Created on 2023-08-30 08:24:16 by Didier Vibert. % Done: 0

The first pass does not perform any velocity fit and is using the fixed velocity, set with the parameter @ LineModelSolve.linemodel.velocityemission@. When this first guess is too far from the actual velocity, it produces a zPdf with a slope promoting high redshift solutions when the modeled velocity dispersion is too small, see for instance #8100. The reverse may happen when the initial velocity guess is too big, see #7655. These biases in the first pass pdf are leading to miss the good z-candidate which cannot be catch up by the second pass performing velocity fitting only on the retained candidates.

To avoid this phenomenon, we have to introduce a velocity fit in the first pass. It has to be parameterized: min, max, step should be different than second pass velocity fitting. Only a few coarse values should be tested. The fit with each velocity value should be performed at all redshifts. We then have to balance the gain in success-rate vs loss of cpu time.

we have then do decide what kind of first-pass pdf we retain:

• a sum over all velocities (marginalized over velocity)
• the one that gives the min chi2 in the plane (z, velocity)
• the best at each z (ie keeping the best velocity at each z)