Bifurcations and qualitative transitions

Start with two resting states

Take a dimensionless state x and time t in seconds. Supply the exact rule dx/dt = (μ − x²)/T, with T = 1 s and a dimensionless control parameter μ. The Greek letter μ is read mu. This is an invented normalized model for learning the reasoning; no apparatus or observations are being fitted.

First hold μ = 0.25. At x = 0, the rate is +0.25 s⁻¹, so x increases. At x = 0.5 and x = −0.5, the rate is zero. Either of these two states can remain constant. Now hold μ = −0.25 instead: μ − x² is negative for every real x, so there is no resting state. Changing one number has changed the number of possible equilibria.

Algebra, functions, graphs, and logarithms prepares substitution, square roots and graph reading; logarithms are not needed here. Follow the root calculation and figure first. Derivatives, Taylor series, and local approximation and Ordinary differential equations are optional return paths for the derivative test and solution check. Their needed ideas are introduced locally.

Separate state, parameter and time

The state x(t) can change during a solution. The parameter μ is held constant during each solution of the main model. Comparing μ = −0.25, 0 and 0.25 compares three different autonomous rules: autonomous means the rate has no explicit time dependence. It does not describe one experiment in which μ is being turned as time passes.

Write f(x; μ) = (μ − x²)/T for the rate. The semicolon separates the state argument from the fixed parameter. Both μ and x² are dimensionless, so they can be subtracted; division by T gives the rate units s⁻¹. Negative x is allowed in this model, which is why its full real neighborhood matters.

An equilibrium x* satisfies f(x*; μ) = 0. The star labels a constant equilibrium value, not multiplication. The polynomial rate is smooth, so any finite initial state has a unique local solution. Local does not promise a solution at every later time; the critical example below will show a finite-time limit.

Count the equilibria before drawing branches

Set the rate to zero: μ − (x*)² = 0, or (x*)² = μ. A square of a real number is nonnegative. For μ < 0 there are no real equilibria. For μ = 0 there is one distinct equilibrium, x* = 0; writing both +0 and −0 does not create two states.

For μ > 0 there are two equilibria, x* = +√μ and x* = −√μ. Thus μ = 0.25 gives ±0.5, while μ = 1 gives ±1. The positive and negative solutions form two branches when we plot equilibrium value against parameter.

The number of real equilibria changes across μ = 0. Their positions also change continuously along each positive-parameter branch. Those are different observations: moving an equilibrium from 0.5 to 1 need not create a new qualitative structure.

Read branches and nearby directions

For a fixed positive μ, put a = √μ. Since μ − x² = (a − x)(a + x), the rate is negative when x < −a, positive when −a < x < a, and negative when x > a. A phase line records these directions on the state axis: a right arrow means increasing x and a left arrow means decreasing x.

Near +a, directions on both sides point toward it. Nearby solutions remain nearby and approach +a: it is locally asymptotically stable. Near −a, directions point away from it, so it is unstable. Stability in the Lyapunov sense means every chosen small neighborhood can be protected by starting sufficiently close; asymptotic stability also requires nearby states to approach the equilibrium. Attraction from just one side does not by itself establish this two-sided requirement on the real line.

The upper panel places parameter μ horizontally and equilibrium state x* vertically. Its solid branch and filled circles indicate the stable positive equilibrium; its dashed branch and hollow squares indicate the unstable negative equilibrium. The cross marks the critical zero. The lower rows are phase lines at three fixed parameter values. Read their arrows for direction only: their equal lengths do not encode rates or elapsed displacements.

Check the critical state from both sides

At μ = 0, the rule becomes dx/dt = −x²/T. Every nonzero state has a negative rate. A positive start moves left toward zero; a negative start moves farther left, away from zero. The cross therefore records a one-sided attracting case, not an equilibrium stable against all sufficiently small real perturbations.

For an initial value x(0) = x₀, the exact solution is x(t) = x₀/(1 + x₀t/T), on the interval containing zero where the denominator stays positive. Differentiation gives −x₀²/[T(1 + x₀t/T)²] = −x(t)²/T, so the formula satisfies the rule and its initial value. The separate initial value x₀ = 0 gives the constant zero solution.

With x₀ > 0, the denominator grows and x approaches zero as time tends to infinity; it does not reach zero in finite time. With x₀ < 0, the denominator vanishes at t = −T/x₀. The solution goes unbounded before it can be continued through that time as a finite real state. A smooth rate defined at every real x still need not give every start a solution for all future time.

Identify the qualitative transition

A bifurcation is a qualitative change in a family’s dynamics as a control parameter changes. Here no equilibria for negative μ, one critical equilibrium at zero, and a stable–unstable pair for positive μ describe such a change. Pair creation or destruction of this kind is called a saddle-node or fold bifurcation. Other bifurcations can change stability without changing the number of equilibria.

In our exact field, the critical conditions are f(0; 0) = 0 and ∂f/∂x(0; 0) = 0. The dependence on the parameter and curvature are nonzero: ∂f/∂μ = 1/T and ∂²f/∂x² = −2/T. These calculations accompany the directly established root and sign structure; a zero state derivative by itself would not establish the transition.

The branches are exact equilibria across alternative fixed parameters. They are not a state trajectory, and the blank negative-parameter part means no equilibrium rather than no allowed state or no solution. The displayed parameter and state windows are drawing choices, not boundaries of the polynomial model.

