• No results found

Online Full Text

N/A
N/A
Protected

Academic year: 2020

Share "Online Full Text"

Copied!
6
0
0

Loading.... (view fulltext now)

Full text

(1)

Matchings Extend to Perfect Matchings on

Hypercube Networks

Y-Chuang Chen

∗†

Kun-Lung Li

Abstract—In this work, we investigate in the prob-lem of perfect matchings with prescribed matchings in the n-dimensional hypercube network Qn. We obtain the following contributions: For any arbitrary match-ing with at most n−1 edges, it can be extended to a perfect matching of Qn for n≥1. Furthermore, for any arbitrary non-forbidden matching with n edges, it also can be extended to a perfect matching of Qn for n 1. It is shown by J. Fink in 2007 that any arbitrary perfect matching of the n-dimensional hy-percubeQn,n≥2, can be extended to a Hamiltonian cycle. Therefore, it leads to a further result that for any arbitrary non-forbidden matching with at mostn edges, it can be extended to a Hamiltonian cycle of Qn for n≥2.

Keywords: matching, perfect matching, Hamiltonian cycle, hypercube

1

Introduction

The underlying topology of an interconnection network

is usually modeled as a graph G = (V, E), in which

the vertex set V(G) represents processors and the edge

setE(G) represents connections between processors. We

use graphs and networks interchangeably. For the fun-damental graph definitions and notations we follow [2].

G= (V, E) is a simple graph ifV is a finite set andE is

a subset of {(a, b)|(a, b) is an unordered pair ofV}. We

say that V is the vertex set and E is the edge set. The

neighborhood ofv,N(v), is{x|(v, x)∈E}. Two vertices

are adjacent if (a, b) E. A path is a sequence of

ad-jacent edges (v0, v1),(v1, v2),· · ·,(vm−1, vm), written as

hv0, v1,· · ·, vmi, in which all the vertices v0, v1,· · ·, vm

are distinct. Acycleis a path with at least three vertices

such that the first vertex is the same as the last one. Hence, the length of a path or cycle is the number of

edges in the path or cycle. Letuandvbe two vertices in

a graph, thedistance fromuto v, denoted bydist(u, v),

is the minimum length of any path fromutov. A cycle

is a Hamiltonian cycle if it traverses every vertex of G

exactly once [4, 5].

This work was supported in part by the National Science

Coun-cil of the Republic of China under Contract NSC 98-2221-E-159-015.

Corresponding author: Y-Chuang Chen, Department of

Information Management, Ming Hsin University of Science and Technology, Hsin Feng, Hsinchu 304, Taiwan, R.O.C. Email:[email protected]

A matching M in graph G is a set of pairwise

non-adjacent edges, that is, every vertex is incident with at

most one edge of M. A vertex is matched if it is

in-cident to an edge in the matching M. Otherwise the

vertex is unmatched. A maximum matchingis a

match-ing that contains the largest possible number of edges.

The matching number of a graph is the size of a

maxi-mum matching. Aperfect matchingis a matching which

matches all vertices of the graph. That is, every ver-tex of the graph is incident to exactly one edge of the

matching. A matching is a forbidden matching if every

vertex of N(v) is matched for some unmatched vertex

v∈V(G). Otherwise, it is a non-forbidden matching. A

perfect matching is also aminimum-size edge cover. For

any graph G without isolated vertices, the sum of the

edge covering number and the matching number equals

to the number of vertices. If the graph G has a perfect

matching, then both of the edge cover number and the

matching number are equal to|V(G)|/2 [11].

Perfect matchings can be applied to network

disclo-sure attacks. The Perfect Matching Disclosure Attack

performs a high rate of success when tracing messages sent through a threshold mix in arbitrary scenarios [12]. Another related work concerning perfect matchings is

matching preclusion. A matching M in graph G is

called an almost perfect matching ifM contains exactly

(|V(G)| −1)/2 edges. An edge setF in graphGis called

a matching preclusion set if G\F has neither a perfect

matching nor an almost perfect matching. If a network has a large size of matching preclusion set, each vertex (process) in this network will has a specific partner in the event of edge (link) faults and the network will be robust. On the side, matching preclusion is also related

to the problem ofconnectivityof networks [3, 10].

