A note on Primes in Short Intervals, , November 27, 2006

1 Introduction

Let

Λ (n)

be von Mangoldt lambda function,

μ (n)

be the Möbius function and

φ (n)

be the Euler's phi function. Let

e (θ) = e^{2 π i θ}

ε > 0

and

ψ (x) = \sum_{n \leq x} Λ (n)

In [ 1 ], Montgomery and Soundararajan studied the distribution of primes in short intervals

\begin{matrix} M_{K} (N; H) : = \sum_{n = 1}^{N} (ψ (n + H) - ψ (n) - H)^{K} \end{matrix}

(1)

under a strong quantitative form of the Hardy-Littlewood prime

k

-tuple conjecture:

Conjecture 1.

\begin{matrix} {\sum_{n \leq x}}^{k} \prod_{i = 1} Λ (n + d_{i}) = S (D) x + E_{k} (x; D) \end{matrix}

(2)

holds with

E_{k} (x; D) ≪_{ε, K} N^{1 / 2 + ε}

uniformly for

1 \leq k \leq K

0 \leq x \leq N

, and distinct

d_{i}

satisfying

1 \leq d_{i} \leq H

. Here

D = {d_{1}, d_{2}, . . ., d_{k}}

, and

S (D) = {\sum_{q_{1}, . . ., q_{k}}}_{1 \leq q_{i} < \infty} (^{k} \prod_{i = 1} \frac{μ (q_{i})}{φ (q_{i})}) {\sum_{a_{1}, . . ., a_{k}}}_{1 \leq a_{i} \leq q_{i}}_{(a_{i}, q_{i}) = 1}_{\sum a_{i} / q_{i} \in Z} e (\sum_{i = 1}^{k} \frac{a_{i} d_{i}}{q_{i}})

is the singular series as in equation (2) of [ 1 ].

They proved

Theorem 1. Under Conjecture 1 ,

\begin{matrix} M_{K} (N; H) = & μ_{k} H^{K / 2} \int_{1}^{N} (log x / H + B)^{K / 2} d x & + O (N (log N)^{K / 2} H^{K / 2} {(\frac{H}{log N})}^{- 1 / (8 K)} + H^{K} N^{1 / 2 + ε}) \end{matrix}

uniformly for

log N \leq H \leq N^{1 / K}

, where

μ_{k} = 1 \cdot 3 \cdot \cdot \cdot (k - 1)

k

is even, and

μ_{k} = 0

k

is odd;

B = 1 - C_{0} - log 2 π

and

C_{0}

denotes Euler's constant.

The first author of [ 1 ] suggested to the present author that Theorem 1 is probably true under an average form of the Hardy-Littlewood prime

k

-tuple conjecture, namely:

Conjecture 2. For

x \geq H

{\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{k} (x, D)^{2} ≪_{k} x^{1 + ε} H^{k} .

Our goal in this paper is to prove Theorem 1 under Conjecture 2 .

Acknowledgement The author would like to thank Prof. Hugh Montgomery for suggesting the problem and the American Institute of Mathematics for provision both financially and spatially.

2 Proof of Theorem 1

Lemma 1. With

Λ_{0} (n) = Λ (n) - 1

and

| A |

denoting the size of a set

A

{\sum_{n \leq x}}^{k} \prod_{i = 1} Λ_{0} (n + d_{i}) = S_{0} (D) x + O (\sum_{J \subset {1, 2, . . ., k}} | E_{| J |} (x; D_{J}) |)

where

S_{0} (D) = {\sum_{q_{1}, . . ., q_{k}}}_{1 < q_{i} < \infty} (^{k} \prod_{i = 1} \frac{μ (q_{i})}{φ (q_{i})}) {\sum_{a_{1}, . . ., a_{k}}}_{1 \leq a_{i} \leq q_{i}}_{(a_{i}, q_{i}) = 1}_{\sum a_{i} / q_{i} \in Z} e (\sum_{i = 1}^{k} \frac{a_{i} d_{i}}{q_{i}})

and

D_{J} = {d_{j}}_{j \in J}

Proof: The left hand side

\begin{matrix} = & {\sum_{n \leq x}}^{k} \prod_{i = 1} (Λ (n + d_{i}) - 1) = \sum_{n \leq x} \sum_{J \subset {1, 2, . . ., k}} (- 1)^{k - | J |} \prod_{i \in J} Λ (n + d_{i}) & = & \sum_{J \subset {1, 2, . . ., k}} (- 1)^{k - | J |} \sum_{n \leq x} \prod_{i \in J} Λ (n + d_{i}) & = & \sum_{J \subset {1, 2, . . ., k}} (- 1)^{k - | J |} (S (D_{J}) x + E_{| J |} (x; D_{J})) & = & S_{0} (D) x + O (\sum_{J \subset {1, 2, . . ., k}} | E_{| J |} (x; D_{J}) |) \end{matrix}

by ( 2 ) and the identity

S_{0} (D) = \sum_{ℐ \subset D} (- 1)^{| ℐ |} S (ℐ)

(see equation (5) of [ 1 ]).

Proof of Theorem 1 under Conjecture 2 : Following [ 1 ], we expand ( 1 ) and have

\begin{matrix} \begin{matrix} M_{K} (N; H) = & \sum_{k = 1}^{K} \frac{1}{k!} {\sum_{1 \leq M_{1}, . . ., M_{k}}}_{\sum M_{i} = K} (\binom{K}{M_{1} \cdot \cdot \cdot M_{k}}) & \times {\sum_{m_{1}, . . ., m_{k}}}_{0 \leq m_{i} < M_{i}}^{k} \prod_{i = 1} (- 1)^{M_{i} - 1 - m_{i}} (\binom{M_{i} - 1}{m_{i}}) L_{k} (m), \end{matrix} \end{matrix}

(3)

where

\begin{matrix} L_{k} (m) : = {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} {\sum_{n = 1}^{N}}^{k} \prod_{i = 1} Λ_{m_{i}} (n + d_{i}) \end{matrix}

(4)

and

Λ_{m} (n) : = Λ (n)^{m} Λ_{0} (n)

. To estimate

L_{k} (m)

, one needs to distinguish between those

i

for which

m_{i} = 0

and those for which

m_{i} > 0

. Following [ 1 ], we set

K = {1, . . ., k}

and introduce

ℋ = {i \in K : m_{i} \geq 1}, ℐ = {i \in K : m_{i} = 0}, J \subset K .

Then

\begin{matrix} \sum_{n \leq x} \prod_{i \in ℐ} Λ_{0} (n + d_{i}) \prod_{i \in ℋ} Λ (n + d_{i}) & = & \sum_{n \leq x} {\sum_{J}}_{ℐ \subset J \subset K} \prod_{i \in J} Λ_{0} (n + d_{i}) = {\sum_{J}}_{ℐ \subset J \subset K} \sum_{n \leq x} \prod_{i \in J} Λ_{0} (n + d_{i}) & = & {\sum_{J}}_{ℐ \subset J \subset K} [S_{0} (D_{J}) x + O (\sum_{J^{'} \subset J} | E_{| J^{'} |} (x; D_{J^{'}}) |)] & = & x {\sum_{J}}_{ℐ \subset J \subset K} S_{0} (D_{J}) + O_{k} (\sum_{K^{'} \subset K} | E_{| K^{'} |} (x; D_{K^{'}}) |) \end{matrix}

by Lemma 1 . We write the above as

f (x) = c x + E_{x, K}

. In general,

\begin{matrix} \int_{1^{-}}^{X} g (x) d f (x) & = & g (X) f (X) - \int_{1^{-}}^{X} f (x) g^{'} (x) d x & = & g (X) [c X + E_{X, K}] - \int_{1^{-}}^{X} (c x + E_{x, K}) g^{'} (x) d x & = & c \int_{1^{-}}^{X} g (x) d x + O (E_{X, K} | g (X) | + \int_{1^{-}}^{X} E_{x, K} | g^{'} (x) | d x) . \end{matrix}

Thus, by integration by parts,

\begin{matrix} \sum_{n = 1}^{N} & (\prod_{i \in ℐ} Λ_{0} (n + d_{i})) (\prod_{i \in ℋ} Λ (n + d_{i}) (log (n + d_{i}))^{m_{i} - 1} (log (n + d_{i}) - 1)) & = & \int_{1^{-}}^{N} \prod_{i \in ℋ} (log (x + d_{i}))^{m_{i} - 1} (log (x + d_{i}) - 1) d f (x) & = & c \int_{1}^{N} \prod_{i \in ℋ} (log (x + d_{i}))^{m_{i} - 1} (log (x + d_{i}) - 1) d x + O (E_{N, K} {log}^{K} N & + \int_{1^{-}}^{N} E_{x, K} \frac{{log}^{K} (x + H)}{x} d x) . \end{matrix}

This is the analogue of equation (65) in [ 1 ]. Now, note that

Λ_{m} (n) = Λ (n) (log n)^{m - 1} (log n - 1)

when

n

is prime. We have, by following the argument in [ 1 ],

\begin{matrix} \begin{matrix} {\sum_{n = 1}^{N}}^{k} \prod_{i = 1} Λ_{m_{i}} (n + d_{i}) & = & ({\sum_{J}}_{ℐ \subset J \subset K} S_{0} (D_{J})) (I_{m} (N) + O (H (log N)^{K - k})) + O (N^{1 / 2 + ε}) & + O (E_{N, K} {log}^{K} N + \int_{1^{-}}^{N} E_{x, K} \frac{{log}^{K} (x + H)}{x} d x) . \end{matrix} \end{matrix}

(5)

where

I_{m} (N) : = \int_{1}^{N} \prod_{i \in ℋ} ((log x)^{m_{i} - 1} (log x - 1)) d x .

Putting ( 5 ) into ( 4 ),

\begin{matrix} \begin{matrix} L_{k} (m) = & I_{m} (N) {\sum_{J}}_{ℐ \subset J \subset K} {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} S_{0} (D_{J}) + O (H^{k} N^{1 / 2 + ε}) & + O ({\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{N, K} {log}^{K} N + \int_{1^{-}}^{N} {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{x, K} \frac{{log}^{K} x}{x} d x) . \end{matrix} \end{matrix}

(6)

Now, we use Conjecture 2 . By Cauchy-Schwarz inequality,

{\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} | E_{k} (x, D) | ≪_{k} x^{1 / 2 + ε} H^{k} .

In particular,

\begin{matrix} {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{x, K} = & {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} \sum_{K^{'} \subset K} | E_{| K^{'} |} (x; D_{K^{'}}) | & ≪_{k} \sum_{j = 0}^{k} {\sum_{d_{1}, . . ., d_{j}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} | E_{j} (x; D) | ≪_{k} x^{1 / 2 + ε} H^{k} \end{matrix}

Applying this to ( 6 ), the second error term is

≪ N^{1 / 2 + ε} H^{k}

while the third error term is

\begin{matrix} = & \int_{H}^{N} {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{x, K} \frac{{log}^{K} (x + H)}{x} d x + \int_{1^{-}}^{H} {\sum_{d_{1}, . . ., d_{k}}}_{1 \leq d_{i} \leq H}_{d_{i} distinct} E_{x, K} \frac{{log}^{K} (x + H)}{x} d x & ≪_{k} & N^{1 / 2 + ε} H^{k} + \int_{1}^{H} H^{k + 1} \frac{{log}^{K} (x + H)}{x} d x ≪_{K} N^{1 / 2 + ε} H^{k} \end{matrix}

H \leq N^{1 / 2}

. Hence, ( 6 ) has an error

O (H^{k} N^{1 / 2 + ε})

and the rest of the proof in [ 1 ] follows. Therefore, we have Theorem 1 under Conjecture 2 .

References

H.L. Montgomery and K. Soundararajan, Primes in Short Intervals, Commun. Math. Phys. 252 (2004), 589-617.

Tsz Ho Chan American Institute of Mathematics 360 Portage Avenue Palo Alto, CA 94306 U.S.A. thchan@aimath.org