Structural identifiability of ODE models by Lie symmetries over finite fields.
symident finds the non-identifiable directions of an ODE model with observables, conditions and events, returns them as exact generators where possible, and reduces the model to identifiable parameters. The observability rank is computed modulo primes in a C++ kernel; closed forms are reconstructed by sparse interpolation and proven over Q or verified at fresh primes.
pip install symidentWheels are built for Linux (x86_64, aarch64, glibc 2.28 or newer; tested on Ubuntu, Debian
and Fedora), macOS (x86_64, arm64) and Windows (x86_64), for Python 3.10 to 3.14. Elsewhere
pip builds from source, which needs a C++17 compiler and CMake; with OpenMP the kernel runs
on several threads. The development version installs with
pip install git+https://github.com/dModverse/symident.
Steady-state analyses (equilibrate=True) whose resting state is a coupled nonlinear
polynomial system use msolve. On Linux and macOS
import symident
symident.install_msolve()builds it with GMP, MPFR and FLINT into a per-user cache (needs a C compiler, make, m4
and curl or wget), where symident finds it. Alternatively point SYMIDENT_MSOLVE at an
msolve executable; on Windows that is the only way.
import symident as si
m = si.Model({"A": "-k1*A", "B": "k1*A - k2*B"}, {"y": "s*B"})
res = si.detect(m, reconstruct=True)
print(res)
#> Result: rank 4 / 5 | 1 non-identifiable direction
res.symmetries[0].generator
#> {'A': 'A', 'B': 'B', 's': '-s'}
print(res.reduce())
#> Reduced 1 of 1 direction.
#> Trafo (non-identity entries)
#> A = 1Model(odes, observables, ...) |
ODEs or a reaction network, observables, trafo, conditions, events, inputs, fixed |
detect(model, ...) |
rank, identifiability and the non-identifiable directions (Symmetries) |
reduce(result, ...), result.reduce() |
reparametrization that removes the directions (Reduction) |
reconst_control(...) |
settings of the saturation and the reconstruction |
switches(...) |
diagnostic and route switches for a with block |
Both results print a report; summary() adds the computation details and
to_dict() returns the raw result.
Simon Beyer, simon.beyer@fdm.uni-freiburg.de. Requests for further functions in model equations, questions and bug reports are welcome by e-mail or as an issue; the supported functions are listed in the user guide.
MIT