Namespaces
Variants
Actions

Difference between revisions of "Banach space"

From Encyclopedia of Mathematics
Jump to: navigation, search
(Importing text file)
 
m (→‎Bases: not so recent)
 
(10 intermediate revisions by 2 users not shown)
Line 1: Line 1:
 
''B-space''
 
''B-space''
  
A complete normed [[Vector space|vector space]]. The function spaces introduced by D. Hilbert, M. Fréchet and F. Riesz between 1904 and 1918 served as the starting point for the theory of Banach spaces. It is in these spaces that the fundamental concepts of strong and weak convergence, compactness, linear functional, linear operator, etc., were originally studied. Banach spaces were named after S. Banach who in 1922 began a systematic study of these spaces, based on axioms introduced by himself, and who obtained highly advanced results.
+
{{MSC|46B|46E15}}
 +
{{TEX|done}}
 +
$
 +
\newcommand{\abs}[1]{\left|#1\right|}
 +
\newcommand{\norm}[1]{\left\|#1\right\|}
 +
\newcommand{\set}[1]{\left\{#1\right\}}
 +
$
  
The theory of Banach spaces developed in parallel with the general theory of linear topological spaces (cf. [[Linear topological space|Linear topological space]]). These theories mutually enriched one another with new ideas and facts. Thus, the idea of semi-norms, taken from the theory of normed spaces, became an indispensable tool in constructing the theory of locally convex linear topological spaces. The ideas of weak convergence of elements and linear functionals in Banach spaces ultimately evolved to the concept of weak topology. The theory of Banach spaces is a thoroughly studied branch of functional analysis, with numerous applications in various branches of mathematics — directly or by way of the theory of operators.
+
A complete normed [[vector space]].  
 +
The  problems involved in Banach spaces are of different types: the geometry  of the unit ball, the geometry of subspaces, the linear topological classification, series and sequences in Banach spaces, best  approximations in Banach spaces, functions with values in a Banach  space, etc. Regarding the theory of operators in Banach spaces it should  be pointed out that many theorems are directly related to the geometry  and the topology of Banach spaces.
  
The problems involved in Banach spaces are of different types: the geometry of the unit ball, the geometry of subspaces, the linear topological classification, series and sequences in Banach spaces, best approximations in Banach spaces, functions with values in a Banach space, etc. Regarding the theory of operators in Banach spaces it should be pointed out that many theorems are directly related to the geometry and the topology of Banach spaces.
+
== History ==
 +
The function spaces introduced by D. Hilbert, M. Fréchet and F. Riesz between 1904 and 1918 served as the starting point  for the theory of Banach spaces. It is in these spaces that the fundamental concepts of strong and weak convergence, compactness, linear  functional, linear operator, etc., were originally studied. Banach spaces were named after S. Banach who in 1922 began a systematic study of these spaces {{cite|Ba}}, based on axioms introduced by himself, and who obtained highly advanced results.
  
Examples. The Banach spaces encountered in analysis are mostly sets of functions or sequences of numbers which are subject to certain conditions.
+
The theory of Banach spaces developed in parallel with the general theory of [[Linear topological space|linear topological spaces]]. These theories mutually  enriched one another with new ideas and facts. Thus, the idea of  semi-norms, taken from the theory of normed spaces, became an  indispensable tool in constructing the theory of locally convex linear  topological spaces. The ideas of weak convergence of elements and linear functionals in Banach spaces ultimately evolved to the concept of weak  topology. The theory of Banach spaces is a thoroughly studied branch of  functional analysis, with numerous applications in various branches of mathematics — directly or by way of the theory of operators.
  
1) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151901.png" />, is the space of numerical sequences <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151902.png" /> for which
+
== Generalities ==
 +
A Banach space $X$ is a [[vector space]] over $\R$ or $\C$ with a [[norm]] $\norm{\cdot}$ which is [[Completeness (in topology)|complete]] with respect to this norm, i.e., every [[Cauchy sequence]] in $X$ converges.  
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151903.png" /></td> </tr></table>
+
For two Banach spaces $X$, $Y$, denote by $B(X,Y)$ the space of linear continuous maps from $X$ to $Y$. It is in itself a Banach space with respect to the norm
 +
$$\norm{T} = \sup_{x \neq 0} \frac{\norm{Tx}}{\norm{x}}.$$
  
with the norm
+
== Examples ==
 +
The Banach spaces encountered in analysis are mostly sets of functions or  sequences of numbers which are subject to certain conditions.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151904.png" /></td> </tr></table>
+
# $\ell_p$, $p \geq 1$, is the space of numerical sequences $\set{\xi_n}$ for which $$ \sum_{n=1}^\infty \abs{\xi_n}^p < \infty$$ with the norm $$ \norm{x} = \left( \sum_{n=1}^\infty \abs{\xi_n}^p \right)^{1/p}. $$
 +
# $m$ is the space of bounded numerical sequences with the norm $$ \norm{x} = \sup_n\abs{\xi_n}.$$
 +
# $c$ is the space of convergent numerical sequences with the norm $$\norm{x} = \sup_n\abs{\xi_n}.$$
 +
# $c_0$ is the space of numerical sequences which converge to zero with the norm $$ \norm{x} = \max_n\abs{\xi_n}.$$
 +
# $C[a,b]$ is the space of continuous functions $x=x(t)$ on $[a,b]$ with the norm $$\norm{x} = \max_{a \leq t \leq b}\abs{x(t)}.$$
 +
# $C[K]$ is the space of continuous functions on a compactum $K$ with the norm $$\norm{x} = \max_{t \in K}\abs{x(t)}$$.
 +
# $C^n[a,b]$ is the space of functions with continuous derivatives up to and including the order $n$, with the norm $$\norm{x} = \sum_{k=0}^n \max_{a \leq t \leq b}\abs{x^{(k)}(t)}. $$
 +
# $C^n[I^m]$ is the space of all functions defined in an $m$-dimensional  cube that are continuously differentiable up to and including the order  $n$, with the norm of uniform boundedness in all derivatives of order at  most $n$. (Cf. [[Hölder space]].)
 +
# $M[a,b]$ is the space of bounded measurable functions with the norm $$\norm{x} = \mathop{\mathrm{ess\;max}}_{a \leq t \leq b} \abs{x(t)}.$$
 +
# $A(D)$ is the space of functions which are [[analytic function|analytic]] in the open unit  disc $D$ and are continuous in the closed disc $\bar{D}$, with the norm $$\norm{x} = \max_{z \in \bar{D}}\abs{x(z)}. $$
 +
# $L_p(S ; \Sigma, \mu)$, $p \geq 1$, is the space of functions $x(s)$  defined on a set $S$ provided with a countably-additive [[measure]] $\mu$,  with the norm $$\norm{x} = \left( \int_S \abs{x(s)}^p \,\mu(\mathrm{d}s) \right)^{1/p}.$$ (Cf. [[Lp spaces|$L^p$ spaces]].)
 +
# $L_p[a,b]$, $p \geq 1$, is a special case of the space $L_p(S ; \Sigma,  \mu)$. It is the space of [[Lebesgue measure|Lebesgue-measurable]] functions, summable of degree $p$, with the norm $$\norm{x} = \left( \int_a^b \abs{x(s)}^p \,\mathrm{d}s \right)^{1/p}.$$
 +
# $AP$ is the Bohr space of almost-periodic functions, with the norm $$\norm{x} = \sup_{-\infty < t < \infty} \abs{x(t)}. $$
  
2) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151905.png" /> is the space of bounded numerical sequences with the norm
+
The  spaces $C[a,b]$, $C^n[a,b]$, $L_p[a,b]$, $c$, $\ell_p$ are separable;  the spaces $M[a,b]$, $m$, $AP$ are non-separable; $C[K]$ is separable if and only if $K$ is a compact metric space.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151906.png" /></td> </tr></table>
+
Other examples include [[Sobolev space]]s and the [[Hardy spaces|Hardy space]] $\mathcal{H}^1$. All [[Hilbert space]]s are a forteriori Banach spaces.
  
3) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151907.png" /> is the space of convergent numerical sequences with the norm
+
== Quotients ==
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151908.png" /></td> </tr></table>
+
A (closed  linear) subspace $Y$ of a Banach space, considered apart from the  enveloping space $X$, is a Banach space. The [[quotient space]] $X/Y$ of a normed space by a subspace $Y$ is a normed space if the norm is defined  as follows. Let $Y_1 = x_1 + Y$ be a coset. Then
 +
$$
 +
\norm{Y_1} = \inf_{y \in Y} \norm{x_1 + y}.
 +
$$
 +
If  $X$ is a Banach space, then $X/Y$ is a Banach space as well.
  
4) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b0151909.png" /> is the space of numerical sequences which converge to zero with the norm
+
In this case, if $Z$ is another normed space and $T\in B(X,Z)$ fulfills $T(Y)=\{0\}$, then there exists $\hat T \in B(X/Y,Z)$ such that $T = \hat T \circ Q$ and $\norm{T}=\norm{\hat T}$, where $Q:X \to X/Y$ is the quotient mapping. {{cite|KR1|Theorem 1.5.8}}
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519010.png" /></td> </tr></table>
+
== Linear functionals, dual space ==
 +
The set of  all continuous [[linear functional]]s defined on the normed space $X$, with the norm
 +
$$
 +
\norm{f} = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}}, \quad x \neq 0  
 +
$$
 +
is said to be the dual space of $X$, and is denoted by $X^*$. It is a Banach space.
  
5) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519011.png" /> is the space of continuous functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519012.png" /> on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519013.png" /> with the norm
+
=== Hahn-Banach theorem ===
 +
Banach spaces satisfy the [[Hahn–Banach theorem]] on the extension of linear functionals: If a linear functional is defined on a subspace $Y$ of a normed space  $X$, it can be extended, while preserving its linearity and continuity,  onto the whole space $X$. Moreover, the extension can be made to have  the same norm:
 +
$$
 +
\norm{f}_X = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}} =  
 +
\norm{f}_Y = \sup_{y \in Y} \frac{\abs{f(y)}}{\norm{y}}.
 +
$$
 +
Even a more general theorem is valid: Let a real-valued function $p(x)$ defined on a linear space satisfy the conditions:
 +
$$
 +
p(x+y) \leq p(x) + p(y), \quad
 +
p(\lambda x) = \lambda p(x), \quad \lambda \geq 0, \quad x,y \in X,
 +
$$
 +
and let $f(x)$ be a real-valued linear functional defined on a subspace $Y \subset X$ and such that
 +
$$
 +
f(x) \leq p(x), \quad x \in Y.
 +
$$
 +
Then there exists a linear functional $F(x)$ defined on the whole of $X$ such that
 +
$$
 +
F(x) = f(x), \quad x \in Y; \quad
 +
F(x) \leq p(x), \quad x \in X.
 +
$$
 +
A consequence of the Hahn–Banach theorem is the  "inverse" formula which relates the norms of $X$ and $X^*$:
 +
$$
 +
\norm{x} = \max_{f \in X^*} \frac{\abs{f(x)}}{\norm{f}},\quad
 +
f \neq 0, \quad
 +
x \in X.
 +
$$
 +
The maximum in this formula is attained for some $f=f_X\in X^*$. Another  important consequence is the existence of a separating set of continuous linear functionals, meaning that for any $x_1 \neq x_2 \in X$ there  exists a linear functional $f$ on $X$ such that $f(x_1) \neq f(x_2)$  (cf.
 +
[[Complete set of functionals]]).
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519014.png" /></td> </tr></table>
+
=== General structure of linear functionals ===
 +
The  general form of a linear functional is known for many specific Banach  spaces. Thus, on $L_p[a,b]$, $p>1$, all linear functionals are given  by a formula
 +
$$
 +
f(x) = \int_a^b x(t)y(t) \,\mathrm{d}t,
 +
$$
 +
where  $y \in L_q[a,b]$, $1/p + 1/q = 1$, and any function $y(t) \in L_q$  defines a linear functional $f$ by this formula, moreover
 +
$$
 +
\norm{f} = \left( \int_a^b \abs{y(t)}^q \,\mathrm{d}t \right)^{1/q}.
 +
$$
 +
Thus,  the dual space of $L_p$ is $L_q$: $L_p^* = L_q$. Linear functionals on  $L_1[a,b]$ are defined by the same formula, but in this case $y \in M$,  so that $L_1^* = M$.
  
6) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519015.png" /> is the space of continuous functions on a compactum <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519016.png" /> with the norm
+
=== Biduals, reflexivity ===
 +
The space $X^{**}$, dual to $X^*$,  is said to be the second dual or bidual. Third, fourth, etc., dual spaces are  defined in a similar manner. Each element in $X$ may be identified with  some linear functional defined on $X^*$:
 +
$$
 +
\text{$F(f) = f(x)$ for all $f \in X^*$ ($F \in X^{**}$, $x \in X$),}
 +
$$
 +
where  $\norm{F} = \norm{x}$. One may then regard $X$ as a subspace of the space $X^{**}$ and $X \subset X^{**} \subset X^\text{IV} \subset  \cdots$, $X^* \subset X^{***} \subset \cdots$. If, as a result of these  inclusions, the Banach space coincides with its second dual, it is  called [[Reflexive space|reflexive]]. In such a case all inclusions are equalities. If $X$ is not reflexive, all inclusions are strict. If the quotient space  $X^{**}/X$ has finite dimension $n$, $X$ is said to be quasi-reflexive  of order $n$. Quasi-reflexive spaces exist for all $n$.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519017.png" /></td> </tr></table>
+
;Reflexivity criteria for Banach spaces
  
7) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519018.png" /> is the space of functions with continuous derivatives up to and including the order <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519019.png" />, with the norm
+
# $X$ is reflexive if and only if for each $f \in X^*$ it is possible to find an $x \in X$ on which the  "sup" in the formula $$ \norm{f} = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}}, \quad x \neq 0, $$ is attained.
 +
#  In reflexive Banach spaces and only in such spaces each bounded set is relatively compact with respect to weak convergence: Any one of its  infinite parts contains a weakly convergent sequence (the Eberlein–Shmul'yan theorem). The spaces $L_p$ and $\ell_p$, $p>1$,  are reflexive. The spaces $L_1$, $\ell_1$, $C$, $M$, $c$, $m$, $AP$ are  non-reflexive.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519020.png" /></td> </tr></table>
+
== Special cases ==
 
+
=== Weakly complete spaces ===
8) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519021.png" /> is the space of all functions defined in an <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519022.png" />-dimensional cube that are continuously differentiable up to and including the order <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519023.png" />, with the norm of uniform boundedness in all derivatives of order at most <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519024.png" />.
+
A Banach space is said to be weakly complete if each weak Cauchy sequence in it weakly converges to an element of the space. Every reflexive space is weakly complete. Moreover, the Banach spaces $L_1$ and $\ell_1$ are weakly complete. The Banach spaces not containing a subspace isomorphic to $c_0$ form an even wider class. These spaces resemble weakly-complete spaces in several respects.
 
 
9) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519025.png" /> is the space of bounded measurable functions with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519026.png" /></td> </tr></table>
 
 
 
10) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519027.png" /> is the space of functions which are analytic in the open unit disc <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519028.png" /> and are continuous in the closed disc <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519029.png" />, with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519030.png" /></td> </tr></table>
 
 
 
11) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519031.png" />, is the space of functions <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519032.png" /> defined on a set <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519033.png" /> provided with a countably-additive measure <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519034.png" />, with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519035.png" /></td> </tr></table>
 
 
 
12) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519036.png" />, is a special case of the space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519037.png" />. It is the space of Lebesgue-measurable functions, summable of degree <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519038.png" />, with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519039.png" /></td> </tr></table>
 
 
 
13) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519040.png" /> is the Bohr space of almost-periodic functions, with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519041.png" /></td> </tr></table>
 
 
 
The spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519042.png" /> are separable; the spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519043.png" /> are non-separable; <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519044.png" /> is separable if and only if <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519045.png" /> is a compact metric space.
 
 
 
A (closed linear) subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519046.png" /> of a Banach space, considered apart from the enveloping space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519047.png" />, is a Banach space. The quotient space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519048.png" /> of a normed space by a subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519049.png" /> is a normed space if the norm is defined as follows. Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519050.png" /> be a coset. Then
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519051.png" /></td> </tr></table>
 
 
 
If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519052.png" /> is a Banach space, then <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519053.png" /> is a Banach space as well. The set of all continuous linear functionals defined on the normed space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519054.png" />, with the norm
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519055.png" /></td> </tr></table>
 
 
 
is said to be the dual space of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519056.png" />, and is denoted by <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519057.png" />. It is a Banach space.
 
 
 
Banach spaces satisfy the [[Hahn–Banach theorem|Hahn–Banach theorem]] on the extension of linear functionals: If a linear functional is defined on a subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519058.png" /> of a normed space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519059.png" />, it can be extended, while preserving its linearity and continuity, onto the whole space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519060.png" />. Moreover, the extension can be made to have the same norm:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519061.png" /></td> </tr></table>
 
 
 
Even a more general theorem is valid: Let a real-valued function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519062.png" /> defined on a linear space satisfy the conditions:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519063.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519064.png" /></td> </tr></table>
 
 
 
and let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519065.png" /> be a real-valued linear functional defined on a subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519066.png" /> and such that
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519067.png" /></td> </tr></table>
 
 
 
Then there exists a linear functional <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519068.png" /> defined on the whole of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519069.png" /> such that
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519070.png" /></td> </tr></table>
 
 
 
A consequence of the Hahn–Banach theorem is the "inverse"  formula which relates the norms of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519071.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519072.png" />:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519073.png" /></td> </tr></table>
 
 
 
The maximum in this formula is attained for some <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519074.png" />. Another important consequence is the existence of a separating set of continuous linear functionals, meaning that for any <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519075.png" /> there exists a linear functional <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519076.png" /> on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519077.png" /> such that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519078.png" /> (cf. [[Complete set of functionals|Complete set of functionals]]).
 
 
 
The general form of a linear functional is known for many specific Banach spaces. Thus, on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519079.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519080.png" />, all linear functionals are given by a formula
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519081.png" /></td> </tr></table>
 
 
 
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519082.png" />, and any function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519083.png" /> defines a linear functional <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519084.png" /> by this formula, moreover
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519085.png" /></td> </tr></table>
 
 
 
Thus, the dual space of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519086.png" /> is <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519087.png" />: <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519088.png" />. Linear functionals on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519089.png" /> are defined by the same formula, but in this case <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519090.png" />, so that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519091.png" />.
 
 
 
The space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519092.png" />, dual to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519093.png" />, is said to be the second dual. Third, fourth, etc., dual spaces are defined in a similar manner. Each element in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519094.png" /> may be identified with some linear functional defined on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519095.png" />:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519096.png" /></td> </tr></table>
 
 
 
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519097.png" />. One may then regard <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519098.png" /> as a subspace of the space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b01519099.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190100.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190101.png" />. If, as a result of these inclusions, the Banach space coincides with its second dual, it is called reflexive. In such a case all inclusions are equalities. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190102.png" /> is not reflexive, all inclusions are strict. If the quotient space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190103.png" /> has finite dimension <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190104.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190105.png" /> is said to be quasi-reflexive of order <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190107.png" />. Quasi-reflexive spaces exist for all <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190108.png" />.
 
 
 
===Reflexivity criteria for Banach spaces.===
 
 
 
 
 
1) <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190109.png" /> is reflexive if and only if for each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190110.png" /> it is possible to find an <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190111.png" /> on which the  "sup"  in the formula
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190112.png" /></td> </tr></table>
 
 
 
is attained.
 
 
 
2) In reflexive Banach spaces and only in such spaces each bounded set is relatively compact with respect to weak convergence: Any one of its infinite parts contains a weakly convergent sequence (the Eberlein–Shmul'yan theorem). The spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190113.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190114.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190115.png" />, are reflexive. The spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190116.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190117.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190118.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190119.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190120.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190121.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190122.png" /> are non-reflexive.
 
