7

The Hausdorff dimension of the Peano curve is know to be two. And I assume it to be a fractal since it's on the List of fractals by Hausdorff dimension. Moreover:

According to Falconer, one of the essential features of a fractal is that its Hausdorff dimension strictly exceeds its topological dimension.

So one can conclude that the topological dimension of the Peano curve must be one. But how could it be? I thought that in this context when we say curve, we mean the image of the curve. But the image of the Peano curve is $[0,1]^2$ and the topological dimension of $[0,1]^2$ is two.

But if the topological dimension of the Peano curve is two, why do we consider it to be a fractal?

Added:

And if the topological dimension of the Peano curve is one, how do we prove it? (For example, the topological dimension of the Koch curve is one because it's homeomorph to $[0,1]$ but it won't work with the Peano curve since $[0,1]$ and $[0,1]^2$ aren't homeomorph.)

Leo
  • 7,800
  • 5
  • 33
  • 68
  • The topology of a space-filling curve isn't the same as the topology of the space it fills, because two points near in space aren't necessarily near along the curve. – user2357112 May 31 '14 at 21:25
  • @user2357112: what definition do you have in mind? I am talking about this: Topological dimension aka Lebesgue covering dimension which is defined for topological spaces. – Leo May 31 '14 at 21:33
  • Same definition. I'm not sure whether the topological dimension ends up being 1 or 2 if you treat the curve as a set of points in the plane, since then you get self-intersections, and I'm not sure whether you get enough of them to raise the dimension. Without self-intersections (perhaps if you consider the graph of the mapping from $[0, 1]$ to $[0, 1]^2$ instead of its image), the topological dimension is definitely 1. – user2357112 May 31 '14 at 21:52
  • 1
    So what is the topology on the "curve" here ? – mercio May 31 '14 at 22:32
  • @mercio: in case you're addressing me, I don't know since I assumed it was about the topology of the curve's image. In case you're addressing user2357112, I guess, you need to add "@user2357112" in front of the comment (else the user won't get the notification). – Leo May 31 '14 at 22:39
  • 1
    Some computations of Hausdorff dimensions of graphs of coordinate functions of space-filling curves are given in http://www.math.unt.edu/~allaart/papers/dimensions.pdf. It is quite likely that if you use their methods, you can compute Hd of graphs of space-filling curves themselves. I am too lazy for such computations. – Moishe Kohan Jun 01 '14 at 18:32

3 Answers3

9

A "curve" is a function. The "Peano curve" is a function whose domain is $[0,1]$ and whose range is $[0,1]^2$. But "topological dimension" and "Hausdorff dimension" apply only to metric spaces. So when you say "the topological or Hausdorff dimension of $X$", then $X$ must be a metric space.

So the question arises: what is meant by "the topological or Hausdorff dimension of the Peano curve"?

  • Do you mean "the topological or Hausdorff dimension of the domain of the Peano curve"? Then the answer in both cases is $1$. Which is boring but sensible.
  • Do you mean "the topological or Hausdorff dimension of the range of the Peano curve"? Then the answer in both cases is $2$. Again boringly sensible.
  • But if you really, really, really mean exactly what is written, "the topological dimension of the Peano curve", well, the Peano curve is a function $$p=(p_1,p_2) :[0,1] \to [0,1]^2 $$ and a function from $[0,1]$ to $[0,1]^2$ is formally a certain subset of $[0,1] \times [0,1]^2 = [0,1]^3$, in this case $$\{(x,p_1(x),p_2(x)) \,\, | \,\, x \in [0,1]\} $$ Which is sensible and not at all boring. The topological dimension is $1$ because it is homeomorphic to $[0,1]$ (any graph of any continuous function is homeomorphic to its domain). But, the Hausdorff dimension is really interesting, I think. I'm not at all sure what it is equal to. I would need a specific formula or description to investigate this.
Lee Mosher
  • 135,265
  • But why then do we mean "the topological dimension of the range of the Koch curve" when we say "the topological dimension of the Koch curve"? (Else continuity would be enough for us to conclude that it equals one yet it's not enough.) – Leo May 31 '14 at 23:02
  • 1
    As far as I can see the Hausdorff dimension of the graph is also 2. It certainly cannot be less, since the projection $[0, 1]^3 \to [0, 1]^2$ is non-expansive. I think covering the graph with suitable families of rectangular regions provides an upper bound on the 2-dimensional Hausdorff measure. – Niels J. Diepeveen Jun 01 '14 at 01:35
  • @LeoSchmidt: My point is that to say "the topological dimenion of the Peano curve" is not sensible; it does not have a well-defined meaning. If you see that written, you have a right to ask the author "What do you mean? The domain of the Peano curve? Its range? Its graph?" If, as is likely to be the case, the author is not available to take your question, then you will have to read carefully to see if he/she implicitly meant one thing or the other. – Lee Mosher Jun 01 '14 at 11:37
  • 1
    On the other hand, unlike the Peano curve, the Koch curve is the image of an injective map. And that does change things. – Lee Mosher Jun 01 '14 at 11:41
  • @NielsDiepeveen: You might be right, for the Peano curve constructed on its Wikipedia page. But there are different constructions of "Peano curves" in the sense of continuous surjections $[0,1] \to [0,1]^2$, and I will guess that different constructions give graphs of different Hausdorff dimensions. – Lee Mosher Jun 01 '14 at 12:17
  • The lower bound applies universally. I would guess that a curve where the dimension of the graph exceeds the dimension of the range needs to have much larger fibres, almost certainly infinite, probably with positive Hausdorff dimension. I don't think any of the usual suspects have that. Maybe composing a cube-filling Peano curve with a projection onto the square works? – Niels J. Diepeveen Jun 01 '14 at 16:14
3

This is an addendum to Lee Mosher's answer, taking off where he stopped.

First of all, there is no such this as "the" Peano curve, as there are many different constructions of surjective continuous maps $g:I=[0,1]\to I^2=[0,1]^2$. Let $G$ denote the graph of such $g$. It is immediate that $G$ is homeomorphic to $[0,1]$ and, hence, has topological dimension $1$. As usual, we equip $G$ with the metric restricted from the one on $R^3$. Let $Hd$ denote the Hausdorff dimension.

Lemma. $Hd(G) \ge 2$.

Proof. $G$ is a subset of $I^3=I\times I^2$. Let $\pi$ denote the projection of $I^3$ to the second factor $I^2$. Then $\pi$ is Lipschitz (with Lipschitz constant $1$); the same, of course, applies to the restriction $\pi|G$. The image of the latter map is $I^2$, since $g$ is surjective. Let $H_2$ denote the 2-dimensional Hausdorff measure. It then follows from the definition of $H_2$ (and the 1-Lipschitz property of $\pi$) that $$ 1=H_2(\pi(G))\le H_2(G). $$ Therefore, $G$ has Hausdorff dimension $\ge 2$. QED

In particular, every such $G$ is a fractal where we define a fractal to be a metric space whose Hausdorff dimension is strictly larger than its topological dimension.

Computing the exact value of Hausdorff dimensions for graphs of particular surjective continuous functions $g: I\to I^2$ is an interesting but nontrivial task. I suspect that somebody already done this. If $\pi_1: I^2\to I$ is the coordinate projection to the $x$-axis, then computation of Hd of the graphs of $\pi_1\circ g$ for various "Peano curves" $g: I\to I^2$ are done here. I suppose that following these methods, one can compute precise values of $Hd(G)$ for graphs of various Peano curves $g: I\to I^2$ as well, but I am too lazy (and busy) for such computations.

Conjecture. For every $\alpha\in [2,3]$ there exists a "Peano curve" $g: I\to I^2$ with the graph $G$, such that $Hd(G)=\alpha$.

Moishe Kohan
  • 111,854
-1

Just because it is a space-filling curve, this curve need not have topological dimension 2. In fact, it has topological dimension 1, so it is a fractal under this definition.

There is more to read on the matter here.

naslundx
  • 9,896
  • And what would be the justification? (The linked text doesn't provide any proof.) – Leo May 31 '14 at 21:41