Improving the performance of Matrix Exponentials in Python
This project exists purely out of necessity. The 'truth benchmark' I will use for all testing here will be the SciPy implementation of matrix exponentials (scipy.linalg.expm).
I would like to point out that the fastest implementation I have found in preparation for this project was from rngantner on GitHub.
The first step in trying to make performant code is to set strict bounds on the problem you are willing to solve. So, the restrictions I will put on my implentation of expm will be:
- The code must be callable from Python v3.5 to v3.7.
- The performance testing will be done on Python v3.6.
- The matrix will be real-valued.
- The performance will only be tested against 2D arrays.
- The performance will not be tested on matricies of size < 10x10.
- The performance will not be tested on matricies of size > 100x100.