Introduction
A quantum finite-level system has an underlying finite-dimensional Hilbert space. Its states are positive semidefinite, unit-trace density matrices, and its observables are Hermitian matrices. This class of quantum systems includes qubits and qudits, finite spin networks, and effective multilevel descriptions of atoms, quantum dots, and superconducting artificial atoms. Finite-level systems provide basic carriers of quantum information and serve as controllable components of quantum memories, quantum processors, and quantum sensors [1 , 2 , 3 ] . Their usefulness depends on our ability to engineer selected degrees of freedom, protect stored information against environmental disturbances, and extract information without severely perturbing the quantities of interest.
In quantum computing and metrology, a prominent goal is to identify and protect degrees of freedom immune to specified environmental noise. In the Heisenberg picture, a decoherence-free subsystem (DFS) comprises degrees of freedom neither driven by the environmental inputs nor detectable through the outputs [4 , 5 , 6 ] . It need not be isolated from other system components, provided their interactions preserve these properties [5 Fig. 2] . This notion is closely related to the concept of decoherence-free subspace in the Schrödinger picture [7 Sec. III-C] , which play a crucial role in protecting quantum information for quantum computation [8 , 9 , 2 , 10 , 11 , 12 ] .
Whereas a DFS protects internal degrees of freedom, quantum back-action-evading (BAE) measurement concerns selected input–output relations. An output field observable is BAE with respect to specified input field observables when their influence on the measured output observable vanishes. This excludes the associated back-action noise from the readout without requiring the entire system to remain undisturbed. Such measurements can enable sensing beyond the standard quantum limit (SQL) [13 , 14 , 4 , 15 , 5 , 16 , 17 ] . Quantum BAE measurements also provide a basis for measurement-based state preparation and feedback stabilization [18 , 14 , 19 , 20 ] .
A related but distinct concept is a quantum nondemolition (QND) observable: a system observable that admits repeated measurements without measurement back-action perturbing its subsequent evolution. Specifically, in the Heisenberg picture, an observable O O O is called a QND observable if [ O ( t ) , O ( s ) ] = 0 [O(t), O(s)] = 0 [ O ( t ) , O ( s )] = 0 for all t , s t, s t , s . Measurement back-action may disturb noncommuting observables, but that disturbance must not feed back into O O O [13 , 14 , 4 , 5 , 16 ] .
Structural decomposition provides a natural approach to visualize all of these important notions. Indeed, in the quantum Kalman canonical form developed in [5 , 6 , 21 , 22 ] for linear quantum systems, a DFS corresponds to the uncontrollable/unobservable subsystem, QND variables corresponds to the uncontrollable/observable subsystem [23 ] , and quantum BAE measurements is characterized by a zero transfer matrix from a certain input to a certain output. Extending these results in linear quantum systems to finite-level quantum systems is challenging, mainly because a finite-level quantum system is modeled in the Heisenberg picture by bilinear quantum stochastic differential equations (QSDEs) where input field increments multiply system observables. Moreover, the multiplication table of the observables involves both the antisymmetric structure tensor, which determines commutators,
This work was partially supported by Hong Kong Research Grant Council (RGC) under Grant No. 15213924 and the Australian Research Council under grant DP240101494.
J. Gu is with the Department of Data Science and Artificial Intelligence, and G. Zhang is with the Department of Applied Mathematics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong (e-mail: joanna.gu@connect.polyu.hk; guofeng.zhang@polyu.edu.hk).
I. R. Petersen is with the School of Engineering, The Australian National University, Canberra, ACT 2601, Australia (e-mail: i.r.petersen@gmail.com). and the symmetric structure tensor, which determines the traceless part of anticommutators [24 ] . Thus, a finite-level quantum system often has more complex dynamics in contrast to a linear one, imposing significant difficulties in system analysis and synthesis.
Structural decomposition for quantum two-level systems has been studied in [25 ] , where DFS, QND variables and quantum BAE measurements have been identified. It should be noted that in the Schrödinger picture, decoherence-free subspaces and noiseless subsystems protect quantum information [8 ] , [9 ] , [2 ] , [26 ] , their dynamical invariance and control properties have been investigated [10 ] , [11 ] ; and model reduction for open quantum finite-level dynamics have recently been developed [27 ] , [28 ] , [29 ] , [30 ] ; see also [31 ] , [32 ] , [33 ] , [34 ] , [35 ] , [36 ] for model reduction of quantum filters.
The purpose of this paper is to conduct a comprehensive study of DFSs, QND variables and quantum BAE measurements of finite-level quantum systems. The main contributions are summarized as follows.
We characterize the real coordinate transformations (Definition 1) that preserve generalized Gell–Mann structure tensors and the physical realizability of finite-level quantum systems (Lemma 2). These transformations are exactly the adjoint representations of unitary conjugations (Theorem 1).
We introduce strongly aligned subspaces (Definition 2), whose associated unital C ∗ C^* C ∗ -algebras admit an exact orthogonal separation into central and noncentral Hermitian components (Lemma 4); based on which we apply the Artin–Wedderburn decomposition to identify a set of system observables whose evolution is decoupled from the input and output fields (Theorem 2).
We formulate relative admissibility (Definition 4) between the generalized Gell–Mann basis λ \lambda λ and another orthonormal Hermitian basis η \eta η of finite-level quantum systems (Lemma 10). Then in Theorem 3 we construct a DFS under the η \eta η -basis, and in Corollary 1 a DFS under the generalized Gell–Mann basis.
We derive QND variables in the sense of linear systems theory in Theorem 4, and further distinguish them in terms of spectral separation of the QND algebra resulting from the output-signal observables (Theorem 5).
We derive a sufficient condition for BAE measurements in finite-level quantum systems in Theorem 6.
A four-qubit nearest-neighbor Heisenberg chain with collective dephasing illustrates the main results.
The remainder of the paper is organized as follows. Section II presents the finite-level quantum systems under consideration. Section III develops admissible transformations, strong alignment, and structural decomposition results. Sections IV, V and VI study DFSs, QND observables and quantum BAE measurements, respectively. Section VII presents a four-qubit illustrative example, and the final section concludes the paper.
Notation. i = − 1 i=\sqrt{-1} i = − 1 . Let R / C \mathbb{R}/\mathbb{C} R / C be the set of real/complex numbers. I n I_n I n is the identity matrix of size n n n . The subscript “n n n ” is often omitted. Let J J J denote the symplectic matrix J = [ 0 I − I 0 ] J=\begin{bmatrix}0&I\\-I&0\end{bmatrix} J = [ 0 − I I 0 ] . Given a matrix A A A , denote its k k k -th row by A k A_k A k or ( A ) k (A)_k ( A ) k , and its transpose, entrywise conjugation, and conjugate transpose by A T A^\mathsf{T} A T , A # A^\# A # , A † A^\dagger A † respectively. Range ( A ) \operatorname{Range}(A) Range ( A ) is the column space and Ker ( A ) \operatorname{Ker}(A) Ker ( A ) is the nullspace of A A A . The commutator of two operators a , b a,b a , b is [ a , b ] = a b − b a [a,b]=ab-ba [ a , b ] = ab − ba , while their anticommutator is { a , b } = a b + b a \{a,b\}=ab+ba { a , b } = ab + ba . The reduced Planck constant ℏ \hbar ℏ is set to 1.
Preliminaries
Let M n ( C ) \mathcal{M}_n(\mathbb{C}) M n ( C ) denote the vector space of all n × n n\times n n × n complex matrices. Denote by S U ( n ) SU(n) S U ( n ) the special unitary group of degree n n n . The Lie algebra s u ( n ) \mathfrak{su}(n) su ( n ) of S U ( n ) SU(n) S U ( n ) is usually written mathematically as the vector space of traceless skew-Hermitian matrices. However, in quantum physics it is often more convenient to use the equivalent traceless Hermitian generator convention [37 ] , [24 ] . Let e 1 , … , e n e_1,\ldots,e_n e 1 , … , e n be the standard basis of C n \mathbb{C}^n C n , and define matrices P j k = e j e k T P_{jk}=e_je_k^\mathsf{T} P j k = e j e k T . A standard traceless Hermitian basis of the Lie algebra s u ( n ) \mathfrak{su}(n) su ( n ) is given by
u j k = P j k + P k j , v j k = i ( P j k − P k j ) , 1 ≤ j < k ≤ n , u_{jk}=P_{jk}+P_{kj},\qquad v_{jk}=i(P_{jk}-P_{kj}),\qquad1\leq j<k\leq n, u j k = P j k + P k j , v j k = i ( P j k − P k j ) , 1 ≤ j < k ≤ n ,
w ℓ = 2 ℓ ( ℓ + 1 ) ( ∑ q = 1 ℓ P q q − ℓ P ℓ + 1 , ℓ + 1 ) , 1 ≤ ℓ ≤ n − 1. w_\ell=\sqrt{\frac{2}{\ell(\ell+1)}}\left(\sum_{q=1}^{\ell}P_{qq}-\ell P_{\ell+1,\ell+1}\right),\qquad1\leq\ell\leq n-1. w ℓ = ℓ ( ℓ + 1 ) 2 ( q = 1 ∑ ℓ P q q − ℓ P ℓ + 1 , ℓ + 1 ) , 1 ≤ ℓ ≤ n − 1.
These s = n 2 − 1 s=n^2-1 s = n 2 − 1 matrices form the generalized Gell-Mann basis of the vector space of traceless Hermitian matrices
Herm 0 ( n ) ≜ { A ∈ M n ( C ) : A = A † , Tr ( A ) = 0 } . \operatorname{Herm}_0(n)\triangleq\{A\in\mathcal{M}_n(\mathbb{C}):A=A^\dagger,\operatorname{Tr}(A)=0\}. Herm 0 ( n ) ≜ { A ∈ M n ( C ) : A = A † , Tr ( A ) = 0 } .
Without particular order, we label these traceless Hermitian generators as λ 1 , … , λ s \lambda_1,\ldots,\lambda_s λ 1 , … , λ s , which by construction satisfy
Tr ( λ j λ k ) = 2 δ j k . (1) \operatorname{Tr}(\lambda_j\lambda_k)=2\delta_{jk}. \tag*{(1)} Tr ( λ j λ k ) = 2 δ j k . (1)
and whose commutation and anticommutation relations are
[ λ a , λ b ] = 2 i ∑ c = 1 s f a b c λ c . (2) [\lambda_a,\lambda_b]=2i\sum_{c=1}^{s}f_{abc}\lambda_c. \tag*{(2)} [ λ a , λ b ] = 2 i c = 1 ∑ s f ab c λ c . (2)
{ λ a , λ b } = 4 n δ a b I n + 2 ∑ c = 1 s d a b c λ c . (3) \{\lambda_a,\lambda_b\}=\frac{4}{n}\delta_{ab}I_n+2\sum_{c=1}^{s}d_{abc}\lambda_c. \tag*{(3)} { λ a , λ b } = n 4 δ ab I n + 2 c = 1 ∑ s d ab c λ c . (3)
From (2), we have
λ a λ b = 2 n δ a b I n + ∑ c = 1 s ( i f a b c + d a b c ) λ c . (4) \lambda_a\lambda_b=\frac{2}{n}\delta_{ab}I_n+\sum_{c=1}^{s}(if_{abc}+d_{abc})\lambda_c. \tag*{(4)} λ a λ b = n 2 δ ab I n + c = 1 ∑ s ( i f ab c + d ab c ) λ c . (4)
The 3rd-order real tensor f a b c f_{abc} f ab c is completely antisymmetric, whereas d a b c d_{abc} d ab c completely symmetric:
f a b c = − f b a c = f b c a , d a b c = d b a c = d b c a . (5) f_{abc}=-f_{bac}=f_{bca},\qquad d_{abc}=d_{bac}=d_{bca}. \tag*{(5)} f ab c = − f ba c = f b c a , d ab c = d ba c = d b c a . (5)
More details can be found [24 Section 3] .
For each a = 1 , … , s a=1,\ldots,s a = 1 , … , s , we define antisymmetric matrices F a F_a F a and symmetric matrices D a ∈ R s × s D_a\in\mathbb{R}^{s\times s} D a ∈ R s × s by
( F a ) b c = f a b c , ( D a ) b c = d a b c , (6) (F_a)_{bc}=f_{abc},\qquad(D_a)_{bc}=d_{abc}, \tag*{(6)} ( F a ) b c = f ab c , ( D a ) b c = d ab c , (6)
based on which define linear maps Θ n − , Θ n + : C s → C s × s \Theta_n^- ,\Theta_n^+:\mathbb{C}^s\to\mathbb{C}^{s\times s} Θ n − , Θ n + : C s → C s × s as
Θ n − ( β ) ≜ [ F 1 T β ⋯ F s T β ] , (7) \Theta_n^-(\beta)\triangleq\begin{bmatrix}F_1^\mathsf{T}\beta&\cdots&F_s^\mathsf{T}\beta\end{bmatrix}, \tag*{(7)} Θ n − ( β ) ≜ [ F 1 T β ⋯ F s T β ] , (7)
Θ n + ( β ) ≜ [ D 1 β ⋯ D s β ] . (8) \Theta_n^+(\beta)\triangleq\begin{bmatrix}D_1\beta&\cdots&D_s\beta\end{bmatrix}. \tag*{(8)} Θ n + ( β ) ≜ [ D 1 β ⋯ D s β ] . (8)
When β \beta β is an s s s -dimensional row vector, then it is understood that Θ n − ( β ) = Θ n − ( β T ) \Theta_n^-(\beta)=\Theta_n^-(\beta^\mathsf{T}) Θ n − ( β ) = Θ n − ( β T ) and Θ n + ( β ) = Θ n + ( β T ) \Theta_n^+(\beta)=\Theta_n^+(\beta^\mathsf{T}) Θ n + ( β ) = Θ n + ( β T ) .
Lemma 1: For every β ∈ C s \beta\in\mathbb{C}^s β ∈ C s ,
Θ n − ( β ) = ∑ j = 1 s β j F j , Θ n + ( β ) = ∑ j = 1 s β j D j . (9) \Theta_n^-(\beta)=\sum_{j=1}^{s}\beta_jF_j,\qquad\Theta_n^+(\beta)=\sum_{j=1}^{s}\beta_jD_j. \tag*{(9)} Θ n − ( β ) = j = 1 ∑ s β j F j , Θ n + ( β ) = j = 1 ∑ s β j D j . (9)
Proof: We prove the first identity by comparing matrix entries. By definition, the ( k , ℓ ) (k,\ell) ( k , ℓ ) -th entry of Θ n − ( β ) \Theta_n^{-}(\beta) Θ n − ( β ) is
( Θ n − ( β ) ) k ℓ = ∑ j = 1 s ( F ℓ ) j k β j = ∑ j = 1 s f ℓ j k β j = ( ∑ j = 1 s β j F j ) k ℓ , \left(\Theta_n^{-}(\beta)\right)_{k\ell}=\sum_{j=1}^{s}(F_\ell)_{jk}\beta_j=\sum_{j=1}^{s}f_{\ell jk}\beta_j=\left(\sum_{j=1}^{s}\beta_jF_j\right)_{k\ell}, ( Θ n − ( β ) ) k ℓ = j = 1 ∑ s ( F ℓ ) j k β j = j = 1 ∑ s f ℓ j k β j = ( j = 1 ∑ s β j F j ) k ℓ ,
where the complete antisymmetry of f i j k f_{ijk} f ij k in (5) has been used. Hence Θ n − ( β ) = ∑ j = 1 s β j F j \Theta_n^{-}(\beta)=\sum_{j=1}^{s}\beta_jF_j Θ n − ( β ) = ∑ j = 1 s β j F j . The proof for Θ n + \Theta_n^{+} Θ n + is identical, using the complete symmetry of d i j k d_{ijk} d ij k .
Define the normalized Hilbert–Schmidt inner product
⟨ X , Y ⟩ n H S ≜ 1 2 Tr ( X † Y ) , X , Y ∈ C n × n . (10) \langle X,Y\rangle_{\mathrm{nHS}}\triangleq\frac{1}{2}\operatorname{Tr}(X^\dagger Y),\quad X,Y\in\mathbb{C}^{n\times n}. \tag*{(10)} ⟨ X , Y ⟩ nHS ≜ 2 1 Tr ( X † Y ) , X , Y ∈ C n × n . (10)
Thus by (1) , the generalized Gell–Mann matrices { λ 1 , … , λ s } \{\lambda_1,\ldots,\lambda_s\} { λ 1 , … , λ s } form an orthonormal basis of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) , under the normalized Hilbert–Schmidt inner product.
Following [24 ] , for an n n n -level open quantum system, we identify the vector of spin observables X = ( X 1 , … , X s ) ⊤ X=(X_1,\ldots,X_s)^\top X = ( X 1 , … , X s ) ⊤ with ( λ 1 , … , λ s ) ⊤ (\lambda_1,\ldots,\lambda_s)^\top ( λ 1 , … , λ s ) ⊤ . Then by Lemma 1, Eq. (2) translates to
[ X , X ⊤ ] ≜ X X ⊤ − ( X X ⊤ ) ⊤ = 2 i Θ n − ( X ) , (11) [X,X^\top]\triangleq XX^\top-(XX^\top)^\top=2i\Theta_n^{-}(X), \tag*{(11)} [ X , X ⊤ ] ≜ X X ⊤ − ( X X ⊤ ) ⊤ = 2 i Θ n − ( X ) , (11)
{ X , X ⊤ } ≜ X X ⊤ + ( X X ⊤ ) ⊤ = 4 n I s + 2 Θ n + ( X ) . (12) \{X,X^\top\}\triangleq XX^\top+(XX^\top)^\top=\frac{4}{n}I_s+2\Theta_n^{+}(X). \tag*{(12)} { X , X ⊤ } ≜ X X ⊤ + ( X X ⊤ ) ⊤ = n 4 I s + 2 Θ n + ( X ) . (12)
Remark 1: In (11) , X X X is understood as a vector of operators. Like (7) , Θ n − ( X ) = [ F 1 ⊤ X ⋯ F s ⊤ X ] ⊤ \Theta_n^{-}(X)=[F_1^\top X\ \cdots\ F_s^\top X]^\top Θ n − ( X ) = [ F 1 ⊤ X ⋯ F s ⊤ X ] ⊤ . Hence, by Lemma 1, Θ n − ( X ) = ∑ j = 1 s λ j F j \Theta_n^{-}(X)=\sum_{j=1}^{s}\lambda_jF_j Θ n − ( X ) = ∑ j = 1 s λ j F j . If X X X is understood as a vector of matrices as each λ j ∈ C n × n \lambda_j\in\mathbb{C}^{n\times n} λ j ∈ C n × n , then the above Θ n − ( X ) \Theta_n^{-}(X) Θ n − ( X ) should be understood as Θ n − ( X ) = [ ( F 1 ⊤ ⊗ I n ) X ⋯ ( F s ⊤ ⊗ I n ) X ] \Theta_n^{-}(X)=[(F_1^\top\otimes I_n)X\ \cdots\ (F_s^\top\otimes I_n)X] Θ n − ( X ) = [( F 1 ⊤ ⊗ I n ) X ⋯ ( F s ⊤ ⊗ I n ) X ] . The same is true for (12) .
The evolution of the spin observables X X X of an n n n -level quantum system driven by n w n_w n w Bosonic input fields can be described in the Heisenberg picture by bilinear QSDEs:
d X = A 0 d t + A X d t + ∑ k = 1 n w B 1 k X d W 1 k + ∑ k = 1 n w B 2 k X d W 2 k , dX=A_0\,dt+AX\,dt+\sum_{k=1}^{n_w}B_{1k}X\,dW_{1k}+\sum_{k=1}^{n_w}B_{2k}X\,dW_{2k}, d X = A 0 d t + A X d t + k = 1 ∑ n w B 1 k X d W 1 k + k = 1 ∑ n w B 2 k X d W 2 k ,
\begin{bmatrix}dY_1\\dY_2\end{bmatrix}
=
\begin{bmatrix}C_1\\C_2\end{bmatrix}X\,dt+
\begin{bmatrix}dW_1\\dW_2\end{bmatrix}. \tag*{(13)}\end{bmatrix}
where matrices A 0 ∈ R s A_0\in\mathbb{R}^{s} A 0 ∈ R s , A , B 1 k , B 2 k ∈ R s × s A,B_{1k},B_{2k}\in\mathbb{R}^{s\times s} A , B 1 k , B 2 k ∈ R s × s and C 1 , C 2 ∈ R n w × s C_1,C_2\in\mathbb{R}^{n_w\times s} C 1 , C 2 ∈ R n w × s , while the column vectors of input operators and output operators are
W j = [ W j 1 ⋮ W j n w ] , Y j = [ Y j 1 ⋮ Y j n w ] , j = 1 , 2. W_j=\begin{bmatrix}W_{j1}\\ \vdots\\ W_{jn_w}\end{bmatrix},\qquad
Y_j=\begin{bmatrix}Y_{j1}\\ \vdots\\ Y_{jn_w}\end{bmatrix},\qquad j=1,2. W j = W j 1 ⋮ W j n w , Y j = Y j 1 ⋮ Y j n w , j = 1 , 2.
Using the compact notation W = [ W 1 ⊤ W 2 ⊤ ] ⊤ \mathcal{W}=[W_1^\top\ W_2^\top]^\top W = [ W 1 ⊤ W 2 ⊤ ] ⊤ , Y = [ Y 1 ⊤ Y 2 ⊤ ] ⊤ Y=[Y_1^\top\ Y_2^\top]^\top Y = [ Y 1 ⊤ Y 2 ⊤ ] ⊤ , and
C = [ C 1 C 2 ] , B μ = { B 1 μ , 1 ≤ μ ≤ n w , B 2 , μ − n w , n w < μ ≤ 2 n w = m , C=\begin{bmatrix}C_1\\C_2\end{bmatrix},\qquad
B_\mu=
\begin{cases}
B_{1\mu},&1\leq\mu\leq n_w,\\
B_{2,\mu-n_w},&n_w<\mu\leq2n_w=m,
\end{cases} C = [ C 1 C 2 ] , B μ = { B 1 μ , B 2 , μ − n w , 1 ≤ μ ≤ n w , n w < μ ≤ 2 n w = m ,
the system (13) can be rewritten as
d X = ( A 0 + A X ) d t + ∑ μ = 1 m B μ X d W μ , dX=(A_0+AX)\,dt+\sum_{\mu=1}^{m}B_\mu X\,dW_\mu, d X = ( A 0 + A X ) d t + μ = 1 ∑ m B μ X d W μ ,
d Y = C X d t + d W . (14) dY=CX\,dt+d\mathcal{W}. \tag*{(14)} d Y = C X d t + d W . (14)
The Itô matrix for the input field quadratures is
d W d W ⊤ = Ω d t , Ω ≜ [ I n w i I n w − i I n w I n w ] ; (15) d\mathcal{W}\,d\mathcal{W}^\top=\Omega\,dt,\qquad
\Omega\triangleq
\begin{bmatrix}
I_{n_w}&iI_{n_w}\\
-iI_{n_w}&I_{n_w}
\end{bmatrix}; \tag*{(15)} d W d W ⊤ = Ω d t , Ω ≜ [ I n w − i I n w i I n w I n w ] ; (15)
i.e., the input fields are initialized in the vacuum state.
Structural decomposition of quantum finite-level systems
In this section, we develop a structure-preserving decomposition theory for the open n n n -level quantum system (14) .
Admissible maps
In this subsection, we establish conditions for coordinate transformations on the quantum finite-level system (14) , under which the commutation and anticommutation relations are preserved by the transformed system observables.
Lemma 2: Given T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s , the vector of transformed observables X ~ = T ⊤ X \widetilde{X}=T^\top X X = T ⊤ X have commutation and anticommutation relations
[ X ~ , X ~ ⊤ ] = 2 i Θ n − ( X ~ ) , (16) [\widetilde{X},\widetilde{X}^\top]=2i\Theta_n^{-}(\widetilde{X}), \tag*{(16)} [ X , X ⊤ ] = 2 i Θ n − ( X ) , (16)
and
{ X ~ , X ~ ⊤ } = 4 n I s + 2 Θ n + ( X ~ ) , (17) \{\widetilde{X},\widetilde{X}^\top\}=\frac{4}{n}I_s+2\Theta_n^{+}(\widetilde{X}), \tag*{(17)} { X , X ⊤ } = n 4 I s + 2 Θ n + ( X ) , (17)
if and only if the transformation matrix T T T satisfies
T ⊤ T = I s . (18) T^\top T=I_s. \tag*{(18)} T ⊤ T = I s . (18)
and for every β ∈ C s \beta\in\mathbb{C}^{s} β ∈ C s ,
Θ n − ( T T β ) = T T Θ n − ( β ) T , (19) \Theta_{n}^{-}(T^{\mathsf{T}}\beta)=T^{\mathsf{T}}\Theta_{n}^{-}(\beta)T, \tag*{(19)} Θ n − ( T T β ) = T T Θ n − ( β ) T , (19)
Θ n + ( T T β ) = T T Θ n + ( β ) T . (20) \Theta_{n}^{+}(T^{\mathsf{T}}\beta)=T^{\mathsf{T}}\Theta_{n}^{+}(\beta)T. \tag*{(20)} Θ n + ( T T β ) = T T Θ n + ( β ) T . (20)
Proof: Partition T T T as T = [ T 1 , … , T s ] T=[T_{1},\ldots,T_{s}] T = [ T 1 , … , T s ] . Then T T β = ∑ j T j k β j T^{\mathsf{T}}\beta=\sum_{j}T_{jk}\beta_{j} T T β = ∑ j T j k β j . Substitution into (7) gives
Θ n − ( T T β ) = ∑ k ( T k T β ) F k = ∑ j β j ∑ k T j k F k . \Theta_{n}^{-}(T^{\mathsf{T}}\beta)=\sum_{k}(T_{k}^{\mathsf{T}}\beta)F_{k}=\sum_{j}\beta_{j}\sum_{k}T_{jk}F_{k}. Θ n − ( T T β ) = k ∑ ( T k T β ) F k = j ∑ β j k ∑ T j k F k .
On the other hand, again by (7), T T Θ n − ( β ) T = ∑ j β j T T F j T T^{\mathsf{T}}\Theta_{n}^{-}(\beta)T=\sum_{j}\beta_{j}T^{\mathsf{T}}F_{j}T T T Θ n − ( β ) T = ∑ j β j T T F j T . As β ∈ C s \beta\in\mathbb{C}^{s} β ∈ C s is arbitrary, (19) is equivalent to
T T F j T = ∑ k = 1 s T j k F k , j = 1 , … , s . (21) T^{\mathsf{T}}F_{j}T=\sum_{k=1}^{s}T_{jk}F_{k},\qquad j=1,\ldots,s. \tag*{(21)} T T F j T = k = 1 ∑ s T j k F k , j = 1 , … , s . (21)
Similarly, it can be easily shown that (20b) is equivalent to
T T D j T = ∑ k = 1 s T j k D k , j = 1 , … , s . (22) T^{\mathsf{T}}D_{j}T=\sum_{k=1}^{s}T_{jk}D_{k},\qquad j=1,\ldots,s. \tag*{(22)} T T D j T = k = 1 ∑ s T j k D k , j = 1 , … , s . (22)
From X ~ = T T X \widetilde{X}=T^{\mathsf{T}}X X = T T X , by (9a) we have [ X ~ , X ~ T ] = T T [ X , X T ] T = 2 i T T Θ n − ( X ) T [\widetilde{X},\widetilde{X}^{\mathsf{T}}]=T^{\mathsf{T}}[X,X^{\mathsf{T}}]T=2\mathrm{i}T^{\mathsf{T}}\Theta_{n}^{-}(X)T [ X , X T ] = T T [ X , X T ] T = 2 i T T Θ n − ( X ) T . As a result, (16) is equivalent to T T Θ n − ( X ) T = Θ n − ( T T X ) T^{\mathsf{T}}\Theta_{n}^{-}(X)T=\Theta_{n}^{-}(T^{\mathsf{T}}X) T T Θ n − ( X ) T = Θ n − ( T T X ) , which is equivalent to
∑ j = 1 s λ j T T F j T = ∑ j = 1 s λ j ∑ k = 1 s T j k F k . (23) \sum_{j=1}^{s}\lambda_{j}T^{\mathsf{T}}F_{j}T=\sum_{j=1}^{s}\lambda_{j}\sum_{k=1}^{s}T_{jk}F_{k}. \tag*{(23)} j = 1 ∑ s λ j T T F j T = j = 1 ∑ s λ j k = 1 ∑ s T j k F k . (23)
Because λ 1 , … , λ s \lambda_{1},\ldots,\lambda_{s} λ 1 , … , λ s are linearly independent, (23) is equivalent to (21), while the latter is equivalent to (20a), as has been established. Similarly, notice that by (9b), { X ~ , X ~ T } = T T { X , X T } T = 4 n T T T + 2 T T Θ n + ( X ) T \{\widetilde{X},\widetilde{X}^{\mathsf{T}}\}=T^{\mathsf{T}}\{X,X^{\mathsf{T}}\}T=\frac{4}{n}T^{\mathsf{T}}T+2T^{\mathsf{T}}\Theta_{n}^{+}(X)T { X , X T } = T T { X , X T } T = n 4 T T T + 2 T T Θ n + ( X ) T . Hence, (17) is equivalent to 4 n I s + 2 Θ n + ( T T X ) = 4 n T T T + 2 T T Θ n + ( X ) T \frac{4}{n}I_{s}+2\Theta_{n}^{+}(T^{\mathsf{T}}X)=\frac{4}{n}T^{\mathsf{T}}T+2T^{\mathsf{T}}\Theta_{n}^{+}(X)T n 4 I s + 2 Θ n + ( T T X ) = n 4 T T T + 2 T T Θ n + ( X ) T , which can be written as
4 n ( I s − T T T ) + 2 ∑ j = 1 s λ j ( ∑ k T j k D k − T T D j T ) = 0. (24) \frac{4}{n}(I_{s}-T^{\mathsf{T}}T)+2\sum_{j=1}^{s}\lambda_{j}\left(\sum_{k}T_{jk}D_{k}-T^{\mathsf{T}}D_{j}T\right)=0. \tag*{(24)} n 4 ( I s − T T T ) + 2 j = 1 ∑ s λ j ( k ∑ T j k D k − T T D j T ) = 0. (24)
As I n , λ 1 , … , λ s I_{n},\lambda_{1},\ldots,\lambda_{s} I n , λ 1 , … , λ s are linearly independent, (24) is equivalent to (19) and (22), while (22) is equivalent to (20) . ▪
Definition 1: A matrix T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s is called an admissible map if it satisfies (19)–(20)
In this paper, an admissible map is also referred to as an admissible coordinate transformation .
Under the transformation X ~ = T T X \widetilde{X}=T^{\mathsf{T}}X X = T T X , where T T T is an admissible coordinate transformation matrix, the system (11) is transformed to
d X ~ = ( A ~ 0 + A ~ X ~ ) d t + ∑ μ = 1 m B ~ μ X ~ d W μ , d\widetilde{X}=(\widetilde{A}_{0}+\widetilde{A}\widetilde{X})dt+\sum_{\mu=1}^{m}\widetilde{B}_{\mu}\widetilde{X}\,dW_{\mu}, d X = ( A 0 + A X ) d t + μ = 1 ∑ m B μ X d W μ ,
d Y = C ~ X ~ d t + d W , (25) dY=\widetilde{C}\widetilde{X}\,dt+dW, \tag*{(25)} d Y = C X d t + d W , (25)
where A ~ 0 = T T A 0 \widetilde{A}_{0}=T^{\mathsf{T}}A_{0} A 0 = T T A 0 , A ~ = T T A T \widetilde{A}=T^{\mathsf{T}}AT A = T T A T , B ~ μ = T T B μ T \widetilde{B}_{\mu}=T^{\mathsf{T}}B_{\mu}T B μ = T T B μ T , and C ~ = C T \widetilde{C}=CT C = C T . The physical parameters (13) transform according to α ~ = T T α \widetilde{\alpha}=T^{\mathsf{T}}\alpha α = T T α , C ~ 1 = C 1 T \widetilde{C}_{1}=C_{1}T C 1 = C 1 T , and C ~ 2 = C 2 T \widetilde{C}_{2}=C_{2}T C 2 = C 2 T . It can be easily verified that B ~ 1 k = Θ n − ( ( C ~ 2 ) k ) \widetilde{B}_{1k}=\Theta_{n}^{-}((\widetilde{C}_{2})_{k}) B 1 k = Θ n − (( C 2 ) k ) , B ~ 2 k = − Θ n − ( ( C ~ 1 ) k ) \widetilde{B}_{2k}=-\Theta_{n}^{-}((\widetilde{C}_{1})_{k}) B 2 k = − Θ n − (( C 1 ) k ) , and A ~ + A ~ T + ∑ k = 1 n w ( B ~ 1 k B ~ 1 k T + B ~ 2 k B ~ 2 k T ) = n 2 Θ n + ( A ~ 0 ) \widetilde{A}+\widetilde{A}^{\mathsf{T}}+\sum_{k=1}^{n_{w}}(\widetilde{B}_{1k}\widetilde{B}_{1k}^{\mathsf{T}}+\widetilde{B}_{2k}\widetilde{B}_{2k}^{\mathsf{T}})=\frac{n}{2}\Theta_{n}^{+}(\widetilde{A}_{0}) A + A T + ∑ k = 1 n w ( B 1 k B 1 k T + B 2 k B 2 k T ) = 2 n Θ n + ( A 0 ) .
Characterization by unitary conjugation
In this subsection, we show that an admissible coordinate transformation T T T is induced by a unitary U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) .
Let U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) be given, and define a matrix T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s by
U λ j U † = ∑ k = 1 s T k j λ k , (26) U\lambda_{j}U^{\dagger}=\sum_{k=1}^{s}T_{kj}\lambda_{k}, \tag*{(26)} U λ j U † = k = 1 ∑ s T k j λ k , (26)
where
T k j = 1 2 Tr ( λ k U λ j U † ) = ⟨ λ k , U λ j U † ⟩ n H S . (27) T_{kj}=\frac{1}{2}\operatorname{Tr}(\lambda_{k}U\lambda_{j}U^{\dagger})=\langle\lambda_{k},U\lambda_{j}U^{\dagger}\rangle_{\mathrm{nHS}}. \tag*{(27)} T k j = 2 1 Tr ( λ k U λ j U † ) = ⟨ λ k , U λ j U † ⟩ nHS . (27)
Since unitary conjugation U ⋅ U † U\mathbin{\cdot}U^{\dagger} U ⋅ U † preserves Hermiticity and trace, it induces the real adjoint representation
Ad : S U ( n ) ⟶ GL R ( Herm 0 ( n ) ) , Ad U ( A ) = U A U † , (28) \operatorname{Ad}:SU(n)\longrightarrow\operatorname{GL}_{\mathbb{R}}(\operatorname{Herm}_{0}(n)),\qquad\operatorname{Ad}_{U}(A)=UAU^{\dagger}, \tag*{(28)} Ad : S U ( n ) ⟶ GL R ( Herm 0 ( n )) , Ad U ( A ) = U A U † , (28)
where GL R ( Herm 0 ( n ) ) \operatorname{GL}_{\mathbb{R}}(\operatorname{Herm}_{0}(n)) GL R ( Herm 0 ( n )) denotes the general linear group of real-linear transformations acting on Herm 0 ( n ) \operatorname{Herm}_{0}(n) Herm 0 ( n ) . Under the generalized Gell–Mann basis, the coordinate matrix of Ad U \operatorname{Ad}_{U} Ad U is the real matrix T T T defined by (26)–(27). Hence, we identify the unitary conjugation Ad U \operatorname{Ad}_{U} Ad U with its coordinate representation T T T .
The following result shows that T = Ad U T=\operatorname{Ad}_{U} T = Ad U for each U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) is an admissible map.
Lemma 3: The matrix T T T defined by (26)–(27) satisfies (19) and (20); in other words, it is admissible.
Proof: As U U U is unitary, we have 2 δ j k = Tr ( λ j λ k ) = Tr [ ( U λ j U † ) ( U λ k U † ) ] = 2 ( T T T ) j k 2\delta_{jk}=\operatorname{Tr}(\lambda_{j}\lambda_{k})=\operatorname{Tr}[(U\lambda_{j}U^{\dagger})(U\lambda_{k}U^{\dagger})]=2(T^{\mathsf{T}}T)_{jk} 2 δ j k = Tr ( λ j λ k ) = Tr [( U λ j U † ) ( U λ k U † )] = 2 ( T T T ) j k , where (1) and (26) have been used. Therefore, T T T satisfies (19). Moreover,
[ U A U † , U B U † ] = U [ A , B ] U † , (29) [UAU^{\dagger},UBU^{\dagger}]=U[A,B]U^{\dagger}, \tag*{(29)} [ U A U † , U B U † ] = U [ A , B ] U † , (29)
and
{ U A U † , U B U † } = U { A , B } U † \{UAU^{\dagger},UBU^{\dagger}\}=U\{A,B\}U^{\dagger} { U A U † , U B U † } = U { A , B } U †
for arbitrary matrices A , B A,B A , B of compatible dimension. On one hand, as given in the proof of Lemma 2, (19) is equivalent to (21). Left- and right-multiplying both sides of (21) by e a T e_{a}^{\mathsf{T}} e a T and e b e_{b} e b of R s \mathbb{R}^{s} R s respectively yield
∑ m , n = 1 s f j m n T n b T m a = ∑ k = 1 s f k a b T j k . (30) \sum_{m,n=1}^{s}f_{jmn}T_{nb}T_{ma}=\sum_{k=1}^{s}f_{kab}T_{jk}. \tag*{(30)} m , n = 1 ∑ s f j mn T nb T ma = k = 1 ∑ s f k ab T j k . (30)
On the other hand, from (29) we have [ U λ j U † , U λ k U † ] = U [ λ j , λ k ] U † [U\lambda_{j}U^{\dagger},U\lambda_{k}U^{\dagger}]=U[\lambda_{j},\lambda_{k}]U^{\dagger} [ U λ j U † , U λ k U † ] = U [ λ j , λ k ] U † , which by (26) is equivalent to
∑ c = 1 s ∑ a , b = 1 s f c a b T a j T b k λ c = ∑ c = 1 s ∑ ℓ = 1 s f ℓ j k T c ℓ λ c . \sum_{c=1}^{s}\sum_{a,b=1}^{s}f_{cab}T_{aj}T_{bk}\lambda_{c}=\sum_{c=1}^{s}\sum_{\ell=1}^{s}f_{\ell jk}T_{c\ell}\lambda_{c}. c = 1 ∑ s a , b = 1 ∑ s f c ab T aj T bk λ c = c = 1 ∑ s ℓ = 1 ∑ s f ℓ j k T c ℓ λ c .
Consequently,
∑ a , b = 1 s f c a b T a j T b k = ∑ ℓ = 1 s f ℓ j k T c ℓ (31) \sum_{a,b=1}^{s}f_{cab}T_{aj}T_{bk}=\sum_{\ell=1}^{s}f_{\ell jk}T_{c\ell} \tag*{(31)} a , b = 1 ∑ s f c ab T aj T bk = ℓ = 1 ∑ s f ℓ j k T c ℓ (31)
due to linear independence of generalized Gell–Mann matrices. However, (30) is actually (31). As this is true for all a , b , c ∈ { 1 , … , s } a,b,c\in\{1,\ldots,s\} a , b , c ∈ { 1 , … , s } , from (29) we get (20a). The anticommutator case can be proved similarly. ▪
In the remainder of this subsection, we aim to show that each admissible map is induced by a unitary matrix. To do this, some preparations are in order. Define the linear map
Φ : R s ⟶ Herm 0 ( n ) , Φ ( v ) = v T X = ∑ a = 1 s v a λ a . (32) \Phi:\mathbb{R}^{s}\longrightarrow\operatorname{Herm}_{0}(n),\qquad\Phi(v)=v^{\mathsf T}X=\sum_{a=1}^{s}v_{a}\lambda_{a}. \tag*{(32)} Φ : R s ⟶ Herm 0 ( n ) , Φ ( v ) = v T X = a = 1 ∑ s v a λ a . (32)
By (3) we have
⟨ Φ ( u ) , Φ ( v ) ⟩ n H S = u T v = ⟨ u , v ⟩ , (33) \langle\Phi(u),\Phi(v)\rangle_{\mathrm{nHS}}=u^{\mathsf T}v=\langle u,v\rangle, \tag*{(33)} ⟨ Φ ( u ) , Φ ( v ) ⟩ nHS = u T v = ⟨ u , v ⟩ , (33)
where ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ is the usual inner product for R s \mathbb{R}^{s} R s . Thus Φ \Phi Φ is an isometry. Introduce two bilinear maps from R s × R s → R s \mathbb{R}^{s}\times\mathbb{R}^{s}\to\mathbb{R}^{s} R s × R s → R s as
f ( u , v ) ≜ Θ n − ( u ) v , d ( u , v ) ≜ Θ n + ( u ) v , u , v ∈ R s . (34) f(u,v)\triangleq\Theta_{n}^{-}(u)v,\qquad d(u,v)\triangleq\Theta_{n}^{+}(u)v,\qquad u,v\in\mathbb{R}^{s}. \tag*{(34)} f ( u , v ) ≜ Θ n − ( u ) v , d ( u , v ) ≜ Θ n + ( u ) v , u , v ∈ R s . (34)
It can be readily verified that
[ Φ ( u ) , Φ ( v ) ] = − 2 i Φ ( f ( u , v ) ) , (35) [\Phi(u),\Phi(v)]=-2i\Phi(f(u,v)), \tag*{(35)} [ Φ ( u ) , Φ ( v )] = − 2 i Φ ( f ( u , v )) , (35)
{ Φ ( u ) , Φ ( v ) } = 4 n ( u T v ) I n + 2 Φ ( d ( u , v ) ) , (36) \{\Phi(u),\Phi(v)\}=\frac{4}{n}(u^{\mathsf T}v)I_{n}+2\Phi(d(u,v)), \tag*{(36)} { Φ ( u ) , Φ ( v )} = n 4 ( u T v ) I n + 2Φ ( d ( u , v )) , (36)
from which we have
Φ ( u ) Φ ( v ) = 2 n ( u T v ) I n + Φ ( d ( u , v ) ) − i Φ ( f ( u , v ) ) . (37) \Phi(u)\Phi(v)=\frac{2}{n}(u^{\mathsf T}v)I_{n}+\Phi(d(u,v))-i\Phi(f(u,v)). \tag*{(37)} Φ ( u ) Φ ( v ) = n 2 ( u T v ) I n + Φ ( d ( u , v )) − i Φ ( f ( u , v )) . (37)
Remark 2: Because we will use the Skolem–Noether Theorem for M n ( C ) M_{n}(\mathbb{C}) M n ( C ) (see e.g., [40 ] ) to establish Theorem 1, we extend the real maps f f f and d d d defined in (34) to complex-bilinear maps C s × C s → C s \mathbb{C}^{s}\times\mathbb{C}^{s}\to\mathbb{C}^{s} C s × C s → C s defined by the same formulas f ( u , v ) = Θ n − ( u ) v f(u,v)=\Theta_{n}^{-}(u)v f ( u , v ) = Θ n − ( u ) v and d ( u , v ) = Θ n + ( u ) v d(u,v)=\Theta_{n}^{+}(u)v d ( u , v ) = Θ n + ( u ) v for all u , v ∈ C s u,v\in\mathbb{C}^{s} u , v ∈ C s . Indeed, they are well defined since the maps Θ n ± \Theta_{n}^{\pm} Θ n ± are actually defined on C s \mathbb{C}^{s} C s .
Let Φ C \Phi_{\mathbb{C}} Φ C denote the complex-linear extension of the isometric map Φ \Phi Φ ; specifically,
Φ C ( u + i v ) ≜ Φ ( u ) + i Φ ( v ) , u , v ∈ R s . (38) \Phi_{\mathbb{C}}(u+iv)\triangleq\Phi(u)+i\Phi(v),\qquad u,v\in\mathbb{R}^{s}. \tag*{(38)} Φ C ( u + i v ) ≜ Φ ( u ) + i Φ ( v ) , u , v ∈ R s . (38)
By means of (36), it can be easily shown that for all u , v ∈ C s u,v\in\mathbb{C}^{s} u , v ∈ C s ,
Φ C ( u ) Φ C ( v ) = 2 n ( u T v ) I n + Φ C ( d ( u , v ) ) − i Φ C ( f ( u , v ) ) . (39) \Phi_{\mathbb{C}}(u)\Phi_{\mathbb{C}}(v)=\frac{2}{n}(u^{\mathsf T}v)I_{n}+\Phi_{\mathbb{C}}(d(u,v))-i\Phi_{\mathbb{C}}(f(u,v)). \tag*{(39)} Φ C ( u ) Φ C ( v ) = n 2 ( u T v ) I n + Φ C ( d ( u , v )) − i Φ C ( f ( u , v )) . (39)
Similarly, given a real subspace V ⊆ R s \mathcal{V}\subseteq\mathbb{R}^{s} V ⊆ R s , define a complex space V C ≜ V + i V ⊆ C s \mathcal{V}_{\mathbb{C}}\triangleq\mathcal{V}+i\mathcal{V}\subseteq\mathbb{C}^{s} V C ≜ V + i V ⊆ C s . Then we use the notation Φ C ( V C ) = Φ ( V ) + i Φ ( V ) \Phi_{\mathbb{C}}(\mathcal{V}_{\mathbb{C}})=\Phi(\mathcal{V})+i\Phi(\mathcal{V}) Φ C ( V C ) = Φ ( V ) + i Φ ( V ) as the complex extension of the real vector space Φ ( V ) \Phi(\mathcal{V}) Φ ( V ) .
Theorem 1: A matrix T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s is admissible if and only if there exists U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) such that (26) holds.
Proof: The “if” part is Lemma 3. Conversely, suppose that T T T is admissible. From (19)–(20) with β = T u \beta=Tu β = T u , we obtain Θ n ± ( T u ) = T Θ n ± ( u ) T T \Theta_{n}^{\pm}(Tu)=T\Theta_{n}^{\pm}(u)T^{\mathsf T} Θ n ± ( T u ) = T Θ n ± ( u ) T T . Consequently,
f ( T u , T v ) = Θ n − ( T u ) T v = T Θ n − ( u ) v = T f ( u , v ) , f(Tu,Tv)=\Theta_{n}^{-}(Tu)Tv=T\Theta_{n}^{-}(u)v=Tf(u,v), f ( T u , T v ) = Θ n − ( T u ) T v = T Θ n − ( u ) v = T f ( u , v ) ,
d ( T u , T v ) = Θ n + ( T u ) T v = T Θ n + ( u ) v = T d ( u , v ) . d(Tu,Tv)=\Theta_{n}^{+}(Tu)Tv=T\Theta_{n}^{+}(u)v=Td(u,v). d ( T u , T v ) = Θ n + ( T u ) T v = T Θ n + ( u ) v = T d ( u , v ) .
In terms of the function Φ C \Phi_{\mathbb{C}} Φ C defined in (37), define a complex-linear map Ψ T : M n ( C ) ⟶ M n ( C ) \Psi_{T}:M_{n}(\mathbb{C})\longrightarrow M_{n}(\mathbb{C}) Ψ T : M n ( C ) ⟶ M n ( C ) by
Ψ T ( c I n + Φ C ( z ) ) ≜ c I n + Φ C ( T z ) , c ∈ C , z ∈ C s . (40) \Psi_{T}(cI_{n}+\Phi_{\mathbb{C}}(z))\triangleq cI_{n}+\Phi_{\mathbb{C}}(Tz),\qquad c\in\mathbb{C},\ z\in\mathbb{C}^{s}. \tag*{(40)} Ψ T ( c I n + Φ C ( z )) ≜ c I n + Φ C ( T z ) , c ∈ C , z ∈ C s . (40)
This map is well defined because M n ( C ) = span C { I n , λ 1 , … , λ s } M_{n}(\mathbb{C})=\operatorname{span}_{\mathbb{C}}\{I_{n},\lambda_{1},\ldots,\lambda_{s}\} M n ( C ) = span C { I n , λ 1 , … , λ s } . Clearly, Ψ T \Psi_{T} Ψ T is unital and bijective. As T T T is real,
Ψ T ( A † ) = Ψ T ( A ) † . (41) \Psi_{T}(A^{\dagger})=\Psi_{T}(A)^{\dagger}. \tag*{(41)} Ψ T ( A † ) = Ψ T ( A ) † . (41)
Furthermore, by (34) and (38) we get Ψ T ( A B ) = Ψ T ( A ) Ψ T ( B ) \Psi_{T}(AB)=\Psi_{T}(A)\Psi_{T}(B) Ψ T ( A B ) = Ψ T ( A ) Ψ T ( B ) for all A , B ∈ M n ( C ) A,B\in M_{n}(\mathbb{C}) A , B ∈ M n ( C ) . Thus Ψ T \Psi_{T} Ψ T is a unital ∗ * ∗ -automorphism of M n ( C ) M_{n}(\mathbb{C}) M n ( C ) . Then by the Skolem–Noether Theorem [40 ] , there exists an invertible matrix V V V such that Ψ T ( A ) = V A V − 1 \Psi_{T}(A)=VAV^{-1} Ψ T ( A ) = V A V − 1 for all A ∈ M n ( C ) A\in M_{n}(\mathbb{C}) A ∈ M n ( C ) . By (40), V A † V − 1 = ( V A V − 1 ) † = ( V − 1 ) † A † V † VA^{\dagger}V^{-1}=(VAV^{-1})^{\dagger}=(V^{-1})^{\dagger}A^{\dagger}V^{\dagger} V A † V − 1 = ( V A V − 1 ) † = ( V − 1 ) † A † V † , which can be rewritten as ( V † V ) A † = A † ( V † V ) (V^{\dagger}V)A^{\dagger}=A^{\dagger}(V^{\dagger}V) ( V † V ) A † = A † ( V † V ) . In other words, V † V V^{\dagger}V V † V commutes with every element of M n ( C ) M_{n}(\mathbb{C}) M n ( C ) . Therefore V † V = γ I n V^{\dagger}V=\gamma I_{n} V † V = γ I n for some γ > 0 \gamma>0 γ > 0 . Replacing V V V by γ − 1 / 2 V \gamma^{-1/2}V γ − 1/2 V gives a unitary matrix. Multiplying γ − 1 / 2 V \gamma^{-1/2}V γ − 1/2 V by a phase e i θ e^{i\theta} e i θ yields U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) . Then Ψ T ( A ) = U A U † \Psi_{T}(A)=UAU^{\dagger} Ψ T ( A ) = U A U † . Applying this identity to (39) and letting c = 0 c=0 c = 0 and z = e j z=e_{j} z = e j establish (26). ■
Strong alignment
In this subsection, we introduce strongly aligned subspaces S ⊆ R s \mathcal{S}\subseteq\mathbb{R}^{s} S ⊆ R s . A strongly aligned subspace is accompanied by another subspace Q ⊆ S ⊥ \mathcal{Q}\subseteq\mathcal{S}^{\perp} Q ⊆ S ⊥ . The spaces of observables Φ ( S ) \Phi(\mathcal{S}) Φ ( S ) and Φ ( Q ) \Phi(\mathcal{Q}) Φ ( Q ) , corresponding to S \mathcal{S} S and Q \mathcal{Q} Q respectively, will be identified with the noncentral and traceless central observable parts, respectively, of a unital C ∗ C^{*} C ∗ -algebra, [41 , 12 ] .
Definition 2: A real subspace S ⊆ R s \mathcal{S}\subseteq\mathbb{R}^{s} S ⊆ R s is called strongly aligned if there exists another real subspace Q ⊆ S ⊥ \mathcal{Q}\subseteq\mathcal{S}^{\perp} Q ⊆ S ⊥ such that
f ( S , S ) ⊆ S , (42) f(\mathcal{S},\mathcal{S})\subseteq\mathcal{S}, \tag*{(42)} f ( S , S ) ⊆ S , (42)
d ( S , S ) ⊆ S ⊕ Q , (43) d(\mathcal{S},\mathcal{S})\subseteq\mathcal{S}\oplus\mathcal{Q}, \tag*{(43)} d ( S , S ) ⊆ S ⊕ Q , (43)
f ( Q , S ⊕ Q ) = { 0 } , (44) f(\mathcal{Q},\mathcal{S}\oplus\mathcal{Q})=\{0\}, \tag*{(44)} f ( Q , S ⊕ Q ) = { 0 } , (44)
d ( Q , S ) ⊆ S , d ( Q , Q ) ⊆ Q , (45) d(\mathcal{Q},\mathcal{S})\subseteq\mathcal{S},\qquad d(\mathcal{Q},\mathcal{Q})\subseteq\mathcal{Q}, \tag*{(45)} d ( Q , S ) ⊆ S , d ( Q , Q ) ⊆ Q , (45)
and the subspace of M n ( C ) M_{n}(\mathbb{C}) M n ( C ) defined as
A ≜ C I n ⊕ Φ C ( Q C ) ⊕ Φ C ( S C ) (46) \mathcal{A}\triangleq\mathbb{C}I_{n}\oplus\Phi_{\mathbb{C}}(\mathcal{Q}_{\mathbb{C}})\oplus\Phi_{\mathbb{C}}(\mathcal{S}_{\mathbb{C}}) \tag*{(46)} A ≜ C I n ⊕ Φ C ( Q C ) ⊕ Φ C ( S C ) (46)
satisfies
Φ ( S ) = A s a ∩ Z ( A ) s a ⊥ . (47) \Phi(\mathcal{S})=\mathcal{A}_{\mathrm{sa}}\cap Z(\mathcal{A})_{\mathrm{sa}}^{\perp}. \tag*{(47)} Φ ( S ) = A sa ∩ Z ( A ) sa ⊥ . (47)
Φ ( Q ) = Z ( A ) s a ∩ { I n } ⊥ . (48) \Phi(\mathcal{Q})=Z(\mathcal{A})_{\mathrm{sa}}\cap\{I_{n}\}^{\perp}. \tag*{(48)} Φ ( Q ) = Z ( A ) sa ∩ { I n } ⊥ . (48)
Here, Z ( A ) Z(\mathcal{A}) Z ( A ) is the center of A \mathcal{A} A , defined as
Z ( A ) ≜ { Z ∈ A : Z A = A Z , ∀ A ∈ A } , (49) Z(\mathcal{A})\triangleq\{Z\in\mathcal{A}:ZA=AZ,\ \forall A\in\mathcal{A}\}, \tag*{(49)} Z ( A ) ≜ { Z ∈ A : Z A = A Z , ∀ A ∈ A } , (49)
the subscript “sa” denotes the part of observables, e.g.,
A s a = { Y ∈ A : Y = Y † } , \mathcal{A}_{\mathrm{sa}}=\{Y\in\mathcal{A}:Y=Y^{\dagger}\}, A sa = { Y ∈ A : Y = Y † } ,
Z ( A ) s a = { Y ∈ Z ( A ) : Y = Y † } , Z(\mathcal{A})_{\mathrm{sa}}=\{Y\in Z(\mathcal{A}):Y=Y^{\dagger}\}, Z ( A ) sa = { Y ∈ Z ( A ) : Y = Y † } ,
and orthogonality in (43) is with respect to the normalized Hilbert–Schmidt inner product (8).
Some algebraic properties of the subspace A \mathcal{A} A defined in (42) are given by the following two lemmas.
Lemma 4: The decomposition of A \mathcal{A} A in (42) is indeed a direct sum.
Proof: By definition, ⟨ I n , Φ C ( w ) ⟩ n H S = 1 n Tr Φ C ( w ) = 0 \langle I_{n},\Phi_{\mathbb{C}}(w)\rangle_{\mathrm{nHS}}=\frac{1}{n}\operatorname{Tr}\Phi_{\mathbb{C}}(w)=0 ⟨ I n , Φ C ( w ) ⟩ nHS = n 1 Tr Φ C ( w ) = 0 for all w ∈ C s w\in\mathbb{C}^{s} w ∈ C s . Moreover, Q ⊥ S \mathcal{Q}\perp\mathcal{S} Q ⊥ S implies p T w = 0 p^{\mathsf T}w=0 p T w = 0 for all p ∈ Q C p\in\mathcal{Q}_{\mathbb{C}} p ∈ Q C and w ∈ S C w\in\mathcal{S}_{\mathbb{C}} w ∈ S C . Thus, the three summands in (42) are mutually orthogonal under the normalized Hermitian inner product (8), and the result follows. ■
Lemma 5: The space A \mathcal{A} A defined in (42) is a finite-dimensional unital C ∗ C^{*} C ∗ -algebra.
Proof: Firstly, I n ∈ A I_{n}\in\mathcal{A} I n ∈ A , and hence A \mathcal{A} A is unital. Secondly, ( c I n ) † = c ∗ I n (cI_{n})^{\dagger}=c^{*}I_{n} ( c I n ) † = c ∗ I n , and by (37) and (32), Φ C ( w ) † = Φ C ( w ∗ ) \Phi_{\mathbb{C}}(w)^{\dagger}=\Phi_{\mathbb{C}}(w^{*}) Φ C ( w ) † = Φ C ( w ∗ ) for all w ∈ C s w\in\mathbb{C}^{s} w ∈ C s , we have A † = A \mathcal{A}^{\dagger}=\mathcal{A} A † = A . Finally, we show multiplicative closure. Products involving I n I_n I n are trivial. If U , V , W ⊆ R s U,V,W \subseteq\mathbb{R}^s U , V , W ⊆ R s are real subspaces with f ( U , V ) ⊆ W f(U,V) \subseteq W f ( U , V ) ⊆ W , then by the complex-bilinear expansion of f ( u x + i u y , v x + i v y ) f(u_x+iu_y,v_x+iv_y) f ( u x + i u y , v x + i v y ) (see Remark 2), simple algebraic manipulations give f ( U C , V C ) ⊆ W C f(U_{\mathbb{C}},V_{\mathbb{C}}) \subseteq W_{\mathbb{C}} f ( U C , V C ) ⊆ W C ; similarly for the map d d d . Hence (41) also holds with S , Q S,Q S , Q replaced by S C , Q C S_{\mathbb{C}},Q_{\mathbb{C}} S C , Q C , provided Q ⊥ S Q \perp S Q ⊥ S as given in Definition 2. Consequently, Φ C ( S C ) Φ C ( S C ) ⊆ A \Phi_{\mathbb{C}}(S_{\mathbb{C}})\Phi_{\mathbb{C}}(S_{\mathbb{C}}) \subseteq\mathcal{A} Φ C ( S C ) Φ C ( S C ) ⊆ A , Φ C ( Q C ) Φ C ( S C ) ⊆ A \Phi_{\mathbb{C}}(Q_{\mathbb{C}})\Phi_{\mathbb{C}}(S_{\mathbb{C}}) \subseteq\mathcal{A} Φ C ( Q C ) Φ C ( S C ) ⊆ A , and Φ C ( Q C ) Φ C ( Q C ) ⊆ A \Phi_{\mathbb{C}}(Q_{\mathbb{C}})\Phi_{\mathbb{C}}(Q_{\mathbb{C}}) \subseteq\mathcal{A} Φ C ( Q C ) Φ C ( Q C ) ⊆ A follow from (38). Finally, equip A \mathcal{A} A with the matrix operator norm ∥ A ∥ = σ max ( A ) \lVert A\rVert=\sigma_{\max}(A) ∥ A ∥ = σ m a x ( A ) , where σ max ( A ) \sigma_{\max}(A) σ m a x ( A ) denotes the largest singular value of A A A . Clearly, ∥ A † A ∥ = ∥ A ∥ 2 \lVert A^\dagger A\rVert=\lVert A\rVert^2 ∥ A † A ∥ = ∥ A ∥ 2 . Thus A \mathcal{A} A is a finite-dimensional unital C ∗ \mathbb{C}^* C ∗ -algebra.
Remark 3: Clearly, [ Φ ( p ) , I n ] = 0 [\Phi(p),I_n]=0 [ Φ ( p ) , I n ] = 0 for all p ∈ R s p\in\mathbb{R}^s p ∈ R s . Moreover, for p ∈ Q p\in Q p ∈ Q , by (35a) and (41c) we have [ Φ ( p ) , Φ ( q ) ] = − 2 i Φ ( f ( p , q ) ) = 0 [\Phi(p),\Phi(q)]=-2i\Phi(f(p,q))=0 [ Φ ( p ) , Φ ( q )] = − 2 i Φ ( f ( p , q )) = 0 for every q ∈ S ⊕ Q q\in S\oplus Q q ∈ S ⊕ Q . As a result, by the form of A \mathcal{A} A defined in (42), we get Φ ( p ) ∈ Z ( A ) \Phi(p)\in Z(\mathcal{A}) Φ ( p ) ∈ Z ( A ) . However, as Φ ( p ) ∈ Herm 0 ( n ) \Phi(p)\in\operatorname{Herm}_0(n) Φ ( p ) ∈ Herm 0 ( n ) , we have Φ ( Q ) ⊆ Z ( A ) s a ∩ { I n } ⊥ \Phi(Q)\subseteq Z(\mathcal{A})_{\mathrm{sa}}\cap\{I_n\}^{\perp} Φ ( Q ) ⊆ Z ( A ) sa ∩ { I n } ⊥ . In this sense, (43b) is the maximality assumption that this inclusion is actually an equality. Moreover, once (43b) holds, Φ ( S ) ⊆ A s a ∩ ( Z ( A ) s a ) ⊥ \Phi(S)\subseteq\mathcal{A}_{\mathrm{sa}}\cap(Z(\mathcal{A})_{\mathrm{sa}})^\perp Φ ( S ) ⊆ A sa ∩ ( Z ( A ) sa ) ⊥ naturally holds (for w ∈ S w\in S w ∈ S : ⟨ Φ ( w ) , c I n + Φ ( p ) ⟩ n H S = 0 + w T p = 0 \langle\Phi(w),cI_n+\Phi(p)\rangle_{n\mathrm{HS}}=0+w^{\mathsf T}p=0 ⟨ Φ ( w ) , c I n + Φ ( p ) ⟩ n HS = 0 + w T p = 0 ), therefore (43a) is again a maximality assumption. Such a “maximality” feature alludes to the word “strongly” in Definition 2.
Artin–Wedderburn decomposition
In this subsection, by applying the Artin–Wedderburn decomposition to the unital C ∗ \mathbb{C}^* C ∗ -subalgebra A ⊆ M n ( C ) \mathcal{A}\subseteq M_n(\mathbb{C}) A ⊆ M n ( C ) studied in the preceding subsection, we present an interpretation of a strongly aligned subspace S S S and the associated subspace Q ⊆ S ⊥ Q\subseteq S^\perp Q ⊆ S ⊥ .
By Lemma 5, A \mathcal{A} A is a finite-dimensional unital C ∗ \mathbb{C}^* C ∗ -algebra. The finite-dimensional Artin–Wedderburn Theorem [41 ] therefore applies. Specifically, there exist positive integers r k , m k r_k,m_k r k , m k and a unitary U U U such that
∑ k = 1 K r k m k = n , (50) \sum_{k=1}^{K}r_km_k=n, \tag*{(50)} k = 1 ∑ K r k m k = n , (50)
and
A ~ ≜ U A U † = ⨁ k = 1 K ( M r k ( C ) ⊗ I m k ) . (51) \widetilde{\mathcal{A}}\triangleq U\mathcal{A}U^\dagger=\bigoplus_{k=1}^{K}\left(M_{r_k}(\mathbb{C})\otimes I_{m_k}\right). \tag*{(51)} A ≜ U A U † = k = 1 ⨁ K ( M r k ( C ) ⊗ I m k ) . (51)
Moreover, as multiplying U U U by a phase e i θ e^{i\theta} e i θ does not change its adjoint action in (46), U U U can always be chosen to satisfy U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) .
By the definition of “center” in (44), the center of A ~ \widetilde{\mathcal{A}} A is
Z ( A ~ ) = ⨁ k = 1 K ( C I r k ⊗ I m k ) , (52) Z(\widetilde{\mathcal{A}})=\bigoplus_{k=1}^{K}\left(\mathbb{C}I_{r_k}\otimes I_{m_k}\right), \tag*{(52)} Z ( A ) = k = 1 ⨁ K ( C I r k ⊗ I m k ) , (52)
whose traceless Hermitian part is
Z 0 = { ⨁ k = 1 K ( z k I r k ⊗ I m k ) : z k ∈ R , ∑ k = 1 K r k m k z k = 0 } . (53) \mathfrak{Z}_0=\left\{\bigoplus_{k=1}^{K}(z_kI_{r_k}\otimes I_{m_k}):z_k\in\mathbb{R},\ \sum_{k=1}^{K}r_km_kz_k=0\right\}. \tag*{(53)} Z 0 = { k = 1 ⨁ K ( z k I r k ⊗ I m k ) : z k ∈ R , k = 1 ∑ K r k m k z k = 0 } . (53)
Consequently, dim R Z 0 = K − 1 \dim_{\mathbb{R}}\mathfrak{Z}_0=K-1 dim R Z 0 = K − 1 . (The notation “dim R V \dim_{\mathbb{R}}V dim R V ” denotes the dimension of V V V as a vector space over the field of real numbers ; i.e., the number of vectors in a basis when only their real linear combinations are allowed. For example, dim R C n = 2 n \dim_{\mathbb{R}}\mathbb{C}^n=2n dim R C n = 2 n .) Clearly by definition, Z 0 \mathfrak{Z}_0 Z 0 is the traceless Hermitian center part of A ~ \widetilde{\mathcal{A}} A . On the other hand, the noncentral traceless Hermitian part of A ~ \widetilde{\mathcal{A}} A is
D = ⨁ k = 1 K ( Herm 0 ( r k ) ⊗ I m k ) , (54) \mathcal{D}=\bigoplus_{k=1}^{K}\left(\operatorname{Herm}_0(r_k)\otimes I_{m_k}\right), \tag*{(54)} D = k = 1 ⨁ K ( Herm 0 ( r k ) ⊗ I m k ) , (54)
with dimension dim R D = ∑ k = 1 K ( r k 2 − 1 ) \dim_{\mathbb{R}}\mathcal{D}=\sum_{k=1}^{K}(r_k^2-1) dim R D = ∑ k = 1 K ( r k 2 − 1 ) .
Lemma 6: D \mathcal{D} D defined in (49) can be written as
D = A ~ s a ∩ ( Z ( A ~ ) ) ⊥ . (55) \mathcal{D}=\widetilde{\mathcal{A}}_{\mathrm{sa}}\cap\left(Z(\widetilde{\mathcal{A}})\right)^\perp. \tag*{(55)} D = A sa ∩ ( Z ( A ) ) ⊥ . (55)
Proof. Clearly, by the definition of Z ( A ~ ) Z(\widetilde{\mathcal{A}}) Z ( A ) in (47), its Hermitian part is
Z ( A ~ ) s a = R I n ⊕ Z 0 . (56) Z(\widetilde{\mathcal{A}})_{\mathrm{sa}}=\mathbb{R}I_n\oplus\mathfrak{Z}_0. \tag*{(56)} Z ( A ) sa = R I n ⊕ Z 0 . (56)
By the definition of A ~ \widetilde{\mathcal{A}} A in (46), every Hermitian element of A ~ \widetilde{\mathcal{A}} A has the form H = ⨁ k = 1 K ( H k ⊗ I m k ) H=\bigoplus_{k=1}^{K}(H_k\otimes I_{m_k}) H = ⨁ k = 1 K ( H k ⊗ I m k ) , where the Hermitian matrix H k = H k † ∈ M r k ( C ) H_k=H_k^\dagger\in M_{r_k}(\mathbb{C}) H k = H k † ∈ M r k ( C ) has the unique decomposition
H k = Tr ( H k ) r k I r k + H k 0 , H k 0 ∈ Herm 0 ( r k ) . H_k=\frac{\operatorname{Tr}(H_k)}{r_k}I_{r_k}+H_k^0,\qquad H_k^0\in\operatorname{Herm}_0(r_k). H k = r k Tr ( H k ) I r k + H k 0 , H k 0 ∈ Herm 0 ( r k ) .
Clearly, the first part contributes to Z ( A ~ ) s a Z(\widetilde{\mathcal{A}})_{\mathrm{sa}} Z ( A ) sa in (51), while the second part belongs to D \mathcal{D} D . Hence, the Hermitian part of A ~ \widetilde{\mathcal{A}} A is
A ~ s a = R I n ⊕ Z 0 ⊕ D . (57) \widetilde{\mathcal{A}}_{\mathrm{sa}}=\mathbb{R}I_n\oplus\mathfrak{Z}_0\oplus\mathcal{D}. \tag*{(57)} A sa = R I n ⊕ Z 0 ⊕ D . (57)
We have D ⊂ A ~ s a \mathcal{D}\subset\widetilde{\mathcal{A}}_{\mathrm{sa}} D ⊂ A sa . For each Z = ⨁ k z k I r k ⊗ I m k ∈ Z 0 Z=\bigoplus_{k}z_kI_{r_k}\otimes I_{m_k}\in\mathfrak{Z}_0 Z = ⨁ k z k I r k ⊗ I m k ∈ Z 0 and D 0 = ⨁ k = 1 K ( H k 0 ⊗ I m k ) D_0=\bigoplus_{k=1}^{K}(H_k^0\otimes I_{m_k}) D 0 = ⨁ k = 1 K ( H k 0 ⊗ I m k ) , where H k 0 ∈ Herm 0 ( r k ) H_k^0\in\operatorname{Herm}_0(r_k) H k 0 ∈ Herm 0 ( r k ) , it is easy to show that ⟨ I n , Z ⟩ H S = 1 2 ∑ k r k m k z k = 0 \langle I_n,Z\rangle_{\mathrm{HS}}=\frac{1}{2}\sum_k r_km_kz_k=0 ⟨ I n , Z ⟩ HS = 2 1 ∑ k r k m k z k = 0 , ⟨ I n , D 0 ⟩ H S = 1 2 ∑ k m k Tr ( H k 0 ) = 0 \langle I_n,D_0\rangle_{\mathrm{HS}}=\frac{1}{2}\sum_km_k\operatorname{Tr}(H_k^0)=0 ⟨ I n , D 0 ⟩ HS = 2 1 ∑ k m k Tr ( H k 0 ) = 0 , and ⟨ Z , D 0 ⟩ H S = 1 2 ∑ k z k m k Tr ( H k 0 ) = 0 \langle Z,D_0\rangle_{\mathrm{HS}}=\frac{1}{2}\sum_k z_km_k\operatorname{Tr}(H_k^0)=0 ⟨ Z , D 0 ⟩ HS = 2 1 ∑ k z k m k Tr ( H k 0 ) = 0 . Now take Y ∈ A ~ s a Y\in\widetilde{\mathcal{A}}_{\mathrm{sa}} Y ∈ A sa with the unique decomposition Y = c I n + Z + D 0 Y=cI_n+Z+D_0 Y = c I n + Z + D 0 as given in (52). By the above orthogonality, Y ⊥ Z ( A ~ ) s a Y\perp Z(\widetilde{\mathcal{A}})_{\mathrm{sa}} Y ⊥ Z ( A ) sa if and only if c = 0 c=0 c = 0 , Z = 0 Z=0 Z = 0 , i.e., if and only if Y ∈ D Y\in\mathcal{D} Y ∈ D . The proof is completed. ■
Recall the adjoint map Ad U \operatorname{Ad}_U Ad U defined in (28). We can easily extend it to be a map on M n ( C ) M_n(\mathbb{C}) M n ( C ) as Ad U ( A ) = U A U † \operatorname{Ad}_U(A)=UAU^\dagger Ad U ( A ) = U A U † for all A ∈ M n ( C ) A\in M_n(\mathbb{C}) A ∈ M n ( C ) . The following result can be easily verified, and thus its proof is omitted.
Lemma 7: Let U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) be given. Then Ad U ( A ) = U A U † \operatorname{Ad}_U(A)=UAU^\dagger Ad U ( A ) = U A U †
(i) fixes I n I_n I n ; i.e., Ad U ( I n ) = I n \operatorname{Ad}_U(I_n)=I_n Ad U ( I n ) = I n ;
(ii) maps A \mathcal{A} A onto A ~ \widetilde{\mathcal{A}} A ;
(iii) maps A s a \mathcal{A}_{\mathrm{sa}} A sa onto A ~ s a \widetilde{\mathcal{A}}_{\mathrm{sa}} A sa ;
(iv) maps Z ( A ) Z(\mathcal{A}) Z ( A ) onto Z ( A ~ ) Z(\widetilde{\mathcal{A}}) Z ( A ) ;
(v) is an isometry for the normalized Hilbert–Schmidt inner product (8).
Consequently, for subspaces V , W ⊆ M n ( C ) V,W\subseteq M_n(\mathbb{C}) V , W ⊆ M n ( C ) ,
Ad U ( V ∩ W ) = Ad U ( V ) ∩ Ad U ( W ) , \operatorname{Ad}_U(V\cap W)=\operatorname{Ad}_U(V)\cap\operatorname{Ad}_U(W), Ad U ( V ∩ W ) = Ad U ( V ) ∩ Ad U ( W ) ,
Ad U ( V ⊥ ) = Ad U ( V ) ⊥ . \operatorname{Ad}_U(V^\perp)=\operatorname{Ad}_U(V)^\perp. Ad U ( V ⊥ ) = Ad U ( V ) ⊥ .
The following result is a consequence of Lemmas 6 and 7.
Lemma 8: Φ ( S ) \Phi(S) Φ ( S ) and Φ ( Q ) \Phi(Q) Φ ( Q ) in (43) are mapped onto Z 0 \mathfrak{Z}_0 Z 0 and D \mathcal{D} D in (52) respectively by Ad U \operatorname{Ad}_U Ad U ; i.e.,
U Φ ( Q ) U † = Z 0 , (58) U\Phi(Q)U^\dagger=\mathfrak{Z}_0, \tag*{(58)} U Φ ( Q ) U † = Z 0 , (58)
U Φ ( S ) U † = D . (59) U\Phi(S)U^\dagger=\mathcal{D}. \tag*{(59)} U Φ ( S ) U † = D . (59)
Proof: First, we derive (53a). Applying Ad U \operatorname{Ad}_U Ad U to (43b) and using Lemma 7,
U Φ ( Q ) U † = Ad U ( Z ( A ) s a ∩ { I n } ⊥ ) = Z ( A ~ ) s a ∩ { I n } ⊥ . (60) U\Phi(\mathcal{Q})U^\dagger=\operatorname{Ad}_U\left(\mathcal{Z}(\mathcal{A})_{\mathrm{sa}}\cap\{I_n\}^{\perp}\right)=\mathcal{Z}(\widetilde{\mathcal{A}})_{\mathrm{sa}}\cap\{I_n\}^{\perp}. \tag*{(60)} U Φ ( Q ) U † = Ad U ( Z ( A ) sa ∩ { I n } ⊥ ) = Z ( A ) sa ∩ { I n } ⊥ . (60)
By (47), a self-adjoint element of Z ( A ~ ) \mathcal{Z}(\widetilde{\mathcal{A}}) Z ( A ) has the form ⨁ k z k I r k ⊗ I m k \bigoplus_k z_k I_{r_k}\otimes I_{m_k} ⨁ k z k I r k ⊗ I m k with z k ∈ R z_k\in\mathbb{R} z k ∈ R , and it is orthogonal to I n I_n I n if and only if ∑ k r k m k z k = 0 \sum_k r_km_kz_k=0 ∑ k r k m k z k = 0 . However, by (48) this set is exactly Z 0 \mathcal{Z}_0 Z 0 . Hence by (54), we get (53a). Next, we derive (53b). Applying Ad U \operatorname{Ad}_U Ad U to (43a) and using Lemma 7, U Φ ( S ) U † = Ad U ( A s a ∩ Z ( A ) s a ⊥ ) = A ~ s a ∩ Z ( A ~ ) s a ⊥ = D U\Phi(\mathcal{S})U^\dagger=\operatorname{Ad}_U\left(\mathcal{A}_{\mathrm{sa}}\cap\mathcal{Z}(\mathcal{A})_{\mathrm{sa}}^\perp\right)=\widetilde{\mathcal{A}}_{\mathrm{sa}}\cap\mathcal{Z}(\widetilde{\mathcal{A}})_{\mathrm{sa}}^\perp=\mathcal{D} U Φ ( S ) U † = Ad U ( A sa ∩ Z ( A ) sa ⊥ ) = A sa ∩ Z ( A ) sa ⊥ = D , where the last equality is (50). ■
Lemma 8 provides an interpretation of the two coordinate subspaces S \mathcal{S} S , Q \mathcal{Q} Q determined by strong alignment. Specifically, the associated subspace Φ ( Q ) \Phi(\mathcal{Q}) Φ ( Q ) is mapped onto the traceless Hermitian center Z 0 \mathcal{Z}_0 Z 0 , while Φ ( S ) \Phi(\mathcal{S}) Φ ( S ) is mapped onto the noncentral traceless Hermitian part D \mathcal{D} D . In this sense, the Artin–Wedderburn decomposition does more than merely block-diagonalize A \mathcal{A} A as shown in (46): it separates its central Hermitian matrices Z 0 \mathcal{Z}_0 Z 0 (namely, Hermitian matrices commuting with each element in A ~ \widetilde{\mathcal{A}} A ) from its noncommutative observables D \mathcal{D} D , where the latter constitute a set of observables whose evolution is decoupled from the environment in the structural-decomposition result to be presented in the next subsection.
