Supernova Detection Efficiency Jos Soto Dual Phase Photon Detection - - PowerPoint PPT Presentation

Supernova Detection Efficiency

José Soto Dual Phase Photon Detection Consortium 18 Decembre 2018

• J. Soto | SN detection efficiency

Content

• Generation of samples: Supernova Livermore and Radiological

model.

• Detector simulation and reconstruction (hit finding).
• Some comments on the reconstructed data for both samples.
• Clustering definition. SN detection efficiency and Background

rate.

• Simple clustering vs Clustering in Single Phase.
• Comments.

2

SN sample – Livermore model

• A ~100k events generated using Livermore is being used to define the PDS

SN trigger.

• Sample generated uniformly over the TPC active volume → the 1m LAr

buffer between cathode and PMTs is not included.

*X is the drift direction. PMTs in X ~ -7m.

*X is the drift direction. PMTs in X ~ -7m.

SN sample – Livermore model

Maximum distance to the PDS in Single Phase geometry is 3m.

• Strong non uniformity of light collection.
• MPV in 4 photons per event (before QE).

Radiological sample

• Sample generated using the model provided by Juergen and Jason. 4130 events of 1ms →

4s of total data.

• Currently, it generates radiologicals within rectangular prisms with sides parallel to the x, y,

and z axes, and within a specified time window (1ms in our case, our readout window).

• It comprises 39ar, 42ar, 85kr, 222rn (prism is the cryostat), neutron prism is the tpc active

volume.

*Above: Number of photons that arrive to the PMT array (left), and number of arrival photons per PMT (right), before applying QE and electronic response.

1 event (all PMTs) 1 event # photons per PMT

• J. Soto | SN detection efficiency

Detector simulation

• Given the number of photons that arrive the photo-cathode (given by the Geant4

simulation), we simulate the PMT response:

–

A quantum efficiency of 0.12 is applied (including TPB re-emission efficiency, from arxiv:1807.07123), to get the number of PE that are re-emitted by the photocathode.

–

Every PE contributed with a SPE signal to the waveform, appyling a Gain of 1e7 (25ADC counts of amplitude per PE). The SPE signal was obtained experimentally from the 3x1x1 PMTs.

–

No saturation nor linearity is applied.

• Other paramteres:

–

Dark Count rate of 1.7kHz (from arXiv:1806.04571), SPE’s generated randomly

• ver the full waveform.

–

Sampling of 250MHz (4ns), 1ms readout window.

–

2V of dynamic range in 4096ADC counts (~0.5mV/ADC).

–

Baseline fluctuation of 1ADC count.

Hit finding

• Hit finding: It is the first step in the reconstruction, it process every

wave-form identifying hits.

–

One hit is characterized by time, amplitude, charge and channel (PMT). It is just one pulse in one waveform.

• How the HitFinder works in our case:

–

AlgoThreshold.cxx is set up as the algorithm.

–

It runs over the waveform, and identify as hit any pulse that crosses a threshold (threshold_start), then the signal is integrated until the baseline is recovered (threshold_end).

• Threshold_start(ADC) = Max(1;1*SigmaPed) = 1ADC.
• Threshold_end(ADC) = Max(1;1*SigmaPed) = 1ADC.

–

SigmaPed is 1ADC, and SPE amplitude in the simulation is around 25PE.

–

We only keep hits with an amplitude larger than 10ADC counts.

–

We don’t consider hits that overlap in time, they would be summed as a single hit. As our signal is very fast, this is not a problem.

–

This set up has been tested in many samples and works fine.

Presenter Name | Presentation Title 7

10ADC

In 1ms readout window. Dual Phase Far Detector 10kt geometry.

Hit finding in the radiological model

Not backtracked hits Not backtracked hits

Single Phase: #PEs per Drift Window Dual Phase: #PEs in 1ms

• Backtracker is not working well → Time propagation is affecting the matching of the art objects.
• Since we cannot backtrack all the hits, we just compare the sum of all backgrounds. Comparing

the total number of PEs:

–

SP: ~ 103 PEs / 3.2ms / workspace

–

DP: ~3·104 PEs/ms/FullVolume = 3·104 x3.2x0.12 = 104 Pes/4.4ms/workspace. DP sees in average one order of magnitude more light from radiological origin than SP. 2xDrift window: from -1.6ms to 1.6ms. Only in the WorkSpace 1x2x6 (12% of the 10kt volume). (Pierre Lasorak)

SN vs BG (ev x<-400m)

SN vs BG (all AV)

Signal median below BGD median. Peak in 1 hit per event.

