Motivation
An explicit solution tells us where one trajectory goes. A dynamical question asks how nearby trajectories behave together. If a state is slightly disturbed, does it remain nearby? Does it return? Does the answer depend on the size or direction of the disturbance? These questions can often be answered without solving the differential equation in closed form.
The distinction between remaining close and returning is essential. The equation keeps every initial state fixed. Its equilibrium at zero is stable because a small disturbance stays small, but the disturbed state never returns to zero. By contrast, both preserves smallness and drives every trajectory to zero. An oscillator provides a two-dimensional version of the first behavior: nearby trajectories remain on nearby closed curves, with no decay of their amplitude.
We study autonomous systems in finite-dimensional real space. The methods proceed from scalar phase lines to linear systems and then to nonlinear energy estimates.
Definitions
Assume is locally Lipschitz on an open set . Local uniqueness gives a maximal solution through each initial state . Write it as for those times for which it exists. Autonomy and uniqueness imply the flow identity whenever both sides are defined. The definition does not assume global existence; a trajectory can reach the boundary of or escape to infinity in finite time.
An equilibrium satisfies . Translating the state lets us place it at zero. All stability definitions below refer to forward time. The norm can be any fixed norm on ; equivalence of finite-dimensional norms preserves the qualitative definitions, although numerical constants change.
Definition 3.1.1
(Stability, attraction, and asymptotic stability)
The equilibrium is Lyapunov stable if for every there exists such that implies the solution exists for all and for all . It is locally attractive if some neighborhood of zero consists of initial states whose forward solutions exist and tend to zero. It is locally asymptotically stable if it is both stable and locally attractive.
The order of the quantifiers in stability requires one initial tolerance to work for every future time. Continuous dependence on a fixed finite interval does not alone establish this. An equilibrium is exponentially stable if, for some , all satisfy for . Global asymptotic stability means stability together with attraction from every initial state in the stated state space, including existence for all forward times.
A continuously differentiable function has orbital derivative . This is the ordinary chain rule along a trajectory. A function is positive definite near zero if and at every other sufficiently near point. Negative definiteness of means away from zero. These terms describe functions, not only quadratic forms.
For a global argument, a continuous positive definite function is called proper here when every sublevel set is compact. Equivalently, as : a failure of this limit would give an unbounded sequence in one sublevel, whereas the limit makes each closed sublevel bounded. Properness says that bounded energy confines the state to a bounded region. It does not say that the differential equation has a solution there; existence and continuation also require regularity of the vector field throughout that region.
Theorem/Proposition
Theorem 3.1.2
(Spectral tests for linear systems)
For , zero is asymptotically stable if and only if every eigenvalue of has strictly negative real part. In that case it is exponentially stable. Zero is stable if and only if all eigenvalues have nonpositive real part and every eigenvalue on the imaginary axis is semisimple, meaning that its Jordan blocks have size one. An eigenvalue with positive real part makes zero unstable.
Concept lensAlgebraic
The Lyapunov equation as a test for decay
For a real square matrix , the condition that every eigenvalue has strictly negative real part is equivalent to an algebraic one: the matrix equation has a symmetric positive definite solution . One direction is the integral formula for constructed in the linearization proof below. Conversely, if such a exists, then satisfies along . Hence , and with it , decays exponentially, and the spectral test excludes every eigenvalue with nonnegative real part. The spectral description asks for eigenvalues; the algebraic one asks for the solution of a linear matrix equation and a positivity check, and it supplies the energy that the nonlinear argument uses.
Theorem 3.1.3
(Lyapunov's direct method)
Suppose is locally Lipschitz near zero, , and is continuously differentiable and positive definite on a neighborhood of zero. If there, zero is stable. If away from zero there, zero is locally asymptotically stable.
For the global conclusion additionally assume that is locally Lipschitz on all of , that is positive definite and proper there, and that for every . Then zero is globally asymptotically stable: every forward solution exists for all and tends to zero, and the equilibrium is stable.
