Martin Robinson

I used `mprof` https://pypi.org/project/memory-profiler/

It could be that the three variables reference the memory in the others, so deleting one has no effect as the memory is still referenced somewhere (this is a guess!)....

"we're allocating memory for return vectors in the solve, this could be done in solver setup" Need to think more on this one, if we re-use this memory for subsequent...

this would work well, be particularly good for the jax evaluator. Would it go in the simplification code? i.e. https://github.com/pybamm-team/PyBaMM/blob/52e2d964ac7fcff26da459f8b07a1d7b8eaf41c2/pybamm/expression_tree/binary_operators.py#L1245

some issues with the 1st and last elements of the calculation, for the example above the first row is `-2*y[0]+y[1]` and the last is `-2y[-2] + y[-1]`, so have three...

sadly I'm not seeing any speed-ups here. @tinosulzer did you want to have a look at this and see if what I implemented is what you were thinking of?, the...

yup, compiling is very slow. My thoughts on this solver is that it is mainly good for parameter estimation or inference, where you need to call the solver many many...

no, jax doesn't directly support sparse matrices. Instead, they implement vjp's, or vector-Jacobian-products, based on the rhs that you supply. These vjps are calculated automatically via automatic differentiation. So my...

nice :) is this in the `develop` branch, or elsewhere? Do you have a benchmark script handy that I can base off of?

Here are the timings for `issue-1235-jax-sparse-mat-vec`, which is branched off `develop`. I just modified the `DFN.py` script as you said, and took the times from the *2nd* solve so that...