Decomposing Graphs into Forests of Paths

with Size Less Than Three

Bor-Liang Chen+, Hung-Lin and Kuo-Ching Huang*
Department of Applied Mathematics
National Chiao Tung University,
Taiwan. Republic of China

Abstract

A forest in which every component is path is caned a path forest. A family

of path forests whose edge sets form a partition of the set of a graph G is called
a path of G. The minimum number of forests in a path
decomposition of a graph G is the path number of G and denoted by p( G). If we
restrict the number of edges in each path to be at most x then we obtain a special
decomposition. The minimum number of path forests in this of decomposition
is denoted In this paper we study We note here that if we
restrict the size to be one, the number the chromatic index of G.

In this paper, we study the special type of path and we obtain

the answers for P2( G) when G is a complete graph, a tree and some other graphs.

1. Introduction.

A path decomposition is a special case of an edge decomposition and is the

type of decomposition we will study in this paper. There are many interesting and
important results and problems in this area. A good survey of them is provided by

+ Research supported by National Science Council of the Republic of China

(NSC 78-D208-M009-D4).
* Research supported by National Science Council of the Republic of China
(NSC 78-D208-M009-20).

A~stralin Journal of Combinatorics 3(1991) pp 55-73

 other (the minimum
number of to decompose ,>r~"\t'H (the minimum
number of forests needed to decompose "r,,,,r,,'1TV (the minimum
number of to dec:OITJlpOi3e illll.!!lli2r (the
minimum number of trees needed to dec:OITJlpOi,e a
these numbers have been found, An
that each cornpcmeJnt
coloring

with c In this paper,

in 4 find and

maximum of

2, Lower bounds on

Given a graph G, it is not difficult to see that G U.~"'VLJ- into at

least forests with the of each than three. Thus we

have the following

~ re(G)/jv(G)l.

For some graphs, this estimate is quite good, but there are also some graphs

56
lOr whICh thIS estlmate IS pretty tar from the exact answer. We can use Proposition
2.1 to obtain bounds on P2( G) for two well-known classes of graphs: complete
graphs and complete bipartite

> > and

The above estimates of and very good. In Section 4,

we will for almost all the lower bounds in t'rcroosltJton 2.2 for P2(K n ) are

also upper So this answers for

Now let us look at G. It is clear that r6:.(G)/21-

This will I-.HVU',~ better vLJUHU''''"'G for the such as graphs,
trees, etc. Thus this with the ideas of the above propositions we
have

max where m (G) is the

2
size of maximum forests in in which the size of each is less than three.

If if there exists a path which joins two vertices x and y

with maximum and all the vertices on this path other than x and y have
degree ~ .6.(G)-l, then P2(G) > the reason for this can be obtained directly

from decomposing this subgraph. In what follows we will call such a path a critical
path.

Let G be a graph with .6.( G) being an even number. Then

57
3. of trees and nets.

A tree is connected

fl. Then

The

3.1.

58
4.

A proper edge-coloring of a graph is an assignment of colors to its edges so

that no two incident edges have the same color. If a graph G can colored by no
more than k colors, then this graph is called k-colorable and the number X / (G) =
min {k: G is k-colorable} is the chromatic index of G. As mentioned in Section 1,
X (G) is the minimum number of matchings required to edge-decompose a graph
f

G. Similarly, P2(G) can be considered to be the minimum number of colors

rc.rnl1-rClrl to color the G so that no proper connected supergraph of K1 2 is
,
induced by edges of one color; call such an edge-coloring a P2-coloring. For
eXcLmTJle, the numbers we put on Figure 3.1 are the colors. If we focus on
the P2-colorings of Kn and then we can use an n x n array to represent the
coloring. It is well-known that a K n,n with proper '-V~JLU.h can be represented by a
latin square of order n. if we consider a P2-coloring, it is slightly different
from a latin square. 4.1 is an of K6 ,6 with P2(K 6,6) 5. As we
have seen in thi s array, L [C. J']' a number occurs in each row and each column
1,

at most twice and furthermore if = £.1/ ,j. / , i * i / and j *j / , then f f.1,J. and

fi / ,j * The number can also be obtained in this way except that the
array L = is symmetric, is for each
= 1, 2, .. " n. Figure 4.2 is an vJ~"Ui-'P

1 1 2 2 3 3
r--
3 1 1 2 2

2 3 5 1 4

2 5 3 4 1 5
4 2 5 3 4 1

5 2 4 3 5 1

Figure 4.1.

59
 ~ 1 2 2 1

1 i~ 3 4
2 3 ~ 1 3
2 4 1 ~ 4
1 2 3 4 ~

We note here that the arrays in and Figure 4.2 provide an upper
bound on P2(K 6 ,6) and and that this upper bound equals the
lower bound in Proposition 2.2 and 2.3, rOQl'\o .... t,uahr

Based on the relationships between and Kn,n we have the following

results.

