betaFeedback

WorksMathematics

Orthogonal Projections and Assouad Dimension: Resolving Fraser's Questions

Abstract

We examine two questions on exceptional orthogonal projections for Assouad dimension. Let 1≤k<d1\le k<d and 1≤q<d1\le q<d. For a set FF contained in a qq-dimensional subspace W⊂RdW\subset\R^d and containing a nonempty relatively open subset of WW, we compute both the Hausdorff and packing dimensions of the exceptional kk-planes as k(d−k)−∣q−k∣−1k(d-k)-|q-k|-1. In particular, a square in R3\R^3 has a one-dimensional set of exceptional plane projections. This disproves the bound k(d−1−k)k(d-1-k) in Question 17.8.1 of Fraser's book as stated, for every 2≤k<d2\le k<d. 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 11 and those of dimension greater than 11 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 Gr⁡(k,d)\operatorname{Gr}(k,d) denote the Grassmannian of kk-dimensional linear subspaces of Rd\mathbb{R}^{d}, and let PVP_V be orthogonal projection onto V∈Gr⁡(k,d)V \in\operatorname{Gr}(k,d). We use the metric ∥PV−PV′∥\|P_V-P_{V'}\|, where the norm is the operator norm. The Grassmannian is a compact smooth manifold of dimension k(d−k)k(d-k) and carries the rotation-invariant Borel probability measure γd,k\gamma_{d,k}. This is G(k,d)G(k,d) in [3] and G(d,k)G(d,k) in several other references; throughout this note the first argument denotes the dimension of the subspace. For a nonempty set F⊂RdF \subset\mathbb{R}^{d}, define its exceptional set by

Ek(F)={V∈Gr⁡(k,d):dim⁡APVF<min⁡{k,dim⁡AF}}.(1)\mathcal{E}_k(F)=\{V\in\operatorname{Gr}(k,d):\dim_A P_VF<\min\{k,\dim_A F\}\}. \tag*{(1)}

Fraser’s Question 17.8.1 [3 Section 17.8], repeated in the update list [5], asks whether

dim⁡HEk(F)≤k(d−1−k)(1≤k<d)(2)\dim_H\mathcal{E}_k(F)\le k(d-1-k)\qquad(1\le k<d) \tag*{(2)}

for every nonempty F⊂RdF \subset\mathbb{R}^{d}. The motivation is a set lying on a line LL: all V⊂L⊥V \subset L^\perp collapse this set to a point, and these subspaces form Gr⁡(k,d−1)\operatorname{Gr}(k,d-1). In that example the projected set is a singleton because VV is orthogonal to LL. Mere transversality to LL does not imply collapse.

The line example does not account for higher ranks. If FF is contained in a kk-plane and has nonempty relative interior there, loss of even one rank gives a projection of dimension less than kk. Theorem 3.1 computes the dimension of the entire set of such projections. It equals k(d−k)−1k(d-k)-1, which exceeds the right-hand side of (2) by k−1k-1 when k≥2k \ge2.

Question 17.8.2 asks whether dim⁡PE1(F)=0\dim_{\mathrm{P}} \mathcal{E}_1(F)=0 for every nonempty planar FF. 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 dim⁡APVF≥min⁡{k,dim⁡AF}\dim_{\mathrm{A}} P_VF \ge\min\{k,\dim_{\mathrm{A}}F\} for γd,k\gamma_{d,k}-almost every VV 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 kk, are not resolved here.

Rank Stratification of the Grassmannian

For a set EE in Euclidean space, let Nr(E)N_r(E) be the least number of balls of radius rr covering EE. The Assouad dimension of EE is defined by

dim⁡AE=inf⁡{s≥0: ∃C>0 ∀x∈E ∀0<r<R, Nr(E∩B(x,R))≤C(R/r)s}.\dim_{\mathrm{A}} E=\inf\left\{s\ge0:\ \exists C>0\ \forall x\in E\ \forall0<r<R,\ N_r(E\cap B(x,R))\le C(R/r)^s\right\}.

