binning / roiwidth vs. selection / radius

[HealthyPear]

What is the relation between these two configuration options?

Does the second overwrite the first?

Does the first need to be <= than the second?

I looked for this because I stumbled upon a paper that says "[...] adopted [...] a circular ROI of 15◦ radius.", but all I can find is roiwidth which seems like the side length of a square.

[omodei]

Was the paper using fermipy to unbinned likelihood analysis, which is defined over circular ROI instead of square.

[HealthyPear]

Thanks for replying!

They do not mention if they did a binned or unbinned likelyhood analysis unfortunately...

That paragraph reads,

We also analyzed the LAT data covering the 100 MeV to 2 TeV energy range to generate the mission-averaged, broadband γ-ray spectrum and monthly binned light curve. We adopted SOURCE class events and a circular ROI of 15◦ radius. All 4FGL-DR4 sources within 20◦ from S5 1027+74 were considered during the likelihood fitting.

Given they to a monthly light curve the number of events in each time bin should not be expected to be that small I think...also they produce the mission averaged spectrum, so it is not strictly a time-series analysis.

[MiltosMichailidis]

Thanks for reaching out.

To summarize briefly, there is some confusion in the literature regarding Fermi-LAT analyses. In most papers, what is reported is the ROI width, which refers to a square region on the sky. However, in practice, every standard Fermi-LAT analysis must define two distinct parameters, and the relationship between them is crucial for obtaining reliable results without caveats (as explained below). There are exceptions - such as weighted unbinned photon analyses - but those follow a different logic. Here, I will focus on the standard cases.

The selection radius defines a circular region on the sky and, in the majority of analyses, must be larger than the ROI width. This parameter controls which photons are loaded into the analysis. It is a circular cut applied before any binning or likelihood evaluation, meaning that only photons within this radius are considered at all.

The ROI width, on the other hand, defines the actual analysis region. It is a square area within which the likelihood is evaluated and the fitting is performed.

The reason the selection radius must be larger than the ROI width is that photons originating from outside the square ROI can still contribute inside it. In particular, bright sources just outside the ROI and diffuse emission from surrounding regions can spill into the analysis region due to the LAT’s relatively broad point-spread function, especially at low energies. If those photons are not loaded, the model cannot properly account for their contribution, which can lead to distorted fits, edge artifacts, and biased background normalizations.

In short, the selection radius is intentionally chosen to be larger so that the likelihood fit inside the square ROI is not artificially constrained by missing photons, allowing the model to correctly account for PSF spill-in and diffuse emission from outside the analysis region.

How much larger the selection radius should be compared to the ROI width depends on the energy range of interest, as this is driven by the energy dependence of the LAT point-spread function.

And there is also a third relevant parameter that defines how far from the ROI center sources are included, but this is a different story.