Madam not working correctly when using in obs.tod

[nraffuzz]

This configuration was used to save TOD components (CMB, wn_1f_100mHz, wn_1f_30mHz, wn, fg) for the Madam mapmaker, and each combination should have been used individually:

obs_multitod.tod_wn_1f_100mHz = np.zeros_like(obs_multitod.tod)
obs_multitod.tod_wn_1f_30mHz  = np.zeros_like(obs_multitod.tod)
obs_multitod.tod_wn           = np.zeros_like(obs_multitod.tod)
obs_multitod.tod_fg           = np.zeros_like(obs_multitod.tod)

In order to save memory the attribute obs_multitod.tod of the Class Observation was used to store CMB instead of using an additional field .tod_cmb, thus leaving .tod unused. In reality, Madam stored all the components in obs_multitod.tod, therefore when Madam produced maps (whether it was binning or destriping), the output showed an excess of power when referring to the case .tod, whereas a correct behaviour otherwise.

Issue is solved by correcting .tod with .tod_cmb. However, in this way, the obs.tod field is not used.

[ziotom78]

Sorry, I don't understand what you mean with «Madam stored all the components in obs_multitod.tod. Madam should read those components from FITS files, perhaps you were referring to the function saving TODs for Madam? Or did you mean, «Madam read all the components…»?

[nraffuzz]

Sorry, I wasn't clear. Basically when calling add_noise_to_observations and scan_map_in_observations "By default, the noise/signal is added to Observation.tod". @marcobortolami and I wanted to save memory by filling obs.tod with CMB only, and instead we don't need a field where all the component-TODs are summed.

Is there a reason for this feature to be set by default? Thanks

[ziotom78]

Oh, I see. Yes, I kept the tod field as the default in order not to introduce breaking changes. But you should be able to inject the noise and signal to whatever field you want through the component parameter, which is accepted both by add_noise_to_observations and scan_map_in_observations:

lbs.add_noise_to_observations(obs, "white", component="tod_wn")
lbs.scan_map_in_observations(obs, maps, component="tod_fg")

Or is the problem due to the fact that there is a bug and the noise/signal are always added to tod regardless of the value of component?

[nraffuzz]

Yeah, the point is exactly that, regardless of the correct usage of component we think the noise/signal is also added to tod, here is said to be added for the noise case and here is said the same for the signal.

A test @marcobortolami and I made was to produce a binned map of CMB only using the Madam mapmaker. Images below show the temperature map and spectrum (the last is EE spectrum instead), which for sure comprehend fg and white noise. (Also the y-axis label is incorrect, those are D_ell)

[nraffuzz]

Also, I forgot to state clearly that make_bin_map correctly works. It just seems that, when using the field .tod with Madam, not only the CMB is added.

[ziotom78]

I created a new PR in the litebird_sim repository to test the behavior of the framework when using component=, but I was not able to reproduce the result.

The two new tests are test_scanning_list_of_obs_in_other_component (in test_scan_map.py) and test_add_noise_to_observations_in_other_field (in test_noise.py). Both inject the signal in a field different from the usual tod and check two conditions:

1. tod should be unaltered (as it was set to zero, it just checks that all the samples are still zero)
2. The new field should have something in it

Both tests pass, so I do not know what's happening here…