Dear Imen,
the difference between the calculated and measured data (which I would
not name relative error...) can depend on two main reasons:
- the systematic error of your measurement, and this is responsibility of
the experimentalist (who should be able to assign a systematic uncertainty
to the measurement... do you know this uncertainty in your case?)
- how much the model in your simulation is reproducing the reality (in
very general way in terms of geometry, source term of the radiation,
conditions of irradiation and so on).
The statistical uncertainty of your simulation has nothing to do with
this... then it has no sense to worsen it by purpose (If i have understood
correctly your doubt). Of course, you can decide to be satisfied when -see
the answer of Stefan some days ago- you are sure that your simulation
converged, without going to a small fraction of percent... but you do this
because you reached the convergence, you are not improving the accuracy
worsen the statistical error.
Inversely, I would proceed to check first if the difference that you find
stays within the error bars that you assign to your measurement. Second, I
would try to understand how adherent to the reality is your model (you could
even -maybe it is not simple- try to assign an error).
Hope this could help,
best,
Anna
Am Sat, 23 Mar 2024 12:44:49 +0000 schrieb imen hammami :
>Thank you to everyone who took the time to respond to my question. The statistical errors obtained exhibit remarkably low values, while the relative errors (difference between calculated and measured data) are slightly higher, ranging from 4% to 20%. Is this acceptable? Or the statistical uncertainty should ideally be within a comparable order of magnitude for optimal accuracy?
>
>Best regards
>
>
>Le ven. 15 mars 2024 à 14:00, Paola Sala <paola.sala_at_orange.fr> a écrit :
>
>>Dear All,
>>thank you Stefan. I might just add a detail:
>>The procedure described by Stefan assumes an obvious prerequisite: the
>>fluka runs must be "independent", meaning that they must start with
>>different random numbers. The Fluka ($FLUPRO/flutil/rfluka ) and flair
>>procedures ensure this independence by input/output of the first/last
>>random number in each cycle.
>>However, if several runs are sent in parallel on different CPUs, or the
>>same run is restarted from zero in a different disk area, the user has
>>to be sure that the initial random for the runs are different, by
>>changing the RANDOMIZE card. Otherwise, results from the parallel runs
>>will be identical. In this case, statistical errors will be magically
>>zero or negligible (can be, due to rounding when writing results to disk).
>>Again, fluka and flair provide utilities (a script in $FLUPRO/flutil, or
>>the loop menu of flair) to handle this automatically.
>>Regards
>>Paola
>>On 3/15/24 09:34, Stefan E. Mueller wrote:
>>> Dear Amandeep Sharma,
>>>
>>> the FLUKA postprocessing routines (either invoked from FLAIR's
>>> "Process"-Button in the "Run"->"Data" menu, or using directly the
>>> executables in $FLUPRO/flutil/*suw) use the Central Limit Theorem to
>>> express the statistical uncertainty of the scorer result as the standard
>>> deviation (sigma) of a normal distribution. For this, the
>>> postprocessing routines need at least 5 independent FLUKA runs.
>>>
>>> See e.g.
>>>
>>>
>>https://indico.cern.ch/event/753612/contributions/3121551/attachments/1974578/3285956/MC_2019.pdf
>>
>>>
>>> (slides 20ff)
>>>
>>> As a (crude) rule of thumb, the simulation has converged if this
>>> uncertainty is better then 5 percent.
>>>
>>> All other uncertainties (often called "systematic uncertainties")
>>> unfortunately have to be (gu)es(s)timated by the user.
>>>
>>> Cheers,
>>>
>>> Stefan
>>>
>>> --
>>> Stefan E. Mueller
>>> Department of Information Services and Computing - Computational Science
>>> and Institute of Radiation Physics
>>> Helmholtz-Zentrum Dresden-Rossendorf
>>> Tel: +49 (0351) 260 3847
>>>
>>Stefan.Mueller_at_hzdr.de
>>>
>>http://www.hzdr.de
>>>
>>> Vorstand: Prof. Dr. Sebastian M. Schmidt, Dr. Diana Stiller
>>> Vereinsregister: VR 1693 beim Amtsgericht Dresden
>>>
>>> On Thu, 14 Mar 2024, Amandeep Sharma wrote:
>>>
>>>> Whether it represents the statistical uncertainity in results or it is
>>>> different from statistical uncertainity?
>>>> Regards !
>>>>
>>>> On Thu, 14 Mar, 2024, 10:15 pm Stefan E. Mueller,
>>>> <
>>stefan.mueller_at_hzdr.de
>>>
>>>> wrote:
>>>> Dear Imen Hammami,
>>>>
>>>> judging from your screenshot, I would say that your uncertainty
>>>> is
>>>> on the order of 0.014% for all your results. No need to multiply
>>>> by 100.
>>>> These are already percentage errors.
>>>>
>>>> Cheers,
>>>>
>>>> Stefan
>>>>
>>>> --
>>>> Stefan E. Mueller
>>>> Department of Information Services and Computing - Computational
>>>> Science
>>>> and Institute of Radiation Physics
>>>> Helmholtz-Zentrum Dresden-Rossendorf
>>>> Tel: +49 (0351) 260 3847
>>>>
>>Stefan.Mueller_at_hzdr.de
>>>>
>>http://www.hzdr.de
>>>>
>>>> Vorstand: Prof. Dr. Sebastian M. Schmidt, Dr. Diana Stiller
>>>> Vereinsregister: VR 1693 beim Amtsgericht Dresden
>>>>
>>>> On Thu, 14 Mar 2024, imen hammami wrote:
>>>>
>>>> > Hi, Could someone provide guidance on how to calculate the
>>>> Monte Carlo
>>>> > error from the output of the USRBIN card, given that we've
>>>> utilized 5*10^7
>>>> > primaries for 5 cycles to calculate the dose? Do I need to
>>>> multiply the
>>>> > values in the red box by 100 to get the exact error value in
>>>> percentage? Any
>>>> > insights would be greatly appreciated.
>>>> >
>>>> > Regards
>>>> >
>>>> >
>>>> >
>>>> >
>>>> >
>>>> >
>>>> > image.png
>>>> >
>>>> >
>>>>
>>>>
>>>>
>>
>>
>
-----------------------------------------------------------------------------------------
Dr. Anna Ferrari
Institute of Radiation Physics | High Energy Density Division
Helmholtz-Zentrum Dresden-Rossendorf e.V. (HZDR)
Bautzner Landstraße 400 | 01328 Dresden | Germany
Tel. +49 351 260 2872 off.
+39 3475802161 mob.
http://www.hzdr.de
Board of Directors: Prof. Dr. Sebastian M. Schmidt, Dr. Diana Stiller
Company Registration Number: VR 1693, Amtsgericht Dresden
__________________________________________________________________________
You can manage unsubscription from this mailing list at
https://www.fluka.org/fluka.php?id=acc_info
- application/x-pkcs7-signature attachment: smime.p7s
Received on Sat Mar 23 2024 - 19:49:14 CET