Professional Documents
Culture Documents
The Binary Goldbach Problem
The Binary Goldbach Problem
ABSTRACT
In this paper the authors will present the results of Baica’s paper “Clarifications of the Author’s
Previous Paper in the Goldbach Conjecture” in a slightly different form and correct some
misprints in it.
1. Introduction
As known, the Goldbach problem admits two different statements:
A) The ternary problem: every odd natural number is the sum of at most three odd primes.
B) The binary problem: every even natural number is the sum of at most two odd primes.
Such decompositions are called Goldbach decompositions.
Concerning the ternary problem, it has been proved (ref. (1)), that if in the original paper of
Hardy-Littlewood (ref. (2)) is assumed the validity of the Extended Riemann Hypothesis
(E.R.H.), then the Goldbach decompositions exists for every odd N > 1,24 x 1050.
This bound was lowered by Lucke in 1926 to N > 1032, and by D. Zinoviev in 1997 to N =
1020.
Finally in 1997 was proved by J.M. Deshouillers that is valid (always under E.R.H.) for every
odd N ≥ 7 (ref. (3)).
Now, very recently, the E.R.H. (together with many others) has been proved in ref. (4), so that
at present the ternary problem can be regarded as completely solved.
As regards the binary problem, the present status is entirely different. As pointed out by Hardy
and Littlewood in their paper above, the circle method is insufficient to prove their conjectural
formula even under the assumption of the E.R.H. , and nobody else, in the intervening 80 years
already past could surmount the difficulties of the method.
In ref. (5) was proved, under the Riemann hypothesis, that every natural number can be written
as a sum of at most four primes; but this is not the binary problem.
According to the six proofs of the Riemann hypothesis given in ref. (4), this is now a new
theorem.
Cleary, it was necessary to develop some new method to deal with the binary problem.
Here, the method of the Laplace transform, developed in several papers along 20 years by one
of the authors of this paper (ref. (6), (7), (8)), has turned out to be a key tool. The results were
shown in ref. (9), and we present them in a slightly different form, correcting by the way some
few misprints and omissions in it.
ϑ ([u ]) = ∑ log p
(2.1) p ≤ [u ]
where p denotes the prime number, we have:
⎧ log p if [u ] = p
Δ ϑ ([u ]) = ϑ ([u ]) − ϑ ([u − 1]) ⎨
(2.2) ⎩o in any other case
Hence, if we form
It is zero everywhere, except at the points where u = p1 t – u = p2, where it has the value
log p1 . log p2
It is evident from the graph that if we wish to obtain this value as an area, we must displace the
graph of Δ ϑ ([u]) to the left in one unity, because then:
p
∫ { ϑ ([u + 1]) - ϑ ([u])} { ϑ ([t - u]) - ϑ ([ t − u − 1])} =
p −1
(The shaded areas of log p1 and log p2 multiply between themselves). This is shown in fig. 2 and
5.
The integrand in (3.3) vanishes if u < 0 or u > t
Hence, we have that
t
v (t) = ∫ Δ ϑ ([u] + 1) Δ ϑ ([t - u]) du
(3.4) o
This is nothing but the convolution of the two functions that appear at right.
Consequently, if we take the Laplace transform of both sides we obtain, by known rules:
(3.6) L { v ( t )} = e s L2 { Δ ϑ ([u])}
Or also:
(3.7) L { v ( t - 1) } = L2 {Δ ϑ ([u])}
1 - e -s ∞
L { ϑ ([u]) } = ∑ ϑ (k) e
- ks
(4.1) s k=0
(4.2) s p
[ n] q μ (q ) e 2 Π i h q
F ( x ) = ∑ log p . x p = ∑ ∑ 2Πi h q
+ An ϑ + 1 4 +ε
(4.3) q =1 h =1 ϕ (q ) ( x − e )
(h.q)=1
[ n] q μ (q )
F (s) = ∑ log p . e − ps = ∑ ∑ + An ϑ + 1 4 +ε
(4.4) q =1 h =1 ϕ (q ) (s + 2 Π i h q )
(h.q)=1
5. Use of (4.4)
Replacement of (4.4) in (4.2), and of (4.2) in (3.6) yields:
⎞ ⎧[ n ]
2 2
⎛ 1 − e −s q μ (q ) 4 +ε⎫
L { v (t) } = e ⎜⎜
s ⎟ ⎨∑
⎟ q =1 ∑ + An ϑ + 1 ⎬
(5.1) ⎝ s ⎠ ⎩ h =1 ϕ (q ) (s + 2 Π i h q ) ⎭
or
⎧ 1 - e-s ⎞ 2 [ n ]
v (t - 1 + 0) + v (t - 1 - 0) -1 ⎪⎛ μ 2 (q)
q
(5.2) v m (t - 1) = = L ⎨⎜⎜ ⎟ +
2 s ⎟ ∑ ∑ 2
h = 1 ϕ (q) (s + 2 πi h q )
2
⎪⎩⎝ ⎠ q
2
⎛ 1 - e-s ⎞ μ (q1 ) μ (q 2 )
+ ⎜⎜ ⎟ ∑ ∑ ∑ ∑
⎟ +
⎝ s ⎠ q1 ≠ q 2 ϕ(q1 ) ϕ(q 2 ) (s + 2 πi h1 q1 ) (s + 2 πi h 2 q 2 )
⎛ 1 - e -s ⎞
+ ⎜⎜ ⎟
2
[ n]
∑ ∑
( )
μ(q) 2 An ϑ + 1/4 + ε ⎛ 1 - e - s ⎞
+ ⎜ ⎟
2
An 4
⎫
2ϑ + 2 + 2 ε ⎪
⎬=
s ⎟ ϕ(q )(s + 2 πi h q ) ⎜⎝ s ⎟⎠
⎝ ⎠ q =1 h ⎪⎭
⎧⎛ 1 - e - s ⎞ 2 ⎧[ n ] μ 2 (q)
⎪
= L ⎨⎜⎜
-1 ⎟ +
s ⎟ ⎨ q∑ ∑ 2
h ϕ (q) (s + 2 πi h q )
2
⎪⎩⎝ ⎠ ⎩ =1
[ n] [ n] μ (q1 ) μ (q 2 )
+ ∑ ∑ ∑ ∑ +
q1 h 1 q 2 h 2 ϕ(q1 ) ϕ(q 2 ) (s + 2 πi h1 q1 ) (s + 2 πi h 2 q 2 )
ϑ+ 1 +ε μ(q) ⎫
+ 2An n
∑∑ ⎬+
ϕ(q) (s + 2 πi h q ) ⎭
⎧⎛ 2 ⎫
⎪⎜ 1 - e - s ⎞⎟ 2 2ϑ + 1/2 + 2 ε ⎪
-1
+ L ⎨⎜
s ⎟ A n ⎬=
⎪⎩⎝ ⎠ ⎪⎭
⎧⎛ 2 ⎫
⎪⎜ 1 - e - s ⎞⎟ ⎪
(5.3) L ⎨⎜ -1
s ⎠ ⎟ {T1 (s) + T2 (s) + T3 (s)}⎬ + T3
⎪⎩⎝ ⎪⎭
By reasons that will become evident in § 10, we shall evaluate in first place
L-1 {T1(s)} and L-1{T2(s)}.
According to the evaluations performed in ref. [10] we can adopt for A the value
(5.4) A = 80
⎧ N μ 2 (q) ⎫
L {T1 (s)} = L ⎨ ∑ ∑ 2
−1 −1
2⎬
⎩ q = 1 h ϕ (q ) (s + 2 πi h/q ) ⎭
N q -1μ 2 (q ) - 2 πi ht/q N μ 2 (q )
= ∑ ∑ 2 e t = ∑ 2 C q (t) . t
(6.1) q =1 h = q ϕ (q ) q = 1 ϕ (q )
where
q -1
- 2 πi h t/q
Cq (t ) = ∑ e
h=0
(h.q) =1
μ 2 (q ) 3 γ (log log n)
2
∑ C q ( t ) < d(t) e log log t
q>n ϕ 2 (q ) N
∞ μ 2 (q )
L−1 {T1 (s)} = ∑ Cq (t ) . t +
q =1 ϕ 2 (q )
(log log N) 2
+ δ1 e 3γ d(t) log log t . t
(6.2) N
∞ μ 2 (q )
S( t ) = ∑ 2
Cq (t)
(6.3) q = 1 ϕ (q )
vanishes for odd t, and for even t can be transformed into an infinite product.
(ref. [14]).
∞ ⎛ ⎞ p -1 p -1
S( t ) = 2Π ⎜⎜1 - 1 2 ⎟⎟ Π t = 1,3203 Π . t = P (t) . t
(6.4)
p =3
⎝ (p - 1) ⎠ p| t p - 2 p| t p - 2
We have computed thus the influence of the infinitude of double poles in (5.2)
7. Evaluation of L-1 {T2(s)}
⎧ N N μ(q1 ) μ(q 2 ) ⎫
F5 (t) = L-1 {T2 (s)} = L-1 ⎨ ∑ ∑ ∑ ∑ ⎬=
⎩ 1
q =1 h 1 q 2 = 2 h 2
ϕ( q1 ) ϕ( q 2 ) (s + 2 πi h /q
1 1 )(s + 2 πi h /q )
1 1 ⎭
N N μ(q1 ) μ(q 2 ) (e − A1 t − e − A 2 t )
= ∑ ∑ ∑ ∑
(7.1) q 1 =1 h 1 q 2 = 2 h2 ϕ(q1 ) ϕ(q 2 ) A 2 - A1
δ2 2
L-1 {T2 (s) } = N (N + 1) 2
(7.2) 4π
We have computed thus the influence of the simple poles with q ≤ N multiplied by themselves.
We have
μ(q)
T3 (s) = 2An ϑ +1 4 + ε L-1 ∑ ∑ =
ϕ(q)(s + 2πi h q )
N μ(q )
= 2An ϑ +1 4 + ε ∑ ∑e
− 2 π i h/q t
=
q ϕ(q) h
N μ(q )
= 2An ϑ +1 4 + ε ∑ Cq (t)
q =1 ϕ(q)
2 ⎧t if 0 < t < 1
⎛ 1 - e-s ⎞ ⎪
F(t) = L ⎜⎜-1 ⎟ = ⎨2 - t if 1 < t < 2
s ⎟
⎝ ⎠ ⎪0 otherwise
(8.2) ⎩
Hence:
⎜F (t) ⎜ ≤ 1 for every t, and as it zero for t > 2, we can leave aside the influence of T3.
9. Return to (5.3)
We must evaluate now
⎧⎛ 1 - e - s ⎞ 2 ⎫
⎪⎜ -1 ⎟ T1 (s)⎪⎬
F1 (t) = L ⎨⎜ ⎟
(9.1) ⎪⎩⎝ s ⎠ ⎪⎭
and
⎧⎛ 1 - e - s ⎞ 2 ⎫
⎪⎜ -1 ⎟ ⎪
F2 (t) = L ⎨⎜ ⎟ T2 (s)⎬
(9.2) ⎪⎩⎝ s ⎠ ⎪⎭
(10.2) D ( t ) = P (t) . t
where
p -1
P ( t ) = 1,3203 Π
(10.3) p| t p-2
is a discontinuous step function of (t), plus a remainder term R(t), given by:
(log log N) 2
R (t) = t δ3 e 3γ d(t) log log t
(10.4) N
that contains a discontinuous factor d(t), so that
(10.5) F4 (t) = D (t) + R (t)
Now, the convolution theorem for the inverse Laplace transform states that if
It follows that
⎧⎛ 1 - e - s ⎞ 2 ⎫
⎪ ⎟ T1 (s)⎪⎬ =
F1 ( t ) = L {f 3 (s) f 4 (s)} = L ⎨⎜⎜
-1 -1
s ⎟
(10.8) ⎪⎩⎝ ⎠ ⎪⎭
t 1 1
= ∫ F(u) F4 (t - u) du = ∫ u F4 (t - u) du + ∫ (2 - u ) F4 (t - u) du
0 0 0
Now:
1 1 1
∫ u F4 (t - u) du = ∫ u (t - u) du . P (t - 0) + ∫ u R (t - u) du
0 0 0
⎧ t 1⎫
= P (t - 0) ⎨ - ⎬ + δ 4 R (t - δ 5 )
(10.9) ⎩ 2 3⎭ 0 < δ4, δ5 < 1
2 2 2
∫ (2 - u) F4 (t - u) du = ∫ (2 - u) (t - u) du . P (t - 1 + 0) + ∫ (2 - u ) R (t - u) du
1 1 1
⎧ t 2⎫
= P (t - 1 + 0) ⎨ - ⎬ + δ 6 R (t - δ 7 )
(10.10) ⎩2 3⎭ 1 < δ6, δ7 < 2
Hence:
⎧ t 1⎫ ⎧ t 2⎫
F1 (t) = P (t - 0) ⎨ - ⎬ + P (t - 1 + 0) ⎨ - ⎬ + δ 4 R (t - δ 5 ) + δ 6 R (t - δ 7 )
(10.11) ⎩ 2 3⎭ ⎩2 3⎭
⎧⎛ 1 − e - s ⎞ 2 ⎫
⎪⎜ ⎪
-1
F2 ( t ) = L ⎨⎜ ⎟
⎟ T2 (s )⎬
⎪⎩⎝ s ⎠ ⎪⎭
But according to the convolution theorem (10.6) – (10.7)
t
F2 (t) = ∫ F(u) . L-1 {T2 (s)} du
0
where F(u) is the function of (8.2) and L-1 {T2 (s)} was calculated in (7.2).
We deduce:
2
F2 (t) = L-1 {T2 (s)} ∫ F(u) du = L-1 {T2 (s)} =
0
δ2 2
= N (N + 1) 2
(11.1) 4π
⎧⎛ 2 ⎫
⎪⎜ 1 − e - s ⎞⎟ ⎪
F5 ( t ) = L ⎨⎜ s ⎟ T3 (s )⎬
-1
⎧ −s ⎞2 ⎫
-1 ⎪⎛
⎜
1− e
⎟ {T1 (s) + T2 (s) + T3 (s) }⎪⎬
v m ( t − 1) = L ⎨⎜ ⎟
⎪⎩⎝ s ⎠ ⎪⎭
= F1 (t) + F2 ( t ) + 2δAN 2ϑ + 1/2 + ε Log t
Hence:
⎧ t 1⎫ ⎧ t 2⎫
v m (t - 1) = P (t - 0) ⎨ - ⎬ + P (t - 1 + 0) ⎨ - ⎬ + δ 4 R (t - δ 5 ) + δ 6 R (t - δ 7 )
⎩ 2 3⎭ ⎩2 3⎭
δ2 2
+ N (N + 1) 2 + 2δAN 2ϑ +1/2 + ε Log t
4π
⎧ t 1⎫ ⎧ t 2⎫
= P (t - 0) ⎨ - ⎬ + P (t - 1 + 0) ⎨ - ⎬ +
(12.1) ⎩ 2 3⎭ ⎩2 3⎭
(log log N) 2 (log log N) 2
t δ 4 δ3e 3γ d(t - δ5 ) log log (t - δ3 ) + δ 6 δ3e 3γ t d(t - δ 7 ) log log (t - δ 7 )
N N
δ
+ 2 N 2 (N + 1) 2 + 2δAN 2ϑ +1/2 + ε Log t
4π
v (t - 1 + 0) + v (t - 1 - 0)
=
2
(log log N) 2
v( t ) = P(t) (t + 1) + δ3 2δ 4 e 3γ t d(t) log log (t + 1 - δ5 )
N
2 δ
− P( t ) + 2 N 2 (N + 1) 2 + 4δAN 2ϑ + 1/2 + ε Log t
(12.2) 3 2Π
The terms in P(t) only can be absorbed changing slightly the values of δ2 and δ3, so we ignore them in
what follows.
As in (12.2) we assume P(t) ≠ 0, then, according to (6.3) and (6.4) we must assume, in what follows, that
t is an even number.
Then
v( t ) = P .(t) .t + δ8 2e 3γ d(t) (log log t) 3 t 4 /5
2ϑ 1
1 54 +
+ δ2 t − 2At 5 10 Log t
(13.2) 2Π
The last term is negligible with respect to the preceding ones, and we obtain roughly:
4
1 5
− t
(13.3) 2Π
(13.2) proves unconditionally the truth of the Hardy – Littlewood conjectural formula for the binary
Goldbach problem, that
v (t) ~ P (t) . t
14.Miscellaneous bounds and majorations
log 2
log log t
lim sup d(t) = d o (t) = t (ref. [12])
(14.2) t →∞
We shall use in what follows very large values of t, so that we adopt the right hand side as an upper
bound for d (t).
Furthermore, the greatest value that v(t) can assume in the case that there is only one solution is 2 log2
t/2 (as was indicated in ref. [9]).
