# BELLE2-NOTE-TE-2020-019 review

# General comments

* I believe, you will need to use “Proc11+prompt” ICHEP data if you are aiming to show results at ICHEP and as Giulio mentioned, please add info about your private MC (for example, scripts/GTs/other-relevant in the note).
* The proc9 official data is a combination of Phase 2(0.5fb-) and Phase 3 (2.65fb-) data and your private MC is all based on phase3.

* Is your slow pion in the note are actually filter out for SVD only track ? Like using CDChits <=0 or via tracking information or other method? Do we have number for slow pions from SVD only (that 2% in last section?). Your motivation mainly based on that and I was woondering about means/sigma of those slow pions.

* Information on SVD dE/dx reconstruction seems incomplete. For example, how hit level dE/dx is measured and what factors from SVD can affect its value and also how they are/will-be corrected/calibrated? This basically important to know in a sense that if we want to improve band separation or band resolution further, what are the possibilities?

* Overall calibration procedure should cover more info like current calibration workflow and future calibration plans. For example what is your plans for other un-calibrated or upcoming data, how much (min) data you need to do these calibrations in future? Also the status of your calibration script/tools for prompt vs major processing etc.

* Are you always throwing "2" highest dE/dx hits for the truncation ? Or it is % based truncation? In either case, can you explain a bit more like why you have selected particular number “2” or %.

* There is no plot that proves beta-gamma university in your study. So please add this important plot of dE/dx vs beta-gamma using your samples [pion (slow+others)/kaon/proton]. Same plot you can use to verify sample vs source dependency (e.g. pion from D* or lambda).

* If possible, adopt same plotting style in both calibration plots.

* Correct me if I am wrong. To recover PID performance in MC there was discussion/proposal of using pion PDF as muons as well since they act similar. This is still true or I am totally outdated here?

* If possible, please also add  
* Efficiency and fake rates as a function of costh
* “costh” dependence of dE/dx means (like plot 36) as well. You can use flat dE/dx region of mid-high-p pions for this plot.
* also “phi” dependence of the same to show detectors all act similarly 

* The MC has some odd "smearing" that seems to be absent in the data. Does it have a cos theta dependence? 
* for ICHEP (if applicable)
* List of plot you want to approve?
# List of minor comments
L65: energy loss +in silicon detector + occurs
L70: same +beta-gamma+ curve
L94: after line 94: It would be nice if you add a table w/ list of cuts you used
L108-109: You build pion PDFs using slow pions <0.5GeV or >=0.5 GeV from D0 pions? OR you just combined both to get a full range over p?
L110: extracted +MC+ signal
L111: truth-matched +(TM)+
L116: Are you saying, you combined dE/dx values from different particle tracks ? (please make it a bit clear)
L120: What is that fixed % (please add number or it is always highest 2 hits)?
L123: but +NOT+? for the two highest ones. Here sometimes you must have less than 8 due to inefficiency, or more than 8 due to sensor overlaps.  What do you do in those cases ? always rejected highest two hits?
L127: truncation of mean and sigma: This truncation is different than normal truncation you did for dE/dx track recons, right? Can you please explain what is done here ? I mean what fraction of dE/dx tracks you removed and also how you calcuated sigma here?
L132: reference [7] is not that useful. May be link to a specific document or a presentation instead?
L171: What formula you used to excluded SVD PID only? Also it is a global PID or binary? Same comment for Kaon ID in D* sample as well.
L184: +MC+signal
L186: What is your explanation on difference between Data vs MC (signal yield as well as shape of momentum distributions of Lambda daughters). ref: Fig 11, 12, 13
L200: Add BIIDP-2269 JIRA ticket as reference and mentioned this is based on official proc10 + your SVD calibration.
L202: Correct GT name: SVDdEdxPayloadsOnly -> SVD_dEdxPayloads_Only
L203-204: How much MC you generated again for the performance study (3fb-1?) ? because luminisotiy corresponds to exp8 test sample is ~1.27/fb
L213-214: The same cut is also used for calibration study right? and if yes then you already mentioned at L208-209.
L238: Why 1.2 GeV specifically? Your calibration band plots above are plotted up-to 1.0 GeV only and from 0.8 GeV onwards separation power is really small as bands are overlapping.
L241:242: please add which PID type you used and also PID cut value. It is binary w/ 0.5 PID cut ? Do you have similar plot as a function of costh?
L259: Add JIRA ticket reference here as well.
L260: Please published this SVD_dEdxPayloads GT. Also you don't need to SET any GTs at analysis level. Or you are saying about private processing here? If processing then "online" GT is also in the chain.
L302: I am not sure how mean/sigma is calculated here. Basically I am bit confused with "truncation" term used to build a track dE/dx vs mean here.
L305: On a different topic: you are saying only 2% of pions have <8 hits and that means slow pions that belongs to SVD only is about 2% only? (just thinking about your slow pion study motivation).
L309: here and at Fig 34, nSVDHits are truncated hits or true track hits? Because you are doing truncation by removing 2 highest hits then you might be seeing sigma distribution for nSVDHits==6 instead of 8!

# On Figures
Fig 5: Split like you did for Fig 6
Fig 12/13: Please add histogram are normalized as D* plots are unscaled.
Fig 17: Couple of questions
* Left is 1.277 /fb and right one is 3/fb so if I scaled right one w/ luminosity I get ~9,150 MC signal means still 20% more. I don’t see this much of difference in your calibration study.
* What about normalizing w/ luminosity.

