To Do

[moble]

• Change quiet to more fine-grained verbosity, and pass along verbose argument to DiffEq solve. Maybe just loglevel would be adequately fine-grained.
• Add more example systems — e.g., GWTC function that returns something like the system in question
• Collect citations in bibtex form using DocumenterCitations.jl
• Clarify the meaning of TaylorT4/T5 in terms of "Formal power series of logarithmic type". In particular, note the "augmentation" as an alternative to evaluation at zero. Or maybe just something simpler with residues.
• I think a slight extension of LabelledArray.jl is what I need; SLVector is almost it, but we also need the PNOrder ot be a type parameter. Look into using StructArrays.jl for the integration, so that each time step may be a StructArray. Rather than having to re-interpret / reset the sol.u field as a PNSystem, StructArrays would do this automatically, which may lead to more threading opportunities. Alternatively, ComponentArrays.jl may be the way to go.
• Support eccentricity
• Switch to using S instead of chi, since this is what Alvi uses; it's easy to convert to chi, but more complicated to convert back if M is time-dependent. Also, it makes me nervous that maybe I'm making a mistake about dimensions somewhere, so this could be a good opportunity to check dimensions.
• Think about Unitful+ComponentArrays
  See this comment et seq. for examples, but note that it can be slow for vectors with heterogeneous units. One workaround mentioned in that thread is to define a Union type, though it's still slower than without units. Maybe things are better by now.
  Maybe just accept unitful inputs to orbital_evolution, strip the units internally, and then replace the units on output.
  It would be interesting to analyze the BBH problem in terms of Buckingham's π theorem.
  Also note the package Dimensionless.jl. This actually goes the opposite way that I would go; I kind of think we need a Dimensionful package to deal with the fact that we have quantities with different dimensions that may nonetheless be measured in the same units when we use geometric units — e.g., when we measure the solar mass in microseconds or kilometers rather than kilograms. Also:
  - It couldbe helpful to formalize things a little more with Huntley's extensions and/or Siano's
  - Maybe account for affine-vs-vector quantities of each dimension
  - Support natural units (hbar=c=1) as well as geometric units (G=c=1)
  - Terence Tao has a writeup that's presumably up to his excellent standard
  - Note Sonin's 2004 generalization
• Re-enable support for TaylorT4/5
• Redefine NS coupling as normalized parameter so we don't have to divide by M^5.
• Check plotly-Firefox issues here and here. That's what's used in the docs, which will have squashed legends for Firefox users.
• Add cubic-in-spin terms