Epic-backward-hcal-l mailing list

[Leszek Kosarzewski <leszek.kosarzewski AT gmail.com>]

Dear Colleagues

Today, we had an update by Brian on neutral jet reconstruction performance with backward HCal. He did many important checks by relating tracks and clusters to MC truth information. The key points are:

1. Neutral jets constitute around 20-25% of all jets.

2. Most neutral jets have just 1 neutral hadron in them, up to 4-5.

3. There are some jets falling out of nHCal acceptance - mostly reconstructed from tracks. Adding neutral cluster sometimes brings them above the E>5 GeV threshold used for jet energy.

4. The jet energy resolution is calculated as the RMS of the energy distribution of reconstructed vs. MC hadron jet. It includes long tails, so it needs to be cut or fitted with a Gaussian. For now it may be overestimated.

5. The energy resolution of charged hadron jets is 30% better when vetoing neutral jets.

6. There is an improvement of neutral jet reconstruction when using neutral clusters. Seems ~half of the missing energy is recovered on average.

7. There is a small deterioration in the jet energy resolution when including hadron clusters on top of charged track jets. This is unavoidable.

8. We can improve it further by including neutral clusters in backward EMCal. It will be useful mostly as a veto, since large energy deposit in EMCal causes a large scattering (deviation from straight line) of the neutron making a combined energy measurement difficult.

9. Track projections should be available with latest simulation campaign. I will check that.

10. We need to understand what causes the jet energy resolution to be large. Is it because of shower fluctuations or leakage?

This is a lot of information and very good progress. I think we should update other experts and request for Brian to give an update at the next calorimetry meeting in 2 weeks. Brian, could you please send me your updated slides?

Alexandr also gave a short update on position resolution study. It is also close to be completed, so we need to prepare a presentation for the next calorimetry meeting as well. For now, could you please send me your current slides? We discussed:

1. The spikes at low energy are because neutron energy is contained in a single tower or in a small cluster. They may overestimate the RMS of the position distribution, so if we generate a very fine distribution of neutron neutron projections this should lower that effect.

We also need to have a presentation about clustering test for the calorimetry meeting.

David reported via email that he made progress and is able to train the GNN algorithm. We can discuss more details next week.

I think we will keep the meeting to be at 1000h on Fridays, so that Brian can connect when he has update ready.

Best regards, Leszek

czw., 18 kwi 2024 o 15:03 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

We are going to have a DSC meeting tomorrow at 1000h BNL time. Sorry for late notification, but I had to receive responses first. The link is below:

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

Brian will connect this time and give an update on the first look at neutral jet reconstruction performance. We are also working towards updates for the calorimetry meeting soon.

Best regards, Leszek

śr., 17 kwi 2024 o 17:21 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

Please answer the Doodle poll:

https://doodle.com/meeting/participate/id/dN59EqNb

Looks like we will shift the meeting to 1000h BNL time on Friday.

Best regards, Leszek

pt., 12 kwi 2024 o 23:02 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

During the meeting today, we discussed energy dependence of the position resolution. It follows the expected 1/sqrt(E) scaling when using most probable value from a Landau fit to the DeltaR_xy distribution. Now we have a few next steps to take:

1. Use RMS value for the scaling vs. energy and tile size. We can also keep the scaling of MPV for reference.

2. The RMS gets smaller at 0.5 GeV, due to clusters made out of single hits. This is just how neutrons seem to interact at low energy. We can try a few options:

 a) generate a more fine grained distribution of neutrons in eta and phi

 b) remove those points from the fit

3. Prepare presentations for next week calorimetry meeting (important to discuss with the experts):

 a) about position resolution study - Alexandr

 b) about clustering check - Subhadip

4. Do a quick check of impact of increasing the tile thickness by factor of 2 (while keeping the steel thickness the same) at eg. E=5 GeV and E=0.5 GeV to check the behavior at high and low energy

5. Since these studies are close to completion, we can think about the next steps. CTU in Prague has experience with light propagation simulation, so perhaps Subhadip could test the performance of putting SiPM in the middle of a tile vs. putting it on the side. You could get help from Ayanabha and Jarda.

David couldn't join this time and I also discussed with Brian that he could start joining regularly. I know that the time is not great for both of them, so we can shift the meeting to a more convenient time. One option is to do it on Thursday after 1200h BNL time. I prepared a Doodle poll:

https://doodle.com/meeting/participate/id/dN59EqNb

Best regards, Leszek

śr., 10 kwi 2024 o 23:07 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

We had a meeting last week and we discussed how to complete the position resolution study. We are going to have a meeting on Friday this week as well at 0900h BNL time:

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

Outcome of the last meeting meeting was:

1. Current clustering scales well with the tile size - only splitting affected. Good enough for now, but we need to give an update on this at the calorimetry meeting soon - please prepare a presentation.

2. We need to see how the position resolution scales with the energy and tile size and where it brakes down. Also a presentation will be needed soon for the calorimetry meeting.

