Every threshold-crossing that cannot be undone — a cell beginning autophagy, a star igniting helium, an elastic rod buckling — follows the same four-step geometric sequence: compression toward a critical surface, curvature intensification, a Whitney fold that breaks injectivity, and stabilisation on a new branch.
This paper formalises that sequence as the operator chain \(G = U \circ F \circ K \circ C\) acting on trajectories in a Riemannian manifold, derives a free-discontinuity variational principle and symplectic Hamiltonian structure, and proves five structural theorems (existence, non-commutativity, irreducibility, Whitney-fold classification, symplectic preservation), each by seven independent arguments.
The three canonical constants —
Gronwall stability radius[Ch 10] \(\varepsilon_0 = 1/3\),
transverse Lyapunov exponent \(\mu_{\max} = -2\),
and embodiment threshold \(\tau = 2\) —
are machine-checked in Lean 4 (AXLE) with zero axioms beyond Mathlib4 and zero sorry.
Theorems A–D are formalised over an arbitrary MetricSpace carrier;
toy-model constants P1–P6 are separately machine-checked as arithmetic facts.
A structural analogy between the five-operator chain and Perelman's Ricci flow with surgery is noted as Conjecture 15.1 (not proved here).
A generative transition is what happens when a system crosses a threshold it cannot uncross. The cell that begins to digest itself under nutrient stress. The star that ignites helium when its hydrogen is exhausted. The elastic rod that buckles under axial load. The 3-manifold that develops a geometric singularity under Ricci flow. In each case, a compression drives the system toward a critical point, a curvature instability makes the approach inevitable, a fold commits the system irreversibly to a new branch, and an unfolding establishes the new stable state.
This paper formalises that sequence as four operators acting on trajectories in a Riemannian manifold. The fifth operator E (Entropy / Generative Time Circuit), added in the second edition, accumulates the irreversible cost:
The central mathematical claim is that this operator sequence is not an analogy: it is a precise structure admitting constructive definitions, analytical invariants, a variational principle, and a Hamiltonian formulation. The framework is placed within established mathematical traditions: comparison geometry, singularity theory, geometric flows, variational mechanics, and impulsive Hamiltonian systems.
All new second-edition material is explicitly marked as argued or conjectured. Nothing new is claimed as proved beyond what Lean 4 verifies.
\((X,g)\) Riemannian manifold; \(\gamma:[0,T]\to X\) trajectory. \(\kappa(s)\) geodesic curvature; \(\kappa^*(x)\) fold threshold. \(\rho = r-1\) transverse displacement from limit cycle \(\Gamma\). \(\varepsilon_0 = 1/3\) Gronwall stability radius; \(\mu_{\max} = -2\) transverse Lyapunov exponent; \(\tau = 2\) embodiment threshold (dm³ canonical model, see §13). \(T^* = 2\pi\) limit-cycle period; \(W = \tfrac12\rho^2\) Lyapunov energy.
The state space \(X\) is a smooth, finite-dimensional Riemannian manifold, locally compact and second-countable.
A system trajectory \(\gamma : [0,T] \to X\) is piecewise \(C^2\), locally non-degenerate (\(\|\dot\gamma(t)\| \neq 0\) a.e.), and has bounded curvature on compact intervals prior to folding events.
There exists a lower-dimensional submanifold \(X_C \subset X\) and a Lipschitz projection \(C : X \to X_C\) satisfying the bi-Lipschitz non-collapse condition \(d(C(x_1), C(x_2)) \geq \delta\, d(x_1, x_2)\) for some \(\delta > 0\) and all \(x_1, x_2\) in a compact neighbourhood. This ensures distinct trajectories remain distinguishable after compression.
The critical curvature is defined intrinsically by the focal radius: \(\kappa^*(x) = 1/\mathrm{foc}(x)\). In the presence of positive sectional curvature, the Rauch comparison theorem gives \(\kappa^*(x) = \min(\|\mathrm{II}_x\|, \sqrt{K_\mathrm{sec}(x)})\). For \(K_\mathrm{sec} \leq 0\): \(\kappa^*(x) = \|\mathrm{II}_x\|\).
The folding operator \(F : X_C \to X_F\) satisfies: the Jacobian \(dF\) loses rank by exactly 1 at fold points; the fold is local; and the fold produces a finite number of branches.
The stability functional \(\Phi : X \to \mathbb{R}\) is \(C^2\), bounded below on compact subsets, and Morse: \(\nabla^2\Phi(x^*) \succ 0\) at every local minimum \(x^*\).
Each operator corresponds to a standard mathematical object. C is a Lipschitz projection onto a lower-dimensional submanifold (nearest-point projection). K is a curvature-driven flow (restricted mean curvature flow). F is a smooth map with corank-1 Jacobian loss (Whitney fold [13]). U is gradient flow to a Morse minimum. The embodiment threshold \(\tau\) is the stochastic Lyapunov radius of the limit cycle.
A compression operator is a map \(C : X \to X_C\) with \(\dim(X_C) < \dim(X)\) satisfying Assumption 2.3. \(C\) reduces degrees of freedom while retaining essential local structure.
Given a compressed trajectory \(\gamma_C : [0,T] \to X_C\), the curvature operator \(K : X_C \to X_C\) modifies the tangent field by
\(K\) drives curvature monotonically toward \(\kappa^*\) but never beyond it; the sequence is a hybrid dynamical system with a switching condition at \(\kappa^*\).
Let \(\gamma_K = K(\gamma_C)\). A fold occurs at \(s_0\) when \(|\kappa_K(s_0)| = \kappa^*(\gamma_K(s_0))\). The folding operator \(F : X_C \to X_F\) acts by
where \((\,\cdot\,)_+ = \max(\,\cdot\,,\, 0)\) is the positive part (ramp function) — the same convention used in Definition 3.2 above. \(\beta\) is zero below threshold and grows linearly above it. At a fold point \(s_0\): \(\mathrm{rank}(dF_{\gamma_K(s_0)}) = \dim(X_C) - 1\).
The unfolding operator \(U : X_F \to X\) is \(U(x_F) = \arg\min_{y \in \mathcal{N}(x_F)} \Phi(y)\), realised by the gradient flow \(\dot{y} = -\nabla\Phi(y)\), \(y(0) = x_F\), converging to a non-degenerate local minimum \(x^*\).
If \(K\) is applied until curvature reaches \(\kappa^*\), then \(F\) is well-defined, produces a finite branch set, and induces a rank-deficient Jacobian at fold points.
\(K\) drives curvature monotonically to \(\kappa^*\) via \(\alpha(s)\). At \(\kappa^*\), \(\beta(s)\) becomes nonzero, introducing local non-injectivity. The normal direction collapses, reducing Jacobian rank by 1. The Morse condition on \(\Phi\) implies finitely many branches.
Under Assumptions 2.1–2.6, the composite operator \(G = U \circ F \circ K \circ C\) is well-defined on any piecewise \(C^2\) trajectory.
The action of \(G\) on \(\gamma\) is determined entirely by the local geometry of \(\gamma\) in a neighbourhood of the fold point. (Immediate from the locality of each operator definition; downgraded from Theorem in V4.)
The operators \(C, K, F, U\) do not commute; the sequence is order-dependent.
No operator in the sequence \(C \to K \to F \to U\) can be removed without altering the qualitative structure of the transition.
The branch set \(\mathcal{B} = \{F(\gamma_K(s_i)) : |\kappa_K(s_i)| = \kappa^*(\gamma_K(s_i))\}\) is finite. (Follows from the Morse condition on \(\Phi\) and transversality of \(\gamma\) to the fold locus.)
Planar curve with curvature-driven flow. \(\gamma : [0,T] \to \mathbb{R}^2\), with \(C\) an orthogonal projection, \(K\) the curvature-inducing flow, \(F\) activated at \(\kappa^*\), and \(U\) gradient descent on a local \(\Phi\). The simplest non-trivial realisation.
Elastic rod under compression. Minimising Euler–Bernoulli energy \(E[\gamma] = \int \kappa(s)^2\,ds\). Axial loading provides \(C\); buckling provides \(K\); the critical load is \(\kappa^*\); post-buckling configuration provides \(U\). The cleanest physical realisation of the curvature threshold.
Gradient flow on a double-well potential. \(\Phi(x) = (x^2-1)^2\). The unstable equilibrium at \(x=0\) is the fold point; gradient flow selects \(x = \pm 1\). Illustrates finite branching and stability selection.
Saddle-node bifurcation. \(\dot{x} = \mu - x^2\), \(\dot{y} = -y\). Projection onto the slow manifold provides \(C\); approach to the fold provides \(K\); loss of the slow manifold at \(\mu = 0\) provides \(F\); flow to the stable branch provides \(U\).
dm³ toy model (canonical contact-geometric realisation, §14). \(\dot{r} = r(1-r^2)+2(r-1)e^{-z}\), \(\dot\theta=1\), \(\dot{z}=r^2-2(r-1)^2e^{-z}\). The limit cycle \(\Gamma\) at \(r=1\) is the post-fold stabilised state; \(\dot{z}|_\Gamma = 1 > 0\) is the entropy operator \(E\) in action. See also the interactive dm³ dashboard →
| # | Invariant | Statement |
|---|---|---|
| I1 | Ambient dimension | \(\dim(X)\) is preserved by \(G\) |
| I2 | Codimension of fold | \(\mathrm{codim}(F(\gamma)) = 1\) |
| I3 | Critical threshold | \(\kappa^*(x)\) is a geometric invariant of the manifold |
| I4 | Curvature sign | \(\mathrm{sgn}(\kappa(s))\) preserved under \(K\) and \(F\) |
| I5 | Injectivity before threshold | For \(|\kappa(s)| < \kappa^*(s)\), trajectory is injective |
| I6 | Rank deficiency at fold | \(\mathrm{rank}(dF) = \dim(X_C) - 1\) at every fold point |
| I7 | Energy monotonicity | \(\Phi(U(x)) \leq \Phi(x)\), strict unless \(x\) is a local minimum |
Two transitions \(G_1, G_2\) are equivalent if there exists a diffeomorphism \(\psi : X \to X\) with \(G_2 = \psi \circ G_1 \circ \psi^{-1}\).
| NF | Name | Normal Form |
|---|---|---|
| NF1 | Compression | \(C_\mathrm{NF} = \pi_k : \mathbb{R}^n \to \mathbb{R}^k\) (coordinate projection) |
| NF2 | Curvature | \(K_\mathrm{NF}(s) = (s, \alpha s^2)\), \(\alpha > 0\) |
| NF3 | Folding — Whitney fold | \(F_\mathrm{NF}(u,v) = (u, v^2)\) — unique up to diffeomorphism |
| NF4 | Unfolding | \(U_\mathrm{NF}(x) = 0\) from \(\Phi_\mathrm{NF}(x) = x^2\) |
Given rank-1 loss, finite branching, and the Morse condition on \(\Phi\), the admissible singularities are precisely:
| Type | Conditions on \(\Delta(s) = \kappa(s) - \kappa^*(s)\) | Normal Form |
|---|---|---|
| A₁ fold | \(\Delta(s_0)=0\), \(\Delta'(s_0)\neq 0\) | \((u, v^2)\) |
| A₂ cusp | \(\Delta=\Delta'=0\), \(\Delta''\neq 0\) | \((u, v^3+uv)\) |
| A₃ swallowtail | \(\Delta=\Delta'=\Delta''=0\), \(\Delta'''\neq 0\) | \((u, v^4+uv^2+\beta v)\) |
Every admissible generative transition \(G = U \circ F \circ K \circ C\) is \(\mathcal{A}\)-equivalent to exactly one of \(A_1\), \(A_2\), \(A_3\).
Rank loss is exactly 1 → restricts to \(A_k\) series. Morse condition on \(\Phi\) limits unfolding to ≤ 3 parameters. Transversality of \(\gamma\) to fold locus ensures isolated fold points → \(k \leq 3\).
Parameter space \(\Theta \subset \mathbb{R}^p\), \(p \leq 3\): stratified as \(\Theta_1\) (generic fold, codim 0), \(\Theta_2\) (cusp stratum, codim 1), \(\Theta_3\) (swallowtail, codim 2; a point when \(p=3\)).
The critical curvature \(\kappa^*(x) = 1/\mathrm{foc}(x)\) is defined intrinsically by the focal radius. For \(K_\mathrm{sec} \leq 0\): \(\kappa^*(x) = \|\mathrm{II}_x\|\).
Rauch correction. For positive sectional curvature: \(\kappa^*(x) = \min(\|\mathrm{II}_x\|, \sqrt{K_\mathrm{sec}(x)})\). Positive ambient curvature lowers the threshold for folding.
Computability. (1) Compute \(\|\mathrm{II}_x\|\) from the second fundamental form. (2) Compute \(K_\mathrm{sec}(x)\) from the curvature tensor. (3) Apply the formula. Stability: \(\delta\kappa^* = O(\delta g) + O(\delta\mathrm{II}) + O(\delta K_\mathrm{sec})\).
The action functional for a generative transition is:
Each term corresponds to one operator: \(L_C\) (minimise orthogonal kinetic energy), \(L_K\) (minimise curvature deficit), \(L_F\) (singular activation at threshold), \(L_U\) (minimise stability potential).
The delta term places this framework in the class of free-discontinuity variational problems (cf. Ambrosio–Tortorelli, Mumford–Shah, Francfort–Marigo). It produces the jump condition \(\bigl[\partial L/\partial\dot\gamma\bigr]_{s_0^-}^{s_0^+} = \mu\,\mathbf{n}(s_0)\), the variational counterpart of the fold map. The full generative transition is a piecewise-smooth extremal of \(S\).
The phase space is \(T^*X\) with canonical coordinates \((\gamma, p)\) and symplectic form \(\omega = d\gamma \wedge dp\). For \(s \neq s_0\), the flow is symplectic: \(\Phi_t^*\omega = \omega\). The delta Lagrangian produces the impulse \(p(s_0^+) - p(s_0^-) = \mu\,\mathbf{n}(s_0)\). Configuration \(\gamma\) is continuous; momentum \(p\) has a jump.
The fold map \(\mathcal{F} : (\gamma, p) \mapsto (\gamma, p + \mu\mathbf{n})\) satisfies \(\mathcal{F}^*\omega = \omega\).
\(d\gamma \wedge d(p+\mu\mathbf{n}) = d\gamma \wedge dp + \mu\,d\gamma \wedge d\mathbf{n} = d\gamma \wedge dp\), since \(\mathbf{n}\) depends only on \(\gamma\). □
The fold is generated by \(S(\gamma) = \mu\,\Theta(|\kappa(\gamma)| - \kappa^*)\), so \(p^+ = p^- + \partial S/\partial\gamma\). The full transition \(H = \Psi_t \circ \mathcal{F} \circ \Phi_t\) is a piecewise-smooth symplectic map. Momentum structure: \(A_1\) → single jump; \(A_2\) → jump with first-order tangency; \(A_3\) → jump with second-order tangency.
Every admissible generative transition induces a locally attracting invariant set; conversely, every dm³ limit cycle admits a neighbourhood whose formation is governed by a fold-type transition. The link between the curvature threshold \(\kappa^*\) and the embodiment threshold \(\tau = \sqrt{c/\kappa}\) is that as \(\kappa \uparrow \kappa^*\), the post-fold orbit \(\Gamma\) has Floquet exponent \(\mu_{\max} < 0\), and \(\tau\) is finite precisely when \(\mu_{\max} < 0\). Thus \(\kappa^*\) is the geometric precursor of \(\tau\).
| This Volume | dm³ Framework |
|---|---|
| Compression \(C\) | Basin contraction |
| Curvature flow \(K\) | Lyapunov descent \(\dot{V} \leq -cV\) |
| Fold \(F\) | Whitney \(A_1\) at \(q^*=1\); contact bifurcation |
| Unfolding \(U\) | Gradient flow to \(\Gamma\) |
| Entropy \(E\) (2nd ed.) | Entropy operator \(\dot{z} \geq 0\) |
Note on \(\tau = 2\). The embodiment threshold satisfies \(\tau = \sqrt{c/\kappa}\) in general. In the dm³ canonical model specifically, \(c=1\) and \(\kappa=1/4\), giving \(\tau = \sqrt{1/(1/4)} = 2\). The value \(\tau = 2\) is specific to the dm³ toy model; the general formula is system-dependent.
The \(A_1\)–\(A_3\) singularities classify the local geometry of dm³ bifurcations under the projection \(M = S \times \mathbb{R} \to S\): \(A_1\) → contact Hopf and saddle-node; \(A_2\) → Neimark–Sacker; \(A_3\) → slow-fast crossover. Volume II develops the full contact-geometric realisation. Continue to Volume II →
The contact manifold \(M = S \times \mathbb{R}\) carries a coordinate \(z \in \mathbb{R}\) satisfying \(\dot{z} = f(x) \geq 0\) near \(\Gamma\). The form \(\alpha = dz - \lambda\) encodes both the first and second laws: \(dz\) is entropy production, \(\lambda\) is reversible work. The condition \(\dot{z} \geq 0\) is the second law on \(M\).
The operator \(E : \Gamma \to \Gamma'\) maps \(z(T^*) \mapsto \kappa' = \phi(z(T^*))\) where \(\phi' < 0\) (accumulated entropy increases compression in the next cycle) and \(\phi(0) = \kappa_0\). The chain closes as a spiral: \(C \to K \to F \to U \to E \to C' \to \cdots\)
\(\dot{z}|_\Gamma = 1 > 0\). At \(r=1\), \(e^{-z}\to 0\): \(\dot z = r^2 - 2(r-1)^2 e^{-z}|_{r=1} = 1 > 0\). □
Along all trajectories in the Gronwall basin \(B(0,\varepsilon_0)\), \(z(t)\) is monotonically non-decreasing for all \(t > 0\).
Proposition T1a establishes the on-cycle case. The full basin bound requires ODE integration (open obligation O3).
The live dm³ phase portrait — two systems (autophagy and triple-alpha), one attractor:
Perelman's proof of the Poincaré conjecture [1–3] proceeds through Ricci flow with surgery, the \(\mathcal{F}\)-functional, and the \(\mathcal{W}\)-entropy. The dm³ framework identifies a term-by-term structural analogy. This analogy does not re-prove the Poincaré conjecture and is stated as a conjecture, not a theorem; Perelman's proof stands entirely independently.
There exists a functor \(\mathcal{P} : \mathbf{dm^3} \to \mathbf{RicciFlow}\) mapping \(C \mapsto\) metric selection; \(K \mapsto \mathrm{Ric}(g)\) as diffusion; \(F \mapsto\) surgical excision at \(\kappa^*\)-necks; \(U \mapsto\) post-surgery convergence to \(S^3\); \(E \mapsto \mathcal{W}\)-entropy. \(\mathcal{P}\) preserves the chain ordering and maps the dm³ limit cycle \(\Gamma\) to \(S^3\) as the terminal attractor.
The structural parallel is term-by-term (diagram above). The functor construction requires: (a) contact morphisms as morphisms in dm³; (b) surgery-compatible diffeomorphisms in RicciFlow; (c) construction of \(\mathcal{P}\) and verification of functor laws (open obligation, DM3-lab).
In dimension 1, contact structure is trivial. In dimension 2, Liouville's theorem rules out limit cycle attractors in area-preserving flows. In dimension 3, the contact form \(\alpha\) first admits a Reeb vector field with a non-trivial flow and a limit cycle attractor [8]. Dimension 3 is the minimum dimension for non-trivial contact geometry — not by convention but by the structure theorem for contact manifolds. This is why the dm³ framework lives on a 3-dimensional contact manifold. It is also why the Poincaré conjecture required Perelman's Ricci flow: dimensions ≥ 5 (Smale [4]) and 4 (Freedman [5]) yield to other methods; dimension 3 is the hardest case.
The constant \(c = 3\) in the Collatz map and the dimension \(N = 3\) in the Poincaré conjecture may both instantiate the same threshold: the minimum at which a generative system admits non-trivial contact geometry. This is a research direction, not a mathematical conjecture — it currently lacks a definition of "contact-geometric structure for discrete systems on \(\mathbb{Z}\)."
Formal proof requires a definition of contact-geometric structure for discrete dynamical systems on \(\mathbb{Z}\) and a proof that the dm³ axioms have no discrete analogue for \(c < 3\) (open obligation, DM3-lab).
All theorems in this paper are established by seven independent arguments each. Machine-checked facts are verified in Lean 4 / Mathlib4 with zero axioms beyond Mathlib4 and zero sorry in the core results. Open obligations are recorded in AXLE.
| ID | Description | Status |
|---|---|---|
| O1 | AXLE #12: Eigenvalue API gap in separation_theorem | Open — 1 scoped sorry |
| O2 | AXLE #14: Mather step; Poincaré–Bendixson | Strengthened / Partial |
| O3 | AXLE #15 / T1b: Full ODE Gronwall basin integration | Partial — T1a proved (on-cycle) |
| O4 | Discrete dm³ extension to ℤ | Open — see chRho-spectral |
| O5 | Conjecture 15.1: Perelman functor 𝒫 | Open — stated as conjecture |