Estimation of the number of twin primes by application of

 the Complementary Prime Sive

 

T a m á s D é n e s mathematician 

free-lance expert

e-mail: tdenest@freemail.hu

 

Published in the journal PU.M.A. Vol. 13 (2002), No. 3.

--- . ---

Abstract.  In the [1] paper we introduced the „Complementary Prime-Sieve” (C.P.S.) which gives necessary and sufficient conditions to generate composite numbers larger than 3 of the forms 6k-1 and 6k+1.  Here we give a proof of S.W.Golomb’s Theorem (see [3]) about the numbers of twin primes, then we deduce an approximate formulae for the T(N) numbers of twin primes in the 1-N interval, also based on C.P.S.

 

Mathematics Subject Classifications (2000).  11A07, 11A41, 11A51

 

 

1. Necessary and sufficient condition for the twin prime theorem

 

Theorem 1.

  are twin primes iff   where k=1,2,3,...

 

Proof:

According to the Theorem 1. in [1] (Every prime numbers larger than 3 are of the forms  6k+1 or  6k-1) there are two options:

 

(1)         p=6k-1                  q=p+2=6k+1

 

(2)         p=6k+1                 q=p+2=6k+3=3(2k+1)     but this is not a prime.

 

Consequently, only case (1) is possible. Inversely the proof is trivial.

Q.E.D.

 

Corollary 1.

(3)   pq = (6k-1)(6k+1) =   if  p,q  are twin primes. This implies the Theorem 2. below:

 

Theorem 2.

() has two prime factors iff  6k-1  and  6k+1 are twin primes.

 

Proof:

If  () has exactly two prime factors, then according to (3) these can only be 6k-1  and  6k+1. Thus 6k-1 and 6k+1 are twin primes.

If 6k-1 and 6k+1  are twin primes, then also according to (3), ( ) cannot have other prime factors.

Q.E.D.

 

Theorem 2. implies the theorem below regarding about the cardinality of twin primes:

 

The cardinality of twin primes is finite iff there is a K natural number for which, every  implies that (36k2 –1) have at least three prime factors.  It can be realized only if at least one of 6k-1 and 6k+1 is a composite number. According to the Theorem 2. in [1] (C.P.S. theorem) it is possible iff  k can be described as one of the following forms:

 

(4)         k=6uv+u+v     or     k=6uv-u-v    or      k=6uv-u+v    or     k=6uv+u-v   

 

With  these states we have proved the Theorem 3. below:

 

Theorem 3.

The cardinality of twin primes is finite iff there is a K  natural number for which every   can be written into one of the forms  (4). 

 

In this way we have also proved  S.W.Golomb’s theorem which he introduced in [3] as an E969 open problem :

„A necessary and sufficient condition that there be infinitely many twin primes is there be infinitely many numbers  k  not of the forms  (4).”

 

 

 

2. Estimation of the number of twin primes in the interval (1-N)

 

Let us order the natural numbers of the forms 6k-1 and  6k+1 according to Table 1. It can be seen easily that Table 1. contains all the natural numbers of the form 6k-1 in the second column and all the natural numbers of the form 6k+1 in the third column, so all the prime numbers as well.[1]

 

According to the Theorem 1. (see above) all the twin primes are in those lines of the table where there are prime numbers in the second and the third column as well. Thus the meaning of Theorem 3. is that in rows after K of Table 1. only one of the numbers can possibly be a prime number. Thus every k K row index can be put into one of the forms (4).

 

 

Table 1.

k
6k-1
6k+1
1
5
7
twin prime
2
11
13
twin prime
3
17
19
twin prime
4
23
25
5
29
31
twin prime
6
35
37
7
41
43
twin prime
8
47
49
9
53
55
10
59
61
twin prime
11
65
67
12
71
73
twin prime
13
77
79
14
83
85
15
89
91
16
95
97
17
101
103
twin prime
18
107
109
twin prime
19
113
115
20
119
121
...
...
...
K
6K-1
6K+1
...
...
...

 

Now, we will show that in Table 1. there are rows in which there are certainly not any twin primes. Firstly we examine the cases k=5r, k=5r+1, k=5r+2, k=5r+3, k=5r+4, which cases obviously generate all  k  row indexes.

 

-If   k=5r  (r=1,2,3,...), then  6k-1=30r-1  and  6k+1=30r+1

 These two sequences contain each fifth row of the Table 1. where there are twin primes for example in the following (see Table 1.):   r= 1,   2,   5,   6,...                                                                 k= 5, 10, 25, 30,...

