3
$\begingroup$

Related to this question.

Let $p$ be prime and $n$ positive integer.

Define $a(n)=(2^n \bmod p)^{p-1} \bmod p^2$

Let $D(n)$ be the base $2$ discrete logarithm of $a(n)$, i.e. given $p,a(n)$ we have $2^{D(n)} \bmod p^2=a(n)$. We can efficiently compute $D(n)=k(p-1)$ via p-adic logarithms and code is given in the linked question.

Let $D(n)$ be computed via p-adic algorithms, probably differing from the smallest logarithm by a small factor.

Strong numerical evidence suggests that three consecutive $D(i)$ satisfy:

$$ (D(n+2)-(D(n+3)+1)) (D(n+1)-(D(n+2)+1))(a_1 D(n+1) + a_2 D(n+2)+a_3 D(n+3)+a_4)\equiv 0 \pmod p \qquad (1)$$

for constants $a_i$. The constants $a_i$ depend on $p$ only and are determined by the first few values of $D(i)$ for which the other factors don't vanish. We don't know closed form for $a_i$.

The numerical evidence is 50 primes greater than $10^6$ and $10^3$ triples per each prime. Also several primes greater than $10^{50}$ were successfully tested.

Q1 Is the above identity true?

Remarks: Let $A=2^n \bmod p$. Then $a(n)=A^{p-1} \bmod p^2$ and $a(n+1)=(2A \bmod p)^{p-1} \bmod p^2$

Finding $n$ given $a(n)$ will break the discrete logarithm, which would be major result.

$\endgroup$
3
  • $\begingroup$ What is your "strong numerical evidence"; Also, what are the values of $a_i$ predicted by these numerical experiments? $\endgroup$
    – Milo Moses
    Apr 21, 2021 at 15:14
  • $\begingroup$ @MiloMoses I edited about the $a_i$. $\endgroup$
    – joro
    Apr 21, 2021 at 17:26
  • $\begingroup$ @MiloMoses I edited with numerical evidence. $\endgroup$
    – joro
    Apr 22, 2021 at 6:19

1 Answer 1

3
$\begingroup$

Using the notation from my answer to the previous question, we have $$D(n+1)−(D(n+2)+1)\not\equiv 0\pmod{p}$$ if and only if $g_0(n+2) = 2 g_0(n+1) - p$ and $g_1(n+2) \equiv 2 g_1(n+1) + 1\pmod{p}$, in which case $$D(n+2) \equiv -(n+2 + \frac{2 g_1(n+1) + 1}{2 g_0(n+1)c})\equiv D(n+1) - 1 - \frac{1}{2 g_0(n+1)c} \pmod{p}.$$ Similarly, $$D(n+2)−(D(n+3)+1)\not\equiv 0\pmod{p}$$ if and only if $$D(n+3) \equiv D(n+2) - 1 - \frac{1}{2 g_0(n+2)c}\equiv D(n+2) - 1 - \frac{1}{4 g_0(n+1)c}\pmod{p}.$$ If both these incongruences hold, then $$D(n+1) - 3D(n+2) + 2D(n+3) + 1 \equiv 0\pmod{p}$$ and so we can always take $(a_1,a_2,a_3,a_4)=(1,-3,2,1)$.

$\endgroup$
2
  • 1
    $\begingroup$ Thanks. I think this deserves paper or a note. You didn't show infinitely many non-Wieferich primes in base 2, right? $\endgroup$
    – joro
    May 9, 2021 at 17:49
  • 1
    $\begingroup$ Btw, this can be extended to $g>2$ -- we can construct a similar product of $2+(g-1)^2$ linear terms over $D(n),D(n+1),D(n+2)$ that is always $0$ modulo $p$. $\endgroup$ May 9, 2021 at 18:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.