why unconditionally stable FDTD becomes unstable when the pulse turns round?

[dovea]

Hello,

I am working on an unconditionally stable FDTD scheme (like ADI-FDTD).
When dt is above the CFL bound, what I noticed is, the scheme works fine for certain time steps -- until the gaussian incident pulse makes the turning (at the lowest or uppermost peak). Then after this turning FDTD becomes unstable. The scheme works fine when dt is below the CFL limit.

Can anyone explain this? What can I do to make the FDTD work stably even after the pulse makes the turning? Any advice is highly appreciated.

 

[Kohi_boy]

Some figures might be helpful in explaining the situation.

Did you check your code again?