On the energy of digraphs

Juan R. Carmona
Facultad de Ciencias – Instituto de ciencias Fı́sicas y Matemáticas
Universidad Austral de Chile, Independencia 631 – Valdivia – Chile.

Abstract
Let D be a simple digraph with eigenvalues z1 , z2 , ..., zn . The energy of
D is defined as E(D) = ni=1 |Re(zi )|, is the real part of the eigenvalue zi .
P
In this paper a lower bound will be obtained for the spectral radius of D,
wich improves some the lower bounds that appear in the literature [10], [25].
This result allows us to obtain an upper bound for the energy of D. Finally,
digraphs are characterized in which this upper bound improves the bounds
given in [10] and [25].
Keywords: Energy of a digraph, upper bound, spectral radius, lower
bound.

1. Introduction and preliminaries

Let D = (V, Γ) denote a digraph of order n, where V = {v1 , v2 , ..., vn } is
the set of vertices of D with |V | = n, and Γ is the set of arcs consisting of
ordered pairs of distinct vertices. We only consider digraphs with no loops
and no miltiple arcs. Two vertices u and v of D are called adjacent if they
are connected by an arc (u, v) ∈ Γ, or (v, u) ∈ Γ, and doubly adjacent if
←→
(u, v), (v, u) ∈ Γ. For any vertex vi , let Ni = {vj ∈ V : (vi , vj ), (vj , vi ) ∈ Γ}
denote the vertices doubly adjacents of vi .
A walk π of length l from vertex u to vertex v is a sequence of vertices
π : u = u0 , u1 , ..., ul = v, where (uk−1 , uk ) is an arc of D for any 1 ≤ k ≤ l.
If u = v then π is called a closed walk. A digraph D is strongly connected if
for every pair of distinct vertices u, v of D, there exists a walk from u to v

Email address: [email protected], [email protected] (Juan R.

Carmona)

Preprint submitted to . September 16, 2019

and a walk from v to u. A strong component of a digraph D is a maximal
subdigraph with respect to the property of being strongly connected.
Denote the number of closed walks of length 2 of associated vertex vi ∈ V
(i) (i) ←→ (1) (2) (n)
by c2 . Note that c2 =  Ni  . The sequence (c2 , c2 , ..., c2 ) is called closed
(1) (2) (n)
walk sequence of length 2 of D. Thus c2 = c2 + c2 + · · · + c2 is the number
(i)
of all closed walks of length 2 of D. Denote by t2 the sum of the all closed
walks of legth 2 of the vertices doubly adjacente to vi . Then,
(i)
X (j)
t2 = c2 ,
←→
vj ∈ N i

A dı́graph D = (V, Γ) is symmetric if for any (u, v) ∈ Γ also (v, u) ∈ Γ,

where u, v ∈ V . A one-to-one correspondence between simple graphs and
←
→ ←
→
symmetric digraphs is given by G → G , where G has the same vertex set
as the graph G, and each edge uv of G is replaced by a pair of symmetric
arcs (u, v) and (v, u). Under this correspondence, a graph can be identified
with a symmetric digraph.
The adjacency matrix A of the digraph D is a 01 matrix of order n with
entries aij , such that aij = 1 if (vi , vj ) ∈ Γ and aij = 0 otherwise. Denote the
characteristic polynomial of matrix A as ΦD (x) and its eigenvalues z1 , z2 , ...zn ,
possibly complex (see [8], [16].) The spectral radius of a digraph D is denoted
by ρ = ρ(D) and defined as

ρ = max {|zi |},

i=1,...,n

where |zi | denote the modulus of zi .

Ivan Gutman in [11], introduced the concept of energy for a simple graph
G, as
n
X
E(G) = |λi |,
i=1

where λ1 , λ2 , ..., λn are the eigenvalues of the graph G. Details of the theory
of graph energy can be found in the reviews [12],[13] and the book [21]. New
results of this theory appear in [1],[4],[14],[17]. It is well known [7] that if G
is a graph with n vertices then

1 ∞ ixΦ0G (ix)
Z  
E(G) = n− dx
π −∞ ΦG (ix)

