Introduction
Standard quantum gate certification protocols on superconducting and trapped-ion processors—most notably Clifford randomized benchmarking (RB) [1 , 2 , 3 ] and interleaved RB [4 ] —compress a noisy quantum operation into a single scalar decay parameter: the Haar-averaged gate infidelity r a v g r_{\mathrm{avg}} r avg [5 , 6 ] , or equivalently the entanglement infidelity 1 − F e 1 - F_{\mathrm{e}} 1 − F e [7 ] . By deliberately twirling gate sequences over the Clifford or Pauli group [12 , 13 , 14 , 15 ] , these benchmarks symmetrize the physical noise channel into an effective depolarizing or Pauli channel, erasing the non-unital polarization drift vector t \boldsymbol{t} t of the Pauli transfer matrix (PTM). At leading order in the dimensionless damping exposure ε = γ T ≪ 1 \varepsilon= \gamma T \ll1 ε = γ T ≪ 1 , both the entanglement infidelity and the average gate infidelity depend solely on the scalar trace rate Γ \Gamma Γ of the Gorini–Kossakowski–Sudarshan–Lindblad (GKSL) dissipator [8 , 9 ] and are invariant under the unitary control trajectory U 0 ( s ) U_{0}(s) U 0 ( s ) implementing the gate [10 , 11 ] :
1 − F e = 1 2 ε Γ + O ( ε 2 ) , r a v g = 1 3 ε Γ + O ( ε 2 ) . (1) 1 - F_{\mathrm{e}} = \frac{1}{2}\varepsilon\Gamma+ \mathcal{O}(\varepsilon^{2}), \qquad r_{\mathrm{avg}} = \frac{1}{3}\varepsilon\Gamma+ \mathcal{O}(\varepsilon^{2}). \tag*{(1)} 1 − F e = 2 1 ε Γ + O ( ε 2 ) , r avg = 3 1 ε Γ + O ( ε 2 ) . (1)
In untwirled quantum algorithms, repetitive stabilizer syndrome-extraction cycles, active qubit reset sequences, and dynamical decoupling protocols, however, physical microwave and optical pulses execute without random Pauli twirling. Whenever the environment is non-unital (t ≠ 0 \boldsymbol{t} \ne0 t = 0 ), the forward survival error probability p + ( ε ) p_{+}(\varepsilon) p + ( ε ) of ∣ 0 ⟩ |0\rangle ∣0 ⟩ and the backward survival error probability p − ( ε ) p_{-}(\varepsilon) p − ( ε ) of ∣ 1 ⟩ |1\rangle ∣1 ⟩ become strongly trajectory-dependent and split at first order in ε \varepsilon ε .
The physical origin of this directional gate-error asymmetry is illustrated in Figure 1 . Consider a superconducting transmon qubit undergoing native longitudinal T 1 T_{1} T 1 energy relaxation toward the ground state ∣ 0 ⟩ |0\rangle ∣0 ⟩ (at the north pole + e z +\boldsymbol{e}_{z} + e z of the Bloch sphere, corresponding to an upward non-unital drift vector t = 2 e z \mathbf{t}=2\mathbf{e}_z t = 2 e z . During a symmetric constant-rate X X X rotation (H 1 = − π 2 σ x H_1=-\frac{\pi}{2}\sigma_x H 1 = − 2 π σ x , Figure 1 (a)), the forward trajectory starting from ∣ 0 ⟩ |0\rangle ∣0 ⟩ and the backward trajectory starting from ∣ 1 ⟩ |1\rangle ∣1 ⟩ spend equal durations in the northern (z > 0 z>0 z > 0 ) and southern (z < 0 z<0 z < 0 ) hemispheres; the time-averaged longitudinal polarization vanishes (m z = 0 m_z=0 m z = 0 ), and the two directional survival error slopes are identical (κ T 1 : = χ + / χ − = 1 \kappa_{T_1}:=\chi_+/\chi_-=1 κ T 1 := χ + / χ − = 1 ). By contrast, during a standard π / 2 \pi/2 π /2 rotation X \sqrt{X} X (H X = − π 4 σ x H_{\sqrt{X}}=-\frac{\pi}{4}\sigma_x H X = − 4 π σ x , Figure 1 (b))—the universal hardware primitive from which arbitrary single-qubit S U ( 2 ) \mathrm{SU}(2) SU ( 2 ) operations are compiled via virtual-Z Z Z phase updates [19 ] —the trajectory starting from ∣ 0 ⟩ |0\rangle ∣0 ⟩ remains entirely within the northern hemisphere (z ≥ 0 z\ge0 z ≥ 0 , near the ground state where T 1 T_1 T 1 decay is suppressed), whereas the antipodal trajectory starting from ∣ 1 ⟩ |1\rangle ∣1 ⟩ remains entirely within the southern hemisphere (z ≤ 0 z\le0 z ≤ 0 , near the excited state where spontaneous emission is rapid). Because the time-averaged longitudinal Bloch coordinate is m z = + 2 / π > 0 m_z=+2/\pi>0 m z = + 2/ π > 0 , ordinary T 1 T_1 T 1 relaxation induces a 12.23 : 1 12.23:1 12.23 : 1 ratio between the backward and forward survival error rates (χ ± = 3 8 ∓ 1 π \chi_\pm=\frac{3}{8}\mp\frac{1}{\pi} χ ± = 8 3 ∓ π 1 , κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 \kappa_{T_1}=(3\pi-8)/(3\pi+8)\approx0.081767 κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 ), even though scalar randomized benchmarking assigns both trajectories the exact same average gate infidelity.
Figure 1. Trajectory mean Bloch vector m \mathbf{m} m versus non-unital bath drift t \mathbf{t} t in the weak-damping Markovian regime. (a) Constant-rate X X X rotation (H 1 = − π 2 σ x H_1=-\frac{\pi}{2}\sigma_x H 1 = − 2 π σ x , q = + 1 q=+1 q = + 1 ): the forward trajectory r + ( s ) \mathbf{r}_+(s) r + ( s ) (solid blue) and backward trajectory r − ( s ) = − r + ( s ) \mathbf{r}_-(s)=-\mathbf{r}_+(s) r − ( s ) = − r + ( s ) (dashed red) spend equal time in the northern and southern hemispheres (m z = 0 m_z=0 m z = 0 , m 1 = 2 π e y \mathbf{m}_1=\frac{2}{\pi}\mathbf{e}_y m 1 = π 2 e y ), coupling only to transverse drift t y ≠ 0 t_y\ne0 t y = 0 . (b) Polar X \sqrt{X} X gate (H X = − π 4 σ x H_{\sqrt{X}}=-\frac{\pi}{4}\sigma_x H X = − 4 π σ x , Q = π / 2 Q=\pi/2 Q = π /2 ): the forward arc from ∣ 0 ⟩ |0\rangle ∣0 ⟩ remains entirely in the northern hemisphere z ≥ 0 z\ge0 z ≥ 0 whereas the antipodal arc from ∣ 1 ⟩ |1\rangle ∣1 ⟩ traverses z ≤ 0 z\le0 z ≤ 0 with m X = 2 π ( e y + e z ) \mathbf{m}_{\sqrt{X}}=\frac{2}{\pi}(\mathbf{e}_y+\mathbf{e}_z) m X = π 2 ( e y + e z ) , so longitudinal T 1 T_1 T 1 relaxation (t = 2 e z \mathbf{t}=2\mathbf{e}_z t = 2 e z ) induces a 12.23 : 1 12.23:1 12.23 : 1 directional survival-error split (κ T 1 ≈ 0.081767 \kappa_{T_1}\approx0.081767 κ T 1 ≈ 0.081767 ) while symmetric X X X rotations remain immune (κ T 1 = 1 \kappa_{T_1}=1 κ T 1 = 1 ).
Summary of main results and physical implications
In this article, we establish a general analytical framework governing state-dependent directional survival errors and their universal complete-positivity bounds across open quantum systems:
Pure-state jump-variance law and two-moment trajectory decomposition (Proposition 1 and Theorem 1): Using first-order toggling-frame perturbation theory for a two-level qubit (d = 2 d=2 d = 2 ) coupled to a time-independent Markovian environment, we show that the instantaneous rate of pure-state survival fidelity loss equals the quantum fluctuation variance of the active traceless jump operators along the control trajectory (Proposition 1). Consequently, the antipodal survival error slopes χ ± = lim ϵ → 0 + p ± ( ϵ ) / ϵ \chi_\pm=\lim_{\epsilon\to0+}p_\pm(\epsilon)/\epsilon χ ± = lim ϵ → 0 + p ± ( ϵ ) / ϵ along any piecewise-continuous Bloch trajectory r ( s ) ∈ S 2 \mathbf{r}(s)\in S^2 r ( s ) ∈ S 2
are determined exclusively by two geometric moments of the control path: its time-averaged Bloch polarization vector m \mathbf{m} m , which couples linearly to the non-unital bath drift t \mathbf{t} t , and its transverse residence covariance matrix G G G , which contracts with the unital bath damping tensor S S S according to χ ± ( D ; r ) = 1 2 Tr ( S G ) ∓ 1 4 m ⋅ t \chi_{\pm}(D; \mathbf{r}) = \frac{1}{2}\operatorname{Tr}(SG) \mp\frac{1}{4}\mathbf{m}\cdot\mathbf{t} χ ± ( D ; r ) = 2 1 Tr ( S G ) ∓ 4 1 m ⋅ t . Maximizing the normalized directional contrast ∣ χ + − χ − ∣ / ( χ + + χ − ) |\chi_{+}-\chi_{-}|/(\chi_{+}+\chi_{-}) ∣ χ + − χ − ∣/ ( χ + + χ − ) over all completely positive GKSL generators yields the sharp geometric cofactor bound β ( r ) = m T G m / det G ≤ 1 \beta(\mathbf{r}) = \sqrt{\mathbf{m}^{T}G\mathbf{m}/\det G} \le1 β ( r ) = m T G m / det G ≤ 1 and the corresponding survival-error selectivity floor κ ( r ) = ( 1 − β ( r ) ) / ( 1 + β ( r ) ) \kappa(\mathbf{r}) = (1-\beta(\mathbf{r}))/(1+\beta(\mathbf{r})) κ ( r ) = ( 1 − β ( r )) / ( 1 + β ( r )) (Theorem 1).
Universal selectivity bound for autonomous swaps and constant-axis rotations (Theorem 2 and Proposition 2): Specializing to time-independent single-qubit π \pi π inversions (autonomous swaps, Theorem 2), we prove that the directional survival-error ratio min ( χ + , χ − ) / max ( χ + , χ − ) \min(\chi_{+},\chi_{-})/\max(\chi_{+},\chi_{-}) min ( χ + , χ − ) / max ( χ + , χ − ) across all Markovian environments is bounded below by the universal constant
κ ⋆ = π − 2 2 π + 2 2 ≈ 0.052456 ( κ ⋆ − 1 ≈ 19.0635 : 1 ) , \kappa_{\star} = \frac{\pi- 2\sqrt{2}}{\pi+ 2\sqrt{2}} \approx0.052456 \qquad(\kappa_{\star}^{-1} \approx19.0635 : 1), κ ⋆ = π + 2 2 π − 2 2 ≈ 0.052456 ( κ ⋆ − 1 ≈ 19.0635 : 1 ) ,
which is uniquely saturated at odd winding ∣ q ∣ = 1 |q|=1 ∣ q ∣ = 1 by the chiral jump operators L ± = ( σ x ± i 2 σ z ) / 6 L_{\pm} = (\sigma_{x} \pm i\sqrt{2}\sigma_{z})/\sqrt{6} L ± = ( σ x ± i 2 σ z ) / 6 . For arbitrary constant-axis rotations H ( Q , ϑ ) H(Q,\vartheta) H ( Q , ϑ ) , block-diagonalization in the trajectory midpoint frame yields an exact closed-form midpoint-sinc formula for β ( ∣ Q ∣ , ϑ ) \beta(|Q|,\vartheta) β ( ∣ Q ∣ , ϑ ) (Proposition 2), revealing both the 12.23 : 1 native-T 1 T_{1} T 1 split on polar X \sqrt{X} X gates and a winding-independent 19 : 3 ≈ 6.33 : 1 \approx6.33 : 1 ≈ 6.33 : 1 native-T 1 T_{1} T 1 split (κ T 1 = 3 / 19 ≈ 0.157895 \kappa_{T_{1}} = 3/19 \approx0.157895 κ T 1 = 3/19 ≈ 0.157895 ) on single-pulse tilted Hadamard gates.
Shaped AWG pulses, SPAM-invariant lock-in metrology, and analytical extensions (§ 6 –§ 8 ): We apply the framework to baseline-subtracted truncated Gaussian and DRAG waveforms [19 , 31 ] (§ 6 ), evaluate quantitative hardware predictions for superconducting transmons and 171 Y b + {}^{171}\mathrm{Yb}^{+} 171 Yb + clock qubits alongside two-branch differential lock-in protocols that cancel additive state-preparation and measurement (SPAM) offsets to O ( ε 2 ) O(\varepsilon^{2}) O ( ε 2 ) for both transverse-drift swaps (8.97× \times × plug-in variance reduction) and longitudinal-T 1 T_{1} T 1 polar X \sqrt{X} X gates (§ 7 ), and derive analytical results for finite damping strength (where entanglement fidelity remains winding-independent through O ( η 3 ) O(\eta^{3}) O ( η 3 ) before splitting at O ( η 4 ) O(\eta^{4}) O ( η 4 ) ), non-autonomous pulse dwell, colored thermal baths, and multilevel qutrit cascade relaxation (§ 8 ).
While state-dependent amplitude-damping errors (∣ 1 ⟩ → ∣ 0 ⟩ |1\rangle\to|0\rangle ∣1 ⟩ → ∣0 ⟩ relaxation asymmetric with respect to ∣ 0 ⟩ |0\rangle ∣0 ⟩ versus ∣ 1 ⟩ |1\rangle ∣1 ⟩ ) and the sensitivity of incoherently damped gates to pulse profiles are well known in quantum control and randomized benchmarking [16 , 17 , 18 , 19 ] , our contribution lies in establishing (i) the exact geometric two-moment trajectory law χ ± = 1 2 Tr ( S G ) ∓ 1 4 m ⋅ t \chi_{\pm} = \frac{1}{2}\operatorname{Tr}(SG) \mp\frac{1}{4}\mathbf{m}\cdot\mathbf{t} χ ± = 2 1 Tr ( S G ) ∓ 4 1 m ⋅ t , (ii) the complete-positivity cofactor bound β ( r ) ≤ 1 \beta(\mathbf{r}) \le1 β ( r ) ≤ 1 , and (iii) the time-independent single-qubit swap selectivity bound κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) \kappa_{\star} = (\pi- 2\sqrt{2})/(\pi+ 2\sqrt{2}) κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) . These results connect three lines of research in quantum gate characterization and open-system control. First, in the theory of Lindblad fidelity relations, Pedersen et al. [10 ] and Abad et al. [11 ] showed that first-order average gate infidelity and entanglement infidelity are path-independent and proportional to the dissipator trace Γ \Gamma Γ , while subsequent studies have analyzed unitarity and non-unital noise in randomized benchmarking and leakage [16 , 17 , 18 ] , thermodynamic dissipation bounds [20 ] , higher-order nonlinearities in open qudit gates [21 ] , and Pauli-twirled Lindbladian noise symmetries [22 , 23 ] . Because scalar fidelity benchmarks and Pauli-twirled channels average over input states, the non-unital drift vector t \mathbf{t} t drops out at first order; our two-moment trajectory law resolves the untwirled regime where t \mathbf{t} t governs state-to-state transition asymmetry. Second, in quantum control and tomography, generalized filter-function formalisms [24 , 25 , 26 ] characterize decoherence via spectral overlap integrals in the toggling frame, while Gate Set Tomography (GST) [27 , 28 , 29 , 30 ] reconstructs full process matrices numerically. Here we show that in the Markovian regime, the zero-frequency toggling-frame filter moments ( m , G ) (\mathbf{m},G) ( m , G ) obey the closed-form matrix cofactor inequality adj ( G ) − m m T ⪰ 0 \operatorname{adj}(G)-\mathbf{m}\mathbf{m}^{T}\succeq0 adj ( G ) − m m T ⪰ 0 , which analytically bounds directional gate selectivity for arbitrary pulse shapes. Third, in biased-noise fault-tolerant architectures such as the XZZX surface code [32 , 33 , 34 ] , decoder weights are tailored to the Pauli noise bias η Z = p Z / ( p X + p Y ) \eta_{Z}=p_{Z}/(p_{X}+p_{Y}) η Z = p Z / ( p X + p Y ) between dephasing and bit-flip errors. Directional transition-error asymmetry ( p + ≠ p − ) (p_{+}\ne p_{-}) ( p + = p − ) represents a complementary state-dependent relaxation bias within the bit-flip sector: incorporating asymmetric ∣ 0 ⟩ ↔ ∣ 1 ⟩ |0\rangle\leftrightarrow|1\rangle ∣0 ⟩ ↔ ∣1 ⟩ transition weights into minimum-weight perfect-matching (MWPM) or belief-propagation decoders avoids the systematic logical-error underestimation of Pauli-twirled approximations (Appendix § 21 ).
The remainder of this paper is organized as follows. § 2 introduces the time-rescaled GKSL master equation in the qubit rotating frame, proves the pure-state jump-variance law, and illustrates the formalism with worked calculations for native T 1 T_1 T 1 relaxation on a X \sqrt{X} X gate and extremal chiral dissipation on an X X X gate. § 4 establishes the general two-moment trajectory law and the geometric cofactor asymmetry bound β ( r ) ≤ 1 \beta(r) \le1 β ( r ) ≤ 1 (Theorem 1). § 5 proves the universal selectivity bound κ ⋆ \kappa_\star κ ⋆ for time-independent swaps (Theorem 2) and derives closed-form midpoint-sinc expressions for arbitrary constant-axis rotations, polar X \sqrt{X} X gates, and single-pulse Hadamard gates (Proposition 2). § 6 extends the theory to realistic microwave arbitrary-waveform-generator (AWG) pulses with truncated Gaussian and DRAG envelopes. § 7 evaluates quantitative hardware predictions for superconducting transmons (ibm_kyiv) and 171 Y b + {}^{171}\mathrm{Yb}^{+} 171 Yb + trapped-ion processors and presents the six-circuit signed-drive differential lock-in protocol. § 8 develops analytical extensions to finite damping strength, non-autonomous pulse dwell, colored thermal baths, and qutrit leakage, and § 9 summarizes our conclusions. Self-contained mathematical proofs, Redfield and Slepian derivations, statistical Cramér–Rao analyses, and numerical verification tables across 1,203,200 random CPTP baths are provided in Appendices § 13 –§ 22 .
Open-system master equation, pure-state variance law, and physical examples
We formulate our analysis in the standard weak-damping Markovian regime relevant to high-fidelity single-qubit gates. Specifically, we consider a two-level quantum system (H ≅ C 2 \mathcal{H} \cong\mathbb{C}^{2} H ≅ C 2 , d = 2 d = 2 d = 2 , neglecting leakage to higher excited levels such as ∣ 2 ⟩ \lvert2\rangle ∣ 2 ⟩ , which is below 3.1 × 10 − 6 3.1 \times10^{-6} 3.1 × 1 0 − 6 per gate under DRAG control [19 ] ) evolving over a gate duration T > 0 T > 0 T > 0 under a piecewise-continuous control Hamiltonian H ( s ) H(s) H ( s ) and a time-independent rotating-frame Markovian Lindblad dissipator D \mathcal{D} D with dimensionless noise exposure ε : = γ T ≪ 1 \varepsilon:= \gamma T \ll1 ε := γ T ≪ 1 , where γ > 0 \gamma> 0 γ > 0 denotes the characteristic bath relaxation rate. Throughout the analytical derivations, we set ℏ = 1 \hbar= 1 ℏ = 1 and parameterize evolution by the dimensionless gate time s : = t / T ∈ [ 0 , 1 ] s := t/T \in[0,1] s := t / T ∈ [ 0 , 1 ] ; when specializing to time-independent constant-amplitude drives in § 5 , the control pulse operates continuously at the quantum speed limit (T = T Q S L T = T_{\mathrm{QSL}} T = T QSL ) without idle dwell.
Time-rescaled GKSL master equation and Dyson remainder
On an operator space over H ≅ C d \mathcal{H} \cong\mathbb{C}^{d} H ≅ C d (d ≥ 2 d \ge2 d ≥ 2 ) equipped with the Hilbert–Schmidt inner product ⟨ ⟨ A ∣ B ⟩ ⟩ : = Tr ( A † B ) \langle\langle A|B\rangle\rangle:= \operatorname{Tr}(A^{\dagger}B) ⟨⟨ A ∣ B ⟩⟩ := Tr ( A † B ) , the time-rescaled GKSL master equation [8 , 9 ] for the dynamical propagator Φ ε ( s ) \Phi_{\varepsilon}(s) Φ ε ( s ) is
d d s Φ ε ( s ) = L ε ( s ) ∘ Φ ε ( s ) , L ε ( s ) = − i ad H ( s ) + ε D ( s ) , (2) \frac{d}{ds}\Phi_{\varepsilon}(s)=\mathcal{L}_{\varepsilon}(s)\circ\Phi_{\varepsilon}(s),\qquad\mathcal{L}_{\varepsilon}(s)=-i\,\operatorname{ad}_{H(s)}+\varepsilon\mathcal{D}(s), \tag*{(2)} d s d Φ ε ( s ) = L ε ( s ) ∘ Φ ε ( s ) , L ε ( s ) = − i ad H ( s ) + ε D ( s ) , (2)
with initial condition Φ ε ( 0 ) = I \Phi_{\varepsilon}(0)=\mathcal{I} Φ ε ( 0 ) = I (where I \mathcal{I} I denotes the superoperator identity), commutator superoperator ad H ( s ) ( ρ ) : = [ H ( s ) , ρ ] \operatorname{ad}_{H(s)}(\rho):=[H(s),\rho] ad H ( s ) ( ρ ) := [ H ( s ) , ρ ] , and Lindblad dissipator
D ( s ) ( ρ ) : = ∑ α γ α ( s ) ( L α ( s ) ρ L α ( s ) † − 1 2 { L α ( s ) † L α ( s ) , ρ } ) , (3) \mathcal{D}(s)(\rho):=\sum_{\alpha}\gamma_{\alpha}(s)\left(L_{\alpha}(s)\rho L_{\alpha}(s)^{\dagger}-\frac{1}{2}\{L_{\alpha}(s)^{\dagger}L_{\alpha}(s),\rho\}\right), \tag*{(3)} D ( s ) ( ρ ) := α ∑ γ α ( s ) ( L α ( s ) ρ L α ( s ) † − 2 1 { L α ( s ) † L α ( s ) , ρ } ) , (3)
where γ α ( s ) ≥ 0 \gamma_{\alpha}(s)\ge0 γ α ( s ) ≥ 0 . Decomposing each jump operator into its trace and traceless parts,
L α ( s ) = c α ( s ) I d + L ~ α ( s ) , c α ( s ) : = 1 d Tr ( L α ( s ) ) , Tr ( L ~ α ( s ) ) = 0 , (4) L_{\alpha}(s)=c_{\alpha}(s)\mathbb{I}_{d}+\widetilde{L}_{\alpha}(s),\qquad c_{\alpha}(s):=\frac{1}{d}\operatorname{Tr}(L_{\alpha}(s)),\qquad\operatorname{Tr}(\widetilde{L}_{\alpha}(s))=0, \tag*{(4)} L α ( s ) = c α ( s ) I d + L α ( s ) , c α ( s ) := d 1 Tr ( L α ( s )) , Tr ( L α ( s )) = 0 , (4)
separates the generator as D ( s ) ( ρ ) = D ~ ( s ) ( ρ ) − i [ Δ H L a m b ( s ) , ρ ] \mathcal{D}(s)(\rho)=\widetilde{\mathcal{D}}(s)(\rho)-i[\Delta H_{\mathrm{Lamb}}(s),\rho] D ( s ) ( ρ ) = D ( s ) ( ρ ) − i [ Δ H Lamb ( s ) , ρ ] , where the traceless jump operators L ~ α ( s ) \widetilde{L}_{\alpha}(s) L α ( s ) set the irreversible dissipation rate and the scalar trace parts c α ( s ) I d c_{\alpha}(s)\mathbb{I}_{d} c α ( s ) I d generate an effective
Lamb-like Hamiltonian shift
Δ H L a m b ( s ) = 1 2 i ∑ α γ α ( s ) ( c α ∗ ( s ) L ~ α ( s ) − c α ( s ) L ~ α † ( s ) ) , \Delta H_{\mathrm{Lamb}}(s)=\frac{1}{2i}\sum_{\alpha}\gamma_{\alpha}(s)\left(c_{\alpha}^{*}(s)\widetilde{L}_{\alpha}(s)-c_{\alpha}(s)\widetilde{L}_{\alpha}^{\dagger}(s)\right), Δ H Lamb ( s ) = 2 i 1 α ∑ γ α ( s ) ( c α ∗ ( s ) L α ( s ) − c α ( s ) L α † ( s ) ) ,
which vanishes whenever L α L_{\alpha} L α is traceless, and is explicitly retained when analyzing non-traceless rank-1 nilpotent jump operators such as L ⋆ : = ∣ ψ ( π / 8 ) ⟩ ⟨ ψ ( 7 π / 8 ) ∣ = − 2 4 I 2 − i 2 4 σ x − 1 2 σ z L_{\star}:=\lvert\psi(\pi/8)\rangle\langle\psi(7\pi/8)\rvert=-\frac{\sqrt{2}}{4}I_{2}-\frac{i\sqrt{2}}{4}\sigma_{x}-\frac{1}{2}\sigma_{z} L ⋆ := ∣ ψ ( π /8 )⟩ ⟨ ψ ( 7 π /8 )∣ = − 4 2 I 2 − 4 i 2 σ x − 2 1 σ z (where ∣ ψ ( θ ) ⟩ : = cos θ ∣ 0 ⟩ + sin θ ∣ 1 ⟩ \lvert\psi(\theta)\rangle:=\cos\theta\lvert0\rangle+\sin\theta\lvert1\rangle ∣ ψ ( θ )⟩ := cos θ ∣ 0 ⟩ + sin θ ∣ 1 ⟩ , c ⋆ : = 1 2 Tr ( L ⋆ ) = − 2 4 c_{\star}:=\frac{1}{2}\operatorname{Tr}(L_{\star})=-\frac{\sqrt{2}}{4} c ⋆ := 2 1 Tr ( L ⋆ ) = − 4 2 , and Γ ⋆ : = 3 4 \Gamma_{\star}:=\frac{3}{4} Γ ⋆ := 4 3 ) in § 8 and Figure 5 (a). The instantaneous traceless dissipation rate and its gate-integrated average are
Figure 5. Exact all-orders finite-ε \varepsilon ε trajectory curvature and winding-number invariance of entanglement fidelity through O ( η 3 ) O(\eta^3) O ( η 3 ) . (a) Finite-damping excursion ∣ b ( ε ) − κ ⋆ a ( ε ) ∣ |b(\varepsilon)-\kappa_\star a(\varepsilon)| ∣ b ( ε ) − κ ⋆ a ( ε ) ∣ versus dimensionless noise exposure ε = γ T \varepsilon=\gamma T ε = γ T for the retimed rank-1 jump L ⋆ = ∣ ψ ( π / 8 ) ⟩ ⟨ ψ ( 7 π / 8 ) ∣ = − 2 4 I 2 − i 2 4 σ x − 1 2 σ z L_\star=|\psi(\pi/8)\rangle\langle\psi(7\pi/8)|=-\frac{\sqrt{2}}{4}\mathbb{I}_2-\frac{i\sqrt{2}}{4}\sigma_x-\frac{1}{2}\sigma_z L ⋆ = ∣ ψ ( π /8 )⟩ ⟨ ψ ( 7 π /8 ) ∣ = − 4 2 I 2 − 4 i 2 σ x − 2 1 σ z at λ ⋆ = ( π − 1 ) / ( 4 π 2 ) \lambda_\star=(\pi-1)/(4\pi^2) λ ⋆ = ( π − 1 ) / ( 4 π 2 ) (solid red, ∝ ε 2 \propto\varepsilon^2 ∝ ε 2 ) and the calibrated fixed-drive traceless chiral jump L + L_+ L + at λ = 0 \lambda=0 λ = 0 (solid blue, exact damped-Bloch solution matching the analytical cubic law Eq. (53) , dashed green), showing that coherent-dissipative frequency pulling shifts the finite-ε \varepsilon ε curve by ∼ 10 − 10 \sim10^{-10} ∼ 1 0 − 10 at ε = 10 − 2 \varepsilon=10^{-2} ε = 1 0 − 2 . (b) Exact entanglement-fidelity splitting ∣ F e ( η , π ) − F e ( η , q π ) ∣ |F_e(\eta,\pi)-F_e(\eta,q\pi)| ∣ F e ( η , π ) − F e ( η , q π ) ∣ from Eq. (49) for q = 3 q=3 q = 3 (solid blue) and q = 5 q=5 q = 5 (dashed red) versus total noise exposure η = ε Γ \eta=\varepsilon\Gamma η = ε Γ , confirming that winding-number invariance of entanglement fidelity persists through O ( η 3 ) O(\eta^3) O ( η 3 ) before splitting at O ( η 4 ) O(\eta^4) O ( η 4 ) according to Eq. (52) (dotted black).
Γ ( s ) : = ∑ α γ α ( s ) Tr ( L ~ α ( s ) L ~ α † ( s ) ) ≥ 0 , Γ t o t : = ∫ 0 1 Γ ( s ) d s , (5) \begin{aligned}
\Gamma(s)&:=\sum_{\alpha}\gamma_{\alpha}(s)\operatorname{Tr}\left(\widetilde{L}_{\alpha}(s)\widetilde{L}_{\alpha}^{\dagger}(s)\right)\geq0,\\
\Gamma_{\mathrm{tot}}&:=\int_{0}^{1}\Gamma(s)\,\mathrm{d}s, \tag*{(5)}
\end{aligned} Γ ( s ) Γ tot := α ∑ γ α ( s ) Tr ( L α ( s ) L α † ( s ) ) ≥ 0 , := ∫ 0 1 Γ ( s ) d s , (5)
while the unshifted jump norm is Γ f u l l ( s ) : = ∑ α γ α ( s ) Tr ( L α ( s ) † L α ( s ) ) = Γ ( s ) + d ∑ α γ α ( s ) ∣ c α ( s ) ∣ 2 \Gamma_{\mathrm{full}}(s):=\sum_{\alpha}\gamma_{\alpha}(s)\operatorname{Tr}(L_{\alpha}(s)^{\dagger}L_{\alpha}(s))=\Gamma(s)+d\sum_{\alpha}\gamma_{\alpha}(s)\lvert c_{\alpha}(s)\rvert^{2} Γ full ( s ) := ∑ α γ α ( s ) Tr ( L α ( s ) † L α ( s )) = Γ ( s ) + d ∑ α γ α ( s ) ∣ c α ( s ) ∣ 2 , with Γ f u l l , t o t : = ∫ 0 1 Γ f u l l ( s ) d s \Gamma_{\mathrm{full,tot}}:=\int_{0}^{1}\Gamma_{\mathrm{full}}(s)\,\mathrm{d}s Γ full , tot := ∫ 0 1 Γ full ( s ) d s .
Equation (2) is written in the frame rotating at the qubit transition frequency ω 0 \omega_{0} ω 0 . When the bare transition frequency far exceeds the Rabi drive amplitude (ω 0 ≫ Ω R a b i \omega_{0}\gg\Omega_{\mathrm{Rabi}} ω 0 ≫ Ω Rabi ) and the bath correlation time satisfies τ B ≪ Ω R a b i − 1 \tau_{B}\ll\Omega_{\mathrm{Rabi}}^{-1} τ B ≪ Ω Rabi − 1 so that the environmental noise power spectrum is flat across the Mollow triplet sidebands ω 0 ± Ω R a b i \omega_{0}\pm\Omega_{\mathrm{Rabi}} ω 0 ± Ω Rabi , fast counter-rotating secular terms average out under the rotating-wave approximation (RWA). In this regime, spontaneous emission L ↓ = ∣ 0 ⟩ ⟨ 1 ∣ L_{\downarrow}=\lvert0\rangle\langle1\rvert L ↓ = ∣ 0 ⟩ ⟨ 1 ∣ , thermal absorption L ↑ = ∣ 1 ⟩ ⟨ 0 ∣ L_{\uparrow}=\lvert1\rangle\langle0\rvert L ↑ = ∣ 1 ⟩ ⟨ 0 ∣ , and pure dephasing L ϕ = σ z / 2 L_{\phi}=\sigma_{z}/\sqrt{2} L ϕ = σ z / 2 remain time-independent in the rotating frame, while zero-duration virtual-Z Z Z phase updates [19 ] rotate the azimuthal drive axis in software. Transforming to the unitary toggling frame Φ ~ ε ( s ) : = U 0 ( s ) † Φ ε ( s ) \widetilde{\Phi}_{\varepsilon}(s):=U_{0}(s)^{\dagger}\Phi_{\varepsilon}(s) Φ ε ( s ) := U 0 ( s ) † Φ ε ( s ) with U 0 ( s ) ( ρ ) : = U 0 ( s ) ρ U 0 ( s ) † \mathcal{U}_{0}(s)(\rho):=U_{0}(s)\rho U_{0}(s)^{\dagger} U 0 ( s ) ( ρ ) := U 0 ( s ) ρ U 0 ( s ) † and D ^ ( s ) : = U 0 ( s ) † D ( s ) U 0 ( s ) \widehat{D}(s):=U_{0}(s)^{\dagger}D(s)U_{0}(s) D ( s ) := U 0 ( s ) † D ( s ) U 0 ( s ) , a first-order Dyson expansion yields
Φ ~ ε ( 1 ) = I + ε ∫ 0 1 D ^ ( s ) d s + R 2 ( ε ) . (6) \widetilde{\Phi}_{\varepsilon}(1)=\mathcal{I}+\varepsilon\int_{0}^{1}\widehat{D}(s)\,\mathrm{d}s+R_{2}(\varepsilon). \tag*{(6)} Φ ε ( 1 ) = I + ε ∫ 0 1 D ( s ) d s + R 2 ( ε ) . (6)
Because the induced operator norm of the dissipator obeys ∥ D ( s ) ∥ o p ≤ 2 Γ f u l l ( s ) \lVert D(s)\rVert_{\mathrm{op}}\leq2\Gamma_{\mathrm{full}}(s) ∥ D ( s ) ∥ op ≤ 2 Γ full ( s ) (reducing to 2 Γ ( s ) 2\Gamma(s) 2Γ ( s ) for traceless jumps c α ≡ 0 c_{\alpha}\equiv0 c α ≡ 0 ), the second-order Dyson remainder is rigorously bounded at all finite noise exposures ε ≥ 0 \varepsilon\geq0 ε ≥ 0 by
∥ R 2 ( ε ) ∥ o p ≤ 2 ε Γ f u l l , t o t 2 exp ( 2 ε Γ f u l l , t o t ) ⟶ c α ≡ 0 2 ε Γ t o t 2 exp ( 2 ε Γ t o t ) . (7) \lVert R_{2}(\varepsilon)\rVert_{\mathrm{op}}\leq2\varepsilon\Gamma_{\mathrm{full,tot}}^{2}\exp(2\varepsilon\Gamma_{\mathrm{full,tot}})
\underset{c_{\alpha}\equiv0}{\longrightarrow}2\varepsilon\Gamma_{\mathrm{tot}}^{2}\exp(2\varepsilon\Gamma_{\mathrm{tot}}). \tag*{(7)} ∥ R 2 ( ε ) ∥ op ≤ 2 ε Γ full , tot 2 exp ( 2 ε Γ full , tot ) c α ≡ 0 ⟶ 2 ε Γ tot 2 exp ( 2 ε Γ tot ) . (7)
Trajectory invariance of average gate fidelity and pure-state jump-variance law
To contrast scalar gate benchmarks with state-resolved directional errors, recall that the entanglement fidelity of Φ ε ( 1 ) \Phi_{\varepsilon}(1) Φ ε ( 1 ) relative to the target unitary U 0 ( 1 ) U_{0}(1) U 0 ( 1 ) is F e ( Φ ε , U 0 ( 1 ) ) : = d − 2 Tr ( U 0 ( 1 ) † Φ ε ) F_{e}(\Phi_{\varepsilon},U_{0}(1)):=d^{-2}\operatorname{Tr}(U_{0}(1)^{\dagger}\Phi_{\varepsilon}) F e ( Φ ε , U 0 ( 1 )) := d − 2 Tr ( U 0 ( 1 ) † Φ ε ) [7 , 10 ] , and the Haar-averaged gate infidelity is r a v g : = d d + 1 ( 1 − F e ) r_{\mathrm{avg}}:=\frac{d}{d+1}(1-F_{e}) r avg := d + 1 d ( 1 − F e ) [5 , 6 ] . Because the Hamiltonian superoperator ad H ( s ) \operatorname{ad}_{H(s)} ad H ( s ) is traceless and Tr ( D ( s ) ) = − d Γ ( s ) \operatorname{Tr}(D(s))=-d\Gamma(s) Tr ( D ( s )) = − d Γ ( s ) on the d 2 d^{2} d 2 -dimensional operator space (see Appendix § 13 for the full derivation [10 , 11 , 35 , 36 ] ), Jacobi’s determinant identity implies that the Liouville determinant of the noisy channel is locked to the integrated trace rate at all orders in ε ≥ 0 \varepsilon\geq0 ε ≥ 0 :
det Φ ε = exp ( − d ε Γ t o t ) . (8) \det\Phi_{\varepsilon}=\exp(-d\varepsilon\Gamma_{\mathrm{tot}}). \tag*{(8)} det Φ ε = exp ( − d ε Γ tot ) . (8)
Expanding Eq. (6) to first order in ε \varepsilon ε recovers the first-order trajectory invariance of entanglement fidelity and average gate infidelity [10 , 11 ] :
1 − F e ( Φ ε , U 0 ( 1 ) ) = ε Γ t o t d + O ( ε 2 ) , r a v g ( Φ ε , U 0 ( 1 ) ) = ε Γ t o t d + 1 + O ( ε 2 ) , (9) \begin{aligned}
1-F_{e}(\Phi_{\varepsilon},U_{0}(1))&=\frac{\varepsilon\Gamma_{\mathrm{tot}}}{d}+\mathcal{O}(\varepsilon^{2}),\\
r_{\mathrm{avg}}(\Phi_{\varepsilon},U_{0}(1))&=\frac{\varepsilon\Gamma_{\mathrm{tot}}}{d+1}+\mathcal{O}(\varepsilon^{2}), \tag*{(9)}
\end{aligned} 1 − F e ( Φ ε , U 0 ( 1 )) r avg ( Φ ε , U 0 ( 1 )) = d ε Γ tot + O ( ε 2 ) , = d + 1 ε Γ tot + O ( ε 2 ) , (9)
both of which are completely independent of the control trajectory U 0 ( s ) U_{0}(s) U 0 ( s ) . Furthermore, by Schur’s lemma, Tr ( D ) = − d Tr I \operatorname{Tr}(D)=-d\operatorname{Tr}I Tr ( D ) = − d Tr I is, up to scalar multiplication, the unique S U ( d ) \mathrm{SU}(d) SU ( d ) -conjugation-invariant linear functional on trace-preserving superoperators (Appendix § 13 ).
