You are on page 1of 20

TRANSACTIONS OF THE

AMERICAN MATHEMATICAL SOCIETY


Volume 352, Number 6, Pages 2581–2600
S 0002-9947(00)02386-2
Article electronically published on February 14, 2000

PARTITIONS INTO PRIMES

YIFAN YANG

Abstract. We investigate
P the asymptotic
Q∞ behavior of the partition function
pΛ (n) defined by ∞ n
n=0 pΛ (n)x =
m −Λ(m) , where Λ(n) denotes
m=1 (1 − x )
the von Mangoldt
p function. Improving a result of Richmond, we show that

log pΛ (n) = 2 ζ(2)n + O( n exp{−c(log n)(log2 n)−2/3 (log3 n)−1/3 }), where
c is a positive constant and logk denotes the k times iterated logarithm. We
also show that the error term can be improved to O(n1/4 ) if and only if the
Riemann Hypothesis holds.

1. Introduction
The asymptotic behavior of partition functions has been extensively studied in
the literature. The most famous result is the asymptotic formula
1 √
p(n) ∼ √ eπ 2n/3 (n → ∞)
(4 3)n
for the ordinary partition function p(n), proved in 1918 by Hardy and Ramanujan
[3]. The asymptotic behavior of more general partition functions has been studied
by many authors, including Ingham [4], Kohlbecker [6], Meinardus [7], [8], Roth
and Szekeres [12], Schwarz [13], [14], and Richmond [10], [11].
Of particular interest are functions related to partitions into primes. As an
application of an asymptotic formula for general partition functions, Roth and
Szekeres [12] showed that the number qP (n) of partitions of n into distinct primes
satisfies
r  1/2   
2 n log log n
log qP (n) = π 1+O .
3 log n log n
A similar, but in some ways more natural, partition function is the function
pΛ (n) defined by
X∞ ∞
Y
pΛ (n)xn = (1 − xm )−Λ(m) ,
n=0 m=1
where
(
log p, m = pr ,
Λ(m) =
0, else,
is the von Mangoldt function. The function pΛ (n) represents a weighted count of
the number of partitions of n into prime powers. This weighted partition function
was first introduced and studied in 1950 by Brigham [1], who proved a conditional

Received by the editors March 3, 1998.


2000 Mathematics Subject Classification. Primary 11P82; Secondary 11M26, 11N05.

2000
c American Mathematical Society
2581
2582 YIFAN YANG

result under the assumption of the Riemann Hypothesis. More recently, Richmond
[11], using an asymptotic formula for general partition functions (see [10]) and
Vinogradov’s zero-free region for the Riemann zeta function, obtained the following
unconditional result.
Theorem A (Richmond). There exists a positive constant c such that for all
sufficiently large n
p  n o
(1.1) log pΛ (n) = 2 ζ(2)n1/2 1 + O exp(−c(log n)4/7 (log log n)−3/7 ) .

Richmond also proved a conditional result.


Theorem B (Richmond). Let θ be the least upper bound for the real parts of the
non-trivial zeros of the Riemann zeta function. Then
p
(1.2) log pΛ (n) = 2 ζ(2)n1/2 + O(nθ/2 ).
In particular, if the Riemann Hypothesis is true, then
p
(1.3) log pΛ (n) = 2 ζ(2)n1/2 + O(n1/4 ).
In our first result we show that the error term in Richmond’s unconditional result
(1.1) can be substantially improved.
Theorem 1. There exists a positive constant c such that for all sufficiently large
n
   
p c log n
(1.4) log pΛ (n) = 2 ζ(2)n 1/2
1 + O exp − ,
(log2 n)2/3 (log3 n)1/3
where logk denotes the k times iterated logarithm.
Note that the exponential factor in (1.4) is much smaller than the corresponding
factor in the error term of the prime number theorem obtained by using the Vino-
gradov zero-free region, namely exp(−(log n)3/5− ). The reason for this unexpect-
edly small error term lies inPthe fact that log pΛ (n) behaves in many respects
P more

like the power series f (x) = n=1 Λ(n)xn than the partial sum Ψ(u) = n≤u Λ(n);
indeed, it would not be hard to show that, as x → 1−, f (x) differs from its approx-
imation 1/(1 − x) by a similarly small error term.
While the estimate of Theorem A can be substantially improved, our next result
shows that Theorem B is best possible.
Theorem 2. We have
p
log pΛ (n) − 2 ζ(2)n1/2 = Ω± (n1/4 ).
Our final result gives a converse to Theorem B.
Theorem 3. Let θ be the least upper bound for the real parts of the zeros of the
Riemann zeta function, and let θ0 be the greatest lower bound for all α for which
p
log pΛ (n) = 2 ζ(2)n1/2 + O(nα/2 )
as n → ∞. Then θ ≤ θ0 .
Combining this result with Theorem B yields the following corollary.
PARTITIONS INTO PRIMES 2583

Corollary. Let θ and θ0 be defined as in Theorem 3. Then θ = θ0 . In particular,


the Riemann Hypothesis is true if and only if
p
log pΛ (n) = 2 ζ(2)n1/2 + O (n1/4+ )
for all  > 0.
Let
X
PΛ (u) = pΛ (n)
n≤u

be the summatory function of pΛ (n). In Section 4 we show that Theorems 1-3 are
true if and only if the corresponding statements with n replaced by u and pΛ (n)
replaced by PΛ (u) are true. Hence it suffices to prove Theorems 1-3 with PΛ (u)
instead of pΛ (n).
Our main tools for proving these results are Abelian and Tauberian theorems
that relate estimates for PΛ (u) to estimates for the Laplace-Stieltjes transform of
PΛ (u), defined by
Z ∞
FΛ (x) = e−xu dPΛ (u).
0
We will state and prove these results in Section 2. In Section 3 we establish some
lemmas relating the behavior of general weighted partition functions pw (n) defined
by
X∞ Y∞
pw (n)xn = (1 − xm )−w(m)
n=0 m=1
to analytic properties of the Dirichlet series
X∞
w(n)
fw (s) = ,
n=1
ns
where w(m) is a non-negative function defined on the set of positive integers. Our
methods here are, to some extent, similar to those in Meinardus [7]. However,
Meinardus made much stronger assumptions on the analytic properties of fw (s).
These assumptions are not satisfied in the case w(n) = Λ(n), so that Meinardus’
results are not applicable to pΛ (n). In Section 4 we complete the proof of Theorems
1-3. In Section 5 we use one of our Tauberian results to give a new proof of Theorem
B that is simpler and more elementary than the original proof of Richmond.

2. Some Abelian and Tauberian Results


Our first result is an elementary Abelian result which generalizes and extends a
result of Freiman [2, p. 276].
Proposition 1. Suppose that P (u) is a non-negative and non-decreasing function
satisfying P (t) = O (et ) for every  > 0. For x > 0 let
Z ∞
F (x) = e−xu dP (u)
0
be the Laplace-Stieltjes transform of P (u). Suppose that for some constants A > 0
and 0 < a < 1 the inequality
(2.1a) log P (u) ≥ Aua − r(u)
2584 YIFAN YANG

holds for all sufficiently large u, where r(u) is a positive differentiable function
satisfying
(R1) r(u)u−a → 0 monotonically as u → ∞,
and
(R2) r(u)/(log u) → ∞ monotonically as u → ∞.
Then
(2.2a) log F (x) ≥ Bx−b − r(B 0 x−b−1 )
for all sufficiently small x, where
(2.3) B = (Aa)1/(1−a) (1 − a)/a, b = a/(1 − a), B 0 = Bb.
Similarly, if
(2.1b) log P (u) ≤ Aua + r(u)
for all sufficiently large u, then with the same notation
(2.2b) log F (x) ≤ Bx−b + 2r(B 0 x−b−1 )
for all sufficiently small x.
As a simple consequence of Proposition 1, we have the following corollary.
Corollary. Let P (u), F (x), A, a, B, b and B 0 be given as in Proposition 1. Sup-
pose that r(u) satisfies conditions (R1) and (R2). Then
log F (x) − Bx−b = Ω+ (r(B 0 x−b−1 ))
as x → 0 implies that
log P (u) − Aua = Ω+ (r(u))
as u → ∞. The same statement holds if Ω+ is replaced by Ω− .
Proof of Proposition 1. Suppose first that (2.1a) holds for sufficiently large u. By
the monotonicity of P (t) and the assumption that P (t) is non-negative, we have
for all x > 0 and all u > 0,
Z ∞ Z ∞
−xt
(2.4) F (x) = e dP (t) = x e−xt P (t) dt
Z ∞
0 0
Z ∞
−xt
≥x e P (t) dt ≥ x e−xt P (u) dt = e−xu P (u).
u u
(Note that the convergence of the integral defining F (x) is ensured by the assump-
tion that P (t) = O (et ) for every  > 0.)
By (2.1a) it follows that
(2.5) F (x) ≥ exp{−xu + Aua − r(u)}
for all sufficiently large u and all x > 0. We choose u = ux to maximize −xu + Aua,
i.e., we let ux be defined by
(2.6) x = Aau−(1−a)
x .
If x is sufficiently small, then ux will be large enough for (2.5) to hold. We note
that
(2.7) xux = x(Aax−1 )1/(1−a) = B 0 x−a/(1−a) = B 0 x−b
PARTITIONS INTO PRIMES 2585

and
(2.8) Auax = A(Aax−1 )a/(1−a) = (Aa)1/(1−a) a−1 x−b = B 0 a−1 x−b .
Substituting these expressions into the right-hand side of (2.5), we obtain
log F (x) ≥ B 0 (−1 + a−1 )x−b − r(B 0 x−b−1 ) = Bx−b − r(B 0 x−b−1 ),
which proves (2.2a). It remains to prove that (2.1b) implies (2.2b).
Let u0 be a positive constant such that (2.1b) and conditions (R1)-(R2) are
satisfied for u ≥ u0 . Given a positive number x, we let ux be defined by (2.6) and
assume that x is small enough so that ux ≥ 2u0 . We write
(Z Z (1−µ)ux Z (1+µ)ux Z ∞ )
u0
(2.9) F (x) = x + + + e−xt P (t) dt
0 u0 (1−µ)ux (1+µ)ux

= x{I1 + I2 + I3 + I4 },
where µ is a parameter to be chosen so that xI3 has the same order of magnitude
as F (x). It turns out that the optimal choice for µ is
q
(2.10) µ = µ(ux ) = K u−ax r(ux ),

where K is a positive constant to be chosen later in terms of A and a. We note


that condition (R1) implies that µ(ux ) → 0 as ux → ∞. In particular, we have
0 < µ < 1/2 if x is sufficiently small in terms of K, which we will henceforth
assume. Ru
The integral I1 is bounded by 0 0 P (t) dt and thus of order O(1). To estimate
I4 , we define φ(t) for t ≥ u0 by
φ(t) = −xt + Ata + r(t).
Since, by condition (R1), r(t)t−a is monotonically decreasing to 0, we have r0 (t)t−a −
−(1−a) −(1−a)
ar(t)t−a−1 ≤ 0 and r(t)t−1 ≤ r(ux )u−a x ux = K −2 µ2 ux for all t ≥ ux .
Thus, for all t ≥ (1 + µ)ux we obtain
φ0 (t) = −x + Aat−(1−a) + r0 (t) ≤ −x + Aat−(1−a) + ar(t)t−1
≤ −x + Aa(1 + µ)−(1−a) u−(1−a)
x + K −2 aµ2 u−(1−a)
x .
By (2.6), the last expression is equal to (−1 + (1 + µ)−(1−a) + O(µ2 ))x. Since
(1+µ)−(1−a) = 1−(1−a)µ+O(µ2), we have −1+(1+µ)−(1−a)+O(µ2 ) ≤ −(1−a)µ/2
for all sufficiently small x. Hence for all sufficiently small x and t ≥ (1 + µ)ux we
obtain
1−a
φ0 (t) ≤ − xµ.
2
It follows that
Z ∞ Z ∞
2 2
xI4 ≤ x eφ(t) dt ≤ − eφ(t) φ0 (t) dt = eφ((1+µ)ux )
(1+µ)ux (1 − a)µ (1+µ)ux (1 − a)µ
2
= exp {−x(1 + µ)ux + A(1 + µ)a uax + r((1 + µ)ux )} .
(1 − a)µ
We substitute (2.7) and (2.8) into the last expression and note that condition (R1)
implies that r((1 + µ)ux )((1 + µ)ux )−a ≤ r(ux )u−a
x , i.e.,

(2.11) r((1 + µ)ux ) ≤ (1 + µ)a r(ux )


2586 YIFAN YANG

for all sufficiently small x. This yields


2  
xI4 ≤ exp B 0 x−b −(1 + µ) + (1 + µ)a a−1 + (1 + µ)a (Aa)−1 r(ux )u−a
(1 − a)µ x

 1 
≤ exp B 0 x−b (a−1 − 1) + (a − 1)µ2 + O(µ3 ) + (1 + µ)a (Aa)−1 r(ux )u−a x
2
+ O(log µ−1 )
 1 
= exp Bx−b 1 − aµ2 + O(µ3 ) + (1 + O(µ))A−1 (1 − a)−1 r(ux )u−a x
2
+ O(log µ−1 ) ,
where we have used the identity B 0 (a−1 − 1) = B which follows from (2.3). Ex-
pressing r(ux )u−a
x in terms of µ using (2.10), the last expression becomes
n  a  o
exp Bx−b 1 − µ2 − K −2 A−1 (1 − a)−1 + O(µ3 ) + O(xb log µ−1 ) .
2
By (2.8), (2.10) and condition (R2) we have
xb log µ−1  u−a −a 2
x log ux = o(ux r(ux )) = o(µ ).