2
 Peña and Rada in [22] showed that for a digraph of n vertices, you have
to n
∞
ixΦ0D (ix)
Z  
1 X
n− dx = |Re(zi )|,
π −∞ ΦD (ix) i=1
extending the concept of energy for the case of digraphs as
n
X
E(D) = |Re(zi )|,
i=1

where z1 , ..., zn are the eigenvalues of D and Re(zi ) denotes the real part of
zi . For more details about the energy of digraphs, see [2],[6],[8],[9],[10],[19],
[22],[23],[24] and thereferences therein.
In [23], Rada generalizes the McClelland inequality for any digraph D
with n vertices, a arcs and c2 closed walks of length 2.
r
n(a + c2 )
E(D) ≤ , (1)
2
←
→
with equality in (1) if and only if D is the direct sum of n2 copies of K2 . in
order to obtain an upper bound for energy of a digraph, Gudiño and Rada in
[10], generalizing the idea in [20], showed that the following relation holds:
p
E(D) ≤ ρ + (n − 1)(a − ρ2 ). (2)
c2
Then, using the inequality ≤ ρ, see [10], they obtained the upper bound
n
s 
c2  c 2 
2
E(D) ≤ + (n − 1) a − . (3)
n n
←
→
Equality holds in (3) if and only if D is either the empty digraph or D = G ,
where G is either n2 K2 , Kn , a non-complete connected strongly regular graph
r
2

c
a−( n2 )
with two non-trivial eigenvalues both with absolute value (n−1)
.
Tian and Cui in [25], improve the upper bound (3) with the following
result
v v !
u n   u n 
u 1 X (i) 2 u 1 X (i)
 2
E(D) ≤ t c2 + t(n − 1) a − c2 . (4)
n i=1 n i=1

3
 ←
→
The equality in (4) holds if and only if D = G , where G is either n2 K2 , Kn ,
a non-complete connected stronglyv regular graph with two non-trivial eigen-
2

u a− ni=1 (c(i)
2 ) 
u P

t n

values both with absolute value (n−1)

, or nK1 .
In this work, motivated by the strategies used in [26] and [25], we get a
(i) (i)
lower bound for the spectral radius ρ in terms of n, c2 and t2 , improving the
known results. Using this result we obtain an upper bound of E(G) in terms
(i) (i)
of n,a, c2 and t2 . In addition, we will show that this bound improving and
generalize the bounds given for graphs and digraphs in [10], [25] and [26].

2. Improving lower bound on the spectral radius of a digraph

Obtaining lower bounds for the spectral radius ρ of a digraph D is essen-
tial to obtain new bounds for E(D), see section 3.
Remark 1. Recall that for an n-by-n matrix A = (aij ), its geometric sym-
√
metrization, denoted by S(A) = (sij ), is the matrix with entries sij = aij aji
for any i, j = 1, 2, ..., n.Thus, it is holds that
n
X
(i)
i.- c2 = sij for any vertex vi ∈ V .
j=1

n
X n 
X 2
(i) (i)
ii.- t2 = c2 .
i=i i=1
p
iii.- ρ(A) ≥ ρ(S(A)) = ρ(S(A2 )).
In [10] obtained the following theorem:
Theorem 2 (Gudiño and Rada). Let D be a digraph with n vertices and
c2 closed walks of length 2. Then
c2
ρ(D) ≥ (5)
n
Equality holds if and only if
←
→
D = G + {possibly some arcs that do not belong to cycles},
c2
where G is a n
-regular graph.

4
 In [25], a better lower bound is presented
(1) (2) (n)
Theorem 3 (Tian and Cui). Let D digraph with n vertices. Also let c2 , c2 , ..., c2
be the closed walk sequence of length 2 of D. Then
v
u P  2
u n (i)
t i=1 c2
ρ(D) ≥ , (6)
n
whit equality in (6) if only if
←
→
D = G + {possibly some arcs that do not belong to cycles},

where each connected component of G is either an r-regular graph or an

(i) 2
  Pn
i=1 c2
(r1 , r2 )-semiregular bipartite graph, satisfying r1 r2 = n
.
Now, we give the following lemma which is important for finding improved
lower bound for the spectral radius of D.
Lemma 4. [23] Let D be a digraph with n vertices, a arcs and c2 closed
walks of length 2. If z1 , z2 , ..., zn are the eigenvalues of D, then
n
X n
X
i.- (Re(zi ))2 − (Im(zi ))2 = c2 ;
i=1 i=1

n
X n
X
2
ii.- (Re(zi )) + (Im(zi ))2 ≤ a.
i=1 i=1

A first main result in this work is the following

(1) (2) (n)
Theorem 5. Let D digraph with n vertices, with sequences c2 , c2 , ..., c2
(1) (2) (n)
and t2 , t2 , ..., t2 . Then
v
u P  2
u n (i)
u i=1 t2
ρ(D) ≥ u
t Pn  (i) 2 , (7)
i=1 c 2

whit equality in (7) if only if

←
→
D = G + {possibly some arcs that do not belong to cycles},

5
 where each connected component of G is either an r-regular graph or an
(i) 2
  Pn
i=1 t2
(r1 , r2 )-semiregular bipartite graph, satisfying r1 r2 = Pn 
(i) 2
 .
i=1 c2

Proof. for the Rayleigh quotient

v
u P  2
u n (i)
i=1 t2
r r
p xT S(A)2 x T 2
c S(A) c u
ρ(S(A)2 ) = maxx6=0 ≥ =u
t Pn  (i) 2 , (8)
xT x cT c
i=1 c2

(1) (2) (n)

where c = (c2 , c2 , ..., c2 )T . Thus, we obtain (7). To prove equality, we
will use the ideas used in the works [25], [15], [10],[5]. Indeed, suppose now
that the equality in (7) holds, then

cT S(A)2 c
ρ(S(A)2 ) = ,
cT c
then c is a positive eigenvector of S(A)2 corresponding to the eigenvalue
ρ(S(A)2 ), either one or two. Next we consider three cases.

Case 1: D is strongly connected.

A is a irreducible matrix in this case. If A > S(A), then ρ(A) >
ρ(S(A)) as A is irreducible (see [3], Corollary 2.1.5), this contradicts
our assumption of equality. Therefore we have that A is a symmetric
←
→
matrix, which implies that D = G . In this case G is a connected simple
graph. Then, similar to the proofs in ([15], Theorem 3.1), one can
easily obtain that G is either an r-regular graph or (r1 , r2 )-semiregular
(i) 2
 Pn
i=1 t2
2
bipartite graph, satisfying r = r1 r2 = Pn  (i) 2 .
i=1 c2

Case 2: D is direct sum of its disjoint strongly connected components

Ps D1 , D2 , ..., Ds .
Let Ak be the nk -by-nk adjacency matrix of Dk and k=1 nk = n. In
this case  
A21
2
 A22 
A = ,
 
. .
 . 
2
As

6
 where the rest of the unspecified entries are 0. Since the equality holds
in (7), we have
s r
p x T S(A)2 x cT S(A)2 c
ρ(S(A)2 ) = max =
x6=0 xT x cT c
v
u s T
uX cnk S(Ak )2 cnk nk
= t
k=1
nk cTnk cnk
v
u s
uX nk ρ (S(Ak )2 ) q
≤ t ≤ max ρ(S(Ak )2 )
k=1
n k
p p
= max ρ(S(Ak )2 ) = ρ(S(A)2 )
pk
= ρ(A)2

which implies that, for every k = 1, 2, ..., s,

v
u s T
p q q uX cnk S(A2k )cnk
2 2 2
ρ(A) = ρ(A ) = ρ(Ak ) = ρ(S(Ak )) = t
k=1
nk

←
→
Then, from Case (1) each Dk = Gk , where each connected component
Gk is either an r-regular graph or (r1 , r2 )-semiregular bipartite graph,
(i) 2

Pn 
i=1t2
satisfying r2 = r1 r2 = Pn  (i) 2 .
i=1 c2

Case 3: Db is a digraph obtained from D by deleting those arcs of D that do

not belong to any cycle.
Then S(A) = S(A(D)), b where A(D)b is the adjacency matrix of D. b
Clearly, D and D b have the same cycle structure. By Theorem 1.2 in
[8], we have that ΦD (x) = ΦDb (x), which implies that D and D b also
have the same eigenvalues. On the other hand, since D b is direct sum of
its some disjoint strongly connected components, then Case (2) implies
that D b =← →
G and each connected component of G is either an r-regular
graph or an (r1 , r2 )-semiregular bipartite graph, satisfying r2 = r1 r2 =
Pn  (i) 2
i=1 t2 ←
→
Pn  (i) 2 . Hence,D = G +{possibly some arcs that do not belong to cycles}.
i=1 c2

7
 ←
→
Conversely, suppose that D = G +{possibly some arcs that do not belong to cycles},
where each connected component of G is either an r-regular graph or an
(i) 2
 
Pn
i=1 t2
2
(r1 , r2 )-semiregular bipartite graph, satisfying r = r1 r2 = Pn  (i) 2
 . It is
i=1 c2
easy to check that the equality in (7) holds.
←
→
The result given in [26] is here re-obtained considering D = G .

