Uncertainty methodology
Uncertainty methodology
My PHYSOR paper using SERPENT has been accepted for publication pending slight revision. One request is giving some background in the uncertainty methodology SERPENT uses. I'm looking at uncertainty in keff, pin power, and number density of certain isotopes. I know that SERPENt doesn't produce uncertainty for ND, so any background into the methodology for how keff and pin power uncertainties are calculated would be useful. Has the methodology been discussed in any published work, or does SERPENT use a standard Monte Carlo uncertainty routine? If you could point me in the right direction it would be greatly appreciated.
Thanks, as always!
Thanks, as always!
 Jaakko Leppänen
 Site Admin
 Posts: 2442
 Joined: Thu Mar 18, 2010 10:43 pm
 Security question 2: 0
 Location: Espoo, Finland
 Contact:
Re: Uncertainty methodology
The uncertainty (statistical error) is calculated similarly for every output parameter. For every source cycle the code calculates an estimate from the collisions and other events that occur during that cycle. At the end of the calculation, these cyclewise estimates are combined to calculate the statistical mean and the associated standard deviation or relative statistical error.
 Jaakko

 Posts: 85
 Joined: Tue Jul 03, 2012 1:25 pm
 Security question 1: No
 Security question 2: 92
Re: Uncertainty methodology
We've been discussing the meaning of your words "the associated standard deviation or relative statistical error" in the previous post in this topic. If I have a result, say, FLUX = [9.33622E+14 0.00050], is the 0.00050 the percentage standard deviation or some other statistical term called "relative statistical error"? If I multiply the error and the value (9.33622E+14 * 0.00050), do I get the standard deviation or something else?
 Jaakko Leppänen
 Site Admin
 Posts: 2442
 Joined: Thu Mar 18, 2010 10:43 pm
 Security question 2: 0
 Location: Espoo, Finland
 Contact:
Re: Uncertainty methodology
All statistical errors in the _res.m output file are relative statistical errors, i.e. standard deviation divided by the mean. So multiplying the two values gives you standard deviation. Keff printed in the runtime output is accompanied by the standard deviation.
 Jaakko

 Posts: 16
 Joined: Thu Mar 14, 2019 3:12 am
 Security question 1: No
 Security question 2: 19
Re: Uncertainty methodology
Hi Jaakko,
From what I understand from your last response  all statistical errors are accompanied by rel statistical error (=stddev/mean).
And, keff is accompanied by std dev (not the statistical error?).
So, in this value:
the stdev for ana_keff = .00016 and for implicit is 3.75E5? Also, if the values following the mean is either stdev (as in case of keff) or statistical error  how can i tell which one is which?
Sorry I am a little bit confused here. Thank you in advance for your help.
Ishi
From what I understand from your last response  all statistical errors are accompanied by rel statistical error (=stddev/mean).
And, keff is accompanied by std dev (not the statistical error?).
So, in this value:
Code: Select all
ANA_KEFF (idx, [1: 6]) = [ 1.19742E+00 0.00016 2.97563E01 0.00016 1.79044E03 0.00269 ];
IMP_KEFF (idx, [1: 2]) = [ 1.19738E+00 3.7E05 ];
Sorry I am a little bit confused here. Thank you in advance for your help.
Ishi
 Jaakko Leppänen
 Site Admin
 Posts: 2442
 Joined: Thu Mar 18, 2010 10:43 pm
 Security question 2: 0
 Location: Espoo, Finland
 Contact:
Re: Uncertainty methodology
What I mean by runtime output is what Serpent prints out in the console when the code is run. All values in _res.m are stddev/mean, including the *_KEFF variables.
 Jaakko

 Posts: 16
 Joined: Thu Mar 14, 2019 3:12 am
 Security question 1: No
 Security question 2: 19
Re: Uncertainty methodology
Thank you for the clarification!