Lecture Notes in Functional Analysis

by William L. Paschke

edition 0.9

image/svg+xml

Contents

Frontmatter

I. Hilbert Space

II. Bounded Operators

III. Compact Operators

IV. The Spectral Theorem

Index and References

C. Weak*-Topology

Throughout, Xnormed linear space\( X \) is a normed linear space. Xnormed linear space* \( X^{*} \) is the dual space (a Banach space). We have a norm, ffunction=equalssupsupremum xvector=equals1one |modulusffunction(xvector)|modulus \( \mathopen{}\left\lVert{}f\right\rVert\mathclose{}= \sup_{ \mathopen{}\left\lVert{}x\right\rVert\mathclose{}= 1 }{} \mathopen{}\left\lvert{}f\mathopen{}\left( x\right)\mathclose{}\right\rvert\mathclose{} \). τnorm topology(Xnormed linear space*) \( τ\mathopen{}\left( X^{*}\right)\mathclose{} \) is the norm topology on Xnormed linear space* \( X^{*} \).

Skirmish III.29

  1. For yvectorelement ofXnormed linear space \( y\in X \), get ŷelement ofbounded linear operators(Xnormed linear space*Ccomplex numbers) \( ŷ\in \mathcal{L}\mathopen{}\left( X^{*}, \mathbb{C}\right)\mathclose{} \) such that ŷ(ffunction)=equalsffunction(yvector) \( ŷ\mathopen{}\left( f\right)\mathclose{}= f\mathopen{}\left( y\right)\mathclose{} \). Note that ŷ=equalsyvector \( \mathopen{}\left\lVert{}ŷ\right\rVert\mathclose{}= \mathopen{}\left\lVert{}y\right\rVert\mathclose{} \) from Hahn-Banach.
  2. ŷ\( ŷ \) is continuous on τnorm topology(Xnormed linear space*) \( τ\mathopen{}\left( X^{*}\right)\mathclose{} \).

Definition III.30

The weak*-topology w* weak*-topology \( \mathop{\mathrm{w}^*} \) on Xnormed linear space* \( X^{*} \) is the one given by the neighborhoods N neighborhood(ffunction) \( \mathop{\mathrm{N}}\mathopen{}\left( f\right)\mathclose{} \) of ffunctionelement ofXnormed linear space* \( f\in X^{*} \) defined as follows: N neighborhood(ffunction)=equalsN neighborhood(ffunctionxvector1onexvectornintegerεpositive real number)=equals{setggroup elementelement ofXnormed linear space*|such that |modulusffunction(xvectoriinteger)-minusggroup element(xvectoriinteger)|modulus<less thanεpositive real number for all(iinteger) }set \[ \mathop{\mathrm{N}}\mathopen{}\left( f\right)\mathclose{}= \mathop{\mathrm{N}}\mathopen{}\left( f, {x}_{1}, \dotsc, {x}_{n}, ε\right)\mathclose{}= \mathopen{}\left\{\, g\in X^{*}\,\middle\vert\, , \mathopen{}\left\lvert{}f\mathopen{}\left( {x}_{i}\right)\mathclose{}-g\mathopen{}\left( {x}_{i}\right)\mathclose{}\right\rvert\mathclose{}\lt ε, , \forall{}\mathopen{}\left( i\right)\mathclose{}, \,\right\}\mathclose{} \].

Remark III.31

  1. These neighborhoods form a neighborhood base and determine a Hausdorff topology on Xnormed linear space* \( X^{*} \).
  2. Each N neighborhood(ffunction) \( \mathop{\mathrm{N}}\mathopen{}\left( f\right)\mathclose{} \) is norm open: w*subsetτnorm topology(Xnormed linear space*) \( \mathop{\mathrm{w}^*}\subseteq τ\mathopen{}\left( X^{*}\right)\mathclose{} \).
  3. w* \( \mathop{\mathrm{w}^*} \) is the weakest topology on Xnormed linear space* \( X^{*} \) making each ŷ\( ŷ \) continuous.
  4. The addition map Xnormed linear space*×Cartesian productXnormed linear space* toXnormed linear space* \( X^{*}\times X^{*} \to X^{*} \) and the scalar multiplication map Ccomplex numbers×Cartesian productXnormed linear space* toXnormed linear space* \( \mathbb{C}\times X^{*} \to X^{*} \) are both w* \( \mathop{\mathrm{w}^*} \) continuous; that is, (tupleXnormed linear space*, w*)tuple \( \mathopen{}\left( X^{*}, \mathop{\mathrm{w}^*}\right)\mathclose{} \) is a topological vector space.

Lemma III.32 (Workhorse Lemma)

Let (sequenceffunctionαcomplex number)sequence \( \mathopen{}\left({f}_{α}\right)\mathclose{} \) be a net in Xnormed linear space* \( {X}^{*} \). Then ffunctionαcomplex numberconverges toffunction \( {f}_{α} \to f \) in w*\( \mathop{\mathrm{w}^*} \) if and only if ffunctionαcomplex number(xvector)converges toffunction(xvector) \( {f}_{α}\mathopen{}\left( x\right)\mathclose{} \to f\mathopen{}\left( x\right)\mathclose{} \) for all xvectorelement ofXnormed linear space \( x\in X \).

Proof. (⇒) Suppose ffunctionαcomplex numberconverges toffunction \( {f}_{α} \to f \) in w*\( \mathop{\mathrm{w}^*} \). For xvectorelement ofXnormed linear space \( x\in X \), εpositive real number>greater than0zero \( ε\gt 0 \), define Vset(ffunctionxvectorεpositive real number)=equals{setggroup elementelement ofXnormed linear space*|such that |modulusffunction(xvector)-minusggroup element(xvector)|modulus<less thanεpositive real number }set \( V\mathopen{}\left( f, x, ε\right)\mathclose{}= \mathopen{}\left\{\, g\in X^{*}\,\middle\vert\, , \mathopen{}\left\lvert{}f\mathopen{}\left( x\right)\mathclose{}-g\mathopen{}\left( x\right)\mathclose{}\right\rvert\mathclose{}\lt ε, \,\right\}\mathclose{} \). Then ffunctionαcomplex numberw*converges toffunction \( {f}_{α} \mathbin{\overset{\mathop{\mathrm{w}^*}}{\to}}f \) implies ffunctionαcomplex numberelement ofVset(ffunctionxvectorεpositive real number) \( {f}_{α}\in V\mathopen{}\left( f, x, ε\right)\mathclose{} \) eventually for all xvectorelement ofXnormed linear space \( x\in X \) and εpositive real number>greater than0zero \( ε\gt 0 \), which implies |modulusffunctionαcomplex number(xvector)-minusffunction(xvector)|modulus<less thanεpositive real number \( \mathopen{}\left\lvert{}{f}_{α}\mathopen{}\left( x\right)\mathclose{}-f\mathopen{}\left( x\right)\mathclose{}\right\rvert\mathclose{}\lt ε \) eventually for all xvectorelement ofXnormed linear space \( x\in X \) and εpositive real number>greater than0zero \( ε\gt 0 \), which implies ffunctionαcomplex number(xvector)converges toffunction(xvector) \( {f}_{α}\mathopen{}\left( x\right)\mathclose{} \to f\mathopen{}\left( x\right)\mathclose{} \) for all xvectorelement ofXnormed linear space \( x\in X \).