We use this definition also for unbounded sets, with all positive scales r<Rr<R. For bounded sets it agrees with the definition restricted to sufficiently small RR. 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 1≤k<d1\le k<d and a qq-dimensional subspace W⊂RdW\subset\mathbb{R}^d, where 1≤q≤d1\le q\le d. For 0≤j≤min⁡{q,k}0\le j\le\min\{q,k\}, define

Sj(W)={V∈Gr⁡(k,d):rank⁡(PV∣W)=j}.S_j(W)=\{V\in\operatorname{Gr}(k,d):\operatorname{rank}(P_V|_W)=j\}.

The kernel of PV∣WP_V|_W is W∩V⊥W\cap V^\perp. In particular, we have

V∈Sj(W)if and only ifdim⁡(W∩V⊥)=q−j.(3)V\in S_j(W)\quad\text{if and only if}\quad\dim(W\cap V^\perp)=q-j. \tag*{(3)}

Lemma 2.1. The stratum Sj(W)S_j(W) is nonempty exactly when

max⁡{0,q+k−d}≤j≤min⁡{q,k}.(4)\max\{0,q+k-d\}\le j\le\min\{q,k\}. \tag*{(4)}

For each such jj, it is a smooth manifold of dimension

dim⁡HSj(W)=dim⁡PSj(W)=k(d−k)−(q−j)(k−j).(5)\dim_{\mathrm{H}}S_j(W)=\dim_{\mathrm{P}}S_j(W)=k(d-k)-(q-j)(k-j). \tag*{(5)}

Proof. Write U=V⊥U=V^\perp, a=q−ja=q-j, and b=d−k−q+jb=d-k-q+j. By (3), A=U∩WA=U\cap W has dimension aa. The image B=PW⊥UB=P_{W^\perp}U has dimension bb, since dim⁡U=d−k=a+b\dim U=d-k=a+b. Thus 0≤a≤q0\le a\le q and 0≤b≤d−q0\le b\le d-q, which are exactly (4). Conversely, choose A∈Gr⁡(a,W)A \in\operatorname{Gr}(a,W) and B∈Gr⁡(b,W⊥)B \in\operatorname{Gr}(b,W^\perp). Here Gr⁡(a,W)\operatorname{Gr}(a,W) denotes the aa-planes in WW. For a linear map T:B→W∩A⊥T:B \to W \cap A^\perp, set

U=A⊕{z+Tz:z∈B}andV=U⊥.(6)U = A \oplus\{z + Tz : z \in B\} \quad\text{and}\quad V = U^\perp. \tag*{(6)}

Then dim⁡U=a+b=d−k\dim U = a+b = d-k and U∩W=AU \cap W = A, so (3) gives V∈Sj(W)V \in\mathcal{S}_j(W). This also proves nonemptiness for every feasible jj.

Every UU with U∩W=AU \cap W = A has a unique representation (6). Indeed, the restriction of PW⊥P_{W^\perp} to U∩A⊥U \cap A^\perp is a linear bijection onto BB. Its inverse sends zz to z+Tzz+Tz, determining TT uniquely. In local Grassmannian charts, smooth orthonormal frames for AA, W∩A⊥W \cap A^\perp, and BB identify these maps TT with real matrices of size jj by bb. Formula (6) is smooth in these coordinates. Conversely, the intersection and image subspaces and this inverse depend smoothly on UU on any chart where the corresponding fixed-rank minors are invertible. Thus the stratum is the total space of a smooth vector bundle over

Gr⁡(a,W)×Gr⁡(b,W⊥),\operatorname{Gr}(a,W) \times\operatorname{Gr}(b,W^\perp),

whose fiber has dimension jbjb. Its dimension is

a(q−a)+b(d−q−b)+jb=aj+bk=k(d−k)−(q−j)(k−j).a(q-a) + b(d-q-b) + jb = aj + bk = k(d-k) - (q-j)(k-j).

This description also applies when a=0a=0 or b=0b=0; 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 F⊂WF \subset W contain a nonempty relatively open subset of WW. For every V∈Gr⁡(k,d)V \in\operatorname{Gr}(k,d), we have

dim⁡AF=qanddim⁡APVF=rank⁡(PV∣W).\dim_{\mathrm{A}} F = q \quad\text{and}\quad\dim_{\mathrm{A}} P_V F = \operatorname{rank}(P_V|_W).

Proof. A subset of a qq-plane has Assouad dimension at most qq: covering a ball by a regular grid gives Nr(E∩B(x,R))≤Cq(R/r)qN_r(E \cap B(x,R)) \le C_q(R/r)^q for 0<r<R0<r<R. A ball contained in FF gives the reverse bound. In fact, volume comparison gives Nr(BW(x,R))≥(R/r)qN_r(B_W(x,R)) \ge(R/r)^q whenever BW(x,R)⊂FB_W(x,R) \subset F, so no exponent less than qq satisfies the definition.

Put j=rank⁡(PV∣W)j = \operatorname{rank}(P_V|_W). The image PVFP_VF is contained in the jj-plane PVWP_VW, which gives dim⁡APVF≤j\dim_{\mathrm{A}} P_VF \le j. For j>0j>0, let σ>0\sigma>0 be the smallest singular value of PV∣W∩ker⁡(PV∣W)⊥P_V|_{W \cap\ker(P_V|_W)^\perp}. If BW(x,R)⊂FB_W(x,R) \subset F, then

BPVW(PVx,σR)⊂PV(BW(x,R))⊂PVF.B_{P_VW}(P_Vx,\sigma R) \subset P_V(B_W(x,R)) \subset P_VF.

The volume argument in dimension jj gives dim⁡APVF≥j\dim_{\mathrm{A}} P_VF \ge j. If j=0j=0, the nonempty image is a singleton.

Flat Sets and the Limitations of the Proposed Bound

Theorem 3.1. Let 1≤k<d1 \le k < d, 1≤q<d1 \le q < d, and let F⊂WF \subset W, where WW is a qq-dimensional subspace of Rd\mathbb{R}^d. Suppose that FF contains a nonempty relatively open subset of WW. Then

dim⁡HEk(F)=dim⁡PEk(F)=k(d−k)−∣q−k∣−1.(7)\dim_{\mathrm{H}} \mathcal{E}_k(F) = \dim_{\mathrm{P}} \mathcal{E}_k(F) = k(d-k)-|q-k|-1. \tag*{(7)}

If instead q=dq=d and FF contains a nonempty open subset of Rd\mathbb{R}^d, then Ek(F)=∅\mathcal{E}_k(F)=\varnothing.

Proof. Set m=min⁡{q,k}m = \min\{q,k\} and j0=max⁡{0,q+k−d}j_0 = \max\{0,q+k-d\}. By Lemma 2.2 and (1), we have

Ek(F)=⋃j=j0m−1Sj(W).(8)\mathcal{E}_k(F) = \bigcup_{j=j_0}^{m-1} \mathcal{S}_j(W). \tag*{(8)}

For q<dq<d, we have j0≤m−1j_0\le m-1: if q≤kq\le k, this follows from k≤d−1k\le d-1, and if k≤qk\le q, it follows from q≤d−1q\le d-1. Thus the union is nonempty. The dimension in (5) strictly increases with jj in this range, since for j<mj<m,

[k(d−k)−(q−j−1)(k−j−1)]−[k(d−k)−(q−j)(k−j)]=q+k−2j−1>0.\left[k(d-k)-(q-j-1)(k-j-1)\right]-\left[k(d-k)-(q-j)(k-j)\right]=q+k-2j-1>0.

Finite the stability of both dimensions and (8) therefore give

dim⁡HEk(F)=dim⁡PEk(F)=k(d−k)−(q−m+1)(k−m+1).\dim_{\mathrm{H}}\mathcal{E}_k(F)=\dim_{\mathrm{P}}\mathcal{E}_k(F)=k(d-k)-(q-m+1)(k-m+1).

One of q−mq-m and k−mk-m is 00, so the last product is ∣q−k∣+1\lvert q-k\rvert+1, proving (7). If q=dq=d, the only feasible rank in (4) is j=kj=k, so no projection is exceptional.

Corollary 3.2. For every 2≤k<d2\le k<d, the inequality in Question 17.8.1 fails for a compact cube contained in a kk-dimensional subspace of Rd\mathbb{R}^d.

Proof. Apply Theorem 3.1 with q=kq=k. It gives that

dim⁡HEk(F)=k(d−k)−1=k(d−1−k)+(k−1)>k(d−1−k).\dim_{\mathrm{H}}\mathcal{E}_k(F)=k(d-k)-1=k(d-1-k)+(k-1)>k(d-1-k).

Example 3.3. Let F=[−1,1]2×{0}⊂R3F=[-1,1]^2\times\{0\}\subset\mathbb{R}^3 and W=R2×{0}W=\mathbb{R}^2\times\{0\}. Write a plane VV as n⊥n^\perp, where n∈S2n\in S^2 is a unit normal, with nn and −n-n representing the same plane. If n∉Wn\notin W, then W∩Rn={0}W\cap\mathbb{R}n=\{0\}, so PV∣WP_V|_W has rank 22 and dim⁡APVF=2\dim_{\mathrm{A}}P_VF=2. If n∈Wn\in W, then its rank is 11, and PVFP_VF is a nondegenerate line segment. Therefore, we have

E2(F)={n⊥:n∈W∩S2}.\mathcal{E}_2(F)=\{n^\perp:n\in W\cap S^2\}.

This is a projective circle, with Hausdorff and packing dimensions 11. The proposed bound is 2(3−1−2)=02(3-1-2)=0.

For q=1q=1, formula (7) reduces to k(d−1−k)k(d-1-k), recovering the motivating line example. For q=kq=k, the largest exceptional stratum has codimension 11 in the full Grassmannian. Consequently any bound valid for all nonempty source sets and fixed d,kd,k must be at least k(d−k)−1k(d-k)-1. This necessary lower limit on a possible bound does not prove a general upper bound of that size. At the full-dimensional endpoint q=dq=d, 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.

Theorem 4.1 (Wu [7], Theorem 1.15). For every nonempty set F⊂R2F\subset\mathbb{R}^2, the set

{L∈Gr⁡(1,2):dim⁡APLF<min⁡{1,dim⁡AF}}\{L\in\operatorname{Gr}(1,2):\dim_{\mathrm{A}}P_LF<\min\{1,\dim_{\mathrm{A}}F\}\}

is at most countable.

No compactness, Borel assumption, or regularity condition on FF 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 F⊂R2F \subset\mathbb{R}^{2}, dim⁡PE1(F)=0\dim_{\mathrm{P}} E_{1}(F)=0.

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 d≥2d \ge2, 1≤k<d1 \le k < d, and let FF be a nonempty subset of a two-dimensional subspace W⊂RdW \subset\mathbb{R}^{d}. Put s=dim⁡AFs=\dim_{\mathrm{A}} F.

  1. If k=1k=1 or 0<s≤10<s \le1, then

dim⁡HEk(F)≤k(d−1−k)anddim⁡PEk(F)≤k(d−1−k).\dim_{\mathrm{H}} E_{k}(F) \le k(d-1-k) \quad\text{and} \quad\dim_{\mathrm{P}} E_{k}(F) \le k(d-1-k).

If s=0s=0, then Ek(F)=∅E_{k}(F)=\varnothing.

  1. If k≥2k \ge2 and 1<s≤21<s \le2, then

Ek(F)=S0(W)∪S1(W)anddim⁡HEk(F)=dim⁡PEk(F)=k(d−1−k)+1,E_{k}(F)=S_{0}(W)\cup S_{1}(W) \quad\text{and} \quad\dim_{\mathrm{H}} E_{k}(F)=\dim_{\mathrm{P}} E_{k}(F)=k(d-1-k)+1,

where an infeasible rank stratum is omitted.

Proof. If rank⁡(PV∣W)=2\operatorname{rank}(P_{V}|_{W})=2, then PV∣WP_{V}|_{W} is bi-Lipschitz onto its image and dim⁡APVF=s\dim_{\mathrm{A}}P_{V}F=s. If its rank is 0, then PVFP_{V}F is a singleton. If its rank is 1, write L=W∩V⊥L=W\cap V^{\perp} and choose a unit vector u∈W∩L⊥u\in W\cap L^{\perp}. For w∈Ww\in W,

PVw=⟨w,u⟩PVu,PVu≠0.(9)P_{V}w=\langle w,u\rangle P_{V}u,\qquad P_{V}u\ne0. \tag*{(9)}

Thus the image in (9) is a similarity of the scalar projection of FF onto Ru\mathbb{R}u.

(1). Suppose first that s>0s>0. Identify WW isometrically with R2\mathbb{R}^{2}, and define

C={L∈Gr⁡(1,W):dim⁡APW∩L⊥F<min⁡{1,s}}.C=\{L\in\operatorname{Gr}(1,W):\dim_{\mathrm{A}}P_{W\cap L^{\perp}}F<\min\{1,s\}\}.

By Theorem 4.1, CC is at most countable. In the present parameter range min⁡{k,s}=min⁡{1,s}\min\{k,s\}=\min\{1,s\}. Equation (9) therefore gives

Ek(F)⊂S0(W)∪⋃L∈C{V∈Gr⁡(k,d):V⊂L⊥}.(10)E_{k}(F)\subset S_{0}(W)\cup\bigcup_{L\in C}\{V\in\operatorname{Gr}(k,d):V\subset L^{\perp}\}. \tag*{(10)}

Each set indexed by LL is a copy of Gr⁡(k,d−1)\operatorname{Gr}(k,d-1) and has both dimensions k(d−1−k)k(d-1-k). The rank-0 set is empty if k>d−2k>d-2; otherwise it is Gr⁡(k,W⊥)\operatorname{Gr}(k,W^{\perp}), of dimension k(d−2−k)k(d-2-k). Countable stability in (10) proves the two upper bounds. If s=0s=0, no nonempty projection has negative Assouad dimension, so (1.1) is empty.

(2). Now min⁡{k,s}=s>1\min\{k,s\}=s>1. 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 d>k≥2d>k\ge2, rank 1 is feasible by (2.2) with q=2q=2. Lemma 2.1 gives its dimension as

k(d−k)−(2−1)(k−1)=k(d−1−k)+1.k(d-k)-(2-1)(k-1)=k(d-1-k)+1.

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 FF: only F⊂WF\subset W and dim⁡AF>1\dim_{\mathrm{A}}F>1 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 Rd\mathbb{R}^{d} when k=1k=1 and d≥3d\ge3. 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. [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. [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. [3]Jonathan M. Fraser. Assouad Dimension and Fractal Geometry, volume 222 of Cambridge Tracts in Mathematics. Cambridge University Press, 2020.
  4. [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. [5]Jonathan M. Fraser. Assouad dimension and fractal geometry: updates on open problems, 2025. January 8, 2025.
  6. [6]Tuomas Orponen. On the Assouad dimension of projections. Proc. Lond. Math. Soc., 122:317–351, 2021.DOI
  7. [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

Paper details

vlnc:W_hwrr8dt2
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 4

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents