4
$\begingroup$

It is known that there exists a finitely additive translation invariant measure on $\mathbb{R}$ that extends the Lebesgue measure. I.e. a function $m:\mathcal{P}(\mathbb{R}) \rightarrow [0,\infty]$ that is

  1. Finitely additive: $m(A \sqcup B) = m(A) + m(B)$

  2. For a Lebesgue measurable set $A$, $m(A)$ is its Lebesgue measure.

  3. For all $A \subset \mathbb{R}$, $\lambda \in \mathbb{R}$ we have $m(A + \lambda) = m(A)$

The proof can be found in Stein & Shakarachi's Functional Analysis (the main tool is the Hahn Banach Theorem, which gives the "Banach Integral"). This got me wondering: can we also find such an m that is homogeneous? Meaning $\forall \lambda \in \mathbb{R}, \, A \subset \mathbb{R} \; m(\lambda A) = \vert \lambda \vert m(A)$. I couldn't see directly from the proof how to do so. Does such a measure even exist? Any reference given is appreciated.

A similar statement in $\mathbb{R}^3$ fails because of the Banach-Tarski Paradox.

$\endgroup$
2
  • $\begingroup$ I think that, to run afoul of the Banach–Tarski paradox, you need a measure invariant not just under translations, but under arbitrary congruences (which is why @YCor's answer isn't ruled out). $\endgroup$
    – LSpice
    Mar 28, 2019 at 2:15
  • $\begingroup$ Under congruences by the group of isometries or oriented isometries yields a "paradox" (no longer a paradox, but obstruction to existence of an invariant measure defined on all subsets) in dimension $\ge 3$. Using congruences by the group of determinant 1 or determinant $\pm 1$ (=measure-preserving) affine self-transformation yields such an obstruction in dimension $\ge 2$. $\endgroup$
    – YCor
    Mar 28, 2019 at 7:57

1 Answer 1

5
$\begingroup$

[Corrected answer, entirely rewritten] Yes. And this also works on $V=\mathbf{R}^n$ (with homogeneity axiom rewritten as $\mu(tY)=|t|^n\mu(Y)$).

Let $\mu_0:\mathcal{P}(V)\to [0,\infty]$ be a translation invariant, finitely additive mesure, extending the Lebesgue measure.

Let $I$ be the set of $Y\subset V$ such that $$\sup_{t\in\mathbf{R}^*}|t|^{-n}\mu_0(tY)<\infty.$$ This is an ideal of the Boolean algebra $\mathcal{P}(V)$. For $Y\in\mathcal{P}(V)\smallsetminus I$, define $\mu(Y)=\infty$.

Let $\nu$ be an invariant mean on the discrete (amenable) group $\mathbf{R}^*$. For $Y\in I$, define $$\mu(Y)=\int_{t\in\mathbf{R}^*}|t|^{-n}\mu(tY)d\nu(t).$$ (This is valid since we integrate a bounded function along the mean.) Then $\mu$ is a finitely additive, translation-invariant measure on $I$, and hence on $\mathcal{P}(V)$ since $I$ is an ideal. From the invariance of $\nu$, we deduce that $\mu$ satisfies $\mu(tY)=|t|^n\mu(Y)$ for all $t\in\mathbf{R}^*$.

$\endgroup$
7
  • $\begingroup$ Thank you for the answer! Could you please briefly explain what you mean by a mean on $\mathcal{P}(V)$? As it seems it can't obtain the value $\infty$, so how can it extend the Lebesgue measure? ($m(\mathbb{R}) = \infty$). $\endgroup$
    – pitariver
    Feb 23, 2019 at 19:20
  • $\begingroup$ @pitariver you're right, my answer is not correct. In a first thought, I can repair the proof so as to work on bounded subsets. I have to think more... $\endgroup$
    – YCor
    Feb 23, 2019 at 19:27
  • $\begingroup$ I think if your approach proves the existence of such mean for $\mathbb{R}^n / \mathbb{Z}^n$ it solves the problem, since form there we can manually extend the mean to $\mathbb{R}^n$. In the case of $\mathbb{R}$ one can define $$ m(E) = \sum_{k=-\infty}^{\infty} \hat{m}(E \cap [k,k+1) - k) $$ where $\hat{m}$ is the promised mean on $\mathbb{R} / \mathbb{Z}$. The proof in the book shows that it works. $\endgroup$
    – pitariver
    Feb 23, 2019 at 19:35
  • $\begingroup$ I rewrote the answer using another approach. $\endgroup$
    – YCor
    Feb 23, 2019 at 19:43
  • $\begingroup$ I think you got it, thanks again for the attention! $\endgroup$
    – pitariver
    Feb 23, 2019 at 20:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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