Re: Particle biasing

From: Alfredo Ferrari <alfredo.ferrari_at_cern.ch>
Date: Sat, 11 Oct 2008 13:03:46 +0200

Dear David

different biasing options/ways must result in the same numbers once the
results converged properly, so you are right in inquiring why you didn't
get compatible numbers. I see two possible problems with your numbers:

a) you made only two runs -> the statistical error estimation is surely
    very poor. That is, the errors reported could well be far from reality.
    A minimum of 5 runs (much better at least 10) are required for a
    meaningful evaluation of the statistical error. For example, instead of
    doing 2 runs of 1,000,000 each, you could have done 10 runs of 200,000
    In that case you would have used the same CPU time, the results would
    have been exactly the same (for the same initial random numbers) as
    for the 2, 1,000,000 each, runs, but statistical errors could have been
    different and surely better estimated. This could explain the
    difference between
      no biasing:
      Tot. response (p/cmq/pr) 4.3660339E-08 +/- 2.982906 %

      sliced geometry with importance biasing:
      Tot. response (p/cmq/pr) 3.8885091E-08 +/- 9.0743139E-02 %
    which is roughly 11% against a combined error of 3%. They could
    still be compatible, particularly if the errors are not properly
    estimated

b) the results with usimbs.f is indeed far out (> 10 times). Obviously
    it should not be, are you sure you used usibms properly?

                   Ciao
                  Alfredo

On Sat, 11 Oct 2008, David Horvath wrote:

> Dear Fluka users,
>
> I try to calculate a shielding for a spent fuel of a teaching reactor,
> and I tried out different ways of importance biasing, but they gave
> different results which i can't explain.
>
> The results
>
> usrtrack_21, inside the shielding:
>
> no biasing:
> Tot. response (p/cmq/pr) 9.8644884E-04 +/- 0.4479266 %
>
> sliced geometry with importance biasing:
> Tot. response (p/cmq/pr) 9.8818401E-04 +/- 0.1642988 %
>
> with usimbs.f:
> Tot. response (p/cmq/pr) 9.8289608E-04 +/- 9.9353164E-02 %
>
> These results are quite the same.
>
> usrtrack_22, last layer of shelding:
>
> no biasing:
> Tot. response (p/cmq/pr) 4.3660339E-08 +/- 2.982906 %
>
> sliced geometry with importance biasing:
> Tot. response (p/cmq/pr) 3.8885091E-08 +/- 9.0743139E-02 %
>
> with usimbs.f:
> Tot. response (p/cmq/pr) 5.2892267E-07 +/- 0.2122020 %
>
> These should be also the same.
>
> For each simulation I used 2 runs with 1,000,000 primaries in each run.
>
> To validate the results, I used MCNP4C with sliced geometry with the
> same importance biasing used in Fluka.
>
> The mcnp results (5,000,000 primaries)
>
> Inside the shielding:
> cell (30<40[0 0 0]) 1.21985E-03 0.0024
> Last layer of shelding:
> cell 109 4.55585E-08 0.0056
>
>
> Is it a normal behavior of the biasing, or I did something wrong?
>
>
> Thank you for your answer.
>
> Best regards,
>
> David HORVATH
> engineering physicist student
> Budapest University of Technology and Economics
>
>
> PS.:
> I attached the input, usrtrack, source.f, usimbs.f files, and the mcnp
> input and output zipped.
>

-- 
+----------------------------------------------------------------------+
|  Alfredo Ferrari                ||  Tel.: +41.22.76.76119            |
|  CERN-AB                        ||  Fax.: +41.22.76.69474            |
|  1211 Geneva 23                 ||  e-mail: Alfredo.Ferrari_at_cern.ch  |
|  Switzerland                    ||                                   |
+----------------------------------------------------------------------+
Received on Sat Oct 11 2008 - 15:35:00 CEST

This archive was generated by hypermail 2.2.0 : Sat Oct 11 2008 - 15:35:01 CEST