[fluka-discuss]: Re: About statistical error

From: Santana, Mario <msantana_at_slac.stanford.edu>
Date: Wed, 25 Sep 2013 10:27:47 -0700

>
>
>Hi,
>
>what is mean by independent run?

By independent run I meant those in which you put a different seed number
in the 'RANDOMIZ' card

>what will do seed of the previous run? does it will run the same random
>number sequence?

Sorry, I don't quite understand the question as it is phrased. But I guess
the answer is NO, the sequence will obviously not be the same for runs
with different seeds.
>
>Also If your number of histories are large and number of cycle are also
>high.
>then making independent run will really change your result much???????

'Much' what? Time consuming? Please don't get confused. My recommendation
was that if you have several CPU's available instead of having many
catenated run on a single machine you could distribute the load over
several machines, but them you would need to remember to give a different
seed to each run through RANDOMIZE
 
>
>Regards,
>Pravin Rai.
>
>> On 9/24/13 1:35 AM, "pkrai_at_barc.gov.in" <pkrai_at_barc.gov.in> wrote:
>>
>>>I have made two run for a same input file.
>>>
>>>1) With number of histories=1000000 and 10 statistically independent
>>>run.
>>>
>>>2) With number of histories=10000000, with only 1 run
>>>
>>>
>>>I am getting same result for usrbdx detector in both cases with only
>>>difference of statistical error.
>>
>> This is normal. If you look at the end of your long simulation there
>>will
>> be a random seed file ranYYY002 which will be identical to that after
>>the
>> 10th cycle of your short simulation, ranYYY011.
>>
>>>
>>>so what is the use of 10 independent run, only for calculation for
>>>error?
>>
>> Please note that what you are describing are not 'independent' runs.
>>Each
>> of these runs starts with a seed generated at the end of the previous
>>run,
>> i.e.. ranYYYnnn
>> Yes, it allows computing the error. Also, you could compute higher order
>> momenta of your estimators.
>> Also, by splitting a long simulation in shorter runs you could start to
>> analyze your output after the fist few runs and decide to stop your
>> simulation if you have enough statistics or if you have done something
>> wrong. Moreover, if the system crashes for any reason, you will have
>>some
>> results.
>>
>> However, the most common use of shorter runs is to send 'truly'
>> independent simulations (each with a different seed number in the input
>> file) to different CPUS, so that you can 'parallelize' your calculation.
>>
>>>
>>>what is mean by a vector of 97 seed? how it initialises random number
>>>generator?
>>
>> The seed file is used to start your next simulation (when you use the
>>-N n
>> -M m flags in the running script)
>>
>> FLUKA uses an updated version of the Marsaglia random generator
>>algorithm
>> to 64-bit pseudo-random sequences with a periodicity of about 10^44
>>
>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>
>
Received on Wed Sep 25 2013 - 20:36:10 CEST