Jamie Sullivan

This should really be a test - for CLASS as well. And we should convert the FD script to a test as well (on a few FD well-behaved parameters)

I am surprised such a simple example fails given the fact Enzyme was applied to ODEs [previously](https://arxiv.org/abs/1812.01892)? I guess they were not using Rodas4 so not hitting BLAS (or whatever...

@marius311 Interested to try it out then - we tried Enzyme even though it is early since the developers (and others) recommended it to us at the [AD workshop](https://diffprogramming.mit.edu/schedule) -...

About to add a cleaned up first scipt version (ie_single_k.jl) that (finally!) works. It is slow and is only tested for a single relatively large k (~0.03 h/Mpc), but it...

The code I pushed has the file I described above, which uses the equations [here](https://arxiv.org/abs/2105.02887#:~:text=Calculations%20of%20the%20evolution%20of,of%20photon%20intensities%20and%20polarization.) (but in conformal newton gauge as I wrote out in the notes). I also turned...

Starting to look at running on GPUs. After confirming I can run the readme examples on NERSC, I tried to do a simple parallelization of the boltsolve call to the...

The idea I had above did not work. I updated that file (/examples/gpu_ensemble_boltsolve.jl.) to do this wrapping (which appears to work fine for the commented out EnsembleThreads() solve, by the...

After perusing the DiffEqGPU.jl github issues I think k must be passed as an Array rather than a scalar value...

Ok so after fixing this I get a new problem that occurs apparently where the solver is actually running (as opposed to just the setup failing). This seems like a...

I looked to see what I am doing differently from the readme example - I tried `Tsit5` instead of `KenCarp4`. Now I get an error that seems a bit more...