Mathematical and Computer Modelling 49 (2009) 703–708

Contents lists available at ScienceDirect

Mathematical and Computer Modelling

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

The role of matrices that are convergent to zero in the study of

semilinear operator systems
Radu Precup
Department of Applied Mathematics, Babeş–Bolyai University, 400084 Cluj, Romania

article info a b s t r a c t

Article history: In this paper we explain the advantage of vector-valued norms and the role of matrices that
Received 26 November 2007 are convergent to zero in the study of semilinear operator systems by means of some basic
Received in revised form 26 March 2008 methods of nonlinear analysis: the contraction principle, Schauder’s fixed point theorem,
Accepted 25 April 2008
the Leray–Schauder continuation principle and Krasnoselskii’s cone fixed point theorem.
A vector version of Krasnoselskii’s theorem is also established.
Keywords:
© 2008 Elsevier Ltd. All rights reserved.
Semilinear operator system
Vector-valued norm
Matrix convergent to zero
Fixed point
Krasnoselskii cone fixed point theorem

1. Introduction

In this paper we are concerning with the solvability of the semilinear operator system

N1 (u1 , u2 ) = u1

(1)
N2 (u1 , u2 ) = u2

in a Banach space X with norm |.|. Here Ni : X 2 → X (i = 1, 2) are given nonlinear operators. Systems of this type arise from
mathematical modelling of many processes from a variety of disciplines, including physics, biology, chemistry, engineering
and other sciences. Thus, initial value problems and boundary value problems for nonlinear competitive or cooperative
differential systems from mathematical biology [4] and mathematical economics [3] can be put in the operator form (1).
It is obvious that system (1) can be viewed as a fixed point problem:

N (u) = u (2)

in the space X 2 , where u = (u1 , u2 ) and N = (N1 , N2 ). Therefore, we may think of applying to (2), in X 2 endowed with a
norm induced by the norm of X , different abstract existence results from nonlinear functional analysis, such as the Banach
contraction principle, the Schauder fixed point theorem, the Leray–Schauder continuation principle, Krasnoselskii’s cone
fixed point theorem, and so on. The aim of this paper is to point out that better results can be obtained for system (1) if in
X 2 we consider the vector-valued norm
 
|u1 |
||u|| = (3)
|u2 |

E-mail address: [email protected].

0895-7177/$ – see front matter © 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mcm.2008.04.006
704 R. Precup / Mathematical and Computer Modelling 49 (2009) 703–708

for u = (u1 , u2 ) ∈ X 2 , instead of a usual scalar norm such as

|u|l = |u1 | + |u2 | ,
|u|m = max {|u1 | , |u2 |} , or
q
|u|e = |u1 |2 + |u2 |2 .
Historically, this was shown by Perov and Kibenko [7] (see also [1,8,9]) in connection with the contraction principle.

Theorem 1 (Perov). Let (E , d) be a complete generalized metric space with d : E × E → Rn and let N : E → E be such that
d (N (u) , N (v)) ≤ M d (u, v)
for all u, v ∈ E and some square matrix M of nonnegative numbers. If the matrix M is convergent to zero, that is M k → 0 as
k → ∞, then N has a unique fixed point u and
d N k (v) , u ≤ M k (I − M )−1 d (N (v) , v)

for every v ∈ E and k ≥ 1.

Let us note the following properties of matrices that are convergent to zero:

Lemma 2. Let M be a square matrix of nonnegative numbers. The following statements are equivalent:
(i) M is a matrix convergent to zero.
(ii) I − M is non-singular and
( I − M ) −1 = I + M + M 2 + · · · .
(iii) |λ| < 1 for every λ ∈ C with det (M − λI ) = 0.
(iv) I − M is non-singular and (I − M )−1 has nonnegative elements.
Proof. The equivalence of (i), (ii) and (iii) is well known; see for example [9,8]. Also the implication from (ii) to (iv) is obvious.
Thus the proof will be complete if we show that (iv) implies (i). Assume (iv). Then from the identity
Sk (I − M ) = I − M k+1
which is true for Sk = I + M + · · · + M k , since M and (I − M )−1 have nonnegative elements, we deduce that
Sk = I − M k+1 (I − M )−1

= (I − M )−1 − M k+1 (I − M )−1

≤ (I − M )−1 .
Thus (Sk )k≥1 is a bounded sequence. Since it is nondecreasing (on elements) by its definition, we deduce that it is convergent.
As a consequence M k → 0 as k → ∞. 
We conclude this introduction by three other well-known abstract results of nonlinear functional analysis. For proofs
and more information we refer the reader to [2,6].

Theorem 3 (Schauder). Let E be a Banach space, D a nonempty closed bounded and convex subset of E, and N : D → D a
completely continuous operator. Then N has at least one fixed point.

