[Fwd: Suggestion for a minor accuracy improvement to glp_exact()]

-------- Forwarded Message --------
From: Brendan McKay <[email protected]>
To: [email protected]
Subject: Suggestion for a minor accuracy improvement to glp_exact()
Date: Mon, 5 Jun 2017 16:59:20 +1000

Version 4.61.

I know there is already movement towards getting exact values
out of glp_exact().  Meanwhile, there is an easy way to improve
the floating-point value for the objective function.  At the moment
each term is separately computed in floating-point and then added
in floating-point.  But at that point in the code the exact values of
the variables and coefficients are still available and it is easy to
compute the objective function exactly and convert it to floating-point
only at the end.  For an objective function with multiple nonzero
terms this can make a significant difference.

Brendan.