Discrete Applied Mathematics 162 (2014) 108–114

Contents lists available at ScienceDirect

Discrete Applied Mathematics

journal homepage: www.elsevier.com/locate/dam

Minimum degree, independence number and pseudo

[2, b]-factors in graphs
Siham Bekkai ∗
USTHB, Faculty of Mathematics, PO Box 32 El-Alia Bab Ezzouar, 16111 Algiers, Algeria

article info abstract

Article history: A pseudo [2, b]-factor of a graph G is a spanning subgraph in which each component C on
Received 13 April 2012 at least three vertices verifies 2 ≤ dC (x) ≤ b, for every vertex x in C . Given an integer
b(δ−1)
Received in revised form 18 August 2013 b ≥ 4, we show that a graph G with minimum degree δ , independence number α > 2
Accepted 24 September 2013
and without isolated vertices possesses a pseudo [2, b]-factor with at most α − 2b (δ − 1)
 
Available online 11 October 2013
components that are edges or vertices. This bound is sharp.
Keywords: © 2013 Elsevier B.V. All rights reserved.
Pseudo [2, b]-factor
Independence number
Minimum degree

1. Introduction

Throughout this paper, graphs are assumed to be finite and simple. For unexplained concepts and notations, the reader
could refer to [1].
Given a graph G, we let V (G) be its vertex set, E (G) its edge set and n its order. The neighborhood of a vertex x in G is
denoted by NG (x) and defined to be the set of vertices of G adjacent to x; the cardinality of this set is called the degree of x in
G. For convenience, we denote by d(x) the degree of a vertex x in G, by δ the minimum degree of G and by α its independence
number. However, if H is a subgraph of G then we write dH (x), δH , and α(H ) for the degree of x in H, the minimum degree,
and the independence number of H respectively. We denote by dG (x, y) the distance between x and y in the graph G.
A factor of G is a spanning subgraph of G, that is a subgraph obtained by edge deletions only. If S is the set of deleted edges,
then this subgraph is denoted G − S. If H is a subgraph of G, then G − H stands for the subgraph induced by V (G) − V (H )
in G. By starting with a disjoint union of two graphs G1 and G2 and adding edges joining every vertex of G1 to every vertex
of G2 , we obtain the join of G1 and G2 , denoted G1 + G2 . For a positive integer p, the graph pG consists of p vertex-disjoint
copies of G. In all that follows, we use disjoint to stand for vertex-disjoint.
In [2], we defined a pseudo 2-factor of a graph G to be a factor each component of which is a cycle, an edge or a vertex. It
can also be seen as a graph partition by a family of vertices, edges and cycles. Graph partition problems have been studied in
lots of papers. They consist of partitioning the vertex set of G by disjoint subgraphs chosen to have some specific properties.
In [3], Enomoto listed a variety of results dealing with partitions into paths and cycles. The emphasis is generally on the
existence of a given partition. In our study of pseudo-factors, however, we take interest in the number of components
that are edges or vertices in a pseudo-factor of G. In [2], we proved that every graph with minimum degree δ ≥ 1 and
independence number α ≥ δ possesses a pseudo 2-factor with at most α − δ + 1 edges or vertices and that this bound
is best possible. Motivated by the desire to know what happens in general cases, we define a pseudo [a, b]-factor (where a
and b are two integers such that b ≥ a ≥ 2) as a factor of G in which each component C on at least three vertices verifies

∗ Fax: +213 21 24 79 07.

E-mail addresses: [email protected], [email protected].

0166-218X/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.dam.2013.09.005
 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114 109

a ≤ dC (x) ≤ b, for every x ∈ C . Clearly, a pseudo [a, b]-factor with no component that is an edge or a vertex is nothing but
an [a, b]-factor. Surveys on factors and specifically [a, b]-factors and connected factors can be found in [4,5]. In the present
work, we study pseudo [2, b]-factors, we consider the case b ≥ 4 and obtain an upper bound (in function of δ , α and b) for
the number of components that are edges or vertices in a pseudo [2, b]-factor of G which minimizes the number of such
b(δ−1)
components. Note that, from a result by Kouider and Lonc [6], we deduce that if α ≤ 2 then G has a [2, b]-factor. Laying
b(δ−1)
down the condition α > 2
, the main result of this paper reads as follows:

Theorem 1. Let b be an integer such that b ≥ 4 and G a graph of minimum  δ ≥ 1 and independence number α with
degree
α > b(δ− 1)
, α (δ )
b
2
. Then G possesses a pseudo [ 2 b ]-factor with at most − 2
− 1 components that are edges or vertices.
The bound given in Theorem 1 is best possible. Indeed, let b be an integer such that b ≥ 4 and let H be a graph with
nonempty set of vertices and empty set of edges. The graph G = H + pK2 , where p > 2b |H |, has minimum degree δ = |H | + 1
and independence number α = p. We can easily verify that G possesses a pseudo [2, b]-factor with α − 2b (δ − 1) edges
 
and we cannot do better. Also, a simple example reaching the bound of Theorem 1, is a graph G obtained by taking a graph
H on n vertices in which every vertex is of degree between 2 and b (b ≥ 4), then taking n additional independent vertices
and joining exactly one isolated vertex to exactly one vertex of H. The graph G has minimumdegree δ = 1, independence
number α = n and can be partitioned into one component that is H and n = α − 2b (δ − 1) vertices (or simply n edges)
and we cannot do better.
Combining Theorem 1 with the results of [2] and [6], we obtain

Corollary 1. Let b ≥ 2 be an integer such that b ̸= 3. Let G be a graph of minimum degreeδ and independence number α and
without isolated vertices. Then G possesses a pseudo [2, b]-factor with at most max(0, α − 2b (δ − 1) ) edges or vertices.