In an untwirled circuit initialized in a specific pure state ∣ ψ i n ⟩ |\psi_{\mathrm{in}}\rangle ∣ ψ in ⟩ , however, the state-to-state survival error probability p e r r ( ε ; ψ i n ) : = 1 − ⟨ ψ o u t ∣ Φ ε ( ∣ ψ i n ⟩ ⟨ ψ i n ∣ ) ∣ ψ o u t ⟩ = ε χ ( ψ i n ) + O ( ε 2 ) p_{\mathrm{err}}(\varepsilon;\psi_{\mathrm{in}}):=1-\langle\psi_{\mathrm{out}}|\Phi_{\varepsilon}(|\psi_{\mathrm{in}}\rangle\langle\psi_{\mathrm{in}}|)|\psi_{\mathrm{out}}\rangle=\varepsilon\chi(\psi_{\mathrm{in}})+O(\varepsilon^2) p err ( ε ; ψ in ) := 1 − ⟨ ψ out ∣ Φ ε ( ∣ ψ in ⟩ ⟨ ψ in ∣ ) ∣ ψ out ⟩ = ε χ ( ψ in ) + O ( ε 2 ) relative to the ideal output state ∣ ψ o u t ⟩ = U 0 ( 1 ) ∣ ψ i n ⟩ |\psi_{\mathrm{out}}\rangle=U_0(1)|\psi_{\mathrm{in}}\rangle ∣ ψ out ⟩ = U 0 ( 1 ) ∣ ψ in ⟩ is governed by a different physical law. Let ∣ ψ ( s ) ⟩ : = U 0 ( s ) ∣ ψ i n ⟩ |\psi(s)\rangle:=U_0(s)|\psi_{\mathrm{in}}\rangle ∣ ψ ( s )⟩ := U 0 ( s ) ∣ ψ in ⟩ and P ( s ) : = ∣ ψ ( s ) ⟩ ⟨ ψ ( s ) ∣ P(s):=|\psi(s)\rangle\langle\psi(s)| P ( s ) := ∣ ψ ( s )⟩ ⟨ ψ ( s ) ∣ denote the evolving pure state along the control trajectory. Substituting ∣ ψ o u t ⟩ ⟨ ψ o u t ∣ = U 0 ( 1 ) ( P ( 0 ) ) |\psi_{\mathrm{out}}\rangle\langle\psi_{\mathrm{out}}|=U_0(1)(P(0)) ∣ ψ out ⟩ ⟨ ψ out ∣ = U 0 ( 1 ) ( P ( 0 )) into the Dyson expansion (6) gives
⟨ ψ o u t ∣ Φ ε ( P ( 0 ) ) ∣ ψ o u t ⟩ = Tr ( P ( 0 ) Φ ~ ε ( 1 ) ( P ( 0 ) ) ) = 1 + ε ∫ 0 1 Tr ( P ( s ) D ( s ) ( P ( s ) ) ) d s + O ( ε 2 ) . (10) \begin{aligned}
\langle\psi_{\mathrm{out}}|\Phi_{\varepsilon}(P(0))|\psi_{\mathrm{out}}\rangle&= \operatorname{Tr}\left(P(0)\widetilde{\Phi}_{\varepsilon}(1)(P(0))\right) \\
&= 1+\varepsilon\int_0^1\operatorname{Tr}\left(P(s)\mathcal{D}(s)(P(s))\right)\,\mathrm{d}s+O(\varepsilon^2). \tag*{(10)}
\end{aligned} ⟨ ψ out ∣ Φ ε ( P ( 0 )) ∣ ψ out ⟩ = Tr ( P ( 0 ) Φ ε ( 1 ) ( P ( 0 )) ) = 1 + ε ∫ 0 1 Tr ( P ( s ) D ( s ) ( P ( s )) ) d s + O ( ε 2 ) . (10)
Because Tr ( P ( s ) [ Δ H L a m b ( s ) , P ( s ) ] ) = 0 \operatorname{Tr}(P(s)[\Delta H_{\mathrm{Lamb}}(s),P(s)])=0 Tr ( P ( s ) [ Δ H Lamb ( s ) , P ( s )]) = 0 and P ( s ) 2 = P ( s ) = ∣ ψ ( s ) ⟩ ⟨ ψ ( s ) ∣ P(s)^2=P(s)=|\psi(s)\rangle\langle\psi(s)| P ( s ) 2 = P ( s ) = ∣ ψ ( s )⟩ ⟨ ψ ( s ) ∣ , the recycling and anticommutator terms of each traceless jump operator L ~ α = L ~ α ( s ) \widetilde{L}_{\alpha}=\widetilde{L}_{\alpha}(s) L α = L α ( s ) evaluate to
Tr ( P L ~ α P L ~ α † ) = ∣ ⟨ ψ ( s ) ∣ L ~ α ∣ ψ ( s ) ⟩ ∣ 2 , \operatorname{Tr}\left(P\widetilde{L}_{\alpha}P\widetilde{L}_{\alpha}^{\dagger}\right)=|\langle\psi(s)|\widetilde{L}_{\alpha}|\psi(s)\rangle|^2, Tr ( P L α P L α † ) = ∣ ⟨ ψ ( s ) ∣ L α ∣ ψ ( s )⟩ ∣ 2 ,
1 2 Tr ( P { L ~ α † L ~ α , P } ) = ⟨ ψ ( s ) ∣ L ~ α † L ~ α ∣ ψ ( s ) ⟩ . (11) \frac{1}{2}\operatorname{Tr}\left(P\{\widetilde{L}_{\alpha}^{\dagger}\widetilde{L}_{\alpha},P\}\right)=\langle\psi(s)|\widetilde{L}_{\alpha}^{\dagger}\widetilde{L}_{\alpha}|\psi(s)\rangle. \tag*{(11)} 2 1 Tr ( P { L α † L α , P } ) = ⟨ ψ ( s ) ∣ L α † L α ∣ ψ ( s )⟩ . (11)
Subtracting the two terms in Eq. (11) establishes the pure-state jump-variance law.
Proposition 1 (Pure-state jump-variance law). For any finite-dimensional open quantum system ( H ≅ C d , d ≥ 2 ) (\mathcal{H}\cong\mathbb{C}^{d},\ d\ge2) ( H ≅ C d , d ≥ 2 ) evolving under the time-rescaled GKSL master equation (2) , the first-order survival error slope χ ( ψ i n ) = lim ε → 0 + p e r r ( ε ; ψ i n ) / ε \chi(\psi_{\mathrm{in}})=\lim_{\varepsilon\to0+}p_{\mathrm{err}}(\varepsilon;\psi_{\mathrm{in}})/\varepsilon χ ( ψ in ) = lim ε → 0 + p err ( ε ; ψ in ) / ε of any pure input state ∣ ψ i n ⟩ |\psi_{\mathrm{in}}\rangle ∣ ψ in ⟩ along the unitary control trajectory P ( s ) = U 0 ( s ) ∣ ψ i n ⟩ ⟨ ψ i n ∣ U 0 ( s ) † P(s)=U_0(s)|\psi_{\mathrm{in}}\rangle\langle\psi_{\mathrm{in}}|U_0(s)^{\dagger} P ( s ) = U 0 ( s ) ∣ ψ in ⟩ ⟨ ψ in ∣ U 0 ( s ) † is given by the integrated quantum variance of the traceless jump operators:
χ ( ψ i n ) = − ∫ 0 1 Tr ( P ( s ) D ( s ) ( P ( s ) ) ) d s = ∫ 0 1 ∑ α γ α ( s ) Var P ( s ) ( L ~ α ( s ) ) d s ≥ 0 , (12) \begin{aligned}
\chi(\psi_{\mathrm{in}})&=-\int_0^1\operatorname{Tr}\left(P(s)\mathcal{D}(s)(P(s))\right)\,\mathrm{d}s \\
&=\int_0^1\sum_{\alpha}\gamma_{\alpha}(s)\operatorname{Var}_{P(s)}(\widetilde{L}_{\alpha}(s))\,\mathrm{d}s\ge0, \tag*{(12)}
\end{aligned} χ ( ψ in ) = − ∫ 0 1 Tr ( P ( s ) D ( s ) ( P ( s )) ) d s = ∫ 0 1 α ∑ γ α ( s ) Var P ( s ) ( L α ( s )) d s ≥ 0 , (12) where Var P ( s ) ( L ) : = ⟨ ψ ( s ) ∣ L † L ∣ ψ ( s ) ⟩ − ∣ ⟨ ψ ( s ) ∣ L ∣ ψ ( s ) ⟩ ∣ 2 ≥ 0 \operatorname{Var}_{P(s)}(L):=\langle\psi(s)|L^{\dagger}L|\psi(s)\rangle-|\langle\psi(s)|L|\psi(s)\rangle|^2\ge0 Var P ( s ) ( L ) := ⟨ ψ ( s ) ∣ L † L ∣ ψ ( s )⟩ − ∣ ⟨ ψ ( s ) ∣ L ∣ ψ ( s )⟩ ∣ 2 ≥ 0 is the quantum fluctuation variance of L L L in the instantaneous pure state ∣ ψ ( s ) ⟩ |\psi(s)\rangle ∣ ψ ( s )⟩ . For a two-level qubit ( d = 2 ) (d=2) ( d = 2 ) evolving along any continuous, non-constant trajectory s ↦ P ( s ) s\mapsto P(s) s ↦ P ( s ) on the Bloch sphere under a time-independent bath with Γ > 0 \Gamma>0 Γ > 0 , the survival error slope is strictly positive ( χ ( ψ i n ) > 0 ) (\chi(\psi_{\mathrm{in}})>0) ( χ ( ψ in ) > 0 ) .
Proof. Inserting Eq. (11) into Eq. (10) immediately yields Eq. (12) . By the Cauchy–Schwarz inequality ∣ ⟨ ψ ( s ) ∣ L ~ α ∣ ψ ( s ) ⟩ ∣ 2 ≤ ⟨ ψ ( s ) ∣ ψ ( s ) ⟩ ⟨ ψ ( s ) ∣ L ~ α † L ~ α ∣ ψ ( s ) ⟩ |\langle\psi(s)|\widetilde{L}_{\alpha}|\psi(s)\rangle|^2\le\langle\psi(s)|\psi(s)\rangle\langle\psi(s)|\widetilde{L}_{\alpha}^{\dagger}\widetilde{L}_{\alpha}|\psi(s)\rangle ∣ ⟨ ψ ( s ) ∣ L α ∣ ψ ( s )⟩ ∣ 2 ≤ ⟨ ψ ( s ) ∣ ψ ( s )⟩ ⟨ ψ ( s ) ∣ L α † L α ∣ ψ ( s )⟩ , the integrand Var P ( s ) ( L ~ α ( s ) ) \operatorname{Var}_{P(s)}(\widetilde{L}_{\alpha}(s)) Var P ( s ) ( L α ( s )) is non-negative and vanishes at time s s s if and only if ∣ ψ ( s ) ⟩ |\psi(s)\rangle ∣ ψ ( s )⟩ is a right eigenstate (instantaneous dark state) of L ~ α ( s ) \widetilde{L}_{\alpha}(s) L α ( s ) . For d = 2 d=2 d = 2 , any non-zero 2 × 2 2\times2 2 × 2 traceless jump operator L ~ α \widetilde{L}_{\alpha} L α has at most two projective eigenstates on the Bloch sphere S 2 S^2 S 2 , so a continuous, non-constant trajectory s ↦ P ( s ) s\mapsto P(s) s ↦ P ( s ) cannot remain in an eigenstate over any subinterval of positive measure, ensuring χ ( ψ i n ) > 0 \chi(\psi_{\mathrm{in}})>0 χ ( ψ in ) > 0 whenever Γ > 0 \Gamma>0 Γ > 0 . □
Proposition 1 has a direct quantum-optical interpretation: the instantaneous rate at which a pure state loses survival fidelity equals the quantum fluctuation variance of the environmental jump operators along the trajectory, vanishing at time s s s if and only if ∣ ψ ( s ) ⟩ |\psi(s)\rangle ∣ ψ ( s )⟩ is an instantaneous dark state of every active jump operator L ~ α ( s ) \widetilde{L}_{\alpha}(s) L α ( s ) . Because a non-trivial 2 × 2 2\times2 2 × 2 jump operator possesses at most two discrete dark states on the Bloch sphere, a continuous non-constant gate trajectory cannot remain trapped in a decoherence-free dark state throughout s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] .
Worked physical examples: native T 1 T_1 T 1 decay on X \sqrt{X} X and extremal chiral dissipation on X X X
Before developing the general tensor theory, we illustrate how the jump-variance law (12) predicts directional error splitting in two concrete physical settings. For orthogonal computational basis states ∣ ψ + ⟩ = ∣ 0 ⟩ |\psi_+\rangle=|0\rangle ∣ ψ + ⟩ = ∣0 ⟩ and ∣ ψ − ⟩ = ∣ 1 ⟩ |\psi_-\rangle=|1\rangle ∣ ψ − ⟩ = ∣1 ⟩ mapped to target output states U 0 ( 1 ) ∣ 0 ⟩ U_0(1)|0\rangle U 0 ( 1 ) ∣0 ⟩ and U 0 ( 1 ) ∣ 1 ⟩ U_0(1)|1\rangle U 0 ( 1 ) ∣1 ⟩ , the forward and backward survival error probabilities are p ± ( ε ) : = 1 − ⟨ ψ ± ∣ U 0 ( 1 ) † Φ ε ( ∣ ψ ± ⟩ ⟨ ψ ± ∣ ) U 0 ( 1 ) ∣ ψ ± ⟩ = ε χ ± + O ( ε 2 ) p_{\pm}(\varepsilon):=1-\langle\psi_{\pm}|U_0(1)^{\dagger}\Phi_{\varepsilon}(|\psi_{\pm}\rangle\langle\psi_{\pm}|)U_0(1)|\psi_{\pm}\rangle=\varepsilon\chi_{\pm}+O(\varepsilon^2) p ± ( ε ) := 1 − ⟨ ψ ± ∣ U 0 ( 1 ) † Φ ε ( ∣ ψ ± ⟩ ⟨ ψ ± ∣ ) U 0 ( 1 ) ∣ ψ ± ⟩ = ε χ ± + O ( ε 2 ) .
Example 1 (Native longitudinal T 1 T_1 T 1 decay on a polar X \sqrt{X} X gate). Consider a standard polar X \sqrt{X} X gate U t a r g e t = exp ( i π 4 σ x ) U_{\mathrm{target}}=\exp(i\frac{\pi}{4}\sigma_x) U target = exp ( i 4 π σ x ) driven by H X = − π 4 σ x H_{\sqrt{X}}=-\frac{\pi}{4}\sigma_x H X = − 4 π σ x on s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] under native longitudinal T 1 T_1 T 1 relaxation (L ↓ = ∣ 0 ⟩ ⟨ 1 ∣ = 1 2 ( σ x + i σ y ) L_{\downarrow}=|0\rangle\langle1|=\frac{1}{2}(\sigma_x+i\sigma_y) L ↓ = ∣0 ⟩ ⟨ 1∣ = 2 1 ( σ x + i σ y ) , Γ = Tr ( L ↓ † L ↓ ) = 1 \Gamma=\operatorname{Tr}(L_{\downarrow}^{\dagger}L_{\downarrow})=1 Γ = Tr ( L ↓ † L ↓ ) = 1 ). The evolving pure-state projectors are P ± ( s ) = 1 2 [ I 2 ± sin ( π s 2 ) σ y ± cos ( π s 2 ) σ z ] P_{\pm}(s)=\frac{1}{2}[\mathbb{I}_2\pm\sin(\frac{\pi s}{2})\sigma_y\pm\cos(\frac{\pi s}{2})\sigma_z] P ± ( s ) = 2 1 [ I 2 ± sin ( 2 π s ) σ y ± cos ( 2 π s ) σ z ] . Since L ↓ † L ↓ = ∣ 1 ⟩ ⟨ 1 ∣ = 1 2 ( I 2 − σ z ) L_{\downarrow}^{\dagger}L_{\downarrow}=|1\rangle\langle1|=\frac{1}{2}(\mathbb{I}_2-\sigma_z) L ↓ † L ↓ = ∣1 ⟩ ⟨ 1∣ = 2 1 ( I 2 − σ z ) , we have Tr ( P ± ( s ) L ↓ † L ↓ ) = 1 2 [ 1 ∓ cos ( π s 2 ) ] \operatorname{Tr}(P_{\pm}(s)L_{\downarrow}^{\dagger}L_{\downarrow})=\frac{1}{2}[1\mp\cos(\frac{\pi s}{2})] Tr ( P ± ( s ) L ↓ † L ↓ ) = 2 1 [ 1 ∓ cos ( 2 π s )] and ∣ Tr ( P ± ( s ) L ↓ ) ∣ 2 = 1 4 sin 2 ( π s 2 ) |\operatorname{Tr}(P_{\pm}(s)L_{\downarrow})|^2=\frac{1}{4}\sin^2(\frac{\pi s}{2}) ∣ Tr ( P ± ( s ) L ↓ ) ∣ 2 = 4 1 sin 2 ( 2 π s ) , giving the instantaneous jump variance
Var P ± ( s ) ( L ↓ ) = 1 2 ∓ 1 2 cos ( π s 2 ) − 1 4 sin 2 ( π s 2 ) . \operatorname{Var}_{P_{\pm}(s)}(L_{\downarrow})=\frac{1}{2}\mp\frac{1}{2}\cos\left(\frac{\pi s}{2}\right)-\frac{1}{4}\sin^2\left(\frac{\pi s}{2}\right). Var P ± ( s ) ( L ↓ ) = 2 1 ∓ 2 1 cos ( 2 π s ) − 4 1 sin 2 ( 2 π s ) .
Integrating over s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] yields χ ± ( X ; T 1 ) = 3 8 ∓ 1 π \chi_{\pm}(\sqrt{X};T_1)=\frac{3}{8}\mp\frac{1}{\pi} χ ± ( X ; T 1 ) = 8 3 ∓ π 1 , corresponding to the 12.23 : 1 native T 1 T_1 T 1 survival-error split (κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 \kappa_{T_1}=(3\pi-8)/(3\pi+8)\approx0.081767 κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 ) shown in Figure 1 (b).
Example 2 (Extremal chiral dissipation on a constant-rate X X X swap). Consider a time-independent single-qubit π \pi π inversion (swap) U t a r g e t = ± i σ x U_{\mathrm{target}}=\pm i\sigma_x U target = ± i σ x driven by H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x on s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] for an odd integer winding number q ∈ { ± 1 , ± 3 , … } q\in\{\pm1,\pm3,\ldots\} q ∈ { ± 1 , ± 3 , … } , where the forward and backward survival errors are
p + ( ε ; q ) : = 1 − ⟨ 1 ∣ Φ ε ( ∣ 0 ⟩ ⟨ 0 ∣ ) ∣ 1 ⟩ = Tr ( ∣ 0 ⟩ ⟨ 0 ∣ Φ ε ( ∣ 0 ⟩ ⟨ 0 ∣ ) ) , p − ( ε ; q ) : = 1 − ⟨ 0 ∣ Φ ε ( ∣ 1 ⟩ ⟨ 1 ∣ ) ∣ 0 ⟩ = Tr ( ∣ 1 ⟩ ⟨ 1 ∣ Φ ε ( ∣ 1 ⟩ ⟨ 1 ∣ ) ) . (13) \begin{aligned}
p_{+}(\varepsilon;q)&:=1-\langle1|\Phi_{\varepsilon}(|0\rangle\langle0|)|1\rangle=\operatorname{Tr}\left(|0\rangle\langle0|\Phi_{\varepsilon}(|0\rangle\langle0|)\right),\\
p_{-}(\varepsilon;q)&:=1-\langle0|\Phi_{\varepsilon}(|1\rangle\langle1|)|0\rangle=\operatorname{Tr}\left(|1\rangle\langle1|\Phi_{\varepsilon}(|1\rangle\langle1|)\right). \tag*{(13)}
\end{aligned} p + ( ε ; q ) p − ( ε ; q ) := 1 − ⟨ 1∣ Φ ε ( ∣0 ⟩ ⟨ 0∣ ) ∣1 ⟩ = Tr ( ∣0 ⟩ ⟨ 0∣ Φ ε ( ∣0 ⟩ ⟨ 0∣ ) ) , := 1 − ⟨ 0∣ Φ ε ( ∣1 ⟩ ⟨ 1∣ ) ∣0 ⟩ = Tr ( ∣1 ⟩ ⟨ 1∣ Φ ε ( ∣1 ⟩ ⟨ 1∣ ) ) . (13)
Every odd q q q has the exact same first-order entanglement infidelity 1 − F e = 1 2 ε Γ + O ( ε 2 ) 1-F_e=\frac{1}{2}\varepsilon\Gamma+O(\varepsilon^2) 1 − F e = 2 1 ε Γ + O ( ε 2 ) . Under the engineered chiral jump operator L + = ( σ x + i 2 σ z ) / 6 L_{+}=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 (Γ = Tr ( L + † L + ) = 1 \Gamma=\operatorname{Tr}(L_{+}^{\dagger}L_{+})=1 Γ = Tr ( L + † L + ) = 1 ), however, the survival errors split as
p ± ( ε ; q ) = ε χ ± ( q ) + O ( ε 2 ) , χ ± ( q ) = 1 3 ± 2 2 3 π q . (14) p_{\pm}(\varepsilon;q)=\varepsilon\chi_{\pm}(q)+O(\varepsilon^2),\qquad\chi_{\pm}(q)=\frac{1}{3}\pm\frac{2\sqrt{2}}{3\pi q}. \tag*{(14)} p ± ( ε ; q ) = ε χ ± ( q ) + O ( ε 2 ) , χ ± ( q ) = 3 1 ± 3 π q 2 2 . (14)
Indeed, for H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x , U 0 ( s ) = cos ( q π s 2 ) I 2 + i sin ( q π s 2 ) σ x U_0(s)=\cos(\frac{q\pi s}{2})\mathbb{I}_2+i\sin(\frac{q\pi s}{2})\sigma_x U 0 ( s ) = cos ( 2 q π s ) I 2 + i sin ( 2 q π s ) σ x evolves ∣ 0 ⟩ , ∣ 1 ⟩ |0\rangle,|1\rangle ∣0 ⟩ , ∣1 ⟩ along P ± ( s ) = 1 2 [ I 2 ± sin ( q π s ) σ y ± cos ( q π s ) σ z ] P_{\pm}(s)=\frac{1}{2}[\mathbb{I}_2\pm\sin(q\pi s)\sigma_y\pm\cos(q\pi s)\sigma_z] P ± ( s ) = 2 1 [ I 2 ± sin ( q π s ) σ y ± cos ( q π s ) σ z ] . Using σ x σ z = − i σ y \sigma_x\sigma_z=-i\sigma_y σ x σ z = − i σ y , we find
L + † L + = 1 6 ( σ x − i 2 σ z ) ( σ x + i 2 σ z ) = 1 6 ( 3 I 2 + 2 2 σ y ) . (15) L_{+}^{\dagger}L_{+}=\frac{1}{6}(\sigma_x-i\sqrt{2}\sigma_z)(\sigma_x+i\sqrt{2}\sigma_z)=\frac{1}{6}(3\mathbb{I}_2+2\sqrt{2}\sigma_y). \tag*{(15)} L + † L + = 6 1 ( σ x − i 2 σ z ) ( σ x + i 2 σ z ) = 6 1 ( 3 I 2 + 2 2 σ y ) . (15)
so Tr ( P ± ( s ) L + † L + ) = 1 2 ± 2 3 sin ( q π s ) \operatorname{Tr}(P_{\pm}(s)L_{+}^{\dagger}L_{+})=\frac{1}{2}\pm\frac{\sqrt{2}}{3}\sin(q\pi s) Tr ( P ± ( s ) L + † L + ) = 2 1 ± 3 2 sin ( q π s ) and ∣ Tr ( P ± ( s ) L + ) ∣ 2 = 1 3 cos 2 ( q π s ) |\operatorname{Tr}(P_{\pm}(s)L_{+})|^2=\frac{1}{3}\cos^2(q\pi s) ∣ Tr ( P ± ( s ) L + ) ∣ 2 = 3 1 cos 2 ( q π s ) . Subtracting gives the instantaneous jump variance
Var P ± ( s ) ( L + ) = 1 2 − 1 3 cos 2 ( q π s ) ± 2 3 sin ( q π s ) , (16) \operatorname{Var}_{P_{\pm}(s)}(L_{+})=\frac{1}{2}-\frac{1}{3}\cos^2(q\pi s)\pm\frac{\sqrt{2}}{3}\sin(q\pi s), \tag*{(16)} Var P ± ( s ) ( L + ) = 2 1 − 3 1 cos 2 ( q π s ) ± 3 2 sin ( q π s ) , (16)
whose integral over s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] yields Eq. (14) . At q = 1 q=1 q = 1 , χ + ( 1 ) ≈ 0.633439 \chi_{+}(1)\approx0.633439 χ + ( 1 ) ≈ 0.633439 , χ − ( 1 ) ≈ 0.033228 \chi_{-}(1)\approx0.033228 χ − ( 1 ) ≈ 0.033228 , and χ − ( 1 ) / χ + ( 1 ) = ( π − 2 2 ) / ( π + 2 2 ) = κ ⋆ ≈ 0.052456 \chi_{-}(1)/\chi_{+}(1)=(\pi-2\sqrt{2})/(\pi+2\sqrt{2})=\kappa_{\star}\approx0.052456 χ − ( 1 ) / χ + ( 1 ) = ( π − 2 2 ) / ( π + 2 2 ) = κ ⋆ ≈ 0.052456 (19.0635 : 1). We now prove that both examples are special cases of a universal two-moment trajectory law.
Two-moment trajectory law and geometric cofactor bound
In the weak-damping Markovian regime for a two-level qubit (d = 2 d=2 d = 2 ) coupled to a time-independent bath, every traceless jump operator L ~ α \widetilde{L}_{\alpha} L α expands in the Pauli operator basis σ = ( σ x , σ y , σ z ) T \boldsymbol{\sigma}=(\sigma_x,\sigma_y,\sigma_z)^{T} σ = ( σ x , σ y , σ z ) T as
L ~ α = ( a α + i b α ) ⋅ σ , a α , b α ∈ R 3 . (17) \widetilde{L}_{\alpha}=(\boldsymbol{a}_{\alpha}+i\boldsymbol{b}_{\alpha})\cdot\boldsymbol{\sigma},\qquad\boldsymbol{a}_{\alpha},\boldsymbol{b}_{\alpha}\in\mathbb{R}^{3}. \tag*{(17)} L α = ( a α + i b α ) ⋅ σ , a α , b α ∈ R 3 . (17)
The Pauli algebra identity ( u ⋅ σ ) ( v ⋅ σ ) = ( u ⋅ v ) I 2 + i ( u × v ) ⋅ σ (\boldsymbol{u}\cdot\boldsymbol{\sigma})(\boldsymbol{v}\cdot\boldsymbol{\sigma})=(\boldsymbol{u}\cdot\boldsymbol{v})\mathbb{I}_2+i(\boldsymbol{u}\times\boldsymbol{v})\cdot\boldsymbol{\sigma} ( u ⋅ σ ) ( v ⋅ σ ) = ( u ⋅ v ) I 2 + i ( u × v ) ⋅ σ yields
L ~ α † L ~ α = ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) I 2 − 2 ( a α × b α ) ⋅ σ , L ~ α L ~ α † = ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) I 2 + 2 ( a α × b α ) ⋅ σ . (18) \begin{aligned}
\widetilde{L}_{\alpha}^{\dagger}\widetilde{L}_{\alpha}&=(\|\boldsymbol{a}_{\alpha}\|^2+\|\boldsymbol{b}_{\alpha}\|^2)\mathbb{I}_2-2(\boldsymbol{a}_{\alpha}\times\boldsymbol{b}_{\alpha})\cdot\boldsymbol{\sigma},\\
\widetilde{L}_{\alpha}\widetilde{L}_{\alpha}^{\dagger}&=(\|\boldsymbol{a}_{\alpha}\|^2+\|\boldsymbol{b}_{\alpha}\|^2)\mathbb{I}_2+2(\boldsymbol{a}_{\alpha}\times\boldsymbol{b}_{\alpha})\cdot\boldsymbol{\sigma}. \tag*{(18)}
\end{aligned} L α † L α L α L α † = ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) I 2 − 2 ( a α × b α ) ⋅ σ , = ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) I 2 + 2 ( a α × b α ) ⋅ σ . (18)
Definition 1 (Markovian bath tensors and control trajectory moments). Any time-independent Markovian qubit bath D \mathcal{D} D with traceless jump expansion (17) is characterized by its scalar trace rate
Γ ≥ 0 \Gamma\ge0 Γ ≥ 0 , symmetric positive-semidefinite unital damping tensor S ∈ R 3 × 3 S \in\mathbb{R}^{3\times3} S ∈ R 3 × 3 , and non-unital polarization drift vector t ∈ R 3 t \in\mathbb{R}^{3} t ∈ R 3 :
Γ = 2 ∑ α γ α ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) = Tr ( S ) , \Gamma= 2\sum_{\alpha}\gamma_{\alpha}\left(\lVert a_{\alpha}\rVert^{2}+\lVert b_{\alpha}\rVert^{2}\right)=\operatorname{Tr}(S), Γ = 2 α ∑ γ α ( ∥ a α ∥ 2 + ∥ b α ∥ 2 ) = Tr ( S ) ,
S : = 2 ∑ α γ α ( a α a α T + b α b α T ) ⪰ 0 , (19) S := 2\sum_{\alpha}\gamma_{\alpha}\left(a_{\alpha}a_{\alpha}^{\mathsf{T}}+b_{\alpha}b_{\alpha}^{\mathsf{T}}\right)\succeq0, \tag*{(19)} S := 2 α ∑ γ α ( a α a α T + b α b α T ) ⪰ 0 , (19)
t : = 8 ∑ α γ α ( a α × b α ) = Tr ( σ D ( I 2 ) ) . t := 8\sum_{\alpha}\gamma_{\alpha}(a_{\alpha}\times b_{\alpha})=\operatorname{Tr}(\sigma\,\mathcal{D}(\mathbb{I}_{2})). t := 8 α ∑ γ α ( a α × b α ) = Tr ( σ D ( I 2 )) .
For orthogonal antipodal input states ∣ ψ + ⟩ = ∣ 0 ⟩ \lvert\psi_{+}\rangle=\lvert0\rangle ∣ ψ + ⟩ = ∣ 0 ⟩ and ∣ ψ − ⟩ = ∣ 1 ⟩ \lvert\psi_{-}\rangle=\lvert1\rangle ∣ ψ − ⟩ = ∣ 1 ⟩ , the evolving pure-state projectors P ± ( s ) : = U 0 ( s ) ∣ ψ ± ⟩ ⟨ ψ ± ∣ U 0 ( s ) † P_{\pm}(s):=U_{0}(s)\lvert\psi_{\pm}\rangle\langle\psi_{\pm}\rvert U_{0}(s)^{\dagger} P ± ( s ) := U 0 ( s ) ∣ ψ ± ⟩ ⟨ ψ ± ∣ U 0 ( s ) † trace out antipodal curves on the Bloch sphere:
P ± ( s ) = 1 2 ( I 2 ± r ( s ) ⋅ σ ) , ∥ r ( s ) ∥ = 1. (20) P_{\pm}(s)=\frac{1}{2}\left(\mathbb{I}_{2}\pm r(s)\cdot\sigma\right),\qquad\lVert r(s)\rVert=1. \tag*{(20)} P ± ( s ) = 2 1 ( I 2 ± r ( s ) ⋅ σ ) , ∥ r ( s )∥ = 1. (20)
The geometry of the control trajectory r : [ 0 , 1 ] → S 2 r:[0,1]\to S^{2} r : [ 0 , 1 ] → S 2 is captured by two integral moments—its time-averaged Bloch polarization vector m ∈ R 3 m\in\mathbb{R}^{3} m ∈ R 3 and its transverse residence covariance matrix (covariance deficit) G ∈ R 3 × 3 G\in\mathbb{R}^{3\times3} G ∈ R 3 × 3 :
m : = ∫ 0 1 r ( s ) d s , m:=\int_{0}^{1}r(s)\,\mathrm{d}s, m := ∫ 0 1 r ( s ) d s ,
G : = I 3 − ∫ 0 1 r ( s ) r ( s ) T d s = ∫ 0 1 [ I 3 − r ( s ) r ( s ) T ] d s . (21) G:=\mathbb{I}_{3}-\int_{0}^{1}r(s)r(s)^{\mathsf{T}}\,\mathrm{d}s=\int_{0}^{1}\left[\mathbb{I}_{3}-r(s)r(s)^{\mathsf{T}}\right]\mathrm{d}s. \tag*{(21)} G := I 3 − ∫ 0 1 r ( s ) r ( s ) T d s = ∫ 0 1 [ I 3 − r ( s ) r ( s ) T ] d s . (21)
Physically, in the affine Bloch–Redfield and Pauli transfer matrix representation (Appendix § 14 ), the matrix S − Γ I 3 S-\Gamma\mathbb{I}_{3} S − Γ I 3 is the unital Bloch relaxation damping matrix that contracts the Bloch sphere along each axis, whereas t = Tr ( σ D ( I 2 ) ) t=\operatorname{Tr}(\sigma\mathcal{D}(\mathbb{I}_{2})) t = Tr ( σ D ( I 2 )) is the non-unital polarization drift vector that drives the maximally mixed state toward the environmental steady state. Geometrically, m m m represents the net vector centroid of the trajectory inside the unit Bloch ball (which couples linearly to the non-unital bath drift t t t ), while the diagonal entries G i i = ∫ 0 1 ( 1 − r i ( s ) 2 ) d s G_{ii}=\int_{0}^{1}(1-r_{i}(s)^{2})\,\mathrm{d}s G ii = ∫ 0 1 ( 1 − r i ( s ) 2 ) d s measure the integrated transverse residence time spent away from each Bloch axis ± e i \pm e_{i} ± e i .
Theorem 1 (Two-moment trajectory law and geometric asymmetry bound). In the weak-damping Markovian regime for a two-level qubit, let r : [ 0 , 1 ] → S 2 r:[0,1]\to S^{2} r : [ 0 , 1 ] → S 2 be any piecewise-continuous Bloch trajectory with moments ( m , G ) (m,G) ( m , G ) defined in Eq. (21) , and let D \mathcal{D} D be any time-independent Markovian bath with tensors ( S , t ) (S,t) ( S , t ) defined in Eq. (19) . Then:
Exact two-moment decomposition: The first-order survival error slopes χ ± ( D ; r ) \chi_{\pm}(\mathcal{D};r) χ ± ( D ; r ) depend on r ( s ) r(s) r ( s ) exclusively through ( m , G ) (m,G) ( m , G ) :
χ ± ( D ; r ) = 1 2 Tr ( S G ) ± 1 4 m ⋅ t , \chi_{\pm}(\mathcal{D};r)=\frac{1}{2}\operatorname{Tr}(SG)\pm\frac{1}{4}m\cdot t, χ ± ( D ; r ) = 2 1 Tr ( S G ) ± 4 1 m ⋅ t , χ + ( D ; r ) + χ − ( D ; r ) = Tr ( S G ) , \chi_{+}(\mathcal{D};r)+\chi_{-}(\mathcal{D};r)=\operatorname{Tr}(SG), χ + ( D ; r ) + χ − ( D ; r ) = Tr ( S G ) , χ + ( D ; r ) − χ − ( D ; r ) = − 1 2 m ⋅ t . (22) \chi_{+}(\mathcal{D};r)-\chi_{-}(\mathcal{D};r)=-\frac{1}{2}m\cdot t. \tag*{(22)} χ + ( D ; r ) − χ − ( D ; r ) = − 2 1 m ⋅ t . (22) Geometric asymmetry bound: For every piecewise-continuous trajectory r ( s ) r(s) r ( s ) , adj ( G ) − m m T ⪰ 0 \operatorname{adj}(G)-mm^{\mathsf{T}}\succeq0 adj ( G ) − m m T ⪰ 0 . Whenever G ≻ 0 G\succ0 G ≻ 0 (det G > 0 \det G>0 det G > 0 ), the supremum of the directional contrast across all non-zero Markovian baths (Γ > 0 \Gamma>0 Γ > 0 ) is
sup Γ > 0 ∣ χ + ( D ; r ) − χ − ( D ; r ) ∣ χ + ( D ; r ) + χ − ( D ; r ) = β ( r ) \sup_{\Gamma>0}\frac{\left\lvert\chi_{+}(\mathcal{D};r)-\chi_{-}(\mathcal{D};r)\right\rvert}{\chi_{+}(\mathcal{D};r)+\chi_{-}(\mathcal{D};r)}=\beta(r) Γ > 0 sup χ + ( D ; r ) + χ − ( D ; r ) ∣ χ + ( D ; r ) − χ − ( D ; r ) ∣ = β ( r ) : = m T G m det G ∈ [ 0 , 1 ] , (23) :=\sqrt{\frac{m^{\mathsf{T}}Gm}{\det G}}\in[0,1], \tag*{(23)} := det G m T G m ∈ [ 0 , 1 ] , (23) and the ratio of the smaller to the larger first-order survival error slope obeys the universal lower bound
inf Γ > 0 min ( χ + , χ − ) max ( χ + , χ − ) = κ ( r ) : = 1 − β ( r ) 1 + β ( r ) . (24) \inf_{\Gamma>0}\frac{\min(\chi_{+},\chi_{-})}{\max(\chi_{+},\chi_{-})}=\kappa(r):=\frac{1-\beta(r)}{1+\beta(r)}. \tag*{(24)} Γ > 0 inf max ( χ + , χ − ) min ( χ + , χ − ) = κ ( r ) := 1 + β ( r ) 1 − β ( r ) . (24)
Moreover, when G ≻ 0 G \succ0 G ≻ 0 and m ≠ 0 \mathbf{m} \ne0 m = 0 , the lower bound κ ( r ) \kappa(r) κ ( r ) in Eq. (24) is uniquely attained (up to a positive scalar rate, global phase e i ϕ e^{i\phi} e i ϕ , and chirality reversal L ~ e x t ↦ L ~ e x t † \widetilde{L}_{\mathrm{ext}} \mapsto\widetilde{L}_{\mathrm{ext}}^{\dagger} L ext ↦ L ext † ) by a single rank-1 traceless jump operator L ~ e x t = ( a ⋆ + i b ⋆ ) ⋅ σ \widetilde{L}_{\mathrm{ext}} = (\mathbf{a}_{\star} + i\mathbf{b}_{\star}) \cdot\boldsymbol{\sigma} L ext = ( a ⋆ + i b ⋆ ) ⋅ σ satisfying a ⋆ T G a ⋆ = b ⋆ T G b ⋆ \mathbf{a}_{\star}^{T}G\mathbf{a}_{\star} = \mathbf{b}_{\star}^{T}G\mathbf{b}_{\star} a ⋆ T G a ⋆ = b ⋆ T G b ⋆ , a ⋆ T G b ⋆ = 0 \mathbf{a}_{\star}^{T}G\mathbf{b}_{\star} = 0 a ⋆ T G b ⋆ = 0 , and G 1 / 2 a ⋆ × G 1 / 2 b ⋆ ∥ G 1 / 2 m G^{1/2}\mathbf{a}_{\star} \times G^{1/2}\mathbf{b}_{\star} \parallel G^{1/2}\mathbf{m} G 1/2 a ⋆ × G 1/2 b ⋆ ∥ G 1/2 m (giving ( a ⋆ × b ⋆ ) ⋅ m = ∓ β ( r ) a ⋆ T G a ⋆ (\mathbf{a}_{\star} \times\mathbf{b}_{\star}) \cdot\mathbf{m} = \mp\beta(r)\mathbf{a}_{\star}^{T}G\mathbf{a}_{\star} ( a ⋆ × b ⋆ ) ⋅ m = ∓ β ( r ) a ⋆ T G a ⋆ ).
Proof sketch and physical mechanism. Evaluating Tr ( P ± ( s ) L ~ α † L ~ α ) − ∣ Tr ( P ± ( s ) L ~ α ) ∣ 2 \operatorname{Tr}(P_{\pm}(s)\widetilde{L}_{\alpha}^{\dagger}\widetilde{L}_{\alpha}) - \lvert\operatorname{Tr}(P_{\pm}(s)\widetilde{L}_{\alpha})\rvert^{2} Tr ( P ± ( s ) L α † L α ) − ∣ Tr ( P ± ( s ) L α ) ∣ 2 via Eq. (18) and integrating over s ∈ [ 0 , 1 ] s \in[0,1] s ∈ [ 0 , 1 ] yields the single-jump variance identity
∫ 0 1 Var P ± ( s ) ( L ~ α ) d s = a α T G a α + b α T G b α ∓ 2 ( a α × b α ) ⋅ m . (25) \int_{0}^{1} \operatorname{Var}_{P_{\pm}(s)}(\widetilde{L}_{\alpha})\,\mathrm{d}s = \mathbf{a}_{\alpha}^{T}G\mathbf{a}_{\alpha} + \mathbf{b}_{\alpha}^{T}G\mathbf{b}_{\alpha} \mp2(\mathbf{a}_{\alpha} \times\mathbf{b}_{\alpha}) \cdot\mathbf{m}. \tag*{(25)} ∫ 0 1 Var P ± ( s ) ( L α ) d s = a α T G a α + b α T G b α ∓ 2 ( a α × b α ) ⋅ m . (25)
Summing over α \alpha α with weights γ α \gamma_{\alpha} γ α establishes Eq. (22) (an independent 4 × 4 4 \times4 4 × 4 PTM proof is given in Appendix B). By convexity of the positive cone of Lindblad generators, the directional contrast ratio is maximized on a single jump operator:
∣ χ + − χ − ∣ χ + + χ − = 2 ∣ ( a × b ) ⋅ m ∣ a T G a + b T G b . (26) \frac{\lvert\chi_{+}-\chi_{-}\rvert}{\chi_{+}+\chi_{-}} = \frac{2\lvert(\mathbf{a}\times\mathbf{b})\cdot\mathbf{m}\rvert}{\mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b}}. \tag*{(26)} χ + + χ − ∣ χ + − χ − ∣ = a T G a + b T G b 2 ∣( a × b ) ⋅ m ∣ . (26)
Transforming to whitened coordinates a ~ = G 1 / 2 a \widetilde{\mathbf{a}} = G^{1/2}\mathbf{a} a = G 1/2 a , b ~ = G 1 / 2 b \widetilde{\mathbf{b}} = G^{1/2}\mathbf{b} b = G 1/2 b and applying the matrix cofactor identity a × b = ( det G ) − 1 / 2 G 1 / 2 ( a ~ × b ~ ) \mathbf{a} \times\mathbf{b} = (\det G)^{-1/2}G^{1/2}(\widetilde{\mathbf{a}} \times\widetilde{\mathbf{b}}) a × b = ( det G ) − 1/2 G 1/2 ( a × b ) , the Cauchy–Schwarz inequality, and the arithmetic–geometric mean inequality gives
2 ∣ ( a × b ) ⋅ m ∣ = 2 ( det G ) − 1 / 2 ∣ ( a ~ × b ~ ) ⋅ ( G 1 / 2 m ) ∣ ≤ m T G m det G ( ∥ a ~ ∥ 2 + ∥ b ~ ∥ 2 ) = β ( r ) ( a T G a + b T G b ) . (27) \begin{aligned}
2\lvert(\mathbf{a}\times\mathbf{b})\cdot\mathbf{m}\rvert&= 2(\det G)^{-1/2}\lvert(\widetilde{\mathbf{a}}\times\widetilde{\mathbf{b}})\cdot(G^{1/2}\mathbf{m})\rvert\\
&\le\sqrt{\frac{\mathbf{m}^{T}G\mathbf{m}}{\det G}(\lVert\widetilde{\mathbf{a}}\rVert^{2}+\lVert\widetilde{\mathbf{b}}\rVert^{2})} \\
&= \beta(r)(\mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b}). \tag*{(27)}
\end{aligned} 2 ∣( a × b ) ⋅ m ∣ = 2 ( det G ) − 1/2 ∣( a × b ) ⋅ ( G 1/2 m )∣ ≤ det G m T G m (∥ a ∥ 2 + ∥ b ∥ 2 ) = β ( r ) ( a T G a + b T G b ) . (27)
The complete proof of the constrained cross-product minimum, the cofactor inequality adj ( G ) − m m T ⪰ 0 \operatorname{adj}(G)-\mathbf{m}\mathbf{m}^{T}\succeq0 adj ( G ) − m m T ⪰ 0 implying β ( r ) ≤ 1 \beta(r)\le1 β ( r ) ≤ 1 , and the SO ( 2 ) \operatorname{SO}(2) SO ( 2 ) gauge uniqueness of L ~ e x t \widetilde{L}_{\mathrm{ext}} L ext is given in Appendix C. □
For an experimentalist characterizing a quantum gate, Theorem 1 states that regardless of how complicated the environmental jump operators are, the directional survival-error split between ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ and ∣ 1 ⟩ \lvert1\rangle ∣ 1 ⟩ is completely determined by the inner product − 1 2 m ⋅ t -\frac{1}{2}\mathbf{m}\cdot\mathbf{t} − 2 1 m ⋅ t between the trajectory centroid and the bath drift vector, and cannot exceed the geometric limit β ( r ) \beta(r) β ( r ) set by the pulse trajectory itself. Physically, the inequality β ( r ) ≤ 1 \beta(r)\le1 β ( r ) ≤ 1 is a direct manifestation of complete positivity: in a general non-CPTP affine linear map on the Bloch sphere, the polarization drift vector t \mathbf{t} t could be chosen arbitrarily large relative to the damping matrix S S S , whereas GKSL complete positivity links non-unital drift t = 8 ∑ α γ α ( a α × b α ) \mathbf{t}=8\sum_{\alpha}\gamma_{\alpha}(\mathbf{a}_{\alpha}\times\mathbf{b}_{\alpha}) t = 8 ∑ α γ α ( a α × b α ) to unital diffusion S = 2 ∑ α γ α ( a α a α T + b α b α T ) S=2\sum_{\alpha}\gamma_{\alpha}(\mathbf{a}_{\alpha}\mathbf{a}_{\alpha}^{T}+\mathbf{b}_{\alpha}\mathbf{b}_{\alpha}^{T}) S = 2 ∑ α γ α ( a α a α T + b α b α T ) via the cross-product bound (27) , preventing min ( χ + , χ − ) \min(\chi_{+},\chi_{-}) min ( χ + , χ − ) from vanishing whenever det G > 0 \det G>0 det G > 0 and β ( r ) < 1 \beta(r)<1 β ( r ) < 1 .