Theorem 4 (Leray–Schauder). Let E be a Banach space, U a bounded open subset of E with 0 ∈ U , and N : U → E a completely
continuous operator. If u 6= λN (u) for all u ∈ U \ U and λ ∈ (0, 1), then N has at least one fixed point.

Theorem 5 (Krasnoselskii). Let (E , |.|0 ) be a Banach space, W a proper wedge of E (i.e., a closed convex set satisfying λW ⊂ W
for all λ ≥ 0 and which is not a linear subspace of E) and let α, β > 0 with α 6= β . Assume that N : W → W is a completely
continuous operator such that
x 6= λN (x) for |x|0 = a and λ ∈ (0, 1) , (4)
x 6= λN (x) for |x|0 = β and λ ∈ (1, ∞) , (5)
and
inf {|N (x)|0 : |x|0 = β} > 0. (6)
Then N has at least one fixed point x with
min {α, β} ≤ |x|0 ≤ max {α, β} .
 R. Precup / Mathematical and Computer Modelling 49 (2009) 703–708 705

Notice that a sufficient condition for (4) is that

|N (x)|0 ≤ |x|0 for |x|0 = α. (7)
Also, a sufficient condition for (5) and (6) is that
|N (x)|0 ≥ |x|0 for |x|0 = β. (8)

2. Application of Perov’s fixed point theorem

First we shall present an application of Perov’s theorem to system (1).

Theorem 6. Assume that for each i ∈ {1, 2}, there exist nonnegative numbers ai and bi such that
|Ni (u1 , u2 ) − Ni (v1 , v2 )| ≤ ai |u1 − v1 | + bi |u2 − v2 | (9)

for all u1 , u2 , v1 , v2 ∈ X . In addition assume that

 
a1 b1
M = is a convergent to zero matrix. (10)
a2 b2

Then (1) has a unique solution u = (u1 , u2 ) in X 2 and Nik (v) → ui as k → ∞ for every v ∈ X 2 and i = 1, 2.

Proof. Condition (9) can be rewritten as

kN (u) − N (v)k ≤ M ku − vk .
Thus Perov’s fixed point theorem applies. Here E := X 2 and d (u, v) := ku − vk . 

3. Application of Schauder’s fixed point theorem

Now we shall relax assumption (9) to an at most linear growth condition, but we shall require instead a compactness
property for Ni .

Theorem 7. Assume that for each i ∈ {1, 2}, the operator Ni is completely continuous and that there exist nonnegative numbers
ai , bi and ci such that
|Ni (u1 , u2 )| ≤ ai |u1 | + bi |u2 | + ci (11)

for all u1 , u2 ∈ X . In addition assume that condition (10) is satisfied. Then (1) has at least one solution u = (u1 , u2 ) with
   
|u1 | c1
≤ ( I − M ) −1 . (12)
|u2 | c2

Proof. Condition (11) can be written in a matrix form as

|N1 (u)|
     
| u1 | c
≤M + 1 . (13)
|N2 (u)| | u2 | c2

We shall apply Schauder’s fixed point theorem to the restriction of N to a subset of X 2 , of the form

D := u = (u1 , u2 ) ∈ X 2 : |u1 | ≤ R1 and |u2 | ≤ R2 .



Thus the existence problem reduces to the invariance condition N (D) ⊂ D. Therefore, we have to find two nonnegative
numbers R1 , R2 such that
|u1 | ≤ R1 , |u2 | ≤ R2 imply |N1 (u)| ≤ R1 , |N2 (u)| ≤ R2 .
According to (13), if |ui | ≤ Ri for i = 1, 2, then

|N1 (u)|
     
R c
≤M 1 + 1 .
|N2 (u)| R2 c2

Hence it would be enough that

     
R1 c R1
M + 1 =
R2 c2 R2
706 R. Precup / Mathematical and Computer Modelling 49 (2009) 703–708

that is
   
R1 c1
= (I − M ) −1
. (14)
R2 c2

Notice that (14) gives us nonnegative constants R1 , R2 , since c1 , c2 ≥ 0 and (I − M )−1 has nonnegative elements as follows
from Lemma 2(iv). 

4. Application of the Leray–Schauder principle

In this section, we assume that X is a Hilbert space with inner product h., .i and norm |.| and we shall refine the result of
Theorem 6 assuming that the operators Ni , i = 1, 2, split into Ni0 + Ni00 with Ni0 satisfying an at most linear growth condition
of type (11) and Ni00 satisfying a sign-type condition. No growth conditions will be required for Ni00 .

Theorem 8. Assume that for each i ∈ {1, 2} , Ni is completely continuous and splits into Ni0 + Ni00 , where

N (u) ≤ ai |u1 | + bi |u2 | + ci

 0 
i (15)

Ni (u) , ui ≤ 0

00 
(16)

for all u = (u1 , u2 ) ∈ X and some nonnegative constants ai , bi and ci . In addition assume that condition (10) is satisfied. Then
2

(1) has at least one solution u = (u1 , u2 ) in X 2 , and any solution of (1) satisfies (12).

