0010:math0010010/CRS

1: \documentclass[12pt,a4paper]{book}

2: \overfullrule=0pt

3: \hfuzz=3pt

4: \tolerance=1000

5: \input amssym

6: \catcode`\@=11

7:

8: \font\decp=cmr6

9: \font\titl=cmr17 at 20pt

10: \font\capit=cmbx12 at 24pt

11: \font\nom=cmr17

12: \font\dedi=cmsl12 at24pt

13:

14: \font\tenmsa=msam10 \font\sevenmsa=msam7 \font\fivemsa=msam5

15: \font\tenmsb=msbm10 at 12pt\font\sevenmsb=msbm7 at 9pt\font\fivemsb=msbm5 at 7pt

16: \newfam\msafam \newfam\msbfam

17: \textfont\msafam=\tenmsa  \scriptfont\msafam=\sevenmsa \scriptscriptfont\msafam=

18: \fivemsa

19: \textfont\msbfam=\tenmsb  \scriptfont\msbfam=\sevenmsb \scriptscriptfont\msbfam=

20: \fivemsb

21:

22: \font\teneufm=eufm10 \font\seveneufm=eufm7 \font\fiveeufm=eufm5

23: \newfam\eufmfam

24: \textfont\eufmfam=\teneufm \scriptfont\eufmfam=\seveneufm\scriptscriptfont

25: \eufmfam=\fiveeufm

26: \def\frak#1{{\fam\eufmfam\relax#1}}

27:

28: \def\hexnumber@#1{\ifcase#1 0\or1\or2\or3\or4\or5\or6\or7\or8\or9\or

29:  A\or B\or C\or D\or E\or F\fi }

30:

31: \edef\msa@{\hexnumber@\msafam} \edef\msb@{\hexnumber@\msbfam}

32:

33: \mathchardef\qed="0\msa@03

34:

35: \def\Bbb#1{{\fam\msbfam\relax#1}}

36:

37:

38:

39: \def\R{{\Bbb R}} \def\C{{\Bbb C}} \def\Z{{\Bbb Z}} \def\N{{\Bbb N}} \def\Q{{\Bbb Q}}

40: \def\K{{\Bbb K}}

41: \catcode`\@=12

42:

43: \newtheorem{teo}{Theorem}[section]

44: \newtheorem{pro}[teo]{Proposition}

45: \newtheorem{cor}[teo]{Corollary}

46: \newtheorem{eje}[teo]{Example}

47: \newtheorem{lem}[teo]{Lemma}

48: \newtheorem{obs}[teo]{Remark}

49: \newtheorem{defin}[teo]{Definition}

50:

51:

52: \newcommand{\bteo}{\begin{teo}}

53: \newcommand{\eteo}{\end{teo}}

54: \newcommand{\bpro}{\begin{pro}}

55: \newcommand{\epro}{\end{pro}}

56: \newcommand{\bcor}{\begin{cor}}

57: \newcommand{\ecor}{\end{cor}}

58: \newcommand{\blem}{\begin{lem}}

59: \newcommand{\elem}{\end{lem}}

60: %\newcommand{\bdefi}[1]{\begin{defin}{\rm{#1}}}

61: \newcommand{\bdefi}{\begin{defin}\rm}

62: \newcommand{\edefi}{\end{defin}}

63: %\newcommand{\beje}[1]{\begin{eje}{\rm{#1}}}

64: \newcommand{\beje}{\begin{eje}\rm}

65: \newcommand{\eeje}{\end{eje}}

66: %\newcommand{\bobs}[1]{\begin{obs}{\rm{#1}}}

67: \newcommand{\bobs}{\begin{obs}\rm}

68: \newcommand{\eobs}{\end{obs}}

69: \newcommand{\bequ}{\begin{equation}}

70: \newcommand{\eequ}{\end{equation}}

71:

72: \def\bdem{{\sl Proof. }}

73: \def\edem{\par\bigskip}

74:

75: \def\S{{\cal S}}

76: \def\SS{{\cal S}_0}

77: \def\LL{L^\infty_0}

78: \def\M{{\cal M}}

79: \def\NN{\hbox{N}}

80:

81: \def\loX{{\Lambda^p_X(w)}}

82: \def\loinX{{\Lambda^{p,\infty}_X(w)}}

83: \def\lox#1#2#3{{\Lambda^{#1}_{#2}(#3)}}

84: \def\loinx#1#2#3{{\Lambda^{#1,\infty}_{#2}(#3)}}

85: \def\llox#1#2{{\Lambda^{#1}(#2)}}

86: \def\lloinx#1#2{{\Lambda^{#1,\infty}(#2)}}

87: \def\lo{{\Lambda^p_u(w)}}

88: \def\loz{{\Lambda^{p_0}_{u_0}(w_0)}}

89: \def\lou{{\Lambda^{p_1}_{u_1}(w_1)}}

90: \def\loin{{\Lambda^{p,\infty}_u(w)}}

91: \def\loinz{{\Lambda^{p_0,\infty}_{u_0}(w_0)}}

92: \def\loinu{{\Lambda^{p_1,\infty}_{u_1}(w_1)}}

93: \def\llo{{\Lambda^p(w)}}

94: \def\lloz{{\Lambda^{p_0}(w_0)}}

95: \def\llou{{\Lambda^{p_1}(w_1)}}

96: \def\lloin{{\Lambda^{p,\infty}(w)}}

97: \def\lloinz{{\Lambda^{p_0,\infty}(w_0)}}

98: \def\lloinu{{\Lambda^{p_1,\infty}(w_1)}}

99: \def\lpw{{L^p(w)}}

100: \def\lpwz{{L^{p_0}(w_0)}}

101: \def\lpwzdec{{L^{p_0}_{\hbox{\decp dec}}(w_0)}}

102: \def\lpwzinc{{L^{p_0}_{\hbox{\decp inc}}(w_0)}}

103: \def\lpwdec{{L^{p}_{\hbox{\decp dec}}(w)}}

104: \def\lpinc{{L^{p}_{\hbox{\decp inc}}}}

105: \def\lpdec{{L^{p}_{\hbox{\decp dec}}}}

106: \def\lpwu{{L^{p_1}(w_1)}}

107: \def\lpwinu{{L^{p_1,\infty}(w_1)}}

108: \def\lpu{{L^p(u)}}

109: \def\lpuz{{L^{p_0}(u_0)}}

110: \def\lpuu{{L^{p_1}(u_1)}}

111: \def\lpuinu{{L^{p_1,\infty}(u_1)}}

112: \def\BBx#1#2{B_{#1}(#2)}

113: \def\BBinx#1#2{B_{#1,\infty}(#2)}

114: \def\BBzu{B_{p_0,p_1}(u_0,u_1)}

115: \def\BBinzu{B_{p_0,p_1,\infty}(u_0,u_1)}

116: \def\BBnux#1{B_{#1}}

117: \def\BBinnux#1{B_{#1,\infty}}

118: \def\Bx#1#2{B_{#1}(#2)}

119: \def\Binx#1#2{B_{#1,\infty}(#2)}

120: \def\Bpu{B_p(u)}

121: \def\Binpu{B_{p,\infty}(u)}

122: \def\Bp{B_p}

123: \def\Binp{B_{p,\infty}}

124: \def\fiq{\phi_Q}

125: \def\Fiu{\Phi_u}

126: \def\limto#1{\smash{\mathop{\longrightarrow}\limits_{#1}}}

127: \def\Fd{\M_{\hbox{\decp dec}}(\R^+)}

128: \def\Fi{\M_{\hbox{\decp inc}}(\R^+)}

129: \def\d{\downarrow}

130: \def\u{\uparrow}

131: \def\dip{\hbox{\decp d}}

132: \def\di{\hbox{\rm d}}

133: \def\Rm{\R^+}

134: \def\Rn{\R^n}

135: %\def\lesssim{\buildrel<\over\sim}

136: \def\lesssim{\mathrel{\mathop<\limits_{\sim}}}

137:

138:

139:

140: \makeindex

141:

142: \begin{document}

143:

144: \title{Recent Developments in the Theory of Lorentz Spaces and

145: Weighted Inequalities}

146: \author{Mar\'{\i}a J. Carro \and Jos\'e A. Raposo \and Javier Soria}

147: \maketitle

148: \newpage

149: \pagenumbering{roman}\pagestyle{empty}

150:

151: \centerline{\titl Recent Developments in the Theory of Lorentz}

152:

153: \medskip

154:

155: \centerline{\titl Spaces and Weighted Inequalities}

156:

157: \vskip 3cm

158:

159: \centerline{\nom Mar\'{\i}a J. Carro \hfill Jos\'e A. Raposo \hfill Javier Soria}

160:

161: \vskip 2cm

162:

163: \centerline{ADDRESS}

164:

165: \medskip

166:

167: \centerline{Dept. Appl. Math. and Analysis}

168: \centerline{Gran Via 585}

169: \centerline{University of Barcelona}

170: \centerline{E-08071 Barcelona, SPAIN}

171:

172: \vskip 1cm

173:

174: \centerline{E-MAIL: {\tt carro@mat.ub.es\hskip 1cm soria@mat.ub.es}}

175:

176: \vfill

177: \noindent

178: {\it MSC2000}: 42B25 (26D10 26D15 46E30 47B38 47G10)

179:

180: \medskip\noindent

181: {\it Keywords}: Hardy operator, Hardy--Littlewood maximal function, Interpolation,  Lorentz spaces, Weighted

182: inequalities.

183:

184: \medskip\noindent

185: This work has been partly supported by DGICYT PB97--0986 and CIRIT 1999SGR00061.

186: \newpage

187: \

188: \newpage

189: \ \vfill

190: \rightline{\dedi Per a la Beatriu}

191: \vfill

192: \newpage\pagestyle{empty}

193: \

194: \newpage

195: \pagestyle{plain}

196: \tableofcontents

197: \newpage\pagestyle{empty}

198: \

199: \newpage

200: \pagestyle{plain}

201: \addcontentsline{toc}{chapter}{Foreword}

202: \ \vskip 2cm

203: \noindent{\capit Foreword}

204: \vskip2cm

205: \noindent

206: On September 4, 1998, Jos\'e Antonio Raposo, the second named author, died when he was only 39. He had

207: just received his Ph.D. degree a few months earlier, under our supervision. He was really happy for all the

208: new projects he had for the future, and so were we, since he was an extraordinary mathematician, and a very

209: valuable friend.

210:

211: This work is an updated version of his thesis, written as a self-contained text, with most of

212: the motivations, examples and applications available in the literature.

213:

214: We want to thank all the people who have encouraged us to write this book, and specially Jos\'e Antonio's

215: family. We also thank  Joan Cerd\`a who has read the whole manuscript and has given us many good advises,

216: improving the final version of these notes.

217:

218: \medskip

219: \rightline{Mar\'{\i}a J. Carro and Javier Soria}

220: \rightline{Barcelona, September 2000}

221:

222: \newpage\pagestyle{empty}

223: \

224: \newpage

225:

226: \pagenumbering{arabic}

227: \addcontentsline{toc}{chapter}{Introduction}

228: \ \vskip 2cm

229: \noindent{\capit Introduction}

230: \vskip2cm

231: \pagestyle{plain}

232:

233: \noindent

234: The main objective of this work is to bring together two well known and, a priori, unrelated theories dealing

235: with weighted inequalities for the Hardy-Littlewood maximal operator

236: $$

237: Mf(x)\index{$M$}=\sup_{x\in Q}{1\over|Q|}\int_Q |f(y)|\,dy,\quad x\in\Rn.

238: $$

239: For this, we consider the boundedness of $M$ in the weighted Lorentz space $\lo$, introduced by G.G. Lorentz

240: (\cite{Lor}) in 1951,

241: \bequ\label{intro}

242: M:\lo\longrightarrow\lo.

243: \eequ

244: Two examples are historically relevant as a motivation:

245: If $w=1$,  (\ref{intro}) corresponds to the study of the boundedness

246: $$

247: M:L^p(u)\longrightarrow L^p(u),

248: $$

249: which was characterized by B. Muckenhoupt (\cite{Muck1}) in 1972, and the solution is given by the so called

250: $A_p$ weights. The second case is when we take in (\ref{intro}) $u=1$. This is a more recent theory, and

251: was completely solved by  M.A. Ari\~no and  B. Muckenhoupt (see \cite{AM2}) in 1991. It turns out that the

252: boundedness

253: $$

254: M:\llo\longrightarrow\llo,

255: $$

256: can be seen to be equivalent to the boundedness of the Hardy operator $A$ restricted to decreasing functions

257: of

258: $L^p(w)$, since the nonincreasing rearrangement of $Mf$ is pointwise equivalent to $Af^*$. The class of

259: weights satisfying this boundedness is known as  $B_p$. Another results related to this problem can be

260: found in \cite{Saw2}, \cite{Ne1}, \cite{CS1}, \cite{CS2}, \cite{Sp1}, \cite{Sp2}, and \cite{HM}.

261:

262:

263:

264: Even though the $A_p$ and $B_p$ classes enjoy some similar features, they come from very different

265: theories, and so are the techniques used on each case: Calder\'on--Zygmund decompositions and covering

266: lemmas for $A_p$,  rearrangement invariant properties and positive integral operators for $B_p$.

267:

268: \medskip

269:

270: It is our aim to give a unified version of these two theories. Contrary to what one could expect, the

271: solution is not given in terms of the limiting cases above considered (i.e., $u=1$ and $w=1$), but in a

272: rather more complicated condition, which reflects the difficulty  of estimating the distribution function of

273: the Hardy-Littlewood maximal operator with respect to general measures (some previous results on this

274: direction can be found in \cite{CHK} and \cite{HK}, for the particular case of the $L^{p,q}(u)$ Lorentz

275: spaces with a power weight $w$).

276:

277: \pagestyle{myheadings}

278: \markboth{INTRODUCTION}{M.J.\ Carro, J.A.\ Raposo, and J.\ Soria}

279:

280: \medskip

281:

282: In order to carry out this program as a self-contained monograph, we study in the first two chapters the

283: main results needed for the rest of the book. Hence, in Chapter~\ref{ch:characteristic} we consider the

284: boundedness of integral operators on monotone functions. The most important  result is Theorem~\ref{te:

285: I.2.13}, where we prove that under very general conditions, the boundedness is determined by the action of

286: the operator on some characteristic functions (see also \cite{BPS} and \cite{CS2}). We then consider the main

287: example in this context, namely the Hardy operator, and introduce the $B_p$ class.

288:

289: \medskip

290:

291: In Chapter~\ref{ch:lor} we make an exhaustive study of the functional properties of the Lorentz spaces over

292: general measure spaces, and arbitrary weights. In particular we consider the discrete case (sequence Lorentz

293: spaces) where we answer several open questions about normability and  duality (see Theorem~\ref{te:

294: II.4.18}). Some previous results were already proved (in the case $w$ decreasing) in \cite{Al} and

295: \cite{AEP}.

296:

297: \medskip

298:

299: Finally, in  Chapter~\ref{ch:maximal} we consider the solution to (\ref{intro}), which is proved in

300: Theorem~\ref{te:principal} (some sufficient conditions had been already obtained in \cite{CS3} and

301: \cite{Ne2}). We see that the characterization that we obtain is based upon the so called

302: $p\,-\,\epsilon$ condition, which is natural since both $A_p$ and $B_p$ enjoy this property. However is worth

303: noticing that  there exist non-doubling weights $u$ for which $M$ satisfies (\ref{intro}) for suitable

304: weights $w$ (see Theorem~\ref{te: pnd}). We also study the weak-type and the restricted weak-type

305: version of (\ref{intro}).

306:

307: \medskip

308:

309: As far as possible, we have always tried to give precise bibliographic information about the results which

310: were previously known, and also about the techniques used in the proofs, although due to the big number of

311: people   working in these theories, this a difficult task. We have also followed the standard notation found

312: in the main reference books (e.g., \cite{BS}, \cite{BL}, \cite{GR}, \cite{Ru2}, \cite{St}, \cite{SW}).

313:

314: \chapter{Boundedness of operators on characteristic functions and the Hardy operator}

315: \label{ch:characteristic}

316: \pagestyle{myheadings}

317: \markboth{CHAPTER 1. BOUNDEDNESS OF OPERATORS}{M.J.\ Carro, J.A.\ Raposo, and J.\ Soria}

318:

319: \section{Introduction}

320: \label{se I: Introduction}

321:

322: This chapter includes general results on boundedness of operators on

323: $L^p\/$ such as the

324: Hardy operator\index{Hardy operator}

325: \bequ

326: \label{hardy}

327: Af(t)\index{$A$}={1\over t}\int_0^t f(s)\,ds,\quad t>0,

328: \eequ

329: and the   Hardy-Littlewood maximal operator\index{Hardy-Littlewood maximal operator}

330: \bequ

331: \label{hardylit}

332: Mf(x)\index{$M$}=\sup_{x\in Q}{1\over|Q|}\int_Q |f(y)|\,dy,\quad x\in\Rn.

333: \eequ

334:

335:

336: The main result is  Theorem~\ref{te: I.2.13} and its Corollaries~\ref{cor: I.2.14}  and

337: \ref{cor: I.2.19} where it is proved that a great variety of operators satisfy that their boundedness

338: is characterized by the restriction to characteristic functions.

339:

340: In the third section, we study the  Hardy operator, recalling some known results that we shall use

341: in the following chapters and including a detailed study of the discrete  Hardy operator\index{Hardy

342: operator!discrete},

343: \bequ

344: \label{dischardy}

345: A_d f(n)\index{$A_d$}={1\over n+1}\sum_{k=0}^n f(k),\qquad n\in\N.

346: \eequ

347:

348: The following standard notations will be used very often: Letters such as $X$ or $Y$ are always

349: $\sigma$-finite measure spaces and ${\cal M}(X)$\index{${\cal M}(X)$} (${\cal M}^+(X)$)\index{${\cal

350: M}^+(X)$} denotes the space of measurable (resp. measurable and nonnegative) functions on

351: $X$.  The distribution function\index{distribution function} of $f$ is

352: $$

353: \lambda_f(t)\index{$\lambda_f$}=\mu(\{x:|f(x)|>t\}),

354: $$

355: the nonincreasing rearrangement\index{nonincreasing rearrangement} is

356: $$

357: f^*(t)\index{$f^*$}=\inf\{s>0:\lambda_f(s)\le t\},

358: $$

359: and

360: $$

361: f^{**}(t)\index{$f^{**}$}={1\over t}\int_0^t f^*(s)\,ds.

362: $$

363: If $f$ is a positive nonincreasing (nondecreasing) function we will write

364: $f\downarrow$ (resp. $f\uparrow$). The space $\lpdec$\index{$\lpdec$} ($\lpinc$)\index{$\lpinc$} is the cone

365: of all nonincreasing (resp. nondecreasing) functions in $L^p$. Two positive quantities $A$ and $B$, are said

366: to be equivalent ($A\approx B$) if there exists a constant $C>1$ (independent of the essential parameters

367: defining

368: $A$ and

369: $B$) such that $C^{-1}A\le B\le CA$. If only $ B\le CA$, we write $B\lesssim A$. The

370: undetermined cases $0\cdot\infty,\,{\infty\over\infty},\,{0\over 0}\/$, will always be taken equal to

371: $0\/$.

372:

373: \section{Boundedness on characteristic functions}

374: \label{se I: characteristic functions}

375:

376: In this section we consider the  problem of characterizing the boundedness of

377: \bequ

378: \label{I.2.1}

379: T\,:\,\big(L^{p_0}(X)\cap L\,,\,\|\cdot\|_{p_0}\big)\longrightarrow L^{p_1}(Y),

380: \eequ

381: where

382: $L\/$ is a subclass of ${\cal M}(X)\/$ and  $T\/$ is an operator (usually linear or sublinear). That

383: is, we are interested in studying the inequality

384: \bequ

385: \label{I.2.2}

386: \|Tf\|_{L^{p_1}(Y)}\le C\|f\|_{L^{p_0}(X)},\quad f\in L.

387: \eequ

388: In some cases inequality  (\ref{I.2.2}) holds for every  $f\in L\/$ if and only if it holds on

389: characteristic functions

390: $f=\chi_A\in L\/$. This happens (see \cite{CS2} and \cite{Sp1}) in the case

391: $L=\lpwzdec$, $X=(\Rm,w_0(t)\,dt)$ and

392: $Y=(\Rm,w_1(t)\,dt)\/$ in the range of indices  $0<p_0\le1,\;p_0\le p_1<\infty\/$ and operators of

393: the type

394: \bequ

395: \label{I.2.3}

396: Tf(r)=\int_0^\infty k(r,t)f(t)\,dt\,,\quad r>0,

397: \eequ

398: for positive kernels $k$. Using this, one can easily obtain a useful characterization of   (\ref{I.2.1}). For

399: example, in the previous case the condition on the weights

400: $w_0,w_1\/$ (nonnegative locally integrable functions

401: in $\R^+$)  for  which $T:\lpwzdec\to\lpwu\/$ (defined by (\ref{I.2.3})) is:

402: $$

403: \bigg(\int_0^\infty\bigg(\int_0^r k(s,t)\,dt\bigg)^{p_1}w_1(s)\,ds\bigg)^{1/p_1}\le

404: C\,\bigg(\int_0^r w_0(s)\,ds\bigg)^{1/p_0}\,,

405: \quad r>0.

406: $$

407: The range $0<p_0\le1,\;p_0\le p_1<\infty\/$ is fundamental and, in fact, the previous principle

408: also works in other situations, as in the case of  increasing functions (see \cite{Sp1})

409: $$

410: T:L^{p_0}_{\hbox{\decp inc}}(w_0)\to\lpwu

411: $$

412: where $T\/$ is as in  (\ref{I.2.3}).

413:

414: As we shall see (Theorem~\ref{te: I.2.13}) all this is a consequence of a general principle that

415: can be applied to a more general class of operators than those of integral type  and for a

416: bigger class of functions that includes the monotone functions.

417:

418: We first need some definitions.

419:

420:

421: \bdefi

422: \label{def: I.2.4}

423: We say that $\emptyset\neq L\subset \M(X)\/$ is a  regular class\index{regular class} in

424: $X\/$ if, for every  $f\in L\/$,

425: \begin{enumerate}

426: \item[(i)] $|\alpha f|\in L\/$, for every  $\alpha\in\R\/$,

427:

428: \item[(ii)] $\chi_{\{|f|>t\}}\in L\/$, for every  $t>0\/$, and

429:

430: \item[(iii)] there exists a sequence of simple functions $(f_n)_n\subset L\/$ such that $0\le f_n(x)\le

431: f_{n+1}(x)\to |f(x)|\hbox{\ a.e.\ }x\in  X\/$.

432: \end{enumerate}

433: \edefi

434:

435: \beje

436: \label{eje: I.2.5}

437:

438: (i) If $X\/$ is an arbitrary measure space,  every functional lattice\index{functional lattice} in $X\/$

439: (i.e., a vector space

440: $L\subset\M(X)\/$ with

441: $g\in L\/$ if $|g|\le|f|,\;f\in L\/$) with the  Fatou property\index{Fatou property} (see \cite{BS}), is a

442: regular class. In particular the  Lebesgue space

443: $L^p(X)\/$ and the  Lorentz space $L^{p,q}(X),\;0<p,q\le\infty\/$ (see (\ref{normlor})) are regular classes.

444:

445: (ii) If $C\/$ is a class formed by measurable sets in $X\/$  containing the empty set,

446: $$

447: L_C=\big\{f\in\M(X)\,:\,\{|f|>t\}\in C,\;\forall t\ge0\big\}

448: $$

449: is a  regular class in $X\/$. To see this we observe that the conditions (i) and (ii) of

450: Definition~\ref{def: I.2.4} are immediate to be checked while to prove (iii) we observe that if

451: $0\le f\in L_C\/$, the simple functions

452: $$

453: f_n=\sum_{k=0}^{n2^{n-1}}k2^{-n}\chi_{\{k2^{-n}<f\le(k+1)2^{-n}\}}+n\chi_{\{f>n\}},\quad

454: n=1,2,\dots,

455: $$

456: form an increasing sequence that converges pointwise to    $f\/$, whose  level sets are also level sets

457: of   $f\/$  and hence they belong to  $C\/$.

458:

459: (iii) If $L\/$ is a  regular class in  $X\/$,

460: $$

461: L^\ast=\big\{f^\ast\,:\,f\in L\big\}

462: $$

463: is a  regular class in $\Rm\/$.

464:

465: (iv) In $\Rm\/$ the positive decreasing functions are a regular class and the same holds for the

466: increasing functions.

467:

468: (v) In $\Rn\/$   radial functions, positive and decreasing  radial functions, or  positive and

469: increasing  radial functions are also regular classes.

470: \eeje

471:

472:

473:

474: \bdefi

475: \label{def: I.2.6}

476: Let  $Y\/$ be a   measure space, $L\/$ a regular class and $T:L\to \M(Y)\/$

477: an operator.

478:

479: (i) $T\/$ is sublinear if $|T(\alpha_1 f_1+\dots+\alpha_n f_n)(y)|\le|\alpha_1

480: Tf_1(y)|+\dots+|\alpha_n Tf_n(y)|\/$ a.e.

481: $y\in Y\/$, for every $\alpha_1,\dots,\alpha_n\in\R$, and every $f_1,\dots,\,f_n\in L\/$ such that $\alpha_1

482: f_1+\dots+\alpha_n f_n\in L\/$.

483:

484: (ii) $T\/$ is monotone if $|Tf(y)|\le|Tg(y)|\hbox{\ a.e.\ }y\in Y\/$, if  $|f(x)|\le|g(x)|\hbox{\

485: a.e.\ }x\in X,\;f,g\in L.\/$

486:

487: (iii) We say that $T\/$ is order continuous\index{order continuous} if it is  monotone and if for every

488: sequence

489: $(f_n)_n\subset L\/$  with

490: $0\le f_n(x)\le f_{n+1}(x)\to f(x)\in L\hbox{\ a.e.\ }x\in X\/$,  we have that $\lim_n

491: |Tf_{n}(y)|=|Tf(y)|\hbox{\ a.e.\ }y\in Y\/$.

492: \edefi

493:

494:

495: \bobs

496: \label{obs: I.2.7}

497: It is immediate that a sublinear and  monotone  operator satisfies the following properties:

498:

499: (i) $T(0)(y)=0\hbox{\ a.e.\ }y\in Y\/$,

500:

501: (ii) $\big|T|f|(y)\big|=|Tf(y)|\hbox{\ a.e.\ }y\in Y,\;f\in L\/$.

502: \eobs

503:

504: \bobs

505: \label{obsI.2.77}

506: (i) Every maximal operator of the form

507: $$

508: T^\ast f(x)=\sup_{T\in B}|Tf(x)|\,,\quad x\in X,\quad f\in L,

509: $$

510: is order continuous, where, for every  $x\in X\/$,

511: $B\/$ is a set of order continuous operators $T$.

512:

513: (ii) In particular every  integral operator $Tf(r)=\displaystyle\int_0^{\infty} k(r,t)f(t)\,dt,\,r>0,\;f\in

514: L\subset

515: \M^+(\Rm)\/$ (with $k:\Rm\times\Rm\to[0,\infty)\/$), is order continuous. For example the Hardy operator

516: (see (\ref{hardy})) and its conjugate

517: $$

518: Qf(r)=\int_r^\infty f(t)\,{dt\over t}.

519: $$

520:

521:

522: (iii)  Also the identity operator

523: $f\mapsto f\/$ is obviously order continuous.

524: \eobs

525:

526: The following  property  of the sublinear and monotone operators will be  fundamental.

527:

528: \bteo

529: \label{te: I.2.8}

530:  Let  $L\subset \M(X)\/$ be a regular class and  $T:L\to \M(Y)\/$ a

531: sublinear and monotone operator. Then, for every simple function     $f\in L\/$,

532: $$

533: |Tf(y)|\le\int_0^\infty\big|T\chi_{\{|f|>t\}}(y)\big|\,dt\,,\quad\hbox{a.e.\ } y\in Y.

534: $$

535: \eteo

536:

537: \bdem

538: By Remark~\ref{obs: I.2.7} (ii) we can assume $f\ge 0\/$. Then,

539: $$

540: f=\sum_{n=1}^N a_n\,\chi_{B_n},

541: $$

542: with $a_1,a_2,\dots,a_N>0\/$ and  $(B_j)_j\/$ is an increasing sequence of measurable sets in

543: $X\/$:

544: $B_1\subset B_2\subset\dots\subset B_N\/$. Set

545: $A_n=\sum_{j=n}^N a_j\,,\;n=1,\dots,N,$\break $A_{N+1}=0\/$ and note that

546: $\{f>t\}=\emptyset\/$ if $t\ge A_1\/$ and $\{f>t\}=B_n\/$ for $A_{n+1}\le t<A_n\,,\;n=1,\dots,N\/$.

547: In particular   $\chi_{B_n}\/$ is in  $L\/$ (by Definition~\ref{def: I.2.4} (ii)) and since  $T\/$

548: is sublinear, it follows that

549: \bequ

550: \label{I.2.9}

551: |Tf(y)|\le\sum_{n=1}^N a_n|T\chi_{B_n}(y)|\,,\quad\hbox{a.e.\ }y\in Y.

552: \eequ

553: On the other hand

554: $T\chi_\emptyset(y)=0\/$ a.e. $y\in Y\/$ (by Remark~\ref{obs: I.2.7} (i)) and hence

555: $$

556: \int_0^\infty\big|T\chi_{\{f>t\}}(y)\big|\,dt=\sum_{n=1}^N\int_{A_{n+1}}^{A_n}|T

557: \chi_{B_n}(y)|\,dt=\sum_{n=1}^N

558: a_n|T\chi_{B_n}(y)|\,,

559: $$

560: since $A_n-A_{n+1}=a_n\/$. The last expression coincides with the right hand side of  (\ref{I.2.9}) and the

561: theorem is proved.$\qquad\qed$

562: \edem

563:

564:

565: \bobs

566: \label{obs: I.2.10}

567: If the  operator   $T\/$ is linear and  positive ($Tf\ge0\/$ if $f\ge0\/$), the inequality

568: in the previous theorem is, in fact, an equality when

569: $f\ge0\/$.

570: \eobs

571:

572: The following equality will be very much useful later on and it has been proved by several authors.

573: See, for example,  \cite{HM}.

574:

575: \blem

576: \label{lem: I.2.11}

577:  If $0<p\le 1\/$,

578: $$

579: \sup_{f\downarrow}\,{\bigg(\displaystyle\int_0^\infty f(t)\,dt\bigg)^p\over\displaystyle\int_0^\infty

580: f^p(t)t^{p-1}\,dt}=p\,.

581: $$

582: \elem

583:

584: We introduce now some new spaces which play an important role in this theory, and make some important

585: comments about its definition.

586:

587: \bdefi\label{def: Lor}

588: If $0<p,q<\infty$, we define the Lorentz space\index{Lorentz space}

589: \bequ\label{normlor}

590: L^{p,q}(X)\index{$L^{p,q}(X)$}=\bigg\{f:\Vert

591: f\Vert_{L^{p,q}(X)}=\bigg(\int_0^{\infty}(t^{1/p}f^*(t)\big)^q\,{dt\over t}\bigg)^{1/q}<\infty\bigg\}.

592: \eequ

593: If $q=\infty$ the space $L^{p,\infty}(X)$\index{$L^{p,\infty}(X)$} is defined with the usual modification.

594: When the space

595: $X$ is clearly understood on the context, we will simply write $\Vert f\Vert_{p,q}.$

596: \edefi

597:

598: \bobs

599: \label{obs: I.2.12}

600:

601: (i) If $1\le q\le p\/$, the functional $\|\cdot\|_{p,q}\/$ is a norm. In general, it is only a quasi-norm

602: (see

603: \cite{SW}).

604:

605: (ii) In the case $1<p<q\le\infty\/$ we can define

606: $$

607: \|f\|_{(p,q)}=\bigg(\int_0^\infty \big(t^{1/p}f^{\ast\ast}(t)\big)^q\,{dt\over t}\bigg)^{1/q}

608: $$

609: (with the usual modification  if $q=\infty\/$) which is a norm (see \cite{BS}) and it is equivalent

610: to the original quasi-norm:

611: \bequ\label{eqlo}

612: \|f\|_{p,q}\le\|f\|_{(p,q)}\le{p\over p-1}\,\|f\|_{p,q},\quad f\in\M(X).

613: \eequ

614:

615: (iii) It is easy to show that $\Vert

616: f\Vert_{p,q}=\displaystyle\bigg(\int_0^{\infty}(t\lambda_f^{1/p}(t))^q\,{dt\over t}\bigg)^{1/q}$ (see

617: Proposition \ref{pro: II.2.5}).

618: \eobs

619:

620: Let us state the main result of this section.

621:

622: \bteo

623: \label{te: I.2.13}

624: Let $L\subset \M(X)\/$ be a regular class, $T:L\to \M(Y)\/$ an order continuous  sublinear

625: operator and  $0<q_0\le1,\,0<p_0<\infty\/$. Then

626: \begin{enumerate}

627: \item[(a)] If $q_0\le q_1\le p_1<\infty$,

628: $$

629: \sup_{f\in L}{\|Tf\|_{L^{p_1,q_1}(Y)}\over \|f\|_{L^{p_0,q_0}(X)}}=\sup_{\chi_B\in

630: L}{\|T\chi_B\|_{L^{p_1,q_1}(Y)}\over \|\chi_B\|_{L^{p_0,q_0}(X)}}.

631: $$

632:

633: \item[(b)] If $q_0<p_1<q_1\le\infty\/$,

634: $$

635: \sup_{f\in L}{\|Tf\|_{L^{p_1,q_1}(Y)}\over \|f\|_{L^{p_0,q_0}(X)}}\le\bigg({p_1\over

636: p_1-q_0}\bigg)^{1/q_0}\,\sup_{\chi_B\in L}{\|T\chi_B\|_{L^{p_1,q_1}(Y)}\over

637: \|\chi_B\|_{L^{p_0,q_0}(X)}}.

638: $$

639: \end{enumerate}

640: \eteo

641:

642: \bdem

643: Let

644: $$

645: C=\sup_{\chi_B\in L}{\|T\chi_B\|_{L^{p_1,q_1}(Y)}\over \|\chi_B\|_{L^{p_0,q_0}(X)}}.

646: $$

647: We have to prove that  $\|Tf\|_{L^{p_1,q_1}(Y)}\le K\,C\,\|f\|_{L^{p_0,q_0}(X)}\/$ for every  $f\in

648: L\/$ with $K=1\/$ (in the case (a)) or $K=\big(p_1/(p_1-q_0)\big)^{1/q_0}\/$ (case (b)). By

649: Remark~\ref{obs: I.2.7} (ii) we can assume $f\ge0\/$ and since there exists an increasing sequence

650: of positive simple functions

651: $(f_n)_n\subset L\/$  converging to  $f\/$ a.e., with

652: $$

653: |Tf_n(y)|\le|Tf_{n+1}(y)|\rightarrow|Tf(y)|\qquad\hbox{a.e.\ }y,

654: $$

655: (Definitions~\ref{def: I.2.4} and \ref{def: I.2.6}), by the monotone convergence theorem, it is

656: enough to prove the previous inequality for

657: $(f_n)_n\/$. That is, we can assume that $0\le f\in L\/$ is a simple function.

658:

659: Let $p=p_1/q_0,\,q=q_1/q_0\/$. By Theorem~\ref{te: I.2.8} and Lemma~\ref{lem: I.2.11},

660: $$

661: \|Tf\|_{p_1,q_1}^{q_0}=\big\||Tf|^{q_0}\big\|_{p,q}\le\bigg\|\int_0^\infty q_0

662: t^{q_0-1}\big|T\chi_{\{f>t\}}(\cdot)\big|^{q_0}\,dt\,\bigg\|_{p,q}.

663: $$

664: In the case (a),

665: $\|\cdot\|_{p,q}\/$ is a norm and we obtain that

666: $$

667: \|Tf\|_{p_1,q_1}^{q_0}\le\int_0^\infty q_0

668: t^{q_0-1}\big\|\big|T\chi_{\{f>t\}}\big|^{q_0}\big\|_{p,q}\,dt.

669: $$

670: In the case  (b),

671: $\|\cdot\|_{(p,q)}\/$ is a norm satisfying (\ref{eqlo})

672: and it follows that

673: \begin{eqnarray*}

674: \|Tf\|_{p_1,q_1}^{q_0}&\le&{p\over p-1}\int_0^\infty q_0

675: t^{q_0-1}\big\|\big|T\chi_{\{f>t\}}\big|^{q_0}\big\|_{p,q}\,dt\\ &=&{p_1\over p_1-q_0}\int_0^\infty

676: q_0 t^{q_0-1}\big\|\big|T\chi_{\{f>t\}}\big|^{q_0}\big\|_{p,q}\,dt.

677: \end{eqnarray*}

678: That is, in any case

679: \begin{eqnarray*}

680: \|Tf\|_{p_1,q_1}^{q_0}&\le& K^{q_0}\int_0^\infty q_0

681: t^{q_0-1}\big\|\big|T\chi_{\{f>t\}}\big|^{q_0}\big\|_{p,q}\,dt\\ &=&K^{q_0}\int_0^\infty q_0

682: t^{q_0-1}\big\|T\chi_{\{f>t\}}\big\|_{p_1,q_1}^{q_0}\,dt\\ &\le& (KC)^{q_0}\int_0^\infty q_0

683: t^{q_0-1}\big\|\chi_{\{f>t\}}\big\|_{p_0,q_0}^{q_0}\,dt\\ &=&(KC)^{q_0}\int_0^\infty p_0

684: t^{q_0-1}\big(\lambda_f(t)\big)^{q_0/p_0}\,dt\\ &=&(KC)^{q_0}\|f\|_{p_0,q_0}^{q_0}.\qquad\qed

685: \end{eqnarray*}

686: \edem

687:

688: Applying the previous result to the strong (i.e., diagonal) case $T:L^{p_0}\rightarrow L^{p_1}\/$ we obtain:

689:

690: \bcor

691: \label{cor: I.2.14}

692:  Let $L\subset \M(X)\/$ be a regular class and  $T:L\to \M(Y)\/$ an order continuous

693: sublinear operator. If

694: $0<p_0\le 1,\,p_0\le p_1<\infty\/$ we have

695: $$

696: \sup_{f\in L}{\|Tf\|_{L^{p_1}(Y)}\over \|f\|_{L^{p_0}(X)}}=\sup_{\chi_B\in

697: L}{\|T\chi_B\|_{L^{p_1}(Y)}\over \|\chi_B\|_{L^{p_0}(X)}}.

698: $$

699: \ecor

700:

701:

702: \bobs

703: \label{obs: I.2.15}

704:  The range of exponents $p_0,p_1\/$ in Corollary~\ref{cor: I.2.14} is optimal. To see this,

705: observe:

706: \begin{enumerate}

707: \item[(i)] The result is not true if $p_0>1\/$. A counterexample is the

708:  Hardy operator $A\/$. Given $1<p_0\le p_1\/$, a necessary and sufficient condition  to have the

709: boundedness

710: $A:L^{p_0}_{\hbox{\decp dec}}(w_0)\to L^{p_1}(w_1)\/$ is (see \cite{Saw2}),

711: \begin{eqnarray}

712: \label{I.2.16}

713:  W_1^{1/p_1}(r)&\le& C\,W_0^{1/p_0}(r),\nonumber\\

714: \bigg(\int_r^\infty{w_1(t)\over

715: t^{p_1}}\,dt\bigg)^{1/p_1}\bigg(\int_0^r\bigg({W_0(t)\over

716: t}\bigg)^{-p_0^\prime}w_0(t)\,dt\bigg)^{1/p_0^\prime}&\le& C.

717: \end{eqnarray}

718: It is easy to see that the condition

719: $$\big\|A\chi_{(0,r)}\big\|_{L^{p_1}(w_1)}\le

720: C\,\big\|\chi_{(0,r)}\big\|_{L^{p_0}(w_0)},\;r>0\/$$

721: obtained from Corollary~\ref{cor: I.2.14}  is equivalent to

722: the inequalities

723: \begin{eqnarray*}

724:  W_1^{1/p_1}(r)&\le& C_1\,W_0^{1/p_0}(r),\quad r>0,\\

725: \bigg(\int_r^\infty\bigg({r\over t}\bigg)^{p_1}w_1(t)\,dt\bigg)^{1/p_1}&\le&

726: C_1\,W_0^{1/p_0}(r),\quad r>0.

727: \end{eqnarray*}

728: Observe that these two last inequalities are true for the weights

729: $w_0(t)=t^{p_0-1}\,,w_1(t)=t^{p_1-1}\chi_{(1,2)}(t)\/$  while these weights do not satisfy

730: (\ref{I.2.16}).

731:

732: \item[(ii)] The statement does not hold if  $p_1<p_0\/$. To see this, it is enough to consider

733: $T=\hbox{\rm Id}:\lpwzdec\to\lpwu\/$. Then,

734: \bequ

735: \label{I.2.17}

736: \sup_{f\downarrow}{\displaystyle\int_0^\infty

737: f^{p_1}(t)w_1(t)\,dt\over\bigg(\displaystyle\int_0^\infty

738: f^{p_0}(t)w_0(t)\,dt\bigg)^{p_1/p_0}}=\sup_{g\downarrow}{\displaystyle\int_0^\infty

739: g(t)w_1(t)\,dt\over\bigg(\displaystyle\int_0^\infty g^{p_0/p_1}(t)w_0(t)\,dt\bigg)^{p_1/p_0}}.

740: \eequ

741: The last supremum is equivalent (c.f.

742: \cite{Saw2}), up to multiplicative constants depending only on  $p_0/p_1\/$, to

743: \bequ

744: \label{I.2.18}

745: \bigg(\int_0^\infty\bigg({W_1(t)\over

746: W_0(t)}\bigg)^{p_1/(p_0-p_1)}w_1(t)\,dt\bigg)^{(p_0-p_1)/p_0}.

747: \eequ

748: On the other hand, if  Corollary~\ref{cor: I.2.14}  were true in this case, (\ref{I.2.17}) would be

749: equal to

750: $$

751: \sup_{r>0}{\|\chi_{(0,r)}\|_{\lpwu}\over\|\chi_{(0,r)}\|_{\lpwz}}=\sup_{r>0}{W_1^{1/p_1}(r)\over

752: W_0^{1/p_0}(r)}\,.

753:  $$

754: This last supremum is finite for the weights

755: $w_i(t)=t^{\alpha_i}\,,\;i=0,1\/$, if

756: $(1+\alpha_1)/p_1=(1+\alpha_0)/p_0\,,\;\alpha_0,\alpha_1>-1\/$, while   (\ref{I.2.18}) is always

757: infinite in this case whatever

758: $\alpha_1,\alpha_2\/$ are.

759: \end{enumerate}

760: \eobs

761:

762: Considering the weak-type case $T:L^{p_0}\rightarrow L^{p_1,\infty}\/$, we  have as a

763: consequence of Theorem~\ref{te: I.2.13}  the following statement.

764:

765:

766: \bcor

767: \label{cor: I.2.19}

768:  Let $L\subset \M(X)\/$ be a  regular class and let $T:L\to \M(Y)\/$ be an order continuous sublinear

769: operator.   If $0<p_0\le 1,\,p_0<p_1<\infty\/$,

770: $$

771: \sup_{f\in L}{\|Tf\|_{L^{p_1,\infty}(Y)}\over \|f\|_{L^{p_0}(X)}}\le\bigg({p_1\over

772: p_1-p_0}\bigg)^{1/p_0}\sup_{\chi_B\in L}{\|T\chi_B\|_{L^{p_1,\infty}(Y)}\over

773: \|\chi_B\|_{L^{p_0}(X)}}.

774:  $$

775: \ecor

776:

777:

778: \bobs

779: \label{obs: I.2.20} The previous corollary is not true if  $p_0>1\/$, as  can be seen in the case

780: \bequ

781: \label{I.2.21}

782: A\,:\,\lpwzdec\longrightarrow\lpwinu.

783: \eequ

784: As  was proved in \cite{CS2} (see also \cite{Ne1}), a necessary and sufficient condition to have

785: (\ref{I.2.21}) is

786: \begin{eqnarray}

787: \label{I.2.22}

788: \bigg(\int_0^r\bigg({W_0(t)\over

789: t}\bigg)^{-p_0^\prime}w_0(t)\,dt\bigg)^{1/p_0^\prime}W_1^{1/p_1}(r)&\le& Cr,\quad r>0,\\

790: W_1^{1/p_1}(r)&\le& CW_0^{1/p_0}(r),\quad r>0,\nonumber

791: \end{eqnarray}

792: that   does not coincide with the condition

793: $$

794: {W_1^{1/p_1}(t)\over t}\le C\,{W_0^{1/p_0}(r)\over r}\,,\quad 0<t<r<\infty\,,

795: $$

796: which can be obtained applying Corollary~\ref{cor: I.2.19}.  For example, the two weights

797: in Remark~\ref{obs: I.2.15} (iii.1) satisfy this last condition but  not

798: (\ref{I.2.22}).

799: \eobs

800:

801:

802: A consequence of  Theorem~\ref{te: I.2.13}

803: and the   Marcinkiewicz interpolation theorem is the following result  on interpolation of

804: operators of restricted weak-type, which is a generalization of the well known

805: Stein-Weiss theorem (Theorem 3.15 in \cite{SW} or Theorem 5.5 in \cite{BS}).

806:

807: \bteo

808: \label{te: I.2.21}

809:  Let $0<p_0,p_1<\infty,\;0<q_0,q_1\le\infty,\;p_0\neq p_1,\;q_0\neq q_1\/$ and let us assume that

810: $T:\big(L^{p_0,1}+L^{p_1,1}\big)(X)\to\M(Y)\/$ is a sublinear order continuous  operator satisfying

811: \begin{eqnarray*}

812: \big\|T\chi_B\big\|_{q_0,\infty}&\le& C_0\|\chi_B\|_{p_0,1},\quad B\subset X,\\

813: \big\|T\chi_B\big\|_{q_1,\infty}&\le& C_1\|\chi_B\|_{p_1,1},\quad B\subset X.

814: \end{eqnarray*}

815:  Then

816: $$

817: T\,:\,L^{p,r}(X)\longrightarrow L^{q,r}(Y),\qquad 0<r\le\infty,

818: $$

819: if

820: $$

821: {1\over p}={1-\theta\over p_0}+{\theta\over p_1},\qquad {1\over q}={1-\theta\over

822: q_0}+{\theta\over q_1},\qquad0<\theta<1.

823: $$

824: \eteo

825:

826: \bdem

827: Since $L^{s,t_0}\subset L^{s,t_1}\/$ if  $t_0<t_1\/$, there exist indices

828: $r_0,r_1\in(0,1)\/$ with $r_i<q_i,\;i=0,1\/$ and such that

829: $$

830: \big\|T\chi_B\big\|_{q_i,\infty}\le C_i\|\chi_B\|_{p_i,r_i},\quad B\subset X,\qquad i=0,1.

831: $$

832: Theorem~\ref{te: I.2.13}

833: tells us that the previous inequality holds for every function   $f\in

834: L^{p_i,r_i}\/$  and the result follows from the general Marcinkiewicz interpolation theorem

835:  (Theorem 5.3.2 in \cite{BL}).$\qquad\qed$

836: \edem

837:

838: \bobs

839: \label{obs: I.2.22}

840:

841: (i) The norm of a characteristic  function    in $L^{p,q}\/$ does not depend (up to constants)

842: on $q\/$. Therefore, the previous  theorem remains true if we substitute the original spaces

843: $L^{p_i,1}\/$ by  $L^{p_i,r_i}\/$ with $0<r_i\le\infty,\;i=0,1\/$.

844:

845: (ii) In the classical result of  Stein-Weiss mentioned above, the previous result is proved for a

846: more restricted set of indices:

847: $$

848: 1\le p_0,p_1<\infty,\qquad 1\le q_0,q_1\le\infty.

849: $$

850:

851: (iii) Corollary~\ref{cor: I.2.14} can be also generalized to consider the case of two operators. The

852: following result gives such extension (the proof is similar to the one given in Corollary~\ref{cor: I.2.14},

853: and it is also based upon some ideas found in \cite{BPS}).

854: \eobs

855:

856: \bteo\label{twoop}

857: Let $L\subset{\cal M}(X)$ be a regular class, and let $T_i:L\longrightarrow {\cal M}(Y_i)$, $i=0,1$ be two

858: order continuous operators. Assume that $T_0$ is positive and linear and $T_1$ is sublinear. Then, for each

859: of the following cases:

860: \begin{enumerate}

861: \item[(a)] $0<p_0\le1\le p_1<\infty$,

862: \item[(b)] $T_1=\hbox{\rm Id}$, $1\le p_1<\infty$, $0<p_0\le p_1$,

863: \item[(c)] $T_0=\hbox{\rm Id}$, $0< p_0\le1$, $p_0\le p_1<\infty$,

864: \item[(d)] $T_0=T_1$,  $0<p_0\le p_1<\infty$,

865: \end{enumerate}

866: we have that

867: $$

868: \sup_{f\in L}{\Vert T_1 f\Vert_{L^{p_1}(Y_1)}\over \Vert T_0 f\Vert_{L^{p_0}(Y_0)}}=

869: \sup_{\chi_B\in L}{\Vert T_1 \chi_B\Vert_{L^{p_1}(Y_1)}\over \Vert T_0 \chi_B\Vert_{L^{p_0}(Y_0)}}.

870: $$

871:

872: \eteo

873:

874: \section{The Hardy operator and the classes $B_p$}

875: \label{se I: Hardy operator}

876:

877: The Hardy operator $A$

878: defined in (\ref{hardy})

879: will play a fundamental role in the following chapters. In particular we shall be interested in the

880: boundedness

881: \bequ

882: \label{I.6.1}

883: A\,:\,\lpwzdec\longrightarrow\lpwu

884: \eequ

885: and, also,

886: \bequ

887: \label{I.6.2}

888: A\,:\,\lpwzdec\longrightarrow\lpwinu.

889: \eequ

890: The  diagonal case $A:L^p_{\hbox{\decp

891: dec}}(w)\to\lpw\,,p>1\/$ was solved by   Ari\~no and Muckenhoupt in \cite{AM1}.

892: The condition on the weight  $w\/$ is known as $B_p\/$. This motivates the following

893: definition.

894:

895: \bdefi

896: \label{def: I.6.3}

897: We write  $w\in B_p\/$\index{$B_p$} if

898: $$

899: A\,:\,L^p_{\hbox{\decp dec}}(w)\longrightarrow\lpw,

900: $$

901: and  $w\in B_{p,\infty}\/$\index{$B_{p,\infty}$} if

902: $$

903: A\,:\,L^p_{\hbox{\decp dec}}(w)\longrightarrow L^{p,\infty}(w).

904: $$

905: Analogously we write

906: $(w_0,w_1)\in B_{p_0,p_1}\/$\index{$B_{p_0,p_1}$} if the boundedness  (\ref{I.6.1}) holds and we say that

907: $(w_0,w_1)\in B_{p_0,p_1,\infty}\/$\index{$B_{p_0,p_1,\infty}$} if  (\ref{I.6.2}) holds.

908: \edefi

909:

910: \bobs

911: \label{obs: I.6.4}

912: Since $\lpw\subset L^{p,\infty}(w)\/$ we have that

913: $$

914: B_{p_0,p_1}\subset B_{p_0,p_1,\infty},\quad0<p_0,p_1<\infty.

915: $$

916: In particular $B_p\subset

917: B_{p,\infty},\,0<p<\infty\/$.

918: \eobs

919:

920: The characterization of the weights satisfying (\ref{I.6.2}) is obtained applying directly  Theorem

921: 3.3 in \cite{CS2}:

922:

923: \bteo

924: \label{te: I.6.5}

925: Let $0<p_1<\infty\/$. Then,

926: \begin{enumerate}

927: \item[(a)]\ \ If $p_0>1\/$ the following conditions are equivalent:

928: \begin{enumerate}

929: \item[(i)] $(w_0,w_1)\in B_{p_0,p_1,\infty},$

930: \item[(ii)] $\bigg(\displaystyle\int_0^r\bigg({W_0(t)\over

931: t}\bigg)^{1-p_0^\prime}\,dt\bigg)^{1/p_0^\prime}W_1^{1/p_1}(r)\le Cr,\quad r>0,$

932: \item[(iii)]

933: \begin{eqnarray*}

934: \bigg(\int_0^r\bigg({W_0(t)\over

935: t}\bigg)^{-p_0^\prime}w_0(t)\,dt\bigg)^{1/p_0^\prime}W_1^{1/p_1}(r)&\le& Cr,\\

936: W_1^{1/p_1}(r)&\le&

937: CW_0^{1/p_0}(r),\quad r>0.\\

938: \end{eqnarray*}

939: \end{enumerate}

940:

941: \item[(b)]\ \ If $p_0\le1\/$ the following conditions are equivalent:

942: \begin{enumerate}

943: \item[(i)] $(w_0,w_1)\in B_{p_0,p_1,\infty},$

944: \item[(ii)] ${W_1^{1/p_1}(r)\over r}\le

945: C{W_0^{1/p_0}(t)\over t},\quad0<t<r.$

946: \end{enumerate}

947: \end{enumerate}

948: \eteo

949:

950:

951: The strong boundedness $A:\lpwzdec\to\lpwu\/$ is not so easy and, in fact, the case

952:  $0<p_1<p_0<1\/$ is still open. The result that follows characterizes the classes $B_p\/$. The

953: proof of

954: $(i) \Leftrightarrow (ii)\/$ can be found in  \cite{AM1}  (case $p\ge1\/$) and it is a consequence of

955: Corollary~\ref{cor: I.2.14}  in the case $p<1\/$. The equivalences with  (iii) and (iv) are proved

956: in  \cite{Sor}.

957:

958: \bteo{(Ari\~no-Muckenhoupt, J. Soria)}

959: \label{te: I.6.6}

960: For $0<p<\infty\/$ the following statements are  equivalent:

961: \begin{enumerate}

962: \item[(i)] $ w\in B_p.$

963: \item[(ii)] $\displaystyle\int_r^\infty\bigg({r\over t}\bigg)^p w(t)\,dt\le

964: C\int_0^r w,\quad r>0.$

965: \item[(iii)] $\displaystyle\int_0^r{t^{p-1}\over W(t)}\,dt\le C{r^p\over W(r)},\quad

966: r>0.$

967: \item[(iv)] $\displaystyle\int_0^r {1\over W^{1/p}(t)}\,dt\le C{r\over W^{1/p}(r)},\quad r>0.$

968: \end{enumerate}

969: \eteo

970:

971: The classes $B_{p_0,p_1}\/$ with $p_0\le1,\,p_0\le p_1\/$ can be characterized  using

972: Corollary~\ref{cor: I.2.14} (see also \cite{CS2}). The case $p_1,p_0>1\/$ was solved by  Sawyer

973: (\cite{Saw2}). Stepanov (\cite{Sp2}) solved the case

974: $0<p_1\le 1<p_0\/$ and Sinnamon and Stepanov (\cite{SS}) solved the case $0<p_1<1=p_0\/$.