 
 
A Banach space is said to be weakly complete if each weak Cauchy sequence in it weakly converges to an element of the space. Every reflexive space is weakly complete. Moreover, the Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190123.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190124.png" /> are weakly complete. The Banach spaces not containing a subspace isomorphic to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190125.png" /> form an even wider class. These spaces resemble weakly-complete spaces in several respects.
 
 
 
A Banach space is said to be strictly convex if its unit sphere <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190126.png" /> contains no segments. Convexity moduli are introduced for a quantitative estimation of the convexity of the unit sphere; these are the local convexity modulus
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190127.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190128.png" /></td> </tr></table>
 
  
 +
=== Strictly convex spaces ===
 +
A Banach space is said to be strictly convex  if its unit sphere $S$ contains no segments. Convexity moduli are  introduced for a quantitative estimation of the convexity of the unit  sphere; these are the local convexity modulus
 +
$$
 +
\delta(x,\epsilon) =
 +
\inf\set{
 +
1 - \norm{\frac{x+y}{2}} :
 +
y \in S,\, \norm{x-y} \geq \epsilon},
 +
\quad x \in S, \quad 0 < \epsilon \leq 2,
 +
$$
 
and the uniform convexity modulus
 
and the uniform convexity modulus
 +
$$
 +
\delta(\epsilon) = \inf_{x \in S} \delta(x,\epsilon).
 +
$$
 +
If  $\delta(x,\epsilon) > 0$ for all $x \in S$ and all $\epsilon >  0$, the Banach space is said to be locally uniformly convex. If  $\delta(x) > 0$, the space is said to be uniformly convex. All  uniformly convex Banach spaces are locally uniformly convex; all locally  uniformly convex Banach spaces are strictly convex. In  finite-dimensional Banach spaces the converses are also true. If a  Banach space is uniformly convex, it is reflexive.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190129.png" /></td> </tr></table>
+
=== Smooth spaces ===
 
+
A Banach space is said to be smooth if for any linearly independent elements $x$ and $y$ the function $\psi(t)=\norm{x+ty}$ is differentiable for all values of $t$. A Banach space is said to be uniformly smooth if its modulus of smoothness
If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190130.png" /> for all <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190131.png" /> and all <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190132.png" />, the Banach space is said to be locally uniformly convex. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190133.png" />, the space is said to be uniformly convex. All uniformly convex Banach spaces are locally uniformly convex; all locally uniformly convex Banach spaces are strictly convex. In finite-dimensional Banach spaces the converses are also true. If a Banach space is uniformly convex, it is reflexive.
+
$$
 
+
\rho(t) = \sup_{x,y \in S}
A Banach space is said to be smooth if for any linearly independent elements <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190134.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190135.png" /> the function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190136.png" /> is differentiable for all values of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190137.png" />. A Banach space is said to be uniformly smooth if its modulus of smoothness
+
\set{\frac{\norm{x + \tau y} + \norm{x - \tau y}}{2} -1},
 
+
\quad \tau > 0,
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190138.png" /></td> </tr></table>
+
$$
 
 
 
satisfies the condition
 
satisfies the condition
 +
$$
 +
\lim_{\tau \rightarrow 0}\frac{\rho(\tau)}{\tau} = 0.
 +
$$
 +
In  uniformly smooth spaces, and only in such spaces, the norm is uniformly [[Frechet derivative|Fréchet differentiable]]. A uniformly smooth Banach space is smooth. The converse is true if the Banach space is finite-dimensional. A Banach space $X$ is uniformly convex (uniformly smooth) if and only if $X^*$ is uniformly smooth (uniformly convex). The following relationship relates the convexity modulus of a Banach space $X$ and the smoothness modulus  of $X^*$:
 +
$$
 +
\rho_{X^*}(\tau) = \sup_{0 < \epsilon \leq 2}
 +
\set{\frac{\epsilon\tau}{2} - \delta_X(\epsilon)}.
 +
$$
 +
If a Banach space is uniformly convex (uniformly smooth), so are all its subspaces and quotient spaces. The Banach spaces $L_p$ and $\ell_p$,  $p>1$, are uniformly convex and uniformly smooth, and
 +
$$
 +
\delta(\epsilon) \simeq
 +
\begin{cases}
 +
\epsilon^2 & (1 < p \leq 2) \\
 +
\epsilon^p & (2 \leq p < \infty);
 +
\end{cases}
 +
$$
 +
$$
 +
\rho(\tau) \simeq
 +
\begin{cases}
 +
\tau^p & (1 < p \leq 2) \\
 +
\tau^2 & (2 \leq p < \infty);
 +
\end{cases}
 +
$$
 +
$$
 +
\left(
 +
f(\epsilon) \simeq \phi(\epsilon) \Leftrightarrow
 +
a < \frac{f(\epsilon)}{\phi(\epsilon)} < b
 +
\right).
 +
$$
 +
The Banach spaces $M$, $C$, $A$, $L_1$, $AP$, $m$, $c$, $\ell_1$ are not strictly convex and are not smooth.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190139.png" /></td> </tr></table>
+
== Linear operators ==
 
 
In uniformly smooth spaces, and only in such spaces, the norm is uniformly Fréchet differentiable. A uniformly smooth Banach space is smooth. The converse is true if the Banach space is finite-dimensional. A Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190140.png" /> is uniformly convex (uniformly smooth) if and only if <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190141.png" /> is uniformly smooth (uniformly convex). The following relationship relates the convexity modulus of a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190142.png" /> and the smoothness modulus of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190143.png" />:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190144.png" /></td> </tr></table>
 
 
 
If a Banach space is uniformly convex (uniformly smooth), so are all its subspaces and quotient spaces. The Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190145.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190146.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190147.png" />, are uniformly convex and uniformly smooth, and
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190148.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190149.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190150.png" /></td> </tr></table>
 
 
 
The Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190151.png" /> are not strictly convex and are not smooth.
 
 
 
 
The following important theorems for linear operators are valid in Banach spaces:
 
The following important theorems for linear operators are valid in Banach spaces:
  
The Banach–Steinhaus theorem. If a family of linear operators <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190152.png" /> is bounded at each point,
+
;The [[Banach–Steinhaus theorem]].  
 
+
If a family of linear operators $T=\set{T_\alpha}$ is bounded at each point,
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190153.png" /></td> </tr></table>
+
$$
 
+
\sup_\alpha \norm{T_\alpha x} < \infty, \quad x \in X,
 +
$$
 
then it is norm-bounded:
 
then it is norm-bounded:
 +
$$
 +
\sup_\alpha \norm{T_\alpha} < \infty.
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190154.png" /></td> </tr></table>
+
;The Banach [[open-mapping theorem]].  
 
+
If a linear continuous operator maps a Banach space $X$ onto a Banach space $Y$ in a one-to-one correspondence, the inverse operator $T^{-1}$ is also continuous.
The Banach open-mapping theorem. If a linear continuous operator maps a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190155.png" /> onto a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190156.png" /> in a one-to-one correspondence, the inverse operator <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190157.png" /> is also continuous.
 
 
 
The closed-graph theorem. If a closed linear operator maps a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190158.png" /> into a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190159.png" />, then it is continuous.
 
 
 
Isometries between Banach spaces occur rarely. The classical example is given by the Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190160.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190161.png" />. The Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190162.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190163.png" /> are isometric if and only if <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190164.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190165.png" /> are homeomorphic (the Banach–Stone theorem). A measure of proximity of isomorphic Banach spaces is the number
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190166.png" /></td> </tr></table>
 
 
 
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190167.png" /> runs through all possible operators which realize a (linear topological) isomorphism between <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190168.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190169.png" />. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190170.png" /> is isometric to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190171.png" />, then <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190172.png" />. However, non-isometric spaces for which <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190173.png" /> also exist; they are said to be almost-isometric. The properties of Banach spaces preserved under an isomorphism are said to be linear topological. They include separability, reflexivity and weak completeness. The isomorphic classification of Banach spaces contains, in particular, the following theorems:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190174.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190175.png" /></td> </tr></table>
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190176.png" /></td> </tr></table>
 
  
<img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190177.png" /> if <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190178.png" /> is a metric compactum with the cardinality of the continuum;
+
;The [[closed-graph theorem]].  
 +
If a closed linear operator maps a Banach space $X$ into a Banach space $Y$, then it is continuous.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190179.png" /></td> </tr></table>
+
== Isometries and isomorphisms ==
  
Each separable Banach space is isomorphic to a locally uniformly convex Banach space. It is not known (1985) if there are Banach spaces which are isomorphic to none of their hyperplanes. There exist Banach spaces which are not isomorphic to strictly convex spaces. Irrespective of the linear nature of normed spaces, it is possible to consider their topological classification. Two spaces are homeomorphic if a one-to-one continuous correspondence, such that its inverse is also continuous, can be established between their elements. An incomplete normed space is not homeomorphic to any Banach space. All infinite-dimensional separable Banach spaces are homeomorphic.
+
[[Isometric mapping|Isometries]]  between Banach spaces occur rarely. The classical example is given by  the Banach spaces $L_1$ and $\ell_2$. The Banach spaces $C[K_1]$ and  $C[K_2]$ are isometric if and only if $K_1$ and $K_2$ are homeomorphic  (the [[Banach–Stone theorem]]). A measure of proximity of isomorphic Banach  spaces is the number
 +
$$
 +
d(X,Y) = \ln\inf\bigl\|T\bigr\|\bigl\|T^{-1}\bigr\|,
 +
$$
 +
where  $T$ runs through all possible operators which realize a (linear topological) isomorphism between $X$ and $Y$. If $X$ is isometric to $Y$, then $d(X,Y)=0$. However, non-isometric spaces for which $d(X,Y)=0$  also exist; they are said to be almost-isometric. The properties of Banach spaces preserved under an isomorphism are said to be linear  topological. They include separability, reflexivity and weak  completeness. The isomorphic classification of Banach spaces contains,  in particular, the following theorems:
 +
$$
 +
L_r \neq L_s; \quad \ell_r \neq \ell_s, \quad r \neq s
 +
$$
 +
$$
 +
L_r \neq \ell_s, \quad r \neq s; \quad
 +
L_r = \ell_s, \quad r = s = 2;
 +
$$
 +
$$
 +
M=m; \quad C[0,1] \neq A(D);
 +
$$
 +
$C[K] = C[0,1]$ if $K$ is a metric compactum with the cardinality of the continuum;
 +
$$
 +
C^n[I^m] \neq C[0,1].
 +
$$
  
In the class of separable Banach spaces, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190180.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190181.png" /> are universal (cf. [[Universal space|Universal space]]). The class of reflexive separable Banach spaces contains even no isomorphic universal spaces. The Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190182.png" /> is universal in a somewhat different sense: All separable Banach spaces are isometric to one of its quotient spaces.
+
Each  separable Banach space is isomorphic to a locally uniformly convex  Banach space. It is not known (1985) if there are Banach spaces which  are isomorphic to none of their hyperplanes. There exist Banach spaces which are not isomorphic to strictly convex spaces. Irrespective of the  linear nature of normed spaces, it is possible to consider their  topological classification. Two spaces are homeomorphic if a one-to-one  continuous correspondence, such that its inverse is also continuous, can  be established between their elements. An incomplete normed space is  not homeomorphic to any Banach space. All infinite-dimensional separable Banach spaces are homeomorphic.
  
