How many isolated oscillations can a polynomial system sustain

← All digests

Family 143 · Dynamical systems and ergodic theory
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-21

1. The question

A pendulum with friction eventually stops, but a system that also receives energy can settle into a repeating oscillation. In a two-dimensional phase portrait, such an oscillation appears as a closed trajectory. If nearby closed trajectories do not fill a surrounding band, the orbit is called a limit cycle.

How many distinct limit cycles can a polynomial differential equation in the plane possess? The question is part of Hilbert’s sixteenth problem. Polynomial formulas of small degree look simple, yet their trajectories need not have simple shapes or simple dependence on the coefficients.

Family 143 contains two different claims. The first is a uniform finiteness theorem: fixing the polynomial degree fixes some finite upper bound on the number of limit cycles, regardless of the coefficients. The second gives an exact answer for a special family of oscillators: classical Liénard systems with a primitive of degree at most five have at most two limit cycles, and two can occur.

These statements have different levels of precision. The general theorem does not provide an effective numerical bound. The restricted theorem supplies the sharp number two.

2. Background and essential definitions

Polynomial vector fields and isolated cycles

Consider

$$\dot x=P(x,y),\qquad \dot y=Q(x,y),$$

where the dot denotes differentiation with respect to time and $P,Q$ are real polynomials. The degree of the vector field is the larger of their total degrees. Its coefficients may be arbitrary real numbers.

A periodic solution is a nonconstant motion that repeats after some positive time. Its geometric image is the closed curve traced by the motion. A limit cycle is a periodic orbit with a neighborhood containing no other periodic orbit. We count that image once, even if the same motion is parametrized with a different starting time.

Isolation is essential. The harmonic oscillator $\dot x=y,\ \dot y=-x$ has circles of periodic trajectories around the origin, but none is a limit cycle: every circle has other periodic circles arbitrarily close to it. Such a region is called a center.

An attracting cycle draws nearby trajectories toward it, while a repelling one pushes them away. The statements in this family also include isolated cycles with more delicate stability behavior; they are not restricted to attractors.

A return map turns an orbit into a zero

Place a short interval across a periodic trajectory, transversely to the direction of motion. Starting at coordinate $r$ on the interval, follow the trajectory until it returns. Its new coordinate defines the Poincaré return map $\Pi(r)$.

The original orbit is a fixed point, and nearby cycles correspond to isolated zeros of

$$D(r)=\Pi(r)-r.$$

A simple zero satisfies $D'(r)\ne0$, or equivalently $\Pi'(r)\ne1$; the corresponding cycle is hyperbolic. A multiple zero may still represent an isolated cycle and must not be discarded when counting all cycles.

The special Liénard class

The companion paper treats

$$\dot x=y-F(x),\qquad \dot y=-x.$$

Eliminating $y$ gives the scalar oscillator equation

$$x''+F'(x)x'+x=0.$$

Thus $F'$ is the damping polynomial, and the restoring force is exactly $x$. A quintic primitive means $\deg F\leq5$, so the damping has degree at most four. This convention matters when comparing statements phrased in terms of the damping rather than its primitive.

3. Development history and previous partial results

Hilbert’s 1900 problem asked about both the number and the arrangement of limit cycles of polynomial planar systems. A basic distinction soon became central: finiteness for each individual system is weaker than a bound uniform over all systems of a given degree. Every member of a collection can have a finite count while the collection of counts remains unbounded.

Dulac announced an individual-finiteness argument in 1923. After a gap was identified, Écalle and Ilyashenko developed proofs using sophisticated asymptotic and complex analysis of return maps. M1 also discusses Yeung’s later objection to one coefficient-closure mechanism in Ilyashenko’s monograph. That discussion concerns a proof mechanism, not an example with infinitely many cycles. The new manuscript states that its own uniformity argument does not assume individual finiteness as an input.

Important partial results controlled particular degenerations. Bautin obtained the sharp bound of three cycles bifurcating from a quadratic focus or center under quadratic perturbations. Ilyashenko and Yakovenko proved finite cyclicity for elementary polycycles in generic parameter families; Kaloshin subsequently supplied quantitative bounds. A polycycle is a closed chain of trajectories joining equilibrium points. The elementary and genericity assumptions limit the singular behavior these theorems address.

Binyamini, Novikov, and Yakovenko established effective bounds in the infinitesimal Hilbert problem, involving zeros of Abelian integrals and suitable perturbations of Hamiltonian ovals. This is a major restricted setting, rather than the unrestricted polynomial problem.

For classical Liénard systems, Lins, de Melo, and Pugh proposed the bound $\lfloor(n-1)/2\rfloor$ when the primitive has degree $n$. Rychkov proved the two-cycle statement for odd quintic primitives. Later counterexamples invalidated the proposed formula in degrees six and above. Li and Llibre established the unrestricted quartic one-cycle bound, and Li and Lu treated certain quintic slow–fast limiting configurations. The gap addressed by M2 is the global quintic case with arbitrary even as well as odd coefficients.

4. What the new manuscripts claim

A degree-only bound in the whole plane

M1 states that, for every integer $d\geq1$, there is a finite nonnegative integer $B(d)$ such that

$$\#\{\text{limit cycles of }(P,Q)\}\leq B(d) \qquad\text{whenever }\max(\deg P,\deg Q)\leq d.$$

The bound applies throughout $\mathbb R^2$. There is no restriction on the size of the coefficients, on how far the cycles lie from the origin, or on their stability. Geometric cycles are counted once, without a multiplicity weight.

This is the uniform boundedness assertion in the second part of Hilbert’s sixteenth problem. The manuscript explicitly does not give an effective formula for $B(d)$, the best possible value, or a classification of arrangements.

Exactly two in the quintic Liénard class

M2 states that every classical Liénard system with an arbitrary real polynomial $F$ of degree at most five has at most two limit cycles. It also supplies a construction attaining two, so the maximum within this class is exactly two.

No oddness assumption is made on $F$, and its coefficients need not satisfy prescribed signs. Isolated nonhyperbolic cycles are included. The result concerns the whole phase plane, not only small oscillations near an equilibrium or cycles arising in a particular perturbation limit.

The restoring force remains $x$. Replacing it by a general polynomial would produce a larger class not covered by this theorem. Likewise, the result does not restore the Lins–de Melo–Pugh formula in degrees where earlier counterexamples already contradict it.

5. Why the advance matters

Provisional importance: 5/5. A uniform degree-only bound would settle a central finiteness assertion in one of Hilbert’s classical problems. The difficulty is not merely to study a complicated individual system, but to control every coefficient degeneration without excluding exceptional cases.

The quintic theorem adds a different kind of advance: a sharp global answer for a natural oscillator class. It shows how an exact count can become accessible when the equations retain a particular structure, even though the general degree-only bound remains non-effective.

For applications, the distinction between existence and a usable bound remains important. M1 would establish that polynomial degree limits oscillatory complexity in principle. It would not by itself give an engineer a practical maximum to insert into a computation. M2 is numerically explicit, but only within its stated Liénard class.

6. The main idea

Keep track of several competing scales

A trajectory passing close to an equilibrium may spend a very long time there. Even the elementary saddle

$$\dot x=x,\qquad \dot y=-\lambda y,\qquad \lambda>0,$$

illustrates the problem. Starting at $(r,1)$, it reaches $(1,r^\lambda)$ after time $\log(1/r)$. If $r$ and $\lambda$ both vary, the time and the contraction exponent $\lambda\log(1/r)$ may grow at different rates.

M1 organizes such passages using nested asymptotic expansions on complex domains. It retains exponentially small corrections and distinguishes continuations on opposite sides of a real path. These details matter because an apparently negligible term can become the first nonzero contribution after leading terms cancel.

The claimed analytic machinery is designed to survive differentiation and implicit solving. Those operations arise when one follows a degenerating orbit, so controlling only the original return map would not be enough.

Count matching configurations uniformly

Instead of describing an entire cycle by one unwieldy formula, M1 decomposes it into finitely many local passages and writes equations matching their endpoints. Several choices of cuts can represent the same cycle, so counting every solution would overcount. The argument relates different hyperbolic cycles to different connected components of suitable sets of matching data.

A separate finite-dimensional counting principle then converts finiteness of isolated solutions in auxiliary systems into a uniform bound on those components. The manuscript emphasizes that the auxiliary systems must include the equations created by differentiation, implicit operations, and projection, with their domains preserved.

A bounded-degree geometric decomposition controls how many passage types and crossings are needed. Finally, a small signed rotation handles a prescribed finite collection of isolated nonhyperbolic cycles, while spatial rescaling accounts for cycles arbitrarily far away. These steps connect the local analytic machinery to the whole-plane statement.

Compare the two halves of a Liénard orbit

M2 uses a more specialized geometry. On each half-plane, the change $u=x^2/2$ turns an orbit into an arch satisfying

$$\frac{du}{dy}=\varphi(u)-y, \qquad \varphi(u)=F(\sqrt{2u})\ \text{or}\ F(-\sqrt{2u}).$$

Here $y$ serves as an endpoint coordinate, not physical time. At a fixed height, an arch has two endpoints; their half-distance is its width, and their average records its midpoint.

The even coefficients of a quintic primitive contribute the same quadratic part to both profiles, while the odd coefficients change sign. M2 fits quadratic comparison profiles to the arch data and transports their slope and curvature as the height varies. A model inequality controls how these fitted quantities change with width.

A periodic orbit requires the two half-orbits to match. Their midpoint difference is therefore a scalar matching function whose isolated zeros count cycles. The manuscript claims that the quintic coefficient structure constrains its changes of direction strongly enough to permit at most two isolated zeros. This comparison counts degenerate cycles directly. A separate small-parameter construction establishes that two cycles really occur.

7. Scope, limitations, and open questions

Uniform finiteness does not determine the exact maximum for a general degree, its asymptotic growth, or all possible spatial arrangements. Those questions remain distinct from the existence of $B(d)$.

The general theorem concerns real planar polynomial vector fields. It is not a statement about arbitrary smooth fields, higher-dimensional systems, or nonisolated families of periodic orbits. The harmonic oscillator’s continuum of circles is compatible with the claim because none is isolated.

For the Liénard result, both the degree convention and the exact restoring force are essential. The maximum two applies to primitive degree at most five, without an oddness assumption. Earlier higher-degree counterexamples remain part of the problem’s history.

The manuscripts present mathematical statements and proposed arguments, rather than a numerical survey of trajectories. A picture showing two apparent cycles would illustrate a system but would not replace either a global upper bound or a proof that the depicted orbits are isolated.

8. Sources and further reading

Hilbert’s Mathematical problems gives the original formulation. Bautin’s quadratic bifurcation theorem and Binyamini, Novikov, and Yakovenko’s constructive infinitesimal result illustrate earlier restricted advances. Ilyashenko’s centennial history supplies broader context.

For the Liénard branch, see Lins, de Melo, and Pugh, On Liénard’s equation; De Maesschalck and Dumortier’s higher-degree counterexamples; and Li and Llibre’s quartic uniqueness theorem.

Source manuscripts

• M1. OpenAI, Uniform bounds for planar polynomial limit cycles (September 24, 2026).
• M2. OpenAI, Two limit cycles for quintic Liénard systems (September 24, 2026).

Sources consulted on 21 October 2026 at repository revision 1581d4c. Download the manuscript citations.