Unexpected behaviour of KSTest with toys ("X" option) for identical histograms

[TomasDado]

Check duplicate issues.

• Checked for duplicates

Description

When passing some identical histograms to KSTest of TH1, the code return probability 0 in case of the "X" option (runing toys) while the probablity of the default KS test is 1 (as expected).

Reproducer

running the following macro reproduces the issue:

void kstest() {

  TFile* f = TFile::Open("kstest.root", "READ");

  TH1D* h1 = f->Get<TH1D>("SR2_2L_el/tt_Conv/nominal/SR2_2L_el_tt_Conv");
  TH1D* h2 = f->Get<TH1D>("SR2_2L_el/tt_Conv/FSRAlphaS_ttbar/SR2_2L_el_tt_Conv_FSRAlphaS_ttbar_Shape_Down");

  std::cout << "integral h1: " << h1->Integral() << ", h2: " << h2->Integral() << "\n";

  const double probX = h1->KolmogorovTest(h2, "X");
  const double prob  = h1->KolmogorovTest(h2);
  std::cout << "KS probability using toys is: " << probX << ", without toys it is: " << prob << "\n";
}

with the input file available in this public afs repository: /afs/cern.ch/user/t/tdado/public/kstest.root

I could not reproduce it with a dummy histograms, so it must be somehow related to the actual values in the bins?

ROOT version

ROOT Version: 6.28/04
Built for linuxx8664gcc on Jul 10 2023, 18:22:00
From v6-28-04@v6-28-04

Installation method

source from LCG

Operating system

CentOS7 (lxplus)

Additional context

No response

[lmoneta]

The problem is that one of the histogram is having all bin errors equal to zero. Is this intentional ?
If not please add the correct errors and the test will work correctly.
In that case the toy option (X) does not work and it returns a zero probability. An error should be probably produced, but this could be probably supported. I can provide a fix for this.
Thank you for the report

[TomasDado]

Hi, thanks for the reply, indeed, in this case the histograms do have 0 uncertainties. I was just surprised by the 0 probability. I think an error message would help, thanks!