Introduction
As quantum error-correction (QEC) experiments continue to improve, it becomes increasingly important to develop methods to predict and assess logical performance over many QEC cycles [1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 ] . One approach is to predict logical performance via measurements or models of the individual, low-level physical components involved in the QEC cycle [12 , 13 , 14 , 15 , 16 ] . Alternatively, logical performance can be characterized directly at the logical level, without consideration of the low-level components [17 , 18 , 19 , 20 , 21 ] . However, both approaches are complicated by the fact that the effective logical dynamics under repeated QEC cycles can be non-Markovian (i.e., retain memory) even when the low-level error model is Markovian [2 ] . Importantly, Ref. [1 ] has established that a stabilizer QEC cycle subject to Pauli-stochastic noise can be modeled by a classical Markov process over syndrome states, with a logical Pauli channel attached to each syndrome transition. As discussed in Ref. [1 ] , studying the family of logical Pauli channels of a QEC cycle provides a direct means of characterizing the logical performance of repeated applications of that QEC cycle. However, these results do not hold in systems in which qubits experience leakage.
Leakage substantially complicates both the prediction and the assessment of logical performance in stabilizer QEC. A Pauli error acts within the computational qubit space, whereas leakage transfers population into noncomputational levels and can leave information there from one QEC cycle to the next. As a result, modeling a general leakage process could require consideration of the full quantum dynamics, rather than the much simpler stochastic description available for Pauli errors. Furthermore, in general the full quantum state, including the leaked degrees of freedom, can influence subsequent gates and thereby affect the errors that occur in later cycles. Because these effects can substantially degrade logical performance, accurately accounting for them is essential. Consequently, modeling repeated QEC in the presence of leakage errors can be considerably more complicated than modeling QEC with Pauli errors. The key simplification in this work comes from Pauli twirling. We develop a version of Pauli twirling for leakage-extended qubits such that twirled error processes can be modeled by Pauli error channels on the computational subspace, conditioned on transitions between leakage patterns. This is formalized in Theorem 1 . In this work we apply it to error processes of QEC cycles, but we hope it will be of independent interest in other contexts as well.
Our second main result is obtained by applying Theorem 1 to the mathematical model developed in § 0.2.2 for a noisy stabilizer-code QEC cycle in the presence of leakage. We show that the complete-cycle channel decomposes into subnormalized logical Pauli maps indexed by joint transitions of the syndrome and the data-leakage pattern. This is formalized in Theorem 2 . One interpretation of this result is that the syndrome and the data-leakage pattern jointly function as a hidden-memory state carried from one QEC cycle to the next. Thus, a single QEC cycle can be viewed as a classical Markov transition rule on the hidden-memory state space, together with a logical Pauli channel conditioned on each transition. In the setting of Ref. [1 ] , where the error model consists of Pauli noise and no leakage, the analogous hidden-memory state is the syndrome alone. Consequently, Theorem 2 can be interpreted as a generalization of the formalism of Ref. [1 ] to account for leakage.
Our third main result comes from applying Theorem 2 to QEC memory experiments. We define a QEC memory experiment to consist of state preparation, k k k repeated QEC cycles, and a final measurement for which one outcome is designated “success.” For each logical Pauli component, we express its contribution to the success probability in terms of a transfer matrix, a state-preparation vector, and a measurement vector. Each transfer matrix is determined by the dynamics of a single QEC cycle and the dependence on k k k enters only through the k k k th power of that matrix. The state-preparation vector depends only on the input state, while the measurement vector depends only on the measurement process and its distinguished success outcome; neither depends on the details of the QEC cycle. This is formalized in Theorem 3 . Altogether, this formulation of memory experiment success probabilities provides a concrete connection between a single-cycle error model and predictions of many-cycle QEC performance. Furthermore, if the action of the transfer matrices can be well approximated by their dominant left and right eigenspaces, this representation could also lead to approximate Markovian models of the kind developed in Ref. [1 ] . Although we do not address this direction in detail here, it is an important matter for future work.
Leakage has long been recognized as an important challenge for fault-tolerant QEC. There has been substantial effort to develop methods for analyzing, suppressing, or removing leakage [22 , 23 , 24 , 25 , 26 , 27 , 28 ] ; detecting leakage and incorporating leakage information into error correction [29 , 30 , 31 ] ; and obtaining tractable representations or simulations of multilevel leakage noise [32 , 33 , 34 ] . Our contribution is complementary to these efforts: for any leakage-handling scheme described by our QEC-cycle model, our framework provides a general way to understand how its dynamics contribute to logical errors.
Pauli twirling has long been used to map general qubit noise to Pauli noise, both for simpler noise characterization and as a tractable approximation in simulations of fault-tolerant circuits and QEC [35 , 36 , 37 ] . For a noisy implementation of a Clifford operation, a random input Pauli can be propagated through the target Clifford to determine a compensating output Pauli, thereby preserving the target operation while Pauli twirling its residual noise. This construction—which we refer to as Pauli twirling with respect to a Clifford—underlies Pauli-frame randomization, randomized compiling, cycle benchmarking, and other noise-characterization protocols [38 , 39 , 40 ] . We note that prior work has generalized some of these ideas to scenarios involving leakage. Randomized-benchmarking protocols have been developed that account for leakage, with some constructions using random relative phases between the computational and leakage subspaces to suppress coherence between them [41 , 42 , 43 ] . The subspace-twirling approximation has been used to obtain a tractable incoherent model of leakage in QEC [32 ] , while, under gate-independent, time-independent Markovian-noise assumptions, independent Pauli twirling has been shown to depolarize each leakage-pattern sector, leading to a classical Markov process on the leakage patterns [44 ] .
This work is organized as follows. Section 2 gives a high-level summary of the results and discusses their implications, limitations, and directions for future work. The appendices provide a detailed, largely self-contained mathematical presentation of the results. Appendix A introduces standard Pauli twirling and the structure of Pauli maps in the absence of leakage. Appendix B develops the theory of Pauli twirling with respect to a Clifford. Appendix C introduces leakage-extended qubits, leakage-extended qubit registers, and other mathematical concepts related to leakage, and then proves Theorem 1 . Appendix D provides the mathematical formulation of a QEC cycle on leakage-extended qubit registers. Appendix E formulates the model of a QEC cycle in terms of logical Pauli maps conditioned on transitions between leakage patterns and syndrome states, and proves Theorem 2 . Appendix F develops the joint syndrome-leakage hidden-memory representation and the transfer-matrix description of repeated QEC experiments, and proves Theorem 3 . Finally, Appendix G contains lemmas related to operator spaces.
Summary of results and discussion
Leakage-extended qubits and leakage-dephased Pauli twirling
Our first main result is a structure theorem for leakage-dephased Pauli twirling. This is formalized in Theorem 1 , and is described in detail in the rest of this subsection.
Throughout this paper, a leakage-extended qubit space is a Hilbert space H = C ⊕ L \mathcal{H} = \mathcal{C} \oplus\mathcal{L} H = C ⊕ L , where C \mathcal{C} C is a two-dimensional computational qubit subspace and L \mathcal{L} L is a one-dimensional leakage subspace (see Definition C.1). Theorem 1 is formulated in terms of a leakage-extended qubit register Q \mathsf{Q} Q , which consists of a set Q Q Q of system (qubit) labels and a leakage-extended qubit space H i \mathcal{H}_i H i for each label i ∈ Q i \in Q i ∈ Q (see Definition C.3). The register’s total Hilbert space is H Q = ⨂ i ∈ Q H i \mathcal{H}_Q = \bigotimes_{i \in Q} \mathcal{H}_i H Q = ⨂ i ∈ Q H i . A leakage pattern S ⊆ Q S \subseteq Q S ⊆ Q specifies which labels correspond to leaked qubits (see Definition C.5) and determines a leakage subspace H Q S \mathcal{H}_Q^S H Q S (see Definition C.7). These subspaces give the decomposition H Q = ⨁ S ⊆ Q H Q S \mathcal{H}_Q = \bigoplus_{S \subseteq Q} \mathcal{H}_Q^S H Q = ⨁ S ⊆ Q H Q S (C.14).
Denote by K Q \mathsf{K}_Q K Q the computational qubit register associated with Q Q Q , obtained by considering only the computational subspace C i \mathcal{C}_i C i for each leakage-extended qubit in Q Q Q (see Definition C.23). The Hilbert space of K Q \mathsf{K}_Q K Q is denoted K Q \mathcal{K}_Q K Q , and it embeds naturally as a subspace of H Q \mathcal{H}_Q H Q . Pauli operators on K Q \mathcal{K}_Q K Q are indexed by the Pauli-label space V K Q V_{\mathsf{K}_Q} V K Q , with binary X X X and Z Z Z labels for each qubit (see Definition A.6 ).
For v ∈ V K Q v \in V_{\mathcal K_Q} v ∈ V K Q , denote by W K Q ( v ) W^{\mathcal K_Q}(v) W K Q ( v ) the corresponding Pauli operator (see Definition A.12), acting on K Q \mathcal K_Q K Q . Denote by W Q ( v ) W^Q(v) W Q ( v ) the corresponding leakage-extended Pauli operator, which acts as the Pauli labeled by v v v on the computational subspace of H Q \mathcal H_Q H Q and as the identity on the leakage subspaces (see Definition C.24). We denote by W N ( v ) = Ad W N ( v ) \mathcal W^N(v) = \operatorname{Ad}_{W^N(v)} W N ( v ) = Ad W N ( v ) the map associated with conjugation by the Pauli W N ( v ) W^N(v) W N ( v ) . For a leakage-extended register Q Q Q , the Pauli map W K Q ( v ) : End ( K Q ) → End ( K Q ) W^{\mathcal K_Q}(v) : \operatorname{End}(\mathcal K_Q) \to\operatorname{End}(\mathcal K_Q) W K Q ( v ) : End ( K Q ) → End ( K Q ) is a standard Pauli map on the computational subspace, while W Q ( v ) : End ( H Q ) → End ( H Q ) W^Q(v) : \operatorname{End}(\mathcal H_Q) \to\operatorname{End}(\mathcal H_Q) W Q ( v ) : End ( H Q ) → End ( H Q ) is a leakage-extended Pauli map.
Consider a Clifford unitary C C C on the computational subspace K Q \mathcal K_Q K Q . Denote by C \mathcal C C the associated superoperator, C ( Y ) = C Y C † \mathcal C(Y) = CYC^\dagger C ( Y ) = C Y C † for any operator Y ∈ End ( K Q ) Y \in\operatorname{End}(\mathcal K_Q) Y ∈ End ( K Q ) . Let Φ K : End ( K Q ) → End ( K Q ) \Phi_{\mathcal K} : \operatorname{End}(\mathcal K_Q) \to\operatorname{End}(\mathcal K_Q) Φ K : End ( K Q ) → End ( K Q ) be a superoperator, whose ideal (noise-free) realization is C \mathcal C C . The map Φ K \Phi_{\mathcal K} Φ K can be twirled into a more convenient form via the Pauli twirl with respect to C C C , denoted T C K Q \mathcal T_C^{\mathcal K_Q} T C K Q (see Definition B.3).
Here we develop an analogous theory of Pauli twirling on leakage-extended qubit registers. Denote by Δ \Delta Δ the leakage-pattern dephasing channel, which removes coherence between distinct leakage subspaces (see Definition C.15). Fix a superoperator Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal H_Q) \to\operatorname{End}(\mathcal H_Q) Φ : End ( H Q ) → End ( H Q ) , a Clifford unitary C C C defined on K Q \mathcal K_Q K Q that is intended to be the nominal action of Φ \Phi Φ in the computational subspace, and an associated symplectic automorphism F C : V K Q → V K Q F_C : V_{\mathcal K_Q} \to V_{\mathcal K_Q} F C : V K Q → V K Q (see Definition B.1). We define the leakage-dephased Pauli twirl (see Definition C.26) of Φ \Phi Φ according to
T C Q ( Φ ) = Δ ∘ [ 1 ∣ V K Q ∣ ∑ v ∈ V K Q W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ] ∘ Δ . (1) \mathcal T_C^Q(\Phi) = \Delta\circ\left[ \frac{1}{\lvert V_{\mathcal K_Q}\rvert} \sum_{v \in V_{\mathcal K_Q}} W^Q(F_Cv) \circ\Phi\circ W^Q(v) \right] \circ\Delta. \tag*{(1)} T C Q ( Φ ) = Δ ∘ ∣ V K Q ∣ 1 v ∈ V K Q ∑ W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ∘ Δ. (1)
Theorem 1 provides a convenient decomposition of Eq. (1) . To state Theorem 1, we introduce the following maps for each leakage pattern R ⊆ Q R \subseteq Q R ⊆ Q . The leakage-subspace projection map P R \mathcal P_R P R restricts an operator on H Q \mathcal H_Q H Q to H Q R \mathcal H_Q^R H Q R , and the leakage-subspace extension map E R \mathcal E_R E R embeds an operator on H Q R \mathcal H_Q^R H Q R as an operator on H Q \mathcal H_Q H Q (see Definitions C.48 and C.50). The filling map F R \mathcal F_R F R replaces the leaked systems with maximally mixed computational qubits (i.e., for i ∈ R i \in R i ∈ R , system i i i is placed in the state 1 / 2 ∈ End ( C i ) 1/2 \in\operatorname{End}(\mathcal C_i) 1/2 ∈ End ( C i ) ), and the leakage map L R \mathcal L_R L R places the systems labeled i ∈ R i \in R i ∈ R in their leakage states (see Definitions C.28 and C.30). For S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , define the map D T , S Φ : End ( K Q ) → End ( K Q ) \mathcal D^\Phi_{T,S} : \operatorname{End}(\mathcal K_Q) \to\operatorname{End}(\mathcal K_Q) D T , S Φ : End ( K Q ) → End ( K Q ) (see Definition C.67) via
D T , S Φ = T C K Q ( F T ∘ P T ∘ Φ ∘ E S ∘ L S ) ∘ C − 1 . (2) \mathcal D^\Phi_{T,S} = \mathcal T_C^{\mathcal K_Q} \left(\mathcal F_T \circ\mathcal P_T \circ\Phi\circ\mathcal E_S \circ\mathcal L_S\right) \circ\mathcal C^{-1}. \tag*{(2)} D T , S Φ = T C K Q ( F T ∘ P T ∘ Φ ∘ E S ∘ L S ) ∘ C − 1 . (2)
Theorem 1 establishes that the leakage-dephased Pauli twirl can be written
T C Q ( Φ ) = ∑ S , T ⊆ Q E T ∘ L T ∘ D T , S Φ ∘ C ∘ F S ∘ P S . (3) \mathcal T_C^Q(\Phi) = \sum_{S,T \subseteq Q} \mathcal E_T \circ\mathcal L_T \circ\mathcal D^\Phi_{T,S} \circ\mathcal C \circ\mathcal F_S \circ\mathcal P_S. \tag*{(3)} T C Q ( Φ ) = S , T ⊆ Q ∑ E T ∘ L T ∘ D T , S Φ ∘ C ∘ F S ∘ P S . (3)
This expression admits the following interpretation. First, the sum decomposes the twirled map into branches indexed by the incoming and outgoing leakage patterns, S S S and T T T , respectively. For a given S → T S \to T S → T branch, the projection map P S \mathcal P_S P S selects the incoming leakage subspace S S S , and then F S \mathcal F_S F S replaces all leaked systems labeled i ∈ S i \in S i ∈ S with maximally mixed computational qubits. Then, the ideal Clifford C C C acts on the computational subspace, followed by the Pauli map D T , S Φ \mathcal D^\Phi_{T,S} D T , S Φ . Finally, L T \mathcal L_T L T produces the outgoing leakage pattern and E T \mathcal E_T E T embeds the result into H Q \mathcal H_Q H Q .
When Φ \Phi Φ is completely positive and trace preserving (CPTP), these Pauli maps D T , S Φ \mathcal D^\Phi_{T,S} D T , S Φ are subnormalized, and the sum of their weights (see Definition A.47) over the outgoing leakage patterns T T T is unity for each incoming pattern S S S (see Proposition C.70). Thus, Eq. (3) describes the leakage-pattern transitions induced by the twirled channel T C ( Φ ) T_{\mathop{\mathrm{C}}}(\Phi) T C ( Φ ) in terms of a classical Markov process, where the weight p ( T ∣ S ) = wt ( D T , S Φ ) p(T|S) = \operatorname{wt}(D^\Phi_{T,S}) p ( T ∣ S ) = wt ( D T , S Φ ) gives the probability of the transition S ↦ T S \mapsto T S ↦ T . For each nontrivial transition, the corresponding action on the computational subspace consists of the ideal Clifford C C C followed by the Pauli channel obtained by normalizing D T , S Φ D^\Phi_{T,S} D T , S Φ (see Definition C.78).
Model of a QEC cycle
Here we introduce the mathematical model of a QEC cycle that is used in the remainder of this paper. Diagram 2.4 gives its circuit representation. This abstract formulation is not the usual way of presenting a QEC cycle, but it provides a convenient framework for our analysis while accommodating a very broad class of QEC-cycle constructions. The results in this paper apply to any QEC cycle that admits this abstract description.
Figure 2.4.
Consider a fixed stabilizer code on n n n physical qubits that encodes k k k logical qubits. We model a QEC cycle for this code using the registers D, A, and O depicted in Diagram 2.4, which are all leakage-extended qubit registers (see Definition C.3). The data register D contains the physical qubits of the code and persists from one QEC cycle to the next. The ancilla register A contains auxiliary qubits used for fault-tolerant syndrome extraction and leakage detection, and is reset between QEC cycles. The outcome register O represents the classical outcomes produced by those processes, and is also reset between QEC cycles. Although the information contained in O is classical, we model O as a leakage-extended qubit register to model all errors associated with the QEC cycle, including leakage, as a channel on the joint D, A, and O system. The subsequent measurement of O is taken to be perfect. We denote by T the joint leakage-extended qubit register comprising D, A, and O (see Definition D.2).
The mathematical model of a QEC cycle consists of a channel Φ : End ( H D ) → End ( H D ) \Phi: \operatorname{End}(\mathcal{H}_{D}) \to\operatorname{End}(\mathcal{H}_{D}) Φ : End ( H D ) → End ( H D ) acting on D alone. The channel Φ \Phi Φ is derived from the extended model shown in Diagram 2.4, read from left to right. The incoming state of D is ρ i n \rho_{\mathrm{in}} ρ in . We use an initialization map I A O \mathcal{I}_{AO} I A O to formalize the introduction of the registers A and O, prepared in their all-zero states (D.7). The first step of the QEC cycle is fault-tolerant syndrome extraction and leakage detection. In the absence of noise or leakage, this is captured by the Clifford unitary C s y n d C_{\mathrm{synd}} C synd , which records syndrome information into the outcome register, possibly using the ancilla qubits in A. This construction appears in Ref. [1 ] . In the presence of noise and leakage, we generalize the Clifford channel C s y n d C_{\mathrm{synd}} C synd to the channel N \mathcal{N} N , which jointly acts on the leakage-extended registers D, A, and O; in the noise- and leakage-free limits, N \mathcal{N} N reproduces C s y n d C_{\mathrm{synd}} C synd on the computational subspace.
Each system in O is then measured perfectly, resulting in one of three outcome labels 0, 1, and L (where L labels the leaked state). The full outcome record is denoted θ ∈ Θ \theta\in\Theta θ ∈ Θ , where Θ : = { 0 , 1 , L } Q \Theta:= \{0,1,L\}^{Q} Θ := { 0 , 1 , L } Q . The measurement and the discard of A and O are formalized by outcome maps { M θ } θ ∈ Θ \{\mathcal{M}_{\theta}\}_{\theta\in\Theta} { M θ } θ ∈ Θ (D.14), with θ \theta θ retained only as a classical outcome label. We use a generalized notion of the QEC cycle’s decoder, which uses the classical outcome record θ \theta θ to determine both the logical correction given by the syndrome record and the subset of data qubits to return to the computational subspace via classical side processing. The Pauli-correction function D : Θ → V K D D:\Theta\to\mathcal{V}_{K_D} D : Θ → V K D (D.16) determines the logical correction, captured by the Pauli channel W D ( D ( θ ) ) \mathcal{W}^{D}(D(\theta)) W D ( D ( θ )) on D (see Definition C.24). The return function K : Θ → 2 D K:\Theta\to2^{D} K : Θ → 2 D (D.17) determines the subset of the systems in the data register that should be returned to the computational subspace as maximally mixed states. This is captured by the leakage-extended operation R K ( θ ) \mathcal{R}_{K(\theta)} R K ( θ ) on D, where K ( θ ) K(\theta) K ( θ ) is the subset of qubits to be returned to the computational subspace.
Thus, the mathematical model of the complete-cycle channel is given by
Φ = ∑ θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ N ∘ I A O , (4) \Phi= \sum_{\theta\in\Theta} \mathcal{R}_{K(\theta)} \circ\mathcal{W}^{D}(D(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{N} \circ\mathcal{I}_{AO}, \tag*{(4)} Φ = θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ N ∘ I A O , (4)
which acts on D alone and is derived in detail in Appendix D.
To define logical error processes, consider an abstract k k k -qubit logical register L and an abstract ( n − k ) (n-k) ( n − k ) -qubit syndrome register S (see Definition A.2). Let C F C_{\mathrm{F}} C F be a Clifford isomorphism (see Definition D.4) that provides the logical–syndrome decomposition of the computational subspace of the data register [1 ] :
C F : K D → K L ⊗ K S . (5) C_{\mathrm{F}}:\mathcal{K}_{D}\to\mathcal{K}_{L}\otimes\mathcal{K}_{S}. \tag*{(5)} C F : K D → K L ⊗ K S . (5)
In the Clifford frame, the nominal Clifford unitary C s y n d C_{\mathrm{synd}} C synd that implements fault-tolerant syndrome extraction and leakage detection in the computational subspace is replaced by the Clifford C f r ∈ End ( K L ⊗ K S ⊗ K A ⊗ K O ) C_{\mathrm{fr}}\in\operatorname{End}(\mathcal{K}_{L}\otimes\mathcal{K}_{S}\otimes\mathcal{K}_{A}\otimes\mathcal{K}_{O}) C fr ∈ End ( K L ⊗ K S ⊗ K A ⊗ K O ) . For every s ∈ Z 2 S s\in\mathbb{Z}_{2}^{S} s ∈ Z 2 S and ∣ ψ ⟩ L ∈ K L |\psi\rangle_{L}\in\mathcal{K}_{L} ∣ ψ ⟩ L ∈ K L , it satisfies
C f r ( ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ 0 ⟩ A ⊗ ∣ 0 ⟩ O ) = ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ α ( s ) ⟩ A ⊗ ∣ E ( s ) ⟩ O , (6) C_{\mathrm{fr}}\left(|\psi\rangle_{L}\otimes|s\rangle_{S}\otimes|0\rangle_{A}\otimes|0\rangle_{O}\right)=|\psi\rangle_{L}\otimes|s\rangle_{S}\otimes|\alpha(s)\rangle_{A}\otimes|E(s)\rangle_{O}, \tag*{(6)} C fr ( ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣0 ⟩ A ⊗ ∣0 ⟩ O ) = ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ α ( s ) ⟩ A ⊗ ∣ E ( s ) ⟩ O , (6)
where E : Z 2 S → Z 2 O E:\mathbb{Z}_{2}^{S}\to\mathbb{Z}_{2}^{O} E : Z 2 S → Z 2 O is an injective encoding of the syndrome record (D.8) that is determined by the ideal Clifford C s y n d C_{\mathrm{synd}} C synd in the original frame (D.10), while ∣ α ( s ) ⟩ A |\alpha(s)\rangle_{A} ∣ α ( s ) ⟩ A is a state of A that may depend on s s s (D.9); its detailed form is unimportant. The interpretation of C s y n d C_{\mathrm{synd}} C synd is that it gives the ideal (noiseless) action of N \mathcal{N} N in the joint computational subspace of the registers D, A, and O. The channel N \mathcal{N} N is otherwise arbitrary, so C s y n d C_{\mathrm{synd}} C synd matters only as the reference Clifford for leakage-dephased Pauli twirling.
The channel N \mathcal{N} N can be simplified by leakage-dephased Pauli twirling, as in Theorem 1. Its twirled form is denoted
N t w = T C s y n d T ( N ) (7) \mathcal{N}^{\mathrm{tw}}=\mathcal{T}^{\mathsf{T}}_{C_{\mathrm{synd}}}(\mathcal{N}) \tag*{(7)} N tw = T C synd T ( N ) (7)
and replacing N \mathcal{N} N by N t w \mathcal{N}^{\mathrm{tw}} N tw in Eq. (2.5) gives the twirled complete-cycle channel Φ t w \Phi^{\mathrm{tw}} Φ tw defined in Eq. (D.20). In the remainder of this paper, we assume that the complete-cycle channel Φ \Phi Φ has the structure exhibited by Φ t w \Phi^{\mathrm{tw}} Φ tw in Eq. (D.24), as formalized in Eq. (E.1). This structure can arise from physically implementing leakage-dephased Pauli twirling, as in Theorem 1, from physical guarantees or approximations, or from applying leakage-dephased Pauli twirling separately to the constituent processes of the QEC cycle. The subsequent results in this paper depend only on the resulting structure in Eq. (E.1), not on how it is obtained.
Logical Pauli maps of a QEC-cycle
Our second main result is that the complete QEC-cycle channel Φ \Phi Φ decomposes into logical Pauli maps conditioned on joint transitions of the syndrome and data-leakage pattern. This is formalized in detail in Theorem 2 and described at a high level in this subsection.
First, for s i n , s o u t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S s in , s out ∈ Z 2 S , define the syndrome transition map S s o u t , s i n : End ( K S ) → End ( K S ) \mathcal{S}_{s_{\mathrm{out},s_{\mathrm{in}}}}:\operatorname{End}(\mathcal{K}_S)\to\operatorname{End}(\mathcal{K}_S) S s out , s in : End ( K S ) → End ( K S ) by S s o u t , s i n ( Y ) : = ⟨ s i n ∣ Y ∣ s i n ⟩ S ∣ s o u t ⟩ ⟨ s o u t ∣ S \mathcal{S}_{s_{\mathrm{out},s_{\mathrm{in}}}(Y):=\langle s_{\mathrm{in}}|Y|s_{\mathrm{in}}\rangle_S|s_{\mathrm{out}}\rangle\langle s_{\mathrm{out}}|_S} S s out , s in ( Y ) := ⟨ s in ∣ Y ∣ s in ⟩ S ∣ s out ⟩ ⟨ s out ∣ S for Y ∈ End ( K S ) Y\in\operatorname{End}(\mathcal{K}_S) Y ∈ End ( K S ) (E.67). Theorem 2 states that there is a family of subnormalized Pauli maps J s o u t , s i n : End ( K L ) → End ( K L ) R o u t , R i n \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}:\operatorname{End}(\mathcal{K}_L)\to\operatorname{End}(\mathcal{K}_L)} J s out , s in : End ( K L ) → End ( K L ) R out , R in , indexed by R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D} R in , R out ⊆ D and s i n , s o u t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S s in , s out ∈ Z 2 S , such that
Φ = ∑ R i n , R o u t ⊆ D s i n , s o u t ∈ Z 2 S E R o u t D ∘ L R o u t D ∘ C F − 1 ∘ J s o u t , s i n ⊗ S s o u t , s i n ∘ C F ∘ F R i n D ∘ P R i n D . R o u t , R i n (8) \Phi=\sum_{\substack{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}\\s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S}}\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{C}_{\mathrm{F}}^{-1}\circ\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}\otimes\mathcal{S}_{s_{\mathrm{out},s_{\mathrm{in}}}\circ\mathcal{C}_{\mathrm{F}}\circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}}\circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}.}} \tag*{(8)} Φ = R in , R out ⊆ D s in , s out ∈ Z 2 S ∑ E R out D ∘ L R out D ∘ C F − 1 ∘ J s out , s in ⊗ S s out , s in ∘ C F ∘ F R in D ∘ P R in D . R out , R in (8)
where C F ( X ) = C F X C F † \mathcal{C}_{\mathrm{F}}(X)=C_{\mathrm{F}}XC_{\mathrm{F}}^\dagger C F ( X ) = C F X C F † moves any operator X ∈ End ( K D ) X\in\operatorname{End}(\mathcal{K}_{\mathrm{D}}) X ∈ End ( K D ) between the computational data register and the logical–syndrome frame. A superscript D, e.g., indicates the specialization of the corresponding leakage-sector map to the data register D. Reading Eq. (8) from right to left, the projection map P R i n D \mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}} P R in D selects the incoming data-leakage subspace, the filling map F R i n D \mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} F R in D replaces the leaked systems labeled i ∈ R i n i\in R_{\mathrm{in}} i ∈ R in with maximally mixed computational qubits, and C F \mathcal{C}_{\mathrm{F}} C F expresses the resulting operator in the logical–syndrome frame. Next, each term J ⊗ S \mathcal{J}\otimes\mathcal{S} J ⊗ S with fixed s i n , s o u t s_{\mathrm{in}},s_{\mathrm{out}} s in , s out applies the syndrome transition s i n → s o u t s_{\mathrm{in}}\to s_{\mathrm{out}} s in → s out (via S \mathcal{S} S ), together with the logical Pauli map J s o u t , s i n R o u t , R i n \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}} J s out , s in R out , R in . Finally, C F − 1 \mathcal{C}_{\mathrm{F}}^{-1} C F − 1 maps operators back into the computational subspace K D \mathcal{K}_{\mathrm{D}} K D , the leakage map L R o u t D \mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}} L R out D leaks into the subspace associated with the outgoing data-leakage pattern, and the extension map E R o u t D \mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}} E R out D embeds that leakage subspace into the full space of the data register.
Here, the maps J s o u t , s i n R o u t , R i n \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}} J s out , s in R out , R in depend jointly on the incoming and outgoing syndrome and data-leakage pattern. When the logical register is considered in isolation, the syndrome and data-leakage pattern together constitute a hidden memory that persists between cycles and determines which logical Pauli map acts. This dependence is the source of the emergent non-Markovianity of the isolated logical dynamics discussed in Ref. [2 ] . To formalize the interpretation of the syndrome and data-leakage pattern as the hidden memory of a QEC cycle, we define the hidden-memory state space M : = 2 D × Z 2 S \mathrm{M}:=2^{\mathrm{D}}\times\mathbb{Z}_2^S M := 2 D × Z 2 S (see Definition F.20). An element m = ( R , s ) ∈ M m=(R,s)\in\mathrm{M} m = ( R , s ) ∈ M specifies a data-leakage pattern R ⊆ D R\subseteq\mathrm{D} R ⊆ D and a syndrome s ∈ Z 2 S s\in\mathbb{Z}_2^S s ∈ Z 2 S . The model of a QEC cycle can then be expressed in terms of logical-Pauli transfer matrices, whose rows and columns are indexed by outgoing and incoming hidden-memory states in M \mathrm{M} M , respectively. These transfer matrices are used to analyze memory experiments, as described in detail in the next subsection.
Application to memory experiments
In this paper, we define a memory experiment to consist of an input state ρ \rho ρ on H D \mathcal{H}_{\mathrm{D}} H D , followed by k ≥ 1 k\geq1 k ≥ 1 QEC cycles, followed by a POVM measurement with a distinguished POVM element E s u c c E_{\mathrm{succ}} E succ that indicates the outcome “success.” Typically, success is understood to mean successful recovery of the logical information encoded in ρ \rho ρ . Modeling each cycle by the complete-cycle channel Φ \Phi Φ in Eq. (2.5), the success probability is
p s u c c ( k ) : = tr [ E s u c c Φ k ( ρ ) ] . (9) p_{\mathrm{succ}}(k):=\operatorname{tr}\left[E_{\mathrm{succ}}\Phi^k(\rho)\right]. \tag*{(9)} p succ ( k ) := tr [ E succ Φ k ( ρ ) ] . (9)
To analyze p s u c c ( k ) p_{\mathrm{succ}}(k) p succ ( k ) , for each logical Pauli label ℓ ∈ V L \ell\in V_{\mathrm{L}} ℓ ∈ V L we first define a logical-Pauli transfer matrix T ℓ ∈ R M × M T_{\ell} \in\mathbb{R}^{M \times M} T ℓ ∈ R M × M (see Definition F.45 ). Each T ℓ T_{\ell} T ℓ is determined by the P ℓ P_{\ell} P ℓ eigenvalues of the logical Pauli maps \mathcal{J}^{R_{\mathrm{out},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}} . The family of transfer matrices { T ℓ } ℓ ∈ V L \{T_{\ell}\}_{\ell\in V_{\mathrm{L}}} { T ℓ } ℓ ∈ V L contains the information needed to calculate Φ k ( ρ ) \Phi^{k}(\rho) Φ k ( ρ ) for any input state ρ \rho ρ and k ≥ 1 k \ge1 k ≥ 1 , as formalized in Corollary F.55 . Then, for each input state ρ \rho ρ we define a state-preparation vector r ℓ ( ρ ) ∈ C M r_{\ell}(\rho) \in\mathbb{C}^{M} r ℓ ( ρ ) ∈ C M (see Definition F.50 ). Finally, for each POVM element E E E , we define a measurement vector e ℓ ( E ) ∈ C M e_{\ell}(E) \in\mathbb{C}^{M} e ℓ ( E ) ∈ C M (see Definition F.58 ). Thus the components of both vectors are indexed by M M M .
Our third main result expresses p s u c c ( k ) p_{\mathrm{succ}}(k) p succ ( k ) in terms of r ℓ ( ρ ) r_{\ell}(\rho) r ℓ ( ρ ) , e ℓ ( E s u c c ) e_{\ell}(E_{\mathrm{succ}}) e ℓ ( E succ ) , and T ℓ T_{\ell} T ℓ for ℓ ∈ V L \ell\in V_{\mathrm{L}} ℓ ∈ V L . This is formalized in Theorem 3 , which states that
p s u c c ( k ) = d L ∑ ℓ ∈ V L e ℓ ( E s u c c ) T ℓ k r ℓ ( ρ ) . (10) p_{\mathrm{succ}}(k) = d_{\mathrm{L}} \sum_{\ell\in V_{\mathrm{L}}} e_{\ell}(E_{\mathrm{succ}}) T_{\ell}^{k} r_{\ell}(\rho). \tag*{(10)} p succ ( k ) = d L ℓ ∈ V L ∑ e ℓ ( E succ ) T ℓ k r ℓ ( ρ ) . (10)
Here d L = dim ( K L ) d_{\mathrm{L}} = \dim(\mathcal{K}_{\mathrm{L}}) d L = dim ( K L ) , and ℓ ∈ V L \ell\in V_{\mathrm{L}} ℓ ∈ V L labels the logical Pauli P ℓ P_{\ell} P ℓ . Matrix multiplication sums over intermediate hidden-memory states, so T ℓ k T_{\ell}^{k} T ℓ k propagates the P ℓ P_{\ell} P ℓ coefficient through k k k cycles. Consequently, d L e ℓ ( E s u c c ) T ℓ k r ℓ ( ρ ) d_{\mathrm{L}}e_{\ell}(E_{\mathrm{succ}})T_{\ell}^{k}r_{\ell}(\rho) d L e ℓ ( E succ ) T ℓ k r ℓ ( ρ ) is the contribution of P ℓ P_{\ell} P ℓ to the success probability, and the sum over ℓ \ell ℓ in Eq. (10) reconstructs the full success probability.
In favorable settings, the individual summands in Eq. (10) may be isolated either using physically motivated assumptions about state preparation and measurement or by adapting the character-weighting techniques of Ref. [45 ] ; if so, the summands may be analyzed separately.
Discussion
We envision several additional applications of these results. First, Theorem 1 is deliberately formulated as an abstract structural result for maps on leakage-extended qubit registers. Because it does not depend on the details of a QEC cycle, we expect it to be useful more broadly in the study of leakage and other multilevel quantum processes. Second, Theorem 2 resolves the logical noise of a QEC cycle into contributions conditioned on the incoming and outgoing syndrome and data-leakage pattern. For any QEC scheme and decoder described by our model, this decomposition makes explicit how each transition of these variables contributes to the resulting logical Pauli map. We hope that this perspective will inform the analysis and design of leakage-aware decoders. Third, Theorem 3 extends the program of Ref. [1 ] , which developed approximate logical-error models for QEC memory experiments. The present results show that the data-leakage pattern plays a role closely analogous to that of the syndrome: both are persistent hidden-memory variables that mediate the effective logical dynamics. In Ref. [1 ] , physically motivated assumptions on the distribution of the syndrome made it possible to establish the existence of accurate approximate Markovian models in a broad class of settings. An important direction for future work is to determine whether analogous assumptions can be justified for the joint syndrome-leakage process and, if so, whether they lead to similarly accurate reduced-memory models.
Several extensions of the QEC-cycle model are particularly important. Although the final Pauli correction and return operation may depend on the complete outcome record of a cycle, our model does not permit an outcome obtained partway through the cycle to determine subsequent operations within that same cycle. Such adaptive decisions arise, for example, in some constructions of flagged fault tolerance. It appears plausible that our results can be generalized to these settings by representing the different adaptive branches as distinct QEC-cycle maps, or by composing several such maps into a larger effective cycle, but we leave a formal treatment to future work. We also do not consider windowed decoding, in which decoding decisions depend on outcome records collected over several QEC cycles. Extending the hidden-memory framework to windowed decoders, perhaps by augmenting the memory state with the decoder’s internal state or by treating an entire decoding window as a composite cycle, is an especially important direction. Finally, we assume that both the decoder-selected Pauli correction and the return map are implemented perfectly. Allowing these operations to be noisy is unlikely to change the essential structure of the results, provided that their errors admit an appropriate leakage-resolved Pauli description, but establishing this extension rigorously is also left for future work.
Context of this work and LLM usage
This project began in July 2026 as an extension of our previous work in Ref. [1 ] . Our original goal was to generalize our previous results to noisy QEC with qubits that also experience leakage. It was clear that in general, allowing arbitrary error processes on leaked levels would make it quite difficult to model QEC and prove useful structural results. However, based on our prior research experience on randomized benchmarking, Pauli twirling, and other randomized methods, it seemed quite plausible that Pauli twirling, once appropriately adapted for leakage, could provide useful simplifications. We made an initial investigation into this question using the LLMs that were available to us at the time, predominantly GPT 5 models. Based on that investigation, we decided it was worthwhile to start a full research project, eventually culminating in this paper.
Throughout the project, we continuously made decisions about the scope, assumptions, organization, and technical strategy of the work. We made these decisions using extensive LLM assistance to suggest ideas, try out different ways to organize results and concepts, fill in details, and find related work. We found that the direct outputs of LLMs were almost always unsuitable for inclusion in the final paper, largely because the LLMs by default would use confusing and ambiguous terminology for specific mathematical objects, and would routinely use vague language and heuristic ideas instead of rigorous proofs. As a result, we found that the best way to efficiently make progress was for us to choose how to break our results down into tractable pieces, essentially by specifying every definition, lemma, and theorem. We chose the organization in such a way the individual lemmas were simple enough that we felt like the proofs output by LLMs were trustworthy, and usable after some (sometimes substantial) editorial effort and manual checking. As such, we have done our best to make sure that the proofs are easily readable and easily checkable.
As part of this project, we also used LLMs to construct formal proofs of our results in LEAN 4. We did not manually check any of the formal proofs, and consequently we do not view them as a true part of this publication. We instead view them as yet another part of our due diligence to guard against errors and inaccurate statements. The proofs are included with the arxiv submission and we encourage any interested parties to verify them.
Acknowledgments
The authors thank Aliza Siddiqui, Shawn Geller, Scott Glancy, and Emanuel Knill for helpful comments and discussions. AJF acknowledges support from the Professional Research Experience Program (PREP) operated jointly by NIST and the University of Colorado. This work includes contributions of the National Institute of Standards and Technology, which are not subject to U.S. copyright.
References
Standard Pauli twirling
The Pauli-twirling construction and its basic properties reviewed in this section are standard in the community [35 , 46 ] . The use of Pauli randomization to simplify noise in quantum circuits is discussed in Refs. [38 , 47 ] . We include the full derivations in order to establish notation and provide a self-contained treatment.
Qubit registers and operator spaces
Definition A.1 (Qubit Hilbert space). A qubit Hilbert space is a Hilbert space that is isomorphic to C 2 \mathbb{C}^{2} C 2 and is equipped with a fixed orthonormal basis, called its computational basis .
Definition A.2 (Qubit register). A qubit register N \mathbf{N} N consists of a finite label set (which we also denote N \mathbf{N} N ) and, for each i ∈ N i \in\mathbf{N} i ∈ N , a qubit Hilbert space q i \mathcal{q}_{i} q i whose computational basis is denoted by { ∣ 0 i ⟩ , ∣ 1 i ⟩ } \{\lvert0_{i}\rangle,\lvert1_{i}\rangle\} {∣ 0 i ⟩ , ∣ 1 i ⟩} . Define the Hilbert space K N \mathcal{K}_{\mathbf{N}} K N of N \mathbf{N} N by
K N = ⨂ i ∈ N q i . (11) \mathcal{K}_{\mathbf{N}} = \bigotimes_{i \in\mathbf{N}} \mathcal{q}_{i}. \tag*{(11)} K N = i ∈ N ⨂ q i . (11) The subscript of K \mathcal{K} K specifies the register’s label set.
Fix a qubit register and associated label set N \mathbf{N} N . Define A \mathcal{A} A to be End ( K N ) \operatorname{End}(\mathcal{K}_{\mathbf{N}}) End ( K N ) equipped with the Hilbert–Schmidt inner product
⟨ A , B ⟩ A = tr ( A † B ) , (12) \langle A,B\rangle_{\mathcal{A}} = \operatorname{tr}(A^{\dagger}B), \tag*{(12)} ⟨ A , B ⟩ A = tr ( A † B ) , (12)
for all A , B ∈ A A,B \in\mathcal{A} A , B ∈ A . Define S \mathcal{S} S to be End ( A ) \operatorname{End}(\mathcal{A}) End ( A ) equipped with the Hilbert–Schmidt inner product
⟨ Φ , Ψ ⟩ S = tr A ( Φ † ∘ Ψ ) , (13) \langle\Phi,\Psi\rangle_{\mathcal{S}} = \operatorname{tr}_{\mathcal{A}}(\Phi^{\dagger} \circ\Psi), \tag*{(13)} ⟨ Φ , Ψ ⟩ S = tr A ( Φ † ∘ Ψ ) , (13)
for all Φ , Ψ ∈ S \Phi,\Psi\in\mathcal{S} Φ , Ψ ∈ S , where Φ † \Phi^{\dagger} Φ † denotes the adjoint with respect to the Hilbert–Schmidt inner product on A \mathcal{A} A .
Pauli-label space and Pauli operators
Define Z 2 = { 0 , 1 } \mathbb{Z}_{2} = \{0,1\} Z 2 = { 0 , 1 } , with arithmetic modulo 2.
Definition A.6 (Pauli-label space). Define Z 2 N \mathbb{Z}_{2}^{\mathbf{N}} Z 2 N to be the Z 2 \mathbb{Z}_{2} Z 2 -valued vector space of functions x : N → Z 2 x:\mathbf{N} \to\mathbb{Z}_{2} x : N → Z 2 , with pointwise operations. Define the Pauli-label space V N \mathcal{V}_{\mathbf{N}} V N of N \mathbf{N} N by
V N = Z 2 N ⊕ Z 2 N . (14) \mathcal{V}_{\mathbf{N}} = \mathbb{Z}_{2}^{\mathbf{N}} \oplus\mathbb{Z}_{2}^{\mathbf{N}}. \tag*{(14)} V N = Z 2 N ⊕ Z 2 N . (14)
For x ∈ Z 2 N x \in\mathbb{Z}_{2}^{\mathbf{N}} x ∈ Z 2 N and i ∈ N i \in\mathbf{N} i ∈ N , write x i = x ( i ) x_{i}=x(i) x i = x ( i ) . For u , v ∈ Z 2 N u,v \in\mathbb{Z}_{2}^{\mathbf{N}} u , v ∈ Z 2 N , define u ⋅ v ∈ Z 2 u \cdot v \in\mathbb{Z}_{2} u ⋅ v ∈ Z 2 by
u ⋅ v = ∑ i ∈ N u i v i , (15) u \cdot v = \sum_{i \in\mathbf{N}} u_{i}v_{i}, \tag*{(15)} u ⋅ v = i ∈ N ∑ u i v i , (15)
where the elements of Z 2 \mathbb{Z}_{2} Z 2 are represented by the integers 0 and 1 in the sum. The space V N \mathcal{V}_{\mathbf{N}} V N is a finite abelian group under componentwise addition. For a = ( x , z ) a=(x,z) a = ( x , z ) and b = ( x ′ , z ′ ) b=(x',z') b = ( x ′ , z ′ ) in V N \mathcal{V}_{\mathbf{N}} V N , one has a + b = ( x + x ′ , z + z ′ ) a+b=(x+x',z+z') a + b = ( x + x ′ , z + z ′ ) . Define the standard symplectic form [ ⋅ , ⋅ ] : V N × V N → Z 2 [\cdot,\cdot]:\mathcal{V}_{\mathbf{N}}\times\mathcal{V}_{\mathbf{N}}\to\mathbb{Z}_{2} [ ⋅ , ⋅ ] : V N × V N → Z 2 by
[ a , b ] = x ⋅ z ′ + z ⋅ x ′ ( m o d 2 ) . (16) [a,b] = x \cdot z' + z \cdot x' \pmod2. \tag*{(16)} [ a , b ] = x ⋅ z ′ + z ⋅ x ′ ( mod 2 ) . (16)
The form is nondegenerate: for every a ∈ V N ∖ { 0 } a \in V_N \setminus\{0\} a ∈ V N ∖ { 0 } , there exists v ∈ V N v \in V_N v ∈ V N such that [ v , a ] = 1 [v,a] = 1 [ v , a ] = 1 . This follows because for a = ( x , z ) ≠ 0 a = (x,z) \ne0 a = ( x , z ) = 0 , one can choose j ∈ N j \in\mathbb{N} j ∈ N such that either z j = 1 z_j = 1 z j = 1 or x j = 1 x_j = 1 x j = 1 . Define e j ∈ Z 2 N e_j \in\mathbb{Z}_2^N e j ∈ Z 2 N by ( e j ) i = δ i , j (e_j)_i = \delta_{i,j} ( e j ) i = δ i , j . If z j = 1 z_j = 1 z j = 1 , then [ ( e j , 0 ) , a ] = 1 [(e_j,0),a] = 1 [( e j , 0 ) , a ] = 1 ; if x j = 1 x_j = 1 x j = 1 , then [ ( 0 , e j ) , a ] = 1 [(0,e_j),a] = 1 [( 0 , e j ) , a ] = 1 . For each i ∈ N i \in\mathbb{N} i ∈ N , define X i , Z i ∈ End ( q i ) X_i,Z_i \in\operatorname{End}(q_i) X i , Z i ∈ End ( q i ) by
X i : = ∣ 0 i ⟩ ⟨ 1 i ∣ + ∣ 1 i ⟩ ⟨ 0 i ∣ , Z i : = ∣ 0 i ⟩ ⟨ 0 i ∣ − ∣ 1 i ⟩ ⟨ 1 i ∣ . (17) X_i := |0_i\rangle\langle1_i| + |1_i\rangle\langle0_i|,\qquad Z_i := |0_i\rangle\langle0_i| - |1_i\rangle\langle1_i|. \tag*{(17)} X i := ∣ 0 i ⟩ ⟨ 1 i ∣ + ∣ 1 i ⟩ ⟨ 0 i ∣ , Z i := ∣ 0 i ⟩ ⟨ 0 i ∣ − ∣ 1 i ⟩ ⟨ 1 i ∣. (17)
For i ∈ N i \in\mathbb{N} i ∈ N and a i = ( x i , z i ) ∈ Z 2 2 a_i = (x_i,z_i) \in\mathbb{Z}_2^2 a i = ( x i , z i ) ∈ Z 2 2 , define the single-qubit Pauli w i ( a i ) ∈ End ( q i ) w_i(a_i) \in\operatorname{End}(q_i) w i ( a i ) ∈ End ( q i ) by
w i ( a i ) = i x i z i X i x i Z i z i . (18) w_i(a_i) = i^{x_i z_i} X_i^{x_i} Z_i^{z_i}. \tag*{(18)} w i ( a i ) = i x i z i X i x i Z i z i . (18)
For a = ( x , z ) ∈ V N a = (x,z) \in V_N a = ( x , z ) ∈ V N , write a i = ( x i , z i ) a_i = (x_i,z_i) a i = ( x i , z i ) for every i ∈ N i \in\mathbb{N} i ∈ N .
Definition A.12. A Pauli operator on N \mathsf{N} N associated with the Pauli label a a a is given by
W N ( a ) = i x ⋅ z ⨂ i ∈ N X i x i Z i z i = ⨂ i ∈ N w i ( a i ) , (19) W^N(a) = i^{x\cdot z}\bigotimes_{i\in\mathbb{N}} X_i^{x_i} Z_i^{z_i} = \bigotimes_{i\in\mathbb{N}} w_i(a_i), \tag*{(19)} W N ( a ) = i x ⋅ z i ∈ N ⨂ X i x i Z i z i = i ∈ N ⨂ w i ( a i ) , (19) where N \mathsf{N} N is a computational (qubit) register.
When it is unambiguous, we suppress the superscript N \mathsf{N} N and write W ( a ) = W N ( a ) W(a) = W^N(a) W ( a ) = W N ( a ) .
Lemma A.14. For all a = ( x , z ) a = (x,z) a = ( x , z ) and b = ( x ′ , z ′ ) b = (x',z') b = ( x ′ , z ′ ) in V N V_N V N ,
W ( a ) W ( b ) = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) W ( a + b ) . (20) W(a)W(b) = (-1)^{z\cdot x'} i^{x\cdot z+x'\cdot z'-(x+x')\cdot(z+z')}W(a+b). \tag*{(20)} W ( a ) W ( b ) = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) W ( a + b ) . (20) Proof. For each i ∈ N i \in\mathbb{N} i ∈ N , write a i = ( x i , z i ) a_i = (x_i,z_i) a i = ( x i , z i ) and b i = ( x i ′ , z i ′ ) b_i = (x'_i,z'_i) b i = ( x i ′ , z i ′ ) . The identities X i 2 = Z i 2 = 1 q i X_i^2 = Z_i^2 = \mathbb{1}_{q_i} X i 2 = Z i 2 = 1 q i and Z i X i = − X i Z i Z_iX_i = -X_iZ_i Z i X i = − X i Z i give
w i ( a i ) w i ( b i ) = i x i z i + x i ′ z i ′ ( − 1 ) z i x i ′ X i x i + x i ′ Z i z i + z i ′ (21) w_i(a_i)w_i(b_i) = i^{x_i z_i+x'_i z'_i}(-1)^{z_i x'_i}X_i^{x_i+x'_i}Z_i^{z_i+z'_i} \tag*{(21)} w i ( a i ) w i ( b i ) = i x i z i + x i ′ z i ′ ( − 1 ) z i x i ′ X i x i + x i ′ Z i z i + z i ′ (21) = ( − 1 ) z i x i ′ i x i z i + x i ′ z i ′ − ( x i + x i ′ ) ( z i + z i ′ ) w i ( a i + b i ) . (22) = (-1)^{z_i x'_i}i^{x_i z_i+x'_i z'_i-(x_i+x'_i)(z_i+z'_i)}w_i(a_i+b_i). \tag*{(22)} = ( − 1 ) z i x i ′ i x i z i + x i ′ z i ′ − ( x i + x i ′ ) ( z i + z i ′ ) w i ( a i + b i ) . (22) Taking tensor products over i ∈ N i \in\mathbb{N} i ∈ N and using Eq. (A.13) gives
W ( a ) W ( b ) = ⨂ i ∈ N w i ( a i ) w i ( b i ) (23) W(a)W(b) = \bigotimes_{i\in\mathbb{N}} w_i(a_i)w_i(b_i) \tag*{(23)} W ( a ) W ( b ) = i ∈ N ⨂ w i ( a i ) w i ( b i ) (23) = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) ⨂ i ∈ N w i ( a i + b i ) (24) = (-1)^{z\cdot x'}i^{x\cdot z+x'\cdot z'-(x+x')\cdot(z+z')}\bigotimes_{i\in\mathbb{N}} w_i(a_i+b_i) \tag*{(24)} = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) i ∈ N ⨂ w i ( a i + b i ) (24) = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) W ( a + b ) . (25) = (-1)^{z\cdot x'}i^{x\cdot z+x'\cdot z'-(x+x')\cdot(z+z')}W(a+b). \tag*{(25)} = ( − 1 ) z ⋅ x ′ i x ⋅ z + x ′ ⋅ z ′ − ( x + x ′ ) ⋅ ( z + z ′ ) W ( a + b ) . (25) Then, Eq. (A.20) proves Eq. (A.15).
Define N = ∣ N ∣ N = |\mathsf{N}| N = ∣ N ∣ to be the number of qubits in N \mathsf{N} N and define d = dim ( K N ) = 2 N d = \dim(\mathcal{K}_N) = 2^N d = dim ( K N ) = 2 N , so that ∣ V N ∣ = 4 N = d 2 = dim A |V_N| = 4^N = d^2 = \dim\mathcal{A} ∣ V N ∣ = 4 N = d 2 = dim A .
Lemma A.21. For all a , b ∈ V N a,b \in V_N a , b ∈ V N , the Pauli operators satisfy
W ( a ) † = W ( a ) , (26) W(a)^\dagger= W(a), \tag*{(26)} W ( a ) † = W ( a ) , (26) W ( a ) 2 = 1 K N , (27) W(a)^2 = \mathbb{1}_{\mathcal{K}_N}, \tag*{(27)} W ( a ) 2 = 1 K N , (27) W ( b ) † W ( a ) W ( b ) = ( − 1 ) [ b , a ] W ( a ) , (28) W(b)^\dagger W(a)W(b) = (-1)^{[b,a]}W(a), \tag*{(28)} W ( b ) † W ( a ) W ( b ) = ( − 1 ) [ b , a ] W ( a ) , (28) tr ( W ( a ) W ( b ) ) = d δ a , b . (29) \operatorname{tr}(W(a)W(b)) = d\delta_{a,b}. \tag*{(29)} tr ( W ( a ) W ( b )) = d δ a , b . (29)
Consequently, each W ( a ) W(a) W ( a ) is unitary and Hermitian, and the family { d − 1 / 2 W ( a ) } a ∈ V N \{d^{-1/2}W(a)\}_{a\in V_N} { d − 1/2 W ( a ) } a ∈ V N is an orthonormal basis for A \mathcal{A} A .
Proof. Eqs. (A.10), (A.11), and (A.13) give Eq. (A.22) and tr ( W ( c ) ) = d δ c , 0 \operatorname{tr}(W(c))=d\delta_{c,0} tr ( W ( c )) = d δ c , 0 for every c ∈ V N c\in V_N c ∈ V N . Setting b = a b=a b = a in Eq. (A.15) gives Eq. (A.23), since ( − 1 ) z ⋅ x + 2 x ⋅ z = 1 (-1)^{z\cdot x+2x\cdot z}=1 ( − 1 ) z ⋅ x + 2 x ⋅ z = 1 .
The exponent of i i i in Eq. (A.15) is unchanged when a a a and b b b are interchanged. Comparing the two instances of Eq. (A.15) and using Eq. (A.9) gives W ( b ) W ( a ) = ( − 1 ) [ b , a ] W ( a ) W ( b ) W(b)W(a)=(-1)^{[b,a]}W(a)W(b) W ( b ) W ( a ) = ( − 1 ) [ b , a ] W ( a ) W ( b ) . Combining this equality with Eqs. (A.22) and (A.23) gives Eq. (A.24).
If a = b a=b a = b , Eq. (A.23) gives tr ( W ( a ) W ( b ) ) = d \operatorname{tr}(W(a)W(b))=d tr ( W ( a ) W ( b )) = d . If a ≠ b a\ne b a = b , then a + b ≠ 0 a+b\ne0 a + b = 0 , and Eq. (A.15) together with tr ( W ( a + b ) ) = 0 \operatorname{tr}(W(a+b))=0 tr ( W ( a + b )) = 0 gives tr ( W ( a ) W ( b ) ) = 0 \operatorname{tr}(W(a)W(b))=0 tr ( W ( a ) W ( b )) = 0 . Hence, Eq. (A.25) holds.
Eqs. (A.22) and (A.23) show that each W ( a ) W(a) W ( a ) is unitary. Eqs. (A.22) and (A.25) give
⟨ W ( a ) , W ( b ) ⟩ A = d δ a , b , \langle W(a),W(b)\rangle_{\mathcal{A}}=d\delta_{a,b}, ⟨ W ( a ) , W ( b ) ⟩ A = d δ a , b ,
for all a , b ∈ V N a,b\in V_N a , b ∈ V N . Since ∣ V N ∣ = dim A |V_N|=\dim\mathcal{A} ∣ V N ∣ = dim A , { d − 1 / 2 W ( a ) } a ∈ V N \{d^{-1/2}W(a)\}_{a\in V_N} { d − 1/2 W ( a ) } a ∈ V N is an orthonormal basis for A \mathcal{A} A . □
Accordingly, we simply write W ( a ) W(a) W ( a ) in lieu of W ( a ) † W(a)^\dagger W ( a ) † or W ( a ) − 1 W(a)^{-1} W ( a ) − 1 going forward.
Pauli representations on operator spaces
For a finite-dimensional Hilbert space X \mathcal{X} X and U ∈ End ( X ) U\in\operatorname{End}(\mathcal{X}) U ∈ End ( X ) , define Ad U : End ( X ) → End ( X ) \operatorname{Ad}_U:\operatorname{End}(\mathcal{X})\to\operatorname{End}(\mathcal{X}) Ad U : End ( X ) → End ( X ) by Ad U ( A ) = U A U † \operatorname{Ad}_U(A)=UAU^\dagger Ad U ( A ) = U A U † . Next, define the map W N : V N → S \mathcal{W}^N:V_N\to\mathcal{S} W N : V N → S by
W N ( v ) ( ⋅ ) = Ad W N ( v ) ( ⋅ ) = W N ( v ) ( ⋅ ) W N ( v ) , (30) \mathcal{W}^N(v)(\cdot)=\operatorname{Ad}_{W^N(v)}(\cdot)=W^N(v)(\cdot)W^N(v), \tag*{(30)} W N ( v ) ( ⋅ ) = Ad W N ( v ) ( ⋅ ) = W N ( v ) ( ⋅ ) W N ( v ) , (30)
for all v ∈ V N v\in V_N v ∈ V N , where the dot denotes an arbitrary operator in A \mathcal{A} A .
When the register is unambiguous, we suppress its superscript and write W \mathcal{W} W for W N \mathcal{W}^N W N .
Lemma A.27. The map W \mathcal{W} W is a unitary representation of the additive group V N V_N V N on A \mathcal{A} A .
Proof. Eq. (A.15) shows that W ( u ) W ( v ) W(u)W(v) W ( u ) W ( v ) is a unit-modulus scalar multiple of W ( u + v ) W(u+v) W ( u + v ) . Therefore, Eqs. (A.26), (G.3), and (G.5) give
W ( u ) ∘ W ( v ) = Ad W ( u ) W ( v ) = Ad W ( u + v ) = W ( u + v ) , \mathcal{W}(u)\circ\mathcal{W}(v)=\operatorname{Ad}_{W(u)W(v)}=\operatorname{Ad}_{W(u+v)}=\mathcal{W}(u+v), W ( u ) ∘ W ( v ) = Ad W ( u ) W ( v ) = Ad W ( u + v ) = W ( u + v ) ,
for all u , v ∈ V N u,v\in V_N u , v ∈ V N . Also, W ( 0 ) = 1 K N \mathcal{W}(0)=\mathbb{1}_{\mathcal{K}_N} W ( 0 ) = 1 K N , so W ( 0 ) = id A \mathcal{W}(0)=\operatorname{id}_{\mathcal{A}} W ( 0 ) = id A . Thus, W \mathcal{W} W is a representation. Furthermore, Eqs. (A.22) and (A.23) show that each W ( v ) W(v) W ( v ) is unitary, while Eq. (A.26) and Lemma G.4 show that each W ( v ) \mathcal{W}(v) W ( v ) is unitary on A \mathcal{A} A . □
Define the map Γ : V N → End ( S ) \Gamma:V_N\to\operatorname{End}(\mathcal{S}) Γ : V N → End ( S ) by
Γ ( v ) ( Φ ) = W ( v ) ∘ Φ ∘ W ( v ) , (31) \Gamma(v)(\Phi)=\mathcal{W}(v)\circ\Phi\circ\mathcal{W}(v), \tag*{(31)} Γ ( v ) ( Φ ) = W ( v ) ∘ Φ ∘ W ( v ) , (31)
for all v ∈ V N v \in V_{\mathrm{N}} v ∈ V N , Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S .
Lemma A.29. The map Γ \Gamma Γ is a unitary representation of the additive group V N V_{\mathrm{N}} V N on S \mathcal{S} S .
Proof. By Lemma A.27, W ( v ) † \mathcal{W}(v)^{\dagger} W ( v ) † is unitary on A \mathcal{A} A for every v ∈ V N v \in V_{\mathrm{N}} v ∈ V N . Therefore, Eqs. (G.3) and (A.28), together with Lemma G.4, show that Γ ( v ) \Gamma(v) Γ ( v ) is unitary on S \mathcal{S} S for every v ∈ V N v \in V_{\mathrm{N}} v ∈ V N . Lemma A.27 and the commutativity of V N V_{\mathrm{N}} V N give W ( v ) ∘ W ( u ) = W ( v + u ) = W ( u + v ) \mathcal{W}(v) \circ\mathcal{W}(u) = \mathcal{W}(v+u) = \mathcal{W}(u+v) W ( v ) ∘ W ( u ) = W ( v + u ) = W ( u + v ) . Therefore, for u , v ∈ V N u,v \in V_{\mathrm{N}} u , v ∈ V N and Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S , Eq. (A.28) gives
( Γ ( u ) ∘ Γ ( v ) ) ( Φ ) = W ( u ) ∘ W ( v ) ∘ Φ ∘ W ( v ) ∘ W ( u ) = W ( u + v ) ∘ Φ ∘ W ( u + v ) = Γ ( u + v ) ( Φ ) . \begin{aligned}
(\Gamma(u) \circ\Gamma(v))(\Phi) &= \mathcal{W}(u) \circ\mathcal{W}(v) \circ\Phi\circ\mathcal{W}(v) \circ\mathcal{W}(u) \\
&= \mathcal{W}(u+v) \circ\Phi\circ\mathcal{W}(u+v) \\
&= \Gamma(u+v)(\Phi).
\end{aligned} ( Γ ( u ) ∘ Γ ( v )) ( Φ ) = W ( u ) ∘ W ( v ) ∘ Φ ∘ W ( v ) ∘ W ( u ) = W ( u + v ) ∘ Φ ∘ W ( u + v ) = Γ ( u + v ) ( Φ ) . Finally, Eq. (A.28) and W ( 0 ) = id A \mathcal{W}(0)=\operatorname{id}_{\mathcal{A}} W ( 0 ) = id A give Γ ( 0 ) = id S \Gamma(0)=\operatorname{id}_{\mathcal{S}} Γ ( 0 ) = id S .
Standard Pauli twirl and Pauli maps
Lemma A.30. Let X \mathcal{X} X be a finite-dimensional complex Hilbert space, let G G G be a finite group, and let ρ : G → U ( X ) \rho:G \to\mathrm{U}(\mathcal{X}) ρ : G → U ( X ) be a unitary representation. Define the operator P G ∈ End ( X ) P_G \in\operatorname{End}(\mathcal{X}) P G ∈ End ( X ) by
P G = 1 ∣ G ∣ ∑ g ∈ G ρ ( g ) , (32) P_G = \frac{1}{\lvert G\rvert}\sum_{g\in G}\rho(g), \tag*{(32)} P G = ∣ G ∣ 1 g ∈ G ∑ ρ ( g ) , (32) and define the invariant subspace X G ⊆ X \mathcal{X}^{G} \subseteq\mathcal{X} X G ⊆ X by
X G = { x ∈ X : ρ ( g ) x = x for every g ∈ G } . (33) \mathcal{X}^{G} = \{x \in\mathcal{X}: \rho(g)x=x\ \text{for every }g\in G\}. \tag*{(33)} X G = { x ∈ X : ρ ( g ) x = x for every g ∈ G } . (33) Then P G P_G P G is the orthogonal projection onto X G \mathcal{X}^{G} X G .
Proof. Left multiplication permutes the elements of G G G , so Eq. (A.31) gives ρ ( g ) P G = P G \rho(g)P_G=P_G ρ ( g ) P G = P G for every g ∈ G g\in G g ∈ G . Thus, ran ( P G ) ⊆ X G \operatorname{ran}(P_G)\subseteq\mathcal{X}^{G} ran ( P G ) ⊆ X G . Conversely, Eq. (A.31) gives P G x = x P_Gx=x P G x = x for every x ∈ X G x\in\mathcal{X}^{G} x ∈ X G . Finally, unitarity of ρ \rho ρ and the fact that inversion permutes the elements of G G G give
P G † = 1 ∣ G ∣ ∑ g ∈ G ρ ( g − 1 ) = P G . P_G^{\dagger}=\frac{1}{\lvert G\rvert}\sum_{g\in G}\rho(g^{-1})=P_G. P G † = ∣ G ∣ 1 g ∈ G ∑ ρ ( g − 1 ) = P G . Hence, P G P_G P G is the orthogonal projection onto X G \mathcal{X}^{G} X G .
Definition A.33 (Standard Pauli twirl). Define the standard Pauli twirl T N : S → S \mathcal{T}^{\mathrm{N}}:\mathcal{S}\to\mathcal{S} T N : S → S by:
T N ( Φ ) = 1 ∣ V N ∣ ∑ v ∈ V N Γ ( v ) ( Φ ) , (34) \mathcal{T}^{\mathrm{N}}(\Phi)=\frac{1}{\lvert V_{\mathrm{N}}\rvert}\sum_{v\in V_{\mathrm{N}}}\Gamma(v)(\Phi), \tag*{(34)} T N ( Φ ) = ∣ V N ∣ 1 v ∈ V N ∑ Γ ( v ) ( Φ ) , (34)
for all Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S .
When the register is unambiguous, we suppress its superscript and write T \mathcal{T} T for T N \mathcal{T}^{\mathrm{N}} T N . Define the invariant subspace S V N ⊆ S \mathcal{S}^{V_{\mathrm{N}}}\subseteq\mathcal{S} S V N ⊆ S of Γ \Gamma Γ by
S V N = { Φ ∈ S : Γ ( v ) ( Φ ) = Φ for every v ∈ V N } . (35) \mathcal{S}^{V_{\mathrm{N}}}=\{\Phi\in\mathcal{S}:\Gamma(v)(\Phi)=\Phi\ \text{for every }v\in V_{\mathrm{N}}\}. \tag*{(35)} S V N = { Φ ∈ S : Γ ( v ) ( Φ ) = Φ for every v ∈ V N } . (35)
Lemma A.36. The standard Pauli twirl T \mathcal{T} T is the orthogonal projection onto S V N \mathcal{S}^{V_N} S V N . Moreover, S V N \mathcal{S}^{V_N} S V N can be expressed as
S V N = span C { W ( e ) : e ∈ V N } , (36) \mathcal{S}^{V_N} = \operatorname{span}_{\mathbb{C}}\{\mathcal{W}(e): e \in V_N\}, \tag*{(36)} S V N = span C { W ( e ) : e ∈ V N } , (36) and the family { W ( e ) } e ∈ V N \{\mathcal{W}(e)\}_{e\in V_N} { W ( e ) } e ∈ V N is an orthogonal basis for S V N \mathcal{S}^{V_N} S V N satisfying
⟨ W ( e ) , W ( f ) ⟩ S = d 2 δ e , f , (37) \langle\mathcal{W}(e),\mathcal{W}(f)\rangle_{\mathcal{S}} = d^2\delta_{e,f}, \tag*{(37)} ⟨ W ( e ) , W ( f ) ⟩ S = d 2 δ e , f , (37) for all e , f ∈ V N e,f \in V_N e , f ∈ V N .
Proof. Apply Lemma A.30 to the unitary representation Γ \Gamma Γ from Lemma A.29, with G = V N G = V_N G = V N , X = S \mathcal{X} = \mathcal{S} X = S , and ρ = Γ \rho= \Gamma ρ = Γ . By Eqs. (A.34) and (A.35), the resulting averaging operator and invariant subspace are T \mathcal{T} T and S V N \mathcal{S}^{V_N} S V N , respectively. Thus, T \mathcal{T} T is the orthogonal projection onto S V N \mathcal{S}^{V_N} S V N , and dim ( S V N ) = tr S ( T ) \dim(\mathcal{S}^{V_N}) = \operatorname{tr}_{\mathcal{S}}(\mathcal{T}) dim ( S V N ) = tr S ( T ) .
For every v ∈ V N v \in V_N v ∈ V N , applying Eqs. (A.28) and (G.10) with X = A \mathcal{X} = \mathcal{A} X = A identifies Γ ( v ) \Gamma(v) Γ ( v ) with L W ( v ) , W ( v ) \mathcal{L}_{\mathcal{W}(v),\mathcal{W}(v)} L W ( v ) , W ( v ) . Therefore, Eq. (G.12) gives
tr S ( Γ ( v ) ) = tr A ( W ( v ) ) tr A ( W ( v ) ) = ∣ tr A ( W ( v ) ) ∣ 2 . \operatorname{tr}_{\mathcal{S}}(\Gamma(v)) = \operatorname{tr}_{\mathcal{A}}(\mathcal{W}(v))\operatorname{tr}_{\mathcal{A}}(\mathcal{W}(v)) = \left|\operatorname{tr}_{\mathcal{A}}(\mathcal{W}(v))\right|^2. tr S ( Γ ( v )) = tr A ( W ( v )) tr A ( W ( v )) = ∣ tr A ( W ( v )) ∣ 2 .
Similarly, Eqs. (A.26) and (G.10), applied with X = K N \mathcal{X} = \mathcal{K}_N X = K N , identify W ( v ) \mathcal{W}(v) W ( v ) with L W ( v ) , W ( v ) \mathcal{L}_{\mathcal{W}(v),\mathcal{W}(v)} L W ( v ) , W ( v ) . Hence, Eq. (G.12) gives
tr A ( W ( v ) ) = tr K N ( W ( v ) ) tr K N ( W ( v ) ) = ∣ tr K N ( W ( v ) ) ∣ 2 . \operatorname{tr}_{\mathcal{A}}(\mathcal{W}(v)) = \operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v))\operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v)) = \left|\operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v))\right|^2. tr A ( W ( v )) = tr K N ( W ( v )) tr K N ( W ( v )) = ∣ tr K N ( W ( v )) ∣ 2 .
Consequently,
tr S ( Γ ( v ) ) = ∣ tr K N ( W ( v ) ) ∣ 4 . \operatorname{tr}_{\mathcal{S}}(\Gamma(v)) = \left|\operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v))\right|^4. tr S ( Γ ( v )) = ∣ tr K N ( W ( v )) ∣ 4 .
Since W ( 0 ) = 1 K N \mathcal{W}(0) = 1_{\mathcal{K}_N} W ( 0 ) = 1 K N , applying Eq. (A.25) with the second label zero gives tr K N ( W ( v ) ) = d δ v , 0 \operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v)) = d\delta_{v,0} tr K N ( W ( v )) = d δ v , 0 for every v ∈ V N v \in V_N v ∈ V N . Taking tr S \operatorname{tr}_{\mathcal{S}} tr S in Eq. (A.34) and using tr S ( Γ ( v ) ) = ∣ tr K N ( W ( v ) ) ∣ 4 \operatorname{tr}_{\mathcal{S}}(\Gamma(v)) = \left|\operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v))\right|^4 tr S ( Γ ( v )) = ∣ tr K N ( W ( v )) ∣ 4 , tr K N ( W ( v ) ) = d δ v , 0 \operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(v)) = d\delta_{v,0} tr K N ( W ( v )) = d δ v , 0 , and ∣ V N ∣ = d 2 |V_N| = d^2 ∣ V N ∣ = d 2 gives
dim ( S V N ) = 1 ∣ V N ∣ ∑ v ∈ V N tr S ( Γ ( v ) ) = d 4 ∣ V N ∣ = ∣ V N ∣ . (38) \dim(\mathcal{S}^{V_N}) = \frac{1}{|V_N|}\sum_{v\in V_N}\operatorname{tr}_{\mathcal{S}}(\Gamma(v)) = \frac{d^4}{|V_N|} = |V_N|. \tag*{(38)} dim ( S V N ) = ∣ V N ∣ 1 v ∈ V N ∑ tr S ( Γ ( v )) = ∣ V N ∣ d 4 = ∣ V N ∣. (38)
For e , v ∈ V N e,v \in V_N e , v ∈ V N , Eqs. (A.26) and (A.28), together with Lemma A.27 and the commutativity of V N V_N V N , give
Γ ( v ) ( W ( e ) ) = W ( v ) ∘ W ( e ) ∘ W ( v ) = W ( v ) ∘ W ( v ) ∘ W ( e ) = W ( e ) . \Gamma(v)(\mathcal{W}(e)) = \mathcal{W}(v)\circ\mathcal{W}(e)\circ\mathcal{W}(v) = \mathcal{W}(v)\circ\mathcal{W}(v)\circ\mathcal{W}(e) = \mathcal{W}(e). Γ ( v ) ( W ( e )) = W ( v ) ∘ W ( e ) ∘ W ( v ) = W ( v ) ∘ W ( v ) ∘ W ( e ) = W ( e ) .
Thus, W ( e ) ∈ S V N \mathcal{W}(e) \in\mathcal{S}^{V_N} W ( e ) ∈ S V N for every e ∈ V N e \in V_N e ∈ V N . Then, Eqs. (G.13), (A.22), and (A.25), applied with X = K N \mathcal{X} = \mathcal{K}_N X = K N , give
⟨ W ( e ) , W ( f ) ⟩ S = ∣ tr K N ( W ( e ) W ( f ) ) ∣ 2 = d 2 δ e , f . \langle\mathcal{W}(e),\mathcal{W}(f)\rangle_{\mathcal{S}} = \left|\operatorname{tr}_{\mathcal{K}_N}(\mathcal{W}(e)\mathcal{W}(f))\right|^2 = d^2\delta_{e,f}. ⟨ W ( e ) , W ( f ) ⟩ S = ∣ tr K N ( W ( e ) W ( f )) ∣ 2 = d 2 δ e , f .
Finally, Eq. (A.39) shows that S V N \mathcal{S}^{V_N} S V N has dimension ∣ V N ∣ |V_N| ∣ V N ∣ . Thus, { W ( e ) } e ∈ V N \{\mathcal{W}(e)\}_{e\in V_N} { W ( e ) } e ∈ V N is an orthogonal basis for S V N \mathcal{S}^{V_N} S V N .
Definition A.40 (Pauli map). A Pauli map on A \mathcal{A} A is a map D ∈ S \mathcal{D} \in\mathcal{S} D ∈ S for which there exists a family { q e } e ∈ V N ⊆ C \{q_e\}_{e \in\mathsf{V}_{\mathrm{N}}} \subseteq\mathbb{C} { q e } e ∈ V N ⊆ C such that
D = ∑ e ∈ V N q e W ( e ) . (39) \mathcal{D} = \sum_{e \in\mathsf{V}_{\mathrm{N}}} q_e \mathcal{W}(e). \tag*{(39)} D = e ∈ V N ∑ q e W ( e ) . (39)
Lemma A.42. Let D \mathcal{D} D be a Pauli map . There is a unique family { q e } e ∈ V N ⊆ C \{q_e\}_{e \in\mathsf{V}_{\mathrm{N}}} \subseteq\mathbb{C} { q e } e ∈ V N ⊆ C satisfying Eq. (A.41). For every e ∈ V N e \in\mathsf{V}_{\mathrm{N}} e ∈ V N ,
q e = 1 d 2 ⟨ W ( e ) , D ⟩ S . (40) q_e = \frac{1}{d^2} \langle\mathcal{W}(e), \mathcal{D} \rangle_{\mathcal{S}}. \tag*{(40)} q e = d 2 1 ⟨ W ( e ) , D ⟩ S . (40)
Proof. Let { r e } e ∈ V N ⊆ C \{r_e\}_{e \in\mathsf{V}_{\mathrm{N}}} \subseteq\mathbb{C} { r e } e ∈ V N ⊆ C satisfy D = ∑ e ∈ V N r e W ( e ) \mathcal{D} = \sum_{e \in\mathsf{V}_{\mathrm{N}}} r_e \mathcal{W}(e) D = ∑ e ∈ V N r e W ( e ) . For every f ∈ V N f \in\mathsf{V}_{\mathrm{N}} f ∈ V N , Eq. (A.38) gives
1 d 2 ⟨ W ( f ) , D ⟩ S = 1 d 2 ∑ e ∈ V N r e ⟨ W ( f ) , W ( e ) ⟩ S = ∑ e ∈ V N r e δ f , e = r f . \frac{1}{d^2} \langle\mathcal{W}(f), \mathcal{D} \rangle_{\mathcal{S}} = \frac{1}{d^2} \sum_{e \in\mathsf{V}_{\mathrm{N}}} r_e \langle\mathcal{W}(f), \mathcal{W}(e) \rangle_{\mathcal{S}} = \sum_{e \in\mathsf{V}_{\mathrm{N}}} r_e \delta_{f,e} = r_f. d 2 1 ⟨ W ( f ) , D ⟩ S = d 2 1 e ∈ V N ∑ r e ⟨ W ( f ) , W ( e ) ⟩ S = e ∈ V N ∑ r e δ f , e = r f .
Thus, every family satisfying Eq. (A.41) satisfies Eq. (A.43), and the family is unique. □
Proposition A.44. Let E ∈ S \mathcal{E} \in\mathcal{S} E ∈ S . For e ∈ V N e \in\mathsf{V}_{\mathrm{N}} e ∈ V N , define q e ∈ C q_e \in\mathbb{C} q e ∈ C by
q e = 1 d 2 ⟨ W ( e ) , E ⟩ S . (41) q_e = \frac{1}{d^2} \langle\mathcal{W}(e), \mathcal{E} \rangle_{\mathcal{S}}. \tag*{(41)} q e = d 2 1 ⟨ W ( e ) , E ⟩ S . (41) Then T ( E ) \mathcal{T}(\mathcal{E}) T ( E ) is a Pauli map satisfying
T ( E ) = ∑ e ∈ V N q e W ( e ) . (42) \mathcal{T}(\mathcal{E}) = \sum_{e \in\mathsf{V}_{\mathrm{N}}} q_e \mathcal{W}(e). \tag*{(42)} T ( E ) = e ∈ V N ∑ q e W ( e ) . (42)
Proof. By Lemma A.36, the family { d − 1 W ( e ) } e ∈ V N \{d^{-1}\mathcal{W}(e)\}_{e \in\mathsf{V}_{\mathrm{N}}} { d − 1 W ( e ) } e ∈ V N is an orthonormal basis for S V N \mathcal{S}^{\mathsf{V}_{\mathrm{N}}} S V N and T \mathcal{T} T is the orthogonal projection onto S V N \mathcal{S}^{\mathsf{V}_{\mathrm{N}}} S V N . Therefore, for every E ∈ S \mathcal{E} \in\mathcal{S} E ∈ S , the orthogonal-projection formula gives
T ( E ) = ∑ e ∈ V N ⟨ d − 1 W ( e ) , E ⟩ S d − 1 W ( e ) = ∑ e ∈ V N 1 d 2 ⟨ W ( e ) , E ⟩ S W ( e ) . \mathcal{T}(\mathcal{E}) = \sum_{e \in\mathsf{V}_{\mathrm{N}}} \langle d^{-1}\mathcal{W}(e), \mathcal{E} \rangle_{\mathcal{S}} d^{-1}\mathcal{W}(e) = \sum_{e \in\mathsf{V}_{\mathrm{N}}} \frac{1}{d^2} \langle\mathcal{W}(e), \mathcal{E} \rangle_{\mathcal{S}} \mathcal{W}(e). T ( E ) = e ∈ V N ∑ ⟨ d − 1 W ( e ) , E ⟩ S d − 1 W ( e ) = e ∈ V N ∑ d 2 1 ⟨ W ( e ) , E ⟩ S W ( e ) .
□
Pauli channels and subnormalized Pauli maps
Definition A.47 (Weight of a Pauli map). For a Pauli map D \mathcal{D} D , define its weight by
wt ( D ) = 1 d tr ( D ( 1 K N ) ) . (43) \operatorname{wt}(\mathcal{D}) = \frac{1}{d} \operatorname{tr}(\mathcal{D}(\mathbb{1}_{\mathcal{K}_{\mathrm{N}}})). \tag*{(43)} wt ( D ) = d 1 tr ( D ( 1 K N )) . (43)
Lemma A.49. Let D \mathcal{D} D be a Pauli map , and let { q e } e ∈ V N ⊆ C \{q_e\}_{e \in\mathsf{V}_{\mathrm{N}}} \subseteq\mathbb{C} { q e } e ∈ V N ⊆ C be its unique coefficient family from Lemma A.42. Then
wt ( D ) = ∑ e ∈ V N q e . (44) \operatorname{wt}(\mathcal{D}) = \sum_{e \in\mathsf{V}_{\mathrm{N}}} q_e. \tag*{(44)} wt ( D ) = e ∈ V N ∑ q e . (44) Consequently, for all A ∈ A A \in\mathcal{A} A ∈ A ,
tr ( D ( A ) ) = wt ( D ) tr ( A ) . (45) \operatorname{tr}(\mathcal{D}(A)) = \operatorname{wt}(\mathcal{D}) \operatorname{tr}(A). \tag*{(45)} tr ( D ( A )) = wt ( D ) tr ( A ) . (45)
Proof. Eqs. (G.3), (A.22), and (A.23) give W ( e ) ( 1 K N ) = 1 K N \mathcal{W}(e)(1_{\mathcal{K}_N})=1_{\mathcal{K}_N} W ( e ) ( 1 K N ) = 1 K N for every e ∈ V N e\in\mathcal{V}_N e ∈ V N . Substituting Eq. (A.41) into Eq. (A.48) therefore gives Eq. (A.50). Finally, unitary invariance of the trace, Eq. (A.41), and Eq. (A.50) give Eq. (A.51).
Definition A.52 (Trace-nonincreasing map). A positive linear map E ∈ S \mathcal{E}\in\mathcal{S} E ∈ S is trace nonincreasing if
tr ( E ( X ) ) ≤ tr ( X ) for every X ⪰ 0. (46) \operatorname{tr}(\mathcal{E}(X))\le\operatorname{tr}(X)\quad\text{for every }X\succeq0. \tag*{(46)} tr ( E ( X )) ≤ tr ( X ) for every X ⪰ 0. (46)
Definition A.54 (Pauli channels and subnormalized Pauli maps). A Pauli map is a Pauli channel if it is completely positive and trace preserving (CPTP). A Pauli map is a subnormalized Pauli map if it is completely positive and trace nonincreasing.
Lemma A.55. Let D \mathcal{D} D be a Pauli map with coefficient family { q e } e ∈ V N \{q_e\}_{e\in\mathcal{V}_N} { q e } e ∈ V N . Then D \mathcal{D} D is a Pauli channel if and only if
q e ∈ [ 0 , 1 ] for all e ∈ V N and ∑ e ∈ V N q e = 1 , (47) q_e\in[0,1]\quad\text{for all }e\in\mathcal{V}_N\quad\text{and}\quad\sum_{e\in\mathcal{V}_N}q_e=1, \tag*{(47)} q e ∈ [ 0 , 1 ] for all e ∈ V N and e ∈ V N ∑ q e = 1 , (47) and D \mathcal{D} D is a subnormalized Pauli map if and only if
q e ∈ [ 0 , 1 ] for all e ∈ V N and ∑ e ∈ V N q e ≤ 1. (48) q_e\in[0,1]\quad\text{for all }e\in\mathcal{V}_N\quad\text{and}\quad\sum_{e\in\mathcal{V}_N}q_e\le1. \tag*{(48)} q e ∈ [ 0 , 1 ] for all e ∈ V N and e ∈ V N ∑ q e ≤ 1. (48)
Proof. Suppose first that q e ≥ 0 q_e\ge0 q e ≥ 0 for every e ∈ V N e\in\mathcal{V}_N e ∈ V N . Then Eq. (A.41) has the Kraus form
D ( A ) = ∑ e ∈ V N ( q e W ( e ) ) A ( q e W ( e ) ) † , (49) \mathcal{D}(A)=\sum_{e\in\mathcal{V}_N}\left(\sqrt{q_e}\mathcal{W}(e)\right)A\left(\sqrt{q_e}\mathcal{W}(e)\right)^\dagger, \tag*{(49)} D ( A ) = e ∈ V N ∑ ( q e W ( e ) ) A ( q e W ( e ) ) † , (49)
for all A ∈ A A\in\mathcal{A} A ∈ A , so D \mathcal{D} D is completely positive.
Conversely, suppose that D \mathcal{D} D is completely positive. By the finite-dimensional Kraus representation theorem, there exist r ∈ N r\in\mathbb{N} r ∈ N and a family { A μ } μ = 1 r ⊆ A \{A_\mu\}_{\mu=1}^{r}\subseteq\mathcal{A} { A μ } μ = 1 r ⊆ A such that
D ( X ) = ∑ μ = 1 r A μ X A μ † , (50) \mathcal{D}(X)=\sum_{\mu=1}^{r}A_\mu X A_\mu^\dagger, \tag*{(50)} D ( X ) = μ = 1 ∑ r A μ X A μ † , (50)
for all X ∈ A X\in\mathcal{A} X ∈ A . Next, substituting Eq. (A.59) into Eq. (A.43) and using Lemma G.4 and Eqs. (A.5) and (G.12) gives
q e = 1 d 2 ∑ μ = 1 r tr A [ X ↦ W ( e ) † A μ X A μ † W ( e ) ] = 1 d 2 ∑ μ = 1 r tr ( W ( e ) † A μ ) tr ( A μ † W ( e ) ) = 1 d 2 ∑ μ = 1 r ∣ tr ( W ( e ) † A μ ) ∣ 2 ≥ 0. (51) \begin{aligned}
q_e&=\frac{1}{d^2}\sum_{\mu=1}^{r}\operatorname{tr}_{\mathcal{A}}\left[X\mapsto\mathcal{W}(e)^\dagger A_\mu X A_\mu^\dagger\mathcal{W}(e)\right]\\
&=\frac{1}{d^2}\sum_{\mu=1}^{r}\operatorname{tr}\left(\mathcal{W}(e)^\dagger A_\mu\right)\operatorname{tr}\left(A_\mu^\dagger\mathcal{W}(e)\right)\\
&=\frac{1}{d^2}\sum_{\mu=1}^{r}\left|\operatorname{tr}\left(\mathcal{W}(e)^\dagger A_\mu\right)\right|^2\ge0. \tag*{(51)}
\end{aligned} q e = d 2 1 μ = 1 ∑ r tr A [ X ↦ W ( e ) † A μ X A μ † W ( e ) ] = d 2 1 μ = 1 ∑ r tr ( W ( e ) † A μ ) tr ( A μ † W ( e ) ) = d 2 1 μ = 1 ∑ r tr ( W ( e ) † A μ ) 2 ≥ 0. (51)
Thus, D \mathcal{D} D is completely positive if and only if q e ≥ 0 q_e\ge0 q e ≥ 0 for every e ∈ V N e\in\mathcal{V}_N e ∈ V N .
Next, Eqs. (A.50) and (A.51) show that D \mathcal{D} D is trace preserving if and only if ∑ e ∈ V N q e = 1 \sum_{e\in\mathcal{V}_N}q_e=1 ∑ e ∈ V N q e = 1 . If D \mathcal{D} D is completely positive, the same equations and Definition A.52 show that D \mathcal{D} D is trace nonincreasing if and only if ∑ e ∈ V N q e ≤ 1 \sum_{e\in V_{\mathrm{N}}}q_e\le1 ∑ e ∈ V N q e ≤ 1 . In either case, nonnegativity and the bound on the sum imply that q e ≤ 1 q_e\le1 q e ≤ 1 for every e ∈ V N e\in V_{\mathrm{N}} e ∈ V N . The lemma follows from Definition A.54. □ \square □
Proposition A.61. The standard Pauli twirl T \mathcal{T} T preserves the classes of completely positive, trace-preserving, and trace-nonincreasing maps.
Proof. For E ∈ S \mathcal{E}\in\mathcal{S} E ∈ S and v ∈ V N v\in V_{\mathrm{N}} v ∈ V N , Eqs. Eqs. (A.26), (A.28), and (G.6) give
Γ ( v ) ( E ) = W ( v ) ∘ E ∘ W ( v ) . \Gamma(v)(\mathcal{E})=\mathcal{W}(v)\circ\mathcal{E}\circ\mathcal{W}(v). Γ ( v ) ( E ) = W ( v ) ∘ E ∘ W ( v ) . Because each adjoint action in this expression is conjugation by a unitary, the transformation E ↦ Γ ( v ) ( E ) \mathcal{E}\mapsto\Gamma(v)(\mathcal{E}) E ↦ Γ ( v ) ( E ) preserves complete positivity, trace preservation, and the class of trace-nonincreasing maps. The average in Eq. (A.34) therefore preserves these classes. □ \square □
Pauli twirling with respect to a Clifford
The material reviewed in this section is standard in the community [38 , 47 , 48 ] . We include the full derivations in order to establish notation and provide a self-contained treatment.
Definition B.1 (Clifford unitary). A unitary C C C on K N \mathcal{K}_{\mathrm{N}} K N is a Clifford unitary if there exist a symplectic automorphism F C : V N → V N F_C:V_{\mathrm{N}}\to V_{\mathrm{N}} F C : V N → V N and a map η C : V N → { + 1 , − 1 } \eta_C:V_{\mathrm{N}}\to\{+1,-1\} η C : V N → { + 1 , − 1 } such that
C ( W ( a ) ) = C W ( a ) C † = η C ( a ) W ( F C a ) , (52) C(W(a))=CW(a)C^\dagger=\eta_C(a)W(F_Ca), \tag*{(52)} C ( W ( a )) = C W ( a ) C † = η C ( a ) W ( F C a ) , (52) for all labels a ∈ V N a\in V_{\mathrm{N}} a ∈ V N , using the convention of Eq. (B.5) for C C C .
Fix a Clifford unitary C C C , a symplectic automorphism F C : V N → V N F_C:V_{\mathrm{N}}\to V_{\mathrm{N}} F C : V N → V N , and a map η C : V N → { + 1 , − 1 } \eta_C:V_{\mathrm{N}}\to\{+1,-1\} η C : V N → { + 1 , − 1 } satisfying Eq. (B.2). The maps F C F_C F C and η C \eta_C η C are unique. To see this, suppose that a symplectic automorphism F ~ C : V N → V N \widetilde{F}_C:V_{\mathrm{N}}\to V_{\mathrm{N}} F C : V N → V N and a map η ~ C : V N → { + 1 , − 1 } \widetilde{\eta}_C:V_{\mathrm{N}}\to\{+1,-1\} η C : V N → { + 1 , − 1 } also satisfy Eq. (B.2). Then, for every a ∈ V N a\in V_{\mathrm{N}} a ∈ V N , the two instances of that equation give η C ( a ) W ( F C a ) = η ~ C ( a ) W ( F ~ C a ) \eta_C(a)W(F_Ca)=\widetilde{\eta}_C(a)W(\widetilde{F}_Ca) η C ( a ) W ( F C a ) = η C ( a ) W ( F C a ) . Equation (A.25) then gives F C a = F ~ C a F_Ca=\widetilde{F}_Ca F C a = F C a , and the same equality leads to η C ( a ) = η ~ C ( a ) \eta_C(a)=\widetilde{\eta}_C(a) η C ( a ) = η C ( a ) .
Definition B.3 (Pauli twirl with respect to C C C ). Define the Pauli twirl with respect to C C C T C N : S → S \mathcal{T}^{\mathrm{N}}_C:\mathcal{S}\to\mathcal{S} T C N : S → S by
T C N ( Φ ) = 1 ∣ V N ∣ ∑ v ∈ V N W ( F C v ) ∘ Φ ∘ W ( v ) , (53) \mathcal{T}^{\mathrm{N}}_C(\Phi)=\frac{1}{|V_{\mathrm{N}}|}\sum_{v\in V_{\mathrm{N}}}\mathcal{W}(F_Cv)\circ\Phi\circ\mathcal{W}(v), \tag*{(53)} T C N ( Φ ) = ∣ V N ∣ 1 v ∈ V N ∑ W ( F C v ) ∘ Φ ∘ W ( v ) , (53) for all Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S .
When the register is unambiguous, we similarly write T C \mathcal{T}_C T C for T C N \mathcal{T}^{\mathrm{N}}_C T C N .
For a unitary operator U ∈ End ( H ) U\in\operatorname{End}(\mathcal{H}) U ∈ End ( H ) , define the associated maps
U ( Y ) : = U Y U † and U − 1 ( Y ) : = U † Y U , (54) \mathcal{U}(Y):=UYU^\dagger\qquad\text{and}\qquad\mathcal{U}^{-1}(Y):=U^\dagger YU, \tag*{(54)} U ( Y ) := U Y U † and U − 1 ( Y ) := U † Y U , (54)
for all Y ∈ End ( H ) Y\in\operatorname{End}(\mathcal{H}) Y ∈ End ( H ) . The convention is that the standard font denotes the unitary and the caligraphic font denotes its adjoint action, as described above.
Define the map R C : S → S R_C:\mathcal{S}\to\mathcal{S} R C : S → S by
R C ( Φ ) = Φ ∘ C † (55) R_C(\Phi)=\Phi\circ\mathcal{C}^{\dagger} \tag*{(55)} R C ( Φ ) = Φ ∘ C † (55)
for all Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S and any unitary C C C with associated adjoint map C † C^{\dagger} C † .
Lemma B.7. The map R C R_C R C is unitary on S \mathcal{S} S . Its inverse is given by
R C − 1 ( Φ ) = Φ ∘ C , (56) R_C^{-1}(\Phi)=\Phi\circ C, \tag*{(56)} R C − 1 ( Φ ) = Φ ∘ C , (56) for all Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S , and
T C = R C − 1 ∘ T ∘ R C . (57) \mathcal{T}_C=R_C^{-1}\circ\mathcal{T}\circ R_C. \tag*{(57)} T C = R C − 1 ∘ T ∘ R C . (57)
Proof. For Φ , Ψ ∈ S \Phi,\Psi\in\mathcal{S} Φ , Ψ ∈ S , Eqs. (A.5), (G.5), and (G.6), together with cyclicity of tr A \operatorname{tr}_{A} tr A , give
⟨ R C ( Φ ) , R C ( Ψ ) ⟩ S = tr A ( C ∘ Φ † ∘ Ψ ∘ C − 1 ) = tr A ( Φ † ∘ Ψ ) = ⟨ Φ , Ψ ⟩ S . (58) \langle R_C(\Phi),R_C(\Psi)\rangle_{\mathcal{S}}=\operatorname{tr}_{A}\left(C\circ\Phi^{\dagger}\circ\Psi\circ C^{-1}\right)=\operatorname{tr}_{A}\left(\Phi^{\dagger}\circ\Psi\right)=\langle\Phi,\Psi\rangle_{\mathcal{S}}. \tag*{(58)} ⟨ R C ( Φ ) , R C ( Ψ ) ⟩ S = tr A ( C ∘ Φ † ∘ Ψ ∘ C − 1 ) = tr A ( Φ † ∘ Ψ ) = ⟨ Φ , Ψ ⟩ S . (58)
Thus, R C R_C R C is unitary. Next, Eq. (G.5) gives C ∘ C − 1 = C − 1 ∘ C = id A C\circ C^{-1}=C^{-1}\circ C=\operatorname{id}_{A} C ∘ C − 1 = C − 1 ∘ C = id A , which proves Eq. (B.8). Rearranging Eq. (B.2) gives C † W ( F C v ) C = η C ( v ) W ( v ) C^{\dagger}\mathcal{W}(F_Cv)C=\eta_C(v)\mathcal{W}(v) C † W ( F C v ) C = η C ( v ) W ( v ) . Since Ad η C ( v ) W ( v ) = W ( v ) \operatorname{Ad}_{\eta_C(v)}\mathcal{W}(v)=\mathcal{W}(v) Ad η C ( v ) W ( v ) = W ( v ) , Eq. (G.5) gives
C − 1 ∘ W ( F C v ) ∘ C = W ( v ) , (59) C^{-1}\circ\mathcal{W}(F_Cv)\circ C=\mathcal{W}(v), \tag*{(59)} C − 1 ∘ W ( F C v ) ∘ C = W ( v ) , (59)
for all v ∈ V N v\in V_{\mathsf{N}} v ∈ V N . For Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S , Eqs. (A.34), (A.26), (A.28), (G.6), (B.4), and (B.11), together with the change of variables u = F C v u=F_Cv u = F C v (which is valid because F C F_C F C is an automorphism of V N V_{\mathsf{N}} V N by Definition B.1), give
( R C − 1 ∘ T ∘ R C ) ( Φ ) = 1 ∣ V N ∣ ∑ u ∈ V N W ( u ) ∘ Φ ∘ C − 1 ∘ W ( u ) ∘ C = 1 ∣ V N ∣ ∑ v ∈ V N W ( F C v ) ∘ Φ ∘ W ( v ) = T C ( Φ ) , (60) \begin{aligned}
\left(R_C^{-1}\circ\mathcal{T}\circ R_C\right)(\Phi)
&=\frac{1}{\lvert V_{\mathsf{N}}\rvert}\sum_{u\in V_{\mathsf{N}}}\mathcal{W}(u)\circ\Phi\circ C^{-1}\circ\mathcal{W}(u)\circ C\\
&=\frac{1}{\lvert V_{\mathsf{N}}\rvert}\sum_{v\in V_{\mathsf{N}}}\mathcal{W}(F_Cv)\circ\Phi\circ\mathcal{W}(v)=\mathcal{T}_C(\Phi), \tag*{(60)}
\end{aligned} ( R C − 1 ∘ T ∘ R C ) ( Φ ) = ∣ V N ∣ 1 u ∈ V N ∑ W ( u ) ∘ Φ ∘ C − 1 ∘ W ( u ) ∘ C = ∣ V N ∣ 1 v ∈ V N ∑ W ( F C v ) ∘ Φ ∘ W ( v ) = T C ( Φ ) , (60)
which holds for every Φ ∈ S \Phi\in\mathcal{S} Φ ∈ S , and thus proves Eq. (B.9).
For E ∈ S \mathcal{E}\in\mathcal{S} E ∈ S , define the map D E ∈ S \mathcal{D}_{\mathcal{E}}\in\mathcal{S} D E ∈ S by D E : = T ( R C ( E ) ) \mathcal{D}_{\mathcal{E}}:=\mathcal{T}(R_C(\mathcal{E})) D E := T ( R C ( E )) . Proposition A.44 shows that D E \mathcal{D}_{\mathcal{E}} D E is a Pauli map. Denote by { q e [ E ] } e ∈ V N ⊆ C \{q_e[\mathcal{E}]\}_{e\in V_{\mathsf{N}}}\subseteq\mathbb{C} { q e [ E ] } e ∈ V N ⊆ C its unique coefficient family from Lemma A.42.
Proposition B.13. For every E ∈ S \mathcal{E}\in\mathcal{S} E ∈ S ,
T C ( E ) = D E ∘ C = ∑ e ∈ V N q e [ E ] W ( e ) ∘ C . (61) \mathcal{T}_C(\mathcal{E})=\mathcal{D}_{\mathcal{E}}\circ C=\sum_{e\in V_{\mathsf{N}}}q_e[\mathcal{E}]\,\mathcal{W}(e)\circ C. \tag*{(61)} T C ( E ) = D E ∘ C = e ∈ V N ∑ q e [ E ] W ( e ) ∘ C . (61) Moreover,
q e [ E ] = 1 d 2 ⟨ W ( e ) , R C ( E ) ⟩ S = 1 d 2 ⟨ W ( e ) ∘ C , E ⟩ S , (62) q_e[\mathcal{E}]=\frac{1}{d^2}\langle\mathcal{W}(e),R_C(\mathcal{E})\rangle_{\mathcal{S}}=\frac{1}{d^2}\langle\mathcal{W}(e)\circ C,\mathcal{E}\rangle_{\mathcal{S}}, \tag*{(62)} q e [ E ] = d 2 1 ⟨ W ( e ) , R C ( E ) ⟩ S = d 2 1 ⟨ W ( e ) ∘ C , E ⟩ S , (62) for all e ∈ V N e\in V_{\mathsf{N}} e ∈ V N .
Proof. Eqs. (B.9) and (B.8), together with the definition D E = T ( R C ( E ) ) \mathcal{D}_{\mathcal{E}}=\mathcal{T}(R_C(\mathcal{E})) D E = T ( R C ( E )) , give T C ( E ) = D E ∘ Ad C \mathcal{T}_C(\mathcal{E})=\mathcal{D}_{\mathcal{E}}\circ\operatorname{Ad}_C T C ( E ) = D E ∘ Ad C . By Lemma A.42, D E = ∑ e ∈ V N q e [ E ] W ( e ) \mathcal{D}_{\mathcal{E}}=\sum_{e\in V_{\mathsf{N}}}q_e[\mathcal{E}]\mathcal{W}(e) D E = ∑ e ∈ V N q e [ E ] W ( e ) . Substituting this equality gives Eq. (B.14). Applying Proposition A.44 to R C ( E ) R_C(\mathcal{E}) R C ( E ) and using the uniqueness in Lemma A.42 gives the first equality in Eq. (B.15). Unitarity of R C R_C R C (from Lemma B.7) and the inverse formula in Eq. (B.8) give
⟨ W ( e ) , R C ( E ) ⟩ S = ⟨ R C − 1 ( W ( e ) ) , E ⟩ S = ⟨ W ( e ) ∘ C , E ⟩ S , (63) \langle\mathcal{W}(e),R_C(\mathcal{E})\rangle_{\mathcal{S}}=\langle R_C^{-1}(\mathcal{W}(e)),\mathcal{E}\rangle_{\mathcal{S}}=\langle\mathcal{W}(e)\circ C,\mathcal{E}\rangle_{\mathcal{S}}, \tag*{(63)} ⟨ W ( e ) , R C ( E ) ⟩ S = ⟨ R C − 1 ( W ( e )) , E ⟩ S = ⟨ W ( e ) ∘ C , E ⟩ S , (63)
which proves the second equality in Eq. (B.15).
Corollary B.17. Let E ∈ S \mathcal{E} \in\mathcal{S} E ∈ S be completely positive and trace nonincreasing. Then D E \mathcal{D}_{\mathcal{E}} D E is a subnormalized Pauli map.
Proof. By Eq. (B.6), R C ( E ) = E ∘ C − 1 R_{\mathcal{C}}(\mathcal{E}) = \mathcal{E} \circ\mathcal{C}^{-1} R C ( E ) = E ∘ C − 1 . The map C − 1 \mathcal{C}^{-1} C − 1 is CPTP, so R C ( E ) R_{\mathcal{C}}(\mathcal{E}) R C ( E ) is completely positive and trace nonincreasing. Proposition A.61 then shows that D E = T ( R C ( E ) ) \mathcal{D}_{\mathcal{E}} = \mathcal{T}(R_{\mathcal{C}}(\mathcal{E})) D E = T ( R C ( E )) is completely positive and trace nonincreasing, while Proposition A.44 shows that it is a Pauli map. The conclusion follows from Definition A.54.
Leakage-dephased Pauli Twirling
Definition C.1 (Leakage-extended qubit space). A leakage-extended qubit space is a Hilbert space H \mathcal{H} H equipped with an orthogonal direct-sum decomposition
H = C ⊕ L . (64) \mathcal{H} = \mathcal{C} \oplus\mathcal{L}. \tag*{(64)} H = C ⊕ L . (64) The computational subspace C \mathcal{C} C is a single-qubit Hilbert space, as in Definition A.1. The leakage subspace L \mathcal{L} L is isomorphic to C \mathbb{C} C .
Definition C.3 (Leakage-extended qubit register). A leakage-extended qubit register Q \mathsf{Q} Q consists of a finite nonempty set of labels, also denoted Q \mathsf{Q} Q , and, for each i ∈ Q i \in\mathsf{Q} i ∈ Q , a leakage-extended qubit space H i \mathcal{H}_i H i . For each i ∈ Q i \in\mathsf{Q} i ∈ Q , denote the computational and leakage subspaces of H i \mathcal{H}_i H i by C i \mathcal{C}_i C i and L i \mathcal{L}_i L i , respectively, and denote the computational basis for C i \mathcal{C}_i C i by { ∣ 0 i ⟩ , ∣ 1 i ⟩ } \{\lvert0_i\rangle,\lvert1_i\rangle\} {∣ 0 i ⟩ , ∣ 1 i ⟩} . Define the total Hilbert space H Q \mathcal{H}_{\mathsf{Q}} H Q of Q \mathsf{Q} Q by
H Q = ⨂ i ∈ Q H i . (65) \mathcal{H}_{\mathsf{Q}} = \bigotimes_{i\in\mathsf{Q}} \mathcal{H}_i. \tag*{(65)} H Q = i ∈ Q ⨂ H i . (65)
Fix a leakage-extended qubit register and associated label set Q \mathsf{Q} Q .
Leakage-subspace decomposition
All tensor products in this subsection are indexed by the indicated labels. We use the canonical unitary identifications obtained by regrouping tensor products with the same labeled systems (tensor factors).
Definition C.5 (Leakage pattern). A leakage pattern of Q \mathsf{Q} Q is a subset of the label set S ⊆ Q \mathsf{S} \subseteq\mathsf{Q} S ⊆ Q . Write S ‾ = Q ∖ S \overline{\mathsf{S}} = \mathsf{Q} \setminus\mathsf{S} S = Q ∖ S , and define
L S = ⨂ i ∈ S L i , C S ‾ = ⨂ i ∈ S ‾ C i . (66) \mathcal{L}_{\mathsf{S}} = \bigotimes_{i\in\mathsf{S}} \mathcal{L}_i,\qquad\mathcal{C}_{\overline{\mathsf{S}}} = \bigotimes_{i\in\overline{\mathsf{S}}} \mathcal{C}_i. \tag*{(66)} L S = i ∈ S ⨂ L i , C S = i ∈ S ⨂ C i . (66) For i ∈ Q i \in\mathsf{Q} i ∈ Q , the condition i ∈ S i \in\mathsf{S} i ∈ S indicates that the system labeled by i i i is leaked, whereas i ∈ S ‾ i \in\overline{\mathsf{S}} i ∈ S indicates that it is computational.
Definition C.7 (Leakage subspace). For a leakage pattern S ⊆ Q \mathsf{S} \subseteq\mathsf{Q} S ⊆ Q , define the corresponding leakage subspace of H Q \mathcal{H}_{\mathsf{Q}} H Q by
H Q S = L S ⊗ C S ‾ ⊆ H Q . (67) \mathcal{H}_{\mathsf{Q}}^{\mathsf{S}} = \mathcal{L}_{\mathsf{S}} \otimes\mathcal{C}_{\overline{\mathsf{S}}} \subseteq\mathcal{H}_{\mathsf{Q}}. \tag*{(67)} H Q S = L S ⊗ C S ⊆ H Q . (67)
For each i ∈ Q i \in\mathrm{Q} i ∈ Q , define the orthogonal projectors on H i \mathcal{H}_i H i by
P i c = 1 c i ⊕ 0 l i , P i ℓ = 0 c i ⊕ 1 l i . (68) P_i^{\mathrm{c}} = 1_{\mathrm{c}_i} \oplus0_{\mathrm{l}_i}, \qquad P_i^{\ell} = 0_{\mathrm{c}_i} \oplus1_{\mathrm{l}_i}. \tag*{(68)} P i c = 1 c i ⊕ 0 l i , P i ℓ = 0 c i ⊕ 1 l i . (68)
We use the same symbols for their extensions to H Q \mathcal{H}_{\mathrm{Q}} H Q , which act as the corresponding projector on the system labeled by i i i and as the identity on every other system. They satisfy
P i c + P i ℓ = 1 H Q , P i c P i ℓ = P i ℓ P i c = 0. (69) P_i^{\mathrm{c}} + P_i^{\ell} = 1_{\mathcal{H}_{\mathrm{Q}}}, \qquad P_i^{\mathrm{c}}P_i^{\ell} = P_i^{\ell}P_i^{\mathrm{c}} = 0. \tag*{(69)} P i c + P i ℓ = 1 H Q , P i c P i ℓ = P i ℓ P i c = 0. (69)
Definition C.11 (Leakage-pattern projector). For a leakage pattern S ⊆ Q S \subseteq\mathrm{Q} S ⊆ Q , define the leakage-pattern projector Π S : H Q → H Q \Pi_S : \mathcal{H}_{\mathrm{Q}} \to\mathcal{H}_{\mathrm{Q}} Π S : H Q → H Q by
Π S = ( ∏ i ∈ S P i ℓ ) ( ∏ i ∈ S ‾ P i c ) . (70) \Pi_S = \left(\prod_{i \in S} P_i^{\ell}\right)\left(\prod_{i \in\overline{S}} P_i^{\mathrm{c}}\right). \tag*{(70)} Π S = ( i ∈ S ∏ P i ℓ ) i ∈ S ∏ P i c . (70) An empty product defaults to 1 H Q 1_{\mathcal{H}_{\mathrm{Q}}} 1 H Q .
By Eq. (C.10), the family { Π S } S ⊆ Q \{\Pi_S\}_{S \subseteq\mathrm{Q}} { Π S } S ⊆ Q is pairwise orthogonal and satisfies
∑ S ⊆ Q Π S = ∏ i ∈ Q ( P i c + P i ℓ ) = 1 H Q . (71) \sum_{S \subseteq\mathrm{Q}} \Pi_S = \prod_{i \in\mathrm{Q}} \left(P_i^{\mathrm{c}} + P_i^{\ell}\right) = 1_{\mathcal{H}_{\mathrm{Q}}}. \tag*{(71)} S ⊆ Q ∑ Π S = i ∈ Q ∏ ( P i c + P i ℓ ) = 1 H Q . (71)
Combining Eqs. (C.8) and (C.12) gives image ( Π S ) = H Q S \operatorname{image}(\Pi_S) = \mathcal{H}_{\mathrm{Q}}^S image ( Π S ) = H Q S for every S ⊆ Q S \subseteq\mathrm{Q} S ⊆ Q . Consequently, the resolution of the identity in Eq. (C.13) gives
H Q = ⨁ S ⊆ Q H Q S . (72) \mathcal{H}_{\mathrm{Q}} = \bigoplus_{S \subseteq\mathrm{Q}} \mathcal{H}_{\mathrm{Q}}^S. \tag*{(72)} H Q = S ⊆ Q ⨁ H Q S . (72)
Leakage-pattern dephasing
Definition C.15 (Leakage-pattern dephasing). Define the leakage-pattern dephasing channel Δ : End ( H Q ) → End ( H Q ) \Delta: \operatorname{End}(\mathcal{H}_{\mathrm{Q}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{Q}}) Δ : End ( H Q ) → End ( H Q ) by
Δ ( X ) = ∑ S ⊆ Q Π S X Π S , (73) \Delta(X) = \sum_{S \subseteq\mathrm{Q}} \Pi_S X \Pi_S, \tag*{(73)} Δ ( X ) = S ⊆ Q ∑ Π S X Π S , (73) for all X ∈ End ( H Q ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{Q}}) X ∈ End ( H Q ) .
Since each Π S \Pi_S Π S is an orthogonal projector, the family { Π S } S ⊆ Q \{\Pi_S\}_{S \subseteq\mathrm{Q}} { Π S } S ⊆ Q is a Kraus family for Δ \Delta Δ . Hence, Δ \Delta Δ is completely positive. Since Π S † Π S = Π S \Pi_S^\dagger\Pi_S = \Pi_S Π S † Π S = Π S for every S ⊆ Q S \subseteq\mathrm{Q} S ⊆ Q , Eq. (C.13) gives ∑ S ⊆ Q Π S † Π S = 1 H Q \sum_{S \subseteq\mathrm{Q}} \Pi_S^\dagger\Pi_S = 1_{\mathcal{H}_{\mathrm{Q}}} ∑ S ⊆ Q Π S † Π S = 1 H Q , so Δ \Delta Δ is also trace preserving. Thus, Δ \Delta Δ is CPTP. Pairwise orthogonality of the leakage-pattern projectors gives
( Δ ∘ Δ ) ( X ) = ∑ S , T ⊆ Q Π S Π T X Π T Π S = ∑ S ⊆ Q Π S X Π S = Δ ( X ) , (74) \begin{aligned}
(\Delta\circ\Delta)(X) &= \sum_{S,T \subseteq\mathrm{Q}} \Pi_S \Pi_T X \Pi_T \Pi_S \\
&= \sum_{S \subseteq\mathrm{Q}} \Pi_S X \Pi_S = \Delta(X), \tag*{(74)}
\end{aligned} ( Δ ∘ Δ ) ( X ) = S , T ⊆ Q ∑ Π S Π T X Π T Π S = S ⊆ Q ∑ Π S X Π S = Δ ( X ) , (74)
for all X ∈ End ( H Q ) X \in\operatorname{End}(\mathcal{H}_{Q}) X ∈ End ( H Q ) .
Lemma C.18. For each R ⊆ Q R \subseteq Q R ⊆ Q , define the operator U R : H Q → H Q U_R : \mathcal{H}_{Q} \to\mathcal{H}_{Q} U R : H Q → H Q by
U R = ∏ i ∈ R ( P i c − P i ℓ ) , (75) U_R = \prod_{i\in R}(P_i^{\mathrm{c}}-P_i^{\ell}), \tag*{(75)} U R = i ∈ R ∏ ( P i c − P i ℓ ) , (75) where U ∅ = 1 H Q U_{\varnothing} = \mathbb{1}_{\mathcal{H}_{Q}} U ∅ = 1 H Q . Every U R U_R U R is a Hermitian unitary, and
Δ = 2 − ∣ Q ∣ ∑ R ⊆ Q Ad U R = 2 − ∣ Q ∣ ∑ R ⊆ Q U R . (76) \Delta= 2^{-|Q|}\sum_{R\subseteq Q}\operatorname{Ad}_{U_R}
=2^{-|Q|}\sum_{R\subseteq Q}U_R. \tag*{(76)} Δ = 2 − ∣ Q ∣ R ⊆ Q ∑ Ad U R = 2 − ∣ Q ∣ R ⊆ Q ∑ U R . (76) Proof. Each P i c − P i ℓ P_i^{\mathrm{c}}-P_i^{\ell} P i c − P i ℓ is Hermitian, and Eq. (C.10) gives ( P i c − P i ℓ ) 2 = 1 H Q (P_i^{\mathrm{c}}-P_i^{\ell})^2=\mathbb{1}_{\mathcal{H}_{Q}} ( P i c − P i ℓ ) 2 = 1 H Q . Since operators supported on distinctly labeled systems commute, every U R U_R U R is a Hermitian unitary.
For i ∈ Q i\in Q i ∈ Q and S ⊆ Q S\subseteq Q S ⊆ Q , Eqs. (C.10) and (C.12) give ( P i c − P i ℓ ) Π S = − Π S (P_i^{\mathrm{c}}-P_i^{\ell})\Pi_S=-\Pi_S ( P i c − P i ℓ ) Π S = − Π S when i ∈ S i\in S i ∈ S , and ( P i c − P i ℓ ) Π S = Π S (P_i^{\mathrm{c}}-P_i^{\ell})\Pi_S=\Pi_S ( P i c − P i ℓ ) Π S = Π S when i ∉ S i\notin S i ∈ / S . Thus, for R , S ⊆ Q R,S\subseteq Q R , S ⊆ Q , U R Π S = ( − 1 ) ∣ R ∩ S ∣ Π S U_R\Pi_S=(-1)^{|R\cap S|}\Pi_S U R Π S = ( − 1 ) ∣ R ∩ S ∣ Π S . Multiplying Eq. (C.13) on the left by U R U_R U R gives U R = ∑ S ⊆ Q ( − 1 ) ∣ R ∩ S ∣ Π S U_R=\sum_{S\subseteq Q}(-1)^{|R\cap S|}\Pi_S U R = ∑ S ⊆ Q ( − 1 ) ∣ R ∩ S ∣ Π S .
If S = T S=T S = T , then ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ = 1 (-1)^{|R\cap S|+|R\cap T|}=1 ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ = 1 for every R ⊆ Q R\subseteq Q R ⊆ Q . If S ≠ T S\ne T S = T , choose i ∈ S △ T i\in S\mathbin{\triangle}T i ∈ S △ T , where △ \triangle △ denotes the symmetric difference. The involution R ↦ R △ { i } R\mapsto R\mathbin{\triangle}\{i\} R ↦ R △ { i } pairs the summands with opposite signs, since i i i belongs to exactly one of S S S and T T T . Consequently,
2 − ∣ Q ∣ ∑ R ⊆ Q ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ = { 1 , S = T , 0 , S ≠ T , (77) 2^{-|Q|}\sum_{R\subseteq Q}(-1)^{|R\cap S|+|R\cap T|}
=
\begin{cases}
1, & S=T,\\
0, & S\ne T,
\end{cases} \tag*{(77)} 2 − ∣ Q ∣ R ⊆ Q ∑ ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ = { 1 , 0 , S = T , S = T , (77) for all S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q . For X ∈ End ( H Q ) X\in\operatorname{End}(\mathcal{H}_{Q}) X ∈ End ( H Q ) , substituting U R = ∑ S ⊆ Q ( − 1 ) ∣ R ∩ S ∣ Π S U_R=\sum_{S\subseteq Q}(-1)^{|R\cap S|}\Pi_S U R = ∑ S ⊆ Q ( − 1 ) ∣ R ∩ S ∣ Π S and using Eq. (C.21) gives
2 − ∣ Q ∣ ∑ R ⊆ Q U R ( X ) = ∑ S , T ⊆ Q [ 2 − ∣ Q ∣ ∑ R ⊆ Q ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ ] Π S X Π T = ∑ S ⊆ Q Π S X Π S = Δ ( X ) . (78) \begin{aligned}
2^{-|Q|}\sum_{R\subseteq Q}U_R(X)
&=\sum_{S,T\subseteq Q}\left[2^{-|Q|}\sum_{R\subseteq Q}(-1)^{|R\cap S|+|R\cap T|}\right]\Pi_SX\Pi_T\\
&=\sum_{S\subseteq Q}\Pi_SX\Pi_S=\Delta(X). \tag*{(78)}
\end{aligned} 2 − ∣ Q ∣ R ⊆ Q ∑ U R ( X ) = S , T ⊆ Q ∑ 2 − ∣ Q ∣ R ⊆ Q ∑ ( − 1 ) ∣ R ∩ S ∣ + ∣ R ∩ T ∣ Π S X Π T = S ⊆ Q ∑ Π S X Π S = Δ ( X ) . (78)
Leakage-extended Paulis and the leakage-dephased Pauli twirl
Definition C.23 (Computational qubit register). Define the computational qubit register associated with Q Q Q , denoted K Q \mathcal{K}_{Q} K Q , to be the qubit register (see Definition A.2) with label set Q Q Q and family of qubit Hilbert spaces { C i } i ∈ Q \{ \mathcal{C}_{i}\}_{i\in Q} { C i } i ∈ Q . Denote its Hilbert space by K Q = ⨂ i ∈ Q C i \mathcal{K}_{Q}=\bigotimes_{i\in Q}\mathcal{C}_{i} K Q = ⨂ i ∈ Q C i .
Let V K Q \mathsf{V}_{\mathcal{K}_{Q}} V K Q be the Pauli-label space of K Q \mathcal{K}_{Q} K Q , as in Definition A.6. For a = ( x , z ) ∈ V K Q a=(x,z)\in\mathsf{V}_{\mathcal{K}_{Q}} a = ( x , z ) ∈ V K Q , write a i = ( x i , z i ) a_i=(x_i,z_i) a i = ( x i , z i ) for every i ∈ Q i\in Q i ∈ Q , and let w i K Q ( a i ) w_i^{\mathcal{K}_{Q}}(a_i) w i K Q ( a i ) and W K Q ( a ) W^{\mathcal{K}_{Q}}(a) W K Q ( a ) be the corresponding Pauli operators from Eqs. (A.10), (A.11), and (A.13), specialized to K Q \mathcal{K}_{Q} K Q (i.e., defined only on the computational register K Q \mathcal{K}_{Q} K Q ) and associated Hilbert space K Q \mathcal{K}_{Q} K Q .
Definition C.24 (Leakage-extended Pauli family). For every a ∈ V K Q a \in V_{K_Q} a ∈ V K Q , define the leakage-extended Pauli W Q ( a ) ∈ End ( H Q ) W^Q(a) \in\operatorname{End}(\mathcal{H}_Q) W Q ( a ) ∈ End ( H Q ) according to
W Q ( a ) = ⨂ i ∈ Q ( w i K Q ( a i ) ⊕ 1 l i ) . (79) W^Q(a)=\bigotimes_{i\in Q}\left(w_i^{K_Q}(a_i)\oplus1_{l_i}\right). \tag*{(79)} W Q ( a ) = i ∈ Q ⨂ ( w i K Q ( a i ) ⊕ 1 l i ) . (79) We call { W Q ( a ) } a ∈ V K Q \{W^Q(a)\}_{a\in V_{K_Q}} { W Q ( a ) } a ∈ V K Q the leakage-extended Pauli family associated with Q \mathrm{Q} Q . The computational-only Pauli family is distinguished by the superscript K Q K_Q K Q instead of Q \mathrm{Q} Q .
The inclusions c i ↪ h i c_i\hookrightarrow h_i c i ↪ h i induce a canonical inclusion K Q ↪ H Q \mathcal{K}_Q\hookrightarrow\mathcal{H}_Q K Q ↪ H Q , corresponding extension by 0 0 0 . We identify K Q \mathcal{K}_Q K Q with its image under this inclusion. For every a ∈ V K Q a\in V_{K_Q} a ∈ V K Q , applying Lemma A.21 to each single-qubit system in Eq. (C.25) shows that W Q ( a ) W^Q(a) W Q ( a ) is a Hermitian unitary and that its restriction to K Q \mathcal{K}_Q K Q is W K Q ( a ) W^{K_Q}(a) W K Q ( a ) .
Fix a Clifford unitary C C C on K Q \mathcal{K}_Q K Q and a symplectic automorphism F C : V K Q → V K Q F_C:V_{K_Q}\to V_{K_Q} F C : V K Q → V K Q associated with C C C , as in Definition B.1 .
Definition C.26 (Leakage-dephased Pauli twirl). Define the leakage-dephased Pauli twirl T C Q : End ( End ( H Q ) ) → End ( End ( H Q ) ) \mathcal{T}^Q_C:\operatorname{End}(\operatorname{End}(\mathcal{H}_Q))\to\operatorname{End}(\operatorname{End}(\mathcal{H}_Q)) T C Q : End ( End ( H Q )) → End ( End ( H Q )) by
T C Q ( Φ ) = Δ ∘ [ 1 ∣ V K Q ∣ ∑ v ∈ V K Q W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ] ∘ Δ , (80) \mathcal{T}^Q_C(\Phi)=\Delta\circ\left[\frac{1}{|V_{K_Q}|}\sum_{v\in V_{K_Q}}\mathcal{W}^Q(F_Cv)\circ\Phi\circ\mathcal{W}^Q(v)\right]\circ\Delta, \tag*{(80)} T C Q ( Φ ) = Δ ∘ ∣ V K Q ∣ 1 v ∈ V K Q ∑ W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ∘ Δ , (80) for all Φ ∈ End ( End ( H Q ) ) \Phi\in\operatorname{End}(\operatorname{End}(\mathcal{H}_Q)) Φ ∈ End ( End ( H Q )) , where W Q ( v ) = Ad W Q ( v ) \mathcal{W}^Q(v)=\operatorname{Ad}_{W^Q(v)} W Q ( v ) = Ad W Q ( v ) .
If Φ \Phi Φ is CPTP, then every map in the sum in Eq. (C.27) is also CPTP. Because the sum contains ∣ V K Q ∣ |V_{K_Q}| ∣ V K Q ∣ terms and is multiplied by 1 / ∣ V K Q ∣ 1/|V_{K_Q}| 1/∣ V K Q ∣ , the bracketed map is CPTP. Since Δ \Delta Δ is CPTP, it follows that T C Q ( Φ ) \mathcal{T}^Q_C(\Phi) T C Q ( Φ ) is also CPTP whenever Φ \Phi Φ is CPTP.
Filling and leakage maps
For a leakage pattern S ⊆ Q S\subseteq Q S ⊆ Q , define c S = ⨂ i ∈ S c i \mathcal{c}_S=\bigotimes_{i\in S}\mathcal{c}_i c S = ⨂ i ∈ S c i . Under the canonical tensor-product identifications, K Q = c S ⊗ c ‾ S \mathcal{K}_Q=\mathcal{c}_S\otimes\overline{\mathcal{c}}_S K Q = c S ⊗ c S .
Definition C.28 (Filling map). For S ⊆ Q S\subseteq Q S ⊆ Q , define the filling map F S : End ( H Q S ) → End ( K Q ) \mathcal{F}_S:\operatorname{End}(\mathcal{H}^S_Q)\to\operatorname{End}(\mathcal{K}_Q) F S : End ( H Q S ) → End ( K Q ) by
F S ( X ) = 1 2 ∣ S ∣ 1 c S ⊗ tr l S ( X ) , (81) \mathcal{F}_S(X)=\frac{1}{2^{|S|}}1_{\mathcal{c}_S}\otimes\operatorname{tr}_{l_S}(X), \tag*{(81)} F S ( X ) = 2 ∣ S ∣ 1 1 c S ⊗ tr l S ( X ) , (81) for all X ∈ End ( H Q S ) X\in\operatorname{End}(\mathcal{H}^S_Q) X ∈ End ( H Q S ) , where tr l S \operatorname{tr}_{l_S} tr l S denotes the partial trace over l S l_S l S with respect to the factorization in Eq. (C.8).
Definition C.30 (Leakage map). For S ⊆ Q S\subseteq Q S ⊆ Q , define the leakage map L S : End ( K Q ) → End ( H Q S ) \mathcal{L}_S:\operatorname{End}(\mathcal{K}_Q)\to\operatorname{End}(\mathcal{H}^S_Q) L S : End ( K Q ) → End ( H Q S ) by
L S ( Y ) = 1 l S ⊗ tr c S ( Y ) , (82) \mathcal{L}_S(Y)=1_{l_S}\otimes\operatorname{tr}_{c_S}(Y), \tag*{(82)} L S ( Y ) = 1 l S ⊗ tr c S ( Y ) , (82) for all Y ∈ End ( K Q ) Y\in\operatorname{End}(\mathcal{K}_Q) Y ∈ End ( K Q ) , where tr c S \operatorname{tr}_{c_S} tr c S denotes the partial trace over c S c_S c S with respect to the factorization K Q = c S ⊗ c ‾ S \mathcal{K}_Q=\mathcal{c}_S\otimes\overline{\mathcal{c}}_S K Q = c S ⊗ c S .
Lemma C.32. For every S ⊆ Q S \subseteq Q S ⊆ Q , the maps F S \mathcal{F}_S F S and L S \mathcal{L}_S L S are CPTP and
L S ∘ F S = id End ( H Q S ) . (83) \mathcal{L}_S \circ\mathcal{F}_S = \operatorname{id}_{\operatorname{End}(\mathcal{H}_Q^S)}. \tag*{(83)} L S ∘ F S = id End ( H Q S ) . (83)
Proof. For every S ⊆ Q S \subseteq Q S ⊆ Q , 1 c S 2 ∣ S ∣ = ⨂ i ∈ S ( 1 c i 2 ) \frac{1_{c_S}}{2^{|S|}} = \bigotimes_{i \in S} \left(\frac{1_{c_i}}{2}\right) 2 ∣ S ∣ 1 c S = ⨂ i ∈ S ( 2 1 c i ) is a density operator on c S c_S c S . Since l S l_S l S is one dimensional, 1 l S 1_{l_S} 1 l S is the unique density operator on l S l_S l S . The map F S \mathcal{F}_S F S is the composition of tr l S \operatorname{tr}_{l_S} tr l S with tensoring by 1 c S 2 ∣ S ∣ \frac{1_{c_S}}{2^{|S|}} 2 ∣ S ∣ 1 c S , and L S \mathcal{L}_S L S is the composition of tr c S \operatorname{tr}_{c_S} tr c S with tensoring by 1 l S 1_{l_S} 1 l S . Hence, both maps are CPTP. For every X ∈ End ( H Q S ) X \in\operatorname{End}(\mathcal{H}_Q^S) X ∈ End ( H Q S ) , Eqs. (C.29) and (C.31) give ( L S ∘ F S ) ( X ) = 1 l S ⊗ tr l S ( X ) = X (\mathcal{L}_S \circ\mathcal{F}_S)(X) = 1_{l_S} \otimes\operatorname{tr}_{l_S}(X) = X ( L S ∘ F S ) ( X ) = 1 l S ⊗ tr l S ( X ) = X , where the second equality uses the fact that l S l_S l S is one dimensional. □
Definition C.34 (Computational replacement channel). For S ⊆ Q S \subseteq Q S ⊆ Q , define the computational replacement channel G S : End ( K Q ) → End ( K Q ) \mathcal{G}_S : \operatorname{End}(\mathcal{K}_Q) \to\operatorname{End}(\mathcal{K}_Q) G S : End ( K Q ) → End ( K Q ) by
G S = F S ∘ L S . (84) \mathcal{G}_S = \mathcal{F}_S \circ\mathcal{L}_S. \tag*{(84)} G S = F S ∘ L S . (84)
Combining Eqs. (C.29) and (C.31) gives
G S ( Y ) = 1 2 ∣ S ∣ 1 c S ⊗ tr c S ( Y ) , (85) \mathcal{G}_S(Y) = \frac{1}{2^{|S|}}1_{c_S} \otimes\operatorname{tr}_{c_S}(Y), \tag*{(85)} G S ( Y ) = 2 ∣ S ∣ 1 1 c S ⊗ tr c S ( Y ) , (85)
for all Y ∈ End ( K Q ) Y \in\operatorname{End}(\mathcal{K}_Q) Y ∈ End ( K Q ) . Define the subgroup of Pauli labels supported on S S S by
V S = { a ∈ V K Q : a i = ( 0 , 0 ) for every i ∈ S ‾ } . (86) V_S = \left\{a \in V_{\mathcal{K}_Q} : a_i = (0,0)\ \text{for every } i \in\overline{S}\right\}. \tag*{(86)} V S = { a ∈ V K Q : a i = ( 0 , 0 ) for every i ∈ S } . (86)
For every i ∈ Q i \in Q i ∈ Q and A ∈ End ( c i ) A \in\operatorname{End}(c_i) A ∈ End ( c i ) , we have that
1 2 1 c i tr ( A ) = 1 4 ∑ a i ∈ Z 2 2 w i K Q ( a i ) A w i K Q ( a i ) . (87) \frac{1}{2}1_{c_i}\operatorname{tr}(A) = \frac{1}{4}\sum_{a_i \in\mathbb{Z}_2^2} w_i^{\mathcal{K}_Q}(a_i) A w_i^{\mathcal{K}_Q}(a_i). \tag*{(87)} 2 1 1 c i tr ( A ) = 4 1 a i ∈ Z 2 2 ∑ w i K Q ( a i ) A w i K Q ( a i ) . (87)
It follows from Eq. (C.38) that
G S = 1 4 ∣ S ∣ ∑ a ∈ V S W K Q ( a ) . (88) \mathcal{G}_S = \frac{1}{4^{|S|}}\sum_{a \in V_S}\mathcal{W}^{\mathcal{K}_Q}(a). \tag*{(88)} G S = 4 ∣ S ∣ 1 a ∈ V S ∑ W K Q ( a ) . (88)
Since ∣ V S ∣ = 4 ∣ S ∣ |V_S| = 4^{|S|} ∣ V S ∣ = 4 ∣ S ∣ , the coefficients in Eq. (C.39) are nonnegative and sum to one. It follows from Definition A.54 that G S \mathcal{G}_S G S is a Pauli channel. Combining Eqs. (C.12) and (C.25) gives
W Q ( a ) Π S = Π S W Q ( a ) , (89) W^Q(a)\Pi_S = \Pi_S W^Q(a), \tag*{(89)} W Q ( a ) Π S = Π S W Q ( a ) , (89)
for all a ∈ V K Q a \in V_{\mathcal{K}_Q} a ∈ V K Q , S ⊆ Q S \subseteq Q S ⊆ Q . Thus, H Q S \mathcal{H}_Q^S H Q S is invariant under every W Q ( a ) W^Q(a) W Q ( a ) .
Definition C.41 (Subspace-S S S Pauli). For a leakage pattern S ⊆ Q S \subseteq Q S ⊆ Q and a Pauli label a ∈ V K Q a \in V_{\mathcal{K}_Q} a ∈ V K Q , define the subspace-S S S Pauli W S Q ( a ) ∈ End ( H Q S ) W_S^Q(a) \in\operatorname{End}(\mathcal{H}_Q^S) W S Q ( a ) ∈ End ( H Q S ) by
W S Q ( a ) = W Q ( a ) ∣ H Q S . (90) W_S^Q(a) = W^Q(a)\big|_{\mathcal{H}_Q^S}. \tag*{(90)} W S Q ( a ) = W Q ( a ) H Q S . (90)
The restriction is well defined because W Q ( a ) W^Q(a) W Q ( a ) preserves the subspace H Q S \mathcal{H}_Q^S H Q S .
Combining Eqs. (C.8) and (C.25) gives
W S Q ( a ) = 1 I S ⊗ ⨂ i ∈ S ‾ w i K Q ( a i ) . (91) W_{S}^{Q}(a)=1_{\mathcal{I}_{S}}\otimes\bigotimes_{i\in\overline{S}}w_{i}^{K_{Q}}(a_{i}). \tag*{(91)} W S Q ( a ) = 1 I S ⊗ i ∈ S ⨂ w i K Q ( a i ) . (91)
For every S ⊆ Q S\subseteq Q S ⊆ Q and a ∈ V K Q a\in V_{K_{Q}} a ∈ V K Q , the subspace-S S S Pauli W S Q ( a ) W_{S}^{Q}(a) W S Q ( a ) is a Hermitian unitary.
Lemma C.44. For every S ⊆ Q S\subseteq Q S ⊆ Q and v ∈ V K Q v\in V_{K_{Q}} v ∈ V K Q ,
F S ∘ W S Q ( v ) = W K Q ( v ) ∘ F S . (92) \mathcal{F}_{S}\circ W_{S}^{Q}(v)=W^{K_{Q}}(v)\circ\mathcal{F}_{S}. \tag*{(92)} F S ∘ W S Q ( v ) = W K Q ( v ) ∘ F S . (92) L S ∘ W K Q ( v ) = W S Q ( v ) ∘ L S . (93) \mathcal{L}_{S}\circ W^{K_{Q}}(v)=W_{S}^{Q}(v)\circ\mathcal{L}_{S}. \tag*{(93)} L S ∘ W K Q ( v ) = W S Q ( v ) ∘ L S . (93)
Proof. The equality in Eq. (C.45) follows from Eqs. (C.29), (A.13), and (C.43), since conjugation by ⨂ i ∈ S w i K Q ( v i ) \bigotimes_{i\in S}w_{i}^{K_{Q}}(v_{i}) ⨂ i ∈ S w i K Q ( v i ) fixes 1 I S / 2 ∣ S ∣ 1_{\mathcal{I}_{S}}/2^{|S|} 1 I S / 2 ∣ S ∣ . The equality in Eq. (C.46) follows from Eqs. (C.31), (A.13), and (C.43), together with invariance of tr C S \operatorname{tr}_{\mathcal{C}_{S}} tr C S under unitary conjugation on C S \mathcal{C}_{S} C S .
Leakage-subspace inclusion, projection, and extension maps
For each S ⊆ Q S\subseteq Q S ⊆ Q , denote by ι S : H Q S ↪ H Q \iota_{S}:\mathcal{H}_{Q}^{S}\hookrightarrow\mathcal{H}_{Q} ι S : H Q S ↪ H Q the canonical inclusion. Its adjoint ι S † : H Q → H Q S \iota_{S}^{\dagger}:\mathcal{H}_{Q}\to\mathcal{H}_{Q}^{S} ι S † : H Q → H Q S is the orthogonal projection onto H Q S \mathcal{H}_{Q}^{S} H Q S , with codomain restricted to H Q S \mathcal{H}_{Q}^{S} H Q S . Hence,
ι S † ι S = 1 H Q S and ι S ι S † = Π S . (94) \iota_{S}^{\dagger}\iota_{S}=1_{\mathcal{H}_{Q}^{S}}
\qquad\text{and}\qquad
\iota_{S}\iota_{S}^{\dagger}=\Pi_{S}. \tag*{(94)} ι S † ι S = 1 H Q S and ι S ι S † = Π S . (94)
Definition C.48 (Leakage-subspace projection map). For S ⊆ Q S\subseteq Q S ⊆ Q , define the leakage-subspace projection map P S : End ( H Q ) → End ( H Q S ) \mathcal{P}_{S}:\operatorname{End}(\mathcal{H}_{Q})\to\operatorname{End}(\mathcal{H}_{Q}^{S}) P S : End ( H Q ) → End ( H Q S ) by
P S ( X ) = ι S † X ι S , (95) \mathcal{P}_{S}(X)=\iota_{S}^{\dagger}X\iota_{S}, \tag*{(95)} P S ( X ) = ι S † X ι S , (95)
for all X ∈ End ( H Q ) X\in\operatorname{End}(\mathcal{H}_{Q}) X ∈ End ( H Q ) .
Definition C.50 (Leakage-subspace extension map). For S ⊆ Q S\subseteq Q S ⊆ Q , define the leakage-subspace extension map E S : End ( H Q S ) → End ( H Q ) \mathcal{E}_{S}:\operatorname{End}(\mathcal{H}_{Q}^{S})\to\operatorname{End}(\mathcal{H}_{Q}) E S : End ( H Q S ) → End ( H Q ) by
E S ( Y ) = ι S Y ι S † , (96) \mathcal{E}_{S}(Y)=\iota_{S}Y\iota_{S}^{\dagger}, \tag*{(96)} E S ( Y ) = ι S Y ι S † , (96)
for all Y ∈ End ( H Q S ) Y\in\operatorname{End}(\mathcal{H}_{Q}^{S}) Y ∈ End ( H Q S ) .
The operator E S ( Y ) \mathcal{E}_{S}(Y) E S ( Y ) is supported on H Q S \mathcal{H}_{Q}^{S} H Q S , and its restriction to H Q S \mathcal{H}_{Q}^{S} H Q S is Y Y Y . Both P S \mathcal{P}_{S} P S and E S \mathcal{E}_{S} E S are completely positive, since each has a single Kraus operator. For Y ∈ End ( H Q S ) Y\in\operatorname{End}(\mathcal{H}_{Q}^{S}) Y ∈ End ( H Q S ) and X ∈ End ( H Q ) X\in\operatorname{End}(\mathcal{H}_{Q}) X ∈ End ( H Q ) with X ⪰ 0 X\succeq0 X ⪰ 0 , Eq. (C.47) gives
tr ( E S ( Y ) ) = tr ( Y ) and tr ( P S ( X ) ) = tr ( Π S X ) ≤ tr ( X ) . (97) \operatorname{tr}(\mathcal{E}_{S}(Y))=\operatorname{tr}(Y)
\qquad\text{and}\qquad
\operatorname{tr}(\mathcal{P}_{S}(X))=\operatorname{tr}(\Pi_{S}X)\leq\operatorname{tr}(X). \tag*{(97)} tr ( E S ( Y )) = tr ( Y ) and tr ( P S ( X )) = tr ( Π S X ) ≤ tr ( X ) . (97)
Thus, E S \mathcal{E}_{S} E S is trace preserving and P S \mathcal{P}_{S} P S is trace nonincreasing. Equation (C.47) also gives
P S ∘ E S = id End ( H Q S ) , (98) \mathcal{P}_{S}\circ\mathcal{E}_{S}=\operatorname{id}_{\operatorname{End}(\mathcal{H}_{Q}^{S})}, \tag*{(98)} P S ∘ E S = id End ( H Q S ) , (98)
( E S ∘ P S ) ( X ) = Π S X Π S . (99) (\mathcal{E}_{S}\circ\mathcal{P}_{S})(X)=\Pi_{S}X\Pi_{S}. \tag*{(99)} ( E S ∘ P S ) ( X ) = Π S X Π S . (99)
Next, combining Eqs. (C.16) and (C.54) gives
Δ = ∑ S ⊆ Q E S ∘ P S , (100) \Delta= \sum_{S \subseteq Q} \mathcal{E}_S \circ\mathcal{P}_S, \tag*{(100)} Δ = S ⊆ Q ∑ E S ∘ P S , (100)
and, for S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , pairwise orthogonality of the leakage subspaces gives
P T ∘ E S = { id End ( H Q S ) T = S 0 End ( H Q S ) → End ( H Q T ) T ≠ S . (101) \mathcal{P}_T \circ\mathcal{E}_S =
\begin{cases}
\operatorname{id}_{\operatorname{End}(\mathcal{H}_Q^S)} & T = S \\
0_{\operatorname{End}(\mathcal{H}_Q^S) \to\operatorname{End}(\mathcal{H}_Q^T)} & T \ne S.
\end{cases} \tag*{(101)} P T ∘ E S = { id End ( H Q S ) 0 End ( H Q S ) → End ( H Q T ) T = S T = S . (101)
Indeed, ι T † ι S = 0 \iota_T^\dagger\iota_S = 0 ι T † ι S = 0 when T ≠ S T \ne S T = S , while Eq. (C.53) covers the case T = S T = S T = S .
Leakage-transition branches
Definition C.57 (Leakage-transition branches). Let Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal{H}_Q) \to\operatorname{End}(\mathcal{H}_Q) Φ : End ( H Q ) → End ( H Q ) be a linear map. For every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , define the S → T S \to T S → T leakage-transition branch Φ T , S : End ( H Q S ) → End ( H Q T ) \Phi_{T,S} : \operatorname{End}(\mathcal{H}_Q^S) \to\operatorname{End}(\mathcal{H}_Q^T) Φ T , S : End ( H Q S ) → End ( H Q T ) by
Φ T , S = P T ∘ Φ ∘ E S , (102) \Phi_{T,S} = \mathcal{P}_T \circ\Phi\circ\mathcal{E}_S, \tag*{(102)} Φ T , S = P T ∘ Φ ∘ E S , (102) and define the family of leakage-transition branches associated with Φ \Phi Φ to be { Φ T , S } S , T ⊆ Q \{\Phi_{T,S}\}_{S,T \subseteq Q} { Φ T , S } S , T ⊆ Q .
Lemma C.59. Let Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal{H}_Q) \to\operatorname{End}(\mathcal{H}_Q) Φ : End ( H Q ) → End ( H Q ) be a C P T P \mathrm{CPTP} CPTP linear map and let { Φ T , S } S , T ⊆ Q \{\Phi_{T,S}\}_{S,T \subseteq Q} { Φ T , S } S , T ⊆ Q be the family of leakage-transition branches associated with Φ \Phi Φ , as in Definition C.57. Then, for every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , the map Φ T , S \Phi_{T,S} Φ T , S is completely positive and trace nonincreasing.
Proof. For every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , combining Eq. (C.58) with the fact that E S \mathcal{E}_S E S is CPTP and that P T \mathcal{P}_T P T is completely positive and trace nonincreasing gives the result.
For a fixed Clifford unitary C C C , denote by T C K Q \mathcal{T}^{K_Q}_C T C K Q the Pauli twirl with respect to C C C , as in Definition B.3, specialized to K Q K_Q K Q and C C C .
