If you check blog post about residuals in Run18 and Run19 FXT, you will see the issue that lead towards reexamining the alignment in Run19.

Pass600 is the result of previous alignment;
Pass605 is the last result after my passes converged well enough.

At this point you can go back and check the results of the residual after the new alignment - very good result. [could it be better though? keep reading]

Pass600

Pass605

Plots above show the profile of the z shift before and after the new alignment. Even though there is improvement of about 2.5mm they do look funny. So look deeper.

Detailed look at the z shift showed the double peak in the distribution across the chosen data which ranges from day 48 (this is day after putting back the GG drivers, therefore after the noise issue) to the last day of Run019.

The double peak could be separated before and after day 53 [February 23] - this is when we started taking 19.6 GeV AuAu data.

This changes was traced to new parameter set on tpcSectorT0offset.20190223.*.C. Reverting those parameters for day 53 gives following results:

Why was this shift added? Should we counter this shift in other place ?

Here is the answer to the above question and the fool story understood:

Run-19 was a first run with all iTPC sectors. Therefore it was important to align the electronics timing for the first time (even though we had a single sector 20 of iTPC in Run-18, Run-19 was still the first look at the whole system). Run-19 started with cosmics and then some pre-collision tuning etc. The AuAu 19.6 GeV first collisions of Run-19 came on day 56, some initial tuning a day or two before. The timing is aligned using the prompt hits. The detector alignment using the cosmics was done by Yuri with cosmic data taken before collisions. The prompt hit based results came after Day 53-54. There was about 2.5 mm worth of timebucket shift. These parameters resulted in different shift seen as Z shift in the above plots for the different cosmic data. After discussing with Yuri we decided that, I will take one more look at the prompt hits, establish the final set of parameters (if any different from what is already in posr 20190223 set) and then apply these to entire Run19 and redo alignment to account for the shift (2.5 mm or so). Those results are below.

So first thing I did was, check the prompts again and see if we need to do one final tune. Below are the plots for inner and outer sectors for the prompt hits (those deviations are set as T0 for each sector separately inner and outer).

Below are the plots of my alignment passes (entire history) for correction parameters x,y,z, a(alpha), b (beta), g(gamma). You will notice that is started way off (especially in Z), that is the big shift that needed to be corrected, it was corrected at pass 6. At that point I payed attention at the double peak in Z and started investigating different options with and without it, hence the jumping in parameters. At pass number 10 the entire dataset has single tpcSectorT0 parameters. At pass 11 I switched the tpcSectorT0 parameter to single value for entire Run19 to those that I obtained (shown on above plots as red points).

Here is a single view summary of the alignment efforts:

SuperSector Aligmnet
After alignment of inner sectors with outer sectors we need to look that entire TPC is well aligned sector by sector. Looking at TpcSuperSector alignment with new prompthit T0 and alignment from cosmics we see following:



Black points are the alignment parameters after putting in the correction from the initial parameters (red). These should be brought to zero.

After few iterations everything was brought to the acceptable level. See the animation of the process below. Notice Sector 3 at the end. That is because I was in parallel tuning the inner vs outer sector 3 parameters by almost doubling the used cosmics data. And I plugged it in when ready. That affected the supersector naturally and was corrected along with final tuning of all together.

Update on June 3rd:

Gene and Babu noticed that there is still some remnant in the data, that looks like negative correction of the Space Charge, which is unphysical and  might be to some remaining misalignment.

I checked the supersector alignment in a very low luminosity 19.6 GeV collision data (low zdc/bbc rate) and also compared the results when implementing SC correction and not implementing, results are below:

There is indication of a supersector misalignment and also some indication that in local Y dirrection having SC correction in even in this low luminosity run has noticable effect.

Next I checked what would removing the SC correction from the run I worked on before (obtained results in above supersector calibration work) had any effect. Results is below:

This indicates that the remaining supersector misalignment is most probably due to me having the SC correction in while doing the alignment last time. The difference seen in the plot above between with (BLACK) and without (RED) SC Correction in chain is about same as seen in the remaining misalignment int the low lumi plot (above the above one). Also the supersector positions have much less scatter without SC correction - perhaps something to think about (the scatter is seen in the low lumi plot though, because I introduced it while making corrections to the alignment with the SC correction in)

UPDATE: August 5, 2020

I wanted to make sure that there is no variation in prompt hit timing during the run. I took the data produced by Gene (event.root files) and grouped them by 50 in each and looked at the spread. Looks like everything is in order here.

I also looked at the slope of the timebucket vs. padrow distribution for every sector in every range (50 files at once) and they are also very consistent.

I did similar study, similar splitting of data as for the above plots, for the supersector alignment parameters. Results are below:

After this I removed the injection runs and plotted these parameters as function of time. In the below plot about 850 non-injection runs are bunched 10 runs for each point:

I also used the limited number of files available from the TFG20c express production of this data and added it on top with the same granularity:

The agreement between dev and TFG20c is pretty good. So I conclude that there is no difference there.

Below is the same plot with the 100 micron limits, question is, what do we blame for the change of the alignment as a function of time?

Drift velocities were also updated to remove the sector 16 which produces outliers in the measurements sometimes. After this Space Charge calibration was also done (although the claim is that the change was insignificant). Gene reproduces a limited sample after these two recalibrations and below are the results (binning in here is not same as in the above plots, because the statistics was different and smaller number of runs):

This shows that the kinks in local X and Z directions are gone, but I see a large offset (somewhat constant) in the Y direction. This needs to be adjusted.