5
$\begingroup$

For a prime $q_k \neq 2$ we can study the corresponding set $\{q_1!+q_k,...,q_{k-1}!+q_k\}$, where $q_1,...q_{k-1}$ are all primes strictly less than the prime $q_k$.

Peter and Mathphile computed that for $q_k=1193$ we have that all the numbers from the corresponding set are composites

I think that $1193$ is rather large as an example of the first prime that generates all composites in the corresponding set and I do not see some reasons, armed with strong principles, of why there shouldn´t be some other primes that generate corresponding sets in which not a single number is prime.

Suppose that $\{r_1,...,r_m,...\}$ is the set of all the prime numbers for which the corresponding sets consist of only the composite numbers.

Are there any reasons and principles for justifying the assertion that $\{r_1,...,r_m,...\}$ is not a finite set?

Edit: This question arrived as the result of thinking about Mathphile´s conjecture.

$\endgroup$

1 Answer 1

5
$\begingroup$

Here's a heuristic for the likelihood that $q_k$ generates only composite numbers in this way.

For given $j<k$, the number $q_j!+q_k$ has size about $q_k \sim k\log k$ when $j<\log k$ or so, and size about $q_j!\approx e^{q_j\log q_j} \approx e^{j(\log j)^2}$ when $j$ is larger. The default probabilities that numbers of this size are prime are about $1/k\log k$ and $1/j(\log j)^2$, respectively.

We know a little more about these particular numbers. We know that $q_j!+q_k$ is not divisible by any of the first $j$ primes, which leads to heuristically modifying their probabilities of being prime upwards by a factor of $\prod_{i=1}^j (1-1/p_i)^{-1} \asymp \log j$. (We also know that $q_j!+q_k$ is the sum of two non-multiples of $p$ for every prime $p$ larger than $q_j$, except for $q_k$ itself, which leads to heuristically modifying their probabilities of being prime downwards by a factor of $\prod_{i=j+1}^{\infty} (1-1/p_j)^{-1} (1-1/(p_j-1)) \asymp 1$.) So the working heuristic probabilities that these numbers are prime change into about $(\log j)/k\log k$ for small $j$ and $1/j\log j$ for larger $j$.

The heuristic probability, then, that all these numbers are composite should be about \begin{multline*} \prod_{j=1}^{\log k} \bigg( 1-\frac{\log j}{k\log k} \bigg) \prod_{j=\log k}^{k} \bigg( 1-\frac1{j\log j} \bigg) \approx 1\bigg/\exp\bigg( \sum_{j=1}^{\log k} \frac{\log j}{k\log k} + \sum_{j=\log k}^{k} \frac1{j\log j} \bigg) \\ \approx 1\bigg/\exp\bigg( \frac{\log\log k}{k} + \log\log k - \log\log\log k \bigg) \approx (\log k)^{-C} \end{multline*} for some constant $C$ (because we were ignoring constants along the way).

Since $\sum_{k=1}^\infty (\log k)^{-C}$ diverges, this heuristic leads to the prediction that there should actually be plenty of primes $q_k$ for which all the numbers $q_1!+q_k,\dots,q_{k-1}!+q_k$ are composite.

$\endgroup$

Your Answer

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

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