Motivation
A stable equilibrium is not the only possible long-time behavior of a planar differential equation. A trajectory may instead return to its starting point and repeat forever. Such a periodic orbit cannot be seen by looking only at the eigenvalues of one equilibrium; it is a geometric object extending across a region of the plane.
Sometimes it is possible to rule out all periodic orbits without solving the system. The idea is to compare two descriptions of the flux across a hypothetical closed trajectory. The vector field is tangent to the trajectory, so its normal flux is zero. Green's theorem turns that zero flux into the integral of a divergence over the enclosed region. If the divergence has a strict sign throughout that region, the two conclusions conflict.
A multiplier can reveal a sign that the original divergence does not have. However, its domain matters. A periodic curve in a region with a hole can enclose points outside that region, where the multiplier may be undefined. We will prove a criterion with explicit hypotheses and then construct a periodic orbit showing why the topology cannot be omitted.
Definitions
Definition 3.3.1
(Planar flow and periodic orbit)
Let be open and let be . A solution of the autonomous system is a differentiable curve satisfying
A nonconstant solution is periodic if some satisfies for all . Its image is a periodic orbit. An equilibrium gives a constant periodic solution for every , but is not called a periodic orbit here.
Because is locally Lipschitz, solutions meeting at the same time agree on their common time interval. Autonomy supplies the shifted version: if , then wherever both sides are defined. Thus a nonconstant periodic solution cannot pass through an equilibrium, and a return to an earlier point makes the intervening time a period.
A nonconstant continuous periodic curve has a least positive period: its periods form a closed subgroup of ; if positive periods accumulated at zero, their integer multiples would be dense, and continuity would force the curve to be constant. Hence the subgroup is discrete and has a smallest positive member. Shifted uniqueness then makes the curve injective on one half-open least-period interval. Its speed never vanishes, and its derivative is periodic, so there is no corner at the joining point. The trace is a regular simple closed curve, enclosing a bounded interior by the Jordan curve theorem.
Definition 3.3.2
(Divergence and a Dulac multiplier)
For a vector field , its divergence is . A Dulac multiplier is a scalar function chosen on the domain so that has a useful sign. The vector field is tangent to every solution curve of , even where is negative or zero.
The multiplier is a proof device; it need not be a conserved quantity or a Lyapunov function. In this note we will not divide by , so the exclusion theorem does not require to be nonzero. What matters is that and are throughout the region on which Green's theorem is applied.
A criterion for excluding closed trajectories
Theorem 3.3.3
(Bendixson–Dulac exclusion)
Let be a connected, simply connected open set, , and . If is strictly positive everywhere on , or strictly negative everywhere on , then has no nonconstant periodic orbit contained in .
Taking gives the Bendixson criterion based on the divergence of itself.
We use the following standard Jordan-interior lemma: if a Jordan curve lies in a connected simply connected open set , its bounded interior lies in . Indeed, an omitted interior point would have winding number with respect to the curve, whereas a contraction of that curve inside would preserve its winding number about the omitted point and give zero. This uses the Jordan curve and winding-number theorems.
The conclusion excludes periodic orbits; it does not by itself prove that every trajectory tends to an equilibrium, remains bounded, or exists for all time. A system can have no periodic orbit and still have trajectories that escape or approach the boundary of its domain.
Proof
(Zero boundary flux versus nonzero area integral)
Suppose a nonconstant periodic orbit lies in . By uniqueness and regularity it is a simple closed curve. Its bounded interior lies in by the Jordan-interior lemma above. In particular the closure of is a compact subset of .
At each boundary point is tangent to . Multiplication by preserves tangency, including where . Thus on , where is the outward unit normal. Since is on an open neighborhood of the compact closure of , Green's theorem applies. Orient positively (counterclockwise) and take the outward normal; the flux integral itself is independent of traversal direction. The flux form gives
The interior has positive area. A continuous function strictly positive everywhere on it has strictly positive integral; a strictly negative one has strictly negative integral. Either case contradicts zero flux. No assumption about the sign of itself was used.
Proof X-ray
Three distinct roles in the argument
ODE uniqueness is used first, and only to make the periodic trace a simple closed curve, so that it has an interior at all. Simple connectedness is used once, to place that whole interior inside , where and are . This is the step that fails in the punctured-plane example below, whose cycle encloses the point where is undefined. Tangency of to its own orbit supplies the zero boundary flux, and multiplication by cannot destroy it. Green's theorem converts that zero into an area integral, and the strict sign of together with the positive area of closes the contradiction. The sign of itself is never used.
Proposition 3.3.4
(A non-strict divergence variant)
Under the same domain and regularity hypotheses, suppose is nonnegative everywhere, or nonpositive everywhere, and is not identically zero on any nonempty open subset of . Then no nonconstant periodic orbit is contained in .
