This paper deals with the study of some operator inequalities involving the power
q-bounded operators along with the most known properties and results, in the
more general framework of locally convex vector spaces.
1. Introduction
Let X be a Hausdorff locally convex vector space over the complex field C. By calibration for the locally convex space X we understand a family P of seminorms generating the topology τP of X, in the sense that this topology is the coarsest with respect to the fact that all the seminorms in P are continuous. Such a family of seminorms was used by the author and Wu [1] and many others in different contexts (see [2–5]).
It is well known that calibration P is characterized by the property that the set
(1)E(p,ϵ)={x∈X:p(x)<ϵ},ϵ>0,p∈P
is a neighborhood subbase at 0. Denote by (X,P) the locally convex space X endowed with calibration P.
Recall that a locally convex algebra is an algebra with a locally convex topology in which the multiplication is separately continuous. Such an algebra is said to be locally m-convex (l.m.c.) if it has a neighborhood base U at 0 such that each U∈U is convex and balanced (i.e., λU⊆U for |λ|≤1) and satisfies the property U2⊆U.
Any algebra with identity will be called unital. It is well known that unital locally m-convex algebra A is characterized by the existence of calibration P such that each p∈P is submultiplicative (i.e., p(xy)≤p(x)p(y), for all x,y∈A) and satisfies p(e)=1, where e is the unit element.
An element a of locally convex algebra A is said to be bounded in A if there exists α∈C such that the set {(αx)n}n≥1 is bounded in A (see [6]). The set of all bounded elements in A will be denoted by A0.
Let C∞:=C∪{∞} be the Alexandroff one-point compactification of C. Following Waelbroeck [7, 8], we introduce the following.
Definition 1.
We call resolvent set in the Waelbroeck sense of an element x from a locally convex unital algebra (X,P) the set of all elements λ0∈C∞ for which there exists V∈Vλ0 such that the following conditions hold:
the element λe-x is invertible in X, for any λ∈V∖{∞};
the set {(λe-x)-1:λ∈V∖{∞}} is bounded in (X,P).
The resolvent set in Waelbroeck sense of an element x will be denoted by ρW(x). The Waelbroeck spectrum of x will be defined as
(2)σW(x):=C∞∖ρW(x).
2. q-Bounded Operators
Following Michael [9] (see also [2, 10]), we introduce the following.
Definition 2.
We say that a linear operator T:X→X is q-bounded (quotient-bounded) with respect to P if for any p∈P there exists cp>0 such that
(3)p(Tx)≤cpp(x),∀x∈X.
Denote by QP(X) the set which consists of all q-bounded operators with respect to calibration P.
For a seminorm p∈P, the application p^:QP(X)→R defined as
(4)p^(T)=inf{r>0:p(Tx)≤rp(x),∀x∈X}
is also a seminorm. Note that
(5)p^(T1T2)≤p^(T1)p^(T2),T1,T2∈QP(X),p∈P.
We denote by P^ the family of seminorms {p^:p∈P}. The space QP(X) will be endowed with a topology τP^ generated by P^. Remark that [9, Proposition 2.4(j)] implies that under this topology QP(X) becomes a Hausdorff locally m-convex topological algebra (in the sense of [9, Definition 2.1]).
If T∈QP(X), the P-spectral radius, denoted by rP(T), is considered as the boundedness radius in the sense of Allan [6] (see also [11–13]),
(6)rP(T)=inf{λ>0:thesequence((λ-1T)n)n∈NhhhhhisboundedinQP(X)((λ-1T)n)n∈N},
where, by common consent, inf∅:=+∞.
The set of all bounded elements in QP(X) will be denoted by (QP(X))0 (see [12]). It easily follows from [6, Proposition 2.14(ii)] that
(7)(QP(X))0={T∈QP(X):rP(T)<∞}.
For T∈(QP(X))0 we denote by ρW(T) the Waelbroeck resolvent set of T and by σW(T) the Waelbroeck spectrum of T. The function
(8)ρW(T)∋λ⟼R(λ,T)∶=(λI-T)-1∈(QP(X))0
is called the resolvent function of T. It is well known that
(9)R(λ,T)=∑n=0∞Tnλn+1.
In this paper we evaluate the behaviour of the power of a q-bounded operator from the algebra (QP(X))0 by some type of approximations. The main results have been announced in [14].
