I have troubles when adding a Gaussian constraint to my signal sample when using HistFactory.
Assuming that from other experiment’s uncertainty, the signal entries satisfies a Gaussian distribution.
And in my model, I want to adding this constraint to my signal model.
I have tried to use AddOverallSys, like:
Thanks a lot for your suggestion.
I find the AddHistoSys. It seems that I need to input two histograms to limit on my signal.
I am still confused about how to define the Gaussian which will be applied to my signal.
Do I need to input a histogram with bin content 1.1 and another histogram with bin content 0.9 ,if I want a Gaussian with the mean 1 and sigma 0.1 ?
The HIstFactory will interpolate the histogram values from the nominal (e.g. 1.0) to the lower (0.9) and upper value (1.1) that represent the 1 sigma deviation.
Depending on the type of interpolation used, you will have effectively a different type of constraint. If the interpolation is linear the constraint will be gaussian, but if the interpolation is exponential, the effective constraint applied will be log-normal.
See paragraph 4.1 of the HIstFactory Users guide (https://cds.cern.ch/record/1456844/files/CERN-OPEN-2012-016.pdf) for the details of the interpolation.
Sorry to trouble you again. I think I didn’t explain my question clearly.
I want a Gaussian constraint to act on my signal sample, and in this way the model should be changed from μS+B to μS*G+B, where G is Gaussian distribution and μ is amplitude.
I tried the function AddHistoSys, like
The alpha_SigErrConstraint is the defined variable. However it works on channel1_model which contains both signal and background. That is not what i hope for.
Is there any way to solve my problem?
I am new about hisfactory. Thanks a lot for suggestions.
I see now your problem. The constraints function will be multiplied to the overall model (signal plus background), but it will affect only the signal, because the background will not depend on the alpha_SigErr parameter.
The same will be if you use AddOverSys. The difference is that in the first case you will have an overall variation for the signal, in the second it will variate following the upper/lower histogram, i.e. variations that are different bin by bin