Inconsistent MadSpin output when using ms_dir option

Since you are reusing the same madspin directory,
It is indeed make sense that some internal seed are kept updated between two consecutive runs to prevent correlation between different sample using the same madspin directory.

Cheers,

Olivier

> On 23 Jun 2021, at 13:55, Jens Roggel <email address hidden> wrote:
>
> New question #697668 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/697668
>
> Dear experts,
>
> I am currently developing a setup for tW samples using MG2.8.1+MadSpin with the MadSTR plugin. While testing my setup, I encountered some inconsistencies in the MadSpin output that I would like to understand.
>
> In order to simplify things, I created a minimal example that allows to reproduce the behaviour without any additional plugin.
> 1.) Event generation via p p > t t~ [QCD]
> 2.) MadSpin run using the default madspin card plus the options "set ms_dir MadSpin" and "set seed 1"
> 3.) Second MadSpin run on the same undecayed events with the same madspin card
>
> When I compare the output of the first and second MadSpin run, I see differences in the LHE files. Naively I would expect that the outputs are identical because the the same input events and the same random seeds were used in both MadSpin runs. Am I missing something?
>
> Thanks in advance.
>
> Best,
> Jens
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Since you are reusing the same madspin directory,
It is indeed make sense that some internal seed are kept updated between two consecutive runs to prevent correlation between different sample using the same madspin directory.

Cheers,

Olivier

> On 23 Jun 2021, at 13:55, Jens Roggel <email address hidden> wrote:
>
> New question #697668 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/697668
>
> Dear experts,
>
> I am currently developing a setup for tW samples using MG2.8.1+MadSpin with the MadSTR plugin. While testing my setup, I encountered some inconsistencies in the MadSpin output that I would like to understand.
>
> In order to simplify things, I created a minimal example that allows to reproduce the behaviour without any additional plugin.
> 1.) Event generation via p p > t t~ [QCD]
> 2.) MadSpin run using the default madspin card plus the options "set ms_dir MadSpin" and "set seed 1"
> 3.) Second MadSpin run on the same undecayed events with the same madspin card
>
> When I compare the output of the first and second MadSpin run, I see differences in the LHE files. Naively I would expect that the outputs are identical because the the same input events and the same random seeds were used in both MadSpin runs. Am I missing something?
>
> Thanks in advance.
>
> Best,
> Jens
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Since you are reusing the same madspin directory,
It is indeed make sense that some internal seed are kept updated between two consecutive runs to prevent correlation between different sample using the same madspin directory.

Cheers,

Olivier

> On 23 Jun 2021, at 13:55, Jens Roggel <email address hidden> wrote:
>
> New question #697668 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/697668
>
> Dear experts,
>
> I am currently developing a setup for tW samples using MG2.8.1+MadSpin with the MadSTR plugin. While testing my setup, I encountered some inconsistencies in the MadSpin output that I would like to understand.
>
> In order to simplify things, I created a minimal example that allows to reproduce the behaviour without any additional plugin.
> 1.) Event generation via p p > t t~ [QCD]
> 2.) MadSpin run using the default madspin card plus the options "set ms_dir MadSpin" and "set seed 1"
> 3.) Second MadSpin run on the same undecayed events with the same madspin card
>
> When I compare the output of the first and second MadSpin run, I see differences in the LHE files. Naively I would expect that the outputs are identical because the the same input events and the same random seeds were used in both MadSpin runs. Am I missing something?
>
> Thanks in advance.
>
> Best,
> Jens
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Hi Olivier,

Is there a way to reset these (internal) seeds so that we have complete control over reproducibility?

Cheers,
Zach

Like this not in a way that I'm aware off.
Resetting the ms_dir directory to his previous state (like intended for gridpack mode) should do the trick.

I have link this question to the following bug report: https://bugs.launchpad.net/mg5amcnlo/+bug/1914822
where other seed issue are reported.

Unfortunately due to lack of time this kind of bug are quite low in priority for the moment (and therefore dormant since a while).

Cheers,

Olivier