Skip to content

Stability of Solutions of Higher-Order ODEs


Proposition

Consider the linear ordinary differential equation (ODE) with constant real coefficients

x(n)(t)+an−1x(n−1)(t)+⋯+a1x′(t)+a0x(t)=g(t),x^{(n)}(t)+a_{n-1}x^{(n-1)}(t)+\dots +a_1x'(t)+a_0x(t)=g(t),

with x′(t)=dxdt(t)x'(t)=\dfrac{\mathrm{d}x}{\mathrm{d}t}(t) and characteristic equation p(r)=rn+an−1rn−1+⋯+a0=0p(r)=r^n+a_{n-1}r^{n-1}+\dots+a_0=0. For the homogeneous equation (g≡0g\equiv0):

  1. Every solution satisfies x(t)→0x(t)\to0 as t→∞t\to\infty if and only if every root of pp has negative real part. The equation is then called asymptotically stable.
  2. If some root has positive real part, some solutions are unbounded: the equation is unstable.
  3. If all roots have real part ≤0\leq0, all solutions are bounded on [0,∞)[0,\infty) if and only if every root with real part 00 is simple; the equation is then called stable. A repeated root on the imaginary axis gives unbounded terms such as tt or tcos⁡βtt\cos\beta t.

For constant g≡cg\equiv c with a0≠0a_0\neq0, every solution tends to the equilibrium c/a0c/a_0 if and only if all roots have negative real part.

In Math V, stable means that all solutions stay bounded, and asymptotically stable that they all converge (case 1). A simple root on the imaginary axis therefore gives a stable equation that is not asymptotically stable.


Interpretation

By the general solution of a linear ODE, every homogeneous solution is a combination of terms tjertt^j\mathrm{e}^{rt}, one family for each root r=α+iβr=\alpha+\mathrm{i}\beta. With Euler's formula, ∣tjert∣=tjeαt|t^j\mathrm{e}^{rt}|=t^j\mathrm{e}^{\alpha t}: the oscillation eiβt\mathrm{e}^{\mathrm{i}\beta t} has modulus 11, and only α\alpha decides growth or decay. Exponential decay beats any polynomial factor tjt^j, so α<0\alpha<0 is enough.

In the complex plane the open left half-plane is shaded; the roots −1 and −2 lie inside it, ±2i lie on the imaginary axis, and 1 ± 2i and a double root 0 are unstable.

The figure places the characteristic roots of the examples below in the complex plane. The roots −1,−2-1,-2 of x′′+3x′+2x=4x''+3x'+2x=4 lie in the shaded half-plane Re⁡r<0\operatorname{Re} r<0, so every solution converges. The simple roots ±2i\pm2\mathrm{i} of x′′+4x=0x''+4x=0 lie on the imaginary axis: solutions stay bounded but do not decay. The roots 1±2i1\pm2\mathrm{i} of x′′−2x′+5x=0x''-2x'+5x=0 lie to the right, and the double root 00 of x′′=0x''=0 lies on the axis but is repeated; both give unbounded solutions.

The criterion is the continuous-time counterpart of stability of an equilibrium: writing the equation as a first-order system, the roots are the eigenvalues of the companion matrix, and "negative real part" plays the role of f′(x∗)<0f'(x^*)<0. For difference equations the boundary is the unit circle ∣λ∣=1|\lambda|=1, not the imaginary axis; see stability of a scalar difference equation.

Typical mistakes: requiring only real roots to be negative and ignoring complex ones; treating purely imaginary roots as stable in the asymptotic sense; and concluding stability from positive coefficients alone when n≥3n\geq3.


Key Results

  • Order 2: x′′+a1x′+a0x=0x''+a_1x'+a_0x=0 is asymptotically stable if and only if a1>0a_1>0 and a0>0a_0>0 (the roots have sum −a1-a_1 and product a0a_0).
  • For any nn, positive coefficients a0,…,an−1a_0,\dots,a_{n-1} are necessary for asymptotic stability, but not sufficient when n≥3n\geq3.
  • Stability is a property of the homogeneous part and does not depend on gg: all solutions x=xp+xhx=x_p+x_h satisfy x(t)−xp(t)→0x(t)-x_p(t)\to0 if and only if all roots have negative real part.
  • The same test applies to the first-order system form: its companion matrix has the roots of pp as eigenvalues, as for the linear vector ODE.

Beyond the course

Many textbooks state the Routh-Hurwitz criterion, which decides from the coefficients alone whether all roots have negative real part. It is not part of Math V.


Examples

Asymptotically stable with forcing. x′′+3x′+2x=4x''+3x'+2x=4: roots −1,−2-1,-2, and the particular solution is xp=2x_p=2. Every solution 2+c1e−t+c2e−2t2+c_1\mathrm{e}^{-t}+c_2\mathrm{e}^{-2t} tends to 22.

Pure oscillation. x′′+4x=0x''+4x=0: roots ±2i\pm2\mathrm{i}, solutions c1cos⁡2t+c2sin⁡2tc_1\cos2t+c_2\sin2t are bounded but do not tend to 00: stable, not asymptotically stable.

Growing oscillation. x′′−2x′+5x=0x''-2x'+5x=0: r=1±2ir=1\pm2\mathrm{i}, solutions et(c1cos⁡2t+c2sin⁡2t)\mathrm{e}^{t}(c_1\cos2t+c_2\sin2t): unstable, matching a1=−2<0a_1=-2<0.

Double root at zero. x′′=0x''=0: r=0r=0 twice, solutions c1+c2tc_1+c_2t, unbounded: unstable although no root has positive real part.

Positive coefficients are not enough. r3+r2+r+1=(r+1)(r2+1)r^3+r^2+r+1=(r+1)(r^2+1) has roots −1,±i-1,\pm\mathrm{i}. All coefficients are positive, yet the simple roots ±i\pm\mathrm{i} lie on the imaginary axis: stable, not asymptotically stable.


Prerequisites



Sources

Aliases: Stability of Solutions of Higher Order ODEs, stability of a linear ODE, asymptotic stability of a higher-order ODE Keywords: negative real part, characteristic roots, Routh-Hurwitz, a_1>0, a_0>0, resonance on the imaginary axis Symbols: alpha, beta, infinity, lambda

Last updated: 27 October 2026 11:52