JAX rewrite?

[htjb]

maxsmooth typically takes order seconds to fit constrained polynomials which means that likelihood functions that use it are quite time consuming. In practice this could probably be sped up by utilising JAX and GPUs. We can take advantage of the autograd functionality to get derivatives of the polynomials and it appears that there are packages written in JAX for quadratic programming (https://jaxopt.github.io/stable/quadratic_programming.html).