Thehypercubeis a popular network because of its

attrac-tive properties, including regularity, symmetry, small di-ameter, strong connectivity, recursive construction, par-titionability, and relatively low link complexity [1, 9]. It

has been widely used in parallel systems, such as the

Con-nection machines, Symult S-series, Intel iPSC, iPSC/2,

and SGI Origin 2000 [6, 13, 14]. The formal

defini-tion of an n-dimensional hypercube is given as follows.

Each vertex v in Qn can be distinctly labeled by a

bi-nary n-bit string, v = vnvn−1· · ·v1. For 1 i n,

we use vi to denote the binary string v

(2)

The Qn consists of all n-bit binary strings

represent-ing its vertices. Two vertices u and v are adjacent if

and only if v = ui with some i. An n-dimensional

hypercube Qn can be constructed from two identical

(n−1)-dimensional hypercubes,Qi,n01andQi,n11, for some

1 i n, where V(Qi,n01) = {vnvn−1· · ·v1|vi = 0}

and V(Qi,n11) ={vnvn−1· · ·v1|vi = 1}. The vertex set

of Qn is V(Qn) = Qi,n01 ∪Q

i,1

n−1, and the edge set is

E(Qn) = E(Qi,n01)∪E(Qi,n11)∪P where P is a set of

edges connecting the vertices ofQi,n01andQi,n11in a one

to one fashion. Ann-dimensional hypercube can be

rep-resented as Qn =Qi,n−01

L

Qi,n11 withi∈1,2,· · ·, n. We

need some more terms, let v be a vertex of Qi,n01 and

Qi,n11, we use v0 to denote the corresponding vertex ofv

inQi,n11andQni,01, respectively. That is, (v, v0) is an edge

of Qn, one ofu, v is inQni,−01, and the other is inQi,n11.

We notice that the hypercube Qn is vertex-symmetric

and edge-symmetric forn≥1.

Some results on perfect matchings of hypercubes are dis-cussed in [7, 8, 10]. In this paper, we shall show that

for any arbitrary matching with at most n−1 edges of

the hypercube Qn for n 1, it can be extended to a

perfect matching of Qn. Furthermore, for any arbitrary

matching with exactly n edges of the hypercube Qn for

n 1, it can be extended to a perfect matching ofQn,

except it is a forbidden matching of Qn. It has shown

that any perfect matching of the hypercube Qn, n 2,

can be extended to a Hamiltonian cycle [7]. As a result, we have a further result that any non-forbidden matching

with at mostnedges of the hypercubeQn,n≥2, can be

extended to a Hamiltonian cycle.

The rest of this paper is organized as follows. Section 2 shows the perfect matchings with prescribed matchings

in then-dimensional hypercubeQnforn≥1. In Section

3, we give the conclusion remarks.

2

Perfect Matchings of Hypercubes

Lemma 1 For the 3-dimensional hypercube Q3, let M

be an arbitrary matching with at most two edges in the

hypercubeQ3. Then, the matchingM can be extended to

a perfect matching ofQ3.

Proof. First, suppose that M contains exactly zero or

one edge, it is clear thatM can be extended to a perfect

matching of Q3. Now, suppose that M has exactly two

edges. Figure 1.(a) shows all the non-isomorphic cases

thatM contains exactly two edges in the hypercube Q3.

Figure 1.(b) shows the perfect matchings of Q3. As a

result, the proof of this lemma is complete.

Theorem 1 Assume that n 1 is an integer. Let M

be an arbitrary matching with at mostn−1 edges in the

(a)

(b)

Figure 1: Perfect matchings of the hypercube Q3.

hypercubeQn. Then, the matchingM can be extended to

a perfect matching ofQn.

Proof. We prove it by induction on n. Suppose that

n = 1,2, it is clear that for any matching with at most

n−1 edges, it can be extended to a perfect matching of

Qn. Suppose thatn= 3, by Lemma 1, for any matching

with at most two edges, it can be extended to a perfect

matching ofQ3.

Assume that the theorem is true forn, which means that

for any arbitrary matchingM with at mostn−1 edges,

M can be extended to a perfect matching of Qn. Now,

we shall show that in the hypercube Qn+1, let M0 be

an arbitrary matching with m n edges in Qn+1, the

matching M0 can be extended to a perfect matching of

Qn+1.

