You are on page 1of 6

Nonseparable Graphs

March 18, 2010

Cut Vertices
A cut vertex of graph G is a vertex v such that when the vertex v and the edges incident with v are removed, the number of connected components are increased, i.e., c(G v) > c(G). A connected graph is said to be 2-connected if it is a single vertex, or a single loop, or it has at least two vertices and any two vertices lie on a common cycle.

Theorem 1.1. Let G be a connected graph G with at least three vertices. Then G has no cut vertex if and only if any two distinct vertices are connected by two internally disjoint paths. Proof. We rst prove the suciency. Since any two vertices of G are connected by two internally disjoint paths, then for each vertex v of G, any two vertices of G v are connected by at least one path, i.e., c(G v) = c(G). Hence G has no cut vertex. Next we prove the necessity. Let u, v be two distinct vertices of G. To show that there are internally disjoint paths between u and v, we apply induction on the distance d(u, v) between u and v. When d(u, v) = 1, i.e., u, v are end-vertices of an edge e in G. Since both u and v are not cut vertices, the edge e is not a cut edge. So e is contained in a cycle C. Thus uev and C\e are two internally disjoint paths between u and v. Now assume that any two vertices having distance less than d are connected by two internally disjoint paths, where d 2. Let d(u, v) = d. Let P := v0 e1 v1 vd1 ed vd be a path from u = v0 to v = vd . Since d(v0 , vd1 ) = d1, there are two internally disjoint paths P1 and P2 from v0 to vd1 in G. Since G has no cut vertex, the subgraph G vd1 is connected. Then there is a uv-path P3 in G vd1 . Let w be the last vertex of P3 that meets P1 P2 . Without loss of generality, we may assume that w lies in P1 . Write P1 = P1 Q1 , P3 = P3 Q3 , where P1 is the sub-path of P from v0 to w, and Q3 is the sub-path of P3 from w to vd ; see Figure below.

P 1 v0 v1 v2 P2 P 3

Q 3 Q 1 vd1 ed vd

...

Then P := P1 Q3 and Q := P2 ed vd are two internally disjoint paths from u to v.

Separation and Blocks


A separation of a connected graph G is a decomposition of G into two connected subgraphs G1 , G2 that have exactly one vertex in common, and no one is contained in another. This common vertex is called a separating vertex of G. A cut vertex is a separating vertex, and a separating vertex is not necessarily a cut vertex. 1

A connected graph G is said to be separable if it has at least one separating vertex; otherwise it is said to be nonseparable. A loop is a nonseparable graph. If a graph G is nonseparable and is not a single loop, then G contains no loops. A nonseparable graph is either a single vertex, or a loop, or a link edge, or a 2-connected graph. A block of a graph is a maximal nonseparable subgraph. Theorem 2.1. Let G be a connected graph. Then G is nonseparable if and only if any two edges lie on a common cycle. Proof. : Suppose G is separable, i.e., G can be decomposed into two connected subgraphs G1 , G2 , no one is contained in another and have exactly one vertex v in common. Let ei be edges of Gi (i = 1, 2) incident with v. If one of e1 , e2 is a loop, it is clear that there is no cycle containing both e1 , e2 ; this is a contradiction. Thus ei are non-loops. Let vi be another end-vertex of ei (i = 1, 2) other than v respectively. It is clear that there is no v1 v2 -path in G v. Hence there is no cycle in G that contains both e1 , e2 . : If G is a loop, then nothing is to be proved. If G is not a loop, then G has no loops. We may assume that G has at least two edges. Let e be an edge with end-vertices v1 , v2 . Subdivide e into two edges by introducing a new vertex w on e to obtain a new graph G . We claim that G is also nonseparable. In fact, suppose G is separable. Then G must be separated at the vertex w into two connected subgraphs G1 , G2 . We may assume that vi belongs to Gi (i = 1, 2). Then w is a cut vertex of G ; subsequently, the edge e is a cut edge of G. Thus G\e has two connected components G1 , G2 with vi V (Gi ). Since G has at least two edges, then either G1 has an edge at v1 or G2 has an edge at v2 , say, G1 has an edge at v1 . Therefore G can be separated at v1 into G1 and G2 e v1 . This is a contradiction. Now, let e1 , e2 be two edges of G. Subdivide ei by introducing a new vertex vi on ei to obtain a new graph G (i = 1, 2). Then G is nonseparable. The resulted graph G is also nonseparable and has at least three vertices. Since Nonseparable graphs have no cut vertices, then by Theorem 1.1, there are two internally disjoint v1 v2 -paths. This means that there is a cycle containing both edges e1 , e2 . Theorem 2.2 (Block-Tree Decomposition). Let G be a connected graph. Then G can be decomposed into blocks such that (a) Any two blocks of G have at most one vertex in common. (b) Every cycle is contained in a block of G. (c) There is no block cycle, i.e., there is no blocks B0 , B1 , . . . , B such that V (Bi ) V (Bi+1 ) = (0 i ), where G +1 = G0 . Proof. (a) Suppose there are two distinct blocks B1 , B2 having k vertices v1 , v2 , . . . , vk in common and k 2. Since Bi are not single loops, then Bi have no loops. Consider the subgraph B := B1 B2 . We shall see that B cannot be separated at a vertex v other than vi . In fact, suppose B is separated at v into G1 , G2 and v V (B1 ). Then both G1 and G2 contain edges of B1 . Thus B1 is separated at v into B1 G1 and B2 G2 ; this is a contradiction. Note that B v1 = (B1 v1 ) (B2 v1 ). Since B1 v1 , B2 v1 are connected and have the vertex v2 in common, then B cannot be separated at v1 . Likewise, B cannot be separated at vi . So B is a block containing both B1 and B2 . This is contradict to the maximality of B1 , B2 . (b) and (c) are equivalent. We prove (c). Suppose there is a sequence B0 , B1 , . . . , B of blocks such that V (Bi ) V (Bi+1 ) = {vi }, 0 i , where B +1 = B0 . Consider the subgraph B := i=0 Bi . It is clear that B is connected and cannot be separated at any vertex other than vi , 0 i . Likewise, B cannot be separated at the vertices vi . So B is nonseparable. This is contradict to the maximality of Bi . Let B denote the set of blocks of a connected graph G, and S the set of separating vertices. Let B(G) denote the bipartite graph whose vertex set has the bipartition {S, B}, and whose edges are the pairs {v, B}, where v belongs to B. Then B(G) is a tree, called the block tree of G. The blocks of G corresponding to the leaves of the block tree B(G) are called end blocks. Any vertex of a block of G other than the separating vertex is called an internal vertex of the block. 2

