Re: Re: [Gate-users] Some news about the strange behaviour

Tue Feb 5 09:35:35 CET 2008

Dear Gate users and developers,

At the end of last year, we reported strange behaviour in the energy spectra of singles and coincidences when using the newest version of Gate (3.1.2). Since then, we have recently received a cluster system in our lab and would like to install Gate on this new system. Has the problem with Gate 3.1.2 been identified and solved?

Thanks in advance,
Jasmine Schirmer and Irene Torres

Original Message:

Hi Yu Chen,
 
 Thanks for your comments !
 
 2007/12/21, Chen, Yu :
 >
 > Thank Simon for thorough investigation.
 > However, it seems that the thing does not become clearer but even more
 > confusing. See below.
 >
 > Simon Stute wrote:
 > > Hi all,
 > >
 > > I dug more to find the reason of the strange behaviour in energy spectra
 > > and found those results :
 > >
 > > - I first installed Gate3.0.0, 3.1.0, 3.1.1 and 3.1.2 on both 32 and 64
 > > bits architectures,and both cluster and single PC.
 > > - I ran PET and SPECT benchmark on all configurations and compared both
 > > energy spectra and results from benchmark.C with root : everything was
 > > identical between all different config.
 >
 > 1, It suggests that either no significant difference between GATE versions
 > or
 > benchmark configuration is not able to reveal any difference.
 >
 
 Well ... perhaps some new benchmarks have to be created ...
 
 > - Then I ran the single_layer.mac simulation on all config and obtained
 > > the same difference in energy spectra as Irene and Jasmine reported.
 >
 > 2, different behaviors come from different comfigurations.
 > I found that major differences between benchmarkPET.mac and
 > single_layer.mac are:
 >
 > 1) physics list,
 > benchmarkPET:
 >
 > # EM P R O C E S S
 > /gate/physics/gamma/selectRayleigh lowenergy
 > /gate/physics/gamma/selectPhotoelectric standard
 > /gate/physics/gamma/selectCompton standard
 >
 > # I N A C T I V E S E C O N D A R Y E L E C T R O N S
 > /gate/physics/setElectronCut 30. cm
 >
 > # I N A C T I V E X - R A Y S
 > /gate/physics/setXRayCut 1. GeV
 > /gate/physics/setDeltaRayCut 1. GeV
 >
 > /gate/physics/verbose 2
 >
 > single_layer:
 >
 > /gate/physics/gamma/selectPhotoelectric standard
 > /gate/physics/gamma/selectCompton standard
 > /gate/physics/gamma/selectRayleigh inactive
 >
 > # High cuts for fast simulation
 >
 > /gate/physics/setElectronCut 3. m
 > /gate/physics/setXRayCut 1. GeV
 > /gate/physics/setDeltaRayCut 1. GeV
 >
 > /gate/physics/verbose 0
 >
 
 Yes, I notice that too and think that the problem may come from here.
 Because here Irene and Jasmine use the standart package instead of more
 appropriate low energy package in this case. So maybe there were some
 changes in geant4 versions in the standart way that create this problem.
 
 
 2), setDepth 1 for benchmarkPET and 5 for single_layer.
 >
 > > Then I found that the change in energy spectra occurs between Gate3.0.0
 > > and 3.1.0. Results from 3.1.0, 3.1.1 and 3.1.2 were the same. (see
 > > attached images)
 > > - I realised then that between Gate3.0.0 and 3.1.0, the associated
 > > Geant4 versions are 8.0.p01 and 8.1.p02. And between these two different
 > > Geant4 versions one of the main changes was the upgrade of the most
 > > important data package G4EMLOW, that move from 3.0 to 4.0, and that
 > > includes major changes.
 >
 > 3, G4EMLOW is a data lib used by EM Low Energy model in Geant4.
 > As we can see, only benchmarkPET used low energy Rayleigh and single_layer
 > used
 > all standard model physics processes for PE and Compton. Any change in
 > G4EMLOW
 > should only affect benchmarkPET not single_layer.
 >
 
 Yes I totally agree with that but I just change the EMLOW package directory
 in the Geant4.8.1.p02 and it solves the problem, so it is strange ...
 
 Any major change of standard EM model between Geant4 8.0.p01 and 8.1.p02?
 >
 
 I don't know about that. We have to investigate in this.
 
 > - I decided then to run the single_layer.mac simulation with Gate3.1.0,
 > > Geant4.8.1.p02 and to change the G4EMLOW4.0 package with G4EMLOW3.0
 > > package from the older version. Then the energy spectra obtained was
 > > more similar than this obtained with Gate3.0.0, but still different (see
 > > energy_3.1.0.EMLOW3.0.jpg image). So I think this little difference is
 > > due to other main changes between Geant4.8.0.p01 and Geant4.8.1.p02, it
 > > is statistical I think.
 >
 > 4, This is really confusing!
 >
 
 Yes, see above.
 
