On the
Infinite Series Characterizing
the
Elimination of Twin Prime Candidates
by
Dennis R. Martin
DP Technology Corp.,
dennis.martin@dptechnology.com
BSME,
Department of
Mathematics,
Copyright © 2006 by Dennis R. Martin
ALL RIGHTS RESERVED
No
part of this document may be reproduced, retransmitted, or redistributed by any
means, without the expressed written consent of Dennis R. Martin.
This document is available in other formats. See www.primenace.com.
Abstract
The pattern of the composite numbers that have a particular lowest prime factor repeats over intervals equal to the primorial of that lowest prime factor. The number of composites having that prime for their lowest factor is constant over those primorial intervals, and the value of that constant for each prime is directly related to the value of the previous prime and its constant composite to primorial ratio.
Those primorial patterns apply to twin primes. The composite numbers that eliminate twin prime candidates can be counted in terms of their lowest prime factors. Those twin prime eliminations repeat over primorial intervals such that the rate of elimination can be represented by an infinite series over all primes 5 and greater. We calculate the partial sums for that series for the first 15 million primes and examine the implications of the convergence of that infinite series.
1 Primorial
Soup
It is known that the pattern of composite numbers repeats with a period equal to the primorial of each prime factor. Dickson [1] refers to remarks by H. J. S. Smith and theorems by J. DesChamps regarding this property, and Weisstein [2] makes note of it. The primorial is analogous to a factorial applied to the sequence of prime numbers. The primorial for the prime pn is the product of all primes up to and including pn, and it is denoted as pn#. By the definition of factorial, p1# = 2, and then for every prime greater than 2:
pn# = pn pn1# (1)
Since pn itself is prime and there can be no composites less than pn having pn for a factor, it is intuitive to start the first primorial interval at pn + 1, such that it ends at pn + pn#. Because in some ways we can think of composite numbers as molecules composed of their constituent prime factor atoms and arranged in a lattice across these primorial intervals, the value pn + pn# will be called the first atomic boundary and labeled an.
If we make a function Tn(x) to count the total number of composites that have pn as their lowest prime factor, then we find that the count in the first and all subsequent primorial intervals is a constant. This constant composite to primorial ratio shall be labeled rn, where rn = Tn(an).
Table
1: Composite to Primorial Ratio and Ratio Summation for the First Twelve Primes
|
n |
pn |
pn# |
an = pn
+ pn# |
rn
= Tn(an) |
Σ (Numerator) |
|
1 |
2 |
2 |
4 |
1 |
1 |
|
2 |
3 |
6 |
9 |
1 |
4 |
|
3 |
5 |
30 |
35 |
2 |
22 |
|
4 |
7 |
210 |
217 |
8 |
162 |
|
5 |
11 |
2310 |
2321 |
48 |
1830 |
|
6 |
13 |
30030 |
30043 |
480 |
24270 |
|
7 |
17 |
510510 |
510527 |
5760 |
418350 |
|
8 |
19 |
9699690 |
9699709 |
92160 |
8040810 |
|
9 |
23 |
223092870 |
223092893 |
1658880 |
186597510 |
|
10 |
29 |
6469693230 |
6469693259 |
36495360 |
5447823150 |
|
11 |
31 |
200560490130 |
200560490161 |
1021870080 |
169904387730 |
|
12 |
37 |
7420738134810 |
7420738134847 |
30656102400 |
6317118448410 |
It is readily apparent in Table 1 that each composite to primorial ratio is related to the previous prime and its ratio by the following:
rn = rn1 (pn1 1) (2)
This relationship can be rigorously proven by observing the number of complete residue systems modulus the primorial of each of the previous primes up to pn that exist within a pn# primorial interval. Note also that Sloane [3] gives this as sequence A005867 in the On-Line Encyclopedia of Integer Sequences (OEIS) with the comment that it corresponds to the local minima of Eulers phi (totient) function.
2 To Be, Or
Not To Be
Twin primes are primes of the form (p, p + 2), such as (3, 5), (5, 7), (11, 13), (17, 19), and so on. The twin prime conjecture is a currently unproven conjecture stating that there are infinitely many twin primes. It is well established that all twin primes after the first are of the form 6n ± 1.
Every pair of positive integers (6n 1, 6n + 1) can be considered to be a twin prime candidate, where the value 6n is the center post. In order to determine when both members of a twin prime candidate are indeed prime, it is important then to understand when and how one or both of them are not. A candidate is eliminated when either or both of the numbers adjacent to its center post are composite. Let us call such a composite an eliminating composite.
It is fairly easy to establish that the primes p1 = 2 and p2 = 3 cannot produce any eliminating composites, and that all composites having p3 = 5 or greater for their lowest prime factor are eliminating composites. See Table 2, for example, and notice that all of the lowest prime factorizations are of the form 5 (6n ± 1) and that starting above 5 there are two such composites in each interval of p3# = 30, which illustrates the relationship presented in the previous section.