Ear Decomposition
Every nonseparable graph other than a single vertex or a link edge contains a cycle. An ear of a subgraph H of a graph G is a path P in G such that P is not closed, the initial and terminal vertices of P lie in H, and internal vertices of P lie outside H.

Proposition 3.1. Let H be a nontrivial subgraph of a nonseparable graph G. Then H is neither a single vertex nor a loop, but has an ear in G. Proof. Since H is a nontrivial subgraph, then H is not a single vertex and cannot be G. If H is a loop e at a vertex v, then G must have an edge not in H; thus G is separated at v into H and G\e; this is a contradiction. If H is a spanning subgraph of G, then V (H) = V (G) and E(G) E(H) = . Since H is neither a single vertex nor a loop, then every edge e E(G) E(H) is a link edge, and subsequently, is an ear of H in G. If H is not a spanning subgraph, since G is connected, there is an edge e with an end-vertex u V (H) and an end-vertex v V (H); see Figure below.

P Hu H
Since G is nonseparable, then G\e is connected, subsequently, there is a (v, H u)-path P in G\e. Thus Q := ueP is an ear of H in G. Proposition 3.2. Let H be a nonseparable subgraph of a graph G, and let P be an ear of H in G. Then H P is nonseparable. Proof. It is clear that H P cannot be separated at any vertex of H. It is also clear that H P cannot be separated at any internal vertex of P . A sequence G0 , G1 , . . . , Gk of graphs is said to be nested if Gi Gi+1 , where 0 i k 1. An ear decomposition of a nonseparable graph G is a nested sequence G0 , G1 , . . . , Gk of nonseparable subgraphs of G such that (i) G0 is a cycle; (ii) Gi+1 = Gi Pi , where Pi is an ear of Gi in G, 0 i k 1; and (iii) Gk = G. Theorem 3.3. Let G be a nonseparable graph. If G is neither a single vertex nor a link edge, then G has an ear decomposition. Proof. Since G is neither a single vertex nor a link edge, then G contains at least two edges. Applying Theorem 2.1, any two edges of G lie on a common cycle. Thus G contains a cycle G0 with at least two vertices. If G0 = G, then G0 is a proper subgraph of G. Since G0 is neither a single vertex nor a link edge, then G0 has an ear P0 in G by Theorem 3.1. Applying Proposition 3.2, G1 := G0 P0 is nonseparable. Similarly, if G1 is a proper subgraph of G, then G1 has an ear in G. Continue this procedure; we obtain nonseparable subgraphs Gi and its ears Pi in G such that Gi+1 := Gi Pi . Since Gi Gi+1 and G is nite, the procedure must end up with Gk = G at some step k. Recall that a digraph D is said to be strongly connected (or just strong) if for any proper subset X the set (X, X c ), consisitng of edges whose orientations have tails in X and heads in X c , is nonempty. V (D),

e u v

Proposition 3.4. A digraph D is strong if and only if for any two vertices u, v of D, there is a directed path from u to v and a directed path from v to u, i.e., the vertices u, v are strongly connected.

