https://github.com/woodward/integrator
A numerical integrator written in Elixir for the solution of sets of non-stiff ordinary differential equations (ODEs).
https://github.com/woodward/integrator
adaptive-stepsize bogacki-shampine dormand-prince elixir elixir-nx numerical nx ode-solver odes ordinary-differential-equations runge-kutta runge-kutta-adaptive-step-size runge-kutta-methods
Last synced: 5 months ago
JSON representation
A numerical integrator written in Elixir for the solution of sets of non-stiff ordinary differential equations (ODEs).
- Host: GitHub
- URL: https://github.com/woodward/integrator
- Owner: woodward
- License: mit
- Created: 2023-04-01T22:46:26.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-12-05T02:44:55.000Z (6 months ago)
- Last Synced: 2024-12-12T12:25:49.280Z (5 months ago)
- Topics: adaptive-stepsize, bogacki-shampine, dormand-prince, elixir, elixir-nx, numerical, nx, ode-solver, odes, ordinary-differential-equations, runge-kutta, runge-kutta-adaptive-step-size, runge-kutta-methods
- Language: Elixir
- Homepage:
- Size: 887 KB
- Stars: 33
- Watchers: 4
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.md
Awesome Lists containing this project
README
# Integrator
A numerical integrator written in Elixir for the solution of sets of non-stiff ordinary differential
equations (ODEs).
## Installation
The package can be installed by adding `integrator` to your list of dependencies in `mix.exs`:
```elixir
def deps do
[
{:integrator, "~> 0.1"},
]
end
```
The docs can be found at .
## Description
Two integrator options are available; `ode45` which is an adaptation of the [Octave
ode45](https://octave.sourceforge.io/octave/function/ode45.html) and [Matlab
ode45](https://www.mathworks.com/help/matlab/ref/ode45.html). The `ode45` integrator utilizes the
[Dormand-Prince](https://en.wikipedia.org/wiki/Dormand%E2%80%93Prince_method) 4th/5th order Runge
Kutta algorithm.
`ode23` is an adaptation of the [Octave
ode23](https://octave.sourceforge.io/octave/function/ode23.html) and [Matlab
ode23](https://www.mathworks.com/help/matlab/ref/ode23.html) The `ode23` integrator uses the
[Bogacki-Shampine](https://en.wikipedia.org/wiki/Bogacki%E2%80%93Shampine_method) 3rd order Runge
Kutta algorithm.
Both `ode45` (which is the default integrator option) and `ode23` utilize an adaptive stepsize
algorithm for computing the integration time step. The time step is computed based on the
satisfaction of a required error tolerance.
This library heavily leverages [Elixir Nx](https://github.com/elixir-nx/nx); many thanks to the
[creators of `Nx`](https://github.com/elixir-nx/nx/graphs/contributors), as without it this library
would not have been possible. The [GNU Octave code](https://github.com/gnu-octave/octave) was also
used heavily for inspiration and was used to generate numerical test cases for the Elixir versions
of the algorithms. Many thanks to [John W. Eaton](https://jweaton.org/) for his tremendous work on
Octave. `Integrator` has been tested extensively during its development, and has a large and growing
test suite.
## Usage
See the Livebook guides for detailed examples of usage. As a simple example, you can integrate the
Van der Pol equation as defined in `Integrator.SampleEqns.van_der_pol_fn/2` from time 0 to 20 with an
intial x value of `[0, 1]` via:
```elixir
t_initial = 0.0
t_final = 20.0
x_initial = Nx.tensor([0.0, 1.0])
solution = Integrator.integrate(&SampleEqns.van_der_pol_fn/2, [t_initial, t_final], x_initial)
```
Then, `solution.output_t` contains a list of output times, and `solution.output_x` contains a list
of values of `x` at these corresponding times.
Options exist for:
- outputting simulation results dynamically via an output function (for applications
such as plotting dynamically, or for animating while the simulation is underway)
- generating simulation output at fixed times (such as at `t = 0.1, 0.2, 0.3`, etc.)
- interpolating intermediate points via quartic Hermite interpolation (for `ode45`) or via cubic
Hermite interpolation (for `ode23`)
- detecting termination events (such as collisions); see the Livebooks for details.
- increasing the simulation fidelity (at the expense of simulation time) via absolute tolerance and
relative tolerance settings
## So why should I care??? A tool to solve ODEs? WTF???
The basic gist of the project is that it is a tool in Elixir (that leverages [Nx](https://github.com/elixir-nx))
to numerically solve sets of ordinary differential equations (ODEs). Science and engineering
problems typically generate either sets of ODEs or partial differential equations (PDEs). So basically
`integrator` lets you solve any scientific or engineering problem which generates ODEs, which is a
HUGE class of problems (FYI, finite element methods are used to solve sets of PDEs).
Fun fact: hundreds (or even thousands) of scientific problems had been formulated in the form of ODEs
since the time that Isaac Newton first invented calculus in the 1600's, but these problems remained
intractable & unsolvable for over three centuries other than a very small handful that were amenable
to a "closed form solution"; i.e., the ODEs could be solved analytically (i.e., via mathematical
manipulations). So there was this tragic dilemma; we could formulate these problems mathematically
since the 1600's - 1800's, but couldn't actually solve them. SAD! :disappointed:
So one of the primary drivers to create the first digital computers in the 1940's - 1960's was
to solve ODEs. The space program, for example, would have been impossible without the numerical
solution of ODEs which represented the space flight trajectories, attitude, & control. And before
the first digital computers, analog computers were used to solve ODEs back in the 1920's - 1940's.
So believe it or not, the first computers were developed and used to solve ODEs, not play League of
Legends. :wink:
These algorithms are battle-tested and in some cases have been around for decades; Matlab and Octave
are just relatively clean implementations of some of these algorithms, so I used them as the basis
for my Elixir versions.