Choosing now K sufficiently large in terms of A and a yields



(2.12) xI4 ≤ exp Bx−b (1 − aµ2 /4)
for all sufficiently small x. By a similar argument we have

(2.13) xI2 ≤ exp Bx−b (1 − aµ2 /4)
for all sufficiently small x.
We now estimate I3 . By (2.1b) we have
Z (1+µ)ux Z (1+µ)ux
e−xt P (t) dt ≤ x e−xt+At
a
+r(t)
xI3 = x dt
(1−µ)ux (1−µ)ux

We observe that t = ux maximizes −xt + Ata and, by condition (R2), r(t) ≤


r((1 + µ)ux ) for all t ≤ (1 + µ)ux . Hence, by (2.7), (2.8) and (2.11),
(2.14) xI3 ≤ 2xµux exp {−xux + Auax + r((1 + µ)ux )}

≤ 2xµux exp Bx−b + (1 + µ)a r(B 0 x−b−1 )
for all sufficiently small x. Combining (2.9), (2.12), (2.13) and (2.14), and noting
that, by (2.7) and (2.10),
 q 

log(xµux ) = log B 0 x−b K u−ax r(u x ) = O log(uax r(ux )) = O(log x−1 ),

we finally obtain

F (x) ≤ O(x) + O(exp Bx−b − aµ2 /4 )

+ exp Bx−b + (1 + µ)a r(B 0 x−b−1 ) + O(log x−1 )

= exp Bx−b + (1 + µ)a r(B 0 x−b−1 ) + O(log x−1 )

≤ exp Bx−b + 2r(B 0 x−b−1 )
for all sufficiently small x, where in the last step we have used the assumption (R2).
This yields the desired estimate (2.2b).
Our next result is a Tauberian counterpart to Proposition 1.
PARTITIONS INTO PRIMES 2587

Proposition 2. Let P (u) and F (x) be defined as in Proposition 1. Suppose that


for some constants B, b > 0 and B 0 = Bb the inequality

(2.15) log F (x) − Bx−b ≤ r(B 0 x−b−1 )
holds for all sufficiently small x, where r(u) is a positive differentiable function
satisfying conditions (R1) and (R2) in Proposition 1 with a = b/(1 + b). Then
p
(2.16) −C ua r(u) ≤ log P (u) − Aua ≤ r(u)
for all sufficiently large u, where A and a are determined by (2.3) and C is a
constant depending on B and b.
Proof. Suppose that (2.15) holds for all sufficiently small x. We first prove the
upper bound for log P (u).
Given a positive number u, by (2.4) and assumption (2.15) we have

(2.17) P (u) ≤ exu F (x) ≤ exp xu + Bx−b + r(B 0 x−b−1 )
for all sufficiently small x. We choose
(2.18) x = xu = Aau−(1−a) ,
where A and a are determined by (2.3), and note that with this choice of x we have
u = ux , where ux is given by (2.6). We assume that u is large enough so that (2.17)
holds for x = xu . Using (2.7) and (2.8) with xu and u in place of x and ux , we see
that
 
1−a
(2.19) xu u + Bx−b
u = B 0 −b
xu (1 + b −1
) = Aau a
x 1 + = Auax
a
and
(2.20) B 0 x−b−1
u = u.
Thus, we obtain the upper bound
(2.21) P (u) ≤ exp{xu u + Bx−b 0 −b−1
u + r(B xu )} = exp{Aua + r(u)}
for all sufficiently large u. This implies the upper bound in (2.16). It remains to
prove the lower bound.
Let u0 be a large positive constant such that (2.21) holds for u ≥ u0 . Assuming
that u ≥ 2u0 , we split the integral defining F (xu ) into four parts as before. Let
I1 , I2 , I3 and I4 pbe defined by (2.9) with x = xu and ux replaced by u. We
let µ = µ(u) = K u−a r(u) be defined as in (2.10), where K is a large positive
constant to be chosen later. Using the upper bound (2.21) and arguing exactly as
in the proof of Proposition 1, we see that the upper bounds (2.12) and (2.13) for
I4 and I2 remain valid for sufficiently large u with xu and u in place of x and ux .
We note that, by (2.3), (2.8) and the definition of µ,
1 1 1 Aa 1
Baµ2 x−b
u = BaK r(u)u
2 −a −b
xu = BaK 2 r(u) 0 = Aa(1 − a)K 2 r(B 0 x−b−1
u ).
4 4 4 B 4
We now choose the constant K large enough so that Aa(1 − a)K 2 /4 ≥ 2. Then by
(2.12) and assumption (2.15), we have
 
1 
xu I4 ≤ exp Bx−b u − Baµ 2 −b
x u ≤ exp Bx−b 0 −b−1
u − 2r(B xu )
4

≤ F (xu ) exp −r(B 0 x−b−1
u ) = o(F (xu )).
2588 YIFAN YANG

Similarly, we see that xu I2 = o(F (xu )). Hence xu I3 ≥ F (xu )(1 + o(1)). On the
other hand, bounding the integral I3 trivially, we obtain
xu I3 ≤ 2µuxu e−xu (1−µ)u P ((1 + µ)u).
It follows that
P ((1 + µ)u) ≥ F (xu ) exp {xu (1 − µ)u − log(2µuxu ) + o(1)} .
By (2.10), (2.18) and conditions (R1) and (R2) we have
log(2µuxu )  log u  r(u).
Thus, using the bound (2.15) for F (xu ), we obtain

P ((1 + µ)u) ≥ exp Bx−bu − r(B xu
0 −b−1
) + xu (1 − µ)u + O(r(u)) .
Substituting (2.18), (2.19) and (2.20) into the last expression, we see that
log P ((1 + µ)u) ≥ Aua − Aaµua + O(r(u)) = A(1 + µ)a ua + O(µua ) + O(r(u)).
p
The last error term can be omitted since r(u) = o( r(u)ua ) = o(µua ) by
p condition
(R1)
p and (2.10). Moreover, by condition (R2), we have µu a
= K r(u)ua ≤
K r((1 + µ)u)(1 + µ)a ua . It follows that
p
log P ((1 + µ)u) ≥ A(1 + µ)a ua − C r((1 + µ)u)(1 + µ)a ua
for all sufficiently large u. Since (1 + µ(u))u is a continuous function of u and
tends to infinity when u → ∞, for all sufficiently large v, there exists a u such that
v = (1 + µ(u))u. Hence we have
p
log P (v) ≥ Av a − C v a r(v)
for all sufficiently large v. This completes the proof of the proposition.
Proposition 2 is not a complete converse to Proposition 1, as the lower bound
in (2.16) is weaker than the lower bound for log P (u) in Proposition 1. Our next
result gives, under a stronger hypothesis, a complete converse.
Proposition 3. Let P (u) and F (x) be defined as in Proposition 1. Suppose that
for some constants B, b > 0 and B 0 = Bb the inequality

(2.15) log F (x) − Bx−b ≤ r(B 0 x−b−1 )

holds for all sufficiently small x, where r(u) is a positive differentiable function
satisfying conditions (R1) and (R2) in Proposition 1 with a = b/(1 + b). Suppose
that, in addition, r(u) satisfies
(R3) r(u)  ua/2
as u → ∞. Suppose further that the function G(x) = log F (x) satisfies

(2.22) G0 (x) = −B 0 x−b−1 + O x−1 r(B 0 x−b−1 )
and
(2.23) G00 (x)  x−b−2
as x → 0. Then
(2.24) −Cr(u) ≤ log P (u) − Aua ≤ r(u)
PARTITIONS INTO PRIMES 2589

for all sufficiently large u, where A and a are determined by (2.3), and C is a
positive constant depending on B, b and the constants implicit in (2.22), (2.23) and
condition (R3).
Proof. The upper bound in (2.24) follows from Proposition 2. Therefore it remains
to prove the lower bound in (2.24). To this end we use a method of Odlyzko [9].
Given a large number u, we let 0 < µ1 < µ2 < 1 be positive numbers to be
chosen later as functions of u. Set u0 = u, u1 = (1 − µ1 )u, u2 = (1 − µ2 )u, and let
x0 , x1 and x2 be defined by (2.18) with u = u0 , u1 and u2 respectively. Thus, we
have x1 = (1 − µ1 )−(1−a) x0 and x2 = (1 − µ2 )−(1−a) x0 . We consider the function
h(t) = exp{x1 u1 − x1 t} − exp{x0 u0 + x1 u1 − x1 u0 − x0 t}
− exp{x2 u2 + x1 u1 − x1 u2 − x2 t}.
Since x0 < x1 < x2 , we have for t ≥ u0
x1 u1 − x1 t ≤ x1 u1 − x1 t + (x1 − x0 )(t − u0 ) = x0 u0 + x1 u1 − x1 u0 − x0 t,
and for t ≤ u2
x1 u1 − x1 t ≤ x1 u1 − x1 t + (x2 − x1 )(u2 − t) = x2 u2 + x1 u1 − x1 u2 − x2 t.
Thus h(t) ≤ 0 for t ≤ u2 and t ≥ u0 . Now we let
Z ∞
H= h(t) dP (t)
0
= ex1 u1 F (x1 ) − ex0 u0 +x1 u1 −x1 u0 F (x0 ) − ex2 u2 +x1 u1 −x1 u2 F (x2 ).
Then we see that
H ≤ (P (u0 ) − P (u2 )) × max h(t)
u2 ≤t≤u0
x1 u1 −x1 t
≤ P (u0 ) × max e = P (u0 )ex1 (u1 −u2 ) .
u2 ≤t≤u0

Hence, we have

(2.25) P (u0 ) ≥ ex1 u1 F (x1 ) − ex0 u0 +x1 u1 −x1 u0 F (x0 )

− ex2 u2 +x1 u1 −x1 u2 F (x2 ) e−x1 (u1 −u2 ) .
We now show that, with a suitable choice of µ1 and µ2 , ex0 u0 +x1 u1 −x1 u0 F (x0 ) +
ex2 u2 +x1 u1 −x1 u2 F (x2 ) ≤ ex1 u1 F (x1 )/2, and thus the last expression has the same
order of magnitude as F (x1 )ex1 u2 .
By assumptions (2.22) and (2.23), there are positive constants C1 and C2 such
that
−B 0 x−b−1 − C1 x−1 r(B 0 x−b−1 ) ≤ G0 (x) ≤ −B 0 x−b−1 + C1 x−1 r(B 0 x−b−1 )
and
G00 (x) ≥ C2 x−b−2
when x is sufficiently small. Thus, Taylor’s formula yields that
1
(2.26) G(x1 ) − G(x0 ) ≥ G0 (x0 )(x1 − x0 ) + min G00 (ξ)(x1 − x0 )2
2 x0 ≤ξ≤x1
≥ (x1 − x0 )(−B 0 x−b−1
0 − C1 x−1 0 −b−1
0 r(B x0 ))
1
+ C2 x−b−21 (x1 − x0 )2 .
2
2590 YIFAN YANG

We now choose µ1 to be of the form µ1 = Kxb0 r(B 0 x−b−1


0 ), where K is a large
constant independent of u0 . We note that, by (2.8) and (2.20), µ1 is also equal to
KB 0 (Aa)−1 u−a
0 r(u0 ). Thus, by condition (R3), we have

−a/2
(2.27) µ1 ≥ KC3 u0

for all sufficiently large u0 , where C3 is a positive constant depending on the con-
stant implicit in condition (R3). Moreover, by condition (R1), this choice of µ1
satisfies µ1 → 0 as u0 → ∞. It follows that when u0 is sufficiently large,
(
−(1−a) ≤ 2(1 − a)µ1 x0 ,
x1 − x0 = x0 ((1 − µ1 ) − 1)
≥ (1 − a)µ1 x0 .

Hence, recalling that, by (2.20), B 0 x−b−1


0 = u0 , we have from (2.26)

G(x1 ) − G(x0 ) ≥ −(x1 − x0 )u0 − 2C1 (1 − a)µ1 r(u0 )


1
+ C2 x−b−2
0 (1 − µ1 )(1−a)(b+2) (1 − a)2 µ21 x20
2
1
≥ −(x1 − x0 )u0 − 2C1 (1 − a)µ1 r(u0 ) + C2 (1 − a)2 µ21 x−b
0
4
for all sufficiently large u0 . We now choose K sufficiently large in terms of B, b, C1
and C2 so that the second term on the right-hand side of the last expression is less
than one half of the last term uniformly for all sufficiently large u0 , and so that the
last term is at least 2 log 4 (which is possible by (2.8) and (2.27)), i.e., so that
1
2C1 (1 − a)µ1 r(B 0 x−b−1
0 )≤ C2 (1 − a)2 µ21 x−b
0
8
and
1
C2 (1 − a)2 µ21 x−b
0 ≥ log 4
8
for all sufficiently large u0 . Then it follows that
1
G(x1 ) − G(x0 ) ≥ −(x1 − x0 )u0 − 2C1 (1 − a)µ1 r(B 0 x−b−1
0 ) + C2 (1 − a)2 µ21 x−b
0
4
1
≥ −(x1 − x0 )u0 + C2 (1 − a)2 µ21 x−b
0
8
≥ −(x1 − x0 )u0 + log 4.
Hence