Fig 18/19: data vs MC difference is because of different sample size? If possible, can you please normalized w/ luminosity and see same plots again?
Fig 22:There is a disagreement in Kaon efficiency and fake rate and same is visible in right side of Fig 20 as well. Do we have any understanding of this (probably one of the reason you mentioned related to SVD strip)?

Fig:23: This is nice!! But can you please explain
* Without and without SVD case is just a different PID selection on same sample (privately processed :proc10+SVD?) or it’s between different samples?
* in BII-2269 sample only K/Pi likelihood were updated. Did you checked same ROC after adding proton hypo from Lambda? This basically means ROC w/ sample produced at BIIDP-2693?

Figure 25: Luminosity is not 3 but 2.86 fb- for the run range you used.
Figure 26/27 and 18/19: If possible, can you adopt same plotting style for both set?
Figure31: Same first comment of Figure 23

# Round 2 comments
Intro;
— Do we have an example case (like any previous analysis? or paper ?), where people used PID information on slow pions?

L63: +transverse+ momentum greater
L86: 1917-3123 -> 01917-03123
>Add word "good run" in case if you have used good runs in your calib/perf study.

L91/184: release-04-01-00 —> release-04-00-01? If rel410 is used then why release and GT combination are different? If rel401 is acutally used, then repharase sentence like;
>**From:** For the sample generation we use master 2019-09-26 GT. We use release-04-01-00 for this purpose.
>**To:** We used official basf2 release 04-00-01 and it’s default GT master 2019-09-26 for MC sample generation.

L111: Appendix A;
L409: "0.1-0.7 GeVc" --> This might be not completely true because you don’t see such background tracks after band merging. One can verify this by plotting dE/dx vs momentum (like Fig 51) for generated kaons but asking PDG of pions.

Fig. 51: Do we have understanding on why we see proton like band in kaon band?

Table II:
> 4. 1.0 —> 1
> 8. +global+ kaon ID (without SVD) > 0.2
> 9. +global+ pion ID ( without SVD) > 0.2

L129: by the distance -> by the distance +so called path length+
L131-137: rephrase this
> **From**: To calculate dE/dx, we require the reconstruction of the path length of the particle in the medium. Both the position and momentum of each particle hit are available through the track representation of fitted tracks. The charge information of each hit is also provided by the hit class as assigned to the track. For a silicon detector, the track’s momentum at its hit position is used to calculate the angle φ between the track and the normal vector to the sensor (Fig. 4) [7]. The traversed distance in the detector medium can be then expressed as s = d/cos φ, where d is the sensor thickness.
>**To**: The path length (s) is defined as s = d/cos φ, where d is the sensor thickness. For a silicon detector, the track’s momentum at its hit position is used to calculate the angle φ between the track and the normal vector to the sensor (Fig. 4) [7]. The position and momentum of each particle hit are available through the track representation of fitted tracks. The charge information of each hit is also provided by the hit class as assigned to the track.

L139-140: rephrase this
>**From:** As we want to get a single dE/dx value for each particle passing through the SVD, we look for a suitable method on how to combine information from different SVD hits.
>**To:** The hit level dE/dx information is then combined to reconstruct the track level dE/dx.

L226/L339: means and width -> mean and +RMS+ width
L234/L297: exp10 -> exp10 (proc11)
L236/299: data_reprocessing_proc11 -> Proc11 GTs (data_reprocessing_proc11, online_proc11)
L258-259: this section is missing (add that back and also see my comment for L314/L316)
L269: can be used -> is used

L286/L328: add these text for ROC (but optional):
>In ROC curve, each point represent a pair of efficiency and fake rate for a given PID cut.

L314/316: Numerator: # of proton track indenfied + with a PID cut+

L324/325: You missed explaining your polar angle study (plot 37/38). You can either rephrase existing lines or add fresh lines. Also re-arrange plot order: 34->35->37->38->39->36

L340-347: This study confirms that dE/dx resolution depends on number of SVD hits.
>**Optional:** Now I see, plotting your dE/dx resolution as a function of SVD hits is also nice/imp plot to add in this note. It should be relatively easy, for example, you can just take your clean sample of proton/kaon/pion and compute/plot the resolution vs SVD hits (for 2,4,6,8…).

L360-364: beta-gamma university is good for kaon/protons but pion are deviating from the curve at low momentum (<300 MeV). This is might be because your histogram mean value in this range have some bias (e.g. bands are assymteric with poor resolution).
>**Optional:** You can consider verifying this by calculating mean via fits in each momentum bin (e.g. mean from gaussian fits in the peak region only)!

# Figures
Fig 2, 3, 11, 12, 13, 15, 16, 21, 22, 32, 33, 42
> Optional: Add Y labels, if easy!

Fig 3: These are plots normalized plots right (like 16) ?If yes, then add that to the caption).
Fig 23/24: Captions are identical (add vs momentum and vs polar angle in the caption).
Fig 25/26: Same comment as Fig 23/24
Fig 27/28: Same comment as Fig 23/24
Fig 39: Add "left plot" and "right plot" in the caption
Fig 46: Fix Y axis label typo (should be dedx mean)
Fig 47/48/49: identical captions (add pion/kaon/proton in the captions)

# References
References
— 10 and 11 are same (remove 11)
— move ref 7 to 2 position up for better referencing number at L136.