Each of the Banach spaces mentioned above, except <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190183.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190184.png" />, contains subspaces without a complement. In particular, in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190185.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190186.png" /> every infinite-dimensional separable subspace is non-complementable, while in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190187.png" /> all infinite-dimensional reflexive subspaces are non-complementable. If all subspaces in a Banach space are complementable, the space is isomorphic to a Hilbert space. It is not known (1985) whether or not all Banach spaces are direct sums of some two infinite-dimensional subspaces. A subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190188.png" /> is complementable if and only if there exists a projection which maps <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190189.png" /> onto <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190190.png" />. The lower bound of the norms of the projections on <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190191.png" /> is called the relative projection constant <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190192.png" /> of the subspace <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190193.png" /> in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190194.png" />. Each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190195.png" />-dimensional subspace of a Banach space is complementable and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190196.png" />. The absolute projection constant <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190197.png" /> of a Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190198.png" /> is
+
In the class of separable Banach spaces, $C[0,1]$ and $A(D)$ are universal (cf. [[Universal space|Universal space]]). The class of reflexive separable  Banach spaces contains even no isomorphic universal spaces. The Banach  space $\ell_1$ is universal in a somewhat different sense: All separable  Banach spaces are isometric to one of its quotient spaces.
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190199.png" /></td> </tr></table>
+
== Non-complementable subspaces ==
 +
Each  of the Banach spaces mentioned above, except $L_2$ and $\ell_2$,  contains subspaces without a complement. In particular, in $m$ and $M$  every infinite-dimensional separable subspace is non-complementable,  while in $C[0,1]$ all infinite-dimensional reflexive subspaces are  non-complementable. If all subspaces in a Banach space are  complementable, the space is isomorphic to a Hilbert space. It is not  known (1985) whether or not all Banach spaces are direct sums of some  two infinite-dimensional subspaces. A subspace $Y$ is complementable if  and only if there exists a projection which maps $X$ onto $Y$. The lower  bound of the norms of the projections on $Y$ is called the relative  projection constant $\lambda(Y,X) $ of the subspace $Y$ in $X$. Each  $n$-dimensional subspace of a Banach space is complementable and  $\lambda(Y_n,X) \leq \sqrt{n}$. The absolute projection constant  $\lambda(Y)$ of a Banach space $Y$ is
 +
$$
 +
\lambda(Y) = \sup_X \lambda(Y,X),
 +
$$
 +
where  $X$ runs through all Banach spaces which contain $Y$ as a subspace. For  any infinite-dimensional separable Banach space $Y$ one has $\lambda(Y)  = \infty$. Banach spaces for which $\lambda(Y) \leq Y < \infty$ form  the class $\mathcal{P}_\lambda$ ($\lambda \geq 1$). The class  $\mathcal{P}_1$ coincides with the class of spaces $C(Q)$ where $Q$ are  extremally-disconnected compacta (cf.
 +
[[Extremally-disconnected space]]).
  
where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190200.png" /> runs through all Banach spaces which contain <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190201.png" /> as a subspace. For any infinite-dimensional separable Banach space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190202.png" /> one has <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190203.png" />. Banach spaces for which <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190204.png" /> form the class <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190206.png" /> (<img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190207.png" />). The class <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190208.png" /> coincides with the class of spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190209.png" /> where <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190210.png" /> are extremally-disconnected compacta (cf. [[Extremally-disconnected space|Extremally-disconnected space]]).
+
== Finite-dimensional case ==
 
+
Fundamental theorems on finite-dimensional Banach spaces:
Fundamental theorems on finite-dimensional Banach spaces. 1) A finite-dimensional space (a [[Minkowski space|Minkowski space]]) is complete, i.e. is a Banach space. 2) All linear operators in a finite-dimensional Banach space are continuous. 3) A finite-dimensional Banach space is reflexive (the dimension of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190211.png" /> is equal to the dimension of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190212.png" />). 4) A Banach space is finite-dimensional if and only if its unit ball is compact. 5) All <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190213.png" />-dimensional Banach spaces are pairwise isomorphic; their set becomes compact if one introduces the distance
+
# A finite-dimensional space is complete, i.e. is a Banach space.  
 
+
# All linear operators in a finite-dimensional Banach space are continuous.  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190214.png" /></td> </tr></table>
+
# A finite-dimensional Banach space is reflexive (the dimension of $X^*$  is equal to the dimension of $X$).  
 
+
# A Banach space is finite-dimensional if and only if its unit ball is compact.
A series
+
# All $n$-dimensional Banach spaces are pairwise isomorphic; their set becomes compact if one introduces the distance
 
+
$$
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190215.png" /></td> <td valign="top" style="width:5%;text-align:right;">(*)</td></tr></table>
+
d(X,Y) = \ln\inf_T\bigl\|T\bigr\|\bigl\|T^{-1}\bigr\|.
 
+
$$
is said to be convergent if there exists a limit <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190216.png" /> of the sequence of partial sums:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190217.png" /></td> </tr></table>
 
  
 +
== Convergence of series ==
 +
A [[series]]
 +
\begin{equation}
 +
\sum_{k=1}^\infty x_k, \quad x_k \in X \label{eq:series}
 +
\end{equation}
 +
is said to be convergent if there exists a limit $S$ of the sequence of partial sums:
 +
$$
 +
\lim_{n \rightarrow \infty}
 +
\norm{S - \sum_{k=1}^n x_k} = 0.
 +
$$
 
If
 
If
 
+
$$
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190218.png" /></td> </tr></table>
+
\sum_{k=1}^\infty \norm{x_k} < \infty,
 
+
$$
the series (*) is convergent, and is said in such a case to be absolutely convergent. A series is said to be unconditionally convergent if it converges when its terms are arbitrarily rearranged. The sum of an absolutely convergent series is independent of the arrangement of its terms. In the case of series in a finite-dimensional space (and, in particular, for series of numbers) unconditional and absolute convergence are equivalent. In infinite-dimensional Banach spaces unconditional convergence follows from absolute convergence but the converse is not true in any infinite-dimensional Banach space. This is a consequence of the Dvoretskii–Rogers theorem: For all numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190219.png" />, subject to the condition <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190220.png" />, there exists in each infinite-dimensional Banach space an unconditionally convergent series <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190221.png" /> such that <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190222.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190223.png" />. In the space <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190224.png" /> (and hence also in any Banach space containing a subspace isomorphic to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190225.png" />), for any sequence <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190226.png" /> that converges to zero, there exists an unconditionally convergent series <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190227.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190228.png" />. In <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190229.png" /> the unconditional convergence of the series <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190230.png" /> implies that
+
the series $\eqref{eq:series}$ is convergent, and is said in such a case to be absolutely convergent. A series is said to be unconditionally convergent if it converges when its terms are arbitrarily rearranged. The sum of an absolutely convergent series is independent of the arrangement of its terms. In the case of series in a finite-dimensional space (and, in particular, for series of numbers) unconditional and absolute convergence are equivalent. In infinite-dimensional Banach spaces unconditional convergence follows from absolute convergence but the converse is not true in any infinite-dimensional Banach space. This is a consequence of the Dvoretskii–Rogers theorem: For all numbers $\alpha_k  \geq 0$, subject to the condition $\sum\alpha_k^2 < \infty$, there exists in each infinite-dimensional Banach space an unconditionally convergent series $\sum x_k$ such that $\norm{x_k} = \alpha_k$,  $k=1,2,\ldots$. In the space $c_0$ (and hence also in any Banach space containing a subspace isomorphic to $c_0$), for any sequence $\alpha_k  \geq 0$ that converges to zero, there exists an unconditionally convergent series $\sum x_k$, $\norm{x_k} = \alpha_k$. In $L_p(S ;  \Sigma, \mu)$ the unconditional convergence of the series $\sum x_k$  implies that
 
+
$$
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190231.png" /></td> </tr></table>
+
\sum_{k=1}^\infty
 
+
\norm{x_k}^s < \infty,
 +
$$
 
where
 
where
 +
$$
 +
s =
 +
\begin{cases}
 +
2 & (1 \leq p \leq 2), \\
 +
p & (p \geq 2).
 +
\end{cases}
 +
$$
 +
In  a uniformly convex Banach space with convexity modulus  $\delta(\epsilon)$ the unconditional convergence of the series $\sum  x_k$ implies that
 +
$$
 +
\sum_{k=1}^\infty\delta(\norm{x_k}) < \infty.
 +
$$
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190232.png" /></td> </tr></table>
+
A series $\sum x_k$ is said to be weakly unconditionally Cauchy if the series of numbers $\sum\abs{f(x_k)}$ converges for each $f \in X^*$. Each weakly unconditionally Cauchy series in $X$ converges if and only if $X$ contains no subspace isomorphic to $c_0$.
 
