Is it a problem when I got a messages: p.d.f value is less than zero during minimization

[anikiten]

Dear experts,

I do unbinned fit of the dimuon mass in the interval 12-50 GeV, not including a signal region
25-30 GeV using polynomial function. The goal is to have background model from the sideband data.

Sometime I got a messages like:

p.d.f value is less than zero (-0.389207), forcing value to zero @ x=dimuonmass2sideband=48.3537, coefList=(a0 = 3.82232,a1 = -0.139657,a2 = 0.00124115)

p.d.f value is less than zero (-0.939635), forcing value to zero @ x=dimuonmass2sideband=48.9306, coefList=(a0 = 3.82232,a1 = -0.139657,a2 = 0.00124115)

but at the end fit converges with the messages

MIGRAD MINIMIZATION HAS CONVERGED.
MIGRAD WILL VERIFY CONVERGENCE AND ERROR MATRIX.
COVARIANCE MATRIX CALCULATED SUCCESSFULLY
FCN=-246.245 FROM MIGRAD STATUS=CONVERGED 163 CALLS 164 TOTAL
EDM=8.7068e-05 STRATEGY= 1 ERROR MATRIX ACCURATE
…

** 9 **HESSE 2000

---

COVARIANCE MATRIX CALCULATED SUCCESSFULLY
FCN=-246.245 FROM HESSE STATUS=OK 23 CALLS 187 TOTAL
EDM=9.43288e-05 STRATEGY= 1 ERROR MATRIX ACCURATE

and my printout at the end:

Number of data events = 233
-log(L) at minimum = -246.245
chi2 = 0.679778

1. RooRealVar:: a0 = 3.85915 +/- 0.455439
2. RooRealVar:: a1 = -0.139186 +/- 0.0142921
3. RooRealVar:: a2 = 0.0012522 +/- 0.000196147
4. RooRealVar:: norm = 306.049 +/- 78.7228
   norm = 306.049 +/- 78.7228

I got background fitted function which describes the sideband data well:

SR2mod13_2017UL_FTEST.pdf (21.0 KB)

My questions are:

1. should I worry then about a such messages reporting a negative pdfs during minimization ? It does not lead to the failure of the fit and minimization gives me finally a reasonable function positively defined in my mass range.

2. what RooFit is doing once at some step it is getting to a negative pdfs ?

Thanks, Sasha.

[devajith]

Hi @anikiten,

Thanks for the question. Let me add @jonas in the loop.