Discrete Mathematics 290 (2005) 105 – 108

www.elsevier.com/locate/disc

Note
A note on internally disjoint alternating paths in
bipartite graphs夡
Dingjun Loua , Akira Saitob , Lihua Tenga
a Department of Computer Science, Zhongshan University, Guangzhou 510275, PR of China
b Department of Computer Science, Nihon University, Sakurajosui 3-25-40, Setagaya-Ku,
Tokyo 156-8550, Japan

Received 3 April 2002; accepted 21 October 2004

Available online 21 December 2004

Abstract
Let G be a balanced bipartite graph with partite sets X and Y, which has a perfect matching, and
let x ∈ X and y ∈ Y . Let k be a positive integer. Then we prove that if G has k internally disjoint
alternating paths between x and y with respect to some perfect matching, then G has k internally
disjoint alternating paths between x and y with respect to every perfect matching.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Matching; Alternating path

For graph-theoretic terminology not defined in this note, we refer the reader to [2]. In
this note, a path which starts from a vertex x and ends at a vertex y is called an xy-path.
For a matching M of a graph G, a trail T = a0 a1 . . . al is said to be an alternating trail
with respect to M if a2i−1 a2i ∈ M for each i with 1
i
 21 l. Note that this implies
a2j a2j +1 ∈ E(G) − M (0
j
 l−12 ). If T is a path (resp. cycle), we call T an alternating
path (resp. alternating cycle). Note that by the definition, if P is an alternating xy-path of
even length, the same path traversed in the opposite direction is not an alternating yx-path
since the first edge belongs to the perfect matching.

夡 Research supported in part by Japan Society for the Promotion of Science, Grant-in-Aid for Scientific Research

(C) 13640138.
E-mail address: [email protected] (A. Saito).

0012-365X/$ - see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.disc.2004.10.019
106 D. Lou et al. / Discrete Mathematics 290 (2005) 105 – 108

For a nonnegative integer n and a graph G of order at least 2n + 2, G is said to be n-

extendable if G has a perfect matching and every set of n independent edges extends to
a perfect matching. The reader can be referred to [3]. In [1], Aldred et al. have found the
following Menger-type relationship between the extendability and the number of internally
disjoint alternating paths between a pair of vertices in a bipartite graph.

Theorem A (Aldred et al. [1]). Let G be a balanced bipartite graph with partite sets X
and Y which has a perfect matching, and let k be a positive integer. Then G is k-extendable
if and only if for every perfect matching M and every pair of vertices x ∈ X and y ∈ Y ,
there exist k internally disjoint xy-paths with respect to M in G.

The sufficiency of Theorem A requires us to check every perfect matching. However, the
following theorem claims that we have only to check just one perfect matching which is
arbitrarily chosen, since it forces the existence of k internally disjoint alternating paths for
all the other perfect matchings.

Theorem 1. Let G be a balanced bipartite graph with partite sets X and Y, which has a
perfect matching, and let x ∈ X and y ∈ Y . Let M0 and M be perfect matchings of G. If
G has k internally disjoint alternating xy-paths with respect to M0 , then G has k internally
disjoint alternating xy-paths with respect to M.

For a vertex x in a graph G, we denote by NG (x) the neighborhood of x in G. In this

note, we deal with a matching as a set of edges. For example, if an edge e belongs to
a matching M, we write e ∈ M. Let T = v0 v1 . . . vl be a trail. For indices i and j with
−
→
0
i < j
l, the subtrail vi vi+1 . . . vj is denoted by vi T vj . The same subtrail, traversed
←−
in the opposite direction, is denoted by vj T vi . For two sets A and B, A$B denotes the
symmetric difference of A and B.

Proof of Theorem 1. Let P1 ,
. . . , Pk be internally disjoint alternating xy-paths with re-
spect to M0 . Let H = (V (G), ki=1 E(Pi )). Then for v ∈ V (H ) = V (G), we have


0 / ki=1 V (Pi )
if v ∈


degH v = 2 if v ∈ ki=1 V (Pi ) − {x, y}

k if v = x or v = y.

Let K = (V (G), E(H )$M0 $M). First, we investigate the degree of the vertices in K. 


Claim 1. For each v ∈ V (G) − ki=1 V (Pi ), degK v = 0 or degK v = 2. Furthermore, if
degK v = 2, then exactly one of the two edges of K incident with v belongs to M.

Proof. Since both M0 and M are perfect
matchings, vv 0 ∈ M0 and vv ∈ M for some

v0 , v ∈ NG (v). Furthermore, since v ∈ / ki=1 V (Pi ), no edge of H is incident with v. If
v = v0 , then vv = vv 0 ∈ (M0 ∩ M) − E(H ), which implies vv ∈ / E(K) and degK v = 0.
If v = v0 , then vv 0 ∈ M0 − (M ∪ E(H )) and vv ∈ M − (M0 ∪ E(H )). Hence we have
NK (v) = {v0 , v } and {vv 0 , vv } ∩ M = {vv }. 
 D. Lou et al. / Discrete Mathematics 290 (2005) 105 – 108 107

k
Claim 2. For each v ∈ i=1 V (Pi ) − {x, y}, degK v = 0 or degK v = 2. Further-
more, if degK v = 2, then exactly one of the two edges of K incident with v belongs
to M.

Proof. Let v ∈ V (H ) − {x, y}. Then degH v = 2. Let NH (v) = {v1 , v2 }. Since each Pi is
an alternating path with respect to M0 (1
i
k), we may assume vv 1 ∈ E(H ) ∩ M0 and
vv 2 ∈ E(H ) − M0 .
Let vv 3 ∈ M. If v3 =v1 , then we have vv 1 ∈ E(H )∩M0 ∩M and vv 2 ∈ E(H )−(M0 ∪M).
Hence NK (v) = {v1 , v2 } and vv 1 ∈ M. If v3 = v2 , then vv 1 ∈ (E(H ) ∩ M0 ) − M and
vv 2 ∈ (E(H ) − M0 ) ∩ M. These imply vv 1 , vv 2 ∈ / E(K) and degK v = 0. Finally, if
v3 = v1 and v3 = v2 , then vv 1 ∈ (E(H ) ∩ M0 ) − M, vv 2 ∈ E(H ) − (M0 ∪ M) and
vv 3 ∈ M − (M0 ∪ E(H )). These imply NK (v) = {v2 , v3 } and vv 3 ∈ M. 

Claim 3. We have degK x = k or degK x = k + 2. If degK x = k, then none of the k edges

in K incident with x belong to M. If degK x = k + 2, then exactly one of the k + 2 edges in
K incident with x belongs to M.

Proof. Let NH (x) = {x1 , . . . , xk } and let xx 0 ∈ M0 . By the definition of an alternating

/ {x1 , . . . , xk }. Hence xx i ∈ E(H ) − M0 (1
i
k) and xx 0 ∈ M0 − E(H ).
path, x0 ∈
Let xx ∈ M. If x = x0 , then xx 0 ∈ (M0 − E(H )) ∩ M and xx i ∈ E(H ) − (M0 ∪
M) (1
i
k). Thus, we have NK (x) = {x1 , . . . , xk } and {xx 1 , . . . , xx k } ∩ M = ∅. Next,
suppose x ∈ {x1 , . . . , xk }. We may assume x = xk . Then xx 0 ∈ M0 − (E(H ) ∪ M),
xx i ∈ E(H ) − (M ∪ M0 ) (1
i
k − 1) and xx k ∈ (E(H ) − M0 ) ∩ M. These imply
NK (x) = {x0 , x1 , . . . , xk−1 } and {xx 0 , xx 1 , . . . , xx k−1 } ∩ M = ∅. If x ∈
/ {x0 , x1 , . . . , xk },
then xx 0 ∈ M0 −(E(H )∪M), xx i ∈ E(H )−(M0 ∪M) (1
i
k) and xx ∈ M −(E(H )∪
M0 ). Hence we have NK (x) = {x0 , x1 , . . . , xk , x } and {xx 0 , xx 1 , . . . , xx k , xx } ∩ M =
{xx }. 

By symmetry, we have the following.

Claim 4. We have degK y = k or degK y = k + 2. Moreover, if degK y = k, then none of the

k edges incident with y belong to M. If degK y = k + 2, then exactly one of the k + 2 edges
incident with y belongs to M.

Using the above claims, we prove the theorem. We consider two cases.
Case 1: degK x = k or degK y = k.
We may assume degK x = k. Let NK (x) = {x1 , . . . , xk }. Then by Claims 1 and 2, for
(i) (i) (i) (i) (i)
each i, 1
i
k, there exists a trail Ti = a0 a1 . . . ali in K with a0 = x, a1 = xi and
(i)
ali ∈ {x, y}. And again by Claims 1 and 2, Ti is an alternating trail with respect to M.
By taking such Ti so that it is as short as possible, we may assume that Ti is either a cycle
(i) (i) (i)
or an xy-path. If ali = x, then ali −1 ali ∈ / M by the assumption of the case and Claim 3.
Then since Ti is alternating, Ti is an odd cycle. This is a contradiction since G is a bipartite
graph. Therefore, each Ti is an alternating xy-path, and by Claims 1 and 2, T1 , . . . , Tk are
internally disjoint.
Case 2: degK x = degK y = k + 2.
108 D. Lou et al. / Discrete Mathematics 290 (2005) 105 – 108

Let NK (x) = {x0 , x1 , . . . , xk , x } and suppose {xx 0 , . . . , xx k , xx } ∩ M = {xx }. By the

same argument as in Case 1, for each i with 0
i
k, we can construct an alternating trail
(i) (i) (i) (i) (i) (i)
Ti = a0 a1 , . . . , ali with a0 = x, a1 = xi and ali ∈ {x, y}. Again by taking Ti so that
it is as short as possible, we may assume that each Ti is either a cycle or an xy-path and that
T0 , . . . , Tk are internally disjoint.
(i)
Suppose ali = x for some i, 0
i
k. We may assume i = k. Since G is a bipartite
(k) (i )
graph, lk is even and alk −1 alk k ∈ M. This forces alk −1 = x . Since T0 , . . . , Tk are internally
disjoint, we still have k internally disjoint alternating xy-paths T0 , . . . , Tk−1 with respect
to M. 

Acknowledgements

The work of this paper was supported by the National Science Foundation of China.
This research was done when A.S. visited Zhongshan University. He is grateful for the
hospitality extended during the stay.

References

[1] R.E.L. Aldred, D.A. Holton, D. Lou, A. Saito, M-alternating paths in n-extendable bipartite graphs, Discrete
Math. 269 (2003) 1–11.
[2] G. Chartrand, L. Lesniak, Graphs & Digraphs, third ed., Wadsworth & Brooks/Cole, Monterey, CA, 1996.
[3] M.D. Plummer, On n-extendable graphs, Discrete Math. 31 (1980) 201–210.