(2.28) ex1 u1 F (x1 ) ≥ 4ex0 u0 +x1 u1 −x1 u0 F (x0 )


for all sufficiently large u0 . Similarly, if we choose µ2 = 2µ1 , then

(2.29) ex1 u1 F (x1 ) ≥ 4ex2 u2 +x1 u1 −x1 u2 F (x2 )

for all sufficiently large u0 . Therefore, by (2.25), (2.28) and (2.29), we have for all
sufficiently large u0
1 x1 u1 1
P (u0 ) ≥ e F (x1 )e−(u1 −u2 )x1 = F (x1 )ex1 u2 = exp {G(x1 ) + x1 u2 + O(1)} .
2 2
PARTITIONS INTO PRIMES 2591

On the other hand, by (2.15) and the definition of x1 and u2 , we have


G(x1 ) + x1 u2 ≥ Bx−b 0 −b−1
1 − r(B x1 ) + x1 u2
= B(1 − µ1 )(1−a)b x−b 0 −b−1
0 − r(B x1 ) + (1 − µ1 )−(1−a) x0 (1 − 2µ1 )u0
= Bx−b −b 0 −b−1
0 + x0 u0 + O(µ1 x0 ) − r(B x1 )
= Bx−b 0 −b−1
0 + x0 u0 + O(r(B x0 )) − r(B 0 x−b−1
1 )
for all sufficiently large u0 . By condition (R2), we have r(B 0 x−b−1
1 ) ≤ r(B 0 x−b−1
0 ).
It follows that, by (2.19) and (2.20),
log P (u0 ) ≥ G(x1 ) + x1 u2 + O(1) ≥ Bx−b 0 −b−1
0 + x0 u0 − Cr(B x0 ) = Aua0 − Cr(u0 )
for all sufficiently large u0 , where C is a constant depending on C1 , C2 , B and b.
This completes the proof of the proposition.

3. Some Lemmas on Dirichlet Series and Mellin Transforms


Let w(n) be a non-negative function defined on the set of positive integers. Let
X∞
w(n)
(3.1) fw (s) =
n=1
ns
be the Dirichlet series generated by w(n), and suppose that fw (s) has finite abscissa
of absolute convergence σa . We define Fw (x) for x > 0 by
Y∞
Fw (x) = (1 − e−nx )−w(n) .
n=1
We note that in the case w(n) = Λ(n), we have fw (s) = −ζ 0 (s)/ζ(s), and Fw (x) is
the generating function for pΛ (n).
In this section we will prove some lemmas relating analytic properties of fw (s)
to the behavior of Fw (x).
Lemma 1. For any κ > max(0, σa ) we have
Z κ+i∞
1
(3.2) log Fw (x) = Γ(s)ζ(1 + s)fw (s)x−s ds.
2πi κ−i∞
Furthermore, for any T > 1 we have
Z κ+iT
1
(3.3) log Fw (x) = Γ(s)ζ(1 + s)fw (s)x−s ds
2πi κ−iT
 
+ O ζ(1 + κ)fw (κ)x−κ T κ+1/2 e−πT /2 ,
where the O-constant is absolute.
Proof. The proof of (3.2) can be found in Meinardus [7, p. 390].
To prove (3.3), we observe that, by Stirling’s formula,
(Z Z κ−iT )
κ+i∞
1
+ Γ(s)ζ(1 + s)fw (s)x−s ds
2πi κ+iT κ−i∞
Z ∞
 ζ(1 + κ)fw (κ)x−κ tκ+1/2 e−πt/2 dt
T
−κ κ+1/2 −πT /2
 ζ(1 + κ)fw (κ)x T e .
Thus (3.3) follows.
2592 YIFAN YANG

Lemma 2. For 0 < x < 1 and Re s > max(0, σa ) we have


Z 1
xs−1 log Fw (x) dx = Γ(s)ζ(1 + s)fw (s) − hw (s),
0

where hw (s) is an entire function.

Proof. Suppose that s satisfies Re s > max(0, σa ). Then


Z 1 Z ∞ Z ∞ 
xs−1 log Fw (x) dx = − xs−1 log Fw (x) dx
0 0 1
Z ∞ Z ∞
X∞ X ∞
w(n) −knx
= − x s−1
e dx
0 1 k
k=1 n=1
∞ X
X ∞  Z 
Γ(s)w(n) w(n) ∞ s−1 −knx
= − x e dx
n=1
ns k s+1 k 1
k=1
= Γ(s)ζ(1 + s)fw (s) − hw (s),
where interchanging the order of integration and summation is justified by the
absolute convergence of the double series involved. The function hw (s) here is
given by
∞ X
X ∞ Z X∞ X∞
w(n) ∞ s−1 −knx w(n)
hw (s) = x e dx = 1+s ns
Γ(s, kn),
n=1
k 1 n=1
k
k=1 k=1
R∞ s−1 −x
where Γ(s, u) = u
x e dx is the incomplete Gamma function. Using the
estimate
Z ∞
|Γ(s, u)| σ1 ,σ2 e−x/2 dx σ1 ,σ2 e−u/2 (σ1 ≤ Re s ≤ σ2 , u ≥ 1)
u

and the bound w(n)  nσa +1 (which follows from the convergence of fw (s) at
s = σa + 1), we see that the double series defining hw (s) converges uniformly in
any strip σ1 ≤ Re s ≤ σ2 , and hence represents an entire function of s.

Lemma 3 (Landau). (i) Let g(n) be a function defined on the set of positive
integers and of constant sign for all sufficiently large n. Suppose that the Dirichlet
series

X
g(n)n−s
n=1

has finite abscissa of absolute convergence σa . Then s = σa is a singularity of the


function represented by the Dirichlet series.
(ii) Let g(x) be an integrable function on [0, 1] and of constant sign for all suffi-
ciently small x. Suppose that the Dirichlet integral
Z 1
g(x)xs−1 dx
0

has finite abscissa of absolute convergence σa . Then s = σa is a singularity of the


function represented by the Dirichlet integral.
PARTITIONS INTO PRIMES 2593

Proof. Part (i) is the classical version of a well-known theorem of Landau (see, e.g.,
Ingham [5, p. 88]). To obtain part (ii), we note that the given integral can be
written as
Z ∞
h(u)u−s du
1
with h(u) = g(1/u)/u and x = 1/u. By the integral version of Landau’s lemma
(Ingham [5, p. 88]) the last integral has a singularity at s = σa .
Lemma 4. Assume that for some constants B > 0 and b > 0
(3.4) log Fw (x) = Bx−b {1 + o(1)}
as x → 0+. Then the Dirichlet series fw (s) has abscissa of convergence b and
satisfies
  