2. Independence number, minimum degree and pseudo [2, b]-factors

First of all, we put aside the case δ = 1 for which we know that we have in G a pseudo [2, b]-factor with at most α edges
or vertices. Indeed, if we regard a cycle as a component each vertex of which is of degree between 2 and b, then  we know
that any graph G can be covered by at most α cycles, edges or vertices (see for instance [7]). So the bound α − 2b (δ − 1)

holds for δ = 1.
From now on, we assume that G has minimum degree δ ≥ 2. Let F be a subgraph of G such that 2 ≤ dF (x) ≤ b for all
x ∈ V (F ). For the sake of simplifying the writing, such a subgraph F will be called a [2, b]-subgraph of G. Denote by D a
smallest component of G − F , set W = G − (D ∪ F ) and choose F in such a manner that:
(a) α(G − F ) is as small as possible;
(b) subject to (a), the number of vertices of D is as small as possible;
(c) subject to (a) and (b), the number of vertices in F is as small as possible.
Notice that a subgraph F satisfying the conditions above exists since G contains a cycle (as δ ≥ 2). We shall show the
following theorem which yields Theorem 1:

Theorem 2. Let b be an integer such that b ≥ 4. Let G be a graph of minimum degree δ ≥ 2 and independence number α
b(δ−1)
such that α > 2 . Then there exists a pseudo [2, b]-factor of G and a [2, b]-subgraph F of this pseudo [2, b]-factor such that
α(G − F ) ≤ α − (δ − 1) .
b 
2
Proof of Theorem 2. Let F be a [2, b]-subgraph of G satisfying the conditions (a), (b) and (c). Denote by u1 , . . . , um (m ≥ 1)
the neighbors of D on F and by Pij a path with internal vertices in D joining two vertices ui and uj with 1 ≤ i, j ≤ m and
i ̸= j. The proof of Theorem 2 will be divided into several claims. The following one, which will intensively be used, reminds
Lemma 1 in [2].

Claim 1. Let F ′ be a [2, b]-subgraph of G which contains the neighbors of D in F and at least one vertex of D. Setting W ′ =
G − (F ′ ∪ D), we have α(W ′ ) > α(W ).
Proof of Claim 1. Set D′ = D − F ′ .
(1) If D′ = ∅, then by the choice of F , we have α(G − F ) ≤ α(G − F ′ ). But α(G − F ) = α(W ) + α(D) ≥ α(W ) + 1 and
α(G − F ′ ) = α(W ′ ), so α(W ) < α(W ′ ).
(2) If D′ ̸= ∅, then F ′ gives a component D′ smaller than D, so again by the choice of F , we have α(W )+α(D) = α(G − F ) <
α(G − F ′ ) = α(W ′ ) + α(D′ ). But as α(D′ ) ≤ α(D), we obtain α(W ) < α(W ′ ). 
In the next claims, we try to learn more about the degrees in F of its vertices.

Claim 2. For every i, 1 ≤ i ≤ m, we have NF (ui ) ∩ {u1 , . . . , um } = ∅.

Proof of Claim 2. Suppose that for some i, NF (ui ) ∩ {u1 , . . . , um } ̸= ∅, then there exists a vertex uj (1 ≤ j ≤ m and j ̸= i)
such that ui uj ∈ E (F ). Put e = ui uj , then (F − e) ∪ Pij is a [2, b]-subgraph. Indeed, none of the vertices of F changes its degree
in (F − e) ∪ Pij and the internal vertices of Pij are of degree 2. So taking F ′ = (F − e) ∪ Pij in Claim 1 we obtain α(W ) > α(W ),
which is absurd. 
110 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114

Claim 3. dF (ui ) ≤ b − 1 for at most one vertex ui , i = 1, . . . , m.

Proof of Claim 3. Suppose to the contrary that there exist at least two distinct vertices uk and ul such that dF (uk ) ≤ b − 1
and dF (ul ) ≤ b − 1. Then taking F ′ = F ∪ Pkl in Claim 1 (notice that in F ′ , dF (uk ) and dF (ul ) are at most b, and the internal
vertices of Pkl are of degree 2 in F ′ so F ′ is a [2, b]-subgraph of G), we obtain α(W ) < α(W ) which is absurd. 

i=1 NF (ui ) such that x is a common neighbor of at least two vertices in {u1 , . . . , um }. We
Let S be the set of vertices x in ∪m
have:

Claim 4. 1. dF (x) ≤ 3 for every x ∈ S.

