The next referee's meeting to discuss LiH scattering analysis will take place on Friday 26th February at 14:00 BST.
For this meeting I propose that John should prepare slides that address the previous meeting's actions and any resultant improvements to the analysis, with a view to moving to review of the paper and detailed word-smithing in future meetings.
We will use the regular MICE Zoom meeting room:
[[computing-software:MICE_zoom_info|MICE Zoom Connection]] (login required)
CR mentioned that we will go through the slides to try to focus a bit more on the physics.
JN mentioned - main changes are that we will go through momentum measurement. Changes to p measurement. Added systematic for deconvolution. New table for raw MC vs raw data. Crosscheck about the angular cut off. All results and tables are included.
Slide 7: JHC: I want to understand what is plotted. Vertically is difference between recon and truth? JN: yes. To be clear it is the difference between recon and truth before a correction is applied. First apply correction as in bullet 5 then apply correction as in bullet 6. JHC there is a strong systematic effect. What you reconstruct is different from the MC truth. Have you tried to figure out what is wrong with that equation. JN yes I spent a lot of time trying to figure this out. In the end I decided just to take an expedient correction factor. JHC: where do you get t_e from eq (1) sl 5? JN: I use the TOFs. JHC: does the MC agree with the data? JN: I have to do some fitting procedure. JHC: Is the MC value the same as the calculated value for t_e? JHC are there any electrons that go from TOF1 to TOF2? JN: yes there are. JHC: what is the systematic effect between truth and recon if I make a difference of 100 ps? PS: this is just energy loss. JN: there is no Bethe Bloch correction between TOF1 and TOF2. JHC: It is very systematic with time of flight. JHC: we need to figure out what is causing this systematic. PS: It's just Bethe Bloch. JHC: Why is there that difference between recon and truth? JN: there are some slightly naive assumptions in eq (1). JHC: please validate that we understand the TOF discrepancy. ACTION: calculate the relationship between TOF12 and momentum at the centre of LiH appropriately modified by Bethe Bloch energy loss to validate the momentum rec vs truth curve.
Slide 9: How close is the recon MC to the MC truth? JN: The mean of each bin is 0 - this is how the correction on slide 7 is generated. JHC: there are two questions - how well does MC reproduce truth? How well does MC reproduce recon? CR: We can provide the residuals plot. JHC: yes please. ACTION: provide the 1D momentum residuals histograms for each TOF bin after the corrections have been done.
Slide 11: JHC: Do they contain the corrections on page 7? JN: yes.JHC: why are they not identical? CR: the selection is done on nominal momentum at absorber centre, so the tof bite is slightly different. PS: there will be some energy spread due to straggling. JHC: why is 240 worse? JN: TOF is less sensitive at higher energy CR: energy straggling is more significant at higher p (goes as gamma^3). PS: what are the stats boxes? JN: cor_mom is the full data; mccalc is the reference. PS: channel with absorber is bigger RMS, which is consistent with the hypothesis. ACTION JN to fix the legend to be LiH and Empty, fix the stats box titles correspondingly
Slide 13: JHC: the sign of the residuals is "truth-deconvolved"? JN: yes I think so. JHC: you need more stats. The deconvolution underestimates the truth by several standard deviations. Why is it different theta x and theta y? JN: unfortunately we can't get any more MC data. JHC: you don't need to run the whole MC. JN: I have tried to do a toy MC. The deconvolution is very precise with large statistics. JHC: this shows that the deconvolution doesn't work. PS: do you use the empty channel data to do the deconvolution? JN: I use empty channel MC. PS: That is where you are lacking statistics? JN: yes. PS: low stats in empty effects the measurements as the tails are very depleted.
Slide 13/14: JHC: what is "upper" and "lower"? JN: it is the difference between blue and black points. JHC: you don't know the truth? JN: yes, it is MC. JHC: when you come out of the deconvolution procedure, what do you do with the error? JN: they are included. JHC: do you use the correlated errors? JN: no. JHC: you need to do correlated errors. ACTION: JN to investigate use of correlated errors in the chi2 calculation. PS: chi2 should be done as a matrix multiplication. CR: How do you use the errors from the deconvolution? JN: The errors on the black points are the output "errors" coming from ROOT. JHC: I want to see the raw distributions. ACTION: please provide the MC deconvolution distributions corresponding to slide 13.
