Paper 4, Section II, F

Graph Theory | Part II, 2009

Let XX denote the number of triangles in a random graph GG chosen from G(n,p)G(n, p). Find the mean and variance of XX. Hence show that p=n1p=n^{-1} is a threshold for the existence of a triangle, in the sense that if pn0p n \rightarrow 0 then almost surely GG does not contain a triangle, while if pnp n \rightarrow \infty then almost surely GG does contain a triangle.

Now let p=n1/2p=n^{-1 / 2}, and let YY denote the number of edges of GG (chosen as before from G(n,p))G(n, p)). By considering the mean of YXY-X, show that for each n3n \geqslant 3 there exists a graph on nn vertices with at least 16n3/2\frac{1}{6} n^{3 / 2} edges that is triangle-free. Is this within a constant factor of the best-possible answer (meaning the greatest number of edges that a triangle-free graph on nn vertices can have)?

Typos? Please submit corrections to this page on GitHub.