1 B |s − b|
(3.5) fw (s) = +o
s − b Γ(b)ζ(1 + b) σ−b
as s → b, while σ = Re s > b.
Proof. By Lemma 2 we have
Z 1 
1
(3.6) fw (s) = xs−1 log Fw (x) dx + hw (s) ,
Γ(s)ζ(1 + s) 0
where hw (s) is an entire function. On the other hand, the hypothesis (3.4) implies
that
Z 1 Z 1 Z 1 
(3.7) xs−1 log Fw (x) dx = B xs−b−1 dx + o xσ−b−1 dx
0 0 0
 
B 1
= +o
s−b σ−b
as σ → b+. Combining these two estimates yields (3.5).
To show that the Dirichlet series fw (s) has abscissa of absolute convergence b, we
observe that, by (3.6) and (3.7), fw (s) is analytic on the half-plane {s : Re s > b},
but has a singularity at the real point s = b. Since w(n) is non-negative, Lemma 3
implies that fw (s) has abscissa of absolute convergence b.
Lemma 5. Suppose that
B 1
gw (s) = fw (s) −
Γ(b)ζ(1 + b) s − b
can be meromorphically continued to a half-plane {s : Re s > σ0 }, where σ0 < b.
Let θ be the least upper bound of the real parts of the singularities of gw (s), and
suppose that gw (s) is analytic at the real point s = θ. Then, for any  > 0,
log Fw (x) − Bx−b = Ω± (x−(θ−) ) as x → 0.
Proof. Let  be given and consider the function
Φ(x) = log Fw (x) − Bx−b − x−(θ−)
with Mellin transform
Z 1
φ(s) = xs−1 Φ(x) dx.
0
2594 YIFAN YANG

Let σa be the abscissa of absolute convergence of φ(s). By Lemma 2, we have for


Re s > b
B 1
(3.8) φ(s) = Γ(s)ζ(1 + s)fw (s) − hw (s) − − ,
s − b s − (θ − )
where hw (s) is an entire function. By the definition of θ, for any δ > 0, the right-
hand side of (3.8) has singularities on the half-plane {s : Re s > θ − δ}; thus we
have
(3.9) σa ≥ θ.
Now, if Φ(x) ≤ 0 for all sufficiently small x, then Lemma 3 implies that the
real point s = σa is a singularity of φ(s). But by the assumption of Lemma 5,
φ(s) is analytic at the real point s = θ and has no singularities to the right of
s = θ. Thus σa < θ, which contradicts (3.9). It follows that Φ(x) changes sign
infinitely often, i.e., log Fw (x) − Bx−b = Ω+ (x−(θ−) ). A similar argument gives
log Fw (x) − Bx−b = Ω− (x−(θ−) ).

4. Proof of Theorems 1-3


We begin by proving two lemmas which show that the function
P pΛ (n) in our
theorems can be replaced by its summatory function PΛ (u) = n≤u pΛ (n).
Lemma 6. For any integer k ≥ 15 and n ≥ 0, we have pΛ (n + k) ≥ pΛ (n).
Proof. We first observe that, for any odd prime power q,

X X
q−1
(pΛ (n + q) − pΛ (n))xn + pΛ (j)x−q+j
n=0 j=0

Y Y
= (x−q − 1) (1 − xm )−Λ(m) = x−q (1 − xq )−(Λ(q)−1) (1 − xm )−Λ(m) .
m=1 m6=q

Since Λ(q) − 1 > 0 when q is an odd prime power, expanding each factor in the last
expression yields a series with non-negative coefficients. Thus we have pΛ (n + q) ≥
pΛ (n) for all n ≥ 0. Iterating this inequality with q = 3 and q = 5, we see that
pΛ (n + 3k + 5l) ≥ pΛ (n) for all non-negative integers n, k and l. Since every
integer ≥ 15 can be written as a linear combination of 3 and 5 with non-negative
coefficients, the lemma follows.
Lemma 7. Let n ∈ N. Let r(u) be a function satisfying r(u)(log u)−1 → ∞ as
u → ∞, and r(u + v)  r(u) uniformly for all sufficiently large u and 0 ≤ v ≤ 15.
Then
p
log pΛ (n) = 2 ζ(2)n1/2 + O(r(n))
as n → ∞ if and only if
p
log PΛ (u) = 2 ζ(2)u1/2 + O(r(u))
as u → ∞. A similar statement holds if O is replaced by Ω± .
Proof. By Lemma 6, we have
X
n
(4.1) pΛ (n) ≤ PΛ (n) = pΛ (k) ≤ (n + 1)pΛ (n + 15)
k=0
PARTITIONS INTO PRIMES 2595

for all integers n ≥ 0. Suppose that, for some positive constant C,


p p
2 ζ(2)u1/2 − Cr(u) ≤ log PΛ (u) ≤ 2 ζ(2)u1/2 + Cr(u)
for all sufficiently large u. Then (4.1) implies that
p p
2 ζ(2)(n − 15)1/2 − log(n − 14) − Cr(n − 15) ≤ log pΛ (n) ≤ 2 ζ(2)n1/2 + Cr(n)
for all sufficiently large integers n. Hence, by the assumptions on r(u), we obtain
p
log pΛ (n) = 2 ζ(2)n1/2 + O(r(n)).
The proof of the converse implication and of the analogous implications between
Ω± -estimates is similar.

By Lemma 7, it suffices to prove Theorems 1-3 with PΛ (u) in place of pΛ (n).

Proof of Theorem 1. By the Vinogradov-Korobov zero-free region for the Riemann


zeta function (Titchmarsh [15, p. 135]), we have ζ(s) 6= 0 for s = σ + it satisfying
C
σ ≥ η(t) = 1 − ,
(log t∗ )2/3 (log2 t∗ )1/3
where logk is the k times iterated logarithm, t∗ = max(10, |t|), and C is a positive
constant. Furthermore, in the same region we have
ζ 0 (s)
(4.2)  log t∗ ,
ζ(s)
provided |s − 1| ≥ δ for some positive constant δ.
Applying Lemma 1 with w(n) = Λ(n) and noting that in this case fw (s) =
−ζ 0 (s)/ζ(s), we obtain, for 0 < x < 1,
Z κ+iT  
1 ζ 0 (s) −s 0
−T ζ (κ) −κ
log FΛ (x) = − Γ(s)ζ(1 + s) x ds + O e x ,
2πi κ−iT ζ(s) ζ(κ)
where 1 < κ < 2 and T > 10 are positive numbers to be chosen later. We shift
the path of integration to the path consisting of the segments γ1 : {σ − iT : κ ≥
σ > η(T )}, γ2 : {η(t) + it : −T ≤ t < −10}, γ3 : {η(t) + it : −10 ≤ t < 10},
γ4 : {η(t) + it : 10 ≤ t < T } and γ5 : {σ + iT : η(t) ≤ σ ≤ κ}.
The estimate
1
|Γ(s)|  (1 + |t|)σ+1/2 e−π|t|/2  e−|t| ( ≤ σ ≤ 2, t ∈ R)
2
(which follows from Stirling’s formula) and (4.2) imply that
Z
 e−T (log T )x−κ
γ1,5

and
Z Z T
 e−t (log t)x−η(t) dt  x−η(T ) .
γ2,4 10

We also have
Z
 x−η(10) .
γ3
2596 YIFAN YANG

Thus, we obtain
(4.3)
   