 > - The change in behaviour do not come from any digitizer module (adder,
 > > readout, any blurring ...), they are still the same since Gate3.0.0 and
 > > even before, I compared all files and algorithms.
 >
 > 5, This may be true. But as parameter setDepth is different, the function
 > of multiple pulses
 > per rsector in readout module may not be tested thoroughly in different
 > configurations.
 >
 > > - For me it is not possible too to have such differences only due to gcc
 > > version or stdlib version. But here I ask myself a question : MJ, you
 > > said you get results with 3.1.2 similar to 3.0.0 but what G4 data
 > > packages do you use ?? Then if you use the good packages version, I do
 > > not understand anything anymore ... Let me know your thoughts about
 > that.
 >
 > 6, That is still the first thing we should know:
 > Is the strange behavior from physics basis (Geant4 etc) or from coding?
 > Since MJ is the only one reported no difference, we do need a definite
 > answer from you!
 >
 
 Yes, it's coming ;-)
 
 > - I will now try to reproduce the problem in intersector coincidences
 > > with the Gate coincidence sorter to see where this problem come from and
 > > to see if it is relative to the energy spectra problem too.
 > >
 >
 > 7, Good.
 >
 > > But now, it is time to go on holidays :-)
 > > Dig, dig, dig and dig more as say MJ !
 >
 > 8. More homework for Irene and Jasmine:
 >
 > Always set bluring to 0.01:
 >
 > 1) set the exact physics from benchmarkPET in your single_layer (even
 > verbose level); run to see ?
 >
 > 2) setDepth 1; same as benchmarkPET now?
 >
 > 3) other assignments yesterday.
 >
 
 Good, that will solve some questions ask above.
 
 Have a wonderful holidays!
 >
 
 You too, thank you,
 Simon
 
 > Cheers,
 > > Simon
 > >
 > > PS : sorry but I send the images in different mails due the size
 > limitation.
 > >
 > > ------------------------------------------------------------------------
 > >
 > >
 > > ------------------------------------------------------------------------
 > >
 > > _______________________________________________
 > > Gate-users mailing list
 > > Gate-users at lists.healthgrid.org
 > > http://lists.healthgrid.org/mailman/listinfo/gate-users
 >
 >
 > Yu Chen, Ph.D.
 > University of Massachusetts Medical School
 > Division of Nuclear Medicine
 > 55 Lake Avenue North
 > Worcester, MA 01655-0243
 > Phone: (508) 856-6123
 > Fax: (508) 856-4572
 >
 >
 >
 >
 > -----Original Message-----
 > From: gate-users-bounces at lists.healthgrid.org on behalf of Simon Stute
 > Sent: Fri 12/21/2007 6:15 AM
 > To: gate-users
 > Subject: [Gate-users] Some news about the strange behaviour
 >
 > Hi all,
 >
 > I dug more to find the reason of the strange behaviour in energy spectra
 > and
 > found those results :
 >
 > - I first installed Gate3.0.0, 3.1.0, 3.1.1 and 3.1.2 on both 32 and 64
 > bits
 > architectures,and both cluster and single PC.
 > - I ran PET and SPECT benchmark on all configurations and compared both
 > energy spectra and results from benchmark.C with root : everything was
 > identical between all different config.
 > - Then I ran the single_layer.mac simulation on all config and obtained
 > the
 > same difference in energy spectra as Irene and Jasmine reported. Then I
 > found that the change in energy spectra occurs between Gate3.0.0 and 3.1.0
 > .
 > Results from 3.1.0, 3.1.1 and 3.1.2 were the same. (see attached images)
 > - I realised then that between Gate3.0.0 and 3.1.0, the associated Geant4
 > versions are 8.0.p01 and 8.1.p02. And between these two different Geant4
 > versions one of the main changes was the upgrade of the most important
 > data
 > package G4EMLOW, that move from 3.0 to 4.0, and that includes major
 > changes.
 > - I decided then to run the single_layer.mac simulation with Gate3.1.0,
 > Geant4.8.1.p02 and to change the G4EMLOW4.0 package with G4EMLOW3.0package
 > from the older version. Then the energy spectra obtained was more similar
 > than this obtained with Gate3.0.0, but still different (see
 > energy_3.1.0.EMLOW3.0.jpg image). So I think this little difference is due
 > to other main changes between Geant4.8.0.p01 and Geant4.8.1.p02, it is
 > statistical I think.
 > - The change in behaviour do not come from any digitizer module (adder,
 > readout, any blurring ...), they are still the same since Gate3.0.0 and
 > even
 > before, I compared all files and algorithms.
 > - For me it is not possible too to have such differences only due to gcc
 > version or stdlib version. But here I ask myself a question : MJ, you said
 > you get results with 3.1.2 similar to 3.0.0 but what G4 data packages do
 > you
 > use ?? Then if you use the good packages version, I do not understand
 > anything anymore ... Let me know your thoughts about that.
 > - I will now try to reproduce the problem in intersector coincidences with
 > the Gate coincidence sorter to see where this problem come from and to see
 > if it is relative to the energy spectra problem too.
 >
 > But now, it is time to go on holidays :-)
 > Dig, dig, dig and dig more as say MJ !
 > Cheers,
 > Simon
 >
 > PS : sorry but I send the images in different mails due the size
 > limitation.
 >
 >

-- 