betaFeedback

WorksPhysics and Astronomy

Resilience of emergent Hamiltonian quantum memories in the presence of noise

Abstract

Emergent Hamiltonians can freeze an entangled state prepared through coherent evolution, since that state is an eigenstate of the Hamiltonian under which it is then stored. We show that the freedom left in this construction allows the storage spectrum to be adapted to a noisy environment without changing the stored state. For a product of Bell pairs prepared from a density wave in an engineered chain, a selection rule prevents local density noise from connecting the target to any of its degenerate partners, so that rescaling the storage Hamiltonian by a factor αα moves every transition this noise drives. Under 1/f1/f noise, the storage time then grows in proportion to αα, a result we confirm by evolving the memory under explicit classical 1/f1/f fields, whereas for an Ohmic bath the most favorable scale depends on its cutoff. Fermionic loss, treated as a solvable model, includes a transition that rescaling leaves at zero frequency, and an energy shift μμ per particle moves it away from zero, with an exact orbital solution giving the resulting lifetimes. Finally, among number-conserving quadratic reference Hamiltonians, αα and μμ are the only controls that keep the storage Hamiltonian nearest neighbor at every time along the preparation, so that the protocol combines local freezing with control over the storage spectrum.

Introduction

Preserving an entangled many-body state, a fundamental aspect of quantum technologies, requires keeping both its local properties and its correlations between distant parts of the system. Several approaches address aspects of this problem. For instance, many-body localization is argued to preserve a memory of the initial local occupations in the presence of sufficiently large disorder [1, 2, 3, 4], although this memory is mostly local and is eroded by coupling to an environment [5, 6, 7]. Quantum error correction, in turn, protects encoded information by identifying and correcting errors [8, 9, 10, 11], while other strategies build protection into the Hamiltonian itself, either by encoding information in decoherence-free subspaces [12, 13] or by separating the protected states from the rest of the spectrum with an energy gap [14, 15, 16]. In Ref. [17], we explored a complementary route based on emergent Hamiltonians: by evolving an initially simple state into an entangled state, one can freeze it by switching to a Hamiltonian for which it is an eigenstate. This protocol preserves the complete state under isolated evolution. It allowed us to prepare and store products of maximally entangled Bell pairs and globally entangled Greenberger-Horne-Zeilinger (GHZ) states.

The construction starts with a state ∣ψ0⟩|\psi_0\rangle that is an eigenstate of a reference Hamiltonian H^0\hat{H}_0. Evolution under a preparation Hamiltonian H^f\hat{H}_f produces ∣ψ(t)⟩=U^(t)∣ψ0⟩|\psi(t)\rangle= \hat{U}(t)|\psi_0\rangle, where U^(t)=e−iH^ft\hat{U}(t) = e^{-i\hat{H}_f t}. The emergent Hamiltonian [18, 19, 20, 21],

M^(t)=U^(t)H^0U^†(t),(1)\hat{M}(t) = \hat{U}(t)\hat{H}_0\hat{U}^{\dagger}(t), \tag*{(1)}

has ∣ψ(t)⟩|\psi(t)\rangle as an eigenstate with the same eigenvalue as ∣ψ0⟩|\psi_0\rangle under H^0\hat{H}_0. A quench from H^f\hat{H}_f to M^(t∗)\hat{M}(t_*) therefore freezes the state reached at any chosen time t∗t_*. More generally, the unitary conjugation in Eq. (1) preserves every eigenvalue and its degeneracy, so that the full storage spectrum is inherited from H^0\hat{H}_0. When one prepares a tensor product of Bell pairs between mirror-symmetric sites of a chain see Sec. III A and Fig. 1(a), this construction gives an exact storage Hamiltonian containing only onsite potentials and nearest-neighbor hopping at every preparation time [17]. Entanglement between distant sites can thus be prepared and frozen without couplings between them, making the protocol an interesting candidate for experimental emulation in platforms with limited connectivity, such as superconducting quantum circuits, where engineered, site-dependent nearest-neighbor couplings have already been used for quantum state transfer in chains [22] and in two-dimensional networks [23].

Nonetheless, practical applications must account for environmental effects, since a bath can drive the stored eigenstate to decay into other states. In Ref. [17], we described this decay with relaxation and dephasing rates fixed by the coherence times T1T_1 and T2T_2, which do not depend on the energies of the transitions they induce. In general, however, the operator that couples to the bath determines which transitions are possible, while their energy differences determine the noise strength they encounter [24, 25]. To lowest order in the coupling, the rate of each transition is the squared matrix element of the coupling operator multiplied by the noise spectrum at the corresponding Bohr frequency, as we illustrate in Figure 1(c) (and later show in Sec. II B). This frequency dependence therefore offers another use for the emergent-Hamiltonian protocol.

Figure 1. Preparing, freezing, and storing a Bell-pair memory. (a) A six-site density wave ∣DW⟩|{\rm DW}\rangle evolves under H^f\hat{H}_{f} into mirror Bell pairs ∣ψ⋆⟩|\psi_{\star}\rangle at t⋆=π/2t_{\star}=\pi/2. A quench to the storage Hamiltonian M^α,μ(t⋆)\hat{M}_{\alpha,\mu}(t_{\star}) freezes this state, which is then coupled to independent local baths through n^ℓ\hat{n}_{\ell} or c^ℓ\hat{c}_{\ell}. (b) One-particle storage orbitals of M^α,μ(t⋆)\hat{M}_{\alpha,\mu}(t_{\star}), with energies εp=αp+μ\varepsilon_{p}=\alpha p+\mu. The Bell target fills the even orbitals. A density operator moves one fermion from an even to an odd orbital, so that ∣Ωf∣≥α|\Omega_{f}|\geq\alpha (blue arrows). Loss removes a fermion from orbital pp and releases αp+μ\alpha p+\mu (orange arrow), which vanishes for p=0p=0 when μ=0\mu=0. (c) Density-noise transition weights for L=10L=10 and α=1,2,4\alpha=1,2,4. Each bar sums the squared matrix elements over sites and over all transitions out of the target at one signed frequency Ωf\Omega_{f}, and is placed at ∣ω∣=∣Ωf∣|\omega|=|\Omega_{f}|. Upward (downward) bars denote energy released (absorbed) by the memory. The 1/f1/f and zero-temperature Ohmic curves, on separate vertical scales, show the spectra sampled by these transitions. Open circles mark the spectral values at the transition frequencies, and the dotted line marks the Ohmic cutoff ωc=10\omega_{c}=10.

Since the storage spectrum is inherited from H^0\hat{H}_0, different reference Hamiltonians sharing the initial state as an eigenstate freeze the same prepared state while placing its transitions at different frequencies. In particular, we noted in Ref. [17] that the gaps inherited from H^0\hat{H}_0 can be tuned to suppress coherent perturbations of the stored state. Here we show that the same freedom also controls its incoherent decay when the noise depends on frequency.

Realistic noise spectra are indeed far from flat. Low-frequency 1/f1/f noise is ubiquitous in solid-state qubits [26]. In superconducting circuits, it is a dominant source of dephasing [27, 28], and classical fluctuating fields are widely used to describe it [29]. At higher frequencies, charge-qubit measurements also reveal a noise spectrum growing approximately linearly with frequency [30]. Changing a transition energy can therefore increase or decrease its rate, even when the state and the environment are left fixed. The same principle underlies dynamical decoupling [29, 31], the control of decay by modulating the system-bath coupling [32], and noise spectroscopy with driven qubits [28, 33]. Moving a transition toward a stronger part of the spectrum can likewise accelerate decay, as in the anti-Zeno effect [34]. In those schemes, time-dependent control sets the frequency window the qubits sample. Here, the storage Hamiltonian sets it statically, and the shift acts on a many-body entangled state without changing it.

In what follows, we assume ideal coherent preparation and examine this freedom during storage. In Sec. II, we review how the Bell pairs are prepared and frozen, introduce the rescaled storage Hamiltonian αM^(t⋆)\alpha\hat{M}(t_{\star}), which multiplies every Bohr frequency by α\alpha, and show that local density noise, which acts on qubits as local dephasing, cannot connect the Bell target to its degenerate partners, so that rescaling moves every transition it drives

[Fig. 1(b)]. Section III shows that, under 1/f1/f noise, the storage time then grows in proportion to α\alpha, a result we confirm by evolving the memory under explicit classical 1/f1/f fields, whereas an Ohmic bath favors a different scale. Particle loss, which we treat for fermions as a solvable model rather than as a description of qubit relaxation, changes the particle number and is therefore sensitive to an energy shift μ\mu per particle, which moves a zero-frequency loss transition away from zero [Fig. 1(b)]. We also show that, among number-conserving quadratic reference Hamiltonians, α\alpha and μ\mu are the only controls compatible with nearest-neighbor storage along the whole preparation. Section IV summarizes our findings and contrasts the Bell memory with the GHZ memory of Ref. [17]. Technical details are compiled in the Appendices, and throughout this work we set ℏ=kB=1\hbar=k_{\mathrm{B}}=1.

Bell-pair storage and noise-induced transitions

Preparing and freezing the Bell pairs

Consider an open chain of LL sites, with LL even, occupied by hard-core bosons, where the operator a^ℓ\hat{a}_{\ell} annihilates a boson at site ℓ=0,…,L−1\ell=0,\ldots,L-1 and n^ℓ=a^ℓ†a^ℓ\hat{n}_{\ell}=\hat{a}_{\ell}^{\dagger}\hat{a}_{\ell} is the local occupation. Through the standard mapping between hard-core bosons and spin-1/21/2 operators [35], each site can be identified with a qubit. As in Ref. [17], we prepare the entangled state with

H^f=∑ℓ=0L−2Jℓ(a^ℓ†a^ℓ+1+a^ℓ+1†a^ℓ),Jℓ=12ℓ+1)(L−ℓ−1).(2)\hat{H}_{f}=\sum_{\ell=0}^{L-2}J_{\ell}\left(\hat{a}_{\ell}^{\dagger}\hat{a}_{\ell+1}+\hat{a}_{\ell+1}^{\dagger}\hat{a}_{\ell}\right),\qquad J_{\ell}=\frac{1}{2}\sqrt{\ell+1)(L-\ell-1)}. \tag*{(2)}

The Jordan–Wigner transformation [36, 37] maps this one-dimensional model to free spinless fermions with annihilation operators c^ℓ\hat{c}_{\ell}, leaving the occupations n^ℓ\hat{n}_{\ell} unchanged. In the one-particle sector, the hopping matrix has exactly the matrix elements of S^x\hat{S}_{x} for a spin s=(L−1)/2s=(L-1)/2, with site ℓ\ell corresponding to Sz=ℓ−sS_{z}=\ell-s, so that each fermion evolves as a large spin precessing about the xx axis. The same engineered hopping underlies perfect state transfer [38, 39, 40].

