Quasi-Linear Stochastic Dynamical Systems

• Quasi-linear stochastic dynamical systems are defined by an affine drift and a linear diffusion term in quantum stochastic differential equations.
• The framework uses a finite-dimensional operator algebra to reduce nonlinearities, ensuring closed-form moment equations and tractable dynamics.
• It supports classical Kalman-like observer design and stability analysis, offering explicit solutions for moment evolution and optimal filter synthesis.

A quasi-linear stochastic dynamical system, in the context of open quantum systems, is characterized by a special structure in which the drift term is affine and the diffusion (dispersion) term is linear in the system variables, all defined over a finite-dimensional operator algebra analogous to the Pauli matrices. This framework, formalized in the context of quantum stochastic differential equations (QSDEs) of Hudson–Parthasarathy type, enables explicit and tractable analysis of moment dynamics, quasi-characteristic functions, long-term cost growth, and optimal observer/filter design for a relevant class of quantum systems.

1. Definition and Structure of Quasi-Linear Quantum Stochastic Differential Equations

A quasi-linear quantum stochastic differential equation (QSDE) arises when self-adjoint system operators $X(t)\in\mathbb{R}^n$ interact with an $m$-channel bosonic input field $W(t)$, under a linear system Hamiltonian $H=E^T X$ and affine coupling operator $L=MX+N$, where $E\in\mathbb{R}^n$, $M\in\mathbb{R}^{m\times n}$, $N\in\mathbb{R}^m$. The Heisenberg equations of motion, in Hudson–Parthasarathy form, read

$dX = G(X)\,dt - i\,[X, L^T]\,dW,$

which can be equivalently rewritten as

$dX(t) = (A X(t) + b)\,dt + B(X(t))\,dW(t),$

where $A\in\mathbb{R}^{n\times n}$ and $b\in\mathbb{R}^n$ are constant, and $B(X)$ is an $n\times m$ matrix whose entries are linear in $X$. This signifies that the drift is affine in $X$ and the diffusion is linear in $X$, establishing the "quasi-linear" character.

2. Algebraic Foundation and Nonlinearity Reduction

The distinctive tractability of quasi-linear QSDEs is rooted in the finite-dimensional operator algebra satisfied by the system variables: $$X_j X_k = \alpha_{jk} I + \sum_{\ell=1}^n \beta_{jk\ell} X_\ell, \qquad j,k = 1,\dots,n,$$ where $\alpha\in\mathbb{C}^{n\times n}$ and $\beta\in\mathbb{C}^{n\times n\times n}$ obey Hermiticity ($\alpha^* = \alpha$, $\beta_\ell^* = \beta_\ell$) and certain quadratic closure constraints. This algebra generalizes the Pauli algebra for $n=3$. As a consequence, any quadratic or higher monomial in $X$ can be reduced to an affine function of $X$ itself, ensuring closure of moment equations at all orders and enabling explicit recursion for calculating higher-order statistics.

3. Moment Dynamics and Evolution Equations

For quasi-linear systems driven by vacuum input fields, the stochastic terms average to zero in expectation, leading to deterministic moment ODEs: $$\dot\mu = A\mu + b,\qquad \dot\Sigma_X = A\Sigma_X + \Sigma_X A^T + V(\mu),$$ where $\mu(t) = \mathbb{E}X(t) \in \mathbb{R}^n$ is the mean vector, $$\Sigma_X(t) = \mathrm{Cov}[X(t)] = \mathbb{E}[X(t) X(t)^T] - \mu(t)\mu(t)^T$$ is the covariance matrix, and $V(\mu)$ is an explicit positive semi-definite matrix function, affine in $\mu$, dependent on the system's noise properties. All moment equations, including higher-order moments, admit analogous closed-form ODEs due to the operator algebra.

