Containment Gives an Order on Extremal Numbers
Extremal graph questions ask how many edges a graph can have while avoiding a specified subgraph. When one forbidden graph sits inside another, avoiding the smaller graph is a stronger restriction: it also prevents the larger graph from appearing. This simple observation gives a useful comparison between extremal numbers, even when the two numbers themselves are not known exactly.
All graphs in this tutorial are finite and simple. A copy of a graph \(H\) in a graph \(G\) means a subgraph isomorphic to \(H\). Subgraphs here are not required to be induced: a copy must contain every edge of \(H\), but it may have additional edges between its vertices. A graph is H-free if it contains no copy of \(H\).
The maximum in this definition exists: on a fixed set of \(n\) vertices there are finitely many simple graphs, and the edgeless graph is \(H\)-free when \(H\) has at least one edge. If \(n\) is too small to contain \(H\), every \(n\)-vertex graph is \(H\)-free, so \(\operatorname{ex}(n,H)=\binom{n}{2}\). The same definition and observation apply to any other forbidden graph with at least one edge.
The Subgraph Comparison Theorem
Proof. Fix \(n\geq 0\). We show that every \(H\)-free graph on \(n\) vertices is also \(F\)-free. Suppose instead that such a graph \(G\) contains a copy of \(F\). Since \(H\subseteq F\), the vertices and edges of that copy of \(F\) include a subgraph isomorphic to \(H\). Therefore \(G\) contains a copy of \(H\), contradicting that \(G\) is \(H\)-free. This proves that the family of \(H\)-free \(n\)-vertex graphs is contained in the family of \(F\)-free \(n\)-vertex graphs. Every edge count in the first family is consequently at most the maximum edge count in the second family. Taking the maximum over the first family gives \(\operatorname{ex}(n,H)\leq\operatorname{ex}(n,F)\). \(\square\)
The implication concerns the forbidden graphs, not an assertion that an \(F\)-free graph must be \(H\)-free. The direction is easy to track by starting with a hypothetical copy of \(F\): it supplies a copy of each of its subgraphs, including \(H\). Thus an \(H\)-free graph cannot contain \(F\), while an \(F\)-free graph may still contain \(H\).
Proof. Apply the theorem to each containment \(H_i\subseteq H_{i+1}\), for \(i=1,\dots,k-1\). Each application gives \(\operatorname{ex}(n,H_i)\leq\operatorname{ex}(n,H_{i+1})\). Placing these inequalities in order gives the stated chain. If \(k=1\), there are no inequalities to apply and the conclusion is just the single term \(\operatorname{ex}(n,H_1)\). \(\square\)
Worked Comparisons
Worked Example: A Chain of Forbidden Cliques
For \(r\leq s\), the complete graph \(K_r\) is a subgraph of \(K_s\): choose any \(r\) vertices of \(K_s\), which have all \(\binom{r}{2}\) edges between them. Consequently, for every \(n\),
Take \(n=5\). A \(K_2\)-free graph has no edges, so \(\operatorname{ex}(5,K_2)=0\). Mantel’s theorem gives \(\operatorname{ex}(5,K_3)=\lfloor 5^2/4\rfloor=\lfloor25/4\rfloor=6\). This value is attained by \(K_{2,3}\), whose parts have sizes \(2\) and \(3\); every edge joins one vertex from each part, so its edge count is \(2\cdot3=6\), and it has no triangle.
The comparison therefore gives \(0\leq6\leq\operatorname{ex}(5,K_4)\). The same \(K_{2,3}\) graph has no \(K_4\), since a clique can use at most one vertex from each part. Hence \(\operatorname{ex}(5,K_4)\geq6\), in agreement with the comparison. This argument does not determine the exact value of \(\operatorname{ex}(5,K_4)\); the comparison transfers a bound, not automatically an exact answer.
Check: \(K_{2,3}\) has five vertices and six edges by the cross-part count \(2\cdot3=6\). Its largest clique has size \(2\), so it is both triangle-free and \(K_4\)-free, as required for the two lower-bound checks.
Worked Example: A Path Compared with a Triangle
Let \(P_3\) be the path on three vertices, with two edges. It is a subgraph of \(K_3\): select any three vertices of the triangle and retain just two of its edges. Therefore
For \(n=4\), Mantel’s theorem gives \(\operatorname{ex}(4,K_3)=\lfloor4^2/4\rfloor=4\). The graph with \(V=\{1,2,3,4\}\) and \(E=\{13,14,23,24\}\) is \(K_{2,2}\). It has \(2\cdot2=4\) edges and is triangle-free, so it attains the bound for \(K_3\). It contains a \(P_3\), for example the path \(1,3,2\), whose two edges are \(13\) and \(23\). Thus this four-edge example verifies the triangle-free count but is not a \(P_3\)-free example.