The initial state is the half-filled density wave ∣DW⟩=∣1010⋯10⟩\lvert\mathrm{DW}\rangle=\lvert1010\cdots10\rangle, which is an eigenstate of the reference Hamiltonian

H^0=∑ℓ=0L−1ℓn^ℓ.(3)\hat{H}_{0}=\sum_{\ell=0}^{L-1}\ell\hat{n}_{\ell}. \tag*{(3)}

Its time evolution under H^f\hat{H}_{f} can therefore be frozen at any time using Eq. (1). In particular, at t⋆=π/2t_{\star}=\pi/2 the state becomes, up to an overall phase,

∣ψ⋆⟩=∏ℓ=0L/2−1∣10⟩ℓ,L−1−ℓ−i∣01⟩ℓ,L−1−ℓ2,(4)\lvert\psi_{\star}\rangle=\prod_{\ell=0}^{L/2-1}\frac{\lvert10\rangle_{\ell,L-1-\ell}-i\lvert01\rangle_{\ell,L-1-\ell}}{\sqrt{2}}, \tag*{(4)}

a product of maximally entangled pairs between mirror sites [17, 41]. The same concentric pairing, up to the phases within each pair, becomes the ground state of nearest-neighbor chains whose couplings decay rapidly away from the center, where it was introduced as an example of volume-law entanglement [42] and is known as the rainbow state [43]. Here, it is instead reached dynamically and stored as an excited eigenstate. Quenching to M^(t⋆)\hat{M}(t_{\star}) freezes these pairs, and the closed form in Appendix A shows that this step requires only onsite terms and nearest-neighbor hopping, even though the pairs extend across the whole chain.

Since the conjugation in Eq. (1) fixes the eigenstates of the storage Hamiltonian but not its energy scale, we are free to use

M^α(t⋆)=αM^(t⋆),α>0,(5)\hat{M}_{\alpha}(t_{\star})=\alpha\hat{M}(t_{\star}),\qquad\alpha>0, \tag*{(5)}

which leaves the Bell state and every other storage eigenstate unchanged while multiplying all energy differences by α\alpha. A second control, the energy shift μ\mu per particle that also appears in Figure 1, leaves every transition that conserves the particle number unchanged, and we therefore introduce it only in Sec. III C, where particle loss makes it necessary.

Coupling to the environment

To see how the storage scale α\alpha affects the memory, we couple each local density to an independent, identical environment Figure 1(a). Since n^ℓ=(1+σ^ℓz)/2\hat{n}_{\ell}=(1+\hat{\sigma}_{\ell}^{z})/2, this corresponds to local dephasing noise in the qubit language, although in the basis of storage eigenstates it mostly drives transitions at nonzero frequencies. We describe the resulting dynamics by the Davies master equation [44, 45, 46],

dρ^dt=−i[M^α(t⋆),ρ^]+∑ℓ,ωS(ω)D[n^ℓ(ω)]ρ^,\frac{d\hat{\rho}}{dt}=-i[\hat{M}_{\alpha}(t_{\star}),\hat{\rho}]+\sum_{\ell,\omega}S(\omega)\mathcal{D}[\hat{n}_{\ell}(\omega)]\hat{\rho},
D[O^]ρ^:=O^ρ^O^†−12{O^†O^,ρ^}.(6)\mathcal{D}[\hat{O}]\hat{\rho}:=\hat{O}\hat{\rho}\hat{O}^{\dagger}-\frac{1}{2}\{\hat{O}^{\dagger}\hat{O},\hat{\rho}\}. \tag*{(6)}

where tt now measures the time elapsed after the quench, with ρ^(0)=∣ψ⋆⟩⟨ψ⋆∣\hat{\rho}(0)=\lvert\psi_{\star}\rangle\langle\psi_{\star}\rvert. The operator n^ℓ(ω)\hat{n}_{\ell}(\omega) collects all density operators at site ℓ\ell that release energy ω\omega, and S(ω)S(\omega) is the corresponding bath rate. A positive (negative) frequency denotes energy released (absorbed) by the memory. This description retains the energy dependence of the bath while allowing us to follow the full density matrix, and the frequency components and the assumptions behind the master equation are detailed in Appendix B.

Let E⋆E_{\star} and EfE_{f} be the energies of the target and of an orthogonal storage eigenstate ∣f⟩\lvert f\rangle, separated by the transition frequency Ωf=E⋆−Ef\Omega_{f}=E_{\star}-E_{f}. (6) then gives the initial rate at which ∣f⟩\lvert f\rangle is populated,

ddt⟨f∣ρ^(t)∣f⟩∣t=0=S(Ωf)∑ℓ∣⟨f∣n^ℓ∣ψ⋆⟩∣2.(7)\left.\frac{d}{dt}\langle f\lvert\hat{\rho}(t)\rvert f\rangle\right|_{t=0}=S(\Omega_{f})\sum_{\ell}\lvert\langle f\lvert\hat{n}_{\ell}\rvert\psi_{\star}\rangle\rvert^{2}. \tag*{(7)}

The energy difference thus selects a frequency in the bath spectrum, while the matrix elements determine how strongly the bath couples the two states. Figure 1(c) groups these matrix elements by frequency and shows that changing α\alpha moves the frequencies while leaving the matrix elements fixed. Rescaling cannot, however, move a transition between degenerate states away from Ωf=0\Omega_{f}=0.

Density noise and degenerate partners

An example of that is the Bell target ∣ψ⋆⟩\lvert\psi_{\star}\rangle, which has orthogonal states at the same many-body energy within its particle-number sector for L≥6L\geq6. Whether these states cause direct decay depends on the density matrix elements. The one-particle eigenstates of M^α(t⋆)\hat{M}_{\alpha}(t_{\star}), which we call storage orbitals, make those elements simple to understand. Orbital p=0,…,L−1p=0,\ldots,L-1 is obtained by evolving site state ∣p⟩\lvert p\rangle under U^(t⋆)\hat{U}(t_{\star}) and has energy αp\alpha p. Let d^p†\hat{d}_{p}^{\dagger} create a fermion in this orbital. The density wave initially occupies the even sites, so the Bell target occupies the even-labeled storage orbitals Figure 1(b).

A local density n^ℓ=c^ℓ†c^ℓ\hat{n}_{\ell}=\hat{c}_{\ell}^{\dagger}\hat{c}_{\ell} is a number-conserving quadratic operator. More generally, any A^=∑ijhijc^i†c^j\hat{A}=\sum_{ij}h_{ij}\hat{c}_{i}^{\dagger}\hat{c}_{j}, with arbitrary coefficients hijh_{ij}, can move only one fermion between orbitals. Apart from its component along the target itself, A^∣ψ⋆⟩\hat{A}|\psi_\star\rangle therefore contains only states obtained by replacing an occupied even orbital pp with an empty odd orbital jj, and the energy released (or absorbed) in this substitution is Ωf=α(p−j)\Omega_f=\alpha(p-j). Since p−jp-j is an odd integer, it cannot vanish, and its smallest possible magnitude is one. Thus ∣Ωf∣≥α|\Omega_f|\geq\alpha for every such transition. As a result, a different state with the same particle number and the same energy must change at least two occupied orbitals, which a single quadratic operator cannot do. Consequently,

⟨m∣A^∣ψ⋆⟩=0for Em=E⋆,if ∣m⟩⊥∣ψ⋆⟩.(8)\langle m|\hat{A}|\psi_\star\rangle=0 \quad\text{for } E_m=E_\star,\quad\text{if } |m\rangle\perp|\psi_\star\rangle. \tag*{(8)}

The target therefore has no direct zero-frequency decay under local density noise. Rescaling α\alpha is thus sufficient to move every density transition that takes it to a different state, despite the many-body degeneracy.

Tuning the storage spectrum of the Bell memory

Density noise with 1/f and Ohmic spectra

Since every density transition out of the Bell target can be moved by rescaling, we now keep the preparation time at t⋆=π/2t_\star=\pi/2 and examine how the storage scale α\alpha, which in practice is bounded by the largest couplings a given device can reach, affects the memory under two bath spectra, summarized in Figure 2. We first consider a symmetric bath rate of strength κ\kappa that decreases as 1/∣ω∣η1/|\omega|^\eta, with exponent η>0\eta>0, above a low-frequency cutoff ωIR\omega_{\mathrm{IR}},

Figure 2. Bell memory under local density noise. In (a,c), solid curves show the target fidelity FF and dashed curves the mean mirror-pair concurrence C‾\overline{C}. For symmetric 1/f1/f noise, (a) shows L=10L=10 at α=0.5,1,2,4\alpha=0.5,1,2,4, and (b) shows the times to reach F=0.95F=0.95, normalized to α=1\alpha=1, at L=8,10,12,14L=8,10,12,14. The gray line is the guide y=αy=\alpha. For the zero-temperature Ohmic bath, (c) shows L=10L=10 at ωc=10\omega_c=10 and α=1,2,4,8\alpha=1,2,4,8, and (d) shows lifetime ratios at ωc=5,10,20\omega_c=5,10,20 for L=10L=10 (solid) and 1414 (dashed). The 1/f1/f parameters are η=1\eta=1, ωIR=10−4\omega_{\mathrm{IR}}=10^{-4} and κ=π10−3\kappa=\pi10^{-3}, and the Ohmic coupling is gO2=10−3g_{\mathrm{O}}^2=10^{-3}.