4. Quasi-Characteristic Function and Invariant State

The quasi-characteristic function (QCF) of the system's quantum state is defined as

$$\Phi(t,u) = \mathbb{E}\,\exp(i u^T X(t)), \qquad u \in \mathbb{R}^n.$$

Exploiting the operator algebra reduction property, the QCF can be computed in closed form: $$\Phi(t,u) = \begin{bmatrix} 1 & 0 \end{bmatrix} \exp\left( i \begin{pmatrix} 0 & u^T \ \alpha u & \beta u \end{pmatrix} \right) \begin{pmatrix} 1 \ \mu(t) \end{pmatrix}.$$ If $A$ is Hurwitz (i.e., all eigenvalues have negative real part), the mean converges: $\mu(t)\to\mu_\infty=-A^{-1}b$, and the invariant state’s QCF is obtained by setting $\mu(t)=\mu_\infty$.

5. Asymptotic Cost and Nonlinear Functionals

For a real symmetric cost matrix $R\in\mathbb{R}^{n\times n}$, the infinite-horizon growth rate of the steady-state quadratic cost is given by

$$\lim_{T\to\infty} \frac{1}{T} \int_0^T \mathbb{E}[ X(t)^T R X(t)]\,dt = \mathrm{Tr}(R\alpha) + \sum_{\ell=1}^n \mathrm{Tr}(R\beta_\ell)\,\mu_{\infty, \ell},$$

where $\alpha$, $\beta_\ell$ originate from the system's operator algebra. More general functionals of Lur’e type also admit a reduction to an affine function of $\mu_\infty$, preserving tractability for a broad class of cost functions.

6. Kalman-like Filtering and Observer Design

In the measurement-based filtering scenario, a linear combination $Z = D Y$ of the output field $Y$ is measured, constructed so $Z$ is a classical diffusion with nonsingular covariance $F = DD^T \succ 0$. A Luenberger observer (Kalman-like filter) for estimating $\xi(t)\in\mathbb{R}^n$ is governed by

$$d\xi = (A\xi + b)dt + K\left( dZ - (C\xi + d)dt \right),$$

where $C=2DJM$, $d=2DJN$, and $K$ is the observer gain. The estimation error $e = X-\xi$ has zero mean, and the error covariance $P = \Re\,\mathbb{E}[ee^T]$ satisfies

$$\dot P = (A-KC)P + P(A-KC)^T + \Sigma(\mu(t)) - (PC^T+B(\mu) D^T)F^{-1}(CP + D B(\mu)^T),$$

with $\Sigma(\mu)$ explicitly determined. Optimizing the gain yields the time-varying Kalman gain: $K_*(t) = (P C^T + B(\mu) D^T) F^{-1},$ with $P$ evolving via a Riccati ODE. The steady-state solution is obtained by solving the algebraic Riccati equation for $P_\infty$ and $K_\infty$.

7. Stability and Convergence Criteria

The system’s mean vector dynamics $\dot\mu = A\mu + b$ converge to $\mu_\infty$ if and only if $A$ is Hurwitz. For the important subclass corresponding to $n=3$ and Pauli operator algebra, $A$ is explicitly

$$A = 2\Theta(E + M^T J N) + 2\sum_{\ell=1}^{3} \Theta_\ell M^T M \Theta_\ell,$$

and, crucially, whenever $\mathrm{rank}\,M \geq 2$, the symmetric part $A + A^T \prec 0$, guaranteeing Hurwitz stability. Under these conditions, the observer Riccati equation admits a stabilizing solution $P_\infty$, and the closed-loop matrix $A-K_\infty C$ remains Hurwitz, ensuring convergence of the observer error.

In summary, the quasi-linear structural features—namely affine drift, linear dispersion, and finite-dimensional operator algebra—enable explicit closed-form (or semi-closed-form) characterization of the system’s statistical dynamics, invariant state, quadratic cost rates, and the synthesis of mean-square optimal quantum observers that reproduce the classical Kalman filter structure in the quantum setting (Vladimirov et al., 2020).