Discrete Applied Mathematics xxx (xxxx) xxx

Contents lists available at ScienceDirect

Discrete Applied Mathematics

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

Burning number of caterpillars

∗,2
Huiqing Liu a ,1 , Xuejiao Hu a ,1 , Xiaolan Hu b ,
a
Hubei Key Laboratory of Applied Mathematics, Faculty of Mathematics and Statistics, Hubei University, Wuhan 430062, PR China
b
School of Mathematics and Statistics & Hubei Key Laboratory of Mathematical Sciences, Central China Normal
University, Wuhan 430079, PR China

article info a b s t r a c t

Article history: Graph burning is a deterministic discrete time graph process that can be interpreted
Received 20 August 2019 as a model for the spread of influence in social networks. The burning number b(G) of
Received in revised form 23 March 2020 a graph G is the minimum number of steps in a graph burning process for G. Bonato
Accepted 28 March 2020 √
et al. (2014) conjectured that b(G) ≤ ⌈ n⌉ for any connected graph G of order n. In
Available online xxxx
this paper, we confirm this conjecture for caterpillars. We also determine the burning
Keywords: numbers of caterpillars with at most two stems and a subclass of the class of caterpillars
Burning number all of whose spine vertices are stems.
Distance domination © 2020 Elsevier B.V. All rights reserved.
Caterpillar

1. Introduction

The burning number of a graph was introduced by Bonato et al. in [3]. Given a graph G, the burning process on G is a
discrete-time process defined as follows. Initially, at time t = 0 all vertices are unburned. At each time step t ≥ 1, one
new unburned vertex xt is chosen to burn (if such a vertex is available), and such a vertex is called the tth fire source of G.
If a vertex is burned, then it remains burned until the end of the process. Once a vertex is burned in step t, in step t + 1
each of its unburned neighbors becomes burned. The process ends when all vertices of G are burned. Suppose that we
burn a graph G in k steps in a burning process. The sequence (x1 , . . . , xk ) is called a burning sequence for G. The burning
number of a graph G, denoted by b(G), is the length of a shortest burning sequence for G. Such a burning sequence is
called an optimal burning sequence for G. For example, for the path P4 (or resp., P5 ) with vertices {v1 , v2 , v3 , v4 } (or resp.,
{v1 , v2 , v3 , v4 , v5 }), the sequence (v2 , v4 ) (or resp., (v2 , v4 , v5 )) is an optimal burning sequence; see Fig. 1. The burning
number can be used as a measure for the speed of spreading fire on the vertex set of graphs.
The following theorem provides an alternative characterization of the burning sequence.

Theorem 1.1 ([3,4,15]). In a graph G, the sequences (x1 , x2 , . . . , xk ) form a burning sequence if and only if, for each pair i and j,
with 1 ≤ i < j ≤ k, d(xi , xj ) ≥ j − i, and the following set equation holds: Nk−1 [x1 ] ∪ Nk−2 [x2 ] ∪ · · · ∪ N1 [xk−1 ] ∪ N0 [xk ] = V (G).
In 2014, Bonato et al. [3] studied the burning number of paths and cycles.
√
Theorem 1.2 ([3]). For a path Pn or a cycle Cn of order n, we have that b(Pn ) = b(Cn ) = ⌈ n⌉.
Based on this result, Bonato et al. made the following conjecture.

∗ Corresponding author.
E-mail address: [email protected] (X. Hu).
1 Partially supported by National Natural Science Foundation of China under Grant Numbers 11571096 and 11971158.
2 Partially supported by National Natural Science Foundation of China under Grant Numbers 11601176 and 11971196.

https://doi.org/10.1016/j.dam.2020.03.062
0166-218X/© 2020 Elsevier B.V. All rights reserved.

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
2 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx

Fig. 1. Burning P4 and P5 (the open circles represent burned vertices).

√
Conjecture 1.3 ([3]). For a connected graph G of order n, b(G) ≤ ⌈ n⌉.

Several bounds on the burning number of graphs are given in [2,3,11,15]. It is shown that the graph burning problem
is NP-complete even for trees with maximum degree three and path-forests (that is, disjoint unions of paths), see [1,15].
But Roshanbin [15] was able to provide the following criterion for a graph to have burning number 2.

Theorem 1.4 ([15]). A graph G of order n satisfies b(G) = 2 if and only if G is of order at least 2, and has maximum degree
n − 1 or n − 2.

Since the burning number is a relatively new parameter, it is also interesting to determine the burning number of
special classes of graphs. Bonato and Lidbetter [5] considered the bounds on the burning numbers of spiders (which are
trees with exactly one vertex of degree strictly greater than two) and path-forests. Sim et al. [16] studied the burning
number of generalized Petersen graphs. Fitzpatrick and Wilm [7] determined the burning number of circulant graphs.
Mitsche et al. [14] established the burning number of graph products. Recently, Mitsche et al. [13] focused on a few
probabilistic aspects of the burning problem. Very recently, Liu et al. [12] considered the burning number of theta graphs.
For more on graph burning and graph searching, see the recent book [6].
In this paper, we focus on the burning number of caterpillars. In Section 2, we present an upper bound on the burning
number of caterpillars, which implies Conjecture 1.3 is true for caterpillars. In Section 3, we determine the burning number
of caterpillars with at most two vertices of degree at least three. In Section 4, we determine the burning number of a
subclass of the class of caterpillars all of whose spine vertices are stems.
At the end of this section, we will introduce some definitions and notation which will be used in this paper. Let
G = (V (G), E(G)) be a graph. For v ∈ V (G), the degree of v , denoted by d(v ), is the number of edges incident with v in
G. For S ⊆ V (G), G − S denotes the subgraph induced by V (G) \ S. For two vertices u and v of G, the distance d(u, v )
between u and v is the number of edges in a shortest path joining u and v . For any positive integer k, let [k] denote the
set {1, 2, . . . , k}. For any vertex v ∈ V (G), the kth closed neighborhood of v in G is the set of vertices whose distance from
v is at most k, and is denoted by Nk [u].

2. Upper bound on the burning number of caterpillars

In this section, we first give the definition of caterpillar, then confirm Conjecture 1.3 for caterpillars.
Let Pn denote the path of order n. A caterpillar is a tree such that removing all leaves gives a path, called the spine. A
non-leaf vertex is called a spine vertex and, more precisely, a stem if it is of degree at least three and a trunk otherwise. Let
L(v ) be the leaves adjacent to a spine vertex v . Then L(v ) = ∅ if v is a trunk. Let T be a caterpillar and let ST = v1 v2 · · · vr
be the spine of T . If r = 1, then T is a star, and b(T ) = 2 by Theorem 1.4. So in the following, we always assume r ≥ 2.
For 1 ≤ i ≤ r, let T (i) be the tree induced by {v1 , v2 , . . . , vi } ∪ (L(v1 ) ∪ L(v2 ) ∪ · · · ∪ L(vi )). Then T (i) is a caterpillar and
|V (T (i) )| ≥ i + 1 as v1 has at least one leaf neighbor.
Caterpillars are an important subclass of trees, and caterpillars were first studied in a series of papers by Harary and
Schwenk [9].
⌈√ ⌉
Theorem 2.1. Let T be a caterpillar of order n, then b(T ) ≤ n .

