arxiv.org · arxiv:0712.3856v1 [math.ca] 22 dec 2007 file: om845.tex, 9 oct. 2001 topics in special...
TRANSCRIPT
arX
iv:0
712.
3856
v1 [
mat
h.C
A]
22
Dec
200
7
FILE: om845.tex, 9 O t. 2001
TOPICS IN SPECIAL FUNCTIONS
G. D. ANDERSON
1, M. K. VAMANAMURTHY, AND M. VUORINEN
Abstra t. The authors survey re ent results in spe ial fun tions, par-
ti ularly the gamma fun tion and the Gaussian hypergeometri fun tion.
1. Introdu tion
Conformal invariants are powerful tools in the study of quasi onformal
mappings, and many of these have expressions in terms of spe ial fun tions.
For instan e, the distortion results in geometri fun tion theory, su h as the
quasi onformal S hwarz Lemma, involve spe ial fun tions. A frequent task
is to simplify ompli ated inequalities, so as to larify the dependen e on im-
portant parameters without sa ri ing sharpness. For these reasons we were
led to study, as an independent subje t, various questions for spe ial fun -
tions su h as monotoneity properties and majorants/minorants in terms of
rational fun tions. These new inequalities gave rened versions of some las-
si al distortion theorems for quasi onformal maps. The lasses of fun tions
that o ur in lude omplete ellipti integrals, hypergeometri fun tions, and
Euler's gamma fun tion. The main part of our resear h is summarized in
[AVV5.
In the later development most of our resear h has involved appli ations
to geometri properties of quasi onformal maps. However, some of the ques-
tions on erning spe ial fun tions, raised in [AVV1, [AVV3, and [AVV5,
relate to spe ial fun tions whi h are useful in geometri fun tion theory in
general, not just to quasi onformal maps. In this survey our goal is to review
the latest developments of the latter type, due to many authors [A1[A9,
[AlQ1, AlQ2, AW, BPR1, BPR2, BPS, BP, EL, K1, K2, Ku.
The methods used in these studies are based on lassi al analysis. One
of the te hni al tools is the Monotone l'Hpital's Rule, stated in the next
paragraph, whi h played an important role in our work [AVV4[AVV5.
The authors dis overed this result in [AVV4, unaware that it had been used
earlier (without the name) as a te hni al tool in dierential geometry. See
[Ch, p. 124, Lemma 3.1 or [AQVV, p. 14 for relevant remarks.
1.1. Lemma. For −∞ < a < b < ∞, let g and h be real-valued fun tions
that are ontinuous on [a, b] and dierentiable on (a, b), with h′ 6= 0 on (a, b).
1991 Mathemati s Subje t Classi ation. Primary 33-02, 33B15, 33C05. Se ondary
33C65, 33E05.
Key words and phrases. Spe ial fun tions, hypergeometri fun tions, gamma fun tion,
beta fun tion, Euler-Mas heroni onstant, ellipti integrals, generalized ellipti integrals,
mean values, arithmeti -geometri mean.
1
This paper is an outgrowth of an invited talk given by the rst author at the 18th
Rolf Nevanlinna Colloquium in Helsinki, Finland, in August 2000.
1
2 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
If g′/h′ is stri tly in reasing (resp. de reasing) on (a, b), then the fun tions
g(x)− g(a)
h(x)− h(a)and
g(x) − g(b)
h(x) − h(b)
are also stri tly in reasing (resp. de reasing) on (a, b).
Graphing of the fun tions and omputer experiments in general played an
important role in our work. For instan e, the software that omes with the
book [AVV5 provides omputer programs for su h experiments.
We begin this survey by dis ussing some re ent results on the gamma
fun tion, in luding monotoneity and onvexity properties and lose approx-
imations for the Euler-Mas heroni onstant. Hypergeometri fun tions have
a very entral role in this survey. We give here a detailed proof of the so-
alled Elliott's identity for these fun tions following an outline suggested by
Andrews, Askey, and Roy in [AAR, p. 138. This identity ontains, as a spe-
ial ase, the lassi al Legendre Relation and has been studied re ently in
[KV and [BPSV. After this we dis uss mean values, a topi related to om-
plete ellipti integrals and their estimation, and we present several sharp
approximations for omplete ellipti integrals. We display inequalities for
hypergeometri fun tions that generalize the Landen relation, and on lude
the paper with a remark on re ent work of geometri mapping properties of
hypergeometri fun tions as a fun tion of a omplex argument.
This survey does not over re ent work on the appli ations of spe ial
fun tions to the hange of distan e under quasi onformal maps. For this
subje t the interested reader may onsult [AVV5.
2. The Γ and Ψ fun tions
Throughout this paper Γ will denote Euler's gamma fun tion, dened by
Γ(z) =
∫
∞
0e−ttz−1 dt, Re z > 0,
and then ontinued analyti ally to the nite omplex plane minus the set
of nonpositive integers. The re urren e formula Γ(z + 1) = z Γ(z) yields
Γ(n+1) = n! for any positive integer n. We also use the fa t that Γ(1/2) =√π. The beta fun tion is related to the gamma fun tion by B(a, b) =
Γ(a)Γ(b)/Γ(a + b). The logarithmi derivative of the gamma fun tion will
be denoted, as usual, by
Ψ(z) ≡ d
dzlog Γ(z) =
Γ′(z)
Γ(z).
The Euler-Mas heroni onstant γ is dened as (see [A2, [TY, [Y)
γ ≡ limn→∞
Dn = 0.5772156649 . . . ; Dn ≡n∑
k=1
1
k− log n.
Then Ψ(1) = Γ′(1) = −γ and Ψ(1/2) = −γ − 2 log 2. For a survey of the
gamma fun tion see [G, and for some inequalities for the gamma and psi
fun tions see [A1.
TOPICS IN SPECIAL FUNCTIONS 3
2.1. Approximation of the Euler-Mas heroni onstant. The onver-
gen e of the sequen e Dn to γ is very slow (the speed of onvergen e is
studied by Alzer [A2). D. W. DeTemple [De studied a modied sequen e
whi h onverges faster and proved
1
24(n + 1)2< Rn − γ <
1
24n2, where Rn ≡
n∑
k=1
1
k− log
(
n+1
2
)
.
Now let
h(n) = Rn − γ, H(n) = n2h(n), n > 1.
Sin e Ψ(n) = −γ − 1/n +∑n
k=1 1/k, we see that
H(n) = (Rn − γ)n2 = (Ψ(n) + 1/n− log(n+1
2))n2.
Some omputer experiments led M. Vuorinen to onje ture that H(n) in-
reases on the interval [1,∞) from H(1) = −γ + 1 − log(3/2) = 0.0173 . . .to 1/24 = 0.0416 . . .. E. A. Karatsuba proved in [K1 that for all integers
n > 1,H(n) < H(n + 1), by lever use of Stirling's formula and Fourier
series. Moreover, using the relation γ = 1− Γ′(2) she obtained, for k ≥ 1,
−ck ≤ γ − 1 + (log k)12k+1∑
r=1
d(k, r)−12k+1∑
r=1
d(k, r)
r + 1≤ ck,
where
ck =2
(12k)!+ 2k2e−k, d(k, r) = (−1)r−1 kr+1
(r − 1)!(r + 1),
giving exponential onvergen e. Some omputer experiments also seem to
indi ate that ((n + 1)/n)2H(n) is a de reasing onvex fun tion.
2.2. Gamma fun tion and volumes of balls. Formulas for geometri
obje ts, su h as volumes of solids and ar lengths of urves, often involve
spe ial fun tions. For example, if Ωn denotes the volume of the unit ball
Bn = x : |x| < 1 in Rn, and if ωn−1 denotes the (n − 1)-dimensional
surfa e area of the unit sphere Sn−1 = x : |x| = 1, n > 2, then
Ωn =πn/2
Γ((n/2) + 1); ωn−1 = nΩn.
