You are on page 1of 5

0.

1 Graph theory problems and solutions


Problem 0.1.1 If graph G can be colored in k colors and k is minimal such number, then there
exists way of length k which goes through each color.

Proof:
Name colours as c1 , c2 , . . . , ck and let order them as c1 > c2 > . . . > ck . Let graph G can
be colored in colors c1 > c2 > . . . > ck (where minimality means order minimality). Let v1 →
v2 → . . . vl be some way where vi has color ci . If there is no elements from G with color cl+1 and
connected with vl , then we can change color of vl from cl to cl + 1 and get graph G0 where G0 < G,
but then by induction we have that graph G0 has some way u1 → u2 → . . . → uk , where color of
ui is ci . So ul+1 6= vl , because otherwise in graph G vartices vl and ul are connected by edge and
have same color. So way u1 → u2 → . . . → uk is suitable for graph G.
So we can continue way v1 → v2 → . . . → vl → vl+1 , where vl+1 colored in color cl+1 . 

Problem 0.1.2 Consider oriented graph G, where for every verticle v number of outgoing oriented
edges from v is not greater than k. Prove that G can be colored in 2k + 1 colors.

Proof:

Lemma 0.1.3 For some verticle x ∈ G, number of ingoing oriented edges to in x is not greater
than number of outgoing oriented vertices.

Proof of lemma 0.1.3:


For any oriented edge → consider numbers −1, +1 on it’s end points, so we get edges {−1} →
{+1}. If we add all numbers, then we will get S = 0, because −1 + 1 = 0 for any oriented edge.
So for some verticle x ∈ G, number of ingoing oriented edges to in x is not greater than number
of outgoing oriented vertices, because otherwise S < 0. 
From lemma 0.1.3 we get that for some verticle x ∈ G, number of ingoing oriented edges to in
x is not greater than number of outgoing oriented vertices, so degree of x is not greater than 2k.
If we considered induced oriented graph G0 := G \ {x} and use induction hypothesis on it, then
we will get some coloring of G0 in 2k + 1, colors. Degree of x is less than 2k + 1, so we can color
x in suitable color from given set of 2k + 1 colors. 

Problem 0.1.4 Let degree of any vertex of graph G is not greater than d. Consider natural
numbers d1 , d2 , . . . , dk , such that d1 + d2 + . . . + dk = d + 1 − k. Then we can partition graph G on
k subgraphs G1 , . . . , Gk , such that induced degree of any vertex in subgraph Gi is not greater than
di , for every 1 ≤ i ≤ k.

Proof:
Induction on k. We will prove case k = 2 at the end.
Let we know that statement is correct for k 0 = 1, 2, . . . , k − 1, so we need to prove it for k 0 = k.
For any set of natural numbers d1 , . . . , dk , with property d1 + d2 + . . . + dk = d + 1 − k, consider
natural number d0k−1 := dk−1 + dk + 1. So we have that d1 + d2 + . . . + dk−2 + d0k−1 = d + 1 − (k − 1)
and from induction hypothesis for k − 1 and numbers d1 , . . . , dk−2 , d0k−1 we get that there exists
partition of G on subgraphs G1 , G2 , . . . , Gk−1 , such that ∀i < k − 1 induced degree of any vertex
in Gi is not greater than di and induced degree of any vertex in Gk−1 is not greater than d0k−1 .
Then we can use induction hypothesis for k = 2 to subgraph Gk−1 (in which degree of any vertex
is not greater than d0k−1 ) and numbers dk−1 , dk , where dk−1 + dk = d0k−1 − 1 = d0k−1 + 1 − 2. So
we can partion Gk−1 in two subgraphs G0k−1 , G0k , where induced degree of graph G0i is not greater
than di , i = k − 1, k. So G1 , . . . , Gk−2 , G0k−1 , G0k – is partition of G.
Consider case k = 2 i.e we have numbers d1 + d2 = d − 1. For any (ordered) partition (G1 , G2 )
consider operations O1 , O2 , where O1 : Is consideration of vertex inside G1 with induced degree

1
in G1 , > d1 and replacing this vertex from G1 to G2 and get new partition (G01 , G02 ). Like the
same define operation O2 , which replaces vertexes from G2 to G1 .
Let statement isn’t correct in case k = 2. One can easily prove that there Q exists partition
(G1 , G2 ) of G and sequence of integers σ1 , σ2 , . . . , σp ∈ {1, 2}, such that (G1 , G2 ) = pi=1 Oσi (G1 , G2 ).
We know that if Oσi (G1 , G2 ) = (G01 , G02 ), then |G01 | − |G1 | = (−1)i , so #{i|σi = 1} = #{i|σi = 2}.
Also easy to check that (from condition d1 + d2 = d − 1) if O1 (G1 , G2 ) = (G01 , G02 ), then number of
edges in G01 is less than number of edges of G1 minus d1 and number of edges 0
Qp in G2 is not greater
than number of edges of G2 plus d2 , same for O2 . So from (G1 , G2 ) = i=1 Oσi (G1 , G2 ) we get
that −#{i|σi = 1}d1 + #{i|σi = 2}d1 > 0. Contradiction. 