By the definition of hypercubes, for each edge in Qn+1,

the labels of two endpoint vertices are different with

exactly one bit. Since M0 contains at most n edges,

there exists one dimensioni∈1,2,· · ·, n+ 1 inQn+1 =

Qi,0

n

L Qi,1

n such that each edge of M0 is distributed in

Qi,0

n or Qi,n1. Let M0 =M0∩Qni,0 andM1 =M0∩Qi,n1.

That is to say that M0 has m

0 = |M0| edges in Qi,n0,

m1 =|M1| edges in Qni,1, andm0+m1 = m. We may

without loss of generality assume that m0 m1. We

divide the proof into the following two cases.

Case 1: m0≥m1>0.

Notice that m0 < n and m1 < n since m0 m1 > 0.

See Figure 2.(a). By inductive hypothesis, M0 can be

extended to a perfect matching of Qi,0

n , say P M0, and

M1 can be extended to a perfect matching ofQi,n1, say

P M1. See Figure 2.(b). Therefore, P M0 ∪P M1 is a

perfect matching ofQn+1.

Case 2: m1= 0.

That is, all the n edges of M0 are distributed in Qi,0

n .

Suppose that m0 < n, by inductive hypothesis, M0

can be extended to a perfect matching of Qi,0

n , and so

M0 can be extended to a perfect matching of Q

n+1.

(3)

Q

n

i,0 Q

n i,1

Q

n+1

: elements of M` Q

n

i,0 Q

n i,1

Q

n+1

(a) (b)

{

m0

e

d

g

es

{

m1

e

d

g

es

{

{

{

2

n

-1

-m0

e

d

g

es

{

2

n

-1

-m1

e

d

g

e

s

m1

e

d

g

es

m0

e

d

g

es

Figure 2: Case 1: m0≥m1>0.

: elements of M`

(a) Qni,0 Qni,1

Qn+1

{

{

Qni,0 Qni,1 Qn+1

{

n-1

e

d

g

es

{

{

2

n

-1-2

e

d

g

e

s

(b)

2

n

-1-(

n

+

1

)

e

d

g

e

s

a

b d c

a

b d c

a'

b' d' c'

n

-1

e

d

g

es

2

n

-1-(

n

+

1

)

e

d

g

e

s

{

2

n

-1-2

e

d

g

e

s

a'

b' d' c'

Figure 3: Case 2: m1= 0.

Let (a, b) be an edge in M0. By inductive

hypothe-sis, M0 \ {(a, b)} can be extended to a perfect

match-ing of Qi,0

n , say P M0, since m−1 n−1. If P M0

contains the edge (a, b), it is clear that M0 can be

ex-tended to a perfect matching of Qn+1. Here, we may

consider that P M0 do not contain the edge (a, b). Let

(a, c) and (b, d) be two edges ofP M0. By inductive

hy-pothesis withn≥3,{(a0, c0),(b0, d0)}can be extended to

a perfect matching of Qi,1

n , say P M1. See Figure 3.(a).

Therefore, (P M0∪P M1∪ {(a, b),(a0, b0),(c, c0),(d, d0)})\

{(a, c),(b, d),(a0, c0),(b0, d0)} is a perfect matching of

Qn+1 as Figure 3.(b), and this theorem follows.

Lemma 2 For the 3-dimensional hypercubeQ3, letM be

an arbitrary matching with exactly three edges in the

hy-percube Q3. Then, the matching M can be extended to a

perfect matching ofQ3, exceptM is a forbidden matching

of Q3.

Proof. Figure 4.(a) shows all the non-isomorphic cases

that M contains exactly three edges in the hypercube

Q3. Figure 4.(b) shows the perfect matchings ofQ3with

the matchingM. As a result, the proof of this lemma is

complete.

Lemma 3 For the 3-dimensional hypercubeQ3, letM be

an arbitrary matching with at most one edge in the

hy-percube Q3, andx, y be any two unmatched vertices with

(a)

(b)

a forbidden matching

Figure 4: Perfect matchings of the hypercube Q3.

