PAH features like 3.3um are fairly compact, e.g. the main 3.3um PAH has an effective resolution of R=77. What this means is:
The fitted FWHM of these features was in fact resolution-dependent (e.g. from IRS low-res, Akari), and will likely need to be tuned for higher resolution of e.g. NIRSPEC/MIRI. Features will get sharper.
Our current assumption that all dust/attenuation features in a science pack are resolution-independent may be incorrect (more so for the most compact feature: 3.3, 11.3, 16.4, etc.).
We should test this on e.g. Akari vs. NIRSPEC when available (@ThomasSYLai ?).
Possible Solutions:
Ensure all science packs reference features are expressed/calibrated at their high resolution limit (R>1000 is probably sufficient).
Add a "coarsening" step of dilating the FWHM of all dust features in a science pack by gaussian-summing the local line FWHM in (and test this!).
This is a bit awkward for features which span segments of different resolution, e.g. Akari + MIRI. If this is implemented, we may need to make these power-tied "ghost" features, similar to how we treat lines.
A more robust solution would be to build the model at "infinite" resolution and directly convolve with the (wavelength+segment-varying) line spread function but this is 1) inefficient, and 2) ill-defined for stitched spectra. So hopefully a simpler coarsening step would be sufficient.
Features that may need this "coarsening" treatment:
Also: this means you'd better not stitch spectra with widely differing resolutions, since we can't really allow PAH features widths to vary like lines can, since they cover much more territory.
PAH features like 3.3um are fairly compact, e.g. the main 3.3um PAH has an effective resolution of R=77. What this means is:
We should test this on e.g. Akari vs. NIRSPEC when available (@ThomasSYLai ?).
Possible Solutions:
This is a bit awkward for features which span segments of different resolution, e.g. Akari + MIRI. If this is implemented, we may need to make these power-tied "ghost" features, similar to how we treat lines.
A more robust solution would be to build the model at "infinite" resolution and directly convolve with the (wavelength+segment-varying) line spread function but this is 1) inefficient, and 2) ill-defined for stitched spectra. So hopefully a simpler coarsening step would be sufficient.
Features that may need this "coarsening" treatment: