1 Some reminders 2011 flagging policy : – Event by event flagging. Correlation between express stream and CosmicCalo – Luminosity blocks flagging. Impact on data loss Study of time development – From cell to noise burst. Further background studies.

2 Reminder – From 14/12 ATLAS weekly ppt Large and coherent signal in a whole partition: – EM endcap (EMEC) and/or hadronic endcap (HEC) and/or Forward Calorimeter (FCAL) – EM barrel (EMB) in 1 or 2 sides Run / LB 363 – CosmicCalo stream EmecA Run / LB 439 – CosmicCalo stream (L1Calo in empty bunches) EmecC HECC Cells up to GeV

3 Reminder (2) - From 14/12 ATLAS weekly ppt Characterized by the yield of channels per event in positive 3  tails (Y 3  ) - Also visible in total energy per LB  map indicates suspicious injection points or weak regions: Noise bursts CosmicCalo :gaussian behavior (0.13% peak) Affected LB =20GeV Also visible in total energy per LB % of events in 4  tails Sampling 1 Sampling 2

4 The common ntuple Software development by Josu Cantero Garcia to provide a unique ntuple designed to contain all relevant variables for noise burst studies: – See ppt at February mini LAr week. – Typical variables : Y3 , LArNoisyRO output, cell energies, ADCs, ATLAS background words… Software available in SVN (LArCalorimeter/LArMonitoring – LArNoiseBursts algorithm). Ongoing work with the help of Marteen Boonekamp to have it automatically produced in the reco chain. Waiting for this, privately produced ntuples for most important runs available in : – ~ trocme/Atlas/Batch/ – Available for everybody ! A lot of plots shown today are extracted from this ntuple.

5 Flagging the events and the LBs

6 Event per event flagging 2 methods so far implemented in LArNoisyROAlg: – LArNoisyRO_Std : >5 “noisy FEBs” (A “noisy FEB” have more than 30 cells with a q factor > 4000) Flag only severe bursts. Still recommended for analysis. Flag stored as WARNING in EventInfo::LAr (and bit 0) Should be perhaps advertised again (together with data integrity) – LArNoisyRO_Sat (Medium/Tight) : > 9/20 channels with Energy > 0.5/1 GeV and q factor = Flag almost all events with large Y3 . Flag stored in bit 1 and 2 in EventInfo::LAr Not yet recommended due to peculiar behavior of q factor in inner wheel : risk to overflag collision events.

7 Event per event (fake) flagging : Std method Due to incorrect computation of q factors (especially in inner wheel), a collision event may be wrongly flagged as “Noise burst” and rejected. – Hypothesis tested by considering correlation of number of events per LB flagged (by Std method) in both Egamma and CosmiCalo. – 1 entry per LB – Different lengths of bunch groups to be taken into account Express/CosmicCalo : 194 / 2663 BX Run (5% of LBs affected by noise burst)

8 Event per event (fake) flagging : : Std method Due to incorrect computation of q factors (especially in inner wheel), a collision event may be wrongly flagged as “Noise burst” and rejected. – Hypothesis tested by considering correlation of number of events per LB flagged (by Std method) in both Egamma and CosmiCalo. – 1 entry per LB Correlation follows statistical rule due to bunch groups lengths. – Not surprising as LArNoisyRO Std flags only very severe noise bursts (cf : requisite of 6 “noisy FEBs”) No apparent overflagging Y = X *194/2663 (different lengths of bunch groups)

9 Event per event (fake) flagging : Sat method Same analysis applied by using the saturated tight and medium methods.

10 Event per event (fake) flagging : Sat method Same analysis applied by using the saturated tight and medium methods. – No apparent fake flagging due to good collisions for SatTight. – Obvious fake flagging due to good collisions for SatMedium. However prefer to wait for new q factor + extended studies + increased number of bunch groups to recommend the use of saturated tight flag. No apparent overflagging! Obvious overflagging! Y = X *194/2663 (different lengths of bunch groups)

11 LB flagging : new 2011 policy flagging LB “flagged” on the basis of CosmicCalo stream: – New set of DQ defects. 2 types of defects available : – Yield of events flagged by LArNoisyRO_Std different from 0 : SEVNOISEBURST (intolerable) – Yield of events flagged by LArNoisyRO_SatTight > 5% or LArNoisyRO_SatMedium > 10% : SEVNOISEBURST (intolerable) – Yield of events flagged by LArNoisyRO_SatTight > 2% or LArNoisyRO_SatMedium > 5% : NOISEBURST (tolerable) Hope to keep it uniform as long as possible.

12 Noise bursts defects in 2011 Tolerable Intolerable Almost no “minor” (tolerable) noise burst : our policy is conservative. Not fully comparable with 2010 but rate of burst seems to be roughly similar. Plots by J.Leveque

13 Trying to understand the origin…

14 From cell to noise burst Run – 23/2 – ISEG week (unstable HVs) – Few bursts observed but nothing dramatic compared to collisions runs (see slide 6) – Also some channels with weird behaviours Not specific to this runs – seen in many circumstances. EMECA 14/10/17 Tot energy = 1.6TeV !