It is well known that for n > 7 both Ωn and ωn de rease to 0 ( f. [AVV5,
2.28). However, neither Ωn nor ωn is monotone for n on [2,∞). On the other
hand, Ω1/(n logn)n de reases to e−1/2
as n → ∞ [AVV1, Lemma 2.40(2).
Re ently H. Alzer [A4 has obtained the best possible onstants a, b, A, B,α, β su h that
aΩn
n+1
n+1 6 Ωn 6 bΩn
n+1
n+1,√
n+A
2π6
Ωn−1
Ωn6
√
n+B
2π,
(
1 +1
n
)α
6Ω2n
Ωn−1Ωn+16
(
1 +1
n
)β
4 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
for all integers n > 1. He showed that a = 2/√π = 1.12837 . . ., b =
√e =
1.64872 . . . , A = 1/2, B = π/2 − 1 = 0.57079 . . . , α = 2 − (log π)/ log 2 =0.34850 . . . , β = 1/2. For some related results, see [KlR.
2.3. Monotoneity properties. In [AnQ it is proved that the fun tion
(2.4) f(x) ≡ log Γ(x+ 1)
x log x
is stri tly in reasing from (1,∞) onto (1−γ, 1). In parti ular, for x ∈ (1,∞),
(2.5) x(1−γ)x−1 < Γ(x) < xx−1.
The proof required the following two te hni al lemmas, among others:
2.6. Lemma. The fun tion
g(x) ≡∞∑
n=1
n− x
(n+ x)3
is positive for x ∈ [1, 4).
2.7. Lemma. The fun tion
(2.8) h(x) ≡ x2Ψ′(1 + x)− xΨ(1 + x) + log Γ(1 + x)
is positive for all x ∈ [1,∞).
It was onje tured in [AnQ that the fun tion f in (2.4) is on ave on
(1,∞).
2.9. Horst Alzer [A2 has given an elegant proof of the monotoneity of the
fun tion f in (2.4) by using the Monotone l'Hpital's Rule and the onvolu-
tion theorem for Lapla e transforms. In a later paper [A3 he has improved
the estimates in (2.5) to
(2.10) xα(x−1)−γ < Γ(x) < xβ(x−1)−γ , x ∈ (0, 1),
where α ≡ 1 − γ = 0.42278 . . . , β ≡ 12
(
π2/6− γ)
= 0.53385 . . . are best
possible. If x ∈ (1,∞), he also showed that (2.10) holds with best onstants
α ≡ 12
(
π2/6− γ)
= 0.53385 . . . , β ≡ 1.
2.11. Elbert and Laforgia [EL have shown that the fun tion g in Lemma 2.6
is positive for all x > −1. They used this result to prove that the fun tion
h in Lemma 2.7 is stri tly de reasing from (−1, 0] onto [0,∞) and stri tly
in reasing from [0,∞) onto [0,∞). They also showed that f ′′ < 0 for x > 1,thus proving the Anderson-Qiu onje ture [AnQ, where f is as in (2.4).
2.12. Berg and Pedersen [BP have shown that the fun tion f in (2.4) is not
only stri tly in reasing from (0,∞) onto (0, 1), but is even a (non onstant)
so- alled Bernstein fun tion. That is, f > 0 and f ′is ompletely monotoni ,
i.e., f ′ > 0, f ′′ < 0, f ′′′ > 0, . . . . In parti ular, the fun tion f is stri tly
in reasing and stri tly on ave on (0,∞).In fa t, they have proved the stronger result that 1/f is a Stieltjes trans-
form, that is, an be written in the form
1
f(x)= c+
∫
∞
0
dσ(t)
x+ t, x > 0,
TOPICS IN SPECIAL FUNCTIONS 5
where the onstant c is non-negative and σ is a non-negative measure on
[0,∞) satisfying∫
∞
0
dσ(t)
1 + t< ∞.
In parti ular, for 1/f they have shown by using Stirling's formula that c = 1.Also they have obtained dσ(t) = H(t)dt, where H is the ontinuous density
H(t) =
tlog |Γ(1− t)|+ (k − 1) log t
(log |Γ(1 − t)|)2 + (k − 1)2π2, t ∈ (k − 1, k), k = 1, 2, . . . ,
0 , t = 1, 2, . . . .
Here log denotes the usual natural logarithm. The density H(t) tends to 1/γas t tends to 0, and σ has no mass at 0.
2.13. In The Lost Notebook and Other Unpublished Papers of Ramanujan
[Ra1, the Indian mathemati al genius, appears the following re ord:
“Γ(1 + x) =√π(x
e
)x
8x3 + 4x2 + x+θx30
1/6,
where θx is a positive proper fra tion
θ0 =30
π3= .9675
θ1/12 = .8071 θ7/12 = .3058
θ2/12 = .6160 θ8/12 = .3014
θ3/12 = .4867 θ9/12 = .3041
θ4/12 = .4029 θ10/12 = .3118
θ5/12 = .3509 θ11/12 = .3227
θ6/12 = .3207 θ1 = .3359
θ∞ = 1.
Of ourse, the values in the above table, ex ept θ∞, are irrational and
hen e the de imals should be nonterminating as well as nonre urring. The
re ord stated above has been the subje t of intense investigations and is
reviewed in [BCK, page 48 (Question 754). This note of Ramanujan led the
authors of [AVV5 to make the following onje ture.
2.14. Conje ture. Let
G(x) = (e/x)xΓ(1 + x)/√π
and
H(x) = G(x)6 − 8x3 − 4x2 − x =θx30
.
Then H is in reasing from (1,∞) into (1/100, 1/30) [AVV5, p. 476.
2.15. In a ni e pie e of work, E. A. Karatsuba [K2 has proved the above
onje ture. She did this by representing the fun tion H(x) as an integral for
whi h she was able to nd an asymptoti development. Her work also led to
6 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
an interesting asymptoti formula for the gamma fun tion:
Γ(x+ 1) =√π(x
e
)x(
8x3 + 4x2 + x+1
30− 11
240x+
79
3360x2+
3539
201600x3
− 9511
403200x4− 10051
716800x5+
47474887
1277337600x6+
a7x7
+ · · ·+ anxn
+∆n+1(x))1/6
,
where ∆n+1(x) = O( 1xn+1 ), as x → ∞, and where ea h ak is given expli itly
in terms of the Bernoulli numbers.
3. Hypergeometri fun tions
Given omplex numbers a, b, and c with c 6= 0,−1,−2, . . . , the Gaussian
hypergeometri fun tion is the analyti ontinuation to the slit plane C \[1,∞) of
(3.1) F (a, b; c; z)=2F1(a, b; c; z)≡∞∑
n=0
(a, n)(b, n)
(c, n)
zn
n!, |z| < 1.
Here (a, 0) = 1 for a 6= 0, and (a, n) is the shifted fa torial fun tion
(a, n) ≡ a(a+ 1)(a + 2) · · · (a+ n− 1)
for n = 1, 2, 3, . . ..The hypergeometri fun tion w = F (a, b; c; z) in (3.1) has the simple
dierentiation formula
(3.2)
d
dzF (a, b; c; z) =
ab
cF (a+ 1, b + 1; c + 1; z).
The behavior of the hypergeometri fun tion near z = 1 in the three ases
a+ b < c, a+ b = c, and a+ b > c, a, b, c > 0, is given by
(3.3)
F (a, b; c; 1) = Γ(c)Γ(c−a−b)Γ(c−a)Γ(c−b) , a+ b < c,
B(a, b)F (a, b; a + b; z) + log(1− z)
= R(a, b) +O((1− z) log(1− z)),
F (a, b; c; z) = (1− z)c−a−bF (c− a, c− b; c; z), c < a+ b,
where R(a, b) = −2γ −Ψ(a)−Ψ(b), R(a) ≡ R(a, 1− a), R(12 ) = log 16, andwhere log denotes the prin ipal bran h of the omplex logarithm. The above
asymptoti formula for the zero-balan ed ase a+ b = c is due to Ramanujan
(see [As, [Be1). This formula is implied by [AS, 15.3.10.
