[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: co2calc.f and ta_iter_1.f




Jim,
that is not true
It depends, of course on the amount of chemical changes
since the previous computation.
two or three iterations are suffient for roundoff error
on a 64-bit machine. Practically, one iteration would 
be sufficient, indeed.

some reminders for optimization:
(trivial) call of subroutines should be done with arrays
not with single variables for vectorization

the choice of phhi and phlo is rather irrelevant
once you are at 10**(-8) the iteration will hardly touch
the bounds. For crazy starting values it is sufficient
to ensure that htotal does not come below , say, 10 %
of the previous guess
It may be more efficient to do 3 iterations for the whole array
than than to perform sophisticated checking for avoiding redundant
algebra


I made the experience that even advenced programmers are not aware
of the fact that sqrt, log, exp sin cos are extremely time consuming.
a**b is even worse
the chemical constants should not have higher temporal resolution
then other fields. I strongly recommend to provide monthly fields
and interpolate for the actual timestep.
Ernst