PSEUDO-RESOLVENTS

ABDUL SAMI AWAN and MIHAI VOICU

The present paper is divided into two sections. In the first section we recall the
notions of resolvent and pseudo-resolvent, and a few remarkable properties. In the
second section we introduce a new concept, the L∞ -type pseudo-resolvent, and
prove a characterization theorem. The generators of L∞ -type pseudo-resolvents
are characterized in spectral terms. The connection between the L∞ -type pseudo-
resolvents and Co -equicontinuous semigroups is pointed out.
AMS 2000 Subject Classification: 47A56.

Key words: pseudo-resolvent, generator, semigroup.

1. INTRODUCTION

The notion of pseudo-resolvent has been introduced by E. Hille in his

remarkable monograph [4]. In 1959, Kato [6] made a few interesting remarks
on pseudo-resolvents and infinitesimal generators of semigroups.
Contraction pseudo-resolvents on Banach spaces have been considered by
Cuculescu ([1], p. 261–281). He investigated excessive functions with respect
to a pseudo-resolvent of integral operators.
Hirsch [5] defined in 1972 the concept of L∞ -type pseudo-resolvent on
Banach spaces. Voicu ([9], [11], [12], [14]) investigated pseudo-resolvents on
locally convex spaces.
In the present paper we define a new concept, the L∞ -type pseudo-
resolvent, and prove a characterization theorem. We give necessary and suffi-
cient conditions under which a linear operator is the generator of a L∞ -type
pseudo-resolvent.
We also prove that the class of generators of L∞ -type pseudo-resolvents
generalizes the class of infinitesimal generators of Co -equicontinuous semi-
groups acting on sequentially complete locally convex spaces.
Throughout the paper, X is a complex Hausdorff locally convex space,
C the complex field, Cs (X) the continuous seminorms on X and L(X) the
algebra of the continuous linear operators.
Definition 1.1. A mapping R : ∆ ⊂ C → L(X) is called pseudo-resolvent if
(1) R(λ) − R(µ) = (µ − λ)R(λ)R(µ)

MATH. REPORTS 10(60), 1 (2008), 1–9

2 Abdul Sami Awan and Mihai Voicu 2

for all λ, µ ∈ ∆. The resolvent equation (1) has a few interesting consequences
stated below.
Proposition 1.1 ([3], p. 208). If R : ∆ → L(X) is a pseudo-resolvent
then
1. R(µ)R(λ) = R(λ)R(µ) for all λ, µ ∈ ∆;
2. ker R(λ), ker(I −λR(λ)), R(λ)(X) and (I −λR(λ))(X) do not depend
on λ ∈ ∆.
Definition 1.2. Let V : D → X be a linear operator and λ ∈ C. We say
that λ ∈ ρ(V ) if
1. λI − V : D → X is one to one;
2. (λI − V )(D) = X and (λI − V )−1 : (λI − V )(D) → X is continuous.
The mapping λ → (λI − V )−1 = R(λ, V ), λ ∈ ρ(V ), is called the
resolvent of V .
Remark 1.1. The mapping λ → R(λ, V ) is not a pseudo-resolvent because
R(λ, V ) ∈
/ L(X)
Proposition 1.2 ([15], p. 211). Let V : D → X be a linear operator.
Suppose that ρ(V ) 6= ∅ and (λI − V )(D) = X for all λ ∈ ρ(V ). Then
R : ρ(V ) → L(X) defined by R(λ) = (λI − V )−1 is a pseudo-resolvent.

2. PSEUDO-RESOLVENTS

Proposition 2.1. The resolvent of a closed linear operator acting on a

complete locally convex space is a pseudo-resolvent.
Proof. Let X be a complete locally convex space, V : D → X a closed
linear operator such that ρ(V ) 6= ∅, and λ ∈ ρ(V ).
Let y ∈ X and (xδ )δ∈∆ ⊂ D a net such that (λI − V )(xδ ) → y. Since
(λI −V )−1 is continuous, (xδ )δ∈∆ is Cauchy, hence convergent. Let x = lim xδ .
δ∈∆
Then lim V (xδ ) = λx−y. In addition, we have (xδ , V (xδ )) → (x, λx−y). Since
δ∈∆
V is closed, x ∈ D and V (x) = λx − y. It follows that (λI − D)(D) = X.
By Proposition 1.2, R : ρ(V ) → L(X) defined by R(λ) = (λI − V )−1 is
a pseudo-resolvent. 
Definition 2.1. A pseudo-resolvent R : ∆ → L(X) is called generated if ∃
a linear operator V : D = R(λ)(X) → X such that ∆ ⊂ ρ(V ), (λI − V )(D) =
X and R(λ) = (λI − V )−1 = R(λ, V ), λ ∈ ∆.
Actually, if such a V exists, it is unique and is called the generator of R.
Proposition 2.2. Let R : ∆ → L(X) be a pseudo-resolvent. The fol-
lowing statements are equivalent:
3 Pseudo-resolvents 3

1. R is generated;
2. ker R(λ) = {0}, λ ∈ ∆.
Proof. 1 ⇒ 2 is a consequence of Definition 2.1. If ker R(λ) = {0} then
λI − R−1 (λ) does not depend on λ ∈ ∆ and V = λI − R−1 (λ) is the generator
of R. For details see ([15], p. 216). 
Example 2.1. Let K(R) the family of compact subsets of R and X =
{x : R → C : x is bounded on the compact subsets of R}. For each K ∈ K(R)
consider the functional defined by
ρK (x) = sup |x(t)| .
t∈K

Then (X, ρK )K∈K(R) is a Hausdorff locally convex space. Let K0 ∈ K(R) and
T : X → X defined by

x (t) for t ∈ K0
T (x)(t) =
0 for t ∈
/ K0 .
It is clear that T is linear and for any K ∈ K(R) we have ρK (T (x)) 6 ρK (x).
Moreover, it follows that T 2 = T , so T is a projection. Let ∆ = C\{0} and
R : ∆ → L(X) defined by R(λ) = λ1 T . Let λ, µ ∈ ∆. Then we have
µ−λ
(2) R(λ) − R(µ) = T
λµ

µ−λ
(3) (µ − λ)R(λ)R(µ) = T.
λµ
Thus, we get
(4) R(λ) − R(µ) = (µ − λ)R(λ)R(µ),
that is R is a pseudo-resolvent.
Remark 2.1. R is not generated because T is not one to one.
We now introduce a remarkable class of generated pseudo-resolvents,
closely connected with semigroups.
Definition 2.2. A pseudo-resolvent R : ∆ → L(X) is called of L∞ -type
if ∃ ω ∈ R such that (ω, ∞) ⊂ ∆ and lim nR(n)(x) = x, ∀ x ∈ X.
n
Remark 2.2. It is clear that a L∞ -type pseudo-resolvent is generated and
R(λ)(X) = X.
Example 2.2. Let X = C([0, 1]) be the set of all continuous real functions
defined on [0, 1]. It is well known that X is a Banach space with respect to
4 Abdul Sami Awan and Mihai Voicu 4

the norm
kf k = sup |f (t)| , f ∈ X.
t∈[0,1]

Consider also the subspace D = {u : u ∈ C 1 ([0, 1]), u(0) = 0} of X and the

linear operator V : D → X defined by V (u) = −u0 . Let λ ∈ R and f ∈ X.
The first order
R x differential equation λu + u0 = f has a unique solution u ∈ D
and u(x) = 0 e λ(t−x) f (t)dt. This is equivalent to saying that ∀ λ ∈ R, the
linear operator λI − V : D → X is a bijection and (λI − V )(u) = f . For any
x ∈ [0, 1] the inequalities
Z x
1
(5) (λI −V )−1 (f )(x)=|u(x)| 6 eλ(t−x) |f (t)| dt 6 (1−e−λx ) kf k , ∀λ > 0,
 
0 λ
do hold. Hence
  1
−1
(6) (λI − V ) (f )(x) 6 kf k , ∀ λ > 0,
 
λ

 −1 
(7) V (f )(x) 6 kf k if λ = 0,

(λI − V )−1 (f )(x) 6 1 e−λ kf k

 
(8) if λ < 0.
|λ|
Hence ρ(V ) = R and, by Definition 2.1, R : R → L(X) defined by R(λ) =
R(λ, V ) = (λI − V )−1 is a generated pseudo-resolvent. On the other hand,
one can remark that R is not a L∞ -type pseudo-resolvent because D is not
dense in X.
Remark 2.3. The class of L∞ -type pseudo-resolvents is strictly included
in the class of generated pseudo-resolvents.
Theorem 2.1. Let R : ∆ → L(X) be a pseudo-resolvent and ω ∈ R
such that (ω, ∞) ⊂ ∆. Then the two statements below are equivalent:
1. 
R is of L∞ -type;
 a) R (λ) (X) = X

