MadGraph5_aMC@NLO

Is it normal that the error bars ar so big using MSTW2008?

Asked by Christian

Hello,

my thesis advisor asked me to write a question about a Simulation i am running.
Well, actually multiple simulations.

Right now I am using the PDFs MSTW2008nlo68cl, ABMP_16_5_nlo and ABMP_16_5_nnlo for fixed order calculations of stable top quarks.

But for some reason the error bars for the transverse momentum of the top and antitop are very high compared to the other error bars of the other differential distributions.

In order to make them small, I am using req_acc_FO = 0.00005. The problem is that for 2 tops in the final state, this takes nearly 6 hours at NLO. 1 hour for LO. I am not sure how long it will take for the 4tops.

When I use req_acc_FO = 0.0005, the error is still very big. (I would upload a image or my HwU file, but i am not sure if it is possible here)

So I should as if it is normal to get so many jumps in the distribution for req_acc_FO=0.0005?

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
Best Olivier Mattelaer (olivier-mattelaer) said : 		#1

Hi,

I do assume in the following that the problem of the "jump"/error are purely related to statistical error and that you are not speaking about scale/PDF error band (correct?) and that the issue is present for the three PDF (is it correct?).

So the first thing to comment about is that the observable that you are looking at is "LO accurate" since it has only contribution for the real contribution and not from the born/virtual. Therefore they are not cancelation of the first order in alpha_s for that differential distribution.

After that FO NLO, is using a substraction method and is using counter-event to remove the singular behavior of the real contribution.
This method (like all method removing singular behavior) is subject to numerical issue. So this always happen that you will have one event (huge contribution) that is going to enter a given histogram while the associate substracted counter event (huge negative contribution) is actually entering the previous/next bin. When this happens (and it will always happens at some point) you have one bin which is suddenly much below the other, followed by the next one much bigger (or in the other direction).
This is typicaly for FO NLO computation, I have never study the impact of such missmatch on non NLO accurate observable (since it is better to just do a LO computation to predict that observable) but it is possible that such missmatch is even more critical in that context.

Cheers,

Olivier

Can you be a bit more precise in your question?

1) which type of error bar are you talking about? is this the statistical error? the scale variation band? the PDF error?
(your observable is technically LO accurate, so I would expect to have larger scale variation band for example)
2) From your question it seems that the error only occur for MSTW PDF, is this really true?

Obviously, the "req_acc_FO" only decrease the statistical uncertainty, theoretical uncertainty will not be impacted),

Revision history for this message
Christian (chris0990) said : 		#2

Thanks Olivier Mattelaer, that solved my question.

Revision history for this message
Christian (chris0990) said : 		#3

Yes, I meant the statistical error.
No, also the other PDFs have a large error, when calculating the differential distribution of the top or antitop.