New algorithm slashes time to run most sophisticated climate models by 10-fold

[Admin]

Climate models can be a million lines of code long and can take months to run on supercomputers. A new algorithm has dramatically shortened that time.

New algorithm slashes time to run most sophisticated climate models by 10-fold : Read more

 

[Unclear Engineer]

Not really sure what this is, based only on this article.

But, it seems like it is using results of previous time steps to extrapolate the first guess at the next step, which then gets refined.

If so, I am wondering if there are any pitfalls in the approach that result from using it in what seems like it might have (mathematically) "chaotic" behaviors - in which small differences in initial conditions lead to different groupings of final results with substantially different parameter values between those groups.

We know that weather models can exhibit mathematically "chaotic" behavior. And we don't yet understand the transition processes from periods of glaciation to periods of warming and back again, which may involve some chaotic aspects.

So, I am wondering if this algorithm might suppress some chaotic behaviors and cause us to miss some potential sets of results.

 

[billslugg]

The premise is there is only one final state that it will settle on, no matter how you get there. But you are right, this may be a bad premise.