Proof
(The same flux identity on an unknown interior)
Indeed, its integral over the proposed orbit interior is nonzero: that interior is a nonempty open subset, and continuity supplies a small disk there on which the sign is strict. The condition is imposed on all open subsets because the orbit interior is unknown in advance. Merely finding one point of strict sign somewhere in is insufficient if a proposed orbit could enclose only a region where the divergence vanishes.
Concept lensGeometric
Divergence as the rate of area change
For the flux argument has a dynamical reading. If is the interior of a periodic orbit , trajectories cannot cross , so the flow carries onto itself and its area is constant in time. The area transport formula
then forces the integral of over to vanish. A flow with strictly negative divergence shrinks every region of positive area, so it cannot keep the interior of a cycle intact.
An explicit periodic orbit
Worked example 3.3.5
(A radial equation determines a stable cycle)
Consider the polynomial system
For , direct calculation gives
At , the radius is constant and the angle advances at unit speed. Thus every point on the unit circle traces the same periodic orbit with least period . To see what nearby trajectories do, set . Then , with solution
The denominator stays positive for forward time, and . The nonzero trajectories therefore approach the unit circle in radial distance, while the origin is an equilibrium. The divergence is , which changes sign on the plane, so the strict-sign criterion does not apply there. We have proved a periodic orbit directly; we have not inferred one merely from a failure of the criterion.
Worked example 3.3.6
(Strict negative divergence in a damped system)
For
we have on all of . The plane is simply connected, so rules out nonconstant periodic orbits. The calculation does not, on its own, establish convergence to the origin. For example, the energy has , which vanishes along the line ; strict decrease at every nonzero point is not available from this particular .
Worked example 3.3.7
(A multiplier in the positive quadrant)
Take
The quadrant is open and simply connected. With ,
The criterion excludes periodic orbits contained in . This example also has no equilibrium in : and force , then . The multiplier is singular on the axes, but those axes are outside ; there is no need to extend across them. For solutions starting in , the equations and also preserve positivity on their interval of existence, because each variable is its initial value times an exponential of an integral. This invariance does not itself assert global existence.
Why a hole matters
Counterexample mode
Strict divergence on a punctured plane need not exclude a cycle
On the punctured plane , write and let
Both and are on . In polar coordinates the equations are and , so the unit circle is a periodic orbit. Yet a direct calculation gives
For instance, the radial component of is , and the polar divergence formula gives ; its angular component has no angular dependence. There is no contradiction: the unit circle encloses the omitted origin, where is undefined, and is not simply connected. Green's theorem cannot be applied to its entire interior with this multiplier. For , on the annulus , the divergence integral is . The outer flux is zero, while the outward normal on the inner boundary points toward the origin and gives flux . That inner flux accounts exactly for the apparent contradiction.
Common mistakes
Common mistake
(Treating a sufficient condition as a converse)
Failure to find a multiplier with strict-sign divergence does not prove that a periodic orbit exists. Conversely, strict negative divergence excludes periodic orbits only when the required domain and regularity conditions hold. The radial polynomial example has a cycle and a sign-changing divergence; the punctured-plane example has strict negative divergence for a multiplier but fails the topology hypothesis.
Common mistake
(Equilibria are not nonconstant periodic orbits)
An equilibrium is a constant solution and therefore formally repeats after any positive time. The theorem excludes nonconstant periodic orbits, not equilibria. In the proof, a nonconstant periodic solution cannot encounter an equilibrium because uniqueness would force it to remain there.
Common mistake
(Excluding cycles is not proving stability)
The criterion also cannot establish stability of an equilibrium by itself. For example, , has strictly positive divergence and no nonconstant periodic orbit, but the origin is unstable and nonzero trajectories escape to infinity. The sign of divergence answers a specific geometric question; it is not a replacement for a Lyapunov or continuation argument.
Summary
A regular closed trajectory has zero normal flux. On a simply connected domain, Green's theorem turns this into zero integral of over its interior, contradicting a strict sign. A Dulac multiplier can expose such a sign on a domain where it is smooth. The hypothesis that the interior lies in that domain is indispensable, and exclusion of periodic orbits is not a full classification of long-time behavior.
Exercises
Quick check
(Show that , has no nonconstant periodic orbit in .)
Solution · Answer
The divergence is . The vector field is on the simply connected plane, so the criterion with applies.
Quick check
(On , consider , . Use to exclude periodic orbits contained in .)
Solution · Answer
On , , so . Both and are there, and the positive quadrant is simply connected. Thus no nonconstant periodic orbit lies wholly in .
Quick check
(For , with , apply the criterion. What changes when ?)
Solution · Answer
The divergence is on the plane when , excluding nonconstant periodic orbits. When the divergence is zero, so the strict-sign criterion no longer applies. Indeed , is periodic whenever .
Quick check
(Suppose a function has at every point of a nonconstant periodic orbit. Prove directly that this is impossible.)
Solution · Answer
Along a periodic solution of period , the chain rule gives . Integrating over one period would give , but . This contradiction uses the strict sign on the orbit and does not require a divergence calculation.