SIEVE WEIGHTS AND THEIR SMOOTHINGS

Notation: n ∼ x if x < n ≤ 2x
Counting twin primes

#{n ∼ x : n, n + 2 prime } = #{n ∼ x : (a, m) = 1},

where
Y
m= p, a = n(n + 2).
√
p≤ 2x

Mobius inversion:
X
1(a,m)=1 = µ(d)
d|(a,m)

gives a formula

#{n ∼ x : n, n + 2 prime }
X X
= µ(d),
n∼x d|(a,m)
X X
= µ(d) 1
d|m n∼x
n(n+2)≡0 (mod d)
√
m = p≤√2x p ≈ e 2x , problems arise when d is big.
Q
want to find a strong correlation:
X
1(a,m)=1 ≈ ρd , ρd = 0, if d > D = xθ
d|(a,m),d≤D

Combinatorial sieve:

ρd = µ(d)1d∈D , d = p1 . . . pr

Selberg’s sieve :
 2
X
1(a,m)=1 ≤  λd  , λ1 = 1, λd = 0, if d > R, D = R2
d|(a,m)
1
2 SIEVE WEIGHTS AND THEIR SMOOTHINGS
 2
X X
#{n ∼ x : n, n + 2 prime } ≤  λd 
n∼x d|(a,m)
X X
= λd1 λd2 1
d1 ,d2 n∼x
d=[d1 ,d2 ] n(n+2)≡0 (mod d)
 
X ρ(d)
= λd1 λd2 s + O(ρ(d)) ,
d1 ,d2 ≤R
d

where ρ(d) = #{n ∈ Z/dZ : n(n + 2) ≡ 0 (mod d)}.

 2
log R/d
λd ≈ cµ(d) 1d≤R
log R

c2 x
#{n ∼ x : n, n + 2 prime } ≤ (8 + o(1)) ,
log2 x

where c2 = 2 p≥3 (1 − p2 )(1 − p1 )−2 , twin prime constant

Q

Counting k-tuple primes

#{n ∼ x : n + h1 , n + h2 , . . . , n + hk all prime }

need
 k
log(R/d)
λd ≈ cµ(d) 1d≤R
R
General
 
X log d
Mf (n; R) = µ(d)f
log R
d|n

supp(f ) ⊂ (−∞, 1], f (1) = 1 =→ Mf (n; R) = 1 if p | n =⇒ p > R

x
#{n ≤ x : p | n =⇒ p > R} ≈
log R

X
M(n; R) = µ(d)
d|n
 SIEVE WEIGHTS AND THEIR SMOOTHINGS 3

If n = 2m, 2 - m,
X
M(n; R) = µ(d)
d|n
X X
= µ(d) + µ(2d)
d|m d|m
d≤R d≤R/2
X
= µ(d)
d|m
R/2<d≤R

= ±1 if m has a unique square free divisor in (R/2,R]

Ford:
x
#{n ≤ x : n has a unique square free divisor in (R/2,R]}  ,
(log R)δ (log log R)3/2
1+log log 2
where δ = 1 − log 2
= 0.086071 · · · < 1. This implies that
X x
M(n; R)2  ,
n≤x
(log R)δ+o(1)

while
x
#{n ≤ x : p | n =⇒ p > R} ≈
log R
If f ∈ C 1 (R), n = pv m, p - m, then
  X  
X log d log(dp)
Mf (n; R) = µ(d)f − µ(d)f
log R log R
d|m d|m
Z log p  
log R X
0 log d
=− µ(d)f u + du
0 log R
d|m

If f ∈ C A (R) and n = pv11 · · · pvAA m, (pj is the j-th smallest prime factor
of n),
Z log p1 Z log pA X  
A
log R log R log d
Mf (n; R) = (−1) ··· µ(d)f u1 + · · · + uA + du
0 0 log R
d|m

log p1 log pA
. ··· M (A) (m; R)
log R log R f
If n is typical, then
Mf (A) (m; R)
Mf (n; R) .
(log R)A
4 SIEVE WEIGHTS AND THEIR SMOOTHINGS

