Journal of Integer Sequences, Vol. 4 (2001), Article 01.1.7 |
Abstract: It would follow from a conjecture of Schinzel that all positive integers are representable as (p+1)/(q+1) with q and p prime. This conjecture is verified to 109, and various results of the calculation are given.
A consequence of an unproved conjecture of Schinzel[1] is that every positive integer n can be represented as n=(p+1)/(q+1), with p and q prime. For n a positive integer, define the function q(n) to be the smallest prime q such that n(q+1)-1 is prime. In other words, let q(n) be the smallest prime q so that n has a representation n=(p+1)/(q+1) with p prime. The sequence q(n) begins 2,2,3,2,3,2,5,2,5,2,3,3,7 (sequence A060324 in [2]; the values of p form sequence A062251). I have verified that q(n) exists for all n < 109.
Generally, q(n) is quite a small prime. For example, letting v(x,q) = #{ n <= x : q=q(n) }, we have, for q <= 31:
q | v(103,q) | v(104,q) | v(105,q) | v(106,q) | v(107,q) | v(108,q) |
2 | 222 | 1634 | 13026 | 108476 | 929119 | 8126474 |
3 | 223 | 1796 | 14962 | 128051 | 1117099 | 9903208 |
5 | 236 | 2085 | 18339 | 162796 | 1456211 | 13149129 |
7 | 93 | 971 | 9276 | 86491 | 800838 | 7418842 |
11 | 102 | 1095 | 11324 | 109516 | 1041573 | 9838207 |
13 | 35 | 524 | 6045 | 62243 | 617983 | 6044694 |
17 | 31 | 522 | 6204 | 66859 | 685210 | 6830034 |
19 | 13 | 261 | 3349 | 38962 | 420793 | 4369435 |
23 | 20 | 316 | 4097 | 46593 | 501096 | 5181342 |
29 | 12 | 261 | 3839 | 46723 | 520540 | 5518907 |
31 | 2 | 67 | 1039 | 14343 | 176355 | 1986081 |
Notice that, for a fixed x, v(x,q) to some extent reflects the number of prime factors of q+1. This makes sense, since the more prime factors q+1 has, the more likely it is that (q+1)n –1 is prime.
The following table gives the maximal values for q(n) (that is, values of n for which q(n') < q(n) for every n' < n).
n | q(n) | (log q(n))/(log n) | (log q(n))/(log log n) |
1 | 2 | - | - |
3 | 3 | 1. | 11.681421 |
7 | 5 | 0.827087 | 2.417554 |
13 | 7 | 0.758654 | 2.065856 |
31 | 13 | 0.746930 | 2.079033 |
51 | 19 | 0.748873 | 2.150632 |
101 | 23 | 0.679396 | 2.050230 |
146 | 41 | 0.745158 | 2.31209 |
311 | 71 | 0.742654 | 2.439409 |
1332 | 109 | 0.65208 | 2.377403 |
2213 | 179 | 0.673502 | 2.540976 |
6089 | 239 | 0.62845 | 2.529593 |
10382 | 269 | 0.604976 | 2.515168 |
11333 | 347 | 0.62657 | 2.618528 |
32003 | 353 | 0.56552 | 2.507828 |
83633 | 443 | 0.537627 | 2.509889 |
143822 | 503 | 0.52378 | 2.513829 |
176192 | 509 | 0.51596 | 2.501489 |
246314 | 617 | 0.517535 | 2.550711 |
386237 | 641 | 0.502404 | 2.530107 |
450644 | 701 | 0.503325 | 2.553224 |
1198748 | 773 | 0.475129 | 2.520164 |
2302457 | 881 | 0.462887 | 2.526093 |
5513867 | 971 | 0.443112 | 2.508225 |
9108629 | 977 | 0.429616 | 2.481671 |
11814707 | 1013 | 0.424976 | 2.480318 |
16881479 | 1019 | 0.416217 | 2.463297 |
18786623 | 1103 | 0.418290 | 2.485805 |
24911213 | 1109 | 0.411678 | 2.473069 |
28836722 | 1223 | 0.413867 | 2.500039 |
34257764 | 1559 | 0.423749 | 2.576361 |
196457309 | 1607 | 0.386581 | 2.502859 |
238192517 | 1709 | 0.385910 | 2.515164 |
482483669 | 1889 | 0.377295 | 2.518416 |
750301568 | 2063 | 0.373455 | 2.529388 |
This table is complete for n < 109 (the first two colums are sequences A060424 and A062252; the corresponding values of p give A062256). The fact that the maximal values of q(n) are so small (apparently less than log n to a fixed power) is supportive of the conjecture that it is always defined. Indeed, on average q(n) was found to be quite a bit smaller. Let Q(x) be the sum of q(n) for all n <= x. We have the following table:
x | Q(x) | Q(x)/(x log x log log x ) 102 | 427
| 0.607145 | 103 | 6680
| 0.500366 | 104 | 101494
| 0.496304 | 105 | 1354578
| 0.481517 | 106 | 17189068
| 0.473833 | 107 | 210240001
| 0.469208 | 108 |
2501065886 | 0.466024 | 109 |
29118770352 | 0.463545 | |
A heuristic argument can be given to explain the behavior seen in this table. We can think of q(n) as representing the k-th prime, where k is the number of primes pi (p1=2, p2=3, etc.) that need to be run through before n(pi+1)–1 is prime. Assuming the pi are small compared to n, the probability of n(pi+1)–1 being prime is about 1/log n. Hence we expect to need to run through about log n primes. Since the log of the n-th prime is roughly log n log log n, we can expect q(n) to be about log n log log n on average.
Finally, let s(x) be the number of n <= x for which q(n) = q(n-1). We have the following table:
x | s(x) | (log(s(x)/x))/(log log x) | (s(x) log x)/ x |
105 | 6881 | -1.095330 | 0.792204 |
106 | 60547 | -1.067996 | 0.836488 |
107 | 539273 | -1.050424 | 0.869205 |
108 | 4874595 | -1.036952 | 0.897934 |
The two right-hand columns both indicate the same thing: that s(x) appears to be approximately x/log x.
1. A. Schinzel and W. Sierpinski, Sur certaines hypothèses concernant les nombres premiers, Acta Arithmetica 4 (1958), 185-208
2. N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences. Published electronically at www.research.att.com/~njas/sequences/.
(Concerned with sequences A060324, A060424, A062251, A062252, A062256 .)
Received March 29, 2001; published in Journal of Integer Sequences July 5, 2001.
Return to Journal of Integer Sequences home page