RooFFTConvPdf: p.d.f normalization integral is zero or

Roofit and RooStats
1

Dear all,
I am trying to convolve a RooExponential with a RooHistPdf; the time constant tau of the RooExponential is
distributed according to a RooHistPdf.
I want then to generate some data from the projected convolved distribution.
I have followed the useful example rf211_paramconv.C.

When I try to run my code (attached here together with the needed data for the tau distribution),
I get the following error:

[#0] ERROR:Eval -- RooAbsReal::logEvalError(tauHPDF_fft) evaluation error, origin : RooHistPdf::tauHPDF_fft[ pdfObs=(tau_shifted_FFTBuffer2) ] message : p.d.f normalization integral is zero or negative server values: pdfObs=(tau_shifted_FFTBuffer2 = -1.485)

at this line:

RooDataHist* d = projtrapConv->generateBinned(t,1000) ;

the error gets repeated over and over.

Later on other errors appear, like:

[#0] ERROR:Eval -- RooAbsReal::logEvalError(trapConv_Int[tau]_Norm[t,tau]) evaluation error, origin : RooRealIntegral::trapConv_Int[tau]_Norm[t,tau][ Int trapConv_Norm(t,tau) d[Ana](tau) ] message : function value is NAN server values: !sumList=(), !intList=(), !anaList=(tau = -2.51149), !jacList=(), !facList=(), !func=trapConv=0, !sumCat=()

Am I missing something?
I am using the following root version:

   ------------------------------------------------------------
  | Welcome to ROOT 6.02/04                http://root.cern.ch |
  |                               (c) 1995-2014, The ROOT Team |
  | Built for macosx64                                         |
  | From tag v6-02-04, 17 January 2015                         |
  | Try '.help', '.demo', '.license', '.credits', '.quit'/'.q' |
   ------------------------------------------------------------

Any help will be greatly appreciated!

Many thanks in advance and best regards,
Marco Bomben
tests_140224-fei4-250um-fl3.8e15-extract_traps-V400-taue-in-ns.root (3.81 KB)
convolution_hist.C (3.43 KB)

2

I will tell our expert to look at it.
I am myself not an expert but looking to your output I see that you have a range with negative numbers:

mean = -2.51149
xmin = -3
xmax = -1.5

and then it seems that you are doing some kind of log evaluation:

[#0] ERROR:Eval -- RooAbsReal::logEvalError(tauHPDF_fft) evaluation error,

It is well known than negative numbers and log are not good friends… could it be the problem ? (just a suggestion … I cannot help you further).

3

Dear Olivier,
the time constant tau of the exponential is indeed negative definite as I need a monotonically decreasing function. If I understand correctly this is the way it works in RooFit if you need an exponential decay function.

Thanks,
Marco

4

Ok, Lorenzo will reply when back next week.

5

Hi,

Your model seems to be ill formed to me, for this you are getting NaN. Can you please write the mathematical expression of your model , i.w. what is the equation of your pdf ?

Thank you

Lorenzo

6

Dear Lorenzo,

in the attached picture you can see which pdf I am trying to model: it is the convolution of a RooExponential E(t;tau) with a RooHistPdf H(tau) which is based on an histogram; the histogram contains the values for tau.

Best regards,
Marco Bomben

7

Hi,

What you wrote is not exactly a convolution. A convolution is defined as

en.wikipedia.org/wiki/Convolution

I don’t think you can use the RooFFTConvPdf in your case for this integral. You should probably use the RooProdPdf and then compute the projection in tau

Best Regards

Lorenzo

8

Many thanks for the explanation. I have used RooProdPdf and it works as expected.
You can consider this issue closed.

Many thanks and best regards,
Marco Bomben