SetParLimits Ignored in Fit

ROOT
1

Hi, I am having trouble fitting a simple TH1D with a Gaussian function. It appears that my SetParLimits calls are not being obeyed. Here is a fragment of my PyROOT code that does this:

[code]
# h is a TH2D, h_binedges is a python list of pairs of floats
projs, fit_gaus = {}, {}
for i in range(len(h_binedges)-1):
projs[i] = h.ProjectionY("_%d" % i,h_binedges[i],h_binedges[i+1]-1)
hs.Add(projs[i])
fg_min,fg_max = projs[i].GetXaxis().GetXmin(),projs[i].GetXaxis().GetXmax()
fg_height = projs[i].GetMaximum()
fg_centre = projs[i].GetMean()
fg_width = projs[i].GetRMS()

    fit_gaus[i] = ROOT.TF1("fit_gaus_%d" % i,"gaus",fg_min,fg_max)
    fit_gaus[i].SetParameters(fg_height, fg_centre, fg_width/3.5)

    fit_gaus[i].SetParLimits(0,0.5*fg_height,2*fg_height)
    fit_gaus[i].SetParLimits(1,fg_min,fg_max)
    fit_gaus[i].SetParLimits(2,1e-3,fg_width*2)
    fit_gaus[i].SetLineWidth(1)
    fit_gaus[i].SetLineColor(ROOT.kRed)
    ftr_p = projs[i].Fit(fit_gaus[i],"SRQ+")
    ftr = ftr_p.Get()
    if int(ftr_p) != 0 or not (0.5*fg_height < ftr.Parameter(0) < 2*fg_height) or not (1e-3 < ftr.Parameter(2) < 2*fg_width) or abs(ftr.Parameter(1) - fg_centre)/ftr.Parameter(2) > 5 :
        refit_func = fit_gaus[i].Clone("refit_%d" % i)
        refit_func.SetLineColor(ROOT.kBlue)
        ftr_p = projs[i].Fit(refit_func,"SRW+")
        ftr = ftr_p.Get()
        fit_gaus[i] = refit_func
    print i, ftr.Parameter(0), ftr.Parameter(1), ftr.Parameter(2)[/code]

After the fit, the printed values of the parameters are often outside of the limits that I set. This happens even on iterations where the “refit” clause was not executed, so I don’t think it’s related to that.

So what could cause my SetParLimits to be ignored here? Any help would be appreciated.

Jean-François

2

Try with the “B” fit option.

3

Yes, you should use the “B” option, since you are using a pre-defined function (“gaus”). IN that case the fit internally re-sets the initial parameter values and limits to optimal ones.
The option “B” is used to avoid this and use the values provided by the user

Best Regards

Lorenzo

4

Ah, thanks for pointing that out. I did see the “B” option, but I think I was mis-led by the wording of the description:

[quote]User defined parameter settings are used for predefined functions like “gaus”, “expo”, “poln”, “landau”.
Use this option when you want to fix one or more parameters for these functions.[/quote]

Can we change the second sentence to be less specifically about “fixing” parameters? Something like:

Jean-François

5

See the TF1::SetParLimits method description if you like misleading information. It simply claims that you must use “B” whenever you want to “set limits” (“predefined functions” are not even mentioned): [quote]The specified limits will be used in a fit operation when the option “B” is specified (Bounds).[/quote]