What is asymptotics of this oscillatory double sum? (Fractal Dimension problem)

147 Views Asked by At

The term Gibbs Phenomenon refers to the peculiar way Fourier Series behave at sharp changes in a function's value.

However, this problem becomes particularly annoying to deal with when trying to estimate the discrete length functional of a function $f(x)$, given by,

$$(1) \quad L^b_a[f(x)]=\frac{b-a}{\omega} \cdot \sum_{n=1}^{\omega} \sqrt{1+\omega^2 \cdot \left(f \left(\frac{(b-a)\cdot n}{\omega} \right)-f\left(\frac{(b-a)\cdot (n-1)}{\omega}\right) \right)^2}$$

Given,

$$(2) \quad f(x)=\sum_{k=1}^{\infty} k^{-p} \cdot \sin(k^p \cdot x)$$

I want to know the asymptotic behavior of $(1)$ as $\omega \rightarrow \infty$. Specifically, I want the techniques used to be clearly explained and possibly reproducible for other functions.

Because of Gibb's Phenomenon, I can't fully trust my numerical results that find $L \sim \omega^{2^{-p}}, \ p \gt v$

However, I do have formal methods pointing towards a power law with the exponent $\beta$ being bounded by,

$$(3) \quad 0\lt \beta \lt 1/p$$

Motivation

The motivation behind the problem is to find the box dimension of the graph of $f(x)$. So a good criteria for a good answer is whether or not it is accuarate enough to yield this number. As mentioned in comments, I don't currently know of a way to approximate the sum except in special cases. However,

$$\sqrt{1+x^2} \sim |x|$$

Is useful since, the derivatives involved are much larger than unity.

(I intend to make this a bounty and add more information)