Proof. Suppose for a contradiction that T is a counterexample of minimal order for which b(T ) is maximal. Let ST =
v1 v2 · · · vr be the spine of T , and let v0 be a leaf neighbor of v1 and vr +1 be a leaf neighbor of vr . Suppose that n = q2 + p
with 1 ≤ p ≤ 2q + 1. If r ≤ 2q − 1, then V (T ) = Nq [vq ], and thus b(T ) ≤ q + 1, a contradiction. So r ≥ 2q. We will prove
the following two facts.

Fact 1. d(v2q ) ≥ 3.

Proof of Fact 1. Assume to the contrary that d(v2q ) = 2. Then |V (T (2q) )| ≥ 2q + 1, and T ′ := T − V (T (2q) ) is a caterpillar.
⌈√ ⌉
Note that |V (T ′ )| ≤ q2 + p − (2q + 1) ≤ q2 as p ≤ 2q + 1. Then b(T ′ ) ≤ |V (T ′ )| ≤ q by the minimality of T .
On the other hand, we note that V (T (2q) ) ⊆ Nq [vq ], and hence, by Theorem 1.1, we have

b(T ) ≤ 1 + b(T ′ ) ≤ q + 1,

which is a contradiction. ■

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx 3

Fig. 2. A caterpillar T with the spine ST = v1 v2 · · · vr .

l ,l
Fig. 3. Trl,s and Tr 1,s,t2 .

Fact 2. T (2q−1) ∼
= P2q .

Proof of Fact 2. Assume to the contrary that T (2q−1) ̸ ∼ = P2q , then |V (T (2q−1) )| ≥⌈√
2q + 1 and ′
⌉ T := T − V (T
(2q−1)
) is a
caterpillar with |V (T ′ )| ≤ q2 + p − (2q + 1) ≤ q2 as p ≤ 2q + 1, and thus b(T ′ ) ≤ |V (T ′ )| ≤ q by the minimality of T .
Hence
b(T ) ≤ 1 + b(T ′ ) ≤ q + 1
by Theorem 1.1, which is a contradiction. ■
If r ≤ 4q − 2, then by Fact 2, we have V (T ) = Nq [v3q−1 ] ∪ Nq−1 [vq−1 ], and thus b(T ) ≤ q + 1, a contradiction. So
r ≥ 4q − 1.
By Fact 1, |V (T (4q−2) )| ≥ 4q. Note that V (T (4q−2) ) ⊆ Nq [v3q−1 ]∪ Nq−1 [vq−1 ] by Fact 2 (see Fig. 2, the red vertices represent
Nq [v3q−1 ] and the blue vertices represent ⌈ N√q−1 [vq−1 ⌉]). Thus T ′ := T − V (T (4q−2) ) is a caterpillar with |V (T ′ )| ≤ q2 + p − 4q ≤
2 ′ ′
(q − 1) as p ≤ 2q + 1, and thus b(T ) ≤ |V (T )| ≤ q − 1 by the minimality of T . Hence
b(T ) ≤ 2 + b(T ′ ) ≤ q + 1
by Theorem 1.1, which is a contradiction.
This completes the proof of Theorem 2.1. ■

Remark 1. Note that our proof of Theorem 2.1 is algorithmic: one can easily follow the steps of the proof and give a
burning sequence of the input graph in polynomial time.
Though Conjecture 1.3 is true for caterpillars. The following burning problem is NP-complete for caterpillars. The proof
of the NP-completeness is more tedious than enlightening (our proof is heavily influenced by the method of Bessy, Bonato
and Janssen [1] for proving the NP-completeness of the burning problem for trees with maximum degree three), so we
postpone it to the Appendix.
Problem: Graph Burning
Instance: A simple graph G of order n and an integer k ≥ 2.
Question: Is b(G) ≤ k? In other words, does G contains a burning sequence (x1 , x2 , . . . , xk )?

Theorem 2.2. The burning problem is NP-complete for caterpillars.

l ,l
3. Burning number of Trl ,s and Tr 1,s,2t

Let T be a caterpillar with the spine ST = v1 v2 · · · vr , where r ≥ 2. Suppose that v0 is a leaf neighbor of v1 and vr +1 is
a leaf neighbor of vr . If T has only one stem vs and vs has l leaf neighbors, T is also denoted by Trl,s , where 1 ≤ s ≤ ⌈ 2r ⌉
(see Fig. 3(a)). In particular, Trl,1 is a broom. If T has exactly two stems vs , vt , and vs has l1 leaf neighbors, vt has l2 leaf
l ,l
neighbors, T is also denoted by Tr 1,s,t2 , where 1 ≤ s < t ≤ r (see Fig. 3(b)). Assume, without loss of generality, that
l ,l
s − 1 ≤ r − t. Particularly, Tr 1,1,2r is a double broom.
Recently, Liu et al. [12] presented the following upper bound on the burning number of linear forests with two
components, which will be useful in the proof of the upper bound on the burning number of caterpillars with at most
two stems.

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
4 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx

Lemma 3.1 ([12]). Let G = Pa1 ∪ Pa2 with a1 ≥ a2 ≥ 1 and a1 + a2 ≤ q2 .

(1) If a1 + a2 ≤ q2 − 1, then b(G) ≤ q;
(2) If a1 + a2 = q2 and a2 ̸ = 2, then b(G) ≤ q.

3.1. Burning number of Trl,s

In this subsection, we will determine the burning number of Trl,s .

⌈√ ⌉
Theorem 3.2. For r ≥ 2, we have b(Trl,s ) = r +2 .

