Re: Strange behavior with biasing

From: Stefan Roesler <sroesler_at_mail.cern.ch>
Date: Wed, 28 Jan 2009 18:20:08 +0100

(smtp1.mi.infn.it [192.84.138.69]); Wed, 28 Jan 2009 18:20:14 +0100 (CET)
Sender: owner-fluka-discuss_at_mi.infn.it

Hi Steffen

If importances of adjacent regions are defined to be equal (BIASING,
What(1)>=0) the modifying parameter M (BIASING, What(1)<0) still changes
the importances. In particular, they are modified in the same way as for
Russian Roulette. Thus, the manual (BIASING, Note 2) should read:

         2) Different importances ...
            ...
            At a boundary crossing, let us call I1 the importance of the
            upstream region, and I2 that of the downstream region.

              - If I2 <= I1, Russian Roulette will be played.
            ...

Thus, in your case of the transport inside the beam pipe, at each region
crossing the importance of the downstream region is divided by M=8. Since
the primaries cross at least two of such boundaries the statistics is very
poor with 50 primaries and the decreasing fluence is simply an artifact.
As you confirmed to me privately, fluence remains constant (as it should,
since the Russian Roulette is compensated by increasing weights) if the
number of primaries is increased.

I confirm your observation that the modifying parameter M introduces an
asymmetry in splitting/Russian roulette at the same boundary.

Cheers
Stefan

On Mon, 26 Jan 2009, Steffen Mueller wrote:

> Dear Fluka users,
>
> I'm getting strange results as soon as I turn biasing on. To make it
> immediately clear what I mean, I prepared some plots:
> https://test-stemu.web.cern.ch/test-stemu/fluka_stuff/
>
> For testing I used a 10GeV pion beam starting at 0,0,0.001 going into the
> negative Z direction inside a beampipe vacuum (R002, R003, R004, R005). One
> run were 50 primaries. The relevant parts of the geometry, including the
> region names can be seen in "ip_regions.eps".
>
> With disabled biasing I get a result, which is shown in
"pions_bias_off.eps" .
> This is as expected, beam is going all the way downstream.
>
> After turning biasing on (nothing else changed), I get the result which is
> shown in "pion.eps". If you compare this with the region boundaries you can
> clearly see that after each boundary several particles get killed, which
> shouldn't happen, given that the bias settings for these regions are the
same.
> Here are the relevant BIASING cards:
>
>
> BIASING 0.0 1.0 1.0 R001 @LASTREG 0.0
> BIASING 0.0 1.0 1.0 R002 0.0
> BIASING 0.0 1.0 1.0 R003 0.0
> BIASING 0.0 1.0 1.0 R004 0.0
> BIASING 0.0 1.0 1.0 R005 0.0
> * ------------------ ADDITIONAL BIASING --------------------------*
> BIASING -1.0 0.0 MUON+ MUON-
> BIASING -1.0 0.4 ELECTRON POSITRON
> BIASING -1.0 0.4 PHOTON
> BIASING -1.0 8.0 PROTON APROTON
> BIASING -1.0 2. NEUTRON
> BIASING -1.0 2. ANEUTRON
> BIASING -1. 8.0 KAONLONG
> BIASING -1. 8.0 PION+
> BIASING -1. 8.0 PION-
> BIASING -1. 8.0 KAON+
> BIASING -1. 8.0 KAON-
> BIASING -1.0 8.0 KAONZERO
> BIASING -1.0 8.0 AKAONZER
>
>
> The full input file is also copied to the above mentioned web page.
>
> Looking into the manual I also have a question regarding BIASING: For
the case
> M>1 it seems to me that RR and splitting becomes asymmetric,e.g:
>
> I2=0.015, I1=0.0125, M=8:
> RR: I1/(M*I2)=0.104
> Splitting: 1+M*(I2/I1 -1) = 1+8*(1.2 - 1) = 2.6
>
> So about 4times more RR than splitting. Therefore I don't understand this
> sentence a bit further down: "If the only biasing is via region importances
> set by WHAT(3),only limited fluctuations arise (all particles of a given
kind
> have about the same weight in the same region), and no window is needed."
>
> As far as I can see it is possible that same particles in same regions
might
> have different weights because of that asymmetry. Please let me know
where I
> didn't get the point.
>
> Last point is, that the manual clearly splits up between I1>I2 and
I1<I2, so I
> wonder what really happens for I1=I2 M=8:
> a) 1 + M* (I2/I1 -1) = 1.0 (great!)
> b) I2/(M*I1) = 1/8 = 0.125 (not so great!)
> c) internal check puts biasing directly to 1.0 (also great!)
>
>
> Please let me know if you need more information.
>
> Cheers,
> Steffen
>
>
>

-- 
___________________________________
Stefan Roesler
CERN, SC/RP
CH-1211 Geneva 23
Switzerland
Phone:  +41-22-7679891
Fax:    +41-22-7669639
E-mail: Stefan.Roesler_at_cern.ch
Received on Thu Jan 29 2009 - 02:25:06 CET

This archive was generated by hypermail 2.2.0 : Thu Jan 29 2009 - 02:25:09 CET