Professional Documents
Culture Documents
Graph
zxm
Abstract
1 Introduction
In a graph with weighting function G(V, E, ω), where ω : E(G) → R \ R− is a map from the edge
set of G to non-negative real number, we can easily find the maximum weight spanning tree of G
P
denoted by Tω . Then define the weighting sum of a graph as ω(G) := e∈E(G) ω(e). Sometimes we
are interesting about the weighting ratio of the maximum spanning tree and the total graph, denoted
as tω (G) := ω(Tω)
ω(G) .
A natural idea is that the maximum spanning tree can find the edge with larger weight and
put this edge into the maximum spanning tree. Thus the maximum spanning tree may be larger
than the uniform weight ratio of entire graph. When we refer to uniform weight ratio, we mean the
weighting ratio of maximum spanning tree when all edges are weighted 1, i.e. a weighting function
ω 1 (Tω1 )
ω 1 : E(G) → 1, and uniform weight ratio of G denoted as u(G) := ω1 (G) = |V|E(G)|
(G)|−1
.
But the fact is the weighting ratio can be much smaller than the uniform weight ratio, and we
have the follow theorem.
tω (G)
Theorem 1. For any ϵ > 0, there exist a graph G and ω, such that u(G) <ϵ
Proof. Consider a graph G constructed by a complete graph Kk and a tree T joined by a vertex with
|T | = t + 1, t ≥ 1. Then we weight edges in Kk 1, and edges in T 0. And in this weighting function
ω, we have:
2
(k − 1)/ k2 2 k 2−k + t
tω (G)
= = · (1)
u(G) (k + t − 1)/( k + t) k k−1+t
2
1
Figure 1: A complete graph Kk and a tree T joined by a vertex
As we have already seen, fixing a graph and its weighting function, we can have the tω (G). Next
we have to consider, when a graph is fixed, how to choose a weighting function, to minimum tω (G).
We denote all the weighting function of G as W , and then we want to get inf ω∈W tω (G).
a1 +b1 x
Lemma 3. If a1 , a2 , b1 , b2 all positive, f (x) = a2 +b2 x . If
b1
≥ f (x)
b2
for all x ≥ 0, then f ′ (x) ≥ 0.
2
Proof. This can be proved easily by derivative with respect to f (x).
Now consider the problem of minimizing Tω (G) when fixed the graph G. Since the map from
E(G) → R \ R− is infinite, it’s not possible to find the minimum Tω (G) from the infinite choices. But
the following lemma makes it possible.
Lemma 4 (0-1 Weighting Function Lemma). Define that W 0−1 is the set of all maps from E(G) →
{0, 1}. Then inf ω∈W tω (G) = minω0−1 ∈W 0−1 tω0−1 (G).
Proof. Suppose that ω ∈ W is the weighting function to obtain the infimum, and 0 ≤ a1 < a2 < · · · <
as ≤ 1 is all the value that ω can get. We may assume that they all lie in [0, 1], because we can times
a positive real number to make them arbitrary small and the spanning tree Tω will stay the same.
We define that
A0i : All edges weighted ai and not in the Tω .
A1i : All edges weighted ai and in the Tω .
Ai : All edges weighted ai , with Ai = A1i ∪ A0i .
Next we want to proof that the ω which achieve the infimum and with the least different kinds of
weighting value (denoted as s), must be a 2-value map, i.e. in this map, s = 2.
By contradiction, if ω is the map to achieve the infimum with the least kinds of different values
s ≥ 3, which are 0 ≤ a1 < a2 < · · · < as ≤ 1, so there exist i, 2 ≤ i ≤ s − 1. Then let a′i such that
ai−1 ≤ a′i ≤ ai+1 , and to replace ai with a′i , and the new weighting function denoted as ω ′ , Tω′ is the
maximum spanning tree about ω ′ . Then we proof that we may assume Tω = Tω′ :
Suppose Alg1 is the process of finding the maximum spanning tree with ω as weighting function,
and Tk is the tree in k-th step;
Suppose Alg2 is the process of finding the maximum spanning tree with ω ′ as weighting function,
and Tk′ is the tree in k-th step;
We want to prove that Tk = Tk′ , k = 1, 2, . . . n. By induction, when k=1, because ai < as and
a′i < as , so the first edge we choose is the edge weighted as , so we can let T1 = T1′ . Now we suppose
that Tk = Tk′ , k ≥ 1, and we want to prove that Tk+1 = Tk+1 ′ . Denote the edge we add to Tk in
k+1-th step is e, then consider the weight of ω(e) = aj :
Case 1. aj ≥ ai+1 : Since there is no edge weighted larger than aj and can be added into Tk
without producing circles and Tk = Tk′ , so does in Tk′ , so e is the weighting largest edge can be put in
Tk′ , so Tk+1
′ = Tk+1
Case 2. aj ≤ ai : We can analyze similarly to the case1, so we have Tk+1 = Tk+1 ′ too.