Structural decomposition
In this subsection, we derive conditions under which there exists a set of observables of the finite-level system (11), whose evolution is decoupled from the input and output fields.
Theorem 2: Let S ⊆ R s \mathcal{S}\subseteq\mathbb{R}^s S ⊆ R s be strongly aligned and let P S P_{\mathcal{S}} P S be its orthogonal projector. Assume that
[ A , P S ] = 0 , (61) [A,P_{\mathcal{S}}]=0, \tag*{(61)} [ A , P S ] = 0 , (61)
B μ P S = 0 , μ = 1 , … , m , (62) B_\mu P_{\mathcal{S}}=0,\qquad\mu=1,\ldots,m, \tag*{(62)} B μ P S = 0 , μ = 1 , … , m , (62)
C P S = 0 , (63) CP_{\mathcal{S}}=0, \tag*{(63)} C P S = 0 , (63)
P S A 0 = 0. (64) P_{\mathcal{S}}A_0=0. \tag*{(64)} P S A 0 = 0. (64)
Define observables
X S ≜ P S X , X ⊥ ≜ ( I s − P S ) X . (65) X_{\mathcal{S}}\triangleq P_{\mathcal{S}}X,\qquad X_{\perp}\triangleq(I_s-P_{\mathcal{S}})X. \tag*{(65)} X S ≜ P S X , X ⊥ ≜ ( I s − P S ) X . (65)
Then system (11) can be rewritten as
d X S = A X S d t , (66) dX_{\mathcal{S}}=AX_{\mathcal{S}}\,dt, \tag*{(66)} d X S = A X S d t , (66)
d X ⊥ = ( A 0 + A X ⊥ ) d t + ∑ μ = 1 m B μ X ⊥ d W μ , (67) dX_{\perp}=(A_0+AX_{\perp})\,dt+\sum_{\mu=1}^{m}B_\mu X_{\perp}\,dW_\mu, \tag*{(67)} d X ⊥ = ( A 0 + A X ⊥ ) d t + μ = 1 ∑ m B μ X ⊥ d W μ , (67)
d Y = C X ⊥ d t + d W . (68) dY=CX_{\perp}\,dt+dW. \tag*{(68)} d Y = C X ⊥ d t + d W . (68)
Hence, the vector of observables X S X_{\mathcal{S}} X S evolves autonomously, decoupled from both the input and output fields.
Proof: By (55a), the space spanned by S \mathcal{S} S and that spanned by S ⊥ \mathcal{S}^{\perp} S ⊥ are invariant under the transformation of the matrix A A A . Moreover, by (15a)–(15b), B μ T = − B μ B_\mu^\mathsf{T}=-B_\mu B μ T = − B μ , and hence by (55b),
P S B μ = − ( B μ P S ) T = 0 , μ = 1 , … , m . (69) P_{\mathcal{S}}B_\mu=-(B_\mu P_{\mathcal{S}})^\mathsf{T}=0,\qquad\mu=1,\ldots,m. \tag*{(69)} P S B μ = − ( B μ P S ) T = 0 , μ = 1 , … , m . (69)
Left-multiplying (11) by P S P_{\mathcal{S}} P S and using (55a), (55d), and (58) we get (57a). Similarly, left-multiplying (11) by I s − P S I_s-P_{\mathcal{S}} I s − P S and using (55a), (55d), and (56), we get (57b). Finally, (55c) and (56) imply that C X = C ( I s − P S ) X = C X ⊥ CX=C(I_s-P_{\mathcal{S}})X=CX_{\perp} C X = C ( I s − P S ) X = C X ⊥ , which establishes (57c). ■
DFSs
Theorem 2 identifies a set of the system observables whose evolution is not influenced by the environment. Unfortunately, as evidenced by (57), it does not give an explicit decomposition of system matrices; this is the purpose of this section.
Block-adapted bases and relative admissibility
In general, a coordinate transformation does not preserve the generalized Gell–Mann tensors ( f , d ) (f,d) ( f , d ) ; instead, it requires these structure tensors to be transformed together with the basis. The purpose of this subsection is to construct a block-adapted decomposition such that the leading block is a DFS under the transformed structure tensors.
Definition 3: Let η = ( η 1 , … , η s ) \eta=(\eta_1,\ldots,\eta_s) η = ( η 1 , … , η s ) be an ordered orthonormal basis of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) with respect to the normalized Hilbert–Schmidt inner product (8). Define its structure tensors
f a b c η ≜ 1 4 i Tr ( [ η a , η b ] η c ) , d a b c η ≜ 1 4 Tr ( { η a , η b } η c ) . (70) f^\eta_{abc}\triangleq\frac{1}{4i}\operatorname{Tr}([\eta_a,\eta_b]\eta_c),\qquad d^\eta_{abc}\triangleq\frac{1}{4}\operatorname{Tr}(\{\eta_a,\eta_b\}\eta_c). \tag*{(70)} f ab c η ≜ 4 i 1 Tr ([ η a , η b ] η c ) , d ab c η ≜ 4 1 Tr ({ η a , η b } η c ) . (70)
The matrices F a η F^\eta_a F a η , D a η D^\eta_a D a η and the maps Θ η , ± \Theta^{\eta,\pm} Θ η , ± are defined from f η , d η f^\eta,d^\eta f η , d η , exactly as in (5)–(6).
Clearly, the generalized Gell–Mann basis ( λ 1 , … , λ s ) (\lambda_1,\ldots,\lambda_s) ( λ 1 , … , λ s ) is a special case, with f λ = f f^\lambda=f f λ = f and d λ = d d^\lambda=d d λ = d .
Lemma 9: For η \eta η in Definition 3, f η f^\eta f η is real and completely antisymmetric, d η d^\eta d η is real and completely symmetric, and
[ η a , η b ] = 2 i ∑ c = 1 s f a b c η η c , [\eta_a,\eta_b]=2i\sum_{c=1}^{s}f^\eta_{abc}\eta_c, [ η a , η b ] = 2 i c = 1 ∑ s f ab c η η c ,
{ η a , η b } = 4 n δ a b I n + 2 ∑ c = 1 s d a b c η η c . (71) \{\eta_a,\eta_b\}=\frac{4}{n}\delta_{ab}I_n+2\sum_{c=1}^{s}d^\eta_{abc}\eta_c. \tag*{(71)} { η a , η b } = n 4 δ ab I n + 2 c = 1 ∑ s d ab c η η c . (71)
Moreover, if
η b = ∑ a = 1 s R a b λ a (72) \eta_b=\sum_{a=1}^{s}R_{ab}\lambda_a \tag*{(72)} η b = a = 1 ∑ s R ab λ a (72)
for some orthogonal matrix R ∈ R s × s R\in\mathbb{R}^{s\times s} R ∈ R s × s , then the transformed and original structured tensors are related by
f a b c η = ∑ a ′ , b ′ , c ′ = 1 s f a ′ b ′ c ′ R a a ′ R b b ′ R c c ′ , f^\eta_{abc}=\sum_{a',b',c'=1}^{s}f_{a'b'c'}R_{aa'}R_{bb'}R_{cc'}, f ab c η = a ′ , b ′ , c ′ = 1 ∑ s f a ′ b ′ c ′ R a a ′ R b b ′ R c c ′ ,
d a b c η = ∑ a ′ , b ′ , c ′ = 1 s d a ′ b ′ c ′ R a a ′ R b b ′ R c c ′ . (73) d^\eta_{abc}=\sum_{a',b',c'=1}^{s}d_{a'b'c'}R_{aa'}R_{bb'}R_{cc'}. \tag*{(73)} d ab c η = a ′ , b ′ , c ′ = 1 ∑ s d a ′ b ′ c ′ R a a ′ R b b ′ R c c ′ . (73)
Proof: Since each η a \eta_a η a is Hermitian, Tr ( [ η a , η b ] η c ) \operatorname{Tr}([\eta_a,\eta_b]\eta_c) Tr ([ η a , η b ] η c ) is purely imaginary, whereas Tr ( { η a , η b } η c ) \operatorname{Tr}(\{\eta_a,\eta_b\}\eta_c) Tr ({ η a , η b } η c ) is real. Thus both f η f^\eta f η and d η d^\eta d η are real. The antisymmetry of f η f^\eta f η and symmetry of d η d^\eta d η are immediate consequences of (59). The operator 1 2 i [ η a , η b ] \frac{1}{2i}[\eta_a,\eta_b] 2 i 1 [ η a , η b ] is traceless Hermitian, and by (59) its coefficient along η c \eta_c η c is
⟨ η c , 1 2 i [ η a , η b ] ⟩ n H S = f a b c η . \left\langle\eta_c,\frac{1}{2i}[\eta_a,\eta_b]\right\rangle_{\mathrm{nHS}}=f^\eta_{abc}. ⟨ η c , 2 i 1 [ η a , η b ] ⟩ nHS = f ab c η .
Similarly, it can be easily shown that { η a , η b } \{\eta_a,\eta_b\} { η a , η b } has trace 4 δ a b 4\delta_{ab} 4 δ ab , the identity component 4 n δ a b I n \frac{4}{n}\delta_{ab}I_n n 4 δ ab I n , and traceless component η c \eta^c η c with coefficient 2 d a b c η 2d^\eta_{abc} 2 d ab c η . This proves (60). (73) follows by substituting the basis expansion (61) into (59). ■
Definition 4: Given η \eta η in Definition 3, a matrix T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s is called ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible if T T T = I s T^{\mathsf T}T=I_s T T T = I s , and for every β ∈ C s \beta\in\mathbb{C}^s β ∈ C s ,
Θ n η , − ( T T β ) = T T Θ n λ , − ( β ) T , Θ n η , + ( T T β ) = T T Θ n λ , + ( β ) T . (74) \Theta^{\eta,-}_n(T^{\mathsf T}\beta)=T^{\mathsf T}\Theta^{\lambda,-}_n(\beta)T,\qquad
\Theta^{\eta,+}_n(T^{\mathsf T}\beta)=T^{\mathsf T}\Theta^{\lambda,+}_n(\beta)T. \tag*{(74)} Θ n η , − ( T T β ) = T T Θ n λ , − ( β ) T , Θ n η , + ( T T β ) = T T Θ n λ , + ( β ) T . (74)
Definition 1 is recovered when η = λ \eta=\lambda η = λ .
The following result shows that a ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible map transforms the generalized Gell-Mann matrices to observables which preserve commutation and anticommutation relations, relative to the basis η \eta η .
Lemma 10: Given η \eta η in Definition 3 and let T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s . Then the transformed observables X ~ = T T X \widetilde{X}=T^{\mathsf T}X X = T T X satisfy
[ X ~ , X ~ T ] = 2 i Θ n η , − ( X ~ ) , [\widetilde{X},\widetilde{X}^{\mathsf T}]=2i\Theta^{\eta,-}_n(\widetilde{X}), [ X , X T ] = 2 i Θ n η , − ( X ) ,
{ X ~ , X ~ T } = 4 n I s + 2 Θ n η , + ( X ~ ) (75) \{\widetilde{X},\widetilde{X}^{\mathsf T}\}=\frac{4}{n}I_s+2\Theta^{\eta,+}_n(\widetilde{X}) \tag*{(75)} { X , X T } = n 4 I s + 2 Θ n η , + ( X ) (75)
if and only if T T T is ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible.
Proof. The proof is identical to that of Lemma 2, with matrices F k , D k F_k,D_k F k , D k replaced by F k η , D k η F_k^\eta,D_k^\eta F k η , D k η . It uses only the linear independence of the matrices I n , λ 1 , … , λ s I_n,\lambda_1,\ldots,\lambda_s I n , λ 1 , … , λ s and the antisymmetry and symmetry established in Lemma 9.
The following lemma gives the construction of ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible maps, which is a counterpart of Lemma 3.
Lemma 11: Given η \eta η in Definition 3, and let U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) , and define T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s by
U † η j U = ∑ k = 1 s T k j λ k , (76) U^\dagger\eta_jU=\sum_{k=1}^{s}T_{kj}\lambda_k, \tag*{(76)} U † η j U = k = 1 ∑ s T k j λ k , (76)
where T k j = ⟨ λ k , U † η j U ⟩ n H S T_{kj}=\langle\lambda_k,U^\dagger\eta_jU\rangle_{\mathrm{nHS}} T k j = ⟨ λ k , U † η j U ⟩ nHS . Then T T T is ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible.
Proof. Clearly, the family ( U † η 1 U , … , U † η s U ) (U^\dagger\eta_1U,\ldots,U^\dagger\eta_sU) ( U † η 1 U , … , U † η s U ) is an orthonormal basis of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) , so T T T = I s T^{\mathsf T}T=I_s T T T = I s . Furthermore,
X ~ j = e j T T T X = ( T e j ) T X = Φ ( T e j ) = U † η j U . (77) \widetilde{X}_j=e_j^{\mathsf T}T^{\mathsf T}X=(Te_j)^{\mathsf T}X=\Phi(Te_j)=U^\dagger\eta_jU. \tag*{(77)} X j = e j T T T X = ( T e j ) T X = Φ ( T e j ) = U † η j U . (77)
Unitary conjugation of (60) therefore gives
[ X ~ a , X ~ b ] = 2 i ∑ c f a b c n X ~ c , [\widetilde{X}_a,\widetilde{X}_b]=2i\sum_c f^{n}_{abc}\widetilde{X}_c, [ X a , X b ] = 2 i c ∑ f ab c n X c ,
{ X ~ a , X ~ b } = 4 n δ a b I n + 2 ∑ c d a b c n X ~ c , \{\widetilde{X}_a,\widetilde{X}_b\}=\frac{4}{n}\delta_{ab}I_n+2\sum_c d^{n}_{abc}\widetilde{X}_c, { X a , X b } = n 4 δ ab I n + 2 c ∑ d ab c n X c ,
which establish (64) via (60). The result then follows from Lemma 10.
Lemma 11 tells us that given the generalized Gell-Mann matrices { λ 1 , … , λ s } \{\lambda_1,\ldots,\lambda_s\} { λ 1 , … , λ s } and another orthonormal basis { η 1 , … , η s } \{\eta_1,\ldots,\eta_s\} { η 1 , … , η s } , each U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) induces a ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible map T T T .
The following result, echoing Lemma 10, says that a ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible map indeed transforms the quantum n n n -level system (11) to a quantum n n n -level system relative to the orthonormal basis { η 1 , … , η s } \{\eta_1,\ldots,\eta_s\} { η 1 , … , η s } .
Lemma 12: Suppose the matrices of the system (11) are given by (15) for physical parameters θ = ( α , C 1 , C 2 ) \theta=(\alpha,C_1,C_2) θ = ( α , C 1 , C 2 ) in (13), and let a matrix T ∈ R s × s T\in\mathbb{R}^{s\times s} T ∈ R s × s be ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible. Define α ~ = T T α \widetilde{\alpha}=T^{\mathsf T}\alpha α = T T α , C ~ i = C i T \widetilde{C}_i=C_iT C i = C i T , and Λ ~ = Λ T \widetilde{\Lambda}=\Lambda T Λ = Λ T . Then A ~ 0 = T T A 0 \widetilde{A}_0=T^{\mathsf T}A_0 A 0 = T T A 0 , A ~ = T T A T \widetilde{A}=T^{\mathsf T}AT A = T T A T , and B ~ μ = T T B μ T \widetilde{B}_\mu=T^{\mathsf T}B_\mu T B μ = T T B μ T are the counterpart of the matrices in (15) under the orthonormal basis η \eta η . Moreover, the Hamiltonian and coupling operators are
H = α T X = α ~ T X ~ , L = Λ X = Λ ~ X ~ . (78) H=\alpha^{\mathsf T}X=\widetilde{\alpha}^{\mathsf T}\widetilde{X},\qquad L=\Lambda X=\widetilde{\Lambda}\widetilde{X}. \tag*{(78)} H = α T X = α T X , L = Λ X = Λ X . (78)
Proof. Using (63), B ~ 1 k = Θ n η , − ( ( C ~ 2 ) k ) \widetilde{B}_{1k}=\Theta^{\eta,-}_n((\widetilde{C}_2)_k) B 1 k = Θ n η , − (( C 2 ) k ) , and B ~ 2 k = − Θ n η , − ( ( C ~ 1 ) k ) \widetilde{B}_{2k}=-\Theta^{\eta,-}_n((\widetilde{C}_1)_k) B 2 k = − Θ n η , − (( C 1 ) k ) . Since T T T is real, Λ ~ # = Λ # T \widetilde{\Lambda}^{\#}=\Lambda^{\#}T Λ # = Λ # T . Using T T T = I s TT^{\mathsf T}=I_s T T T = I s and (63) for complex arguments gives the η \eta η -basis forms of R k R_k R k and Q k Q_k Q k , and hence the η \eta η -basis version of (15d). Finally, as α ~ T X ~ = α T T T T X = α T X \widetilde{\alpha}^{\mathsf T}\widetilde{X}=\alpha^{\mathsf T}TT^{\mathsf T}X=\alpha^{\mathsf T}X α T X = α T T T T X = α T X , and Λ ~ X ~ = Λ T T T X = Λ X \widetilde{\Lambda}\widetilde{X}=\Lambda TT^{\mathsf T}X=\Lambda X Λ X = Λ T T T X = Λ X , which prove (67).
Definition 5: Let S ⊂ R s \mathcal{S}\subset\mathbb{R}^s S ⊂ R s be strongly aligned in the sense of Definition 2, and with Artïn–Wedderburn data { ( r k , m k ) } k = 1 K \{(r_k,m_k)\}_{k=1}^K {( r k , m k ) } k = 1 K as given in (45). Set p ≜ ∑ k = 1 K ( r k 2 − 1 ) p\triangleq\sum_{k=1}^K(r_k^2-1) p ≜ ∑ k = 1 K ( r k 2 − 1 ) . An orthonormal basis η \eta η of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) is called block adapted if
span R { η 1 , … , η p } = D , (79) \operatorname{span}_{\mathbb{R}}\{\eta_1,\ldots,\eta_p\}=\mathcal{D}, \tag*{(79)} span R { η 1 , … , η p } = D , (79)
and
span R { η p + 1 , … , η p + K − 1 } = J 0 , (80) \operatorname{span}_{\mathbb{R}}\{\eta_{p+1},\ldots,\eta_{p+K-1}\}=\mathcal{J}_0, \tag*{(80)} span R { η p + 1 , … , η p + K − 1 } = J 0 , (80)
where D \mathcal{D} D and J 0 \mathcal{J}_0 J 0 are given in (49) and (48), respectively.
Remark 4: Block-adapted bases always exist. By Lemma 6, D \mathcal{D} D and J 0 \mathcal{J}_0 J 0 are mutually orthogonal subspaces of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) . One may therefore choose orthonormal bases of these two subspaces and complete their union to get an orthonormal basis of Herm 0 ( n ) \operatorname{Herm}_0(n) Herm 0 ( n ) .
The following is the main result of this section.
Theorem 3: Assume the hypotheses of Theorem 2. Let U ∈ S U ( n ) U\in SU(n) U ∈ S U ( n ) realize (46). Then let η \eta η be the associated block-adapted basis, and define the corresponding T T T as in Lemma 11. The following statements hold.
(i) T T T is ( λ → η ) (\lambda\to\eta) ( λ → η ) -admissible. Thus, the transformed observables X ~ = T T X \widetilde{X}=T^{\mathsf T}X X = T T X satisfies the commutation and anticommutation relations (64), and the transformed system is physically realizable with respect to the block-adapted basis η \eta η .
(ii) The first p p p columns of T T T form an orthonormal basis of S \mathcal{S} S . Columns p + 1 , … , p + K − 1 p+1,\ldots,p+K-1 p + 1 , … , p + K − 1 form an orthonormal basis of Q \mathcal{Q} Q . The remaining s − p − K + 1 s-p-K+1 s − p − K + 1 columns form an orthonormal basis of ( S ⊕ Q ) ⊥ (\mathcal{S}\oplus\mathcal{Q})^\perp ( S ⊕ Q ) ⊥ .
(iii) Partition
T = [ T S T ⊥ ] , T S ∈ R s × p . (81) T=[T_{\mathcal{S}}\ \ T_{\perp}],\qquad T_{\mathcal{S}}\in\mathbb{R}^{s\times p}. \tag*{(81)} T = [ T S T ⊥ ] , T S ∈ R s × p . (81)
The transformed system matrices have the block forms
A ~ = T T A T = [ A S 0 0 A ⊥ ] , (82) \widetilde{A}=T^{\mathsf T}AT=
\begin{bmatrix}
A_{\mathcal{S}}&0\\
0&A_{\perp}
\end{bmatrix}, \tag*{(82)} A = T T A T = [ A S 0 0 A ⊥ ] , (82)
B ~ μ = T T B μ T = [ 0 0 0 B μ , ⊥ ] , (83) \widetilde{B}_\mu=T^{\mathsf T}B_\mu T=
\begin{bmatrix}
0&0\\
0&B_{\mu,\perp}
\end{bmatrix}, \tag*{(83)} B μ = T T B μ T = [ 0 0 0 B μ , ⊥ ] , (83)
C ~ = C T = [ 0 C ⊥ ] , (84) \widetilde{C}=CT=[0\ \ C_{\perp}], \tag*{(84)} C = C T = [ 0 C ⊥ ] , (84)
A ~ 0 = T T A 0 = [ 0 A 0 , ⊥ ] . (85) \widetilde{A}_0=T^{\mathsf T}A_0=
\begin{bmatrix}
0\\
A_{0,\perp}
\end{bmatrix}. \tag*{(85)} A 0 = T T A 0 = [ 0 A 0 , ⊥ ] . (85)
(iv) Denote the transformed system observables by X ~ = [ X ~ S X ~ ⊥ ] = T T X \widetilde{X}=\begin{bmatrix}\widetilde{X}_{\mathcal{S}}\\\widetilde{X}_{\perp}\end{bmatrix}=T^{\mathsf T}X X = [ X S X ⊥ ] = T T X . Then
d X ~ S = A S X ~ S d t , (86) d\widetilde{X}_{\mathcal{S}}=A_{\mathcal{S}}\widetilde{X}_{\mathcal{S}}\,dt, \tag*{(86)} d X S = A S X S d t , (86)
d X ~ ⊥ = ( A 0 , ⊥ + A ⊥ X ~ ⊥ ) d t + ∑ μ = 1 m B μ , ⊥ X ~ ⊥ d W μ , (87) d\widetilde{X}_{\perp}=(A_{0,\perp}+A_{\perp}\widetilde{X}_{\perp})\,dt+\sum_{\mu=1}^{m}B_{\mu,\perp}\widetilde{X}_{\perp}\,dW_\mu, \tag*{(87)} d X ⊥ = ( A 0 , ⊥ + A ⊥ X ⊥ ) d t + μ = 1 ∑ m B μ , ⊥ X ⊥ d W μ , (87)
d Y = C ⊥ X ~ ⊥ d t + d W . (88) dY=C_{\perp}\widetilde{X}_{\perp}\,dt+dW. \tag*{(88)} d Y = C ⊥ X ⊥ d t + d W . (88)
Clearly, the subsystem in (72a) is a DFS of the finite-level quantum system 11 in the η \eta η -basis.
Proof: Part (i) follows from Lemmas 11 and 12. For Part (ii), if j ≤ p j \le p j ≤ p , then by (68) we get η j ∈ D \eta_j \in\mathcal{D} η j ∈ D . Moreover, by (53b), U † D U = Φ ( S ) U^\dagger\mathcal{D} U = \Phi(\mathcal{S}) U † D U = Φ ( S ) , which, together with (66), gives
Φ ( T e j ) = U † η j U ∈ Φ ( S ) . (89) \Phi(Te_j) = U^\dagger\eta_j U \in\Phi(\mathcal{S}). \tag*{(89)} Φ ( T e j ) = U † η j U ∈ Φ ( S ) . (89)
Thus T e j ∈ S Te_j \in\mathcal{S} T e j ∈ S . Clearly, these p p p columns are orthonormal, and as p = dim R D = dim S p = \dim_{\mathbb{R}} \mathcal{D} = \dim\mathcal{S} p = dim R D = dim S , they form an orthonormal basis of S \mathcal{S} S . The assertion concerning Q \mathcal{Q} Q follows in the same way from U † Z 0 U = Φ ( Q ) U^\dagger\mathcal{Z}_0 U = \Phi(\mathcal{Q}) U † Z 0 U = Φ ( Q ) in Lemma 8.
For Part (iii), by (70) and item (ii) proved above,
P S = T S T S T , P S T S = T S , T S ⊥ T P S = 0. (90) P_{\mathcal{S}} = T_{\mathcal{S}}T_{\mathcal{S}}^{\mathsf{T}}, \qquad P_{\mathcal{S}}T_{\mathcal{S}} = T_{\mathcal{S}}, \qquad T_{\mathcal{S}}^{\perp\mathsf{T}}P_{\mathcal{S}} = 0. \tag*{(90)} P S = T S T S T , P S T S = T S , T S ⊥ T P S = 0. (90)
Thus, by (55a), T S ⊥ T A T S = T S ⊥ T ( I s − P S ) A P S T S = 0 T_{\mathcal{S}}^{\perp\mathsf{T}} A T_{\mathcal{S}} = T_{\mathcal{S}}^{\perp\mathsf{T}}(I_s - P_{\mathcal{S}}) A P_{\mathcal{S}} T_{\mathcal{S}} = 0 T S ⊥ T A T S = T S ⊥ T ( I s − P S ) A P S T S = 0 , and T S T A T S ⊥ = T S T P S A ( I s − P S ) T S ⊥ = 0 T_{\mathcal{S}}^{\mathsf{T}} A T_{\mathcal{S}}^\perp= T_{\mathcal{S}}^{\mathsf{T}} P_{\mathcal{S}} A(I_s - P_{\mathcal{S}})T_{\mathcal{S}}^\perp= 0 T S T A T S ⊥ = T S T P S A ( I s − P S ) T S ⊥ = 0 . Therefore A ~ \widetilde{A} A has the block form (71a). Next, by (55b) and (74), B μ T S = B μ P S T S = 0 B_\mu T_{\mathcal{S}} = B_\mu P_{\mathcal{S}}T_{\mathcal{S}} = 0 B μ T S = B μ P S T S = 0 . Moreover, B μ T = − B μ B_\mu^{\mathsf{T}} = -B_\mu B μ T = − B μ implies P S B μ = 0 P_{\mathcal{S}}B_\mu= 0 P S B μ = 0 , and hence T S T B μ = T S T P S B μ = 0 T_{\mathcal{S}}^{\mathsf{T}}B_\mu= T_{\mathcal{S}}^{\mathsf{T}}P_{\mathcal{S}}B_\mu= 0 T S T B μ = T S T P S B μ = 0 . This proves the form B ~ μ \widetilde{B}_\mu B μ in (71b). Similarly, by (55c)–(55d), and (74), C T S = C P S T S = 0 CT_{\mathcal{S}} = CP_{\mathcal{S}}T_{\mathcal{S}} = 0 C T S = C P S T S = 0 , and T S T A 0 = T S T P S A 0 = 0 T_{\mathcal{S}}^{\mathsf{T}}A_0 = T_{\mathcal{S}}^{\mathsf{T}}P_{\mathcal{S}}A_0 = 0 T S T A 0 = T S T P S A 0 = 0 . Thus (71c)–(71d) hold. Part (iv) follows by substitution of the above transformed matrices into the transformed system.
DFS under the Gell–Mann basis
The following result, which is a consequence of Theorem 3, constructs a DFS of the finite-level quantum system 11 with respect to the generalized Gell–Mann basis.
Corollary 1: Assume the hypotheses of Theorem 2. Let U ∈ S U ( n ) U \in\mathrm{SU}(n) U ∈ SU ( n ) realize the Artin–Wedderburn decomposition (46), define T = Ad U † T = \operatorname{Ad}_{U^\dagger} T = Ad U † , and set p = dim R D = ∑ k = 1 K ( r k 2 − 1 ) p = \dim_{\mathbb{R}} \mathcal{D} = \sum_{k=1}^{K}(r_k^2 - 1) p = dim R D = ∑ k = 1 K ( r k 2 − 1 ) . Suppose that D \mathcal{D} D is spanned by a subset of the selected generalized Gell–Mann basis. After relabeling that subset, we assume
D = span R { λ 1 , … , λ p } . (91) \mathcal{D} = \operatorname{span}_{\mathbb{R}}\{\lambda_1,\ldots,\lambda_p\}. \tag*{(91)} D = span R { λ 1 , … , λ p } . (91)
Then T T T is admissible in the sense of Definition 1, its first p p p columns form an orthonormal basis of S \mathcal{S} S , and the transformed matrices have the block forms (71). Consequently, the transformed system has the form (72) while retaining the original generalized Gell–Mann structure tensors ( f , d ) (f,d) ( f , d ) .
A sufficient condition for (75) is the following: exactly one Artin–Wedderburn block has r k ≥ 2 r_k \ge2 r k ≥ 2 with multiplicity m k = 1 m_k = 1 m k = 1 , and the blocks are ordered so that that block occupies the leading diagonal corner.
Proof: By Lemma 3, T = Ad U † T = \operatorname{Ad}_{U^\dagger} T = Ad U † is admissible. For j = 1 , … , p j = 1,\ldots,p j = 1 , … , p , (75) gives λ j ∈ D \lambda_j \in\mathcal{D} λ j ∈ D . Hence, by Lemma 8,
Φ ( T e j ) = U † λ j U ∈ U † D U = Φ ( S ) , j = 1 , … , p . (92) \Phi(Te_j) = U^\dagger\lambda_j U \in U^\dagger\mathcal{D} U = \Phi(\mathcal{S}), \qquad j = 1,\ldots,p. \tag*{(92)} Φ ( T e j ) = U † λ j U ∈ U † D U = Φ ( S ) , j = 1 , … , p . (92)
Therefore T e j ∈ S Te_j \in\mathcal{S} T e j ∈ S . Since T T T is orthogonal, the vectors T e 1 , … , T e p Te_1,\ldots,Te_p T e 1 , … , T e p are orthonormal. Moreover, by (53b), p = dim R D = dim S p = \dim_{\mathbb{R}} \mathcal{D} = \dim\mathcal{S} p = dim R D = dim S , so they form an orthonormal basis of S \mathcal{S} S . The block forms then follow from the same projector calculation used in the proof of items (iii) and (iv) of Theorem 3.
For the sufficient condition, let that unique block have size r ≥ 2 r \ge2 r ≥ 2 and multiplicity one, and be placed in the leading r × r r \times r r × r corner. Every other block has r k = 1 r_k = 1 r k = 1 , and hence Herm 0 ( r k ) = { 0 } \operatorname{Herm}_0(r_k) = \{0\} Herm 0 ( r k ) = { 0 } . It follows that D = Herm 0 ( r ) ⊕ 0 n − r \mathcal{D} = \operatorname{Herm}_0(r) \oplus\mathbf{0}_{n-r} D = Herm 0 ( r ) ⊕ 0 n − r . The generalized Gell–Mann matrices { u j k , v j k : 1 ≤ j < k ≤ r } ∪ { w 1 , … , w r − 1 } \{u_{jk},v_{jk}:1 \le j < k \le r\} \cup\{w_1,\ldots,w_{r-1}\} { u j k , v j k : 1 ≤ j < k ≤ r } ∪ { w 1 , … , w r − 1 } are supported on the leading r × r r \times r r × r corner and form an orthonormal basis of this subspace. Indeed, their number is 2 ( r 2 ) + ( r − 1 ) = r 2 − 1 = dim R Herm 0 ( r ) 2\binom{r}{2} + (r-1) = r^2 - 1 = \dim_{\mathbb{R}}\operatorname{Herm}_0(r) 2 ( 2 r ) + ( r − 1 ) = r 2 − 1 = dim R Herm 0 ( r ) . After relabeling these matrices as λ 1 , … , λ p \lambda_1,\ldots,\lambda_p λ 1 , … , λ p , (75) holds. ■
QND observables
In this section, we study QND variables of the quantum finite-level system (11).
QND observables in the sense of linear systems theory
Given a matrix V = [ v 1 ⋯ v r ] ∈ R s × r V = [v_1 \ \cdots\ v_r] \in\mathbb{R}^{s \times r} V = [ v 1 ⋯ v r ] ∈ R s × r (r ≤ s r \le s r ≤ s ), define a vector of observables
Z ≜ V T X = [ Z 1 ⋮ Z r ] . (93) Z \triangleq V^{\mathsf{T}}X =
\begin{bmatrix}
Z_1\\
\vdots\\
Z_r
\end{bmatrix}. \tag*{(93)} Z ≜ V T X = Z 1 ⋮ Z r . (93)
Left-multiplying system (11) by V T V^{\mathsf{T}} V T gives the following QSDE
d Z = ( V T A 0 + V T A X ) d t + ∑ μ = 1 m V T B μ X d W μ . (94) dZ = (V^{\mathsf{T}}A_0 + V^{\mathsf{T}}AX)\,dt + \sum_{\mu=1}^{m} V^{\mathsf{T}}B_\mu X\,dW_\mu. \tag*{(94)} d Z = ( V T A 0 + V T A X ) d t + μ = 1 ∑ m V T B μ X d W μ . (94)
Theorem 4: Let V ∈ R s × r V \in\mathbb{R}^{s \times r} V ∈ R s × r have full column rank, and define Z = V T X Z = V^{\mathsf{T}}X Z = V T X as in (77). Then the following statements hold.
(i) The components of the vector of observables Z Z Z mutually commute if and only if
Θ n − ( v i ) v j = 0 , i , j = 1 , … , r . (95) \Theta_n^{-}(v_i)v_j = 0, \qquad i,j = 1,\ldots,r. \tag*{(95)} Θ n − ( v i ) v j = 0 , i , j = 1 , … , r . (95)
(ii) In addition, if
Θ n + ( v i ) v j ∈ Range ( V ) , i , j = 1 , … , r (96) \Theta_n^{+}(v_i)v_j \in\operatorname{Range}(V), \qquad i,j = 1,\ldots,r \tag*{(96)} Θ n + ( v i ) v j ∈ Range ( V ) , i , j = 1 , … , r (96)
holds, then
A Z ≜ span C { I n , Z 1 , … , Z r } (97) \mathcal{A}_Z \triangleq\operatorname{span}_{\mathbb{C}}\{I_n,Z_1,\ldots,Z_r\} \tag*{(97)} A Z ≜ span C { I n , Z 1 , … , Z r } (97)
is a commutative unital C ∗ C^* C ∗ -algebra.
(iii) If
V T B μ = 0 , μ = 1 , … , m , (98) V^{\mathsf{T}}B_\mu= 0, \qquad\mu= 1,\ldots,m, \tag*{(98)} V T B μ = 0 , μ = 1 , … , m , (98)
and
A T Range ( V ) ⊆ Range ( V ) (99) A^{\mathsf{T}}\operatorname{Range}(V) \subseteq\operatorname{Range}(V) \tag*{(99)} A T Range ( V ) ⊆ Range ( V ) (99)
hold, then the QSDE (78) reduces to
d Z = ( b + K Z ) d t , (100) dZ = (b + KZ)\,dt, \tag*{(100)} d Z = ( b + K Z ) d t , (100)
where the matrices
K = V T A V ( V T V ) − 1 , b = V T A 0 . (101) K = V^{\mathsf{T}}AV(V^{\mathsf{T}}V)^{-1}, \qquad b = V^{\mathsf{T}}A_0. \tag*{(101)} K = V T A V ( V T V ) − 1 , b = V T A 0 . (101)
In addition, if (79) holds, then
[ Z i ( t ) , Z j ( s ) ] = 0 (102) [Z_i(t),Z_j(s)] = 0 \tag*{(102)} [ Z i ( t ) , Z j ( s )] = 0 (102)
for every i , j i,j i , j and all t , s ≥ 0 t,s \ge0 t , s ≥ 0 .
(iv) Moreover, denote Y M = M Y Y_M = MY Y M = M Y for some M ∈ R p × m M \in\mathbb{R}^{p \times m} M ∈ R p × m (p ≤ m p \le m p ≤ m ). If
M J M T = 0 , (103) MJM^{\mathsf{T}} = 0, \tag*{(103)} M J M T = 0 , (103)
and
ker ( V T ) ⊆ ker ( M C ) (104) \operatorname{ker}(V^{\mathsf{T}}) \subseteq\operatorname{ker}(MC) \tag*{(104)} ker ( V T ) ⊆ ker ( M C ) (104)
hold, then
d Y M = C Z Z d t + M d W (105) dY_M=C_ZZ\,dt+M\,dW \tag*{(105)} d Y M = C Z Z d t + M d W (105)
describes the dynamics of the commuting observables Y M Y_M Y M , where
C Z = M C V ( V ⊤ V ) − 1 ∈ R p × r , (106) C_Z=MCV(V^\top V)^{-1}\in\mathbb{R}^{p\times r}, \tag*{(106)} C Z = M C V ( V ⊤ V ) − 1 ∈ R p × r , (106)
(v) Finally, define the observability matrix for the linear dynamics described by (84) and (89),
O Z ≜ [ C Z C Z K ⋮ C Z K r − 1 ] . (107) \mathcal{O}_Z\triangleq
\begin{bmatrix}
C_Z\\
C_ZK\\
\vdots\\
C_ZK^{r-1}
\end{bmatrix}. \tag*{(107)} O Z ≜ C Z C Z K ⋮ C Z K r − 1 . (107)
If
rank O Z = r , (108) \operatorname{rank}\mathcal{O}_Z=r, \tag*{(108)} rank O Z = r , (108)
then the vector Z Z Z is observable from Y M Y_M Y M in the standard linear-systems sense.
Proof: (i) From (9a) we have [ Z , Z ⊤ ] = [ V ⊤ X , X ⊤ V ] = V ⊤ [ X , X ⊤ ] V = 2 i V ⊤ Θ n − ( X ) V [Z,Z^\top]=[V^\top X,X^\top V]=V^\top[X,X^\top]V=2iV^\top\Theta_n^-(X)V [ Z , Z ⊤ ] = [ V ⊤ X , X ⊤ V ] = V ⊤ [ X , X ⊤ ] V = 2 i V ⊤ Θ n − ( X ) V . Thus, similar to the proof for Lemma 2,
[ Z i , Z j ] = 0 , i , j = 1 , … , r . (109) [Z_i,Z_j]=0,\qquad i,j=1,\ldots,r. \tag*{(109)} [ Z i , Z j ] = 0 , i , j = 1 , … , r . (109)
if and only if (79) holds.
(ii) Consider the space A Z \mathcal{A}_Z A Z defined in (81). Under item (i) proved above and the matrix operator norm ∥ A ∥ = σ max ( A ) \lVert A\rVert=\sigma_{\max}(A) ∥ A ∥ = σ m a x ( A ) , to show that A Z \mathcal{A}_Z A Z is commutative unital C ∗ \mathrm{C}^* C ∗ -algebra, it is necessary and sufficient to show that A Z \mathcal{A}_Z A Z is closed under the anticommutation relations. Since V V V has full column rank, by (80) there is a unique vector Γ i j ∈ R r \Gamma_{ij}\in\mathbb{R}^r Γ ij ∈ R r satisfying Θ n + ( v i ) v j = V Γ i j \Theta_n^+(v_i)v_j=V\Gamma_{ij} Θ n + ( v i ) v j = V Γ ij , where the matrices Γ i j = ( V ⊤ V ) − 1 V ⊤ Θ n + ( v i ) v j \Gamma_{ij}=(V^\top V)^{-1}V^\top\Theta_n^+(v_i)v_j Γ ij = ( V ⊤ V ) − 1 V ⊤ Θ n + ( v i ) v j . Consequently, by (9b) and (77),
{ Z i , Z j } = 4 n v i ⊤ v j I n + 2 Γ i j ⊤ Z . (110) \{Z_i,Z_j\}=\frac{4}{n}v_i^\top v_jI_n+2\Gamma_{ij}^\top Z. \tag*{(110)} { Z i , Z j } = n 4 v i ⊤ v j I n + 2 Γ ij ⊤ Z . (110)
As a result, if both (79)–(80) hold, then by (93)–(94), Z i Z j = 2 n v i ⊤ v j I n + Γ i j ⊤ Z ∈ A Z Z_iZ_j=\frac{2}{n}v_i^\top v_jI_n+\Gamma_{ij}^\top Z\in\mathcal{A}_Z Z i Z j = n 2 v i ⊤ v j I n + Γ ij ⊤ Z ∈ A Z .
(iii) As V V V is of full column rank, by (83), the matrix K ∈ R r × r K\in\mathbb{R}^{r\times r} K ∈ R r × r in (85) satisfies V ⊤ A = K V ⊤ V^\top A=KV^\top V ⊤ A = K V ⊤ . Then by (77)
V ⊤ A X = K V ⊤ X = K Z . (111) V^\top AX=KV^\top X=KZ. \tag*{(111)} V ⊤ A X = K V ⊤ X = K Z . (111)
By (82) and (95), the QSDE (78) reduces to (84). The solution of (84) with the initial condition Z ( 0 ) = Z Z(0)=Z Z ( 0 ) = Z is
Z ( t ) = e K t Z + ∫ 0 t e K ( t − τ ) b d τ . Z(t)=e^{Kt}Z+\int_0^t e^{K(t-\tau)}b\,d\tau. Z ( t ) = e K t Z + ∫ 0 t e K ( t − τ ) b d τ .
Consequently,
[ Z ( t ) , Z ( s ) ⊤ ] = e K t [ Z , Z ⊤ ] e K ⊤ s . (112) [Z(t),Z(s)^\top]=e^{Kt}[Z,Z^\top]e^{K^\top s}. \tag*{(112)} [ Z ( t ) , Z ( s ) ⊤ ] = e K t [ Z , Z ⊤ ] e K ⊤ s . (112)
As proved in item (i) above, if (79) holds, then we have [ Z , Z ⊤ ] = 0 [Z,Z^\top]=0 [ Z , Z ⊤ ] = 0 , whose substitution into (96) gives [ Z ( t ) , Z ( s ) ⊤ ] = 0 [Z(t),Z(s)^\top]=0 [ Z ( t ) , Z ( s ) ⊤ ] = 0 for all t , s ≥ 0 t,s\geq0 t , s ≥ 0 . Componentwise, (86) holds for all i , j = 1 , … , r i,j=1,\ldots,r i , j = 1 , … , r and t , s ≥ 0 t,s\geq0 t , s ≥ 0 .
(iv) Under (88), there exists a matrix C Z C_Z C Z such that M C = C Z V ⊤ MC=C_ZV^\top M C = C Z V ⊤ . In fact, C Z C_Z C Z defined in (90) is a valid choice. As a result, we get (89). Moreover, by (87), the components of the output observable vector Y M Y_M Y M mutually commute.
(v) The conclusion follows as in the well-known linear systems theory [42 ] . ■
QND observables in the sense of spectral separation
According to Theorem 4, if (92) does not hold, the observables { Z 1 , … , Z r } \{Z_1,\ldots,Z_r\} { Z 1 , … , Z r } in (84) cannot be completely reconstructed from the output field vector (89), in the sense of linear systems theory [42 ] . However, for the finite-dimensional C ∗ \mathrm{C}^* C ∗ -algebra A Z \mathcal{A}_Z A Z of commuting observables defined in (81), a different notion of observability is often more appropriate. Specifically, instead of asking whether each Z i Z_i Z i can be reconstructed linearly from Y M Y_M Y M , one may ask whether a vector of selected output-signal observables (to be defined in (98) below) is able to distinguish joint spectral subspaces of A Z \mathcal{A}_Z A Z . Equivalently, one asks whether a vector of selected output-signal observables is able to generate algebra A Z \mathcal{A}_Z A Z . In this sense, we say the commutative observable algebra A Z \mathcal{A}_Z A Z is observable. This operator-algebraic notion of observability is developed in this subsection.
According to Theorem 4, if (79)–(80) hold, then all the elements in A Z \mathcal{A}_Z A Z defined in (81) are mutually commutative. Consequently, there is a family of minimal mutually orthogonal projectors Π 1 , … , Π q ∈ A Z \Pi_1,\ldots,\Pi_q\in\mathcal{A}_Z Π 1 , … , Π q ∈ A Z such that
A Z = span { Π 1 , … , Π q } , \mathcal{A}_Z=\operatorname{span}\{\Pi_1,\ldots,\Pi_q\}, A Z = span { Π 1 , … , Π q } ,
Π α Π β = δ α β Π α , ∑ α = 1 q Π α = I n . (113) \Pi_\alpha\Pi_\beta=\delta_{\alpha\beta}\Pi_\alpha,\qquad\sum_{\alpha=1}^{q}\Pi_\alpha=I_n. \tag*{(113)} Π α Π β = δ α β Π α , α = 1 ∑ q Π α = I n . (113)
Recall the selected output equation (89), and define the output-signal observables
F ℓ = ∑ j = 1 r ( C Z ) ℓ j Z j ∈ A Z , ℓ = 1 , … , p . (114) F_\ell=\sum_{j=1}^{r}(C_Z)_{\ell j}Z_j\in\mathcal{A}_Z,\qquad\ell=1,\ldots,p. \tag*{(114)} F ℓ = j = 1 ∑ r ( C Z ) ℓ j Z j ∈ A Z , ℓ = 1 , … , p . (114)
These are system observables, rather than output-field quadratures. The terminology “output-signal observables” refers to their role in the selected output equation. Actually from the output equation (89) we have
d Y M , ℓ ( t ) = F ℓ ( t ) d t + ( M d W ( t ) ) ℓ , ℓ = 1 , … , p . dY_{M,\ell}(t)=F_\ell(t)\,dt+(M\,dW(t))_\ell,\qquad\ell=1,\ldots,p. d Y M , ℓ ( t ) = F ℓ ( t ) d t + ( M d W ( t ) ) ℓ , ℓ = 1 , … , p .
Thus, F ℓ ( t ) F_\ell(t) F ℓ ( t ) is the system-dependent signal in the measured-output QSDE [31 Sec. 2.1] . The spectral-separation criterion below concerns the system algebra generated by these signal observables; it does not assert exact reconstruction from a finite noisy measurement record.
Theorem 5. The selected mutually commuting output-signal observables { F 1 , … , F p } \{F_1,\ldots,F_p\} { F 1 , … , F p } generate A Z \mathcal{A}_Z A Z if and only if the joint signal-value vectors (namely f j α f_{j\alpha} f j α for each observable F j F_j F j , j = 1 , … , p j=1,\ldots,p j = 1 , … , p )
f α = [ f 1 α ⋮ f p α ] , α = 1 , … , q f_\alpha=
\begin{bmatrix}
f_{1\alpha}\\
\vdots\\
f_{p\alpha}
\end{bmatrix},
\qquad\alpha=1,\ldots,q f α = f 1 α ⋮ f p α , α = 1 , … , q
satisfy
f α ≠ f β whenever α ≠ β . (115) f_\alpha\ne f_\beta\qquad\text{whenever }\alpha\ne\beta. \tag*{(115)} f α = f β whenever α = β . (115)
Equivalently, the ∗ ^* ∗ -algebra
alg ∗ ( I n , F 1 , … , F p ) = A Z . (116) \operatorname{alg}^*(I_n,F_1,\ldots,F_p)=\mathcal{A}_Z. \tag*{(116)} alg ∗ ( I n , F 1 , … , F p ) = A Z . (116)
Proof: (Sufficiency). The proof is algebraic. By means of (97) and the form of F ℓ F_\ell F ℓ , direct algebraic manipulations yield
∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = ∑ γ = 1 q [ ∑ ℓ = 1 p ( f ℓ γ − f ℓ β ) 2 ] Π γ = ∑ γ = 1 q ∥ f γ − f β ∥ 2 Π γ . \sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2
=
\sum_{\gamma=1}^{q}\left[\sum_{\ell=1}^{p}(f_{\ell\gamma}-f_{\ell\beta})^2\right]\Pi_\gamma
=
\sum_{\gamma=1}^{q}\lVert f_\gamma-f_\beta\rVert^2\Pi_\gamma. ℓ = 1 ∑ p ( F ℓ − f ℓ β I n ) 2 = γ = 1 ∑ q [ ℓ = 1 ∑ p ( f ℓ γ − f ℓ β ) 2 ] Π γ = γ = 1 ∑ q ∥ f γ − f β ∥ 2 Π γ .
from which we have
∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 ∥ f α − f β ∥ 2 = ∑ γ = 1 q ∥ f γ − f β ∥ 2 ∥ f α − f β ∥ 2 Π γ . \frac{\sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2}{\lVert f_\alpha-f_\beta\rVert^2}
=
\sum_{\gamma=1}^{q}\frac{\lVert f_\gamma-f_\beta\rVert^2}{\lVert f_\alpha-f_\beta\rVert^2}\Pi_\gamma. ∥ f α − f β ∥ 2 ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = γ = 1 ∑ q ∥ f α − f β ∥ 2 ∥ f γ − f β ∥ 2 Π γ .
This, together with
( ∑ γ a γ Π γ ) ( ∑ δ b δ Π δ ) = ∑ γ a γ b γ Π γ , \left(\sum_\gamma a_\gamma\Pi_\gamma\right)\left(\sum_\delta b_\delta\Pi_\delta\right)
=
\sum_\gamma a_\gamma b_\gamma\Pi_\gamma, ( γ ∑ a γ Π γ ) ( δ ∑ b δ Π δ ) = γ ∑ a γ b γ Π γ ,
yields
∏ β = 1 β ≠ α q ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 ∥ f α − f β ∥ 2 = Π α , α = 1 , … , q . (117) \prod_{\substack{\beta=1\\\beta\ne\alpha}}^{q}
\frac{\sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2}{\lVert f_\alpha-f_\beta\rVert^2}
=
\Pi_\alpha,\qquad\alpha=1,\ldots,q. \tag*{(117)} β = 1 β = α ∏ q ∥ f α − f β ∥ 2 ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = Π α , α = 1 , … , q . (117)
where (99) has been used. By (101), all orthogonal projectors Π α \Pi_\alpha Π α can be obtained from the output-signal observables { F 1 , … , F p } \{F_1,\ldots,F_p\} { F 1 , … , F p } .
(Necessity.) We prove it by contradiction. Suppose that f α = f β f_\alpha=f_\beta f α = f β for some α ≠ β \alpha\ne\beta α = β . As the observables F ℓ F_\ell F ℓ mutually commute, every element of B ≜ alg ∗ ( I n , F 1 , … , F p ) \mathcal{B}\triangleq\operatorname{alg}^{*}(I_n,F_1,\ldots,F_p) B ≜ alg ∗ ( I n , F 1 , … , F p ) is a polynomial in them, and, by mean of spectral decomposition such polynomial has the form
P ( F 1 , … , F p ) = ∑ γ = 1 q P ( f γ ) Π γ . P(F_1,\ldots,F_p)=\sum_{\gamma=1}^{q}P(f_\gamma)\Pi_\gamma. P ( F 1 , … , F p ) = γ = 1 ∑ q P ( f γ ) Π γ .
