Seventy Years of Fractal Projections

Abstract: Seventy years ago, John Marstrand published a paper which, among other things, relates the Hausdorff dimension of a plane set to the dimensions of its orthogonal projections onto lines. For some time this paper attracted little attention, but over the past 40 years Marstrand's projection theorems have become the prototype for many results in fractal geometry with numerous variants and applications and they continue to motivate leading research.

• The paper’s main contribution is its comprehensive review of Marstrand-type projection theorems and their extensions using advanced harmonic analysis.
• It employs methodologies such as energy integrals and Fourier transforms to obtain precise bounds for exceptional sets in fractal projections.
• The survey also highlights applications to self-similar, self-affine, and nonlinear projection settings, linking them to ergodic theory and additive combinatorics.

Survey of Fractal Projections: Seventy Years Since Marstrand

Introduction and Historical Context

The seventy years since Marstrand's seminal 1954 paper have witnessed the evolution of projection theorems from foundational results in geometric measure theory into a central theme in fractal geometry. Marstrand’s projection theorems, relating the Hausdorff dimension of a planar set to those of its orthogonal projections, remained largely unnoticed for decades but, spurred by developments in ergodic theory, harmonic analysis, and additive combinatorics from the 1980s onwards, have become a paradigm for “Marstrand-type” results in both theoretical and applied settings.

Marstrand’s original bounds have been greatly refined and generalized, with modern proofs utilizing potential-theoretic methods, Fourier analysis, and energy integrals. Current research addresses high-dimensional analogues, exceptions, projections under alternative dimensional definitions, and the geometry of projections under structural regularity assumptions on the sets, such as self-similarity or self-affinity.

The Marstrand Projection Theorem and Extensions

Let $E \subset \mathbb{R}^2$ be a Borel (or analytic) set, and let $\operatorname{proj}_\theta$ denote orthogonal projection onto the line through the origin at angle $\theta$. Marstrand’s theorem asserts two principal claims:

1. $$\dim_{\rm H} \operatorname{proj}_\theta E \leq \min\{\dim_{\rm H} E, 1\}$$ for all $\theta$, and equality holds for almost every $\theta$.
2. If $\dim_{\rm H} E > 1$, then $\operatorname{proj}_\theta E$ has positive Lebesgue measure for almost all $\theta$.

These extend to projections from $\mathbb{R}^n$ onto $m$-dimensional subspaces $V \in G(n,m)$ (the Grassmannian), again with equality for almost every projection and positive $m$-dimensional measure when the dimension of $E$ exceeds $m$.

The proofs rely on energy integrals

$I^s(\mu) := \iint \frac{d\mu(x)d\mu(y)}{|x - y|^s}$

and the transformation of these via the Fourier transform. These analytic techniques decisively shifted the field from combinatorial or geometric methods to harmonic and probabilistic frameworks.

Figure 1: Projection of a set $E$ onto a line in direction $\theta$.

Exceptional Sets and Sharp Bounds

Although the set of exceptional directions for which the "expected" dimension drop fails is of measure zero, a core question became the quantitative nature of these exceptional sets. Kaufman's and Falconer's potential-theoretic and Fourier transform approaches led to explicit upper bounds for the Hausdorff dimension of the set of "bad" projections; e.g., if $0 \leq s < \dim_{\rm H} E < m$, then the set of $V$ for which $\dim_{\rm H} \operatorname{proj}_V E < s$ has dimension at most $m(n-m) - (m-s)$.

Perhaps most striking, Oberlin's conjecture predicted that for planar sets and $0 \leq s \leq \min\{\dim_{\rm H} E, 1\}$, the set of $\theta$ with $\dim_{\rm H} \operatorname{proj}_\theta E < s$ has dimension at most $\max\{2s - \dim_{\rm H} E, 0\}$. This conjecture has now been resolved and shown to be sharp, facilitated by deep connections to the Furstenberg set problem and modern sum-product techniques.

In higher dimensions, optimal bounds are in some regimes unavailable, but partial results and sharp estimates (especially in low dimensions or under additional hypotheses) establish a nuanced dividing line between generic and exceptional behavior.

Alternative Dimension Definitions and Projection Theorems

Projection theorems have been generalized to other notions of fractal dimension:

• Box-counting Dimension: The almost sure box-counting dimension of projections may be strictly less than the minimum of the original set's box dimension and the target dimension, as explicit constructions reveal. Sharp dimension profiles and capacities with respect to suitable kernels underpin modern theorems in this context.
• Packing Dimension and Dimension Profiles: Packing dimension projection results closely parallel those for box dimension, especially when using covering decompositions. There are sharp upper bounds for the dimensions of exceptional projections, and refined lower bounds under measure-theoretic refinements.
• Intermediate and Assouad Dimension: Recently introduced notions such as intermediate dimensions interpolate between Hausdorff and box dimension, and analogous projection theorems, along with exceptional set estimates, have been developed. For Assouad dimension, highly non-generic phenomena can occur: there is no Marstrand-type almost-sure theorem, and the dimension of projections can behave pathologically.
• Fourier Dimension: The Fourier dimension of projections relates elegantly via the invariance of the Fourier transform under orthogonal projections, but Marstrand-type theorems for Fourier dimension remain open. For so-called Salem sets (where Fourier and Hausdorff dimensions coincide), all projections attain the expected dimension, but in general, the picture is subtler.

Restricted Direction Families and Nonlinear Projections

A rich line of inquiry considers projections along parameterized curves or submanifolds within the Grassmannian. For families with non-degeneracy or curvature, projection theorems extend with lower bounds for the Hausdorff dimension and sharp exceptional set estimates. For example, for non-degenerate families in $\mathbb{R}^3$, the almost sure dimension of line or plane projections matches the expected minimum, accompanied by sharp bounds for exceptional parameter values. Techniques encompass decoupling, circular maximal functions, and modern Fourier analytic decoupling.

Self-Similar, Self-Affine, and Self-Conformal Sets

For sets with analytic or algebraic regularity—such as attractors of iterated function systems (IFSs)—the dimension of projections in specific or all directions becomes tractable:

• Self-Similar Sets with Dense Rotations: If the rotation group generated by the IFS is dense, then all projections realize the minimal of the set and target dimension, and, up to countable exceptions, this persists under images by all $C^1$ maps without singularities.
• Self-Affine Sets: For strongly irreducible (and typically, strongly separated) self-affine IFSs, both the set and almost all projections realize the affinity dimension. In higher dimensions ($n \geq 4$), new algebraic obstructions appear: the projection dimension can genuinely depend on the target subspace, with only a finite stratification of possible values.
• Self-Conformal Sets: For conformal IFSs with dense group actions, analogous projection theorems are obtained via dynamical systems and CP-chain constructions.

These structural results resolve the behavior of projections in all or almost all directions for broad and important classes of fractals.

Radial and Nonlinear Projections

Orthogonal projections are a special case; radial projections, i.e., mapping points to their direction from a fixed center, present further subtlety. Recent work establishes Marstrand-type theorems and sharp exceptional set bounds for radial projections, both in the plane and in higher dimensions, with strong analogies to the orthogonal case but via novel projective geometric and Fourier analytic techniques.

Further Directions

Additional directions of extension and research include:

• Projections of Random Sets: For random fractals, such as Mandelbrot (fractal) percolation, zero-one laws ensure almost sure Marstrand-type conclusions for all projections in many settings.
• Generalized and Nonlinear Projections: The development of transversality frameworks (Peres-Schlag) allows Marstrand-type results to transfer to many classes of nonlinear images, under suitable non-degeneracy hypotheses.
• Projections in Non-Euclidean and Anisotropic Geometries: The study extends to Heisenberg groups, normed spaces, Riemannian and hyperbolic geometry, and even infinite-dimensional Banach spaces, with tailored projection theorems valid in these contexts.
• Visibility and Geometric Tomography: Dimension bounds for visible parts of fractals (from infinity or from points) remain a vibrant and partially unresolved area.

Numerical and Qualitative Highlights

Several strong or surprising claims are emphasized and, where resolved in the recent literature, proven to be sharp:

• For planar sets $E$ of Hausdorff dimension $s$, the set of directions for which the projection’s dimension falls below $t$ has Hausdorff dimension at most $\max\{2t - s, 0\}$—this is both strong in its explicitness and proven to be optimal.
• For typical self-similar sets with dense rotations, the dimension of all orthogonal projections attains the maximal possible value in all directions, not almost all.
• For self-affine sets in low (2 and 3) dimensions, under mild irreducibility, all projections almost surely realize the affinity dimension; in higher dimensions, the exceptional behavior is algebraically stratified.
• Exceptional set dimensions for projections, both in orthogonal and radial settings, are now known to satisfy sharp Kaufman-type and Oberlin-type bounds.

Implications and Future Directions

From a theoretical perspective, projection theorems play an essential role in connecting fractal geometry to harmonic analysis, probability, additive combinatorics, and dynamical systems. The analytic strategies imported from potential theory and harmonic analysis allow the field to address questions in geometric measure theory that were previously tractable only in special or low-dimensional settings. Applications range from quantitative Diophantine approximation to embedding theorems, and from topology to the spectral theory of random Schrödinger operators.

Future directions include:

• Extending sharp exceptional set bounds to mixed (e.g., nonorthogonal, nonlinear, or stochastic) projection families in higher dimensions,
• Characterizing the dimension and measure-theoretic properties of projections for classes of dynamically defined measures,
• Analyzing projection phenomena in a broader class of metric and non-Archimedean spaces, including advancements in the context of Heisenberg and Carnot groups, and
• Deepening connections to number theory, e.g., via restricted projection results in Diophantine contexts.

Conclusion

Seventy years after Marstrand’s initial insights, the study of fractal projections stands out as a mature field with robust analytic methods, precise dimension and measure-theoretic results, and rich interactions with several branches of mathematics. The landscape for projections of fractal sets is both highly structured when regularity is present (e.g., for self-similar or self-affine sets) and sharply quantified in terms of exceptional directions and parameter regimes. Open problems in higher-dimensional and nonlinear contexts continue to stimulate the development of advanced techniques across analysis, geometry, and probability.