next up previous
Next: Dual spaces for Quotients Up: Linear Functionals Previous: Hahn-Banach Theorem

Banach Limits

Consider $ c \subset \ell^\infty$. Define $ L : c \rightarrow \mathbb{F}$ by $ L(x_1, x_2, \ldots ) = \displaystyle \lim_{n \rightarrow \infty} x_n$.

A moment's thought shows $ L$ is linear and $ \Vert L\Vert = 1$. Moreover, if $ x_i \geq 0$ for all $ i$, then $ L(x) \geq 0$.

Theorem 10.2.1   There is a linear functional $ L: \ell^\infty \rightarrow \mathbb{F}$ such that
  1. $ \Vert L\Vert = 1$
  2. If $ x = (x_1,x_2, \ldots) \in c$, then $ L(x) = \displaystyle \lim_{n \rightarrow \infty} x_n$.
  3. If $ x = (x_1,x_2, \ldots) \in \ell^\infty$ with $ x_i \geq 0$, then $ L(x) \geq 0$
  4. If $ x = (x_1,x_2, \ldots) \in \ell^\infty$ then $ L(x) = L(x_2,x_3, x_4, \ldots)$
(Not unique)

L is called the Banach Limit

Proof.
  1. $ \mathbb{F}= \mathbb{R}$. Define $ ' : \ell^\infty \rightarrow \ell^\infty$ by $ (x_1, x_2, \ldots )' = (x_2, x_3, \ldots)$. This is a linear operator of norm 1.
    Let $ M = \{ x - x' : x \in \ell^\infty \}$. As $ M$ is the kernel of $ I - '$, $ M$ is a linear subspace.
    Let $ \bf {1} = (1,1, \ldots) \in \ell^\infty$.
    Claim: $ dist(\bf {1}, M) = 1$.
    As $ (0,0,\ldots) \in M$, $ dist(\bf {1},M) \leq 1$. Let $ x = x_1,x_2, \ldots) \in \ell^\infty$. If for any $ n$, $ x_{n+1} \geq x_n$, then the $ n^{th}$ entry of $ x - x'$ will be nonpositive. Then,

    $\displaystyle \Vert\bf {1} - (x-x')\Vert _\infty $

    is at least

    $\displaystyle \vert 1 - (x_n - x_{n+1}) \vert \geq 1 $

    It remains only to consider $ x$ in $ \ell^\infty$ with $ x_{n+1} < x_n$ for each $ n$. As $ x_n \rightarrow \alpha$ for some $ \alpha \in \mathbb{R}$ for $ n \rightarrow \infty$, the $ n^{th}$ entry of $ x - x'$ goes to 0. Thus, $ \Vert\bf {1} - (x-x')\Vert _\infty \geq 1$.
    By corollary (4), there is $ L \in X^*$ with $ \ker L \geq M$ and $ L(\bf {1}) = 1$ and $ \Vert L\Vert = 1$.
    Claim: $ c_0 \subset \ker L$.
    Let $ x \in c_0$. Consider $ x',x'',x'',\ldots$. This converges to $ (0,0,\ldots)$ in $ \ell^\infty$. Why? Let $ \epsilon > 0$. Pick $ N$ such that $ \forall n \geq N$, $ \vert x_n\vert \leq \epsilon $. Then, if we ``prime" $ x$ $ N$ or more times, we get a sequence with all entries at most $ \epsilon $ in absolute value. But, $ L(x) = L(x') = L(x'') = \cdots$. Since $ \vert L(x^{(n)})\vert \leq \Vert x^{(n)}\Vert _\infty \rightarrow 0$,

    $\displaystyle 0 = L(x) = L(x') = \cdots $

    proving the claim.
    For $ x \in c$, if $ \alpha = \displaystyle \lim_{n \rightarrow \infty} x_n$, then $ x - \alpha \bf {1} \in c_0$, as $ L(x) = L(\alpha \bf {1}) = \alpha$.
    It remains to prove (3). Suppose $ x \in \ell^\infty$, $ x_n \geq 0$ for all $ n$ and $ L(x) < 0$. WLOG, $ \Vert x\Vert = 1$. So, $ 0 \leq x_n < 1$ for all $ n$ and $ \bf {1} - x$ has entries between 0 and 1. So, $ \Vert\bf {1} - x\Vert _\infty \leq 1$ and $ L(\bf {1} - x) = L(\bf {1}) - L(x) > 1$, which is a contradiction.
  2. $ \mathbb{F}= \mathbb{C}$
    Every $ z \in \ell^\infty$ can be written as $ x + iy$, $ x,y \in \ell^\infty$. Define $ L: \ell^\infty \rightarrow \mathbb{C}$ by

    $\displaystyle L(z) = L_1(x) + i L_1(y) $

    where $ L$ is the functional from case 1. A moments thought shows $ L$ is $ \mathbb{C}$-linear and properties 2,3, and 4 hold.
    Clearly, $ \Vert L\Vert \leq 2$ and $ L(\bf {1}) = 1$ shows $ \Vert L\Vert \geq 1$.
    As $ \ell^\infty = L^\infty(\mathbb{N}, P(\mathbb{N}), \mu)$, the simple functions in $ \ell^\infty$ are those of the form

    $\displaystyle f = \sum_{k=1}^n \alpha_k \chi_{E_k} $

    where $ E_k$ are pairwise disjoint subsets of $ \mathbb{N}$. Then,

    $\displaystyle L(f) \leq \max \vert\alpha_k\vert $

    and $ \Vert f\Vert _\infty = \max\vert\alpha_k\vert$, so $ \Vert L\Vert = 1$ on the simple functions. The simple functions are dense in an $ L^p$ space, so $ \Vert L\Vert = 1$.
$ \qedsymbol$


next up previous
Next: Dual spaces for Quotients Up: Linear Functionals Previous: Hahn-Banach Theorem
Brian Bockelman 2005-12-12