Locally Lipschitz on all of means that a local Lipschitz estimate is available near every state, not merely near zero. It does not require a single global Lipschitz constant. For instance, the polynomial vector field satisfies the stated local requirement everywhere even though its derivative is unbounded. Local existence at every state still does not by itself exclude finite-time escape; that is the role of the proper energy in this theorem.
Theorem 3.1.4
(Linearization at a strictly decaying equilibrium)
Let be continuously differentiable near zero, with , and let . If every eigenvalue of has strictly negative real part, the nonlinear equilibrium is locally exponentially stable. Eigenvalues with zero real part leave the general nonlinear conclusion undecided by this criterion.
Proof sketch or proof idea
Proof
(Jordan blocks and the spectral tests)
For the linear theorem, a Jordan block contributes
If , exponential decay dominates each polynomial factor. Since there are finitely many blocks, for suitable positive constants. If a positive real part occurs, a real invariant direction or plane supplies arbitrarily small initial states whose trajectories leave a fixed neighborhood. On the imaginary axis, a nontrivial Jordan block produces unbounded polynomial growth; a block of size one produces bounded rotations or constants. These observations give both directions of the spectral tests. Nondecaying modes prevent attraction even when they remain bounded.
Proof
(Energy barriers, continuation, and attraction)
For the direct method, choose a closed ball of radius inside the neighborhood on which the assumptions hold. Positivity and compactness give . Continuity at zero permits a smaller ball on which . Along a trajectory cannot increase, so a state beginning in that smaller ball cannot reach the sphere of radius . Choosing the ball strictly inside the domain also traps the solution in a compact subset, and the continuation theorem gives existence for all forward time. This proves stability.
For attraction under strict decrease, keep a trajectory in a smaller compact ball. For any sufficiently small , stability supplies such that entering guarantees remaining in . If the trajectory never enters the smaller ball, compactness of the annulus gives there. This would force eventually negative. Therefore the trajectory enters and remains inside every prescribed small neighborhood, proving convergence to zero.
For the global conclusion, fix any initial state . Local Lipschitz regularity throughout the state space gives its unique maximal solution. Energy nonincrease traps it in the compact set . If its maximal forward time were finite, continuity of on would bound , making Cauchy as . Its limit belongs to , and local existence at that limit extends the solution beyond , a contradiction. Thus the solution exists for every forward time.
Stability follows from the local sphere argument already proved. To show attraction from the arbitrary , choose and its stability radius . If the trajectory never enters that smaller ball, it stays in the compact set . Continuity and strict negativity give a uniform bound on this set, contradicting after sufficiently long time. It must therefore enter the smaller ball; autonomy and stability keep it within thereafter. Since was arbitrary, the trajectory tends to zero. The order of the argument matters: first obtain continuation, then use an arbitrarily long-time energy contradiction.
Proof X-ray
Where each hypothesis enters the direct method
Positive definiteness is used exactly once, to make the sphere minimum positive. For that minimum is zero and the barrier disappears. Nonincrease of then makes the set where inside the -ball a trap, which gives stability and, because the trap is compact, continuation. Attraction needs a second, separate move. On the compact annulus between and the outer radius, the continuous function is strictly negative, so it attains a maximum . A trajectory staying there up to time would satisfy , which is negative for large . In the global version properness does the work of the small ball: it makes compact, first to exclude finite-time escape and then to supply the same uniform bound . Without compactness, could tend to zero along an escaping trajectory and no contradiction would follow.
The linearization theorem reduces to the direct method once a suitable quadratic energy is found.
Proof
(A quadratic Lyapunov function from the linear part)
For nonlinear linearization, write with . The decaying linear system supplies the symmetric positive definite matrix
Convergence follows from exponential decay. For nonzero , . Differentiating the integrand and integrating from zero to infinity proves the matrix identity. With ,
Near zero, , so . If , then . Integration gives while the trajectory stays in the neighborhood; the invariant-sublevel argument ensures it does. Taking square roots yields an exponential estimate for . This proof explains precisely why terms smaller than linear cannot overcome a strictly decaying linear part near the equilibrium.
