Code for my, TMA4900 Industrial Mathematics, Master’s Thesis
Based on Specialization-Project-fall-2021, LICENCE copy.
Project depends on Matrix LSQ
Please see Solvers for more documentation.
Please see Default constants for documentation.
Please see Helpers for documentation.
In POD
eigh - scipy.linalg.eigh is not compliantly stable and it can also be quite slow fractional_matrix_power - scipy.linalg.fractional_matrix_power is really slow (is in the else). At least slower than eigh, sparsity of a_mean is lost in input where a_mean.A is called giving the np.array and the unction can use much RAM if a_mean is large ~ 10_000 x 10_000. Testing if case against each other on case with n_free = 12_960 and ns = 15_625
- gives 3:44 in eigh for corr_mat (times in mm:ss)
- gives 20:02 in fractional_matrix_power and 2:09 in eigh for k_mat if ns <= n_free - is not necessary because corr_mat and k_mat have the same eigenvalues, but it gives the smallest matrix between corr_mat and k_mat
In _sym_mls_params_setup of QuadrilateralSolver
The construction and thereby the evaluation of the Legendre Polynomials is not optimal, however, rewriting this is out of scoop for the current thesis.