Proof. According to the Leray–Schauder continuation principle, it suffices to show that the set of all solutions in X 2 to the
equations

λN1 (u1 , u2 ) = u1

(17)
λN2 (u1 , u2 ) = u2
when λ ∈ [0, 1] is bounded. Let u = (u1 , u2 ) be any solution of (17). Then, for each i ∈ {1, 2}, from (15) and (16), we obtain
|ui |2 = λ hNi (u) , ui i ≤ λ Ni0 (u) , ui ≤ Ni0 (u) |ui |

  

≤ (ai |u1 | + bi |u2 | + ci ) |ui | .

Hence
|ui | ≤ ai |u1 | + bi |u2 | + ci .
These inequalities can be put in the form
     
| u1 | |u1 | c
≤M + 1
| u2 | |u2 | c2

which by (10) immediately yields (12). 

5. Application of Krasnoselskii’s theorem

In this section the notation |.|0 stands for any norm on X 2 . Also, for a square matrix M of nonnegative numbers, the
notation M ≤ 1 means that µM is convergent to zero for every µ ∈ (0, 1), or equivalently, that |λ| ≤ 1 for every λ ∈ C
with det (M − λI ) = 0.

Theorem 9. Let K be a proper wedge of the Banach space (X , |.|) and let Ni : K 2 → K be completely continuous maps, i = 1, 2.
Assume that there are α, β ∈ (0, ∞) with α 6= β , and nonnegative numbers ai , bi , a0i , b0i (i = 1, 2) such that

|Ni (u)| ≤ ai |u1 | + bi |u2 | for |u|0 = α, i = 1, 2 (18)

|ui | ≤ ai |N1 (u)| + bi |N2 (u)|
0 0
for |u|0 = β, i = 1, 2. (19)
If in addition M ≤ 1 and M 0 ≤ 1 where

a01 b01
   
a1 b1
M = and M0 =
a2 b2 a02 b02

then (1) has at least one solution u = (u1 , u2 ) ∈ K × K satisfying

min {α, β} ≤ |u|0 ≤ max {α, β} .
 R. Precup / Mathematical and Computer Modelling 49 (2009) 703–708 707

Proof. Notice that conditions (18) and (19) can be written in the form

kN (u)k ≤ M kuk for |u|0 = α (20)

and respectively

M 0 kN (u)k ≥ kuk for |u|0 = β. (21)

First we show that u 6= λN (u) for all u ∈ K with |u|0 = α and λ ∈ (0, 1) . Indeed, otherwise, if u = λN (u) for some u
2

with |u|0 = α and λ ∈ (0, 1), then kuk = λ kN (u)k ≤ λM kuk. Thus (I − λM ) kuk ≤ 0. Since λM is convergent to zero, we
deduce that kuk ≤ (I − λM )−1 0 = 0, whence u = 0. This contradicts |u|0 = α > 0.
Next we prove that u 6= λN (u) for all u ∈ K 2 with |u|0 = β and λ ∈ (1, ∞) . Indeed, if u = λN (u) for some
u with
|u|0 = β and λ ∈ (1, ∞), then kuk = λ kN (u)k and so M 0 kuk = λM 0 kN (u)k ≥ λ kuk. Consequently, I − λ1 M 0 kuk ≤ 0,

−1
whence kuk ≤ I − λ1 M 0 0. Hence u = 0, which contradicts |u|0 = β > 0.
Finally, we note that (21) guarantees that inf{|N (u)|0 : |u|0 = β} > 0. Therefore we may apply Krasnoselskii’s theorem
to E = X 2 and W = K 2 . 

6. The vector-valued norm versus scalar norms

The aim of this section is to show that the results in Sections 2–4, obtained by using the vector-valued norm (3), are
better than those established by means of any scalar norm in X 2 .
10 . Let us first consider the scalar norm |u|l = |u1 | + |u2 |. Then, if N1 , N2 satisfy the Lipschitz conditions (9), we obtain

|N (u) − N (v)|l ≤ max {a1 + a2 , b1 + b2 } |u − v|l (22)

for all u, v ∈ X . Similarly, if N1 , N2 satisfy (11), then

2

|N (u)|l ≤ max {a1 + a2 , b1 + b2 } |u|l + c1 + c2 . (23)

Finally, if N1 , N2 are as in Theorem 7, and u is any solution of (17), then

|u|l ≤ max {a1 + a2 , b1 + b2 } |u|l + c1 + c2 . (24)

Thus, the Banach contraction principle, Schauder’s fixed point theorem and the Leray–Schauder continuation principle can
be applied, provided that

α := max {a1 + a2 , b1 + b2 } < 1. (25)

2 . If in X we consider the scalar norm |u|m = max {|u1 | , |u2 |}, then the corresponding formulas for (22)–(24) are,
0 2

respectively,

|N (u) − N (v)|m ≤ max {a1 + b1 , a2 + b2 } |u − v|m

|N (u)|m ≤ max {a1 + b1 , a2 + b2 } |u|m + max {c1 , c2 }
|u|m ≤ max {a1 + b1 , a2 + b2 } |u|m + max {c1 , c2 } .
Thus, with this choice of a scalar norm in X 2 , the above three results of nonlinear functional analysis apply provided that

β := max {a1 + b1 , a2 + b2 } < 1. (26)

p
30 . Similarly, if in X 2 we consider the euclidean norm |u|e = |u1 |2 + |u2 |2 , then the corresponding formulas for (22)–
(24) are, respectively,
q
|N (u) − N (v)|e ≤ a21 + a22 + b21 + b22 |u − v|e
q q
|N (u)|e ≤ a21 + a22 + b21 + b22 |u|e + c12 + c22
q q
|u|e ≤ a21 + a22 + b21 + b22 |u|e + c12 + c22 .

Thus, the applicability condition for the above abstract results is the following inequality:

γ := a21 + a22 + b21 + b22 < 1. (27)

The following examples show that, in general, the condition that M is a matrix convergent to zero is weaker than
conditions (25)–(27).
708 R. Precup / Mathematical and Computer Modelling 49 (2009) 703–708

Example 1. Let
 
a a
M = .
b b

Then the characteristic roots of M are λ1 = 0 and λ2 = a + b. Hence M is convergent to zero if and only if a + b < 1. On
the other hand,
α = a + b, β = max {2a, 2b} , γ = 2 a2 + b 2 .

Hence each one of the conditions β < 1 and γ < 1 is more restrictive than the condition that M is convergent to zero.

Example 2. Let
 
a b
M = .
a b

Now the characteristic roots of M are λ1 = 0, λ2 = a + b and so again M is convergent to zero if and only if a + b < 1. Also,
in this case
α = max {2a, 2b} , β = a + b, γ = 2 a2 + b 2 .

Hence each one of the conditions α < 1 and γ < 1 is more restrictive than the condition that M is convergent to zero.

Example 3. Let
 
a b
M = .
b a

Then, the characteristic roots of M are λ1 = a − b, λ2 = a + b and so again M is convergent to zero if and only if a + b < 1.
Now
α = a + b, β = a + b, γ = 2 a2 + b 2 .

Hence the condition γ < 1 is more restrictive than the condition that M is convergent to zero.

Example 4. Assume
 
a b
M = .
0 c

Then, the characteristic roots of M are λ1 = a, λ2 = c and M is convergent to zero if and only if max {a, c } < 1. Here
α = max {a, b + c } , β = max {a + b, c } , γ = a2 + b 2 + c 2
which show that in general, each one of the conditions α < 1, β < 1 and γ < 1 is more restrictive than the condition that
M is convergent to zero.
Similar considerations can be given in connection with Theorem 9. More exactly, one can test that for different choices of
the norm |.|0 on X 2 , and different types of matrices M , M 0 ≤ 1, conditions (20) and (21) are less restrictive than (7) and (8).
Therefore, we may conclude that for different types of estimations, the use of the vector-valued norm and, correspondingly,
of the matrices convergent to zero, is more appropriate when treating systems of equations.
Notice that our approach can be adapted in order to treat systems of n equations (n ≥ 3). For related topics we refer the
reader to the forthcoming paper [5].

References

[1] C. Avramescu, Sur l’existence des solutions convergentes pour des équations intégrales, An. Univ. Craiova Ser. V (2) (1974) 87–98.
[2] A. Granas, J. Dugundji, Fixed Point Theory, Springer, New York, 2003.
[3] B.S. Jensen, The Dynamic Systems of Basic Economic Growth Models, Kluwer, Dordrecht, 1994.
[4] J.D. Murray, Mathematical Biology, Springer, Berlin, 1989.
[5] D. Muzsi, R. Precup, Nonresonance and existence for systems of semilinear operator equations (in press).
[6] D. O’Regan, R. Precup, Theorems of Leray–Schauder Type and Applications, Gordon and Breach, Amsterdam, 2001.
[7] A.I. Perov, A.V. Kibenko, O a certain general method for investigation of boundary value problems, Izv. Akad. Nauk SSSR 30 (1966) 249–264 (in Russian).
[8] R. Precup, Methods in Nonlinear Integral Equations, Kluwer, Dordrecht, 2002.
[9] I.A. Rus, Principles and Applications of the Fixed Point Theory (Romanian), Dacia, Cluj, 1979.