975:

976:

977: We shall also be interested in the  boundedness  of the discrete  Hardy operator $A_d\/$\index{Hardy

978: operator!discrete} acting on sequences  $f=\big(f(n)\big)_{n\ge0}\/$ in the form

979: $$

980: A_df(n)={1\over n+1}\sum_{k=0}^n f(k),\qquad n=0,1,2,\dots

981: $$

982: In the next chapter, it will be very much useful to know the boundedness of

983: \bequ

984: \label{I.6.7}

985: A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w),

986: \eequ

987: and also,

988: \bequ

989: \label{I.6.8}

990: A_d\,:\,\ell^{p,\infty}_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w),

991: \eequ

992: where

993: $w=\big(w(n)\big)_n\/$ is a weight in  $\N^\ast\/$, that is, a sequence of positive numbers, and

994: $\ell^p(w)\/$\index{$\ell^p(w)\/$}  is the  Lebesgue space $L^p\/$ in $\N^\ast\/$, with measure $\sum_n

995: w(n)\delta_{\{n\}}\/$, that is,

996: $$

997: \ell^p(w)=\bigg\{f=\big(f(n)\big)_n\subset\C\,:\,\|f\|_{\ell^p(w)}=\bigg(\sum_{n=0}^\infty|f(n)|^p

998: w(n)\bigg)^{1/p}<\infty\,\bigg\}.

999: $$

1000: $\ell^p_{\hbox{\decp dec}}(w)\/$\index{$\ell^p_{\hbox{\decp dec}}(w)$} is the class of positive and

1001: decreasing sequences $f\/$  (we shall use the notation

1002: $f\d\/$) in $\ell^p(w)\/$ while  $\ell^{p,\infty}(w)\/$\index{$\ell^{p,\infty}(w)\/$} is the weak version

1003: of  $\ell^p(w)\/$ and it is defined by the semi-norm

1004: $$

1005: \|f\|_{\ell^{p,\infty}(w)}=\sup_{t>0}t^{1/p}f^\ast_w(t),

1006: $$

1007: where $f^\ast_w\/$ is the decreasing rearrangement of  $f=\big(f(n)\big)_n\/$

1008: in the measure space

1009: $\big(\N^\ast,\sum_n w(n)\delta_{\{n\}}\big)\/$. Similarly as  was done in  $\Rm\/$, for each

1010: weight

1011: $w\/$ (resp. $w_0,u,\dots\/$) in $\N^\ast\/$ we denote by  $W\/$ (resp. $W_0,U,\dots\/$)   the

1012: sequence,

1013: \bequ

1014: \label{I.6.9}

1015: W(n)=\sum_{k=0}^n w(k),\qquad n=0,1,2,\dots

1016: \eequ

1017: With this notation, it can be easily seen that

1018: \bequ

1019: \label{I.6.10}

1020: \|f\|_{\ell^{p,\infty}(w)}=\sup_{n\ge0}W^{1/p}(n)f(n),\qquad f\d.

1021: \eequ

1022:

1023:

1024: It is now the moment to recall an important result due to    E. Sawyer

1025: which  will be used several times in what follows. The original proof can be seen in \cite{Saw2}

1026: and other proofs in  \cite{CS2} and \cite{Sp2}.

1027:

1028: \bteo

1029: \label{te: I.5.7}

1030: If $1<p<\infty\/$ we have that

1031: \begin{eqnarray*}

1032: \sup_{f\downarrow}{\|f\|_{L^1(w_1)}\over\|f\|_{L^p(w_0)}}&\approx&\bigg(\int_0^\infty\bigg({W_1(t)\over

1033: W_0(t)}\bigg)^{p^\prime-1}w_1(t)\,dt\bigg)^{1/p^\prime}\\ &\approx&\bigg(\int_0^\infty\bigg({W_1(t)\over

1034: W_0(t)}\bigg)^{p^\prime}w_0(t)\,dt\bigg)^{1/p^\prime}+{W_1(\infty)\over W_0^{1/p}(\infty)}.

1035: \end{eqnarray*}

1036: \eteo

1037:

1038: Let us now formulate a discrete version of the above result.

1039:

1040: \bteo

1041: \label{te: I.6.11}  Let $w=\big(w(n)\big)_n,\;v=\big(v(n)\big)_n\/$ be weights in $\N^\ast\/$ and let

1042: $$

1043: S=\sup_{f\d}{\sum_{n=0}^\infty f(n)v(n)\over\big(\sum_{n=0}^\infty f(n)^p w(n)\big)^{1/p}}.

1044: $$

1045: Then,

1046: \begin{enumerate}

1047: \item[(i)] If $0<p\le1\/$,

1048: $$

1049: S=\sup_{n\ge0}{V(n)\over W^{1/p}(n)},

1050: $$

1051: with $W\/$ defined by (\ref{I.6.9}) and  $V\/$ analogously.

1052:

1053: \item[(ii)] If $1<p<\infty\/$,

1054: \begin{eqnarray*}

1055: S&\approx&\bigg(\int_0^\infty\bigg({\widetilde{V}(t)\over\widetilde{W}(t)}\bigg)^{p^\prime-1}

1056: \tilde{v}(t)\,dt\bigg)^{1/p^\prime}\\

1057: &\approx&\bigg(\int_0^\infty\bigg({\widetilde{V}(t)\over\widetilde{W}(t)}\bigg)^{p^\prime}

1058: \tilde{w}(t)\,dt\bigg)^{1/p^\prime}+{\widetilde{V}(\infty)\over\widetilde{W}^{1/p}(\infty)},

1059: \end{eqnarray*}

1060: where $\tilde{v}\/$ is the weight in $\Rm\/$ defined by

1061: $$

1062: \tilde{v}=\sum_{n=0}^\infty v(n)\chi_{[n,n+1)}

1063: $$

1064: and $\widetilde{V}(t)=\displaystyle\int_0^t\tilde{v}(s)\,ds\/$ and analogously for

1065: $\tilde{w}$ and $\widetilde{W}\/$.

1066:

1067: Moreover, the constants implicit in the symbol $\approx\/$ only depend on $p\/$.

1068: \end{enumerate}

1069: \eteo

1070:

1071: \bdem       (i) is obtained applying  Corollary~\ref{cor: I.2.14}  with

1072: $p_1=1,\,p_0=p,\,X=Y=\N^\ast,\,T=\hbox{\rm Id}\/$ to the regular class $L\/$ of decreasing

1073: sequences in

1074: $\N^\ast\/$.

1075:

1076: (ii) can be deduced from Theorem~\ref{te: I.5.7}  observing that

1077: \bequ

1078: \label{I.6.12}

1079: S=\sup_{\tilde{f}\downarrow}{\displaystyle\int_0^\infty\tilde{f}(t)\tilde{v}(t)\,dt\over\bigg(\displaystyle\int_0^\infty

1080: \tilde{f}^p(t)\tilde{w}(t)\,dt\bigg)^{1/p}}.

1081: \eequ

1082: To see this, note that if  $f=\big(f(n)\big)_n\/$ is a decreasing sequence in  $\N^\ast\/$ and we

1083: define

1084: $\tilde{f}=\sum_{n=0}^\infty f(n)\chi_{[n,n+1)}\in\Fd\/$, it is obvious that

1085: $$

1086: {\displaystyle\sum_{n=0}^\infty f(n)v(n)\over\bigg(\displaystyle\sum_{n=0}^\infty f(n)^p

1087: w(n)\bigg)^{1/p}}={\displaystyle\int_0^\infty\tilde{f}(t)\tilde{v}(t)\,dt\over\bigg(\displaystyle\int_0^\infty

1088: \tilde{f}^p(t)\tilde{w}(t)\,dt\bigg)^{1/p}}.

1089: $$

1090: Therefore $S\/$ is less than or equal to the second member in

1091:  (\ref{I.6.12}). On the other hand, if $g\ge0\/$ is a  decreasing function     in $\Rm\/$

1092: and we define the  decreasing sequence $f(n)=\big(\int_n^{n+1}g^p(s)\,ds\big)^{1/p},\;n=0,1,\dots\/$, we

1093: obtain

1094: $$

1095: \int_0^\infty g^p(t)\tilde{w}(t)\,dt=\sum_n f(n)^p w(n),

1096: $$

1097: while by  H\"older's inequality

1098: \begin{eqnarray*}

1099: \int_0^\infty g(t)\tilde{v}(t)\,dt&=&\sum_n v(n)\int_n^{n+1}g(t)\,dt\\

1100: &\le&\sum_n

1101: v(n)\bigg(\int_n^{n+1}g^p(t)\,dt\bigg)^{1/p}=\sum_n v(n)f(n).

1102: \end{eqnarray*}

1103: Hence,

1104: $$

1105: {\displaystyle\int_0^\infty g(t)\tilde{v}(t)\,dt\over\bigg(\displaystyle\int_0^\infty

1106: g^p(t)\tilde{w}(t)\,dt\bigg)^{1/p}}\le{\displaystyle\sum_{n=0}^\infty

1107: f(n)v(n)\over\bigg(\displaystyle\sum_{n=0}^\infty f(n)^p w(n)\bigg)^{1/p}}\le S.

1108: $$

1109: Thus

1110: (\ref{I.6.12}) is proved  and  (ii) follows by applying  Theorem~\ref{te: I.5.7}.$\qquad\qed$

1111: \edem

1112:

1113: The following result characterizes the boundedness   (\ref{I.6.7}).

1114:

1115: \bteo

1116: \label{te: I.6.13}

1117: \ \par

1118: \begin{enumerate}

1119: \item[(a)] If  $0<p\le1\/$ we have that $A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow

1120: \ell^{p,\infty}(w)\/$ if and only if

1121: \bequ

1122: \label{I.6.14}

1123: {W^{1/p}(n)\over n+1}\le C\,{W^{1/p}(m)\over m+1},\qquad 0\le m\le n.

1124: \eequ

1125:

1126: \item[(b)] If $1<p<\infty\/$ the following statements are equivalent:

1127: \begin{enumerate}

1128: \item[(i)] $ A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w),$

1129: \item[(ii)] $\tilde{w}=\sum_{n=0}^\infty w(n)\chi_{[n,n+1)}\in B_p,$

1130: \item[(iii)] $\sum_{k=0}^n{1\over W^{1/p}(k)}\le C\,{n+1\over W^{1/p}(n)},\quad

1131: n=0,1,2,\dots$

1132: \end{enumerate}

1133: \end{enumerate}

1134: \eteo

1135:

1136: \bdem      The boundedness  of $A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow

1137: \ell^{p,\infty}(w)\/$ is equivalent (c.f. (\ref{I.6.10})) to the inequality

1138: $$

1139: W^{1/p}(n)A_d f(n)\le C\|f\|_{\ell^p(w)},\qquad f\d,\,n=0,1,2,\dots

1140: $$

1141: Equivalently

1142: \bequ

1143: \label{I.6.15}

1144: {W^{1/p}(n)\over n+1}\sup_{f\d}{\sum_{k=0}^n f(k)\over\big(\sum_{k=0}^\infty f(k)^p

1145: w(k)\big)^{1/p}}\le C.

1146: \eequ

1147: Hence, (a) is obtained applying  Theorem~\ref{te: I.6.11} (i). Also observe that the condition

1148: (\ref{I.6.14}) is obtained applying (\ref{I.6.15}) to sequences of the form

1149: $f=(1,1,1,\dots,1,0,0,\dots)\/$ (characteristic functions in  $\N^\ast\/$)  and thus it is also

1150: necessary in the case   $p>1\/$.

1151:

1152: To prove  (b), that is the case $p>1\/$, we note, as we already did in  Theorem~\ref{te: I.6.11},

1153: that the discrete supremum in  (\ref{I.6.15}) can be substituted by a supremum on decreasing

1154: functions in

1155: $\Rm\/$ with weights

1156: $\chi_{(0,n+1)}$ and $\tilde{w}\/$:

1157: $$

1158: \sup_{g\d}{\displaystyle\int_0^{n+1} g(t)\,dt\over\bigg(\displaystyle\int_0^\infty

1159: g^p(t)\tilde{w}(t)\,dt\bigg)^{1/p}}.

1160: $$

1161: Therefore

1162: (\ref{I.6.15}) is equivalent to

1163: $$

1164: \widetilde{W}^{1/p}(n+1)Ag(n+1)\le C\|g\|_{L^p(\tilde{w})},\qquad g\d,\;n=0,1,2,\dots

1165: $$

1166: In fact

1167: the previous inequality is true (with constant $2C\/$) substituting $n+1\/$ by any value

1168: $t>0\/$. To see this, if $0<t<1\/$,

1169: \begin{eqnarray*}

1170: \widetilde{W}^{1/p}(t)Ag(t)&=&\big(w(0)t\big)^{1/p}\,t^{-1}\int_0^t g(s)\,ds\\

1171: &\le&\big(w(0)t\big)^{1/p}\,t^{-1}\bigg(\int_0^t g^p(s)\,ds\bigg)^{1/p}t^{1/p^\prime}\\

1172: &\le&\bigg(\int_0^t

1173: g^p(s)\tilde{w}(s)\,ds\bigg)^{1/p}\le\|g\|_{L^p(\tilde{w})},

1174: \end{eqnarray*}

1175: and if  $t\in[n,n+1),\;n\ge1,\/$ we obtain, using  (\ref{I.6.14}),

1176: $$

1177: \widetilde{W}^{1/p}(t)Ag(t)\le\widetilde{W}^{1/p}(n+1)Ag(n)\le

1178: 2C\widetilde{W}^{1/p}(n)Ag(n)\le\|g\|_{L^p(\tilde{w})}.

1179: $$

1180: Summarizing, in the case $p>1\/$ the

1181: boundedness  of $A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w)\/$

1182: is equivalent to the inequality

1183: $$

1184: \widetilde{W}^{1/p}(t)Ag(t)\le C\|g\|_{L^p(\tilde{w})},\qquad t>0,

1185: $$

1186: which holds for every function

1187: $g\d\/$ in $\Rm\/$. But this means that we have the boundedness of the continuous Hardy operator

1188: $$

1189: A\,:\,L^p_{\hbox{\decp dec}}(\tilde{w})\longrightarrow L^{p,\infty}(\tilde{w}),

1190: $$

1191: which is equivalent  (Definition~\ref{def: I.6.3}) to  $\tilde{w}\in B_{p,\infty}=B_p\/$ (the classes

1192: $B_p\/$ and

1193: $B_{p,\infty}\/$ coincide if  $p>1\/$, see \cite{Ne1}). We have then proved the equivalence (i)

1194: and (ii). To see the equivalence with  (iii) let us first observe that this condition

1195: also implies (\ref{I.6.14}) since if  $m\le n\/$,

1196: $$

1197: {m+1\over W^{1/p}(m)}\le\sum_{k=0}^m{1\over W^{1/p}(k)}\le\sum_{k=0}^n{1\over W^{1/p}(k)}

1198: $$

1199: and by

1200: (iii), the last expression is bounded above by  $C(n+1)/W^{1/p}(n)\/$. Hence, in the proof of

1201: (ii)$\Leftrightarrow$(iii) we can assume that (\ref{I.6.14}) holds. With this (and taking into

1202: account the definition of

1203: $\tilde{w}\/$) it is immediate to show the equivalence between the condition

1204: $\tilde{w}\in B_p\/$ (that is (ii)), that by Theorem~\ref{te: I.6.6}  (iv) is

1205: $$

1206: \int_0^r{1\over\widetilde{W}^{1/p}(t)}\,dt\le C\,{r\over\widetilde{W}^{1/p}(r)},\qquad r>0,

1207: $$

1208: and condition  (iii):

1209: $$

1210: \sum_{k=0}^n{1\over W^{1/p}(k)}\le C\,{n+1\over W^{1/p}(n)},\quad n=0,1,2,\dots

1211: $$

1212: It is enough to discretize the  integral and to note that

1213: $$

1214: \int_n^{n+1}{1\over\widetilde{W}^{1/p}(t)}\,dt\approx{1\over W^{1/p}(n)},\quad n=0,1,2,\dots\qquad\qed

1215: $$

1216: \edem

1217:

1218:

1219: The characterization of the  boundedness  (\ref{I.6.8}) can be seen very easily and it is

1220: equivalent to the condition (b)(iii) of the previous  theorem.

1221:

1222: \bteo

1223: \label{te: I.6.16}

1224: For $0<p<\infty\/$ we have that

1225: $$

1226: A_d\,:\,\ell^{p,\infty}_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w)

1227: $$

1228: if and only if

1229: $$

1230: \sum_{k=0}^n{1\over W^{1/p}(k)}\le C\,{n+1\over W^{1/p}(n)},\quad n=0,1,2,\dots

1231: $$

1232: \eteo

1233:

1234: \bdem      The boundedness  in the statement is  equivalent to the inequality

1235: \bequ

1236: \label{I.6.17}

1237: \|A_d f\|_{\ell^{p,\infty}(w)}\le C,\qquad f\d,\;\|f\|_{\ell^{p,\infty}(w)}\le1.

1238: \eequ

1239: Now, if  $f\/$ is decreasing $\|f\|_{\ell^{p,\infty}(w)}=\sup_n W^{1/p}(n)f(n)\/$ (cf.\

1240: (\ref{I.6.10})) and $\|f\|_{\ell^{p,\infty}(w)}\le1\/$ implies $f(n)\le W^{-1/p}(n),$ for every $ n\/$.

1241: On the other hand the sequence  $W^{-1/p}\/$ is decreasing and with norm  1. Therefore

1242: $f=W^{-1/p}\/$ is the sequence that attains the maximum in the left hand side of  (\ref{I.6.17}) and the

1243: characterization will be

1244: $$

1245: \|A_d W^{-1/p}\|_{\ell^{p,\infty}(w)}=C<\infty.

1246: $$

1247: Taking into account that  $A_d W^{-1/p}\/$ is decreasing  and  (\ref{I.6.10}) for the norm in

1248: $\ell^{p,\infty}(w)\/$ of such sequences,  we obtain the condition of the statement. $\qquad\qed$

1249: \edem

1250:

1251: \bcor

1252: \label{cor: I.6.18}

1253: If $1<p<\infty\/$ the following conditions are equivalent:

1254: \begin{enumerate}

1255: \item[(i)] $ \tilde{w}=\sum_{n=0}^\infty w(n)\chi_{[n,n+1)}\in B_p,$

1256: \item[(ii)] $ \sum_{k=0}^n{1\over W^{1/p}(k)}\le C\,{n+1\over W^{1/p}(n)},\quad n=0,1,2,\dots,$

1257: \item[(iii)] $ A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w),$

1258: \item[(iv)] $ A_d\,:\,\ell^{p,\infty}_{\hbox{\decp dec}}(w)\longrightarrow \ell^{p,\infty}(w),$

1259: \item[(v)] $ A_d\,:\,\ell^p_{\hbox{\decp dec}}(w)\longrightarrow \ell^p(w).$

1260: \end{enumerate}

1261: \ecor

1262:

1263: \bdem      The first four equivalences are a consequence of the two previous theorems. On the other

1264: hand, (v) implies (iii) and (i) implies

1265: $$

1266: A\,:\,L^p_{\hbox{\decp dec}}(\tilde{w})\longrightarrow L^p(\tilde{w}).

1267: $$

1268: It is immediate to check that this  boundedness  of the  Hardy operator in  $\Rm\/$ implies the

1269: corresponding boundedness for the discrete  Hardy operator

1270: $A_d\/$, that is (v). $\qquad\qed$

1271: \edem

1272:

1273: \chapter{Lorentz Spaces}

1274: \label{ch:lor}

1275: \pagestyle{myheadings}

1276: \markboth{CHAPTER 2. LORENTZ SPACES}{M.J.\ Carro, J.A.\ Raposo, and J.\ Soria}

1277:

1278: \section{Introduction}

1279: \label{se:introlor}

1280:

1281: If $(X,\mu)$ is a measure space, $w$ is a weight in $\R^+$ (see Definition \ref{defgen} below) and

1282: $0<p<\infty$, the Lorentz space\index{Lorentz space} $\Lambda_X^p(w)$\index{$\Lambda_X^p(w)$} is

1283: defined as the class of all measurable functions $f$ in $X$ for which

1284: \bequ

1285: \label{normx}

1286: \Vert f \Vert_{\Lambda_X^p(w)}\buildrel{\hbox{\decp def.}}\over=\bigg(\int_0^{\infty}

1287: (f^*(t))^p\,w(t)\,dt\bigg)^{1/p}<\infty,

1288: \eequ

1289: where $f^*$ is the nonincreasing rearrangement\index{nonincreasing rearrangement} of

1290: $f$ with respect to $\mu$. These spaces were first introduced by G.G.\ Lorentz in

1291: \cite{Lor} for the case $X=(0,1)$. By choosing $w$ properly, one obtains the

1292: spaces $L^{p,q}$\index{$L^{p,q}$} defined in (\ref{normlor}). As we shall see, we also have a

1293: weak-type version denoted by

1294: $\Lambda^{p,\infty}_X(w)$\index{$\Lambda^{p,\infty}_X(w)$}.

1295:

1296: In this chapter we are going to consider the study of analytical properties of these

1297: function spaces, giving a complete description of some previously known partial

1298: results. Already in the work of Lorentz (\cite{Lor}) there exists a characterization

1299: of when (\ref{normx}) defines a norm. Later, A. Haaker (\cite{Haa})

1300: extends this result to the case of $X=\R^+$ and $w\notin L^1$, and considers also

1301: the existence of a dual space. E. Sawyer (\cite{Saw2}) gives

1302: several equivalent conditions on $w$ so that $\Lambda_{\R^n}^p(w)$ is a Banach

1303: space, when $p>1$, and M.J. Carro, A. Garc\'{\i}a del Amo, and J. Soria

1304: (\cite{CGS}) study the normability in the case $p\ge1$ and $X$ is a nonatomic

1305: space. For these same conditions, J. Soria (\cite{Sor}) characterizes the

1306: normability of the weak-type Lorentz spaces\index{Lorentz space!weak-type}. We

1307: present a throughout account of all these properties in the general setting of

1308: resonant spaces, allowing us to also consider the discrete

1309: case $X=\N$ and the sequence spaces $\ell^p$.  These discrete spaces (which are also

1310: known as $d(w,p)$\index{$d(w,p)$}) had been previously studied for decreasing

1311: weights $w$ (and hence $\Vert\cdot\Vert_{\Lambda^p(w)}$ is always a norm if

1312: $p\ge 1$). Here we will consider general weights and also the weak-type spaces

1313: $d^{\infty}(w,p)$\index{$d^{\infty}(w,p)$}.

1314:

1315: This chapter is divided into several sections. After the definition and first

1316: properties of section~\ref{se:def} we consider in section~\ref{se:cuasi}

1317: the quasi-normed Lorentz spaces (for which only minimal assumptions on the

1318: monotonicity of the primitive of $w$ are required). Density properties for simple

1319: functions and absolute continuity of the quasi-norm are also considered.

1320: Section~\ref{se:dual} is the largest and we study the duality results. In

1321: particular we characterize  when $\Lambda^*=\Lambda'$ and when these are the

1322: trivial space. In section~\ref{se:norma} we give necessary and sufficient

1323: conditions to have that the Lorentz spaces are Banach spaces. Finally in

1324: section~\ref{se:oper} we study interpolation properties and the boundedness of

1325: certain operators.

1326:

1327: We fix now the main definitions and notations used in what follows.

1328: \bdefi

1329: \label{defgen}

1330: We denote by $\R^+=(0,\infty)$. The letters $w,\ \tilde w,\ w_0,...,$ are used for

1331: weight\index{weight} functions in $\R^+$ (nonnegative locally integrable functions

1332: in $\R^+$). For a given weight $w$ we write $W(r)=\displaystyle\int_0^rw(t)\,dt<\infty$, $0\le r<\infty$.

1333:

1334: If $(X,\mu)$ is a measure space,  $f\in {\cal M}(X)$, we denote

1335: the distribution function\index{distribution function} and the nonincreasing

1336: rearrangement\index{nonincreasing rearrangement} of

1337: $f$ as

1338:  $\lambda_f(t)$ and $f^*(t)$, and

1339: if the measure $d\mu(t)=w(t)\,dt$, we will  write $\lambda_f^w$ and $f^*_w$ to

1340: show the dependence on the weight $w$.

1341:

1342:  If

1343: $(X,\mu)=(\R^m,w(t)dt)\/$ we write $L^{p,q}(w)$\index{$L^{p,q}(w)$} instead of $L^{p,q}(X)\/$.

1344: \edefi

1345:

1346:

1347:

1348: \section{$\Lambda_X^p(w)$ spaces}

1349: \label{se:def}

1350:

1351: In this section $(X,\mu)\/$ denotes, except if otherwise mentioned, a general measure space.

1352:

1353: \bdefi\label{def:norm}

1354: Let $w\/$ be a weight in $\R^+$. For $0<p<\infty\/$ we define the functional

1355: $\|\cdot\|_\loX:\M(X)\rightarrow[0,\infty]\/$ as

1356: $$\|f\|_\loX=\bigg(\int_0^\infty (f^\ast(t))^p w(t)\,dt\bigg)^{1/p},\quad f\in \M(X).$$ The Lorentz

1357: space\index{$\Lambda^p(w)$}

1358: $\Lambda^p(w)=\loX\/$ is the class

1359: $$\loX=\big\{f\in \M(X)\,:\,\|f\|_\loX<\infty\big\}.$$ Observe that $\|f\|_\loX=\|f^\ast\|_{L^p(w)}\/$. This

1360: allows us to extend the previous definition. For

1361: $0<p,q\le\infty\/$ set

1362: $$\lox{p,q}{X}{w}\index{$\lox{p,q}{X}{w}$}=\big\{f\in

1363: \M(X)\,:\,\|f\|_{\lox{p,q}{X}{w}}=\|f^\ast\|_{L^{p,q}(w)}<\infty\big\}.$$

1364: \edefi

1365:

1366: From now on, the notation $\loX\/$ or $\lox{p,q}{X}{w}\/$ without any reference to $w$, means that $w\/$ is a

1367: weight in $\R^+$ not identically zero on $\big(0,\mu(X)\big)\/$. The symbol $\Lambda\/$ will denote any of the

1368: spaces previously defined. Sometimes we will write $\Lambda^p\/$ as a shorthand for $\Lambda^p_X(w)\/$ if

1369: $X\/$ and

1370: $w\/$ are clearly determined. Similarly $\Lambda(w),\Lambda_X,\Lambda^p(w)\/$, etc.

1371: Observe that

1372: $\Lambda^{\infty,q}=\{0\}\/$ if $0<q<\infty\/$.

1373:

1374: These spaces were introduced by Lorentz in \cite{Lor2} and \cite{Lor} for the case $X=(0,l)\subset\Rm\/$.

1375: They are invariant under rearrangement\index{rearrangement invariant} and generalize the  $L^p\/$

1376: Lebesgue spaces and

1377: $L^{p,q}\/$ (see Examples~\ref{Eje2.3}(i) and \ref{Eje2.3}(ii)). The spaces $\Lambda^p(w)\/$ we have defined

1378: are usually called the \lq\lq classical\rq\rq\ Lorentz spaces  to distinguish them from $L^{p,q}\/$.

1379:

1380: \bobs\label{Obs2.2}

1381:

1382: (i) $f^\ast(t)=0\/$ if $t\ge\mu(X)\/$. Hence, the behavior of the weight $w\/$ in

1383: $\big[\mu(X),\infty\big)\/$ is irrelevant. We will assume, without loss of generality, that the weight $w\/$

1384: vanishes in

1385: $\big[\mu(X),\infty\big)\/$ if $\mu(X)<\infty\/$. Observe that, in this way, we have that  $w\in

1386: L^1(\Rm)\/$ if

1387: $\mu(X)<\infty\/$.

1388:

1389: (ii) If $w\notin L^1(\Rm)\/$ then $\lim_{t\to\infty}f^\ast(t)=0\/$ if $f\in\Lambda^{p,q}(w)\/$

1390: ($p<\infty\/$).

1391:

1392: (iii) Simple functions with finite support are in $\Lambda^{p,q}(w)\/$. If $w\in L^1\/$ then

1393: $L^\infty\subset\Lambda^{p,q}(w)\/$ and every simple function is in  $\Lambda^{p,q}(w)\/$.

1394:

1395: (iv) Observe that $\lox{p,p}{}{w}=\Lambda^p(w),\,0<p\le\infty\/$.

1396: \eobs

1397:

1398: \beje\label{Eje2.3}

1399:

1400: (i) In the case $w=1\/$ we have, for $0<p,q\le\infty,\;\Lambda^{p,q}_X(1)=L^{p,q}(X)\/$. Here

1401: $W(t)=t,\;t\ge0\/$.

1402:

1403: (ii) If $0<p,q<\infty,\;\Lambda^q_X(t^{q/p-1})=L^{p,q}(X)\/$ with  equality of \lq\lq norms\rq\rq. In this

1404: case

1405: $W(t)={p\over q}t^{q/p},\;t\ge0\/$.

1406:

1407: (iii) If $w=\chi_{(0,1)}\/$, the space $\Lambda^1_X(w)=\Lambda^1_X\big(\chi_{(0,1)}\big)\/$ contains

1408: $L^\infty(X)\/$ and $W(t)=t,\;0\le t<1\/$. In this case the functional $\|\cdot\|_{\Lambda^1}\/$ is a norm

1409: and the space is a

1410: \lq\lq  Banach function space" (see Definition~\ref{def4.3}).

1411:

1412: (iv) If $X=\N^\ast=\{0,1,2,\dots\}\/$ and  we consider the counting measure, then measurable functions in

1413: $X\/$ are sequences

1414: $f=\big(f(n)\big)_{n\ge0}\subset\C\;$ and

1415: $$

1416: \|f\|_{\Lambda_X^p(w)}=\bigg(\sum_{n=0}^\infty \big(f^\ast(n)\big)^p\Omega_n\bigg)^{1/p},

1417: $$

1418: where for each

1419: $n=0,1,2,\dots,\;\Omega_n=\displaystyle\int_n^{n+1} w(s)\,ds=W(n+1)-W(n)\/$. Thus, $\Lambda_X^p(w)\/$

1420: depends only on the sequence of positive numbers $\Omega=(\Omega_n)_{n=0}^\infty\/$

1421:  and it is usually denoted as

1422: $d(\Omega,p)\/$. In the weak-type case,

1423: $$

1424: \|f\|_{\Lambda_X^{p,\infty}(w)}=\sup_{n\ge0}\bigg(\sum_{k=0}^n\Omega_k\bigg)^{1/p}f^\ast(n),

1425: $$

1426: and we will use the symbol $d^\infty(\Omega,p)\/$ to denote

1427: $\Lambda_X^{p,\infty}(w)\/$. Given $d(\Omega,p)\/$ or $d^\infty(\Omega,p)\/$

1428: we will assume that $\Omega\/$ is always a sequence of positive numbers.

1429:

1430: It is clear that both $d(\Omega,p)\/$ and $d^\infty(\Omega,p)\/$ are always contained in

1431: $\ell^\infty(\N^\ast)\/$ and if

1432: $\Omega\in \ell^1(\N^\ast)\/$ we have in fact that $d(\Omega,p)=d^\infty(\Omega,p)

1433: =\ell^\infty(\N^\ast)\/$ (with

1434: equivalent norms). The only interesting case is hence  $\Omega\notin\ell^1(\N^\ast)\/$.

1435: In this case, the

1436: space is contained in

1437: $c_0(\N^\ast)\/$, the space of sequences that converge to $0\/$ and, for each $f\in

1438: d(\Omega,p)\/$ ($f\in

1439: d^\infty(\Omega,p)\/$),

1440: $\big(f^\ast(n)\big)_{n\ge0}\/$ is the nonincreasing rearrangement of the sequence

1441: $\big(f(n)\big)_{n\ge0}\/$.

1442:

1443: If the sequence $\Omega=(\Omega_n)_n\/$ of the previous example is decreasing and $p\ge1\/$,

1444:  $d(\Omega,p)\/$

1445: is a Banach space. This case has been studied by several authors (\cite{Ga}, \cite{CH},

1446: \cite{Al}, \cite{Po},

1447: \cite{NO}, \cite{AM2}, etc.). The case  $\Omega\/$ increasing is considered in \cite{AEP}.

1448:  We will assume no a

1449: priori conditions on $\Omega$.

1450: \eeje

1451:

1452: The following lemma (see \cite{CS1}) will be very useful.

1453:

1454: \blem

1455: \label{le: II.2.4}

1456:  Let $0<p<\infty\/$ and $v\ge0\/$ be a measurable function in $\R^+\/$. Let $V(r)=\displaystyle\int_0^r

1457: v(s)\,ds,\,0\le r<\infty\/$. Then, for every decreasing function $f\/$ we have that

1458: $$

1459: \int_0^\infty f^p(s) v(s)\,ds=\int_0^\infty pt^{p-1}V\big(\lambda_f(t)\big)\,dt.

1460: $$

1461: \elem

1462:

1463:

1464:

1465: The following result gives several equivalent expressions for the functional

1466:  $\|\cdot\|_{\lox{p,q}{X}{w}}$.  In particular, we see that it only depends on  $W\/$.

1467:

1468: \bpro

1469: \label{pro: II.2.5}

1470: For $0<p,q<\infty\/$ and  $f\/$ measurable in $X\/$,

1471: \begin{enumerate}

1472: \item[(i)] $\|f\|_\lox{p,q}{X}{w}=\bigg(\displaystyle\int_0^\infty pt^{q-1}W^{q/p}

1473: \big(\lambda_f(t)\big)\,dt\bigg)^{1/q},$

1474: \item[(ii)] $\|f\|_\lox{p}{X}{w}=\bigg(\displaystyle\int_0^\infty pt^{p-1}W\big(\lambda_f(t)\big)\,dt

1475: \bigg)^{1/p},$

1476: \item[(iii)] $\|f\|_\loinx{p}{X}{w}=\sup_{t>0}tW^{1/p}\big(\lambda_f(t)\big)=\sup_{t>0}

1477: f^\ast(t)W^{1/p}(t).$

1478: \end{enumerate}

1479: \epro

1480:

1481: \bdem

1482:

1483: (i) Since every function and its decreasing rearrangement have the same distribution function (see

1484: \cite{BS}) we have that,

1485: $$

1486: W\big(\lambda_f(t)\big)=W\big(\lambda_{f^\ast}(t)\big)=\int_0^{\lambda_{f^\ast}(t)}w(s)\,ds

1487: =\int_{\{f^\ast>t\}}w(s)\,ds=\lambda^w_{f^\ast}(t).

1488: $$

1489: By Lemma~\ref{le: II.2.4} and since  $\lambda^w_{f^\ast}\/=(f^\ast)^\ast_w\/$,

1490: we obtain that

1491: \begin{eqnarray*}

1492: \bigg(\int_0^\infty pt^{q-1}W^{q/p}\big(\lambda_f(t)\big)\,dt\bigg)^{1/q}&=&\bigg(\int_0^\infty

1493: pt^{q-1}\big(\lambda^w_{f^\ast}(t)\big)^{q/p}\,dt\bigg)^{1/q}\\

1494: &=&\bigg(\int_0^\infty t^{q/p-1}\int_0^{(f^\ast)^\ast_w(t)}qs^{q-1}\,ds\,dt\bigg)^{1/q}\\ &

1495: =&\bigg(\int_0^\infty\big((f^\ast)^\ast_w(t)\big)^q

1496: t^{q/p-1}\,dt\bigg)^{1/q}\\ &=&\|f\|_{\lox{p,q}{X}{w}}.

1497: \end{eqnarray*}

1498:

1499: (ii) It is an immediate consequence of  (i).

1500:

1501: (iii) For the first inequality we observe that

1502: $$

1503: \|f\|_{\loinX}=\|f^\ast\|_{L^{p,\infty}(w)}=\sup_{t>0}t\big(\lambda^w_{f^\ast}(t)\big)^{1/p}

1504: =\sup_{t>0}tW^{1/p}\big(\lambda_f(t)\big).

1505: $$

1506: The second inequality for characteristic functions is trivial and, by monotonicity, the

1507: general case follows.$\qquad\qed$

1508: \edem

1509:

1510: \bobs

1511: \label{obs: II.2.6}

1512:

1513: (i)  If we compare  \ref{pro: II.2.5}   (i) and  \ref{pro: II.2.5} (ii) we see that, for

1514: $q<\infty\/$,

1515: $\|f\|_{\lox{p,q}{X}{w}}=\|f\|_\lox{q}{X}{\tilde{w}}\/$ where

1516: $\tilde{w}(t)=W^{q/p-1}(t)w(t),\;0<t<\mu(X)\/$. Therefore, every  Lorentz space as defined here

1517: reduces to

1518: $\loX\/$ and its weak version $\loinX\/$.

1519:

1520: (ii) From \ref{pro: II.2.5} (iii), we deduce that  $\lox{p,\infty}{X}{w}=

1521: \Lambda^{q,\infty}_X\big((q/p)\tilde{w}\big)\/$ for $0<p,q<\infty\/$.

1522: \eobs

1523:

1524:

1525:

1526: For the spaces $L^{p,\infty}(X)\/$ it is known that the quasi-norm $\|f\|_{p,\infty}\/$

1527: is, for every $q<p\/$, equivalent to the functional

1528: $$

1529: \sup_{E\subset X}\big\|f\chi_E\big\|_q\,\mu(E)^{1/p-1/q}.

1530: $$ This is the so-called  Kolmogorov condition

1531: (see e.g.\ \cite{GR}). An analogous version for

1532: $\Lambda^{p,\infty}(w)\/$ also holds.

1533:

1534: \bpro

1535:  If $0<q<p<\infty\/$ and $f\in\M(X)\/$,

1536: $$

1537: \|f\|_{\Lambda^{p,\infty}_X(w)}\le\sup_{E\subset X}\big\|f\chi_E\big\|_{\Lambda^q_X(w)}W

1538: \big(\mu(E)\big)^{1/p-1/q}\le\bigg({p\over

1539: p-q}\bigg)^{1/q}\,\|f\|_{\Lambda^{p,\infty}_X(w)},

1540: $$

1541: where the supremum is taken over all measurable sets $E\subset X\/$.

1542: \epro

1543:

1544: \bdem

1545: Let

1546: $$

1547: S=\sup_{E\subset X}\big\|f\chi_E\big\|_{\Lambda^q_X(w)}W\big(\mu(E)\big)^{1/p-1/q}.

1548: $$

1549: To show the first inequality,  let  $E=\{|f|>t\}\/$, $t>0\/$.

1550: Then,

1551: \begin{eqnarray*}

1552: S&\ge&\big\|f\chi_E\big\|_{\Lambda^q_X(w)}W\big(\mu(E)\big)^{1/p-1/q}\\ &

1553: =&\bigg(\int_0^\infty\big(\big(f\chi_E\big)^\ast(s)\big)^q

1554: w(s)\,ds\bigg)^{1/q}W\big(\mu(E)\big)^{1/p-1/q}\\

1555: &\ge&\bigg(t^q\int_0^{\mu(E)} w(s)\,ds\bigg)^{1/q}

1556: W\big(\mu(E)\big)^{1/p-1/q}\\

1557: &=&t\,W\big(\mu(E)\big)^{1/p}=t\,W\big(\lambda_f(t)\big)^{1/p},

1558: \end{eqnarray*}

1559: and taking the supremum in $t>0\/$

1560: we get $\|f\|_{\Lambda^{p,\infty}_X(w)}\le S\/$ (c.f.

1561: \ref {pro: II.2.5} (iii)).

1562:

1563: To prove the second inequality, for each  $f\in\M(X),\,E\subset X\/$, let $a=

1564: \|f\|_{\Lambda^{p,\infty}_X(w)}W\big(\mu(E)\big)^{-1/p}\/$. Then, by \ref{pro: II.2.5} (iii),

1565: \begin{eqnarray*}

1566: \big\|f\chi_E\big\|^q_{\Lambda^q_X(w)}&=&\int_0^\infty qt^{q-1}W\big(\lambda_{f\chi_E}(t)\big)\,dt

1567: \\ &=&\int_0^a

1568: qt^{q-1}W\big(\lambda_{f\chi_E}(t)\big)\,dt+\int_a^\infty qt^{q-1}W\big(\lambda_{f\chi_E}(t)

1569: \big)\,dt\\ &\le &W\big(\mu(E)\big)\int_0^a qt^{q-1}\,dt+\int_a^\infty q

1570: t^{q-1}{\|f\|^p_{\Lambda^{p,\infty}_X(w)}\over t^p}\,dt\\ &=&{p\over p-q}\,

1571: \|f\|^q_{\Lambda^{p,\infty}_X(w)}W\big(\mu(E)\big)^{1-q/p}.

1572: \end{eqnarray*}

1573:  Hence,

1574: $$

1575: \big\|f\chi_E\big\|_{\Lambda^q_X(w)}W\big(\mu(E)\big)^{1/p-1/q}\le\bigg({p\over p-q}\bigg)^{1/q}

1576: \,\|f\|_{\Lambda^{p,\infty}_X(w)}.\quad\qed

1577: $$

1578: \edem

1579:

1580: In the following proposition we state some elementary properties for these spaces (see \cite{BS}).

1581:

1582: \bpro

1583: \label{pro: II.2.8}

1584:  For $0<p,q\le\infty\/$ and $f,g,f_k,\;k\ge1,\/$ measurable functions in  $X\/$, we have that:

1585: \begin{enumerate}

1586: \item[(i)]  $|f|\le|g|\Rightarrow\|f\|_{\lox{p,q}{X}{w}}\le\|g\|_{\lox{p,q}{X}{w}}\/$,

1587:

1588: \item[(ii)] $\|tf\|_{\lox{p,q}{X}{w}}=|t|\|f\|_{\lox{p,q}{X}{w}},\quad t\in\C\/$,

1589:

1590: \item[(iii)]  $0\le f_k\le f_{k+1}\limto{k}f\/$ a.e. $\Rightarrow\|f_k\|_{\lox{p,q}{X}{w}}

1591: \limto{k}\|f\|_{\lox{p,q}{X}{w}}\/$,

1592:

1593: \item[(iv)]  $\big\|\liminf_k|f_k|\big\|_{\lox{p,q}{X}{w}}\le\liminf_k\|f_k\|_{\lox{p,q}{X}{w}}\/$,

1594:

1595: \item[(v)]  $\lox{p,q_0}{X}{w}\subset\lox{p,q_1}{X}{w}$ continuously, $0<q_0\le q_1\le\infty\/$,

1596:

1597: \item[(vi)]  If\ $W\big(\mu(X)\big)<\infty\/$ then, for $0<p_0<p_1\le\infty\/$,

1598: we have that  $\lox{p_1,q}{X}{w}\subset\lox{p_0,r}{X}{w},\;0<r\le\infty\/$ continuously,

1599:

1600: \item[(vii)]  $\chi_E\in\lox{p,q}{X}{w}\/$ if $\mu(E)<\infty\/$.

1601: \end{enumerate}

1602: \epro

1603:

1604: The following property connects the norm convergence $\|\cdot\|_{\lox{p,q}{X}{w}}\/$

1605: with the convergence  in measure  and it is related to the  \lq\lq completeness\rq\rq\ property of

1606: our spaces.

1607:

1608: \bpro

1609: \label{pro: II.2.9}

1610: Assume that  $W>0\/$ in $(0,\infty)\/$, let  $\Lambda=\lox{p,q}{X}{w}\/$ be

1611: a Lorentz space and  let $(f_n)_n\/$ be a sequence of measurable functions in  $X\/$.

1612: \begin{enumerate}

1613: \item[(i)] If   $\lim_{m,n}\|f_m-f_n\|_{\Lambda}=0\/$ then $(f_n)_n\/$  is a Cauchy sequence in measure

1614: and there exists

1615: $f\in

1616: \M(X)\/$ such that

1617: $\lim\|f-f_n\|_{\Lambda}=0\/$.

1618:

1619: \item[(ii)] If  $f\in \M(X)\/$ and  $\lim_{n}\|f-f_n\|_{\Lambda}=0\/$ then $(f_n)_n\/$

1620: converges to $f\/$ in measure and there exists a partial $(f_{n_k})_k\/$ convergent to

1621: $f\/$ a.e.

1622: \end{enumerate}

1623: \epro

1624:

1625: \bdem

1626: The case  $q<p=\infty\/$ is  trivial and in  the case $p=q=\infty,\;\Lambda^{p,q}

1627: =L^\infty\/$ and the result is already known. If $p<\infty\/$ it is immediate,

1628: by Proposition~\ref{pro: II.2.5}, that

1629: $$

1630: W\big(\lambda_f(r)\big)\le C_{p,q}{\|f\|_{\lox{p,q}{X}{w}}^p\over r^p},\quad r>0,

1631: \quad0<q\le\infty.

1632: $$

1633: Using  the hypothesis of  (i) we obtain then that

1634: $W\big(\lambda_{f_n-f_m}(r)\big)\limto{m,n}0\/$,   for every $r>0\/$, which (since $W>0\/$) implies

1635: $\lambda_{f_n-f_m}(r)\limto{m,n}0,\;r>0\/$, that is, $(f_n)_n\/$ is a  Cauchy sequence

1636: in measure. We know  that this implies the convergence in measure of $(f_n)_n\/$

1637: to some measurable function $f\/$ and the existence of a partial $(f_{n_k})_k\/$ converging to $f\/$

1638: a.e. By Proposition~\ref{pro: II.2.8} (iv) we have that  $\|f-f_n\|_\Lambda

1639: \le\liminf_{k}\|f_{n_k}-f_n\|_\Lambda\/$ and, thus,

1640: $\lim_n\|f-f_n\|_{\Lambda}=0\/$.

1641:

1642: The proof of  (ii) is analogue.$\qquad\qed$

1643: \edem

1644:

1645:

1646: The functional $\|\cdot\|_{\Lambda}\/$ is not, in general, a quasi-norm and, in fact,

1647: $\Lambda\/$ could not even be a vector space. The following lemma characterizes the

1648: quasi-normability of these spaces, which, as we will see, only depends on  the

1649: weight $w\/$ and on the measure space $X\/$.

1650:

1651: \blem

1652: \label{le: II.2.10}

1653:  If $0<p<\infty$ and $0<q\le\infty\/$, the space $\Lambda^{p,q}_X(w)\/$

1654: is quasi-normed if and only if

1655: \bequ\label{delta2}

1656: 0<W\big(\mu(A\cup B)\big)\le C\big( W\big(\mu(A)\big)+W\big(\mu(B)\big) \big),

1657: \eequ for every pair of measurable sets $A,B\subset X\/$ with $\mu(A\cup B)>0\/$.

1658: \elem

1659:

1660: \bdem

1661: Sufficiency:  The hypothesis implies that  $W\big(\mu(A)\big)>0\/$ if $\mu(A)>0\/$.

1662: If $\|f\|_{\Lambda^{p,q}}=0\/$, by

1663: Proposition~\ref{pro: II.2.5}, we have that

1664: $W\big(\lambda_f(t)\big)=0,\;t>0,\/$ and hence, $\lambda_f(t)=0\/$ for every $t>0\/$,

1665: that is $f=0\/$ a.e. It remains to show the quasi-triangular inequality and it   suffices to prove

1666: it for positive functions.  Let   $0\le

1667: f,g\in\Lambda^{p,q}(w)\/$  and $t>0\/$. Then $\{f+g>t\}\subset\{f>t/2\}\cup\{g>t/2\}\/$  and by

1668: hypothesis

1669: $$

1670: W\big(\lambda_{f+g}(t)\big)\le C\big(W\big(\lambda_f(t/2)\big)+W\big(\lambda_g(t/2)\big)\big).

1671: $$

1672: Since $C\/$ does not depend on  $t\/$, it satisfies that (Proposition~\ref{pro: II.2.5})

1673: $$

1674: \|f+g\|_{\Lambda^{p,q}(w)}\le C_{p,q}\big(\|f\|_{\Lambda^{p,q}(w)}+\|g\|_{\Lambda^{p,q}(w)}

1675: \big).

1676: $$

1677: Necessity: If $A,B\/$ are two measurable sets with

1678: $\mu(A\cup B)>0,\;\chi_{A\cup B}\le\chi_A+\chi_B\/$ and since  $\Lambda^{p,q}(w)\/$

1679: is quasi-normed, we have that (Proposition~\ref{pro: II.2.5}),

1680: \begin{eqnarray*}

1681:  0<C_{p,q}W^{1/p}\big(\mu(A\cup B)\big)&=&\|\chi_{A\cup B}\|_{\Lambda^{p,q}(w)}\\

1682: &\le&\|\chi_A+\chi_B\|_{\Lambda^{p,q}(w)}\\ &\le&

1683: C\big(\|\chi_A\|_{\Lambda^{p,q}(w)}+\|\chi_B\|_{\Lambda^{p,q}(w)}\big)\\ &=&C^\prime_{p,q}

1684: \big(W^{1/p}\big(\mu(A)\big)+W^{1/p}\big(\mu(B)\big)\big)

1685: \end{eqnarray*}

1686:  which is equivalent to the condition of the statement.$\quad\qed$

1687: \edem

1688:

1689: The previous result motivates the following definition.

1690:

1691: \bdefi

1692:  Let $w\/$ be a weight in $\Rm\/$. We write $W\in\Delta_2(X)\/$\index{$\Delta_2(X)$}

1693: (or $W\in\Delta_2(\mu)\/$)\index{$\Delta_2(\mu)$} if $W\/$ satisfies (\ref{delta2}). If

1694: $X=\R\/$ we shall simply write $W\in\Delta_2\/$\index{$\Delta_2$}.

1695: \edefi

1696:

1697: Therefore, Lemma~\ref{le: II.2.10} tells us that, for every  $0<p<\infty,\;0<q\le\infty\/$,

1698: $$

1699: \Lambda^{p,q}_X(w)\hbox{\rm is quasi-normed\ }\Leftrightarrow W\in\Delta_2(X).

1700: $$

1701:

1702:

1703: \bpro

1704: \label{pro: II.2.12}

1705: The following conditions are equivalent:

1706: \begin{enumerate}

1707: \item[(i)] $W\in\Delta_2\/$,

1708:

1709: \item[(ii)] $W(2r)\le CW(r),\;r>0\/$,

1710:

1711: \item[(iii)] $W(t+s)\le C\big(W(t)+W(s)),\;t,s>0\/$,

1712: \end{enumerate}

1713:

1714: \noindent and in any of these cases, $W(t)>0,\;t>0\/$.

1715: \epro

1716:

1717: This proposition (whose proof is trivial) shows that $W\in\Delta_2$ is something easy to check. The condition

1718: $W\in\Delta_2\/$ is sufficient to have the quasi-normability of

1719: $\Lambda^{p,q}_X(w)\/$,  independent of the measure space $X\/$ and  this is the content of

1720: the following theorem. The second part was proved in \cite{CS1} and the proof of (i) is immediate.

1721:

1722: \bteo

1723: \label{te: II.2.13}

1724: Let $0<p<\infty,\,0<q\le\infty\/$.

1725: \begin{enumerate}

1726: \item[(i)] If $W\in\Delta_2,\;\Lambda^{p,q}(w)\/$ is quasi-normed.

1727:

1728: \item[(ii)]  If $X\/$ is nonatomic, $\lox{p,q}{X}{w}\/$ is quasi-normed if and only if

1729: $W\in\Delta_2\/$.

1730: \end{enumerate}

1731: \noindent That is,  $\Delta_2\subset\Delta_2(X)\/$ for every $X\/$ and if  $X\/$

1732: is nonatomic, $\Delta_2=\Delta_2(X)\/$ (we are  assuming that  $w\/$ is zero outside of

1733: $\big(0,\mu(X)\big))\/$.

1734:

1735: \eteo

1736:

1737: \bobs

1738:

1739: For the case $p=q=\infty\/$, it  is obvious that the following properties are equivalent:

1740: \begin{enumerate}

1741: \item[(i)] $\Lambda^\infty_X(w)$  is quasi-normed.

1742: \item[(ii)] $\Lambda^\infty_X(w) =L^\infty(X)$ (with equality of norms).

1743: \item[(iii)] $\big(\Lambda^\infty_X(w),\|\cdot\|_{\Lambda^\infty_X(w)}\big)$

1744: is a Banach space.

1745: \item[(iv)] $ W\big(\mu(A)\big)>0$ if $\mu(A)>0,\;A\subset X.$

1746: \end{enumerate}

1747: Any of these conditions hold if  $W\in\Delta_2(X)\/$.

1748: \eobs

1749:

1750:

1751:

1752: Let us recall the following definition that we shall use quite often in what follows.

1753:

1754: \bdefi

1755: \label{def: II.2.15}

1756: A resonant measure space\index{resonant space} $X\/$  (\cite{BS})

1757: is a $\sigma$-finite space  which is nonatomic or it is a union (at most  countable)  of atoms

1758: with equal measure.

1759: \edefi

1760:

1761: If $X\/$ is nonatomic we  have already characterized (Theorem~\ref{te: II.2.13})

1762: when $\Lambda_X\/$ is quasi-normed. The following theorem completes the characterization for the

1763: case

1764: $X\/$ is a resonant measure space. Its proof is immediate.

1765:

1766: \bteo

1767: \label{te: II.2.16}

1768: Let $X\/$ be an atomic measure space whose atoms have  measure  $b>0\/$.

1769: Then $W\in\Delta_2(X)\/$ if and only if

1770: $$

1771: W\big(2nb)\le C\,W(nb),\qquad n\ge1.

1772: $$

1773: In particular, if $0<p<\infty\/$  the spaces  $d(\Omega,p)\/$ and $d^\infty(\Omega,p)\/$

1774: (Examples~\ref{Eje2.3}(iv)) are quasi-normed if and only if

1775: $$

1776: \sum_{n=1}^{2N}\Omega_{n-1}\le C\sum_{n=1}^N\Omega_{n-1},\quad N=1,2,\dots

1777: $$

1778: \eteo

1779:

1780:

1781: \section{Quasi-normed Lorentz spaces}

1782: \label{se:cuasi}

1783:

1784:

1785: In this section $(X,\mu)\/$ will be   an arbitrary

1786: $\sigma$-finite measure space and the  Lorentz spaces will be defined on the measure space   $X\/$.

1787: We shall study the topology and some elementary properties of the quasi-normed

1788: Lorentz spaces. As we know,  the condition to have that

1789: $\Lambda_X(w)\/$ is quasi-normed is  $W\in\Delta_2(X)\/$ (Lemma~\ref{le: II.2.10}).

1790:

1791: As in every quasi-normed space, we consider in  $\Lambda\/$

1792: the topology induced by the quasi-norm $\|\cdot\|_\Lambda\,:\;G\subset\Lambda\/$ is an open set

1793: if for every function

1794: $f\in G\/$ there exists

1795: $r>0\/$ such that  $\big\{g\in\Lambda\,:

1796: \,\|f-g\|_\Lambda<r\big\}\subset G\/$. We know that there exists in $\Lambda\/$ a translation

1797: invariant distance  $d\/$

1798:   and an exponent  $p\in(0,1]\/$ so that

1799: \bequ

1800: \label{II.3.1}

1801: d(f,g)\le\|f-g\|_\Lambda^p\le 2 d(f,g),\quad f,g\in\Lambda.

1802: \eequ

1803: This is a common fact in every quasi-normed space (see \cite{BL}). Inequality (\ref{II.3.1}) tells

1804: us that the topology of

1805: $\Lambda\/$ coincides with the one induced by the metric $d\/$ and hence,

1806:  $\Lambda\/$ is a metric space. From Proposition~\ref{pro: II.2.9}  it can be also deduced that it

1807: is complete.

1808: $\Lambda\/$ is then a quasi-Banach space. These and other facts are summarized in the following

1809: theorem. Its proof is obvious using the previous considerations and Proposition~\ref{pro: II.2.9}

1810: (see also  \cite{BL} for property (iii) and \cite{BS} for  (iv)).

1811: \bigbreak

1812: \bteo

1813: \label{te: II.3.2}

1814: \

1815: Every quasi-normed  Lorentz space $\Lambda\/$ is quasi-Banach.

1816: In particular $\Lambda\/$ is an  F-space (topological vector space which is

1817: metrizable with a translations invariant measure and complete). Every function in  $\Lambda\/$ is

1818: finite a.e. and if

1819: $(f_n)_n\subset\Lambda\/$,

1820: \begin{enumerate}

1821: \item[(i)]

1822: $(f_n)_n$ is a Cauchy sequence, if and only if  $\lim_{m,n}\|f_n-f_m\|_\Lambda=0\/$ and then it is also a

1823: Cauchy sequence in measure.

1824:

1825: \item[(ii)] $\Lambda-\lim f_n=f\/$ if and only if  $\lim_{n}\|f-f_n\|_\Lambda=0\/$ and then  $(f_n)_n\/$

1826: converges to

1827: $f\/$ in measure  and there exists a partial that converges to

1828: $f\/$ a.e.

1829:

1830: \item[(iii)] If  $p=\log(2)/\log(2C)\/$ where $C\/$ is the  constant of

1831: the quasi-norm $\|\cdot\|_\Lambda\/$

1832: in $\Lambda\/$, there exists a $p$-norm in $\Lambda\/$ equivalent to

1833: $\|\cdot\|_\Lambda^p\/$. If $\sum_n\|f_n\|_\Lambda^p<\infty\/$ the  series $\sum_n f_n\/$

1834: is convergent in $\Lambda\/$.

1835:

1836: \item[(iv)] If $\Lambda_X\subset\widetilde{\Lambda}_X\/$ and both are quasi-normed  Lorentz

1837: spaces, the embedding is continuous.

1838:

1839: \item[(v)] If $F\/$ is another quasi-normed space,  a linear operator $T:\Lambda\to F\/$

1840: is continuous if and only if   $\sup_{\|f\|_\Lambda\le1}\|Tf\|_F<\infty\/$.

1841: \end{enumerate}

1842: \eteo

1843:

1844:

1845: \subsection{Absolutely continuous norm}

1846:

1847: We now study the equivalent property to the dominated convergence theorem  of the   $L^p\/$

1848: spaces: If

1849: $\lim_n f_n(x)=f(x)\hbox{\ a.e.\ }x\/$ and

1850: $|f_n|\le g\in L^p\/$ then  $\|f_n-f\|_p\limto{n}0\/$. In general, in a Banach function   space

1851: (on

1852: $X\;\sigma$-finite)  a function $g\/$ as before  is said to have   an  absolutely

1853: continuous norm. A space in which every function has    absolutely

1854: continuous norm satisfies the dominated convergence theorem (see \cite{BS}).

1855:

1856: \bdefi

1857:  Let $\big(E,\|\cdot\|\big),\;E\subset\M(X),\/$ be a quasi-normed

1858: space. A function $f\in E\/$   is said to have     absolutely

1859: continuous norm\index{absolutely

1860: continuous norm} if

1861: $$

1862: \lim_{n\to\infty}\big\|f\chi_{A_n}\big\|=0,

1863: $$

1864: for every decreasing sequence of

1865:  measurable sets $(A_n)_n\/$ with $\chi_{A_n}\to0\/$ a.e. If every function in

1866: $E\/$ has this property, we say that  $E\/$ has   an absolutely

1867: continuous norm.

1868: \edefi

1869:

1870: The connection of this property with the dominated convergence theorem is clear from the following

1871: proposition.

1872: Its proof is

1873: (except on some minor modifications) identical to the one in  \cite{BS} (pp. 14--16) for

1874:  Banach function spaces and we omit it.

1875:

1876: \bpro

1877: If $\Lambda\/$ is a quasi-normed  Lorentz space and   $f\in\Lambda\/$, the following statements are

1878: equivalent:

1879: \begin{enumerate}

1880: \item[(i)] $f\/$ has  absolutely continuous norm.

1881:

1882: \item[(ii)] $\lim_n\big\|f\chi_{E_n}\big\|_\Lambda=0\/$ if $(E_n)_n\/$ is a sequence of measurable sets

1883: with

1884: $\chi_{E_n}\to0\/$ a.e.

1885:

1886: \item[(iii)] $\lim_n\|f_n\|_\Lambda=0\/$ if $|f_n|\le|f|\/$ and $\lim_n f_n=0\/$ a.e.

1887:

1888: \item[(iv)] $\lim_n\|g-g_n\|_\Lambda=0\/$ if $|g_n|\le|f|\/$ and $\lim_n g_n=g\/$ a.e.

1889: \end{enumerate}

1890: \epro

1891:

1892: The following result shows that, except in the special case $\mu(X)=\infty\/$ and  $w\in L^1\/$,

1893: the spaces  $\loX\/$ have   absolutely continuous norm.

1894:

1895: \bteo

1896: \label{te: II.3.5}

1897: Let $0<p<\infty\/$ and  $\Lambda^p_X(w)\/$ be a quasi-normed space. Then,

1898: \begin{enumerate}

1899: \item[(i)] If $\mu(X)<\infty,\;\Lambda^p_X(w)\/$ has   absolutely continuous norm.

1900:

1901: \item[(ii)] If $\mu(X)=\infty,\;\Lambda^p_X(w)\/$ has   absolutely continuous norm if and only if

1902:  $w\notin L^1\/$.

1903: \end{enumerate}

1904: \eteo

1905:

1906: \bdem

1907:  (i) Let us assume  that $\mu(X)<\infty\/$ and let us  see that if $0\le f\in\loX\/$

1908: and  $(g_n)_n\/$ is a sequence of measurable functions  satisfying

1909: $$

1910: 0\le g_n\le f\in\Lambda,\quad g_n\to0\hbox{\rm \  a.e.},

1911: $$

1912: then   $\lim_n\|g_n\|_\Lambda=0\/$.

1913: Since the space is of finite measure, the pointwise convergence implies the convergence in measure

1914:   and we have that  $\lim_n\lambda_{g_n}(t)=0,\;0<t<\infty\/$.

1915: In particular $W\big(\lambda_{g_n}(t)\big)\to0,\;t>0\/$. Since

1916: $W\big(\lambda_{g_n}(t)\big)\le W\big(\lambda_f(t)\big)\/$ and $f\in\Lambda^p\/$,

1917: from Proposition~\ref{pro: II.2.5}(ii) and the dominated convergence theorem, it follows that

1918: $\lim_n\|g_n\|_\Lambda=0\/$.

1919:

1920: (ii) Assume that  $w\notin L^1,\;0\le f\in\loX\/$ and  $(g_n)_n\/$ is as above.

1921: The hypothesis on $w\/$ implies $\lambda_f(t)<\infty,\;t>0\/$ and the sets

1922: $E_k=\{f\le1/k\}\/$ have complement of finite measure and, if $f_k=f\chi_{E_k}\/$

1923: we have that $f_k^\ast\le1/k\/$ and also $f_k^\ast\le f^\ast\/$. By the

1924: dominated convergence theorem $\|f_k\|_\Lambda\to0\/$.

1925: Therefore, given $\epsilon>0\/$ there exists a measurable set  $E\subset X\/$ with

1926: $\mu(X\setminus E)<\infty\/$ and such that

1927: $\|f\chi_E\|_\Lambda<\epsilon\/$. Then the functions  $g_n\chi_{X\setminus E}\/$

1928: are as in  (i) and  $\|g_n\chi_{X\setminus E}\|_\Lambda\to0\/$.

1929: We also have that

1930: $$

1931: \limsup_n\|g_n\|_\Lambda\le C\limsup_n\big(\|g_n\chi_{X\setminus E}\|_\Lambda

1932: +\|g_n\chi_E\|_\Lambda\big)\le C\|f\chi_E\|_\Lambda\le C\epsilon.

1933: $$

1934: And, since this is true for every  $\epsilon>0\/$, we obtain that $\lim_n\|g_n\|_\Lambda=0\/$.

1935: This proves the sufficiency in (ii). To show the necessity, let us assume that  $\mu(X)=\infty,\;w\in

1936: L^1\/$ and let us see that

1937: $\loX\/$ has not  absolutely continuous norm.  In this case, since  $X=\bigcup_{n=1}^\infty X_n\/$

1938: with

1939: $\mu(X_n)<\infty\/$ for every

1940: $n\/$, the sets

1941: $E_n=\bigcup_{k=n}^\infty X_k\/$ satisfy $\chi_{E_n}\to0\/$ a.e. and also,

1942: $\chi_{E_n}^\ast=1\/$. Thus, $\lim_n\|\chi_{E_n}\|_{\Lambda^p}=\|w\|_1^{1/p}>0\/$ and the function

1943: $1\in\loX\/$ does not have   absolutely continuous norm.$\quad\qed$

1944: \edem

1945:

1946:

1947:

1948:

1949: \bcor

1950: \label{co: II.3.6}

1951: If $0<p<\infty\/$ and $\lox{p}{X}{w}\/$ is a quasi-normed space,

1952: every function in this space which vanishes in the complement of a set of finite measure, has

1953: absolutely continuous norm.

1954: \ecor

1955:

1956: The analogous question in the case of the weak-type space  $\loinX\/$ is solved

1957:  in the following theorem, for the case when  $X\/$ is a resonant measure space:

1958: these spaces do not have, except in the trivial case,   absolutely continuous norm.

1959:

1960: \bteo

1961: \label{te: II.3.7}

1962: If $X\/$ is a resonant measure space, $0<p<\infty\/$ and  $W\in \Delta_2(X),\;\loinX\/$

1963: has    absolutely continuous norm if and only if  $X\/$ is a finite union of atoms.

1964: \eteo

1965:

1966: \bdem

1967: If $X\/$ is a finite union of atoms the proof is trivial.

1968: Hence, we shall assume that this is not the case and we consider two possibilities:

1969:

1970: (i) If  $X\/$ is nonatomic, since the function $W^{-1/p}\/$ is decreasing and continuous in

1971: $\R^+\/$,

1972: there exists a measurable function

1973: $f\ge0\/$ in $X\/$ such that

1974: $f^\ast(t)=W^{-1/p}(t),\;0<t<\mu(X)\/$. Therefore

1975: $\|f\|_{\Lambda^{p,\infty}(w)}=\sup_t W^{1/p}(t)f^\ast(t)=1\/$ and  $f\in\Lambda^{p,\infty}(w)\/$.

1976: The sets

1977: $E_n=\chi_{\{f>n\}},\;n=1,2,\dots\/$ form a decreasing sequence with  $\chi_{E_n}\to0\/$ a.e. and,

1978: however,

1979: $\|f\chi_{E_n}\|_{\Lambda^{p,\infty}}=1\/$ for every

1980: $n\/$, since

1981: $$

1982: \big(f\chi_{E_n}\big)^\ast(t)=f^\ast(t)\chi_{\big[0,\lambda_f(n)\big)}(t)

1983: =W^{-1/p}(t)\chi_{\big[0,\lambda_f(n)\big)}(t),\quad0<t<\infty.

1984: $$

1985: Hence, $f\/$ does not have an

1986:   absolutely continuous norm.

1987:

1988: (ii) If $X=\bigcup_{n=0}^\infty X_n\/$ with $X_n\/$ an atom of measure  $b>0\/$ and  $X_n\cap

1989: X_m=\emptyset,\;n\ne m,\/$ the function

1990: $$

1991: f=\sum_{n=0}^\infty W^{-1/p}\big((n+1)b\big)\chi_{X_n}

1992: $$

1993: is in $\Lambda^{p,\infty}_X(w)\/$

1994: and has norm  $1\/$. The sets $E_N=\bigcup_{n=N}^\infty

1995: X_n,\;N=1,2,\dots\/$ form a decreasing sequence with  $\chi_{E_n}\to0\/$ a.e. and,

1996:  for each  $N\/$,

1997: $$

1998: \big(f\chi_{E_N}\big)^\ast(nb)=W^{-1/p}\big((n+N+1)b\big),\quad n=0,1,2,\dots

1999: $$

2000: Using that $f^\ast\/$ is constant on every interval $\big[nb,(n+1)b\big)\/$ and

2001: Theorem~\ref{te: II.2.16}, we obtain

2002: \begin{eqnarray*}

2003: \big\|f\chi_{E_N}\big\|_{\Lambda^{p,\infty}}&=&\sup_{t>0}W^{1/p}(t)\big(f\chi_{E_N}\big)^\ast(t)

2004: \\ &=&\sup_{n\ge1}W^{1/p}(nb)W^{-1/p}\big((n+N)b\big)\\

2005: &\ge&\big(W(Nb)W^{-1}(2Nb)\big)^{1/p}\\ &\ge& C^{-1/p}.

2006: \end{eqnarray*}

2007: Then,

2008: $\lim_N\big\|f\chi_{E_N}\big\|_{\Lambda^{p,\infty}}\neq0\/$ and

2009: $\Lambda^{p,\infty}_X(w)\/$ does not have absolutely continuous norm. $\quad\qed$

2010: \edem

2011:

2012:

2013:

2014:

2015: \bdefi

2016: $$

2017: L^\infty_0(X)\index{$L^\infty_0(X)$}=\big\{f\in L^\infty(X)\,:\,\mu\big(\{f\neq0\}\big)<\infty\big\}.

2018: $$

2019: \edefi

2020:

2021: We can now state a positive partial result for the spaces  $\Lambda^{p,\infty}\/$.

2022:

2023: \bpro

2024: \label{pro: II.3.9}

2025: If $0<p<\infty\/$ and $W\in\Delta_2(X)\/$,

2026: every function in  $L^\infty_0(X)\/$ has  absolutely continuous norm in

2027: $\Lambda^{p,\infty}_X(w)\/$.

2028: \epro

2029:

2030: \bdem

2031:  If $f\in L^\infty_0\/$, let  $Y=\{f\neq0\}\subset X\/$.

2032: Then, if $(A_n)_n\/$ is a sequence of measurable sets with  $\chi_{A_n}\to0\/$

2033: a.e., the functions $\big(f\chi_{A_n}\big)_n\/$ are zero in the complement of  $Y\/$

2034: and $\lim_n f(x)\chi_{A_n}(x)=0\hbox{\ a.e.\ }x\in Y\/$. Since  $Y\/$ has finite measure, it

2035: follows from    Egorov's theorem,  that the convergence of the previous

2036: functions is quasi-uniform. Thus,  since  $\epsilon>0\/$, there exists a set

2037: $Y_\epsilon\subset Y\/$ of measure less than

2038: $\epsilon\/$ and such that $f\chi_{A_n}\to0\/$ uniformly in $X\setminus Y_\epsilon\/$. Therefore,

2039: there exists

2040: $n_0\in\N\/$ such that

2041: $$

2042: \big|f(x)\chi_{A_n}(x)\big|<\epsilon,\quad x\in X\setminus Y_\epsilon,\;n\ge n_0.

2043: $$

2044:

2045: Let $n\ge n_0\/$. Then:

2046:

2047: (i) if  $t\ge\|f\|_\infty\/$,

2048: $$

2049: tW^{1/p}\big(\lambda_{f\chi_{A_n}}(t)\big)=tW^{1/p}(0)=0,

2050: $$

2051:

2052: (ii) if $\epsilon\le t\le\|f\|_\infty\/$,

2053: $$

2054: tW^{1/p}\big(\lambda_{f\chi_{A_n}}(t)\big)\le tW^{1/p}\big(\mu(Y_\epsilon)\big)

2055: \le\|f\|_\infty W^{1/p}(\epsilon),

2056: $$

2057:

2058: (iii) if $0\le t<\epsilon\/$,

2059: $$

2060: tW^{1/p}\big(\lambda_{f\chi_{A_n}}(t)\big)\le \epsilon W^{1/p}\big(\mu(Y)\big)

2061: $$

2062: and by Proposition~\ref{pro: II.2.5} (iii) we get that

2063: $$

2064: \big\|f\chi_{A_n}\big\|_{\Lambda^{p,\infty}_X(w)}\le

2065: \max\big\{\|f\|_\infty W^{1/p}(\epsilon),\epsilon W^{1/p}\big(\mu(Y)\big)\big\},\qquad n\ge n_0,

2066: $$

2067: that is,

2068: $$

2069: \limsup_n \big\|f\chi_{A_n}\big\|_{\Lambda^{p,\infty}_X(w)}\le

2070: \max\big\{\|f\|_\infty W^{1/p}(\epsilon),\epsilon W^{1/p}\big(\mu(Y)\big)\big\}.

2071: $$

2072: Since this inequality holds for every $\epsilon>0\/$ and  $\lim_{t\to0}W(t)=0\/$, we conclude that

2073: $$

2074: \lim_n\big\|f\chi_{A_n}\big\|_{\Lambda^{p,\infty}_X(w)}=0.\qquad\qed

2075: $$

2076: \edem

2077:

2078: \subsection{Density  of the  simple functions and    $L^\infty\/$}

2079:

2080: Let us now study the density  of the  simple functions and also of  $L^\infty\/$

2081:  in the quasi-normed  Lorentz spaces. This question is solved in a positive way for the spaces

2082: $\Lambda^p\/$ (except when

2083: $\mu(X)=\infty,\;w\in L^1\/$).  The behavior of the density in the spaces

2084: $\Lambda^{p,\infty}\/$ is more irregular. It is not true here that the simple functions are dense and,

2085: in fact,  not even $L^\infty\/$ is always dense in  $\Lambda^{p,\infty}\/$.

2086:

2087: \bdefi

2088: Let us denote by  $\S=\S(X)\/$\index{$\S(X)$}  the class of simple functions in $X\/$. That is

2089: $$

2090: \S=\big\{f\in\M(X)\,:\,\hbox{card}\big(f(X)\big)<\infty\big\}.

2091: $$

2092: $\SS\/$\index{$\SS(X)$} will be the simple functions  with support in a set of finite measure:

2093: $$

2094: \SS=\SS(X)=\big\{f\in\S\,:\,\mu\big(\{f\neq0\}\big)<\infty\big\}.

2095: $$

2096: \edefi

2097:

2098: It is clear that $\SS\subset\LL\subset\Lambda\/$ for every  Lorentz space $\Lambda\/$.

2099:

2100: \blem

2101: \label{lem: II.3.11}

2102:  If $f\in\M(X)\/$ there exists a sequence $(s_n)_n\subset\S\/$ satisfying

2103: \begin{enumerate}

2104: \item[(i)] $\lim_n s_n(x)=f(x),\;x\in X$,

2105:

2106: \item[(ii)] $\big(|s_n(x)|\big)_n$ is an increasing sequence and $|s_n(x)|\le|f(x)|$, for every $x\in X$,

2107:

2108: \item[(iii)] $|f-s_n|\le|f|,\;n\in\N.$

2109: \end{enumerate}

2110: Moreover, if $f\/$ is bounded then $s_n\to f\/$ uniformly.

2111:  In particular $\S\/$ is dense in  $L^\infty\/$.

2112: \elem

2113:

2114:

2115:

2116: \bteo

2117: \label{te: II.3.12}

2118: If $\Lambda(w)\/$ is a quasi-normed Lorentz space,

2119: $$

2120: \SS\subset\LL\subset\Lambda(w)

2121: $$

2122: and  $\SS\/$ is dense in $\LL\/$ with the topology of $\Lambda\/$. If $w\in L^1\/$,

2123: $$

2124: \S\subset L^\infty\subset\Lambda(w)

2125: $$

2126: and $\S\/$ is dense in $L^\infty\/$ but,

2127: in this case, $\LL\/$ is not dense in  $L^\infty\/$, if $\mu(X)=\infty\/$ (always with the topology

2128: of $\Lambda\/$).

2129: \eteo

2130:

2131: \bdem

2132: The embeddings are trivial. On the other hand, if $f\in\LL\/$

2133: and $E=\{f\neq0\},\;\mu(E)<\infty\/$ and if $(s_n)_n\subset\S\/$ is the sequence of the previous

2134: lemma, since $|s_n|\le|f|\/$, these functions have also support in  $E\/$

2135: and  $(s_n)_n\subset\SS\/$. Besides $\|f-s_n\|_{L^\infty}\to0\/$ and thus,

2136: $$

2137: \|f-s_n\|_{\Lambda}\le\big\|\|f-s_n\|_{L^\infty}\chi_E\big\|_{\Lambda}

2138: =\|f-s_n\|_{L^\infty}\|\chi_E\|_{\Lambda}\to0.

2139: $$

2140: This proves that $\SS\/$ is dense in

2141: $\LL\/$. If $w\in L^1,\;L^\infty\subset\Lambda\/$ and this embedding is continuous

2142: (Theorem~\ref{te: II.3.2}). Since $\S\/$ is dense in $\big(L^\infty,\|\cdot\|_{L^\infty}\big)\/$

2143: (Lemma~\ref{lem: II.3.11}), we have that $\S\/$ is dense in $L^\infty\/$ with the topology of

2144: $\Lambda\/$.  In this case $1\in L^\infty\subset\Lambda\/$ and if  $g\in\LL,\;|g-1|=1\/$ in

2145: a set of infinite measure if  $\mu(X)=\infty\/$. Thus,  $(1-g)^\ast\ge 1\/$ and  $\|1-g\|_{\Lambda}

2146: \ge C_{\Lambda}>0\/$. Therefore,  $1\/$ cannot be a limit in

2147: $\Lambda\/$ of functions in  $\LL\/$ and this set is not dense in $L^\infty.\quad\qed$

2148: \edem

2149:

2150:

2151: \bteo\label{te: II.3.13}

2152:  Let  $0<p<\infty\/$ and $W\in\Delta_2(X)\/$. Then,

2153: \begin{enumerate}

2154: \item[(i)] if  $w\notin L^1,\;\LL\/$ is dense in $\Lambda^p(w)\/$,

2155:

2156: \item[(ii)] if  $w\in L^1,\;L^\infty\/$ is dense in $\Lambda^p(w)\/$. But, in this case,

2157: $\LL(X)\/$ is not dense in $\Lambda^p_X(w)\/$, if $\mu(X)=\infty\/$.

2158: \end{enumerate}

2159: \eteo

2160:

2161: \bdem

2162: Let us see that $L^\infty\/$ ($\LL\/$ in the case $w\notin L^1\/$)

2163: is dense in $\Lambda^p\/$. If $f\in\Lambda^p\/$,  Proposition~\ref{pro: II.2.5} tells us that

2164: $\lim_{t\to\infty}\lambda_f(t)=0\/$. If for each  $n=1,2,\dots\/$ we define $f_n

2165: =f\chi_{\{|f|\le n\}}\/$, since $f-f_n=

2166: f\chi_{\{|f|>n\}},$ then $(f-f_n)^\ast=f^\ast\chi_{\big[0,\lambda_f(n)\big)}\/$  and we have that

2167: $\lim_n(f-f_n)^\ast(t)=0,\;t>0\/$. On the other hand $(f-f_n)^\ast\le f^\ast\/$ and by the dominated

2168: convergence theorem, we get

2169: $\|f-f_n\|_{\Lambda^p}\to0\/$.  This proves that

2170: $L^\infty\cap\Lambda^p\/$  is dense in  $\Lambda^p\/$. If $w\in L^1\/$,  the first of these spaces

2171: is

2172: $L^\infty\/$ and we are done. If $w\notin L^1\/$

2173:  we only have to see that every function in  $L^\infty\cap\Lambda^p\/$ is limit (in

2174: $\Lambda^p\/$) of functions in  $\LL\/$. But if $g\in L^\infty\cap\Lambda^p\/$

2175: and  $w\notin L^1,\;\lim_{t\to\infty}g^\ast(t)=0\/$ and the functions

2176: $g_n=g\chi_{\{|g|>g^\ast(n)\}}\in\LL,\;n=1,2,\dots,\/$ satisfy $(g-g_n)^\ast\le

2177: g^\ast(n)\to0\/$ and since  $(g-g_n)^\ast\le g^\ast\/$, the dominated

2178: convergence theorem shows that $\|g-g_n\|_{\Lambda^p}\to0\/$.

2179:

2180: It only remains to see that $\LL\/$ is not dense in $\Lambda^p(w)\/$ if $w\in L^1\/$ and

2181: $\mu(X)=\infty\/$. But, in this case, we know by  Theorem~\ref{te: II.3.12}, that $\LL\/$

2182: is not dense in  $L^\infty\/$ and, hence, it can neither be dense in $\Lambda^p\supset

2183: L^\infty.\quad\qed$

2184: \edem

2185:

2186:

2187: In the weak-type case we have the following result of density.

2188:

2189: \bteo

2190: If $X\/$ is a resonant measure space, $0<p<\infty\/$ and $\Lambda^{p,\infty}_X(w)\/$

2191: is a quasi-normed space,

2192: \begin{enumerate}

2193: \item[(i)] $\LL\/$ is dense in  $\Lambda^{p,\infty}\cap L^\infty\/$  if and only if  $\mu(X)<\infty\/$

2194: and then they coincide (the density is considered with respect to the topology of

2195: $\Lambda^{p,\infty}\/$).

2196:

2197: \item[(ii)] $\Lambda^{p,\infty}\cap L^\infty\/$ is dense in $\Lambda^{p,\infty}\/$

2198: if and only if  $X\/$ is atomic and then they coincide.

2199: \end{enumerate}

2200: \noindent Hence, in both cases, if these two spaces do not coincide, the smaller one is not dense in

2201: the other.

2202: \eteo

2203:

2204: \bdem

2205:  (i) If $\mu(X)<\infty\/$ both spaces coincide and there is nothing to prove. If $\mu(X)=\infty\/$

2206: and

2207: $X\/$ is not atomic, we can take   $f\in L^\infty\cap\Lambda^{p,\infty}\/$ with

2208: $$

2209: f^\ast=W^{-1/p}(1)\chi_{[0,1)}+W^{-1/p}\chi_{[1,\infty)}.

2210: $$

2211: If $g\in\LL\/$ and  $E=\{g=0\}\/$

2212: then $b=\mu(X\setminus E)<\infty\/$ and

2213: $$

2214: (f-g)^\ast(t)\ge\big((f-g)\chi_E\big)^\ast(t)=\big(f\chi_E\big)^\ast(t)\ge f^\ast(b+t)

2215: =W^{-1/p}(t+b),\quad t>1,

2216: $$

2217: and we have that,

2218: $$

2219: \|f-g\|_{\Lambda^{p,\infty}}=\sup_{t}W^{1/p}(t)(f-s)^\ast(t)\ge \sup_{t>1}{W^{1/p}(t)

2220: \over W^{1/p}(t+b)}.

2221: $$

2222: If $w\in L^1\/$ the above supremum is  $1\/$ and, in the opposite case, since

2223: $W\in\Delta_2,\;W(t+b)\le C\big(W(t)+W(b)\big)\/$ (cf. Theorem~\ref{te: II.2.13} and

2224: Proposition~\ref{pro: II.2.12})  we obtain that

2225: $\|f-g\|_{\Lambda^{p,\infty}}\ge C^{-1/p}\/$ and  $f\/$ cannot be the limit of functions in $\LL\/$.

2226: This shows that $\LL\/$ is not dense in  $L^\infty

2227: \cap\Lambda^{p,\infty}\/$ in the nonatomic case. If $\mu(X)=\infty\/$ and  $X\/$ is an atomic

2228: resonant measure space,

2229: $X=\bigcup_{n=0}^\infty X_n\/$ with every $X_n\/$ an atom of measure  $b>0\/$ and  $X_n\cap

2230: X_m=\emptyset,\;n\ne m,\/$ and the function

2231: $$

2232: f=\sum_{n=0}^\infty W^{-1/p}\big((n+1)b\big)\chi_{X_n}

2233: $$

2234: is in  $\Lambda_X^{p,\infty}(w)

2235: \cap L^\infty\/$ and   has norm $1\/$. If $g\in\LL\/$, then its support has finite measure (a finite

2236: union

2237: $X_{n_1}\cup\dots\cup X_{n_k}\/$ of atoms) and there exists $N\in\N\/$

2238: such that $g=0\/$ in $\bigcup_{n=N}^\infty X_n\/$. Hence,

2239: $(f-g)^\ast(nb)\ge f^\ast\big((n+N)b\big)=W^{-1/p}\big((n+N+1)b\big),\;n=0,1,2,\dots,\/$

2240: proceeding as in the proof of  Theorem~\ref{te: II.3.7}  we can conclude that

2241: $\|f-g\|_{\Lambda^{p,\infty}}\ge C>0\/$. Thus,  $f\/$ is not the limit of functions in $\LL\/$

2242: and consequently this set is not dense in

2243: $\Lambda^{p,\infty}\cap L^\infty\/$ and (i) is proved.

2244:

2245: (ii) As before, if $X\/$ is atomic both spaces coincide and there is nothing to be proved.  If

2246: $X\/$ is nonatomic, there exists $0\le f\in\M(X)\/$ with $f^\ast=W^{-1/p}\/$ in $[0,\mu(X))\/$.

2247: Thus,

2248: $\|f\|_{\Lambda^{p,\infty}}=1\/$ and $f\in\Lambda^{p,\infty}\setminus

2249:  L^\infty\/$. If $g\in L^\infty(X)\/$ with $\|g\|_\infty=b,$ then $|f-g|\ge|f|-b$

2250: and, hence, $(f-g)^\ast(t)\ge f^\ast(t)-b=W^{-1/p}(t)-b\/$ if $0\le t<\lambda_f(b)=t_0\/$.

2251: Therefore,

2252: $$

2253: \|f-g\|_{\Lambda^{p,\infty}}=\sup_t W^{1/p}(t)(f-g)^\ast(t)\ge\sup_{0<t<t_0}\big(1-b

2254: W^{1/p}(t)\big)=1,

2255: $$

2256: since $t_0=\lambda_f(b)>0\/$. We conclude that $f\/$ is not the limit in $\Lambda^{p,\infty}\/$ of

2257: functions in

2258: $L^\infty(X)\/$. This proves that

2259: $L^\infty\cap\Lambda^{p,\infty}\/$  is not dense in  $\Lambda^{p,\infty}\/$  in this case.

2260: $\quad\qed$

2261: \edem

2262:

2263:

2264:

2265: \section{Duality}

2266: \label{se:dual}

2267:

2268:

2269: In this section $(X,\mu)\/$ will be a $\sigma$-finite measure space.

2270: $d(\Omega,p)\/$  and its weak version   $d^\infty(\Omega,p)\/$ will be the

2271: Lorentz spaces on  $\N^\ast\/$ introduced in  Example~\ref{Eje2.3}(iv).

2272: Here $\Omega=(\Omega_n)_{n=0}^\infty\/$ is a sequence of positive numbers $\Omega_0\neq0\/$.

2273:

2274: We shall study the dual spaces and the associate spaces of the

2275: Lorentz spaces on $X\/$ with special attention to the case when $X\/$ is a resonant measure space

2276: (Definition~\ref{def: II.2.15}).

2277: We shall describe the associate space and we shall deduce a necessary  and sufficient condition to

2278: have that the dual and the associate spaces  coincide. In some cases, we shall also give a

2279: description of the Banach envelope  of

2280: $\Lambda\/$, which is of interest if this space is not a normed space. In our study, we include

2281:  the sequence Lorentz spaces $d(\Omega,p)\/$ with  $\Omega\/$ arbitrary. As far as we know, only the

2282: Banach case has been previously studied  (that is, the case $\Omega\/$ decreasing).

2283:

2284: We shall introduce the associate space generalizing the definition that   can be found in

2285: \cite{BS} in the context of  Banach function spaces.

2286:

2287: \bdefi

2288: \label{def:II.4.1}

2289:  If $\|\cdot\|:\M(X)\to[0,\infty]\/$ is a  positively homogeneous functional

2290: and  $E=\big\{f\in\M(X)\,:\,\|f\|<\infty\big\}\/$, we define the

2291:  associate \  \lq\lq norm\rq\rq

2292: $$

2293: \|f\|_{E^\prime}=\sup\bigg\{\int_X|f(x)g(x)|\,d\mu(x)\,:\,\|g\|\le1,\;g\in E\bigg\},\quad f\in\M(X).

2294: $$

2295: The associate space\index{associate space} of  $E\/$ is then

2296: $E^\prime=\big\{f\in\M(X)\,:\,\|f\|_{E^\prime}<\infty\big\}\/$.

2297: \edefi

2298:

2299: \bobs

2300: \label{Obs: II.4.2}

2301: The following properties are immediate:

2302:

2303: (i)  $\|\cdot\|_{E^\prime}\/$ is subadditive and positively homogeneous and, if $E\/$

2304: contains the characteristic functions of sets of finite measure,

2305: $\big(E^\prime,\|\cdot\|_{E^\prime}\big)\/$ is a normed space.

2306:

2307: (ii) If $(E,\|\cdot\|)\/$ is a lattice ($\|f\|\le\|g\|\/$, if $|f|\le|g|\/$)

2308: then  $E^\prime\/$ is also a lattice. If $\|\cdot\|\/$ has the  Fatou property, the same happens to

2309: $\|\cdot\|_{E^\prime}\/$.

2310:

2311: (iii) If we denote by  $E^{\prime\prime}=(E^\prime)^\prime\/$ we have that

2312: $E\subset E^{\prime\prime}\/$ and   $\|f\|_{E^{\prime\prime}}\le\|f\|\/$ for every  $f\in\M(X)\/$.

2313: \eobs

2314:

2315:

2316: \bdefi

2317: \label{def4.3}

2318: A  Banach function space (BFS)\index{Banach function space} $E\/$ in $X\/$  is a subspace of   $\M(X)\/$

2319: defined by

2320: $E=\{f:\|f\|<\infty\}\/$ where

2321: $\|\cdot\|=\|\cdot\|_E\/$ is a norm (called \lq\lq Banach function norm")

2322: that satisfies the following properties, for $f,g,f_n\in E,\;A\subset X\/$ measurable (see \cite{BS}):

2323: \begin{enumerate}

2324: \item[(i)]  $\|f\|\le\|g\|\/$ if $|f|\le|g|\/$,

2325:

2326: \item[(ii)]  $0\le f_n\le f_{n+1}\to f\Rightarrow\|f_n\|\to\|f\|\/$,

2327:

2328: \item[(iii)]  $\chi_A\in E\/$ if $\mu(A)<\infty\/$ and $E\neq\{0\}\/$,

2329:

2330: \item[(iv)]  $\displaystyle\int_A|f(x)|\,d\mu(x)\le C_A\|f\|\/$ if $\mu(A)<\infty\/$.

2331: \end{enumerate}

2332: \edefi

2333:

2334: The result that   follows establishes that the associate space  $\Lambda^\prime_X\/$

2335: of a quasi-normed Lorentz space is a   Banach function space whenever  $X\/$ is a resonant measure

2336: space.

2337:

2338: \bteo

2339: If $\Lambda_X\/$ is a quasi-normed  Lorentz space on  a resonant measure space $X\/$, the associate

2340: space

2341: $\Lambda^\prime_X\/$  is a  rearrangement invariant Banach function space\index{rearrangement invariant}.

2342: For every

2343: $f\in\Lambda^\prime_X\/$,

2344: \bequ\label{asso}

2345: \|f\|_{\Lambda^\prime_X}=\sup\bigg\{\int_0^\infty f^\ast(t) g^\ast(t)\,dt\,:\,\|g\|_{\Lambda_X}\le1\bigg\}.

2346: \eequ

2347: Moreover, a function $f\in\M(X)\/$ is in $\Lambda^\prime_X\/$  if and only if

2348: $\displaystyle\int_X|f(x)g(x)|\,d\mu(x)<\infty\/$ for every

2349: $g\in\Lambda_X\/$ and  $\Lambda^\prime_X\ne\{0\}\/$  if and only if  $\Lambda_X\subset

2350: L^1_{\hbox{\decp loc}}(X)\/$.

2351: \eteo

2352:

2353: \bdem

2354: Let  us first prove  (\ref{asso}). If $f,g\in\M(X)$, we have

2355: that $\displaystyle\int_X|f(x)g(x)|\,d\mu(x)\le\displaystyle\int_0^\infty  f^\ast(t) g^\ast(t)\,dt\/$ and

2356: thus,

2357: $$

2358: \|f\|_{\Lambda^\prime_X}\le\sup\bigg\{\int_0^\infty f^\ast (t)

2359: g^\ast(t)\,dt\,:\,\|g\|_{\Lambda_X}\le1\bigg\}.

2360: $$

2361: To prove the other inequality, we observe that  for

2362: $g\in\Lambda_X\/$, with  $\|g\|_{\Lambda_X}\le1\/$ we have, since $X\/$ is resonant,

2363: \begin{eqnarray*}

2364: \int_0^\infty f^\ast(t) g^\ast(t)\,dt&=&\sup_{h^\ast=g^\ast}\int_X|f(x)h(x)|\,d\mu(x)\\

2365: &\le&\sup_{\|h\|_{\Lambda_X}\le1}\int_X|f(x)h(x)|\,d\mu(x)

2366: =\|f\|_{\Lambda^\prime_X}.

2367: \end{eqnarray*}

2368: Hence, the functional

2369: $\|\cdot\|_{\Lambda^\prime_X}\/$ is rearrangement invariant.

2370:

2371: Let us now prove that  $f\in\Lambda^\prime_X\/$ if and only if

2372: $\displaystyle\int_X|f(x)g(x)|\,d\mu(x)<\infty\/$  for every $g\in\Lambda_X\/$. The necessity is a

2373: consequence of the definition of

2374: $\Lambda^\prime_X\/$. The sufficiency follows from the closed graph theorem (see

2375: \cite{Ru2})  since under the hypothesis, the linear operator $T_f(g)=fg,\;T_f:\Lambda_X\to

2376: L^1(X)\/$ is well defined and  both are  F-spaces continuously embedded in  $\M(X)\/$

2377: (from which one can immediately see that the graph is closed). $T_f\/$ is then continuous and this

2378: proves    (Theorem~\ref{te: II.3.2}

2379: (v)) that $\displaystyle\int_X|f(x)g(x)|\,d\mu(x)\le C\|g\|_{\Lambda_X}\/$  and thus

2380: $f\in\Lambda^\prime_X\/$.

2381:

2382: Let us now see that  $\Lambda^\prime_X\/$ is a   Banach function space. That

2383: $\|\cdot\|_{\Lambda^\prime_X}\/$  is a norm is immediate. Properties

2384: (i) and  (ii) in the definition of BFS (Definition~\ref{def4.3}) are also trivial.  Property  (iv)

2385: of that definition is also very easy, since if  $\mu(A)<\infty\/$ and $f\in\Lambda^\prime_X\/$,

2386: $$

2387: \int_A|f(x)|\,d\mu(x)\le\|\chi_A\|_{\Lambda_X}\|f\|_{\Lambda^\prime_X}.

2388: $$

2389: It only remains to prove (iii). That is that

2390: $\chi_A\in\Lambda^\prime_X\/$ if $\mu(A)<\infty\/$ and

2391: $\Lambda^\prime_X\neq\{0\}\/$. In the atomic case  it is trivial and if $X\/$  is not atomic we have

2392: that, if

2393: $0\neq f\in\Lambda^\prime_X\/$ there exists $t>0\/$ such that, if

2394: $E=\{|f|>t\},$ then $\mu(E)>0\/$. Hence, $t\chi_E\le|f|\/$ and   $\chi_E\in\Lambda^\prime_X\/$. That is, if

2395: $\Lambda^\prime_X\neq\{0\}\/$ there exists

2396: $\chi_E\in\Lambda^\prime_X\/$ with $b=\mu(E)>0\/$. Since $\|\cdot\|_{\Lambda^\prime_X}\/$ is

2397: rearrangement invariant and monotone, we have that

2398: $\chi_A\in\Lambda^\prime\/$ for every measurable set  $A\/$ with $\mu(A)\le b\/$. If

2399: $\infty>\mu(A)>b\/$,  the above is also true since $A=\bigcup_{n=1}^N A_n\/$ with

2400: $\mu(A_n)\le b\/$ and $\|\chi_A\|_{\Lambda^\prime_X}\le\sum_n\|\chi_{A_n}\|_{\Lambda^\prime_X}<\infty\/$.

2401:

2402: It only remains to prove that  $\Lambda^\prime_X\ne\{0\}\/$  if and only if $\Lambda_X\subset

2403: L^1_{\hbox{\decp loc}}(X)\/$. The necessity is immediate since

2404: $\Lambda^\prime_X\ne\{0\}\/$ and  $\mu(A)<\infty\/$ then we have already seen

2405: that $\chi_A\in\Lambda^\prime_X\/$ and, in particular

2406: $\displaystyle\int_A|f(x)|\,d\mu(x)<\infty,\;f\in\Lambda_X\/$. On the other hand if

2407: $\Lambda_X\subset L^1_{\hbox{\decp  loc}}(X)\/$ and  $0<\mu(A)<\infty\/$ we have that

2408: $\displaystyle\int_X|f(x)\chi_A(x)|\,d\mu(x)<\infty\/$ for every  $f\in\Lambda_X\/$ and that implies

2409: $\chi_A\in\Lambda^\prime_X\/$ and $\Lambda^\prime_X\neq\{0\}.\quad\qed$

2410: \edem

2411:

2412:

2413:

2414: \bobs

2415:  If in the previous theorem, we omit the condition that   $X\/$

2416: is a resonant measure space   we  can still prove that

2417: $\Lambda^\prime_X\/$ is a  Banach space  with monotone norm and with the Fatou property.

2418: The convergence in $\Lambda^\prime_X\/$ implies the convergence in measure on sets of finite measure

2419:  (same for the  Cauchy sequences) and it is also true that  $f\in\Lambda^\prime_X\/$ if and only if

2420: $\displaystyle\int_X|f(x)g(x)|\,d\mu(x)<\infty\/$ for every

2421: $g\in\Lambda_X\/$. The last statement of the theorem is not true  (in general) in this case.

2422: \eobs

2423:

2424:

2425: The result that follows describes the associate space of  $\Lambda_X\/$  in the case $X\/$

2426: nonatomic. But first, we need to define the Lorentz space $\Gamma\/$.\index{Lorentz space!Gamma}

2427:

2428: \bdefi

2429: \label{def: II.4.6}

2430: If $0<p\le\infty\/$ we define

2431: $$

2432: \Gamma^p_X(w)\index{$\Gamma^p_X(w)$}=\bigg\{f\in\M(X)\,:\,\|f\|_{\Gamma^p_X(w)}=\bigg(\int_0^\infty

2433: (f^{\ast\ast}(t))^pw(t)\,dt\bigg)^{1/p}<\infty\,\bigg\}.

2434: $$

2435: The weak-type version of the previous space is

2436: $$

2437: \Gamma^{p,\infty}_X(w)\index{$\Gamma^{p,\infty}_X(w)$}=\bigg\{f\in\M(X)\,:\,\|f\|_{\Gamma^{p,\infty}_X(w)}=\sup_{t>0}

2438: W^{1/p}(t)f^{\ast\ast}(t)<\infty\,\bigg\}.

2439: $$

2440: The last definition can be extended in the following way. If $\Phi\/$  is an arbitrary function in

2441: $\R^+\/$ we write

2442: $$

2443: \Gamma^{p,\infty}_X(\di\Phi)\index{$\Gamma^{p,\infty}_X(\di\Phi)$}=\bigg\{f\in\M(X)\,:\,\|f\|_{\Gamma^{p,\infty}_X(\dip\Phi)}

2444: =\sup_{t>0}\Phi^{1/p}(t)f^{\ast\ast}(t)<\infty\,\bigg\}.

2445: $$

2446: \edefi

2447:

2448: We can always assume that the function $\Phi\/$ in the previous definition is increasing since

2449: otherwise it can be  substituted by

2450: $\widetilde{\Phi}(t)=\sup_{0<s<t}\Phi(s),\;t>0,\/$ which is increasing and satisfies

2451: $\|f\|_{\Gamma^{p,\infty}_X(\dip\Phi)}=\|f\|_{\Gamma^{p,\infty}_X(\dip\widetilde{\Phi})}\/$.

2452:

2453:  Condition (ii) of the following result  is a direct consequence of   Sawyer's formula

2454: stated in  Theorem~\ref{te: I.5.7} while  (i) was proved in

2455: \cite{CS2}.

2456:

2457: \bteo

2458: \label{te: II.4.7}

2459:  Let $X\/$  be a nonatomic measure space and let  $w\/$ be an arbitrary weight in  $\R^+\/$.

2460: \begin{enumerate}

2461: \item[(i)] If $0<p\le1\/$, then

2462: $$

2463: \Lambda_X^p(w)^\prime=\Gamma^{1,\infty}_X(\di\Phi)\qquad\hbox {(with equal norms),}

2464: $$

2465: where

2466: $\Phi(t)=tW^{-1/p}(t),\;t>0\/$.

2467:

2468: \item[(ii)] If $1<p<\infty\/$ and $f\in\M(X)\/$, then

2469: \begin{eqnarray*}

2470: \|f\|_{\Lambda_X^p(w)^\prime}&\approx&\bigg(\int_0^\infty\bigg({1\over W(t)}\int_0^t

2471: f^\ast(s)\,ds\bigg)^{p^\prime}w(t)\,dt\bigg)^{1/p^\prime}+{\displaystyle\int_0^\infty f^\ast(t)\,dt\over

2472: W^{1/p}(\infty)}\\ &\approx&\bigg(\int_0^\infty\bigg({1\over W(t)}\int_0^t

2473: f^\ast(s)\,ds\bigg)^{p^\prime-1}f^\ast(t)\,dt\bigg)^{1/p^\prime}.

2474: \end{eqnarray*}

2475:

2476: \item[(iii)] If $0<p<\infty\/$, then

2477: $$

2478: \Lambda_X^{p,\infty}(w)^\prime=\Lambda^1(W^{-1/p})\qquad\hbox{(with equal norms).}

2479: $$

2480: \end{enumerate}

2481: \eteo

2482:

2483: \bdem

2484:  Since  $X\/$ is nonatomic and  $\sigma$-finite, every decreasing function in

2485: $[0,\mu(X))\/$ equals  a.e. to the decreasing rearrangement of a function in

2486: $\M(X)\/$. Besides,   $X\/$ is resonant  and hence,

2487: $$

2488: \|f\|_{\lox{p,q}{X}{w}^\prime}=\sup_{g\in\lox{p,q}{X}{w}}{\displaystyle\int_X|f(x)g(x)|\,d\mu(x)\over\|g\|_{\lox{p,q}{X}{w}}

2489: }=\sup_{g\downarrow}{\displaystyle\int_0^\infty

2490: g(t)f^\ast(t)\,dt\over\|g\|_{L^{p,q}(w)}}.

2491: $$

2492: The first case (i) is solved applying Corollary~\ref{cor: I.2.14}

2493: (with $p_1=1,\,T=\hbox{\rm Id}\/$) to the regular class  $L=\Fd\/$  or

2494: Theorem 2.12 in \cite{CS1}. The second case  (ii) is an immediate consequence of Theorem~\ref{te: I.5.7}

2495: (E. Sawyer). Finally, (iii) corresponds to  $q=\infty\/$ and we have that,

2496: $$

2497: \|f\|_{\lox{p,\infty}{X}{w}^\prime}=\sup\bigg\{\int_0^\infty f^\ast(t)

2498: g(t)\,dt\,:\,\|g\|_{L^{p,\infty}}(w)=1,\;g\downarrow\bigg\}.

2499: $$

2500: Now, if

2501: $g\downarrow,\;\|g\|_{L^{p,\infty}(w)}=\sup_t W^{1/p}(t)g(t)\/$ and  $\|g\|_{L^{p,\infty}}(w)=1\/$

2502: implies $g\le W^{-1/p}\/$, and therefore

2503: $$

2504: \|f\|_{\lox{p,\infty}{X}{w}^\prime}\le\int_0^\infty f^\ast(t) W^{-1/p}(t)\,dt=\|f\|_{\lox{1}{}{W^{-1/p}}}.

2505: $$

2506: On the other hand $W^{-1/p}\/$  is decreasing,

2507: $\big\|W^{-1/p}\big\|_{L^{p,\infty}}(w)=1\/$ and we have the equality. $\quad\qed$

2508: \edem

2509:

2510:

2511: \bobs

2512: \label{obs: II.4.8}

2513:

2514: (i) If $p>1\/$ and  $\tilde{w}(t)=t^{p^\prime}W^{-p^\prime}(t)w(t),\;t>0\/$, then

2515: (ii) of the previous theorem can be stated of the following way:

2516: \begin{eqnarray*}

2517: \lox{p}{X}{w}^\prime&=&\Gamma^{p^\prime}_X(\tilde{w}),\qquad\hbox{\rm if\ }w\notin L^1,\\

2518: \lox{p}{X}{w}^\prime&=&\Gamma^{p^\prime}_X(\tilde{w})\cap L^1(X),\qquad\hbox{\rm if\ }w\in L^1.

2519: \end{eqnarray*}

2520: It is assumed that the norm  in the intersection space is the maximum of the sum  of both norms  and

2521: in these equalities we are assuming the equivalence of the norms.

2522:

2523: (ii) If $w\notin L^1,\;p>1\/$ and $\tilde{w}\/$ (as above defined) is in $B_{p^\prime}\/$,  the  Hardy

2524: operator $A\/$ satisfies the boundedness $A:L^{p^\prime}_{\hbox{\decp  dec}}(\tilde{w})\to

2525: L^{p^\prime}(\tilde{w})\/$ and then,

2526: $$

2527: \|f\|_{\Lambda^p_X(w)^\prime}\approx\|f\|_{\Lambda_X^{p^\prime}(\tilde{w})}\,,\quad f\in\M(X).

2528: $$

2529: It is easy to see that this condition on the weight    $w\/$ is equivalent to

2530: \bequ

2531: \label{II.4.9}

2532: \bigg(\int_0^r\bigg({W(t)\over t}\bigg)^{-p^\prime}w(t)\,dt\bigg)^{1/p^\prime}W^{1/p}(r)\ge

2533: Cr,\quad r>0,

2534: \eequ

2535: which is the opposite inequality to the condition   $w\in

2536: B_{p,\infty}=B_p\/$ (Theorem~\ref{te: I.6.5}). Since one of the embedding always holds, it follows

2537: that condition  (\ref{II.4.9})  is necessary  and sufficient  (in the case

2538: $w\notin L^1\/$) to have the identity

2539: $$

2540: \lox{p}{X}{w}'=\Lambda^{p^\prime}_X(\tilde{w})\qquad\hbox{\rm (with equivalent norms).}

2541: $$

2542:

2543:

2544: (iii) If $1<p<\infty\/$ the space $\Lambda_X^{p^\prime}(w^{1-p^\prime})\/$ is embedded in

2545: $\lox{p}{X}{w}^\prime\/$ since, by  H\"older's inequality,

2546: \begin{eqnarray*}

2547: \int_X |f(x)g(x)|\,d\mu(x)&\le&\int_0^\infty (f^\ast(t) w^{1/p}(t))(g^\ast(t)

2548: w^{-1/p}(t))\,dt\\

2549: &\le&\|f\|_{\Lambda^p(w)}\|g\|_{\Lambda^{p^\prime}(w^{1-p^\prime})}.

2550: \end{eqnarray*}

2551: If $\tilde{w}\/$ is as before,  we have that

2552: \bequ

2553: \label{II.4.10}

2554: \lox{p^\prime}{X}{w^{1-p^\prime}}\subset\lox{p}{X}{w}^\prime\subset

2555: \Gamma^{p\prime}_X(\tilde{w})\subset\Lambda^{p^\prime}_X(\tilde{w}).

2556: \eequ

2557:

2558: (iv)  The  comments made after    Definition~\ref{def: II.4.6}

2559:   tell  us that, for every

2560: $0<p\le1\/$,

2561: $$

2562: \Lambda_X^p(w)^\prime=\Gamma^{1,\infty}_X(\di\Phi_p)\qquad\hbox{\rm (with equal  norms),}

2563: $$

2564:  where

2565: $\Phi_p(t)=\sup_{0<s<t}sW^{-1/p}(s),\;t>0\/$. Note that

2566: $$

2567: \Phi_p(t)={t\over\inf_{0<s<t}{t\over s}W^{1/p}(s)}={t\over\inf_{s>0}\max\big(1,{t\over

2568: s}\big)W^{1/p}(s)}={t\over W_p(t)},

2569: $$

2570: where

2571: $W_p(t)=\inf_{s>0}\max\big(1,{t\over s}\big)W^{1/p}(s),\;t>0\/$. It is easy to check that this

2572: function is quasi-concave (\cite{BS}). In fact, it is the biggest quasi-concave function majorized

2573: by

2574: $W^{1/p}\/$ and consequently it is called the greatest concave minorant of  $W^{1/p}\/$ (see

2575: \cite{CPSS}).  Hence, we can write,

2576: $$

2577: \|f\|_{\Lambda^p_X(w)^\prime}=\sup_{t>0}{1\over W_p(t)}\int_0^t f^\ast(s)\,ds,\qquad f\in\M(X).

2578: $$

2579: \eobs

2580:

2581: As a first consequence of Theorem~\ref{te: II.4.7} we obtain the  characterization of the weights

2582: $w\/$  for which $\Lambda^\prime(w)=\{0\}\/$.

2583:

2584:

2585:

2586: \bteo

2587: \label{te: II.4.11}

2588: If $X\/$ is nonatomic:

2589: \begin{enumerate}

2590: \item[(i)] If $0<p\le1,\;(\Lambda^p_X(w))^\prime\ne\{0\}\Leftrightarrow\sup_{0<t<1}{\displaystyle t^p\over

2591: \displaystyle W(t)}<\infty.$

2592: \item[(ii)] If $1<p<\infty,\;(\Lambda^p_X(w))^\prime\ne\{0\}\Leftrightarrow\displaystyle

2593: \int_0^1\bigg({\displaystyle t\over \displaystyle W(t)}\bigg)^{p^\prime-1}\,dt<\infty.$

2594: \item[(iii)] If $0<p<\infty,\;(\Lambda^{p,\infty}_X(w))^\prime\ne\{0\}\Leftrightarrow\displaystyle \int_0^1

2595: {\displaystyle 1\over \displaystyle W^{1/p}(t)}\,dt<\infty.$

2596: \end{enumerate}

2597: \eteo

2598:

2599: \bdem

2600:  Since $\Lambda^\prime\/$ is a BFS, it is not identically zero if and only if

2601: it contains the functions $\chi_E\/$ with $\mu(E)<\infty\/$, that is  if and only if

2602: $\|\chi_E\|_{\Lambda^\prime}<\infty,\;\mu(E)<\infty\/$. The conditions are then obtained

2603: applying Theorem~\ref{te: II.4.7}.~$\quad\qed$

2604: \edem

2605:

2606: \bobs

2607: Since $W\/$ is continuous in $\R^+$,  the condition of the previous theorem only depends on the

2608: local behavior of  $w\/$ at  $0\/$.

2609: \eobs

2610:

2611:

2612: \bdefi

2613: A weight $w\/$ in $\R^+\/$ is called  regular\index{regular weight} (see \cite{Re}) if it satisfies

2614: $$

2615: {W(t)\over t}\le C\,w(t),\qquad t>0,

2616: $$

2617: with $C>0\/$ independent of $t\/$. A sequence of positive numbers  $(\Omega_n)_{n=0}^\infty\/$ is

2618: said, analogously, to be regular if\index{regular sequence}

2619: $$

2620: {1\over n+1}\sum_{k=0}^n\Omega_k\le C\,\Omega_n,\qquad n=0,1,2,\dots

2621: $$

2622: \edefi

2623:

2624: Every increasing sequence   and every power sequence    are regular. In

2625: \cite{Re} it is proved that if  $w\/$ is regular and decreasing,  the spaces

2626: $\lox{p^\prime}{X}{w^{1-p^\prime}}\/$ and $\lox{p}{X}{w}^\prime\/$ coincide. In the following

2627: result, we extend this to the case of an arbitrary  weight $w\/$.

2628:

2629:

2630: \bteo

2631: \label{te: II.4.14}

2632: Let  $1<p<\infty\/$ and  $X\/$ be nonatomic. Then:

2633: \begin{enumerate}

2634: \item[(i)] If $w\notin L^1,\;\lox{p}{X}{w}^\prime=\lox{p^\prime}{X}{w^{1-p^\prime}}\/$

2635: if and only if there exists  $C>0\/$ such that, for $r>0$,

2636: \bequ\label{equiv}

2637: \int_0^r w^{1-p^\prime}(t)\,dt\le C\bigg(r^{p^\prime}W^{1-p^\prime}(r)+\int_0^r

2638: t^{p^\prime}W^{-p^\prime}(t)w(t)\,dt\bigg).

2639: \eequ

2640:

2641: \item[(ii)] If $w\/$ is regular,

2642: $$

2643: \lox{p^\prime}{X}{w^{1-p^\prime}}=\lox{p}{X}{w}^\prime=\Gamma^{p\prime}_X(\tilde{w})

2644: =\Lambda^{p^\prime}_X(\tilde{w}),

2645: $$

2646: where

2647: $\tilde{w}(t)=t^{p^\prime}W^{-p^\prime}(t)w(t),\;t>0\/$.

2648:

2649: \item[(iii)] If  $w\/$ is increasing, $\lox{p}{X}{w}^\prime=\lox{p^\prime}{X}{w^{1-p^\prime}}\/$ with

2650: equality of norms.

2651: \end{enumerate}

2652: \eteo

2653:

2654: \bdem

2655:  (i) We have observed in  Remark~\ref{obs: II.4.8} that we always have

2656: $\lox{p^\prime}{X}{w^{1-p^\prime}}\subset\lox{p}{X}{w}^\prime=\Gamma^{p^\prime}_X(\tilde{w})\/$.

2657: Thus, the equality of these two spaces is equivalent to the embedding

2658: $$

2659: \Gamma^{p^\prime}_X(\tilde{w})\subset\lox{p^\prime}{X}{w^{1-p^\prime}},

2660: $$

2661: which   is also equivalent to

2662: the opposite inequality for the   Hardy operator,

2663: $$

2664: \|g\|_{L^{p^\prime}(w^{1-p^\prime})}\le C\|Ag\|_{L^{p^\prime}(\tilde{w})},\qquad g\d.

2665: $$

2666: A necessary  and sufficient condition for it  can be found in \cite{CPSS}  or in \cite{Ne1}. This

2667: condition is

2668: $$

2669: \int_0^r w^{1-p^\prime}(t)\,dt\le C\bigg(\widetilde{W}(r)+r^{p^\prime}\int_r^\infty

2670: t^{-p^\prime}\tilde{w}(t)\,dt\bigg),

2671: $$

2672: which is (\ref{equiv}), since

2673: $\displaystyle\widetilde{W}(r)=\int_0^r t^{p^\prime}W^{-p^\prime}(t)w(t)\,dt\/$,     and

2674: $$

2675: r^{p^\prime}\int_r^\infty t^{-p^\prime}\tilde{w}(t)\,dt=r^{p^\prime}\int_r^\infty

2676: W^{-p^\prime}(t)w(t)\,dt={1\over p^\prime-1}r^{p^\prime}W^{1-p^\prime}(r).

2677: $$

2678:

2679: (ii) If  $w\/$ is regular we have that

2680: $$

2681: w^{1-p^\prime}(t)=w(t)w^{-p^\prime}(t)\le Cw(t)t^{p^\prime}W^{-p^\prime}(t)=C\tilde{w}(t).

2682: $$

2683: Hence,

2684: $\Lambda^{p^\prime}(\tilde{w})\subset\Lambda^{p^\prime}(w^{1-p^\prime})\/$  and by

2685: Remark~\ref{obs: II.4.8} (iii), we conclude the equality of the four spaces in the statement.

2686:

2687: (iii) If $f\in\M(X)\/$ and  $w\/$ is increasing, the function $g_0=(f^\ast w^{-1})^{p^\prime-1}\/$

2688: is decreasing in  $\R^+\/$ and we have that

2689: \begin{eqnarray*}

2690: \|f\|_{\Lambda^p(w)^\prime}&=&\sup_{g\d}{\displaystyle\int_0^{\infty} f^\ast(t)

2691: g(t)\,dt\over\bigg(\displaystyle\int_0^{\infty} g^p(t)

2692: w(t)\,dt\bigg)^{1/p}}\\

2693: &\ge&{\displaystyle\int_0^{\infty} f^\ast(t)

2694: g_0(t)\,dt\over\bigg(\displaystyle\int_0^{\infty} g_0^p(t) w(t)\,dt\bigg)^{1/p}}

2695: =\|f\|_{\Lambda^{p^\prime}(w^{1-p^\prime})}.

2696: \end{eqnarray*}

2697: Since the opposite inequality is always true  (Remark~\ref{obs: II.4.8}  (iii)), we get

2698: the result.$\qquad\qed$

2699: \edem

2700:

2701:

2702: The following result describes the biassociate space of  $\Lambda^p\/$ in the case $p\le1\/$. See

2703: also \cite{CPSS}.

2704:

2705: \bteo

2706: \label{te: II.4.15}

2707: Let $X\/$ be nonatomic and  let  $w\/$ be a weight in  $\R^+\/$. Let $0<p\le1\/$ and $W_p\/$

2708: the greatest concave minorant of  $W^{1/p}\/$ (Remark~\ref{obs: II.4.8} (iv)). Then there exists a decreasing

2709: weight

2710: $\tilde{w}_p\/$ with  ${1\over2}\widetilde{W}_p\le W_p\le\widetilde{W}_p\/$ and such that

2711: $\lox{p}{X}{w}^{\prime\prime}=\lox{1}{X}{\tilde{w}_p}\/$. Moreover,

2712: $$

2713: {1\over2}\|\cdot\|_{\lox{1}{X}{\tilde{w}_p}}\le\|\cdot\|_{\lox{p}{X}

2714: {w}^{\prime\prime}}\le\|\cdot\|_{\lox{1}{X}{\tilde{w}_p}}.

2715: $$

2716: \eteo

2717:

2718: \bdem

2719: Let us note that  $\|g\|_{ \Lambda^p(w)^\prime }=\sup_{t>0}W^{-1/p}(t)\displaystyle\int_0^t

2720: g^\ast(s)\,ds\/$ (see Theorem~\ref{te: II.4.7}) and this norm is less than or equal to

2721: $1\/$, if and only if   $\displaystyle\int_0^t g^\ast(s)\,ds\le W^{1/p}(t),\;t>0\/$. By Lemma~\ref{le:

2722: II.2.4} and  Proposition~\ref{pro: II.2.5}, we obtain, for $f\in\M(X)\/$,

2723: \begin{eqnarray*}

2724: \|f\|_{ \Lambda^p(w)^{\prime\prime} }&=&\sup\bigg\{\int_0^\infty f^\ast(t) g^\ast(t)\,dt\,:\,

2725: \|g\|_{\Lambda^p(w)^\prime}\le1\bigg\}\\ &=&\sup\bigg\{\int_0^\infty f^\ast(t)

2726: \tilde{w}(t)\,dt\,:\,\tilde{w}\d,\;\int_0^t \tilde{w}(s)\,ds\le W^{1/p}(t),\;\forall t>0\bigg\}\\

2727: &=&\sup\bigg\{\int_0^\infty\widetilde{W}\big(\lambda_f(t)\big)\,dt\,:\,\widetilde{W}\in\Upsilon

2728: \bigg\},

2729: \end{eqnarray*}

2730: where

2731: $\Upsilon=\big\{\widetilde{W}\,:\,\tilde{w}\d,\,\widetilde{W}\le W^{1/p}\big\}\/$. Every function

2732: in

2733: $\Upsilon\/$ is quasi-concave and it is majorized by   $W^{1/p}\/$. Hence,

2734: $\widetilde{W}\le W_p\/$ for every $\widetilde{W}\in\Upsilon\/$.  On the other hand $W_p\/$  is

2735: quasi-concave and there exists a concave function

2736: $\widetilde{W}_p(t)=\displaystyle\int_0^t\tilde{w}_p(s)\,ds,\,t>0,\/$ with ${1\over2}\widetilde{W}_p\le

2737: W_p\le\widetilde{W}_p\/$ (see \cite{BS}). In particular $\tilde{w}_p\d\/$ and

2738: ${1\over2}\widetilde{W}_p\in\Upsilon\/$. It follows that

2739: \begin{eqnarray*}

2740: {1\over2}\|f\|_{\lox{1}{X}{\tilde{w}_p}}&=&\int_0^\infty{1\over2}\widetilde{W}_p

2741: \big(\lambda_f(t)\big)\,dt\le\|f\|_{

2742: \Lambda^p(w)^{\prime\prime}

2743: }\\

2744: &\le&\int_0^\infty\widetilde{W}_p\big(\lambda_f(t)\big)\,dt=\|f\|_{\lox{1}{X}{\tilde{w}_p}}.

2745: \qquad\qed

2746: \end{eqnarray*}

2747: \edem

2748:

2749: It is convenient now to recall  the definition of the discrete Hardy operator  $A_d\/$

2750: that has already appeared in (\ref{dischardy}). This operator acts on sequences $f=\big(f(n)\big)_n\/$  by

2751: $$

2752: A_d f(n)={1\over n+1}\sum_{k=0}^n f(k),\qquad n=0,1,2,\dots

2753: $$

2754: We can also describe the associate space when  $X\/$ is  resonant  totally atomic.  If

2755: $\mu(X)<\infty,\;\lox{p,q}{X}{w}=L^\infty(X)\/$ (with equivalent norms) and in this case it is not

2756: of interest. If

2757: $\mu(X)=\infty\/$ there are only, up to isomorphisms, two spaces: $d(\Omega,p)\/$ and

2758: $d^\infty(\Omega,p)\/$ (Example~ \ref{Eje2.3} (iv)) and the most interesting case is when

2759: $\Omega\notin\ell^1\/$. Up to now, the space

2760: $d(\Omega,p)^\prime\/$ had been identified only in some special cases. In \cite{Al} it is solved

2761: for

2762: $p\ge1\/$ with $\Omega\/$  decreasing and regular. In \cite{Po} and \cite{NO} the case

2763: $0<p<1\/$ and

2764: $\Omega\d\/$ is studied, while  in \cite{AEP} the associate is described under the condition

2765: $\Omega\u\/$  unbounded (this seems to be the unique reference where the sequence

2766: $\Omega\/$ is not decreasing). Here, we shall identify these   spaces in the most general

2767: case:

2768:

2769: \bteo

2770: \label{te: II.4.16}

2771: Let $W_n=\sum_{k=0}^n\Omega_k,\;n=0,1,2,\dots\/$ Then, for every

2772: $f=\big(f(n)\big)_{n=0}^\infty\subset\C\/$,

2773: \begin{enumerate}

2774: \item[(i)] If $0<p\le1\/$,

2775: $$

2776: \|f\|_{d(\Omega,p)^\prime}=\sup_{n\ge0}\,W_n^{-1/p}\sum_{k=0}^n f^\ast(k).

2777: $$

2778:

2779:

2780: \item[(ii)] If $1<p<\infty\/$ and $\Omega\notin\ell^1\/$,

2781: $$

2782: C_1 \|f\|_{d(\Omega,p)^\prime}\le\bigg(\sum_{n=0}^\infty\big(A_d

2783: f^\ast(n)\big)^{p^\prime}\widetilde{\Omega}_n\bigg)^{1/p^\prime}\le C_2

2784: \|f\|_{d(\Omega,p)^\prime},

2785: $$

2786: with

2787: $\widetilde{\Omega}_0=W_0^{1-p^\prime},\;\widetilde{\Omega}_n=(n+1)^{p^\prime}

2788: \big(W_{n-1}^{1-p^\prime}-W_n^{1-p^\prime}\big),\;n\ge1\/$,

2789: and the constants

2790: $C_1$ and $C_2\/$ depending only on  $p\/$.

2791:

2792: \item[(iii)] If $0<p<\infty\/$,

2793: $$

2794: \|f\|_{d^\infty(\Omega,p)^\prime}=\sum_{n=0}^\infty f^\ast(n)W_n^{-1/p}=\|f\|_{d(W^{-1/p},1)}.

2795: $$

2796: \end{enumerate}

2797: \eteo

2798:

2799: \bdem

2800: By definition

2801: $$

2802: \|f\|_{d(\Omega,p)^\prime}=\sup_g{\sum_{n=0}^\infty

2803: f(n)g(n)\over\|g\|_{d(\Omega,p)}}=\sup_{g\d}{\sum_{n=0}^\infty

2804: f^\ast(n)g(n)\over\bigg(\sum_{n=0}^\infty g(n)^p\Omega_n\bigg)^{1/p}}.

2805: $$

2806: Then  (i) can be directly deduced  applying the first part of  Theorem~\ref{te: I.6.11}. Applying the second

2807: expression of the same theorem in  (ii), we obtain, for $p>1\/$,

2808: $$

2809: \|f\|_{d(\Omega,p)^\prime}^{p^\prime}\approx\int_0^\infty\bigg({\widetilde{V}(t)\over\widetilde{W}(t)}

2810: \bigg)^{p^\prime}\tilde{w}(t)\,dt,

2811: $$

2812: where, $\tilde{v}=\sum_{n=0}^\infty f^\ast(n)\chi_{[n,n+1)},$ $\tilde{w}=\sum_{n=0}^\infty

2813: \Omega_n\chi_{[n,n+1)},$ $\widetilde{V}(t)=\displaystyle\int_0^t\tilde{v}(s)\,ds,$ $\widetilde{W}(t)=

2814: \displaystyle\int_0^t

2815: \tilde{w}(s)\,ds,\;t>0\/$.

2816: Being  $f^\ast\/$ decreasing we have, for $n\ge1$ and  $t\in[n,n+1)\/$,

2817: $$

2818: {1\over

2819: 2}\big(f^\ast(0)+\dots+f^\ast(n)\big)\le\widetilde{V}(t)\le\big(f^\ast(0)+\dots+f^\ast(n)\big).

2820: $$

2821: On the other hand

2822: \begin{eqnarray*}

2823: \int_n^{n+1}{\tilde{w}(t)\over\widetilde{W}^{p^\prime}(t)}\,dt&=&\int_0^1\Omega_n\big(\Omega_0+\dots+

2824: \Omega_{n-1}+\Omega_n

2825: t\big)^{-p^\prime}\\

2826: &=&{W_{n-1}^{1-p^\prime}-W_n^{1-p^\prime}\over

2827: p^\prime-1}=C_p{\widetilde{\Omega}_n\over(n+1)^{p^\prime}}.

2828: \end{eqnarray*}

2829: Hence,

2830: \begin{eqnarray*}

2831: \|f\|_{d(\Omega,p)^\prime}^{p^\prime}&\approx&\int_0^1\bigg({\widetilde{V}(t)\over\widetilde{W}(t)}

2832: \bigg)^{p^\prime}\tilde{w}(t)\,dt+\sum_{n=1}^\infty\int_n^{n+1}\bigg({\widetilde{V}(t)\over\widetilde{W}(t)}

2833: \bigg)^{p^\prime}\tilde{w}(t)\,dt\\

2834: &\approx& f^\ast(0)^{p^\prime}\Omega_0^{1-p^\prime}+\sum_{n=1}^\infty

2835: \bigg(\sum_{k=0}^n f^\ast(k)\bigg)^{p^\prime}{\widetilde{\Omega}_n\over(n+1)^{p^\prime}}\\

2836: &=&\sum_{n=0}^\infty\big(A_d f^\ast(n)\big)^{p^\prime}\widetilde{\Omega}_n.

2837: \end{eqnarray*}

2838:

2839: It only remains to prove  (iii). Note that

2840: \begin{eqnarray*}

2841: \|f\|_{d^\infty(\Omega,p)^\prime}&=&\sup\bigg\{\sum_{n=0}^\infty

2842: f^\ast(n)g^\ast(n)\,:\,\|g\|_{d^\infty(\Omega,p)}\le1\bigg\}\\ &=&\sup\bigg\{\sum_{n=0}^\infty

2843: f^\ast(n)g^\ast(n)\,:\,g^\ast(k)\le W_k^{-1/p},\;\forall k\ge0\bigg\},

2844: \end{eqnarray*}

2845: and this supremum is attained at the sequence

2846: $W^{-1/p}=\big(W_n^{-1/p}\big)_n\/$. Hence,

2847: $$

2848: \|f\|_{d^\infty(\Omega,p)^\prime}=\sum_{n=0}^\infty f^\ast(n)W_n^{-1/p}.\qquad\qed

2849: $$

2850: \edem

2851:

2852: \bobs

2853: \label{obs: II.4.17}

2854:

2855: (i) Both  $d(\Omega,p)\/$ and $d^\infty(\Omega,p)\/$  are continuously embedded in

2856: $\ell^\infty\/$ and, thus,  $\ell^1\subset d(\Omega,p)^\prime\/$ (and also $\ell^1\subset

2857: d^\infty(\Omega,p)^\prime\/$) for $0<p<\infty\/$. In particular, the associate space of these

2858: spaces is never trivial.

2859:

2860: (ii) The argument used in Remark~\ref{obs: II.4.8} (iii) is also useful here and,

2861: for

2862: $1<p<\infty\/$,

2863: $$

2864: d(\Omega^{1-p^\prime},p^\prime)\subset d(\Omega,p)^\prime,

2865: $$

2866: and

2867: $\|f\|_{d(\Omega,p)^\prime}\le\|f\|_{d(\Omega^{1-p^\prime},p^\prime)}\/$ for every sequence

2868: $f\/$. If $\Omega\u\/$ we can  use the argument in the proof of

2869: Theorem~\ref{te: II.4.14} (iii) and conclude (as in \cite{AEP})

2870: that the two previous spaces (and their norms) are equal.

2871: \eobs

2872:

2873: In \cite{Al}, Allen proves, in the case $\Omega\d,\;p>1\/$, that

2874: $d(\Omega,p)^\prime=d(\Omega^{1-p^\prime},p^\prime)\/$ if $\Omega\/$ is regular. The following

2875: theorem extends this result to the general case.

2876:

2877:

2878: \bteo

2879: \label{te: II.4.18}

2880:  Let $1<p<\infty\/$ and $\Omega\notin\ell^1\/$. Then

2881: $d(\Omega,p)^\prime=d(\Omega^{1-p^\prime},p^\prime)\/$ if and only if

2882: $$

2883: \sum_{k=0}^n\Omega_k^{1-p^\prime}\le C\sum_{k=0}^n\big(A_d\Omega(k)\big)^{1-p^\prime},\qquad

2884: n=0,1,2,\dots

2885: $$

2886: In particular

2887: $d(\Omega,p)^\prime=d(\Omega^{1-p^\prime},p^\prime)\/$ if $\Omega\/$ is regular.

2888: \eteo

2889:

2890: \bdem

2891: The equality of the two   spaces is equivalent (see Remark~\ref{obs: II.4.17}) to the

2892: embedding

2893: $d(\Omega,p)^\prime\subset d(\Omega^{1-p^\prime},p^\prime)\/$ that, by Theorem~\ref{te:

2894: II.4.16} (ii), holds if and only if the inequality

2895: \bequ

2896: \label{II.4.19}

2897: \bigg(\sum_{n=0}^\infty g(n)^{p^\prime}\Omega_n^{1-p^\prime}\bigg)^{1/p^\prime}\le

2898: C\bigg(\sum_{n=0}^\infty

2899: \big(A_dg(n)\big)^{p^\prime}\widetilde{\Omega}_n\bigg)^{1/p^\prime}

2900: \eequ

2901: holds for every positive and decreasing sequence $g=\big(g(n)\big)_n\/$. Here

2902: $\widetilde{\Omega}\/$ is the sequence defined by $\widetilde{\Omega}_0=\Omega_0^{1-p^\prime}\/$,

2903: $$

2904: {\widetilde{\Omega}_n\over(n+1)^{p^\prime}}=\bigg(\sum_{k=0}^{n-1}\Omega_k\bigg)^{1-p^\prime}-

2905: \bigg(\sum_{k=0}^n\Omega_k\bigg)^{1-p^\prime},\qquad n=1,2,3,\dots

2906: $$

2907: The class of positive and decreasing sequences  is regular (Definition~\ref{def: I.2.4}) in $\N^\ast\/$ and

2908: we can use Theorem~\ref{twoop} with

2909: $T_0=A_d\/$ (which is order continuous, linear and positive) to characterize  (\ref{II.4.19}).

2910: The condition  is obtained \lq\lq applying" the inequality to the sequences

2911: $(1,1,\dots,1,0,0,\dots)\/$ (decreasing characteristic functions in $\N^\ast\/$) and it is

2912: equivalent to

2913: \begin{eqnarray*}

2914: \sum_{k=0}^n\Omega_k^{1-p^\prime}\lesssim

2915: \Omega_0^{1-p^\prime}&+&\sum_{k=1}^n(k+1)^{p^\prime}\big(W_{k-1}^{1-p^\prime}-W_k^{1-p^\prime}\big)\\

2916: &+&(n+1)^{p^\prime}\sum_{k=n+1}^\infty\big(W_{k-1}^{1-p^\prime}-W_k^{1-p^\prime}\big),

2917: \end{eqnarray*}

2918: $n=1,2,\dots\/$, with $W_k=\sum_{j=0}^k\Omega_j\/$. The second term is equivalent, for

2919: $n\ge1,\/$ to the expression

2920: \begin{eqnarray*}

2921: \Omega_0^{1-p^\prime}+\sum_{k=1}^n(k+1)^{p^\prime}(W_{k-1}^{1-p^\prime}-W_k^{1-p^\prime})

2922: +(n+1)^{p^\prime}W_n^{1-p^\prime}&\approx&\sum_{k=0}^{n-1}\bigg({W_k\over

2923: k+1}\bigg)^{1-p^\prime}\\ &\approx&\sum_{k=0}^n\bigg({W_k\over k+1}\bigg)^{1-p^\prime}.

2924: \end{eqnarray*}

2925: And  the condition of the theorem is obtained.

2926:

2927: The second assert is immediate, since every regular weight trivially satisfies  the given

2928: condition.$\quad\qed$

2929: \edem

2930:

2931: In \cite{NO}, M. Nawrocki and A. Orty\'{n}ski prove, for the case $\Omega\/$ decreasing, that the

2932: associate of

2933: $d(\Omega,p),\;p\le1\/$, is $\ell^\infty\/$ if the condition

2934: $\sum_{k=0}^n\Omega_k\ge C(n+1)^p\/$ holds. We now extend this result proving that this condition

2935: also works for the  general case. Moreover, this condition is also necessary.

2936:

2937: \bteo

2938: \label{te: II.4.20}

2939: Let $0<p\le1\/$. Then  $d(\Omega,p)^\prime\subset\ell^\infty\/$,  and these two spaces are equal, if

2940: and only if there exists

2941: $C>0\/$ such that

2942: $$

2943: \sum_{k=0}^n\Omega_k\ge C(n+1)^p,\qquad n=0,1,2,\dots

2944: $$

2945: \eteo

2946:

2947: \bdem

2948: By Theorem~\ref{te: II.4.16} (i), we have that, with $W_n=\sum_{k=0}^n\Omega_k\/$,

2949: $$

2950: \|f\|_{d(\Omega,p)^\prime}\ge W_0^{-1/p}f^\ast(0)=W_0^{-1/p}\|f\|_{\ell^\infty}.

2951: $$

2952: Therefore,

2953: $d(\Omega,p)^\prime\subset\ell^\infty\/$.

2954:

2955: If the inequality of the theorem holds,

2956: $$

2957: W_n^{-1/p}\sum_{k=0}^nf^\ast(k)={n+1\over W_n^{1/p}}A_d f^\ast(n)\le C^{-1/p}A_d f^\ast(n)\le

2958: C^{-1/p}\|f\|_{\ell^\infty},

2959: $$

2960: for all  $n=0,1,2,\dots\/$

2961: Hence,

2962: $\|f\|_{d(\Omega,p)^\prime}=\sup_n W_n^{-1/p}\sum_{k=0}^nf^\ast(k)\le C^{-1/p}\|f\|_{\ell^\infty}\/$

2963: and we have that  $\ell^\infty= d(\Omega,p)^\prime\/$.

2964:

2965: Conversely, if $\ell^\infty\subset d(\Omega,p)^\prime\/$, then $1\in d(\Omega,p)^\prime\/$ and

2966: $\sup_n{n+1\over W_n^{1/p}}=\|1\|_{d(\Omega,p)^\prime}$

2967: $=C<\infty\/$, which  implies the result.$\qquad\qed$

2968: \edem

2969:

2970:

2971: The following result describes the biassociate of  $d(\Omega,p)\/$ in the case $p\le1\/$ in an

2972: analogous way as we did in the nonatomic case  (Theorem~\ref{te: II.4.15}). We omit here the

2973: proof.

2974:

2975:

2976: \bteo

2977: \label{te: II.4.21}

2978: Let $0<p\le1\/$ and let us denote by $W_n=\sum_{k=0}^n\Omega_k,\;n=0,1,2,\dots\/$

2979: Then there exists a decreasing sequence

2980: $\widetilde{\Omega}=(\widetilde{\Omega}_n)_n\/$ satisfying

2981: $$

2982: {1\over2}\sum_{k=0}^n\widetilde{\Omega}_k\le\inf_{m\ge0}\max\bigg(1,\,{n+1\over

2983: m+1}\bigg)W^{1/p}_m\le\sum_{k=0}^n\widetilde{\Omega}_k,

2984: $$

2985: and such that

2986: $d(\Omega,p)^{\prime\prime}=d\big(\widetilde{\Omega},1\big)\/$ with equivalent norms. More

2987: precisely,

2988: $$

2989: {1\over2}\|\cdot\|_{d(\widetilde{\Omega},1)}\le\|\cdot\|_{d(\Omega,p)^{\prime\prime}}

2990: \le\|\cdot\|_{d(\widetilde{\Omega},1)}\;.

2991: $$

2992: \eteo

2993:

2994: Let us study now the topologic dual  and its connection with the associate space.

2995:

2996: \bdefi

2997: If $(E,\|\cdot\|)\/$ is a quasi-normed space,  we define the dual\index{dual space}

2998: $E^\ast\/$ in the usual way:

2999: $$

3000: E^\ast=\{u:E\to\C\,:\,u\hbox{\rm \ linear and continuous}\}.

3001: $$

3002: If $u\in E^\ast\/$ we denote by

3003: $\|u\|=\|u\|_{E^\ast}=\sup\big\{|u(f)|\,:\,\|f\|\le1,\;f\in E\big\}\in[0,\infty)\/$.

3004: \edefi

3005:

3006: The  dual $E^\ast\/$  of a quasi-normed space $\big(E,\|\cdot\|_E\big)\/$  is a Banach space.

3007: If $E^\ast\/$ separates points, every $f\in E\/$ can be identified with the linear and continuous

3008: form

3009: $\tilde{f}:E^\ast\to\C\/$  such that $\tilde{f}(u)=u(f),\;u\in E^\ast\/$. We have then a continuous

3010: injection,

3011: $$

3012: \big(E,\|\cdot\|_E\big)\hookrightarrow\big(E^{\ast\ast},\|\cdot\|_{E^{\ast\ast}}\big),

3013: $$

3014: and the constant of this embedding  is less than or equal to $1\/$ since, for $f\in E\/$,

3015: $$

3016: \|f\|_{E^{\ast\ast}}=\|\tilde{f}\|_{(E^\ast,\|\cdot\|_{E^\ast})^\ast}=\sup_{u\in

3017: E^\ast}{\big|u(f)\big|\over\|u\|_{E^\ast} }\le\|f\|_E.

3018: $$

3019:

3020: The  Mackey topology\index{Mackey topology} in   $E\/$ associated to the dual pair  $(E,E^\ast)\/$, is

3021: defined as having as a local basis the convex envelope of the balls in  $E\/$ (see \cite{Ko}). It is the

3022: finest locally convex topology in

3023: $E\/$ having

3024: $E^\ast\/$ as its topological dual. The completion

3025: $\widetilde{E}\/$ with this topology is called   Mackey completion or also

3026: Banach envelope of

3027: $E\/$. In a quasi-normed space whose dual separates points, this topology corresponds to the one

3028: induced in     $E\/$ by

3029: $\big(E^{\ast\ast},\|\cdot\|_{E^{\ast\ast}}\big)\/$ and the  Mackey completion  is then the closure

3030: of $E\/$ in

3031: $\big(E^{\ast\ast},\|\cdot\|_{E^{\ast\ast}}\big)\/$.

3032:

3033: In our case $E=\Lambda\/$ is a quasi-normed  Lorentz space. If $f\in\Lambda^\prime\/$

3034: the application $u_f:\Lambda\to\C\/$ such that $u_f(g)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ is obviously

3035: linear and continuous with norm equal to    $\|f\|_{\Lambda^\prime}\/$. Thus,  $\Lambda^\prime\/$

3036: is isometrically isomorphic to a subspace of

3037: $\Lambda^\ast\/$ and, in fact, we shall identify the functions of  $\Lambda^\prime\/$

3038: as linear and continuous forms in  $\Lambda\/$:

3039: $$

3040: \Lambda^\prime\subset\Lambda^\ast.

3041: $$

3042: On the other hand, $\Lambda\/$ is continuously embedded in

3043: $\big(\Lambda^{\prime\prime},\|\cdot\|_{\Lambda^{\prime\prime}}\big)\/$ (Remark~\ref{Obs:

3044: II.4.2}), and therefore (if $\Lambda^\ast\/$ separates points) we are working with three independent

3045: topologies in

3046: $\Lambda\/$ and two embeddings with constant $1\/$:

3047: \begin{eqnarray*}

3048: \big(\Lambda,\|\cdot\|_{\Lambda}\big)&\hookrightarrow&\big(\Lambda^{\ast\ast},\|

3049: .\|_{\Lambda^{\ast\ast}}\big),\\

3050: \big(\Lambda,\|\cdot\|_{\Lambda}\big)&\hookrightarrow&\big(\Lambda^{\prime\prime},

3051: \|\cdot\|_{\Lambda^{\prime\prime}}\big).

3052: \end{eqnarray*}

3053:

3054: \bpro

3055: \label{pro: II.4.23}

3056: If $\Lambda\/$ is a quasi-normed  Lorentz space whose dual separates points,

3057: $$

3058: \big(\Lambda,\|\cdot\|_{\Lambda}\big)\hookrightarrow

3059: \big(\Lambda,\|\cdot\|_{\Lambda^{\ast\ast}}\big)\hookrightarrow

3060: \big(\Lambda,\|\cdot\|_{\Lambda^{\prime\prime}}\big),

3061: $$

3062: and  $\|f\|_{\Lambda^{\prime\prime}}\le\|f\|_{\Lambda^{\ast\ast}}

3063: \le\|f\|_{\Lambda},\;f\in\Lambda\/$.

3064: If in addition, $\Lambda^\prime=\Lambda^\ast\/$, then

3065: $\Lambda^{\prime\prime}\/$ is isometrically identified with a subspace of

3066: $\Lambda^{\ast\ast}\/$. In particular,

3067: $\|f\|_{\Lambda^{\prime\prime}}=\|f\|_{\Lambda^{\ast\ast}},$ for every $ f\in\Lambda,\/$  in this case.

3068: \epro

3069:

3070: \bdem

3071: If $f\in\Lambda\/$, then $f\in\Lambda^{\ast\ast}\/$ and

3072: $$

3073: \|f\|_{\Lambda}\ge\|f\|_{\Lambda^{\ast\ast}}=

3074: \sup_{u\in\Lambda^\ast}{|u(f)|\over\|u\|_{\Lambda^\ast}}

3075: \ge\sup_{g\in\Lambda^\prime}{\int|fg|\over\|g\|_{\Lambda^\prime}}=\|f\|_{\Lambda^{\prime\prime}}

3076: .

3077: $$

3078:

3079: If $\Lambda^\prime=\Lambda^\ast\/$, every continuous and linear form $u\in\Lambda^\ast\/$ is of the

3080: form

3081: $u(f)=u_g(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ with $g\in\Lambda^\prime\/$. Also, to every function

3082: $f\in\Lambda^{\prime\prime}\/$ we can associate  the  linear form

3083: $\tilde{f}:\Lambda^\prime=\Lambda^\ast\to\C\/$ defined by

3084: $\tilde{f}(g)=\tilde{f}(u_g)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ and with norm

3085: $$

3086: \|\tilde{f}\|_{\Lambda^{\ast\ast}}=\sup_{u_g\in\Lambda^\ast}

3087: {\big|\tilde{f}(u_g)\big|\over\|u_g\|_{\Lambda^\ast}}=

3088: \sup_{g\in\Lambda^{\prime}}{\int|fg|\over\|g\|_{\Lambda^{\prime}}}

3089: =\|f\|_{\Lambda^{\prime\prime}}.\quad\qed

3090: $$

3091: \edem

3092:

3093: We shall now study, among other, the two following questions: (i) To characterize the weights  $w\/$

3094: for which

3095: $\Lambda(w)^\ast=\{0\}\/$. (ii)  When is

3096: $\Lambda^\prime=\Lambda^\ast\/$? As in the case of the Lebesgue space  $L^p\/$,

3097: Radon-Nikodym  theorem is the key for the following   result.

3098:

3099: \bpro

3100: \label{pro: II.4.24}

3101: Let $\Lambda\/$  be a quasi-normed  Lorentz space on  $X\/$. If

3102: $u\in\Lambda^\ast\/$, there exists a unique function

3103: $g\in\Lambda^\prime\/$ with $\|g\|_{\Lambda^\prime}\le\|u\|\/$, satisfying

3104: $$

3105: u(f)=\int_X f(x)g(x)\,d\mu(x),\quad f\in\LL.

3106: $$

3107: \epro

3108:

3109: \bdem

3110: Let us first assume that  $\mu(X)<\infty\/$ (in this case $\S\subset

3111: L^\infty\subset\Lambda\/$). For every  $E\subset X\/$ let

3112: $$

3113: \sigma(E)=u(\chi_E)\in\C.

3114: $$

3115: Then $\sigma\/$ is a complex measure in  $X\/$ (the fact that it is

3116:  $\sigma$-additive is a consequence of the fact that  $\mu(X)<\infty\/$), absolutely continuous with

3117: respect to the $\sigma$-finite measure    $\mu\/$ of $X\/$, since  $\mu(E)=0\/$

3118: implies

3119: $\|\chi_E\|_{\Lambda}=0\/$ and thus  $u(\chi_E)=0\/$. By Radon-Nikodym theorem, there exists a

3120: function

3121: $g\in L^1(X)\/$ so that

3122: $\sigma(E)=\displaystyle\int_E g(x)\,d\mu(x)\/$  for every measurable set  $E\subset X\/$. In particular

3123: (since

3124: $u\/$ is linear) $u(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ for $f\in\S\subset\Lambda\/$. This also holds

3125: if

3126: $f\in L^\infty(X)\subset\Lambda\/$ since there exists a sequence

3127: $(s_n)_n\subset\S\subset\Lambda\/$ converging to  $f\/$ in $\Lambda\/$  and also uniformly (see the

3128: proof of Theorem~\ref{te: II.3.12}).

3129:

3130: If $\mu(X)=\infty\/$, the previous argument is true in any subset

3131: $Y\subset X\/$ of finite measure: we define a complex measure $\sigma_Y\/$ in $Y\/$ and we obtain

3132: the existence of a function

3133: $g_Y\in L^1(Y)\/$ with $u(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ for every $f\in L^\infty(X)\/$

3134: supported in

3135: $Y\/$. If $Y_1,Y_2\/$ are two sets of finite measure,  necessarily   $\displaystyle\int_E

3136: g_{Y_1}(x)\,d\mu(x)=\displaystyle\int_E g_{Y_2}(x)\,d\mu(x)\/$ for every measurable set  $E\subset Y_1\cap

3137: Y_2\/$ and it follows that

3138: $g_{Y_1}(x)=g_{Y_2}(x)\/$ a.e.

3139: $x\in Y_1\cap Y_2\/$. Hence, since $X$ is $\sigma$-finite we can assure the existence of a function

3140: $g\in L^1_{\hbox{\decp loc}}(X)\/$ such that  $u(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ for every $f\in

3141: L^\infty_0(X)\/$.

3142:

3143: To show that  $\|g\|_{\Lambda^\prime}\le\|u\|\/$, let $\alpha\in\M(X)\/$ with $|\alpha|=1,\;\alpha

3144: g=|g|\/$ and let also  $(X_n)_n\/$ be an increasing sequence of sets of finite measure with

3145: $\bigcup_n X_n=X\/$. If $f\in\Lambda\/$, we consider  $f_n=|f|\chi_{\big\{|f|\le n\big\}\cap X_n}\/$.

3146: Then $0\le f_n\le f_{n+1}\to|f|\/$ and every  $f_n\/$ is bounded and with support in a set of

3147: finite measure. Then,

3148: \begin{eqnarray*}

3149: \int_X|f(x)g(x)|\,d\mu(x)&=&\lim_n\bigg|\int_X\alpha(x) f_n(x) g(x)\,d\mu(x)\bigg|\\

3150: &=&\lim_n\big|u(\alpha

3151: f_n)\big|\le\|u\|\lim_n\|f_n\|_{\Lambda}=\|u\|\|f\|_{\Lambda},

3152: \end{eqnarray*}

3153: from which

3154: $g\in\Lambda^\prime\/$ and  $\|g\|_{\Lambda^\prime}\le\|u\|\/$.

3155:

3156: Finally, the uniqueness of  $g\/$ is also clear, since if  $g_1\/$  is another function with

3157: the same properties than $g\/$, we have that

3158: $\displaystyle\int_E g(x)\,d\mu(x)=\displaystyle\int_E g_1(x)\,d\mu(x)=u(\chi_E)\/$ for every measurable

3159: set  $E\subset X\/$ of finite measure, which implies that  $g=g_1.\quad\qed$

3160: \edem

3161:

3162:

3163: \bobs

3164:  If $\mu(X)=\infty\/$ and $w\in L^1\/$ every simple function, regardless of the measure of its support, is

3165: in

3166: $\Lambda_X(w)\/$. We can then define, for every linear form $u\in\Lambda^\prime\/$,

3167: $$

3168: \sigma(E)=u(\chi_E),\quad E\subset X\hbox{\ measurable}.

3169: $$

3170: but this set function, defined

3171: on the whole  $\sigma$-algebra of $X\/$,  is not $\sigma$-additive in general.

3172: To have this property, we need that, for every family of disjoint measurable sets

3173: $(E_n)_n\/$, the functions

3174: $\chi_{\bigcup_1^N E_n}\/$ have to converge to

3175: $\chi_{\bigcup_1^\infty E_n}\/$ in $\Lambda$. Contrary to what happens if $\mu(X)<\infty$, this is not true

3176: in general in this case. For example, in

3177: $\Lambda^1_{\R}\big(\chi_{(0,1)}\big)\/$ the sets $E_n=(n,n+1)\/$ give a counterexample.

3178:

3179: This explains why we need the restriction  \lq\lq support of finite measure" in the previous

3180: proposition.

3181: \eobs

3182:

3183: \bcor

3184:  If $\Lambda\/$  is a quasi-normed Lorentz space on

3185: $X,\;\Lambda^\prime=\{0\}\/$ if and only if every functional  $u\in\Lambda^\ast\/$ is zero on

3186: $\LL(X)\/$. In particular, if $X\/$ is resonant, $\Lambda_X^\ast\/$ separates points if and only if

3187: $\Lambda_X^\prime\neq\{0\}\/$.

3188: \ecor

3189:

3190: \bcor

3191: \label{cor: II.4.27}

3192: If $\Lambda\/$  is a quasi-normed Lorentz space  and  $f\in\Lambda\/$

3193: has  absolutely continuous norm, then

3194: $\|f\|_{\Lambda^{\ast\ast}}=\|f\|_{\Lambda^{\prime\prime}}\/$.

3195: \ecor

3196:

3197: \bdem

3198: Since  $f\/$ is  pointwise limit  of a sequence $(f_n)_n\/$ in $\LL\/$ with

3199: $|f_n|\le|f|\/$ and it has absolutely continuous norm, we have that  $f=\lim_n f_n\/$  in

3200:  $\Lambda\/$

3201: and, by continuity,  (Proposition~\ref{pro: II.4.23}), we also have the convergence in

3202: $\Lambda^{\prime\prime}\/$ and in

3203: $\Lambda^{\ast\ast}\/$. Hence, it is sufficient to prove the result assuming that  $f\in\LL\/$. By

3204: Proposition~\ref{pro: II.4.24}, for every  $u\in\Lambda^\ast\/$ there exists $g\in\Lambda^\prime\/$

3205: with

3206: $\|g\|_{\Lambda^{\prime}}\le\|u_g\|_{\Lambda^\ast}=\|u\|_{\Lambda^\ast}\/$ and such that

3207: $u(f)=u_g(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$. Then,

3208: $$

3209: \|f\|_{\Lambda^{\ast\ast}}=\sup_{u\in\Lambda^\ast}{|u(f)|

3210: \over\|u\|_{\Lambda^\ast}}=\sup_{g\in\Lambda^\prime}{|u_g(f)|

3211: \over\|u_g\|_{\Lambda^\ast}}=\sup_{g\in\Lambda^\prime}{\displaystyle\int|f(x)g(x)|\,d\mu(x)\over\|g\|_{\Lambda^\prime}}

3212: =\|f\|_{\Lambda^{\prime\prime}}.\quad\qed

3213: $$

3214: \edem

3215:

3216: We can give now a representation of the    dual of  $\Lambda^p_X\/$  for an arbitrary

3217: $\sigma$-finite measure space $X$.

3218:

3219: \bteo

3220: If $0<p<\infty\/$ and $W\in\Delta_2(X)\/$,

3221: $$

3222: \Lambda^p_X(w)^\ast=\Lambda^p_X(w)^\prime\oplus\Lambda^p_X(w)^s\,,

3223: $$

3224: where

3225: $$

3226: \Lambda^p_X(w)^s=\big\{u\in\Lambda^p(w)^\ast\,:\,|u(f)|\le C\lim_{t\to\infty}f^\ast(t),\forall

3227: f\in\Lambda^p_X(w)\big\}.

3228: $$

3229: This last subspace is formed by functionals that are zero

3230: on the functions whose support is of finite measure and it is not zero only if

3231: $\mu(X)=\infty\/$ and

3232: $w\in L^1\/$.

3233: \eteo

3234:

3235: \bdem

3236: That every functional in  $\Lambda^p(w)^s\/$  is zero on functions with support of finite measure

3237: is immediate from the definition, since for such a function $f\/$ we have that

3238: $\lim_{t\to\infty}f^\ast(t)=0\/$. This also tells us that

3239: $\Lambda^p(w)^\prime\cap\Lambda^p(w)^s=\{0\}\/$ since if  $u=u_g\/$ is in the previous intersection

3240: we have that $\displaystyle\int_X f(x)g(x)\,d\mu(x)=u(f)=0\/$ for every  $f\in\Lambda^p\/$ with support of

3241: finite measure. Then it follows that $\displaystyle\int_X f(x)g(x)\,d\mu(x)=0\/$ for every

3242: $f\in\Lambda^p\/$ and hence

3243: $u=0\/$. Let us see now the decomposition $\Lambda^\ast=\Lambda^\prime+\Lambda^s\/$. To this end, let

3244: $u\in\Lambda^p(w)^\ast\/$. By Proposition~\ref{pro: II.4.24}, there exists $g\in\Lambda^p(w)^\prime\/$ such

3245: that the continuous linear functional

3246: $u_g(f)=\displaystyle\int_X f(x)g(x)\,d\mu(x)\/$ coincides with $u\/$ in $\LL\/$. If $f\in\Lambda^p\/$ and

3247: $Y=\{f\neq0\}\/$ has finite measure, then $u(f)=u_g(f)\/$ since $u$ and $u_g\/$ are continuous linear forms

3248: on

3249: $\Lambda^p_Y(w)\/$  and coincide in $\LL(Y)\/$, which is dense in  $\Lambda^p_Y(w)\/$ (Theorem~\ref{te:

3250: II.3.13}).  Let

3251: $u^s=u-u_g\/$. Then $u^s\/$ is zero on the functions of  $\Lambda^p\/$ supported in a set of finite

3252: measure. Therefore, if

3253: $f\/$ any function in

3254: $\Lambda^p\/$ and

3255: $a=\lim_{t\to\infty}f^\ast(t)\/$ and $f_n=f\chi_{\{|f|\le a+1/n\}},\;n=1,2,\dots,\/$ we have

3256: $u^s(f)=u^s(f_n)\/$ for every $n\/$ (since $f-f_n\/$  has support of finite measure). Thus,

3257: $$

3258: |u^s(f)|=|u^s(f_n)|\le\|u^s\|\|f_n\|_{\Lambda^p},\quad n=1,2,\dots

3259: $$

3260: We now have two cases: (i)

3261: $w\notin L^1\/$. Then $a=0\/$ and  $\Lambda^p(w)\/$ has absolutely continuous norm

3262: (Theorem~\ref{te: II.3.5}). Since $|f_n|\le|f|\/$ and  $|f_n|\to0\/$ a.e. we get that

3263: $\|f_n\|_{\Lambda^p}\to0\/$ and

3264: $|u^s(f)|=0=a\/$. (ii) $w\in L^1\/$. Then

3265: $\|f_n\|_{\Lambda^p}\le\|f_n\|_\infty\|w\|_1^{1/p}\le(a+1/n)\|w\|_1^{1/p}\/$ and we have that

3266: $|u^s(f)|\le\|u^s\|\|w\|_1^{1/p}a=Ca=C\lim_{t\to\infty}f^\ast(t)\/$. In any of these two cases,

3267: there exists $C\in(0,+\infty)\/$ (independent of $f\/$) such that  $|u^s(f)|\le

3268: C\lim_{t\to\infty}f^\ast(t)\/$. That is,  $u^s\in\Lambda^p(w)^s\/$ and we have

3269: $\Lambda^\ast=\Lambda^\prime\oplus\Lambda^s\/$.

3270:

3271: It remains to prove that  $\Lambda^p(w)^s\neq\{0\}\/$ if and only if  $\mu(X)=\infty\/$ and $w\in

3272: L^1\/$. If

3273: $\mu(X)<\infty\/$ or if  $w\notin L^1\/$ every function

3274: $f\in\Lambda^p(w)\/$ satisfies $\lim_{t\to\infty}f^\ast(t)=0\/$ and therefore, for

3275: $u\in\Lambda^s\/$, we get $u(f)=0,$ for every $ f\in\Lambda^p\/$. That is $u=0\/$ or equivalently

3276: $\Lambda^s=\{0\}\/$. If $\mu(X)=\infty\/$ and $w\in L^1\/$, the functional

3277: $p(f)=\lim_{t\to\infty}f^\ast(t),\;p:\Lambda^p\to[0,+\infty),\/$ is a seminorm:

3278: \begin{eqnarray*}

3279: (i)&\quad &p(\lambda f)=|\lambda p(f)|\quad\hbox{\rm (obvious)},\\ (ii)&\quad&

3280: p(f+g)=\lim_{t\to\infty}(f+g)^\ast(t)\le\lim_t\big(f^\ast(t/2)+g^\ast(t/2)\big)=p(f)+p(g).

3281: \end{eqnarray*}

3282: Since

3283: $p(1)=1\/$ (the constant function $1\/$ is in

3284: $\Lambda^p(w)\/$ in this case) $p\/$ is not zero and there exists a nonzero linear form  $u\/$

3285: on

3286: $\Lambda^p\/$ satisfying

3287: \bequ

3288: \label{II.4.29}

3289: |u(f)|\le p(f)=\lim_{t\to\infty}f^\ast(t),\quad\forall f\in\Lambda^p

3290: \eequ

3291: (see \cite{Ru2}

3292: for example). In particular, $u\/$ is continuous since

3293: $$

3294: \|f\|_{\Lambda^p(w)}=\bigg(\int_0^\infty(f^\ast(t))^p w(t)\,dt\bigg)^{1/p}\ge p(f)\|w\|_1^{1/p},

3295: \quad\forall

3296: f,

3297: $$

3298: and it follows that $|u(f)|\le p(f)\le C\|f\|_{\Lambda^p},$ for every $ f\/$. Finally,

3299: (\ref{II.4.29}) tells us that  $u\in\Lambda^s\/$ and hence $\Lambda^s\neq\{0\}.\quad\qed$

3300: \edem

3301:

3302: We shall state two immediate consequences of the last theorem that solve, in the case

3303: $\Lambda=\Lambda^p\/$, the two questions about duality we were looking for.

3304:

3305: \bcor

3306: \label{cor: II.4.30}

3307:  Let $0<p<\infty\/$ and  $W\in\Delta_2(X)\/$.

3308: \begin{enumerate}

3309: \item[(i)] If $\mu(X)<\infty\/$ or $w\notin L^1\/$,

3310: $$

3311: \Lambda^p_X(w)^\prime=\Lambda^p_X(w)^\ast.

3312: $$

3313: In particular

3314: $d(\Omega,p)^\ast=d(\Omega,p)^\prime\/$  if $\Omega\notin\ell^1\/$ and $d(\Omega,p)\/$ is

3315: quasi-normed.

3316:

3317: \item[(ii)] If $\mu(X)=\infty\/$ and $w\in L^1\/$,

3318: $$

3319: \Lambda^p_X(w)^\prime\varsubsetneq\Lambda^p_X(w)^\ast.

3320: $$

3321: In particular

3322: $\Lambda^p_X(w)^\ast\neq\{0\}\/$ in this case.

3323: \end{enumerate}

3324: \ecor

3325:

3326: As a consequence of this last result, Theorems~\ref{te: II.4.16}, \ref{te: II.4.18}, \ref{te:

3327: II.4.20} and Remark~\ref{obs: II.4.17},  remain true (in the case $d(\Omega,p)\/$

3328: quasi-normed,

3329: $\Omega\notin\ell^1\/$) if we substitute  $d(\Omega,p)^\prime\/$ by

3330: $d(\Omega,p)^\ast\/$.

3331:

3332: \bcor

3333:  Let $X\/$ be a nonatomic measure space and let $W\in\Delta_2\/$.

3334: \begin{enumerate}

3335: \item[(i)] If $0<p\le1,\;\Lambda^p_X(w)^\ast=\{0\}\/$ if and only if  the two following conditions hold:

3336: \begin{eqnarray*}

3337: i.1)&\quad&\mu(X)<\infty\hbox{\ or\ else\ }\mu(X)=\infty\hbox{\ and\ }w\notin L^1, \hbox{\ and }\\

3338: i.2)&\quad&\sup_{0<t<1}{t^p\over W(t)}=\infty.

3339: \end{eqnarray*}

3340:

3341:

3342: \item[(ii)] If $1<p<\infty,\;\Lambda^p_X(w)^\ast=\{0\}\/$ if and only if it holds:

3343: \begin{eqnarray*}

3344: ii.1)&\quad&\mu(X)<\infty\hbox{\ or\ else\ }\mu(X)=\infty\hbox{\ and }w\notin L^1, \hbox{\ and }\\

3345: ii.2)&\quad&\int_0^1\bigg({t\over W(t)}\bigg)^{p^\prime-1}\,dt=\infty.

3346: \end{eqnarray*}

3347: \end{enumerate}

3348: \ecor

3349:

3350: The spaces $d(\Omega,p)\/$ have been studied up to now in the case $\Omega\d\/$. As we shall see,

3351: this condition asserts (if

3352: $p\ge1\/$) that

3353: $\|\cdot\|_{d(\Omega,p)}\/$ is a norm. In the case $\Omega\d,p<1\/$ there has been interest in

3354: finding the     Banach envelope of $d(\Omega,p)\/$ (see for example  \cite{Po} and

3355:  \cite{NO}). The following theorem solves this question in the general case of $w\notin

3356: L^1\/$.

3357:

3358: \bteo

3359: \label{te: II.4.32} Let $0<p\le1\/$.

3360: \begin{enumerate}

3361: \item[(i)] Let $X\/$ be nonatomic and let us assume that

3362: \begin{enumerate}

3363: \item[(a)] $ W\in\Delta_2,$

3364: \item[(b)] $ w\notin L^1\hbox{\ or\ }\mu(X)<\infty,$

3365: \item[(c)] $\sup_{0<t<1}{t^p\over  W(t)}<\infty.$

3366: \end{enumerate}

3367: Then, there exists a decreasing weight $\tilde{w}\/$

3368: with

3369: $$

3370: \widetilde{W}(t)\approx\inf_{s>0}\max(1,t/s)W^{1/p}(s),t>0,\/

3371: $$

3372: and such that the

3373:  Mackey topology in  $\Lambda^p_X(w)\/$ is the one induced by the norm

3374: $\|\cdot\|_{\Lambda^1_X(\tilde{w})}\/$. The Banach envelope of  $\Lambda^p(w)\/$ is

3375: $\Lambda^1(\tilde{w})\/$ if $\tilde{w}\notin L^1\/$, and it is the space

3376: $$

3377: \Lambda^1(\tilde{w})_0=\big\{f\in\Lambda^1(\tilde{w})\,:\,\lim_{t\to\infty}f^\ast(t)=0\big\}

3378: $$

3379: otherwise.

3380:

3381: \item[(ii)] If $\Omega\notin\ell^1\/$ and the space $d(\Omega,p)\/$ is quasi-normed, there exists a

3382: decreasing sequence

3383: $\widetilde{\Omega}=(\widetilde{\Omega}_n)_n\/$ such that

3384: $$

3385: \sum_{k=0}^n\widetilde{\Omega}_k\approx\sup_{m\ge0}\max\bigg(1,\,{n+1\over

3386: m+1}\bigg)\bigg(\sum_{k=0}^m\Omega_k\bigg)^{1/p},\;n=0,1,2,\dots,

3387: $$

3388: and the norm

3389: $\|\cdot\|_{d(\widetilde{\Omega},1)}\/$  induces the  Mackey topology in $d(\Omega,p)\/$. If

3390: $\widetilde{\Omega}\notin\ell^1\/$   the Mackey completion of

3391: $d(\Omega,p)\/$ is   $d(\widetilde{\Omega},1)\/$, and this space is  $c_0=c_0(\N^\ast)\/$ otherwise.

