hypergeo: The hypergeometric function
In hypergeo: The Gauss Hypergeometric Function

Description Usage Arguments Details Note Author(s) References See Also Examples

The Hypergeometric and generalized hypergeometric functions as defined by Abramowitz and Stegun. Function hypergeo() is the user interface to the majority of the package functionality; it dispatches to one of a number of subsidiary functions.

1	hypergeo(A, B, C, z, tol = 0, maxiter=2000)

`A,B,C`	Parameters for `hypergeo()`
`z`	Primary argument, complex
`tol`	absolute tolerance; default value of zero means to continue iterating until the result does not change to machine precision; strictly positive values give less accuracy but faster evaluation
`maxiter`	Integer specifying maximum number of iterations

The hypergeometric function as defined by Abramowitz and Stegun, equation 15.1.1, page 556 is

[omitted; see PDF]

where (a)_n=Gamma(a+n)/Gamma(a) is the Pochammer symbol (6.1.22, page 256).

Function hypergeo() is the front-end for a rather unwieldy set of back-end functions which are called when the parameters A, B, C take certain values.

The general case (that is, when the parameters do not fall into a “special” category), is handled by hypergeo_general(). This applies whichever of the transformations given on page 559 gives the smallest modulus for the argument z.

Sometimes hypergeo_general() and all the transformations on page 559 fail to converge, in which case hypergeo() uses the continued fraction expansion hypergeo_contfrac().

If this fails, the function uses integration via f15.3.1().

Abramowitz and Stegun state:

“The radius of convergence of the Gauss hypergeometric series ... is |z|=1” (AMS-55, section 15.1, page 556).

This reference book gives the correct radius of convergence; use the ratio test to verify it. Thus if |z|>1, the hypergeometric series will diverge and function genhypergeo() will fail to converge.

However, the hypergeometric function is defined over the whole of the complex plane, so analytic continuation may be used if appropriate cut lines are used. A cut line must join z=1 to (complex) infinity; it is conventional for it to follow the real axis in a positive direction from z=1 but other choices are possible.

Note that in using the package one sometimes draws a “full precision not achieved” warning from gamma(); and complex arguments are not allowed. I would suggest either ignoring the warning (the error of gamma() is unlikely to be large) or to use one of the bespoke functions such as f15.3.4() and tolerate the slower convergence, although this is not always possible.

Robin K. S. Hankin

Abramowitz and Stegun 1955. Handbook of mathematical functions with formulas, graphs and mathematical tables (AMS-55). National Bureau of Standards

hypergeo_powerseries, hypergeo_contfrac, genhypergeo

#  equation 15.1.3, page 556:
f1 <- function(x){-log(1-x)/x}
f2 <- function(x){hypergeo(1,1,2,x)}
f3 <- function(x){hypergeo(1,1,2,x,tol=1e-10)}
x <- seq(from = -0.6,to=0.6,len=14)
f1(x)-f2(x)
f1(x)-f3(x)  # Note tighter tolerance

# equation 15.1.7, p556:
g1 <- function(x){log(x + sqrt(1+x^2))/x}
g2 <- function(x){hypergeo(1/2,1/2,3/2,-x^2)}
g1(x)-g2(x)  # should be small 
abs(g1(x+0.1i) - g2(x+0.1i))  # should have small modulus.

# Just a random call, verified by Maple [ Hypergeom([],[1.22],0.9087) ]:
genhypergeo(NULL,1.22,0.9087)


# Little test of vectorization (warning: inefficient):
hypergeo(A=1.2+matrix(1:10,2,5)/10, B=1.4, C=1.665, z=1+2i)


# following calls test for former bugs:
hypergeo(1,2.1,4.1,1+0.1i)
hypergeo(1.1,5,2.1,1+0.1i)
hypergeo(1.9, 2.9, 1.9+2.9+4,1+0.99i) # c=a+b+4; hypergeo_cover1()