P2(K n ) + 1 ~P2(Kn ,n)'

(3) P2(Kn)~ + P2(K n ,n)·(4) P2(Kin,)~' (5)
2· P2(K n ) + 1.

Obviously, we (1). can be obtained by putting a number

(new) in the diagonal of the array which conreST>owdS to Since K2 n \K n,n is a
disjoint union of two Kn's, hence we have (3). (4) is a direct result of the direct

product of the array corresponding to Kn,n and a latin square of order i. Finally, by
Figure we obtain (5). Q.E.D.

60
 f
M

L' ar r ay for p,)-{:oi1onrlg of Kn based on {I, 2 , (Kn)}

M: array for of based on {P2{K n )+1,' " 2'P2(K n )}
f new color.
4.3.

These inequalities will help us in the number But we need

some
A Steiner a where S is a
v-set and t is a collection of 3---€lement triple) of S such that every
2---element subset of S occurs in of t. It is well-known that an
if v is a

3 (mod
can be partitioned into a resolvable
(briefly RSTS(v)) This is to the
fact that K 6k + 3 can be into 3k+ 1 collections of 2k+ 1 mutually

disjoint triangles. Thus we

9t +

Let (S,t) an RSTS(6k+3). we consider k 2t. Then there

are 6t+ 1 parallel classes. The idea of the is as follows: start with two
parallel classes; from each if we take one away from each triangle, then
we obtain a forest of the type we need, so is the other class. if we can
suitably combine the edges we took away, we obtain one more path forest of the
we want. If the above process can be done, then S 3, + 2. By

61
Proposition . (4t-l) - 1 1 9t + Hence

we have the nrn,nlOl'Yl left is how to combine the U,,",A'v\iUC'F,

Let two 1, .. " 4t+ 1} and

Construct G

... 4t+l,

}
start

We
it been shown

obtain the

9t+6.

62
 Proof. 9t+2 P2(K 12t + 3) ~ P2(K 12t + 2 ) 9t+2. 9t+6 P2(K 12t + 9 )

P2(K 12t + 8 ) P2(K 12t + 7) ~ 9t+6. Q.E.D.

We can use the construction of a (12t+6) x (12t+6) array to show the

following

Proposition Let L

and M be two arrays which correspond to -C()10n~s of which are based on

the colors {I, 2, ... , P2(K 6t + 3)} and {P2(K 6t + 3)+1,

2P2(K6t+3) }

Now if t is odd, then P2(K 12t + 6) 9t+4 and P2(K 12t + 6 ) ~

2'P2(K + )+ 1
6t 3
2.(3t+l~I) + 1 9t + 4 (proposition 4.1.). Hence we

have P2(K 12t + 6) 9t+4.

If t is even, then 3t+ I is odd. This implies that if the construction in

Proposition 4.2 is used, we have one match~ in the nAnArn'nAn, of K6t + 3 which
has 2t+ I Of course, as in the proof of 4.2 we can combine two
matching and M2 from two parallel classes together to obtain a path forest of

the we need. Let this path forest be denoted by al-br--cl' ,-",,---, l~'-

where {apb l }, {a2,b 2}, "', {a2t+l'b2t+l} are in the

matching MI and {bl,c]}, {b 2,c 2}, "', {b2t+l,C2t+l} are in the matching M2.
Now in Land M of Figure 4.3, we take the edges of Ml and out respectively, i.e.

the following cells of the array in Figure 4.3 are empty:

(ai ,\), (bi'ai ), (bi'6t+3+ci ), (c i ,bi +6t+3), (b i +6t+3,ci ), (ci +6t+3, bi)' (a i +6t+3,
bj +6t+3) and (b i +6t+3,ci +6t+3). i = 1, 2, "', 2t+1. By letting all of the these
8· cells be filled with a common number f, then we have an array which is

63
corespldi~ to Since the induced the correS1PoIll.OUlg to f
of need. Thus +
conclude

Now

the M 9t.
we

64
 1 1 2 3 8 5 4 6 9

8 4 4 5 5 6 6 7 9 3 1

6 8 9 3 1

6 9 1 1 8 7 5

1 8 4 5 5 6 6 7 9
6 7 7 8 9
M
8 7 4 6 9 1 1

9 2 1 8 4 5 6

2 1 6 7 7 8 8
- - --
3 7 4 6 9 1 1

5 6 6 7 2 4
7 8 9 9 1 7

4.5.

we for each 0,2, 6, 7, 8, 9, 11 We

will need another TtJ,'nn1,nllO when n 4,

Before we go any we need some more definitions. Let is a v--set.

