Meeting minutes for first period:

2016


07/06/2016:
  • New transport forward matrix is obtained:
    • For y_focal: use 56 data points and 23 parameters.
    • For phi_focal: use 56 data points and only 14 parameters (0th, 1st, 2nd and cross terms).
    • Result agree good between simulation and data, plots are here.
  • Jianping suggested to use SNAKE to go from focal plane to target plane, then project to sieve plane or HRS entrance. We know the HRS well, so we can separate HRS effect from the broken septum.
    • This will not work because the coupling of variables at target, make it becomes only 3 variables.
  • I will spend time working on Nilanga's method.
  • Dp reconstructed: dp = offset + 0.073*x_fp + 0.073*y_fp - 1.46*y_fp**2 + 0.511*theta_fp + 0.292*theta_fp*x_fp + 3.65*theta_fp**2.

07/13/2016:
  • 2D plot for all conbination of 4 variables at focal plane, which can find here.
  • From 2D plot for theta_fp vs x_fp, there is a dependent between these two variables. In dp_reconstructed (which is given above), this theta_fp vs x_fp is taken into account. So it made 4 variables become 3. But we need one more variable to tell us about y_tg?
  • The spread of x_fp (~30cm with different phi_tg) is not a big concern because in dp_res, the coefficient  theta_fp dependent is a factor of 10 bigger than that of x_fp.
  • The question still, how can we get y_tg?
  • To do:
    • Get momentum distribution and compare with simulation. Now dp_res is trusted. Can add an offset into dp_res to match with energy loss due to ice and material.
    • Plot 4 variables at focal plane for 2nd period.
    • Maybe can start to add 2 more variables at focal plane.

07/20/2016:
  • New forward matrix for both y_fp and phi_fp. Results agree pretty well, plots can find here.
    • Changes: remove dp offset in simulation when I do the optimization.
    • Use 56 data points which include: 1A, 1B, 1C, 1D and 2C, 2D, 2E.
    • 23 paramerters for y_fp, 18 parameters for phi_fp.
    • Can it be better if I use all given data points?
  • Momentum reconstruction:
    • Add in offset due to shift in ice thickness + known materials (offset=0.00405).
    • Add in x_fp^2 term to shift both -2 and +2 dp to the same direction. Result can find here.
    • The dp =-2% is not so good. Is it due to septum mismatch? Need to check with Vince. Is it due to systematic? Is it the same as the other dp =-2%?
  • y_target reconstruction:
    • y_tg
    • Vince mentioned that y_tg resolution at 1.1 GeV is 4mm. So the result make sense. The spread in y_tg for upstream and downstream foil is due to the correlation between y_tg and phi_tg become stronger for these 2 foils.
    • We can't separate the foils because there is overlap between foils.
  • 2D plot at focal plane for second period (good septum) can find here. theta_fp ~ theta_tg; y_fp ~ phi_tg.
  • To do:
    • Try to optimize forward matrix again with all given data points.
    • Reconstruct Z_target, hope can get better separation between foils.
    • Check septum setting mismatch.

07/27/2016:
  • During meeting, discuss about the complete procedure for optic study:
    • Get dp_res, y_res.
      • How to get dp_res? Each sieve hole, known: theta_tg, phi_tg, E_beam --> E' --> dp. The optimized dp =f(x_fp,y_fp,th_fp,ph_fp). It should be equal to dp calculated.
      • How to constrain y_tg? Which term to add?
    • Get forward matrix as a function of (dp,y,theta,phi)_tg.
  • Meeting with Nilanga:
    • Suppose y_tg is strongly depend on ph_fp and y_fp, 2D plot. Nilanga suggested to make a modification to y_tg to make each foil seat at the right place, by scaling y_tg. y_tg look better, but the resolution is worse. --> Need more work on this.
    • Get an universal cut which can be applied for all cases (multifoil, multi-dp). The cut will take the y_tg, dp acceptance into account. From 5 edge to 3 edge, I need to make a spline interpolation to make the cut continous. But the function depend on the y_tg range.

08/03/2016:
  • Reconstructed momentum:
    • The y_fp^2 term seem bend dp too much, that's why dp width is broaden. Vince suggested that it maybe due to the constrain on y_tg (we only have 3 foils), when we go up to 2nd order in y_fp make the dp diverge.
    • Without y_fp^2 term, dp look much better: width is smaller and the tail on high energy side is better, plot is here. Is this too good to believe?
  • To do:
    • Check dp spectra for all momentum setting and all foils target.
    • Optimize y_tg further to see the improvement.

