Confusion on simulation output result

Asked by SAUMYEN KUNDU

Hi,
I am simulating radiative Bhaba at 1TeV ILC with ISR and beamstrahlung. What I found in different run is different cross-sections varying from ~50pb to ~90pb. Do you think just due to the different seed for random number generation? Below is one such result.
|=============================================================================|
| It Calls Integral[fb] Error[fb] Err[%] Acc Eff[%] Chi2 N[It] |
|=============================================================================|
| [VAMP2] set chain: use chained weights.
   1 4690 5.2038349E+04 2.32E+04 44.65 30.58* 0.78
   2 4376 3.5869327E+04 2.69E+03 7.50 4.96* 1.33
   3 3876 6.1412020E+04 9.01E+03 14.67 9.13 0.75
   4 4584 7.5462024E+04 1.29E+04 17.04 11.54 0.75
   5 4124 9.3124213E+04 2.74E+04 29.45 18.91 0.48
|-----------------------------------------------------------------------------|
   5 21650 4.0005074E+04 2.50E+03 6.26 9.21 0.48 4.91 5
|-----------------------------------------------------------------------------|
   6 9016 7.4465492E+04 6.62E+03 8.90 8.45* 0.54
   7 9016 7.6760487E+04 9.30E+03 12.11 11.50 0.49
   8 9016 1.3509076E+05 4.71E+04 34.85 33.09 0.21
|-----------------------------------------------------------------------------|
   8 27048 7.6013984E+04 5.36E+03 7.05 11.60 0.21 0.82 3
|=============================================================================|

What concerned me is with the polarised beam I expected the CS to be on the lesser side it got increased. So, I kept the seed number same to find the CS is actually less.
Now I note the error or error %. Error is 5pb which to me lokked like to be very high. Do you think this is consistent? Even the accuracy (ACC) is 11.6 and Chi^2 is 0.82.

Also, I'm getting a warning saying:
Warning: Encountered events with excess weight: 10302 events ( 10.302 %)
| Maximum excess weight = 1.437E+04
| Average excess weight = 9.287E-01

Again I am sceptic if this is good result or not.

The reason I doubtful is this same kinda problem was there in other packages namely, Calchep. There thwy said this CS is not trustworthy as the error % is high due to the collinear divergence. Even the event generation was taking too indefinite for higher event number. Whizard work like a charm so far. Just this doubt. I suppose Whizard takes care of the hard-collinear divergence by trancating the series at 3rd order. Am I right?

Here is my script-content that I'm using:
#-----------------------------------------------------------
model = SM
$integration_method = "vamp2"
process Bhaba = "e-", "e+" => "e-","e+", "A"
sqrts = 1 TeV
beams = e1, E1 => circe2 => isr
!
!beams_pol_density = @(+1), @(-1)
!beams_pol_fraction = 80%, 30%
!
?keep_beams = true
!isr_order = 3
?isr_handler = true
$isr_handler_mode = "recoil"
isr_alpha = 0.0072993
isr_mass = me
!
$circe2_file = "/home/pdas/1000_B1b_ws_ee018.circe"
$circe2_design = "ILC"
?circe2_polarized = false
!
cuts = let subevt @non_isr_photon = [ select if Index > 2 [A] ] in all E > 8 GeV [@non_isr_photon] and all Theta > 5.5 degree [@non_isr_photon] and all Theta < 174.5 degree [@non_isr_photon]
 and all Pt < 500 GeV [A]
! and all 174.5 degree > Theta > 5.5 degree [e1:E1]
! and all Dist > 0.4 [e1,E1]
!
analysis = record e_photon (eval E [extract index 1 [sort by - Pt [A]]]) ;
    record pt_photon (eval Pt [extract index 1 [sort by - Pt [A]]])
histogram e_photon (0 GeV, 500 GeV, 10 GeV)
histogram pt_photon (0 GeV, 500 GeV, 10 GeV)
!
n_events = 100000
seed = 11683
sample_format = lhef
simulate (Bhaba_Pol)
compile_analysis {$out_file = "Bhaba.dat"}
#-----------------------------------------------------------

Regards,
Saumyen

Question information

Language:
English Edit question
Status:
Solved
For:
WHIZARD Edit question
Assignee:
Juergen Reuter Edit question
Solved by:
Juergen Reuter
Solved:
Last query:
Last reply:
Revision history for this message
Juergen Reuter (j.r.reuter) said :
#1