Proposition C.60. Let Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal{H}_Q) \to\operatorname{End}(\mathcal{H}_Q) Φ : End ( H Q ) → End ( H Q ) be a linear map. Let { Φ T , S } S , T ⊆ Q \{\Phi_{T,S}\}_{S,T \subseteq Q} { Φ T , S } S , T ⊆ Q and { ( T C K Q ( Φ ) ) T , S } S , T ⊆ Q \{(\mathcal{T}^{K_Q}_C(\Phi))_{T,S}\}_{S,T \subseteq Q} {( T C K Q ( Φ ) ) T , S } S , T ⊆ Q be the families of leakage-transition branches associated with Φ \Phi Φ and T C K Q ( Φ ) \mathcal{T}^{K_Q}_C(\Phi) T C K Q ( Φ ) , respectively, as in Definition C.57. Then, for every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q ,
( T C K Q ( Φ ) ) T , S = L T ∘ T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ F S . (103) (\mathcal{T}^{K_Q}_C(\Phi))_{T,S} = \mathcal{L}_T \circ\mathcal{T}^{K_Q}_C(\mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S) \circ\mathcal{F}_S. \tag*{(103)} ( T C K Q ( Φ ) ) T , S = L T ∘ T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ F S . (103)
Proof. Fix S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q . Combining Eqs. (C.40), (C.42), (C.49), and (C.51) gives
W Q ( v ) ∘ E S = E S ∘ W S Q ( v ) (104) \mathcal{W}^{Q}(v) \circ\mathcal{E}_S = \mathcal{E}_S \circ\mathcal{W}^{Q}_S(v) \tag*{(104)} W Q ( v ) ∘ E S = E S ∘ W S Q ( v ) (104)
P T ∘ W Q ( F C v ) = W T Q ( F C v ) ∘ P T (105) \mathcal{P}_T \circ\mathcal{W}^{Q}(F_Cv) = \mathcal{W}^{Q}_T(F_Cv) \circ\mathcal{P}_T \tag*{(105)} P T ∘ W Q ( F C v ) = W T Q ( F C v ) ∘ P T (105)
for every v ∈ V K Q v \in V_{K_Q} v ∈ V K Q . Combining Eqs. (C.55) and (C.56) also gives
P T ∘ Δ = P T , Δ ∘ E S = E S . (106) \mathcal{P}_T \circ\Delta= \mathcal{P}_T,\qquad\Delta\circ\mathcal{E}_S = \mathcal{E}_S. \tag*{(106)} P T ∘ Δ = P T , Δ ∘ E S = E S . (106)
Expanding T C K Q \mathcal{T}_{C}^{K_Q} T C K Q by Eq. (B.4) and using Eqs. (C.45), (C.46), and (C.33) gives
L T ∘ T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ F S = 1 ∣ V K Q ∣ ∑ v ∈ V K Q L T ∘ W K Q ( F C v ) † ∘ F T ∘ Φ T , S ∘ L S ∘ W K Q ( v ) ∘ F S = 1 ∣ V K Q ∣ ∑ v ∈ V K Q W T Q ( F C v ) ∘ ( L T ∘ F T ) ∘ Φ T , S ∘ W S Q ( v ) ∘ ( L S ∘ F S ) = 1 ∣ V K Q ∣ ∑ v ∈ V K Q W T Q ( F C v ) ∘ Φ T , S ∘ W S Q ( v ) . (107) \begin{aligned}
\mathcal{L}_T \circ\mathcal{T}_{C}^{K_Q}(\mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S) \circ\mathcal{F}_S
= \frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{L}_T \circ\mathcal{W}^{K_Q}(F_Cv)^\dagger\circ\mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S \circ\mathcal{W}^{K_Q}(v) \circ\mathcal{F}_S \\
&= \frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{W}^{Q}_T(F_Cv) \circ(\mathcal{L}_T \circ\mathcal{F}_T) \circ\Phi_{T,S} \circ\mathcal{W}^{Q}_S(v) \circ(\mathcal{L}_S \circ\mathcal{F}_S) \\
&= \frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{W}^{Q}_T(F_Cv) \circ\Phi_{T,S} \circ\mathcal{W}^{Q}_S(v). \tag*{(107)}
\end{aligned} L T ∘ T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ F S = ∣ V K Q ∣ 1 v ∈ V K Q ∑ L T ∘ W K Q ( F C v ) † ∘ F T ∘ Φ T , S ∘ L S ∘ W K Q ( v ) ∘ F S = ∣ V K Q ∣ 1 v ∈ V K Q ∑ W T Q ( F C v ) ∘ ( L T ∘ F T ) ∘ Φ T , S ∘ W S Q ( v ) ∘ ( L S ∘ F S ) = ∣ V K Q ∣ 1 v ∈ V K Q ∑ W T Q ( F C v ) ∘ Φ T , S ∘ W S Q ( v ) . (107)
Substituting Eq. (C.58) into Eq. (C.65) and applying Eqs. (C.62) and (C.63) termwise gives
1 ∣ V K Q ∣ ∑ v ∈ V K Q W T Q ( F C v ) ∘ Φ T , S ∘ W S Q ( v ) = P T ∘ [ 1 ∣ V K Q ∣ ∑ v ∈ V K Q W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ] ∘ E S = P T ∘ Δ ∘ [ 1 ∣ V K Q ∣ ∑ v ∈ V K Q W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ] ∘ Δ ∘ E S = P T ∘ T C Q ( Φ ) ∘ E S = ( T C Q ( Φ ) ) T , S . (108) \begin{aligned}
\frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{W}^{Q}_T(F_Cv) \circ\Phi_{T,S} \circ\mathcal{W}^{Q}_S(v)
= \mathcal{P}_T \circ\left[\frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{W}^{Q}(F_Cv) \circ\Phi\circ\mathcal{W}^{Q}(v)\right] \circ\mathcal{E}_S \\
&= \mathcal{P}_T \circ\Delta\circ\left[\frac{1}{|V_{K_Q}|} \sum_{v \in V_{K_Q}} \mathcal{W}^{Q}(F_Cv) \circ\Phi\circ\mathcal{W}^{Q}(v)\right] \circ\Delta\circ\mathcal{E}_S \\
&= \mathcal{P}_T \circ\mathcal{T}_{C}^{Q}(\Phi) \circ\mathcal{E}_S = \left(\mathcal{T}_{C}^{Q}(\Phi)\right)_{T,S}. \tag*{(108)}
\end{aligned} ∣ V K Q ∣ 1 v ∈ V K Q ∑ W T Q ( F C v ) ∘ Φ T , S ∘ W S Q ( v ) = P T ∘ ∣ V K Q ∣ 1 v ∈ V K Q ∑ W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ∘ E S = P T ∘ Δ ∘ ∣ V K Q ∣ 1 v ∈ V K Q ∑ W Q ( F C v ) ∘ Φ ∘ W Q ( v ) ∘ Δ ∘ E S = P T ∘ T C Q ( Φ ) ∘ E S = ( T C Q ( Φ ) ) T , S . (108)
The second equality uses Eq. (C.64), the third uses Eq. (C.27), and the last uses Eq. (C.58). Combining Eqs. (C.65) and (C.66) proves Eq. (C.61).
Leakage-transition Pauli maps
Definition C.67 (Leakage-transition Pauli-map family). Let Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal{H}_Q) \to\operatorname{End}(\mathcal{H}_Q) Φ : End ( H Q ) → End ( H Q ) be a linear map and let { Φ T , S } S , T ⊆ Q \{\Phi_{T,S}\}_{S,T \subseteq Q} { Φ T , S } S , T ⊆ Q be the family of leakage-transition branches associated with Φ \Phi Φ , as in Definition C.57. For every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q , define D T , S Φ : End ( K Q ) → End ( K Q ) \mathcal{D}^{\Phi}_{T,S} : \operatorname{End}(\mathcal{K}_Q) \to\operatorname{End}(\mathcal{K}_Q) D T , S Φ : End ( K Q ) → End ( K Q ) by
D T , S Φ = T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ C − 1 , (109) \mathcal{D}^{\Phi}_{T,S} = \mathcal{T}_{C}^{K_Q}(\mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S) \circ C^{-1}, \tag*{(109)} D T , S Φ = T C K Q ( F T ∘ Φ T , S ∘ L S ) ∘ C − 1 , (109) and define the leakage-transition Pauli-map family associated with Φ \Phi Φ to be { D T , S Φ } S , T ⊆ Q \{\mathcal{D}^{\Phi}_{T,S}\}_{S,T \subseteq Q} { D T , S Φ } S , T ⊆ Q .
By Proposition B.13, applied with E = F T ∘ Φ T , S ∘ L S \mathcal{E} = \mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S E = F T ∘ Φ T , S ∘ L S , the map D T , S Φ \mathcal{D}^{\Phi}_{T,S} D T , S Φ is a Pauli map for every S , T ⊆ Q S,T \subseteq Q S , T ⊆ Q . Combining Eq. (C.68) with C − 1 ∘ C = id End ( K Q ) C^{-1} \circ C = \operatorname{id}_{\operatorname{End}(\mathcal{K}_Q)} C − 1 ∘ C = id End ( K Q ) gives
T C K Q ( F T ∘ Φ T , S ∘ L S ) = D T , S Φ ∘ C . (110) \mathcal{T}_{C}^{K_Q}(\mathcal{F}_T \circ\Phi_{T,S} \circ\mathcal{L}_S) = \mathcal{D}^{\Phi}_{T,S} \circ C. \tag*{(110)} T C K Q ( F T ∘ Φ T , S ∘ L S ) = D T , S Φ ∘ C . (110)
Proposition C.70. Let Φ : End ( H Q ) → End ( H Q ) \Phi: \operatorname{End}(\mathcal{H}_Q) \to\operatorname{End}(\mathcal{H}_Q) Φ : End ( H Q ) → End ( H Q ) be a CPTP map and let { D T , S Φ } S , T ⊆ Q \{\mathcal{D}^{\Phi}_{T,S}\}_{S,T \subseteq Q} { D T , S Φ } S , T ⊆ Q be the leakage-transition Pauli-map family associated with Φ \Phi Φ , as in Definition C.67. Then D T , S Φ \mathcal{D}^{\Phi}_{T,S} D T , S Φ
is a subnormalized Pauli map for every S , T ⊆ Q S,T \subseteq\mathrm{Q} S , T ⊆ Q , and
∑ T ⊆ Q wt ( D T , S Φ ) = 1. (111) \sum_{T\subseteq\mathrm{Q}} \operatorname{wt}\left(D^\Phi_{T,S}\right)=1. \tag*{(111)} T ⊆ Q ∑ wt ( D T , S Φ ) = 1. (111)
for all S ⊆ Q S \subseteq\mathrm{Q} S ⊆ Q .
Proof. Denote by { Φ T , S } S , T ⊆ Q \{\Phi_{T,S}\}_{S,T\subseteq\mathrm{Q}} { Φ T , S } S , T ⊆ Q the family of leakage-transition branches associated with Φ \Phi Φ , as in Definition C.57. Fix S , T ⊆ Q S,T \subseteq\mathrm{Q} S , T ⊆ Q and set E = F T ∘ Φ T , S ∘ L S \mathcal{E}=\mathcal{F}_T\circ\Phi_{T,S}\circ\mathcal{L}_S E = F T ∘ Φ T , S ∘ L S . Lemma C.59 shows that Φ T , S \Phi_{T,S} Φ T , S is completely positive and trace nonincreasing. Lemma C.32 then shows that E \mathcal{E} E is completely positive and trace nonincreasing. Next, Eq. (C.68), Proposition B.13 (applied with this E \mathcal{E} E ), and C ∘ C − 1 = id End ( K Q ) \mathcal{C}\circ\mathcal{C}^{-1}=\operatorname{id}_{\operatorname{End}(\mathcal{K}_{\mathrm{Q}})} C ∘ C − 1 = id End ( K Q ) give D T , S Φ = D E D^\Phi_{T,S}=D_{\mathcal{E}} D T , S Φ = D E . Corollary B.17 therefore shows that D T , S Φ D^\Phi_{T,S} D T , S Φ is a subnormalized Pauli map. Since S , T S,T S , T were arbitrary, this holds for every S , T ⊆ Q S,T \subseteq\mathrm{Q} S , T ⊆ Q .
Let τ = 2 − ∣ Q ∣ 1 K Q \tau=2^{-|\mathrm{Q}|}1_{\mathcal{K}_{\mathrm{Q}}} τ = 2 − ∣ Q ∣ 1 K Q , and fix S ⊆ Q S\subseteq\mathrm{Q} S ⊆ Q . The state τ \tau τ is invariant under C \mathcal{C} C and every W K Q ( v ) W_{\mathcal{K}_{\mathrm{Q}}}(v) W K Q ( v ) . Definition A.47, linearity of the trace, Eqs. (B.4) and (C.69), and unitary invariance of the trace give the first two equalities in
∑ T ⊆ Q wt ( D T , S Φ ) = ∑ T ⊆ Q tr ( D T , S Φ ( τ ) ) = ∑ T ⊆ Q tr ( ( F T ∘ Φ T , S ∘ L S ) ( τ ) ) = ∑ T ⊆ Q tr ( ( P T ∘ Φ ∘ E S ∘ L S ) ( τ ) ) = tr ( ( ∑ T ⊆ Q Π T ) Φ ( ( E S ∘ L S ) ( τ ) ) ) = tr ( Φ ( ( E S ∘ L S ) ( τ ) ) ) = tr ( τ ) = 1. (112) \begin{aligned}
\sum_{T\subseteq\mathrm{Q}}\operatorname{wt}\left(D^\Phi_{T,S}\right)
&=\sum_{T\subseteq\mathrm{Q}}\operatorname{tr}\left(D^\Phi_{T,S}(\tau)\right)\\
&=\sum_{T\subseteq\mathrm{Q}}\operatorname{tr}\left((\mathcal{F}_T\circ\Phi_{T,S}\circ\mathcal{L}_S)(\tau)\right)\\
&=\sum_{T\subseteq\mathrm{Q}}\operatorname{tr}\left((\mathcal{P}_T\circ\Phi\circ\mathcal{E}_S\circ\mathcal{L}_S)(\tau)\right)\\
&=\operatorname{tr}\left(\left(\sum_{T\subseteq\mathrm{Q}}\Pi_T\right)\Phi((\mathcal{E}_S\circ\mathcal{L}_S)(\tau))\right)\\
&=\operatorname{tr}\left(\Phi((\mathcal{E}_S\circ\mathcal{L}_S)(\tau))\right)=\operatorname{tr}(\tau)=1. \tag*{(112)}
\end{aligned} T ⊆ Q ∑ wt ( D T , S Φ ) = T ⊆ Q ∑ tr ( D T , S Φ ( τ ) ) = T ⊆ Q ∑ tr ( ( F T ∘ Φ T , S ∘ L S ) ( τ ) ) = T ⊆ Q ∑ tr ( ( P T ∘ Φ ∘ E S ∘ L S ) ( τ ) ) = tr T ⊆ Q ∑ Π T Φ (( E S ∘ L S ) ( τ )) = tr ( Φ (( E S ∘ L S ) ( τ )) ) = tr ( τ ) = 1. (112)
The third equality uses trace preservation of F T \mathcal{F}_T F T from Lemma C.32 and Eq. (C.58). The fourth uses the trace identity in Eq. (C.52). The fifth uses Eq. (C.13). The last uses trace preservation of Φ \Phi Φ , E S \mathcal{E}_S E S , and L S \mathcal{L}_S L S .
Corollary C.73. Under the assumptions of Proposition C.70, the map ∑ T ⊆ Q D T , S Φ \sum_{T\subseteq\mathrm{Q}}D^\Phi_{T,S} ∑ T ⊆ Q D T , S Φ is a Pauli channel on K Q \mathcal{K}_{\mathrm{Q}} K Q for every S ⊆ Q S\subseteq\mathrm{Q} S ⊆ Q .
Proof. Each summand is a Pauli map with nonnegative coefficients, and Eqs. (C.71) and (A.50) imply that the coefficients of the sum add to one.
Theorem 1. Let Φ : End ( H Q ) → End ( H Q ) \Phi:\operatorname{End}(\mathcal{H}_{\mathrm{Q}})\to\operatorname{End}(\mathcal{H}_{\mathrm{Q}}) Φ : End ( H Q ) → End ( H Q ) be a linear map and let { D T , S Φ } S , T ⊆ Q \{D^\Phi_{T,S}\}_{S,T\subseteq\mathrm{Q}} { D T , S Φ } S , T ⊆ Q be the leakage-transition Pauli-map family associated with Φ \Phi Φ , as in Definition C.67. Then
T C Q ( Φ ) = ∑ S , T ⊆ Q E T ∘ L T ∘ D T , S Φ ∘ C ∘ F S ∘ P S . (113) \mathcal{T}^{\mathrm{Q}}_{\mathcal{C}}(\Phi)=\sum_{S,T\subseteq\mathrm{Q}}\mathcal{E}_T\circ\mathcal{L}_T\circ D^\Phi_{T,S}\circ\mathcal{C}\circ\mathcal{F}_S\circ\mathcal{P}_S. \tag*{(113)} T C Q ( Φ ) = S , T ⊆ Q ∑ E T ∘ L T ∘ D T , S Φ ∘ C ∘ F S ∘ P S . (113)
Proof. Combining Eqs. (C.27) and (C.17) gives T C Q ( Φ ) = Δ ∘ T C Q ( Φ ) ∘ Δ \mathcal{T}^{\mathrm{Q}}_{\mathcal{C}}(\Phi)=\Delta\circ\mathcal{T}^{\mathrm{Q}}_{\mathcal{C}}(\Phi)\circ\Delta T C Q ( Φ ) = Δ ∘ T C Q ( Φ ) ∘ Δ . Next, denote by { ( T C Q ( Φ ) ) T , S } S , T ⊆ Q \{(\mathcal{T}^{\mathrm{Q}}_{\mathcal{C}}(\Phi))_{T,S}\}_{S,T\subseteq\mathrm{Q}} {( T C Q ( Φ ) ) T , S } S , T ⊆ Q the family of leakage-transition branches associated with T C Q ( Φ ) \mathcal{T}^{\mathrm{Q}}_{\mathcal{C}}(\Phi) T C Q ( Φ ) , as in
Definition C.57. Substituting Eq. (C.55) on both sides of T C Q ( Φ ) \mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi) T C Q ( Φ ) gives
T C Q ( Φ ) = ( ∑ T ⊆ Q E T ∘ P T ) ∘ T C Q ( Φ ) ∘ ( ∑ S ⊆ Q E S ∘ P S ) \mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi)=\left(\sum_{T\subseteq Q}\mathcal{E}_{T}\circ\mathcal{P}_{T}\right)\circ\mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi)\circ\left(\sum_{S\subseteq Q}\mathcal{E}_{S}\circ\mathcal{P}_{S}\right) T C Q ( Φ ) = T ⊆ Q ∑ E T ∘ P T ∘ T C Q ( Φ ) ∘ S ⊆ Q ∑ E S ∘ P S = ∑ S , T ⊆ Q E T ∘ ( T C Q ( Φ ) ) T , S ∘ P S . =\sum_{S,T\subseteq Q}\mathcal{E}_{T}\circ\left(\mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi)\right)_{T,S}\circ\mathcal{P}_{S}. = S , T ⊆ Q ∑ E T ∘ ( T C Q ( Φ ) ) T , S ∘ P S .
Combining Eqs. (C.61) and (C.69) gives
( T C Q ( Φ ) ) T , S = L T ∘ D T , S Φ ∘ C ∘ F S . (114) \left(\mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi)\right)_{T,S}=\mathcal{L}_{T}\circ\mathcal{D}^{\Phi}_{T,S}\circ\mathcal{C}\circ\mathcal{F}_{S}. \tag*{(114)} ( T C Q ( Φ ) ) T , S = L T ∘ D T , S Φ ∘ C ∘ F S . (114)
Finally, substituting Eq. (C.75) termwise gives Eq. (C.74).
Leakage-pattern Markov process
Fix a CPTP map Φ : End ( H Q ) → End ( H Q ) \Phi:\operatorname{End}(\mathcal{H}_{Q})\to\operatorname{End}(\mathcal{H}_{Q}) Φ : End ( H Q ) → End ( H Q ) . Denote by { D T , S Φ } S , T ⊆ Q \{\mathcal{D}^{\Phi}_{T,S}\}_{S,T\subseteq Q} { D T , S Φ } S , T ⊆ Q the leakage-transition Pauli-map family associated with Φ \Phi Φ , as in Definition C.67.
Definition C.76 (Leakage-pattern transition weights). For S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q , define the leakage-pattern transition weight p ( T ∣ S ) p(T|S) p ( T ∣ S ) by
p ( T ∣ S ) = wt ( D T , S Φ ) . (115) p(T|S)=\operatorname{wt}\left(\mathcal{D}^{\Phi}_{T,S}\right). \tag*{(115)} p ( T ∣ S ) = wt ( D T , S Φ ) . (115)
The conclusions of Proposition C.70 give p ( T ∣ S ) ≥ 0 p(T|S)\ge0 p ( T ∣ S ) ≥ 0 for all S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q and ∑ T ⊆ Q p ( T ∣ S ) = 1 \sum_{T\subseteq Q}p(T|S)=1 ∑ T ⊆ Q p ( T ∣ S ) = 1 for every S ⊆ Q S\subseteq Q S ⊆ Q . Thus, p p p is a Markov transition kernel on the set of leakage patterns.
Definition C.78 (Conditional leakage-transition maps). For S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q satisfying p ( T ∣ S ) > 0 p(T|S)>0 p ( T ∣ S ) > 0 , define the map C T , S Φ : End ( K Q ) → End ( K Q ) \mathcal{C}^{\Phi}_{T,S}:\operatorname{End}(\mathcal{K}_{Q})\to\operatorname{End}(\mathcal{K}_{Q}) C T , S Φ : End ( K Q ) → End ( K Q ) by
C T , S Φ = 1 p ( T ∣ S ) D T , S Φ . (116) \mathcal{C}^{\Phi}_{T,S}=\frac{1}{p(T|S)}\mathcal{D}^{\Phi}_{T,S}. \tag*{(116)} C T , S Φ = p ( T ∣ S ) 1 D T , S Φ . (116)
Corollary C.80. For every S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q satisfying p ( T ∣ S ) > 0 p(T|S)>0 p ( T ∣ S ) > 0 , C T , S Φ \mathcal{C}^{\Phi}_{T,S} C T , S Φ is a Pauli channel and
T C Q ( Φ ) = ∑ S , T ⊆ Q p ( T ∣ S ) > 0 p ( T ∣ S ) E T ∘ L T ∘ C T , S Φ ∘ C ∘ F S ∘ P S . (117) \mathcal{T}^{\mathrm{Q}}_{\mathrm{C}}(\Phi)=\sum_{\substack{S,T\subseteq Q\\p(T|S)>0}}p(T|S)\mathcal{E}_{T}\circ\mathcal{L}_{T}\circ\mathcal{C}^{\Phi}_{T,S}\circ\mathcal{C}\circ\mathcal{F}_{S}\circ\mathcal{P}_{S}. \tag*{(117)} T C Q ( Φ ) = S , T ⊆ Q p ( T ∣ S ) > 0 ∑ p ( T ∣ S ) E T ∘ L T ∘ C T , S Φ ∘ C ∘ F S ∘ P S . (117)
Proof. Fix S , T ⊆ Q S,T\subseteq Q S , T ⊆ Q , and let { q v } v ∈ V K Q \{q_{v}\}_{v\in V_{\mathcal{K}_{Q}}} { q v } v ∈ V K Q be the unique Pauli-map coefficient family of D T , S Φ \mathcal{D}^{\Phi}_{T,S} D T , S Φ , as in Lemma A.42. Proposition C.70 implies that D T , S Φ \mathcal{D}^{\Phi}_{T,S} D T , S Φ is a subnormalized Pauli map, so Lemma A.55 gives q v ≥ 0 q_{v}\ge0 q v ≥ 0 for every v ∈ V K Q v\in V_{\mathcal{K}_{Q}} v ∈ V K Q . Combining Eqs. (C.77) and (A.50) gives
p ( T ∣ S ) = ∑ v ∈ V K Q q v . p(T|S)=\sum_{v\in V_{\mathcal{K}_{Q}}}q_{v}. p ( T ∣ S ) = v ∈ V K Q ∑ q v .
Suppose that p ( T ∣ S ) > 0 p(T|S)>0 p ( T ∣ S ) > 0 . By Eq. (C.79), the Pauli-map coefficient family of C T , S Φ \mathcal{C}^{\Phi}_{T,S} C T , S Φ is { q v / p ( T ∣ S ) } v ∈ V K Q \{q_{v}/p(T|S)\}_{v\in V_{\mathcal{K}_{Q}}} { q v / p ( T ∣ S ) } v ∈ V K Q . These coefficients lie in [ 0 , 1 ] [0,1] [ 0 , 1 ] and sum to one. Hence, C T , S Φ \mathcal{C}^{\Phi}_{T,S} C T , S Φ is a Pauli channel, by Lemma A.55.
Suppose instead that p ( T ∣ S ) = 0 p(T|S)=0 p ( T ∣ S ) = 0 . Since every q v q_{v} q v is nonnegative and their sum is zero, q v = 0 q_{v}=0 q v = 0 for every v ∈ V K Q v\in V_{\mathcal{K}_{Q}} v ∈ V K Q . Hence, D T , S Φ = 0 \mathcal{D}^{\Phi}_{T,S}=0 D T , S Φ = 0 . Substituting D T , S Φ = p ( T ∣ S ) C T , S Φ \mathcal{D}^{\Phi}_{T,S}=p(T|S)\mathcal{C}^{\Phi}_{T,S} D T , S Φ = p ( T ∣ S ) C T , S Φ for the positive-weight terms and D T , S Φ = 0 \mathcal{D}^{\Phi}_{T,S}=0 D T , S Φ = 0 for the zero-weight terms in Eq. (C.74) gives Eq. (C.81).
For a density operator ρ ∈ End ( H Q ) \rho\in\operatorname{End}(\mathcal{H}_{\mathsf{Q}}) ρ ∈ End ( H Q ) , define the scalars π S \pi_{S} π S and π T ′ \pi'_{T} π T ′ by
π S = tr ( P S ( ρ ) ) , and π T ′ = tr ( P T ( T C Q ( Φ ) ( ρ ) ) ) , (118) \pi_{S}=\operatorname{tr}(\mathcal{P}_{S}(\rho)), \qquad\text{and} \qquad\pi'_{T}=\operatorname{tr}\left(\mathcal{P}_{T}(\mathcal{T}^{\mathsf{Q}}_{\mathcal{C}}(\Phi)(\rho))\right), \tag*{(118)} π S = tr ( P S ( ρ )) , and π T ′ = tr ( P T ( T C Q ( Φ ) ( ρ )) ) , (118)
for all S , T ⊆ Q S,T\subseteq\mathsf{Q} S , T ⊆ Q . The families { π S } S ⊆ Q \{\pi_{S}\}_{S\subseteq\mathsf{Q}} { π S } S ⊆ Q and { π T ′ } T ⊆ Q \{\pi'_{T}\}_{T\subseteq\mathsf{Q}} { π T ′ } T ⊆ Q can be understood as the probability distributions over leakage patterns of the incoming state ρ \rho ρ and the outgoing state T C Q ( Φ ) ( ρ ) \mathcal{T}^{\mathsf{Q}}_{\mathcal{C}}(\Phi)(\rho) T C Q ( Φ ) ( ρ ) , respectively.
Corollary C.83. The families { π S } S ⊆ Q \{\pi_{S}\}_{S\subseteq\mathsf{Q}} { π S } S ⊆ Q and { π T ′ } T ⊆ Q \{\pi'_{T}\}_{T\subseteq\mathsf{Q}} { π T ′ } T ⊆ Q are probability distributions on the set of leakage patterns, and
π T ′ = ∑ S ⊆ Q p ( T ∣ S ) π S , (119) \pi'_{T}=\sum_{S\subseteq\mathsf{Q}}p(T|S)\pi_{S}, \tag*{(119)} π T ′ = S ⊆ Q ∑ p ( T ∣ S ) π S , (119) for all T ⊆ Q T\subseteq\mathsf{Q} T ⊆ Q .
Proof. Since ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 , Eq. (C.49) gives π S ≥ 0 \pi_{S}\geq0 π S ≥ 0 for every S ⊆ Q S\subseteq\mathsf{Q} S ⊆ Q . Combining Eqs. (C.52) and (C.13) gives ∑ S ⊆ Q π S = tr ( ρ ) = 1 \sum_{S\subseteq\mathsf{Q}}\pi_{S}=\operatorname{tr}(\rho)=1 ∑ S ⊆ Q π S = tr ( ρ ) = 1 . Thus, { π S } S ⊆ Q \{\pi_{S}\}_{S\subseteq\mathsf{Q}} { π S } S ⊆ Q is a probability distribution. Next, fix T ⊆ Q T\subseteq\mathsf{Q} T ⊆ Q . Applying P T \mathcal{P}_{T} P T to Eq. (C.81), using Eq. (C.56) termwise, and taking the trace gives
π T ′ = ∑ S ⊆ Q p ( T ∣ S ) > 0 p ( T ∣ S ) tr [ ( L T ∘ C T , S Φ ∘ C ∘ F S ) ( P S ( ρ ) ) ] \pi'_{T}=\sum_{\substack{S\subseteq\mathsf{Q}\\p(T|S)>0}}p(T|S)\operatorname{tr}\left[\left(\mathcal{L}_{T}\circ\mathcal{C}^{\Phi}_{T,S}\circ\mathcal{C}\circ\mathcal{F}_{S}\right)\left(\mathcal{P}_{S}(\rho)\right)\right] π T ′ = S ⊆ Q p ( T ∣ S ) > 0 ∑ p ( T ∣ S ) tr [ ( L T ∘ C T , S Φ ∘ C ∘ F S ) ( P S ( ρ ) ) ] = ∑ S ⊆ Q p ( T ∣ S ) > 0 p ( T ∣ S ) π S = ∑ S ⊆ Q p ( T ∣ S ) π S . (120) =\sum_{\substack{S\subseteq\mathsf{Q}\\p(T|S)>0}}p(T|S)\pi_{S}=\sum_{S\subseteq\mathsf{Q}}p(T|S)\pi_{S}. \tag*{(120)} = S ⊆ Q p ( T ∣ S ) > 0 ∑ p ( T ∣ S ) π S = S ⊆ Q ∑ p ( T ∣ S ) π S . (120) The second equality uses trace preservation of F S \mathcal{F}_{S} F S and L T \mathcal{L}_{T} L T from Lemma C.32, trace preservation of C \mathcal{C} C , and trace preservation of C T , S Φ \mathcal{C}^{\Phi}_{T,S} C T , S Φ from Corollary C.80. The last equality inserts the terms satisfying p ( T ∣ S ) = 0 p(T|S)=0 p ( T ∣ S ) = 0 .
Since p ( T ∣ S ) ≥ 0 p(T|S)\geq0 p ( T ∣ S ) ≥ 0 , Eq. (C.84) gives π T ′ ≥ 0 \pi'_{T}\geq0 π T ′ ≥ 0 for every T ⊆ Q T\subseteq\mathsf{Q} T ⊆ Q . Summing Eq. (C.84) over T T T gives ∑ T ⊆ Q π T ′ = ∑ S ⊆ Q π S ∑ T ⊆ Q p ( T ∣ S ) = 1 \sum_{T\subseteq\mathsf{Q}}\pi'_{T}=\sum_{S\subseteq\mathsf{Q}}\pi_{S}\sum_{T\subseteq\mathsf{Q}}p(T|S)=1 ∑ T ⊆ Q π T ′ = ∑ S ⊆ Q π S ∑ T ⊆ Q p ( T ∣ S ) = 1 .
Model of a QEC cycle on leakage-extended qubits
Joint registers and Cliffords
Definition D.1 (Joint register of qubit registers). Let N 1 \mathsf{N}_{1} N 1 and N 2 \mathsf{N}_{2} N 2 be qubit registers with disjoint label sets. For j = 1 , 2 j=1,2 j = 1 , 2 and i ∈ N j i\in\mathsf{N}_{j} i ∈ N j , denote the qubit Hilbert space at label i i i and its computational basis by q i \mathfrak{q}_{i} q i and { ∣ 0 i ⟩ , ∣ 1 i ⟩ } \{\lvert0_{i}\rangle,\lvert1_{i}\rangle\} {∣ 0 i ⟩ , ∣ 1 i ⟩} , respectively. Define their joint register N \mathsf{N} N to be the qubit register with label set
N = N 1 ⊔ N 2 \mathsf{N}=\mathsf{N}_{1}\sqcup\mathsf{N}_{2} N = N 1 ⊔ N 2
and family of qubit Hilbert spaces { q i } i ∈ N \{\mathfrak{q}_{i}\}_{i\in\mathsf{N}} { q i } i ∈ N , equipped with these computational bases.
Definition D.2 (Joint register of leakage-extended qubit registers). Let Q 1 \mathsf{Q}_{1} Q 1 and Q 2 \mathsf{Q}_{2} Q 2 be leakage-extended qubit registers with disjoint label sets. For j = 1 , 2 j=1,2 j = 1 , 2 and i ∈ Q j i\in\mathsf{Q}_{j} i ∈ Q j , denote the leakage-extended qubit space at label i i i by h i \mathfrak{h}_{i} h i , its specified decomposition by h i = c i ⊕ l i \mathfrak{h}_{i}=\mathfrak{c}_{i}\oplus\mathfrak{l}_{i} h i = c i ⊕ l i , and the computational basis of c i c_i c i by { ∣ 0 i ⟩ , ∣ 1 i ⟩ } \{\lvert0_i\rangle,\lvert1_i\rangle\} {∣ 0 i ⟩ , ∣ 1 i ⟩} . Define their joint register Q \mathbf{Q} Q to be the leakage-extended qubit register with label set
Q = Q 1 ⊔ Q 2 \mathbf{Q}=\mathbf{Q}_1\sqcup\mathbf{Q}_2 Q = Q 1 ⊔ Q 2
and family of leakage-extended qubit spaces { H i } i ∈ Q \{\mathcal{H}_i\}_{i\in\mathbf{Q}} { H i } i ∈ Q , equipped with these decompositions and computational bases.
We implicitly use the canonical unitary identifications obtained by regrouping tensor products with the same labeled systems (tensor factors). For both constructions, the Hilbert space of the joint register is identified with the tensor product of the Hilbert spaces of the two constituent registers. If N \mathbf{N} N is the joint register of the qubit registers N 1 \mathbf{N}_1 N 1 and N 2 \mathbf{N}_2 N 2 , we identify its Pauli labels with pairs of constituent Pauli labels by restriction to the two label sets. Under these identifications,
V N = V N 1 × V N 2 , (121) V_{\mathbf{N}}=V_{\mathbf{N}_1}\times V_{\mathbf{N}_2}, \tag*{(121)} V N = V N 1 × V N 2 , (121)
[ ( a 1 , a 2 ) , ( b 1 , b 2 ) ] = [ a 1 , b 1 ] + [ a 2 , b 2 ] , (122) [(a_1,a_2),(b_1,b_2)]=[a_1,b_1]+[a_2,b_2], \tag*{(122)} [( a 1 , a 2 ) , ( b 1 , b 2 )] = [ a 1 , b 1 ] + [ a 2 , b 2 ] , (122)
W N ( ( a 1 , a 2 ) ) = W N 1 ( a 1 ) ⊗ W N 2 ( a 2 ) , (123) W^{\mathbf{N}}((a_1,a_2))=W^{\mathbf{N}_1}(a_1)\otimes W^{\mathbf{N}_2}(a_2), \tag*{(123)} W N (( a 1 , a 2 )) = W N 1 ( a 1 ) ⊗ W N 2 ( a 2 ) , (123)
for a j , b j ∈ V N j a_j,b_j\in V_{\mathbf{N}_j} a j , b j ∈ V N j , j = 1 , 2 j=1,2 j = 1 , 2 , with the symplectic-form sum taken in Z 2 \mathbb{Z}_2 Z 2 .
The family { W N 1 ( a 1 ) ⊗ W N 2 ( a 2 ) } ( a 1 , a 2 ) ∈ V N 1 × V N 2 \{W^{\mathbf{N}_1}(a_1)\otimes W^{\mathbf{N}_2}(a_2)\}_{(a_1,a_2)\in V_{\mathbf{N}_1}\times V_{\mathbf{N}_2}} { W N 1 ( a 1 ) ⊗ W N 2 ( a 2 ) } ( a 1 , a 2 ) ∈ V N 1 × V N 2 is the product Pauli family of N 1 \mathbf{N}_1 N 1 and N 2 \mathbf{N}_2 N 2 . If Q \mathbf{Q} Q is the joint register of leakage-extended qubit registers Q 1 \mathbf{Q}_1 Q 1 and Q 2 \mathbf{Q}_2 Q 2 , then K Q \mathcal{K}_{\mathbf{Q}} K Q is the joint register of the computational qubit registers K Q 1 \mathcal{K}_{\mathbf{Q}_1} K Q 1 and K Q 2 \mathcal{K}_{\mathbf{Q}_2} K Q 2 .
Both constructions and the corresponding identifications extend immediately to any finite nonempty family of registers of the same type with pairwise disjoint label sets. The product Pauli family likewise consists of tensor products of one Pauli operator from each constituent qubit register. The preceding definitions and results formulated for an arbitrary qubit register or leakage-extended qubit register therefore apply to joint registers through these identifications. Whenever a finite family of registers is treated jointly in the remainder of this discussion, its constituent label sets are assumed to be pairwise disjoint.
Definition D.4 (Clifford isomorphism). Let M \mathbf{M} M and N \mathbf{N} N be qubit registers. A unitary isomorphism
C : K M → K N (124) C:\mathcal{K}_{\mathbf{M}}\to\mathcal{K}_{\mathbf{N}} \tag*{(124)} C : K M → K N (124) is a Clifford isomorphism if there exist a symplectic isomorphism F C : V M → V N F_C:V_{\mathbf{M}}\to V_{\mathbf{N}} F C : V M → V N and a map η C : V M → { + 1 , − 1 } \eta_C:V_{\mathbf{M}}\to\{+1,-1\} η C : V M → { + 1 , − 1 } such that
C ( W M ( a ) ) = C W M ( a ) C † = η C ( a ) W N ( F C a ) , (125) C(W^{\mathbf{M}}(a))=CW^{\mathbf{M}}(a)C^\dagger=\eta_C(a)W^{\mathbf{N}}(F_Ca), \tag*{(125)} C ( W M ( a )) = C W M ( a ) C † = η C ( a ) W N ( F C a ) , (125) for all $a\in V_{\mathbf{M}}.
When M = N \mathbf{M}=\mathbf{N} M = N , Definition D.4 agrees with Definition B.1. The uniqueness argument following Definition B.1 also applies to F C F_C F C and η C \eta_C η C . Equation (D.6) shows that adjoints, compositions of composable Clifford isomorphisms, and tensor products of Clifford isomorphisms are Clifford. Their induced symplectic isomorphisms are, respectively, F C − 1 F_C^{-1} F C − 1 , F D ∘ F C F_D\circ F_C F D ∘ F C for D ∘ C D\circ C D ∘ C , and the product of the constituent symplectic isomorphisms.
QEC-cycle construction
Fix a stabilizer code on n n n physical qubits that encodes k k k logical qubits. Let D, A, and O be leakage-extended qubit registers, as in Definition C.3, where ∣ D ∣ = n |\mathrm{D}|=n ∣ D ∣ = n . Denote by T the joint register of D, A, and O, and also the set of joint labels. Let K D \mathrm{K}_{\mathrm{D}} K D , K A \mathrm{K}_{\mathrm{A}} K A , K O \mathrm{K}_{\mathrm{O}} K O , and K T \mathrm{K}_{\mathrm{T}} K T be the computational qubit registers associated with D, A, O, and T, respectively, as in Definition C.23. Then K T \mathrm{K}_{\mathrm{T}} K T is the joint register of K D \mathrm{K}_{\mathrm{D}} K D , K A \mathrm{K}_{\mathrm{A}} K A , and K O \mathrm{K}_{\mathrm{O}} K O .
Let L and S be qubit registers (and the associated label sets) as in Definition A.2, where ∣ L ∣ = k |\mathrm{L}|=k ∣ L ∣ = k and ∣ S ∣ = n − k |\mathrm{S}|=n-k ∣ S ∣ = n − k . Choose a Clifford isomorphism
C F : K D → K L ⊗ K S . C_{\mathrm{F}}:\mathcal{K}_{\mathrm{D}}\to\mathcal{K}_{\mathrm{L}}\otimes\mathcal{K}_{\mathrm{S}}. C F : K D → K L ⊗ K S .
A QEC cycle on D is modeled by a CPTP linear map End ( H D ) → End ( H D ) \operatorname{End}(\mathcal{H}_{\mathrm{D}})\to\operatorname{End}(\mathcal{H}_{\mathrm{D}}) End ( H D ) → End ( H D ) . Denote by ∣ 0 ⟩ A ∈ K A |0\rangle_{\mathrm{A}}\in\mathcal{K}_{\mathrm{A}} ∣0 ⟩ A ∈ K A and ∣ 0 ⟩ O ∈ K O |0\rangle_{\mathrm{O}}\in\mathcal{K}_{\mathrm{O}} ∣0 ⟩ O ∈ K O the all-zero product computational-basis vectors. Define the initialization map I A O : End ( H D ) → End ( H T ) \mathcal{I}_{\mathrm{AO}}:\operatorname{End}(\mathcal{H}_{\mathrm{D}})\to\operatorname{End}(\mathcal{H}_{\mathrm{T}}) I AO : End ( H D ) → End ( H T ) as
I A O ( Z ) : = Z ⊗ ∣ 0 ⟩ ⟨ 0 ∣ A ⊗ ∣ 0 ⟩ ⟨ 0 ∣ O , (126) \mathcal{I}_{\mathrm{AO}}(Z):=Z\otimes|0\rangle\langle0|_{\mathrm{A}}\otimes|0\rangle\langle0|_{\mathrm{O}}, \tag*{(126)} I AO ( Z ) := Z ⊗ ∣0 ⟩ ⟨ 0 ∣ A ⊗ ∣0 ⟩ ⟨ 0 ∣ O , (126)
for all Z ∈ End ( H D ) Z\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) Z ∈ End ( H D ) . The tensor factors in Eq. (D.7) are identified by the canonical labeled tensor-product identification. The restriction of I A O \mathcal{I}_{\mathrm{AO}} I AO to End ( K D ) \operatorname{End}(\mathcal{K}_{\mathrm{D}}) End ( K D ) has range in End ( K T ) \operatorname{End}(\mathcal{K}_{\mathrm{T}}) End ( K T ) and is denoted by the same symbol. Both maps are CPTP.
For s ∈ Z 2 S s\in\mathbb{Z}_2^{\mathrm{S}} s ∈ Z 2 S , denote by ∣ s ⟩ S |s\rangle_{\mathrm{S}} ∣ s ⟩ S the corresponding product computational-basis vector. Let C f r C_{\mathrm{fr}} C fr be a unitary on K L ⊗ K S ⊗ K A ⊗ K O \mathcal{K}_{\mathrm{L}}\otimes\mathcal{K}_{\mathrm{S}}\otimes\mathcal{K}_{\mathrm{A}}\otimes\mathcal{K}_{\mathrm{O}} K L ⊗ K S ⊗ K A ⊗ K O that is Clifford with respect to the product Pauli family of L, S, K A \mathcal{K}_{\mathrm{A}} K A , and K O \mathcal{K}_{\mathrm{O}} K O . Further assume that there exists an injective map
E : Z 2 S → Z 2 O , (127) E:\mathbb{Z}_2^{\mathrm{S}}\to\mathbb{Z}_2^{\mathrm{O}}, \tag*{(127)} E : Z 2 S → Z 2 O , (127)
and unit vectors ∣ α ( s ) ⟩ A ∈ K A |\alpha(s)\rangle_{\mathrm{A}}\in\mathcal{K}_{\mathrm{A}} ∣ α ( s ) ⟩ A ∈ K A (for s ∈ Z 2 S s\in\mathbb{Z}_2^{\mathrm{S}} s ∈ Z 2 S ) such that, for every s ∈ Z 2 S s\in\mathbb{Z}_2^{\mathrm{S}} s ∈ Z 2 S and ∣ ψ ⟩ L ∈ K L |\psi\rangle_{\mathrm{L}}\in\mathcal{K}_{\mathrm{L}} ∣ ψ ⟩ L ∈ K L ,
C f r ( ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ 0 ⟩ A ⊗ ∣ 0 ⟩ O ) = ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ α ( s ) ⟩ A ⊗ ∣ E ( s ) ⟩ O . (128) C_{\mathrm{fr}}\left(|\psi\rangle_{\mathrm{L}}\otimes|s\rangle_{\mathrm{S}}\otimes|0\rangle_{\mathrm{A}}\otimes|0\rangle_{\mathrm{O}}\right)
=|\psi\rangle_{\mathrm{L}}\otimes|s\rangle_{\mathrm{S}}\otimes|\alpha(s)\rangle_{\mathrm{A}}\otimes|E(s)\rangle_{\mathrm{O}}. \tag*{(128)} C fr ( ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣0 ⟩ A ⊗ ∣0 ⟩ O ) = ∣ ψ ⟩ L ⊗ ∣ s ⟩ S ⊗ ∣ α ( s ) ⟩ A ⊗ ∣ E ( s ) ⟩ O . (128)
Define the syndrome-readout unitary C s y n d C_{\mathrm{synd}} C synd on K T \mathcal{K}_{\mathrm{T}} K T according to
C s y n d : = ( C F † ⊗ 1 K A ⊗ K O ) C f r ( C F ⊗ 1 K A ⊗ K O ) . (129) C_{\mathrm{synd}}:=\left(C_{\mathrm{F}}^\dagger\otimes\mathbb{1}_{\mathcal{K}_{\mathrm{A}}\otimes\mathcal{K}_{\mathrm{O}}}\right)C_{\mathrm{fr}}\left(C_{\mathrm{F}}\otimes\mathbb{1}_{\mathcal{K}_{\mathrm{A}}\otimes\mathcal{K}_{\mathrm{O}}}\right). \tag*{(129)} C synd := ( C F † ⊗ 1 K A ⊗ K O ) C fr ( C F ⊗ 1 K A ⊗ K O ) . (129)
The Clifford properties of C f r C_{\mathrm{fr}} C fr and C F C_{\mathrm{F}} C F imply that C s y n d C_{\mathrm{synd}} C synd is Clifford with respect to the Pauli family of K T \mathcal{K}_{\mathrm{T}} K T .
Denote by N : End ( H T ) → End ( H T ) \mathcal{N}:\operatorname{End}(\mathcal{H}_{\mathrm{T}})\to\operatorname{End}(\mathcal{H}_{\mathrm{T}}) N : End ( H T ) → End ( H T ) the CPTP map representing a noisy syndrome-readout operation, whose ideal (noiseless) implementation corresponds to the adjoint action of C s y n d C_{\mathrm{synd}} C synd . The unitaries C s y n d C_{\mathrm{synd}} C synd and C f r C_{\mathrm{fr}} C fr are specified only on the computational spaces K \mathcal{K} K , whereas N \mathcal{N} N is defined on the full space H T \mathcal{H}_{\mathrm{T}} H T . No further restriction on the action of N \mathcal{N} N is imposed at this point.