A latin of order v based on S is array with entries such that in
each row and each column every element S occurs
let latin square L is to be ~nril!@. provided

that eij = for every 1 ~ i, j ~ v, A latin square L [eij] is ~= i for

each i E S. It is well-known that an idempotent commutative square of order

v exists if and if v is odd. In case v 2k, let H

65
 in II
latin square such that for
on h. Since all the holes

commutative latin

we

4.7.

B,

4.6.

66
 1 1 4 5 10 9 4 b c 15 16 12 13 3 18 17 11 11 14 14 10 2 8 6 7
b b 2 5 6 4 3 3 c 17 15 15 13 13 11 18 16 10 17 14 7 1 8 9
b c 3 3 6 7 5 2 2 12 12
16 13 17 17 11 18 15 15 10 14 1 9 4 8
b c 4 8 6 3 14 18
14 18 11 12 16 10 13 5 9 2 3
b b 5 5 9 16 14
10 11 16 10 12 18 17 18 13 3 3 4 6
6 6 9 1 10 14
17 11 15 18 16 12 13 16 7 7 8
7 1 6 8 2 15 18 17 11 18 12 10 14 16
6 b b 8 8 2 9 12 7 17 16 15 11 18 18 12 14
4 6 a b c 8 8 1 14 3 15 13 16 13 18 16 11 12 10 17 2
3 8 b c a a 1 9 5 10 16 15 10 18 15 17 17 11 13
b 3 9 7 8 c 18 17 6 4 10 17 13 16 1 5
c c 7 1 9 a b 7 4 4 5 14 14 10 15 5 11 15
17 16 10 6 2 8 18 1 1 4 11 15 13 10
16 15 18 14 14 9 1 8 11 b 16 11 17 13
12 12 16 14 10 17 8 9 1 9 18 2 13 15 17
15 13 11 2 12 14 5 17 3 10 16 10 9
3 13 17 16 15 7 3 10 6 7 14 14 18 18
18 13 17 11 10 18 8 14 15 16 4 1 12 16 12
171111 12 18 17 13 15 10 7 2 6
11 18 15 13 13 16 10 17 14 4 8 8 1 2 7
11 16 16 18 17 3 13 14 14 10 7 b c c 9 8 1
14 10 15 17 17 11 16 18 18 12 16 8 8 7 b c a a 1 2 9 4
14 17 10 10 18 11 15 12 15 12 6 9 7 8 c a b 4 4 5 1
10 16 18 11 11 15 12 17 7 1 1 9 a b 6 6 5 4
2 7 1 5 12 18 12 17 13 15 11 16 11 10 9 2 1 4 6 a a b
8 1 9 3 7 10 18 10 17 16 5 15 11 13 16 14 2 1 8 2 4 6 a b c
6 8 4 14 12 17 11 111 15 10 18 16 8 3 9 5 5 b c
7 9 6 16 14 2 13 5 15 10 13 17 9 18 12 6 1 4 1 4 b c c

4.7.

If t is let L = [£..J be an Id€:mI)ot,ent commutative latin square of order

1J
t. Now using and defined in construct a (12t+4)x(12t+1) array
as in 4.8 where if A (i,j) =
x
(x-I).9+A 1(i,j) if AI(i,j)~{abc} .g and ex =M+ (x-I)·9, 1 ~ x
~ t. It is a routine matter to check that 9t+3 whenever t is odd.

67
 Al B1

A2

B.1
lj

A.
J
U

T B
'1 '2 '3 1 J 't,

1.8.

If t is even, let L [£..J be

1J
commutative square of order t = with
2x2 holes. Then

and Blare now defined 4.7. we conclude that

t44. Since the of 9t+1 = 9t-+-8 is

we omit it here and put the arrays which {'()rrOQn()rlfl to

K2~) :)
= 19, 17 and = 26 in the

68
 A B1

ij
B2

B3

T
1]

ilk

D
'1 '2 )3
~h

thE' fact commutative square of order 4 with holes doesn't

we are not to obtain and now, But, we

believe this is of tedious work in constructions and not

it

'-JUH"~ the above results we have the

and

except possibly if 3v+ 1 E

69
 also obtain the on
direct constructions. Since the techIll!QU<es we will not go any further in
this direction. As when 9t which can be
obtained latin square of order t.
result is to the eXlstelrlCe ~-1act,onz.lO of balanced r>Arn .... 'A...

with 12t vertices shown

On n",ht·r.~ of

11

Solution of Kirkman's
RUI-''-'I.'''LUU on Mathematics

Discrete Math.

361-366.
K. Ushio and ..... «''-''-A ... ,,,,,,.,010 of cornplete multipartite

Graphs and Combinatorics 385-387.

70
'~Y

1 1 4 4 3 7 6 6 a 8 9
1 2 2 5 3 6 a 7 7 9 8 4
1 2 3 3 4 6 7 9 8 8 5
4 2 3 1 1 a 7 8 9 6 6 5
4 5 3 5 9 8 6 7 9 2
3 4 5 8 8 a 6 9 7 2
7 6 6 a a 8 2 2 4 4 5 9
6 7 7 9 8 2 1 3 5 4
6 7 9 8 8 2 1 5 4 3
7 8 6 6 4 2 2 3
a 9 8 6 7 9 4 3 5 1
8 8 a 6 9 7 5 5 4 2 3
9 4 5 5 2 2 9 4 3 1 1

P2(K 13 ) 10

114437669 11 11 10 10 13 13 18 17 15 14 16 19 5
1 3 6 9 7 10 8 18 14 14 15 16 16 17 19 13 12 11 4
1 2 3 4 6 7 10 8 9 5 16 17 17 18 18 19 19 15 13 12 11 14
423 1 1 9 7 8 10 6 6 5 16 19 11 11 12 12 13 13 18 17 15 14
453 1 9 10 8 7 10 13 12 11 18 14 14 15 15 16 16 17 19 2
3 1 5 8 8 9 6 10 7 12 11 14 16 17 17 18 18 19 19 15 2
766998 2 2 4 5 18 17 15 14 16 19 11 11 12 12 13 10
6 9 7 7 10 8 2 1 135 16 17 19 13 12 11 18 14 14 15 15 16 4
6 7 10 8 8 9 2 1 554 19 19 15 13 1211141617171818 3
9 7 8 10 6 6 1 2 2 13 13 18 17 15 14 16 19 11 11 3
9 10 8 6 7 10 3 5 2 3 15 15 16 16 17 19 13 12 11 18 14 14 1
8 8 9 6 10 7 5 5 4 3 17 18 18 19 19 15 13 12 11 14 16 17 1
11 18 5 5 13 13 18 16 19 12 15 17 6 6 9 9 8 2 1 1 4 4 3 10
11 14 16 16 12 12 17 17 19 13 15 18 7 7 10 8 1 4 2 2 5 3 9
10 14 17 19 11 11 15 19 15 13 16 18 6 7 8 8 9 1 5 3 3 4 12
10 15 17 11 18 14 14 13 13 18 16 19 9 7 8 6 6 4 3 5 1 1 12
13 15 18 11 14 16 16 12 12 17 17 19 9 10 8 6 10 4 5 3 1 2 5 7
13 16 18 12 14 17 19 11 11 15 19 15 8 8 9 6 10 3 3 4 1 5 2 7
18 16 19 12 15 17 11 18 14 14 13 13 2 1 1 4 4 3 7 7 9 9 10 5
17 17 19 13 15 18 11 14 16 16 12 12 1 4 2 2 5 3 7 6 6 8 10 9
15 19 15 13 16 18 12 14 17 19 11 11 1 2 5 3 3 4 7 6 10 10 9 8
14 13 13 18 16 19 12 15 17 11 18 14 4 3 5 1 $) 6 10 7 7 8
16 12 12 17 17 19 13 15 18 11 14 16 4 5 3 1 2 5$)8 10 7 8 6
19 11 11 15 19 15 13 16 18 12 14 17 3 3 4 1 5 2 10 10 9 7 8 6
5 4 14 14 2 2 10 4 3 3 1 2 10 9 12 12 7 7 5 9 8 8 6 6

71
 11 h e g b a a d d c 8 7 546 9 1
11 g h g h a c b b f c 6 2 8 5
II g 12 h e 532 7
e h 12 f g d b 8 i1 2
g h 13 rl c c a b b 5 5 6 6
g f g 13 c d 7 8 8 9
b a a d d c e f g h 11 11 12 12
a e b b e c d f g h 18 14 14 15
rt b e c c d e rl 2 h f 14 17 17
d b c e a f 2 h g 14 16 10 11
d f a beg g h h 3 13 12 11 18
(' c d a lJ e h h 12 11 14
8 6 9 2 5 11 18 14 14 1 1 e
9 3 .5 !3 11 14 16 16 12 12 g h g a e b b f
5 9 5 681 h a b e c c d
3 3 8 6 9 12 15 17 11 18 14 e 2 f d b c e a
6 2 2 7 9 3 15 18 11 J l~ f g h f 3 d e c b
'-' 9 9 131618]2 g 3
( .....
8 4 i d 11 e g
() 2 171710 11 ]S a e b b e c 1] d g h 14 16 16
9 l 1S llj IS 1 j
8 .', 1:4 13 1 ~ 1R
B b 16 12 12
4
6 8 2 4 10 1111 IS
16 1710 13 12 11 6 7 9 :1
iO 10 15 J 1 I 11 9 q 5
12 13 13 18 J7 IS
I') 15 J 7 i ') S I) ()

4 5 e
4 16 17 17 18 18 ~ 6 7 b b a
4 16 10 11 I! 12 h h 9 b
") 1 11 18 I', ::.
a
12 11 114 1;- 2.
c c

26