The cause of this issue might be that the likelihood values grow quite large and thus minuit does not converge properly as the values leave the [0, 1]-range by several orders of magnitude.
@maxnoe suggested to divide the likelihood used by the dof to lower the value. The likelihood is constructed to asymptotically resemble a chi2 so this changes the approach to chi2 per dof which does not change the minimum.
This does, ofc., not make the test more robust in any way.
See #2415
The cause of this issue might be that the likelihood values grow quite large and thus minuit does not converge properly as the values leave the [0, 1]-range by several orders of magnitude.
@maxnoe suggested to divide the likelihood used by the dof to lower the value. The likelihood is constructed to asymptotically resemble a chi2 so this changes the approach to chi2 per dof which does not change the minimum.
This does, ofc., not make the test more robust in any way.