Define Θ : = { 0 , 1 , L } O \Theta:=\{0,1,\mathrm{L}\}^{\mathrm{O}} Θ := { 0 , 1 , L } O . For each i ∈ O i\in\mathrm{O} i ∈ O , define the orthogonal projectors on H i = C i ⊕ L i \mathcal{H}_i=\mathcal{C}_i\oplus\mathcal{L}_i H i = C i ⊕ L i by
P i 0 : = ∣ 0 i ⟩ ⟨ 0 i ∣ ⊕ 0 L i , P i 1 : = ∣ 1 i ⟩ ⟨ 1 i ∣ ⊕ 0 L i , P i L : = P i ℓ . (130) P_i^0:=|0_i\rangle\langle0_i|\oplus0_{\mathcal{L}_i},\qquad P_i^1:=|1_i\rangle\langle1_i|\oplus0_{\mathcal{L}_i},\qquad P_i^{\mathrm{L}}:=P_i^{\ell}. \tag*{(130)} P i 0 := ∣ 0 i ⟩ ⟨ 0 i ∣ ⊕ 0 L i , P i 1 := ∣ 1 i ⟩ ⟨ 1 i ∣ ⊕ 0 L i , P i L := P i ℓ . (130)
For θ ∈ Θ \theta\in\Theta θ ∈ Θ , define the outcome projector P θ O ∈ End ( H O ) P_{\theta}^{\mathrm{O}} \in\operatorname{End}(\mathcal{H}_{\mathrm{O}}) P θ O ∈ End ( H O ) by
P θ O : = ⨂ i ∈ O P i θ i . (131) P_{\theta}^{\mathrm{O}} := \bigotimes_{i \in\mathrm{O}} P_{i}^{\theta_i}. \tag*{(131)} P θ O := i ∈ O ⨂ P i θ i . (131)
Equations (C.9) and (D.11) give
P θ O P θ ′ O = δ θ , θ ′ P θ O , ∑ θ ∈ Θ P θ O = 1 H O . (132) P_{\theta}^{\mathrm{O}} P_{\theta'}^{\mathrm{O}} = \delta_{\theta,\theta'} P_{\theta}^{\mathrm{O}}, \qquad\sum_{\theta\in\Theta} P_{\theta}^{\mathrm{O}} = 1_{\mathcal{H}_{\mathrm{O}}}. \tag*{(132)} P θ O P θ ′ O = δ θ , θ ′ P θ O , θ ∈ Θ ∑ P θ O = 1 H O . (132)
for all θ , θ ′ ∈ Θ \theta, \theta' \in\Theta θ , θ ′ ∈ Θ . Define the outcome-map family { M θ } θ ∈ Θ \{\mathcal{M}_{\theta}\}_{\theta\in\Theta} { M θ } θ ∈ Θ , where M θ : End ( H T ) → End ( H D ) \mathcal{M}_{\theta} : \operatorname{End}(\mathcal{H}_{\mathrm{T}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{D}}) M θ : End ( H T ) → End ( H D ) is given by
M θ ( Z ) : = tr A O [ ( 1 H D ⊗ H A ⊗ P θ O ) Z ( 1 H D ⊗ H A ⊗ P θ O ) ] , (133) \mathcal{M}_{\theta}(Z) := \operatorname{tr}_{\mathrm{AO}}\left[(1_{\mathcal{H}_{\mathrm{D}} \otimes\mathcal{H}_{\mathrm{A}}} \otimes P_{\theta}^{\mathrm{O}}) Z (1_{\mathcal{H}_{\mathrm{D}} \otimes\mathcal{H}_{\mathrm{A}}} \otimes P_{\theta}^{\mathrm{O}})\right], \tag*{(133)} M θ ( Z ) := tr AO [ ( 1 H D ⊗ H A ⊗ P θ O ) Z ( 1 H D ⊗ H A ⊗ P θ O ) ] , (133)
for all Z ∈ End ( H T ) Z \in\operatorname{End}(\mathcal{H}_{\mathrm{T}}) Z ∈ End ( H T ) . Each M θ \mathcal{M}_{\theta} M θ is completely positive and trace nonincreasing, and Eq. (D.13) gives
∑ θ ∈ Θ M θ ( ⋅ ) = tr A O ( ⋅ ) . (134) \sum_{\theta\in\Theta} \mathcal{M}_{\theta}(\mathord{\cdot}) = \operatorname{tr}_{\mathrm{AO}}(\mathord{\cdot}). \tag*{(134)} θ ∈ Θ ∑ M θ ( ⋅ ) = tr AO ( ⋅ ) . (134)
Let { W D ( v ) } v ∈ V K D \{W^{\mathrm{D}}(v)\}_{v \in\mathsf{V}_{K_{\mathrm{D}}}} { W D ( v ) } v ∈ V K D be the leakage-extended Pauli family associated with D \mathrm{D} D , as in Definition C.24. Fix a Pauli-correction function
D : Θ → V K D . (135) D : \Theta\to\mathsf{V}_{K_{\mathrm{D}}}. \tag*{(135)} D : Θ → V K D . (135)
Fix a return function
K : Θ → 2 D , K ( θ ) ⊆ D , (136) K : \Theta\to2^{\mathrm{D}}, \qquad K(\theta) \subseteq\mathrm{D}, \tag*{(136)} K : Θ → 2 D , K ( θ ) ⊆ D , (136)
for all θ ∈ Θ \theta\in\Theta θ ∈ Θ . For any J ⊆ D J \subseteq\mathrm{D} J ⊆ D , define c J : = ⨂ i ∈ J c i c_J := \bigotimes_{i \in J} c_i c J := ⨂ i ∈ J c i , and define the return map R J : End ( H D ) → End ( H D ) \mathcal{R}_J : \operatorname{End}(\mathcal{H}_{\mathrm{D}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{D}}) R J : End ( H D ) → End ( H D ) by
R J ( Z ) : = 1 2 ∣ J ∣ 1 c J ⊗ tr J ( Z ) , (137) \mathcal{R}_J(Z) := \frac{1}{2^{|J|}}1_{c_J} \otimes\operatorname{tr}_J(Z), \tag*{(137)} R J ( Z ) := 2 ∣ J ∣ 1 1 c J ⊗ tr J ( Z ) , (137)
for all Z ∈ End ( H D ) Z \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) Z ∈ End ( H D ) , where 1 c J 1_{c_J} 1 c J is extended by zero to all of ⨂ i ∈ J H i \bigotimes_{i \in J}\mathcal{H}_i ⨂ i ∈ J H i , tr J \operatorname{tr}_J tr J is the partial trace over the physical systems with labels in J J J , and the output tensor factors are identified by the canonical labeled tensor-product identification. The map R J \mathcal{R}_J R J is CPTP.
For our purposes—with the stabilizer code, registers, computational bases, and Clifford frame fixed as above—the specification of a noisy QEC cycle consists of the syndrome-readout unitary C s y n d C_{\mathrm{synd}} C synd , the syndrome-readout map N \mathcal{N} N , the Pauli-correction function D D D , and the return function K K K .
Define the associated complete-cycle map Φ : End ( H D ) → End ( H D ) \Phi: \operatorname{End}(\mathcal{H}_{\mathrm{D}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{D}}) Φ : End ( H D ) → End ( H D ) by
Φ : = ∑ θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ N ∘ I A O . (138) \Phi:= \sum_{\theta\in\Theta} \mathcal{R}_{K(\theta)} \circ W^{\mathrm{D}}(D(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{N} \circ\mathcal{I}_{\mathrm{AO}}. \tag*{(138)} Φ := θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ N ∘ I AO . (138)
The map Φ \Phi Φ is completely positive. For every Z ∈ End ( H D ) Z \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) Z ∈ End ( H D ) , Eq. (D.15) and trace preservation of the remaining constituent maps give
tr ( Φ ( Z ) ) = ∑ θ ∈ Θ tr ( M θ ( N ( I A O ( Z ) ) ) ) = tr ( Z ) . (139) \operatorname{tr}(\Phi(Z)) = \sum_{\theta\in\Theta} \operatorname{tr}\left(\mathcal{M}_{\theta}\left(\mathcal{N}(\mathcal{I}_{\mathrm{AO}}(Z))\right)\right) = \operatorname{tr}(Z). \tag*{(139)} tr ( Φ ( Z )) = θ ∈ Θ ∑ tr ( M θ ( N ( I AO ( Z )) ) ) = tr ( Z ) . (139)
Thus Φ \Phi Φ is a completely positive trace-preserving map on End ( H D ) \operatorname{End}(\mathcal{H}_{D}) End ( H D ) .
Leakage-subspace structure of the twirled complete cycle
Let V K T V_{K_{\mathrm{T}}} V K T be the Pauli-label space of K T K_{\mathrm{T}} K T as in Definition A.6. Let F s y n d : V K T → V K T F_{\mathrm{synd}}: V_{K_{\mathrm{T}}} \to V_{K_{\mathrm{T}}} F synd : V K T → V K T be the symplectic automorphism associated with C s y n d C_{\mathrm{synd}} C synd , as in Definition B.1. Denote by T C s y n d T \mathcal{T}^{\mathrm{T}}_{C_{\mathrm{synd}}} T C synd T the leakage-dephased Pauli twirl, as in Definition C.26, specialized to T \mathrm{T} T , C s y n d C_{\mathrm{synd}} C synd , and F s y n d F_{\mathrm{synd}} F synd . Define the twirled syndrome-readout map N t w : End ( H T ) → End ( H T ) \mathcal{N}^{\mathrm{tw}}: \operatorname{End}(\mathcal{H}_{\mathrm{T}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{T}}) N tw : End ( H T ) → End ( H T ) by
N t w : = T C s y n d T ( N ) . (140) \mathcal{N}^{\mathrm{tw}} := \mathcal{T}^{\mathrm{T}}_{C_{\mathrm{synd}}}(\mathcal{N}). \tag*{(140)} N tw := T C synd T ( N ) . (140)
Because N \mathcal{N} N is CPTP, Eq. (C.27) then implies that N t w \mathcal{N}^{\mathrm{tw}} N tw is also CPTP.
Next, define the twirled complete-cycle map Φ t w : End ( H D ) → End ( H D ) \Phi^{\mathrm{tw}}: \operatorname{End}(\mathcal{H}_{D}) \to\operatorname{End}(\mathcal{H}_{D}) Φ tw : End ( H D ) → End ( H D ) by
Φ t w : = ∑ θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ N t w ∘ I A O . (141) \Phi^{\mathrm{tw}} := \sum_{\theta\in\Theta} \mathcal{R}_{K(\theta)} \circ\mathcal{W}^{D}(D(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{N}^{\mathrm{tw}} \circ\mathcal{I}_{\mathrm{AO}}. \tag*{(141)} Φ tw := θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ N tw ∘ I AO . (141)
The map Φ t w \Phi^{\mathrm{tw}} Φ tw is completely positive. It follows from Eq. (D.15)—together with trace preservation of I A O \mathcal{I}_{\mathrm{AO}} I AO , N t w \mathcal{N}^{\mathrm{tw}} N tw , tr A O \operatorname{tr}_{\mathrm{AO}} tr AO , W D ( D ( θ ) ) \mathcal{W}^{D}(D(\theta)) W D ( D ( θ )) , and R K ( θ ) \mathcal{R}_{K(\theta)} R K ( θ ) for every θ ∈ Θ \theta\in\Theta θ ∈ Θ —that Φ t w \Phi^{\mathrm{tw}} Φ tw is trace preserving.
For each S ⊆ T S \subseteq\mathrm{T} S ⊆ T , let P S T \mathcal{P}^{\mathrm{T}}_{S} P S T , E S T \mathcal{E}^{\mathrm{T}}_{S} E S T , F S T \mathcal{F}^{\mathrm{T}}_{S} F S T , and L S T \mathcal{L}^{\mathrm{T}}_{S} L S T denote, respectively, the leakage-subspace projection map, leakage-subspace extension map, filling map, and leakage map (as in Definitions C.48, C.50, C.28, and C.30, respectively), specialized to T \mathrm{T} T . Denote by T C s y n d K T \mathcal{T}^{K_{\mathrm{T}}}_{C_{\mathrm{synd}}} T C synd K T the Pauli twirl with respect to C s y n d C_{\mathrm{synd}} C synd from Definition B.3, specialized to K T K_{\mathrm{T}} K T , C s y n d C_{\mathrm{synd}} C synd , and F s y n d F_{\mathrm{synd}} F synd .
Using the Pauli twirl with respect to C s y n d C_{\mathrm{synd}} C synd (whose noisy implementation is given by N \mathcal{N} N ), we define, for every U , T ⊆ T U,T \subseteq\mathrm{T} U , T ⊆ T , the superoperator
D T , U N : = T C s y n d K T ( F T T ∘ P T T ∘ N ∘ E U T ∘ L U T ) ∘ C s y n d − 1 . (142) \mathcal{D}^{\mathcal{N}}_{T,U} := \mathcal{T}^{K_{\mathrm{T}}}_{C_{\mathrm{synd}}} \left(\mathcal{F}^{\mathrm{T}}_{T} \circ\mathcal{P}^{\mathrm{T}}_{T} \circ\mathcal{N} \circ\mathcal{E}^{\mathrm{T}}_{U} \circ\mathcal{L}^{\mathrm{T}}_{U}\right) \circ C_{\mathrm{synd}}^{-1}. \tag*{(142)} D T , U N := T C synd K T ( F T T ∘ P T T ∘ N ∘ E U T ∘ L U T ) ∘ C synd − 1 . (142)
The family { D T , U N } U , T ⊆ T \{\mathcal{D}^{\mathcal{N}}_{T,U}\}_{U,T\subseteq\mathrm{T}} { D T , U N } U , T ⊆ T is the leakage-transition Pauli-map family associated with N \mathcal{N} N , as in Definition C.67, specialized to T \mathrm{T} T , C s y n d C_{\mathrm{synd}} C synd , and F s y n d F_{\mathrm{synd}} F synd . Proposition C.70 then implies that every D T , U N \mathcal{D}^{\mathcal{N}}_{T,U} D T , U N is a subnormalized Pauli map. Combining (140) with Theorem 1 leads to
N t w = ∑ U , T ⊆ T E T T ∘ L T T ∘ D T , U N ∘ C s y n d ∘ F U T ∘ P U T . (143) \mathcal{N}^{\mathrm{tw}} = \sum_{U,T\subseteq\mathrm{T}} \mathcal{E}^{\mathrm{T}}_{T} \circ\mathcal{L}^{\mathrm{T}}_{T} \circ\mathcal{D}^{\mathcal{N}}_{T,U} \circ C_{\mathrm{synd}} \circ\mathcal{F}^{\mathrm{T}}_{U} \circ\mathcal{P}^{\mathrm{T}}_{U}. \tag*{(143)} N tw = U , T ⊆ T ∑ E T T ∘ L T T ∘ D T , U N ∘ C synd ∘ F U T ∘ P U T . (143)
for the twirled version of the channel N \mathcal{N} N that captures noisy syndrome readout.
For each R ⊆ D R \subseteq\mathrm{D} R ⊆ D , denote by P R D \mathcal{P}^{\mathrm{D}}_{R} P R D and F R D \mathcal{F}^{\mathrm{D}}_{R} F R D , respectively, the leakage-subspace projection map and filling map, as in Definitions C.48 and C.28, specialized to D \mathrm{D} D . For each R ⊆ D R \subseteq\mathrm{D} R ⊆ D , define U R ⊆ T U_{R} \subseteq\mathrm{T} U R ⊆ T to be the image of R R R under the canonical inclusion D ↪ T \mathrm{D} \hookrightarrow\mathrm{T} D ↪ T .
Proposition D.23. The twirled complete-cycle map Φ t w \Phi^{\mathrm{tw}} Φ tw satisfies
Φ t w = ∑ R ⊆ D T ⊆ T ∑ θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R N ∘ C s y n d ∘ I A O ∘ F R D ∘ P R D . (144) \Phi^{\mathrm{tw}} =
\sum_{\substack{R\subseteq\mathrm{D}\\T\subseteq\mathrm{T}}}
\sum_{\theta\in\Theta}
\mathcal{R}_{K(\theta)} \circ\mathcal{W}^{\mathrm{D}}(D(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{E}^{\mathrm{T}}_{T} \circ\mathcal{L}^{\mathrm{T}}_{T} \circ\mathcal{D}^{\mathcal{N}}_{T,U_{R}} \circ C_{\mathrm{synd}} \circ\mathcal{I}_{\mathrm{AO}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}. \tag*{(144)} Φ tw = R ⊆ D T ⊆ T ∑ θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R N ∘ C synd ∘ I AO ∘ F R D ∘ P R D . (144)
Proof. Combining Eqs. (D.7), (C.49), and (C.29) gives
F U T ∘ P U T ∘ I A O = { I A O ∘ F R D ∘ P R D , U = U R for some R ⊆ D , 0 U ∩ ( A ⊔ O ) ≠ ∅ (145) \mathcal{F}^{\mathsf{T}}_{U} \circ\mathcal{P}^{\mathsf{T}}_{U} \circ\mathcal{I}_{\mathrm{AO}} =
\begin{cases}
\mathcal{I}_{\mathrm{AO}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}, & U = U_{R}\text{ for some }R \subseteq\mathrm{D},\\
0 & U \cap(\mathrm{A} \sqcup\mathrm{O}) \ne\varnothing
\end{cases} \tag*{(145)} F U T ∘ P U T ∘ I AO = { I AO ∘ F R D ∘ P R D , 0 U = U R for some R ⊆ D , U ∩ ( A ⊔ O ) = ∅ (145)
for all U ⊆ T U \subseteq\mathrm{T} U ⊆ T . Substituting Eq. (D.22) into Eq. (D.20) and applying Eq. (D.25) termwise gives Eq. (D.24), completing the proof. □
Syndrome-resolved logical structure of the QEC cycle
Setup
For the remainder of this document, assume that the complete-cycle map Φ \Phi Φ in (138) can be expressed as follows: there exists a family { D T , U } U , T ⊆ T \{\mathcal{D}_{T,U}\}_{U,T \subseteq\mathrm{T}} { D T , U } U , T ⊆ T of subnormalized Pauli maps on End ( K T ) \operatorname{End}(\mathcal{K}_{\mathrm{T}}) End ( K T ) such that
Φ = ∑ R ⊆ D , T ⊆ T , θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R ∘ C s y n d ∘ I A O ∘ F R D ∘ P R D , (146) \Phi= \sum_{R \subseteq\mathrm{D},\ T \subseteq\mathrm{T},\ \theta\in\Theta} \mathcal{R}_{K(\theta)} \circ\mathcal{W}^{\mathrm{D}}(D(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{E}^{\mathrm{T}}_{T} \circ\mathcal{L}^{\mathrm{T}}_{T} \circ\mathcal{D}_{T,U_R} \circ\mathcal{C}_{\mathrm{synd}} \circ\mathcal{I}_{\mathrm{AO}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}, \tag*{(146)} Φ = R ⊆ D , T ⊆ T , θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R ∘ C synd ∘ I AO ∘ F R D ∘ P R D , (146)
that further satisfies
∑ T ⊆ T wt ( D T , U ) = 1 , (147) \sum_{T \subseteq\mathrm{T}} \operatorname{wt}(\mathcal{D}_{T,U}) = 1, \tag*{(147)} T ⊆ T ∑ wt ( D T , U ) = 1 , (147)
for all U ⊆ T U \subseteq\mathrm{T} U ⊆ T .
Fix one such family. Propositions D.23 and C.70 show that Φ t w \Phi^{\mathrm{tw}} Φ tw and the family { D T , U N } U , T ⊆ T \{\mathcal{D}^{\mathsf{N}}_{T,U}\}_{U,T \subseteq\mathrm{T}} { D T , U N } U , T ⊆ T give one such realization. The arguments below only invoke the fact that each D T , U \mathcal{D}_{T,U} D T , U is a subnormalized Pauli map and that Eqs. (E.1) and (E.2) hold; they do not depend on how these conditions are achieved.
Define Δ D \Delta^{\mathrm{D}} Δ D to be the leakage-pattern dephasing channel, as in Definition C.15, specialized to D. Equation (C.55), specialized to D, gives
Δ D = ∑ R ⊆ D E R D ∘ P R D . (148) \Delta^{\mathrm{D}} = \sum_{R \subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}. \tag*{(148)} Δ D = R ⊆ D ∑ E R D ∘ P R D . (148)
For T ⊆ T T \subseteq\mathrm{T} T ⊆ T , define T D : = T ∩ D T_{\mathrm{D}} := T \cap\mathrm{D} T D := T ∩ D .
Proposition E.4 (Complete-cycle leakage-dephasing invariance). The map Φ \Phi Φ satisfies
Φ = Φ ∘ Δ D = Δ D ∘ Φ = Δ D ∘ Φ ∘ Δ D . (149) \Phi= \Phi\circ\Delta^{\mathrm{D}} = \Delta^{\mathrm{D}} \circ\Phi= \Delta^{\mathrm{D}} \circ\Phi\circ\Delta^{\mathrm{D}}. \tag*{(149)} Φ = Φ ∘ Δ D = Δ D ∘ Φ = Δ D ∘ Φ ∘ Δ D . (149)
Proof. Equations (C.8), (C.51), (C.31), and (D.14) give
ran ( M θ ∘ E T T ∘ L T T ) ⊆ ran ( E T D D ) , (150) \operatorname{ran}\left(\mathcal{M}_{\theta} \circ\mathcal{E}^{\mathrm{T}}_{T} \circ\mathcal{L}^{\mathrm{T}}_{T}\right) \subseteq\operatorname{ran}\left(\mathcal{E}^{\mathrm{D}}_{T_{\mathrm{D}}}\right), \tag*{(150)} ran ( M θ ∘ E T T ∘ L T T ) ⊆ ran ( E T D D ) , (150)
for all T ⊆ T T \subseteq\mathrm{T} T ⊆ T , θ ∈ Θ \theta\in\Theta θ ∈ Θ . Equations (C.40) and (C.51), specialized to D, give
W D ( v ) ( ran ( E S D ) ) ⊆ ran ( E S D ) , (151) \mathcal{W}^{\mathrm{D}}(v)\left(\operatorname{ran}\left(\mathcal{E}^{\mathrm{D}}_{S}\right)\right) \subseteq\operatorname{ran}\left(\mathcal{E}^{\mathrm{D}}_{S}\right), \tag*{(151)} W D ( v ) ( ran ( E S D ) ) ⊆ ran ( E S D ) , (151)
for all S ⊆ D S \subseteq\mathrm{D} S ⊆ D , v ∈ V K D v \in V_{K_{\mathrm{D}}} v ∈ V K D . Equations (D.18), (C.8), and (C.51) give
ran ( R K ( θ ) ∘ E T D D ) ⊆ ran ( E T D ∖ K ( θ ) D ) , (152) \operatorname{ran}\left(\mathcal{R}_{K(\theta)} \circ\mathcal{E}^{\mathrm{D}}_{\mathcal{T}_{\mathrm{D}}}\right) \subseteq\operatorname{ran}\left(\mathcal{E}^{\mathrm{D}}_{\mathcal{T}_{\mathrm{D}}\setminus K(\theta)}\right), \tag*{(152)} ran ( R K ( θ ) ∘ E T D D ) ⊆ ran ( E T D ∖ K ( θ ) D ) , (152)
for all T ⊆ T T \subseteq\mathcal{T} T ⊆ T , θ ∈ Θ \theta\in\Theta θ ∈ Θ . Next, Eqs. (E.6), (E.7), and (E.8) give
ran ( R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ) ⊆ ran ( E T D ∖ K ( θ ) D ) , (153) \operatorname{ran}\left(\mathcal{R}_{K(\theta)} \circ\mathcal{W}^{\mathrm{D}}(\mathcal{D}(\theta)) \circ\mathcal{M}_{\theta} \circ\mathcal{E}^{\mathrm{T}}_{T} \circ\mathcal{L}^{\mathrm{T}}_{T}\right) \subseteq\operatorname{ran}\left(\mathcal{E}^{\mathrm{D}}_{\mathcal{T}_{\mathrm{D}}\setminus K(\theta)}\right), \tag*{(153)} ran ( R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ) ⊆ ran ( E T D ∖ K ( θ ) D ) , (153)
for all T ⊆ T T \subseteq\mathcal{T} T ⊆ T , θ ∈ Θ \theta\in\Theta θ ∈ Θ . Substituting Eq. (E.3) and applying Eq. (C.56) termwise, specialized to D, gives
P R D ∘ Δ D = P R D , Δ D ∘ E R D = E R D , (154) \mathcal{P}^{\mathrm{D}}_{R} \circ\Delta^{\mathrm{D}} = \mathcal{P}^{\mathrm{D}}_{R}, \qquad\Delta^{\mathrm{D}} \circ\mathcal{E}^{\mathrm{D}}_{R} = \mathcal{E}^{\mathrm{D}}_{R}, \tag*{(154)} P R D ∘ Δ D = P R D , Δ D ∘ E R D = E R D , (154)
for all R ⊆ D R \subseteq\mathrm{D} R ⊆ D . Right composing Eq. (E.1) with Δ D \Delta^{\mathrm{D}} Δ D and applying P R D ∘ Δ D = P R D \mathcal{P}^{\mathrm{D}}_{R} \circ\Delta^{\mathrm{D}} = \mathcal{P}^{\mathrm{D}}_{R} P R D ∘ Δ D = P R D from Eq. (E.10) termwise gives Φ ∘ Δ D = Φ \Phi\circ\Delta^{\mathrm{D}} = \Phi Φ ∘ Δ D = Φ . Left composing Eq. (E.1) with Δ D \Delta^{\mathrm{D}} Δ D , using Eq. (E.9), and applying Δ D ∘ E T D ∖ K ( θ ) D = E T D ∖ K ( θ ) D \Delta^{\mathrm{D}} \circ\mathcal{E}^{\mathrm{D}}_{\mathcal{T}_{\mathrm{D}}\setminus K(\theta)} = \mathcal{E}^{\mathrm{D}}_{\mathcal{T}_{\mathrm{D}}\setminus K(\theta)} Δ D ∘ E T D ∖ K ( θ ) D = E T D ∖ K ( θ ) D from Eq. (E.10) to each term in the summand gives Δ D ∘ Φ = Φ \Delta^{\mathrm{D}} \circ\Phi= \Phi Δ D ∘ Φ = Φ . These identities imply that Δ D ∘ Φ ∘ Δ D = Φ \Delta^{\mathrm{D}} \circ\Phi\circ\Delta^{\mathrm{D}} = \Phi Δ D ∘ Φ ∘ Δ D = Φ , proving Eq. (E.5). □
Clifford-frame representatives of complete-cycle leakage-subspace branches
Denote by { Φ R o u t , R i n } R i n , R o u t ⊆ D \{\Phi_{R_{\mathrm{out}},R_{\mathrm{in}}}\}_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}} { Φ R out , R in } R in , R out ⊆ D the family of leakage-subspace branches associated with Φ \Phi Φ , as in Definition C.57, specialized to D.
Definition E.11 (Clifford-frame representatives of complete-cycle leakage-subspace branches). For every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D , define the Clifford-frame representative Φ ~ R o u t , R i n : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} : \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) \to\operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) Φ R out , R in : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) according to
Φ ~ R o u t , R i n : = C F ∘ F R o u t D ∘ Φ R o u t , R i n ∘ L R i n D ∘ C F − 1 . (155) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} := \mathcal{C}_{F} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\Phi_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{C}^{-1}_{F}. \tag*{(155)} Φ R out , R in := C F ∘ F R out D ∘ Φ R out , R in ∘ L R in D ∘ C F − 1 . (155)
Lemma C.59 establishes that every Φ R o u t , R i n \Phi_{R_{\mathrm{out}},R_{\mathrm{in}}} Φ R out , R in is completely positive and trace nonincreasing. Unitarity of C F \mathcal{C}_{F} C F implies that C F \mathcal{C}_{F} C F and its inverse are CPTP. Lemma C.32 and Eq. (E.12) therefore establish that every Φ ~ R o u t , R i n \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} Φ R out , R in is completely positive and trace nonincreasing.
Proposition E.13 (Reconstruction from Clifford-frame representatives). For every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D ,
Φ R o u t , R i n = L R o u t D ∘ C F − 1 ∘ Φ ~ R o u t , R i n ∘ C F ∘ F R i n D . (156) \Phi_{R_{\mathrm{out}},R_{\mathrm{in}}} = \mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{C}^{-1}_{F} \circ\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{F} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(156)} Φ R out , R in = L R out D ∘ C F − 1 ∘ Φ R out , R in ∘ C F ∘ F R in D . (156) Moreover,
Φ = ∑ R i n , R o u t ⊆ D E R o u t D ∘ Φ R o u t , R i n ∘ P R i n D \Phi= \sum_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\Phi_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}} Φ = R in , R out ⊆ D ∑ E R out D ∘ Φ R out , R in ∘ P R in D = ∑ R i n , R o u t ⊆ D E R o u t D ∘ L R o u t D ∘ C F − 1 ∘ Φ ~ R o u t , R i n ∘ C F ∘ F R i n D ∘ P R i n D . (157) = \sum_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{C}^{-1}_{F} \circ\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{F} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(157)} = R in , R out ⊆ D ∑ E R out D ∘ L R out D ∘ C F − 1 ∘ Φ R out , R in ∘ C F ∘ F R in D ∘ P R in D . (157)
Proof. Unitarity of C F \mathcal{C}_{F} C F gives
C F − 1 ∘ C F = id End ( K D ) and C F ∘ C F − 1 = id End ( K L ⊗ K S ) . (158) \mathcal{C}^{-1}_{F} \circ\mathcal{C}_{F} = \operatorname{id}_{\operatorname{End}(\mathcal{K}_{\mathrm{D}})} \qquad\text{and} \qquad\mathcal{C}_{F} \circ\mathcal{C}^{-1}_{F} = \operatorname{id}_{\operatorname{End}(\mathcal{K}_{\mathrm{L}}\otimes\mathcal{K}_{\mathrm{S}})}. \tag*{(158)} C F − 1 ∘ C F = id End ( K D ) and C F ∘ C F − 1 = id End ( K L ⊗ K S ) . (158)
Substituting Eq. (E.12) into the right-hand side of Eq. (E.14) and applying Eqs. (E.16) and (C.33) reduces that side to Φ R o u t , R i n \Phi_{R_{\mathrm{out}},R_{\mathrm{in}}} Φ R out , R in . Substituting Eq. (E.3) into both occurrences of Δ D \Delta^{\mathrm{D}} Δ D in the equality Φ = Δ D ∘ Φ ∘ Δ D \Phi= \Delta^{\mathrm{D}} \circ\Phi\circ\Delta^{\mathrm{D}} Φ = Δ D ∘ Φ ∘ Δ D , distributing the compositions over the finite sums, and applying Eq. (C.58) termwise gives the first equality in Eq. (E.15). Substituting Eq. (E.14) termwise gives the second equality.
For each R i n ⊆ D R_{\mathrm{in}} \subseteq\mathrm{D} R in ⊆ D , define U R i n ⊆ T U_{R_{\mathrm{in}}} \subseteq\mathrm{T} U R in ⊆ T to be the image of R i n R_{\mathrm{in}} R in under the canonical inclusion D ↪ T \mathrm{D} \hookrightarrow\mathrm{T} D ↪ T . For each such R i n R_{\mathrm{in}} R in , denote by G U R i n T \mathcal{G}^{\mathrm{T}}_{U_{R_{\mathrm{in}}}} G U R in T the computational replacement channel, as in Definition C.34 , specialized to T \mathrm{T} T and U R i n U_{R_{\mathrm{in}}} U R in .
Lemma E.17 (Explicit form of the Clifford-frame representatives). For every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D ,
Φ ~ R o u t , R i n = ∑ T ⊆ T ∑ θ ∈ Θ C F ∘ F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R i n ∘ C s y n d ∘ G U R i n T ∘ I A O ∘ C F − 1 . (159) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}
=
\sum_{\mathrm{T}\subseteq\mathrm{T}}
\sum_{\theta\in\Theta}
\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}}
\circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{out}}}
\circ\mathcal{R}_{K(\theta)} \circ\mathcal{W}^{\mathrm{D}}(D(\theta)) \circ\mathcal{M}_{\theta}
\circ\mathcal{E}^{\mathrm{T}}_{\mathrm{T}} \circ\mathcal{L}^{\mathrm{T}}_{\mathrm{T}} \circ\mathcal{D}_{\mathrm{T},U_{R_{\mathrm{in}}}}
\circ\mathcal{C}_{\mathrm{synd}} \circ\mathcal{G}^{\mathrm{T}}_{U_{R_{\mathrm{in}}}} \circ\mathcal{I}_{\mathrm{AO}} \circ\mathcal{C}_{\mathrm{F}}^{-1}. \tag*{(159)} Φ R out , R in = T ⊆ T ∑ θ ∈ Θ ∑ C F ∘ F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R in ∘ C synd ∘ G U R in T ∘ I AO ∘ C F − 1 . (159)
Proof. Fix R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D . For R ≠ R i n R \ne R_{\mathrm{in}} R = R in , Eq. (C.56), specialized to D \mathrm{D} D , gives P R D ∘ E R i n D = 0 \mathcal{P}^{\mathrm{D}}_{R} \circ\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} = 0 P R D ∘ E R in D = 0 . Left composing this equality with F R D \mathcal{F}^{\mathrm{D}}_{R} F R D and right composing with L R i n D \mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} L R in D gives the R ≠ R i n R \ne R_{\mathrm{in}} R = R in case below. For R = R i n R = R_{\mathrm{in}} R = R in , Eq. (C.53), specialized to D \mathrm{D} D , reduces the map F R D ∘ P R D ∘ E R i n D ∘ L R i n D \mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R} \circ\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} F R D ∘ P R D ∘ E R in D ∘ L R in D to F R i n D ∘ L R i n D \mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} F R in D ∘ L R in D . Let G R i n D \mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}} G R in D denote the computational replacement channel, as in Definition C.34 , specialized to D \mathrm{D} D and R i n R_{\mathrm{in}} R in . Equation (C.35), specialized to D \mathrm{D} D , gives F R i n D ∘ L R i n D = G R i n D \mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} = \mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}} F R in D ∘ L R in D = G R in D . Hence, for every R ⊆ D R \subseteq\mathrm{D} R ⊆ D , we find that
F R D ∘ P R D ∘ E R i n D ∘ L R i n D = { G R i n D R = R i n , 0 End ( K D ) → End ( K D ) R ≠ R i n . (160) \mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R} \circ\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}}
=
\begin{cases}
\mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}} & R = R_{\mathrm{in}},\\
0_{\operatorname{End}(\mathcal{K}_{\mathrm{D}})\to\operatorname{End}(\mathcal{K}_{\mathrm{D}})} & R \ne R_{\mathrm{in}}.
\end{cases} \tag*{(160)} F R D ∘ P R D ∘ E R in D ∘ L R in D = { G R in D 0 End ( K D ) → End ( K D ) R = R in , R = R in . (160)
Right composing Eq. (E.1) with E R i n D ∘ L R i n D \mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}} E R in D ∘ L R in D and applying Eq. (E.19) termwise annihilates the summands with R ≠ R i n R \ne R_{\mathrm{in}} R = R in . The remaining summands lead to
Φ ∘ E R i n D ∘ L R i n D = ∑ T ⊆ T ∑ θ ∈ Θ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R i n ∘ C s y n d ∘ I A O ∘ G R i n D . (161) \Phi\circ\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{in}}}
=
\sum_{\mathrm{T}\subseteq\mathrm{T}}
\sum_{\theta\in\Theta}
\mathcal{R}_{K(\theta)} \circ\mathcal{W}^{\mathrm{D}}(D(\theta)) \circ\mathcal{M}_{\theta}
\circ\mathcal{E}^{\mathrm{T}}_{\mathrm{T}} \circ\mathcal{L}^{\mathrm{T}}_{\mathrm{T}} \circ\mathcal{D}_{\mathrm{T},U_{R_{\mathrm{in}}}}
\circ\mathcal{C}_{\mathrm{synd}} \circ\mathcal{I}_{\mathrm{AO}} \circ\mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(161)} Φ ∘ E R in D ∘ L R in D = T ⊆ T ∑ θ ∈ Θ ∑ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R in ∘ C synd ∘ I AO ∘ G R in D . (161)
For every Y ∈ End ( K D ) Y \in\operatorname{End}(\mathcal{K}_{\mathrm{D}}) Y ∈ End ( K D ) , Eq. (C.36), specialized to D \mathrm{D} D with S = R i n S = R_{\mathrm{in}} S = R in and to T \mathrm{T} T with S = U R i n S = U_{R_{\mathrm{in}}} S = U R in , together with Eq. (D.7), gives
( I A O ∘ G R i n D ) ( Y ) = G R i n D ( Y ) ⊗ ∣ 0 ⟩ ⟨ 0 ∣ A ⊗ ∣ 0 ⟩ ⟨ 0 ∣ O = G U R i n T ( Y ⊗ ∣ 0 ⟩ ⟨ 0 ∣ A ⊗ ∣ 0 ⟩ ⟨ 0 ∣ O ) = ( G U R i n T ∘ I A O ) ( Y ) . (162) (\mathcal{I}_{\mathrm{AO}} \circ\mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}})(Y)
=
\mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}}(Y) \otimes|0\rangle\langle0|_{\mathrm{A}} \otimes|0\rangle\langle0|_{\mathrm{O}}
=
\mathcal{G}^{\mathrm{T}}_{U_{R_{\mathrm{in}}}}(Y \otimes|0\rangle\langle0|_{\mathrm{A}} \otimes|0\rangle\langle0|_{\mathrm{O}})
=
(\mathcal{G}^{\mathrm{T}}_{U_{R_{\mathrm{in}}}} \circ\mathcal{I}_{\mathrm{AO}})(Y). \tag*{(162)} ( I AO ∘ G R in D ) ( Y ) = G R in D ( Y ) ⊗ ∣0 ⟩ ⟨ 0 ∣ A ⊗ ∣0 ⟩ ⟨ 0 ∣ O = G U R in T ( Y ⊗ ∣0 ⟩ ⟨ 0 ∣ A ⊗ ∣0 ⟩ ⟨ 0 ∣ O ) = ( G U R in T ∘ I AO ) ( Y ) . (162)
Since the foregoing expression holds for every Y ∈ End ( K D ) Y \in\operatorname{End}(\mathcal{K}_{\mathrm{D}}) Y ∈ End ( K D ) , it follows that
I A O ∘ G R i n D = G U R i n T ∘ I A O . (163) \mathcal{I}_{\mathrm{AO}} \circ\mathcal{G}^{\mathrm{D}}_{R_{\mathrm{in}}}
=
\mathcal{G}^{\mathrm{T}}_{U_{R_{\mathrm{in}}}} \circ\mathcal{I}_{\mathrm{AO}}. \tag*{(163)} I AO ∘ G R in D = G U R in T ∘ I AO . (163)
Substituting the relation above into each summand of Eq. (E.20), left composing with C F ∘ F R o u t D ∘ P R o u t D \mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{out}}} C F ∘ F R out D ∘ P R out D , and right composing with C F − 1 \mathcal{C}_{\mathrm{F}}^{-1} C F − 1 gives Eq. (E.18), by Eqs. (C.58) and (E.12).
Syndrome-conditioned Pauli reduction of sector branches
Denote by V L \mathcal{V}_{\mathrm{L}} V L and V S \mathcal{V}_{\mathrm{S}} V S the Pauli-label spaces of L and S, respectively, as in Definition A.6 .
Definition E.23 (Syndrome-subspace selection maps). For each s ∈ Z 2 S s \in\mathbb{Z}_{2}^{\mathrm{S}} s ∈ Z 2 S , define the map Z s : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) \mathcal{Z}_{s} : \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) \to\operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) Z s : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) according to
Z s ( Y ) : = ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) Y ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) , (164) \mathcal{Z}_{s}(Y) := (\mathbb{1}_{\mathrm{L}} \otimes|s\rangle\langle s|_{\mathrm{S}})Y(\mathbb{1}_{\mathrm{L}} \otimes|s\rangle\langle s|_{\mathrm{S}}), \tag*{(164)} Z s ( Y ) := ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) Y ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) , (164) for all Y ∈ End ( K L ⊗ K S ) Y \in\operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) Y ∈ End ( K L ⊗ K S ) , where 1 L \mathbb{1}_{\mathrm{L}} 1 L denotes the identity on K L \mathcal{K}_{\mathrm{L}} K L .
Definition E.25 (Syndrome-dephasing map). Define the map Z S : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) \mathcal{Z}_{\mathrm{S}} : \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) \to\operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) Z S : End ( K L ⊗ K S ) → End ( K L ⊗ K S ) according to
Z S : = ∑ s ∈ Z 2 S Z s , (165) \mathcal{Z}_{\mathrm{S}} := \sum_{s \in\mathbb{Z}_{2}^{\mathrm{S}}} \mathcal{Z}_{s}, \tag*{(165)} Z S := s ∈ Z 2 S ∑ Z s , (165) with Z s \mathcal{Z}_{s} Z s as defined in Definition E.23 .
Lemma E.27. For every X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) and s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S ,
Z S ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ X ⊗ ∣ s ⟩ ⟨ s ∣ S . (166) \mathcal{Z}_{\mathrm{S}}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right) = \delta_{s,s'}X \otimes|s\rangle\langle s|_{\mathrm{S}}. \tag*{(166)} Z S ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ X ⊗ ∣ s ⟩ ⟨ s ∣ S . (166)
Proof. For every t , s , s ′ ∈ Z 2 S t,s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} t , s , s ′ ∈ Z 2 S and X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) , Eq. (E.24) gives
Z t ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ t , s δ t , s ′ X ⊗ ∣ t ⟩ ⟨ t ∣ S . \mathcal{Z}_{t}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right) = \delta_{t,s}\delta_{t,s'}X \otimes|t\rangle\langle t|_{\mathrm{S}}. Z t ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ t , s δ t , s ′ X ⊗ ∣ t ⟩ ⟨ t ∣ S .
Summing this equality over t t t and invoking Eq. (E.26) recovers Eq. (E.28). □
Lemma E.29. The map Z S \mathcal{Z}_{\mathrm{S}} Z S satisfies
Z S = 1 2 ∣ S ∣ ∑ z ∈ Z 2 S Ad 1 L ⊗ W S ( 0 , z ) , (167) \mathcal{Z}_{\mathrm{S}} = \frac{1}{2^{|\mathrm{S}|}} \sum_{z \in\mathbb{Z}_{2}^{\mathrm{S}}} \operatorname{Ad}_{\mathbb{1}_{\mathrm{L}} \otimes W^{\mathrm{S}}(0,z)}, \tag*{(167)} Z S = 2 ∣ S ∣ 1 z ∈ Z 2 S ∑ Ad 1 L ⊗ W S ( 0 , z ) , (167) and is a CPTP Pauli channel.
Proof. For every s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S ,
1 2 ∣ S ∣ ∑ z ∈ Z 2 S ( − 1 ) z ⋅ ( s + s ′ ) = ∏ i ∈ S [ 1 2 ∑ z i ∈ Z 2 ( − 1 ) z i ( s i + s i ′ ) ] = ∏ i ∈ S 1 + ( − 1 ) s i + s i ′ 2 = δ s , s ′ . (168) \frac{1}{2^{|\mathrm{S}|}}\sum_{z\in\mathbb{Z}_{2}^{\mathrm{S}}}(-1)^{z\cdot(s+s')}=\prod_{i\in\mathrm{S}}\left[\frac{1}{2}\sum_{z_i\in\mathbb{Z}_2}(-1)^{z_i(s_i+s_i')}\right]=\prod_{i\in\mathrm{S}}\frac{1+(-1)^{s_i+s_i'}}{2}=\delta_{s,s'}. \tag*{(168)} 2 ∣ S ∣ 1 z ∈ Z 2 S ∑ ( − 1 ) z ⋅ ( s + s ′ ) = i ∈ S ∏ [ 2 1 z i ∈ Z 2 ∑ ( − 1 ) z i ( s i + s i ′ ) ] = i ∈ S ∏ 2 1 + ( − 1 ) s i + s i ′ = δ s , s ′ . (168)
Combining the foregoing relation with Eqs. (A.10), (A.13), (E.31), and (E.28) gives, for every X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) and s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S ,
[ 1 2 ∣ S ∣ ∑ z ∈ Z 2 S Ad 1 L ⊗ W S ( 0 , z ) ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = 1 2 ∣ S ∣ ∑ z ∈ Z 2 S ( − 1 ) z ⋅ ( s + s ′ ) X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S = δ s , s ′ X ⊗ ∣ s ⟩ ⟨ s ∣ S = Z S ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (169) \begin{aligned}
\left[\frac{1}{2^{|\mathrm{S}|}}\sum_{z\in\mathbb{Z}_{2}^{\mathrm{S}}}\operatorname{Ad}_{\mathbb{1}_{\mathrm{L}}\otimes W^{\mathrm{S}}(0,z)}\right]\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right) \\
&= \frac{1}{2^{|\mathrm{S}|}}\sum_{z\in\mathbb{Z}_{2}^{\mathrm{S}}}(-1)^{z\cdot(s+s')}X\otimes|s\rangle\langle s'|_{\mathrm{S}} \\
&= \delta_{s,s'}X\otimes|s\rangle\langle s|_{\mathrm{S}}=\mathcal{Z}_{\mathrm{S}}\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right). \tag*{(169)}
\end{aligned} 2 ∣ S ∣ 1 z ∈ Z 2 S ∑ Ad 1 L ⊗ W S ( 0 , z ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = 2 ∣ S ∣ 1 z ∈ Z 2 S ∑ ( − 1 ) z ⋅ ( s + s ′ ) X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S = δ s , s ′ X ⊗ ∣ s ⟩ ⟨ s ∣ S = Z S ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (169)
The operators X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S X \otimes|s\rangle\langle s'|_{\mathrm{S}} X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S (with X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) and s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S ) span End ( K L ⊗ K S ) \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) End ( K L ⊗ K S ) , so Eq. (E.30) follows from Eq. (E.32). Equation (E.30) and Definition A.54 establish that Z S \mathcal{Z}_{\mathrm{S}} Z S is a Pauli channel. Lemma A.55 therefore implies that Z S \mathcal{Z}_{\mathrm{S}} Z S is CPTP.
Lemma E.33. For every Pauli map D \mathcal{D} D on End ( K L ⊗ K S ) \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) End ( K L ⊗ K S ) ,
D ∘ Z S = Z S ∘ D . (170) \mathcal{D} \circ\mathcal{Z}_{\mathrm{S}} = \mathcal{Z}_{\mathrm{S}} \circ\mathcal{D}. \tag*{(170)} D ∘ Z S = Z S ∘ D . (170)
Proof. Because V L × V S V_{\mathrm{L}} \times V_{\mathrm{S}} V L × V S is abelian, applying Eq. (A.26) and Lemma A.27 to K L ⊗ K S \mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}} K L ⊗ K S shows that the maps Ad W L ( ℓ ) ⊗ W S ( r ) \operatorname{Ad}_{W_{\mathrm{L}}(\ell)\otimes W_{\mathrm{S}}(r)} Ad W L ( ℓ ) ⊗ W S ( r ) , with ( ℓ , r ) ∈ V L × V S (\ell,r) \in V_{\mathrm{L}} \times V_{\mathrm{S}} ( ℓ , r ) ∈ V L × V S , commute with one another under composition. Combining Eqs. (A.41) and (E.30) proves Eq. (E.34).
For every leakage pattern R ⊆ D R \subseteq\mathrm{D} R ⊆ D , denote by V R D ⊆ V K D V_{R}^{\mathrm{D}} \subseteq V_{\mathrm{K}_{\mathrm{D}}} V R D ⊆ V K D the subgroup of Pauli labels supported on R R R , specialized to K D \mathcal{K}_{\mathrm{D}} K D . For every T ⊆ T T \subseteq\mathsf{T} T ⊆ T , denote by V T T ⊆ V K T V_{T}^{\mathrm{T}} \subseteq V_{\mathrm{K}_{\mathrm{T}}} V T T ⊆ V K T the corresponding subgroup specialized to K T \mathcal{K}_{\mathrm{T}} K T .
Proposition E.35. There exists a family of Pauli maps on End ( K D ) \operatorname{End}(\mathcal{K}_{\mathrm{D}}) End ( K D ) , denoted
{ B s R o u t , R i n } R i n , R o u t ⊆ D s ∈ Z 2 S , \left\{\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\right\}_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}^{s\in\mathbb{Z}_{2}^{\mathrm{S}}}, { B s R out , R in } R in , R out ⊆ D s ∈ Z 2 S , such that, for every R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D , we have that
Φ ~ R o u t , R i n = ∑ s ∈ Z 2 S C F ∘ B s R o u t , R i n ∘ C F − 1 ∘ Z s . (171) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}
=
\sum_{s\in\mathbb{Z}_{2}^{\mathrm{S}}}
\mathcal{C}_{\mathrm{F}}\circ\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\circ\mathcal{C}_{\mathrm{F}}^{-1}\circ\mathcal{Z}_{s}. \tag*{(171)} Φ R out , R in = s ∈ Z 2 S ∑ C F ∘ B s R out , R in ∘ C F − 1 ∘ Z s . (171)
Proof. Fix R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D , X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) , and s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S . For T ⊆ T T \subseteq\mathsf{T} T ⊆ T , write
T D : = T ∩ D , T A : = T ∩ A , and T O : = T ∩ O . T_{\mathrm{D}} := T \cap\mathrm{D}, \qquad T_{\mathrm{A}} := T \cap\mathrm{A}, \qquad\text{and} \qquad T_{\mathrm{O}} := T \cap\mathrm{O}. T D := T ∩ D , T A := T ∩ A , and T O := T ∩ O .