S1/f(ω)=κmax⁡(∣ω∣,ωIR)η.(9)S_{1/f}(\omega)=\frac{\kappa}{\max(|\omega|,\omega_{\mathrm{IR}})^\eta}. \tag*{(9)}

As indicated by Eq. (8), this rate, although largest near zero frequency, cannot drive the Bell target into a degenerate partner, and the remaining transitions, which occur at ∣Ωf∣≥α|\Omega_f|\geq\alpha, encounter weaker noise as α\alpha increases. We follow both the complete stored state, through the target fidelity F(t)=⟨ψ⋆∣ρ^(t)∣ψ⋆⟩F(t)=\langle\psi_\star|\hat{\rho}(t)|\psi_\star\rangle, and the entanglement within its mirror pairs, through the mean concurrence C‾\overline{C} defined in Appendix B 1, which averages the two-qubit entanglement over the pairs. For η=1\eta= 1, used in all 1/f1/f data shown here, both quantities decay more slowly as α\alpha is increased [Fig. 2(a)]. The time τF,0.95\tau_{F,0.95} at which the fidelity reaches F=0.95F = 0.95 grows in proportion to α\alpha for all sizes in Figure 2(b), consistent with the scaling S1/f(αω)=α−ηS1/f(ω)S_{1/f}(\alpha\omega) = \alpha^{-\eta}S_{1/f}(\omega) obeyed by every rate out of the target, whose transitions all lie far above ωIR\omega_{\mathrm{IR}}. The details of the density-noise calculation are given in Appendix B 2.

Increasing an energy scale need not always improve storage, however, and to illustrate how the bath spectrum determines the outcome, we also consider a zero-temperature Ohmic rate,

SO(ω)=gO2ωe−ω/ωcΘ(ω),(10)S_{\mathrm{O}}(\omega) = g_{\mathrm{O}}^{2}\omega e^{-\omega/\omega_{c}}\Theta(\omega), \tag*{(10)}

where gO2g_{\mathrm{O}}^{2} sets the coupling strength, ωc\omega_c is the cutoff, and Θ\Theta is the Heaviside function. At zero temperature, the bath can absorb energy released by the memory but cannot supply energy to it, and its rate grows with frequency up to ωc\omega_c and decays beyond it. Moving the Bell transitions through this spectrum can therefore either accelerate or suppress the decay [Fig. 2(c)]. Below the cutoff, increasing α\alpha pushes the transitions toward stronger noise and accelerates the decay, so that storage improves only once most of them lie beyond ωc\omega_c. For ωc=20\omega_c = 20, the lifetimes at all displayed α>1\alpha> 1 remain shorter than at α=1\alpha= 1, whereas for the lower cutoff ωc=5\omega_c = 5 the largest scales shown yield longer storage [Fig. 2(d)]. The same Bell state can thus benefit from different storage scales, depending on the spectrum of the bath.

Storage under classical 1/f fields

We next test the storage-scale control beyond the Davies description, by evolving the memory under explicit time-dependent 1/f1/f fields, a standard description of low-frequency dephasing [28, 29]. For each realization, the Bell state evolves under

H^ξ(t)=M^α(t)+∑ℓξℓ(t)n^ℓ,(11)\widehat{H}_{\xi}(t) = \widehat{M}_{\alpha}(t) + \sum_{\ell}\xi_{\ell}(t)\hat{n}_{\ell}, \tag*{(11)}

where the real fields ξℓ(t)\xi_{\ell}(t) at different sites are independent. Each field is built from a spectral density Pξ(ν)\mathcal{P}_{\xi}(\nu), specified at positive frequencies, as a sum of cosines with sampled frequencies νq>0\nu_q > 0 with integration widths Δνq\Delta\nu_q and phases ϕℓq\phi_{\ell q} drawn independently and uniformly from [0,2π)[0,2\pi),

ξℓ(t)=∑q2Pξ(νq)Δνqcos⁡(νqt+ϕℓq).(12)\xi_{\ell}(t) = \sum_q \sqrt{2\mathcal{P}_{\xi}(\nu_q)\Delta\nu_q}\cos(\nu_q t + \phi_{\ell q}). \tag*{(12)}

We take Pξ(ν)=Aϵ2/νη\mathcal{P}_{\xi}(\nu) = A_{\epsilon}^{2}/\nu^{\eta} with η=1\eta= 1 inside the sampled band, which corresponds to κ=πAϵ2\kappa= \pi A_{\epsilon}^{2} in Eq. (9), so that both descriptions refer to the same noise strength. The frequency range and the evolution procedure are given in Appendix B 3.

In the interaction picture, the matrix elements of n^ℓ\hat{n}_{\ell} between storage eigenstates acquire phases at the Bohr frequencies, so that field components near these frequencies drive transitions most efficiently. The Davies equation averages the effect of the fluctuations into fixed transition rates at those frequencies, whereas the explicit fields retain their temporal correlations, including the slow, nearly static components that dominate 1/f1/f noise.

Figure 3 compares the evolution averaged over 4000 field realizations with the Davies dynamics for L=6L = 6. As in Sec. III A, increasing α\alpha slows the loss of both fidelity and mirror-pair entanglement [Figure 3(a) and 3(b)]. At short times, the explicit fields produce a faster, partly oscillatory drop of the fidelity that the Davies description does not capture, and this transient shrinks as α\alpha increases. Beyond it, the fidelity decays approximately exponentially, appearing as nearly straight lines on the logarithmic axis of Figure 3(a), at rates that differ from the Davies prediction by less than 3% for every α\alpha. Fits over the shaded interval also give the same dependence on the storage scale in both descriptions, with rates that decrease close to, though not exactly as, 1/α1/\alpha [Figure 3(c)]. The improvement under 1/f1/f noise therefore persists when the memory evolves under explicit time-dependent fluctuations.

Figure 3. Bell storage under classical 1/f1/f fields and the Davies master equation. Results are for L=6L = 6 and α=0.5,1,2,4\alpha= 0.5,1,2,4, and the classical means use 4000 independent realizations per α\alpha. (a) Target fidelity on a logarithmic vertical axis: solid curves with standard-error bands show the classical-field means, and dashed curves show the Davies results. The fitting interval 50≤t≤20050 \le t \le200 is shaded. (b) Mean mirror-pair concurrence: markers show the classical results with bootstrap standard-error bars, and dashed curves show the Davies results. The two-site density matrices are averaged over realizations before concurrence is calculated. (c) Decay rates Γ(α)\Gamma(\alpha) from straight-line fits to −ln⁡F(t)-\ln F(t) over the shaded interval in (a), normalized to the rate at α=1\alpha= 1 in each description. Crosses with bootstrap standard-error bars show classical fields, and open circles show Davies dynamics. The field parameters are Aϵ2=10−3A_{\epsilon}^{2} = 10^{-3}, corresponding to κ=π10−3\kappa= \pi10^{-3} as in Figure 2, η=1\eta= 1, ωIR=10−4\omega_{\mathrm{IR}} = 10^{-4}, and an upper frequency cutoff of 32.

Fermionic loss and the energy shift μ\mu

The density-noise results rely on the selection rule of Sec. II C, which keeps every transition out of the Bell target at a nonzero frequency. We now consider a channel for which this is not the case, namely linear fermionic loss with coupling operator c^ℓ\hat{c}_{\ell} at each site. Removing a fermion from storage orbital pp changes the particle number and releases the orbital energy αp\alpha p [Figure 1(b)]. Unlike the energy differences probed by density noise, this energy is measured from the zero of the one-particle energies, which the reference Hamiltonian of Eq. (3) places on site ℓ=0\ell= 0. Since the density wave occupies this site, the Bell target fills orbital p=0p = 0, and loss has a zero-frequency transition that no choice of α\alpha can move.

This zero follows from our choice of reference Hamiltonian, not from a property of the Bell target. The density wave is also an eigenstate of αH^0+μN^\alpha\hat{H}_{0} + \mu\hat{\mathcal{N}} for any α\alpha and μ\mu, where N^=∑ℓn^ℓ\hat{\mathcal{N}} = \sum_{\ell}\hat{n}_{\ell} is the total particle number, and since H^f\hat{H}_{f} conserves N^\hat{\mathcal{N}}, Eq. (1) turns this reference Hamiltonian into

M^α,μ(t∗)=αM^(t∗)+μN^.(13)\hat{M}_{\alpha,\mu}(t_{*}) = \alpha\hat{M}(t_{*}) + \mu\hat{\mathcal{N}}. \tag*{(13)}

The Bell target and the storage orbitals are unchanged, while the orbital energies become ϵp=αp+μ\epsilon_{p} = \alpha p + \mu. Density noise cannot detect μ\mu, since the shift cancels in the energy difference between any two states with the same particle number. Removing a fermion, however, now releases αp+μ\alpha p + \mu, which moves the formerly zero-frequency transition away from zero. The complementary density wave ∣0101⋯01⟩\lvert0101\cdots01\rangle makes the same point. It also evolves into mirror Bell pairs but fills the odd orbitals, so that at μ=0\mu= 0 its smallest removal energy is α\alpha, whereas a bath able to add fermions would fill the empty orbital p=0p = 0 at zero frequency. In both cases, μ\mu is what keeps the transitions that change the particle number away from zero frequency.

Since the storage Hamiltonian is quadratic and the loss is linear, the problem is exactly solvable [47, 48], and the identical, independent baths simplify it further. Written in terms of the storage orbitals, the local loss operators combine into a single loss channel per orbital, with rate Γp−=Sloss(ϵp)\Gamma^{-}_{p} = S_{\mathrm{loss}}(\epsilon_{p}) set by the removal rate of the bath at the released energy (see Appendix § 0.2.3 for details). The unoccupied orbitals then empty independently, and since the target survives only if all of them keep their fermion, its fidelity is

F(t)=exp⁡[−t∑p evenΓp].(14)F(t)=\exp\left[-t\sum_{p\ {\rm even}}\Gamma_p\right]. \tag*{(14)}

The storage energies therefore enter only through where the occupied orbitals sample the spectrum of the bath, and for a frequency-independent removal rate neither α\alpha nor μ\mu would change the lifetime.