 
In a uniformly convex Banach space with convexity modulus <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190233.png" /> the unconditional convergence of the series <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190234.png" /> implies that
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190235.png" /></td> </tr></table>
 
 
 
A series <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190236.png" /> is said to be weakly unconditionally Cauchy if the series of numbers <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190237.png" /> converges for each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190238.png" />. Each weakly unconditionally Cauchy series in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190239.png" /> converges if and only if <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190240.png" /> contains no subspace isomorphic to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190241.png" />.
 
 
 
A sequence of elements <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190242.png" /> of a Banach space is said to be minimal if each one of its terms lies outside the closure of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190243.png" />, the linear hull of the remaining elements. A sequence is said to be uniformly minimal if
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190244.png" /></td> </tr></table>
 
 
 
If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190245.png" />, the series is said to be an Auerbach system. In each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190246.png" />-dimensional Banach space there exists a complete Auerbach system <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190247.png" />. It is not known (1985) whether or not a complete Auerbach system exists in each separable Banach space. For each minimal system there exists an adjoint system of linear functionals <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190248.png" />, which is connected with <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190249.png" /> by the biorthogonality relations: <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190250.png" />. In such a case the system <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190251.png" /> is said to be biorthogonal. A set of linear functionals is said to be total if it annihilates only the zero element of the space. In each separable Banach space there exists a complete, minimal system with a total adjoint. Each element <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190252.png" /> can formally be developed in a series by the biorthogonal system:
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190253.png" /></td> </tr></table>
 
  
 +
A  sequence of elements $\set{e_k}_1^\infty$ of a Banach space is said to  be minimal if each one of its terms lies outside the closure of $X^{(n)}  = [e_k]_{k \neq n}$, the linear hull of the remaining elements. A  sequence is said to be uniformly minimal if
 +
$$
 +
\rho(e_n ; X^{(n)}) \geq \gamma\norm{e_n}, \quad
 +
0 < \gamma \leq 1, \quad
 +
n = 1, 2, \ldots.
 +
$$
 +
If  $\gamma=1$, the series is said to be an Auerbach system. In each  $n$-dimensional Banach space there exists a complete Auerbach system  $\set{e_k}_1^n$. It is not known (1985) whether or not a complete  Auerbach system exists in each separable Banach space. For each minimal  system there exists an adjoint system of linear functionals $\set{f_n}$,  which is connected with $\set{e_k}$ by the biorthogonality relations:  $f_i(e_j) = \delta_{ij}$. In such a case the system $\set{e_k,f_k}$ is  said to be biorthogonal. A set of linear functionals is said to be total  if it annihilates only the zero element of the space. In each separable  Banach space there exists a complete, minimal system with a total  adjoint. Each element $x \in X$ can formally be developed in a series by  the biorthogonal system:
 +
$$
 +
x \sim \sum_{k=1}^\infty f_k(x)e_k,
 +
$$
 
but in the general case this series is divergent.
 
but in the general case this series is divergent.
  
A system of elements <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190254.png" /> is said to be a basis in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190255.png" /> if each element <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190256.png" /> can be uniquely represented as a convergent series
+
== Bases ==
 
+
A system of elements $\set{e_k}_1^\infty$ is said to be a basis in $X$ if each element $x \in X$ can be uniquely represented as a convergent series
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190257.png" /></td> </tr></table>
+
$$
 
+
x = \sum_{k=1}^\infty \alpha_k e_k, \quad \alpha_k = \alpha_k(x).  
Each basis in a Banach space is a complete uniform minimal system with a total adjoint. The converse is not true, as can be seen from the example of the system <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190258.png" /> in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190259.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190260.png" />.
+
$$
 
+
Each basis in a Banach space is a complete uniform minimal system with a total adjoint. The converse is not true, as can be seen from the example of the system $\set{e^{int}}_{-\infty}^\infty$ in $C[0,2\pi]$ and $L_1[0,2\pi]$.
A basis is said to be unconditional if all its rearrangements are also bases; otherwise it is said to be conditional. The system <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190261.png" /> in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190262.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190263.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190264.png" />, is a conditional basis. The Haar system is an unconditional basis in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190265.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190266.png" />. There is no unconditional basis in the spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190267.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190268.png" />. It is not known (1985) whether or not each Banach space contains an infinite-dimensional subspace with an unconditional basis. Any non-reflexive Banach space with an unconditional basis contains a subspace isomorphic to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190269.png" /> or <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190270.png" />.
 
 
 
Two normalized bases <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190271.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190272.png" /> in two Banach spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190273.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190274.png" /> are said to be equivalent if the correspondence <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190275.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190276.png" /> may be extended to an isomorphism between <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190277.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190278.png" />. In each of the spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190279.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190280.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190281.png" /> all normalized unconditional bases are equivalent to the natural basis. Bases constructed in Banach spaces which have important applications are not always suitable for solving problems, e.g. in the theory of operators. T-bases, or summation bases, have been introduced in this context. Let <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190282.png" /> be the matrix of a regular summation method (cf. [[Regular summation methods|Regular summation methods]]). The system of elements <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190283.png" /> is said to be a <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190284.png" />-basis corresponding to the given summation method if each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190285.png" /> can be uniquely represented by a series
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190286.png" /></td> </tr></table>
 
 
 
which is summable to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190287.png" /> by this method. The trigonometric system <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190288.png" /> in <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190289.png" /> is a summation basis for the methods of Cesàro and Abel. Each <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190290.png" />-basis is a complete minimal (not necessarily uniformly minimal) system with a total adjoint. The converse is not true. Until recently (the 1970's) one of the principal problems of the theory of Banach spaces was the basis problem dealt with by Banach himself: Does a basis exist in each separable Banach space? The question of existence of a basis in specifically defined Banach spaces remained open as well. The first example of a separable Banach space without a basis was constructed in 1972; bases in the spaces <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190291.png" /> and <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/b/b015/b015190/b015190292.png" /> have been constructed.
 
 
 
====References====
 
<table><TR><TD valign="top">[1]</TD> <TD valign="top">  S. Banach,   "Sur les opérations dans les ensembles abstraits et leur application aux équations intégrales"  ''Fund. Math.'' , '''3'''  (1922)  pp. 133–181</TD></TR><TR><TD valign="top">[2]</TD> <TD valign="top"> S.S. Banach,  "A course of functional analysis" , Kiev  (1948)  (In Ukrainian)</TD></TR><TR><TD valign="top">[3]</TD> <TD valign="top">  N. Dunford,   J.T. Schwartz,  "Linear operators. General theory" , '''1''' , Interscience  (1958)</TD></TR><TR><TD valign="top">[4]</TD> <TD valign="top">  M.M. Day,  "Normed linear spaces" , Springer  (1958)</TD></TR><TR><TD valign="top">[5]</TD> <TD valign="top">  N. Bourbaki,  "Elements of mathematics. Topological vector spaces" , Addison-Wesley  (1977)  (Translated from French)</TD></TR><TR><TD valign="top">[6]</TD> <TD valign="top">  I.M. Singer,  "Bases in Banach spaces" , '''1–2''' , Springer  (1970–1981)</TD></TR><TR><TD valign="top">[7]</TD> <TD valign="top">  J. Lindenstrauss,  L. Tzafriri,  "Classical Banach spaces" , '''1–2''' , Springer  (1977–1979)</TD></TR><TR><TD valign="top">[8]</TD> <TD valign="top">  J.J. Diestel,  "Geometry of Banach spaces. Selected topics" , Springer  (1975)</TD></TR><TR><TD valign="top">[9]</TD> <TD valign="top">  B. Beauzamy,  "Introduction to Banach spaces and their geometry" , North-Holland  (1985)</TD></TR></table>
 
 
 
  
 +
A basis is said to be unconditional if  all its rearrangements are also bases; otherwise it is said to be  conditional. The system $\set{e^{int}}_{-\infty}^\infty$ in  $L_p[0,2\pi]$, $p>1$, $p \neq 2$, is a conditional basis. The Haar  system is an unconditional basis in $L_p$, $p > 1$. There is no  unconditional basis in the spaces $C$ and $L_1$. It is not known (1985)  whether or not each Banach space contains an infinite-dimensional  subspace with an unconditional basis. Any non-reflexive Banach space  with an unconditional basis contains a subspace isomorphic to $\ell_1$  or $c_0$.
  
====Comments====
+
Two normalized bases $\set{e_k^\prime}$ and  $\set{e_k^{\prime\prime}} $ in two Banach spaces $X_1$ and $X_2$ are  said to be equivalent if the correspondence $e_k^\prime \leftrightarrow  e_k^{\prime\prime}$, $k=1,2,\ldots$, may be extended to an isomorphism  between $X_1$ and $X_2$. In each of the spaces $\ell_2$, $\ell_1$, $c_0 $  all normalized unconditional bases are equivalent to the natural basis.  Bases constructed in Banach spaces which have important applications  are not always suitable for solving problems, e.g. in the theory of  operators. $T$-bases, or summation bases, have been introduced in this  context. Let $\set{t_{i,j}}_1^\infty$ be the matrix of a [[Regular summation methods|regular summation method]]. The system of elements $\set{e_n} \subset X$ is  said to be a $T$-basis corresponding to the given summation method if  each $x \in X$ can be uniquely represented by a series
The second dual of a space is also called the bidual.
+
$$
 +
x \sim \sum_{k=1}^\infty \alpha_k e_k,
 +
$$
 +
which is summable to $x$ by this method. The trigonometric system  $\set{e^{int}}_{-\infty}^\infty$ in $C[0,2\pi]$ is a summation basis for  the methods of Cesàro and Abel. Each  $T$-basis is a complete minimal  (not necessarily uniformly minimal) system with a total adjoint. The converse is not true. Until the 1970s, one of the principal problems of the theory of Banach spaces was the basis problem dealt with by Banach himself: Does a basis exist in each separable Banach space?  The question of existence of a basis in specifically defined Banach spaces remained open as well. The first example of a separable Banach space without a basis was constructed in 1972; bases in the spaces  $C^n(I^m)$ and $A(D)$ have been constructed.
  