Universal selectivity bound for swaps and constant-axis quantum gates
Time-independent single-qubit swaps and winding-number compression
We now specialize to time-independent constant-amplitude control Hamiltonians H ∈ s u ( 2 ) H\in\mathfrak{su}(2) H ∈ su ( 2 ) that invert the computational basis states ∣ 0 ⟩ ↔ ∣ 1 ⟩ \lvert0\rangle\leftrightarrow\lvert1\rangle ∣ 0 ⟩ ↔ ∣ 1 ⟩ up to phase. Up to an azimuthal rotation of the transverse axes around e z \mathbf{e}_{z} e z , any such time-independent swap Hamiltonian takes the form H q : = − q π 2 σ x H_q := -\frac{q\pi}{2}\sigma_x H q := − 2 q π σ x for an odd integer winding number q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 (corresponding to total rotation angle Q = q π Q=q\pi Q = q π ), which generates the meridian Bloch trajectory r q ( s ) = ( 0 , sin ( q π s ) , cos ( q π s ) ) T \mathbf{r}_q(s)=(0,\sin(q\pi s),\cos(q\pi s))^{T} r q ( s ) = ( 0 , sin ( q π s ) , cos ( q π s ) ) T .
Theorem 2 (Universal selectivity bound of single-qubit swaps). *In the weak-damping Markovian regime, let H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 ) be any time-independent single-qubit swap Hamiltonian. Then for each odd winding number q q q , the trajectory moments (21) , geometric asymmetry parameter β q \beta_q β q , and
selectivity bound κ q \kappa_q κ q are
m q = ( 0 , 2 q π , 0 ) T , G q = diag ( 1 , 1 2 , 1 2 ) , m_q=\left(0,\frac{2}{q\pi},0\right)^{T},\qquad G_q=\operatorname{diag}\left(1,\frac{1}{2},\frac{1}{2}\right), m q = ( 0 , q π 2 , 0 ) T , G q = diag ( 1 , 2 1 , 2 1 ) ,
det G q = 1 4 , m q T G q m q = 2 q 2 π 2 . (28) \det G_q=\frac{1}{4},\qquad m_q^{T}G_qm_q=\frac{2}{q^{2}\pi^{2}}. \tag*{(28)} det G q = 4 1 , m q T G q m q = q 2 π 2 2 . (28)
β q = 2 2 π ∣ q ∣ , κ q : = inf Γ > 0 min ( χ + , χ − ) max ( χ + , χ − ) = π ∣ q ∣ − 2 2 π ∣ q ∣ + 2 2 . (29) \beta_q=\frac{2\sqrt{2}}{\pi|q|},\qquad\kappa_q:=\inf_{\Gamma>0}\frac{\min(\chi_+,\chi_-)}{\max(\chi_+,\chi_-)}=\frac{\pi|q|-2\sqrt{2}}{\pi|q|+2\sqrt{2}}. \tag*{(29)} β q = π ∣ q ∣ 2 2 , κ q := Γ > 0 inf max ( χ + , χ − ) min ( χ + , χ − ) = π ∣ q ∣ + 2 2 π ∣ q ∣ − 2 2 . (29)
Over all time-independent single-qubit swap Hamiltonians and all Markovian baths (Γ > 0 \Gamma>0 Γ > 0 ), the survival-error ratio obeys the universal lower bound
min ( χ + ( D ; H q ) , χ − ( D ; H q ) ) max ( χ + ( D ; H q ) , χ − ( D ; H q ) ) ≥ κ ⋆ : = π − 2 2 π + 2 2 , (30) \frac{\min(\chi_+(\mathcal{D};H_q),\chi_-(\mathcal{D};H_q))}{\max(\chi_+(\mathcal{D};H_q),\chi_-(\mathcal{D};H_q))}\ge\kappa_\star:=\frac{\pi-2\sqrt{2}}{\pi+2\sqrt{2}}, \tag*{(30)} max ( χ + ( D ; H q ) , χ − ( D ; H q )) min ( χ + ( D ; H q ) , χ − ( D ; H q )) ≥ κ ⋆ := π + 2 2 π − 2 2 , (30)
with κ ⋆ ≈ 0.052456 , κ ⋆ − 1 ≈ 19.0635 , \kappa_\star\approx0.052456,\qquad\kappa_\star^{-1}\approx19.0635, κ ⋆ ≈ 0.052456 , κ ⋆ − 1 ≈ 19.0635 ,
with equality at ∣ q ∣ = 1 |q|=1 ∣ q ∣ = 1 attained by a single traceless jump operator if and only if, up to a positive rate and global phase e i ϕ e^{i\phi} e i ϕ , it equals one of the two chiral partners L ± : = 1 6 ( σ x ± i 2 σ z ) L_\pm:=\frac{1}{\sqrt{6}}(\sigma_x\pm i\sqrt{2}\sigma_z) L ± := 6 1 ( σ x ± i 2 σ z ) .
Proof. For H = Q 2 n ⋅ σ H=\frac{Q}{2}\boldsymbol{n}\cdot\boldsymbol{\sigma} H = 2 Q n ⋅ σ , the swap condition ⟨ 0 ∣ e − i H ∣ 0 ⟩ = cos ( Q / 2 ) − i sin ( Q / 2 ) ( n ⋅ e z ) = 0 \langle0|e^{-iH}|0\rangle=\cos(Q/2)-i\sin(Q/2)(\boldsymbol{n}\cdot\boldsymbol{e}_z)=0 ⟨ 0∣ e − i H ∣0 ⟩ = cos ( Q /2 ) − i sin ( Q /2 ) ( n ⋅ e z ) = 0 forces Q = q π Q=q\pi Q = q π (q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 ) and n ⋅ e z = 0 \boldsymbol{n}\cdot\boldsymbol{e}_z=0 n ⋅ e z = 0 . Choosing n = − e x \boldsymbol{n}=-\boldsymbol{e}_x n = − e x yields r q ( s ) = ( 0 , sin ( q π s ) , cos ( q π s ) ) T \boldsymbol{r}_q(s)=(0,\sin(q\pi s),\cos(q\pi s))^T r q ( s ) = ( 0 , sin ( q π s ) , cos ( q π s ) ) T , whose time integrals over s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] give Eq. (28) . Substituting Eq. (28) into Eqs. (23) –(24) of Theorem 1 yields Eqs. (29) –(30) since κ q \kappa_q κ q is strictly increasing in ∣ q ∣ ≥ 1 |q|\ge1 ∣ q ∣ ≥ 1 . Finally, the extremality conditions of Theorem 1 in ( G q 1 / 2 m q ) ⊥ = span { e x , e z } (G_q^{1/2}m_q)^\perp=\operatorname{span}\{\boldsymbol{e}_x,\boldsymbol{e}_z\} ( G q 1/2 m q ) ⊥ = span { e x , e z } select a ~ = 1 6 e x \tilde{\boldsymbol{a}}=\frac{1}{\sqrt{6}}\boldsymbol{e}_x a ~ = 6 1 e x and b ~ = ± 1 6 e z \tilde{\boldsymbol{b}}=\pm\frac{1}{\sqrt{6}}\boldsymbol{e}_z b ~ = ± 6 1 e z , giving a ⋆ = 1 6 e x \boldsymbol{a}_\star=\frac{1}{\sqrt{6}}\boldsymbol{e}_x a ⋆ = 6 1 e x , b ⋆ = ± 1 3 e z \boldsymbol{b}_\star=\pm\frac{1}{\sqrt{3}}\boldsymbol{e}_z b ⋆ = ± 3 1 e z , or L ± = 1 6 ( σ x ± i 2 σ z ) L_\pm=\frac{1}{\sqrt{6}}(\sigma_x\pm i\sqrt{2}\sigma_z) L ± = 6 1 ( σ x ± i 2 σ z ) . For the tilted jump family L θ : = 1 6 [ σ x + i 2 ( cos θ σ z − sin θ σ y ) ] L_\theta:=\frac{1}{\sqrt{6}}[\sigma_x+i\sqrt{2}(\cos\theta\,\sigma_z-\sin\theta\,\sigma_y)] L θ := 6 1 [ σ x + i 2 ( cos θ σ z − sin θ σ y )] , the directional contrast is ∣ χ + − χ − ∣ / ( χ + + χ − ) = β q ∣ cos θ ∣ |\chi_+-\chi_-|/(\chi_++\chi_-)=\beta_q|\cos\theta| ∣ χ + − χ − ∣/ ( χ + + χ − ) = β q ∣ cos θ ∣ , attaining β q \beta_q β q only at θ ∈ { 0 , π } \theta\in\{0,\pi\} θ ∈ { 0 , π } .
In plain physical terms, Theorem 2 proves that no Markovian environment can cause one computational basis state to fail more than 19.0635 times as often as the other during a constant-rate single-qubit π \pi π swap; observing a larger directional ratio in an experiment immediately diagnoses departures from the autonomous two-level Markovian regime (such as non-autonomous pole idling or multilevel leakage).
Figure 2 illustrates the physical mechanism behind the 1 / ∣ q ∣ 1/|q| 1/∣ q ∣ suppression of directional asymmetry in Eq. (29) . A multi-winding swap (∣ q ∣ ≥ 3 |q|\ge3 ∣ q ∣ ≥ 3 ) executes ∣ q ∣ − 1 |q|-1 ∣ q ∣ − 1 complete 2 π 2\pi 2 π Rabi nutation cycles whose eastern (y > 0 y>0 y > 0 ) and western (y < 0 y<0 y < 0 ) hemisphere segments cancel identically in the time integral for m q m_q m q , leaving only the final π \pi π half-cycle to contribute net drift coupling ∝ 1 / ∣ q ∣ \propto1/|q| ∝ 1/∣ q ∣ . Meanwhile, the unital covariance deficit G q = diag ( 1 , 1 2 , 1 2 ) G_q=\operatorname{diag}(1,\frac{1}{2},\frac{1}{2}) G q = diag ( 1 , 2 1 , 2 1 ) is completely independent of q q q , so the average survival error slope χ ˉ = 1 2 ( χ + + χ − ) = 1 3 \bar{\chi}=\frac{1}{2}(\chi_++\chi_-)=\frac{1}{3} χ ˉ = 2 1 ( χ + + χ − ) = 3 1 remains fixed while χ + ( q ) \chi_+(q) χ + ( q ) and χ − ( q ) \chi_-(q) χ − ( q ) pinch symmetrically toward χ ˉ \bar{\chi} χ ˉ as ∣ q ∣ |q| ∣ q ∣ increases (Figure 2 (b)).
Figure 2. Winding-number compression of directional survival errors for time-independent swaps. (a) Toggling-frame Bloch component r y ( s ) = sin ( q π s ) r_y(s)=\sin(q\pi s) r y ( s ) = sin ( q π s ) for constant-amplitude swaps H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x with odd winding numbers q = + 1 q=+1 q = + 1 (solid blue), q = + 3 q=+3 q = + 3 (dash-dotted green), and q = − 1 q=-1 q = − 1 (dashed red); horizontal dotted lines mark the trajectory centroids m y ( q ) = 2 / ( q π ) m_y(q)=2/(q\pi) m y ( q ) = 2/ ( q π ) . (b) Directional survival error slopes χ + ( q ) \chi_+(q) χ + ( q ) (blue circles) and χ − ( q ) \chi_-(q) χ − ( q ) (red squares) under the extremal chiral jump L + = ( σ x + i 2 σ z ) / 6 L_+=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 (Γ = 1 \Gamma=1 Γ = 1 ), their constant average χ ˉ = 1 / 3 \bar{\chi}=1/3 χ ˉ = 1/3 (black dotted), and the selectivity ratio κ q = χ − ( q ) / χ + ( q ) \kappa_q=\chi_-(q)/\chi_+(q) κ q = χ − ( q ) / χ + ( q ) (green triangles, right axis), demonstrating that higher odd windings cancel opposing-hemisphere drift contributions and compress directional error asymmetry as 1 / ∣ q ∣ 1/|q| 1/∣ q ∣ .
Arbitrary constant-axis rotations, polar X \sqrt{X} X , and Hadamard gates
Theorem 1 applies equally to non-swap constant-axis single-qubit gates generated by H ( Q , ϑ ) = Q 2 ( sin ϑ σ x + cos ϑ σ z ) H(Q,\vartheta)=\frac{Q}{2}(\sin\vartheta\,\sigma_x+\cos\vartheta\,\sigma_z) H ( Q , ϑ ) = 2 Q ( sin ϑ σ x + cos ϑ σ z ) with total rotation angle ∣ Q ∣ > 0 |Q|>0 ∣ Q ∣ > 0 and polar axis tilt ϑ ∈ ( 0 , π ) \vartheta\in(0,\pi) ϑ ∈ ( 0 , π ) . Defining sinc 0 : = sinc ( ∣ Q ∣ / 2 ) \operatorname{sinc}_0:=\operatorname{sinc}(|Q|/2) sinc 0 := sinc ( ∣ Q ∣/2 ) and sinc 1 : = sinc ( ∣ Q ∣ ) \operatorname{sinc}_1:=\operatorname{sinc}(|Q|) sinc 1 := sinc ( ∣ Q ∣ ) (where sinc ( x ) : = sin ( x ) / x \operatorname{sinc}(x):=\sin(x)/x sinc ( x ) := sin ( x ) / x ), block-diagonalizing G G G in the trajectory midpoint frame (Appendix § 16 ) yields an exact closed-form midpoint-sinc formula.
Proposition 2 (Midpoint-sinc formula for constant-axis rotations). For any constant-axis single-qubit gate generated by H ( Q , ϑ ) = Q 2 ( sin ϑ σ x + cos ϑ σ z ) H(Q,\vartheta)=\frac{Q}{2}(\sin\vartheta\,\sigma_x+\cos\vartheta\,\sigma_z) H ( Q , ϑ ) = 2 Q ( sin ϑ σ x + cos ϑ σ z ) with ∣ Q ∣ > 0 |Q|>0 ∣ Q ∣ > 0 and ϑ ∈ ( 0 , π ) \vartheta\in(0,\pi) ϑ ∈ ( 0 , π ) , the geometric asymmetry parameter (23) evaluates in closed form to
β ( ∣ Q ∣ , ϑ ) = 2 N ( ∣ Q ∣ , ϑ ) D ( ∣ Q ∣ , ϑ ) , \beta(|Q|,\vartheta)=\sqrt{\frac{2N(|Q|,\vartheta)}{D(|Q|,\vartheta)}}, β ( ∣ Q ∣ , ϑ ) = D ( ∣ Q ∣ , ϑ ) 2 N ( ∣ Q ∣ , ϑ ) ,
N ( ∣ Q ∣ , ϑ ) : = 2 ( 1 − sinc 0 2 ) cos 2 ϑ + sinc 0 2 ( 1 − sinc 1 ) sin 2 ϑ , N(|Q|,\vartheta):=2(1-\operatorname{sinc}_0^{2})\cos^{2}\vartheta+\operatorname{sinc}_0^{2}(1-\operatorname{sinc}_1)\sin^{2}\vartheta, N ( ∣ Q ∣ , ϑ ) := 2 ( 1 − sinc 0 2 ) cos 2 ϑ + sinc 0 2 ( 1 − sinc 1 ) sin 2 ϑ ,
D ( ∣ Q ∣ , ϑ ) : = [ 2 − ( 1 − sinc 1 ) sin 2 ϑ ] [ 2 ( 1 − sinc 0 2 ) cos 2 ϑ + ( 1 − sinc 1 ) sin 2 ϑ ] . (31) D(|Q|,\vartheta):=[2-(1-\operatorname{sinc}_1)\sin^{2}\vartheta]
\left[2(1-\operatorname{sinc}_0^{2})\cos^{2}\vartheta+(1-\operatorname{sinc}_1)\sin^{2}\vartheta\right]. \tag*{(31)} D ( ∣ Q ∣ , ϑ ) := [ 2 − ( 1 − sinc 1 ) sin 2 ϑ ] [ 2 ( 1 − sinc 0 2 ) cos 2 ϑ + ( 1 − sinc 1 ) sin 2 ϑ ] . (31)
Evaluating Eq. (31) (plotted in Figure 3 (a)) for specific hardware gate families reveals a sharp physical contrast between equatorial swaps and hemisphere-biased rotations:
Figure 3. Constant-axis geometric asymmetry parameter β ( ∣ Q ∣ , ϑ ) \beta(|Q|,\vartheta) β ( ∣ Q ∣ , ϑ ) and non-autonomous pulse-dwell scaling κ ( δ ) \kappa(\delta) κ ( δ ) . (a) Geometric asymmetry parameter β ( ∣ Q ∣ , ϑ ) \beta(|Q|,\vartheta) β ( ∣ Q ∣ , ϑ ) from Eq. (31) versus normalized rotation angle ∣ Q ∣ / π |Q|/\pi ∣ Q ∣/ π for equatorial drive (ϑ = π / 2 \vartheta=\pi/2 ϑ = π /2 , solid blue), Hadamard tilt (ϑ = π / 4 \vartheta=\pi/4 ϑ = π /4 , dashed red), and ϑ = π / 6 \vartheta=\pi/6 ϑ = π /6 (dash-dotted green), with markers at polar X \sqrt{X} X , X q = 1 X_{q=1} X q = 1 , and H q = 1 H_{q=1} H q = 1 . (b) Selectivity bound κ ( δ ) \kappa(\delta) κ ( δ ) for wait-flip-wait pulse schedules with active quantum-speed-limit flip fraction δ = T Q S L / T ∈ ( 0 , 1 ] \delta=T_{\mathrm{QSL}}/T\in(0,1] δ = T QSL / T ∈ ( 0 , 1 ] (§ 8 ): one-sided wait-then-flip κ d w e l l ( δ ) \kappa_{\mathrm{dwell}}(\delta) κ dwell ( δ ) (solid blue) with linear slope ( 3 π 2 − 16 ) / ( 8 π 2 ) δ ≈ 0.172358 δ (3\pi^2-16)/(8\pi^2)\delta\approx0.172358\delta ( 3 π 2 − 16 ) / ( 8 π 2 ) δ ≈ 0.172358 δ (dashed red), contrasted against symmetric wait-flip-wait κ s y m ( δ ) \kappa_{\mathrm{sym}}(\delta) κ sym ( δ ) (dash-dotted green) and the autonomous bound κ ⋆ ≈ 0.052456 \kappa_\star\approx0.052456 κ ⋆ ≈ 0.052456 (dotted gray), showing that asymmetric pole idling degrades directional error bounds linearly in δ \delta δ .
Equatorial rotations (ϑ = π / 2 \vartheta=\pi/2 ϑ = π /2 ) and polar X \sqrt{X} X gates. For any rotation around an equatorial axis (ϑ = π / 2 \vartheta=\pi/2 ϑ = π /2 ), the factor ( 1 − sinc 1 ) (1-\operatorname{sinc}_1) ( 1 − sinc 1 ) cancels in Eq. (31) , leaving
β ( ∣ Q ∣ , π / 2 ) = 2 sinc 0 2 1 + sinc 1 = 4 ( 1 − cos ∣ Q ∣ ) ∣ Q ∣ ( ∣ Q ∣ + sin ∣ Q ∣ ) . (32) \beta(|Q|,\pi/2)=\sqrt{\frac{2\operatorname{sinc}_0^2}{1+\operatorname{sinc}_1}}=\sqrt{\frac{4(1-\cos|Q|)}{|Q|(|Q|+\sin|Q|)}}. \tag*{(32)} β ( ∣ Q ∣ , π /2 ) = 1 + sinc 1 2 sinc 0 2 = ∣ Q ∣ ( ∣ Q ∣ + sin ∣ Q ∣ ) 4 ( 1 − cos ∣ Q ∣ ) . (32)
For the principal polar X \sqrt{X} X gate (H X = − π 4 σ x H_{\sqrt{X}}=-\frac{\pi}{4}\sigma_x H X = − 4 π σ x , ∣ Q ∣ = π / 2 |Q|=\pi/2 ∣ Q ∣ = π /2 , k = 0 k=0 k = 0 ), the trajectory remains entirely in the northern hemisphere (z ≥ 0 z\ge0 z ≥ 0 , Figure 1 (b)), producing a non-diagonal covariance deficit G X G_{\sqrt{X}} G X with G y z = − 1 / π G_{yz}=-1/\pi G y z = − 1/ π :
m X = ( 0 , 2 π , 2 π ) T , G X = ( 1 0 0 0 1 2 − 1 π 0 − 1 π 1 2 ) , m_{\sqrt{X}}=\left(0,\frac{2}{\pi},\frac{2}{\pi}\right)^T,\qquad
G_{\sqrt{X}}=
\begin{pmatrix}
1 & 0 & 0 \\
0 & \frac{1}{2} & -\frac{1}{\pi} \\
0 & -\frac{1}{\pi} & \frac{1}{2}
\end{pmatrix}, m X = ( 0 , π 2 , π 2 ) T , G X = 1 0 0 0 2 1 − π 1 0 − π 1 2 1 ,
det G X = 1 4 − 1 π 2 , m X T G X m X = 4 ( π − 2 ) π 3 . (33) \det G_{\sqrt{X}}=\frac{1}{4}-\frac{1}{\pi^2},\qquad
m_{\sqrt{X}}^T G_{\sqrt{X}}m_{\sqrt{X}}=\frac{4(\pi-2)}{\pi^3}. \tag*{(33)} det G X = 4 1 − π 2 1 , m X T G X m X = π 3 4 ( π − 2 ) . (33)
Consequently, while a full X X X rotation (∣ Q ∣ = π |Q|=\pi ∣ Q ∣ = π ) has m z = 0 m_z=0 m z = 0 and zero directional split under native longitudinal T 1 T_1 T 1 decay (κ T 1 = 1 \kappa_{T_1}=1 κ T 1 = 1 ), the polar X \sqrt{X} X gate has m z = + 2 / π m_z=+2/\pi m z = + 2/ π and experiences a 12.23 : 1 native-T 1 T_1 T 1 directional survival ratio (κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 \kappa_{T_1}=(3\pi-8)/(3\pi+8)\approx0.081767 κ T 1 = ( 3 π − 8 ) / ( 3 π + 8 ) ≈ 0.081767 ), along with a universal complete-positivity bound β ( π / 2 , π / 2 ) = 4 / π ( π + 2 ) ≈ 0.995259 \beta(\pi/2,\pi/2)=4/\sqrt{\pi(\pi+2)}\approx0.995259 β ( π /2 , π /2 ) = 4/ π ( π + 2 ) ≈ 0.995259 and κ X = ( π ( π + 2 ) − 4 ) / ( π ( π + 2 ) + 4 ) ≈ 0.002376 \kappa_{\sqrt{X}}=(\sqrt{\pi(\pi+2)}-4)/(\sqrt{\pi(\pi+2)}+4)\approx0.002376 κ X = ( π ( π + 2 ) − 4 ) / ( π ( π + 2 ) + 4 ) ≈ 0.002376 (420.87 : 1). Conversely, implementing the same target unitary X \sqrt{X} X via a reversed − 3 π / 2 -3\pi/2 − 3 π /2 over-rotation (∣ Q ∣ = 3 π / 2 |Q|=3\pi/2 ∣ Q ∣ = 3 π /2 , k = 1 k=1 k = 1 ) drives the state through the opposite hemisphere for two-thirds of the pulse duration, reducing m z m_z m z to − 2 / ( 3 π ) -2/(3\pi) − 2/ ( 3 π ) and moderating the native T 1 T_1 T 1 ratio to κ T 1 = ( 9 π − 8 ) / ( 9 π + 8 ) ≈ 0.558917 \kappa_{T_1}=(9\pi-8)/(9\pi+8)\approx0.558917 κ T 1 = ( 9 π − 8 ) / ( 9 π + 8 ) ≈ 0.558917 (with β ( 3 π / 2 , π / 2 ) = 4 / 3 π ( 3 π − 2 ) ≈ 0.478170 \beta(3\pi/2,\pi/2)=4/\sqrt{3\pi(3\pi-2)}\approx0.478170 β ( 3 π /2 , π /2 ) = 4/ 3 π ( 3 π − 2 ) ≈ 0.478170 and κ ≈ 0.353024 \kappa\approx0.353024 κ ≈ 0.353024 ), which provides a direct experimental calibration knob to modulate directional T 1 T_1 T 1 asymmetry.
Single-pulse tilted Hadamard gates (ϑ = π / 4 \vartheta=\pi/4 ϑ = π /4 ). For a single-pulse Hadamard gate driven around the tilted axis n = ( e x + e z ) / 2 \boldsymbol{n}=(\boldsymbol{e}_x+\boldsymbol{e}_z)/\sqrt{2} n = ( e x + e z ) / 2 (ϑ = π / 4 \vartheta=\pi/4 ϑ = π /4 , ∣ Q ∣ = q π |Q|=q\pi ∣ Q ∣ = q π for odd q ≥ 1 q\ge1 q ≥ 1 ), the q = 1 q=1 q = 1 moments are m H a d = ( 1 2 , − 2 π , 1 2 ) T \boldsymbol{m}_{\mathrm{Had}}=(\frac{1}{2},-\frac{\sqrt{2}}{\pi},\frac{1}{2})^{T} m Had = ( 2 1 , − π 2 , 2 1 ) T and
G H a d = ( 5 8 2 2 π − 1 8 2 2 π 3 4 2 2 π − 1 8 2 2 π 5 8 ) , det G H a d = 9 32 − 3 4 π 2 . (34) G_{\mathrm{Had}}=
\begin{pmatrix}
\frac{5}{8} & \frac{\sqrt{2}}{2\pi} & -\frac{1}{8}\\
\frac{\sqrt{2}}{2\pi} & \frac{3}{4} & \frac{\sqrt{2}}{2\pi}\\
-\frac{1}{8} & \frac{\sqrt{2}}{2\pi} & \frac{5}{8}
\end{pmatrix},
\qquad
\det G_{\mathrm{Had}}=\frac{9}{32}-\frac{3}{4\pi^2}. \tag*{(34)} G Had = 8 5 2 π 2 − 8 1 2 π 2 4 3 2 π 2 − 8 1 2 π 2 8 5 , det G Had = 32 9 − 4 π 2 3 . (34)
with m H a d T G H a d m H a d = ( π 2 − 2 ) / ( 4 π 2 ) \boldsymbol{m}_{\mathrm{Had}}^{T}G_{\mathrm{Had}}\boldsymbol{m}_{\mathrm{Had}}=(\pi^2-2)/(4\pi^2) m Had T G Had m Had = ( π 2 − 2 ) / ( 4 π 2 ) . For arbitrary odd winding q ≥ 1 q\ge1 q ≥ 1 , Eq. (31) gives
β H a d ( q ) : = β ( q π , π / 4 ) = 8 ( q 2 π 2 − 2 ) 3 ( 3 q 2 π 2 − 8 ) , (35) \beta_{\mathrm{Had}}(q):=\beta(q\pi,\pi/4)=\sqrt{\frac{8(q^2\pi^2-2)}{3(3q^2\pi^2-8)}}, \tag*{(35)} β Had ( q ) := β ( q π , π /4 ) = 3 ( 3 q 2 π 2 − 8 ) 8 ( q 2 π 2 − 2 ) , (35)
β H a d ( 1 ) ≈ 0.985474 \beta_{\mathrm{Had}}(1)\approx0.985474 β Had ( 1 ) ≈ 0.985474 , κ H a d ( 1 ) ≈ 0.007316 \kappa_{\mathrm{Had}}(1)\approx0.007316 κ Had ( 1 ) ≈ 0.007316 , approaching β H a d ( ∞ ) = 2 2 / 3 ≈ 0.942809 \beta_{\mathrm{Had}}(\infty)=2\sqrt{2}/3\approx0.942809 β Had ( ∞ ) = 2 2 /3 ≈ 0.942809 and κ H a d ( ∞ ) = 17 − 12 2 ≈ 0.029437 \kappa_{\mathrm{Had}}(\infty)=17-12\sqrt{2}\approx0.029437 κ Had ( ∞ ) = 17 − 12 2 ≈ 0.029437 as q → ∞ q\to\infty q → ∞ . For any odd winding q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 , the trajectory starting from ∣ 0 ⟩ |0\rangle ∣0 ⟩ under H ( q π , ϑ ) H(q\pi,\vartheta) H ( q π , ϑ ) is r q ( s ) = ( sin ϑ cos ϑ ( 1 − cos q π s ) , − sin ϑ sin q π s , cos 2 ϑ + sin 2 ϑ cos q π s ) T \boldsymbol{r}_q(s)=(\sin\vartheta\cos\vartheta(1-\cos q\pi s),-\sin\vartheta\sin q\pi s,\cos^2\vartheta+\sin^2\vartheta\cos q\pi s)^T r q ( s ) = ( sin ϑ cos ϑ ( 1 − cos q π s ) , − sin ϑ sin q π s , cos 2 ϑ + sin 2 ϑ cos q π s ) T . Because ∫ 0 1 cos ( q π s ) d s = 0 \int_0^1\cos(q\pi s)\,\mathrm{d}s=0 ∫ 0 1 cos ( q π s ) d s = 0 and ∫ 0 1 cos 2 ( q π s ) d s = 1 / 2 \int_0^1\cos^2(q\pi s)\,\mathrm{d}s=1/2 ∫ 0 1 cos 2 ( q π s ) d s = 1/2 for every odd q q q , both the time-averaged longitudinal polarization m z = ∫ 0 1 r z ( s ) d s = cos 2 ϑ m_z=\int_0^1r_z(s)\,\mathrm{d}s=\cos^2\vartheta m z = ∫ 0 1 r z ( s ) d s = cos 2 ϑ and the mean squared longitudinal projection ⟨ r z 2 ⟩ : = ∫ 0 1 r z ( s ) 2 d s = cos 4 ϑ + 1 2 sin 4 ϑ \langle r_z^2\rangle:=\int_0^1r_z(s)^2\,\mathrm{d}s=\cos^4\vartheta+\frac{1}{2}\sin^4\vartheta ⟨ r z 2 ⟩ := ∫ 0 1 r z ( s ) 2 d s = cos 4 ϑ + 2 1 sin 4 ϑ are independent of q q q . Using ∥ r q ( s ) ∥ 2 = 1 \|\boldsymbol{r}_q(s)\|^2=1 ∥ r q ( s ) ∥ 2 = 1 , the transverse residence trace evaluates to
G x x + G y y = ∫ 0 1 ( 2 − r x ( s ) 2 − r y ( s ) 2 ) d s = 1 + ⟨ r z 2 ⟩ = 1 + cos 4 ϑ + 1 2 sin 4 ϑ = 2 − sin 2 ϑ ( 1 − 3 8 sin 2 ϑ ) . (36) \begin{aligned}
G_{xx}+G_{yy}=\int_0^1(2-r_x(s)^2-r_y(s)^2)\,\mathrm{d}s=1+\langle r_z^2\rangle\\
&=1+\cos^4\vartheta+\frac{1}{2}\sin^4\vartheta=2-\sin^2\vartheta\left(1-\frac{3}{8}\sin^2\vartheta\right). \tag*{(36)}
\end{aligned} G xx + G y y = ∫ 0 1 ( 2 − r x ( s ) 2 − r y ( s ) 2 ) d s = 1 + ⟨ r z 2 ⟩ = 1 + cos 4 ϑ + 2 1 sin 4 ϑ = 2 − sin 2 ϑ ( 1 − 8 3 sin 2 ϑ ) . (36)
At the Hadamard tilt ϑ = π / 4 \vartheta=\pi/4 ϑ = π /4 (cos 2 ϑ = sin 2 ϑ = 1 / 2 \cos^2\vartheta=\sin^2\vartheta=1/2 cos 2 ϑ = sin 2 ϑ = 1/2 ), Eq. (36) gives m z = 1 / 2 m_z=1/2 m z = 1/2 , ⟨ r z 2 ⟩ = 1 / 4 + 1 / 8 = 3 / 8 \langle r_z^2\rangle=1/4+1/8=3/8 ⟨ r z 2 ⟩ = 1/4 + 1/8 = 3/8 , and G x x + G y y = 5 / 8 + 3 / 4 = 11 / 8 G_{xx}+G_{yy}=5/8+3/4=11/8 G xx + G y y = 5/8 + 3/4 = 11/8 for all odd q q q . Under longitudinal Γ 1 \Gamma_1 Γ 1 relaxation
( S T 1 = diag ( 1 / 2 , 1 / 2 , 0 ) (S_{T_1}=\operatorname{diag}(1/2,1/2,0) ( S T 1 = diag ( 1/2 , 1/2 , 0 ) and t T 1 = 2 e z ) \mathbf{t}_{T_1}=2\mathbf{e}_z) t T 1 = 2 e z ) , Theorem 1 yields χ ± = 1 4 ( G x x + G y y ) ∓ 1 2 m z = 11 32 ∓ 1 4 \chi_{\pm}=\frac{1}{4}(G_{xx}+G_{yy})\mp\frac{1}{2}m_z=\frac{11}{32}\mp\frac{1}{4} χ ± = 4 1 ( G xx + G y y ) ∓ 2 1 m z = 32 11 ∓ 4 1 (χ + = 3 / 32 \chi_+=3/32 χ + = 3/32 , χ − = 19 / 32 \chi_-=19/32 χ − = 19/32 ), establishing a winding-independent 19 : 3 ≈ 6.33 : 1 19:3\approx6.33:1 19 : 3 ≈ 6.33 : 1 directional error split (κ T 1 = 3 / 19 ≈ 0.157895 \kappa_{T_1}=3/19\approx0.157895 κ T 1 = 3/19 ≈ 0.157895 ).
Shaped Gaussian and DRAG control pulses
In experimental superconducting and semiconductor quantum processors, microwave arbitrary-waveform generators (AWGs) never apply ideal rectangular envelopes because sharp turn-on and turn-off edges have broad Fourier spectra that excite off-resonant transitions. Instead, hardware control stacks (such as IBM Qiskit Pulse) shape single-qubit rotations using zero-pedestal truncated Gaussian and derivative-removal-by-adiabatic-gate (DRAG) waveforms [19 , 31 ] . Because Theorem 1 holds for any piecewise-continuous trajectory r ( s ) r(s) r ( s ) , it applies directly to time-dependent AWG pulse envelopes (Table 1 ). In the language of generalized open-loop filter-function theory [24 , 25 , 26 ] , the moments ( m , G ) (m,G) ( m , G ) represent the exact zero-frequency (DC) first- and second-order toggling-frame filter moments for non-unital and unital Markovian noise.
Table 1: Geometric asymmetry parameter β ( r ) \beta(r) β ( r ) , universal selectivity bound κ ( r ) = ( 1 − β ) / ( 1 + β ) \kappa(r)=(1-\beta)/(1+\beta) κ ( r ) = ( 1 − β ) / ( 1 + β ) , maximum directional contrast κ − 1 \kappa^{-1} κ − 1 , and native longitudinal-T 1 T_1 T 1 survival ratio κ T 1 \kappa_{T_1} κ T 1 across rectangular, baseline-subtracted truncated Gaussian (σ ∈ { T / 6 , T / 4 , T / 3 } \sigma\in\{T/6,T/4,T/3\} σ ∈ { T /6 , T /4 , T /3 } ), and tilted-quadrature DRAG (σ = T / 4 \sigma=T/4 σ = T /4 , η D R A G = 0.02 \eta_{\mathrm{DRAG}}=0.02 η DRAG = 0.02 ) single-qubit pulses in the weak-damping Markovian regime (d = 2 d=2 d = 2 ). Time-symmetric X X X inversions maintain exact native-T 1 T_1 T 1 symmetry (κ T 1 = 1 \kappa_{T_1}=1 κ T 1 = 1 ) across all pulse envelopes while narrowing the Gaussian width suppresses transverse chiral noise coupling; hemisphere-biased X \sqrt{X} X and tilted Hadamard pulses exhibit pronounced native-T 1 T_1 T 1 directional asymmetry (κ T 1 ∈ [ 0.081767 , 0.157895 ] \kappa_{T_1}\in[0.081767,0.157895] κ T 1 ∈ [ 0.081767 , 0.157895 ] ).
Table 1.
Consider a baseline-subtracted truncated Gaussian envelope on s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] with root-mean-square width σ ∈ { T / 6 , T / 4 , T / 3 } \sigma\in\{T/6,T/4,T/3\} σ ∈ { T /6 , T /4 , T /3 } ,
g e n v ( s ) : = Z g − 1 [ exp ( − ( s − 1 / 2 ) 2 2 ( σ / T ) 2 ) − exp ( − 1 8 ( σ / T ) 2 ) ] , (37) g_{\mathrm{env}}(s):=Z_g^{-1}\left[\exp\left(-\frac{(s-1/2)^2}{2(\sigma/T)^2}\right)-\exp\left(-\frac{1}{8(\sigma/T)^2}\right)\right], \tag*{(37)} g env ( s ) := Z g − 1 [ exp ( − 2 ( σ / T ) 2 ( s − 1/2 ) 2 ) − exp ( − 8 ( σ / T ) 2 1 ) ] , (37)
where subtracting the pedestal exp [ − 1 / ( 8 ( σ / T ) 2 ) ] \exp[-1/(8(\sigma/T)^2)] exp [ − 1/ ( 8 ( σ / T ) 2 )] enforces continuous zero-amplitude boundaries g e n v ( 0 ) = g e n v ( 1 ) = 0 g_{\mathrm{env}}(0)=g_{\mathrm{env}}(1)=0 g env ( 0 ) = g env ( 1 ) = 0 matching experimental AWG waveforms, and Z g Z_g Z g normalizes the unit area ∫ 0 1 g e n v ( s ) d s = 1 \int_{0}^{1} g_{\mathrm{env}}(s)\,\mathrm{d}s=1 ∫ 0 1 g env ( s ) d s = 1 . The instantaneous rotation angle θ G ( s ) = Θ ∫ 0 s g e n v ( u ) d u \theta_{G}(s)=\Theta\int_{0}^{s}g_{\mathrm{env}}(u)\,\mathrm{d}u θ G ( s ) = Θ ∫ 0 s g env ( u ) d u (Θ = π \Theta=\pi Θ = π for X X X , Θ = π / 2 \Theta=\pi/2 Θ = π /2 for X \sqrt{X} X ) evaluates in closed form via the error function:
θ G ( s ) Θ = 1 Z g { π 2 σ T [ erf ( s − 1 / 2 2 σ / T ) + erf ( 1 2 2 σ / T ) ] − s e − 1 / [ 8 ( σ / T ) 2 ] } . (38) \frac{\theta_{G}(s)}{\Theta}=\frac{1}{Z_g}\left\{\sqrt{\frac{\pi}{2}}\frac{\sigma}{T}\left[\operatorname{erf}\left(\frac{s-1/2}{\sqrt{2}\sigma/T}\right)+\operatorname{erf}\left(\frac{1}{2\sqrt{2}\sigma/T}\right)\right]-s e^{-1/[8(\sigma/T)^2]}\right\}. \tag*{(38)} Θ θ G ( s ) = Z g 1 { 2 π T σ [ erf ( 2 σ / T s − 1/2 ) + erf ( 2 2 σ / T 1 ) ] − s e − 1/ [ 8 ( σ / T ) 2 ] } . (38)
When DRAG quadrature driving [19 , 31 ] is added via Ω y ( t ) = β D R A G Ω ˙ x ( t ) \Omega_y(t)=\beta_{\mathrm{DRAG}}\dot{\Omega}_x(t) Ω y ( t ) = β DRAG Ω ˙ x ( t ) to suppress ∣ 2 ⟩ \lvert2\rangle ∣ 2 ⟩ leakage, its action within the two-level dressed qubit subspace is described by the frame-tilted unitary U d r e s s ( s ) = exp ( − i ϕ D R A G ( s ) 2 σ y ) exp ( i θ G ( s ) 2 σ x ) U_{\mathrm{dress}}(s)=\exp\left(-i\frac{\phi_{\mathrm{DRAG}}(s)}{2}\sigma_y\right)\exp\left(i\frac{\theta_G(s)}{2}\sigma_x\right) U dress ( s ) = exp ( − i 2 ϕ DRAG ( s ) σ y ) exp ( i 2 θ G ( s ) σ x ) with instantaneous tilt angle ϕ D R A G ( s ) = η D R A G Θ g e n v ( s ) \phi_{\mathrm{DRAG}}(s)=\eta_{\mathrm{DRAG}}\Theta g_{\mathrm{env}}(s) ϕ DRAG ( s ) = η DRAG Θ g env ( s ) , which slightly mixes G x x G_{xx} G xx and G z z G_{zz} G z z while preserving time-reversal symmetry on an X X X gate (η D R A G = 0.02 \eta_{\mathrm{DRAG}}=0.02 η DRAG = 0.02 in Table 1).