3. How to calculate the resolution from DeltaR_xy? Use mean, MPV (most probable value) form Landau fit or RMS?

 a) We discussed that RMS might overestimate the uncertainty and maybe we need to check other measures.

 b) After further consideration my answer is keep RMS as the measure of resolution, because:

In x,y the distribution will look like "mexican hat", the RMS is a good measure of spread here. When we calculate DeltaR_xy we get absolute values anyway, so RMS still measures deviation from 0. That's why we can keep it as it is.

Best regards, Leszek

czw., 4 kwi 2024 o 11:07 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

As a reminder, we are going to have a DSC meeting tomorrow at 0900h BNL time. Here's the Zoom link:

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

Please let me know if you have updates. We need to finalize the position resolution study and make progress on neutron reconstruction with machine learning.

Best regards, Leszek

pt., 29 mar 2024 o 08:40 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Hi David and Subhadip

In that case we can skip the meeting. David, if you need help, we can still connect and I can provide assistance in accessing or reading edm4eic data. As far as I know in the Python code, you can just change the names of the data collections to HCalEndcapN* to access the nHCal data.

You can find my basic codes here:

https://github.com/lkosarz/ePICSimDataAnalysisTest/

Let me know if that helps.

Best regards, Leszek

czw., 28 mar 2024, 19:49 użytkownik Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał:

Dear Colleagues

We are planning to have a DSC meeting tomorrow at 0900h BNL time to discuss updates. Sorry, for late notice, I was on travel. I know, there are holidays in Czech Republic, so I wanted to ask if Alexandr and Subhadip would like to connect this time. If not, then let me know, so we can talk later. Also, David, what is your situation? Are you ready to discuss your update tomorrow? Please let me know if you are ready for discussion tomorrow.

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

Best regards, Leszek

pt., 15 mar 2024 o 23:56 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

I wanted to summarize the meeting below:

1. Leszek - presented progress on tile testing effort at OSU.

2. Subhadip did additional check of clustering without requiring exactly 1 truth and 1 reco cluster. There are some discrepanciencies for ~0.4% of cases, probably due to splitting. There will be a combined effort looking at that - Cluster Splitting Task Force. Please follow these activities.

3. David was able to fix all the problems with the reproduction of Daniel-Han's results and now will do tests for backward HCal. The plan is to show first results (when they are ready) at the calorimetry meeting to spark discussion on the topic.

Best regards, Leszek

pt., 15 mar 2024 o 09:36 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear All

We are in the meeting now, please note the time change.

Best regards, Leszek

śr., 13 mar 2024 o 19:42 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

We are planning to have a DSC meeting on Friday this week at 0900h BNL time - note the time change vs. Europe. Here's the meeting link:

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

The plan is to have a work in progress updates on:

1. Clustering

2. Position resolution study

3. Neutron reconstruction with ML

David, any progress on communication with Daniel or LFHCAL group? Could you please include me in your e-mails?

Next week, there will be a STAR Collaboration meeting, so we will skip the DSC meeting.

Best regards, Leszek

pt., 8 mar 2024 o 10:56 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

Today we had a very productive meeting. Thanks for the updates! Here are the main points discussed and tasks to do:

It looks like clustering is performing very well - good enough for jet study.

Here are the check for Subhadip and Alexandr

1. Check the performance for all events insted of requiring exactly 1 reco and 1 truth cluster

2. Check logWeightBase impact on clustering performance - adjust it based on the Awes et. al. paper
3. Do position resolution study for: 5cm x 5cm tiles, 15 cm x 15 cm, 25 cm x 25 cm - try to describe the scaling of resolution with tile size using a fit
4. See if adjusting the localDistXY to values greater than 15 cm affects clustering with eg. 20 cm tiles
5. Adjusting logWeightBase may reduce the difference between reco and truth clusters for the outer high material-budget region
6. Check nhits vs. energy for reco and truth clusters
7. Calculate average efficiency and resolution vs. energy to see the scaling - try to describe it with a fit ~1/sqrt(E)

David

1. Contact Daniel and LFHCAL group to understand the difference between their training results

2. Check the ML parameters and send them to us - some may impact the training performance

Best regards, Leszek

śr., 6 mar 2024 o 22:57 Leszek Kosarzewski <leszek.kosarzewski AT gmail.com> napisał(a):

Dear Colleagues

We are going to have a DSC meeting on Friday at 0900h. Here's the link:

https://osu.zoom.us/j/99011057540?pwd=QmNhTDBrNmMyckdWYnJqcmFLMHl5QT09

As we discussed last time, we plan to have work in progress updates on the tasks:

1. Clustering optimization. I contacted experts via calorimetry mailing list. Make sure you follow this discussion. There is also a way to pass parameters to eicrecon, without the need to recompile it: https://eicrecon.epic-eic.org/#/table_flags/flags_view , but please make sure they get passed to the algorithm. I have a feeling that some hardcoded values may remain unaffected.

2. Position resolution study - progress.

3. Neutron reconstruction with ML - update.

Best regards, Leszek