Klein, Plotkin, Rao

Version & licenses
Creative Commons License

Cleaning the proof of Klein, Plotkin, Rao's theorem on excluded minors and network decomposition

Guyslain

The purpose of this note is to present the seminal paper by Klein, Plotkin and Rao [1].

Clustering of graphs with no $K_{r,r}$ minor

We prove the following theorem. The proof is not really different from the original proof, I just tried to remove all the details that were not necessary, and reformulate the useful ideas in a way that I find more intuitive. James Lee gave a short proof of this theorem at a workshop in Banff in December 2011 , but I can't remember the details.

Theorem 1: (Klein , Plotkin, Rao [1]) Given a graph $G$, two parameters $\delta$ and $r$, then we can either find a $K_{r,r}$ minor in $G$, or find a decomposition of $G$ into components $C_1,\ldots,C_k$ such that :

  • for any $i$ and any $u,v \in C_i$, $d_G(u,v) \leq 4 \delta r(r+1)$,
  • $\sum_{i \neq j} \left|E[C_i,C_j]\right| \leq \frac{r}{\delta} \left|E(G)\right|$

Proof: The main ingredient in the proof is the construction of the decomposition. This is done by using successive breadth-first-search (BFS) in smaller and smaller parts of the graph.

Let $G_0 = G$, and define iteratively $G_{i+1}$ from $G_i$, if $G_i$ has diameter greater than $4 \delta r(r+1)$, by the following construction. Choose an arbitrary vertex $r_i$ in $G_i$, and let $\mathcal{T}_i$ be a BFS tree from $r_i$. Then choose some value $l$ and take $V_{i+1} = \left\{v \in V(G_i) : d_{G_i}(r_i,v) \in [l, l+ \delta -1]\right\}$, and let $G_{i+1}$ be a component of $G[V_{i+1}]$. Hence $G_{i+1}$ is a component of the graph induced by $\delta$ consecutive layers in $\mathcal{T}_i$. The main lemma in the original proof is the following:

Lemma 2: (Klein , Plotkin, Rao [1]) If $G_{r+1}$ exists and there is $u,v \in V(G_{r+1})$ such that $d_G(u,v) \geq 4 \delta r (r+1)$, then there is a $K_{r,r}$-minor in $G$ (and we can find it algorithmically).

Proof: Assume wlog $r \geq 2$. We prove the following by iterating on $i$ from $r+1$ to $1$: there is a $K_{r+1-i,r}$-minor in $G_i$. In addition, denote $S^i_j, j\in \{i,\ldots,r\}$ and $T^i_j, j \in \{1,\ldots,r\}$ the supernodes of the minor. Then we also assume that there is $u^i_j \in T^i_j$ for all $i \in \{1,\ldots,r\}$ with the properties:

  • $d_G(u^i_j,u^i_k) \geq 4 i \delta$ for all $j \neq k$,
  • $u^i_j$ is at distance more than $\delta$ in $G_i$ from any other supernode.

The base case is when $i=r+1$, then by assumption there is $u,v$ and a $(u,v)$-path in $G_r$ of length at least $4 \delta r (r+1)$ (because $d_G(u,v) \geq 4 \delta r (r+1)$ and the distance are longer in $G_r$). Clearly one can choose $r$ vertices at distance $4 \delta (r+1)$ from each other on this path. This defines $u^{r+1}_1$ to $u^{r+1}_r$, and by partitioning the path we get $T^{r+1}_1$ to $T^{r+1}_r$ as wanted.

Now suppose that we have iteratively built the $K_{r+1-i,r}$-minor in $G_{i+1}$, for $i \in \{1,\ldots, r-1\}$. For each $j \in \{1,\ldots,r\}$, define $P^i_j$ as the unique directed path in $\mathcal{T}_i$ from the root to $u^{i+1}_j$. Such a path has length at least $(2 r (r+1) -1) \delta > 4 \delta$ (otherwise vertices of $G_r$ would be at distance at most $2 \delta r(r+1)$ from $r_i$ in $G$, contradicting the assumption). Define:

  • $u^i_j$ to be the vertex at distance $2 \delta$ from $u^{i+1}_j$ on $P^i_j$.
  • $T^i_j = T^{i+1}_j \cup Q^i_j$ where $Q^i_j$ is the subpath of $P^i_j$ of length $4 \delta$, starting from $u^{i+1}_j$,
  • $S^i_i = \bigcup_j P^i_j \setminus Q^i_j$,
  • $S^i_j = S^{i+1}_j$ for any $j \in \{i+1,\ldots, r\}$.