Corollary 6. Let G be a nonempty graph with degree sequence d1 , d2 , , dn

and 2-degree sequence t1 , t2 , , tn . Then
sP
n
t2
λ1 (G) ≥ Pni=1 i2 ,
i=1 di

with equality if and only if G is a pseudo-regular graph or a pseudo-semiregular

bipartite graph.

The following remark allows us to prove that the bound given in (7) is better
than the bound (6) given in [25] and consequently better than the bound
given in [10].
Remark 7. Note that
v
v u P  2
u n (i)
u i=1 t2
u n  
u1 X (i)
2
t c ≤ t P  2 .
u
n i=1 2 n (i)
i=1 c2

Proof. In efect, by Cauchy-Schwarz inequality, we have

n
!2 n 
X X 2
(i) (i)
t2 ≤n t2 .
i=1 i=1

Using this inequality and Remark 1 (part ii),

v v
u P  2 u P 2 v
u n (i) n (i)
t t u n  
u
u i=1 2 u i=1 2 u 1 X (i) 2
t Pn  (i) 2 ≥ t Pn  (i) 2 = n c2 .
u u t
i=1 c2 n i=1 c2 i=1

8
3. An upper bound for the energy of a digraph
In this section, using the strategies given in articles [5],[25] and [26], we
will construct a lower bound for the energy of digraph D, using the result
obtained in 5.
(1) (2) (n)
Theorem 8. Let D digraph with n vertices, a arcs, with sequences c2 , c2 , ..., c2
(1) (2) (n)
and t2 , t2 , ..., t2 . Then

v v
u P  2 u 
Pn  (i) 2

u n (i)
u i=1 t2 i=1 t2
u
u
E(D) ≤ u + u(n − 1)  a −   2 . (9)

t Pn   2 t 
(i) Pn (i)
i=1 c2 i=1 c2

←→
The equality in (9) holds if and only if D = G , where G is either n2 K2 , Kn ,
a non-complete connected strongly
v regular graph with two non-trivial eigen-
(i) 2

u a− i=1 (t2 ) 
u Pn

(i) 2
i=1 ( 2 )
t Pn
c
values both with absolute value (n−1)
, or nK1 .

Proof. Let ρ = z1 , z2 , ..., zn be the eigenvalues of the digraph D such that

Re(z1 ) ≥ Re(z2 ) ≥ · · · ≥ Re(zn ). By Lemma 4 (part (ii)), we have
n
X
(Re(zi ))2 ≤ a − ρ2 , (10)
i=2

where a is the number of arcs. Using (10) together with the Cauchy-Schwartz
inequality, we obtain the inequality
v
Xn u
Xn
(Re(zi ))2 ≤ (n − 1)(a − ρ2 ).
u p
|Re(zi )| ≤ (n − 1)
t
i=2 i=2

Thus, we must have

p
E(D) ≤ ρ + (n − 1)(a − ρ2 ). (11)
p √
Now, consider the function f (x) = x + (n − 1)(a − x2 ), x ∈ [0, a]. p It is
easy to see that the function f (x) increases strictly on the interval 0, na
p a √ 
and decreases strictly on n
, a . At this point, we have to analyze two
cases:

9
 Pn  2
(i)
i=1 t2
Case 1. a ≤ n P  2
n (i)
i=1 c2
Then by Theorem 5 and inequality (10), we have
v
u P  2
u n (i)
i=1 t2 √
r
a u
≤ t P  2 ≤ ρ ≤ a.
u
n n (i)
i=1 c2

s !
Pn  (i) 2
t
i=1 2
p a √ 
Thus f (ρ) ≤ f Pn  (i) 2 , because f is decreasing in n
, a .
i=1 c2

