[ANN] ConvolutionInterpolations.jl at v1.0.0

Package Announcements
1

Hi fellow Julia people :slight_smile:

I’m proud to announce that the API of ConvolutionInterpolations.jl is finally stable at v1.0.0.

I’m very pleased with the final feature-set of the package.
Notably, since JuliaCon Mainz 2026, I’ve succesfully added:

  • Faster and more accurate kernel evaluation through direct application of Horner’s scheme (thanks again @mkitti).
  • Fast repeated high-order integrals up to 8th integral in arbitrary dimensions.
  • The capability for fully scattered data through fitting a gridded interpolant.

The latter allows for high-order derivatives and integrals of fully scattered data.
Notably, the scattered fit is achieved through a sparse kernel solve, which, contrary to dense radial-basis function solves, scales very well, both in setup and evaluation time. :slight_smile:

I’m working on a Quarto-based documentation website here: https://convinterp.org/
This site will document both the Julia reference and the Python port which I’m working on too.

I’m very grateful to all of the input I got in the initial thread here on discourse for my package, which motivated sustained development and improvements, as well as the positive experience at JuliaCon Mainz. :slight_smile: I should also note that this package was developed with assistance from Anthropic’s Claude models.

I hope this package will be useful to the community! It was a joy to develop it, and it was motivated from the pleasure of pure math, and not a specific application.

2

Having examined my claims regarding the RBF-comparison, I have to be more careful:
There’s no clear overall winner in a head-to-head comparison with ScatteredInterpolation.jl: While my package evaluates more rapidly, and is more accurate in 2D, RBF in 3D solves faster and have higher accuracy/lower error there. In any case, what my package does uniquely is the scattered integral capability. :slight_smile: