owlbarn/owl_ode

Owl's Differential Equation Solvers

owl_ode's Issues

Add option to deal with events

incorrect handling of matrix-shaped states

This fails, despite what the README.md says about integration of matrices:

open Owl
open Owl_ode

let tspec = Types.(T1 { t0 = 0.; duration = 1.; dt = 1E-3 })
let solver = Owl_ode_sundials.cvode ~stiff:false ~relative_tol:0. ~abs_tol:1E-5
let n = 10
let batch_size = 3
let dx x _ = Mat.(neg x)
let x0 = Mat.gaussian batch_size n
let times, states = Ode.odeint solver dx x0 tspec ()

For reference, the README.md says:

We also support temporal integration of matrices. That is, cases in which the state x is a matrix of dimensions of dimensions NxM. By default, in the output, we flatten and stack the states vertically (i.e., ts has dimensions Tx1 and xs has dimensions TxNM.

Just logging this issue so I don't forget, will try to take a look later ─ probably a bug introduced in Dec 2019 with a commit called Arr integration.

Document the interface in more details

The only thing that prevents us to release the current implementation is the lack of documentation, we should document the interfaces cleanly without sparing details.

RK4 seems broken

We need to double check, but I am getting inconsistent results (while the other methods seem to consistently agree)

Cleanup examples

Lost interface documentation of native and symplectic modules

It looks like the changes to owl_ode.mli introduced in #14 have made the part of documentation of the Native and Symplectic modules disappear (not the autocompletion or merlin information, just odoc's documentation).

This may be a chance to document better the interface in a new native.mli and symplectic.mli module, although it would be nice to be able to re-enable odoc documentation there

Integrating backwards in time

The current design does not naturally support integrating odes backwards in time. This will require some thinking about typetspec and how it affects the different solvers.