Home | News | Documentation | Download

Is variable constraint supported?


Dear RooFit users and experts,

I am scanning an invariant mass spectrum using a background+signal model. I would like to introduce a nuisance parameter modelling the amount of spurious signal events induced by an imperfect background model. Assuming the spurious signal is equally present across the entire spectrum, I could construct my model as follows:

RooWorkspace* wksp = new RooWorkspace("wksp")
//The background model
//The signal model
//The gaussian constraint on spurSigSize
//The constrained bkg+sig model
wksp->factory("PROD::bkgPlusSigModel(SUM::freeModel(bkgN[500.,0.,1000.]*bkgModel, sum::sigSize(sigN[0.,-1000,1000],spurSigSize)*sigModel),ssConstraint)");

As I understand it, the factory interface will interpret the two numbers 0,10 in the Gaussian constructor as RooConst objects, and this indeed works for me. Now, my question is: what if the constraint on spurSigSize is tighter at certain signal pole masses? Then the width of the constraint gaussian should be variable:


So I could change the constraint at a whim using wksp->var("spurSigSpread")->setVal(15).
But will RooFit still interpret the last argument as a RooConst? If not, is there some other way to implement a variable constraint?

The only alternative I see is to create a new constraint pdf (and hence a new background+signal model) for each pole mass, but this seems rather tedious, especially when scanning a large range of signal hypotheses.


Well this was a big misunderstanding! :sweat_smile: For the books, I thought RooConst referred to a Roo-Constraint , not a Roo-Constant. Since I had never used this class before, I inferred its meaning from the RooFit tutorial on constraints. Anyways, I hope the lines above are helpful for someone. Of course one can use RooRealVar arguments to construct a variable constraint, but a “variable RooConst” is a contradiction in terms.


Just a comment. You can define a variable as a RooConst or called RooRealVar->setConstant(true).
For the fitting the two things are equivalent.



This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.