Architecture Problems

Architecture and Version Problems

I recently did two things related to hardware system architecture and code versions: create Solaris executables and bring over the latest arve from GLAST. Both of these were done last week (~7/7/98).

Joanne quickly tried the Solaris version (since those are the zippiest SLAC unix boxes) and found strange behviour: it was as if pions were not interacting (though it seems they manage to deposit all their energy, unlike muons).

Particle Response

A new version of arve was just released from GLAST, so I grabbed that and created a new nld version on NT only. I proceeded to run 100 particles through the various versions and got the following results, where the first number in the table is the mean energy and the second is the rms (s) energy

100 5 GeV particles	Old Arve NT	New Arve NT	Sun	AIX
e	4.353, 0.325	4.351,0.332	4.358,0.384	4.322,0.338
mu	1.649,8x10^-4	1.649,8x10^-4	1.649,8x10^-4	1.649,8x10^-4
pi	3.884,0.969	3.033,0.912	4.600,0.901	3.952,1.151

Assuming the randoms sequences are not the same on all machines, I conclude that the electron and muon response is consistent across all versions and platforms. The Solaris result is quite different from the others in that the energy is much higher, and the distribution is not pion-like (it is way too narrow). I have faint hopes that the Solaris behaviour is due to the hurried build of the libf2c library by Tom Pavel (Toby doesn't see how this can do anything - he is probably right). Tom has promised to remake it this week when he can find a moment between BABAR meetings. There is a hefty difference between the old and new arve versions.

Toby tried one of the GISMO tools (shower.exe) in the new arve and found that it reported back for all the incident energy, though that does not necessarily account for visible energy. I tried the old arve version here, but it crashed. I guess I should try harder! Other than that I have no ideas for how to account for the energy difference - this remains an open wound!

Error Conditions

We are also seeing a number of warning and error messages, which come in two groups: one is for particles stoppping due to exceeding MAXSTEP; the other is related to geometry and tolerances.

stoppping: I looked a 3 of these and they were all very low energy nucleons. In practice, not a problem, but in our copious spare time we might want to see how to fix this.
tolerances: again these break down into two errors classes: assertion failures in Intersection.cxx and confusion (the code claims the particle is leaving the World, but that it is still in some known named volume)
- Joanne and I noticed that the confusion errors all seemed to come at the outer radius of the coil, which coincides with the inner radius of the HAD calorimeter. I tried a test of moving the HAD back 1 cm, and the error messages went away. Joanne has a test job which has the assert failure; she will try this geometry mod on it.
- another disturbing feature is that I found that what I thought were cosmetic recompiles affected the energy distribution and errror messages, though this may depend on how the random number sequences are initiated. If they are always the same, this is a bad sign; if they are different for each run, then this is OK.

R.Dubois Last Modified: 01/13/04 12:57