logo

Sequence Spaces (ℓp spaces) 📂Banach Space

Sequence Spaces (ℓp spaces)

Definition 1

For 1p<1 \le p < \infty, distance space (p,dp)( \ell^{p} , d^{p} ) is defined as follows:

(i) Set of converging sequences:

p:={{xn}nNC(i=1xip)1p<} \ell^{p} := \left\{ \left\{ x_{n} \right\}_{n \in \mathbb{N}} \subset \mathbb{C} \left| \left( \sum_{i=1}^{\infty} | x_{i} |^{p} \right)^{{1} \over {p}} < \infty \right. \right\}

(ii) Distance function:

dp(xn,yn):=(i=1xiyip)1p,{xn},{yn}p d^{p} ( x_{n} , y_{n} ) := \left( \sum_{i = 1}^{\infty} | x_{i} - y_{i} |^{p} \right)^{ {{1} \over {p}} },\quad \left\{ x_{n} \right\} , \left\{ y_{n} \right\} \in \ell^{p}

For p=p = \infty, distance space (,d)( \ell^{\infty} , d^{\infty} ) is defined as follows:

(i)’ Set of bounded sequences:

:={{xn}nN  supiNxi<} \ell^{\infty} := \left\{ \left\{ x_{n} \right\}_{n \in \mathbb{N}} \ \left| \ \sup_{i \in \mathbb{N}} | x_{i} | < \infty \right. \right\}

(ii)’ Distance function:

d(xn,yn):=supiNxiyi,{xn},{yn}d^{\infty} ( x_{n} , y_{n} ) := \sup_{i \in \mathbb{N}} | x_{i} - y_{i} |,\quad \left\{ x_{n} \right\} , \left\{ y_{n} \right\} \in \ell^{\infty}

Explanation

p\ell^{p} is called a sequence space, and p\ell^{p} is pronounced as [ell_p]. The TeX code for \ell is \ell.

The difference between space p\ell^{p} and LpL^{p} space is merely whether it’s about sequences or functions, series or integrals. This includes Young’s inequality, Cauchy-Schwarz inequality, Hölder’s inequality, Minkowski’s inequality, as well as completeness. Since the facts themselves are similar, the methods of proof are also largely the same, so if one is thoroughly studied, there is no real need to study the other.

On the other hand, \ell^{\infty} is, in reality, the same as when pp \to \infty, proven to be unnecessary to define separately. Most properties of p\ell^{p} and \ell^{\infty} are almost the same, so there’s no need to think of them separately.

However, a notable exception is divisibility.

Theorem 1

Let’s say 1p0<1 \le p_{0} < \infty.

  • (a): p0\ell^{p_{0}} is a divisible space.
  • (b): \ell^{\infty} is an indivisible space.

This difference occurs because p0\ell^{p_{0}} has convergence as a condition, whereas \ell^{\infty} only has boundedness as a condition.

Proof

(a)

Strategy: A converging sequence always allows to select i0i_{0} so that i=i0ai\displaystyle \left| \sum_{i = i_{0}}^{\infty} a_{i} \right| becomes sufficiently small. Based on this i0i_{0}, it divides into finite and infinite parts, and then explicitly finds a subset that makes lp0l^{p_{0}} a divisible space by utilizing the fact that any finite sequence must converge to a point.

Claim: A countable set Mp0M \subset \ell^{p_{0}} that satisfies M=p0\overline{M} = \ell^{p_{0}} exists.


Consider a set of complex sequences that repeat only 00 from a certain j0j_{0}

M:={{mj}p0  mjQ+iQ,mj=0, j>j0, j0N} M : = \left\{ \left\{ m_{j} \right\} \in \ell^{p_{0}} \ | \ m_{j} \in \mathbb{Q} + i \mathbb{Q} , m_{j} = 0 , \ j>j_{0} , \ j_{0} \in \mathbb{N} \right\}

Since MM is a countable set and Mp0\overline{M} \subset \ell^{p_{0}} obviously holds, it is sufficient to show p0M\ell^{p_{0}} \subset \overline{M}. According to the definition of lp0l^{p_{0}}, every sequence x:=(x1,x2,)p0x : = ( x_{1} , x_{2} , \cdots ) \in \ell^{p_{0}} must have

