Lipschitz surjections between self-similar sets
Abstract: We give a characterization of the existence of Lipschitz surjections between dust-like self-similar sets of the same dimension under the assumption that the similarity ratios are commensurable. Our results apply more generally to graph-directed fractals.
Sign up to identify related papers:
Summary
- The paper characterizes the existence of Lipschitz surjections between dust-like self-similar sets, stating that a Lipschitz surjection $X → Y$ exists if and only if the union of similarity ratios $R(Φ) ∪ R(Ψ)$ is commensurable.
- An important corollary from the main theorem suggests that if either set of similarity ratios is commensurable, then the existence of a Lipschitz surjection in one direction ensures its existence in the reverse direction, even though Lipschitz surjections are not generally symmetric.
- The methodology involves a reduction to homogeneous primitive systems, addressing the general framework of graph-directed iterated function systems (GD-IFSs)
Overview and main result
This paper, by Attila Gáspár (2608.16471), studies the existence of Lipschitz surjections between dust-like self-similar sets of equal Hausdorff dimension. Recall that a compact set X⊂Rd is dust-like if it is generated by contracting similarities S1,…,Sn with pairwise disjoint pieces Si(X). Prior work established that a Lipschitz surjection X→Y exists whenever dimHX>dimBY (Balka–Keleti), and that when dimHX=dimHY, the existence of a Lipschitz surjection is equivalent to the existence of a bi-Lipschitz map provided either X or Y is homogeneous (all similarity ratios equal). The question of whether this equivalence persists without homogeneity was answered negatively by a counterexample of Ruan–Xiao.
The paper's main contribution is a characterization under a strictly weaker hypothesis: commensurability of similarity ratios. A set of ratios is commensurable if all ratios logr1/logr2 are rational. The main theorem states:
Let X and S1,…,Sn0 be dust-like self-similar sets with IFSs S1,…,Sn1, S1,…,Sn2, both of dimension S1,…,Sn3. If either S1,…,Sn4 or S1,…,Sn5 is commensurable, then there exists a Lipschitz surjection S1,…,Sn6 if and only if S1,…,Sn7 is commensurable.
A notable structural feature is the symmetry of the right-hand condition in S1,…,Sn8 and S1,…,Sn9. This yields a corollary that is arguably counterintuitive: if one of the two sets has commensurable ratios and they share dimension, then the existence of a Lipschitz surjection in one direction forces its existence in the reverse direction — even though Lipschitz surjections are not symmetric objects in general. The author explicitly leaves open whether this reversibility holds without any commensurability assumption.
The results are proved in the more general framework of graph-directed iterated function systems (GD-IFSs) in the sense of Mauldin–Williams: finite directed graphs whose edges carry contracting similarities, with attractors satisfying Si(X)0. The paper assumes GD-IFSs are dust-like and strongly connected, which guarantees all attractors have common dimension Si(X)1 with positive finite Si(X)2-Hausdorff measure. In this setting, the role of individual ratios is played by cycle ratios Si(X)3, and the theorem holds verbatim with Si(X)4 the set of cycle ratios.
The graph-directed generality is not cosmetic: the proof genuinely requires it, since the key reduction (Proposition on powering homogeneous systems below) has no analogue for ordinary IFSs.
Reduction to homogeneous primitive systems
The first substantive ingredient shows that commensurability of cycle ratios lets one replace an arbitrary GD-IFS by a primitive one — homogeneous, strongly connected, with cycle-length GCD 1 — having the same attractor.
The argument combines three steps. First, a combinatorial potential lemma: if every directed cycle in a strongly connected graph has positive integer weight, then edge weights can be shifted by a potential so as to become nonnegative integers, positive on edges into a chosen base vertex. Applying this with weights Si(X)5 (where Si(X)6 generates the cycle-ratio group), rescaling each attractor by a power of Si(X)7, and conjugating similarities yields a homogeneous system whose cycles have ratio less than 1 but which may contain isometric edges (ratio exactly 1). Isometries are eliminated by an edge-composition ("elimination") procedure that preserves strong connectivity, dust-likeness, the GCD invariant, and crucially cannot delete the base vertex because it was chosen to lack incoming isometric edges. Finally, edges of ratio Si(X)8 are subdivided into paths of Si(X)9 edges of ratio X→Y0, introducing intermediate vertices whose attractors are scaled copies of existing ones.
Two further facts about primitive dust-like GD-IFSs are recorded: the system can be powered to any ratio X→Y1 via length-X→Y2 walk refinements (with adjacency matrix X→Y3 still primitive); and if one component of the measure vector lies in the ring X→Y4 where X→Y5, then all components do. The latter follows from a linear-algebra argument: the Perron eigenspace of the integer adjacency matrix is one-dimensional, forcing the measure vector to coincide with a solution over X→Y6, and X→Y7 is algebraic so ring and field coincide.
Measure-linear surjections
The second section establishes, without any commensurability assumption, that a Lipschitz map X→Y8 with positive-measure image gives rise to a Lipschitz surjection from a disjoint union of scaled copies of the X→Y9 onto each dimHX>dimBY0, minimizing domain measure among such maps, whose pushforward is a constant multiple of Hausdorff measure. This generalizes Ruan–Xiao's work; notably, the resulting covering lemma (each small cell's image contains a cell of comparable diameter) is stronger than its IFS predecessor because homogeneity of dimHX>dimBY1 is not assumed.
Technically, the proof uses a martingale argument due to Wang in place of Ruan–Xiao's Proposition 3.4: the conditional measures of dimHX>dimBY2 on level-dimHX>dimBY3 cells form a bounded martingale, converging almost everywhere by Doob's theorem. Combined with the fact that almost every coding walk contains every cyclic walk (an exponential decay estimate using strong connectivity), one extracts a point dimHX>dimBY4 along which the preimage decomposition stabilizes, yielding boundedness of the minimal-domain function dimHX>dimBY5 on dimHX>dimBY6; a compactness/Arzelà–Ascoli limit argument then produces the surjective minimizer, and an equality-chasing argument over cells propagates the constant-density property across all vertices.
From surjections to commensurability
The forward implication splits into two cases depending on which system has commensurable ratios. Both exploit the same mechanism: the quantity dimHX>dimBY7 along a nested sequence of cells takes values in a finite set up to scaling, because the integer counting vectors dimHX>dimBY8 recording which level-dimHX>dimBY9 cells of dimHX=dimHY0 map into dimHX=dimHY1 are bounded independently of dimHX=dimHY2. Consequently, changes of dimHX=dimHY3 are either zero or uniformly bounded below (a discreteness phenomenon analogous to one used by Falconer–Marsh for bi-Lipschitz maps).
When dimHX=dimHY4 is commensurable, choosing dimHX=dimHY5 whose coding walk contains every cyclic walk dimHX=dimHY6 of dimHX=dimHY7 forces dimHX=dimHY8 to be eventually constant along long stretches, and comparing values at two repeats of dimHX=dimHY9 yields X0 — a rational power of X1. When instead X2 is commensurable, the argument runs along powers of a cycle X3 in the graph of X4: the covering lemma guarantees large image cells inside X5, minimality of the measure-linear surjection forces these preimages to lie inside X6 entirely, and finiteness of cell types up to similarity lets one compare two levels, again producing X7.
From commensurability to surjections
The reverse direction adapts Xi–Xiong's mass decomposition technique to the graph-directed setting. After reducing both systems to homogeneous primitive form and powering them to a common ratio X8 (using X9 from commensurability), the key lemma states: given a cell Y0 at some level and target measures Y1 drawn from a finite set within the ring Y2, for sufficiently deep refinement Y3 admits a partition into subcells realizing those exact measures.
The proof reduces to a purely arithmetic statement (essentially reused from Xi–Xiong) about splitting sequences of numbers from finite sets into two subsequences with equal sums, together with the identity Y4 for large Y5 when Y6 and Y7 ranges over nonnegative integer vectors, plus Perron–Frobenius convergence Y8 ensuring positivity of the partition vectors.
The surjection itself is built as a limit of cell-to-cell correspondences: starting from Y9 disjoint scaled copies of logr1/logr20 (whose existence at deep levels follows from strong connectivity plus a counting argument showing the attractor is not a single cycle of points), one recursively partitions preimage cells according to the measures of image cells using the mass decomposition lemma, maintaining the invariant logr1/logr21. Nested closed cells with vanishing diameters define a pointwise limit map, which is Lipschitz (separated points at depth logr1/logr22 sit at distance logr1/logr23 while images lie in cells of diameter logr1/logr24), has dense image by construction, hence is surjective by compactness. Extending constantly on the complement of the clopen set logr1/logr25 preserves Lipschitz continuity.
This construction also recovers, as a special case, a simpler proof of the earlier counterexample of Ruan–Xiao: two dust-like subsets of logr1/logr26, both of dimension logr1/logr27 with golden-ratio-related measure roots generating different rings logr1/logr28-incompatibly (hence not bi-Lipschitz equivalent by the Xi–Xiong criterion), yet admitting Lipschitz surjections in both directions since all ratios are powers of logr1/logr29.
Limitations and open questions
The characterization requires at least one of the two systems to have commensurable ratios (cycle ratios, in the graph-directed case), and dust-likeness throughout; the paper does not address self-similar sets with overlaps or with incommensurable ratios on both sides. The author explicitly states that he does not know whether the surjection-reversal corollary remains valid when neither X0 nor X1 has commensurable ratios. Additionally, the cited generalization of the bi-Lipschitz characterization to totally disconnected sets satisfying the open set condition involves substantially more complicated conditions, and no analogue of the present surjection characterization is claimed beyond the dust-like, strongly connected regime — where Mauldin–Williams theory ensures well-behaved Hausdorff measures; without strong connectivity the dimensional analysis becomes more delicate.
Conclusion
The paper replaces the homogeneity assumption in the Balka–Keleti/Ruan–Xiao equivalence between Lipschitz surjectivity and bi-Lipschitz equivalence with the weaker, symmetric condition of commensurable ratios, yielding a clean characterization of Lipschitz surjectivity between equal-dimensional dust-like self-similar sets and their graph-directed generalizations. The proof synthesizes a graph-theoretic reduction to primitive homogeneous systems, a martingale-based construction of measure-linear surjections, and a graph-directed adaptation of Xi–Xiong's mass decomposition method, and it supplies a transparent explanation of the known failure of the equivalence without commensurability.
Paper to Video (Beta)
No one has generated a video about this paper yet.
Whiteboard
No one has generated a whiteboard explanation for this paper yet.
Paper Prompts
Sign up for free to create and run prompts on this paper.
Top Community Prompts
Open Problems
Continue Learning
- What conditions are needed for the existence of a Lipschitz surjection between self-similar sets of unequal Hausdorff dimensions?
- Can the commensurable condition can be relaxed or extended to other types of relationships between similarity ratios?
- How does the graph-directed iterated function system (GD-IFS) framework generalize the results obtained for ordinary iterated function systems (IFSs)?
- In what scenarios would the reversibility of Lipschitz surjections hold without the commensurability assumption?
- Find recent papers about the equivalence of Lipschitz surjectivity and bi-Lipschitz equivalence for totally disconnected sets.