WorksMathematics
Orthogonal Projections and Assouad Dimension: Resolving Fraser's Questions
Abstract
We examine two questions on exceptional orthogonal projections for Assouad dimension. Let and . For a set contained in a -dimensional subspace and containing a nonempty relatively open subset of , we compute both the Hausdorff and packing dimensions of the exceptional -planes as . In particular, a square in has a one-dimensional set of exceptional plane projections. This disproves the bound in Question 17.8.1 of Fraser's book as stated, for every . The planar packing question, Question 17.8.2, is already answered affirmatively by Wu's countability theorem. Using that theorem, we also describe the distinction between planar sets of Assouad dimension at most and those of dimension greater than when they are embedded in higher-dimensional spaces. The computations use elementary Grassmannian geometry; we make no claim of a new planar projection theorem or of an optimal bound for arbitrary higher-dimensional sets.
Introduction
Let denote the Grassmannian of -dimensional linear subspaces of , and let be orthogonal projection onto . We use the metric , where the norm is the operator norm. The Grassmannian is a compact smooth manifold of dimension and carries the rotation-invariant Borel probability measure . This is in [3] and in several other references; throughout this note the first argument denotes the dimension of the subspace. For a nonempty set , define its exceptional set by
Fraser’s Question 17.8.1 [3 Section 17.8], repeated in the update list [5], asks whether
for every nonempty . The motivation is a set lying on a line : all collapse this set to a point, and these subspaces form . In that example the projected set is a singleton because is orthogonal to . Mere transversality to does not imply collapse.
The line example does not account for higher ranks. If is contained in a -plane and has nonempty relative interior there, loss of even one rank gives a projection of dimension less than . Theorem 3.1 computes the dimension of the entire set of such projections. It equals , which exceeds the right-hand side of (2) by when .
Question 17.8.2 asks whether for every nonempty planar . Orponen [6] proved that this exceptional set has Hausdorff dimension 0. Subsequently, Wu [7 Theorem 1.15] proved that it is at most countable. Thus the packing question has an affirmative answer by an existing theorem. Section 4 records this consequence and applies it to sets contained in a two-dimensional subspace. Wu’s theorem is an external premise here, rather than a result proved anew.
The general lower bound for -almost every is established in [2 Theorem 2.9]; see also [3 Theorem 10.2.1]. General exceptional-set estimates obtained from weak tangents are discussed in [2 Theorem 2.10] and [4 Section 2]. These almost-everywhere assertions allow positive-dimensional exceptional sets. The more recent projection results for the Assouad spectrum [1] concern a different dimension and do not remove the rank obstruction in (2).
This note keeps the original Assouad dimension as the source and image dimension. The packing dimension in Question 17.8.2 measures a set of directions, not the dimension of the projected set. We give complete elementary proofs of the flat computations and the deductions from Wu’s theorem. The higher-dimensional line-projection case of (2), and an optimal replacement for general , are not resolved here.
Rank Stratification of the Grassmannian
For a set in Euclidean space, let be the least number of balls of radius covering . The Assouad dimension of is defined by
We use this definition also for unbounded sets, with all positive scales . For bounded sets it agrees with the definition restricted to sufficiently small . Assouad dimension is monotone under inclusion and invariant under bi-Lipschitz maps; it is not in general nonincreasing under Lipschitz maps [3 Chapter 2]. Hausdorff and packing dimensions below are the usual dimensions for the specified metric, with both dimensions of the empty set taken to be 0.
Fix and a -dimensional subspace , where . For , define
The kernel of is . In particular, we have
Lemma 2.1. The stratum is nonempty exactly when
For each such , it is a smooth manifold of dimension
Proof. Write , , and . By (3), has dimension . The image has dimension , since . Thus and , which are exactly (4). Conversely, choose and . Here denotes the -planes in . For a linear map , set
Then and , so (3) gives . This also proves nonemptiness for every feasible .
Every with has a unique representation (6). Indeed, the restriction of to is a linear bijection onto . Its inverse sends to , determining uniquely. In local Grassmannian charts, smooth orthonormal frames for , , and identify these maps with real matrices of size by . Formula (6) is smooth in these coordinates. Conversely, the intersection and image subspaces and this inverse depend smoothly on on any chart where the corresponding fixed-rank minors are invertible. Thus the stratum is the total space of a smooth vector bundle over
whose fiber has dimension . Its dimension is
This description also applies when or ; the corresponding Grassmannian or matrix space is a point.
On compact subsets of a smooth coordinate chart, the chart and its inverse are Lipschitz for the Grassmannian metric. A countable cover by such subsets gives the upper bounds in (5), by countable stability of Hausdorff and packing dimensions. A nonempty coordinate ball contains a bi-Lipschitz image of a Euclidean ball of the stated dimension, giving the matching lower bounds. In dimension 0, the nonempty stratum is locally discrete and has both dimensions 0.
Lemma 2.2. Let contain a nonempty relatively open subset of . For every , we have
Proof. A subset of a -plane has Assouad dimension at most : covering a ball by a regular grid gives for . A ball contained in gives the reverse bound. In fact, volume comparison gives whenever , so no exponent less than satisfies the definition.
Put . The image is contained in the -plane , which gives . For , let be the smallest singular value of . If , then
The volume argument in dimension gives . If , the nonempty image is a singleton.
Flat Sets and the Limitations of the Proposed Bound
Theorem 3.1. Let , , and let , where is a -dimensional subspace of . Suppose that contains a nonempty relatively open subset of . Then
If instead and contains a nonempty open subset of , then .
Proof. Set and . By Lemma 2.2 and (1), we have
For , we have : if , this follows from , and if , it follows from . Thus the union is nonempty. The dimension in (5) strictly increases with in this range, since for ,
Finite the stability of both dimensions and (8) therefore give
One of and is , so the last product is , proving (7). If , the only feasible rank in (4) is , so no projection is exceptional.
Corollary 3.2. For every , the inequality in Question 17.8.1 fails for a compact cube contained in a -dimensional subspace of .
Proof. Apply Theorem 3.1 with . It gives that
Example 3.3. Let and . Write a plane as , where is a unit normal, with and representing the same plane. If , then , so has rank and . If , then its rank is , and is a nondegenerate line segment. Therefore, we have
This is a projective circle, with Hausdorff and packing dimensions . The proposed bound is .
For , formula (7) reduces to , recovering the motivating line example. For , the largest exceptional stratum has codimension in the full Grassmannian. Consequently any bound valid for all nonempty source sets and fixed must be at least . This necessary lower limit on a possible bound does not prove a general upper bound of that size. At the full-dimensional endpoint , the exceptional set is empty, rather than the value obtained by formally extending (7) outside its range.
Planar exceptions and planar sets in higher dimensions
We use the following theorem with its full set-theoretic scope.
No compactness, Borel assumption, or regularity condition on is imposed in this theorem. The proof in [7 Section 5.2] passes from a set to a microscopic limit supporting a measure with ergodic scenery, and uses the countable-exception theorem for projections of that scenery. We invoke the set theorem directly; no such measure hypotheses are added to our statements.
Corollary 4.2. For every nonempty , .
Proof. If the exceptional set is nonempty, write it as a finite or countable union of singletons. Each singleton has packing dimension 0, and packing dimension is countably stable. Theorem 4.1 gives the conclusion; the empty case follows from our convention. Thus Question 17.8.2 is answered by Wu’s theorem.
Proposition 4.3. Let , , and let be a nonempty subset of a two-dimensional subspace . Put .
If or , then
If , then .
If and , then
where an infeasible rank stratum is omitted.
Proof. If , then is bi-Lipschitz onto its image and . If its rank is 0, then is a singleton. If its rank is 1, write and choose a unit vector . For ,
Thus the image in (9) is a similarity of the scalar projection of onto .
(1). Suppose first that . Identify isometrically with , and define
By Theorem 4.1, is at most countable. In the present parameter range . Equation (9) therefore gives
Each set indexed by is a copy of and has both dimensions . The rank-0 set is empty if ; otherwise it is , of dimension . Countable stability in (10) proves the two upper bounds. If , no nonempty projection has negative Assouad dimension, so (1.1) is empty.
(2). Now . Every image of rank at most 1 has Assouad dimension at most 1 and is exceptional. Rank 2 is not exceptional by bi-Lipschitz invariance. Since , rank 1 is feasible by (2.2) with . Lemma 2.1 gives its dimension as
If rank 0 is feasible, its dimension is smaller, so the same value holds for the union.
The second part does not require relative interior or any measure regularity of : only and are used. The first part gives the original proposed bound for a restricted class of sources. It does not establish that bound for arbitrary sets in when and . Together with the flat computation, it distinguishes the known planar answer from the different rank geometry of higher-dimensional projections.
Acknowledgments
The author acknowledges support from his institution, including the Analysis, Probability & Fractals Laboratory (Grant No. LR18ES17).
Statements and Declarations
Competing Interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.
Data Availability
No data was used for the research described in this article.
References
- [1]Kenneth J. Falconer, Jonathan M. Fraser, and Antti Käenmäki. On the Marstrand projection theorem for the Assouad spectrum, 2026. Preprint, arXiv:2606.28830.arxiv.org/abs/2606.28830
- [2]Jonathan M. Fraser. Distance sets, orthogonal projections, and passing to weak tangents. Israel J. Math, 226:851–875, 2018.arxiv.org/abs/1703.06900
- [3]Jonathan M. Fraser. Assouad Dimension and Fractal Geometry, volume 222 of Cambridge Tracts in Mathematics. Cambridge University Press, 2020.
- [4]Jonathan M. Fraser. A nonlinear projection theorem for Assouad dimension and applications. J. Lond. Math. Soc., 107:777–797, 2023.arxiv.org/abs/2004.12001
- [5]Jonathan M. Fraser. Assouad dimension and fractal geometry: updates on open problems, 2025. January 8, 2025.
- [6]Tuomas Orponen. On the Assouad dimension of projections. Proc. Lond. Math. Soc., 122:317–351, 2021.DOI
- [7]Meng Wu. Projection theorems with countably many exceptions and applications to the exact overlaps conjecture, 2025. Preprint, arXiv:2503.21923.arxiv.org/abs/2503.21923