Header logo is

Probabilistic Filtering ODE Solver


A numerical ODE solver implemented in Matlab which returns a Gaussian probability distribution over the solution. It's reliable, fast and mostly API compatible.

This software release supplements the paper:

Michael Schober, Simo Särkkä, Philipp Hennig: "A probabilistic model for the numerical solution of initial value problems", 2017.

The numerical implementation is provided in the Matlab programming environment. The probabilistic Nordsieck method is implemented in the function


whose interface resembles other numerical differential equation solvers available in Matlab. Other functions in


implement additional functionality associated with the filter output, such as computing the smoothing distribution and sampling from the predictive posterior. To reproduce the illustrative plots from Section 2 of the paper, open a Matlab instance and type

setup; Sec2Figure

Code to reproduce the benchmark comparison is partially copyrighted by Matlab and, thus, cannot be publicly released. We are working to provide an alternative soon. Further symbolic algebra code written in Python is provided to check the derivations.

Author(s): Michael Schober
Department(s): Probabilistic Numerics
Research Projects(s): Probabilistic Solvers for Ordinary Differential Equations
Publication(s): Probabilistic {ODE} Solvers with Runge-Kutta Means
A probabilistic model for the numerical solution of initial value problems
Authors: Michael Schober
Maintainers: Michael Schober
Release Date: 2017-05-19
License: Apache License, Version 2.0 (Apache-2.0)
Repository: https://github.com/mpi-is/pfos/