Reducing Step Size With Non-Default Solvers
Posted: Sat May 04, 2013 6:10 pmHi -
I am simulating a deterministic model and have been having trouble with badly scaled matrices and diverging simulations for large shocks. I determined that the problem is the step size I am using is too large because a key function has an asymptote and when it takes too large of a step things go haywire (reducing slowc and increasing maxit_ helps a lot).
I also experimented with putting my model into bytecode and using different stack_solve_algo values and found that solvers 2 (Newton with Generalized Minimal Residual) and 3 (Newton with Stabilized Bi-Conjugate Gradient) work much better than the standard solver. However, with these solvers with large enough shocks things still diverge. I have tried to use options slowc and maxit_ for these solvers but when I put in these options in things won't work, even for small shocks. Does anyone know how to adjust the stepsize with these solvers? Any other trouble shooting suggestions?
Thanks so much.
Adam
I am simulating a deterministic model and have been having trouble with badly scaled matrices and diverging simulations for large shocks. I determined that the problem is the step size I am using is too large because a key function has an asymptote and when it takes too large of a step things go haywire (reducing slowc and increasing maxit_ helps a lot).
I also experimented with putting my model into bytecode and using different stack_solve_algo values and found that solvers 2 (Newton with Generalized Minimal Residual) and 3 (Newton with Stabilized Bi-Conjugate Gradient) work much better than the standard solver. However, with these solvers with large enough shocks things still diverge. I have tried to use options slowc and maxit_ for these solvers but when I put in these options in things won't work, even for small shocks. Does anyone know how to adjust the stepsize with these solvers? Any other trouble shooting suggestions?
Thanks so much.
Adam