Table 2: Eliminations of Twin Prime Candidates by p3 = 5 in the Interval up to 217
|
Candidate Center Post |
Eliminating Composite |
Prime Factorization |
Lowest Prime Factorization |
Residues |
|
|
(mod 30) |
(mod 6) |
||||
|
24 |
25 |
5 5 |
5 (6
1) |
25 |
1 |
|
36 |
35 |
5 7 |
5 (6
+ 1) |
5 |
5 |
|
54 |
55 |
5 11 |
5 (12
1) |
25 |
1 |
|
66 |
65 |
5 13 |
5 (12
+ 1) |
5 |
5 |
|
84 |
85 |
5 17 |
5 (18
1) |
25 |
1 |
|
96 |
95 |
5 19 |
5 (18
+ 1) |
5 |
5 |
|
114 |
115 |
5 23 |
5 (24
1) |
25 |
1 |
|
126 |
125 |
5 5 5 |
5 (24
+ 1) |
5 |
5 |
|
144 |
145 |
5 29 |
5 (30
1) |
25 |
1 |
|
156 |
155 |
5 31 |
5 (30
+ 1) |
5 |
5 |
|
174 |
175 |
5 5 7 |
5 (36
1) |
25 |
1 |
|
186 |
185 |
5 37 |
5 (36
+ 1) |
5 |
5 |
|
204 |
205 |
5 41 |
5 (42
1) |
25 |
1 |
|
216 |
215 |
5 43 |
5 (42
+ 1) |
5 |
5 |
There are 30 / 6 = 5 candidates out of every interval of 30, so that means that three candidates out of every interval of 30 are not eliminated by p3 = 5. If those candidates are to be eliminated, they must be eliminated by a composite having p4 = 7 or higher as its lowest prime factor.
The composites having 7 as their lowest prime factor have a pattern that repeats over intervals of p4# = 210, and there are r4 = 8 of them within each such interval. Up to a4 = 217, for instance, those eight composites are (49, 77, 91, 119, 133, 161, 203, 217) while in the next interval to 427 they are (259, 287, 301, 329, 343, 371, 413, 427). This pattern is summarized in Table 3.
Table 3: Factorization of Any p4# = 210 Primorial Interval for the Lowest Prime Factor p4 = 7
|
Primorial Interval Member |
Lowest Prime Factorization |
Residues |
||
|
Modulo
p4# |
Modulo
p3# |
Modulo
p2# |
||
|
210n + 49 |
7 (30n + 7) |
49
(mod 210) |
19 (mod 30) |
1 (mod 6) |
|
210n + 77 |
7 (30n + 11) |
77
(mod 210) |
17 (mod 30) |
5 (mod 6) |
|
210n + 91 |
7 (30n + 13) |
91
(mod 210) |
1 (mod 30) |
1 (mod 6) |
|
210n + 119 |
7 (30n + 17) |
119
(mod 210) |
29 (mod 30) |
5 (mod 6) |
|
210n + 133 |
7 (30n + 19) |
133
(mod 210) |
13 (mod 30) |
1 (mod 6) |
|
210n + 161 |
7 (30n + 23) |
161
(mod 210) |
11 (mod 30) |
5 (mod 6) |
|
210n + 203 |
7 (30n + 29) |
203
(mod 210) |
23 (mod 30) |
5 (mod 6) |
|
210(n + 1) + 7 |
7 (30(n + 1) + 1) |
7
(mod 210) |
7 (mod 30) |
1 (mod 6) |
There are 210 / 6 = 35 candidates out of every interval of 210. With an interval starting above 7, there would be 2 (210 / 30) = 14 eliminating composites produced by p3 = 5. That matches the count from Table 2, which conveniently listed them up to 217. Add those 14 to the 8 eliminating composites produced by p4 = 7 and that is still only 22. At least 35 22 = 13 candidates are not eliminated by p3 or p4 in each interval of 210. If those twin prime candidates are to be eliminated, it has to be by a composite with a lowest prime factor of p5 = 11 or higher.
But
there would only be 13 candidates left if all 22 of those eliminating
composites corresponded to exactly one unique elimination of a twin prime. They
do not. There are cases of double
eliminations where both of the numbers in the candidate pair are composite.
Two such double eliminations can be seen in the preceding tables. The value 205
in Table 2 and 210n + 203 from Table 3,
when n = 0, will both be adjacent to
the center post 204, while the 215 from Table 2 and 210(n + 1) + 7 from Table 3, when n
= 0, will doubly eliminate the center post 216.
3 The Double
Elimination Round
Let E″m,n represent the double eliminations made by a lower prime pm in combination with the prime pn. We can calculate the twin prime count p2(x) by subtracting the total composite count contributed by each lowest prime factor and then adding back in the double eliminations within that range. For n ≥ 3 such that pn ≤ x1/2 and 3 ≤ m ≤ (n 1) that count is as follows, where the initial 1 is to account for the first twin prime (3, 5) and the \ in (x \ 6) indicates integer division.
p2(x) = 1 + (x \ 6) STn(x) + SE″m,n(x) (3)
If we account for the eliminations of twin prime candidates by lower prime factors pm < pn in this way, then pn2 is the earliest that a prime factor pn can generate a previously unaccounted for eliminating composite. For example, Table 2 lists the composites 35 and 175, which have both 5 and 7 as factors, as being eliminations by p3 = 5. Those composites are not included in the factorizations in Table 3. Table 1 also includes 55, while Table 2 includes the factorization representing 77, both of which have 11 as a factor. Since those composites are counted there, the first composite that has 11 as a factor that needs to be counted in one of the E″m,5 components assigned to the prime factor p5 is its square 121.
To give an example using (3), let us evaluate the twin prime count up to x = 217. For this we need to consider eliminations by p5 = 11 and p6 = 13, because 112 and 132 are less than 217. The square of 17 is 289, though, so we do not need to consider p7, and since 13 17 = 221 we know that 169 is the only eliminating composite that can be attributed to p6 in this interval. On the other side of its center post is the prime 167, so it can be thought of as a single elimination.
The eliminating composites less than 217 having 11 for their lowest prime factor are 121, 143, 187, and 209. The corresponding candidate center posts are 120, 144, 186, and 210. The factorization for 119 is listed in Table 3, and that is adjacent to 120, so it is a double elimination with p4, and 144 and 186 are adjacent to 145 and 185, which would be counted as eliminations by p3. The only one of the four candidates that is singly eliminated by p5 is the center post at 210.
For x = 217, as can be deduced from the discussion above, E″3,4(217) = 2; E″3,5(217) = 2; and E″4,5(217) = 1. There were no double eliminations involving p6 so E″m,6(217) = 0.
p2(217) = 1 + (217 \ 6) (14 + 8 + 4 + 1) + (2 + 2 + 1) = 1 + 36 27 + 5 = 15
Those fifteen, along with the next twin prime after 217, are listed in Table 4.
Table 4: The First Fifteen Twin Primes Up to 217, Plus One More
|
p2 |
Center Post |
Twin Prime |
|
p2 |
Center Post |
Twin Prime |
|
1 |
4 |
(3, 5) |
|
9 |
102 |
(101, 103) |
|
2 |
6 |
(5, 7) |
|
10 |
108 |
(107, 109) |
|
3 |
12 |
(11, 13) |
|
11 |
138 |
(137, 139) |
|
4 |
18 |
(17, 19) |
|
12 |
150 |
(149, 151) |
|
5 |
30 |
(29, 31) |
|
13 |
180 |
(179, 181) |
|
6 |
42 |
(41, 43) |
|
14 |
192 |
(191, 193) |
|
7 |
60 |
(59, 61) |
|
15 |
198 |
(197, 199) |
|
8 |
72 |
(71, 73) |
|
16 |
228 |
(227, 229) |
From the periodic primorial property presented in part 1, it follows that the pattern of double eliminations of twin prime candidates will repeat over intervals of the primorial of the higher of the two lowest prime factors involved in each of the eliminating composites.
4 Double
Jeopardy: The Primorial Patterns of Double Eliminations
We previously noted that p3 and p4 both eliminate the twin prime candidates centered at 204 and 216. If we check every multiple of p4# = 210 thereafter we find a similar double elimination. 413 is 7 59 and 415 is 5 83, while 425 = 52 13 and 427 = 7 61, thus the center posts at 414 and 426 are doubly eliminated by p3 and p4. 623 = 7 89 and 625 = 54, while 635 is 5 127 and 637 is 72 13, thus the center posts at 624 and 636 are doubly eliminated by p3 and p4 also. These results are summarized in Table 5.
Table 5: Double Eliminations by p3 = 5 and p4 = 7 within Intervals of p4# = 210
|
Center
Post |
Interval Member |
Factorization |
Residues (modulus) |
||
|
(210) |
(30) |
(6) |
|||
|
210n + 204 |
210n + 203 |
7 (30n + 29) |
203 |
|
5 |
|
210n + 205 |
5 (42n + 41) |
|
25 |
1 |
|
|
210(n + 1) + 6 |
210(n + 1) + 5 |
5 (42(n + 1) + 1) |
|
5 |
5 |
|
210(n + 1) + 7 |
7 (30(n + 1) + 1) |
7 |
|
1 |
|
120 was a double elimination by p4 and p5, and 144 and 186 were double eliminations by p3 and p5. The primorial p5# = 2310. If we check by 2430, we see that 2429 = 7 347 while 2431 = 11 13 17. For 2454 and 2496, obviously to one side is a number divisible by 5, while the values on the other side, 2453 and 2497, both have 11 for their lowest prime factor. There are many more double eliminations for p5 across intervals of its primorial. They are listed in Tables 6 and 7.
