Definition (Graph Copy)

We say a Graph has a copy of another graph if has a subgraph that is isomorphic to .

Definition (Turan Numbers)

Define

as the greatest number of edges that a graph on vertices that does not have a copy of can have.

Examples

Consider . Then we cannot have any copies like

graph LR;
	a((a)) o--o b((b));
	b o--o c((c));

which means we cannot have vertices of Degree more than . Otherwise, we would get a copy of the Complete Graph . This tell us our graph is a collection of disjoint s and maybe s. In particular, we only care about the edges, so at most we can have .

More generally, for , then must have maximum degree . By the Handshake Lemma, then the number of edges . Furthermore, if this is not an integer, then

Let be the following graph

graph TD;
	a((a)) o--o b((b));
	c((c)) o--o d((d))

Then

Extremal Triangles

Likewise . The way to look at it is to consider the maximal (in terms of edges) graph with no . Consider where and a non-adjacent vertex . Potentially, . We can replace with .

We claim this creates no triangles and it can only increase the number of edges. If this edit creates a new triangle, then we have:

graph LR;
	w((w)) o--o x((x));
	x o--o y((y));
	y o--o w;

Since , then . But then as , then by construction, . This implies that we have another triangle,

graph LR;
	v((v)) o--o x((x));
	x o--o y((y));
	y o--o v;

which is a contradiction on the assumption does not have any triangles. Then gained and lost edges. But as was maximal, then

creating more edges.

What if we continued this on every non-adjacent ? Then every non neighbor of connects to every neighbor of .

graph LR;
	v((v)) o--o Nv[[Nv]];
	v o--o Nv;
	v o--o Nv;
	v o--o Nv;
	Nv o--o a((1));
	Nv o--o a;
	Nv o--o a;
	Nv o--o b((2));
	Nv o--o b;
	Nv o--o c((...));

but this is a complete Bipartite Graph. But no matter graph we start with, we can always turn it to a bipartite graph. So if vertices are and , then

but to maximize this, and must be as close as possible. If even, then , and odd, then (or ).

Extremal Completes

What about ? Let . If we cannot have edges and it is . The complete bipartite graph example is quite good. We have edges over at most . One way to show we have no triangles is to show a left and right part. Indeed, for any three vertices, by Pidgeonhole, at least vertices must be in the same part, and thus those two cannot be connected.

To generalize this, consider a -partite graph, where the vertices are grouped into parts and there are edges between every pair of vertices in different parts but none in the same part.

Given this graph, we do not have a in it. For any vertices, by the Pidgeonhole Principle, at least two are in the same part, but these cannot have an edge. Thus we cannot have .

Each part should generally have as close to equal as possible^[1]. In particular, the part size you need is either

This leads to Turan’s Theorem.

Bipartite Dichotomy

If is not bipartite, then will be on the order of .

Proof:

By theorem, if is not bipartite, then there is an odd cycle. However, we can construct complete bipartite which has edges. Thus .


If is bipartite, then is no longer valid. This means we’ll need to drop some edges. In particular, instead of , we have edges (sub quadratic).