LT Separation Meeting - 26/08/22
Stephen Kay
Garth Huber
Vijay Kumar
Ali Usman
Muhammad Junaid
Garth has a set of slides - https://redmine.jlab.org/attachments/download/1593/LT_sep_iterations.pdf
See also Bill's slides - https://redmine.jlab.org/attachments/download/1594/Kaon_LT_tutorial_18nov28.pdf
Start from Bill's code - https://github.com/billlee77/omega_analysis
- Prior to LT separation, need *final* normalised yields
- Most of the work involved is actually in getting to this point!
- If things are converging, iteration steps should be quick
- This includes diamond cut already
- One could analyse the variations across only the larger, high epsilon diamond
- All efficincies included
- LT
- FADC DT
- Cryotarget
- Yield corrections
- ALL tested for reliablity over a wide range and applied
- All kinematic offsets determined and finalised
- Need one thing kept constant between iterations
- The normalised yields and distributions
- Cross section varies across experimental acceptance
- Need to choose a functional form that will reasonably account for this
- Don't know this in advance
- Make a choice, start iteration process with it
- Can check previous analyses for guidance on forms to try
- Can be different for each diamond plot
- So long as it it well understood for each individual diamond
- On diamond plot, t bins slice across diagonally
- W and Q2 different for each t bin
- Results quote different Q2/W value for each t bin
- t min varies across the diamond
- t min varies even within a t bin
- In SIMC, replace physics_pion.f with physics_iterate.f
- Need to make some good guesses of intial paramter values for your first iteration
- Need some Q2 dependence, t dependence etc
- Will likely have to go back and change these functions a lot
- Each Q2/W should be done separately
- Can't expect the procedure to work globally
- *Keep notes organised!*
- *Keep all output!*
- Store each iteration in its own directory
- Step 1 - SIMC distributions
- Run SIMC for large #events
- Generate spectrometer and physics variables using functional form and fitpar
- Do this setting by setting for a given Q2/W diamond
- E.g. Left/Centre/Right all different
- Compare to data
- Step 2 - Combine Left/Centre/Right settings at high and low epsilon for each W/Q2/t/phi/epsilon bin for data and MC
- Get statistical errors for each bin
- Get yield, error in yield for data and simulation
- Calculate ratio and error in ratio for each bin
- Process enough events to minimise SIMC statistical error!
- Shouldn't be dominated by simulation statistical error!
- Step 3 - Calculate average kinematics
- Mean data values of W/Q2/Theta/epsilon for each t bin at high an low epsilon
- Need this for data and MC
- Values will differ between high and low epsilon slghtly
- Will also change slightly as the model is iterated
- Step 4a - Inspect and understand data closely
- Examine variation on phi between data and MC within each t bin
- Deviations between data/MC are usually indicated as wiggles in the ratio
- Want R ~ 1 across a broad kinematic range
- Some plots earlier on might look ok, only by examining the ratio very closely in individual bins can some differences be found
- How things vary should guide what changes in the next iteration
- E.g. Is it the interference terms that are off?
- Step 4b - Even closer examination
- Each t bin (for one epsilon), binned in theta
- Interference terms depend upon theta
- Interference terms vanish in parallel kinematics
- Explicitly chose a functional form that makes this true
- In lowest bin, should have minimal interference terms
- Wiggles get larger as theta increases
- Phi distributions for each t bin, subdivided into 8 theta bins
- A lot of plots
- More than you probably want to put in thesis
- BUT a tech note/report is encouraged
- Even here, you probably want to show only a few plots as an example
- Step 5 - Calculate unseparated cross section
- Using fitpar for the iteration, evaluate the model at *average* kinematics of the data for each t-bin
- Calculate using formula and fit parameters, calculate the cross section at the average values of W/Q2/Theta and epsilon for each bin
- Read over Blok PRC 78 (2008) 045202 carefully
- Each t/phi bin at both high and low epsilon
- The weight in SIMC is NOT the unseparated two fold cross section
- It is the five fold cross section (d5sigma)
- Fitting procedure iterated until the experimental cross section changed by less than a prescribed amount (~1%)
- Step 6 - Fit Rosenbluth Eqn
- Each t bin is fit separately
- Fit result gives L, T, LT, TT cross sections for each t bin
- Step 7 - Fit L, T, LT, TT values to get new fitpar
- Compare t bins with each other
- Fit with physics_iterate.f functions to give next iteration model parameters
- Now, we're back at the top. Feed these back into step 1
- Repeat steps 1-7 until separated cros sections are stable
- *Don't re-run SIMC*
- Recalculate weight for each event
- Any theses older than Tanja's are likely using different procedures
- Tanja's procedures were improved on earlier ones
- Some slight modifications after this
- Only Marco and Bill used Tanja's procedure
- Evaluating if fit equations are ok
- Procedure usually works ok, but for some kinematics, pi-/pi+ sigma wouldn't converge
- Compare fitpar from different Q2/W and see if they were slowly varying
- If not, could use their variation as a pointer to an alternative functional form variety to try
- Things will behave in a consistent manner if the correct solution is found
- General pointers
- Read Blok paper carefully
- Review Bill's slides
- A single Q2/W iteration should only take 1-2 hours
- Fitpar should converge in a few iterations to give 0.5