Every locally compact second countable group $G$ has a regular left-invariant measure $h$, the Haar measure. On the other hand the Birkhoff–Kakutani Theorem asserts that such groups also admit a compatible left-invariant metric $d$. Let us denote by $B(id_G,r)$ the $d$-open ball around the identity of radius $r$.

I'm interested in the relations between $h$ and $d$. More precisely I'm curious about the function $$Gr_d:\mathbb R\rightarrow \mathbb R,\quad r\mapsto h(B(id_G,r)).$$

Clearly this function is monotone increasing right-continuous but in general (?) not continuous. The function $Gr_d$ is not continuous at $r\in\mathbb R$ if and only if the boundary of the $r$-ball has positive measure. However since it is monotone there are only countably many point of discontinuity.

My first question is: given a l.c.s.c. group $G$ is there a left-invariant metric $d$ such that $Gr_d$ is continuous?

In most of the groups I know, there exists a canonical metric for which the growth $Gr_d$ is continuous. But in general I have no idea how to construct such a metric.

The second question is about how bad things can go: does anyone have an example of a l.c.s.c. group G and a compatible left-invariant metric $d$ for which $Gr_d$ is not continuous? Can $\mathbb R^2$ have such metrics? Given an at most countable subset $C\subset \mathbb R$ can I find a metric on $\mathbb R^2$ (or your favorite l.c.s.c. group) for which the set of discontinuity points of $Gr_d$ is exactly $C$?

Edit: Here it is what I meant but didn't write (thanks for the comments!):

  • The group $G$ is non-discrete
  • The metric $d$ is proper

And this third point would make me even happier

  • The closure of any open ball is the closed ball of the same radius
  • $\begingroup$ Do you want to assume $G$ is connected? Otherwise any discrete group is a counterexample... $\endgroup$ – Nate Eldredge Feb 16 '17 at 15:16
  • $\begingroup$ $Gr_d$ is continuous at every $r<0$, and is continuous at zero iff $G$ is non-discrete. $\endgroup$ – YCor Feb 16 '17 at 15:32
  • $\begingroup$ There's a largest left-invariant metric on $\mathbf{R}$ such that $d(x,0)\le x$ for all $x\ge 0$ and $d(x,0)\le 1$ for all $x\in [1,2]$. This metric is compatible and indeed $d(x,0)=\min(x,1)$ for all $x\in [0,2]$. The 1-sphere equals all $\pm[1,2]$ and has positive measure, so we have discontinuity then. $\endgroup$ – YCor Feb 16 '17 at 15:35
  • $\begingroup$ @NateEldredge Yes, sorry, I want to assume that $G$ is not discrete. $\endgroup$ – Alessandro Carderi Feb 16 '17 at 17:19
  • 1
    $\begingroup$ @LSpice The multiplicative group of nonzero $p$-adics is direct product of a compact group $K$ by $\mathbf{Z}$. So it's somewhat a decorated version of a discrete group. In addition, you should specify the metric, because the standard $p$-adic norm is not left-invariant by multiplication. $\endgroup$ – YCor Feb 16 '17 at 19:49

Since Wikipedia tells me that a proper metric space is one in which every closed ball is compact, I think that a modification of my comment will work: namely, the additive group of a $p$-adic field (i.e., complete, non-Archimedean, non-trivially discretely valued field with finite residue field), with its usual metric. (Maybe it violates some spirit of the question, though. For example, it is essentially maximally non-Euclidean; an $(n + 1)$-point subset of $\mathbb Q_p$ embeds isometrically in $\mathbb R^n$, but in no smaller Euclidean space.)

EDIT: I originally suggested the multiplicative group, but @YCor pointed out that doesn't work.

  • 1
    $\begingroup$ You need to specify the metric on the multiplicative group. $\endgroup$ – YCor Feb 16 '17 at 19:54
  • 1
    $\begingroup$ And clearly $\mathbf{Q}_p^*$ admits no proper compatible left-invariant ultrametric. $\endgroup$ – YCor Feb 16 '17 at 19:56
  • $\begingroup$ @YCor, oh, right, I ignored the crucial 'left-invariant' part. I guess the additive group (with its usual metric) should still work, though. $\endgroup$ – LSpice Feb 16 '17 at 20:14
  • 1
    $\begingroup$ But in the additive group, the open 1-ball is closed, i.e. equal to its closure, but not equal to the "closed 1-ball". $\endgroup$ – YCor Feb 16 '17 at 20:22
  • $\begingroup$ I actually don't know which question you claim to answer. $\endgroup$ – YCor Feb 16 '17 at 20:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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