Matrix inverse

[RRiva]

Hi @gbarter, thanks for this nice code! By skimming through it I noticed that you have computed a couple of matrix inverses. Generally, this has bad numerical properties, as it's both slow and inaccurate. You should instead solve a linear system.

I suggest to:

• At raft_model L399 replace np.matmul(np.linalg.inv(M_tot), C_tot) with np.linalg.solve(M_tot, C_tot).
• At raft_model L540 replace np.matmul(np.linalg.inv(Z[:,:,ii]), F_tot[:,ii] ) with np.linalg.solve(Z[:,:,ii], F_tot[:,ii]).

Furthermore, if M_tot and Z are reasonably sparse, you might get an even faster solution using spsolve.

By the way, the routines from scipy.linalg might be faster than the ones from numpy on some architectures.

[mattEhall]

Hi @RRiva, thanks for this tip. I included your recommended switch to np.linalg.solve in this recent commit: https://github.com/WISDEM/RAFT/commit/bdc1b8b7a2f3f1754dc2e9429d6149bec8132376