The $G(n,p)$ random graph model creates graphs with $n$ vertices and each possible edge exists independently with probability $p\in (0,1)$.

Much is known about the (expected) size of a largest clique in these graphs, and it has been shown that the expected number of maximal cliques of size $d$ is

$${n\choose d} p^{d \choose 2} (1-p^d)^{n-d}.$$

To find the expected number of maximal cliques, just sum over all values of $d$ to get

$$\mu (G(n,p)) := \sum_{d=1}^n {n\choose d} p^{d \choose 2} (1-p^d)^{n-d}.$$

I'm interested in the asymptotic growth of $\mu (G(n,p))$ as a function $n$. For example, if $p$ is fixed, does $\mu$ grow polynomially in $n$?

A quasi-polynomial bound is not hard to show. Since the expected maximum clique size is roughly $\omega(G(n,p))\approx\frac{2 \log n}{\log(1/p)}$, the number of cliques (not necessarily maximal) is $O( {n \choose \omega(G(n,p))+1}) = O(n^{\omega(G(n,p))+1})=O(n^{\frac{2 \log n}{\log(1/p)}+1})$.

  • $\begingroup$ Do you really want $p$ fixed as $n \to \infty$? I always see these analyses done either in the limit that $np$ is fixed, that $p \propto n^{-\alpha}$ or that $p = c\log n/n$. $\endgroup$ – korrok May 28 '14 at 16:32
  • $\begingroup$ @korrok Sure, why not. For gits and shiggles. $\endgroup$ – Austin Buchanan May 28 '14 at 16:39

Since $\mu$ is the sum of $n$ terms, we can estimate it up to a factor of $n$ by the maximal term: $$ \max_{d=1}^n \binom{n}{d} p^{\binom{d}{2}} (1-p^d)^{n-d} \leq \sum_{d=1}^n \binom{n}{d} p^{\binom{d}{2}} (1-p^d)^{n-d} \leq n\max_{d=1}^n \binom{n}{d} p^{\binom{d}{2}} (1-p^d)^{n-d}. $$ In order to find out which term is the largest, we can use the estimate $$ \log_2 \binom{n}{d} p^{\binom{d}{2}} (1-p^d)^{n-d} \approx nh(d/n) + \binom{d}{2} \log_2 p - (n-d)p^d, $$ which is true up to polynomial factors for $d = \omega(1)$. We can estimate the magnitude of the largest term by differentiating this quantity with respect to $d$. If you follow this path, you should obtain an estimate for $\mu$ which is tight up to polynomial factors in $n$.

  • $\begingroup$ What is the term $h(d/n)$? $\endgroup$ – Austin Buchanan May 28 '14 at 20:24
  • $\begingroup$ @AustinBuchanan It's the binary entropy function. $\endgroup$ – Yuval Filmus May 28 '14 at 20:36

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.