⚜ PRINCIPIA ORTHOGONA · Vol VI · Roots · WP-122 ← WP-121 · What ETF Connect Actually Couples
Vol VI · Roots · WP-122 · 2026-09-15 · Dynamical systems · Instruments

The Return Map Was in the Exercise

Strogatz computes the Poincaré map of this corpus's transverse attractor in closed form, on page 282, as a worked example. The series has been using its derivative at one point and nothing else. Here is what the whole map gives — a bound tightened by a factor of eighty-five thousand, a basin certificate of the kind WP-62 asked for, and one paragraph whose arithmetic does not close.
Methodtextbook instruments applied to the corpus's own objects
eight verification blocks, standard library only
Claim typean instrument audit, not a new theorem
every result quoted is Strogatz's or follows in one line
Verified 2026-09-15book6/wp122-verify.py, exit 0
closed form against RK4 to 1e-15 relative
The series has an attracting cycle, a period, a transverse exponent and a Floquet multiplier, and it has had them for two years. What it has never had is a return map. The map exists, it is elementary, it is on page 282 of the book already sitting in WP-22's bibliography, and it answers questions the corpus has been answering with estimates.

1 · The map

The transverse system, as ch-strogatz establishes, is Strogatz's Example 7.1.1: $\dot r = r(1-r^2)$, $\dot\theta = 1$. Because $\dot\theta = 1$, a trajectory crosses any ray $\theta = \text{const}$ at intervals of exactly $2\pi$, so that ray is a surface of section and the first-return map is available in closed form. Example 8.7.1, page 282:

P(r) = [ 1 + e^(-4π) ( r^(-2) − 1 ) ]^(-1/2)

Block [1] of the verify script integrates $\dot r = r(1-r^2)$ over $t \in [0, 2\pi]$ from eight starting radii between $0.02$ and $40$ and compares. Agreement is at the $10^{-15}$ level — the map is not an approximation to the flow, it is the flow, sampled once per turn.

Where the corpus's number lives Differentiating at the fixed point gives $P'(1) = e^{-4\pi} = 3.487342356\times10^{-6}$. That is the transverse Floquet multiplier ch-feynman uses for its periodic-orbit weight and ch-grothendieck measures as an index candidate. It is one number read off a function the series has never written down.

The bookkeeping, checked first

Strogatz's Liapunov exponent for a $p$-cycle (p. 374, Example 10.5.1) is $\lambda = \frac1p\ln|(f^p)'(x_0)|$; per unit time for a flow, $\frac1{T^*}\ln|\text{multiplier}|$. Block [2] runs it: $\frac{1}{2\pi}\ln e^{-4\pi} = -2$ exactly. The series' use of $\mu_{\max} = -2$ alongside $e^{-4\pi}$ is correct — they are one fact in two units. That is stated here before §4, which is about the one place they are conflated.

2 · What the whole map buys: the Grönwall radius

chEps-gronwall Proof IV writes $\rho = r-1$, $f(\rho) = -\rho(1+\rho)(2+\rho)$ — which block [3] confirms is $r(1-r^2)$ rewritten — bounds $|P'|$ by Grönwall as $\exp\!\big((\mu_{\max} + 6\varepsilon)T^*\big)$ on $|\rho| \le \varepsilon$, and reads off $\varepsilon_0 = 1/3$ as the radius where that bound reaches 1. The derivation is correct. The bound is very loose.

quantity on |ρ| ≤ 1/3value
Grönwall bound on |P′| at ε = 1/31.000000
exact sup |P′|, attained at ρ = −1/31.176970 × 10−5
the bound exceeds the truth by8.496 × 104
radius at which |P′| actually reaches 1r = 0.015049224

So $P$ is a contraction from $r \approx 0.0150$ outward — a ball about three times wider in $\rho$ than $|\rho| \le 1/3$, and the contraction inside it is five orders of magnitude stronger than the bound admits.

What ε0 = 1/3 is, and what it is not It is the radius at which a Grönwall estimate saturates. Block [4] iterates the exact map twenty times from initial radii spanning ten decades, $10^{-6}$ to $10^{4}$, and every one lands on $\Gamma$ to $10^{-8}$ — the basin of the uncoupled transverse flow is all of $r > 0$, with no boundary but the axis. Whatever $\varepsilon_0$, $r_\star$ and $\kappa^*$ are thresholds of, it is not convergence of this flow. Vol I's own note already warns that writing $\varepsilon_0 \approx r_\star$ is an identification it is not, and The r★ Mess records what that constant has cost; this paper adds only the exact number that settles the $\varepsilon_0$ half.
A defect the same block turned up in its own first draft Fixed-step RK4 on $\dot r = r - r^3$ returns nan for $r_0 \gtrsim 30$; the equation is stiff out there and the integrator overshoots. The first version of block [4] tested if abs(r - 1.0) > 1e-8: fail — and every comparison against nan is false, so two nan rows passed as convergence. The guard now tests isfinite first. The nan rows are kept in the printed table on purpose.

3 · Three more instruments, and what each one is for here

instrumentwherethe corpus question it answers
Trapping region + Poincaré–Bendixson§7.3, pp. 205–206Example 7.3.1 perturbs this very system to $\dot r = r(1-r^2) + \mu r\cos\theta$ and proves a closed orbit survives for $\mu < 1$, inside $0.999\sqrt{1-\mu} < r < 1.001\sqrt{1+\mu}$. The proof is two inequalities that hold for every $\theta$ because $|\cos| \le 1$. WP-62 asks what it takes for a basin result to be “publishable as a certificate rather than an illustration”; this is the analytic answer, and block [7] checks the annulus at five values of $\mu$.
Dulac's criterion§7.2, pp. 201–204Rules out closed orbits in a simply connected region by exhibiting one function $g$ with $\nabla\cdot(g\mathbf{f}) < 0$. The series asserts $\Gamma$ is the cycle and has nowhere proved there is no other. One mention in the corpus, in WP-92.
Liapunov functions§7.2, p. 203The same $V$-decreases argument the contact chapters use informally, with the two conditions stated as conditions. Strogatz is candid that there is no systematic way to construct one, which is itself worth quoting in a chapter that produces one.

4 · One paragraph whose arithmetic does not close

chRho-spectral, Argument V of seven, reads: “the linearised Poincaré map at $\Gamma$ has eigenvalues $\{\mu_{\max}, \pm i\} = \{-2, \pm i\}$… the first zero at $t_1 \approx 14.135$ gives $t_1/(2\pi) \approx 2.25 \approx \tau\cdot\varepsilon_0^{-1}\cdot(1/3)$ — within the dm³ scaling of the first spectral gap.” Block [5] runs it.

as writtencomputed
$t_1/(2\pi)$2.249611375552
$\varepsilon_0^{-1}\cdot(1/3)$1.000000000000the two factors cancel
$\tau\cdot\varepsilon_0^{-1}\cdot(1/3)$2.000000000000so the right-hand side is $\tau$
relative gap12.4806 %
Three separate things, and each has a fix on a numbered page

The units. $-2$ is an exponent and $e^{-4\pi}$ is the multiplier; a return map's eigenvalues are multipliers. The two numbers differ by a factor of $5.7\times10^{5}$. Strogatz p. 374 gives the conversion, and §1 above shows the corpus elsewhere applies it correctly.

The count. Strogatz p. 281: a surface of section for an $n$-dimensional flow is $(n-1)$-dimensional, so the linearisation of $P$ is $(n-1)\times(n-1)$ — one multiplier for the planar transverse system, two on a contact 3-manifold. A spectrum listed as three numbers fits neither, and a modulus-one entry would assert a neutral direction this system does not have.

The scaling. $\varepsilon_0^{-1}\cdot(1/3) = 1$ identically, so $\varepsilon_0$ does not appear in the expression at all; the right-hand side is $\tau = 2$ wearing two cancelling factors. And $2.2496$ against $2$ is a 12.5 % gap presented as agreement. This is vocabulary-correspondence in its exact form — two numbers near each other, joined by an expression that looks structural and is not.

Scope. The other six arguments on that page were not examined and nothing here is a verdict on them, on RH, or on whether a contact reading of the critical strip is worth pursuing. The page also carries no MODEL / OPEN tag on any of the seven, which the house rules require.

5 · The one the corpus cannot borrow

Every result above is about the uncoupled transverse system. The full model is $\dot r = f(r)(1-e^{-z})$, $\dot\theta = 1$, $\dot z = 1$, and it has no Poincaré map at all: $\dot z = 1$ everywhere, so no trajectory ever returns to a section $\{z = z_0\}$. What the series computes over one period is a time-$2\pi$ flow map, not a first-return map. The two agree on the transverse coordinate and nowhere else, and the drift of the per-period exponent with base point — $-4\pi + 2e^{-z_0}(1-e^{-2\pi})$, reaching $-4\pi$ only in the limit — is exactly that gap, measured.

Open, and handed forward Find the right substitute for a return map on a flow with a monotone coordinate. The candidates are standard — a section in the extended phase space, an asymptotic (pullback) attractor in the non-autonomous sense, or a rescaling of $z$ that compactifies the sweep — and choosing among them is the work. Until then, “Poincaré map” should not appear in the series for an object on the 3-manifold. In chEps-gronwall's planar reduction the term is correct, because there the time-$2\pi$ map is the first return.

6 · What this paper does not claim

References

  1. S. H. Strogatz, Nonlinear Dynamics and Chaos, 2nd ed., Westview 2015 / CRC 2018, ISBN 978-0-8133-4910-7. §7.1 p. 199 · §7.2 pp. 201–204 · §7.3 pp. 205–206 · §8.7 pp. 281–282 · §10.5 pp. 373–374.
  2. book7/strogatz-citations-verify.py — every page and example number above, checked against the book on 2026-09-17; exit 0, thirteen of thirteen. They had not been checked before that run. One thing the check turned up and this paper did not know: on p. 282 Strogatz computes $P$ and defers the integral itself to Exercise 8.7.1, so the title is exact rather than figurative.
  3. book6/wp122-verify.py — 8 blocks, standard library only, exit 0. Closed form against RK4 at eight radii; exponent-from-multiplier; Grönwall bound against exact supremum and the unit-slope radius by bisection; twenty iterates of $P$ over ten decades with an isfinite guard; Argument V's arithmetic; the section-dimension count; the trapping annulus at five values of $\mu$; a control block.
  4. ch-strogatz — which of the helix's constants came with Example 7.1.1, and which are the contact structure's.
  5. WP-22 · chEps-gronwall · chRho-spectral · The r★ Mess · WP-62 · WP-92 · ch-feynman · ch-grothendieck