6

Assume that $G$ is a graph which ${\lvert{E(G)}\rvert}>{\lvert{V(G)}\rvert}+3$. Prove $G$ has two disjoint cycles.


Answer:

I tried to prove it when $G$ is connected. consider the smallest cycle of length $k$. It exists because ${\lvert{E(G)}\rvert}>{\lvert{V(G)}\rvert}-1$. Now, I divided vertices into $2$ groups. Group $A$ consists of $k$ vertices in the smallest cycle and group $B$ consists of the other $n-k$ vertices.

Now, between $n-k$ vertices are exactly $n-k-1$ edges. Because, If it is more then we have another cycle (the proof is done) and if it is less then $G$ is disconnected (contradiction). So there are at least $5$ edges between group $A$ and $B$. Now, what should I do? should I write every possibility to prove it? Are there any easy ways to continue the proof?

What about if $G$ is disconnected? I think we should do it in each component. Am I right?

Thank you so much for your answers.

Sepehr Omidvar
  • 358
  • 3
  • 14

2 Answers2

3

Suppose the result is false; take a minimal counterexample $G$ with $n$ vertices and exactly $n+4$ edges (if there were more edges, we could delete some).

Then $G$ cannot contain a $3$-cycle or $4$-cycle. Otherwise, removing the edges of such a cycle, we have a graph with $n$ vertices and at least $n$ edges, which must still contain another cycle.

Also, $G$ has minimum degree $3$. If $G$ had a vertex $v$ of degree $1$, we could delete it and the edge out of it, getting a smaller counterexample (since the leaf was not part of any cycle). If $G$ had a vertex $v$ of degree $2$, with neighbors $w_1, w_2$, we could replace $v$ and its two edges with the single edge $w_1w_2$. This also gives a smaller counterexample, since any cycle containing $v$ would have had to use both edges, and now it can use the edge $w_1w_2$ instead. Also, because $G$ had no $3$-cycle, this still gives a simple graph.

Since $G$ has minimum degree $3$ and $n$ vertices, it must have at least $\frac 32n$ edges, but we know it has exactly $n+4$. Therefore $n+4 \ge \frac32n$, or $n \le 8$.

However, if we take a vertex $v$, we know it must have at least $3$ neighbors $w_1, w_2, w_3$. These have no edges between them (because $G$ has no $3$-cycles) and each has $2$ more neighbors (by minimum degree $3$) which are all distinct (because $G$ has no $4$-cycles), giving us at least $10$ vertices already. This is a contradiction, so the result must be true.

Misha Lavrov
  • 159,700
  • Ahh, this is very simple proof. Nicely done. – nonuser Jan 28 '21 at 20:17
  • Very nice! You have used that line of reasoning before, to solve another problem, right? I remember because it was very clever and counterintuitive and I filed that one away in memory. – Mike Jan 28 '21 at 20:21
  • @Aqua By the way, while it is your decision, I don't think it's necessary to delete good questions that have been closed as duplicates. They stick around as pointers to the other question and may help people find it, especially if you asked the question in a slightly different way. – Misha Lavrov Jan 28 '21 at 20:21
  • 1
    @Mike It's a common line of reasoning in some graph theory results. I once taught a week-long class on different proofs of Brooks' theorem from this article, and many of them have the minimal-counterexample structure. – Misha Lavrov Jan 28 '21 at 20:23
  • What problem is that @Mike – nonuser Jan 28 '21 at 20:27
  • I dont remember precisely @Aqua it was a while ago – Mike Jan 28 '21 at 20:30
  • Do you have any idea wich problem is Mike having in mind? @MishaLavrov – nonuser Jan 28 '21 at 20:33
  • @MishaLavrov Not particularly, sorry. It sounds plausible that I've solved problems on MSE with similar methods before. Could also be this one, which shares a different idea with this proof. – Misha Lavrov Jan 28 '21 at 20:37
  • @MishaLavrov Do you have an idea (again, shorther solution) for this one https://math.stackexchange.com/questions/3967104/3-regular-connected-graph-g-has-independent-subgraph-h-such-that-g-h-is-s – nonuser Jan 29 '21 at 09:16
2

The case by case analysis can be made to work. The key is to organize your cases carefully. It's not clear a priori how this is done, but it turns out that a good way is to look at how each connected component obtained by the removal of the cycle intersects the cycle.

Theorem: Any graph satisfying $|E| \ge |V|+4$ has two (edge) disjoint cycles.

Proof: It suffices to prove that every graph satisfying $|E| = |V|+4$ has two (edge) disjoint cycles. This clearly implies the original statement since we can always delete edges until we have $|E|=|V|+4$, and any cycles of the remaining subgraph will be cycles of the original graph as well.

