Signal Uncertainties and RECAST

Hi All,

I wanted to open this thread to spur some discussion about best practices/recommendations for including signal-related uncertainties in RECAST. Apologies if this has been considered extensively in the past and I’m just not aware of it. Beware, long post ahead!

Perhaps I’m wrong, but I think many people assume that the only barrier to RECASTing an analysis for some new signal model is generating the samples (if needed) and producing the necessary derivations. From there, it’s more or less just the push of a button.

But it’s important not to overlook the theoretical uncertainties that we typically assign to our signal processes (factorization+renormalization scales, etc.). Additionally, it’s not uncommon for an analysis to assign other signal uncertainties on stuff like the modeling of ISR jets, for example, which are not necessarily captured by the standard theoretical uncertainties. Of course the obvious issue is that these are both signal-dependent and typically evaluated outside of the standard analysis workflow.

Suppose I originally designed my search to target Model A, which features a hard ISR jet emission. I will probably use SimpleAnalysis to determine the theoretical uncertainties for this model and some “auxiliary” measurement/procedure to derive uncertainties on the modeling of the ISR emissions by the generator.

Now I want to RECAST this search for Model B, which in general can be completely unrelated to Model A. How do I define these signal uncertainties? Is the RECAST user expected to rerun SimpleAnalysis on their own for the theory uncertainties and somehow patch them into the preserved workflow? What if it doesn’t make any sense to talk about ISR emissions in Model B, where the jet(s) could instead be coming from e.g. boson decays? Or if it still makes sense in Model B, am I expected to somehow run the standalone procedure for evaluating this uncertainty myself?

In principle, one could incorporate SimpleAnalysis and any other “auxiliary” procedures into the workflow in order to calculate these uncertainties on-the-fly, but are people actually doing this?

I’m not necessarily asking for answers to all of these questions, but rather hoping to get us thinking more about this subtle topic as RECAST becomes increasingly widespread in our search programs. And if there are examples of people handling this very carefully already, that’s fantastic!


1 Like