====References====
+
==References==
<table><TR><TD valign="top">[a1]</TD> <TD valign="top"> Z. Semanedi,  "Banach spaces of continuous functions" , Polish Sci. Publ.  (1971)</TD></TR></table>
+
{|
 +
|-
 +
|valign="top"|{{Ref|Ba}}||valign="top"|  S. Banach,  "Sur les opérations dans les ensembles abstraits et leur  application aux équations intégrales"  ''Fund. Math.'', '''3'''  (1922)  pp. 133–181  JFM {{ZBL|48.0201.01}}
 +
|-
 +
|valign="top"|{{Ref|Ba2}}||valign="top"|  S.S. Banach,  "A course of functional analysis", Kiev  (1948)  (In Ukrainian) 
 +
|-
 +
|valign="top"|{{Ref|Be}}||valign="top"|  B. Beauzamy,  "Introduction to Banach spaces and their geometry",  North-Holland  (1985)  {{MR|0889253}}  {{ZBL|0585.46009}}
 +
|-
 +
|valign="top"|{{Ref|Bo}}||valign="top"|  N. Bourbaki,  "Elements of mathematics. Topological vector spaces",  Addison-Wesley  (1977)  (Translated from French)  {{MR|0583191}}  {{ZBL|1106.46003}} {{ZBL|1115.46002}} {{ZBL|0622.46001}}  {{ZBL|0482.46001}}
 +
|-
 +
|valign="top"|{{Ref|Da}}||valign="top"|  M.M. Day,  "Normed linear spaces", Springer  (1958)  {{MR|0094675}}  {{ZBL|0082.10603}}
 +
|-
 +
|valign="top"|{{Ref|Di}}||valign="top"|  J.J. Diestel,  "Geometry of Banach spaces. Selected topics", Springer  (1975)  {{MR|0461094}}  {{ZBL|0307.46009}}
 +
|-
 +
|valign="top"|{{Ref|DuSc}}||valign="top"|  N. Dunford,  J.T. Schwartz,  "Linear operators. General theory",  '''1''', Interscience  (1958)  {{MR|0117523}} 
 +
|-
 +
|valign="top"|{{Ref|LiTz}}||valign="top"|  J. Lindenstrauss,  L. Tzafriri,  "Classical Banach spaces",  '''1–2''', Springer  (1977–1979)  {{MR|0500056}}  {{ZBL|0362.46013}}
 +
|-
 +
|valign="top"|{{Ref|KR1}}||valign="top"| R.V. Kadison, J.R. Ringrose, "Fundamentals of the Theory of Operator Algebras", Volume I: Elementary Theory, AMS (1997) {{MR|1468229}}
 +
|-
 +
|valign="top"|{{Ref|Se}}||valign="top"| Z. Semanedi,  "Banach spaces of continuous functions", Polish Sci. Publ.  (1971)  
 +
|-
 +
|valign="top"|{{Ref|Si}}||valign="top"|  I.M. Singer,  "Bases in Banach spaces", '''1–2''', Springer  (1970–1981)  {{MR|0298399}} {{MR|0268648}}  {{ZBL|0198.16601}}  {{ZBL|0189.42901}}
 +
|-
 +
|}

Latest revision as of 23:54, 13 April 2019

B-space

2020 Mathematics Subject Classification: Primary: 46B Secondary: 46E15 [MSN][ZBL] $ \newcommand{\abs}[1]{\left|#1\right|} \newcommand{\norm}[1]{\left\|#1\right\|} \newcommand{\set}[1]{\left\{#1\right\}} $

A complete normed vector space. The problems involved in Banach spaces are of different types: the geometry of the unit ball, the geometry of subspaces, the linear topological classification, series and sequences in Banach spaces, best approximations in Banach spaces, functions with values in a Banach space, etc. Regarding the theory of operators in Banach spaces it should be pointed out that many theorems are directly related to the geometry and the topology of Banach spaces.

History

The function spaces introduced by D. Hilbert, M. Fréchet and F. Riesz between 1904 and 1918 served as the starting point for the theory of Banach spaces. It is in these spaces that the fundamental concepts of strong and weak convergence, compactness, linear functional, linear operator, etc., were originally studied. Banach spaces were named after S. Banach who in 1922 began a systematic study of these spaces [Ba], based on axioms introduced by himself, and who obtained highly advanced results.

The theory of Banach spaces developed in parallel with the general theory of linear topological spaces. These theories mutually enriched one another with new ideas and facts. Thus, the idea of semi-norms, taken from the theory of normed spaces, became an indispensable tool in constructing the theory of locally convex linear topological spaces. The ideas of weak convergence of elements and linear functionals in Banach spaces ultimately evolved to the concept of weak topology. The theory of Banach spaces is a thoroughly studied branch of functional analysis, with numerous applications in various branches of mathematics — directly or by way of the theory of operators.

Generalities

A Banach space $X$ is a vector space over $\R$ or $\C$ with a norm $\norm{\cdot}$ which is complete with respect to this norm, i.e., every Cauchy sequence in $X$ converges.

For two Banach spaces $X$, $Y$, denote by $B(X,Y)$ the space of linear continuous maps from $X$ to $Y$. It is in itself a Banach space with respect to the norm $$\norm{T} = \sup_{x \neq 0} \frac{\norm{Tx}}{\norm{x}}.$$

Examples

The Banach spaces encountered in analysis are mostly sets of functions or sequences of numbers which are subject to certain conditions.

  1. $\ell_p$, $p \geq 1$, is the space of numerical sequences $\set{\xi_n}$ for which $$ \sum_{n=1}^\infty \abs{\xi_n}^p < \infty$$ with the norm $$ \norm{x} = \left( \sum_{n=1}^\infty \abs{\xi_n}^p \right)^{1/p}. $$
  2. $m$ is the space of bounded numerical sequences with the norm $$ \norm{x} = \sup_n\abs{\xi_n}.$$
  3. $c$ is the space of convergent numerical sequences with the norm $$\norm{x} = \sup_n\abs{\xi_n}.$$
  4. $c_0$ is the space of numerical sequences which converge to zero with the norm $$ \norm{x} = \max_n\abs{\xi_n}.$$
  5. $C[a,b]$ is the space of continuous functions $x=x(t)$ on $[a,b]$ with the norm $$\norm{x} = \max_{a \leq t \leq b}\abs{x(t)}.$$
  6. $C[K]$ is the space of continuous functions on a compactum $K$ with the norm $$\norm{x} = \max_{t \in K}\abs{x(t)}$$.
  7. $C^n[a,b]$ is the space of functions with continuous derivatives up to and including the order $n$, with the norm $$\norm{x} = \sum_{k=0}^n \max_{a \leq t \leq b}\abs{x^{(k)}(t)}. $$
  8. $C^n[I^m]$ is the space of all functions defined in an $m$-dimensional cube that are continuously differentiable up to and including the order $n$, with the norm of uniform boundedness in all derivatives of order at most $n$. (Cf. Hölder space.)
  9. $M[a,b]$ is the space of bounded measurable functions with the norm $$\norm{x} = \mathop{\mathrm{ess\;max}}_{a \leq t \leq b} \abs{x(t)}.$$
  10. $A(D)$ is the space of functions which are analytic in the open unit disc $D$ and are continuous in the closed disc $\bar{D}$, with the norm $$\norm{x} = \max_{z \in \bar{D}}\abs{x(z)}. $$
  11. $L_p(S ; \Sigma, \mu)$, $p \geq 1$, is the space of functions $x(s)$ defined on a set $S$ provided with a countably-additive measure $\mu$, with the norm $$\norm{x} = \left( \int_S \abs{x(s)}^p \,\mu(\mathrm{d}s) \right)^{1/p}.$$ (Cf. $L^p$ spaces.)
  12. $L_p[a,b]$, $p \geq 1$, is a special case of the space $L_p(S ; \Sigma, \mu)$. It is the space of Lebesgue-measurable functions, summable of degree $p$, with the norm $$\norm{x} = \left( \int_a^b \abs{x(s)}^p \,\mathrm{d}s \right)^{1/p}.$$
  13. $AP$ is the Bohr space of almost-periodic functions, with the norm $$\norm{x} = \sup_{-\infty < t < \infty} \abs{x(t)}. $$

The spaces $C[a,b]$, $C^n[a,b]$, $L_p[a,b]$, $c$, $\ell_p$ are separable; the spaces $M[a,b]$, $m$, $AP$ are non-separable; $C[K]$ is separable if and only if $K$ is a compact metric space.

Other examples include Sobolev spaces and the Hardy space $\mathcal{H}^1$. All Hilbert spaces are a forteriori Banach spaces.

Quotients

A (closed linear) subspace $Y$ of a Banach space, considered apart from the enveloping space $X$, is a Banach space. The quotient space $X/Y$ of a normed space by a subspace $Y$ is a normed space if the norm is defined as follows. Let $Y_1 = x_1 + Y$ be a coset. Then $$ \norm{Y_1} = \inf_{y \in Y} \norm{x_1 + y}. $$ If $X$ is a Banach space, then $X/Y$ is a Banach space as well.

In this case, if $Z$ is another normed space and $T\in B(X,Z)$ fulfills $T(Y)=\{0\}$, then there exists $\hat T \in B(X/Y,Z)$ such that $T = \hat T \circ Q$ and $\norm{T}=\norm{\hat T}$, where $Q:X \to X/Y$ is the quotient mapping. [KR1, Theorem 1.5.8]

Linear functionals, dual space

The set of all continuous linear functionals defined on the normed space $X$, with the norm $$ \norm{f} = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}}, \quad x \neq 0 $$ is said to be the dual space of $X$, and is denoted by $X^*$. It is a Banach space.

