Kharkov, 1892: a student of Chebyshev asks whether a system settles or flies apart — and answers it without solving the equation of motion. Build an energy-like V(x)>0 whose rate V̇≤0, and the origin is stable by that fact alone. This panel is runnable: it solves the Lyapunov equation AᵀP+PA=−I to certify a linear system, and it measures the Lyapunov exponent of the logistic map — the number whose sign is the border of chaos.
source A. M. Lyapunov, The General Problem of the Stability of Motion (Kharkov Mathematical Society, 1892) — doctoral thesis, a student of P. L. Chebyshev. Room: THE NUMBER. Rendered, not quoted.
Aleksandr Mikhailovich Lyapunov (1857–1918), a student of Chebyshev at St Petersburg, defended The General Problem of the Stability of Motion at Moscow University in 1892 (the memoir published by the Kharkov Mathematical Society, where he was professor). Its radical move: decide stability without integrating the equations.
He sits in THE NUMBER beside his teacher Chebyshev and Chebyshev’s other students Markov and the makers of the continued fractions — the same St Petersburg line. His name also names the Lyapunov condition in the Central Limit Theorem.
Find a scalar V(x) — an energy — with V(x)>0 for x≠0, V(0)=0, and whose rate along the flow is V̇(x)≤0. Then energy never increases, so the state cannot run away: the origin is stable. If V̇<0 strictly, it bleeds to zero — asymptotically stable.
No trajectory is ever computed. A single inequality on one function replaces solving the whole system — the “second method” that carries Lyapunov’s name.
The direct method is the spine of modern control theory (Lyapunov functions certify controllers and adaptive laws). The Lyapunov exponent became the yardstick of chaos — weather predictability, orbital mechanics, turbulence.
And the name reaches back into THE NUMBER itself: the Lyapunov condition is a sufficient hypothesis for the Central Limit Theorem, sharpening the work of his teacher Chebyshev and his fellow student Markov.
Top: a linear system ẋ = A x, A a 2×2 matrix — the engine solves AᵀP+PA=−I and reads off stability from P. Bottom: the logistic map x → r·x(1−x) — the engine measures its Lyapunov exponent λ.
Presets fill the four boxes; edit them freely. r = 4 is full chaos, r = 2 is a superstable fixed point.
1 · The direct method (linear). Solve the Lyapunov equation for the symmetric P, test whether it is positive-definite (both leading minors > 0). If so, V = xᵀP x is a valid energy and V̇ = −|x|² < 0 proves asymptotic stability.
| P | col 1 | col 2 |
|---|---|---|
| row 1 | — | — |
| row 2 | — | — |
leading minors: — · residual AᵀP+PA: —
V̇(x) at x=(1,1): — eigenvalues of A: —
2 · The Lyapunov exponent (nonlinear). λ = mean of ln|f′(x)| along an orbit of x→r·x(1−x), f′=r(1−2x). λ>0 is chaos; λ<0 is a stable attractor.
λ(r=4.00) = — regime: —
| A | P>0 (certifies) | trajectory | agree |
|---|
The certificate (P positive-definite) is checked against the actual simulated trajectory ẋ=Ax: a stable A must shrink to 0, an unstable A must blow up. Certificate and reality must agree on every row.
The real wall the method admits: finding V can be hard for a general nonlinear system — but for a linear A it is not a search at all. The Lyapunov equation hands you P by solving a linear system, as the engine does live.
“You need the explicit solution x(t) to judge stability.” Cut. The direct method needs none — V>0, V̇≤0 decides it without ever integrating.
“A positive V̇ is fine as long as V is positive.” Cut. V̇>0 means energy is growing — the state runs away. The sign of the rate is the whole test.
“A bounded orbit means the system is not chaotic.” Cut. The logistic map stays in [0,1] forever yet has λ=ln2>0 — bounded and chaotic.
The red move: make the certificate accept V̇>0 as stable — drop the requirement that P be positive-definite, so a divergent system gets stamped stable.
Flip the sign test and the certificate lies about the unstable presets: it calls them stable while the simulated trajectory (5) actually diverges. The witness compares the two and turns red. The attack is real and it is caught.