15 3 TeV in a EMEC cell (2) Run Event Rather quiet except in this EMECA cell : 2% of cells in 3  tails (0.13% expected) Not spotted by LArNoisyRO_Std (6 noisy FEBs - EventInfo::WARNING) Spotted by LArNoisy_SatTight (>19 cells with saturated q factor and E>1GeV) NB : just after this 3 TeV event (!), there is a -1.4TeV event (!), explaining the summed energy of 1.6TeV of previous slide.

16 3 TeV in a EMEC cell ( the same one)

17 From cell to noise burst : time development of the burst Consider only events around this particular one and use BCID as a clock (more reliable than event time): – Found 4 events within 24 BX ( ) – Development of noise bursts indicated by positive and negative Y3  (2% -> 4% -> 2%) – Superimposed are the samples of dramatic cells EMECA 14/10/17 when available (if not, means that the cells is within 3  noise). Presence of huge undershoot. “Positive noise burst” delayed by ~14 BX with respect to this single cell peak (seed?) Previous event display is at BCID=1206 ?

18 Occupancy cumulated over the whole LB EMECA 14/10/17 is here “Burst” well localized in a given HV sector + neighbouring ones. Topology different from the standard one (ring at  ~ 1.4)

19 Similar behaviors in other runs/conditions? Method : – Consider runs with several noise burst. – Identify bursts (events with a large Y3  ). – Try to find events with similar timestamp (within 1 second) and extract the cell with largest energy at the lowest BCID – Study ADC evolution of this potential “seed” cell. Limitations: – Use of BCID reliable with usual trigger rate but no longer at large rate, BCID is no longer a reliable “time estimator” (1 turn = 3563 BCID = 90  s). – With aphysical/distorted pulse, searching the cell with the largest energy is perhaps not optimal (should use raw ADC instead). – Due to trigger constraints/time window, no assurance to see the burst from its beginning. Due to this, no clear evidence of a seed in a lot of bursts, but managed to find some. Examples hereafter in different conditions and partitions.

20 Looking for a seed cell No trip known at this time

21 Looking for a seed cell (2) In this case, we arrive probably few BCIDs too late No trip known at this time but 3 trips in EMECA in the end of the run. Localization to be checked.

22 Looking for a seed cell (3) No trip known at this time

23 Looking for a seed cell (4) No trip known at this time

24 Looking for a seed cell (5) Same run as in the first case

25 To summarize In some cases, it appears that at the beginning of a burst a fast and large signal is observed in one (or very few) cell. It is followed by a large undershoot and a come back to baseline when the noise burst is developing. – It was observed in almost all partitions (embc, emeca, emcc). – It was observed when the HV was known to be unstable but also in apparent stable conditions. – It is very difficult to assess whether it is always the case.

26 And what about background?

27 And what about background? The empty bunches are known to be not perfectly clean of backgrounds/collisions. A single event may fake a noise burst : – Energy deposit → large Y3  – If out of time event (or background), event may be flagged by LArNoisyRO (Sat or Std). It is important to understand the level of background/collisions in CosmicCalo. NB : we observe a permanent (  ) structure more or less stable among time (ex : band at  ~ 1.4, weaknesses in particular  ) → it is excluded that poor background can explain the noise bursts phenomena (it really exists!). It can however induce an overflagging of some quiet LBs. A lot of work with ATLAS non collision background task force to try to understand conditions and use other subdetectors data. – Thanks to D.Berge, M.Huthinen, J.Boyd!

28 Looking for background with all subdetectors New bit available in EvenInfo to flag collision/halo candidate on the basis of MBTS/LAr/Bcm/Muon time diff, Pixel/SCT/Lucid multiplicities: – Code by J.Boyd available here: /RecBackgroundAlgs/trunk/src/BackgroundWordFiller.cxx#L58 – Not yet fully available and commissioned on ATLAS side. – Today’s results very preliminary from a private processing of 400k of events of run

29 Correlation between all detectors Tuning needed? Afterglow effect? LAr reliable to spot collisions/halo

30 Collision/backgrounds contribution to Y3  tails? Ratio <1 indicate samples enriched of events flagged as “Collisions/halo” by LAr (signal on both endcaps) – 154 events/400k. No clear impact.

31 Collision/backgrounds contribution to Y3  tails? Ratio <1 indicate samples enriched of events flagged as “Collisions/halo” by Pixel (large multiplicities) – 114k/400k. No clear impact.

32 Collision/backgrounds contribution to Y3  tails? (2) Ratio <1 indicate samples enriched of events flagged as “Collisions/halo” by SCT (large multiplicities) – 184k/400k. No clear impact.

33 Collision/backgrounds contribution to Y3  tails? (3) Ratio <1 indicate samples enriched of events flagged as “Collisions/halo” by Bcm (signal on both sides) – 135 events/400k. No clear impact.

34 Conclusion Event by event flagging: – LArNoisyRO_Std/LArNoisyRO_SatTight seems to be rather reliable (no overflagging). – LArNoisyRO_Std still strongly recommended from now on. – Prefer to wait the n ew q factor to also recommend LArNoisyRO_SatTight LB flagging policy : – Uniform, constant and very conservative policy. – Data loss relatively under control (3-5%) Some facts about time development : – Existence of a seed cell to be confirmed/understood Background studies: – Noise bursts not beam halo/collision (confirmation). – No correlation found neither.