Dimensions of orthogonal projections of typical self-affine sets and measures

This paper establishes that for typical self-affine sets and measures generated by contractive affine iterated function systems, the Hausdorff and box-counting dimensions of their orthogonal projections coincide and are determined by a specific pressure function, while also characterizing the conditions under which the projected measures possess exact dimensionality.

The Gist

Imagine you have a magical, infinitely complex sculpture made of rubber bands and gears. This sculpture is a self-affine set. It's built by taking a shape, shrinking it in different directions (stretching it one way, squishing it another), and sticking copies of it together in a specific pattern. Because the pattern repeats forever, the shape has a "fractal" nature—it looks messy and detailed no matter how much you zoom in.

Now, imagine you have a very bright light source. You shine this light on your sculpture from different angles, and you look at the shadow it casts on a wall.

This paper is about figuring out exactly how "thick" or "complex" that shadow is.

The Main Characters

1. The Sculpture ($K_a$): This is the fractal shape. It's built using a set of rules (matrices) that tell us how to shrink and rotate the pieces. The "translation vector" ($a$) is just the position where we place these pieces. The authors say that if you move the pieces around randomly (which they call "typical"), the shadow behaves in a very predictable way.
2. The Shadow ($P_W K_a$): This is the orthogonal projection. Think of it as the shadow cast on a specific wall (a subspace $W$).
3. The Dimension: In math, "dimension" isn't just 1D (a line) or 2D (a square). Fractals have "fractional" dimensions. A dimension of 1.5 means it's more complex than a line but not quite a full sheet of paper. The paper asks: What is the dimension of the shadow?

The Big Discovery: The "Magic Formula"

For a long time, mathematicians knew how to calculate the dimension of the original sculpture. But calculating the dimension of the shadow was tricky. Sometimes the shadow looks like a simple line (dimension 1), even if the sculpture is complex. Sometimes it looks like a full sheet (dimension 2).

The authors found a "Magic Formula" (called a pressure function) that predicts the dimension of the shadow.

• The Analogy: Imagine you are trying to guess the size of a shadow. You have a calculator that takes the "stretching rules" of your sculpture and the "angle of the wall" as inputs. The calculator spits out a number. The authors proved that for almost any random placement of the sculpture pieces, the shadow's dimension is exactly that number.
• The Surprise: Usually, when you project a complex shape, you lose some complexity (the shadow is smaller than the object). The authors figured out exactly how much complexity is lost based on the geometry of the stretching rules.

The Measure Problem: "Is the Shadow Uniform?"

The paper also looks at measures. Instead of just the shape, imagine the sculpture is made of "dust" where some parts are denser than others (like a cloud that is thick in the middle and thin at the edges). When you cast a shadow of this dusty cloud, is the dust in the shadow spread out evenly?

• The "Exact Dimensional" Question: If the dust in the shadow is perfectly uniform, we say it is "exact dimensional." It's like a perfectly smooth sheet of paper.
• The Bad News: The authors found that for some very specific, weirdly constructed fractals, the shadow is not uniform. It's like a sheet of paper with some spots being thick and others thin, even though the original object was built by simple rules. This happens when the "stretching rules" are a bit chaotic (specifically, when they are "antidiagonal" matrices, which swap the x and y axes in a tricky way).
• The Good News: However, if the dust is distributed in a simple, random way (like a Bernoulli measure, which is like flipping a coin to decide where to put the dust), then the shadow is perfectly uniform.

The "Pivot" and the "Angle"

To solve these problems, the authors used some heavy-duty math tools, but we can visualize them:

• The Pivot: Imagine the shadow wall has a grid. The "pivot" is the specific spot on the grid where the shadow "locks in." The authors realized that the dimension of the shadow depends on which "lock" the shadow hits.
• The Angle: They also looked at the angles between the stretching directions. If the stretching directions are all aligned with the wall, the shadow is big. If they are perpendicular, the shadow is tiny. The paper calculates the "worst-case" and "best-case" angles to find the true dimension.

Why Does This Matter?

This isn't just about abstract shapes. These ideas apply to:

• Image Compression: Understanding how complex patterns look when simplified.
• Physics: Modeling how fluids or materials behave in complex, repeating structures.
• Data Science: Understanding the true "size" or complexity of high-dimensional data when we try to visualize it on a 2D screen (which is essentially a projection).

The Takeaway

The paper says: "If you build a fractal using random rules, its shadow will have a predictable size, calculated by a specific formula. However, if the dust on the fractal is arranged in a very specific, tricky way, the shadow might be messy and uneven. But if the dust is random, the shadow will be perfectly smooth."

They essentially gave us a map to predict the "thickness" of shadows cast by the most complex, self-repeating shapes in mathematics.

Technical Summary

Here is a detailed technical summary of the paper "Dimensions of Orthogonal Projections of Typical Self-Affine Sets and Measures" by De-Jun Feng and Yu-Hao Xie.

1. Problem Statement and Context

The paper investigates the fractal dimensions of orthogonal projections of typical self-affine sets and ergodic stationary measures associated with affine Iterated Function Systems (IFS).

• Setting: Let $\{T_i\}_{i=1}^m$ be a tuple of $d \times d$ invertible real matrices with $\|T_i\| < 1/2$. Let $a = (a_1, \dots, a_m) \in \mathbb{R}^{md}$ be a translation vector. The IFS is defined by $f_i^a(x) = T_i x + a_i$. The attractor is $K_a = \bigcup f_i^a(K_a)$.
• Goal: Determine the Hausdorff and box-counting dimensions of the orthogonal projection $P_W(K_a)$ onto a linear subspace $W \subset \mathbb{R}^d$, and the local dimensions of the projected measure $(P_W \pi_a)_* \mu$, where $\mu$ is an ergodic $\sigma$-invariant measure on the symbolic space $\Sigma$.
• Typicality: The results hold for Lebesgue almost every translation vector $a$ (denoted $L^{md}$-a.e.) and, in some cases, for almost every subspace $W$ with respect to the natural invariant measure on the Grassmann manifold $G(d, k)$.
• Background: While Marstrand's theorem guarantees that for self-similar sets with dense rotations, the projection dimension is $\min\{k, \dim_H K\}$, self-affine sets are more complex due to non-uniform scaling. Previous work (Falconer, Jordan-Pollicott-Simon) established that for typical $a$, $\dim_H K_a = \min\{d, \dim_{AFF}(T)\}$, where $\dim_{AFF}$ is the affinity dimension. This paper extends these results to projections and measures, addressing whether the dimension is preserved or drops, and whether the projected measure is "exact dimensional."

2. Methodology

The authors employ a sophisticated combination of thermodynamic formalism, multiplicative ergodic theory, and linear algebra (specifically exterior algebra and pivot position vectors).

• Subadditive Potentials and Pressure: The authors define a family of subadditive potentials $\psi_s^W(I) = \sup_{J \in \Sigma^*} \phi_s(T_I^* P_{T_J^* W})$, where $\phi_s$ is the singular value function. They relate the dimension of projections to the zero point of the topological pressure associated with these potentials.
• Oseledets' Multiplicative Ergodic Theorem: A central tool is the application of Oseledets' theorem to the matrix cocycle $x \mapsto T_{x_1}^*$. This allows the authors to analyze the asymptotic growth rates of singular values of projected matrices $T_{x|n}^* P_W$.
• Pivot Position Vectors: To handle the interaction between the subspace $W$ and the Lyapunov filtration, the authors introduce pivot position vectors relative to an ordered basis adapted to the Oseledets splitting. This combinatorial tool helps classify the possible values of the projected dimension.
• Transversality: The proofs rely on transversality arguments (similar to Falconer's and Solomyak's) to show that for almost every translation vector $a$, the projected set behaves "generically," avoiding exceptional overlaps that would lower the dimension.
• Counter-Example Construction: To demonstrate the failure of exact dimensionality in general cases, the authors construct specific examples using antidiagonal matrices and Markov measures, exploiting the periodicity of the Lyapunov exponents.

