Delphes Reconstructing too few photons

[Jake]

Hi,

I recently upgraded to Delphes 3.4.1 from 3.3.3. I am generating events with MadGraph, hadronizing through Pythia8 and finally running Delphes with the delphes_card_ATLAS_PileUp.tcl card. My generated events include 20000 photons. Before the upgrade roughly 17000 were being reconstructed by Delphes, but after the upgrade only ~3500 are reconstructed. I've tried using the cards from v3.3.3 as well but obtain the same result; I expect to get the same result with both anyway since it looks like the photon efficiency and isolation parameters are the same for both cards.

Is there anything in the newest version that could be causing fewer photons to be reconstructed? I suspect the problem is not before reconstruction since since I have all 20000 photons in the "Particle" branch of the Delphes tree. I'm running root v6.10.04.

Thanks a lot in advance

[Michele Selvaggi]

Sorry for the very late reply, there has been an issue with the ticket notification system.
Can you attach a small event file so I can have a look?

[Jake]

Hi,

Thanks for your reply.

I've found that the issue arises when trying to include pileup, and not if I use the default card with no pileup. I have generated 200 p+p -> photon+photon events at 14 TeV and run the result through Delphes 3.3.3 and 3.4.1 with pileup. I have attached the run card from MadGraph, the Delphes card, the Delphes log files, and the ROOT files containing the final events.

[Jake]

Hi,

In case the ticketing system is fixed, I would like to ping this as I am still having issues.

Thanks.

[Pavel Demin]

For information.

Here is a link to the diff between the 3.3.3 and 3.4.1 versions of the delphes_card_ATLAS_PileUp.tcl card:

https://cp3.irmp.ucl.ac.be/projects/delphes/changeset?reponame=git&new=4b4207ec4b11bb359483461a6a825513636f647b%40cards%2Fdelphes_card_ATLAS_PileUp.tcl&old=0a6c72bd88c52c787017f271cd3063befc49fd46%40cards%2Fdelphes_card_ATLAS_PileUp.tcl

Looks like the only difference is in the configuration of the isolation modules.

In the Calorimeter module, there are more changes between the two versions:

https://cp3.irmp.ucl.ac.be/projects/delphes/changeset?reponame=git&new=298734ecdd212c321586b6f290117a6a765941e0%40modules%2FCalorimeter.cc&old=f8299bc5bd3d1b2b1d6db200ae33abe5f2d70a99%40modules%2FCalorimeter.cc

Looking at the code, I'd say that the new version should provide better results. I think that 85% photon reconstruction efficiency in the presence of pile-up is too high.

[Jake]

Hi Pavel,

Thanks a lot for your response. It seems like a factor of 5-6 difference in the photon reconstruction efficiency is rather high, and perhaps points to a bug in the earlier version. I'm looking through the modules and the changes to try to understand this large difference. Perhaps you can help point me to the source of this large discrepancy. Why do you think the new version produces better results? Was there indeed a bug in the pileup subtraction of 3.3.3 that motivated the changes in the new version that caused this difference? Is there some way to validate the results of the new version?

Thanks for your time.

[Pavel Demin]

Perhaps you can help point me to the source of this large discrepancy.

The code outputting photons is on lines 553-571:

https://cp3.irmp.ucl.ac.be/projects/delphes/browser/git/modules/Calorimeter.cc#L553

This code changed significantly between the 3.3.3 and 3.4.1 versions.

Was there indeed a bug in the pileup subtraction of 3.3.3 that motivated the changes in the new version that caused this difference?

If I'm not mistaken, in Delphes the pileup subtraction doesn't affect the photons.

In Delphes, the photons are defined as the excesses of energy in the electromagnetic calorimeter that can't be associated with the charged particles.

In 3.3.3 there were some problems with the calculation of these excesses of energy that we tried to fix in 3.4.0.

Is there some way to validate the results of the new version?

We need an input from the experiments for the validation. In this particular case, we need to know the efficiency of the photon reconstruction in the presence of pile-up. Once we have this information, we can compare it with the results produced by Delphes.

[Jake]

Hi Pavel,

I have indeed noticed an increase in photon reconstruction efficiency in 3.4.1 when no pileup is added, and after some tests (outlined below) I suspect this is the primary effect of the new Calorimeter module.

The results of my new tests seem to suggest to me that the problem is not with the difference in the modules; I'm simulating 1000 diphoton events at 14 TeV with MadGraph. Using the 3.4.1 default installation I obtain 0 photons reconstructed with mean pileup of 140 overlayed. I've tried compiling and running 3.4.1 simulations using the PileUpMerger, TrackPileUpSubtractor, and Isolation modules from 3.3.3 and noticed no difference (still no photons reconstructed). Separately, I used the 3.4.1 versions of these three modules but the 3.3.3 version of the Calorimeter module. With no pileup I get the same number of photons as 3.3.3 (the 3.4.1 version of the Calorimeter module
reconstructs 50-100 more photons than the 3.3.3 version), but with pileup I once again get no photons reconstructed.

Given that using the 3.3.3 version of the modules in 3.4.1 produces no changes, could it be that this issue is an unintentional side-effect of the upgrade?

The final thing I've noticed is that the only class shared by these modules that has been modified since 3.3.3 is DelphesClasses.cc. It's not possible for me to easily use the 3.3.3 version in 3.4.1 instead since there are many altered definitions in the modules that depend on this class, but I'm wondering if the problem arises in this class?

Thanks

[Pavel Demin]

I've just checked the number of photons at the outputs of the modules in delphes_card_ATLAS_PileUp.tcl and found that it's PhotonIsolation that removes most of the photons.

The diff between the 3.3.3 and 3.4.1 versions of the delphes_card_ATLAS_PileUp.tcl linked in my first comment shows that the configuration of the PhotonIsolation module changed.

I'd say that something is wrong with the configuration of the EFlowMergerAllTracks module but I don't know how to fix it.

I'll ask Michele to check it.

[Michele Selvaggi]

Yes, I have fixed it on github.
Can you try now please?

​https://github.com/delphes/delphes/blob/master/cards/delphes_card_ATLAS_PileUp.tcl#L379

[Jake]

Hi Pavel and Michele,

This new card fixed the problem! I now have slightly more photons reconstructed with 3.4.1 in the presence of pileup, which is consistent with the increase I saw without pileup due to the new Calorimeter module.

Thanks a lot for your help!