2. ∀ p ∈ Cs (X) , ∃ q ∈ Cs (X)
 b) such that lim sup p (nR (n) (x)) ≤ q (x) ∀ x ∈ X.

n
Proof. 1 ⇒ 2 is a simple consequence of Definition 2.2.
2 ⇒ 1. Let µ ∈ ∆ and x ∈ X. By the resolvent equation we have
1 n
(9) nR(n)R(µ)(x) = nR(n)(x) + R(µ)(x), n 6= µ
µ−n n−µ
By 2 b), the sequence (nR(n)(x))n∈N, n>ω is bounded in X and letting n → ∞
in (9) we get
lim nR(n)R(µ)(x) = R(µ)(x).
n
5 Pseudo-resolvents 5

Let now a net xj ∈ R(µ)(X), j ∈ J, such that xj → x. For n > ω and j ∈ J

we have
(10) nR(n)(x) − x = nR(n)(x − xj ) + nR(n)(xj ) − xj + xj − x.
Let p, q ∈ Cs (X) given by 2 b). Then by (10) we have
p(nR(n)(x) − x) 6 p(nR(n)(x − xj )) + p(nR(n)(xj ) − xj ) + p(xj − x).
Moreover,
(11) limsup p(nR(n)(x)−x) 6 q(x−xj ) + limsup p(nR(n)(xj )−xj )+p(xj −x).
n n
Hence limsup p(nR(n)(x) − x) = 0. Consequently,
n
lim nR(n)(x) = x
n
In conclusion, R is a L∞ -type pseudo-resolvent. 
Theorem 2.2. Let R : ∆ → L(X) be a L∞ -type pseudo-resolvent, V :
D → X its generator, and W : ∆ → L(X) defined by W (λ) = λ(λR(λ) − I).
Then the following
n assertions hold. o
1. D = x ∈ X : ∃ lim W (n)(x) ;
n
2. V (x) = lim W (n)(x), x ∈ D.
n

Proof. Let R : ∆ → L(X) be a L∞ -type pseudo-resolvent. Put

n o
M = x ∈ X : ∃ lim W (n)(x) .
n
Let µ ∈ ∆, x ∈ X and n ∈ N ∩ ∆. By the resolvent equation,
n(nR(n)R(µ)(x) − R(µ)(x)) = µnR(n)R(µ)(x) − nR(n)(x)
Letting n → ∞ we get
lim W (n)R(µ)(x) = µR(µ)(x) − x = V R(µ)(x).
n
Therefore,
R(µ)(X) = D ⊂ M.
Let now x ∈ M and z = lim W (n)(x). Then
n
R(µ)(z) = lim R(µ)W (n)(x) = lim W (n)R(µ)(x) = µR(µ)(x) − x.
n n
Thus, we have
R(µ)(z) = µR(µ)(x) − x and x = R(µ)(µx − z).
In conclusion, x ∈ D and the proof is complete. 
Theorem 2.3. A linear operator V : D → X is the generator of a
L∞ -type pseudo-resolvent if and only
6 Abdul Sami Awan and Mihai Voicu 6

1. V is closed and D = X.
2. ∃ ω ∈ R such that (ω, ∞) ⊂ ρ(V ) and (λI −V )(D) = X for all λ > ω.
3. ∀ p ∈ Cs (X), ∃ q ∈ Cs (X) such that
limsup p(n(nI − V )−1 (x)) 6 q(x), ∀ x ∈ X.
n

In this case, the mapping R : (ω, ∞) → L(X) defined by R(λ) = (λI −

V )−1 is a pseudo-resolvent of L∞ -type whose generator is V.
Proof. If R : ∆ → L(X) is a L∞ -type pseudo-resolvent whose generator
is V, then the three conditions on V are consequences of Definition 2.2 and
Theorem 2.1. If V fulfills the three conditions, one can define R : (ω, ∞) →
L(X) by R(λ) = (λI − V )−1 .
It follows from the last condition that ∀ p ∈ Cs (X) ∃ q ∈ Cs (X) such that
limsup p(nR(n)(x)) 6 q(x), ∀ x ∈ X.
n

Since R(λ)(X) = D, R fulfills the conditions of Theorem 2.1. Conse-

quently, R is a pseudo-resolvent of L∞ -type and V is its generator. 
Remark 2.4. If R0 : ∆0 → L(X) and R : ∆ → L(X) are pseudo-resolvents
with a common generator V : D → X, then
R0 (λ) = R(λ) = (λI − V )−1 for all λ ∈ ∆0 ∩ ∆.