(⇐) Suppose ffunctionαcomplex number(xvector)converges toffunction(xvector) \( {f}_{α}\mathopen{}\left( x\right)\mathclose{} \to f\mathopen{}\left( x\right)\mathclose{} \) for all xvectorelement ofXnormed linear space \( x\in X \). Let xvectoriintegerelement ofXnormed linear space \( {x}_{i}\in X \) for iintegerelement of{set1one2twoninteger}set \( i\in \mathopen{}\left\{\, 1, 2, \dotsc, n\,\right\}\mathclose{} \) and let εpositive real number>greater than0zero \( ε\gt 0 \) be given. ffunctionαcomplex number(xvector)converges toffunction(xvector) \( {f}_{α}\mathopen{}\left( x\right)\mathclose{} \to f\mathopen{}\left( x\right)\mathclose{} \) for all xvector\( x \) implies ffunctionαcomplex number(xvectoriinteger)converges toffunction(xvectoriinteger) \( {f}_{α}\mathopen{}\left( {x}_{i}\right)\mathclose{} \to f\mathopen{}\left( {x}_{i}\right)\mathclose{} \) for all iinteger\( i \). This implies ffunctionαcomplex numberelement ofN neighborhood(ffunctionxvectoriintegerεpositive real number) \( {f}_{α}\in \mathop{\mathrm{N}}\mathopen{}\left( f, {x}_{i}, ε\right)\mathclose{} \) for all iinteger\( i \) eventually. Then for all iinteger\( i \), there exists αcomplex numberiinteger \( {α}_{i} \) such that for all αcomplex number>greater thanαcomplex numberiinteger \( α\gt {α}_{i} \), ffunctionαcomplex numberelement ofN neighborhood(ffunctionxvectoriintegerεpositive real number) \( {f}_{α}\in \mathop{\mathrm{N}}\mathopen{}\left( f, {x}_{i}, ε\right)\mathclose{} \). Pick βcomplex number>greater thanαcomplex numberiinteger \( β\gt {α}_{i} \) for all iinteger\( i \). Then ffunctionβcomplex numberelement ofintersectioniinteger=1oneninteger N neighborhood(ffunctionxvectoriintegerεpositive real number) =equalsN neighborhood(ffunctionxvector1onexvectornintegerεpositive real number) \( {f}_{β}\in \bigcap_{i=1}^{n}{} \mathop{\mathrm{N}}\mathopen{}\left( f, {x}_{i}, ε\right)\mathclose{} = \mathop{\mathrm{N}}\mathopen{}\left( f, {x}_{1}, \dotsc, {x}_{n}, ε\right)\mathclose{} \) which gives ffunctionαcomplex numberw*converges toffunction \( {f}_{α} \mathbin{\overset{\mathop{\mathrm{w}^*}}{\to}}f \).

Example III.33

Let Xnormed linear space=equalsL1Lebesgue space(Rreal numbers) \( X= \mathrm{L}^{\mathrm{1}}\mathopen{}\left( \mathbb{R}\right)\mathclose{} \). Then Xnormed linear space*=equalsLLebesgue space(Rreal numbers) \( X^{*}= \mathrm{L}^{\mathrm{∞}}\mathopen{}\left( \mathbb{R}\right)\mathclose{} \). For FL1 functionelement ofXnormed linear space \( F\in X \), set ffunctionninteger(FL1 function)=equalsintegralnintegerinfinityFL1 function \( {f}_{n}\mathopen{}\left( F\right)\mathclose{}= \int _{n}^{\infty}{}F \).

  1. ffunctionnintegerw*converges to0zero \( {f}_{n} \mathbin{\overset{\mathop{\mathrm{w}^*}}{\to}}0 \). As nintegerconverges toinfinity \( n \to \infty \), we have integralnintegerinfinityFL1 functionconverges to0zero \( \int _{n}^{\infty}{}F \to 0 \) for all FL1 functionelement ofL1Lebesgue space(Rreal numbers) \( F\in \mathrm{L}^{\mathrm{1}}\mathopen{}\left( \mathbb{R}\right)\mathclose{} \) by the dominated convergence theorem. This gives ffunctionninteger(FL1 function)converges to0zero \( {f}_{n}\mathopen{}\left( F\right)\mathclose{} \to 0 \) and ffunctionnintegerw*converges to0zero \( {f}_{n} \mathbin{\overset{\mathop{\mathrm{w}^*}}{\to}}0 \) by Lemma III.32.
  2. On the other hand, (sequenceffunctionninteger)sequence \( \mathopen{}\left({f}_{n}\right)\mathclose{} \) does not converge to 0zero\( 0 \) in τnorm topology(Xnormed linear space*) \( τ\mathopen{}\left( X^{*}\right)\mathclose{} \). Indeed, let FL1 functionninteger \( {F}_{n} \) be the indicator function of [intervalninteger, ninteger+plus1one]interval \( \mathopen{}\left[n, n+1\right]\mathclose{} \). Then ffunctionninteger(FL1 functionninteger)=equals1one=equalsFL1 functionninteger \( {f}_{n}\mathopen{}\left( {F}_{n}\right)\mathclose{}= 1= \mathopen{}\left\lVert{}{F}_{n}\right\rVert\mathclose{} \), so ffunctionnintegergreater than or equal to1one \( \mathopen{}\left\lVert{}{f}_{n}\right\rVert\mathclose{}\geq 1 \) for every ninteger\( n \).
  3. An even easier example of the same phenomenon is furnished by an orthonormal sequence (sequenceeunit vectorninteger)sequence \( \mathopen{}\left({e}_{n}\right)\mathclose{} \) in an infinite-dimensional Hilbert space HHilbert space\( H \). Define eunit vectornintegerelement ofHHilbert space* \( {e}_{n}\in {H}^{*} \) by eunit vectorninteger(xvector)=equalsxvector, eunit vectorninteger \( {e}_{n}\mathopen{}\left( x\right)\mathclose{}= \mathopen{}\left\langle{}x, {e}_{n}\right\rangle\mathclose{} \). Then eunit vectornintegerw*converges to0zero \( {e}_{n} \mathbin{\overset{\mathop{\mathrm{w}^*}}{\to}}0 \) by Bessel's inequality, while eunit vectorninteger=equals1one \( \mathopen{}\left\lVert{}{e}_{n}\right\rVert\mathclose{}= 1 \).