3. The Main Results
We continue to employ the notations from the previous sections and we will introduce two types of operatorial approximations for operators from the algebra (QP(X))0 which approximate a given operator T on a convergent power bounded series. The power boundedness problem for operators acting on Banach spaces was largely developed in various frameworks by many authors (see [15–17]).
In the following, using the functional calculus from the (QP(X))0 algebra (see [7, 8]), some important boundedness properties are obtained. Denote N*=N∖{0}. First we have the following.
Theorem 3.
If T∈(QP(X))0 satisfies
(10)supp^∈P^p^(Tk)≤C,
for k∈N*, then
(11)supp^∈P^p^[R(λ,T)k]≤C(|λ|-1)k,
for k∈N* and for all λ∈C with |λ|>1.
Proof.
Assume that supp^∈P^p^(Tk)≤C for k∈N*. Since
(12)R(λ,T)=∑j=0∞Tjλj+1,
for |λ|>1, then, by using the generalized binomial formula, we get
(13)R(λ,T)k=λ-k(I-Tλ)-k=1λk∑j=0∞(j+k-1j)Tjλj,
from where we deduce
(14)p^[R(λ,T)k]≤C|λ|k·∑j=0∞(j+k-1j)(1|λ|)j=C|λ|k·1(1-1/|λ|)k=C(|λ|-1)k,
for any k∈N* and any p^∈P^. Therefore, the conclusion is verified.
Conversely, we have the following.
Theorem 4.
If T∈(QP(X))0 and
(15)supp^∈P^p^[R(λ,T)]≤C|λ|-1,
for all λ∈C with |λ|>1, then
(16)supp^∈P^p^(Tk)≤Ce(k+1),
for k∈N*.
Proof.
Let us suppose condition p^[R(λ,T)]≤C/(|λ|-1) is true for all p^∈P^, for any k∈N* and |λ|>1. For k∈N* fixed, by choosing the integration path Γ:|λ|=1+1/k, with the aid of the functional calculus from the algebra (QP(X))0, we obtain
(*)Tk=12πi∫ΓλkR(λ,T)dλ.
Thus, for all p^∈P^, we have
(17)p^(Tk)≤12π∫Γ|λ|kp^(R(λ,T))dλ≤12π·maxλ∈Γ|λ|k·maxλ∈ΓC|λ|-1·∫Γdλ≤12π·(1+1k)k·Ck·2π(1+1k)≤Ce(k+1)
which implies the desired result.
Moreover, we can formulate the following.
Theorem 5.
If T∈(QP(X))0 and
(18)supp^∈P^p^[R(λ,T)k]≤C(|λ|-1)k,
for k∈N* and for all λ∈C with |λ|>1, then
(19)supp^∈P^p^(Tk)≤Ck!ekkk≤C2π(k+1),k∈N*.
Proof.
Integrating (*) by parts j-1 times, for j>2, we obtain
(20)Tk=(-1)j-12πi∫Γ(j+1)!λk+j-1(k+1)⋯(k+j-1)R(λ,T)jdλ.
Now choosing Γ the circle of radius 1+j/k and by using the hypothesis, for j→∞, we get
(21)supp^∈P^p^(Tk)≤Ck!ekkk≤C2π(k+1).
The last inequality was obtained by using Stirling’s approximation.
Now, for T∈(QP(X))0 we introduce (see [18]) the following.
Definition 6.
The Yosida approximation Y(λ,T) of T, for λ∈ρW(T)∩C, is defined as
(22)Y(λ,T)=λTR(λ,T).
Next theorem shows how an operator from the (QP(X))0 algebra is related to its Yosida approximation.
Theorem 7.
The Yosida approximation Y(λ,T) is analytic for λ∈ρW(T)∩C and the series representation
(23)Y(λ,T)=∑j=0∞Tj+1λj
converges for |λ|>rP(T). Moreover,
Y(λ,T)=λ2R(λ,T)-λI;
if there exists p^∈P^ such that rP(T)<p^(T), then
(24)p^(Y(λ,T)-T)≤p^(T2)|λ|-p^(T),
for |λ|>p^(T);
σW(Y(λ,T))={z/(1-z/λ),z∈σW(T)}.
Proof.