Worked examples
Worked example 3.1.5
(A phase line with two different equilibrium types)
For , equilibria are and . The right hand side is negative on , positive on , and negative on . Thus arrows point away from zero and toward one on both sides near one. Uniqueness prevents a nonconstant trajectory from crossing an equilibrium: if it reached the equilibrium, uniqueness through that meeting point would force it to coincide with the constant solution.
An initial state in therefore increases and stays below one. It has a limit . If , the continuous positive rate near would keep it increasing at a fixed positive speed, a contradiction. Hence . Initial states just above one decrease to one by the same argument. This establishes local attraction; the inward arrows also establish stability. States below zero instead escape to in finite forward time, so attraction of one is not global on .
Worked example 3.1.6
(A one-sided attracting equilibrium)
Consider . The equilibria are and . The right hand side is positive on both and , and negative on . Thus the arrows point right on both sides of : they point toward from the left but away from it on the right. This is called one-sided attraction, or semistability; it is not Lyapunov stability.
Let . The vector field is locally Lipschitz, so uniqueness prevents the trajectory from crossing or reaching an equilibrium at a finite time unless it is constant. It therefore increases within . Because it remains in this compact interval, the continuation theorem gives existence for all . It has a limit . If , continuity would make eventually bounded below by a positive constant, contradicting convergence. Hence .
If , the same argument confines the increasing trajectory to and gives . If , it decreases within and also tends to . Thus is asymptotically stable. To see why is unstable, fix . Every sufficiently small positive initial value tends to , so it eventually leaves the -neighborhood, however small it was initially. Attraction from the left cannot satisfy a definition requiring all nearby initial states to remain close.
Quick check
(For , find the sign of the right-hand side on the three intervals cut by and . Which equilibrium attracts from only one side? Is it stable?)
Use uniqueness to prevent crossing an equilibrium, and use a bounded monotone trajectory to identify its limit.
Solution · Answer
The signs are positive on and negative on and . Hence attracts from both sides and is asymptotically stable. Zero attracts from the right and repels from the left: for example, each initial state in decreases to . Therefore zero is unstable, despite attracting positive initial states. Boundedness between the initial point and the relevant equilibrium justifies continuation and the limit argument as above.
Worked example 3.1.7
(A center is stable without attraction)
For , , let . Then . Every trajectory remains on its initial circle, so zero is stable: choose . A nonzero trajectory never approaches zero because its norm is constant. The eigenvalues are and , both semisimple. Bounded imaginary-axis modes agree with stability, while their lack of decay prevents asymptotic stability.
By contrast, , has only the eigenvalue zero but a nontrivial Jordan block. Its solutions are , so arbitrarily small nonzero causes eventual departure from any fixed ball. The sign of real parts alone is insufficient when a zero real part occurs.
Counterexample mode
The same linearization with three nonlinear answers
The scalar equations , , and all have derivative zero at the origin. The first has
so zero is globally asymptotically stable. The convergence is slower than exponential. For the second, the analogous denominator is , and every nonzero initial state escapes in finite forward time. Zero is unstable. In the third equation every trajectory is constant, giving stability without attraction. A zero linearization can therefore accompany each of these different outcomes.
Worked example 3.1.8
(A nonlinear energy with strict decay)
Consider , . With ,
In the region , the final term is at most , so . A sufficiently small circular sublevel set lies inside that region and is forward invariant. Thus there, proving local exponential stability. The calculation states the region in which the inequality holds; it does not extrapolate that particular bound to all states. The Jacobian is , so the linearization theorem gives the same local conclusion through a separate argument.
Worked example 3.1.9
(A global decay estimate with continuation justified)
For , choose . The vector field is locally Lipschitz everywhere, and is globally positive definite and proper. Its orbital derivative satisfies
The global theorem first guarantees forward existence for every initial value. Multiplying the differential inequality by then gives , hence for all . This example gives global exponential decay, whereas a merely strict negative derivative can give slower decay, as demonstrates.
