An equilibrium solves , where is a parameter. As the parameter changes, equilibria may move, meet, appear or disappear. A picture of solution curves can suggest what happens, but it does not by itself distinguish a genuine branch from an algebraic artifact, or determine stability.
For a scalar autonomous equation, the sign of the vector field on either side of an equilibrium determines local motion. The implicit function theorem controls the nearby equilibrium set. Together they give a precise saddle-node theorem and explain which hypotheses separate it from other bifurcations.
Equilibria and scalar stability
Consider
where is real-valued and on a neighborhood of . The parameter is fixed along each solution. An equilibrium at parameter is a point with . The definitions of stability and attraction are relative to this fixed autonomous equation; they do not say that itself evolves.
Definition 3.5.1
(Hyperbolic equilibrium and bifurcation point)
An equilibrium at parameter is hyperbolic if . A point with is a bifurcation point of equilibria if there is a neighborhood of such that, for every neighborhood of and every , the number of equilibria in is not constant for .
A hyperbolic equilibrium is never a bifurcation point. Since , the implicit function theorem gives a rectangle around in which the equilibria form a single continuous curve of the parameter . Every small neighborhood of therefore contains exactly one equilibrium for all parameters near . A change in the local count requires , which is the first hypothesis of the theorem below.
Theorem 3.5.2
(The scalar hyperbolic stability test)
If and , then is locally exponentially asymptotically stable. If , then is unstable. When , this test alone is inconclusive.
Proof
(The vector field points toward or away from the equilibrium)
Write and . If , differentiability supplies and a small interval with . Along a solution in that interval,
Consequently by the elementary integrating-factor inequality. This bound also prevents a trajectory starting sufficiently near zero from leaving the interval; local existence and compact continuation then give existence for all positive time and exponential convergence.
If , choose a small interval with for . A solution starting at any increases at least exponentially as long as it stays in that interval, and therefore reaches its outer endpoint in finite time. Arbitrarily small perturbations leave a fixed neighborhood, proving instability.
A nondegenerate saddle-node
Theorem 3.5.3
(Local saddle-node theorem)
Suppose
There are a fixed neighborhood of and such that, for , there are exactly two equilibria in that neighborhood on the side where
and none on the other side. At there is exactly one equilibrium in that same neighborhood. On the two-branch side, writing , the equilibria satisfy
One branch is locally exponentially stable and the other unstable. More precisely, for sufficiently small , the branch for which is stable; the other is unstable.
Proof
(Solve for the parameter instead of the state)
Because , the implicit function theorem writes every nearby equilibrium uniquely as , with and . Differentiating gives
Fix a rectangle where this graph contains every equilibrium. Thus has a strict local minimum if and a strict local maximum if it is negative. More precisely, continuity of makes have opposite nonzero signs on the two sides of in a sufficiently small interval. The graph is strictly monotone on each side. Choose small enough that each monotone side covers the corresponding interval of parameter values adjacent to the vertex. A horizontal line , with , intersects each side exactly once when , and neither side when that ratio is negative. Strict extremality gives as the unique solution when . This proves the local count; the side condition in the theorem is equivalent.
Taylor's formula gives
Inverting this relation on the two monotone sides gives the displayed square-root asymptotics. No differentiability of at is asserted.
For stability, differentiate the graph identity to get
The last equality uses . Thus the two branches have opposite nonzero signs of ; the scalar stability test gives the conclusion. All assertions concern equilibria in a sufficiently small neighborhood, not possible distant equilibria.
Proof X-ray
Why the parameter is the useful dependent variable
At the bifurcation , so the implicit function theorem cannot produce as a function of ; the normal form shows that no such function exists. The hypothesis is used exactly once, to solve for instead, and from then on the problem is about the graph of one real function. The condition reappears as , a critical point, and becomes , which makes that critical point a strict turn with two monotone sides. Counting equilibria is now counting intersections of a horizontal line with a parabola-like graph. Stability needs no new idea: the identity transfers the opposite signs of on the two sides into opposite signs of on the two branches. Each dropped hypothesis breaks a specific step, as the counterexamples below show.
Worked examples
Worked example 3.5.4
(The normal form and its semistable limiting equilibrium)
For , the nondegeneracy derivatives at are and . If the equilibria are ; makes stable and unstable. If there is no equilibrium.
At , the hyperbolic test is silent. The explicit solution is
on its maximal interval. Positive initial values tend to zero, while for negative initial values the solution tends to as . The equilibrium is one-sided attractive but not Lyapunov stable. The parameter-side conclusion does not classify the degenerate equilibrium automatically.
Worked example 3.5.5
(Exchanging stability is a different geometry)
For , the equilibrium branches are and . For , and ; for their stability is reversed. Both branches exist on both sides and cross at zero: this is the transcritical pattern.
Here , so the saddle-node theorem does not apply. At the equation is again ; linearization alone does not resolve that equilibrium.
Worked example 3.5.6
(A pitchfork and stability at the critical parameter)
For , is an equilibrium for every , and are additional equilibria when . The derivative makes zero stable for and unstable for , while both nonzero branches are stable. Again , so the saddle-node theorem is inapplicable.
At , solve to obtain
It tends to zero from either side, so the origin is asymptotically stable even though . For every nonzero initial value its decay is algebraic, not exponential; also proves Lyapunov stability. A zero eigenvalue does not by itself decide stability.
Counterexample mode
Dropping a nondegeneracy condition changes the count
For , at we have and . There is one equilibrium at and none for either nonzero parameter. This shows that cannot be dropped from the two-branch theorem.
For , instead but . Each parameter has exactly one real equilibrium . Neither example satisfies the saddle-node hypotheses, and neither has the asserted two-on-one-side local pattern.
Common mistakes
Common mistake
(Reading the side of the branches from one derivative)
The branches lie on the side where , so the side depends on the ratio of the two derivatives. The equations and both have , yet their branches exist for and for respectively, because in the first and in the second. Likewise, which branch is stable is decided by the sign of , not by the sign of .
Common mistake
(Treating the local count as a global count)
The theorem counts equilibria in a fixed small neighborhood of . For , the right side at is , so besides the double root at zero there is the root , where . That equilibrium is hyperbolic and persists for all small . Hence the equation has three equilibria for small and one for small , not two and none.
Exercises
Quick check
(For , find all equilibria and their stability when . Which side of zero contains equilibria, and which hypothesis determines that side?)
Solution · Answer
For , the roots are . Since , the negative root is stable and the positive root unstable. No real equilibrium exists for . The sign of determines the parameter side. Here , , so the theorem's sign condition is .
Quick check
(For , verify the saddle-node hypotheses at . Determine the leading asymptotic form and stability of its two nearby branches for small ; make no claim about distant roots.)
Solution · Answer
, and at the origin. Thus for small , . Because , the positive branch is stable and the negative branch unstable. The result is local; the cubic can have other equilibria away from zero.
Quick check
(For , classify the equilibrium at zero for , , and .)
Solution · Answer
At zero, , giving exponential stability for and instability for . At the explicit formula above proves asymptotic stability, with algebraic, non-exponential decay for every nonzero initial value.
Quick check
(Why can one not apply the implicit function theorem to solve as a single differentiable function through ? Which variable can be solved for smoothly instead?)
Solution · Answer
The zero set has two roots for each small and none for , so there is no such function on a two-sided neighborhood. Also , so the usual implicit-function hypothesis for solving for fails. Since , one can solve smoothly for instead.