SIAM J. MATH. ANAL.

(C) 1987 ’Society for Industrial and Applied Mathematics

Vol. 18, No. 5, September 1987 OO2

THE BEHAVIOR AT UNIT ARGUMENT OF THE

Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

HYPERGEOMETRIC FUNCTION 3F2 :t:

WOLFGANG BOHRING’
Abstract. The behavior at z of the generalized hypergeometric function 3F2(a, b, c’, e,f’, z) is investi-
gated. First the analytic continuation near z is obtained for the general case when s e +f- a b c is
not equal to an integer. The corresponding continuation formulas for the special cases when s is equal to
an integer are then derived by appropriate limiting processes. When f c or e c, the formulas immediately
reduce to the well-known continuation formulas of the Gaussian hypergeometric function.

Key words, special functions, hypergeometric series, hypergeometric functions, continuation formulas,
hypergeometric differential equations

AMS(MOS) subject classifications. 33A30, 34A20, 34A30, 30B40

1. Introduction. The behavior near z 1 of the Gaussian hypergeometric function

or series
(a),,(b),,
(1.1) 2Fl(a, b; e; z)= z"
=o (e),,n!

is given by a well-known continuation formula which may be written

r(a)r(b) r(a)r(b)r(s)
2Fl(a, b; e; z)= 2Fl(a, b; 1- s; 1- z)
r(e) r’(a+s)r(b+s)
(1.2) +F(-s)(1-z)2F(a+s,b+s; l+s; l-z)
(larg (1 z) < r),
(1.3) s=e-a-b.
As it stands, (1.2) is valid if s is not equal to an integer, otherwise it is the starting
point from which the relevant formulas may be derived by a limiting process. It is the
aim of the present work to obtain a corresponding formula for the generalized hyper-
geometric function 3FE(a, b, c; e, f; z), including the formulas for the exceptional cases
when
(1.4) s e +f- a b c
is equal to an integer.
Like the EFt, the 3F2 is also a particular solution of a certain linear differential
equation, now of the third order, with three regular singular points at z -0, 1, o. The
characteristic exponents 0, 1, s and the local power series solutions of this differential
equation relative to the point z 1 are known from the detailed investigation by
Netrlund [8]. The analytic structure near z 1 of the general solution is therefore
known and, if s is not equal to an integer, is given by
(1.5) A + B(1 z)+ O({1 z}2)+ C(1 z)S{1 + O(1 z)}.
It is more difficult to obtain the connecting constants A, B, C for the function 3F2(a,
b, c; e, f; z). While the value of C (and of K in (1.6) below) can be inferred from

* Received b y the editors October 21,1985; accepted for publication (in revised form) September 17,1986.

" Physikalisches Institut, Universifiit Heidelberg, D-6900 Heidelberg, West Germany.

1227
 1228 WOLFGANG BOHRING

[8], A and B are not known in general. This matter has recently received attention in
the special case when s 0. Then, as z-+ 1 in the sector larg (1- z) < r, we have
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(1.6)
r(a)r(b)r(c)
(a,b,c z)
r(e)r(f) 3F2 e,f
L+M(1-z)+O({1-z}2)+K In (1-z){l+O(1-z)}
where
(1.7)
and, if Re (c) > 0,
(e-c).(f-c),.
(1.8) L=2q(1)-O(a)-b(b)+ Y
.,=1 m(a).,(b).,
Here q,(x) denotes the logarithmic derivative F’(x)/F(x) of the gamma function. The
expression (1.8) for L, contained without proof in Ramanujan’s note books [9], has
recently been proved by Evans and Stanton [3], but, as they state themselves, "because
of the inductive nature" of their proofs, their paper "unfortunately sheds little light
on how Ramanujan might have made this remarkable discovery." A new proof along
different lines is therefore of interest. Also, the third connecting constant M is not
known. Its value will be given in (4.9) below.
In the present paper we first consider the general case when s is not equal to an
integer and determine the connecting constants A, B, C of (1.5) for the function
3F2(a, b, c; e,f; z). The results for the exceptional case when s is equal to zero or,
more generally, equal to any integer are then obtained by a limiting process.
A lemma is provided in 2. The main theorem for the general case is derived in
{} 3. The results for the exceptional cases are supplied in {} 4.
2. A lemma. In order to avoid inconvenient and unnecessary restrictions of
the denominator parameters, we follow Olver [6] and consider
{F(e)F(f)}-laFE(a, b, c; e,f; z) rather than aFt_ itself, but without explicitly introducing
a new symbol for this quantity. As a function of z, it has a finite value at z 1 if
Re(e+f-a-b-c)>O. Following Wimp [11], [12] we may consider this value as a
function of the five parameters. This suggests the following.
DEFINITION 1. For those points of the parameter space which satisfy Re (e +f-
a b c) > 0 let the symbol 3F2(a, b, c; e, f) be defined by

1
(2.1)
F(e)F(f) 3F2 e,f F(e)F(f) 3F2 e,f
and let it then be defined for the other points by analytic continuation.
Using this definition we supply for later application the following.
LEMMA 1. With s e + f a b c there holds

(2.2)
r(e)r(f) 3F= e,f =r(c)r(a+s)r(b+s)
Proof. If Re (s) > 0 and Re (c) > 0 we have from [2] or [5]

(2.3)
1
(a,b, cl) =r(c)r(a+s)r(b+s),F(e-c,f-c,s)
F(e)F(f) 3F2 e,f
1
r(s)
a+s,b+s
 THE HYPERGEOMETRIC FUNCTION 3F2 1229

Here the conditions on s and c are needed to ensure the convergence of the hyper-
geometric series of unit argument on the left- and right-hand side, respectively. By
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

means of Definition 1 and analytic continuation with respect to the parameters we get
rid of these restrictions, which completes the proof.

3. Derivation of the main theorem. The generalized hypergeometric function

3F2(a, b, c; e,f; z) is a particular solution of the third order linear differential equation
(3.1) {D(zD+ e- 1)(zD+ f 1)- (zD+ a)(zD+ b)(zD+ c)}w(z) =0
where D d dz. Relative to the regular singular point z 1 of the differential equation,
the Frobenius ansatz

(3.2) w(z)= 2 g,(r)(1-z) r+",

n-----0

written with powers of 1-z rather than z-1 for later convenience, yields for the
coefficients g, the recurrence relation

(r+ n)(r+ n- 1)(r+ n+ a+ b+ c-e-f)g,(r)

(r+ n- 1){(r+n-2)(2r+2n- 1-e-f+2a+2b+2c)
(3.3)
-ef +ab+bc+ca+a+b+c+ 1}g._l(r)
-(r+ n-2+ a)(r+ n-2+ b)(r+ n-2+ c)g,__(r),
valid for n=0,1,2,...,provided that we have defined g_2(r)=g_l(r)=O while
go(0) 0. The equation for n 0 then leads to the indicial equation which determines
the characteristic exponents

(3.4) r {0, 1, s}, s as in (1.4).

We now assume in this section that s is not equal to an integer, so that the exponent
r s yields a solution of the differential equation. Of the two exponents which are
integers the larger one always gives a solution, but for the smaller one we have to
check if (3.3) is consistent for all the n. Since the difference of the exponents is 1, the
equation for n 1 is crucial. With (1.4) we have from (3.3)

(r+ 1)r(r+ 1-s)gl(r) r{(r- 1)(2r+ 1-e-f+2a+2b+2c)

(3.5)
ef+ ab+ bc+ ca + a + b+ c+ 1}go(r).
For r 0 not only the left-hand side vanishes but the right-hand side happens to vanish
also, and so (3.5) is consistent irrespective of the value of gl(0). Therefore gl(0),
besides go(0), may be assigned arbitrarily. The exponent r=0 in this way yields a
solution containing two constants of integration, where changing gl(0) is equivalent
to adding a constant multiple of the solution with the exponent r 1 mentioned above.
Thus we have shown that

r(a)r(b)r(c)
(3.6)
r(e)r(f) 3F2 e,f
z
=og"(0)(1-z)"+(1-z),=og"(s)(1-z)"
 1230 WOLFGANG BHRING

where the connecting constants go(0), gl(0), go(s) still have to be determined while
the other coefficients g,(r) are then given by the recurrence relation
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

gn(r)
(r-s+ n)(r+ n)(r+ n- 1)
(3.7) ((r+n- 1){(r+n-2)(2r-s+2n- l+a+b+c)
-ef+ ab+ bc+ ca + a + b+ c+ 1}g,,_l(r)
-(r+ n-2+ a)(r+ n-2 + b)(r+ n-2+ c)g,,_2(r)),
to be used for r 0 with n 2, 3, 4, ,
or for r s with n 1, 2, 3, , and g_l(S) 0.
The connecting constant go(s), which multiplies the (at z 1 singular contribution,
,
can most conveniently be determined by Darboux’s method [6] as follows. The left-hand
side of (3.6), when written in the form Y uz has coefficients

F(a+ n)r(b + n)F(c + n)

(3.8) u,
F(e+ n)F(f+ n)F(l+ n)
with the asymptotic behavior

as n- o, which follows by means of the formula

n
r(y +
nX-Y 1+O

The leading singular term on the right-hand side of (3.6) is go(s)(1- z) s, which, when
expanded in powers of z by means of the formula

(3.11) (1 z) S_._
y. (-s). Ztl
,=o n!
and written in the form v,z", has coefficients
go(s) F(-s+ n)
(3.12) v. r(-s) r(l+n)
with the asymptotic behavior

(3.13) v. 1+0
r(_s---5.
as n o. Thus un and v, have, in view of the definition of s according to (1.4), the
same n-dependence of the leading asymptotic term, as expected, and comparison of
the constant factors in (3.9) and (3.13) yields
(3.14) go(s)=r(-s).
This result has been obtained in a different way by Nerlund [8]. More recently such
connection problems have been treated by Naundorf [7] and by Sch5fke and Schmidt
[10].
 THE HYPERGEOMETRIC FUNCTION 3F2 1231

The connecting constants go(0) and gl(0), which appear in the (at z- 1) regular
term, will now be determined. If Re (s)> 0, the singular contribution vanishes and
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(3.6) evaluated at z- 1 yields

(3.15) go(0) =r(a)F(b)F(C)3Fu

r(e)r(f) e,f
1

In a similar way, if Re (s) > 1, the derivative with respect to z of (3.6) gives
r(a+ 1)r(b + 1)F(c + 1) 1, b+ 1, c+ 1
3F2[a+e+l,f+l
(3.16) g,(0)
r(e + 1)r(f+ 1)
By analytic continuation with respect to the parameters, Definition 1 and Lemma 1,
we then finally obtain

(3.17) go(0)
r(a)r(b)r(s)
F(a+s)F(b+s) 3F2 (e-c,f s
-c,
a+s,b+s /’
r(a+ 1)r(b + )r(s- a)
(3.18) g,(O)
r(a+s)r(b+s) 3F2
e-c,f
a+s, b+s -c,s-1)
All the other coefficients g,(r) appearing in (3.6) are now uniquely defined by the
recurrence relation (3.7). On the computer they may efficiently be evaluated in this
way, but for analytical work, in particular the investigation of the so far excluded
exceptional cases when s is equal to an integer, an explicit representation of all the
g, (r) is desirable.
When r=0 we may consider the nth derivative with respect to z of (3.6) and
proceed essentially in the same way as above with gl(0). As a generalization of (3.17),
(3.18) we then immediately obtain

(3.19) g"(O)=(-1)’r(a+n)r(b+n)r(s-n)r(a+s)r(b+s)n,. F(e-c’f-c’s-n)a+s, b+s

When r- s we suspect that the g, (s) can be represented in a similar form. From
(3.7) and (3.14) we have, after rearrangement of the terms,
(3.20) gl(S) -r(-s 1){s2+ (a + b)s- ef+ ab+ c(a + b+ s)}.
By the definition of s the factor of c is equal to e /f-c. Making use of this fact we
lose the symmetry with respect to the numerator parameters and obtain
gl(S) -r(-s 1){(a + s)( b + s) -(e c)(f c)}
(3.21)
1)(a + s)(b + s) 3F2( e a+s,b+s
c, f- c, 1

To proceed further, it is convenient to get rid of some factors and to consider y,(r)
defined by
(3.22) n !g,(r) (-1)"r(s -2r- n)y,(r)
rather than g, (r) itself. In view of r {0, s}, the recurrence relation for y, (r) then reads
y.(r) ={(r+n-2)(2n- l+2r-s+a+b+c)
(3.23) -ef+ ab+ bc+ ca + a + b+ c+ 1}y._l(r)
+(s- r- n + 1)(a + r+ n 2)(b + r+ n 2)(c + r+ n 2)y._2(r),
 1232 WOLFGANG BOHRING

and we are interested in its special solution which, according to (3.14), (3.17), (3.18),
(3.21), has the starting values
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

