Roadmap

gotranx is under active development and there are several features that we want to add in the future.

The plan is to have all the features from the old gotran implemented in gotranx (and some more). This includes

• [ ] More numerical schemes

  • [x] Forward Euler

  • [x] Generalized Rush Larsen

  • [x] Hybrid Generalized Rush Larsen

  • [ ] Simplified Implicit Euler

• [ ] Code generation for more languages

  • [x] Python

  • [x] C

  • [x] jax (working but missing tests)

  • [ ] C++

  • [ ] Julia (in progress)

  • [ ] CUDA

  • [ ] OpenCL

  • [ ] Rust

  • [ ] Latex

  • [ ] Markdown

• [x] Converters between commonly used ODE markup languages

  • [x] Myokit (still some limited support for unit conversion, see issue #26)

  • [x] CellML (supported via MyoKit)

• [ ] Add support for lookup tables. A master student has currently implemented support for this in legacy gotran, see https://www.mn.uio.no/ifi/studier/masteroppgaver/bmi/automated-code-generation-for-simulating-cardiac-c.html

• [ ] Better handling of singularities, see ongoing work here finsberg/gotranx#68

  • The Rush-Larsen schemes linearize by differentiating through intermediates, which multiplies removable singularities in terms like x/(exp(x) - 1). remove_singularities should be extended over the linearized block.

  • sympy.cse additionally hoists subexpressions out of Piecewise branches, so a guarded singular term can become an unconditionally evaluated temporary (verified on the pinned sympy 1.14: the derivative of Piecewise((0, V < -40), ((V+40)/(exp(-(V+40)/10)-1), True)) CSEs to a temporary 1/(1 - x0) that is singular at V = -40 and is now evaluated unconditionally) – harmless today, since numpy’s where evaluates both branches anyway and in C the IEEE-754 inf is discarded by the ternary unless the user traps FP exceptions, but it is a second, distinct interaction with CSE worth recording alongside the one above.

If you have additional feature requests, please open an issue