METHODS OF MODERN MATHEMATICAL PHYSICS I: FUNCTIONAL ANALYSIS Revised and Enlarged Edition wMrcHAEL neED anny sion seria Papen of Manas @ ‘ACADEMIC PRESS, INC San igo” Now Yok Betaey Boon Londen Syaney Tee romTo RS. Philips and A. S. Wightman, Mentors, Colleagues, Friends Conmaur © 1900, 2¥ Acabe Pres, ne SR Beale: ha Dio, cass 92101 ony of Cones Can Febenton Dts thd mater materi physic. Ve Fantini nd ened ion lace tinal feces, onTes Feta 2 oe nat optare 1 Mamata phys 1 Sinn, Bary on SEF sss M5 (405) 1970 jet Casino: 6-02, 47-02, 42-02 Preface ‘This book i the fist of ¢ muvolume serie devoted oan exposition of une ‘ional analysis methods in modem mathematical physics, It deeribes the nde ‘mental principles of funcdonal analysis and estenally self-contained, a ‘hough here are occasional references out volumes, We hve included fo application when we thought tha hey would provide metvstion forthe read. {Later volumes deseribe various advanced topics in functional analysis ad give ‘umerous applications in classical physics, modem physics, and paral cifferen- tial equats. ‘This revised and enlarged eition liters from the fit in two major way. ‘int, many colleagues have suggested tous that woul be helpfl inca some material onthe Fourier transform in Volume Iso tht thi important topic an be conveniently included in «standard funtion eaalyis couse using is book, Thus, we have included inthis dion Sectons 1X1, 1X2, and pat of 12 fom Volume and some ational material, ogetir with evan mts and problems. Secondly, we have iced vary of supplementary mei atte ead ofthe book. Some ofthese supplementary sections provide profs of theorems in Chapters I-IV which were omitedin the it eon, While these roofs make Chapter I-IV more sel-omsined, we sil recommiend tha sti ‘ents wih no previous experience with this material consult more elmentry texts. Other supplementary sections provide expository material to ad the in stucior andthe student (for example, "Appleton of Compact Operators") Sill ote sections inoduce and develop new materi (or example,“ Misimi- ation of Functional"). Te gves us please to thank many indivi: The sidents who wok our couse in 1970-1971 and especialy J. E. Taylor for constructive comments abou the lectres and lectus ots. L, Gros, Kut, and especialy D. Rulle for reading parts ofthe manv script and fr making numerous soggestions and conectins,F, Armstrong, E.Epstsn, B. Farol, and H. Wertz for excellent ping 1M, Goldberger, E. Nelson, M. Simon, E. Stn and A. Wightman fr ak apd encouragement Mice Reso Baan Sivon ‘April 1980 Introduction Mathematics bas its rots in numerology, geometry, and physics. Since the time of Newton, the search for mathematical mols for physical phenomena fas been a source of mathematical problems. In fact, woe branches of mathematics have growa out of allempts to analyze particular physical ‘Satins, An example isthe developmen of harmonic analysis from Fouriers ‘work onthe heat equation “Although mathematics and physics have grown apart in this cen Physics has continoed to stimulate mathematical reseatch, Partly because Of this the influence of physics on mathematics swell uaderstood, However, the contributions of mathematics to physis are nota well understood, Ii common ality to suppose that mathematics ie impestant for physics on ‘because itis «useful tool for making computations. Actually, mathematics Plays @ more subtle role which inthe long run i more important. When © cessful mathematical models created fora physic phenomenon, tht i, 8 model which can be used for accurate computations and predictions the ‘mathematical structure of the mode self provides new way of thinking about the phenomenon. Pu slightly diferent, when a made! is successful ‘tis natural f think ofthe physical quantities i terms of the mathematica object which represent them and toaterpre similar or secondary phenomena in terms ofthe same mode. Because ofthis, an investigation of te lateral mathematical structure ofthe mode! an aler and enlarge our understanding ‘ofthe physical phenomenon. OF course, the oustanding example of th it Newtonian mechanics which provided roch a clear and coherent picture of celestial motions that it was used to interpret practically al physical Phenomena. The medel itself became central to an understanding othe physical workd andi was dificult to give it upin the late nineteenth eatur, ‘ven in the face of contradictory evidence. A tore movern example ofthis fnfluence of mathematics on physics i the ure of gzoup theory to classy cementary particles,vi mTRooUETION ‘The analysis of mathematical models for physical phenomena ie part of the subject mater of mathematical pis. BY analysis meant both the rigorous derivation of explicit formulas and investigations of the iteral ‘mathematical srcture ofthe model. Ia both eases the mathematical prob lems which ase lend to more general mathematical questions not asocated ‘wth any particular medel. Although these general questions are sometimes problems in pure mathematics, they are usualy classified 2s mathematical physics since hey arise vom problems in phic Mathematical physics has tredonally been concerned with the mathe- rates of casial physics: mechanics, uid dynamics, acoustics, potential ‘heots, adopts, The main mathematical tol for the sty of these Ivanches of physics isthe theory of erdinary and paral diferent equations and elated areas ike integral equations and the calelus of vaation. This {lasseal mathematical phsis has long been pat of eurrcula im mathematics tnd phyics departments, However, since 1926 the fonts of physics have been concentrated increasingly in quantum mechanis andthe subject opened sp by the quantum theory: atomic physics, nuclear physics, solid sate phys, elementary particle phsss, The central mathematical discipline for the study of thee branches of physics is functional analyst, though the theotes of woup representations and several complex variables ae also importent. Von Neumann began the analysis ofthe amework of quantim ‘mechanics in the year folowing 1926, but there were few attempts to study the structure of specie quantum systems (exceptions woul be some ofthe work of Friedrich and Relich). This situation changed in the eary 1950s Shen Kato proved the sell-adjointnes of atomic Hamitonans aad Girding 8nd Wightman formulated the axioms for quantum eld theory. These events emonsirated the usefulness of functional analysis ad pointed out the many ‘itcult matiematical questions azsing in modern physics. Since then the range and breadth of both the functional analysis techniques ved and the subjects discussed inmodern mathematical phsishave increasedenormou “The problems range Irom the coneree, for example how to compute of ‘estimate the point spectrum of « particular operator, 10 the general, for ‘example the fepresenation theory of Cvalgebras The techniques used and the general approach to the subject have Become more absrit. Although Insome ates the physics to well understood tha the problems ae exercises Im pure mathematics, there are oer areas where neither the physics nor the mathematical modes are well understood. These developments have had Several seious elects nt the est of which the dficuty of communication between mathematicians and physicist. Physics are offen dismayed atthe breadth of background and increasing mathematical sophistication which are required to understand the models. Mathematicians are often frustrated by theic own inability to understand the physic ad the inability of physics to formulate the problems ina way that mathematicans can understand. ‘A few specific remarks are appropiate. The precequisie Tor reading this volume is roughly the mathematieal sophistication acquired in a (piel Undergraduate mathematics education in the United Sates. Chapter | intended asa review of background material. We expect that the reader wil have some acquaintance wth parts ofthe material covered in Chapters I-IV and have occasionally omited proofs in thse chapers when they seem ‘inspiring and unimportant forthe reader. “The material inthis book escent fora two-semestercoure. Although ve taught most ofthe material ina special one semester coure at Princeton whch met five days a week, we do not recommend a repetition ofthat, eter for faculty or students In order thatthe material may be easly adapied for lectures, he have writen mast ofthe ehaptes 19 thatthe eater rections ‘contain the basic topic while the later sections contain more specialized and Idvaneed topics and applications. For example, one can give students the basic ideas about unbounded operators in nine oF ten lectures from Sections -# of Chapter VIL On the other hand, by doing the details of the proofs and by adding material from the notes and problems, Chapter VII could easily become a nesemeter course by el ach chapter ofthis bookends with 2 lag st of problems. Some ofthe problems il paps in the text (ies are marked with a dagger). Others develop kermate proof to the theorems in the text or introduce new material We have also included haréer problems (indicated bya sarin order to challenge the reader. We strongly encourage student todo the problems ts tite but teue that mathematics is learned by doing it, not by watching other peopl doit. We hope that these volumes will provide physicists with an access 10 ‘moder abstract techniques and that mathematicians vil benefit by learning the advanced techniques side by sie with thee applications.Contents Preface Iraraduction Contents of Other Volumes 1 PREUMNARIES 1. Sats and fctions 2. “Mein and nomed linear spaces ‘Append Lim sup and lim inf 3. The Lebesgue iepral 4 Abaract measure theory ‘5. Tho convergence arguments 5. Equconinuty ‘ates Problems pene spaces The geometry of Hilden space The Rite lemma Ortonormal bases Tensor products of Hilbert spaces Ergodic theory: an introduction Date Problems n 2 » 26 2 ey 2 6 aie sanacn spaces 1. Definition and examples 2. Duals and double dats 3. The Hatn- Banach theorem 4 Operations on Banach spaces 5. The Bair category theorem and it consequences Nove Problems ‘ve TOPOLOGICAL Spaces 1. General notions 2. Nats and convergence 3. Compaciness ‘Appenie The Stone-Weierarass theorem 46° Measure theory on compact spaces 5. Weak topologies on Banach spaces ‘Appendie Weak and strong measurailty Neer Problems Ye mca CONVEX SPACES 1. General properties 2 Frecher spaces 3. Function of rap decease andthe tempered dsributions ‘pends The N-representaton for and 4 Indctive linia: generalized factions and weak solations of paral diferental equations Flee point theorem 5. & Applcains of fed poi evens Topologies on oclly conver spaces: duality theory and the strong dual topology ‘AppendisPolars and the Mackey-Arens theorem Problems ” seuaan 03 108 m us 7 19 ne BI 133 MI ws 130 13 192 167 19 mm We souNoeD OPERATORS Topologies on bounded aperators Adonis The spectrum Postive operators and the polar decomposition Compact operators The race clas and Hilbert-Schide ideals Deter Problems THe SPecraAL THeoReM ‘The cominuous functional calculus The spectral measures Special projections Ergodic theory revised: Koopmanim Dower Problems ‘ue unsaunoco oreeaToRS 1. Domains, graphs, adjoins, and spectrum 2. Symmetric and seleadioint operators: the base criterion for self adjoinnest 3. The specral theorem 4 Stone's theorem 5. Formal manipulation is touchy business: Nelson's cumple 6 Quart forms 7. Comergence of unbounded operators 3 The Trower product formula 9. The polar decomposition for closed operators 10. Tenor producer 11, Three mathematical problems In quantum mechanics Note Problems 12 18s 188 19s 198 23 216 2 2 2 237 2a 2s 29 255 259 2 275 2a 295 27 302 anTHE FOURER TRANSFOR 1. The Fourier ranform on SARC) and (6) comvtutions 2. The ange ofthe Fourerranform: Classical spaces 4, Tera ote Fair tanform: Aavy Problems SUPPLEMENTARY MATERAL 112. Applications ofthe Riesz lemma ILE. Basic properties of L paces 13. Prof ef Tehonas theorem IA. The Rese~ Markov theorem for X = (0 WV. Mmction of functionals VS. Proofs of sme theorems i nolnear functional analysis YES. Appicaions of campect operators VIN. Menetne convergence for forms VIIES. More onthe Troter product formula aes ofthe masimam principle over Problems Ls of Symbols der a8 525 ae as 20 aa 305 as 38 a 36 2 7 a7 20 395 Contents of Other Volumes Volume Ii: Fourier Analysis, Sel-Adjoininess IX The Fourier Transform X- Self-Adjoinness and the Existence of Dynamics Volume i: Scattering Theory XL Scattering Theory Volume IV: Analysis of Operators, XII Perturbation of Point Spectra XII. Spectral Anabsis‘The eg shuld nat be cored fhe fads tate on tate pres Sorraing the preeies ree] 14 Sate and functions We assume tht the render is familar with sts and functions but itis ppropriate to standardize ovr terminology and to introduce here abbrevie- ‘ons that wil occu throughout the book UX isa set, xe X means that xis an element of; x X means that xis fot in X. The clause “forall xin Xi abbreviated (Vx © X) and“ there friss an xe such that” is abbreviated Gre-X). The symbol (e|P)) Stands forthe st of x obeying the condition (or conditions) P(x). I Ais 8 Subset of X (denoted 4 ©, the symbol X\4 represents the complement of ‘Ain X, that X\A= (26 Xlx4 4). More genealy it A and B are subsets OF X, then A\B = (xe 4,24 B). When we discuss sts with «topology, A wil always denote th closure ofthe set A. Final the st of ordered pais (spine X.ye¥) i called the Cartan product of X and Y and ie denoted Xx ¥ ‘We will ure the words “function” and “mapping” interchangeably, order fo emphasize that certain functions f depend on two variables, we will sometimes write f(,.). The symbol /(,) denote the function of one Variable obtained’ by picking a Ssed value of y Tor the second variable. A2 PRELIMINARIES linear funtion wil lobe called an operator or lner transformation. Out functions wil alvays be single valued so a function from a set X to another set, denoted ty f°» Yor X.Y or er+/(s) wil ave one aad only one value in ¥ for each xe X. 1 4 ©'X, then fla] = /)|se A) a subset of Y and 8] = (x] (0) €B is subset of X sf Be ¥. [2] wl ually be called the range of and wll be denoted Ran J. is calles the domain off ‘A function f wil be ale injective (or one-one) if foreach y Raa f there fs at most xeX such that fla) = y; fis called surjective (or onto) if Ran f= ¥ If fis both injective and suretv, we wll ny iti bijective. The restction of fo a subset oft domain wil be denoted by fT. IX 5 4 we defie the characterise fnction 7s) 28 it xe4 u-lo ig ea ‘There are two st theoretic notions which are sihlly deeper than mere notation, so we will dicus them to some extent. A relation Rona Se X Io 8 (er Related) 10 Definition A relation is called an equivalence relation i i satisfies () (ixeX) eRe [reese] Gi) Os, y €X) xy implies Re oymmetic) (i) (Ye y,26X) sRyand ye implies Re [eansitie) “The set of elements in X that ae elated to given xe X is called the ‘quialence clas of x, denoted ural a(x) sear to prove: ‘Theorem 1 Let be an equivalence relation on set X. Then each 6X belongs to unique equivalence cla “Thus, under an equivalence relation, ast divides up in a natural way into isola subse Example 1 (the integers med 3)__Let be the integers and write xRy iy isa multiple of 3. This equivalence relation divides the integes ino three equivalence eases (o1=1 MC =3 21, Ble a4 12, Example 2 (the real projective line) Let denote the real line and tet. be the nonzero vector in (= R x R). We write xRy i there ie come eR with x= ay. The equivalence classes are ines rough he origin with (0,0) removed). Next, we dicus Zoe's lemma Definition A selation on set X which i reer, transitive, and ant symmetric (that i, xAy and yRe implies x= y) ip ealled 2 partial ordering. IPRs paral ordering, we often write x- ven By Mey) = IF Ie Example 2 Let M be the unit cic in RF, that isthe set of al pars of seal numbers (a, 6) with 2 + = 1, and let, 464. BD. 0) = a= aS OF “Another possible metic is dsp, p'} = are length between the points, (eee Figure 1). i—4 Powe kt Themect an Example 3 Let M = C{0, 1], the continuous eealvalved functions on [0,1 mith ether of the mete (ha) pas fe) ot aitod=ff 70a [Now that we have a notion of distance, we can say what we mesa by convergence. Definition A sequence of elements (x57, of a metric space is called 2 Cauchy sequence if (Ye > OK) n,m 2 N implies dy. X=) . “Although efery convergent sequence i a Cauchy sequence the following ‘example shows that the converte need not be true. Let be the rational ‘numbers wih the usual metic (hati, dx, ») = [x ~ y) and let = be any inratonal amber (Unt i, * €R\Q). Find a sequence of rationals x, with 8" in B Then x, isa Cauchy sequence of numbers in Q, bu it eannot converge in Q to some ye @ (for, if x,-+) in ©, thea x,-+) In R, 50 We vould have y= 3°), Definition A metric space in which all Cauchy sequences converge is called complete For example, Ris complete, but Q is not tsan be showa (Sections 3 and 15) that ¢C{O, 1], ,) i complete but C10, 1), >is not. The example of © and ® sugges What we need to do to an incomplete space X to make i complete. We need to enlarge X by adding “all posible limits of Cauchy Sequenees" The erginal space X should be dense inthe larger space whee Definition A set Bina metric pace M is calla deme if every me Misa lini of elements in 2. OF cours, ifthe incomplete space isnot already contained in a larger complete spice (ike @ is contained in R) i isnot clear what all posible limits” means. That this “completion” can be done it the content of & ‘theorem that we shal shorty state; but fie ome definitions Definition function f from a metric space (Xd) toa metre space (hp) sealed cominaous at x if) P25 f(x) whenever 25 Es, ‘We have already bad an example ofa soquene of elents in C0, 1 with Lrt+0 bt fH O. Thus the identity function from CCTO, 1], dz) t0 610,11, 4,) i8 nor continuous but the identity from which preserves the metic, hat is th), Hy) = ds.) is called an isometry. Iti atomatically conindous. (Xd and V9) are ssid to be metic if such an isometry ext, Isometric spaces are esetialy identical as mew spaces a theorem con- ering ony the metre structure of CX, d) wll hold in all spaces isometric We now state precisely in which seas an incomplete space ca be fattened cout to be compete ‘Theorem 1 If (M.d> isan incomplete mew space, itis posible wo find «compte metric space Aso that Mis isometric tos dense subset off. Sketch of proof Consider the Cauchy sequences (x, of elements of M. Call two sequences, (4, Um) equivalent lites. diy 3) =O. Let BY be the family of equivalence clases of Cauchy sequences under this equivalence relation. One can show that for any two Cavchy sequences lms dl, 3) fests and depends only onthe equivalence cases of x, and (2) This tit Gein metic on MF and Ms compete. Finally, map Minto by taking = Into the constant scquence in which each x, equals 3. M is dense ia A a ‘map isometric B To complete our disutsion of metic spaces, we want to inttoduce the notions of open and closed sets. The reads should Kep the example of ope0 ‘ad closed Sots on thera in n mind Definition Let (Xd) be a metre space: (2) ‘The set (x] 6X, dix,y) <7) scaled the open ball, (951), of radius ‘about the pois () A et OX is called open if (¥y¢ OK > 0) B(y:") = 0. (©) Aste We X ise a meighbornod of y€ Nif By: 7) for some reo, (@ Let BEX. A point x is called a limit point of, if (> 0) _Bex37). 0 (Ea) # B, that is, i a limit point of F if E contains points ‘ther thanx arbiter nea x (e) Aset FX is called elosed fF contains alts limit points, (0) Ge X,x€6 isealle an lnterior point of Gif Gis «neighborhood of“The reader can prove for himself the following collection of elementary Thooram 14 Let (X,) bea metric space: (2) Ase, 0, open if and ony if X10 is closed (©) q+ xif and only if foreach neighborhood N of x, there exsts an Ms th m= M implies x € (@)_ The set of interior poims ofa sti open (4), Theunion ofa set E with ts limit pots i closed set (denoted by E and called the elovare of). (©) Asetis open if and only iit is «neighborhood of each ofits points (One of the main wses of open sets isto check for convergence using ‘Theorem 14: and in particular to check for continuity via the folowing criteria, the proof of which we leave as an exercie: Theorem 5 A function /¢) from a metric space X to another space Y |s continuous if and only i fo all open sets O = ¥, JO} is open Finally, we warn the reaer that often in incomplete metric epaces, closed sets may nt sppear tobe closed a rst linet. For example, (1) x lose in (O.1) (withthe usual mew), ‘Wecomplete thi tecton wit a dizcston of two of the central concepts of functional analysis: normed linear spaces nd bounded linear transformations, Definition A normed linear space isa vector space, ¥, over R (or C) and a function, jf from V 10 R which sass: © tolzoforatlein ¥ (i) Ish =0ifand only ite —0 Gi) jae = [2} Te} for all via Vand ain B (or ©) (0) Tost ws Fol + fw forall and win V Definition A bounded nea transformation (or bounded operator) from 1 normed linear space Nips |, aly © TI, Teh ITs, ~ Tania = IT X)le SIT iy ~aely St which proves that Tayisa Cauchy sequence in Vy. Since V; i complete, Tx, yor some . Set ‘Te'= y, Wermst fist show that Ins dfitonsindepenent ofthe sequence sya xchosen Ifx,~ x and x; then the sequence jy tey tocar 50 Tx, Tx)...» for some by the above argument Tha it T= lim Tx. Moreover, we ean show 0 defined is bounded beeause sig = tim ily (ee Problem 8) ‘sTmClin§, Gee Appendix to 12) = Cis, “Thus Tis bounded, The proofs of linearity and uniqueness are let to the reader. ‘Wie can use this theorem to give avery elegant definition ofthe Riemann lnegral. Let PC\a 6} be the faily of bounded piecewise continuous Tune. tions o a), which ae continuous fom height tha lim, ,, (8) =J0) and for which img, /(2) exis a each y and i equal to. /() for all but ‘ily many y. Norm PC withthe nosre te sp foal Let, 5%, Bea partition of the interval fa, B, x9 = a,x, =. Let xls) be the characteristic function of (x.y, ) except for ys) which is the sharactrsi function of fx,~..%:h A function on (a, 6] of the form Eis siad) with 5 real is calle sep fmetion to ee why, dra its praph). ‘The set of all step functions forall possible frite partitions is « normed lest space withthe notm 2x sl [Eee = spoon Denote this space by Slab]. Its a nice exercise (Problem 10) to prove that Sta.8} is dense In PCle, 6). For any step fonction, Siu, me define 1(Ssut9) = $0109 ‘he ntuve va ofthe nepal f(s] de. near transformation from Sle.) othe real numbers, and Nee Avpandixto 42 Lim sp and in int Pda -tste-n mat if =n s[Zsado- ‘isa bounded inet wansformaton. Since the eel numbers are complete, 1 «an be uniquely extended to, the completion of $ (bythe BLT. theorem) “The extended transformation Tf), restricted to PC is called the Riemann Integral and is denoted by wn ve While this method does not appear asthe most intutve dfiton af the ‘Riemann integral i wil he seen upon reflection that the proof is realy just. the “usual” proof put ito the language of completion aad the BLL T- {heorem. It ifestrates a main point of geacral philosophy in functional analysis: In order fo define something on worl linea space, it soften Convenient to define it on dense set and extend it by the BLT. theorem. “The eader should ty his hand at constvetng the Riemann Stiles intra (Problem 11). By using the same method, we ean define the Riemann integral for cominuous fusetions taking values in any complete normed linear space, in particular, for complex-valued Tustin. dix to 1.2 Lim sup and lim int Lim sup an lim inf ate notions which may be unfair othe reer. 0 we summarize thei definition and properties Definition Let 4< # be nonfnite bounded set. Let i pi) = set of limit points of 4. Then the lilt superior f 4 defined by lim sup A = HA = ups im 4} Sitar lim int = nC) = if in pea) Remarks 1. When A is Bounded, lim pl(A) is always nonempty by the Bolzano-Weierras theorem2.16 Ais not bounded above, one defines Fi A = +00. If Ais bounded shove and lim pa) = @ one defines Tis d= ~ <2 2 Ti Ais actualy in im pt). For let b= Tis A and let e> 0 be given, Wie can find ae lim p(4) 50 [b~ a] <2. Since ae im pa), we can find de A with |a~ A} <2; 50 given &, we find de A with |B~ Aj ee, that, elim pa). Ti 4 has avery simple altenative characterization, whose proof we lave to th reader, Proposition Let b= [i 4. Then for ¢> 0, An (ola>b +4) ie finite and An fala> 6— 8) is init For a sequence (a), We say be lim pa) ifr all and al thee isan n> W with [b~ a <2. We deine (a) = sup im pa) Finally let us summarize the properties of Tl for bounded ets itis & sf exercise fo desde Which extend Io unbounded se), Proposition (3) lime, +6) fia, + Tm 5, () Ti ab, < (it ain) ia, 20 (@) Tinea) = elit ay (@) Fiteo) = ctim ey La The Lebesgue integral Wie have just seen that Cla, 8] has two quite reasonable metres omit. fa Section L3 we will ee that It sa complete metic space inthe metric 49) = 39 78) ~ 069 In the other matric. we considered, (a) = Uf wit tal = {EM a, Cla B} mr compete Tose th for C1, 1 I be ee at in Figure £3, snot hard ose that fs Caehy inl, buk i doesnot ».