 

-If k=5r+1 (r=1,2,3,...), then  6k-1=30r+5=5(6r+1) which is not a prime number, so these rows definitely don’t contain twin primes (see Table 1.). For example:  r=1,   2,   3,   4,...        k=6, 11, 16, 21,...  

 

-If   k=5r+2 (r=0,1,2,3,...), then   6k-1=30r+11  and  6k+1=30r+13 

 These two sequences also come up in each fifth rows of Table 1. starting from the second row, where there are twin primes. For example (see Table 1.):   r=0,   1,   2,   3,   6,...                k=2,   7, 12, 17,  32,...

 

 

-If   k=5r+3 (r=0,1,2,3,...), then   6k-1=30r+17  and  6k+1=30r+19

 These two sequences also come up in each fifth rows of Table 1. starting from the 3rd row, where there are twin primes. For example (see Table 1.):   r=0,   3,   4,   6,...                           k=3, 18, 23, 33,...

 

-If   k=5r+4 (r=0,1,2,3,...), then  6k-1=30r+23  and  6k+1=30r+25=5(6r+5), which is not a prime number, so in these rows there will certainly not be any twin primes.

 

According to the above mentioned in the rows k=5r+1, and k=5r+4 of Table 1. there are not any twin primes  (k=4,6,9,11,14,16,...) thus   (5)    all the twin primes are contained in the  k=5r,  k=5r+2,  k=5r+3    (k=0,1,2,3,...)    rows of the Table 1.

 

Consequently there is an upper bound for the number of twin primes (denote T(N)):                                                            

 

The number of rows in Table 1. to N (denote ) is:    

According to the previous deduction twin primes can occur in maximum  part of these rows:

 

(6)                                                  T(N)

 

Let us examine the rows of the forms  (5) of Table 1. with the aid of  C.P.S.

There are definitely not any twin primes in rows (5), if  the  k  rowindex can be put into one of the forms of (4). Because the (4) relations are symmetric for u,v we can examine the cases when u=constant.

 

 (7)   u=1 and  k=5r= 6uv+u+v k=7v+1 holds, if  v=5a+2    k=35a+15

 (8)   u=1 and  k=5r= 6uv-u-v   k=5v-1    5r=5v-1, it can never be realised

 (9)   u=1 and  k=5r= 6uv+u-v  k=5v+1   5r=5v+1, it can never be realised

(10)  u=1 and  k=5r= 6uv-u+v   k=7v-1  holds, if  v=5a+3,   k=35a+20

 

(11)  u=1 és  k=5r+2= 6uv+u+v k=7v+1 holds, if v=5a+3,  k=35a+22

(12)  u=1 és  k=5r+2= 6uv-u-v   k=5v-1    5r=5v-3, it can never be realised

(13)  u=1 és  k=5r+2= 6uv+u-v  k=5v+1   5r=5v-1, it can never be realised

(14)  u=1 és  k=5r+2= 6uv-u+v  k=7v-1  holds, if v=5a+4,  k=35a+27

 

(15)  u=1 és  k=5r+3= 6uv+u+v k=7v+1 holds, if v=5a+1,  k=35a+8

(16)  u=1 és  k=5r+3= 6uv-u-v   k=5v-1    5r=5v-4, it can never be realised

(17)  u=1 és  k=5r+3= 6uv+u-v  k=5v+1   5r=5v-2, it can never be realised

(18)  u=1 és  k=5r+3= 6uv-u+v  k=7v-1   holds, if v=5a+2,  k=35a+13

 

As a consequence of deductions  (7), (10), (11), (14), (15), (18) there are not any twin primes in    part of rows (5) containing twin primes. In such a way we can make formulae (6) more precisely:

 

(19)                                  T(N)

 

 

The case  u=2  leads us to the general formulae.

 

(20)  u=2 and  k=5r= 6uv+u+v  k=13v+2  holds, if  v=5a+1,   k=65a+15

(21)  u=2 and  k=5r= 6uv-u-v    k=11v-2   holds, if  v=5a+2,  k=55a+20

(22)  u=2 and  k=5r= 6uv+u-v   k=11v+2  holds, if  v=5a+3,  k=55a+35

(23)  u=2 and  k=5r= 6uv-u+v   k=13v-2   holds, if  v=5a+4,  k=65a+50

 

(24)  u=2 and  k=5r+2= 6uv+u+v k=13v+2 holds, if  v=5a,  k=65a+2

(25)  u=2 and  k=5r+2= 6uv-u-v  k=11v-2   holds, if  v=5a+4,  k=55a+42

(26)  u=2 and  k=5r+2= 6uv+u-v k=11v+2  holds, if  v=5a,  k=55a+2

(27)  u=2 and  k=5r+2= 6uv-u+v k=13v-2   holds, if  v=5a+3,  k=65a+37

 

