Can I use JAX for evaluating hessians of complicated functions?

[infinite-pursuits]

Hi there,
First time using JAX. Apologies already for the noob questions.
I need to compute hessians for complicated functions. Getting hessians in functional form by hand takes some time. I am thinking if JAX can compute all this for me?
Is there a guarantee that the answer is going to be correct? What is the degree of approximation?

[georgedahl]

If you are new to JAX, some of your answers might be in the https://jax.readthedocs.io/en/latest/notebooks/autodiff_cookbook.html

[mattjj]

To answer the specific questions:

Is there a guarantee that the answer is going to be correct?

Yes, insofar as any software can be guaranteed bug-free. JAX differentiation is extensively tested (we have hundreds of thousands of tests). Moreover, you can always check the result yourself by testing against numerical differentiation in random directions.

What is the degree of approximation?

It's accurate to machine precision.