login

Year-end appeal: Please make a donation to the OEIS Foundation to support ongoing development and maintenance of the OEIS. We are now in our 61st year, we have over 378,000 sequences, and we’ve reached 11,000 citations (which often say “discovered thanks to the OEIS”).

A209407
Primes p(i) such that p(i+1)/p(i) > p(k+1)/p(k) for all k>i, where p(i) is the i-th prime.
1
3, 7, 13, 23, 31, 47, 113, 139, 199, 211, 293, 317, 523, 1327, 1669, 1951, 2179, 2477, 2971, 3271, 4297, 4831, 5591, 5749, 5953, 6491, 6917, 7253, 8467, 9551, 9973, 10799, 11743, 15683, 19609, 31397, 34061, 35617, 35677, 43331, 44293, 45893, 48679, 58831
OFFSET
1,1
COMMENTS
p(i) belongs to the sequence if p(i+1)/p(i) > p(k+1)/p(k) for all k>i.
It follows from the prime number theorem that p(i+1)/p(i) converges to 1 as i tends to infinity. a(n) is an infinite sequence therefore. The a(n) constitute "record holders" for the relative size of the prime number gaps.
The values a(n) given above were obtained by comparing p(i+1)/p(i) with p(k+1)/p(k) for 1<=i<=5949 and k ranging from i+1 to 200000 for given i.
In order to show that these values are correct one has to analyze the error terms in the formula p(k) ~ k*log(k) and extend the "test range" if needed. Using Dusart's bound: n*(log(n)+loglog(n)-1) < p(n) < n*(log(n)+loglog(n)) for n>=6 one gets
p(k+1)/p(k) < f(k):=(1+1/k)*(log(k+1)+loglog(k+1))/(log(k)+loglog(k)-1) for all k>=6. However this bound tends to 1 like 1+1/log(k) as k->oo. In order to verify, for example, that the term a(9)=199=p(46) is correct one must make sure that p(k+1)/p(k) < p(47)/p(46) = 211/199 =~ 1.0603 for all k>47. However f(10^7)~=1.06666 still, so k <= 10^7 is not sufficient to validate a(9). a(8) however is validated by checking the range k<=10^7.
In order to validate terms up to a(n)=31397 for example one even needs k<=10^20 roughly which needs considerable computational power.
This can be improved with another of Dusart's bounds: there is always a prime in (x, x + x/(25log^2 x)) for x > 396738. Hence it suffices to check up to the higher of exp(1/(25 (prime(i+1)/prime(i)-1))) and 396738. - Charles R Greathouse IV, Mar 06 2013
LINKS
Charles R Greathouse IV, Table of n, a(n) for n = 1..67
EXAMPLE
The smallest prime belonging to the sequence is p(2)=3 because p(3)/p(2) = 5/3 > 7/5, 11/7, 13/11, 17/13,... p(1)=2 does not belong to the sequence since p(2)/p(1) = 3/2 <5/3 = p(3)/p(2).
PROG
(PARI) {np=200000; a=vector(44); q=vector(np, k, prime(k+1)/prime(k)); m=n=0;
while(n<=44, if(q[m++]>vecmax(vector(np-m, j, q[m+j])), a[n++]=prime(m)))} \\ computes the first 44 terms of sequence.
(PARI) list(lim)=my(v=List([3]), u=List([2/3]), mn=.04/log(lim)^2, p=7, t); forprime(q=11, nextprime(lim+1), t=(q-p)/p; if(t>mn, if(t>u[#v], v[#v]=p; u[#u]=t, listput(v, p); listput(u, t))); p=q); t=u[#u]; forstep(i=#u-1, 6, -1, if(u[i]>t, t=u[i], v[i]=3)); Set(v) \\ valid for lim > 396738; Charles R Greathouse IV, Jun 25 2014
CROSSREFS
Sequence in context: A134197 A053001 A053607 * A124129 A101301 A103116
KEYWORD
nonn
AUTHOR
Thomas Nordhaus, Mar 08 2012
STATUS
approved