For v ∈ V K T v \in V_{\mathrm{K}_{\mathrm{T}}} v ∈ V K T , denote by v D ∈ V K D v_{\mathrm{D}} \in V_{\mathrm{K}_{\mathrm{D}}} v D ∈ V K D , v A ∈ V K A v_{\mathrm{A}} \in V_{\mathrm{K}_{\mathrm{A}}} v A ∈ V K A , and v O ∈ V K O v_{\mathrm{O}} \in V_{\mathrm{K}_{\mathrm{O}}} v O ∈ V K O , respectively, the restriction of v v v to D, A, and O. Evaluating Eq. (E.18) on X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S X \otimes|s\rangle\langle s'|_{\mathrm{S}} X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S gives
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = ∑ T ⊆ T ∑ θ ∈ Θ [ C F ∘ F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R i n ∘ C s y n d ∘ G U R i n T ∘ I A O ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (172) \begin{aligned}
\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right)
={}&
\sum_{T\subseteq\mathsf{T}}\sum_{\theta\in\Theta}
\Bigl[
\mathcal{C}_{\mathrm{F}}\circ\mathcal{F}_{R_{\mathrm{out}}}^{\mathrm{D}}\circ\mathcal{P}_{R_{\mathrm{out}}}^{\mathrm{D}}\circ\mathcal{R}_{\mathrm{K}}(\theta)\\
&\circ\mathcal{W}^{\mathrm{D}}(D(\theta))\circ\mathcal{M}_{\theta}\circ\mathcal{E}_{T}^{\mathrm{T}}\circ\mathcal{L}_{T}^{\mathrm{T}}\\
&\circ\mathcal{D}_{T,U_{R_{\mathrm{in}}}}\circ\mathcal{C}_{\mathrm{synd}}\circ\mathcal{G}_{U_{R_{\mathrm{in}}}}^{\mathrm{T}}\circ\mathcal{I}_{\mathrm{AO}}\circ\mathcal{C}_{\mathrm{F}}^{-1}
\Bigr]\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right). \tag*{(172)}
\end{aligned} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = T ⊆ T ∑ θ ∈ Θ ∑ [ C F ∘ F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ D T , U R in ∘ C synd ∘ G U R in T ∘ I AO ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (172)
For every T ⊆ T T \subseteq\mathsf{T} T ⊆ T and e ∈ V K T e \in V_{\mathrm{K}_{\mathrm{T}}} e ∈ V K T , define
q e T , U R i n : = 1 ( dim K T ) 2 ⟨ W K T ( e ) , D T , U R i n ⟩ S . (173) q_{e}^{T,U_{R_{\mathrm{in}}}} := \frac{1}{(\dim\mathcal{K}_{\mathrm{T}})^{2}}\left\langle\mathcal{W}^{\mathrm{K}_{\mathrm{T}}}(e),\mathcal{D}_{T,U_{R_{\mathrm{in}}}}\right\rangle_{\mathrm{S}}. \tag*{(173)} q e T , U R in := ( dim K T ) 2 1 ⟨ W K T ( e ) , D T , U R in ⟩ S . (173)
By the standing assumption on { D T , U } U , T ⊆ T \{\mathcal{D}_{T,U}\}_{U,T\subseteq\mathsf{T}} { D T , U } U , T ⊆ T in Appendix E, every D T , U R i n \mathcal{D}_{T,U_{R_{\mathrm{in}}}} D T , U R in is a Pauli map. Thus, Lemma A.42, specialized to K T \mathcal{K}_{\mathrm{T}} K T , gives
D T , U R i n = ∑ e ∈ V K T q e T , U R i n W K T ( e ) . (174) \mathcal{D}_{T,U_{R_{\mathrm{in}}}}
=
\sum_{e\in V_{\mathrm{K}_{\mathrm{T}}}}
q_{e}^{T,U_{R_{\mathrm{in}}}}\mathcal{W}^{\mathrm{K}_{\mathrm{T}}}(e). \tag*{(174)} D T , U R in = e ∈ V K T ∑ q e T , U R in W K T ( e ) . (174)
Equation (C.39), specialized to T \mathsf{T} T with S = U R i n \mathcal{S}=U_{R_{\mathrm{in}}} S = U R in , gives
G U R i n T = 1 4 ∣ U R i n ∣ ∑ u ∈ V U R i n T W K T ( u ) . (175) \mathcal{G}^{\mathsf{T}}_{U_{R_{\mathrm{in}}}}=\frac{1}{4^{|U_{R_{\mathrm{in}}}|}}\sum_{u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}}}\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(u). \tag*{(175)} G U R in T = 4 ∣ U R in ∣ 1 u ∈ V U R in T ∑ W K T ( u ) . (175)
For every e ∈ V K T e\in V_{\mathsf{K}_{\mathsf{T}}} e ∈ V K T and u ∈ V U R i n T u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , define
v e , u : = e + F s y n d u . (176) v_{e,u}:=e+F_{\mathrm{synd}}u. \tag*{(176)} v e , u := e + F synd u . (176)
Combining Eq. (B.2), specialized to C s y n d C_{\mathrm{synd}} C synd and F s y n d F_{\mathrm{synd}} F synd , with Eqs. (A.15) and (G.5) gives
W K T ( e ) ∘ C s y n d ∘ W K T ( u ) = W K T ( e ) ∘ W K T ( F s y n d u ) ∘ C s y n d = W K T ( e + F s y n d u ) ∘ C s y n d = W K T ( v e , u ) ∘ C s y n d . (177) \begin{aligned}
\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(e)\circ C_{\mathrm{synd}}\circ\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(u)
&=\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(e)\circ\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(F_{\mathrm{synd}}u)\circ C_{\mathrm{synd}}\\
&=\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(e+F_{\mathrm{synd}}u)\circ C_{\mathrm{synd}}
=\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(v_{e,u})\circ C_{\mathrm{synd}}. \tag*{(177)}
\end{aligned} W K T ( e ) ∘ C synd ∘ W K T ( u ) = W K T ( e ) ∘ W K T ( F synd u ) ∘ C synd = W K T ( e + F synd u ) ∘ C synd = W K T ( v e , u ) ∘ C synd . (177)
Substituting Eqs. (E.39) and (E.40) into Eq. (E.37) and applying Eq. (E.42) termwise gives
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = ∑ T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V U R i n T q e T , U R i n 4 ∣ U R i n ∣ [ C F ∘ F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ∘ W K T ( v e , u ) ∘ C s y n d ∘ I A O ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (178) \begin{aligned}
\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}(X\otimes|s\rangle\langle s'|_{S})
={}&\sum_{T\subseteq\mathsf{T}}\sum_{\theta\in\Theta}\sum_{e\in V_{\mathsf{K}_{\mathsf{T}}}}\sum_{u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}}}
\frac{q^{T,U_{R_{\mathrm{in}}}}_{e}}{4^{|U_{R_{\mathrm{in}}}|}}\\
&\quad\left[
C_{\mathrm{F}}\circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{R}_{\mathsf{K}}(\theta)
\circ\mathcal{W}^{\mathrm{D}}(D(\theta))\circ\mathcal{M}_{\theta}\circ\mathcal{E}^{\mathsf{T}}_{T}\circ\mathcal{L}^{\mathsf{T}}_{T}\right.\\
&\qquad\left.\circ\mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(v_{e,u})\circ C_{\mathrm{synd}}\circ\mathcal{I}_{\mathrm{AO}}\circ C_{\mathrm{F}}^{-1}
\right](X\otimes|s\rangle\langle s'|_{S}). \tag*{(178)}
\end{aligned} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V U R in T ∑ 4 ∣ U R in ∣ q e T , U R in [ C F ∘ F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ∘ W K T ( v e , u ) ∘ C synd ∘ I AO ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (178)
For every e ∈ V K T e\in V_{\mathsf{K}_{\mathsf{T}}} e ∈ V K T and u ∈ V U R i n T u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , write
( v e , u ) O = ( x e , u , O , z e , u , O ) , (v_{e,u})_{\mathrm{O}}=(x_{e,u,\mathrm{O}},z_{e,u,\mathrm{O}}), ( v e , u ) O = ( x e , u , O , z e , u , O ) ,
where x e , u , O , z e , u , O ∈ Z 2 O x_{e,u,\mathrm{O}},z_{e,u,\mathrm{O}}\in\mathbb{Z}^{\mathrm{O}}_{2} x e , u , O , z e , u , O ∈ Z 2 O . For every e ∈ V K T e\in V_{\mathsf{K}_{\mathsf{T}}} e ∈ V K T and u ∈ V U R i n T u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , define the map r e , u : Z 2 S → Z 2 O r_{e,u}:\mathbb{Z}^{\mathrm{S}}_{2}\to\mathbb{Z}^{\mathrm{O}}_{2} r e , u : Z 2 S → Z 2 O via
r e , u ( t ) : = E ( t ) + x e , u , O , (179) r_{e,u}(t):=E(t)+x_{e,u,\mathrm{O}}, \tag*{(179)} r e , u ( t ) := E ( t ) + x e , u , O , (179)
for all t ∈ Z 2 S t\in\mathbb{Z}^{\mathrm{S}}_{2} t ∈ Z 2 S . For every T ⊆ T T\subseteq\mathsf{T} T ⊆ T , define the map θ T : Z 2 O → Θ \theta_{T}:\mathbb{Z}^{\mathrm{O}}_{2}\to\Theta θ T : Z 2 O → Θ according to
[ θ T ( r ) ] i : = { L , i ∈ T O , r i , i ∈ O ∖ T O , (180) [\theta_{T}(r)]_{i}:=
\begin{cases}
\mathsf{L}, & i\in T_{\mathrm{O}},\\
r_{i}, & i\in\mathrm{O}\setminus T_{\mathrm{O}},
\end{cases} \tag*{(180)} [ θ T ( r ) ] i := { L , r i , i ∈ T O , i ∈ O ∖ T O , (180)
for all r ∈ Z 2 O r\in\mathbb{Z}^{\mathrm{O}}_{2} r ∈ Z 2 O , i ∈ O i\in\mathrm{O} i ∈ O . For every e ∈ V K T e\in V_{\mathsf{K}_{\mathsf{T}}} e ∈ V K T and u ∈ V U R i n T u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , Eq. (A.13) gives
W K T ( v e , u ) = W K D ( ( v e , u ) D ) ⊗ W K A ( ( v e , u ) A ) ⊗ W K O ( ( v e , u ) O ) . (181) \mathcal{W}^{\mathsf{K}_{\mathsf{T}}}(v_{e,u})=
\mathcal{W}^{\mathsf{K}_{\mathrm{D}}}((v_{e,u})_{\mathrm{D}})
\otimes\mathcal{W}^{\mathsf{K}_{\mathrm{A}}}((v_{e,u})_{\mathrm{A}})
\otimes\mathcal{W}^{\mathsf{K}_{\mathrm{O}}}((v_{e,u})_{\mathrm{O}}). \tag*{(181)} W K T ( v e , u ) = W K D (( v e , u ) D ) ⊗ W K A (( v e , u ) A ) ⊗ W K O (( v e , u ) O ) . (181)
For every e ∈ V K T e\in V_{\mathsf{K}_{\mathsf{T}}} e ∈ V K T and u ∈ V U R i n T u\in V^{\mathsf{T}}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , taking outer products in Eq. (D.9), and applying Eqs. (D.10), (D.7), (A.13), and (E.46), gives
( W K T ( ( v e , u ) T ) ∘ C s y n d ∘ I A O ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = ( W K D ( ( v e , u ) D ) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) ⊗ W K A ( ( v e , u ) A ) ( ∣ α ( s ) ⟩ ⟨ α ( s ′ ) ∣ A ) ⊗ ( − 1 ) z e , u , O ⋅ ( E ( s ) + E ( s ′ ) ) ∣ r e , u ( s ) ⟩ ⟨ r e , u ( s ′ ) ∣ O . (182) \begin{aligned}
\left(\mathcal{W}^{K\mathrm{T}}((v_{e,u})_{\mathrm{T}})\circ\mathcal{C}_{\mathrm{synd}}\circ\mathcal{I}_{\mathrm{AO}}\circ\mathcal{C}_{\mathrm{F}}^{-1}\right)\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right)
={}&\left(\mathcal{W}^{K\mathrm{D}}((v_{e,u})_{\mathrm{D}})\circ\mathcal{C}_{\mathrm{F}}^{-1}\right)\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right)\\
&\otimes\mathcal{W}^{K\mathrm{A}}((v_{e,u})_{\mathrm{A}})\left(|\alpha(s)\rangle\langle\alpha(s')|_{\mathrm{A}}\right)\\
&\otimes(-1)^{z_{e,u,\mathrm{O}}\cdot(E(s)+E(s'))}|r_{e,u}(s)\rangle\langle r_{e,u}(s')|_{\mathrm{O}}. \tag*{(182)}
\end{aligned} ( W K T (( v e , u ) T ) ∘ C synd ∘ I AO ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = ( W K D (( v e , u ) D ) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) ⊗ W K A (( v e , u ) A ) ( ∣ α ( s )⟩ ⟨ α ( s ′ ) ∣ A ) ⊗ ( − 1 ) z e , u , O ⋅ ( E ( s ) + E ( s ′ )) ∣ r e , u ( s )⟩ ⟨ r e , u ( s ′ ) ∣ O . (182)
Substituting Eq. (E.47) termwise into Eq. (E.43) leads to
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = ∑ T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V R i n T q e T , U R i n 4 ∣ U R i n ∣ [ C F ∘ F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ M θ ∘ E T T ∘ L T T ] ( ( W K D ( ( v e , u ) D ) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) ⊗ W K A ( ( v e , u ) A ) ( ∣ α ( s ) ⟩ ⟨ α ( s ′ ) ∣ A ) ⊗ ( − 1 ) z e , u , O ⋅ ( E ( s ) + E ( s ′ ) ) ∣ r e , u ( s ) ⟩ ⟨ r e , u ( s ′ ) ∣ O ) . (183) \begin{aligned}
\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right)
={}&\sum_{T\subseteq\mathsf{T}}\sum_{\theta\in\Theta}\sum_{e\in V_{K_{\mathsf{T}}}}\sum_{u\in V^{\mathsf{T}}_{R_{\mathrm{in}}}}
\frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4^{|U_{R_{\mathrm{in}}}|}}
\left[\mathcal{C}_{\mathrm{F}}\circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{R}_{K(\theta)}\right.\\
&\left.\qquad\circ\mathcal{W}^{\mathrm{D}}(D(\theta))\circ\mathcal{M}_{\theta}\circ\mathcal{E}^{\mathsf{T}}_{\mathsf{T}}\circ\mathcal{L}^{\mathsf{T}}_{\mathsf{T}}\right]\\
&\quad\left(\left(\mathcal{W}^{K\mathrm{D}}((v_{e,u})_{\mathrm{D}})\circ\mathcal{C}_{\mathrm{F}}^{-1}\right)\left(X\otimes|s\rangle\langle s'|_{\mathrm{S}}\right)\right.\\
&\qquad\otimes\mathcal{W}^{K\mathrm{A}}((v_{e,u})_{\mathrm{A}})\left(|\alpha(s)\rangle\langle\alpha(s')|_{\mathrm{A}}\right)\\
&\left.\qquad\otimes(-1)^{z_{e,u,\mathrm{O}}\cdot(E(s)+E(s'))}|r_{e,u}(s)\rangle\langle r_{e,u}(s')|_{\mathrm{O}}\right). \tag*{(183)}
\end{aligned} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V R in T ∑ 4 ∣ U R in ∣ q e T , U R in [ C F ∘ F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ M θ ∘ E T T ∘ L T T ] ( ( W K D (( v e , u ) D ) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) ⊗ W K A (( v e , u ) A ) ( ∣ α ( s )⟩ ⟨ α ( s ′ ) ∣ A ) ⊗ ( − 1 ) z e , u , O ⋅ ( E ( s ) + E ( s ′ )) ∣ r e , u ( s )⟩ ⟨ r e , u ( s ′ ) ∣ O ) . (183)
Writing T = T D ⊔ T A ⊔ T O T=T_{\mathrm{D}}\sqcup T_{\mathrm{A}}\sqcup T_{\mathrm{O}} T = T D ⊔ T A ⊔ T O and applying Eqs. (C.31) and (C.51) to each tensor factor under the canonical labeled tensor-product identifications leads to
E T T ∘ L T T = ( E T D D ∘ L T D D ) ⊗ ( E T A A ∘ L T A A ) ⊗ ( E T O O ∘ L T O O ) . (184) \mathcal{E}^{\mathsf{T}}_{\mathsf{T}}\circ\mathcal{L}^{\mathsf{T}}_{\mathsf{T}}
=
\left(\mathcal{E}^{\mathrm{D}}_{T_{\mathrm{D}}}\circ\mathcal{L}^{\mathrm{D}}_{T_{\mathrm{D}}}\right)
\otimes
\left(\mathcal{E}^{\mathrm{A}}_{T_{\mathrm{A}}}\circ\mathcal{L}^{\mathrm{A}}_{T_{\mathrm{A}}}\right)
\otimes
\left(\mathcal{E}^{\mathrm{O}}_{T_{\mathrm{O}}}\circ\mathcal{L}^{\mathrm{O}}_{T_{\mathrm{O}}}\right). \tag*{(184)} E T T ∘ L T T = ( E T D D ∘ L T D D ) ⊗ ( E T A A ∘ L T A A ) ⊗ ( E T O O ∘ L T O O ) . (184)
Next, for a ∈ Z 2 O a\in\mathbb{Z}_{2}^{\mathrm{O}} a ∈ Z 2 O and I ⊆ O I\subseteq\mathrm{O} I ⊆ O , write a ∣ I : = ( a i ) i ∈ I ∈ Z 2 I a|_{I}:=(a_i)_{i\in I}\in\mathbb{Z}_{2}^{I} a ∣ I := ( a i ) i ∈ I ∈ Z 2 I . For every T ⊆ T T\subseteq\mathsf{T} T ⊆ T and r , r ′ ∈ Z 2 O r,r'\in\mathbb{Z}_{2}^{\mathrm{O}} r , r ′ ∈ Z 2 O , Eqs. (C.31) and (C.51), specialized to O \mathrm{O} O , give
( E T O O ∘ L T O O ) ( ∣ r ⟩ ⟨ r ′ ∣ O ) = δ r ∣ T O , r ′ ∣ T O ∣ θ T ( r ) ⟩ ⟨ θ T ( r ′ ) ∣ O . (185) \left(\mathcal{E}^{\mathrm{O}}_{T_{\mathrm{O}}}\circ\mathcal{L}^{\mathrm{O}}_{T_{\mathrm{O}}}\right)\left(|r\rangle\langle r'|_{\mathrm{O}}\right)
=\delta_{r|_{T_{\mathrm{O}}},r'|_{T_{\mathrm{O}}}}|\theta_{T}(r)\rangle\langle\theta_{T}(r')|_{\mathrm{O}}. \tag*{(185)} ( E T O O ∘ L T O O ) ( ∣ r ⟩ ⟨ r ′ ∣ O ) = δ r ∣ T O , r ′ ∣ T O ∣ θ T ( r )⟩ ⟨ θ T ( r ′ ) ∣ O . (185)
Suppose that s ≠ s ′ s\ne s' s = s ′ . Equation (E.44) and injectivity of E E E give r e , u ( s ) ≠ r e , u ( s ′ ) r_{e,u}(s)\ne r_{e,u}(s') r e , u ( s ) = r e , u ( s ′ ) for every e ∈ V K T e\in V_{K_{\mathsf{T}}} e ∈ V K T and u ∈ V R i n T u\in V^{\mathsf{T}}_{R_{\mathrm{in}}} u ∈ V R in T . If their restrictions to T O T_{\mathrm{O}} T O differ, then the Kronecker delta in Eq. (E.50) vanishes. Otherwise, their restrictions to O ∖ T O \mathrm{O}\setminus T_{\mathrm{O}} O ∖ T O differ, so that θ T ( r e , u ( s ) ) ≠ θ T ( r e , u ( s ′ ) ) \theta_{T}(r_{e,u}(s))\ne\theta_{T}(r_{e,u}(s')) θ T ( r e , u ( s )) = θ T ( r e , u ( s ′ )) . Equations (D.12) and (D.14) therefore show that M θ \mathcal{M}_{\theta} M θ annihilates the corresponding term for every θ ∈ Θ \theta\in\Theta θ ∈ Θ . Thus, every summand in Eq. (E.48) vanishes when s ≠ s ′ s\ne s' s = s ′ .
Now, suppose s = s ′ s=s' s = s ′ . In this case, r e , u ( s ) = r e , u ( s ′ ) r_{e,u}(s)=r_{e,u}(s') r e , u ( s ) = r e , u ( s ′ ) , the phase in Eq. (E.48) is one, and Eqs. (E.50), (D.12), and (D.14) contribute the factor δ θ , θ T ( r e , u ( s ) ) \delta_{\theta,\theta_{T}(r_{e,u}(s))} δ θ , θ T ( r e , u ( s )) . Lemma C.32, specialized to A, shows that L T A A \mathcal{L}^{\mathrm{A}}_{T_{\mathrm{A}}} L T A A is trace preserving, and Eq. (C.52), specialized to A, shows that E T A A \mathcal{E}^{\mathrm{A}}_{T_{\mathrm{A}}} E T A A is trace preserving. Since W K A ( ( v e , u ) A ) \mathcal{W}^{K\mathrm{A}}((v_{e,u})_{\mathrm{A}}) W K A (( v e , u ) A ) is unitary and ∣ α ( s ) ⟩ A |\alpha(s)\rangle_{\mathrm{A}} ∣ α ( s ) ⟩ A is a unit vector, the traced A factor is tr A ( ∣ α ( s ) ⟩ ⟨ α ( s ) ∣ A ) = 1 \operatorname{tr}_{\mathrm{A}}(|\alpha(s)\rangle\langle\alpha(s)|_{\mathrm{A}})=1 tr A ( ∣ α ( s )⟩ ⟨ α ( s ) ∣ A ) = 1 . Combining the cases s ≠ s ′ s\ne s' s = s ′ and s = s ′ s=s' s = s ′ and substituting Eqs. (E.49) and (E.50) into Eq. (E.48), with Eq. (D.14), gives
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ∑ T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V U R i n T q e T , U R i n 4 ∣ U R i n ∣ δ θ , θ T ( r e , u ( s ) ) [ C F ∘ F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ E T D D ∘ L T D D ∘ W K D ( ( v e , u ) D ) ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (186) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X \otimes|s\rangle\langle s'|_{S}\right)
= \delta_{s,s'} \sum_{T\subseteq\mathcal{T}} \sum_{\theta\in\Theta} \sum_{e\in V_{K_T}} \sum_{u\in V^{T}_{U_{R_{\mathrm{in}}}}} \frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4|U_{R_{\mathrm{in}}}|}\delta_{\theta,\theta_T(r_{e,u}(s))}
\left[
\mathcal{C}_{F}\circ\mathcal{F}^{D}_{R_{\mathrm{out}}}\circ\mathcal{P}^{D}_{R_{\mathrm{out}}}\circ\mathcal{R}_{K(\theta)}
\circ\mathcal{W}^{D}(D(\theta))\circ\mathcal{E}^{D}_{T_D}\circ\mathcal{L}^{D}_{T_D}
\circ\mathcal{W}^{K_D}((v_{e,u})_{D})\circ\mathcal{C}_{F}^{-1}
\right]\left(X\otimes|s\rangle\langle s|_{S}\right). \tag*{(186)} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ T ⊆ T ∑ θ ∈ Θ ∑ e ∈ V K T ∑ u ∈ V U R in T ∑ 4∣ U R in ∣ q e T , U R in δ θ , θ T ( r e , u ( s )) [ C F ∘ F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ E T D D ∘ L T D D ∘ W K D (( v e , u ) D ) ∘ C F − 1 ] ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (186)
For every T ⊆ T T\subseteq\mathcal{T} T ⊆ T , θ ∈ Θ \theta\in\Theta θ ∈ Θ , e ∈ V K T e\in V_{K_T} e ∈ V K T , and u ∈ V U R i n T u\in V^{T}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , we have that
F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ W D ( D ( θ ) ) ∘ E T D D ∘ L T D D ∘ W K D ( ( v e , u ) D ) = F R o u t D ∘ P R o u t D ∘ R K ( θ ) ∘ E T D D ∘ L T D D ∘ W K D ( ( v e , u ) D + D ( θ ) ) = δ R o u t , T D ∖ K ( θ ) G T D ∪ K ( θ ) D ∘ W K D ( ( v e , u ) D + D ( θ ) ) , (187) \mathcal{F}^{D}_{R_{\mathrm{out}}}\circ\mathcal{P}^{D}_{R_{\mathrm{out}}}\circ\mathcal{R}_{K(\theta)}\circ\mathcal{W}^{D}(D(\theta))\circ\mathcal{E}^{D}_{T_D}\circ\mathcal{L}^{D}_{T_D}\circ\mathcal{W}^{K_D}((v_{e,u})_{D})
= \mathcal{F}^{D}_{R_{\mathrm{out}}}\circ\mathcal{P}^{D}_{R_{\mathrm{out}}}\circ\mathcal{R}_{K(\theta)}\circ\mathcal{E}^{D}_{T_D}\circ\mathcal{L}^{D}_{T_D}\circ\mathcal{W}^{K_D}((v_{e,u})_{D}+D(\theta))
= \delta_{R_{\mathrm{out}},T_D\setminus K(\theta)}\mathcal{G}^{D}_{T_D\cup K(\theta)}\circ\mathcal{W}^{K_D}((v_{e,u})_{D}+D(\theta)), \tag*{(187)} F R out D ∘ P R out D ∘ R K ( θ ) ∘ W D ( D ( θ )) ∘ E T D D ∘ L T D D ∘ W K D (( v e , u ) D ) = F R out D ∘ P R out D ∘ R K ( θ ) ∘ E T D D ∘ L T D D ∘ W K D (( v e , u ) D + D ( θ )) = δ R out , T D ∖ K ( θ ) G T D ∪ K ( θ ) D ∘ W K D (( v e , u ) D + D ( θ )) , (187)
where Eqs. (C.62), (C.46), (C.43), (A.15), and (G.5), specialized to D D D , give the first equality, while Eqs. (D.18), (C.31), (C.51), (C.56), (C.35), and (C.36) give the second.
For T ⊆ T T\subseteq\mathcal{T} T ⊆ T , e ∈ V K T e\in V_{K_T} e ∈ V K T , u ∈ V U R i n T u\in V^{T}_{U_{R_{\mathrm{in}}}} u ∈ V U R in T , and t ∈ Z 2 S t\in\mathbb{Z}_{2}^{S} t ∈ Z 2 S , define
R T , v e , u ( t ) : = T D ∖ K ( θ T ( r e , u ( t ) ) ) (188) R_{T,v_{e,u}}(t):=T_D\setminus K(\theta_T(r_{e,u}(t))) \tag*{(188)} R T , v e , u ( t ) := T D ∖ K ( θ T ( r e , u ( t ))) (188)
Λ T , v e , u ( t ) : = T D ∪ K ( θ T ( r e , u ( t ) ) ) , (189) \Lambda_{T,v_{e,u}}(t):=T_D\cup K(\theta_T(r_{e,u}(t))), \tag*{(189)} Λ T , v e , u ( t ) := T D ∪ K ( θ T ( r e , u ( t ))) , (189)
ν T , v e , u ( t ) : = ( v e , u ) D + D ( θ T ( r e , u ( t ) ) ) . (190) \nu_{T,v_{e,u}}(t):=(v_{e,u})_{D}+D(\theta_T(r_{e,u}(t))). \tag*{(190)} ν T , v e , u ( t ) := ( v e , u ) D + D ( θ T ( r e , u ( t ))) . (190)
Substituting Eq. (E.52) termwise into Eq. (E.51), the factor δ θ , θ T ( r e , u ( s ) ) \delta_{\theta,\theta_T(r_{e,u}(s))} δ θ , θ T ( r e , u ( s )) restricts the sum over θ \theta θ to θ = θ T ( r e , u ( s ) ) \theta=\theta_T(r_{e,u}(s)) θ = θ T ( r e , u ( s )) . Equation (E.53) therefore gives
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ∑ T ⊆ T , e ∈ V K T , u ∈ V U R i n T R T , v e , u ( s ) = R o u t q e T , U R i n 4 ∣ U R i n ∣ ( C F ∘ G Λ T , v e , u ( s ) D ∘ W K D ( ν T , v e , u ( s ) ) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (191) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X\otimes|s\rangle\langle s'|_{S}\right)
= \delta_{s,s'}\sum_{\substack{T\subseteq\mathcal{T},\ e\in V_{K_T},\ u\in V^{T}_{U_{R_{\mathrm{in}}}}\\ R_{T,v_{e,u}}(s)=R_{\mathrm{out}}}} \frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4|U_{R_{\mathrm{in}}}|}\left(\mathcal{C}_{F}\circ\mathcal{G}^{D}_{\Lambda_{T,v_{e,u}}(s)}\circ\mathcal{W}^{K_D}(\nu_{T,v_{e,u}}(s))\circ\mathcal{C}_{F}^{-1}\right)\left(X\otimes|s\rangle\langle s|_{S}\right). \tag*{(191)} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ T ⊆ T , e ∈ V K T , u ∈ V U R in T R T , v e , u ( s ) = R out ∑ 4∣ U R in ∣ q e T , U R in ( C F ∘ G Λ T , v e , u ( s ) D ∘ W K D ( ν T , v e , u ( s )) ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (191)
For every t ∈ Z 2 S t\in\mathbb{Z}_{2}^{S} t ∈ Z 2 S , define the map
B t R o u t , R i n : End ( K D ) → End ( K D ) \mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{t}:\operatorname{End}(\mathcal{K}_{D})\to\operatorname{End}(\mathcal{K}_{D}) B t R out , R in : End ( K D ) → End ( K D )
according to
B t R o u t , R i n : = ∑ T ⊆ T , e ∈ V K T , u ∈ V U R i n T R T , v e , u ( t ) = R o u t q e T , U R i n 4 ∣ U R i n ∣ G Λ T , v e , u ( t ) D ∘ W K D ( ν T , v e , u ( t ) ) . (192) \mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{t}:=\sum_{\substack{T\subseteq\mathcal{T},\ e\in V_{K_T},\ u\in V^{T}_{U_{R_{\mathrm{in}}}}\\ R_{T,v_{e,u}}(t)=R_{\mathrm{out}}}} \frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4|U_{R_{\mathrm{in}}}|}\mathcal{G}^{D}_{\Lambda_{T,v_{e,u}}(t)}\circ\mathcal{W}^{K_D}(\nu_{T,v_{e,u}}(t)). \tag*{(192)} B t R out , R in := T ⊆ T , e ∈ V K T , u ∈ V U R in T R T , v e , u ( t ) = R out ∑ 4∣ U R in ∣ q e T , U R in G Λ T , v e , u ( t ) D ∘ W K D ( ν T , v e , u ( t )) . (192)
For every Λ ⊆ D \Lambda\subseteq\mathrm{D} Λ ⊆ D and ν ∈ V Λ D \nu\in\mathrm{V}_{\Lambda}^{\mathrm{D}} ν ∈ V Λ D , Eq. (C.39), specialized to D \mathrm{D} D , and Eqs. (A.15) and (G.5) give
G Λ D ∘ W K D ( ν ) = 1 4 ∣ Λ ∣ ∑ m ∈ V Λ D W K D ( m ) ∘ W K D ( ν ) = 1 4 ∣ Λ ∣ ∑ m ∈ V Λ D W K D ( m + ν ) . (193) \mathcal{G}_{\Lambda}^{\mathrm{D}} \circ\mathcal{W}^{\mathrm{KD}}(\nu) = \frac{1}{4^{|\Lambda|}} \sum_{m \in\mathrm{V}_{\Lambda}^{\mathrm{D}}} \mathcal{W}^{\mathrm{KD}}(m) \circ\mathcal{W}^{\mathrm{KD}}(\nu) = \frac{1}{4^{|\Lambda|}} \sum_{m \in\mathrm{V}_{\Lambda}^{\mathrm{D}}} \mathcal{W}^{\mathrm{KD}}(m+\nu). \tag*{(193)} G Λ D ∘ W KD ( ν ) = 4 ∣Λ∣ 1 m ∈ V Λ D ∑ W KD ( m ) ∘ W KD ( ν ) = 4 ∣Λ∣ 1 m ∈ V Λ D ∑ W KD ( m + ν ) . (193)
Since ∣ V Λ D ∣ = 4 ∣ Λ ∣ |\mathrm{V}_{\Lambda}^{\mathrm{D}}| = 4^{|\Lambda|} ∣ V Λ D ∣ = 4 ∣Λ∣ , Definition A.54 implies that G Λ D ∘ W K D ( ν ) \mathcal{G}_{\Lambda}^{\mathrm{D}} \circ\mathcal{W}^{\mathrm{KD}}(\nu) G Λ D ∘ W KD ( ν ) is a Pauli channel. Thus, Eq. (E.55) shows that every B t R o u t , R i n \mathcal{B}_{t}^{R_{\mathrm{out}},R_{\mathrm{in}}} B t R out , R in is a Pauli map. Equations (E.54) and (E.55) give
Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ( C F ∘ B s R o u t , R i n ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (194) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right) = \delta_{s,s'}\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right)\left(X \otimes|s\rangle\langle s|_{\mathrm{S}}\right). \tag*{(194)} Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ( C F ∘ B s R out , R in ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) . (194)
For every t ∈ Z 2 S t \in\mathbb{Z}_{2}^{\mathrm{S}} t ∈ Z 2 S , Definition E.23 gives
Z t ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ δ t , s X ⊗ ∣ s ⟩ ⟨ s ∣ S . (195) \mathcal{Z}_{t}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right) = \delta_{s,s'}\delta_{t,s}X \otimes|s\rangle\langle s|_{\mathrm{S}}. \tag*{(195)} Z t ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ δ t , s X ⊗ ∣ s ⟩ ⟨ s ∣ S . (195)
Applying Eq. (E.58) termwise gives
[ ∑ t ∈ Z 2 S C F ∘ B t R o u t , R i n ∘ C F − 1 ∘ Z t ] ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ( C F ∘ B s R o u t , R i n ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) = Φ ~ R o u t , R i n ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (196) \begin{aligned}
\left[\sum_{t \in\mathbb{Z}_{2}^{\mathrm{S}}} \mathcal{C}_{\mathrm{F}} \circ\mathcal{B}_{t}^{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{\mathrm{F}}^{-1} \circ\mathcal{Z}_{t}\right]\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right)
&= \delta_{s,s'}\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right)\left(X \otimes|s\rangle\langle s|_{\mathrm{S}}\right) \\
&= \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(X \otimes|s\rangle\langle s'|_{\mathrm{S}}\right). \tag*{(196)}
\end{aligned} t ∈ Z 2 S ∑ C F ∘ B t R out , R in ∘ C F − 1 ∘ Z t ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) = δ s , s ′ ( C F ∘ B s R out , R in ∘ C F − 1 ) ( X ⊗ ∣ s ⟩ ⟨ s ∣ S ) = Φ R out , R in ( X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S ) . (196)
where the last equality is Eq. (E.57). The operators X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S X \otimes|s\rangle\langle s'|_{\mathrm{S}} X ⊗ ∣ s ⟩ ⟨ s ′ ∣ S , with X ∈ End ( K L ) X \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) and s , s ′ ∈ Z 2 S s,s' \in\mathbb{Z}_{2}^{\mathrm{S}} s , s ′ ∈ Z 2 S , span End ( K L ⊗ K S ) \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) End ( K L ⊗ K S ) . Since R i n R_{\mathrm{in}} R in and R o u t R_{\mathrm{out}} R out were arbitrary, Eq. (E.59) proves Eq. (E.36) for every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D .
An explicit construction of the family of Pauli maps { B s R o u t , R i n } R i n , R o u t ⊆ D s ∈ Z 2 S \{\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\}_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}{}_{s\in\mathbb{Z}_{2}^{\mathrm{S}}} { B s R out , R in } R in , R out ⊆ D s ∈ Z 2 S appears in Eq. (E.55).
Corollary E.60 (Syndrome dephasing of Clifford-frame sector branches). For every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D ,
Φ ~ R o u t , R i n = Φ ~ R o u t , R i n ∘ Z S . (197) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} = \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{Z}_{\mathrm{S}}. \tag*{(197)} Φ R out , R in = Φ R out , R in ∘ Z S . (197)
Proof. For every s , t ∈ Z 2 S s,t \in\mathbb{Z}_{2}^{\mathrm{S}} s , t ∈ Z 2 S , Definition E.23 gives
Z s ∘ Z t = δ s , t Z t . \mathcal{Z}_{s} \circ\mathcal{Z}_{t} = \delta_{s,t}\mathcal{Z}_{t}. Z s ∘ Z t = δ s , t Z t .
Right composing Eq. (E.36) by Z t \mathcal{Z}_{t} Z t therefore gives
Φ ~ R o u t , R i n ∘ Z t = C F ∘ B t R o u t , R i n ∘ C F − 1 ∘ Z t . \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{Z}_{t} = \mathcal{C}_{\mathrm{F}} \circ\mathcal{B}_{t}^{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{C}_{\mathrm{F}}^{-1} \circ\mathcal{Z}_{t}. Φ R out , R in ∘ Z t = C F ∘ B t R out , R in ∘ C F − 1 ∘ Z t .
Summing this equality over t t t , and applying Eq. (E.36) and Definition E.25, gives Eq. (E.61).
Corollary E.62. Let { B s R o u t , R i n } R i n , R o u t ⊆ D s ∈ Z 2 S \{\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\}_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}{}_{s\in\mathbb{Z}_{2}^{\mathrm{S}}} { B s R out , R in } R in , R out ⊆ D s ∈ Z 2 S be the family of Pauli maps constructed in the proof of Proposition E.35. Every B s R o u t , R i n \mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} B s R out , R in is a subnormalized Pauli map, and for every
R i n ⊆ D R_{\mathrm{in}} \subseteq\mathrm{D} R in ⊆ D and s ∈ Z 2 S s \in\mathbb{Z}_2^{\mathrm{S}} s ∈ Z 2 S , we have that
∑ R o u t ⊆ D wt ( B s R o u t , R i n ) = 1. (198) \sum_{R_{\mathrm{out}}\subseteq\mathrm{D}} \operatorname{wt}\left(\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\right)=1. \tag*{(198)} R out ⊆ D ∑ wt ( B s R out , R in ) = 1. (198)
Proof. Fix R i n ⊆ D R_{\mathrm{in}} \subseteq\mathrm{D} R in ⊆ D and s ∈ Z 2 S s \in\mathbb{Z}_2^{\mathrm{S}} s ∈ Z 2 S . Use the coefficients q e T , U R i n q_{e}^{T,U_{R_{\mathrm{in}}}} q e T , U R in , the labels v e , u v_{e,u} v e , u , and the sets and labels R T , v e , u ( s ) R_{T,v_{e,u}}(s) R T , v e , u ( s ) , Λ T , v e , u ( s ) \Lambda_{T,v_{e,u}}(s) Λ T , v e , u ( s ) , and ν T , v e , u ( s ) \nu_{T,v_{e,u}}(s) ν T , v e , u ( s ) defined in Eqs. (E.38), (E.41), and (E.53). Each D T , U R i n \mathcal{D}_{T,U_{R_{\mathrm{in}}}} D T , U R in is a subnormalized Pauli map by the standing assumption in Appendix E. Thus, Eq. (A.57) gives q e T , U R i n ≥ 0 q_{e}^{T,U_{R_{\mathrm{in}}}} \ge0 q e T , U R in ≥ 0 for every T ⊆ T T \subseteq\mathsf{T} T ⊆ T and e ∈ V K T e \in V_{K_{\mathrm{T}}} e ∈ V K T . Moreover, Eqs. (A.50) and (E.2)—together with U = U R i n U=U_{R_{\mathrm{in}}} U = U R in —give
∑ T ⊆ T ∑ e ∈ V K T q e T , U R i n = ∑ T ⊆ T wt ( D T , U R i n ) = 1. (199) \sum_{T\subseteq\mathsf{T}}\sum_{e\in V_{K_{\mathrm{T}}}}q_{e}^{T,U_{R_{\mathrm{in}}}}=\sum_{T\subseteq\mathsf{T}}\operatorname{wt}\left(\mathcal{D}_{T,U_{R_{\mathrm{in}}}}\right)=1. \tag*{(199)} T ⊆ T ∑ e ∈ V K T ∑ q e T , U R in = T ⊆ T ∑ wt ( D T , U R in ) = 1. (199)
These inequalities, along with Eqs. (E.55) and (E.56), establish that every B s R o u t , R i n \mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} B s R out , R in has nonnegative Pauli coefficients. Combining Eqs. (E.55) and (E.56) with linearity of the weight functional wt ( ⋅ ) \operatorname{wt}(\cdot) wt ( ⋅ ) leads to
wt ( B s R o u t , R i n ) = ∑ T ⊆ T , e ∈ V K T , u ∈ V U R i n R T , v e , u ( s ) = R o u t q e T , U R i n 4 ∣ U R i n ∣ . (200) \operatorname{wt}\left(\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\right)=\sum_{\substack{T\subseteq\mathsf{T},\,e\in V_{K_{\mathrm{T}},\,u\in V_{U_{R_{\mathrm{in}}}}\\R_{T,v_{e,u}}(s)=R_{\mathrm{out}}}}\frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4^{|U_{R_{\mathrm{in}}}|}}.} \tag*{(200)} wt ( B s R out , R in ) = T ⊆ T , e ∈ V K T , u ∈ V U R in R T , v e , u ( s ) = R out 4 ∣ U R in ∣ q e T , U R in . ∑ (200)
Each triple ( T , e , u ) (T,e,u) ( T , e , u ) determines exactly one value of R T , v e , u ( s ) R_{T,v_{e,u}}(s) R T , v e , u ( s ) . Since ∣ V U R i n T ∣ = 4 ∣ U R i n ∣ \lvert V_{U_{R_{\mathrm{in}}}}^{\mathrm{T}}\rvert=4^{|U_{R_{\mathrm{in}}}|} ∣ V U R in T ∣ = 4 ∣ U R in ∣ , summing Eq. (E.65) over R o u t ⊆ D R_{\mathrm{out}}\subseteq\mathrm{D} R out ⊆ D gives
∑ R o u t ⊆ D wt ( B s R o u t , R i n ) = ∑ T ⊆ T ∑ e ∈ V K T ∑ u ∈ V U R i n T q e T , U R i n 4 ∣ U R i n ∣ = ∑ T ⊆ T ∑ e ∈ V K T q e T , U R i n = 1 , \sum_{R_{\mathrm{out}}\subseteq\mathrm{D}}\operatorname{wt}\left(\mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}}\right)
=\sum_{T\subseteq\mathsf{T}}\sum_{e\in V_{K_{\mathrm{T}}}}\sum_{u\in V_{U_{R_{\mathrm{in}}}}^{\mathrm{T}}}\frac{q_{e}^{T,U_{R_{\mathrm{in}}}}}{4^{|U_{R_{\mathrm{in}}}|}}
=\sum_{T\subseteq\mathsf{T}}\sum_{e\in V_{K_{\mathrm{T}}}}q_{e}^{T,U_{R_{\mathrm{in}}}}=1, R out ⊆ D ∑ wt ( B s R out , R in ) = T ⊆ T ∑ e ∈ V K T ∑ u ∈ V U R in T ∑ 4 ∣ U R in ∣ q e T , U R in = T ⊆ T ∑ e ∈ V K T ∑ q e T , U R in = 1 ,
where the last equality is Eq. (E.64). By Eq. (A.50), the sum of the Pauli coefficients of each map B s R o u t , R i n \mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} B s R out , R in equals its weight. Since those coefficients are nonnegative and each weight is at most one, every B s R o u t , R i n \mathcal{B}_{s}^{R_{\mathrm{out}},R_{\mathrm{in}}} B s R out , R in is a subnormalized Pauli map, by Definition A.54. □
Complete-cycle logical transition form
Definition E.66 (Syndrome transition maps). For every s i n , s o u t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^{\mathrm{S}} s in , s out ∈ Z 2 S , define the map S s o u t , s i n : End ( K S ) → End ( K S ) \mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}:\operatorname{End}(\mathcal{K}_{\mathrm{S}})\to\operatorname{End}(\mathcal{K}_{\mathrm{S}}) S s out , s in : End ( K S ) → End ( K S ) by
S s o u t , s i n ( Y ) : = ⟨ s i n ∣ Y ∣ s i n ⟩ S ∣ s o u t ⟩ ⟨ s o u t ∣ S , (201) \mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}(Y):=\langle s_{\mathrm{in}}|Y|s_{\mathrm{in}}\rangle_{\mathrm{S}}|s_{\mathrm{out}}\rangle\langle s_{\mathrm{out}}|_{\mathrm{S}}, \tag*{(201)} S s out , s in ( Y ) := ⟨ s in ∣ Y ∣ s in ⟩ S ∣ s out ⟩ ⟨ s out ∣ S , (201) for all $Y\in\operatorname{End}(\mathcal{K}_{\mathrm{S}}).