Now suppose for the sake of contradiction that the statement is false. Let $G$ be a counter-example. Let $C$ be any cycle in $G$. Removing the edges of $C$ must leave a cycle-free graph since any remaining cycles would be edge-disjoint with $C$. Therefore $G\backslash C$ is a forest. Let $\{T_1,\cdots,T_c\}$ denote the components of the forest. The number of components $c$ of the forest satisfy $|E|-|C| = |V|-c$. This implies that $|C|=4+c$.

Now, consider the vertices common to each component $T_i$ and $C$, which we will denote as $V_i$. On the one-hand, we must have $$\sum_{i=1}^c|V_i| = |C|,$$ since each vertex of $C$ belongs in a unique component tree. On the other-hand, since $|C|=4+c$, it follows at least some of the $|V_i|$ must be larger than $1$, i.e., some of the trees must form chords of $C$.

Now we split into a few cases. In each case, we will construct two disjoint cycles, contrary to assumptions. For vertices $u,v\in V_i$, I will write $u\rightarrow v$ to indicate a path in $C$ from $u$ to $v$ and $u\Rightarrow v$ a path from $u$ to $v$ in $T_i$.

Note that the reasoning below is formal, but the various cases have easy and intuitive geometric interpretations. I suggest drawing out each of the cases if you find it difficult to follow.

  1. Suppose that there exists some $i$ such that $|V_i|\ge 4$. Let $v_1,v_2,v_3,v_4 \in V_i$ be four arbitrary intersection points, taken in sequence along $C$, i.e., $C = v_1\rightarrow v_2 \rightarrow v_3 \rightarrow v_4 \rightarrow v_1$. Then there exists a path $v_1\Rightarrow v_2$ and a path $v_3\Rightarrow v_4$. If these two paths are edge disjoint in $T_i$, then we are done, since our disjoint cycles are $v_1\rightarrow v_2 \Rightarrow v_1$ and $v_3\rightarrow v_4 \Rightarrow v_3$. Otherwise, there exists vertices $u_0,u_1$ where the paths $v_2\Rightarrow v_1$ and $v_3\Rightarrow v_4$ first and last meet. Then $v_1\Rightarrow u_1 \Rightarrow v_4 \rightarrow v_1$ and $v_2 \Rightarrow u_0 \Rightarrow v_3 \rightarrow v_1$ are disjoint cycles.

  2. Suppose that there exists some $i$ such that $|V_i|=3$. Let $v_1,v_2,v_3 \in C$ be the points of intersection. There must exist some other component $T_j$ such that $|V_j| \ge 2$. Let $u_1,u_2 \in V_i$. Then $C\backslash\{u_1,u_2\}$ splits into two disconnected components. Without loss of generality, suppose that $v_1$ lies in one component and $v_2,v_3$ in the other. Then $v_2\Rightarrow v_3 \rightarrow v_2$ and $u_1\Rightarrow u_2 \rightarrow v_1 \rightarrow u_1$ are disjoint cycles.

  3. Finally, suppose that all components satisfy $|V_i|\le 2$. There must exist at least four components such that $|V_i|=2$, say $|V_1|=|V_2|=|V_3|=|V_4|=2$. Let their respective intersection points be $\{v_1(i),v_2(i)\}$ for $i=1,2,3,4$. We will say that $V_i$ splits $V_j$ if $v_1(j)$ and $v_2(j)$ lies within distinct connected components of $C\backslash\{v_1(i),v_2(i)\}$. Note that this relation is symmetric. If there exists indices $i,j \in \{1,2,3,4\}$ such that $V_i$ and $V_j$ are not split, then we have disjoint cycles $v_1(i) \rightarrow v_2(i) \Rightarrow v_1(i)$ and $v_1(j) \rightarrow v_2(j) \Rightarrow v_1(j)$. So suppose that $V_i$ and $V_j$ are split for all $i,j$. Then without loss of generality, we can write the ordering of the vertices in $C$ as $$C=v_1(1)\rightarrow v_1(2) \rightarrow v_1(3) \rightarrow v_1(4) \rightarrow v_2(1) \rightarrow v_2(2) \rightarrow v_2(3) \rightarrow v_2(4) \rightarrow v_1(1).$$ In this case, the disjoint cycles are given by $v_1(1) \Rightarrow v_1(2) \Rightarrow v_2(2) \rightarrow v_2(1) \Rightarrow v_1(1)$ and $v_1(3)\rightarrow v_1(4) \Rightarrow v_2(4) \rightarrow v_2(3) \Rightarrow v_1(3)$.

EuYu
  • 42,696
  • Not an easy reading but this is some great work! +1 – Oldboy Oct 24 '20 at 11:08
  • Sorry, I couldn't follow your answer. Can you explain this $\sum_{i=1}^c|V_i| = |C|$ please? – Sepehr Omidvar Oct 25 '20 at 15:35
  • @SepehrOmidvar I explained it in the sentence below. Each vertex in $C$ belongs in a unique component of the forest. Therefore the $V_i$ must partition $C$. – EuYu Oct 25 '20 at 16:33