SM value using Reweighting

Asked by Abdellah Tnourji on 2021-06-01

Dear Expert,

I generated two EFT sample: ttbar event using SMEFT@NLO_LO model (MadGraph version: MG5_aMC_v2_9_1_2) .

I used MadGraph to reweight the generated events with a different set of ctq8 and ctg as following :

1. Generate a sample with differents value of ctq8 using reweighting [all other operators are zero], 500k event
2. Generate another sample with different value of ctg using reweighting [all other operators are zero], 500k event

My question :

When I compute the value of the observable I used in my analysis, spin correlation ckk, for ctg=0, and ctq8=0, I got different results?

For instance :

   ctq8=0 ckk : 0.329719 +/- 0.0042406 (stat)
   ctg=0 ckk : 0.391155 +/- 0.00421376 (stat)

I think that the value of ckk for ctg_0 and ctq8_0 should be compatible within their stat error?

Thanks in advance
A.Tnourji

Question information

Language:
English Edit question
Status:
Answered
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said :
#1

Hi,

1) Why do you think that theoretical uncertainty does not play any role here?
2) Which re-weighting are you using? the one with sum over helicity or the helicity by helicity one (default)?
if the second one are you sure that all SM helicity are populated in both cases?
3) How do you compute statistical uncertainty? Note that such estimator gives an hint of what the statistical uncertainty is but this estimator can not really be trusted since they are many case where such value is under-estimated (typically when something goes bad with the re-weighting)

Cheers,

Olivier

> On 1 Jun 2021, at 22:30, Abdellah Tnourji <email address hidden> wrote:
>
> New question #697345 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/697345
>
> Dear Expert,
>
> I generated two EFT sample: ttbar event using SMEFT@NLO_LO model (MadGraph version: MG5_aMC_v2_9_1_2) .
>
> I used MadGraph to reweight the generated events with a different set of ctq8 and ctg as following :
>
> 1. Generate a sample with differents value of ctq8 using reweighting [all other operators are zero], 500k event
> 2. Generate another sample with different value of ctg using reweighting [all other operators are zero], 500k event
>
> My question :
>
> When I compute the value of the observable I used in my analysis, spin correlation ckk, for ctg=0, and ctq8=0, I got different results?
>
> For instance :
>
> ctq8=0 ckk : 0.329719 +/- 0.0042406 (stat)
> ctg=0 ckk : 0.391155 +/- 0.00421376 (stat)
>
> I think that the value of ckk for ctg_0 and ctq8_0 should be compatible within their stat error?
>
> Thanks in advance
> A.Tnourji
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Abdellah Tnourji (atnourji) said (last edit ):
#2

Hello Olivier,

1) Why do you think that theoretical uncertainty does not play any role here?

yes, theoretical uncertainty has an impact, I only mentioned stat errors because I knew how to compute them.

2) Which re-weighting are you using? the one with sum over helicity or the helicity by helicity one (default)?
if the second one are you sure that all SM helicity are populated in both cases?

I am running MG using python2.7 and I do not think that I am using the helicity option for re-weighting. Please find below the procedure I used to generate the first sample.

==> if the second one are you sure that all SM helicity are populated in both cases?

I am not sure I understand the question, but after I generate the sample, I used a python script to read the LHE file and generate a root file. then, I reweighted my histogram using, for example, weight_ctg_0 .

Here is a plot showing the distribution of weights: https://cernbox.cern.ch/index.php/s/5zgBPygks2Bz0ew
(Red => weight_ctg_0, Blue => weight_ctq8_0)

3) How do you compute statistical uncertainty?

Ckk observable is computed using the average of cos(theta, +)*cos(theta, -) distribution times -9 e.g, ckk = -9 * < cos(theta, +)*cos(theta, -) >.

The stat error is computed using the standard error of the mean of cos(theta, +)*cos(theta, -) distribution times 9.

Best,
A.Tnourji

I am running MG interface using : python2.7 ./bin/mg5_aMC

proc_card :
--------------------------------------------------------------------
import model SMEFTatNLO-LO_1_5_ctq8
define p = p b b~
generate p p > t t~ QED=0 QCD=2 NP=2
output ctq8_smeft_lo
--------------------------------------------------------------------
where SMEFTatNLO-LO_1_5_ctq8 is a restricted card where ctq8=1.5 and all other operates are zero

reweighing_card :
----------------------------------------------------------------------------------------
change mode LO
change helicity False
launch --rwgt_name=ctq8_neg_1_5
  set DIM64F 4 -1.5
launch --rwgt_name=ctq8_neg_1_2
  set DIM64F 4 -1.2
launch --rwgt_name=ctq8_neg_1
  set DIM64F 4 -1
launch --rwgt_name=ctq8_neg_0_5
  set DIM64F 4 -0.5
launch --rwgt_name=ctq8_neg_0_2
  set DIM64F 4 -0.2
launch --rwgt_name=ctq8_neg_0_09
  set DIM64F 4 -0.09
launch --rwgt_name=ctq8_neg_0
  set DIM64F 4 0
launch --rwgt_name=ctq8_0_09
  set DIM64F 4 0.09
launch --rwgt_name=ctq8_0_2
  set DIM64F 4 0.2
launch --rwgt_name=ctq8_0_5
  set DIM64F 4 0.5
launch --rwgt_name=ctq8_1
  set DIM64F 4 1
launch --rwgt_name=ctq8_1_2
  set DIM64F 4 1.2
launch --rwgt_name=ctq8_1_5
  set DIM64F 4 1.5
----------------------------------------------------------------------------------------

I am also using MadSpin for decay particles :
--------------------------------------------------------------------
decay t > w+ b, w+ > l+ vl NP=0
decay t~ > w- b~, w- > l- vl~ NP=0
decay w+ > l+ vl
decay w- > l- vl~
--------------------------------------------------------------------

Revision history for this message
Abdellah Tnourji (atnourji) said :
#3

Hello,

I was wondering if you have had a chance to look at my error yet? your feedback will be of great assistance.

Cheers,
A.Tnourji

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said :
#4

if theoretical uncertainty have an impact, the question is if your computation agrees or not within theoretical uncertainty.
If it does then problem solves (and/or you can redo your computation to use the same theoretical hyppothesis to be able to do such comparison without theoretical uncertainty).

3) How do you compute statistical uncertainty?

I'm not sure to understand your answer but it can be correct.

Now from your weight distribution, you can see that you have negative weights in both sample, this will increase significantly the statistical error (even if they are not that many of those). Is this due to the PDF? (in theory you should use LO pdf for LO prediction and those type of pdf should be definite positive).

From the distribution of weights, you can also expect that the statistical uncertainty to be much bigger for the blue curve compare to the red curve since you have a much worse distribution of weights.

2) Which re-weighting are you using? the one with sum over helicity or the helicity by helicity one (default)?

From your script it seems that you re-weight according to the sum over helicity and not helicity by helicity.
So be sure that you do not use any helicity information from the leshouches event since this will not be correct.

Can you help with this problem?

Provide an answer of your own, or ask Abdellah Tnourji for more information if necessary.

To post a message you must log in.