Re: On biasing

Dear Beatrice

In order to find out if results are reliable you should calculate the=20
statistical uncertainty for your quantity of interest over a number of=20
runs (more than three). Results are generally reliable if the error is=20
below 10%. Note that errors above 20% can still mean a true uncertainty of=
=20
a factor of few.

Contour plots with visible particle "tracks" usually point to significant=
=20
statistical fluctuations that can also be introduced by the application of=
=20
excessive biasing.

Cheers
Stefan

On Mon, 25 Jul 2011, beatrice pomaro wrote:

> Dear Dr. Sala,
> =A0=A0=A0=A0=A0 you were not misunderstanding my problem, I actually did =
wrong: I should
> have decreased the importance factor through the thickness, as you sugges=
ted.
> I give in attachment the new subroutine USIMBS: I have modified the expon=
ent for
> the law describing the index FIMP in that now it is (FRADI-FRADF)/lambda =
instead
> of (FRADF-FRADI)/lambda, in order to guarantee it is negative. I ran 10^7
> primaries, instead of 10^6, and merged results over 3 cycles, again. I al=
so tried
> with 10^6 particles to compare the statistics with respect to the previou=
s
> biasing (wrong way) where 10^6 particles were run; the statistics was alm=
ost the
> same.
> I then ran 10^7 particles with the new biasing: the statistics was better=
 and the
> contours more homogeneous. I provide the new results (below the old ones)=
 in the
> .doc file in terms of neutron fluence and deposited energy, again.
> Do you think that, with this statistics, the results are reliable?=A0=A0=
=A0
> If, so, it seems to me to conclude that the different regions (aggregate,=
 cement
> paste and interfacial zone) are not differently crossed by neutrons or ha=
ve
> different deposition of energy. Am I correct?
> Thank you again for your help,
> sincerely,
> Beatrice
>=20
>=20
>=20
> At 13.45 22/07/2011, Paola Sala wrote:
> Dear Beatrice,
> if I understand correctly your routine, it seems to me that it is
> doing
> the opposite of what I suggested. My suggestion was to kill particl=
es
> far
> from the aggregate : "assign a small importance to "far" regions=A0=
 " ,
> but
> you are instead increasing the importance as the radius increases.
>
> Maybe I was not clear : in your case, the usual way of biasing, tha=
t
> is
> incrementing artificially the number of particles,=A0 has no meanin=
g,
> because you are interested in a zone on the surface of the shieldin=
g,
> a
> zone that is reached without attenuation. In your case, you can try
> to
> gain speed, reduce cpu time per primary, by reducing the time spent
> in
> particles far from the aggregate, therefore reducing the importance
> at
> large radii.
>
> If you succeed, for a given number of primaries you should get the
> same
> statistivcal error but with a reduced cpu time
> Hope it helps, and I hope I did not misunderstand your setup
>
> Paola
>
> > Dear Dr. Sala,
> >=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0 thank you for your last expla=
nations on the problem I
> > brought on Flukadiscuss; I have used the biasing card by modifyin=
g
> > the USIMBS routine, because my geometry is quite complicated in t=
he
> > investigated area, due to the presence of small aggregates in the
> > concrete shielding.
> > In the subroutine I ask biasing to be applied only at regions fro=
m
> 7
> > to 67, which correspond to the regions of the investigated small
> area
> > in front of the target, where I modelled concrete with its
> aggregates
> > and cement paste. Also I ask to use biasing only for neutrons
> (JTRACK=3D8).
> > In the subroutine, the importance factor between two regions, FIM=
P,
> > is defined by an exponential law depending on:
> > FRADF radial coordinate at the end of the track of a particle
> > FRADI radial coordinate at the beginning of the track of the
> particle
> > the normalizing length is the absorption length for concrete
> (assumed
> > value: 40cm)
> > However, after biasing this way, I do not envisage a significant
> > change in the results.
> > I send enclose the project files without and with biasing; in the
> > Word file are the results in both cases in terms of neutron fluen=
ce
> > and deposited energy; neutron fluence is reported also in
> > 1-dimensional graph in direction of the beam into the shielding.
> > Can you suggest me how to better results in the small investigate=
d
> area?
> >
> > Any help would be appreciated,
> > Thank you and kind regards,
> > Beatrice Pomaro
> >
> > At 12.20 11/05/2011, Paola Sala wrote:
> >>Dear beatrice,
> >>
> >> > Dear Dr. Sala,
> >> >=A0=A0=A0=A0=A0=A0 thank you for your help; I see one has to as=
sign different
> >> > importance factors to different regions, but I can not
> understand the
> >> > criterion to do that.
> >>Very roughly : for deep penetration studies ( NOT your case) the
> aim of
> >>importance biasing is to keep an high number of tracked particles
> even
> >>after a large amount of absorbing material, since the statistical
> error
> >>depend on how many different tracks are followed. To achieve this=