Evaluating the trajectory moments for Eq. (38) establishes three practical pulse-design principles:
Time-symmetric X X X envelopes guarantee exact native-T 1 T_1 T 1 balance: Any time-symmetric pulse envelope satisfying g e n v ( 1 − s ) = g e n v ( s ) g_{\mathrm{env}}(1-s)=g_{\mathrm{env}}(s) g env ( 1 − s ) = g env ( s ) for a π \pi π inversion (Θ = π \Theta=\pi Θ = π ) enforces the reflection symmetry θ G ( 1 − s ) = π − θ G ( s ) \theta_G(1-s)=\pi-\theta_G(s) θ G ( 1 − s ) = π − θ G ( s ) , which identically zeroes out longitudinal drift coupling (m z = 0 m_z=0 m z = 0 , G y z = 0 G_{yz}=0 G y z = 0 ) and guarantees κ T 1 = 1 \kappa_{T_1}=1 κ T 1 = 1 regardless of pulse width or DRAG quadrature (Table 1). Conversely, intentionally skewing a pulse envelope (g e n v ( 1 − s ) ≠ g e n v ( s ) g_{\mathrm{env}}(1-s)\ne g_{\mathrm{env}}(s) g env ( 1 − s ) = g env ( s ) ) breaks m z = 0 m_z=0 m z = 0 even on a full π \pi π swap, allowing experimentalists to continuously tune the directional T 1 T_1 T 1 survival ratio χ − / χ + \chi_-/\chi_+ χ − / χ + without altering the target unitary U t a r g e t = − i σ x U_{\mathrm{target}}=-i\sigma_x U target = − i σ x or gate duration T T T .
Pole lingering passively filters worst-case transverse chiral noise on X X X gates: During the smooth turn-on and turn-off ramps of a Gaussian pulse near s = 0 s=0 s = 0 and s = 1 s=1 s = 1 , the Bloch vector dwells near the poles ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ and ∣ 1 ⟩ \lvert1\rangle ∣ 1 ⟩ (± e z \pm e_z ± e z ) where sin θ G ( s ) ≈ 0 \sin\theta_G(s)\approx0 sin θ G ( s ) ≈ 0 , and sweeps rapidly across the equator at s = 1 / 2 s=1/2 s = 1/2 . Narrowing the Gaussian width therefore reduces transverse drift coupling m y m_y m y from 2 / π ≈ 0.636620 2/\pi\approx0.636620 2/ π ≈ 0.636620 (rectangular) to 0.449869 (σ = T / 3 \sigma=T/3 σ = T /3 ), 0.419731 (σ = T / 4 \sigma=T/4 σ = T /4 ), and 0.338831 (σ = T / 6 \sigma=T/6 σ = T /6 ), while increasing G y y G_{yy} G y y from 0.5 to 0.672058, 0.698198, and 0.763008. Both effects reduce β \beta β and raise the universal selectivity bound κ G a u s s ( X ) \kappa_{\mathrm{Gauss}}(X) κ Gauss ( X ) from κ ⋆ ≈ 0.052456 \kappa_\star\approx0.052456 κ ⋆ ≈ 0.052456 (19.06 : 1) to 0.120088 (8.33 : 1 at σ = T / 3 \sigma=T/3 σ = T /3 ), 0.133768 (7.48 : 1 at σ = T / 4 \sigma=T/4 σ = T /4 ), and 0.179238 (5.58 : 1 at σ = T / 6 \sigma=T/6 σ = T /6 ).
Pulse-width dependence of native-T 1 T_1 T 1 asymmetry on X \sqrt{X} X gates: On a polar X \sqrt{X} X pulse (Θ = π / 2 \Theta=\pi/2 Θ = π /2 ), the initial Gaussian ramp lingers near the ground state ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ (+ z +z + z ) while the final ramp approaches the equator (z = 0 z=0 z = 0 ). This asymmetric pole dwell increases m z m_z m z and moderates the native T 1 T_1 T 1 directional survival ratio from 12.23 : 1 (κ T 1 ≈ 0.081767 \kappa_{T_1}\approx0.081767 κ T 1 ≈ 0.081767 , rectangular) to 8.44 : 1 (κ T 1 ≈ 0.118418 \kappa_{T_1}\approx0.118418 κ T 1 ≈ 0.118418 at σ = T / 4 \sigma=T/4 σ = T /4 ) and 7.52 : 1 (κ T 1 ≈ 0.133034 \kappa_{T_1}\approx0.133034 κ T 1 ≈ 0.133034 at σ = T / 6 \sigma=T/6 σ = T /6 ), providing a directly testable pulse-width signature.
Experimental predictions and lock-in protocol on superconducting and trapped-ion processors
Cross-platform hardware predictions: superconducting transmons, trapped ions, and engineered reservoirs
Table 2 compares the analytical first-order predictions of Theorem 1 against exact 4 × 4 4\times4 4 × 4 matrix exponentiation exp ( L ε ) \exp(\mathcal{L}_{\varepsilon}) exp ( L ε ) across superconducting transmons, optical-clock trapped ions, and engineered reservoirs. To keep experimental sources distinct, we distinguish between two hardware parameter sets: (i) in McKay et al. [19 ] , a 13.3 ns Gaussian DRAGZ X π / 2 = X \mathit{X}_{\pi/2}=\sqrt{X} X π /2 = X pulse was benchmarked on a single-qubit fixed-frequency transmon with T 1 ≈ 45 μ s T_1\approx45\,\mu\mathrm{s} T 1 ≈ 45 μ s and T 2 ≈ 50 μ s T_2\approx50\,\mu\mathrm{s} T 2 ≈ 50 μ s , achieving an interleaved randomized benchmarking error of 1.95 ( 3 ) × 10 − 4 1.95(3)\times10^{-4} 1.95 ( 3 ) × 1 0 − 4 and leakage to ∣ 2 ⟩ \lvert2\rangle ∣ 2 ⟩ of 3.1 ( 6 ) × 10 − 6 3.1(6)\times10^{-6} 3.1 ( 6 ) × 1 0 − 6 per gate (which corresponds to ϵ M c K a y = ( 13.3 n s ) / ( 45 μ s ) ≈ 2.9556 × 10 − 4 \epsilon_{\mathrm{McKay}}=(13.3\,\mathrm{ns})/(45\,\mu\mathrm{s})\approx2.9556\times10^{-4} ϵ McKay = ( 13.3 ns ) / ( 45 μ s ) ≈ 2.9556 × 1 0 − 4 and Δ p ≈ 1.8816 × 10 − 4 \Delta p\approx1.8816\times10^{-4} Δ p ≈ 1.8816 × 1 0 − 4 , row 3 of Table 2 ); and (ii) on the 127-qubit ibm_kyiv Eagle processor, Kim et al. [37 ] report a device-wide median longitudinal relaxation time T 1 = 288 μ s T_1=288\,\mu\mathrm{s} T 1 = 288 μ s and median T 2 = 127 μ s T_2=127\,\mu\mathrm{s} T 2 = 127 μ s . Pairing the median `ibm_kyiv relaxation time T 1 = 288 μ s T_1=288\,\mu\mathrm{s} T 1 = 288 μ s with a representative 25 ns single-qubit pulse duration (or a 13.3 ns fast DRAGZ duration) yields the representative modern transmon exposures
Table 2. Predicted first-order directional survival split ∣ Δ p ( 1 ) ∣ = ϵ ∣ χ + − χ − ∣ |\Delta p^{(1)}|=\epsilon|\chi_{+}-\chi_{-}| ∣Δ p ( 1 ) ∣ = ϵ ∣ χ + − χ − ∣ versus exact 4 × 4 4\times4 4 × 4 Lindblad master-equation exponentiation ∣ Δ p e x a c t ∣ = ∣ p + ( ϵ ) − p − ( ϵ ) ∣ |\Delta p^{\mathrm{exact}}|=|p_{+}(\epsilon)-p_{-}(\epsilon)| ∣Δ p exact ∣ = ∣ p + ( ϵ ) − p − ( ϵ ) ∣ , finite-ϵ \epsilon ϵ error ratio min ( p + , p − ) / max ( p + , p − ) \min(p_{+},p_{-})/\max(p_{+},p_{-}) min ( p + , p − ) / max ( p + , p − ) , and ideal 3 σ 3\sigma 3 σ shot budget N t o t ( 3 σ ) N_{\mathrm{tot}}^{(3\sigma)} N tot ( 3 σ ) across representative superconducting transmon [19 , 31 , 37 ] , trapped-ion 171 Y b + ^{171}\mathrm{Yb}^{+} 171 Yb + [38 , 39 ] , and engineered circuit-QED reservoir scenarios. Both the single-device DRAGZ benchmark of McKay et al. [19 ] (13.3 ns, T 1 = 45 μ s T_1=45\,\mu\mathrm{s} T 1 = 45 μ s ) and representative modern transmon combinations pairing 25 ns or 13.3 ns gate durations with the median T 1 = 288 μ s T_1=288\,\mu\mathrm{s} T 1 = 288 μ s of ibm_kyiv [37 ] are shown.
ϵ t r a n s m o n = T g a t e T 1 = 25 n s 288 μ s ≈ 8.680556 × 10 − 5 (39) \epsilon_{\mathrm{transmon}}=\frac{T_{\mathrm{gate}}}{T_1}=\frac{25\,\mathrm{ns}}{288\,\mu\mathrm{s}}\approx8.680556\times10^{-5} \tag*{(39)} ϵ transmon = T 1 T gate = 288 μ s 25 ns ≈ 8.680556 × 1 0 − 5 (39)
(and ϵ f a s t = ( 13.3 n s ) / ( 288 μ s ) ≈ 4.618056 × 10 − 5 \epsilon_{\mathrm{fast}}=(13.3\,\mathrm{ns})/(288\,\mu\mathrm{s})\approx4.618056\times10^{-5} ϵ fast = ( 13.3 ns ) / ( 288 μ s ) ≈ 4.618056 × 1 0 − 5 at T g a t e = 13.3 n s T_{\mathrm{gate}}=13.3\,\mathrm{ns} T gate = 13.3 ns ; rows 1–2 of Table 2 ). Under longitudinal T 1 T_1 T 1 decay (L ↓ = ∣ 0 ⟩ ⟨ 1 ∣ L_{\downarrow}=|0\rangle\langle1| L ↓ = ∣0 ⟩ ⟨ 1∣ , Γ = 1 \Gamma=1 Γ = 1 ), the bath tensors S T 1 = diag ( 1 2 , 1 2 , 0 ) S_{T_1}=\operatorname{diag}\left(\frac{1}{2},\frac{1}{2},0\right) S T 1 = diag ( 2 1 , 2 1 , 0 ) and t T 1 = ( 0 , 0 , 2 ) T t_{T_1}=(0,0,2)^{\mathsf{T}} t T 1 = ( 0 , 0 , 2 ) T yield
χ + ( X ; T 1 ) = 3 8 − 1 π ≈ 0.056690 , \chi_{+}(\sqrt{X};T_1)=\frac{3}{8}-\frac{1}{\pi}\approx0.056690, χ + ( X ; T 1 ) = 8 3 − π 1 ≈ 0.056690 ,
χ − ( X ; T 1 ) = 3 8 + 1 π ≈ 0.693310 , \chi_{-}(\sqrt{X};T_1)=\frac{3}{8}+\frac{1}{\pi}\approx0.693310, χ − ( X ; T 1 ) = 8 3 + π 1 ≈ 0.693310 ,
κ ( X ; T 1 ) = 3 π − 8 3 π + 8 ≈ 0.081767 (40) \kappa(\sqrt{X};T_1)=\frac{3\pi-8}{3\pi+8}\approx0.081767 \tag*{(40)} κ ( X ; T 1 ) = 3 π + 8 3 π − 8 ≈ 0.081767 (40)
or a directional error ratio χ − / χ + ≈ 12.23 \chi_{-}/\chi_{+}\approx12.23 χ − / χ + ≈ 12.23 , predicting a single-gate survival split Δ p = 2 π ϵ t r a n s m o n ≈ 5.5262 × 10 − 5 \Delta p=\frac{2}{\pi}\epsilon_{\mathrm{transmon}}\approx5.5262\times10^{-5} Δ p = π 2 ϵ transmon ≈ 5.5262 × 1 0 − 5 for a rectangular X \sqrt{X} X pulse and 5.1318 × 10 − 5 5.1318\times10^{-5} 5.1318 × 1 0 − 5 for a Gaussian X \sqrt{X} X pulse (σ = T / 4 \sigma=T/4 σ = T /4 ). Note that pure dephasing (L ϕ = σ z / 2 L_{\phi}=\sigma_z/\sqrt{2} L ϕ = σ z / 2 , contributing when T 2 < 2 T 1 T_2<2T_1 T 2 < 2 T 1 ) is unital (t ϕ = 0 t_{\phi}=0 t ϕ = 0 ) and leaves the directional split Δ p = ϵ ∣ χ + − χ − ∣ \Delta p=\epsilon|\chi_{+}-\chi_{-}| Δ p = ϵ ∣ χ + − χ − ∣ completely unchanged at first order, while a residual thermal excited-state population n t h = 0.01 n_{\mathrm{th}}=0.01 n th = 0.01 modifies the T 1 T_1 T 1 ratio only slightly to κ t h = [ 3 π ( 1 + 2 n t h ) − 8 ] / [ 3 π ( 1 + 2 n t h ) + 8 ] ≈ 0.091594 \kappa_{\mathrm{th}}=[3\pi(1+2n_{\mathrm{th}})-8]/[3\pi(1+2n_{\mathrm{th}})+8]\approx0.091594 κ th = [ 3 π ( 1 + 2 n th ) − 8 ] / [ 3 π ( 1 + 2 n th ) + 8 ] ≈ 0.091594 . Resolving Δ p \Delta p Δ p at 3 σ 3\sigma 3 σ statistical significance under ideal readout requires a total shot budget
N t o t ( 3 σ ) = 18 [ p + ( 1 − p + ) + p − ( 1 − p − ) ] ( Δ p ) 2 ≃ 18 ( χ + + χ − ) ( χ − − χ + ) 2 ϵ 2 (41) N_{\mathrm{tot}}^{(3\sigma)}=\frac{18[p_{+}(1-p_{+})+p_{-}(1-p_{-})]}{(\Delta p)^2}\simeq\frac{18(\chi_{+}+\chi_{-})}{(\chi_{-}-\chi_{+})^2\epsilon^2} \tag*{(41)} N tot ( 3 σ ) = ( Δ p ) 2 18 [ p + ( 1 − p + ) + p − ( 1 − p − )] ≃ ( χ − − χ + ) 2 ϵ 2 18 ( χ + + χ − ) (41)
or N t o t ( 3 σ ) = 3.84 × 10 5 N_{\mathrm{tot}}^{(3\sigma)}=3.84\times10^5 N tot ( 3 σ ) = 3.84 × 1 0 5 ideal shots on ibm_kyiv (Table 2 ).
Comparing superconducting transmons with 171 Y b + ^{171}\mathrm{Yb}^{+} 171 Yb + hyperfine clock-state trapped-ion processors [38 , 39 ] highlights how directional gate-error splitting serves as a cross-platform noise diagnostic. In 171 Y b + ^{171}\mathrm{Yb}^{+} 171 Yb + race-track processors, stimulated-Raman single-qubit gates have duration T g a t e ∼ 5 μ s T_{\mathrm{gate}}\sim5\,\mu\mathrm{s} T gate ∼ 5 μ s , average gate infidelity r a v g = 2.5 ( 3 ) × 10 − 5 r_{\mathrm{avg}}=2.5(3)\times10^{-5} r avg = 2.5 ( 3 ) × 1 0 − 5 [39 ] , and magnetic-dipole hyperfine relaxation T 1 ≳ 300 s T_1\gtrsim300\,\mathrm{s} T 1 ≳ 300 s (ϵ T 1 ≤ 1.67 × 10 − 8 \epsilon_{T_1}\le1.67\times10^{-8} ϵ T 1 ≤ 1.67 × 1 0 − 8 , Δ p T 1 ≤ 1.06 × 10 − 8 \Delta p_{T_1}\le1.06\times10^{-8} Δ p T 1 ≤ 1.06 × 1 0 − 8 ; Table 2 ). Because elastic Rayleigh and symmetric Raman photon scattering during the pulse are unital (t = 0 t=0 t = 0 ), hyperfine detuning asymmetry ω H F / Δ l a s e r ∼ 10 − 3 \omega_{\mathrm{HF}}/\Delta_{\mathrm{laser}}\sim10^{-3} ω HF / Δ laser ∼ 1 0 − 3 –10 − 2 10^{-2} 1 0 − 2 induces a fractional non-unital optical-pumping imbalance ∣ η s c ∣ ≲ 10 − 2 |\eta_{\mathrm{sc}}|\lesssim10^{-2} ∣ η sc ∣ ≲ 1 0 − 2 for ϵ s c ≲ r a v g ≈ 2.5 × 10 − 5 \epsilon_{\mathrm{sc}}\lesssim r_{\mathrm{avg}}\approx2.5\times10^{-5} ϵ sc ≲ r avg ≈ 2.5 × 1 0 − 5 , giving a native directional split Δ p s c ≲ 2 × 10 − 7 \Delta p_{\mathrm{sc}}\lesssim2\times10^{-7} Δ p sc ≲ 2 × 1 0 − 7 (> 10 10 >10^{10} > 1 0 10 shots); any larger observed split on 171 Y b + ^{171}\mathrm{Yb}^{+} 171 Yb + cleanly diagnoses residual optical-pumping beam leakage. Conversely, to test the 19.0635 : 1 universal selectivity bound κ ⋆ \kappa_{\star} κ ⋆ directly on a full X q = 1 X_{q=1} X q = 1 swap in circuit QED, coupling a transmon to a lossy readout resonator via bichromatic sideband drives synthesizes the engineered chiral jump operator L + = ( σ x + i 2 σ z ) / 6 L_{+}=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 at controlled exposure ϵ ∈ [ 10 − 3 , 10 − 2 ] \epsilon\in[10^{-3},10^{-2}] ϵ ∈ [ 1 0 − 3 , 1 0 − 2 ] , requiring only 3.33 × 10 4 3.33\times10^4 3.33 × 1 0 4 to 3.32 × 10 3 3.32\times10^3 3.32 × 1 0 3 shots (Table 2 ).
Six-circuit SPAM-invariant lock-in protocols and calibration error budget
On real hardware, native state-preparation errors ( e 0 , e 1 ) (e_0,e_1) ( e 0 , e 1 ) and readout assignment errors ( f 0 , f 1 ) (f_0,f_1) ( f 0 , f 1 ) are of order 10 − 2 10^{-2} 1 0 − 2 , exceeding the single-gate transmon split Δ p ≈ 5.5 × 10 − 5 \Delta p\approx5.5\times10^{-5} Δ p ≈ 5.5 × 1 0 − 5 by two orders of magnitude. Extracting a directional gate-error split immune to first-order SPAM asymmetry requires contrasting two control trajectories that implement the same target population transfer (up to global phase) and incur identical first-order unital damping Σ = p + + p − = ε Tr ( S G ) \Sigma=p_++p_-=\varepsilon\operatorname{Tr}(SG) Σ = p + + p − = ε Tr ( S G ) , but possess different trajectory centroids m \boldsymbol{m} m along the active bath drift vector t \boldsymbol{t} t .
Crucially, the action of reversing the drive Hamiltonian H ↦ − H H\mapsto-H H ↦ − H (Q ↦ − Q Q\mapsto-Q Q ↦ − Q for H ( Q ) = − Q 2 σ x H(Q)=-\frac{Q}{2}\sigma_x H ( Q ) = − 2 Q σ x ) must be distinguished from antipodal state inversion ∣ 0 ⟩ ↔ ∣ 1 ⟩ \lvert0\rangle\leftrightarrow\lvert1\rangle ∣ 0 ⟩ ↔ ∣ 1 ⟩ : whereas exchanging the initial states ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ and ∣ 1 ⟩ \lvert1\rangle ∣ 1 ⟩ negates the full Bloch vector r − ( s ) = − r + ( s ) \boldsymbol{r}_-(s)=-\boldsymbol{r}_+(s) r − ( s ) = − r + ( s ) , reversing the drive sign H ( Q ) ↦ H ( − Q ) = − H ( Q ) H(Q)\mapsto H(-Q)=-H(Q) H ( Q ) ↦ H ( − Q ) = − H ( Q ) reflects the trajectory starting from ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ across the x x x –z z z plane:
r + ( s ; Q ) = ( 0 , sin ( Q s ) , cos ( Q s ) ) T ⟹ r + ( s ; − Q ) = ( 0 , − sin ( Q s ) , + cos ( Q s ) ) T . (42) \boldsymbol{r}_+(s;Q)=\left(0,\sin(Qs),\cos(Qs)\right)^{T}
\qquad\Longrightarrow\qquad
\boldsymbol{r}_+(s;-Q)=\left(0,-\sin(Qs),+\cos(Qs)\right)^{T}. \tag*{(42)} r + ( s ; Q ) = ( 0 , sin ( Q s ) , cos ( Q s ) ) T ⟹ r + ( s ; − Q ) = ( 0 , − sin ( Q s ) , + cos ( Q s ) ) T . (42)
Consequently, under H ↦ − H H\mapsto-H H ↦ − H , the transverse centroid component m y ( Q ) = ( 1 − cos Q ) / Q m_y(Q)=(1-\cos Q)/Q m y ( Q ) = ( 1 − cos Q ) / Q is odd in Q Q Q (m y ( − Q ) = − m y ( Q ) m_y(-Q)=-m_y(Q) m y ( − Q ) = − m y ( Q ) ), whereas the longitudinal centroid component m z ( Q ) = ( sin Q ) / Q m_z(Q)=(\sin Q)/Q m z ( Q ) = ( sin Q ) / Q is even in Q Q Q (m z ( − Q ) = + m z ( Q ) m_z(-Q)=+m_z(Q) m z ( − Q ) = + m z ( Q ) ). This parity distinction dictates two distinct six-circuit SPAM-invariant lock-in protocols depending on whether the bath drift is transverse or longitudinal (Figure 4(a)):
Protocol 1(a) (Signed-winding X X X swaps q = ± 1 q=\pm1 q = ± 1 for transverse bath drift t y ≠ 0 t_y\ne0 t y = 0 ): For a full π \pi π swap H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (Q = − q π Q=-q\pi Q = − q π with q ∈ { + 1 , − 1 } q\in\{+1,-1\} q ∈ { + 1 , − 1 } ), both trajectories map ∣ 0 ⟩ ↔ ∣ 1 ⟩ \lvert0\rangle\leftrightarrow\lvert1\rangle ∣ 0 ⟩ ↔ ∣ 1 ⟩ (U ( ± π ) = ± i σ x U(\pm\pi)=\pm i\sigma_x U ( ± π ) = ± i σ x ) with m z ( ± π ) = 0 m_z(\pm\pi)=0 m z ( ± π ) = 0 and opposite transverse centroids m y ( q ) = 2 q π m_y(q)=\frac{2}{q\pi} m y ( q ) = q π 2 . Reversing q = + 1 ↔ − 1 q=+1\leftrightarrow-1 q = + 1 ↔ − 1 therefore flips the transverse directional split Δ q = 1 2 ε m y ( q ) t y = Δ 1 / q \Delta_q=\frac{1}{2}\varepsilon m_y(q)t_y=\Delta_1/q Δ q = 2 1 ε m y ( q ) t y = Δ 1 / q while preserving Σ + 1 = Σ − 1 \Sigma_{+1}=\Sigma_{-1} Σ + 1 = Σ − 1 .
Protocol 1(b) (Two-branch lock-in k ∈ { 0 , 1 } k\in\{0,1\} k ∈ { 0 , 1 } for polar X \sqrt{X} X under longitudinal T 1 T_1 T 1 drift t = 2 e z t=2e_z t = 2 e z ): For a polar X \sqrt{X} X gate under longitudinal T 1 T_1 T 1 relaxation (t = 2 e z t=2e_z t = 2 e z ), simple drive reversal Q = Q= Q =
+ π / 2 ↦ − π / 2 +\pi/2 \mapsto-\pi/2 + π /2 ↦ − π /2 fails for two reasons: (i) because m z ( ± π / 2 ) = + 2 / π m_z(\pm\pi/2)=+2/\pi m z ( ± π /2 ) = + 2/ π is even in Q Q Q , the T 1 T_1 T 1 directional split 1 2 ε m z t z \frac{1}{2}\varepsilon m_z t_z 2 1 ε m z t z is identical for Q = ± π / 2 Q=\pm\pi/2 Q = ± π /2 and cancels in their difference, and (ii) exp ( ± i π 4 σ x ) \exp(\pm i\frac{\pi}{4}\sigma_x) exp ( ± i 4 π σ x ) map ∣ 0 ⟩ |0\rangle ∣0 ⟩ to opposite equatorial states ∣ ∓ y ⟩ |\mp y\rangle ∣ ∓ y ⟩ rather than the same target state. Instead, to modulate m z m_z m z while implementing the exact same target unitary up to sign, we contrast the principal branch k = 0 k=0 k = 0 (Q 0 = + π / 2 Q_0=+\pi/2 Q 0 = + π /2 , U ( Q 0 ) = exp ( i π 4 σ x ) U(Q_0)=\exp(i\frac{\pi}{4}\sigma_x) U ( Q 0 ) = exp ( i 4 π σ x ) ), spending all its time in z ≥ 0 z\geq0 z ≥ 0 with m z ( 0 ) = + 2 / π m_z^{(0)}=+2/\pi m z ( 0 ) = + 2/ π against the reversed over-rotated branch k = 1 k=1 k = 1 (Q 1 = − 3 π / 2 Q_1=-3\pi/2 Q 1 = − 3 π /2 , U ( Q 1 ) = exp ( − i 3 π 4 σ x ) = − U ( Q 0 ) U(Q_1)=\exp(-i\frac{3\pi}{4}\sigma_x)=-U(Q_0) U ( Q 1 ) = exp ( − i 4 3 π σ x ) = − U ( Q 0 ) ), which traverses the southern hemisphere z ≤ 0 z\leq0 z ≤ 0 for two-thirds of the pulse and has m z ( 1 ) = sin ( − 3 π / 2 ) − 3 π / 2 = − 2 3 π = − 1 3 m z ( 0 ) m_z^{(1)}=\frac{\sin(-3\pi/2)}{-3\pi/2}=-\frac{2}{3\pi}=-\frac{1}{3}m_z^{(0)} m z ( 1 ) = − 3 π /2 s i n ( − 3 π /2 ) = − 3 π 2 = − 3 1 m z ( 0 ) at matched pulse duration T T T . Following the test gate U ( Q k ) U(Q_k) U ( Q k ) (k ∈ { 0 , 1 } k\in\{0,1\} k ∈ { 0 , 1 } ) with a fixed reference X \sqrt{X} X analysis pulse U a n = exp ( i π 4 σ x ) U_{\mathrm{an}}=\exp(i\frac{\pi}{4}\sigma_x) U an = exp ( i 4 π σ x ) maps the ideal composite gate to a full population inversion U a n U ( Q k ) = ± i σ x U_{\mathrm{an}}U(Q_k)=\pm i\sigma_x U an U ( Q k ) = ± i σ x prior to Z Z Z -basis readout. Because both branches share the exact same diagonal covariance deficit diag ( G ( 0 ) ) = diag ( G ( 1 ) ) = ( 1 , 1 / 2 , 1 / 2 ) T \operatorname{diag}(G^{(0)})=\operatorname{diag}(G^{(1)})=(1,1/2,1/2)^T diag ( G ( 0 ) ) = diag ( G ( 1 ) ) = ( 1 , 1/2 , 1/2 ) T , they experience identical first-order unital damping Σ ( 0 ) = Σ ( 1 ) = 3 4 ε \Sigma^{(0)}=\Sigma^{(1)}=\frac{3}{4}\varepsilon Σ ( 0 ) = Σ ( 1 ) = 4 3 ε under longitudinal T 1 T_1 T 1 relaxation and pure dephasing while exhibiting opposite directional splits Δ X ( 1 ) = − 1 3 Δ X ( 0 ) \Delta_{\sqrt{X}}^{(1)}=-\frac{1}{3}\Delta_{\sqrt{X}}^{(0)} Δ X ( 1 ) = − 3 1 Δ X ( 0 ) (giving differential lever arm Δ m z = m z ( 0 ) − m z ( 1 ) = 8 3 π \Delta m_z=m_z^{(0)}-m_z^{(1)}=\frac{8}{3\pi} Δ m z = m z ( 0 ) − m z ( 1 ) = 3 π 8 ).
Protocol 1 (Six-circuit two-branch differential lock-in estimators). Let v p r e p : = 1 − e 0 − e 1 v_{\mathrm{prep}}:=1-e_0-e_1 v prep := 1 − e 0 − e 1 (η p r e p : = e 0 − e 1 \eta_{\mathrm{prep}}:=e_0-e_1 η prep := e 0 − e 1 ) and v r o : = 1 − f 0 − f 1 v_{\mathrm{ro}}:=1-f_0-f_1 v ro := 1 − f 0 − f 1 (V S P A M : = v r o v p r e p V_{\mathrm{SPAM}}:=v_{\mathrm{ro}}v_{\mathrm{prep}} V SPAM := v ro v prep ) denote the preparation and readout visibilities, and let R 0 ( α ) ( m ) R_0^{(\alpha)}(m) R 0 ( α ) ( m ) and R 1 ( α ) ( m ) R_1^{(\alpha)}(m) R 1 ( α ) ( m ) denote the probabilities of recording readout outcome “1” when preparing ∣ 0 ⟩ |0\rangle ∣0 ⟩ or ∣ 1 ⟩ |1\rangle ∣1 ⟩ and executing m = 0 m=0 m = 0 (two SPAM calibration circuits) or m = 1 m=1 m = 1 (four gate circuits across the two trajectory branches α \alpha α ). Forming Y m ( α ) : = R 0 ( α ) ( m ) + R 1 ( α ) ( m ) Y_m^{(\alpha)}:=R_0^{(\alpha)}(m)+R_1^{(\alpha)}(m) Y m ( α ) := R 0 ( α ) ( m ) + R 1 ( α ) ( m ) and Z m ( α ) : = R 0 ( α ) ( m ) − R 1 ( α ) ( m ) Z_m^{(\alpha)}:=R_0^{(\alpha)}(m)-R_1^{(\alpha)}(m) Z m ( α ) := R 0 ( α ) ( m ) − R 1 ( α ) ( m ) gives Z 0 = − V S P A M Z_0=-V_{\mathrm{SPAM}} Z 0 = − V SPAM and (full derivation in Appendix § 21 ):
Signed-winding X X X -swap lock-in (α = q ∈ { + 1 , − 1 } \alpha=q\in\{+1,-1\} α = q ∈ { + 1 , − 1 } , transverse bath drift t y ≠ 0 t_y\neq0 t y = 0 ): With Σ q : = p + ( q ) + p − ( q ) = Σ 1 \Sigma_q:=p_+(q)+p_-(q)=\Sigma_1 Σ q := p + ( q ) + p − ( q ) = Σ 1 and Δ q : = p − ( q ) − p + ( q ) = Δ 1 / q + O ( ε 2 ) \Delta_q:=p_-(q)-p_+(q)=\Delta_1/q+\mathcal{O}(\varepsilon^2) Δ q := p − ( q ) − p + ( q ) = Δ 1 / q + O ( ε 2 ) ,
Y 1 ( q ) = 2 f 0 + v r o ( 1 − η p r e p − Δ q + η p r e p Σ q ) , (43) Y_1^{(q)}=2f_0+v_{\mathrm{ro}}(1-\eta_{\mathrm{prep}}-\Delta_q+\eta_{\mathrm{prep}}\Sigma_q), \tag*{(43)} Y 1 ( q ) = 2 f 0 + v ro ( 1 − η prep − Δ q + η prep Σ q ) , (43)
Z 1 ( q ) = V S P A M ( 1 − Σ q ) , Z 0 = − V S P A M , (44) Z_1^{(q)}=V_{\mathrm{SPAM}}(1-\Sigma_q),\qquad Z_0=-V_{\mathrm{SPAM}}, \tag*{(44)} Z 1 ( q ) = V SPAM ( 1 − Σ q ) , Z 0 = − V SPAM , (44)
g ^ A ′ : = Y 1 ( − 1 ) − Y 1 ( + 1 ) − 2 Z 0 = Δ 1 v p r e p + O ( ε 2 ) . (45) \widehat{g}_{A'}:=\frac{Y_1^{(-1)}-Y_1^{(+1)}}{-2Z_0}=\frac{\Delta_1}{v_{\mathrm{prep}}}+\mathcal{O}(\varepsilon^2). \tag*{(45)} g A ′ := − 2 Z 0 Y 1 ( − 1 ) − Y 1 ( + 1 ) = v prep Δ 1 + O ( ε 2 ) . (45)
Two-branch X \sqrt{X} X lock-in (α = k ∈ { 0 , 1 } \alpha=k\in\{0,1\} α = k ∈ { 0 , 1 } , Q 0 = + π / 2 Q_0=+\pi/2 Q 0 = + π /2 vs. Q 1 = − 3 π / 2 Q_1=-3\pi/2 Q 1 = − 3 π /2 at matched duration T T T followed by a fixed reference X \sqrt{X} X analysis pulse, longitudinal T 1 T_1 T 1 drift): With common analysis-pulse error ( Σ a n , Δ a n ) (\Sigma_{\mathrm{an}},\Delta_{\mathrm{an}}) ( Σ an , Δ an ) cancelling in the branch difference and Δ X ( 1 ) = − 1 3 Δ X ( 0 ) + O ( ε 2 ) \Delta_{\sqrt{X}}^{(1)}=-\frac{1}{3}\Delta_{\sqrt{X}}^{(0)}+\mathcal{O}(\varepsilon^2) Δ X ( 1 ) = − 3 1 Δ X ( 0 ) + O ( ε 2 ) ,
g ^ X : = 3 ( Y 1 ( k = 0 ) − Y 1 ( k = 1 ) ) − 4 Z 0 = Δ X ( 0 ) v p r e p + O ( ε 2 ) . (46) \widehat{g}_{\sqrt{X}}:=\frac{3(Y_1^{(k=0)}-Y_1^{(k=1)})}{-4Z_0}=\frac{\Delta_{\sqrt{X}}^{(0)}}{v_{\mathrm{prep}}}+\mathcal{O}(\varepsilon^2). \tag*{(46)} g X := − 4 Z 0 3 ( Y 1 ( k = 0 ) − Y 1 ( k = 1 ) ) = v prep Δ X ( 0 ) + O ( ε 2 ) . (46)
For transverse-drift X X X swaps, subtracting opposite-helicity drives (q = + 1 q=+1 q = + 1 versus q = − 1 q=-1 q = − 1 , lever arm Δ m y = 4 / π \Delta m_y=4/\pi Δ m y = 4/ π ) triples the directional signal compared to unsigned multi-winding subtraction (q = 1 q=1 q = 1 versus q = 3 q=3 q = 3 , Δ m y = 4 / ( 3 π ) \Delta m_y=4/(3\pi) Δ m y = 4/ ( 3 π ) ), achieving an asymptotic 3 2 = 9 × 3^2=9\times 3 2 = 9 × variance reduction (8.97 × 8.97\times 8.97 × plug-in variance reduction, 8.81 × 8.81\times 8.81 × Cramér–Rao reduction for Δ 1 / v p r e p \Delta_1/v_{\mathrm{prep}} Δ 1 / v prep , and 8.54 × 8.54\times 8.54 × Cramér–Rao reduction for Δ 1 \Delta_1 Δ 1 at the benchmark parameter vector θ 0 \boldsymbol{\theta}_0 θ 0 in Figure 4 (a)). For polar X \sqrt{X} X gates under longitudinal T 1 T_1 T 1 decay, the two-branch estimator g ^ X \widehat{g}_{\sqrt{X}} g X in Eq. (46) achieves lever arm Δ m z = 8 3 π \Delta m_z=\frac{8}{3\pi} Δ m z = 3 π 8 (two-thirds of a full signed swap), yielding N t o t Var ( g ^ X ) = 0.9915 N_{\mathrm{tot}}\operatorname{Var}(\widehat{g}_{\sqrt{X}})=0.9915 N tot Var ( g X ) = 0.9915 (4.00 × 4.00\times 4.00 × lower variance than unsigned q ∈ { 1 , 3 } q\in\{1,3\} q ∈ { 1 , 3 } subtraction) and agreeing with exact 4 × 4 4\times4 4 × 4 PTM exponentiation on ibm_kyiv parameters to ∼ 10 − 4 \sim10^{-4} ∼ 1 0 − 4 relative accuracy (Appendix § 21 ). When the k = 1 k=1 k = 1 branch (Q 1 = − 3 π / 2 Q_1=-3\pi/2 Q 1 = − 3 π /2 ) is implemented at fixed Rabi amplitude by tripling the pulse duration to 3 T 3T 3 T rather than tripling the drive amplitude at fixed duration T T T , its unital damping scales as Σ ( 1 ) = 3 Σ ( 0 ) \Sigma^{(1)}=3\Sigma^{(0)} Σ ( 1 ) = 3 Σ ( 0 ) while its directional split becomes Δ X ( 1 ) ( 3 T ) = − Δ X ( 0 ) ( T ) \Delta_{\sqrt{X}}^{(1)}(3T)=-\Delta_{\sqrt{X}}^{(0)}(T) Δ X ( 1 ) ( 3 T ) = − Δ X ( 0 ) ( T ) , and the η p r e p ( Σ ( 0 ) − Σ ( 1 ) ) \eta_{\mathrm{prep}}(\Sigma^{(0)}-\Sigma^{(1)}) η prep ( Σ ( 0 ) − Σ ( 1 ) ) cross-term is removed either by the six-circuit calibration ratio ( Z 1 ( 1 ) − Z 1 ( 0 ) ) / Z 0 (Z_1^{(1)}-Z_1^{(0)})/Z_0 ( Z 1 ( 1 ) − Z 1 ( 0 ) ) / Z 0 or by standard gate-count slope fitting (Appendix § 21 ).
Figure 4. Two-branch differential lock-in protocols and numerical CPTP verification of the swap sector C 1 \mathcal{C}_1 C 1 . (a) Root-mean-square estimation error σ ( Δ ^ / v p r e p ) \sigma(\widehat{\Delta}/\sqrt{v_{\mathrm{prep}}}) σ ( Δ / v prep ) versus total measurement shots N t o t N_{\mathrm{tot}} N tot under native state-preparation and measurement (SPAM) errors θ 0 = ( 0.02 , 0.06 , 0.01 , 0.02 , 0.02 , 0.018 ) \boldsymbol{\theta}_0=(0.02,0.06,0.01,0.02,0.02,0.018) θ 0 = ( 0.02 , 0.06 , 0.01 , 0.02 , 0.02 , 0.018 ) for the signed-winding X X X -swap lock-in estimator g ^ A ′ \widehat{g}_{A'} g A ′ (q = ± 1 q=\pm1 q = ± 1 , Protocol 1(a), solid blue; 8.97 × 8.97\times 8.97 × plug-in variance reduction and 8.54 × 8.54\times 8.54 × Cramér–Rao reduction for Δ 1 \Delta_1 Δ 1 over unsigned q ∈ { 1 , 3 } q\in\{1,3\} q ∈ { 1 , 3 } subtraction g ^ A \widehat{g}_A g A , dashed red) and the two-branch X \sqrt{X} X longitudinal-T 1 T_1 T 1 lock-in estimator g ^ X \widehat{g}_{\sqrt{X}} g X (k ∈ { 0 , 1 } k\in\{0,1\} k ∈ { 0 , 1 } , Q 0 = + π / 2 Q_0=+\pi/2 Q 0 = + π /2 versus Q 1 = − 3 π / 2 Q_1=-3\pi/2 Q 1 = − 3 π /2 , Protocol 1(b), dash-dotted green). (b) Random sample of 300 multi-jump CPTP Markovian baths (Γ = 1 \Gamma=1 Γ = 1 , blue circles, drawn from 1,203,200 random baths audited in Appendix J) in the ( χ + , χ − ) (\chi_+,\chi_-) ( χ + , χ − ) plane for an X q = 1 X_{q=1} X q = 1 swap, confirming that every physical bath lies strictly inside the sector κ ⋆ χ + ≤ χ − ≤ κ ⋆ − 1 χ + \kappa_\star\chi_+\le\chi_-\le\kappa_\star^{-1}\chi_+ κ ⋆ χ + ≤ χ − ≤ κ ⋆ − 1 χ + bounded by the chiral extremizer L + L_+ L + (gold star).
