complex infinity

[Samuel A. Figueroa]

	Date:     Tue, 7 Jan 92 15:05:13 EST
	From: Doug Gwyn (VLD/VMB) <uunet!BRL.MIL!gwynauunet.UU.NET>
	To: Dik.Winteracwi.nl
	Cc: ncegatimbuk.cray.com, dghavalidgh.com
	Subject:  Re:  complex infinity
	Message-Id:  <9201071505.aa13527aVGR.BRL.MIL>

	Unlike the artificial inclusion of +Inf and -Inf in IEEE f.p. real
	number arithmetic, the complex point at infinity has genuinely nice
	properties.  IEEE f.p. was wrong to take the tack it did when it
	introduced so much complexity as well as incompatibility with proper
	mathematical usage.

I am not sure I would go so far as to say that "IEEE f.p. was wrong" with
regard to signed infinities.  A while back I implemented an algorithm to
compute the eigenvalues of symmetric tridiagonal matrices.  One version ran
40% faster on a Sun SPARC workstation, while another version ran 15% faster
and was provably more accurate.  The speedup and increased accuracy was
solely due to the availability of signed infinities and zeros.