3. Key Contributions and Results

A. Dimensions of Projected Self-Affine Sets (Theorem 1.1)

For a fixed tuple of matrices $T$ and a subspace $W \in G(d, k)$:

1. Coincidence of Dimensions: For $L^{md}$-a.e. $a$, the Hausdorff dimension $\dim_H P_W(K_a)$ and the box-counting dimension $\dim_B P_W(K_a)$ coincide.
2. Formula: This common dimension is determined by the zero point of a specific pressure function:
   $$\dim_H P_W(K_a) = \dim_{AFF}(T, W)$$
   where $\dim_{AFF}(T, W)$ is defined via the convergence of $\sum \phi_s(P_W T_I)$.
3. Finiteness of Values: As $W$ varies over $G(d, k)$, $\dim_{AFF}(T, W)$ takes only finitely many values.
4. Generic Behavior: For $\gamma_{d,k}$-a.e. $W$, $\dim_{AFF}(T, W) = \min\{k, \dim_{AFF}(T)\}$.
5. Positive Measure: If a specific pressure condition is positive, the projected set has positive $k$-dimensional Hausdorff measure.

B. Dimensions of Projected Measures (Theorem 1.2)

For an ergodic $\sigma$-invariant measure $\mu$:

1. Existence of Local Dimensions: For $L^{md}$-a.e. $a$, the local dimension of the projected measure $(P_W \pi_a)_* \mu$ exists $\mu$-a.e.
2. Upper Bound: The local dimension is bounded above by a quantity $S(\mu, T, W)$, which is derived from the Lyapunov exponents and the pivot position of $W$.
3. Exact Dimensionality (Bernoulli/Supermultiplicative): If $\mu$ is a Bernoulli product measure or a supermultiplicative measure (and fully supported), then $(P_W \pi_a)_* \mu$ is exact dimensional for $L^{md}$-a.e. $a$. In this case:
   $$\dim (P_W \pi_a)_* \mu = \min\{k, \dim_{LY}(\mu, T)\}$$
   for $\gamma_{d,k}$-a.e. $W$.
4. Failure of Exact Dimensionality (General Case): Crucially, the authors show that for general ergodic measures, the projected measure may not be exact dimensional. They construct an example (Example 8.1) where $S(\mu, T, W) \neq \overline{S}(\mu, T, W)$, implying the local dimension varies depending on the point $x$.

C. Specific Criteria for Dimension Drop

The paper provides verifiable criteria (Propositions 7.1–7.4) for when the projected dimension is strictly less than the generic value $\min\{k, \dim_{AFF}(T)\}$ or $\min\{k, \dim_{LY}(\mu, T)\}$. These conditions typically involve the reducibility of the matrix tuple or specific invariance properties of the subspace $W$ under the transpose matrices $T_i^*$.

4. Significance and Implications

• Resolution of Open Questions: The paper resolves the question of whether typical self-affine projections preserve dimension. It confirms that for "nice" measures (Bernoulli/supermultiplicative), the dimension is preserved (up to the subspace dimension), but for general ergodic measures, dimension drop can occur, and the projected measure may fail to be exact dimensional.
• Counter-Intuitive Phenomenon: The discovery that $(P_W \pi_a)_* \mu$ can fail to be exact dimensional for almost every $a$ (even when the original measure is ergodic) is a significant and unexpected result in fractal geometry. It contrasts with the behavior of self-similar sets with dense rotations.
• Unified Framework: The authors unify the study of set dimensions and measure dimensions under a single framework involving subadditive pressures and Oseledets filtrations.
• Generalization: The results extend beyond orthogonal projections to general linear projections of rank $k$.
• Independence: The work was done independently and simultaneously with Morris and Sert [45], who obtained similar results, validating the robustness of the findings.

5. Conclusion

Feng and Xie provide a comprehensive theory for the dimensions of projections of typical self-affine sets and measures. They establish that while the dimension of the projected set is generically determined by a pressure function, the behavior of projected measures is more subtle: they are exact dimensional only under specific structural assumptions (like supermultiplicativity). The paper highlights the critical role of the interplay between the geometry of the subspace $W$ and the Lyapunov filtration of the matrix cocycle, offering precise criteria for when exceptional dimension drops occur.