2. If S contains a vertex x such that dF (x) = 3, then
(a) For every k, 1 ≤ k ≤ m, we have dF (uk ) = b.
(b) For every y ∈ ∪m i=1 NF (ui ) − {x} we have dF (y) = 2.

Proof of Claim 4. 1. Suppose that dF (x) ≥ 4 for some x ∈ S. By definition, x is the neighbor in F of at least two vertices say
ui and uj with 1 ≤ i, j ≤ m, i ̸= j. Put e = xui and e′ = xuj . Then in F ′ = (F − e − e′ ) ∪ Pij only x changes its degree but it
remains at least 2. So F ′ is a [2, b]-subgraph which leads to a contradiction by Claim 1.
2. Let x be in NF (ui ) ∩ NF (uj ) (1 ≤ i, j ≤ m and i ̸= j) such that dF (x) = 3. Suppose that there exists uk (which will be the
only one by Claim 3) such that dF (uk ) ≤ b − 1, we can always assume that k ̸= i. Then taking F ′ = (F − e) ∪ Pik , where
e = xui , in Claim 1 gives a contradiction.
Furthermore, we suppose that there exists y ∈ NF (uk ) − {x}, with 1 ≤ k ≤ m (we can suppose without loss of generality
that k ̸= i) such that dF (y) ≥ 3. Then setting e = xui , e′ = yuk and taking F ′ = (F − e − e′ ) ∪ Pik in Claim 1 gives a
contradiction. Notice that F ′ is a [2, b]-subgraph: indeed, only x and y lose 1 in their degree but they remain of degree at
least 2 in F ′ and the internal vertices of Pik are of degree 2 in F ′ . 
Claim 4 implies that S is an independent set in F and we will deduce later that it is also independent in G. But before that,
we take a look at the neighbors of {u1 , . . . , um } which are not in S. For each ui (i = 1, . . . , m), set NF∗ (ui ) = {x ∈ NF (ui );
x ̸∈ S }.

i=1 NF (ui ) such that dF (x) ≥ 3, then these vertices are in the neighborhood of the same
Claim 5. 1. If there exist vertices x in ∪m ∗

uk , 1 ≤ k ≤ m.
2. If there exist k, 1 ≤ k ≤ m, such that dF (uk ) ≤ b − 1 and x ∈ ∪m i=1 NF (ui ) such that dF (x) ≥ 3, then x ∈ NF (uk ).
∗

Proof of Claim 5. 1. Suppose that there exist x ∈ NF∗ (uk ) such that dF (x) ≥ 3, x′ ∈ NF∗ (uj ) such that dF (x′ ) ≥ 3 and
1 ≤ j, k ≤ m, j ̸= k. Then the subgraph (F − e − e′ ) ∪ Pkj , where e = uk x and e′ = uj x′ is a [2, b]-subgraph of G. Taking
F ′ = (F − e − e′ ) ∪ Pkj in Claim 1, we obtain a contradiction.
2. Suppose that there exist k, 1 ≤ k ≤ m, such that dF (uk ) ≤ b − 1 and x ∈ ∪m i=1 NF (ui ) with dF (x) ≥ 3. By Claim 4(2), x ̸∈ S.
Suppose that x ∈ NF∗ (ui ), with i ̸= k. Notice that the fact that dF (uk ) ≤ b − 1 forces dF (ui ), by Claim 3, to be equal to b.
Taking F ′ = (F − e) ∪ Pik , where e = xui , in Claim 1, we obtain α(W ) > α(W ) which is absurd. 
Looking more closely at the structure of D, we can say more about the degrees of the vertices in ∪m i=1 NF [ui ], where
NF [ui ] = NF (ui ) ∪ {ui } is the closed neighborhood of ui . First, we remark that D has minimum degree at most 1.

Remark 1. δD ≤ 1.
Proof. Suppose, by contradiction, that δD ≥ 2 then taking a longest path in D provides a cycle C which verifies α(D − C ) <
α(D). Put F ′ = F ∪ C , then F ′ is a [2, b]-subgraph of G. Moreover, α(G − F ′ ) = α(D − C ) + α(W ) < α(D) + α(W ) = α(G − F )
and this contradicts the choice of F . 
Two cases are to consider, the case where D is a tree (a single vertex is a trivial tree) and the case where D contains a
cycle. The following claim deals with this latter case.

Claim 6. Suppose that D contains a cycle. Then

1. dF (x) = 2 for all x ∈ ∪m i = 1 N F ( ui ) .
2. dF (ui ) = b for all i, 1 ≤ i ≤ m.
Proof of Claim 6. 1. Suppose that there exists a vertex x ∈ NF (ui ) such that dF (x) ≥ 3 and let Q be an edge or a path with
internal vertices in D − C joining ui and C . Then taking F ′ = (F − e)∪ Q ∪ C , where e = xui , in Claim 1 gives a contradiction.
Notice that dF ′ (x) ≥ 2 and that ui does not change its degree (nor do the other vertices of F ) then F ′ is a [2, b]-subgraph
of G.
2. Suppose that dF (uk ) ≤ b − 1 for some k, 1 ≤ k ≤ m and let Q be an edge or a path with internal vertices in D − C joining
uk and C . Then, taking F ′ = F ∪ Q ∪ C in Claim 1 gives a contradiction. 
If D is a tree and δD ̸= 0, then D has at least two leaves, say x0 and y0 . We relabel u1 , . . . , um1 , with m1 ≤ m, the vertices
in NF (x0 ) ∪ NF (y0 ).
 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114 111

