Re: lojban as a programming language [was Re: [lojban] Lojban for lay programmers]

[Jim Carter <jimc@MATH.UCLA.EDU>]

On Thu, 24 Jan 2002, John Cowan wrote:
> The additive inverse of NaN is NaN, I think, depending on your
> definition of inverse:  NaN + NaN is not 0, but 0 - NaN is definitely
> NaN.

On this sidetrack we're looking for X such that X + NaN == 0, and if I
remember the standard right, anything + NaN == NaN, so NaN lacks an
additive inverse.

> As for not being a group, if they weren't a group over the defined
> IEEE operations, that would mean that something not an IEEE-float
> was being delivered, which is self-contradictory, since every
> bit combination has an IEEE meaning.

In 64 bit IEEE float, let X = 1; Y = 1e-17 (and the LSB is about 2e-16).
If IEEE floats were closed under addition, the "first law" for groups, then
X+Y has to have a IEEE representation.  Let Z = X+Y.  But Z == 1; the sum
is truncated on the right to 56 bits.  So we've just proved that Y == 0,
because 0 is the group member (provably unique) such that W + 0 == W for
every W in the group.  Oops.

Unsigned integer arithmetic is truly in the ring of 2^N, but all the other
data types (signed integer, 32 and 64 bit float) do not fit the neat
type categories (groups, rings and fields) where useful theorems can be
proved.  Nonetheless floating point is useful, because it approximates real
numbers, and within reasonable limits the computer's answer is similar to
the true answer if authentic real or rational numbers had been used.

James F. Carter          Voice 310 825 2897    FAX 310 206 6673
UCLA-Mathnet;  6115 MSA; 405 Hilgard Ave.; Los Angeles, CA, USA  90095-1555
Email: jimc@math.ucla.edu    http://www.math.ucla.edu/~jimc (q.v. for PGP key)