Merge pull request #192 from patrick-kidger/some-fixes2

Minor fixes

Author: patrick-kidger

diffrax/step_size_controller/adaptive.py

@@ -465,6 +465,10 @@ def adapt_step_size(
         #
 
         def _scale(_y0, _y1_candidate, _y_error):
+            # In case the solver steps into a region for which the vector field isn't
+            # defined.
+            _nan = jnp.isnan(_y1_candidate).any()
+            _y1_candidate = jnp.where(_nan, _y0, _y1_candidate)
             _y = jnp.maximum(jnp.abs(_y0), jnp.abs(_y1_candidate))
             return _y_error / (self.atol + _y * self.rtol)
 

docs/further_details/faq.md

@@ -4,6 +4,7 @@
 
 - Use `scan_stages=True`, e.g. `Tsit5(scan_stages=True)`. This is supported for all Runge--Kutta methods. This will substantially reduce compile time at the expense of a slightly slower run time.
 - Set `dt0=<not None>`, e.g. `diffeqsolve(..., dt0=0.01)`. In contrast `dt0=None` will determine the initial step size automatically, but will increase compilation time.
+- Prefer `SaveAt(t0=True, t1=True)` over `SaveAt(ts=[t0, t1])`, if possible.
 - It's an internal (subject-to-change) API, but you can also try adding `equinox.internal.noinline` to your vector field (s). eg. `ODETerm(noinline(...))`. This stages the vector field out into a separate compilation graph. This can greatly decrease compilation time whilst greatly increasing runtime.
 
 ### The solve is taking loads of steps / I'm getting NaN gradients / other weird behaviour.
@@ -24,7 +25,7 @@ diffeqsolve(
 )
 ```
 
-In practice, [`diffrax.Tsit5`][] is usually a better solver than [`diffrax.Dopri5`][]. And the default adjoint method ([`diffrax.DirectAdjoint`][]) is usually a better choice than [`diffrax.BacksolveAdjoint`][].
+In practice, [`diffrax.Tsit5`][] is usually a better solver than [`diffrax.Dopri5`][]. And the default adjoint method ([`diffrax.RecursiveCheckpointAdjoint`][]) is usually a better choice than [`diffrax.BacksolveAdjoint`][].
 
 ### I'm getting a `CustomVJPException`.