Hahn-Banach theorem

Banach spaces satisfy the Hahn–Banach theorem on the extension of linear functionals: If a linear functional is defined on a subspace $Y$ of a normed space $X$, it can be extended, while preserving its linearity and continuity, onto the whole space $X$. Moreover, the extension can be made to have the same norm: $$ \norm{f}_X = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}} = \norm{f}_Y = \sup_{y \in Y} \frac{\abs{f(y)}}{\norm{y}}. $$ Even a more general theorem is valid: Let a real-valued function $p(x)$ defined on a linear space satisfy the conditions: $$ p(x+y) \leq p(x) + p(y), \quad p(\lambda x) = \lambda p(x), \quad \lambda \geq 0, \quad x,y \in X, $$ and let $f(x)$ be a real-valued linear functional defined on a subspace $Y \subset X$ and such that $$ f(x) \leq p(x), \quad x \in Y. $$ Then there exists a linear functional $F(x)$ defined on the whole of $X$ such that $$ F(x) = f(x), \quad x \in Y; \quad F(x) \leq p(x), \quad x \in X. $$ A consequence of the Hahn–Banach theorem is the "inverse" formula which relates the norms of $X$ and $X^*$: $$ \norm{x} = \max_{f \in X^*} \frac{\abs{f(x)}}{\norm{f}},\quad f \neq 0, \quad x \in X. $$ The maximum in this formula is attained for some $f=f_X\in X^*$. Another important consequence is the existence of a separating set of continuous linear functionals, meaning that for any $x_1 \neq x_2 \in X$ there exists a linear functional $f$ on $X$ such that $f(x_1) \neq f(x_2)$ (cf. Complete set of functionals).

General structure of linear functionals

The general form of a linear functional is known for many specific Banach spaces. Thus, on $L_p[a,b]$, $p>1$, all linear functionals are given by a formula $$ f(x) = \int_a^b x(t)y(t) \,\mathrm{d}t, $$ where $y \in L_q[a,b]$, $1/p + 1/q = 1$, and any function $y(t) \in L_q$ defines a linear functional $f$ by this formula, moreover $$ \norm{f} = \left( \int_a^b \abs{y(t)}^q \,\mathrm{d}t \right)^{1/q}. $$ Thus, the dual space of $L_p$ is $L_q$: $L_p^* = L_q$. Linear functionals on $L_1[a,b]$ are defined by the same formula, but in this case $y \in M$, so that $L_1^* = M$.

Biduals, reflexivity

The space $X^{**}$, dual to $X^*$, is said to be the second dual or bidual. Third, fourth, etc., dual spaces are defined in a similar manner. Each element in $X$ may be identified with some linear functional defined on $X^*$: $$ \text{$F(f) = f(x)$ for all $f \in X^*$ ($F \in X^{**}$, $x \in X$),} $$ where $\norm{F} = \norm{x}$. One may then regard $X$ as a subspace of the space $X^{**}$ and $X \subset X^{**} \subset X^\text{IV} \subset \cdots$, $X^* \subset X^{***} \subset \cdots$. If, as a result of these inclusions, the Banach space coincides with its second dual, it is called reflexive. In such a case all inclusions are equalities. If $X$ is not reflexive, all inclusions are strict. If the quotient space $X^{**}/X$ has finite dimension $n$, $X$ is said to be quasi-reflexive of order $n$. Quasi-reflexive spaces exist for all $n$.

Reflexivity criteria for Banach spaces
  1. $X$ is reflexive if and only if for each $f \in X^*$ it is possible to find an $x \in X$ on which the "sup" in the formula $$ \norm{f} = \sup_{x \in X} \frac{\abs{f(x)}}{\norm{x}}, \quad x \neq 0, $$ is attained.
  2. In reflexive Banach spaces and only in such spaces each bounded set is relatively compact with respect to weak convergence: Any one of its infinite parts contains a weakly convergent sequence (the Eberlein–Shmul'yan theorem). The spaces $L_p$ and $\ell_p$, $p>1$, are reflexive. The spaces $L_1$, $\ell_1$, $C$, $M$, $c$, $m$, $AP$ are non-reflexive.

Special cases

Weakly complete spaces

A Banach space is said to be weakly complete if each weak Cauchy sequence in it weakly converges to an element of the space. Every reflexive space is weakly complete. Moreover, the Banach spaces $L_1$ and $\ell_1$ are weakly complete. The Banach spaces not containing a subspace isomorphic to $c_0$ form an even wider class. These spaces resemble weakly-complete spaces in several respects.

Strictly convex spaces

A Banach space is said to be strictly convex if its unit sphere $S$ contains no segments. Convexity moduli are introduced for a quantitative estimation of the convexity of the unit sphere; these are the local convexity modulus $$ \delta(x,\epsilon) = \inf\set{ 1 - \norm{\frac{x+y}{2}} : y \in S,\, \norm{x-y} \geq \epsilon}, \quad x \in S, \quad 0 < \epsilon \leq 2, $$ and the uniform convexity modulus $$ \delta(\epsilon) = \inf_{x \in S} \delta(x,\epsilon). $$ If $\delta(x,\epsilon) > 0$ for all $x \in S$ and all $\epsilon > 0$, the Banach space is said to be locally uniformly convex. If $\delta(x) > 0$, the space is said to be uniformly convex. All uniformly convex Banach spaces are locally uniformly convex; all locally uniformly convex Banach spaces are strictly convex. In finite-dimensional Banach spaces the converses are also true. If a Banach space is uniformly convex, it is reflexive.

Smooth spaces

A Banach space is said to be smooth if for any linearly independent elements $x$ and $y$ the function $\psi(t)=\norm{x+ty}$ is differentiable for all values of $t$. A Banach space is said to be uniformly smooth if its modulus of smoothness $$ \rho(t) = \sup_{x,y \in S} \set{\frac{\norm{x + \tau y} + \norm{x - \tau y}}{2} -1}, \quad \tau > 0, $$ satisfies the condition $$ \lim_{\tau \rightarrow 0}\frac{\rho(\tau)}{\tau} = 0. $$ In uniformly smooth spaces, and only in such spaces, the norm is uniformly Fréchet differentiable. A uniformly smooth Banach space is smooth. The converse is true if the Banach space is finite-dimensional. A Banach space $X$ is uniformly convex (uniformly smooth) if and only if $X^*$ is uniformly smooth (uniformly convex). The following relationship relates the convexity modulus of a Banach space $X$ and the smoothness modulus of $X^*$: $$ \rho_{X^*}(\tau) = \sup_{0 < \epsilon \leq 2} \set{\frac{\epsilon\tau}{2} - \delta_X(\epsilon)}. $$ If a Banach space is uniformly convex (uniformly smooth), so are all its subspaces and quotient spaces. The Banach spaces $L_p$ and $\ell_p$, $p>1$, are uniformly convex and uniformly smooth, and $$ \delta(\epsilon) \simeq \begin{cases} \epsilon^2 & (1 < p \leq 2) \\ \epsilon^p & (2 \leq p < \infty); \end{cases} $$ $$ \rho(\tau) \simeq \begin{cases} \tau^p & (1 < p \leq 2) \\ \tau^2 & (2 \leq p < \infty); \end{cases} $$ $$ \left( f(\epsilon) \simeq \phi(\epsilon) \Leftrightarrow a < \frac{f(\epsilon)}{\phi(\epsilon)} < b \right). $$ The Banach spaces $M$, $C$, $A$, $L_1$, $AP$, $m$, $c$, $\ell_1$ are not strictly convex and are not smooth.

Linear operators

The following important theorems for linear operators are valid in Banach spaces:

The Banach–Steinhaus theorem.

If a family of linear operators $T=\set{T_\alpha}$ is bounded at each point, $$ \sup_\alpha \norm{T_\alpha x} < \infty, \quad x \in X, $$ then it is norm-bounded: $$ \sup_\alpha \norm{T_\alpha} < \infty. $$

The Banach open-mapping theorem.

If a linear continuous operator maps a Banach space $X$ onto a Banach space $Y$ in a one-to-one correspondence, the inverse operator $T^{-1}$ is also continuous.

The closed-graph theorem.

If a closed linear operator maps a Banach space $X$ into a Banach space $Y$, then it is continuous.

Isometries and isomorphisms

Isometries between Banach spaces occur rarely. The classical example is given by the Banach spaces $L_1$ and $\ell_2$. The Banach spaces $C[K_1]$ and $C[K_2]$ are isometric if and only if $K_1$ and $K_2$ are homeomorphic (the Banach–Stone theorem). A measure of proximity of isomorphic Banach spaces is the number $$ d(X,Y) = \ln\inf\bigl\|T\bigr\|\bigl\|T^{-1}\bigr\|, $$ where $T$ runs through all possible operators which realize a (linear topological) isomorphism between $X$ and $Y$. If $X$ is isometric to $Y$, then $d(X,Y)=0$. However, non-isometric spaces for which $d(X,Y)=0$ also exist; they are said to be almost-isometric. The properties of Banach spaces preserved under an isomorphism are said to be linear topological. They include separability, reflexivity and weak completeness. The isomorphic classification of Banach spaces contains, in particular, the following theorems: $$ L_r \neq L_s; \quad \ell_r \neq \ell_s, \quad r \neq s $$ $$ L_r \neq \ell_s, \quad r \neq s; \quad L_r = \ell_s, \quad r = s = 2; $$ $$ M=m; \quad C[0,1] \neq A(D); $$ $C[K] = C[0,1]$ if $K$ is a metric compactum with the cardinality of the continuum; $$ C^n[I^m] \neq C[0,1]. $$

Each separable Banach space is isomorphic to a locally uniformly convex Banach space. It is not known (1985) if there are Banach spaces which are isomorphic to none of their hyperplanes. There exist Banach spaces which are not isomorphic to strictly convex spaces. Irrespective of the linear nature of normed spaces, it is possible to consider their topological classification. Two spaces are homeomorphic if a one-to-one continuous correspondence, such that its inverse is also continuous, can be established between their elements. An incomplete normed space is not homeomorphic to any Banach space. All infinite-dimensional separable Banach spaces are homeomorphic.

In the class of separable Banach spaces, $C[0,1]$ and $A(D)$ are universal (cf. Universal space). The class of reflexive separable Banach spaces contains even no isomorphic universal spaces. The Banach space $\ell_1$ is universal in a somewhat different sense: All separable Banach spaces are isometric to one of its quotient spaces.