The asymptoti formula (3.3) gives a pre ise des ription of the behavior
of the fun tion F (a, b; a + b; z) near the logarithmi singularity z = 1. Thissingularity an be removed by an exponential hange of variables and the
transformed fun tion will be nearly linear.
3.4. Theorem. [AQVV For a, b > 0, let k(x) = F (a, b; a+b; 1−e−x), x > 0.Then k is an in reasing and onvex fun tion with k′((0,∞)) = (ab/(a + b),Γ(a+ b)/(Γ(a)Γ(b))).
3.5. Theorem. [AQVV Given a, b > 0, and a + b > c, d ≡ a + b − c, the
fun tion ℓ(x) = F (a, b; c; 1 − (1 + x)−1/d), x > 0, is in reasing and onvex,
with ℓ′((0,∞)) = (ab/(cd), Γ(c)Γ(d)/(Γ(a)Γ(b))).
TOPICS IN SPECIAL FUNCTIONS 7
3.6. Gauss ontiguous relations and derivative formula. The six fun -
tions F (a±1, b; c; z), F (a, b±1; c; z), F (a, b; c±1; z) are alled ontiguous to
F (a, b; c; z). Gauss dis overed 15 relations between F (a, b; c; z) and pairs of
its ontiguous fun tions [AS, 15.2.1015.2.27, [R2, Se tion 33. If we apply
these relations to the dierentiation formula (3.2), we obtain the following
useful formulas.
3.7. Theorem. For a, b, c > 0, z ∈ (0, 1), let u = u(z) = F (a − 1, b; c; z),v = v(z) = F (a, b; c; z), u1 = u(1− z), v1 = v(1− z). Then
zdu
dz= (a− 1)(v − u),(3.8)
z(1 − z)dv
dz= (c− a)u+ (a− c+ bz)v,(3.9)
and
(3.10)
ab
cz(1− z)F (a+ 1, b+ 1; c + 1; z) = (c− a)u+ (a− c+ bz)v.
Furthermore,
(3.11) z(1−z)d
dz
(
uv1+u1v−vv1)
=(1−a−b)[
(1−z)uv1−zu1v−(1−2z)vv1)]
.
Formulas (3.8)-(3.10) in Theorem 3.7 are well known. See, for example,
[AAR, 2.5.8. On the other hand, formula (3.11), whi h follows from (3.8)-
(3.9) is rst proved in [AQVV, 3.13 (4).
Note that the formula
(3.12) z(1− z)dF
dz= (c− b)F (a, b− 1; c; z) + (b− c+ az)F (a, b; c; z)
follows from (3.9) if we use the symmetry property F (a, b; c; z) = F (b, a; c; z).
3.13. Corollary. With the notation of Theorem 3.7, if a ∈ (0, 1), b = 1−a <c, then
uv1 + u1v − vv1 = u(1) =(Γ(c))2
Γ(c+ a− 1)Γ(c− a+ 1).
4. Hypergeometri differential equation
The fun tion F (a, b; c; z) satises the hypergeometri dierential equation
(4.1) z(1 − z)w′′ + [c− (a+ b+ 1)z]w′ − abw = 0.
Kummer dis overed solutions of (4.1) in various domains, obtaining 24 in
all; for a omplete list of his solutions see [R2, pp. 174, 175.
4.2. Lemma. (1) If 2c = a+ b+1 then both F (a, b; c; z) and F (a, b; c; 1− z)satisfy (4.1) in the lens-shaped region z : 0 < |z| < 1, 0 < |1 − z| < 1.(2) If 2c = a+ b+ 1 then both F (a, b; c; z2) and F (a, b; c; 1 − z2) satisfy the
dierential equation
(4.3) z(1− z2)w′′ + [2c − 1− (2a+ 2b+ 1)z2]w′ − 4abzw = 0
in the ommon part of the disk z : |z| < 1 and the lemnis ate z : |1−z2| <1.
8 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
Proof. By Kummer ( f. [R2, pp. 174-177), the fun tions F (a, b; c; z)and F (a, b; a+ b+1− c; 1− z) are solutions of (4.1) in z : 0 < |z| < 1 and
z : 0 < |1 − z| < 1, respe tively. But a + b + 1 − c = c under the stated
hypotheses. The result (2) follows from result (1) by the hain rule.
4.4. Lemma. The fun tion F (a, b; c;√1− z2) satises the dierential equa-
tion
Z3(1− Z)zw′′ − Z(1− Z) + [c− (a+ b+ 1)Z]Zz2w′ − abz3w = 0,
in the subregion of the right half-plane bounded by the lemnis ate r2 =2cos(2ϑ), −π/4 ≤ ϑ ≤ π/4, z = reiϑ. Here Z =
√1− z2, where the
square root indi ates the prin ipal bran h.
Proof. From (4.1), the dierential equation for w = F (a, b; c; t) is givenby
t(1− t)d2w
dt2+ [c− (a+ b+ 1)t]
dw
dt− abw = 0.
Now put t =√1− z2. Then
dz
dt= − t
z,dt
dz= −z
t,d2t
dz2= − 1
t3
and
dw
dt= − t
z
dw
dz,d2w
dt2=
t2
z2d2w
dz2− 1
z3dw
dz.
So
t(1− t)[ t2
z2w′′ − 1
z3w′
]
+[
c− (a+ b+ 1)t](
− t
z
)
w′ − abw = 0.
Multiplying through by z3 and repla ing t by Z ≡√1− z2 gives the result.
If w1 and w2 are two solutions of a se ond order dierential equation, then
their Wronskian is dened to be W (w1, w2) ≡ w1w′
2 − w2w′
1.
4.5. Lemma. [AAR, Lemma 3.2.6 If w1 and w2 are two linearly independent
solutions of (4.1), then
W (z) = W (w1, w2)(z) =A
zc(1− z)a+b−c+1,
where A is a onstant.
(Note the misprint in [AAR, (3.10), where the oe ient x(1 − x) is
missing from the rst term.)
4.6. Lemma. If 2c = a+ b+ 1 then, in the notation of Theorem 3.7,
(4.7) (c− a)(uv1 + u1v) + (a− 1)vv1 = A · z1−c(1− z)1−c.
Proof. If 2c = a+ b+ 1 then by Lemma 4.2(1), both v(z) and v(1 − z)are solutions of (4.1). Sin e W (z) = W (v1, v)(z) = v′(z)v1(z) − v(z)v′1(z),we have
z(1− z)W (z) = z(1− z)(v′v1 − vv′1)
= (c− a)(uv1 + u1v) + (2a+ b− 2c)vv1
= (c− a)(uv1 + u1v) + (a− 1)vv1.
TOPICS IN SPECIAL FUNCTIONS 9
Next, sin e 2c = a + b + 1, Lemma 4.5 shows that zc(1 − z)cW (z) = A,and the result follows.
Note that in the parti ular ase c = 1, a = b = 12 the right side of (4.7) is
onstant and the result is similar to Corollary 3.13. This parti ular ase is
Legendre's Relation (5.3), and this proof of it is due to Duren [Du.
4.8. Lemma. If a, b > 0, c ≥ 1, and 2c = a + b + 1, then the onstant Ain Lemma 4.6 is given by A = (Γ(c))2/(Γ(a)Γ(b)). In parti ular, if c = 1then Lemma 4.6 redu es to Legendre's Relation (5.8) for generalized ellipti
integrals.
Proof. The idea of the proof is to repla e the possibly unbounded hyper-
geometri fun tions in formula (4.7) by bounded or simpler ones. Therefore
we onsider three ases.
Case (1): c ≥ 2. Now a+ b ≥ c+ 1 ≥ 3. By (3.3) or [AS, 15.3.3, we have
u(z) = (1−z)2−cF (c+1−a, c−b; c; z), u1(z) = z2−cF (c+1−a, c−b; c; 1−z),
v(z) = (1− z)1−cF (c− a, c− b; c; z), v1(z) = z1−cF (c− a, c− b; c; 1 − z).
