Re: FLUKA: USRBDX results and running speed
Fan,
Stefan was smarter than me and spotted not only your errors but also
those in the sample input. I never liked that example anyway, but now it
really looks that we have to provide something more useful to a new user.
But here I would like to address your question about speed.
You are insisting on using the ICARUS setting, but I insist on saying that
1) it was designed for very exotic purposes and not for the normal user
2) it is far from being representative of any FLUKA application, unless
you are a member of the ICARUS collaboration or are designing an
apparatus similar to ICARUS
3) for your problem, and for most common problems, you don't need its
collection of time-consuming choices
Here is the list from the manual of what ICARUS implies:
- EMF on
==> You don't need that, since you are scoring only neutrons and
have not requested photonuclear interactions.
Discard electrons and photons and you will cut your CPU time by a large
amount. From your PEMF data file, you are following electrons down to 5
keV! That will take veeeeeery long, and will contribute nothing to your
neutron score.
- Rayleigh scattering and inelastic form factor corrections
to Compton scattering activated
==> Same thing as above.
- Detailed photoelectric edge treatment and fluorescence
photons activated
==> Idem.
- Low energy neutron transport on
(high energy neutron threshold at 19.6 MeV)
==> OK, this one you do need, but you get it also with NEW-DEFA.
- Fully analogue absorption for low energy neutrons
==> This you can get also by a card GLOBAL, with WHAT(2) = -1.
- Particle transport threshold set at 100 keV, except neutrons
(19.6 MeV), antineutrons (50 MeV), and (anti)neutrinos
(0, but discarded)
==> Not needed. You can gain CPU time by setting proton and pion
thresholds at an energy where they cannot produce neutrons.
Let's say, something like 6 MeV should be more than enough.
- Multiple scattering threshold at minimum allowed energy
both for primary and secondary charged particles
==> Not needed. The defaults are sufficient for your problem.
- Delta ray production on with threshold 100 keV
==> Not needed. Think that in your problem each delta ray is followed
until it reaches 5 keV!
- Restricted Landau fluctuations on both for hadrons/muons
and EM particles
==> Not needed.
- Tabulation ratio for hadron/muon dp/dx set at 1.04, fraction
of the kinetic energy to be lost in a step set at 0.05,
number of dp/dx tabulation points set at 80
==> The default is good enough for your problem
- Heavy particle e+/e- pair production activated with full
explicit production (0 threshold --> 2 m_electron)
==> Not needed, since yours is a low-energy problem and this kind
of effect is irrelevant. Taking the default is not sufficient in this
case: you should add a card PAIRBREM with WHAT(1) = -3. to explicitely
switch OFF this option. You will not gain much during transport time,
but you will save a considerable initialization time.
- Heavy particle bremsstrahlung activated with explicit
photon production above 300 keV
==> Same thing as for the previous item
- Muon photonuclear interactions activated with explicit
generation of secondaries
==> A very exotic effect that you certainly don't need. OK, you
probably will not have many muons anyway.
- Heavy fragment transport activated
==> Not needed.
I hope to have convinced you. Try again with a more reasonable setting,
and I bet that FLUKA will not be much slower than other codes.
Maybe a little, but not a factor of 9 as quoted by you. After all,
when you run MCNPX and GCALOR you don't ask them to track 5 keV electrons
or to simulate muon photonuclear interactions!
Best regards,
Alberto
On Fri, 6 Apr 2001, Fan Lei wrote:
> >
> > I just had a quick look into your input. Your beam momentum is 1GeV/c;
> > therefore the maximum kinetic energy a neutron can have is only about
> > 0.43GeV. In contrast to what you wrote you have defined the inner shell
> > (region 3) to be Aluminum and the outer Tungsten (region 4). Furthermore I
> > suggest that you use a lower energy limit for the usrbdx-binning which
> > coincides with one of the low-energy neutron group boundaries (see
> > manual).
>
> Hi Stefan,
>
> Thanks for pointing out the mistakes in my .inp file. This shows that one
> realy OUGHT
> to read the manual before rushing into any simulations.
>
> An update on the performance:
>
> FLUKA ( ICARUS defaults): 1487 s
>
> same problem with similar cuts:
> MCNPX: 262 s
> GCALOR: 32 s
>
> I am a bit surprised on the difference in speed of these codes. Are there
> any benchmarks
> on these codes available?
>
> Regards.
>
> Fan
--
Alberto Fassò
CERN-EP/AIP, CH-1211 Geneve 23 (Switzerland)
Phone: (41 22) 767 2398 Fax: (41 22) 767 9480 Alberto.Fasso@cern.ch
Partial thread listing: