TL;DR
While the "popcorn function" is a staple of undergraduate real analysis, its fractal dimensions have remained surprisingly elusive. This paper finally settles the matter, proving that the graph of the popcorn function has a box dimension of 4/3 and a maximal Assouad dimension of 2, using a clever blend of number theory and probability.
What Problem Does This Paper Address?
The popcorn function (also known as Thomae’s function or the raindrop function) is a mathematical celebrity. Defined as $f(x) = 1/q$ when $x$ is a rational number $p/q$ in lowest terms, and $f(x) = 0$ otherwise, it is famous for being continuous at every irrational point but discontinuous at every rational point.
Despite its pedagogical importance, its fractal properties—specifically its "dimensions"—were not fully understood. In fractal geometry, dimension tells us how a set fills space as we zoom in. For the popcorn graph, which looks like a series of spikes of decreasing height, the question was: how "thick" is this cloud of points? Does it behave more like a 1D line, a 2D plane, or something in between?
Methods
The authors employ a rigorous "box-counting" approach, but the irregular nature of the popcorn function makes standard fractal techniques insufficient. Instead, they bridge three distinct fields:
- Fractal Geometry: They use the definitions of Box dimension (global scaling) and Assouad dimension/spectrum (local scaling).
- Number Theory (Diophantine Approximation): To count how many points fall into a specific "box," the authors rely on Duffin-Schaeffer type estimates. These help quantify how rational numbers with specific denominators are distributed across the unit interval. They also utilize estimates of the Euler totient function ($\phi(n)$) to count the number of points at each "height" $1/n$.
- Probability Theory: To handle the overlapping of intervals when calculating lower bounds, they apply the Chung-Erdős inequality. This probabilistic tool allows them to estimate the measure of a union of sets even when those sets have complex intersections.
Datasets
As a work of pure mathematics, this paper does not use empirical datasets. The "data" consists of the mathematical properties of the popcorn function $f : [0, 1] \to \mathbb{R}$ and the set of rational numbers.
Key Insights and Ideas
- The 4/3 Threshold: The discovery that the box dimension is exactly 4/3 is the paper's centerpiece. This value indicates that the graph is significantly "thicker" than a simple 1D curve (dimension 1) but doesn't quite fill the 2D plane.
- Local vs. Global Density: A major "aha" moment comes from the Assouad dimension result. While the global box dimension is 4/3, the local Assouad dimension is 2. This means that if you zoom into the "right" spots (near the base of the graph), the set of points can appear as dense as a full 2D area.
- The "Horizontal View": The authors analyze the graph by slicing it into horizontal strips. By counting points in strips of height $\delta$, they transform a complex 2D geometry problem into a 1D counting problem involving denominators, which is where the number theory tools shine.
- Stability under Closure: The authors show that the "full popcorn set" (which includes the entire x-axis) has the same dimensions as the graph itself, demonstrating that the "fuzz" of the rational points dominates the geometry.
Key Results
- Box Dimension: The upper and lower box dimensions of the popcorn graph ($G_f$) are proven to be 4/3.
- Assouad Dimension: The Assouad dimension is proven to be 2, the maximum possible for a subset of $\mathbb{R}^2$.
- Assouad Spectrum: The paper provides a precise formula for the Assouad spectrum ($\dim_A^\theta$), showing a phase transition:
- For $\theta \in (0, 2/3)$, the dimension is $\frac{4/3 - \theta}{1 - \theta}$.
- For $\theta \in [2/3, 1)$, the dimension plateaus at 2.
- Comparison: For context, the simpler set $E = \{1/n : n \in \mathbb{N}\}$ has a box dimension of 1/2. The popcorn graph is significantly more complex due to the distribution of numerators.
Limitations and Future Work
The paper focuses specifically on the classic popcorn function on the interval $[0, 1]$. While the results are definitive for this case, the authors note that the "modified lower dimension" of such sets is not stable under closure, hinting at deeper topological complexities.
Future research might explore:
- Higher Dimensions: Calculating dimensions for multi-variable versions of the popcorn function.
- General Number-Theoretic Functions: Applying these Diophantine-probabilistic hybrid methods to other functions defined by rational properties (e.g., variants of the Riemann zeta function or other arithmetic functions).
- Other Spectra: Investigating different types of fractal spectra (like the multifractal spectrum) for the popcorn graph.
Summary updated