TH::Sumw2()

ROOT
1

Hi,

I observe that when I fill into a histogram’s underflow or overflow bin, the number of entries is updated as expected but the sum of weights is not:

root [0] TH1D h("h", "h", 10, -0.5, 9.5)
root [1] h.Sumw2()
root [2] h.GetEntries()
(const Double_t)0.00000000000000000e+00
root [3] h.GetSumOfWeights()
(const Double_t)0.00000000000000000e+00
root [4] h.Fill(15, 2)
(Int_t)(-1)
root [5] h.GetEntries()
(const Double_t)1.00000000000000000e+00
root [6] h.GetSumOfWeights()
(const Double_t)0.00000000000000000e+00

To my understanding, I believe this is incorrect behavior because the normalization will be wrong.

In general, I believe each event should be normalized by w/(sum of w) where w = event weight. The sum of weights should include weights from all events filled regardless of whether it went into under/overflow. This can be seen in the extreme case where e.g. events only enter the overflow, where otherwise the supposed normalization factor has a 0 in the denominator i.e. is infinite.

Am I missing something obvious?

Thanks,
Eric

2

Hi,
the histogram normalization is computed normally using only the entries in the histogram range (i.e. excluding underflow and overflow).
TH1::GetSumOfWeights does not consider the underflow and overflows.
This is documented in root.cern.ch/root/htmldoc/TH1.ht … mOfWeights

Best Regards
Lorenzo

3

[quote=“moneta”]Hi,
the histogram normalization is computed normally using only the entries in the histogram range (i.e. excluding underflow and overflow).
TH1::GetSumOfWeights does not consider the underflow and overflows.
This is documented in root.cern.ch/root/htmldoc/TH1.ht … mOfWeights

Best Regards
Lorenzo[/quote]

Hi Lorenzo,

Thanks, I see that now in the documentation. But why is that the case? I believe the argument I gave about that the sum of weights should include underflow and overflow bins was correct.

Eric

4

Hi,

your argument does not make much sense to me, because if the are are no events in the histograms, there is nothing to normalize.
I remind you that also all the histogram statistics, sum of weights, sum of weights square, mean , standard deviation, etc… it is computed using only the events in the histogram range.
See root.cern.ch/root/htmldoc/TH1.html#TH1:GetStats

Lorenzo