Dear Saumyen,
indeed this is not a good integration, i.e. it is not trustworthy. Some remarks in order to improve it:
(1)
These are way too few iterations, though this is "only" a 2->3 process, it has ISR and beamstrahlung on, so this is a 9-dim integration with a tricky phase space. It would be good to take 7 or 8 iterations with 100k calls each, and then 3 iterations with 200k calls, jsut as a rule of thumb.
(2)
Your cuts only cut off the soft photon singularity and the collinear one w.r.t. the beam, but not the collinear singularity with the final state leptons, you need an electromagnetic separation criterion, not only to get a theoretically well-defined cross section but also to get a clean particle identification in the tracker/ECAL.
(3)
Please also take a look at the discussion here from the same process studied by the CLICdp collaboration:
https://answers.launchpad.net/whizard/+question/685180
Cheers,
    JRR

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#2

Hi Juergen,
Just as a quick response to your answer, I would like to ask what do you mean by "an electromagnetic separation criterion"? Do you mean that I should put some cut on electron-pair? Even in the discussion of ticket you shared https://answers.launchpad.net/whizard/+question/685180
Prof. Ulrike Schnoor also metioned there ideally shouldn't be any cut (although I couldn 't see his SINDARIN script). Then how should I put separating cut?

Regards,
Saumyen

Revision history for this message
Juergen Reuter (j.r.reuter) said :
#3

I don't know your background, but electromagnetic separation means that you can distinguish something in the electromagnetic calorimeter of the detector as two different objects, similar to jet separation for hadronic objects. What you need is a Delta-R cut (Dist) or an angular cut (Theta > 5 degrees [A: e1:E1]) or a pT cut (Pt > X GeV [A, e1:E1]).
I hope you understand that we cannot write your analysis.

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#4

I am extremely sorry, Juergen, if I anyhow offended you or too many questions disturbed you. I never intended any of those. Sorry!
Earlier also I used the DeltaR(e-,e+) cut as in the script. But that results in very low cross-section (~26fb) and the accuracy was also of the same nature. So I commented those. I again ran with the following cuts to get a CS ~5fb:
cuts = let subevt @non_isr_photon = [ select if Index > 2 [A] ] in all E > 8 GeV [@non_isr_photon] and all Theta > 5.5 degree [@non_isr_photon] and all Theta < 174.5 degree [@non_isr_photon]
 and all Pt < 500 GeV [A]
 and all 174.5 degree > Theta > 5.5 degree [e1:E1]
 and all Dist > 0.4 [e1:E1,E1:e1:A]

Anyway, I will try other things too and let you if it is fine.

Sorry again.

Thanks and regards,
Saumyen

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#5

Hi Juergen,
So I tried with some extra cuts and with higher number of iterations to get the following out output. The Whizard.log and the SINDARIN script you'll find in here:
https://www.dropbox.com/sh/ue9tcnhmpkcda22/AACwebpsWkFzkvra0T9aSCAOa?dl=0

I have also kept the some histos and the file phasespace channels.

Do you think the integration result I am getting is good? I think it may not be a good result still because, although the error% is less (1.18%) the efficiency is very less (0.16%).

From the diagrams of the phasespace channels I saw only some on-shell electrons are collinear. Is there way to put a cut on that?

Thanks and regards,
Saumyen

Revision history for this message
Juergen Reuter (j.r.reuter) said :
#6

If I'm not mistaken there is stil a singularity that is only covered due to the electron mass, namely if the photon becomes very hard, and the e+e- pair very soft. You should also demand minimal energies on your leptons or a lower invariant mass cut on the e+e- pair.
The integration you showed is ok-ish, but not very well-behaved.

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#7

Thank you so much Juergen for the explanation.
So, I modified the cuts a little bit. As you can see in the script put in the Drpbox directory (old files has a suffix 'Old'). I added two cuts namely ,
and all M <= -me*2 [beam elec, elec]
and all M > 5 GeV [elec,elec:A]
 The output has an error of 0.25% and efficiency of 1.78%. They are not pretty good even compared to the neutrino background (0.78% error & 8% efficiency), I understand. But better than the previous result. By the way, what 'Acc' stand for? I was thinking accuracy. Is it so?
And from the ticket (from Dr. Ulrike) you shared, choosing the size of the phase space may inhance the accuracy of the output. I didn't understand that much, but do you think I should explore that too? Sorry if this an inappropriate question.

With regards,
Saumyen

Revision history for this message
Best Juergen Reuter (j.r.reuter) said :
#8

Yes, acc = accuracy, with details explained in Sec. 5.7.1 of the WHIZARD manual.
Your integration looks stable and fine now.

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#9

Thank you so much, Juergen.

Regards,
Saumyen

Revision history for this message
SAUMYEN KUNDU (saumyen.k) said :
#10

Thanks Juergen Reuter, that solved my question.