Proof. The suciency is trivial. For necessity, let Vu be the set of all vertices w such that there exists a directed path from u to w in D. Then for any vertex w Vu c there is no directed path from u to w . Clearly, Vu = , since we allow directed path of length zero. If Vu = V (D), then (Vu , Vu c ) is nonempty, for D is strong. Let e (Vu , Vu c ) c be an edge with an end-vertex w1 Vu and an end-vertex w2 Vu , and let P be a directed path from u to w1 . Then Q := P ew2 is a directed path from u to w2 ; this is contradict to that there is no directed path from u to w2 . Hence Vu = V (D); analogously, Vv = V (D). This means that u, v are strongly connected. Proposition 3.5. A connected digraph is strong if and only if each of its block is strong. Proof. Trivial. Proposition 3.6. Let P be an ear of a subgraph H in a digraph D. If H is strongly connected and P is a directed path, then H P is also strongly connected. Proof. Let P = v0 e1 v1 el vl be directed from v0 to vl . For two vertices u, v H P , if u, v H, nothing is to be proved, since H is strongly connected. If u, v P , say, u = ui and v = vj with i < j, let P0 be a directed path from vl to v0 in H. Then P1 := vi ei+1 vi+1 ej vj is a directed path from u to v, and P2 := vj ej+1 vj+1 el vl P0 e1 v1 ei1 vi is a directed from v to u in H P . So H P is strongly connected. Theorem 3.7. Every connected graph G without cut edges has a strong orientation. Proof. It suces to show that each block of G has a strong orientation. Without loss of generality, we may assume that G is nonseparable and has no cut edges. If G is a single vertex or a loop, it is trivial that G can be oriented to be strongly connected. Note that G cannot be a link edge; applying Theorem 3.3, G has an ear decomposition (G0 , G1 , . . . , Gk ), where Pi is an ear of Gi in G and Gi+1 = Gi Pi , 0 i k 1. Now orient edges of G0 and Pi so that G0 becomes a directed cycle and Pi becomes a directed path. Initially, G0 is strongly connected. Applying Lemma 3.6, we see that all Gi+1 = Gi Pi are strongly connected. Hence G = Gk is strongly connected.

Ear Decomposition of Digraphs


Let D = (G, ) be a digraph and H a subdigraph of D. A directed ear of H in D is a directed path P in D whose distinct initial and terminal vertices lie in H and internal vertices lie outside H. A direction of a path P = v0 e1 v1 e v is an orientation P on P such that P (ei , vi )P (ei+1 , vi ) = 1, i.e., vi is neither a source nor a sink, 1 i 1. A direction P of P with P (e1 , v0 ) = 1 is usually called a positive direction of P . A direction of a walk W = v0 e1 v1 e v is a function W on the pairs (ei , vi1 ), (ei , vi ) (1 i l) such that W (ei , vi )W (ei+1 , vi ) = 1, 1 i l 1. A direction W of W with W (e1 , v0 ) = 1 is usually called a positive direction of W .

Proposition 4.1. Let H be a nontrivial strong subdigraph of a nonseparable strong digraph D. Then H has a directed ear in D. Proof. Since D is nonseparable and H is a nontrivial subgraph of D, then H has ears in D by Proposition 3.1. Among these ears we choose an ear P having minimal number of reversing edges. We claim that such an ear P is actually a directed ear of H in D. Let P = v0 e1 v1 ek vk . If k = 1, then either P or P 1 is a directed ear for H. Suppose P is not a directed path. We must have k 2. Let ei be an edge whose orientation is from vi to vi1 . Then one of the vertices vi1 and vi is outside H. Since D is strong, there is a directed path P from vi1 to vi in D. Then Q := v0 e1 v1 ei1 P ei+1 vi+1 ek vk is a walk from v0 to vk , having less number of reversing edges comparing with the walk P . If P H = , then Q is disjoint from H, except v0 , vk H. Let Q be a path followed the walk Q from v0 to vk , having no vertices repeating. Then Q is an ear of H in D, having less number of reversing edges comparing with P ; this is contradict to the choice of P . If P H = , let u, v be the rst and last vertices in Q such that u, v H respectively, let P1 be the subpath of P from vi1 to u, and let P2 be the subpath of P from v to vi . If u = v, then 4

