matexp

This program solves systems of differential equations for the NEURON simulator using the matrix-exponential method of integration. This is a new method of integration. The solution is faster and more accurate than NEURONs built in "sparse" solver. This method is only applicable to systems which are linear and time-invariant, such as Markov kinetic models. This method is also limited to systems with one or two inputs.

This program uses the NMODL file format (".mod" files). The input kinetic model is an NMODL file, and the solution is written to a new NMODL file.

For more information about how this program works see DETAILS.md.

Installation

Prerequisites:

• Compiler for the target system.
  • CPU requires g++
  • Cuda requires the cupy python package.

$ pip install matexp

Usage

$ matexp --help
usage: matexp [-h] [-v] [--plot] -t TIME_STEP [-c CELSIUS] [-e ERROR]
              [-i NAME MIN MAX] [--log [INPUT]] [--target {host,cuda}]
              [-f {32,64}]
              INPUT_PATH OUTPUT_PATH

positional arguments:
  INPUT_PATH            input filename for the unsolved NMODL file
  OUTPUT_PATH           output filename for the solution

options:
  -h, --help            show this help message and exit
  -v, --verbose         print diagnostic information, give twice for trace mode
  --plot                show the propagator matrix

simulation parameters:
  -t TIME_STEP, --time_step TIME_STEP
  -c CELSIUS, --celsius CELSIUS
                        default: 37°
  -e ERROR, --error ERROR
                        maximum error per time step. default: 10^-4

input specification:
  -i NAME MIN MAX, --input NAME MIN MAX
  --log [INPUT]         scale input logarithmically, for chemical concentrations

computer specification:
  --target {host,cuda}  default: host
  -f {32,64}, --float {32,64}
                        default: 64