Proof. Suppose that r + 2 ⌈= q2 +⌉ p, where 1 ≤ p ≤ 2q + 1. If r = 2, then s = ⌈1 as 1 ⌉

√ √ ≤ s ≤ ⌈ 2r ⌉. By
l
Theorem 1.4, b(T2,1 ) = 2 = r + 2 . Therefore we may assume r ≥ 3. Then q ≥ 2 and r + 2 = q + 1. Note
that b(Trl,s ) ≥ b(Pr +2 ) = q + 1, and hence, in the following, we will show that b(Trl,s ) ≤ q + 1.
Suppose to the contrary that Trl,s is a minimal counterexample. If r ≤ 2q − 1, then V (Trl,s ) = Nq [vq ], and thus
b(Trl,s ) ≤ q + 1, a contradiction. So r ≥ 2q.
First we assume s ≤ 2q − 1, let x1 = vq , then the tree induced by Nq [x1 ] is a T2ql ′ l
−1,s , and T := Tr ,s − N q [x1 ] is a path
2 2
⌈√ ⌉
Pr +2−(2q+1) . Since p ≤ 2q + 1, we have |V (T )| = r + 2 − (2q + 1) = q + p − (2q + 1) ≤ q , and thus b(T ′ ) ≤
′
|V (T ′ )| ≤ q
l ′
by Theorem 1.2. Hence b(Tr ,s ) ≤ 1 + b(T ) ≤ q + 1 by Theorem 1.1, which is a contradiction.
Now we assume s ≥ 2q. Then r ≥ 4q − 1 as s ≤ ⌈ 2r ⌉. Recall that r + 2 = q2 + p ≤ (q + 1)2 . Hence 4q + 1 ≤ r + 2 ≤ (q + 1)2 ,
i.e., q ≥ 2. Moreover, if q = 2, then r + 2 = 9 and s = 4. It is easy to check that (v3 , v7 , v0 ) is a burning sequence of T7l ,4 ,
and hence b(T7l ,4 ) = 3 = q + 1, which is a contradiction. Therefore q ≥ 3. Set x1 = vs , then the tree induced by Nq [x1 ] is
l ′ l ′ 2 2
a T2q −1,q , and T := Tr ,s − Nq [x1 ] = Ps−q ∪ Pr +1−s−q . Thus |V (T )| = r + 1 − 2q = q + p − 1 − 2q ≤ q as p ≤ 2q + 1.
r
Since s − q ≥ q ≥ 3 and r + 1 − s − q ≥ ⌊ 2 ⌋ + 1 − q ≥ q ≥ 3, then b(Ps−q ∪ Pr +1−s−q ) ≤ q by Lemma 3.1, and thus
b(Trl,s ) ≤ 1 + b(T ′ ) ≤ q + 1 by Theorem 1.1, which is a contradiction.
Therefore, we complete the proof of Theorem 3.2. ■

l ,l
3.2. Burning number of Tr 1,s,t2

l ,l
In this subsection, we will establish the exact value of the burning number of Tr 1,s,t2 for r ≥ 2. Note that if r = 2, then
s = 1 and t = 2.
l ,l l ,l
First we determine the burning numbers of two graphs T21,1,22 and T71,3,25 .

l ,l l ,l
Theorem 3.3. b(T21,1,22 ) = 3 and b(T71,3,25 ) = 4.

l ,l l ,l
Proof. First we show that b(T21,1,22 ) = 3. By Theorem 1.4, b(T21,1,22 ) ≥ 3. On the other hand, (v1 , v2 , v3 ) is a burning sequence
l ,l l ,l
for T21,1,22 . Then b(T21,1,22 ) = 3.
l ,l l ,l
Next, we show that b(T71,3,25 ) = 4. Let (x1 , x2 , . . . , xk ) be an optimal burning sequence for T71,3,25 . By Theorem 1.4,
k ≥ 3. In the following, we show that k ≥ 4. Assume to the contrary that k = 3. Then N2 [x1 ] ∪ N1 [x2 ] ∪ N0 [x3 ] =
l ,l
V (T71,3,25 ) by Theorem 1.1. In particular, {v0 , v1 , . . . , v8 } ⊆ N2 [x1 ] ∪ N1 [x2 ] ∪ N0 [x3 ]. Then |N2 [x1 ] ∩ {v0 , v1 , . . . , v8 }| = 5,
|N1 [x2 ] ∩ {v0 , v1 , . . . , v8 }| = 3 and |N0 [x3 ] ∩ {v0 , v1 , . . . , v8 }| = 1. Thus {x1 , x2 , x3 } ⊆ {v0 , v1 , . . . , v8 }, and x1 ̸= v4 . Hence
L(v3 ) ⊆ N1 [x2 ] or L(v5 ) ⊆ N1 [x2 ], say L(v3 ) ⊆ N1 [x2 ], that is, x2 = v3 . Then L(v5 ) ⊆ N2 [x1 ], that is, x1 ∈ {v4 , v5 , v6 }. Note
that N1 [x2 ] ∩ {v0 , v1 , . . . , v8 } = {v2 , v3 , v4 } as |N1 [x2 ] ∩ {v0 , v1 , . . . , v8 }| = 3, x3 ∈ {v0 , v1 }, say x3 = vi for some 0 ≤ i ≤ 1.
But now v1−i ∈ / N2 [x1 ] ∪ N1 [x2 ] ∪ N0 [x3 ], which is a contradiction.
l ,l l ,l
On the other hand, we note that (v3 , v6 , v7 , v8 ) is a burning sequence for T71,3,25 , thus b(T71,3,25 ) = 4. ■
l ,l
Now we determine the burning number of Tr 1,s,t2 for r ≥ 3 and (r , s, t) ̸ = (7, 3, 5).

Theorem 3.4. For r ≥ 3 and (r , s, t) ̸ = (7, 3, 5), we have

⌈√
l ,l
⌉
b(Tr 1,s,t2 ) = r +2 .

⌈√ ⌉
Proof. Suppose that r + 2 = q2 + p, where 1 ≤ p ≤ 2q + 1. Then q ≥ 2 and r +2 = q + 1. Note that
l ,l s,t
b(Tr 1,s,t2 ) ≥ b(Pr +2 ) = q + 1, we only need to show that b(Tn,l ) ≤ q + 1 in the following. Suppose to the contrary
1 ,l2
l ,l l ,l l ,l
that Tr 1,s,t2 is a counterexample of minimal order for which b(Tr 1,s,t2 ) is maximal. If r ≤ 2q − 1, then V (Tr 1,s,t2 ) = Nq [vq ], and
l ,l
thus b(Tr 1,s,t2 ) ≤ q + 1, a contradiction. So r ≥ 2q. We consider three cases.

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx 5

Case 1. s ≥ 2q + 2.
In this case, t ≥ 2q + 3 and r − t ≥ 2q + 1, then r ≥ 4q + 4 ≥ 12.
l ,l l ,l2
Set x1 = vq , then the tree induced by Nq [x1 ] is a P2q+1 , and T ′ := Tr 1,s,t2 − Nq [x1 ] = Tr 1−(2q+1),s−(2q+1),t −(2q+1) is a caterpillar
with spine ST = v2q+2 v2q+3 · · · vr . Since p ≤ 2q + 1, |V (ST )| = r − (2q + 1) = (q + p − 2) − (2q + 1) ≤ q2 − 2. Note
′ ′ 2
l ,l
that |V (ST ′ )| = r − (2q + 1) ≥ 2q + 3 ≥ 7 and (r − (2q + 1)) − (t − (2q + 1)) = r − t ≥ 2q + 1 ≥ 5, then T ′ ̸ = T71,3,25
⌈√ ⌉ l ,l l ,l
if |V (ST ′ )| = 7. Thus b(T ′ ) ≤ |V (ST ′ )| + 2 ≤ q by the minimality of Tr 1,s,t2 . Hence b(Tr 1,s,t2 ) ≤ 1 + b(T ′ ) ≤ q + 1 by
Theorem 1.1, which is a contradiction.

Case 2. 2q ≤ s ≤ 2q + 1.
In this case, since r ≥ s + t − 1, r ≥ 2s ≥ 4q ≥ 8. We will consider three subcases according to the value of t.

Subcase 2.1. t ≥ 4q + 1.
l
Set x1 = v3q−1 and x2 = vq−1 . Then the tree induced by Nq [x1 ] is a T2q1 −1,s−(2q−1) , the tree induced by Nq−1 [x2 ] is a P2q−1 ,
l ,l l
and T ′ := Tr 1,s,t2 − Nq [x1 ] − Nq−1 [x2 ] = Tr 2−4q,t −4q is a caterpillar with spine ST ′ = v4q+1 v4q+2 · · · vr . Since p ≤ 2q + 1, then
⌈√ ⌉
|V (ST ′ )| = r − 4q = (q2 + p − 2) − 4q ≤ (q − 1)2 − 2, and thus b(T ′ ) ≤ |V (ST ′ )| + 2 ≤ q − 1 by Theorem 3.2. Hence
l ,l
b(Tr 1,s,t2 ) ≤ 2 + b(T ′ ) ≤ q + 1, by Theorem 1.1, which is a contradiction.

Subcase 2.2. 4q − 1 ≤ t ≤ 4q.

l
In this case, set x1 = vq+3 , x2 = v3q+3 , then the tree induced by Nq [x1 ] is a T2q1 −1,s−3 , the tree induced by Nq−1 [x2 ] is
l l ,l
a T2q2 −3,t −(2q+1)−3 , and T ′ := Tr 1,s,t2 − Nq [x1 ] − Nq−1 [x2 ] = P3 ∪ Pr −4q−1 . Since p ≤ 2q + 1, then |V (T ′ )| = r + 2 − 4q =
2 2
⌈√ ⌉ l ,l
(q + p) − 4q ≤ (q − 1) , and thus b(T ′ ) ≤ |V (T ′ )| ≤ q − 1 by Lemma 3.1. Hence b(Tr 1,s,t2 ) ≤ 2 + b(T ′ ) ≤ q + 1 by
Theorem 1.1, which is a contradiction.

Subcase 2.3. t ≤ 4q − 2.
l ,l
In this case, set x1 = v3q−1 and x2 = vq−1 , then the tree induced by Nq [x1 ] is a T2q1 −21,s−(2q−1),t −(2q−1) , the tree induced
l ,l
by Nq−1 [x2 ] is a path P2q−1 , and T ′ := Tr 1,s,t2 − Nq [x1 ] − Nq−1 [x2 ] = Pr +2−4q . Since p ≤ 2q + 1, then |V (T ′ )| = r + 2 − 4q =
⌈√ ⌉ l ,l
(q2 + p) − 4q ≤ (q − 1)2 , and thus b(T ′ ) ≤ |V (T ′ )| ≤ q − 1 by Theorem 1.2. Hence b(Tr 1,s,t2 ) ≤ 2 + b(T ′ ) ≤ q + 1 by
Theorem 1.1, which is a contradiction.

Case 3. s ≤ 2q − 1.
In this case, we consider the following three subcases according to the value of t.

Subcase 3.1. t ≥ 2q + 2.
l l ,l
In this subcase, t ≥ 2q + 2 ≥ 6. Set x1 = vq , then the tree induced by Nq [x1 ] is a T2q1 −1,s , and T ′ := Tr 1,s,t2 − Nq [x1 ] =
l
Tr 2−(2q+1),t −(2q+1) is a caterpillar with spine ST ′ = v2q+2 v2q+3 · · · vr . Since p ≤ 2q + 1, then |V (ST ′ )| = r − (2q + 1) =
⌈√ ⌉ l ,l
(q2 + p − 2) − (2q + 1) ≤ q2 − 2, and thus b(T ′ ) ≤ |V (ST ′ )| + 2 ≤ q by Theorem 3.2. Hence b(Tr 1,s,t2 ) ≤ 1 + b(T ′ ) ≤ q + 1
by Theorem 1.1, which is a contradiction.

Subcase 3.2. 2q ≤ t ≤ 2q + 1.
l
If s ≤ 2q − 3, set x1 = v3q−1 and x2 = vq−1 , then the tree induced by Nq [x1 ] is a T2q2 −1,t −(2q−1) , the tree induced by
l l ,l
Nq−1 [x2 ] is a T2q1 −3,s , and T ′ := Tr 1,s,t2 − Nq [x1 ] − Nq−1 [x2 ] = Pr +2−4q . Since p ≤ 2q + 1, then |V (T ′ )| = r + 2 − 4q =
⌈√ ⌉ l ,l
(q2 + p) − 4q ≤ (q − 1)2 , and thus b(T ′ ) ≤ |V (T ′ )| ≤ q − 1 by Theorem 1.2. Hence b(Tr 1,s,t2 ) ≤ 2 + b(T ′ ) ≤ q + 1 by
Theorem 1.1, which is a contradiction.
Therefore s ≥ 2q − 2. If q = 2, then 2 ≤ s ≤ 3, 4 ≤ t ≤ 5 and r = q2 + p − 2 ≤ (q + 1)2 − 2 = 7. If
l ,l
s = 2 and t = 4, then r ≥ s + t − 1 = 5. Now (v3 , vr , v0 ) is a burning sequence for Tr 1,2,24 . If s = 2 and t = 5, then
l ,l
r ≥ s + t − 1 = 6, and thus (v6 , v2 , v0 ) is a burning sequence for Tr 1,2,24 . If s = 3, then t = 4 as (r , s, t) ̸ = (7, 3, 5). In this
l ,l l ,l
case, r ≥ s + t − 1 = 6, and (v3 , v7 , v0 ) is a burning sequence for Tr 1,3,24 . So in either case, we have b(Tr 1,s,t2 ) ≤ 3 = q + 1,
l ,l
which is a contradiction. Hence, we may assume q ≥ 3. Set x1 = vq+3 , then the tree induced by Nq [x1 ] is a T2q1 −21,s−3,t −3 ,
l ,l
and T := ′
− Nq [x1 ] = P3 ∪ Pr −1−(2q+1) . Since p ≤ 2q + 1, then |V (T )| = r + 2 − (2q + 1) = q + p − (2q + 1) ≤ q2 , and
Tr 1,s,t2 ′ 2
⌈√ ⌉ l ,l
thus b(T ′ ) ≤ |V (T ′ )| ≤ q by Lemma 3.1. Hence b(Tr 1,s,t2 ) ≤ 1 + b(T ′ ) ≤ q + 1 by Theorem 1.1, which is a contradiction.