(f) is however the best to keep in mind whea thinking intuitively. The hive wanted the poem to oe of defig fn extended sot fe We ms i dos what te fo aves, Wy not SS? Theresa cal eumple (abo Prob ohh shows t ‘Men can ves sate manta be vain ona oti fndantos (ad otto Bei such aggro al i postback up» wit al int santero wd pz, move the pss round by vetaton ad trandaton and resemble peo 5 wo bl of ads one (Ganach To para) Th, cana veut an so some amy of wl hes meshes What broertes do we vant To have? We woud ke bo ff} a Fla) tbe mesial = 030 we wel ie to ae ere ‘Ae inl Ride a Abo, when fr cominoay, we wan fer 6 bein sso shold contain the open a Fal we wat ohae Ga) = fans Af the 4y ate mutually disjoint (to meet our intuitive notion of size) 0 we would he [sey dye @ if each Asin @ Definition The Boal ses of isthe smalls family of subsets of with the following properties (The ful isclosed vader complements (i) The fay is elosed under eountable unions. (Gi) The family contain each open interval, 13 Te Levengue integra Torre that such «smallest family exists we note thai {4a collec ion of Tames obeying (, (i), and (i), then 50 does (yea See Ths the Jmersection of ll families obeying (i) is he smallest sich taal "Now we define the Lebesgue measures of sets in the Borel sets in Definition Let 4 be the family ofall countable unions of disjoint open Intervals (which sus the fai of open set) sad let 4( Grea) = £0-29 Cohan or any Be een = nt) “This nti of size has Four crucial properties: “Theorem 18 (@) A@)=0 ©) Msc and the A, are mutually disioiet 4, 0 Aq = 2 all mn), then Us dy) = Fe id (© MB) = infin) |Be Kis open) (@) MB) =SuplHO)|Ce'B, Cis compact) ‘The infite sum in () contains only postive terms, soit either converges to a finite number or diverges to infinity, in which case we sett equal to >. (6) and () say that any Bore set canbe approximated “from the outside” by open sts and from the aside by compact ses. We remind the eade hat (or the real ln a sets compact if and only its lose and bounded ‘We have thes extended the usual tio of sie of intevals and we define the family of funtion we wl consider in the obvious ay Definition A function fiscalled a Bort faction i and only iff, 8) 88 Borel set forall 3 Its often convenient to allow our functions to take the values co on eval wt in which ease we require /~"[ cal] to be Borel. Proposition fis a Bore if and only if, for all Be a, f"(B]ew (ste Problem 14),This ast proposition implies that the composition of two Borel Functions is Bore. Many books deal witha slighiy larger elas of functions than the Forel clas. They first define a set M fo be measurable i one can waite Mo A, BU A where Bis Hore and 4, <2, with B, Borel and y(8,)~ 0 (ius they add and subtract “unimportant” set rom Borel se) messue- ble unction then defied as a function, for which f(a 8) i always measurable. It is no Tonger true tat fg is measurable if and gare, and ‘many technical problems rie dn any event, e deal ony with Borel ses and Jincions nd woe the nords Borel and measurable interchangeably Borel functions are closed under many operations: Proposition (a) If f.g are Borel, then so are f+ 9, fo,manlfe9) and rin fg} Is Borel and Xe Ry ifs Bore. 1) If each fy is Borel, w= 1,2... and fdp)-ofp) forall p, then f is Bore. Since | f| = max(f, ~f}, || is measurable if fis. ‘wretched above ess one can Stine fick may be). uf fides on we wiie fe and dei fs ff, ~ [fe wbere Josieat; = mnt o.0) 260) be sto fans on 9) ‘chr oe tend te oe whe el ine bydening tn eso ote of 6) I fe-24a 0) we wre ff ff de We the ‘Theorem 1.9 Let fand g be measurable functions Then © Whge 2%), s0 a f+ 9 and if forall 16. ©) Wot sfand fo 2", then ge 2 1 [Ue parm fades fads tf and garein 2 @ iffael sepia a6) zat). I tot hard tos ta the hi oh Ua a+ el) andthe int tom he eh intel) oe the +0 ad CeO) respecte Sina) dos ot ne he Joins g ant kis eawal to dene nad) = 6-0) ae 0, Foe ‘Notion of si fori, one xa conta meer jo Drone FR thatisa map 8 = ]wiha( J) = $2 puBsaB A oo {nd 2) 0. By conarasion te mere Ra he any Pepe (8) = sup(u(C)|C= B,C compact), = inf(O)1Be 0, 0 open)‘Also, (cfr any compact tC. A measure with hese tw regularity fropere i alld Dorel menue In para, ye OD) = a8 + 0) Se), One can then coseot a integral f= [da (Wl abo wet {ich bas proper (ah) of Theorem 19 elle a Lebesgue Sie tear La 8} a) and LR) ean be formed as before. These Space equvaene cats of uncon ae complete nthe mete fa) = F791 ay and analogues ofthe monotone and dominated convergence theorem hol. The continous functions Cla, ] form a dense subspace of Litre, pt iret, 6) a) sh completion CeO] with the mee pra) =f J9 de wre we nes oly we the Riemann Sts inurl eine pce Problem 1 Lats comer hee examples whieh lusrae te vay of Lebesgue sieie eure Example 1 Suppose «is continuously dieeatiale. Then (0, 6) = [Bdrds) ts where di Lebesgue measure, 50 it sto be expected (and is Ral fren jolZee ‘Thus, thee measures can essentially be described in tems of Lebesgue Lxample 2 Suppote ta ihe ara tution ot Here b =! Poste asd BOW O¢t0) The mews one Sar yey te cn 1B) = Oe Ben 0) 200 Be se Solo one cipey eel a soc el at [7-10 This weaned i town a he ae mene Gc ao Pinca tatter tk ay vse Too oe Ba {ft wOlsn pep = and ony) = 40) Ava ea weet ee et acted Sexel yh vu) esa in ate enna crac! Nets ho eet fcc of tte be ve ots “uncon” age pa 1x eed es dont of veges) Example 3__Ourlastexample makes use of fanypatbologicsl function, {which we Sst construct. Lt Se the subset of [0 1) S=GDUEDEGDUG HAY pe that is, remove the mide hid of what isnot in Sat each stage sd adit 10S, see Figure. The Lebesgue measure of Sis} + 24) + Gf) == Let C= [0, 1S Ichas Lebesue measure. C, which is know 8 the Cantor Set is easy to describe if we write each x¢ (0,1 nite base three decimal ‘expansion. Then Cif and only if this base 3 expansion his no I's. Thus C ‘san uncountabe set of measure 0. To see this, map Cia 8 one-one way oo (0,1 by changing into I's and viewing the end fevult as abate 2 number. Now eonsruct (2) a8 follows: set a(x) = on (I.4);a02)—1 on (he): a(s) 4 on Gf), ete see Figure 16, Extend 3 to [01] by making con: Pomel Th Ctr sion Yinwows. Then 2k a nonconstant continuous fuaction with the strange Property that (Gs) exists ae. (wth respect to Lebesgue measur) snd 20 8c. Now, we can form the measure pe. Since ais coatinuous, jut) =O for any stp) with ony one pont. Neverthsles, concentrated nce set Cin the sense that (0, 1NC) = nS) = 0. On ie ater band, the Lebesgue measure of is zo. Thus, and Lebesgue measure live” on completly ierent Sts In sense we now make precise, these tet example afe model ofthe ‘most general Lebespue Stiles measures. Suppose jis Bors measure caty PRELIMINARIES let P= (x\ als #0), that i, Pi the set of pure points of Since js Bore [C) < efor any compact st, Pisa countable set, Define y= FM = MP) “Then fo i mensre 3 an =~ i ste Hage ha he property pent 0 forall p hat it Bat no pre points and iy, has only pure poi athe see that p(X) = Dae hl Definition A Borel measure won Ris called continuous it es n pure points jis called & pe plat measure if p(X) = Dens a) for aay Borel re “Thus, we have seen: Theorem 113 Any Bore! measure can be decomposed uniquelynto a Am fp + Hn WHEE Ho i CRLINUOUS A Hig sa PE ont mea, ‘We have thus generalized Example 2by allowing sums of irae measures, 1s there any generalization of Examples | and 3? Defrition We say hat bay cominaoes wh rect (98) ‘Lebesgue measure if there isa function, f, locally L? (that i, f2 | f(3)| dx < 20 foray finer, 9 that Josn= fara foray Bor! function 9 in LCR de). We he writ d= ‘This definition generalizes Example 1; we wil eventually make a diferent (ut equvalent) definition of absolute continuity. Definition We say sis singular relative to Lebesgue measure if and only IRS) = 0 for some aetS where RIS has Lebesque measure 0. “The fundamental results: ‘Theorem 114 (Lebesgue decomposition theorem) Let 4 be a Borel Themre, Then 1 fn fy it nique way Wit 4, absolutly continous vert Lebesgue measre and ith singular relative to Lebesgue measure, ‘Thus Theorems 1.13 and. 14 tell us that any measure jon Rb canonical Ascomposition = Hg + He + Hg WHE Hy i Pur Point, Hy absolutly continuous with respect 10 Lebeogue meastte, ad pg is continuous and Singular reatve to Lebesgue measure, This decomposition will eur in a , the following ‘cil theorems holt ‘Theorem 1:18 {rmonotone convergence theorem) If fe 2M, did, Os fie) s fy) sand fH) = linen (2), then fe." if and only if lion fal = 2 and im tbat ease limes fy =O 20d Un el, = Wi Theorem 116 (dominated convergence theotem) If fe L(M, di), lim 40) = f(s) [land if theresa Gel with KC3)] = Wala) form then fel 29d iene Juhi =O. Theorem 147 (Fitou's lemma) IF fre", eich fils) 20 and sf lim th co, then fe) = Lim fa) sin 2 andy if ‘Note tn Fate's lemma nothing is suid about lim. If fll ‘Theorem 118 (Riese-Fisher theorem) LM, dis complete. (One also has the idea of mutually singular: Definition Let u,v be two measures on a space M with e-eld ®. We sy that peand v are mutually singular if hee isa Se A with (A) = 0, wM\4)=0 is usful to tke a leaker looking definition of absolate continuity Which Isessemialy the opposite of singular, Definition We say + is sbioutely continuous wrt. if and only it ed) Opis (4) = “That this definition nthe same the previous one is a consequence of “Theorem 19 (Redon-Niodyn hem) _» Healy coninues TE eaten ony ther mene non fv that are [seeds ane) forany meso t4fiwiaydterned a. (3% 9) Finally the Lebesgue decomposition theorem hasan abstract form: ‘Theorem 1.20 (Lebesgue decomposition theorem) Let gv be two measures on a measure space (M, a). Then can be witen uniquely as 1 Bet Ying WHS HBB yy P6 mutually Singular and, absolutely ‘There is one inal subject in measure theory which we must consider and ‘hat involes changing the oder of integration n a lil integral, We fst rust consider what functions can be multiply integrated Definition Let (M. a, (N,) be two sets with associated ‘Then the enfield, 8 ¥ of subsets of M > W is defined to be the smalls, ‘fed containing (Rx FIR ®, Fe #) Notice tha if f2 Mx N+ is measurable (w.rt. #@.), then for any ime M, the funciion n+, m is measurable (wet #) It» isa measure fon N such tha {fm ) da) exits forall m, then one can show that ‘ms[ som m) dh) smeasurale (wt 3), There ie dietatalogue of the ‘et that absolute convergent sums ea be rearranged at wil ‘Theorem 121 (Fubinis theorem) Let f-be 4 mesurable function on ADDON. Let ube a measure on M, va measure on N. Then {,(J, vm) nd W implies Us) SGD] For functions on [0, 1), every equicontinuous family is uniformly egu- continous (Problem 3 ‘We can combine the convergece theorems (Theocems 126 and 1.27, the iagonaliation trick of Sesion 1 and the above remark (Problem 31) t9 prove the beautiful: ‘Theorem 1.28 (Ascols theorem) Let/,beafamily of uniformly bounded equicontinuous functions on [0,1]. Then Some subsequence Ju converges uniformly on (0,1) Proof Let gu.4yy... be a numbering of the rationals. Since the fs are uniformly bounded, |gn)| = Cf all mand m. Ths, by the diagonalization ‘tik, wean find a subsequence with /s(g) converging asco foreach m. By Theorem 126, the Ju comverge pointwise everywhere and then by “Theotem 1.2, the are uniformly convergent At this point, we discuss no applications in detail. However, we mention twoetamples to which we sal retira which show the varey of appistions In Seaton V1, we define a metric on all functions, 6. analyte ina repion D. In Theorem 25 we ute equiconinity arguments Yo prove that certain subset of Op ate compact. Ia Chapter XX, we will discuss the limit of the fee energy per unt volume ofa “lative gas” in a box as the volume oes to infinity, Our proof that thi imi exists for aare class of ineratons wil proced in hte stops: I) The interactions wil be given @ mesic andi willbe shown that strity rite range interaction are dase inal” allowable” interactions in this etl. (2) I for fed volime A the ree eneey per unit ‘volume Fs treated asa fanetion onthe metic space of allonabeintractions, the (Fa) ae equicontinuous. (3) The lity.» Fy() wil be shown to exist i ‘is ante range interaction. Then equcontnuty arguments will be used {o ell us that im... F(0) exits for any allowable interaction ® (and the limit wil be continuous in Sen £1 Fora cet of he tie 2c ee the aiom of ie, «intent ote vn we recommen. Haima, Nowe et They, Vn Nos ‘tng Renol,Pnowon Now Joey 158 Secon 12 For ado cut of metrics notons, se A. Gas, fa sano Abaca Aason Wel, Resting, Manche 150° es onan’: Eke of tT of Fuel teat V1, Grek res {557 For nro near pcs and npr ror nin of be Rea ine (eae Dice, Fontes of Meyda, Academe re, New Yok od Erin, Seer Lovmisand§ Sunt, Adonced Cae Aon Wey, Rede, Sestins 13,14 For cio f Labs Integra, tJ Winsome uso th abou nga Hal, New Yor, 1 of W Ro, Peis of Modo ‘ty Mem NewYork, 95 Festa eae teoy we prea re San 5.1. Berean, Mase ond magaon, Malan, New York, 1963 td 1H Roylen, Rel day, Masala, New Yr, "oy, Van Neston, Rio Pinson, New se, nor Opersrs, Caper 3 Wey Citerocs), New Yer. 58. "Fors Gnomon oth Banach Tar prone RRs, Elements a Mate sata 80, Dover, New Yok 1980, oR Robin, Pd Mat, 34 (94, “Werote tht ncn cont al ore et lly: Start withthe pen setae hi ‘omplamnt tie Sele, Ald col non le ele Fa nd It ‘Svcs nutter open et) eG tr Ten acouneteuies Focal Ga and hw Soper Fe Net a Gc fer comity many Sept an lt dn fr oe ne Gye Fa, Ca xe ees ‘ded Intheend asleep thei coat ina ee ‘tthe protic show fe Brel uso lt a lel unde alatets nd conning lcntinuou feos As wih te Bore shal corucion ‘suse, We ote hws then sy Bor mena Boe ‘ton sequal almostevoyabee wet 10a ponent of continues tons (Probl 04 By remvig he mie ce at he th te a (one can conse coed tof "Pocupprass oestrogen sac thet han asc sas) wich we dann inevion eh slab ith he Fr Seo! Soe Der atin, ‘Shaper, Herma, Pr G99, 1956 195 196) oa Gel i i tea Nate TCI an Neen Rel Pe, Sein £5 Toot ting Aa hoe i feetns ona aia com ct mee space, or mote eal song cote soma norm pce Which 8 Su any matte for expe sma teplogl eo. “heen gouty mesg the nate of he theodyransat cet tack atessse arse RG “A Pett he Fe Eryn Se Ie Ecerae"1 Mot, Phy 5 (19601215122 The pol we outed fr aie bn, Sieh ne Sc In Chaper XX; defo G. Gna and. Mile "Sasa ‘Mec Laie Sytem.” Corman Muh Py 96), 3708 ‘ae fon Deo, eqoconialy ely end ene of he proc a he ‘usnann mapping tecrem see Aion, Comper tan pet, Mame Fifew Yor 53, srs of glenn ensione cae” soa aie 1. Bd aunresampl tothe semen: Esty mei taste relation ee fete. What wrong nih te oc) dpe imples (yang)? 12 Very mat he proposed mei of Eames 1-3 Sein 1.2 et wei 1. Letina Cauchy gunna metic pce (29) Supe hat or some se 4 Lett bea aun in mec tee ad lB he. Supa ta ery ib seiner st rabaessdbsobeogrs comeing fe roe a ee ile the des of he proof of Theor 13. 16 ove Therm Lan 5 17. Prove Theown 16 8 Poesy emia me pce then Oa inn A) At Complete poo of Tear 1. 110, Bove ts ee a Fe Bah = ena 11. (0) Lets ben oti 8 I. ioe of and arith sa C0 E tae foc any sequence 0 4 "S41, Poe ta any monaco ton (0) Bete 00 50.1199 d=aeol se 1 & sn) = §, nots abn rove fit BIT analy if Sade vain (6) Lars of tated savaton on 1 Comal 8 Rian Sil itera Sra 12, Powe the vv proprio nin enn the apen to Seton 12. 1 Const ¥ he Vist aon: Cal wo nant sy 01 eat {ete Labgur manure Prove int Lebogue esl (Hi: Prove 1) saint ue of amine of 1) 116-1) Let dea Bore fncon rove fer ny BE. ) Let fand 9 be Be fonction, Prov that fv Bo 15.) Lelie fort umber Prove at respite fic mga of eins ef) i 0 hen Pier Aeron ¢ tes im Cone ta theta of ny gue of Bore forts Boe (e) Vag ta and (0, prove any pose mt of ue f Borel anions (6) Wang), pov the dominated conertnce thor fey the mentor cn “16, Pov tate Done ae etn 2 Hae he rat ai wi ie ‘lode Hanah the property. see of unfory bende uncon: Sn'F and poate en fe U7, Poet erry to Theor 112. 118 (2) ove at foray open stn 1s na in of niu Fanti (B tec ben teeta th roe yoann ners nash 4 oem (Gee the put of Lateaue ee. (o Prove Clint! deme mb)19.0 Lat folnin wont Prone, 44 Fst at anion 2, Contact a aun bound futon on [01] someing 020 E36 cm ie ete i) "ha omer 0 pln 0. fuetion fica» up fenton i mente an continuous ep a 37 Amel sac Xs cal oly ode i far every X san be coee by ay nate ier feats anf spcewsecostn on thecorglemet tay els. Rove that poste conerpent uray eucoinae ere ‘hese jampn call unguar cats t-and Sos ale ey Tunis on sly bund meee nore comes ‘Stns ene ec od Baha foray (a) Using the Heine Bowel propery, pote atx comlnoes fron on 11 Ebewa-nal set conan (Proyecto suman fees 2 la wif rom which Is 9] Uh I fllows. Another useful geometric equality ithe paralelogram law Problem 4): Deb oP + De P= abe + 2Iyt In Section 12 we defined normed linear spaces and observed that every normed linear space is a metric spac. The following theorem shows that ‘ery inne produc space i «normed liner space. ‘Theorem 12 Every inzer product space V is normed linear space with the most bx = G2)" Proof Since V is a vector space, we need only verify that Ul has all the properties ofa norm. Al ofthese proper, excep the trengle inequality, {oli immediately Trom the properties ()-() of inner products. Seppore xiye Yo Thea Deby =G.99+ 9) +0940.) Ge) +2 Re +O.) $66.94 21G 9140.3) 565.8) +259! +009) by the Schwarz inequality. Thus bet yi s(bel + WD which proves the triangle inequality. ‘This theorem shows that we havea natural metre 45,3) = ETD in ¥. We thus have the notions of convergence, completeness, and densiy eine for meric spaces in Section 2 In parila, we can alvays complete ¥ to a normed linear space P in which ¥ is isometriclly embedded as 8 dense subset. In fc, P is alo an inner product space since the inner product an be extended from ¥t0 P by continuity (Problem I) Definition A complete inner product space is called a Hert space. Inner product spaces are sometimes calle pre Hilbert spaces. Definition Two Filbert spies 2, and 2, are sui to be iamerpie it there is linear operator U from 2; ono 33 such that (Ux, UD, = (5 De, forall, ye 9; Such an operator i called itary We elaborate thes ideas and show the reader what types of Hilbert spaces tis ly to meet by series of examples. Example 2 eve!) Defies 0] 19 bet wtf compe aud tent tctonson (Bs ements eae Wie ee a ner pod Gar f Feo) a tm tah nr mae nese (7G) s 41/001 + Hotel? so that 7G}a() is in (0, 6]. A proof similar to the RiexFisher theorem (Theorem 1.12) shows that Lf, ] is compete and is therefore a Hilbert Space. Itt not to dificult to show (Problem 2) that La, 8} 6 the com Pletion of Ca, 6 inthe norm (f yeoran)"* wiExample 3 (¢3) Define 10 be the st of sequences), of complex ‘numbers which satisfy J. [1y/*< 0 with the inner product In Section 1.3 we will se that any Hilbert space that has countable dense set and i nt finite dimensional is isomorphic to. In this sense, €, ithe ‘canonical example of Hilbert space Example 4 (LR di)) Lely bea Borel measure on FR, dy isthe set of complex-valued measurabe functions on which ay er i/G)* << ce, LR, du) sa Hilbert space under the inner product 0) |, Fo Example § (ret sum) Suppose that 2, and 2, are Hilbert spaces. “Thea the et of pars x, with x€ 2,» 2 isa Hilbert space with nner product (si 1D. G2 IDV Cie, +01 Der “his space is called the rect sam ofthe spaces 2, and 2, and is denoted by 26,2. If and jr are mutually singular Borel measures on Rand i= 4, + a, then ER, dh) isomorphic im a natural way to LR, di) @ 2G. du) (Problem 3). We can also construct countable direst sume as follows. Suppose Ff, is sequence of Hilbert spaces. Let 3 denote the ef sequenes (i), with xy € 2%, which satay Eiatnes 2 is a Hilbert space under the natural nner product and is denoted by wbx, Example 6 (vector-valued functions) Suppose (¥, ita measure space and 2 isa Hilbert space. Let DX dt 2°) be these of measurable ane: tions on X with values. 3 which sais J, Uebe- ails) <0 Thi ti ibe sce ith einer rat 60)~ [U0 ares) OF course, we have not sid what it means fr a vector-valued function to be measurable. For this definition and related mates see Problem 12 and the ppendis to Section V5. 2 “The Riese lemma t Tn the examples in Section HL we showed several ways of constructing new Hilbert spaces from old ones. Another way to do this i to restit attention toa closed subspace -# ofthe given Hilbert space 2. Under the ‘natural inner product tat it inherits asa subspace of 9, .€ is a Hilbert, suce. We denote by .A* the set of vectors in 3° which are orthogonal AF is caled the orthogonal complement of = It follows from the linearity ‘ofthe inner product that isa linear subspace of 2° and an elementary argument (Problem 6) shows that = is cloned, Thos "is also Hilbert Space. and 4" have only the zero clement in common. The following theorem shows that there are vectors perpendicular to any closed. proper subspace, indeed there are enough of them 0 that RM Malt ylre med) ‘This important geometric property is one of the mun reasons tat Hilbert, spaces are ease to handle than Banach spaces (Chapter I}. In the follow lemma and theorem, the reader should keep the fiteimensiona casein mind (2e Figure 1). of “he potion of on 1 Asapleen this setion beacon 34,2 smnent spaces Lemma Let 9 bea Hilbert space, closed subspace of 2, and suppote x€ 3. Then there ens in a unique element z close o Proof Let d= inky gt ~ yl Choose a sequence (Jy € Me 8 that k= yn Yel? = I= 2) = 00-9 221, — 28 + 2p, aI = Het ek 2m aI + 2p, 2B 4b 1G + DIE $21y,—xP +2), —P 40? ea 4 ad 4d? 0 The second equality follows from the parallelogram law the inequality follows fom the fact that (J, + 3) Thus (j) ie Cavey and nce

given by Keune tw We wil often suppress the isomorphism and simply write 3° = 4 @ ‘We have already defined in Seeuon 2 what we mean by a bounded linear transformation from one Hilbert space 2 to another 2”. We will denote by 20, 27 the set of such transformations. 22F, 267) is clearly a vector Space andi becomes a Banach space under the norm i= sp se “The proof ofthe fact, though not diflcult, is postponed until Chapter II hee i proven in greater generality. For the time being we are ineested In the special case where 2°" = C Definition The space 22", C) is called the dual space of 3 and is ‘denoted by 2°, The cements of 2* are called continous linear fanctonls ‘The following important theorem which characterizes 2¢* is due to FF. Ries and M, Freche. Theorem A (te Rise na) For each Pea te wie Jee atta Wes cara ne Inston r= es Pn, tet be te of 0 sh tT) 0. By he emi ot FF tal oop ie en Teosb =O) fr tod Teitelnl este oath Ten te pao ese eee nnare ct of Date yom TED a, We sey hans bs i ri Pen T= Jeg Msthumes fran tee a= Te) =) = TED) = O18) Sith nti (an (ye ima on apo dt) cage nt pene ad Bt en Doyen ye cm eres = (22.4) «22 y(n) Ta ‘Thus T(x) = (ir, x) for all x € 9. IF Thx) = (", x) abso, then ly" — yx]? = Ty aOR yo ty eer fog vue Yoo ta ie. Drtew Sone a 174 pT] = lo up, el = Da and t7t= gg on») = re) =o‘We note that the Schwarz inequality shows that the convene ofthe Riesz emia is tue. Namely, each ye 3€ defines a continuous linear fanctionl T, on 3 by 7/0)= (2). The Ries lemma has the following corollary which is very important in applications. Corollary Let B(,-) bea fonction from Hx HF to € which satis: (Bes, ay + 3) = 20x, 9) + BBle,2) i) Blak + By 2)= BC, 2) + BBL) 2) Gi) 1865.95 Coat I for all x y,2€ 2, ¢C. The theresa unique bounded liner transfor: mation 4, fom 2 to 2 so that Boxy) =(As,y) forall xyexe “The norm of isthe smallest constant C such that i) holds Proof Fixx then () and (i) show that B(x.) ie 8 continuous linear fetional on 2: Thus by the Ries lemma there sn x ¢ so that Bts.9)= 0.9) forall year Define Ax = x isnot dificult show tht 4 isa continuous linear operator with the ight properties (Problem 8). ‘A tilinear function on 2 obeying (end (i) is alld a sexquitinar form. ‘We have already defined what it means fora set of vectors to be ortho= normal In this section we develop this ides further in particule we waat (0 extend the iea ofa "bas," so wel for fite-dimensional vector spaces, to complete inner product spaces. I isan orthonormal et in 8 Hilbert space 2 and no other ortonormal st contains S as 4 proper subset, thea 8 is talled an orthonormal bass (o complete orthonormal system) for 2, Theorem ILS Every Hilbert space 2 has an orthonormal basis ‘Proof Consider the collection € of orthonormal sts in ¥. We order by Inclusions that i, we say S, <5, if S, © Sy. With this definition of <, @ is partly ordered; itis also nonempty since if vis any element of ¥, the set consisting ony of ois an orthonormal et, Now let (Sana any Tncry ordered suet of . Then Jeu, 8 a onthonormal st which contains each S, and shu an uper bound for (Su Since vey ine ly ‘rere beet of hasan uper bund, weean apply Zav'slemma Theorem 13)andconcade tat bass masa ment hat, an orthonormal Sytem not property cotained in anyother erhonormal sem. H “The following theorem shows that asin the fnite- dimensional ease every clemeat of« Hilbert space can be exprested a neat combination (posibiy Infinite) of basis elements ‘Theorem 1.6 Let bea Hilbert spaceand $= (x).e4an orthonormal ‘basi. Then foreach ye, E.G ay and wir= Ela or au) “The equality in (1) means that te sum onthe righthand side converges (independent of orden to yin 2. Convery, Sag el? < e2.e,€Cs then Sees, converge to an clement of 2. Proof We have already shown in Section 1L1 (Besse inequality) that for any Bite subset A° = A, Dae [Co 7)/? 5 IDI. Thus (4,9) #0 for at ‘most countable number af in a which we order in some Way 8,8, hans» Furthermore, since PJoy [(x,.))F* i monotone increasing and pounded it converges to a finite limit 86 N= 29. Let Y= Does yn) ‘Then for n> m, Denna Feo Emo ‘Therefore (y,) is & Cauchy sequence and converges to an element 3° of 2. Obierve that on vingetin (r= Fsanrmts) $015) -Ox,)=0 ‘And if a for some ¢ we bave tin p= fru rm oaTherefore y~ y’is orthogonal wo al the x in S. Since Sis a complete orho- normal system we must nave yy" = 0. Thus ra few, sn yh ote ab- =n (v= ° oe) bP Fie. MP soa 2) Weont ney pet eee ‘We note that (1.2) sealed Parseva’s elation. The coefients (sy. ace often called the Fourier coeficents of y with respect to the base (i) Tae reason for this terminology wil become apparent shorty. We now describe «useful procedure, called Gram. Schmid orthogonaliza- tion, for constructing an orthonormal set from an arbitrary sequence of independent vector. Suppose the independent vector, uyy Ae VED and deine = mille y= malig =e ‘The family (0) sam orthonormal se and has the property that foreach, (u)Jex and (6) span the same vector spce. In particular, the et offic Tinea combinations ofall thes the same a theft linear combinations of the ws (ce Figur 112), Foe Gran-Smitotogonsitin. We semark that the Legendre polynomials (op to constant multiples) are obtained by applying the Gram-Sehmidt proces tothe functions I, 2, Pn, 0m the ater (=I, {) with the wal? inner product. Definition A meric space which has a countable dense bss sid tobe separable ‘Most Hiter spies that arse in practice are separable. The following theorem characterizes them up to isomorphism. ‘Theorem IL7 A Hilbert space 2 is separable if and only if t has 2 countable orthonormal bass. If there are N'< > element in, then 2 ie ‘isomorphic to C%. If there are countably many elements in S, then 2 is isomorphic to ¢, (Example 3, Section II). Proof Suppose 2 is separable and let (x) be a countable dense set. By throwing ovt some of the 2s we can gets subcllction of independent ‘vectors whote span (nite linear combination) isthe same as the (x) and is thus dense Applying the Gram-Schmidtprocedue to this subcolecton we obtain a countable complete orthonormal system, Conversely i () 8 complete orthonornal system for a Hilbert space 2 then it follows from ‘Theorem ILS that the et of inte inear combinations of the J, with rational coaicients is dense in 2. Since thi st is countable, 2 is separable ‘Suppose 2 is separable and (2, is a complete orthonormal system. We define a map #2 +f by Wn ‘Theorem IL6 shows that thie map is well defined and onto Its easy to show itis unitary. The proot that 2° Is iomorphic to Cif 5 has N element ie similar. B [Notice that inthe separable case, the Gram-Schmidt proces allows us to construct an orthonormal basis without using Zor’ emma. ‘We conclude this section with an example tht shows how Hilbert spaces rote naturally from probleme in classical analy. I (eis an integrable function on (0, 2s) we can define the numbers _— cnabn [femmes “The formal series D2. 