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

E. Constructions

Proposition I.78

Suppose HHilbert space\( H \) is a Hilbert space, Tclosed subspace1one \( {T}_{1} \) and Tclosed subspace2two \( {T}_{2} \) are closed orthogonal subspaces, and Tclosed subspace\( T \) is a closed subspace such that Tclosed subspace1onesubsetTclosed subspace \( {T}_{1}\subseteq T \). Then Tclosed subspace1oneTclosed subspace2two \( {T}_{1}\oplus {T}_{2} \) is closed, and there exists exactly one closed subspace, denoted Tclosed subspacedirect minusTclosed subspace1one \( T\ominus {T}_{1} \), such that (Tclosed subspaceTclosed subspace1one)subsetTclosed subspace \( \mathopen{}\left(T\ominus {T}_{1}\right)\mathclose{}\subseteq T \), (Tclosed subspaceTclosed subspace1one)Tclosed subspace1one \( \mathopen{}\left(T\ominus {T}_{1}\right)\mathclose{}\perp {T}_{1} \), and Tclosed subspace=equalsTclosed subspace1one(Tclosed subspaceTclosed subspace1one) \( T= {T}_{1}\oplus \mathopen{}\left(T\ominus {T}_{1}\right)\mathclose{} \). This subspace is called the orthogonal complement of Tclosed subspace1one \( {T}_{1} \) with respect to Tclosed subspace\( T \).

Proof. Exercise.

Exercise I.79

For HHilbert space\( H \) a Hilbert space and Tclosed subspacesubsetHHilbert space \( T\subseteq H \) closed, HHilbert spaceTclosed subspace=equalsTclosed subspace \( H\ominus T= {T}^{\perp} \).

Definition I.80

The external direct sum of a finite number of Hilbert spaces HHilbert space1one\( {H}_{1} \), …, HHilbert spaceninteger\( {H}_{n} \) with inner products ·, ·1one\( \mathopen{}\left\langle{}\cdot, \cdot\right\rangle\mathclose{}_{1} \), …, ·, ·ninteger\( \mathopen{}\left\langle{}\cdot, \cdot\right\rangle\mathclose{}_{n} \), written HHilbert space1onedirect sumdirect sumHHilbert spaceninteger \( {H}_{1}\oplus \dotsb\oplus {H}_{n} \), is the set of ninteger\( n \)-tuples {set(tuplexvector1one, , xvectorninteger)tuple|such that xvectoriintegerelement ofHHilbert spaceiinteger }set \( \mathopen{}\left\{\, \mathopen{}\left({x}_{1}, \dotsc, {x}_{n}\right)\mathclose{}\,\middle\vert\, , {x}_{i}\in {H}_{i}, \,\right\}\mathclose{} \).

Exercise I.81

The direct sum of Hilbert spaces is a Hilbert space with

  1. (tuplexvector1one, , xvectorninteger)tuple+plus(tupleyvector1one, , yvectorninteger)tuple=equals(tuplexvector1one+plusyvector1one, , xvectorninteger+plusyvectorninteger)tuple \( \mathopen{}\left({x}_{1}, \dotsc, {x}_{n}\right)\mathclose{}+\mathopen{}\left({y}_{1}, \dotsc, {y}_{n}\right)\mathclose{}= \mathopen{}\left({x}_{1}+{y}_{1}, \dotsc, {x}_{n}+{y}_{n}\right)\mathclose{} \),
  2. (tuplexvector1one, , xvectorninteger)tuple, (tupleyvector1one, , yvectorninteger)tuple=equalssummationiinteger=1oneninteger xvectoriinteger, yvectoriintegeriinteger \( \mathopen{}\left\langle{}\mathopen{}\left({x}_{1}, \dotsc, {x}_{n}\right)\mathclose{}, \mathopen{}\left({y}_{1}, \dotsc, {y}_{n}\right)\mathclose{}\right\rangle\mathclose{}= \sum_{i=1}^{n}{} \mathopen{}\left\langle{}{x}_{i}, {y}_{i}\right\rangle\mathclose{}_{i} \),
  3. (tuplexvector1one, , xvectorninteger)tuple=equals (summationiinteger=1oneninteger (xvectoriintegeriinteger) 2two ) 1one2two \( \mathopen{}\left\lVert{}\mathopen{}\left({x}_{1}, \dotsb, {x}_{n}\right)\mathclose{}\right\rVert\mathclose{}= {\mathopen{}\left(\sum_{i=1}^{n}{} {\mathopen{}\left(\mathopen{}\left\lVert{}{x}_{i}\right\rVert\mathclose{}_{i}\right)\mathclose{}}^{2} \right)\mathclose{}}^{\frac{1}{2}} \).

