Andraž Jelinčič

I would summarise the complete behaviour in 3 rules: 1. We always have `t1-t0 =prev_dt`. 3. If the step was rejected, then `next_dt` must be `< t1-t0`. These can be...

Thanks, that's good to know. I will keep the number of `eqx.error_if`s to a minimum. In practice it seems that your proposal doesn't lead to desirable behaviour. I compared our...

Fair enough, I'll get rid of it then. On the flip side, as I mentioned in #484, it seems like it was possible for the step size to increase after...

Well technically this prevents it only from shrinking below `t1-t0`, which can still be smaller than `prev_dt`. But as you said before, probably all of these choices can bring pluses...

Thanks for spotting this, Sam. Idk how the `h = t0 - t1` slipped past me, it's just clearly wrong. In terms of VBT, I think `_tree._levy_diff` should handle this...

Hi! I think you just need to add a `jnp.abs` on the line https://github.com/patrick-kidger/diffrax/blob/80d1f34f7608629246600831f9397eb6fc5973a3/test/helpers.py#L103 If that doesn't work, I can take a closer look.

Great to see the others passed!! As far as I know, Milstein is the only SDE solver in Diffrax which uses derivatives, so that could be the cause. But otherwise...

I completely agree. In fact I've been eyeing that split by PyTree for a while now and am intending to refactor it soonish. In addition I am intending to add...