For a lower-bound check on the other extremal number, take \(V=\{1,2,3,4\}\) and \(E=\{12,34\}\). Each vertex has degree \(1\), so there is no two-edge path \(P_3\). This graph has \(2\) edges, giving \(\operatorname{ex}(4,P_3)\geq2\). The comparison supplies the upper bound \(\operatorname{ex}(4,P_3)\leq4\); no claim of equality follows from this comparison alone.
Check: The \(K_{2,2}\) edge count is \(2\cdot2=4\), and \(1,3,2\) uses exactly the edges \(13\) and \(32=23\). The graph with edges \(12\) and \(34\) has no adjacent pair of edges, so it cannot contain \(P_3\).
Worked Example: A 4-Cycle Compared with a Clique
The cycle \(C_4\) is a subgraph of \(K_4\): order four vertices around the cycle and retain its four cycle edges from the six edges of the complete graph. Hence
For \(n=5\), define \(G\) by \(V=\{1,2,3,4,5\}\), with parts \(A=\{1,2,3\}\) and \(B=\{4,5\}\), and let \(E\) consist of all edges joining \(A\) to \(B\). Explicitly, \(E=\{14,15,24,25,34,35\}\), so \(|E|=3\cdot2=6\). The graph has no \(K_4\), because any clique has at most one vertex in each of the two parts. Thus \(\operatorname{ex}(5,K_4)\geq6\).
However, \(G\) contains the 4-cycle \(1,4,2,5,1\), using the edges \(14,24,25,15\). It is therefore not a \(C_4\)-free witness. This illustrates the theorem’s direction: a \(C_4\)-free graph must be \(K_4\)-free, but a \(K_4\)-free graph need not be \(C_4\)-free.
Check: The six listed edges are precisely the \(3\cdot2=6\) cross-part pairs. The four edges in the displayed cycle are distinct members of that edge set, and the two parts prevent any four vertices from forming a clique.
How to Use the Comparison
The theorem is useful when one forbidden graph is easier to analyze than another. If \(H\subseteq F\) and a known result gives an upper bound for \(\operatorname{ex}(n,F)\), then the comparison gives the same upper bound for \(\operatorname{ex}(n,H)\). For example, \(P_3\subseteq K_3\), so Mantel’s theorem immediately gives \(\operatorname{ex}(n,P_3)\leq\lfloor n^2/4\rfloor\). This bound may not be sharp, but it follows without solving the \(P_3\)-free problem separately.
In the other direction, any \(H\)-free construction is also \(F\)-free whenever \(H\subseteq F\). Therefore a lower bound for \(\operatorname{ex}(n,H)\) also gives a lower bound for \(\operatorname{ex}(n,F)\). In the clique example, the triangle-free graph \(K_{2,3}\) is automatically \(K_4\)-free, so its six edges establish a lower bound for both extremal numbers.
Containment alone does not imply strict inequality, equality, or an exact value. The \(H\)-free class can be a proper subset of the \(F\)-free class while both classes have a maximum attained at the same edge count. To prove equality or strictness requires additional information about the extremal graphs or sharper bounds.
Common Pitfall: Reversing the Inequality
A common error is to reason that because \(H\) is smaller than \(F\), forbidding \(F\) is the stronger restriction. The theorem’s hypothesis is about which copies contain which: every copy of \(F\) contains a copy of \(H\), so forbidding \(H\) is the stronger restriction. The resulting inequality is \(\operatorname{ex}(n,H)\leq\operatorname{ex}(n,F)\), not the reverse.
For a concrete check, take \(H=K_3\), \(F=K_4\), and \(n=4\). Mantel’s theorem gives \(\operatorname{ex}(4,K_3)=\lfloor16/4\rfloor=4\). The graph on \(\{1,2,3,4\}\) with edges \(\{12,13,14,23,24\}\) has five edges and is \(K_4\)-free, since the edge \(34\) is missing. Thus \(\operatorname{ex}(4,K_4)\geq5>4=\operatorname{ex}(4,K_3)\), directly ruling out the reversed inequality.
Another pitfall is silently changing “subgraph” to “induced subgraph.” A triangle contains \(P_3\) as a subgraph because two of its edges form a path, even though the three vertices induce a triangle rather than an induced \(P_3\). Extremal numbers in this comparison use ordinary, non-induced forbidden subgraphs.
Check Your Understanding
For each question, identify the containment direction before comparing the extremal numbers.
- If \(C_5\subseteq K_5\), which of \(\operatorname{ex}(n,C_5)\) and \(\operatorname{ex}(n,K_5)\) is at most the other? Explain why.
- Explain why a \(P_4\)-free graph must be \(P_3\)-free, or give a counterexample.
- For \(P_3\subseteq K_3\), state an upper bound for \(\operatorname{ex}(n,P_3)\) obtained from Mantel’s theorem.
- Give a reason that a \(K_4\)-free graph need not be \(C_4\)-free.
- Does \(H\subseteq F\) guarantee \(\operatorname{ex}(n,H)<\operatorname{ex}(n,F)\)? Explain what the comparison theorem does guarantee.