An experimental error-budget audit confirms that five potential systematic biases on a superconducting transmon remain well below Δ p ≈ 5.5 × 10 − 5 \Delta p\approx5.5\times10^{-5} Δ p ≈ 5.5 × 1 0 − 5 :
Coherent over-rotation (δ Q ± \delta Q_{\pm} δ Q ± ): By the skew-symmetry identity (73) in Appendix § 17 , coherent rotation-angle calibration errors δ Q ± \delta Q_{\pm} δ Q ± vanish at first order in survival probabilities after a swap and enter at second order, δ p c o h ( δ Q ± ) ≈ 1 4 ( δ Q ± ) 2 \delta p_{\mathrm{coh}}^{\mathrm{(}}{\delta Q_{\pm}}\mathrm{)} \approx\frac{1}{4}(\delta Q_{\pm})^2 δ p coh ( δ Q ± ) ≈ 4 1 ( δ Q ± ) 2 . Standard Rabi amplitude calibration (∣ δ Q ± ∣ ≲ 10 − 3 |\delta Q_{\pm}| \lesssim10^{-3} ∣ δ Q ± ∣ ≲ 1 0 − 3 ) limits the differential coherent bias to ∣ δ Q + 2 − δ Q − 2 ∣ / 4 ≲ 2.5 × 10 − 7 ≪ Δ p |\delta Q_{+}^{2}-\delta Q_{-}^{2}|/4 \lesssim2.5 \times10^{-7} \ll\Delta p ∣ δ Q + 2 − δ Q − 2 ∣/4 ≲ 2.5 × 1 0 − 7 ≪ Δ p .
Transmon ∣ 2 ⟩ |2\rangle ∣2 ⟩ state leakage: Measured DRAGZ leakage per X \sqrt{\mathrm{X}} X gate is 3.1 ( 6 ) × 10 − 6 3.1(6) \times10^{-6} 3.1 ( 6 ) × 1 0 − 6 [18 , 19 ] , which is 18 × 18\times 18 × smaller than Δ p ≈ 5.5 × 10 − 5 \Delta p \approx5.5 \times10^{-5} Δ p ≈ 5.5 × 1 0 − 5 on ibm_kyiv and 61 × 61\times 61 × smaller than Δ p ≈ 1.88 × 10 − 4 \Delta p \approx1.88 \times10^{-4} Δ p ≈ 1.88 × 1 0 − 4 on the T 1 = 45 μ s T_{1}=45\,\mu\mathrm{s} T 1 = 45 μ s device of Ref. [19 ] .
Spectator Z Z ZZ Z Z crosstalk: Static Z Z ZZ Z Z coupling shifts the effective drive detuning and is eliminated in isolated single-qubit characterization by initializing spectator qubits in their ground state ∣ 0 ⟩ |0\rangle ∣0 ⟩ .
Behavior of multi-pulse swap chains (k = 2 m + 1 k=2m+1 k = 2 m + 1 ): Applying an odd train H q k H_q^k H q k of k = 2 m + 1 k=2m+1 k = 2 m + 1 identical swaps alternates each trajectory between ∣ 0 ⟩ → ∣ 1 ⟩ |0\rangle\to|1\rangle ∣0 ⟩ → ∣1 ⟩ on odd steps and ∣ 1 ⟩ → ∣ 0 ⟩ |1\rangle\to|0\rangle ∣1 ⟩ → ∣0 ⟩ on even steps. Consequently, the cumulative survival error probabilities obey
p ± ( k ) = p ˉ ( k ) ( 1 ± μ q k ) + O ( k 2 ε 2 ) , p ˉ ( k ) = k 2 ( χ + + χ − ) ε , (47) p_{\pm}^{(k)}=\bar{p}^{(k)}\left(1\pm\frac{\mu_q}{k}\right)+O(k^2\varepsilon^2),\qquad\bar{p}^{(k)}=\frac{k}{2}(\chi_{+}+\chi_{-})\varepsilon, \tag*{(47)} p ± ( k ) = p ˉ ( k ) ( 1 ± k μ q ) + O ( k 2 ε 2 ) , p ˉ ( k ) = 2 k ( χ + + χ − ) ε , (47)
where μ q = ( χ + − χ − ) / ( χ + + χ − ) \mu_q=(\chi_{+}-\chi_{-})/(\chi_{+}+\chi_{-}) μ q = ( χ + − χ − ) / ( χ + + χ − ) . Because p ˉ ( k ) ∝ k \bar{p}^{(k)}\propto k p ˉ ( k ) ∝ k , the 2 m 2m 2 m paired forward-and-backward steps cancel in the directional difference, leaving the absolute split p − ( k ) − p + ( k ) = Δ 1 + O ( k 2 ε 2 ) p_{-}^{(k)}-p_{+}^{(k)}=\Delta_{1}+O(k^2\varepsilon^2) p − ( k ) − p + ( k ) = Δ 1 + O ( k 2 ε 2 ) constant in k k k while the relative contrast μ q / k \mu_q/k μ q / k decreases as 1 / k 1/k 1/ k . Constructive k k k -fold accumulation of the directional split ( p − ( k ) − p + ( k ) = k Δ 1 + O ( k 2 ε 2 ) ) (p_{-}^{(k)}-p_{+}^{(k)}=k\Delta_{1}+O(k^2\varepsilon^2)) ( p − ( k ) − p + ( k ) = k Δ 1 + O ( k 2 ε 2 )) is achieved for transverse bath drift by alternating the drive sign (+ H 1 +H_{1} + H 1 on odd steps, − H 1 -H_{1} − H 1 on even steps so that every step traverses the same transverse hemisphere), or for longitudinal T 1 T_{1} T 1 drift by active reset between pulses.
Rejection of 1 / f 1/f 1/ f T 1 T_{1} T 1 and readout telegraph drift: Because transmon T 1 T_{1} T 1 and readout assignment fidelities fluctuate over minutes to hours, interleaving the two lock-in branch circuits shot-by-shot within each hardware repetition buffer (at 1 1 1 –10 k H z 10\,\mathrm{kHz} 10 kHz repetition rates) acts as a lock-in chopper that rejects slow 1 / f 1/f 1/ f drift.
Extensions to finite damping, pulse dwell, colored baths, and multilevel systems
A natural physical question is how the leading-order autonomous qubit bounds generalize when any of the four baseline regime assumptions—weak damping (ε → 0 + \varepsilon\to0^{+} ε → 0 + ), continuous quantum-speed-limit driving (T = T Q S L T=T_{\mathrm{QSL}} T = T QSL ), white Markovian noise, or strictly two-level evolution (d = 2 d=2 d = 2 )—is extended to broader experimental conditions. Each extension yields an exact analytical law.
Finite damping strength and exact all-orders damped-Bloch solution
When the dimensionless damping exposure ε > 0 \varepsilon>0 ε > 0 is finite, let a ( ε ) : = max ( p + ( ε ) , p − ( ε ) ) a(\varepsilon):=\max(p_{+}(\varepsilon),p_{-}(\varepsilon)) a ( ε ) := max ( p + ( ε ) , p − ( ε )) and b ( ε ) : = min ( p + ( ε ) , p − ( ε ) ) b(\varepsilon):=\min(p_{+}(\varepsilon),p_{-}(\varepsilon)) b ( ε ) := min ( p + ( ε ) , p − ( ε )) denote the larger and smaller exact survival error probabilities of a constant-amplitude swap H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x . Because λ min ( G q ) = 1 / 2 \lambda_{\min}(G_q)=1/2 λ m i n ( G q ) = 1/2 guarantees χ max ≥ 1 4 Γ \chi_{\max}\geq\frac{1}{4}\Gamma χ m a x ≥ 4 1 Γ , combining Theorem 2 with the explicit Dyson remainder bound (7) yields a rigorous non-perturbative lower bound valid for every Markovian bath with η r e m ( ε ) : = 8 ε Γ e 2 ε Γ < 1 \eta_{\mathrm{rem}}(\varepsilon):=8\varepsilon\Gamma e^{2\varepsilon\Gamma}<1 η rem ( ε ) := 8 ε Γ e 2 ε Γ < 1 (proof in Appendix § 17 ):
b ( ε ) a ( ε ) ≥ κ q − η r e m ( ε ) 1 + η r e m ( ε ) ≥ κ q ( 1 − C q Γ e 2 ε Γ ) , (48) \frac{b(\varepsilon)}{a(\varepsilon)}\geq\frac{\kappa_q-\eta_{\mathrm{rem}}(\varepsilon)}{1+\eta_{\mathrm{rem}}(\varepsilon)}\geq\kappa_q\left(1-C_q\Gamma e^{2\varepsilon\Gamma}\right), \tag*{(48)} a ( ε ) b ( ε ) ≥ 1 + η rem ( ε ) κ q − η rem ( ε ) ≥ κ q ( 1 − C q Γ e 2 ε Γ ) , (48)
C q : = 8 ( 1 + κ q ) κ q = 16 π ∣ q ∣ π ∣ q ∣ − 2 2 . C_q:=\frac{8(1+\kappa_q)}{\kappa_q}=\frac{16\pi|q|}{\pi|q|-2\sqrt{2}}. C q := κ q 8 ( 1 + κ q ) = π ∣ q ∣ − 2 2 16 π ∣ q ∣ .
where C 1 = 16 π / ( π − 2 2 ) ≈ 160.51 C_1=16\pi/(\pi-2\sqrt{2})\approx160.51 C 1 = 16 π / ( π − 2 2 ) ≈ 160.51 at ∣ q ∣ = 1 |q|=1 ∣ q ∣ = 1 .
Moreover, for the extremal chiral jump operator L + = ( σ x + i 2 σ z ) / 6 L_{+}=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 (Γ = 1 \Gamma=1 Γ = 1 ) under H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (Q = q π Q=q\pi Q = q π ), the full 4 × 4 4\times4 4 × 4 Liouvillian L ε = − i a d H q + ε D [ L + ] \mathcal{L}_{\varepsilon}=-i\,\mathrm{ad}_{H_q}+\varepsilon\mathcal{D}[L_{+}] L ε = − i ad H q + ε D [ L + ] integrates in exact closed form at arbitrary damping strength ε ≥ 0 \varepsilon\ge0 ε ≥ 0 (Appendix E). With damped Rabi frequency Ω ε , q = Q 2 − ε 2 / 9 \Omega_{\varepsilon,q}=\sqrt{Q^2-\varepsilon^2/9} Ω ε , q = Q 2 − ε 2 /9 and phase lag δ q = Q − Ω ε , q \delta_q=Q-\Omega_{\varepsilon,q} δ q = Q − Ω ε , q , the exact entanglement fidelity F e ( ε , Q ) F_e(\varepsilon,Q) F e ( ε , Q ) and survival error probabilities p ± ( ε , q ) = 1 2 [ C ( ε , Q ) ± J ( ε , Q ) ] p_\pm(\varepsilon,q)=\frac{1}{2}[C(\varepsilon,Q)\pm J(\varepsilon,Q)] p ± ( ε , q ) = 2 1 [ C ( ε , Q ) ± J ( ε , Q )] are given at all orders in ε \varepsilon ε by
F e ( ε , Q ) = 1 4 [ 1 + e − 2 ε / 3 ( 1 + 2 cos δ q ) ] , (49) F_e(\varepsilon,Q)=\frac{1}{4}\left[1+e^{-2\varepsilon/3}(1+2\cos\delta_q)\right], \tag*{(49)} F e ( ε , Q ) = 4 1 [ 1 + e − 2 ε /3 ( 1 + 2 cos δ q ) ] , (49)
C ( ε , Q ) = 1 − e − 2 ε / 3 ( cos δ q − ε 3 Ω ε , q sin δ q ) , (50) C(\varepsilon,Q)=1-e^{-2\varepsilon/3}\left(\cos\delta_q-\frac{\varepsilon}{3\Omega_{\varepsilon,q}}\sin\delta_q\right), \tag*{(50)} C ( ε , Q ) = 1 − e − 2 ε /3 ( cos δ q − 3 Ω ε , q ε sin δ q ) , (50)
J ( ε , Q ) = 2 2 ε Q [ 1 + e − 2 ε / 3 ( cos δ q − 2 ε 3 Ω ε , q sin δ q ) ] 3 ( Q 2 + ε 2 / 3 ) . (51) J(\varepsilon,Q)=\frac{2\sqrt{2}\varepsilon Q\left[1+e^{-2\varepsilon/3}\left(\cos\delta_q-\frac{2\varepsilon}{3\Omega_{\varepsilon,q}}\sin\delta_q\right)\right]}{3(Q^2+\varepsilon^2/3)}. \tag*{(51)} J ( ε , Q ) = 3 ( Q 2 + ε 2 /3 ) 2 2 εQ [ 1 + e − 2 ε /3 ( cos δ q − 3 Ω ε , q 2 ε sin δ q ) ] . (51)
Expanding Eq. (49) in powers of η : = ε Γ = ε \eta:=\varepsilon\Gamma=\varepsilon η := ε Γ = ε shows an all-orders structural property (Figure 5 (b)): the winding-number invariance of entanglement fidelity F e ( η , q π ) F_e(\eta,q\pi) F e ( η , q π ) holds not merely at first order, but identically through third order O ( η 3 ) O(\eta^3) O ( η 3 ) , splitting only at fourth order O ( η 4 ) O(\eta^4) O ( η 4 ) :
F e ( η , q π ) − F e ( η , q ′ π ) = ( 1 q ′ 2 − 1 q 2 ) [ η 4 1296 π 2 − η 5 1944 π 2 ] + O ( η 6 ) . (52) \begin{aligned}
F_e(\eta,q\pi)-F_e(\eta,q'\pi)
&=\left(\frac{1}{q'^2}-\frac{1}{q^2}\right)\left[\frac{\eta^4}{1296\pi^2}-\frac{\eta^5}{1944\pi^2}\right]+O(\eta^6). \tag*{(52)}
\end{aligned} F e ( η , q π ) − F e ( η , q ′ π ) = ( q ′2 1 − q 2 1 ) [ 1296 π 2 η 4 − 1944 π 2 η 5 ] + O ( η 6 ) . (52)
Simultaneously, expanding Eqs. (50) –(51) for the calibrated q = 1 q=1 q = 1 swap shows that non-commutativity between the Rabi drive − i ad H 1 -i\,\operatorname{ad}_{H_1} − i ad H 1 and the dissipator ε D [ L + ] \varepsilon\mathcal{D}[L_+] ε D [ L + ] pulls the effective Rabi frequency to Ω ε , 1 = π 2 − ε 2 / 9 \Omega_{\varepsilon,1}=\sqrt{\pi^2-\varepsilon^2/9} Ω ε , 1 = π 2 − ε 2 /9 , bending the exact survival-error curve slightly below the first-order ray κ ⋆ a ( ε ) \kappa_\star a(\varepsilon) κ ⋆ a ( ε ) at third order (Figure 5 (a)):
b ( ε ) − κ ⋆ a ( ε ) = − 2 ( 4 π 2 − 39 ) 81 π 2 ( π + 2 2 ) ε 3 + O ( ε 4 ) ≈ − 1.417626 × 10 − 4 ε 3 < 0. (53) \begin{aligned}
b(\varepsilon)-\kappa_\star a(\varepsilon)=-\frac{\sqrt{2}(4\pi^2-39)}{81\pi^2(\pi+2\sqrt{2})}\varepsilon^3+O(\varepsilon^4) \\
&\approx-1.417626\times10^{-4}\varepsilon^3<0. \tag*{(53)}
\end{aligned} b ( ε ) − κ ⋆ a ( ε ) = − 81 π 2 ( π + 2 2 ) 2 ( 4 π 2 − 39 ) ε 3 + O ( ε 4 ) ≈ − 1.417626 × 1 0 − 4 ε 3 < 0. (53)
Because the cubic coefficient is − 1.417626 × 10 − 4 -1.417626 \times10^{-4} − 1.417626 × 1 0 − 4 , at circuit-QED exposures ε ≤ 10 − 2 \varepsilon\le10^{-2} ε ≤ 1 0 − 2 the finite-noise excursion for the calibrated traceless jump L + L_{+} L + (λ = 0 \lambda= 0 λ = 0 ) is of order 10 − 10 10^{-10} 1 0 − 10 . By contrast, for the non-traceless rank-1 nilpotent jump L ⋆ : = ∣ ψ ( π / 8 ) ⟩ ⟨ ψ ( 7 π / 8 ) ∣ = − 2 2 I 2 − i 2 4 σ x − 1 2 σ z L_{\star} := |\psi(\pi/8)\rangle\langle\psi(7\pi/8)| = -\frac{\sqrt{2}}{2}I_{2} - \frac{i\sqrt{2}}{4}\sigma_{x} - \frac{1}{2}\sigma_{z} L ⋆ := ∣ ψ ( π /8 )⟩ ⟨ ψ ( 7 π /8 ) ∣ = − 2 2 I 2 − 4 i 2 σ x − 2 1 σ z (which shares the same traceless part L ~ ⋆ ∝ L − = L + † \widetilde{L}_{\star} \propto L_{-} = L_{+}^{\dagger} L ⋆ ∝ L − = L + † and first-order ratio κ ⋆ \kappa_{\star} κ ⋆ , but carries trace c ⋆ = − 2 4 c_{\star} = -\frac{\sqrt{2}}{4} c ⋆ = − 4 2 and Lamb shift Δ H L a m b = − 1 8 σ x \Delta H_{\mathrm{Lamb}} = -\frac{1}{8}\sigma_{x} Δ H Lamb = − 8 1 σ x ), retiming the drive as H ε ( λ ) = π 2 ( 1 + λ ε ) σ x H_{\varepsilon}(\lambda) = \frac{\pi}{2}(1+\lambda\varepsilon)\sigma_{x} H ε ( λ ) = 2 π ( 1 + λ ε ) σ x at λ ⋆ = ( π − 1 ) / ( 4 π 2 ) ≈ 0.054247 \lambda_{\star} = (\pi-1)/(4\pi^{2}) \approx0.054247 λ ⋆ = ( π − 1 ) / ( 4 π 2 ) ≈ 0.054247 produces a second-order O ( ε 2 ) O(\varepsilon^{2}) O ( ε 2 ) excursion b ( ε ) − κ ⋆ a ( ε ) = c min a ( ε ) 2 + O ( ε 3 ) b(\varepsilon)-\kappa_{\star}a(\varepsilon)=c_{\min}a(\varepsilon)^{2}+O(\varepsilon^{3}) b ( ε ) − κ ⋆ a ( ε ) = c m i n a ( ε ) 2 + O ( ε 3 ) with c min = − 2 / ( π + 2 2 ) 2 ≈ − 6.6464 × 10 − 3 c_{\min}=-\sqrt{2}/(\pi+2\sqrt{2})^{2}\approx-6.6464\times10^{-3} c m i n = − 2 / ( π + 2 2 ) 2 ≈ − 6.6464 × 1 0 − 3 (solid red curve in Figure 5 (a); full derivation in Appendix § 17 .§ 2 ).
Quantum-speed-limit pulse dwell scaling
When a control schedule includes idle dwell intervals at the poles before or after an active flip of duration T Q S L : = π / Ω max T_{\mathrm{QSL}} := \pi/\Omega_{\max} T QSL := π / Ω m a x set by the Mandelstam–Tamm quantum speed limit [40 , 41 ] , the active flip fraction is δ : = T Q S L / T ∈ ( 0 , 1 ] \delta:= T_{\mathrm{QSL}}/T \in(0,1] δ := T QSL / T ∈ ( 0 , 1 ] . For a general wait-flip-wait schedule idling at ∣ 0 ⟩ |0\rangle ∣0 ⟩ for fraction w + = 1 + η w 2 ( 1 − δ ) w_{+} = \frac{1+\eta_{w}}{2}(1-\delta) w + = 2 1 + η w ( 1 − δ ) , flipping at rate − π 2 δ σ x -\frac{\pi}{2\delta}\sigma_{x} − 2 δ π σ x over duration δ \delta δ , and idling at ∣ 1 ⟩ |1\rangle ∣1 ⟩ for w − = 1 − η w 2 ( 1 − δ ) w_{-} = \frac{1-\eta_{w}}{2}(1-\delta) w − = 2 1 − η w ( 1 − δ ) with wait asymmetry η w ∈ [ − 1 , 1 ] \eta_{w} \in[-1,1] η w ∈ [ − 1 , 1 ] , integrating the trajectory moments (Appendix § 18 ) yields
β ( δ , η w ) = 8 δ π 2 + η w 2 ( 1 − δ ) 2 1 − δ / 2 . (54) \beta(\delta,\eta_{w})=\sqrt{\frac{8\delta}{\pi^{2}}+\eta_{w}^{2}\frac{(1-\delta)^{2}}{1-\delta/2}}. \tag*{(54)} β ( δ , η w ) = π 2 8 δ + η w 2 1 − δ /2 ( 1 − δ ) 2 . (54)
Physically, one-sided pole idling (η w = ± 1 \eta_{w}=\pm1 η w = ± 1 ) exposes one state to longitudinal relaxation while the antipodal state rests in a dark state, causing the selectivity bound κ d w e l l ( δ ) = ( 1 − β ( δ , ± 1 ) ) / ( 1 + β ( δ , ± 1 ) ) = 3 π 2 − 16 8 π 2 δ + O ( δ 2 ) ≈ 0.172358 δ \kappa_{\mathrm{dwell}}(\delta)=(1-\beta(\delta,\pm1))/(1+\beta(\delta,\pm1))=\frac{3\pi^{2}-16}{8\pi^{2}}\delta+O(\delta^{2})\approx0.172358\,\delta κ dwell ( δ ) = ( 1 − β ( δ , ± 1 )) / ( 1 + β ( δ , ± 1 )) = 8 π 2 3 π 2 − 16 δ + O ( δ 2 ) ≈ 0.172358 δ to scale linearly with the speed-limit fraction δ \delta δ , whereas symmetric pole idling (η w = 0 \eta_{w}=0 η w = 0 ) suppresses transverse drift coupling and drives κ s y m ( δ ) = ( π − 2 2 δ ) / ( π + 2 2 δ ) → 1 \kappa_{\mathrm{sym}}(\delta)=(\pi-2\sqrt{2}\sqrt{\delta})/(\pi+2\sqrt{2}\sqrt{\delta})\to1 κ sym ( δ ) = ( π − 2 2 δ ) / ( π + 2 2 δ ) → 1 (Figure 3 (b)).
Colored thermal baths and calibration-free quantum thermometry
When the environment has a frequency-dependent spectral density matrix S ( ν ) = S ( s ) ( ν ) + i S ( a ) ( ν ) ⪰ 0 S(\nu)=S^{(s)}(\nu)+iS^{(a)}(\nu)\succeq0 S ( ν ) = S ( s ) ( ν ) + i S ( a ) ( ν ) ⪰ 0 (violation of the flat-spectrum white-noise approximation, such as 1 / f 1/f 1/ f flux noise or a resonant thermal reservoir), first-order Redfield perturbation theory [43 ] (Appendix § 19 , Proposition 5) shows that the survival error sum χ + + χ − \chi_{+}+\chi_{-} χ + + χ − couples exclusively to the frequency-even classical symmetric spectrum S ( s ) ( ν ) S^{(s)}(\nu) S ( s ) ( ν ) , whereas the directional split χ + − χ − \chi_{+}-\chi_{-} χ + − χ − couples exclusively to the frequency-odd quantum skew-symmetric spectrum S ( a ) ( ν ) S^{(a)}(\nu) S ( a ) ( ν ) . Consequently, any classical symmetric noise source (S ( a ) ( ν ) ≡ 0 S^{(a)}(\nu)\equiv0 S ( a ) ( ν ) ≡ 0 , including quasistatic 1 / f 1/f 1/ f phase noise) produces zero first-order directional split (χ + = χ − \chi_{+}=\chi_{-} χ + = χ − , κ = 1 \kappa=1 κ = 1 ). Conversely, consider an x x x -axis swap H q = − π 2 σ x H_{q}=-\frac{\pi}{2}\sigma_{x} H q = − 2 π σ x , whose trajectory centroid m q = ( 0 , 2 π , 0 ) T m_{q}=(0,\frac{2}{\pi},0)^{\mathsf{T}} m q = ( 0 , π 2 , 0 ) T lies purely along e y e_{y} e y . For a standard longitudinal thermal bath quantized along e z e_{z} e z (t ∥ e z t\parallel e_{z} t ∥ e z ), m q ⋅ t = 0 m_{q}\cdot t=0 m q ⋅ t = 0 and the x x x -swap directional split vanishes identically. However, if the thermal boson bath at frequency ω b \omega_{b} ω b and temperature T b a t h T_{\mathrm{bath}} T bath (γ ↓ / γ ↑ = exp [ ℏ ω b / ( k B T b a t h ) ] \gamma_{\downarrow}/\gamma_{\uparrow}=\exp[\hbar\omega_{b}/(k_{B}T_{\mathrm{bath}})] γ ↓ / γ ↑ = exp [ ℏ ω b / ( k B T bath )] ), obeying the Kubo–Martin–Schwinger detailed-balance relation [44 , 45 ] is quantized along the transverse e y e_{y} e y axis (L ↓ , y = γ ↓ ∣ + y ⟩ ⟨ − y ∣ L_{\downarrow,y}=\sqrt{\gamma_{\downarrow}}|+y\rangle\langle-y| L ↓ , y = γ ↓ ∣ + y ⟩ ⟨ − y ∣ , L ↑ , y = γ ↑ ∣ − y ⟩ ⟨ + y ∣ L_{\uparrow,y}=\sqrt{\gamma_{\uparrow}}|-y\rangle\langle+y| L ↑ , y = γ ↑ ∣ − y ⟩ ⟨ + y ∣ ), so that S y = 1 2 ( γ ↓ + γ ↑ ) diag ( 1 , 0 , 1 ) S_{y}=\frac{1}{2}(\gamma_{\downarrow}+\gamma_{\uparrow})\operatorname{diag}(1,0,1) S y = 2 1 ( γ ↓ + γ ↑ ) diag ( 1 , 0 , 1 ) and t y = 2 ( γ ↓ − γ ↑ ) e y t_{y}=2(\gamma_{\downarrow}-\gamma_{\uparrow})e_{y} t y = 2 ( γ ↓ − γ ↑ ) e y )—or equivalently, if the qubit is coupled to a longitudinal e z e_{z} e z thermal bath and one either prepares and measures the equatorial basis states ∣ ± y ⟩ |\pm y\rangle ∣ ± y ⟩ (via X \sqrt{X} X basis rotations) during a σ x \sigma_{x} σ x -dressed spin-locked swap or executes a single polar X \sqrt{X} X gate (m z = 2 / π m_{z}=2/\pi m z = 2/ π , where the prefactor δ / ( 3 π ∣ q ∣ ) \delta/(3\pi|q|) δ / ( 3 π ∣ q ∣ ) becomes δ / ( 3 π ) \delta/(3\pi) δ / ( 3 π ) )—then Tr ( S y G q ) = 3 4 ( γ ↓ + γ ↑ ) \operatorname{Tr}(S_{y}G_{q})=\frac{3}{4}(\gamma_{\downarrow}+\gamma_{\uparrow}) Tr ( S y G q ) = 4 3 ( γ ↓ + γ ↑ ) and − 1 2 m q ⋅ t y = − 2 π ( γ ↓ − γ ↑ ) -\frac{1}{2}m_{q}\cdot t_{y}=-\frac{2}{\pi}(\gamma_{\downarrow}-\gamma_{\uparrow}) − 2 1 m q ⋅ t y = − π 2 ( γ ↓ − γ ↑ ) . The overall system-bath coupling strength then cancels out in the normalized directional contrast, yielding a calibration-free quantum thermometer:
∣ χ + ( q ) − χ − ( q ) ∣ χ + ( q ) + χ − ( q ) = 8 3 π ∣ q ∣ tanh ( ℏ ω b 2 k B T b a t h ) ≤ 8 3 π ∣ q ∣ . (55) \frac{|\chi_{+}(q)-\chi_{-}(q)|}{\chi_{+}(q)+\chi_{-}(q)}=\frac{8}{3\pi|q|}\tanh\left(\frac{\hbar\omega_{b}}{2k_{B}T_{\mathrm{bath}}}\right)\le\frac{8}{3\pi|q|}. \tag*{(55)} χ + ( q ) + χ − ( q ) ∣ χ + ( q ) − χ − ( q ) ∣ = 3 π ∣ q ∣ 8 tanh ( 2 k B T bath ℏ ω b ) ≤ 3 π ∣ q ∣ 8 . (55)
Multilevel qutrit cascade shelving and leakage diagnostics
Finally, when evolution extends beyond the two-level qubit subspace to a three-level transmon qutrit (d = 3 d=3 d = 3 , H = span { ∣ 0 ⟩ , ∣ 1 ⟩ , ∣ 2 ⟩ } \mathcal{H}=\operatorname{span}\{|0\rangle,|1\rangle,|2\rangle\} H = span { ∣0 ⟩ , ∣1 ⟩ , ∣2 ⟩} ), the Lie algebra s u ( 3 ) \mathfrak{su}(3) su ( 3 ) possesses non-vanishing completely symmetric structure constants d a b c ≠ 0 d_{abc}\ne0 d ab c = 0 . Whereas on a qubit (d = 2 d=2 d = 2 , d a b c = 0 d_{abc}=0 d ab c = 0 ) the entanglement fidelity F e F_{\mathrm{e}} F e is identically invariant under bath chirality reversal L ~ α ↔ L ~ α † \widetilde{L}_{\alpha}\leftrightarrow\widetilde{L}_{\alpha}^{\dagger} L α ↔ L α † at all orders in ε \varepsilon ε , on a qutrit undergoing cascade energy relaxation ε D [ ∣ 0 ⟩ ⟨ 1 ∣ ] + ε η D [ ∣ 1 ⟩ ⟨ 2 ∣ ] \varepsilon\mathcal{D}[|0\rangle\langle1|]+\varepsilon\eta\mathcal{D}[|1\rangle\langle2|] ε D [ ∣0 ⟩ ⟨ 1∣ ] + ε η D [ ∣1 ⟩ ⟨ 2∣ ] with decay ratio η = γ 12 / γ 01 \eta=\gamma_{12}/\gamma_{01} η = γ 12 / γ 01 , two endpoint-matched 2 π 2\pi 2 π shelving rotations H 02 H_{02} H 02 and H 12 H_{12} H 12 share the same first-order infidelity 1 + η 3 ε \frac{1+\eta}{3}\varepsilon 3 1 + η ε but split at second order:
F e ( ε ; H 02 ) − F e ( ε ; H 12 ) = 6 − 4 η − 3 η 2 144 ε 2 + O ( ε 3 ) , (56) F_{\mathrm{e}}(\varepsilon;H_{02})-F_{\mathrm{e}}(\varepsilon;H_{12})=\frac{6-4\eta-3\eta^2}{144}\varepsilon^2+O(\varepsilon^3), \tag*{(56)} F e ( ε ; H 02 ) − F e ( ε ; H 12 ) = 144 6 − 4 η − 3 η 2 ε 2 + O ( ε 3 ) , (56)
which vanishes and reverses sign at the critical cascade ratio η c r i t = ( 22 − 2 ) / 3 ≈ 0.896805 \eta_{\mathrm{crit}}=(\sqrt{22}-2)/3\approx0.896805 η crit = ( 22 − 2 ) /3 ≈ 0.896805 (Appendix § 19 , Theorem 3). Additional mathematical complements, including the three-mode qubit Slepian concentration pencil [46 ] and multilevel bipartite Cauchy constructions (Appendix § 20 ), the statistical Cramér–Rao and repetition-code decoding theory (Appendix § 21 ), and the 12-digit analytical gate audit and 1,203,200-bath random CPTP Lindblad numerical verification (Figure 4 (b) and Appendix § 22 , Table 3 –Table 4 ), are detailed in the Appendices.
Table 3. Complete 12-digit analytical and multi-start L-BFGS-B numerical gate audit for single-qubit rectangular, Gaussian (σ = T / 4 \sigma=T/4 σ = T /4 ), and tilted-quadrature DRAG (σ = T / 4 \sigma=T/4 σ = T /4 , η D R A G = 0.02 \eta_{\mathrm{DRAG}}=0.02 η DRAG = 0.02 ) trajectories in the weak-damping Markovian regime (d = 2 d=2 d = 2 , rank ( G ) = 3 \operatorname{rank}(G)=3 rank ( G ) = 3 ).
Table 4. Adversarial numerical verification of Theorem 1 and Theorem 2 across 1,203,200 random multi-jump CPTP Markovian baths (8 × \times × 150,400, seeds 202601–202608), multi-start L-BFGS-B optimization, superoperator Fréchet derivatives, and Richardson-extrapolated 4 × \times × 4 Lindblad channel exponentiation exp ( L 0 + ε D ) \exp(\mathcal{L}_0+\varepsilon\mathcal{D}) exp ( L 0 + ε D ) at ε 0 = 10 − 7 \varepsilon_0=10^{-7} ε 0 = 1 0 − 7 .
Conclusion and outlook
We have shown that while twirled gate benchmarks such as Clifford randomized benchmarking and average gate fidelity assign a single path-independent decay rate to a noisy quantum gate at leading order, untwirled single-qubit circuits exhibit trajectory-dependent directional survival-error asymmetries governed by a compact geometric law. For any piecewise-continuous qubit control path r ( s ) ∈ S 2 \boldsymbol{r}(s)\in S^2 r ( s ) ∈ S 2 in a Markovian environment, the forward and backward survival-error slopes χ ± \chi_{\pm} χ ± depend on the trajectory solely through its time-averaged Bloch vector m \boldsymbol{m} m and transverse residence covariance matrix G G G , obeying the cofactor inequality β ( r ) = m T G m / det G ≤ 1 \beta(\boldsymbol{r})=\sqrt{\boldsymbol{m}^{T}G\boldsymbol{m}/\det G}\leq1 β ( r ) = m T G m / det G ≤ 1 (Theorem 1). For time-independent single-qubit π \pi π swaps, complete positivity links non-unital drift to unital diffusion and enforces the selectivity bound κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) ≈ 0.052456 \kappa_{\star}=(\pi-2\sqrt{2})/(\pi+2\sqrt{2})\approx0.052456 κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) ≈ 0.052456 (19.0635 : 1, Theorem 2). On superconducting transmon processors, native longitudinal T 1 T_1 T 1 relaxation induces a 12.23 : 1 directional survival-error split (κ T 1 ≈ 0.081767 \kappa_{T_1}\approx0.081767 κ T 1 ≈ 0.081767 ) on polar X \sqrt{X} X gates and a winding-independent 19 : 3 ≈ 6.33 \approx6.33 ≈ 6.33 : 1 split (κ T 1 = 3 / 19 \kappa_{T_1}=3/19 κ T 1 = 3/19 ) on single-pulse Hadamard gates while leaving symmetric rectangular, Gaussian, and DRAG X X X rotations balanced (κ T 1 = 1 \kappa_{T_1}=1 κ T 1 = 1 ), and our two-branch differential lock-in protocols enable experimental extraction of both transverse-drift swap splits and longitudinal-T 1 T_1 T 1 X \sqrt{X} X splits free of first-order SPAM bias.
These results open several immediate directions for experimental quantum characterization and fault-tolerant architecture design. First, in fault-tolerant quantum error correction, standard minimum-weight perfect-matching (MWPM) and belief-propagation decoders for surface and XZZX codes [32 , 33 , 34 ] assume symmetric bit-flip transition probabilities (p + = p − p_{+}=p_{-} p + = p − ) obtained from Pauli-twirled benchmarks; incorporating the analytical trajectory-dependent directional error rates χ ± ( D ; r ) \chi_{\pm}(\mathcal{D};\boldsymbol{r}) χ ± ( D ; r ) directly into asymmetric branch weights on the decoding graph can prevent systematic logical-error underestimation (Remark 2 in Appendix § 21 ) and enhance fault-tolerant thresholds under native T 1 T_1 T 1 damping. Second, extending the two-moment trajectory law and matrix cofactor bound from s u ( 2 ) \mathfrak{su}(2) su ( 2 ) to the 15-dimensional Lie algebra s u ( 4 ) \mathfrak{su}(4) su ( 4 ) will enable analytical bounds on state-dependent transition asymmetries in two-qubit entangling gates such as cross-resonance and tunable-coupler CZ and iSWAP operations.
Data and code availability
All verification, simulation, table-generation, and figure-generation code, datasets, deterministic random seeds, and LaTeX sources are permanently archived in an open-access Zenodo repository at https://doi.org/10.5281/zenodo.24116128 [42 ] . Simulation and verification scripts are released under the MIT License, and manuscript text and figures are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
Author contributions
A.P. conceived the theoretical framework, proved the mathematical theorems, performed the analytical and numerical calculations, designed the lock-in protocol, and wrote the manuscript.
Acknowledgements
The author thanks colleagues in the open quantum systems and quantum control communities for helpful discussions on gate characterization and randomized benchmarking. This research received no external funding. The author declares no competing financial or non-financial interests.
Superoperator trace, Liouville determinant, and first-order trajectory invariance
This appendix provides the self-contained derivation of the Liouville determinant identity (8) , first-order scalar trajectory invariance (9) , and uniqueness of the SU ( d ) \operatorname{SU}(d) SU ( d ) -invariant trace functional stated in § 3.1 .
Lemma 1 (Superoperator trace, Liouville determinant, and Schur uniqueness). On the d 2 d^2 d 2 -dimensional operator space B ( H ) \mathcal{B}(\mathcal{H}) B ( H ) (H ≅ C d \mathcal{H}\cong\mathbb{C}^{d} H ≅ C d , d ≥ 2 d\geq2 d ≥ 2 ) equipped with the Hilbert–Schmidt inner product ⟨ ⟨ A ∣ B ⟩ ⟩ = Tr ( A † B ) \langle\langle A|B\rangle\rangle=\operatorname{Tr}(A^\dagger B) ⟨⟨ A ∣ B ⟩⟩ = Tr ( A † B ) :
Every elementary multiplication superoperator S A , B ( X ) : = A X B S_{A,B}(X):=AXB S A , B ( X ) := A X B has superoperator trace Tr ( S A , B ) = Tr ( A ) Tr ( B ) \operatorname{Tr}(S_{A,B})=\operatorname{Tr}(A)\operatorname{Tr}(B) Tr ( S A , B ) = Tr ( A ) Tr ( B ) . Consequently, Tr ( ad H ( s ) ) = 0 \operatorname{Tr}(\operatorname{ad}_{H}(s))=0 Tr ( ad H ( s )) = 0 and Tr ( D ( s ) ) = − d Γ ( s ) \operatorname{Tr}(\mathcal{D}(s))=-d\Gamma(s) Tr ( D ( s )) = − d Γ ( s ) , where Γ ( s ) = ∑ α γ α ( s ) Tr ( L ~ α ( s ) † L ~ α ( s ) ) \Gamma(s)=\sum_{\alpha}\gamma_{\alpha}(s)\operatorname{Tr}(\widetilde{L}_{\alpha}(s)^\dagger\widetilde{L}_{\alpha}(s)) Γ ( s ) = ∑ α γ α ( s ) Tr ( L α ( s ) † L α ( s )) depends solely on the traceless jump parts L ~ α ( s ) \widetilde{L}_{\alpha}(s) L α ( s ) .
The dynamical propagator Φ ε ( 1 ) \Phi_{\varepsilon}(1) Φ ε ( 1 ) of the time-rescaled GKSL master equation (2) obeys det Φ ε ( 1 ) = exp ( − d ε Γ t o t ) \det\Phi_{\varepsilon}(1)=\exp(-d\varepsilon\Gamma_{\mathrm{tot}}) det Φ ε ( 1 ) = exp ( − d ε Γ tot ) at all orders in ε ≥ 0 \varepsilon\geq0 ε ≥ 0 , and its first-order entanglement infidelity 1 − F e ( Φ ε , U 0 ( 1 ) ) = ε d Γ t o t + O ( ε 2 ) 1-F_{\mathrm{e}}(\Phi_{\varepsilon},U_{0}(1))=\frac{\varepsilon}{d}\Gamma_{\mathrm{tot}}+O(\varepsilon^{2}) 1 − F e ( Φ ε , U 0 ( 1 )) = d ε Γ tot + O ( ε 2 ) is independent of the unitary trajectory U 0 ( s ) U_{0}(s) U 0 ( s ) .
Up to scalar multiplication, D ↦ Tr ( D ) = − d Γ \mathcal{D}\mapsto\operatorname{Tr}(\mathcal{D})=-d\Gamma D ↦ Tr ( D ) = − d Γ is the unique SU ( d ) \operatorname{SU}(d) SU ( d ) -conjugation-invariant linear functional on trace-preserving superoperators.
Proof. In the orthonormal matrix-unit basis { E i j : = ∣ i ⟩ ⟨ j ∣ } i , j = 0 d − 1 \{E_{ij}:=|i\rangle\langle j|\}_{i,j=0}^{d-1} { E ij := ∣ i ⟩ ⟨ j ∣ } i , j = 0 d − 1 of B ( H ) \mathcal{B}(\mathcal{H}) B ( H ) , the superoperator trace of S A , B ( X ) = A X B S_{A,B}(X)=AXB S A , B ( X ) = A X B is
Tr ( S A , B ) = ∑ i , j = 0 d − 1 Tr ( E i j † A E i j B ) = ∑ i , j = 0 d − 1 ⟨ i ∣ A ∣ i ⟩ ⟨ j ∣ B ∣ j ⟩ = Tr ( A ) Tr ( B ) . (57) \operatorname{Tr}(S_{A,B})=\sum_{i,j=0}^{d-1}\operatorname{Tr}(E_{ij}^{\dagger}AE_{ij}B)=\sum_{i,j=0}^{d-1}\langle i|A|i\rangle\langle j|B|j\rangle=\operatorname{Tr}(A)\operatorname{Tr}(B). \tag*{(57)} Tr ( S A , B ) = i , j = 0 ∑ d − 1 Tr ( E ij † A E ij B ) = i , j = 0 ∑ d − 1 ⟨ i ∣ A ∣ i ⟩ ⟨ j ∣ B ∣ j ⟩ = Tr ( A ) Tr ( B ) . (57)
Consequently, Tr ( ad H ( s ) ) = Tr ( H ( s ) ) Tr ( I d ) − Tr ( I d ) Tr ( H ( s ) ) = 0 \operatorname{Tr}(\operatorname{ad}_{H}(s))=\operatorname{Tr}(H(s))\operatorname{Tr}(\mathbb{I}_{d})-\operatorname{Tr}(\mathbb{I}_{d})\operatorname{Tr}(H(s))=0 Tr ( ad H ( s )) = Tr ( H ( s )) Tr ( I d ) − Tr ( I d ) Tr ( H ( s )) = 0 . Substituting the trace-traceless decomposition L α ( s ) = c α ( s ) I d + L ~ α ( s ) L_{\alpha}(s)=c_{\alpha}(s)\mathbb{I}_{d}+\widetilde{L}_{\alpha}(s) L α ( s ) = c α ( s ) I d + L α ( s ) (Tr ( L ~ α ( s ) ) = 0 \operatorname{Tr}(\widetilde{L}_{\alpha}(s))=0 Tr ( L α ( s )) = 0 ) into Eq. (3) cancels the d 2 ∣ c α ( s ) ∣ 2 d^{2}|c_{\alpha}(s)|^{2} d 2 ∣ c α ( s ) ∣ 2 terms identically:
Tr ( D ( s ) ) = ∑ α γ α ( s ) [ ∣ Tr ( L α ( s ) ) ∣ 2 − 1 2 Tr ( L α ( s ) † L α ( s ) ) Tr ( I d ) − 1 2 Tr ( I d ) Tr ( L α ( s ) † L α ( s ) ) ] = − d ∑ α γ α ( s ) Tr ( L ~ α ( s ) † L ~ α ( s ) ) = − d Γ ( s ) . (58) \begin{aligned}
\operatorname{Tr}(\mathcal{D}(s))&=\sum_{\alpha}\gamma_{\alpha}(s)\left[\left|\operatorname{Tr}(L_{\alpha}(s))\right|^{2}-\frac{1}{2}\operatorname{Tr}(L_{\alpha}(s)^{\dagger}L_{\alpha}(s))\operatorname{Tr}(\mathbb{I}_{d})-\frac{1}{2}\operatorname{Tr}(\mathbb{I}_{d})\operatorname{Tr}(L_{\alpha}(s)^{\dagger}L_{\alpha}(s))\right]\\
&=-d\sum_{\alpha}\gamma_{\alpha}(s)\operatorname{Tr}(\widetilde{L}_{\alpha}(s)^{\dagger}\widetilde{L}_{\alpha}(s))=-d\Gamma(s). \tag*{(58)}
\end{aligned} Tr ( D ( s )) = α ∑ γ α ( s ) [ ∣ Tr ( L α ( s )) ∣ 2 − 2 1 Tr ( L α ( s ) † L α ( s )) Tr ( I d ) − 2 1 Tr ( I d ) Tr ( L α ( s ) † L α ( s )) ] = − d α ∑ γ α ( s ) Tr ( L α ( s ) † L α ( s )) = − d Γ ( s ) . (58)
Applying Jacobi’s determinant formula to the time-rescaled master equation (2) gives
det Φ ε ( 1 ) = exp ( ∫ 0 1 Tr ( L ε ( s ) ) d s ) = e − d ε Γ t o t , (59) \det\Phi_{\varepsilon}(1)=\exp\left(\int_{0}^{1}\operatorname{Tr}(\mathcal{L}_{\varepsilon}(s))\,\mathrm{d}s\right)=e^{-d\varepsilon\Gamma_{\mathrm{tot}}}, \tag*{(59)} det Φ ε ( 1 ) = exp ( ∫ 0 1 Tr ( L ε ( s )) d s ) = e − d ε Γ tot , (59)
which establishes Eq. (8) . Taking d − 2 Tr ( ⋅ ) d^{-2}\operatorname{Tr}(\cdot) d − 2 Tr ( ⋅ ) of the Dyson expansion (6) and using cyclic invariance Tr ( D ^ ( s ) ) = Tr ( U 0 ( s ) † D ( s ) U 0 ( s ) ) = Tr ( D ( s ) ) = − d Γ ( s ) \operatorname{Tr}(\widehat{\mathcal{D}}(s))=\operatorname{Tr}(U_{0}(s)^{\dagger}\mathcal{D}(s)U_{0}(s))=\operatorname{Tr}(\mathcal{D}(s))=-d\Gamma(s) Tr ( D ( s )) = Tr ( U 0 ( s ) † D ( s ) U 0 ( s )) = Tr ( D ( s )) = − d Γ ( s ) yields Eq. (9) . Finally, by Schur’s lemma on the irreducible SU ( d ) \operatorname{SU}(d) SU ( d ) -module decomposition B ( H ) = C I d ⊕ s l ( d , C ) \mathcal{B}(\mathcal{H})=\mathbb{C}\mathbb{I}_{d}\oplus\mathfrak{sl}(d,\mathbb{C}) B ( H ) = C I d ⊕ sl ( d , C ) , the commutant of the adjoint representation { U : U ∈ SU ( d ) } \{U:U\in\operatorname{SU}(d)\} { U : U ∈ SU ( d )} is spanned by the identity and traceless projectors P I P_{\mathbb{I}} P I and P s l P_{\mathfrak{sl}} P sl ; since Tr ( P I D ) = 0 \operatorname{Tr}(P_{\mathbb{I}}\mathcal{D})=0 Tr ( P I D ) = 0 by trace preservation (D † ( I d ) = 0 \mathcal{D}^{\dagger}(\mathbb{I}_{d})=0 D † ( I d ) = 0 ), any SU ( d ) \operatorname{SU}(d) SU ( d ) -conjugation-invariant linear functional of D \mathcal{D} D is a scalar multiple of Tr ( P s l D ) = Tr ( D ) = − d Γ \operatorname{Tr}(P_{\mathfrak{sl}}\mathcal{D})=\operatorname{Tr}(\mathcal{D})=-d\Gamma Tr ( P sl D ) = Tr ( D ) = − d Γ . ▫
Pauli transfer matrix proof of the two-moment trajectory law
This appendix establishes an independent Pauli transfer matrix (PTM) derivation of the two-moment decomposition (22) in Theorem 1. In the normalized Pauli basis { B 0 , B 1 , B 2 , B 3 } : = { I 2 / 2 , σ x / 2 , σ y / 2 , σ z / 2 } \{B_0, B_1, B_2, B_3\} := \{\mathbb{I}_2/\sqrt{2}, \sigma_x/\sqrt{2}, \sigma_y/\sqrt{2}, \sigma_z/\sqrt{2}\} { B 0 , B 1 , B 2 , B 3 } := { I 2 / 2 , σ x / 2 , σ y / 2 , σ z / 2 } , a qubit GKSL dissipator D \mathcal{D} D with traceless jump operators L ~ α = ( a α + i b α ) ⋅ σ \tilde{L}_{\alpha} = (\mathbf{a}_{\alpha} + i\mathbf{b}_{\alpha}) \cdot\boldsymbol{\sigma} L ~ α = ( a α + i b α ) ⋅ σ has PTM entries [ D ] 0 j = Tr ( B 0 D ( B j ) ) = 0 [\mathcal{D}]_{0j} = \operatorname{Tr}(B_0\mathcal{D}(B_j)) = 0 [ D ] 0 j = Tr ( B 0 D ( B j )) = 0 (by trace preservation),
[ D ] i 0 = Tr ( B i D ( B 0 ) ) = 1 2 Tr ( σ i D ( I 2 ) ) = 1 2 t i ( i ∈ { 1 , 2 , 3 } ) , [\mathcal{D}]_{i0} = \operatorname{Tr}(B_i\mathcal{D}(B_0)) = \frac{1}{2}\operatorname{Tr}(\sigma_i\mathcal{D}(\mathbb{I}_2)) = \frac{1}{2}t_i \qquad(i \in\{1,2,3\}), [ D ] i 0 = Tr ( B i D ( B 0 )) = 2 1 Tr ( σ i D ( I 2 )) = 2 1 t i ( i ∈ { 1 , 2 , 3 }) ,
and unital 3 × 3 3 \times3 3 × 3 damping block [ D ] i j = 1 2 Tr ( σ i D ( σ j ) ) = [ S − Tr ( S ) I 3 ] i j [\mathcal{D}]_{ij} = \frac{1}{2}\operatorname{Tr}(\sigma_i\mathcal{D}(\sigma_j)) = [S-\operatorname{Tr}(S)\mathbb{I}_3]_{ij} [ D ] ij = 2 1 Tr ( σ i D ( σ j )) = [ S − Tr ( S ) I 3 ] ij ( i , j ∈ { 1 , 2 , 3 } ) (i,j \in\{1,2,3\}) ( i , j ∈ { 1 , 2 , 3 }) :
[ D ] P T M = ( 0 0 T 1 2 t S − Tr ( S ) I 3 ) . (60) [\mathcal{D}]_{\mathrm{PTM}} =
\begin{pmatrix}
0 & \mathbf{0}^{T}\\
\frac{1}{2}\mathbf{t} & S-\operatorname{Tr}(S)\mathbb{I}_3
\end{pmatrix}. \tag*{(60)} [ D ] PTM = ( 0 2 1 t 0 T S − Tr ( S ) I 3 ) . (60)
With the evolving pure-state Pauli vectors v ± ( s ) : = [ U 0 ( s ) ] P T M v ± ( 0 ) = 1 2 ( 1 , ± r ( s ) T ) T \mathbf{v}_{\pm}(s) := [\mathcal{U}_0(s)]_{\mathrm{PTM}}\mathbf{v}_{\pm}(0) = \frac{1}{\sqrt{2}}(1,\pm\mathbf{r}(s)^{T})^{T} v ± ( s ) := [ U 0 ( s ) ] PTM v ± ( 0 ) = 2 1 ( 1 , ± r ( s ) T ) T and ∥ r ( s ) ∥ 2 = 1 \|\mathbf{r}(s)\|^2=1 ∥ r ( s ) ∥ 2 = 1 , integrating − v ± ( 0 ) T [ U 0 ( s ) † D U 0 ( s ) ] P T M v ± ( 0 ) = − v ± ( s ) T [ D ] P T M v ± ( s ) -\mathbf{v}_{\pm}(0)^{T}[\mathcal{U}_0(s)^{\dagger}\mathcal{D}\mathcal{U}_0(s)]_{\mathrm{PTM}}\mathbf{v}_{\pm}(0)=-\mathbf{v}_{\pm}(s)^{T}[\mathcal{D}]_{\mathrm{PTM}}\mathbf{v}_{\pm}(s) − v ± ( 0 ) T [ U 0 ( s ) † D U 0 ( s ) ] PTM v ± ( 0 ) = − v ± ( s ) T [ D ] PTM v ± ( s ) over s ∈ [ 0 , 1 ] s\in[0,1] s ∈ [ 0 , 1 ] gives
χ ± ( D ; r ) = − 1 2 ∫ 0 1 [ ± 1 2 r ( s ) ⋅ t + r ( s ) T ( S − Γ I 3 ) r ( s ) ] d s = 1 2 Tr ( S ∫ 0 1 ( I 3 − r ( s ) r ( s ) T ) d s ) + 1 4 ( ∫ 0 1 r ( s ) d s ) ⋅ t = 1 2 Tr ( S G ) + 1 4 m ⋅ t . (61) \begin{aligned}
\chi_{\pm}(\mathcal{D};\mathbf{r})=-\frac{1}{2}\int_{0}^{1}\left[\pm\frac{1}{2}\mathbf{r}(s)\cdot\mathbf{t}+\mathbf{r}(s)^{T}(S-\Gamma\mathbb{I}_3)\mathbf{r}(s)\right]\,ds \\
&=\frac{1}{2}\operatorname{Tr}\left(S\int_{0}^{1}(\mathbb{I}_3-\mathbf{r}(s)\mathbf{r}(s)^{T})\,ds\right)+\frac{1}{4}\left(\int_{0}^{1}\mathbf{r}(s)\,ds\right)\cdot\mathbf{t} \\
&=\frac{1}{2}\operatorname{Tr}(SG)+\frac{1}{4}\mathbf{m}\cdot\mathbf{t}. \tag*{(61)}
\end{aligned} χ ± ( D ; r ) = − 2 1 ∫ 0 1 [ ± 2 1 r ( s ) ⋅ t + r ( s ) T ( S − Γ I 3 ) r ( s ) ] d s = 2 1 Tr ( S ∫ 0 1 ( I 3 − r ( s ) r ( s ) T ) d s ) + 4 1 ( ∫ 0 1 r ( s ) d s ) ⋅ t = 2 1 Tr ( S G ) + 4 1 m ⋅ t . (61)
recovering Eq. (22) of Theorem 1.
Constrained cross-product minimum and extremal jump uniqueness for Theorem 1
This appendix proves the constrained cross-product minimization identity, the matrix cofactor inequality adj ( G ) − m m T ⪰ 0 \operatorname{adj}(G)-\mathbf{m}\mathbf{m}^{T}\succeq0 adj ( G ) − m m T ⪰ 0 , and the uniqueness of the rank-1 extremal jump operator L ~ e x t \tilde{L}_{\mathrm{ext}} L ~ ext in Theorem 1.
Lemma 2 (Constrained cross-product minimum and trajectory cofactor inequality). For any symmetric positive-definite matrix G ≻ 0 G\succ0 G ≻ 0 on R 3 \mathbb{R}^{3} R 3 and any target cross-product vector w ∈ R 3 \mathbf{w}\in\mathbb{R}^{3} w ∈ R 3 , the constrained quadratic form a T G a + b T G b \mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b} a T G a + b T G b over a , b ∈ R 3 \mathbf{a},\mathbf{b}\in\mathbb{R}^{3} a , b ∈ R 3 satisfying a × b = w \mathbf{a}\times\mathbf{b}=\mathbf{w} a × b = w has the exact minimum
min a , b ∈ R 3 a × b = w ( a T G a + b T G b ) = 2 w T adj ( G ) w . (62) \min_{\substack{\mathbf{a},\mathbf{b}\in\mathbb{R}^{3}\\ \mathbf{a}\times\mathbf{b}=\mathbf{w}}}\left(\mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b}\right)=2\sqrt{\mathbf{w}^{T}\operatorname{adj}(G)\mathbf{w}}. \tag*{(62)} a , b ∈ R 3 a × b = w min ( a T G a + b T G b ) = 2 w T adj ( G ) w . (62) Consequently, for every piecewise-continuous Bloch trajectory r : [ 0 , 1 ] → S 2 \mathbf{r}:[0,1]\to S^{2} r : [ 0 , 1 ] → S 2 with moments ( m , G ) (\mathbf{m},G) ( m , G ) , the matrix cofactor inequality adj ( G ) − m m T ⪰ 0 \operatorname{adj}(G)-\mathbf{m}\mathbf{m}^{T}\succeq0 adj ( G ) − m m T ⪰ 0 holds, implying β ( r ) = m T G m / det G ≤ 1 \beta(\mathbf{r})=\sqrt{\mathbf{m}^{T}G\mathbf{m}/\det G}\leq1 β ( r ) = m T G m / det G ≤ 1 whenever det G > 0 \det G>0 det G > 0 .
Proof. To verify Eq. (62) , define whitened vectors a ~ : = G 1 / 2 a \tilde{\mathbf{a}}:=G^{1/2}\mathbf{a} a ~ := G 1/2 a and b ~ : = G 1 / 2 b \tilde{\mathbf{b}}:=G^{1/2}\mathbf{b} b ~ := G 1/2 b . The cofactor transformation identity ( M a ) × ( M b ) = adj ( M T ) ( a × b ) (M\mathbf{a})\times(M\mathbf{b})=\operatorname{adj}(M^{T})(\mathbf{a}\times\mathbf{b}) ( M a ) × ( M b ) = adj ( M T ) ( a × b ) applied to M = G 1 / 2 M=G^{1/2} M = G 1/2 gives a ~ × b ~ = ( det G ) 1 / 2 G − 1 / 2 w = adj ( G ) 1 / 2 w \tilde{\mathbf{a}}\times\tilde{\mathbf{b}}=(\det G)^{1/2}G^{-1/2}\mathbf{w}=\operatorname{adj}(G)^{1/2}\mathbf{w} a ~ × b ~ = ( det G ) 1/2 G − 1/2 w = adj ( G ) 1/2 w . Applying the arithmetic–geometric mean inequality and Lagrange’s identity ∥ a ~ × b ~ ∥ ≤ ∥ a ~ ∥ ∥ b ~ ∥ \|\tilde{\mathbf{a}}\times\tilde{\mathbf{b}}\|\leq\|\tilde{\mathbf{a}}\|\|\tilde{\mathbf{b}}\| ∥ a ~ × b ~ ∥ ≤ ∥ a ~ ∥∥ b ~ ∥ yields
a T G a + b T G b = ∥ a ~ ∥ 2 + ∥ b ~ ∥ 2 ≥ 2 ∥ a ~ ∥ ∥ b ~ ∥ ≥ 2 ∥ a ~ × b ~ ∥ = 2 w T adj ( G ) w , (63) \mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b}=\|\tilde{\mathbf{a}}\|^{2}+\|\tilde{\mathbf{b}}\|^{2}\geq2\|\tilde{\mathbf{a}}\|\|\tilde{\mathbf{b}}\|\geq2\|\tilde{\mathbf{a}}\times\tilde{\mathbf{b}}\|=2\sqrt{\mathbf{w}^{T}\operatorname{adj}(G)\mathbf{w}}, \tag*{(63)} a T G a + b T G b = ∥ a ~ ∥ 2 + ∥ b ~ ∥ 2 ≥ 2∥ a ~ ∥∥ b ~ ∥ ≥ 2∥ a ~ × b ~ ∥ = 2 w T adj ( G ) w , (63)
with equality if and only if a ~ ⊥ b ~ ⊥ adj ( G ) 1 / 2 w \tilde{\mathbf{a}}\perp\tilde{\mathbf{b}}\perp\operatorname{adj}(G)^{1/2}\mathbf{w} a ~ ⊥ b ~ ⊥ adj ( G ) 1/2 w and ∥ a ~ ∥ = ∥ b ~ ∥ \|\tilde{\mathbf{a}}\|=\|\tilde{\mathbf{b}}\| ∥ a ~ ∥ = ∥ b ~ ∥ . Because the integrated jump variance (25) is non-negative for every a , b ∈ R 3 \mathbf{a},\mathbf{b}\in\mathbb{R}^{3} a , b ∈ R 3 ( a T G a + b T G b ≥ 2 ∣ w ⋅ m ∣ ) (\mathbf{a}^{T}G\mathbf{a}+\mathbf{b}^{T}G\mathbf{b}\geq2|\mathbf{w}\cdot\mathbf{m}|) ( a T G a + b T G b ≥ 2∣ w ⋅ m ∣ ) , minimizing the left-hand side over a × b = w \mathbf{a}\times\mathbf{b}=\mathbf{w} a × b = w via Eq. (62) (and taking the limit G + ϵ I 3 → G ⪰ 0 G+\epsilon\mathbb{I}_3\to G\succeq0 G + ϵ I 3 → G ⪰ 0 when G G G is singular) establishes
w T adj ( G ) w ≥ ( w ⋅ m ) 2 ⟺ adj ( G ) − m m T ⪰ 0. (64) \mathbf{w}^{T}\operatorname{adj}(G)\mathbf{w}\geq(\mathbf{w}\cdot\mathbf{m})^{2}\quad\Longleftrightarrow\quad\operatorname{adj}(G)-\mathbf{m}\mathbf{m}^{T}\succeq0. \tag*{(64)} w T adj ( G ) w ≥ ( w ⋅ m ) 2 ⟺ adj ( G ) − m m T ⪰ 0. (64)
Evaluating Eq. (64) at w = G m w=Gm w = G m and using G adj ( G ) G = ( det G ) G G\operatorname{adj}(G)G=(\det G)G G adj ( G ) G = ( det G ) G gives ( det G ) ( m T G m ) ≥ ( m T G m ) 2 (\det G)(m^{T}Gm)\geq(m^{T}Gm)^{2} ( det G ) ( m T G m ) ≥ ( m T G m ) 2 , or m T G m ≤ det G m^{T}Gm\leq\det G m T G m ≤ det G , proving β ( r ) ≤ 1 \beta(r)\leq1 β ( r ) ≤ 1 . Finally, in Eq. (27) , equality holds if and only if a ~ × b ~ ∥ G 1 / 2 m \tilde{a}\times\tilde{b}\parallel G^{1/2}m a ~ × b ~ ∥ G 1/2 m , a ~ ⊥ b ~ \tilde{a}\perp\tilde{b} a ~ ⊥ b ~ , and ∥ a ~ ∥ = ∥ b ~ ∥ \lVert\tilde{a}\rVert=\lVert\tilde{b}\rVert ∥ a ~ ∥ = ∥ b ~ ∥ ; any two such equal-norm orthogonal pairs in the two-dimensional plane ( G 1 / 2 m ) ⊥ (G^{1/2}m)^{\perp} ( G 1/2 m ) ⊥ differ only by an S O ( 2 ) \mathrm{SO}(2) SO ( 2 ) gauge rotation ( a ~ + i b ~ ↦ e − i φ ( a ~ + i b ~ ) ) (\tilde{a}+i\tilde{b}\mapsto e^{-i\varphi}(\tilde{a}+i\tilde{b})) ( a ~ + i b ~ ↦ e − i φ ( a ~ + i b ~ )) and orientation reflection ( b ~ ↦ − b ~ (\tilde{b}\mapsto-\tilde{b} ( b ~ ↦ − b ~ , i.e., L ~ e x t ↦ L ~ e x t † ) \tilde{L}_{\mathrm{ext}}\mapsto\tilde{L}_{\mathrm{ext}}^{\dagger}) L ~ ext ↦ L ~ ext † ) .
Midpoint-frame block-algebra derivation for constant-axis rotations
This appendix derives the midpoint-sinc formula (31) of Proposition 2 for general constant-axis rotations H ( Q , ϑ ) = Q 2 ( n ⋅ σ ) H(Q,\vartheta)=\frac{Q}{2}(n\cdot\sigma) H ( Q , ϑ ) = 2 Q ( n ⋅ σ ) with n = ( sin ϑ , 0 , cos ϑ ) T n=(\sin\vartheta,0,\cos\vartheta)^{T} n = ( sin ϑ , 0 , cos ϑ ) T in Section 4.2. Introduce the right-handed orthonormal frame e ∥ : = n e_{\parallel}:=n e ∥ := n , e 1 : = ( cos ϑ , 0 , − sin ϑ ) T e_{1}:=(\cos\vartheta,0,-\sin\vartheta)^{T} e 1 := ( cos ϑ , 0 , − sin ϑ ) T , e 2 : = ( 0 , 1 , 0 ) T e_{2}:=(0,1,0)^{T} e 2 := ( 0 , 1 , 0 ) T , and rotate { e 1 , e 2 } \{e_{1},e_{2}\} { e 1 , e 2 } by the midpoint angle Q / 2 Q/2 Q /2 to u 1 : = cos ( Q / 2 ) e 1 + sin ( Q / 2 ) e 2 u_{1}:=\cos(Q/2)e_{1}+\sin(Q/2)e_{2} u 1 := cos ( Q /2 ) e 1 + sin ( Q /2 ) e 2 and u 2 : = − sin ( Q / 2 ) e 1 + cos ( Q / 2 ) e 2 u_{2}:=-\sin(Q/2)e_{1}+\cos(Q/2)e_{2} u 2 := − sin ( Q /2 ) e 1 + cos ( Q /2 ) e 2 . Parameterizing time relative to the pulse midpoint via u : = s − 1 / 2 ∈ [ − 1 / 2 , 1 / 2 ] u:=s-1/2\in[-1/2,1/2] u := s − 1/2 ∈ [ − 1/2 , 1/2 ] expresses the trajectory symmetrically as
r ( u ) = cos ϑ e ∥ − sin ϑ [ cos ( Q u ) u 1 + sin ( Q u ) u 2 ] . (65) r(u)=\cos\vartheta\,e_{\parallel}-\sin\vartheta\left[\cos(Qu)u_{1}+\sin(Qu)u_{2}\right]. \tag*{(65)} r ( u ) = cos ϑ e ∥ − sin ϑ [ cos ( Q u ) u 1 + sin ( Q u ) u 2 ] . (65)
Integrating over u ∈ [ − 1 / 2 , 1 / 2 ] u\in[-1/2,1/2] u ∈ [ − 1/2 , 1/2 ] in the orthonormal basis { e ∥ , u 1 , u 2 } \{e_{\parallel},u_{1},u_{2}\} { e ∥ , u 1 , u 2 } decouples the odd sine mode u 2 u_{2} u 2 , giving m m i d = ( cos ϑ , − sinc 0 sin ϑ , 0 ) T m_{\mathrm{mid}}=(\cos\vartheta,-\operatorname{sinc}_{0}\sin\vartheta,0)^{T} m mid = ( cos ϑ , − sinc 0 sin ϑ , 0 ) T and the block-diagonal covariance deficit
G m i d = diag ( G 2 × 2 , 1 − 1 − sinc 1 2 sin 2 ϑ ) , G 2 × 2 = ( sin 2 ϑ sinc 0 sin ϑ cos ϑ sinc 0 sin ϑ cos ϑ 1 − 1 + sinc 1 2 sin 2 ϑ ) . (66) G_{\mathrm{mid}}=\operatorname{diag}\left(G_{2\times2},1-\frac{1-\operatorname{sinc}_{1}}{2}\sin^{2}\vartheta\right),\qquad
G_{2\times2}=\begin{pmatrix}
\sin^{2}\vartheta& \operatorname{sinc}_{0}\sin\vartheta\cos\vartheta\\
\operatorname{sinc}_{0}\sin\vartheta\cos\vartheta& 1-\frac{1+\operatorname{sinc}_{1}}{2}\sin^{2}\vartheta
\end{pmatrix}. \tag*{(66)} G mid = diag ( G 2 × 2 , 1 − 2 1 − sinc 1 sin 2 ϑ ) , G 2 × 2 = ( sin 2 ϑ sinc 0 sin ϑ cos ϑ sinc 0 sin ϑ cos ϑ 1 − 2 1 + sinc 1 sin 2 ϑ ) . (66)
The determinant of the 2 × 2 2\times2 2 × 2 block G 2 × 2 G_{2\times2} G 2 × 2 simplifies to
det G 2 × 2 = sin 2 ϑ ( 1 − 1 + sinc 1 2 sin 2 ϑ − sinc 0 2 cos 2 ϑ ) = 1 2 sin 2 ϑ [ 2 ( 1 − sinc 0 2 ) cos 2 ϑ + ( 1 − sinc 1 ) sin 2 ϑ ] . (67) \begin{aligned}
\det G_{2\times2}
&=\sin^{2}\vartheta\left(1-\frac{1+\operatorname{sinc}_{1}}{2}\sin^{2}\vartheta-\operatorname{sinc}_{0}^{2}\cos^{2}\vartheta\right)\\
&=\frac{1}{2}\sin^{2}\vartheta\left[2(1-\operatorname{sinc}_{0}^{2})\cos^{2}\vartheta+(1-\operatorname{sinc}_{1})\sin^{2}\vartheta\right]. \tag*{(67)}
\end{aligned} det G 2 × 2 = sin 2 ϑ ( 1 − 2 1 + sinc 1 sin 2 ϑ − sinc 0 2 cos 2 ϑ ) = 2 1 sin 2 ϑ [ 2 ( 1 − sinc 0 2 ) cos 2 ϑ + ( 1 − sinc 1 ) sin 2 ϑ ] . (67)
Thus det G m i d = det G 2 × 2 ⋅ G 33 = 1 4 sin 2 ϑ D ( ∣ Q ∣ , ϑ ) \det G_{\mathrm{mid}}=\det G_{2\times2}\cdot G_{33}=\frac{1}{4}\sin^{2}\vartheta\,D(|Q|,\vartheta) det G mid = det G 2 × 2 ⋅ G 33 = 4 1 sin 2 ϑ D ( ∣ Q ∣ , ϑ ) , while the numerator quadratic form m m i d T G m i d m m i d m_{\mathrm{mid}}^{T}G_{\mathrm{mid}}m_{\mathrm{mid}} m mid T G mid m mid (which equals m 2 T G 2 × 2 m 2 m_{2}^{T}G_{2\times2}m_{2} m 2 T G 2 × 2 m 2 ) evaluates to
m 2 T G 2 × 2 m 2 = sin 2 ϑ [ ( 1 − sinc 0 2 ) cos 2 ϑ + 1 − sinc 1 2 sinc 0 2 sin 2 ϑ ] = 1 2 sin 2 ϑ N ( ∣ Q ∣ , ϑ ) . (68) \begin{aligned}
m_{2}^{T}G_{2\times2}m_{2}
&=\sin^{2}\vartheta\left[(1-\operatorname{sinc}_{0}^{2})\cos^{2}\vartheta+\frac{1-\operatorname{sinc}_{1}}{2}\operatorname{sinc}_{0}^{2}\sin^{2}\vartheta\right]\\
&=\frac{1}{2}\sin^{2}\vartheta\,N(|Q|,\vartheta). \tag*{(68)}
\end{aligned} m 2 T G 2 × 2 m 2 = sin 2 ϑ [ ( 1 − sinc 0 2 ) cos 2 ϑ + 2 1 − sinc 1 sinc 0 2 sin 2 ϑ ] = 2 1 sin 2 ϑ N ( ∣ Q ∣ , ϑ ) . (68)
Dividing Eq. (68) by det G m i d \det G_{\mathrm{mid}} det G mid cancels 1 2 sin 2 ϑ \frac{1}{2}\sin^{2}\vartheta 2 1 sin 2 ϑ and establishes Eq. (31) .
First-order tangent cone, finite-damping Dyson bound, and exact damped-Bloch solution
This appendix establishes the Bouligand tangent-cone closure of the survival-error map, the coherent over-rotation immunity identity used in § 7.2 , and the exact finite-ε \varepsilon ε derivations for § 8.1 .
First-order Bouligand tangent cone and coherent over-rotation immunity
Proposition 3 (First-order tangent cone and coherent calibration immunity). Let Π s w a p ( Φ ) : = ( p + ( Φ ) , p − ( Φ ) ) \Pi_{\mathrm{swap}}(\Phi):=(p_{+}(\Phi),p_{-}(\Phi)) Π swap ( Φ ) := ( p + ( Φ ) , p − ( Φ )) denote the directional survival-error pair of a noisy swap channel Φ \Phi Φ . Along any regular sequence of channels Φ n = exp ( − i ad H q + δ H n + D n ) → U q \Phi_{n}=\exp(-i\,\operatorname{ad}_{H_{q}+\delta H_{n}}+D_{n})\to U_{q} Φ n = exp ( − i ad H q + δ H n + D n ) → U q whose coherent Hamiltonian calibration error satisfies ∥ δ H n ∥ ∞ = o ( ∥ D n ∥ H S 1 / 2 ) \lVert\delta H_{n}\rVert_{\infty}=o(\lVert D_{n}\rVert_{\mathrm{HS}}^{1/2}) ∥ δ H n ∥ ∞ = o (∥ D n ∥ HS 1/2 ) , determinant convergence det Φ n = e − 2 Γ n → 1 \det\Phi_{n}=e^{-2\Gamma_{n}}\to1 det Φ n = e − 2 Γ n → 1 forces τ n : = ∥ D n ∥ H S → 0 \tau_{n}:=\lVert D_{n}\rVert_{\mathrm{HS}}\to0 τ n := ∥ D n ∥ HS → 0 , and the first-order Bouligand tangent cone of Π s w a p \Pi_{\mathrm{swap}} Π swap at U q U_{q} U q is the closed sector
C q = { ( x + , x − ) ∈ R + 2 ∣ κ q x + ≤ x − ≤ κ q − 1 x + } . (69) \mathcal{C}_{q}=\left\{(x_{+},x_{-})\in\mathbb{R}_{+}^{2}\mid\kappa_{q}x_{+}\leq x_{-}\leq\kappa_{q}^{-1}x_{+}\right\}. \tag*{(69)} C q = { ( x + , x − ) ∈ R + 2 ∣ κ q x + ≤ x − ≤ κ q − 1 x + } . (69)
Proof. With L 0 = − i ad H q \mathcal{L}_0=-i\,\operatorname{ad}_{H_q} L 0 = − i ad H q and Δ n = − i ad δ H n + D n \Delta_n=-i\,\operatorname{ad}_{\delta H_n}+D_n Δ n = − i ad δ H n + D n , Duhamel expansion gives
Φ n = U q ( 1 ) + ∫ 0 1 U q ( 1 − s ) Δ n U q ( s ) d s + R 2 ( Δ n ) , (70) \Phi_n=U_q(1)+\int_0^1 U_q(1-s)\,\Delta_n U_q(s)\,ds+R_2(\Delta_n), \tag*{(70)} Φ n = U q ( 1 ) + ∫ 0 1 U q ( 1 − s ) Δ n U q ( s ) d s + R 2 ( Δ n ) , (70)
where the second-order remainder obeys
∥ R 2 ( Δ n ) ∥ o p ≤ ( ∥ ad δ H n ∥ o p 2 + ∥ D n ∥ o p 2 ) exp ( ∥ Δ n ∥ o p ) , (71) \lVert R_2(\Delta_n)\rVert_{\mathrm{op}}\leq\left(\lVert\operatorname{ad}_{\delta H_n}\rVert_{\mathrm{op}}^2+\lVert D_n\rVert_{\mathrm{op}}^2\right)\exp\left(\lVert\Delta_n\rVert_{\mathrm{op}}\right), \tag*{(71)} ∥ R 2 ( Δ n ) ∥ op ≤ ( ∥ ad δ H n ∥ op 2 + ∥ D n ∥ op 2 ) exp ( ∥ Δ n ∥ op ) , (71)
so that under ∥ δ H n ∥ o p = o ( τ n 1 / 2 ) \lVert\delta H_n\rVert_{\mathrm{op}}=o(\tau_n^{1/2}) ∥ δ H n ∥ op = o ( τ n 1/2 ) ,
∥ R 2 ( Δ n ) ∥ o p τ n ≤ ( 4 ∥ δ H n ∥ o p 2 τ n + τ n ) e ∥ Δ n ∥ o p ⟶ n → ∞ 0. (72) \frac{\lVert R_2(\Delta_n)\rVert_{\mathrm{op}}}{\tau_n}\leq\left(\frac{4\lVert\delta H_n\rVert_{\mathrm{op}}^2}{\tau_n}+\tau_n\right)e^{\lVert\Delta_n\rVert_{\mathrm{op}}}\mathop{\longrightarrow}_{n\to\infty}0. \tag*{(72)} τ n ∥ R 2 ( Δ n ) ∥ op ≤ ( τ n 4 ∥ δ H n ∥ op 2 + τ n ) e ∥ Δ n ∥ op ⟶ n → ∞ 0. (72)
Because the 3 × 3 3\times3 3 × 3 Bloch generator A n : = [ − i ad δ H n ] P T M , 3 × 3 A_n:=[-i\operatorname{ad}_{\delta H_n}]_{\mathrm{PTM},3\times3} A n := [ − i ad δ H n ] PTM , 3 × 3 of any coherent Hamiltonian perturbation δ H n \delta H_n δ H n is real skew-symmetric (A n T = − A n A_n^{\mathsf{T}}=-A_n A n T = − A n ), its quadratic form with the symmetric state trajectory r q ( s ) r_q(s) r q ( s ) vanishes identically at every point along the trajectory:
− 1 2 ∫ 0 1 r q ( s ) T A n r q ( s ) d s = 0. (73) -\frac{1}{2}\int_0^1 r_q(s)^{\mathsf{T}}A_n r_q(s)\,ds=0. \tag*{(73)} − 2 1 ∫ 0 1 r q ( s ) T A n r q ( s ) d s = 0. (73)
Coherent over-rotation and axis-tilt errors therefore contribute zero first-order survival error (δ p ± c o h = O ( ∥ δ H n ∥ 2 ) \delta p_{\pm}^{\mathrm{coh}}=\mathcal{O}(\lVert\delta H_n\rVert^2) δ p ± coh = O (∥ δ H n ∥ 2 ) ), leaving the normalized limit ray
Π s w a p ( Φ n ) τ n = ( χ + ( D ^ n ; H q ) , χ − ( D ^ n ; H q ) ) + o ( 1 ) ∈ C q , (74) \frac{\Pi_{\mathrm{swap}}(\Phi_n)}{\tau_n}=\left(\chi_+(\widehat{D}_n;H_q),\chi_-(\widehat{D}_n;H_q)\right)+o(1)\in C_q, \tag*{(74)} τ n Π swap ( Φ n ) = ( χ + ( D n ; H q ) , χ − ( D n ; H q ) ) + o ( 1 ) ∈ C q , (74)
where D ^ n = D n / τ n \widehat{D}_n=D_n/\tau_n D n = D n / τ n , establishing Eq. (69) .
Finite-damping Dyson bound, exact Liouvillian exponentiation, and boundary jets
Proposition 4 (Finite-damping Dyson ratio bound and exact damped-Bloch integration). *For any constant-amplitude swap H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 ) and any time-independent Markovian bath D D D with η r e m ( ε ) = 8 ε Γ e 2 ε Γ < 1 \eta_{\mathrm{rem}}(\varepsilon)=8\varepsilon\Gamma e^{2\varepsilon\Gamma}<1 η rem ( ε ) = 8 ε Γ e 2 ε Γ < 1 , the exact finite-ε \varepsilon ε survival errors a ( ε ) = max ( p + ( ε ) , p − ( ε ) ) a(\varepsilon)=\max(p_+(\varepsilon),p_-(\varepsilon)) a ( ε ) = max ( p + ( ε ) , p − ( ε )) and b ( ε ) = min ( p + ( ε ) , p − ( ε ) ) b(\varepsilon)=\min(p_+(\varepsilon),p_-(\varepsilon)) b ( ε ) = min ( p + ( ε ) , p − ( ε )) satisfy Eq. (48) . Furthermore, for L + = ( σ x + i 2 σ z ) / 6 L_+=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 , the exact all-orders channel exp ( L ε ) \exp(\mathcal{L}_\varepsilon) exp ( L ε ) is given by Eqs. (49) –(51) .
Proof. To prove the finite-ε \varepsilon ε Dyson guarantee Eq. (48) , observe that G q = diag ( 1 , 1 2 , 1 2 ) G_q=\operatorname{diag}(1,\frac{1}{2},\frac{1}{2}) G q = diag ( 1 , 2 1 , 2 1 ) has minimum eigenvalue λ min ( G q ) = 1 2 \lambda_{\min}(G_q)=\frac{1}{2} λ m i n ( G q ) = 2 1 and S ⪰ 0 S\succeq0 S ⪰ 0 has Tr ( S ) = Γ \operatorname{Tr}(S)=\Gamma Tr ( S ) = Γ , so Eq. (22) yields the universal lower bound χ max : = max ( χ + , χ − ) ≥ 1 2 ( χ + + χ − ) = 1 2 Tr ( S G q ) ≥ 1 2 λ min ( G q ) Tr ( S ) = 1 4 Γ \chi_{\max}:=\max(\chi_+,\chi_-)\geq\frac{1}{2}(\chi_++\chi_-)=\frac{1}{2}\operatorname{Tr}(SG_q)\geq\frac{1}{2}\lambda_{\min}(G_q)\operatorname{Tr}(S)=\frac{1}{4}\Gamma χ m a x := max ( χ + , χ − ) ≥ 2 1 ( χ + + χ − ) = 2 1 Tr ( S G q ) ≥ 2 1 λ m i n ( G q ) Tr ( S ) = 4 1 Γ (saturated by pure dephasing L ϕ = σ z / 2 L_\phi=\sigma_z/\sqrt{2} L ϕ = σ z / 2 ). Combining χ min : = min ( χ + , χ − ) ≥ κ q χ max \chi_{\min}:=\min(\chi_+,\chi_-)\geq\kappa_q\chi_{\max} χ m i n := min ( χ + , χ − ) ≥ κ q χ m a x with the Dyson remainder bound R 2 ( ε ) ≤ 2 ε 2 Γ 2 e 2 ε Γ R_2(\varepsilon)\leq2\varepsilon^2\Gamma^2e^{2\varepsilon\Gamma} R 2 ( ε ) ≤ 2 ε 2 Γ 2 e 2 ε Γ of Eq. (7) gives b ( ε ) ≥ ε χ min − R 2 ( ε ) ≥ κ q ε χ max − R 2 ( ε ) b(\varepsilon)\geq\varepsilon\chi_{\min}-R_2(\varepsilon)\geq\kappa_q\varepsilon\chi_{\max}-R_2(\varepsilon) b ( ε ) ≥ ε χ m i n − R 2 ( ε ) ≥ κ q ε χ m a x − R 2 ( ε ) and a ( ε ) ≤ ε χ max + R 2 ( ε ) a(\varepsilon)\leq\varepsilon\chi_{\max}+R_2(\varepsilon) a ( ε ) ≤ ε χ m a x + R 2 ( ε ) . Dividing numerator and denominator by ε χ max ≥ 1 4 ε Γ \varepsilon\chi_{\max}\geq\frac{1}{4}\varepsilon\Gamma ε χ m a x ≥ 4 1 ε Γ implies R 2 ( ε ) / ( ε χ max ) ≤ 8 ε Γ e 2 ε Γ = η r e m ( ε ) R_2(\varepsilon)/(\varepsilon\chi_{\max})\leq8\varepsilon\Gamma e^{2\varepsilon\Gamma}=\eta_{\mathrm{rem}}(\varepsilon) R 2 ( ε ) / ( ε χ m a x ) ≤ 8 ε Γ e 2 ε Γ = η rem ( ε ) , and therefore
b ( ε ) a ( ε ) ≥ κ q − η r e m ( ε ) 1 + η r e m ( ε ) = κ q ( 1 − ( 1 + κ q ) η r e m ( ε ) κ q ( 1 + η r e m ( ε ) ) ) ≥ κ q ( 1 − C q ε Γ e 2 ε Γ ) , (75) \frac{b(\varepsilon)}{a(\varepsilon)}\geq\frac{\kappa_q-\eta_{\mathrm{rem}}(\varepsilon)}{1+\eta_{\mathrm{rem}}(\varepsilon)}
=\kappa_q\left(1-\frac{(1+\kappa_q)\eta_{\mathrm{rem}}(\varepsilon)}{\kappa_q(1+\eta_{\mathrm{rem}}(\varepsilon))}\right)
\geq\kappa_q\left(1-C_q\varepsilon\Gamma e^{2\varepsilon\Gamma}\right), \tag*{(75)} a ( ε ) b ( ε ) ≥ 1 + η rem ( ε ) κ q − η rem ( ε ) = κ q ( 1 − κ q ( 1 + η rem ( ε )) ( 1 + κ q ) η rem ( ε ) ) ≥ κ q ( 1 − C q ε Γ e 2 ε Γ ) , (75)
where C q = 8 ( 1 + κ q ) / κ q = 16 π ∣ q ∣ / ( π ∣ q ∣ − 2 2 ) C_q=8(1+\kappa_q)/\kappa_q=16\pi\lvert q\rvert/(\pi\lvert q\rvert-2\sqrt{2}) C q = 8 ( 1 + κ q ) / κ q = 16 π ∣ q ∣ / ( π ∣ q ∣ − 2 2 ) , proving Eq. (48) (for general jump operators with c α ≠ 0 c_\alpha\neq0 c α = 0 , replacing ε Γ e 2 ε Γ \varepsilon\Gamma e^{2\varepsilon\Gamma} ε Γ e 2 ε Γ by Γ f u l l Γ ε e 2 ε Γ f u l l \frac{\Gamma_{\mathrm{full}}}{\Gamma}\varepsilon e^{2\varepsilon\Gamma_{\mathrm{full}}} Γ Γ full ε e 2 ε Γ full in η r e m ( ε ) \eta_{\mathrm{rem}}(\varepsilon) η rem ( ε ) yields the corresponding bound).
