Scribd Logo
Upload

Welcome to Scribd!

  • [Mechanics_ Analysis] M.a. Krasnosel'Skii, P.P. Zabreyko, E.I. Pustylnik, P.E. Sobolev - Integral Operators in Spaces of Summable Functions (1976, Noordhoff) - Libgen.li

    Uploaded by

    Guillermo García Sáez
    6 views533 pages

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    6 views533 pages

    [Mechanics_ Analysis] M.a. Krasnosel'Skii, P.P. Zabreyko, E.I. Pustylnik, P.E. Sobolev - Integral Operators in Spaces of Summable Functions (1976, Noordhoff) - Libgen.li

    Uploaded by

    Guillermo García Sáez

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    Download as pdf
    of 533
    Integral operators in spaces of summable functions M.A. Krasnoselskii, P. P. Zabreiko E.I. Pustylnik, P. E. Sbolevskii Translated by T. Ando Hokkaido University Sapporo, Japan NOORDHOFF INTERNATIONAL PUBLISHING — LEYDEN Copyright © 1976 Noordhoff International Publishing A division of A. W. Sijthoff International Publishing Company B.V. Leyden, The Netherlands All rights reserved. No part of this publication may be reproduced, stored ina retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner. ISBN: 90 286 0294 1 Printed in The Netherlands by Kon. Drukkerij van de Garde B.V., Zaltbommel Contents Chapter 1. Linear operators in L, spaces... 2... 2.2... I hetspace ECE eae eae eae See Seeee 1.1. Description of the spaces 1.2. Criteria for compactness 1.3. Continuous linear functionals and weak convergence 1.4. Semi-ordering in the spaces S and L, 1.5. Projections and bases of Haar type 1.6. Operators in the spaces L, 2. Continuous linear operators. 6. oe 2.1. Linear operators 2.2. Regular operators 2.3. The M. Riesz interpolation theorem 2.4. Interpolation theorems for regular operators 2.5. Classes of L-characteristics of linear operators 2.6. On a property of regular operators 2.7. The Marcinkiewics interpolation theorem 3. Compact linear operators. . 2 6. 1 6 oe ee ee 3.1. Compact linear operators 3.2. Compactness and adjoint operators 3.3. Properties of operators compact in measure 3.4. Interpolation properties of compactness 3.5. Strongly continuous linear operators Chapter 2. Continuity and compactness of linear integral operators . . 4. General theorems on continuity on integral operators... .. « 4.1. Linear integral operators 4.2. Regular operators XI 48 64 Contents VI 4.3. Example of a non-regular operator 4.4. The adjoint operator 4.5. Operators with symmetric kernels 4.6. Products of integral operators 4.7. Truncations of kernels of integral operators . General theorems on compactness of integral operators . . . . . 81 5.1. Problem setting 5.2. Regular operators acting from Lo to Ly, and from L,, to Ly 5.3. Regular operators acting from L,, to Lg,, where 0 < a < 1, 0 0) 6.7. Integral operators acting from L, to C 6.8. vo-Cobounded linear operators 6.9. Compactness of vo-cobounded operators 6.10. Interpolation properties of wo-boundedness 6.11. On weakly compact operators in L, Contents 7. Integral operators with kernels satisfying conditions of Kantorovic 7.1. Simplest criteria 7.2. Theorems with intermediate conditions 7.3. Lemmas 7.4. Applications of theorems on adjoint operators 7.5. Fundamental theorems 7.6. Conditions of ‘Kantorovic’ type 7.7. Summability of kernels of integral operators 8. Operators of potential type... 1. ee 144 8.1. Definitions 8.2. Simplest theorems on continuity and compactness of potentials 8.3. Interpolation theorem of Stein-Weiss 8.4. Limit theorems on continuity of potentials 8.5. Operators of potential type 8.6. The logarithmic potential 8.7. Iterates of operators of potential type 8.8. Generalizations to the case of distinct dimensions 8.9. Potentials with respect to non-Lebesgue measure Chapter 3. Fractional powers of selfadjoint operators. . . . ... . 173 9. Splitting of linear operators... .... Bet pea ee teat 9.1. Square root of selfadjoint operators 9.2. Splitting of an operator 9.3. L-Characteristic of a square root 9.4. Representation of compact operators 9.5. Square root of integral operator 9.6. Example 9.7. Investigation of integral operators by means of properties of iterated kernels 9.8. Remark on Mercer’s theorem 10. Fractional powers of bounded operators... 1... 66s ee 191 10.1. The spectral function 10.2. Fractional powers of bounded selfadjoint operators vil Contents 12. 10.3. The fundamental theorem 10.4. Operators in real spaces 10.5. Fractional powers of compact operators 10.6. L-Characteristics of fractional powers of operators 10.7. Fractional powers of integral operators . Unbounded selfadjoint operators. . 2... ee ee ee 205 11.1. Closed operators 11.2. Adjoint operators 11.3. Integration with respect to spectral functions 11.4. The fundamental theorem on spectral representation of un- bounded selfadjoint operators 11.5. Functions of selfadjoint operators 11.6. Commuting selfadjoint operators 11.7. Integrals of operator-functions 11.8. Integral representation of fractional powers of an operator Properties of fractional powers of unbounded operators... . . 222 12.1. Problem setting 12.2. The moment inequality for fractional powers 12.3. Subordinate operators 12.4. Subordination of fractional powers 12.5. Heinz’ first inequality 12.6. Heinz’ second inequality 12.7. Fractional powers of projected operators 12.8. On a special class of selfadjoint operators 12.9. Theorems on splitting 12.10. Theorems on fractional powers 12.11. L-Characteristic of fractional powers Chapter 4. Fractional powers of operators of positive type... . . « 248 13. VII Semi-groups of operators... 2 6 oe 248 13.1. Vector-functions and operator-functions 13.2. Unbounded operators 13.3. Resolvents 14. Contents 13.4. Definition of a semi-group 13.5. Generator of a semi-group 13.6. Theorem of Hille-Phillips-Miyadera 13.7. Analytic-semi-groups 13.8. Estimates for the operators A"7(t) Fractional powers of positive-type operators... .. 1... 279 14.1. Positive-type operators 14.2. Negative fractional powers 14.3. Positive fractional powers 14.4. A moment inequality 14.5. Operators subordinate to fractional powers of a positive- type operator 14.6. General theorems on subordination 14.7. Estimates for elements of the form BA™ ‘x 14.8. Comparison of fractional powers of two operators 14.9. Fractional powers of positive-type generators 14.10. Compactness of fractional powers 14.11. Supplementary remarks |. Moment inequalities and L-characteristics of fractional powers . . 306 15.1. Lorentz spaces 15.2. Linear operators 15.3. Interpolation theorems 15.4. Fundamental theorems 15.5. L-Characteristics of fractional powers 15.6. One more theorem on compactness . Fractional powers of elliptic operators... oe ee 324 16.1. Elliptic differential expressions 16.2. Elliptic operators 16.3. Positive-type elliptic operators 16.4. Multiplicative inequalities and fractional powers of elliptic operators 16.5. L-Characteristics of negative fractional powers of elliptic operators x Contents 16.6. 16.7. Further theorems On integral representations of fractional powers of elliptic operators Chapter 5. Non-linear integral operators... .........- 349 17. The superposition operators. . 6. 0 oo ee 349 19. 17.1. 17.2. 17.3. 17.4. 17.5. 17.6. 17.7, 17.8. 18.1. 18.2. 18.3. 18.4. 18.5. 18.6. 18.7. 18.8. On functions which are continuous in one variable Simplest properties of the superposition operator Fundamental theorems Examples General form of L-characteristics of superposition operators Uniform continuity of the superposition operator Improvement of superposition operators Supplementary remarks . Conditions for continuity of integral operators... 1... 4% 373 Definitions and simple properties Conditions for continuity of Uryson operators General theorem on continuity of Uryson operators On a property of Uryson operators Regular Uryson operators Special examples Uryson operators with values in the space of bounded functions On uniform continuity of Uryson operators Conditions for complete continuity of an Uryson operator . . . . 398 19.1. 19.2. 19.3. 19.4. 19.5. 19.6. Problem setting Hammerstein operators Complete continuity of regular Uryson operators acting from Ly to Ly, Be (0, 1] Complete continuity of regular Uryson operators acting from L, toL,,«>0,0 1, the space L, is not normable; however it becomes a complete metric space with metric p(x, y) = (x — yllz)'/*. Convergence in L, is usually called convergence in mean with exponent 1/a. Together with the spaces L, there is usually considered the space M of essentially bounded functions x(t) with norm Ilo = ess sup|x(s)]. (1.3) sen This space is also denoted by Lo, because sae lixlle = Ilo (EL) = M). (1.4) a In contrast to the spaces L, (0 < « < 00), the space Ly is non-separable. The space C of uniformly continuous functions on Q is the most important subspace of the space Lo. The space C is separable if the set Q is bounded. In the sequel it is necessary to examine different spaces L, simul- taneously. In such cases Hélder’s inequality: Ix-yllare S lla yllp Ee Ly ye Ly), (1.5) valid for any a, B < [0, 00), plays a basic role. From Hélder’s inequality it follows, in particular, that the function A(a) = ||x\|, for any fixed element x(s) is continuous and logarithmically convex. This means that for any 7 € (0, 1) the following inequality is fulfilled: U>tllece S lll2y “le, (1.6) where a(t) = (1 — tay + Toy. § 1 The spaces L, The inequality (1.5) implies that for a, < a the space L,, is imbedded in the space L,,, ic. L,, < L,, and Illa S klix, = (EL, k= (mes Q)*-*). (L7) 1.2 Criteria for compactness Let D be a measurable subset of 2. Denote by Pp the linear operator, defined by the relation ‘x(s), if se D 0, if s¢D. (1.8) Ppx(s) = { It is clear that PZ = Pp and! ||Ppll,+. = 1, if mes(D) # 0. The norm in the spaces L, for « > 0 possesses the property of absolute continuity. This means that for each fixed function x(s) €L, lim ||Ppxllz = 0. mes D0 A family IN of functions in L, is said to have equi-absolutely continuous norms, if for any ¢ > 0 there can be found a 6 > 0 such that mes(D) < 6 and x(s)€ MN imply the inequality ||Ppx(s)||, < & It is clear that convergence of a sequence of functions x,(s) to x9(s) in the norm of L,, implies convergence of this sequence in measure (but not almost everywhere!). The converse is not true. What is valid for the spaces L, (0 < « < 0) is that a sequence x,(s), convergent to a function x(s) in measure, converges in the norm of L, (0 < a < 0) if and only if the functions x,(s) have equi-absolutely continuous norms. This assertion immediately implies the following simple criterion for compactness in the spaces L, (0 < « < 00). Lemma 1.1: A family M of functions x(s)eL, (0 < « < 0) is compact if and only if it is compact in measure and It has equi-absolutely continuous norms. 1 ||Alla_,¢ denotes norm of A as an operator acting from La to Lp (see 2.1°). 1 Linear operators in L, spaces In the study of concrete families of functions, a direct verification of com- pactness in measure is often difficult to achieve. Hence the following scheme of arguments is usually applied. First it is shown that the set Nt of functions in L, which is in question is compact in some space L,,, where a > «. It then follows that 9 is compact in measure. To check equi-absolute continuity of norms of the functions in a set It, it is sometimes convenient to use the following criterion of Vallée-Poussin: The norms of functions x(s) in a set Mt are equi-absolutely continuous, if and only if there exists an even continuous function ®(u), satisfying the condition -_ P(u) lim = 0, uto Ul Sor which IPL SA (xe M). Now examine the space Ly. The norm in this space does not possess the property of absolute continuity. In fact, for each non-zero function x(s) € Lo, it is possible to find sets D, < Q such that mes(D,)>0 but ||Pp,*llo = [Ixllo (# = 1, 2,-..). Lemma 1.2: A bounded set M of functions in Lo is compact if and only if for any & > O there exists a partition of the set Q into a finite number of subsets Q4, ..., Quy 2=2,02,0...09, such that for any function x(s)eM ess sup|x(s’) — x(s")| f(xo) follows from ||x, — xl] > 0. There does not exists any non-zero continuous linear functional on the spaces L,, where a > 1. In the sequel, we will describe the continuous linear functionals on those spaces L,, where « € (0, 1]. For continuous linear functionals f(x), defined on a Banach space E (in particular, on a space L,, «€ [0, 1]), we introduce the notion of the norm of a functional: If] = sup |f@)I. last The totality of continuous linear functionals on a Banach space E also constitutes a Banach space. This space is denoted by E*; it is called the dual space (conjugate, adjoint) of E. The expression (9) = J x(9)¥@)ds ) is called the scalar product of the functions x(s), y(s). Hélder’s inequality 5 1 Linear operators in L, spaces implies that |(x, y)| < 0 if xeL,, yeL,-, (0S 1). It is of course possible for the scalar product to be finite in other cases. Let 0 S$ a S 1. The following fact plays an important role; a(t) €L;_., if the inequality |(x, a)| < 0o is fulfilled for all functions x(s) ¢ L,. Further- more llall1-2 = sup |(x, a)|. (1.10) Wxllest As it turns out, each continuous linear functional f(x), defined on L, (0 < a S 1), can be written in the form f(x) = (x, 4), where ae L,_,, and ||f|| = |lal|,-.. In this connection the space Ly-_, is called the dual of L,. The expression f(x) = (x, a) for a€L, becomes a continuous linear functional on Lj; however there exist other continuous linear functionals on Lo; L, can be considered as a subspace of the space L% dual to Lo. A space E is called reflexive, if ' (E*)* = E. The above observations show that the spaces L, for 0 < « < | are reflexive, but those with « = 0,1 are not reflexive. The space L, is Hilbertian. A sequence x, € E is called weakly convergent, if, for every f € E*, the sequence f(x,,) converges. A sequence x, € E converges weakly to an element Xo; if for every f € E*, the sequence f(x,) converges to f(xo). If each weakly convergent sequence converges weakly to some limit, the space E is called weakly complete. A space E is called weakly compact, if it is possible to choose from each bounded sequence a weakly convergent subsequence. Each reflexive space is weakly complete and weakly compact. Hence the spaces L, for « € (0, 1) are weakly complete and weakly compact. The space L, is weakly complete but does not possess the property of weak compactness. LEMMA 1.3:7 A set Mt of functions in L, is weakly compact if and only if the norms of the functions in Mt are equi-absolutely continuous. + The following relation is understood in the sense that each functional g © E** has the form 9(f) = f(x), where xo is some element of E. ? J, P, Natanson [1], N. Dunford and J. T. Schwartz [1]. 6 $1 The spaces L, In the sequel we will consider a special notion of weak convergence for the space Lo, one which is distinct from what is defined above. Let F be a linear subspace of the space E*. A sequence x, € E is called F-weakly convergent, if, for each f € F, the sequence f(x,) converges. The notions of F-weak completeness and F-weak compactness are introduced in the obvious way. If F is separable the space E is F-weakly compact. In the space Ly we will always mean by ‘weak convergence’ the weak convergence, defined by the subspace F = L,. Since L, is separable, the space Ly is weakly compact with respect to this convergence. It can be shown that Ly also possesses the property of weak completeness with respect to this convergence. In conclusion we note that weak convergence does not imply conver- gence in measure, and convergence in measure does not imply weak con- vergence. But if a sequence x, converges weakly and converges in measure, then the weak limit coincides with the limit in measure. Sometimes it is convenient to use the following remark: if a sequence x, converges weakly to Xo and is compact in measure, then it converges to Xo in measure. 1.4 Semi-ordering in the spaces S and L, It is convenient to consider the spaces S and L, (0 S a < 0) in their natural semi-ordering: a function x(s) is smaller than a function y(s), if x(s) < y(s) for almost all s ¢ Q. This semi-ordering possesses the following properties: 1°) xSy, y Sz imply x Sz. 2°) x Sy, y Sx imply x =y. 3°) x Sy implies that x + zy +z for any z. 4°) x Sy, 420 implies Ax < dy. 5°) If x,(s) S y,(5) (2 = 1, 2,...) and the sequences x,(s) and y,(s) con- verge in measure (or in the norm of L,) respectively to x*(s) and y*(s) as n— oo, then x*(s) < y*(s). In other words, inequality is preserved under the limit operation. In a semi-ordered space (in particular, in S or L,) there are defined the notions of supremum and infimum of a set It. An element z, (z2) is called the supremum (infimum) of a set I and is denoted by sup Mt (inf M), if x S z, (x 2 z,) for each xe Mt and if x < u (x = u) being valid for all elements x € M, implies z, < u (u = z2). WA Each set in S or L, which is bounded from above (below), has a supre- 7 1 Linear operators in L, spaces mum (infimum). If the set Yt contains not more than a countable number of elements then the following formula is valid: 2,(s) = sup M = sup x(s). xe® In the general case the relation takes the form: 2,(s) = sup {x,(s)} in MN. Similar where {x,(s)} is any countable set of functions of Nt, dense formulas are valid for the infimum. Let us present several properties of the spaces L,, connected with their semi-ordering. 1°) Each element x L, is uniquely represented in the form: X=X,—X- (X4,x- 20), where x4 = sup(x, 0), x_ = sup(—x, 0). Here loa lle + ella S 2 MIM 2°) 0S x Sy implies |x|, < lvl. 3°) In the space ZL; the norm ||x||, possesses the property lx+yli = lel + ll Gy» 20). 4°) In the space Ly the norm ||x||9 possesses the property IIsup {x1, x2}llo = max{[|x: Ilo» llallo} 1s 2 2 0). 5°) Let «€(0, 00). Then each monotone sequence x, (n = 1, 2,...) which is bounded in norm converges to some element x*. * For MC Lo this refers to denseness in the S metric; the latter property actually suf- fices for MCLa0 S a < co [Ed]. 8 § 1 The spaces L, For a = 0 this assertion is not true: the sequence will merely converge to some element almost everywhere. 1.5 Projections and bases of Haar type In the constructions of this section the condition « < 1 is essential. We are interested in special bases in the spaces L,. Their construction is connected with certain finite dimensional projections, acting in the spaces L, (averaging operators). Suppose that the set Q is partitioned into a number of disjoint parts: 2=02,02,0...UQ,. (1.11) Define the operator P by the relation Px(s) = f K(s, 0) x(a)do (1.12) a where 1 K(s,¢) = 4 mes Q; 0. ,ifs,oeQ, ((=1,2-..,9) , for other s,0€Q. In other words Px(s) = - ; [rca (seQ, i=1,2,...,q). (1.13) mes Q; a It is easy to see that the operator P is defined on all summable functions. Its values belong to a finite dimensional space—the linear span of the totality of characteristic functions of sets Q,,...,Q,. Thus the operator (1.12) acts in each space L,(Q) (0 S$ « S 1). First let « > 0. Then for each function x(s) ¢ L, we have, by Hélder’s inequality, [[Px(s)/tds = ¥ (mes 24)!" | f x(0) do!" ay a zt Ms SX (mes 2°" f |x(o)|**do- (mes Q)"-* = f |x(0)|*"do i X Q i 1 1 Linear operators in L, spaces ie. Pxlla S IIxlla (xe L,). (1.14) In the case a = 0, inequality (1.14) follows in trivial way from (1.13). Consequently, the operator (1.12) is continuous in each space L,(Q) O 0 there can be taken sets 2{”, ..., Q" in the partition (1.15) such that [mes [(Q'” U ... UV Q) 0 F] — mes(F)| xo in La, for « > 0. (Ed.) 11 1 Linear operators in L, spaces where a is some positive number. Since the sequence P,, is regular, it is possible to find ng such that for each n > ng there can be found sets 2” in the partition (1.15) satisfying mes(of? a F) = 2Molo + Peso eee Qi = 2lzollo + %) ed Then for s¢Q the following inequality will be fulfilled: 1 Patol) = Sop J zo(a)do - 2; 1 z aap | zo(0) do — J I) 2M%i9 0 F 2019—F mes(Q{” 0 F) Kole mes Q{") — mes(Q® 9 F) > % = ——_2 —— = llzollp ——_*>"___, > —__ 2 =. =e tes (QM) 7 mes(Q() free This means that the sequence P,Z) does not converge to zero. We arrived at a contradiction. Lemma 1.5: Let E be an arbitrary separable subspace of the space Lo(Q). Then there exists a regular sequence of operators P, which is strongly con- vergent to the identity operator in E: lim ||P,,x(s) — x(s)llb =0 (xe E). (1.19) Proor: Let a sequence x;,X2,... be dense in the ball ||x||) <1 of the space E. Denote by D{" the set of points t at which inSx(s)<(G+D/n (= —-n, -n4+1,...,0,1,...,0). Take as the n-th partition (1.15) the family Q¢”, ..., 2, of disjoint sets such that each set D\ (k = 1,...,) can be represented in a form of a union of some sets in ‘this family. It is clear that with this choice of the sets Q(” (i = 1,..., q(n); n = 1, 2,...) the sequence P,X_ converges to Xp 12 § 1+ The spaces L, as n > 0 for any m. As is easily seen, the sets Q{” can be chosen to have such small measure that the sequence P, becomes regular. Now let x(s) be an arbitrary function in E. Then for any ¢ > 0 there is a function x,,(s), such that ||x — cxpllo < ¢/2, where c = ||x||o. Hence for sufficiently large n the following inequality will be fulfilled: Pix — xllo S Pile — eX m)llo + €llPuXm — Xmllo + IX — C%mllo < & The lemma has been proved. In particular, for the inclusion C(Q) < L,(Q; {P,}) it is sufficient that the maximum diameter of the sets Q{” in the partition (1.15) converges to Zero as n — 00. The assertions of Lemmas 1.4 and 1.5 are equivalent to an assertion on expansion of functions x(s) in the form x(s) = Pyx(s) + (Pz — Py)x(s) +... + (Pa — Py) x(9) +... (1.20) Naturally this raises the problem of whether the series (1.20) is an expansion with respect to some basis. In order that the series (1.20) be an expansion with respect to a basis, it is necessary that each of operators P,,, — P, (n = 1, 2,...) be a pro- jection on a one-dimensional subspace. For this it is, in turn, sufficient that the (n + 1)-st partition be obtained from the n-th partition (1.15) by replacing one of the sets 2{” by two disjoint subsets whose union is equal to 2”. Consequently, the constructions of the preceding section reveal the possibility of constructing various systems of functions which are bases simultaneously in all the spaces L, (0 < « S< 1) as well as in some space Ec Ly, containing the space C. In particular, with a suitable choice of partitions (1.15) we arrive at the well known Haar basis [1] (see also M. A. Krasnoselskii and Y. B. Rutickii [5]). 1.6 Operators in the spaces L, The study of integral, integro-differential and other types of equations is usually simplified, if they can be considered as operator equations in 13 1 Linear operators in L,, spaces function spaces. A successful choice of spaces guarantees ‘good’ properties of the operators in the equations: continuity, complete continuity, differ- entiability, and so on. In the study of operators in the spaces L, it is natural to seek those a, B for which these operators possess needed properties. The notion of L- characteristic of a given operator then becomes a convenient method for the description and study of properties of operators in the spaces L, for various «. Let A be some concrete operator, for instance, an integral operator Ax(t) = J K[t, s, x(s)] ds, a acting from some space of functions, defined on a set Q, to a space of functions defined, generally speaking, on another set 2*. The set L(A; def.) (Fig. 1.1) of all points {«, 6} in the quadrant a, 6 = 0 such that the operator A acts from L, to Ly, is called the L-characteristic of the operator A. It possesses the important property of extrapolation: if {a 9, Bo} € L(A; def.) then {a, B} € L(A; def.) for a S %, BZ Bo. In fact, if A acts from L,, to L,,, then it will also act from the smaller space L, (& $ a) to the larger space L, (6 = By). My MM) asa, VL (A, det.) HTL 1 L(ALR) Y (B=5(a:A; det.) B= COAL) _ « Figure 1.1 ef Figure 1.2 § 1 The spaces L, The property of extrapolation implies that the function E(a) = E(a; A; def.) = inf B (1.21) (a, B)e L(A; def.) is defined either on some semi-interval [0, %)) or on some interval [0, #] and is non-decreasing. It is clear that {{a, B}; B > &(a)} < L(A; def.) < {{a, B}: B 2 E(w}. (1.22) The set L(A; def.) possesses some simple properties resulting immediately from the definition. 1°) For the L-characteristics of any two operators A and B the following inclusion is valid: L(A + B; def.) > L(A; def.) > L(B; def.) and so is the inequality: &(a; A + B; def.) S max {&(a; A; def.), €(a; B; def.)}. If &(a; A) # E(a; B) for all «, then L(A + B; def.) = L(A; def.) 0 L(B; def). 2°) The following inequality is valid: E(a; C; def.) < (a; A; def.) + E(w; B; def.), where C denotes the operator, defined by the relation Cx = Ax- Bx. 3°) Let A and B be two operators. The L-characteristic L(AB; def.) contains the set of points {«, 8}, for which there exists a number y, such that {a, y} € L(B; def.), {y, B} ¢ L(A; def.). Hence (if ¢(«; A; def.) is continuous from the right) 1 Linear operators in L, spaces &(a; AB; def.) < ElE(a; B; def.); A, def.]. It is natural to examine subsets of the L-characteristic L(A; def.), consisting of points {«, B} such that the operator A acts from L, to L, and possesses some supplementary property ft: continuity, compactness etc. These sub- sets will be denoted by L(A; 8) (Fig. 1.2) and will also be called L- characteristics of the operator A. For most of the properties, studied in the sequel, the sets L(A; 9) will possess the property of extrapolation. As with the L-characteristic L(A; def.), the sets L(A; ) in such cases are defined in essence by monotone functions &(;A;R) = inf {a B) eL(4,%) It is clear that (a; A; R) = (a; A; def.). We will often study the L- characteristics L(A; cont.) and L(A; comp.), corresponding to the properties of continuity and compactness of the operator A. Consider, as a simple example, an operator A defined by the relation Ax(s) = a(s)x(s), where a(s) is a measurable function. Assume that a(s)eL,. Then (1.5) implies that Ax(s) is a function in L,,,, if x(s)eL,, and Ax)Ilp+a S lal, Ix) lee WLLL L(A; Cont.) Y y 7 a(t)eL NU L. ° rove icacconty L(A;Cont.) 7 I altie MLNAL Subs % Ry « Figure 1.3 § 2 > Continuous linear operators In other words, the L-characteristic L(A; cont.) contains the set of points {a, B}, for which B = y + a. The L-characteristic L(A; cont.) of this operator can be determined completely. Let yo be such that a(s)€L, for y > yo, but a(s) ¢L, for y < yo. It is not difficult to see that for the case a(s) € L,, the set L(A) = L(A; cont.) coincides with the set of points {«, 8}, for which B = yo + a, while in the case a(s)¢L,,, L(A) coincides with the set of points {a, 8}, for which B> Yo + & (Fig. 1.3). The notion of L-characteristics has been systematically applied by many authors to yield a geometric description of interpolation theorems for linear operators. A number of assertions on general properties of L- characteristics are mentioned in the paper of P. P. Zabreiko and M. A. Krasnoselskii [1]. § 2 Continuous linear operators 2.1. Linear operators‘ In this section we recall certain fundamental definitions concerning linear operators, acting from one space E, to another space E,. An operator A is linear, if it is additive and homogeneous: A(ayX1 + 0X2) = &AXy + AX. (2.1) A linear operator A is continuous, if it transforms sequences convergent in norm to sequences convergent in norm. If E, and E, are Banach spaces, a continuous linear operator also transforms each weakly convergent sequence to a weakly convergent sequence. We assign to a continuous linear operator A its norm ||All: ||Al = sup | Axlle,. (2.2) Welle, $1 + General problems in the theory of linear operators in function spaces are discussed in many books on functional analysis (see S. Banach [1], L. A. Ljusternik and V. I. Sobolev [1], L. V. Kantorovic and G. P. Akilov [1], A. Zaanen [3], E. Hille and R. Phillips [1], V. I. Smirnov [1]. See the presentation of these problems and the bibliography in N. Dunford and J. T. Schwartz [1]. 17 1 Linear operators in L, spaces Formula (2.2) makes sense in the case of any Banach spaces E, and E>. We will also apply it to the case in which E, and E, are spaces L, and L,, where the numbers o and f can be greater than 1. The norm of a linear operator A, acting from L, to Ls, will be denoted by ||All.+5- In this paragraph we are interested in problems connected with the continuity of linear operators, acting from L, to Ls. Proving the continuity of such operators is equivalent to giving a proof of the inequality Axl SMllxl. (xe L,). (2.3) It is clear that the infimum of those M for which (2.3) is fulfilled is equal to [Alle It turns out that there does not exist any non-zero continuous linear operator acting from L, to L, if « > 1 and f < a. For the proof it suffices to show that any continuous linear operator A acting from L, to L, vanishes on all characteristic functions «p. Partition the set D into n parts D,,..., D, of equal measure. It is clear that pila = 2 “(mes D)* (i = 1, 2, ...5 0), hence |Akp,llp S |Allaspn” “(mes D)® (i = 1, 2,..., 0). First let 6 S$ 1. Then [Avy = 13 Ano SE 1Av op S HAlp (mes DY nt If 8 > 1, then Uv oly = IEA olp S (AK lp"? S VA glmes Dyn! These inequalities and the arbitrariness of n imply that Axp = 0. This assertion means that the L-characteristic L(A; cont.) of each non-zero linear operator A is situated in the domain hatched in Fig. 2.1. In the sequel, we usually study only those parts of L-characteristics, which 18 § 2. Continuous linear operators 1 Figure 2.1 lie in the strip 0 < « < 1, 0 S B < oo. Furthermore in many important cases only those parts of L-characteristics situated in the unit square are of interest. In the study of a linear operator A acting from one Banach space E, to another Banach space E, the adjoint operator A*, which acts from E} to E%, plays an important role. This operator is defined by the relation [A*fI(x) = f(x) (xe Ey, fe ED). (2.4) In case of a linear operator A acting from L, to Ls, 0 < a, B <1, the operator A* acts from L;_, to L;_, and the relation (2.4) has the form (Ax, y) = (x A*y) (WE Ly ye Ly-p). (2.5) If 0 <« <1, B = 0, then the adjoint operator A* will act from (L)* to L,-,. Since L, is a subspace of (Lo)*, A* can be treated as an operator from L, to Ly_,; this operator will also be denoted by A*. Finally, we recall that WAlle,+e2 = 1A*lle er (2.6) This relation holds even in the case when A* is an operator from L, to L,_,. 19 1 Linear operators in L, spaces 2.2 Regular operators A direct verification of the continuity of a linear operator (i.e. the establish- ment of inequalities of the type (2.3)) often requires that one overcomes considerable difficulties. In this connection we are interested in obtaining sufficient conditions for continuity which can be expressed in terms of simpler properties. An operator A acting from a function space E, to a space E), is called positive, if it transforms non-negative functions to non-negative functions. It is easy to establish for positive linear operators the inequality: |Ax| S Ajx. (2.7) THEOREM 2.1: Let A be a positive linear operator from L, to Ls. Then A is continuous. Proor: Let us show first that for some constant M the following inequality is fulfilled: Axl S MIxl, (Ee Ly, x 2 0). (2.8) In the contrary case there exists a sequence x,€L,, Xue S 1, x, 2 0, for which: WAxyllp 20-2". (2.9) Put 21 Uo = Lake (2.10) wi 2" It is easy to see that uy eL,. For « < 1, this follows from the absolute convergence of the series (2.10), while for « > 1 it is proved directly: « t/a {Lz ¥ io dts z 2 “fs (O|"dt < z QM" < «. a 20 § 2 > Continuous linear operators On the other hand, uo x,/2, hence it follows that 1 Au = ¥ Ax, and consequently 1 Aull = Hence by (2.9) |Augllp 2 0 and we arrive at a contradiction. Now let x(t) be an arbitrary function in L,. By (2.7) and (2.8) Ax@llp S ACxDIlp SM Il. The theorem has been proved !. Theorem 2.1 means that for positive operators the sets L(4; def.) and L(A; cont.) coincide. A linear operator A is called regular, if it can be written in the form A=A,— Az, where A, and A, are positive operators. Regular operators acting from L, to Ly are continuous by Theorem 2.1. In the subsequent chapters we mainly deal with regular operators. 1 Theorem 2.1 (for 0 S a, 6 S 1) becomes a special case of a more general assertion proved in the paper of I. A. Bakhtin, M. A. Krasnoselskii and V. Y. Stecenko [1]. In this paper it is proved that a linear operator A acting from a Banach space E, with a repro- ducing cone K, to a Banach space E, with a normal cone Ks, is continuous if AK, C Kz. This assertion becomes, on the one hand, a generalization of Banach’s theorem [1] on continuity of an integral operator and, on the other hand, of Krein’s theorem (see M. G. Krein and M. A. Rutman [1]) on continuity of a positive functional on a conic body. 2 A detailed analysis of regular linear operators is developed in the monograph of L. V. Kantorovic, B. Z. Vulikh and A. G. Pinsker [1]. In this monograph a number of assertions more general than Theorem 2.2 and 2.3 are proved. 21 1 Linear operators in L, spaces For the regularity of an operator A it is necessary and sufficient that there exists a positive linear operator B satisfying the inequality |Ax] S B(x). (2.11) An operator, satisfying the condition (2.11), will be called a positive majorant of the operator A. THEOREM 2.2: A linear operator A, acting from L, to Lg is regular if and only if for each non-negative function u(t) €L, it is possible to take a non- negative function v(t) €L, such that Ax) S(t) OS x(t) Sud). (2.12) Proor: Assume that the operator A is regular. Then it is possible to take a positive majorant B of the operator A. For each u(t) € L,, u(t) 2 0, put v(t) = Bu(z). Then (2.11) implies that, for x(t) as in (2.12), Ax(t) S$ |Ax(1)| S B(x) S Bult) = vo). Thus necessity has been proved. Let us prove sufficiency. By (2.12) the set of functions {Ax(t); 0 S x(t) S u(o} is bounded above by some element v(t) in Ly. Hence there exists sup Ax. Define the operator A, for non-negative u(t)eL, by the osxsu relation: Ayu(t)= sup Ax(t). (2.13) Osx(t)Su(t) It is clear that A, transforms non-negative functions u(t) to non-negative functions and satisfies A,u(t) = Au(t) (u(t) = 0). Let us show that A, is an additive operator on non-negative functions in L,. In fact, if x = xy + 2,0 Sy, S$ x1,0 5 y2 S x2, then A,x = sup Ay= sup A(y, + y2) Osysx OS Sx = sup (Ay; + Ay2)= sup Ay; + sup Ay, = Ayx, + Ayx2. Osusx OsySx1 OSy2Sx2 § 2 Continuous linear operators On the other hand, if 0 < y S x, + x2, there exist elements y, and y.' for which 0 < y, S x; (i = 1,2), 1 + 2 = y. Hence Ay = Ay, + Ay2 implies that Ay S sup Ay, + sup Ay, oSysm 0Sy25x2 and consequently Aix S Ayxy + Ayxp Additivity of the operator A on non-negative functions has been proved. By (2.13) the operator A, is also homogeneous: A,(Au(t)) = AAu(t) «(A= 0). Extend the operator A, by the relation Ax = Ayx4 — Ayx_ where x4 =sup{x, 0}, x_ = —inf{x, 0}, to a linear? operator, defined on the whole of L,. By Theorem 2.1 A, is continuous. (2.13) implies that the operator 4, = A, — A is also positive and continuous. Since A = A; — Ag, A is regular. The theorem has been proved. Observe that the operator B = A, + A, becomes a positive majorant of the operator A. Furthermore it can be shown that B becomes the ‘least’ majorant of the operator A. This means that for any other positive majorant B, of the operator A the following relation is fulfilled: Bx) S Bx) (x() 20). 1 It suffices, for instance, to put y; = inf {x,y}, 2 =y¥— Y- ? Linearity follows upon noting that, by the additivity on non-negative functions, Ax’ — Aix" = A,x+ — Aix- whenever x’ — x” = x, x’, x” 2 0. (Ed.) 23 1 Linear operators in L, spaces Thus, it has also been proved that each regular operator has a least majorant. Let A be a regular operator acting from L, toL,(0<«<1,0SfS1), ie. A=A4,— A), where A, and A, are positive linear operators. Then the adjoint operator A* acting from L,_, to L,_, will also be regular because At = At AS, and the adjoint operator of a positive operator is positive. THEOREM 2.3: Each continuous linear operator acting from L, to Lo or from L, to L, 0 SB < 1), is regular. Proor: Consider first an operator A acting from L, to Lo. Since A is continuous, for any u(t) = 0 the set of functions {Ax(t);0 S x(t) S u(t)} is bounded in norm: ||Ax(1)l|o S ||Alla+o lulz. Consequently Ax) Sot) OS x) Sul), where v(t) = ||Allq+ollulle¢Lo. Regularity of the operator follows now from Theorem 2.2. Let A be a continuous operator, acting from L, to L, (0 S B < 1). Then the adjoint operator A* acting from L,_, to Lo, is regular, and consequently the operator A, which is adjoint to the operator A*, is also regular. The theorem has been proved. 2.3. The M. Riesz interpolation theorem In this section we study operators acting from L,(Q) to L,(Q*), where « and f are arbitrary non-negative numbers. The remarkable theorem that L-characteristics of linear operators A acting from L, to Ly are convex sets is due to M. Riesz. More precisely, the L-characteristics L(A; cont.) contain, together with any two points! * M. Riesz [1] considered only values ao, Bo, 1 and Bx, in the interval [0, 1]. The general theorem was obtained by A. Calderon and A. Zygmund [1, 2]. 24 § 2 Continuous linear operators MMM L(A; Cont.) Figure 2.2 {0, Bo} and {a,, B,}, the whole segment joining them (see Fig. 2.2). This theorem was established by M. Riesz in the form of the so-called inter- polation theorem. THEOREM 2.4: Let A be a linear operator, which is continuous simultaneously From Ly, to Ly, and from L,, to L,,: Axllp6 S Mo Ii*llao» (2.14) Axle, S My Mlle (2.15) Then for any t € (0, 1) the operator A acts continuously from Ly, to Ly where a(t) = (1 — t)%q + to, Bt) = (1 — t)Bo + tB1, (2.16) moreover in the case of complex spaces L,, Axl a) S Mo" Millllacey (2.17) while in case of real spaces L,, [Axl S CBos Bi, MoM il llacey (2.18) where the constant C(Bo, B;,t) does not depend on the operator A. 25 1 Linear operators in L, spaces Proor: We confine ourselves to giving a detailed proof for indices a, Bo, a, B, in the interval [0, 1]. Consider first the case of complex spaces L,, Lg. Let % > 4. Let us show that the inequality (2.17) is valid for all simple functions X(S) = €4Ky(S) +... + Calm (S)> where K,(5s),...,Km(S) are characteristic functions of disjoint sets Q; < Q @=1,2,...,m). The inequality (2.17) with a fixed function x(s) € L,,) is equivalent to the inequality \(4x, 9) SM5“Mi lela (2.19) Ilia where the supremum is taken over all y(t) in some set dense in Lj _p,,). Hence to prove (2.17), it suffices to establish the inequality (Ax, DLS MoM xIhace Pla ce for every simple function y(t): WO) = AAD) + + date(s where 7,(1),..., z,(t) are characteristic functions of disjoint sets QF < Q* G=1,...,n). Consider the function ®(z) of a complex variable z = u + iv which is defined by the relation (2) = (Alix? x], [PO *y) == ceded" "4 POL Ave) xa a4] (2.20) where a(2) = oe Ble) = Yo — 28x 2.21 Bee) =) It is clear that ®(z) is analytic in the strip 0 < Rez <1. 