Subcase 3.3. t ≤ 2q − 1.
l ,l l ,l
Set x1 = vq , then the tree induced by Nq [x1 ] is a T2q1 −21,s,t , and T ′ := Tr 1,s,t2 − Nq [x1 ] = Pr +2−(2q+1) . Since p ≤ 2q + 1,
2 2
⌈√ ⌉
then |V (T )| = r + 2 − (2q + 1) = (q + p) − (2q + 1) ≤ q , and thus b(T ′ ) ≤
′
|V (T ′ )| ≤ q by Theorem 1.2. Hence
l ,l
b(Tr 1,s,t2 ) ≤ 1 + b(T ′ ) ≤ q + 1 by Theorem 1.1, which is a contradiction.
Therefore, we complete the proof of Theorem 3.4. ■

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
6 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx

Fig. 4. Burning sequence of CTr .

4. Burning number of CTr

In this section, we will determine the burning number of a subclass of the class of caterpillars all of whose spine
vertices are stems.
Let CTr be a caterpillar with the spine P = v1 v2 · · · vr such that d(vi ) ≥ 3 for i ∈ {1, 2, . . . , r − 2, r } and d(vr −1 ) = 3.
l ,l
In particular, if d(v1 ) = d(v2 ) = · · · = d(vr ) = 3, then CTr is a comb graph Combr . If r = 2, then CT2 is a T21,1,22 , so for
the remainder of this section, we always assume that r ≥ 3. Note that since d(vi ) ≥ 3, the vertex v1 has at least two leaf
neighbors, v0 and v1′ ; the vertex vr has at least two leaf neighbors, vr +1 and vr′ ; and the vertex vi has at least one leaf
neighbor, vi′ for 2√ ≤ i ≤ r − 1 (see Fig. 4). √
Assume k = ⌈ r − 1⌉ + 1, then k − 1 ≥ r − 1, i.e., r − 1 ≤ (k − 1)2 . Set xk = vr′ −1 , xk−1 = vr , and xk−i = vr −1−i2 +i
for 2 ≤ i ≤ k − 2. Set x1 = vr −1−(k−1)2 +(k−1) if r ≥ (k − 1)2 − k + 3, and set x1 = v0 otherwise (see Fig. 4).
Since d(vr −1 ) = 3, vr −1 has only one leaf neighbor vr′ −1 . Note that vr′ −1 = N0 [xk ], {vr −1 , vr } ∪ L(vr ) ⊆ N1 [xk−1 ],
{vr −2 }∪ L(vr −2 ) ⊆ N2 [xk−2 ]. For 2 ≤ i ≤ k − 3, we have d(xk−i , xk−(i+1) ) = (r − 1 − i2 + i) − (r − 1 − (i + 1)2 + (i + 1)) = 2i, then
{vr −1−(i+1)2 +(i+1) , . . . , vr −1−i2 +i } ∪ L(vr −1−(i+1)2 +(i+1) ) ∪ · · · ∪ L(vr −1−i2 +i ) ⊆ Ni [xk−i ] ∪ Ni+1 [xk−(i+1) ]. If x1 = vr −1−(k−1)2 +(k−1) ,
then d(x1 , x2 ) = 2(k − 2), thus {vr −1−(k−1)2 +(k−1) , . . . , vr −1−(k−2)2 +(k−2) } ∪ L(vr −1−(k−1)2 +(k−1) ) ∪ · · · ∪ L(vr −1−(k−2)2 +(k−2) ) ⊆
Nk−2 [x2 ] ∪ Nk−1 [x1 ]. Recall that r − 1 ≤ (k − 1)2 , then d(v0 , x1 ) = r − 1 − (k − 1)2 + (k − 1) ≤ k − 1, thus
{v1 , . . . , vr −2−(k−1)2 +(k−1) }∪ L(v1 ) ∪· · ·∪ L(vr −2−(k−1)2 +(k−1) ) ⊆ Nk−1 [x1 ]. If x1 = v0 , then d(x1 , x2 ) = r − 1 − (k − 2)2 + (k − 2) ≤
2(k − 2) as r ≤ (k − 1)2 − k + 2, and thus {v1 , . . . , vr −2−(k−2)2 +(k−2) } ∪ L(v1 ) ∪ · · · ∪ L(vr −2−(k−2)2 +(k−2) ) ⊆ Nk−1 [x1 ] ∪ Nk−2 [x2 ].
So, in either case, we have V (CTr ) = N0 [xk ] ∪ N1 [xk−1 ] ∪ · · · ∪ Nk−1 [x1 ]. Thus (x1 , x2 , . . . , xk ) is a burning sequence for CTr
by Theorem 1.1. Hence, we have the following result.
⌈√ ⌉
Theorem 4.1. For r ≥ 3, we have b(CTr ) ≤ r − 1 + 1.

Remark
⌈√ ⌉2. Let Tr be a caterpillar with the spine P = v1 v2 · · · vr such that d(vi ) ≥ 3 for 1 ≤ i ≤ r. If d(v2 ) = 3, then
⌈√r − 1⌉ + 1 is also an upper bound for b(Tr ) as we can relabel the spine vertices. But⌈if√d(vr −1⌉) > 3 and d(v2 ) > 3,
r − 1 + 1 is not an upper bound in general. For example, if d(vr −4 ) = 3, then k = r − 1 + 1 is still an upper
bound for b(Tr ) as (x1 , x2 , . . . , xk ) is a burning sequence for Tr , where xk = vr′ −4 , xk−1 = vr , xk−2 = vr −2 , xk−i = vr −1−i2 +i
for 3 ≤ i ≤ k − 2, and x1 = vr −1−(k−1)2 +(k−1) if r ≥ (k − 1)2 − k + 3, and x1 = v0 otherwise. Another example, if r = 5 and
⌈√ ⌈√
d(vi ) = 4 for 1 ≤ i ≤ 5, then
⌉ ⌉
r − 1 + 1 is not an upper bound for b(Tr ) as b(Tr ) = 4 but r − 1 + 1 = 3.
Let (x1 , x2 , . . . , xk ) be a burning sequence for CTr , where k ≥ 3. Define a surjective function f : V (CTr ) →
{x1 , x2 , . . . , xk }, with f (w) = xk−i if w ∈ Ni [xk−i ]. Then the tree induced by f −1 (xk−i ) is a connected subgraph Ti , and
V (CTr ) = ∪ki=−01 V (Ti ). Note that the distance between each vertex of Ti and xk−i is at most i for 1 ≤ i ≤ k − 1. If there
exists i0 and j0 > i0 such that d(xi0 , xj0 ) = 1, then V (Tk−j0 ) ⊆ V (Tk−i0 ), and thus V (CTr ) = ∪ki=−01 V (Ti ) \ V (Tk−j0 ). Let
{xi1 , xi2 , . . . , xil } ⊆ {x1 , x2 , . . . , xk } such that V (Tk−ij1 ) ̸⊆ V (Tk−ij2 ) and V (Tk−ij2 ) ̸⊆ V (Tk−ij1 ) for 1 ≤ j1 ̸= j2 ≤ l, and
V (CTr ) = ∪lj=1 V (Tk−ij ). Then d(xij , xij+1 ) ≥ 2 for 1 ≤ j ≤ l − 1, and xk is a leaf of CTr if xk ∈ {xi1 , xi2 , . . . , xil }.
Denote A = {x1 , x2 , . . . , xk }, and B = {xi1 , xi2 , . . . , xil }. Let L = {xi : xi is a leaf of CTr , 1 ≤ i ≤ k}, BL = B ∩ L and
BS = B \ BL , and let S = A \ L. Suppose that xij = vaj for xij ∈ BS , and xij = va′ j for xij ∈ BL . Assume without loss of generality
that a1 < a2 < · · · < al . Note that {vaj , vaj +1 , . . . , vaj+1 , va′ j , va′ +1 , . . . , va′ j+1 } ⊆ V (Tk−ij ) ∪ V (Tk−ij+1 ). If xij , xij+1 ∈ BS , then
j

we say Tk−ij and Tk−ij+1 are adjacent. We write Tk−ij ∼ Tk−ij+1 for ‘‘Tk−ij is adjacent to Tk−ij+1 ’’.
For example, let (x1 , x2 , . . . , xk ) be the burning sequence of CTr described in the previous paragraph. Then Ti ∼ Ti+1
for 2 ≤ i ≤ k − 2, and |V (Ti ) ∩ V (Ti+1 )| = 2 for 2 ≤ i ≤ k − 3, |V (Tk−2 ) ∩ V (Tk−1 )| = 2 if r ≥ (k − 1)2 − k + 3 and
|V (Tk−2 ) ∩ V (Tk−1 )| = 2(k − 2)2 + 2(k − 2) − 2r + 4 otherwise. Another example, let (v3 , v6 , v7 , v8 ) be a burning sequence
l ,l
for T71,3,25 . Note that N0 [v8 ] ⊆ N1 [v7 ] ⊆ N2 [v6 ], V (CTr ) = V (T1 ) ∪ V (T2 ), T1 ∼ T2 and |V (T1 ) ∩ V (T2 )| = 3 + l2 .
In the following, we consider a lower bound for the burning number of the comb graph Combr .
⌈√ ⌉
Theorem 4.2. For r ≥ 3, we have b(Combr ) ≥ r − 1 + 1.

Proof. Suppose that (x1 , x2 , . . . , xk ) is an optimal burning sequence for Combr . Then k ≥ 3. We may assume without
loss of generality that xi ∈ S for 1 ≤ i ≤ k − 1. Otherwise, there exists i (1 ≤ i ≤ k − 1) such that xi = va′ i ∈ L, then
(x1 , . . . , xi−1 , vai , xi+1 , xk ) is still a burning sequence for Combr .

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx 7

Let {xi1 , xi2 , . . . , xil } ⊆ {x1 , x2 , . . . , xk } such that V (Tk−ij ) ̸ ⊆ V (Tk−ij ) and V (Tk−ij ) ̸ ⊆ V (Tk−ij ) for 1 ≤ j1 ̸ = j2 ≤ l, and
1 2 2 ∑l 1 ∑k−1
V (Combr ) = ∪lj=1 V (Tk−ij ). Note that |V (T0 )| = 1, |V (Ti )| ≤ 4i for 1 ≤ i ≤ k − 1. Then |V (Tk−ij )| ≤ 1 + j=1 i=k−l+1 4i =
∑l ∑k−1
4kl − 4k − 2l2 + 2l + 1 if xk ∈ {xi1 , xi2 , . . . , xil } and j=1 |V (Tk −ij
)| ≤ i=k−l 4i = 4kl − 2l − 2l otherwise. 2

Fact 3. For any 1 ≤ j ≤ l − 1, |V (Tk−ij ) ∩ V (Tk−ij+1 )| ≥ 2 if Tk−ij ∼ Tk−ij+1 .

Proof of Fact 3. If d(xij , xij+1 ) = 2, then it is easy to check that |V (Tk−ij ) ∩ V (Tk−ij+1 )| ≥ 2. So we assume d(xij , xij+1 ) ≥ 3.
If va′ −1 ∈ V (Tk−ij ), then xij+1 = vaj+1 ∈ V (Tk−ij ) as d(va′ −1 , xij ) = d(vaj+1 , xij ). Thus vaj+1 , vaj+1 −1 ∈ V (Tk−ij ) ∩ V (Tk−ij+1 ),
j+1 j+1
i.e., |V (Tk−ij ) ∩ V (Tk−ij+1 )| ≥ 2. Therefore va′ −1 ∈
/ V (Tk−ij ). Since va′ j ∈ V (Tk−ij ), there exists at (aij ≤ at ≤ aij+1 − 2) such
j +1

/ V (Tk−ij ). Then va′ t +1 ∈ V (Tk−ij+1 ). Note that d(va′ t , xij ) = d(vat +1 , xij ) and d(vat , xij+1 ) =
that va′ t ∈ V (Tk−ij ) and va′ t +1 ∈
d(va′ t +1 , xij+1 ), then vat , vat +1 ∈ V (Tk−ij ) ∩ V (Tk−ij+1 ), i.e., |V (Tk−ij ) ∩ V (Tk−ij+1 )| ≥ 2. ■

|V (Tk−ij ) ∩ V (Tk−ij+1 )|. Note that if xk ∈ {xi1 , . . . , xil }, then xk ∈ L, l ≤ k, and there are l − 3
∑
Denote m = Tk−i ∼Tk−i
j j +1
∑l
pairs of adjacent trees, thus m ≥ 2(l − 3) by Fact 3. Hence j=1 |V (Tk−ij )|− m ≤ 4kl − 4k − 2l2 + 2l + 1 − 2(l − 3) ≤ 2k2 − 4k + 7.