(a)

(b)

x

y

x y x

y

y x

x

y

x y x

y

y x

Figure 5: Perfect matchings of the hypercubeQ3\ {x, y}.

dist(x, y)is odd. Then, the matchingM can be extended

to a perfect matching ofQ3\ {x, y}.

Proof. Firstly, suppose that|M|= 0, it is clear that for

any two unmatched vertices x, y with dist(x, y) is odd,

there is a perfect matching ofQ3\ {x, y}. Now, suppose

that |M|= 1. Figure 5.(a) shows all the non-isomorphic

cases thatM contains exactly one edge andx, y be any

two unmatched vertices withdist(x, y) is odd in the

hy-percubeQ3. Figure 5.(b) shows the perfect matchings of

Q3\ {x, y} with the matching M. Hence, the proof of

this lemma is complete.

Lemma 4 Assume that n 3 is an integer. Let M be

an arbitrary matching with at mostn−2 edges in the

hy-percubeQn, and x, y be any two unmatched vertices with

dist(x, y)is odd. Then, the matchingM can be extended

to a perfect matching ofQn\ {x, y}.

Proof. We prove it by induction on n. Suppose that

n = 3, by Lemma 3, for any matching with at most

one edge and x, y be any two unmatched vertices with

dist(x, y) is odd, it can be extended to a perfect

match-ing ofQ3\ {x, y}.

Assume that the theorem is true forn, which means that

for any arbitrary matching M with at most n−2 edges

(4)

odd, M can be extended to a perfect matching of Qn\

{x, y}. Now, we shall show that in the hypercubeQn+1,

let M0 be an arbitrary matching with m n1 edges

in Qn+1 and x, y be any two unmatched vertices with

dist(x, y) is odd, the matching M0 can be extended to a

perfect matching ofQn+1\ {x, y}.

By the definition of hypercubes, for each edge in Qn+1,

the labels of two endpoint vertices are different with

ex-actly one bit. Since M0 contains at most n1 edges,

there exists one dimension i∈1,2,· · ·, n+ 1 of Qn+1 =

Qi,0

n

L Qi,1

n such that each edge of M0 is distributed in

Qi,0

n or Qi,n1. Let M0 =M0∩Qni,0 andM1 =M0∩Qi,n1.

That is to say that M0 has m

0 = |M0| edges in Qi,n0,

m1 =|M1| edges in Qni,1, andm0+m1 = m. We may

without loss of generality assume thatm0 ≥m1 and

di-vide the proof into the following three cases.

Case 1: Both xand y are in Qi,1

n .

By Theorem 1,M0can be extended to a perfect matching

ofQi,0

n , sayP M0, sincem0≤m≤n−1. Sincem0≥m1

andm≤n−1,m1≤n−2. By inductive hypothesis,M1

can be extended to a perfect matching of Qi,1

n \ {x, y},

sayP M0

1. Therefore, P M0∪P M10 is a perfect matching

ofQn+1\ {x, y}.

Case 2: Both xand y are in Qi,0

n .

Case 2.1: m0≤n−2.

By inductive hypothesis,M0can be extended to a perfect

matching ofQi,0

n \ {x, y}, sayP M00. By Theorem 1, M1

can be extended to a perfect matching ofQi,1

n , sayP M1.

Therefore, P M0

0∪P M1 is a perfect matching of Qn+1\

{x, y}.

Case 2.2: m0=n−1.

Let (a, b) be an edge of M0, by inductive hypothesis,

M0\ {(a, b)} can be extended to a perfect matching of

Qi,0

n \ {x, y}, say P M00, since m01 = n−2. Firstly,

suppose that (a, b) P M0

0. By Theorem 1, Qi,n1 has

a perfect matching, say P M1. Then, P M0

0∪P M1 is a

perfect matching of Qn+1 \ {x, y}. Now suppose that

(a, b) ∈/ P M0

0. We may let {(a, c),(b, d)} ⊆ P M00.

By Theorem 1, {(a0, c0),(b0, d0)} can be extended to a

perfect matching of Qi,1

n , say P M1, as Figure 6.(a).

Then, (P M0

0 P M1 ∪ {(a, b),(a0, b0),(c, c0),(d, d0)}) \

{(a, c),(b, d),(a0, c0),(b0, d0)} is a perfect matching of

Qn+1\ {x, y}as Figure 6.(b).