Remark I.82

It makes sense to use the same notation for the external direct sum HHilbert space=equalsHHilbert space1oneHHilbert spaceninteger \( H= {H}_{1}\oplus \dotsb\oplus {H}_{n} \) and the (internal) direct sum Kcompact operator=equalsHHilbert space1oneHHilbert spacenintegersubsetHHilbert space \( K= {H}_{1}\oplus \dotsb\oplus {H}_{n}\subseteq H \) since HHilbert space\( H \) is isomorphic to Kcompact operator\( K \) in the following way: Let Econvex setiinteger \( {E}_{i} \) be the orthogonal projection of Kcompact operator\( K \) onto the subspace HHilbert spaceiinteger \( {H}_{i} \). Note that summationiinteger=1onenintegerEconvex setiinteger=equalsI \( \sum_{i=1}^{n}{}{E}_{i}= I \). Now define U:mapsKcompact operatortoHHilbert space \( U : K \to H \) as xvectoris mapped to(tupleEconvex set1onetimesxvector, , Econvex setnintegertimesxvector)tuple \( x\mapsto \mathopen{}\left({E}_{1}x, \dotsc, {E}_{n}x\right)\mathclose{} \). U\( U \) is clearly surjective. U\( U \) preserves norms since U(xvector) 2two =equalssummationiinteger=1oneninteger Econvex setiinteger(xvector) 2two =equals (summationiinteger=1oneninteger Econvex setiinteger )(xvector) 2two =equals xvector 2two \( {\mathopen{}\left\lVert{}U\mathopen{}\left( x\right)\mathclose{}\right\rVert\mathclose{}}^{2}= \sum_{i=1}^{n}{} {\mathopen{}\left\lVert{}{E}_{i}\mathopen{}\left( x\right)\mathclose{}\right\rVert\mathclose{}}^{2} = {\mathopen{}\left\lVert{}\mathopen{}\left(\sum_{i=1}^{n}{} {E}_{i} \right)\mathclose{}\mathopen{}\left( x\right)\mathclose{}\right\rVert\mathclose{}}^{2}= {\mathopen{}\left\lVert{}x\right\rVert\mathclose{}}^{2} \).

Definition I.83

If Vvector space1one\( {V}_{1} \), …, Vvector spaceninteger\( {V}_{n} \) are vector spaces, a function on Vvector space1one×Cartesian product×Cartesian productVvector spaceninteger \( {V}_{1}\times \dotsb\times {V}_{n} \) is multilinear if it is linear on each Vvector spaceiinteger \( {V}_{i} \).

Theorem I.84

If Vvector space1one\( {V}_{1} \), …, Vvector spaceninteger\( {V}_{n} \), Vvector space\( V \) are vector spaces, there exists a unique (up to isomorphism) vector space Wvector space\( W \) and a map γmapping:mapsVvector space1one×Cartesian product×Cartesian productVvector spacenintegertoWvector space \( γ : {V}_{1}\times \dotsb\times {V}_{n} \to W \) with the property that if φmultilinear map:mapsVvector space1one×Cartesian product×Cartesian productVvector spacenintegertoVvector space \( φ : {V}_{1}\times \dotsb\times {V}_{n} \to V \) is a multilinear map, there exists a unique linear map ψlinear map:mapsWvector spacetoVvector space \( ψ : W \to V \) such that φmultilinear map=equalsψlinear mapcompositionγmapping \( φ= ψ\circ γ \).

Definition I.85

The vector space Wvector space\( W \) and mapping γmapping\( γ \) of Theorem I.84 is called the algebraic tensor product of Vvector space1one\( {V}_{1} \), …, Vvector spaceninteger\( {V}_{n} \) and is written Vvector space1onetensor producttensor productVvector spaceninteger \( {V}_{1}\otimes \dotsb\otimes {V}_{n} \). If xvectoriintegerelement ofVvector spaceiinteger \( {x}_{i}\in {V}_{i} \), then γmapping((xvector1one, , xvectorninteger)) \( γ\mathopen{}\left( \mathopen{}\left({x}_{1}, \dotsc, {x}_{n}\right)\mathclose{}\right)\mathclose{} \) is written xvector1onexvectorninteger \( {x}_{1}\otimes \dotsb\otimes {x}_{n} \).

Although the simple tensors xvector1onexvectorninteger \( {x}_{1}\otimes \dotsb\otimes {x}_{n} \) are the image of γmapping\( γ \), they are not the entire tensor product space. Elements of Vvector space1oneVvector spaceninteger \( {V}_{1}\otimes \dotsb\otimes {V}_{n} \), may, however, be regarded as (not always unique) sums of simple tensors.

Remark I.86

Alternately, the algebraic tensor product may be thought of as the quotient space (Vvector space1one×Cartesian product×Cartesian productVvector spaceninteger)/Nsubspace \( \mathopen{}\left({V}_{1}\times \dotsb\times {V}_{n}\right)\mathclose{}/N \) where Nsubspace\( N \) is the subspace spanned by elements of multilinear form.

Exercise I.87

If Vvector space\( V \) and Wvector space\( W \) are vector spaces with xvector1one \( {x}_{1} \) and xvector2two \( {x}_{2} \) in Vvector space\( V \) and with yvector1one \( {y}_{1} \) and yvector2two \( {y}_{2} \) in Wvector space\( W \), the following hold for λcomplex numberelement ofCcomplex numbers \( λ\in \mathbb{C} \):

  1. (λcomplex numbertimesxvector1one)yvector1one=equalsxvector1one(λcomplex numbertimesyvector1one)=equalsλcomplex numbertimes(xvector1oneyvector1one) \( \mathopen{}\left(λ{x}_{1}\right)\mathclose{}\otimes {y}_{1}= {x}_{1}\otimes \mathopen{}\left(λ{y}_{1}\right)\mathclose{}= λ\mathopen{}\left({x}_{1}\otimes {y}_{1}\right)\mathclose{} \).
  2. (xvector1one+plusxvector2two)yvector1one=equalsxvector1oneyvector1one+plusxvector2twoyvector1one \( \mathopen{}\left({x}_{1}+{x}_{2}\right)\mathclose{}\otimes {y}_{1}= {x}_{1}\otimes {y}_{1}+{x}_{2}\otimes {y}_{1} \).
  3. xvector1one(yvector1one+plusyvector2two)=equalsxvector1oneyvector1one+plusxvector1oneyvector2two \( {x}_{1}\otimes \mathopen{}\left({y}_{1}+{y}_{2}\right)\mathclose{}= {x}_{1}\otimes {y}_{1}+{x}_{1}\otimes {y}_{2} \).

In general, if Xnormed linear space\( X \) and Ynormed linear space\( Y \) are normed spaces, there are a number of possible norms that may be defined on Xnormed linear spaceYnormed linear space \( X\otimes Y \) (see Istratescu [4] and Murphy [8]). In the case of Hilbert spaces, however, a natural inner product helps simplify the situation.

Theorem I.88

If (tupleHHilbert space, ·, ·HHilbert space)tuple \( \mathopen{}\left(H, \mathopen{}\left\langle{}\cdot, \cdot\right\rangle\mathclose{}_{H}\right)\mathclose{} \) and (tupleKHilbert space, ·, ·KHilbert space)tuple \( \mathopen{}\left(K, \mathopen{}\left\langle{}\cdot, \cdot\right\rangle\mathclose{}_{K}\right)\mathclose{} \) are Hilbert spaces, then HHilbert spaceKHilbert space \( H\otimes K \) is an inner product space with inner product hvector1onekvector1one, hvector2twokvector2two=equalshvector1one, hvector2twoHHilbert spacetimeskvector1one, kvector2twoKHilbert space \[ \mathopen{}\left\langle{}{h}_{1}\otimes {k}_{1}, {h}_{2}\otimes {k}_{2}\right\rangle\mathclose{}= \mathopen{}\left\langle{}{h}_{1}, {h}_{2}\right\rangle\mathclose{}_{H}\mathopen{}\left\langle{}{k}_{1}, {k}_{2}\right\rangle\mathclose{}_{K} \] and induced norm hvector1onekvector1one=equalshvector1oneHHilbert spacetimeskvector1oneKHilbert space . \[ \mathopen{}\left\lVert{}{h}_{1}\otimes {k}_{1}\right\rVert\mathclose{}= \mathopen{}\left\lVert{}{h}_{1}\right\rVert\mathclose{}_{H}\mathopen{}\left\lVert{}{k}_{1}\right\rVert\mathclose{}_{K} \text{.} \]

Proof. Follows from the algebraic properties of simple tensors and complex conjugation.

Definition I.89

If HHilbert space\( H \) and KHilbert space\( K \) are Hilbert spaces, the Hilbert space derived by completion of HHilbert spaceKHilbert space \( H\otimes K \) with respect to the norm induced by the inner product of Theorem Theorem I.88 is called the Hilbert space tensor product of HHilbert space\( H \) and KHilbert space\( K \), and is denoted HHilbert space(complete) tensor productKHilbert space \( H\mathbin{\hat{\otimes}} K \).

