Suggestion: New solver Jones-modes

[henedler]

The following modes are currently supported by the DP3 solvers: scalar, scalarphase, scalaramplitude, diagonal, diagonalphase, diagonalamplitude, fulljones, tec, tecandphase, rotation and rotation+diagonal .

In a few cases, it would be useful to have additional solve types to reduce free parameters by leveraging known physical effects:

• FaradayRotation+diagonal would help to greatly stabilize the estimate of the polarization misalingment and faraday rotation in the LOFAR LBA calibrator pipeline when also including international stations. The idea is to not solve for one rotation per channel, but to enforce that the rotation follows a RM*lambda^2 frequency dependence. The resulting h5parms would be a diagonal matrix and a faraday000 soltab.
• FaradayRotation: Fitting a rotation that follows the RM*lambda^2 frequency dependence. Useful e.g. for DD-calibration in a regime where FR is not negligible (LBA IS, LBA decameter).
• TEC+clock would be useful to derive the clock delay of the individual stations in the calibrator pipeline. In general, the function of the form anu+bnu^(-1) shows many comparable local minima, these can be mitigated when using wideband data or constraining the clock delay to be less than one wrap (~LOFAR 2)

Currently, physical effects such as clock+tec or Faraday Rotation are oftentimes fit to the h5parm solutions in solutions space. This indirect method is not optimal since we don't take into account the visibility weights.

Cheers,
Henrik

[maaijke]

Hi Henrik,
If it is not too much work it could probably be a good idea to implement these functionalities, but we might need some more discussion on the usability. Clock/TEC fitting on the visibility phases will not lead to optimal solutions, not only because of the local minima issues you mentioned yourself (although it might well work in LOFAR2 if there is a common clock) but also because you are ignoring second order effects like cable reflections and the constant phase offset. Proper Faraday rotation including the 1/frequency^2 effect is probably easier , so it would certainly be interesting to see how that goes.