26 § 2 > Continuous linear operators By Hadamard’s three lines theorem’ the following inequality is valid for each t € (0, 1): |6(2)| S sup|(iv)|*~*-sup|G(1 + i)|. (2.22) Let us estimate the expressions #(iv) and &(1 + iv). For Rez = 0 we have the relation _% |. _1-Bo |, bo- Bs wm yt HO = as tT whence |®(iv)| S aolatt))~ 1+ io(as—aa)/e See nant e-t eta Seenseeai |(Af]x] S0lMD~ 1 Hees aod/ate) | y](A~BOd/CABEAD)~ 1 +40(BO—BINCA-BLOD. yy} S [Al] x|%0/20~ 1420600. yy 1 Bo)/(1— (2) ~ 1+ iv(Bo~ B1)/(1~ BCE} I fp] #00 800) o(Bo- Bs)/(1~ BC) x “Yili-po IA AL] x] 40/2~ 1 #12006). 9H |] yA POMA-PEOD By (2.14) it follows from this inequality that [PCiv)] S Mo || |x| GOl*P~ 4900/00 « | | y| APOE POT yg, = Mol xl" fag* Ly POM EPO | or [BGS] S Mo lees I vo-P. (2.23) * For any function f(z), analytic in the strip 0 < Re(z) < 1 and continuous in the strip 0 S Re(z) S 1, the following inequality is valid: IF@)I_-S sup |f@)I** sup [fC + iyi a,. Hence the sequence Ax,(s) converges to the element Ax in L,,-norm and it follows that y = Ax. Proceeding to the limit n > oo in the inequality VAXullae) S MoM Mallacon we obtain the inequality (2.17) for the function x(s) € Li). Now let a = a. Then the inequality (2.17) follows immediately from (1.7): WAZ pc) S WAZ p “WAR, S Mo Mi lll The theorem for the case of complex spaces L, has been proved. Now let the spaces L, be real and let A be a given operator. Denote by A the extension of the operator A to complex valued functions z(t) = x(t) + iy(t) which is defined by the relation A(x + iy) = Ax + iAy. The inequalities IAG + Vllgo S VAXIp9 + AY Ilp0 S Moll loo + lylleo) S 2Mollx + Wllao 28 § 2. Continuous linear operators imply that HAZIlp, S$ 2Mo llZllao- Similarly WAzIlp, $ 2M; llzlla,- Applying to the operator A the part of the theorem, already proved for the complex cases, we deduce that the operator A acts from Lj) to Lg.) and that NAZI pe) S 2Mo-M i lellacey whence follows the inequality VAX S 2Mo "Mi lla (2.25) The theorem has been proved for indices %9, Bo, %, and , in the interval 0, 1]. This proof is borrowed from the paper of A. Calderon and A. Zygmund. Their proof, as we have already noted, embraces also the case of indices greater than 1. The basic change made in the proof is that then instead of the analytic function (2.20) the following subharmonic function is considered: H@) = (AllXO" tT, LyPO-*y), where _ (1 = 2)tto + zat fe age _ 1-1 = 2k Bo — 2kB, BE) = THe and k satisfies the inequality Bok <1, Bik <1. 29 1 Linear operators in L, spaces Here instead of Hadamard’s three lines theorem its generalization to sub- harmonic functions is used. As was noted earlier, the assertion of Theorem 2.3 means that the L- characteristic L(A; cont.) is a convex set. Hence it follows that the function &(«; A; cont.) is convex and, in particular, continuous in a. In concluding this section we note that the inequality (2.17) is in general invalid in the case of real spaces. But even for real spaces the inequality (2.17) is fulfilled for special pairs of indices «,, B; (the inequality (2.17) is trivially fulfilled if 7) = a, ; non-trivial assertions are presented in the sequel in 10.4°). In the next section it will be shown that for special classes of operators the inequality (2.17) is always valid in the case of real spaces. 2.4 Interpolation theorems for regular operators In this section interpolation theorems for positive and regular operators are proved. THEOREM 2.5:! Let A be a positive linear operator acting simultaneously from a space L,, to a space Ly, and from a space L,, to a space Ly,: Axlp6 S Moll*llagy Alla, S Milita» where By and B, can be any non-negative numbers. Then for any t € (0, 1) the operator A acts from Ly.) to Ly), where a(t) = (1 = t)o + tay, P(t) = (1 — t)Bo + tA, and Axo S$ MO~"M3 I xllacey (2.26) This shows that for positive operators the inequality (2.17) is valid for real spaces as well as for complex spaces. 1 A matrix analogue of Theorem 2.5 is indicated in the book of G. Hardy, D. Littlewood and G. Polya [1]. In general form the theorem was proved in the paper of P. P. Zabreiko and E. I. Pustylnik [2]. A related assertion was previously obtained by E. I. Pustylnik [4]. 30 § 2 Continuous linear operators ProorF: It suffices to prove the theorem for the case % > a,. As in the proof of Theorem 2.4, it suffices to establish the inequality (2.26) for simple functions X(S) = C4 y(S) +... + Cm (S) (2.27) where the x;,(s) are characteristic functions of disjoint sets Q;< Q @=1,2,...,m). Let us first prove the inequality: 1Ax(] S CAC?) LAC 470] (2.28) By (2.27) this inequality can be rewritten in the form | Ledee(o| s [zee any “TE lel Ax(oy. (2.29) i= = =1 Let t =f be such that all the numbers (Ax;)(to) are finite. Put 4; = (i = 1,2,..., m). It is clear that all the 4; are non-negative. Hence Hélder’s inequality is valid: [Dawid SCL hal/O-P AY Cooly O <1 <1). tet i=1 i=1 Setting u, = |e[~ 9" -sen ¢,, v; = lel", we obtain the inequality [Ded S$ (Lleol Ay -(X fed" A)" i=1 11 i=1 which, in view of linearity of the operator A, is equivalent to the inequality |Ax(a)] S CAC x2 V)] [A> for t = to. Since for almost all t € Q* the functions Ax;(t) are finite, the inequality (2.28) is valid for almost all te Q*. Hence VAX pce S WAC 1 FAC )F ll pcos 31 1 Linear operators in L, spaces whence by Hdlder’s inequality (1.5) W Axl pce) SACP) 55 AC )I5,- (2.30) From (2.14), (2.15) and (2.30) it follows immediately that W Az pce) S MoU el 25 MG I balls, S Mo"M3 Illacey The inequality (2.26) has been proved for all step functions x(t). Hence it follows (see the proof of Theorem 2.4) that it is valid for all functions x(t) € L,.,). The theorem has been proved. Theorem 2.5 and the definition of regular operators imply immediately: THEOREM 2.6: Let an operator A admit a positive majorant B: |Ax| S$ B(|x\). (2.31) Let the positive operator B act simultaneously from L,, to Ly, and from L,, to Ly,: IBxll6o S Noll*llaor Bla, S Ni lille Then for any + € (0, 1) the operator A acts continuously from Ly. to Ley where a(t) = (1 = 1) + 104, A(t) = (1 — t)Bo + thr, and WAxllpcey S NONG Ilacoy Let A be a regular linear operator, acting from L,, to L,, and from L,, to Lg, % 2 %. Denote by B the ‘least’ positive majorant of A as an operator from L,, to Ly, (see 2.2°). By (2.13) the operator B will also be the ‘least’ positive majorant of A as an operator from L,, to Lg,. Then by Theorem 2.5 it follows that B is a positive operator acting from L,,, to Ly) 32 § 2 Continuous linear operators (0 , by, by satisfy the inequalities 0Sa, a,. In fact, when x(s)eL,,2< 1-5; 1 1 lAx(p)| = fx, s)x(s)ds| $ refs" x()lds 1 S 1% fs“ dsyt “AY x. 6 Consequently 1 |Axllp S fi Ill. (2.35) — a,/BY [1 — b/( — o]'~* Applying Theorem 2.4 to the operator A, we conclude that it acts from L, to L, if the numbers «, f satisfy the inequalities MT y+ a, — (1 — by) a, —- 1—}b:, B> ast by Boo, B= “| B>a,. (2.36) by Denote by I’ the boundary of the set of points, whose coordinates satisfy these inequalities. The inequality (2.35) and the Riesz interpolation theorem imply that Axllp SM [lle (2.37) where M is some constant, depending, for given a, B, only on the distance of the point {«, 8} from I’. Thus the set L(A; cont.) contains all points {a, B}, for which three inequalities of (2.36) are fulfilled. It turns out that the set of interior points of the L-characteristic of the operator A coincides with the set of points {a, 8} for which the inequalities (2.36) are fulfilled (Fig. 2.3). For the proof it suffices to show that any point {a, B}, for which one of the inequalities 34 § 2 Continuous linear operators a>1—b, aa ) 224% < -(d- + B abet ¢ b, by B v. Consider the function 1 s*ds Axo(t) = |aee (2.38) é To estimate Axo(z), perform the change of variables 5 = 1(279/(1~%) y in (2.38): po (@2-ai)/(b1~b2) yen Axg(t) = 1 lt O#br~ 1NMea—anyib.= be l+u du. (2.39) bin be 0 For v > 1 — b, the integral in the right side of (2.39) diverges and the operator A does not act from the space L,, « > 1 — b2, to any Lg. In other Figure 2.3 Figure 2.4 35 1 Linear operators in L, spaces words, the set L(A;cont.) does not contain any point {a, 8} with a>1—b). Let 1 — b) > v> 1 — dy. (2.39) implies that AXo(t) Z cot Wart Ort 9— N@2—as)/ (bs ba) where 1 yoke on J Tern o Hence Ax,(t) is not summable with power 1/8, if a,— by - Bsa, + = ( +¥—1. (2.40) 2 This means that the operator A does not act from L,, 1 — by < a < 1 — by, to any space L, if f satisfies the inequality (2.40). In other words the set L(A; cont.) does not contain any point {«, 6} for which a,—a, a — a ata,—(1—5. b, by a, — ( ) bbs ,l-b ot 6 (t St) <1). Consequently the function Ax(t) is not summable with any power r, for which Ir Say. In other words, the set L(A; cont.) does not contain any point {a, 8} for which B S$ a. 36 § 2 Continuous linear operators Consider now an arbitrary open set L, satisfying the conditions of the theorem; denote by (a) the convex continuous function &(a) = inf Bp. {a,pjeL Denote by I) the support line to the set L at a point {a, €(a)} and by L® the set of points in the strip 0 < a < 1, B = 0, lying above Ij» (Fig. 2.4). Take a countable set of numbers «,, «2, ..., dense in the interval [0, 1]. It is clear that 2 L=(\Lo. net Let A, be an operator of the form (2.32) with a kernel K,(t, s) of the form (2.33), the interior part of whose L-characteristic coincides with the set L°”., Consider the positive operator B, defined by the relation 1 Mes B= i Ai and let us show that it satisfies the conditions of the theorem. Let {a, B} eZ. Since the distances from this point to the boundaries Tq of the sets L®” are bounded from below, inequality (2.37) implies that the following inequalities are fulfilled: lAxlp SMIxla @=1,2,...). Hence for any xe L, the series atl = Ae i et converges in L,-norm, so Bx eL,. Thus, {x, B} ¢ L(B; cont.). Now let a point {«, B} not belong to the closure of L. Then obviously there exists a number ip such that {«, B} ¢ Lo’. This means that 4,, does not act from L, to Ly. The inequalities Bx= ze Aix, 37 1 Linear operators in L, spaces valid for positive functions in L,, imply that the operator B does not act from L, to Lg, either, i.e. {«, 8} ¢ L(B; cont.). The theorem has been proved. 2.6 Ona property of regular operators THEOREM 2.8:' Let A be a regular linear operator acting from L, to Ly (0 Sa < , 0 < B < ©). Then each set of functions M with equi-absolutely continuous norms in L, is transformed by the operator A to a set of functions with equi-absolutely continuous norms in Ly. ProorF: The assertion of the theorem is obvious in the case of « = 0. Let a > 0. It suffices to consider positive operators A. It can also be assumed that the functions x(s)¢ Mt are non-negative. From the hypothesis of the theorem it follows that Ixia Sa Ox(s)e M). Let ¢ > 0 be given. Equi-absolute continuity of the norms of the functions x(s)eM in L, implies the existence of a 59 such that Pox(s)lle < (we M) & PF Alla+g whenever mes(D) < do. Put hy = a/55- Write each function x(s)¢ 2 in the form X(s) = x4(8) + x2(s) where _ f(s), if [x(s)| S Ao *1(9) = i if |x(s)| > ho. + This theorem was proved by P. P. Zabreiko and E. I. Pustylnik (P. P. Zabreiko [2]). 38 § 2 Continuous linear operators It is clear that |x,(s)| S Ao and that the measure of the set D(e; x) of points s for which x,(s) # 0, is less than 69. Hence it follows that € Ia )la = Poe, ODla S Fale From this inequality it follows that for an arbitrary set D* < Q* PpAxllp S 2’ho||Pp-Auollp + e/2 (x(s)e M), where uo(s) = 1. Hence there exists a 5 > 0 such that IPrAxla 0 Sle(lrae = fatleto's Aldh = facet; AY”) ah d é ; = fae: A)dh! =r a 1X(z; h)dh. ° 6 * J, Marcinkiewicz [1]. 39 1 Linear operators in L, spaces Hence for B > 0 . B llzllp = iF AO PIB YC; nar} (et) € Ly). (2.43) é The function MA =A VE; WR=[ f AvParye (2.44) 2@12h) is obviously bounded if z(r) € Ls. The converse does not hold; for instance the function z(t) = t~* (0 S$ t S 1) does not belong to Ls, but the function (2.44) is bounded. Boundedness of the function (2.44) implies only that the function z(t) belongs to all spaces L,,, for ¢ > 0. Denote by M, (f > 0) the totality of all functions z(t), for which the function (2.44) is bounded, and put IzOlli, = sup AAG; AY. O

    You might also like