crntk

A C99-compliant implementation of stochastic chemical reaction network (chemical master equation) operations.

Purpose

This library provides a straight-forward interface to specify a chemical reaction network and its conservation laws. The corresponding chemical master equation, $p'=Ap$, is then exposed to the user via a few operations. In particular, matrix-free evaluations of $Ax$ and $xA$ are made available, along with a few preconditioners. Because of the matrix-free formulation, the memory-requirements are independent of the number of reactions in the network, depending only on the conservation laws and initial conditions. Additionally, we represent the state-space exactly with a hyperplane rather than with a mostly empty bounding hypercube, which leads to significantly less memory-usage.

That said, this library is meant for exact computations on small chemical reaction networks only. Any time a copy number can be approximated with a concentration, or rare events are unimportant, alternative approaches should be considered (such as a stochastic simulation algorithm).

Documentation

This library is under heavy development at the moment. For now, only the C99 interface is provided. The intent is that in the near future, more elegant Python and Matlab wrappers will be made available. Nonetheless, the C99 interface is entirely self-contained, with no external dependencies. While my focus is on developing these wrappers, the examples (examples/*.c) and the header file (include/crntk.h) are both heavily commented and provide a full overview of the library and its capabilities. Similarly, documentation can be easily generated by running doxygen. Please feel free to email me with questions if you have any.

Notes

To get running quickly, from any terminal (OSX, Linux, Windows+WSL, or Windows+msys):

git clone https://github.com/jasondark/crntk

cd crntk

make examples

./bin/nullspace 10 1 1 1e-6

If that works, you're in business. The Makefile uses gcc by default and also compiles with -fopenmp: the crntk_id_apply() and crntk_tr_apply() methods are fully parallelized. If your state-space is very small, your program might run faster by disabling OpenMP (e.g. removing that -fopenmp flag).

Funding Acknowledgment

The core ideas of this software were developed while I was a graduate student at the University of California, Merced, where I received both intramural funding from the applied mathematics department and extramural funding from an NSF grant. The current implementation was developed while a postdoc at the University of California, Irvine, where my salary was derived from a DARPA grant. The content is solely the responsibility of the author and does not necessarily represent the official views of either of these agencies.