Table 6: Double Eliminations by p3 = 5 and p5 = 11 within Intervals of p5# = 2310
|
Center
Post |
Interval Member |
Factorization |
Residues (modulus) |
||
|
(2310) |
(30) |
(6) |
|||
|
2310n + 144 |
2310n + 143 |
11
(210n + 13) |
143 |
|
5 |
|
2310n + 145 |
5 (462n + 29) |
|
25 |
1 |
|
|
2310n + 186 |
2310n + 185 |
5 (462n + 31) |
|
5 |
5 |
|
2310n + 187 |
11
(210n + 17) |
187 |
|
1 |
|
|
2310n + 474 |
2310n + 473 |
11
(210n + 43) |
473 |
|
5 |
|
2310n + 475 |
5 (462n + 95) |
|
25 |
1 |
|
|
2310n + 516 |
2310n + 515 |
5 (462n + 103) |
|
5 |
5 |
|
2310n + 517 |
11
(210n + 47) |
517 |
|
1 |
|
|
2310n + 804 |
2310n + 803 |
11
(210n + 73) |
803 |
|
5 |
|
2310n + 805 |
5 (462n + 161) |
|
25 |
1 |
|
|
2310n + 1134 |
2310n + 1133 |
11
(210n + 103) |
1133 |
|
5 |
|
2310n + 1135 |
5 (462n + 227) |
|
25 |
1 |
|
|
2310n + 1176 |
2310n + 1175 |
5 (462n + 235) |
|
5 |
5 |
|
2310n + 1177 |
11
(210n + 107) |
1177 |
|
1 |
|
|
2310n + 1506 |
2310n + 1505 |
5 (462n + 301) |
|
5 |
5 |
|
2310n + 1507 |
11
(210n + 137) |
1507 |
|
1 |
|
|
2310n + 1794 |
2310n + 1793 |
11
(210n + 163) |
1793 |
|
5 |
|
2310n + 1795 |
5 (462n + 359) |
|
25 |
1 |
|
|
2310n + 1836 |
2310n + 1835 |
5 (462n + 367) |
|
5 |
5 |
|
2310n + 1837 |
11
(210n + 167) |
1837 |
|
1 |
|
|
2310n + 2124 |
2310n + 2123 |
11
(210n + 193) |
2123 |
|
5 |
|
2310n + 2125 |
5 (462n + 425) |
|
25 |
1 |
|
|
2310n + 2166 |
2310n + 2165 |
5 (462n + 433) |
|
5 |
5 |
|
2310n + 2167 |
11
(210n + 197) |
2167 |
|
1 |
|
Table 7: Double Eliminations by p4 = 7 and p5 = 11 within Intervals of p5# = 2310
|
Center
Post |
Interval Member |
Factorization |
Residues (modulus) |
||
|
(2310) |
(210) |
(6) |
|||
|
2310n + 120 |
2310n + 119 |
7 (330n + 17) |
|
119 |
5 |
|
2310n + 121 |
11
(210n + 11) |
121 |
|
1 |
|
|
2310n + 342 |
2310n + 341 |
11
(210n + 31) |
341 |
|
5 |
|
2310n + 343 |
7 (330n + 49) |
|
133 |
1 |
|
|
2310n + 582 |
2310n + 581 |
7 (330n + 83) |
|
161 |
5 |
|
2310n + 583 |
11
(210n + 53) |
583 |
|
1 |
|
|
2310n + 1728 |
2310n + 1727 |
11
(210n + 157) |
1727 |
|
5 |
|
2310n + 1729 |
7 (330n + 247) |
|
49 |
1 |
|
|
2310n + 1968 |
2310n + 1967 |
7 (330n + 281) |
|
77 |
5 |
|
2310n + 1969 |
11
(210n + 179) |
1969 |
|
1 |
|
|
2310n + 2190 |
2310n + 2189 |
11
(210n + 199) |
2189 |
|
5 |
|
2310n + 2191 |
7 (330n + 313) |
|
91 |
1 |
|
If we count up the entries in the three previous tables, we have 2 double eliminations by p3 and p4 within Intervals of p4#; 12 double eliminations by p3 and p5 within Intervals of p5#; and 6 double eliminations by p4 and p5 within intervals of p5#. Is there a pattern to those values? Three counts are not many to work with. Fortunately, since we are only concerned with tabulating the counts for a certain number of lowest prime factors, and not with completely factorizing every candidate pair, it is easy to construct a program that can quickly give us more such counts.
Table 8 shows the numerical results of just such a program which considered prime factors up to p11 = 31 within the interval to a11 + 2 = 200560490163. It was necessary to go to an + 2 for each count in case an + 1 was a potential center post that was eliminated by an + 2 being a composite of one of the factors under consideration. To shorten the notation, let a″n = an + 2.
Table 8: Double Elimination Counts within Primorials, E″m,n(a″n),
With
Lower Prime pm Along the
Left and Higher Prime pn
Across the Top
|
m,n |
|
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
|
|
p |
7 |
11 |
13 |
17 |
19 |
23 |
29 |
31 |
|
3 |
5 |
2 |
12 |
120 |
1440 |
23040 |
414720 |
9123840 |
255467520 |
|
4 |
7 |
|
6 |
60 |
720 |
11520 |
207360 |
4561920 |
127733760 |
|
5 |
11 |
|
|
30 |
360 |
5760 |
103680 |
2280960 |
63866880 |
|
6 |
13 |
|
|
|
270 |
4320 |
77760 |
1710720 |
47900160 |
|
7 |
17 |
|
|
|
|
2970 |
53460 |
1176120 |
32931360 |
|
8 |
19 |
|
|
|
|
|
44550 |
980100 |
27442800 |
|
9 |
23 |
|
|
|
|
|
|
757350 |
21205800 |
|
10 |
31 |
|
|
|
|
|
|
|
15904350 |
|
|
Total |
2 |
18 |
210 |
2790 |
47610 |
901530 |
20591010 |
592452630 |
For pm = 5 and pn = 23, double elimination number 414720 occurs at a9 + 1 = 223092894.
For all entries in the Table 8 that have an entry to their left such that 3 < m < (n 1), the following relationship holds:
E″m,n(a″n)
= (pn1 1) E″m,n1(a″n1) (4)
The logic for proving equation (4) would be the same as that used to prove equation (2), namely that only certain residues in the complete residue system of previous primorials are lined up in the correct positions necessary to produce an eliminating composite when the members of those previous primorials are multiplied by the next prime to initially generate the primorial interval corresponding to that next prime.
What about the entries in Table 8 that do not have a value to their left on which to base the next double elimination count? To find out, let us use E″n to represent the total number of double eliminations that are already attributed to a lower prime pm < pn within each pn# primorial interval, and then E′n to represent those eliminations (single or double) that can be attributed to pn itself within each pn# primorial interval.
