# [R] increase thr range in R

Spencer Graves spencer.graves at pdf.com
Tue Dec 14 17:56:21 CET 2004

```Sebastian:

I routinely deal with situations like that in one of two ways:

(1) Can you work with log(besselI)?  If yes, that should solve the
problem.

(2) What do you do with the numbers returned from besselI?  I
assume they are later used to compute numbers in a more sensible range
by division or differencing.  There are many asymptotic approximations,
etc., that should work quite well to simplify the computations in those
extreme cases.

hope this helps.  spencer graves

Earl F. Glynn wrote:

>IEEE floating point provides a "double-extended" type often called a long
>double.  The exact size of a double-extended can vary but the minimum
>specified by the spec is 80 bits.  A PC with the IEEE extended type gives an
>approximate range from 3.4E-4932 to 1.1E4932, which would give you the range
>you want.  Other architectures, such as a Tru64 alpha, give an even larger
>extended.
>
>Reference:  See the "IEEE Standard" in
>"What Every Computer Scientist Should Know About Floating-Point Arithmetic"
>http://docs.sun.com/source/806-3568/ncg_goldberg.html
>
>The "note" under R's ?numeric help says "All real numbers are stored in
>double precision format," so R apparently does not yet support the extended
>type.
>
>str(.Machine) in R gives a sizeof.longdouble, but I don't know how to use
>such a longdouble in native R -- perhaps someone else could enlighten us.  C
>or C++ (or other languges) would support the long double type and the
>computations you'd like to do.
>
>efg
>Earl F. Glynn
>Stowers Institute for Medical Research
>
>=========
>"Sebastian Kaiser" <KINGSEBI at gmx.net> wrote in message
>news:26416.1103030099 at www35.gmx.net...
>Hello Everybody in order to get some needed results out of my function i
>need to get my besselI function evaluated at some values which normally gave
>Inf or 0 (expon.scaled NAN) back. So I would like to increase the range in R
>from approxamittly 1e+320 to aabout 1e+500 or something like that. Is there
>any possibility or pacckage to do this easily?
>Thank You
>Sebastian Kaiser
>Institut for Statistics in Munich Germany
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://stat.ethz.ch/mailman/listinfo/r-help