We assume that $G\in G(n,p),p=\frac{\ln n +\ln \ln n +c(n)}{n}$. Then the following fact is well known:

\begin{eqnarray} Pr [G\mbox{ has a Hamiltonian cycle}]= \begin{cases} 1 & (c(n)\rightarrow \infty) \\ 0 & (c(n)\rightarrow - \infty) \\ e^{-e^{-c}} & (c(n)\rightarrow c) \end{cases} \end{eqnarray}

I want to know results about the number of Hamiltonian cycles on random graphs.

Q1. How many is the expected number of Hamiltonian cycles on $G(n,p)$?

Q2. What is the probability $Pr [G \textrm{ has a *unique* Hamiltonian cycle}]$ for the edge probability $p$ on $G(n,p)$?

  • 9
    $\begingroup$ You can probably answer Q1 yourself. Hint: Linearity of expectation. $\endgroup$ Oct 8, 2012 at 13:15

1 Answer 1


As Yuval said, Q1 is easy to answer using linearity of expectation (spoiler: $(n-1)!p^n$) . I don't know the exact answer to Q2, but it might be good enough if you know it's very low: for the range of $p$ where there is at least one cycle, it holds that $P[\text{there is more than one cycle} | \text{there is at least one cycle}]>1-1/n^{\log n}$ or so. In other words, once there is one cycle, there are many. The reason is that once there is one cycle, there are around $n^2$ ways to create another cycle from it by exchanging two edges of the cycle by the two "crossing" edges (this is called a "2-flip" or something in some of the relevant literature). For any pair of edges, the chance you can do that is $p^2$. So for all of these to fail, the chance is $(1-p^2)^{n^2}$ which is roughly $e^{-(pn)^2}$, which is pretty tiny.


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.