Is Amp doing better than ResAmp for NR?

[vijayvarma392]

For the NR case in the example notebook, it looks like Amp method is doing better:

Why? These kinds of things should be checked in the NR tests, for example by making a plot like this for all NR cases.

[md-arif-shaikh]

No, Amp does not do better than ResAmp, it's just that ResAmp is using extrema closer to merger which Amp does not find at all. As we have noticed earlier, using extrema close to merger gives unphysical features in spline and therefore also makes ecc non-monotonic. That's why we use num_orbits_to_exclude_before_merger. But the default value is 1 which is not good enough here. If we make num_orbits_to_exclude_before_merger=2 then they give the same result.

[md-arif-shaikh]

This raises two questions:

• Do we want to increase the default num_orbits_to_exclude_before_merger value?
• If with the default num_orbits_to_exclude_before_merger value, measured ecc is non-monotonic, do we want to add an option to try increasing num_orbits_to_exclude_before_merger iteratively up to a certain value to see if the non-monotonicity goes away?

[vijayvarma392]

Ok, I suggest studying all of the NR cases with num_orbits_to_exclude_before_merger = 1, 1.5, 2, to get a better idea of what works.

I personally would avoid the iterative option, I'm worried it can be a bit finicky/unstable.

[md-arif-shaikh]

current plots in the notebook are with value 1 and the behaviour is pretty much always that Amp looks better than ResAmp near the merger because of the reasons above. So I guess we definitely want to increase it a little bit. I will try 1.5 and 2 and see which works better