Figure 4 shows the resulting rates and lifetimes for the same decreasing form as Eq. (9), Sloss(ω)=κ/max⁡(∣ω∣,ωIR)ηS_{\mathrm{loss}}(\omega)=\kappa/\max(|\omega|,\omega_{\mathrm{IR}})^\eta with η=1\eta=1. At μ=0\mu=0, orbital p=0p=0 is removed at zero frequency, with rate κ/ωIRη\kappa/\omega_{\mathrm{IR}}^\eta, which exceeds all other removal rates by about four orders of magnitude [Figure 4(a)]. The lifetime is then set by the infrared cutoff alone and changes with neither α\alpha nor LL [Figure 4(b) and 4(c)]. Any μ>0\mu>0 lifts this orbital, together with all the others, to a weaker part of the spectrum, after which the lifetime no longer depends on ωIR\omega_{\mathrm{IR}}. The gain relative to μ=0\mu=0 therefore grows as the cutoff is lowered, and at L=100L=100, α=1\alpha=1 and ωIR=10−4\omega_{\mathrm{IR}}=10^{-4} it amounts to a factor of about 4.4×1034.4\times10^3 for μ=2\mu=2 and 6.2×1036.2\times10^3 for μ=5\mu=5. Once μ>0\mu>0, increasing α\alpha lengthens storage further by moving the higher orbitals to weaker noise [Figure 4(b)]. The dependence on system size is weak as well [Figure 4(c)]. Since ∑p even1/(αp+μ)\sum_{p\ {\rm even}}1/(\alpha p+\mu) grows only logarithmically with LL, the lifetime for μ=2\mu=2 decreases by a factor of three between L=10L=10 and L=1000L=1000, whereas a frequency-independent removal rate would reduce it in proportion to 1/L1/L, by a factor of one hundred. All of these gains follow from changing the storage energies alone, with the preparation, the Bell target and the bath held fixed.

Figure 4. Bell memory under fermionic loss. (a) Removal rates Γp−\Gamma^{-}_{p} of the occupied (even) storage orbitals at L=100L = 100 and α=1\alpha= 1, for μ=0,2,5\mu= 0, 2, 5. (b) Time τF,0.95\tau_{F,0.95} to reach F=0.95F = 0.95 versus α\alpha at L=100L = 100. (c) The same time versus LL at α=1\alpha= 1. The dotted line shows the 1/L1/L decrease expected for a frequency-independent removal rate. The removal rate is Sloss(ω)=κ/max⁡(∣ω∣,ωIR)ηS_{\mathrm{loss}}(\omega) = \kappa/\max(|\omega|,\omega_{\mathrm{IR}})^{\eta}, with η=1\eta= 1, ωIR=10−4\omega_{\mathrm{IR}} = 10^{-4} and κ=π10−7\kappa= \pi10^{-7}. Since all lifetimes scale as 1/κ1/\kappa, the value of κ\kappa only sets the unit of time.

We stress that fermionic loss serves here as a solvable model, and its conclusions should not be transferred directly to the relaxation of qubits. In the hard-core boson, or qubit, language, local loss is described by a^ℓ=P^ℓc^ℓ\hat{a}_\ell=\hat{P}_\ell\hat{c}_\ell rather than by c^ℓ\hat{c}_\ell [see Eq. (17)]. Since the Jordan–Wigner string is unitary, the total loss rate out of the target is unchanged, ∑ℓ∥a^ℓ∣ψ⋆⟩∥2=∑ℓ∥c^ℓ∣ψ⋆⟩∥2\sum_\ell\|\hat{a}_\ell|\psi_\star\rangle\|^2=\sum_\ell\|\hat{c}_\ell|\psi_\star\rangle\|^2, but its distribution over frequencies is not. Whereas c^ℓ\hat{c}_\ell removes a single storage orbital at energy αp+μ\alpha p+\mu, the string also rearranges the remaining particles, so that a^ℓ\hat{a}_\ell connects the target to many final states with different energies. As a result, the orbital solution of Eq. (14) no longer holds. Since every final state still has one particle fewer than the target, μ\mu shifts all loss frequencies by the same amount, but some of these final states lie above the target in storage energy, so that at μ=0\mu=0 the qubit loss spectrum extends to negative frequencies, down to −3α-3\alpha for L=6L=6 and −6α-6\alpha for L=8L=8. A shift that clears zero frequency for fermions thus therefore need not clear it for qubits, and at α=1\alpha=1 and μ=2\mu=2 about 4% and 1% of the qubit loss weight, for L=6L=6 and L=8L=8, is still at zero frequency.

In particular, in superconducting circuits a uniform term μN^\mu\hat{\mathcal{N}} is already present in the laboratory frame through the qubit frequency, so that relaxation samples the bath near that frequency rather than near zero. The loss results therefore illustrate how μ\mu controls a zero-frequency channel when the coupling removes one storage orbital at a time, as for fermions, rather than predicting qubit lifetimes.

Completeness of the local storage family

So far, we have used α\alpha to move the density transitions and μ\mu to move the zero-frequency loss transition, and it is natural to ask whether other reference Hamiltonians offer further control while keeping the storage Hamiltonian local. Since a^H^0+μN^\hat{a}\hat{H}_0+\mu\hat{\mathcal{N}} has the density wave as an eigenstate for any α\alpha and μ\mu, the family of Eq. (13) extends to every preparation time,

M^α,μ(t⋆)=αU^(t⋆)H^0U^†(t⋆)+μN^.(15)\hat{M}_{\alpha,\mu}(t_\star)=\alpha\hat{U}(t_\star)\hat{H}_0\hat{U}^\dagger(t_\star)+\mu\hat{\mathcal{N}}. \tag*{(15)}

As shown in Ref. [17], the emergent Hamiltonian U^(t⋆)H^0U^†(t⋆)\hat{U}(t_\star)\hat{H}_0\hat{U}^\dagger(t_\star) contains only onsite potentials and nearest-neighbor hopping at every preparation time [see Eq. (19)], and multiplying these terms by α\alpha and adding μN^\mu\hat{\mathcal{N}} preserves this locality. A single reference Hamiltonian can therefore freeze, with local couplings, whichever state the preparation is stopped at, not only the Bell product, so that the storage step does not have to be redesigned for each target. We now show that, among number-conserving quadratic reference Hamiltonians, requiring locality at every preparation time leaves only α\alpha and μ\mu.

Let H^ref\hat{H}_{\mathrm{ref}} be such a reference, and let hrefh_{\mathrm{ref}} and hfh_f be the one-particle matrices of H^ref\hat{H}_{\mathrm{ref}} and H^f\hat{H}_f, with (hf)ℓ,ℓ+1=(hf)ℓ+1,ℓ=Jℓ(h_f)_{\ell,\ell+1}=(h_f)_{\ell+1,\ell}=J_\ell. At t=0t=0 the emergent Hamiltonian is the reference itself, which must therefore be nearest neighbor and have the density wave as an eigenstate. Since neighboring sites of the density wave have different occupations, any hopping term in H^ref\hat{H}_{\mathrm{ref}} removes a particle from an occupied site to an empty neighbor and produces an occupation string that no other term can cancel, so that only site potentials survive, H^ref=∑ℓvℓn^ℓ\hat{H}_{\mathrm{ref}}=\sum_\ell v_\ell\hat{n}_\ell. During the preparation, H^f\hat{H}_f turns these potentials into hopping. To first order in tt, the one-particle matrix of the emergent Hamiltonian acquires the term −it[hf,href]-it[h_f,h_{\mathrm{ref}}], which connects only neighboring sites, with amplitudes Jℓ(vℓ+1−vℓ)J_\ell(v_{\ell+1}-v_\ell), while to second order it acquires −(t2/2)[hf,[hf,href]]-(t^2/2)[h_f,[h_f,h_{\mathrm{ref}}]], which also connects sites two bonds apart, with

[hf,[hf,href]]ℓ,ℓ+2=JℓJℓ+1(vℓ+2−2vℓ+1+vℓ).(16)[h_f,[h_f,h_{\mathrm{ref}}]]_{\ell,\ell+2}=J_\ell J_{\ell+1}(v_{\ell+2}-2v_{\ell+1}+v_\ell). \tag*{(16)}

Since every coupling JℓJ_\ell is nonzero, locality at all times requires this second difference of the potentials to vanish, so that the potentials change by the same amount α\alpha across every bond. Writing vℓ=μ+αℓv_\ell=\mu+\alpha\ell then gives H^ref=αH^0+μN^\hat{H}_{\mathrm{ref}}=\alpha\hat{H}_0+\mu\hat{\mathcal{N}}. Conversely, every reference of this form is local at all times by Eq. (19), so that Eq. (15) is the complete family of local storage Hamiltonians, with the remaining details given in Appendix A2.

Locality at every preparation time is a stronger condition than the Bell memory alone requires. At the Bell point, every nearest-neighbor quadratic Hamiltonian with mirror-symmetric potentials and with hopping amplitudes that change sign under the reflection ℓ→L−1−ℓ\ell\to L-1-\ell keeps the Bell target as an eigenstate, a family with 3L/2−13L/2-1 real parameters (Appendix A 2). Which local Hamiltonians share a given eigenstate is a question of broad interest [49, 50, 51, 52], and these additional parameters would allow more general storage spectra for the Bell memory, a possibility we do not explore here. Within the family of Eq. (15), the preparation time t⋆t_\star selects the state to be stored, while α\alpha and μ\mu set its storage energies without changing that state or introducing longer-range couplings.