Hen e
A = (c− a)[(1− z)F (c + 1− a, c− b; c; z)F (c − a, c− b; c; 1 − z)
+zF (c+ 1− a, c− b; c; 1 − z)F (c − a, c− b; c; z)]
+(a− 1)F (c − a, c− b; c; z)F (c − a, c− b; c; 1 − z).
Now, sin e a+ b− c = c− 1, letting z → 0, from (3.3) we get
A = (c− a)Γ(c)Γ(c− 1)
Γ(a)Γ(b))+ (a− 1)
Γ(c)Γ(c − 1)
Γ(a)Γ(b)
= (c− 1)Γ(c)Γ(c − 1)
Γ(a)Γ(b)=
(Γ(c))2
Γ(a)Γ(b),
as laimed.
Case (2): 1 < c < 2. Now, 1 < c < a+ b < c+ 1 < 3. Then
A = (c−a)[(1−z)c−1u(z)F (c−a, c−b; c; 1−z)+zc−1u1(z)F (c−a, c−b; c; z)]
+(a− 1)F (c − a, c− b; c; z)F (c − a, c− b; c; 1 − z).
Now letting z → 0, from (3.3), as in Case (1), we get
A = (c− a)Γ(c)Γ(c− 1)
Γ(a)Γ(b)+ (a− 1)
Γ(c)Γ(c − 1)
Γ(a)Γ(b)
= (c− 1)Γ(c)Γ(c − 1)
Γ(a)Γ(b)=
(Γ(c))2
Γ(a)Γ(b),
as laimed.
Case (3): c = 1. Now a+ b = 1. Then
A = (1− a)[u(z)v1(z) + u1(z)v(z) − v(z)v1(z)]
= (1− a)u1(z)v(z) + (1− a)v1(z)[u(z) − v(z)].
10 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
From [R1, Ex. 21(4), p.71 we have
u(z)− v(z) = F (a− 1, b; c; z) − F (a, b; c; z)
=c− b
czF (a, b; c + 1; z) − zF (a, b; c; z),
so that
u(z)− v(z)
z=
c− b
cF (a, b; c + 1; z) − F (a, b; c; z) → −b/c,
as z → 0. Also, by (3.3), zv1(z) → 0 as z → 0. Hen e, letting z → 0, we get
A = (1− a)u1(1) = (1− a)Γ(c)Γ(c + 1− a− b)
Γ(c+ 1− a)Γ(c− b)
= (1− a)(Γ(c))2
(1 − a)Γ(a)Γ(b)=
(Γ(c))2
Γ(a)Γ(b),
as laimed.
Note that, in Case (3), Γ(c) = Γ(1) = 1,Γ(b) = Γ(1−a), and thus by [AS,
6.1.17 A = 1/(Γ(a)Γ(1 − a)) = (sinπa)/π.
For rational triples (a, b, c) there are numerous ases where the hyper-
geometri fun tion F (a, b; c; z) redu es to a simpler fun tion (see [PBM).
Other important parti ular ases are generalized ellipti integrals, whi h we
will now dis uss. For a, r ∈ (0, 1), the generalized ellipti integral of the rst
kind is given by
Ka = Ka(r) =π
2F (a, 1− a; 1; r2)
= (sinπa)
∫ π/2
0(tan t)1−2a(1− r2 sin2 t)−a dt,
K′
a = K′
a(r) = Ka(r′).
We also dene
µa(r) =π
2 sin(πa)
K′
a(r)
Ka(r), r′ =
√
1− r2.
The invariant of the linear dierential equation
(4.9) w′′ + pw′ + qw = 0,
where p and q are fun tions of z, is dened to be
I ≡ q − 1
2p′ − 1
4p2
( f. [R2,p.9). If w1 and w2 are two linearly independent solutions of (4.9),
then their quotient w ≡ w2/w1 satises the dierential equation
Sw(z) = 2I,
where Sw is the S hwarzian derivative
Sw ≡(
w′′
w′
)
′
− 1
2
(
w′′
w′
)2
and the primes indi ate dierentiations ( f. [R2, pp. 18,19).
TOPICS IN SPECIAL FUNCTIONS 11
From these onsiderations and the fa t that Ka(r) and K′
a(r) are linearlyindependent solutions of (4.3) (see [AQVV, (1.11)), it follows that w = µa(r)satises the dierential equation
Sw(r) =−8a(1− a)
(r′)2+
1 + 6r2 − 3r4
2r2(r′)4.
The generalized ellipti integral of the se ond kind is given by
Ea = Ea(r) ≡π
2F (a− 1, 1 − a; 1; r2)
= (sinπa)
∫ π/2
0(tan t)1−2a(1− r2 sin2 t)1−a dt
E′
a = E′
a(r) = Ea(r′),
Ea(0) =π
2, Ea(1) =
sin(πa)
2(1 − a).
For a = 12 , Ka and Ea redu e to K and E, respe tively, the usual ellipti
integrals of the rst and se ond kind, respe tively. Likewise µ1/2(r) = µ(r),the modulus of the well-known Grötzs h ring in the plane [LV.
4.10. Corollary. The generalized ellipti integrals Ka and Ea satisfy the dif-
ferential equations
r(r′)2d2Ka
dr2+ (1− 3r2)
dKa
dr− 4a(1− a)rKa = 0,(4.11)
r(r′)2d2Ea
dr2+ (r′)2
dEa
dr+ 4(1− a)2rEa = 0,(4.12)
respe tively.
Proof. These follow from (4.3).
For a = 12 these redu e to well-known dierential equations [AVV5, pp.
474-475, [BF.
5. Identities of Legendre and Elliott
In geometri fun tion theory the omplete ellipti integrals K(r) and E(r)play an important role. These integrals may be dened, respe tively, as
K(r) = π2F (12 ,
12 ; 1; r
2), E(r) = π2F (12 ,−1
2 ; 1; r2),
for −1 < r < 1. These are Ka(r) and Ea(r), respe tively, with a = 12 . We
also onsider the fun tions
K′ = K
′(r) = K(r′), 0 < r < 1,
K(0) = π/2, K(1−) = +∞,
and
E′ = E
′(r) = E(r′), 0 6 r 6 1,
where r′ =√1− r2. For example, these fun tions o ur in the following
quasi onformal ounterpart of the S hwarz Lemma [LV:
12 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
5.1. Theorem. For K ∈ [1,∞), let w be a K-quasi onformal mapping of
the unit disk D = z : |z| < 1 into the unit disk D′ = w : |w| < 1 with
w(0) = 0. Then
|w(z)| 6 ϕK(|z|),where
(5.2) ϕK(r) ≡ µ−1( 1
Kµ(r)
)
and µ(r) ≡ πK′(r)
2K(r).
This result is sharp in the sense that for ea h z ∈ D and K ∈ [1,∞) there is
an extremal K-quasi onformal mapping that takes the unit disk D onto the
unit disk D′with w(0) = 0 and |w(z)| = ϕK(|z|) (see [LV, p. 63).
It is well known [BF that the omplete ellipti integrals K and E satisfy
the Legendre relation
(5.3) EK′ + E
′K−KK
′ =π
2.
For several proofs of (5.3) see [Du.
In 1904, E. B. Elliott [E ( f. [AVV3) obtained the following generalization
of this result.
5.4. Theorem. If a, b, c > 0 and 0 < x < 1 then
(5.5) F1F2 + F3F4 − F2F3 =Γ(a+ b+ 1)Γ(b+ c+ 1)
Γ(a+ b+ c+ 32 )Γ(b+
12)
.
where
F1 = F
(
1
2+ a,−1
2− c; 1 + a+ b;x
)
,
F2 = F
(
1
2− a,
1
2+ c; 1 + b+ c; 1− x
)
,
F3 = F
(
1
2+ a,
1
2− c; 1 + a+ b;x
)
,
F4 = F
(
− 1
2− a,
1
2+ c; 1 + b+ c; 1 − x
)
.
Clearly (5.3) is a spe ial ase of (5.5), when a = b = c = 0 and x =r2. For a dis ussion of generalizations of Legendre's Relation see Karatsuba
and Vuorinen [KV and Balasubramanian, Ponnusamy, Sunanda Naik, and
Vuorinen [BPSV.
Elliott proved (5.5) by a lever hange of variables in multiple integrals.
Another proof was suggested without details in [AAR, p. 138, and here we
provide the missing details.
Proof of Theorem 5.4. In parti ular, let y1 ≡ F3, y2 ≡ x−a−b(1 −x)b+cF2. Then by [R2, pp. 174, 175 or [AAR, (3.2.12), (3.2.13), y1 and y2are linearly independent solutions of (4.1).
By (3.12),
(5.6) x(1−x)y′1 =(
a+ b+ c+1
2
)
F1 + [−(a+ b+ c+1
2) + (a+
1
2)x]F3,
TOPICS IN SPECIAL FUNCTIONS 13
and by (3.9),
x(1− x)y′2 =x(1− x)[
− (a+ b)x−a−b−1(1− x)b+c
− (b+ c)x−a−b(1− x)b+c−1]
F2
− x−a−b(1− x)b+c[(
a+ b+ c+1
2
)
F4
+[
− (a+ b+ c+1
2
)
+(
c+1
2
)
(1− x))F2
]
.
(5.7)
Multiplying (5.7) by y1 and (5.6) by y2 and subtra ting, we obtain
x(1− x)(y2y′
1 − y1y′
2) =(
a+ b+ c+1
2
)
x−a−b(1− x)b+cF1F2
+[
−(
a+ b+ c+1
2
)
+(
a+1
2
)
x]
x−a−b(1− x)b+cF2F3
+ x(1− x)[(a+ b)x−a−b−1(1− x)b+c
+ (b+ c)x−a−b(1− x)b+c−1]F2F3
+ x−a−b(1− x)b+c(
a+ b+ c+1
2
)
F3F4
+ x−a−b(1− x)b+c[
− (a+ b+ c+1
2
)
+(
c+1
2
)
(1− x)]
F2F3
=(
a+ b+ c+1
2
)
x−a−b(1− x)b+cF1F2
+ x−a−b(1− x)b+c[
− (a+ b+ c+1
2
)
+(
a+1
2
)
x+ (a+ b)(1− x)
+ (b+ c)x−(
a+ b+ c+1
2
)
+(
c+1
2
)
(1− x)]
F2F3
+ x−a−b(1− x)b+c(
a+ b+ c+1
2
)
F3F4
=(
a+ b+ c+1
2
)
x−a−b(1− x)b+cF1F2
+ x−a−b(1− x)b+c[
−(
a+ b+ c+ 12
)]
F2F3
+(
a+ b+ c+1
2
)
x−a−b(1− x)b+cF3F4.
So
x(1− x)W (y2, y1) =A
xa+b(1− x)−b−c
= x−a−b(1− x)b+c(
a+ b+ c+1
2
)
[F1F2 + F3F4 − F2F3]
by Lemma 4.5. Thus
F1F2 + F3F4 − F2F3 = A,
where A is a onstant.
14 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
Now, by (3.3),
F1F2 tends to F(1
2+ a,−1
2− c; a+ b+ 1; 1
)
=Γ(a+ b+ 1)Γ(b+ c+ 1)
Γ(b+ 12)Γ(a+ b+ c+ 3
2 )
as x → 1, sin e 12 + a+ (−1
2 − c) = a− c < a+ b+ 1.Next
F3F4 − F3F2 = F3(F4 − F2),
where F4 − F2 ∼ onst · (1− x)2 +O((1 − x)3), and
F3 =Γ(a+ b+ 1)Γ(b + c+ 1)
Γ(b+ 12 )Γ(a+ b+ c+ 1
2)
if a+ 12 + 1
2 − c < a+ b+ 1, or −c < b, i.e., b > 0 or c > 0. If −c = b = 0,then
F3 =R(a+ 1
2 ,12 − c)
B(a+ 12 ,
12 − c)
+O((1− x) log(1− x))
by (3.3). In either ase the produ t F3(F4 − F2) tends to 0 as x → 1. Thethird ase a+ 1
2+12−c > a+b+1 is impossible sin e we are assuming that b, c
are nonnegative. Thus A = Γ(a+b+1)Γ(b+c+1)/(Γ(b+ 12)Γ(a+b+c+ 3
2)),as desired.
The generalized ellipti integrals satisfy the identity
(5.8) EaK′
a + E′
aKa −KaK′
a =π sin(πa)
4(1− a).
This follows from Elliott's formula (5.5) and ontains the lassi al relation
of Legendre (5.3) as a spe ial ase.
Finally, we re ord the following formula of Kummer [Kum, p. 63, Form.
30:
F (a, b; a+ b− c+ 1; 1− x)F (a+ 1, b+ 1; c + 1;x)
+c
a+ b− c+ 1F (a, b; c;x)F (a + 1, b+ 1; a+ b− c+ 2; 1 − x)
= Dx−c(1− x)c−a−b−1, D =Γ(a+ b− c+ 1)Γ(c+ 1)
Γ(a+ 1)Γ(b+ 1).
This formula, like Elliott's identity, may be rewritten in many dierent ways
if we use the ontiguous relations of Gauss. Note also the spe ial ase c =a+ b− c+ 1.
6. Mean values
The arithmeti -geometri mean of positive numbers a, b > 0 is the limit
AGM(a, b) = lim an = lim bn,
where a0 = a, b0 = b, and for n = 0, 1, 2, 3, ...,
an+1 = A(an, bn) ≡ (an + bn)/2, bn+1 = G(an, bn) ≡√
anbn,
are the arithmeti and geometri means of an and bn, resp. For a mean value
M , we also onsider the t-modi ation dened as
Mt(a, b) = M(at, bt)1/t.
TOPICS IN SPECIAL FUNCTIONS 15
For example, the power mean of a, b > 0 is
At(a, b) =
(
at + bt
2
)1/t
,
and the logarithmi mean is
L(a, b) =a− b
log(b/a).
The power mean is the t-modi ation of the arithmeti mean A1(a, b).
The onne tion between mean values and ellipti integrals is provided by
Gauss's amazing result
AGM(1, r′) =π
2K(r).
This formula motivates the question of nding minorant/majorant fun tions
for K(r) in terms of mean values. For a xed x > 0 the fun tion t 7→Lt(1, x), t > 0, in reases with t by [VV, Theorem 1.2 (1). The two-sided
inequality
L3/2(1, x) > AG(1, x) > L(1, x)
holds; the se ond inequality was pointed out in [CV, and the rst one, due
to J. and P. Borwein [BB2, proves a sharp estimate settling a question raised
in onne tion with [VV. Combined with the identity above, this inequality
yields a very pre ise inequality for K(r).Several inequalities between mean values have been proved re ently. See,
for instan e, [AlQ2, [QS, [S1, [S2, [S3, [T, [C, and [Br.
Finally, we remark that the arithmeti -geometri mean, together with
Legendre's Relation, played a entral role in a rapidly onverging algorithm
for the number π in [Sa. See also [BB1, H, Le, Lu.
7. Approximation of ellipti integrals
E ient algorithms for the numeri al evaluation of K(r) and E(r) are
based on the arithmeti -geometri mean iteration of Gauss. This fa t led to
some lose majorant/minorant fun tions for K(r) in terms of mean values in
[VV.
Next, let a and b be the semiaxes of an ellipse with a > b and e entri ity
e =√a2 − b2/a, and let L(a, b) denote the ar length of the ellipse. Without
loss of generality we take a = 1. In 1742, Ma laurin ( f. [AB) determined
that
L(1, b) = 4E(e) = 2π · 2F1(12 ,−1
2 ; 1; e2).