Remark I.90

Sometimes the notation HHilbert spacealgebraic tensor productKHilbert space \( H\odot K \) is used for the algebraic tensor product, and HHilbert spaceKHilbert space \( H\otimes K \) for its completion.

Example I.91

Let μmeasure1one \( {μ}_{1} \) and μmeasure2two \( {μ}_{2} \) be measures on Rreal numbers\( \mathbb{R} \). The algebraic tensor product L2Lebesgue space(Rreal numbersμmeasure1one)L2Lebesgue space(Rreal numbersμmeasure2two) \( \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{1}\right)\mathclose{}\otimes \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{2}\right)\mathclose{} \) is composed of equivalence classes of functions, square integrable on Rreal numbers2two \( {\mathbb{R}}^{2} \) with respect to the measure μmeasure1onetimesμmeasure2two \( {μ}_{1}{μ}_{2} \). For ffunction\( f \) and ggroup element\( g \) in L2Lebesgue space(Rreal numbersμmeasure1one)L2Lebesgue space(Rreal numbersμmeasure2two) \( \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{1}\right)\mathclose{}\otimes \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{2}\right)\mathclose{} \), ffunction, ggroup element=equalsintegral integral ffunction(xvectoryvector) ¯complex conjugatetimesggroup element(xvectoryvector) d μmeasure1one(xvector) d μmeasure2two(yvector) \[ \mathopen{}\left\langle{}f, g\right\rangle\mathclose{}= \int {} \int {} \overline{ f\mathopen{}\left( x, y\right)\mathclose{} }g\mathopen{}\left( x, y\right)\mathclose{} \,\mathrm{d} {μ}_{1}\mathopen{}\left( x\right)\mathclose{} \,\mathrm{d} {μ}_{2}\mathopen{}\left( y\right)\mathclose{} \] Since L2Lebesgue space(Rreal numbersμmeasure1one)L2Lebesgue space(Rreal numbersμmeasure2two) \( \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{1}\right)\mathclose{}\otimes \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{2}\right)\mathclose{} \) contains all step functions on Rreal numbers2two \( {\mathbb{R}}^{2} \), L2Lebesgue space(Rreal numbersμmeasure1one)L2Lebesgue space(Rreal numbersμmeasure2two)L2Lebesgue space(Rreal numbers2twoμmeasure1one×Cartesian productμmeasure2two). \[ \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{1}\right)\mathclose{}\mathbin{\hat{\otimes}} \mathrm{L}^{\mathrm{2}}\mathopen{}\left( \mathbb{R}, {μ}_{2}\right)\mathclose{}\simeq \mathrm{L}^{\mathrm{2}}\mathopen{}\left( {\mathbb{R}}^{2}, {μ}_{1}\times {μ}_{2}\right)\mathclose{}\text{.} \]

Exercise I.92

If MmatrixHHilbert space \( {M}_{H} \) and MmatrixKHilbert space \( {M}_{K} \) are dense subspaces of HHilbert space\( H \) and KHilbert space\( K \) respectively, then MmatrixHHilbert spaceMmatrixKHilbert space \( {M}_{H}\otimes {M}_{K} \) is dense in HHilbert spaceKHilbert space \( H\mathbin{\hat{\otimes}} K \).

Exercise I.93

If Eorthonormal basisHHilbert space \( {E}_{H} \) and Eorthonormal basisKHilbert space \( {E}_{K} \) are orthonormal bases for HHilbert space\( H \) and KHilbert space\( K \) respectively, then {seteunit vectorHHilbert spaceeunit vectorKHilbert space|such thateunit vectorHHilbert spaceelement ofEorthonormal basisHHilbert spaceeunit vectorKHilbert spaceelement ofEorthonormal basisKHilbert space}set \( \mathopen{}\left\{\, {e}_{H}\otimes {e}_{K}\,\middle\vert\, {e}_{H}\in {E}_{H}, {e}_{K}\in {E}_{K}\,\right\}\mathclose{} \) is an orthonormal basis for HHilbert spaceKHilbert space \( H\mathbin{\hat{\otimes}} K \). What does this say about the dimension of HHilbert spaceKHilbert space \( H\mathbin{\hat{\otimes}} K \) ?

Exercise I.94

If HHilbert space\( H \) and KHilbert space\( K \) are non-trivial, then HHilbert spaceKHilbert space \( H\mathbin{\hat{\otimes}} K \) is separable if and only if HHilbert space\( H \) and KHilbert space\( K \) are separable.

Exercise I.95

The inner product space defined in Theorem I.88 is complete if and only if HHilbert space\( H \) or KHilbert space\( K \) is finite dimensional.


Previous: Orthonormal Sets back to top Next: Bounded Operators