Conclusion

Emergent Hamiltonians freeze a prepared state by making it an eigenstate, and we have shown that the freedom left in this construction can be used to adapt the memory to its environment. For the density wave evolved under the engineered chain, every state reached during the preparation can be frozen with onsite potentials and nearest-neighbor hopping, and among number-conserving quadratic references that achieve this at every preparation time, only an energy scale α\alpha and an energy shift μ\mu per particle remain. For the Bell memory, a selection rule lets α\alpha move every transition that local density noise drives, which under 1/f1/f noise lengthens storage in proportion to α\alpha, while μ\mu moves the zero-frequency transition of fermionic loss to weaker noise. Since both controls act on the storage energies alone, with the prepared state and the environment held fixed, they can complement other means of protecting entanglement.

The Bell target owes this behavior to its position inside a degenerate spectrum, where the selection rule, rather than an energy gap, keeps its transitions away from zero frequency. The GHZ memory of Ref. [17], which we analyze in Appendix § 0.3.4, illustrates the opposite situation. There, one-axis twisting prepares the GHZ state from a spin product, and the target becomes the unique ground state of a storage Hamiltonian whose terms act on all LL spins, so that an equilibrium bath at zero temperature cannot excite it. At finite temperature, rescaling then suppresses thermal absorption across a gap that depends on the noise operator. Local σ^z\hat{\sigma}_z noise connects the target only to storage eigenstates with one spin changed, at energy α\alpha, whereas excitation exchange with the bath changes all spins and connects it to states at energies α(L−1)\alpha(L-1) and α\alpha, so that larger systems keep the GHZ state for a shorter time under the first coupling and for a longer time under the second. The same spectral freedom therefore applies, but through the familiar mechanism of gap protection [14, 15] and at the cost of locality.

The structure of the storage family also has a natural reading from the perspective of Hamiltonian learning. A local Hamiltonian reconstructed from a single eigenstate is typically determined only up to its overall scale and to conserved quantities with zero variance in that state, such as the particle number [51, 52]. Our two controls play exactly these roles, with α\alpha setting the scale and μ\mu multiplying the particle number, although this minimal freedom comes from requiring locality during the whole preparation rather than from the Bell target alone (Sec. IIID). Since the robustness of such reconstructions has recently been proposed as a probe of quantum chaos and ergodicity [52], it would be interesting to understand how much freedom remains for storing states prepared by interacting dynamics, for which the emergent Hamiltonian is generally nonlocal and only approximate local forms are available [17].

Acknowledgments

Numerical calculations were performed on the Carya and Sabine clusters at the University of Houston’s Research Computing Data Core. This work also used ACES at Texas A&M High Performance Research Computing through allocation PHY240046 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by U.S. National Science Foundation grants 2138259, 2138286, 2138307, 2137603, and 2138296. R.M. acknowledges that the research was funded in part by the Robert A. Welch Foundation, Grant #L-E-0001-19921203.

The data and analysis scripts supporting this work are available on Zenodo [53].

Exact local storage Hamiltonian

This appendix collects the exact results on which the storage Hamiltonian of the main text rests. Appendix A 1 derives the closed form of the emergent Hamiltonian at every preparation time, which contains only onsite potentials and nearest-neighbor hopping, and which Sec. II A uses to freeze the Bell product. Appendix A 2 completes the proof of Sec. IIID that α\alpha and μ\mu are the only controls that preserve this locality at every preparation time, and characterizes the larger family of local Hamiltonians that share the Bell target at the Bell point alone.

Exact local form

We first review the noiseless preparation of the Bell product and the closed form of its emergent Hamiltonian. For the open chain, the Jordan–Wigner transformation [36, 37] is

a^ℓ=P^ℓc^ℓ,\hat{a}_{\ell}=\hat{P}_{\ell}\hat{c}_{\ell},
P^ℓ=exp⁡(iπ∑m<ℓn^m)=∏m<ℓ(1−2n^m).(17)\hat{P}_{\ell}=\exp\left(i\pi\sum_{m<\ell}\hat{n}_{m}\right)=\prod_{m<\ell}(1-2\hat{n}_{m}). \tag*{(17)}

The parity strings cancel between neighboring sites, a^ℓ+1†a^ℓ=c^ℓ+1†c^ℓ\hat{a}_{\ell+1}^{\dagger}\hat{a}_{\ell}=\hat{c}_{\ell+1}^{\dagger}\hat{c}_{\ell}, and n^ℓ=c^ℓ†c^ℓ\hat{n}_{\ell}=\hat{c}_{\ell}^{\dagger}\hat{c}_{\ell}, which gives the free-fermion form of Eq. (2).

All operators used here are quadratic, A^=∑ℓmAℓmc^ℓ†c^m\hat{A}=\sum_{\ell m}A_{\ell m}\hat{c}_{\ell}^{\dagger}\hat{c}_{m}, so that their commutators, and their conjugation by U^(t)\hat{U}(t), follow from those of the one-particle matrices AA. We denote by S^a\hat{S}_{a} the quadratic operator built in this way from the spin-ss matrix SaS_{a}, with s=(L−1)/2s=(L-1)/2 and site ℓ\ell corresponding to Sz=ℓ−sS_{z}=\ell-s. The engineered couplings JℓJ_{\ell} are the matrix elements of SxS_{x} in this basis [38, 39], so that H^f=S^x\hat{H}_{f}=\hat{S}_{x} and H^0=sN^+S^z\hat{H}_{0}=s\hat{N}+\hat{S}_{z}. Rotation about the xx axis then gives

U^(t)S^zU^†(t)=cos⁡tS^z−sin⁡tS^y,(18)\hat{U}(t)\hat{S}_{z}\hat{U}^{\dagger}(t)=\cos t\hat{S}_{z}-\sin t\hat{S}_{y}, \tag*{(18)}

and in the site basis the emergent Hamiltonian becomes [17]

M^(t)=∑ℓ=0L−1[s+(ℓ−s)cos⁡t]n^ℓ+isin⁡t∑ℓ=0L−2Jℓ(a^ℓ+1†a^ℓ−a^ℓ†a^ℓ+1).(19)\hat{M}(t)=\sum_{\ell=0}^{L-1}[s+(\ell-s)\cos t]\hat{n}_{\ell} +i\sin t\sum_{\ell=0}^{L-2}J_{\ell}(\hat{a}_{\ell+1}^{\dagger}\hat{a}_{\ell}-\hat{a}_{\ell}^{\dagger}\hat{a}_{\ell+1}). \tag*{(19)}

Only onsite potentials and nearest-neighbor (purely imaginary, yet Hermitian) hopping appear, at every preparation time. At the Bell point, t=π/2t=\pi/2, the potential becomes uniform and the hopping reaches its largest amplitude,

M^(π/2)=sN^−S^y.(20)\hat{M}(\pi/2)=s\hat{N}-\hat{S}_{y}. \tag*{(20)}

Completeness of the local family

Here we give the details of the argument in Sec. III D, for a Hermitian, number-conserving quadratic reference H^ref=∑ab(href)abc^a†c^b\hat{H}_{\mathrm{ref}}=\sum_{ab}(h_{\mathrm{ref}})_{ab}\hat{c}_{a}^{\dagger}\hat{c}_{b}, with the preparation Hamiltonian of Eq. (2) and the initial density wave held fixed. Since conjugation by U^(t)\hat{U}(t) maps quadratic operators to quadratic operators, the emergent Hamiltonian has the coefficient matrix

u(t)hrefu†(t)=href−it[hf,href]−t22[hf,[hf,href]]+O(t3),(21)u(t)h_{\mathrm{ref}}u^{\dagger}(t)=h_{\mathrm{ref}}-it[h_{f},h_{\mathrm{ref}}]-\frac{t^{2}}{2}[h_{f},[h_{f},h_{\mathrm{ref}}]]+O(t^{3}), \tag*{(21)}

with u(t)=e−ihftu(t)=e^{-ih_{f}t}. If this matrix is tridiagonal for every tt, so is each of its Taylor coefficients. For the site potentials left by the eigenstate condition, href=diag⁡(v0,…,vL−1)h_{\mathrm{ref}}=\operatorname{diag}(v_{0},\ldots,v_{L-1}), the first commutator has the elements

[hf,href]ab=(hf)ab(vb−va)[h_{f},h_{\mathrm{ref}}]_{ab}=(h_{f})_{ab}(v_{b}-v_{a}), which vanish unless aa and bb are neighbors. Applying the commutator once more, sites ℓ\ell and ℓ+2\ell+2 are connected only through site ℓ+1\ell+1, which gives Eq. (16) for ℓ=0,…,L−3\ell=0,\ldots,L-3. Conversely, since [H^f,N^]=0[\hat{H}_{f},\hat{N}]=0, the reference αH^0+μN^\alpha\hat{H}_{0}+\mu\hat{N} has the emergent Hamiltonian αM^(t)+μN^\alpha\hat{M}(t)+\mu\hat{N}, which keeps the evolved density wave as an eigenstate and, by Eq. (19), remains nearest neighbor at every tt.

Locality at a single preparation time is a weaker condition. At the Bell point, consider the one-particle operator K=u(π/2)Πu†(π/2)K=u(\pi/2)\Pi u^{\dagger}(\pi/2), with Π=diag⁡[(−1)ℓ]\Pi=\operatorname{diag}[(-1)^{\ell}]. Since Π\Pi equals +1+1 on the sites occupied by the density wave and −1-1 on the empty ones, the occupied storage orbitals span the +1+1 eigenspace of KK, and a quadratic Hamiltonian can be Bell target if and only if its one-particle matrix mm commutes with KK. Because u(π/2)u(\pi/2) rotates SzS_{z} into −Sy-S_{y}, KK is a rotation by π\pi about the yy axis, up to a phase, and therefore connects only mirror sites, KL−1−ℓ,ℓ∝(−1)ℓK_{L-1-\ell,\ell}\propto(-1)^{\ell}. For a tridiagonal mm, the condition [m,K]=0[m,K]=0 requires mirror-symmetric potentials, mℓℓ=mL−1−ℓ,L−1−ℓm_{\ell\ell}=m_{L-1-\ell,L-1-\ell}, and hopping amplitudes that change sign under reflection, mL−1−a,L−1−b=−mabm_{L-1-a,L-1-b}=-m_{ab} for neighboring aa and bb. This leaves L/2L/2 potentials, one complex amplitude for each of the L/2−1L/2-1 pairs of mirror bonds, and a purely imaginary hopping across the central bond, for a total of 3L/2−13L/2-1 real parameters. The storage family of Eq. (13), with uniform potential and imaginary hopping proportional to JℓJ_{\ell} [Eq. (A4)], is the two-parameter subfamily that remains local at all preparation times.

Noise models and calculations

This appendix specifies the density-noise calculations behind Figure 2 and Figure 3, so that each result can be reproduced. We first show how the Davies equation is built from the frequency components of the coupling operators, how it gives the transition rates of Eq. (7), and why its zero-frequency part leaves the Bell target unchanged. Appendix B1 then defines the mirror-pair concurrence, Appendix B2 gives the construction and parameters of the Davies evolution, and Appendix B3 describes the classical 1/f1/f fields that test it.

Let Π^E\hat{\Pi}_{E} project onto the complete eigenspace at energy EE of the storage Hamiltonian. The frequency component of a coupling operator A^\hat{A} is

A^(ω):=∑E′−E=ωΠ^EA^Π^E′.(22)\hat{A}(\omega):=\sum_{E^{\prime}-E=\omega}\hat{\Pi}_{E}\hat{A}\hat{\Pi}_{E^{\prime}}. \tag*{(22)}

It lowers the system energy by ω\omega, and transitions with the same energy difference are summed before forming the dissipator D\mathcal{D} of Eq. (6). The Davies equation describes weakly coupled relaxation on time scales long compared with the decay of bath correlations, over which terms oscillating at differences of distinct Bohr frequencies average out [44, 54]. Since 1/f1/f noise has slowly decaying correlations, Sec. III B tests this description against explicit classical fields. For the initial target projector ρ^⋆=∣ψ⋆⟩⟨ψ⋆∣\hat{\rho}_{\star}=|\psi_{\star}\rangle\langle\psi_{\star}| and an orthogonal energy eigenstate ∣f⟩|f\rangle, the Hamiltonian term of the population derivative vanishes because the target is an eigenstate, and the anticommutator terms vanish because ⟨f∣ψ⋆⟩=0\langle f|\psi_{\star}\rangle=0. The remaining contribution is

⟨f∣D[n^ℓ(ω)]ρ^⋆∣f⟩=∣⟨f∣n^ℓ(ω)∣ψ⋆⟩∣2=δω,Ωf∣⟨f∣n^ℓ∣ψ⋆⟩∣2.(23)\langle f|\mathcal{D}[\hat{n}_{\ell}(\omega)]\hat{\rho}_{\star}|f\rangle=|\langle f|\hat{n}_{\ell}(\omega)|\psi_{\star}\rangle|^{2} =\delta_{\omega,\Omega_{f}}|\langle f|\hat{n}_{\ell}|\psi_{\star}\rangle|^{2}. \tag*{(23)}

Summing the bath rates gives Eq. (7). The density selection rule also implies

n^ℓ(0)∣ψ⋆⟩=⟨ψ⋆∣n^ℓ∣ψ⋆⟩∣ψ⋆⟩.(24)\hat{n}_{\ell}(0)|\psi_{\star}\rangle=\langle\psi_{\star}|\hat{n}_{\ell}|\psi_{\star}\rangle|\psi_{\star}\rangle. \tag*{(24)}

Thus the zero-frequency density dissipator leaves the target projector unchanged.

Mirror-pair concurrence

For each mirror pair (ℓ,L−1−ℓ)(\ell,L-1-\ell), we trace out the other sites to obtain its physical two-site density matrix ρ^ℓ,L−1−ℓ\hat{\rho}_{\ell,L-1-\ell}, whose concurrence is [55, 56]

Cℓ,L−1−ℓ=max⁡{0,λ1−λ2−λ3−λ4},C‾=2L∑ℓ=0L/2−1Cℓ,L−1−ℓ.(25)C_{\ell,L-1-\ell}=\max\{0,\lambda_{1}-\lambda_{2}-\lambda_{3}-\lambda_{4}\}, \qquad \overline{C}=\frac{2}{L}\sum_{\ell=0}^{L/2-1}C_{\ell,L-1-\ell}. \tag*{(25)}

where λ1≥⋯≥λ4\lambda_{1}\ge\cdots\ge\lambda_{4} are the square roots of the eigenvalues of ρ^ℓ,L−1−ℓ(σ^y⊗σ^y)ρ^ℓ,L−1−ℓ∗(σ^y⊗σ^y)\hat{\rho}_{\ell,L-1-\ell}(\hat{\sigma}_{y}\otimes\hat{\sigma}_{y})\hat{\rho}_{\ell,L-1-\ell}^{*}(\hat{\sigma}_{y}\otimes\hat{\sigma}_{y}), with the complex conjugation taken in the two-site occupation basis. For the classical fields of Appendix B 3, the two-site matrices are averaged over realizations before each concurrence is calculated.

Bell density noise: Davies evolution

The density-noise results of Figure 2 follow from Eq. (6) with the 1/f1/f rate of Eq. (9), for η=1\eta=1, κ=π 10−3\kappa=\pi\,10^{-3} and ωIR=10−4\omega_{\mathrm{IR}}=10^{-4}, or with the zero-temperature Ohmic rate of Eq. (10), for gO2=10−3g_{\mathrm{O}}^{2}=10^{-3} and the cutoffs given in the figure. The first rate allows transitions in both energy directions, whereas the second only lets the memory release energy.

Since density coupling preserves the particle number, we work in the half-filled sector of the storage-orbital basis. A basis state with NN particles has energy α∑p∈occp+μN\alpha\sum_{p\in\mathrm{occ}}p+\mu N, so that at N=L/2N=L/2 all states with the same sum of occupied labels form a complete energy eigenspace. We construct the projectors onto these eigenspaces, form each n^ℓ(ω)\hat{n}_{\ell}(\omega) using Eq. (22), and evolve ρ^(0)=ρ^⋆\hat{\rho}(0)=\hat{\rho}_{\star} under the resulting generator, which gives the fidelity and the reduced two-site density matrices. We set μ=0\mu=0, since it cancels from all energy differences at fixed particle number. The times τF,0.95\tau_{F,0.95} in Figure 2(b) are first downward crossings of F=0.95F=0.95, interpolated between consecutive recorded times.

Bell storage under classical fields

Each independent field realization evolves according to

iddt∣ψξ(t)⟩=[M^α(t⋆)+∑ℓξℓ(t)n^ℓ]∣ψξ(t)⟩,i\frac{\mathrm{d}}{\mathrm{d}t}|\psi_{\xi}(t)\rangle= \left[\hat{M}_{\alpha}(t_{\star})+\sum_{\ell}\xi_{\ell}(t)\hat{n}_{\ell}\right]|\psi_{\xi}(t)\rangle,
ρ^(t)=Eξ[∣ψξ(t)⟩⟨ψξ(t)∣],(26)\hat{\rho}(t)=\mathbb{E}_{\xi}[|\psi_{\xi}(t)\rangle\langle\psi_{\xi}(t)|], \tag*{(26)}

with ∣ψξ(0)⟩=∣ψ⋆⟩|\psi_{\xi}(0)\rangle=|\psi_{\star}\rangle. The random-phase fields of Eq. (12) use the spectral density

Pξ(ν)={Aϵ2/νη,ωIR≤ν≤32,0,otherwise.(27)\mathcal{P}_{\xi}(\nu)= \begin{cases} A_{\epsilon}^{2}/\nu^{\eta}, & \omega_{\mathrm{IR}}\le\nu\le32,\\ 0, & \text{otherwise}. \end{cases} \tag*{(27)}

where ν>0\nu>0, Aϵ2=10−3A_{\epsilon}^{2}=10^{-3}, η=1\eta=1 and ωIR=10−4\omega_{\mathrm{IR}}=10^{-4}. The upper limit exceeds every frequency at which density noise drives transitions for L=6L=6 and α≤4\alpha\le4, which is at most 5α=205\alpha=20. The continuum covariance of these fields is

E[ξℓ(t+s)ξm(t)]=δℓm∫0∞Pξ(ν)cos⁡(νs) dν.(28)\mathbb{E}[\xi_{\ell}(t+s)\xi_{m}(t)]=\delta_{\ell m}\int_{0}^{\infty}\mathcal{P}_{\xi}(\nu)\cos(\nu s)\,\mathrm{d}\nu. \tag*{(28)}

and its Fourier transform is

∫−∞∞eiωsE[ξℓ(t+s)ξℓ(t)] ds=πPξ(∣ω∣).(29)\int_{-\infty}^{\infty}e^{i\omega s}\mathbb{E}[\xi_{\ell}(t+s)\xi_{\ell}(t)]\,\mathrm{d}s=\pi\mathcal{P}_{\xi}(|\omega|). \tag*{(29)}

Within the sampled band, these fields therefore correspond to the Davies rate of Eq. (9) with κ=πAϵ2\kappa=\pi A_{\epsilon}^{2}.

We use logarithmic frequency bins with 40 bins per decade below 0.050.05, and uniform bins of width 0.010.01 from 0.050.05 to 3232. The frequencies νq\nu_{q} in Eq. (12) are the geometric midpoints of the logarithmic bins and the arithmetic midpoints of the uniform ones, with Δνq\Delta\nu_{q} the bin width, and the phases are drawn independently for every site and frequency. Each realization is evolved with time step Δt=0.02\Delta t=0.02, using half a step under the storage Hamiltonian, a full step under the onsite noise evaluated at the midpoint, and another storage half-step.

For each α=0.5,1,2,4\alpha=0.5,1,2,4, we average 4000 independent L=6L=6 realizations. Figure 3(c) compares straight-line fits to −ln⁡F(t)-\ln F(t) over 50≤t≤20050\le t\le200, normalizing each description by its own α=1\alpha=1 rate. The classical error bars are standard errors from 4000 bootstrap resamples of whole realizations.

Independent orbital dynamics for fermionic loss

This appendix derives the target fidelity of Eq. (14), used for fermionic loss in Sec. III C and Figure 4. Appendix C1 shows that, for identical and independent baths, local loss acts on the stored state as a single loss channel for each storage orbital, and Appendix C2 solves for the orbital occupations and combines them into the fidelity, also allowing for baths that add particles.

From local loss to orbital loss

Let UℓpU_{\ell p} be the amplitude at site ℓ\ell of the normalized storage orbital pp, which has energy εp=αp+μ\varepsilon_p=\alpha p+\mu. The corresponding fermion operators satisfy

d^p†=∑ℓUℓpc^ℓ†,c^ℓ=∑pUℓpd^p,\hat{d}_p^\dagger=\sum_{\ell}U_{\ell p}\hat{c}_{\ell}^\dagger,\qquad\hat{c}_{\ell}=\sum_p U_{\ell p}\hat{d}_p,
∑ℓUℓp∗Uℓq=δpq.(30)\sum_{\ell}U_{\ell p}^*U_{\ell q}=\delta_{pq}. \tag*{(30)}

In this basis,

M^α,μ(t⋆)=∑pεpd^p†d^p,∣ψ⋆⟩=∏p evend^p†∣0⟩,(31)\hat{M}_{\alpha,\mu}(t_\star)=\sum_p\varepsilon_p\hat{d}_p^\dagger\hat{d}_p,\qquad|\psi_\star\rangle=\prod_{p\ {\rm even}}\hat{d}_p^\dagger|0\rangle, \tag*{(31)}

where the product is ordered by increasing pp. Since [M^α,μ,d^p]=−εpd^p[\hat{M}_{\alpha,\mu},\hat{d}_p]=-\varepsilon_p\hat{d}_p, the component of local loss that releases energy ω\omega is

c^ℓ(ω)=∑p: εp=ωUℓpd^p.(32)\hat{c}_{\ell}(\omega)=\sum_{p:\,\varepsilon_p=\omega}U_{\ell p}\hat{d}_p. \tag*{(32)}

Since the orbital energies are distinct for α≠0\alpha\ne0, each frequency selects one orbital, c^ℓ(εp)=Uℓpd^p\hat{c}_{\ell}(\varepsilon_p)=U_{\ell p}\hat{d}_p, and this single operator includes every many-body transition that removes a fermion from orbital pp. For independent sites with the same removal rate S−(ω)≡Sloss(ω)S_-(\omega)\equiv S_{\rm loss}(\omega), the normalization of the orbital then gives

S−(εp)∑ℓD[Uℓpd^p]ρ^=Γp−D[d^p]ρ^,S_-(\varepsilon_p)\sum_{\ell}\mathcal{D}[U_{\ell p}\hat{d}_p]\hat{\rho}=\Gamma_p^-\mathcal{D}[\hat{d}_p]\hat{\rho},
Γp−:=S−(εp).(33)\Gamma_p^-:=S_-(\varepsilon_p). \tag*{(33)}

If the bath can also add particles, with rate S+(ω)S_+(\omega) for a process that releases energy ω\omega from the memory, the rate for adding a fermion to orbital pp is Γp+:=S+(−εp)\Gamma_p^+:=S_+(-\varepsilon_p), and the master equation becomes

dρ^dt=−i[M^α,μ(t⋆),ρ^]+∑pΓp−D[d^p]ρ^+∑pΓp+D[d^p†]ρ^.(34)\frac{d\hat{\rho}}{dt}=-i[\hat{M}_{\alpha,\mu}(t_\star),\hat{\rho}]+\sum_p\Gamma_p^-\mathcal{D}[\hat{d}_p]\hat{\rho}+\sum_p\Gamma_p^+\mathcal{D}[\hat{d}_p^\dagger]\hat{\rho}. \tag*{(34)}

Occupation equations and target fidelity

Define n^p=d^p†d^p\hat{n}_p=\hat{d}_p^\dagger\hat{d}_p and np(t)=Tr⁡[ρ^(t)n^p]n_p(t)=\operatorname{Tr}[\hat{\rho}(t)\hat{n}_p]. The Hamiltonian in Eq. (34) commutes with n^p\hat{n}_p, and so do d^q\hat{d}_q and d^q†\hat{d}_q^\dagger for q≠pq\ne p, so that jumps in the other orbitals do not change npn_p. For jumps in orbital pp, the fermion algebra gives

D†[d^p](n^p)=−n^p,D†[d^p†](n^p)=1−n^p,(35)\mathcal{D}^\dagger[\hat{d}_p](\hat{n}_p)=-\hat{n}_p,\qquad\mathcal{D}^\dagger[\hat{d}_p^\dagger](\hat{n}_p)=1-\hat{n}_p, \tag*{(35)}

where the adjoint dissipator acts on an observable X^\hat{X} as D†[O^](X^)=O^†X^O^−{O^†O^,X^}/2\mathcal{D}^\dagger[\hat{O}](\hat{X})=\hat{O}^\dagger\hat{X}\hat{O}-\{\hat{O}^\dagger\hat{O},\hat{X}\}/2. Taking the expectation value yields

n˙p=−Γp−np+Γp+(1−np).(36)\dot{n}_p=-\Gamma_p^-n_p+\Gamma_p^+(1-n_p). \tag*{(36)}

Thus, for rp:=Γp−+Γp+>0r_p:=\Gamma_p^-+\Gamma_p^+>0,

np(t)=Γp+rp+[np(0)−Γp+rp]e−rpt.(37)n_p(t)=\frac{\Gamma_p^+}{r_p}+\left[n_p(0)-\frac{\Gamma_p^+}{r_p}\right]e^{-r_pt}. \tag*{(37)}

An orbital with rp=0r_p=0 retains its initial occupation. The target is an orbital occupation state, with np⋆=1n_p^\star=1 for even pp and np⋆=0n_p^\star=0 for odd pp, and Eq. (34) keeps its density matrix diagonal in this basis. A loss jump changes only np=1n_p=1 to 0, at rate Γp−\Gamma_p^-, and a gain jump only np=0n_p=0 to 1, at rate Γp+\Gamma_p^+. Since neither rate depends on the occupations of the other orbitals, the initially fixed occupations remain independent random variables, and the probability of matching the target in orbital pp is

qp(t)=np⋆np(t)+(1−np⋆)[1−np(t)],(38)q_p(t)=n_p^\star n_p(t)+(1-n_p^\star)[1-n_p(t)], \tag*{(38)}

and the target fidelity is the probability of matching in every orbital,

F(t)=∏pqp(t).(39)F(t)=\prod_p q_p(t). \tag*{(39)}

For pure loss, with Γp+=0\Gamma_p^+=0, empty orbitals remain empty and occupied ones survive with probability e−Γp−te^{-\Gamma_p^-t}, which gives

F(t)=exp⁡[−t∑p evenΓp−],τf=−ln⁡f∑p evenΓp−.(40)F(t)=\exp\left[-t\sum_{p\ {\rm even}}\Gamma_p^-\right],\qquad\tau_f=\frac{-\ln f}{\sum_{p\ {\rm even}}\Gamma_p^-}. \tag*{(40)}

where τf\tau_f is the time to reach fidelity 0<f<10<f<1. Figure 4 evaluates these expressions with Γp−=Sloss(αp+μ)\Gamma_p^-=S_{\rm loss}(\alpha p+\mu) and the parameters given in its caption, with τF,0.95=τ0.95\tau_{F,0.95}= \tau_{0.95}.

GHZ storage under thermal noise

This appendix supports the comparison between the Bell memory and the GHZ memory of Ref. [17] made in Sec. IV. For the GHZ target, rescaling the storage Hamiltonian protects the memory through an energy gap rather than through a selection rule, and we show that the size of this gap, and hence what α\alpha achieves, depends on how the noise couples to the spins. Appendix D1 describes the storage Hamiltonian and its spectrum, Appendix D2 the two thermal couplings and the resulting fidelities in Figure 5, and Appendix D3 the master equations and the quantum trajectories used to compute them.

Figure 5. GHZ memory under thermal noise, for local σ^z\hat{\sigma}^{z} coupling (top row) and excitation exchange (bottom row). (a), (c) Target fidelity for α=0.5\alpha= 0.5, 11, 1.51.5 and 22, at L=8L = 8 for σ^z\hat{\sigma}^{z} coupling and L=4L = 4 for exchange. (b), (d) Dependence on system size at α=1\alpha= 1, for L=4L = 4, 66 and 88. Curves are means over 150 quantum trajectories, and bands show standard errors. The independent Ohmic baths have T=1T = 1, ωc=5\omega_c = 5 and gO2=0.02g_{\mathrm{O}}^2 = 0.02, and the time axis for exchange is logarithmic.