6 2x)" ei called the Fourie series off The lasial problem is for which fand in what sense does the Fourie series of|‘Feonverpe to" This problem which originated with Fourier a 1811 has had {ich and eventful histor. It has given Te 1 an ene branch of modern ‘mathematis (abstract harmon anals). Furthermore, some of the mize results onthe elaeial ase ave just been proven recently (oe the Notes) AS anexarple of classical result we state Probleme 18 and 15) ‘Theorem 1L8 Suppose that /() is petiole of period 2x and is con ‘inuously dierentable, Then the functions Sy ce converge uniformly tofias Mae “This theorem gives sufficient condition for the Fourier series of a function to converge uniformly. But, finding the exact clase of functions whose Fourier seis converge uniformly or converge pointwise has proved to be & hand problem. We can, however. get a nice answer to this question if we change our notion af “convergence” and thi is jst where Hilbert spaces come in. The collection of functions (2n)"" <™)=. i early an orthor ‘normal st in 10, 2x}. If we knew that it was a complet orthonormal st, then Theorem IL6 would allow us to conclade that forall fonctions in 0, 2n). 10 on Yoon sem sensor resonate mi (97e LESSENS Tp a ae aor Theorem WL9 If fe (0, 2x), then TMy cy (2)°! & converges to fine norm as Mov, Proof We nee to know that the period, contauovsly diferentiable fne- 2] ae dense in 0,2} Ta Problem 2 the reader asked to show that thé sep functions are dense. Buta step function can be approximated (in 2) by 8 CHO, 2) function by rounding of the corners ina smooth way tnd by changing it atone end to make periodic, The reader should con- ‘ince himself hat this canbe done so that the esulting function ie abitreriy lose to the step function in ? norm, To show that (2e)""" «=. sa complete set we need only show that (eg) = 0 forall n implies 9 “0. Suppose fe C10 25), then by Theorem iis Seon as in (Seonrirer) 0 it (0%, g)=0 forall n But, then g is orthogonal to ll inthe dene set iO. 2] which implies g = 0. Thus (2x)""* <2, isa complete orthor ‘normal set and it follows from Thearem IL6 thatthe Fourier eres of every 1Z(0, 2s] finction converges inthe -norm tthe function. ey “This thgorem shows that the “natural” notion of convergence for Fourier series is convergence and illustrates oe ofthe basi principles of functional analysis: nmely to choose an abstract space anda notion of convergence hat is appropriate to he problem at hand, «space in which one can prove nce ‘theorems, By doing ths one avoids some hard problems; thi has both advantages and disadvantage, WA. Tensor produets of Hilbert spac ‘We detribed in Sections U1 and 11.2 several ways of making new Hirt sacs from old ones. In this section we describe the tensor product 2, © 2 of two Hilbert spaces 27, and 2°, ‘The construction of the tensor produce which we use Is ot the! most elegant, but is very dict. The reader can casly extend our profs to constrct the tensor product 2, @ 2", © x, of initely many Hilbert spaces. el ot and 2¢, be Hilbert spaces. For etch 9, €2%), o1¢ 25, et 1%: os denoe the conjugate bilinear form which ate on 3, © 97, by (018 Nn 12) = Wi 2b. 02) {Let be the sto nite linear combinations of such conjugate linear forms; we deine an inner product (,") on by defining @O¥.n81)= (0.4.09 and extending by linearity to 6 Proposition’ —(,/)is well ened and postive definite Proof To show that (is well defined, we must show that 2 2) doesnot ‘depend on which finite linear combinations are used to express 1 and 1. Todo this tis suena show tha if is a ite sam which the aro form, then (fn) = Oforallye€, To see tat this is true, etn = FI, €(@.0 4), then en ($ ceo.) eH 6D “0 sinc the zo form, Thus) well ene, Now, suppose A= FH dns @ ). Then (ml and (a span sub- spaces My © Fy and My © Fy respectively. If we let (0), and (1, be ‘orthonormal Bases for My and My, we can express each interme of he cach i terms of the 9s obtaining CD —(E C10; OV.T 60:8 Ve) AL epeinl Op OMe Yad nF lew? sf, 2)=0,thenallthe ey, = Oand Ais the zero form, Thus (is positive Aefinite. Definition We define 2 @ 2 to be the completion of # under the er product (,") defined above, 2, @ 2, scaled the tensor product of Proposition 2 If (o4) end (fare orthonormal bass for 2°, and 2; respectively, then (99) isan orthonormal basis for 2, @ 2 ‘Proof To simply notation, we consider the casein which both 3, and fare infinite dimensional and separable. The other cases are similar. The set {e844} cearly orthonormal and therfore we ned only show that is centane in the closed space §spaned by fo, 4) Let @@ Vee. Since {ea} and (4) are bas = Sv and b= Ydrby where S lal? <> and Ela? < o. Ths Sled < oo. Therefre by Theorem 16 here tea vsor = Bachan 0c in 5. By Sires compuction 1e0V-Tadnovd-+0 as M,N. 8 To show how the tensor product arses naturally, we will show how i is related to Hilbert spaces with which te reader is already familiar, Fist lt (My yd and CAs, g) be measure spaces. We suppose that LM, snd L(M,, dj) ate separable (cee Problems 24 and 25 ofthis chap Problem $3 of Chapter IV). Let (96) and {¥s)} be bases for XM dy) and 174 di) respectively. Then (2049) is certainly an orthonormal set in QA, Mas diy @ do). The fact that (ons) i actualy a Beis an be seen a follows. Suppose that f(x,y) € LM, x Ma. di @ du) and fora an) Pts eae hic be resiten J, [7c anen) worangr=0 Se air 0 ple a fue FR Viou62) d(x) = 0 ht ont 8,6, wit 5) 8 Tas 6 SJ fd pines cb hr alien Doin neni ke el Re Fenteea yre nt Seite) ease BM, x May dy @ dts) Now let Ui @¥, = eMeWA) ‘Then U takes an orthonormal bass for 2M; diz) ® LAM, di) ono an ‘orthonormal bass for LM, « Mf, dh © de) and extends oniguely £08 Unitary mapping of 20H, dad @ LOM) onto FM, x Mas de, dust mmperr sPaces Notice that if fe (My di), 6 CM, du) then UY @0)=ULa%OL ave) -u(Za4ney,) = Eadie) = feo) Because of this property, we often say that L2My % Mz, diy © ee) and Lidl) LAM di) are naturally” isomorphic, Let My= Rand duc= Lebesgue measure, then we have shown that 2(R2) s naturally i0- morphic © FRO L). Let us return for a moment to Example 6 of Section IL: (fn) is & measure space and 2 separable Hilbert space with bass In Problem. 12, the eade is asked to show that each LM, du; 2°) sa iit mim Simsten of ite near combinations of vectors ofthe form faa Js) € ECM, di. We now define vu furen- $100 ‘Then U isa wol-deined map from a dense set in (Mi) 2” onto a dense set in (Mats 27) which preserves norms, so U extends uniquely to & unitary operator from 17 di) ® 2" to LM, dh 7). Notice that Under this map, UG) © ) flag forall @e "In this sense, Uiscalle the natural isomorphism Between LM, dad@ 2 and LM, di 30). We summariae this dsesson ina theorem Theorem MAD Let ($j, and (M2) be measure spaces so that BO day) and BOG, di ate separable Then (2) There is @ unique isomorphism from LM, dis) @ LCM, dis) to BO, Madey Odin) 9 that }orfy (). 120" is separable Hilbert spac, then there i wiguefsomorphism from Edy, duJ@ 2" © LM, digi 2°) 50 that {0 @9™=f le. (Theres a unigue isomorphism from L(M, x My.dt, @di) to (Ah, dass 2, da) such that/(, ys takensntothefuclions+=fs, A “Taner product of Wibrtapeeee 53 Example 1 ‘The Hilbert space in the quantum-mechanicl description ‘ofa singe Schrgingee partici of spin one-hal i Z@,dx; C) thai he Se of pairs (U3), ¥s(0) of square-integrable fonctions (xis Lebesgue measure). By what we have shown above, L(@, de; C*)is naturally isomor Phie to LR @ Example 2 (Fock spaces) Let 2 be a Hilbert space and denote by 2° the mfold tensor product 2° = @.#@'@2. Set "= C and eine Fo) = Be (2) is called the Fock space over 2; it willbe separable if 3 is. For ‘example, i = LR), then an element ye (2) isa sequence of functions P= Wo WG He AD Wabi Ad ott Wat Eisen ce Actually, i not (20) tf, but two ofits subspaces which are wed ‘most requeatl in quantum eld theory. These to subspaces are constructed 5 follows: Lt 2 be the permutation group on m elements and let (gy bea bash for 2°. For each 2 ¢ 9, we deine an operator (which we also dente ‘by 6) on basis elements of 2° by 204, 84,8") = Pays BP" Phan «extends by linearity ta bounded operator (of norm one) on J" so we can efne 8, = (Und Seve, 0: Its an easy exercise Problem 23) to show that ‘82 =S, ind St = 5,505, isan orthogonal projection (he reader unfamilar ‘with adjoint and orthogonal projections should look up thei definitions and clementary properties in Chapter VI). The range of Sy i called the mol Symmes tensor product of 2. Inthe case where 20 = L'(@) and 20" BQ) @ “© LR) = LER), S,€% is just the subspace of AR) of all, functions left invariant under ny permutation of he viable We now define (20 scale the symmetric Fock space over 3 ofthe Boson Fock space over 2.Lett) be the fnstion fom 4 (, =I) whch sone on evn ermal sons admins one on of pormstton, Dene d= Cit) Sooo ae then isa otopoal projection on, aldo sl em ‘ome tensor product of Inthe ce bere a © E(@). A" the subasof 2) coming o those function old under echange oF te oon Te bape Fone Bae ‘scaled the antisymmetric Fock space over © or the Fermion Fock space over #. In this section we give brief introduction to ergodic theory. For our 0) Show (FoF) ne Sn and pve ‘Beet F hs eo 1. at be onpramee woup of mean preening adoration a4 mesure Seace

apy he Ce mean epi tere oft) poe he ten i he for iy EYE Moho a eo iting 1 A a AiN)SEL YY = raf eo whee Piatt mcrae fen oisivr=7 a 2. Conier the un ele, Cle} = 1 with Leberg metre, Let TU) =e Stow that Tego and ony 18 aon 2. Suppose i comput mere sace th mete pan se mere Lat Tbe restr peering ese aloe wth he sna propery a fh) pe Fer exarpehe rp Tin Prien 2 when atl) Show ff anit etn en EEF cere non 48 0%) worm euicntiouou and hen ete mean edie and Aol heres) 12, Letigd a beast f vet ina Hie sac, 10 tat a= le 9 he ne trl a) oan operat om Za) Poe a Eker suas focany fe.0 (a) Prone that 5 depen ets (8) Prove at = Sand Se See ow hat ot 9, {9} Bo pana) p80) et (Min) be ote mance space Lat y= We RAD < 2), a Tyee egal and ony ir pct a M0 whew LAY MUNNICMOD, ie ete my fears ces of nae th ition (a) Prove that oly depos o the eguinlence aes of and Yin dy (©) Prove tat yi te usin, Y= 2) a mei ace {G Povettr (404 wears and on wth beep a eprable fe este pce (ht Mu) with A) <2) wth 4M oe ‘pul (ie: Tae ap uncut Crean pot ft a efor 0.11) ron ht 8 (of Thorn 10 low rm prs (2) a). Pro he pote theorem ing the exter f topo se, Banach Spaces educa airmen of math fos maps. 1 afer ambi an ty chert gm hs ple ayer the safer of apo a xm pe, 8 ‘hme fs he gone GH. Harty LA Definition and examptes + We defined normed linear spaces in Section 12. Since normed linear spaces are metic spaces, they may have the propery of Being complete Definition A complete normed linear space i called a Banach space. ‘Banach spaces have many of the properties of R they are vector spaces, they have a notion of distance provided by the norm, and every Cauchy Sequence ha limit In general the norm does not arise fom an inne product. (Gee Problem 4 of Chapter I), 20 Banach spaces aze not necessarily Hilbert, ces and wil ot have al ofthe sme nice geomatscal propenies. In order {o acquaint the reader with the types of Banach spaces he is Iiely to em ‘eounter, we discus several examples in deta, Example t (L°(R) and ts subspaces) Let L°(R) be these of equivalence ‘clases of) complex-valued meaturable fonction on such that | 0) 0. Then 120.) 0103) thn +) 04 Bac 9.4 55) sis Lon-mieztgont 9) Pol, 28) + opty, + 62) Thus, forall #>0 and y.99 € Y, Henne dyois s+ —0a‘We can therefore find real number a sueh that nya un) eset Hye) -a09) ‘We now define Ie) = a It may be easly veried thatthe resulting eeasion sarises J) p() for all'v ¥. This shows that 2 can be extended one dimension at time ‘Wernow proceed with the Zoras lemma argument. et & be the colton ofextensions «off which satisfy es) = pls) on the subspace where they are ened. We partially order € by sting ey , Thos A= 0 which isa contradiction, ‘Therefore @ ie dense and X ie separabe. H ‘The example of and /= shows that the converse ofthis theorem does not hold Icidenly, Theorem TILT provides a proof that ¢, isnot the dual of Cen sine és separable and fis no WA. Operations on Banach spaces ‘We have already seen several ways in which new Banach spaces ean aie from old ones. The socstive dials of «Banach space are Banach spaces and the Bounded operators from one Banach space to another form a Banach space. Also any clsed linear subspace of a Banach space i Banach pace ‘Thee are two other ways of constructing new Banach spaces which we will reed: direct sums and quotient spaces. ‘Let A be an index set (not necesaily countable, an suppose that for cach we 4, Xia Banach space Let, Xt “Then X with the norm Altes Wesdt = bee, isa Banach space. I is called the direct sum of the spaces X, and is often Wetter X= Dye Kee We remark tht the Hilbert space ditet sum and the Banach spas direc sum are not necessarily the same. For example if We take a countable numer of copes of. the Banach space det sm i 2, hil the Hilbert space dinet sum i. However, i one has fine number Of Hilbert spaces, thei Hilbert space dire sum and ther Banach space ‘get sum are somorphi in the sense of Section IIL ‘Let M bea closed linea subspace of «Banach space X.1f X were a Hilbert space, we could write Y= Mf @ M>, The Banach space that we now define ‘can sometimes tak the place of M inthe Banach space ease where there Js no orthogonality. IF and y are elements of X, we will write x~y if x —y EM. The relation ~ isan equivalence relation; we denote the st of ‘quivalence eases by 7M. As usual we denote the equivalence as contan- Ing by [x]. We define addition and scalar multiplication of equivalence lass by ats] + Bb] = ox + 6) “The are category thaoram andi connaguancet 18 hich makes sense since the equivalence class onthe ght only depends on the equivalence classes from which x and y are chose, not on the elements themselves, With these operations X7M, becomes complex vector space (dhe cats M isthe zero element. Now define ole = in x mie tis not hard to show that fly 8 norm on XM. Is} =O implies [x] = 0 because M is closed. We wll show that X/M wih this norm is complete by ‘sing Theorem 113, Le (lx), bean absolutely sommable sequence in XYM. Thats Estn-mee For each m, choose m6 M9 that by -m ‘Then (x, — mis absolutly summable in X. Since Xie comple, (x5 ~ ms suramable, Ut yo tin $6,— m9 Then 40 Naw Stra-vif =| fx-»- Eo ‘This proves that fe) is sumumable, Using Theorem 11.3 again we concude that X/M is complete. X/ is called the quotient pace of by The reer should workout the easy detail of the following example Example Let =, 1} and le M = (/1/0) = 0). Then X/M = ©, NLS The Baie category theorem and Many questions in Banach space theory involve proving tat sets have nonempty interiors. For exampleProposition Let ¥and ¥ be normed linear space. Then nar map T= Vis bounded if and only if THis bas 8 nonempty interio. Proof Suppose tat T's given an the set in question contains the ball Gls BF for rome where BE mee X| xl <7) However, sino (BF] ~ FT[BE}, we need only show that T(2¥1 ea nighbor hood of zero for some r. Finally, by the “translation argument” of the proposition, iis sufcent to show that T(BP] has 2 nonempty interior for Since Tis onto, 0,04 s0 some TUB has» nonempty interior. Now the hard work begins, snc we vant T(8) 10 have nonempty intro. By sealing and tralting we can suppose that Bis contained in T{Bj}; we will show that TTB) = 718s) ‘which wll complete the prof. Lety eT} Picks, €8, 09 — Tay © By TBs ah Now ik 5 € By so that YT Tese oe By indotion, we choose Bt that yo E rye Bae Then x =D, xis ands in Band ye Etuets Thus yeTIBL ‘Theorem IIL11 (inverse mapping theorem) A continuous bition of ‘one Banach space onto another hata continuous inverse, Proof Ti open s0 T For an application ofthis sult see Problem 19. Definition Let Thea mappingota normed near space intoa normed linear space Y, The graph of T, denote by T(T), i defined as FT) =x leese xx % y= Tal ‘Theorem HILA2 (closed graph theorem) Let ¥ and Y be Banach spaces and Talinear map of Tinto ¥. Then Tis bounded if and only ithe graph of Tis closed Proof Suppose that F(T) isclosed. Then, sine Tislnea, F(T) ea subspace of the Banach space X( ¥. By assumption T(7) is eosed and thos a Banach space inthe norm Mex Tap = bet +P Consider the continuous mp Ty TT TG Tae MGT) Te 1 i bition so by the inverse mapping theorem TI;* is continuous. But Te Mige Ths, 0 Tis continuous, The conver i tv To avoid future confsion, we emphasize thatthe Tin this theorem is Implictly assumed to be defined on all of. We wil ater desl with irene: ormations defined on algebraic subspaces of X (tall of X) with closed graphs which are not continuous. To appreciate what the closed graph theorem relly does, consider the thre statements, (@) x, converges to some element x () Te, converges to some element y (@ Trey.A priv t0 prove that Tis continuous one mst show that (2) implies (8) ‘ad (¢, What the closed graph theorem says tha tb suffice to prove that (2) and () imply The following eorllry ofthe closed graph theorem has importa quences in matheraial physic, Corollary (the Helinger-Toeplit theorem) Let A be an evenhere defined liearoperaior on a Hilbert space with (x, A)) = Aa.9) forall ‘rand yim, Then i bounded Proof We wll prove that F(A) is closed. Suppose that (xy, AB) + (4.99 We need only prove that (x) ea), tht rs. that y= x. Bat, for aay rer, (ey) lim te, An) nde) = 6.49) Ax and TU) is closed. lim (42.39) [As we shall te, this theorem isthe cause of much technical pin beease in quantum mechanics thee ae operators (like the energy) Which are une bounded bu which we want to obey (A= Any) in some sense. The Helinger-Toeplitz theorem tls us that such operators «cannot be everyhere defined. Thus such operators are defined on subapaces (D(A) of and defining what ane means by A+ B or AB may be dieu For example, 4+ 8 is pris: only defied on D(A) 0 D(B) which may (O) even inthe case where Both D(4) and (8) are dense. We reluen to these questions in Chapters VIIT and X noves comedies rng the 1950 cansting oly Thos Opts {Ener Monae, Wana, 932A gue rena recs fx heater intns chapter sth bok Pardon Meer Safa A. Padma Hl, New York Sis tiecatwypenr ype Holton che Tope lorie td se th Her nega ote pei eve wae ~ fara(for) (fue) (=i Suppore XisaBanch soe. Onto he ay of hinge Banach pace operas tom Feet #2) sta eat a gra Ths oe canes. estate! iy nse ‘iscranimperta easel (3, a, woure na spn bets se ‘Thee hoy of petra ed nets Seon 2 _ Toe prot tha LL may be font in Revs’ ook Ge the [Not for Charter Dor ray be oer ing the vain af ny conven sac Probie 23 and 26 of Chapel and Pb 1 of Chap In Scion oe {eu the ns of sce abit of 2 Section I} ‘The Mauch trem dts bck a he wrk of Waly in “Ober nae Funktion Opertnen" Stabe, dad Wi. en Mater R12 My (G32, 26-297 ad Uber Spee nearer Genera Uren Vien Urbane ten Mora Ma, Py 9 40-9 mde ven ao 68S, Banach, “ures otc Ln I Sade Mate 1 (39, 31-2 2b. Ano eal ofthe conte sppesionf the Hi nah tte ay foun nth took by Prcdnanmentaed above Thre shown be teats rach evento poe the esters Ge utian far he Dsl ble sh” da ar Se 391109). 12) Tae peices Reto prope ‘ae ans eau main Pnd Ms (03) 90- 8a Bach Theenes turlesmmbes de res egae” nt Mah 16 9598 Bat See ‘usteoremwopeoentys anh and Stas Sue pre econceaoe ‘sings Pond Moth 9,92, 3-1 Ter dco ee a thene eo ‘estunen Lara ook eth Neto Scon 3) Tere oem ne {SwgA const etre ee i aed ean Al her at ee it ny i ofl cde Aa tha it Felina comple metic ace deme atc 2.1 ic spaces, or sont somthing tre anes eeyabre"ifit te ona realty sxe i he fol oa free er. Thre ae ‘ne smn er ha etn fn. Cong Lace aa, ‘at pp 13-96 Seam, Ne York. 6. Waring Terence 2 [0,1] tet eatery ith mente I Tha he wo aio he, Lege de a ‘iter elope sce es complete met snes ae the rope tha rele a Se hc ae ners Fr cal ay oma 1, Prove that ER) i Beach pce 12-2) Rove and yar separable bt at (8) ove iat etorae ‘succes sumtin the" To hota Cae) segue SO ‘Ripert only mona show fas aegance comers) "4. Proe tha al noms on ae equal (Mt: Ute ft ha beware it Compact inthe Eins tepsan) Prove hat othe completion of rove hat i le he he he asta shen By ‘Mai. “Eu ‘bas norm Dis Aa. a ‘Prove that = ba tat 2 by ing the hana hoe (0) Ponte nonzre bounde naruto 26 £2 whch vanes once (e) Pov tat the Bsns nr eon LR) auch hat MY) = 0) foraeh fs CO 9, Suppoe hat 2 i iter sca hat ia bounded een on 2 1, Brave that tein fn an (8) 20 fim a shales Cah tolas ey age forall, n. » 2 rove the satemnt inthe example atthe df Seton & Vetheunfo boundete incl proviean aerate pronf the Helin ‘Toc thee Let.xtes Banach pace Giveancample ofan eeryrber find but dominant Tia enone Show eet) tat nt Sa, Le a tes arte lie gr wth an honor ue et Ba ‘Stuer of cement sand prove athe flog fata aretha (fend 2 vee (0 seam, nor ch me 2.4 ys oat ‘Asse $f Beach sce led went bountlf and ony i fora eX pacit cee cle sow tunked Wad sly Hope te Prova set stony Sodediand ony ts wey ounces econ V7, rove that spel continu miler fntonalo Banach sc sity en the Hensel hoe lei foes Bishi (nam Let XbeaBanachspuein ter of he or i oy Seppe ta he {or vane Prove tat threes Dh Fh Di ‘Why doen 2a pe Wala the Bue here? ‘Prove hat ay counts itrcton of dens open sti compete mer seit ‘ee (a) Prove tht Bach space Xi reine ad li eee i: Io ad ound a inl on" hh vanes o 2 (Prove tha whoever Hs a onde Baneh space, (0) 1a fete ate te pcan ah Sha eatin of Lette ines nto on tea Banach one Prov ta ker omer (sin te atmos aa hat the rater! projec "= Mer? = ted to by f= Ete ‘Banach sae cl wll centr ach e> 0, her 8> 0,60 tat Betsigiet aed Wrote t= nly br— ye; ths te voit bls ‘try rode Wes cen Problem 15 of Chap, tat every uiomiy(2) Prove les hat Land £8) ate teem cave, (© Prove hat ay Hilbert pice tery comes 1) Proven EU astm conn orp Hit Prove foro, Be, rehash > fan Bhs ula [BI by ha pon aes Bir = er SVR + wre 11 aca nr cone fr all 1 (6) Prove Ran YRan 4 ial af % Hu Given th Tato shoe ay eee by eng (4) Cone ta any separable Banke pce i tesla homer some 4) By wing) with 2 replaced by 3,4, show that db aca an isometry. 1M. Lec bea Banchepce and it ber slobpaceo Le Mn Reda by Pavexity=0) ‘Given Bound ines tinal fon XY, eine =f) yA NK =e ‘Prov tata ran eric aomarsi (37 ano 1.) Let be Basch space wh separable dal and M9 mesure suc ith [Dts seni for peel he try Le aa rouse esy of LM, 2) dace i eo lan 18 (2) Prove M0) ah ce ae tal wararpi. 0) Lees tens Bancn gran ltl po Prove at ME ‘naunly womanly omer to eB) Fs show a te oup i orve ha ey bounded lat weornaton Fat te EX, de) ‘Pott lem und = Um fo sre fe XM B*). rove a nt Spat se whee EZ, Ply we Poin 9a 30 eat Ue per ‘Serb nach sac, 122 La Sea lng of 6,1) Suppose ht imps that 2701 fersome p> Powe tat S248 forse poIV: Topological Spaces xenon now what ae at eh ugh nthe become one ‘rol he ote mer of pre expos kien iu ws eral notions ‘The abut notions of tin and convergence ar he bet ad butter of functional anata. The purely mec spe frmltions hat we hve tod ‘hus far ae sadly lacing in soe case, so rnc to introduce mre teneral concepts. Ii possible to describe whats known as topolopeal Space prely in terms of converges bit rey awa Inn, ne sully defies opoopal space by sbsrasig the notion a ope rewic spaces. Conernce then toons 2 deve concept. We ducus Convergence in Section 1V2. This ston const rary of defo at we inoduce an extensive language need to desert oploil nis We ure the rear tea the language by returning to thi section when ecesary rater than by Definition A topological space sa set S with a distinguished family of subsets J" called open es withthe properties: ©) F tect under fite nro, at i 4, Be, en AnBes, ve (Fis closed under arbitrary wwions thats if Age F forall in some index set, then Jour Gi) Gerandser. Fincalleds opaogy for S We wll occasionally write (S,7) for «topolo- cal space In contaisinction to Borel structure, topological structures are nor symmetric between intersection and union and involve not mevely countable operations but abitrary operations, ‘The prime example of topological space isa metric space. The open sets, J, are those sei, Mc, with the property (Yee MYGr > 0) (plots 3) <1) © At After digesting continuous fonctions, we wil describe othe family of examples. We rst mention, however, tw trivial examples ‘Given act, the family fal sobets of Sis a topoogy; iis called the dis crete topology. J = (2, ) i aleo-a topology: i called the indscrete topotogy. "The ailyof al topologies on aset Sis ordered ina natural way F< Fy M,C inthe sense of setsheoretic inclusion. If F< Fy, we say Fy ia weaker topology than J. (The term weaker comes from the fact that tore sequences converge ia than in 7,550 37 convergence isa weaker roton than 7, convergence) Definition A family @ < Fis caleds bate if and only if any Tes ofthe form 7 = Uy, for some family (8) For example the ball ina metic pace are always a base. We now take a whole family of definitions directly from mete spaces: Definition A set W scaled a neighborhood ofa point xe 5, topological space, i there exists an open set U with re UN. A family 7 of subsets ofS, topologies space is called a nelghborhood tase at if cach Ne isa neighborhood of xand if given any neighborhood of x, thre ian We with Nc M. Equivalently,” isa neighborhood base at x if and oaly if (MV © Af for some Ne) isthe family ofall reighborhoode of x, For example, if @ is a base for (Ne |xe.