3392: \end{enumerate}

3393: \eteo

3394:

3395: \bdem

3396: By Corollary~\ref{cor: II.4.30} the dual and the associate space of

3397: $\Lambda=\Lambda^p_X(w)\/$ ($\Lambda=d(\Omega,p)\/$ in the case (ii)) coincide, and moreover

3398: $\Lambda^p(w)^\prime=\Lambda^p(w)^\ast\neq\{0\}\/$  by Theorem~\ref{te: II.4.11}. It follows then

3399: from Proposition~\ref{pro: II.4.23}

3400: that the  Mackey topology in

3401: $\Lambda\/$ (the topology of the  bidual norm)  is the one induced by the norm of the biassociate

3402: space, and since $\Lambda^{\prime\prime}\/$ is complete, the Mackey completion

3403: $\widetilde{\Lambda}\/$ is contained in  $\Lambda^{\prime\prime}\/$. By Theorem~\ref{te: II.4.15}

3404: (Theorem~\ref{te: II.4.21} in the case (ii)),

3405: $\Lambda^{\prime\prime}=\Lambda^1(\tilde{w})\/$ ($\Lambda^{\prime\prime}=d(\widetilde{\Omega},1)\/$ resp.).

3406: If

3407: $\tilde{w}\notin L^1\/$ ($\widetilde{\Omega}\notin\ell^1\/$),  Theorem~\ref{te: II.3.13}  tells us

3408: that

3409: $\Lambda\/$ is dense in  $\Lambda^{\prime\prime}\/$ (since $\LL\subset\Lambda\/$) and hence

3410: $\widetilde{\Lambda}=\Lambda^{\prime\prime}\/$. If, on the contrary $\tilde{w}\in L^1\/$,

3411: $\Lambda^1(\tilde{w})_0\/$ is closed in

3412: $\Lambda^1(\tilde{w})\/$ and it contains  the space  $\Lambda^p(w)\/$. If $f\in\Lambda^1(\tilde{w})_0\/$, the

3413: functions

3414: $f_n=f\chi_{\{|f|>f^\ast(n)\}}\/$ have support in a set of finite measure and converge to $f\/$

3415: in $\Lambda^1(\tilde{w})\/$. Each of these functions $f_n\/$ has absolutely continuous norm

3416: (Corollary~\ref{co: II.3.6}) and thus   $f_n\/$ is limit in

3417: $\Lambda^1(\tilde{w})\/$ of function in  $L^\infty_0\/$. It follows that this   space is dense

3418: in $\Lambda^1(\tilde{w})_0\/$ and since  $L^\infty_0\subset\Lambda^p(w)\/$, we conclude that the

3419: Banach envelope of

3420: $\Lambda^p(w)\/$ is $\Lambda^1(\tilde{w})_0\/$. In the atomic case,

3421: $\widetilde{\Omega}\in\ell^1\/$ implies

3422: $\Lambda^{\prime\prime}=d(\widetilde{\Omega},1)=\ell^\infty\/$ and

3423: $\widetilde{\Lambda}=c_0.\qquad\qed$

3424: \edem

3425:

3426: \bobs

3427: Conditions  (a) and  (c) in  (i) of the previous theorem are  natural, since if

3428: $W\notin\Delta_2\/$ there is no topology in

3429: $\Lambda^p(w)\/$ and, if condition (c) fails, $\Lambda^\ast=\{0\}\/$ (Theorem~\ref{te: II.4.11}) and it

3430: does not make sense to consider the  Mackey topology in  $\Lambda^p(w)\/$  with \lq\lq respect to the dual

3431: pair

3432: $\big(\Lambda^p(w),\Lambda^p(w)^\ast\big)\/$\rq\rq. The conditions on  $\Omega\/$  in

3433: (ii) are not at all restrictive, since if

3434: $\Omega\in\ell^1,\;d(\Omega,1)=\ell^\infty\/$ and the  Mackey completion is, in this case, the same

3435: space

3436: $d(\Omega,p)=\ell^\infty\/$.

3437: \eobs

3438:

3439: N. Popa proves in  \cite{Po}, for the case $\Omega\d\/$, that the Mackey completion of

3440: $d(\Omega,p)\/$ ($0<p<1\/$) is $\ell^1\/$ if and only if  $d(\Omega,p)\subset\ell^1\/$. By Theorem~

3441: \ref{te: II.4.32} we can extend this result to the  general case.

3442:

3443: \bcor

3444:  Let $0<p\le1,\;\Omega\notin\ell^1\/$. Then the  Banach envelope of

3445: $d(\Omega,p)\/$ is $\ell^1\/$ if and only if

3446: $d(\Omega,p)\subset\ell^1\/$.

3447: \ecor

3448:

3449: \bdem

3450: If $d(\Omega,p)\subset\ell^1\/$ then

3451: $\ell^\infty=\big(\ell^1\big)^\prime\subset d(\Omega,p)^\prime\/$ and it follows (Theorem~

3452: \ref{te: II.4.20}) that these last two spaces are equal, and hence

3453: $d(\Omega,p)^{\prime\prime}=\ell^1\/$. Since

3454: $d(\Omega,p)^{\prime\prime}=d(\widetilde{\Omega},1)\/$ for some sequence

3455: $\widetilde{\Omega}\/$ (Theorem~\ref{te: II.4.21}), it follows that

3456: $\widetilde{\Omega}\notin\ell^1\/$ and, by the previous theorem, the  Banach envelope of

3457: $d(\Omega,p)\/$ is

3458: $d(\Omega,p)^{\prime\prime}=\ell^1\/$.

3459:

3460: The converse is immediate.$\qquad\qed$

3461: \edem

3462:

3463:

3464:

3465: Let us now study the weak-type spaces $\Lambda^{p,\infty}\/$. First we shall consider the

3466: nonatomic case $X\/$.

3467:

3468: \bteo

3469: \label{te: II.4.35}

3470: If $0<p<\infty,\;W\in\Delta_2\/$ and $X\/$ is nonatomic, then

3471: $\Lambda_X^{p,\infty}(w)^\prime=\Lambda_X^{p,\infty}(w)^\ast\/$ if and only if

3472: these two spaces are zero.

3473: \eteo

3474:

3475:

3476: \bdem

3477: The sufficiency is obvious. To see the necessity, we only need to prove that

3478: $\loinx{p}{X}{w}^\prime\neq\loinx{p}{X}{w}^\ast\/$ if the first of these spaces is not  trivial.

3479:

3480: But if $\loinx{p}{X}{w}^\prime\neq\{0\}\/$, the function $W^{-1/p}\/$ is locally integrable in

3481: $[0,\infty)\/$ (Theorem~\ref{te: II.4.11}) and we can define the seminorm

3482: $$

3483: H(f)=\limsup_{t\to0}{\displaystyle\int_0^t f^\ast(s)\,ds\over\displaystyle\int_0^t

3484: W^{-1/p}(s)\,ds},\qquad f\in\Lambda^{p,\infty}.

3485: $$

3486: If $f\in\loinx{p}{X}{w}\/$ we have that

3487: $f^\ast(t)\le\|f\|_{\Lambda^{p,\infty}}W^{-1/p}(t)\/$, and therefore $H\/$ is well defined and it is

3488: continuous. Moreover, it is not zero because there exists   $f_0\in\Lambda^{p,\infty}\/$

3489: with

3490: $f_0^\ast=W^{-1/p}\/$ in $[0,\mu(X))\/$ and hence  $H(f_0)=1\/$. By  Hahn-Banach theorem, there exists a

3491: nonzero $u\in\Lambda^\ast\/$  such that $|u(f)|\le H(f),$ for every $ f\/$. This linear form is not in

3492: $\Lambda^\prime\/$, because it is zero over all functions  $f\in\LL\/$:

3493: $$

3494: H(f)=\limsup_{s\to0}{\displaystyle\int_0^s f^\ast(t)\,dt\over\displaystyle\int_0^s

3495: W^{-1/p}(t)\,dt}\le\|f\|_{L^\infty}\limsup_{s\to0}\displaystyle{s\over \displaystyle\int_0^s

3496: W^{-1/p}(t)\,dt}=0.

3497: $$

3498: Thus

3499: $\loinx{p}{X}{w}^\ast\neq\loinx{p}{X}{w}^\prime.\qquad\qed$

3500: \edem

3501:

3502:

3503: \bteo

3504:  Let $0<p<\infty\/$ and  $d^\infty(\Omega,p)\/$ be quasi-normed. Let

3505: $W_n=\sum_{k=0}^n\Omega_k$, $n=0,1,\dots\/$ Then, if

3506: $\Omega\notin\ell^1\/$ and $W^{-1/p}\notin\ell^1\/$, we have that  $d^\infty(\Omega,p)^\ast\neq

3507: d^\infty(\Omega,p)^{\prime}\/$.

3508: \eteo

3509:

3510: \bdem The seminorm

3511: $$

3512: H(f)=\limsup_{n\to\infty}{\sum_{k=0}^n f^\ast(k)\over\sum_{k=0}^n W_k^{-1/p}},\qquad f\in

3513: d^\infty(\Omega,p),

3514: $$

3515: is continuous and nonzero, and if

3516: $W^{-1/p}\notin\ell^1\/$, it is zero on all finite sequences. Thus, using and analogous argument

3517: as in the previous proof,   there exists  $u\in d^\infty(\Omega,p)^\ast\setminus

3518: d^\infty(\Omega,p)^\prime.\qquad\qed$

3519: \edem

3520:

3521:

3522: In the nonatomic cases, it remains to characterize the identity

3523: $\Lambda_X^{1,\infty}(w)^\ast=\{0\}\/$. We first need the following results.

3524:

3525: \blem

3526: \label{le: II.4.37}

3527: Let $W\in\Delta_2\/$. If $\mu(X)=\infty\/$ and $w\in L^1\/$, then

3528: $\Lambda^{1,\infty}(w)^\ast\neq\{0\}\/$.

3529: \elem

3530:

3531: \bdem

3532: By   hypothesis, we have that if

3533: $H(f)=\lim_{t\to\infty}f^\ast(t),$ then $H:\Lambda^{1,\infty}(w)\to[0,\infty)\/$

3534: is a continuous seminorm in

3535: $\Lambda^{1,\infty}(w)\/$ not identically zero (since  $1\in\Lambda^{1,\infty}(w)\/$ and

3536: $H(1)=1\neq0\/$). By  Hahn-Banach theorem it follows that

3537: $\Lambda^{1,\infty}(w)^\ast\neq\{0\}.\quad\qed$

3538: \edem

3539:

3540:

3541:

3542: \blem

3543: \label{le: II.4.38}

3544: Let $X\/$ be nonatomic and  $f\in\Lambda_X^{1,\infty}(w)\/$. If

3545: $\lim_{t\to\infty}f^\ast(t)=0\/$ there exists  $F\in\Lambda_X^{1,\infty}(w)\/$ satisfying:

3546: \begin{enumerate}

3547: \item[(i)] $ |f(x)|\le F(x)$ a.e. $x\in X$.

3548: \item[(ii)] $

3549: F^\ast(t)=\|f\|_{\Lambda^{1,\infty}}W^{-1}(t),\quad 0<t<\mu(X).$

3550: \end{enumerate}

3551: \elem

3552:

3553: \bdem

3554: We can assume $\|f\|_{\Lambda^{1,\infty}}=1\/$. Let

3555: $A=\{f\neq0\},\,a=\mu(A)\/$. Since $f^\ast(t)\limto{t\to\infty}0\/$, there exists a measure

3556: preserving transformation

3557: $\sigma:A\to(0,a)\/$ such that  $|f(x)|=f^\ast\big(\sigma(x)\big)\/$ a.e. $x\in

3558: A\/$ (see Theorem~II.7.6 in \cite{BS}). Define

3559: $$

3560: F_0(x)=W^{-1}\big(\sigma(x)\big)\chi_A(x),\quad x\in X.

3561: $$

3562: Since $\sigma\/$ is measure preserving,

3563: $F_0^\ast(t)=W^{-1}(t),\;0<t<a\/$. Besides, since $f^\ast\le W^{-1}\/$, we have that

3564: $$

3565: F_0(x)\ge f^\ast\big(\sigma(x)\big)=|f(x)|\quad\hbox{a.e.\ } x\in A.

3566: $$

3567: If $a=\mu(X)\/$ the function

3568: we are looking for is clearly, $F=F_0\/$. If on the contrary

3569: $a<\mu(X)\/$, we take  $0\le F_1\in\M(X\setminus A)\/$, with $F_1^\ast(t)=W^{-1}(a+t),\;0\le

3570: t<\mu(X\setminus A)\/$. It is then immediate to check that the function

3571: $F=F_0+F_1\chi_{X\setminus A}\/$  satisfies the statement.$\quad\qed$

3572: \edem

3573:

3574: Let us define now the seminorm  $\NN\/$ whose properties (Proposition~\ref{Pro: II.4.40}) will be

3575: very useful.

3576:

3577: \bdefi

3578: \label{def: II.4.39}

3579: If $(E,\|\cdot\|)\/$ is a quasi-normed space, we define the seminorm

3580: $\NN:E\to[0,+\infty)\/$ by

3581: $$

3582: \NN(f)=\NN_E(f)=\inf\bigg\{\sum_{k=1}^K\|f_k\|\,:\,(f_k)_k\subset E,\,f=\sum_k f_k\bigg\},\quad

3583: f\in E.

3584: $$

3585: \edefi

3586:

3587: It can be proved  that  $\NN\/$ is the bidual norm:

3588: $$

3589: \NN(f)=\sup_{u\in\Lambda^\ast}{|u(f)|\over\|u\|_{\Lambda^\ast}}=\|f\|_{\Lambda^{\ast\ast}}.

3590: $$

3591:

3592: \bpro

3593: \label{Pro: II.4.40}

3594:  Let $\Lambda\/$ be a quasi-normed Lorentz space, and

3595: $\NN=\NN_\Lambda\/$. Then for  $f,g\in\Lambda\/$ we have,

3596: \begin{enumerate}

3597: \item[(i)] $\NN(f)\le\|f\|_\Lambda.$

3598: \item[(ii)] $ |f|\le|g|\Rightarrow \NN(f)\le

3599: \NN(g).$

3600: \item[(iii)] $\NN(f)=\NN(|f|).$

3601: \item[(iv)] $

3602: \NN(f)=\inf\bigg\{\sum_{k=1}^K\|f_k\|_\Lambda\,:\,|f|\le\sum_k|f_k|\bigg\}.$

3603: \end{enumerate}

3604: Moreover

3605: $\Lambda^\ast=\{0\}\/$ if and only if  $\NN=0\/$.

3606: \epro

3607:

3608: \bdem

3609:  (i) is immediate. To see (ii), if $g=\sum_k g_k\/$ we have that

3610: $f=(f/g)g=\sum_k(f/g)g_k\/$,  hence

3611: $\NN(f)\le\sum_k\big\|(f/g)g_k\big\|_\Lambda\le\sum_k\|g_k\|_\Lambda\/$ and thus

3612: $\NN(f)\le\NN(g)\/$. (iii) and (iv) are  corollaries of  (ii).

3613:

3614: Let us now prove the last statement. If there exists  $0\neq u\in\Lambda^\ast\/$, we can find

3615: $f\in\Lambda\/$ with $u(f)\neq 0\/$. For every finite decomposition

3616: $f=\sum_k f_k\/$ we then have that, $|u(f)|\le\sum_k|u(f_k)|\le\|u\|\sum_k\|f_k\|_\Lambda\/$. Hence

3617: $\sum_k\|f_k\|_\Lambda\ge|u(f)|/\|u\|\/$ and taking the infimum, we obtain

3618: $\NN(f)\ge|u(f)|/\|u\|>0\/$, that is, $\NN\neq0\/$. Conversely, since $\NN\/$ is a continuous

3619: seminorm in

3620: $\Lambda\/$ (by  (i)), if

3621: $\NN\neq0\/$ there exists (Hahn-Banach theorem, \cite{Ru2}) a nonzero continuous linear form $u\/$  in

3622: $\Lambda.\quad\qed$

3623: \edem

3624:

3625:

3626:

3627:

3628:

3629: In the following result, proved by  A. Haaker in  \cite{Haa} for the case $X=\Rm\/$,

3630: we give a necessary condition  to have that the  dual of

3631: $\Lambda^{1,\infty}(w)\/$ is zero. As we shall see in  Theorem~\ref{te: II.4.45}, this condition is

3632: also sufficient. Our proof is based on the proof of A.  Haaker.

3633:

3634: \blem

3635: \label{lem: II.4.41}

3636: Let $X\/$ be nonatomic and $W\in\Delta_2\/$. If

3637: $\Lambda^{1,\infty}_X(w)^\ast=\{0\}\/$ and  $\epsilon>0\/$, there exists $n\in\N\/$ such that

3638: $$

3639: \int_{nt}^s W^{-1}(r)\,dr\le\epsilon\int_t^s W^{-1}(r)\,dr,\quad 0<t<nt<s\le\mu(X).

3640: $$

3641: \elem

3642:

3643: \bdem

3644: We can assume that  $W\/$ is strictly increasing (otherwise, we substitute this function  by

3645: $W(t)(1+2t)/(1+t)\/$). Let $0\le f\in\Lambda^{1,\infty}(w)\/$ with $f^\ast=W^{-1}\/$ in

3646: $\big(0,\mu(X)\big)\/$. By Proposition~\ref{Pro: II.4.40},

3647: $\NN(f)=0\/$  and there exist positive functions $f_1,\dots,f_{n-1}\in\Lambda^{1,\infty}(w)\/$

3648: with

3649: $$

3650: f\le\sum_k f_k\/,

3651: $$

3652: and such that, if  $a_k=\|f_k\|_{\Lambda^{1,\infty}},\;k=1,\dots,n-1\/$, then

3653: $$

3654: \sum_k a_k<\epsilon.

3655: $$

3656: By  Lemma~\ref{le: II.4.37}, $\mu(X)<\infty\/$  or $\mu(X)=\infty\/$ and $w\notin

3657: L^1\/$. In any case $\lim_{t\to\infty}f_k^\ast(t)=0,\;k=1,\dots,n-1\/$. We can then take functions

3658: $g_1,\dots,g_{n-1}\in\Lambda^{1,\infty}(w)\/$ with $g_k\ge f_k\/$ a.e. and

3659: $g_k^\ast(t)=a_kW^{-1}(t),\;0<t<\mu(X)\/$ (using Lemma~\ref{le: II.4.38}). Summarizing,

3660: \begin{eqnarray*}

3661: (i)&\quad &f\le g_1+\dots+g_{n-1}\hbox{\ \ a.e.},\\ (ii)&\quad&

3662: g_k^\ast(t)=a_kW^{-1}(t),\;0<t<\mu(X),\\ (iii)&\quad& a_1+\dots+a_{n-1}<\epsilon.

3663: \end{eqnarray*}

3664: If

3665: $s,t>0\/$, with $nt<s<\mu(X)\/$, we define

3666: \begin{eqnarray*}

3667:  F&=&\big\{x\in X\,:\,W^{-1}(s)<f(x)\le W^{-1}(t)\big\},\\ E_k&=&\big\{x\in

3668: F\,:\,g_k(x)\le a_kW^{-1}(t)\big\},\quad k=1,\dots,n-1,\\ E&=&\bigcap_{k=1}^{n-1}E_k.

3669: \end{eqnarray*}

3670: Since

3671: $f^\ast=W^{-1}\/$ and this function  is strictly decreasing,

3672: $$

3673: \mu(F)=\lambda_{f^\ast}\big(W^{-1}(s)\big)-\lambda_{f^\ast}\big(W^{-1}(t)\big)=s-t.

3674: $$

3675: Moreover,

3676: $$

3677: \mu(F\setminus E_k)=\mu\big(\big\{x\in

3678: F\,:\,g_k(x)>a_kW^{-1}(t)\big\}\big)\le\lambda_{g_k^\ast}\big(a_kW^{-1}(t)\big)=t,

3679: $$

3680: and

3681: $\mu(F\setminus E)=\mu\big((F\setminus E_1)\cup\dots\cup(F\setminus E_{n-1})\big)\le(n-1)t\/$.

3682: Hence,

3683: $$

3684: \mu(E)\ge\mu(F)-(n-1)t=s-nt.

3685: $$

3686: Since $E\subset F\/$, if $G=\big\{W^{-1}(s)<f\le

3687: W^{-1}(s-\mu(E))\big\}\/$, the distribution function  of $f\chi_E\/$ majorizes

3688: $f\chi_G\/$ and we have that

3689: \begin{eqnarray}

3690: \label{II.4.42}

3691: \int_E

3692: f(x)\,d\mu(x)&=&\int_0^\infty(f\chi_E)^\ast(r)\,dr\ge\int_0^\infty(f\chi_G)^\ast(r)\,dr\\ \nonumber

3693: &=&\int_{s-\mu(E)}^s

3694: W^{-1}(r)\,dr\ge\int_{nt}^s W^{-1}(r)\,dr.

3695: \end{eqnarray}

3696: On the other hand,

3697: \begin{eqnarray*}

3698: \int_E f(x)\,d\mu(x)&\le&\sum_{k=1}^{n-1}\int_E

3699: g_k(x)\,d\mu(x)\\

3700: &\le&\sum_k\int_{E_k}g_k(x)\,d\mu(x)=\sum_k\int_0^\infty(g_k\chi_{E_k})^\ast(r)\,dr,

3701: \end{eqnarray*}

3702: but $\mu(E_k)\le\mu(F)=s-t\/$

3703: and in this set

3704: $g_k\le a_kW^{-1}(t)=g_k^\ast(t)\/$. Hence, if $H_k=\{g_k^\ast(t)\ge g_k>g_k^\ast(s)\}\/$,

3705: the distribution function of  $\chi_{H_k}\/$ majorizes   the   distribution function of

3706:  $\chi_{E_k}\/$ (observe that

3707: $g_k^\ast=a_kW^{-1}\/$ is strictly decreasing) and we have that

3708: $$

3709: \int_0^{\infty}(g_k\chi_{E_k})^\ast(r)\,dr\le\int_t^s g_k^*(r)\,dr=a_k\int_t^s W^{-1}(r)\,dr\/,

3710: $$

3711: and hence,

3712: $$

3713: \int_E f(x)\,d\mu(x)\le\sum_k a_k\int_t^s W^{-1}(r)\,dr<\epsilon\int_t^s W^{-1}(r)\,dr.

3714: $$

3715: This and (\ref{II.4.42}) end the proof.$\quad\qed$

3716: \edem

3717:

3718:

3719:

3720: \bdefi

3721: A function  $f\in\M(X)\/$ is a step function\index{step function} if it is

3722:  of the form

3723: $$

3724: f=\sum_{n=1}^\infty a_n\chi_{E_n},

3725: $$

3726: where $(a_n)_n\subset\C\/$ and $(E_n)_n\/$ is a sequence of  pairwise disjoint

3727: measurable sets in  $X\/$.

3728: \edefi

3729:

3730: The proof of the following result is totally analogous to that of Lemma 7 in  \cite{Cw1}, where  the author

3731: considers the case $W(t)=t^{1/p}\/$.

3732:

3733: \blem

3734: \label{lem: II.4.44}

3735: Every function  $0\le f\in\Lambda^{1,\infty}_X(w)\/$ is majorized by a step  function

3736: $F\in\Lambda^{1,\infty}\/$.

3737: \elem

3738:

3739: Now, we can prove the following result.

3740:

3741: \bteo

3742: \label{te: II.4.45}

3743: Let $X\/$ be nonatomic and  $W\in\Delta_2\/$. If $0<p<\infty\/$,

3744: $$

3745: \Lambda^{p,\infty}_X(w)^\ast=\{0\}

3746: $$

3747: if and only if

3748: \bequ

3749: \label{II.4.46}

3750: \lim_{t\to0^+}\,\sup_{0<r<\mu(X)}{W^{1/p}(tr)\over tW^{1/p}(r)}=0.

3751: \eequ

3752: \eteo

3753:

3754: \bdem

3755: By Remark~\ref{obs: II.2.6}, we can assume $p=1\/$.

3756:

3757: {\sl Sufficiency.} We assume that (\ref{II.4.46}) holds and we shall prove that

3758: $\NN=\NN_{\Lambda^{1,\infty}}=0\/$. By Proposition~\ref{Pro: II.4.40}  we have

3759: $\Lambda^{p,\infty}_X(w)^\ast=\{0\}$. Let then  $f\in\Lambda^{1,\infty}\/$ and let us see that

3760: $\NN(f)=0\/$. By Lemma~\ref{lem: II.4.44} and Proposition~\ref{Pro: II.4.40} (ii),  we can assume

3761: that

3762: $f\/$ is a positive step function:

3763: $$

3764: f=\sum_n a_n\chi_{A_n},\quad (a_n)_n\subset(0,\infty),\;A_n\cap A_m=\emptyset,\;n\neq m.

3765: $$

3766: Observe that either $\mu(X)<\infty\/$ or, by (\ref{II.4.46}),

3767: $W(\infty)=\infty\/$. Therefore, $\lim_{t\to\infty}f^\ast(t)=0\/$, and hence

3768: $\mu(A_n)<\infty\/$ for every $n=1,2,\dots\/$ Also by (\ref{II.4.46}), for every  $\epsilon>0\/$

3769: there exists

3770: $m\in\N\/$ such that

3771: $$

3772: 2^m{W(2^{-m}r)\over W(r)}<\epsilon,\quad 0<r<\mu(X).

3773: $$

3774: That is,

3775: $$

3776: W^{-1}(2^m t)<\epsilon\,2^{-m}W^{-1}(t),\quad 0<t<2^{-m}\mu(X).

3777: $$

3778: Let us divide the sets  $A_n\/$ into $2^m\/$ pairwise disjoint measurable subsets

3779: $(A_n^k)_{k=1}^{2^m}\/$  and with equal measure. For every

3780: $k=1,2,\dots,2^m\/$, we have

3781: $$

3782: f_k=\sum_n a_n\chi_{A_n^k}.

3783: $$

3784: Thus $f=\sum_k f_k\/$ and for  $0<t<2^{-m}\mu(X)\/$ ($f_k^\ast\/$

3785: is zero outside this  interval) we have,

3786: \begin{eqnarray*}

3787: f_k^\ast(t)&=&f^\ast(2^m t)\le W^{-1}(2^m t)\|f\|_{\Lambda^{1,\infty}}\\

3788: &<&\epsilon\,

3789: 2^{-m}W^{-1}(t)\|f\|_{\Lambda^{1,\infty}},\quad k=1,\dots,2^m.

3790: \end{eqnarray*}

3791: Therefore,

3792: $\|f_k\|_{\Lambda^{1,\infty}}\le\epsilon\,2^{-m}\|f\|_{\Lambda^{1,\infty}}\/$ and,

3793: $$

3794: \NN(f)\le\sum_k\|f_k\|_{\Lambda^{1,\infty}}\le\epsilon\,\|f\|_{\Lambda^{1,\infty}}.

3795: $$

3796: Since this holds for every  $\epsilon>0\/$ we conclude that $\NN(f)=0\/$.

3797:

3798: {\sl Necessity.} Let us assume now that $\Lambda^{1,\infty}_X(w)^\ast=\{0\}$. By

3799: Lemma~\ref{lem: II.4.41}, there exists

3800: $n\in\N\/$ such that

3801: \bequ

3802: \label{II.4.47}

3803: \int_{nt}^s W^{-1}(r)\,dr\le{1\over 2}\int_t^s W^{-1}(r)\,dr,\quad t,s>0,\;nt<s<\mu(X).

3804: \eequ

3805: To see

3806: (\ref{II.4.46}) we shall prove that for  $K\in\N\/$,

3807: $$

3808: \sup_{0<r<\mu(X)}{W(tr)\over tW(r)}\le 4n\,2^{-K},

3809: $$

3810: if $0<t<n^{-2K}\/$. To this end, let us define for

3811: $0<r<\mu(X)\/$

3812: $$

3813: A_k=\int_{n^k tr}^r W^{-1}(x)\,dx,\quad k=0,1,\dots,K.

3814: $$

3815: By  (\ref{II.4.47}),

3816: $$

3817: A_0\ge 2A_1\ge 4A_2\ge\dots\ge 2^K A_K

3818: $$

3819: and

3820: $$

3821: {1\over 2}\ge{A_1\over A_0}=1-{A_0-A_1\over A_0}\ge1-2^{-K}{A_0-A_1\over A_K}.

3822: $$

3823: But

3824: $A_0-A_1=\displaystyle\int_{tr}^{ntr} W^{-1}(x)\,dx\le(n-1)trW^{-1}(tr)\/$ and

3825: $A_K\ge(1-n^Kt)rW^{-1}(r)\/$. Hence,

3826: $$

3827: {1\over 2}\ge1-2^{-K}{(n-1)tW(r)\over(1-n^K t)W(rt)}.

3828: $$

3829: And then,

3830: $$

3831: {W(tr)\over tW(r)}\le 2^{1-K}{n-1\over 1-n^K t}\le 4n\,2^{-K}.\quad\qed

3832: $$

3833: \edem

3834:

3835:

3836:

3837:

3838: \bobs

3839:

3840: (i) Condition (\ref{II.4.46}) appears for the first time in  \cite{Haa} where A. Haaker proves the

3841: previous result for the case

3842: $X=\Rm,\;w\notin L^1\/$.

3843:

3844: (ii) In the proof, we have seen that the condition of Lemma~\ref{lem: II.4.41} implies

3845: (\ref{II.4.46}). Therefore such condition is also  equivalent to

3846: $\Lambda^{1,\infty}(w)^\ast=\{0\}\/$.

3847: \eobs

3848:

3849:

3850:

3851: \bcor

3852:  Let $0<p<\infty,\;X\/$  be nonatomic and    $\Lambda=\Lambda_X^{p,\infty}(w)\/$ be

3853: quasi-normed. Then:

3854: \begin{enumerate}

3855: \item[(i)] If $\lim_{t\to0^+}\,\sup_{0<r<\mu(X)}{W^{1/p}(tr)\over tW^{1/p}(r)}=0\/$,

3856: $$

3857: \{0\}=\Lambda^\prime=\Lambda^\ast.

3858: $$

3859:

3860: \item[(ii)] If the previous condition fails,  two cases can occur:

3861: \begin{enumerate}

3862: \item[(ii.1)] If $\displaystyle\int_0^1 W^{-1/p}(s)\,ds=\infty\/$,

3863: $$

3864: \{0\}=\Lambda^\prime\varsubsetneq\Lambda^\ast.

3865: $$

3866:

3867: \item[(ii.2)] If $\displaystyle\int_0^1 W^{-1/p}(s)\,ds<\infty\/$,

3868: $$

3869: \{0\}\neq\Lambda^\prime\varsubsetneq\Lambda^\ast.

3870: $$

3871: \end{enumerate}

3872: \end{enumerate}

3873: \ecor

3874:

3875: \bdem

3876: It is an immediate consequence of  Theorems \ref{te: II.4.11}, \ref{te: II.4.35}, and  \ref{te:

3877: II.4.45}.$\qquad\qed$

3878: \edem

3879:

3880:

3881:

3882: \beje

3883: \ \par

3884: (i) We can apply the previous result to  identify the dual and associate space of the

3885: Lorentz spaces $L^{p,q}(X)\/$ ($X\/$ nonatomic), observing that

3886: $L^{p,q}=\Lambda^q\big(t^{q/p-1}\big)\/$ and $L^{p,\infty}=\Lambda^{p,\infty}(1)\/$. We obtain,

3887: then, by other methods, results already known (see  \cite{Hu, CS, Cw1, Cw2}):

3888: \medskip

3889:

3890: (i.1) $L^{p,q}(X)^\prime=L^{p,q}(X)^\ast=\{0\}\/$, if $0<p<1,\;0<q<\infty\/$,

3891:

3892: (i.2) $L^{1,q}(X)^\prime=L^{1,q}(X)^\ast=L^\infty(X)\/$, if $0<q\le1\/$,

3893:

3894: (i.3) $L^{1,q}(X)^\prime=L^{1,q}(X)^\ast=\{0\}\/$, if $1<q<\infty\/$,

3895:

3896: (i.4) $L^{p,q}(X)^\prime=L^{p,q}(X)^\ast=L^{p^\prime,\infty}(X)\/$, if $1<p<\infty,\;0<q\le1\/$,

3897:

3898: (i.5) $L^{p,q}(X)^\prime=L^{p,q}(X)^\ast=L^{p^\prime,q^\prime}(X)\/$, if $1<p<\infty,\;1<q<\infty\/$,

3899:

3900: (i.6) $L^{p,\infty}(X)^\prime=L^{p,\infty}(X)^\ast=\{0\}\/$, if $0<p<1\/$,

3901:

3902: (i.7) $\{0\}=L^{1,\infty}(X)^\prime\varsubsetneq L^{1,\infty}(X)^\ast\/$,

3903:

3904: (i.8) $\{0\}\neq L^{p,\infty}(X)^\prime\varsubsetneq L^{p,\infty}(X)^\ast\/$, if $1<p<\infty\/$.

3905:

3906: \medskip (ii) If $w=\chi_{(0,1)},\;\Lambda^p_{\Rm}(w)\supset\Lambda^p_{\Rm}(1)=L^p(\Rm)\/$.

3907: $\Lambda^p_{\Rm}(w)^\ast\neq\{0\}\/$ for every

3908: $p\in(0,\infty)\/$ while, as for  $L^p,\;\Lambda^p_{\Rm}(w)^\prime\neq\{0\}\/$, if and

3909: only if

3910: $p\ge1\/$. In the weak-type case,

3911: $\Lambda^{p,\infty}_{\Rm}(w)^\ast\neq\{0\}\/$ for every $p\in(0,\infty)\/$ and the associate space

3912: is not zero only if $p>1\/$.

3913: \eeje

3914:

3915:

3916: \section{Normability}

3917: \label{se:norma}

3918:

3919: In this section $(X,\mu)\/$ is an arbitrary  $\sigma$-finite measure space. We shall study when

3920: $\lox{p,q}{X}{w}\/$ is a  Banach space in the sense that there exists a norm in

3921: $\lox{p,q}{X}{w}\/$ equivalent to the functional $\|\cdot\|_{\lox{p,q}{X}{w}}\/$. In the nonatomic

3922: case, some results are already known:  G.G. Lorentz (\cite{Lor},

3923: 1951) proved that, for

3924: $p\ge1\/$, $\|\cdot\|_{\lox{p}{(0,l)}{w}}\/$ is a norm if and only if $w\/$ is decreasing (that is,

3925: equal  a.e. to a decreasing  function) in $(0,l)\/$. A. Haaker (\cite{Haa}, 1970)

3926: characterizes the normability of

3927: $\lox{p,\infty}{\Rm}{w},\,0<p<\infty\/$ (see also \cite{Sor}), and gives some partial results for

3928: $\lox{p}{\Rm}{w},\,p<\infty\/$. E. Sawyer (\cite{Saw2}, 1990) gives a necessary and sufficient

3929: condition to have that $\loX\/$ is normed in the case

3930: $1<p<\infty,\,X=\Rn\/$, while in  (\cite{CGS}, 1996) the authors solve the case

3931: $0<p<\infty,\,\mu(X)=\infty,\;X\/$ nonatomic.

3932:

3933: In the case $X=\N^\ast\/$ (spaces $d(\Omega,p)\/$) we do not know any previous result concerning

3934: normability. Except some isolated case, in all the references we know on the spaces $d(\Omega,p)\/$,

3935: it is assumed that the sequence

3936: $\Omega\/$ is decreasing which, in the case $p\ge1\/$, ensures that

3937: $\|\cdot\|_{d(\Omega,p)}\/$ is already a norm.

3938:

3939: Here, we solve the  general case for a resonant measure space $X\/$  unifying in this way

3940: all the previous results and including the unknown results for the spaces

3941: $d(\Omega,p)\/$. In the atomic case we shall reduce the problem to  $\Rn\/$.

3942:

3943: Let us start with some general results.

3944:

3945:

3946:

3947:

3948: \bteo

3949: \label{te: II.5.1}

3950: If $1\le p<\infty\/$ and $w\d\/$, then $\|\cdot\|_{\Lambda^p_X(w)}\/$ is a norm.

3951: \eteo

3952:

3953: \bdem

3954: In the nonatomic case,

3955: \begin{eqnarray*}

3956: \|f+g\|_{\Lambda^p_X(w)}&=&\bigg(\int_0^\infty\big((f+g)^\ast(t)\big)^p

3957: w(t)\,dt\bigg)^{1/p}\\

3958: &=&\sup_{h^\ast=w}\bigg(\int_X|f(x)+g(x)|^p h(x)\,d\mu(x)\bigg)^{1/p},

3959: \end{eqnarray*}

3960: and the

3961: triangular inequality is immediate. On the other hand if  $X\/$ is $\sigma$-finite,

3962: we know (retract method) that there exists a nonatomic space  $\bar{X}\/$ and a linear

3963: application

3964: $F:\M(X)\to\M(\bar{X})\/$ such that $F(f)^\ast=f^\ast\/$ (c.f. \cite{BS})  and the

3965: result follows immediately from the atomic case.$\qquad\qed$

3966: \edem

3967:

3968:

3969: The following result gives sufficient conditions to have the normability.

3970:

3971: \bteo

3972: \label{te: II.5.2}

3973: Let  $X$ be a $\sigma$-finite measure space. Then,

3974: \begin{enumerate}

3975: \item[(i)] if $1\le p<\infty\/$ and  $w\in B_{p,\infty},\;\Lambda^p_X(w)\/$ is normable,

3976:

3977: \item[(ii)] if $0<p<\infty\/$ and $w\in B_p,\;\Lambda^{p,\infty}_X(w)\/$ is normable.

3978: \end{enumerate}

3979: \eteo

3980:

3981: \bdem (i) If $p=1\/$ and $w\in B_{p,\infty}=B_{1,\infty}\/$, there exists  a decreasing

3982: function

3983: $\tilde{w}\/$ with $\widetilde{W}\approx W\/$ (\cite{CGS}). It follows that

3984: $\|\cdot\|_{\Lambda^1(\tilde{w})}\/$ is a norm in $\Lambda^1(w)\/$ equivalent to the original

3985: one. If

3986: $p>1\/$ and $w\in B_{p,\infty}\/$,  the Hardy operator

3987: $A\/$ satisfies $A:L^p_{\hbox{\decp dec}}(w)\to L^p(w)\/$  and it follows that the functional

3988: $\|f\|=\|f^{\ast\ast}\|_{L^p(w)}\/$ is an equivalent norm to the    original quasi-norm.

3989:

3990: (ii) If $w\in B_p\/$ we have that  $A:L^{p,\infty}_{\hbox{\decp dec}}(w)\to L^{p,\infty}(w)\/$ (see

3991: \cite{Sor}) and the functional $\|f\|=\|f^{\ast\ast}\|_{L^{p,\infty}(w)}\/$ is an equivalent

3992: norm.$\qquad\qed$

3993: \edem

3994:

3995: \bobs

3996: \label{obs: 5.3}

3997: The functional $\|\cdot\|_{\Lambda^{p,\infty}}\/$ is not a norm, except in some trivial cases. In

3998: fact, if

3999: $(X,\mu)\/$ is an arbitrary measure space and there exist two measurable sets $E\subset F\subset

4000: X\/$ with

4001: $1<a=W\big(\mu(F)\big)/W\big(\mu(E)\big)<\infty\/$ (where $w\/$ is an arbitrary weight in

4002: $\Rm\/$), we have that

4003: $\|f+g\|_{\Lambda^{p,\infty}}>\|f\|_{\Lambda^{p,\infty}}+\|g\|_{\Lambda^{p,\infty}}\/$ for

4004: $f=a\chi_E+\chi_{F\setminus E},\;g=\chi_E+a\chi_{F\setminus E}\/$ (assuming without loss of

4005: generality that

4006: $\mu(F\setminus E)\le\mu(E)\/$). Then, the following statements are equivalent:

4007: \medskip

4008:

4009: (i)\ \  $\|\cdot\|_{\Lambda^{p,\infty}}\/$ is a norm.

4010:

4011: (ii)\ \ $\|\cdot\|_{\Lambda^{p,\infty}}=C\|\cdot\|_{L^\infty}\/$.

4012:

4013: (iii)\ \ The restriction of $W\/$ to the range of  $\mu\/$ is constant.

4014: \eobs

4015:

4016:

4017:

4018:

4019:

4020: \bteo

4021: \label{te: II.5.4}

4022: A Lorentz space $\Lambda\/$   is normable if and only if

4023: $\Lambda=\Lambda^{\prime\prime}\/$,  with equivalent  norms. In particular, every  normable Lorentz

4024: space  $\Lambda\/$  is a  Banach function space with the norm

4025: $\|\cdot\|_{\Lambda^{\prime\prime}}\/$.

4026: \eteo

4027:

4028: \bdem  The sufficiency is obvious. Conversely, if $\Lambda\/$ is

4029: normable with a norm $\|\cdot\|\/$, it has to be equivalent to

4030: $\|\cdot\|_{\Lambda^{\ast\ast}}\/$, (by  Hahn-Banach theorem), and by  Corollary~\ref{cor: II.4.27}, we

4031: have

4032: $\|f\|_{\Lambda}\approx\|f\|_{\Lambda^{\prime\prime}}\/$ for every function  $f\/$ with absolutely

4033: continuous norm in

4034: $\Lambda\/$. Since every $|f|\in\Lambda\/$ is a  pointwise limit of an increasing sequence of functions

4035: in

4036: $\LL\/$  (which have absolutely continuous norm by Corollary~\ref{co: II.3.6}  and

4037: Proposition~\ref{pro: II.3.9}) and the functionals

4038: $\|\cdot\|_{\Lambda}\/$ and $\|\cdot\|_{\Lambda^{\prime\prime}}\/$  have the Fatou property, it

4039: follows that

4040: $\|f\|_{\Lambda}\approx\|f\|_{\Lambda^{\prime\prime}}\/$ for every $f\in\Lambda\/$.  Finally, it is

4041: immediate to prove that the norm

4042: $\|\cdot\|_{\Lambda^{\prime\prime}}\/$ satisfies  the properties of

4043: Definition~\ref{def4.3}.$\qquad\qed$

4044: \edem

4045:

4046: \bobs

4047: \label{obs: 5.5}

4048:  If $\|\cdot\|_{\Lambda}\/$ is a norm, then

4049: $(\Lambda,\|\cdot\|_{\Lambda})\/$ is a  Banach function space and

4050: $\|\cdot\|_{\Lambda}=\|\cdot\|_{\Lambda^{\prime\prime}}\/$ (see \cite{BS}).

4051: \eobs

4052:

4053: From now on, we shall consider the case $X\/$ resonant. We shall see how the previous conditions

4054: are also necessary (in the atomic case). As an  immediate consequence of the previous

4055: theorem   and the    representation theorem of  Luxemburg (\cite{BS}) we have the following result.

4056:

4057: \bcor

4058: \label{cor: II.5.6}

4059: Let $X\/$ be a nonatomic measure space, $0<p,q\le\infty\/$. If $\Lambda_X^{p,q}\/$ is

4060: normable, so is $\Lambda^{p,q}_{(0,\mu(X))}\/$, and if

4061: $\|\cdot\|_{\Lambda_X^{p,q}}\/$ is a norm the same holds for

4062: $\|\cdot\|_{\Lambda_{(0,\mu(X))}^{p,q}}\/$.

4063: \ecor

4064:

4065:

4066:

4067: \blem

4068: \label{lem: II.5.7}

4069: Let  $A\subset\R\/$ be a measurable set, and let $w\/$ be a weight with  $w=w\chi_{(0,|A|)}\/$.

4070: Then,

4071: \begin{enumerate}

4072: \item[(i)]\ \ $\|\cdot\|_{\Lambda_A(w)}\/$ is a norm if and only if  $\|\cdot\|_{\Lambda_\R(w)}\/$ is a

4073: norm,

4074:

4075: \item[(ii)]\ \ $\Lambda_A(w)\/$ is normable if and only if  $\Lambda_{\R}(w)\/$ is normable.

4076: \end{enumerate}

4077: \elem

4078:

4079: \bdem

4080: Since $\Lambda_A(w)\subset\Lambda_\R(w)\/$ the  \lq\lq if\rq\rq\  implication is immediate. To see

4081: the other part,  let us assume that

4082: $$

4083: \bigg\|\sum_{j=1}^n f_j\bigg\|_{\Lambda_A(w)}\le C\sum_j\|f_j\|_{\Lambda_A(w)},\quad\forall

4084: f_1,\dots,f_n\in\Lambda_A(w),

4085: $$

4086: and let us first see that we have an analogous inequality   substituting

4087: $A\/$ by an arbitrary set  $Y\subset\R\/$ with $|Y|\le|A|$. By monotonicity

4088: we can assume

4089: $|Y|<|A|\/$. Then there exists an open set  $G\supset Y\/$, with $b=|G|<|A|\/$. Let

4090: $\tilde{A}\subset A\/$ with $|\tilde{A}|=b\/$ and let  $\sigma_1:\tilde{A}\to(0,b)\/$ be a measure

4091: preserving transformation (Proposition~{7.4} in \cite{BS}). Since $G\/$ is a countable union of

4092:    intervals, it is easy to construct another measure preserving transformation  $\sigma_2:(0,b)\to

4093: G\/$. Then

4094: $\sigma=\sigma_2\circ\sigma_1:\tilde{A}\to G\/$ preserves the measure and for each

4095: $f_1,\dots,f_n\in\M(Y)\subset\M(G)\/$, the functions

4096: $\tilde{f}_j=f_j\circ\sigma\in\M(A),\;j=1,\dots,n,\/$ satisfy $\tilde{f}_j^\ast=f_j^\ast,

4097: $ for every $

4098: j\/$ and

4099: $(\tilde{f}_1+\dots+\tilde{f}_n)^\ast=(f_1+\dots+f_n)^\ast\/$. Hence

4100: $$

4101: \|f_1+\dots+f_n\|_{\Lambda_Y(w)}=\|\tilde{f}_1+\dots+\tilde{f}_n\|_{\Lambda_A(w)}\le

4102: C\sum_j\|\tilde{f}_j\|_{\Lambda_A(w)}=C\sum_j\|f_j\|_{\Lambda_Y(w)}.

4103: $$

4104:

4105: If now  $f_1,\dots,f_n\/$ are arbitrary measurable functions in $\R\/$ and we choose

4106: $Y\subset\big\{|f_1+\dots+f_n|\ge(f_1+\dots+f_n)^\ast(|A|)\big\}\/$, with

4107: $|Y|=|A|\/$, since $w\/$ is supported in  $(0,|A|)\/$ we have that

4108: $$

4109: \|f_1+\dots+f_n\|_{\Lambda_\R(w)}=\big\|(f_1+\dots+f_n)\chi_Y\big\|_{\Lambda_\R(w)}=\bigg\|\sum_j

4110: f_j\chi_Y\bigg\|_{\Lambda_Y(w)},

4111: $$

4112: and it follows that

4113: $$

4114: \|f_1+\dots+f_n\|_{\Lambda_\R(w)}\le C\sum_j\big\|f_j\chi_Y\big\|_{\Lambda_Y(w)}\le

4115: C\sum_j\|f_j\|_{\Lambda_\R(w)}.

4116: $$

4117:

4118: If $C=1\/$ we conclude that $\|\cdot\|_{\Lambda_\R(w)}\/$ is a norm. If $1<C<\infty\/$, the

4119: previous inequality proves that

4120: $\Lambda_\R(w)\/$ is normable (the functional $\NN\/$ defined in \ref{def: II.4.39}

4121: would be, for example, an equivalent norm).$\qquad\qed$

4122: \edem

4123:

4124: The following result characterizes the normability in the nonatomic case.

4125:

4126: \bteo

4127: \label{te: II.5.8}

4128:  Let $(X,\mu)\/$ be a nonatomic measure space,

4129: $0<p<\infty,\;w=w\chi_{(0,\mu(X))}\/$. Then:

4130: \begin{enumerate}

4131: \item[(i)]\ \ $\|\cdot\|_{\Lambda^p_X(w)}\/$ is a norm if and only if  $p\ge1,\;w\d\/$.

4132:

4133: \item[(ii)]\ \ $\Lambda^p_X(w)\/$ is normable if and only if  $p\ge1,\;w\in B_{p,\infty}\/$.

4134:

4135: \item[(iii)]\ \ $\|\cdot\|_{\Lambda^{p,\infty}_X(w)}\/$ is not a norm (except if  $\mu(X)=0\/$).

4136:

4137: \item[(iv)]\ \ $\Lambda^{p,\infty}_X(w)\/$ is normable if and only if $w\in B_p\/$.

4138: \end{enumerate}

4139: \eteo

4140:

4141: \bdem

4142: That the conditions are sufficient has been already seen in  Theorems~\ref{te: II.5.1}, \ref{te: II.5.2},

4143: and Remark~\ref{obs: 5.3}. To see the necessity, we use Corollary~\ref{cor: II.5.6}  and

4144:  Lemma~\ref{lem: II.5.7} to conclude that we can substitute the space $X\/$ by  $\R\/$. Then (i)

4145: follows from  \cite{CGS} and \cite{Lor}; (ii) from  \cite{CGS} and \cite{Saw2}; (iii) from

4146: Remark~\ref{obs: 5.3}, and  (iv) follows from \cite{Sor}.$\qquad\qed$

4147: \edem

4148:

4149:

4150: We shall now study the normability of the sequence   Lorentz spaces $d(\Omega,p)\/$ and

4151: $d^\infty(\Omega,p)\/$.

4152:

4153: \bteo

4154: \label{te: II.5.9}

4155: Let $\Omega=(\Omega_n)_{n=0}^\infty\subset[0,\infty)\/$.

4156: \begin{enumerate}

4157: \item[(i)]\ \ If $0<p<1,\;\|\cdot\|_{d(\Omega,p)}\/$ is a norm if and only if

4158: $\Omega=(\Omega_0,0,0,\dots)\/$.

4159:

4160: \item[(ii)]\ \ If $1\le p<\infty,\;\|\cdot\|_{d(\Omega,p)}\/$ is a norm if and only if $\Omega\d\/$.

4161: \end{enumerate}

4162: \eteo

4163:

4164: \bdem

4165: The sufficiency in  (i) is obvious and in  (ii)   has already been proved in  Theorem~\ref{te:

4166: II.5.1}. Let us see that if  $\|\cdot\|_{d(\Omega,p)}\/$ is a norm, necessarily

4167: $\Omega\d\/$. For $n\in\N$ and $t\in(0,1),\/$ let $f=(1,1,\dots,1,t,0,0,0,\dots)\/$  and let

4168: $g=(1,\dots,1,t,1,0,0\dots).\/$

4169: Then,

4170: \begin{eqnarray*}

4171: \|f\|_{d(\Omega,p)}&=&\|g\|_{d(\Omega,p)}=\big(\Omega_0+\dots+\Omega_n+t^p\Omega_{n+1}\big)^{1/p},

4172: \\

4173: \|f+g\|_{d(\Omega,p)}&=&\big(2^p\Omega_0+\dots+2^p\Omega_{n-1}+(1+t)^p(\Omega_n+

4174: \Omega_{n+1})\big)^{1/p}.

4175: \end{eqnarray*}

4176: Form the inequality

4177: $\|f+g\|_{d(\Omega,p)}\le\|f\|_{d(\Omega,p)}+\|g\|_{d(\Omega,p)}\/$ it follows that

4178: $$

4179: \Omega_{n+1}\le{2^p-(1+t)^p\over(1+t)^p-2^p t^p}\Omega_n,

4180: $$

4181: and letting  $t\/$ tend to  $1\/$

4182: we obtain that  $\Omega_{n+1}\le\Omega_n\/$ and  (ii) is proved.

4183:

4184: Let us assume now that  $\|\cdot\|_{d(\Omega,p)}\/$ is a norm and  $p<1\/$. Then (Remark~\ref{obs:

4185: 5.5}) it coincides with the norm of the biassociate space. Then since  by  Theorem~\ref{te: II.4.16}

4186: (i), $d({\Omega},p)'=d(\widetilde{\Omega},1)'$, where

4187: $\widetilde{\Omega}_0=\Omega_0^{1/p},\;\widetilde{\Omega}_n=\big(\sum_{k=0}^n

4188: \Omega_k\big)^{1/p}-\big(\sum_{k=0}^{n-1}\Omega_k\big)^{1/p},\;n=1,2,\dots\/$, we obtain that

4189: $\Vert\cdot\Vert_{d({\Omega},p)''}=\Vert\cdot\Vert_{d(\widetilde{\Omega},1)''}$. Applying this equality of

4190: norms to

4191: $f=(1,t,0,0,\dots),\;t\in(0,1),\/$ we obtain that

4192: \begin{eqnarray*}

4193: (\Omega_0+t^p\Omega_1)^{1/p}=\|f\|_{d(\Omega,p)}&=&\|f\|_{d(\widetilde{\Omega},1)^{\prime\prime}}

4194: \le\|f\|_{d(\widetilde{\Omega},1)}\\

4195: &=&\Omega_0^{1/p}+t\big((\Omega_0+\Omega_1)^{1/p}

4196: -\Omega_0^{1/p}\big),

4197: \end{eqnarray*}

4198: and hence

4199: $$

4200: {(\Omega_0+t^p\Omega_1)^{1/p}-\Omega_o^{1/p}\over t}\le C<\infty.

4201: $$

4202: If $\Omega_1\neq0\/$ the limit, when $t\to0\/$, of the left hand side is  $+\infty\/$, and thus

4203: $\Omega_1=0\/$ and,  since the sequence is decreasing, we finally obtain

4204: $\Omega_n=0,\;n\ge1.\qquad\qed$

4205: \edem

4206:

4207:

4208: Let us consider now the normability of  $d(\Omega,p)\/$.

4209:

4210: \bteo

4211: \label{te: II.5.10}

4212: Let $\Omega=(\Omega_n)_{n=0}^\infty\subset[0,\infty)\/$ and let us denote

4213: $W_n=\sum_{k=0}^n\Omega_k,\;n=0,1,2,\dots\/$

4214: \begin{enumerate}

4215: \item[(i)]\ \ If $0<p<1,\;d(\Omega,p)\/$ is normable if and only if  $\Omega\in\ell^1\/$.

4216:

4217: \item[(ii)]\ \ $d(\Omega,1)\/$ is normable if and only if

4218: $$

4219: {W_n\over n+1}\le C{W_m\over m+1},\qquad 0<m<n.

4220: $$

4221:

4222: \item[(iii)]\ \ If $1<p<\infty,\;d(\Omega,p)\/$ is normable if and only if

4223: $$

4224: \sum_{k=0}^n{1\over W^{1/p}_k}\le C{n+1\over W^{1/p}_n}.\qquad n=0,1,2,\dots

4225: $$

4226: \end{enumerate}

4227:

4228: That is, if $p<1,\;d(\Omega,p)\/$ is not  normable except in the trivial case $\Omega\in\ell^1\/$

4229: and then

4230: $d(\Omega,p)=\ell^\infty\/$.

4231: \eteo

4232:

4233:

4234:

4235: \bdem (i) If $\Omega\in\ell^1,\;d(\Omega,p)=\ell^\infty\/$ and there is nothing to prove.

4236: Conversely, if $d(\Omega,p)\/$ is normable,

4237: $\|\cdot\|_{d(\Omega,p)^{\prime\prime}}\/$  is a norm in this space (Theorem~\ref{te: II.5.4}) which is

4238: majorized by the norm of

4239: $d(\widetilde{\Omega},1)\/$, with

4240: $\widetilde{\Omega}_0=\Omega_0^{1/p},\;\widetilde{\Omega}_n=\big(\sum_{k=0}^n\Omega_k\big)^{1/p}

4241: -\big(\sum_{k=0}^{n-1}\Omega_k\big)^{1/p},\;n=1,2,\dots\/$

4242: (see the proof of the previous theorem). We have then the inequality,

4243: $$

4244: \bigg(\sum_{n=0}^\infty g(n)^p\Omega_n\bigg)^{1/p}\le C\sum_{n=0}^\infty

4245: g(n)\widetilde{\Omega}_n,\qquad g\d.

4246: $$

4247: If $r=1/p>1\/$ the previous expression  is equivalent to

4248: $$

4249: S=\sup_{g\d}{\sum_{n=0}^\infty g(n)\Omega_n\over\big(\sum_{n=0}^\infty

4250: g(n)^r\widetilde{\Omega}_n\big)^{1/r}}<\infty.

4251: $$

4252: By Theorem~\ref{te: I.6.11},

4253: $$

4254: S\approx\bigg(\int_0^\infty\bigg({W(t)\over\widetilde{W}(t)}\bigg)^{p/(1-p)}w(t)\,dt\bigg)^{1-p},

4255: $$

4256: with

4257: $w=\sum_{k=0}^\infty\Omega_k\chi_{[k,k+1)},\;W(t)=\displaystyle\int_0^t w(s)\,ds\/$ and analogously

4258: $\tilde{w}$ and $\widetilde{W}\/$. Using that $W\in \Delta_2\/$ (because $d(\Omega,p)\/$ is

4259: quasi-normed), one can easily see that the above integrand is  comparable, in every  interval

4260: $[n,n+1)\/$, to the function  ${w/ W}\/$. We have then

4261: $$

4262: \int_1^\infty {w(t)\over W(t)}\,dt=\log {W(\infty)\over W(1)}\lesssim S<\infty,

4263: $$

4264: which implies  $w\in L^1\/$ and  $\Omega\in\ell^1\/$.

4265:

4266: (ii) and  (iii): Let $w(t)=\sum_{k=0}^\infty\Omega_k\chi_{[k,k+1)}(t),\;t>0\/$.

4267: Then the condition of the statement implies  $w\in B_{p,\infty}\/$ (c.f. Theorem~\ref{te: I.6.5} and

4268: Corollary~\ref{cor: I.6.18})  and by  Theorem~\ref{te: II.5.2}, $d(\Omega,p)=\Lambda^p_{\N^\ast}(w)\/$ is

4269: normable. To see the converse, let

4270: $f\ge0\/$ be a decreasing sequence in

4271: $d(\Omega,p)\/$. If $n\ge1\/$, let us define $g_n=\sum_{j=-n}^n f_j\/$ where, for each

4272: $j\in\N^\ast,\;f_j(k)=f(k+j)\chi_{\{k\ge -j\}}(k),\;k=0,1,2,\dots\/$ Then

4273: \bequ

4274: \label{II.5.11}

4275: \|g_n\|_{d(\Omega,p)}\le

4276: C\sum_{j=-n}^n\|f_j\|_{d(\Omega,p)}\le(2n+1)C\|f\|_{d(\Omega,p)},\quad\forall n,

4277: \eequ

4278: and on the other hand, we have

4279: $g_n\ge\big(f(0)+\dots+f(n)\big)\chi_{\{0,\dots,n\}}\/$ and hence

4280: $\|g_n\|_{d(\Omega,p)}\ge\big(f(0)+\dots+f(n)\big)\big\|\chi_{\{0,\dots,n\}}\big\|_{d(\Omega,p)}

4281: =\big(f(0)+\dots+f(n)\big)W_n^{1/p}\/$.

4282: Combining this with  (\ref{II.5.11}) we obtain that

4283: $$

4284: A_d f(n)W_n^{1/p}\le 2C\|f\|_{d(\Omega,p)}=2C\|f\|_{\ell^p(\Omega)},\qquad n\in\N,\;f\d,

4285: $$

4286: which is equivalent to the boundedness of  $A_d:\ell^p_{\hbox{\decp

4287: dec}}(\Omega)\to\ell^{p,\infty}(\Omega)\/$. Now, by Theorem~\ref{te: I.6.13}, the conditions of the

4288: statement follow.$\qquad\qed$

4289: \edem

4290:

4291: To finish this section, we characterize the normability in the weak  case.

4292:

4293: \bteo

4294: \label{te: II.5.12}

4295:  Let $0<p<\infty,\;\Omega=(\Omega_n)_{n=0}^\infty\subset[0,\infty)\/$ and let us denote by

4296:  $W_n=\sum_{k=0}^n\Omega_k,\;n=0,1,2,\dots\/$ Then:

4297: \begin{enumerate}

4298: \item[(i)]\ \ $\|\cdot\|_{d^\infty(\Omega,p)}\/$ is  a norm if and only if  $\Omega=(\Omega_0,0,0,\dots).\/$

4299:

4300: \item[(ii)]\ \ $d^\infty(\Omega,p)\/$ is normable if and only if

4301: $$

4302: \sum_{k=0}^n{1\over W_k^{1/p}}\le C{n+1\over W_n^{1/p} },\qquad n=0,1,2,\dots

4303: $$

4304: \end{enumerate}

4305: \eteo

4306:

4307: \bdem

4308:  (i) is a consequence of  Remark~\ref{obs: 5.3}. To see   (ii) it is enough to observe that the

4309: normability of  $d^\infty(\Omega,p)\/$ is equivalent to the boundedness of

4310: \bequ

4311: \label{II.5.13}

4312: A_d:\ell^{p,\infty}_{\hbox{\decp dec}}(\Omega)\to\ell^{p,\infty}(\Omega),

4313: \eequ

4314:  since if

4315: (\ref{II.5.13}) holds, then $\|f\|=\|A_d f^\ast\|_{\ell^{p,\infty}(\Omega)}\/$ is a norm in

4316: $d^\infty(\Omega,p)\/$ equivalent to the original quasi-norm, and if  $d^\infty(\Omega,p)\/$ is

4317: normable,  the same argument  used in the last part of the previous theorem

4318: (changing

4319: $\|f\|_{d(\Omega,p)}\/$ by

4320: $\|f\|_{d^\infty(\Omega,p)})\/$ shows (\ref{II.5.13}). Applying now Theorem~\ref{te: I.6.16}

4321: we conclude the proof.$\qquad\qed$

4322: \edem

4323:

4324:

4325: \section{Interpolation of operators}

4326: \label{se:oper}

4327:

4328:

4329: In this section $(X,\mu)$ and $(\bar{X},\bar{\mu})\/$ are $\sigma$-finite measure spaces, although this

4330: will not be  necessary in the interpolation theorems  for the

4331: spaces

4332: $\lox{p,q}{X}{w}\/$. In some cases (Theorem~\ref{te: II.6.3}) no conditions on the weight

4333: $w\/$ will be required. In the more general result (Theorem~\ref{te: II.6.5}) we will assume

4334: that the spaces involved are quasi-normed  ($W\in\Delta_2(X)\/$).

4335: This  theorem is a generalization of  Marcinkiewicz theorem

4336: adapted to the context of the   $\lox{p,q}{}{w}\/$ spaces.

4337:

4338: Finally, we shall see how some of the results of chapter~\ref{ch:characteristic} on boundedness of

4339: order continuous operators can be extended to the context of  Lorentz spaces.

4340:

4341: We start by recalling some concepts connected with the real method of interpolation. In this section

4342: we call functional lattice to any class $A\/$ formed by measurable functions

4343: and defined by

4344: $$

4345: A=\big\{f\,:\,\|f\|_A<\infty\big\},

4346: $$

4347: where $\|\cdot\|_A\/$ is a nonnegative functional that acts on measurable functions and satisfies

4348: $\|rf\|_A=r\|f\|_A,\,r>0,\/$ and $\|f\|_A\le\|g\|_A\/$ if $|f(x)|\le|g(x)|\/$ a.e. $x\/$. In

4349: particular the  Lorentz spaces $\lox{p,q}{X}{w}\/$ are of this type. In what follows

4350: $A,B,A_0,A_1,B_0,B_1\/$ denote arbitrary functional lattices. We shall write $A\approx B\/$, if

4351: $A=B\/$ and$\|\cdot\|_A\approx\|\cdot\|_B\/$. Let us now recall the definition of the $K$-functional

4352: associated to the pair $(A_0,A_1)\/$. For every function  $f\in A_0+A_1\/$ and

4353: $t>0\/$

4354: $$

4355: K(f,t,A_0,A_1)=\inf_{f=f_0+f_1}\big(\|f_0\|_{A_0}+t\|f_1\|_{A_1}\big),

4356: $$

4357: where the infimum extends over all possible decompositions $f=f_0+f_1,\;f_i\in A_i,\;i=0,1\/$.

4358: We shall simply write

4359: $K(t,f)\/$ if  it is clear which pair    $(A_0,A_1)\/$ we are working with. The main properties of

4360: $K\/$ are the following:

4361:

4362: (i)\ \ \ \ $K(rf,t)=rK(f,t),\quad r>0\/$,

4363:

4364: (ii)\ \ \ $|f|\le|g|\Rightarrow K(f,t)\le K(g,t)\/$,

4365:

4366: (iii)\ \ $A_i\approx B_i,\;i=0,1\Rightarrow K(f,t,A_0,A_1)\approx K(g,t,B_0,B_1)\/$.

4367:

4368: \noindent Properties (i) and (iii) are immediate while  (ii) can be seen in  \cite{KPS}. For each

4369: $0<\theta<1,\;0<q\le\infty,\/$ we define the following  \lq\lq norm\rq\rq\ in $A_0+A_1\/$:

4370: $$

4371: \|f\|_{(A_0,A_1)_{\theta,q}}=\bigg(\int_0^\infty\big(t^{-\theta}K(f,t,A_0,A_1)\big)^q\,{dt\over

4372: t}\bigg)^{1/q},

4373: $$

4374: ($\sup_t t^{-\theta}K(f,t,A_0,A_1)\/$ if

4375: $q=\infty\/$). It is then  natural to define the space

4376: $$

4377: (A_0,A_1)_{\theta,q}=\big\{f\in A_0+A_1\,:\,\|f\|_{(A_0,A_1)_{\theta,q}}<\infty\big\}.

4378: $$

4379: We say that the operator $T\/$, defined in $A_0+A_1\/$,  and with values in $B_0+B_1\/$, is

4380: quasi-additive if there exists $k>0\/$ such that

4381: $$

4382: \big|T(f+g)(x)\big|\le k\big(|Tf(x)|+|Tg(x)|\big)\quad\hbox{a.e.\ }x,

4383: $$

4384: for every pair of functions

4385: $f,g,f+g\in A_0+A_1\/$.

4386:

4387: The fundamental result of this theory is the following.

4388:

4389: \bteo

4390: \label{te: II.6.1} Let $T\/$ be a quasi-additive operator defined in  $A_0+A_1\/$

4391: and such that

4392: \begin{eqnarray*}

4393: &T&\,:\,A_0\longrightarrow B_0,\\ &T&\,:\,A_1\longrightarrow B_1.

4394: \end{eqnarray*}

4395: Then, for

4396: $0<\theta<1,\;0<q\le\infty,\/$ we have,

4397: $$

4398: T\,:\,(A_0,A_1)_{\theta,q}\longrightarrow(B_0,B_1)_{\theta,q}.

4399: $$

4400: \eteo

4401:

4402: Our purpose now is to identify the space  $(A_0,A_1)_{\theta,q}\/$ or, equivalently, the functional

4403: $\|\cdot\|_{(A_0,A_1)_{\theta,q}}\/$ when

4404: $A_0,A_1\/$ are Lorentz spaces. As we shall see, $\|\cdot\|_{(A_0,A_1)_{\theta,q}}\/$ will be,

4405: under appropriate conditions, equi\-valent to the  \lq\lq norm\rq\rq of a certain Lorentz space.

4406:

4407:

4408: \bteo

4409: \label{te: II.6.2}

4410:  If $0<p<\infty\/$ and $f\in\M(X)\/$,

4411: $$

4412: K\big(f,t,\Lambda^p_X(w),L^\infty(X)\big)\approx K\big(f^\ast,t,L^p(w),L^\infty(w)\big),\qquad

4413: t>0,

4414: $$

4415: with constants depending only on  $p\/$. In

4416: particular, for  $0<\theta<1,\,0<q\le\infty\/$,

4417: $$

4418: \big(\Lambda^p_X(w),L^\infty(X)\big)_{\theta,q}\approx\Lambda^{\bar{p},q}_X(w)

4419: $$

4420: where,

4421: $$\quad{1\over\bar{p}}={1-\theta\over p}.$$

4422: \eteo

4423:

4424: \bdem

4425:  If $f=f_0+f_1\/$ with $f_0\in\Lambda^p_X(w),\,f_1\in L^\infty(X)\/$, we have that

4426: $f^\ast(s)\le f_0^\ast(s)+f_1^\ast(0)=f_0^\ast(s)+\|f_1\|_{L^\infty(X)},\;s>0\/$. Hence,

4427: $$

4428: K\big(f^\ast,t,L^p(w),L^\infty(w)\big)\le

4429: \|f_0^\ast\|_{L^p(w)}+t\|f_1\|_{L^\infty(X)}=\|f_0\|_{\Lambda^p_X(w)}+t\|f_1\|_{L^\infty(X)}.

4430: $$

4431: Taking the infimum over all decompositions

4432: $f=f_0+f_1\in\Lambda^p_X(w)+L^\infty(X)\/$ we obtain,

4433: $$

4434: K\big(f^\ast,t,L^p(w),L^\infty(w)\big)\le K\big(f,t,\Lambda^p_X(w),L^\infty(X)\big).

4435: $$

4436:

4437: To prove the converse inequality, if $f\in\M(X),\,t>0\/$ let $a=(f^\ast)^\ast_w(t^p)\/$ and let

4438: $$

4439: f_0=\bigg(f-a{f\over|f|}\bigg)\chi_{\{|f|>a\}},\qquad f_1=f-f_0.

4440: $$

4441: Then

4442: $(f^\ast_0)^\ast_w=\big((f^\ast)^\ast_w-a\big)\chi_{[0,t^p)}\/$ while

4443: $f^\ast_1\le a\/$. Since $f=f_0+f_1\/$ we have that

4444: \begin{eqnarray*}

4445: K\big(f,t,\loX,L^\infty(X)\big)&\le&\|f_0\|_\loX+t\|f_1\|_{L^\infty(X)}\\

4446: &\le&\|f^\ast_0\|_{L^p(w)}+ta\\ &=& \big\|(f^\ast_0)^\ast_w\big\|_p+ta\\

4447: &=&\bigg(\int_0^{t^p}\big((f^\ast)^\ast_w(s)-a\big)^p\,ds

4448: \bigg)^{1/p}+

4449: \bigg(\int_0^{t^p}a^p\,ds\bigg)^{1/p}\\

4450: &\le& C_p\bigg(\int_0^{t^p}\big((f^\ast)^\ast_w(s)\big)^p\,ds\bigg)^{1/p}.

4451: \end{eqnarray*}

4452: Since the last expression is equivalent to $K\big(f^\ast,t,L^p(w),L^\infty(w)\big)\/$ (see

4453: \cite{BL}), the first part of the theorem is proved.

4454:

4455: The second part is an immediate consequence of the previous one. To see this, observe that the \lq\lq

4456: norm\rq\rq in

4457: $\big(\Lambda^p_X(w),L^\infty(X)\big)_{\theta,q}\/$ is defined using the $K$-functional and we

4458: have

4459: $$

4460: \|f\|_{\big(\Lambda^p_X(w),L^\infty(X)\big)_{\theta,q}}=\|f^\ast\|_{\big(L^p(w),

4461: L^\infty(w)\big)_{\theta,q}},

4462: $$

4463: and the last \lq\lq norm\rq\rq is (see

4464: \cite{BL}),

4465: $$

4466: \|f^\ast\|_{L^{\bar{p},q}(w)}=\|f\|_{\Lambda_X^{\bar{p},q}(w)}.\qquad\qed

4467: $$

4468: \edem

4469:

4470: A consequence of this result is the following interpolation theorem with

4471: $L^\infty(X)\/$.

4472:

4473: \bteo

4474: \label{te: II.6.3}

4475: If $0<p,\bar{p}<\infty\/$ and  $T\/$ is a quasi-additive operator in

4476: $\lox{p}{X}{w}+L^\infty(X)\/$ such that,

4477: \begin{eqnarray*}

4478: T\,&:&\,L^\infty(X)\longrightarrow L^\infty(X),\\

4479: T\,&:&\,\lox{p}{X}{w}\longrightarrow\lox{\bar{p},\infty}{\bar{X}}{\bar{w}},

4480: \end{eqnarray*}

4481: then for

4482: $q,\bar{q}\in(0,\infty)\/$ satisfying  $q/p=\bar{q}/\bar{p}>1,\/$ we have

4483: $$

4484: T\,:\,\lox{q,r}{X}{w}\longrightarrow\lox{\bar{q},r}{\bar{X}}{\bar{w}},\qquad0<r\le\infty.

4485: $$

4486: \eteo

4487:

4488: \bdem

4489: The argument used in the first part of the previous theorem to prove the inequality

4490: $$

4491: K\big(f^\ast,t,L^p(w),L^\infty(w)\big)\le K\big(f,t,\Lambda^p_X(w),L^\infty(X)\big),

4492: $$

4493: still works if we substitute the spaces

4494: $L^p(w)$ and $\Lambda^p_X(w)\/$ by $L^{\bar{p},\infty}(\bar{w})\/$ and

4495: $\Lambda^{\bar{p},\infty}_{\bar{X}}(\bar{w})\/$ respectively. Therefore, with

4496: $1/\bar{q}=(1-\theta)/\bar{p}\/$, it follows   that (see \cite{BL}),

4497: \begin{eqnarray*}

4498: \|Tf\|_{\lox{\bar{q},r}{\bar{X}}{\bar{w}}}&=&\big\|(Tf)^\ast\big\|_{L^{\bar{q},r}(\bar{w})}=

4499: \big\|(Tf)^\ast\big\|_{ \big(L^{\bar{p},\infty}(\bar{w}),L^\infty(w)\big)_{\theta,r}}\\

4500: &\le&\|Tf\|_{\big(\lox{\bar{p},\infty}{}{\bar{w}},L^\infty\big)_{\theta,r}}

4501: \end{eqnarray*}

4502: and the rest is a consequence of Theorem~\ref{te: II.6.1}

4503:  and Theorem~\ref{te: II.6.2}.$\qquad\qed$

4504: \edem

4505:

4506:

4507:

4508: \bobs

4509: \label{obs: 6.4}

4510: Observe that in the last two results, it is not necessary that the

4511: Lorentz spaces involved are quasi-normed.

4512: If we assume that these spaces are quasi-normed (that is $W\in\Delta_2\/$) we can  use the

4513: reiteration theorem to obtain a more general result analogous to the  interpolation  theorem  of

4514: Marcinkiewicz.

4515: \eobs

4516:

4517: \bteo

4518: \label{te: II.6.5}

4519:  Let $0<p_i,q_i,\bar{p}_i,\bar{q}_i\le\infty,\;i=0,1,\/$ with $p_0\ne

4520: p_1,\,\bar{p}_0\ne\bar{p}_1\/$ and let  $T\/$ be a quasi-additive operator defined in

4521: $\lox{p_0,q_0}{X}{w}+\lox{p_1,q_1}{X}{w}\/$ satisfying

4522: \begin{eqnarray*}

4523:  T\,&:&\,\lox{p_0,q_0}{X}{w}\longrightarrow\lox{\bar{p}_0,\bar{q}_0}{\bar{X}}

4524: {\bar{w}},\\

4525: T\,&:&\,\lox{p_1,q_1}{X}{w}\longrightarrow\lox{\bar{p}_1,\bar{q}_1}{\bar{X}}{\bar{w}}.

4526: \end{eqnarray*}

4527: Assume that $W\in\Delta_2(X)$ and $\bar{W}\in\Delta_2(\bar{X})\/$. Then, for

4528: $0<\theta<1,\,0<r\le\infty\/$,

4529: $$

4530: T\,:\,\lox{p,r}{X}{w}\longrightarrow\lox{\bar{p},r}{\bar{X}}{\bar{w}},

4531: $$

4532: where

4533: $$

4534: {1\over p}={1-\theta\over p_0}+{\theta\over p_1},\quad{1\over

4535: \bar{p}}={1-\theta\over\bar{p}_0}+{\theta\over\bar{p}_1}.

4536: $$

4537: \eteo

4538:

4539: \bdem

4540: Let $0<s<\min\{p_0,p_1\}\/$ and take  $\theta_0,\theta_1\in(0,1)\/$ such that

4541: $1/p_i=(1-\theta_i)/s,\,i=0,1\/$. Then,  by Theorem~\ref{te: II.6.2} it follows,

4542: $$

4543: \big(\lox{p_0,q_0}{X}{w},\lox{p_1,q_1}{X}{w}\big)_{\theta,r}\approx\bigg(\big(\lox{s}{X}{w},

4544: L^\infty(X)\big)_{\theta_0,q_0},\big(\lox{s}{X}{w},L^\infty(X)\big)_{\theta_1,q_1}

4545: \bigg)_{\theta,r}.

4546: $$

4547: Since  $W\in\Delta_2(X)\/$ we have that  $\lox{s}{X}{w}\/$ is a quasi-Banach space

4548: and we can apply the reiteration  theorem (Theorem~{3.11.5} in \cite{BL}) to identify the above

4549: space:

4550: $$

4551: \big(\lox{s}{X}{w},\lox{\infty}{X}{w}\big)_{\eta,r}

4552: $$

4553: with

4554: $\eta=(1-\theta)\theta_0+\theta\theta_1\/$. Applying again Theorem~\ref{te: II.6.2}, we finally

4555: obtain

4556: $$

4557: \big(\lox{p_0,q_0}{X}{w},\lox{p_1,q_1}{X}{w}\big)_{\theta,r}\approx\lox{p,r}{X}{w}

4558: $$

4559: since

4560: $(1-\eta)/s=(1-\theta)/p_0+\theta/p_1\/$. The same argument works for

4561: $\lox{\bar{p_i},\bar{q_i}}{\bar{X}}{\bar{w}}\/$ and analogously,

4562: $$

4563: \big(\lox{\bar{p}_0,\bar{q}_0}{\bar{X}}{\bar{w}},\lox{\bar{p}_1,\bar{q}_1}{\bar{X}}

4564: {\bar{w}}\big)_{\theta,r}\approx\lox{\bar{p},r}{\bar{X}}{\bar{w}}.

4565: $$

4566: The result   now follows from Theorem~\ref{te: II.6.1}. $\qquad\qed$

4567: \edem

4568:

4569: \bobs

4570: \label{obs: 6.6}

4571:  J. Cerd\`a and J. Mart\'{\i}n (\cite{CM}) have proved, under some conditions on the weights  $w_0,w_1\/$

4572: that,

4573: $$

4574: K\big(f,t,\lox{p_0,r_0}{}{w_0},\lox{p_1,r_1}{}{w_1}\big)\approx

4575: K\big(f^\ast,t,L^{p_0,r_0}(w_0),L^{p_1,r_1}(w_1)\big)

4576: $$

4577:  for $0<p_0,p_1,r_0,r_1\le\infty\/$. This allows them to obtain a more general interpolation

4578: theorem including  the case

4579: $$

4580: T\,:\,\big(\lox{p_0,q_0}{X}{w_0},\lox{p_1,q_1}{X}{w_1}\big)\longrightarrow

4581: \big(\lox{\bar{p}_0,\bar{q}_0}{\bar{X}}{\bar{w}_0},\lox{\bar{p}_1,\bar{q}_1}

4582: {\bar{X}}{\bar{w}_1}\big),

4583: $$

4584: with $w_0\ne w_1,\bar{w}_0\ne\bar{w}_1\/$.

4585: \eobs

4586:

4587: The following Marcinkiewicz interpolation type result is a  direct consequence of  Theorem~\ref{te:

4588: II.6.5}.

4589:

4590: \bcor

4591: \label{cor: II.6.7}

4592:  Let $0<p_0<p_1\le\infty,\,W\in\Delta_2(X)$, and $\bar{W}\in\Delta_2(\bar{X})\/$.

4593: If $T\/$ is a quasi-additive operator in

4594: $\lox{p_0}{X}{w}+\lox{p_1}{X}{w}\/$ such that,

4595: \begin{eqnarray*}

4596: T\,&:&\,\lox{p_0}{X}{w}\longrightarrow\lox{p_0,\infty}{\bar{X}}{\bar{w}},\\

4597: T\,&:&\,\lox{p_1}{X}{w}\longrightarrow\lox{p_1,\infty}{\bar{X}}{\bar{w}},

4598: \end{eqnarray*}

4599: then, for $p_0<p<p_1\/$,

4600: $$

4601: T\,:\,\lox{p}{X}{w}\longrightarrow\lox{p}{\bar{X}}{\bar{w}}.

4602: $$

4603: \ecor

4604:

4605:

4606: Before ending this chapter, we shall extend some of the results of

4607: section~\ref{se I: characteristic functions}

4608: about boundedness of order  continuous operators.

4609: The first result is a generalization of  Corollary~\ref{cor: I.2.14}.

4610:

4611: \bteo

4612: \label{te: II.6.8}

4613:  Let $L\subset\M(X)\/$ be a regular class and let $T:L\to\M(\bar{X})\/$ be a

4614: sublinear order continuous operator. If $0<p_0\le1,\;p_0\le p_1<\infty\/$ and

4615: $w_1\in B_{p_1/p_0,\infty}\/$, we have

4616: \bequ

4617: \label{II.6.9}

4618: \|Tf\|_{\Lambda^{p_1}_{\bar X}(w_1)}\le C\|f\|_{\Lambda^{p_0}_X(w_0)},\qquad f\in L,

4619: \eequ

4620: if and only if there exists $C_0<\infty\/$ such that

4621: $$

4622: \big\|T\chi_B\big\|_{\Lambda^{p_1}_{\bar X}(w_1)}\le

4623: C_0\big\|\chi_B\big\|_{\Lambda^{p_0}_X(w_0)},\qquad \chi_B\in L.

4624: $$

4625: \eteo

4626:

4627: \bdem

4628: We shall assume the last inequality and we shall prove (\ref{II.6.9}). By monotonicity, it is

4629: sufficient to prove it for a

4630:  simple function $f\ge0\/$. Proceeding as in the proof of  Theorem~\ref{te: I.2.13},

4631: $$

4632: \|Tf\|_{\Lambda^{p_1}_{\bar X}(w_1)}^{p_0}=\big\||Tf|^{p_0}\big\|_{\Lambda^{p_1/p_0}_{\bar

4633: X}(w_1)}\le\bigg\|\int_0^\infty p_0

4634: t^{p_0-1}\big|T\chi_{\{f>t\}}(\cdot)\big|^{p_0}\,dt\bigg\|_{\Lambda^{p_1/p_0}_{\bar X}(w_1)}.

4635: $$

4636: But the condition $w_1\in B_{p_1/p_0,\infty}\/$ implies that

4637: $\|\cdot\|_{\Lambda^{p_1/p_0}_{\bar X}(w_1)}\/$ is equivalent to a  Banach function norm

4638: (Theorem~\ref{te: II.5.2}) and it follows,

4639: \begin{eqnarray*}

4640: \|Tf\|_{\Lambda^{p_1}_{\bar X}(w_1)}^{p_0}&\le& C_1\int_0^\infty  p_0

4641: t^{p_0-1}\big\|\big|T\chi_{\{f>t\}}\big|^{p_0}\big\|_{\Lambda^{p_1/p_0}_{\bar X}(w_1)}\,dt\\

4642: &=&C_1\int_0^\infty  p_0 t^{p_0-1}\big\|T\chi_{\{f>t\}}\big\|_{\Lambda^{p_1}_{\bar

4643: X}(w_1)}^{p_0}\,dt\\

4644: &\le& C_1C_0^{p_0}\int_0^\infty p_0

4645: t^{p_0-1}\big\|\chi_{\{f>t\}}\big\|_{\Lambda^{p_0}_X(w_0)}^{p_0}\,dt\\

4646: &=&C^{p_0}\int_0^\infty p_0

4647: t^{p_0-1}W_0\big(\lambda_f(t)\big)\,dt\\

4648: &=&C^{p_0}\|f\|_{\Lambda^{p_0}_X(w_0)}^{p_0}.\qquad\qed

4649: \end{eqnarray*}

4650:

4651: \edem

4652:

4653:

4654: Using the same idea, one can easily prove an analogous result for the weak-type case:

4655:

4656:

4657: \bteo

4658: \label{te: II.6.10}

4659:  Let $L\subset\M(X)\/$ be a  regular class and let  $T:L\to\M(\bar{X})\/$ be a

4660: sublinear order continuous operator. If $0<p_0\le1,\;0<p_1<\infty\/$ and

4661: $w_1\in B_{p_1/p_0}\/$, we have

4662: $$

4663: \|Tf\|_{\Lambda^{p_1,\infty}_{\bar X}(w_1)}\le C\|f\|_{\Lambda^{p_0}_X(w_0)},\qquad f\in L,

4664: $$

4665: if and only if, there exists $C_0<\infty\/$ such that

4666: $$

4667: \big\|T\chi_B\big\|_{\Lambda^{p_1,\infty}_{\bar X}(w_1)}\le

4668: C_0\big\|\chi_B\big\|_{\Lambda^{p_0}_X(w_0)},\qquad \chi_B\in L.

4669: $$

4670: \eteo

4671:

4672: Combining the previous results with the general interpolation theorem (Theorem~\ref{te: II.6.5})

4673: we obtain a   generalization  of Stein and Weiss theorem on restricted weak-type operators (\cite{SW}).

4674:

4675: \bteo

4676: \label{te: II.6.11}

4677:  Let $0<p_0,p_1,q_0,q_1\le\infty,\;p_0\neq p_1,\;q_0\neq q_1\/$ and let us assume that

4678: $T\,:\,\big(\Lambda^{p_0}_X(w)+\Lambda^{p_1}_X(w)\big)\to\M(\bar{X})\/$ is a

4679: sublinear order continuous operator satisfying

4680: \begin{eqnarray*}

4681: \big\|T\chi_B\big\|_{\Lambda^{q_0,\infty}(\bar{w})}&\le& C_0\|\chi_B\|_{\Lambda^{p_0}(w)},\quad

4682: B\subset X,\\

4683: \big\|T\chi_B\big\|_{\Lambda^{q_1,\infty}(\bar{w})}&\le& C_1\|\chi_B\|_{\Lambda^{p_1}(w)},\quad

4684: B\subset X.

4685: \end{eqnarray*}

4686: Then, if $W,\bar{W}\in\Delta_2\/$, we have

4687: $$

4688: T\,:\,\Lambda^{p,r}_X(w)\longrightarrow \Lambda^{q,r}_{\bar{X}}(\bar{w}),\qquad 0<r\le\infty,

4689: $$

4690: if

4691: $${1\over p}={1-\theta\over p_0}+{\theta\over p_1},\qquad {1\over q}={1-\theta\over

4692: q_0}+{\theta\over q_1},\qquad0<\theta<1.

4693: $$

4694: \eteo

4695:

4696: \bdem

4697:  If $\bar{W}\in\Delta_2\/$ there exists $t>0\/$ such that $\bar{w}\in B_t\/$ (see

4698: \cite{CGS}). Since the classes $B_p\/$ are increasing in $p\/$, there exists an index $r\in(0,1)\/$

4699: with

4700: $r<p_i,\;i=0,1\/$ and such that $\bar{w}\in B_{q_i/r},\;i=0,1\/$. Since

4701: $\big\|\chi_B\big\|_{\Lambda^{p_i}}=C_{p_i,r}\big\|\chi_B\big\|_{\Lambda^{p_i,r}},\;B\subset X,\/$

4702: we have

4703: $$

4704: \big\|T\chi_B\big\|_{\Lambda^{q_i,\infty}(\bar{w})}\le

4705: C_i^\prime\|\chi_B\|_{\Lambda^{p_i,r}(w)},\quad B\subset X,\quad i=0,1.

4706: $$

4707: But

4708: $\Lambda^{p_i,r}(w)=\Lambda^r(\tilde{w}_i)\/$ with $\tilde{w}_i=W^{r/p_i-1}w\/$ (Remark~\ref{obs:

4709: II.2.6}) and by  Theorem~\ref{te: II.6.10} it follows that

4710: $$

4711: T\,:\,\Lambda^{p_i,r}_X(w)\longrightarrow \Lambda^{q_i,\infty}_{\bar{X}}(\bar{w}),\qquad i=0,1.

4712: $$

4713: Applying then  Theorem~\ref{te: II.6.5} we obtain the result.$\qquad\qed$

4714: \edem

4715:

4716:

4717: \bobs

4718: \label{obs: 6.12}

4719:

4720: (i) Since

4721: $\big\|\chi_B\big\|_{\Lambda^{p_i}}=C_{p_i,r}\big\|\chi_B\big\|_{\Lambda^{p_i,r}},\;B\subset X,\/$

4722: the previous  theorem is still true if  we substitute the spaces

4723: $\Lambda^{p_i}(w)\/$ by  $\Lambda^{p_i,r_i}(w)\/$, with $0<r_i\le\infty,\;i=0,1\/$.

4724:

4725: (ii) An analogous result holds, without the hypothesis $W\in\Delta_2\/$, if we change the space

4726: $\Lambda^{p_1}(w)\/$ by $L^\infty\/$, and with  $1/p=(1-\theta)/p_0\/$. This is true since, in this

4727: case, one can use Theorem~\ref{te: II.6.2} (where this hypothesis is not needed) to identify the

4728: interpolated space.

4729: \eobs

4730:

4731:

4732: \chapter{The  Hardy-Littlewood maximal operator in weighted Lorentz spaces}

4733: \label{ch:maximal}

4734: \pagestyle{myheadings}

4735: \markboth{CHAPTER 3. MAXIMAL OPERATOR}{M.J.\ Carro, J.A.\ Raposo, and J.\ Soria}

4736:

4737: \section{Introduction}

4738:

4739: In the previous chapter, we have introduced and studied the Lorentz spaces $\Lambda^{p,q}_X(w)\/$.

4740: Our purpose in this new chapter is to study the boundedness of the Hardy-Littlewood maximal

4741: operator of the type

4742: \bequ

4743: \label{bpp}

4744: M\,:\,\lo\longrightarrow\lo,

4745: \eequ

4746: and its weak version, $M:\lo\to\loin\/$ (see below for the definition of these spaces). Our goal is to find

4747: necessary and/or sufficient conditions on the weights $u\/$ and $w\/$ (in $\R^n\/$ and $\R^m\/$

4748: respectively) to have (\ref{bpp}). We shall also obtain some positive result

4749: for the nondiagonal case

4750: \bequ

4751: \label{bpq}

4752: M\,:\,\loz\longrightarrow\loinu.

4753: \eequ

4754:

4755: If $w=1\/$,    (\ref{bpp}) is equivalent to $M:L^p(u)\longrightarrow L^p(u)\/$ and

4756: this problem was completely solved by Muckenhoupt (\cite{Muck1}), who obtained the condition

4757: $u\in

4758: A_p,\;1<p<\infty\/$ (see (\ref{Ap})). On the other hand, if $u=1\/$, the characterization of (\ref{bpp}) is

4759: equivalent to the boundedness of the Hardy operator $A:\lpwdec\to L^p(w)\/$ and was

4760: obtained ($p>1\/$) by Ari\~no and Muckenhoupt (\cite{AM1}). The condition, in this case, is

4761: $w\in B_p\/$ (see Definition~\ref{def: I.6.3}).

4762:

4763: The problem (\ref {bpq}) and the corresponding strong-type boundedness when $w_0,w_1\/$ are power

4764: weights ($w_i(t)=t^{\alpha_i}\/$)  reduces to $M:L^{p_0,q_0}(u)\to

4765: L^{p_1,q_1}(u)\/$ and was solved, for some cases, by Chung, Hunt, and Kurtz in \cite{CHK, HK}

4766: (see also \cite {L}). The complete solution for the general diagonal case (\ref{bpp}) has been

4767: an open question so far, although there are some partial results due to Carro-Soria (\cite{CS3}) and

4768: Neugebauer (\cite{Ne2}). In this chapter, we prove (see Theorem~\ref{te:principal})  a complete

4769: characterization of (\ref{bpp}) and give necessary and sufficient conditions to have the weak-type

4770: boundedness.

4771:

4772: The chapter is organized as follows: in section~\ref{se: gr} we present some general results which

4773: will be used in what follows, in section~\ref{se:dfcd}, we show that

4774: (\ref{bpp}) holds if and only if there exists $q\in(0,p)\/$  such that

4775: $$

4776: {W\big(u\big(\bigcup_j Q_j\big)\big)\over W\big(u\big(\bigcup_j E_j\big)\big)}\le

4777: C\max_j\bigg({|Q_j|\over|E_j|}\bigg)^q,

4778: $$

4779: for every finite family of cubes  and sets $(Q_j,E_j)_j\/$, with $E_j\subset Q_j\/$.

4780:

4781: In many cases this condition can be simplified. Among other results, we shall see that although the

4782: weight  $w\/$ in (\ref{bpp}) has to be in some class $B_p\/$, the weight  $u\/$  need not be even in

4783: $A_\infty\/$ (see (\ref{Ain})). In fact, we shall show that (\ref{bpp}) can be true with weights $u\/$  which

4784: are not doubling.\index{weight!doubling}

4785:

4786: In section~\ref{se:wtin}, we study the weak boundedness. For example, we prove that if $u_0=u_1=u\in A_1\/$,

4787:   (\ref{bpq}) is equivalent to the case $u_0=u_1=1\/$. Finally, in section~\ref{se:applic}, we

4788: completely solve the problem

4789: $$

4790: M\,:\,L^{p,q}(u)\longrightarrow L^{r,s}(u),

4791: $$

4792: and also (\ref{bpq}), whenever $u_0=u_1=u\/$ is a power weight.

4793: \medskip

4794:

4795: \noindent

4796: {\bf Notation:} Letters $u,u_0,u_1,\dots\/$, will be used to denote weights in $\R^n\/$. They will

4797: be nonnegative measurable functions, not identically zero and integrable on sets of finite

4798: measure. If $A\subset\R^n\/$ is measurable, we shall denote by $u(A)\/$ the measure of  $A\/$ in

4799: the space

4800: $X=\big(\R^n, u(x)dx\big)\/$; that is,

4801: $$

4802: u(A)=\int_A u(x)\,dx.

4803: $$

4804:

4805: The distribution function of $f\in\M(\R^n)\/$ in this space,  will be denoted by  $\lambda_f^u\/$

4806: and the decreasing  rearrangement by  $f^\ast_u\/$. $\Lambda^{p,q}_u(w)\/$ will be the  Lorentz

4807: space

4808: \break

4809: $\Lambda^{p,q}_X(w)\/$. If  $u=1\/$ we simply write $\Lambda^{p,q}(w)\/$. Hence, the \lq\lq

4810: norm\rq\rq\ of $f\/$ in the weighted Lorentz space  $\Lambda^p_u(w)\/$ will be given by\index{$\lo$}

4811: $$

4812: \|f\|_{\lo}=\bigg(\int_0^\infty\big(f^\ast_u(t)\big)^p w(t)\,dt\bigg)^{1/p},

4813: $$

4814: and, in the weak space, \index{$\loin$}

4815: $$

4816: \|f\|_{\loin}=\sup_{t>0}W^{1/p}(t)f^\ast_u(t)=\sup_{t>0}t\,W^{1/p}\big(\lambda_f^u(t)\big).

4817: $$

4818: Let us also recall that  $w\/$ is a weight in  $\R^+\/$  with support in  $[0,u(\R^n)]\/$ and that

4819: $$

4820: 0\le W(t)=\int_0^t w(s)\,ds<\infty,\;t>0.

4821: $$

4822: Finally, the    Hardy-Littlewood maximal  operator $M\/$\index{Hardy-Littlewood maximal operator} is

4823: defined by

4824: $$

4825: Mf(x)=\sup_{x\in Q}{1\over|Q|}\int_Q|f(y)|\,dy,\qquad x\in\R^n,\;f\in\M(\R^n),

4826: $$

4827: where the supremum extends over all cubes $Q\/$, with sides parallel to the axes.

4828:

4829: \section{Some general results}

4830: \label{se: gr}

4831:

4832:

4833: It is known that the decreasing rearrangement of $Mf\/$,  with respect to the Lebesgue measure, is

4834: equivalent (\cite{BS}) to the function  $f^{\ast\ast}\/$:

4835: \bequ

4836: \label{rmax}

4837: (Mf)^\ast(t)\approx f^{\ast\ast}(t)=Af^\ast(t),\qquad t>0.

4838: \eequ

4839: Since every decreasing and positive function in  $\R^+\/$ is equal  a.e. to the decreasing

4840: rearrangement  of a measurable function in  $\R^n\/$, we deduce that the boundedness of

4841: $M:\llo\to\lloin\/$ (resp.

4842: $M:\llo\to\llo\/$) is equivalent to the boundedness of  $A:L^p_{\hbox{\decp dec}}(w)\to

4843: L^{p,\infty}(w)\/$ (resp. $A:L^p_{\hbox{\decp dec}}(w)\to L^p(w)\/$). Therefore

4844: (see section~\ref{se I: Hardy

4845: operator}) a necessary and sufficient condition is $w\in B_{p,\infty}\/$ (resp.

4846: $w\in B_p\/$). On the other hand, when  $w=1\/$, the boundedness

4847: $M:\Lambda^p_u(w)\to\Lambda^{p,\infty}_u(w)\/$  is equivalent to  $M:L^p(u)\to L^{p,\infty}(u)\/$

4848: which is known to be   $u\in A_p\/$ (for  $p\ge1\/$), the Muckenhoupt class of weights  (\cite {Muck1,

4849: MW, CF}). This motivates the following definition:

4850:

4851: \bdefi

4852: If $0<p<\infty\/$,  we write

4853: $w\in B_p(u)\/$\index{$B_p(u)$} (respectively  $w\in B_{p,\infty}(u)\/$)\index{$B_{p,\infty}(u)$} if the

4854: boundedness

4855: $M:\lo\to\lo\/$ (resp. $M:\lo\to\loin\/$) holds. We shall also write $u\in A_p(w)\/$\index{$A_p(w)$} if

4856: $M:\lo\to\loin\/$ holds.

4857: \edefi

4858:

4859: That is, $u\in A_p(w)\Leftrightarrow w\in B_{p,\infty}(u)\/$.

4860: It is clear then that  $A_p(1)=A_p,\;1\le p<\infty,\/$ and that

4861: $B_p(1)=B_p,\;B_{p,\infty}(1)=B_{p,\infty},\;0<p<\infty\/$. On the other hand, since

4862: $\Lambda^p\subset\Lambda^{p,\infty}\/$, we always have  that $B_p(u)\subset

4863: B_{p,\infty}(u),\;0<p<\infty\/$. With this terminology, our purpose will be to identify the classes

4864: $B_p(u)\/$ and

4865: $B_{p,\infty}(u)\/$, or equivalently to identify  the classes  $A_p(w)\/$.

4866:

4867: Using (\ref{rmax}), the nondiagonal case (with $u=1\/$) is also very simple: the boundedness of

4868: $M:\lox{p_0,q_0}{}{w_0}\to\lox{p_1,q_1}{}{w_1}\/$ is equivalent to the boundedness of the Hardy

4869: operator

4870: $A:L^{p_0,q_0}_{\hbox{\decp dec}} (w_0)\to L^{p_1,q_1}(w_1)\/$, which is known in most of the

4871: cases.  %(v�ase la secci�n 6 del Cap�tulo 1).

4872:

4873: The natural question now is to see if there exists an analogue to (\ref{rmax})

4874: when the rearrangement  $f^\ast\/$ is taken with respect to a weight  $u\ne1\/$; that is, if

4875: $(Mf)^\ast_u(t)\approx Af^\ast_u(t)\/$ or $(Mf)^\ast_u(t)\le C Af^\ast_u(t)\/$, when  $u\/$ is a

4876: weight in

4877: $\R^n\/$ satisfying certain conditions. However, it was shown in \cite{CS3} (see also

4878: \cite{ LN1, LN2}) that  $(Mf)^\ast_u\approx Af^\ast_u\/$  if and only if   $u\approx 1\/$, and

4879: $(Mf)^\ast_u\le C\big(Af^{\ast p}_u\big)^{1/p}\/$  if and only if $M:L^p(u)\to

4880: L^{p,\infty}(u),\;p>1\/$ (equivalently if  $u\in A_p\/$). The following theorem generalizes this

4881: result, characterizing

4882: $M:\lo\to\loin\/$ in terms of an expression like  (\ref{rmax}).

4883: \bteo

4884: \label{adpp}

4885: If $0<p<\infty\/$,  $M:\lo\to\loin\/$ is bounded if and only if

4886: $$

4887: (Mf)^\ast_u(t)\le C\bigg({1\over W(t)}\int_0^t\big(f^\ast_u\big)^p(s) w(s)\,ds\bigg)^{1/p},

4888: \qquad t>0,\;f\in\M(\Rn).

4889: $$

4890: \eteo

4891:

4892: \bdem

4893: The inequality of the statement implies

4894: $$

4895: W^{1/p}(t)(Mf)^\ast_u(t)\le C\bigg(\int_0^t\big(f^\ast_u\big)^p (s) w(s)\,ds\bigg)^{1/p}

4896: \le C\|f\|_{\lo},\qquad t>0,

4897: $$

4898: or, equivalently, $\|Mf\|_{\loin}\le C\|f\|_{\lo}\/$. This proves that this condition is sufficient.

4899: On the other hand, let us assume that

4900: $M:\lo\to\loin\/$ is bounded and let  $f\in\lo\/$. If  $f=f_0+f_1\/$,  with

4901: $f_1\in L^\infty\/$, we have, for every  $t>0\/$,

4902: \begin{eqnarray*}

4903: (Mf)^\ast_u(t)&\le&(Mf_0)^\ast_u(t)+(Mf_1)^\ast_u(0)

4904: \\

4905: &\le& W^{-1/p}(t)\|Mf_0\|_{\loin}+\|Mf_1\|_{L^\infty(u)}

4906: \\

4907: &\le& CW^{-1/p}(t)\big(\|f_0\|_{\lo}+W^{1/p}(t)\|f_1\|_{L^\infty(u)}\big),

4908: \end{eqnarray*}

4909: and taking the infimum over all decompositions $f=f_1+f_0\/$ we obtain

4910: \bequ

4911: \label{rmaxu}

4912: (Mf)^\ast_u(t)\le CW^{-1/p}(t)K\big(f,W^{1/p}(t),\lo,L^\infty(u)\big),\quad t>0.

4913: \eequ

4914: By Theorem~\ref{te: II.6.2}, $K\big(f^\ast_u,W^{1/p}(t),L^p(w),L^\infty(w)\big)\/$

4915: is equivalent to the   $K$-functional  %\break

4916:   and therefore (see \cite {BL})

4917: equivalent to

4918: $$

4919: \bigg(\int_0^{W(t)}\big((f^\ast_u)^\ast_w\big)^p(s)\,ds\bigg)^{1/p}=

4920: \bigg(\int_0^t(f^\ast_u)^p (s) w(s)\,ds\bigg)^{1/p}.

4921: $$

4922: This and (\ref{rmaxu})  give us the inequality we are looking for. $\qquad\qed$

4923: \edem

4924:

4925: \bobs

4926: The same argument works for the nondiagonal case and hence (\ref{bpq})   is

4927: equivalent, if $0<p_0,p_1<\infty\/$, to the inequality

4928: $$

4929: (Mf)^\ast_{u_1}(t)\le C\bigg({1\over

4930: W_1^{p_0/p_1}(t)}\int_0^{W_0(t)}\big((f^\ast_{u_0})^\ast_{w_0}\big)^{p_0}(s)\,ds

4931: \bigg)^{1/p_0},\quad

4932: t>0.

4933: $$

4934: \eobs

4935:

4936: Next result gives a necessary and sufficient condition to have the weak-type inequality,

4937: $$

4938: \|Mf\|_{\loinu}\le C\|f\|_{\loz}

4939: $$

4940: on characteristic functions $f=\chi_E,\;E\subset\Rn\/$.

4941:

4942: \bteo

4943: \label{adfc}

4944: Let $0<p_0,p_1<\infty\/$. Then,

4945: \bequ

4946: \label{Mcar}

4947: \big\|M\chi_E\big\|_{\loinu}\le C\big\|\chi_E\big\|_{\loz},\qquad E\subset\Rn\hbox{\ measurable,}

4948: \eequ

4949: if and only if, for every finite family of cubes  $(Q_j)_{j=1}^J\/$ and every family of measurable

4950: sets $(E_j)_{j=1}^J\/$, with

4951: $E_j\subset Q_j,$ for every $ j,\/$ we have that

4952: \bequ

4953: \label{dQjEj}

4954: \frac{W_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)}{W_0^{1/p_0}\big(u_0\big(\bigcup_j

4955: E_j\big)\big)}\le C\max_j\frac{|Q_j|}{|E_j|}.

4956: \eequ

4957: \eteo

4958:

4959: \bdem

4960: To prove the necessary condition, let us consider $f=\chi_E\/$ with $E=\bigcup_j E_j\/$. If

4961: $t>0\/$ is such that ${1/ t}>\max_j{|Q_j|\over|E_j|}\/$, then

4962: $$

4963: {1\over|Q_j|}\int_{Q_j}f(x)\,dx\ge{|E_j|\over|Q_j|}>t,

4964: $$

4965: and we have that  $Q_j\subset\{Mf>t\}\/$.

4966: This holds for every  $j=1,\dots,J\/$ and hence, $\bigcup_j Q_j\subset\{Mf>t\}\/$. Therefore,

4967: $$

4968: tW_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)\le tW_1^{1/p_1}\big(\lambda^{u_1}_{Mf}(t)\big)

4969: \le C\|f\|_{\loz}=CW_0^{1/p_0}\big(u_0\big(\bigcup_j E_j\big)\big),

4970: $$

4971: and consequently

4972: $$

4973: {W_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)\over W_0^{1/p_0}\big(u_0\big(\bigcup_j E_j\big)

4974: \big)}\le {C\over t}.

4975: $$

4976: Since  ${1/ t}>\max_j{|Q_j|\over|E_j|}\/$ is arbitrary, we have shown (\ref{dQjEj}).

4977:

4978: Conversely, let us assume that  (\ref{dQjEj}) holds. Then, this inequality also holds

4979: if the families  $(Q_j)_j,\;(E_j)_j\/$ are countable nonfinite. If $f=\chi_E\in\M(\Rn)\/$ and

4980: $t>0\/$, for every $x\in\{Mf>t\}\/$, there exists a cube  $Q\/$, with $x\in Q\/$ and such that

4981: $ \displaystyle\int_Q f(x)\,dx={|E\cap Q|}>t\/$. By definition we have that

4982: $Mf(y)>t$, for every $ y\in Q\/$ and hence, $Q\subset\{Mf>t\}\/$. Using again this argument, we

4983: obtain a countable  family of cubes

4984: $(Q_j)_j\/$ such that  $\{Mf>t\}=\bigcup_j Q_j\/$ and satisfying (with $E_j=E\cap Q_j\/$)

4985: ${|E_j|/|Q_j|}>t,\;j=1,2,\dots\/$ Then,

4986: $t\le\inf_j{|E_j|\over|Q_j|}=\big(\sup_j{|Q_j|\over|E_j|}\big)^{-1}\/$ and we have, applying

4987:  (\ref{dQjEj}),

4988: \begin{eqnarray*}

4989: tW_1^{1/p_1}\big(\lambda^{u_1}_{Mf}(t)\big)&\le&\displaystyle{W_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)

4990: \over\sup_j{|Q_j|\over|E_j|}}\\

4991: &\le& C W_0^{1/p_0}\big(u_0\big(\bigcup_j E_j\big)\big)\leq C

4992: W_0^{1/p_0}\big(u_0(E)\big).

4993: \end{eqnarray*}

4994: Taking now the supremum in  $t\/$ we obtain that

4995: $$

4996: \|Mf\|_{\loinu}\le C W_0^{1/p_0}\big(u_0(E)\big)=C\|f\|_{\loz}.\qquad\qed

4997: $$

4998: \edem

4999:

5000: \bobs

5001: \label{ob:cd}

5002: In the condition  (\ref{dQjEj}),  we can assume that the sets $(E_j)_j\/$

5003: are disjoint. In fact, for every finite family of cubes $(Q_j)_j\/$ there exists a subfamily

5004: $(Q_{j_k})_k\/$ of disjoint cubes such that $\bigcup_j Q_j\subset\bigcup_k Q^\ast_{j_k}\/$,

5005: where every $Q^\ast_{j_k}\/$ is a dilation of  $Q_{j_k}\/$ with side three times bigger (see for

5006: example \cite{St}). The sets  $(E_{j_k})_k\/$ are then disjoint and, if the condition holds, we

5007: have

5008:

5009:

5010: \begin{eqnarray*}

5011: W_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)\over W_0^{1/p_0}\big(u_0\big(\bigcup_j E_j\big)

5012: \big) &\le \displaystyle{W_1^{1/p_1}\big(u_1\big(\bigcup_k Q_{j_k}^\ast\big)\big)\over

5013: W_0^{1/p_0}\big(u_0\big(\bigcup_k E_{j_k}\big)\big)}

5014: \le C\max_k{\big|Q_{j_k}^\ast\big|\over\big|E_{j_k}\big|} \\

5015: &\le  C C_n\max_k{\big|Q_{j_k}\big|\over\big|E_{j_k}\big|}\le C C_n\max_j{|Q_j|\over|E_j|}.

5016: \end{eqnarray*}

5017: \eobs

5018:

5019: \bcor

5020: \label{cor: cnad}

5021: If $M:\loz\to\loinu,\;0<p_0,p_1<\infty\/$, then

5022: $$

5023: {W_1^{1/p_1}\big(u_1(Q)\big)\over|Q|}\le C{W_0^{1/p_0}\big(u_0(E)\big)\over|E|},\quad E\subset Q,

5024: $$

5025: for every cube  $Q\subset\Rn\/$. In particular $W_0(t)>0,\;t>0\/$, and $u_0(x)>0\/$

5026: a.e. $x\in\Rn.\/$

5027: \ecor

5028:

5029:

5030: The  following two propositions are a consequence of these results.

5031:

5032: \bpro

5033: \label{pro: cnu_0}

5034: Let $0<p_0,p_1<\infty\/$ and let us assume that  $M:\loz\to\loinu\/$, then $u_0\notin L^1(\R^n)$.

5035: \epro

5036:

5037: \bdem

5038: By Corollary~\ref{cor: cnad} we have that, for every cube  $Q\/$,

5039: \bequ

5040: \label{dEQ}

5041: \frac{|E|}{|Q|}\le C\frac{W_0^{1/p_0} \big(u_0(E)\big)}{W_1^{1/p_1} \big(u_1(Q)\big)},

5042: \quad E\subset Q.

5043: \eequ

5044: Let us assume that $u_0(\Rn)<\infty\/$. Then,

5045: if  $a\in\big(0,u_1(\Rn)\big)\/$ and $b\in(0,1)\/$ are such that

5046: $$

5047: C{W_0^{1/p_0}\big(b u_0(\Rn)\big)\over W_1^{1/p_1}(a)}<5^{-n},

5048: $$

5049: with $C$ as in (\ref{dEQ}), we have that, if $u_1(Q)\ge a\/$,  the inequality ${u_0(E)/ u_0(Q)}\le b\/$

5050: implies

5051: $$

5052: {|E|\over|Q|}\le C{W_0^{1/p_0}\big(u_0(E)\big)\over W_1^{1/p_1}\big(u_1(Q)\big)}\le

5053: C{W_0^{1/p_0}\big(b u_0(\Rn)\big)\over W_1^{1/p_1}(a)}<5^{-n};

5054: $$

5055: that is, if $Q\/$ is an arbitrary cube with  $u_1(Q)\ge a\/$,

5056: \bequ

5057: \label{dEQu_0}

5058: E\subset Q,\;|E|\ge5^{-n}|Q|\Rightarrow u_0(E)>b u_0(Q).

5059: \eequ

5060:

5061: Let now  $Q_0\/$ be a cube with  $u_1(3Q_0)\ge a\/$ (for each cube  $Q\/$,  $kQ$

5062: denotes another cube with the same center $Q\/$ and side $k\/$ times bigger than  $Q\/$). Let

5063: $\tilde{Q}\subset3Q_0\/$ be a cube whose interior is disjoint with  $Q_0\/$ and such that

5064: $|\tilde{Q}|=|Q_0|\/$. Then,  $5\tilde{Q}\supset 3Q_0\/$ and hence, $u_1(5\tilde{Q})\ge a\/$, and we

5065: have by (\ref{dEQu_0})

5066: $$

5067: u_0(\tilde{Q})>b u_0(5\tilde{Q})\ge b u_0(Q_0).

5068: $$

5069: Therefore, $u_0(3Q_0)\ge u_0(Q_0)+u_0(\tilde{Q})\ge(1+b)u_0(Q_0)=\alpha u_0(Q_0)\/$,

5070: where $\alpha=1+b>1\/$. By the same argument, $u_0(9Q_0)\ge\alpha u_0(3Q_0)\ge\alpha^2 u_0(Q_0)\/$

5071: and, in general, $u_0(3^n Q_0)\ge\alpha^n u_0(Q_0)\/$. Since $\lim_n u_0(3^n Q_0)=u_0(\Rn)\/$

5072: and

5073: $\alpha>1\/$ we have that $u_0(\Rn)=\infty\/$ (observe that $u_0(Q_0)>0\/$ by  (\ref{dEQ}))

5074: contradicting the initial assumption. $\qed$

5075: \edem

5076:

5077: \bpro

5078: Let $0<p_0,p_1<\infty\/$. If $M:\lox{p_0}{u}{w}\to\loinx{p_1}{u}{w}\/$,

5079: then $p_1\le p_0\/$. If,  in addition, $w\notin L^1(\R^+)$, then $p_1=p_0\/$.

5080: \epro

5081:

5082: \bdem

5083: By Corollary~\ref{cor: cnad}, we have  that

5084: $$

5085: W^{1/p_1-1/p_0}\big(u(Q)\big)\le C,

5086: $$

5087: for every cube  $Q\subset\Rn\/$, and by  Proposition~\ref{pro: cnu_0}

5088: $$

5089: W^{1/p_1-1/p_0}(r)\le C,\quad r>0.

5090: $$

5091: Since $\lim_{t\to0}W(t)=0\/$, we obtain that $p_1\le p_0\/$. If $w\notin L^1(\R^+)$,

5092: $\lim_{t\to\infty}W(t)=\infty\/$ and the previous inequality holds  only if $p_0=p_1.\qquad\qed$

5093: \edem

5094:

5095: \bpro

5096: \label{cnalpq}

5097:  Let $0<p,q,r<\infty\/$. If $M:\lox{p,q}{u}{w}\to\lox{p,r}{u}{w}\/$, then

5098:  $r\ge q\/$.

5099: \epro

5100:

5101: \bdem

5102:  $|f|\le Mf\/$ for every $f\in\M(\Rn)\/$ and the hypothesis implies that

5103: $\lox{p,q}{u}{w}\subset\lox{p,r}{u}{w}\/$. Since $\big(\Rn,u(x)dx\big)\/$ and

5104: $\big(\R^m,w(t)dt\big)\/$ are nonatomic spaces, and

5105: $\|f\|_{\lox{p,q}{u}{w}}=\|f^\ast_u\|_{L^{p,q}(w)}\/$, the previous embedding implies that

5106: $$

5107: \bigg(\int_0^b g^r(t) t^{r/p-1}\,dt\bigg)^{1/r}\le C\bigg(\int_0^b g^q(t)

5108: t^{q/p-1}\,dt\bigg)^{1/q},\qquad g\d,

5109: $$

5110: with $b=W\big(u(\R^n)\big)\/$. Equivalently,

5111: $$

5112: \sup_{g\d}{\displaystyle\int_0^b g(t)t^{r/p-1}\,dt\over\bigg(\displaystyle\int_0^b

5113: g(t)^{q/r}t^{q/p-1}\,dt\bigg)^{r/q}}<\infty.

5114: $$

5115: Now, by  Theorem~\ref{te: I.5.7} in \cite{Saw2}, this supremum is finite if  $r<q.\quad\qed$

5116: \edem

5117:

5118: The following result is a consequence of the interpolation theorems developed in section~\ref{se I:

5119: characteristic functions} of the previous chapter.

5120:

5121: \bteo

5122: \label{te:cnadfc}

5123: Let $0<p_0,p_1<\infty\/$ and let us assume that  $W_1\in\Delta_2\/$. Then, the boundedness

5124: $M:\Lambda^{p_0}_{u_0}(w_0)\to\Lambda^{p_1,\infty}_{u_1}(w_1)\/$ on characteristic  functions:

5125: $$

5126: \big\|M\chi_E\big\|_{\Lambda^{p_1,\infty}_{u_1}(w_1)}\le

5127: C\|\chi_E\|_{\Lambda^{p_0}_{u_0}(w_0)},\qquad E\subset\Rn,

5128: $$

5129: implies

5130: $$

5131: M:\Lambda^{q_0,r}_{u_0}(w_0)\longrightarrow\Lambda^{q_1,r}_{u_1}(w_1),\qquad0<r\le\infty,

5132: $$

5133: for  $p_i<q_i<\infty,\;i=0,1,\;p_1/p_0=q_1/q_0\/$. In particular, we have  that

5134: $M:\lox{q_0}{u_0}{w_0}\to\lox{q_1}{u_1}{w_1}\/$ if $p_1\ge p_0\/$.

5135: \eteo

5136:

5137: \bdem

5138: Since $M:L^\infty\to L^\infty\/$, the statement is an immediate consequence of Theorem~\ref{te: II.6.11}

5139: (see also Remark~\ref{obs: 6.12}). $\qquad\qed$

5140: \edem

5141:

5142: Combining Theorem~\ref{adfc}  with the results of section~\ref{se:oper} of the previous chapter, on

5143: boundedness of continuous order operators on  Lorentz spaces, we obtain the following characterization for

5144: the weak-type boundedness  in the case

5145: $0<p_0\le1$ and $w_1\in B_{p_1/p_0}\/$.

5146:

5147: \bteo

5148: \label{cnsadnd}

5149:  If $0<p_0\le1,\;0<p_1<\infty\/$, and  $w_1\in B_{p_1/p_0}\/$ we have the boundedness

5150: $M:\loz\to\loinu\/$ if and only if,  for every finite family of cubes $(Q_j)_{j=1}^J\/$ and every

5151: family of measurable sets

5152: $(E_j)_{j=1}^J\/$ with $E_j\subset Q_j,$ for every $ j,\/$ we have that

5153: $$

5154: {W_1^{1/p_1}\big(u_1\big(\bigcup_j Q_j\big)\big)\over W_0^{1/p_0}\big(u_0\big(\bigcup_j E_j\big)

5155: \big)}\le C\max_j{|Q_j|\over|E_j|}.

5156: $$

5157: \eteo

5158:

5159: \bdem

5160:  Since $M\/$ is an order continuous operator (Definition~\ref{def: I.2.6}) on the regular class $L=\loz\/$,

5161: we have, by  Theorem~\ref{te: II.6.10},  that the boundedness   is equivalent

5162: to (\ref{Mcar}) and thus,  condition (\ref{dQjEj}) is necessary

5163: and sufficient.$\qquad\qed$

5164: \edem

5165:

5166:

5167: \section{Strong-type boundedness in the diagonal case}

5168: \label{se:dfcd}

5169:

5170: In this section we shall give a characterization of the  $B_p(u)\/$ class in the more general case;

5171: that is, we shall obtain a necessary and sufficient condition to have the boundedness

5172: $$

5173: M\,:\,\lo\longrightarrow\lo.

5174: $$

5175: To this end, we first need the two following technical lemmae:

5176:

5177: \blem

5178: \label{le:Bqp}

5179:  Let $0<p<\infty\/$ and let us assume that, for every cube $Q\subset\Rn\/$,

5180: $$

5181: {W\big(u(Q)\big)\over|Q|^p}\le C{W\big(u(E)\big)\over|E|^p},\qquad E\subset Q,

5182: $$

5183: with $C\/$ independent of $Q\/$. Then,  $w\in B_q\/$ for every $q>p\/$. In particular

5184: $W\in\Delta_2\/$.

5185: \elem

5186:

5187: \bdem

5188: The measure $u(x)dx\/$ is $\sigma$-finite and nonatomic and, therefore,

5189: \break $\big(\Rn,u(x)dx\big)\/$ is a resonant measure space. In these spaces, it holds that

5190: $$

5191: \int_0^\infty f^\ast(s) g^\ast(s)\,ds=\sup_{h^\ast=g^\ast}\int f(x)h(x)u(x)\,dx

5192: $$

5193: (see \cite{BS}) for every measurable functions $f,g\/$. In our case, we have that,

5194: for $0<t<u(Q)\/$,

5195:

5196: \begin{eqnarray*}

5197: \big(u^{-1}\chi_Q\big)^{\ast\ast}_u(t)&=&{1\over t}\int_0^t\big(u^{-1}\chi_Q\big)^\ast_u(s)\,ds\\

5198: &=&{1\over t}\sup\bigg\{\int_Qu^{-1}(x)\chi_E(x)\, u(x)\,dx\,:\,u(E)=t,\;E\subset Q\bigg\}\\

5199: &=&{1\over t}\sup\big\{|E|\,:\,u(E)=t,\;E\subset Q\big\}\\

5200: &\le& C{|Q|\over W^{1/p}\big(u(Q)\big)}

5201: {W^{1/p}(t)\over t}.

5202: \end{eqnarray*}

5203:

5204: Since the function $\big(u^{-1}\chi_Q\big)^{\ast\ast}_u\/$

5205: is decreasing and,

5206: $$

5207: \big(u^{-1}\chi_Q\big)^{\ast\ast}_u(t)\ge\big(u^{-1}\chi_Q\big)^{\ast\ast}_u\big(u(Q)\big)

5208: ={|Q|\over

5209: u(Q)},

5210: $$

5211: we obtain that

5212: $$

5213: {W^{1/p}\big(u(Q)\big)\over u(Q)}\le C{W^{1/p}(t)\over t},\qquad 0<t<u(Q),

5214: $$

5215: which is equivalent to  ${W^{1/p}(r)\over r}\le C{W^{1/p}(t)\over t},\;0<t<r<\infty\/$ and

5216: hence (see, for example \cite{Sor}) $w\in\bigcup_{q>p}B_q.\qquad\qed$

5217: \edem

5218:

5219: We shall also need the following result due to  Hunt-Kurtz (\cite{HK}).

5220:

5221: \blem

5222: \label{HK}

5223:  If $t>0,\;E\subset\Rn,\/$ and we denote by

5224: $E_t=\big\{M\chi_E>t\big\}\/$, there exists a constant $\alpha>1\/$, depending only on the dimension,

5225: such that

5226: $$

5227: \big(E_t\big)_s\supset E_{\alpha ts},\qquad s,t\in(0,1).

5228: $$

5229: \elem

5230:

5231: It is known that the boundedness of  $M:L^p(u)\to L^p(u),\;p>1,\/$ implies $M:L^{p-\epsilon}(u)

5232: \to L^{p-\epsilon}(u)\/$ for some $\epsilon>0\/$. Analogously,  $M:\Lambda^p(w)\to\Lambda^p(w)\/$

5233: is equivalent to  $w\in B_p\/$, that implies  $w\in B_{p-\epsilon}\/$.

5234: As we are going to see next,

5235: this also holds for the general case $M:\lo\to\lo\/$; that is, if $w\in

5236: B_p(u),\;0<p<\infty,\/$ there exists $\epsilon>0\/$ such that $w\in B_{p-\epsilon}(u)\/$.

5237: In fact, in our next theorem, we prove a stronger result: if $M:\lo\to\lo\/$ on characteristics

5238: functions, then

5239: $w\in B_{p-\epsilon}\/$. Part of its proof follows the same patterns developed in Theorem~2 of

5240: \cite{HK}.

5241:

5242: \bteo

5243: \label{apfciaq}

5244:  Let $0<p,r<\infty\/$ and let us assume that

5245: $$

5246: \big\|M\chi_E\big\|_{\Lambda^{p,r}_u(w)}\le C\|\chi_E\|_{\Lambda^p_u(w)},\qquad E\subset\Rn.

5247: $$

5248: Then, there exists $q\in(0,p)\/$ such that  $M:\lox{q}{u}{w}\to\lox{q}{u}{w}\/$.

5249: \eteo

5250:

5251:

5252: \bdem

5253: First, we shall prove that the inequality of the statement is also true if we substitute

5254: the indices $p,r$ by some others $\tilde p$ and $\tilde r$ with $\tilde p<p$ and $\tilde r<r$.

5255:  To this end, let us observe that the hypothesis of the theorem is equivalent (c.f.

5256: Proposition~\ref{pro: II.2.5}) to the inequality

5257: \bequ

5258: \label{u(E_t)}

5259: \int_0^1 t^{r-1}W^{r/p}\big(u(E_t)\big)\,dt\le BW^{r/p}\big(u(E)\big),\qquad E\subset\Rn,

5260: \eequ

5261: with $E_t,\;t>0,\/$ defined as in the previous lemma, and   $B\/$ is a constant independent of

5262: $E\/$. Let

5263: $\alpha>1\/$, the constant, depending only on the dimension, of Lemma~\ref{HK}. We shall prove that,

5264: for every

5265: $n=0,1,2,\dots\/$, the  inequality

5266: \bequ

5267: \label{u(E_t)2}

5268: \int_0^1 t^{r-1}W^{r/p}\big(u(E_t)\big){1\over n!}\log^n {1\over t}\,dt\le B(B\alpha^r)^n

5269:  W^{r/p}\big(u(E)\big),\qquad E\subset\Rn,

5270: \eequ

5271: holds.  If $n=0\/$, (\ref {u(E_t)2}) is  (\ref{u(E_t)}). By induction, we only have to show that

5272: (\ref {u(E_t)2})  implies an analogous inequality with $n+1\/$ instead of  $n\/$. To see this, and

5273: since

5274: $s\in(0,1)\/$, we apply (\ref {u(E_t)2}) to the set $E=E_s\/$, obtaining

5275: \bequ

5276: \label{u((E_s)_t)}

5277: \int_0^1 t^{r-1}W^{r/p}\big(u\big((E_s)_t\big)\big){1\over n!}\log^n {1\over t}\,dt

5278: \le B(B\alpha^r)^n W^{r/p}\big(u(E_s)\big).

5279: \eequ

5280: By Lemma~\ref{HK}, the left hand side of (\ref{u((E_s)_t)}) is greater than or equal to

5281: \begin{eqnarray*}

5282: &&\int_0^{1/\alpha}t^{r-1}W^{r/p}\big(u(E_{\alpha st})\big){1\over n!}\log^n{1\over t}\,dt\\

5283: =(\alpha s)^{-r}&&\int_0^s x^{r-1}W^{r/p}\big(u(E_x)\big){1\over n!}\log^n{\alpha s\over x}\,dx\\

5284: \ge(\alpha s)^{-r}&&\int_0^s x^{r-1}W^{r/p}\big(u(E_x)\big){1\over n!}\log^n{s\over x}\,dx.

5285: \end{eqnarray*}

5286: Thus, from (\ref{u((E_s)_t)}) it follows that

5287: $$

5288: {1\over s}\int_0^s x^{r-1}W^{r/p}\big(u(E_x)\big){1\over n!}\log^n{s\over x}\,dx

5289: \le  (B\alpha^r)^{n+1}s^{r-1}W^{r/p}\big(u(E_s)\big).

5290: $$

5291: Integrating in $s\in(0,1)\/$ both members of this inequality and by  (\ref{u(E_t)}),

5292: we deduce that

5293: $$

5294: \int_0^1{1\over s}\int_0^s x^{r-1}W^{r/p}\big(u(E_x)\big){1\over n!}\log^n{s\over x}

5295: \,dx\,ds\le B(B\alpha^r)^{n+1}W^{r/p}\big(u(E)\big),

5296: $$

5297: and changing the order of integration in the left hand side, we obtain

5298: $$

5299: \int_0^1 x^{r-1}W^{r/p}\big(u(E_x)\big){1\over(n+1)!}\log^{n+1}{1\over x}\,dx\le

5300: B(B\alpha^r)^{n+1}W^{r/p}\big(u(E)\big),

5301: $$

5302: as we wanted to prove.

5303:

5304: Let now $R\in(0,1)\/$. The inequality (\ref{u(E_t)2}) can be written in the following form:

5305: $$

5306: \int_0^1 t^{r-1}W^{r/p}\big(u(E_t)\big){\big({R\over B\alpha^r}\log{1\over t}\big)^n\over n!}

5307: \,dt\le BR^n W^{r/p}\big(u(E)\big),

5308: $$

5309: and summing in  $n\/$ we obtain, with $\delta=R/(B\alpha^r)>0\/$,

5310: $$

5311: \int_0^1 t^{r-\delta-1}W^{r/p}\big(u(E_t)\big)\,dt\le{B\over 1-R}W^{r/p}\big(u(E)\big),

5312: $$

5313: equivalently

5314: $$

5315: \big\|M\chi_E\big\|_{\Lambda^{\tilde{p},\tilde{s}}_u(w)}\le\widetilde{C}

5316: \|\chi_E\|_{\Lambda^{\tilde{p}}_u(w)},\qquad E\subset\Rn,

5317: $$

5318: where $\tilde{p}=p(r-\delta)/r<p,\;\tilde{s}=r-\delta\/$ (see Proposition~\ref{pro: II.2.5}).

5319: In particular,

5320: $$

5321: \big\|M\chi_E\big\|_{\Lambda^{\tilde{p},\infty}_u(w)}\le

5322: \widetilde{C}\|\chi_E\|_{\Lambda^{\tilde{p}}_u(w)},\qquad E\subset\Rn.

5323: $$

5324: Also, by Theorem~\ref{adfc} and  Lemma~\ref{le:Bqp} we have that $W\in\Delta_2\/$. Then, we can

5325: apply   Theorem~\ref{te:cnadfc} (with $u_0=u_1=u,\;w_0=w_1=w,\;p_0=p_1=\tilde{p}\/$) to conclude

5326: the result.

5327: $\qquad\qed$

5328: \edem

5329:

5330: An immediate consequence of Theorem~\ref{apfciaq} and Lemma~\ref{le:Bqp}

5331: is the following:

5332:

5333: \bcor

5334: \label{B(p-epsilon)}

5335: Let $0<p<\infty\/$ and let $u\/$ be an arbitrary weight in $\Rn\/$. Then,

5336: \begin{enumerate}

5337: \item If $w\in B_p(u)\/$ there exists $q<p\/$ such that $w\in B_q(u)\/$.

5338:

5339: \item  $B_p(u)\subset B_p\/$, that is, the boundedness $M:\lo\to\lo\/$ implies $M:\llo\to\llo\/$.

5340: \end{enumerate}

5341: \ecor

5342:

5343: We can now obtain the characterization of the boundedness

5344: $M:\lo\to\lo\/$  or equivalently of the classes  $B_p(u)\/$ for every   $u\/$.

5345:

5346: \bteo

5347: \label{te:principal}

5348: Let $u,w\/$ be weights in  $\Rn\/$ and $\R^+\/$ respectively. If $0<p<\infty\/$

5349: the following results are equivalent:

5350: \begin{enumerate}

5351: \item[(i)] $M:\lo\to\lo\/$.

5352:

5353: \item[(ii)]  $\big\|M\chi_E\big\|_{\Lambda^p_u(w)}\le C\|\chi_E\|_{\Lambda^p_u(w)},

5354: \qquad E\subset\Rn\/$.

5355:

5356: \item[(iii)]  $M:\lox{q}{u}{w}\to\lox{q}{u}{w}\/$, with $q\in(0,p)\/$.

5357:

5358: \item[(iv)]  There exists $q\in(0,p)\/$ such that $\big\|M\chi_E\big\|_{\Lambda^{q,\infty}_u(w)}

5359: \le C\|\chi_E\|_{\Lambda^q_u(w)},\quad E\subset\Rn\/$.

5360:

5361: \item[(v)]  There exists $q\in(0,p)\/$ such that, for every finite family of cubes  $(Q_j)_{j=1}^J\/$

5362: and every family of measurable sets

5363: $(E_j)_{j=1}^J\/$ with $E_j\subset Q_j,$ for every $ j,\/$ we have that

5364: \bequ

5365: \label{cprincipal}

5366: {W\big(u\big(\bigcup_j Q_j\big)\big)\over W\big(u\big(\bigcup_j E_j\big)\big)}\le

5367: C\max_j\bigg({|Q_j|\over|E_j|}\bigg)^q.

5368: \eequ

5369:

5370: \item[(vi)]  $\big((M\chi_E)^\ast_u(t)\big)^q\le C\displaystyle{W\big(u(E)\big)\over W(t)},\;t>0,\;E

5371: \subset\Rn,\/$ with $q\in(0,p)\/$ independent of $t$ and $E\/$.

5372: \end{enumerate}

5373: \eteo

5374:

5375: \bdem

5376:

5377: (i) $\Rightarrow$ (ii) is immediate and  (ii)$\Rightarrow$(iii) is given in Theorem~\ref{apfciaq}.

5378: (iii)$\Rightarrow$(iv) is also immediate and   (iv)$\Rightarrow$(i) is a consequence of

5379: Theorem~\ref{te:cnadfc} (since $W\in\Delta_2\/$ by Theorem~\ref{adfc} and  Lemma~\ref{le:Bqp}).  The

5380: equivalence (iv)$\Leftrightarrow$(v) is Theorem~\ref{adfc}. On the other hand,  (vi) implies, for

5381: $E\subset\Rn\/$,

5382: $$

5383: \big\|M\chi_E\big\|_{\Lambda^{q,\infty}_u(w)}=\sup_{t>0}W^{1/q}

5384: (t)(M\chi_E)^\ast_u(t)\le CW^{1/q}\big(u(E)\big)=C\|\chi_E\|_{\Lambda^q_u(w)},

5385: $$

5386: which is condition  (iv). Finally, (iii) implies $M:\lox{q}{u}{w}

5387: \to\loinx{q}{u}{w}\/$ and by Theorem~\ref{adpp}, we get  (vi).$\qquad\qed$

5388: \edem

5389:

5390: \bobs

5391: \label{ob:disj}

5392: \item As we mentioned in Remark~\ref{ob:cd}, we can assume that the sets $(E_j)_j\/$

5393: in  (\ref{cprincipal}) are  disjoint. If the weight $u\/$ is doubling\index{weight!doubling} (i.e.,

5394: $u(2Q)\le Cu(Q)$), then  also the cubes  $(Q_j)_j\/$ can be taken disjoint.

5395:

5396: \item  Let us assume that the weight $w\/$ satisfies the following property: for every $\alpha>1\/$

5397: there exists a constant $C_\alpha\/$ such that

5398: \bequ

5399: \label{Sum}

5400: {W\big(\sum_j r_j\big)\over W\big(\sum_j t_j\big)}\le C_\alpha\max_j\bigg({W(r_j)

5401: \over W(t_j)}\bigg)^\alpha,

5402: \eequ

5403: for every finite family of   positive numbers $\big\{(r_j,t_j)\big\}_{j=1}^m\/$,

5404: with $0<t_j<r_j,\;j=1,\dots,m\/$. Then, we only need to check condition  (\ref{cprincipal}) for a

5405: unique cube

5406: $Q_j\/$ and a unique set  $E_j\/$, that is, it is equivalent to the inequality

5407: \bequ

5408: \label{cps}

5409: \frac{W\big(u(Q)\big)}{|Q|^q }\le C\frac{W\big(u(E)\big)}{ |E|^q },\qquad E\subset Q,

5410: \eequ

5411: for  $q<p$, and for every cube  $Q\subset\Rn\/$. To see this, observe that this condition is a

5412: consequence of (\ref{cprincipal}) and, if  (\ref{cps}) holds and  $(E_j)_j\/$

5413: is a disjoint family with

5414: $E_j\subset Q_j\/$, then

5415: \begin{eqnarray*}

5416: {W\big(u\big(\bigcup_j Q_j\big)\big)\over W\big(u\big(\bigcup_j E_j\big)\big)}&

5417: \le& {W\big(\sum_j u(Q_j)\big)\over W\big(\sum_j u(E_j)\big)}\le C_\alpha\max_j\bigg({W(u(Q_j))

5418: \over

5419: W(u(E_j))}\bigg)^\alpha\\  &\le&  C C_\alpha\max_j\bigg({|Q_j|\over|E_j|}\bigg)^{\alpha q},\\

5420: \end{eqnarray*}

5421: and it is enough to take $\alpha>1\/$ with $\alpha q<p\/$. We observe that

5422: every power weight  $w(t)=t^\alpha,\;\alpha>-1,\/$ satisfies the above condition.

5423:

5424: \item  If $w=1,\;p>1,\/$ condition (\ref{cprincipal})  says that $u\in A_p\/$, since this

5425: last condition is equivalent  (see, for example \cite{St}) to the existence of  $q\in(1,p)\/$

5426: such that

5427: $$

5428: {u(Q)\over|Q|^q}\le C{u(E)\over|E|^q},\qquad E\subset Q,

5429: $$

5430: for every cube  $Q\/$,  which is equivalent to (\ref{cprincipal}) (by (ii)).

5431:

5432: \item If $u=1\/$, then it is immediate to see that  (\ref{cprincipal}) is equivalent to  $w\in

5433: B_p\/$.

5434: \eobs

5435:

5436: This last observation can be generalized in the following way.

5437:

5438: \bteo

5439: \label{sol-uA_1}

5440: If $u\in A_1\/$ then,

5441: $$

5442: M:\lo\to\lo\;\Leftrightarrow\; M:\llo\to\llo,\qquad0<p<\infty.

5443: $$

5444: With more generality, if $0<p<\infty,$ then $B_p(u)=B_p\/$ if and only if $u\in\bigcap_{q>1}A_q\/$.

5445: \eteo

5446:

5447:

5448:

5449:

5450: \bdem

5451: We already know, by  Corollary~\ref{B(p-epsilon)},  that $B_p(u)\subset B_p\/$.

5452: On the other hand, if $w\in B_p\/$, there exists $l<p\/$ such that $w\in B_l\/$ and we

5453: have that ${W(r)/ W(t)}\le	C\big({r/ t}\big)^l,\;0<t<r<\infty\/$. Let $s>1\/$ be such

5454: that  $sl<p\/$. If

5455: $u\in\bigcap_{q>1}A_q\/$, in  particular $u\in A_s\/$, and it holds that  ${u(Q)/

5456: u(E)}\le C\big({|Q|/ |E|}\big)^s,\;E\subset Q\/$. Therefore, for every family

5457: $(Q_j)_j\/$ of cubes and $E_j\subset Q_j\/$ (pairwise disjoint),

5458: \begin{eqnarray*}

5459: {W\big(u\big(\bigcup_j Q_j\big)\big)\over W\big(u\big(\bigcup_j E_j\big)\big)}&

5460: \le& {W\big(\sum_j u(Q_j)\big)\over W\big(\sum_j u(E_j)\big)}\le C\bigg({\sum_j u(Q_j)\over\sum_j

5461: u(E_j)}\bigg)^l\\

5462: &\le &C\max_j\bigg({u(Q_j)\over u(E_j)}\bigg)^l \le   C\max_j\bigg({|Q_j|\over

5463: |E_j|}\bigg)^{sl},\\

5464: \end{eqnarray*}

5465: which implies that $w\in B_p(u)\/$ (Theorem~\ref{te:principal} (v)).

5466:

5467: Conversely, let us assume now that $B_p(u)=B_p\/$. For every $q<p\/$ the weight

5468: $w(t)=t^{q-1}\/$ is in  $B_p=B_p(u)\/$ and by  (\ref{cprincipal}) we have that, with $q_1\in(q,p)\/$,

5469: $$

5470: \bigg({u(Q)\over u(E)}\bigg)^q={W\big(u(Q)\big)\over W\big(u(E)\big)}\lesssim

5471: \bigg({|Q|\over|E|}\bigg)^{q_1},

5472: $$

5473: and thus,

5474: $$

5475: {u(Q)\over|Q|^{q_1/q}}\le C{u(E)\over|E|^{q_1/q}},\qquad E\subset Q,

5476: $$

5477: for every cube $Q\/$ and hence, if $r=q_1/q\in(1,p/q)\/$, we obtain $u\in\bigcap_{s>r}A_s\/$.

5478: Since this argument works for  $q<p\/$, we deduce that   $u\in\bigcap_{s>1}A_s.\qquad\qed$

5479: \edem

5480:

5481: \bobs

5482:

5483: (i) We know that the boundedness of  $M:\llo\to\llo\/$ holds if and only if $w\in B_p\/$ and, on

5484: the other hand,

5485: $M:\Lambda^p_u\to\Lambda^p_u,\;p>1\/$ (that is,

5486: $M:L^p(u)\to L^p(u)\/$)

5487: if and only if  $u\in A_p\/$. Then, one could think that the boundedness $M:\lo\to\lo,\;p>1,\/$

5488: holds if and only if  $u\in A_p,\;w\in B_p\/$, that is,   $B_p(u)=B_p\/$ if $u\in A_p\/$. The

5489: previous result shows that this is not true. In fact, as a consequence of it,

5490: $B_p(u)\neq

5491: B_p,\;1<p<\infty,\/$ if $u\in A_p\setminus A_q\/$ with $1<q<p\/$.

5492:

5493: (ii)  The fact that $B_p\subset B_p(u)\/$ if $u\in\bigcap_{q>1}A_q\/$ was already proved in

5494: \cite{CS3} and \cite{Ne2}, for the case $p\ge1\/$.

5495: \eobs

5496:

5497:

5498: Using the same idea one can analogously prove the following result which together with  Theorem~

5499: \ref{sol-uA_1} improves Corollary~3.3 in  \cite{CS3} and  Theorem~4.1 in  \cite{Ne2}, to consider the whole

5500: range $0<q<\infty$.

5501:

5502: \bteo

5503: If $1<p<\infty\/$ and $u\in A_p\/$, then $B_{q/p,

5504: \infty}\subset B_q(u),\;0<q<\infty\/$.

5505: \eteo

5506:

5507: By Proposition~\ref{pro: cnu_0} we know that the boundedness  of  $M:\lo\to\lo\/$

5508: implies

5509: $u(\Rn)=\infty\/$. This is, essentially, the best we can say about  $u\/$, since there are examples in

5510: which

5511: $u\/$ is not in any $A_p\/$ class. In fact, the following result proves something stronger: the

5512: weight

5513: $u\/$ could be not doubling. See also Proposition~\ref{W(r)/r^epsilon}.

5514:

5515: \bteo

5516: \label{te: pnd}

5517:  If $u(x)=e^{|x|},\;x\in\R,\/$ and $w=\chi_{(0,1)}\/$,

5518: we have that $M:\lox{q}{u}{w}\to\lox{q}{u}{w}\/$ for every $q>1\/$. Therefore, it is not necessary,

5519: in general, that the weight

5520: $u\/$ is doubling to have the boundedness $M:\lo\to\lo\/$.

5521: \eteo

5522:

5523: \bdem

5524: Since the weight $w\/$ satisfies condition (\ref{Sum}) of Remark~\ref{ob:disj},

5525: it is enough to show that, for every cube  $Q\subset\R\/$ we have,

5526: \bequ

5527: \label{WuQE}

5528: \frac{W(u(Q))}{|Q|}\le C\frac{W(u(E))}{|E|},\qquad E\subset Q.

5529: \eequ

5530: If $u(E)\ge1\/$ then $u(Q)\ge1\/$, $W(u(E))=W(u(Q))=1\/$, and (\ref{WuQE}) holds

5531: (with $C=1\/$) trivially. Thus, we can assume that $u(E)<1\/$. Then,  $W(u(E))=u(E)\/$ and

5532: (\ref{WuQE})

5533: is equivalent, by Lebesgue differentiation theorem, to

5534: \bequ

5535: \label{W(u(Q))}

5536: {W(u(Q))\over|Q|}\le Cu(x),\qquad\hbox{a.e.\ }x\in Q.

5537: \eequ

5538: To prove  (\ref{W(u(Q))}), we can assume $Q=(a,b)\/$ with $0\le a<b\/$ since   $u\/$

5539: is an even function:

5540: for cubes of the form $Q=(-a,b)\/$ we have

5541: $$

5542: {W(u(Q))\over|Q|}\le{W\big(2u(0,b)\big)\over|(0,b)|}\le 2{W\big(u(0,b)\big)\over|(0,b)|},

5543: $$

5544: and we use that the essential infimum of $u\/$ in $(0,b)\/$ and in  $(-a,b)\/$ coincide.

5545: Let then  $Q=(a,b)\/$ with $0\le a<b\/$ and let us see  (\ref {W(u(Q))}). The infimum of  $u\/$ in

5546: $Q\/$

5547:  is $e^a\/$ and, hence, we have to show that  ${W(u(Q))/|Q|}\le Ce^a\/$. If

5548: $e^b-e^a=u(Q)\ge1\/$

5549: then $b\ge\log(1+e^a)\/$ and hence

5550: $$

5551: {W(u(Q))\over|Q|}={1\over b-a}\le{1\over\log(1+e^a)-a}={1\over\log(1+e^{-a})}\le Ce^a.

5552: $$

5553: If, on the contrary, $e^b-e^a=u(Q)<1\/$, we have that  $b-a<e^b-e^a<1\/$ and therefore,

5554: $$

5555: {W(u(Q))\over|Q|}={u(Q)\over|Q|}={e^b-e^a\over b-a}\le e^b=e^a e^{b-a}\le e e^a.\qquad\qed

5556: $$

5557: \edem

5558:

5559:

5560: \section{Weak-type inequality}

5561: \label{se:wtin}

5562:

5563: In this section, we shall study necessary  and/or sufficient conditions

5564: for the weak-type inequality to hold,

5565: $$

5566: M\,:\,\lo\longrightarrow\loin.

5567: $$

5568: The following result (see \cite{CS3})  gives  a necessary condition.

5569:

5570: \bteo

5571: \label{te:c7p}

5572: Let  $0<p_0,p_1<\infty\/$ and

5573: let us assume that $M:\loz\longrightarrow\loinu\/$. Then, there exists a constant $C>0\/$ such that

5574: \bequ

5575: \label{c7p}

5576: \big\|u_0^{-1}\chi_Q\big\|_{(\loz)^\prime}\big\|\chi_Q\big\|_\lou\le C|Q|

5577: \eequ

5578: for every cube  $Q\subset\Rn\/$. Here,  $\|\cdot\|_{(\loz)^\prime}\/$ denotes the norm in the

5579: associate space (c.f. Definition~\ref{def:II.4.1},  Theorem~\ref{te: II.4.7}).

5580: \eteo

5581:

5582: In what follows, we shall study condition (\ref{c7p}) and we shall obtain some important consequences. For

5583: example, combining the previous statement with  some of the results  in chapter~\ref{ch:lor},

5584: we obtain a condition that reduces (depending on  $w_0\/$) the range of indices

5585: $p_0\/$ for which the boundedness of  $M:\loz\to\loinu\/$ can be true. To this end, we

5586:  define the index

5587: $p_w\in[0,\infty)\/$ as follows:

5588: \bequ

5589: \label{defpw}

5590: p_w=\inf\bigg\{p>0\,:\,{t^p\over W(t)}\in L^{p^\prime-1}\bigg((0,1),{dt\over t}\bigg)\bigg\},

5591: \eequ

5592: (where $p^\prime=\infty\/$ if $0<p\le1\/$).

5593:

5594: \bteo

5595: Let $0<p_1<\infty\/$ and let us assume that we have the boundedness of

5596: $M:\loz\longrightarrow\loinu\/$. Then, $p_0\ge p_{w_0}\/$. If $p_{w_0}>1\/$ the previous inequality

5597: is strict.

5598: \eteo

5599:

5600: \bdem

5601: If $M:\loz\longrightarrow\loinu\/$, we have, by  Theorem~\ref{te:c7p},

5602: $(\Lambda^{p_0}_{u_0}(w_0))^\prime\neq\{0\}\/$. Then,  Theorem~\ref{te: II.4.11}

5603: implies

5604:  ${t^{p_0}/W(t)}\in L^{p_0^\prime-1}\big((0,1),{dt/ t}\big)\/$. That is,

5605: $$

5606: p_0\in\bigg\{p>0\,:\,{t^p\over W(t)}\in L^{p^\prime-1}\bigg((0,1),{dt\over t}\bigg)\bigg\}=I.

5607: $$

5608: The result we are looking for,   follows from the fact that  $I\/$ is an  interval in $[0,\infty)\/$,

5609: unbounded in the right hand side, and open in the left hand side by

5610: $p_{w_0}\/$, if

5611: $p_{w_0}>1.\qquad\qed$

5612: \edem

5613:

5614:

5615: \bteo

5616: If $p_0<1\/$ there are no weights $u_0,u_1\/$ such that  $M:L^{p_0}(u_0)

5617: \to L^{p_1,\infty}(u_1),\;0<p_1<\infty\/$ is bounded.

5618: \eteo

5619:

5620: \bdem

5621: It is enough to observe that  $L^{p_0}(u_0)=\Lambda^{p_0}_{u_0}(1)\/$ and that, if $w=1,\;p_w=1\/$

5622: (c.f. (\ref{defpw})). $\qquad\qed$

5623: \edem

5624:

5625: To find equivalent integral expression to (\ref{c7p}), it will be useful to associate to each weight

5626: $u\/$ in

5627: $\Rn\/$ the family of functions

5628: $\{\phi_Q\}_Q\/$ defined in the following way. For every cube  $Q\subset\Rn\/$,

5629: \bequ

5630: \label{defphiQ}

5631: \phi_Q(t)=\phi_{Q,u}(t)={u(Q)\over|Q|}\int_0^t\big(u^{-1}\chi_Q\big)^\ast_u(s)\,ds\,,\quad t\ge 0.

5632: \eequ

5633: It will be also very useful the right derivative of the function $\phi_Q$

5634: \bequ

5635: \label{phiQprime}

5636: \phi_Q^\prime(t)={u(Q)\over|Q|}\big(u^{-1}\chi_Q\big)^\ast_u(t),\quad t\ge 0.

5637: \eequ

5638:  Observe that $\phi_Q^\prime(t)=0\/$ if $t\ge u(Q)\/$, and that  $\phi_Q^\prime(t)

5639: \le\phi_Q(t)/t,\;t\ge0\/$. Now, we can give an equivalent expression of  (\ref{c7p}) in

5640: terms of these functions.   These integral expressions are quite similar to those for the classes

5641: $B_{p_0,p_1,\infty}\/$ (Theorem~\ref{te: I.6.5}).

5642:

5643: \bpro

5644: \label{ce7p}

5645: Let $\phi_Q=\phi_{Q,u_0}\/$.

5646: \begin{enumerate}

5647: \item[(a)] If $1<p_0<\infty\/$ each of the following expressions are equivalent  to (\ref{c7p}):

5648: \begin{enumerate}

5649: \item[(i)]

5650: $$ \bigg(\int_0^{u_0(Q)}W_0^{1-p_0^\prime}(s)\,d\fiq^{p_0^\prime}(s)\bigg)^{1/p_0^\prime}

5651: W_1^{1/p_1}\big(u_1(Q)\big)\le Cu_0(Q).

5652: $$

5653: \item[(ii)]

5654: \begin{eqnarray*}

5655: \bigg(\int_0^{u_0(Q)}\bigg({W_0\over\fiq}\bigg)^{-p_0^\prime}(t)

5656: w_0(t)\,dt\bigg)^{1/p_0^\prime}W_1^{1/p_1}\big(u_1(Q)\big)&\le& Cu_0(Q)\\

5657:  W_1^{1/p_1}\big(u_1(Q)\big)&\le& CW_0^{1/p_0}\big(u_0(Q)\big).

5658: \end{eqnarray*}

5659: \end{enumerate}

5660: \item[(b)] If $0<p_0\le1\/$, (\ref{c7p}) is equivalent to

5661: $$

5662: {W_1^{1/p_1}\big(u_1(Q)\big)\over u_0(Q)}\le C{W_0^{1/p_0}(t)\over\fiq(t)},

5663: \qquad 0<t\le u_0(Q).

5664: $$

5665: \end{enumerate}

5666: We are assuming both in  (a) and   (b) that the inequalities are true for every cube

5667: $Q\subset\Rn\/$, and that the constant $C\/$ is independent of  $Q\/$.

5668: \epro

5669:

5670: \bdem

5671: Observe that, by definition, $\displaystyle\int_0^t\big(u_0^{-1}

5672: \chi_Q\big)^\ast_{u_0}(s)\,ds=\displaystyle{|Q|\over u_0(Q)}\fiq(t)\/$. Hence,  by Theorem~\ref{te: II.4.7}

5673: (with

5674: $r=u_0(Q)\/$),

5675: \begin{eqnarray*}

5676: \big\|u_0^{-1}\chi_Q\big\|_{(\loz)^\prime}&\approx&\displaystyle{|Q|\over u_0(Q)}

5677: \bigg(\int_0^{u_0(Q)}W_0^{1-p_0^\prime}(s)\,d\fiq^{p_0^\prime}(s)\bigg)^{1/p_0^\prime}\\

5678: &\approx&\displaystyle{|Q|\over u_0(Q)}\bigg(\int_0^{u_0(Q)}

5679: \bigg({W_0\over\fiq}\bigg)^{-p_0^\prime}(s)w_0(s)\,ds\bigg)^{1/p_0^\prime}\\

5680: &\qquad+&

5681: {\displaystyle\int_0^\infty(u_0^{-1}\chi_Q)^\ast_{u_0}(s)\,ds\over

5682: W_0^{1/p_0}\big(u_0(Q)\big)},\\

5683: \end{eqnarray*}

5684: in the case $p_0>1\/$, and

5685: $$

5686: \big\|u_0^{-1}\chi_Q\big\|_{(\loz)^\prime}={|Q|\over u_0(Q)}\sup_{t>0}

5687: {\phi_Q(t)\over W_0^{1/p_0}(t)},

5688: $$

5689: in the case $0<p_0\le1\/$. Since  $\big\|\chi_Q\big\|_\lou=W_1^{1/p_1}

5690: \big(u_1(Q)\big)\/$ and \break

5691: $\displaystyle\int_0^\infty\big(u_0^{-1}\chi_Q\big)^\ast_{u_0}(s)\,ds=\displaystyle\int_{\Rn}u_0^{-1}(x)\chi_Q

5692: (x)u_0(x)\,dx=|Q|\/$, the given expression can be immediately deduced  from (\ref{c7p}).

5693: $\qquad\qed$

5694: \edem

5695:

5696: \bobs

5697: In many cases, condition (\ref{c7p}) is also sufficient.

5698: For example:

5699:

5700: (i) If $w_0=w_1=1,\;p_0=p_1=p\ge1\/$ inequality (\ref{c7p}) is

5701: $$

5702: \big\|u_0^{-1}\chi_Q\big\|_{L^{p^\prime}(u_0)}\big\|\chi_Q\big\|_{L^p(u_1)}\le C|Q|,

5703: $$

5704: or equivalently,

5705: \bequ\label{Ap}

5706: \bigg({1\over|Q|}\int_Q u_1(x)\,dx\bigg)\bigg({1\over|Q|}\int_Q u_0^{-1/(p-1)}(x)\,dx\bigg)^{p-1}\le C,

5707: \eequ

5708: in the case $p>1\/$, and

5709: $$

5710: {u_1(Q)\over|Q|}\le C{u_0(E)\over|E|},\qquad E\subset Q,

5711: $$

5712: in the case $p=1\/$.

5713: This is the so-called  $A_p\/$\index{$A_p$} condition for the pair  $(u_1,u_0)\/$ and it is known to be

5714: sufficient for the boundedness

5715: $M:L^p(u_0)\to L^{p,\infty}(u_1)\/$ or, in other terms,

5716: $M:\lox{p}{u_0}{1}\to\loinx{p}{u_1}{1}\/$ (see \cite{GR}).

5717:

5718: (ii) If $u_0=u_1=1,$ then $ \fiq(t)=t,\,0\le t\le1,\/$ and condition (\ref{c7p})

5719: (or  any of the expressions of  Proposition~\ref{ce7p}) is equivalent to  $(w_0,w_1)\in

5720: B_{p_0,p_1,\infty}\/$ (Theorem~\ref{te: I.6.5}). That is, the condition is also sufficient.

5721:

5722: (iii) If $u_0=u_1=u,\,p_0=p_1=q\ge1\/$ and $w_0(t)=w_1(t)=t^{q/p-1},\,1<p<\infty,\/$

5723: we are in the case $M:L^{p,q}(u)\longrightarrow L^{p,\infty}(u)\/$. Then condition  (\ref{c7p})

5724: is equivalent to the inequality

5725: \bequ

5726: \label{Lpq7p}

5727: \big\|u^{-1}\chi_q\big\|_{L^{p^\prime,q^\prime}(u)}\big\|\chi_Q\big\|_{L^{p,q}(u)}

5728: \le C|Q|.

5729: \eequ

5730: This case was studied by  Chung, Hunt, and Kurtz in  \cite{CHK} (see also \cite{HK})

5731: where they proved that (\ref{Lpq7p}) is a necessary and sufficient condition.

5732: \eobs

5733:

5734:

5735: The study of the properties of the functions $\{\phi_Q\}\/$

5736: will allow us to better understand condition  (\ref{c7p}), and obtain some consequences. In the

5737: following proposition, we summarize some of these properties.

5738:

5739: \bpro

5740: \label{6pphiQ}

5741: Let $u\/$ be a weight in  $\Rn\/$ and,

5742: for every cube  $Q\subset\Rn\/$, let $\phi_Q=\phi_{Q,u}\/$. Then,

5743: \begin{enumerate}

5744: \item[(i)] $\fiq(t)={u(Q)\over|Q|}\max\big\{|E|\,:\,E\subset Q,\,u(E)=t\big\},\quad

5745: 0\le t\le u(Q). $

5746: \item[(ii)] ${\fiq}_{|\,[0,u(Q)]}:\big[0,u(Q)\big]\longrightarrow\big[0,u(Q)\big]$ is strictly increasing,

5747: onto, concave and absolutely continuous.

5748:

5749: \item[(iii)] $\fiq(t)\ge t\in[0,u(Q)]\hbox{\ and\ }u\in A_1\hbox{\ if and only if\ }\fiq(t)

5750: \le Ct,\quad 0\le t\le u(Q).  $

5751: \item[(iv)] ${If\ }1<p<\infty,\,0<q\le p^\prime,\quad u\in A_p

5752: \Leftrightarrow\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{L^{p^\prime,q}}\le C.  $

5753: \item[(v)] $\hbox{If\ }u\in A_p,\;1\le p<\infty,\hbox{\ then\ } \fiq^\prime

5754: \big(u(Q)t\big)<Ct^{-1/p^\prime},\;0<t\le1. $

5755: \item[(vi)] $\hbox{If\ }1<p<\infty\hbox{\ and\ }\fiq^\prime\big(u(Q)t\big)

5756: <Ct^{-1/p^\prime},\;0<t\le1,\hbox{\ then\ }u\in\bigcap_{q>p}A_q. $

5757: \end{enumerate}

5758: In (iii), (iv), (v), and (vi), $C\/$ represents a constant not depending on  $Q\/$.

5759: \epro

5760:

5761:

5762: \bdem

5763:

5764: (i) Let $0\le t\le u(Q)\/$ and $E\subset Q\/$ be a measurable set such that $u(E)=t\/$.

5765:  Then

5766: \begin{eqnarray*}

5767: |E|&=&\int_{\Rn}u^{-1}(x)\chi_Q(x)\chi_E(x)\,

5768: u(x)\,dx\le\int_0^\infty\big(u^{-1}\chi_Q\big)^\ast_u(s)\big(\chi_E\big)^\ast_u(s)\,ds\\

5769: &=&\int_0^t\big(u^{-1}

5770: \chi_Q\big)^\ast_u(s)\,ds=\displaystyle{|Q|\over

5771: u(Q)}\fiq(t).\\

5772: \end{eqnarray*}

5773: On the other hand, since  $\big(Q,u(x)\chi_Q(x)dx\big)\/$

5774: is a strongly resonant measure space (see \cite{BS}), there exists $f\ge0\/$  measurable in

5775: $Q\/$ with

5776: $f^\ast_u=\big(\chi_E\big)^\ast_u=\chi_{[0,t)}\/$  and such that

5777: \begin{eqnarray}

5778: \label{eiphiQ}

5779: {|Q|\over u(Q)}\fiq(t)&=&\int_0^\infty\big(u^{-1}\chi_Q\big)^\ast_u(s)

5780: \big(\chi_E\big)^\ast_u(s)\,ds\nonumber\\

5781: &=&\int_{Q}f(x)u^{-1}(x)\chi_Q(x)u(x)\,dx.

5782: \end{eqnarray}

5783: It follows that  $f=\chi_{R}\/$, with $R\subset Q,\;u(R)=t\/$ and, by

5784: (\ref{eiphiQ}), ${|Q|\over u(Q)}\fiq(t)=|R|\/$.

5785:

5786: (ii) Let us see that ${\fiq}_{|\,[0,u(Q)]}\/$ is strictly increasing (the rest of the properties

5787: are obvious). It is enough to prove that  $\big(u^{-1}\chi_Q\big)^\ast_u(t)>0\/$ for $0<t<u(Q)\/$.

5788: On the contrary, there would exist  $t_0\in\big(0,u(Q)\big)\/$ such that

5789: $\big(u^{-1}\chi_Q\big)^\ast_u(t_0)=0\/$ and hence $u\big(\big\{u^{-1}\chi_Q>0\big\}\big)\le

5790: t_0\/$. Since $t_0<u(Q)\/$, we would have that $u\big(\big\{x\in

5791: Q\,:\,u(x)=+\infty\big\}\big)=u\big(\big\{u^{-1}\chi_Q=0\big\}\big)>0\/$,

5792: which contradicts the fact that  $u\/$ is locally integrable.

5793:

5794: (iii) Since  $\fiq\/$ is concave in  $[0,u(Q)],\;t^{-1}\fiq(t)\/$ is decreasing and, thus,

5795: $t^{-1}\fiq(t)\ge u(Q)^{-1}\fiq\big(u(Q)\big)=1\/$ for $0\le t\le u(Q)\/$. The second part of

5796: (iii) follows from  (i) and the fact that $u\in A_1\/$ if and only if

5797: ${u(Q)\over|Q|}\le C{u(E)\over|E|}\/$.

5798:

5799: (iv) By definition (see \cite{St}), $u\in A_p\/$

5800: \begin{eqnarray*}

5801: &\Leftrightarrow&{u(Q)\over|Q|}\bigg({1\over|Q|}\int_Q u^{-p^\prime/p}(x)\,dx\bigg)^{p/p^\prime}

5802: \le

5803: C\\

5804: &\Leftrightarrow&{u(Q)^{1/p}\over|Q|}\bigg(\int_{\Rn}\big(u^{-1}(x)

5805: \chi_Q(x)\big)^{p^\prime}u(x)\,dx\bigg)^{1/{p^\prime}}\le

5806: C_1\\

5807: &\Leftrightarrow&{u(Q)^{1/p}\over|Q|}\bigg(\int_0^\infty{\big(u^{-1}\chi_Q\big)^\ast_u}^{p^\prime}(s)\,ds

5808: \bigg)^{1/{p^\prime}}\le

5809: C_1

5810: \Leftrightarrow\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{p^\prime}\le C_1.

5811: \end{eqnarray*}

5812: Since

5813: $L^{p^\prime,q}\subset L^{p^\prime}\/$ if $0<q\le p^\prime\/$, we have proved the sufficiency in

5814: (iv). Conversely, if $u\in A_p\/$ it is known that  $u\in A_{p-\epsilon}\/$ for some

5815: $\epsilon>0\/$.  Therefore,

5816: by the previous equivalence, $\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{(p-\epsilon)^\prime}

5817: \le

5818: C\/$. Observe that

5819: $\fiq^\prime\big(u(Q)t\big)={u(Q)\over|Q|}\big(u^{-1}\chi_Q\big)^\ast_u\big(u(Q)t\big)=0\/$ for

5820: $t\ge1\/$. Since $L^{p_1,r}(X)\subset L^{p_0,q}(X)\/$ if $p_0<p_1\/$ and $X\/$ is of finite measure

5821: (see \cite{BS}), we have that

5822: $$

5823: \big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{p^\prime,q}\le\big\|\fiq^\prime\big(u(Q)\,

5824: .\big)\big\|_{(p-\epsilon)^\prime}\le

5825: C.

5826: $$

5827:

5828: (v) The case $p=1\/$ is a consequence of  (iii) and the inequality $\fiq^\prime(t)

5829: \le\fiq(t)/t\/$. If $u\in A_p,\;1<p<\infty\/$ then, by (iv), we have that

5830: $\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{p^\prime,\infty}\le\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{p^\prime}\le

5831: C\/$. Since $\fiq^\prime\big(u(Q)\,\cdot\big)\/$ is decreasing, right continuous and equal to zero in

5832: $[1,\infty)\/$,

5833: $\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{p^\prime,\infty}=\sup_{0<t<1}t^{1/p^\prime}

5834: \fiq^\prime\big(u(Q)t\big)\/$

5835: and we obtain (v).

5836:

5837: (vi) The hypothesis implies  $\big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{L^{p^\prime,\infty}}

5838: \le C\/$. But $\fiq^\prime\big(u(Q)\,\cdot\big)\/$ is supported in $[0,1]\/$ and therefore,

5839: $$

5840: \big\|\fiq^\prime\big(u(Q)\,\cdot\big)\big\|_{L^{q^\prime}}\le\big\|\fiq^\prime\big(u(Q)

5841: \,\cdot\big)\big\|_{L^{p^\prime,\infty}}\le C,

5842: $$

5843: for $q>p\/$ and (vi) follows from (iv).$\qquad\qed\/$

5844: \edem

5845:

5846:

5847: Let us see a useful consequence of the two above propositions.

5848:

5849: \bpro

5850: \label{pro: cnadnd}

5851:  If $0<p_0,p_1<\infty\/$ and $M:\loz\to\loinu\/$, then

5852: \begin{eqnarray*}

5853: (i)\quad&\displaystyle{W_1^{1/p_1}\big(u_1(Q)\big)\over u_0(Q)}\le

5854: C\displaystyle{W_0^{1/p_0}(t)\over\fiq(t)},

5855: \qquad 0<t\le u_0(Q),\\

5856: (ii)\quad&\displaystyle{W_1^{1/p_1}\big(u_1(Q)\big)\over u_0(Q)}\le

5857: C\displaystyle{W_0^{1/p_0}\big(u_0(S)\big)

5858: \over u_0(S)},\quad S\subset Q.

5859: \end{eqnarray*}

5860: Here $\phi_Q=\phi_{Q,u_0}\/$ and we are assuming that the previous inequalities are satisfied for

5861: every cube  $Q\subset\Rn\/$, with

5862: $C\/$ independent of  $Q\/$.

5863: \epro

5864:

5865: \bdem

5866:  (ii) is a consequence of  (i) and Proposition~\ref{6pphiQ} (iii).

5867: Hence, we only have to show  (i). If $p_0\le1\/$, (i) is Proposition~\ref{ce7p} (b).

5868: For $p_0>1\/$ we have, by Proposition~\ref{ce7p} (a.i),

5869: $$

5870: \bigg(\int_0^t W_0^{1-p_0^\prime}(s)\,d\fiq^{p_0^\prime}(s)\bigg)^{1/p_0^\prime}W_1^{1/p_1}

5871: \big(u_1(Q)\big)\le Cu_0(Q),

5872: $$

5873: for $0<t\le u_0(Q)\/$. Since $W_0\/$ is increasing, it follows that

5874: $$

5875: W_0^{-1/p_0}(t)\bigg(\int_0^t d\fiq^{p_0^\prime}(s)\bigg)^{1/p_0^\prime}W_1^{1/p_1}

5876: \big(u_1(Q)\big)\le Cu_0(Q),

5877: $$

5878: and we obtain (i).$\qquad\qed$

5879: \edem

5880:

5881: The following result establishes that in the boundedness

5882: $M:\lox{p_0}{u}{w_0}\to\loinx{p_1}{u}{w_1}\/$ we can always assume that  $u\equiv1$.

5883:

5884: \bteo

5885: \label{te: MubMb}

5886:  If $0<p_0,p_1<\infty\/$ and $M:\lox{p_0}{u}{w_0}\to\loinx{p_1}{u}{w_1}\/$,

5887: we have that

5888: $$

5889: M\,:\,\lloz\longrightarrow\lloinu,

5890: $$

5891: that is, $(w_0,w_1)\in B_{p_0,p_1,\infty}\/$.

5892: \eteo

5893:

5894: \bdem

5895: If $p_0>1\/$ then, by Proposition~\ref{ce7p} (a.ii) and using that  $\fiq(t)\ge t,\;0<t<u(Q)\/$

5896: (Proposition~\ref{6pphiQ}  (iii)) we have the inequalities

5897: \begin{eqnarray*}

5898: \bigg(\int_0^{u(Q)}\bigg({W_0(t)\over t}\bigg)^{-p_0^\prime}w_0(t)\,dt\bigg)^{1/p_0^\prime}

5899: W_1^{1/p_1}\big(u(Q)\big)&\le& Cu(Q),\\

5900: W_1^{1/p_1}\big(u(Q)\big)&\le& CW_0^{1/p_0}\big(u(Q)\big),

5901: \end{eqnarray*}

5902: for every cube $Q\subset\Rn\/$. But we have proved  (Proposition~\ref{pro: cnu_0}) that

5903: $u(\Rn)=\infty\/$.

5904: Then, for every  $r>0\/$ there exists a cube  $Q\/$ with $u(Q)=r\/$. Therefore, the two previous

5905: inequalities are equivalent to the condition (a.iii) of Theorem~\ref{te: I.6.5}  and hence,

5906: $(w_0,w_1)\in\BBinnux{p_0,p_1}\/$.

5907:

5908: If $p_0\le1\/$  we apply Proposition~\ref{ce7p}(b)

5909: to obtain the expression of  Theorem~\ref{te: I.6.5} to conclude the same.$\qquad\qed\/$.

5910: \edem

5911:

5912: \bcor

5913: \label{Bpinfi(u)}

5914:  $B_{p,\infty}(u)\subset B_{p,\infty},\;0<p<\infty\/$.

5915: \ecor

5916:

5917: \bobs

5918: \label{obs7p}

5919: (i) If $M:\Lambda^1_u(w)\to\Lambda^{1,\infty}_u(w)\/$ then,

5920: by  Corollary~\ref{Bpinfi(u)}, $w\in B_{1,\infty}\/$. The bigger class $B_p\/$ contained in

5921: $B_{1,\infty}\/$ is $B_1\/$ and if  $w\in B_1\/$, Theorem~\ref{cnsadnd} characterizes the previous

5922: boundedness. That is, the problem $M:\Lambda^1_u(w)\to\Lambda^{1,\infty}_u(w)\/$ only remains open

5923: in the case

5924: $w\in B_{1,\infty}\setminus B_1\/$.

5925:

5926: (ii) Condition (\ref{c7p}) is sufficient in the case

5927: $M:\Lambda^1_u(w)\to\Lambda^{1,\infty}_u(w)\/$ if the weight  $w\/$ satisfies

5928: $$

5929: \sum_{j=1}^n{W(t_j)\over W(r_j)}r_j\le C{W\big(\sum_j t_j\big)

5930: \over W\big(\sum_j r_j\big)}\sum_j r_j,

5931: $$

5932: for every finite families of numbers  $\big\{(t_j,r_j)\big\}_j\/$ with $0<t_j<r_j,\;j=1,

5933: \dots,n\/$. To see this, we observe that  if the previous condition and  (\ref{c7p}) hold (which

5934: implies  (\ref{dEQ})) we have, for every finite family of disjoint cubes  $(Q_j)_j\/$ and every

5935: family of sets

5936: $(E_j)_j\/$ with

5937: $E_j\subset Q_j\/$,

5938: $$

5939: \sum_j{|E_j|\over|Q_j|}u(Q_j)\le C\sum_j{W(u(E_j))\over W(u(Q_j))}u(Q_j)

5940: \le C{W\big(u\big(\bigcup E_j\big)\big)\over W\big(u\big(\bigcup Q_j\big)\big)}\sum_j u(Q_j)

5941: .

5942: $$

5943: This also holds (since $W\in\Delta_2\/$) if the cubes  $Q_j\/$

5944: are \lq\lq almost\rq\rq\ disjoint (if $\sum_j\chi_{Q_j}\le k=k_n\/$). If $0\le f\in\M(\Rn)\/$  is

5945: bounded and with compact support and   $t>0\/$, the level set  $E_t=\{Mf>t\}\/$ is contained

5946: (see \cite{LN1}) in a finite union of cubes $(Q_j)_j\/$ such that $\sum_j\chi_{Q_j}\le k=k_n\/$

5947: (only depending on the dimension) and  ${1\over|Q_j|}\int_{Q_j}f(x)\,dx>t,$ for every $ j\/$. Then,

5948: \begin{eqnarray*}

5949: t\sum_j u(Q_j)&\le&\int_{\R^n} f(x)\bigg(\sum_j{u(Q_j)\over|Q_j|}u^{-1}(x)\chi_{Q_j}(x)\bigg)u(x)\,dx\\

5950: &\le&\|f\|_{\Lambda^1_u(w)}\bigg\|\sum_j{u(Q_j)\over|Q_j|}u^{-1}\chi_{Q_j}

5951: \bigg\|_{(\Lambda^1_u(w))^\prime}\\

5952: \end{eqnarray*}

5953: and, to prove that  $\|Mf\|_{\Lambda^{1,\infty}_u(w)}\le C\|f\|_{\Lambda^1_u(w)}\/$ is bounded,

5954: we only have to see that

5955: $$

5956: \bigg\|\sum_j{u(Q_j)\over|Q_j|}u^{-1}\chi_{Q_j}\bigg\|_{(\Lambda^1_u(w))^\prime}

5957: \lesssim{\sum_j u(Q_j)\over W\big(u\big(\bigcup_j Q_j\big)\big)},

5958: $$

5959: (since then we have that $tW(u(E_t))\le tW\big(u\big(\bigcup_j Q_j\big)\big)

5960: \le C\|f\|_{\Lambda^1_u(w)}\/$).

5961: But, since $\sum_j\chi_{Q_j}\le k\/$,

5962: \begin{eqnarray*}

5963: \int_0^t\bigg(\sum_j{u(Q_j)\over|Q_j|}u^{-1}\chi_{Q_j}\bigg)_u^\ast(s)\,ds&=\sup_{u(E)=t}

5964: \displaystyle\int_E\sum_j{u(Q_j)\over|Q_j|}u^{-1}(x)\chi_{Q_j}(x)\,u(x)\,dx\\

5965: &=\sup_{u(E)=t}\displaystyle\sum_j{|E\cap Q_j|\over|Q_j|}u(Q_j)\\

5966: &\lesssim\displaystyle{W(t)\over W\big(u\big(\bigcup_j Q_j\big)\big)}\sum_j u(Q_j),

5967: \end{eqnarray*}

5968: and  (see Theorem~\ref{te: II.4.7}(i))    the inequality we are looking for follows.

5969:

5970: (iii) If $w\in L^1(\R^+)$, the condition on  $w\/$ of the previous observation can be weakened up:

5971: it is sufficient that it holds \lq\lq near  0\rq\rq, that is, with $\sum r_j<\epsilon\/$ (for some

5972: fixed $\epsilon>0\/$).

5973:

5974: (iv) It is easy to see (using the two previous observations) that, for the weights

5975: $w=\chi_{(0,1)},\;w(t)=\big(\log^+(1/t)\big)^\alpha,\;w(t)=\big(\log^+\log^+(1/t)\big)^\alpha,

5976: \;w(t)=t^\alpha\/$,

5977: condition (\ref{c7p}) is also equivalent (in the case

5978: $p_0=p_1=1,\;u_0=u_1,\;w_0=w_1=w\/$.)

5979: \eobs

5980:

5981: \bteo

5982: \label{Bq/p}

5983:  Let $1\le p<\infty,\,0<q<\infty\/$. Then,

5984:

5985: \medskip

5986: (i)  $B_{q/p}\subset B_{q,\infty}(u)\/$ implies $u\in\bigcap_{r>p}A_r\/$.

5987:

5988: (ii) If $q\le1,\;B_{q,\infty}\subset B_{q,\infty}(u)\/$ implies $u\in A_1\/$.

5989: \eteo

5990:

5991:

5992: \bdem

5993: (i) It is immediate to check (Theorem~\ref{te: I.6.6}) that the weight $w(t)=t^{r-1}\/$ is in

5994: $B_{q/p}\subset B_{q,\infty}(u)\/$, for  $0<r<q/p\/$. Hence, by  Proposition~\ref{pro:

5995: cnadnd} (i),   we have that,

5996: for every cube  $Q\subset\Rn\/$ and $0<t<u(Q)\/$,

5997: $$

5998: \fiq(t)\le Cu(Q){W^{1/q}(t)\over W^{1/q}\big(u(Q)\big)}=C u(Q)^{1-r/q}\,t^{r/q}.

5999: $$

6000: Since $\fiq^\prime(t)\le \fiq(t)/t,\;0<t<u(Q),\/$ we obtain that  $\fiq^\prime\big(u(Q)t\big)

6001: \le Ct^{r/q-1},\;0<t<1,\/$  and by  Proposition~\ref{6pphiQ}(vi), $u\in A_s\/$ for every

6002: $s>q/r\/$. Since this holds for  $0<r<q/p\/$, we conclude that $u\in\bigcap_{r>p}A_r\/$.

6003:

6004: (ii) Observe that  (using, for example,  Theorem~\ref{te: I.6.5})  $w(t)=t^{q-1}\in B_{q,\infty}

6005: \subset B_{q,\infty}(u)\/$. From  Proposition~\ref{pro: cnadnd}(i) (with

6006: $w_0=w_1=t^{q-1}\;,u_0=u_1=u\;,p_0=p_1=q\/$)  we obtain now that  $\fiq(t)\le Ct,\;0<t<u(Q),\/$ and

6007: the result  is a consequence of Proposition~\ref{6pphiQ} (iii). $\qquad\qed$

6008: \edem

6009:

6010: The following result  is related to the  $A_\infty\/$ class.

6011: Let us recall that (see, for example, \cite{St})   $u\in A_\infty=\bigcup_{p\ge1}A_p\/$ if and only

6012: if there exist constants

6013: $\epsilon,C>0\/$ so that

6014: \bequ\label{Ain}

6015: {u(Q)\over|Q|^\epsilon}\ge C{u(E)\over|E|^\epsilon}

6016: \eequ

6017: for every cube  $Q\/$ and every measurable set  $E\subset Q\/$. In general,

6018: the boundedness of $M:\lo\rightarrow\loin\/$

6019: does not imply $u\in A_\infty\/$ (Theorem~\ref{te: pnd}). But we can still  give a sufficient

6020: condition on the weight $w\/$.

6021:

6022: \bpro

6023: \label{W(r)/r^epsilon}

6024: Let us assume that there exist constants $\epsilon,C>0\/$ so that

6025: $$

6026: {W(r)\over r^\epsilon}\ge C{W(t)\over t^\epsilon},\quad0<t<r<\infty.

6027: $$

6028: Then $A_p(w)\subset A_\infty,\;0<p<\infty\/$.

6029: \epro

6030:

6031: \bdem

6032: We can assume  $\epsilon<p\/$. If $u\in A_p(w)\/$, by Proposition~\ref{pro: cnadnd},

6033: $$

6034: \fiq(t)\le C_1{W^{1/p}(t)\over W^{1/p}\big(u(Q)\big)}u(Q)\le C_2

6035: \bigg({t\over u(Q)}\bigg)^{\epsilon/p}u(Q),

6036: $$

6037: for every $Q\/$ and $0<t<u(Q)\/$. Then,

6038: $$

6039: \fiq^\prime\big(u(Q)t\big)\le{1\over u(Q)t}\fiq\big(u(Q)t\big)\le C_2 t^{\epsilon/p-1},\quad0<t<1,

6040: $$

6041: and, from Proposition~\ref{6pphiQ} (vi), it follows that  $u\in A_{q^\prime}\/$, for the range

6042: $q<p/(p-\epsilon).\qquad\qed$

6043: \edem

6044:

6045: \bobs The condition on $w\/$ of  Proposition~\ref{W(r)/r^epsilon} is not equivalent to

6046:   $w\in\bigcup_p B_p\/$ (as it happens with the classes

6047: $A_p\/$). For example, the weight $w=\chi_{(0,1)}\in B_{1,\infty}\/$ does not satisfy that

6048: condition. Even if we impose  $w\notin L^1(\R^+)$ as the example

6049: $w(t)={1/( 1+t)}\/$ shows.

6050:

6051: Power weights $w(t)=t^\alpha\/$  trivially satisfy the condition of

6052: Proposition~\ref{W(r)/r^epsilon}. Another nontrivial example, of a function satisfying such

6053: condition (with

6054: $\epsilon=1/2\/$) is $w(t)=1+\log^+{(1/ t)}\/$.

6055: \eobs

6056:

6057: Up to now, we have only seen necessary conditions for the weak-type boundedness.

6058: Let us see now some sufficient conditions. For example, from  Theorem~\ref{adpp}, we deduce the two

6059: following results.

6060:

6061: \bteo

6062:  Let $0<p,q<\infty\/$ and let us assume that $M:\lo\to\loin\/$ is bounded.

6063: Let $\tilde{w}\/$ another weight. Then we have that

6064: $M:\lox{q}{u}{\tilde{w}}\to\Lambda^{q,\infty}_u(\tilde{w})\/$ is bounded in each of the following

6065: cases:

6066: \begin{enumerate}

6067: \item[(i)] If $0<q\le p\/$ and the condition

6068: $$

6069: {\widetilde{W}^{1/q}(r)\over W^{1/p}(r)}\le C{\widetilde{W}^{1/q}(t)\over W^{1/p}(t)},

6070: \qquad 0<t<r<\infty

6071: $$

6072: holds.

6073:

6074:

6075: \item[(ii)] If $p<q<\infty\/$ and it satisfies

6076: $$

6077: \bigg({1\over\widetilde{W}(t)}\int_0^t\bigg({W(s)\over\widetilde{W}(s)}\bigg)^{q\over q-p}

6078: \tilde{w}(s)\,ds\bigg)^{q-p\over q}\le C{W(t)\over\widetilde{W}(t)},\qquad t>0.

6079: $$

6080: \end{enumerate}

6081: \eteo

6082:

6083: \bdem

6084: Let us fix  $t>0\/$ and let us consider the weights

6085: $$

6086: w_1(s)={w(s)\over W(t)}\chi_{(0,t)}(s),\quad w_0(s)={\tilde{w}(s)\over

6087: \widetilde{W}(t)}\chi_{(0,t)}(s),\quad s>0.

6088: $$

6089: Under the hypothesis of  (i) we have, by Theorem~\ref{twoop} (d)

6090: $$

6091: \sup_{g\d}{\bigg(\displaystyle\int_0^\infty g^p(t)

6092: w_1(t)\,dt\bigg)^{1/p}\over\displaystyle\bigg(\int_0^\infty g^q(t) w_0(t)\,dt\bigg)^{1/q}}

6093: =\sup_{s>0}{W_1^{1/p}(s)\over W_0^{1/q}(s)}\le C.

6094: $$

6095: Therefore,

6096: $$

6097: \bigg({1\over W(t)}\int_0^t (f^\ast_u)^p(s) w(s)\,ds\bigg)^{1/p}\le C\bigg({1\over \widetilde{W}(t)}

6098: \int_0^t (f^\ast_u)^q (s)\tilde{w}(s)\,ds\bigg)^{1/q}

6099: $$

6100: and we obtain  (i).

6101:

6102: To see  (ii), observe that the hypothesis implies, with $r=q/p>1\/$,

6103: $$

6104: B=B(t)=\bigg(\int_0^\infty\bigg({W_1(s)\over W_0(s)}\bigg)^{r^\prime}w_0(s)\,ds\bigg)^{1/r^\prime}+

6105: {W_1(\infty)\over W^{1/r}_0(\infty)}\le C+1<\infty,

6106: $$

6107: and by Theorem~{I.5.7} in \cite{Saw2}, we have

6108: \begin{eqnarray*}

6109: \sup_{g\d}{\bigg(\displaystyle\int_0^\infty g^p

6110: (s)w_1(s)\,ds\bigg)^{1/p}\over\bigg(\displaystyle\int_0^\infty g^q(s) w_0(s)\,ds\bigg)^{1/q}}

6111: &=&\Bigg(\sup_{g\d}{\displaystyle\int_0^\infty g(s) w_1(s)\,ds\over\bigg(\displaystyle\int_0^\infty g^r(s)

6112: w_0(s)\,ds\bigg)^{1/r}}\Bigg)^{1/p}\\  &\le& (C^\prime B)^{1/p}\le

6113: \big(C^\prime(C+1)\big)^{1/p}=C^{\prime\prime}<\infty.

6114: \end{eqnarray*}

6115: In particular,

6116: $$

6117: \bigg({1\over W(t)}\int_0^t\big(f^\ast_u\big)^p(s) w(s)\,ds\bigg)^{1/p}\le C^{\prime\prime}\bigg({1\over

6118: \widetilde{W}(t)}\int_0^t\big(f^\ast_u\big)^q(s) \tilde{w}(s)\,ds\bigg)^{1/q},

6119: $$

6120: for every $t>0\/$ and $f\in\M(\Rn)\/$ and by Theorem~\ref{adpp} we get the result.$\qquad\qed$

6121: \edem

6122:

6123:

6124: \bcor

6125: Let $0<p<\infty\/$ and let us assume that  $M:\lo\to\loin\/$ is bounded. If $0<q\le p\/$ and

6126: $\tilde{w}\/$ is a weight such that  $\widetilde{W}^{1/q}/W^{1/p}\/$  is decreasing, then

6127: $M\,:\,\lox{q}{u}{\tilde{w}}\longrightarrow\loinx{q}{u}{\tilde{w}}\/$ is also bounded.

6128: \ecor

6129:

6130: The following result completes Theorem~\ref{te: MubMb}.

6131:

6132: \bteo

6133: \label{aua_1}

6134: If $u\in A_1\/$, we have the boundedness $M:\Lambda^{p_0}_u(w_0)

6135: \to\Lambda^{p_1,\infty}_u(w_1)\/$, if and only if

6136: $M:\Lambda^{p_0}(w_0)\to\Lambda^{p_1,\infty}(w_1)\/$. In particular

6137: $B_{p,\infty}(u)=B_{p,\infty},\;0<p<\infty,\/$ if $u\in A_1\/$.

6138: \eteo

6139:

6140: \bdem

6141: The \lq\lq only if\rq\rq\  condition is Theorem~\ref{te: MubMb}.

6142: To see the other implication, let us observe that if   $u\in A_1\/$  we have, by

6143: Theorem~\ref{adpp} (applied to

6144: $w\equiv1,\;p=1\/$) that $(Mf)_u^\ast(t)\lesssim Af_u^\ast(t),\;t>0,\;f\in\M(\Rn),\/$ where

6145:  $A\/$ the Hardy operator. Since the boundedness of

6146: $M:\Lambda^{p_0}(w_0)\to\Lambda^{p_1,\infty}(w_1)\/$ is equivalent to  $A:L_{\hbox{\decp dec}}^{p_0}(w_0)\to

6147: L^{p_1,\infty}(w_1)\/$, we get

6148: $$

6149: W^{1/p_1}_1(t)(Mf)_u^\ast(t)\lesssim\|Af_u^\ast\|_{L^{p_1,\infty}(w_1)}\lesssim

6150: \|f_u^\ast\|_{L^{p_0}(w_0)}=\|f\|_{\Lambda_u^{p_0}(w_0)},\qquad t>0,

6151: $$

6152: which is equivalent to  $\|Mf\|_{\Lambda^{p_1,\infty}_u(w_1)}\lesssim\|f\|_{\Lambda^{p_0}_u(w_0)}.

6153: \qquad\qed$

6154: \edem

6155:

6156: Using the same idea, we can prove the following result analogous to Theorem~3.2 in \cite{CS3}.

6157:

6158: \bteo

6159: \label{ B_{p_0,p_1}}

6160:  Let $0<p_0,p_1<\infty,\;1\le p<\infty\/$ and let us assume $u\in A_p\/$.

6161:  Then,

6162: \begin{enumerate}

6163: \item[(i)] If  $(w_0,w_1)\in B_{p_0,p_1}\/$, we have that  $M:\Lambda^{pp_0}_u(w_0)

6164: \to\Lambda^{pp_1}_u(w_1)\/$ is bounded.

6165:

6166: \item[(ii)] If $(w_0,w_1)\in B_{p_0,p_1,\infty}\/$,  then $M:\Lambda^{pp_0}_u(w_0)\to

6167: \Lambda^{pp_1,\infty}_u(w_1)\/$ is bounded.

6168: \end{enumerate}

6169: \eteo

6170:

6171: \bdem

6172:  If $u\in A_p\/$, by Theorem~\ref{adpp}, $(Mf)_u^\ast(t)^p\lesssim

6173: A\big((f_u^\ast)^p\big)(t)\/$. The hypothesis in (i) implies that

6174: $A:L_{\hbox{\decp dec}}^{p_0}(w_0)\to

6175: L^{p_1}(w_1)\/$ and it follows that

6176: \begin{eqnarray*}

6177: \|Mf\|^p_{\Lambda^{pp_1}_u(w_1)}&=&\bigg(\int_0^\infty\big((Mf)_u^\ast\big)^{pp_1}(t)w_1(t)\,dt

6178: \bigg)^{1/p_1}\lesssim\big\|A\big((f_u^\ast)^p\big)\big\|_{L^{p_1}(w_1)}\\

6179: &\lesssim&\|(f_u^\ast)^p\|_{L^{p_0}(w_0)}=\|f\|^p_{\Lambda^{pp_0}_u(w_0)}.

6180: \end{eqnarray*}

6181: Analogously, one can easily prove  (ii). $\qquad\qed$

6182: \edem

6183:

6184:

6185: If $p>1,\;p_0\le p_1\/$,  (ii) can be improved in the following way:

6186:

6187: \bcor

6188: Let $0<p_0\le p_1<\infty\/$. If $u\in A_p,\;p>1,\/$

6189: and  $(w_0,w_1)\in B_{p_0,p_1,\infty}\/$, then

6190: $M:\Lambda^{pp_0}_u(w_0)\to\Lambda^{pp_1}_u(w_1)\/$ is bounded.

6191: \ecor

6192:

6193: \bdem

6194: If  $u\in A_p\/$ then $u\in A_q\/$ for some $q\in(1,p)\/$,

6195: and by the previous theorem, $M:\Lambda^{qp_0}_u(w_0)\to\Lambda^{qp_1,\infty}_u(w_1)\/$ is bounded.

6196: Applying now the interpolation theorem (Theorem~\ref{te: II.6.3}) we obtain the boundedness of

6197: $M:\Lambda^{pp_0}_u(w_0)\to\Lambda^{pp_1,pp_0}_u(w_1)\subset\Lambda^{pp_1}_u(w_1).\qquad\qed$

6198: \edem

6199:

6200:

6201: \bobs

6202: In \cite{Ne2},  Neugebauer introduces the classes $A_p^\ast\/$\index{$A_p^\ast\/$}

6203: which are defined by

6204: $$

6205: A_p^\ast=\big\{\,u\in A_\infty\,:\,p=\inf\{q\ge1\,:\,u\in A_q\}\,\big\},\qquad1\le p<\infty.

6206: $$

6207: These classes are pairwise disjoint and the union of all of them is $A_\infty\/$.

6208: With this terminology, Theorem~\ref{sol-uA_1} states that $B_p(u)=B_p,\;0<p<\infty,\/$ if

6209: $u\in A_1^\ast\/$ (and this condition is also necessary). From Theorems~\ref{Bq/p} and \ref{

6210: B_{p_0,p_1}} we can deduce   (using the property   $w\in B_p\Rightarrow w\in

6211: B_{p-\epsilon}\/$) that, for

6212: $1<p<\infty,\;0<q<\infty\/$,

6213: \medskip

6214:

6215: (a)  $u\in A^\ast_p\,\Rightarrow\,B_{q/p}\subset B_q(u)\/$,

6216:

6217: (b) $B_{q/p}=B_q(u)\,\Rightarrow u\in A_p^\ast\/$.

6218:

6219: \medskip

6220: This cannot be improved (as it happens in the case $p=1\/$), that is, it is not true that the

6221: characterization of the classes

6222: $B_q(u)\/$ when

6223: $u\in A_\infty\/$, is

6224: $B_q(u)=B_{p/q}\/$ for  $p\/$ such that $u\in A^\ast_p\/$. In fact, it is easy to check that the

6225: weight

6226: $u(x)=1+|x|,\;x\in\R,\/$ is in  $A_2^\ast\/$ and, by Theorem~\ref{te:principal} (see also Remark~

6227: \ref{ob:disj}(ii)),

6228: $w=\chi_{(0,1)}\in B_{3/2}(u)\/$ (condition  (\ref{cprincipal}) holds with $q=1\/$). However,

6229: $w\notin B_{3/4}\/$ and thus, $B_{3/2}(u)\neq B_{3/4}\/$.

6230: \eobs

6231:

6232: We shall study now a sufficient condition to have the weak-type boundedness.

6233: To this end, we shall need the following notation. We shall associate to each weight  $u \/$ in

6234: $\Rn\/$ a function

6235: $\Fiu\/$ which is connected with the family of functions $\big\{\fiq\big\}_Q\/$ introduced in

6236: (\ref{defphiQ}). This new function is defined by

6237: \bequ

6238: \label{def-Phi_u}

6239: \Fiu(t)=\sup_{Q}\fiq^\prime\big(u(Q)\,t\big),\quad0<t<\infty,

6240: \eequ

6241: where the supremum is taken over all cubes $Q\subset\Rn\/$. Let us observe that

6242: $$

6243: \Fiu(t)=\sup_{Q}{u(Q)\over|Q|}\big(u^{-1}\chi_Q\big)^\ast_u\big(u(Q)\,t\big),\quad0<t<\infty.

6244: $$

6245: Therefore, $\Fiu(t)=0\/$ if $t\ge1\/$. Moreover, by Proposition~\ref{6pphiQ}(iii),

6246: $\displaystyle\int_0^t\Fiu(s)\,ds\ge t,\;0<t<1\/$. We know that if $u\equiv1,\;(Mf)_u^\ast(t)\approx

6247: A(f_u^\ast)(t),\;t>0,\/$ where $A\/$ is the  Hardy operator. And although this does not happen when

6248: $u\ne1\/$ (see \cite{CS3}) there exists  the following  positive partial result  due

6249: to  Leckband and Neugebauer (\cite{LN1}).

6250:

6251: \bteo

6252: \label{te:LN1}

6253: Let $u\/$ be a weight in $\Rn\/$. Then,

6254: for each $f\in\M(\Rn)\/$ we have that

6255: $$

6256: \big(Mf\big)^\ast_u(t)\le C\int_0^\infty\Fiu(s)f_u^\ast(st)\,ds,\quad0<s<\infty,

6257: $$

6258: where $C\/$ is a constant depending only on the dimension.

6259: \eteo

6260:

6261: Using this theorem,  we can find a sufficient condition to have the boundedness

6262: $M:\Lambda^{p_0}_u(w_0)\to\Lambda^{p_1,\infty}_u(w_1)\/$. Observe the analogy with the

6263: corresponding expressions of Proposition~\ref{ce7p}.

6264:

6265: \bteo

6266: \label{csamdnd}

6267:  Let $0<p_1<\infty\/$.

6268: \begin{enumerate}

6269: \item[(a)] If  $1<p_0<\infty\/$ and there exists a constant $C<\infty\/$ such that

6270:

6271: (i)

6272: $\displaystyle\bigg(\int_0^r\bigg(\int_0^{t/r}\Fiu(s)\,ds\bigg)^{p_0^\prime}W_0^{-p_0^\prime}(t)w_0(t)

6273: \,dt\bigg)^{1/p_0^\prime}W_1^{1/p_1}(r)\le C,\quad r>0,$

6274:

6275: (ii) $W_1^{1/p_1}(r)\le CW_0^{1/p_0}(r),\quad r>0,$

6276:

6277: then $M:\Lambda^{p_0}_u(w_0)\to\Lambda^{p_1,\infty}_u(w_1)\/$ is bounded.

6278:

6279: \item[(b)] If $0<p_0\le1\/$, we have the same if the following condition holds

6280: $$

6281: \int_0^{t/r}\Fiu(s)\,ds\le C{W_0^{1/p_0}(t)\over W_1^{1/p_1}(r)},\quad0<t<r<\infty.

6282: $$

6283: \end{enumerate}

6284: \eteo

6285:

6286: \bdem

6287: Let us consider the operator

6288: $$

6289: Tg(t)=\int_0^\infty\Fiu(s)g(st)\,ds=\int_0^\infty k(t,s)g(s)\,ds,

6290: $$

6291: (with $k(t,s)=(1/t)\Fiu(s/t)\/$) acting on functions $g\d\/$. By Theorem~\ref{te:LN1},

6292: the boundedness of

6293: \bequ

6294: \label{adT}

6295: T\,:\,L^{p_0}_{\hbox{\decp dec}}(w_0)\longrightarrow L^{p_1,\infty}(w_1)

6296: \eequ

6297: implies the result. The characterization of  (\ref{adT})  can be obtained as a

6298: direct application of  Theorem~4.3 in \cite{CS2}.

6299: $\qquad\qed$

6300: \edem

6301:

6302: \bobs

6303: \label{pobs}

6304:

6305: (i) The sufficient condition of the previous theorem is not necessary in general.

6306: For example, if $w_0=w_1=w=\chi_{(0,1)},\;p_0=p_1=1\/$, such condition is

6307: $\displaystyle\int_0^t\Fiu(s)\,ds\lesssim t,\;0<t<1,\/$ and by  Proposition~\ref{6pphiQ}, it is equivalent

6308: to

6309: $u\in A_1\/$. That is,

6310: $A_1\subset A_1(\chi_{(0,1)})\/$. However, for this weight,   condition (\ref{c7p}) is

6311: necessary and sufficient in this case (see Remark~\ref{obs7p}). This condition is

6312: (Proposition~\ref{ce7p}b)

6313: $\fiq(t)\lesssim\displaystyle{u(Q)\over W(u(Q))}W^{-1}(t),\;t>0,\/$ and, by Proposition~\ref{6pphiQ}(i), it

6314: is equivalent to

6315: $$

6316: {W\big(u(Q)\big)\over|Q|}\le C{W\big(u(E)\big)\over|E|},\qquad E\subset Q.

6317: $$

6318: It is a simple exercise to check that (with $w=\chi_{(0,1)}\/$) the weight $u(x)=1+|x|,\;x\in\R,\/$

6319: satisfies this condition. However,  $u\notin A_p\/$ if $p\le2\/$, and it follows that

6320: $A_1(\chi_{(0,1)})\neq A_1\/$, that is, the condition of Theorem~\ref{csamdnd} is not necessary.

6321:

6322: (ii) Let us assume that the weight $u\/$ satisfies the following property: for every $r>0\/$  there

6323: exists a cube

6324: $Q_r\/$ so that

6325:

6326: \medskip

6327: \ \ \ (a)\ \ $u(Q_r)\approx r\/$,

6328:

6329: \medskip

6330: \ \ \ (b)\ \ $\displaystyle\int_0^t\Fiu(s)\,ds\approx\displaystyle\int_0^t\phi_{Q_r}(u(Q_r)s)\,ds,\qquad

6331: 0<t<1\/$.

6332:

6333: Then the condition of the previous theorem is equivalent to  (\ref{c7p}) and,

6334: hence, both are equivalent to the boundedness of

6335: $M:\Lambda^{p_0}_u(w_0)\to\Lambda^{p_1,\infty}_u(w_1)\/$. This follows immediately from the

6336: expressions of Theorem~\ref{csamdnd} and by  Proposition~\ref{ce7p} (with

6337: $u_0=u_1=u\/$) since, fixed $r>0\/$  one can substitute, in the expressions of

6338: Theorem~\ref{csamdnd}, $r\/$ by

6339: $u(Q_r)\/$ and the integral

6340: $\displaystyle\int_0^{t/r}\Fiu(s)\,ds\/$ by $\displaystyle\int_0^{t/r}\fiq(u(Q_r)s)\,ds\/$ to obtain the

6341: condition of  Proposition~\ref{ce7p} and viceversa.

6342:

6343: (iii) Every power weight $u(x)=|x|^\alpha,\;x\in\Rn,\;\alpha>0\/$ satisfies the previous condition.

6344: In fact, if $Q\/$ is a cube centered at the origin, one can easily see that

6345: $$

6346: \fiq(u(Q)t)\approx\Fiu(t)\approx t^{-\alpha\over n+\alpha},\qquad 0<t<1.

6347: $$

6348: Therefore, the characterization of the boundedness of

6349: $M:\lox{p_0}{u}{w_0}\to\lox{p_1,\infty}{u}{w_1}\/$

6350: is given in this case by the expressions of Theorem~\ref{csamdnd} (which is now equivalent to

6351: (\ref{c7p}). See also Theorem~5.7.

6352: \eobs

6353:

6354:

6355: \section{Applications}

6356: \label{se:applic}

6357:

6358: We can use now the result of the previous sections,

6359:  to characterize in its total generality, the boundedness of

6360: \bequ

6361: \label{aLpq}

6362: M:L^{p,q}(u)\to L^{r,s}(u),

6363: \eequ

6364: completing results of \cite{Muck1,  CHK, HK, L}.

6365:

6366: \bteo Let $p,r\in(0,\infty),\;q,s\in(0,\infty]\/$.

6367: \begin{enumerate}

6368: \item[(a)] If either $p<1\/$,   $p\neq r\/$ or $s<q\/$, there are no weights $u\/$

6369:  satisfying the boundedness   $M:L^{p,q}(u)\to L^{r,s}(u)\/$.

6370:

6371: \item[(b)] The boundedness of

6372: $$

6373: M\,:\,L^{1,q}(u)\longrightarrow L^{1,s}(u)

6374: $$

6375: only holds if  $q\le1,\;s=\infty\/$ and, in this case, a necessary and sufficient condition is

6376: $u\in A_1\/$.

6377:

6378: \item[(c)] If $p>1\/$ and $0<q\le s\le\infty\/$, a necessary and sufficient condition

6379: to have the boundedness of

6380: $$

6381: M\,:\,L^{p,q}(u)\longrightarrow L^{p,s}(u)

6382: $$

6383: is:

6384: \begin{enumerate}

6385: \item[(1)]  If $q\le1,\;s=\infty\,:\quad \displaystyle{u(Q)\over|Q|^p}\le C{u(E)\over|E|^p},

6386: \quad E\subset Q\/$.

6387:

6388: \item[(2)] If $q>1\/$ or   $s<\infty\,:\quad u\in A_p\/$.

6389: \end{enumerate}

6390: \end{enumerate}

6391: \eteo

6392:

6393: \bdem

6394: (a) By Proposition~\ref{cnalpq}, the boundedness of

6395: \bequ

6396: \label{alpqrs}

6397: M\,:\,L^{p,q}(u)\longrightarrow L^{r,s}(u)

6398: \eequ

6399: implies $s\ge q\/$. On the other hand, (\ref{alpqrs}) implies the boundedness of  $M:L^{p,q}(u)\to

6400: L^{r,\infty}(u)\/$

6401: which is  equivalent to $M:\lox{q}{u}{t^{q/p-1}}\to\lox{r,\infty}{u}{1}\/$ and from

6402: Corollary~\ref{cor: cnad} it follows that

6403: \bequ

6404: \label{u(Q)|Q|}

6405: {\big(u(Q)\big)^{1/r}\over|Q|}\le C{\big(u(E)\big)^{1/p}\over|E|},\qquad E\subset Q.

6406: \eequ

6407: By the Lebesgue differentiation theorem, we have that necessarily  $p\ge1\/$. Moreover, applying

6408: (\ref{u(Q)|Q|})  with $E=Q\/$ we obtain that $\big(u(Q)\big)^{1/r-1/p}\le C\/$ and since  $u(Q)\/$

6409: can take any value in $(0,\infty)\/$ (Proposition~\ref{pro: cnu_0}), we get  $p=r\/$.

6410:

6411: (b) Since the norm of a characteristic function in $L^{1,q}\/$ does not depend on  $q\/$, the

6412: boundedness in (b) implies

6413: $\|M\chi_E\|_{L^{1,s}(u)}\le C\|\chi_E\|_{L^1(u)},\;E\subset Q\/$. If $s<\infty\/$ we obtain, from

6414: Theorem~\ref{apfciaq}, the boundedness of

6415: $M:L^p(u)\to L^p(u)\/$ with $p<1\/$ and, by (a), we have a contradiction.  Therefore,

6416: $s=\infty\/$. On the other hand, the boundedness of  $M:L^{1,q}(u)\to L^{1,\infty}(u)\/$ is the same

6417: than

6418: $M:\lox{q}{u}{t^{q-1}}\to\lox{q,\infty}{u}{t^{q-1}}\/$ and by Corollary~\ref{cor: cnad} we obtain

6419: ${u(Q)/|Q|}\le C{u(E)/|E|},\;E\subset Q,\/$ which is  $u\in A_1\/$. We know that this

6420: condition is sufficient if

6421: $q\le1\/$. But from Corollary~\ref{Bpinfi(u)} we get $t^{q-1}\in B_{q,\infty}\/$,  and this is

6422: only possible (see Theorem~\ref{te: I.6.5}) if $q\le1\/$.

6423:

6424: (c) If $q\le s<\infty\/$, by Theorem~\ref{apfciaq} and using interpolation, the boundedness of

6425:  $M:L^{p,q}(u)\to L^{p,s}(u)\/$ is equivalent to the boundedness of $M:L^p(u)\to L^p(u)\/$ and a

6426: necessary and sufficient condition is

6427: $u\in A_p\/$. In the case $M:L^{p,q}(u)\to L^{p,\infty}(u)\/$ (that is

6428: $s=\infty\/$) we have two possibilities: (i) if $q>1\/$ a necessary and sufficient condition is

6429: (see \cite{CHK}) $u\in A_p\/$, and  (ii) if $q\le1\/$, from Corollary~\ref{cor: cnad} we obtain

6430: the condition

6431: ${u(Q)/|Q|^p}\le{u(E)/|E|^p}\/$. In \cite{CHK} it is shown that this condition is

6432: sufficient in the case

6433: $q=1\/$ and  (since

6434: $L^{p,q}\subset L^{p,1}\/$ if $q\le1\/$) also in the case

6435: $q\le1.\qquad\qed$

6436: \edem

6437:

6438: \bobs

6439:

6440: (i) The cases  $M:L^1(u)\to L^{1,\infty}(u)\/$ and $M:L^{p,q}(u)\to L^{p,s}(u)\/$

6441: with  $1<p\le q\le s\le\infty\/$ were solved by  Muckenhoupt in \cite{Muck1}, giving rise to the

6442: $A_p\/$ classes. Chung, Hunt, and Kurtz (\cite{CHK, HK}) solved the case $M:L^{p,q}(u)\to L^{p,s}(u)\/$  with

6443: $p>1,\;1\le q\le s\le\infty\/$ and  Lai (\cite{L}) proved the necessity of the condition  $p=r\/$

6444: to have (\ref{aLpq}).

6445:

6446: (ii) The conditions obtained  in the cases of weak-type inequalities

6447: ($q,s<\infty\/$) coincide with  (\ref{c7p}).

6448: \eobs

6449:

6450: The following result characterizes the boundedness of

6451: \bequ

6452: \label{bLpqu}

6453: M:\lox{p_0}{u}{w_0}\to\lox{p_1,\infty}{u}{w_1}

6454: \eequ

6455: when $u(x)=|x|^\alpha,\;x\in\Rn\/$.

6456:

6457: \bteo

6458: Let $u(x)=|x|^\alpha,\;x\in\Rn,\;\alpha>-n\/$.

6459: \begin{enumerate}

6460: \item[(a)] If $\alpha\le0\/$, {\rm(\ref{bLpqu})} is equivalent to $(w_0,w_1)\in B_{p_0,p_1,\infty}\/$.

6461:

6462: \item[(b)] If $\alpha>0\/$    {\rm(\ref{bLpqu})} holds if and only if  $(\bar{w}_0,\bar{w}_1)\in

6463: B_{p_0,p_1,\infty}\/$,

6464:  where for each  $i=0,1\/$,

6465: $$

6466: \bar{w}_i(t)=w_i\big(t^{n+\alpha\over n}\big)\,t^{\alpha\over n},\qquad t>0.

6467: $$

6468: \end{enumerate}

6469: \eteo

6470:

6471: \bdem

6472:   (a) is consequence of Theorem~\ref{aua_1}, since  $u(x)=

6473: |x|^\alpha\in A_1\/$ if $\alpha\le0\/$. To prove (b) we use the condition of

6474: Theorem~\ref{csamdnd}

6475:  that  (by Remark~\ref{pobs}(iii)) is necessary and sufficient.  The final condition is obtained

6476: making the change of variables $t=\bar{t}^{(n+\alpha)/n},\;r=\bar{r}^{(n+\alpha)/n}\/$

6477: and comparing the expression we get with the corresponding to the classes

6478: $B_{p_0,p_1,\infty}\/$ (Theorem~\ref{te: I.6.5}).$\qquad\qed$

6479: \edem

6480:

6481: \vfill\eject

6482: \markboth{BIBLIOGRAPHY}{M.J. Carro, J.A. Raposo, and J. Soria}

6483:

6484: \begin{thebibliography}{CPSS}

6485: \addcontentsline{toc}{chapter}{Bibliography}

6486:

6487: \bibitem[Al]{Al} G.D. Allen, {\sl Duals of Lorentz spaces}, Pacific J.  Math. {\bf 177} (1978), 287--291.

6488:

6489: \bibitem[AEP]{AEP} M.A. Ari\~no, R. Eldeeb, and N.T. Peck, {\sl The Lorentz sequence spaces $d(w,p)\/$ where

6490: $w\/$ is increasing}, Math. Ann. {\bf 282} (1988), 259--266.

6491:

6492: \bibitem[AM1]{AM1} M.A. Ari\~no and  B. Muckenhoupt, {\sl Maximal functions on classical Lorentz spaces and

6493: Hardy's  inequality with weights for nonincreasing functions}, Trans. Amer. Math. Soc. {\bf 320} (1990),

6494: 727--735.

6495:

6496: \bibitem[AM2]{AM2} M.A. Ari\~no and  B. Muckenhoupt, {\sl A characterization of the dual of the classical

6497: Lorentz  sequence space $d(w,q)\/$}, Proc. Amer. Math. Soc. {\bf 102} (1991), 87--89.

6498:

6499: \bibitem[BPS]{BPS} S. Barza, L.E. Persson, and J. Soria, {\sl Sharp weighted multidimensional integral

6500: inequalities for monotone functions}, Math. Nachr.  {\bf 210} (2000), 43--58.

6501:

6502: \bibitem[BS]{BS} C. Bennet and R. Sharpley, {\sl Interpolation of Operators}, Academic Press, 1988.

6503:

6504: \bibitem[BL]{BL} J. Bergh and J. L\"ofstr\"om, {\sl Interpolation Spaces. An Introduction}, Springer-Verlag, New

6505: York, 1976.

6506:

6507: \bibitem[CH]{CH} J.R. Calder and J.B. Hill,  {\sl A collection of sequence spaces}, Trans. Amer. Math. Soc. {\bf

6508: 152} (1970), 107--118.

6509:

6510: \bibitem[CGS]{CGS} M.J. Carro, A. Garc\'{\i}a del Amo, and J. Soria,  {\sl Weak-type

6511: weights and normable Lorentz spaces}, Proc. Amer. Math. Soc. {\bf 124} (1996),

6512: 849--857.

6513:

6514: \bibitem[CPSS]{CPSS} M.J. Carro, L. Pick, J. Soria, and V. Stepanov,  {\sl On embeddings between classical

6515: Lorentz spaces}, Math. Inequal. Appl. (To appear).

6516:

6517: \bibitem[CS1]{CS1} M.J. Carro and J. Soria,  {\sl Weighted Lorentz spaces and the Hardy operator}, J. Funct.

6518: Anal. {\bf 112} (1993), 480 --494.

6519:

6520: \bibitem[CS2]{CS2} M.J. Carro and J. Soria,  {\sl Boundedness of some integral operators}, Canad. J. Math. {\bf

6521: 45} (1993), 1155--1166.

6522:

6523: \bibitem[CS3]{CS3} M.J. Carro and J. Soria,  {\sl The Hardy-Littlewood maximal function and weighted Lorentz

6524: spaces}, J. London Math. Soc. {\bf 55} (1997), 146--158.

6525:

6526: \bibitem[CM]{CM} J. Cerd\`a and J. Mart\'{\i}n,  {\sl Interpolation restricted to decreasing functions and

6527: Lorentz spaces}, Proc. Edinburgh Math. Soc.  {\bf 42} (1999),  243--256.

6528:

6529: \bibitem[CHK]{CHK} H.M. Chung, R.A. Hunt, and D.S. Kurtz,  {\sl The Hardy-Littlewood maximal function on

6530: $L(p,q)$ spaces with weights}, Indiana Univ. Math. J. {\bf 31} (1982), 109--120.

6531:

6532: \bibitem[CF]{CF} R.R. Coifman and C. Fefferman,  {\sl Weighted norm inequalities for maximal functions and

6533: singular integrals}, Studia Math. {\bf 51} (1974), 241--250.

6534:

6535: \bibitem[Cw1]{Cw1} M. Cwikel,  {\sl On the conjugates of some function spaces}, Studia Math. {\bf 45}

6536: (1973), 49--55.

6537:

6538: \bibitem[Cw2]{Cw2} M. Cwikel,  {\sl The dual of weak $L^p\/$}, Ann. Inst. Fourier (Grenoble) {\bf 25}

6539: (1975), 81--126.

6540:

6541: \bibitem[CS]{CS} M. Cwikel and Y. Sagher,  {\sl $L(p,\infty)^\ast\/$}, Indiana Univ. Math. J. {\bf 21} (1972),

6542: 781--786.

6543:

6544: \bibitem[GR]{GR} J. Garc\'{\i}a Cuerva and J.L. Rubio de Francia, {\sl Weighted Norm Inequalities and Related

6545: Topics}, vol. 46, North-Holland, 1985.

6546:

6547: \bibitem[Ga]{Ga} A.J.H. Garling,  {\sl A class of reflexive symmetric BK-spaces}, Canad. J.  Math. {\bf 21}

6548: (1969), 602--608.

6549:

6550: \bibitem[Ha]{Haa} A. Haaker,  {\sl On the conjugate space of Lorentz space}, University

6551: of Lund (1970).

6552:

6553: %\bibitem[HLP]{HLP} G.H. Hardy, J.E. Littlewood, and  G. Polya, {\sl Inequalities}, Cambridge Univ. Press,

6554: %1952.

6555:

6556: \bibitem[HM]{HM} H. Heinig and L. Maligranda,  {\sl Weighted inequalities for monotone and concave functions},

6557: Studia Math. {\bf 116} (1995), 133--165.

6558:

6559: \bibitem[Hu]{Hu} R.A. Hunt, {\sl On $L(p,q)$ spaces}, Enseignement Math. {\bf 12} (1966), 249--276.

6560:

6561: \bibitem[HK]{HK} R.A. Hunt and D.S. Kurtz, {\sl The Hardy-Littlewood maximal function on $L(p,1)$}, Indiana

6562: Univ. Math. J. {\bf 32} (1983), 155--158.

6563:

6564: \bibitem[K\"o]{Ko} G. K\"othe, {\sl Topological Vector Spaces I}, Springer-Verlag, 1969.

6565:

6566: \bibitem[KPS]{KPS} S.G. Krein, Ju.I. Petunin, and E.M. Semenov, {\sl Interpolation of Linear Operators},

6567: Transl. Math. Monographs, vol. 54, 1982.

6568:

6569: \bibitem[La]{L} Q. Lai, {\sl A note on weighted maximal inequalities}, Proc. Edimburgh Math. Soc. {\bf 40}

6570: (1997), 193--205.

6571:

6572: \bibitem[LN1]{LN1} M.A. Leckband and C.J. Neugebauer, {\sl A general maximal operator and the

6573: $A_p$-condition}, Trans. Amer. Math. Soc. {\bf 275} (1983), 821--831.

6574:

6575: \bibitem[LN2]{LN2} M.A. Leckband and C.J. Neugebauer, {\sl Weighted iterates and variants of the

6576: Hardy-Littlewood maximal operator}, Trans. Amer. Math. Soc. {\bf 279} (1983), 51--61.

6577:

6578: \bibitem[Lo1]{Lor2} G.G. Lorentz,  {\sl Some new functional spaces}, Ann. of

6579: Math. {\bf 51} (1950), 37--55.

6580:

6581: \bibitem[Lo2]{Lor} G.G. Lorentz,  {\sl On the theory of spaces $\Lambda$}, Pacific J.

6582: Math. {\bf 1} (1951), 411--429.

6583:

6584: \bibitem[Mu]{Muck1} B. Muckenhoupt,  {\sl Weighted norm inequalities for the Hardy maximal function}, Trans.

6585: Amer. Math. Soc. {\bf 165} (1972), 207--227.

6586:

6587: \bibitem[MW]{MW} B. Muckenhoupt and R.L. Wheeden,  {\sl Weighted norm inequalities for fractional

6588: integrals}, Trans. Amer. Math. Soc. {\bf 192} (1974), 261--275.

6589:

6590: \bibitem[NO]{NO} M. Nawrocki and A. Orty\'{n}ski, {\sl The Mackey topology and complemented subspaces of Lorentz

6591: sequence spaces $d(w,p)\/$ for $0<p<1\/$}, Trans. Amer. Math. Soc. {\bf 287} (1985), 713--722.

6592:

6593: \bibitem[Ne1]{Ne1} C.J. Neugebauer,  {\sl Weighted norm inequalities for average operators of monotone

6594: functions}, Publ. Mat. {\bf 35} (1991), 429--447.

6595:

6596: \bibitem[Ne2]{Ne2} C.J. Neugebauer,  {\sl Some classical operators on Lorentz spaces}, Forum Math. {\bf 4}

6597: (1992), 135--146.

6598:

6599: \bibitem[Po]{Po} N. Popa,  {\sl Basic sequences and subspaces in Lorentz sequence spaces without local

6600: convexity}, Trans. Amer. Math. Soc. {\bf 263} (1981), 431--456.

6601:

6602: \bibitem[Re]{Re} S. Reiner,  {\sl On the duals of Lorentz function and sequence spaces}, Indiana  Univ. Math. J.

6603: {\bf 31} (1982), 65--72.

6604:

6605: \bibitem[Ru]{Ru2} W. Rudin,  {\sl Functional Analysis}, McGraw-Hill, Inc. International Series in Pure and

6606: Applied Mathematics, 1991.

6607:

6608: \bibitem[Sa]{Saw2} E. Sawyer,  {\sl Boundedness of classical operators on classical

6609: Lorentz spaces}, Studia Math. {\bf 96} (1990), 145--158.

6610:

6611: \bibitem[SS]{SS} G. Sinnamon and V. Stepanov,  {\sl The weighted Hardy inequality: new proofs and the case

6612: $p=1$}, J. London Math. Soc. {\bf 154} (1996), 89--101.

6613:

6614: \bibitem[So]{Sor} J. Soria,  {\sl Lorentz spaces of weak-type}, Quart. J.  Math.

6615: Oxford {\bf 49} (1998), 93--103.

6616:

6617: \bibitem[Stn]{St} E.M. Stein,  {\sl Harmonic Analysis. Real-Variable Methods, Orthogonality, and Oscillatory

6618: Integrals}, Princeton University Press, 1993.

6619:

6620: \bibitem[SW]{SW} E.M. Stein and G. Weiss,  {\sl Introduction to Fourier Analysis on Euclidean Spaces}, Princeton

6621: University Press, 1971.

6622:

6623: \bibitem[Stp1]{Sp1} V. Stepanov,  {\sl Integral operators on the cone of monotone functions}, Institute for

6624: Applied Mathematics. F-E.B. of the USSR Academy of Sciences, Khabarovsk, 1991.

6625:

6626: \bibitem[Stp2]{Sp2} V. Stepanov, {\sl The weighted Hardy's inequality for nonincreasing functions}, Trans. Amer.

6627: Math. Soc. {\bf 338} (1993), 173--186.

6628:

6629:

6630:

6631: \end{thebibliography}

6632: \vfill\eject

6633: \addcontentsline{toc}{chapter}{Index}

6634: \markboth{INDEX}{M.J. Carro, J.A. Raposo, and J. Soria}

6635:

6636: \input{CRS.ind}

6637:

6638: \end{document}