Non-complementable subspaces

Each of the Banach spaces mentioned above, except $L_2$ and $\ell_2$, contains subspaces without a complement. In particular, in $m$ and $M$ every infinite-dimensional separable subspace is non-complementable, while in $C[0,1]$ all infinite-dimensional reflexive subspaces are non-complementable. If all subspaces in a Banach space are complementable, the space is isomorphic to a Hilbert space. It is not known (1985) whether or not all Banach spaces are direct sums of some two infinite-dimensional subspaces. A subspace $Y$ is complementable if and only if there exists a projection which maps $X$ onto $Y$. The lower bound of the norms of the projections on $Y$ is called the relative projection constant $\lambda(Y,X) $ of the subspace $Y$ in $X$. Each $n$-dimensional subspace of a Banach space is complementable and $\lambda(Y_n,X) \leq \sqrt{n}$. The absolute projection constant $\lambda(Y)$ of a Banach space $Y$ is $$ \lambda(Y) = \sup_X \lambda(Y,X), $$ where $X$ runs through all Banach spaces which contain $Y$ as a subspace. For any infinite-dimensional separable Banach space $Y$ one has $\lambda(Y) = \infty$. Banach spaces for which $\lambda(Y) \leq Y < \infty$ form the class $\mathcal{P}_\lambda$ ($\lambda \geq 1$). The class $\mathcal{P}_1$ coincides with the class of spaces $C(Q)$ where $Q$ are extremally-disconnected compacta (cf. Extremally-disconnected space).

Finite-dimensional case

Fundamental theorems on finite-dimensional Banach spaces:

  1. A finite-dimensional space is complete, i.e. is a Banach space.
  2. All linear operators in a finite-dimensional Banach space are continuous.
  3. A finite-dimensional Banach space is reflexive (the dimension of $X^*$ is equal to the dimension of $X$).
  4. A Banach space is finite-dimensional if and only if its unit ball is compact.
  5. All $n$-dimensional Banach spaces are pairwise isomorphic; their set becomes compact if one introduces the distance

$$ d(X,Y) = \ln\inf_T\bigl\|T\bigr\|\bigl\|T^{-1}\bigr\|. $$

Convergence of series

A series \begin{equation} \sum_{k=1}^\infty x_k, \quad x_k \in X \label{eq:series} \end{equation} is said to be convergent if there exists a limit $S$ of the sequence of partial sums: $$ \lim_{n \rightarrow \infty} \norm{S - \sum_{k=1}^n x_k} = 0. $$ If $$ \sum_{k=1}^\infty \norm{x_k} < \infty, $$ the series $\eqref{eq:series}$ is convergent, and is said in such a case to be absolutely convergent. A series is said to be unconditionally convergent if it converges when its terms are arbitrarily rearranged. The sum of an absolutely convergent series is independent of the arrangement of its terms. In the case of series in a finite-dimensional space (and, in particular, for series of numbers) unconditional and absolute convergence are equivalent. In infinite-dimensional Banach spaces unconditional convergence follows from absolute convergence but the converse is not true in any infinite-dimensional Banach space. This is a consequence of the Dvoretskii–Rogers theorem: For all numbers $\alpha_k \geq 0$, subject to the condition $\sum\alpha_k^2 < \infty$, there exists in each infinite-dimensional Banach space an unconditionally convergent series $\sum x_k$ such that $\norm{x_k} = \alpha_k$, $k=1,2,\ldots$. In the space $c_0$ (and hence also in any Banach space containing a subspace isomorphic to $c_0$), for any sequence $\alpha_k \geq 0$ that converges to zero, there exists an unconditionally convergent series $\sum x_k$, $\norm{x_k} = \alpha_k$. In $L_p(S ; \Sigma, \mu)$ the unconditional convergence of the series $\sum x_k$ implies that $$ \sum_{k=1}^\infty \norm{x_k}^s < \infty, $$ where $$ s = \begin{cases} 2 & (1 \leq p \leq 2), \\ p & (p \geq 2). \end{cases} $$ In a uniformly convex Banach space with convexity modulus $\delta(\epsilon)$ the unconditional convergence of the series $\sum x_k$ implies that $$ \sum_{k=1}^\infty\delta(\norm{x_k}) < \infty. $$

A series $\sum x_k$ is said to be weakly unconditionally Cauchy if the series of numbers $\sum\abs{f(x_k)}$ converges for each $f \in X^*$. Each weakly unconditionally Cauchy series in $X$ converges if and only if $X$ contains no subspace isomorphic to $c_0$.

A sequence of elements $\set{e_k}_1^\infty$ of a Banach space is said to be minimal if each one of its terms lies outside the closure of $X^{(n)} = [e_k]_{k \neq n}$, the linear hull of the remaining elements. A sequence is said to be uniformly minimal if $$ \rho(e_n ; X^{(n)}) \geq \gamma\norm{e_n}, \quad 0 < \gamma \leq 1, \quad n = 1, 2, \ldots. $$ If $\gamma=1$, the series is said to be an Auerbach system. In each $n$-dimensional Banach space there exists a complete Auerbach system $\set{e_k}_1^n$. It is not known (1985) whether or not a complete Auerbach system exists in each separable Banach space. For each minimal system there exists an adjoint system of linear functionals $\set{f_n}$, which is connected with $\set{e_k}$ by the biorthogonality relations: $f_i(e_j) = \delta_{ij}$. In such a case the system $\set{e_k,f_k}$ is said to be biorthogonal. A set of linear functionals is said to be total if it annihilates only the zero element of the space. In each separable Banach space there exists a complete, minimal system with a total adjoint. Each element $x \in X$ can formally be developed in a series by the biorthogonal system: $$ x \sim \sum_{k=1}^\infty f_k(x)e_k, $$ but in the general case this series is divergent.

Bases

A system of elements $\set{e_k}_1^\infty$ is said to be a basis in $X$ if each element $x \in X$ can be uniquely represented as a convergent series $$ x = \sum_{k=1}^\infty \alpha_k e_k, \quad \alpha_k = \alpha_k(x). $$ Each basis in a Banach space is a complete uniform minimal system with a total adjoint. The converse is not true, as can be seen from the example of the system $\set{e^{int}}_{-\infty}^\infty$ in $C[0,2\pi]$ and $L_1[0,2\pi]$.

A basis is said to be unconditional if all its rearrangements are also bases; otherwise it is said to be conditional. The system $\set{e^{int}}_{-\infty}^\infty$ in $L_p[0,2\pi]$, $p>1$, $p \neq 2$, is a conditional basis. The Haar system is an unconditional basis in $L_p$, $p > 1$. There is no unconditional basis in the spaces $C$ and $L_1$. It is not known (1985) whether or not each Banach space contains an infinite-dimensional subspace with an unconditional basis. Any non-reflexive Banach space with an unconditional basis contains a subspace isomorphic to $\ell_1$ or $c_0$.

Two normalized bases $\set{e_k^\prime}$ and $\set{e_k^{\prime\prime}} $ in two Banach spaces $X_1$ and $X_2$ are said to be equivalent if the correspondence $e_k^\prime \leftrightarrow e_k^{\prime\prime}$, $k=1,2,\ldots$, may be extended to an isomorphism between $X_1$ and $X_2$. In each of the spaces $\ell_2$, $\ell_1$, $c_0 $ all normalized unconditional bases are equivalent to the natural basis. Bases constructed in Banach spaces which have important applications are not always suitable for solving problems, e.g. in the theory of operators. $T$-bases, or summation bases, have been introduced in this context. Let $\set{t_{i,j}}_1^\infty$ be the matrix of a regular summation method. The system of elements $\set{e_n} \subset X$ is said to be a $T$-basis corresponding to the given summation method if each $x \in X$ can be uniquely represented by a series $$ x \sim \sum_{k=1}^\infty \alpha_k e_k, $$ which is summable to $x$ by this method. The trigonometric system $\set{e^{int}}_{-\infty}^\infty$ in $C[0,2\pi]$ is a summation basis for the methods of Cesàro and Abel. Each $T$-basis is a complete minimal (not necessarily uniformly minimal) system with a total adjoint. The converse is not true. Until the 1970s, one of the principal problems of the theory of Banach spaces was the basis problem dealt with by Banach himself: Does a basis exist in each separable Banach space? The question of existence of a basis in specifically defined Banach spaces remained open as well. The first example of a separable Banach space without a basis was constructed in 1972; bases in the spaces $C^n(I^m)$ and $A(D)$ have been constructed.

References

[Ba] S. Banach, "Sur les opérations dans les ensembles abstraits et leur application aux équations intégrales" Fund. Math., 3 (1922) pp. 133–181 JFM Zbl 48.0201.01
[Ba2] S.S. Banach, "A course of functional analysis", Kiev (1948) (In Ukrainian)
[Be] B. Beauzamy, "Introduction to Banach spaces and their geometry", North-Holland (1985) MR0889253 Zbl 0585.46009
[Bo] N. Bourbaki, "Elements of mathematics. Topological vector spaces", Addison-Wesley (1977) (Translated from French) MR0583191 Zbl 1106.46003 Zbl 1115.46002 Zbl 0622.46001 Zbl 0482.46001
[Da] M.M. Day, "Normed linear spaces", Springer (1958) MR0094675 Zbl 0082.10603
[Di] J.J. Diestel, "Geometry of Banach spaces. Selected topics", Springer (1975) MR0461094 Zbl 0307.46009
[DuSc] N. Dunford, J.T. Schwartz, "Linear operators. General theory", 1, Interscience (1958) MR0117523
[LiTz] J. Lindenstrauss, L. Tzafriri, "Classical Banach spaces", 1–2, Springer (1977–1979) MR0500056 Zbl 0362.46013
[KR1] R.V. Kadison, J.R. Ringrose, "Fundamentals of the Theory of Operator Algebras", Volume I: Elementary Theory, AMS (1997) MR1468229
[Se] Z. Semanedi, "Banach spaces of continuous functions", Polish Sci. Publ. (1971)
[Si] I.M. Singer, "Bases in Banach spaces", 1–2, Springer (1970–1981) MR0298399 MR0268648 Zbl 0198.16601 Zbl 0189.42901
How to Cite This Entry:
Banach space. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Banach_space&oldid=11548
This article was adapted from an original article by M.I. KadetsB.M. Levitan (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article