Corollary 2.1. Let X be a complete locally convex space. A linear ope-

rator V : D → X is the generator of a L∞ -type pseudo-resolvent if and only
if it fulfills the three conditions below.
1. V is closed and D = X.
2. ∃ ω ∈ R such that (ω, ∞) ⊂ ρ(V ).
3. ∀ p ∈ Cs (X), ∃ q ∈ Cs (X) such that
limsup p(n(nI − V )−1 (x)) 6 q(x), ∀ x ∈ X.
n

Proof. If λ ∈ ρ(V ) then, according to Proposition 2.1,

(λI − V )(D) = X. 

Corollary 2.2. Let X be a Banach space. A linear operator V : D →

X is the generator of a L∞ -type pseudo-resolvent if and only if the three
conditions below hold.
1. V is closed and D = X.
2. ∃ ω ∈ R such that (ω, ∞) ⊂ ρ(V ).
3. The sequence (n(nI − V )−1 )n>ω is equicontinuous.
7 Pseudo-resolvents 7

Proof. Necessity: By Theorem 2.3, the sequence (n(nI − V )−1 )n>ω is

pointwise bounded and, consequently, uniformly bounded, i.e., equicontinuous.
Sufficiency: By Theorem 2.3 the three conditions are also sufficient. 
Remark 2.5. Corollary 2.3 covers the similar result given by F. Hirsch
in [5].
Example 2.3. Let X = C([0, ∞)) denote the set of all complex continuous
functions defined on [0, ∞). For each n ∈ N consider the seminorm pn : X →
R defined by
pn (u) = sup |u(x)| .
x∈[0,n]
Thus, X is a Frèchet space whose topology is given by the family of seminorms
(pn )n∈N . Let i ∈ X be the identity function (i.e., i(x) = x, ∀ x ∈ [0, ∞)) and
the linear operator V : X → X defined by V (u) = −i2 u, i.e.,
V (u)(x) = −x2 u(x), ∀ x ∈ [0, ∞).
Let λ ∈ C with the property that Re(λ) > 0 and u ∈ X such that
(λI − V )(u) = 0X , λI(u) − V (u) = 0X ,
2
(λ + x )u(x) = 0, ∀x ∈ [0, ∞), u(x) = 0, ∀x ∈ [0, ∞), u = 0X ,
hence λI − V is one to one.
w
Let now w ∈ X be an arbitrary function and u = λ+i 2 . Then (λI −

V )(u) = w which shows that (λI − V )(X) = X. Let now u ∈ X, x ∈ [0, n]

and n ∈ N. Then we have
((λI − V )(u)) (x) = (λ + x2 )u(x) and |(λI − V )(u)(x)| > |λ| |u(x)| .
Hence
pn (λI − V )(u) > |λ| pn (u)
which is equivalent to
pn (λ(λI − V )−1 (w)) 6 pn (w), ∀ w ∈ X.
Therefore, V fulfills the conditions of Theorem 2.3 and, consequently, is
the generator of a L∞ -type pseudo-resolvent R : {λ ∈ C : Re(λ) > 0} → L(X)
defined by
w
R(λ)(w) = (λI − V )−1 (w) = for all w ∈ X.
λ + i2
Theorem 2.3 suggests a connection between the L∞ -type pseudo-resolvents
and semigroups. We have in this sense the following result.
Theorem 2.4. Let X be a sequentially complete locally convex space and
V : D → X a linear operator. The statements below are equivalent.
1. V is the infinitesimal generator of a Co -equicontinuous semigroup.
8 Abdul Sami Awan and Mihai Voicu 8

