Synopsis
pyMOR is a BSD-licensed software library for building model order reduction applications in the Python programming language. All algorithms in pyMOR are formulated in terms of abstract interfaces, allowing generic implementations to work with different backends, from NumPy/SciPy to external partial differential equation solver packages.
Features
Currently, the following model reduction algorithms are provided by pyMOR:
- A generic reduction routine for projection of arbitrary high-dimensional discretizations onto reduced spaces, preserving (possibly nested) affine decompositions of operators and functionals for efficient offline/online decomposition.
- Efficient error estimation for linear affinely decomposed problems.
- Empirical interpolation of arbitrary operators (with efficient evaluation of projected interpolated operators if the operator supports restriction to selected degrees of freedom).
- Parallel adaptive greedy and POD algorithms for reduced space construction.
- Empirical-Interpolation-Greedy and DEIM algorithms for generation of interpolation data for empirical operator interpolation.
- Balanced-based and interpolation-based reduction methods for first-order, second-order, and port-Hamiltonian linear time-invariant systems.
- Model order reduction using artificial neural networks.
- Data-driven model order reduction with Dynamic Mode Decomposition, Loewner matrices, Eigensystem Realization, and parametric AAA.
- Gram-Schmidt algorithm supporting re-orthogonalization for improved numerical accuracy.
- Time-stepping and Newton algorithms, as well as generic iterative linear solvers.
- Low-rank alternating direction implicit (LR ADI) method for large-scale Lyapunov equations and bindings for matric equations solvers in SLICOT (via slycot) and Py-M.E.S.S.
- Eigenvalue/pole computation using the implicitly restarted Arnoldi method and the subspace accelerated dominant pole (SAMDP) algorithm.
- Modal truncation for linear time-invariant systems.
- Time-dependent parameters.
All these algorithms are formulated in terms of abstract interfaces for seamless integration with external high-dimensional PDE solvers. Bindings for the following PDE solver libraries are available:
Pure Python implementations of discretizations using the NumPy/SciPy scientific computing stack are implemented to provide an easy-to-use sandbox for experimentation with new model reduction approaches. pyMOR offers:
- Structured 1D and 2D grids, as well as an experimental Gmsh-based grid, implementing the same abstract grid interface.
- Finite element and finite volume operators based on this interface.
- SciPy/SuperLU based iterative and direct solvers for sparse systems.
- OpenGL, matplotlib, and k3d based visualizations of solutions.
References
- M. Ohlberger, S. Rave, S. Schmidt, S. Zhang. "A Model Reduction Framework for Efficient Simulation of Li-Ion Batteries". Springer Proceedings in Mathematics & Statistics Vol. 78: Finite Volumes for Complex Applications VII-Elliptic, Parabolic and Hyperbolic Problems, Berlin, June 2014.
- R. Milk, S. Rave, F. Schindler. "pyMOR - Generic Algorithms and Interfaces for Model Order Reduction". SIAM Journal on Scientific Computing 38(5): S194--S216, 2016.
Links
- Official website,
- Development of pyMOR can be tracked on GitHub,
- Online documentation.
Contact
For assistance with, and contributions to pyMOR, the developers can be contacted via GitHub discussions.