You have it right.
In cases like this does the smaller one being zero vs non-zero matter?
If it is carried along the effect is not a reduction of accuracy.
The problem (if any) may bethe way that numbers are stored. Floats have
basically 63bits including a sign bit and 11 bit exponent bits. The
examples are well within these limits. Complex numbers, if I remember
correctly involve 2 words plus some linkage which may involve saving a
bit for indicating "real/imaginary " properties.
Don Kelly
But I don’t think anyone would want this
to happen in every calculation, in any
algorithm. It’s okay to have a function
you call every time you might actually
want to consider the smaller one 0 but
not with every intermediate result.
From my point of view, there are use cases,
and they can be dealt with by users.
This one isn’t even worth a new primitive
in my opinion. I do think there are better
ways to bloat the language.
Am 19.02.21 um 21:50 schrieb Raul Miller:
So we
have to deal with epsilon issues.
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
