I'm trying to evaluate the probability that some even integer $2n$ has no Goldbach partitions using the following approach...

First, visualize the distribution of primes from $1$ to $n$ as a binary string representing odd integers, with $1$ for primes and $0$ for composites:

$$ 011101101101001... $$

Now in order for $2n$ to have no Goldbach partitions, the following binary sequence (from $n$ to $2n$), spelled backward, would have to be of the form:

$$ x000x00x00x0xx0... $$

This means that prime numbers from $n$ to $2n$ would have to be distributed within the $x$'s.

In order to get the probability that $2n$ has no Goldbach partitions, we can calculate the ratio of the number of possible distribution of primes from $n$ to $2n$ within the $x$'s, over the total number of possible distribution of primes from $n$ to $2n$.

Number of primes from $n$ to $2n$: $$ \pi\left(2n\right)-\pi\left(n\right) $$

Number of odd integers from $n$ to $2n$: $$ \frac{n+1}{2} $$

Number of odd composites from $1$ to $n$ (number of $x$'s): $$ \frac{n+3}{2}-\pi\left(n\right) $$

Number of possible distributions of primes from $n$ to $2n$: $$ \frac{\left(\frac{n+1}{2}\right)!}{\left(\pi\left(2n\right)-\pi\left(n\right)\right)!\left(\frac{n+1}{2}-\left(\pi\left(2n\right)-\pi\left(n\right)\right)\right)!} $$

Number of possible distribution of primes from $n$ to $2n$ within the $x$'s: $$ \frac{\left(\frac{n+3}{2}-\pi\left(n\right)\right)!}{\left(\pi\left(2n\right)-\pi\left(n\right)\right)!\left(\left(\frac{n+3}{2}-\pi\left(n\right)\right)-\left(\pi\left(2n\right)-\pi\left(n\right)\right)\right)!} $$

Probability $P$ that $2n$ has no Goldbach partitions:

$$ P\approx\frac{\left(\frac{n+3}{2}-\pi\left(n\right)\right)!\left(\frac{n+1}{2}-\pi\left(2n\right)+\pi\left(n\right)\right)!}{\left(\frac{n+1}{2}\right)!\left(\frac{n+3}{2}-\pi\left(2n\right)\right)!}$$


1: This approach does not take into account the fact that $3$ consecutive odd integers cannot be all primes. I need to show whether this fact decrease or increase the value of $P$.

2: A quick search about this subject shows some other formulas to approximate $P$, like for example:

$$ \prod_{a=2}^{n/2}\left(1-\frac 1{\log a\cdot \log (n-a)}\right) $$

but the values for $P$ obtained from those formulas are incredibly higher than what i get with mine. Could it be only because of the fact from question 1, or is there a problem with the whole approach ?

Some results for $P$ using my formula:

enter image description here

  • $\begingroup$ I am not sure whether the probability for $2n=8194$ is realistic. $\endgroup$ – Peter Jun 4 at 17:28
  • 2
    $\begingroup$ It's difficult to see how a probability can be negative. My personal handwaving might suggest $\left( 1- \frac{1}{(\log_e (2n))^2}\right)^n$ which is a lot higher than yours $\endgroup$ – Henry Jun 4 at 17:36
  • $\begingroup$ Sorry for the negatives, should have been 0, it only happens a couple of times below 100, because there is less non-primes in the first half than the number of primes in the second half $\endgroup$ – François Huppé Jun 4 at 17:41
  • $\begingroup$ Goldbach only cares about certain subsets of primes for any given n. $\endgroup$ – Roddy MacPhee Jun 5 at 14:11

Few things need accounting for:

$$n^2\bmod 3\to p\equiv q\bmod 6\\n^2\bmod 2\to p\equiv q \bmod 4\\p\equiv q\bmod 4\land p\equiv q\bmod 6\implies p\equiv q\bmod 12\\p\not\equiv q\bmod 4\land p\not\equiv q\bmod 6\implies p\not\equiv q\bmod 12\\ \gcd(n,p)=gcd(n,q)=1$$

and a few others for a start. There are also gaps in this new arithmetic progression, lining up with other gaps in another ( possibly different) arithmetic progression, to consider. These eliminate gaps, the relevant primes could possibly hide in.

It's literally pigeonhole principle, if the number of primes of a specific form above n, outnumber the gaps not covered by gaps in another ( possibly different) arithmetic progression of primes below n, then it's a shut case, it will cause a partition.

The exception to your first rule is of course 3,5,7.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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