Different minuit results for different atlas releases

ROOT
1

Dear all,

I’m using Minuit for fitting in two different ATLAS releases:

  • Release 1: AnalysisBase,22.2.113 (ROOT 6.26/08)
  • Release 2: AnalysisBase,25.2.45 (ROOT 6.34/04)

I’ve observed that the resulting chi²/NDF values from the fits can differ by as much as 50% between the two releases, which is quite concerning given the sensitivity of our analysis.

Do you know if there are any major changes in the Minuit implementation between these ROOT versions that could explain such a large discrepancy? Or are there known issues or behavioral differences that might affect the fit quality?

Any insights would be greatly appreciated.

Best regards,
Diallo

2

Hi @diboye,

Thank you for your question. Would it be possible for you to add a small reproducer so that we can check this on our side?

I’m adding @jonas in the loop.

Cheers,
Dev

3

Thanks a lot Dev, for your reply. The package is quite big, so I am not sure if you want to run it.
However, here are the log files from the 2 releases:
The one for rel22: https://drive.google.com/file/d/1lNYWj-bjs0nDbJRTknkWMaD_Xe8-vniV/view?usp=sharing
and the one for rel25: https://drive.google.com/file/d/16ApUbWeji83q2PsCThn5_mOD63KR3TWD/view?usp=sharing
But the log files are also very big, so it might not be easy to extract the useful information so if you still want to run the code, I can point you out to it.

Best,

Diallo.

4

Maybe a diff from the two files could help.

Best,

Diallo.

5

I’m sorry I don’t have a clear understanding why. @moneta might be able to help you.

Cheers,
Dev

6

No worry Dev, thanks for trying. Let’s wait for @moneta then.

Best,

Diallo.