same parameter, different event number, different result.

Asked by YiCong Sui

Dear Kirtimaan,

I ran the same process with same parameters multiple times(p p > w+ w- in sm model).In four of the runs i set events number to 1. In another 3 runs i set events number as 100, 10000 and 1000000. the results are as follow:

cross section(pb) Events
72.5 ± 0.72 100
72.52 ± 0.16 10000
72.57 ± 0.016 1000000
73.38 ± 0.67 1
72.38 ± 0.83 1
71.96 ± 0.7 1
73.06 ± 0.74 1

Why is it that all the results are different? is that normal? some of the diverse is beyond the tolerance range.

I also used another model(EWeff_UFO) to run the process p p > w w w,
the diverse of results is even larger, many times of the error.

 the results are as follow:

cross section(pb) Events
48.36 ± 1.2 1
52.91 ± 2.4 1
51.76 ± 1.3 1
55.69 ± 0.047 100000
55.67 ± 0.047 100000
55.8 ± 0.044 100000
56.1 ± 0.016 1000000

 I want to know the reason for such diverse of results and wether such diverse is normal or not. And is there a link between the accuracy of the cross section calculation and the events number?

Thank you
Best regards

YiCong Sui

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
Last query:
Last reply:
Revision history for this message
Jure Drobnak (jure-drobnak) said :
#1

Hi YiCong,

I think I might be experiencing a similar issue, possibly of the same nature as you but in a p p > j j analysis.
See my thread https://answers.launchpad.net/madgraph5/+question/233960

Best
Jure Drobnak

Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said :
#2

Hi YiCong,

The error between the various channel of integration are consider as independent and therefore combine quadratically.
This is in fact a bit optimistic, but on the other hand combine them linearly is clearly pessimistic.
So I'm never worry when two runs are above the three sigma limit.
In addition, I will never trust the result of any run make with less that 1k events.

So for your first example, I don't see anything even close to be worried about.
On the second process, your latest runs starts to be surprinsingly high compare to the other one, but nothing really alarming in fact.

> And is there a link between the accuracy of the cross section calculation and the events number?

Yes of course. More PS point will be probe and therefore more precise the cross-section will be.
In addition this forces to have more precise sub-dominant contribution in order to have to correct contribution(in terms of number of events) for those sub-dominant contribution.

Cheers,

Olivier

Revision history for this message
YiCong Sui (syckd) said :
#3

Hi Olivier,
Thank you for answering my question. I know what it is now.

Cheers,
YiCong Sui