F( a + r)F( b + r)
(3.24) yo(r)
F(a + s)F(b + s) 3F2 (
e c, f c, s
a + s, b + s r)
(3.25) yl(r)
F(a+ r+l)F(b+ r+l)
F(a+s)r(b+s) 3122
e-c,f
-c,s-r-1)
a+s, b+s
The factor of y,-l(r) in (3.23) may be rewritten, using the definition of s, as
(n-2)(2n-l+4r-s+a+b+c)-(e+r)(f+r)+(a+ r)(b+r)
(3.26)
+(b+ r)(c+ r)+(c+ r)(a+ r)+ a+ b+c+3r+ l.
We then may see that the recurrence relation for y,(r) can be obtained from that for
y,(0) by the simultaneous substitutions
(3.27) a-a+r, b-b+r, cc+r, ee+r, f-f+r,
and as a consequence,
(3.28) s s- r.

The same is true for the starting values (3.24), (3.25). Since we already know from
(3.19) that
r(a+n)r(b+n) e-c,f -c,
s-n)

-
(3.29) yn(0)
F(a+s)F(b+s) a+s, b+s
it follows by the substitutions (3.27)-(3.28) that

(3.30) y,(r) r(a+r+n)r(b+r+n).r

)I’(b + 3F (e-c,f-c,s-r-n)
s) a+s, b+s
By means of (3.22) we obtain the required explicit representation for g,(r) which is
valid for both r 0 and r s and so we have proved the following.
THEOREM 1. If S e+ f-a-b-c is not equal to an integer, then the analytic
continuation of the 3122-series near z 1 in the sector larg (1 z)l < r is given by theformula

(3.31)
r(a)r(b)r(c)
r(e)r(f) 3F e,f(a,
b, c
=0
g.(O)(1-z)"+(1-z) Z
=0
g.(s)(1-z)"

where the series on the right-hand side converge inside the circle [z-11= 1 and the
coefficients g.(r), with re {0, s}, are
(3.32) g.(r)=(_l).F(a+r+n)r(b+r+n)r(s-2r-n)
F(a+s)F(b+s)n 3F2 (e-c,f -c,s-r-n)
a+s,b+s
We may observe that when f= c or e c then the 3F2 on the right-hand side of
(3.32) becomes equal to 1 and the Theorem 1 reduces to the continuation formula for
the 2F1.
The recurrence (3.7) for the quantities (3.32) can also be deduced from the work
of Lewanowicz [4] and is, essentially, a special case of the recursion relation for the
Wilson polynomials.
 THE HYPERGEOMETRIC FUNCTION 3F2 1233

4. The exceptional cases. When s is an integer, the hypergeometric series is called

s-balanced. In such a case the continuation formula may be derived from Theorem 1
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

by an approximate limiting process. For this purpose it is convenient to make the

simultaneous substitutions a -> a e, b --> b e, c -> c e, e -> e e, f->f- e, and as a
consequence, s-> s + e and then to consider the resulting equations for integral s,
separately for s-> 0 or s <_-0, respectively. Since the series representation of the 3F2 in
(3.32) is needed, a restriction of the parameter c now comes in to ensure the convergence.
By performing the limit e -> 0, which is a somewhat lengthy but standard [5] procedure,
we may derive the following results.
COROLLARY 1. If S--e + f -a- b-c is equal to an integer >= O, then the analytic
continuation of the 3F2-series near z 1 in the sector larg (1 z)[ < 7r is given by theformula

(4.1)
r(a)r(b)r()
r(e)r(/) 3
b,
e,f (a, c[ )
Y k,,(1-z)"+(1-z)’ 2 {p.+q. ln(1-z)}(1-)"
=0 =0

where the series on the right-hand side converges inside the circle z- 1[---1 and the
coefficients are

(4.2) k-(-1)’r(a+n)r(b+n)(t-n-1)!r(a+t)r(b+t)n. F:(e-c’f-c’t-n)a+t,b+t

(t--iin! t).((_l)t
(a+ t).(b+ (e-C)m(f -c),.(-n)
m=O (a+ t)m(b+ t)mm!