E″n = SE″m,n(a″n) {for m = 3 to (n 1) (5)
E′n = rn E″n (6)
For example, E″5 = E″3,5(a″5) + E″4,5(a″5) = 12 + 6 = 18, which means that E′5 = 48 18 = 30.
Once again a word of caution; E′n does not represent the number of single eliminations. Some of the eliminations counted by E′n may be double eliminations that are later attributed to a higher prime. That is, they may be included in a subsequent E″q term for a prime pq > pn. For instance, 168 was a single elimination by p6, as is 168 + p6#, since 30197 is prime and 30199 = 13 2323. But 168 + 2p6# is a double elimination because 60227 = 229 263.
What E′n does represent is the unique twin prime elimination count. It is the number of twin prime eliminations over intervals of the primorial pn# that we get when we take out those that were double counted because the other half of the pair was eliminated by an earlier prime. For example, E′4 = 8 2 = 6. The two that are subtracted out initially are the eliminations of 204 and 216, which are counted in the E′3 component for p3. One of those six that remain will be counted again in E″5 (that being the elimination of 120) but then that over-count is corrected in the computation of E′5. Table 9 shows the E′n value for the first eleven primes. Since there are no eliminations by p2 whatsoever we can say that E″2 = 0 so that E′3 = r3 = 2.
Table
9: Unique Twin Prime Elimination Counts Over Primorial Intervals
|
n |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
|
pn |
5 |
7 |
11 |
13 |
17 |
19 |
23 |
29 |
31 |
|
rn |
2 |
8 |
48 |
480 |
5760 |
92160 |
1658880 |
36495360 |
1021870080 |
|
E″n |
0 |
2 |
18 |
210 |
2790 |
47610 |
901530 |
20591010 |
592452630 |
|
E′n |
2 |
6 |
30 |
270 |
2970 |
44550 |
757350 |
15904350 |
429417450 |
Notice that the E′n values in Table 9 all match the values that appear in Table 8 as the double elimination count for that prime when paired with the next higher prime.
E″n1,n(a″n) = E′n1 (7)
When equation (7) is combined with those presented earlier, it is straightforward to derive that:
E′n = (pn1 2) E′n1 (8)
This relationship is clearly demonstrated in the last row in Table 9. The integer sequences formed by E″n and by E′n have been submitted to the On-Line Encyclopedia of Integer Sequences as sequences A121407 and A121406, respectively.
4 Double
Vision: The Implications of Primorial Patterns on Twin Primes
Equation (8) allows us to calculate a twin prime elimination ratio, (E′n / pn#), for each prime factor. Starting with (E′3 / p3#) = 2 / 30, we can easily calculate subsequent ratios.
(E′n / pn#)
= (E′n1 / pn1#) (pn1 2) / pn (9)
The first several values of the ratios formed by equation
(9) are listed in Table 10.
Table
10: Twin Prime Elimination to Primorial Ratios and Ratio Summation
|
n |
pn |
E′n |
pn# |
» (E′n / pn#) |
» Σ (E′n / pn#) |
|
3 |
5 |
2 |
30 |
0.06666667 |
0.06666667 |
|
4 |
7 |
6 |
210 |
0.02857143 |
0.09523810 |
|
5 |
11 |
30 |
2310 |
0.01298701 |
0.10822511 |
|
6 |
13 |
270 |
30030 |
0.00899101 |
0.11721612 |
|
7 |
17 |
2970 |
510510 |
0.00581771 |
0.12303383 |
|
8 |
19 |
44550 |
9699690 |
0.00459293 |
0.12762676 |
|
9 |
23 |
757350 |
223092870 |
0.00339477 |
0.13102153 |
|
10 |
29 |
15904350 |
6469693230 |
0.00245829 |
0.13347982 |
|
11 |
31 |
429417450 |
200560490130 |
0.00214109 |
0.13562091 |
|
12 |
37 |
12453106050 |
7420738134810 |
0.00167815 |
0.13729905 |
|
13 |
41 |
435858711750 |
304250263527210 |
0.00143257 |
0.13873162 |
|
14 |
43 |
16998489758250 |
13082761331670030 |
0.00129930 |
0.14003093 |
|
15 |
47 |
696938080088250 |
614889782588491410 |
0.00113344 |
0.14116436 |
|
16 |
53 |
31362213603971250 |
32589158477190044730 |
0.00096235 |
0.14212671 |
One application for these ratios is to estimate values of the twin prime counting function, p2(x). For x ≥ 4 and n ≥ 3 such that pn < x1/2 the estimation is given by:
p2(x) » 1 + (x / 6) S ((x pn) (E′n / pn#)) (10)
For p2(217), that value would be would be calculated as:
1 + (217 / 6) 212 2 / 30 210 6 / 210 206 30 / 2310 204 270 / 30030
The
result is p2(217) » 12.5, which is about a 16.7% error from the
correct count of 15 that we found earlier. Table 11 shows the results from
equation (10) for powers of ten up to 1016. The values were
calculated using the primes up to 108 as published by
Table
11: Twin Prime Counting Function Es
|
x |
p2(x) Estimated |
p2(x) Actual[a] |
% Error |
|
101 |
2.7 |
2 |
35.0 |
|
102 |
8.7 |
8 |
8.8 |
|
103 |
33.2 |
35 |
5.1 |
|
104 |
194.4 |
205 |
5.2 |
|
105 |
1234.6 |
1224 |
0.9 |
|
106 |
8662.4 |
8169 |
6.0 |
|
107 |
63973.6 |
58980 |
8.5 |
|
108 |
489460.4 |
440312 |
11.2 |
|
109 |
3873936.6 |
3424506 |
13.1 |
|
1010 |
31382177.5 |
27412679 |
14.5 |
|
1011 |
259468905.1 |
224376048 |
15.6 |
|
1012 |
2180467679.1 |
1870585220 |
16.6 |
|
1013 |
18579615197.2 |
15834664872 |
17.3 |
|
1014 |
160206996577.1 |
135780321665 |
18.0 |
|
1015 |
1395577415965.0 |
1177209242304 |
18.5 |
|
1016 |
12265983873368.7 |
10304195697298 |
19.0 |
a. Source: Sloane's A007508;
Ribenboim 1996, p. 263; Nicely 1998, 1999; Sebah 2002; as referenced in Weisstein, Eric W.
