Goal : make a fine QA of run 7 in order to :

1. remove junk runs
2. decrease the number of runs/events to analyze
3. from list1(all runs)  ➞  list2(event-based QA)  ➞  list3(track-based QA) ➞ list4 (D0 candidates-based QA)

This can be done by making a cut on nsigma of a given distribution VS sequence to remove the outliers.

The variables to QA that I've chosen are :

1. V_r = √(V_x² + V_y²)
2. V_z
3. widths of V_r and V_z fit
4. ResV_r, ResV_z : primary vertex resolution 
5. mean number of tracks (per sequence) with SSD hits , same for SVT

note :

1. the rank Id will be set to the first vertex (see plot below)
2. we can also monitor the sequences vs TriggerId
3. there are still some hard cuts used :
   
   1. |V_z|<30 cm
   2. triggerId = 200001,200003, 200013
   3. p_t >0.1 GeV/c
   4. |η|<1.2
   5. NTPCfit / NTPCposs >0.51
   6. |DCA_XY|<3 cm

A RunId is coded as (for example) : 8102017, where :

• is the year
• is the day number
• is the sequence Id

A) preparation plots

Fig. 1 : Vertex Z resolution vs vertex rank

Vertices with rank >0 have large vertex Z resolution 

Fig. 2 : Vertex Z position vs sequence id of day 8109

From this plot, a double gauss is used to get the mean and width of the distribution, for each sequence

Fig3 . : <Vz> vs sequence Id for day 8109 : the error bar are the width of the double gauss fit (the narrower one)

The rest of the plots for day 102 to day 109 is here

note : plots for day 102 are almost empty because the vertex Z distribution is not centered at 0, 

B) other preparation plots

Not exhaustive but :

1. Vr vs sequence id : here
2. Resolution Vz vs sequence id : here ; for this kind of distribution, I will use a landau fit. The fit is not really good but the MPV of the Landau fit is closed the peak of this distribution
3. SSD hits vs sequence : here

note : the plots for the SSD hits is similar to previous studies

C) plots for Vr

The distribution of Vr vs sequence is here.

The results of a double gaus fit is here

update 9/5 : Vr for day 100 to day 120 : here

D) results for level 1 QA

1. This pass will be based only on the primary vertex resolution in XY and Z
2. a 2σ window is chosen to keep or reject runnumbers : outliers for Vr , outliers for Vz
3. for ResVz : list before and list with runs to remove 
4. for ResVr : list before and list with runs to remove 
5. final list to remove , matched between ResVr and ResVz

E) improvements for level1 QA : here 

F) QA -level 2

We used a cut based on the number of silicon hits per tracks.

At this step, I'm evaluating the mean (per run number) of the following quantity :

Bool_t conditionK1 = (NoFSvtHits>1);

Bool_t conditionK2 = (NoFSsdHits==1 && NoFSvtHits>0);  

Bool_t conditionK3 = conditionK1 || conditionK2;

So a track should have 2 silicon hits :

1. 2 SVT hits
2. or 1 SSD hit and 1 SVT hit

A preliminary plot is here (for day 98 to day 109)