SUNDIALS v7.4.0 Release Experiments

This repository contains experiments used in the SUNDIALS v7.4.0 release paper.

Prerequisites

A C and C++ compiler, CMake, and OpenMP are needed to build the SUNDIALS-based experiments. For the Julia experiment, the Julia language is required.

The specific compilers and library versions used for the release paper are:

GNU gcc/g++:

$ cc --version
cc (GCC) 8.5.0 20210514 (Red Hat 8.5.0-26)
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Julia:

$ julia --version
julia version 1.8.3

Building

To build the SUNDIALS-based C/C++ experiments:

git clone https://github.com/sundials-codes/release-7.4.0-experiments.git && cd release-7.4.0-experiments
cmake -S . -B builddir
cmake --build builddir -j8 

The Julia experiment is self-contained and will install the Julia packages it requires.

Running

Gray–Scott

Slurm batch scripts are provided to run the Gray–Scott experiments. To generate a reference solution and the operator splitting results, set the CMake option SUNDIALS_PRECISION to EXTENDED and execute the following commands:

cd Gray-Scott
mkdir data
sbatch ref.sh
sbatch splitting.sh

Finally, return the SUNDIALS_PRECISION to DOUBLE for the LSRK experiments and run

sbatch lsrk.sh

The solutions and runtimes are writtin into the data directory. A Python script is included to convert these into CSV tables:

python3 postprocess.py --method Splitting --grid_pts_1d 128 --plot
python3 postprocess.py --method LSRK --grid_pts_1d 1024 --plot

Lotka Volterra

To run the ARKODE version as done for the paper:

$ cd builddir
$ ./run_experiment.sh

To run the Julia version as done for the paper:

$ cd Lotka-Volterra
$ julia lotka_volterra_zygote.jl
OrdinaryDiffEq computed ||u(t_f)||: 1.3714668933550804
Discrete SciMLSensitivity computed sensitivities L2 norm: 0.7644151699219316
Discrete ForwardDiff computed sensitivities L2 norm: 0.2999335581117122 0.7031147934185299
Discrete Zygote (reverse mode) computed sensitivities L2 norm: 0.2999335581120665 0.703114793419319