Non-uniformity of the light detection

• Plot in the left:

–

If we compare the light signal (red) with the background (blue), we see that the maximum of the signal distribution is in just 1 hit, and around 10% of events doesn’t provide any hit.

–

On the other hand, the background mean is around 10hits per event.

• However, the plot in the right shows the same distribution for events at a maximum

distance of 3m to the PMTs (like SP). In this case we can separate well both signals.

SN Clustering

• Reminder:

–

A Cluster is a group optical hits, and will define the PDS based trigger.

–

One Cluster is aimed to identify one SN neutrino interaction.

• We need to define the rules to cluster hits:

–

Length of the time window where to look for the hit coincidence in all PMTs.

–

Define a threshold in several variables to monitor:

• #hits
• #PMTs with a signal over a certain

threshold.

• Maximum distance between PMTs with

signal.

• …
• With the Cluster definition, we can calculate:

–

The Detection Efficiency (proportion of physic events identified as clusters).

–

And the Background Rate (temporary rate of cluster finding in the background).

• We are now starting to study several candidate variables

to define the cluster and its efficiency to select SN events.

SN Clustering

• Clustering should be created looping along the wave-forms, and checking if the conditions of

a cluster to be form are fulfilled… Thus we need to have defined a cluster in advance, and the analysis is very heavy and slow (we need to go tick by tick).

• To do a fast analysis, I split the wave-forms in windows of a constant length (τ), and check

some candidate variables to define a cluster, and store in a tree → Easier to analyze.

• Then we can get the Detection efficiency and Background Rate just doing a query over the

tree.

τ

Cluster candidate hits>10 #hits>5 #PMTs>3 #hits>3 13 hits in 4 PMTs

✔ ✔ ✔

8 hits in 4 PMTs

✘ ✔ ✔

4 hits in 2 PMTs ✘ ✘

✔

DE&BGR DE&BGR DE&BGR

Background level at 250ns

Looking at the number of hits per cluster at different time windows, we see a very different behavior when looking at the whole active volume in comparison when considering the closer part to the PMTs.

• The

detection efficiency drop very fast at very low thresholds (plot on the top) when considering the full

• volume. If we consider only

the events at a maximum distance of 3m to the PTM array, the distribution is much flat.

• At 250ns: We drop from

90% efficiency to 25%, at a level of 30hits per event.

The Non-uniformity in the light detection, affects the detection efficiency.

To evaluate the DE and BG rate, we propose a figure of merit:

–

DE/Sqrt(BG)

–

DE/Sqrt(ε); if BG=0 We will look for a maximum

• f this parameter to

maximize DE and minimize the BGR. We found the maximum at 125ns and 40 hits, and 250ns at 60hits.

Maximum at 125ns and 60 PEs.

Maximum at 125ns and 33 PMTs

The maximum distance between hits inside a cluster seems to be a good candidate to low the background rate, wihtout affecting the efficiency, but this an artifact of having two separated samples.

• J. Soto | SN detection efficiency

Optical Hit Clustering in Single Phase

• 4 parameters define the clustering:

A. TimeWindowOpt: Maximum distance in time between two hits to be included in the same cluster.

• B. PositionOpt: Maximum distance in

cm between two hits to be included in the same cluster.

• C. BucketSize: Maximum duration of a

cluster.

• D. OpHitInCluster: Minimum number
• f hits in a cluster.

B A A A A A

Hit out of cluster Hit out of cluster

C

Cluster found Cluster found

• J. Soto | SN detection efficiency

• J. Soto | SN detection efficiency

Time window (ns) #PMTs #hits Detection Efficiency BackGround Rate (0.25Hz) 125 40 11.4% 250 60 12% 125 33 8% 0.5 250 47 5.8% Time window (ns) #hits Distance between hits (m) Detection Efficiency BackGround Rate (0.25Hz) 125 33 1.5 12% 250 46 1.5 14.6% 125 33 2.5 13.9% 250 48 2.5 15%

Simple clustering Single Phase like clustering

• J. Soto | SN detection efficiency

Some comments

• If we take the best configuration of the previous ones:

–

SP like clustering with a distance between hits of 2.5m, and a time window

• f 250ns: We obtain a DE=15%, but it would be 75% if we only consider the

events at maximum distance of 3m from PTMs.

–

As a result, the detected events are very non uniformly distributed over the active volume.

Position of the detected events

• J. Soto | SN detection efficiency

Comments and next steps

• The detection efficiency we can reach is very limited due to the

non uniformity of the detected light.

– 10% of events without any hit.

• To do: Compute the real DE and BGR using Pierre’s code, to be

able to compare with SP.

• Explore some improvements for the next iteration of the TDR?

Reflected light?