2. V is the generator of a L∞ -type pseudo-resolvent R : (0, ∞) → L(X)

such that (nR(n))k , n > 1, k ∈ N, be equicontinuous.
Proof. 1 ⇒ 2. Let T : [0, ∞) → L(X) be a semigroup whose generator is
V . By the characterization theorem for Co -equicontinuous semigroups ([15],
p. 246), D = X, (0, ∞) ⊂ ρ(V ) and (λI − V )(D) = X, ∀λ > 0. Moreover,
we have
Z ∞
−1
R(λ, V )(x) = (λI − V ) (x) = e−λt T (t)(x)dx, ∀ λ > 0 and x ∈ X.
0

Consequently, the family (λR(λ, V ))k , k ∈ N, λ > 0, is equicontinuous. By

Theorem 2.3, the mapping R : (0, ∞) → L(X) given by R(λ) = R(λ, V ) is a
L∞ -type pseudo-resolvent whose generator is V .
2 ⇒ 1. By Theorem 2.3, D = X, (0, ∞) ⊂ ρ(V ) and (λI − V )(D) = X,
∀ λ > 0. Now, it is sufficient to use the characterization theorem for Co -
equicontinuous semigroups mentioned above. Therefore, V is the generator of
a Co -equicontinuous semigroup S : [0, ∞) → L(X). Let W : (0, ∞) → L(X)
introduced above by W (λ) = λ(λR(λ) − I). In this context, we have S(t)(x) =
lim etW (n) (x), ∀t > 0 and x ∈ X. 
n
Remark 2.6. The class of generators of L∞ -type pseudo-resolvents is a
proper extension of the class of the infinitesimal generators of Co -equicontinuous
semigroups acting on sequentially complete locally convex spaces.

REFERENCES

[1] I. Cuculescu, Markov Processes and Excessive Functions. Ed. Academiei, Bucureşti,
1968. (Romanian)
[2] E.B. Davies, One Parameter Semigroups. Academic Press, London, 1980.
[3] K. Engel and R. Nagel, One Parameter Semigroups for Linear Evolution Equations.
Springer-Verlag, Berlin, 1999.
[4] E. Hille, Functional Analysis and Semigroups. Amer. Math. Soc., Colloq. Publ. 31.,
Providence, RI, 1948.
[5] F. Hirsch, Families résolvants, genérateurs, cogenérateurs, potentiels. Ann. Inst. Fourier
22 (1972), 89–210.
[6] T. Kato, Remarks on pseudo-resolvents and infinitesimal generators of semigroups.
Proc. Japan Acad. 35 (1959), 467–468.
[7] W. Arendt, A. Garabosch, G. Greiner, U. Groh, H.P. Lotz, U. Moustakas, R. Nagel,
F. Neubrander and U Schlotterbeck, One-parameter Semigroups of Positive Operators.
Lecture Notes in Math. 1184. Springer-Verlag, Berlin, 1986.
[8] A. Pazy, Semigroups of Linear Operators and Application to Partial Differential Equa-
tions. Appl. Math. Sci. 44. Springer-Verlag, New York, 1983.
[9] M. Voicu, Positive resolvents and the property of dispersiveness in locally convex lattices.
Rev. Roumaine Math. Pures Appl. 33 (1988), 5, 471–477.
9 Pseudo-resolvents 9

[10] M. Voicu, Dissipative operators and resolvents. Rev. Roumaine Math. Pures Appl. 42
(1997), 5-6, 449–459.
[11] M. Voicu, Dissipative and accretive operators on locally convex spaces. Rend. Circ. Mat.
Palermo (2) Suppl. No. 52, Vol. II (1998), 805–815.
[12] M. Voicu, Resolvents on locally convex spaces. In: Order Structures in Functional Analy-
sis, 4, pp. 204–258. Ed. Acad. Române, Bucureşti, 2001.
[13] M. Voicu, Pseudo-resolvents and semigroups on locally convex spaces. In: Order Struc-
tures in Functional Analysis, 5, pp. 198–232. Ed. Acad. Române, Bucureşti, 2006.
[14] M. Voicu, Pseudo-resolvents on locally convex spaces. Rend. Circ. Mat. Palermo (2)
Suppl. No. 76 (2005), 655–665.
[15] K. Yosida, Functional Analysis. Springer-Verlag, Berlin, 1980.

Received 21 May 2007 School of Mathematical Sciences

68-B, New Muslim Town
Lahore, Pakistan
[email protected]
and
Technical University of Civil Engineering
Department of Mathematics
B-dul Lacul Tei 124
020396 Bucharest, Romania
[email protected]