Ackermann's formula

Ackermann's formula is an explicit algebraic method for computing the state-feedback gain matrix \(K\) that places the closed-loop poles of a single-input linear system at arbitrarily specified locations. Discovered by Jürgen Ackermann in 1972, the formula provides a direct computational route to pole placement without requiring iterative numerical optimization or the solution of a Riccati equation. For a system in state-space form \(\dot{x} = Ax + Bu\) with scalar input \(u\), the feedback law \(u = -Kx\) yields the closed-loop dynamics \(\dot{x} = (A - BK)x\), and Ackermann's formula gives \(K\) explicitly in terms of \(A\), \(B\), and the desired characteristic polynomial.

The Formula

Let the desired closed-loop characteristic polynomial be

\phi(s) = s^n + \alpha_{n-1} s^{n-1} + \cdots + \alpha_1 s + \alpha_0

and let \(\phi(A) = A^n + \alpha_{n-1} A^{n-1} + \cdots + \alpha_1 A + \alpha_0 I\) be the matrix polynomial obtained by substituting \(A\) for \(s\). Let \(\mathcal{C} = [B \; AB \; A^2B \; \cdots \; A^{n-1}B]\) be the controllability matrix. Then Ackermann's formula states:

K = \begin{bmatrix} 0 & 0 & \cdots & 0 & 1 \end{bmatrix} \mathcal{C}^{-1} \phi(A)

The formula is valid if and only if the pair \((A, B)\) is controllable — that is, if and only if \(\mathcal{C}\) has full rank. When the system is not controllable, no state-feedback gain can place the poles arbitrarily, and Ackermann's formula fails because \(\mathcal{C}\) is singular. This reveals a deep structural fact: pole placement is possible precisely to the extent that the system's internal dynamics are reachable through its input.

Significance and Limitations

Ackermann's formula is significant primarily as a theoretical guarantee and a pedagogical tool. It proves constructively that controllability implies arbitrary pole assignability for single-input systems — a foundational result in modern control theory. The formula also generalizes to multi-input systems, though the multi-input case requires additional degrees of freedom and is typically handled through more sophisticated algorithms rather than direct matrix inversion.

In practice, however, Ackermann's formula is rarely used for high-order systems. The direct computation of \(\mathcal{C}^{-1}\) and \(\phi(A)\) is numerically ill-conditioned for systems of order greater than about ten: the controllability matrix becomes nearly singular, and the matrix polynomial \(\phi(A)\) amplifies rounding errors. Practical pole placement relies instead on numerically stable algorithms such as the QR method or the Schur method, which solve the same problem without explicit matrix inversion. Ackermann's formula thus occupies a curious position: it is a beautiful exact result that is computationally fragile, a symbol of what is theoretically possible and practically precarious.

The formula also connects to broader themes in system design. Pole placement via state feedback assumes that the full state vector is available for measurement, which is rarely true in practice. When only outputs are measured, the feedback law must be combined with a state estimator — typically a Luenberger observer or a Kalman filter — leading to the separation principle: the controller and observer can be designed independently. Ackermann's formula places the controller poles; a dual formula places the observer poles. The two together form the complete architecture of modern output-feedback control.

Ackermann's formula is often taught as a computational recipe — a sequence of matrix operations that produces a gain vector. This pedagogy misses the point. The formula's true content is not the computation but the theorem it embodies: that the geometry of controllability and the algebra of characteristic polynomials are two faces of the same structure. The controllability matrix is not merely a test for whether pole placement is possible; it is the coordinate system in which the pole-placement problem becomes trivial. Any presentation of Ackermann's formula that does not emphasize this geometric duality — between the reachability of states and the assignability of dynamics — has reduced a structural insight to a mere algorithm.