MHB Proving Turan's Theorem (Dual Version) and its Implications for $ex(n, K_{p+1})$

  • Thread starter Thread starter joypav
  • Start date Start date
  • Tags Tags
    Dual Theorem
Click For Summary
The discussion focuses on proving the dual version of Turan's Theorem, which establishes that for a graph G with n vertices and edges equal to those of T'(n,q), the independence number α(G) is at least q, with equality implying G is isomorphic to T'(n,q). The thread explores how this dual version leads to the original Turan's Theorem, specifically showing that if the number of edges in G exceeds that of T(n,p), then the size of the largest clique ω(G) must be at least p+1. It emphasizes the relationship between the edge count of G and the independence number of its complement, leading to the conclusion that ω(G) equals α(¬G). The proof is presented as a logical progression from the dual theorem to the implications for Turan's Theorem.
joypav
Messages
149
Reaction score
0
I have already proved that for a graph $G$ with $n$ vertices and $|E(T'(n,q))|$ edges, $\alpha (G) \geq q$. Additionally, if $\alpha (G) = q$ then it must be that $G \cong T'(n,q)$.

Apparently this is the "dual version" of Turan's Theorem. How does this theorem imply Turan's?

That $ex(n, K_{p+1}) = |E(T(n,p))|$

Where:

- $T'(n,q)$ : $q$ disjoint cliques with size as equal as possible
- $\alpha (G)$ : independence number of $G$
- $ex(n, K_{p+1})$ : max number of edges in an n-vertex graph with no $K_{p+1}$ subgraph
 
Physics news on Phys.org
For completeness, here is the proof I wrote.
I'm not sure it is correct! May be some mistakes in the details.

Known Theorem:
Define $T'(n,q)$ to be q disjoint cliques with sizes of vertex sets as equal as possible. Let G be a graph with n vertices and $|E(T'(n,q))|$ edges. Then,
$$\alpha (G) \geq q$$
and if $\alpha (G) = q$ then $G \cong T'(n,q)$.To prove Turan's theorem, it suffices to show that if $|E(G)|>|E(T(n,p))|$ then $\omega (G) \geq p+1$, (where $\omega (G)$ is the size of the largest clique in G).\\
Assume $|E(G)|>|E(T(n,p))|$ then $|E(\overline{G})| \leq |E(T'(n,p))|$. By the in class proof, $\alpha (\overline{G}) \geq p+1$.
Notice that $\omega(G) = \alpha(\overline{G})$. Then we have, $\omega(G) = \alpha(\overline{G}) \geq p+1$, completing the proof.
 
Hello, I'm joining this forum to ask two questions which have nagged me for some time. They both are presumed obvious, yet don't make sense to me. Nobody will explain their positions, which is...uh...aka science. I also have a thread for the other question. But this one involves probability, known as the Monty Hall Problem. Please see any number of YouTube videos on this for an explanation, I'll leave it to them to explain it. I question the predicate of all those who answer this...