In 1883, Muir ( f. [AB) proposed that L(1, b) ould be approximated by
the expression 2π[(1 + b3/2)/2]2/3. Sin e this expression has a lose resem-
blan e to the power mean values studied in [VV, it is natural to study the
sharpness of this approximation. Close numeri al examination of the error
in this approximation led Vuorinen [V2 to onje ture that Muir's approxi-
mation is a lower bound for the ar length. Letting r =√1− b2, Vuorinen
asked whether
(7.1)
2
πE(r) = 2F1
(
12 ,−1
2 ; 1; r2)
>
(1 + (r′)3/2
2
)2/3
16 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
for all r ∈ [0, 1].In [BPR1 Barnard and his oauthors proved that inequality (7.1) is true.
In fa t, they expanded both fun tions into Ma laurin series and proved that
the dieren es of the orresponding oe ients of the two series all have the
same sign.
Later, the same authors [BPR2 dis overed an upper bound for E that
omplements the lower bound in (7.1):
(7.2)
2
πE(r) = 2F1
(
12 ,−1
2 ; 1; r2)
6
(1 + (r′)2
2
)1/2, 0 6 r 6 1.
See also [BPS.
In [BPR2 the authors have onsidered 13 histori al approximations (by
Kepler, Euler, Peano, Muir, Ramanujan, and others) for the ar length of an
ellipse and determined a linear ordering among them. Their main tool was
the following Lemma 7.3 on generalized hypergeometri fun tions. These
fun tions are dened by the formula
pFq(a1, a2, · · · , ap; b1, b2, · · · , bq; z) ≡ 1 +
∞∑
n=1
Πpi=1(ai, n)
Πqj=1(bj , n)
· zn
n!,
where p and q are positive integers and in whi h no denominator parameter
bj is permitted to be zero or a negative integer. When p = 2 and q = 1, thisredu es to the usual Gaussian hypergeometri fun tion F (a, b; c; z).
7.3. Lemma. Suppose a, b > 0. Then for any ǫ satisfying ab1+a+b < ǫ < 1,
3F2(−n, a, b; 1 + a+ b, 1 + ǫ− n; 1) > 0
for all integers n > 1.
7.4. Some inequalities for K(r). At the end of the pre eding se tion we
pointed out that upper and lower bounds an be found for K(r) in terms
of mean values. Another sour e for the approximation of K(r) is based on
the asymptoti behavior at the singularity r = 1, where K(r) has logarith-mi growth. Some of the approximations motivated by this aspe t will be
dis ussed next.
Anderson, Vamanamurthy, and Vuorinen [AVV2 approximated K(r) bythe inverse hyperboli tangent fun tion arth, obtaining the inequalities
(7.5)
π
2
(
arth r
r
)1/2
< K(r) <π
2
arth r
r,
for 0 < r < 1. Further results were proved by Laforgia and Sismondi [LS.
Kühnau [Ku and Qiu [Q proved that, for 0 < r < 1,
9
8 + r2<
K(r)
log(4/r′).
Qiu and Vamanamurthy [QVa proved that
K(r)
log(4/r′)< 1 +
1
4(r′)2 for 0 < r < 1.
TOPICS IN SPECIAL FUNCTIONS 17
Several inequalities for K(r) are given in [AVV5, Theorem 3.21. Later Alzer
[A3 showed that
1 +( π
4 log 2− 1)
(r′)2 <K(r)
log(4/r′),
for 0 < r < 1. He also showed that the onstants
14 and π/(4 log 2) − 1 in
the above inequalities are best possible.
For further renements, see [QVu1, (2.24) and [Be.
Alzer and Qiu [AlQ1 have written a related manus ript in whi h, besides
proving many inequalities for omplete ellipti integrals, they have rened
(7.5) by proving that
π
2
(arth r
r
)3/4< K(r) <
π
2
arth r
r.
They also showed that 3/4 and 1 are the best exponents for (arth r)/r on
the left and right, respe tively.
One of the interesting tools of these authors is the following lemma of
Biernaki and Krzy» [BK (for a detailed proof see [PV1):
7.6. Lemma. Let rn and sn, n = 1, 2, . . . be real numbers, and let the power
series R(x) =∑
∞
n=1 rnxnand S(x) =
∑
∞
n=1 snxnbe onvergent for |x| <
1. If sn > 0 for n = 1, 2, . . ., and if rn/sn is stri tly in reasing (resp.
de reasing) for n = 1, 2, . . ., then the fun tion R/S is stri tly in reasing
(resp. de reasing) on (0, 1).
7.7. Generalized ellipti integrals. For the ase of generalized ellipti
integrals some inequalities are given in [AQVV. B. C. Carlson has intro-
du ed some standard forms for ellipti integrals involving ertain symmetri
integrals. Approximations for these fun tions an be found in [CG.
8. Landen inequalities
It is well known ( f. [BF) that the omplete ellipti integral of the rst
kind satises the Landen identities
K
(
2√r
1 + r
)
= (1 + r)K(r), K
(
1− r
1 + r
)
=1 + r
2K
′(r).
Re all that K(r) = π2F (12 ,
12 ; 1; r
2). It is thus natural to onsider, as sug-
gested in [AVV3, the problem of nding an analogue of these formulas for
the zero-balan ed hypergeometri fun tion F (a, b; c; r) for a, b, c > 0 and
a+ b = c, at least when the parameters (a, b, c) are lose to (12 ,12 , 1). From
(3.3) it is lear that F (a, b; c; r2) has a logarithmi singularity at r = 1, ifa, b > 0, c = a + b ( f. [AAR). Some renements of the growth estimates
were given in [ABRVV and [PV1.
Qiu and Vuorinen [QVu1 proved the following Landen-type inequalities:
For a, b ∈ (0, 1), c = a+ b,
F
(
a, b; c;
(
2√r
1 + r
)2)
6 (1 + r)F (a, b; c; r2)
6 F
(
a, b; c;
(
2√r
1 + r
)2)
+1
B(R− log 16)
18 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
and
1 + r
2F (a, b; c; 1 − r2) 6 F
(
a, b; c;
(
1− r
1 + r
)2)
61 + r
2
[
F (a, b; c; 1 − r2) +1
B(R− log 16)
]
,
with equality in ea h instan e if and only if a = b = 12 . Here B = B(a, b),
the beta fun tion, and R = R(a, b) = −2γ−Ψ(a)−Ψ(b), where Ψ is as given
in Se tion 2.
9. Hypergeometri series as an analyti fun tion
For rational triples (a, b, c) the hypergeometri fun tion often an be ex-
pressed in terms of elementary fun tions. Long lists with su h triples on-
taining hundreds of fun tions an be found in [PBM. For example, the
fun tions
f(z) ≡ zF (1, 1; 2; z) = − log(1− z)
and
g(z) ≡ zF (1, 1/2; 3/2; z2) =1
2log
(
1 + z
1− z
)
have the property that they both map the unit disk into a strip domain.
Observing that they both orrespond to the ase c = a + b one may ask
(see [PV1, PV2) whether there exists δ > 0 su h that zF (a, b; a + b; z) andzF (a, b; a + b; z2) with a, b ∈ (0, δ) map into a strip domain.
Membership of hypergeometri fun tions in some spe ial lasses of univa-
lent fun tions is studied in [PV1, PV2, BPV2.
Referen es
[AS M. Abramowitz and I. A. Stegun, eds: Handbook of Mathemati al Fun -
tions with Formulas, Graphs and Mathemati al Tables, Dover, New York, 1965.
[AB G. Almkvist and B. Berndt: Gauss, Landen, Ramanujan, the arithmeti -
geometri mean, ellipses, pi, and the Ladies Diary, Amer. Math. Monthly 95
(1988), 585608.
[A1 H. Alzer: On some inequalities for the gamma and psi fun tions,Math. Comp.
66 (1997), 373389.
[A2 H. Alzer: Inequalities for the gamma and polygamma fun tions, Abh. Math.
Sem. Univ. Hamburg 68, (1998), 363372.
[A3 H. Alzer: Sharp inequalities for the omplete ellipti integral of the rst kind,
Math. Pro . Camb. Phil. So . 124 (1998), 309314.
[A4 H. Alzer: Inequalities for the gamma fun tion, Pro . Amer. Math. So . 128
(2000), 141147.