∑l / {xi1 , xi2 , . . . , xil }, then l ≤2 k − 1, and there are l 2− 1 pairs of adjacent trees, thus m ≥ 2(l − 1) by Fact 3. Hence
If xk ∈
|V (Tk−ij )| − m ≤ 4kl − 2l − 2l − 2(l − 1) ≤ 2k − 4k + 4. Since 2r + 2 is even, and 2r + 2 = |V (Combr )| ≤
∑jl=1 2 2 2
ij )| − m, we have 2r + 2 ≤ 2k − 4k + 6 in either case. Thus k − 2k + 3 ≥ r + 1, i.e., (k − 1) ≥ r − 1, and
j=1 |V (Tk−√
hence k ≥ ⌈ r − 1⌉ + 1. ■
By Theorems 4.1 and 4.2, the fact that b(CTr ) ≥ b(Combr ), we can determine the burning number of CTr .
⌈√ ⌉
Theorem 4.3. For r ≥ 3, we have b(CTr ) = r − 1 + 1.

CRediT authorship contribution statement

Huiqing Liu: Supervision. Xuejiao Hu: Writing - original draft. Xiaolan Hu: Writing - review & editing.

Acknowledgments

The authors would like to thank the anonymous referees for their valuable comments and suggestions.

Appendix. Proof of Theorem 2.2

Let T be a caterpillar and let ST = v1 v2 · · · vr be the spine of T . If r is odd, then the center of T is {v r +1 }, and if r
2
is even, then the center of T is {v r , v r +1 }. Every vertex in the center of T is called a central vertex of T . Every vertex in
2 2

L(v1 ) ∪ L(vr ) is called an end-vertex of T .

Theorem 2.2. The burning problem is NP-complete for caterpillars.

Proof. Given a graph G of order n and a sequence (x1 , x2 , . . . , xk ) of the vertices

⋃k in G, we can easily find Nk−i [xi ] in
polynomial time for 1 ≤ i ≤ k. Thus we can check in polynomial time if V (G) = i=1 Nk−i [xi ]. Hence the burning problem
is in NP.
In the following, we show the NP-completeness of the burning problem for caterpillars by a reduction from the
following variant of the 3-Partition problem.
Problem: Distinct 3-Partition ∑3n
Instance: A finite set X = {a1 , a2 , . . . , a3n } of positive distinct integers, and a positive integer B such that i=1 ai = nB,
and B/4 < ai < B/2 for 1 ≤ i ≤ 3n.
Question: Is there any partition of X into n triples such that the elements in each triple add up to B.
Suppose we are given an instance of the Distinct 3-Partition problem, that is, we ∑are given a non-empty finite set
3n
X = {a1 , a2 , . . . , a3n } of distinct positive integers and a positive integer B such that i=1 i = nB, and B/4 < ai < B/2
a
for 1 ≤ i ≤ 3n. Since the Distinct 3-Partition problem is NP-complete in the strong sense (see [8,10]), without loss of
generality we assume that B is bounded above by a polynomial in the length of the input. Assume that the maximum of
the set X is m which is by assumption bounded above by B/2. Then |X | ≤ m ≤ B/2. We now construct a caterpillar as
follows.
Let Om = {1, 3, . . . , 2m − 1}, Y = {2a∑ i − 1 : ai ∈ X }, Z = Om \Y , and |Z | = k. Then Y ⊆ Om , 3n = |X | = |Y | ≤ m,
|Z | = m − |Y | = m − 3n and 2nB − 3n = 3n i=1 (2ai − 1) is the sum of the numbers in Y . For 1∑ ≤ i ≤ k, let Qi′ be a path of
m
order li , where li is the i’th largest number in Z . Note that the sum of the numbers in Y ∪ Z is i=1 (2i − 1) = m2 , then the

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
8 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx

sum of the numbers in Z is m2 − 2nB + 3n. For 1 ≤ i ≤ m + 1, we define Ti to be a caterpillar all of whose spine vertices
are stems and STi = vi1 vi2 . . . vi2(2m+1−i)−1 , that is, |V (STi )| = 2(2m + 1 − i) − 1. Let yi = vi2m+1−i be the central vertex of Ti .
The longest path of Ti , which denoted by DTi , is of order 2(2m + 1 − i) + 1 ∈ Om . We take Qi to be a path of order 2B − 3,
for 1 ≤ i ≤ n. Then we combine the graphs that we created above from left to right in the following order:

Q1 , T1 , Q2 , T2 , . . . , Qn , Tn , Q1′ , Tn+1 , Q2′ , Tn+2 , . . . , Qk′ , Tn+k , Tn+k+1 , . . . , Tm+1

such that each graph in this order is joined by an edge from one of its end-vertices to an end-vertices of the next graph
in the presented order. The resulting graph is named C (X ); note that it is a caterpillar.
Let P be the longest path of C (X ), then
n k m+1
∑ ∑ ∑
|V (P)| = |Qi | + |Qi′ | + |DTi |
i=1 i=1 i=1
m+1
∑
= n(2B − 3) + m2 − 2nB + 3n + (2(2m + 1 − i) + 1)
i=1

= (2m + 1)2 .
Since m is bounded by B and by assumption, B is bounded above by a polynomial in terms of n, then C (X ) is obtained
in polynomial time in terms of the length of the input. Now we need to show that, there is a partition of X into n triples
such that the numbers in each triple add up to B if and only if b(C (X )) ≤ 2m + 1. Note that there is a partition of X into
triples such that the elements in each triple add up to B if and only if we can decompose the paths Q1 , Q2 , . . . , Qn into
subpaths of orders {2ai − 1 : ai ∈ X }.
First, assume that there is a partition of X into n triples such that the numbers in each triple add up to B. That is,
Q1 , Q2 , . . . , Qn can be partitioned
⋃into smaller
⋃ ⋃kpaths of orders {2ai − 1 : ai ∈ X }. Since Om = Y ∪ Z , we conclude that there
n
is a partition for the subgraph ( i=1 Qi ) ( i=1 Qi′ ) in terms of the subpaths {Pl : l ∈ Om }. Now, for m + 2 ≤ i ≤ 2m + 1,
let xi be the central vertex of a path Pl in such a partition, where l = 2(2m + 1 − i) − 1 ∈ Om = Y ∪ Z . For 1 ≤ i ≤ m + 1,
let xi = yi (the central vertex of Ti ). Thus we have that
2m+1
⋃
V (C (X )) = N2m+1−i [xi ].
i=1

That is, (x1 , x2 , . . . , x2m+1 ) forms a burning sequence of C (X ). Hence b(C (X )) ≤ 2m + 1.

Conversely, suppose that b(C (X )) ≤ 2m + 1. Since C (X ) contains a path P of order (2m + 1)2 , b(C (X )) ≥ b(P) = 2m + 1
by Theorem 1.2. Thus b(C (X )) = 2m + 1. Assume that (x1 , x2 , . . . , x2m+1 ) is an optimal burning sequence for C (X ). Then
(x1 , x2 , . . . , x2m+1 ) is also an optimal burning sequence for P. Therefore, we have the following fact.