This implies that the inequality (9) holds. On the other hand, if the
equality in (9) holds, then
v
u P  2
u n (i)
u i=1 t2
ρ = t P  2 ,
u
n (i)
i=1 c2

later by theorem 5, we have that

←
→
D = G + {possibly some arcs that do not belong to cycles},

where each connected component of G is either an r-regular graph or

(i) 2
  Pn
i=1 t2
an (r1 , r2 )-semiregular bipartite graph, satisfying r1 r2 = Pn  (i) 2
 .
i=1 c2
Noting that c2 ≤ a. By Theorem 2.1 in [27], we obtain
v v
u P  2 u 
Pn  (i) 2

u n (i)
u i=1 t2 i=1 t2
u
u
E(D) = E(G) ≤ u + (n − 1) c −
u  
t Pn  (i) 2 t  2 Pn  (i) 2 
i=1 c2 i=1 c2
v v
u P  2 u 
Pn  (i) 2

u n (i)
u i=1 t2 i=1 t2
u
u
≤ u t Pn  (i) 2 + t(n − 1) a − Pn  (i) 2  = E(D),
u  

i=1 c2 i=1 c2

10
 which implies c2 = a, this way we have to
v v
u P  2 u 
Pn  (i) 2

u n (i)
u i=1 t2 i=1 t2
u
u
E(D) = u + u(n − 1) c 2 −   2 .

t Pn   2 t 
(i) Pn (i)
i=1 c2 i=1 c2

Using the Theorem 2.1 in [27], we obtain the conditions of equality.

Pn  (i) 2
i=1 t2
Case 2: a > n P  2 .
n (i)
i=1 c2
Using Remark 7, we have
v
v u P  2
u n (i)
u 1 X (i) 2 u i=1 t2
u n   r
a
0≤t c2 ≤u  2 ≤ .
n i=1 t P n (i) n
i=1 c 2

vu P  2 
v  u n (i)
u i=1 t2
u n  
u 1 X (i) 2 
Therefore, we have to f  t c2  ≤ f t P  2 
 u
,
n i=1 n (i)
i=1 c2
 pa
because, f is increasing in 0, n . Then by Theorem 2 in [5] the
inequality (9) holds.
Assume now that equality holds in (9), then we have that
vu P  2 
v  u n (i)
u i=1 t2
u n  
u 1 X (i) 2 
E(D) = f t c2  = f  u
2,

n i=1 t  
 Pn (i)
i=1 c2

Thus, the characteristics of D are obtained from the conditions of the

Theorem 2 in [5]
←
→
The result given in [27] is here re-obtained considering D = G .
Corollary 9. Let G be a nonempty graph with n vertices, m edges, degree
sequence d1 , d2 , , dn and 2-degree sequence t1 , t2 , , tn . Then
sP s
n 2
 Pn 2 
ti t
E(G) ≤ Pn 2 + (n − 1) 2m − Pni=1 i2 .
i=1

i=1 di i=1 di

11
Equality holds if and only if one of the following statements holds:
(1) G ∼
= n2 K2 ;
(2) G ∼
= Kn ;
(3) G is a non-bipartite
 q connected p seudo-regular
 graph with three distinct
q
2 2
eigenvalues p, 2m−p , − 2m−p , where p > m
p
n−1 n−1 n
.

Remark 10. Consider the collection of digraphs of n vertices, a arcs and c2

walks of length 2 denoted and defined by:

Γ = {D : an < (c2 )2 }.

Si D ∈ Γ, then
v
v u P  2
u n (i)
X (i) 2 u i=1 t2
u n  
√
r
a c2 u 1
≤ ≤t c2 ≤ t P  2 ≤ ρ ≤ a.
u
n n n i=1 n (i)
i=1 c2

p a √ 
Since the function f is strictly decreasing on the interval n
, a , we have
that:
vu P  2 
u n (i)
v 
u i=1 2 t u n  
u 1 X (i) 2 c 
 2
E(D) ≤ f (ρ) ≤ f t P  2  ≤ f
 u   t c2  ≤f
n (i) n i=1 n
i=1 c2

In this way, we can affirm that for all G ∈ Γ, the bound given in (9) is
better than the bound (4) given in [25] and consequently better than the bound
(3) given in [10].