Proposition E.68. There exists a family
\left\{\mathcal{J}_{s_{\mathrm{out},s_{\mathrm{in}}}^{R_{\mathrm{out}},R_{\mathrm{in}}}\right\}_{\substack{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}\\s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^{\mathrm{S}}}}}
of subnormalized Pauli maps on End ( K L ) \operatorname{End}(\mathcal{K}_{\mathrm{L}}) End ( K L ) such that, for every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D ,
Φ ~ R o u t , R i n = ∑ s i n , s o u t ∈ Z 2 S J s o u t , s i n R o u t , R i n ⊗ S s o u t , s i n . (202) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}=\sum_{s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_{2}^{S}}\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\otimes\mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}. \tag*{(202)} Φ R out , R in = s in , s out ∈ Z 2 S ∑ J s out , s in R out , R in ⊗ S s out , s in . (202)
Proof. Fix R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D} R in , R out ⊆ D . For every s i n ∈ Z 2 S s_{\mathrm{in}}\in\mathbb{Z}_{2}^{S} s in ∈ Z 2 S , denote by
{ b q , s i n R o u t , R i n } q ∈ V K D \left\{b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\right\}_{q\in V_{\mathrm{K}_{\mathrm{D}}}} { b q , s in R out , R in } q ∈ V K D
the unique coefficient family of B s i n R o u t , R i n \mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{in}}} B s in R out , R in from Lemma A.42, specialized to K D \mathcal{K}_{\mathrm{D}} K D . Corollary E.62 and Eq. (A.57) imply that b q , s i n R o u t , R i n ≥ 0 b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\geq0 b q , s in R out , R in ≥ 0 for every q ∈ V K D q\in V_{\mathrm{K}_{\mathrm{D}}} q ∈ V K D . Then, the relations
B s i n R o u t , R i n = ∑ q ∈ V K D b q , s i n R o u t , R i n W K D ( q ) , (203) \mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{in}}}=\sum_{q\in V_{\mathrm{K}_{\mathrm{D}}}}b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\mathcal{W}^{\mathrm{K}_{\mathrm{D}}}(q), \tag*{(203)} B s in R out , R in = q ∈ V K D ∑ b q , s in R out , R in W K D ( q ) , (203)
∑ q ∈ V K D b q , s i n R o u t , R i n ≤ 1 (204) \sum_{q\in V_{\mathrm{K}_{\mathrm{D}}}}b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\leq1 \tag*{(204)} q ∈ V K D ∑ b q , s in R out , R in ≤ 1 (204)
hold. Renaming the summation index s s s in Eq. (E.36) as s i n s_{\mathrm{in}} s in , substituting Eq. (E.70) for B s i n R o u t , R i n \mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{in}}} B s in R out , R in in each summand, and distributing the composition over the resulting sum give
Φ ~ R o u t , R i n = ∑ s i n ∈ Z 2 S ∑ q ∈ V K D b q , s i n R o u t , R i n C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s i n . (205) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}=\sum_{s_{\mathrm{in}}\in\mathbb{Z}_{2}^{S}}\sum_{q\in V_{\mathrm{K}_{\mathrm{D}}}}b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\mathcal{C}_{F}\circ\mathcal{W}^{\mathrm{K}_{\mathrm{D}}}(q)\circ\mathcal{C}_{F}^{-1}\circ\mathcal{Z}_{s_{\mathrm{in}}}. \tag*{(205)} Φ R out , R in = s in ∈ Z 2 S ∑ q ∈ V K D ∑ b q , s in R out , R in C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s in . (205)
By the Clifford assumption on C F \mathcal{C}_{F} C F , fix a symplectic isomorphism κ = ( κ L , κ S ) : V K D → V L × V S \kappa=(\kappa_{\mathrm{L}},\kappa_{\mathrm{S}}):V_{\mathrm{K}_{\mathrm{D}}}\to V_{\mathrm{L}}\times V_{\mathrm{S}} κ = ( κ L , κ S ) : V K D → V L × V S such that, for every q ∈ V K D q\in V_{\mathrm{K}_{\mathrm{D}}} q ∈ V K D ,
C F ∘ W K D ( q ) ∘ C F − 1 = W L ( κ L ( q ) ) ⊗ W S ( κ S ( q ) ) . (206) \mathcal{C}_{F}\circ\mathcal{W}^{\mathrm{K}_{\mathrm{D}}}(q)\circ\mathcal{C}_{F}^{-1}=\mathcal{W}^{\mathrm{L}}(\kappa_{\mathrm{L}}(q))\otimes\mathcal{W}^{\mathrm{S}}(\kappa_{\mathrm{S}}(q)). \tag*{(206)} C F ∘ W K D ( q ) ∘ C F − 1 = W L ( κ L ( q )) ⊗ W S ( κ S ( q )) . (206)
Next, define the map x : V S → Z 2 S \mathbf{x}:V_{\mathrm{S}}\to\mathbb{Z}_{2}^{S} x : V S → Z 2 S by x ( x , z ) : = x \mathbf{x}(x,z):=x x ( x , z ) := x . For every s i n ∈ Z 2 S s_{\mathrm{in}}\in\mathbb{Z}_{2}^{S} s in ∈ Z 2 S and q ∈ V K D q\in V_{\mathrm{K}_{\mathrm{D}}} q ∈ V K D ,
C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s i n = W L ( κ L ( q ) ) ⊗ S s i n + x ( κ S ( q ) ) , s i n . (207) \mathcal{C}_{F}\circ\mathcal{W}^{\mathrm{K}_{\mathrm{D}}}(q)\circ\mathcal{C}_{F}^{-1}\circ\mathcal{Z}_{s_{\mathrm{in}}}=\mathcal{W}^{\mathrm{L}}(\kappa_{\mathrm{L}}(q))\otimes\mathcal{S}_{s_{\mathrm{in}}+\mathbf{x}(\kappa_{\mathrm{S}}(q)),s_{\mathrm{in}}}. \tag*{(207)} C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s in = W L ( κ L ( q )) ⊗ S s in + x ( κ S ( q )) , s in . (207)
To verify Eq. (E.74), Eqs. (A.10) and (A.13), specialized to S \mathrm{S} S , give W S ( r ) ( ∣ s ⟩ ⟨ s ∣ S ) = ∣ s + x ( r ) ⟩ ⟨ s + x ( r ) ∣ S \mathcal{W}^{\mathrm{S}}(r)(|s\rangle\langle s|_{\mathrm{S}})=|s+\mathbf{x}(r)\rangle\langle s+\mathbf{x}(r)|_{\mathrm{S}} W S ( r ) ( ∣ s ⟩ ⟨ s ∣ S ) = ∣ s + x ( r )⟩ ⟨ s + x ( r ) ∣ S for every r ∈ V S r\in V_{\mathrm{S}} r ∈ V S and s ∈ Z 2 S s\in\mathbb{Z}_{2}^{S} s ∈ Z 2 S . This equality and Eqs. (E.73), (E.24), and (E.67) give, for every X ∈ End ( K L ) X\in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X ∈ End ( K L ) and Y ∈ End ( K S ) Y\in\operatorname{End}(\mathcal{K}_{\mathrm{S}}) Y ∈ End ( K S ) ,
( C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s i n ) ( X ⊗ Y ) = ⟨ s i n ∣ Y ∣ s i n ⟩ S W L ( κ L ( q ) ) ( X ) ⊗ ∣ s i n + x ( κ S ( q ) ) ⟩ ⟨ s i n + x ( κ S ( q ) ) ∣ S = ( W L ( κ L ( q ) ) ⊗ S s i n + x ( κ S ( q ) ) , s i n ) ( X ⊗ Y ) . \begin{aligned}
\left(\mathcal{C}_{F}\circ\mathcal{W}^{\mathrm{K}_{\mathrm{D}}}(q)\circ\mathcal{C}_{F}^{-1}\circ\mathcal{Z}_{s_{\mathrm{in}}}\right)(X\otimes Y)
&=\langle s_{\mathrm{in}}|Y|s_{\mathrm{in}}\rangle_{\mathrm{S}}\mathcal{W}^{\mathrm{L}}(\kappa_{\mathrm{L}}(q))(X)\otimes|s_{\mathrm{in}}+\mathbf{x}(\kappa_{\mathrm{S}}(q))\rangle\langle s_{\mathrm{in}}+\mathbf{x}(\kappa_{\mathrm{S}}(q))|_{\mathrm{S}}\\
&=\left(\mathcal{W}^{\mathrm{L}}(\kappa_{\mathrm{L}}(q))\otimes\mathcal{S}_{s_{\mathrm{in}}+\mathbf{x}(\kappa_{\mathrm{S}}(q)),s_{\mathrm{in}}}\right)(X\otimes Y).
\end{aligned} ( C F ∘ W K D ( q ) ∘ C F − 1 ∘ Z s in ) ( X ⊗ Y ) = ⟨ s in ∣ Y ∣ s in ⟩ S W L ( κ L ( q )) ( X ) ⊗ ∣ s in + x ( κ S ( q ))⟩ ⟨ s in + x ( κ S ( q )) ∣ S = ( W L ( κ L ( q )) ⊗ S s in + x ( κ S ( q )) , s in ) ( X ⊗ Y ) .
Since simple tensors span End ( K L ⊗ K S ) \operatorname{End}(\mathcal{K}_{\mathrm{L}}\otimes\mathcal{K}_{\mathrm{S}}) End ( K L ⊗ K S ) , this proves Eq. (E.74).
For every s i n , s o u t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_{2}^{S} s in , s out ∈ Z 2 S , define the map J s o u t , s i n : End ( K L ) → End ( K L ) R o u t , R i n \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}:\operatorname{End}(\mathcal{K}_{\mathrm{L}})\to\operatorname{End}(\mathcal{K}_{\mathrm{L}})} J s out , s in : End ( K L ) → End ( K L ) R out , R in by
J s o u t , s i n : = ∑ q ∈ V K D s i n + x ( κ S ( q ) ) = s o u t b q , s i n R o u t , R i n W L ( κ L ( q ) ) . R o u t , R i n (208) \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out},s_{\mathrm{in}}}:=\sum_{\substack{q\in V_{\mathrm{K}_{\mathrm{D}}}\\s_{\mathrm{in}}+\mathbf{x}(\kappa_{\mathrm{S}}(q))=s_{\mathrm{out}}}}b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}}\mathcal{W}^{\mathrm{L}}(\kappa_{\mathrm{L}}(q)).} \tag*{(208)} J s out , s in := ∑ q ∈ V K D s in + x ( κ S ( q )) = s out b q , s in R out , R in W L ( κ L ( q )) . R out , R in (208)
Grouping the terms in Eq. (E.75) by κ L ( q ) \kappa_L(q) κ L ( q ) gives Pauli coefficients that are sums of the nonnegative coefficients b q , s i n R o u t , R i n b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}} b q , s in R out , R in . Their sum is at most ∑ q ∈ V K D b q , s i n R o u t , R i n \sum_{q\in V_{\mathrm{KD}}}b^{R_{\mathrm{out}},R_{\mathrm{in}}}_{q,s_{\mathrm{in}}} ∑ q ∈ V KD b q , s in R out , R in , which is at most one, by Eq. (E.71). Hence, every J s o u t , s i n R o u t , R i n \mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}} J s out , s in R out , R in is a subnormalized Pauli map, by Definition A.54.
Substituting Eq. (E.74) termwise into Eq. (E.72), grouping by the value of s i n + x ( κ S ( q ) ) s_{\mathrm{in}}+x(\kappa_S(q)) s in + x ( κ S ( q )) , and applying Eq. (E.75) give Eq. (E.69).
The members of the family { J s o u t , s i n R o u t , R i n } s i n , s o u t ∈ Z 2 S R i n , R o u t ⊆ D \left\{\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right\}^{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}_{s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S} { J s out , s in R out , R in } s in , s out ∈ Z 2 S R in , R out ⊆ D constructed in the foregoing proof are given explicitly by Eq. (E.75).
Corollary E.76. Denote by { J s o u t , s i n R o u t , R i n } s i n , s o u t ∈ Z 2 S R i n , R o u t ⊆ D \left\{\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right\}^{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}_{s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S} { J s out , s in R out , R in } s in , s out ∈ Z 2 S R in , R out ⊆ D the family constructed in the proof of Proposition E.68. For every R i n ⊆ D R_{\mathrm{in}}\subseteq\mathrm{D} R in ⊆ D and s i n ∈ Z 2 S s_{\mathrm{in}}\in\mathbb{Z}_2^S s in ∈ Z 2 S ,
∑ R o u t ⊆ D ∑ s o u t ∈ Z 2 S wt ( J s o u t , s i n R o u t , R i n ) = 1. (209) \sum_{R_{\mathrm{out}}\subseteq\mathrm{D}}\sum_{s_{\mathrm{out}}\in\mathbb{Z}_2^S}\operatorname{wt}\left(\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right)=1. \tag*{(209)} R out ⊆ D ∑ s out ∈ Z 2 S ∑ wt ( J s out , s in R out , R in ) = 1. (209)
Proof. Fix R i n ⊆ D R_{\mathrm{in}}\subseteq\mathrm{D} R in ⊆ D and s i n ∈ Z 2 S s_{\mathrm{in}}\in\mathbb{Z}_2^S s in ∈ Z 2 S . For each R o u t ⊆ D R_{\mathrm{out}}\subseteq\mathrm{D} R out ⊆ D , as s o u t s_{\mathrm{out}} s out varies over Z 2 S \mathbb{Z}_2^S Z 2 S , the conditions in Eq. (E.75) partition the summation indices in Eq. (E.70). Grouping terms with the same adjoint map in Eq. (E.75) does not change the total coefficient sum. Combining Eqs. (A.50), (E.70), and (E.75) and then applying Eq. (E.63) gives
∑ R o u t ⊆ D ∑ s o u t ∈ Z 2 S wt ( J s o u t , s i n R o u t , R i n ) = ∑ R o u t ⊆ D wt ( B s i n R o u t , R i n ) = 1. \sum_{R_{\mathrm{out}}\subseteq\mathrm{D}}\sum_{s_{\mathrm{out}}\in\mathbb{Z}_2^S}\operatorname{wt}\left(\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right)=\sum_{R_{\mathrm{out}}\subseteq\mathrm{D}}\operatorname{wt}\left(\mathcal{B}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{in}}}\right)=1. R out ⊆ D ∑ s out ∈ Z 2 S ∑ wt ( J s out , s in R out , R in ) = R out ⊆ D ∑ wt ( B s in R out , R in ) = 1.
For every linear map A : End ( K L ) → End ( K L ) \mathcal{A}:\operatorname{End}(\mathcal{K}_L)\to\operatorname{End}(\mathcal{K}_L) A : End ( K L ) → End ( K L ) and every s i n , s o u t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S s in , s out ∈ Z 2 S , Eqs. (E.24), (E.26), and (E.67) give
Z S ∘ ( A ⊗ S s o u t , s i n ) = A ⊗ S s o u t , s i n = ( A ⊗ S s o u t , s i n ) ∘ Z S . \mathcal{Z}_S\circ\left(\mathcal{A}\otimes\mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right)=\mathcal{A}\otimes\mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}=\left(\mathcal{A}\otimes\mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right)\circ\mathcal{Z}_S. Z S ∘ ( A ⊗ S s out , s in ) = A ⊗ S s out , s in = ( A ⊗ S s out , s in ) ∘ Z S .
For every R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D} R in , R out ⊆ D , applying this identity termwise to Eq. (E.69) gives
Φ ~ R o u t , R i n = Z S ∘ Φ ~ R o u t , R i n = Φ ~ R o u t , R i n ∘ Z S = Z S ∘ Φ ~ R o u t , R i n ∘ Z S . (210) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}=\mathcal{Z}_S\circ\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}=\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\circ\mathcal{Z}_S=\mathcal{Z}_S\circ\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\circ\mathcal{Z}_S. \tag*{(210)} Φ R out , R in = Z S ∘ Φ R out , R in = Φ R out , R in ∘ Z S = Z S ∘ Φ R out , R in ∘ Z S . (210)
Theorem 2. Let Φ \Phi Φ be the complete QEC-cycle channel, assumed to satisfy Eq. (E.1), and let { J s o u t , s i n R o u t , R i n } s i n , s o u t ∈ Z 2 S R i n , R o u t ⊆ D \left\{\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right\}^{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}}_{s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S} { J s out , s in R out , R in } s in , s out ∈ Z 2 S R in , R out ⊆ D be the family constructed in the proof of Proposition E.68. Then
Φ = ∑ R i n , R o u t ⊆ D s i n , s o u t ∈ Z 2 S E R o u t D ∘ L R o u t D ∘ C F − 1 ∘ ( J s o u t , s i n R o u t , R i n ⊗ S s o u t , s i n ) ∘ C F ∘ F R i n D ∘ P R i n D . (211) \Phi=\sum_{\substack{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}\\s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S}}\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}}\circ\mathcal{C}_{F}^{-1}\circ\left(\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\otimes\mathcal{S}_{s_{\mathrm{out}},s_{\mathrm{in}}}\right)\circ\mathcal{C}_{F}\circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}}\circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(211)} Φ = R in , R out ⊆ D s in , s out ∈ Z 2 S ∑ E R out D ∘ L R out D ∘ C F − 1 ∘ ( J s out , s in R out , R in ⊗ S s out , s in ) ∘ C F ∘ F R in D ∘ P R in D . (211)
Proof. Substituting Eq. (E.69) into Eq. (E.15) gives Eq. (E.79).
Leakage–syndrome-dephased operators and hidden-memory dynamics
Leakage–syndrome dephasing
Definition F.1 (Leakage–syndrome-dephasing map). Define the map D L S : End ( H D ) → End ( H D ) \mathcal{D}_{\mathrm{LS}} : \operatorname{End}(\mathcal{H}_{\mathrm{D}}) \to\operatorname{End}(\mathcal{H}_{\mathrm{D}}) D LS : End ( H D ) → End ( H D ) according to
D L S : = ∑ R ⊆ D E R D ∘ L R D ∘ C F − 1 ∘ Z S ∘ C F ∘ F R D ∘ P R D . (212) \mathcal{D}_{\mathrm{LS}} := \sum_{R \subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_{R} \circ\mathcal{L}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1} \circ\mathcal{Z}_{\mathrm{S}} \circ\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}. \tag*{(212)} D LS := R ⊆ D ∑ E R D ∘ L R D ∘ C F − 1 ∘ Z S ∘ C F ∘ F R D ∘ P R D . (212)
Lemma F.3. The map D L S \mathcal{D}_{\mathrm{LS}} D LS is completely positive and trace preserving (CPTP).
Proof. For every R ⊆ D R \subseteq\mathrm{D} R ⊆ D , the maps P R D \mathcal{P}^{\mathrm{D}}_{R} P R D and E R D \mathcal{E}^{\mathrm{D}}_{R} E R D are completely positive. Lemma C.32, unitarity of C F \mathcal{C}_{\mathrm{F}} C F , and Lemma E.29 show that the remaining factors in every summand of Eq. (F.2) are completely positive. Hence D L S \mathcal{D}_{\mathrm{LS}} D LS is completely positive.
Equation (C.52), specialized to D, Lemma C.32, together with unitarity of C F \mathcal{C}_{\mathrm{F}} C F and Lemma E.29, show that every factor in each summand other than P R D \mathcal{P}^{\mathrm{D}}_{R} P R D is trace preserving. Therefore, for every X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) , Eq. (E.3) and trace preservation of Δ D \Delta^{\mathrm{D}} Δ D give
tr [ D L S ( X ) ] = ∑ R ⊆ D tr [ P R D ( X ) ] = tr [ Δ D ( X ) ] = tr ( X ) . \operatorname{tr}[\mathcal{D}_{\mathrm{LS}}(X)] = \sum_{R \subseteq\mathrm{D}} \operatorname{tr}[\mathcal{P}^{\mathrm{D}}_{R}(X)] = \operatorname{tr}[\Delta^{\mathrm{D}}(X)] = \operatorname{tr}(X). tr [ D LS ( X )] = R ⊆ D ∑ tr [ P R D ( X )] = tr [ Δ D ( X )] = tr ( X ) .
Lemma F.4. For every R ⊆ D R \subseteq\mathrm{D} R ⊆ D , we have that
( C F ∘ F R D ∘ P R D ) ∘ D L S = Z S ∘ ( C F ∘ F R D ∘ P R D ) . (213) \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right) \circ\mathcal{D}_{\mathrm{LS}} = \mathcal{Z}_{\mathrm{S}} \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right). \tag*{(213)} ( C F ∘ F R D ∘ P R D ) ∘ D LS = Z S ∘ ( C F ∘ F R D ∘ P R D ) . (213) Proof. Fix R ⊆ D R \subseteq\mathrm{D} R ⊆ D and let G R D \mathcal{G}^{\mathrm{D}}_{R} G R D be the computational replacement channel, as in Definition C.34, specialized to D and R R R . Substituting Eq. (F.2) into ( C F ∘ F R D ∘ P R D ) ∘ D L S (\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}) \circ\mathcal{D}_{\mathrm{LS}} ( C F ∘ F R D ∘ P R D ) ∘ D LS and renaming its summation variable at T T T , we obtain a sum in which Eq. (E.19) applies with R i n = T R_{\mathrm{in}} = T R in = T . All summands with T ≠ R T \ne R T = R vanish, and the summand with T = R T = R T = R gives
( C F ∘ F R D ∘ P R D ) ∘ D L S = ( C F ∘ G R D ∘ C F − 1 ) ∘ Z S ∘ ( C F ∘ F R D ∘ P R D ) . (214) \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right) \circ\mathcal{D}_{\mathrm{LS}} = \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{G}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right) \circ\mathcal{Z}_{\mathrm{S}} \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right). \tag*{(214)} ( C F ∘ F R D ∘ P R D ) ∘ D LS = ( C F ∘ G R D ∘ C F − 1 ) ∘ Z S ∘ ( C F ∘ F R D ∘ P R D ) . (214) Conjugating the expansion in Eq. (C.39) by C F \mathcal{C}_{\mathrm{F}} C F and applying Eq. (E.73) show that C F ∘ G R D ∘ C F − 1 \mathcal{C}_{\mathrm{F}} \circ\mathcal{G}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1} C F ∘ G R D ∘ C F − 1 is a Pauli channel on End ( K L ⊗ K S ) \operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) End ( K L ⊗ K S ) . Equation (E.34) therefore gives
( C F ∘ G R D ∘ C F − 1 ) ∘ Z S = Z S ∘ ( C F ∘ G R D ∘ C F − 1 ) . (215) \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{G}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right) \circ\mathcal{Z}_{\mathrm{S}} = \mathcal{Z}_{\mathrm{S}} \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{G}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right). \tag*{(215)} ( C F ∘ G R D ∘ C F − 1 ) ∘ Z S = Z S ∘ ( C F ∘ G R D ∘ C F − 1 ) . (215) Equations (E.16), (C.35), and (C.33) give
( C F ∘ G R D ∘ C F − 1 ) ∘ ( C F ∘ F R D ∘ P R D ) = C F ∘ F R D ∘ L R D ∘ F R D ∘ P R D = C F ∘ F R D ∘ P R D . (216) \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{G}^{\mathrm{D}}_{R} \circ\mathcal{C}_{\mathrm{F}}^{-1}\right) \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right) = \mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{L}^{\mathrm{D}}_{R} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R} = \mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}. \tag*{(216)} ( C F ∘ G R D ∘ C F − 1 ) ∘ ( C F ∘ F R D ∘ P R D ) = C F ∘ F R D ∘ L R D ∘ F R D ∘ P R D = C F ∘ F R D ∘ P R D . (216) Combining Eqs. (F.6), (F.7), and (F.8) gives Eq. (F.5).
Corollary F.9. The map D L S \mathcal{D}_{\mathrm{LS}} D LS is idempotent:
D L S ∘ D L S = D L S . \mathcal{D}_{\mathrm{LS}} \circ\mathcal{D}_{\mathrm{LS}} = \mathcal{D}_{\mathrm{LS}}. D LS ∘ D LS = D LS .
Proof. For every s , t ∈ Z 2 s,t \in\mathbb{Z}_2 s , t ∈ Z 2 , Definition E.23 gives Z s ∘ Z t = δ s , t Z t \mathcal{Z}_s \circ\mathcal{Z}_t = \delta_{s,t}\mathcal{Z}_t Z s ∘ Z t = δ s , t Z t . Summing this equality over s , t ∈ Z 2 s,t \in\mathbb{Z}_2 s , t ∈ Z 2 and applying Eq. (E.26) give Z S 2 = Z S \mathcal{Z}_S^2 = \mathcal{Z}_S Z S 2 = Z S . Substituting Eq. (F.5) into Eq. (F.2) and using Z S 2 = Z S \mathcal{Z}_S^2 = \mathcal{Z}_S Z S 2 = Z S give
D L S ∘ D L S = ∑ R ⊆ D E R D ∘ L R D ∘ C F − 1 ∘ Z S 2 ∘ C F ∘ F R D ∘ P R D = D L S . \mathcal{D}_{\mathrm{LS}} \circ\mathcal{D}_{\mathrm{LS}} = \sum_{R \subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_R \circ\mathcal{L}^{\mathrm{D}}_R \circ\mathcal{C}^{-1}_{\mathrm{F}} \circ\mathcal{Z}_S^2 \circ\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R = \mathcal{D}_{\mathrm{LS}}. D LS ∘ D LS = R ⊆ D ∑ E R D ∘ L R D ∘ C F − 1 ∘ Z S 2 ∘ C F ∘ F R D ∘ P R D = D LS .
Definition F.10 (Leakage–syndrome-dephased operators). An operator X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) is leakage–syndrome dephased if D L S ( X ) = X \mathcal{D}_{\mathrm{LS}}(X) = X D LS ( X ) = X .
Corollary F.9 and Definition F.10 imply that D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) is leakage–syndrome dephased for every X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) .
Proposition F.11. The complete physical QEC-cycle channel Φ \Phi Φ satisfies
Φ = Φ ∘ D L S = D L S ∘ Φ . (217) \Phi= \Phi\circ\mathcal{D}_{\mathrm{LS}} = \mathcal{D}_{\mathrm{LS}} \circ\Phi. \tag*{(217)} Φ = Φ ∘ D LS = D LS ∘ Φ. (217) Moreover, Φ ( X ) \Phi(X) Φ ( X ) is leakage–syndrome dephased for every X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) .
Proof. Right composing Eq. (E.15) by D L S \mathcal{D}_{\mathrm{LS}} D LS and substituting Eq. (F.5) give
Φ ∘ D L S = ∑ R i n , R o u t ⊆ D E R o u t D ∘ L R o u t D ∘ C F − 1 ∘ Φ ~ R o u t , R i n ∘ Z S ∘ C F ∘ F R i n D ∘ P R i n D . (218) \Phi\circ\mathcal{D}_{\mathrm{LS}} = \sum_{R_{\mathrm{in}},R_{\mathrm{out}}\subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{L}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{C}^{-1}_{\mathrm{F}} \circ\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\mathcal{Z}_S \circ\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(218)} Φ ∘ D LS = R in , R out ⊆ D ∑ E R out D ∘ L R out D ∘ C F − 1 ∘ Φ R out , R in ∘ Z S ∘ C F ∘ F R in D ∘ P R in D . (218) Substituting Eq. (E.61) into Eq. (F.13) and applying Eq. (E.15) give Φ ∘ D L S = Φ \Phi\circ\mathcal{D}_{\mathrm{LS}} = \Phi Φ ∘ D LS = Φ . Equations (C.58), (E.12), (E.16), and (C.33) give, for every R i n , R o u t ⊆ D R_{\mathrm{in}}, R_{\mathrm{out}} \subseteq\mathrm{D} R in , R out ⊆ D ,
Φ ~ R o u t , R i n ∘ ( C F ∘ F R i n D ∘ P R i n D ) = ( C F ∘ F R o u t D ∘ P R o u t D ) ∘ Φ ∘ E R i n D ∘ P R i n D . (219) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}} \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}\right) = \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{out}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{out}}}\right) \circ\Phi\circ\mathcal{E}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}. \tag*{(219)} Φ R out , R in ∘ ( C F ∘ F R in D ∘ P R in D ) = ( C F ∘ F R out D ∘ P R out D ) ∘ Φ ∘ E R in D ∘ P R in D . (219) Fix R ⊆ D R \subseteq\mathrm{D} R ⊆ D . Summing Eq. (F.14) with R o u t = R R_{\mathrm{out}} = R R out = R over R i n ⊆ D R_{\mathrm{in}} \subseteq\mathrm{D} R in ⊆ D , applying Eq. (E.3), and using Φ ∘ Δ D = Φ \Phi\circ\Delta^{\mathrm{D}} = \Phi Φ ∘ Δ D = Φ from Eq. (E.5) give
( C F ∘ F R D ∘ P R D ) ∘ Φ = ∑ R i n ⊆ D Φ ~ R , R i n ∘ ( C F ∘ F R i n D ∘ P R i n D ) . (220) \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R\right) \circ\Phi= \sum_{R_{\mathrm{in}}\subseteq\mathrm{D}} \widetilde{\Phi}_{R,R_{\mathrm{in}}} \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R_{\mathrm{in}}} \circ\mathcal{P}^{\mathrm{D}}_{R_{\mathrm{in}}}\right). \tag*{(220)} ( C F ∘ F R D ∘ P R D ) ∘ Φ = R in ⊆ D ∑ Φ R , R in ∘ ( C F ∘ F R in D ∘ P R in D ) . (220) Left composing Eq. (F.15) by Z S \mathcal{Z}_S Z S and applying Eq. (E.78) give
Z S ∘ ( C F ∘ F R D ∘ P R D ) ∘ Φ = ( C F ∘ F R D ∘ P R D ) ∘ Φ , (221) \mathcal{Z}_S \circ\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R\right) \circ\Phi= \left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R\right) \circ\Phi, \tag*{(221)} Z S ∘ ( C F ∘ F R D ∘ P R D ) ∘ Φ = ( C F ∘ F R D ∘ P R D ) ∘ Φ , (221) for every R ⊆ D R \subseteq\mathrm{D} R ⊆ D . Substituting Eq. (F.16) into Eq. (F.2) and applying Eqs. (E.16), (C.33), and (E.3), together with Δ D ∘ Φ = Φ \Delta^{\mathrm{D}} \circ\Phi= \Phi Δ D ∘ Φ = Φ from Eq. (E.5), give
D L S ∘ Φ = ∑ R ⊆ D E R D ∘ L R D ∘ C F − 1 ∘ Z S ∘ C F ∘ F R D ∘ P R D ∘ Φ = ∑ R ⊆ D E R D ∘ L R D ∘ F R D ∘ P R D ∘ Φ = Δ D ∘ Φ = Φ . (222) \begin{aligned}
\mathcal{D}_{\mathrm{LS}} \circ\Phi
&= \sum_{R\subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_R \circ\mathcal{L}^{\mathrm{D}}_R \circ\mathcal{C}^{-1}_{\mathrm{F}} \circ\mathcal{Z}_S \circ\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R \circ\Phi\\
&= \sum_{R\subseteq\mathrm{D}} \mathcal{E}^{\mathrm{D}}_R \circ\mathcal{L}^{\mathrm{D}}_R \circ\mathcal{F}^{\mathrm{D}}_R \circ\mathcal{P}^{\mathrm{D}}_R \circ\Phi= \Delta^{\mathrm{D}} \circ\Phi= \Phi. \tag*{(222)}
\end{aligned} D LS ∘ Φ = R ⊆ D ∑ E R D ∘ L R D ∘ C F − 1 ∘ Z S ∘ C F ∘ F R D ∘ P R D ∘ Φ = R ⊆ D ∑ E R D ∘ L R D ∘ F R D ∘ P R D ∘ Φ = Δ D ∘ Φ = Φ. (222)
Combining Φ ∘ D L S = Φ \Phi\circ\mathcal{D}_{\mathrm{LS}}=\Phi Φ ∘ D LS = Φ with Eq. (F.17) gives Eq. (F.12). Equation (F.17) gives D L S ( Φ ( X ) ) = Φ ( X ) \mathcal{D}_{\mathrm{LS}}(\Phi(X))=\Phi(X) D LS ( Φ ( X )) = Φ ( X ) for every X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) . Definition F.10 therefore implies that Φ ( X ) \Phi(X) Φ ( X ) is leakage–syndrome dephased.
Logical blocks and standard form
Definition F.18 (Clifford-frame representatives). For every X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) and for every R ⊆ D R \subseteq\mathrm{D} R ⊆ D , define the operator X ~ R ∈ End ( K L ⊗ K S ) \widetilde{X}_{R} \in\operatorname{End}(\mathcal{K}_{\mathrm{L}} \otimes\mathcal{K}_{\mathrm{S}}) X R ∈ End ( K L ⊗ K S ) by
X ~ R : = ( C F ∘ F R D ∘ P R D ) ( X ) . (223) \widetilde{X}_{R}:=\left(\mathcal{C}_{\mathrm{F}} \circ\mathcal{F}^{\mathrm{D}}_{R} \circ\mathcal{P}^{\mathrm{D}}_{R}\right)(X). \tag*{(223)} X R := ( C F ∘ F R D ∘ P R D ) ( X ) . (223) Define the Clifford-frame family associated with X X X by { X ~ R } R ⊆ D \{\widetilde{X}_{R}\}_{R\subseteq\mathrm{D}} { X R } R ⊆ D .
Definition F.20 (Hidden-memory index). Define the hidden-memory index set by M : = Z 2 D × Z 2 S \mathrm{M}:=\mathbb{Z}_{2}^{\mathrm{D}} \times\mathbb{Z}_{2}^{\mathrm{S}} M := Z 2 D × Z 2 S . For m = ( R , s ) ∈ M m=(R,s)\in\mathrm{M} m = ( R , s ) ∈ M , write R ( m ) : = R R(m):=R R ( m ) := R and s ( m ) : = s s(m):=s s ( m ) := s .
Definition F.21 (Logical blocks). For X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) , let { X ~ R } R ⊆ D \{\widetilde{X}_{R}\}_{R\subseteq\mathrm{D}} { X R } R ⊆ D be the Clifford-frame family associated with X X X as in Definition F.18. For every m ∈ M m \in\mathrm{M} m ∈ M , define the logical block X m L ∈ End ( K L ) X^{\mathrm{L}}_{m} \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) X m L ∈ End ( K L ) by
X m L : = tr S [ Z s ( m ) ( X ~ R ( m ) ) ] . (224) X^{\mathrm{L}}_{m}:=\operatorname{tr}_{\mathrm{S}}\left[\mathcal{Z}_{s(m)}\left(\widetilde{X}_{R(m)}\right)\right]. \tag*{(224)} X m L := tr S [ Z s ( m ) ( X R ( m ) ) ] . (224) Define the logical-block family associated with X X X to be { X m L } m ∈ M \{X^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { X m L } m ∈ M .
Lemma F.23. For X ∈ End ( H D ) X \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) X ∈ End ( H D ) , denote by { X m L } m ∈ M \{X^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { X m L } m ∈ M the logical-block family associated with X X X , as in Definition F.21, and denote by { D L S ( X ) ~ R } R ⊆ D \{\widetilde{\mathcal{D}_{\mathrm{LS}}(X)}_{R}\}_{R\subseteq\mathrm{D}} { D LS ( X ) R } R ⊆ D the Clifford-frame family associated with D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) , as in Definition F.18. Then, for every R ⊆ D R \subseteq\mathrm{D} R ⊆ D ,
D L S ( X ) ~ R = ∑ s ∈ Z 2 S X ( R , s ) L ⊗ ∣ s ⟩ ⟨ s ∣ S . (225) \widetilde{\mathcal{D}_{\mathrm{LS}}(X)}_{R}
=
\sum_{s\in\mathbb{Z}_{2}^{\mathrm{S}}}
X^{\mathrm{L}}_{(R,s)} \otimes|s\rangle\langle s|_{\mathrm{S}}. \tag*{(225)} D LS ( X ) R = s ∈ Z 2 S ∑ X ( R , s ) L ⊗ ∣ s ⟩ ⟨ s ∣ S . (225)
Proof. Denote by { X ~ R } R ⊆ D \{\widetilde{X}_{R}\}_{R\subseteq\mathrm{D}} { X R } R ⊆ D the Clifford-frame family associated with X X X , as in Definition F.18, and fix R ⊆ D R \subseteq\mathrm{D} R ⊆ D . Equations (F.5), (F.19), and (E.26) give
D L S ( X ) ~ R = Z S ( X ~ R ) = ∑ s ∈ Z 2 S Z s ( X ~ R ) . \widetilde{\mathcal{D}_{\mathrm{LS}}(X)}_{R}
=
\mathcal{Z}_{\mathrm{S}}\left(\widetilde{X}_{R}\right)
=
\sum_{s\in\mathbb{Z}_{2}^{\mathrm{S}}}
\mathcal{Z}_{s}\left(\widetilde{X}_{R}\right). D LS ( X ) R = Z S ( X R ) = s ∈ Z 2 S ∑ Z s ( X R ) .
For every s ∈ Z 2 S s \in\mathbb{Z}_{2}^{\mathrm{S}} s ∈ Z 2 S , Eq. (E.24) gives
Z s ( X ~ R ) = ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) Z s ( X ~ R ) ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) . \mathcal{Z}_{s}\left(\widetilde{X}_{R}\right)
=
(1_{\mathrm{L}}\otimes|s\rangle\langle s|_{\mathrm{S}})
\mathcal{Z}_{s}\left(\widetilde{X}_{R}\right)
(1_{\mathrm{L}}\otimes|s\rangle\langle s|_{\mathrm{S}}). Z s ( X R ) = ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) Z s ( X R ) ( 1 L ⊗ ∣ s ⟩ ⟨ s ∣ S ) .
Since ∣ s ⟩ ⟨ s ∣ S |s\rangle\langle s|_{\mathrm{S}} ∣ s ⟩ ⟨ s ∣ S has rank one and trace one, this equality implies that
Z s ( X ~ R ) = tr S [ Z s ( X ~ R ) ] ⊗ ∣ s ⟩ ⟨ s ∣ S . \mathcal{Z}_{s}\left(\widetilde{X}_{R}\right)
=
\operatorname{tr}_{\mathrm{S}}\left[\mathcal{Z}_{s}\left(\widetilde{X}_{R}\right)\right]
\otimes|s\rangle\langle s|_{\mathrm{S}}. Z s ( X R ) = tr S [ Z s ( X R ) ] ⊗ ∣ s ⟩ ⟨ s ∣ S .
Substituting these equalities into ∑ s ∈ Z 2 S Z s ( X ~ R ) \sum_{s\in\mathbb{Z}_{2}^{\mathrm{S}}}\mathcal{Z}_{s}(\widetilde{X}_{R}) ∑ s ∈ Z 2 S Z s ( X R ) and applying Eq. (F.22) give Eq. (F.24).
For every A ∈ End ( K L ) A \in\operatorname{End}(\mathcal{K}_{L}) A ∈ End ( K L ) and s , t ∈ Z 2 S s,t \in\mathbb{Z}_{2}^{S} s , t ∈ Z 2 S , one has
∣ s ⟩ ⟨ s ∣ S ∣ t ⟩ ⟨ t ∣ S ∣ s ⟩ ⟨ s ∣ S = δ s , t ∣ s ⟩ ⟨ s ∣ S , tr S ( ∣ s ⟩ ⟨ s ∣ S ) = 1. |s\rangle\langle s|_{S}|t\rangle\langle t|_{S}|s\rangle\langle s|_{S}=\delta_{s,t}|s\rangle\langle s|_{S}, \qquad\operatorname{tr}_{S}(|s\rangle\langle s|_{S})=1. ∣ s ⟩ ⟨ s ∣ S ∣ t ⟩ ⟨ t ∣ S ∣ s ⟩ ⟨ s ∣ S = δ s , t ∣ s ⟩ ⟨ s ∣ S , tr S ( ∣ s ⟩ ⟨ s ∣ S ) = 1.
Equation (E.24) therefore gives
tr S [ Z s ( A ⊗ ∣ t ⟩ ⟨ t ∣ S ) ] = δ s , t A . (226) \operatorname{tr}_{S}\left[\mathcal{Z}_{s}\left(A\otimes|t\rangle\langle t|_{S}\right)\right]=\delta_{s,t}A. \tag*{(226)} tr S [ Z s ( A ⊗ ∣ t ⟩ ⟨ t ∣ S ) ] = δ s , t A . (226)
For X ∈ End ( H D ) X\in\operatorname{End}(\mathcal{H}_{D}) X ∈ End ( H D ) , let { X m L } m ∈ M \{X^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { X m L } m ∈ M and { ( D L S ( X ) ) m L } m ∈ M \{(\mathcal{D}_{\mathrm{LS}}(X))^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} {( D LS ( X ) ) m L } m ∈ M be the logical-block families associated with X X X and D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) , respectively, as in Definition F.21. Fix m ∈ M m\in\mathrm{M} m ∈ M . Equations (F.22), (F.24), and (F.25), together with Definition F.20, give
( D L S ( X ) ) m L = tr S [ Z s ( m ) ( ( D L S ( X ) ) R ( m ) ) ] = ∑ s ∈ Z 2 S tr S [ Z s ( m ) ( X ( R ( m ) , s ) L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = X ( R ( m ) , s ( m ) ) L = X m L . \begin{aligned}
(\mathcal{D}_{\mathrm{LS}}(X))^{\mathrm{L}}_{m}
&=\operatorname{tr}_{S}\left[\mathcal{Z}_{s(m)}\left((\mathcal{D}_{\mathrm{LS}}(X))_{R(m)}\right)\right]\\
&=\sum_{s\in\mathbb{Z}_{2}^{S}}\operatorname{tr}_{S}\left[\mathcal{Z}_{s(m)}\left(X^{\mathrm{L}}_{(R(m),s)}\otimes|s\rangle\langle s|_{S}\right)\right]\\
&=X^{\mathrm{L}}_{(R(m),s(m))}=X^{\mathrm{L}}_{m}.
\end{aligned} ( D LS ( X ) ) m L = tr S [ Z s ( m ) ( ( D LS ( X ) ) R ( m ) ) ] = s ∈ Z 2 S ∑ tr S [ Z s ( m ) ( X ( R ( m ) , s ) L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = X ( R ( m ) , s ( m )) L = X m L .
Corollary F.26. For X ∈ End ( H D ) X\in\operatorname{End}(\mathcal{H}_{D}) X ∈ End ( H D ) , let { X m L } m ∈ M \{X^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { X m L } m ∈ M be the logical-block family associated with X X X as in Definition F.21. Then
D L S ( X ) = ∑ m ∈ M ( E R ( m ) D ∘ L R ( m ) D ∘ C F − 1 ) ( X m L ⊗ ∣ s ( m ) ⟩ ⟨ s ( m ) ∣ S ) . (227) \mathcal{D}_{\mathrm{LS}}(X)=\sum_{m\in\mathrm{M}}\left(\mathcal{E}^{\mathrm{D}}_{R(m)}\circ\mathcal{L}^{\mathrm{D}}_{R(m)}\circ\mathcal{C}_{F}^{-1}\right)\left(X^{\mathrm{L}}_{m}\otimes|s(m)\rangle\langle s(m)|_{S}\right). \tag*{(227)} D LS ( X ) = m ∈ M ∑ ( E R ( m ) D ∘ L R ( m ) D ∘ C F − 1 ) ( X m L ⊗ ∣ s ( m )⟩ ⟨ s ( m ) ∣ S ) . (227)
Proof. Let { ( D ~ L S ( X ) ) R } R ⊆ D \{(\widetilde{\mathcal{D}}_{\mathrm{LS}}(X))_{R}\}_{R\subseteq D} {( D LS ( X ) ) R } R ⊆ D be the Clifford-frame family associated with D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) as in Definition F.18. Equations (F.2), (F.5), and (F.19) give
D L S ( X ) = ∑ R ⊆ D ( E R D ∘ L R D ∘ C F − 1 ) ( ( D ~ L S ( X ) ) R ) . \mathcal{D}_{\mathrm{LS}}(X)=\sum_{R\subseteq D}\left(\mathcal{E}^{\mathrm{D}}_{R}\circ\mathcal{L}^{\mathrm{D}}_{R}\circ\mathcal{C}_{F}^{-1}\right)\left((\widetilde{\mathcal{D}}_{\mathrm{LS}}(X))_{R}\right). D LS ( X ) = R ⊆ D ∑ ( E R D ∘ L R D ∘ C F − 1 ) ( ( D LS ( X ) ) R ) .
Substituting Eq. (F.24) and reindexing the resulting double sum by m = ( R , s ) ∈ M m=(R,s)\in\mathrm{M} m = ( R , s ) ∈ M , using Definition F.20, give Eq. (F.27).
Single-cycle hidden-memory update
For m i n , m o u t ∈ M m_{\mathrm{in}},m_{\mathrm{out}}\in\mathrm{M} m in , m out ∈ M , define J m o u t , m i n \mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}} J m out , m in by
J m o u t , m i n : = J s ( m o u t ) , s ( m i n ) R ( m o u t ) , R ( m i n ) . (228) \mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}:=\mathcal{J}^{R(m_{\mathrm{out}}),R(m_{\mathrm{in}})}_{s(m_{\mathrm{out}}),s(m_{\mathrm{in}})}. \tag*{(228)} J m out , m in := J s ( m out ) , s ( m in ) R ( m out ) , R ( m in ) . (228)
Proposition F.29. For every X ∈ End ( H D ) X\in\operatorname{End}(\mathcal{H}_{D}) X ∈ End ( H D ) , let { X m L } m ∈ M \{X^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { X m L } m ∈ M and { ( Φ ( X ) ) m L } m ∈ M \{(\Phi(X))^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} {( Φ ( X ) ) m L } m ∈ M be the logical-block families associated with X X X and Φ ( X ) \Phi(X) Φ ( X ) , respectively, as in Definition F.21. Then, for every m o u t ∈ M m_{\mathrm{out}}\in\mathrm{M} m out ∈ M ,
( Φ ( X ) ) m o u t L = ∑ m i n ∈ M J m o u t , m i n ( X m i n L ) . (229) (\Phi(X))^{\mathrm{L}}_{m_{\mathrm{out}}}=\sum_{m_{\mathrm{in}}\in\mathrm{M}}\mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}\left(X^{\mathrm{L}}_{m_{\mathrm{in}}}\right). \tag*{(229)} ( Φ ( X ) ) m out L = m in ∈ M ∑ J m out , m in ( X m in L ) . (229)
Proof. Let { ( D ~ L S ( X ) ) R } R ⊆ D \{(\widetilde{\mathcal{D}}_{\mathrm{LS}}(X))_{R}\}_{R\subseteq D} {( D LS ( X ) ) R } R ⊆ D and { ( Φ ~ ( X ) ) R } R ⊆ D \{(\widetilde{\Phi}(X))_{R}\}_{R\subseteq D} {( Φ ( X ) ) R } R ⊆ D be the Clifford-frame families associated with D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) and Φ ( X ) \Phi(X) Φ ( X ) , respectively, as in Definition F.18. Evaluating Eq. (F.15) at D L S ( X ) \mathcal{D}_{\mathrm{LS}}(X) D LS ( X ) , using Φ ∘ D L S = Φ \Phi\circ\mathcal{D}_{\mathrm{LS}}=\Phi Φ ∘ D LS = Φ from Eq. (F.12), and applying Eq. (F.19) give, for every R o u t ⊆ D R_{\mathrm{out}}\subseteq D R out ⊆ D ,
Φ ~ ( X ) R o u t = ∑ R i n ⊆ D Φ ~ R o u t , R i n ( D L S ( X ) R i n ) . (230) \widetilde{\Phi}(X)_{R_{\mathrm{out}}}=\sum_{R_{\mathrm{in}}\subseteq D}\widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(\mathcal{D}_{\mathrm{LS}}(X)_{R_{\mathrm{in}}}\right). \tag*{(230)} Φ ( X ) R out = R in ⊆ D ∑ Φ R out , R in ( D LS ( X ) R in ) . (230)
For s i n , s o u t , t ∈ Z 2 S s_{\mathrm{in}},s_{\mathrm{out}},t\in\mathbb{Z}_2^S s in , s out , t ∈ Z 2 S , Eq. (E.67) gives
S s o u t , s i n ( ∣ t ⟩ ⟨ t ∣ S ) = δ s i n , t ∣ s o u t ⟩ ⟨ s o u t ∣ S . (231) S_{s_{\mathrm{out}},s_{\mathrm{in}}}\left(|t\rangle\langle t|_S\right)=\delta_{s_{\mathrm{in}},t}|s_{\mathrm{out}}\rangle\langle s_{\mathrm{out}}|_S. \tag*{(231)} S s out , s in ( ∣ t ⟩ ⟨ t ∣ S ) = δ s in , t ∣ s out ⟩ ⟨ s out ∣ S . (231)
Applying Eq. (E.69) to D L S ( X ) R i n \mathcal{D}_{\mathrm{LS}}(X)_{R_{\mathrm{in}}} D LS ( X ) R in and substituting Eq. (F.24) termwise, with its summation index renamed t t t , yield a sum over s i n , s o u t , t s_{\mathrm{in}},s_{\mathrm{out}},t s in , s out , t . By Eq. (F.32), only the terms with t = s i n t=s_{\mathrm{in}} t = s in remain. Thus, for every R i n , R o u t ⊆ D R_{\mathrm{in}},R_{\mathrm{out}}\subseteq D R in , R out ⊆ D ,
Φ ~ R o u t , R i n ( D L S ( X ) R i n ) = ∑ s i n , s o u t ∈ Z 2 S J s o u t , s i n R o u t , R i n ( X ( R i n , s i n ) L ) ⊗ ∣ s o u t ⟩ ⟨ s o u t ∣ S . (232) \widetilde{\Phi}_{R_{\mathrm{out}},R_{\mathrm{in}}}\left(\mathcal{D}_{\mathrm{LS}}(X)_{R_{\mathrm{in}}}\right)=\sum_{s_{\mathrm{in}},s_{\mathrm{out}}\in\mathbb{Z}_2^S}\mathcal{J}^{R_{\mathrm{out}},R_{\mathrm{in}}}_{s_{\mathrm{out}},s_{\mathrm{in}}}\left(X^{\mathrm{L}}_{(R_{\mathrm{in}},s_{\mathrm{in}})}\right)\otimes|s_{\mathrm{out}}\rangle\langle s_{\mathrm{out}}|_S. \tag*{(232)} Φ R out , R in ( D LS ( X ) R in ) = s in , s out ∈ Z 2 S ∑ J s out , s in R out , R in ( X ( R in , s in ) L ) ⊗ ∣ s out ⟩ ⟨ s out ∣ S . (232)
Substituting Eq. (F.33) termwise into Eq. (F.31), using Eq. (F.28), and reindexing the double sum over R i n R_{\mathrm{in}} R in and s i n s_{\mathrm{in}} s in by m i n = ( R i n , s i n ) m_{\mathrm{in}}=(R_{\mathrm{in}},s_{\mathrm{in}}) m in = ( R in , s in ) give
Φ ~ ( X ) R o u t = ∑ s o u t ∈ Z 2 S [ ∑ m i n ∈ M J ( R o u t , s o u t ) , m i n ( X m i n L ) ] ⊗ ∣ s o u t ⟩ ⟨ s o u t ∣ S . (233) \widetilde{\Phi}(X)_{R_{\mathrm{out}}}=\sum_{s_{\mathrm{out}}\in\mathbb{Z}_2^S}\left[\sum_{m_{\mathrm{in}}\in\mathcal{M}}\mathcal{J}_{(R_{\mathrm{out}},s_{\mathrm{out}}),m_{\mathrm{in}}}\left(X^{\mathrm{L}}_{m_{\mathrm{in}}}\right)\right]\otimes|s_{\mathrm{out}}\rangle\langle s_{\mathrm{out}}|_S. \tag*{(233)} Φ ( X ) R out = s out ∈ Z 2 S ∑ [ m in ∈ M ∑ J ( R out , s out ) , m in ( X m in L ) ] ⊗ ∣ s out ⟩ ⟨ s out ∣ S . (233)
Fix m o u t ∈ M m_{\mathrm{out}}\in\mathcal{M} m out ∈ M . Substituting Eq. (F.34) with R o u t = R ( m o u t ) R_{\mathrm{out}}=R(m_{\mathrm{out}}) R out = R ( m out ) into Eq. (F.22) gives a sum over s o u t s_{\mathrm{out}} s out . Termwise application of Eq. (F.25) leaves only the term with s o u t = s ( m o u t ) s_{\mathrm{out}}=s(m_{\mathrm{out}}) s out = s ( m out ) , giving Eq. (F.30).
Corollary F.35. For every X ∈ End ( H D ) X\in\operatorname{End}(\mathcal{H}_D) X ∈ End ( H D ) , let { X m L } m ∈ M \{X^{\mathrm{L}}_m\}_{m\in\mathcal{M}} { X m L } m ∈ M be the logical-block family associated with X X X , as in Definition F.21. Then
Φ ( X ) = ∑ m i n , m o u t ∈ M ( E R ( m o u t ) D ∘ L R ( m o u t ) D ∘ C F − 1 ) [ J m o u t , m i n ( X m i n L ) ⊗ ∣ s ( m o u t ) ⟩ ⟨ s ( m o u t ) ∣ S ] . (234) \Phi(X)=\sum_{m_{\mathrm{in}},m_{\mathrm{out}}\in\mathcal{M}}\left(\mathcal{E}^{D}_{R(m_{\mathrm{out}})}\circ\mathcal{L}^{D}_{R(m_{\mathrm{out}})}\circ\mathcal{C}_{F}^{-1}\right)\left[\mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}\left(X^{\mathrm{L}}_{m_{\mathrm{in}}}\right)\otimes|s(m_{\mathrm{out}})\rangle\langle s(m_{\mathrm{out}})|_S\right]. \tag*{(234)} Φ ( X ) = m in , m out ∈ M ∑ ( E R ( m out ) D ∘ L R ( m out ) D ∘ C F − 1 ) [ J m out , m in ( X m in L ) ⊗ ∣ s ( m out )⟩ ⟨ s ( m out ) ∣ S ] . (234)
Proof. Let { ( Φ ( X ) ) m L } m ∈ M \{(\Phi(X))^{\mathrm{L}}_m\}_{m\in\mathcal{M}} {( Φ ( X ) ) m L } m ∈ M be the logical-block family associated with Φ ( X ) \Phi(X) Φ ( X ) , as in Definition F.21. The factorization in Eq. (F.12) gives Φ ( X ) = D L S ( Φ ( X ) ) \Phi(X)=\mathcal{D}_{\mathrm{LS}}(\Phi(X)) Φ ( X ) = D LS ( Φ ( X )) . Applying Eq. (F.27) to Φ ( X ) \Phi(X) Φ ( X ) , followed by termwise substitution of Eq. (F.30), gives Eq. (F.36).
Repeated-cycle logical-block dynamics
Proposition F.37. For every integer t ≥ 1 t\geq1 t ≥ 1 , define Φ t \Phi^t Φ t to be the t t t -fold composition of Φ \Phi Φ . For every such t t t and every X ∈ End ( H D ) X\in\operatorname{End}(\mathcal{H}_D) X ∈ End ( H D ) , let { X m L } m ∈ M \{X^{\mathrm{L}}_m\}_{m\in\mathcal{M}} { X m L } m ∈ M and { ( Φ t ( X ) ) m L } m ∈ M \{(\Phi^t(X))^{\mathrm{L}}_m\}_{m\in\mathcal{M}} {( Φ t ( X ) ) m L } m ∈ M be the logical-block families associated with X X X and Φ t ( X ) \Phi^t(X) Φ t ( X ) , respectively, as in Definition F.21. Then, for every m t ∈ M m_t\in\mathcal{M} m t ∈ M ,
( Φ t ( X ) ) m t L = ∑ m 0 , … , m t − 1 ∈ M ( J m t , m t − 1 ∘ ⋯ ∘ J m 1 , m 0 ) ( X m 0 L ) . (235) (\Phi^t(X))^{\mathrm{L}}_{m_t}=\sum_{m_0,\ldots,m_{t-1}\in\mathcal{M}}\left(\mathcal{J}_{m_t,m_{t-1}}\circ\cdots\circ\mathcal{J}_{m_1,m_0}\right)\left(X^{\mathrm{L}}_{m_0}\right). \tag*{(235)} ( Φ t ( X ) ) m t L = m 0 , … , m t − 1 ∈ M ∑ ( J m t , m t − 1 ∘ ⋯ ∘ J m 1 , m 0 ) ( X m 0 L ) . (235) For t = 1 t=1 t = 1 , the composition in Eq. (F.38) is the single map J m 1 , m 0 \mathcal{J}_{m_1,m_0} J m 1 , m 0 .
Proof. When t = 1 t=1 t = 1 , Eq. (F.38) reduces to Eq. (F.30). Fix t ≥ 2 t\ge2 t ≥ 2 , and assume that Eq. (F.38) holds with t t t replaced by t − 1 t-1 t − 1 . Let { ( Φ t − 1 ( X ) ) m L } m ∈ M \{(\Phi^{t-1}(X))^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} {( Φ t − 1 ( X ) ) m L } m ∈ M be the logical-block family associated with Φ t − 1 ( X ) \Phi^{t-1}(X) Φ t − 1 ( X ) , as in Definition F.21. Fix m t ∈ M m_t\in\mathrm{M} m t ∈ M . Applying Eq. (F.30) with X X X replaced by Φ t − 1 ( X ) \Phi^{t-1}(X) Φ t − 1 ( X ) and substituting the induction hypothesis termwise give
( Φ t ( X ) ) m t L = ∑ m t − 1 ∈ M J m t , m t − 1 ( ( Φ t − 1 ( X ) ) m t − 1 L ) = ∑ m 0 , … , m t − 1 ∈ M ( J m t , m t − 1 ∘ ⋯ ∘ J m 1 , m 0 ) ( X m 0 L ) . \begin{aligned}
(\Phi^t(X))^{\mathrm{L}}_{m_t}
&=\sum_{m_{t-1}\in\mathrm{M}}\mathcal{J}_{m_t,m_{t-1}}\left((\Phi^{t-1}(X))^{\mathrm{L}}_{m_{t-1}}\right)\\
&=\sum_{m_0,\ldots,m_{t-1}\in\mathrm{M}}\left(\mathcal{J}_{m_t,m_{t-1}}\circ\cdots\circ\mathcal{J}_{m_1,m_0}\right)(X^{\mathrm{L}}_{m_0}).
\end{aligned} ( Φ t ( X ) ) m t L = m t − 1 ∈ M ∑ J m t , m t − 1 ( ( Φ t − 1 ( X ) ) m t − 1 L ) = m 0 , … , m t − 1 ∈ M ∑ ( J m t , m t − 1 ∘ ⋯ ∘ J m 1 , m 0 ) ( X m 0 L ) .
Memory experiments from logical-Pauli transfer matrices
Define d L : = dim ( K L ) = 2 ∣ L ∣ d_{\mathrm{L}}:=\dim(\mathcal{K}_{\mathrm{L}})=2^{|\mathrm{L}|} d L := dim ( K L ) = 2 ∣ L ∣ and denote by tr L \operatorname{tr}_{\mathrm{L}} tr L the trace on K L \mathcal{K}_{\mathrm{L}} K L . For ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , define P ℓ : = W L ( ℓ ) P_\ell:=W^{\mathrm{L}}(\ell) P ℓ := W L ( ℓ ) , using the specialization of the Pauli notation in Eq. (A.13) to L. For ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , define the linear functional c ℓ : End ( K L ) → C c_\ell:\operatorname{End}(\mathcal{K}_{\mathrm{L}})\to\mathbb{C} c ℓ : End ( K L ) → C by
c ℓ ( A ) : = 1 d L tr L ( P ℓ A ) , (236) c_\ell(A):=\frac{1}{d_{\mathrm{L}}}\operatorname{tr}_{\mathrm{L}}(P_\ell A), \tag*{(236)} c ℓ ( A ) := d L 1 tr L ( P ℓ A ) , (236)
for all A ∈ End ( K L ) A\in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) A ∈ End ( K L ) . Combining Eqs. (F.39) and (A.25), specialized to L, gives,
A = ∑ ℓ ∈ V L c ℓ ( A ) P ℓ , (237) A=\sum_{\ell\in\mathrm{V}_{\mathrm{L}}}c_\ell(A)P_\ell, \tag*{(237)} A = ℓ ∈ V L ∑ c ℓ ( A ) P ℓ , (237)
for all A ∈ End ( K L ) A\in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) A ∈ End ( K L ) .
Definition F.41 (Pauli eigenvalues). Let D : End ( K L ) → End ( K L ) \mathcal{D}:\operatorname{End}(\mathcal{K}_{\mathrm{L}})\to\operatorname{End}(\mathcal{K}_{\mathrm{L}}) D : End ( K L ) → End ( K L ) be a Pauli map, and let { q e } e ∈ V L \{q_e\}_{e\in\mathrm{V}_{\mathrm{L}}} { q e } e ∈ V L be its unique Pauli-map coefficient family, as in Lemma A.42, specialized to L. For ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , define the P ℓ P_\ell P ℓ -eigenvalue of D \mathcal{D} D by
λ ℓ ( D ) : = ∑ e ∈ V L q e ( − 1 ) [ e , ℓ ] . (238) \lambda_\ell(\mathcal{D}):=\sum_{e\in\mathrm{V}_{\mathrm{L}}}q_e(-1)^{[e,\ell]}. \tag*{(238)} λ ℓ ( D ) := e ∈ V L ∑ q e ( − 1 ) [ e , ℓ ] . (238)
Lemma F.43. Let D : End ( K L ) → End ( K L ) \mathcal{D}:\operatorname{End}(\mathcal{K}_{\mathrm{L}})\to\operatorname{End}(\mathcal{K}_{\mathrm{L}}) D : End ( K L ) → End ( K L ) be a Pauli map. Then, for every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L ,
D ( P ℓ ) = λ ℓ ( D ) P ℓ . (239) \mathcal{D}(P_\ell)=\lambda_\ell(\mathcal{D})P_\ell. \tag*{(239)} D ( P ℓ ) = λ ℓ ( D ) P ℓ . (239) If D \mathcal{D} D is a subnormalized Pauli map, then λ ℓ ( D ) ∈ R \lambda_\ell(\mathcal{D})\in\mathbb{R} λ ℓ ( D ) ∈ R for every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L .
Proof. Let { q e } e ∈ V L \{q_e\}_{e\in\mathrm{V}_{\mathrm{L}}} { q e } e ∈ V L be the unique Pauli-map coefficient family of D \mathcal{D} D , as in Lemma A.42. Substituting Eq. (A.41) and using Eqs. (G.3), (A.22), and (A.24), specialized to L, give
D ( P ℓ ) = ∑ e ∈ V L q e Ad P e ( P ℓ ) = ∑ e ∈ V L q e ( − 1 ) [ e , ℓ ] P ℓ = λ ℓ ( D ) P ℓ , \mathcal{D}(P_\ell)=\sum_{e\in\mathrm{V}_{\mathrm{L}}}q_e\operatorname{Ad}_{P_e}(P_\ell)=\sum_{e\in\mathrm{V}_{\mathrm{L}}}q_e(-1)^{[e,\ell]}P_\ell=\lambda_\ell(\mathcal{D})P_\ell, D ( P ℓ ) = e ∈ V L ∑ q e Ad P e ( P ℓ ) = e ∈ V L ∑ q e ( − 1 ) [ e , ℓ ] P ℓ = λ ℓ ( D ) P ℓ ,
which proves Eq. (F.44). If D \mathcal{D} D is subnormalized, then Eq. (A.57), specialized to L, gives q e ∈ [ 0 , ∞ ) q_e\in[0,\infty) q e ∈ [ 0 , ∞ ) for every e ∈ V L e\in\mathrm{V}_{\mathrm{L}} e ∈ V L . It follows from Eq. (F.42) that λ ℓ ( D ) ∈ R \lambda_\ell(\mathcal{D})\in\mathbb{R} λ ℓ ( D ) ∈ R for every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L . ∎
Definition F.45 (Logical-Pauli transfer matrices). For ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , define T ℓ ∈ C M × M T_{\ell} \in\mathbb{C}^{\mathrm{M}\times\mathrm{M}} T ℓ ∈ C M × M by
[ T ℓ ] m o u t , m i n ≔ λ ℓ ( J m o u t , m i n ) , (240) [T_{\ell}]_{m_{\mathrm{out}},m_{\mathrm{in}}} \coloneqq\lambda_{\ell}(\mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}), \tag*{(240)} [ T ℓ ] m out , m in : = λ ℓ ( J m out , m in ) , (240) for every m i n , m o u t ∈ M m_{\mathrm{in}},m_{\mathrm{out}} \in\mathrm{M} m in , m out ∈ M .
Proposition E.68, Eq. (F.28), Lemma F.43, and Eq. (F.46) imply that T ℓ ∈ R M × M T_{\ell} \in\mathbb{R}^{\mathrm{M}\times\mathrm{M}} T ℓ ∈ R M × M for every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L .
Lemma F.47. For every A ∈ End ( K L ) A \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) A ∈ End ( K L ) , ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , and m i n , m o u t ∈ M m_{\mathrm{in}},m_{\mathrm{out}} \in\mathrm{M} m in , m out ∈ M ,
c ℓ ( J m o u t , m i n ( A ) ) = [ T ℓ ] m o u t , m i n c ℓ ( A ) . (241) c_{\ell}(\mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}(A)) = [T_{\ell}]_{m_{\mathrm{out}},m_{\mathrm{in}}}c_{\ell}(A). \tag*{(241)} c ℓ ( J m out , m in ( A )) = [ T ℓ ] m out , m in c ℓ ( A ) . (241) Proof. Fix A ∈ End ( K L ) A \in\operatorname{End}(\mathcal{K}_{\mathrm{L}}) A ∈ End ( K L ) , ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , and m i n , m o u t ∈ M m_{\mathrm{in}},m_{\mathrm{out}} \in\mathrm{M} m in , m out ∈ M . By Proposition E.68 and Eq. (F.28), Lemma F.43 applies to J m o u t , m i n \mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}} J m out , m in . Applying this map termwise to Eq. (F.40) and using Eq. (F.44) give
J m o u t , m i n ( A ) = ∑ a ∈ V L λ a ( J m o u t , m i n ) c a ( A ) P a . (242) \mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}}(A) = \sum_{a\in\mathrm{V}_{\mathrm{L}}}\lambda_a(\mathcal{J}_{m_{\mathrm{out}},m_{\mathrm{in}}})c_a(A)P_a. \tag*{(242)} J m out , m in ( A ) = a ∈ V L ∑ λ a ( J m out , m in ) c a ( A ) P a . (242) Combining Eqs. (F.39) and (A.25), specialized to L \mathrm{L} L , gives c ℓ ( P a ) = δ ℓ , a c_{\ell}(P_a)=\delta_{\ell,a} c ℓ ( P a ) = δ ℓ , a . Applying c ℓ c_{\ell} c ℓ termwise to Eq. (F.49) leaves only the term with a = ℓ a=\ell a = ℓ . Using Eq. (F.46) then gives Eq. (F.48). □
Definition F.50 (Logical-Pauli state-preparation vectors). For ρ ∈ End ( H D ) \rho\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) ρ ∈ End ( H D ) with ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and tr ( ρ ) = 1 \operatorname{tr}(\rho)=1 tr ( ρ ) = 1 , let { ρ m L } m ∈ M \{\rho_m^{\mathrm{L}}\}_{m\in\mathrm{M}} { ρ m L } m ∈ M be the logical-block family associated with ρ \rho ρ , as in Definition F.21 . For ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , define the state-preparation vector r ℓ ( ρ ) ∈ C M r_{\ell}(\rho) \in\mathbb{C}^{\mathrm{M}} r ℓ ( ρ ) ∈ C M by
[ r ℓ ( ρ ) ] m ≔ c ℓ ( ρ m L ) , (243) [r_{\ell}(\rho)]_m \coloneqq c_{\ell}(\rho_m^{\mathrm{L}}), \tag*{(243)} [ r ℓ ( ρ ) ] m : = c ℓ ( ρ m L ) , (243) for all m ∈ M m \in\mathrm{M} m ∈ M .
Lemma F.52. For every ρ ∈ End ( H D ) \rho\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) ρ ∈ End ( H D ) with ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and tr ( ρ ) = 1 \operatorname{tr}(\rho)=1 tr ( ρ ) = 1 , and every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L ,
r ℓ ( Φ ( ρ ) ) = T ℓ r ℓ ( ρ ) . (244) r_{\ell}(\Phi(\rho)) = T_{\ell}r_{\ell}(\rho). \tag*{(244)} r ℓ ( Φ ( ρ )) = T ℓ r ℓ ( ρ ) . (244) Proof. Fix ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L and m o u t ∈ M m_{\mathrm{out}} \in\mathrm{M} m out ∈ M . Applying c ℓ c_{\ell} c ℓ termwise to Eq. (F.30) with X = ρ X=\rho X = ρ , substituting Eq. (F.48), and using Eq. (F.51) give
[ r ℓ ( Φ ( ρ ) ) ] m o u t = ∑ m i n ∈ M [ T ℓ ] m o u t , m i n [ r ℓ ( ρ ) ] m i n = [ T ℓ r ℓ ( ρ ) ] m o u t . (245) [r_{\ell}(\Phi(\rho))]_{m_{\mathrm{out}}} = \sum_{m_{\mathrm{in}}\in\mathrm{M}}[T_{\ell}]_{m_{\mathrm{out}},m_{\mathrm{in}}}[r_{\ell}(\rho)]_{m_{\mathrm{in}}} = [T_{\ell}r_{\ell}(\rho)]_{m_{\mathrm{out}}}. \tag*{(245)} [ r ℓ ( Φ ( ρ )) ] m out = m in ∈ M ∑ [ T ℓ ] m out , m in [ r ℓ ( ρ ) ] m in = [ T ℓ r ℓ ( ρ ) ] m out . (245) Since m o u t m_{\mathrm{out}} m out was arbitrary, Eq. (F.54) gives Eq. (F.53). □
Corollary F.55. For every ρ ∈ End ( H D ) \rho\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) ρ ∈ End ( H D ) with ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and tr ( ρ ) = 1 \operatorname{tr}(\rho)=1 tr ( ρ ) = 1 , every ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L , and every integer k ≥ 1 k \ge1 k ≥ 1 ,
r ℓ ( Φ k ( ρ ) ) = T ℓ k r ℓ ( ρ ) . (246) r_{\ell}(\Phi^k(\rho)) = T_{\ell}^k r_{\ell}(\rho). \tag*{(246)} r ℓ ( Φ k ( ρ )) = T ℓ k r ℓ ( ρ ) . (246) Proof. Fix ρ ∈ End ( H D ) \rho\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) ρ ∈ End ( H D ) with ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and tr ( ρ ) = 1 \operatorname{tr}(\rho)=1 tr ( ρ ) = 1 , and fix ℓ ∈ V L \ell\in\mathrm{V}_{\mathrm{L}} ℓ ∈ V L . The case k = 1 k=1 k = 1 follows from Lemma F.52 . Fix k ≥ 2 k \ge2 k ≥ 2 and assume that Eq. (F.56) holds, with k k k replaced by k − 1 k-1 k − 1 . Since Φ \Phi Φ is a channel, Φ k − 1 ( ρ ) \Phi^{k-1}(\rho) Φ k − 1 ( ρ ) is a state. Applying Lemma F.52 to Φ k − 1 ( ρ ) \Phi^{k-1}(\rho) Φ k − 1 ( ρ ) and substituting the induction hypothesis give
r ℓ ( Φ k ( ρ ) ) = T ℓ r ℓ ( Φ k − 1 ( ρ ) ) = T ℓ T ℓ k − 1 r ℓ ( ρ ) = T ℓ k r ℓ ( ρ ) . (247) r_{\ell}(\Phi^k(\rho)) = T_{\ell}r_{\ell}(\Phi^{k-1}(\rho)) = T_{\ell}T_{\ell}^{k-1}r_{\ell}(\rho) = T_{\ell}^k r_{\ell}(\rho). \tag*{(247)} r ℓ ( Φ k ( ρ )) = T ℓ r ℓ ( Φ k − 1 ( ρ )) = T ℓ T ℓ k − 1 r ℓ ( ρ ) = T ℓ k r ℓ ( ρ ) . (247)
Definition F.58 (Logical-Pauli measurement vectors). For E ∈ End ( H D ) E \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) E ∈ End ( H D ) with 0 ⪯ E ⪯ 1 H D 0 \preceq E \preceq1_{\mathcal{H}_{\mathrm{D}}} 0 ⪯ E ⪯ 1 H D , let { E m L } m ∈ M \{E^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { E m L } m ∈ M be the logical-block family associated with E E E , as in Definition F.21. For ℓ ∈ V L \ell\in V_{\mathrm{L}} ℓ ∈ V L and any m ∈ M m \in\mathrm{M} m ∈ M , define the measurement vector e ℓ ( E ) ∈ C M e_{\ell}(E) \in\mathbb{C}^{\mathrm{M}} e ℓ ( E ) ∈ C M as
[ e ℓ ( E ) ] m : = 2 ∣ R ( m ) ∣ c ℓ ( E m L ) . (248) [e_{\ell}(E)]_{m} := 2^{|R(m)|}c_{\ell}(E^{\mathrm{L}}_{m}). \tag*{(248)} [ e ℓ ( E ) ] m := 2 ∣ R ( m ) ∣ c ℓ ( E m L ) . (248)
Lemma F.60. Let E , Y ∈ End ( H D ) E,Y \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) E , Y ∈ End ( H D ) satisfy 0 ⪯ E ⪯ 1 H D 0 \preceq E \preceq1_{\mathcal{H}_{\mathrm{D}}} 0 ⪯ E ⪯ 1 H D , Y ⪰ 0 Y \succeq0 Y ⪰ 0 , and tr ( Y ) = 1 \operatorname{tr}(Y)=1 tr ( Y ) = 1 . Let { E m L } m ∈ M \{E^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { E m L } m ∈ M and { Y m L } m ∈ M \{Y^{\mathrm{L}}_{m}\}_{m\in\mathrm{M}} { Y m L } m ∈ M be the logical-block families associated with E E E and Y Y Y , respectively, as in Definition F.21, and assume that Y Y Y is leakage–syndrome dephased. Then
tr ( E Y ) = ∑ m ∈ M 2 ∣ R ( m ) ∣ tr L ( E m L Y m L ) (249) \operatorname{tr}(EY)=\sum_{m\in\mathrm{M}}2^{|R(m)|}\operatorname{tr}_{\mathrm{L}}(E^{\mathrm{L}}_{m}Y^{\mathrm{L}}_{m}) \tag*{(249)} tr ( E Y ) = m ∈ M ∑ 2 ∣ R ( m ) ∣ tr L ( E m L Y m L ) (249) = d L ∑ ℓ ∈ V L e ℓ ( E ) T r ℓ ( Y ) . (250) =d_{\mathrm{L}}\sum_{\ell\in V_{\mathrm{L}}}e_{\ell}(E)^{\mathsf{T}}r_{\ell}(Y). \tag*{(250)} = d L ℓ ∈ V L ∑ e ℓ ( E ) T r ℓ ( Y ) . (250) Proof. Denote by { E ~ R } R ⊆ D \{\widetilde{E}_{R}\}_{R\subseteq\mathrm{D}} { E R } R ⊆ D the Clifford-frame family associated with E E E as in Definition F.18. Since Y Y Y is leakage–syndrome dephased, Definition F.10 gives Y = D L S ( Y ) Y=\mathcal{D}_{\mathrm{LS}}(Y) Y = D LS ( Y ) . Substituting Eq. (F.27) with X = Y X=Y X = Y into tr ( E Y ) \operatorname{tr}(EY) tr ( E Y ) gives
tr ( E Y ) = ∑ m ∈ M tr [ E ( E R ( m ) D ∘ L R ( m ) D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ( m ) ⟩ ⟨ s ( m ) ∣ S ) ] . (251) \operatorname{tr}(EY)=\sum_{m\in\mathrm{M}}\operatorname{tr}\left[E\left(\mathcal{E}^{\mathrm{D}}_{R(m)}\circ\mathcal{L}^{\mathrm{D}}_{R(m)}\circ\mathcal{C}^{-1}_{\mathrm{F}}\right)\left(Y^{\mathrm{L}}_{m}\otimes|s(m)\rangle\langle s(m)|_{\mathrm{S}}\right)\right]. \tag*{(251)} tr ( E Y ) = m ∈ M ∑ tr [ E ( E R ( m ) D ∘ L R ( m ) D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ( m )⟩ ⟨ s ( m ) ∣ S ) ] . (251) Fix R ⊆ D R\subseteq\mathrm{D} R ⊆ D and C ∈ End ( H D R ) C\in\operatorname{End}(\mathcal{H}^{R}_{\mathrm{D}}) C ∈ End ( H D R ) . Since l R \mathfrak{l}_{R} l R is one dimensional, there is a unique C ‾ R ∈ End ( c R ) \overline{C}_{R}\in\operatorname{End}(\mathfrak{c}_{R}) C R ∈ End ( c R ) such that C = 1 l R ⊗ C ‾ R C=1_{\mathfrak{l}_{R}}\otimes\overline{C}_{R} C = 1 l R ⊗ C R . For B ∈ End ( K D ) B\in\operatorname{End}(\mathcal{K}_{\mathrm{D}}) B ∈ End ( K D ) , Eqs. (C.29) and (C.31), specialized to D \mathrm{D} D and R R R , respectively, give
tr [ C L R D ( B ) ] = tr [ C ‾ R tr c R ( B ) ] = tr [ ( 1 c R ⊗ C ‾ R ) B ] = 2 ∣ R ∣ tr [ F R D ( C ) B ] . (252) \operatorname{tr}\left[C\mathcal{L}^{\mathrm{D}}_{R}(B)\right]=\operatorname{tr}\left[\overline{C}_{R}\operatorname{tr}_{\mathfrak{c}_{R}}(B)\right]=\operatorname{tr}\left[(1_{\mathfrak{c}_{R}}\otimes\overline{C}_{R})B\right]=2^{|R|}\operatorname{tr}\left[\mathcal{F}^{\mathrm{D}}_{R}(C)B\right]. \tag*{(252)} tr [ C L R D ( B ) ] = tr [ C R tr c R ( B ) ] = tr [ ( 1 c R ⊗ C R ) B ] = 2 ∣ R ∣ tr [ F R D ( C ) B ] . (252) Fix m ∈ M m\in\mathrm{M} m ∈ M , and set R : = R ( m ) R:=R(m) R := R ( m ) and s : = s ( m ) s:=s(m) s := s ( m ) . Applying Eqs. (C.49) and (C.51), followed by Eq. (F.64), gives the first two transformations in the following calculation. Cyclicity of the trace, followed by Eq. (F.19), give the next two. The remaining transformations follow from Eq. (E.24), the defining property of the partial trace, and Eq. (F.22):
tr [ E ( E R D ∘ L R D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = tr [ P R D ( E ) ( L R D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ ( F R D ∘ P R D ) ( E ) C F − 1 ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ C F ( ( F R D ∘ P R D ) ( E ) ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ E ~ R ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ Z s ( E ~ R ) ( Y m L ⊗ 1 K S ) ] = 2 ∣ R ∣ tr L ( tr S [ Z s ( E ~ R ) ] Y m L ) = 2 ∣ R ( m ) ∣ tr L ( E m L Y m L ) . (253) \begin{aligned}
\operatorname{tr}\left[E\left(\mathcal{E}^{\mathrm{D}}_{R}\circ\mathcal{L}^{\mathrm{D}}_{R}\circ\mathcal{C}^{-1}_{\mathrm{F}}\right)\left(Y^{\mathrm{L}}_{m}\otimes|s\rangle\langle s|_{\mathrm{S}}\right)\right]
&=\operatorname{tr}\left[\mathcal{P}^{\mathrm{D}}_{R}(E)\left(\mathcal{L}^{\mathrm{D}}_{R}\circ\mathcal{C}^{-1}_{\mathrm{F}}\right)\left(Y^{\mathrm{L}}_{m}\otimes|s\rangle\langle s|_{\mathrm{S}}\right)\right]\\
&=2^{|R|}\operatorname{tr}\left[\left(\mathcal{F}^{\mathrm{D}}_{R}\circ\mathcal{P}^{\mathrm{D}}_{R}\right)(E)\mathcal{C}^{-1}_{\mathrm{F}}\left(Y^{\mathrm{L}}_{m}\otimes|s\rangle\langle s|_{\mathrm{S}}\right)\right]\\
&=2^{|R|}\operatorname{tr}\left[\mathcal{C}_{\mathrm{F}}\left(\left(\mathcal{F}^{\mathrm{D}}_{R}\circ\mathcal{P}^{\mathrm{D}}_{R}\right)(E)\right)\left(Y^{\mathrm{L}}_{m}\otimes|s\rangle\langle s|_{\mathrm{S}}\right)\right]\\
&=2^{|R|}\operatorname{tr}\left[\widetilde{E}_{R}\left(Y^{\mathrm{L}}_{m}\otimes|s\rangle\langle s|_{\mathrm{S}}\right)\right]\\
&=2^{|R|}\operatorname{tr}\left[\mathcal{Z}_{s}(\widetilde{E}_{R})\left(Y^{\mathrm{L}}_{m}\otimes1_{\mathcal{K}_{\mathrm{S}}}\right)\right]\\
&=2^{|R|}\operatorname{tr}_{\mathrm{L}}\left(\operatorname{tr}_{\mathrm{S}}\left[\mathcal{Z}_{s}(\widetilde{E}_{R})\right]Y^{\mathrm{L}}_{m}\right)\\
&=2^{|R(m)|}\operatorname{tr}_{\mathrm{L}}\left(E^{\mathrm{L}}_{m}Y^{\mathrm{L}}_{m}\right). \tag*{(253)}
\end{aligned} tr [ E ( E R D ∘ L R D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = tr [ P R D ( E ) ( L R D ∘ C F − 1 ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ ( F R D ∘ P R D ) ( E ) C F − 1 ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ C F ( ( F R D ∘ P R D ) ( E ) ) ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ E R ( Y m L ⊗ ∣ s ⟩ ⟨ s ∣ S ) ] = 2 ∣ R ∣ tr [ Z s ( E R ) ( Y m L ⊗ 1 K S ) ] = 2 ∣ R ∣ tr L ( tr S [ Z s ( E R ) ] Y m L ) = 2 ∣ R ( m ) ∣ tr L ( E m L Y m L ) . (253)
Substituting Eq. (F.65) termwise into Eq. (F.63) gives Eq. (F.61).
For m ∈ M m \in\mathsf{M} m ∈ M , substituting Eq. (F.40) with A = E m L A=E_m^{\mathrm{L}} A = E m L and A = Y m L A=Y_m^{\mathrm{L}} A = Y m L into tr L ( E m L Y m L ) \operatorname{tr}_{\mathrm{L}}(E_m^{\mathrm{L}}Y_m^{\mathrm{L}}) tr L ( E m L Y m L ) and applying Eq. (A.25) termwise, specialized to L, leads to
tr L ( E m L Y m L ) = ∑ ℓ , ℓ ′ ∈ V L c ℓ ( E m L ) c ℓ ′ ( Y m L ) tr L ( P ℓ P ℓ ′ ) = d L ∑ ℓ ∈ V L c ℓ ( E m L ) c ℓ ( Y m L ) . (254) \operatorname{tr}_{\mathrm{L}}(E_m^{\mathrm{L}}Y_m^{\mathrm{L}})=\sum_{\ell,\ell'\in\mathcal{V}_{\mathrm{L}}}c_{\ell}(E_m^{\mathrm{L}})c_{\ell'}(Y_m^{\mathrm{L}})\operatorname{tr}_{\mathrm{L}}(P_{\ell}P_{\ell'})
=d_{\mathrm{L}}\sum_{\ell\in\mathcal{V}_{\mathrm{L}}}c_{\ell}(E_m^{\mathrm{L}})c_{\ell}(Y_m^{\mathrm{L}}). \tag*{(254)} tr L ( E m L Y m L ) = ℓ , ℓ ′ ∈ V L ∑ c ℓ ( E m L ) c ℓ ′ ( Y m L ) tr L ( P ℓ P ℓ ′ ) = d L ℓ ∈ V L ∑ c ℓ ( E m L ) c ℓ ( Y m L ) . (254)
Substituting Eq. (F.66) termwise into Eq. (F.61), interchanging the finite sums, and using Eqs. (F.51) and (F.59) give Eq. (F.62).
Fix ρ ∈ End ( H D ) \rho\in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) ρ ∈ End ( H D ) with ρ ⪰ 0 \rho\succeq0 ρ ⪰ 0 and tr ( ρ ) = 1 \operatorname{tr}(\rho)=1 tr ( ρ ) = 1 , and fix E y ∈ End ( H D ) E_y \in\operatorname{End}(\mathcal{H}_{\mathrm{D}}) E y ∈ End ( H D ) with 0 ⪯ E y ⪯ 1 H D 0 \preceq E_y \preceq1_{\mathcal{H}_{\mathrm{D}}} 0 ⪯ E y ⪯ 1 H D , where y y y is an outcome label. The operator ρ \rho ρ is intended to represent the prepared state, and E y E_y E y is intended to represent the POVM element associated with the outcome labeled by y y y . For every integer k ≥ 1 k \ge1 k ≥ 1 , define the probability of outcome y y y after k k k QEC cycles by
p y ( k ) : = tr [ E y Φ k ( ρ ) ] . (255) p_y(k):=\operatorname{tr}[E_y\Phi^k(\rho)]. \tag*{(255)} p y ( k ) := tr [ E y Φ k ( ρ )] . (255)
Theorem 3. For every integer k ≥ 1 k \ge1 k ≥ 1 , the outcome probability p y ( k ) p_y(k) p y ( k ) satisfies
p y ( k ) = d L ∑ ℓ ∈ V L e ℓ ( E y ) T ℓ k r ℓ ( ρ ) . (256) p_y(k)=d_{\mathrm{L}}\sum_{\ell\in\mathcal{V}_{\mathrm{L}}}e_{\ell}(E_y)T_{\ell}^{k}r_{\ell}(\rho). \tag*{(256)} p y ( k ) = d L ℓ ∈ V L ∑ e ℓ ( E y ) T ℓ k r ℓ ( ρ ) . (256)
Proof. Fix an integer k ≥ 1 k \ge1 k ≥ 1 , and set Y : = Φ k ( ρ ) Y:=\Phi^k(\rho) Y := Φ k ( ρ ) . Since Φ \Phi Φ is a channel, Y Y Y is a state. By Proposition F.11 , Y Y Y is leakage–syndrome dephased. Applying Eq. (F.62) with E = E y E=E_y E = E y and using Eq. (F.67) give
p y ( k ) = d L ∑ ℓ ∈ V L e ℓ ( E y ) r ℓ ( Y ) . (257) p_y(k)=d_{\mathrm{L}}\sum_{\ell\in\mathcal{V}_{\mathrm{L}}}e_{\ell}(E_y)r_{\ell}(Y). \tag*{(257)} p y ( k ) = d L ℓ ∈ V L ∑ e ℓ ( E y ) r ℓ ( Y ) . (257)
By Eq. (F.56), r ℓ ( Y ) = T ℓ k r ℓ ( ρ ) r_{\ell}(Y)=T_{\ell}^{k}r_{\ell}(\rho) r ℓ ( Y ) = T ℓ k r ℓ ( ρ ) for every ℓ ∈ V L \ell\in\mathcal{V}_{\mathrm{L}} ℓ ∈ V L . Substituting these equalities termwise into Eq. (F.69) gives Eq. (F.68).
Elementary operator-space lemmas
Let X \mathcal{X} X be a finite-dimensional Hilbert space. Define A \mathcal{A} A to be End ( X ) \operatorname{End}(\mathcal{X}) End ( X ) equipped with the Hilbert–Schmidt inner product
⟨ A , B ⟩ A = tr X ( A † B ) , (258) \langle A,B\rangle_{\mathcal{A}}=\operatorname{tr}_{\mathcal{X}}(A^\dagger B), \tag*{(258)} ⟨ A , B ⟩ A = tr X ( A † B ) , (258)
for any A , B ∈ A A,B \in\mathcal{A} A , B ∈ A . For Φ ∈ End ( A ) \Phi\in\operatorname{End}(\mathcal{A}) Φ ∈ End ( A ) , write Φ † \Phi^\dagger Φ † for the adjoint of Φ \Phi Φ with respect to the inner product on A \mathcal{A} A . Define S \mathcal{S} S to be End ( A ) \operatorname{End}(\mathcal{A}) End ( A ) equipped with the Hilbert–Schmidt inner product
⟨ Φ , Ψ ⟩ S = tr A ( Φ † ∘ Ψ ) , (259) \langle\Phi,\Psi\rangle_{\mathcal{S}}=\operatorname{tr}_{\mathcal{A}}(\Phi^\dagger\circ\Psi), \tag*{(259)} ⟨ Φ , Ψ ⟩ S = tr A ( Φ † ∘ Ψ ) , (259)
for any Φ , Ψ ∈ S \Phi,\Psi\in\mathcal{S} Φ , Ψ ∈ S . Next, for a unitary U ∈ A U \in\mathcal{A} U ∈ A , define Ad U ∈ S \operatorname{Ad}_U \in\mathcal{S} Ad U ∈ S by
Ad U ( B ) = U B U † = : U ( B ) , (260) \operatorname{Ad}_U(B)=UBU^\dagger=:\mathcal{U}(B), \tag*{(260)} Ad U ( B ) = U B U † =: U ( B ) , (260)
for any B ∈ A B \in\mathcal{A} B ∈ A .
Lemma G.4 (Unitary adjoint action). Let U , V ∈ A U,V \in\mathcal{A} U , V ∈ A be unitary. Then
Ad U ∘ Ad V = Ad U V , (261) \operatorname{Ad}_U \circ\operatorname{Ad}_V = \operatorname{Ad}_{UV}, \tag*{(261)} Ad U ∘ Ad V = Ad U V , (261) ( Ad U ) † = Ad U † = ( Ad U ) − 1 . (262) (\operatorname{Ad}_U)^\dagger= \operatorname{Ad}_{U^\dagger} = (\operatorname{Ad}_U)^{-1}. \tag*{(262)} ( Ad U ) † = Ad U † = ( Ad U ) − 1 . (262) Consequently, Ad U \operatorname{Ad}_U Ad U is a unitary element of S \mathcal{S} S .
Proof. For B ∈ A B \in\mathcal{A} B ∈ A , applying Eq. (G.3) first to Ad V \operatorname{Ad}_V Ad V and then to Ad U \operatorname{Ad}_U Ad U gives
( Ad U ∘ Ad V ) ( B ) = U ( V B V † ) U † = U V B ( U V ) † = Ad U V ( B ) . (263) \begin{aligned}
(\operatorname{Ad}_U \circ\operatorname{Ad}_V)(B) &= U(VBV^\dagger)U^\dagger\\
&= UVB(UV)^\dagger\\
&= \operatorname{Ad}_{UV}(B). \tag*{(263)}
\end{aligned} ( Ad U ∘ Ad V ) ( B ) = U ( V B V † ) U † = U V B ( U V ) † = Ad U V ( B ) . (263)
Since B ∈ A B \in\mathcal{A} B ∈ A was arbitrary, Eq. (G.7) proves Eq. (G.5).
For B , C ∈ A B,C \in\mathcal{A} B , C ∈ A , Eqs. (G.1) and (G.3), together with cyclicity of the trace, give
⟨ Ad U ( B ) , C ⟩ A = tr X ( ( U B U † ) † C ) = tr X ( U B † U † C ) = tr X ( B † U † C U ) = ⟨ B , Ad U † ( C ) ⟩ A . (264) \begin{aligned}
\langle\operatorname{Ad}_U(B),C\rangle_{\mathcal{A}} &= \operatorname{tr}_{\mathcal{X}}\left((UBU^\dagger)^\dagger C\right) \\
&= \operatorname{tr}_{\mathcal{X}}(UB^\dagger U^\dagger C) \\
&= \operatorname{tr}_{\mathcal{X}}(B^\dagger U^\dagger C U) \\
&= \langle B,\operatorname{Ad}_{U^\dagger}(C)\rangle_{\mathcal{A}}. \tag*{(264)}
\end{aligned} ⟨ Ad U ( B ) , C ⟩ A = tr X ( ( U B U † ) † C ) = tr X ( U B † U † C ) = tr X ( B † U † C U ) = ⟨ B , Ad U † ( C ) ⟩ A . (264)
By the defining property of the adjoint, Eq. (G.8) implies that
( Ad U ) † = Ad U † . (\operatorname{Ad}_U)^\dagger= \operatorname{Ad}_{U^\dagger}. ( Ad U ) † = Ad U † .
Applying Eq. (G.5) with ( U , V ) = ( U † , U ) (U,V)=(U^\dagger,U) ( U , V ) = ( U † , U ) and ( U , V ) = ( U , U † ) (U,V)=(U,U^\dagger) ( U , V ) = ( U , U † ) , together with Eq. (G.3), gives
Ad U † ∘ Ad U = Ad U ∘ Ad U † = id A . (265) \operatorname{Ad}_{U^\dagger} \circ\operatorname{Ad}_U = \operatorname{Ad}_U \circ\operatorname{Ad}_{U^\dagger} = \operatorname{id}_{\mathcal{A}}. \tag*{(265)} Ad U † ∘ Ad U = Ad U ∘ Ad U † = id A . (265)
Combining the relation ( Ad U ) † = Ad U † (\operatorname{Ad}_U)^\dagger=\operatorname{Ad}_{U^\dagger} ( Ad U ) † = Ad U † with Eq. (G.9) proves Eq. (G.6) and shows that Ad U \operatorname{Ad}_U Ad U is a unitary element of S \mathcal{S} S .
Next, for all A , B , D ∈ A A,B,D \in\mathcal{A} A , B , D ∈ A , define the multiplication map
L B , D ∈ S , L B , D ( A ) = B A D . (266) \mathcal{L}_{B,D} \in\mathcal{S}, \qquad\mathcal{L}_{B,D}(A)=BAD. \tag*{(266)} L B , D ∈ S , L B , D ( A ) = B A D . (266)
Lemma G.11 (Trace of L B , D \mathcal{L}_{B,D} L B , D ). For B , D ∈ A B,D \in\mathcal{A} B , D ∈ A ,
tr A ( L B , D ) = tr X ( B ) tr X ( D ) . (267) \operatorname{tr}_{\mathcal{A}}(\mathcal{L}_{B,D})=\operatorname{tr}_{\mathcal{X}}(B)\operatorname{tr}_{\mathcal{X}}(D). \tag*{(267)} tr A ( L B , D ) = tr X ( B ) tr X ( D ) . (267) Consequently, for unitary U , U ′ ∈ A U,U' \in\mathcal{A} U , U ′ ∈ A ,
⟨ Ad U , Ad U ′ ⟩ S = ∣ tr X ( U † U ′ ) ∣ 2 . (268) \langle\operatorname{Ad}_U,\operatorname{Ad}_{U'}\rangle_{\mathcal{S}}=\left|\operatorname{tr}_{\mathcal{X}}(U^\dagger U')\right|^2. \tag*{(268)} ⟨ Ad U , Ad U ′ ⟩ S = tr X ( U † U ′ ) 2 . (268)
Proof. Let n = dim ( X ) n = \dim(\mathcal{X}) n = dim ( X ) , let { ∣ j ⟩ } j = 1 n \{\lvert j\rangle\}_{j=1}^{n} {∣ j ⟩ } j = 1 n be an orthonormal basis for X \mathcal{X} X , and set E j k = ∣ j ⟩ ⟨ k ∣ E_{jk} = \lvert j\rangle\langle k\rvert E j k = ∣ j ⟩ ⟨ k ∣ . The family { E j k } 1 ≤ j , k ≤ n \{E_{jk}\}_{1\leq j,k\leq n} { E j k } 1 ≤ j , k ≤ n is an orthonormal basis for A \mathcal{A} A , and
E j k † B E j k = ⟨ j ∣ B ∣ j ⟩ ∣ k ⟩ ⟨ k ∣ . E_{jk}^{\dagger} B E_{jk} = \langle j\lvert B\rvert j\rangle\lvert k\rangle\langle k\rvert. E j k † B E j k = ⟨ j ∣ B ∣ j ⟩ ∣ k ⟩ ⟨ k ∣ .
The trace of an endomorphism of A \mathcal{A} A is the sum of its diagonal matrix elements in any orthonormal basis. Therefore, Eqs. (G.1) and (G.10) give
tr A ( L B , D ) = ∑ j , k = 1 n ⟨ E j k , L B , D ( E j k ) ⟩ A = ∑ j , k = 1 n tr X ( E j k † B E j k D ) = ∑ j , k = 1 n ⟨ j ∣ B ∣ j ⟩ ⟨ k ∣ D ∣ k ⟩ = tr X ( B ) tr X ( D ) . (269) \begin{aligned}
\operatorname{tr}_{\mathcal{A}}(\mathcal{L}_{B,D}) &= \sum_{j,k=1}^{n} \langle E_{jk},\mathcal{L}_{B,D}(E_{jk})\rangle_{\mathcal{A}} \\
&= \sum_{j,k=1}^{n} \operatorname{tr}_{\mathcal{X}}\left(E_{jk}^{\dagger} B E_{jk}D\right) \\
&= \sum_{j,k=1}^{n} \langle j\lvert B\rvert j\rangle\langle k\lvert D\rvert k\rangle\\
&= \operatorname{tr}_{\mathcal{X}}(B)\operatorname{tr}_{\mathcal{X}}(D). \tag*{(269)}
\end{aligned} tr A ( L B , D ) = j , k = 1 ∑ n ⟨ E j k , L B , D ( E j k ) ⟩ A = j , k = 1 ∑ n tr X ( E j k † B E j k D ) = j , k = 1 ∑ n ⟨ j ∣ B ∣ j ⟩ ⟨ k ∣ D ∣ k ⟩ = tr X ( B ) tr X ( D ) . (269)
Equation (G.14) proves Eq. (G.12). For unitary U , U ′ ∈ A U,U'\in\mathcal{A} U , U ′ ∈ A , Eqs. (G.3) and (G.10) give
Ad U † U ′ = L U † U ′ , U ′ † U and t r X ( U ′ † U ) ‾ = tr X ( U † U ′ ) . \operatorname{Ad}_{U^{\dagger}U'}=\mathcal{L}_{U^{\dagger}U',U'^{\dagger}U}
\quad\text{and}\quad
\overline{\operatorname{tr_{\mathcal{X}}(U'^{\dagger}U)}}=\operatorname{tr}_{\mathcal{X}}(U^{\dagger}U'). Ad U † U ′ = L U † U ′ , U ′ † U and t r X ( U ′ † U) = tr X ( U † U ′ ) .
Equations (G.2), (G.5), (G.6), and (G.12) together with these equalities, give
⟨ Ad U , Ad U ′ ⟩ S = tr A [ ( Ad U ) † ∘ Ad U ′ ] = tr A ( Ad U † U ′ ) = tr A ( L U † U ′ , U ′ † U ) = tr X ( U † U ′ ) tr X ( U ′ † U ) = ∣ tr X ( U † U ′ ) ∣ 2 . (270) \begin{aligned}
\langle\operatorname{Ad}_{U},\operatorname{Ad}_{U'}\rangle_{\mathcal{S}}
&= \operatorname{tr}_{\mathcal{A}}\left[(\operatorname{Ad}_{U})^{\dagger}\circ\operatorname{Ad}_{U'}\right] \\
&= \operatorname{tr}_{\mathcal{A}}(\operatorname{Ad}_{U^{\dagger}U'}) \\
&= \operatorname{tr}_{\mathcal{A}}(\mathcal{L}_{U^{\dagger}U',U'^{\dagger}U}) \\
&= \operatorname{tr}_{\mathcal{X}}(U^{\dagger}U')\operatorname{tr}_{\mathcal{X}}(U'^{\dagger}U) \\
&= \left\lvert\operatorname{tr}_{\mathcal{X}}(U^{\dagger}U')\right\rvert^{2}. \tag*{(270)}
\end{aligned} ⟨ Ad U , Ad U ′ ⟩ S = tr A [ ( Ad U ) † ∘ Ad U ′ ] = tr A ( Ad U † U ′ ) = tr A ( L U † U ′ , U ′ † U ) = tr X ( U † U ′ ) tr X ( U ′ † U ) = tr X ( U † U ′ ) 2 . (270)
Equation (G.15) proves Eq. (G.13).