Minutes of Technical Forum @ TRIUMF
(Draft 1.1 - Oct.02 M.Asai)

15:30 - 17:30, September 5th, 2003 (Friday)
The TRIUMF auditorium

[1] Recent and current developments (J.Apostolakis)

[2] Release plan (G.Cosmo)

[3]Discussion on new requirements

[3.1] Requirements on new functionalities

Req.0101 Access to the Track properties before the processes
Requester : G.Santin (ESA/ESTEC)
Responsible G4 developer : H.Kurashige

Requested additional access method before DoIt() invokation for all
processes was evaluated as likely to create performance overhead.
Hisaya recommended using a wrapper class for use with any process.
(action TBD)
Hans-Peter proposed to add new option in the hadronic processes to
enable this for hadronic processes ? which is expected to have only
small performance impact for typical use cases.

Req.0102 Development of X-ray surface processes
Requester : G.Santin (ESA/ESTEC)
Responsible G4 developer : P.Gumplinger

Peter will remain in contact with Giovanni Santin for collaborating
in the development.

Req.0103 Geometry construction: input from external models
Requester : G.Santin (ESA/ESTEC)
Responsible G4 developer : G.Cosmo

Several potential developments were presented, but a relative
priority was not established, as well as concerning resources
to be assigned to them for possible future projects.
A discussion at the architects forum and Steering Board will be
brought forward.

[3.2] Requirements on robustness / testing / validation

Req.0104 Ensure that events are reproducible between debug and
optimized modes
Requester : F.Ranjard (CERN/LHCb)
Responsible G4 Developer : J.Apostolakis

Observations: Some Unix compilers (eg Sun) appear to provide such
consistency between debug and optimized modes. However the 80-bit
hardware of the FPU of x86 contrasts with the size of doubles stored
as 64-bits in memory. Changes in storing of intermediate results
can thus lead to small differences in the results, summing up until
they can influence a Monte Carlo choice.
Another source of differences is the initialization of variable in
debug mode. Such changes should be detected using appropriate tools
(eg valgrind) and fixed.
Actions: Investigate use of combined (-g -O on Linux GNU-GCC) for
addressing underlying debugging issues. (LHCb)

Req.0105 Handle unexpected conditions more robustly
Requester : F.Ranjard (CERN/LHCb)
Responsible G4 developer : J.Apostolakis

Due to lack of complete information this was not addressed. It was
postponed until the next forum.

Req.0106 Setup statistical test suite for most sensitive physics quantities
Requester : R.Debois (SLAC/GLAST), et.al.
Responsible G4 developer : J.Apostolakis

It was well understood that Geant4 should establish a statistical test
suites for Geant4 releases, extending the current testing undertaken at
different development stages. John Apostolakis will organize a task force
inside the collaboration in the coming weeks. It is clear, however, that
with the limitation of finite resources, it is impossible to cover all
physics use cases and the effect of all physics processes at all energy
range or materials. The collaboration encourages users to undertake their
own acceptance testing for their particular use and domain.
The collaboration noted positively the user communityfs wish that it should
publicize its test suite, for the use of users themselves.

[3.2] Requirements on user support

Req.0107 Installation kit which contains all packages
Requester : B.Arjomandy (Loma Linda University Medical Center)
Responsible G4 developer : G.Cosmo / G.Folger

Because of the copyrights and the large variety of potential installations,
it is not easy to put everything into one piece.
The Geant4 collaboration undertook to study changes to ease installation and
its instruction to the degree possible.

Req.0108 Creation of user group / discussion forum in medical field in
particular in dosimetry
Requester : E.Brauer-Krisch (PSI), et.al.
Responsible G4 developer : M.G.Pia

Possibility of having new threads in HyperNews is discussed.
Also, it was agreed that Geant4 collaboration should promote any possible
information exchange among users.

[4] Action list, summary and next Forum (M.Asai)

Next proto Technical Forum will be held at CERN on October 7th. John
Apostolakis will make an announcement with details.

Meeting Date: 

Thursday, September 4, 2003