It doesn't seem to exist a good option for the linear system solver initial guess that improves the algorithm's performance. Options tried as initial guess: using previous value yn or make one euler explicit step (no extra cost as function evaluation has to be done anyway) to obtain another initial guess.
Currently, as it hasn't been tested enough, the code uses the gmres solver default: a vector of zeros.
Functions to modify:
_getAdditionalSolverParameters
_euler_semiimplicit
_midpoint_semiimplicit
Code TODO comment (remove once issue is closed):
TODO: The option of doing an explicit step doesn't seem to be effective (maybe because with
# extrapolation we are taking too big steps for the explicit solver to be close to the solution).
# Doing the _euler_explicit doesn't cost extra function evaluations
# This idea came from the lsode solver.
It doesn't seem to exist a good option for the linear system solver initial guess that improves the algorithm's performance. Options tried as initial guess: using previous value yn or make one euler explicit step (no extra cost as function evaluation has to be done anyway) to obtain another initial guess.
Currently, as it hasn't been tested enough, the code uses the gmres solver default: a vector of zeros.
Functions to modify: _getAdditionalSolverParameters _euler_semiimplicit _midpoint_semiimplicit
Code TODO comment (remove once issue is closed):
TODO: The option of doing an explicit step doesn't seem to be effective (maybe because with