[A5 H. Alzer: Inequalities for the volume of the unit ball in Rn
, J. Math. Anal.
Appl. 252 (2000), 353363.
[A6 H. Alzer: A power mean inequality for the gamma fun tion, Monatsh. Math.
131 (2000), 179188.
[A7 H. Alzer: Inequalities for the Hurwitz zeta fun tion, Pro . Royal So . Edinb.
130A (2000), 12271236.
[A8 H. Alzer: Mean value inequalities for the polygamma fun tion, Aeqationes
Math. 61 (2001), 151161.
[A9 H. Alzer: Sharp inequalities for the beta fun tion, Indag. Math. (N.S.) 12
(2001), 1521.
[AlQ1 H. Alzer and S.-L. Qiu: Monotoni ity theorems and inequalities for the
omplete ellipti integrals, Manus ript (2000).
TOPICS IN SPECIAL FUNCTIONS 19
[AlQ2 H. Alzer and S.-L. Qiu: Inequalities for means in two variables, Manus ript
(2000).
[AW H. Alzer and J. Wells: Inequalities for the polygamma fun tions, SIAM J.
Math. Anal. 29 (1998), 14591466 (ele troni ).
[ABRVV G. D. Anderson, R. W. Barnard, K. C. Ri hards, M. K. Vamana-
murthy, and M. Vuorinen: Inequalities for zero-balan ed hypergeometri
fun tions, Trans. Amer. Math. So . 347 (1995), 17131723.
[AnQ G. D. Anderson and S.-L. Qiu: A monotoneity property of the gamma
fun tion, Pro . Amer. Math. So . 125 (1997), 33553362.
[AQVV G. D. Anderson, S.-L. Qiu, M. K. Vamanamurthy, and M. Vuori-
nen: Generalized ellipti integrals and modular equations, Pa i J. Math.
192 (2000), 137.
[AQVa G. D. Anderson, S.-L. Qiu, M. K. Vamanamurthy: Ellipti integral in-
equalities, with appli ations. Constr. Approx. 14 (1998), no. 2, 195207.
[AQVu G. D. Anderson, S.-L. Qiu, and M. Vuorinen: Pre ise estimates for
dieren es of the Gaussian hypergeometri fun tion, J. Math. Anal. Appl. 215
(1997), 212234.
[AVV1 G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen: Spe ial
fun tions of quasi onformal theory, Exposition. Math. 7 (1989), 97138.
[AVV2 G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen: Fun tional
inequalities for hypergeometri fun tions and omplete ellipti integrals, SIAM
J. Math. Anal. 23 (1992), 512524.
[AVV3 G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen: Hypergeo-
metri fun tions and ellipti integrals, in Current Topi s in Analyti Fun tion
Theory, ed. by H. M. Srivastava and S. Owa, World S ienti , London, 1992,
pp. 4885.
[AVV4 G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen: Inequalities
for quasi onformal mappings in spa e, Pa i J. Math. 160 (1993), 118.
[AVV5 G. D. Anderson, M. K. Vamanamurthy, and M. Vuorinen: Conformal
Invariants, Inequalities, and Quasi onformal Maps, J. Wiley, 1997.
[AAR G. Andrews, R. Askey, R. Roy: Spe ial Fun tions, En y lopedia of Math-
emati s and its Appli ations, Vol. 71, Cambridge U. Press, 1999.
[As R. Askey: Ramanujan and hypergeometri and basi hypergeometri series,
Ramanujan Internat. Symposium on Analysis, De ember 26-28, 1987, ed. by
N. K. Thakare, 1-83, Pune, India, Russian Math. Surveys 451 (1990), 37-86.
[BPV1 R. Balasubramanian, S. Ponnusamy, and M. Vuorinen: Fun tional in-
equalities for the quotients of hypergeometri fun tions, J. Math. Anal. Appl.
218 (1998), 256268.
[BPV2 R. Balasubramanian, S. Ponnusamy, and M. Vuorinen: On hypergeo-
metri fun tions and fun tion spa es, J. Comp. Appl. Math. (to appear).
[BPSV R. Balasubramanian, S. Ponnusamy, Sunanda Naik, and M. Vuori-
nen: Elliott's identity and hypergeometri fun tions, Preprint 284, April
2001, University of Helsinki, 22 pp.
[B R. W. Barnard: On appli ations of hypergeometri fun tions, Continued
fra tions and geometri fun tion theory (CONFUN)(Trondheim, 1997), J.
Comp. Appl. Math. 105 (1999), no. 12, 18.
[BPR1 R. W. Barnard, K. Pear e, and K. C. Ri hards: A monotoni ity prop-
erty involving 3F2 and omparisons of the lassi al approximations of ellipti al
ar length, SIAM J. Math. Anal. 32 (2000), 403419 (ele troni ).
[BPR2 R. W. Barnard, K. Pear e, and K. C. Ri hards: An inequality involving
the generalized hypergeometri fun tion and the ar length of an ellipse, SIAM
J. Math. Anal. 31 (2000), no. 3, 693699 (ele troni ).
[BPS R. W. Barnard, K. Pear e, and L. S hovane : Inequalities for the
perimeter of an ellipse, J. Math. Anal. Appl. 260 (2001), 295306.
[Be A. F. Beardon: The hyperboli metri of a re tangle, Ann. A ad. S i. Fenn.
Ser. A I 26 (2001), 401407.
20 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
[BP C. Berg and H. Pedersen: A ompletely monotone fun tion related to the
gamma fun tion, J. Comp. Appl. Math. 133 (2001), 219230.
[Be1 B. C. Berndt: Ramanujan's Notebooks, Vol. II, Springer-Verlag, Berlin, 1989.
[Be2 B. C. Berndt: Ramanujan's Notebooks, Vol. IV, Springer-Verlag, Berlin,
1993.
[BBG B. C. Berndt, S. Bhargava, and F. G. Garvan: Ramanujan's theories
of ellipti fun tions to alternative bases, Trans. Amer. Math. So . 347 (1995),
41634244.
[BCK B. C. Berndt, Y.-S. Choi, and S.-Y. Kang: The problems submitted by
Ramanujan to the Journal of the Indian Mathemati al So iety. Continued fra -
tions: from analyti number theory to onstru tive approximation (Columbia,
MO, 1998), 1556, Contemp. Math., 236, Amer. Math. So ., Providen e, RI,
1999.
[BK M. Biernaki and J. Krzy»: On the monotoni ity of ertain fun tionals in
the theory of analyti fun tions, Ann. Univ. M. Curie-Skªodowska 2 (1955),
135147.
[BB1 J. M. Borwein and P. B. Borwein: Pi and the AGM, Wiley, New York,
1987.
[BB2 J. M. Borwein and P. B. Borwein: Inequalities for ompound mean itera-
tions with logarithmi asymptotes, J. Math. Anal. Appl. 177 (1993), 572582.
[Br P. Bra ken: An arithmeti -geometri mean inequality. Exposition. Math. 19
(2001), 273279.
[BF P. F. Byrd and M. D. Friedman: Handbook of Ellipti Integrals for En-
gineers and S ientists, 2nd ed., Grundlehren Math. Wiss., Vol. 67, Springer-
Verlag, Berlin, 1971.
[CG B. C. Carlson and J. L. Gustafson: Asymptoti approximations for sym-
metri ellipti integrals, SIAM J. Math. Anal. 25 (1994), 288303.
[CV B. C. Carlson and M. Vuorinen: An inequality of the AGM and the
logarithmi mean, SIAM Rev. 33 (1991), Problem 91-17, 655.
[Ch I. Chavel: Riemannian Geometry A Modern Introdu tion, Cambridge
Tra ts in Math. 108, Cambridge Univ. Press, 1993.
[C S.-Y. Chung: Fun tional means and harmoni fun tional means, Bull. Aus-
tral. Math. So . 57 (1998), 207220.
[De D. W. DeTemple: A qui ker onvergen e to Euler's onstant, Amer. Math.
