Riccati equation

The Riccati equation is a nonlinear first-order differential equation of the form $\dot{P} = PA + A^T P - PBR^{-1}B^T P + Q$, named after Jacopo Riccati (1676–1754). In control theory, the algebraic Riccati equation (ARE) — the steady-state form where $\dot{P} = 0$ — is the central equation of LQR design and Kalman filtering. Its solution $ is the unique positive-definite matrix that yields the optimal feedback gain = R^{-1}B^T P$, minimizing a quadratic cost over state and control trajectories.

The Riccati equation is remarkable because it compresses an infinite-horizon optimization problem into a single matrix equation. But this compression comes with fragility: the equation has a solution only when the system satisfies certain controllability and observability conditions, and small perturbations in the matrices $, $, $, or $ can destroy the existence of a positive-definite solution. The ARE is therefore not merely an algebraic convenience; it is a diagnostic test for whether the control problem is well-posed.

The Riccati equation is control theory's Faustian bargain: it gives you the optimal controller in closed form, but only if your system is well-behaved enough to deserve one. The systems that need optimal control most — the unstable, the uncertain, the barely controllable — are precisely the systems for which the Riccati equation has no solution, or worse, has a solution that looks optimal but is numerically illusory.