If Π α = P ( F 1 , … , F p ) \Pi_\alpha=P(F_1,\ldots,F_p) Π α = P ( F 1 , … , F p ) for some polynomial P ( F 1 , … , F p ) P(F_1,\ldots,F_p) P ( F 1 , … , F p ) , multiplication by Π α \Pi_\alpha Π α and Π β \Pi_\beta Π β , respectively, yield P ( f α ) = 1 P(f_\alpha)=1 P ( f α ) = 1 , and P ( f β ) = 0 P(f_\beta)=0 P ( f β ) = 0 , contradicting f α = f β f_\alpha=f_\beta f α = f β . The same argument excludes Π β \Pi_\beta Π β too. Thus, neither Π α \Pi_\alpha Π α nor Π β \Pi_\beta Π β belongs to B \mathcal{B} B . On the other hand, both Π α \Pi_\alpha Π α and Π β \Pi_\beta Π β belong to A Z \mathcal{A}_{Z} A Z . We conclude that B \mathcal{B} B is a proper subalgebra of A Z \mathcal{A}_{Z} A Z , therefore reaching an contradiction.
BAE measurement
In this section, we study quantum BAE measurement for the n n n -level quantum system (11). A sufficient condition is given by the following result.
Theorem 6: Consider system (11). Define input field observables, output field observables and output matrix
W g = g T W , Y h = h T Y , C h = h T C . (118) W_g=g^{\mathsf{T}}W,\qquad Y_h=h^{\mathsf{T}}Y,\qquad C_h=h^{\mathsf{T}}C. \tag*{(118)} W g = g T W , Y h = h T Y , C h = h T C . (118) where g ∈ R m × r g g\in\mathbb{R}^{m\times r_g} g ∈ R m × r g and h ∈ R m × r h h\in\mathbb{R}^{m\times r_h} h ∈ R m × r h (r g , r h ≤ m r_g,r_h\le m r g , r h ≤ m ). Assume an orthogonal completion of S = [ g g ⊥ ] S=[g\ \ g_\perp] S = [ g g ⊥ ] satisfies
S T S = I m , S T Ω S = Ω . (119) S^{\mathsf{T}}S=I_m,\qquad S^{\mathsf{T}}\Omega S=\Omega. \tag*{(119)} S T S = I m , S T Ω S = Ω. (119) Define matrices
B g , j = ∑ k = 1 m g k j B k , j = 1 , … , r g . (120) B_{g,j}=\sum_{k=1}^{m}g_{kj}B_k,\qquad j=1,\ldots,r_g. \tag*{(120)} B g , j = k = 1 ∑ m g k j B k , j = 1 , … , r g . (120) Then define a set
R g ≜ span R { M ℓ ⋯ M 1 B g , j v : v ∈ R s , j = 1 , … , r g , M j ∈ { A , B 1 , … , B m } , ℓ = 0 , 1 , … } . (121) \mathcal{R}_g\triangleq\operatorname{span}_{\mathbb{R}}\left\{M_\ell\cdots M_1B_{g,j}v:v\in\mathbb{R}^{s},\ j=1,\ldots,r_g,\ M_j\in\{A,B_1,\ldots,B_m\},\ \ell=0,1,\ldots\right\}. \tag*{(121)} R g ≜ span R { M ℓ ⋯ M 1 B g , j v : v ∈ R s , j = 1 , … , r g , M j ∈ { A , B 1 , … , B m } , ℓ = 0 , 1 , … } . (121) (ℓ = 0 \ell=0 ℓ = 0 means no left multiplication of { A , B 1 , … , B m } \{A,B_1,\ldots,B_m\} { A , B 1 , … , B m } on the vector B g , j v B_{g,j}v B g , j v .) If
h T g = 0 r h × r g (122) h^{\mathsf{T}}g=0_{r_h\times r_g} \tag*{(122)} h T g = 0 r h × r g (122) and
C h R g = { 0 } , (123) C_h\mathcal{R}_g=\{0\}, \tag*{(123)} C h R g = { 0 } , (123) then the output field observables Y h Y_h Y h are back-action evading with respect to the input field observables W g W_g W g ; i.e., W g W_g W g have neither:
a direct feedthrough path to Y h Y_h Y h ; nor
a system-variable-mediated path through the bilinear dynamics.
Proof: Denote W ⊥ = g ⊥ T W W_\perp=g_\perp^{\mathsf{T}}W W ⊥ = g ⊥ T W . By (103), the transformed input field observables
[ W g W ⊥ ] = S T W \begin{bmatrix}W_g\\W_\perp\end{bmatrix}=S^{\mathsf{T}}W [ W g W ⊥ ] = S T W also satisfy (12). Thus, they are genuine quantum input quadratures. Choose a full-row-rank matrix N ∈ R q × s N\in\mathbb{R}^{q\times s} N ∈ R q × s (q ≤ s q\le s q ≤ s ) such that ker N = R g \ker N=\mathcal{R}_g ker N = R g , and define a vector of observables Z = N X Z=NX Z = N X . By definition, R g \mathcal{R}_g R g is invariant under A A A and every B μ B_\mu B μ , hence there exist matrices A ˉ \bar{A} A ˉ and B ˉ μ \bar{B}_\mu B ˉ μ satisfying N A = A ˉ N NA=\bar{A}N N A = A ˉ N and N B μ = B ˉ μ N NB_\mu=\bar{B}_\mu N N B μ = B ˉ μ N , for μ = 1 , … , m \mu=1,\ldots,m μ = 1 , … , m . Equivalently, for the transformed input matrices in (104), there exist matrices B ˉ g , a \bar{B}_{g,a} B ˉ g , a and B ˉ ⊥ , b \bar{B}_{\perp,b} B ˉ ⊥ , b such that N B g , a = B ˉ g , a N NB_{g,a}=\bar{B}_{g,a}N N B g , a = B ˉ g , a N , and N B ⊥ , b = B ˉ ⊥ , b N NB_{\perp,b}=\bar{B}_{\perp,b}N N B ⊥ , b = B ˉ ⊥ , b N . By construction of R g \mathcal{R}_g R g in (105), Range ( B g , a ) ⊆ R g = ker N \operatorname{Range}(B_{g,a})\subseteq\mathcal{R}_g=\ker N Range ( B g , a ) ⊆ R g = ker N . Therefore, N B g , a = 0 NB_{g,a}=0 N B g , a = 0 . Since N N N has full row rank, it has a right inverse N ~ = N T ( N N T ) − 1 \widetilde{N}=N^{\mathsf{T}}(NN^{\mathsf{T}})^{-1} N = N T ( N N T ) − 1 such that N N ~ = I q N\widetilde{N}=I_q N N = I q . Therefore,
B ˉ g , a = B ˉ g , a N N ~ = N B g , a N ~ = 0. (124) \bar{B}_{g,a}=\bar{B}_{g,a}N\widetilde{N}=NB_{g,a}\widetilde{N}=0. \tag*{(124)} B ˉ g , a = B ˉ g , a N N = N B g , a N = 0. (124) Consequently, left-multiplying system (11) by N N N gives
d Z = ( N A + A ˉ Z ) d t + ∑ b = 1 m − r g B ˉ ⊥ , b Z d ( W ⊥ ) b . (125) dZ=(NA+\bar{A}Z)\,dt+\sum_{b=1}^{m-r_g}\bar{B}_{\perp,b}Z\,d(W_\perp)_b. \tag*{(125)} d Z = ( N A + A ˉ Z ) d t + b = 1 ∑ m − r g B ˉ ⊥ , b Z d ( W ⊥ ) b . (125) According to (102), (104) and (108), the increment d W g dW_g d W g does not appear in (109). Next, (107) means ker N = R g ⊆ ker C h \ker N=\mathcal{R}_g\subseteq\ker C_h ker N = R g ⊆ ker C h . Hence, there exists C ˉ h ∈ R r h × q \bar{C}_h\in\mathbb{R}^{r_h\times q} C ˉ h ∈ R r h × q such that C h = C ˉ h N C_h=\bar{C}_hN C h = C ˉ h N . Consequently, C h X = C ˉ h Z C_hX=\bar{C}_hZ C h X = C ˉ h Z . By (102), the selected output satisfies
d Y h = h T d Y = C h X d t + h T d W = C ˉ h Z d t + h T g d W g + h T g ⊥ d W ⊥ . dY_h=h^{\mathsf{T}}dY=C_hX\,dt+h^{\mathsf{T}}dW
=\bar{C}_hZ\,dt+h^{\mathsf{T}}g\,dW_g+h^{\mathsf{T}}g_\perp\,dW_\perp. d Y h = h T d Y = C h X d t + h T d W = C ˉ h Z d t + h T g d W g + h T g ⊥ d W ⊥ . Using (106), we obtain
d Y h = C ˉ h Z d t + h T g ⊥ d W ⊥ . dY_h=\bar{C}_hZ\,dt+h^{\mathsf{T}}g_\perp\,dW_\perp. d Y h = C ˉ h Z d t + h T g ⊥ d W ⊥ . As the dynamics of Z Z Z in (109) have no W g W_g W g -driven term, and the selected output Y h Y_h Y h has no feedthrough W g W_g W g term either, the output observables Y h Y_h Y h are back-action evading against the input observables W g W_g W g . ■
Remark 5: If Y h Y_h Y h is intended to represent a vector of simultaneously measured output fields, one should also require @page 11/16 starts_mid=1 ends_mid=1 Proof: (Sufficiency). The proof is algebraic. By means of (97) and the form of F ℓ F_\ell F ℓ , direct algebraic manipulations yield
∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = ∑ γ = 1 q [ ∑ ℓ = 1 p ( f ℓ γ − f ℓ β ) 2 ] Π γ = ∑ γ = 1 q ∥ f γ − f β ∥ 2 Π γ . \sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2
=
\sum_{\gamma=1}^{q}\left[\sum_{\ell=1}^{p}(f_{\ell\gamma}-f_{\ell\beta})^2\right]\Pi_\gamma
=
\sum_{\gamma=1}^{q}\lVert f_\gamma-f_\beta\rVert^2\Pi_\gamma. ℓ = 1 ∑ p ( F ℓ − f ℓ β I n ) 2 = γ = 1 ∑ q [ ℓ = 1 ∑ p ( f ℓ γ − f ℓ β ) 2 ] Π γ = γ = 1 ∑ q ∥ f γ − f β ∥ 2 Π γ . from which we have
∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 ∥ f α − f β ∥ 2 = ∑ γ = 1 q ∥ f γ − f β ∥ 2 ∥ f α − f β ∥ 2 Π γ . \frac{\sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2}{\lVert f_\alpha-f_\beta\rVert^2}
=
\sum_{\gamma=1}^{q}\frac{\lVert f_\gamma-f_\beta\rVert^2}{\lVert f_\alpha-f_\beta\rVert^2}\Pi_\gamma. ∥ f α − f β ∥ 2 ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = γ = 1 ∑ q ∥ f α − f β ∥ 2 ∥ f γ − f β ∥ 2 Π γ . This, together with
( ∑ γ a γ Π γ ) ( ∑ δ b δ Π δ ) = ∑ γ a γ b γ Π γ , \left(\sum_\gamma a_\gamma\Pi_\gamma\right)\left(\sum_\delta b_\delta\Pi_\delta\right)
=
\sum_\gamma a_\gamma b_\gamma\Pi_\gamma, ( γ ∑ a γ Π γ ) ( δ ∑ b δ Π δ ) = γ ∑ a γ b γ Π γ , yields
∏ β = 1 β ≠ α q ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 ∥ f α − f β ∥ 2 = Π α , α = 1 , … , q . (126) \prod_{\substack{\beta=1\\\beta\ne\alpha}}^{q}
\frac{\sum_{\ell=1}^{p}(F_\ell-f_{\ell\beta}I_n)^2}{\lVert f_\alpha-f_\beta\rVert^2}
=
\Pi_\alpha,\qquad\alpha=1,\ldots,q. \tag*{(126)} β = 1 β = α ∏ q ∥ f α − f β ∥ 2 ∑ ℓ = 1 p ( F ℓ − f ℓ β I n ) 2 = Π α , α = 1 , … , q . (126) where (99) has been used. By (101), all orthogonal projectors Π α \Pi_\alpha Π α can be obtained from the output-signal observables { F 1 , … , F p } \{F_1,\ldots,F_p\} { F 1 , … , F p } .
(Necessity.) We prove it by contradiction. Suppose that f α = f β f_\alpha=f_\beta f α = f β for some α ≠ β \alpha\ne\beta α = β . As the observables F ℓ F_\ell F ℓ mutually commute, every element of B ≜ alg ∗ ( I n , F 1 , … , F p ) \mathcal{B}\triangleq\operatorname{alg}^{*}(I_n,F_1,\ldots,F_p) B ≜ alg ∗ ( I n , F 1 , … , F p ) is a polynomial in them, and, by mean of spectral decomposition such polynomial has the form
P ( F 1 , … , F p ) = ∑ γ = 1 q P ( f γ ) Π γ . P(F_1,\ldots,F_p)=\sum_{\gamma=1}^{q}P(f_\gamma)\Pi_\gamma. P ( F 1 , … , F p ) = γ = 1 ∑ q P ( f γ ) Π γ . If Π α = P ( F 1 , … , F p ) \Pi_\alpha=P(F_1,\ldots,F_p) Π α = P ( F 1 , … , F p ) for some polynomial P ( F 1 , … , F p ) P(F_1,\ldots,F_p) P ( F 1 , … , F p ) , multiplication by Π α \Pi_\alpha Π α and Π β \Pi_\beta Π β , respectively, yield P ( f α ) = 1 P(f_\alpha)=1 P ( f α ) = 1 , and P ( f β ) = 0 P(f_\beta)=0 P ( f β ) = 0 , contradicting f α = f β f_\alpha=f_\beta f α = f β . The same argument excludes Π β \Pi_\beta Π β too. Thus, neither Π α \Pi_\alpha Π α nor Π β \Pi_\beta Π β belongs to B \mathcal{B} B . On the other hand, both Π α \Pi_\alpha Π α and Π β \Pi_\beta Π β belong to A Z \mathcal{A}_{Z} A Z . We conclude that B \mathcal{B} B is a proper subalgebra of A Z \mathcal{A}_{Z} A Z , therefore reaching an contradiction.
BAE measurement In this section, we study quantum BAE measurement for the n n n -level quantum system (11). A sufficient condition is given by the following result.
Theorem 6: Consider system (11). Define input field observables, output field observables and output matrix
W g = g T W , Y h = h T Y , C h = h T C , (127) W_g=g^{\mathsf{T}}W,\qquad Y_h=h^{\mathsf{T}}Y,\qquad C_h=h^{\mathsf{T}}C, \tag*{(127)} W g = g T W , Y h = h T Y , C h = h T C , (127) where g ∈ R m × r g g\in\mathbb{R}^{m\times r_g} g ∈ R m × r g and h ∈ R m × r h h\in\mathbb{R}^{m\times r_h} h ∈ R m × r h (r g , r h ≤ m r_g,r_h\le m r g , r h ≤ m ). Assume an orthogonal completion of S = [ g g ⊥ ] S=[g\ \ g_\perp] S = [ g g ⊥ ] satisfies
S T S = I m , S T Ω S = Ω . (128) S^{\mathsf{T}}S=I_m,\qquad S^{\mathsf{T}}\Omega S=\Omega. \tag*{(128)} S T S = I m , S T Ω S = Ω. (128) Define matrices
B g , j = ∑ k = 1 m g k j B k , j = 1 , … , r g . (129) B_{g,j}=\sum_{k=1}^{m}g_{kj}B_k,\qquad j=1,\ldots,r_g. \tag*{(129)} B g , j = k = 1 ∑ m g k j B k , j = 1 , … , r g . (129) Then define a set
R g ≜ span R { M ℓ ⋯ M 1 B g , j v : v ∈ R s , j = 1 , … , r g , M j ∈ { A , B 1 , … , B m } , ℓ = 0 , 1 , … } . (130) \mathcal{R}_g\triangleq\operatorname{span}_{\mathbb{R}}\left\{M_\ell\cdots M_1B_{g,j}v:v\in\mathbb{R}^{s},\ j=1,\ldots,r_g,\ M_j\in\{A,B_1,\ldots,B_m\},\ \ell=0,1,\ldots\right\}. \tag*{(130)} R g ≜ span R { M ℓ ⋯ M 1 B g , j v : v ∈ R s , j = 1 , … , r g , M j ∈ { A , B 1 , … , B m } , ℓ = 0 , 1 , … } . (130) (ℓ = 0 \ell=0 ℓ = 0 means no left multiplication of { A , B 1 , … , B m } \{A,B_1,\ldots,B_m\} { A , B 1 , … , B m } on the vector B g , j v B_{g,j}v B g , j v .) If
h T g = 0 r h × r g (131) h^{\mathsf{T}}g=0_{r_h\times r_g} \tag*{(131)} h T g = 0 r h × r g (131) and
C h R g = { 0 } , (132) C_h\mathcal{R}_g=\{0\}, \tag*{(132)} C h R g = { 0 } , (132) then the output field observables Y h Y_h Y h are back-action evading with respect to the input field observables W g W_g W g ; i.e., W g W_g W g have neither:
a direct feedthrough path to Y h Y_h Y h ; nor
a system-variable-mediated path through the bilinear dynamics.
Proof: Denote W ⊥ = g ⊥ T W W_\perp=g_\perp^{\mathsf{T}}W W ⊥ = g ⊥ T W . By (103), the transformed input field observables
[ W g W ⊥ ] = S T W \begin{bmatrix}W_g\\W_\perp\end{bmatrix}=S^{\mathsf{T}}W [ W g W ⊥ ] = S T W also satisfy (12). Thus, they are genuine quantum input quadratures. Choose a full-row-rank matrix N ∈ R q × s N\in\mathbb{R}^{q\times s} N ∈ R q × s (q ≤ s q\le s q ≤ s ) such that ker N = R g \ker N=\mathcal{R}_g ker N = R g , and define a vector of observables Z = N X Z=NX Z = N X . By definition, R g \mathcal{R}_g R g is invariant under A A A and every B μ B_\mu B μ , hence there exist matrices A ˉ \bar{A} A ˉ and B ˉ μ \bar{B}_\mu B ˉ μ satisfying N A = A ˉ N NA=\bar{A}N N A = A ˉ N and N B μ = B ˉ μ N NB_\mu=\bar{B}_\mu N N B μ = B ˉ μ N , for μ = 1 , … , m \mu=1,\ldots,m μ = 1 , … , m . Equivalently, for the transformed input matrices in (104), there exist matrices B ˉ g , a \bar{B}_{g,a} B ˉ g , a and B ˉ ⊥ , b \bar{B}_{\perp,b} B ˉ ⊥ , b such that N B g , a = B ˉ g , a N NB_{g,a}=\bar{B}_{g,a}N N B g , a = B ˉ g , a N , and N B ⊥ , b = B ˉ ⊥ , b N NB_{\perp,b}=\bar{B}_{\perp,b}N N B ⊥ , b = B ˉ ⊥ , b N . By construction of R g \mathcal{R}_g R g in (105), Range ( B g , a ) ⊆ R g = ker N \operatorname{Range}(B_{g,a})\subseteq\mathcal{R}_g=\ker N Range ( B g , a ) ⊆ R g = ker N . Therefore, N B g , a = 0 NB_{g,a}=0 N B g , a = 0 . Since N N N has full row rank, it has a right inverse N ~ = N T ( N N T ) − 1 \widetilde{N}=N^{\mathsf{T}}(NN^{\mathsf{T}})^{-1} N = N T ( N N T ) − 1 such that N N ~ = I q N\widetilde{N}=I_q N N = I q . Therefore,
B ˉ g , a = B ˉ g , a N N ~ = N B g , a N ~ = 0. (133) \bar{B}_{g,a}=\bar{B}_{g,a}N\widetilde{N}=NB_{g,a}\widetilde{N}=0. \tag*{(133)} B ˉ g , a = B ˉ g , a N N = N B g , a N = 0. (133) Consequently, left-multiplying system (11) by N N N gives
d Z = ( N A + A ˉ Z ) d t + ∑ b = 1 m − r g B ˉ ⊥ , b Z d ( W ⊥ ) b . (134) dZ=(NA+\bar{A}Z)\,dt+\sum_{b=1}^{m-r_g}\bar{B}_{\perp,b}Z\,d(W_\perp)_b. \tag*{(134)} d Z = ( N A + A ˉ Z ) d t + b = 1 ∑ m − r g B ˉ ⊥ , b Z d ( W ⊥ ) b . (134) According to (102), (104) and (108), the increment d W g dW_g d W g does not appear in (109). Next, (107) means ker N = R g ⊆ ker C h \ker N=\mathcal{R}_g\subseteq\ker C_h ker N = R g ⊆ ker C h . Hence, there exists C ˉ h ∈ R r h × q \bar{C}_h\in\mathbb{R}^{r_h\times q} C ˉ h ∈ R r h × q such that C h = C ˉ h N C_h=\bar{C}_hN C h = C ˉ h N . Consequently, C h X = C ˉ h Z C_hX=\bar{C}_hZ C h X = C ˉ h Z . By (102), the selected output satisfies
d Y h = h T d Y = C h X d t + h T d W = C ˉ h Z d t + h T g d W g + h T g ⊥ d W ⊥ . dY_h=h^{\mathsf{T}}dY=C_hX\,dt+h^{\mathsf{T}}dW
=\bar{C}_hZ\,dt+h^{\mathsf{T}}g\,dW_g+h^{\mathsf{T}}g_\perp\,dW_\perp. d Y h = h T d Y = C h X d t + h T d W = C ˉ h Z d t + h T g d W g + h T g ⊥ d W ⊥ . Using (106), we obtain
d Y h = C ˉ h Z d t + h T g ⊥ d W ⊥ . dY_h=\bar{C}_hZ\,dt+h^{\mathsf{T}}g_\perp\,dW_\perp. d Y h = C ˉ h Z d t + h T g ⊥ d W ⊥ . As the dynamics of Z Z Z in (109) have no W g W_g W g -driven term, and the selected output Y h Y_h Y h has no feedthrough W g W_g W g term either, the output observables Y h Y_h Y h are back-action evading against the input observables W g W_g W g . ■
Remark 5: If Y h Y_h Y h is intended to represent a vector of simultaneously measured output fields, one should also require
h T J h = 0 h^{\mathsf T}Jh=0 h T J h = 0 for the preservation of commutation relations. A common normalization h T h = I r h h^{\mathsf T}h=I_{r_h} h T h = I r h can also be imposed. These conditions concern physical joint measurability of the selected output quadratures Y h Y_h Y h and have nothing to do with the BAE measurement conditions presented in Theorem 6.
Four-qubit-chain illustrative example
This section illustrates the results of Sections III–VI using a four-qubit Heisenberg chain with collective dephasing. The example exhibits three coexisting structures: a logical-qubit observable subsystem (DFS), a conserved commutative magnetization algebra (QND observables), and a homodyne output that evades the conjugate measurement’s back action (BAE measurement). Four-qubit decoherence-free encodings and their control by exchange interactions are established construction; see for example [43 ] , [44 ] , [45 ] .
