Harmonic-Coupled Riccati Equation (HCRE)

• HCRE is a matrix equation coupling multiple Riccati-like matrices via harmonic means, emerging in networked estimation and periodic control.
• It establishes rigorous existence, uniqueness, and convergence properties, enabling efficient computation in both finite- and infinite-dimensional settings.
• Key applications include distributed sensor network filtering, target tracking, and optimal harmonic LQ control for periodic systems.

The Harmonic-Coupled Riccati Equation (HCRE) is a class of matrix equations emerging from two distinct but technically interrelated domains: distributed filtering for networked systems (notably, consensus-on-information algorithms) and infinite-dimensional control of periodic systems. Its fundamental feature is the coupling of multiple Riccati-like matrices, either through harmonic means or via block structures induced by system periodicity. This leads to a nontrivial fixed-point condition, central to state estimation error analysis, performance characterization of distributed filters, and optimal periodic control design. The HCRE generalizes classical Riccati equations both structurally and analytically; recent developments provide rigorous existence, uniqueness, and convergence results, as well as efficient computational methods for large- and infinite-dimensional settings (Qian et al., 2022, Riedinger et al., 2022).

1. Formal Definition and Mathematical Structure

The HCRE as developed in the context of distributed Kalman-type filtering assumes $N$ local agents (sensors) indexed by $i = 1, \ldots, N$ with individual observation matrices $C_i$, process noise $Q > 0$, measurement covariance $R_i > 0$, a fixed system state transition $A \in \mathbb{R}^{n \times n}$ ($A$ invertible), and an inter-agent weight matrix $L = [l_{ij}]$ that is row-stochastic and primitive. The set of $N$ positive definite matrices $(P_1, \ldots, P_N)$ solve the coupled system:

$$P_i = A \left( \sum_{j=1}^N l_{ij} P_j^{-1} + \sum_{j=1}^N l_{ij} C_j^T R_j^{-1} C_j \right)^{-1} A^T + Q, \quad i = 1, \ldots, N.$$

This structure couples the Riccati equations of each agent via harmonic means of the covariance inverses, in contrast to traditional additive (arithmetic) coupling. Equivalently, introducing the "harmonic-fused" local Gramian $\tilde P_i = (\sum_j l_{ij}P_j^{-1})^{-1}$, one obtains:

$$P_i = A \left( \tilde P_i^{-1} + \tilde C_i^T \tilde R_i^{-1} \tilde C_i \right)^{-1} A^T + Q$$

with

$$\tilde C_i = [\,\text{sgn}(l_{i1})C_1^T, \ldots, \text{sgn}(l_{iN})C_N^T]^T,\quad \tilde R_i = \operatorname{diag}(1/l_{i1} R_1, \ldots, 1/l_{iN} R_N).$$

A comparable structure appears in the harmonic Riccati equations for linear time-periodic systems (infinite-dimensional setting):

$$(\mathcal{A} - \mathcal{N})^* X + X (\mathcal{A} - \mathcal{N}) - X \mathcal{B}\mathcal{R}^{-1} \mathcal{B}^* X + \mathcal{Q} = 0,$$

where $\mathcal{A}, \mathcal{B}, \mathcal{Q}, \mathcal{R}$ are block Toeplitz (harmonic) operators and $\mathcal{N}$ is a frequency diagonal shift (Riedinger et al., 2022).

2. Origin: Distributed Filtering and Infinite-Dimensional Control

The HCRE in multi-agent state estimation arises from the matrix iterative law underpinning the Consensus-on-Information Distributed Filtering (CIDF) algorithm. Each agent maintains prior and posterior error covariances $P_{i,k|k-1}$, $P_{i,k|k}$, fusing them via matrix harmonic mean weighted by $L$. The Kalman correction is locally applied at each node. The iteration:

$$P_{i,k+1|k} = A\Bigl(\sum_j l_{ij}P_{j,k|k-1}^{-1} + \sum_j l_{ij} C_j^T R_j^{-1} C_j \Bigr)^{-1} A^T + Q$$

with correction via

$$P_{j, k|k}^{-1} = P_{j, k|k-1}^{-1} + C_j^T R_j^{-1} C_j$$

leads, in the $k \to \infty$ limit, to the fixed-point system of HCRE (Qian et al., 2022).

In infinite-dimensional settings, HCRE emerges in the harmonic (Fourier) lifting approach to LTP (linear time-periodic) systems, recasting periodic control and filtering problems into an equivalent time-invariant problem on $\ell^2(\mathbb{C}^n)$ (Riedinger et al., 2022). The associated Riccati or Lyapunov equation becomes block-coupled through the harmonic operator structure.