Theorem III.34 (Alaoglu's Theorem)

Let Xnormed linear space\( X \) be a normed linear space. Then the closed unit ball in Xnormed linear space* \( X^{*} \) is compact in w*\( \mathop{\mathrm{w}^*} \).

Proof. For xvectorelement ofXnormed linear space \( x\in X \), define Ixvector=equals{setzelement ofCcomplex numbers|such that |modulusz|modulusless than or equal toxvector }set \( {I}_{x}= \mathopen{}\left\{\, z\in \mathbb{C}\,\middle\vert\, , \mathopen{}\left\lvert{}z\right\rvert\mathclose{}\leq \mathopen{}\left\lVert{}x\right\rVert\mathclose{}, \,\right\}\mathclose{} \). Then Ixvector \( {I}_{x} \) is compact in Ccomplex numbers\( \mathbb{C} \). Let I=equalsproductxvectorelement ofXnormed linear spaceIxvector \( I= \prod_{x\in X}{}{I}_{x} \). I\( I \) is compact by the Tychonoff product theorem (with the product topology). Let Bball=equals{setffunctionalelement ofXnormed linear space*|such that ffunctionalless than or equal to1one }set \( B= \mathopen{}\left\{\, f\in X^{*}\,\middle\vert\, , \mathopen{}\left\lVert{}f\right\rVert\mathclose{}\leq 1, \,\right\}\mathclose{} \). Define J:mapsBballembeds inI \( J : B \hookrightarrow I \), ffunctionalis mapped to{setffunctional(xvector)|such that xvectorelement ofXnormed linear space }set \( f\mapsto \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, , x\in X, \,\right\}\mathclose{} \).

Claim 1: BballJ(Bball) \( B\simeq J\mathopen{}\left( B\right)\mathclose{} \). J\( J \) is injective, as J(ffunctional)=equalsJ(gfunctional) \( J\mathopen{}\left( f\right)\mathclose{}= J\mathopen{}\left( g\right)\mathclose{} \) implies {setffunctional(xvector)|such thatxvectorelement ofXnormed linear space}set=equals{setgfunctional(xvector)|such thatxvectorelement ofXnormed linear space}set \( \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{}= \mathopen{}\left\{\, g\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \) implies ffunctional=equalsgfunctional \( f= g \). To show J\( J \) is bicontinuous (both it and its inverse are continuous) onto J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \), let (sequenceffunctionalαcomplex number)sequence \( \mathopen{}\left({f}_{α}\right)\mathclose{} \) be a net in Bball\( B \) converging in w* \( \mathop{\mathrm{w}^*} \) to ffunctionalelement ofBball \( f\in B \). By Lemma III.32, this is if and only if ffunctionalαcomplex number(xvector)converges toffunctional(xvector) \( {f}_{α}\mathopen{}\left( x\right)\mathclose{} \to f\mathopen{}\left( x\right)\mathclose{} \) for all xvectorelement ofXnormed linear space \( x\in X \) if and only if {setffunctionalαcomplex number(xvector)|such thatxvectorelement ofXnormed linear space}setconverges to{setffunctional(xvector)|such thatxvectorelement ofXnormed linear space}set \( \mathopen{}\left\{\, {f}_{α}\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \to \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \) if and only if J(ffunctionalαcomplex number)converges toJ(ffunctional) \( J\mathopen{}\left( {f}_{α}\right)\mathclose{} \to J\mathopen{}\left( f\right)\mathclose{} \). This shows J\( J \) and J1inverse \( {J}^{-1} \) are continuous, and the claim is proven.

Thus Bball\( B \) is compact if and only if J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \) is compact if and only if J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \) is closed.

Claim 2: J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \) is closed. Let {setJ(ffunctionalαcomplex number)}set \( \mathopen{}\left\{\, J\mathopen{}\left( {f}_{α}\right)\mathclose{}\,\right\}\mathclose{} \) be a net in J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \) and suppose {setJ(ffunctionalαcomplex number)}setconverges to{setffunctional(xvector)|such thatxvectorelement ofXnormed linear space}set \( \mathopen{}\left\{\, J\mathopen{}\left( {f}_{α}\right)\mathclose{}\,\right\}\mathclose{} \to \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \). Then {setffunctionalαcomplex number(xvector)|such thatxvectorelement ofXnormed linear space}setconverges to{setffunctional(xvector)|such thatxvectorelement ofXnormed linear space}set \( \mathopen{}\left\{\, {f}_{α}\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \to \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, x\in X\,\right\}\mathclose{} \). We want to show ffunctionalelement ofBball \( f\in B \).

First, we have ffunctional(xvector+plusyvector)=equalsffunctional(xvector)+plusffunctional(yvector) \( f\mathopen{}\left( x+y\right)\mathclose{}= f\mathopen{}\left( x\right)\mathclose{}+f\mathopen{}\left( y\right)\mathclose{} \) and ffunctional(λcomplex numbertimesxvector)=equalsλcomplex numbertimesffunctional(xvector) \( f\mathopen{}\left( λx\right)\mathclose{}= λf\mathopen{}\left( x\right)\mathclose{} \) because these are limits of nets and thus we can use properties of nets in topological vector spaces. That is ffunctional(xvector+plusyvector)=equalslimlimitαcomplex number ffunctionalαcomplex number(xvector+plusyvector) , \[ f\mathopen{}\left( x+y\right)\mathclose{}= \lim_{α}{} {f}_{α}\mathopen{}\left( x+y\right)\mathclose{} \text{,} \] ffunctional(xvector)+plusffunctional(yvector)=equalslimlimitαcomplex number ffunctionalαcomplex number(xvector) +pluslimlimitαcomplex number ffunctionalαcomplex number(yvector) , \[ f\mathopen{}\left( x\right)\mathclose{}+f\mathopen{}\left( y\right)\mathclose{}= \lim_{α}{} {f}_{α}\mathopen{}\left( x\right)\mathclose{} +\lim_{α}{} {f}_{α}\mathopen{}\left( y\right)\mathclose{} \text{,} \] etc.

For xvectorelement ofXnormed linear space \( x\in X \), ffunctional(xvector)element ofIxvector \( f\mathopen{}\left( x\right)\mathclose{}\in {I}_{x} \) implies |modulusffunctional(xvector)|modulusless than or equal toxvector \( \mathopen{}\left\lvert{}f\mathopen{}\left( x\right)\mathclose{}\right\rvert\mathclose{}\leq \mathopen{}\left\lVert{}x\right\rVert\mathclose{} \) for all xvector\( x \). This gives ffunctional=equalssupsupremum xvectorless than or equal to1one |modulusffunctional(xvector)|modulusless than or equal to1one \( \mathopen{}\left\lVert{}f\right\rVert\mathclose{}= \sup_{ \mathopen{}\left\lVert{}x\right\rVert\mathclose{}\leq 1 }{}\mathopen{}\left\lvert{}f\mathopen{}\left( x\right)\mathclose{}\right\rvert\mathclose{}\leq 1 \). Thus we have ffunctionalelement ofBball \( f\in B \). From this, we get {setffunctional(xvector)|such that xvectorelement ofXnormed linear space }set=equalsJ(ffunctional)element ofJ(Bball) \( \mathopen{}\left\{\, f\mathopen{}\left( x\right)\mathclose{}\,\middle\vert\, , x\in X, \,\right\}\mathclose{}= J\mathopen{}\left( f\right)\mathclose{}\in J\mathopen{}\left( B\right)\mathclose{} \). Finally, then, J(Bball) \( J\mathopen{}\left( B\right)\mathclose{} \) is closed.


Previous: Trace Class Operators back to top Next: Trace-Class/Compact Duality