I'm preparing a presentation that would enable high-school level students to grasp that the (self-similarity) dimension of an object needs not be an integer. The first example we look at is the Sierpiński triangle and with some effort we learn that its dimension is $$s := \log(3)/\log(2) \approx 1.585.$$ After this I thought that it would be nice to mention what the actual Hausdorff $s$-measure of the triangle is, but all I found was measure estimates for a certain class of Sierpiński carpets and some estimates of the Sierpiński triangle.

I am literally shocked to learn that we apparently do not know the exact value of Hausdorff $s$-measure of the Sierpiński triangle! Especially since it's such a concrete and symmetric object. To comply to the idea of this site hosting questions instead of rants, I formulate my bafflement as follows:

Why is the $s$-measure of the Sierpiński triangle and other self-similar fractals so hard to calculate?

Are we missing a link to some complicated machinery or is the problem connected to some deep problem that one would expect to remain unsolved?

  • $\begingroup$ That's an interesting question! But I want to add that I would not know what to make of the fact that some specific set would have $0.7664$-dimenaional measure $1$. I suspect that the same is true for others and this may explain that some people would not put too much effort in such calculations. $\endgroup$
    – Dirk
    Oct 21, 2019 at 18:52
  • 4
    $\begingroup$ I don't know this subject at all, but is there any reason to expect the answer to be a closed-form number? If the answer is not a closed-form number, what would one mean by "knowing the exact value"? $\endgroup$ Oct 21, 2019 at 20:46
  • 4
    $\begingroup$ @TimothyChow One could ask for a relatively fast algorithm for computing the digits of the number. Based on the reference above, it looks like it's not even known to one significant digit, so "closed-form" isn't really the issue.. $\endgroup$
    – verret
    Oct 21, 2019 at 20:55
  • 1
    $\begingroup$ @verret : Granted. I was reacting to the sentence, "I am literally shocked to learn that we apparently do not know the exact value." $\endgroup$ Oct 21, 2019 at 21:53
  • 6
    $\begingroup$ Do you know exactly how many disks of radius 1 it would take to cover a disc of radius 1000000? I think this is the flavour of the question you’re asking. (Yes I know with Hausdorff dimension you’re allowed to have sets of different diameters, but morally...) $\endgroup$ Oct 22, 2019 at 4:44

1 Answer 1


The latest I could find is

Móra, Péter, Estimate of the Hausdorff measure of the Sierpinski triangle, Fractals 17, No. 2, 137-148 (2009). ZBL1178.28007.

where the best values are given as $$ 0.77 \le \mathscr{H}^s(\Lambda) \le 0.81794 . $$ He also proves an upper estimate $\mathscr{H}^s(\Lambda) \le 0.819161232881177$ of which he says "everybody can check it easily".

He also explains why this is harder and more technical than the reasons found in the comments here; for instance, the obvious upper bound gives only $\mathscr{H}^s(\Lambda) \le 1$.

  • 4
    $\begingroup$ That's actually the paper behind the second links in the question. $\endgroup$
    – Dirk
    Oct 24, 2019 at 14:07
  • 2
    $\begingroup$ He says it is not so easy to check the best published upper bound because it was published in Chinese. $\endgroup$
    – user44143
    Oct 24, 2019 at 15:00
  • 1
    $\begingroup$ Paywall-free version of the papers: lower bound, upper bound. $\endgroup$ Oct 25, 2019 at 17:45

Your Answer

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

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