6
$\begingroup$
  • Let $a(n)$ be A106483 (i.e., primes $p$ such that $2p^2-1$ is also prime).

  • Let $b(n)$ be an integer sequence such that $b(n) = B$ after the whole transformation where we start with $A = n$, $B = 1$, $C$ and then for $i$ from $1$ to $n-1$ apply $C := \gcd(A, B)$, $A := A + 2n - C$, $B := B + C$.

I conjecture that odd number $k$ belongs to $a(n)$ iff $b(2k+1)=6k$.

Here is the PARI/GP program to check it numerically:

b(n) = my(A = n, B = 1, C); for(i=1, n-1, C = gcd(A, B); A += 2*n - C; B += C); B
my(x=3, z=3); for(k=1,299, while(!(isprime(x) && isprime(2*x^2-1)), x++); while(!(z%2 && b(2*z+1)==6*z), z++); print(x==z); x++; z++;);

Conjecture was verified with the given program up to $a(1000)=58901$ with no counterexamples.

Is there a way to prove it?

$\endgroup$
5
  • 5
    $\begingroup$ Since $\gcd(A,B)= \gcd(A+B,B )$ and each step $A+B$ increase by $2n$ so before the $i$'th step $A+B= (2i-1) n+1$, we can express b(n) more simply as starting with $B=1$ and then for $i$ from $1$ to $n-1$ apply $B:= B + \gcd( (2i-1)n+1, B)$. This kind of successively-adding-gcds dynamical system is very bizarre and it's not clear what its behavior should be. $\endgroup$
    – Will Sawin
    Commented May 3 at 14:31
  • 2
    $\begingroup$ Well, after an embarrassing mistake, let me offer a heuristic I noticed that may be helpful. It seems when the above iff condition holds, the last $A$ value in $b(n)$ is equal to $8k^2+2$, and conversely when $A \not= 8k^2+2$ then $k \not\in A106483$. $\endgroup$
    – Ben Burns
    Commented May 3 at 20:30
  • 2
    $\begingroup$ @BenBurns: As Will pointed out, the last value for $A+B$ is $(2k+1)+1+2(2k+1)2k=8k^2+6k+2$. Hence last $B=6k$ iff last $A=8k^2+2$. $\endgroup$ Commented May 3 at 20:40
  • 2
    $\begingroup$ @WillSawin "This kind of successively-adding-gcds dynamical system is very bizarre and it's not clear what its behavior should be." how can you authoritatively say such things?!? $\endgroup$ Commented May 4 at 0:11
  • 4
    $\begingroup$ @mathworker21 The phrase "not clear" often carries an implied "to me". (Of course one could worry that this is not a helpful thing to say, but I was confident enough that a high enough percentage of mathematicians would find this strange and hard to analyze to be worth the effort of mentioning to a question asker who is stated to be not a mathematician.) $\endgroup$
    – Will Sawin
    Commented May 4 at 0:17

1 Answer 1

6
$\begingroup$

Below I will prove that the proposed test is necessary, that is, if $k\in\text{A106483}$ then $b(2k+1)=6k$.

Following the simplification proposed by Will Sawin in the comments, the test for a given odd $k$ consists in setting $B_0 := 1$ and then iteratively computing for $i=1,2,\dots,2k$ $$B_i = B_{i-1} + \gcd((2i-1)(2k+1)+1,B_{i-1}),$$ and eventually testing whether $B_{2k} = 6k$.

Let $k$ be an odd prime such that $2k^2-1$ is also prime.

Trivially we have $B_1 = 2$, $B_2 = 4$, $B_3=8$. Since $(2i-1)(2k+1)+1$ is even, we observe that $B_i$ is even for all $i\geq 1$, implying that $2\mid \gcd((2i-1)(2k+1)+1,B_{i-1})$ for all $i\geq 2$.

Let $t>3$ be the smallest integer such that $\gcd((2t-1)(2k+1)+1,B_{t-1}) > 2$. Then $B_{t-1} = B_3 + 2(t-4) = 2t$ and thus $$2 < \gcd((2t-1)(2k+1)+1,B_{t-1}) = 2\gcd(k,t),$$ implying that $t=k$.

Now, we can derive by induction on $i$ that $$B_i = \begin{cases} 2(i+1), & \text{if } 3\leq i < k;\\ 2(i+k), & \text{if } k\leq i \leq 2k. \end{cases} $$ The base case $i\leq k$ is already proved above. For $i>k$, by induction we have \begin{split} B_i & = 2(i-1+k) + \gcd((2i-1)(2k+1)+1,2(i-1+k)) \\ &= 2(i-1+k) + 2\gcd(2k^2-1,i-1+k) \\ &= 2(i+k), \end{split} where we used the fact that $2k^2-1$ is prime.

It remains to note that $B_{2k}=6k$ as expected.

$\endgroup$
1
  • 3
    $\begingroup$ If $k$ is prime but $2k^2 - 1$ is composite, then $B_{2k} > 6k$. Your formula for $B_i$ holds if $i \le k$, and $B_i$ is still even. Thus $B_{2k} \ge 6k$. $B_{2k} = 6k$ if and only if $\gcd(2k^2 - 1, u) = 1$ for all $2k \le u \le 3k - 1$. Let $2k^2 - 1 = ab$. Then $a < 3k/2$ or $b < 3k/2$. If $a \le k$, then there is a multiple of $a$ in the range $[2k, 3k - 1]$. If $k < a < 3k/2$, then $2a$ is in this range. $\endgroup$ Commented May 4 at 7:13

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .