On Jun 13, 2006, at 6:06 AM, Jose M. Llorens wrote:
> You are right, the term "noisy" is confusing. I have attached the
> results
> I get after running the program.
I see nothing at all surprising here. Perhaps I overlooked it. Could
you please elaborate? While I didn't check the values, it looks to me
like the single and double precision versions track quite closely
until the result gets down to around 10**(-42). Soon thereafter, the
single precision version underflows to zero, while the double
precision version continues decreasing until it gets to around 10**
(-321), at which point it underflows to zero. Both sets of results
look "smooth", at least based on my eyeball examination of the
columns of numbers.
Is there a chance that you are unaware that single and double
precision may (and do with IEEE formats) differ in dynamic range as
well as in precision? The results here show that range difference.
Single can't represent numbers smaller than about 10**(-43), while
double goes down to about 10**(-322). The exact boundaries depend on
whether underflow is gradual or abrupt, but that's a detail here.
If that is what is bothering you, that is just a property of floating
point arithmetic, with specific application to the IEEE formats used
in most current machines. In that case, this has essentially nothing
to do with Fortran per se.
If that's not it, you'll need to elaborate more explicitly, because
my eyes aren't seeing anything else worthy of note in any regard,
much less "noisy".
--
Richard Maine | Good judgment comes from experience;
[log in to unmask] | experience comes from bad judgment.
| -- Mark Twain
|