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

B. Orthogonal Projection

The key to many computations to follow in this chapter can be found in Exercise I.45.

Exercise I.45

Prove that in an inner product space, xvector+plusyvector2two=equalsxvector2two+plus2twotimesRereal part(xvector, yvector)+plus yvector2two \( {\mathopen{}\left\lVert{}x+y\right\rVert\mathclose{}}^{2}= {\mathopen{}\left\lVert{}x\right\rVert\mathclose{}}^{2}+2\Re\mathopen{}\left( \mathopen{}\left\langle{}x, y\right\rangle\mathclose{}\right)\mathclose{}+{\mathopen{}\left\lVert{}y\right\rVert\mathclose{}}^{2} \).

Theorem I.46 (Parallelogram Law)

For any xvector\( x \) and yvector\( y \) in a normed linear space Xnormed linear space\( X \), xvector+plusyvector2two+plusxvector-minusyvector2two=equals2twotimes(xvector2two+plusyvector2two)\( {\mathopen{}\left\lVert{}x+y\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}}^{2}= 2\mathopen{}\left({\mathopen{}\left\lVert{}x\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}y\right\rVert\mathclose{}}^{2}\right)\mathclose{} \).

Theorem I.47

For any xvector\( x \) and yvector\( y \) in a normed linear space Xnormed linear space\( X \), if xvector, yvector=equals0zero\( \mathopen{}\left\langle{}x, y\right\rangle\mathclose{}= 0 \) then xvector+plusyvector2two=equalsxvector2two+plusyvector2two\( {\mathopen{}\left\lVert{}x+y\right\rVert\mathclose{}}^{2}= {\mathopen{}\left\lVert{}x\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}y\right\rVert\mathclose{}}^{2} \).

Definition I.48

For an element xvector\( x \) and a subset Ssubset\( S \) of a normed linear space Xnormed linear space\( X \), the distance from xvector\( x \) to Ssubset\( S \) in Xnormed linear space\( X \), written ddistance(xvectorSsubset)\( \operatorname{d}\mathopen{}\left( x, S\right)\mathclose{} \), is infinfimumyvectorelement ofSsubsetd(xvectoryvector)\( \inf_{y\in S}{}\operatorname{d}\mathopen{}\left( x, y\right)\mathclose{} \).

Theorem I.49

For a closed convex (non-empty) subset Econvex set\( E \) of a Hilbert space HHilbert space\( H \) and a vector xvector\( x \) in HHilbert space\( H \), there exists a vector yvector\( y \) in Econvex set\( E \) such that xvector-minusyvector<less thanxvector-minuswvector\( \mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}\lt \mathopen{}\left\lVert{}x-w\right\rVert\mathclose{} \) for all wvector\( w \) in Econvex set\( E \) distinct from yvector\( y \). That is, there exists a unique point in Econvex set\( E \) nearest to xvector\( x \).

Proof. Let (sequenceyvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left({y}_{n}\right)\mathclose{}_{n=1}^{\infty} \) be a sequence of elements of Econvex set\( E \) such that (sequencexvector-minusyvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left(\mathopen{}\left\lVert{}x-{y}_{n}\right\rVert\mathclose{}\right)\mathclose{}_{n=1}^{\infty} \) converges to d(xvectorEconvex set)\( \operatorname{d}\mathopen{}\left( x, E\right)\mathclose{} \). We claim that (sequenceyvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left({y}_{n}\right)\mathclose{}_{n=1}^{\infty} \) is Cauchy. Indeed, using Theorem I.46, 2twotimes(xvector-minusyvectorninteger2two+plusxvector-minusyvectorminteger2two)=equalsyvectorminteger-minusyvectorninteger2two+plus2twotimes(xvector-minusyvectorminteger+plusyvectorninteger 2two)2twogreater than or equal toyvectorminteger-minusyvectorninteger2two+plus4fourtimesd(xvectorEconvex set)2two . \[ 2\mathopen{}\left({\mathopen{}\left\lVert{}x-{y}_{n}\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}x-{y}_{m}\right\rVert\mathclose{}}^{2}\right)\mathclose{}= {\mathopen{}\left\lVert{}{y}_{m}-{y}_{n}\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}2\mathopen{}\left(x-\frac{{y}_{m}+{y}_{n}}{2}\right)\mathclose{}\right\rVert\mathclose{}}^{2}\geq {\mathopen{}\left\lVert{}{y}_{m}-{y}_{n}\right\rVert\mathclose{}}^{2}+4{\operatorname{d}\mathopen{}\left( x, E\right)\mathclose{}}^{2} \text{.} \] Note that yvectorminteger+plusyvectorninteger 2two\( \frac{{y}_{m}+{y}_{n}}{2} \) is in Econvex set\( E \) because Econvex set\( E \) is convex. So yvectorminteger-minusyvectorninteger2twoless than or equal to2twotimes((xvector-minusyvectorninteger2two-minusd(xvectorEconvex set)2two)+plus(xvector-minusyvectorminteger2two-minusd(xvectorEconvex set)2two)) \[ {\mathopen{}\left\lVert{}{y}_{m}-{y}_{n}\right\rVert\mathclose{}}^{2}\leq 2\mathopen{}\left(\mathopen{}\left({\mathopen{}\left\lVert{}x-{y}_{n}\right\rVert\mathclose{}}^{2}-{\operatorname{d}\mathopen{}\left( x, E\right)\mathclose{}}^{2}\right)\mathclose{}+\mathopen{}\left({\mathopen{}\left\lVert{}x-{y}_{m}\right\rVert\mathclose{}}^{2}-{\operatorname{d}\mathopen{}\left( x, E\right)\mathclose{}}^{2}\right)\mathclose{}\right)\mathclose{} \] and it follows that (sequenceyvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left({y}_{n}\right)\mathclose{}_{n=1}^{\infty} \) is Cauchy.

As a Hilbert space, HHilbert space\( H \) is complete, so the sequence (sequenceyvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left({y}_{n}\right)\mathclose{}_{n=1}^{\infty} \) converges to some element yvector\( y \) of HHilbert space\( H \). Because Econvex set\( E \) is closed, yvectorelement ofEconvex set\( y\in E \). Thus xvector-minusyvector=equalsd(xvectorEconvex set)\( \mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}= \operatorname{d}\mathopen{}\left( x, E\right)\mathclose{} \), and for any element wvector\( w \) of Econvex set\( E \) other than yvector\( y \), 2twotimes(xvector-minusyvector2two+plusxvector-minuswvector2two)=equalswvector-minusyvector2two+plus4fourtimes xvector-minuswvector+plusyvector2two 2two>greater than4fourtimesd(xvectorEconvex set)2two . \[ 2\mathopen{}\left({\mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}}^{2}+{\mathopen{}\left\lVert{}x-w\right\rVert\mathclose{}}^{2}\right)\mathclose{}= {\mathopen{}\left\lVert{}w-y\right\rVert\mathclose{}}^{2}+4{\mathopen{}\left\lVert{}x-\frac{w+y}{2}\right\rVert\mathclose{}}^{2}\gt 4{\operatorname{d}\mathopen{}\left( x, E\right)\mathclose{}}^{2} \text{.} \] So 2twotimesxvector-minuswvector2two>greater than2twotimesd(xvectorEconvex set)2two \( 2{\mathopen{}\left\lVert{}x-w\right\rVert\mathclose{}}^{2}\gt 2{\operatorname{d}\mathopen{}\left( x, E\right)\mathclose{}}^{2} \) and xvector-minuswvector>greater thand(xvectorEconvex set) \( \mathopen{}\left\lVert{}x-w\right\rVert\mathclose{}\gt \operatorname{d}\mathopen{}\left( x, E\right)\mathclose{} \).

