⚜ PRINCIPIA ORTHOGONA · Vol VI · Roots · WP-120 WP-122 · The Return Map Was in the Exercise →
Vol VI · Roots · WP-120 · 2026-09-15 · Dynamical systems · Instruments

How Many Closed Orbits

The series calls Γ the limit cycle ten to seventeen times a page and has never counted them. Two instruments settle it, both from chapters the corpus already owns. One of the two toy models turns out to carry a second circular orbit that nothing in the corpus names — and the neutral line at $z=0$ is where the two collide.
Methodindex theory and Dulac's criterion, applied to the corpus's own fields
eight verification blocks, standard library only
Claim typea count, with proofs; one elementary new object
no claim about the three-dimensional flows — see §5
Verified 2026-09-15book6/wp120-verify.py, 23 checks, exit 0
divergence formulas against numerical differentiation
A definite article is a claim. Across the corpus Γ = {r = 1} is the limit cycle, and the only place uniqueness is asserted anywhere in the series is one row of chRho-spectral's open-obligation table, about the discrete Collatz cycle. For the continuous Γ nothing has been proved, nothing has been claimed, and the word has been standing in for both.

1 · When the count is free

If $\dot r$ depends on $r$ alone and $\dot\theta > 0$, there is nothing to prove. A closed orbit needs $r(t)$ periodic; $r$ is monotone wherever $\dot r \ne 0$; a monotone periodic function is constant. So the closed orbits are exactly the circles $r = $ root of $f$ in $r > 0$, and the count is a root count. Dulac is not needed and index theory is not needed.

Applied to Book 6's $\dot r = f(r)(1-e^{-z})$ — the modulation is a positive factor for $z>0$, so the roots are $f$'s own — both canonical closures give exactly one, at $r=1$. That is the whole answer for that model, and it took a root scan.

2 · Vol II has two, and the neutral line is where they meet

Volume II's toy model is $\dot r = r(1-r^2) + 2(r-1)e^{-z}$. Write $a = 2e^{-z}$ and factorise — block [5] checks this is an identity:

r(1 − r²) + a(r − 1) = −(r − 1)(r² + r − a)

Besides $r=1$ there is a second positive root. The frozen slice has two circular orbits, not one:

r₂(a) = ( −1 + √(1 + 4a) ) / 2, with a = 2e^(−z), so r₂(z) = ( −1 + √(1 + 8e^(−z)) ) / 2

and the two radial eigenvalues are $a-2$ at $r=1$ and $(1-r_2)\sqrt{1+4a}$ at $r=r_2$. Both checked against numerical differentiation at six values of $z$.

za = 2e−zr₂eigenvalue at Γeigenvalue at r₂
1.500.4462600.33442215−1.553740+1.110746
0.750.9447330.59303847−1.055267+0.889649
0.251.5576020.84447074−0.442398+0.418209
0.002.0000001.000000000.0000000.000000
−0.252.5680511.17870511+0.568051−0.599986
−0.754.2340001.61754576+2.234000−2.615363
The neutral line, renamed $r_2 = 1$ exactly at $a = 2$, that is at $z = 0$, and both eigenvalues vanish together there. So the corpus's neutral line is not merely a place where an eigenvalue changes sign: it is the collision of two circular orbits, which pass through each other and exchange stability — a transcritical bifurcation of cycles. For $z > 0$ the inner circle $r_2 < 1$ is repelling, and it is the basin boundary of $\Gamma$ in that slice. For $z < 0$ the roles are swapped and $\Gamma$ is the repeller.

Book 6's model has no second orbit at all. The two toy models the series runs in parallel are not variants of one another: a multiplicative modulation $f(r)(1-e^{-z})$ preserves the root set, an additive coupling $+\,a(r-1)$ adds a root. Whichever is meant, the difference has to be stated where both are cited.

A guard, before anyone matches a decimal $r_2(z)$ is strictly increasing in $a$ with $r_2 \to 0$ and $r_2 \to 1$ at the ends, so as $z$ runs over $(0,\infty)$ it takes every value in $(0,1)$ exactly once. Block [6] finds the $z$ that produces $0.25$, $0.5$, $0.773$, $0.882$ and $0.99$, to $10^{-9}$ each. A one-parameter family that sweeps an interval meets every constant in it, so a numerical agreement between $r_2(z)$ and a stored constant — $r_\star$, $\kappa^*$, anything — is not evidence. It would become evidence only if the $z$ at which it matched were independently fixed. None is.

3 · When you do need Dulac

The moment $\dot r$ depends on $\theta$, the free count is gone. Strogatz's Example 7.3.1 is the case at hand, because it perturbs the corpus's own field:

Ṛ = r(1 − r²) + μ r cosθ, θ̇ = 1

Plain Bendixson fails: $\nabla\cdot F = 2 - 4r^2 + 2\mu\cos\theta$ changes sign at $r = 1/\sqrt2$. A Dulac function is needed, and Strogatz is candid on p. 204 that there is no algorithm for finding one, listing $g = 1$, $1/x^ay^b$, $e^{ax}$, $e^{ay}$ as the candidates that occasionally work. The polar analogue of the second works here:

g(r) = 1/r³ ⇒ div(gF) = −(1/r) [ (1 + μ cosθ) / r² + 1 ]

strictly negative on all of $r>0$ exactly when $|\mu| < 1$ — the same threshold under which Example 7.3.1's trapping annulus works. Block [2] checks the closed form against numerical differentiation at nine points, scans a grid at six values of $\mu$, and confirms the sign flips once $\mu = 1.5$.

Index theory supplies the missing half

Dulac as printed requires a simply connected region and concludes no closed orbits. The punctured plane is not simply connected, and there is a closed orbit, so the theorem does not apply as written. Two things bridge the gap.

First, Theorem 6.8.2, p. 180: any closed orbit must enclose fixed points whose indices sum to $+1$. Block [3] scans $r \in (0.01, 4]$ on an $800\times720$ grid and finds $\min|F| = 0.0168$, so the origin is the only fixed point; the winding number on circles of radius $0.05$ through $5.0$ is $+1$ on every one. A closed orbit avoiding the origin would enclose nothing and sum to zero. So every closed orbit encircles the origin, and any two are nested with an annulus between them inside $r > 0$.

And second, one extra line of Strogatz's own proof If $C_1$ lies inside $C_2$, both closed orbits, and the annulus $A$ between them lies in $R$, then Green's theorem gives $$\iint_A \nabla\cdot(g\mathbf{F})\,dA \;=\; \oint_{C_2} g\mathbf{F}\cdot\mathbf{n}\,d\ell \;-\; \oint_{C_1} g\mathbf{F}\cdot\mathbf{n}\,d\ell \;=\; 0,$$ because $\mathbf{F}$ is tangent to each orbit and so orthogonal to $\mathbf{n}$. If $\nabla\cdot(g\mathbf{F})$ has one sign on $R$, the left-hand side cannot vanish. Hence $R$ contains at most one closed orbit.

This is not the theorem printed on p. 204. It is that theorem's proof run on an annulus instead of a disc, and it is the form that gives uniqueness rather than absence. A reader should check the line rather than take it on the citation.

4 · Exactly one

Existence for $|\mu| < 1$ is Example 7.3.1's trapping annulus, checked in WP-122 block [7]. Uniqueness is the paragraph above. Together, for every $|\mu| < 1$, the perturbed system has exactly one closed orbit. Block [4] corroborates by integrating from five starting radii between $0.05$ and $3.0$ at three values of $\mu$ and finding a spread of $10^{-13}$ on the return ray — corroboration, not proof: three values of $\mu$ against a theorem that covers all of them.