,
> >>particles are cloned at boundaries, according to particle
> importance. A
> >>rough criterion is to keep the fluence constant, achieved for
> instance by
> >>doubling or 3-pling the importance every interaction length.
> >>Note that paricles going into a region with smoller importance ar=
e
> >>randomly killed according to the importance ratio.
> >>Please have a look at the lectures from the fluka courses for mor=
e
> info.
> >>
> >> > On the other hand, would you, please, explain to me better how
> to do
> >> > practically point 1)=A0 use higher e.m. thresholds in regions =
far
> from
> >> > the scoring one
> >>EMFCUT works by region. You can segment your geometry, and put hi=
gh
> >>electron-photon thresholds in regions far from the aggregate. How
> far and
> >>how high? have a look at the range of electrons and photons in
> concrete.
> >>Keep a region around the aggregate with the same transport
> threshold as
> >>the aggregate, to preserve electronic equilibrium.
> >>
> >>
> >>and point 3) use biasing to kill particles that
> >> > travel deep in the shield.
> >>
> >>It is very unlikely, for instance, that a neutron that is 2 m dee=
p
> in the
> >>part of the shield oopposite to the aggregate will ever reach the
> >>aggregate itself. So: if -again- you segmant your hgeometry, you
> can
> >>choose
> >>either to eliminate part of your shiled, or to assign a small
> importance
> >>to "far" regions (with the BIASING card) , so that only a fractio=
n
> of the
> >>particles will be transported there.
> >>Hope this helps
> >>Paola
> >>
> >> > Thank you again,
> >> > Sincerely
> >> > Beatrice
> >> >
> >> > At 15.48 10/05/2011, Paola Sala wrote:
> >> >>Dear Beatrice
> >> >>sorry, but I do not have brilliant, or even simple, ideas for
> your
> >> case.
> >> >>If I understand correctly your geometry, your region of intere=
st
> is on
> >> >> the
> >> >>surface of the shielding, directly in front of the target. see=
n
> the
> >> beam
> >> >>energy, I assume that the beam is stopped in the target.
> >> >>Therefore, you do not have a deep penetration problem, where t=
he
> >> biasing
> >> >>card (or equivalent) would be useful.
> >> >>=A0=A0 By the way, how the biasing card is used in your input =
is
> >> meaningless:
> >> >>importanmce biasing works by splitting ok killing particles
> according
> >> to
> >> >>the relative importance of regions: if all regions have the sa=
me
> >> >>importance nothing happens.
> >> >>My only suggestion is to try to speed up the problem: use high=
er
> e.m.
> >> >>thresholds in regions far from the scoring one, use a thinner
> >> shielding
> >> >> or
> >> >>use biasing to kill particles that travel deep in the shield.
> >> >>Ciao
> >> >>Paola
> >> >>
> >> >>
> >> >>
> >> >>=A0 >
> >> >>=A0 >=A0 Dear Fluka experts,
> >> >>=A0 >=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0 I would like to put forwar=
d a problem in biasing.
> My
> >> >> problem
> >> >>=A0 >=A0 is in attachment: the geometry is a concrete room aro=
und a
> UC2
> >> >>=A0 >=A0 target. Near the point of coordinates (400, 500, 370)=
 a
> small
> >> >> portion
> >> >>=A0 >=A0 of the concrete shielding is created at a mesoscale:
> aggregates
> >> and
> >> >>=A0 >=A0 paste. Results of neutron flux and energy deposition =
are
> reported
> >> as
> >> >>=A0 >=A0 well. I guess with biasing I can better my results, t=
hat is
> I can
> >> >> get
> >> >>=A0 >=A0 no more neutron tracks but more homogeneous contours;=
 which
> >> biasing
> >> >>=A0 >=A0 technique should I apply? I tried to use the card but=
 maybe
> you
> >> can
> >> >>=A0 >=A0 suggest which is the most suitable way to do that bec=
ause I
> am
> >> not
> >> >>=A0 >=A0 that expert on biasing results.
> >> >>=A0 >=A0 Thank you for your help,
> >> >>=A0 >=A0 my best regards,
> >> >>=A0 >=A0 Beatrice
> >> >>=A0 >
> >> >>=A0 >=A0 **************************************************
> >> >>=A0 >=A0 Ing. Beatrice Pomaro
> >> >>=A0 >
> >> >>=A0 >=A0 Universita' degli Studi di Padova
> >> >>=A0 >=A0 Dipartimento di Costruzioni e Trasporti
> >> >>=A0 >=A0 Via F. Marzolo, 9 - 35131 Padova (Italy)
> >> >>=A0 >=A0 tel.: +39 049 8275605
> >> >>=A0 >=A0 e-mail: pomaro_at_dic.unipd.it
> >> >>=A0 >
> >> >>=A0 >
> >> >
> >>
> >>
> >>Paola Sala
> >>INFN Milano
> >>tel. Milano +39-0250317374
> >>tel. CERN=A0=A0 +41-227679148
> >
> > **************************************************
> > Ing. Beatrice Pomaro
> >
> > Universita' degli Studi di Padova
> > Dipartimento di Costruzioni e Trasporti
> > Via F. Marzolo, 9 - 35131 Padova (Italy)
> > tel.: +39 049 8275605
> > e-mail: pomaro_at_dic.unipd.it
> >
>=20
>
> Paola Sala
> INFN Milano
> tel. Milano +39-0250317374
> tel. CERN=A0=A0 +41-227679148
>=20
> **************************************************
> Ing. Beatrice Pomaro
>=20
> Universita' degli Studi di Padova
> Dipartimento di Costruzioni e Trasporti
> Via F. Marzolo, 9 - 35131 Padova (Italy)
> tel.: +39 049 8275605
> e-mail: pomaro_at_dic.unipd.it
>=20
>=20
>
---1970678318-274367657-1311628070=3D:1699--
Received on Mon Jul 25 2011 - 23:36:56 CEST