The model
Let H = ( C 2 ) ⊗ 4 \mathcal{H}=(\mathbb{C}^{2})^{\otimes4} H = ( C 2 ) ⊗ 4 . Then n = 2 4 = 16 n=2^{4}=16 n = 2 4 = 16 and s = n 2 − 1 = 255 s=n^{2}-1=255 s = n 2 − 1 = 255 . For a ∈ { x , y , z } a\in\{x,y,z\} a ∈ { x , y , z } , let σ a ( j ) \sigma_{a}^{(j)} σ a ( j ) denote the Pauli matrix σ a \sigma_a σ a acting on qubit j j j . Define the nearest-neighbor exchange operators E j , j + 1 E_{j,j+1} E j , j + 1 and chain Hamiltonian H c h H_{\mathrm{ch}} H ch as
E j , j + 1 = ∑ a = x , y , z σ a ( j ) σ a ( j + 1 ) , j = 1 , 2 , 3 , E_{j,j+1}=\sum_{a=x,y,z}\sigma_{a}^{(j)}\sigma_{a}^{(j+1)},\qquad j=1,2,3, E j , j + 1 = a = x , y , z ∑ σ a ( j ) σ a ( j + 1 ) , j = 1 , 2 , 3 ,
H c h = J 12 E 12 + J 23 E 23 + J 34 E 34 , H_{\mathrm{ch}}=J_{12}E_{12}+J_{23}E_{23}+J_{34}E_{34}, H ch = J 12 E 12 + J 23 E 23 + J 34 E 34 ,
where J 12 , J 23 , J 34 ∈ R J_{12},J_{23},J_{34}\in\mathbb{R} J 12 , J 23 , J 34 ∈ R . Introduce the collective spin components
J a ≜ 1 2 ∑ j = 1 4 σ a ( j ) , J 2 ≜ J x 2 + J y 2 + J z 2 , a = x , y , z . (135) J_a\triangleq\frac{1}{2}\sum_{j=1}^{4}\sigma_{a}^{(j)},\qquad J^{2}\triangleq J_x^{2}+J_y^{2}+J_z^{2},\qquad a=x,y,z. \tag*{(135)} J a ≜ 2 1 j = 1 ∑ 4 σ a ( j ) , J 2 ≜ J x 2 + J y 2 + J z 2 , a = x , y , z . (135)
It can be readily shown that [ σ a ( j ) + σ a ( j + 1 ) , E j , j + 1 ] = 0 [\sigma_{a}^{(j)}+\sigma_{a}^{(j+1)},E_{j,j+1}]=0 [ σ a ( j ) + σ a ( j + 1 ) , E j , j + 1 ] = 0 and [ σ a ( k ) , E j , j + 1 ] = 0 [\sigma_{a}^{(k)},E_{j,j+1}]=0 [ σ a ( k ) , E j , j + 1 ] = 0 if k ∉ { j , j + 1 } k\notin\{j,j+1\} k ∈ / { j , j + 1 } . Hence we have
[ H c h , J a ] = 0 , [ H c h , J 2 ] = 0 , a = x , y , z . (136) [H_{\mathrm{ch}},J_a]=0,\qquad[H_{\mathrm{ch}},J^{2}]=0,\qquad a=x,y,z. \tag*{(136)} [ H ch , J a ] = 0 , [ H ch , J 2 ] = 0 , a = x , y , z . (136)
Assume a single vacuum field channel with coupling operator L = κ J z L=\sqrt{\kappa}J_z L = κ J z , where κ > 0 \kappa>0 κ > 0 . Let c z ∈ R 255 c_z\in\mathbb{R}^{255} c z ∈ R 255 satisfy Φ ( c z ) = J z \Phi(c_z)=J_z Φ ( c z ) = J z . Then by (32) we get
L = Φ ( κ c z ) J z = ( κ c z ) T X . (137) L=\Phi(\sqrt{\kappa}c_z)J_z=(\sqrt{\kappa}c_z)^{\mathsf T}X. \tag*{(137)} L = Φ ( κ c z ) J z = ( κ c z ) T X . (137)
Consequently, system matrices in (15) are
C 1 = 2 κ c z T , C 2 = 0 , A 0 = 0 , C_1=2\sqrt{\kappa}c_z^{\mathsf T},\qquad C_2=0,\qquad A_0=0, C 1 = 2 κ c z T , C 2 = 0 , A 0 = 0 ,
B 1 = 0 , B 2 = − 2 κ Θ ( c z ) , (138) B_1=0,\qquad B_2=-2\sqrt{\kappa}\Theta(c_z), \tag*{(138)} B 1 = 0 , B 2 = − 2 κ Θ ( c z ) , (138)
and the input–output equations in (11) are
d Y 1 = 2 κ J z d t + d W 1 , d Y 2 = d W 2 . (139) dY_1=2\sqrt{\kappa}J_z\,dt+dW_1,\qquad dY_2=dW_2. \tag*{(139)} d Y 1 = 2 κ J z d t + d W 1 , d Y 2 = d W 2 . (139)
The Y 1 Y_1 Y 1 -quadrature carries the magnetization signal J z J_z J z , whereas the conjugate Y 2 Y_2 Y 2 -quadrature has no system-dependent signal. For H c h H_{\mathrm{ch}} H ch an L L L given above, in the Heisenberg picture, the evolution of the system operator O O O follows the QSDE, [24 Section 2] ,
d O = L ( O ) d t + i [ L , O ] d W 2 , (140) dO=\mathcal{L}(O)dt+i[L,O]dW_2, \tag*{(140)} d O = L ( O ) d t + i [ L , O ] d W 2 , (140)
where the superoperators
L ( O ) = K ( O ) + D L ( O ) , \mathcal{L}(O)=\mathcal{K}(O)+\mathcal{D}_L(O), L ( O ) = K ( O ) + D L ( O ) ,
K ( O ) = − i [ O , H c h ] , D L ( O ) = − 1 2 [ L , [ L , O ] ] . (141) \mathcal{K}(O)=-i[O,H_{\mathrm{ch}}],\qquad\mathcal{D}_L(O)=-\frac{1}{2}[L,[L,O]]. \tag*{(141)} K ( O ) = − i [ O , H ch ] , D L ( O ) = − 2 1 [ L , [ L , O ]] . (141)
Clearly
d J z = 0 , (142) dJ_z=0, \tag*{(142)} d J z = 0 , (142)
i.e., the measured magnetization J z J_z J z is conserved. Moreover, from (112) we have c z T d X = 0 c_z^{\mathsf T}dX=0 c z T d X = 0 . Comparing with (10) and using (113), we get
c z T A = 0 , c z T B μ = 0 , μ = 1 , 2. (143) c_z^{\mathsf T}A=0,\qquad c_z^{\mathsf T}B_\mu=0,\qquad\mu=1,2. \tag*{(143)} c z T A = 0 , c z T B μ = 0 , μ = 1 , 2. (143)
Similarly, given v ∈ R 255 v\in\mathbb{R}^{255} v ∈ R 255 , by means of B 2 = − B 1 T B_2=-B_1^{\mathsf T} B 2 = − B 1 T , (113) and (33), substituting O = Φ ( v ) = v T X O=\Phi(v)=v^{\mathsf T}X O = Φ ( v ) = v T X into (115) and comparing with (10) we obtain
L ( Φ ( v ) ) = Φ ( A T v ) , \mathcal{L}(\Phi(v))=\Phi(A^{\mathsf T}v), L ( Φ ( v )) = Φ ( A T v ) ,
Φ ( B 2 v ) = − i κ [ J z , Φ ( v ) ] , \Phi(B_2v)=-i\sqrt{\kappa}[J_z,\Phi(v)], Φ ( B 2 v ) = − i κ [ J z , Φ ( v )] ,
C 1 v = κ Tr ( J z Φ ( v ) ) . (144) C_1v=\sqrt{\kappa}\operatorname{Tr}(J_z\Phi(v)). \tag*{(144)} C 1 v = κ Tr ( J z Φ ( v )) . (144)
Hence, due to the isometric property of Φ \Phi Φ defined in (32),
B 2 v = 0 ⟺ [ J z , Φ ( v ) ] = 0. (145) B_2v=0\Longleftrightarrow[J_z,\Phi(v)]=0. \tag*{(145)} B 2 v = 0 ⟺ [ J z , Φ ( v )] = 0. (145)
An autonomous logical-qubit observable subsystem (DFS)
The singlet encoding
Define the two-qubit states
∣ s ⟩ = ∣ 01 ⟩ − ∣ 10 ⟩ 2 , ∣ t 0 ⟩ = ∣ 01 ⟩ + ∣ 10 ⟩ 2 , |s\rangle=\frac{|01\rangle-|10\rangle}{\sqrt{2}},\qquad|t_0\rangle=\frac{|01\rangle+|10\rangle}{\sqrt{2}}, ∣ s ⟩ = 2 ∣01 ⟩ − ∣10 ⟩ , ∣ t 0 ⟩ = 2 ∣01 ⟩ + ∣10 ⟩ ,
∣ t + ⟩ = ∣ 00 ⟩ , ∣ t − ⟩ = ∣ 11 ⟩ . (146) |t_+\rangle=|00\rangle,\qquad|t_-\rangle=|11\rangle. \tag*{(146)} ∣ t + ⟩ = ∣00 ⟩ , ∣ t − ⟩ = ∣11 ⟩ . (146)
Then define orthonormal states
∣ 0 L ⟩ = ∣ s ⟩ 12 ∣ s ⟩ 34 , |0_L\rangle=|s\rangle_{12}|s\rangle_{34}, ∣ 0 L ⟩ = ∣ s ⟩ 12 ∣ s ⟩ 34 ,
∣ 1 L ⟩ = 1 3 ( ∣ t + ⟩ 12 ∣ t − ⟩ 34 − ∣ t 0 ⟩ 12 ∣ t 0 ⟩ 34 + ∣ t − ⟩ 12 ∣ t + ⟩ 34 ) . (147) |1_L\rangle=\frac{1}{\sqrt{3}}\left(|t_+\rangle_{12}|t_-\rangle_{34}-|t_0\rangle_{12}|t_0\rangle_{34}+|t_-\rangle_{12}|t_+\rangle_{34}\right). \tag*{(147)} ∣ 1 L ⟩ = 3 1 ( ∣ t + ⟩ 12 ∣ t − ⟩ 34 − ∣ t 0 ⟩ 12 ∣ t 0 ⟩ 34 + ∣ t − ⟩ 12 ∣ t + ⟩ 34 ) . (147)
It can be easily verified that
J a ∣ 0 L ⟩ = J a ∣ 1 L ⟩ = 0 , a = x , y , z . (148) J_a|0_L\rangle=J_a|1_L\rangle=0,\qquad a=x,y,z. \tag*{(148)} J a ∣ 0 L ⟩ = J a ∣ 1 L ⟩ = 0 , a = x , y , z . (148)
As a result, by (110), J 2 ∣ 0 L ⟩ = J 2 ∣ 1 L ⟩ = 0 J^{2}|0_L\rangle=J^{2}|1_L\rangle=0 J 2 ∣ 0 L ⟩ = J 2 ∣ 1 L ⟩ = 0 . Hence, { ∣ 0 L ⟩ , ∣ 1 L ⟩ } \{|0_L\rangle,|1_L\rangle\} { ∣ 0 L ⟩ , ∣ 1 L ⟩} form an orthonormal basis for the two-dimensional total-spin-zero subspace. Moreover, the rank-2 projector
P 0 = ∣ 0 L ⟩ ⟨ 0 L ∣ + ∣ 1 L ⟩ ⟨ 1 L ∣ (149) P_0=|0_L\rangle\langle0_L|+|1_L\rangle\langle1_L| \tag*{(149)} P 0 = ∣ 0 L ⟩ ⟨ 0 L ∣ + ∣ 1 L ⟩ ⟨ 1 L ∣ (149)
is the spectral projector of J 2 J^{2} J 2 associated with eigenvalue zero. Hence the second equation in (111) implies
[ H c h , P 0 ] = 0. (150) [H_{\mathrm{ch}},P_0]=0. \tag*{(150)} [ H ch , P 0 ] = 0. (150)
Moreover, by (123)–(124),
J a P 0 = P 0 J a = 0 , a = x , y , z . J_aP_0=P_0J_a=0,\qquad a=x,y,z. J a P 0 = P 0 J a = 0 , a = x , y , z .
Thus the logical subspace spanned by { ∣ 0 L ⟩ , ∣ 1 L ⟩ } \{|0_L\rangle,|1_L\rangle\} { ∣ 0 L ⟩ , ∣ 1 L ⟩} is invariant under the exchange Hamiltonian H c h H_{\mathrm{ch}} H ch and dark with respect to collective dephasing J z J_z J z .
Define logical Pauli operators on this subspace, extended by zero on its orthogonal complement:
σ ˉ x = ∣ 0 L ⟩ ⟨ 1 L ∣ + ∣ 1 L ⟩ ⟨ 0 L ∣ , σ ˉ y = − i ∣ 0 L ⟩ ⟨ 1 L ∣ + i ∣ 1 L ⟩ ⟨ 0 L ∣ , σ ˉ z = ∣ 0 L ⟩ ⟨ 0 L ∣ − ∣ 1 L ⟩ ⟨ 1 L ∣ . (151) \begin{aligned}
\bar{\sigma}_x &= |0_L\rangle\langle1_L| + |1_L\rangle\langle0_L|,\\
\bar{\sigma}_y &= -i|0_L\rangle\langle1_L| + i|1_L\rangle\langle0_L|,\\
\bar{\sigma}_z &= |0_L\rangle\langle0_L| - |1_L\rangle\langle1_L|. \tag*{(151)}
\end{aligned} σ ˉ x σ ˉ y σ ˉ z = ∣ 0 L ⟩ ⟨ 1 L ∣ + ∣ 1 L ⟩ ⟨ 0 L ∣ , = − i ∣ 0 L ⟩ ⟨ 1 L ∣ + i ∣ 1 L ⟩ ⟨ 0 L ∣ , = ∣ 0 L ⟩ ⟨ 0 L ∣ − ∣ 1 L ⟩ ⟨ 1 L ∣. (151)
By (123)–(124),
P 0 σ ˉ a = σ ˉ a P 0 = σ ˉ a , J z σ ˉ a = σ ˉ a J z = 0 , σ ˉ a σ ˉ b = δ a b P 0 + i ∑ c = x , y , z ϵ a b c σ ˉ c , (152) \begin{aligned}
P_0\bar{\sigma}_a &= \bar{\sigma}_aP_0 = \bar{\sigma}_a,\qquad J_z\bar{\sigma}_a = \bar{\sigma}_aJ_z = 0,\\
\bar{\sigma}_a\bar{\sigma}_b &= \delta_{ab}P_0 + i\sum_{c=x,y,z}\epsilon_{abc}\bar{\sigma}_c, \tag*{(152)}
\end{aligned} P 0 σ ˉ a σ ˉ a σ ˉ b = σ ˉ a P 0 = σ ˉ a , J z σ ˉ a = σ ˉ a J z = 0 , = δ ab P 0 + i c = x , y , z ∑ ϵ ab c σ ˉ c , (152)
where ϵ a b c \epsilon_{abc} ϵ ab c is the completely antisymmetric Levi–Civita tensor [37 ] . Direct, yet tedious, calculation on this logical basis gives
P 0 E 12 P 0 = P 0 E 34 P 0 = − P 0 − 2 σ ˉ z , P 0 E 23 P 0 = − P 0 + σ ˉ z + 3 σ ˉ x . (153) \begin{aligned}
P_0E_{12}P_0 &= P_0E_{34}P_0 = -P_0 - 2\bar{\sigma}_z,\\
P_0E_{23}P_0 &= -P_0 + \bar{\sigma}_z + \sqrt{3}\bar{\sigma}_x. \tag*{(153)}
\end{aligned} P 0 E 12 P 0 P 0 E 23 P 0 = P 0 E 34 P 0 = − P 0 − 2 σ ˉ z , = − P 0 + σ ˉ z + 3 σ ˉ x . (153)
Consequently, the restricted Hamiltonian is
P 0 H c h P 0 = h 0 P 0 + h x σ ˉ x + h z σ ˉ z , h 0 = − ( J 12 + J 23 + J 34 ) , h x = 3 J 23 , h z = J 23 − 2 J 12 − 2 J 34 . (154) \begin{aligned}
P_0H_{\mathrm{ch}}P_0 &= h_0P_0 + h_x\bar{\sigma}_x + h_z\bar{\sigma}_z,\\
h_0 &= -(J_{12}+J_{23}+J_{34}),\\
h_x &= \sqrt{3}J_{23},\\
h_z &= J_{23}-2J_{12}-2J_{34}. \tag*{(154)}
\end{aligned} P 0 H ch P 0 h 0 h x h z = h 0 P 0 + h x σ ˉ x + h z σ ˉ z , = − ( J 12 + J 23 + J 34 ) , = 3 J 23 , = J 23 − 2 J 12 − 2 J 34 . (154)
b) Verification of strong alignment: Denote
Q L = P 0 − 1 8 I 16 . Q_L=P_0-\frac{1}{8}I_{16}. Q L = P 0 − 8 1 I 16 .
Then by (152) we get
Tr ( σ ˉ a σ ˉ b ) = 2 δ a b , [ σ ˉ a , σ ˉ b ] = 2 i ∑ c = x , y , z ϵ a b c σ ˉ c , { σ ˉ a , σ ˉ b } = 2 δ a b P 0 = 1 4 δ a b I 16 + 2 δ a b Q L . (155) \begin{aligned}
\operatorname{Tr}(\bar{\sigma}_a\bar{\sigma}_b) &= 2\delta_{ab},\\
[\bar{\sigma}_a,\bar{\sigma}_b] &= 2i\sum_{c=x,y,z}\epsilon_{abc}\bar{\sigma}_c,\\
\{\bar{\sigma}_a,\bar{\sigma}_b\} &= 2\delta_{ab}P_0=\frac{1}{4}\delta_{ab}I_{16}+2\delta_{ab}Q_L. \tag*{(155)}
\end{aligned} Tr ( σ ˉ a σ ˉ b ) [ σ ˉ a , σ ˉ b ] { σ ˉ a , σ ˉ b } = 2 δ ab , = 2 i c = x , y , z ∑ ϵ ab c σ ˉ c , = 2 δ ab P 0 = 4 1 δ ab I 16 + 2 δ ab Q L . (155)
Define real coordinate subspaces S , Q ⊂ R 255 S,Q\subset\mathbb{R}^{255} S , Q ⊂ R 255 by
Φ ( S ) = span R { σ ˉ x , σ ˉ y , σ ˉ z } , Φ ( Q ) = span R { Q L } . (156) \Phi(S)=\operatorname{span}_{\mathbb{R}}\{\bar{\sigma}_x,\bar{\sigma}_y,\bar{\sigma}_z\},\qquad
\Phi(Q)=\operatorname{span}_{\mathbb{R}}\{Q_L\}. \tag*{(156)} Φ ( S ) = span R { σ ˉ x , σ ˉ y , σ ˉ z } , Φ ( Q ) = span R { Q L } . (156)
Let Φ ( s a ) = σ ˉ a \Phi(s_a)=\bar{\sigma}_a Φ ( s a ) = σ ˉ a for a = x , y , z a=x,y,z a = x , y , z , and Φ ( q L ) = Q L \Phi(q_L)=Q_L Φ ( q L ) = Q L . Since Tr ( P 0 ) = 2 \operatorname{Tr}(P_0)=2 Tr ( P 0 ) = 2 , Q L Q_L Q L is traceless Hermitian. By (33) and (155) , normalized Hilbert–Schmidt inner products give
s a T s b = δ a b , q L T s a = 0 , q L T q L = 7 8 . (157) s_a^{\mathsf T}s_b=\delta_{ab},\qquad q_L^{\mathsf T}s_a=0,\qquad q_L^{\mathsf T}q_L=\frac{7}{8}. \tag*{(157)} s a T s b = δ ab , q L T s a = 0 , q L T q L = 8 7 . (157)
Consequently, Q ⊆ S ⊥ Q\subseteq S^\perp Q ⊆ S ⊥ . Define the projector
P S = ∑ a = x , y , z s a s a T (158) P_S=\sum_{a=x,y,z}s_as_a^{\mathsf T} \tag*{(158)} P S = a = x , y , z ∑ s a s a T (158)
on the subspace Q Q Q . By (35), (155) , and the facts Q L σ ˉ a = 7 8 σ ˉ a Q_L\bar{\sigma}_a=\frac{7}{8}\bar{\sigma}_a Q L σ ˉ a = 8 7 σ ˉ a and Q L 2 = 3 4 Q L + 7 64 I 16 Q_L^2=\frac{3}{4}Q_L+\frac{7}{64}I_{16} Q L 2 = 4 3 Q L + 64 7 I 16 , we obtain
f ( s a , s b ) = − ∑ c ϵ a b c s c , d ( s a , s b ) = δ a b q L , f ( q L , S ⊕ Q ) = { 0 } , d ( q L , s a ) = 7 8 s a , d ( q L , q L ) = 3 4 q L . (159) \begin{aligned}
f(s_a,s_b)&=-\sum_c\epsilon_{abc}s_c,\qquad d(s_a,s_b)=\delta_{ab}q_L,\\
f(q_L,S\oplus Q)&=\{0\},\qquad d(q_L,s_a)=\frac{7}{8}s_a,\qquad d(q_L,q_L)=\frac{3}{4}q_L. \tag*{(159)}
\end{aligned} f ( s a , s b ) f ( q L , S ⊕ Q ) = − c ∑ ϵ ab c s c , d ( s a , s b ) = δ ab q L , = { 0 } , d ( q L , s a ) = 8 7 s a , d ( q L , q L ) = 4 3 q L . (159)
These identities verify (41) in Definition 2.
By (156) , the associated unital C ∗ \mathbb{C}^* C ∗ -algebra in (46) is
A = span C { I 16 , Q L , σ ˉ x , σ ˉ y , σ ˉ z } = P 0 M 16 ( C ) P 0 ⊕ C ( I 16 − P 0 ) . (160) \begin{aligned}
\mathcal{A}&=\operatorname{span}_{\mathbb{C}}\{I_{16},Q_L,\bar{\sigma}_x,\bar{\sigma}_y,\bar{\sigma}_z\}\\
&=P_0M_{16}(\mathbb{C})P_0\oplus\mathbb{C}(I_{16}-P_0). \tag*{(160)}
\end{aligned} A = span C { I 16 , Q L , σ ˉ x , σ ˉ y , σ ˉ z } = P 0 M 16 ( C ) P 0 ⊕ C ( I 16 − P 0 ) . (160)
where the facts span C { I 16 , Q L } = span C { P 0 , I 16 − P 0 } \operatorname{span}_{\mathbb{C}}\{I_{16},Q_L\}=\operatorname{span}_{\mathbb{C}}\{P_0,I_{16}-P_0\} span C { I 16 , Q L } = span C { P 0 , I 16 − P 0 } and P 0 M 16 ( C ) P 0 = span C { P 0 , σ ˉ x , σ ˉ y , σ ˉ z } P_0M_{16}(\mathbb{C})P_0=\operatorname{span}_{\mathbb{C}}\{P_0,\bar{\sigma}_x,\bar{\sigma}_y,\bar{\sigma}_z\} P 0 M 16 ( C ) P 0 = span C { P 0 , σ ˉ x , σ ˉ y , σ ˉ z } have been used to establish the second equality. Clearly, the center Z ( A ) = span C { I 16 , Q L } Z(\mathcal{A})=\operatorname{span}_{\mathbb{C}}\{I_{16},Q_L\} Z ( A ) = span C { I 16 , Q L } . Using (157) and the traceless property of the logical Pauli operators, we obtain
Z ( A ) s a ∩ { I 16 } ⊥ = Φ ( Q ) , A s a ∩ Z ( A ) s a ⊥ = Φ ( S ) . (161) Z(\mathcal{A})_{\mathrm{sa}}\cap\{I_{16}\}^{\perp}=\Phi(Q),\qquad
\mathcal{A}_{\mathrm{sa}}\cap Z(\mathcal{A})_{\mathrm{sa}}^{\perp}=\Phi(S). \tag*{(161)} Z ( A ) sa ∩ { I 16 } ⊥ = Φ ( Q ) , A sa ∩ Z ( A ) sa ⊥ = Φ ( S ) . (161)
As a result, the required exact central–noncentral separation (43) also holds, and S S S is strongly aligned.
c) Verification of (55): By (152) , we get [ J z , Φ ( s a ) ] = 0 [J_z,\Phi(s_a)]=0 [ J z , Φ ( s a )] = 0 . Then by (145) and (138) , we have B μ s a = 0 B_\mu s_a=0 B μ s a = 0 . Moreover, by (152) , c z ⊥ s a c_z\perp s_a c z ⊥ s a and hence C 1 s a = 0 C_1s_a=0 C 1 s a = 0 . Finally, C 2 = 0 C_2=0 C 2 = 0 and A 0 = 0 A_0=0 A 0 = 0 are already given in (138) . Consequently, (55b)–(55d) hold.
Next, we look at the drift condition (55a). From (150) , (152) , (154) , and (141) , we get
K ( σ ˉ a ) = K ( P 0 σ ˉ a P 0 ) = i [ h x σ ˉ x + h z σ ˉ z , σ ˉ a ] ∈ Φ ( S ) , D L ( σ ˉ a ) = 0. (162) \begin{aligned}
\mathcal{K}(\bar{\sigma}_a)&=\mathcal{K}(P_0\bar{\sigma}_aP_0)
=i[h_x\bar{\sigma}_x+h_z\bar{\sigma}_z,\bar{\sigma}_a]\in\Phi(S),\\
D_L(\bar{\sigma}_a)&=0. \tag*{(162)}
\end{aligned} K ( σ ˉ a ) D L ( σ ˉ a ) = K ( P 0 σ ˉ a P 0 ) = i [ h x σ ˉ x + h z σ ˉ z , σ ˉ a ] ∈ Φ ( S ) , = 0. (162)
Hence L ( Φ ( S ) ) ⊆ Φ ( S ) \mathcal{L}(\Phi(S))\subseteq\Phi(S) L ( Φ ( S )) ⊆ Φ ( S ) . For X , Y ∈ Herm 0 ( 16 ) X,Y\in\operatorname{Herm}_0(16) X , Y ∈ Herm 0 ( 16 ) ,
⟨ X , K ( Y ) ⟩ n H S = − i 2 Tr ( X Y H c h − X H c h Y ) = 1 2 Tr ( i ( X H c h − H c h X ) Y ) = − ⟨ K ( X ) , Y ⟩ n H S . \begin{aligned}
\langle X,\mathcal{K}(Y)\rangle_{\mathrm{nHS}}
&=-\frac{i}{2}\operatorname{Tr}(XYH_{\mathrm{ch}}-XH_{\mathrm{ch}}Y)\\
&=\frac{1}{2}\operatorname{Tr}(i(XH_{\mathrm{ch}}-H_{\mathrm{ch}}X)Y)
=-\langle\mathcal{K}(X),Y\rangle_{\mathrm{nHS}}.
\end{aligned} ⟨ X , K ( Y ) ⟩ nHS = − 2 i Tr ( X Y H ch − X H ch Y ) = 2 1 Tr ( i ( X H ch − H ch X ) Y ) = − ⟨ K ( X ) , Y ⟩ nHS .
That is, the Hamiltonian superoperator K \mathcal{K} K is skew-adjoint with respect to the normalized Hilbert–Schmidt inner product (8). Similarly, from ⟨ X , D L ( Y ) ⟩ n H S = ⟨ D L ( X ) , Y ⟩ n H S \langle X,D_L(Y)\rangle_{\mathrm{nHS}}=\langle D_L(X),Y\rangle_{\mathrm{nHS}} ⟨ X , D L ( Y ) ⟩ nHS = ⟨ D L ( X ) , Y ⟩ nHS we know that the Lindblad superoperator D L D_L D L is self-adjoint. Now let X ∈ Φ ( S ) X\in\Phi(S) X ∈ Φ ( S ) and Y ∈ Φ ( S ) ⊥ Y\in\Phi(S)^\perp Y ∈ Φ ( S ) ⊥ , the orthogonal complement being taken in Herm 0 ( 16 ) \operatorname{Herm}_0(16) Herm 0 ( 16 ) . Then by (162)
⟨ X , L ( Y ) ⟩ n H S = − ⟨ K ( X ) , Y ⟩ n H S + ⟨ D L ( X ) , Y ⟩ n H S = 0. \langle X,\mathcal{L}(Y)\rangle_{\mathrm{nHS}}
=-\langle\mathcal{K}(X),Y\rangle_{\mathrm{nHS}}
+\langle D_L(X),Y\rangle_{\mathrm{nHS}}=0. ⟨ X , L ( Y ) ⟩ nHS = − ⟨ K ( X ) , Y ⟩ nHS + ⟨ D L ( X ) , Y ⟩ nHS = 0.
