Chapter 2 gave us flat $\mathbb{R}^3$ with a contact structure. Now the manifold bends. Instead of carrying the contact form on flat space, we carry it on a curved 3-dimensional surface — a contact 3-manifold. The dynamics that were linear in $\mathbb{R}^3$ become nonlinear. A preferred class of curves appears — the Reeb orbits — and one of them is an attractor. This is the setting of the dm³ toy ODE, the helical convergence that anchored Book 3.
This chapter is where the book's geometry becomes genuinely curved, and where the central engine — the dm³ contact manifold and its attractor — first appears in full.
The move from Chapter 2 is this: there the space was flat, and flatness has a cost — nothing global ever happens on it. A manifold looks flat up close but can curve overall; the surface of a sphere is the standard image. Put the contact structure on a curved 3-manifold and you get the dm³ system, whose defining feature is an attractor: a preferred loop that nearby trajectories are pulled toward, wherever in the outer basin they start.
What follows explains what makes a manifold curved, the Reeb vector field (the flow the structure generates), the dm³ manifold itself, the attractor theorem, convergence in the simulator, the G-chain on a curved manifold, a note on Book 3, and what Chapter 4 adds. The attractor theorem is the chapter's heart; the rest supports it.
In Chapter 2, the extended phase space was flat $\mathbb{R}^3$. "Flat" means that parallel lines stay parallel, angles are preserved over long distances, and the Pythagorean theorem holds everywhere. The contact form on flat space is the prototype $\alpha = dz - y\,dx$.
A manifold is a space that looks flat in every small neighbourhood but may curve globally. The surface of a sphere is a 2-manifold: zoom in on any small patch and it looks like a plane, but globally it closes up and has no boundary. A contact 3-manifold is the same idea in one dimension higher, with the contact structure along for the ride.
The curvature changes the dynamics. On flat space, every Reeb orbit (the canonical flow of the contact structure) is a straight line in the $z$-direction. On a curved manifold, Reeb orbits can close up, spiral, or converge — and which behaviour occurs depends on the global shape of the manifold and the specific contact form placed on it.
Given a contact form $\alpha$ on a 3-manifold $M$, there is a canonical vector field attached to it — the Reeb vector field $R_\alpha$ — defined by two conditions:
The first condition says $R_\alpha$ lies in the kernel of $d\alpha$ — it is "invisible" to the symplectic part of the contact structure. The second says $\alpha$ measures $R_\alpha$ as having unit length. Together they pick out a unique vector field at every point of $M$.
The flow lines of $R_\alpha$ are called Reeb orbits. They are the intrinsic dynamics of the contact structure — not imposed from outside, but generated by the geometry of $\alpha$ itself.
A smooth curve $\gamma: \mathbb{R} \to M$ is a Reeb orbit if $\dot{\gamma}(t) = R_\alpha(\gamma(t))$ for all $t$. A closed Reeb orbit (also called a periodic orbit) satisfies $\gamma(T) = \gamma(0)$ for some $T > 0$.
On the standard contact structure $\alpha = dz - y\,dx$ on $\mathbb{R}^3$, the Reeb vector field is simply $\partial_z$ — every orbit is a vertical line, none are closed. On a curved manifold the picture is richer. The key theorem governing closed Reeb orbits is Weinstein's conjecture (proved by Taubes in 2007): on any closed contact 3-manifold, there exists at least one closed Reeb orbit. The circadian clock of Book 3 Chapter 3 is this theorem, instantiated in biology.
The specific contact 3-manifold used in dm³ has coordinates $(r, \theta, z)$ — a cylindrical coordinate system where $r > 0$ is the radial distance, $\theta \in [0, 2\pi)$ is the angle, and $z \in \mathbb{R}$ is the height. The contact form is constructed so that its Reeb flow produces convergence to the unit circle $r = 1$.
The dynamical system placed on this manifold is:
This is not arbitrary. Each term is geometrically motivated:
The $r(1-r^2)$ term in $\dot{r}$ is the radial ODE from Chapter 1, pulling every radius toward $r = 1$. The correction term $2(r-1)e^{-z}$ encodes the coupling between radial and axial dynamics — it vanishes when $r = 1$ (on the attractor) and decays exponentially as $z \to \infty$. The $\dot{\theta} = 1$ says the orbit winds at constant angular speed: one full revolution per unit time. The $\dot{z}$ equation advances height at a rate that also settles when $r \to 1$.
For System (3.1), every trajectory with initial condition $r(0) > 1$ converges exponentially to the unit circle $r = 1$ at rate $\mu = -2$. The attractor is the Legendrian helix $\gamma^* = \{r = 1,\, \dot{\theta} = 1\}$, winding at unit speed around the cylinder $r = 1$. It is Legendrian, not Reeb: its tangent is $\dot\gamma = (0,1,1)$ in $(r,\theta,z)$, and $\alpha(\dot\gamma) = \dot z - r^2\dot\theta = 1 - r^2 = 0$ at $r=1$, so $\gamma^*$ lies inside the contact distribution. The Reeb field of $\alpha = dz - r^2 d\theta$ is $\partial_z$, which is transverse to that distribution and has no closed orbits — consistent with the flat case noted above, to which this manifold is contactomorphic (Ch 23, §23.7).
On the inner side, the true basin boundary is $r^* \approx 0.8$. Trajectories with $r(0) \in (r^*, 1)$ also converge to $\gamma^*$; trajectories with $r(0) < r^*$ diverge inward.
Earlier printings called $\gamma^*$ a closed Reeb orbit. The Reeb field of $\alpha = dz - r^2d\theta$ is $\partial_z$, whose orbits are vertical lines, and the attractor satisfies $\alpha(\dot\gamma)=0$ — the complementary condition. A curve cannot be both: Reeb is transverse to $\ker\alpha$, Legendrian lies in it. The correction strengthens the chapter rather than weakening it, because Legendrian is the condition Ch 22 and Ch 23 need. Verified in ch24-verify.py, block [1].
The rate $\mu = -2$ is the Lyapunov exponent of the linearised flow near the attractor — exactly in the $z\to\infty$ limit, where the coupling term $2(r-1)e^{-z}$ and its derivative vanish; at finite $z$ the linearisation is $-2 + 2e^{-z}$ — the $\mu$ operator in the dm³ recurrence ladder ($\pi \to \phi \to \mu \to \eta \to \Delta \to \Sigma \to \Omega$). It is not a coincidence. The Lyapunov exponent of the contact Reeb flow is the operator $\mu$, instantiated on this manifold.
An earlier Gronwall estimate suggested the inner basin boundary was the symmetric ball $|r - 1| < 1/3 = \varepsilon_0$. Numerical integration (DOP853 integrator, see the GTCT repository) revealed the true boundary is $r^* \approx 0.8$ — asymmetric because the correction term $2(r-1)e^{-z}$ is not symmetric in $r$ around $r=1$. The Lean 4 formalisation in AXLE has been updated to reflect this. $\varepsilon_0 = 1/3$ remains the stability radius[Ch 10] for the linearised system; $r^*$ is the true nonlinear boundary.
The simulation below integrates System (3.1) numerically. The left panel shows the $(r, z)$ portrait — a cross-section of the cylinder. The right panel shows a 3D projection of the helix in $(r, \theta, z)$ space, viewed from a fixed angle. Seed new orbits by clicking the left panel.
Fig 3.1 — Left: $(r,z)$ portrait. Dashed lines: attractor $r=1$ (gold) and basin boundary $r^* \approx 0.8$ (red). Right: helical orbits in $(r, \theta, z)$ space. All outer orbits converge to the unit helix.
We saw in Chapter 2 how the G-chain $G = U \circ F \circ K \circ C$ maps onto the contact structure of flat $\mathbb{R}^3$. On a curved manifold the same mapping holds, but each operator now has geometric content:
C is the contact form $\alpha$ itself — the compression of the full ODE to a single geometric constraint on the manifold.
K is the non-integrability condition, now carrying curvature. The quantity $\alpha \wedge d\alpha$ is not just non-zero; its magnitude encodes how strongly the contact planes twist as you move along the manifold. At the attractor $r = 1$, this twist is maximal — the contact planes rotate fastest precisely where the helix lives.
F is the Reeb flow — the solution operator that carries a point along the Reeb orbit. It builds the trajectory from the contact structure.
U is the projection $\pi: M \to \Sigma$ to the base surface $\Sigma = \{z = 0\}$. The 3D helix projects down to the circle $r = 1$ in the plane — the visible, observable attractor.
Book 3 arrived at this manifold from the biology: circadian clocks (Chapter 3), neural oscillators (Chapter 4), immune memory (Chapter 5) — all were instances of the contact structure on a 3-manifold, interpreted domain by domain. The helical attractor was the unifying object behind all of them.
Book 4 arrives at the same manifold from the mathematics: starting from the 1D ODE, promoting time to a coordinate (Chapter 2), curving the resulting flat space (this chapter). The two paths meet here.
What Book 4 adds is what comes next: Chapter 4 asks what happens when you add a fourth coordinate. The helix is no longer the destination — it is the floor.
We are sitting on a contact 3-manifold with one Legendrian helix as attractor. Chapter 4 introduces a fourth coordinate $w$. The manifold becomes 4-dimensional. The contact structure meets its natural partner — the symplectic structure — and a new object appears: the symplectisation of the contact manifold, a 4-manifold that contains the 3-manifold as a hypersurface.
The question Chapter 4 answers: does the helical attractor persist when you embed the 3-manifold into 4D? The answer is yes — but the proof requires a new tool that does not exist in 3D: holomorphic curves. That is where the dimensional climb truly begins.
The angular period of the dm³ attractor is $T^* = 2\pi$. That $\pi$ is not a label — it is a measurement. One full revolution of the helix $\dot{\theta} = 1$ takes exactly $2\pi$ units of time, returning the angle $\theta$ to its starting value. Note that $z$ does not return: $\dot z = 1$ on the attractor, so $\gamma^*$ is a helix and not a closed curve in $M$. It closes in the $(r,\theta)$ reduction, and $T^*=2\pi$ is the period of that reduction — which is the period the Floquet multiplier of Ch 24 is computed over.
Now ask: what is the simplest space in which a closed orbit with period $2\pi$ and a contact structure naturally coexist? The answer is $S^3$ — the unit 3-sphere.
$S^3$ carries a canonical contact structure. Write $z_j = x_j + iy_j$ for $j = 1, 2$. The standard contact form on $S^3$ is:
The Reeb vector field of $\alpha_{\text{std}}$ generates the Hopf flow: $(z_1, z_2) \mapsto (e^{it}z_1, e^{it}z_2)$. Every orbit is a circle of period $2\pi$. The orbit space — the set of all Reeb orbits — is $S^2$. This is the Hopf fibration: $S^3 \to S^2$ with fibers $S^1$.
$(S^3, \ker\alpha_{\text{std}})$ is the unique (up to contactomorphism) tight contact structure on $S^3$. Every closed Reeb orbit has period $2\pi$. The dm³ contact manifold $M = \mathbb{R}^2_+ \times \mathbb{R}$ carries its helical attractor at $r=1$ with angular period $T^* = 2\pi$. Being contactomorphic to the standard structure on $\mathbb{R}^3$ (Ch 23, §23.7), $M$ is locally contactomorphic to $(S^3,\ker\alpha_{\text{std}})$ away from a point — every contact 3-manifold is locally standard, by Darboux. What that does not transfer is the closed-orbit statement: the $2\pi$ of the Hopf flow is the period of a closed Reeb orbit, while the $2\pi$ of the dm³ attractor is the angular period of a Legendrian helix. The two numbers agree; the objects do not, and the agreement is not evidence.
This is why $\pi$ appears in the dm³ system as the first rung of the recurrence ladder — it is not the area constant or the ratio of circumference to diameter arriving by coincidence. It is the period of the Reeb orbit on the contact 3-manifold, and that manifold is, at its canonical core, $S^3$.
The dm³ coordinates $(r, \theta, z)$ describe a solid cylinder. The attractor $r=1$ is a surface inside it. But the geometry near the attractor — the contact planes, the Reeb flow, the period $2\pi$ — is that of $S^3$. The cylinder is a chart. The hypersphere is the canonical model. When Chapter 4 adds a fourth coordinate $w$, it will be adding the ambient $\mathbb{R}^4$ that $S^3$ has always been sitting inside.
The dimensional ladder now has a sharper picture. The 3D step is not "some curved manifold" — it is $S^3$, the unit hypersphere in $\mathbb{R}^4$. Chapter 4 is the step into $\mathbb{R}^4$ itself, where $S^3$ lives as a hypersurface. The climb from 3D to 4D is the move from the sphere to its ambient space.