Arbitrary pixelation schemes?

[SharperJBCA]

Is it possible to setup MADAM to use pixelation schemes other than HEALPix? E.g., some arbitrary WCS pixelation scheme on a cartesian grid?

[keskitalo]

With certain limitations, yes. Madam outputs are Healpix FITS maps and it would require a nontrivial amount of work to change that. However, your calling code could easily embed your pixel numbers into any Healpix map with enough pixels and re-interpret them upon reading the maps. If you are using Madam only for destriping, you could also project the destriped timelines from your code.

There are two important issues about embedding other pixelizations into the nested Healpix pixelizations Madam expects:

• you must destripe at the same resolution as the maps are made: nside_cross = nside_map. Otherwhise Madam will translate pixel numbers between resolutions assuming Healpix relations.
• consecutive pixels numbers should be near each other on the sky. Otherwise the submaps used internally in Madam will be fragmented and can cause poor performance

[SharperJBCA]

Ah okay I see, that sounds like a reasonable work around and thanks for the quick response.

On your second point, does that mean it is not possible to simply stop Madam from subdividing the maps internally? If not I suppose the best option to minimise fragmentation would be to simply have nside_submap = 1?

[keskitalo]

Even at nside_submap = 1 there are 12 distinct submaps: https://github.com/hpc4cmb/libmadam/blob/master/src/parameter_control.f90#L255

However, the issue with performance is not limited to submaps. If pixel numbers are completely random, the memory access patterns will lead to repeated cache misses and cause the code to run much slower than possible. Just something to consider before launching large production runs.