.C(zbiry, ..) correctly now, thanks to Alexios Galanos.
Using cospi(nu) instead of cos(pi*nu), for
negative nu, both in BesselJ() and BesselY() may
increase accuracy.
Rd-crossrefs: replace \linkS4class{mpfr} by
\link[Rmpfr:mpfr-class]{mpfr}.
The exponential rescaling for the asymptotic version of the
I() function has been changed (for non-real
complex arguments) to be compatible to the scaling for
BesselI().
deprecate bI() to use besselIs() instead –
enable log = TRUE option and test it.
new besselJs(), i.e., Taylor series for J_n(x);
using lssum() for alternating exp/log sum, also works with
Rmpfr's mpfr() high precision numbers.
use C routine registration, adding safety and a bit of speed.
improved documentation for Bessel functions Ai,Bi, H, I,J,K,Y.
besselI.nuAsym() and besselK.nuAsym() now allow maximal
k.max = 5 (instead of 4).
new besselKasym() (9.7.2 of A.&S. p.378).
All the TOMS 644 routines get an optional argument
verbose indicating the verbosity level of output notably from
C code. The output itself may change and is not part of the API.
fix BesselI(<cmplx>, nu < 0, expo..=TRUE) bug thanks
to Hiroyuki Kawakatsu.
better warning message (r29).
the TOMS 644 routines Bessel[HIYJK] and Airy[AB] now all
return NaN with a warning instead of stopping with an error
in case of low level error ierr == 4.
BesselK(0,*) is zero now, and BesselH(m,
z=0,..) returns Inf/NaN combinations.
Both no longer end with an error (from ierr == 1).
BesselK(<negative>, 3) and BesselJ(*) now also
return the correct imaginary part instead of just a numeric value.
besselIs(z, nu) now also works for complex z.
Adapt to new standard: move vignettes to own ‘vignettes/’ directory.
Improved vignette
BesselY(0, nu) should not give an error message
from zbesy(), but either -Inf (real 0) or something like
"Inf + NaN i".
Cosmetic, mainly for R CMD check
besselI.nuAsym() and besselK.nuAsym()
asymptotic versions for large nu and/or x
See the ‘ChangeLog’ file in the sources.