Kaya Unalmis

As discussed in #1508 indexing operations are bottlenecks for the transforms. Below is brief recap and my suggested solution that will also greatly reduce/simplify DESC code. 1. Recall there is...

then just store full tensor product mode set and a mask of the nonzero modes (instead of deleting them) in basis. Then if you need only nonzero modes in at...

> @unalmis can you take a look at the `Bounce{1,2}D` stuff? I didn't mark any attributes as static and things seem to work, I think because the `Bounce` objects are...

So just for computing stuff, e.g. `obj.compute`, exactly 25% of the total computation time is spent on doing these surface integrals. They require an FFTable grid anyway, so using the...

Here are the steps that should be taken to debug 1. Has this error ever occurred on the branch `Gamma_c` where this quantity is developed on? JAX version? 2. If...

Ok I ran optimizations leading up to ISHW, so commit 5cd7ebde563258f754a0401d9da6aa143bc3376f should not have this issue. State of branch at that commit https://github.com/PlasmaControl/DESC/tree/5cd7ebde563258f754a0401d9da6aa143bc3376f

> with use_jit=False (and commenting out the jit in constraint wrappers) I'm unable to reproduce There is also a jit call wrapping the compute function in `_compute`. When you could...

I think this is some jax issue; and the caching suggest this is problem dependent. In any case, I suggest trying on #1290, and if the issue disappears then can...

I accidentally ran the [the tutorial's optimization cell block 6](https://github.com/PlasmaControl/DESC/blob/ku/fourier_bounce_neo/docs/notebooks/tutorials/EffectiveRipple.ipynb) another time after the optimization completes successfully, and I get the same error. JIT caching is not done there, and...

> BUT, the error does not happen if I replace EffectiveRipple with ForceBalance Just in case I should point out that the "fix" from #1412 is not on #1003 ;...