(4.3) {(l+n-m)+(l+ t+n)-(a+ t+n)-(b+ t+n)}

+(-1) +’n! E
m=,,+l
(e-C)m(f-c).(m-n-1)!
(a+t),.(b+t)mm! )
ifRe(c)>-t-n,
(a + t),,(b +
(4.4) q,, _(_1),
(t+n)!n! t),,3F2(e-c,f-c
a+t,b+t
-n

COROLLARY 2. If s e + f a b c is equal to an integer =<0, then the analytic

continuation of the 3F2-series near z 1 in the sector ]arg (1 z)l < 7r is given by theformula

(4.5)
r(a)r(b)r(c) b,
r(e)r(f) 3F e,f
z
(a, cl )
t--1
=(l-z)-’ Z
n=0
h,,(1-z)"+ Z
n=0
{u.+v. ln(1-z)}(1-z)"

where the series on the right-hand side converges inside the circle ]z-1] 1 and the
coefficients are

(4.6) h" (-1)"

(a- t)"(b- t)"(t-
3F)_(e-c’f -c’ -n
a-t,b-t 1),
 1234 WOLFGANG BOHRING

(a-t)t+,,(b-t),+,,
u. n t(t+
Downloaded 01/01/13 to 150.135.135.70. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php

(-1)t (e-c),,(f-c),,,(-t-n)m
t+,,

’m=0
(a-t),,,(b-t)mmV.
(4.7) {(1 + t+ n- m)+ if(1 + n)- ,(a + n)- (b + n)}
(e-C)m(f-C)m(m-t-n-1)!
+ (-1)"(+ n)! 2
,=+,,+1 (a-t),(b-t)mm!
ifRe(c)>-n,
v,,=-(-1)’(a-t)t+"(b-t)t+" 3F_ ( )
(4.8) e-c’f-c’-t-n 1
+ n !n a-t,b-t
When 0, the empty sums in (4.1) or (4.5) have to be interpreted as 0.
If f= c or e =c, the formulas in Corollaries 1 and 2 immediately reduce to the
corresponding formulas for the Gaussian hypergeometric function, which appear, apart
from some minor differences of presentation, as (15.3.10)-(15.3.12) in 1], for instance.
Taking the leading terms of (4.1) or (4.5) with 0 we may obtain (1.6) with the
constants K and L as expected and the third connecting constant
M { ab (e c)(f- c)}{2 + 2,(1) ,(a + 1) ,(b + 1)}
(4.9)
(e-c),(f-c)m
+(e-c)(f-c)-ab (Re (c) > -1).
2 (m-1)m(a)r,,(b)

REFERENCES
M. ABRAMOWITZ AND I. A. STEGUN, Handbook of Mathematical Functions, Dover, New York, 1965.
[2] W. N. BAILEY, Generalized Hypergeometric Series, Stechert-Hafner, New York, 1964.
[3] R. J. EVANS AND D. STANTON, Asymptotic formulas for zero-balanced hypergeometric series, this
Journal, 15 (1984), pp. 1010-1020.
14] S. LEWANOWICZ, Recurrence relations for hypergeometric functions of unit argument, Math. Comp., 45
(1985), pp. 521-535.
[5] Y. L. LUKE, The Special Functions and Their Approximations, Vol. 1, Academic Press, New York, 1969.
[6] F. W. J. OLVER, Asymptotics and Special Functions, Academic Press, New York, 1974.
[7] F. NAUNDORF, Ein Verfahren zur L6sung des Zusammenhangproblems bei linearen Differential-
gleichungen zweiter Ordnung mit mehreren singuliiren Stellen, Z. Angew. Math. Mech., 59 (1979),
pp. 273-275.
[8] N. NRLUND, Hypergeometric functions, Acta Math., 94 (1955), pp. 289-349.
[9] S. RAMANUJAN, Notebooks, 2 volumes, Tata Institute of Fundamental Research, Bombay, India, 1957.
10] R. SCHFKE AND D. SCHMIDT, The connection problem for general linear ordinary differential equations
at two regular singular points with applications in the theory of special functions, this Journal, 11
(1980), pp. 848-862.
[11] J. WIMP, The computation of3F2(1), Internat. J. Comput. Math., 10 (1981), pp. 55-62.
[12] Computation with Recurrence Relations, Pitman, Boston, 1984.