-
-
Notifications
You must be signed in to change notification settings - Fork 65
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Upgrade implicit solvers to the latest nlsolve #99
Conversation
Looks like this is very far along, just some test setup issues now? ISSEM is a 1/2 order algorithm. ImplicitEM is a 1/2 order algorithm. ImplicitRKMil is an order 1 algorithm. So the tests have too high of convergence coefficients, which seems to be the failure. |
Actually on additive noise those should all be 1.0 with possible super convergence. Since that was all in the same file and only on the additive test, I think this is pointing to a possible issue with how the time values are set there. |
src/perform_step/sdirk.jl
Outdated
J, W = calc_W!(integrator, cache, γdt, repeat_step) | ||
if nlsolve! isa NLNewton | ||
J, nlcache.W = calc_W!(integrator, cache, γdt, repeat_step) | ||
uf.t = t |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this should go before?
src/perform_step/kencarp.jl
Outdated
J, W = calc_W!(integrator, cache, γdt, repeat_step) | ||
if nlsolve! isa NLNewton | ||
J, nlcache.W = calc_W!(integrator, cache, γdt, repeat_step) | ||
uf.t = t |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
this should go before?
tol = cache.tol | ||
if nlsolve! isa NLNewton | ||
J, nlcache.W = calc_W!(integrator, cache, dt*theta, repeat_step) | ||
uf.t = t |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think the current time needs to be set before the Jacobian is calculated in order to make sure it's at the correct time point.
Upping the number of Monte Carlo runs got rid of the superconvergence and made the results match the theoretical results perfectly, so I think those are good now. Fingers crossed tests pass. |
Something seems wrong in the split tests |
No description provided.