Geometric Anatomy of Emergence

• Geometric Anatomy of Emergence is a framework that defines emergent behaviors using explicit geometric, spectral, and topological observables rather than abstract novelty.
• It integrates models from complex networks, morphogenesis, quantum systems, and learning dynamics to demonstrate thresholds, persistence, and scaling laws in emergent phenomena.
• Practical insights include quantifying emergent features using measures like combinatorial curvature, modular Fubini–Study metrics, and singular-value knee analysis to predict macroscopic structural organization.

“Geometric anatomy of emergence” denotes a family of research programs in which emergent organization is read off from explicit geometric, spectral, topological, or variational structures rather than from informal appeals to novelty alone. In the cited literature, emergence is anatomized through “curvature heterogeneity, community modularity, K-core hierarchies, small-world metric scaling and finite diffusion dimension,” through the “Quantum Modular Geometric Tensor,” through a “Gromov–Hausdorff-type potential,” through the “catastrophe set,” through “scale-invariant collapse,” and through the “singular-value knee” (Wu et al., 2014, Huang et al., 2021, Morozova et al., 2014, Gastaldo et al., 2021, Billa, 17 Feb 2026, Faruque et al., 2024). A recurrent theme is that the relevant geometry may be produced by non-equilibrium growth rules, by operator spectra, by broken symmetry, by local cellular events, or by training dynamics, and may or may not presuppose an ambient or latent space.

1. Recurring formal motifs

Across the literature, the geometric object used to describe emergence varies sharply by domain, but the methodological pattern is stable: specify microscopic update rules or a dynamical law, define a geometric observable, and show that the observable organizes macroscopic behavior. This suggests that “geometric anatomy” is best understood as a comparative framework rather than a single formalism (Wu et al., 2014, Morozova et al., 2014, Reitz et al., 2023, Billa, 17 Feb 2026).

Domain	Geometric object	Emergence signature
Complex networks	simplicial complex of triangles; combinatorial curvature; spectral dimension	small-world behavior, modularity, K-core hierarchy, finite $d_S$
Morphogenesis	cell–surface matrix; star-convex cells; catastrophe set	event tree, moving boundary, regenerative growth, gastrulation models
Quantum and holographic systems	QMGT; basis-independent spectra; center of a representation	kinematic-space metric, scale-dependent $D_{\rm eff}$, emergent manifold
Learning and dynamics	RankMe; metric order; singular-value knee	capability emergence, topological emergence, cross-embodiment detection

Two broad orientations recur. In one, geometry is not postulated in advance: the “growing geometrical network” develops “without ever embedding the network in a pre-existing metric space,” the basis-independent path integral “never had to choose” a position basis, and spontaneous symmetry breaking yields an emergent manifold as the spectrum of a commutative center (Wu et al., 2014, Reitz et al., 2023, Ojima, 2011). In the other, an explicit ambient or latent geometry is assumed from the outset, as in the “two-dimensional hyperbolic disk of curvature $K=-1$” of geometric preferential attachment or the “star-convex” cellular geometry of morphogenesis (Zuev et al., 2015, Morozova et al., 2014). This distinction is substantive: it separates emergent geometry from geometry-guided emergence.

2. Networks, graphs, and combinatorial geometry

In “Emergent Complex Network Geometry,” Wu et al. define a two-parameter model with $m\in\{2,3,\dots,\infty\}$, the maximum number of triangles incident on a link, and $p\in[0,1]$, the probability of a local “closure” step. The process starts with a single triangle and alternates between triangle-addition and local closure, producing “a growing simplicial complex of triangles; ignoring higher simplices yields the ‘geometrical network’ of nodes + links.” The key geometric observable is the combinatorial curvature

$R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$

The model has sharp limiting regimes: $m=\infty,p=0$ gives the “pure triangle-tree, Barabási–Albert limit” with

$$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$

and asymptotically vanishing clustering; $m=2$, for any $p$, gives “planar random geometry” with

$D_{\rm eff}$0

Euler characteristic $D_{\rm eff}$1, and an exponential negative curvature tail. For intermediate $D_{\rm eff}$2, the network develops “a broad degree distribution,” “hierarchical clustering,” “a nontrivial community structure of high modularity $D_{\rm eff}$3,” “a rich $D_{\rm eff}$4-core hierarchy,” logarithmic distance growth $D_{\rm eff}$5, and finite spectral dimension, with numerical estimates $D_{\rm eff}$6 for $D_{\rm eff}$7 and $D_{\rm eff}$8 (Wu et al., 2014).

A complementary construction appears in “Emergence of Soft Communities from Geometric Preferential Attachment,” where the geometry is latent rather than emergent. Nodes live in a hyperbolic disk with radial coordinate $D_{\rm eff}$9 for popularity and angular coordinate $K=-1$0 for similarity. A new node chooses $K=-1$1 by “geometric preferential attachment” to regions of high node density and then links to the $K=-1$2 closest nodes in hyperbolic distance. The attachment kernel reduces exactly to the one of the PS model and to pure preferential attachment with exponent $K=-1$3, yielding

$K=-1$4

The model combines “scale-free degree distributions, strong clustering, and community structure,” with communities defined as angular runs between gaps exceeding $K=-1$5. Validation against the December 2009 Internet AS topology shows matching degree CCDFs, clustering spectra, angular nonuniformity, and mean community separation $K=-1$6 (Zuev et al., 2015).

Taken together, these models isolate two different geometric routes to network emergence. One route generates curvature, modularity, and finite diffusion dimension from non-equilibrium local moves alone; the other uses hidden hyperbolic geometry to unify popularity, similarity, clustering, and “soft communities.” This suggests that network emergence can be anatomized either as spontaneous combinatorial geometry or as the dynamical occupation of a latent geometric substrate.

3. Morphogenesis, cellular form, and layered growth

In the morphogenetic literature, emergence is formulated through local cell state, event selection, and global boundary evolution. Morozova and Shubin associate to each cell $K=-1$7 an $K=-1$8 “cell–surface code matrix”

$K=-1$9

with $m\in\{2,3,\dots,\infty\}$0 monosaccharide types in the illustrative model. A full cell state is $m\in\{2,3,\dots,\infty\}$1, and a “cell event” $m\in\{2,3,\dots,\infty\}$2 acts on this state by division, growth or shrinkage, shift, or differentiation. Development becomes a rooted, time-ranked event tree $m\in\{2,3,\dots,\infty\}$3, while the “morphogenetic field” is a map $m\in\{2,3,\dots,\infty\}$4 selecting the next admissible event from the current cell state and context. The local code is updated by operators $m\in\{2,3,\dots,\infty\}$5, and the aggregate of local events induces a continuous moving boundary $m\in\{2,3,\dots,\infty\}$6 with normal velocity $m\in\{2,3,\dots,\infty\}$7 satisfying

$m\in\{2,3,\dots,\infty\}$8

The resulting program connects cell-surface codes, event selection, and a PDE-like evolution of embryo shape (Morozova et al., 2012).

Morozova and Penner recast morphogenesis in a metric-measure and variational language. A cell is “star-convex with respect to a distinguished ‘MTOC point’ $m\in\{2,3,\dots,\infty\}$9,” so its shape is encoded by the radial function

$p\in[0,1]$0

The organism at time $p\in[0,1]$1 is the union $p\in[0,1]$2, equipped with the induced Euclidean metric and an atomic measure $p\in[0,1]$3. Over the configuration space $p\in[0,1]$4 of MTOC centers they define a Hilbert bundle with fiber $p\in[0,1]$5, where $p\in[0,1]$6 stores the radial shape function and epigenetic surface fields. Morphogenesis is postulated to follow from stationarity of an action

$p\in[0,1]$7

with $p\in[0,1]$8 taken as a “Gromov–Hausdorff-type distance.” In this framework, development and regenerative growth are governed by Euler–Lagrange equations on a bundle over configuration space (Morozova et al., 2014).

A different but related geometry of biological emergence is Thom’s catastrophe-theoretic program. Here the state-space is a finite-dimensional smooth manifold $p\in[0,1]$9, local growth events are “the universal unfolding of a degenerate critical point in a potential $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$0,” and the elementary catastrophes include the fold, cusp, swallowtail, hyperbolic umbilic, elliptic umbilic, parabolic umbilic, and butterfly. Local dynamics can be gradient flows,

$R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$1

or Hamiltonian flows on $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$2. The “catastrophe set” $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$3 is a stratified algebraic subset of nongeneric maps, and its codimension is the number of unfolding parameters. Gastrulation examples include “blastopore invagination” via the swallowtail and “limb-bud emergence” via the elliptic umbilic (Gastaldo et al., 2021).

Layered cellular automata provide a discrete analog of morphogenetic emergence. García-Morales defines $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$4-decomposable automata by expressing the alphabet size $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$5 as a product $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$6 and introducing layer variables

$R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$7

The resulting tree diagrams distinguish graded rules, in which layers evolve independently, from non-graded rules with directed couplings. Using the Boolean majority rule as a building block, the formalism produces “nested domains,” “stable coexistence of domains,” “Wolfram CA inside a domain,” and two-dimensional “Matryoshka” structures. The diagram serves as a “blueprint” specifying how many layers, which couplings, and which neighborhood scales generate the observed forms (García-Morales, 2016).

4. Quantum-information and spacetime emergence

In holography and quantum information, the “Emergence of Kinematic Space from Quantum Modular Geometric Tensor” generalizes the ordinary Quantum Geometric Tensor by replacing the Hamiltonian with the modular Hamiltonian. For an eigenstate $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$8 of $R_i=1-\tfrac{k_i}{2}+\tfrac{t_i}{3}.$9, the QMGT is

$m=\infty,p=0$0

with decomposition

$m=\infty,p=0$1

The symmetric part $m=\infty,p=0$2 is a modular Fubini–Study metric and the antisymmetric part $m=\infty,p=0$3 is a modular Berry curvature. For a spherical entangling surface in a CFT, the nonzero components are mixed $m=\infty,p=0$4–$m=\infty,p=0$5 derivatives, and the induced line element reproduces the standard kinematic-space metric. In AdS$m=\infty,p=0$6/CFT$m=\infty,p=0$7, connected Wilson-line correlators coincide at leading order with two-point functions of modular Hamiltonians and compute mutual-information differences “even for two disjoint intervals at finite separation, and including the one-loop quantum correction in bulk gravity” (Huang et al., 2021).

A more radical operator-based program appears in “Emergence of Spacetime from Fluctuations.” Here the Euclidean path integral is formulated entirely in the eigenbasis of the Laplacian with a sharp spectral cutoff $m=\infty,p=0$8, so the partition function is basis independent and meaningful even in “pre-geometric” regimes that “do not admit a mathematical representation of the physical degrees of freedom in terms of fields that live on a spacetime.” The Hawking–Gilkey heat-kernel expansion

$m=\infty,p=0$9

recovers the Einstein–Hilbert action plus higher-curvature corrections. When a spacetime picture is available, the effective dimension follows from the eigenvalue density or, with species-dependent masses, from

$$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$0

Mass thresholds produce jumps or crossovers in $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$1, and the effective volume is read from the spectral gap. The theory therefore gives an explicit mechanism by which “the effective spacetime dimension can depend on the energy scale” (Reitz et al., 2023).

Ojima’s symmetry-breaking account gives another route from algebra to geometry. Starting from a $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$2-algebra $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$3, a compact Lie symmetry group $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$4, and a symmetry-breaking state with residual subgroup $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$5, the GNS representation acquires a nontrivial center. In the augmented algebra $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$6, one has the “condensation condition”

$$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$7

The emergent manifold is then

$$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$8

and the “logical extension” embeds a constant observable $$P(k)=\frac{12}{k(k+1)(k+2)}\sim k^{-3},\quad \gamma=3,$$9 into a function over the spectrum. In this picture, order parameters become coordinates, and space(-time) arises from condensation and symmetry breakdown rather than from a background manifold (Ojima, 2011).

Covariant loop quantum gravity provides a discrete quantum-gravitational version of the same problem. For the Euclidean EPRL/FK vertex amplitude on the 4-simplex boundary graph, with cylindrical symmetry and $m=2$0, the amplitude is evaluated on Livine–Speziale coherent intertwiners and compared to classical Regge data reconstructed from the areas

$m=2$1

The numerical study shows that “already at spins $m=2$2 the spin-network amplitude reorganizes into a sharp peak on classical Regge geometry,” with $m=2$3, $m=2$4, and $m=2$5 peaked near their classical values even for $m=2$6–10. At the same time, “degenerate configurations” appear as secondary peaks at small $m=2$7 and $m=2$8 (Bayle et al., 2016).

5. Quantifying emergence in learning and dynamical data

In neural-network training, Billa introduces five geometric measures—RankMe, Fisher effective rank, the local learning coefficient (LLC), Hessian top-$m=2$9, and gradient-covariance rank—computed on task-conditioned diagnostic examples. RankMe is defined from the singular values of a hidden-layer activation matrix $p$0 by

$p$1

Across “five model scales (405K–85M parameters), 120+ emergence events in eight algorithmic tasks,” training begins with a “universal representation collapse to task-specific floors that are scale-invariant across a 210X parameter range”; for modular arithmetic, the minimum is

$p$2

Collapse propagates “top-down through layers” with “zero exceptions (32/32 consistency for all 4-, 6-, 8-layer models).” Temporal precedence is stratified as “Representation geometry $p$3 Gradient geometry $p$4 Loss-landscape geometry”: RankMe leads in 75–100% of hard-task cases, Fisher-rank in $p$5, LLC in $p$6, Hessian top-$p$7 in only 17%, and gradient-covariance rank in $p$8. Yet prediction is limited: within-easy-task concordance is 27%, the swap-test is only 26% correct, and on Pythia “task-specific precursors fail” even though global collapse patterns replicate (Billa, 17 Feb 2026).

In topological dynamics, Carvalho, Rodrigues, and Varandas define topological emergence from the covering numbers of the ergodic-measure set $p$9. With $D_{\rm eff}$00 a weak$D_{\rm eff}$01-compatible metric on probability measures,

$D_{\rm eff}$02

and

$D_{\rm eff}$03

The main dichotomy is dimensional: for orientation-preserving homeomorphisms of $D_{\rm eff}$04 or $D_{\rm eff}$05, “Every $D_{\rm eff}$06 satisfies $D_{\rm eff}$07,” whereas for a residual set of conservative homeomorphisms on a compact manifold $D_{\rm eff}$08 of dimension $D_{\rm eff}$09, one has

$D_{\rm eff}$10

Metric emergence, defined relative to a reference measure $D_{\rm eff}$11, satisfies an intermediate-value property: for any $D_{\rm eff}$12, there exists an invariant measure with $D_{\rm eff}$13 (Carvalho et al., 2022).

For noisy trajectory data, “Singular knee identification to support emergence recognition” treats emergence as low-dimensional structure embedded in Gaussian noise. Given a demeaned and normalized data matrix $D_{\rm eff}$14, one computes

$D_{\rm eff}$15

and locates the knee in the singular-value scree plot by a triangle method. The local knee angle

$D_{\rm eff}$16

is the preferred scalar indicator. Random-matrix noise bounds come from Marčenko–Pastur: $D_{\rm eff}$17 Across Gaussian noise, random walks, Cucker–Smale, Vicsek, 1D cellular automata, and bird-flocking video, the “angle subtended at the singular value knee emerges with the most potential for supporting cross-embodiment emergence detection.” Reported values range from $D_{\rm eff}$18 for pure noise to $D_{\rm eff}$19 for a deterministic spiral, with a tracked bird flock at $D_{\rm eff}$20 and $D_{\rm eff}$21 (Faruque et al., 2024).

6. Thresholds, persistence, and boundary conditions

The most explicit threshold theory in the corpus appears in “Morphogenesis Across $D_{\rm eff}$22: Overlays, Emergence Thresholds, and Weak Self-Similarity in the Partition Graph.” For the partition graph $D_{\rm eff}$23, whose vertices are partitions of $D_{\rm eff}$24 and whose edges are elementary unit-transfers, Ferrers-translation maps

$D_{\rm eff}$25

are “induced graph embeddings.” As a result, every finite rooted induced motif persists to all higher levels, and every “overlay-monotone finitely witnessed property has a stable emergence threshold.” The extremal local invariants

$D_{\rm eff}$26

are nondecreasing, and the canonical motifs $D_{\rm eff}$27, $D_{\rm eff}$28, $D_{\rm eff}$29, $D_{\rm eff}$30, and $D_{\rm eff}$31 have exact or stable thresholds $D_{\rm eff}$32. The family $D_{\rm eff}$33 is therefore “theorem-level weakly self-similar” (Lyudogovskiy, 26 Mar 2026).

Several boundary conditions recur across domains. A common misconception is that geometric emergence must always rely on a pre-existing ambient geometry; in fact, some frameworks explicitly generate structure “without ever embedding the network in a pre-existing metric space” or “without ever referring to a position basis,” while others are built precisely on a hyperbolic disk or on star-convex cells (Wu et al., 2014, Reitz et al., 2023, Zuev et al., 2015, Morozova et al., 2014). Another misconception is that a geometric signal automatically yields fine-grained prediction. In controlled algorithmic tasks, RankMe gives the clearest precursor relation, but “geometry encodes a coarse difficulty ranking (easy vs. hard), but not the next task to emerge in fine order,” and the precursor relation disappears on Pythia when “task-training alignment” is absent (Billa, 17 Feb 2026). A third misconception is that higher-dimensional geometric richness is universal across dynamical settings: the topological-emergence results show exactly the opposite dependence on ambient dimension, with zero emergence in dimension one and maximal generic emergence in dimensions greater than one (Carvalho et al., 2022). Even when a classical geometric peak is present, as in the Euclidean LQG vertex, degenerate configurations may remain visible rather than disappearing outright (Bayle et al., 2016).

These results collectively suggest a disciplined use of the term “emergence.” In the strongest cases, one has explicit theorems on persistence, thresholds, or scaling laws; in others, one has variational principles, spectral diagnostics, or empirical precursor relations with stated failure modes. The unifying contribution of geometric anatomy is not a single ontology of emergence, but a precise vocabulary for how microscopic rules reorganize into curvature, metric, topology, spectral dimension, catastrophe structure, modular geometry, or low-rank collective modes.