Structured singular value

The structured singular value \(\mu\) is a generalization of the singular value that measures the robust stability of a system against structured uncertainty. Introduced by John Doyle in 1982, it addresses a limitation of the H-infinity norm: while H-infinity analysis treats all uncertainty as a single unstructured block, real systems often have uncertainty with structure — multiple independent perturbations, real parameter variations, or dynamically constrained uncertainties. The structured singular value computes exactly how large such structured uncertainty can grow before destabilizing the closed-loop system.

Formally, for a matrix \(M\) and a set of admissible uncertainty structures \(\Delta\), \(\mu_\Delta(M)\) is defined as the inverse of the smallest uncertainty magnitude that makes \(I - M\Delta\) singular. When \(\Delta\) is the set of all complex matrices of a given size, \(\mu\) reduces to the maximum singular value. For block-diagonal structures with mixed real and complex blocks, \(\mu\) is more complex — indeed, computing \(\mu\) exactly is NP-hard in general, and practical algorithms use upper and lower bounds.

The structured singular value is the central analysis tool in robust control. Together with H-infinity synthesis, it forms the \(\mu\)-synthesis loop: design a controller using H-infinity methods, analyze its robustness using \(\mu\), scale the problem to reduce conservatism, and iterate. The structured singular value transforms the abstract guarantee of robust control into a computable certificate.

The structured singular value is the mathematical admission that not all uncertainty is equal. Some uncertainties are correlated, some are constrained, some are real parameters that cannot vary arbitrarily. Treating structured uncertainty as unstructured is not conservative; it is wrong. The structured singular value is the correction.

See also: Robust control, H-infinity control, Control theory, Singular value decomposition, Loop shaping, Sensitivity function