(j>Nxjp0)1p0<ε2 \left( \sum_{j > N} | x_{j} |^{p_{0}} \right)^{ {{1} \over {p_{0}}} } < {{ \varepsilon } \over {2}}

that satisfies for any ε>0\varepsilon > 0. Then, for each xx

(x1m1p0++xNmNp0)1p0<ε2 \left( |x_{1} - m_{1}|^{p_{0}} + \cdots + |x_{N} - m_{N}|^{p_{0}} \right)^{ {{1} \over {p_{0}}} } < {{\varepsilon} \over {2}}

that satisfies also exists

dp0(x,m)=(jNxjmjp0+j>Nxjp0)1p0<ε2+ε2=ε d^{p_{0}} ( x, m) = \left( \sum_{j \le N} |x_{j} - m_{j}|^{p_{0}} + \sum_{j > N} |x_{j}|^{p_{0}} \right)^{{1} \over {p_{0}}} < {{ \varepsilon} \over {2}} + {{ \varepsilon} \over {2}} = \varepsilon

for all ε>0\varepsilon >0, Bdp0(x;ε)MB^{d^{p_{0}}} (x ; \varepsilon ) \cap M \ne \emptyset, so xMx \in \overline{M}

(b)

Strategy: Define manageable bounded function eIe_{I} \in \ell^{\infty} and functions ψ\psi on them, and use their injectiveness to calculate cardinality.

Claim: No countable set Mp0M \subset \ell^{p_{0}} satisfying M=\overline{M} = \ell^{\infty} exists.


It is sufficient to show that all MM \subset \ell^{\infty} satisfying M=\overline{ M} = \ell^{\infty} are uncountable.

  • Part 1.

    Let’s define function eI:I{0,1}e_{I} : I \to \left\{ 0, 1 \right\} with domain INI \subset \mathbb{N} as

    eI(j):={1,jI0,j(NI) e_{I} (j) := \begin{cases} 1 & , j \in I \\ 0 & , j \in ( \mathbb{N} \setminus I ) \end{cases}

    For example, if I=2N={2,4,6,}I = 2 \mathbb{N} = \left\{ 2, 4, 6 , \cdots \right\} , then the function value appears as e2N(1)=0e_{2 \mathbb{N}} (1) = 0, e2N(2)=1e_{2 \mathbb{N}} (2) = 1, e2N(3)=0e_{2 \mathbb{N}} (3) = 0, e2N(4)=1e_{2 \mathbb{N}} (4) = 1 .

    Considering the set of these functions A:={eI  IN}A: = \left\{ e_{I} \ | \ I \subset \mathbb{N} \right\}, since the function value cannot exceed [0,1][0,1], AA \subset \ell^{\infty} holds.

  • Part 2.

    Define function ϕ:P(N)A\phi : \mathscr{P} ( \mathbb{N} ) \to A as ϕ(I):=eI\phi (I) : =e_{I}. Then, if I,INI , I’ \subset \mathbb{N} is IiI \ne i ', ϕ(I)=eIeI=ϕ(I)\phi (I) = e_{I} \ne e_{I’} = \phi (I’) holds, so ϕ\phi is injective, and therefore

    AP(N)=20=1 |A| \ge | \mathscr{P} ( \mathbb{ N} ) | = 2^{\aleph_{0}} = \aleph_{1}

    For any x=Mx \in \ell^{\infty} = \overline{M} and ε>0\varepsilon >0, Bd(x;ε)MB_{d^{\infty}} (x ; \varepsilon ) \cap M \ne \emptyset holds

    Bd(eI;13)M B_{d^{\infty}} \left( e_{I} ; {{1} \over {3}} \right) \cap M \ne \emptyset

  • Part 3.

    Since Bd(eI;13)M\displaystyle B_{d^{\infty}} \left( e_{I} ; {{1} \over {3}} \right) \cap M \ne \emptyset holds

    ψ(eI)(Bd[eI;13]M) \psi ( e_{I} ) \in \left( B_{d^{\infty}} \left[ e_{I} ; {{1} \over {3}} \right] \cap M \right)

    we can define function ψ:AM\psi : A \to M that satisfies. Assuming ψ\psi is not injective, for ψ(eI)=ψ(eI)\psi ( e_{I}) = \psi ( e_{I’ }), we have

    ψ(eI)=ψ(eI)[Bd(eI;13)Bd(eI;13)] \psi ( e_{I}) = \psi ( e_{I’ }) \in \left[ B_{d^{\infty}} \left( e_{I} ; {{1} \over {3}} \right) \cap B_{d^{\infty}} \left( e_{I’} ; {{1} \over {3}} \right) \right]

    According to the triangle inequality

    1=d(eI,eI)d(eI,ψ(eI))+d(ψ(eI),eI)13+13=23 1 = d^{\infty} ( e_{I} , e_{I’} ) \le d^{\infty} ( e_{I} , \psi (e_{I}) ) + d^{\infty} ( \psi (e_{I}) , e_{I’} ) \le {{1} \over {3}} + {{1} \over {3}} = {{2} \over {3}}

    It concludes 123\displaystyle 1 \le {{2} \over {3}}, which is a contradiction, hence ψ\psi is injective. Also, since ψ:AM\psi : A \to M is injective, MA=1|M| \ge |A| = \aleph_{1}, and MM cannot be countable.


  1. Kreyszig. (1989). Introductory Functional Analysis with Applications: p11. ↩︎ ↩︎