Consequently, if we know that
1,3203 t1/5
d( t ) <
(14.5) e 3γ (log log t ) 3
or if
From (14.6) and (14.2) follows that the Goldbach hypothesis is valid for every value of t such that
(14.8) d (t) ≥ d o ( t )
If we adopt the equality sign, we have an equation whose root is
Improved Lemma 1
∞ μ 2 (q ) μ 2 (d ) μ (q )
∑ C q (t) = ∑ ∑
q>N ϕ 2 (q ) d t ϕ(d ) N d<q ϕ 2 (q )
(L1) ( q , t ) =1
Now
μ (q) 1 1
∑ < ∑ < ∑
N d<q ϕ 2 (q ) N d<q
2
ϕ (q ) N d<q
2
ϕ (q )
(L2) ( q , t ) =1 ( q , t ) =1
n 5
≤ e γ log log n +
(L3) ϕ(n ) 2 log log n (ref. [13])
2
1 < 2 γ (log log n) 1
∑ ∑ e +5 ∑
N d<q
2
ϕ (q ) N d < q n 2 2 N d < q n log log n
2
∞ (log log u) 2 5
∞ du
< e2γ ∫ du + ∫
N d −1 u2 2 N d −1 u 2 log log u
(log log N / d ) 2 5 1
~ e2γ +
(L4) ( N / d) 2 log log N / d ⋅ N / d
2
μ (q) 2 γ (log log N / d ) 5 d
∑ < e ⋅d +
N d<q
2
ϕ (q ) N 2 log log N / d ⋅ N
(L5) ( q , t ) =1
Replacing in (L1)
(log log N) 2
< d(t) e 3γ log log t
N
We have that
1 1 1 q1 q 2
= =
A 2 − A1 2 πi h 2 − h1 2 πi (h 2 q1 - h1 q 2 )
q 2 q1
Hence
1 q1 q 2 q q
= ≤ 1 2
A 2 − A1 2 π h 2 q1 - h1 q 2 2π
e -A1 t − e − A 2 t ≤ 2
so that
e -A1 t − e − A 2 t q q
≤ 1 2
A 2 - A1 π
and
1 N N
1 N (N + 1) N ( N + 1) N 2 (N + 1) 2
< q q = π =
π q∑=1 q∑=1 1 2 2 2 4π
1 2
Lemma 3.
F( t ) ~ G ( t ) as t → ∞
is that
be continuous at t = 1.
By (4.2) we have:
1 − e −s
f (s) = L{Δϑ([u ])} = − ps
∑ log p ⋅ e =
s p
−s ⎫
−1 ⎧1 − e
Δϑ([u ]) = L ⎨ g N (s) ⎬
⎩ s ⎭
and
1 − e −s
L{Δϑ([u ])} = g N (s)
s
This is asymptotically equal to gN(s) as s→0, and besides, Δϑ([u]) fulfills condition [A]. Hence
N μ(q )
Δϑ([u ]) ~ L−1{g N (s)} = ∑ Cq (t)
q =1 ϕ(q )
This is the result used in § 7 bis.
REFERENCES:
(7). A. Peretti. The Goldbach problem (I). Notes on Number Theory and Discrete Math.
– Vol. 2 (1996) Nº1, p. 53-68.
- Idem, Vol. 2 (1996) Nº2, p. 1-26.
- Idem, Vol. 3 (1980) Nº4, p. 194-204.
(9). M. Baica. Clarifications of the author’s previous paper in the Goldbach conjecture –
Buletinul Stiintific Al Universitati “Politehnika” Din Timisoara – Tom 47 (61) 1
(2002).
(11). A. Prudnikov, Yu. Brychkov, O. Marichev. Integrals and Series. Vol. 4. Direct
Laplace Transforms – Vol. 5: Inverse Laplace transforms Gordon and Breach
Science Publishers (1992).
(14). R. Vaughan. The Hardy – Littlewood method – Cambridge Univ. Press. 1982-1997.
Fig. 1
ϑ ([u]) – ϑ ([u – 1])
4
3 log 13
log 7 log 11
2
log 5
log 2 log 3
1
0
1 2 l o g 23 l o g 34 5 lo g 5 6 7 lo g 7 8 9 10 11lo g 1112 13
lo g 13 14 15
Fig. 2
ϑ ([u + 1]) – ϑ ([u])
4
3
log 13
log 7 log 11
2
log 5
log 2 log 3
1
0
1 l o g 22 l o g 33 4 lo g 55 6 lo g 7 7 8 9 10l o g 11 11 12lo g 13 13 14 15
Fig. 3
ϑ ([12 – u]) – ϑ ([12 – u – 1])
4
3
log 11 log 7
2
log 5
log 3 log 2
1
0
l o g 11 lo g 7 lo g 5
1 2 3 4 5 6 7 8l o g 3
9 lo g 2 10 11 12 13 14 15
Fig. 4
ϑ ([10 – u]) – ϑ ([10 – u – 1])
4
3
log 7
2
log 5
log 3 log 2
1
0
1 2 l o g 73 4 lo g 5 5 6 lo g 3 7 lo g 2 8 9 10 11 12 13 14 15
Fig. 5
{ϑ ([u + 1]) – ϑ ([u])} {ϑ ([12 – u]) – ϑ ([12 – u – 1])}
4
log 7 . log 5 log 7 . log 5
3