−1
0
−T ζ (κ) −κ

log FΛ (x) = ζ(2)x +O e x + O e−T (log T )x−κ + O x−η(T ) .
ζ(κ)
We now assume that x is sufficiently small so that log3 x−1 is defined and ≥ 1. We
set κ = 1 + 1/ log x−1 and
C log x−1
T = ,
(log2 x−1 )2/3 (log 3 x−1 )1/3
where C is the constant appearing in the definition of η(t). Then x−κ = ex−1 and
ζ 0 (κ)/ζ(κ)  log x−1 . Hence, noting that
C
1 − η(T ) ∼
(log2 x−1 )2/3 (log3 x−1 )1/3
as x → 0, we obtain from (4.3)
  
−1 −1 (C/2) log x−1
(4.4) log FΛ (x) = ζ(2)x + O x exp − .
(log2 x−1 )2/3 (log3 x−1 )1/3
We now apply Proposition 2 with B = B 0 = ζ(2), b = 1 and the quantities A and
a determined by (2.3), i.e.,
p
A = a−1 (Ba/(1 − a))
1−a
(4.5) a = 1/2 and = 2 ζ(2).
Setting y = B 0 x−b−1 = ζ(2)x−2 and
 
(C/5) log y
r(y) = y exp −
1/2
,
(log2 y)2/3 (log3 y)1/3
it easily follows from (4.4) that
| log FΛ (x) − ζ(2)x−1 | ≤ r(ζ(2)x−2 )
for all sufficiently small x. Hence, by Proposition 2, we may conclude that, with
c = C/10,
p
log PΛ (u) = Aua + O( r(u)ua )
  
p c log u
= 2 ζ(2)u + O u exp −
1/2 1/2
.
(log2 u)2/3 (log3 u)1/3
This completes the proof of Theorem 1.
Proof of Theorem 2. We will prove only the Ω− -estimate; the Ω+ -estimate can be
proved similarly.
Applying
p the corollary to Proposition 1 with P (u) = PΛ (u), F (x) = FΛ (x),
A = 2 ζ(2), a = 1/2, B = B 0 = ζ(2), b = 1 and r(u) = u1/4 , we see that it suffices
to show that
(4.6) log FΛ (x) − ζ(2)x−1 = Ω− (x−1/2 ).
Let θ be the least upper bound of the real parts of the zeros of the Riemann zeta
function. If the Riemann Hypothesis is false, i.e., if θ > 1/2, we apply Lemma 5
with w(n) = Λ(n), fw (s) = −ζ 0 (s)/ζ(s), B = ζ(2) and b = 1. The function gw (s)
in Lemma 5 then is given by −ζ 0 (s)/ζ(s) − 1/(s − 1). Thus gw (s) is a meromorphic
function in Re s > 0, whose singularities are exactly the zeros of ζ(s). In particular,
the least upper bound of the real parts of these singularities is equal to the least
PARTITIONS INTO PRIMES 2597

upper bound θ of the real parts of the zeros of ζ(s). Moreover, since ζ(s) has no
zeros on the real line, gw (s) is analytic at s = θ. Thus, Lemma 5 implies that, for
any  > 0,
log FΛ (x) − ζ(2)x−1 = Ω± (x−(θ−) ).
In particular, we can choose  = θ − 1/2 and the assertion of Theorem 2 follows.
We therefore assume that the Riemann Hypothesis is true.
Let 1/2 + iγ be one of the non-trivial zeros of the Riemann zeta function and let
m be its multiplicity. Let C be a positive constant such that
(4.7) C < m|Γ(1/2 + iγ)ζ(3/2 + iγ)|
and define
Φ(x) = log FΛ (x) − ζ(2)x−1 + Cx−1/2 .
If (4.6) is false, then there exists x0 > 0 such that Φ(x) ≥ 0 holds for all x ≤ x0 .
By Lemma 3, the integral
Z 1
φ(s) = Φ(x)xs−1 dx
0
has a singularity at the real point s = σa , the abscissa of absolute convergence of
φ(s). On the other hand, applying Lemma 2 with fw (s) = −ζ 0 (s)/ζ(s), we obtain
ζ 0 (s) ζ(2) C
(4.8) φ(s) = −Γ(s)ζ(1 + s) − hΛ (s) − + ,
ζ(s) s − 1 s − 1/2
where hΛ (s) is an entire function. Since the last expression has no singularity on
the real axis to the right of 1/2, the abscissa of absolute convergence σa of the
integral φ(s) must be ≤ 1/2. For σ > 1/2 we have
Z 1 Z x0
(4.9) |φ(σ + iγ)| ≤ |Φ(x)|xσ−1 dx + Φ(x)xσ−1 dx
x0 0
Z1 Z 1
= {|Φ(x)| − Φ(x)}xσ−1 dx + Φ(x)xσ−1 dx
x0 0
Z1
≤ 2|Φ(x)|x−1/2 dx + φ(σ).
x0

We multiply both sides by σ − 1/2 and let σ → 1/2+. The right-hand side becomes
limσ→1/2+ (σ − 1/2)φ(σ), which, by (4.8), is equal to

ζ 0 (σ) ζ(2) C

lim (σ − 1/2) − Γ(σ)ζ(1 + σ) − + = C.
σ→1/2+ ζ(σ) σ − 1 σ − 1/2
On the other hand, applying (4.8) to φ(σ + iγ), we see that the left-hand side of
(4.9) becomes (after multiplying by σ − 1/2 and letting σ → 1/2+)

ζ 0 (σ + iγ)
lim (σ − 1/2) − Γ(σ + iγ)ζ(1 + σ + iγ)
σ→1/2+ ζ(σ + iγ)

ζ(2) C
− + = m|Γ(1/2 + iγ)ζ(3/2 + iγ)|.
σ − 1 + iγ σ − 1/2 + iγ
Hence m|Γ(1/2 + iγ)ζ(3/2 + iγ)| ≤ C, which contradicts (4.7). Therefore (4.6)
holds, and the proof of Theorem 2 is complete.
2598 YIFAN YANG

Proof of Theorem 3. Let θ be the least upper bound for the real parts of the zeros
of the Riemann zeta function. It suffices to show that if α is a real number such
that the estimate
p
(4.10) log PΛ (u) = 2 ζ(2)u1/2 + O(uα/2 )
holds as u → ∞, then α ≥ θ.
By Proposition 1, (4.10) implies that
log FΛ (x) = ζ(2)x−1 + O(x−α ).
Thus the integral
Z 1 
φ(s) = xs−1 log FΛ (x) − ζ(2)x−1 dx
0

defines an analytic function on the half-plane {s : Re s > α}. On the other hand,
by Lemma 2 we have, for Re s > 1,
ζ 0 (s) ζ(2)
(4.11) φ(s) = −Γ(s)ζ(1 + s) − − hΛ (s),
ζ(s) s−1
where hΛ (s) is an entire function. Since φ(s) is analytic in Re s > α, the latter
relation remains valid in this larger region, and the right-hand side of (4.11) is
analytic in Re s > α. Therefore ζ(s) cannot have zeros on the half-plane {s : Re s >
α}, i.e., we have θ ≤ α, as claimed.

