Changelog

All notable changes to multipoles are documented here. The format is based on Keep a Changelog.

Unreleased

MultipoleExpansion.field(x, y, z): the field E = -∇Φ, computed analytically.
center argument to expand about any point instead of the center of charge. This makes interior expansions possible where the center of charge coincides with a charge.
validity_radius and is_valid(x, y, z): where the expansion converges.
ExpansionValidityWarning when evaluating the expansion where it does not converge, and the charge_tol argument to control how much of a continuous charge density may be neglected for this check.
dipole_moment and quadrupole_moment: the cartesian dipole vector and traceless quadrupole tensor.
total_charge for discrete charge distributions (previously only for continuous ones).
multipole_moments_array: the multipole moments as array, indexed by [l, m].
A readable repr, type hints for the public API, and ExpansionValidityWarning, InvalidChargeDistributionException and InvalidExpansionException exported from multipoles.

Much faster: computing the moments of a 101³ grid takes 0.17 s instead of 2.4 s for l_max=4, and 0.3 s instead of 24 s for l_max=10. Evaluating the expansion is about 20 times faster.
Evaluating the potential at points inside the charge distribution (exterior expansion) or outside of it (interior expansion) now emits an ExpansionValidityWarning. This includes evaluating on the whole grid with Phi[:, :, :]. The results there were never meaningful.
total_charge of continuous charge distributions is now integrated with Simpson's rule, consistent with the multipole moments, instead of a plain sum over the grid.
multipoles no longer uses scipy.special for the spherical harmonics.

Import error with SciPy >= 1.17, where scipy.special.sph_harm was removed.
Phi(x, y, z, l_max=...) ignored l_max.
exterior=False alone selected the exterior instead of the interior expansion.
Phi[...] used the exterior expansion for interior expansions and modified internal state.
Evaluating Phi[...] for discrete charges failed with an AttributeError; it now raises a clear TypeError.
Invalid input now raises errors instead of silently producing NaN: zero total charge, charge at the center of an interior expansion, grids from np.meshgrid without indexing='ij', and mismatched grid shapes.
Grids with descending coordinates gave wrong results.
Evaluation accepts arrays of any shape, including 0-d arrays, and is vectorized.