Case 3: x is inQi,0

n and y is inQi,n1.

Case 3.1: m0≤n−2.

InQn+1, sincen≥3, there exists an edge (c, c0) between

Qi,0

n andQi,n1such thatcis an unmatched vertex ofQi,n0,

c 6= x, dist(c, x) is odd, c0 is an unmatched vertex of

Qi,1

n , c06=y, anddist(c0, y) is odd. See Figure 7.(a). By

inductive hypothesis, M0 can be extended to a perfect

matching of Qi,0

n \ {x, c}, say P M00, and M1 can be

ex-tended to a perfect matching of Qi,1

n \ {y, c0}, sayP M10.

Therefore, P M0

0∪P M10 ∪ {(c, c0)} is a perfect matching

(a)

Qni,0 Q Qni,1 n+1 (b) Qni,0

Qni,1 Qn+1

{

n-2

ed

g

e

s

a

b

x

y c

d a'

b' c'

d'

{

2

n

-1-(

n

+

1

)

e

d

g

e

s

{

2

n

-1-

2

e

d

g

e

s

{

n-2 ed

g

e

s

a

b

x

y c

d a'

b' c'

d'

{

n-1

{

2

-

2

e

d

g

e

s

2

n

-1-(

n

+

1

)

e

d

g

e

s

Figure 6: Case 2.2: m0=n−1.

(a) (b)

Qni,0 Qni,1 Qn+1

Qni,0 Q

n i,1 Qn+1

{

m0

e

d

g

e

s

x y

c c'

{

m0

e

d

g

es

x y

{

2

n

-1-(

m0

+

1)

e

d

g

es

c c'

{

m1

ed

g

es

{

m1

ed

g

es

{

2

n

-1-(

m1

+

1)

e

d

g

es

Figure 7: Case 3.1: m0≤n−2.

ofQn+1\ {x, y}as Figure 7.(b).

Case 3.2: m0=n−1.

By Theorem 1,M0can be extended to a perfect matching

of Qi,0

n , say P M0. Let (x, c) be an edge of P M0. By

inductive hypothesis,Qi,1

n \{c0, y}has a perfect matching,

say P M0

1, since dist(c0, y) is odd. Therefore, (P M0

P M0

1∪ {(c, c0)})\ {(c, x)}is a perfect matching ofQn+1\

{x, y}.

In the following theorem, we shall show that for any

ar-bitrary non-forbidden matching withm≤nedges in the

hypercubeQn,M can be extended to a perfect matching

ofQn, exceptM is a forbidden matching ofQn.

Theorem 2 Assume that n 1 is an integer. Let M

be an arbitrary matching with at most nedges in the

hy-percube Qn. Then, the matchingM can be extended to a

perfect matching ofQn, exceptM is a forbidden matching

of Qn.

Proof. We prove it by induction on n. Suppose that

n = 1,2, it is clear that for any matching with at most

nedges, it can be extended to a perfect matching ofQn.

Suppose thatn= 3, by Lemma 1 and 2, for any matching

with at most three edges, it can be extended to a perfect

matching ofQ3, except it is a forbidden matching of Q3.

Assume that the theorem is true forn, which means that

for any arbitrary matching M with at mostnedges, M

(5)

is a forbidden matching of Qn. Now, we shall show that

in the hypercubeQn+1, letM0 be an arbitrary matching

withm≤n+ 1 edges inQn+1, the matching M0 can be

extended to a perfect matching of Qn+1, exceptM0 is a

forbidden matching ofQn+1. By Theorem 1, M0 can be

extended to a perfect matching ofQn+1ifm≤n. Hence,

we may only consider that m = n+ 1 in the following

proof of this theorem.

By the definition of hypercubes, the labels of two

end-point vertices of each edge in Qn+1 is exactly different

with one bit. SinceM0 containsn+ 1 edges, there exists

one dimensioni∈ {1,2,· · ·, n+1}ofQn+1=Qi,n0

L Qi,1

n

such that at most one edge of M0 is distributed between

Qi,0

n andQi,n1. LetM0=M0∩Qni,0,M1=M0∩Qi,n1, and

Mc=M0\(M0∪M1). That is that there arem0=|M0|

edges inQi,0

n ,m1=|M1|edges inQni,1,mc=|Mc|= 0 or

1 edge between Qi,0

n and Qi,n1, andm0+m1+mc =m.

We may without loss of generality assume thatm0≥m1.

Hence,m1≤ b(n+ 1)/2c ≤n−1 for n≥3. We divide

the proof into the following three cases.

Case 1: m0=n+ 1.

Assume thatM0 ={(ai, bi)|i = 1,2,· · ·, n+ 1}. We let

M0

1 ={(a0i, b0i)|i = 1,2,· · ·, n+ 1} and P = {(c, c0)|c

(V(Qi,0

n )\{ai, bi|i= 1,2,· · ·, n+1})}. Then,M0∪M10∪P

is a perfect matching ofQn+1.

Case 2: m0=n.

Firstly, suppose that mc = 1. Assume that M0 =

{(ai, bi)|i = 1,2,· · ·, n}. We let M10 = {(a0i, b0i)|i =

1,2,· · ·, n} and P = {(c, c0)|c (V(Qi,0

n )\ {ai, bi|i =

1,2,· · ·, n})}. Then,M0∪M10∪P is a perfect matching

ofQn+1. Now, suppose thatm1= 1. We divide this case

into two subcases.

Case 2.1: M0 is a non-forbidden matching of Qi,n0.

By inductive hypothesis,M0can be extended to a perfect

matching of Qi,0

n , say P M0. By another, since m1 = 1

and n 3, M1 can be extended to a perfect matching

of Qi,1

n , say P M1. Therefore, P M0∪P M1 is a perfect

matching ofQn+1.

Case 2.2: M0 is a forbidden matching of Qi,n0.

Let M0 = {(a1, b1),(a2, b2),· · ·,(an, bn)} and c be the

common neighbor ofa1, a2,· · ·, an in Qi,n0. Sincem1= 1

andn≥3, among the (a1, b1),(a2, b2),· · ·,(an, bn), there

exists a matching element, say (an, bn), and a

neigh-bor unmatched vertex of bn, say d, in Qi,n0 such that

d0 is an unmatched vertex. Here, it is clear that c0

is also an unmatched vertex, otherwise M is a

forbid-den matching of Qn+1. By inductive hypothesis, since

(M0∪ {(bn, d)})\ {(an, bn)} is a non-forbidden

match-ing of Qi,0

n with n edges, it can be extended to a

per-fect matching of Qi,0

n , say P M0. By Lemma 4, M1

can be extended to a perfect matching of Qi,1

n \ {c0, d0},

say P M0

1. See Figure 8.(a). Therefore, (P M0∪P M10

{(an, bn),(c, c0),(d, d0)})\{(an, c),(bn, d)}forms a perfect

Q

n

i,0 Q

n i,1

Qn+1

{

(a) 2 n -1 -( n + 1 ) e d g e s

{

2 n -1-2 e dg es

{

n edges

c d a n bn a 1 b1 a 2 b2 c' d' Qn

i,0 Q

n i,1 Q n+1

{

(b) 2 n -1 -( n + 1 ) e d ge s

{

2 n -1-2 e d g es

{

n e dges

c d an bn a1 b1 a2 b2 c' d'

Figure 8: Case 2.2: M0 is a forbidden matching ofQi,n0.

(a)

Qni,0 Q

n i,1 Qn+1

{

m0

e d g e s

{

{

m1

e d g e s

{

2 n -1-( m1 + 1 ) e d g e s 2 n -1-( m0 + 1 ) e d g e s c an c' a'n (b)

Qni,0

Qni,1 Qn+1

{

m0

e d g e s

{

{

m1 e d g e s

{

2 n -1-( m1 + 1 ) e d g e s 2 n -1-( m0 + 1 ) e d g e s c an c' a'n

Figure 9: Case 3.1: an anda0n are unmatched vertices.

matching ofQn+1as Figure 8.(b).

Case 3: m0≤n−1.

Firstly, suppose that mc = 0. M0 can be extended to

a perfect matching of Qi,0

n , sayP M0. By another, since

m1 ≤n−1 for n≥3,M1 can be extended to a perfect

matching ofQi,1

n , sayP M1. Therefore,P M0∪P M1 is a

perfect matching of Qn+1. Now, suppose that mc = 1.

Let Mc = {(c, c0)} where c is in Qi,n0. Then, in Qi,n0,

among thenneighbors ofc, denoted bya1, a2,· · ·, an, if

there exists at least an unmatched vertex, say an, such

that M0∪ {(c, an)} is a non-forbidden matching ofQi,n0

anda0

n is also an unmatched vertex as Figure 9.(a), this

shall be discussed in the following Case 3.1. Otherwise,

ai or a0i is matched in Qn+1 fori= 1,2,· · ·, nas Figure

10.(a), this shall be discussed in Case 3.2.

Case 3.1: an and a0n are unmatched vertices.

Then, by inductive hypothesis,M0∪ {(c, an)} can be

ex-tended to a perfect matching of Qi,0

n , say P M0. Since

m0 m1 and mc = 1, m1 ≤ b(n+1)21c ≤ n−2 for

n≥3. Hence,M1∪ {(c0, a0n)}can be extended to a

per-fect matching ofQi,1

n , sayP M1. See Figure 9.(a).

There-fore, (P M0∪P M1∪ {(c, c0),(an, an0)})\ {(c, an),(c0, a0n)}

is a perfect matching ofQn+1 as Figure 9.(b).

Case 3.2: ai or a0i is matched in Qn+1 for i =

1,2,· · ·, n.

Since m0 m1, m0 n 1, and m0 +

m1 = n, we may without loss of generality

as-sume that M0 = {(a1, b1),(a2, b2),· · ·,(ai, bi)} and

M1 = {(a0i+1, d0i+1),(a0i+2, d0i+2),· · ·,(a0n, d0n)} where i

(6)

(a)

Qn

i,0 Q

n i,1

Qn+1

(b) Qn

i,0

Qn i,1

Qn+1

c

g a1

b1

a2

b2 bi

ai

c' a 'n

{

2

n

-1-(

n

+1

)

ed

g

e

s

g'

{

2

n

-1-(

n

-i

)-1

e

d

g

e

s

c

g a1

b1

a2

b2 bi

ai

c' a 'n

{

2

n

-1-(

n

+

1

)

ed

g

es

g'

{

2

n

-1-(

n

-i

)-1

e

d

g

e

s

a'i+1

a'i+2 a'i+2 a'i+1

d 'n d 'i+2 d'i+1

an

bn

bi+1

ai+1 an

bn

bi+1

ai+1

d'n d'i+2 d'i+1

Figure 10: Case 3.2: ai or a0i is matched in Qn+1 for

i= 1,2,· · ·, n.

be a forbidden matching of Qi,0

n , note the c is

the common neighbor of a1, a2,· · ·, an. Then,

{(c, a1),(a2, b2),(a3, b3),· · ·,(an, bn)} is a non-forbidden

matching ofQi,0

n withnedges, and it can be extended to

a perfect matching ofQi,0

n , sayP M0. Let (b1, g)∈P M0.

On account ofdist(c, g) = 3 and inQi,1

n , the distance ofc0

and each of the matched vertices ofQi,1

n is at most 2, sog0

is unmatched. By Lemma 4,M1can be extended to a

per-fect matching ofQi,1

n \{c0, g0}, sayP M10, sincem1≤n−2

and dist(c0, g0) is odd. See Figure 10.(a). Therefore,

(P M0∪P M10∪ {(a1, b1),(c, c0),(g, g0)})\ {(a1, c),(b1, g)}

is a perfect matching ofQn+1. See Figure 10.(b).

It has shown in [7] that every perfect matching of then

-dimensional hypercube withn≥2 can be extended to a

Hamiltonian cycle. Consequently, we have the following corollary.

Corollary 1 Assume that n 2 is an integer. Let M

be an arbitrary matching with at most nedges in the

hy-percube Qn. Then, the matchingM can be extended to a

Hamiltonian cycle ofQn, exceptM is a forbidden

match-ing of Qn.

3

Conclusion Remarks

Perfect matchings can be applied to network disclosure attacks, such as the Perfect Matching Disclosure Attack. Perfect matching problem is also applicable to connec-tivity of networks. In this paper, we assign an arbitrary

non-forbidden matching with at mostn edges to form a

perfect matching or a Hamiltonian cycle on the

hyper-cubeQn forn≥2.

An open problem is that on the hypercubeQn, if we

as-sign a matching with more thannedges, how can we

re-strict the prescribed matching to obtain a perfect match-ing or a Hamiltonian cycle.

References

[1] Bhuyan, L. and Agrawal, D.P., “Generalized hyper-cube and hyperbus structures for a computer

net-work,” IEEE Transactions on Computers 33, pp.

323-333 (1984).

[2] Bondy, J.A. and Murty, U.S.R., Graph theory with applications, North Holland, New York, (1980).

[3] Brigham, R.C., Harary, F., Violin, E.C., and Yellen,

J., “Perfect-matching preclusion,” Congressus

Nu-merantium174, pp. 185-192 (2005).

[4] Chen, Y.C., Huang, Y.Z., Hsu, L.H., and Tan, Jimmy J.M., “A family of Hamiltonian and

Hamil-tonian connected graphs with fault tolerance,” The

Journal of Supercomputing, DOI

10.1007/s11227-009-0316-3 (2009).

[5] Chen, Y.C., Tsai, C.H., Hsu, L.H., and Tan, Jimmy J.M., “On some super fault-tolerant

Hamil-tonian graphs,”Applied Mathematics and

Computa-tion148, pp. 729-741 (2004).

[6] Fang, J.F., “The bipancycle-connectivity of the

hy-percube,” Information Sciences 178, pp. 4679-4687

(2008).

[7] Fink, J., “Perfect matchings extend to Hamilton

cy-cles in hypercubes,”Journal of Combinatorial

The-ory Series B97, pp. 1074-1076 (2007).

[8] Kreweras, G., “Matchings and Hamiltonian cycles

on hypercubes,” Bulletin of the Institute of

Combi-natorics and its Applications16, pp. 87-91 (1996).

[9] Leu, Y.R. and Kuo, S.Y., “Distributed fault-tolerant ring embedding and reconfiguration in hypercubes,”

IEEE Transactions on Computers 48, pp. 81-88

(1999).

[10] Park, J.H. and Son, S.H., “Conditional matching preclusion for hypercube-like interconnection

net-works,”Theoretical Computer Science410, pp.

2632-2640 (2009).

[11] Pemmaraju, S. and Skiena, S., Implementing dis-crete mathematics: Combinatorics and graph theory with mathematica, Addison-Wesley, pp. 178, (1990).

[12] Troncoso, C., Gierlichs, B., Preneel, B., and Ver-bauwhede, I., “Perfect matching disclosure attacks,” Privacy Enhancing Technologies - 8th International

Symposium, PETS 20085134, pp. 2-23 (2008).

[13] Tsai, C.H., “Fault-free cycles passing through

pre-scribed paths in hypercubes with faulty edges,”

Ap-plied Mathematics Letters22, pp. 852-855 (2009).

[14] Xiang, D., Zhang, Y., and Sun, J.G., “Unicast-based fault-tolerant multicasting in wormhole-routed

hy-percubes,” Journal of Systems Architecture 54, pp.

References

Related documents

Understanding the gap between research and practice: Chemistry Understanding the gap between research and practice: Chemistry faculty's awareness and reported implementation

• Researchers are advised to ensure that they follow best practice guidance for the correct management of their records in order to ensure compliance against both the Data

The paper argues that there is a breakup of community social structures as evident from changes in family relations in critical domains such as extra-marital relations, sexual

Recent work suggests that learning-related emotions (LREs) play a crucial role in performance especially in the first year of university, a period of transition for

We use dis-aggregated unemployment rate data corresponding to six demographic groups: White Total, White Males, White Females, Black Total, Black Males, and Black Females.. To

To condemn the embargo is not to declare love of Fidel Castro, or to declare support for Communism, or to endorse every action of the Cuban Government.. The member states of the

Performance of soil moisture dynamics from the SWBM dataset (best parameters) and all reference datasets (ERA-Land, MERRA-Land, GLDAS-2-Noah, CLM4), assessed against several

With few scores given for other haz- ard mitigation items that could be or were implemented that applied to drought (scores 1 and 3 in the rubric (Table 5), the stand-alone