Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

It is well-known that for any ultrafilter $\mathscr{u}$ in $\mathbb{N}$, the map\begin{equation}a\mapsto \lim_{\mathscr{u}}a\end{equation} is a multiplicative linear functional, where $\lim_{\mathscr{u}}a$ is the limit of the sequence $a$ along $\mathscr{u}$.

I vaguely remember someone once told me that every multiplicative linear functional on $\ell^{\infty}$ is of this form. That is, given a multiplicative linear functional $h$ on $\ell^{\infty}$, there is an ultrafilter $\mathscr{u}$ such that \begin{equation} h(a)=\lim_{\mathscr{u}}a \end{equation} for all $a\in\ell^{\infty}$.

However, I cannot find a proof to this. I can show that if $h$ is the evaluation at $n$, then $h$ corresponds to the principal ultrafilter centered at $n$, but there are other kinds of multiplicative functionals (all these must vanish on any linear combinations of point masses though).

Can somebody give a hint on how to do this latter case? Thanks!

share|cite|improve this question
up vote 14 down vote accepted

There are several ways of doing this, but I'll go with the most "elementary".

Let $\varphi$ be a nonzero multiplicative functional on $\ell^\infty(\mathbb{N})$. Since $\varphi(1)=\varphi(1^2)=\varphi(1)^2$, we get that $\varphi(1)=1$ (it cannot be zero, because then $\varphi=0$).

Now let $a\in\ell^\infty(\mathbb{N})$ such that $a(n)\in\{0,1\}$ for all $n$. Write $\alpha=\varphi(a)$. As $a(1-a)=0$, we have $$ 0=\varphi(a(1-a))=\varphi(a)\varphi(1-a)=\alpha(1-\alpha). $$ So either $\alpha=0$ or $\alpha=1$.

Note that we can write $a=1_A$, $A\subset\mathbb{N}$, where $A=\{n: a(n)=1\}$. Now define $$ \mathcal U=\{A:\ \varphi(1_A)=1\}. $$ We can see that

  • $\mathbb{N}\in\mathcal U$ (since $\varphi(1)=1$)
  • $A\in\mathcal U\ \iff\ A^c\not\in\mathcal U$ (because $1_A\,1_{A^c}=0$)
  • If $A,B\in\mathcal U$, then $A\cap B\in\mathcal U$ (because $1_{A\cap B}=1_A\,1_B$)
  • If $A\in\mathcal U$ and $A\subset B$, then $B\in\mathcal U$ (because $1_A=1_A\,1_B$)

In other words, $\mathcal U$ is an ultrafilter.

Now let $c\in\ell^\infty(\mathbb{N})$ be positive, i.e. $0\leq c\leq 1$. Define sets $$ A_j^{(n)}=\{m:\ \frac{j}{2^n}\leq c(m)<\frac{(j+1)}{2^n}\},\ \ j=0,1,\ldots,2^n-1. $$ For fixed $n$, these sets are pairwise disjoint and $$\tag{1}\bigcup_jA_j^{(n)}=\mathbb{N}.$$ As $\mathcal U$ is an ultrafilter, for each $n$ there is exactly one $j(n)$ such that $A_{j(n)}^{(n)}\in\mathcal U$, and none of the others is.

Define $$ c_n=\sum_{j=0}^{2^n-1}\,\frac{j}{2^n}\,1_{A_j^{(n)}}. $$ By definition, $\|c-c_n\|\leq 2^{-n}$, so $c_n\to c$ in norm. As $\varphi$ is norm-continuous, we have $\varphi(c)=\lim_n\varphi(c_n)$. And $$ \varphi(c_n)=\sum_{j=0}^{2^n-1}\,\frac{j}{2^n}\,\varphi(1_{A_j^{(n)}})=\frac{j(n)}{2^n}, $$ so $$ \varphi(c)=\lim_n \ c(j(n))=\lim_{\mathcal U}\ c. $$ Last step is to extend by linearity to all of $\ell^\infty(\mathbb{N})$.

share|cite|improve this answer
Argh! I was in the process of typing in an answer along these lines already! Oh well, wasted effort … +1 from me then. – Harald Hanche-Olsen Jul 26 '12 at 19:30
I'm sorry about that. This idea is more or less clear in my mind, because until a few years ago ultrafilters were a complete mistery to me; and then in 2009 I attended a talk by Vladimir Pestov where these ideas were crystal clear. – Martin Argerami Jul 26 '12 at 19:55
This is crystal clear, Martin! – Jon Bannon Jul 26 '12 at 20:00
Heh. No need to apologize. It's all part of the game. – Harald Hanche-Olsen Jul 26 '12 at 20:15
The other immediate way that comes to mind is to notice that $\ell^\infty(\mathbb{N})$ is isomorphic to $C(\beta\mathbb{N})$ (continuous functions on the Stone-Cech compactification of $\mathbb{N}$. Then the multiplicative functionals are the point evaluations; functionals coming from principal filters are the evaluations at points of $\mathbb{N}$, and those coming from free ultrafilters are those evaluations at the corona. But now that I think about it, the computations above are probably still necessary. – Martin Argerami Jul 27 '12 at 1:16

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.