Explanation of machine-dependent constants
beta = Radix for the floating-point system
maxexp = Smallest power of beta that overflows
XSMALL = Positive argument such that 1.0 - X = 1.0 to
machine precision for all ABS(X) .LE. XSMALL.
XINF = Largest positive machine number; approximately
betamaxexp
XMAX = Largest argument acceptable to BESI0; Solution to
equation:
W(X) * (1+1/(8X)+9/(128X2) = betamaxexp
where W(X) = EXP(X)/SQRT(2PIX)
Approximate values for some important machines are: beta maxexp XSMALL
CRAY-1 (S.P.) 2 8191 3.55E-15 Cyber 180/855 under NOS (S.P.) 2 1070 3.55E-15 IEEE (IBM/XT, SUN, etc.) (S.P.) 2 128 2.98E-8 IEEE (IBM/XT, SUN, etc.) (D.P.) 2 1024 5.55D-17 IBM 3033 (D.P.) 16 63 6.95D-18 VAX (S.P.) 2 127 2.98E-8 VAX D-Format (D.P.) 2 127 6.95D-18 VAX G-Format (D.P.) 2 1023 5.55D-17
XINF XMAX
CRAY-1 (S.P.) 5.45E+2465 5682.810 Cyber 180/855 under NOS (S.P.) 1.26E+322 745.893 IEEE (IBM/XT, SUN, etc.) (S.P.) 3.40E+38 91.900 IEEE (IBM/XT, SUN, etc.) (D.P.) 1.79D+308 713.986 IBM 3033 (D.P.) 7.23D+75 178.182 VAX (S.P.) 1.70D+38 91.203 VAX D-Format (D.P.) 1.70D+38 91.203 VAX G-Format (D.P.) 8.98D+307 713.293
Error returns
The program returns XINF for BESI0 for ABS(ARG) .GT. XMAX.
Intrinsic functions required are:
ABS, SQRT, EXP
Authors: W. J. Cody and L. Stoltz Mathematics and Computer Science Division Argonne National Laboratory Argonne, IL 60439
Latest modification: June 7, 1988
Type | Intent | Optional | Attributes | Name | ||
---|---|---|---|---|---|---|
double precision | :: | ARG | ||||
double precision | :: | RESULT | ||||
integer | :: | JINT |