(28)  u=2 and  k=5r+3= 6uv+u+v k=13v+2  holds, if  v=5a+2,  k=65a+28

(29)  u=2 and  k=5r+3= 6uv-u-v   k=11v-2  holds, if  v=5a,  k=55a-2

(30)  u=2 and  k=5r+3= 6uv+u-v  k=11v+2 holds, if  v=5a+1,  k=55a+13

(31)  u=2 and  k=5r+3= 6uv-u+v  k=13v-2  holds, if  v=5a,  k=65a-2

 

Connections (20)-(31) show that  starting from a certain row (which rows are all different) each 55th and 65th rows definitely do not contain twin primes.

That is according to (19),  part of the rows containing twin primes potentially fall out as well. We can see that it’s true for any  u in general that su part of the rows containing twin primes potentially definitely do not contain twin primes, where

 

(32)                                        

 

We showed in C.P.S. method that if we want to generate prime numbers until N then we have to „run”  u  from 1 until   . Thus, the number of rows without twin primes below  1-N is:

 

(33)                             

 

But of the statement explained in C.P.S. method, all the  u,v pairs generate  k  by  m(N) multiplicity in K(N) steps. Thus, the number of rows not containing twin primes (denote S(N)) is the following:

 (34)                   S(N)    

 

This causes the estimation from (19) to T(N) below:

 (35)                  T(N)

 

 

(36)                 

 

 

We use the estimated formulae below  (see [2]  2.pp.):

(37)                      

 

(38)                  

 

From (38) and (36) the result is the following:                                                                                                                                                                                                                                                                                                         

 

(39)                  

 

 

(40)                    =   

 

(41)            

 

 

(42)                          

 

The result of Hardy and Littlewood (see [4]) for the number of twin primes until  x  is the following (x is an arbitrary natural number and T(x) the number of twin primes until  x):

 

(41)                 T(x)

                                 where    means the prime numbers are greater than  2 until  x.

 

Let us denote the real number of twin primes until  N to  RT(N). Than the comparison of the results above are contained in Table 2.

 

Table 2.

N
RT(N)
T(N)


T(N) rel.err.
(%)

T(x)
T(x) rel.err.
(%)
100
0.1
8
8
-
6
25
1.000
0.165
35
42
20
27
23
2.000
0.23
61
75
22
45
26
3.000
0.27
81
101
24
61
25
4.000
0.29
103
131
27
76
26
5.000
0.30
126
156
23
90
29
10.000
0.35
205
268
30
155
24
30.000
0.42
467
660
41
372
21
50.000
0.44
705
964
36
563
21
1.000.000
0.56
8.169
9.774
19
6.915
16
2.000.000
0.58
14.871
16.102
8
12.541
16
3.000.000
0.59
20.933
20.734
1
17.803
15
4.000.000
0.60
26.861
27.451
2
22.847
15
5.000.000
0.61
32.464
31.398
3
27.739
15
8.000.000
0.62
48.619
46.103
5
41.797
14

 

Acknowledgment

 

I am happy to thank  Dr S.W.Golomb for the informative letter in which he explained that he has published the theorem in [3]: "A necessary and sufficient condition that there be infinitely many twin primes is that there be infinitely many numbers  n  not of the form  ."  and he wrote up-to-date historical background to it. 

 

References

[1] T.Dénes:  Complementary prime-sive, PU.M.A. Volume 12, Number 2., May 2002.

http://www.titoktan.hu/_raktar/_e_vilagi_gondolatok/PUMA-CPS.htm

 

[2] P. Erdős, J. Surányi: Selected chapters from the theory of numbers, (in Hungarian) Polygon, Szeged, 1996.

 

[3]  S.W.Golomb: Problem E969: American Math. Monthly, (vol 58. no.5. p 338.) 1951.

 

[4] S.W. Golomb: The Twin Prime Constant, American Math. Monthly, (vol 67. no.8. p 767-769) 1960.

 

[5] W.Sierpinski: A Selection of Problems in the Theory of Numbers, Pergamon Press, Oxford, U.K., 1964.

 

[6] Maria Suzuki: Alternative Formulations of the Twin Prime Problem, The Amer. Math. Monthly, Vol. 107. january 2000.

 



[1] This statement is the consequence of the theorem: „All prime numbers are of the forms of 6k-1 or 6k+1.”  (see  [1]  Theorem 1.)