To derive the exact all-orders damped-Bloch formulas Eqs. (49) –(51) under L + = ( σ x + i 2 σ z ) / 6 L_+=(\sigma_x+i\sqrt{2}\sigma_z)/\sqrt{6} L + = ( σ x + i 2 σ z ) / 6 (Γ = 1 \Gamma=1 Γ = 1 ) and H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (Q = q π Q=q\pi Q = q π ), we write the 4 × 4 4\times4 4 × 4 PTM generator of L ε = − i ad H q + ε D [ L + ] \mathcal{L}_\varepsilon=-i\operatorname{ad}_{H_q}+\varepsilon\mathcal{D}[L_+] L ε = − i ad H q + ε D [ L + ] in the basis { B 0 , B 1 , B 2 , B 3 } \{B_0,B_1,B_2,B_3\} { B 0 , B 1 , B 2 , B 3 } :
[ L ε ] P T M = ( 0 0 0 0 0 − 2 ε 3 0 0 − 2 2 3 ε 0 − ε Q 0 0 − Q − ε 3 ) . (76) [\mathcal{L}_\varepsilon]_{\mathrm{PTM}}=
\begin{pmatrix}
0 & 0 & 0 & 0\\
0 & -\frac{2\varepsilon}{3} & 0 & 0\\
-\frac{2\sqrt{2}}{3}\varepsilon& 0 & -\varepsilon& Q\\
0 & 0 & -Q & -\frac{\varepsilon}{3}
\end{pmatrix}. \tag*{(76)} [ L ε ] PTM = 0 0 − 3 2 2 ε 0 0 − 3 2 ε 0 0 0 0 − ε − Q 0 0 Q − 3 ε . (76)
Thus [ exp ( L ε ) ] 11 = e − 2 ε / 3 [\exp(\mathcal{L}_{\varepsilon})]_{11}=e^{-2\varepsilon/3} [ exp ( L ε ) ] 11 = e − 2 ε /3 , while the coupled ( B 2 , B 3 ) (B_2,B_3) ( B 2 , B 3 ) block
K y z : = ( − ε Q − Q − ε / 3 ) = − 2 ε 3 I 2 + ( − ε / 3 Q − Q ε / 3 ) (77) K_{yz}:=\begin{pmatrix}-\varepsilon& Q \\ -Q & -\varepsilon/3\end{pmatrix}=-\frac{2\varepsilon}{3}I_2+\begin{pmatrix}-\varepsilon/3 & Q \\ -Q & \varepsilon/3\end{pmatrix} \tag*{(77)} K y z := ( − ε − Q Q − ε /3 ) = − 3 2 ε I 2 + ( − ε /3 − Q Q ε /3 ) (77)
exponentiates in closed form via Euler’s identity with damped frequency Ω ε , q = Q 2 − ε 2 / 9 \Omega_{\varepsilon,q}=\sqrt{Q^2-\varepsilon^2/9} Ω ε , q = Q 2 − ε 2 /9 :
e s K y z = e − 2 ε s / 3 [ cos ( Ω ε , q s ) I 2 + sin ( Ω ε , q s ) Ω ε , q ( − ε / 3 Q − Q ε / 3 ) ] . (78) e^{sK_{yz}}=e^{-2\varepsilon s/3}\left[\cos(\Omega_{\varepsilon,q}s)I_2+\frac{\sin(\Omega_{\varepsilon,q}s)}{\Omega_{\varepsilon,q}}\begin{pmatrix}-\varepsilon/3 & Q \\ -Q & \varepsilon/3\end{pmatrix}\right]. \tag*{(78)} e s K y z = e − 2 ε s /3 [ cos ( Ω ε , q s ) I 2 + Ω ε , q sin ( Ω ε , q s ) ( − ε /3 − Q Q ε /3 ) ] . (78)
Evaluating the diagonal entries at s = 1 s=1 s = 1 yields Eqs. (49) –(50) , and integrating the affine drift column via K y z − 1 ( exp ( K y z ) − I 2 ) ( − 2 2 ε / 3 , 0 ) T K_{yz}^{-1}(\exp(K_{yz})-I_2)(-2\sqrt{2}\varepsilon/3,0)^T K y z − 1 ( exp ( K y z ) − I 2 ) ( − 2 2 ε /3 , 0 ) T yields Eq. (51) .
In the high-winding limit ∣ q ∣ → ∞ |q|\to\infty ∣ q ∣ → ∞ at fixed scaled exposure u : = ε / ( ∣ q ∣ π ) > 0 u:=\varepsilon/(|q|\pi)>0 u := ε / ( ∣ q ∣ π ) > 0 , Eqs. (50) –(51) give lim ∣ q ∣ → ∞ sup u > 0 p + ( u ∣ q ∣ π , q ) / p − ( u ∣ q ∣ π , q ) = ( 3 + 2 ) 2 = 5 + 2 6 ≈ 9.898979 \lim_{|q|\to\infty}\sup_{u>0}p_+(u|q|\pi,q)/p_-(u|q|\pi,q)=(\sqrt{3}+\sqrt{2})^2=5+2\sqrt{6}\approx9.898979 lim ∣ q ∣ → ∞ sup u > 0 p + ( u ∣ q ∣ π , q ) / p − ( u ∣ q ∣ π , q ) = ( 3 + 2 ) 2 = 5 + 2 6 ≈ 9.898979 . Under a retimed drive H ε ( λ ) = − π 2 ( 1 + λ ε ) σ x H_{\varepsilon}(\lambda)=-\frac{\pi}{2}(1+\lambda\varepsilon)\sigma_x H ε ( λ ) = − 2 π ( 1 + λ ε ) σ x , the boundary expansion b ( ε ) = κ ⋆ a ( ε ) + c L + ( λ ) a ( ε ) 2 + O ( ε 3 ) b(\varepsilon)=\kappa_\star a(\varepsilon)+c_{L_+}(\lambda)a(\varepsilon)^2+O(\varepsilon^3) b ( ε ) = κ ⋆ a ( ε ) + c L + ( λ ) a ( ε ) 2 + O ( ε 3 ) for L + L_+ L + has second-order curvature c L + ( λ ) = 2 [ ( 3 π 2 λ + 1 ) 2 − 1 ] / ( π + 2 2 ) c_{L_+}(\lambda)=\sqrt{2}[(3\pi^2\lambda+1)^2-1]/(\pi+2\sqrt{2}) c L + ( λ ) = 2 [( 3 π 2 λ + 1 ) 2 − 1 ] / ( π + 2 2 ) , which vanishes at λ = 0 \lambda=0 λ = 0 and leaves the third-order term (53) .
Finally, for the rank-1 nilpotent jump L ⋆ = ∣ ψ ( π / 8 ) ⟩ ⟨ ψ ( 7 π / 8 ) ∣ = − 2 4 I 2 − i 2 4 σ x − 1 2 σ z L_\star=|\psi(\pi/8)\rangle\langle\psi(7\pi/8)|=-\frac{\sqrt{2}}{4}I_2-\frac{i\sqrt{2}}{4}\sigma_x-\frac{1}{2}\sigma_z L ⋆ = ∣ ψ ( π /8 )⟩ ⟨ ψ ( 7 π /8 ) ∣ = − 4 2 I 2 − 4 i 2 σ x − 2 1 σ z (c = − 2 4 c=-\frac{\sqrt{2}}{4} c = − 4 2 , Γ = 3 4 \Gamma=\frac{3}{4} Γ = 4 3 ) with retimed drive H ε ( λ ) = π 2 ( 1 + λ ε ) σ x H_{\varepsilon}(\lambda)=\frac{\pi}{2}(1+\lambda\varepsilon)\sigma_x H ε ( λ ) = 2 π ( 1 + λ ε ) σ x , the Lamb shift Δ H L a m b = − 1 8 σ x \Delta H_{\mathrm{Lamb}}=-\frac{1}{8}\sigma_x Δ H Lamb = − 8 1 σ x gives the PTM generator (with Ω ~ λ : = π ( 1 + λ ε ) − ε / 4 \widetilde{\Omega}_{\lambda}:=\pi(1+\lambda\varepsilon)-\varepsilon/4 Ω λ := π ( 1 + λ ε ) − ε /4 )
[ L ε , L ⋆ ( λ ) ] P T M = ( 0 0 0 0 0 − ε / 2 0 0 2 2 ε 0 − 3 ε / 4 − Ω ~ λ 0 0 Ω ~ λ − ε / 4 ) . (79) [\mathcal{L}_{\varepsilon,L_\star(\lambda)}]_{\mathrm{PTM}}=
\begin{pmatrix}
0 & 0 & 0 & 0\\
0 & -\varepsilon/2 & 0 & 0\\
\frac{\sqrt{2}}{2}\varepsilon& 0 & -3\varepsilon/4 & -\widetilde{\Omega}_{\lambda}\\
0 & 0 & \widetilde{\Omega}_{\lambda} & -\varepsilon/4
\end{pmatrix}. \tag*{(79)} [ L ε , L ⋆ ( λ ) ] PTM = 0 0 2 2 ε 0 0 − ε /2 0 0 0 0 − 3 ε /4 Ω λ 0 0 − Ω λ − ε /4 . (79)
Expanding the ( B 2 , B 3 ) (B_2,B_3) ( B 2 , B 3 ) block in powers of ε \varepsilon ε (where a ( ε ) = A ⋆ ε + a 2 ( λ ) ε 2 + O ( ε 3 ) a(\varepsilon)=A_\star\varepsilon+a_2(\lambda)\varepsilon^2+O(\varepsilon^3) a ( ε ) = A ⋆ ε + a 2 ( λ ) ε 2 + O ( ε 3 ) with A ⋆ = ( π + 2 2 ) / ( 4 π ) A_\star=(\pi+2\sqrt{2})/(4\pi) A ⋆ = ( π + 2 2 ) / ( 4 π ) and P ( λ ) : = 16 π 4 λ 2 − 8 π 2 ( π + 1 ) λ − 3 π 2 + 2 π P(\lambda):=16\pi^4\lambda^2-8\pi^2(\pi+1)\lambda-3\pi^2+2\pi P ( λ ) := 16 π 4 λ 2 − 8 π 2 ( π + 1 ) λ − 3 π 2 + 2 π ) gives
a 2 ( λ ) = P ( λ ) − 8 2 ( 4 π λ + π − 1 ) 64 π 2 , (80) a_2(\lambda)=\frac{P(\lambda)-8\sqrt{2}(4\pi\lambda+\pi-1)}{64\pi^2}, \tag*{(80)} a 2 ( λ ) = 64 π 2 P ( λ ) − 8 2 ( 4 π λ + π − 1 ) , (80)
b 2 ( λ ) = P ( λ ) + 8 2 ( 4 π λ + π − 1 ) 64 π 2 , (81) b_2(\lambda)=\frac{P(\lambda)+8\sqrt{2}(4\pi\lambda+\pi-1)}{64\pi^2}, \tag*{(81)} b 2 ( λ ) = 64 π 2 P ( λ ) + 8 2 ( 4 π λ + π − 1 ) , (81)
yielding the quadratic curvature c ( λ ) = [ b 2 ( λ ) − κ ⋆ a 2 ( λ ) ] / A ⋆ 2 c(\lambda)=[b_2(\lambda)-\kappa_\star a_2(\lambda)]/A_\star^2 c ( λ ) = [ b 2 ( λ ) − κ ⋆ a 2 ( λ )] / A ⋆ 2 :
c ( λ ) = 2 [ ( 4 π 2 λ − ( π − 1 ) ) 2 − 1 ] ( π + 2 2 ) 3 , c min = − 2 ( π + 2 2 ) 3 ≈ − 0.006646418567 , (82) c(\lambda)=\frac{\sqrt{2}[(4\pi^2\lambda-(\pi-1))^2-1]}{(\pi+2\sqrt{2})^3},\qquad c_{\min}=-\frac{\sqrt{2}}{(\pi+2\sqrt{2})^3}\approx-0.006646418567, \tag*{(82)} c ( λ ) = ( π + 2 2 ) 3 2 [( 4 π 2 λ − ( π − 1 ) ) 2 − 1 ] , c m i n = − ( π + 2 2 ) 3 2 ≈ − 0.006646418567 , (82)
which attains its minimum at λ ⋆ = ( π − 1 ) / ( 4 π 2 ) ≈ 0.054247175635 \lambda_\star=(\pi-1)/(4\pi^2)\approx0.054247175635 λ ⋆ = ( π − 1 ) / ( 4 π 2 ) ≈ 0.054247175635 (Figure 5 (a)), where the exact second- and third-order jet coefficients (a 2 ⋆ ≈ − 0.10831902 a_2^\star\approx-0.10831902 a 2 ⋆ ≈ − 0.10831902 , b 2 ⋆ ≈ − 0.00718212 b_2^\star\approx-0.00718212 b 2 ⋆ ≈ − 0.00718212 , c 3 ⋆ ≈ + 0.006344064267 c_3^\star\approx+0.006344064267 c 3 ⋆ ≈ + 0.006344064267 ) are
a 2 ⋆ = − 4 π 3 + 8 2 π 2 − 3 π − 8 2 64 π 3 , a_2^\star=-\frac{4\pi^3+8\sqrt{2}\pi^2-3\pi-8\sqrt{2}}{64\pi^3}, a 2 ⋆ = − 64 π 3 4 π 3 + 8 2 π 2 − 3 π − 8 2 ,
b 2 ⋆ = − 4 π 3 − 8 2 π 2 − 3 π + 8 2 64 π 3 , (83) b_2^\star=-\frac{4\pi^3-8\sqrt{2}\pi^2-3\pi+8\sqrt{2}}{64\pi^3}, \tag*{(83)} b 2 ⋆ = − 64 π 3 4 π 3 − 8 2 π 2 − 3 π + 8 2 , (83)
c 3 ⋆ = − 8 2 π 4 + 32 π 3 − 90 2 π 2 − 360 π + 3 2 6 ( π + 2 2 ) 5 . c_3^\star=-\frac{8\sqrt{2}\pi^4+32\pi^3-90\sqrt{2}\pi^2-360\pi+3\sqrt{2}}{6(\pi+2\sqrt{2})^5}. c 3 ⋆ = − 6 ( π + 2 2 ) 5 8 2 π 4 + 32 π 3 − 90 2 π 2 − 360 π + 3 2 .
Moments and selectivity scaling of non-autonomous wait-flip-wait schedules
This appendix derives the non-autonomous pulse-dwell moments and selectivity scaling (54) presented in Section 7.2. For a wait-flip-wait schedule idling at ∣ 0 ⟩ |0\rangle ∣0 ⟩ for duration w + = 1 + η w 2 ( 1 − δ ) w_+=\frac{1+\eta_w}{2}(1-\delta) w + = 2 1 + η w ( 1 − δ ) , flipping at rate − π 2 δ σ x -\frac{\pi}{2\delta}\sigma_x − 2 δ π σ x over active duration δ ∈ ( 0 , 1 ] \delta\in(0,1] δ ∈ ( 0 , 1 ] , and idling at ∣ 1 ⟩ |1\rangle ∣1 ⟩ for w − = 1 − η w 2 ( 1 − δ ) w_-=\frac{1-\eta_w}{2}(1-\delta) w − = 2 1 − η w ( 1 − δ ) with wait asymmetry η w ∈ [ − 1 , 1 ] \eta_w \in[-1,1] η w ∈ [ − 1 , 1 ] , summing I 3 − r ( s ) r ( s ) T \mathbb{I}_3-\mathbf{r}(s)\mathbf{r}(s)^T I 3 − r ( s ) r ( s ) T over the two pole-idle intervals of total length 1 − δ 1-\delta 1 − δ (where I 3 − e z e z T = diag ( 1 , 1 , 0 ) \mathbb{I}_3-\mathbf{e}_z\mathbf{e}_z^T=\operatorname{diag}(1,1,0) I 3 − e z e z T = diag ( 1 , 1 , 0 ) ) and the active flip interval of length δ \delta δ yields
m ( δ , η w ) = ( 0 , 2 δ π , η w ( 1 − δ ) ) T , G ( δ ) = diag ( 1 , 1 − δ 2 , δ 2 ) , det G ( δ ) = δ 2 ( 1 − δ 2 ) . (84) \mathbf{m}(\delta,\eta_w)=\left(0,\frac{2\delta}{\pi},\eta_w(1-\delta)\right)^T,\qquad G(\delta)=\operatorname{diag}\left(1,1-\frac{\delta}{2},\frac{\delta}{2}\right),\qquad\det G(\delta)=\frac{\delta}{2}\left(1-\frac{\delta}{2}\right). \tag*{(84)} m ( δ , η w ) = ( 0 , π 2 δ , η w ( 1 − δ ) ) T , G ( δ ) = diag ( 1 , 1 − 2 δ , 2 δ ) , det G ( δ ) = 2 δ ( 1 − 2 δ ) . (84)
Evaluating the quadratic form
m ( δ , η w ) T G ( δ ) m ( δ , η w ) = 4 δ 2 π 2 ( 1 − δ 2 ) + δ 2 η w 2 ( 1 − δ ) 2 (85) \mathbf{m}(\delta,\eta_w)^T G(\delta)\mathbf{m}(\delta,\eta_w)=\frac{4\delta^2}{\pi^2}\left(1-\frac{\delta}{2}\right)+\frac{\delta}{2}\eta_w^2(1-\delta)^2 \tag*{(85)} m ( δ , η w ) T G ( δ ) m ( δ , η w ) = π 2 4 δ 2 ( 1 − 2 δ ) + 2 δ η w 2 ( 1 − δ ) 2 (85)
and dividing by det G ( δ ) = δ 2 ( 1 − δ 2 ) \det G(\delta)=\frac{\delta}{2}\left(1-\frac{\delta}{2}\right) det G ( δ ) = 2 δ ( 1 − 2 δ ) yields Eq. (54) , from which the one-sided wait-then-flip (η w = ± 1 \eta_w=\pm1 η w = ± 1 ) and symmetric wait-flip-wait (η w = 0 \eta_w=0 η w = 0 ) selectivity bounds follow:
κ d w e l l ( δ ) = 1 − β ( δ , ± 1 ) 1 + β ( δ , ± 1 ) = 3 π 2 − 16 8 π 2 δ + O ( δ 2 ) , (86) \kappa_{\mathrm{dwell}}(\delta)=\frac{1-\beta(\delta,\pm1)}{1+\beta(\delta,\pm1)}=\frac{3\pi^2-16}{8\pi^2}\delta+O(\delta^2), \tag*{(86)} κ dwell ( δ ) = 1 + β ( δ , ± 1 ) 1 − β ( δ , ± 1 ) = 8 π 2 3 π 2 − 16 δ + O ( δ 2 ) , (86)
κ s y m ( δ ) = π − 2 2 δ π + 2 2 δ = 1 − 4 2 π δ + O ( δ ) . (87) \kappa_{\mathrm{sym}}(\delta)=\frac{\pi-2\sqrt{2}\sqrt{\delta}}{\pi+2\sqrt{2}\sqrt{\delta}}=1-\frac{4\sqrt{2}}{\pi}\sqrt{\delta}+O(\delta). \tag*{(87)} κ sym ( δ ) = π + 2 2 δ π − 2 2 δ = 1 − π 4 2 δ + O ( δ ) . (87)
Colored thermal baths, Redfield spectral parity separation, and multilevel qutrit cascade dynamics
This appendix provides the complete derivations of the colored-bath spectral parity selection rule, the calibration-free quantum thermometer (55) of § 8.3 , and the all-orders single-qubit chiral fidelity invariance versus the multilevel qutrit (d = 3 d=3 d = 3 ) cascade shelving split (56) of § 8.4 .
Qubit chiral fidelity invariance versus qutrit cascade shelving anomaly
Theorem 3 (Qubit chiral fidelity invariance versus d ≥ 3 d \ge3 d ≥ 3 chiral anomaly and η c r i t \eta_{\mathrm{crit}} η crit ).
All-orders single-qubit chiral fidelity invariance (d = 2 d=2 d = 2 ): For every piecewise-continuous control Hamiltonian H ( s ) H(s) H ( s ) , every time-independent Markovian bath D [ { L ~ α } ] \mathcal{D}[\{\widetilde{L}_{\alpha}\}] D [{ L α }] , and all ϵ ≥ 0 \epsilon\ge0 ϵ ≥ 0 , replacing every traceless jump operator by its Hermitian conjugate (L ~ α ↦ L ~ α † \widetilde{L}_{\alpha}\mapsto\widetilde{L}_{\alpha}^{\dagger} L α ↦ L α † ) leaves the entanglement fidelity identically invariant:
F e ( Φ ϵ [ { L ~ α } ] , U 0 ( 1 ) ) ≡ F e ( Φ ϵ [ { L ~ α † } ] , U 0 ( 1 ) ) ( ∀ ϵ ≥ 0 ) . (88) F_e\left(\Phi_{\epsilon}[\{\widetilde{L}_{\alpha}\}],U_0(1)\right)\equiv F_e\left(\Phi_{\epsilon}[\{\widetilde{L}_{\alpha}^{\dagger}\}],U_0(1)\right)\qquad(\forall\ \epsilon\ge0). \tag*{(88)} F e ( Φ ϵ [{ L α }] , U 0 ( 1 ) ) ≡ F e ( Φ ϵ [{ L α † }] , U 0 ( 1 ) ) ( ∀ ϵ ≥ 0 ) . (88) Qutrit (d = 3 d=3 d = 3 ) cascade shelving fidelities and critical ratio η c r i t \eta_{\mathrm{crit}} η crit : On H = span { ∣ 0 ⟩ , ∣ 1 ⟩ , ∣ 2 ⟩ } \mathcal{H}=\operatorname{span}\{|0\rangle,|1\rangle,|2\rangle\} H = span { ∣0 ⟩ , ∣1 ⟩ , ∣2 ⟩} under cascade decay ϵ D = ϵ D [ ∣ 0 ⟩ ⟨ 1 ∣ ] + ϵ η D [ ∣ 1 ⟩ ⟨ 2 ∣ ] \epsilon\mathcal{D}=\epsilon\mathcal{D}[|0\rangle\langle1|]+\epsilon\eta\mathcal{D}[|1\rangle\langle2|] ϵ D = ϵ D [ ∣0 ⟩ ⟨ 1∣ ] + ϵη D [ ∣1 ⟩ ⟨ 2∣ ] (η ≥ 0 \eta\ge0 η ≥ 0 ), the endpoint-matched 2 π 2\pi 2 π shelving rotations H 02 ( 2 π ) = π ( ∣ 0 ⟩ ⟨ 2 ∣ + ∣ 2 ⟩ ⟨ 0 ∣ ) H_{02}(2\pi)=\pi(|0\rangle\langle2|+|2\rangle\langle0|) H 02 ( 2 π ) = π ( ∣0 ⟩ ⟨ 2∣ + ∣2 ⟩ ⟨ 0∣ ) and H 12 ( 2 π ) = π ( ∣ 1 ⟩ ⟨ 2 ∣ + ∣ 2 ⟩ ⟨ 1 ∣ ) H_{12}(2\pi)=\pi(|1\rangle\langle2|+|2\rangle\langle1|) H 12 ( 2 π ) = π ( ∣1 ⟩ ⟨ 2∣ + ∣2 ⟩ ⟨ 1∣ ) share the same first-order entanglement infidelity 1 − F e = 1 + η 3 ϵ + O ( ϵ 2 ) 1-F_e=\frac{1+\eta}{3}\epsilon+O(\epsilon^2) 1 − F e = 3 1 + η ϵ + O ( ϵ 2 ) , while their second-order entanglement-fidelity difference is
F e ( ϵ ; H 02 ( 2 π ) , η ) − F e ( ϵ ; H 12 ( 2 π ) , η ) = 6 − 4 η − 3 η 2 144 ϵ 2 + O ( ϵ 3 ) , (89) F_e(\epsilon;H_{02}(2\pi),\eta)-F_e(\epsilon;H_{12}(2\pi),\eta)=\frac{6-4\eta-3\eta^2}{144}\epsilon^2+O(\epsilon^3), \tag*{(89)} F e ( ϵ ; H 02 ( 2 π ) , η ) − F e ( ϵ ; H 12 ( 2 π ) , η ) = 144 6 − 4 η − 3 η 2 ϵ 2 + O ( ϵ 3 ) , (89) which vanishes and reverses sign at the irrational cascade ratio
η c r i t = 22 − 2 3 ≈ 0.896805253274. (90) \eta_{\mathrm{crit}}=\frac{\sqrt{22}-2}{3}\approx0.896805253274. \tag*{(90)} η crit = 3 22 − 2 ≈ 0.896805253274. (90)
Proof. In the 4 × 4 4\times4 4 × 4 PTM representation (60) of Appendix § 14 , trace preservation ([ L ϵ ( s ) ] 0 j = 0 [\mathcal{L}_{\epsilon}(s)]_{0j}=0 [ L ϵ ( s ) ] 0 j = 0 ) makes Φ ϵ ( 1 ) \Phi_{\epsilon}(1) Φ ϵ ( 1 ) lower-block-triangular with diagonal blocks 1 1 1 and M ϵ ( 1 ) = T exp ∫ 0 1 ( [ L 0 ( s ) ] 3 × 3 + ϵ [ S − Tr ( S ) I 3 ] ) d s M_{\epsilon}(1)=\mathcal{T}\exp\int_0^1([\mathcal{L}_0(s)]_{3\times3}+\epsilon[S-\operatorname{Tr}(S)\mathbb{I}_3])\,\mathrm{d}s M ϵ ( 1 ) = T exp ∫ 0 1 ([ L 0 ( s ) ] 3 × 3 + ϵ [ S − Tr ( S ) I 3 ]) d s . Because Hermitian conjugation of the traceless jump operators L ~ α = ( α α + i β α ) ⋅ σ ↦ L ~ α † = ( α α − i β α ) ⋅ σ \widetilde{L}_{\alpha}=(\boldsymbol{\alpha}_{\alpha}+i\boldsymbol{\beta}_{\alpha})\cdot\boldsymbol{\sigma}\mapsto\widetilde{L}_{\alpha}^{\dagger}=(\boldsymbol{\alpha}_{\alpha}-i\boldsymbol{\beta}_{\alpha})\cdot\boldsymbol{\sigma} L α = ( α α + i β α ) ⋅ σ ↦ L α † = ( α α − i β α ) ⋅ σ negates β α ↦ − β α \boldsymbol{\beta}_{\alpha}\mapsto-\boldsymbol{\beta}_{\alpha} β α ↦ − β α , it reverses the non-unital drift t ↦ − t t\mapsto-t t ↦ − t in the off-diagonal first column of [ D ] P T M [\mathcal{D}]_{\mathrm{PTM}} [ D ] PTM while leaving the unital 3 × 3 3 \times3 3 × 3 damping tensor S = 2 ∑ α γ α ( a α a α T + b α b α T ) S = 2\sum_{\alpha}\gamma_{\alpha}(a_{\alpha}a_{\alpha}^{\mathsf{T}} + b_{\alpha}b_{\alpha}^{\mathsf{T}}) S = 2 ∑ α γ α ( a α a α T + b α b α T ) and hence M ε ( 1 ) M_{\varepsilon}(1) M ε ( 1 ) and F e ( Φ ε , U 0 ( 1 ) ) = 1 4 [ 1 + Tr ( R ( 1 ) T M ε ( 1 ) ) ] F_{\mathrm{e}}(\Phi_{\varepsilon},U_{0}(1)) = \frac{1}{4}[1+\operatorname{Tr}(R(1)^{\mathsf{T}}M_{\varepsilon}(1))] F e ( Φ ε , U 0 ( 1 )) = 4 1 [ 1 + Tr ( R ( 1 ) T M ε ( 1 ))] identically invariant at all orders in ε ≥ 0 \varepsilon\ge0 ε ≥ 0 .
For a three-level qutrit (d = 3 d=3 d = 3 ), by contrast, the second-order Dyson coefficient of entanglement fidelity is
c 2 ( a b ) ( 2 π , η ) = 1 d 2 ∫ 0 1 d s 1 ∫ 0 s 1 d s 2 Tr ( D ^ ( s 1 ) D ^ ( s 2 ) ) = 1 18 ∫ 0 1 d s 1 ∫ 0 1 d s 2 Tr ( D ^ ( s 1 ) D ^ ( s 2 ) ) , (91) c_{2}^{(ab)}(2\pi,\eta)=\frac{1}{d^{2}}\int_{0}^{1}\mathrm{d}s_{1}\int_{0}^{s_{1}}\mathrm{d}s_{2}\operatorname{Tr}\bigl(\widehat{D}(s_{1})\widehat{D}(s_{2})\bigr)=\frac{1}{18}\int_{0}^{1}\mathrm{d}s_{1}\int_{0}^{1}\mathrm{d}s_{2}\operatorname{Tr}\bigl(\widehat{D}(s_{1})\widehat{D}(s_{2})\bigr), \tag*{(91)} c 2 ( ab ) ( 2 π , η ) = d 2 1 ∫ 0 1 d s 1 ∫ 0 s 1 d s 2 Tr ( D ( s 1 ) D ( s 2 ) ) = 18 1 ∫ 0 1 d s 1 ∫ 0 1 d s 2 Tr ( D ( s 1 ) D ( s 2 ) ) , (91)
where the second equality uses Tr ( D ^ ( s 1 ) D ^ ( s 2 ) ) = Tr ( D ^ ( s 2 ) D ^ ( s 1 ) ) \operatorname{Tr}(\widehat{D}(s_{1})\widehat{D}(s_{2}))=\operatorname{Tr}(\widehat{D}(s_{2})\widehat{D}(s_{1})) Tr ( D ( s 1 ) D ( s 2 )) = Tr ( D ( s 2 ) D ( s 1 )) . Under H 02 ( 2 π ) H_{02}(2\pi) H 02 ( 2 π ) , the evolving states are ∣ 0 ( s ) ⟩ = cos ( π s ) ∣ 0 ⟩ − i sin ( π s ) ∣ 2 ⟩ \lvert0(s)\rangle=\cos(\pi s)\lvert0\rangle-i\sin(\pi s)\lvert2\rangle ∣ 0 ( s )⟩ = cos ( π s ) ∣ 0 ⟩ − i sin ( π s ) ∣ 2 ⟩ , ∣ 1 ( s ) ⟩ = ∣ 1 ⟩ \lvert1(s)\rangle=\lvert1\rangle ∣ 1 ( s )⟩ = ∣ 1 ⟩ , and ∣ 2 ( s ) ⟩ = − i sin ( π s ) ∣ 0 ⟩ + cos ( π s ) ∣ 2 ⟩ \lvert2(s)\rangle=-i\sin(\pi s)\lvert0\rangle+\cos(\pi s)\lvert2\rangle ∣ 2 ( s )⟩ = − i sin ( π s ) ∣ 0 ⟩ + cos ( π s ) ∣ 2 ⟩ ; under H 12 ( 2 π ) H_{12}(2\pi) H 12 ( 2 π ) , ∣ 0 ( s ) ⟩ = ∣ 0 ⟩ \lvert0(s)\rangle=\lvert0\rangle ∣ 0 ( s )⟩ = ∣ 0 ⟩ , ∣ 1 ( s ) ⟩ = cos ( π s ) ∣ 1 ⟩ − i sin ( π s ) ∣ 2 ⟩ \lvert1(s)\rangle=\cos(\pi s)\lvert1\rangle-i\sin(\pi s)\lvert2\rangle ∣ 1 ( s )⟩ = cos ( π s ) ∣ 1 ⟩ − i sin ( π s ) ∣ 2 ⟩ , and ∣ 2 ( s ) ⟩ = − i sin ( π s ) ∣ 1 ⟩ + cos ( π s ) ∣ 2 ⟩ \lvert2(s)\rangle=-i\sin(\pi s)\lvert1\rangle+\cos(\pi s)\lvert2\rangle ∣ 2 ( s )⟩ = − i sin ( π s ) ∣ 1 ⟩ + cos ( π s ) ∣ 2 ⟩ . Evaluating the superoperator trace inner products over ( s 1 , s 2 ) ∈ [ 0 , 1 ] 2 (s_{1},s_{2})\in[0,1]^{2} ( s 1 , s 2 ) ∈ [ 0 , 1 ] 2 yields the exact quadratic polynomials
c 2 ( 02 ) ( 2 π , η ) = 10 + 16 η + 7 η 2 144 , c 2 ( 12 ) ( 2 π , η ) = 4 + 20 η + 10 η 2 144 . (92) c_{2}^{(02)}(2\pi,\eta)=\frac{10+16\eta+7\eta^{2}}{144},\qquad c_{2}^{(12)}(2\pi,\eta)=\frac{4+20\eta+10\eta^{2}}{144}. \tag*{(92)} c 2 ( 02 ) ( 2 π , η ) = 144 10 + 16 η + 7 η 2 , c 2 ( 12 ) ( 2 π , η ) = 144 4 + 20 η + 10 η 2 . (92)
Subtracting c 2 ( 12 ) ( 2 π , η ) c_{2}^{(12)}(2\pi,\eta) c 2 ( 12 ) ( 2 π , η ) from c 2 ( 02 ) ( 2 π , η ) c_{2}^{(02)}(2\pi,\eta) c 2 ( 02 ) ( 2 π , η ) gives ( 6 − 4 η − 3 η 2 ) / 144 (6-4\eta-3\eta^{2})/144 ( 6 − 4 η − 3 η 2 ) /144 , whose unique positive root is η c r i t = ( 22 − 2 ) / 3 \eta_{\mathrm{crit}}=(\sqrt{22}-2)/3 η crit = ( 22 − 2 ) /3 .
Redfield spectral parity separation and KMS quantum thermometry
Proposition 5 (Colored-bath odd-spectrum selection rule and bath thermometry). *For a single qubit coupled weakly via H i n t = ∑ j ∈ { x , y , z } σ j ⊗ B j H_{\mathrm{int}}=\sum_{j\in\{x,y,z\}}\sigma_{j}\otimes B_{j} H int = ∑ j ∈ { x , y , z } σ j ⊗ B j to a stationary Gaussian bath with two-time correlation matrix C j k ( τ ) : = ⟨ B j ( τ ) B k ( 0 ) ⟩ B C_{jk}(\tau):=\langle B_{j}(\tau)B_{k}(0)\rangle_{B} C j k ( τ ) := ⟨ B j ( τ ) B k ( 0 ) ⟩ B and spectral density matrix S ( ν ) = ∫ − ∞ ∞ C ( τ ) e i ν τ d τ = S ( s ) ( ν ) + i S ( a ) ( ν ) ⪰ 0 S(\nu)=\int_{-\infty}^{\infty}C(\tau)e^{i\nu\tau}\,\mathrm{d}\tau=S^{(s)}(\nu)+iS^{(a)}(\nu)\succeq0 S ( ν ) = ∫ − ∞ ∞ C ( τ ) e i ν τ d τ = S ( s ) ( ν ) + i S ( a ) ( ν ) ⪰ 0 (S ( s ) ( − ν ) = S ( s ) ( ν ) S^{(s)}(-\nu)=S^{(s)}(\nu) S ( s ) ( − ν ) = S ( s ) ( ν ) real symmetric, S ( a ) ( − ν ) = − S ( a ) ( ν ) S^{(a)}(-\nu)=-S^{(a)}(\nu) S ( a ) ( − ν ) = − S ( a ) ( ν ) real skew-symmetric), the first-order Redfield survival error slopes satisfy
χ + + χ − = ∫ − ∞ ∞ Tr ( S ( s ) ( ν ) F e v e n ( ν ) ) d ν 2 π , χ + − χ − = − ∫ − ∞ ∞ Tr ( S ( a ) ( ν ) F o d d ( ν ) ) d ν 2 π , (93) \chi_{+}+\chi_{-}=\int_{-\infty}^{\infty}\operatorname{Tr}\bigl(S^{(s)}(\nu)F_{\mathrm{even}}(\nu)\bigr)\frac{\mathrm{d}\nu}{2\pi},\qquad\chi_{+}-\chi_{-}=-\int_{-\infty}^{\infty}\operatorname{Tr}\bigl(S^{(a)}(\nu)F_{\mathrm{odd}}(\nu)\bigr)\frac{\mathrm{d}\nu}{2\pi}, \tag*{(93)} χ + + χ − = ∫ − ∞ ∞ Tr ( S ( s ) ( ν ) F even ( ν ) ) 2 π d ν , χ + − χ − = − ∫ − ∞ ∞ Tr ( S ( a ) ( ν ) F odd ( ν ) ) 2 π d ν , (93) where F e v e n ( − ν ) = F e v e n ( ν ) ⪰ 0 F_{\mathrm{even}}(-\nu)=F_{\mathrm{even}}(\nu)\succeq0 F even ( − ν ) = F even ( ν ) ⪰ 0 is real symmetric and F o d d ( − ν ) = − F o d d ( ν ) F_{\mathrm{odd}}(-\nu)=-F_{\mathrm{odd}}(\nu) F odd ( − ν ) = − F odd ( ν ) is real skew-symmetric. Under an autonomous swap H q = − q π 2 σ x H_{q}=-\frac{q\pi}{2}\sigma_{x} H q = − 2 q π σ x coupled to a transverse thermal bath at frequency ω b \omega_{b} ω b and temperature T b a t h T_{\mathrm{bath}} T bath (γ ϕ = 0 \gamma_{\phi}=0 γ ϕ = 0 ), the directional survival slopes evaluate to
χ ± ( q ) = 3 8 ( γ ↓ + γ ↑ ) ∓ 1 q π ( γ ↓ − γ ↑ ) , (94) \chi_{\pm}(q)=\frac{3}{8}(\gamma_{\downarrow}+\gamma_{\uparrow})\mp\frac{1}{q\pi}(\gamma_{\downarrow}-\gamma_{\uparrow}), \tag*{(94)} χ ± ( q ) = 8 3 ( γ ↓ + γ ↑ ) ∓ q π 1 ( γ ↓ − γ ↑ ) , (94) which yields the calibration-free quantum thermometer (55) of the main text.
Proof. In the toggling frame of the control unitary U 0 ( s ) U_{0}(s) U 0 ( s ) with R j l ( s ) : = 1 2 Tr ( σ j U 0 ( s ) † σ l U 0 ( s ) ) ∈ S O ( 3 ) R_{jl}(s):=\frac{1}{2}\operatorname{Tr}(\sigma_{j}U_{0}(s)^{\dagger}\sigma_{l}U_{0}(s))\in\mathrm{SO}(3) R j l ( s ) := 2 1 Tr ( σ j U 0 ( s ) † σ l U 0 ( s )) ∈ SO ( 3 ) , the second-order Dyson expansion of the reduced density matrix ρ ~ ( 1 ) = Tr B [ U t o t ( 1 ) ( ρ i n ⊗ ρ B ) U t o t ( 1 ) † ] \widetilde{\rho}(1)=\operatorname{Tr}_{B}[U_{\mathrm{tot}}(1)(\rho_{\mathrm{in}}\otimes\rho_{B})U_{\mathrm{tot}}(1)^{\dagger}] ρ ( 1 ) = Tr B [ U tot ( 1 ) ( ρ in ⊗ ρ B ) U tot ( 1 ) † ] expresses the survival error probabilities p ± = 1 − ⟨ ψ ± ∣ ρ ~ ± ( 1 ) ∣ ψ ± ⟩ p_{\pm}=1-\langle\psi_{\pm}|\widetilde{\rho}_{\pm}(1)|\psi_{\pm}\rangle p ± = 1 − ⟨ ψ ± ∣ ρ ± ( 1 ) ∣ ψ ± ⟩ for ρ i n , ± = 1 2 ( I 2 ± σ z ) \rho_{\mathrm{in},\pm}=\frac{1}{2}(I_{2}\pm\sigma_{z}) ρ in , ± = 2 1 ( I 2 ± σ z ) as double time integrals over 0 ≤ s 2 ≤ s 1 ≤ 1 0\le s_{2}\le s_{1}\le1 0 ≤ s 2 ≤ s 1 ≤ 1 . Transforming to the Fourier representation C j k ( T ( s 1 − s 2 ) ) = ∫ − ∞ ∞ S j k ( ν ) e − i ν T ( s 1 − s 2 ) d ν 2 π C_{jk}(T(s_{1}-s_{2}))=\int_{-\infty}^{\infty}S_{jk}(\nu)e^{-i\nu T(s_{1}-s_{2})}\frac{\mathrm{d}\nu}{2\pi} C j k ( T ( s 1 − s 2 )) = ∫ − ∞ ∞ S j k ( ν ) e − i ν T ( s 1 − s 2 ) 2 π d ν and symmetrizing over ( s 1 , s 2 ) ∈ [ 0 , 1 ] 2 (s_{1},s_{2})\in[0,1]^{2} ( s 1 , s 2 ) ∈ [ 0 , 1 ] 2 :
The σ z \sigma_{z} σ z -independent unital commutator term Tr ( [ σ ~ j ( s 1 ) , [ σ ~ k ( s 2 ) , I 2 ] ] σ z ) = 0 \operatorname{Tr}([\widetilde{\sigma}_{j}(s_{1}),[\widetilde{\sigma}_{k}(s_{2}),I_{2}]]\sigma_{z})=0 Tr ([ σ j ( s 1 ) , [ σ k ( s 2 ) , I 2 ]] σ z ) = 0 vanishes, while the state-even survival sum χ + + χ − \chi_{+}+\chi_{-} χ + + χ − depends on Tr ( σ z [ σ ~ j ( s 1 ) , [ σ ~ k ( s 2 ) , σ z ] ] ) \operatorname{Tr}(\sigma_{z}[\widetilde{\sigma}_{j}(s_{1}),[\widetilde{\sigma}_{k}(s_{2}),\sigma_{z}]]) Tr ( σ z [ σ j ( s 1 ) , [ σ k ( s 2 ) , σ z ]]) , which is symmetric under ( j , s 1 ) ↔ ( k , s 2 ) (j,s_{1})\leftrightarrow(k,s_{2}) ( j , s 1 ) ↔ ( k , s 2 ) and pairs exclusively with the real frequency-even spectrum S ( s ) ( ν ) S^{(s)}(\nu) S ( s ) ( ν ) via the positive-semidefinite filter matrix F e v e n ( ν ) F_{\mathrm{even}}(\nu) F even ( ν ) .
