), derived distributions (Gaussian, 
exponential,spherical) are very similar but not identical and I just 
would trust gsl more than root. The only exception was the 
PHG4ParticleGeneratorVectorMeson which uses a Breit-Wigner distribution 
which doesn't exist in gsl.

Some of the single particle generators did not use the base class and 
had lots of duplicated code - I modified them to use the base class and 
removed duplicated methods/data members.

All of this made small changes in the Fun4All_G4_sPHENIX.C macro 
necessary and the old macro will not work with the newest build (3 
subsequent lines 141-143 need to change)

from:
gen->set_embedflag(1);
gen->set_seed(uniqueseed);
gen->set_verbosity(0);

to:
gen->Embed(1);
gen->Verbosity(0);

and you might want to take out the RANDOMSEED gymnastics but that's not 
required.

Chris


--
*************************************************************

Christopher H. Pinkenburg       ;    pinkenburg AT bnl.gov
                                ;    http://www.phenix.bnl.gov/~pinkenbu

Brookhaven National Laboratory  ;    phone: (631) 344-5692
Physics Department Bldg 510 C   ;    fax:   (631) 344-3253
Upton, NY 11973-5000

*************************************************************