"Twin Primes." From MathWorld--A
Wolfram Web Resource. http://mathworld.wolfram.com/TwinPrimes.html
Notice the estimation undercounts for 103 and 104 and over-counts for all of the other entries, with the percent over-count increasing for the estimates after 104. The estimates vary because the distribution of the double eliminations is not regular within the primorial intervals. In the case of the double eliminations by p3 and p4, for instance, both double eliminations come near the end of each p4# primorial interval. However, there is no reason to believe that the percent error will always continue to increase past 1016. The rate of increase in the error seems to be stabilizing towards the end of Table 11, and it may be the case that it will decrease again at some point.
That topic will be left for later research, for what is even more significant than using the individual ratios to estimate p2(x) is that we can take a summation of the (E′n / pn#) ratios. Let us call this value E2, for the twin prime elimination constant, where for n = 3 to ∞ its limit is:
E2 = S (E′n / pn#) (11)
Expansion of equation (11) gives the following:
E2 = (2 / 30) + (6 / 210) + (30 / 2310) + (270 / 30030) + (2970 / 510510) +
The first few of those partial sums were also shown in Table 10. The individual ratios are getting smaller, so it appears that the summation will converge to a definite value. But what value?
Thoerem 1: The value of the twin prime elimination constant E2 as given by equation (11) must converge to a value of 1 / 6 from below, such that it never exceeds 1 / 6.
Proof: The numerators of each term in E2 are the unique twin prime elimination counts over primorial intervals, such that all double eliminations of twin prime candidates are counted as one instance of a twin prime elimination. Since the ratio of the total number of candidates is 1 / 6, their total unique elimination ratio could never exceed 1 / 6, because that would imply that more candidates had been eliminated than actually existed.
But by the prime number theorem, the density of the prime numbers tends to thin out as they go off to infinity. This means that the proportion of composites to primes is always increasing. This increase can be seen back in Table 1, where the summation of the composite counts will continue to grow such that the quotient of the summation over the primorial will forever keep getting closer and closer to 1.0. Likewise the proportion of composites that eliminate twin primes will forever continue to increase and approach 1 / 6 as the primes go off to infinity.
Q.E.D.
The numerical evidence presented in Table 12 supports the assertion of Theorem 1.
Derbyshire [5] discusses how Euler showed that ζ(2) converges to p2 / 6, where ζ(x) is the Riemann Zeta Function and the p in the numerator is the familiar constant » 3.1415927 and does not represent a prime number counting function.
ζ(2) = 1 + (1/22) + (1/32) + (1/42) + (1/52) + (1/62) + (1/72) + (12)
Since ζ(2) = p2 / 6 it follows that ζ(2) / p2 = 1 / 6. Table 12 divides each term in the ζ(2) summation by p2 in order to compare them to the individual terms for E2 and it shows a comparison of their partial sums for a select group of terms.
The first term in ζ(2) / p2, for example, is p-2 » 0.101321183642. That is much larger than the first term in E2, which is (2 / 30). But by the second term the former is already smaller than the latter: (1 / 4p2) » 0.025330295911 which is less than (6 / 210) » 0.028571428571. Each of the successive terms shown for E2 is larger than its counterpart in ζ(2) / p2 as well.
