21-801 Advanced Topics in Discrete Math: Graph Theory

Fall 2010
Prof. Andrzej Dudek
notes by Brendan Sullivan
October 18, 2010

Contents
0 Introduction
1 Matchings
1.1 Matchings in Bipartite Graphs . . . .
1.2 Matchings in General (Simple) Graphs
1.3 Tree Packing . . . . . . . . . . . . . .
1.4 Path Covering (for digraphs) . . . . .
1.5 Connectivity . . . . . . . . . . . . . .

Introduction

Matchings

1.1

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

1
1
2
5
6
6

Matchings in Bipartite Graphs

Definition 1.1. A matching is ************

Theorem 1.2 (Hall, 1930). ************
Proof. **********
Definition 1.3. A k-factor of a graph G is a k-regular spanning subgraph of G.
Note: a 1-factor is a complete matching and a 2-factor divides G into cycles.
Theorem 1.4. Any k-regular bipartite graph has a 1-factor.
Proof. Let S A. Then e(S, N (S)) = k|S| and e(N (S), A) = k|N (S)|. Certainly, e(N (S), A) e(S, N (S))
so |N (S)| |S|. Thus, Halls condition is satisfied and so a matching, i.e. a 1-factor.
Theorem 1.5 (Petersen, 1891). Every regular graph of positive, even degree has a 2-factor.
Proof. Let G be 2k-regular. Then we can find an Eulerian tour through G (i.e. a closed walk through
vertices, of the form v0 v1 v` = v0 , that visits every edge). Replace every v by (v , v + ) and add edge
e?i = vi vi+1
********

1.2

Matchings in General (Simple) Graphs

For a given graph G, let q(G) denote the number of odd components of G.
Theorem 1.6 (Tutte, 1947). G has a 1-factor q(G S) |S| S V (Tuttes Condition, or TC).
Proof. () If G has a 1-factor then
() Suppose TC holds and G has no 1-factor. Add edges to G to form G? such that G? has no 1-factor
but G? + e contains a 1-factor for any possible additional edge e. Then q(G? S) q(G S) |S| for all
S V . If S = then q(G? ) = 0 and |V | is even. Consider U = {v V : dG? (v) = n 1} where n = |V |.
Notice U 6= V otherwise G? is a complete graph on n = 2k vertices, so it would have a matching.
Claim: G? U is a disjoint union of complete graphs. Proof : Suppose not. Then x, y, z such that
xy, yz E(G? ) and xz
/ E(G? ). Since y
/ U , such that yw
/ E(G? ). Let M1 be the matching in G? + xz
?
and M2 be the matching in G + yw. Let H = M1 M2 . Notice H is a disjoint union of even cycles.
Case 1: xz and yw belong to different cycles. ****
Case 2: xz and yw belong to the same cycle. ******
Theorem 1.7 (Petersen). Every bridgeless cubic graph has a 1-factor.
Proof. Pick S, arbitrary. Let C be an odd component of G S. Then e(C, S) 3. Let e be the number of
edges from S to odd components. Then 3q(G S) e 3|S| by our assumptions. Thus, TC holds and so
G has a 1-factor.
*****
Goal for today is to state and prove a theorem stronger than Tuttes theorem in that it implies Tuttes
theorem and tells us some other stuff.
Definition 1.8. A graph G = (V, E) is factor-critical if G 6= and G v has a 1-factor v V .
Definition 1.9. Let Cg be the components of G. A vertex set S V is called matchhable to CGS if the
graph obtained by contracting components of G S to single vertices and deleting edges within S contains a
matching of S.
The following is Theorem 2.2.3 from Diestel.
Theorem 1.10. Every graph G = (V, E) contains a vertex set S V with the following two properties:
1. S is matchable to CGS
2. Every component of G S is factor-critical
Given such an S, the graph contains a 1-factor |S| = |CGS |.
Why does this imply Tuttes theorem? The first property of S implies |S| |CGS | and the second
condition implies |CGS | = q(GS). Tuttes condition then implies |CGS | = q(GS) |S|, so |S| = |CGS |.
Proof. The 1-factor part follows from properties (1) and (2):
() If 1-factor then q(G S) |S| q(G S) so |S| = |CGS |.
() If |S| = |CGS | then match S to one vertex of each component in CGS and then use factor-criticality
to find a matching in each component with one vertex removed (accounting for the matchability to S).
Now, to show existence of S, we use induction on |G|.
Base case: |G| = 0. Take S = .
Indcutive step: Let G be given, |G| > 0 and suppose the theorem holds for graphs with fewer vertices.
Conisder the sets T V where Tuttes condition fails the worst, i.e.
d(T ) := dG (T ) := q(G T ) |T |
is a maximum. So d(T ) d() 0. Let S be a largest such set.
2

Claim 1: Every component C CGS =: C is odd.

Proof of Claim 1: Suppose some C C is even. Pick a vertex c C and let T := S {c}. WWTS
d(T ) d(S) to obtain a contradiction. Notice C {c} has odd order and so it has at least one odd
component which is also a component of G T . Then
d(T ) = q(G T ) |T | q(G S) + 1 (|S| + 1) = q(G S) |S| = d(S)
and this contradicts our assumption that S was the largest set that maximumized d. This proves Claim 1.
Claim 2: Every C C is factor-critical.
Proof of Claim 2: Suppose C C and c C such that C 0 := C {c} has no 1-factor. By the inductive
hypothesis, S 0 V (C 0 ) such that q(C 0 S 0 ) > |S 0 | (using the fact that the current theorem implies Tuttes
Theorem). Notice |C 0 | is even, so if |S 0 | is even then q(C 0 S 0 ) is even (since |C 0 S 0 | is even, too); similarly,
if |S 0 | is odd then q(C 0 S 0 ) is odd. Thus, q(C 0 S 0 ) |S 0 | + 2, using the previously established inequality.
Furthermore, we have two equalities involving |T | := S {c} S 0 :
q(G T ) = q(G S) + q(C 0 S 0 )

and

|T | = |S| + 1 + |S 0 |

Then,
d(T ) = q(G T ) |T | = q(G S) 1 + q(C 0 S 0 ) |S| 1 |S 0 | q(G S) |S| = d(S)
This proves Claim 2.
Claim 3: S is matchable to CGS .
Proof of Claim 3: Suppose not. Then S 0 S such that |NC (S 0 )| < |S 0 | by Halls Theorem. Let T = (S 0 )c
and S = S 0 T . So,
d(T ) = q(G T ) |T | q(G S) |NC (S 0 )| |T | > q(G S) |S 0 | |T | = q(G S) |S|
This proves Claim 3 and completes the proof.
Let M be any matching and kM := number of edges in M with at least 1 end in S, and let kG := number
of edges in M with both ends in G S. Notice M satisfies kS |S| and kG 21 (|V | |S| |C|). Any
maximum matching satisfies these at equality.
Theorem 1.11 (Gallai-Edmonds Structure Theorem). Let G = (V, E) be any graph. Let D be the set of
vertices which are not covered by at least one maximal matching. Let A be the vertices in V D which are
adjacent to at least 1 vertex in D. Let C = V D A. Then
1. The components of G[D] are factor critical.
2. G[C] has a perfect matching
3. The bipartite graph on A CG[D] has positive surplus viewed from A; that is, N (S) > |S| for every
S A (S 6= ).
4. Any maximal matching has
a near perfect matching of components of G[D]
perfect matchings on components of G[C]
matches each vertex with distinct components of G[D]
5. |M | =

1
2

(|V | c(G[D]) + |A|), where c() is the number of components.

Proof. *******************

Definition 1.12. H has the Erd

os-Posa property if there f : N R, k 7 f (k), such that k either G
contains k disjoint subgraphs, each isomorphic to a graph in H, or there is a set U V (G) with |U | f (k)
such that G U has no subgraph in H.
Goal: prove class of all cycles has E-P property (with f (k) 4k log k). For the rest of today, consider
(
4krk if k 2
rk := log k + log log k + 4
sk :=
1
if k 1
Lemma 1.13. Let k N and let H be a cubic (3-regular) multigraph (loops and multiple edges allowed). If
|H| sk then H contains k disjoint cycles.
Proof. Induction on k. Base case: k 1 trivial. Inductive step: Let k 2 be given and let C be a
shortest cycle in H. CLaim: H C contains a subdivision of a cubic multigraph H 0 with |H 0 | |H| 2|C|.
*** subdivision picture ***
Proof of claim: Let m be the number of edges between C and H C. Since H is 3-regular and the
average degree of C is 2, m |C|. Now, consider the following sequence of bipartitions of V , {V1 , V2 }. Start
with V1 = V (C). If H[V2 ] has a vertex of degree 1, move it to V1 . Then the number of crossing edges
decreases by 1 each time. Suppose you can do this n times, but no more. Then {V1 , V2 } is crossed by
m n edges. Hence H[V2 ] has at most m n vertices of degree < 3 and these vertices have degree = 2
(otherwise we moved it over to V1 ). Now suppress the vertices of degree 2 in H[V2 ] (i.e. delete such a
vertex v and add an edge between its neighbors). This yields a cubic graph multigraph H 0 . Notice
|H 0 | |H|

|C|
|{z}

original cycle

n
|{z}

move-over provess

(m n)
| {z }

|H| 2|C|

suppress degree 2s

This proves the claim.

Now, we just have to show |H 0 | sk1 . Corollary 1.3.5 (in Diestel) says if (G) 3 then g(G) 2 log |G|
(where g() is the girth, i.e. length of shortest cycle). So |C| 2 log |H|. Since |H| sk 6 and x 4 log x
is increasing for x 6, we get
|H 0 | |H| 2|C| |H| 4 log |H| sk 4 log sk
To complete the proof WWTS sk 4 log sk sk1 . For k = 2 we have 4 2 (1 + 0 + 4) 4 log 40 1 = s1 .
Also, notice rk 4 log k for k 3 (for k = 3 use a calculator, and for k 4 its obvious). So for k 3
sk 4 log sk = 4krk 4 log(4krk ) = 4(k 1)rk + 4 log k + 4 log log k + 16 (8 + 4 log k + 4 log rk )
sk1 + 4 log log k + 8 4 log(4 log k) = sk1

Theorem 1.14 (Erd

os, Posa, 1965). The class of all cycles has E-P property.
Proof. Let f (k) := sk + k 1. Let k be given and G be any graph (and assume G has a cycle, otherwise its
trivial). So it has a maximal (with respect to the subgraph relation) subgraph H where all degrees in H are
either 2 or 3. Let U be its set of degree 3 vertices and let C be the set of cycles in G that avoid U and meet
H in exactly 1 vertex. Let Z V (H) U be the set of vertices in a member of C. For each z Z, pick one
cycle Cz C and let C 0 = {Cz : z Z}. The cycles in C 0 are disjoint by maximality of H (otherwise take
part of cycles until first meeting point and add to H). Let D be the set of 2-regular components of H that
avoid Z. Then C 0 D is a set of disjoint cycles. So if |C 0 D| k then were done. Otherwise, take one
vertex from each D-cycle and add it to Z to get a set X of size k 1 which meet all cycles in C and all
2-regular components of H. Consider any cycle of G which avoids X. It has to meet H by maximality. It
has to meet U because: it cant be all in H (otherwise in D), it cant meet H in just one vertex (otherwise
in C), and it cant connect 2 vertices of H U with a path outside of H, so it must hit U . So every cycle
in G meets X U . We know |X| k 1. If |U | < sk then we have < f (k) vertices meeting each cycle. If
|U | sk , suppress all degree 2 vertices in H to get a 3-regular multigraph H 0 with |H 0 | = |U | sk . Apply
the lemma.
4

1.3

Tree Packing

Let G be a given graph.

Theorem 1.15 (Mengers Theorem). If G is k-edge connected, then k disjoint paths between any 2 vertices
in G.
Question: How many edge-disjoint spanning trees exist in G?
Necessary condition: k-edge connectivity.
Is this condition sufficient? No. Consider k = 2 and take a cycle with 4 vertices.
Another necessary condition: for all partitions of V (G) into r sets, each spanning tree has r 1 cross-edges
(edges with ends in different partitions).
Theorem 1.16 (Nash-Williams 1961, Tutte 1961). A multigraph G has k edge-disjoint spanning trees
G has k(r 1) cross edges for any partition of size r.
Corollary 1.17. Every 2k-edge connected multigraph G has k edge-disjoint spanning trees.
Pr
Proof of Corollary. ***** picture G has 21 i=1 2k = kr k(r 1). Open: is this bound sharp?
Set up for proof of theorem: Let G be a given multigraph and k N. Let F be the set of all
k-tuples F = (F1 , F2 , . . . , Fk ) where the Fi s are edge-disjoint spanning forests such that kF k := |E[F ]| =
|E[F1 ] E[Fk ]| is as large as possible.
If F F and e E \ E[F ] then Fi + e for i = 1, 2, . . . , k constains a cycle. For some fixed i, take e0 in
this cycle (e0 6= e). Then setting Fi0 := Fi + e e0 and Fj0 := Fj for j 6= i yields a new F 0 = (F10 , . . . , Fk0 )
such that F 0 F. We say F 0 is obtained from F by the replacement of e0 with e. Note: for every path
x . . . y Fi0 !xFi y.
Consider the fixed k-tuple F 0 = (F10 , . . . , Fk0 ) F. [
Let F 0 be the set of all k-tuples that can be obtained
from F 0 by a series of edge replacements. Let E 0 :=
(E \ E[F ]) and G0 := (V, E 0 ).
F F 0

Lemma 1.18. For any e0 E \ E[F 0 ] there exists U V (G) that is connected in every Fi0 and contains
the ends of e0 .
Proof. *** we believe the lemma
Proof of Theorem. () Induction on |G|. Base case: |G| = 2. Done. Induction step: Suppose for each
partition P of V (G), k(|P | 1) cross edges. We will construct k edge-disjoint spanning trees.
Fix a k-tuple F 0 = (F10 , . . . , Fk0 ) F. If each Fi0 is a tree, done; otherwise,
kF 0 k =

k
X

kFi0 k < k(|G| 1)

i=1

(Recall: k k denotes # of edges.) We have kGk k(|G| 1) by assumption, when we consider P to be single
vertices. Thus, e0 E \ E[F 0 ]. By the Lemma , U V (G) that is connected in each Fi0 and contains
ends of e0 . In particular, |U | 2.
Since every partition of the contracted multigraph G \ U induces a partition of G with the same # of
cross edges, G \ U has k(|P | 1) cross edges, with respect to any partition P . By induction, G \ U has k
disjoint spanning trees T1 , . . . , Tk . In each Ti , replace VU by the spanning tree Fi0 G[U ].
Apparently the other direction is obvious.
We say subgraphs G1 , . . . , Gk partition G if their edge sets form a partition of E(G).
Question: Into how many connected spanning subgraphs can we partition a given G?
If we can answer that question, then we can answer the question: Into how few acyclic spanning subgraphs
can we partition G? Or, for a given k, which graphs can be partitioned into k forests?
Necessary: U V (G) induces k(|U | 1) edges.
5

Theorem 1.19 (Nash-Williams 1961). A multigraph G can be partitioned into at most k forests
kG[U ]k k(|U | 1) for all u V (G).
Proof. We will show: every k-tuple F = (F1 , . . . , Fk ) F partitions G. Suppose otherwise; then e
E \ E[F ]. Use the Lemma By the Lemma, U V connected in every Fi and containing ends of e.
Therefore, G[U ] has |U | 1 edges in each Fi in addition to e, so kG[U ]k > k(|U | 1), a contradiction.

1.4

Path Covering (for digraphs)

Definition 1.20. A path partition in a digraph D is a family of vertex disjoint directed paths that cover all
of the vertices of D. We let (D) denote the maximum size of an independent set in D.
Theorem 1.21 (Gallai-Milgram 1960). Every digraph D has a path partition with (D) paths.
Proof. By induction on |D|. Will show: if P is a path partition with |P| > (D) then Q with |Q| = |P| 1
and Start(Q) Start(P) where Start(P) is the set of starting vertices of paths in P.
****** picture
Let Pu be a path starting at u, for some u Start(P). Since |P| > (D), uv
~ where v Start(P). If
len(Pu ) = 0, replace Pv by uvPv and were done. If len(Pu ) 1, then uw
~ Pu . Let D0 = D u. Notice
(D0 ) (D). Let Q be a path partition of D0 . Notice |Q| = |P| > (D) (D0 ). By induction, Q such
that |Q0 | = |Q| 1 and Start(Q0 ) Start(P) {u} + {w}.
Let be the size of a maximal matching in a bipartite digraph. Then the min size of a path cover is
n 2 + = n where X has n vertices. ***** picture
Corollary 1.22. K
onigs Theorem
Corollary 1.23 (Dilworth 1960). In every finite poset (P, ), max size of an antichain = min size of chain
partition.
Proof. Let e be a chain partition and A the max antichain. Certainly |e| |A|. WWTS |A| chains suffice.
Use Gallai-Milgram on D with edges {(x, y) : x < y}. In this graph, antichain independent set and chain
cover path cover.

1.5

Connectivity

Definition 1.24. G is k-connected if the minimum size of a separator is k. The connectivity (G) =
max k such that G is k-connected.
Definition 1.25. A block is a maximal connected (sub)graph with no cut-vertex.
Examples of blocks are K1 , bridges, maximal 2-connected subgraphs, etc. We can form a natural block
graph that is a bipartite graph with one set of vertices as the blocks and the other set as the cut vertices
that blocks share.
***** picture
Proposition 1.26. The block graph of a connected graph is a tree.
Proof. G is connected the block graph is connected. Can the block graph have cycles? No, by the
maximality of blocks.
Proposition 1.27. A graph is 2-connected sequence cycle = G0 , G1 , . . . , Gn = G such that Gi+1 is
obtained from Gi by adding a Gi -path.
***** picture ***** ear decomposition

Proof. () Trivial, since Gi 2-connected Gi+1 2-connected.

() Given a sequence G0 , G1 , . . . , Gi , suppose Gi 6= G. Then e E(G) E(Gi ). Let e = xy with
x V (Gi ). We know G x is connected so path from y to z V (Gi ). Adding e gives Gi+1 .
Theorem 1.28 (Tutte 1961). G is 3-connected seq G0 , G1 , . . . , Gn = G such that Gi+1 has an edge
e = xy with deg(x), deg(y) 3 and Gi = Gi+1 /e.
Lemma 1.29. If G is 3-connected with |V (G)| > 4 then e E(G) such that G/e is 3-connected.
Proof. Suppose not. Then e = xy E(G), G/e has a cut-set of 2 vertices. We know, then, that Vxy S,
the cut-set, since (G) 3. Let S = {Vxy , z}. Then X = {x, y, z} is a cut-set of G every vertex in X
has an edge to every component of G X. Let C be the smallest component of G X over all {x, y, z}.
Let w N (z) C. By assumption, G/f has a cut-set of size 2, so v such that {v, z, w} is a cut-set of
G and each of these vertices has an edge to every component of G {v, z, w}. Since x, y are connected,
component D that does not contain x and y. Thus, =
6 D N (w) V (C), so D ( C, contradiction our
assumption that C was the smallest such component. ******* picture *********
Proof of theorem. () by Lemma
() K4 is 3-connected. Suppose Gi is 3-connected but Gi+1 is not. Then Gi = Gi+1 /e (e = xy) where
Gi+1 is 2-connected. Let S be a cut-set of size 2; let C1 , C2 be two components of Gi+1 S. Since x, y
are connected, we may assume V (C1 ) {x, y} = . Also, C2 cannot contain both x and y (otherwise S
is a cutset of Gi ), nor can it contain any v
/ {x, y} (otherwise Vxy will be disconnected from C1 in Gi by
removing 2 vertices). This is a contradiction of the degree assumption!
Theorem 1.30 (Tuttes Wheel Theorem). Every 3-connected graph can be obtained by the following procedure:
Start with K4
Given Gi pick a vertex v
Split into v 0 and v 00 and add edge {v 0 , v 00 }
Today we decide whether k-connectivity is equivalent to having k independent paths.
Definition 1.31. Let A, B V (G). An A B path is a path P = (u, . . . , v) where P A = {u} and
P B = {v}.
A set S is an A B separator if there is no A B path in G S.
Theorem 1.32 (Menger 19217). The minimum size of an A B separator = maximum number of disjoint
A B paths.
Proof. Let k = min size of an A B separator. Clearly #paths k. We will construct k disjoint A B
paths, by induction on |E(G)|.
Base: |E(G)| = 0. Here |A B| = k and k vertices form trivial paths.
Inductive: Suppose xy = e E(G). Consider G/e. Put Vxy in A or B (or both) if x or y is in A or B.
Suppose the max # of disjoit A B paths in G is k 1. Then the same holds in G/e. By induction,
A B separator S 0 of size k 1 and Vxy S 0 (otherwise S 0 is an A B separator in G).
Now, S = S 0 \ {Vxy } {x, y} is a separator in G of size k. Consider G0 = G e. Note: every A S separator
and every S B separator is also an A B separator; therefore, the mnimum size of an A S separator
is k. By induction, k disjoint paths from A to S, likewise from S to B. These paths cannot intersect
outside A S B. Since |S| = k, combine the 2 sets of paths. Done.
Definition 1.33. Suppose B V (G) and a V (G) \ B. An a B fan is a set of paths from a to B that
intersect only at a.

Corollary 1.34 (Fan Theorem). Min # of vertices needed to separate a from B = max size of an a B
fan.
Proof. Apply Mengers Theorem to A = N (a) and B.
Corollary 1.35 (Local Version of Mengers Theorem).
= max # internally disjoint a b paths.

1. If ab
/ E(G), then min size of a b separator

2. If a 6= b, min # edges needed to separate a from b = max # edge disjoint a b paths.

Proof.

1. Apply Mengers Theorem to A = N (a) and B = N (b).

2. Apply Mengers Theorem to the line graph of G: A = E(a) := {e E(G) : e is incident to a} and
B = E(b).
Corollary 1.36 (Global Version of Mengers Theorem).
between any 2 vertices.

1. G is k-connected k independent paths

2. G is k-edge-connected k edge-disjoint paths.

Proof.
1. Done except when ab E(G) (rest follows from Local Version 1). Suppose ab E(G) and let
G0 = G ab. If G0 has k 1 disjoint a b paths, were done, so suppose otherwise. Then we know the
max # disjoint a b paths in G0 is k 2 and so, by Mengers Theorem, an a b separator S of
size k 2. Since |V (G)| > k, w
/ S {a, b}. S is either an a w separator or a b w separator
(otherwise an a b path not hitting S), but S {b} is an a w separator in G of size k 1. This is
a contradiction.
2. Follows from the Local Version 2.
Definition 1.37. A graph G is k-linked if for any two sets of size k (say, with vertices {a1 , . . . , ak } and
{b1 , . . . , bk }) we can find disjoint paths from ai to bi .
Observation: k-linked k-connected.
Question: If a graph is f (k)-connected, can this be enough to guarantee k-linked? Is this even possible? If
so, for which f (k) is this true?
2

Theorem 1.38 (Jung, Larman, Mani 1970). If a graph is 210k , this be enough to guarantee k-linked.
Observation: If graph is k-connected then the average degree minimum degree k.
Proposition 1.39. If a graph has average degree d, then it has a subgraph with all degrees > d2 .
Proof sketch, algorithmic. Algorithm for finding that subgraph: if we have any vertex of degree d2 , throw
it away. Question: Why does this stop before all vertices gone? Answer: As we do this, the average degree
is nondecreasing (basically...). The condition E d2 n is preserved.
New E = Old E degree of deleted vtx Old E

d
d
d
d
n = (n 1)
2
2
2
2

Proposition 1.40. If all degrees in graph are then graph has cycle of length + 1.
Proof. Suppose you take the longest path and let v be the last vertex. All neighbors of v must fall back onto
path, otherwise theres a longer path. Since there are such neighbors, choose the furthest one from v
along the path and close the path to make a cycle. This has length + 1.
8

Corollary 1.41. If average degree d then we have a cycle of length

d
2

+ 1.

Definition 1.42. A graph has a topological Kr minor if r branch vertices and

connecting them.

r
2

vertex-disjoint paths

Question: What average degree, if any, is enough to guarantee existence of a topological Kr minor?
Remark 1.43. Tur
ans Theorem says average degree
does not depend on n.

r2
r1 n

Kr subgraph. We hope for a bound that

Lemma 1.44. Average degree 2(2) we have a topological K r minor.

m
Proof. Consider

only r 3. Induction (on m): Prove the statement: average degree 2 where m =
r
r, r + 1, . . . , 2 then we have a topological minor with r vertices and m edges (topologically, meaning we
have r branch vertices and some m vertex-disjoint paths between them).
Base case: Given average degree d 2r , find a topological minor with r vertices and r edges, i.e. an
r-cycle. By the previous proposition/observation, we have a cycle of length 2r1 + 1 r. We can turn
this topologically into an r-cycle by choosing r of the vertices if the length is > r.
Inductive step: Assume true for m 1. Given average degree d 2m . Would be nice if we could get
a connected set U such that inside N (u) the average degree is 2m1 . Then, by induction we can find a
topological copy of r vertices with m 1 edges. Connected back to U gives us an extra edge to make m,
since U is connected. Now, we need to find such a set U .
Since the average degree 2m in all of G some component of G has average degree 2m , so WOLOG G
m1
. Suppose
is connected. Pick U maximal such that U is connected and if U is contracted then edges
vtxs 2
m1
deg(v) inside N (U ) is < 2
. Then what if we were to add v to U ? U would still be connected, and after
contracting U + v then nw edges 2m1 (vtxs) 2m1 . This contradicts U being maximal. We know such
a U exists because we can pick U to be any one vertex with high degree.


3k
Theorem 1.45. If a graph is 2( 2 ) + 2k -connected then it is k-linked.

Proof. Fix any vertex sets {a1 , . . . , ak } and {b1 , . . . , bk } and find disjoint paths between each ai and bi .
3k
Find a topological K3k ; notice its still 2( 2 ) -connected. Mengers Theorem allows us to connect 3k
branch vertices with 2k vertex-disjoint paths **** picture **** while minimizing the # of edges not on the
topological K3k . Let c1 , . . . , ck be the unused branch vertices. **** picture **** show there cant be crossing
of topological T path and Menger M path.
Theorem 1.46 (Thomas, Wollan 2005). 2k-connected and average degree 10k k-linked.
Note: 10k-connected implies hypotheses of theorem.
Corollary 1.47. Average degree 8r2 topological Kr minor.
Proof. (Maders Theorem, Diestel Thm 1.4.3) have subgraph which is r2 -connected and has average
degree 5r2 12 r2 -linked. Pick r branch vertices and r 1 neighbors of each. This is r2 vertices and can
dictate links between all 21 r2 pairs of vertices.
Let G = (V, E) with an enumeration of the edges e1 , e2 , . . . , em . We want to define a vector space with
m dimensions.
Definition 1.48. Formally, we let G = (V, E) be a fixed graph with |V | = n and |E| = m. The edge
space E(G) is the vector space over F2 of all functions f : E F2 with the usual vector addition on F2 (so
this corresponds to the symmetric difference of two subgraphs of G). Note: a basis for the edge space is
{{e1 }, {e2 }, . . . , {em }}.

For any subspace F E(G), let

F = {D E(G) : hF, Di = 0 F F}
where
hF, F 0 i =

m
X

i 0i

for F = (1 , . . . , m ), F 0 = (01 , . . . , 0m )

i=1

Notice dim F + dim F

= m.

Definition 1.49. The cycle space C = C(G) is the subspace of E(G) spanned by all cycles in G.
Question: What is dim C? Goal: Prove dim C = m n + 1 (when G is connected, otherwise we just
consider each component separately).
Proposition 1.50. The induced cycles in G generate its entire cycle space.
Proof. By induction on the number of vertices in a given cycle.
Proposition 1.51. TFAE:
1. F C(G).
2. F is a disjoint union of (edges sets of ) cycles.
3. All vertex degrees of the graph (V, F ) are even.
Proof. (1)(3): Symmetric difference preserves the even parity.
(3)(2): Induction on |F |. If F 6= then F contains a cycle C. Remove those edges and repeat.
(2)(1): By definition, disjoin union is a sum of vectors.
Definition 1.52. The cut space of G is *****************
Proposition 1.53. Together with , the cuts in G form a subspace C ? . This space is generated by cuts of
the form E(v).
Example 1.54. ******* picture
Proof. Let C ? denote the set of all cuts in G, plus . WWTS D, D0 C ? D + D0 C ? . Recall
D + D0 = DD0 . Set V1 = (V1 V10 ) (V2 V20 ) and V2 = (V1 V20 ) (V2 V10 ). Then D + D0 corresponds

to all edges between VP

1 and V2 , so it is also a cut. (pick the diagonal, essentially)
Next, E(V1 , V2 ) = vV1 E(v). ********** picture
Definition 1.55. A minimal non-empty cut in G is a bond.
Remember: minimal in the sense of containment of the sets of crossing edges.
Example 1.56. ************** pictures
Observation: A cut is a bond (in a connected graph) both sides of the corresponding vertex
partition are conneceted induced subgraphs.
Proposition 1.57. Every cut is a disjoint union of bonds.
Proof. Take D C ? . Look at the components of V1 and V2 .... ************
Theorem 1.58. The cycle space C and the cut space C ? of any graph satisfy
1. C = (C ? ) and
2. C ? = C
10

Proof. (1) WWTS C (C ? ) . Note that any cycle in G has an even number of edges in each cut. Also,
observe that hF, F 0 i = 0 F and F 0 have an even number of edges in common. So hC, Di = 0 for all
D C?.
For the other direction, WWTS F
/ C F
/ (C ? ) . So v V (F ) such that deg(v) is odd. So then
hE(v), F i = 1.
(2) It suffices to show that C ? = ((C ? ) ) . [This is true, for free, assuming some knowledge of finitedimensional vector spaces.]
First, F C ? F 0 (C ? ) we have hF, F 0 i = 0. Next, dim C ? + dim(C ? ) = m = dim(C ? ) +
dim((C ? ) ) .
Definition 1.59. Let G be a given connected graph and let T be a spanning tree of G. Let e E(G) \ E(T ).
Then Ce is the fundamental cycle with respect to T .
Definition 1.60. Given G and T a spanning tree and e E(T ), then De is the fundamental cut.
Theorem 1.61. Let G be a fixed connected graph and let T be a fixed spanning tree of G. Then the
corresponding fundamental cycles and cuts form a basis of C and C ? , respectively. Also, dim C = n m + 1
and dim C ? = n 1.
Proof. Pick e E(T ) and try to write the fundamental cut as a sum. You cant! So the set of fundamental
cuts is a linearly independent subset of C ? , and thus dim C ? n 1. Similarly, dim C m n + 1. Now,
dim C ? + dim C = m = (n 1) + (m n + 1) dim C ? + dim C
so theyre all equal.

11