Thus L ( Φ ( S ) ⊥ ) ⊆ Φ ( S ) ⊥ \mathcal{L}(\Phi(S)^\perp)\subseteq\Phi(S)^\perp L ( Φ ( S ) ⊥ ) ⊆ Φ ( S ) ⊥ . Since Φ \Phi Φ is an isometric isomorphism, Φ ( S ) ⊥ = Φ ( S ⊥ ) \Phi(S)^\perp=\Phi(S^\perp) Φ ( S ) ⊥ = Φ ( S ⊥ ) , and by (144) and (162) we get A T S ⊆ S A^{\mathsf T}S\subseteq S A T S ⊆ S and A T S ⊥ ⊆ S ⊥ A^{\mathsf T}S^\perp\subseteq S^\perp A T S ⊥ ⊆ S ⊥ ; i.e., [ A T , P S ] = 0 [A^{\mathsf T},P_S]=0 [ A T , P S ] = 0 , which is equivalent to [ A , P S ] = 0 [A,P_S]=0 [ A , P S ] = 0 . This is (55a). Collecting the preceding conclusions yields precisely (55). Thus, all hypotheses of Theorem 2 are satisfied.
d) Artin–Wedderburn decomposition and DFS: Choose U ∈ S U ( 16 ) U\in SU(16) U ∈ S U ( 16 ) such that U ∣ 1 L ⟩ = e 1 U|1_L\rangle=e_1 U ∣ 1 L ⟩ = e 1 , U ∣ 0 L ⟩ = e 2 U|0_L\rangle=e_2 U ∣ 0 L ⟩ = e 2 , and the orthogonal complement Range ( I 16 − P 0 ) \operatorname{Range}(I_{16}-P_0) Range ( I 16 − P 0 ) is mapped onto span { e 3 , … , e 16 } \operatorname{span}\{e_3,\ldots,e_{16}\} span { e 3 , … , e 16 } . With the sign convention for the generalized Gell–Mann basis used in Section II, we have
U σ ˉ x U † = u 12 , U σ ˉ y U † = v 12 , U σ ˉ z U † = w 1 , U\bar{\sigma}_xU^\dagger=u_{12},\qquad U\bar{\sigma}_yU^\dagger=v_{12},\qquad U\bar{\sigma}_zU^\dagger=w_1, U σ ˉ x U † = u 12 , U σ ˉ y U † = v 12 , U σ ˉ z U † = w 1 ,
U P 0 U † = I 2 ⊕ 0 14 , U Q L U † = 7 8 I 2 ⊕ ( − 1 8 ) I 14 . (163) UP_0U^\dagger=I_2\oplus0_{14},\qquad UQ_LU^\dagger=\frac{7}{8}I_2\oplus\left(-\frac{1}{8}\right)I_{14}. \tag*{(163)} U P 0 U † = I 2 ⊕ 0 14 , U Q L U † = 8 7 I 2 ⊕ ( − 8 1 ) I 14 . (163)
Consequently, applying unitary conjugation to A \mathcal{A} A in the second equality in (160) and Φ ( S ) \Phi(S) Φ ( S ) and Φ ( Q ) \Phi(Q) Φ ( Q ) in (156) and using (53), we get
U A U † = M 2 ( C ) ⊕ C I 14 , U\mathcal{A}U^\dagger=M_2(\mathbb{C})\oplus\mathbb{C}I_{14}, U A U † = M 2 ( C ) ⊕ C I 14 ,
D = Herm 0 ( 2 ) ⊕ 0 14 = span R { u 12 , v 12 , w 1 } , \mathcal{D}=\operatorname{Herm}_0(2)\oplus0_{14}
=\operatorname{span}_{\mathbb{R}}\{u_{12},v_{12},w_1\}, D = Herm 0 ( 2 ) ⊕ 0 14 = span R { u 12 , v 12 , w 1 } ,
J 0 = span R { I 2 ⊕ ( − I 14 ) } . (164) \mathcal{J}_0=\operatorname{span}_{\mathbb{R}}\{I_2\oplus(-I_{14})\}. \tag*{(164)} J 0 = span R { I 2 ⊕ ( − I 14 )} . (164)
The Artin–Wedderburn data are ( r 1 , m 1 ) = ( 2 , 1 ) (r_1,m_1)=(2,1) ( r 1 , m 1 ) = ( 2 , 1 ) and ( r 2 , m 2 ) = ( 1 , 14 ) (r_2,m_2)=(1,14) ( r 2 , m 2 ) = ( 1 , 14 ) . Moreover, dim D = 3 \dim\mathcal{D}=3 dim D = 3 and dim J 0 = 1 \dim\mathcal{J}_0=1 dim J 0 = 1 , corresponding exactly to S S S and Q Q Q .
Relabel the original Gell–Mann matrices so that λ 1 = u 12 \lambda_1 = u_{12} λ 1 = u 12 , λ 2 = v 12 \lambda_2 = v_{12} λ 2 = v 12 , and λ 3 = w 1 \lambda_3 = w_1 λ 3 = w 1 . Then by (164) , D = span R { λ 1 , λ 2 , λ 3 } D = \operatorname{span}_{\mathbb{R}}\{\lambda_1,\lambda_2,\lambda_3\} D = span R { λ 1 , λ 2 , λ 3 } . Thus, the unique noncommutative block has multiplicity m 1 = 1 m_1 = 1 m 1 = 1 and occupies the leading corner, so the conditions of Corollary 1 hold. Writing σ ˉ = ( σ ˉ x , σ ˉ y , σ ˉ z ) T \bar{\sigma} = (\bar{\sigma}_x,\bar{\sigma}_y,\bar{\sigma}_z)^{\mathsf{T}} σ ˉ = ( σ ˉ x , σ ˉ y , σ ˉ z ) T , direct calculation by (162) and (155) gives the DFS:
d σ ˉ = A S σ ˉ d t , \mathrm{d}\bar{\sigma}=A_S\bar{\sigma}\,\mathrm{d}t, d σ ˉ = A S σ ˉ d t ,
A S = T S T A T S = [ 0 − 2 h z 0 2 h z 0 − 2 h x 0 2 h x 0 ] . (165) A_S=T_S^{\mathsf{T}}A T_S=
\begin{bmatrix}
0 & -2h_z & 0\\
2h_z & 0 & -2h_x\\
0 & 2h_x & 0
\end{bmatrix}. \tag*{(165)} A S = T S T A T S = 0 2 h z 0 − 2 h z 0 2 h x 0 − 2 h x 0 . (165)
On the encoded subsystem σ ˉ \bar{\sigma} σ ˉ , the Bloch vector precesses about ( h x , 0 , h z ) (h_x,0,h_z) ( h x , 0 , h z ) with angular frequency 2 h x 2 + h z 2 2\sqrt{h_x^2+h_z^2} 2 h x 2 + h z 2 . We have shown that the transformed system has a DFS of the form in (72). We verified Corollary 1.
The conserved QND magnetization algebra
The spectra of J z J_z J z are { − 2 , − 1 , 0 , 1 , 2 } \{-2,-1,0,1,2\} { − 2 , − 1 , 0 , 1 , 2 } . Let Π m \Pi_m Π m denote the spectral projector corresponding to the spectrum m m m . Calculations show that the rank of Π m \Pi_m Π m is
d m = Tr ( Π m ) = ( 4 2 − m ) , m = − 2 , − 1 , 0 , 1 , 2. (166) d_m=\operatorname{Tr}(\Pi_m)=\binom{4}{2-m},\qquad m=-2,-1,0,1,2. \tag*{(166)} d m = Tr ( Π m ) = ( 2 − m 4 ) , m = − 2 , − 1 , 0 , 1 , 2. (166)
Define traceless observables
Q m ≜ Π m − d m 16 I 16 , m = − 2 , − 1 , 0 , 1 , 2 , (167) Q_m\triangleq\Pi_m-\frac{d_m}{16}I_{16},\qquad m=-2,-1,0,1,2, \tag*{(167)} Q m ≜ Π m − 16 d m I 16 , m = − 2 , − 1 , 0 , 1 , 2 , (167)
and their coordinate vectors by Φ ( v m ) = Q m \Phi(v_m)=Q_m Φ ( v m ) = Q m . Denote
V = [ v − 2 v − 1 v 0 v 1 ] ∈ R 255 × 4 , V=[\,v_{-2}\ v_{-1}\ v_0\ v_1\,]\in\mathbb{R}^{255\times4}, V = [ v − 2 v − 1 v 0 v 1 ] ∈ R 255 × 4 ,
Z = V T X = [ Q − 2 , Q − 1 , Q 0 , Q 1 ] T . (168) Z=V^{\mathsf{T}}X=[Q_{-2},Q_{-1},Q_0,Q_1]^{\mathsf{T}}. \tag*{(168)} Z = V T X = [ Q − 2 , Q − 1 , Q 0 , Q 1 ] T . (168)
As Π m \Pi_m Π m are mutually orthogonal, simple algebraic manipulations yield that Q − 2 , Q − 1 , Q 0 , Q 1 Q_{-2},Q_{-1},Q_0,Q_1 Q − 2 , Q − 1 , Q 0 , Q 1 are linearly independent. As a result, the matrix V V V has full column rank.
a) Commutativity and conservation: As spectral projectors Π m \Pi_m Π m are mutually orthogonal, Q − 2 , Q − 1 , Q 0 , Q 1 Q_{-2},Q_{-1},Q_0,Q_1 Q − 2 , Q − 1 , Q 0 , Q 1 mutually commute too. From (167) we have
Q m Q ℓ = ( δ m ℓ − d ℓ 16 ) Q m − d m 16 Q ℓ + d m 16 ( δ m ℓ − d ℓ 16 ) I 16 . (169) Q_mQ_\ell=\left(\delta_{m\ell}-\frac{d_\ell}{16}\right)Q_m-\frac{d_m}{16}Q_\ell+\frac{d_m}{16}\left(\delta_{m\ell}-\frac{d_\ell}{16}\right)I_{16}. \tag*{(169)} Q m Q ℓ = ( δ m ℓ − 16 d ℓ ) Q m − 16 d m Q ℓ + 16 d m ( δ m ℓ − 16 d ℓ ) I 16 . (169)
Moreover, by (33) , v m T v ℓ = d m 2 ( δ m ℓ − d ℓ / 16 ) v_m^{\mathsf{T}}v_\ell=\frac{d_m}{2}\left(\delta_{m\ell}-d_\ell/16\right) v m T v ℓ = 2 d m ( δ m ℓ − d ℓ /16 ) . Consequently, comparing (169) with the product formula (37) yields
f ( v m , v ℓ ) = 0 , f(v_m,v_\ell)=0, f ( v m , v ℓ ) = 0 ,
d ( v m , v ℓ ) = ( δ m ℓ − d ℓ 16 ) v m − d m 16 v ℓ ∈ Range ( V ) . (170) d(v_m,v_\ell)=\left(\delta_{m\ell}-\frac{d_\ell}{16}\right)v_m-\frac{d_m}{16}v_\ell\in\operatorname{Range}(V). \tag*{(170)} d ( v m , v ℓ ) = ( δ m ℓ − 16 d ℓ ) v m − 16 d m v ℓ ∈ Range ( V ) . (170)
These are the hypotheses of Theorem 4(i)–(ii). Clearly,
A Z = span C { I 16 , Q − 2 , Q − 1 , Q 0 , Q 1 } = span C { Π − 2 , Π − 1 , Π 0 , Π 1 , Π 2 } (171) \mathcal{A}_Z=\operatorname{span}_{\mathbb{C}}\{I_{16},Q_{-2},Q_{-1},Q_0,Q_1\}
=\operatorname{span}_{\mathbb{C}}\{\Pi_{-2},\Pi_{-1},\Pi_0,\Pi_1,\Pi_2\} \tag*{(171)} A Z = span C { I 16 , Q − 2 , Q − 1 , Q 0 , Q 1 } = span C { Π − 2 , Π − 1 , Π 0 , Π 1 , Π 2 } (171)
is a five-dimensional commutative unital C ∗ \mathrm{C}^* C ∗ -algebra. A set of its minimal projectors is precisely { Π − 2 , Π − 1 , Π 0 , Π 1 , Π 2 } \{\Pi_{-2},\Pi_{-1},\Pi_0,\Pi_1,\Pi_2\} { Π − 2 , Π − 1 , Π 0 , Π 1 , Π 2 } . This verifies (97) .
Because J z J_z J z commutes with H c h H_{\mathrm{ch}} H ch and L L L , each Q m Q_m Q m defined in (167) commutes with H c h H_{\mathrm{ch}} H ch and L L L too. As a result, (141) and (144) give
V T A = 0 , V T B μ = 0 , V T A 0 = 0. (172) V^{\mathsf{T}}A=0,\qquad V^{\mathsf{T}}B_\mu=0,\qquad V^{\mathsf{T}}A_0=0. \tag*{(172)} V T A = 0 , V T B μ = 0 , V T A 0 = 0. (172)
The conditions in Theorem 4(iii) therefore hold with K = 0 K=0 K = 0 and b = 0 b=0 b = 0 . Consequently (100) becomes Z ( t ) ≡ Z ( 0 ) Z(t)\equiv Z(0) Z ( t ) ≡ Z ( 0 ) , and (102) holds. Thus, the complete magnetization QND algebra A Z \mathcal{A}_Z A Z is conserved, not merely its generator J z J_z J z .
b) Linear observability: Choose M = [ 1 0 ] M=[\,1\ 0\,] M = [ 1 0 ] . Then Y M = Y 1 Y_M=Y_1 Y M = Y 1 and M J M T = 0 MJM^{\mathsf{T}}=0 M J M T = 0 . Using
J z = ∑ m m Π m , ∑ m Q m = 0 , J_z=\sum_m m\Pi_m,\qquad\sum_m Q_m=0, J z = m ∑ m Π m , m ∑ Q m = 0 ,
and (166) –(168) , we get
J z = − 4 Q − 2 − 3 Q − 1 − 2 Q 0 − Q 1 = γ T Z , (173) J_z=-4Q_{-2}-3Q_{-1}-2Q_0-Q_1=\gamma^{\mathsf{T}}Z, \tag*{(173)} J z = − 4 Q − 2 − 3 Q − 1 − 2 Q 0 − Q 1 = γ T Z , (173)
where γ = [ − 4 − 3 − 2 − 1 ] T \gamma=[-4\ -3\ -2\ -1]^{\mathsf{T}} γ = [ − 4 − 3 − 2 − 1 ] T . Hence c z = V γ c_z=V\gamma c z = V γ and M C = C 1 = 2 κ γ T V T MC=C_1=2\sqrt{\kappa}\gamma^{\mathsf{T}}V^{\mathsf{T}} M C = C 1 = 2 κ γ T V T . This verifies (104) , as required in Theorem 4(iv). Also, (105) –(106) become
d Y 1 = C Z Z d t + d W 1 , C Z = 2 κ γ T . (174) \mathrm{d}Y_1=C_ZZ\,\mathrm{d}t+\mathrm{d}W_1,\qquad C_Z=2\sqrt{\kappa}\gamma^{\mathsf{T}}. \tag*{(174)} d Y 1 = C Z Z d t + d W 1 , C Z = 2 κ γ T . (174)
Since K = 0 K=0 K = 0 , by (107) we get Rank O Z = Rank C Z = 1 < 4 \operatorname{Rank}\mathcal{O}_Z=\operatorname{Rank}C_Z=1<4 Rank O Z = Rank C Z = 1 < 4 . Hence, the vector Z Z Z is not observable from Y 1 Y_1 Y 1 in the standard linear-systems sense.
c) Spectral separation: Let F = C Z Z = 2 κ J z F=C_ZZ=2\sqrt{\kappa}J_z F = C Z Z = 2 κ J z , where (173) –(174) have been used. Clearly, F F F has five distinct spectral values f m = 2 κ m f_m=2\sqrt{\kappa}m f m = 2 κ m . (115) holds naturally. The spectral decomposition of F F F is
F = ∑ k = − 2 2 f k Π k , f k = 2 κ k . (175) F=\sum_{k=-2}^{2}f_k\Pi_k,\qquad f_k=2\sqrt{\kappa}k. \tag*{(175)} F = k = − 2 ∑ 2 f k Π k , f k = 2 κ k . (175)
Define degree-4 polynomials
p m ( x ) ≜ ∏ ℓ = − 2 ℓ ≠ m 2 x − f ℓ f m − f ℓ , m = − 2 , … , 2. (176) p_m(x)\triangleq\prod_{\substack{\ell=-2\\\ell\ne m}}^{2}\frac{x-f_\ell}{f_m-f_\ell},\qquad m=-2,\ldots,2. \tag*{(176)} p m ( x ) ≜ ℓ = − 2 ℓ = m ∏ 2 f m − f ℓ x − f ℓ , m = − 2 , … , 2. (176)
Clearly, p m ( f k ) = δ m k p_m(f_k)=\delta_{mk} p m ( f k ) = δ mk . To evaluate these polynomials at the matrix F F F , we rewrite p m ( x ) p_m(x) p m ( x ) as
p m ( x ) = ∑ r = 0 4 a r ( m ) x r . (177) p_m(x)=\sum_{r=0}^{4}a_r^{(m)}x^r. \tag*{(177)} p m ( x ) = r = 0 ∑ 4 a r ( m ) x r . (177)
On the other hand, by (175) ,
F r = ∑ k = − 2 2 f k r Π k , r ∈ Z . (178) F^r=\sum_{k=-2}^{2}f_k^r\Pi_k,\qquad r\in\mathbb{Z}. \tag*{(178)} F r = k = − 2 ∑ 2 f k r Π k , r ∈ Z . (178)
Using (177) –(178) , we obtain
p m ( F ) = ∑ r = 0 4 a r ( m ) F r = ∑ r = 0 4 a r ( m ) ∑ k = − 2 2 f k r Π k = ∑ k = − 2 2 ( ∑ r = 0 4 a r ( m ) f k r ) Π k = ∑ k = − 2 2 p m ( f k ) Π k = ∑ k = − 2 2 δ m k Π k = Π m . (179) \begin{aligned}
p_m(F)&=\sum_{r=0}^{4}a_r^{(m)}F^r=\sum_{r=0}^{4}a_r^{(m)}\sum_{k=-2}^{2}f_k^r\Pi_k\\
&=\sum_{k=-2}^{2}\left(\sum_{r=0}^{4}a_r^{(m)}f_k^r\right)\Pi_k=\sum_{k=-2}^{2}p_m(f_k)\Pi_k\\
&=\sum_{k=-2}^{2}\delta_{mk}\Pi_k=\Pi_m. \tag*{(179)}
\end{aligned} p m ( F ) = r = 0 ∑ 4 a r ( m ) F r = r = 0 ∑ 4 a r ( m ) k = − 2 ∑ 2 f k r Π k = k = − 2 ∑ 2 ( r = 0 ∑ 4 a r ( m ) f k r ) Π k = k = − 2 ∑ 2 p m ( f k ) Π k = k = − 2 ∑ 2 δ mk Π k = Π m . (179)
By (176) , (179) and the identity I 16 = ∑ m = − 2 2 Π m I_{16}=\sum_{m=-2}^{2}\Pi_m I 16 = ∑ m = − 2 2 Π m we get
Π m = p m ( F ) = ∏ ℓ = − 2 ℓ ≠ m 2 F − f ℓ I 16 f m − f ℓ = ∏ ℓ = − 2 ℓ ≠ m 2 F − 2 κ ℓ I 16 2 κ ( m − ℓ ) \Pi_m=p_m(F)=\prod_{\substack{\ell=-2\\\ell\ne m}}^{2}\frac{F-f_\ell I_{16}}{f_m-f_\ell}=\prod_{\substack{\ell=-2\\\ell\ne m}}^{2}\frac{F-2\sqrt{\kappa}\ell I_{16}}{2\sqrt{\kappa}(m-\ell)} Π m = p m ( F ) = ℓ = − 2 ℓ = m ∏ 2 f m − f ℓ F − f ℓ I 16 = ℓ = − 2 ℓ = m ∏ 2 2 κ ( m − ℓ ) F − 2 κ ℓ I 16
for all m = − 2 , … , 2 m=-2,\ldots,2 m = − 2 , … , 2 . Therefore, every projector Π m \Pi_m Π m is a polynomial in F F F of degree at most four, and consequently A Z = span C { I 16 , F , F 2 , F 3 , F 4 } \mathcal{A}_Z=\operatorname{span}_{\mathbb{C}}\{I_{16},F,F^2,F^3,F^4\} A Z = span C { I 16 , F , F 2 , F 3 , F 4 } . Equivalently, the output-signal observable F F F separates spectral projections of A Z \mathcal{A}_Z A Z , and alg ∗ ( I 16 , F ) = A Z \operatorname{alg}^{*}(I_{16},F)=\mathcal{A}_Z alg ∗ ( I 16 , F ) = A Z , which is (116) . Theorem 5 is verified.
The above distinction is essential: linear observability concerns a linear model composed of (100) and (105) , whereas algebraic observability uses the finite joint spectrum of the commuting observables. The Homodyne signal distinguishes the five magnetization labels.
Quantum BAE measurements
For the single field channel, let
h = e 1 , g = e 2 , g ⊥ = − e 1 , S = [ g g ⊥ ] = [ 0 − 1 1 0 ] . h=e_1,\quad g=e_2,\quad g_\perp=-e_1,\quad S=[\,g\ g_\perp\,]=\begin{bmatrix}0&-1\\1&0\end{bmatrix}. h = e 1 , g = e 2 , g ⊥ = − e 1 , S = [ g g ⊥ ] = [ 0 1 − 1 0 ] .
Then W g = W 2 W_g=W_2 W g = W 2 , Y h = Y 1 Y_h=Y_1 Y h = Y 1 , B g = B 2 B_g=B_2 B g = B 2 , C h = C 1 C_h=C_1 C h = C 1 . Moreover, h T g = 0 h^{\mathsf T}g=0 h T g = 0 excludes direct feedthrough from W 2 W_2 W 2 to Y 1 Y_1 Y 1 . Clearly, (103), and both h T J h = 0 h^{\mathsf T}Jh=0 h T J h = 0 and h T h = 1 h^{\mathsf T}h=1 h T h = 1 also hold.
It remains to verify the system-mediated decoupling condition. By (143) , we get C h A = 0 C_hA=0 C h A = 0 , C h B 1 = 0 C_hB_1=0 C h B 1 = 0 , C h B 2 = 0 C_hB_2=0 C h B 2 = 0 ; that is, C h C_h C h annihilates every matrix in { A , B 1 , B 2 } \{A,B_1,B_2\} { A , B 1 , B 2 } from the left. Let M j ∈ { A , B 1 , B 2 } M_j\in\{A,B_1,B_2\} M j ∈ { A , B 1 , B 2 } . It follows that for every v ∈ R 255 v\in\mathbb{R}^{255} v ∈ R 255 ,
C h M ℓ ⋯ M 1 B g v = 0 , ℓ = 0 , 1 , … . C_hM_\ell\cdots M_1B_gv=0,\qquad\ell=0,1,\ldots. C h M ℓ ⋯ M 1 B g v = 0 , ℓ = 0 , 1 , … .
Thus (107) holds. All conditions of Theorem 6 are therefore satisfied. Actually, as already derived in (139) , the output quadrature
d Y h = 2 κ J z d t + d W 1 , (180) \mathrm{d}Y_h=2\sqrt{\kappa}J_z\,\mathrm{d}t+\mathrm{d}W_1, \tag*{(180)} d Y h = 2 κ J z d t + d W 1 , (180)
is back-action evading with respect to the input quadrature W g = W 2 W_g=W_2 W g = W 2 . Theorem 6 is verified. Moreover, by (142) , J z ( t ) ≡ J z ( 0 ) J_z(t)\equiv J_z(0) J z ( t ) ≡ J z ( 0 ) , the output equation (180) takes the particularly simple form Y 1 ( t ) = 2 κ J z ( 0 ) t + W 1 ( t ) Y_1(t)=2\sqrt{\kappa}J_z(0)t+W_1(t) Y 1 ( t ) = 2 κ J z ( 0 ) t + W 1 ( t ) . Thus the input field quadrature W 2 W_2 W 2 has neither a direct nor a system-mediated effect on the output quadrature Y 1 Y_1 Y 1 . We verified Theorem 6.
Coexistence of the three structures
The three structures exhibited above are governed by two commutation properties of the model: the rotation invariance (136) of the exchange Hamiltonian H c h H_{\mathrm{ch}} H ch , and the fact that the coupling operator L L L is a function of the conserved magnetization J z J_z J z . The first property makes the total-spin-zero subspace invariant under H c h H_{\mathrm{ch}} H ch and, together with J z P 0 = 0 J_zP_0=0 J z P 0 = 0 , yields the autonomous logical qubit in Subsection VII-B, which is the ( r , m ) = ( 2 , 1 ) (r,m)=(2,1) ( r , m ) = ( 2 , 1 ) DFS. These two properties together yield both the QND magnetization algebra of Subsection VII-C, which is conserved and, although not linearly observable, is generated by the single homodyne signal F F F , and the quantum BAE measurement of Subsection VII-D.
Conclusion
We have investigated the structure of quantum finite-level systems in the Heisenberg picture, by focusing on their decoherence-free subsystems (DFSs), quantum nondemolition (QND) variables and quantum back-action evading (BAE) measurements. One line of future research is to apply these analytical results to system design of quantum finite-level systems. The Tavis–Cummings model is studied [7 ] , where for a special case investigated in Section III-C, a real orthogonal coordinate transformation T T T decomposes the whole system into the controllable/observable subsystem and a uncontrollable/unobservable subsystem (namely, DFS). Interestingly, the columns of T T T corresponding to the DFS are exactly the dark states which are decoherence-free subspace in the Schrödinger picture. However, to the best of our knowledge, the connection between decoherence-free subsystems in the Heisenberg picture and decoherence-free subspaces in the Schrödinger picture does not exist for general quantum systems. Thus, another line of future research is to connect these results in the Heisenberg picture to relevant results in the Schrödinger picture, in the hope of obtaining a unified theory.