Bubble Tree: Hierarchical Structures in Geometry

• Bubble tree is a family of tree-structured constructions that organize bubble-like limits in geometry, algebraic compactification, and computational pangenomics.
• The approach systematically captures energy concentration, scale separation, and degeneration phenomena via iterated blow-ups and hierarchical decompositions.
• Applications span from proving compactness in Ricci shrinkers and harmonic maps to modeling soap-bubble clusters and enhancing genomic variation algorithms.

“Bubble tree” denotes a family of tree-structured constructions in which bubbles, bubble-like limits, or bubble-like subgraphs are organized by attachment, scale, or nesting. In geometric analysis, it is the rooted tree produced by iterated blow-up at concentration points, with a base object at the root and bubbles attached at finer scales. In algebraic geometry, it becomes a compactification by vector bundles on trees of surfaces. In computational pangenomics, it denotes a representation of nesting relations among snarls and ultrabubbles. In foam geometry, the term also appears as an analogy for soap-bubble clusters whose central multiple-torus bubble acts as a trunk and attached bubbles as branches (Buzano et al., 2022, Markushevich et al., 2011, Sena et al., 9 Apr 2026, Delbary, 4 Mar 2025).

1. Bubble-tree convergence as a compactness mechanism

For gradient Ricci shrinkers, a bubble tree is a scale-by-scale description of curvature concentration. A gradient Ricci shrinker is a complete Riemannian manifold $(M^n,g,f)$ satisfying

$\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$

Under a lower Perelman entropy bound and uniform local curvature energy bounds, sequences of shrinkers admit subsequential limits in the pointed orbifold Cheeger–Gromov sense, and each orbifold singular point gives rise to a finite number of point-scale sequences whose rescalings converge to Ricci-flat ALE bubbles. The resulting structure is a tree whose vertices are ALE bubbles and whose edges represent neck regions connecting bubbles and the root (Buzano et al., 2022).

A central feature of this theory is the exclusion of hidden curvature in the intermediate scales. The neck theorem shows that thin annuli with small total energy are quantitatively close to annular regions in a flat cone $\mathbb{R}^n/\Gamma$, and the energy estimate in necks implies that no energy concentrates in neck regions. Consequently, the local energy identity takes the form

$$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$

so all curvature energy is accounted for by the orbifold limit and the bubbles. Direct consequences include an Euler characteristic identity and a local diffeomorphism finiteness theorem (Buzano et al., 2022).

The same term is used in the broader compactness theory of harmonic-type variational problems. In that setting, a bubble tree records the successive extraction of nontrivial limiting maps at concentration points, together with the hierarchy of scales at which those limits appear. The tree formalism is therefore not merely topological; it is a bookkeeping device for energy quantization, neck analysis, and the topology of degeneration.

2. Harmonic maps, conformal maps, and no-neck limits

For harmonic maps from a compact Riemann surface into a compact locally CAT(1) space, bubble tree convergence holds under a uniform energy bound. The limiting object is a bubble tree domain carrying bubble tree maps, and the theory establishes both energy quantization and the no-neck property. The proof is geometric rather than PDE-based: it relies on local convexity of the target, together with an $\epsilon$-regularity theorem, an energy gap theorem, a removable singularity theorem, and an isoperimetric inequality for conformal harmonic maps with small image (Breiner et al., 2018).

For associative Smith maps, bubble trees arise in a conformally invariant first-order theory on $3$-manifolds. An associative Smith map $u$ satisfies the Smith equation

$J \circ \wedge^2 du = |du|\, du \circ *,$

and bounded $3$-energy sequences can be conformally rescaled to yield bubble trees of such maps. The analytic package includes quantitative $C^{1,\beta}$-regularity under small $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$0-energy, a removable singularity result, and an energy gap on $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$1. When the $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$2-structure is closed, both the $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$3-energy and the homotopy are preserved in the bubble tree limit, and the zero-neck-length statement identifies the bubble tree limit as a connected union of the base map and the bubbles (Cheng et al., 2019).

For $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$4 branched conformal immersions of closed Riemann surfaces with bounded area and Willmore energy, the bubble tree construction persists even when the conformal structures degenerate. Local blow-up analysis at concentration points produces a finite rooted tree of bubbles, and globally the limit is a stratified surface consisting of the base and attached spheres. The compactness theorem yields convergence in Hausdorff distance to a continuous $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$5 limit map on the stratified surface, with no-neck and no-energy-loss across collars and bubble junctions (Chen et al., 2011).

A further refinement appears for the $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$6-functional on closed non-spherical surfaces. There, the “simple bubble tree” is the first non-trivial bubbling configuration: the base map is constant and the nontrivial energy concentrates into a single bubble. The energy gap theorem states that for every critical point $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$7 on a surface of positive genus,

$\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$8

so the bubble energy level $\operatorname{Ric}_g + \nabla^2 f = \frac{1}{2} g.$9 is isolated from genuine critical points on non-spherical domains. Near the manifold of adapted bubbles, sequences of almost critical points satisfy Łojasiewicz inequalities that quantify the bubble scale, the $\mathbb{R}^n/\Gamma$0-distance to the bubble manifold, and the excess of energy above $\mathbb{R}^n/\Gamma$1 (Malchiodi et al., 2020).

3. Algebraic bubble trees and compactification of moduli

In algebraic geometry, bubble trees arise in the compactification of moduli spaces of rank-$\mathbb{R}^n/\Gamma$2 vector bundles on surfaces. The guiding idea is to replace boundary points represented by torsion-free sheaves with vector bundles on trees of surfaces, providing an algebraic counterpart of the bubbling of vector bundles and connections in differential geometry. A weighted tree $\mathbb{R}^n/\Gamma$3 assigns a charge $\mathbb{R}^n/\Gamma$4 to each vertex, summing to a fixed total charge $\mathbb{R}^n/\Gamma$5, and this combinatorial datum governs the geometry of the limiting surface (Markushevich et al., 2011).

A tree of surfaces $\mathbb{R}^n/\Gamma$6 is built by associating to the root a blowup of the original smooth projective surface $\mathbb{R}^n/\Gamma$7, to internal non-root vertices blowups of $\mathbb{R}^n/\Gamma$8 at prescribed points away from a distinguished line, and to terminal vertices copies of $\mathbb{R}^n/\Gamma$9. Adjacent components are glued along lines corresponding to the edges of the tree, and there is a contraction morphism $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$0 that collapses all components except the root. A tree-like bundle, or $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$1-bundle, is then a compatible collection of rank-$$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$2 bundles on the components, with prescribed Chern classes and admissibility conditions on non-root components (Markushevich et al., 2011).

The compactification is constructed functorially through families of $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$3-surfaces and $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$4-bundles, with parameter spaces built using the Fulton–MacPherson compactification of configuration spaces, Hilbert schemes, Grassmannian embeddings, and quotient constructions under group actions. The main result is the existence of a separated algebraic space $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$5 of finite type corepresenting the functor of families of limit $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$6-bundles and compactifying the moduli space of stable rank-$$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$7 bundles with fixed Chern classes (Markushevich et al., 2011).

The example $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$8, $$\lim_{i\to\infty} \int_{B_{g_i}(p_i,r)} |\mathrm{Rm}_{g_i}|^{n/2} = \int_{B_{g_\infty}(p_\infty,r)} |\mathrm{Rm}_{g_\infty}|^{n/2} + \sum_{q}\sum_k \int_{V^k} |\mathrm{Rm}_{h^k}|^{n/2},$$9, $\epsilon$0 is especially explicit. The classical moduli space $\epsilon$1 is isomorphic to $\epsilon$2, but the bubble-tree compactification replaces singular boundary sheaves by bundles on trees of surfaces of several combinatorial types. In this case, $\epsilon$3 is isomorphic to the blowup of $\epsilon$4 along the Veronese surface, and the strata correspond to distinct tree types recording whether bubbling occurs at one point, two points, or in a more iterated pattern (Markushevich et al., 2011).

4. Soap-bubble clusters and tree-like foam topologies

In the geometry of soap bubbles, “bubble tree” is used more topologically than analytically. Numerical constructions of stable soap bubble clusters with torus and higher-genus bubbles show how a central multiple-torus bubble can act as a trunk with attached bubbles as branches. A genus-$\epsilon$5 example is a $\epsilon$6-bubble cluster with three exterior bubbles, one central torus-like bubble, and two interior bubbles. Higher genus examples are obtained by assembling copies of the base cluster around Platonic solids whose vertices have valence $\epsilon$7, namely the tetrahedron, cube, and dodecahedron, yielding respectively a triple torus, a fivefold torus, and an elevenfold torus (Delbary, 4 Mar 2025).

These constructions are governed by Plateau’s laws and verified numerically with Surface Evolver. The genus is computed from the Euler characteristic,

$\epsilon$8

and the stability diagnostics include Hessian eigenvalue analysis and perturbation tests. In this literature, the multiple torus bubble is described as analogous to what Sullivan and Morgan dubbed “bubble trees,” with the multi-torus as the trunk and repeated attached regions as branches (Delbary, 4 Mar 2025).

A later extension generalizes the construction from Platonic solids to prisms and Archimedean solids. The method starts from a convex polyhedral scaffold, introduces three concentric homothetic copies $\epsilon$9, $3$0, and $3$1, inserts inner double-bubble prisms or related linkers, and minimizes surface area under volume constraints. For a general convex polyhedron with $3$2 faces, the resulting cluster has $3$3 bubbles and one bubble of genus $3$4. For semiregular $3$5-prisms, the formulas become $3$6 bubbles and genus $3$7. The great rhombicosidodecahedron example produces a cluster with $3$8 bubbles and a torus bubble of genus $3$9 (Fabrice, 31 Jan 2026).

A related planar theory studies the space of planar soap bubble clusters with fixed topology. There, clusters with tree-like combinatorial type offer a model for the space’s structure, and the space of equilibrium clusters with positive second variation is a smooth $u$0-dimensional manifold, locally parametrized by the areas. Earlier work of Moukarzel showed that every equilibrium cluster can be realized non-canonically as a generalized Voronoi partition, linking foam geometry to computational geometry and moduli questions (Morgan, 2016).

5. Bubble Tree in computational pangenomics

In computational biology, “Bubble Tree” refers to a decomposition for nested variation structures in genome graphs. Earlier work constructed the Bubble Tree via a cactus decomposition, representing nesting relations among snarls and ultrabubbles. More recent work replaces that viewpoint with an SPQR-tree framework and states explicitly that the SPQR-tree generalizes and subsumes the role of the Bubble Tree because it is a finer and more informative decomposition exposing all $u$1-separators of the underlying undirected graph (Sena et al., 9 Apr 2026).

This reinterpretation is algorithmic rather than geometric. Nearly all bubble-like subgraphs—superbubbles in directed graphs, and snarls and ultrabubbles in bidirected graphs—are linked to $u$2-separators. The SPQR-tree decomposition of a biconnected graph encodes all such separation pairs through S-, P-, Q-, and R-nodes, together with skeleton graphs and expansion graphs. Dynamic-programming-style traversals of the SPQR-tree then test whether the expansion associated to a separation pair satisfies the required acyclicity, minimality, reachability, and tiplessness conditions (Sena et al., 9 Apr 2026).

The main complexity results are sharp. There are first linear-time algorithms for identifying all snarls and all ultrabubbles, and the same unified framework yields a completely different linear-time algorithm for finding all superbubbles. A crucial ingredient is the linear-time computation of all feedback arcs in tipless bidirected graphs. The representation of all snarls is also linear in the size of the graph, even though the total number of snarls can be quadratic (Sena et al., 9 Apr 2026).

An implementation of the unified SPQR-tree framework in C++, BubbleFinder, was evaluated on large pangenomic datasets. Reported performance includes being up to two times faster than vg for snarls while identifying all snarls, and up to $u$3 times faster than BubbleGun for superbubbles. In this setting, the Bubble Tree is not a compactification or a blow-up limit; it is a hierarchical decomposition of genomic variation sites (Sena et al., 26 Nov 2025).

6. Scope of the term and neighboring but distinct usages

Because “bubble” and “tree” are common scientific words, “bubble tree” is easily conflated with unrelated topics. Cavitation in a synthetic tree concerns spontaneous or triggered cavitation in water-filled microcavities of a hydrogel, with a stable bubble created in a microsecond timescale and a later diffusion-driven expansion that fills the cavity; the subject is the dynamics of metastable water under tension, not a bubble-tree construction (Vincent et al., 2011). “Dynamics of Investor Spanning Trees Around Dot-Com Bubble” studies minimum and maximum spanning trees built from investor-specific net-volume correlations around the Nokia dot-com episode; here “bubble” refers to a financial bubble and “tree” to a graph-theoretic spanning tree (Ranganathan et al., 2017). “Bubble $u$4Ar and Its New Breathing Modes” concerns bubble nuclei with depleted central density and three density distribution modes—micro-bubble, bubble, and cluster resonance—rather than a bubble tree (Ren et al., 2024).

At the same time, the term can be extended informally to hierarchical physical systems of interacting bubbles. A theoretical study of a central bubble surrounded by a circular bubble cluster states that the principles observed there extend to more complex, hierarchical “bubble tree” systems with implications for how energy is distributed, absorbed, or focused (Zhang et al., 2023). This suggests a common structural motif across the literature: hierarchy, localized concentration, and scale-separated attachment. What varies from field to field is the object being organized—maps, ALE spaces, surfaces, soap films, or bubble-like subgraphs—rather than the underlying tree logic.