Proof of Dilworth’s theorem :
See formulation here : [2].
Consider next operation O : for any partial ordered set G, O(G) – is consideration of minimal
partitioning of G in ordering chains.
So consider maximal antichain A ⊆ G. Consider any v ∈ G \ A, let v 0 be maximal (or minimal
if there is no maximal elements) element in set G \ {v ∪ X}. By induction hypothesis number of
chains in O(G \ {v 0 }) is equivalent to |X|. Consider all elements Y from chains in O(G \ {v 0 })
which are less than corresponding elements in set X in same chain. By induction if Y 6= 0, then
number of chains in O(G \ Y ) is |X|, so from construction of v we have that chains from O(G \ Y )
can be glued to chains from Y and total number of chains in G is X. Assume that Y = 0. Then
easy to see that elements from antichain X are minimal in G and any minimal element in G
belongs to X. Consider any x ∈ X. If number of chains in O(G \ {x}) is |X| − 1, then easy to
see that number of chains in G is |X|, otherwise number of chains in O(G \ {x}) is |X|, so from
minimality of x and same ideas as in previous cases we have that maximal antichain X 0 of G \ {x}
has size X and consists of minimal elements from G \ {x}. So we can add x to some chain in
O(G \ {x}), because x is comparable to some element in X 0 and is < than it. 

Remark 0.1.5 Note that dual Dilworth’s theorem [2] has same proof, where we only need to
change definition of operation O.

Proof of Brook’s theorem :


See formulation here : [3].
First one can easily prove Brook’s theorem for d = 3.
Induction on d. Consider any vertex v ∈ G and from induction hypothesis we can color graph
G \ {v} in d colors. Consider graphs G \ {v} = G1 ∪ G2 ∪ . . . ∪ Gd , such that vertices from Gi have
same color ci and ci 6= cj , for every i 6= j. If there exists i, such that ∃vi ∈ Gi is not connected to
vertex v, by edge, then we can color v in color ci and get coloring of G in d colors. From condition
that degree of v is ≤ d we get that v is connected to Gi by exactly one edge for every i. Consider
induced subgraph G0 := {v} ∪ G1 ∪ G2 ∪ . . . ∪ Gd−1 . If there exists verticle x ∈ G0 with induced
degree ≥ d, then easy to see that x is not connected to graph Gd by edge, so if v = x then can
color v in color cd and get coloring of G, and if x 6= v, then color x in cd and minimize number
|G0 |.
So we can think that induced degree of any vertex in G0 is ≤ d−1. So from induction hypothesis
if graph G0 has not complete graph subgraph Kd then we can recolor it in d − 1 colors and so
graph G is colored in d colors. So graph G0 consists of complete graph subgraph Kd . Easy to see
that v ∈ Kd because G0 \ {v} can be colored in d − 1 colors. Also note that Kd intersects every
subgraph Gi , i < d by exactly one vertex.
Assume that G can’t be colored in d colors. Name graph Kd from previous discussion as Kd (n),
so if we will consider other graphs G \ Gi , then we’ll get another complete graphs Kd (i). Vertex v
is connected to Gi ’s by exactly one edge, so graph ∪i Kd (i) is exactly complete graph with d + 1
vertices. Contradiction. 

2
0.2 Proofs of some problems from book
Here I considered solution to some exercises from book [1, Chapter 8].
Proof of (i) page 2:
Consider some coloring of graph G in 4 + 1 colors. Then consider next process : Consider two
vertices’s A, B with same color and connected by one edge, degree of A is ≤ 4, so there exists
color c which doesn’t appear in set of neighbors of A and A itself (because A and B have same
color). So we can change color of A on c.
So after each step of this process we will get new coloring with less number of equi-colored
vertices. So this process must stabilised and we will get good coloring in 4 + 1 colors. 
Proof of (ii) page 2:
In fact we will prove that this induced subgraph H is uniquely defined (in some sense). Consider
next operation O, where O(H) – is subgraph H 0 of H, such that |H 0 | = |H| − 1 and vertex from
H \ H 0 has less than VE neighbors from H, and if there is no such subgraph H 0 , then H 0 := H.
Easy to see that O(OV (G)) = OV (G), so we only need to prove that OV (G) 6= 0. Consider sets
Xi = Oi (G) \ Oi+1 (G), let l be such that Xl 6= 0, X l+1 = 0, so we get that G = X1 ∪ X2 ∪ . . . ∪ Xl .
Let ki := |Xi |, we have that number of edges from any vertex in Xi to verteces in X1 ∪X2 ∪. . .∪Xi is
less than VE . So number of edges in graph G is less than VE k1 + VE k2 + . . . + VE kl = E. Contradiction.

Proof of (vi) page 3:
Induction on n. If there exist verticle v, such that any directed edge between v and u goes to
v, i.e u → v, ∀u 6= v, then from induction hypothesis we have that there exists Hamiltonian path
in G \ {v}, so we can add v to it’s end and get Hamiltonian path in G. If there is no such vertices
v, then consider oriented cycle C, so if we consider c ∈ C and Hamiltonian path in G \ {c}, then
we can easily construct Hamiltonian path in G from this data. 
Proof of Example 1 on page 5:
It’s combination of facts (ii) and (iii) on page 2. 
Proof of Example 6 on page 9:
Consider maximal i, such that v1 and vi are connected by edge (where verticle vj has degree
di ). So consider case when i − 1 > d1 . In this case easy to see that there exists 1 < j < i, such
that vertices vi , vj are not connected. Also easy to prove that there exists vertex vk , such that
k 6= 1, i, j and vk is connected to vj by edge, but not connected to vi . So we can delete edges
v1 → vi , vj → vk and add edges vi → vk , v1 → vj . After this operation vector [d1 , d2 , . . . , vn ]
stay same, but number min{i : v1 → vi } decreases. So after some number of same operations we
can think that i − 1 = d1 . So we have that after deleting v1 graph G \ {v1 } has degree vector
[d2 − 1, d3 − 1, . . . , dd1 +1 − 1, dd1 +2 , . . . , dn ]. 
Proof of Example 8 on page 11:
Consider some partition of G in two connected subgraphs G1 , G2 (where |G1 | > |G2 | and |G2 |
−1)
is maximal). If |G2 | < VD−1 , then |G1 | ≥ V − VD−1 = (D−1)(V D
+ 1. G is connected, so there exists
v ∈ G1 , such that v is connected to G2 by edge. If graph G1 \ {v} is connected, then consider
another partition G1 \ {v}, G2 ∪ {v} with bigger |G2 |. So we can think that G1 \ {v} has connected
components I1 , I2 , . . . , Il . Verticle v must be connected to every Ii , so l ≤ D. And verticle v is
connected to G2 by edge, so l < D.
Let given that |I1 | ≥ |I2 | ≥ . . . ≥ |Il |. Then |I1 | ≥ |G1l|−1 ≥ VD−1 and we can consider partition
G01 := I1 = G1 \ [{v} ∪i>1 Ii ], G02 := G2 ∪ {v} ∪i>1 Ii of graph G in two connected components,
where |G01 | ≥ VD−1 and |G02 | > |G2 |. So after some of similar operations we can get partition of G
in two connected subgraphs G∗1 , G∗2 with |G∗i | ≥ VD−1 . 
Proof of Example 11 on page 15:
Case when Ai are single element sets is trivial. So there exists element a ∈ {1, 2, . . . , n} such
that Ai 6= a, ∀i. Consider n−1 - element set N := {1, 2, . . . , n}\{a} and it’s subsets Bi := Ai ∩N .
In case if sets Bi are pairwise different we get that x = a. So let some two sets from {Bi }i are same,
so we have n − 1 or less different sets in {Bi }i name them as {Bj0 }j . By induction hypothesis

3
6 Bj0 2 \ {x0 }, ∀j1 6= j2 . So easy to see that ∀i 6= j,
there exists x0 ∈ N , such that Bj0 1 \ {x0 } =
Ai \ {x0 } =
6 Aj \ {x0 }. 

4
Bibliography

1. Pranav A.Sriram, http://artofproblemsolving.com/community/c6t240f6h601134_


olympiad_combinatorics_book

2. Wikipedia, Dilworth’s theorem, https://en.wikipedia.org/wiki/Dilworth%27s_theorem

3. Wikipedia, Brook’s theorem, https://en.wikipedia.org/wiki/Brooks%27_theorem

You might also like