This code performs Laplace and Matern interpolation where missing data are on a one, two, or three dimensional grid. Matern kernels generalize the radial basis function approach to interpolation, but interpolation using these kernels involves systems of equations that are dense. By using the Green's function representation, and substituting the finite-difference operator, we replace the dense operator with a sparse one and thus obtain an approximation to the kernel.
TL;DR: Substituting a discrete Laplace approximation to Matern kernel reduces the computational complexity of gridded interpolation from
This package is unregistered, so please install using
pkg> add https://github.com/vishwas1984/LaplaceInterpolation.jl
Jupyter Notebooks which illustrate the speed and accuracy of the approximation
are located in the /Notebooks directory.
To run the examples yourself, clone the repo, navigate to the Notebooks directory, start julia and use
julia> using Pkg
] activate .
julia> Pkg.add(url="https://github.com/vishwas1984/LaplaceInterpolation.jl")
julia> Pkg.instantiate()
julia> include("run_notebooks.jl")
which will start a jupyter notebook for you, with all relevant dependencies.
Please refer to the Examples notebooks for useage. Github displays notebooks with output, if you do not wish to run them yourself.
Below we show an example of a three dimensional x-ray scattering experiment on a crystalline structure in which peaks at integer locations are removed from the dataset and interpolated. The plot shows a one-dimensional cut of the 3D data along the h-axis. The image on the left shows the data with and without interpolation (the original data is in red, Green and Orange respectively show Laplace and Matern interpolated data). The right hand side image is a blown-up version of that on the left, using a linear scale in the y-axis.
| Bragg Peaks | Matern and Laplace Interpolation |
|---|---|
 |
 |
For large three-dimensional datasets such as this one, an RBF kernel interpolation
fails on a laptop computer with contemporary (2021) architecture. The gridded interpolation
approach we give here can tackle the above
Laplacians.jl authored by Dan Spielman.
This work appears in the Journal of Open Source Software. Clicking "Cite this repository" on the right hand side (under "About") will format a bibliography entry for you.
Contributions are welcome in the form of issues, pull requests. Support questions can be directed to vhebbur at anl.gov.
This material is based upon work supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences.