Quantum Kalman Decomposition

• Quantum Kalman Decomposition is a canonical framework that partitions linear quantum systems into controllable, observable, and uncontrollable-unobservable subsystems using physically admissible transformations.
• It reveals key quantum resource modes, including decoherence-free subsystems and quantum-nondemolition variables, which are essential for robust quantum control and measurement.
• Implementation methods such as one-sided symplectic SVD, Gramian-based approaches, and Gram–Schmidt constructions enable systematic block-diagonalization of system matrices.

The quantum Kalman decomposition is a structural canonical form for linear quantum systems that partitions the system’s degrees of freedom in terms of controllability and observability, under the constraints imposed by quantum mechanics. Unlike the classical Kalman decomposition, the quantum variant requires coordinate transformations that are physically allowable—specifically, Bogoliubov transformations in the doubled-up (annihilation/creation) representation or real symplectic transformations in the real quadrature operator representation. This decomposition exposes the system’s fundamental subsystems: those that are controllable and observable, those that are only controllable or only observable, and those that are neither. Quantum Kalman decomposition thereby provides a unifying framework connecting classical linear systems theory with quantum information-theoretic resources such as decoherence-free subsystems and quantum-nondemolition variables (Zhang et al., 2016, Grivopoulos et al., 2016, Zhang et al., 2023).

1. Mathematical Formulation and Quantum Constraints

A general linear quantum system consisting of $n$ quantum harmonic oscillators coupled to $m$ input bosonic fields is described by quantum stochastic differential equations (QSDEs):

$$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$

where $\breve a = (a, a^\#)^\top$ is the doubled-up vector of annihilation and creation operators, and

$$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$

with $(\cdot)^\flat$ denoting the $J$-weighted adjoint. The admissible coordinate transformations $T$ on $\breve a$ are restricted to those preserving canonical commutation relations:

$T\,J_n\,T^\dagger = J_n.$

Such $m$0 are called Bogoliubov transformations. In the real quadrature representation $m$1, only real symplectic transformations $m$2 with $m$3 are permitted (Zhang et al., 2016).

2. Invariant Subspace Structure: Fourfold Partition

Analogous to the classical case, the system’s state space decomposes into four invariant subspaces, determined by controllability and observability:

• $m$4: controllable and observable
• $m$5: controllable but unobservable
• $m$6: observable but uncontrollable
• $m$7: neither controllable nor observable

These subspaces are defined via the controllability and observability matrices (or their Gramian analogs):

• Controllability matrix: $m$8
• Observability matrix: $m$9

The column space of $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$0 gives the controllable subspace; the kernel of $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$1 gives the unobservable subspace. The four subspaces are characterized by intersections and complements of these spaces, with quantum constraints requiring, for instance, that $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$2 and $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$3 appear in conjugate pairs (Zhang et al., 2016, Grivopoulos et al., 2016, Zhang et al., 2023).

3. Canonical Kalman Form for Quantum Systems

A central result is that, for any linear quantum system, there exists a physically admissible (symplectic or Bogoliubov) transformation that block-diagonalizes the system matrices into Kalman canonical form:

For $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$4 in the real quadrature frame, the coordinate transformation $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$5 ($$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$6 symplectic) yields:

• $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$7, partitioned according to the dimensions of the respective subspaces.
• The system matrices $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$8 acquire a block structure, with each block associated to one of the four subspaces. Explicitly, the modes with both controllability and observability couple to both input and output; those with only controllability or observability couple only in the corresponding direction; the uncontrollable-unobservable modes are dynamically isolated (Grivopoulos et al., 2016, Zhang et al., 2016, Zhang et al., 2023).

The canonical form of the $$\dot{\breve a}(t) = \mathbb{A} \, \breve a(t) + \mathbb{B}\, \breve b(t),\qquad \breve b_{\rm out}(t) = \mathbb{C}\, \breve a(t) + \mathbb{D}\, \breve b(t)$$9-matrix, for instance, is:

$\breve a = (a, a^\#)^\top$0

Off-diagonal and zero blocks clarify the selective coupling between subsystems (Grivopoulos et al., 2016).

