Error-surfce: 20 iterations in unstable case

[Patrick LEE]

Originally Posted by mchandler

If we ran with a shorter time step the model would probably be more stable, but it would also slow down a lot....always trade-offs unfortunately.

This is not a stability problem. The reason the iteration fails to converge in the UNSTABLE case is that the eqaution actually becomes meaningless (as some of quantities become NANs). The issue here is NOT a numerical convergence problem. Usually, this iteration procedure to solve this equation converges in less than 5 iterations (in Model II).

However, the porblem seems to couple to other problems. Consider these:
(1) I have solved one of the non-convergence problem by instructing the compiler to compile Model II so that Model II becomes LESS accurate (as I have mentioned in another thread in this forum).

(2) I proved that there is a buffer overflow bug in Model II last year (as I said in another thread in this forum). This proof starts by tracking down the sources of the NANs that lead to this non-convergence problem.

Patrick.