However, the head start given by that larger first term has
ζ(2) / p2 converging
towards 1 / 6 much more rapidly than E2 is. As shown at the end of
Table 12, the partial sums for latter have not caught up to the former even as
far out as nearly 15 million terms. These values were again computed using the
first 15 million primes up to 275604541
as given by
Table 12: Comparison of the Convergence of E2 to ζ(2) / p2
|
n+2 |
p(n+2) |
E2 |
ζ(2) / p2 |
||
|
n-th Term |
n-th Partial Sum |
n-th Term |
n-th Partial Sum |
||
|
3 |
5 |
0.066666666667 |
0.066666666667 |
0.101321183642 |
0.101321183642 |
|
4 |
7 |
0.028571428571 |
0.095238095238 |
0.025330295911 |
0.126651479553 |
|
5 |
11 |
0.012987012987 |
0.108225108225 |
0.011257909294 |
0.137909388847 |
|
6 |
13 |
0.008991008991 |
0.117216117216 |
0.006332573978 |
0.144241962824 |
|
7 |
17 |
0.005817711700 |
0.123033828916 |
0.004052847346 |
0.148294810170 |
|
8 |
19 |
0.004592930290 |
0.127626759206 |
0.002814477323 |
0.151109287493 |
|
9 |
23 |
0.003394774562 |
0.131021533768 |
0.002067779258 |
0.153177066751 |
|
10 |
29 |
0.002458285028 |
0.133479818795 |
0.001583143494 |
0.154760210246 |
|
11 |
31 |
0.002141086959 |
0.135620905755 |
0.001250878810 |
0.156011089056 |
|
12 |
37 |
0.001678149238 |
0.137299054993 |
0.001013211836 |
0.157024300892 |
|
13 |
41 |
0.001432566423 |
0.138731621416 |
0.000837365154 |
0.157861666047 |
|
14 |
43 |
0.001299304430 |
0.140030925846 |
0.000703619331 |
0.158565285377 |
|
15 |
47 |
0.001133435780 |
0.141164361626 |
0.000599533631 |
0.159164819008 |
|
16 |
53 |
0.000962351134 |
0.142126712760 |
0.000516944815 |
0.159681763823 |
|
|
|||||
|
25 |
97 |
0.000403126748 |
0.147518146114 |
0.000191533428 |
0.162355777128 |
|
26 |
101 |
0.000379178625 |
0.147897324739 |
0.000175904833 |
0.162531681960 |
|
|
|||||
|
168 |
997 |
0.000017399629 |
0.158010351238 |
0.000003676919 |
0.166058132858 |
|
169 |
1009 |
0.000017158207 |
0.158027509445 |
0.000003633016 |
0.166061765874 |
|
|
|||||
|
1229 |
9973 |
0.000000981730 |
0.161772250511 |
0.000000067299 |
0.166584123955 |
|
1230 |
10007 |
0.000000978198 |
0.161773228710 |
0.000000067190 |
0.166584191145 |
|
|
|||||
|
9592 |
99991 |
0.000000062771 |
0.163528456887 |
0.000000001102 |
0.166656101922 |
|
9593 |
100003 |
0.000000062762 |
0.163528519650 |
0.000000001101 |
0.166656103023 |
|
|
|||||
|
78498 |
999983 |
0.000000004361 |
0.164486199406 |
0.000000000016 |
0.166665375893 |
|
78499 |
1000003 |
0.000000004361 |
0.164486203767 |
0.000000000016 |
0.166665375909 |
|
n+2 |
p(n+2) |
E2 |
ζ(2)
/ p2 |
||
|
n-th Term |
n-th Partial Sum |
n-th Term |
n-th Partial Sum |
||
|
664579 |
9999991 |
3.2041434E10 |
0.165064596726 |
2.2940857E13 |
0.166666514207 |
|
664580 |
10000019 |
3.2041338E10 |
0.165064597046 |
2.2940788E13 |
0.166666514207 |
|
|
|||||
|
5761455 |
99999989 |
2.4531971E11 |
0.165440068281 |
3.0523632E15 |
0.166666649081 |
|
5761456 |
100000007 |
2.4531966E11 |
0.165440068305 |
3.0523621E15 |
0.166666649081 |
|
|
|||||
|
14999999 |
275604533 |
7.9967276E12 |
0.165564699483 |
4.5031655E16 |
0.166666659912 |
|
15000000 |
275604541 |
7.9967273E12 |
0.165564699491 |
4.5031649E16 |
0.166666659912 |
5 Double the
Pleasure, Double the Fun: Locating Twin Primes
How can these primorial patterns help in locating twin primes? A good place to look for twin primes is in the interval between the first atomic boundary for that prime and the next composite that corresponds to its square, that is, between pn# + pn + 2 and pn# + pn2 2. The +/ 2 is specified so as to leave out the adjacent values, since pn would eliminate them. But pn is not capable of eliminating any other candidates in that interval.
The numerical results for this interval look promising, at least in the beginning. The twin prime (41, 43) falls between 37 and 53 while (227, 229) and (239, 241) are both between 219 and 257. The twin primes (2339, 2341) and (2381, 2383) are between 2323 and 2429, and there are two twins between 30045 and 30197 as well: (30089, 30091) and (30137, 30139). And if we check the interval from 510529 to 510797, we find a total of five twin primes: (510551, 510553), (510581, 510583), (510611, 510613), (510617, 510619), and (510707, 510709).
Because the pattern of the composites having prime factors lower than pn repeats, what happens is that all of the twin primes that exist in the interval from pn + 2 to pn2 2 get through and their corresponding candidates survive past pn in the interval from pn# + pn + 2 to pn# + pn2 2. If those corresponding candidates are to be eliminated at all, it must be by a prime factor greater than pn.
There are three twin primes (9699731, 9699733), (9699887, 9699889), and (9699917, 9699919) between 9699711 and 9700049. They correspond to 19# + (41, 43); 19# + (197, 199); and 19# + (227, 229). Unfortunately, however, there are no twin primes between 223092895 and 223093397. Of the 83 candidates total in that interval, 62 are eliminated by primes less than p9 = 23. The other 21 are eliminated by various primes greater than 23.
But just because there are no twin primes in a particular interval does not mean that the trend completely stops there. If we look in the interval from 6469693261 to 6469694069 above 29#, we find five twin primes again: (6469693331, 6469693333), (6469693511, 6469693513), (6469693661, 6469693663), (6469694039, 6469694041), and (6469694057, 6469694059).