10/12/2016:
  • Reconstruct y_tg does not work. The resolution is so bad and I can not do better than that. So Nilanga suggested to make 4 forward matrices from target for focal plane.
  • Requirement for these matrices
    • The good ratio between number of data points and number of parameters. Over-constaint problem, oscillation between data points.
    • Try to minimize higher order terms: the 3 cross terms.
    • A reason to use a term?
    • When do optimization, each data should has a separate uncertainty. So chi square can make sense. Because hole size change with each setting.
  • Several matrices were obtained:
    • 62 data points and 60 parameters, plots and the coefficients can be found here.
    • 74 data points and 52 parameters, here.
    • 88 data points and 49 parameters, here.
    • 88 data points and 37 parameters, here.
    • 88 data points and 20 parameters (only offset and first order), here.
  • One problem with 4 forward matrix is how to get 4D cut at focal plane which does not depend on any target variable? The cut should be a function of focal plane variables only.
    • The solution is: create a look up table from simulation. In simulation we do (more detail can be found here):
      • Simulate whole phase space: (180 mrad and 100 mrad). 22 cm long target. No elastic, no radiative correction.
      • Cut at sieve plane to choose only 1CDE & 2CDE.
      • Cut momentum from -3.5 to 4.5%.
      • Cut y_tg: -18 to 18 cm.
      • Create a 4D array to store each 4D-bin, each bin has (xfoc, yfoc, phfoc, thfoc) and a content which is there is an event.
    • From experimental data, look at each bin and compare that bin with that one in simulation. If there is an event in the "look-up" table, then accept this experimental event.
      • Apply this method to simulation and experimental data with multifoil and sieve slit in, here.
      • Problem: why 11F is not cut out?
  • For 88 data and 37 parameters set, chi square from optimization for y_focal is too big, it is either uncertainty for each data point is underestimate or the fitting is bad. Suggestion: add 1 more parameter for y_focal and reduce 1 parameter from theta_focal (because chi quare for theta is too small).
  • Jianping suggested to get the functional form for each theta separately, then all theta-dependent parameters will go away. Because we only have 2 rows, which is 2 values of theta_target, and there are so many theta-dependent-parameter.
  • To do:
    • Add 1 parameter for y_focal and reduce 1 parameter for theta_focal.
    • Try with 1 value of theta functional form.

10/26/2016:
  • Add in data points from 3rd row: there are total 107 data points.
    • Remove 10 bad data points: 1863 (21D, 32D), 1866 (32A), 1869 (11A, 23E, 32A), 1809 (22F, 23E, 23F), 1816 (23E).
    • Reason for removing:
      • holes that are partly seen (small statistic, hard to know exact center hole location).
      • Points that are bending too much, strong bending effect from septum. And not stay in interested area.
  • There are 97 data points and 39 parameters at this moment.
    • With add in 3rd row, theta^2 term can be added in.
    • Issue I had before is that y_fp reduce chi square is bad. With theta^2 term add in and more data point, reduce chi square for all variables now are reasonable. Sieve plot with new matrix can be find here.
  • Use above matrix to create a look up table. There are 2 cases:
    1. Look up table for sieve slit in, no target collimator.
    2. Look up table for production run, where target collimator is in.
  • Get cross section using this look up table:
    • Simulation:
      • Simulate elastic carbon run, 1899 ( dp =0%), center foil only. Cut at x_sieve and y_sieve. Save into root file. This give me number of event (called tacc) that initially stay within cut at target.
      • Use look up table to cut event in root file. For event survive inside 4D cut (look up table cut), sum all cross sections (called xsc).
      • To get cross section, cross section = xsc/tacc.
    • Data:
      • Apply 4D cut to carbon run with PID included. Get number of event survive, called ncarbon.
      • Apply same cut to no target run. Get number of event survive, called nhe.
      • Get cross section, cross section = (ncarbon*scale_for_carbon - nhe*scale_for_he), density, Solid angle.

12/28/2016
  • Have been trying to figure out 40% difference between simulation and data when using look-up table method.
  • I tried:
    • Change transport functions --> not help (here).
    • Get cross section with different binning: issue is that both simulation and data cross section decrease as number of binning increase, detail here and here. --> Solution for simulation is that statistic for rootfile to create the table is not enough. After increase number of simulated event, cross section from simulation is stable, cross section results can be found here.
    • Remain issue is cross section from data. Seem the acceptance cut a lot of event from data???