Fitting To Regions in RooFit

[sweta]

Hi all,
I am having some problems doing unbinned ML Fit in regions. I have data in range [70,120] of which [86,96] is my blind region (“blind”) and both side of it are the sidebands ("unblindReg_1 , unblindReg_2). The integral of data in [70,120] is 2637 ( blind: 614 and sidebands:2023). I am using a power law as my model(RooGenericPdf). I need to fit my model in the side bands → extract the parameter value (p) → Find normalisation in blind region from the model with parameter value p.

1.First I fit the model to data sideband regions without any extended normalisation going into the fit. The fit looks fine.
2. Take a model with parameter value same as obtained from step1 and extend it with a norm parameter → fitTo data in blind region → obtain the normalisation. Here the normalisation is coming out to be 3115 , even greater than integral of data in entire range!

I have attached the plots and the code snippet. It would be really helpful to know what i am missing here or is there a better way to do what i am trying to do.

###Step1: Fiting in sidebands without an extended norm.
p = r.RooRealVar(“p”,“p”,-1,-5,0)
data_bkg = r.RooGenericPdf(“data_bkg”,“data_bkg”,“Hm^p”,r.RooArgList(Hm,p))
fitresult = data_bkg.fitTo(data_obs,r.RooFit.Range(“unblindReg_1,unblindReg_2”),r.RooFit.Save())

###Step2:Fiting in blind region with parameter from side band fits with an extended norm.
p = r.RooRealVar(“p”,“p”,-3.1562)
data_bkg2 = r.RooGenericPdf(“data_bkg2”,“data_bkg2”,“Hm^p”,r.RooArgList(Hm,p))
blind_norm = r.RooRealVar(“blind_norm”,“blind_norm”,500,0,3500)
data_bkg2_=r.RooAddPdf(“data_bkg2_”,“data_bkg2_”,r.RooArgList(data_bkg2),r.RooArgList(blind_norm))
fitresult = data_bkg2_.fitTo(data_obs,r.RooFit.Range(“blind”),r.RooFit.Save())

Thanks,
Sweta

[jonas]

Hi @sweta,

If you create a RooAddPdf, it’s coefficients/yields are still interpreted as applying to the full range, even if you only fit a subrange.

In the documentation to RooAbsPdf::fitTo(), you will see an option called SumCoefRange(). If you set this to your “blind” range, you should get the expected result.

Hope that works!
Jonas

[sweta]

Hi @jonas
Thanks. It worked for the blind range. I further tried using this on the sidebands by extending the sideband model of step1

##Fiting in sidebands with an extended norm.
p= r.RooRealVar(“p”,“p”,-1,-5,0)
data_bkg1 = r.RooGenericPdf(“data_bkg1”,“data_bkg1”,“Hm^p”,r.RooArgList(Hm,p))
side_norm = r.RooRealVar(“side_norm”,“side_norm”,500,0,3000)
data_bkg1_=r.RooAddPdf(“data_bkg1_”,“data_bkg1_”,r.RooArgList(data_bkg1),r.RooArgList(side_norm))
fitresult = data_bkg1_.fitTo(data_obs,r.RooFit.Range(“unblindReg_1,unblindReg_2”),r.RooFit.SumCoefRange(“unblindReg_1,unblindReg_2”),r.RooFit.Save())

Here again the normalisation coefficient comes greater than expected if i dont put the ‘SumCoefRange(…)’ in the fitTo and it comes out to be smaller than expected when i include ‘r.RooFit.SumCoefRange(“unblindReg_1,unblindReg_2”)’ in the fit.
Is it that the 'SumCoefRange() works for a single range? If so how to extract normalisation coeff from fit in sidebands or in general multiple ranges?

Thanks again,
Sweta