3. Existence, Uniqueness, and Convergence

The existence and uniqueness theory for HCRE in networked estimation requires two key conditions (Qian et al., 2022):

• Collective observability: $A$ invertible, and $(A, [C_1^T \ldots C_N^T]^T)$ globally observable.
• Primitivity: $L$ is a primitive row-stochastic matrix (some power $L^m > 0$ elementwise).

Under these, the paper proves:

Theorem 1 (Uniqueness): There exists a unique set of positive definite matrices $\{P_i\}_{i=1}^N$ solving the HCRE.

Convergence under monotone, bounded matrix iterations starting from any positive definite initialization is established via contraction mappings built on information-theoretic bounds and the decay properties implied by $L$'s primitivity. For the infinite-dimensional harmonic Riccati case, monotonic convergence of a Kleinman-like iteration is proved under standard stabilizability/detectability assumptions, with explicit operator-norm error bounds (Riedinger et al., 2022).

4. Computational Algorithms

For networked (finite-dimensional) HCRE, convergence of the CIDF iteration allows the direct computation of fixed points by iterating:

$$P_{i,k+1} = A \Bigl( \sum_j l_{ij} P_{j, k}^{-1} + \sum_j l_{ij} C_j^T R_j^{-1} C_j \Bigr)^{-1} A^T + Q.$$

For infinite-dimensional harmonic settings, the solution leverages block Toeplitz and diagonal harmonic representatives:

• Floquet factorization: To explicitly characterize spectral properties and construct solution representations, using state transition matrices over the system period.
• Truncation strategies: Block banded truncation and avoidance of spurious spectrum as explained by spectral “wings” behavior, enabling reliable recovery of the true solution.
• Sylvester equation embedding: Casting the problem into a large but finite linear system, solvable via standard methods, with guaranteed error decay as truncation is refined.
• Kleinman iteration: Recursively improves stabilizing gains and associated Riccati/Lyapunov solutions, converging monotonically to the HCRE fixed point (Riedinger et al., 2022).

5. Analytical Consequences and Connections to Lyapunov Equations

Once a steady-state HCRE solution $\{P_i\}_{i=1}^N$ is established, the error covariance of the distributed estimator network satisfies a block DLE (discrete-time Lyapunov equation). Define steady-state posterior/prior covariances and let:

$$\mathcal{A} = [l_{ij} A \bar P_i P_j^{-1}]_{i, j=1}^N, \quad \Gamma = [l_{ij}A \bar P_i C_j R_j^{-1}]_{i, j}.$$

Then the joint error covariance satisfies

$$\mathcal{P} = \mathcal{A} \mathcal{P} \mathcal{A}^T + \Gamma R \Gamma^T + (1_N 1_N^T) \otimes Q$$

and $\mathcal{A}$ is Schur stable as established using Perron–Frobenius theory. Solving this DLE with the computed HCRE $\{P_i\}$ yields the precise steady-state MSE for the network (Qian et al., 2022).

6. Applications and Numerical Examples

Documented applications include:

• Distributed filtering: HCRE underpins provable consensus performance in sensor networks. Numerical examples—both small, structured (e.g., three-node, scalar) networks and large random graphs with observable/unobservable clusters—demonstrate convergence, sharpness of bounds, and the reduction of conservativeness relative to previous approaches (Qian et al., 2022).
• Target tracking: Monte Carlo experiments validate HCRE-based DLE predictions of network MSE matching empirical performance across scenarios.
• Harmonic LQ trajectory tracking: In the LTP setting, HCRE-based design allows optimal feedback tracking of periodic references with explicit construction of the periodic gain from harmonic-domain Riccati solutions. Numerical evidence shows rapid convergence in phasor (Fourier) modes and stable tracking under widely varying references (Riedinger et al., 2022).

7. Limitations and Scope

Critical assumptions for HCRE solvability are collective observability, primitivity or Hurwitzicity (in the infinite-dimensional setting, invertibility of $(\mathcal{A} - \mathcal{N})$), and the availability of an initial stabilizing gain in Kleinman-type iterations. While the provided algorithms efficiently approximate HCRE solutions to arbitrary precision, accuracy must be balanced against truncation order. In the infinite-dimensional case, spurious spectrum management and convergence monitoring via Lyapunov residuals are essential (Riedinger et al., 2022). The theory currently holds for linear, time-invariant interconnections and LTP system classes, with extensions to nonlinear or time-varying couplings an open direction.

---

References:

(Qian et al., 2022): Harmonic-Coupled Riccati Equations and its Applications in Distributed Filtering (Riedinger et al., 2022): Solving Infinite-Dimensional Harmonic Lyapunov and Riccati equations