June 4-5/26 PionLT/KaonLT Analysis Meeting Notes ------------------------------------------------ (Notes by GH) This week: Gabriel will go first Please remember to post your slides at: https://redmine.jlab.org/projects/kltexp/wiki/Kaon_LT_Meetings Thursday: Present ----------------- Regina - Garth Huber, Nathan Heinrich, Alicia Postuma, Dex Yadlowski, Vijay Kumar, Nermin Sadoun Virginia - Richard Trotta JMU - Gabriel Niculescu, Ioana Niculescu Ohio - Julie Roche CUA - Tanja Horn, Chi Kin Tam Gabriel ------- Bin Centering in KaonLT analysis - goal is to determine Sigma0/Lambda and Lambda*/Lambda ratios for high epsilon, higher energy data, taking advantage of full statistics (no diamond cut applied) - believe they understand the analysis workflow, looked at a variety of sources, including: Mohring, Volmer, Horn theses, Blok paper, V. Klimenko (Hall B 2024 thesis) - Mohrning applies an explicit bin centering correction, however, to apply it here, would need to preserve all SIMC tries, not just successes - looked more carefully at the Volmer+Blok MC Ratio Method that we use - justification of Ratio Method needs clarification - similar to Cauchy's Mean Value Theorem, which states there is a point where f(x-mean)=f-mean(x), but this point is not necessarily x-mean - one can argue that the relation is okay provided the bin size is sufficiently small - did a pseudo-data study, generated MC data divided into 5 t-bins spanning 5-95% coverage range, nearly equal statistics/bin - for each bin, find average Q2,W - calculate sigma-ave(Q2,W) vs sigma(Q2-ave, W-ave), find they differ by 2-3%, giving a measure of the non-linearity of the cross section, clearly depends on the granularity of the bin - looking at the convolution: f(x)_measured_distr=Int(a,b)[ g(y)_physics_distr * r(x,y)_det response]dy - going from g(y) to f(x) is straighforward, the backward transformation f(x) to g(y) is more complicated - tried various functions for g(y), e.g. Lorenzian+Gaussian kernel, and see if can recover g(y) from f(x) using the MC Ratio method - sees no problems at ~5% level, but random deviations between g_extracted and g_input with sigma~3% - did only 1 iteration, would things be better if do more iterations? - tried g(y) with phi-distr+Gaussian kernel, fit g=a(1+b*cos(phi)+c*cos(2phi)) and do up to 20 iterations - find a strong correlation between a,c terms, fit did not converge with more iterations - Julie: when doing this on data, one uses the experimental error bars and the improvement of chi-square vs iteration to prevent over-fitting, need to quantify whether the fit is "good enough" - if the cross section no longer changes, then it's good enough - also asks about the wrap-around from phi=-pi to phi=+pi in his analysis, where the data wrap-around, but the fit function has a rapid decline to zero at these points - Tanja: agrees that the bin size is very important to the accuracy of the MC ratio procedure - It also depends on whether the cross section is changing rapidly (e.g. at the phi=-pi,+pi points in his study) while the exclusive reactions generally vary slowly with kinematics, the only real exception being pi+n sigL at -tmin and we need to take care to bin more finely there - inclusive studies are more susceptible to this, as it's much harder to identify experimental background leaking into the signal. But in exclusive reactions the MM reconstruction is much cleaner in comparison - Garth: the PionLT t-phi binning is not larger than Fpi-2 binning, the bin sizes between 12 GeV and 6 GeV data are about the same. We paid attention to this when designing the experiment - Vijay: did not find this problem of non-convergence in his LT-sep fitting - found that parameters did not change at all after a few iterations, both - the fit parameters and the experimental yields were stable, no need to keep iterating if nothing changes - *NB* suggestion to Gabi was to incorporate the experimental errors in the procedure as per Julie's suggestion, as the pseudodata study seems to exhibit instabilities we do not see in practice with the data - other than this, no easy answers were offered Friday: Present --------------- Regina - Garth Huber, Alicia Postuma, Nathan Heinrich, Nacer Hamdi, Dex Yadlowski, Vijay Kumar, Nermin Sadoun CUA - Chi Kin Tam, Tanja Horn Virginia - Richard Trotta Nathan ------ Difficulties converting scripts for JLab new batch submission system, from SWIF to SWIF2 SWIF: add-jsub Deprecation The legacy add-jsub command has been deprecated and will be removed on June 15. Please transition to the add-job command or the import features as soon as possible. Hanjie Liu: If it's using the Auger job submission format, it needs to be modified to SWIF2 format. https://jlab.servicenowservices.com/kb?id=kb_article_view&sysparm_article=KB0015611 Scientific Computing - SWIF - Scientific Workflow Indefatigable Factotum - Knowledge Portal Scientific Computing - Workflows - SWIF2 is the recommended tool for managing groups of related compute jobs. It manages inputs and outputs Brad's hands-on tutorial: https://indico.jlab.org/event/1069/contributions/18589/attachments/14051/22727/Farm_Job_Examples-swif_slurm-May2026.pdf Nathan's new scripts: https://github.com/heinricn/UTIL_BATCH/blob/master/batch_scripts/run_batch_FullReplay_PionLT_newswif.sh - no test batch queue is available, so progress is slow - enountering EXIT_CODE=13, which indicates a permission issue, but no information is printed out to diagnose the problem Richard's template for new system: https://github.com/trottar/farm_template Alicia's: https://github.com/acpostuma/UTIL_BATCH/blob/LTSep_Analysis_2022/batch_scripts/run_batch_uchannel.sh Alicia didn't immediately see a difference between this and Nathan's script but this one ran. - PionLT Q2=5.0 LT-sep - did 2 iterations, brought down high epsilon Data/MC ratios a lot, but not uniformly - fixed a 5 MeV error in P_HMS in SIMC input file - was due to Junaid's ad-hoc high momentum P_HMS offset, which was not removed although it should have been - Q2=6.0 - still missing some ROOT files, which requires job submission scripts to work first Nermin ------ PionLT LD2 Lumi studies - last week: showed HMS scaler yields combined over all 6.4, 9.2 GeV runs gave boiling correction: (-3.75% +/- 0.19%)/100uA - SHMS was not consistent w/ HMS - plotted normalized yield vs rate to see if the discrepancy was a rate issue - HMS rates up to 220kHz, with highest 220kHz point ~2% lower than linear trend - SHMS rates up to 1.25 MHz, non-linear dependence with over-correction at high rates - removed 1 HMS point and all SHMS points >200 kHz, then both spectrometers have similar trend, although deviations up to 3sigma are still seen - fit only <200kHz data: HMS (-3.137 +/- 0.382%)/100uA SHMS (-2.439 +/- 0.564%)/100uA Combined (-2.753 +/- 0.315%)/100uA Apply the combined slope to data: obtain a residual slope of (-0.273 +/- 0.683%)/100uA - Garth: red and black sets (one HMS, one SHMS) disgree by up to 3 sigma, so they're not consistent within statistical uncertainties, indicates an additional source of systematic error - Nermin: planning to add a systematic error to the -0.273%/100uA slope result using Nathan's standard-deviation method - *NB* Nathan: should look also at the 6.4 GeV tracked/untracked data using EDTM instead of scaler with ELLT (single arm scans) - if the -2.753%/100uA slope is correct, then the 6.4 GeV tracked data shold be flat - Nermin: planning to compare TLT from EDTO to ELLT*CPULT for these scans - Garth: it's possible CPULT will have a problem, the real criterion will be to see what slope the tracked/untracked data give Vijay ----- Low Q2 PionLT systematic studies - pi+ decay systematics - shows a plot of SIMC Decay_ON/Decay_OFF vs -t for mid-eps center setting - decay correction is 10% - trend vs -t looks small, but the SIMC statistical errors are large, running more SIMC statistics to see this better - Garth: the idea is to see of the decay fraction changes much across the acceptance, the pion decay fraction presumably depends on the spectrometer path length in the simulation, and the question is how sensitive we are to errors in that determination - if the dependence across the acceptance is small, one can argue the systematic uncertainty is small - if there is a 5% error in the correction, the systematic uncertainty for the data is 5%x10%correction=0.05% - Tanja seems generally happy with this approach, agrees the uncertainty appears to be <0.1% - target thickness systematic - looked up Dave Meekins target geometry report (2018), and calculated the variation in the caliper measurements - obtained a variation of +/-0.027% - the same target stack was apparently used for KaonLT and PionLT, only a different cell in the stack - EDTM TLT systematic - Nathan: the number he came up with was somewhat ad-hoc, probably reasonably correct also for Summer 2019 data - Vijay plans to look at TLT vs rate and determine the standard-deviation of TLT at a given rate - there are no CoinLumi studies in the Summer 2019 data to look at - *NB* Nathan agrees this is a sensible approach, however you need to be careful to have good beam current cuts to be sure you are comparing rates at as close as possible to the same rate - might have to make the current cuts tighter than usual for this study - Tanja: asked whether these systematics are included in the draft paper that was circulated to the spokespersons - No, these are still in progress. Garth will redo the model fits with the systematics after they are done. The parameter central values will probably not change much, but the error bars will be more realistic - the circulated paper is to give the spokespersons a good idea of what the paper will look like, and to invite comments Alicia ------ u-channel - mostly working on slides for Hall A/C (u-channel) and JLUO (BSA paper) talks - has new Q_effective including Aerogel Knock-On Correction - recalculating yields and running a few iterations to get some plots for her talk Nacer ----- KaonLT Low Q2 systematic studies - varying acceptance cuts, MM variation cuts, pi+n leakage normalization cuts - MM variation cut was too large, took +/-5% of 1.08 GeV cut value, not the width of the cut, giving a 50 MeV varation, which is a large change to where in the MM histo the cut is applied - Garth suggested applying the % variation to the width of the cut instead - similarly the pi+n leakage variation cut (to select normalization) of +/-5 MeV seemed large - Vijay used 2 MeV cut variation in his MM cut study Chi Kin ------- KaonLT analysis issues - HMS saturation correction - compared w/ Alicia's implementation: applied only to half of the focal plane if (hsxpfp>0) hsdelta += 3.0%&hsxpfp - applying this correction straightens the MM vs hsxpfp "banana", gives a reduction in Lambda peak sigma by ~5 MeV - also compared C0 parameter vs P_HMS - lots of scatter, hard to see trend - linear fit C0 = -0.393*P_cent + 1.018 - revisit pi+ leakthrough subtraction w/ improved MM resolution - showed a plot of pi+ leakthrough distribution, sees a small Lambda peak in pion dist, apparently the pi+ sample is not sufficiently clean - *NB* Richard: the pi+ selection cuts need to be tightened before subtracting from K+ data - will compare background subtraction results w/ Richard - Heep offsets - Garth & Kin offsets somewhat different, but give simiar Heep replay result - replayed some physics data to see if MM shift is reduced however, used CKT individual beam energy fit offsets, not global offsets Q2=3.0 W=3.14: high epsilon MM shift is a bit smaller. Was: 10 MeV. Now: 6-7 MeV low epsilon ~2 MeV shift - *NB* Garth: suggests to try Garth and Kin's global offsets from 3 highest energies to physics data, instead of individual energy offsets Richard ------- KaonLT analysis - replaying Q2=2.115 setting, ran into batch difficulties similar to Nathan - switched to Q2=3.0 W=3.14 w/ new HGC-hole and Diamond cuts - big background, even at low epsilon (where RFtime cut is available) - HMS saturation correction won't make a big difference here due to P_HMS - the pi+n peak in K+ data is small, an empirical background fit is still probably needed - Garth: as Gabriel has noted earlier, there is no guarantee that the pi+ leakthrough spectrum determined from PID cuts is exactly the same as the actual leakthrough in the K+ data - *NB* would it help to normalize the pi+Delta0 region instead of pi+n, in the cases where the pi+n peak is too small? - Richard will look into this Next Meetings ------------- - Thurs: June 11 @ 13:00 Eastern/11:00 Regina - KaonLT will be first - Fri: June 12 @ 12:00 Eastern/10:00 Regina - we will continue where we left off