7
$\begingroup$

Let $n > 1$ and $m > 0$ be two integers and $P_n$ be the $n^{th}$ prime.

Prove: $$P_{n+m} \ge P_n + P_m .$$

Can you give a hint, reference, comment, or proof?

$\endgroup$
7
  • 5
    $\begingroup$ This is equivalent to the well-known conjecture $\pi(x+y) \le \pi(x)+\pi(y)$, which is now widely believed to be false because it contradicts the prime $k$-tuples conjecture. $\endgroup$ Jun 27, 2018 at 2:41
  • $\begingroup$ $\pi(x)$ is not equivalent to $P_x$ $\endgroup$ Jun 27, 2018 at 2:42
  • 3
    $\begingroup$ I didn't claim that $\pi(x)=P_x$; I claimed that your inequality is equivalent to my inequality. $\endgroup$ Jun 27, 2018 at 2:47
  • $\begingroup$ I am sorry, how these are equivalent, may you show me? $\endgroup$ Jun 27, 2018 at 2:50
  • $\begingroup$ @GregMartin I am sorry, May You see answer below and my comment below? $\endgroup$ Jun 27, 2018 at 7:45

1 Answer 1

13
$\begingroup$

This is an expanded version of my previous answer. It shows, among other things, that the OP's conjecture contradicts the $k$-tuple conjecture for $k=459$.

1. Let $r\geq 0$ be a fixed integer. I claim that the following two statements are equivalent (for integral variables $x,y,n,m$): $$\forall x,y\geq 2:\pi(x+y)\leq\pi(x)+\pi(y)+r\tag{1}$$ $$\forall n,m\geq 2: P_n+P_m-1\leq P_{n+m+r-1}\tag{2}$$

$(1)\Rightarrow(2)$: Let $n,m\geq 2$ be arbitrary, and apply $(1)$ for $x:=P_n-1$ and $y:=P_m-1$, which are at least $2$. We get $\pi(P_n+P_m-2)\leq\pi(P_n-1)+\pi(P_m-1)+r=n+m+r-2$, whence $P_n+P_m-2<P_{n+m+r-1}$. That is, $P_n+P_m-1\leq P_{n+m+r-1}$, which is $(2)$.

$(2)\Rightarrow(1)$: Let $x,y\geq 2$ be arbitrary, and apply $(2)$ for $n:=\pi(x)+1$ and $m:=\pi(y)+1$, which are at least $2$. We get $x+y\leq P_n+P_m-2\leq P_{n+m+r-1}-1$, whence $\pi(x+y)\leq n+m+r-2$. That is, $\pi(x+y)\leq\pi(x)+\pi(y)+r$, which is $(1)$.

2. The OP's conjecture is weaker than $(1)$ and $(2)$ for $r=0$, but stronger than $(1)$ and $(2)$ for $r=1$. At any rate, the $k$-tuple conjecture is inconsistent with $(1)$ and $(2)$ for $r=1$, hence it is also inconsistent with the OP's conjecture. Indeed, there exists an admissible $459$-tuple in $\{0,1,\dots,3240\}$, see here. Hence, by the $k$-tuple conjecture, there exists a $y\geq 2$ such that $\pi(3241+y)-\pi(y)\geq 459$. However, $(1)$ for $r=1$ and $x=3241$ would yield that $\pi(3241+y)-\pi(y)\leq\pi(3241)+1=458$, contradicting the previous inequality.

3. In fact the $k$-tuple conjecture is inconsistent with $(1)$ and $(2)$ for any integer $r\geq 0$. More precisely, assuming the $k$-tuple conjecture, Hensley and Richards proved in 1973 that for any $x\geq 2$ we have $$\sup_{y\geq x}\bigl(\pi(x+y)-\pi(x)-\pi(y)\bigr)\geq\bigl(\log 2-o(1)\bigr)\frac{x}{(\log x)^2}.$$

$\endgroup$
4
  • 3
    $\begingroup$ Nice. I will stop typing up my answer. $\endgroup$
    – kodlu
    Jun 27, 2018 at 3:55
  • $\begingroup$ My comment is true? $P_{m+n}$ may $>>$ $P_{n+m-1}$ but $P_n+P_m \approx P_n+P_m-1$ ?? $\endgroup$ Jun 27, 2018 at 4:25
  • 1
    $\begingroup$ @Freeman: $(1)$ is believed to be false, because it contradicts the $k$-tuple conjecture. For a nice short explanation see en.wikipedia.org/wiki/… $\endgroup$
    – GH from MO
    Jun 27, 2018 at 15:12
  • $\begingroup$ I am sorry. I think your proof maybe false at (2)==> (1). I think (2) ==> $\pi(x+y) \le \pi(x)+\pi(y)+1$ ? $\endgroup$ Jun 27, 2018 at 16:21

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.