Case 3. aj = ai : Then we can find that when putting any edge weighted larger than ai+1 will
produce circles in Tk and so does in Tk+1 . So e is the weighting largest edge we can put in Tk+1 .
Since ω ′ (e) = a′i ≥ ai−1 , so e is also the largest weighting edge can be added in Tk′ , so Tk+1 = TK+1
′ .
So we can find that Tω = Tω′ , i.e. we can change weight of all edges weighting ai in the interval
[ai−1 , ai+1 ] and won’t change the maximum spanning tree.
3
And we have that:
ω(Tω ) |A1 |a1 + · · · + |A1i |ai + · · · + |A1s |as
= 1 (7)
ω(G) |A1 |a1 + · · · + |A1 |ai + · · · + |As |as
|A1 |
So when |Aii | ≥ ω(T ω) ′
ω(G) , we assume ω changes the edges weighted ai re-weight as ai−1 , and by above
analysis, Tω = Tω′ . So
ω(Tω′ ) |A11 |a1 + · · · + |A1i |ai−1 + · · · + |A1s |as
=
ω ′ (G) |A1 |a1 + · · · + |A1 |ai−1 + · · · + |As |as
|A1 |a1 + · · · + |A1i |ai + · · · + |A1s |as
≤ 1 (8)
|A1 |a1 + · · · + |A1 |ai + · · · + |As |as
ω(Tω )
=
ω(G)
The inequality is by lemma 3. So we find ω ′ is also the weighting function that minimize the weight
ratio but with s − 1 kinds of value, which is a contradiction about the definition of .
|A1 | ω(Tω′ )
Similarly, if |Aii | ≤ ω′ (G) , we can let ω ′ change all edges weighting ai re-weight ai+1 , then we can
ω(T )
ω′
also have ω′ (G) ≤ ω(Tω) ′
ω(G) , and ω has s − 1 kinds of value, which is a contradiction about the definition
of too. So s = 2.
|A11 |a1 +|A12 |a2 |A11 |
Now 0 ≤ a1 < a2 ≤ 1 is two different value of ω, with ω(T ω)
(G) = |A1 |a1 +|A2 |a2 . If ω(Tω )
|A1 | ≤ ω(G) , then
|A1 |
we can let a1 = 0 and get |A22 | ≤ ω(T ω) 1
ω(G) , then we multiply all weighting of edges by a2 , which doesn’t
change the ratio and the inequatlity is by lemma 3 .We finally get a 0-1 weighting function.
|A1 | |A11 |+|A12 |
If |A11 | < ω(Tω)
ω(G) , then let a1 = a2 , and get ω(Tω )
|A1 |+|A2 | ≤ ω(G) , then we multiply all weighting of edges
by a12 . We also get a 0-1 weighting function.
4
Now we find the weighting function of minimize the ratio of any chordal graph.
Proof. By induction on k, when k=3, we assume that a1 , a2 , a3 i is the weight of the three edges, we
may assume that a1 ≤ a2 ≤ a3 , and then ω(T ω) a2 +a3 2
ω(G) = a1 +a2 +a+3 ≤ 3 , established. Then when k ≤ 4, we
P
suppose that the V (G) = {v1 , v2 , . . . , vk }, and Ωi = j̸=i ω(vi vj ), i.e. the sum of all edges adjacent
to vertex vi . We may assume that Ω1 ≤ Ω2 ≤ · · · ≤ Ωk
Denote G′ = G \ v1 , G′ = Kk−1 , and
ω(T ) ω(T ′ ) + a∗
2
ω(Kk−1 ) + a∗
≥ ≥ k−1
ω(Kk ) ω(Kk−1 ) + Ω1 ω(Kk−1 ) + Ω1
2 ∗
k−1 ω(Kk−1 ) + a
≥ 2
ω(Kk−1 ) + k−1 ω(Kk−1 ) (10)
2(k − 2) (k − 2)a∗ 2(k − 2) k − 2 ω(Kk−1 )
= + ≥ + k−1
ω(Kk−1 )
k(k − 1) kω(Kk−1 ) k(k − 1) k 2
2
=
k
ω(Kk−1 )
Case 2. a∗ < :
(k−1
2 )
We have:
ω(T ) ω(T ′ ) + a∗
2
ω(Kk−1 ) + a∗
≥ ≥ k−1 (11)
ω(Kk ) ω(Kk−1 + Ω1 ω(Kk−1 ) + (k − 1)a∗
To prove the left part ≥ k2 , equally to proof
2k
2ω(Kk−1 ) + 2(k − 1)a∗ ≤ ω(Kk−1 ) + ka∗ (12)
k−1
ω(Kk−1 )
Which is equal to proof a∗ ≤ . By assumption we know its correct.