By evaluating Y(λ,T) in terms of the resolvent R(λ,T), for |λ|>rP(T) we obtain
(25)Y(λ,T)=λTR(λ,T)=λT(λI-T)-1=λT·∑j=0∞Tjλj+1=∑j=0∞Tj+1λj
from where it follows that the assertion of the theorem is true. Moreover,
(26)Y(λ,T)=∑j=0∞Tj+1λj=λI+T+T2λ+⋯+Tn+1λn+⋯-λI=λ2∑j=0∞Tjλj+1-λI=λ2R(λ,T)-λI,
so (1) is true.
To prove (2) one can observe that, from
(27)Y(λ,T)=∑j=0∞Tj+1λj,
it follows that
(28)Y(λ,T)-T=∑j=0∞T2λ(Tjλj)
on a set for which |λ|>rP(T). Moreover,
(29)p^(Y(λ,T)-T)≤∑j=0∞p^(T2λ)p^(Tjλj)≤p^(T2λ)·∑j=0∞p^(Tλ)j=p^(T2λ)11-p^(T/λ)=p^(T2)|λ|-p^(T),
for |λ|>p^(T)>rP(T).
A simple reasoning shows that R(λ,T)∈(QP(X))0; then it follows Y(λ,T)∈(QP(X))0.
From [19, Theorem 3.1.14], for |λ|>|z|, we have
(30)σW(Y(λ,T))=Y(λ,σW(T)),
for all z∈σW(T), and
(31)Y(λ,z)=∑j=0∞zj+1λj
on |λ|>|z|, which could be written as Y(λ,z)=z/(1-z/λ), for any z∈σW(T), so (3) is proved.
Below we state an equivalence between a power bounded operator from the (QP(X))0 algebra and the power of its Yosida approximation.
Theorem 8.
Let T∈(QP(X))0 and Y(λ,T) its Yosida approximation. Then the following assertions are equivalent:
supp^∈P^p^(Tk)≤c, for any k∈N*;
supp^∈P^p^(Y(λ,T)k)≤c/(1-1/|λ|)k, for any k∈N* and for all λ∈C with |λ|>1.
Proof.
Property (i) implies rP(T)≤1 so that the argumentation given in the proof of Theorem 7 implies that any λ∈C with |λ|>1 belongs to the resolvent set of T. Hence, using the generalized binomial formula, we get
(32)Y(λ,T)k=∑j=0∞(k+j-1j)Tj+kλj.
Now, by applying (i) again we obtain
(33)p^(Y(λ,T)k)≤c∑j=0∞(k+j-1j)(1|λ|)j=c(1-1/|λ|)k
for any p^∈P^, whence by passing to supremum, the inequality (ii) holds.
Conversely, (i) is a direct consequence of (ii).
For μ∈ρW(T), consider now the following Möbius transformation (see [20]):
(34)ψλ(μ)={(λ-1)μλ-μ,ifλ≠∞,μ,ifλ=∞.
Definition 9.
The Möbius approximation of T is defined as
(35)A(λ,T):=ψλ(T).
Proposition 10.
A(λ,T) is holomorphic in λ∈ρW(T)∩C and satisfies
(36)A(λ,T)=(1-1λ)Y(λ,T),λ≠0.
Proof.
Let λ∈ρW(T)∩C∖{0}. By evaluating the right member of the above equality, we get successively
(37)(1-1λ)Y(λ,T)=(1-1λ)λTR(λ,T)=(λ-1)TR(λ,T)=(λ-1)TλI-T=A(λ,T)
for λ≠∞. If λ=∞, then from Definition 9 we have A(λ,T)=T. On the other side (1-1/λ)Y(λ,T) converges to T, when λ→∞.
A similar result as in Theorem 8 is given below.
Theorem 11.
Let T∈(QP(X))0 and A(λ,T) its approximation as above. Then the following assertions are equivalent:
supp^∈P^p^(Tk)≤C, for any k∈N*;
supp^∈P^p^(A(λ,T)k)≤C, for any k∈N* and for every λ∈C with |λ|>1.
Proof.
From Theorem 8, for T∈(QP(X))0,
(38)supp^∈P^p^(Tk)≤C
is equivalent to
(39)supp^∈P^p^(Y(λ,T)k)≤C(1-1/|λ|)k.
The conclusion follows taking into account that
(40)A(λ,T)k=(1-1λ)k·Y(λ,T)k,
for k∈N*.
4. Application
For L>0 let X:=C[0,L] be the space of continuous functions on [0,L] endowed with the norm |u|L:=max[0,L]|u(t)|.
Consider T:X→X, given by
(41)Tu(t)=∫0tu(s)ds.
Following [19], we see that the resolvent of T is given by
(42)R(λ,T)u(t)=1λu(t)+1λ2∫0te(t-s)/λu(s)ds,
the Yosida approximation of T is
(43)Y(λ,T)u(t)=∫0te(t-s)/λu(s)ds,
and the Möbius approximation of T is
(44)A(λ,T)u(t)=(1-1λ)∫0te(t-s)/λu(s)ds.
Remark that, for all u∈C[0,L], we have
(45)|Tu|L=maxt∈[0,L]|Tu(t)|≤maxt∈[0,L]∫0t|u(s)|ds≤maxt∈[0,L]|u(t)|∫0Lds=|u|L·L.
The above implies that T is a contraction for L≤1.
If L>1, then we can introduce for each ɛ>0 the following norm on C[0,L]:
(46)∥u∥ɛ:=maxt∈[0,L]et/ɛ|u(t)|,u∈C[0,L].
Then a simple computation gives that
(47)∥Tu∥ɛ<ɛ∥u∥ɛ,u∈C[0,L].
On the other hand,
(48)∥u∥ɛ≤|u|L≤eL/ɛ∥u∥ɛ.
Remark that, by Theorem 11, for all λ>1, we get
(49)|A(λ,T)|L=(λ-1)(eT/λ-1)≤1
if and only if |T|L≤1.
It is clear that for estimating the powers of T it seems to be better to use the Yosida approximation or Möbius approximation than the resolvent approximation.
Conflict of Interests
The author declares that there is no conflict of interests regarding the publication of this paper.
Acknowledgment
The author is grateful to the anonymous referees for their very careful reading and for useful suggestions that helped in better exposing this material.
LemleL. D.WuL. M.Uniqueness of C0-semigroups on a general locally convex vector space and an application201182348549610.1007/s00233-010-9285-3MR27960392-s2.0-79956076549MooreR. T.Banach algebras of operators on locally convex spaces196975687310.1090/S0002-9904-1969-12147-6MR0236723PaterF.Properties of multipliers on special algebras with application to signal processing201154313318PaterF.A multiplier algebra representation with application to harmonic signal models1479Proceedings of the A.I.P. International Conference on Numerical Analysis and Applied MathematicsSeptember 2012Kos, Greece10751078KostićM.Abstract Volterra equations in locally convex spaces20125591797182510.1007/s11425-012-4477-9MR29608632-s2.0-84865546366AllanG. R.A spectral theory for locally convex alebras196515399421MR0176344WaelbroeckL.196031Académie Royale de Belgique. Classe des sciences. Mémoires. Col.WaelbroeckL.Algébrés commutatives: éléments réguliers195794249MichaelE. A.Locally multiplicatively-convex topological algebras195219521179MR0051444ZBL0047.35502JosephG. A.Boundedness and completeness in locally convex spaces and algebras19772415063MR0512300BonalesF. G.MendozaR. V.Extending the formula to calculate the spectral radius of an operator199812619710310.1090/S0002-9939-98-04430-XMR14591102-s2.0-33646977393PaterF.BinzarT.On some ergodic theorems for a universally bounded operator201026197102MR26767232-s2.0-77954185176PaterF.LemleL. D.On some multiplication operator algebra problem with application to stochastic signal models1558Proceedings of the International Conference on Numerical Analysis and Applied MathematicsSeptember 2013Rhodes, Greece16611664AIP Conference Proceedings10.1063/1.4825835PaterF.LemleL. D.BinzarT.On some Yosida type approximation theorems1168Proceedings of the International Conference on Numerical Analysis and Applied MathematicsSeptember 2009Rethymno, Crete521524AIP Conference ProceedingsKatznelsonY.TzafririL.On power bounded operators198668331332810.1016/0022-1236(86)90101-1MR859138ZBL0611.470052-s2.0-0000974959NagyB.ZemanekJ.A resolvent condition implying power boundedness19991342143151MR16882222-s2.0-0003379338NevanlinnaO.On the growth of the resolvent operator for power bounded operators, Linear Operators199738247264YosidaK.On the differentiability and the representation of one-parameter semi-group of linear operators19481152110.2969/jmsj/00110015MR0028537ZBL0037.35302NevanlinnaO.1993Basel, SwitzerlandBirkhäuser10.1007/978-3-0348-8547-8MR1217705ShieldsA. L.On Möbius bounded operators1978403-4371374MR515218