Questions/comments on CCE tutorial

[moble]

I've gone through the CCE tutorial, and it looks pretty good. I think it should be very helpful to users in the workshop. I have a few fairly minor comments and questions about it, though.

• Under "Acquiring the CCE module", it says that there should be a tarball in every release starting from late May 2024. I guess that release doesn't exist yet, but when it does, I assume it will be in the "Assets" section. If so, that should probably be mentioned, because there can be a lot of text to look through, especially for people who haven't looked at a release for the assets before.
• I'm not convinced that the HDF5 type for the "Legend" attribute is very important, and it kind of hits like this tutorial is going to be more of a slog than it really is.
• I think the iterative angular solve warning I got was at least a couple orders of magnitude larger than the 2e-8 in the example. Is there some level at which I should be worried that the result is just garbage?
• Going from the ExampleInput.yaml that @nilsdeppe pointed me to, my output seemed to have excessively fine time steps. This was also something I saw in data that Nils had run himself. Throughout inspiral the timestep is ~4e-2, and drops as low as ~5e-3 after merger; compare these to 0.5 and 0.05, respectively, for a typical extrapolated waveform. It really makes BMS correction extraordinarily slow. The tutorial suggests ScriOutputDensity is important, but that's already just 1. Is there some other default we could tweak to lighten the load a bit?
• It looks like the datasets in the CCE output always include ell=0 and ell=1 modes, even when the spin is nonzero. This may be surprising (it was to me), so I think it should be mentioned in the "Output from CCE" section, right under where the h5ls output is listed. And maybe remind the reader that there's a "Legend" attribute that's relevant.
• In the BMS-correction code in scri, there's an option to correct by the Christodoulou mass. Where is this number found?
• The link to the scri tutorial should probably point directly to the relevant section, because there's a lot on that page, and it's not clear where to find the part about adjust the BMS frame.

[nilsdeppe]

Thanks, Mike!

1. Legend attribute: it turns out this was an issue! Deborah had the encoding incorrect or something like that and everything exploded in a cryptic manner. Do you have a suggestion on how to make sure this is clear but maybe not intrusive?
2. Output density: So this is a slightly bigger improvement that needs to be looked at. @keefemitman suggested limiting the time step size to 0.1M. However, CCE currently doesn't have an option to output every N steps, only N times per step. Presumably what's happening is you're getting output at 0.1M density and this is just way too often? What I want to do but haven't had time yet is to make it so we can output data every N steps in addition to N times per step. I'll check to see how difficult that is to add.

[keefemitman]

@moble the very fine time steps stem from the error based time stepping that CCE utilizes; this is why I added the t_interpolate option (https://github.com/moble/scri/blob/62df4db6099bd3d11035e2a6325d2524e6281168/scri/SpEC/file_io/__init__.py#L485) to create_abd_from_h5 because, for practical purposes, we will always want to use the time array of the worldtube from the Cauchy evolution as t_interpolate. (At least this is what we decided on a while back in the vacuum call for the catalog).

[knelli2]

Thanks for going through this @moble! I can definitely implement these suggestions. Yeah I think unfortunately the legend stuff is important since our H5 IO looks for it. I also think @keefemitman's solution for the time stepping is the way to go, and we can address outputting every N steps in the future.