Common mistakes
Common mistake
(Nonincrease does not give attraction)
An inequality proves stability under positive definiteness, but it does not alone prove attraction. The oscillator has identically zero orbital derivative and no attraction. Even with damping, a zero-derivative set may contain points other than the equilibrium; concluding convergence from that situation requires an invariance argument with additional hypotheses. Strict negative definiteness in the theorem avoids this extra step.
Common mistake
(A local minimum is not positive definiteness)
Likewise, a function that merely has a local minimum need not be strictly positive away from the equilibrium. For example, vanishes along a line and cannot control the distance in the direction. The sphere-minimum step in the proof explains the role of positive definiteness. Compact sublevels are also necessary in the stated global argument to prevent a trajectory from escaping to infinity while its energy remains bounded.
Common mistake
(Regularity near zero is not regularity everywhere)
Nor can regularity near zero replace regularity at distant states. Set for , but , and take . This field is locally Lipschitz near zero and has at every nonzero state, with proper . Yet no classical solution starts at . Such a solution would have , so for small positive time; there forces , whose right derivative at zero is , a contradiction. The algebraic energy sign does not repair the missing existence hypothesis at the discontinuity.
Do not read a plotted trajectory as a stability proof. A picture follows finitely many initial states for finite time, whereas stability quantifies over every sufficiently close state and every future time. The phase line, spectral argument, or energy estimate supplies the missing quantifiers.
Summary
Stability controls all future departures; attraction concerns the eventual limit. Strictly negative spectral real parts produce exponential decay for a linear system and survive small nonlinear terms near an equilibrium. Imaginary-axis modes require care, and zero real parts can leave nonlinear behavior undecided. A positive definite Lyapunov function turns inequalities along trajectories into geometric confinement and, under strict decrease, convergence.
Exercises
Quick check
(Classify zero for , . State separately whether it is stable, attractive, or exponentially stable.)
Examine both coordinates.
Solution · Answer
The solution is . Its Euclidean norm never exceeds its initial norm, so zero is stable. A nonzero does not decay, so zero is not attractive and therefore not asymptotically or exponentially stable. One equilibrium can be stable even though every neighborhood contains other equilibria.
Quick check
(Prove global asymptotic stability of zero for using a Lyapunov function, and show why no uniform local exponential estimate is possible.)
Use and the explicit solution.
Solution · Answer
The polynomial field is locally Lipschitz on all of . The function is globally positive definite, has compact sublevels, and satisfies for . The global direct method gives forward existence and global asymptotic stability. For any fixed nonzero small initial value, the ratio decays polynomially. For every , tends to infinity, contradicting the bound required for exponential stability.
Quick check
(Classify the equilibria of by a phase line.)
Record the signs on all four intervals separated by the equilibria.
Solution · Answer
The equilibria are . The signs of are positive on , negative on , positive on , and negative on . Arrows point toward and from both sides, making each locally asymptotically stable. They point away from zero, making it unstable. Monotonicity and trapping between neighboring equilibria justify the local limits.
Quick check
(For , , show local exponential stability with and give one explicit region where your derivative bound holds.)
Bound by a constant times when both coordinates are small.
Solution · Answer
Here . On and , . Hence . Any sufficiently small circular sublevel contained in this square is invariant. Integrating and taking square roots gives for those initial states, proving local exponential stability.
Quick check
(For and , use to prove global asymptotic stability and obtain an exponential bound for the Euclidean norm.)
Check the vector field on the whole state space and explain why the decay estimate holds for every forward time.
Solution
The polynomial vector field is locally Lipschitz on all of . The function is , globally positive definite, and proper. Along solutions,
so it is strictly negative away from zero. The global theorem supplies continuation and global asymptotic stability. Integrating the stronger inequality gives , and taking square roots yields for every .