Skip to content

martinjrobins/diffsl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

b92e1f7 · Jan 29, 2025
Jan 17, 2025
Jan 29, 2025
Jan 3, 2025
Jan 3, 2025
Mar 14, 2024
Jan 29, 2025
Nov 2, 2023
Mar 9, 2024
Mar 19, 2024
Jan 29, 2025
Mar 6, 2024
Jan 3, 2025
Jan 17, 2025
Mar 31, 2024
Jun 11, 2024

Repository files navigation

DiffSL

CI build status badge

A compiler for a domain-specific language for ordinary differential equations (ODEs) of the following form:

M ( t ) d u d t = F ( u , t )

As an example, the following code defines a classic DAE testcase, the Robertson (1966) problem, which models the kinetics of an autocatalytic reaction, given by the following set of equations:

d x d t = 0.04 x + 10 4 y z d y d t = 0.04 x 10 4 y z 3 10 7 y 2 0 = x + y + z 1

The DiffSL code for this problem is as follows:

in = [k1, k2, k3]
k1 { 0.04 }
k2 { 10000 }
k3 { 30000000 }
u_i {
  x = 1,
  y = 0,
  z = 0,
}
dudt_i {
  dxdt = 1,
  dydt = 0,
  dzdt = 0,
}
M_i {
  dxdt,
  dydt,
  0,
}
F_i {
  -k1 * x + k2 * y * z,
  k1 * x - k2 * y * z - k3 * y * y,
  1 - x - y - z,
}
out_i {
  x,
  y,
  z,
}

DiffSL Language Features

See the DiffSL language documentation for a full description.

  • Tensor types:
    • Scalars (double precision floating point numbers)
    • Vectors (1D arrays of scalars)
    • N-dimensional tensor of scalars
    • Sparse/dense/diagonal tensors
  • Tensor operations:
    • Elementwise operations
    • Broadcasting
    • Tensor contractions/matmul/translation etc via index notation

Usage

Generally the easiest way to make use of DiffSL is via an ode solver that supports the language, for example the diffsol library. Please see the diffsol documentation and consult the DiffSL language documentation for more information.

If you are writing your own ode solver and want to make use of the DiffSL compiler, please either get in touch by opening an issue, contacting the author or by looking at the diffsol source code.

Dependencies

To use the llvm backend (optional) You will need to install the LLVM project. The easiest way to install this is to use the package manager for your operating system. For example, on Ubuntu you can install these with the following command:

sudo apt-get install llvm

Installation

You can install DiffSL using cargo. By default the cranelift backend will be used. To use the llvm backend (which generates more optimised code), you will need to indicate the llvm version you have installed using a feature flag. For example, for llvm 16:

cargo add diffsl --features llvm16-0

Other versions of llvm are also supported given by the features llvm15-0, llvm16-0, llvm17-0, llvm18-0.