(k−1
2 )
As analyzed above, we can find a much more simple proof of theorem 6 by theorem 5.
5
|S|−o(S)
Proof. Since G in complete graph, any subset S of V (G), G[S] is also complete. Thus e(G[s]) =
|S|−1 2
= |s| .
(|S|
2 )
ω(Tω ) 2 2
So inf ω∈W ω(G) = minS⊆V (G) s = k
Now we can describe the 0-1 weighting function minimize the weighting ratio of any chordal.
Chordal graph has good properties, the following one is useful.
Lemma 7. G is a chordal with C1 , C2 , . . . , Ck are its maximal cliques, for any clique Ci , there exist
an ordering σ of {1, 2, . . . , k} such that σ(1) = i and Cσ(t) ∩ (∪s<t Cσ(s) ) is a clique.
3.2 The weighting function to minimize weighting ratio of any chordal graph
Then we have the following descriptions about the weighting function of chordal graph that minimize
the weighting ratio.
Before that, we define a clique graph of any graph:
Algorithm 8. For any graph, we can make a clique graph by following algorithm:
For any connected graph G, and G′ = ∅, T = ∅.
1. Choose any maximal clique C, C ̸⊂ G′ .
2.Assume that C joint G’ at few maximal cliques: S1 , S2 , . . . , Sk , and for any Si , there exist Ci
which is the maximal clique of G’ such that Si ⊆ Ci .
3.G′ = G′ ∪ C, T = T ∪ {vc }, there vc is a vertex respect clique C in graph T, and adjacent vc
with the vertices representing C1 , . . . , Ck .
Denote H(G)
e := T .
Theorem 11. If G is a chordal graph induced by C1 , C2 , . . . , Ck which are maximal cliques satisfying
Ci ∩ (∪j<i Cj ) is a clique, s is the size of maximum clique. Suppose that w∗ ∈ arg minω∈W0−1 ω(T ω)
ω(G) ,
F is the graph induced by edges that weighted 1. Then F satisfying following properties:
1): F is chordal graph and H(F e ) is the clique graph of F, then H(F
e ) is a tree, with all the leaves
of H(F
e ) is the maximal clique of G.
2): dF (V ) ≥ 2s .
1
3): s−1 ≤ tω (G) ≤ 2s
6
Proof. By induction on the number of maximal cliques k. When k = 1, it has been proved in theorem
6. Now we assume that it’s established when G has k maximal cliques, which also means given any
graph with t maximal cliques denoted Ft with t < k, there exist s, s < t connected maximal cliques
included in Ft satisfying the above properties denoted as Fs ⊆ Ft , with v(F s )−1 v(Ft )−1
e(Fs ) ≤ e(Ft ) . We need
to proof that given any subset S of V(G), there exist t connected cliques denoted as F such that
v(F )−1 |S|−o(S)
e(F ) ≤ e(G[S]) and these t cliques also satisfying above properties.