References
[1] E. Andrade, J.R. Carmona, G. Infante, M. Robbiano, New lower bounds
for the energy of matrices and graphs.(2019) arXiv:1903.01326
[2] S.K. Ayyaswamy, S. Balachandran, I. Gutman, Upper bound for the
energy of strongly connected digraphs, Applicable Anal. Discrete Math.
5 (2011) 37-45.

12
 [3] Berman A, Plemmons RJ. Nonnegative matrices in the mathematical
sciences. New York (NY): Academic; 1979; Philadelphia (PA): SIAM;
1994.

[4] Ş. B. Bozkurt Altındağ, D. Bozkurt. Lower bounds for the energy of
(bipartite) graphs, MATCH Commun. Math. Comput. Chem. 77 (2017):
9-14.

[5] Ş. B. Bozkurt Altındağ, D. Bozkurt and X-D. Zhang, On the spectral
radius and the energy of a digraph, Linear and Multilinear Algebra, 63
(2015), No. 10, 2009-2016

[6] R. Brualdi, Spectra of digraphs, Linear Algebra Appl. 432 (2010) 2181-
2213.

[7] C.A. Coulson, On the calculation of the energy in unsaturated hydro-

carbon molecules, Proc. Cambridge Phil. Soc. 36 (1940) 201-203.

[8] D.M. Cvetković, M. Doob, H. Sachs, Spectra of Graph, Theory and

Application ,Academic Press, New York, (1980).

[9] R. Cruz, H. Giraldo and J. Rada, An upper bound for the energy of
radial digraphs, Linear Algebra Appl., 442 (2014), 75-81.

[10] E. Gudiño, J. Rada, A lower bound for the spectral radius of a digraph,
Linear Algebra Appl. 433 (2010) 233-240.

[11] I. Gutman, The energy of a graph, Ber. Math.Statist. Sekt. Forschungsz.

Graz. 103 (1978) 1-22.

[12] I. Gutman, The energy of a graph: Old and new results, in: A. Betten,
A. Kohnert, R. Laue, A. Wassermann (Eds.), Algebraic Combinatorics
and Applications, Springer, Berlin, 2001, pp. 196-211.

[13] I. Gutman, X. Li, J. Zhang, Graph energy, in: M. Dehmer, F. Emmert-

Streib (Eds.), Analysis of Complex Networks. From Biology to Linguis-
tics, WileyVCH, Weinheim, 2009, pp. 145-174.

[14] I. Gutman , B. Furtula , E. Zogic , E. Glogic , Resolvent energy of graphs,

MATCH Commun. Math. Comput. Chem. 75 (2) (2016) 279-290.

13
[15] Hong Y, Zhang XD. Sharp upper and lower bounds for the largest eigen-
value of the Laplacian matrices of trees. Discrete Math. 2005;296:187-
197.

[16] R. Horn, C. Johnson, Matrix Analysis, Cambridge University Press,

(1985).

[17] A. Jahanbani, Lower bounds for the energy of graphs. AKCE Interna-
tional Journal of Graphs and Combinatorics. 15 (2018): 88-96.

[18] H. Liu, M. Lu and F. Tian, Some upper bounds for the energy of graphs,
J. Math. Chem. 41 (2007) 45-57

[19] W. López, J. Rada, Equienergetic digraphs, Int. J. Pure Appl. Math. 36

(3) (2007) 361-372.

[20] V. Moulton, J.H. Koolen, Maximal energy graphs, Adv. Appl. Math. 26
(2001) 47-52.

[21] Li X, Shi Y, Gutman I. Graph energy. New York (NY): Springer; 2012.

[22] I Peña and J. Rada. Energy of digraphs. Linear and Multilinear Algebra,
56: 565-579, 2008.

[23] J. Rada, The McClelland inequality for the energy of digraphs, Linear
Algebra Appl. 430 (2009) 800-804.

[24] J.Rada, Bounds for the energy of normal digraphs,Linear Multilinear

Algebra,60 (3)(2012) 323-332.

[25] G-X. Tian, S-Y Cui, On upper bounds for the energy of digraphs, Linear
Algebra and its Applications 438 (2013) 4742-4749.

[26] A. Yu, M. Lu and F. Tian, On the spectral radius of graphs, Linear

Algebra and its Applications, 387(2004), 41-49.

[27] A. Yu, M. Lu, F. Tian, New upper bound for the energy of graphs,
MATCH Commun. Math. Comput. Chem. 53 (2005), 441448.

14