Filippo Vicentini

Change spin convention to 1=↑ -1=↓, relax requirement of ordered Hilbert space values

1

Fixes #710 WIP, requires #1730 to pass tests. ```python In [4]: nk.operator.spin.sigmaz(nk.hilbert.Spin(0.5, 2), 0).get_conn(np.array([-1, -1])) Out[4]: (array([[-1, -1]]), array([-1.])) ``` Removes the requirement that the local values of the Hilbert...

Hilbert with dtypes

1

Prototyping Hilbert spaces with dtypes. On top of #1720 it's very, very straightforward. Maybe it's worth doing this (even if I keep asking myself why should we do that). ```python...

Remove numba4jax, numba callback closures from operators

5

I wrote Numba4jax to allow HamiltonianSampler to work with operators back when jax could not call back into python. This necessitated also to introduce those pesky `get_conn_flattened_closure` functions that returned...

Refining Continuous Operators: make them Jax-compatible pytrees

1

As the Jax operator interface for discrete operator is turning out very successful and extremely useful and easy to use, I'm thinking about turning ContinuousOperators into something very similar. This...

Automatic checkpointing

5

A long way to go... Will need some help to make this play nice with sharding and MPI. Must decide what to do with loggers. Right now we override the...

Reduce number of vjp calls in nk.jax.vjp for R->C functions

3

Noticed by some user on slack some time ago. Might lead to some speedups, but unsure about that. Should benchmark.