A Lyapunov function often measures an energy. Friction makes the energy decrease, but it may stop decreasing at an instant without the motion having stopped. For a damped oscillator, the derivative of energy vanishes whenever velocity is zero, even when the restoring force immediately starts the motion again. A condition requiring strict decrease at every non-equilibrium point would miss this useful example.
LaSalle's principle asks a sharper question: which trajectories can remain forever in the set where the energy derivative vanishes? A compact trapping region guarantees that the long-time behavior has somewhere to accumulate. Invariance then excludes points that can only be visited momentarily.
The setting
Definition 3.4.1
(Positive invariance and -limit sets)
Let be open, and let be locally Lipschitz. Write for the maximal solution of from at time zero. A compact set is positively invariant if every solution starting in exists for all and remains in .
For , its -limit set is
A set is invariant here if, for every , its solution exists for all and remains in . This is stronger than positive invariance. For a set , write .
The local Lipschitz assumption provides uniqueness and continuous dependence. Compact positive invariance is an assumption about the actual forward trajectories; a bounded formula for alone does not establish it.
Theorem 3.4.2
(LaSalle's invariance principle on a compact set)
Suppose is nonempty, compact and positively invariant, and satisfies
Let , and let be the union of all invariant subsets of . Then is nonempty and compact, and every satisfies
This is convergence to a set. It does not by itself imply convergence to one point, nor does it assert that every point of lies in .
Why limit points must stay in the zero-derivative set
Proof
(Compactness supplies a complete invariant limit set)
Fix . Since is compact, every sequence has a subsequence for which converges. Thus is nonempty. It is closed: a diagonal choice of times tending to infinity approximates a convergent sequence of its points. Hence is compact.
We next justify invariance in both time directions. Fix and . Choose with . Continuous dependence for time gives . To go backwards, compactness provides a subsequence with ; continuous dependence gives . Uniqueness identifies with the solution through at time . Write this point as . If , uniqueness gives , so the backward pieces fit together to define a complete solution in . Thus is invariant.
Finally, . Otherwise there are times tending to infinity whose distance is bounded below by some positive number; a convergent subsequence in would have a limit in , a contradiction.
Concept lensAnalytic
The limit set as an intersection of orbit tails
Equivalently,
Each tail closure is a nonempty compact subset of , and the closures decrease as grows, so their intersection is nonempty and compact by the finite intersection property. Each tail closure is also connected, being the closure of a continuous image of an interval; a decreasing intersection of compact connected sets is connected. Thus is in addition connected, so a single trajectory can never have exactly two isolated equilibria as its limit set.
Proof
(The energy becomes constant on the limit set)
Along the forward orbit, is nonincreasing and bounded below on compact , so it tends to a finite limit . By continuity, for every . Since the full orbit through stays in , for every . Differentiating at zero gives . Therefore is an invariant subset of and lies in .
To see that itself is compact, use the equivalent description
It is invariant by uniqueness. It is closed. If in , then because is closed. While the backward solution from exists, continuous dependence gives for each fixed negative , so . A finite backward maximal time is impossible: the orbit would remain in the compact set , contradicting the continuation theorem applied to . Thus the backward solution exists for all time. The same argument for positive time, or positive invariance of together with continuous dependence, shows that it stays in . Thus is a closed subset of compact . It is nonempty because it contains an -limit set. Finally,
Proof X-ray
Where each hypothesis enters
The key move is the first paragraph of the second proof. equals on all of , and contains the complete orbit of each of its points, so is constant and its derivative at , which is , vanishes. Both ingredients are needed: knowing the value at one point says nothing about , and invariance is what turns a statement about values into a statement about the derivative. Compactness of is used three times: to make nonempty, to bound below so that is finite, and to exclude finite-time escape, including in the backward continuation that builds complete orbits. Uniqueness is used only to glue the backward pieces into one orbit. The sign is used once, to make monotone. For below, the monotone limit exists (), but is empty, so there is no point at which to differentiate.
A useful local stability consequence
Theorem 3.4.3
(A local asymptotic stability test)
Let be locally Lipschitz near an equilibrium with , and let there satisfy , for , and . Suppose some sufficiently small closed ball lies in the domain and the largest subset of in that ball whose full trajectories remain there is . Then is asymptotically stable.
Proof
(Build a compact trapping sublevel)
Choose a closed ball on which the hypotheses hold. The positive minimum exists. For , the compact set is positively invariant: at a first time reaching the value of would be at least , contradicting monotonicity. It cannot escape in finite time while it stays in compact . Its largest invariant subset of is . LaSalle gives convergence to zero for initial data in .
For stability, fix any sufficiently small and choose . Continuity at zero supplies with and on . The first-exit argument traps these trajectories in . Combining stability and convergence for all sufficiently small initial data proves asymptotic stability.
Two examples of zero energy loss
Worked example 3.4.4
(A damped oscillator)
Consider
Setting gives the usual damped oscillator , . Here , and every closed disk centered at zero is positively invariant and compact. The set is the horizontal diameter of the chosen disk. A complete trajectory in must have for all , so ; hence the largest invariant subset is the origin. LaSalle gives for every initial condition, since each initial point lies in some invariant disk.
Notice why at does not make that point an equilibrium: if , then and the trajectory immediately leaves . The vector field, not the derivative's value at a single point, determines invariance.
Worked example 3.4.5
(A pendulum in a potential well)
For , let , and
Choose and the compact region . Energy nonincrease keeps . A first exit through is impossible because there ; the energy inequality also bounds , so compact continuation gives global existence and is positively invariant. In , an orbit can stay only where . The energy restriction excludes , leaving only . Every trajectory starting in converges to the downward equilibrium. This is a local result in one potential well, not a claim that every real-angle initial condition approaches the same representative.
Worked example 3.4.6
(A limit set that keeps moving)
Consider
and write . Then and , so in polar coordinates and . Take . Since ,
The annulus is compact and positively invariant, because on and on . In the set is the unit circle (the other zero, , lies outside ), and the unit circle is itself invariant: it is the periodic orbit . Hence is the whole circle and LaSalle gives . No trajectory starting off the circle converges to a point, because keeps it rotating. Here the conclusion that trajectories approach is exactly the right statement; contains no equilibrium at all.
Why the hypotheses cannot be dropped
Counterexample mode
Nonincrease alone does not imply attraction
For , , the positive definite function has everywhere. Nonzero trajectories travel on circles forever. The origin is stable but not attractive: the largest invariant subset of is the entire disk, not just the origin. Positive definiteness and nonincrease prove stability here, not convergence.
The first example keeps compactness but loses strict decrease; the next keeps strict decrease but loses compactness.
Counterexample mode
A decreasing function without compact trapping
On the real line, take and . Then , but
There is no compact forward trapping set containing this orbit and no -limit point in . Strict decrease alone cannot substitute for compactness. This example has no equilibrium; it demonstrates failure of the compactness step, not a failure of the local equilibrium test.
Common mistake
(Convergence to a set is not convergence to a point)
A set-valued conclusion can be the best available one. For , with on a compact invariant disk, is the vertical diameter and every point of it is an equilibrium. Each trajectory approaches that diameter, but the limit point depends on its initial -coordinate. In other systems a limit set can contain continuing motion, so distance to must not be silently replaced by convergence to a single equilibrium.
Common mistake
(Stopping at the zero-derivative set)
LaSalle's conclusion is about , not about . In the damped oscillator, is the whole horizontal diameter, but trajectories only pass through its nonzero points. Concluding that solutions approach some point of , or merely that they approach , discards the information that makes the principle useful. The step that cannot be skipped is the invariance check: substitute for all time into the equation, as in , and keep only the points whose complete orbits stay in .
Exercises
Quick check
(For , , use to prove global convergence to the origin.)
Solution · Answer
, which vanishes only at zero. Each closed disk is compact and positively invariant, so LaSalle gives convergence from every initial point. Stability follows from the same disks; the origin is globally asymptotically stable.
Quick check
(For , , find for and determine the largest invariant subset of . Classify the origin.)
Solution · Answer
. On , remaining there requires , so the largest invariant subset is the origin. Closed disks are compact and positively invariant; LaSalle and the Lyapunov stability estimate give global asymptotic stability.
Quick check
(For , with on , find , , and the limit of the solution from .)
Solution · Answer
, and every point of is an equilibrium, so . The exact solution is and converges to , not necessarily to the origin.
Quick check
(In the pendulum example, explain why the region defined only by on the entire plane is not compact, even for . Why is restricting legitimate for initial data in the displayed ?)
Solution · Answer
The points have zero energy for all integers , hence the full sublevel is unbounded. For the specified , the energy at either boundary is at least two; a trajectory with initial energy at most cannot reach that boundary. Its velocity is bounded by the same energy inequality, giving compact trapping.