Given any induced subgraph F with D1 , D2 , . . . , Dr are its maximal cliques , and di := |Di |. In
which {Dj }j∈J are the maximal cliques which are leaves in the clique graph. If r ≤ k − 1, we may
assume that Di ⊆ Ci , and F ′ = ∪ri=1 Ci . By induction hypothesis there exist r′ ≤ r connected cliques
|S|−o(S) v(F ′ )−o(F ′ )
{Ci }i∈I construct H, |I| = r′ and I ′ ⊆ [r] such that v(H)−1
e(H) = minS⊆V (F ′ ) e(G[S]) ≤ e(F ) , and
H satisfying above properties, established. We may assume that Di ∩ (∪j<i Dj ) is a clique. So suppose
Pi = Di ∩ (∪j<i Dj ),and pj = |Pj |. Now we have:
ω(Tω ) d1 + · · · + dk − p2 − · · · − pk
= d1
(13)
ω(G) + · · · + d2k − p22 − · · · − p2k
2
If a vertex v of Ci also included in another maximal clique, we call v is in the public part of Ci , while
if it isn’t included in any other maximal cliques, we call it in the non-public part. When add vertices
to any clique in {Dj }j∈J (for example D1 in one of them) in non-public part, the ratio becomes:
ω ′ (Tω′ ) d1 + · · · + dk − p2 − · · · − pk + 1
= (14)
ω ′ (G) d1
+ · · · + d2k − p22 − · · · − p2k + d1
2
ω ′ (Tω′ )
So the molecular of ω(T ω) d1 dk
p2 pk
ω(G) − ω ′ (G) is d1 (d1 +· · ·+dk −p2 −· · ·−pk )−( 2 +· · ·+ 2 − 2 −· · ·− 2 ),
which is monotonically increasing with respect to d1 , so when d1 is large enough, adding vertex in
non-public part of D1 , the ratio will decrease and with adding vertices it will keep decreasing, while
it will increase when d1 in small and deleting the vertices will keep decreasing the ratio. So adding all
non-public vertices of Ci to Di or delete all non-public vertices of Ci from Di will meets the smallest
ratio about changing the non-public part.
So if there exist a clique Di ∈ {Dj }j∈J such that delete all its non-public part will meet the
smallest ratio, then we will find ∪j̸=i Dj achieve smaller ratio will k − 1 maximal cliques, then by
inductive hypothesis as mentioned above, we have find what we want.
Then if any clique Di ∈ {Dj }j∈J is large enough and adding vertices in non-public part achieve
′ ′
the smallest ratio, then from lemma 2 we can have that d1j < dj1−1 ≤ ωω′(T ω)
(G) for any {Dj }j∈J , otherwise
we can delete a vertex in non-public part to achieve a smaller ratio.
Next we consider the public part of {Dj }j∈J . When adding a vertex in public part, for example
in ∩i∈I Di (∩i∈I Di ̸= ∅), now we may assume that 1 ∈ I. Then denote ai := |Di | − |Pi |, then the ratio
becomes:
ω ′ (Tω′ ) d1 + · · · + dk − p2 − · · · − pk + 1
′
= d1 dk p2 pk
(15)
ω (G)
P
2 + · · · + 2 − 2 − · · · − 2 + d1 + i∈I\{1} ai
′ ′
For d1j < ω(Tω) P 1 ω(Tω ) ω (Tω ) ω(Tω )
ω(G) , so d1 + i∈I\{1} ai < ω(G) , from lemma 2 we can know that ω ′ (G) ≤ ω(G) . So in this
case, adding vertices in public part will decrease the ratio. So in this way we will have k maximal
7
cliques. If these k maximal cliques in not connected, and L1 , . . . , Lq are its component, so
Theorem 12. If G is a graph with n vertices and m edges such that tk (n) + 1 ≤ m ≤ tk+1 (n), k ≥ 2.
Suppose that F is the maximum chordal subgraph of G, when n is large enough, there must be a k + 1
clique in F.
Proof. Assume that F has no k + 1 clique, then the maximum clique of F denoted as s is at most k.
1
Then from the theorem 11, we know that tw (F ) ≥ s−1 . Since tω (G) is the minimum weighting ratio,
then when all edges in F weighted 1, the ratio can’t be smaller than tω (F ). So
1 1 v(F ) − 1 n−1
≤ ≤ tω (F ) ≤ = (18)
k−1 s−1 e(F ) e(F )
So we have that:
e(F ) ≤ (k − 1)(n − 1) (19)
√
1 p k+1
(k − )n + 2(k + 1)a/k − − O( n) ≥ (k − 1)(n − 1) (21)
k 2
8
4 Some explorations on chordal graph
Theorem 13. If G is a chordal graph, and G is not a complete graph, then there exist two vertices
u, v ∈ V (G), and N (u), V (v) are both clique.
Corollary 15. If H is chordal graph and not complete, then there exist e1 ∈ E(H), but e2 ∈
/ E(H),
and H − e1 + e2 is chordal graph.
References