The state-odd directional difference χ + − χ − \chi_{+}-\chi_{-} χ + − χ − depends on the anticommutator Tr ( σ z [ σ ~ j ( s 1 ) , { σ ~ k ( s 2 ) , I 2 } ] ) \operatorname{Tr}(\sigma_{z}[\widetilde{\sigma}_{j}(s_{1}),\{\widetilde{\sigma}_{k}(s_{2}),I_{2}\}]) Tr ( σ z [ σ j ( s 1 ) , { σ k ( s 2 ) , I 2 }]) , which is antisymmetric in the Pauli indices via ϵ a b z R j a ( s 1 ) R k b ( s 2 ) \epsilon_{abz}R_{ja}(s_{1})R_{kb}(s_{2}) ϵ ab z R j a ( s 1 ) R k b ( s 2 ) and pairs exclusively with the imaginary frequency-odd spectrum i S ( a ) ( ν ) iS^{(a)}(\nu) i S ( a ) ( ν ) via the skew-symmetric filter matrix F o d d ( ν ) F_{\mathrm{odd}}(\nu) F odd ( ν ) .
In the Markovian thermal regime at transition frequency ω b \omega_{b} ω b , an x x x -axis swap H q = − q π 2 σ x H_{q}=-\frac{q\pi}{2}\sigma_{x} H q = − 2 q π σ x between ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ and ∣ 1 ⟩ \lvert1\rangle ∣ 1 ⟩ has m q = ( 0 , 2 q π , 0 ) T ∥ e y \boldsymbol{m}_{q}=(0,\frac{2}{q\pi},0)^{\mathsf{T}}\parallel\boldsymbol{e}_{y} m q = ( 0 , q π 2 , 0 ) T ∥ e y and G q = diag ( 1 , 1 / 2 , 1 / 2 ) G_{q}=\operatorname{diag}(1,1/2,1/2) G q = diag ( 1 , 1/2 , 1/2 ) , so a longitudinal thermal bath ( t ∥ e z ) (\boldsymbol{t}\parallel\boldsymbol{e}_{z}) ( t ∥ e z ) gives m q ⋅ t = 0 \boldsymbol{m}_{q}\cdot\boldsymbol{t}=0 m q ⋅ t = 0 . When the thermal bath is quantized along the transverse e y \boldsymbol{e}_{y} e y axis (L ↓ , y = γ ↓ ∣ + y ⟩ ⟨ − y ∣ L_{\downarrow,y}=\sqrt{\gamma_{\downarrow}}\lvert+y\rangle\langle-y\rvert L ↓ , y = γ ↓ ∣ + y ⟩ ⟨ − y ∣ , L ↑ , y = γ ↑ ∣ − y ⟩ ⟨ + y ∣ L_{\uparrow,y}=\sqrt{\gamma_{\uparrow}}|{-y}\rangle\langle{+y}| L ↑ , y = γ ↑ ∣ − y ⟩ ⟨ + y ∣ , its bath tensors (19) are S y = 1 2 ( γ ↓ + γ ↑ ) diag ( 1 , 0 , 1 ) S_y=\frac{1}{2}(\gamma_{\downarrow}+\gamma_{\uparrow})\operatorname{diag}(1,0,1) S y = 2 1 ( γ ↓ + γ ↑ ) diag ( 1 , 0 , 1 ) and t y = 2 ( γ ↓ − γ ↑ ) e y t_y=2(\gamma_{\downarrow}-\gamma_{\uparrow})e_y t y = 2 ( γ ↓ − γ ↑ ) e y (physically realized in circuit QED by continuous resonant σ y \sigma_y σ y spin-locking at Rabi frequency ω b \omega_b ω b or, for a standard longitudinal bath S z = 1 2 ( γ ↓ + γ ↑ ) diag ( 1 , 1 , 0 ) S_z=\frac{1}{2}(\gamma_{\downarrow}+\gamma_{\uparrow})\operatorname{diag}(1,1,0) S z = 2 1 ( γ ↓ + γ ↑ ) diag ( 1 , 1 , 0 ) and t z = 2 ( γ ↓ − γ ↑ ) e z t_z=2(\gamma_{\downarrow}-\gamma_{\uparrow})e_z t z = 2 ( γ ↓ − γ ↑ ) e z , by preparing and measuring the equatorial states ∣ ± y ⟩ |\pm y\rangle ∣ ± y ⟩ under a σ x \sigma_x σ x swap so that G x x + G y y = 1 + 1 2 = 3 2 G_{xx}+G_{yy}=1+\frac{1}{2}=\frac{3}{2} G xx + G y y = 1 + 2 1 = 2 3 and m z = 2 q π m_z=\frac{2}{q\pi} m z = q π 2 ). In either realization, 1 2 Tr ( S G ) = 3 8 ( γ ↓ + γ ↑ ) \frac{1}{2}\operatorname{Tr}(SG)=\frac{3}{8}(\gamma_{\downarrow}+\gamma_{\uparrow}) 2 1 Tr ( S G ) = 8 3 ( γ ↓ + γ ↑ ) and 1 4 m ⋅ t = 1 q π ( γ ↓ − γ ↑ ) \frac{1}{4}m\cdot t=\frac{1}{q\pi}(\gamma_{\downarrow}-\gamma_{\uparrow}) 4 1 m ⋅ t = q π 1 ( γ ↓ − γ ↑ ) , so Theorem 1 gives Eq. (94) . Substituting the Kubo–Martin–Schwinger (KMS) detailed-balance ratio [44 , 45 ]
γ ↓ − γ ↑ γ ↓ + γ ↑ = 1 − e − ℏ ω b / ( k B T b a t h ) 1 + e − ℏ ω b / ( k B T b a t h ) = tanh ( ℏ ω b 2 k B T b a t h ) (95) \frac{\gamma_{\downarrow}-\gamma_{\uparrow}}{\gamma_{\downarrow}+\gamma_{\uparrow}}=\frac{1-e^{-\hbar\omega_b/(k_BT_{\mathrm{bath}})}}{1+e^{-\hbar\omega_b/(k_BT_{\mathrm{bath}})}}=\tanh\left(\frac{\hbar\omega_b}{2k_BT_{\mathrm{bath}}}\right) \tag*{(95)} γ ↓ + γ ↑ γ ↓ − γ ↑ = 1 + e − ℏ ω b / ( k B T bath ) 1 − e − ℏ ω b / ( k B T bath ) = tanh ( 2 k B T bath ℏ ω b ) (95)
into ∣ χ + ( q ) − χ − ( q ) ∣ / ( χ + ( q ) + χ − ( q ) ) |\chi_+(q)-\chi_-(q)|/(\chi_+(q)+\chi_-(q)) ∣ χ + ( q ) − χ − ( q ) ∣/ ( χ + ( q ) + χ − ( q )) cancels the overall coupling prefactor γ ↓ + γ ↑ \gamma_{\downarrow}+\gamma_{\uparrow} γ ↓ + γ ↑ and establishes Eq. (55) .
Three-mode qubit Slepian concentration pencil and spectral geometry
This appendix presents the Fourier-mode Slepian concentration pencil formulation of the qubit selectivity bound (Theorem 2) and its generalization to multilevel (d ≥ 5 , 7 d\ge5,7 d ≥ 5 , 7 ) bipartite Cauchy Fourier models.
Proposition 6 (Three-mode qubit Slepian pencil). Under the autonomous swap Hamiltonian H q = − q π 2 σ x H_q=-\frac{q\pi}{2}\sigma_x H q = − 2 q π σ x (q ∈ 2 Z + 1 q\in2\mathbb{Z}+1 q ∈ 2 Z + 1 ), expanding any jump operator L ∈ B ( C 2 ) L\in\mathcal{B}(\mathbb{C}^2) L ∈ B ( C 2 ) in the σ x \sigma_x σ x -eigenbasis ∣ ± ⟩ = ( ∣ 0 ⟩ ± ∣ 1 ⟩ ) / 2 |\pm\rangle=(|0\rangle\pm|1\rangle)/\sqrt{2} ∣ ± ⟩ = ( ∣0 ⟩ ± ∣1 ⟩) / 2 decomposes the evolving matrix elements ⟨ ψ ± ( s ) ∣ L ∣ ψ ± ( s ) ⟩ \langle\psi_\pm(s)|L|\psi_\pm(s)\rangle ⟨ ψ ± ( s ) ∣ L ∣ ψ ± ( s )⟩ into three harmonic Fourier modes { e + i q π s , 1 , e − i q π s } \{e^{+iq\pi s},1,e^{-iq\pi s}\} { e + i q π s , 1 , e − i q π s } . For any rank-1 jump operator L = ∣ Φ d a r k ⟩ ⟨ Ψ ∣ L=|\Phi_{\mathrm{dark}}\rangle\langle\Psi| L = ∣ Φ dark ⟩ ⟨ Ψ∣ whose target state ∣ Φ d a r k ⟩ |\Phi_{\mathrm{dark}}\rangle ∣ Φ dark ⟩ is orthogonal to the equator or dark with respect to the active subspace, parameterized by the normalized three-mode vector x ∈ C 3 x\in\mathbb{C}^3 x ∈ C 3 (∥ x ∥ = 1 \lVert x\rVert=1 ∥ x ∥ = 1 ), the directional survival error slopes take the quadratic pencil form
χ ± ( L ) = x † ( I 3 ± H ^ q ) x , H ^ q = 2 π q ( 0 − i 0 i 0 − i 0 i 0 ) , spec ( H ^ q ) = { − 2 2 π ∣ q ∣ , 0 , + 2 2 π ∣ q ∣ } . (96) \chi_\pm(L)=x^\dagger(I_3\pm\widehat{H}_q)x,\qquad\widehat{H}_q=\frac{2}{\pi q}\begin{pmatrix}0&-i&0\\ i&0&-i\\ 0&i&0\end{pmatrix},\qquad\operatorname{spec}(\widehat{H}_q)=\left\{-\frac{2\sqrt{2}}{\pi|q|},0,+\frac{2\sqrt{2}}{\pi|q|}\right\}. \tag*{(96)} χ ± ( L ) = x † ( I 3 ± H q ) x , H q = π q 2 0 i 0 − i 0 i 0 − i 0 , spec ( H q ) = { − π ∣ q ∣ 2 2 , 0 , + π ∣ q ∣ 2 2 } . (96)
Maximizing the Rayleigh quotient ∣ x † H ^ q x ∣ / ( x † x ) |x^\dagger\widehat{H}_qx|/(x^\dagger x) ∣ x † H q x ∣/ ( x † x ) immediately recovers β q = 2 2 π ∣ q ∣ \beta_q=\frac{2\sqrt{2}}{\pi|q|} β q = π ∣ q ∣ 2 2 and κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) \kappa_\star=(\pi-2\sqrt{2})/(\pi+2\sqrt{2}) κ ⋆ = ( π − 2 2 ) / ( π + 2 2 ) .
Proof. In the σ x \sigma_x σ x -eigenbasis ∣ ± ⟩ |\pm\rangle ∣ ± ⟩ where U 0 ( s ) ∣ ± ⟩ = e ± i q π s / 2 ∣ ± ⟩ U_0(s)|\pm\rangle=e^{\pm iq\pi s/2}|\pm\rangle U 0 ( s ) ∣ ± ⟩ = e ± i q π s /2 ∣ ± ⟩ , the evolving states are ∣ ψ + ( s ) ⟩ = 1 2 ( e i q π s / 2 ∣ + ⟩ + e − i q π s / 2 ∣ − ⟩ ) |\psi_+(s)\rangle=\frac{1}{\sqrt{2}}(e^{iq\pi s/2}|+\rangle+e^{-iq\pi s/2}|-\rangle) ∣ ψ + ( s )⟩ = 2 1 ( e i q π s /2 ∣ + ⟩ + e − i q π s /2 ∣ − ⟩) and ∣ ψ − ( s ) ⟩ = 1 2 ( e i q π s / 2 ∣ + ⟩ − e − i q π s / 2 ∣ − ⟩ ) |\psi_-(s)\rangle=\frac{1}{\sqrt{2}}(e^{iq\pi s/2}|+\rangle-e^{-iq\pi s/2}|-\rangle) ∣ ψ − ( s )⟩ = 2 1 ( e i q π s /2 ∣ + ⟩ − e − i q π s /2 ∣ − ⟩) . Flipping ∣ ψ + ( s ) ⟩ ↦ ∣ ψ − ( s ) ⟩ |\psi_+(s)\rangle\mapsto|\psi_-(s)\rangle ∣ ψ + ( s )⟩ ↦ ∣ ψ − ( s )⟩ flips the relative sign between even and odd Fourier harmonics; the time integrals ∫ 0 1 e i ( m − n ) q π s / 2 d s \int_0^1 e^{i(m-n)q\pi s/2}\,ds ∫ 0 1 e i ( m − n ) q π s /2 d s between opposite-parity harmonics evaluate to ± 2 i π q ( m − n ) \frac{\pm2i}{\pi q(m-n)} π q ( m − n ) ± 2 i , yielding the tridiagonal Hermitian Toeplitz Slepian matrix H ^ q \widehat{H}_q H q in Eq. (96) , whose eigenvalues are 0 0 0 and ± 2 π ∣ q ∣ 1 2 + 1 2 = ± 2 2 π ∣ q ∣ \pm\frac{2}{\pi|q|}\sqrt{1^2+1^2}=\pm\frac{2\sqrt{2}}{\pi|q|} ± π ∣ q ∣ 2 1 2 + 1 2 = ± π ∣ q ∣ 2 2 .
Statistical theory of the six-circuit SPAM-invariant lock-in protocol, Cramér–Rao bounds, and decoding graph weights
This appendix provides the complete mathematical derivation of the six-circuit SPAM-invariant signed-winding lock-in protocol (Protocol 1 in § 7.2 ), its exact Fisher information and Cramér–Rao variance bounds, and the logical failure excess of Pauli-twirled decoding weights in repetition and surface codes.
Proposition 7 (SPAM-invariant signed-winding lock-in protocol and Cramér–Rao gain). Under gate-independent Markovian bit-flip SPAM with state-preparation errors ( e 0 , e 1 ) (e_0,e_1) ( e 0 , e 1 ) ( v p r e p = 1 − e 0 − e 1 , η p r e p = e 0 − e 1 ) (v_{\mathrm{prep}}=1-e_0-e_1,\ \eta_{\mathrm{prep}}=e_0-e_1) ( v prep = 1 − e 0 − e 1 , η prep = e 0 − e 1 ) and readout assignment errors ( f 0 , f 1 ) (f_0,f_1) ( f 0 , f 1 ) ( v r o = 1 − f 0 − f 1 , V S P A M = v r o v p r e p ) (v_{\mathrm{ro}}=1-f_0-f_1,\ V_{\mathrm{SPAM}}=v_{\mathrm{ro}}v_{\mathrm{prep}}) ( v ro = 1 − f 0 − f 1 , V SPAM = v ro v prep ) , let R 0 ( q ) ( m ) R_0^{(q)}(m) R 0 ( q ) ( m ) and R 1 ( q ) ( m ) R_1^{(q)}(m) R 1 ( q ) ( m ) denote the probabilities of recording outcome “1” after preparing ∣ 0 ⟩ \lvert0\rangle ∣ 0 ⟩ or ∣ 1 ⟩ \lvert1\rangle ∣ 1 ⟩ and applying m ∈ { 0 , 1 } m\in\{0,1\} m ∈ { 0 , 1 } swaps of signed winding q q q . With a q : = p − ( q ) a_q:=p_-(q) a q := p − ( q ) , b q : = p + ( q ) b_q:=p_+(q) b q := p + ( q ) , Σ q : = a q + b q \Sigma_q:=a_q+b_q Σ q := a q + b q , and Δ q : = a q − b q = Δ 1 / q + O ( ε 2 ) \Delta_q:=a_q-b_q=\Delta_1/q+O(\varepsilon^2) Δ q := a q − b q = Δ 1 / q + O ( ε 2 ) , the sum and difference observables Y m ( q ) : = R 0 ( q ) ( m ) + R 1 ( q ) ( m ) Y_m^{(q)}:=R_0^{(q)}(m)+R_1^{(q)}(m) Y m ( q ) := R 0 ( q ) ( m ) + R 1 ( q ) ( m ) and Z m ( q ) : = R 0 ( q ) ( m ) − R 1 ( q ) ( m ) Z_m^{(q)}:=R_0^{(q)}(m)-R_1^{(q)}(m) Z m ( q ) := R 0 ( q ) ( m ) − R 1 ( q ) ( m ) obey
Y 0 = 2 f 0 + v r o ( 1 + η p r e p ) , Z 0 = − V S P A M , (97) Y_0=2f_0+v_{\mathrm{ro}}(1+\eta_{\mathrm{prep}}),\qquad Z_0=-V_{\mathrm{SPAM}}, \tag*{(97)} Y 0 = 2 f 0 + v ro ( 1 + η prep ) , Z 0 = − V SPAM , (97) Y 1 ( q ) = 2 f 0 + v r o ( 1 − η p r e p − Δ q + η p r e p Σ q ) , Z 1 ( q ) = V S P A M ( 1 − Σ q ) , (98) Y_1^{(q)}=2f_0+v_{\mathrm{ro}}(1-\eta_{\mathrm{prep}}-\Delta_q+\eta_{\mathrm{prep}}\Sigma_q),\qquad Z_1^{(q)}=V_{\mathrm{SPAM}}(1-\Sigma_q), \tag*{(98)} Y 1 ( q ) = 2 f 0 + v ro ( 1 − η prep − Δ q + η prep Σ q ) , Z 1 ( q ) = V SPAM ( 1 − Σ q ) , (98) and the six-circuit signed-winding estimator ( q ∈ { + 1 , − 1 } ) (q\in\{+1,-1\}) ( q ∈ { + 1 , − 1 }) satisfies
g ^ A ′ : = Y 1 ( − 1 ) − Y 1 ( + 1 ) − 2 Z 0 = Δ 1 v p r e p + O ( ε 2 ) , (99) \widehat{g}_{A'}:=\frac{Y_1^{(-1)}-Y_1^{(+1)}}{-2Z_0}=\frac{\Delta_1}{v_{\mathrm{prep}}}+O(\varepsilon^2), \tag*{(99)} g A ′ := − 2 Z 0 Y 1 ( − 1 ) − Y 1 ( + 1 ) = v prep Δ 1 + O ( ε 2 ) , (99) achieving an asymptotic 3 2 = 9 × 3^2=9\times 3 2 = 9 × plug-in variance reduction relative to the unsigned multi-winding estimator
g ^ A : = 3 ( Y 1 ( 3 ) − Y 1 ( 1 ) ) − 2 Z 0 ( q ∈ { 1 , 3 } ) . \widehat{g}_A:=\frac{3(Y_1^{(3)}-Y_1^{(1)})}{-2Z_0}\quad(q\in\{1,3\}). g A := − 2 Z 0 3 ( Y 1 ( 3 ) − Y 1 ( 1 ) ) ( q ∈ { 1 , 3 }) . Proof. In the two-state classical transition-matrix representation on { ∣ 0 ⟩ , ∣ 1 ⟩ } \{\lvert0\rangle,\lvert1\rangle\} {∣ 0 ⟩ , ∣ 1 ⟩} , preparing ∣ j ⟩ \lvert j\rangle ∣ j ⟩ ( j ∈ { 0 , 1 } ) (j\in\{0,1\}) ( j ∈ { 0 , 1 }) produces the initial population vector p 0 = ( 1 − e 0 , e 0 ) T \mathbf{p}_0=(1-e_0,e_0)^T p 0 = ( 1 − e 0 , e 0 ) T for j = 0 j=0 j = 0 and p 1 = ( e 1 , 1 − e 1 ) T \mathbf{p}_1=(e_1,1-e_1)^T p 1 = ( e 1 , 1 − e 1 ) T for j = 1 j=1 j = 1 . One swap of winding q q q acts via the stochastic transition matrix
T s w a p ( q ) = ( b q 1 − a q 1 − b q a q ) , (100) T_{\mathrm{swap}}^{(q)}=
\begin{pmatrix}
b_q & 1-a_q\\
1-b_q & a_q
\end{pmatrix}, \tag*{(100)} T swap ( q ) = ( b q 1 − b q 1 − a q a q ) , (100) and readout outcome “1” is recorded with probability vector r r o T = ( f 0 , 1 − f 1 ) = ( f 0 , f 0 + v r o ) \mathbf{r}_{\mathrm{ro}}^T=(f_0,1-f_1)=(f_0,f_0+v_{\mathrm{ro}}) r ro T = ( f 0 , 1 − f 1 ) = ( f 0 , f 0 + v ro ) . Direct multiplication R j ( q ) ( 0 ) = r r o T p j R_j^{(q)}(0)=\mathbf{r}_{\mathrm{ro}}^T\mathbf{p}_j R j ( q ) ( 0 ) = r ro T p j and R j ( q ) ( 1 ) = r r o T T s w a p ( q ) p j R_j^{(q)}(1)=\mathbf{r}_{\mathrm{ro}}^TT_{\mathrm{swap}}^{(q)}\mathbf{p}_j R j ( q ) ( 1 ) = r ro T T swap ( q ) p j yields Eqs. (97) –(98) . Because Σ − 1 = Σ + 1 + O ( ε 2 ) \Sigma_{-1}=\Sigma_{+1}+O(\varepsilon^2) Σ − 1 = Σ + 1 + O ( ε 2 ) while Δ − 1 = − Δ + 1 + O ( ε 2 ) \Delta_{-1}=-\Delta_{+1}+O(\varepsilon^2) Δ − 1 = − Δ + 1 + O ( ε 2 ) , subtracting Y 1 ( + 1 ) Y_1^{(+1)} Y 1 ( + 1 ) from Y 1 ( − 1 ) Y_1^{(-1)} Y 1 ( − 1 ) cancels 2 f 0 2f_0 2 f 0 , v r o ( 1 − η p r e p ) v_{\mathrm{ro}}(1-\eta_{\mathrm{prep}}) v ro ( 1 − η prep ) , and v r o η p r e p Σ q v_{\mathrm{ro}}\eta_{\mathrm{prep}}\Sigma_q v ro η prep Σ q , leaving Y 1 ( − 1 ) − Y 1 ( + 1 ) = 2 v r o Δ 1 + O ( ε 2 ) Y_1^{(-1)}-Y_1^{(+1)}=2v_{\mathrm{ro}}\Delta_1+O(\varepsilon^2) Y 1 ( − 1 ) − Y 1 ( + 1 ) = 2 v ro Δ 1 + O ( ε 2 ) . Dividing by − 2 Z 0 = 2 v r o v p r e p -2Z_0=2v_{\mathrm{ro}}v_{\mathrm{prep}} − 2 Z 0 = 2 v ro v prep gives Eq. (99) .
For N t o t N_{\mathrm{tot}} N tot total measurement shots allocated equally across the 6 Bernoulli circuits ( n = N t o t / 6 (n=N_{\mathrm{tot}}/6 ( n = N tot /6 shots per circuit), each empirical frequency R ^ j ( q ) ( m ) \widehat{R}_j^{(q)}(m) R j ( q ) ( m ) has binomial variance V j ( q ) ( m ) / n = R j ( q ) ( m ) ( 1 − R j ( q ) ( m ) ) / n V_j^{(q)}(m)/n=R_j^{(q)}(m)(1-R_j^{(q)}(m))/n V j ( q ) ( m ) / n = R j ( q ) ( m ) ( 1 − R j ( q ) ( m )) / n . By the delta method, the asymptotic variance of g ^ A ′ \widehat{g}_{A'} g A ′ is dominated by the numerator difference Y ^ 1 ( − 1 ) − Y ^ 1 ( + 1 ) \widehat{Y}_1^{(-1)}-\widehat{Y}_1^{(+1)} Y 1 ( − 1 ) − Y 1 ( + 1 ) with lever arm 2 v r o 2v_{\mathrm{ro}} 2 v ro , whereas the unsigned estimator g ^ A \widehat{g}_A g A uses q ∈ { 1 , 3 } q\in\{1,3\} q ∈ { 1 , 3 } where Δ 1 − Δ 3 = 2 3 Δ 1 \Delta_1-\Delta_3=\frac{2}{3}\Delta_1 Δ 1 − Δ 3 = 3 2 Δ 1 has lever arm 2 3 v r o \frac{2}{3}v_{\mathrm{ro}} 3 2 v ro . Because the lever arm of g ^ A ′ \widehat{g}_{A'} g A ′ is 3 × 3\times 3 × larger at identical Bernoulli readout noise, its asymptotic variance is 3 2 = 9 × 3^2=9\times 3 2 = 9 × smaller. At the benchmark parameter vector θ 0 = ( e 0 , e 1 , f 0 , f 1 , Σ 1 , Δ 1 ) = ( 0.02 , 0.06 , 0.01 , 0.02 , 0.02 , 0.018 ) \boldsymbol{\theta}_0=(e_0,e_1,f_0,f_1,\Sigma_1,\Delta_1)=(0.02,0.06,0.01,0.02,0.02,0.018) θ 0 = ( e 0 , e 1 , f 0 , f 1 , Σ 1 , Δ 1 ) = ( 0.02 , 0.06 , 0.01 , 0.02 , 0.02 , 0.018 ) plotted in Figure 4 (a), exact evaluation of the 6 × 6 6\times6 6 × 6 binomial Fisher information matrix I ( θ 0 ) I(\boldsymbol{\theta}_0) I ( θ 0 ) gives plug-in variances N t o t Var ( g ^ A ) = 3.959401 N_{\mathrm{tot}}\operatorname{Var}(\widehat{g}_A)=3.959401 N tot Var ( g A ) = 3.959401 versus N t o t Var ( g ^ A ′ ) = 0.441165 N_{\mathrm{tot}}\operatorname{Var}(\widehat{g}_{A'})=0.441165 N tot Var ( g A ′ ) = 0.441165 (an 8.9749 × 8.9749\times 8.9749 × plug-in variance reduction), Cramér–Rao lower bounds for g = Δ 1 / v p r e p g=\Delta_1/v_{\mathrm{prep}} g = Δ 1 / v prep of 3.599532 3.599532 3.599532 versus 0.408712 0.408712 0.408712 (an 8.8070 × 8.8070\times 8.8070 × reduction), and Cramér–Rao lower bounds for the bare split Δ 1 \Delta_1 Δ 1 of 3.268356 3.268356 3.268356 versus 0.382758 0.382758 0.382758 (an 8.5390 × 8.5390\times 8.5390 × reduction).
For the polar X \sqrt{X} X gate under longitudinal T 1 T_1 T 1 relaxation ( t = 2 ε z (t=2\varepsilon_z ( t = 2 ε z , Protocol 1(b)), let U ( Q k ) = exp ( − i Q k 2 σ x ) U(Q_k)=\exp(-i\frac{Q_k}{2}\sigma_x) U ( Q k ) = exp ( − i 2 Q k σ x ) denote the k = 0 k=0 k = 0 principal branch ( Q 0 = + π / 2 , m z ( 0 ) = + 2 / π ) (Q_0=+\pi/2,\ m_z^{(0)}=+2/\pi) ( Q 0 = + π /2 , m z ( 0 ) = + 2/ π ) and the k = 1 k=1 k = 1 reversed over-rotated branch ( Q 1 = − 3 π / 2 , m z ( 1 ) = − 2 / ( 3 π ) = − 1 3 m z ( 0 ) ) (Q_1=-3\pi/2,\ m_z^{(1)}=-2/(3\pi)=-\frac{1}{3}m_z^{(0)}) ( Q 1 = − 3 π /2 , m z ( 1 ) = − 2/ ( 3 π ) = − 3 1 m z ( 0 ) ) at matched duration T T T , followed by a fixed reference X \sqrt{X} X analysis pulse U a n = exp ( i π 4 σ x ) U_{\mathrm{an}}=\exp(i\frac{\pi}{4}\sigma_x) U an = exp ( i 4 π σ x ) with error parameters ( Σ a n , Δ a n ) (\Sigma_{\mathrm{an}},\Delta_{\mathrm{an}}) ( Σ an , Δ an ) . Because the composite ideal unitary U a n U ( Q k ) = ± i σ x U_{\mathrm{an}}U(Q_k)=\pm i\sigma_x U an U ( Q k ) = ± i σ x is a population swap and the two-gate survival errors compose additively at first order in ε \varepsilon ε , the composite error sum and directional split of branch k ∈ { 0 , 1 } k\in\{0,1\} k ∈ { 0 , 1 } are Σ t o t ( k ) = Σ X ( k ) + Σ a n + O ( ε 2 ) \Sigma_{\mathrm{tot}}^{(k)}=\Sigma_{\sqrt{X}}^{(k)}+\Sigma_{\mathrm{an}}+O(\varepsilon^2) Σ tot ( k ) = Σ X ( k ) + Σ an + O ( ε 2 ) and Δ t o t ( k ) = − Δ X ( k ) + Δ a n + O ( ε 2 ) \Delta_{\mathrm{tot}}^{(k)}=-\Delta_{\sqrt{X}}^{(k)}+\Delta_{\mathrm{an}}+O(\varepsilon^2) Δ tot ( k ) = − Δ X ( k ) + Δ an + O ( ε 2 ) (where the sign on Δ X ( k ) \Delta_{\sqrt{X}}^{(k)} Δ X ( k ) reflects the fact that U a n U_{\mathrm{an}} U an inverts the equatorial populations into the Z Z Z readout basis). Since $\Sigma_{\sqrt{X}}^{(0)}=\Sigma_{\sqrt{X}}^{(1)}=\frac{3}{4}\varepsilon
and Δ X ( 1 ) = − 1 3 Δ X ( 0 ) \Delta_{\sqrt{X}}^{(1)}=-\frac{1}{3}\Delta_{\sqrt{X}}^{(0)} Δ X ( 1 ) = − 3 1 Δ X ( 0 ) , subtracting Y 1 ( k = 1 ) Y_{1}^{(k=1)} Y 1 ( k = 1 ) from Y 1 ( k = 0 ) Y_{1}^{(k=0)} Y 1 ( k = 0 ) cancels 2 f 0 2f_{0} 2 f 0 , v r o ( 1 − η p r e p ) v_{\mathrm{ro}}(1-\eta_{\mathrm{prep}}) v ro ( 1 − η prep ) , Δ a n \Delta_{\mathrm{an}} Δ an , and η p r e p Σ t o t ( k ) \eta_{\mathrm{prep}}\Sigma_{\mathrm{tot}}^{(k)} η prep Σ tot ( k ) , leaving Y 1 ( k = 0 ) − Y 1 ( k = 1 ) = 4 3 v r o Δ X ( 0 ) + O ( ε 2 ) Y_{1}^{(k=0)}-Y_{1}^{(k=1)}=\frac{4}{3}v_{\mathrm{ro}}\Delta_{\sqrt{X}}^{(0)}+O(\varepsilon^{2}) Y 1 ( k = 0 ) − Y 1 ( k = 1 ) = 3 4 v ro Δ X ( 0 ) + O ( ε 2 ) and proving Eq. (46) . At θ 0 \theta_{0} θ 0 , g ^ X \hat{g}_{\sqrt{X}} g ^ X has plug-in variance N t o t Var ( g ^ X ) = 0.991450 N_{\mathrm{tot}}\operatorname{Var}(\hat{g}_{\sqrt{X}})=0.991450 N tot Var ( g ^ X ) = 0.991450 and Cramér–Rao bound 0.930127 0.930127 0.930127 (reflecting its ( 3 / 2 ) 2 = 2.25 × (3/2)^{2}=2.25\times ( 3/2 ) 2 = 2.25 × smaller squared lever arm relative to a full signed swap), and exact 4 × 4 4\times4 4 × 4 PTM exponentiation at ibm_kyiv parameters (ε = 8.6806 × 10 − 5 \varepsilon=8.6806\times10^{-5} ε = 8.6806 × 1 0 − 5 , Δ X ( 0 ) = 5.5260 × 10 − 5 \Delta_{\sqrt{X}}^{(0)}=5.5260\times10^{-5} Δ X ( 0 ) = 5.5260 × 1 0 − 5 ) yields g ^ X v p r e p = 5.5253 × 10 − 5 \hat{g}_{\sqrt{X}}v_{\mathrm{prep}}=5.5253\times10^{-5} g ^ X v prep = 5.5253 × 1 0 − 5 (relative error 1.3 × 10 − 4 1.3\times10^{-4} 1.3 × 1 0 − 4 ). If branch k = 1 k=1 k = 1 is instead executed at fixed Rabi amplitude over duration 3 T 3T 3 T (so Σ X ( 1 ) = 3 Σ X ( 0 ) \Sigma_{\sqrt{X}}^{(1)}=3\Sigma_{\sqrt{X}}^{(0)} Σ X ( 1 ) = 3 Σ X ( 0 ) and Δ X ( 1 ) = − Δ X ( 0 ) \Delta_{\sqrt{X}}^{(1)}=-\Delta_{\sqrt{X}}^{(0)} Δ X ( 1 ) = − Δ X ( 0 ) ), then Y 1 ( 0 ) − Y 1 ( 1 ) − 2 Z 0 + Y 0 + Z 0 2 Z 0 Z 1 ( 1 ) − Z 1 ( 0 ) Z 0 = Δ X ( 0 ) v p r e p + O ( ε 2 ) \frac{Y_{1}^{(0)}-Y_{1}^{(1)}}{-2Z_{0}}+\frac{Y_{0}+Z_{0}}{2Z_{0}}\frac{Z_{1}^{(1)}-Z_{1}^{(0)}}{Z_{0}}=\frac{\Delta_{\sqrt{X}}^{(0)}}{v_{\mathrm{prep}}}+O(\varepsilon^{2}) − 2 Z 0 Y 1 ( 0 ) − Y 1 ( 1 ) + 2 Z 0 Y 0 + Z 0 Z 0 Z 1 ( 1 ) − Z 1 ( 0 ) = v prep Δ X ( 0 ) + O ( ε 2 ) again isolates the X \sqrt{X} X split without first-order SPAM bias.
Remark 2 (Alternating swap chains, MWPM decoding weights, and classical repetition-code failure excess). After an odd number k = 2 m + 1 k=2m+1 k = 2 m + 1 of untwirled swaps, the cumulative forward and backward transition error probabilities satisfy p ± ( k ) = p ˉ ( k ) ( 1 ± μ q / k ) + O ( k 2 ε 2 ) p_{\pm}^{(k)}=\bar{p}^{(k)}(1\pm\mu_{q}/k)+O(k^{2}\varepsilon^{2}) p ± ( k ) = p ˉ ( k ) ( 1 ± μ q / k ) + O ( k 2 ε 2 ) with μ q = ( χ + − χ − ) / ( χ + + χ − ) \mu_{q}=(\chi_{+}-\chi_{-})/(\chi_{+}+\chi_{-}) μ q = ( χ + − χ − ) / ( χ + + χ − ) . In a 3-bit repetition code (or within a single bit-flip syndrome chain of an XZZX surface code [32 , 33 , 34 ] ) protecting a balanced prior over ∣ 000 ⟩ |000\rangle ∣000 ⟩ and ∣ 111 ⟩ |111\rangle ∣111 ⟩ , the true majority-vote logical failure probability for p ± = p ˉ ( 1 ± δ r e l ) p_{\pm}=\bar{p}(1\pm\delta_{\mathrm{rel}}) p ± = p ˉ ( 1 ± δ rel ) (δ r e l = μ q / k \delta_{\mathrm{rel}}=\mu_{q}/k δ rel = μ q / k ) is
p L t r u e = 1 2 [ 3 p + 2 ( 1 − p + ) + p + 3 + 3 p − 2 ( 1 − p − ) + p − 3 ] = 3 p ˉ 2 ( 1 + δ r e l 2 ) − 2 p ˉ 3 ( 1 + 3 δ r e l 2 ) , (101) p_{L}^{\mathrm{true}}=\frac{1}{2}\left[3p_{+}^{2}(1-p_{+})+p_{+}^{3}+3p_{-}^{2}(1-p_{-})+p_{-}^{3}\right]=3\bar{p}^{2}(1+\delta_{\mathrm{rel}}^{2})-2\bar{p}^{3}(1+3\delta_{\mathrm{rel}}^{2}), \tag*{(101)} p L true = 2 1 [ 3 p + 2 ( 1 − p + ) + p + 3 + 3 p − 2 ( 1 − p − ) + p − 3 ] = 3 p ˉ 2 ( 1 + δ rel 2 ) − 2 p ˉ 3 ( 1 + 3 δ rel 2 ) , (101)
whereas the Pauli-twirled approximation replaces both p + p_{+} p + and p − p_{-} p − by p ˉ \bar{p} p ˉ , predicting p L t w i r l = 3 p ˉ 2 − 2 p ˉ 3 p_{L}^{\mathrm{twirl}}=3\bar{p}^{2}-2\bar{p}^{3} p L twirl = 3 p ˉ 2 − 2 p ˉ 3 . At leading order in p ˉ ≪ 1 \bar{p}\ll1 p ˉ ≪ 1 , Pauli twirling systematically underestimates the logical failure rate by the factor
p L t r u e p L t w i r l = 1 + μ q 2 k 2 + O ( p ˉ ) ≤ 1 + 8 π 2 q 2 k 2 , (102) \frac{p_{L}^{\mathrm{true}}}{p_{L}^{\mathrm{twirl}}}=1+\frac{\mu_{q}^{2}}{k^{2}}+O(\bar{p})\leq1+\frac{8}{\pi^{2}q^{2}k^{2}}, \tag*{(102)} p L twirl p L true = 1 + k 2 μ q 2 + O ( p ˉ ) ≤ 1 + π 2 q 2 k 2 8 , (102)
which reaches 1 + 8 π 2 ≈ 1.810569 1+\frac{8}{\pi^{2}}\approx1.810569 1 + π 2 8 ≈ 1.810569 (an 81.1% logical error excess) at k = q = 1 k=q=1 k = q = 1 for the chiral extremizer L + L_{+} L + . Assigning state-dependent directional log-likelihood branch weights w ± = ln [ ( 1 − p ± ) / p ± ] w_{\pm}=\ln[(1-p_{\pm})/p_{\pm}] w ± = ln [( 1 − p ± ) / p ± ] in minimum-weight perfect-matching (MWPM) and belief-propagation decoders eliminates this bias.
High-precision analytical gate audit and 1,203,200-bath random CPTP numerical verification
This appendix documents the complete 12-digit analytical and multi-start L-BFGS-B numerical optimization audit across all single-qubit gate families (Table 3 ) and the adversarial numerical verification of Theorem 1 and Theorem 2 across 1,203,200 random multi-jump CPTP Markovian baths (Table 4 ).
In the numerical audit (Table 4 ), 150,400 random multi-jump CPTP Lindblad generators D = ∑ α = 1 K γ α D [ L α ] \mathcal{D}=\sum_{\alpha=1}^{K}\gamma_{\alpha}\mathcal{D}[L_{\alpha}] D = ∑ α = 1 K γ α D [ L α ] (1 ≤ K ≤ 4 1\leq K\leq4 1 ≤ K ≤ 4 , normalized to Γ = 1 \Gamma=1 Γ = 1 ) were sampled for each of the 8 primary gate trajectories (1,203,200 random baths in total, deterministic random seeds 202601–202608). For each generator, the first-order slopes χ ± ( D ; r ) \chi_{\pm}(\mathcal{D};r) χ ± ( D ; r ) were evaluated by three independent methods: (i) the analytical two-moment trace formula (22) of Theorem 1, (ii) exact block-matrix superoperator Fréchet differentiation of exp ( L 0 + ε D ) \exp(\mathcal{L}_0+\varepsilon\mathcal{D}) exp ( L 0 + ε D ) at ε = 0 \varepsilon=0 ε = 0 , and (iii) two-point Richardson-extrapolated 4 × \times × 4 matrix exponentiation at ε 0 = 10 − 7 \varepsilon_0=10^{-7} ε 0 = 1 0 − 7 . Across all 1,203,200 random CPTP baths, the two-moment formula (22) agreed with exact superoperator Fréchet derivatives to machine precision (≤ 5.00 × 10 − 16 \leq5.00 \times10^{-16} ≤ 5.00 × 1 0 − 16 ), zero violations of κ ( r ) \kappa(r) κ ( r ) occurred, and multi-start L-BFGS-B optimization over the Stiefel manifold of jump operators converged to the analytical bound κ ( r ) = ( 1 − β ( r ) ) / ( 1 + β ( r ) ) \kappa(r)=(1-\beta(r))/(1+\beta(r)) κ ( r ) = ( 1 − β ( r )) / ( 1 + β ( r )) to all 12 significant digits in every case.