¥) isa neighborhood ba at x. We emphasize that neighbothoods need not be ‘pe, In metic space, the cloned balls of radio peatr than zero ate a neighborhood base2 Iv: ToPoLosICAL spaces Definition Ast C= 5,2 topological space isle close i and only ‘Tits the complement of en ope st. ‘The properties of the family ofall losed sets can be read off from the properties of 7 Definition Let 4c S, x topological space. The close of A, Zs the ‘oles closed set containing 4. The nei of 4, 4°, the ares open set contined in 4. The boundary of 4 isthe et 4° = n (SAT “Tata smallest closed set containing exists follows from the fact hat 'sclosed unde arbitrary unions ‘As examples, me consider several topologies on R Example 1 The ordinary metric topology. Example 2 _ Consider the family of sets ofthe form (x,y) |x € 0} where 1s fised and Ois an open set of in the usual topology. This family of sess "he base fora topology whove open sets are the sete C sich that for each YER, (41x y)€ C)isopen in Rin the usual topology In an intuitive ease, Which we shal shortly make pecs, this topology ithe "product" ofthe usual topology in ne actor and the disrete topology in the other factor Example 3 Let F consist of the empty set and all ses containing 0:0). A neighborhood base for (, x ctinuous foreach in 2 A neigh> borhood base for Ois given explicitly bythe sts Oyen Wa ties o6 82) = (Wee ON 0, Ysa te aritary, and n= 1,2... Ths the neighbor: hoods in the weak topology are cylinders inal bat tly many dimensions ‘That is there ie a subspace (he orthogonal complement of yor) whore complement, Ms fnite dimensional and so thal @e Nyne M implies orneN, (On R* consider the maps xy, m5 slven by w(x.3) = 5 y. The wesk topology defined by, and 2 andthe usual topology fon Ras rectangles (8) (c,d) base ors open sets and ths he weak topology ithe usual” topology onIv: ToPoLocieaL spaces Example 7 The weak topology can be used 10 topologize Cartesian rodvets. Recall {SJog i 4 family of sts, S= Xeer Sis the family OF al (sgn with xyeS. For cach , we have a mips: $-rS, given by Ssiglar)=%. It each S, has a topology J. we dine the. prodect topoiogy. Xai’ a8 the weak topology generated bythe projections n, We now return to our listing of defisitions by clashing spaces by how wel open ses separate points and cloed sete Definition (@) A topological space is called 27, space if and only if for al x andy, 4. there is an open et O with ye 0, x4 0. Equivalently, a space 1, and only if x} 8 closed foreach x (©) A topological space is called Hanson (or 7) if and ony i forall x and yx # thee are open sts 04, such that x6 O,,'€ 0, and O00 =o. (© A topological space i aed regular (or T,) sf and only tts 7, and forall xand C, closed, with x #4, there are open sete 0, 0, such tet 4 €0,, Ce Og, and 0; 9.0, ~ Bi Equivalently, a space es i the lose neighborhoods of any point are a neighborhood bse (8) topological space i sled normal (or T,) and only i ts T, and for all Cy Cy, closed, with C, 9 Cy = there are open sets Oy, Or Wit C, € Oy, C= 05, and O, nO, Obvioul Proposition T=T,=T,=T, ‘We remark thatthe two most important ation are Hausdorff and normal AX this time, we avoid discussing another say of separating st, nately With continuous fnetions. Urysohe’s lemm (Theorem IV) deals with tis question, ‘We nest consider various countably critei: Definition (Atopic spc cae separable a ony it has countable A topologies space $s called frst oust if and only if each point ‘xe Sas a countable neighborhood bate, (a) topotoial spaces called second countable if and only i Shas & ‘countable bse “The relation between these topological notions and metric spaces i et forward inthe elementary’ Proposition (a) Every metric space is fist countable (@) A metic space is Second countable if an only i itis separable (@)_ Any second countable topeogial space is separable, ‘Waring These are separable spaces that are not second countable (se Problem 7). To ad to the confusion, sme athors use" separable” to mean second countable. By seporable we alvays mean that there exis a coumble dns se, “The geomet idea of being connected hes a topological formulation Definition A topological space Sis called discomectd if and only i it ‘contains a nonempty proper subset, C, which s both open and closed equi lenily, Sis disconnected and ony i itcan be writen asthe union of to ‘isjoint nonempty closed sets IS nt disconnected itis ealedcomectd We examine connectivity in Problems 3 and 6 As a final topological notion, we consider resting topologies to subs. Definition Let (5,7) be a topological space and tet A= S. The felatve topology on Ais the family of sts 74 = (On A\Oe 7). A subst (Berd iscalled relatively open if Be F, and telaiely closed if A\B 6 F'y V2. Nate and convergence In this ection we introduce new objects, called nets, in order to handle limit operations in general topological spaces. Although ets seem on first scqusintance to be bute, the propositions in this section show how natural they are Definition A directed system is an index set together with an ordering hich sais Mx fe hen there exists 759 that y>2 and y> B. GH <8 paral ordering.Definition A net ina topological space Sis mapping from a dteted system 110 5; we denote it bY (8) It we choose the positive integers with the usual onde as dirested sytem, the et on tht directed sytem ee jus sequenes in, x0neteareageeralizae ‘ion of the notion of sequen. IF Ia) isa proposition depending on an index sna directed set 7 we sy P() seventy te if there a Pt true > f. We sy Pa) is requently tre if i 2 eventaly fale, ats Pfr any 8 there am 2 > with PG) tue Definition _Anet(x),rina topological space Sis sid to converge to 8 point eS (writen x,~3) if for any neighborhood WV of ther is & Belsothat eNifa> 2. Thos 5-4 iFand ony if xi eventually in any neighborhood of I, is frequently in any neighborhood of x, we say that xis a ctr point of {Notice that the notions of limit and cluster point generalize the same notions for sequences in a metric space. Theorem IVA Leta bea set ina topolosial space S, Then, point xis in the closure of 4 if snd only if there eet (sae with xy@ A, 50 that roof We fist observe that His just the set of points such that any rephborhood of x conais a point of A. The set certainly contains and is complement isthe largest open set not containing any points of 4 Now suppose x, = wher each x, «-4.Thenany neighborhood of contains some nad Hence some points of 4 that isa lint pot of 4. s0 ve A CConversey, suppose x A. Let I'be the coleton of neighborhoods of x withthe ordering Ny implies Fa) > f(hat is Fa) is eventually larger than any fed 6), the above ‘We then have the following proposition which shows th Aefinition ithe right one. Proposition A point xin a topological space Sis a cluster pointofanet (6) fand only ir some subset of (x) converges Of course, subsequences are subnets of sequences. But itis aio possible for ‘sequence in a topological space to have no converget subsequences but to have convergent subats (sce Problem 12. 13 Compactness + ‘The eader no doubt remembersthe special role hat closed bounded subsets of played in elementary analysis. In this section we wl std the topelog cal abstraction ofthis concept 1 Aseplmen 9h ion epi on 35Definition We say a topological space (S; > is compact if any open cover of 5 bas a finite subcover, that iif for any family @ with Sa Une U, there is a finite subset (Uj .., U,) =a with S= Utes Ur ‘Asubse Of topological spaces called x compat sti compact space in the relative topology Henceforth in our discussion we will lvays suppose that all compact spaces are Hausdorf, although occasionally we wll epeat this coiton for emphasis: ‘Since we have & considerable amount of material to diss, it perhaps sf to describe brie ly the contents ofthe neXt two Sections After studying some equivalent formulations of compacnes and some elementary proper tie of compact spices, we turn to some ofthe pil of functional enalyi We ft state and discuss Tychonos theorem, We then turn to the study of continuous functions on compact sets After showing that a compact Fiavsdort space X has los of continuous Fonction (Urysoh’s lemma), we discuss the Banach space COX) of continuous functions. We state the ‘Stone-Welertass theorem but defer it intructive proof to an appendix. In the next section, we determine the dual of C(X). Using the Rest Markov theorem, we wll prove that C(X)® i Mencal with -€(X), the fay of signed measures on. "We fist reformulate the notion of compactness by taking complements of open es: Definition A topoogial space $s sud to have the ite otersection property (ip) if and only if any fay of closed Sets F with pny AD for any finite subfamily (Fite © # sashes par FA: Proposition (ip. criterion) Sis compact if and only i Sha the Fi.p. roof, Let $ be given and ot # = {S\FIFe #). Then # has the property hat (tay # Dil and only sf 4” has no Bite subcover and the propery that (pus FAB if and only i. is not a cover. The reader iid to vend his way through the double negatives to complete the pref. [A somewhat deeper reformulation is ‘Theorom IV.3 (The Bolzano-Weirstrass theorem) A space Sis com- pact and only every net in Shas a convergent subnet. Proof Suppose that every net has «convergent subnet and let be an open ‘cover. Lets suppose hat has no nit subcover and derive contradiction. 1a Compuciness 9 (Order the finite subfamilies & of by inclusion; & is thereby a drcted et. For each $= (Fy ---» Fa) ©, pick Xp Uf, F. By assumption, the net sp has a chster points. Sice 4 ina cover, we can find Ue with xe U. Slace xp i frequently in U we can find a frite subfamily € 40 that (Wy F and pe U. Since (U}<9, Ue Yous ©, and 50 24€ User ‘which sa contradiction. ‘Suppose that $s compact and let (yey bea net. I (,) has no cluster points then for any xe S, there isan open set U, containing x and an 256 1 ‘thy, #U,ifa> a, Te family (U,| v6 5) open cover of S,s0 we can find) --1 3 0 that Jf Ua, 5 Since Fis directed, we can fd 69 8, fort 1m But jf Ugg Um. ns m which is impossible since Urey Uy, =: This contradiction establishes that (y)gy has & cluster point snd thurs convergent subnet. Second countable spaces are compact if and only if every sequence has 2 ‘convergent subeequene (this can be shown by mimicking the above proc). Example 1__‘The nitbllin/, isnot compact inthe meri topology: NO subset of quence of orthonormal elements can converge Example 2 Let S~(o.)e¢3| lol < 1m). It is easy to see that ¢ sequence of elements ofS converges fad only each component conver Using the diagonalization trick, we conclude that every sequence has @ com Vergent subsequence. Therefore, by the Bolzano-Weerstrass theorem, 5s compact. Warning Compact is not the same as closed and bounded in a general Banach space, tn act the unt bll a Banach space ke compact (in the norm topology) if and only ifthe space is finite cimensional (ee Probie 4 of Chapter V. ‘We now mention two simple “hereditary” properties of compact spaces (see Problem 38) Proposition (a) A closed subset of a compact spaces compactin the relative topology. (A continuous image of «compact space i compact. Corollary Any continuous function on a sompact space takes on its ‘maximum and minimum values, Tha iy there ae x, 40 that Lia.)=sep 0) and fe.)= iat 0)‘The followin theorem is often useful Theorem IVA Let Sand Tbe compact Hausdorf spaces; let fi S-+T tbe continuous bijection, Then i homeamorphitn, We nee the folowing lemma: Lemma If Tis Hausdorff and S.< Tis compact, then S's lott Proof Let x4 $. We an find net (yen $ with x,-4 x Since limits are unique in Hausdorff spaces, xs the only user pinto he oe. But since $ Is compact, the net has a clister pont a S, hat, x6. Thus $= S. B Preaf of Theorem 1V.4 We need oly prove fis open or equivalently, since {isa bijection, that /[Cisclored if Ceclosed, Buti Ce Sisclosed, then © is compact. By the lst proposition, /(C} is compact. The result now fellows from she emma, Proposition F(a, is 8 ay Aa ith the product topology is compact. roof Let legs BE net in A = Xtay Ai X= (Hh AEs. Sine Ay 's compact, we can ind a subnet bryalien, 80 that {ef} converges to an ie, By a finte induction, we cand’ suboet (yy), 30 th My Conerses to an sje, for each J. Then [tqu) converges in A to x= (Gost 0 AIS compact by the Bolzano Weierstrass erterion. I ily of compact sets, then “This last proposition snot deep; what is deep i that it remains re for an arbitrary product of compet spaces: “Theorem IVS (Tycho theorem) Let (Adyar bea collation of compact pac Then Kye) Ay it compac i the oduct (hat meh) ‘ooo Siac this theorem has amily complicate proof well-read inthe text= book iteratre, we refer he reader othe references given nthe Notes, Let vs, however, make several comments. We fst remark that itis this theorem that Supports the Teling thatthe weak topology isthe “natural” topology for Xe, Another a prior candidate the "box topology,” which is generated by ets ofthe form X, Uy, whee each Us open in A, isnot a topology for 1a Compactnese $81 which Tychonof's theorem hos, Secondly, we not that this theorem depends rucilly on the sion of choice (Zorn lem). In fact ite kown tat, st ‘heoretially speaking, Tychonofs theorem implies Zor's lemma. Fal, ‘note that inthe special cas of countably many mere spaces, Theorem V.5 can be proven by the method of the proposition and the diagonal tick of Section 13 ‘Next, we would lke to discuss functions on compact Hausdorff paces. We frst show that compact Hausdorff spaces have sang separation prepertcsia ‘the ens of separating closed ets with open sete, We ten ite hese separation ropes to construct continuo functions: ‘Theorem IV. Any compact Hausdorff space Xis normal (7). Proof We first prove X i regular (Ty). Let pe X and let C= X be cosed ‘with p¢ C Since is Hauser we ean nd, for any © C, open and disjoint Sets, U, and F,, s0 that ye U,, and pe V,. The (U,)jac cover C, which is compact. Thus U,.-.,U, Gover Le U= Ute Ug: Y= (hes Py ‘Then Uand Var open ad disjoint with C © U and pe. Thsshowsthat fis requler. Now et C, Dbe closed and disjoint. By repeating the above argument With D replacing and since Xs regular” replacing" sine Xi Hausdor." ‘we prove that fs normal. Normal spaces always have lots of continuous function for ‘Theorem I1V.7 (Urysohn’s lemma) Let Cand D be closed disjoint sets 8 normal space, X. Then, thee is a continuous funtion from Xt R with 0 s f(a) 5 I forall x such that (=) = Oif xe Cand fs) |ifxe D. Sketch of proof Using the normality of, one constructs by induction for ‘ach dyadic rational (hat is, r= 2, km integers, 0 = k = 2" open ves, U,, with Ce U,e U,< U,e U, = X\Difr ‘Ran p will be Banach space and thus closed Clearly loge 5 Ila #0 VAL/Dleg $ HL Heyyr= Thus. tis ‘enough to show that given ge Ramp, we can nde Cy(X) with 9 = p() and lew [len (femenber the defiition of quotient norm), Since age Ran p, we know that g = p(h) for some hy © Cy). Let 4g = min legs) So that fh) = g and AG) < lite or all. Let hy = max{~ lle Then, Hslegin = Ilo 80d Cy) = g. Thi completes the proof. Tn this appendix we prove Theorem 1V.9n the case 1 ¢B. The general proof islet as an exerise. Interestingly enough, the ist step in the prof isthe proof of the clasical Weierstrass theorem (which is 4 special ease of the beneral theorem!) Lemmat ‘The polynomials are dense in Cla 8] for any fit el nam bers, Proof See Problems 19 and 2, ‘This can now be use to prove tht Bis lattice, where Deft from nA subset Se CX) is called a latin if for all 9S, inf fg} and fv g = max 9} aren. Lomma2 Any closed subalgebea B of Cq(X) with 1 BisProof We show that fe B, then {| «B. The result then follows fom the formulas: fu 9 = 41/~ 9] +440), Fag= ~U-f)v (~a)) Without loss suppose that i lg 1. By the elasia! Weiertrass theorem, we can find a sequence of polynomials Px) converging uniformly to |] on 1,1, for txampl | P42) — [x||-< Ifa forall x (0,1. Since Ile lit follows that IPD) |f|fe-< Um ie [J] =litene BAU). Since Bian algebra with 16, AU) eB Since B's closed, |7Fe 8. Finally, he fll StoneWeieestras theorem i a consequence of Lemma 2 and he following theorem which sof some nterest in el Let X be a compact Hausdorff space. Any lattice © Cal] which i eloed subspace contain ing I and which sepacates points ial of CX). ‘Proof Lat he Cy(X) and let ebe given. We seth fe ¥ with thf <8 Suppose we can show for any xe X, there isf.€2 with fe) = Me) and hrsf. Then for each x, Bnd U,, 2% open neighborhood of © with ng) fp) ~ forall ye U, (by the coninity of hf). The U cover X40 ho Am obeys fU) t= Ky). Moreover, since any' ye U, f08 some f, 0) — {io)~€5 Ky). Thos if hie B, then f~lim,f,» sp, J, i64 Borel function because FM co = YF a.) is open. The monotone convergence theorem bas the fllowing net genralia-Theorem 1V.15 (monotone convergence theorem for ets) Let be a ‘teglar Borel measure on a compact Hausdorff space X- Let (fees bea increasing et of continuous functions. Then f= him, foe L'X dy ifand only if sup, Ul <2 and in that case lng I Before leaving measute theory on compact spaces, we should identily the ‘ual space of C(X). OF course, not every continuous linear functional on CCX) is postive linear functional, bat the major result we are heading toward is that any ¢ © CX)" isthe dierence of two positive linea func Tionals. This depends on & simple “Iatice theoretic” sesh about CX) Lemma Letfgeca(X) with/g 2 0. Suppore he Cy(X)and0 sh f+. ‘Then, we can wete A= by + hy with O-< by cf OS 9. hy, hye Ca) Proof Let hy = min(f 8}. Then 0 hy =f and if hy =~ hy then hy 2 0. Moceover, i h(x) =x), then his) =O a) and if OO) =f), then 43) = M2) ~F0) Sf) +2) ~F1a) = 0,508, <0. bet £0 G(8%. The f can be wn = 050. ind fore tet Racing, Mareen.) e602 V7vand wig temic és a Prof Forfe 1X), = (6 CX) | 20 ee) = Ihe: 854 OIF isconvex ands one it iscalledsition Let X be a compact Hausdorff space, Then fy, ,(X) ‘is conver and. (1) is a convex cone Proof A convex combination of postive linear functionals clearly «pos ive linear Funetional. Moreover, |¢1 = 1) isa positive linear functional Bo (= NEL ihe ee [At fst sight, this geometric fact may appear a litle strange since the reser used to thinking of the nit sphere, (=| x= 1) at "round and ete we are saying a pice ois absolutely ft! The moral stat every norm |S nt te Eucligean norm (the parallogam law implies that ina Hilbert space, if nlm y= I, and x Ay, then ex + (1 #)yl <1). In fat, vith the norm Id, Bin, [| has 2 aie sphere wih Bat fees, sce Figure IV.2. This i nota coincidence; (1...) is 8 compat tat when Bien the discrete topology, and Re with the norm considered it presely Meh Now, we want to extend “tpolopieal measure theory" to a larger clas of spaces Definition A topological space, Xie called locally compact if and only ‘every pont pe X, has a compact neighborhood By thinking of Lebesgue measure on R, we realize that we wat orl the condition n(X) < co which We required when X was compact. Weft deine "he Bare sets in Xa locally compact space, tobe the a-ring & generated by sompact Gy ses. Note that, general, may got be a tet of. However, i ‘ isaccompact, hat sa countable union of compas set, X isin Definition A Baie measure on X,s locally compact spac, 8 measure ‘on the Bure sets for which p(C) < 00 fr any compact Bare st C. WV Weak opoag nach spacer Given any Bare measure 1 on X, and given a compact G, set C= X there ‘indice by retvetion a Bare measure on C. Convers itis easy tet ‘that family of measures (ji) one for each compact Ge, with the property that eC) = nol YD if Ye Ce D, defines a Batre measur. This association allows us to prove theorems inthe locally compact case from their compact ‘ase analogues Definition Let X bea locally compact space. x(X), the algebra of eon- timous fonctions of compact support, isthe st of functions that vanish ‘outside some compact et. C1), the algebra of continuous functions vansh- {ng at is the set of Fe CCT) with the property that for any ¢> 0, teres & compact set D,= X such that [/(3)|- Inthe next chapter, we wll nda topology on xX) for which the duals jst the complex Baire measures. Notice that this topology is ner given by Tr fo 1) snot compet inthe norm I> its completion is C(x) and fis Sali > heft measutes. V5 Weak topologtes on Banach spaces Definition Let X be a Banach space with dual spice X*. The weak topology on X's the weakest topology on Xin which each functional in Xs “Thus neighborhood basa zr forthe weak opology it given by the sets ‘ofthe form Nese) [EA] <0 FD 1 Asap ote tion beg on 14tm Ww: ToPoLocicaL spaces: ‘hats, neighborhoods of zero contain cylinders with nite dimensional open bases. A net (4) converges weakly to, writen 7.5 and only if x.) + @0) for ale X*, For infinite dimensional Banach spaces, the weak topology doesnot arise from a met. Thisis one ofthe main reasons we have introduced topological pues Before considering examples, let us note thre elementary properties of {he weak topolory Proposition (a) The weak opologyis weaker than the norm topology, that i, every weakly open sets norm open, () Every weakly convergent sequence norm bounded. (©) ‘The weak topology is 2 Hauadoot topology Proof (a)ollows from is) 11 Ux (isa consequence ofthe uniform boundedness principle: and (e)folows from the Hab-Banach theorem. We leave the details to the reader ‘We emphasize sat (0) sony true for sequences, In Problem 39, he reader isasked to constrct a counterexample to the aalogovs net statment, Let vs consider two examples in Doth of them, we wil describe what i means for sequences to converge. This does no completely describe the ‘topology, but it wil give the reader an impressionistic view of the underying topology Example Let 2 bea Hilbert space. Let (gue) be an orthonormal bai flor 2 Given a sequence, € let 0 = Coa, b> be the coordinates ‘of. We claim Win the weak topology i and oly (0) YP = 6 for ach 2 and (b) [yal is Bounded. For suppose "then (8) follows by Asfnion and (e) Comes fom (i) af the proposition. On the other hand, let(a)and (hold and et F= 2 te the subspace of ft nea combinations ofthe 2s. By (8), Co.) Co, 99 it pe F. Using the Fact that ie dene, (@) and‘an 3 argument, the weak convergence follows, Example 2 Let bea compact Hausdorff space and consier the weak topology on 1X). Let) bea sequence in CCX). We claim .-»/in the weak topology ind only if (3) (2) f(a) for each xe X, and (lis bounded. For if ff, then (a) holds since fm f(x) i an element of G(X) and (6) ‘comes trom (i) ofthe preposition: On the other hand if a) and (9) nod then [£(2)] sup, Vile whichis! with respect to any Bite measure VE Weak toptopee on tanachspnces 118 ‘Thus, by the dominated convergence theorem, for any 1 ¢M X), Je -+ [fa Since any ¢€.4(X) is fine linear combination of measures in A we conctde that =f weakly. ‘We have seen thatthe weak topology is weaker than the norm topology; cualy, itis very weak indeed! To sets, we note tht having few open ses sth same as having few closed ses and this nthe same a bg closures. In Problem 40, the reader wll prove thatthe wesk closure of the unt sphere, (ee X | sl'= I) in Xis the unit ball, | [el =I) inany infinite dimensional Banach space ‘We will shortly study general “dual topologies. As a special ease of| “Theorem 1V.20, we sa Theorem 1V.19 A linear Functional / on & Banach space is weakly cone Ninuows if and only fis norm continous, While this theorem follows fom Theorem 1V.20, it has a simple direst proof (Problem 4) Finally we should like to discuss the weak- topology and prove 4 com paciness theorem which wil often beef we tous Suppose P= Xi the ‘ual of some Banach space X. Y*= X°*, of out, induces the weak topology ‘on Y, but we may insiead consider the topology induced by X acting on X" cexpiily: Definition Let X* be the dual of « Banach space. The weak apology isthe weakest topology on X in which all the functions é+0(3), 3 Tae continuous. Notice tha the weak» topology is even weaker than the weak topology. As ‘one might expect, is reflexive ifand ony ifthe weak and weak-+ topologies Goincide, and many characteriatons of relenivity depend on relations Involving the weak and wesk- topologies. ‘To avoid confusion and to be able to sate our nex theorem nits natural setting, let us introduce anew notion: Definition Let X’bea vector space and let ¥be family of liner fone Yionals on ¥ which separates points of X. Then the Y-weak topology on X ‘writen oC, Y), sth weakest topology on X for which al the functonals in Y are continuous,Because ¥ is assumed to separate points, o(, ¥) i 8 Hausdorff topology on X. For example, the weak topology on X's the o( KX) topology wile the a(X*,X) topology i the weak topology on X*. The eX.) toplogy ‘pend ny on he ear pce nerd by T, 0 we beef spon that Ysa vector space. Example The weak topology on A(X), X a compact Hausdorf space, soften called the vague topology. To get an iden of how weak it i, Jet us show the liner combinations of point matics are weak» dense in #(X), In Problem 41, the reader is asked to show they are actally noe lose. Suppose that isa given measure. We ist show that every weak neighborhood of y contains a sum of point measures, or equivalent, given Sines fo ad that We e88 Bad ay ny ANd foo Kq 30th mn- fal For then $28,, will beim the vague neighborhood NU. ofo6)-+ Without loss, suppose thats ate linearly independent. Far each consider the vector f= C/(B), on fe) # RI the () d0 not span RE ere is an 8 ay) #0 688 with af, 90 for all x, thats, Etesa.fi=0 contradicting tineat independence’ Thus, the f, pan RY So we cam find xy Ry and gy 53 th in. Add =Eah, ‘So, wf) = Las 44/4) which proves our claim, (Kee foe fe oa ‘The o(X*,X) topology is of eourse weaker than the noem topology on X* so all the (X*, Xpcontinuous linear Tanctonals are in X*™ In general, however ot all €*" is weak continuous on X°s ia fets ‘Theorem 1V.20_ The o(¥, ¥) continuous lace functionals on X reciely Yin particular the only weak» continuous functionals on X* are the elements of X Proof Suppose that ¢ is a a(X, ¥) continuous Functional on X, Then [ie 1) > Ce] yay cert lyse for some cand some jue Yue Y. Now suppose that (a) = Ofer i= am Then [eH 1 fot alle> 0, which implies hat) =O. Av a'esuit 2 isto fnetional ? fon X[K where K'= (x|y(x) = 0, |, -.m). Elementary sbtract algebra som Ju pt th al ic of HE Tae? = Bs 0 ta Finally, we conclude this setion with is most important result, a result which is perhaps the mot important consequence of Tychonof’s theorem: ‘Theorem IV. 21 (ihe Benach-Alzoglu theorem) Let X* be the dual of rome Banach space, X, Then the unit Ballin X" is compact inthe weakew ‘opoloay. Proof For each re X, tet By = (L.€C | [il < 1x). Each B, is compact, 0, by Tychonos theorem, B= Xp_4 B.S compact in the product topology [Now what is BY An element of Bs just an assignment ofa number x) € for each x in X, thats bis function from X to € with [4G)1 [xl In articular, the uni ball), isa subset of B, namely those be B which are |sthe relative topology induced on (1°), bythe product topology [reciely the weakest topology making Z++7{3) continuous for ‘each x, thats, the weak-» topology. ‘Thus, we must ony show that (X"); is closed in the product topology, Suppose that 6 net in (X*), with 4,» Since |) = Ix, we ned only show (i linear. Bu his is easy if xy X and dye C, then ax wa) = i 20 + wy) =i 8) + wl) 2000) HO) Appendix to {5 Weak and strong measurabilty In Section TL, we briefly dscused vector-valued measurable functions with values na infinite dimensional Hilbert space 9. f was called measur ble (in Problem 12 of Chapter TT) i, /() was a completralued measue able function foreach ye 2. Tie notion might be called meak measurably ‘Another natural candidate for measurably isthe a priori stronger notion of ‘measurably which requtes that f~![C] be measurable for each open set Cex. Throughout this book, by @wetor-alued meturable fonction, we will ‘ean a function mearurable inthe weak gone, However, tatty the fea’: ‘arora curiosity, «bie comparison ofthe various notions of measurably of vecoesvalued fanetions seems in order Definition Let f be a function on & measure space CM, ®) taking values in a Banach space E:(©) Fs called strongly measurable if and only if thee i.e sequence of functions so that (2) ~/(2) in norm for ae re M and each f, takes only fnitely many values, each value being taken on ast in (i) fis called Borel mensrabeif/-"[C]=2 foreach open set Cin EG the metre space topology on). (i) fs called weakly measerable if an only its) ia complexalued ‘measurable function foreach 7 E* Proposition (a) A pointwise limit of & sequence of Borel measurable functions a Borel measurable function. (0) La/bea function rom Af 19 E If fis strongly measurable then fis Borel measrable (6) Let bes Tunction from M ta; fis Borel measurable, its weakiy neaserbe Proof (2) Let f+ pointwise in norm, Let C bea open set in. Let CL (x1Bt., © C) where Bs the all f rads about x Then, ra=0 0 artca £0 fs Borel measurable () This isa direct consequence of 2) and the definitions. [e) The composition of Bore! functions Borel. ‘Theorem 1V.22 Let bea separable Hilbert space. Let f be function froma measure space (M,y,3) 1.9, Then te following thre statements are equivalent: (a) fissteonpy measurable () fis Borel measurable (6) Fis weakly measrable ‘Proof By the lat proposition, we need only show that) implies (a). Let. ‘Wom be an orthonormal bass for 3. Let 2, ~ (Was Jta), ach eis complovsalved measurable function. 11 easy fo construct y(t aled, [a2 = [a] forall wand img 8,8) = a) Oral € Define fy = ter ane fy is Bete valued ad fy +f in norm sof 1 strongly messorabe Example Let ©, be a copy of the complex numbers © and et 2 = Brea Cy thatis, # consis functions yon R, nonzero at aly count- ably many 1 with Zag ||? < co. Let ¢, be ziven by enh Es 10 otherwise “Then (Jem it an orthonoeml bass for 2. Let f: 8+ be defined by J19)~ 91 For any Ve 2, (Hf) = 0 excep fora countable set 30/0) {pimeasirable. Thus /a) ie weakly measurable. But fis not tongly measur= able; for i= lim f, pointe in norm, then Ran fe [Ran I each f ‘wee nite valued, Ran f would be separable, whic it 8 not ores Seen) _ are tates a ata an oe oe dae, tem Sea dl py, as Neca teo Sint ee ni ery sey erode kde ister, nano tt alr ca eer cod SESE Geese ny peed ep nts me ee ctemrtcsor a ena Roe ren Fos Er oa as Noe You s6W Ph Fo Se cer wo isto et tt Fate nt a he F-F camer ara as tt hap ses er 8 a ym et ted hey poi ee ee a sere oopadazen ee emia Cua san atone ae ov ms ea Sg LR SAE Pe ea ecenc agent oe wd Neen a Ts famine nec crse ny we Ceara Sues een em icon eaie Tee eng Ticimnlangnala tte ay oat Aye esate cepeapeowmpemene oat Tings Latina e wrieopal cs nay eran ant Sar TAEE ROUT Ntafane yt ey pn oe BIg Ton fp otis credeentodumntesanetelae oad Semele oearieet ane Distat apo (ort ac dni open ove hate be (0) Poets send Coulter hat 7) pnt soma Ghat To ‘RefreKe 9p. 108 (0) Prove ta Rad R arene homeo foray > 1 (©) Prove that 8 § forty pope apace ‘ar Wha been to Risse po emored? -Atapolc sce Xi cle rv comece gen 4.762 hr Sur fncten an ar Bt] eX wD) Ae (2 Show nai ro Comet, conan (8) Le Xa bese ar of easton y= in on ~ (ven he ete op yaaa the ne Lat X= ay (nix Show ts a Y= ih the tpsogy F pnt by al st of te form a6} sha bea aban toe te E33 acon (2 CF ar ond cou, Prove ha stone of parable ee sepa et we te es en ying eo of ean 7 6) Yb separ ean emanate tan 7 eps ong tise (0) Xiscompecr (G) Xaver fst bor second countable Prove Thorn 12. Le be the Ranch sp eon te sens By Ingen by 8a Prove has no Wek coneret sharin ut th th a ek con sent tne 13. Givean example sow that a pointe i of tf Be ects Ry 1H, Show that he pce of the example in Section 1V2 net compat bt Lindt (Ge Prot 15. Lathe iy of conosco on 2 wh he sapere) = 0 {Cais ete ge Prove ran sissies ps wth tea bu for whch a's lode 116, Prove the conan of the Stone-Weestass hoe ith ase where we dost “17 Let be a ideo Cf ih cles. Let Y= (re HY Le) = for al fe Prove tha Yisclsed and hat =e Cal = 008 18, Prove te Tete theorem nth cate wen Xie aan orm (Stee ‘Doma Ket, Chater 7, robles OF 1. La fb 2 cominaas fron on (4:4 wh J) = /( HO. Lat af) be 8 Seo ffenane th ys) de~ yeh n S03 at for any 88 fi fauinente)=0 Prove that in Jifante 70d =s0 foc any £614, Band a te convenes orm. 20, Let) = ("1b 2 whe af (1 — 2 de Us Pre 9 pov that $y enoianus fncion on [a tm payor kt (teens rem compe saya ot fr LO, 2 {The Lae polo aru compte repo Str (—1 1. 70) ph amen we coe sol 1 pe (Hint Use your knowledge Cech Ccrde aise) 2. Prove Din’ hem: et Yes compet Hamat ace Supt f Sah often ah = pote Tan sae ly ian clit coon 28 Let X ec compact Hada pce, Coney T= (=) were 6 “toi notin CalO= f opener = # Dard Oisopenin Verse Om iO weampact Prove tat Fon consi! Had sce! sae the opt ‘compton 24, Prove the Stne- Weis thre for aly compact space Xt fe ached Subba of Cte conus reese fancies anshing at. 28 oF Sepa plns an freak x three with) #0, mo = C3)Sooper coninouraneon/.0 2/2, o9 wth MCh= 1B) ‘Rear Use esc Xo Probie 1 fone Mand 3, 26.) Prova any ely comes Hadas (Prove ta any cond county eompac Meade apace nema (6) Peet any oompat sly compas Hassan ‘nar Tee et ey como oer ih ae Haus bt or ee Katey, Chap Probie © 2. wou 6 wa ope sce« optoges! run the map xy 9" of Oe CG jail omtnance fun fons opto! pp ld un Toray coninuou ff any wecan Sn ightorood N06 Che et) ‘nih 7) JO) « tify eh, Petit any continous anton om sepa ‘Sonapalprup i unomly comin, 124 (0) Let af bam ales of read ote continu ution on wich ‘paras sos ae ced in Vo Form eu" » nace Rls] SHft0) ‘nt pode tocog. MR» yy ting = go the pit nono rates [/ ea Prove hat he nag of Rin ars homeo fo ‘Tad eaty conti naan compen oppo 0 Atl eat t Ths cmp 8 emacs space Two conpecientons/ R= andy: > Yurecon- Street hte heme ¥-= othe =. Prove th tre a one-one coespondens betwen conasceaton of Rnd slphat IF Cy sting he cantons (@ ifswlake = Ce he corpatitog ne bts the conto i tae te Stow Cech comport, Prove that Rit ue oma Feson of Rin te flowing sera’ nena compaction = aed fen Stone Ch compacisntiong: =H wecan nd > Vento Srisuetne wihihsg= 29, Lat a> te mati sgce with no wed pum Suppone hat sone Finis ns vl ominous Show at compe 10.0) Prove ta even met sue nvm. (0) Prove hat ey sod tins mer space, 191, Prove te igen statement of Thee 16 22 Let a be ace of manbr wh he fawing propery If Blue a= for Site fiber yao, Prove tha re gu, (ee) mej iivihan [bed 1, La. tor space ih Yaa fons separating pint, Prove haf theo Htoplogy oer omamere the Peas conte seo ners Ie lsat bor Pb robt whore ve tact spa Te ‘Sens mcnson te number of Semen is ase he 34 Lee te ral Bach pce and et Cte eit allo 2 wh te ek topo ‘Prove that contortion on Cc ior aoe pain. inthe denen acing bee fons on 25, Let Xbea Beach ape, X* ul Lat a> edema with a+ Le + ithe mek sn Lt aim oem i ect Li) 2? 25 Prove tha ide Xin te (1%, tp 37 Let TCU) = UN oe. Wey Tia post peseig or poste 1720 sienoee/20 I Th potve we wrte TaklfST 20 we wrle T= (Prove tr any T= 0 stomatal ction aad ha ITT (Las, ean mesg my of map Prove hat Somer operice nom, {Caden St omega ontn nem. 138, Prove test propio a Sesion 12. 19, Find Bach cea a weal severe net wich et arm Bounded 110, Lt te nine denon Ranch pe wih te ak oplgy Prove tae ‘Soe of tet per the na 1, Lt x be compas Had pte Prove th these of comer ine Het ‘rbimion plt entues oem edn 12, Prove Theorem 1.19 det #43 (0) at X be compat with a outa Lt be Be meno Prov ta op eprops (Re Ltt bes ote bae fats For alm vi Ber Aa ee 9 wn =O def nt. Use then oie cau tin CX) Te eh ha Goi deme ln 40. (Ene erat of 10 thee ere Xe ony oy apc (Hi: Poe 4 Do any ty poten in Kellys ook,': Locally Convex Spaces Masher rie Freer har ou 1 them they ron thr ome lone andra some eily eet ‘Tw. Gote We have already discussed several nonnonmed topologies on vector spaces in the investigation of weak topologies in Section 1V.S. We have alo lade to the fat that i i a loally compact topological space, the Baire rreasues on X ate the dusl space ofthe continuous functions of compact Support, n(X), when tis given sulable nonnormed topology. Our goal in his chaper ito discus & general clase of topologaed vector spaces which includes these examples and also the spaces of dstibtions which aie in 8 ide variety of functional situations and physical problems. The ides behind the topologies we discs is quite simple, Suppose that, iosead of one norm, we havea family of norms (eax Where A is tome Index set. We should like a topology in which a nt {3 converges to = if and only if pny ~ 2)-20 foreach faed 26 4. However, iis weil to ‘weaken one condition onthe norm. Recall that [xl|= 0 ips x= O and {hat this condition is needed for limite to be unig, tat ifr the induced topology to be Hausdorf Suppose that (oi aml of objets obey- ing all the norm conditions except x=0 when p(s) ~0 for some 2. But suppose instead that x = O whenever pe) =O for ll; then tis easy to se VA Gane properion 188 timits ate unique ina topology where convergence means py ~ x)-+0 for ‘each fixed a. We thus deine: Definition A seminorm on a vector space V is map p ¥~+(0 2) obeying ) wet) sols) +90) plex) = IalpG) for ae (or RY [A faily of seminorms (nee «sid to separate points i (i) pls) =O forall ze Ailes Definition A locally convex space isa vector space X (ver Ror C) with 2 family (nea of seminorms separating points, The natural topology on 8 focaly cones space isthe weakest topology in which all the are con tinuous and in which the operation of addition is continuous. We temporarily defer giving examples or the explanation of the term “jelly conven.” We also note that many authors donot require the semi. norms to separate points but add t as an extra condition The significance of the separation contin ie that it implies (Problem 6): Proposition The natural topology of locally convex space (with our efntion) i Hausdort. ‘A oeighborhood base at 0 forthe natral topology is given by the sets (Messen ones Que Ave» 0) where an la) 66 hoon) 6 4. The notion of “Thus, a net xp if and only if ay ~ 3) 0 for completeness extends naturally: Definition A net (x) in locally convex space X is called Cavchy if And only if, forall «> 0, and foreach seminar p, tere isa By so that pulp) <0 if fly > Bo. X 18 called compete if every Cavehy et ‘converge. “The important strocture on a locally convex space isthe natural topology athe than the particule seminorms sed to generate the topology. We call‘0 families of seminorms (Paes4 and {dye 8 8 Veto space X eqlalnt Af they generate the same natal topology. It ie often useful 10 know (Problem 66) Proposition Let (piee4 0d (py. 4 be two files of eminorms. The following statements are equivalent (2) The faies are equivalent familie ofseminorms. (©) Each pis continuous i the natural topology and each dy is con- tinuous inthe p-aatural topology. ©) Foreich aed tee ar fi Bye B and C>0 1 that for alt PAA) 5 Cd.) #04 00) and for cach Be B, there ae yg A and D > Oo that forall 6 X 4402) 5 Dioula) ++ pe) Tre appearance of expressions ike Clip) +--+ dy (n)isquitecommon in the theory of focally convex spaces I is ths wef to conier families ‘of seminorms with a special property Definition A family (7iee4 of seminorms on a vector space scaled ‘rete if and only ior all 2: Be 4 there va 724 and 4 C0 that 0+ 92) Cosa) for all xe ¥. Equivalently, by induction, fr alo 4 thea 2nd Ds that me seed terisay Pale) +04 al) Doe) forallxey. Forctample i yalee4isa directed fly, hen (x(a) 0) ‘sa neighborhood bas 0 One ean alvays find diected sets of seminorns Proposition Every locally convex spce hat dicted family of semi- ‘norms equivalent tothe family defining the space. Proof I Upsoeg defines the space, let B be the set offiite subsets of. IF FeB, let dy = Syurpe Then ij)pyp i Gcted and uaa fo the inti see Vs General poperiaa 7 ‘We consider briefly two examples. In Sections V3 and VA we wil cuss several other examples; in particular, the technical appendix of Section V.3 is seul to the reader desiring experience with equivalent semi-norms and directed sete of eeminorms. Example 4 Let X bea vector space and suppose that Yis ast of ina functionals on X separating pola In Section IV, we introduced the {o(%.Y-opology Kis preset the loealy convex topology generated bythe Seminorme ([email protected] Y) where pds) = |e). While this topology is given by Seminorms, is never given by norms i Yhas infiate algebrae dimension (roblem 2 Example 2 Let D bea region of the complex plane, that it, Dis cone reve and open, Let 6 be the veetor space ofall ngle-alued) analytic functions in D. For any compact Ce D, kt pf) = sieve |f(@I- Om topologized bythe seminars pe, isa locally convex space whichis compete For suppose fa is a pe-Cauchy net forall C. The, f(2)-»f(2 uniformly ‘on compacts. By classical theorem of Weierstrass, /isanalyte(essetilly because is analyte and only iit obeys the Cauchy integral formula which is preserved by uniform iit). Let, une ff veoires ‘tei Pad ett aia ote We are now prepared to discuss the reason forthe name locally convexand the associated, geometrizal ideas and constuction, The neighborhoods Nausea BV special geometric properties: Definition A set CV, vector spe, is called convex ix and ye ¥, Osis implies 1+ (I= ye. Ci called babaced (or eeled) ice C tnd [a] implies Axe C. Finaly, Cis calle absorbing (or absorbent) if Ure = ¥, that is if for every x6 ¥, 2x6 Cfor some s> 0. If Cis convex and Vis a vector space over the reals balanced means ooly that =e C whenever x6 C:if Visa vector space over the complex numbers, balanced means ee C whenever 8 [0, 2x) and x C (0 circled is more sable name) isan elementary aplication ofthe definitions tose that the a, ae are convex; in fact:Proposition Af pay. sPu a0 minors on a vector space ¥ ten (lla. oul) blaced, comes, boring st “This proposition i @ onehalf of the basi theorem Theorem VA Let ¥be a vector space with « Hauidort topology in which addvion and scalar mulipieation are separately continuous Then Js locally convex space (Chats, bs a topology piven by a family of semi- norm) if and only if 0 has a neighborhood base of balanced, convex, bsorbing sets “The proof ofthe other half ofthe theorem, that i, that V has a topology generated by seminorms if 0 has 2 neighborhood base of balanced, conven, absorbing ses relies on the following technical device Definition Let C be an absorbing subset ofa vector space with the ditional property that ifr C and 0-1 <1, then te C. The Minkowski functional or gauge of is the map p:¥ (0, ) given by ks) ~ ina 20) = foplalxe Lemma (a) If 2 0, then plex) = 96) forthe gauge af any st C. (©) p obeys plz =)) $06) + (9) if Cis conver. (_pobeys alin) = [ipo Cis cicea © (lee) NC etelats) sh “The proof ofthis beautiful lia eft to the problems Proof of Theorem V1 Let 4 bea neighborhood bese a 0 containing only ‘comes, balanced, absorbing Sts Tor each Ue lt py Be the gauge of U. By (band ©) ofthe lea, p isa seminorm and by (8) the neighborhoods ‘of On the erignal topology are the same as those in the Tealy convex topoiony given by the seminorms (| Ue}, Since addition is separately continuous in both topologies, the neighborhoods about any point ate FMenical in the two topologies. VA Game sropeniae 128 ‘normed linear spaces, linear map from X to Y's continuous if and conly if is bounded. A. similar reno holds in locally convex spaces (Problem 9): ‘Theorem V.2__ Let X and Y be locally convex spaces wth families of seminorms (Pee td (ye 9 Then a liner map T: XY, is continuous ‘and only if forall Be B, there area. 246 A and C> O with AT) 5 Clog (ad 44 96009) ee (odes ate diecte, then Tis continuous if and only if for all Pe B afr) = Dole) forsome ae 4,and D>. Finally, we conclude this introduction by discusing two applications of the Hahn-Banach theorem (Theorem IIL.5) to Toally convex spaces. Fist: Theorem V.3 Let X bes locally convex space and let Y= X bea sub- space. Leté: ¥ +R (or Cif X isa complex space) te linear and continuous. “Then, there is continuous linear map L: X= R (or C) with roof The relative topology on Y i gven by the restrictions ofthe cone tinuous seminorms to Y. Thus, [¢3)| = Cp() for some continuous sem norm. Applying Theorem ILS or IIS, we obtain or result. ‘Thus, locally convex spaces possess many continuous linear functions; in fact enough to separate points. We denote by X" the family of continuous Tinga functionals on X snd cali the topoogial dol "The second application ofthe theorem is more geometric in nature and is related tothe idea of siping a closed hyperplane between disjoint convex set; se Figure V1. A hyperplane isthe se of points where €(2) = for some ‘reat elued Vines Fontinal (even inthe complex cas). Definition We say that two sets 4 and B in a locally convex space are separated by & hyperplane if there is 4 continuous reabvalued functional ¢ Andan ae R with 3) 2 for xe d and €() > a for xe. I 4x) <0 for eA and €3) > afor xe, we say that and B aresteletly separate‘Theorem V4 (Separating hyperplane theorem) Let A and B be dite joint convex sets ina locally convex space X. Then (3) 14s open, they cam be separated bya hyperplane ib) A aad a bh open they can ey erated ya hyper (©) IF Ais compact and B is closed, they can be strictly separated by a byperpane Proof (a) Pick eA ~B={y~s|ye4,2€8). Let C= 4-8 + (2). ‘Then Cis open and thus absorbing convex, Oe C and x¢C since A and B ate disjoint Let pc be the Minkowak functional for ¢-"Then pele +9) = Pde) +c) and peas) = apes) it a>0. Define ¢ on fislieR) oy Gx) = 2 Since x ¢C, ple) 10 that Cs) < els). Ths by Theorem ILS, © as an estersion to all of X with £0) pc). Since Cn (Che ©M=1, 1,6 iscontinooss. By the inequality, 2) 1 ify C, Thus for any eA and Be B, Ca) <2) + (I ~e(e). Since Cs) = 1, pela sink 6) 9 ¢ ented and B (Oi ieny ose that ono ear fnctonl nd ioe, the A nape and ia ope a eet nto pom hyo taal (© Sine 4a B are da 048 = A= § Sine dis compa tis cytostatic Tha eran pesos Uae to SeihocUita at" Wand = 310 ten aaa open cone st tat aye sy pare by pepe Sho °c 4"and Be ths hypepae so spate and BB In Chapter XIV we wil discuss a “algebraic Habn-Banach theorem,” ie «form of the separation theorem which makes no mention of open sets oF Continuous fanetons V2. Fréchet spaces In Section ILS, we saw that complete metre spaces have special properties which imply strong results for Banach spaces. Its thus of interest single ‘ut those loaly conver spaces that are also complete metic spaces. First, we rust ask which locally convex spaces are mtrzable, tht, havea topology generated by a metic: These ae ot only the spaces whose topology i given bya norm, for ifp isa metric, p(x, 0) need not bea norm since p(x, 0) need sot equal ips 0) Let X be & locally convex space. The following are (2) X is merizabe (©) 0 hast countable neighborhood base {The topology on X i generated by some countable family of semi Proof We show (#)=(0)> (0). (a)>(b) isa property of any metric space. {@)= (follows from the fact hat Fie any neighborhood base of convex, balanced ss, he gaupes of Use % generate the topology and the fact hat (Ona. countable neighborhood base, we can finda countable neighborhood base ofconves, balanced sts (=) Letfpaens.., Dea family of eminorms generating the topology. Dine pon frfae=e 7 wore Sole “ Since al +2) <1 for any a>0, px.) <<. It is easy to se that is 2 ‘etic and that it generates the same topology 88 the (,) (Problem 108). In ation, the two notions of complete are the same (Problem 106):Proposition A net x,)isCauehy nthe metre pof(V.1)f and only iit 's Cauchy in each p,. Thus a metrabl loclly convex space X is complete asa metre space and ony iF iis complete asa locally conven space Definition A complete metrizable locally convex spice i called a riche spe, AAs complete metric spaces, Feéchel spaces obey the Bair categor the and thos one can prove theorems analogous to those found in Section ‘ThooremV.6 If X and Y are Fréchet spaces and fiX—+ Yi acon ‘inuous liner surjeton, then fis open Theorem V.7__ Let X and Y be Fréchet spaces; let F be a family of ‘oatigus linear maps fom Xt0 ¥sothat fr continuous seminorm p on Y and every x © X [(F(2)|F € #7 is bounded. Then, foreach p there acon tinuows seminorm d on X and a C'> 0 so that, HE) $ Cae) oral xe X and Fe Foran application of Theorem V.6, se Problem 12. As an application of ‘Theorem V.7, we rst notice thatthe corllry of Theorem ILD goes through without ange: Corollary IF X is a Fetchet space, 8 sparsely continous bilinear functional, Bis jointly continuous, that, [BU 9)| = Cov Yes) Tor some continous seminars ‘We can alto prov the following corollary of Theorem V.7 whose proof canbe viewed a8 Theorem L27 in dispute Theorem V8 Let X be a Fréchet space and lt fe X* be a sequence converging to fe X* in the o(X, X)}topalogy. Then fa-ef uniformly on compact subi of X. Proof Since f(s) is convergent, iis bounded, s0 we can find & continuous seminorm p on X so that || Cots). Given a compact subset D of X, and ane pick fine cover of D by sets Ui...» Uys0 tat x ye Uy implies Vs Functons ofrapd deerate and helper ditt 1 Als —)) SBC. Now pick mye Usand Nson> Nimplies [fx) ~ Stel < 3 for = ,....mm Then an 3 argument proves sp fie) FoOl | 9 90 that for Sige F, (DING) ~ 180), hat is, f DYNA) de = [/NSONG) de. AL fest sight tis appears ileal out iteration by pas imps hat (Dg (= AE) 9 we take S= (=D,Operation 2 Let Tes, ne. The weak derivative, D'T; oF the derivative Inthe sense of darts is etned by wng= (wT In symbolic notation: a Jormeaseas 7 0) fe ‘We have thus defined a notion of derivative which concen with the ‘ordinary derivative on Of, and for which integration by parts without boundary tems at co holds by fat Example 8 Let ‘Then iscominsous but snot everywhere diferetiable inthe clascal sense Since =. (e)9) de = Ip, ge. 40 ithab a derivative in." By defii- 79)t0) = ~o(2) = = = foe de ee ee mol BS (eho a) so dla = 5,8, 00 has a derivative itis described in Example 5. Coven This last example shows that even 2 nonfunction like 5 is the second | = Clu + 17. For (aj es deine ‘Bhla) = Feo yay. Then a, ys Ematan sef 0 oma se( Be rniea)" (Bere fatten “Thus B defies aco F aod sthereis aT F with T(z. 04.) = Se TUG.) = ,. The weak convergence of yd 10 Tv easy Wie can now easily prove many intresting theorems about 7 with this machinery which has to important simplieations: (1) Sequences are sir fo deal with than fonctions. 2) The two conditions in. flhof at co nd the C* condition, ee eeplaced by single fall-off condition in 8 Corollary 1 is dense in" in he of", ) topology. Pro, SayanbideeF and comers weakly to Tes” ay N10 if hari Corollary 2% is separable in the Fréchet topology. 1 is separable in the e( 7,7) topology (and also inthe 9", 7) topology We introduce in Section V7. Corollary 3 The regularity theorem for distributions Theorem V.10. Proof Agtin we only consider he cask = I Since [ln $ CW(L+2°Y"Ia wwe conclude that dvi C( + 1), using A and a? and the estimate in the proof of Lemma 3 (More detailed studies of the 4, show aula ~ Dove 1)-") Let Te" and et (6) be its Hermite coeficients. Then [del Ser for some m Leta, = (r+ 1)" hy. Then Soy lie = ES (2+)? < co, 20 S-a,9, converges uniformly to some continuous function F on RF bas Hermite coefiients (as an element of 7"), (a) Extend a! A and N= (ids? 42° =I) 10 2% Then Teorere raga ‘Thus Tean be writen asa sum of polynomials times weak derivatives of polynomially bounded continuous futons. Simple manipulations (Problem 537) now complete the proot. Corollary 6 (ouciar theorem) Every joinly continuous bilinear func ional BC) on SAR x RP) is ofthe form BU a) = TUFSe) fr some Tere", Proof, Since B is ointy continuous, | BUS) < Cl Iya for some re, SE15. Then [Blb,, 49) $ Cr+ IY + I= Clea, 6) + 1) where BD = Cy on aes Bad TE [As a result bp = Bigs, $y) ae the Hermit coeficens of « distribution eH") with Tian) = T. 44) =Beagy- Let f= Lends 9 Leydy Sine these expansions comergem 2 TUG 9) ~F 2664101044) =F, ocpbasy = BUD) VA Inductive lnite: generalized functions and weak aoltons of partldiereta, Jan inwitve sense, the distributions of he las ection had the restition of being polynomially bounded at infinity. We taw this in Theorem VIO hich od us any Te "isthe derivauve of polynomially bounced function ‘The growth ofa tempered distebution Te” isin some sens dual to te n)sthat Msconver, ‘slaned, and open in X, and Ny 6 Kyay = Nacy- LeU N= Use Na Te cary to se Nye and NX 6,. Thus 6, it neighborhood inthe ‘elatve topology. That the Flative topology is finer than the given topology follows fiom the definition of @- This proves (cand ths (3). (0) i eany. A reference for (is given inthe notes. B Definition The focally convex space X constructed in Theorem V.1S is called the strict indctve int ofthe spaces X ‘We remark that ifeach X, is «proper closed subspace of Nay. then X is not metrzable (Problem 48}. One of the nice properis of set ineucive Tie ‘Theorem VA6__LetX be thestrict inductive limit ofthe lolly convex spaces (Xn, Then a linear map T from X to a leally convex space Yis continuous if and only each ofthe restrictions TX, i continuous Proof Tis continuous ach restriction continuous. Conversely, suppose {20h restiton is continuous. LetN bea Balanced, conse, open st in ¥ ‘Then THIN] 0X, = (7 1Xy)"(N]isopenin since T |X, iscontnuous Since T~![N] hs Balanced and convex, ts open 30,7 continuous Example 1 Let x(@) be the continuows functions on ® which have compact support. Lets, be those Functions i (with suppor in [—. mh, normed with I'l. Topologize x with the inductive limit topology. By the last theorem, the dual of with this topology is pecs the complex Baire measures on R. This consiraction works [or a(X) when X is any esompact Toclly compact space [Now let be an open connected se in RY, C5(0) the infinitely ifereniable functions with compact support in AL Let K, be an increasing family of compact sets with JK:= 01 Put the Fréchet topology on C@(K,) generated by the IDI, norms. The set CE(Q) withthe inductive limit Topology ‘obtained by CFO) = JCE(K,) i denoted 3g, This topology is independent ofthe choice ofthe K, (Prober 46) Sequential convergence in tly Simple. t X= UX, has a svt inductive iit teach is closed proper subspace of X,qy. Then a sequen Je X converses (0 f6X if and only ial the fa aren some Xand ff inthe topology ofthat Xy- In particular, a sequence fe Se onvergss tof if and only ill the f.'anef have suppor inside some fed compact K and DY, conserges uniformly to D'/ foreach multiader a Proof Us =f and suppose oral that herein fy With fa Xy. Then ies easy to construct subsequences ofthe a, Say 3, = fy and the Ky Y= Nay mith ge ai Yipee Ys ose, we ca use the Hahn-Banach theorem fo inde X*s0that/, = O0n Yand 7(g.) =~ S321 490. Let C= Su, On any Xy, ths sum is effectively fine, so ¢ is continuous on tach Xa ene, by Theorem V.16, on X. Since gyf. and © X". (9) ‘Sonserges But ig,) =a this contadicion proves tha all the fy a2 12 sone fe We are now ready to deine the ditrbutions on 0: Definition _A generalized fenton (or distbuton i a contineous linear fonctonal on The space ofall continuous linear functionals on is denoted by 9. 