Q := P2 ei P1 is a directed ear of H in D; see the left Figure below. If u = v, we have two cases: (i) u = v0 , then the walk Q1 := v0 e1 v1 vi2 ei1 P1 contains an ear of H in D, having less number of reversing edges comparing with P ; (ii) u = vk , then the walk Q2 := P2 ei+1 vi+1 ek vk contains a directed ear of H in D, having less number of reversing edges comapring with P ; all these are contradict to the choice of P ; see the right Figure below.

u H

P v0 vl vi 1 P ei vi

P v0 u v H vl P vi 1 P ei vi

A directed ear decomposition of a nonseparable strong digraph D is nested sequence (D0 , D1 , . . . , Dk ) of nonseparable strong subdigraphs of D such that (i) D0 is a directed cycle, (ii) Di+1 = Di Pi , where Pi is a directed ear of Di in D, 0 i k 1, and (iii) Dk = D. Theorem 4.2. Every nontrivial, nonseparable, strong digraph D has a directed ear decomposition. Proof. It is obviously true if D is a directed loop. If D is not a directed loop, then D contains at least two vertices u, v. There is a directed path P from u to v and a directed path Q from v to u. Then W := P Q1 is a closed directed walk containing both u, v. Of course, W contains a directed cycle D0 that contains both u, v. Clearly, D0 is strongly connected and non separable. By Proposition 4.1, D0 has a directed ear P0 in D. Then by Proposition 3.2 and Proposition 3.6, D1 := D0 P0 is strongly connected and nonseparable. Continue this procedure; we obtain a nested sequence (D0 , D1 , . . . , Dk ) of strongly connected nonseparable subdigraphs of D such that Di+1 = Di i , where Pi is a directed ear of Di in D, 0 i k 1, and Dk = D. This is a directed ear decomposition of D. A feedback set of a digraph D is an edge subset S of D such that D\S contains no directed cycles. A feedback set S of a digraph D is said to be minimal if for each edge e S the subdigraph D\S + e contains at least one directed cycle. Each such directed cycle intersects S at the only edge e, and is called a fundamental directed cycle of D with respect to S. A minimal feedback set S of a digraph D is said to be coherent if every edge of D is contained in some fundamental directed cycle of D with respect to S. If a digraph D admits a coherent feedback set, then every component of D must be strongly connected, for each edge of D is contained in a fundamental directed cycle. Theorem 4.3. Every strongly connected digraph D admits a coherent feedback set. Proof. If D is separable, then each of its block is strongly connected, and we may consider each of its blocks. So without loss of generality, we may assume that D is nonseparable. By Theorem 4.2, D has a directed ear decomposition (D0 , D1 , . . . , Dk ), where D0 is a directed cycle, Di+1 = Di Pi , Pi is a directed ear of Di in D, 0 i k 1, and Dk = D. Choose an edge e0 from D0 and set S0 := {e0 }. If D1 \e0 contains no directed cycles, set S1 := S0 . If D1 \e0 contains a directed cycle, then the directed cycle must contain the path P0 ; choose an edge e1 from P0 and set S1 := S0 e1 . Thus D\S1 contains no directed cycles. In general, if Di \Si1 contains no directed cycles, set Si := Si1 . If Di \Si1 contains a directed cycle, then the directed cycle must contain the whole path Pi1 ; choose an edge ei from Pi1 and set Si := Si1 ei . Then Di \Si contains no directed cycles. Finally, we have a coherent feedback set S = Sk for D. Proposition 4.4. Every strong digraph D has a strong spanning subgraph of at most 2|V (D)| 2 edges.

Proof. Delete all loops of D if necessary; so we may assume that D contains no loops. If D is a single vertex, it is clearly true. If D is not a single vertex, then each block B of D is strong. Consider a directed ear decomposition of B. Delete from B the edges in the directed ears of length one; we obtain a strong spanning subdigraph H of B, and a directed ear decomposition (D0 , D1 , . . . , Dk ) of H, where D0 is a directed cycle, Di+1 = Di Pi , Pi is a directed ear of length at least two, and Dk = H. Since each ear contains at least one internal vertex and |V (D)| 2, we see that k |V (H)| |V (D0 )| |V (H)| 2. Since D0 is a cycle and Pi are paths, then |E(D0 )| = |V (D0 )|, |E(Pi )| = |V (Pi )| 1, 0 i k 1. Thus
k1 k1

|E(H)| = |E(D0 )| +
i=0

|E(Pi )| = |V (D0 )| +
i=0

|V (Pi )| 1

= |V (H)| + k 2|V (H)| 2. Now the union of the strong subdigraphs H (one for each block B of D) is a strong spanning subdigraph of D. Since each block B has a strong spanning subdigraph H and |E(H)| 2|V (H)| 2, it follows that the union H H has the number of edges: E H =
H

|E(H)|
H

2|V (H)| 2 = 2
H

|V (H)| 1 = 2|V (D)| 2.

You might also like