S: 1,0.3,0.22,0.12,0.06 N: 7,201,478,2595,17950 (no idea how to put this in a tally)

I've got a question here where S is not shrunk by the same fraction throughout, I know how to work out the dimension Fractal when the Shrink is continuous E.G 1/3 each time but not when it's shrunk different each time. Can anyone help?


  • $\begingroup$ It would help if you give some context to your question. I have no idea what you're talking about. For example, the word "Shrink" is used as a noun and not defined, the expression "E.G 1/3" seems code for something, "S" is used as a noun and not defined, "S:1,0.3,0.22,0.12,0.16" seems code for something, etc. $\endgroup$ – Dave L. Renfro Feb 7 '14 at 15:38
  • $\begingroup$ where N(s) is the number of copies of the cauliflower of linear size s, which make up the original. $\endgroup$ – Unknownstarz Feb 7 '14 at 15:40
  • $\begingroup$ Ah, cauliflowers. $\endgroup$ – TonyK Feb 9 '14 at 19:43

In the definition of box-counting dimension, the dimension $d$ is given by $$d = \lim_{\varepsilon \rightarrow 0} \frac{\log N(\varepsilon)}{\log 1/\varepsilon},$$ where $N(\varepsilon)$ is the smallest number of sets of diameter $\varepsilon$ required to cover the set. In numerical applications, $N(\varepsilon)$ is often estimated using any of a number of equivalent formulations. It sounds like your assignment presents the following estimates:

\begin{align} N(1) &= 7 \\ N(0.3) &= 201 \\ N(0.22) &= 478 \\ N(0.12) &= 2595 \\ N(0.06) &= 17950 \end{align}

It might be tempting to simply use the smallest given value of $\varepsilon$ to conclude that your dimension is best approximated by $$\frac{\log(17950)}{\log(1/0.06)} \approx 3.48.$$ One problem with this is that we would typically like to take this sparse data, given over a broad range of $\varepsilon$ values and extrapolate a limit. In this example, if we compute $\log(N(\varepsilon))/\log(1/\varepsilon))$ for the other given values of $\varepsilon$, we see that the computations are decreasing sharply. Thus, the usual approach is to plot $\log(N(\varepsilon))$ versus $\log(1/\varepsilon)$ like so:

enter image description here

Thus, the $x$ coordinates are \begin{align} \log(1) &= 0 \\ \log(1/0.3) &\approx 1.2 \\ \log(1/0.22) &\approx 1.5 \\ \log(1/0.12) &\approx 2.1 \\ \log(1/0.06) &\approx 2.8 \end{align} The $y$ coordinates are $\log(N(1)), \log(N(0.3)),\ldots,\log(N(0.06))$. The line is a least squares fit to this data. The ridiculously good fit suggests that the problem has been manufactured. Since we're interested in how $\log(N(\varepsilon))$ changes in response to a change of $\log(1/\varepsilon)$, the slope of that line represents the fractal dimension. Any number of techniques can be used to compute the least squares fit to be $y\approx 1.945 + 2.79 x$.

| cite | improve this answer | |

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.