A second route, and it is older — added 2026-09-16

The argument above was built here from index theory and an annulus form of Dulac. There is a classical theorem that reaches the same kind of conclusion without either, and it was published in 1928: Liénard's Theorem (Strogatz p. 212). For $\ddot x + f(x)\dot x + g(x) = 0$ with $f$ even, $g$ odd and positive on $x>0$, and $F(x) = \int_0^x f$ having exactly one positive zero at $x = a$, negative below it and positive and nondecreasing above it with $F \to \infty$, the system has a unique, stable limit cycle surrounding the origin. Strogatz cites Jordan & Smith (1987), Grimshaw (1990) and Perko (1991) for the proof.

The two routes are not interchangeable, and the difference is recorded in ch-van-der-pol. Dulac with $g = 1/r^3$ works here because this cycle is a circle and the field is radial plus rotation; on van der Pol, whose cycle is not a circle, the same $g$ changes sign on the same grid and the route fails. Liénard's theorem never assumed a circle. This paper's result stands as proved; what it does not have is reach.

5 · And the three-dimensional flows have no periodic orbit to be unique

Book 6's model has $\dot z = 1$ identically, so $z(t) = z_0 + t$ is strictly increasing and no trajectory returns to its starting point. That flow has no periodic orbit at all. $\Gamma = \{r=1\}$ is a helix, and a helix is a closed orbit of the $(r,\theta)$ projection, not of the flow — which is the same fact WP-122 reports as the absence of a return map, seen from the other side.

Vol II's $\dot z = r^2 - 2(r-1)^2e^{-z}$ is $1$ on $\Gamma$ and positive on a tube around it: block [7] scans $|r-1| \le 0.4$, $z \in [0,12]$ and finds $\min \dot z = 0.040000$, at $(r,z) = (0.600, 0.00)$, with the analytic bound $(1-\delta)^2 > 2\delta^2$ holding whenever $\delta < 1/(1+\sqrt2) = 0.4142$. So no periodic orbit lies in that tube either. Outside it $\dot z$ does change sign for Vol II, so this is a statement about a neighbourhood of $\Gamma$; Book 6's $\dot z = 1$ needs no neighbourhood.

What follows for the vocabulary “Limit cycle” is correct for the planar transverse system and for the frozen slices. For the object on the contact 3-manifold it is loose: that object is a helical orbit, and the series' own WP-22 says as much in a parenthesis — “a periodic orbit of period $T^*=2\pi$ (a helix in $(r,\theta,z)$, since $\dot\theta=\dot z=1$)” — where the parenthesis contradicts the phrase it qualifies. Counting them settles which reading is meant: on the 3-manifold there are none.

Carried into the source, 2026-09-15. WP-22's abstract and §2 now state that the flow has no periodic orbit and that $T^*$ is the period of the $(r,\theta)$ projection; the revision also names the sense of “degenerate,” says where $z$ is frozen in Theorem 5.1, and cites Example 7.1.1 where its data are used. Listed in that paper's own Corrections section, dated. No result was withdrawn.

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. Theorem 6.8.1 p. 179 · Theorem 6.8.2 p. 180 · Dulac's criterion p. 204 · Example 7.3.1 p. 206 · Liénard's Theorem p. 212, Example 7.4.1 p. 213.
  2. book6/wp120-verify.py — 8 blocks, 23 checks, standard library only, exit 0. Polar divergence against Cartesian numerical divergence; $g=1/r^3$ analytic against numerical, scanned at six values of $\mu$ and failing at $\mu=1.5$; the fixed-point scan and winding numbers; the uniqueness corroboration; the factorisation as an identity and both eigenvalue formulas; the sweep guard; the $\dot z$ tube.
  3. WP-122 — the closed-form return map, the Grönwall radius measured, and the trapping annulus that supplies existence here.
  4. ch-strogatz · WP-22 · vol2-toymodel · vol2-contact · Sessão S2 · chRho-spectral · WP-92