We check the hypothesis. Firstly, by the triangle inequality,

$$ \begin{align*} d_G(u^i_j,u^i_k) &\geq d_G(u^{i+1}_j,u^{i+1}_k) - d_G(u^i_j,u^{i+1}_j) - d_G(u^i_k,u^{i+1}_k) \\ &\geq 4 \delta (i+1)- 2 \delta - 2 \delta \\ &\geq 4 \delta i \end{align*} $$

Then by construction, $u^i_j$ is at distance more than $\delta$ from $G_{i+1}$ and from $S^i_i$ in $G_i$ (we used a path in a BFS tree), and at distance (in $G_i$) at least $4 \delta$ from any $u^i_k$, $k \neq j$. Hence $u^i_j$ is at distance more than $\delta$ in $G_i$ to any supernode except $T^i_j$.

It remains to check that it gives indeed a $K_{r+1-i,r}$ minor. Most of it is clear by construction, the hard part is to prove the disjointness of the supernodes. The only not-obvious part is for what happen in $G_{i+1}$ when we add the $P^i_j$ to the previous supernodes. But then, because $u^{i+1}_j$ is at distance more than $\delta$ in $G_{i+1}$ from any other supernode by hypothesis, $P^i_j$ is disjoint from any supernode from level $i+1$ except $T^{i+1}_j$.

Given a graph $G$, let $R$ be a set of vertices intersecting each component of $G$ exactly once. Define $E_0,\ldots, E_{\delta-1}$, with $E_i = \left\{(u,v) : 2 \delta r^2 \leq d(R,v) - 1 = d(R,u) \equiv i \pmod{\delta} \right\}$ (hence it is the union of edge sets between two levels of a BFS, every $l$ levels). These sets are disjoint, $\sum_{i=0}^{\delta-1} E_i \subset E(G)$ hence there is $\iota$ such that $\left|E_\iota\right| \leq \left|E(G)\right| / \delta$. Define an elementary clustering of $G$ as the graph $G' = G \setminus E_\iota$. Then each connected component $C$ of $G'$ has either diameter $4 \delta r(r+1)$ (in particular the component containing $r$), or has $\{d(r,v) : v \in V(C)\} \subseteq \left[l,l+\delta-1\right]$ for some $l$.

As a consequence, for any graph $G$, after $r$ successive operations of elementary clustering, by Lemma 2,

  • either we find a $K_{r,r}$ minor,
  • or for each remaining component $C$, for each $u,v \in V(C)$, $d_G(u,v) \leq 4 \delta r(r+1)$, and we removed at most $\frac{r}{\delta} |E(G)|$ edges.

Moreover this construction can be implemented using breadth-first-search in polynomial-time.

Flow-cut gap in minor-closed classes of graphs

Any graph $F$ is a minor of $K_{r,r}$ for some $r = r(F)$ big enough. Hence Theorem 1 is a decomposition theorem for any forbidden-minor class of graphs. We use this to prove that the flow-cut gap of any such class is constant. First we recall some definition.

A multicommodity flow instance is a pair of graphs $G,H$ on a common set of vertices $V(G)$, together with a capacity function $c: E(G) \to \mathbb{N}$ and a demand function $D: E(H) \to \mathbb{N}$. A solution is a family $(f_h)_{h \in E(H)}$ of flows indexed by the demand edges $E(H)$, such that:

  • $f_h$ is an $(s,t)$-flow, where $h = st \in E(H)$, and $\left|f_h\right| \leq D_h$ ($\left|f_H\right|$ denotes the value of the flow),
  • $\sum_{h \in E(H)} f_h \leq c$.

The throughput of a solution is defined by $\min_h \left|f_H\right|/D_h$. The Maximum Throughput problem is: given a multicommodity flow instance, find a solution maximizing its throughput.

An easy way to bound the maximum throughput, denoted by $\tau$, is to find a bottleneck in the graph, which is a cut such that the demand through it is larger than its capacity. We define:

$$ \Gamma = \min_{U \subset V : D(\delta_H(U)) > 0} \frac{c(\delta_G(U))}{D(\delta_H(U))}$$

and we have $\tau \leq \Gamma$. The flow-cut gap for $G,H,c,D$ is defined by $\frac{\Gamma}{\tau}$.

A harder but optimal way to bound the maximum throughput is given by the so-called Japanese Theorem, which is an immediate consequence of linear duality:

Theorem 3: $ \tau = \min_{l} \sum_{e \in E(G)} c_e \cdot l_e $ where the minimum is taken over all $l : E(G) \to \mathbb{Q}_+$ such that $\sum_{h = st \in E(H)} D_H \cdot d_l(s,t) = 1$.

We actually want to work with $l$ integral values, hence by scaling we rewrite it as

$$ \tau = \min_{0 \neq l \in \mathbb{N}^{E(G)}} \frac{\sum_{e \in E(G)} c_e \cdot l_e}{\sum_{h = st \in E(H)} D_H \cdot d_l(s,t)} $$

Theorem 4: Let $G,H,c,D$ be some multicommodity flow instance, such that $H$ does not contain a $K_{r,r}$-minor, and define $\Gamma$, $\tau$ and $l$ as above. Then the flow-cut gap is at most $2 r^2(r+1) \log D(E(H))$

Proof: Fix a parameter $\delta = \frac{r \cdot c(E(G))}{\Gamma \cdot D(E(H))}$ and apply Theorem 1 to $G,c,l$ (note how $c$ can be encoded by parallel edges, and $l$ by series edges), to get a decomposition of $G$ into components $C_1,\ldots,C_k$ with:

  • for any $u,v \in C_j$, $d_l(u,v) \leq 4 \delta r (r+1)$,
  • $\sum_{i \neq j} c(E[C_i,C_j]) \leq \frac{c(E(G)) \cdot r}{\delta}$.

Consider first the demands between distinct components among $C_1,\ldots,C_k$. By definition of $\Gamma$, for a component $C_i$ we have

$$ \sum_{st \in \delta_H(C_i)} D(st) \leq \frac{c(\delta(C_i))}{\Gamma}$$

Then by summing over all components:

$$ \sum_{i \neq j} \sum_{st \in E[C_i,C_j]} D(st) \leq \frac{1}{2 \Gamma} \sum_{i} c(\delta(C_i)) \leq \frac{r}{2 \Gamma \cdot \delta} c(E(G)) = \frac{1}{2} D(E(H)) $$

Here the second inequality comes from the properties of the decomposition. Hence at least half of the demand is contained in components of diameter $4 \delta r(r+1)$ in $G$. We deduce:

$$ \sum_i \sum_{st \in C_i} D(st) d_l(s,t) \leq 4 \delta r(r+1) \cdot \frac{1}{2} D(E(G)) = 2 r^2(r+1) \frac{C(E(G))}{\Gamma} $$

Now, by applying recursively this argument to the instance $G,H',c,D'$, where $H',D'$ are obtained from $H,D$ by removing all the demands internal to some component $C_i$ of the decomposition, we get that

$$ \sum_{st \in E(H)} D(st) d_l(s,t) \leq 2 \delta r^2(r+1) \frac{C(E(G))}{\Gamma} \log D(E(H)) $$

Using this in the flow-cut gap formula we get:

$$\frac{\Gamma}{\tau} = \Gamma \cdot \frac{\sum_{st \in E(H)} D(st) d_l(s,t)}{\sum_{e \in E(G)} c_e \cdot l_e} \leq 2 r^2 (r+1) \log D(E(H))$$

1.Philip Klein, Serge A. Plotkin, Satish Rao: Excluded minors, network decomposition, and multicommodity flow.
In: Proceedings of the twenty-fifth annual ACM Symposium on Theory of Computing. , 682 -- 690.
(1993).