R devel - Oct 2000 - bug in pnorm (PR#699)

Full_Name: James Michael Rath
Version: all (I think)
OS: doesn't matter
Submission from: (NULL) (129.116.226.162)


The code for pnorm in R was adapted from a Fortran library published in the ACM
TOMS journal.  The published version had a typographical error, though, which
was
pointed out in a second article published three years after the original.

The error was that a macro/variable named SIXTEN had a value of 1.6e0 instead of
the
correct 1.6e1 (that is, 16 not 1.6).  The author of the code for R (Ross
Ithaka?)
must have worked from an outdated version of SPECFUN and not the corrected one
that
now appears in CALGO, STATLIB, et cetera.

There is a macro in pnorm.c (in the nmath directory) which looks like:
#define SIXTEN	1.6	/* Magic Cutoff */
It should be something like:
#define SIXTEN	16.0	/* Magic Cutoff */

I know this appears in R version 1.1.1.  I know it also appears in older
versions.
I do not know if it appears in newer ones.


The relevant references from ACM TOMS are:

Cody, W. D.  "Algorithm 715: SPECFUN - A portable FORTRAN package of
special
  function routines and test drivers," in ACM TOMS 19, 22-32, 1993.

Price, David T.  "Remark on algorithm 715," in ACM TOMS 22(2), 258,
1996.


Jim


-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

ratjamm@alum.mit.edu writes:
> There is a macro in pnorm.c (in the nmath directory) which looks like:
> #define SIXTEN	1.6	/* Magic Cutoff */
> It should be something like:
> #define SIXTEN	16.0	/* Magic Cutoff */
> 
> I know this appears in R version 1.1.1.  I know it also appears in older
> versions.
> I do not know if it appears in newer ones.
Hmm. I tried making the appropriate change and plotting 

plot(x,(pnorm(x+.00001)-pnorm(x-.00001))/dnorm(x))

for 

x<-seq(0,-16,-.01)

and

x<-seq(0,16,.01)

I'll be darned if I can spot the difference. Where would you expect to
find it?

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._