Claim 7. Suppose that D is a tree and that there exist two vertices x0 and y0 in D with dD (x0 ) = dD (y0 ) = 1 such that NF (x0 ) =
NF (y0 ). Then

1. For all k, 1 ≤ k ≤ m1 , dF (uk ) ≥ b − 1.

m
2. If there exists a vertex x ∈ ∪i=11 NF (ui ) such that dF (x) ≥ 3, then it is the only one.
m
3. If there exists a vertex uk (with 1 ≤ k ≤ m1 ) such that dF (uk ) = b − 1, then for every vertex x ∈ ∪i=11 NF (ui ) we have
dF (x) = 2.

Proof of Claim 7. Let P be a path in D joining x0 to y0 .

1. Suppose that there exists a vertex ui (1 ≤ i ≤ m1 ) such that dF (ui ) ≤ b − 2. Then taking F ′ = F ∪ ui x0 Py0 ui in Claim 1
gives a contradiction.
m
2. Suppose that there exists a vertex x ∈ ∪i=11 NF (ui ) such that dF (x) ≥ 3. If x ∈ S, then Claim 4 gives what desired. If x ̸∈ S,
m
then x ∈ NF (uk ) for some 1 ≤ k ≤ m1 . Suppose that there exists y ∈ ∪i=11 NF (ui ) such that dF (y) ≥ 3 so y ∈ NF∗ (uj ), for
∗

some 1 ≤ j ≤ m1 . By Claim 5(1), j = k. Taking F = (F − e − e ) ∪ uk x0 Py0 uk , where e = xuk and e′ = yuk , in Claim 1, we
′ ′

obtain a contradiction.
m
3. Finally, suppose that there exist a vertex uk (1 ≤ k ≤ m1 ) such that dF (uk ) = b − 1 and a vertex x ∈ ∪i=11 NF (ui )
such that dF (x) ≥ 3. By Claim 5(2), x ∈ NF (uk ). Put e = xuk . Then taking F = (F − e) ∪ uk x0 Py0 uk in Claim 1 gives a
∗ ′

contradiction. 

A path I in F with V (I ) ⊂ V (F ), E (I ) ⊂ E (F ) and such that every internal vertex x of I has dF (x) = 2 is called an interval
(or a segment) of F . We say that two disjoint intervals I (1) and I (2) in F are path-independent if there exists no path internally
disjoint from F ∪ D joining a vertex in I (1) to a vertex in I (2) . We say that t intervals I (1) , I (2) , . . . , I (t ) (t ≥ 2) in F are path-
independent if they are pairwise path-independent. The following claim will be very useful. It is a shorter version of Lemma
2 in [2] with a short proof.

Claim 8. Let I (1) , I (2) , . . . , I (t ) (t ≥ 2) be t disjoint intervals in F , containing no neighbor of D and such that α(W ∪ I (i) ) = α(W )
for every i = 1, . . . , t. If I (1) , I (2) , . . . , I (t ) are path-independent, then α(D ∪ W ∪ I (1) ∪ I (2) ∪ · · · ∪ I (t ) ) = α(W ∪ D).

Proof of Claim 8. Let Wi be the union of components of W with neighbors in I (i) (i = 1, . . . , t). By hypothesis, the intervals
I (i) are pairwise path-independent so Wi ∩ Wj = ∅, for all 1 ≤ i, j ≤ t , i ̸= j. Hence W −∪ti=1 Wi , W1 ∪ I (1) , . . . , Wt ∪ I (t ) form
a partition of W ∪I (1) ∪· · ·∪I (t ) and it follows that α(W ∪I (1) ∪· · ·∪I (t ) ) = α(W1 ∪I (1) )+· · ·+α(Wt ∪I (t ) )+α(W −∪ti=1 Wi ). On
(i) (i) (1)
t other hand, as α(W ∪It ) = α(W ) for every i = 1, . . . , t, one has α((1W
the i ∪ I ) = α(Wi ). This yields α(W ∪ I ∪· · ·∪I (t ) ) =
)
i=1 α(Wi ) + α(W − ∪i=1 Wi ) = α(W ). We finally get α(W ∪ D ∪ I ∪ . . . ∪ I ) = α(W ∪ D) because the intervals I (i)
(t )

(with i = 1, . . . , t) contain no neighbor of D. 

Let s be the vertex of S (if it exists) such that dF (s) = 3. We put s aside before applying the procedure described hereafter.
Provided always that s exists, we set NF′ (ui ) = NF (ui ) − {s} if s ∈ NF (ui ) and NF′ (ui ) = NF (ui ) otherwise (i = 1, . . . , m).
For uk , 1 ≤ k ≤ m, denote by xki (i = 1, . . . , |NF′ (uk )|) its neighbors that belong to NF′ (uk ). Using this notation, we can
have xki = xlj , for some 1 ≤ i ≤ |NF′ (uk )| and 1 ≤ j ≤ |NF′ (ul )| (1 ≤ k, l ≤ m, k ̸= l), in case xki ∈ NF′ (uk ) ∩ NF′ (ul ) ⊂ S.
From now on, let m′ = m1 if D is a tree with at least two leaves and m′ = m otherwise. Starting at a vertex uk (1 ≤ k ≤ m′ ),
we go to one of its neighbors in F (say x) then to a neighbor y (y ̸= uk ) of x in F , then to a neighbor (not x) of y in F and so on.
We consider this kind of path as a sense of ‘‘orientation’’. Let uk (1 ≤ k ≤ m′ ) be such that dF (uk ) = b and all its neighbors
that are in NF′ (uk ) are of degree 2 in F . Starting at xk1 and following the chosen orientation we go over from a vertex to its
k,+ k,+ k,+
neighbor until meeting a vertex which we call yk1 , such that dF (y1 ) ≥ 3 or y1 = uj for some j, 1 ≤ j ≤ m (where y1 is
the successor of yk1 following the chosen orientation). This gives an interval xk1 . . . yk1 which we denote by P1k = [xk1 , yk1 ]F .
We repeat the process using the other neighbors of uk that are in NF′ (uk ). At the pth step, we consider a vertex xkp ∈ Xp =
NF′ (uk ) − (∪i=1 V (Pik )) and construct a path Ppk = xkp . . . ykp containing xkp and such that dF (ykp,+ ) ≥ 3 or ykp,+ = uj for some
p−1

j, 1 ≤ j ≤ m. When Xr becomes empty at the rth step (r ≥ p), then we consider another vertex ul (l ̸= k). We choose as long
as possible ul such that dF (ul ) = b and all vertices in NF′ (ul ) are of degree 2 in F . We choose a vertex in NF′ (ul ) − ∪ir=−11 V (Pik ),
′
and we do the same construction, until the vertices in NF′ (ul ) are all in (∪ir=−11 V (Pik )) ∪ (∪ri=−11 V (Pil )). Denote by P the set
of paths obtained so far. When it is no longer possible to choose a vertex up , 1 ≤ p ≤ m′ , such that dF (up ) = b and all
vertices in NF′ (up ) have degree 2 in F , then we take the vertex uq of degree at most b − 1 or a vertex uq such that NF′ (uq ) has
vertices of degree at least 3 in F . Notice that uq exists only if s does not (see Claim 4(2)) and if both a vertex uq of degree at
j
most b − 1 (which would be the only one by Claim 3) and vertices xi of degree at least 3 exist, then these vertices are in the
q q q
neighborhood of uq (see Claim 5). Put Nq = {xi ∈ NF′ (uq ) − V (P) such that dF (xi ) = 2}. Starting at a vertex xi ∈ Nq , we
repeat the construction described above until Nq becomes empty. We update the set P at each step.
By construction all the vertices of Pik are of degree 2 in F so V (Pik ) ∩ V (Pjl ) = ∅ for every couple Pik , Pjl of paths in P (they
are disjoint), moreover no vertex in Pik is adjacent in F to a vertex in Pjl , for all Pik , Pjl in P.
112 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114

We divide the set P into three subsets, each containing the paths Pik = [xki , yki ]F of Type 1, Type 2 or Type 3, defined as
follows:
k,+
Type 1 If yi = uj , with j ̸= k.
k,+
Type 2 If yi = uj , with j = k.
k,+
Type 3 If yi ̸= uj for every j, 1 ≤ j ≤ m.
For technical reasons, in case D is a trivial tree or a tree having no couple of leaves with the same neighborhood in F ,
we stop the procedure described above when there remains no vertex up (1 ≤ p ≤ m′ ) such that dF (up ) = b, or when the
remaining vertex up (1 ≤ p ≤ m′ ) has in its neighborhood NF′ (up ) a vertex of degree at least 3. We consider Q the subset of
P, of paths obtained till then. Let P1 = Q in this case and P1 = P in the others.
We show in what follows that the addition of a path of P1 to W ∪ D augments α(W ∪ D) by at least 1.

Claim 9. For each Pik ∈ P1 , we have α(W ∪ D ∪ Pik ) > α(W ∪ D).

Proof of Claim 9. Let Pik be a path in P1 .

1. If D contains a cycle, then taking F ′ = F − Pik gives what is desired. Indeed, in this case all the vertices ui are of degree b
(by Claim 6); as b ≥ 4, after the deletion of Pik , the degree of the vertices ui (1 ≤ i ≤ m′ ) remains at least 2. Moreover,
by construction of Pik , the degree of no vertex in F becomes smaller than 2, after deletion of Pik . So F ′ is a [2, b]-subgraph
of G. F ′ contradicts condition (c) in the choice of F (because |V (F ′ )| < |V (F )|) so α(G − F ′ ) > α(G − F ) which yields
α(W ∪ D ∪ Pik ) > α(W ∪ D).
2. Suppose that D is a tree possessing two vertices x0 and y0 of degree 1 in D, having the same neighborhood in F
(NF (x0 ) = NF (y0 )). In this case, if Pik is of Type 1 or 3, then we reason as in (1) and we obtain what desired. If Pik is
of Type 2, then (1) is no more efficient if dF (uk ) = b − 1 (because the degree of uk may become smaller than 2 when
Pik is deleted). So we take F ′ = (F − Pik ) ∪ uk x0 Py0 uk , where P is a path with internal vertices in D joining x0 to y0 . The
subgraph F ′ is a [2, b]-subgraph of G (we have dF (uk ) = dF ′ (uk )) which, by Claim 1, gives what is desired.
3. In the other cases, as b ≥ 4 and by the choice of the subset P1 , the deletion of any path Pik ∈ P1 , gives a [2, b]-subgraph.
Reasoning as in (1), we get what is desired. 

Notice that as D is independent from Pik (by construction) and from W , one has α(W ∪ D ∪ Pik ) = α(W ∪ Pik ) + α(D).
Hence the conclusion in Claim 9 is equivalent to α(W ∪ Pik ) > α(W ). For each path Pik = [xki , yki ]F in P1 and following the
chosen orientation, let vik be the first vertex of Pik such that α(W ∪ [xki , vik ]F ) > α(W ). Notice that vik is well defined by
′ ′
Claim 9. Denote by Pi k the interval [xki , vik ]F of Pik and by P′ the set of the intervals Pi k . In what follows, we take interest in
the path-independence of the intervals of P′ .

′ ′ ′ ′
Claim 10. Let Pi k and Pj l be two distinct intervals [xki , vik ]F and [xlj , vjl ]F in P′ such that 1 ≤ k, l ≤ m′ , k ̸= l. Then Pi k and Pj l
are path-independent.
′ ′
Proof of Claim 10. By way of contradiction, suppose that there exist two vertices aki ∈ Pi k and alj ∈ Pj l such that aki and
alj are joined by Q which is an edge in G or a path with internal vertices in W . Choose aki and alj so as to minimize the sum
dF (xki , aki ) + dF (xlj , alj ). Recall that by construction xy ̸∈ E (F ) for every x ∈ Pik and y ∈ Pjl .
The segments [xki , aki [F and [xlj , alj [F verify the hypothesis of Claim 8. Indeed, by the choice of aki and alj , they are path-
independent. Furthermore, as [xki , aki [F ⊂ [xki , vik [F and [xlj , alj [F ⊂ [xlj , vjl [F , and by the choice of vik and vjl , we have α(W ∪
[xki , aki [F ) = α(W ) and α(W ∪ [xlj , alj [F ) = α(W ). So by Claim 8, we obtain

α(W ∪ [xki , aki [F ∪[xlj , alj [F ) = α(W ). (⋆)

Also, taking the [2, b]-subgraph F ′ = (F −([xki , aki [F ∪[xlj , alj [F ))∪ Q ∪ Pkl , in Claim 1, gives α((W − Q )∪[xki , aki [F ∪[xlj , alj [F ) >
α(W ). But as α(W ∪ [xki , aki [F ∪[xlj , alj [F ) ≥ α((W − Q ) ∪ [xki , aki [F ∪[xlj , alj [F ) we get α(W ∪ [xki , aki [F ∪[xlj , alj [F ) > α(W ), which
contradicts (⋆). 

When k = l in the previous claim, then we consider the structure of D. If D contains a cycle or D is a tree with two leaves
′
x0 and y0 such that NF (x0 ) = NF (y0 ), then the following claim gives the path-independence of any couple of segments Pi k
′
and Pj k in P′ .

′ ′
Claim 11. Let Pi k and Pj k be two distinct segments of P′ . Suppose that D contains a cycle or D is a tree with two leaves x0 and
′ ′
y0 such that NF (x0 ) = NF (y0 ). Then Pi k and Pj k are path-independent, for every k, 1 ≤ k ≤ m′ .
 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114 113

′ ′
Proof of Claim 11. Let Pi k and Pj k (with 1 ≤ i, j ≤ m′ , i ̸= j) be two segments in P′ . By way of contradiction, suppose that
′ ′
there is a path Q internally disjoint from F ∪ D joining a vertex aki ∈ Pi k to a vertex akj ∈ Pj k and choose aki and akj so that the
sum dF (xki , aki ) + dF (xkj , akj ) is minimum.
The segments [xki , aki [F and [xkj , akj [F verify the hypothesis of Claim 8. Indeed, they are path-independent, by the choice
of aki and akj . Furthermore, as [xki , aki [F ⊂ [xki , vik [F and [xkj , akj [F ⊂ [xkj , vjk [F , and by the choice of vik and vjk , we have α(W ∪
[xki , aki [F ) = α(W ) and α(W ∪ [xkj , akj [F ) = α(W ). So by Claim 8, we obtain

α(W ∪ [xki , aki [F ∪[xkj , akj [F ) = α(W ). (⋆⋆)

′
On the other hand, if D contains a cycle C , then let Q be a path with internal vertices in D − C joining uk to a vertex on C .
Suppose that D is a tree with two leaves x0 and y0 such that NF (x0 ) = NF (y0 ). Then let P be a path with internal vertices in
D joining x0 to y0 . Taking F ′ = (F − ([xki , aki [F ∪[xkj , akj [F )) ∪ Q ∪ Q ′ ∪ C in the first case and F ′ = (F − ([xki , aki [F ∪[xkj , akj [F )) ∪
Q ∪ uk x0 Py0 uk in the second one and using Claim 1, we obtain in both cases α(W ∪ [xki , aki [F ∪[xkj , akj [F ) ≥ α((W − Q ) ∪
[xki , aki [F ∪[xkj , akj [F ) > α(W ), which contradicts (⋆⋆). 
Suppose now that D is either a trivial tree or D has no leaves with the same neighborhood in F .
′ ′
• If all couples of distinct segments (Pi k , Pj k ) (k, 1 ≤ k ≤ m′ , 1 ≤ i, j ≤ |NF′ (uk )|) in P′ are path-independent, then we have
′
finished. It is particularly the case if s exists. Indeed, suppose to the contrary that there exist two distinct segments Pi k and
′k
Pj (k, 1 ≤ k ≤ m′ , 1 ≤ i, j ≤ |NF′ (uk )|) in P′ that are path-dependent, that is there is a path Q internally disjoint from D∪F
′ ′
joining a vertex aki ∈ Pi k to a vertex akj ∈ Pj k . We choose these vertices so as to minimize the sum dF (xki , aki ) + dF (xkj , akj ).
Reasoning as in the previous claims using Claim 8 and taking in Claim 1 F ′ = (F − [xki , aki [F ∪[xkj , akj [F ) ∪ Q ∪ Pkr − ur s
where ur is a neighbor of s such that r ̸= k and subtracting ur s means deleting the edge ur s not its extremities, we get a
contradiction.
It is also the case if there exists a vertex ur (1 ≤ r ≤ m) that is of degree at most b − 1 in F or that has in its neighborhood
NF (ur ) a vertex x such that dF (x) ≥ 3. Recall that in our case, this vertex is supposed to be put apart in the procedure we
′ ′
have used. So, suppose that there is a path Q internally disjoint from D ∪ F joining a vertex aki ∈ Pi k to a vertex akj ∈ Pj k
(k ̸= r). We choose these vertices so as to minimize the sum dF (xki , aki ) + dF (xkj , akj ). Here again, using Claims 1 and 8 with
F ′ = (F − [xki , aki [F ∪[xkj , akj [F ) ∪ Q ∪ Pkr if dF (ur ) ≤ b − 1 or F ′ = (F − [xki , aki [F ∪[xkj , akj [F ) ∪ Q ∪ Pkr − ur x if dF (ur ) = b
and dF (x) ≥ 3 where x ∈ NF′ (ur ), we get a contradiction.
• If not, then this case is treated in the following claim.

Claim 12. Suppose that D is a trivial tree or a tree with no leaves having the same neighborhood in F . Suppose moreover that
′ ′
there exist two distinct segments Pi k and Pj k (k, 1 ≤ k ≤ m′ , 1 ≤ i, j ≤ |NF′ (uk )|) in P′ that are path-dependent. Then there exists
′ ′
no other couple of segments (Ppl , Pql ) (l ̸= k, 1 ≤ l ≤ m′ , 1 ≤ p, q ≤ |NF′ (ul )|, p ̸= q) in P′ that are path-dependent.
′ ′
Proof of Claim 12. The proof is basically the same as the previous. Let aki and akj be two vertices in Pi k and Pj k respectively
that are joined by a path Q internally disjoint from F ∪ D and chosen so as to minimize the sum dF (xki , aki ) + dF (xkj , akj ).
′ ′
Suppose to the contrary that there exist two distinct segments Ppl , Pql (l ̸= k, 1 ≤ l ≤ m′ , 1 ≤ p, q ≤ |NF′ (ul )|) and a path
′ ′
Q ′ internally disjoint from F ∪ D joining a vertex alp ∈ Ppl to a vertex alq ∈ Pql and choose these vertices in such a way
that dF (xlp , alp ) + dF (xlq , alq ) is minimum. Then using Claim 8 with four intervals and taking F ′ = (F − ([xki , aki [F ∪[xkj , akj [F ) −
([xlp , alp [F ∪[xlq , alq [F ))∪ Q ∪ Q ′ ∪ Pkl in Claim 1 yield a contradiction. As k ̸= l, then Claim 10 guarantees the path-independence
′ ′
of the segments Pr k , Pt l for r ∈ {i, j}, t ∈ {p, q}. 
By the claims above, we have that P′ contains several segments that are path-independent. Furthermore, the following
remark claims that an additional segment can be considered when needed, particularly when S contains a vertex of
degree 3.

Remark 2. If there exists a vertex s ∈ S such that dF (s) = 3 (s is unique by Claim 4(2)). We consider two cases:
(i) Suppose that s is in the neighborhood of three vertices uk , ul and up , with 1 ≤ k, l, p ≤ m and k, l, p pairwise distinct.
Then setting P ∗ = {s} we have that P ∗ is path-independent from any path in P′ and α(W ∪ P ∗ ) > α(W ).
(ii) Suppose that s is in the neighborhood of exactly two vertices, say uk and ul , k ̸= l, 1 ≤ l, k ≤ m. If furthermore P
does not contain paths of Type 3, then there exists a path P ∗ that is path-independent from any segment in P′ included in
a path of Type 1 or Type 2. Furthermore α(W ∪ P ∗ ) > α(W ).
Proof. (i) First, taking F ′ = (F −{s})∪ Pkl in Claim 1 we obtain α(W ∪{s}) > α(W ). Of course, since by Claim 4(2) dF (ui ) = b
for all i = 1, . . . , m, we have that F ′ is a [2, b]-subgraph of G. As s ∈ NF (uk ) ∩ NF (ul ) ∩ NF (up ) ⊂ NF (uk ) ∩ NF (ul ), we can
write {s} = Pik or {s} = Pil as suitable to apply Claim 10 and show the path-independence of {s} from any segment in P′ .
114 S. Bekkai / Discrete Applied Mathematics 162 (2014) 108–114

(ii) Let us start from s and go forward following the chosen orientation from a vertex of degree 2 in F to a vertex of degree
2 in F , until coming across a vertex y whose successor y+ is of degree at least 3. We have that y+ ̸∈ {u1 , . . . , um }, otherwise,
going in the opposite direction, the segment [y, s[F (where s is not taken) is a path of Type 3. Moreover, y+ ̸∈ ∪m i=1 NF (ui )
because since s exists then by Claim 4 every vertex in ∪m i=1 NF (u i ) is of degree 2. So y+
∈ V ( F ) − ∪m
i=1 N F [ u i ] . The path
′ ′
P = s . . . y can be considered as a path deriving from uk (P = Pi k ) or deriving from ul (P = Pi l ) and hence reasoning as in
Claim 9, taking F ′ = F − P, we obtain α(W ∪ P ) > α(W ). Let v be the first vertex of P following the chosen orientation such
that α(W ∪ [s, v]F ) > α(W ). Setting P ∗ = [s, v]F , we can show its path-independence with any segment in P′ included in
a path of Type 1 or Type 2, like in Claim 10. 
Finally, we count the number of pairwise path-independent segments in P′ , those whose independence
 is guaranteed

b(δ−1)
by Claims 10–12, we distinguish different cases according to the structure of D and in any case get at least 2
(recall
that m ≥ δ − 1 by Remark 1) path-independent segments, adding when necessary the path P (in particular when s exists).
′ ∗

By Claim 9 and Remark 2, each of the segments in P′ ∪ {P ∗ }, when added to W ∪ D, augment α(W ∪ D). Put L = P′ ∪ {P ∗ }.
Recall that the segments of L are independent from D by construction. For each P ∈ L, let WP be the union of components
of W that contain a neighbor of P.
We have that
 
α(W ∪P ∈L P ) = α(W − ∪P ∈L WP ) + α(WP ∪ P )
P ∈L

≥ α(W − ∪P ∈L WP ) + α(WP ) + |L|
P ∈L
 
b
≥ α(W ) + (δ − 1) .
2
Hence
  
α = α(G) = α(W ∪ D ∪ F ) ≥ α W ∪ D ∪P ∈L P
  
≥ α(D) + α W ∪P ∈L P
 
b
≥ α(D) + α(W ) + (δ − 1)
2
 
b
= α(W ∪ D) + (δ − 1) .
2

So α(W ∪ D) = α(G − F ) ≤ α − (δ − 1) and the proof of Theorem 2 is achieved. 

b 
2

Proof of Theorem 1. Since by Theorem 2 we have α(G − F ) ≤ α − 2b (δ − 1) , the subgraph of G induced by V (W ∪ D) =

 

V (G − F ) can be covered by at most α − 2 (δ − 1) cycles, edges or vertices (see for instance [7]). Denote by E the set of
b 

cycles, edges or vertices covering G − F . The graph F ∪ E is a pseudo [2, b]-factor of G with at most α − 2b (δ − 1) edges
 
or vertices. This completes the proof of Theorem 1. 

Acknowledgment

The question solved in this paper was proposed by M. Kouider.

References

[1] J.A. Bondy, U.S.R. Murty, Graph theory with applications, MacMillan and Co., London, 1976.
[2] S. Bekkai, M. Kouider, On Pseudo 2-factors, Discrete Appl. Math. 157 (2009) 774–779.
[3] H. Enomoto, Graph partition problems into cycles and paths, Discrete Math. 233 (2001) 93–102.
[4] M.D. Plummer, Graph factors and factorization: 1985–2003: A survey, Discrete Math. 307 (2007) 791–821.
[5] M. Kouider, P.D. Vestergaard, Connected Factors in Graphs-a Survey, Graphs Combin. 21 (2005) 1–26.
[6] M. Kouider, Z. Lonc, Stability number and [a, b]-factors in graphs, J. Graph Theory 46 (2004) 254–264.
[7] L. Posá, On the circuits of finite graphs, MTA Mat. Kut. Int. Kozl. 8 (1964) 355–361.