MadGraph5_aMC@NLO

weights' sum 1% away from cross section

Asked by Gauthier

Dear MG Team,

I'm producing a LHE file with about 50k unweighted events at NLO using v2.3.0.

The lhe file contains information about the cross section and weights at several places:
1. in the <init> header: the total cross section per process (and the maximum weight)
2. in each <event>: the event weight times the number of events
3. in each <wgt id="1001">: idem

The weights in 2. and 3. match at the 1e-5 level or so, but they add up to a total cross section that differs at the 1e-2 level or so from 1.

It is definitely not critical but 1% can easily exceeds the MC uncertainty.
I would guess 1. (which is also quoted in various other places) is the most accurate.
Is that right?

Thanks a lot,
Gauthier

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Rikkert Frederix Edit question
Solved by:
Gauthier
Solved:
Last query:
Last reply:
Revision history for this message
Rikkert Frederix (frederix) said : 		#1

Dear Gauthier,

Indeed, the one in the <init> block is the most precise. If you have many negatively weighted events, the statistical uncertainty from summing the weights of the events can be *significantly* larger than what you would naively expect from the number of events requested.

Cheers,
Rikkert

Revision history for this message
Gauthier (gauthier.d) said : 		#2

Hi Rikkert,

Thanks for your prompt answer!

Is there a reason why the weights are not chosen as
sign{<weight>} * <total cross section> / ( <#positive weights> - <#negative weights> )
so that their sum gives exactly the most accurate estimate of the total cross section?

Cheers,
Gauthier

Revision history for this message
Rikkert Frederix (frederix) said : 		#3

Hi Gauthier,

That would require a reprocessing of the event file after it has been created. And the results should be statistically equivalent. Moreover, if you would generate more events from the same grids (using the "-o" option), the events will have exactly the same weights as the original event file (up to the usual sign, of course).

Best,
Rikkert

Revision history for this message
Gauthier (gauthier.d) said : 		#4

Thanks a lot!
Cheers,
Gauthier