Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

motion accuracy



p1788
although Bill Walster's statement "Functions have to be valid, everything else is a matter of QoI" is appealing , I think that p1788 shall not be worse than 754 concerning accuracy.
Therefore I submit te following motion:
Table 1 in chapter 2 lists the forward mode elementary functions.
1. move the square function sqr(x) behind  the function recip
2. Now the table consists out of 6 parts mhich here will be called
	- the basic operations
	- the case fct
	- power fcts
	- trig/hyp fcts
	- integer fcts
	- absmax fcts
3. the basic operations are provided as internal type fcts and as mixed type formatOf fcts 4. the basic operations, the case fct, integer fcts as well as absmax fcts shall be "tightest" 5. the other functions of table 1, the recommended fcts of tables 6 and table 7 as well as all reverse mode fcts shall be "valid" and should be "accurate"

Jürgen

--
o Prof. Dr. Juergen Wolff von Gudenberg, Lehrstuhl fuer Informatik II
    / \          Universitaet Wuerzburg, Am Hubland, D-97074 Wuerzburg
InfoII o         Tel.: +49 931 / 31 86602
  / \  Uni       E-Mail: wolff@xxxxxxxxxxxxxxxxxxxxxxxxxxx
 o   o Wuerzburg