"Our greatest responsibility is to be good ancestors."

-Jonas Salk

Wednesday, July 26, 2017

A grumble

A random grumble about coupled models found during the cleanup. This may provide some insight into one of the key failure points of my checkered career as a scientist.

==

It's the clocking when you couple systems that makes for messes like these. Even if co-designed, the component models necessarily have different time steps.

Not only is it a kluge to make the ice the main(), as far as I understand nobody has investigated the dynamic implications of various coupling strategies. 

I am very confident that somebody smarter than me can figure out a predictor-corrector (a sort of huge Runge-Kutta step) scheme for the coupler strategy. Whether it is worth the bother is unclear to me. Perhaps the extant coupling is good enough. 

But to be honest, if anybody has done any serious math defending the proposal that ANY of the various ad hoc coupling strategies of climate models (or perhaps of multiphysics models in general) does no significant damage to the coupled system dynamics, I haven't seen it. Maybe it's obvious to some. Or maybe it's hard, but someone has worked through it. Or maybe it's just a crude hack that needs a major rethinking.


I really wanted to approach this experimentally, as my math isn't subtle enough to answer the question to my own satisfaction. But I put too much faith in a piece of DoE middleware, not for the first time, and fell flat on my face.

No comments: