Polynomial limit cycles: uniform bounds and the quintic Liénard case

Source and verification

This article explains family 143 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

🏷️ Limit cycles and return maps

For a planar polynomial vector field

$$\dot{x}=P(x,y),\dot{y}=Q(x,y),$$

a periodic orbit is the image of a nonconstant periodic solution. A limit cycle is a periodic orbit isolated from all other periodic orbits in some neighborhood. Geometric images are counted once, regardless of multiplicity or stability. A center surrounded by periodic trajectories therefore contributes no limit cycles.

A small transverse interval through a periodic orbit carries a first-return map $\Pi$. Nearby cycles correspond to isolated zeros of $\Pi (r)-r$. A simple zero has ${\Pi }^{\prime }(r)\ne 1$ and gives a hyperbolic cycle. Multiple isolated zeros also count.

The two manuscripts address different levels of uniformity: a degree-dependent bound for arbitrary polynomial fields, and an exact bound for a special quintic family. (OpenAI, 2026) (OpenAI, 2026a)

🏷️ Uniform degree bounds and an exact special case

General statement in the first manuscript

For each integer $d\ge 1$, there is a finite integer $B(d)$ bounding the number of limit cycles of every real planar polynomial vector field of degree at most $d$, throughout the plane. No coefficient, stability, or location restriction is imposed. The proof does not provide an effective formula for $B(d)$. (OpenAI, 2026b)

Individual finiteness for every field does not imply a uniform bound over all coefficients. The elementary analogy is a family of finite sets whose cardinalities tend to infinity. Degenerating coefficients and cycles escaping to infinity must both be handled.

The companion concerns the Liénard system

$$\dot{x}=y-F(x),\dot{y}=-x,$$

where $F$ is a real polynomial of degree at most five. It claims that the exact maximum number of limit cycles is two, without parity, coefficient-sign, amplitude, or hyperbolicity assumptions. (OpenAI, 2026a)

Differentiating the first equation gives

$$x^{\prime \prime }+F^{\prime }(x)x^{\prime }+x=0.$$

Thus the degree-five hypothesis is on the primitive $F$; the damping polynomial has degree at most four, and the restoring force is exactly $x$. The result is not an exact count for arbitrary quintic planar fields.

🏷️ A direct two-cycle construction

The companion’s lower bound has an accessible perturbative proof. Set

$$\begin{array}{c}{F}_{\epsilon }(x)=\epsilon f(x),\\ f(x)=4x-\frac{20}{3}{x}^3+\frac{8}{5}{x}^5.\end{array}$$

At $\epsilon =0$, the orbit through $(0,s)$ is $x=s\sin t,y=s\cos t$. For $E=(x^2+y^2)/2$, the exact identity is $\dot{E}=-\epsilon xf(x)$.

Let $P(s,\epsilon )$ be the positive-axis return ordinate and $\tau (s,\epsilon )$ the return time. Smooth dependence and transversality give

$$\begin{array}{c}\frac{P(s,\epsilon {)}^2-s^2}{2}=\epsilon Q(s,\epsilon ),\\ Q(s,\epsilon )=-{\int }_0^{\tau (s,\epsilon )}x(t)f(x(t))dt.\end{array}$$

The integral defines the quotient smoothly even at $\epsilon =0$. Using the integrals of ${\sin }^{2}t,{\sin }^{4}t,{\sin }^{6}t$ over one period gives

$$Q(s,0)=-\pi s^2(s^2-1)(s^2-4).$$

Its positive zeros at $1$ and $2$ are simple, with derivatives $6\pi$ and $-48\pi$. The implicit function theorem preserves two distinct roots for small positive $\epsilon$. At such a root,

$$s(P_s-1)=\epsilon Q_s\ne 0.$$

The corresponding cycles are isolated; the inner one repels and the outer one attracts. This proves existence of two cycles. It does not by itself exclude distant additional cycles, which is the separate upper-bound problem. (OpenAI, 2026a)

🏷️ The quintic upper-bound mechanism

The upper proof keeps the positive and negative half-orbits separate. In either half-plane, the substitution $u=x^2/2$ gives an arch equation

$$\begin{array}{c}\frac{du}{dy}=\varphi (u)-y,\\ \varphi (u)=F(\pm \sqrt{2u}).\end{array}$$

At a fixed height, the two endpoint ordinates determine a half-width and a midpoint displacement. Even coefficients of $F$ contribute the same quadratic part to the two profiles, while odd coefficients change sign.

The manuscript proves that the midpoint data and its width derivative admit a unique global fit by a quadratic comparison profile. Both the range of this fit and its nonvanishing Jacobian matter: a merely local fit would not control all amplitudes.

Transport equations for the fitted slope and curvature reduce comparison of half-orbits to monotonicity. A model estimate uses endpoint variation, a Riccati equation, and a Schwarzian identity. The quintic coefficient structure then gives a sign partition in which the relevant matching function can have at most two isolated zeros. The source counts degenerate zeros directly rather than deleting them through a generic perturbation.

🏷️ Uniformity through degenerations

For the general theorem, a saddle already exhibits competing scales. The system $\dot{x}=x,\dot{y}=-\lambda y$ sends $(r,1)$ to $(1,r^{\lambda })$, with flight time $\log (1/r)$. As both $r$ and $\lambda$ vary, flight time and contraction depth can behave very differently.

The first manuscript encodes passage maps together with asymptotic expansions, derivative bounds, and complex-domain information. It proves separation and implicit-closure properties for that class; formal real asymptotic flatness alone would not suffice. (OpenAI, 2026b)

A finite geometric decomposition for bounded-degree fields supplies passage templates. Matching endpoints gives auxiliary equation systems. The proposed counting principle converts absolute isolated-zero finiteness for the required differentiated systems into uniform bounds on connected components of parameter fibers.

The finite templates are chosen before the sequences used in asymptotic contradiction arguments. Sequence-dependent local charts are not treated as a finite cover of coefficient space. Finally, a signed rotation transfers the hyperbolic count to arbitrary isolated cycles, and spatial rescaling addresses the whole plane.

🏷️ Key steps and proof outlines

An integrating factor reduces the quintic count to one sign change

Write the two half-plane profiles as ${\varphi }_{\pm }(u)=d_{0}u+b_0{u}^2\pm p(u)$, where $p(u)=e{u}^{1/2}+c{u}^{3/2}+a{u}^{5/2}$ and $a\ge 0$ after reflection and time reversal. The even coefficients $d_0,b_0$ remain unrestricted. Let $M_{\pm }(r)$ be the fitted midpoints at height zero and ${\kappa }_{\pm }(r)$ the fitted curvatures. The source’s global quadratic-fit theorem supplies a smooth model function $\hat{v}(r,M,\kappa )$ satisfying ${\hat{v}}_M>0$, ${\hat{v}}_{\kappa }>0$, and $M_{\pm }^{\prime }=\hat{v}(r,M_{\pm },{\kappa }_{\pm })$.

In the case $e\ge 0,c<0$,

$$p^{\prime \prime \prime }(u)=\frac{3e-3cu+15a{u}^2}{8u^{5/2}}>0.$$

The transport theorem then makes $q(r)={\kappa }_{+}(r)-{\kappa }_{-}(r)$ nondecreasing. For $\Delta =M_{+}-M_{-}$, subtraction and the fundamental theorem of calculus yield

$${\Delta }^{\prime }=A(r)\Delta +B(r)q(r),B(r)>0.$$

Here $A$ averages ${\hat{v}}_M$ along the midpoint segment and $B$ averages ${\hat{v}}_{\kappa }$ along the curvature segment. Multiplying by $\exp (-\int A)$ gives a derivative with the sign of $q$. The transformed function decreases, may be constant on an interval, then increases. It therefore has at most two isolated zeros. A zero interval contributes no isolated cycles, and a multiple minimum zero is counted correctly. This is the decisive zero-counting mechanism in “The quintic upper bound.” (OpenAI, 2026a)

The other coefficient cases retain strict boundary information

For $e,c\ge 0$, either $p=0$ and the half-profile matching function vanishes identically, or $p(u)>0$ for all $u>0$. In the latter case, two arches with the same endpoints would have difference $w$ satisfying a linear equation with strictly positive forcing and zero values at both endpoints. The integrating-factor formula rules that out.