Fact 4. Each xi must be in P, and for each v ∈ P, there exists only one j (1 ≤ j ≤ 2m + 1) such that v ∈ Nk−j [xj ].

Fact 5. For 1 ≤ i ≤ m + 1, we have xi = yi (the central vertex of Ti ).

Proof of Fact 5. We prove this by induction on i. For i = 1, suppose that x1 ̸ = y1 . Since |V (DT1 )| = 4m + 1, there
exist i1 , i2 such that xi1 , xi2 ∈ V (DT1 ). By a similar argument as that of Fact 3, there exists v ∈ V (DT1 ) such that
v ∈ N2m+1−i1 [xi1 ] ∩ N2m+1−i2 [xi2 ], which is a contradiction to Fact 4. Therefore x1 = r1 . Assume that xj = yj for 1 ≤ j ≤ m.
Suppose that xj+1 ̸ = yj+1 . Since |V (DTj+1 )| = 4m − 2j + 1, there exist i1 and i2 such that xi1 , xi2 ∈ V (DTj+1 ). By a similar
argument as that of Fact 3, there exists v ∈ V (DTj+1 ) such that v ∈ N2m+1−i1 [xi1 ] ∩ N2m+1−i2 [xi2 ], which is a contradiction
to Fact 4. Therefore xj+1 = yj+1 . Hence Fact 5 is proved by induction. ■

By Fact 5, N2m+1−i [xi ] = V (Ti ) for 1 ≤ i ≤ m + 1. Denote W = ∪ni=1 V (Qi ) and W ′ = ∪ki=1 V (Qi′ ). Then (W ∪ W ′ ) ⊆
2m+1
∪ [xj ]. Since Qi (1 ≤ i ≤ n) and Qj′ (1 ≤ j ≤ k) are disjoint union of paths, then
j=m+2 N2m+1−j

N2m+1−i [xi ] ∩ (W ∪ W ′ )|≤ 2(2m + 1 − i) + 1 for m + 2 ≤ i ≤ 2m + 1.

∑m
− 1) = m2 = i2m
∑ +1
On the other hand, |W | + |W ′ | = i=1 (2m =m+2 (2(2m + 1 − i) + 1). Thus |N2m+1−i [xi ] ∩ (W ∪ W )| =
′

′ ′
2(2m + 1 − i) + 1. Hence there exists a partition of W ∪ W into subpaths {Pl : l ∈ Om }. If there is a path Qi of order
l ∈ Om \Y = Z that is partitioned by a union of paths of orders in Om rather than by Pl itself, then Pl must have covered some
part of a path Qj′ with j ̸ = i, or must be used in partitioning Q1 , Q2 , . . . , Qn . Hence, we can easily modify the partition by
switching the place of Pl and those paths that have covered Pl (as they have equal lengths). Therefore, we have decreased
the number of such displaced paths in our partition for W ∪ W ′ . Since the number of Qi′ ’s, where 1 ≤ i ≤ k, is finite, we
will end up after a finite number of switches with a partition for W ∪ W ′ in which every Qi′ , 1 ≤ i ≤ k, is covered by
itself. Therefore, we have the following fact.

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.
 H. Liu, X. Hu and X. Hu / Discrete Applied Mathematics xxx (xxxx) xxx 9

Fact 6. There is a partition of W ∪ W ′ into subpaths of orders in Om in which each path Qi (1 ≤ i ≤ n) is decomposed
into paths of orders in Y , and each path Qj′ (1 ≤ j ≤ k) is covered by itself.
Since each Qi is of order 2B − 3, there must be a partition of Y into triples such that the elements in each triple add up
to 2B − 3. Equivalently, there must be a partition of X into triples such that the elements in each triple add up to B. Since
C (X ) is a caterpillar, we have a polynomial time reduction from the Distinct 3-Partition problem to the Graph Burning
problem for caterpillars. ■

References

[1] S. Bessy, A. Bonato, J. Janssen, D. Rautenbach, E. Roshanbin, Burning a graph is hard, Discrete Appl. Math. 232 (2017) 73–87.
[2] S. Bessy, A. Bonato, J. Janssen, D. Rautenbach, E. Roshanbin, Bounds on the burning number, Discrete Appl. Math. 235 (2018) 16–22.
[3] A. Bonato, J. Janssen, E. Roshanbin, Burning a graph as a model of social contagion, Lecture Notes in Comput. Sci. 8882 (2014) 13–22.
[4] A. Bonato, J. Janssen, E. Roshanbin, How to burn a graph, Internet Math. 1–2 (2016) 85–100.
[5] A. Bonato, T. Lidbetter, Bounds on the burning numbers of spiders and path-forests, Theoret. Comput. Sci. (2018) http://dx.doi.org/10.1016/j.
tcs.2018.05.035.
[6] A. Bonato, P. Prałat, Graph Searching Games and Probabilistic Methods, CRC Press, 2017.
[7] S.L. Fitzpatrick, L. Wilm, Burning circulant graphs, arXiv:1706.03106.
[8] M.R. Garey, D.S. Johnson, Computers and Intractability: A Guide to the Theory of NP-Completeness, W.H. Freeman, 1979.
[9] F. Harary, A.J. Schwenk, The number of caterpillars, Discrete Math. 6 (1973) 359–365.
[10] H. Hulett, T.G. Will, G.J. Woeginger, Multigraph realizations of degree sequences: Maximization is easy, minimization is hard, Oper. Res. Lett.
36 (2008) 594–596.
[11] M.R. Land, L.Y. Lu, An upper bound on the burning number of graphs, in: A. Bonato, F. Graham, Pralat P. (Eds.), Algorithms and Models for the
Web Graph. WAW 2016, in: Lecture Notes in Computer Science, vol. 10088, Springer, Cham, 2016.
[12] H.Q. Liu, R.T. Zhang, X.L. Hu, Burning number of theta graphs, Appl. Math. Comput. 361 (2019) 246–257.
[13] D. Mitsche, P. Prałat, E. Roshanbin, Burning graphs: a probabilistic perspective, Graphs Combin. 33 (2017) 449–471.
[14] D. Mitsche, P. Prałat, E. Roshanbin, Burning number of graph products, Theoret. Comput. Sci. 746 (2018) 124–135.
[15] E. Roshanbin, Burning a Graph as a Model of Social Contagion (Ph.D. thesis), Dalhousie University, 2016.
[16] K.A. Sim, T.S. Tan, K.B. Wong, On the burning number of generalized Petersen graphs, Bull. Malays. Math. Sci. Soc. 41 (2018) 1657–1670.

Please cite this article as: H. Liu, X. Hu and X. Hu, Burning number of caterpillars, Discrete Applied Mathematics (2020),
https://doi.org/10.1016/j.dam.2020.03.062.