2
$\begingroup$

I want to prove the following statement:

Let $u$ be a vertex in a $2$-connected graph $G$. Then $G$ has two spanning trees such that for every vertex $v$, the $u,v$-paths in the trees are independent.

I tried to show this, but surprisingly, I have proved another statement.

A graph with $\vert V(G) \vert \geq 3$ is $2$-connected iff for any two vertices $u$ and $v$ in $G$, there exist at least two independent $u,v$-paths.

And I can assure that it is true, since I could find it from other papers.
I think this one may help me proving the desired statement, but I have no idea how to use it properly.
Would you help me find a such way, or suggest another proof of the first statement?

$\endgroup$
3
  • $\begingroup$ What does 'independent' mean? If you mean internally vertex-disjoint, then this is false (taking $G$ to be a cycle). $\endgroup$
    – Tony Huynh
    Commented Oct 8, 2021 at 0:43
  • $\begingroup$ Yes, it means internally vertex-disjoint, but it is true for cycle $G=\{u,v_1,\cdots,v_n\}$. If we choose $T_1=\{u,v_1,\cdots,v_{n-1}\}$ and $T_2=\{u,v_n,\cdots,v_2\}$, we can see that for any $v \in V(G)-\{u\}$ with fixed $u$, there are two internally vertex-disjoint paths one from $T_1$ and one from $T_2$. $\endgroup$
    – okw1124
    Commented Oct 8, 2021 at 0:46
  • $\begingroup$ @TonyHuynh Unfortunately, I found a counterexample of my proof. I think we have to think about this problem much more. // It's not a counterexample. I think my proof is valid, but not certain... $\endgroup$
    – okw1124
    Commented Oct 8, 2021 at 5:42

2 Answers 2

1
$\begingroup$

Yes, this is true. We will prove the following stronger lemma.

Lemma. Let $G$ be a $2$-connected graph and $u \in V(G)$. Then $G$ contains two spannings trees $T_1$ and $T_2$ such that for all $a,b \in V(G) \setminus \{u\}$ (possibly $a=b$), either

  • the $ua$-path in $T_1$ and the $ub$-path in $T_2$ are internally-disjoint, or
  • the $ub$-path in $T_1$ and the $ua$-path in $T_2$ are internally disjoint.

Proof. We proceed by induction on $|V(G)|+|E(G)|$. Since $G$ is $2$-connected, $G$ has an ear decomposition $(C, P_1, \dots, P_k)$, with $u \in V(C)$. If $G=C$, let $e_1$ and $e_2$ be the two edges incident to $u$. It is easy to check that we may take $T_1=C \setminus e_1$ and $T_2=C \setminus e_2$. Thus, we may assume $k \geq 1$. If the last ear $P_k$ is just an edge $e$, then $G \setminus e$ is $2$-connected, and we can apply induction. Thus, we may assume $|V(P_k)| \geq 3$. Suppose $P_k=x_1x_2 \dots x_\ell$. Observe that $G'=G-\{x_2, \dots, x_{\ell-1}\}$ is $2$-connected. By induction, $G'$ contains two spanning trees $T_1'$ and $T_2'$ such that for all $a,b \in V(G') \setminus \{u\}$, either

  • the $ua$-path in $T_1'$ and the $ub$-path in $T_2'$ are internally-disjoint, or
  • the $ub$-path in $T_1'$ and the $ua$-path in $T_2'$ are internally disjoint.

In particular, either

  • the $ux_1$-path in $T_1'$ and the $ux_\ell$-path in $T_2'$ are internally-disjoint, or
  • the $ux_\ell$-path in $T_1'$ and the $ux_1$-path in $T_2'$ are internally disjoint.

In the first case, $T_1:=T_1' \cup x_1x_2 \dots x_{\ell-1}$ and $T_2:=T_2' \cup x_2x_3 \dots x_\ell$ are the required spanning trees of $G$. In the second case, $T_1:=T_1' \cup x_2x_3 \dots x_\ell$ and $T_2:=T_2' \cup x_1x_2 \dots x_{\ell-1}$ are the required spanning trees of $G$.

$\endgroup$
5
  • $\begingroup$ But if each of $ux_1$ path intersects with both of $ux_2$ paths, can we say $ux_i$ paths are internally-disjoint? $\endgroup$
    – okw1124
    Commented Oct 8, 2021 at 1:17
  • $\begingroup$ In addition, with my brief guess, if so, one of $T_1$ and $T_2$ must have a cycle. $\endgroup$
    – okw1124
    Commented Oct 8, 2021 at 1:21
  • $\begingroup$ I think, maybe, I found a proper algorithm to draw such two trees. I will upload the sketch of my proof as soon as possible. $\endgroup$
    – okw1124
    Commented Oct 8, 2021 at 3:20
  • $\begingroup$ I think our proofs have exactly the same essence. In fact, 'oriented path' in my proof is the result of your induction. But your one is much shorter, and looks nice! Thanks a lot. $\endgroup$
    – okw1124
    Commented Oct 11, 2021 at 7:05
  • $\begingroup$ It seems as if the case that $a$ is an internal vertex of the last ear and $b \in V(G') \setminus \{u\}$ needs more details (which I didn't write down). So, the proofs are probably the same length. $\endgroup$
    – Tony Huynh
    Commented Oct 11, 2021 at 7:43
1
$\begingroup$

Let's use a new concept oriented path in a cycle $C$ or an open ear $P_k$.

In $C$: Label the vertices in $C$ in clockwise direction by $u,v_1,\cdots,v_n$. Then the clockwise-oriented path of $C$ ($c_1$ path) is $uv_1\cdots v_n$, and the counterclockwise-oriented path of $C$ ($c_2$ path) is $uv_n\cdots v_1$.

In $P_k$: Since $P_k$ is open, it has two distinct vertices $x_1$ and $x_2$ intersecting with $\begin{cases} C \cup \bigcup_{j=1}^{k-1}P_j & k \geq 2 \\ C & k=1\end{cases}$. WLOG assume that by following $P_k$ in clockwise direction from $x_1$, we can reach $x_2$. Then label the vertices in $P_k$ in clockwise direction by $x_1,y_1,\cdots,y_\ell,x_2$. Then $c_1$ path of $P_k$ is $x_1y_1\cdots y_\ell$, and $c_2$ path of $P_k$ is $x_2y_\ell\cdots y_1$.

Now, we will construct two graphs $T_1$ and $T_2$: $T_1$ is a union of all $c_1$ paths of $C,P_1,\cdots,P_k$, and $T_2$ is a union of all $c_2$ paths of $C,P_1,\cdots,P_k$. And let's denote this event $E$.

  1. $T_1$ and $T_2$ are connected. We deleted just one edge from a cycle or an ear to construct $c_1$ and $c_2$ path. It cannot disconnect $G$. Also, by the same reason, $T_1$ and $T_2$ are acyclic.

  2. $V(T_1)=V(T_2)=V(G)$.

Thus, $T_1$ and $T_2$ are spanning tree of $G$.
We will use induction on $\vert V(G) \vert + \vert E(G) \vert$.

Basis: The smallest $2$-connected graph is $G=K_3$ with $\vert V(G) \vert + \vert E(G) \vert=6$. By applying $E$, $T_1=uv_1v_2$ and $T_2=uv_2v_1$. Obviously, it satisfies the given statement.

Induction hypothesis: Any graph $H$ with $\vert V(H) \vert + \vert E(H) \vert \leq m-1$ has two spanning trees obtained by $E$ such that for any $v \in U - \{u\}$, two $uv$-paths are independent. Note that since tree has a unique path between any pair of vertices, so we can assure that there are exactly two independent $uv$-paths, one from $T_1$ and another from $T_2$.

Now, for $2$-connected $G$, there is an open ear decomposition $(C,P_1,\cdots,P_k)$ of $G$. Let's delete edges in $P_k$ from $G$, and then delete isolated vertices. Let $G'$ be the resulting graph. By hypothesis, $G'$ has two spanning trees $T_1'$ and $T_2'$ made by $E$, containing two independent $uv'$-paths for any $v' \in V(G')-\{u\}$.

Since $P_k$ is an open ear, it has at least two vertices $x_1$ and $x_2$ intersecting with $G'$. As we did above, WLOG assume that we can reach $x_2$ from $x_1$ by following $P_k$ in clockwise direction. If $P_k$ is an edge, $T_1'$ and $T_2'$ are the desired trees. So assume $V(P_k) \geq 3$. Label the vertices in $P_k$ by $y_1,\cdots,y_n$ in clockwise direction. Now, by hypothesis, there are $ux_1$-path in $T_1'$ and $ux_2$-path in $T_2'$. Let $T_1=T_1' \cup x_1y_1\cdots y_n$ and $T_2=T_2' \cup x_2y_n\cdots y_1$. Then for any vertex $y_t$ in $P_k$, there are two $uy_t$-paths one from $T_1$ and another from $T_2$. Let's denote them $p_1$ and $p_2$ respectively.

  1. Obviously, $p_1$ and $p_2$ does not intersect in $P_k$.
  2. If they intersect at $z \in V(G')$, then among $p_1 \cup p_2$, two cycles appear. Let's denote $C_1$ a cycle containing $u$, and $C_2$ a cycle containing $y_t$. Since $C_2$ should be an open ear, there is another ear (or a union of ears) $P'$ having a path $q=zz_1\cdots z_s$ in common with $C_2$. But then we must attach $P'$ to $C_1$ before attaching $C_2$ to $C_1$ (or it's a closed ear). Then $C_2-q$ is an ear having intersection points $z$ and $z_s$ with $P'$. But then according to the algorithm, we cannot start drawing an oriented path from $z$, but have to start from $z_s$. Contradiction.

As a conclusion, $p_1$ and $p_2$ are independent. Now we obtained the desired $T_1$ and $T_2$ in $G$.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .