Riesz's lemma

In mathematics, Riesz's lemma (after Frigyes Riesz) is a lemma in functional analysis. It specifies (often easy to check) conditions that guarantee that a subspace in a normed vector space is dense. The lemma may also be called the Riesz lemma or Riesz inequality. It can be seen as a substitute for orthogonality when the normed space is not an inner product space.

Statement

If $X$ is a reflexive Banach space then this conclusion is also true when $\alpha =1.$ ^[2]

Metric reformulation

As usual, let $d(x,y):=\|x-y\|$ denote the canonical metric induced by the norm, call the set $\{x\in X:\|x\|=1\}$ of all vectors that are a distance of $1$ from the origin the unit sphere, and denote the distance from a point $u$ to the set $Y\subseteq X$ by $d(u,Y)~:=~\inf _{y\in Y}d(u,y)~=~\inf _{y\in Y}\|u-y\|.$ The inequality $\alpha \leq d(u,Y)$ holds if and only if $\|u-y\|\geq \alpha$ for all $y\in Y,$ and it formally expresses the notion that the distance between $u$ and $Y$ is at least $\alpha .$ Because every vector subspace (such as $Y$ ) contains the origin $0,$ substituting $y:=0$ in this infimum shows that $d(u,Y)\leq \|u\|$ for every vector $u\in X.$ In particular, $d(u,Y)\leq 1$ when $\|u\|=1$ is a unit vector.

Using this new notation, the conclusion of Riesz's lemma may be restated more succinctly as: $d(u,Y)\geq \alpha$ holds for some unit vector $u\in X.$

Using this new terminology, Riesz's lemma may also be restated in plain English as:

Given any closed proper vector subspace of a normed space

X,

for any desired minimum distance

\alpha

less than

1,

there exists some vector in the unit sphere of

X

that is at least this desired distance away from the subspace.

Special cases

When $X=\{0\}$ , it is a trivial vector space, so then it has no proper vector subspace $Y,$ and so Riesz's lemma holds vacuously for all real numbers $\alpha \in \mathbb {R} .$ The remainder of this section will assume that $X\neq \{0\},$ which guarantees that a unit vector exists.

The inclusion of the hypotheses $0<\alpha <1$ can be explained by considering the three cases: $\alpha \leq 0$ , $\alpha =1,$ and $\alpha >1.$ The lemma holds when $\alpha \leq 0$ since every unit vector $u\in X$ satisfies the conclusion $\alpha \leq 0\leq d(u,Y)\leq 1=\|u\|.$ The hypotheses $0<\alpha$ is included solely to exclude this trivial case and is sometimes omitted from the lemma's statement.

Riesz's lemma is always false when $\alpha >1$ because for every unit vector $u\in X,$ the required inequality $\|u-y\|\geq \alpha$ fails to hold for $y:=0\in Y$ (since $\|u-0\|=1<\alpha$ ). Another consequence of $d(u,Y)>1$ being impossible is that the inequality $d(u,Y)\geq 1$ holds if and only if equality $d(u,Y)=1$ holds.

Proof

The proof^[3] can be found in functional analysis texts such as Kreyszig.^[4]^[5]

Proof

Consider any $v\in X-Y$ and denote its distance from $Y$ by $a$ . Clearly, $a>0$ since $Y$ is closed. Take any $\alpha \in (0,1)$ . By the definition of an infimum, there is a $y_{0}\in Y$ such that

a\leq \left\|v-y_{0}\right\|\leq {\frac {a}{\alpha }}

1

(note that $a/\alpha >a$ since $0<\alpha <1$ ). Let

z=c\left(v-y_{0}\right)\quad {\text{ where }}\quad c={\frac {1}{\left\|v-y_{0}\right\|}}.

Then $\|z\|=1$ , and we show that $\|z-y\|\geq \alpha$ for every $y\in Y$ . We have

{\begin{aligned}\|z-y\|&=\left\|c\left(v-y_{0}\right)-y\right\|\\&=c\left\|v-y_{0}-c^{-1}y\right\|\\&=c\left\|v-y_{1}\right\|\end{aligned}}

where

y_{1}=y_{0}+c^{-1}y.

The form of $y_{1}$ shows that $y_{1}\in Y$ . Hence $\left\|v-y_{1}\right\|\geq a$ , by the definition of $a$ . Writing $c$ out and using (1), we obtain

\|z-y\|=c\left\|v-y_{1}\right\|\geq ca={\frac {a}{\left\|v-y_{0}\right\|}}\geq {\frac {a}{a/\alpha }}=\alpha .

Since $y\in Y$ was arbitrary, this completes the proof.

Reflexivity

If we set $\alpha =1$ , then the statement of Riesz’s lemma becomes:

For every closed proper vector subspace

Y

of

X,

there exists some vector

u

of unit norm that satisfies

d(u,Y)=1.

When $X$ is a Banach space, then this statement is true if and only if $X$ is a reflexive space.^[2] That is, a Banach space $X$ is reflexive if and only if for every closed proper vector subspace $Y,$ there is some vector $u$ on the unit sphere of $X$ that is always at least a distance of $1=d(u,Y)$ away from the subspace. This gives a geometrically intuitive definition for reflexivity.

Examples

Since every finite dimensional normed space is a reflexive Banach space, Riesz’s lemma does hold for $\alpha =1$ when the normed space is finite-dimensional, as will now be shown. When the dimension of $X$ is finite then the closed unit ball $B\subseteq X$ is compact. Since the distance function $d(\cdot ,Y)$ is continuous, its image on the closed unit ball $B$ must be a compact subset of the real line, proving the claim. The "perpendicular" vector may be found pictorially by drawing a unit sphere that is supported by $Y$ at the origin.

For example, if the reflexive Banach space $X=\mathbb {R} ^{3}$ is endowed with the usual $\|\cdot \|_{2}$ Euclidean norm and if $Y=\mathbb {R} \times \mathbb {R} \times \{0\}$ is the $x{\text{-}}y$ plane then the points $u=(0,0,\pm 1)$ satisfy the conclusion $d(u,Y)=1.$ If $Z=\{(0,0)\}\times \mathbb {R}$ is $z$ -axis then every point $u$ belonging to the unit circle in the $x{\text{-}}y$ plane satisfies the conclusion $d(u,Z)=1.$ But if $X=\mathbb {R} ^{3}$ was endowed with the $\|\cdot \|_{1}$ taxicab norm (instead of the Euclidean norm), then the conclusion $d(u,Z)=1$ would be satisfied by every point $u=(x,y,0)$ belonging to the “diamond” $|x|+|y|=1$ in the $x{\text{-}}y$ plane (a square with vertices at $(\pm 1,0,0)$ and $(0,\pm 1,0)$ ).

Non-examples

Equivalently stated, a Banach space is non-reflexive if and only if Riesz’s lemma does not hold for $\alpha =1$ .

For example, the Lebesgue space $\ell _{\infty }(\mathbb {N} )$ of all bounded sequences is non-reflexive, and $\ell _{1}(\mathbb {N} )$ is non-reflexive, too. More examples on the reflexivity page.

Applications

Characterization of finite dimension

Riesz's lemma guarantees that for any given $0<\alpha <1,$ every infinite-dimensional normed space contains a sequence $x_{1},x_{2},\ldots$ of (distinct) unit vectors satisfying $\|x_{n}-x_{m}\|>\alpha$ for $m\neq n;$ or stated in plain English, these vectors are all separated from each other by a distance of more than $\alpha$ while simultaneously also all lying on the unit sphere. Such an infinite sequence of vectors cannot be found in the unit sphere of any finite dimensional normed space (just consider for example the unit circle in $\mathbb {R} ^{2}$ ).

This sequence can be constructed by induction for any constant $0<\alpha <1.$ Start by picking any element $x_{1}$ from the unit sphere. Let $Y_{n-1}$ be the linear span of $\{x_{1},\ldots ,x_{n-1}\}$ and (using Riesz's lemma) pick $x_{n}$ from the unit sphere such that

d\left(x_{n},Y_{n-1}\right)>\alpha

where

d(x_{n},Y)=\inf _{y\in Y}\|x_{n}-y\|.

This sequence $x_{1},x_{2},\ldots$ contains no convergent subsequence, which implies that the closed unit ball is not compact.

Thus, if $X$ is a normed vector space, then $X$ is finite dimensional if and only if the closed unit ball in $X$ is compact.

In particular, the identity operator on a Banach space $X$ is compact if and only if $X$ is finite-dimensional.^[6]

More generally, a topological vector space $X$ is finite dimensional if and only if it is locally compact.^[7] This classical result is also attributed to Riesz.

Proof sketch: Let $C$ be a compact neighborhood of the origin in $X.$ By compactness, there are $c_{1},\ldots ,c_{n}\in C$ such that $C~\subseteq ~\left(c_{1}+{\tfrac {1}{2}}C\right)\cup \cdots \cup \left(c_{n}+{\tfrac {1}{2}}C\right).$

We claim that the finite dimensional subspace $Y$ spanned by $\{c_{1},\ldots ,c_{n}\}$ is dense in $X,$ or equivalently, its closure is $X.$ Since $X$ is the union of scalar multiples of $C,$ it is sufficient to show that $C\subseteq Y.$ By induction, for every $m,$ $C~\subseteq ~Y+{\frac {1}{2^{m}}}C.$ But compact sets are bounded, so $C$ lies in the closure of $Y.$ This proves the result.

For a different proof based on Hahn–Banach theorem see Crespín (1994).

Banach space theory

The spectral properties of compact operators acting on a Banach space are similar to those of matrices. Riesz's lemma is essential in establishing this fact.

A particular case of the previous section is that the identity operator on a Banach space $X$ is a compact operator if and only if $X$ is finite-dimensional.^[6]

As detailed in the article on infinite-dimensional Lebesgue measure, Riesz's lemma is useful in showing the non-existence of certain measures on infinite-dimensional Banach spaces.^[6]

References

^ Rynne & Youngson 2008, p. 47.
^ ^a ^b Diestel 1984, p. 6.
^ "Riesz's lemma". PlanetMath.
^ Kreyszig 1978.
^ Based on notes by Paul Garrett
^ ^a ^b ^c Kreyszig (1978, Theorem 2.5-3, 2.5-5)
^ Tao, Terence (24 May 2011). "Locally compact topological vector spaces".

Diestel, Joe (1984). Sequences and series in Banach spaces. New York: Springer-Verlag. ISBN 0-387-90859-5. OCLC 9556781.
Crespín, Daniel (1994). "Hahn–Banach theorem implies Riesz theorem" (PDF). Portugaliae Mathematica. 51 (2): 217–218. MR 1277990.
Hashimoto, Kazuo; Nakamura, Gen; Oharu, Shinnosuke (1986-01-01). "Riesz's lemma and orthogonality in normed spaces" (PDF). Hiroshima Mathematical Journal. 16 (2). Hiroshima University - Department of Mathematics. doi:10.32917/hmj/1206130429. ISSN 0018-2079.
Kreyszig, Erwin (1978). Introductory functional analysis with applications. New York: John Wiley & Sons. ISBN 0-471-50731-8. OCLC 2818701.
Riesz, Frederic; Sz.-Nagy, Béla (1990) [1955]. Functional Analysis. Translated by Boron, Leo F. New York: Dover Publications. ISBN 0-486-66289-6. OCLC 21228994.
Rynne, Bryan P.; Youngson, Martin A. (2008). Linear Functional Analysis (2nd ed.). London: Springer. ISBN 978-1848000049. OCLC 233972987.

v t e Banach space topics
Types of Banach spaces	Asplund Banach list Banach lattice Grothendieck Hilbert Inner product space Polarization identity (Polynomially) Reflexive Riesz L-semi-inner product (B Strictly Uniformly) convex Uniformly smooth (Injective Projective) Tensor product (of Hilbert spaces)
Banach spaces are:	Barrelled Complete F-space Fréchet tame Locally convex Seminorms/Minkowski functionals Mackey Metrizable Normed norm Quasinormed Stereotype
Function space Topologies	Banach–Mazur compactum Dual Dual space Dual norm Operator Ultraweak Weak polar operator Strong polar operator Ultrastrong Uniform convergence
Linear operators	Adjoint Bilinear form operator sesquilinear (Un)Bounded Closed Compact on Hilbert spaces (Dis)Continuous Densely defined Fredholm kernel operator Hilbert–Schmidt Functionals positive Pseudo-monotone Normal Nuclear Self-adjoint Strictly singular Trace class Transpose Unitary
Operator theory	Banach algebras C-algebras Operator space Spectrum C-algebra radius Spectral theory of ODEs Spectral theorem Polar decomposition Singular value decomposition
Theorems	Anderson–Kadec Banach–Alaoglu Banach–Mazur Banach–Saks Banach–Schauder (open mapping) Banach–Steinhaus (Uniform boundedness) Bessel's inequality Cauchy–Schwarz inequality Closed graph Closed range Eberlein–Šmulian Freudenthal spectral Gelfand–Mazur Gelfand–Naimark Goldstine Hahn–Banach hyperplane separation Kakutani fixed-point Krein–Milman Lomonosov's invariant subspace Mackey–Arens Mazur's lemma M. Riesz extension Parseval's identity Riesz's lemma Riesz representation Robinson-Ursescu Schauder fixed-point Sobczyk's theorem
Analysis	Abstract Wiener space Banach manifold bundle Bochner space Convex series Differentiation in Fréchet spaces Derivatives Fréchet Gateaux functional holomorphic quasi Integrals Bochner Dunford Gelfand–Pettis regulated Paley–Wiener weak Functional calculus Borel continuous holomorphic Measures Lebesgue Projection-valued Vector Weakly / Strongly measurable function
Types of sets	Absolutely convex Absorbing Affine Balanced/Circled Bounded Convex Convex cone (subset) Convex series related ((cs, lcs)-closed, (cs, bcs)-complete, (lower) ideally convex, (Hx), and (Hwx)) Linear cone (subset) Radial Radially convex/Star-shaped Symmetric Zonotope
Subsets / set operations	Affine hull (Relative) Algebraic interior (core) Bounding points Convex hull Extreme point Interior Linear span Minkowski addition Polar (Quasi) Relative interior
Examples	Absolute continuity AC $ba(\Sigma )$ c space Banach coordinate BK Besov $B_{p,q}^{s}(\mathbb {R} )$ Birnbaum–Orlicz Bounded variation BV Bs space Continuous C(K) with K compact Hausdorff Hardy H^p Hilbert H Morrey–Campanato $L^{\lambda ,p}(\Omega )$ ℓ^p $\ell ^{\infty }$ L^p $L^{\infty }$ weighted Schwartz $S\left(\mathbb {R} ^{n}\right)$ Segal–Bargmann F Sequence space Sobolev W^k,p Sobolev inequality Triebel–Lizorkin Wiener amalgam $W(X,L^{p})$
Applications	Differential operator Finite element method Mathematical formulation of quantum mechanics Ordinary Differential Equations (ODEs) Validated numerics