Slide 15: JHC: what are the units? JN: (Difference in probability per ~10 mrad bin)^2 JHC: the peak of the fit is offset at 240. JN: yes. JHC: is the asymmetry in slide 13 reflected on slide 15. JN: yes I believe so. CR: Note that the normalisation is different so it is challenging to do the comparison. JHC: I am trying to get an idea what is the actual error on each bin? JHC: Calculating on the fly, errors on bin by bin basis are about 10 %. JN: yes. JHC: ACTION: Please provide a table of bin-by-bin errors. JHC: I am not convinced based on the evidence that the deconvolution procedure is adequate. JN: well, the errors are consistent with other errors in the experiment. CR: if the table of errors shows deconvolution systematic is comparable to other errors, then the deconvolution is adequate. JHC: Agreed.
Slide 17: JHC: I like this table. Thank you. There are two interpretations - 1. G4 MC and G4 model of MCS are good. Therefore if you want to do scattering in LiH ask G4 and do it at you leisure. 2. G4 MC is wrong and G4 model is wrong in a way that "cancels" the wrongness. The deconvolution adds complexity that is not helpful. JN: it is nice to access the scattering distribution. KL: I don't think that there is anything wrong with what is presented. JHC: it is a bit more indirect. CR: It is desirable to present the deconvolution as well. JHC: What do we make of the Kolmogorov table? JN: I need to check. KL: Do we need KS test? JN: it may be okay not to present it JHC: Is the chi2 purely statistical? JN: no systematics are included.
JHC: Looking at fig. 11 e/f (MC vs data theta x and theta y distributions). ACTION Remake the table on slide 17 but with stat errors only.
JHC: Is this a shape test or an absolute test? JN: it is an absolute comparison.
CR: TOF systematic - if the MC happened to reconstruct at 70 ps higher than the data, how would that comparison of slide 17 be dealt with in the analysis? Say one is making the comparison. PS: the problem is that the events can migrate from one bin to another. JHC: it's not a strong function of momentum. PS: well, yes, we have quantified that. JN: clarifies, the plot shown in fig. 11(e/f) are data in the 200 ps time of flight around 200 MeV/c. JHC: can you add tables for all the intermediate points? JN: it would mean quite a bit of CPU time. I don't have the systematics analysis done for all the data points. I will need to drive lots more MC. CR: I am not sure that it would be beneficial.
Slide 18: JHC showed some slides. JN: we should look at the distribution upstream. JHC: no. CR: showed toy MC data. 97.7 % transmission in the central bin. JN: this is consistent with the data. JN: ACTION: I will include the acceptance distributions at other momenta. JHC: one could consider doing an iteration as the estimate of the acceptance can be affected by the acceptance - feedback loop? PS: No the G4 simulation should be good enough. CR/JHC: We agree that we don't need to do that.
Slide 21: JHC: What does it mean? JN: It is the ratio of the number of tracks downstream to the number of tracks upstream. JHC: So the sample downstream includes the scattering upstream. What do you do about inefficiencies in the tracker? JN: They are included in the acceptance plot. CR: we expect them to be small.
Slide 22: JHC: ACTION please show the probability plots on a linear vertical scale as well. I can't tell how good the agreement is.
Slide 25: JHC: what do we conclude from the figures? JN: we have pretty good agreement. PS: Better agreement with Geant than Moliere. JHC: is there a ripple? JN: not sure how much I should read into these. JHC: It depends on the systematic. JN: the errors on the black dots are the total errors. The errors on the blue dots are statistical. JHC: there are strong correlations between deviations. PS: they are correlated by the deconvolution procedure. ACTION: please repeat the normalised residuals plots for forwards convolution case.
Slide 26: JHC: ACTION need to include correlated errors when calculating chi2.
Slide 28: JHC: the deconvolution shows that we have "unfitted" the data. KL: we should not include the P-value and the Kolmogorov. PS: I have just included the chi2 when writing the paper. JHC: the deconvolution seems to indicate that we don't agree well with Moliere. How confident are we? JHC: ACTION we need to check the Moliere calculation.
CR: We had a question as to how wide the distributions should be measured to - in particular concerns about the acceptance of the tracker and the width of the empty dataset. KL: I believe 0.045 mrad would be a more valid limit. MB: ACTION change the plots for the paper to 45 mrad but keep the plots in the note 60 mrad. JHC: yes I am happy with that. JHC: can we do probability per mrad not probability per 3 mrad. JN: That would be an endeavour. Why? JHC: ACTION don't have to change the analysis, just change the units. Plots should be probability per mrad. JN: Okay.
General discussion at the end. There is a range of validity for forward convolution and a systematic error associated. ACTION: add this as a source of uncertainty.
Agree to distribute a paper at the CM57 assuming that nothing turns up following from the actions outlined above. Formally we have to write to Maurizio. Request that the referees help with the wording. Target journal PRD. Possibly EPJC but average refereeing time 6 months. We will have a follow up meeting on Friday 19th March at 14:00 to review the actions outlined above and discuss wording of the paper.
Paul Soler - Link to draft 0.1 of LiH MCS paper