Definition I.50

Let Wclosed subspace\( W \) be a closed subspace of a Hilbert space HHilbert space\( H \). The orthogonal projection of an element xvectorelement ofHHilbert space\( x\in H \) on Wclosed subspace\( W \) is the point in Wclosed subspace\( W \) nearest to xvector\( x \). The operator that maps each element of HHilbert space\( H \) to its orthogonal projection on Wclosed subspace\( W \) is called the projection of HHilbert space\( H \) on Wclosed subspace\( W \).

Exercise I.51

Show that the projection of a Hilbert space onto a closed subspace is a linear operator.

Definition I.52

For an inner product space Vvector space\( V \), elements xvector\( x \) and yvector\( y \) are orthogonal, denoted xvectororthogonalyvector\( x\perp y \), if xvector, yvector=equals0zero\( \mathopen{}\left\langle{}x, y\right\rangle\mathclose{}= 0 \). .

Definition I.53

Let Ssubset \( S \) be a subset of an inner product space Vinner product space\( V \). The orthogonal complement of Ssubset\( S \) in Vinner product space\( V \), written Ssubsetorthogonal complement\( {S}^{\perp} \), is the set of all vectors xvector\( x \) in Vinner product space\( V \) such that xvectoryvector\( x\perp y \) for all vectors yvector\( y \) in Ssubset \( S \).

Exercise I.54

Prove that for any subset Ssubset \( S \) of an inner product space Vinner product space\( V \), Ssubset\( {S}^{\perp} \) is a subspace of Vinner product space\( V \).

Exercise I.55

Prove that for any subset Ssubset \( S \) of an inner product space Vinner product space\( V \), Ssubsetsubset(Ssubset)\( S\subseteq {\mathopen{}\left( {S}^{\perp}\right)\mathclose{}}^{\perp} \).

Exercise I.56

Prove that for any subspace Wsubspace \( W \) of an inner product space Vinner product space\( V \), WsubspaceintersectionWsubspace=equals{set0zero}set\( W\cap {W}^{\perp}= \mathopen{}\left\{\, 0\,\right\}\mathclose{} \).

Proposition I.57

For any subset Ssubset \( S \) of an inner product space Vinner product space\( V \), Ssubset\( {S}^{\perp} \) is a closed subspace of Vinner product space\( V \).

Proof. The subset Ssubset\( {S}^{\perp} \) is a subspace by Exercise I.54. To show that it is closed, consider a sequence (sequencexvectorninteger)sequenceninteger=1oneinfinity\( \mathopen{}\left({x}_{n}\right)\mathclose{}_{n=1}^{\infty} \) of elements of Ssubset\( {S}^{\perp} \) that converges in Vinner product space\( V \) to a vector xvector\( x \). Then using Proposition I.3, |modulusxvector, yvector|modulus=equals|modulusxvector-minusxvectorninteger, yvector|modulusless than or equal toyvectortimesxvector-minusxvectorninteger \[ \mathopen{}\left\lvert{}\mathopen{}\left\langle{}x, y\right\rangle\mathclose{}\right\rvert\mathclose{}= \mathopen{}\left\lvert{}\mathopen{}\left\langle{}x-{x}_{n}, y\right\rangle\mathclose{}\right\rvert\mathclose{}\leq \mathopen{}\left\lVert{}y\right\rVert\mathclose{}\mathopen{}\left\lVert{}x-{x}_{n}\right\rVert\mathclose{} \] for all yvector\( y \) in Ssubset\( S \). Taking limits on each side completes the proof.

Definition I.58

A vector space Vvector space\( V \) is the direct sum of two subspaces Usubspace\( U \) and Wsubspace\( W \), which we write as Vvector space=equalsUsubspacedirect sumWsubspace\( V= U\oplus W \), if each vector vcomplex number\( v \) in Vvector space\( V \) can be written uniquely as the sum of a vector xvector\( x \) in Usubspace\( U \) and a vector yvector\( y \) in Wsubspace\( W \).

Theorem I.59

For any closed subspace Wclosed subspace\( W \) of a Hilbert space HHilbert space\( H \), we have HHilbert space=equalsWclosed subspaceWclosed subspace\( H= W\oplus {W}^{\perp} \).

Proof. Take an element xvector\( x \) in the Hilbert space HHilbert space\( H \). By Theorem I.49 there exists a point yvector\( y \) in Wclosed subspace\( W \) nearest to xvector\( x \). In particular, xvector-minusyvector=equalsminminimumwvectorelement ofWclosed subspacexvector-minuswvector\( \mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}= \min_{w\in W}{}\mathopen{}\left\lVert{}x-w\right\rVert\mathclose{} \). We claim the orthogonal decomposition we seek is xvector=equalsyvector+plus(xvector-minusyvector)\( x= y+\mathopen{}\left(x-y\right)\mathclose{} \).

We must first show that xvector-minusyvector\( x-y \) is in Wclosed subspace\( {W}^{\perp} \). Indeed, for wvectorelement ofWclosed subspace\( w\in W \) and εpositive real number>greater than0zero\( ε\gt 0 \), xvector-minusyvector+plusεpositive real numbertimeswvector2two=equalsxvector-minusyvector2two+plus2twotimesεpositive real numbertimesRereal part(xvector-minusyvector, wvector)+plusεpositive real number2twotimeswvector2twogreater than or equal toxvector-minusyvector2two \[ {\mathopen{}\left\lVert{}x-y+εw\right\rVert\mathclose{}}^{2}= {\mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}}^{2}+2ε\Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}+{ε}^{2}{\mathopen{}\left\lVert{}w\right\rVert\mathclose{}}^{2}\geq {\mathopen{}\left\lVert{}x-y\right\rVert\mathclose{}}^{2} \] since yvector-minusεpositive real numbertimeswvector\( y-εw \) is in Wclosed subspace\( W \). So, for εpositive real number>greater than0zero\( ε\gt 0 \), 2twotimesεpositive real numbertimesRereal part(xvector-minusyvector, wvector)+plusεpositive real number2twotimeswvector2twogreater than or equal to0zero \( 2ε\Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}+{ε}^{2}{\mathopen{}\left\lVert{}w\right\rVert\mathclose{}}^{2}\geq 0 \). Dividing by εpositive real number\( ε \) gives 2twotimesRereal part(xvector-minusyvector, wvector)+plusεpositive real numbertimeswvector2twogreater than or equal to0zero \( 2\Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}+ε{\mathopen{}\left\lVert{}w\right\rVert\mathclose{}}^{2}\geq 0 \). Thus Rereal part(xvector-minusyvector, wvector)greater than or equal to0zero \( \Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}\geq 0 \) for all wvectorelement ofWclosed subspace\( w\in W \). Repeating the same argument with wvector\( {-}w \) in place of wvector\( w \) leads to the conclusion Rereal part(xvector-minusyvector, wvector)less than or equal to0zero \( \Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}\leq 0 \) for all wvectorelement ofWclosed subspace\( w\in W \), meaning, in fact, Rereal part(xvector-minusyvector, wvector)=equals0zero \( \Re\mathopen{}\left( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}\right)\mathclose{}= 0 \) for all wvectorelement ofWclosed subspace\( w\in W \). Repeating the above with iimaginary unittimeswvector\( \mathrm{i}w \) and iimaginary unittimeswvector\( {-}\mathrm{i}w \) shows that xvector-minusyvector, wvector=equals0zero \( \mathopen{}\left\langle{}x-y, w\right\rangle\mathclose{}= 0 \) for all wvectorelement ofWclosed subspace\( w\in W \). Thus, xvector-minusyvector\( x-y \) is in Wclosed subspace\( {W}^{\perp} \).

By Exercise I.56, the decomposition xvector=equalsyvector+plus(xvector-minusyvector)\( x= y+\mathopen{}\left(x-y\right)\mathclose{} \) is unique (since two different decompositions would give you something non-zero in Wclosed subspaceintersectionWclosed subspace\( W\cap {W}^{\perp} \)). Thus HHilbert space=equalsWclosed subspaceWclosed subspace\( H= W\oplus {W}^{\perp} \).

Exercise I.60

For Ssubset\( S \) a subset of an inner product space Vinner product space\( V \), Ssubset=equalsspanspan(Ssubset)=equalsspanspan(Ssubset)¯\( {S}^{\perp}= {\operatorname{span}\mathopen{}\left( S\right)\mathclose{}}^{\perp}= {\overline{\operatorname{span}\mathopen{}\left( S\right)\mathclose{}}}^{\perp} \).

Proposition I.61

For a subset Ssubset \( S \) of a Hilbert space HHilbert space\( H \), (Ssubset)=equalsspanspan(Ssubset)¯ \( {\mathopen{}\left( {S}^{\perp}\right)\mathclose{}}^{\perp}= \overline{\operatorname{span}\mathopen{}\left( S\right)\mathclose{}} \).

Proof. Let Wclosed subspace=equalsspanspan(Ssubset)¯\( W= \overline{\operatorname{span}\mathopen{}\left( S\right)\mathclose{}} \) so that Wclosed subspace\( W \) is a closed subspace of HHilbert space\( H \). We claim that Wclosed subspace=equals(Wclosed subspace)\( W= {\mathopen{}\left( {W}^{\perp}\right)\mathclose{}}^{\perp} \). Indeed, Wclosed subspacesubset(Wclosed subspace)\( W\subseteq {\mathopen{}\left( {W}^{\perp}\right)\mathclose{}}^{\perp} \) by Exercise I.55. Using Theorem I.59, for xvectorelement of(Wclosed subspace)\( x\in {\mathopen{}\left( {W}^{\perp}\right)\mathclose{}}^{\perp} \), write xvector=equalsyvector+pluswvector\( x= y+w \) where wvectorelement ofWclosed subspace\( w\in W \) and yvectorelement ofWclosed subspace\( y\in {W}^{\perp} \). Then xvector, yvector=equals0zero\( \mathopen{}\left\langle{}x, y\right\rangle\mathclose{}= 0 \), and so xvector, yvector=equalswvector+plusyvector, yvector=equalswvector, yvector+plusyvector, yvector=equals0zero+plusyvector, yvector . \[ \mathopen{}\left\langle{}x, y\right\rangle\mathclose{}= \mathopen{}\left\langle{}w+y, y\right\rangle\mathclose{}= \mathopen{}\left\langle{}w, y\right\rangle\mathclose{}+\mathopen{}\left\langle{}y, y\right\rangle\mathclose{}= 0+\mathopen{}\left\langle{}y, y\right\rangle\mathclose{} \text{.} \] So yvector, yvector=equals0zero\( \mathopen{}\left\langle{}y, y\right\rangle\mathclose{}= 0 \) and thus yvector=equals0zero\( y= 0 \). Hence xvector=equalswvector\( x= w \), so xvectorelement ofWclosed subspace\( x\in W \), and thus Wclosed subspace=equals(Wclosed subspace)\( W= {\mathopen{}\left( {W}^{\perp}\right)\mathclose{}}^{\perp} \). Finally, since Wclosed subspace=equalsSsubset\( {W}^{\perp}= {S}^{\perp} \) by Exercise I.60, we have (Ssubset)=equalsspanspan(Ssubset)¯ \( {\mathopen{}\left( {S}^{\perp}\right)\mathclose{}}^{\perp}= \overline{\operatorname{span}\mathopen{}\left( S\right)\mathclose{}} \).


Previous: Completeness back to top Next: Finite Dimensional Spaces