4. Explicit Construction Methods

Three core methodologies for effecting the decomposition have been developed:

1. One-Sided Symplectic SVD of the Observability Matrix: Given a real quadrature LQSS, a one-sided symplectic SVD (Xu–Long–Pope 2003) is applied to the observability matrix $\breve a = (a, a^\#)^\top$1. The symplectic matrix $\breve a = (a, a^\#)^\top$2 then yields the required change of coordinates $\breve a = (a, a^\#)^\top$3 (Grivopoulos et al., 2016).
2. Direct Orthogonal-Symplectic Gramian Approach: The controllability and observability Gramians, $\breve a = (a, a^\#)^\top$4 and $\breve a = (a, a^\#)^\top$5, are block-diagonalized using real orthogonal, block-symplectic matrices constructed from their image and kernel spaces. This method yields a concrete, algorithmic synthesis of the Kalman decomposition in quadrature coordinates (Zhang et al., 2023).
3. Gram–Schmidt Construction in the Doubling Picture: For the doubled-up (Bogliubov) representation, the Kalman transformation matrix $\breve a = (a, a^\#)^\top$6 is constructed to ensure that its columns span bases of the four invariant subspaces, with $\breve a = (a, a^\#)^\top$7 and $\breve a = (a, a^\#)^\top$8 paired as required by quantum constraints (Zhang et al., 2016).

Each method yields the same canonical structure and is illustrated by explicit computation for prototype systems.

5. Quantum Resource Modes and Physical Implications

The decomposition transparently identifies quantum subsystems of operational significance:

• Decoherence-Free Subsystems (DFS): The $\breve a = (a, a^\#)^\top$9 modes are decoupled both from control and observation, representing DFSs that evolve autonomously and do not couple to environmental noise or measurement (Zhang et al., 2016, Grivopoulos et al., 2016).
• Quantum-Nondemolition (QND) Variables: Each $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$0 mode, by its construction, commutes with itself at all times and can be measured without back-action disturbance on its conjugate partner. All $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$1 observables together form a quantum-mechanics-free subsystem (QMFS), useful in quantum metrology and measurement (Zhang et al., 2016).
• Back-Action-Evading (BAE) Measurements: The structure of the canonical form reveals subsystems (notably $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$2/$$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$3 conjugate pairs) where BAE measurements are possible: certain input quadratures can be inferred from output without ever feeding back to disturb conjugate variables (Zhang et al., 2016, Grivopoulos et al., 2016).

These correspondences provide a systematic way to engineer and certify quantum resource subsystems in practical applications.

6. Illustrative Examples

Typical case studies include:

• Three-Mode Optomechanical System: For $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$4, with Hamiltonian $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$5 and coupling $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$6, the Kalman transformation identifies one co mode, one $$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$7/$$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$8 conjugate pair, and one DFS ($$\mathbb{A} = -i\,J_n\,\Omega - \frac12\,\mathbb{C}^\flat\,\mathbb{C}, \quad \mathbb{B} = -\mathbb{C}^\flat, \quad \mathbb{C}=\Delta(C_-, C_+), \quad \mathbb{D}=I_{2m}$$9). The specific transformation yields

$(\cdot)^\flat$0

which directly reveal the system’s modal structure (Grivopoulos et al., 2016, Zhang et al., 2023).

• Passive Two-Mode Optical System: With $(\cdot)^\flat$1 and $(\cdot)^\flat$2, the passive system reduces to a DFS (dark mode) and a co mode, with the absence of $(\cdot)^\flat$3/$(\cdot)^\flat$4 pairs (Zhang et al., 2016).

Such explicit examples demonstrate the practical computation of the decomposition and the elucidation of quantum resource modes.

7. Significance and Extensions

The quantum Kalman decomposition offers a rigorous canonical analysis tool for quantum control, quantum measurement theory, and quantum information processing. Its ability to reveal DFSs, QND variables, QMFSs, and BAE structures unifies classical systems analysis with fundamental quantum resources. The framework is applicable to both passive and active systems and applies to physical implementations from circuit QED to optomechanics. The methodology has been extended from the original construction (Zhang et al., 2016, Grivopoulos et al., 2016) to recent Gramian-based algorithms (Zhang et al., 2023), providing concrete paths for automated decomposition and resource certification in high-dimensional systems.