Rory Conlin

I've used https://github.com/tdkorn/sphinx-github-style in a few other projects and find it works fairly well.

I haven't seen any particularly clear explanation in the literature, the least bad is probably Hirshman's 1986 paper on free boundary VMEC (doi:10.1016/0010-4655(86)90058-5). That was using NESTOR to compute the...

For profiles it probably makes the most sense to "stretch" them so that their dependence on rho doesn't change (same for contracting, since we may not want to cut finite...

With @unalmis recent updates to grid tags, I think we could add a `requires_meshgrid` thing to the boozer qtys and then vmap them over rho, allowing you to do multiple...

Maybe `FixCoilCurrent(s)` to make it more explicit and avoid confusion?

> > define torsion as zero at corners > > Was this specifically for the test or should this be defined in the torsion compute function? I realized I don't...

> > ```diff > > | benchmark_name | dt(%) | dt(s) | t_new(s) | t_old(s) | > > | -------------------------------------- | ---------------------- | ---------------------- | ---------------------- | ---------------------- | >...

This might be a duplicate of #900. If we make the objective work for generic surfaces then it could work for coil/plasma or coil/vessel distance

I think this is probably a better implementation: ```python npts = 100 ncoil = 10 pts = jnp.array(np.random.random((ncoil, npts,3))) # or output of coilset.compute("x") def body(i): dx = pts[i] -...

For dealing with symmetry: I think the easiest option is to have `CoilSet.compute` return data for all the coils (ie via duplication, rotation, flipping etc), including virtual coils, similar to...