summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorzimmerma <zimmerma@280ebfd0-de03-0410-8827-d642c229c3f4>2007-10-15 14:11:22 +0000
committerzimmerma <zimmerma@280ebfd0-de03-0410-8827-d642c229c3f4>2007-10-15 14:11:22 +0000
commitae15e86d54d3c912578b238f91e2ca377fcb1413 (patch)
tree720523c52266741494e29fc42d8f12e234e49ce2
parent7583bbab6d952e34100ed5208faf9394facffc8b (diff)
downloadmpfr-ae15e86d54d3c912578b238f91e2ca377fcb1413.tar.gz
added item in Efficiency section
git-svn-id: svn://scm.gforge.inria.fr/svn/mpfr/trunk@4895 280ebfd0-de03-0410-8827-d642c229c3f4
-rw-r--r--TODO6
1 files changed, 6 insertions, 0 deletions
diff --git a/TODO b/TODO
index 29565992d..e4f248bee 100644
--- a/TODO
+++ b/TODO
@@ -292,6 +292,12 @@ Table of contents:
possible through stripping low zero bits or limbs could check for
that (this would be less efficient but easier).
+- try the idea of the paper "Reduced Cancellation in the Evaluation of Entire
+ Functions and Applications to the Error Function" by W. Gawronski, J. Mueller
+ and M. Reinhard, to be published in SIAM Journal on Numerical Analysis: to
+ avoid cancellation in say erfc(x) for x large, they compute the Taylor
+ expansion of erfc(x)*exp(x^2/2) instead (which has less cancellation),
+ and then divide by exp(x^2/2) (which is simpler to compute).
##############################################################################
6. Miscellaneous