Monthly 100(5) (1993), 468470.
[Du P. L. Duren: The Legendre relation for ellipti integrals, in Paul Halmos:
Celebrating 50 years of Mathemati s, ed. by J. H. Ewing and F. W. Gehring,
Springer-Verlag, New York, 1991, pp. 305315.
[EL Á. Elbert and A. Laforgia: On some properties of the gamma fun tion,
Pro . Amer. Math. So . 128 (2000), 26672673.
[E E. B. Elliott: A formula in luding Legendre's EK′+ KE′
− KK′=
1
2π,
Messenger of Math. 33 (1904), 3140.
[G W. Gauts hi: The in omplete gamma fun tions sin e Tri omi. Tri omi's
ideas and ontemporary applied mathemati s (Rome/Turin, 1997), 203237,
Atti Convegni Lin ei, 147, A ad. Naz. Lin ei, Rome, 1998.
[H M. Hirs hhorn: A new formula for π, Gaz. Austral. Math. So . 25 (1998),
8283.
[K1 E. A. Karatsuba: On the omputation of the Euler onstant γ, Numer.
Algorithms 24 (2000), 8387.
[K2 E. A. Karatsuba: On the asymptoti representation of the Euler gamma
fun tion by Ramanujan, J. Comp. Appl. Math. 135.2 (2001), 225240.
[KV E. A. Karatsuba and M. Vuorinen: On hypergeometri fun tions and
generalizations of Legendre's relation, J. Math. Anal. Appl. 260 (2001), 623-
640.
[KlR D. A. Klain and G.-C. Rota: A ontinuous analogue of Sperner's theorem.
Comm. Pure Appl. Math. 50 (1997), no. 3, 205223.
TOPICS IN SPECIAL FUNCTIONS 21
[Ku R. Kühnau: Eine Methode, die Positivität einer Funktion zu prüfen, Z.
Angew. Math. Me h. 74 (1994), no. 2, 140143.
[Kum E. E. Kummer: Über die hypergeometris he Reihe, J. Reine Angew. Math.
15 (1836), 3983 and 127172.
[LS A. Laforgia and S. Sismondi: Some fun tional inequalities for omplete
ellipti integrals. Rend. Cir . Mat. Palermo (2) 41 (1992), no. 2, 302308.
[LV O. Lehto and K. I. Virtanen: Quasi onformal Mappings in the Plane, 2nd
ed., Springer-Verlag, New York, 1973.
[Le D. C. van Leijenhorst, Algorithms for the approximation of π, Nieuw
Ar hief Wisk. 14 (1996), 255274.
[Lu A. Lupas: Some BBP-fun tions, Manus ript 2000.
[PV1 S. Ponnusamy and M. Vuorinen: Asymptoti expansions and inequalities
for hypergeometri fun tions, Mathematika 44 (1997), 278301.
[PV2 S. Ponnusamy and M. Vuorinen: Univalen e and onvexity properties
for onuent hypergeometri fun tions, Complex Variables Theory Appl. 36
(1998), 73-97.
[PV3 S. Ponnusamy and M. Vuorinen: Univalen e and onvexity properties for
Gaussian hypergeometri fun tions, Ro ky Mountain J. Math. 31 (2001), 327
353.
[PBM A. P. Prudnikov, Yu. A. Bry hkov, and O. I. Mari hev: Integrals and
Series, Vol. 3: More Spe ial Fun tions, trans. from the Russian by G. G. Gould,
Gordon and Brea h, New York, 1988; see Math. Comp. 65 (1996), 13801384
for errata.
[Q S.-L. Qiu: The proof of a onje ture on the rst ellipti integrals (in Chinese),
J. Hangzhou Inst. of Ele t. Eng. 3 (1993), 2936.
[QS S.-L. Qiu and J.-M. Shen: On two problems on erning means, J. Hangzhou
Inst. of Ele t. Eng. 3 (1997), 17.
[QVa S.-L. Qiu and M. K. Vamanamurthy: Sharp estimates for omplete ellipti
integrals, SIAM J. Math. Anal. 27 (1996), 823834.
[QVV1 S.-L. Qiu, M. K. Vamanamurthy, and M. Vuorinen: Some inequalities
for the Hers h-Puger distortion fun tion, J. Inequalities Appli . 4 (1999), 115
139.
[QVV2 S.-L. Qiu, M. K. Vamanamurthy, and M. Vuorinen: Some inequalities
for the growth of ellipti integrals, SIAM J. Math. Anal. 29 (1998), 12241237.
[QVu1 S.-L. Qiu and M. Vuorinen: Landen inequalities for hypergeometri fun -
tions, Nagoya Math. J. 154 (1999), 3156.
[QVu2 S.-L. Qiu and M. Vuorinen: Innite produ ts and the normalized quotients
of hypergeometri fun tions, SIAM J. Math. Anal. 30 (1999), 10571075.
[QVu3 S.-L. Qiu and M. Vuorinen: Dupli ation inequalities for the ratios of hy-
pergeometri fun tions, Forum Math. 12 (2000), 109133.
[R1 E. D. Rainville: Spe ial Fun tions, Ma Millan, New York, 1960.
[R2 E. D. Rainville: Intermediate Dierential Equations, 2nd ed., Ma millan,
1964.
[Ra1 S. Ramanujan: The Lost Notebook and Other Unpublished Papers, Introdu -
tion by G. Andrews, Springer-Verlag, New York, 1988.
[Ra2 S. Ramanujan: Colle ted papers, ed. by G. S. Hardy, P. V. Seshu Aiyar, and
B. M. Wilson, AMS Chelsea Publ. 2000, with a ommentary by B. Berndt,
357426.
[Sa E. Salamin: Computation of π using arithmeti -geometri mean, Math.
Comp. 135 (1976), 565570.
[S1 J. Sándor: On ertain inequalities for means, J. Math. Anal. Appl. 189
(1995), 602606.
[S2 J. Sándor: On ertain inequalities for means, II, J. Math. Anal. Appl. 199
(1996), 629635.
[S3 J. Sándor: On ertain inequalities for means, III, Ar h. Math. (Basel) 76
(2001), 3440.
22 G. D. ANDERSON, M. K. VAMANAMURTHY, AND M. VUORINEN
[TY S. R. Tims and J. A. Tyrell: Approximate evaluation of Euler's onstant,
Math. Gaz. 55 (1971), 6567.
[T Gh. Toader: Some mean values related to the arithmeti -geometri mean, J.
Math. Anal. Appl. 218 (1998), 358368.
[UK J. S. Ume and Y.-H. Kim: Some mean values related to the quasi-arithmeti
mean, J. Math. Anal. Appl. 252 (2000), 167176.
[VV M. K. Vamanamurthy and M. Vuorinen: Inequalities for means. J. Math.
Anal. Appl. 183 (1994), 155166.
[V1 M. Vuorinen: Geometri properties of quasi onformal maps and spe ial fun -
tions, IIII, Bull. So . S i. Lett. ód¹ Sér. Re h. Déform. 24 (1997), 758.
[V2 M. Vuorinen: Hypergeometri fun tions in geometri fun tion theory, in Pro-
eedings of the Spe ial Fun tions and Dierential Equations, pp. 119-126, ed.
by K. R. Srinivasa, R. Jagannathan, and G. Van der Jeugy, Allied Publishers,
New Delhi, 1998.
[Y R. M. Young: Euler's onstant, Math. Gaz. 75(472) (1991), 187190.
ANDERSON:
Department of Mathemati s
Mi higan State University
East Lansing, MI 48824, USA
email: andersonmath.msu.edu
FAX: +1-517-432-1562
VAMANAMURTHY:
Department of Mathemati s
University of Au kland
Au kland, NEW ZEALAND
email: vamanamumath.au kland.nz
FAX: +649-373-7457
VUORINEN:
Department of Mathemati s
University of Helsinki
P.O. Box 4 (Yliopistonkatu 5)
FIN-00014, FINLAND
e-mail: vuorinen s .fi
FAX: +358-9-19123213