Recent changes to linked boundary conditions only in non-optimised init_opt_bc

[David Dickinson]

Need to port "recent" changes to linked boundary condition initialisation from non-optimised form to optimised form. Would then ideally remove non-optimised form (essentially same code, so no more complicated for optimised version).

[David Dickinson]

This difference currently leads to different results in nonlinear simulations depending on the choice for init_opt_bc

[David Dickinson]

The above image shows a comparison of identical collisionless simulations expect for the choice of opt_init_bc

[David Dickinson]

On closer inspection this difference is very surprising, as far as I can tell there are no differences between the values calculated by the optimised and non-optimised versions for any of the arrays calculated by these routines.

[David Dickinson]

It seems that the timesteps follow a slightly different trajectory in the early phase and that this may then lead to a somewhat different late time behaviour. It's still not clear why the timestep trajectory is different -- floating point order may potentially be slightly different between identical repeat runs but I don't think we'd usually expect such substantial variation between repeat runs.

[David Dickinson]

See issue #69 for a generalised continuation of this topic.