[Gate-users] Reproducible Artifact With Ranlux64 - Not Present With Mersenne Twister

[Geron Bindseil]

Dear GATE Users,

I have encountered a reproducible artifact when using the Ranlux64 random
number generator (auto seed). The artifact does not appear when using the
Mersenne Twister random number generator (also auto seed).

I have attached two sinograms (centre axial plane) of uniform cylinder
source PET emission data for both generators. (Note: The Mersenne Twister
sinogram was generated from a simulation that was run for a shorter time
and thus fewer counts.)

In the Ranlux64 image, there is a hot spot at the centre of the sinogram.
This appears to stretch out in other axial planes of the sinogram (not
shown).

A few other users have reported unexpected results in PET data when using
Ranlux64 over this listserve. I suggest that the default random number
generator (used with the benchmarks and other example code that new users
start with) be changed to Mersenne Twister in the next GATE release. I will
also update the GATE Wiki to add a recommendation to avoid using Ranlux64.

Thanks,
Geron Bindseil

---

Geron Bindseil
PhD Candidate, Physics
University of Western Ontario
London, Canada
gbindsei at uwo.ca
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MT.png
Type: image/png
Size: 5920 bytes
Desc: not available
URL: <http://lists.opengatecollaboration.org/mailman/private/gate-users/attachments/20111123/51bd2330/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: RanLux64.png
Type: image/png
Size: 11949 bytes
Desc: not available
URL: <http://lists.opengatecollaboration.org/mailman/private/gate-users/attachments/20111123/51bd2330/attachment-0003.png>