Higher multiples of those high density base primorials do well, too. There are four twin primes between 2 29# + 29 + 2 = 12939386491 and 2 29# + 292 2 = 12939387299. Those four correspond to 2 29# plus the pairs (227, 229); (311, 313); (599, 601); and (641, 643).
So
while no theorem will be offered here as to where twin primes might always be
found, the evidence points to the intervals between N pn# + pn + 2 and N pn#
+ pn2 2 as being favorable.
Finally, consider that there is one twin prime between 37# + 37 + 2 and 37# + 372 2, that being (7420738134911, 7420738134913). It corresponds to 37# + (101, 103). The other corresponding candidates in this and similar such intervals are eliminated by primes larger than the prime of the base primorial, and in some cases much larger.
For example, in the twin prime candidate that corresponds to 37# + (599, 601), the first half of the pair, the value 7420738135409, is prime, while the other, 7420738135411, is a product of 2459383 3017317. That such eliminations exist further validates our infinite series. While the elimination ratio for the prime 2459383 is only about 1.563 109, that is still not zero, and because that ratio can never be zero, it leads us to another theorem.
Theorem 2: It would take infinitely many primes to eliminate all of the twin prime candidates greater than any particular value.
Proof: As given by equation (11), we have an infinite series over the primes p3 and greater that represents the rate of elimination of twin prime candidates. As given by equation (9), each term in that infinite series is generated by multiplying the previous term by (pn1 2) / pn. Since the previous prime pn1 is greater than 2 for all n > 3, the multiplier ((pn1 2) / pn) is always greater than zero, and thus each and every term in the infinite series in non-zero.
It can never be the case that all of the twin prime candidates past a particular value are entirely eliminated by only prime factors lower than a certain prime. If that was the case, then those eliminating composites would repeat in subsequent primorial intervals such that the elimination ratio for larger primes would have to be zero, and that would be a contradiction.
Not only would it take infinitely many primes to eliminate all twin prime candidates from a particular value onward, but it would take all of the primes 5 and greater to eliminate all of the candidates from any point out to infinity.
Q.E.D.
5 Open
Questions and Conclusions
What does theorem 2 imply about the infinitude of the twin primes themselves? Let us assume that the number of twin primes is finite. That would mean that there is a last or maximum twin prime, (pmax 2, pmax). If that is true, then there must be a prime number pe whose square, the first eliminating composite that it can generate, is greater than that last twin prime, i.e., pe2 > pmax. There cannot be any twin primes in the primorial interval from pe2 to pe2 + pe#.
There would be an integer number of primorial intervals for each of the primes up to pe within an interval of pe#. Therefore the elimination rate for the primes less than or equal to pe within that interval would precisely match the summation of E2 up to and including the prime pe. The elimination rate for the primes greater than pe within that interval would have to be (1/6) minus the elimination rate of the primes up to and including the prime pe.
But that would match the elimination rate for all of the primes greater than pe out to infinity. There are only a finite number of primes greater than pe, though, which could be the lowest prime factor of a composite within that interval. That is, only the primes up to a pi ≤ (pe2 + pe#)1/2 could account for eliminations of twin prime candidates between pe2 and pe2 + pe#. Now, the pattern for those composites would not repeat within intervals of pe#; it would repeat within intervals of the larger primorials within pi# > pe#. So while the elimination rate for those primes would tend toward the summation of their individual (E′n / pn#) ratios, the actual rate could increase in more narrow intervals. But then by the time a primorial interval from pi2 to pi2 + pi# was completed, the elimination rate for them within that pi# primorial interval would precisely match the summation of their individual ratios, and now the primes from pi up to (pi2 + pi#)1/2 would have to exceed their primorial ratios to eliminate the remaining candidates in this larger interval.
Is it possible for the twin prime elimination rate of a finite number of lowest prime factors to forever exceed their primorial elimination ratios within earlier and smaller primorials in this way? It also seems interesting to note that the number of unique twin prime eliminations for each prime is always an even number over its primorial, whereas the total number of twin prime candidates in any primorial of p3# or larger is always odd. Could a finite number of prime factors always produce an odd number of twin prime eliminations exactly as needed from a certain point forever onward out to infinity so as to make the number of twin primes finite?
While
it seems highly unlikely that could be the case, an actual proof has been
elusive. Shanks [6] stated that the evidence for the twin prime conjecture was
overwhelming. At the very least, the infinite series introduced here adds a
strong argument to that arsenal. But the primorial patterns presented here also
have the potential to be applied to other prime gaps, and to sieving in
general. Perhaps they may eventually lead to other useful insights regarding the
distribution of prime numbers.
7 Acknowledgments
I
am grateful to the following individuals for their assistance and advice: Dr.
Peter Gibson, The University of Alabama in
References:
1.
Dickson, Leonard E. History
of the Theory of Numbers Volume I: Divisibility and Primality.
2.
Weisstein, Eric W.
"
3.
Sloane, Neil J. A. Sequence
A005867 in The On-Line Encyclopedia of Integer Sequences. http://www.research.att.com/~njas/sequences/A005867
4.
Caldwell, Chris K. The Prime
Pages, http://primes.utm.edu/.
5.
Derbyshire, J. Prime
Obsession: Bernhard Riemann and the Greatest Unsolved Problem in Mathematics.
6.
Shanks, D. Solved
and Unsolved Problems in Number Theory, 4th ed.