(soft) Conjecture:
( )
2k
P
x m≤x M(n; R)
X
Mf (n; R)2k . max , .
n≤x
log R (log R)2k
↑ ↑
almost primes typical integers

Conjecture :
X
M(n; R) ∼ ck x(log R)Ek
n≤x

Ek
need Ek − 2kA < 0 ⇔ A > 2k
for Mf (n; R)2k to behave like a sieve
weight.
Two questions:
1. Ek =?
2. Is (soft) conjecture true?
For the second question, we have

Theorem 1 (GKM ’16 ). Suppose f ∈ C A (R), f, f 0 , . . . , f (A) are uni-

formly bounded, supp(f ) ⊂ (−∞, 1],
Ek
a) If A > 2k
+ 1, then

 
X
2k 3/2 x 2
Mf (n; , R) α , x ≥ R ≥ α, 1 ≥ α ≥
n≤x
log R log R
∃p|n,p≤Rα
X x
Mf (n; R)2k ∼ ck,f x ≥ R2k log2 R, R → ∞


,
n≤x
log R

Ek
b) If A ≤ 2k
+ 1, then
X
Mf (n; R)2k  x(log R)Ek −2k(A−1)
n≤x

For the first question on Ek :

(1) k = 1, E1 = 0, (Dress, Iwaniec and Tenenbaum, ’83)
(2) k = 2, E2 = 0, (Motohashi, ’04)
(3) ∀k, Ek exists, (Balazard, Naimi, and Pétermann, ’08, de la
Bréteche, ’01)
 SIEVE WEIGHTS AND THEIR SMOOTHINGS 5
 2k
X X X
M(n; R)2k =  µ(d)
n≤x n≤x d|n,d≤R
X X
= µ(d1 ) · · · µ(d2k ) 1
d1 ,...,d2k ≤R n≤x
d1 ···d2k |n
X µ(d1 ) · · · µ(dk )
=x + O(R2k )
d1 ,...,dk ≤R
[d 1 , . . . , d2k ]

2k
X µ(d1 ) · · · µ(d2k ) Y R si
Z Z
2k+ x
x≥R ∼ ··· ds
(2πi)2k <(s1 )= log1 R <(s2k )= log1 R d ,...,d
[d1 , . . . , d2k ] i=1 si dsi i
1 2k

[d1 , . . . , d2k ] = p =⇒ ∃I 6= ∅, s.t.

di = p, i ∈ I
dj = 1, j 6∈ I
1 X
(−1)|I| 1+s , sI = si
p I i∈I

2k Q
R si I even6=∅ ζ(1 + sI )
X µ(d1 ) · · · µ(d2k ) Y
si = F (s) Q
d ,...,d
[d1 , . . . , d2k ] i=1 si di I odd ζ(1 + sI )
1 2k

Shift s2k to the left, poles when sI = 0, 2k ∈ I I even,

X
order = (−1)|I|
I6=∅,2k∈I,sI =0

Denote s2k = L(s1 , . . . , s2k ), where L is a linear form. Shift s2k−1 ,

pick poles when sI = 0 and 2k − 1 ∈ I. s2k−1 = LP 1 (s1 , . . . , s2k−2 )
s2k = L̃(s1 , . . . , s2k−2 ). Shift 2k − m variables, si = m j=1 ai,j sj ,
m X
X X
sI = 0 ⇔ ( ai,j )sj = 0 ⇔ ai,j = 0, ∀j
j=1 i∈I i∈I
 
X
(−1)|I| − 2k + m
 
Ek = max 
m≥1  
A contains Im =∅,⊂{1,...,2k}
I6P
i∈I ai,j =0∀j

optimal: m = 1, (a1 , . . . , a2k ) = (1, −1, . . . , 1, −1)

 
2k
Ek = − 2k
k