How to construct Poisson p.d.f. with shifted argument

Suppose I have a histogram which consist of Gaussian backround and Poisson signal. I want to construct composite model Signal+Background p.d.f. and then fit it to a given data. The problem is my Poisson signal is shifted along x-axis. So I want to add shift variable to RooPoisson as a parameter. I know that there are many ways to do this. I think that it is a common problem with p.d.f.'s that have no shift in their nature (unlike those whicch already has it e.g. gaus, Cauchy and etc. ) although this seems to be simple. So I want to know what is the simplest way to add x-shift as a parameter to an existing p.d.f. in RooFit?

What do I mean by shifted argument is just Pois( (x - shift), lambda ).

To complete the picture I provide my code below.

void ENoise( )
{
    //Get data hist
    TFile* f = TFile::Open( "ENoise/tree_noise.root" );
    TTree* tree = (TTree*)f->Get( "tree_noise" );
    Double_t xMin = -150.;
    Double_t xMax = 300;
    TH1* hist = new TH1F( "hist", "Histogram from noise tree", 100, xMin, xMax );
    tree->Project( "hist", "integral", "mode == \"E1LSB\"" );
    //Create a variable of interest
    RooRealVar x( "x", "x", xMin, xMax );
    //Build data from hist
    RooDataHist data( "data", "Noise integral (1LSB)", x, hist );
    //Construct gaussian bkg
    RooRealVar mean( "mean", "Pedestal Mean", 0, -100, 100 );
    RooRealVar sigma( "sigma", "Pedestal Sigma", 30, 0, 100 );
    RooGaussian bkg( "bkg", "Backgroud", x, mean, sigma );
    //Construct Poisson signal. Here's the problem!!!
    RooRealVar lambda( "lambda", "Signal Mean", 1, 0, 10 );
    RooPoisson sig( "sig", "Signal", x, lambda );
    //signal fraction
    RooRealVar fSig( "fsig", "Signal fraction", 0.5, 0, 1 );
    //build composite model
    RooAddPdf model( "model", "model", RooArgList( sig, bkg), fSig );
    //Fit model to data in specific range
    model.fitTo( data, Range( xMin, 200 ) );
    //Draw the model and components
    TCanvas* c = new TCanvas();

    RooPlot* xFrame = x.frame();
    data.plotOn( xFrame );
    model.plotOn( xFrame );
    model.plotOn( xFrame, Components("bkg"), LineStyle( kDashed ) );
    model.plotOn( xFrame, Components("sig"), LineStyle( kDashed ) );
    xFrame->Draw();

    c->SetLogy();
}


WHAT HAVE I TRIED

I tried to use reparametrization trick in order to shift variable x:

    RooRealVar shift( "shift", "shift", 100. );
    RooFormulaVar xShift( "xShift", "x - shift", RooArgList( x, shift) );
    RooRealVar lambda( "lambda", "Signal Mean", 10, 0, 100 );
    RooPoisson pois( "pois", "Poisson", xShift, lambda );

But during the minimization it is just freezing terminal. Doesn’t work.

Also I tried RooClassFactory::makePdf in order to create my own p.d.f.
In the ROOT session

root [0] RooClassFactory::makePdf("ShiftedPoisson", "x,lambda,x0");

NOTE: there are no spaces between parameter names.
Then it is necessary to implement evaluate method in ShiftedPoisson.cxx file. In my case it is as follows:

 Double_t ShiftedPoisson::evaluate() const 
 { 
   // ENTER EXPRESSION IN TERMS OF VARIABLE ARGUMENTS HERE 
   return TMath::Poisson( (x - x0), lambda ); 
 } 

Then I just include ShiftedPoisson.cxx in my program and do

    //x definded above
    RooRealVar lambda( "lambda", "Signal Mean", 90, 1, 800 );
    RooRealVar x0( "x0", "Shift in signal", 25, -600, 40 );
    ShiftedPoisson sig( "sig", "Signal", x, lambda, x0 );

And it works pretty well.


Thank you in advance

Hi,

I guess the problem is due to the fact that the Poisson is a discrete distribution and there is problematic if you want to shift the observables.
I am not sure why you are using a Poisson in this case, for representing a signal. I think a Gamma distribution will be more appropriate.

Lorenzo

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