For $e<0,c\ge 0$, the identity $p^{\prime \prime }(u)=(-e+3cu+15a{u}^2)/(4u^{3/2})>0$ separates the fitted curvatures. At every matching zero, ${\hat{v}}_{\kappa }>0$ gives ${\Delta }^{\prime }>0$, so there is at most one zero. Two upward crossings would require an intervening non-upward crossing.

For $e,c<0$, the relevant quantity is $u{p}^{\prime \prime }-p^{\prime }>0$. The transport theorem gives opposite strict inequalities for the fitted intercepts ${\lambda }_{\pm }-h{\kappa }_{\pm }$ at positive height $h$. Strictness cannot be preserved merely by taking $h\downarrow 0$. Along a fixed-peak characteristic, the derivative of the intercept has the sign of ${\lambda }_{\pm }-{\varphi }_{\pm }^{\prime }(h)$. The fitted parameters stay bounded, whereas ${\varphi }_{+}^{\prime }(h)\to -\infty$ and ${\varphi }_{-}^{\prime }(h)\to +\infty$. Integrating from one small positive height preserves a definite strict gap at zero. The resulting slope ordering again forces the curvature ordering at a matching zero and hence ${\Delta }^{\prime }>0$.

Signed rotation transfers a hyperbolic bound to all isolated cycles

The general-degree manuscript first bounds hyperbolic cycles through its finite passage templates and component-counting theorem. To include degenerate isolated cycles, perturb $V=(P,Q)$ by $V_{\mu }=V+\mu (-Q,P)$. For a periodic solution $z(t)$, its parameter variation $u$ satisfies $u^{\prime }=DV(z)u+JV(z)$, where $J$ is a quarter-turn. The transverse determinant $w=\det (V,u)$ obeys

$$w^{\prime }=(\mathrm{div}V)w+|V{|}^2,w(T)>0.$$

Thus the return displacement $D(r,\mu )$ has $D_{\mu }\ne 0$. Its zero set is locally $\mu =\psi (r)=a{r}^m+O(r^{m+1})$, with $a\ne 0$ because the original cycle is isolated. An odd $m$ yields one simple nearby root for either sign of $\mu$; an even $m$ yields two for its favorable sign.

For a finite collection, let $O$ count odd-order cycles and $E_{+},E_{-}$ the even-order cycles favoring each sign. One common small perturbation produces at least $O+2E_{+}$ or $O+2E_{-}$ hyperbolic cycles. The larger count is at least the original total. This does not claim every original cycle persists for the chosen sign. Disjoint neighborhoods keep the resulting cycles distinct, and the perturbation preserves polynomial degree.

Finally, any finite collection anywhere in the plane can be moved inside one fixed square by affine spatial scaling, again preserving degree. A uniform square bound therefore bounds every finite collection, hence the whole set. This closes the global and degenerate-cycle interfaces; the finite-template and component-counting results remain the substantive source inputs behind that uniform bound. (OpenAI, 2026b)

See Also

Entropy for the standard map also requires control of degenerating derivative growth. Mass-action permanence studies global bounds for another structured class of polynomial differential equations.

🏷️ Manuscripts and verification scope

• Uniform bounds for planar polynomial limit cycles — pinned manuscript and companion source.
• Two limit cycles for quintic Liénard systems — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. Despite its general heading, the scope selects only the exact two-cycle theorem for quintic Liénard systems. The arbitrary-degree uniform bound is outside the documented selection. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026a. Two limit cycles for quintic Liénard systems.

🐻  OpenAI 2026b. Uniform bounds for planar polynomial limit cycles.