9 and wil denote Jy and De respectively ‘Theorem V.16 rales dc int: Corollary A nea functional Fon Sis continuous if and only ifor cach compact K'c BY, there is constant Cand ileer js hat IT@1 SCE 1a for all = C5(K). Example 2 Let fhe an arbitrary continuous function on Rand define Dies by (Dre) = (= 1" FFD 0N) de “Then foreach compact set K and gC” K) NOP} Cid*ee sup IF] fo Dife Die, Thus Si contains the weak derivatives ofall continuous Fuetons. Example Consider 9. Lat tx — Then Sie StH — IT) 2 Coy Sx ~ aK N= DYE). Tis in 9. For lt g¢ CS(—m, ml). Then |Scor| F101, “hist example shows that 2 consis dsibtions T which re not the ath dervatine of any continvous function. Ths. there is no ect Analogue of Theorem V0 for 2 but there aloe reguaty theorem (Problem 24). Teresa 4 milear theorem for @ Ge Problems 59, 6) ‘One ean ako carry operations from 10. by the method td in Section V3. Thos for exarpie if poe) 82 Polynomial of eal epee Kin n vatble, pena) Dyesae™ te paral deena operator D) = nan. Dexter to by te ormala womne=7[ 3-00] Formula (V-5) may aso be used i hea, are x-dependent C= functions. “The extension of partial differential operators to is particularly weful in the theory of partial diferenal equations. Let bea continua function, ‘A kets continuously difrentable function v (ve wete we C') fr whieb Dw =f i called a sit solation. Te 9" and plDIT =f with pO) Gefined by (V5), then Tis called a weak Slaton of the paral difleential ‘auation. Thediferencebetneen strict and weak solutions is only smathnes, for: Proposition Iu, then p(D defined by (V5) sequal to the classical value of p(D. I particular, fue C*and fiscontinuous, usa weak solution of p(Dyu if and only fii strict solution Proof An elementary integration by pacts. The following example shows that not every weak solution isa strict sclution: Example Let /(x) be the characteristic function of [0 1]. We will show that ule 1) =/Ge~ ef) a Weak solution of ty ~ Puy, =. Rather than ase ‘he definition (V5) dicey (which sa useful execs), me make vse of the fact that the operator p(D) in (V8) s continuous on 2" Since fe L(), we can find a Sequence fn Dy with fm» in 1B). His then easy to se that Sife, =e) ale, 1) in (29). But p(D)e, can be computed las, that = . a Fuoeena-o ZunasIn Problem 47, we dicus the notion of «distribution being only a function ‘of et and prove that any sich dsibuton Tastes, Foe oat gat “The concept of weak solution is particularly wseful because i is often easy to prove weak Solutions exist ce Section IX.5 for the constant coefficient case) In the case of eliptic equations (ie Section IX.6), one can prove & Feqularity theorem which asures that under certain condition, every weak solution i strict solution. Combining these two techniques one concludes the existence of strong solutions for eliptic parts! dierent! equations Because ofthe phenomenon of the last example, the case of hyperbolic ‘equations is ot So eas. VS. Fixed point theorems? Wie wish to consider the solation of equations of the form x= Ts in a variety of types of applications. For example, an inhomogeneous ince equation /2) = 918) +{Klx.y)/)dy is of the form f= TF with T/=9+ Kf, an afine linear map. The famous "bootstrap equations proposed in particle physics are of the forms S= (5) where Sis the ‘Smatricand Tis very complicated operator. A Lorentz invariance condition fn vacuum expectation vats tS the form Ws 989) = WA og Ax) here Aisa fed Lorent transformation ‘We want to discuss a variety of existence theorems for such equations —so-alled xed point theorems-and in Section V6 we dicurs some applications. We study them here because several of them are sated quite ‘atually inthe language of locally conver spaces, We fist consider non linea theorems, that i theorems that make no assumptions onthe linearity of the map T involved, and then one simple theorem which employs Tineary. Definition Let 7: x-+X bea map on set X.A point x6 X for which Te = is alled Bxed pot of T “The fist nonlinear theorem is very simple and is probably familia to the reader VS iced plat thors Definition Let (S,p) be a metric space. A map T: S$ for which (Tx, Ty) = pbx, 3) called 2 contraction. If there sa K' 0. <1 sos SI, oF equivalently 02 #< 00, 12120, As we shall se in Section X17, for lage variety of potenti the amplitede Ayn 0 the boundary value ofa function analytic in the cut s plane (gore V.3). In plane asneral there may be poles along the negative ax which we ignore for diols Thon done image ric) ae 0) fr some analytic function in (larg s #0. Moreover, A(z) is el for <0 so the Schwa rection principle mphs thatthe physct! amplitude obeys Im A+ 10,0) = 5 La + 0,0) ~ 406 ~ 10,09) u Suppose, agnin for simplicity, that A(s,0)—0 at infinity. In general, his ‘sfalse and one must mocity the argument below by what isknown as making subtractions (se Section X17) Then, by the Cauchy integral theorem, 1g ano) M9. 0)- ef Mw here Cis the contour in Figure VA we now make the le cic bigs 4 bier. tll make no contbaton sce weasel 0), 5° by shrinking the aight seston ote ea ane Fd (Aeneas here DAs ty = 0) = (1214s + 1,0) ~ ACE = 10,0)]= Im ACs + 10,0) is the" econ” of “This "dispersion elation” (V.78) canbe proven to hold with ¢= fy for all eal positive fora small lass of pote. This cas includes suns of ‘Yukawa potentials which are theught to be the nonrelativistic analogues of nuclear forces. Moreover, in that case, fr Jp fied, Dy 1) the boundary value ofan analyte function in a plane cut from #= os) to so. a) a A190) 79 function depending on the potential whichis explicily known and is called the Mandeltam boundary, We can write a dispersion relation for D ee t= QING) = 4A} ld esl Tara ie puting our two dispersion relations together, we obtain the “ Mandelstam somnenon — comb fae fh areA wm anne shel erent ‘Tht aon weeny sen ft ly oper Tie ant cst of toot he nary smanat ta seen eais ie rata an ore ME fore ast wae ct Une nn een Siena mny ria sels eer reponse Pata mnt puis ee ot et ee ne tom tr eso «Sperone en 9) ‘urn msn, eee be emi frien bam an lente ice tated ne tae ered ‘nines opcode et eed Potten peri | To a e oe eon Im Ao) ¢ AE. 0 where (6) sa fonction of E dependent on normalization of 4 and dts the angular measure onthe sphere. One can extend this ation to nonzero Band thus obtain a nonlinear integral elation betwen DX, and A( #) (Quadratic in), By taking the discontinuity of D and using the Mandelstam representa ton for 4 one fins the elation pa To) where Tis an explicit but complicated function ofp. fp obeys p = Tip)and has the proper decrease to make certain inegrals converge, then one ean show that defined by (V7) obeys unitary. Ths, the existence of (1) ‘withthe ight analy and unitary properties equivalent othe extence ‘offied point of 7. Of coun nthe nonelatstc case ope knows that such Dist since one can show that the seatering amplitude fora superposition fof Yukawa potesals has the right analy and unitarity properties, In the rate case, for example, 2°” scattering, there are two addi ional complizations:©) Crosing symmetry. Replace the momenta piBs in Figute Vil by the energy-momentum fourvetors p,=(/s? + pp) where is the mass ofthe pion. Given a four-ector a (a8), we define @! a a; Tor cxample, p= 8, an expresion of the relativistic. energy-momentum ‘elation Now one defies the Mandelsam variables, #=(py +93)" = 4p" +12) Af p is the center of mass momentum, t= (pp)? 46 ~4p2¥6050—1) and = (n= pa? = HE —44(—c080~ 1). OF ‘course, fad ware not independent fore} 1 += 4 Croming ayy expresses deep fact of relativistic quantum theory, namely thatthe analy ‘ealy continued amplitide for say 22" 2°». symmette function of Sand t, anda symumetc function of sand u when the change of variables (5) (5 is made by using w= 4y2 —2~ 1, This automaticaly implies auditonal branch cus in the domain of the function Ae, ). For example, Analogous to the cut in the domain ofthe nontelatvstic atmplte, running ftom £0 to E =o, is 2 eut in 4(,0) ronning from += 4° to + 2 Crossing symmetry imps tha thee must ako be cot running from 342 tou= c, or equvalealy, since 1~ 0, from #0 to =e. The analogue of nonrelativistic Mandelsam analytic then expresed by the ‘elaine Mandeltam relation Miootd = 3 fo de dt 1 1 ato 7 aaa) 1 this formula and p must obey of) ls). Te ast two terms ae just issu) plea) ae ew afer change of variable, This Mandestam relation is just the expression of ‘rossing symmetry plus certain analytity properties (W) Inelastic processes eis characteris of relavsic systems that if ‘heres sicient energy available, large numbers of particles can be produced, For example if s>16j%, the reaction 2° 3°-+2" 4° 4R" 4x is VA Aoplictions of teed pant heoreme 158 posible. Unitarity comes from a connection between the interference of the later and unstated waves and th total amouat scattered in all prov sees, It thus gives elation between an A(,0) and a sum of terms, one of which comes from x°-+ + 2°-4 3°. Even when 4y? << 16", an in- laste process n" + n° vn +2" is posible. A complete treatment would equte consideration of ll the w+ m= + x amplitudes. For simplicity we Consors model when ther are no =* and #~. Unitary i thus a nonlinear fauality on 4 only when dy? <5 16? “The “bootstrap hypothesis of Chew and Mandelstam isthe philosophical ea that there sony one set of amplitudes for al processes withthe" waa” Analytety properties and which obey all the unstarity equations (coupling ‘arlous pres) In practice, one approximates the equations, for example by replacing the coup unitarity equations with inequalities when s = 161° {5 we have done above, Whether one accepts the bootstrap philosophy or not, the various bootstrap equations are of intrest since they can be Viewed [sanexpresson of the constrains paced by unitary, crossing symmetry, and {nalycty onthe ampltode. Even if these do not determine the amplitude (and we do not subscribe tothe booltrap philosophy), they do put severe festctions on the amplitude fact, ts not ea, a pri that an functions ‘AG exist which obey the requisite analyety and crossing symmetry, ‘dase unitarity 4y-z <1, and the unitarity inequalities for 16y" « 8 "The existence of such functions has been established by Atkinson in & bei application ofthe Leray-Schauder-Tychonoff theorem. The basic idea of the proof the flowing. One seeks afontion 1) fo put into the Mandestam repreenation. If 4 obeyed elastic unitarity everywhere, one would have Pls. = (TNs. where T* is 8 complisted nonlinear map. Sine elastic unitarity is only ‘beyed in certain regions, one has this equality only in certain regions of the St pane. In general lt) = (T*9NS 1) ef) where secing = 0 in rian regions is equivalent to laste unitary. Ifo obeys certain other condition, any solution of p= T"'p +» obeying certain integrabiliycondi- tions yields an 4 obeying elastic uttarityn ys ss 16" andthe inelastic ‘unitarity inequalities in + 162. Thus, the existence of solutions of these “approximate boottrep eqstions would follow ip = Tp had a solution with ‘Tp Tip 41. What Atkinson doesisto construct a comvenseS,, depending ‘on of uniformly bounded equicontinuous functions which seompact nthe Iie topology and so that 7: $= Sand is continuous The Leray-Schavder~ "Tychonoff theorem then provides the existence of soluons of ofthe approx- mate bootstrap equations.0) Vs LOCALLY CONVEX SPACES . Determining the phase of the scattering amplitude ‘According to quantum scatering theory (ae Chapter XII), the difer- cia” Satteing eros section at fixed energy is given by a function DO= IFO! where F(0) isa complexvalved function ofthe scattering angle 0. At eneries her theres only elastic scattering, F must obey the nonlinear "untaity relation” In F()= [FDC sind, a, doy tee 0 te anton of 0 and determine by he per aoe in Figue Vs. = cos0, is gen interme off =cO8 8, == cn and ay Speen (1 = 2 2) cos gy ous VA The ange Inexperiments one measures D() while (0) is of gest theoretical interest. ‘There are two questions one immediately wants to atk: (1) Does the unitarity relation place any restriction on the possible funetions D(@) which come from functions # obeying the unitarity relation, 2) Given D(Q).isF determined by the condition |F(@)| = [D()|* and the unitary condition. The reader Should realize by now that these questions are really the existence and uniqueness aspects of a singe question Introduce the variables z, cos. Let Ky, 24:3) be the Jacobian ofthe transformation from topologies. A lea convex topology 70m Xs called duhis “The bate duity theorem which ientifes all dual topologies sys The focal ram V.22. (Mackey-Aens theorem) Let (X,Y be a dual pair. A convex topology 7 on X isa dualX, ¥) topology if and only if a(t, EF SK, Y) Proof For the geometric proof, se the appendix to this section, ‘Thus the dual topologies are precisely those between the weak topology and the Mackey topology (incase) ‘We have seen how to recover the norm topology on a Banach space X in terms ofthe dual pair CX, X"). How about the norm topology on X°2 It isnot the (X*, X) topology unless X is eetexive, for th «(X*,X) duals X. ‘Gay, the norm topology on X* isthe topology of uniform convergence on the unt ballin X and oe ned ocly convex notion that singles out sts contained inside ball. This ned is met by te notion of bounded set Before ‘efining this notion, we noe ‘Theorem V.23 Let Ebea locally convex space with dual F. Tae folowing are equivalent fora set 4 = E (@) Forany neighborhood, U, of Oe E, nV (nx|x U} for some (6) The poar of 4,4” (which we define in the pendix to this section, is absorbing {@). For any continuous smiorm p on E.SuPa<4PC2)< (@) For any 2 Fe subse, [| <2 Proof Tae equivalence of (a) and (6) and of (b) and (f) are esietally raters of definition. Theorem V4 sys that () implies (8). So suppose (8) holds and a continuous seminorm pi given, Let Ky ~ (x © Els) =O} and let E be the vector space E/K,. Then p "its" to 4 norm on E,. Let x be the Canonical map of E--Ey and let 4,= {4}. It i easy t0 see that V2 Topologies om oe convex spacer 18 P48) < 0 if and only i upse4y A) < 0, Let E, be the completion OfE,E, isa Banach space LatZ 6 (E,)" Then C+ 6 B30 89 G01 = sup [ee mk <2 ‘Thus by the Banach-Steinhaus principle, up, ls) <2. Definition A set A E,a locally convex spac, i called bounded if and ‘only if one, and hence ll, ofthe conditions (@)-(8) of Theorem V.23 old Condition (4) makes it clea that the notion of boundedness the same in \E, F)-dval topologies ad thus a notion astcinted most naturally with 4 duality on E rather than single topology on Example 3 If X isa Banach space, 4 X is bounded if and only if Sup sea [et < 2, which holds if and ony if 4s contained in some multiple ‘of th unit ball. Thus for any bounded linear map on X. sup ITs < C4IT Example 4 Let X be a striet inductive limit of spaces Xy with X, 2 proper closed subspace of Nyq3- If i6a sequence with 2 ¢ X, by using the onstruction inthe proof of Theorem V.IT, we can find linear functional, eX" with sup 3.) = co. Thus any bounded wt A © X must actualy bes bounded sub of some X,. Ths or example, A = is bounded if and only if () there i a compact Kc so that suppfe K when feds (i) euP sea IDF © 2 for any 265 ‘The first example tls us how to generalize the norm topology on X" Definition Let be lsly convex space. Let Fbe it dul The strong toplogy,f(F, Eon Fis the topology of uniform convergence om Bounded nubs of tat ithe topeogy generated by te semioras (plac Es bound) where p)) = sopra 7D) ‘Any ofE, Fcompact set C of B is bounded because the fe Fare con- tinuous functions on C. Thus the strony topology ACFE) is stonge than the ‘Mackey iopology ‘We have found a topology UF, E) only dependent onthe dual pair so that the norm topology on X* i the BX, X) topology when X ik @Banach space. Given E, locally convex space, we ca form its dual B* and ut the ACE, ) topology on it. The dual oF E* in this topology is aed the ouble dal oF E and is dented by E** when itis given the ALE*, E*) topo logy. One can map E into E** by the standard duality map, p: B+ E* by (3X2) = E(x). This map 18 not always continuous (Problem 34) 1 itis a Tepolgicalsomorphism, we Say Eis reflexive, that Eis elev i (i) The AE, £) dul of E* iE. (i) The {E,E*) topology on E isthe given topology “The fllowing eiterion for telenivity is oRen wef Lemma Let tea locally convex space, Then Bis esve iP and only if al ofthe following old (a) Every o(8, Blosed, hounded set of Ei (6, (©) Every o(E*,B)closed, bounded set of E* i oE*, ) compact (©) Eas the Mackey (thats, (E,E*) topology. Proof It's not hard to se (Problem $5) that () holds if and only if the ‘UE, E) and t,E) topologies are identical and () se the BlE,E*) and 1(E,£) topologies are the same. Now, let E te reflexive. Sine E= E*, the BLE, E) topology is dual topoogy, 20 ACE, £) = HE, E) bythe Mackey ‘Arens theorem, Since, * © B, BLE", E)» (E*.E) Silay, ince the dsl (OF Eis Band Ebas the UE, E*) topology, BlE, E*) = «(EE and Eas the Mackey topolory. Convery, let (a6) hold. By (a), E= E** as vector spaces since #(E*,E) = x(E, in that ease, and the Mackey topology i al topology by Theorem V2. By (b) and), E has the BCE E*) topology. ‘Thus Ee reexve. B Using this emma, one can prove (Problems 56,57) tha: ‘Theorem V.24_— The spaces HR"), 9p. and 6p are reflexive. ‘In genera, the Mackey topology ona spe is much stronger than the weak topology, s0 ts mich harder fora se fo be Mackey compact then weakly compact. For example, te unit bal in an infinite dimensional Banach space {sever norm compact (Problem 4) bu ite weakly compact X selene “Thus general theorems on compactness in the Mackey topology ate pt clay strong statements. A mos seul one is: Theorem V.25 In Y1R), 9g, and Bp, any closed bounded st is com: psc in the sual Fréchet topology) ‘Avoendix V3 Pola andthe Machay-Arens theorem 167 Proof LC € 9(R) be closed and bounded. Since sup jc lle = E <2, 1/G)~f)| $ Elx— yl whenever feC. As a rel C is @ uniformly equicontinuous family of uniformly bounded functions. Similarly, wore) re uniformly equicontinuous and uniformly bounded. By the Ascoli theorem tnd a diagonalization trick, any sequence in Chas a convergent subsequence. Since C is closed andthe topology is met, tis proves that C is compact ‘The proofs for MUR" and 0, ae similar. For Ce dq, we note that since C ‘bounded, C= C*(K) fr some K and then ute the iment ‘A particulary useful consequence ofthis lst theorem and Theorem V.8 is Theorem V.28 A sequence in $7, 9° oF i converges in the weak topology if and only if converge in the strong topology. Proof For # and 6 tis follows diretly from Theorems V8 and Theorem 'V.25. In the cae of 9, we notice that any bounded set C'< @ is in C5(K), ‘which is 2 Fréchet space and then apply Theorem V8 and Theorem V2. 1B Even though Theorem V.26 js esenialy a corollary to Theorem V.25, we single tout as a theorem because i very seul in applications, We caution the reader to hed the word sequence. The theorem does not hold if "net eee [Appendix toV.7.Polare and the Mackey-Arens theorem In this technical appendix, we prove the Mackey-Arens theorem by introducing the machinery of pola es Definition Let CE, F) be a dual pit. Let A1= E, The pola of 4, 4° i er I [fe] <1 Vee 4). we want F tobe explicit, we write (A Examples (1) Let bea Hilbert space in duality with ise IF isa subspace, 4° = 44. (@) IF Bis Banach space and Fits dul, (al Bele SA)" = 091 We sk) is easy to prove the following simple propertiss of polar stsLemma t Let (E,F) bea dual pair Then (2) is conver, balanced, and oF, ) closed () ae B then Baa (©) W280, GaP = [4 @) (Uead= Neat ola set are simply elated to dass Theorem V.27__Let te oly comer ste and a eghborhood base t 0, Coser te dal ur CE), where Ey. the algebra dul thee ofall nent maps of into Cen the oflog eso Ei oe 0" where the ples are then ate to Proof €Btqiscontinuousif and onl if ¢(2| <1 forall xinsome Ue, ies ifand only if? © U"forsome Ue. ‘Thoorem V.28 (ihe bipolar theorem) Let E and F bea dual pir. Thea sing the o(E, FMopology on E, we have e-uh® where ach(E), the absolutely convex hull of E, isthe smallest balanced ‘conven se containing E, thai, wena, and the closure isin the o(E,F) topology. Proof Lat Ee=3Eh{E). Clearly EE" and sine (EY is conven lance, and oE, F-losed, Ee < (E.On the other handy i x4 Ee, we ‘an find ¢eF with Ree) © I for ex Ee and Re (x) > | (Thecrem V4) Since Ee is balanced, siPyere [M0] SI $0 CEE” But then |¢(9)| > 1 implies S62" Lemma 2 The Mackey topology i dul topology. Proof Weuse Theorem V.27 to compute the (E, F)-Sul of. Te (pe 88 C ans through all o(F, E}-compact, absolutely conven ses of F generate the HE, Friopology, Consider C= Fc ES, Since the retivtion of the (Eby, BD topology to Fis (FE), Cis et, E-compact and 80 oy E)- closed in Ey. Ths, by Theorem V28, (Cig, = C. But C*= (ree = Theses (C*IC ina comer, balanced o(F,E-compact subset of F) thus form a neighborhood base at 06 E for EF. Therefore, Be Yr, = Om ‘Lemma 3 (the Bourbaki-Aluoglu theorem) Let Uc E be a balanced, convex neighborhood of 0 in some CE, F> dual topology. Then Uj is a ‘oF, Epcompact set in F Proof This is exentlly » restatement of the Banach-Alaogly theorem (Theorem IV.21}; see Problem 38. Lemma 4 Every dul topology is weaker than the Mackey topology ‘Proof Let bea semiorm on Fin some given dul topology. We wil show that pe for some o(F E)compact, comer sie, C, in FL = (e[ oon th Then Ui Blanes, cones and e(, Fos by an ppeation of Theorem VA we Problem 22). Thus,(U")'~ U by thedouble Polar herem. Let C= U" F. By Lemma 3, Cis fF, compact and is omer By definition (UY = (elect <1) = Us 50,e=0. W ‘We are now ready for Proof of Theorem ¥.22. Sige the o(E,F) and (EF) topologies are dual topologies (Lemma 2 and Theorem 1V.20) any Jin betwen is also a dal topology. By denition, o(E, F) isthe weakest posible dual topology and by Lema 4, x6, F) i the srongest possible dal topology. Sect V1 For nr tenes on oly cove sant st: Chom book (ecto Seton Ve ele 1 Nel, Lr Topol Soe, a Nostra eit, Prmstn: Now Kay, 1983. RStbe Topo Yor Spe, Sia Watue Be and Now York, 1% Aad W. Rotem, Tp! Pier Spaces, (Come Une Pres, London an Now Yar, 164 Steer Topol eco ‘Sharer, Mail, New York, 1966.‘Ths Roberto tok a del yae monograph, ad fhe reining books ‘in ine ranttion of Kes Geo cite as este ote ie robles ie Kee Remota bat ha ny, ut ot aap Kel nso “Theis formulation of the Ha-Ranach hoe in tems of eating cone 8 isin Manu, "Uter toner Menges inten nomenon Rau Sa an (3, 84 Tae more modes form (Theorem V)'s ew Mai 2ae Pan ‘kr kaovten Mergen in ie noroner Rau" Sa Mar € (990, 1 {and Katuun "En Bene Sr Soe on M. set er omeve Menge" Po Toclycomex spas pole vter pain we he Hahah hore ls nd the pace ih ae tpcopel dab The sce O ‘(tc seem acta a ers semen ea (© Let inns Be fay famines 0a sme Be a = pi (4) Prove tha ay lly omen opto on Rte wl ope int Ue Uheesuan ofllnion andthe olowngcomtrcton Pekan pup Dard tt Y= alpad~ 0); am Ys 1- WY, 9) nok ee aod Posen) #0. tat Fe lis pe) 0) dns Sm dete (o) Lethe ey conten sce Show tht ay sa foal ile ‘mendoal beac of Yara coninnon exes ao (4) Prove tay Sie dmencnlalapce oe aay ones pee sed 4. Teepuposeof exert prove at vey oly opt ay cone inf mens (@) et U ben comrct eghborood of. Show toe ca 4 hat Ue Urea Ua tha ate dmeinal ses Mh UM 0 (Powe hal Ue dee GP fray, © Prowinn =A, (@ Coming tat = x= 44 5 Let nea Mach pace, 19 df ula doe al. La eth i ball (of and "he wt alia Prove thatTH Vr LOCALLY CONVEX sPaces (0) ote 2%) deme a 2TH: Ue eho Ve) on A the clare ieee) (©) Fis ce and ony te an tall oC, X*-compe. 16 Prove the tee srepstion te being of etn V. 17 Move athe fs (7 a oo areal (et Ue the Cay ‘a ome eps oe ahs po) 1H Lat Cheam stare nt wih ne Cf x6 Cand 057]. Lp te Ninna incest Pv at “ (one wren ) aD) 4) Kanda ral €Cando-st<1, Abe + WE {© fn alton ol fem alee {3 fender Ce ges 19. Prove Te V2. 110.) Compe he rat of Theron V5 (©) Prove the propel wing Toe V5 1, Lean be abctig set whe ope Yloalyconvesseaces with Tet ad continous. Let Tt te iat ) T ise and ony Ra =. (0) Ran T= Xa the 1) oe, 2 apo and oly Tine, (0) Late > beth natural map of 7 to 7 then icons when 6 en theo, 7 teployand gen the 7, pao (@ Powe! 1) Conte 2) deme ia the 4", Pande of, oon. 2 Let Xe zal conen pce, with dil 2 (6) Prove that any cee ip nok) cee (©) Prove ht all dal, 79 tooled ues. {9 mebeu clad comen tet of Prove tat Co 4° cre (Ho: Use ‘te veprtion hese)(Prove hat ak > topos ase theta olen ee 121, Pov ety a 8 (Exe Scion V3) in Prove tat der a come 12.) Provetat natal) ioe on po tn tS Saas en ten tb anesneos SSDS Sy arse ct oeeaaee amend (Prove foul. 121.) Let FeO, ove hat ound map of #2 {8 Lec het enna stn ta fe fo. Pv h Fis C™ 8 12 Gunes pow bt Pee 26 Lee Fe) lt TW) SCE ane ieee Map F = C1IB: “QCA (name f ff Fare CAR) he Banach a fanaa Foren fh ip fin fora henna ifr Bane hee (Ue the Ha Banch and Ri-Marho hin 0 prove hat Pena be wen ef fora whe rt cmp af oe oth 2.6) Latest une mn =f 00-6 PE de ‘rove hat = Gin tea fa tutone (Prove the reptiyerem for J) ung Polen 24 an 250). $e at IF FAD feral fe AB) wi pp ee 27. Let tetany san operate on (Bed be erative pean on 7 roe hat (Us heaves pole nthe oop die 125.0) Rove tat YA dete by Operation Seon Veer wth (Vp) = Pointe” wrepnied san cert of 7 (a) Show tat he sappae fs focton ge 7 a Stone Heme 1 age. Probleme 7 10) Ls 2 0-0 Hr at recite OF ei tat tel “pe cOler a sel (0 Large 1 ih Fp Orr a1 wh 8 mPa ha here (pd wit Desir ptoraltso tat WDM pic sOlecahee Ts td Frnt iam co tare ison mnie g won. Sipe ta po = Oa pe wth D>) ~0fo [8 Poet 79 Le Ferm ob) ITOLSeF weor te {et pp T= 0) Find eontans (Cy that TH)= Ercan) fora Yc Min Pk 9, oth eer and et 8-9, 2000, (© Prove Theor Vi 2%. Let Pe Gulf, 76 7 UR). Laden the divave on By mbipting the kon ot munipesion snd" pave PPV 2 PF 3A mao $9 is ell tol 2899 $9 where spp p=. A map S29 jeals lal app STA whens agp Poa. Go tar 9° telo Ponta SS oe (©) Which oe operation ot cal? "B.A dutbon TY ad obe ford at mes nit Ire s¢,F bromo forse Cad i ao te fod tt” If for ome fe a foray D>O, tone ace ad C9 ta IT 2,E wor—+ CF weotpt ‘esay Tis of orem odes ott ot ve at ot. We sy Tisot eden fol rer at mon tte ere est (4) Prove tha he eerste) of Example te fr (8) Prove that any oer enrmataton oi i) hs er ote han "2, Prove that ot ety fm om LARD (p<) of te frm PU, Srivtewonad for sane sy)1 1 Win 92. We Fg) =f cn faa de for some GL i+ p= 20 phd ark 1) Te <9 exam se in te hit eis Your's 2)The na them ue fo La eery azar fermon LD he fermUzo) = Fis )feWo) 6 fortome Fe E-G) an eedingereo oes form he Dunford Pethecrem whch yy Let te spre Beach fonce,T-E-*L") Then tees mama unction gon Shera Eo {fa aaa lh ITE and TONS) = ee, aes. Fors pot of te interes oven ee Tvs et Seon V3) pp. 68-7, Prove that seperately continous multinearrm n F6-% Fy iy on tncos ia he Fee Petet pens nnd the proof of he ae hore nthe appedis f V3 1 mus a ‘ona ety Bf fa sept emus fen Tuono Fe So ge heels Peete) te Mi TH 92M. Provide he deh fhe root of Lemna in be spn to Seton V3. Compete Poof Crary 3 nthe spend to Son V3. Dee (R= it C* fant on R whl rai a LR), Pa Wesenies en on 28, whee fee De Prove nt 3,0 compe 18) Powe nat CF not om a ld ore, Le = U4 C* atin om RY For any ep and JE, et Mee up OK) erm 89 with (Prove a cone (8 Prove hat heat ton © soins 0 4° ina aaa Hamels (Prove that Tein an only IT como ope Pro pat) of There V5 een Lenn ofthe append Scion trove ta a a teeter te Toran Sued sp if Ulgne= Dee ste) Rove ta and omer hat i here continuous et ion (2) Prove that AR) and) re psopaly Homeric. 4. Prove tha he some 84 (gy 08 Fat aie ay = Ett al 48 et W be at edie tof % with ech proper cloned sinc of ‘typeset {) wa ceumabe deen fry twee 0. Pick meu Gwe hag not bounded (0) how hat el mould nde i were aiherhod ae {9 Conte Xi ot mee. “4. a) Suppone Xi the srtindotie it of pace Suppo (Ya sensing fay faucet otha for any nr ian Hh Xe No Prove tat ‘nthe ndtv ae (@) EXKSA = Reh eompactend open Prove taf asa teplow shen ‘sre fay hen he ttn is topeog to CEC he Me Ben ike DY norms (Pro tha he oplog 9 ndeendent 0 te ele of he nase 407 tet deb corte on We ya iain Pe 97 sided pee Sebel tn ofan endo fray tanaon Casto Die Gre de Rebate AD Letrtea nebo on wh ey Let Fee ton (© “Socined hte fncion Pipi.) 09 Prove that Ti lnependen (ES ene (fi: Looks Poin 2. (o Ler en detruton n AR which tant of. Prove ti tinder at nape cori taken fr (@ rT Prove tat (0) = iy and at Oa tion wh ony a fence of (Conese nat so (4, Let Tad 5 te sormmaig ayo meal Let f= T= Gioia seep sefe (@ Sion that 7a st Conacton Prove aS tas a ed oe, (© Lerten contacto frame Prove hat Ths wie ed po. 18, Let X= (ig efi 11% (6) Prove at a canoe cone sot of OS EO Yb ob Et. Pvt nami aft (e Prove at Fasc contigo extension tal 2,1 Suppose hat i pup with am ablan styroup 220 ht GIN i abla (or ‘ample te any of ron sad anuon of) Let be compet cane Se aia cern f Fore ean ato mel (a) Let Crm ne CTex~ forall n= M7) rove that Cy i compact, covwe, and rene. (ey Seppe ono rein he ame contin GIN. Poe ta TC (@ Prow mat ihecls woo restora gee Ge ‘rove Theor V2 dey ‘lly convene 2 ca 4 Make pee es broil mai a8) iy eaen sce Yn) nee map TY wih oes teed st in (GLa bea Mackey pce: rove thatthe he Macey opel, Comes t=) (2) Leviys0 ins meta lay cones spe, Poe ta he fp €R with grec ao tev Mit: Let Us tea comae neghtorord by tin 2c, Pakmson =m impli eles Pet am tem ee mer] \o) Poveda every mia comesspac na May ace in: U8) indie hae» emergence ound) (a) Prove at steetnce in X= rh ya rope oe aap of svat iene Macey. (oy Pte at tt nbc of ksh pss is a Macey pace 11 Conte at he ator opto on 7 nd te he ce ope Let Ete ely conve pce Define the maa topes (Eon E* 2 fegus Le te tefamiy of tanod cones neorodeaf0<& Fore le ete Ev pour of U" £* The Opener tr pay. rove (3) "Th tral osogy weaker ten he A, ) ply, a” Boy nude) (The econ of he patra topo ono ein onsen on Tp = cE, > conan tnd ope (eT nar of hema necton pn CE, > om Ran p to i abenys Let Ete = anak space wt he wa tpg Prove the inion p= LEP bea ul ar, rove ht vy of Phd bonded set of Ei AE, Fe ‘pat fad ony the, E208 BP 8) psp on Fase sea (a) Lee be eth pce. rove ht ay 8 coed unde at of Ei {E*.Eheanct (in Minch Banc Ano heen, ovking ep oy Shetty A Blt (asl eR, Let pede {eL—eh bt 05m (Gr) Prove tat fo} 0 CE ‘to: Ofer allie»). Devel orm gs Ticaomorpie Prove Wt ke dane of CP expeinla): #01 Rs Gen ian tif (9) Let Bnei fptonl on C5 Provethat eit PeCHt(w al = (ade win eg) = TU pty mache Ingthe apps to Saion Vand wing Pete 9. (Lat (bea fonction! nq, wih seaely copious, Prove a tee a Pe Bhs wih Ba) = TSO) 4, Prove tht CER) enemy, that coms an expt funtion of compact fopor which i infnic ofoentable Hint! Fit show tat the fenton Toe nino thc were nthe carters encase 0, =).Bounded Operators 1 war the mathematical sh where ema ingt Ni ppl fer «method ee ‘inginaeow Eurp. Te proprio an ramration ere fr steno a ‘jer nth sonpsed of sph ten. Ta he ten we vl tpn 8 {ning toma an forthe fing et mong trend and ater A the afer ‘ocr inc mad ote btn, erg te ropes wit i. “ota Sift Gallo rel VIA Topoteg We have already introduced 2X, ), the Banach space of operators frm fone Banach space to another. In this chapter we will study UX, Y)more ‘osly. We emphasize the cate which wilarise mort frequently Inter namely, ZF, 2) OF) where 2 isa separable Hilbert space. Theorem IL Shows that 2(1, Y) is Banach spec with the norm Ws rt 2 “The induced topology on 4%, ¥) is alle the uniform operator topology (oF ‘norm topology). In this topology the map (4, B= BA of 20%, Y) % £(Y,2)-» £4, Z) soil cominuous. ‘We naw introduce two new topologies on UX, 7), the weak and strong ‘operator topologies There are ater interesting and useful topologies on “2(X, Y), but we dla thet introduction anil wenced them ina later volume {sce however the discussion atthe end of Section 6 andthe Note) "The strong operator topology isthe weakest topology on (x, ¥) such thatthe maps Fogo Vs Topologies on bounded epmatre 185 siven by £,(7) = Tx are continuovs forall ¢ X:A neighborhood bass atthe rgin isgven by sets ofthe form {S]SeLO6M, ISeily ) and only if [7x — Txl-+0 fr all xe X, The map (4, BD» AB is epa- ely but not jointly continuous if, Y, and Z ae ininite dimensional se Problem 6a, b). We sometimes dente song limits by the symbol sim. "The weak operator topology on 2X, Y)is the weakest topology such that the mape ELE NC sven by , 47) = (Ts) at all continuous forall ze X,/€ Y*.A basisat the ‘vain is given by sts ofthe form (S|Se2OM), le¢Tapl T) i and only if (Tex) ~¢(73)| +0 for each Ze Yeand x6 X. Notice that in the ease £(2F), TT weakly just means thatthe" matrix lements” (), T,») converge f(y, Tn the weak topology the map (4, B)~ AB is separalely, but not jointly continuous if X, ¥, and Z ate infaite dimensional (see Problem 6). Remark The reader should not confue the weak operator topology on ‘Cx, ¥) withthe weak (Banach space) topology on (x, ¥). The former it the weakest opology such thatthe bounded linear functionals on 20, Yo {he form (2) are continuous forall xe X and ¢« Y*. The latter is the ‘weakest topology sich that ol bounded linear funetonals on CY, ¥) ate Continuous (se Section VI} [Notice thatthe weak opeator topology is weaker than the strong operator topology whichis weaker than the uniform operator topology. Tn general, the wenk and strong operator topologies on 2X, Ywll ot be fst countable fo that questions of compactness, net convergence, and sequential comer sence are complicated, The following simple example usrates the dierent topologies on 12)ample Costumed opto () 7b oy bot (Usk gu) Then T,~+0 uniformly Lets, be defined 89 SG ban I= 0,200 Sant bngs0 9) Then $,-+0 strongly but not uniformly (i) Let Wy be defined by Waban) ‘Then W,0 in the weak operator topology but notin the sions ‘or uniform topologies ‘The folowingresltinhe Hilbertspacecaseissometimes sefuland provides 8 nice application of the uniform boundedness theorem. ‘Theorem VLA Let (x) denote the bounded operators on a Hilbert space of: Let T, be a sequence of bounded operators and suppose that (Tex,3) converges as 12 for etch x, ye. Then there exits Te 2) such that 7, . Proof We begin by showing that foreach x, sup, [Tex < 2. Since for any E20, Ty) converges we have sup [x39] < 20 For each n, .'¢-£F,C), and since sup, |(7.2Xp)|e < 2, the uniform ‘boundedness theorem implies thatthe operator norms of the T, xin 2(2,C) sr uniformly bounded. But the norm of T,as an operator in (2, C)sthe fame as ite norm in 3: thus {Toa uniformly Bounded. ‘Now, we use the uniform boundedness theorem agai, Since sup ITerle < we conclude 0p Tiga <2 Viz Aajolote 8 Define Bs 9) = Finest and 1 (T%.9). Then itis easily verted that Bx, i sesau- [ace 9) age, wag = IT bear ‘The second equality uses corollary of the Habs-Banach theorem. H ‘Weare mostly interested in the case where isa bounded linear transforms tion of a Hilbert space > to sell. The Banach spice adjoint of Tis the mapping of 2°" 10 2¢%. Let C-2¢ 2° be the map which assigns to etch 1¥€F, the Bounded linear functional (9, ) in 37%. isa comgate linear Isometty which i surjective by the Riesz lemma. Now define a map T*: roe by recore Then T° sisi (D)= (CHT) = (THY) = (CC, 9) = (TE) Ti called the Hert space adjoint of T, but usually we wil just call the adjoint and lt the *eistingushit fom T™. Notice thatthe map T+ T* is ‘coniupate incr, that, 27» 3, This becaute Cs conjugate linear, We summarize the properties of the map TT ‘Theorem VES (a) T-T*isa conjugnte linear isomeric Romorphism of 209) onto (2). () (Isr = sere, (@ (Ty =T, (@), 16 Thas a bounded inverse, Tt then T* hs a bounded inverse and =o Via Adie 7 (©The map T+ 7*isalways continuous inthe weak and uniform opera: {or topologies but is only continuous inthe strong operator topology i 2 fs Anite dimensional. © iren= iv. Proof (a) follows from Theorem VIL2 andthe fet that C isan isometry (@) and (e) ate easily checked. Since T-*T= [= TT we have From () TAT Fa Pm OT hich proves (, ‘Continuity of T- Tin the weak and wniform operator topologies rv. Inthe case 2° = here isa counter example which shows hat T-> T* is not continuous inthe srong operator topology. The general infinite dimensional seis similar. Let, be right sift on £, by m paces. Then I converges weakly But ot strongly to zero. However, WS = Vg converges strongly to zero. Thus "+0, but FE = W, doesnot converge strongly to 20. (O Note that [TTI < \TH IT] = ITH and [TOTES sup (x, 7°72) = sup (Tel? = 17 Definition A bounded opertior T on a Hilbert space is called self. join if T= 7 Seltadjont operators pay a major role in functional analysis and mathe- ‘matical physics and much of our time is devoted to staying them. Chapter VIL is devoted to proving a structure theorem for bounded teladjint ‘operators. In Chapter VII we introduce unbounded selEadjint operators land continue thee stuy in Chapter X. We remind the reaer that on C2 linear transformation i sefadjoin if and only ifs mati in any ortho- ‘normal basis is invariant under the opevation of election eros the digo followed by complex conjugation, ‘An important clas of operators on Hilbert space is that ofthe projections Definition IF Pe 212") and P? =P, then P iscalled projeton If in addition P= P*, then Pi called an orthogonal projection, ‘Notice that he range of projection is always closed subspace on which P ace ike the identity. I in ation P is orthogonal, then? ate ike the 20 ‘operator on (Ran PY Wx = y+ =, with ye Ran Pand ze (Ran P)* is he IT, the the series on the right converges in norm and iti easily ‘checked that for such J its limit is indeed the inverse of (7 T). Thus, 2 [i +c, [R\(T)\-+0.Ire(7) were empry,Ry(T) would bean etre bounded analytic Fonction, By Liowvill's theorem, A(T) would be 210 whichis & contradiction, Thus, o(7)is not empty. ‘The series (V1.2) i called the Newmann series for R,(T). The proof ofthe corollary shows that o(7) is contained In the closed dse of radivs 17 ‘Actually, we can say more about o(7).‘82 Vic BOUNDED OPERATORS Definition Let 17) = 0p 3) UT); called the spectral raias of T, ‘Theorem VLE Let X be aBanachspace, Pe2(X). Then lites [TH texiss and is equal to (7). IF isa Hilbert space and 4 eeleadjon, then Ha) Proof The reader can check that limya [TIM exits by following the clever subaddluvity agumeat outlined im Problem 1. The crux ofthe proof (ofthe theorem is fo establish thatthe radius of convergence ofthe Laurent series of RT) about 2 is just o(TY"* First notice thatthe Yadis of con vergence cannet be smaller than 7(7)" since me have proven tht R(T) i analytic on p(T)and (| [2] > 1(7)}& p(T). On the other hand, (V1.2) just the Laurent sties about 2o and we have sen that where ic converges abso- Iutely. RT) exists, Since a Laurent series converges absolutely inside the ele of convergence, we conlide that the radi of convergence cannot be larger than 1(7-". That (7) = im, Tl!" follows from the vector-valued session of Hadamards theorem which say that the radi of convergence of (V1.2) is jus the inverse of Tyres =tim yee ‘iol if is Hilbert space and 4 is seladoint, then II? = 28 by part (/) of Theorem VIS. This impis that 4") = 50 in 8 i LAP" A “The flloning theorem is sometimes well in determining spectra ‘Theorem VI.7 (Philips) Let ¥ bea Banach space, Te 2(X). Then (Th oT) and RAT} = R(T). AF 28 is @ Hilbert space, then o(T*) = lz ocTyand RT) = KATY We note that the Hilbert space cat follows fom () of Theorem VI. ‘We now workout in ome dtal an example which lustre the various King of spect Example Let The the operator ony which acts by Theda = reed Via The spectrom 180 The adjoint of T, 7 ats on 4 by Pa Bay) = OE kasd Wetirat observe that T= [T= 1s that all with |] > 1 aren (Tpand (AT) Suppose [3] 0 Jus, 1s [2] = \/F2 we would lke t0 deine || = y/4*A, To do tht we must Show that we can take square r00t of postive Operators. We bepin with & lemma. Lemma The powerseres for /T— ¥ about zero converges absolutely for I complex numbers» satisfying [2] $1 Proof Let fU=2 1 tee + 22 +0 be the power series of VT bout the efgin. Since /1—2 is analyte for |2| <1, the series converges absolutely there. Th derivatives of T= atthe origin are all negative, 10 the e, are negative if > 1. Ths Sieaia2- Se ~ im Satwhet im, = mean he mit tx appoaches one from slow. Sic his is tive forall 8 Er lel £2, which mpl tha the series converge abo. Inty fr [2] = ‘Theorem VES (squate root lemma) Let Aes2(H) and A= 0. Then ere is a unique Be (27) with B20 and B= 4. Furthermore, B ‘commutes with every Bounded operator which commutes with 4 Proof Its sufiient to consider the case where [<1 Since MAb sup (l= Ae, 0) 1 the above lemma implies thatthe sees 4 (f= 4) + e,(0~ AP cone ‘eres in orm to an operator Since the convergence is absolute we ean ‘square the series and rearrange tems which proves that BA. Further more, singe 0514's J ve have 0:5 (@, (I= Ae) fr all 96.8 with igi 1 Thes tem 1+ § sted arebano where we have wsed the fact that e, <0 and the estimate in the lemma Thus, ‘120. Since the series for B converges absolutely, it commutes with any operator that commutes with 4 ‘oppose there isa, with B= Oand (BY = A. The since Bm (BY = 48 2 commutes with A and thus with 8. Therefore (B88 - BY + (B- BH B~ B) = (8 ~ BKB-—B)=0 (VI) Since both terms in (VI3) ar postive, they mut both be zo, s ther fer ence (B~ BY? =0.Since B~ B'sseiadjoin, |B ~ B'\*~ WB - BY| = 0, SOB B= 0.9 We are now ready to define | Definition Let 4€ £0"), Then [al = JAA ‘The reader should be wary of the emotion connotations of the symbol I-[. While itis ive that [24| = [2] [Al for 46. is im general fle that VLA Peltive potters and the pote decompesion 17 4a) = [4 [8] or that [4] = [4%], Furthermore tis not eve in pene hat 443] [4] + [3] (roblem 16) Infact wie itis own hat 18 ‘oem condauous oe Problem 15,8 not known whee i Lipset {bats wheter | |4] ~ [5] Tc 1A ~ Bl forsome constant Gower ee Probiem 17. "The analogue of te complex aumbers of modulus one i ie mre complicated. At fst one mip expect hat the unitary operators would be Sufi, bt the following example shows that his ot he eae Example Let bethe right shift operator on/,.Then || = JAA =I sof we write A= UlA| we must have U= A. However, 4 is nol unitary Since (10,0, snot ints range. Definition An operator U'€ (at) is called an isometry if Ux ~ forall x 2 Uisalld a partial isometry if Ui an isometry when restricted to the closed subspace (Ker U)* ‘Thos, if Uisa partial isomety, 2 canbe writen as 2° = Ker U@ (Ker U)! and #¢ = Ran UG (Ran UY and U ea unitary operator between (Ker U)*, ‘the intl subspace of U, and Ran U, the final subspace of U. Is not hard to see that U* isa partial isometry from Ran U to (Ker U)* which ats as the inverse of the map U: (Ker U)* = Ran U. Proposition Let U’be a pata isometry. Then P, = U*VandP, = UU* are respectively the projestions onto the intial and final swbepace of Conversely, i Ue 2() with U*U and UU projections, then U is partial Isomety ‘The prof ofthe proposition i et to Problem 18. We are now ready to prove the analogue ofthe decomposition = [|e ‘Theorem VLA0 (Polar decomposition) Let A be a bounded linear ‘operator ona Hilbert space 2°. Then thre i paral isometry U such that A= UA}. Uisuniquely determined bythe condition that Ker U = Ker 4 Moreover, Ren U = Ran Proof Define U: Ran [Al + Ran A by U(\A] ¥) = AV. Since (, |AI? W= (6, Ata) = FAIVi: BOUNDED oPeRATORS Us welldefined, that is if [Aly = [Ald then Ay = Ad. U's isomeric and so extends to an isometry of Rai [4] to Ran. Extend U to all of by defining ite be zero on (Ran |). Since || i selt-adjont, (Ran [A1)* = Ker [4]. Furthermore, [4| =O f and only if ay ~0'so that Ker |] = Ker A: Thus Ker U= Ker A, Uniqueness sft tothe reader, In Problem 20 of Chapter VI, the reader wil prove that is tong init of polynomials in A and A* so that Us in the "von Neumann alba” enerated by 4 VIS Compact operators Many problems in ciasical mathematical physic can te handled by refor ‘mulating them in terms of integral equations. A famous example is the Dirichlet problem discussed atthe end ofthis section. Consider the simple operator K, defined in C[O, 1} by (509) = [kts reddy oy where th function Xx, 3) s continuous onthe square Oy $1. Ke) called the Kernel of he itera operator. Since ssone = (ps1) ) swe ee that vote) $0 K isa bounded operator on C[0, I. &’has another property which very important. Let By denote the functions ein C01] such that ol. 0, we ean find > Osuch that [x —2'| ‘verget subsequence (he limit may no be in KIB). Another way of Saying ‘this is that the set (2s precompat; that is it closure is compact in ‘C(O, 1. Tis clear thatthe ehoie of M Was not important so what we have shown is that K takes bounded sts into precompect sts. I is this property which makes the #0 clled "Fredholm alternative” hold for nice inte ‘equation ike (VLA), This section is devoted to studying such operators. Definition Let X and Y be Banach spaces. An operator Te (X,Y) is called compact (or completely continuous) if takes bounded sets in Xiao precompect ses in ¥. Equivalently, Tis compact if and only if for every bounded sequence (x) € X(T) has a subsequence convergent in Y. ‘Theintegral operator (VI4) is one example ofa compact operator. Another class of examples it: Example (ini rnk operator) _Suppon hat the range of Ti faite ‘Genter, That every vector inthe range of Teen be ten Ts = ya for some ted family (lyn Ys an bonded eqs in Ye be Geponding ot ve hounded ce T's bounded The ssl sob ‘equence tok lows one ro ext a omergentsubeguens fom (Te) sich proves that Tis comp ‘An important propery of compact operators i given by (compare robs Jem 34): ‘Theorem VI.tt A compact operator maps weakly convergent sequences into norm convergent sequences. Proof Suppose x7 x. By the uniform boundedness theorem, the I) a bounded. Let yy Try. Then €(,) = ¢y) =(T’Aa, =) for any ZY ‘Thus, converges weshly to y= Tx in ¥. Suppose that y, doesnot converge {o yin norm. Then, there is ane > O and a subsequence {yg} of (,) 9 that lig — 71 2 Since the sequence (x) bounded and Tis compact fy} has 8 subsequence which converges 10 2 7). This subsequence must then also converge weakly to J, but this is imposible since p, converges weakly (0 ‘Thus y, converges to yin norm.We note that if X is reflexive then the converse of Theorem VII holds (Problem 20) The following theorem is important since one an vse it 10 rove that an operator is compact by exhibiting it a @ norm imi of compect ‘operators or a8 an adjoint of a compact operator. ‘Theorem VIA2 Let Xand ¥ be Bansch spaces, Te LOX, ¥), (a) IF (7) are compact and T,~T in the norm topology, thn Tis compact (@) Tis compact if and ony if Tis compact. (@) WS e-2(7,2) with Za Banach space and if Tor Sis compact, then STis compact. Proof (a) Lt (xq) bea sequence inthe unit all of X. Since 7 s compact {or each, we can se the diagonalization trick of 1.5 to finda subsequence of ‘egocall it (x80 that Ty» yf cach nas k=, Sine eel <1 and UTh~ Ti--Ovan e/argument shows that the sequence {3} is Cauchy, #9 Yo72y- Tei nt dieu to show using an e8 argument that Tk. Thus T compact (b) See the Notes and Problem 36. (©) The proof is elementary (Problem 37). Weare most imerested inthe ease where Tis « compact operator from separable Hilbert space to isl, so we will nt pursue the general cate any further thowever, See the discussion in the Notes), We denote the Banach spuce of compact operators on a separable Hilbert space by Comt3P), By the {st example end Theorem VII? the norm lit of sequence of finite rank ‘operators is compact. The converse i lo tre in the Hilbert space cae ‘Theorem VAS Lett bea separable Hilbert space Thenevery compact ‘operator on 2 isthe norm limit of a sequence of operators of finite rank. Proof Let (fy be an orthonormal set in 26. Define =), ITU gg ee Cleat. (2) is monotone decreasing soit converges toa limit 22 0. We fist show that A= 0. Choose a sequenet Weir... Wall= with VIS Compact operators 20 ITVal1= A. Since YO, THe +0 by Theorem VILL. Thus, A= 0. As & result Len ote fn norm since 2 i just the norm ofthe iference. H We have dscused wide variety of properties of compact operators but we have nt yet described any property Which explains ou special interest in ‘em, The basic principle whigh makes compact operators important isthe Fredholm alternative: If 4 is compact, then ether AQ = basa solution for (I= A)" iss. This is not a property shared by all bounded linear transformations For example, if 4 isthe operator (4s) =xets) 08 30,2] then 49 = 9 has no solutions but (F~ A)-* doesnot exist (asa bounded ‘operator. In terms of “solving equations” the Fredholm alternative is ‘expecially ac: Teel s that ifor any g there is at most one y with » = ‘9 + A¥, thea there isalways exactly one. Thats, compactness and uniqueness together imply existence; for an example, se the diseussion of the Dirichlet ‘problem atthe end ofthe section. 'As one might expect, since the Fredholm altemative holds for ite- imensonal matics, it is possible to prove the Fredholm alternative for ‘compact operators (inthe Hilbert space case) by using the fact that any ‘Compact operator 4 can be written a 4 = F-+ R where F bas finite rank and ‘Rhas small norm, Comptcness combines very nicely with analytic 0 we first prove an elegant result which sof great use in itself ee Sections XL, XLT, XIIL4, and XIILS). ‘Theorem VEU (analytic Fredholm theorem) Let D bean open con: nected subset of © Let/: D+ 2) be an analytic operatorvaled function ‘ch that f() is compact foreach + © D. Then, either () C= f10)"* exis for no 26 D. (©) U=/(0)"* exists forall 26 DIS where Sis a discrete subset of D (je. et which has no limit points in D) In this ease, U—/(@))" 8 mero- ‘morphic in D, analytic in DS, the residues a the poles are ite rank opera- tors, and if 6S thea /()¥ = W has a nonzero solution in. Proof We will prove that nest any Z ithe (2) oF (b) hols. 4 simple con- noctedness argument allows one to convert this into a statement about all of D(Protlem 21). Given 29 D, choose an r so that |2— zl O. Let 2, be the postive square root of and st y= AVA/i. A short calculation Shows thatthe g, are orthonormal and tha eS o The proof shows that the singular vals of are pecsly the eigenvalues of al Wie conclude with a clasial example Example (Dirichlet problem) The main impetus for the study of ‘compat operators arote from the vse of integral equations in stfemping 10 folve the classical Boundary value problems of mathematic! physics We briefly describe this method. Let D bean open Bounded region in R with & smooth boundary surface 2D. The Dirichlet problem fr Laplace's equation 's: given a continuous fuetion on 2D. find & function a twice dillerentable Jn D and continuous on D, which sais Qua)=0 xed Ma)=/0)-x€2d Let K(s,)) =~», 9)2eh~y1? where isthe outer normal to 2D at the point y€ 8D. Then asa function of z, K(x, )) satises A, Ux, 3) = 0 in {he tntroe which sugpests that we ty to wt as «superposition 00 [kts neoras0 cs) here of) is Some continuous function on 3D and dS is the usu measure. Indeed, for x€-D, the integral makes pevfecly good sense and Vis Compact operators 28 Aux) =0 in D. Furthermore, if xis any point in @D and xy from Inside D, scan be proven that vn eted +f ea siondsy) 4 tq om oie , he in ele by ao A, Aeon ason ‘exits and isa continuous fanetion on 20 if gis continuous function on 1D. The proof depends on the fact thatthe boundary of Dis smooth which implies that for «2D, (2 ym.) = el ~ yas x49. ‘Since we wish és) =) on 3, the whole question reduces to whether We can find 9 so that Fe) = — 910) + J Kis sd0() dS), xe 3D Let T: C10) CD) be defined by Tom {ke 00() A519) [Not only is Thounded bu (4s we will shortly se) Tisalso compact. Thus by the Fredholm alternative, eter 2~ Lien the pot specrum of Tin which ‘ite there isa Ye 10D) such that (I T)g=0, of f= (I~ Tho bis a ‘unique sotion foreach fe CI@D). Iu is defined by (VLG) with ¥ replacing {then w=0 in D by the maximum prineple. Further, dum is continuous ‘ross 0D and therefore equals zevo on 0D. By an integration by pats this implies that w = outside 2D. Therefore, by (V6), 26(3) =O 08 2D, 80 the fint alternative does not hod “The idea ofthe compactness proof i he fllowing. Let 2m) Koad SoS 1f3 > 0 the kernel Ki continuous, so, bythe discussion atthe beginning of this section, the corresponding integra operators Ty are compact. To prove ‘that Tis compact, we ned only show that [TT] =0 a8 60, By the cating 1.0) ON ef, IR) = Kl 48)we must only show thatthe integral converges to zero uniformly in x a8 {50.To prove this, divide the integration region into the set whee [x=] 26 {and its complement. For fixed the kernels converge uniformly onthe fre region. By using the fact that Kis integrable, the contribution from the second region canbe made arbraiy small fore suiiently small. VES The Wace class and Milbert-Sehmidt ideale Inthe las section we saw that compat operators have many nice properties tnd afe sefl for applications. I is thezeore important to have elective sriteria for determining when a given operator is compact or, beter yet, feeral statements about whole classes of operators. In this section we will prove thatthe integral operator me) J xenrorduon on LM) i compact 1K) 28 Md 8d. i el Sev the eats, 0 whch gine ina Toren WEE Stove that Cont) he compat operon seule Hike sac farm Banach par. At te condaon of te scdon, we vl compet te oan ou dal of Comex) Tae aston tee dees Screen he mest Boach pce ology on 10") ane weak opt teplog an pea forte of te nracure of aban von Newrase “Miwa sn ne snr Thence gel of te al oon of te sm of te goal cmon of man, bt mse sam ae ne nt ah pent sive ata, Thecnsteconot he bases anegos (sth season ofthe Letesgs ingle ona des dn for =O i har aloe ink nog So Ton 2 tod toe tafe ‘Piacoa nd festa lar uncon Sasi tae thts thon the poo operon; Asstt has aly.) We then dee he tac saw, tea 4 (such a [| Coe We Sil enshow hat) haar acon on wih ei ropes Theorem VIIB Let > be a separable Hilbert space, (92, anortho- normal bass. Then for any pesiive operator de 2(3") we deine tr A 33-1 (0,, og. The number 4 is called the trace of A andisindependent ‘ofthe orthonormal basis chosen. The trace has the following properties ‘VLA The ace late and ber Sohmitdaale 7 (0) WA + Bwrdd, () Wd) = dir A forall 220. (© wwau ©) WOSA SB tena sed, Proof Given an orthonormal basis (wf. define tg(A) = Dry (@x- AB. (else, i another orthonormal bass then wsl= Sonded= Eure 28 (Eiw..0%0r) = 5 (Z1arv..0or) Evy o Banas rie Since ll the terms are positive, interchanging the sums is allowed. Properties (a, (0), and (2) are obvious To prove () we nate that i (p,) isan orthonormal bass, then 50 i (Uo). Thus, (VAUD) = tra) = 4). wuau Definition An operator 4 e (2) is called trace clas if and only if te || <0, The family ofall rave class operators it denoted by J. ‘The basic properties of, ae given in te folowing: Theorem VI J is 4 odeal in 2130, that is, G2) J is. vector space (0) aed; and Be 200), then BES, and BAe, ©) Wad then Are, Proof (a) Sine {id|~ 3] [Al for Re, 4, ie closed under sale ‘multiplication. Now, suppose that A and B ae inf, we wish Lo prove thatA+ Bey, Let U, Vand W be the pata someteies arising from the polar decompositions Ae ulate A= vA) wa} Then Seen As Bled = Ste 04+ Bed Buen vevisivats Elemis! Ee. uviaioot s En larAvved tle = (Zuvarev-veur)" (Ear) Thu wea tow Enareronse sua un ‘we can conclude that ° Lloclae niga su tal ve alco and thus 4 + Be.J,. To show (VI), we need only prove that (U'VAIV* U) ste} Pickingan onhonormal basis, (9) with ach gn Ker U or (Ker U)! we see that U(U* AI PU) = ||. Silly, picking an orthonormal ass, (Ya Wit each Yq in Ker V* or (Kee Pe we find tV|A1Y") = weld (©) By the emma proven below, each Be 2") can be writen asa linear combination af Tout unitary operators $0 by (a) me need only show that Aes, implies UA e J, and AU ed, if Uis unitary. But |UAl = [4] and [AU = U"" [Aid 80 by part (0) of Theorem VIB, AU and UA are in Fy (Leta = UA) and 4* = V)4*| be he polar decomposition of 4 and AP Then [At] = V* [4] UP. IPA ES, then |] Fy, 30 by past (6) [4 € Foard a= aL A “To complete the proof of part (b) above we need the following femma which we will use in other context ater. Lemma __ Every (a0) canbe writen ata liner combination of fue nia operatrs. Proof Since B= 40+ B*)— {{(8~ B*], B can be writen as 9 linear combination of two selFadjoin operators. So, suppose 4s seladoint and without los of generality assume (AIL 1. Then A fY/T= 4? are unitary and A= HA +i T= 4) 444 T= “The proot ofthe following theorem is left to the reader (Problem 23), mm VLMD Let Il, be defined in 44 by NA, Banach space with norm oy and ANS Aly |Al. Thea We note that 45 i or closed unde the operator norm l=. The connetion ‘between the ace class operators ad the compact operators is simple Theorem Vi2t__ Every A..F; is compact. A compact operator A i in Ay ifand only FE 2 = ce where (hey re the singular vals of 4 Proof Since AeI,, [Ale Jy, 90 (IAL) Suppose ¥ €[0,, | Ag < 2 for any el and Til I hen we save swat) Suan ee ee sup(lAVIIV EL... 99) IVI 10 as Novo Therefore TZ. (6, Av it norm convergent to 4. Ths 4 is compact. “he send part ‘f the theorem fellows easly from the canoe derived in Theorem VLT Problem 28) 8 Corollary The finite rank operators ace dense in ‘The second class of operators which we will discus are the Hlbert= Schmidt operator, the analogue of 2Definition An operator Te £12) scaled Hibert-Schmid if and only ‘tte ToT oo, The family of al Hilbert Schmit operators i denoted by I ‘By argumens analogous to those we used for J, one can prove ‘Theorem () Jais wisea. (@) Ih, Be, then for any orthonormal basis (oh Eewmy ‘absolutely summable, and its iit, denoted by (A, By is independent of the orthonormal base chose, (o) whiner prot, ia Hilbert space. (8) Whale = Ja), = (04a) then WANS IAG SHAK, and 24K, dts (6)_ Every AF, is compact and a compat operator, 4 is in J if and only iS co where yar the guar values of (0) The nite rank operators te | ledense in J {@) 464; sfand only i (Aga efor some orthonormal bass) {8} 469; itand only A “BC with By Cin We note that J, is not -Hilosed. The important fact about J, isthat when 2° = LM, di), has a concrete realization. Theorem VIZ3 Let (WM, u) be a measure space and 2° = L4M, dy ‘Then 4-€ 20) is Milber- Schmidt iFand ony i here i function Kel Modus) with ante) = fs 909440) Moreover WAIE = J 1XCe 91 4 0) Proof Let Ke LAM x M,du@ di and let Ay be the associated integral ‘operator. tis easy to Se (Problem 25) that Ay isa well-defined operator on and th [des Ks oy Vis Thera clan and ibe. Sehmitidele 24 Let (9,32. be an orthonormal basis for L3(M, dh). Then {9,()70)kan= isan orthonormal base for L4(M » Mu @ ds) 50 E sn 0teineD Ke Ene) “Then each Ky the integral Kernel ofa init rank opeator. Un fat, Ay, Dees Seely Mae Sine Ue = Kla20 we have dy ~ Aggl*0 a8 {N52 bY WL). Thus yi compact and in fet wate Eden? £ fel? =e Thus Aye, nd Egle = Kh "Weave shown tht the map K- 4,e an ieometry of L'(M >, dd) into a0 ts ange closed. But the finite rank operators clearly come rom erels and since they are dense in J, the range of Ke Ay is all of Fy. W “hie theorem provides simple sfcent condition for an operator 1 be ‘compact and is therefore vty Useful. Notice that the condition isnot neces tary, Also, we have scent condition foran operator on 2 = L'(M, da) to bean integral operator. This condition i also not necessary. Now, we retorn 10 defining the trace on ‘Theorem VI24 Ae, and (gata is any orthonormal basis, then Sins x. 4g.) converges absolutely and the limit i independent of the hoi of bas Proof We write A= U|4|*?[4|", Then 1, Ava $1) Al2 UF 01 A)! sh Eves anat s(Geiaruren) "(Ea)" se since 4|"U" and [4 arein 4, the sum converge. The proof of the independence of basis is ideatcal to that fort when A > 0. W Definition The map tr: F,-+€ given by tr4 (2) any orthonormal bas called the tac‘We remark that its not tre that P2.s |(eys ga] < 2 for some ortho- noxmal basis implies €.f. For to bein J) the sum must be ite oral ‘orhonormal bases, The spectral theorem which we wil prove in the next hapter wil ell us that any selPadjint dean be witten A= Ay — A where both Ay and 4. are postive and 4. A =O. Not surprisingly, AeA, I and only if (4) < eb, A.) < 2 and in this cae = thy te A ‘We collect the properties ofthe tae. ‘Theorem VL25 (a) te) lina (aid (©) WAB= te BA AES, and Be 2OP), roof (a) and (b) ace obvious. To prove (cit is sulicient to consider the ‘ase where Bis unitary since any bounded operators the sum of four uniter Te har ease WAB= $0,480) = Suton Sau) wee ve Beran IA Jy the map Bott Bisa linear functional on (oF), These are ot al he Continuous linea functionals on <2(2) but such Functional do Yield the ene dual of Come), the compet operators. We cam also hold Be ZU) fed and obtain a Hineat functional on F, given by the map ‘drsir BA. Theset of hese furtionals is jst thedul of, (with te operator orm topology. We state this as theorem; the interested reader ean follow the outline ofthe proof given in Problem 30 Theorem VI26 (2) J, =[Com(x)}*. That i, the map Amt(4) ie an isometric inomorphitm of J, onto [Com (21 (EF) = FT That ithe map Br-+tB-) i an isometric fomorphism of Zor) onto Ft We now return othe distinction between the weak operator topology on (oF) (ore Seton VI) and the weak Banach space topology, 1. the eX2120), 2(30)) topology. If Fis the Family of fie rank operators, then FCF and cach Fe # can be realized a near Functional on 22) via the ‘ual action of J, on 20). The topology on (x) generated by these functionals, that of PC"), 7) just the weak operator topology. The set (F isnot closed in the Z2UEV-norm, Asa matter of fact the (30) noo on just 30th closure of in this norm just J, The weak topolony ‘on a) generated by the functionals in Py, thats (2120), J), scaled the wtrameak topology on (2°). Notice that it is stronger than the weak ‘operator topology, since more Funcionals are required to be continuous, but weaker than the weak Banach epace topology on (2), since J 16 not the entre dial of GP) In fact, since 2(2P) = 9%, the ulttaweak topology on ‘2(ot) is jus he weak topology. This realization of ZF) a the dual of ‘the Banach space of linear fancionalscoatinuousin the o( (20), #) topology isvald for larger clan of algebras than jst-2(2). Problem 31 gives another trample: the muitipheation algebra Lon Ly, We wil study such algebras in ‘etal in Chapter XVII. We study the J, spaces for p #1, 22 in Sections ea and XT, Sectin VL} reader be tered te may opts nebo cn 212), the wa, aang, aed unform oar tpales the wenk Burch ie rang ean ert Seton Vi Laren wei ener ito feats eve metedn ep one i of in Non iorn tok eat hat om ean "some Sot ao naw at ‘ropes ef cig spt lon ram proper hr sec fo ear he nam i fangs op hp Fro wb petit PO ‘Ste duan alvns ino nh see enw sos lt hee we ane at {Splone ot hand Inger isthe wen, ong, tir opto ops wh _Seimorar in Volume fan Thesrsneat a uatong topoope il hy oak ‘Shenae da nh on Neuman sigs The weak, son. an ua epee Xeplogc were imsdced in} ton Neumann, Za Alger er Fucomopertionn (Su tiworic Normale Operon” Math An 102 (92190, 37D Secon 12 ps cen fo sain epee o finite desir! wie eed FR ior Fe Do! Pcr Space, Yon eon neha, Pn, New Som 198 Sion 7.2 Thedeiions around fect ao ews or etoundet operon Theoem Vs hole ong arn reat Te lets (Frou, anomaly cso2 Vi: BOUNDED OPERATORS ‘he theoy of Ranch spaced anc function dese in rea el in ‘oe ern the, ide a9 Frm ope Buch sue nate. pon f Thr Vi. cnt founda Fora “Arai, Ace ro, New Yor, 6S Ys Sime sure or ame: oni Hie, Pp) eh rn “contin pce umf dente any eo?) which ater the pin sperm anh Fea {ecnan Othe ators cha Keo or Rene, Nagy wee denon at we pe ‘Seon VIL? One important cist th with ou xin the contacto athe plat perm eed oat edt Se VE4__‘Thw pour dsempoion tt sine geome rane fe Her esttemains on Any les oman Aan Wea ewe ost OS whee 1's echopnat ard mma Bytes these Scan be tought st ‘tion eerecon, esnsthistnn actin per rhognl ona "heen ays lenin opi an py “The states tha eng gly fafa, 4 ye ‘ta eeu c (41+ [8 ne robe 1 att hat) = [a ae comet perros Rntion that for 24s hed ena e+ Nfs tee fr earl peut AaB Sete fn ht For CDN CS Gynt fran real ano t= sey wich ascend oper ved "uncios ate antes ts be sted: F_ Krum“ ber konese Marrone” tin 2 a1) 142 a) ent and § Shean,” Monson end Cover Oper anc” Tan, er Mah Soe 79 (989, 301 Seo V14__The pol ofthe econ pat of Ten Vi a be fondin Youes Sostiamseappon of AAs Ao ons er abo Pen "na ey ea eat thence operates se tack Fels epee Fredan contre oving eatin he ar Beye) A9 Rend erg and are en contours nd =< a (Pow patio fom 2129 GP) 2009 tay ominaos ‘Shon #4) een te og toon (6) Sion (Aranda be af 4B, Bove dB AB, (6) Ltn, Bb sees ot B rove tat AB AB B-Givean eamplewbere Ae AB Cie ance oso hat se ig tans operat ot ee ‘rove Th bund eerybee sete, an at ety hen Ran ce 18.0) Lette aseteadht banded optar cer ace Pett hie ‘ahs fae ead hate Speen creo won eens ecropea, “ (From th oof Thorn VLE drive x wn bt dependent) bond fer ‘benom ote wschen of ant ope at oot ACE (oat ce petro ber ss, ret Ai up 4,21 ne ante at Reh Ad) ~ MO 6 AEE BD =f. oy 40 pn andthe plore hat 1.481 op a pt Wt (Finan carl wich shows that the concn ef sel nat eae Ai sotedeadoat 12 St ees aif Yer ne cor ame ['r98 samp on Cite fos. What the orm of 7? tH Let Pe 13) Pron th el x an 1 fa flows (6) Setar og land rove ht exe ae (8 Fox sled poeta mt 2 my Seg ad ae pose igs Sed Orem Ug) conceal Ries (© Pre thins ad he se ey. 12, Prove he propionate end of Seton VL 1.) Ghe sn eam wbich tows ht lear tacformtion om C* cn pone ‘ey Daves mary and wr condi fort nx ati be pov. 14a) Prove that i 42, dame omen Vk -= VR aor (0) Stppose A, Seonh for a seuence (4) reve tht V4, + VA sony 15.) Lat As Ai oem, Pov ta | [4 90 {8 Sep fal a aoe usr re at A= (©) Fan ample which shows tat 0 weal comin on 2F)- 2). rove tt fe tat Kort D+@.—Mi Siler +01+ Ker) Renark: Tis ample deo B. Neon Nal (Bist a0 (tne: Se bie 6) 18.) Prove propeitin reeing Theor VLIO. (8) Prove he enueaa a Theorem Vi. 19. Wete he mati“ asthe prota fren an postive symmetries “20. Suppor tat Xie sete Banach paced ta 7X» tov na pe Sor Move that fT tae west convergtSquncs it nom soe oo ‘tenn then Ts compe. 12, Compete poof Theorem VL. by extending te ea olf,1m, 1m Vi: BOUNDED OPERATORS, sing he Stones theorem prove ta every Fel ile on Get} one-man ere onto om of petro al nk (Proved 141 < Ht, (© Stoyose an Caney seu. Sow ta ats ia that ‘ula «ss Ten coe te po! of Prem VIB by shag that the Wish oc (2 Use the canst form gen by Theses VT 0 prov the ond emeta (0) Prove tbe cry to There VL, et KEL Md ad A he ine opee topd= [Kerr 40 rove tat Ae wll dened ed at < ie (3) Prove hat. App.) < © eal ertnsel bth th A (8) Prob an ag 7, that Cape) < fron aed tonal i, Prove tat AB) = OIC A Be-4y Prova) ABS At UB Prove tht Ae J fan of A= BC wth Band Cia J. “Tega of hi pobem to pow There VI25 (3) Et Bonded foal on Com). Let (9 be he operator on 2 nae ito (fh Sho hat en ios boned ea pert eo vingteteran —— BPMAG AL Be awa i[§ enors] brotha BEF an BP < hen co Pra mabe iin Cont wn (e) Prove at I= emia {e) Lary beutounde nt fons 6o 4, Show tt er opto O29 A198) (0 Prove that Ar-tB4) Donde sft on 4, wich ges wth g sedi tle = 2 ak > Be cre pce ne LM 208 2 gn ~ Kort) ‘rove tt te topo 001 nd bythe mak operator ploy on 127 ‘etal othe wats ply indeed on "ty Ls 106 doy 22 Le Co, acon £40, 1 an Probie 31 Fad a out 0,1 emer in eye csetr wpb o ie ce heh weer ne 30. Comes J iter sae witht produ (A hy = WA"B) Let Am Laan ae ie aps of 12) te 202) gen by Lam AB, RAB) (a) Prove hat dna ta hamomarpim of £2 ito 4. (8) Prove tat 40> is coo et brorergism af 2) io 25). {© Stppse tat Ce_2U) and bes Chum LyC for al ACA), Prove tat CT name Be, “24, Show that na bert pn 4 map. 2- F comaous fhe domaine gen {he oea topology nthe age he nor opie hat = ls Te oe arbrary mt and sly 7h fe kt (Compe ith Tees 35 (9) Seppo Tian get In 2) 0 tat pls Pecente a (Leet the canna eae ps of (27 to if Bh domain nd ‘anges ives tbe weak topo 2, Use) o Theorem VP ade pole decmpition to pov) of Theor VID when k= Pa Her space 17, Prove art (@) fThore VLI2- 3% Let Pad Q be oiogoal projtons onto sbpaces a BF Suppose tat P= OP. Gi Pow 12.1 0.70,P+ OPO and P+ 0-270 ace othoonl pro- pion (2) How are ange of the proetions in) ete to and 109, Le.PandQteonbogsal protons ono supa nd na Hier sue 2, ‘Provethat im ean he ool eon a 9 10. Lat 4 ane ce ein 2130, #0 Prove Cont) = F by rovig tat Sn) ae rank oper 7 ‘Remark: We ise (Chaces Vl Poti 3) hat theo om close eas when 2 spublewe), Con”), FUP. 41 od prejetin on RE which ot an eter rokton 42 tet 6 BUN) Prove hth 2 hat in oO ot A Stee 0d R=) led bat ot al of ts pen eat oF2) Vir BOUNDED OPERATORS. Le ant subspaces Bane pce Yc hat M+ N= and Mn N= (GLP beth proc of gto. Ponta Pr Ged en on “46 (a) Deine he ural rng, 7) of bound epee Ton a Hert ee, 20) MT) =i. Tage 2, = Powe that off) NG in Fe Soe hat fAigon eae oT" T. oe eMC ae sw hat a) nhs not ancient of rT, wecan fdas Sotat HI Ml 20) (0) Fd ap nampe here N(7) a sed and of) © NCP (©) Fé mete whew (7) # = NC omar "Thee dep re of Has a (7 coven 15) Lag {beam oho as fr Mie pace Let A Beam opertor BP WWin0 noe, (Lert ey orn bi fora ert pce an ht A be compact Prove cei gg lMb120 we ne, 65) Lat 20 with 4 compact Prove that 4 cmp (Hit Use lem 45) (L042 Pro it 4b corp if Biscompac Hew: Poe tat ‘Scapoc ing Probie 4 and prt 0) 47 Le and 2° be vo iter space IFT bended ear map fm 240 Seite Sede 7p wales eae ad Fl umber, > 0, ed ethane a en = lr © 20 a 6 fud (a nt a 9 ne Sela ee RTE te ron nanan o Glen peo Se rN $9614. I = ry 2252 poet nemap tings oii esata nth SGantonny ur Ose snd bs) (0 Sian gear how ht te it Pal 2,30, nd eon (Gab ate mh fo atic ye and Eden The Spectral Theorem Manat! pike amon re hard wl a ear, end lb ce i sting bts raming, Job Lace Seed Rap tte Bsa of Wn VILE The continuous functional calevoe In this chapter, we wil discuss the speteal theorem iis many’ guises. ‘This structure theorem isa concrete desrption of al selFadjint operators. ‘There are several apparently distinct formulations of the spectral theorem, Im some sense they are all equivalent ‘The form we prefer says that every bounded selfadjoint operator is rmukiplication operator. (We emphasize the word bounded since we wil del txtensively with unbounded self-adjoint operators in the next chapter; there [sa spectaltheocem for unbounded operators which we dics in Section ‘VIIL3) This mesos that given a bounded sleadjint operator ona Hiltet space 2, we can alvays find a measure 4 on @ measure space M and 2 Unitary operator Us 2 = 2M; 0 that (WAU™' 0) = FIO) for some bounded real-valued measurable funetion Fon M, "This is clearly 2 generalization of the finte-imensional theotem, which says any selfadjoin mm matrix can be cagonalized. or in an abtract form: Given slfadjiat operator 4 on an mdimensonsl complex space,‘ere isa unitary operator U: ¥-+€* and esl numbers Ay... 480 that WALD, = Af for each f= fy. fin In practice, M wil Be a union of copes of and F willbe x, so the core ‘ofthe proot of the theorem willbe the coastructon of certain measures. “This wil be done in Section VIL. by using the Riest-Markov theorem. Our orl inthis section willbe to make sense out of (4), for fa continuous Tuneion. In the next Section, we wil comider the measures defined by the Functional fm (AV) for fixed 6 2 Given a fixed operator 4, for which f ean we define f(A)? First, suppose that A isan arbitrary bounded operator IF/(0) = Se, isa polynomial, we want /(4) = Dt, aya Suppose that fs) = Fy, ia power series with radius of convergence RAT TAL 0,4 can also be epeesented as maliplication by on LA, di). ‘Thus, we explicitly see the nonunigueness ofthe measure inthis case. We ‘an also see when more than one measure is needed: one ean represent & frite dimensional selfadoint operator as mutilation on L(@. i) ‘only one measure if and only if has no repeated eigenvalue, Example 2 _ Ustd te compact and savin. The Hier Sebi theorem tls ut thete compte crthonoimal tf specter Uhalzs th j= ope helen repaid egenane 2) Sore petal see Example 3 Let af =/%(~0, co), that isthe st of sequences, with 3 [ag < 0. Let L: 90 = 38 by (Lal apo that iy Lait the le P= R with (Ra) a,» Lee = R + L which is seladjoint. Can ve represent 4 as 2 multiplication operator? Map 2 into 120, 1 by Us (eg) Siew yet" Then. ULU"* is multiplication by e-?" and URU"? 4s multiplication by ¢"™ $0 UAU~* is multiplication by 2 cos(2nx). The cesar transformations needed to represent as multiplication by x on 1 di) & PAR, dy) ae left Fr the problems and jy have support in (23, ViLE The spect masnnes 28 Example 4 Consider/~!dix on (Rd). This it an unbounded opera: tor and thus not src within the contest of this section, but we wll prove an analogue of Theorem VIL3 in Section VIIL3, We thus tek an operator U tnd a measure oj (it turns ou that only one wis needed) with U: LR, de) = PR, dh) 30 that of £10 The Fourier wasfrm (UPR) = 227!” f fe de which we say in Chapter TX precisely does the trick. Thus, the Fourier unsform is one cxampl ofa spectral representation. ‘We now iavesgute the connection between spectral measures and the spectrum. Definition If (2. family of measures, the support of (i) is the ‘complement of the largest open set B with 4(B) = O fora 40 supp x) ~ (sop Proposition Let A be a self-adjoint operator and (1) spectral ensues, Then 1 family of| 0) = supp i ‘Ther is also a simple description of o(4) in terms ofthe more general ‘multiplication operators discussed ater Theorem VIL3: Definition Let Fea real-valued function on a measure space (M1). ‘We say 1s inthe essa rane of Fiend ony it Hoa ~2< Fm <1 49>0 for alle>0. Proposition Let F be « bounded real-valued function on a measure space (M, 1). Let Ty be the operator on L(M, given by (Trak) = Fomalo) ‘Then of) isthe essential range of F Proof See Problem 175.‘We can now see exactly what information is contained in the spectrum. |A onitary invariant of a eleadjoint operator 4 it property Ps that Pid) = PUUAU~') fr all unitary operators U. Ths, unitary invariants are “intrinsic” properties of self-adjoint operators, that i, properties inde- pendent of “representation.” An example of such unitary invariants the Spectum o(4). However, the spectrum is 8 poor invariant for example, "ulipcation by x on EX, 1}.de) and an operator with a complete st of ‘genfuncions having al aorals im (, 1] 8 eigenvalues are very difereat ‘even though both have spectrum 0, 1] 'At the conclsion of this section, we wile that there f 4 canonical choice of spectral measures” which forms @ complete st of unitary in “arian, thats et of properses which distinguish two selEadoint opera tore and B uness = UBU for some unitary operator U. This explaing why o(4) is soch a bad invariant fr diferent sorts of measres can have the Same support IF we wish to find better invariants which are, however, simpler than measures, it reasonable to fist decompo spectral measure in some natural way and then pass to supports. Recall Theorem 1.13 which says that any measure yn R has @ unique decomposition into Wg Hae + Ming WHEE Ng 6 re Point measure, a. is absoetly Continuous with repect to Lebesgue measite, and iy comtinuous and Singular with respect to Lebesgue measure. These thre pieces are mutually BAR di) = 2, dig) @ BB, dg) @ PAB, dig) It's eay to see (Problem 18) that any y-€L1®. di hasan absolutely con- indous spctsal measure du, if and only if ye LAR, du). and similarly for pure point an singular measures I). family of spectral measures, wwe can sum BE, LIB, ding) by defining Definition Let 4 bea hounded self-adjoint operator om 2. Let y= (la pure point), 2, (Vli i absolutely continuous}, yoy = ‘lag is continuous sng). We have hus proven: ‘Theorem VILA = yy Bite OA mg. Each of these subspaces is Invatiant under 4. A 72, haba complete St of eigenvectors, A TF, has ‘only absolutely cootinuout spectral measures and A Pty, has only core ‘nuous singular specttal measures. [VILE The spec measures Definition eA) = (dI2 is an eigenvalue of 4) Hanh) OA Eg BH) eld) mat 1200) aan( A= 4 1, ‘These sets are called the pore pont, continuous, absolutely continous, and singular (or continuous singular) spectrum respectively. ‘White it may happen that cu ug yp #0 it oly te bese ve did ot deine oy a8 (4 bu ater he acta et of geval One always has Proposition cun(A)= 0x14) UJ eun(A) 1A) = FD U ant) “The sets need not be disjoint, however. The reader should be warned that am) may have nonzero Lebesgue measure (Problem 7). For many pur posts, breaking up the spectrum in thi way gives tiefol information, In Section VIL3, we introduce another breakup which i ls useful As we cused inthe notes to Seon VE3, sme authors use a oton of "continous spectrum” nbich i dain from the above, sanely they {eine the continuous ectrim tobe the set of he (7) which ae neither a the pont spectrum nor inthe residual spectrum. To tate the difeence teiween the wo detnton weet = Cc LO, Nandetine 4:9) = G.s/ta). With ou definition the pont = sn bath he pe pot and thecontnuous spectrom. The other author suige d= tothe pot apectm and their continuous spectrum (0.9) (hI Final, we tur tothe question of mahing canonical choles for the spectral measures,» subject whch goes under the tile of “muh ‘theor.” We wl dsribe the base rit witho prot 1. Mutiptcty fee operators We must first ask when 4 is unitarily equivalent to multiplication by x on 1B(R, di), that, when only one speceal meature is needed. A Took at Example | tells ws this happens inthe Hate dimensional ease only when ‘has no eepeated eigenvalues, 40 we define