Background subtraction systematics

jtenavidal commented 11 months ago

What is the impact of the background subtraction mehtod?

To quantify it, we must be able to configure the background multiplicity from the command line

jtenavidal commented 11 months ago

DONE

jtenavidal commented 11 months ago

Next steps

[x] Add plotting more than one dataset at once
[ ] Compute difference between data due to multiplicity change
[x] Add back in plotting
[x] Add in run scripts . For now do +1 and -1 in multiplicity

jtenavidal commented 9 months ago

Fixed Background test. This will be used for systematics. Systematics due to multiiplicity choice are quantified already.

jtenavidal commented 9 months ago

I added a new script that stores the bkg plots in a nice format. Difference is calculated

Example (few events used for calculation) 1p0pi

jtenavidal commented 9 months ago

When increasing the statistics, the error shows up to be around 20%. It seems too big

jtenavidal commented 9 months ago

Repeating test without acceptance weights: 1p0pi case:

It clearly describes it without acceptance!!!

1p1pim

jtenavidal commented 9 months ago

Applying acceptance correction after background subtraction, 1p0pi case (1GeV Carbon)

This fixes the issue!

jtenavidal commented 9 months ago

zoomed in

jtenavidal commented 9 months ago

Acceptance after fiducial before background

jtenavidal commented 9 months ago

I HAVE A FEELING it is the angle shift in the acceptance? somehting else is odd not the method itself makes sense as it is now

jtenavidal commented 9 months ago

https://github.com/e4nu/e4nuanalysiscode/blob/b956d1c2e92e23cd73a4e7a8868dbbb21d1c3396/src/analysis/AnalysisI.cxx#L55

This creates many issues. I need a way to quickly double check. This will also help for the presentation and so on.

Direct comparison with previous analysis. Even store important histograms of previous analysis in git so we can easily compare to it. Validation tool
We need to be able to define this flip in one single place, tired of breaking the code

I will open a separate issue for this

jtenavidal commented 9 months ago

No acceptance

1p0pi 1GeV

Contribution from multiplicity 2 bkg

Contribution from multiplicity 3 bkg

Contribution from multiplicity 4 bkg

1p0pi 2GeV, max multiplicity 4

1p0pi 2GeV, max mult 4, mult 3 contribution

1p0pi 2GeV, max mult 4, mult 2 contribution

1p0pi 2GeV, max multiplicity 2

jtenavidal commented 9 months ago

1p0pi 2GeV, max multiplicity 2 The agreement improves significantly. It seems there's an issue with higher order multiplicity events

jtenavidal commented 9 months ago

1p0pi 2GeV, 5000000 events, 2 max substraction.

Max mult 3

Max mult 4

jtenavidal commented 9 months ago

TO DO We need a plot showing the method works. For it I need True Reco : true to compare against No Bkg subtracted: how bad we would do Bkg subtracted: min mult +1 , min mult +2 and min mult +3

Total: 5 ROOT files Observable

1p0pi 1GeV

1p0pi 2GeV

jtenavidal commented 9 months ago

Also need to re-think logic of the background histograms. Are we really storing correctly the plots? If we store it before substraction we have

mult 4 events
mult 3 events
mult 2 events We look into contribution to signal
mult 4 -> mult 1 (aren't we missing mult 4 into 2 into 1?)
mult 3 -> mult 1 (aren't we missing mult 3 into 2 into 1?)
mult 2 -> mult 1 (this always contributes the same way) Because before subtraction they are only existing in one place Need to store it after background

Also, https://github.com/e4nu/e4nuanalysiscode/blob/7c01d836224d8ebeeedac2e046533ac6db6b72fc/src/physics/Event.cxx#L72

We assume linearity https://github.com/e4nu/e4nuanalysiscode/blob/7c01d836224d8ebeeedac2e046533ac6db6b72fc/src/analysis/AnalysisI.cxx#L235

jtenavidal commented 9 months ago

1p0pi on C, 2GeV