5. A New Proof of Theorem B


Let θ be the least upper bound for the real parts of the zeros of the Riemann
zeta function. We may assume 1/2 ≤ θ < 1, for if θ = 1, the asserted estimate
follows from Theorem 1.
By Lemma 1 we have, for 0 < x < 1 and κ > 1,
Z κ+i∞
1 ζ 0 (s) −s
log FΛ (x) = − Γ(s)ζ(1 + s) x ds.
2πi κ−i∞ ζ(s)
By standard arguments (as, for example, in the proof of Theorems 28 of Ingham
[5]) we now shift the line of integration to the vertical line Re s = −1/2, taking into
account the residues of the integrand at s = 1, 0 and at the zeta-zeros. This gives
X ζ 0 (0)
log FΛ (x) = ζ(2)x−1 − Γ(ρ)ζ(1 + ρ)x−ρ − log x−1
ρ
ζ(0)
Z −1/2+i∞
1 ζ 0 (s) −s
− Γ(s)ζ(1 + s) x ds,
2πi −1/2−i∞ ζ(s)
where ρ runs over non-trivial zeta-zeros, counted with multiplicities.
By the symmetry of the zeta-zeros and the assumption that θ < 1, we have
|ζ(1 + ρ)| ≤ ζ(1 + (1 − θ))  1
for all zeta-zeros ρ. Using Stirling’s formula and the bound
X
1  log T,
T ≤Im ρ≤T +1
PARTITIONS INTO PRIMES 2599

we obtain
X X
Γ(ρ)ζ(1 + ρ)x−ρ  e−|γ| x−β  x−θ .
ρ ρ=β+iγ

Also, for Re s = −1/2 we have ζ 0 (s)/ζ(s)  log(|s| + 2) and ζ(1 + s)  (|s| + 1)1/2
(Theorems 9 and 27 in [5]). Thus
Z −1/2+i∞
ζ 0 (s) −s
Γ(s)ζ(1 + s) x ds
−1/2−i∞ ζ(s)
Z ∞
 x1/2 e−|t| (|t| + 1)1/2 log(|t| + 2) dt  x1/2 .
−∞
Hence, we obtain
(5.1) log FΛ (x) = ζ(2)x−1 + O(x−θ ).
Similarly, we have, for κ > 1,
Z
d 1 κ+i∞
ζ 0 (s) −s−1
(5.2) log FΛ (x) = sΓ(s)ζ(1 + s) x ds
dx 2πi κ−i∞ ζ(s)
−2
= −ζ(2)x + O(x−(θ+1) )
and
Z
d2 1 κ+i∞
ζ 0 (s) −s−2
(5.3) log FΛ (x) = − s(1 + s)Γ(s)ζ(1 + s) x ds
dx2 2πi κ−i∞ ζ(s)
= 2ζ(2)x−3 + O(x−(θ+2) ).
By (5.1)-(5.3), the function F (x) = FΛ (x) satisfies the hypotheses (2.15), (2.22)
and (2.23) of Proposition 3 with b = 1, B = B 0 = ζ(2) and r(u) = Kuθ/2, where
K is a sufficiently large constant. Moreover, the function r(u) satisfies conditions
(R1), (R2) and (R3) of that proposition. Hence the conclusion (2.24) of Proposition
3 holds with A and a given by (4.5). Therefore
p
log PΛ (u) = Aua + O(r(u)) = 2 ζ(2)u1/2 + O(uθ/2 )
as u → ∞, which is the claimed result.

Acknowledgment
The author wishes to thank Professor A. Hildebrand for helpful advice on the
exposition of the paper.

References
1. N. A. Brigham, On a certain weighted partition function, Proc. Amer. Math. Soc. 1 (1950),
192–204. MR 11:582c
2. G. A. Freiman, Inverse problems of the additive theory of numbers, Izv. Akad. SSSR. Ser.
Mat. 19 (1955), 275-284. (Russian) MR 17:239c
3. G. H. Hardy and S. Ramanujan, Asymptotic formulae in combinatory analysis, Proc. London
Math. Soc. (2) 17 (1918), 75–115.
4. A. E. Ingham, A Tauberian theorem for partitions, Ann. of Math. (2) 42 (1941), 1075–1090.
MR 3:166a
5. , The distribution of prime numbers, Cambridge University Press, Cambridge, 1990.
MR 91f:11064
6. E. E. Kohlbecker, Weak asymptotic properties of partitions, Trans. Amer. Math. Soc 88
(1958), 346–365. MR 20:2309
2600 YIFAN YANG

7. G. Meinardus, Asymptotische Aussagen über Partitionen, Math. Z. 59 (1954), 388–398.


MR 16:17e
8. , Über Partitionen mit Differenzenbedingungen, Math. Z. 61 (1954), 289–302.
MR 16:905a
9. A. M. Odlyzko, Explicit Tauberian estimates for functions with positive coefficients, J. Com-
put. Appl. Math. 41 (1992), 187–197. MR 94f:11086
10. B. Richmond, Asymptotic relations for partitions, J. Number Theory 7 (1975), 389–405.
MR 52:3095
11. , A general asymptotic result for partitions, Canad. J. Math. 27 (1975), 1083–1091.
MR 52:5604
12. K. F. Roth and G. Szekeres, Some asymptotic formulae in the theory of partitions, Quart. J.
Math., Oxford Ser. (2) 5 (1954), 241–259. MR 16:797b
13. W. Schwarz, Schwache asymptotische Eigenschaften von Partitionen, J. Reine Angew. Math.
232 (1968), 1–16. MR 38:4433
14. , Asymptotische Formeln für Partitionen, J. Reine Angew. Math. 234 (1969), 174–178.
MR 40:7217
15. E. C. Titchmarsh, The theory of the Riemann zeta-function, 2nd ed., The Clarendon Press,
Oxford University Press, New York, 1986. MR 88c:11049

Department of Mathematics, University of Illinois, Urbana, Illinois 61801


E-mail address: yfyang@math.uiuc.edu

You might also like