Simultaneous fits with conditional pdfs

[Rishabh_Mehta]

Dear experts,

I have a resolution function pdf which models deltat residuals and is conditional on deltaterr. I do not multiply this with the deltaterr distribution to make it full, rather I simply fit this pdf using the ConditionalObservables option in fitTo.

When I simply fit this, the fit converges without fail and the results are good. I have two datasets, and the fits converge individually on both of them.

Now I need to do a simultaneous fit on both datasets sharing some parameters between the individual pdfs. I use RooSimWSTool to split the base pdf into two and create a combined pdf. I create a simultaneous dataset using the Indexing procedure. Now when I do the simultaneous fit, multiple things go wrong.

Firstly, the fit quality on both the individual datasets is pretty bad, even when I don’t have any shared parameters. Secondly, the roofitresult shows deltaterr as one of the parameters, while it should have been a conditional variable. The individual separate fits did not have this issue. I believe the first issue is related to second, roofit treats deltaterr as a floating parameter instead of a conditional observable.

My question is, what can be done to mitigate this? I can in principle create a full pdf by multiplying the conditional pdf with a histogram pdf of deltaterr, but it will bring some other complications in the analysis.

[Danilo]

Hello,

I am adding in the loop @jonas .

Best,
D

[Rishabh_Mehta]

Thanks @Danilo .

For now, I have arrived at something which tells me why doing this might lead to bad fit result. The deltaterr distribution is fundamentally different in both the datasets. Maybe roofit treats simultaneous fits like a composite pdf (signal, bkg) (not sure why this would be) and this leads to biases due to Punzi effect. Not exactly sure if this is the case, but atleast this tells me I will have to revert to a full pdf at some point in later stage.