Use a local rate test with its conditions

For a continuously differentiable one-dimensional autonomous rate, a nonzero state derivative at an equilibrium supplies a local test: negative derivative means local asymptotic stability, while positive derivative means instability. Here ∂f/∂x = −2x/T, so the positive branch has derivative −2√μ/T and the negative branch has derivative +2√μ/T.

To see the local approximation, write x = x* + η, where η is a small dimensionless deviation. Since (x*)² = μ, the exact equation is dη/dt = [−2x*η − η²]/T. Its linearized rule retains only −2x*η/T. At the positive equilibrium this linear rule predicts exponential decay with characteristic time T/(2√μ). That is a local linearized timescale, not an exact time for every nonlinear trajectory to settle.

As positive μ approaches zero, the magnitude of the nonzero derivative decreases and this linearized timescale grows. At μ = 0 the linear term vanishes, so the nonzero-derivative test and its finite timescale formula no longer apply. The exact one-sided solution, rather than a zero-slope verdict, decides the critical behavior.

Keep the conclusion within the model

As a contrast, supply dx/dt = −x³/T for every value of an unused parameter μ. Zero is its only equilibrium; negative states increase toward zero and positive states decrease toward zero. It is asymptotically stable even though the state derivative there is zero. Nothing changes when this unused parameter changes, so that zero derivative does not mark a bifurcation.

If a parameter instead depends on time, the rule becomes a different, nonautonomous problem. A frozen-parameter diagram can identify equilibria of its instantaneous rate rules, but it does not prove that an evolving state follows a branch, jumps at zero, or retains a previous state. Initial data, the parameter history and the resulting differential equation must be checked.

This supplied example establishes a mathematical transition in one exact family. Using the idea for an actual physical system requires a justified model, relevant state and parameter ranges, and evidence about those quantities. The diagram alone supplies no empirical tipping threshold, uncertainty interval, hysteresis or route to chaos. The sign, domain and parameter checks are the reasoning to carry to a new model.

Worked examples

Try it yourself

1. For fixed μ = 0.04 and T = 1 s, find the real equilibria and each state derivative. Classify their local stability.

   Show the answer and reasoning

   The equilibria are −0.2 and +0.2. Their derivatives are +0.4 s⁻¹ and −0.4 s⁻¹; the negative root is unstable and the positive root is locally asymptotically stable.

   Because 0.2² = 0.04, both roots satisfy the zero-rate equation. Evaluating −2x/T at each root keeps the signs separate. The nonzero-derivative test applies to this smooth autonomous model.

2. At fixed μ = 0.25, x = 0.4 and T = 1 s, find dx/dt. Which direction belongs on the phase line? Does a 48 px arrow imply a 48-unit displacement?

   Show the answer and reasoning

   The rate is +0.09 s⁻¹ and the arrow points right. Its drawn length gives no elapsed displacement.

   0.25 − 0.4² = 0.09. The positive sign means increasing dimensionless x. The figure uses equal arrow lengths only for direction; it supplies no elapsed time with which to infer a displacement.

3. At μ = 0, a learner counts +√0 and −√0 as two stable equilibria because the derivative is zero. Correct both claims on the full real state line.

   Show the answer and reasoning

   There is one distinct equilibrium, x* = 0, and it is unstable in the two-sided Lyapunov sense. It attracts from the right and repels to the left.

   Both written roots equal zero. The rate −x²/T is negative on either side: a positive state moves toward zero, while a negative state moves away. Arbitrarily small negative starts eventually leave a fixed small neighborhood, so one-sided attraction does not establish stability.

4. The solid upper branch includes (μ, x*) = (0.25, 0.5) and (1, 1). Does it show one state increasing from 0.5 to 1 over 0.75 s? What do its axes actually show?

   Show the answer and reasoning

   No. The axes show a dimensionless fixed parameter and a dimensionless equilibrium value across different models; neither axis is elapsed time.

   Each branch point solves μ − (x*)² = 0 at its own fixed μ. The difference 1 − 0.25 = 0.75 is a parameter difference, not a duration. A time history requires an initial state and a specified evolution rule.

5. In the separate family dx/dt = −x³/T for every value of an unused μ, the derivative at zero is always zero. Does that establish a bifurcation when μ crosses zero?

   Show the answer and reasoning

   No. The rate rule, sole equilibrium and asymptotic stability are unchanged for every μ.

   For x > 0 the rate is negative and for x < 0 it is positive, so nearby states approach zero from both sides. A zero derivative leaves the linear test inconclusive, but the nonlinear signs decide this case. There is no parameter-dependent qualitative change in this family.

6. Replace the fixed parameter by μ(t) = 0.25 − t/(4 s). Its values at 0, 1 and 2 s are 0.25, 0 and −0.25. Does the static diagram alone prove that a state follows the stable branch and jumps at 1 s?

   Show the answer and reasoning

   No. This is a different nonautonomous problem, and neither branch tracking nor a jump time follows from the frozen-parameter diagram alone.

   The expression t/(4 s) is dimensionless. The listed parameter values are correct, but a state history also requires an initial condition and a solution of dx/dt = [μ(t) − x²]/T. Instantaneous equilibria of the frozen rules are not automatically that evolving solution.

What to carry forward

• Solve the equilibrium equation and count distinct real roots at each fixed parameter.
• A bifurcation changes qualitative structure; a moving root or changing rate need not do so.
• Phase-line signs and conditional local tests establish stability; one-sided attraction and a zero derivative need separate analysis.
• An equilibrium-branch diagram varies a parameter, while a trajectory evolves in time under a specified rule and initial data.
• Keep exact model domains, finite-time limits and physical evidence separate from the displayed window.