diff --git a/docs/guide/numerics/boundaries.md b/docs/guide/numerics/boundaries.md index 7d98eb4c1..91e7ff265 100644 --- a/docs/guide/numerics/boundaries.md +++ b/docs/guide/numerics/boundaries.md @@ -889,5 +889,5 @@ field = apply_split_step(field, boundary.temporal_rate, step=0.01) spectrum = apply_band_mask(np.ones((1, 1, grid.n_omega)), boundary.band_mask) # And the run says so if the absorber is too weak for the walk-off it will produce. -warn_walkoff(grid, group_delay_span=1e-12) # SolpropWarning: widen trange, or ... +warn_walkoff(grid, group_delay_span=1e-11) # SolpropWarning: widen trange, or ... ``` diff --git a/src/solprop/driver/routes/hollow.py b/src/solprop/driver/routes/hollow.py index f7d2241bf..32104923c 100644 --- a/src/solprop/driver/routes/hollow.py +++ b/src/solprop/driver/routes/hollow.py @@ -147,9 +147,8 @@ The transform is one synthesis matrix and its dual, which represent an orthonormal basis and nothing else, so a set that is not one is propagating a different problem -from the one it describes. The bound is loose because the overlaps are integrated -adaptively by :func:`scipy.integrate.dblquad`, whose own accuracy on a set of Bessel -profiles is around 1e-9. +from the one it describes. This is the tolerance on the supplied mode set's +overlaps, separate from the quadrature tolerance used to compute them. """ _MINIMUM_RADIAL_ORDER = 32 diff --git a/src/solprop/translate/stage.py b/src/solprop/translate/stage.py index 9a2241f0a..1b257f625 100644 --- a/src/solprop/translate/stage.py +++ b/src/solprop/translate/stage.py @@ -24,7 +24,7 @@ A z table has no such split (:attr:`~solprop.translate.problem.PropagationProblem.device_ready`), and :func:`require_device_ready` is the refusal an accelerator build raises. The design is -``docs/dev/kernel-contract.md`` section 3, "How a stage is evaluated"; the bundles are +``docs/dev/kernel-contract.md`` section 1, "Identity and validation"; the bundles are frozen by ``tests/backends/test_driver.py``. """