Storage Hamiltonian and spectrum

For LL spins, let S^a=12∑ℓσ^ℓa\hat{S}_a=\frac{1}{2}\sum_{\ell}\hat{\sigma}_\ell^a be the collective spin, with σ^ℓa\hat{\sigma}_\ell^a the Pauli operator at site ℓ\ell along a=x,y,za=x,y,z, not to be confused with the quadratic operators S^a\hat{S}_a of Appendix A1. Starting from the product state ∣y−⟩⊗L|y_-\rangle^{\otimes L}, with ∣y±⟩=(∣0⟩±i∣1⟩)/2|y_\pm\rangle=(|0\rangle\pm i|1\rangle)/\sqrt{2}, one-axis twisting under H^f=\hat{H}_f= −S^z2-\hat{S}_z^2 [57] reaches the GHZ state at t⋆=π/2t_\star= \pi/2. Taking S^y\hat{S}_y as the reference Hamiltonian, which has this product as its unique ground state, the construction of Ref. [17] gives

U^G=eiπS^z2/2,∣ψG⟩=U^G∣y−⟩⊗L,\hat{U}_{G} = e^{i\pi\hat{S}_z^2/2}, \qquad|\psi_G\rangle= \hat{U}_{G}|y-\rangle^{\otimes L},
M^G(α)=αU^GS^yU^G†.(41)\hat{M}_{G}(\alpha) = \alpha\hat{U}_{G}\hat{S}_y\hat{U}_{G}^{\dagger}. \tag*{(41)}

For even LL, the target is, up to an overall phase,

∣ψG⟩=12[∣y−⟩⊗L−i(−1)L/2∣y+⟩⊗L].(42)|\psi_G\rangle= \frac{1}{\sqrt{2}}\left[|y-\rangle^{\otimes L} - i(-1)^{L/2}|y+\rangle^{\otimes L}\right]. \tag*{(42)}

The storage eigenstates follow by applying U^G\hat{U}_{G} to the product eigenstates of S^y\hat{S}_y, and a product with rr spins in ∣y+⟩|y+\rangle gives an eigenstate of energy α(r−L/2)\alpha(r-L/2). The GHZ target, with r=0r = 0, is therefore the unique ground state of M^G(α)\hat{M}_{G}(\alpha) for α>0\alpha> 0, and every spin changed relative to it costs an energy α\alpha. Unlike the Bell storage Hamiltonian, however, M^G(α)\hat{M}_{G}(\alpha) is not local, since conjugation by U^G\hat{U}_{G} attaches to each single-spin term of S^y\hat{S}_y a string of σ^z\hat{\sigma}^{z} operators on all other spins, so that every term of the storage Hamiltonian acts on all LL spins.

Thermal noise and storage fidelity

We compare local σ^z\hat{\sigma}^{z} noise with a coupling that exchanges spin excitations with the bath through σ^ℓ±=(σ^ℓx±iσ^ℓy)/2\hat{\sigma}_{\ell}^{\pm} = (\hat{\sigma}_{\ell}^{x} \pm i\hat{\sigma}_{\ell}^{y})/2. Since the target is the unique ground state, a bath at zero temperature cannot excite it, and we therefore take independent Ohmic baths at temperature TT, with the spectral function J(ω)=gO2ωe−ω/ωcJ(\omega) = g_{\mathrm{O}}^2\omega e^{-\omega/\omega_c} of Eq. (10) for ω>0\omega> 0 and the Bose occupation nB(ω)=(eω/T−1)−1n_{\mathrm{B}}(\omega) = (e^{\omega/T}-1)^{-1}. Emission and absorption then occur with the rates

ST(ω)={J(ω)[1+nB(ω)],ω>0,gO2T,ω=0,J(∣ω∣)nB(∣ω∣),ω<0.(43)S_T(\omega) = \begin{cases} J(\omega)[1+n_{\mathrm{B}}(\omega)], & \omega> 0,\\ g_{\mathrm{O}}^2T, & \omega= 0,\\ J(|\omega|)n_{\mathrm{B}}(|\omega|), & \omega< 0. \end{cases} \tag*{(43)}

where the zero-frequency value is the limit of J(ω)nB(ω)J(\omega)n_{\mathrm{B}}(\omega) as ω→0+\omega\to0^{+}. As in the main text, positive ω\omega denotes energy released by the memory, so that thermal excitation uses the rate at negative frequency.

The two couplings reach very different parts of the spectrum. Since σ^z\hat{\sigma}^{z} commutes with U^G\hat{U}_{G} and exchanges ∣y+⟩|y+\rangle and ∣y−⟩|y-\rangle at site ℓ\ell, it changes a single spin of the storage eigenstates and connects the target only to states at energy α\alpha or above it. Conjugation by U^G\hat{U}_{G} instead attaches a σ^z\hat{\sigma}^{z} string to σ^ℓ±\hat{\sigma}_{\ell}^{\pm}, which changes every other spin as well, so that exchange connects the target only to states at energies α(L−1)\alpha(L-1) and αL\alpha L. Rescaling therefore suppresses thermal absorption for both couplings, but across a gap α\alpha for σ^z\hat{\sigma}^{z} noise and across gaps that grow with LL for exchange.

We follow the fidelity FGHZ(t)=⟨ψG∣ρ^(t)∣ψG⟩F_{\mathrm{GHZ}}(t) = \langle\psi_G|\hat{\rho}(t)|\psi_G\rangle for T=1T = 1, ωc=5\omega_c = 5 and gO2=0.02g_{\mathrm{O}}^2 = 0.02. Increasing $\alpha lengthens storage for both couplings Figure 5(a) and Figure 5(c), since the absorption rate J(Δ)nB(Δ)J(\Delta)n_{\mathrm{B}}(\Delta) decreases with the gap △\triangle. The dependence on system size, however, is opposite for the two couplings Figure 5(b) and Figure 5(d). For σ^z\hat{\sigma}^{z} noise, a larger system offers more sites through which the bath can excite the memory at the same energy, and the fidelity falls sooner, whereas for exchange the gap grows with LL, which reduces thermal absorption and gives longer storage over the sizes and times shown. In both cases the preparation and the GHZ target remain fixed, and α\alpha changes only the energy that the bath must supply.

Davies evolution and quantum trajectories

For both couplings, the memory starts in ∣ψG⟩⟨ψG∣\lvert\psi_G\rangle\langle\psi_G\rvert and evolves under M^G(α)\hat{M}_G(\alpha), with frequency components formed as in Eq. (22). For local σ^ℓz\hat{\sigma}^{z}_{\ell} coupling, the Davies equation is

dρ^dt=−i[M^G(α),ρ^]+∑ℓ,ωST(ω)D[σ^ℓz(ω)]ρ^.(44)\frac{d\hat{\rho}}{dt}=-i[\hat{M}_G(\alpha),\hat{\rho}]+\sum_{\ell,\omega}S_T(\omega)D[\hat{\sigma}^{z}_{\ell}(\omega)]\hat{\rho}. \tag*{(44)}

For excitation exchange, the components σ^ℓ−(Δ)\hat{\sigma}^{-}_{\ell}(\Delta) with Δ≥0\Delta\geq0 lower the energy of the memory as it gives an excitation to the bath, and their adjoints describe absorption, so that

dρ^dt=−i[M^G(α),ρ^]+∑ℓ,Δ>0ST(Δ)D[σ^ℓ−(Δ)]ρ^+∑ℓ,Δ>0ST(−Δ)D[σ^ℓ−(Δ)†]ρ^+gO2T∑ℓ{D[σ^ℓ−(0)]+D[σ^ℓ−(0)†]}ρ^.(45)\begin{aligned} \frac{d\hat{\rho}}{dt}={}&-i[\hat{M}_G(\alpha),\hat{\rho}]\\ &+\sum_{\ell,\Delta>0}S_T(\Delta)D[\hat{\sigma}^{-}_{\ell}(\Delta)]\hat{\rho}\\ &+\sum_{\ell,\Delta>0}S_T(-\Delta)D[\hat{\sigma}^{-}_{\ell}(\Delta)^\dagger]\hat{\rho}\\ &+g_O^2T\sum_{\ell}\{D[\hat{\sigma}^{-}_{\ell}(0)]+D[\hat{\sigma}^{-}_{\ell}(0)^\dagger]\}\hat{\rho}. \tag*{(45)} \end{aligned}

As for the Bell memory, each frequency component includes complete degenerate eigenspaces, and different sites act as independent channels.

We solve both equations with quantum trajectories, also known as the Monte Carlo wave-function method [58, 59, 60, 61, 62], in which pure states evolve under a non-Hermitian effective Hamiltonian interrupted by random quantum jumps, and the average over trajectories reproduces the density matrix of the master equation. We diagonalize M^G(α)\hat{M}_G(\alpha) in the full 2L2^L-dimensional spin space, take as jump operators the frequency components above, each multiplied by the square root of its rate, and draw the jump times from the decay of the norm under the effective Hamiltonian, so that the fidelity is the trajectory average of ∣⟨ψG∣ϕ(t)⟩∣2\lvert\langle\psi_G|\phi(t)\rangle\rvert^2. At L=4L=4, these averages agree with direct evolution of the density matrix within statistical error.

References

References

  1. [1]R. Nandkishore and D. A. Huse, Many-body localization and thermalization in quantum statistical mechanics, Annu. Rev. Condens. Matter Phys. 6, 15 (2015).DOI
  2. [2]D. A. Abanin, E. Altman, I. Bloch, and M. Serbyn, Colloquium: Many-body localization, thermalization, and entanglement, Rev. Mod. Phys. 91, 021001 (2019).
  3. [3]M. Schreiber, S. S. Hodgman, P. Bordia, H. P. Lüschen, M. H. Fischer, R. Vosk, E. Altman, U. Schneider, and I. Bloch, Observation of many-body localization of interacting fermions in a quasirandom optical lattice, Science 349, 842 (2015).DOI
  4. [4]J. Smith, A. Lee, P. Richerme, B. Neyenhuis, P. W. Hess, P. Hauke, M. Heyl, D. A. Huse, and C. Monroe, Many-body localization in a quantum simulator with programmable random disorder, Nat. Phys. 12, 907 (2016).DOI
  5. [5]E. Levi, M. Heyl, I. Lesanovsky, and J. P. Garrahan, Robustness of many-body localization in the presence of dissipation, Phys. Rev. Lett. 116, 237203 (2016).DOI
  6. [6]M. H. Fischer, M. Maksymenko, and E. Altman, Dynamics of a many-body-localized system coupled to a bath, Phys. Rev. Lett. 116, 160401 (2016).DOI
  7. [7]H. P. Lüschen, P. Bordia, S. S. Hodgman, M. Schreiber, S. Sarkar, A. J. Daley, M. H. Fischer, E. Altman, I. Bloch, and U. Schneider, Signatures of many-body localization in a controlled open quantum system, Phys. Rev. X 7, 011034 (2017).
  8. [8]P. W. Shor, Scheme for reducing decoherence in quantum computer memory, Phys. Rev. A 52, R2493 (1995).pubmed.ncbi.nlm.nih.gov/9912632
  9. [9]E. Knill and R. Laflamme, Theory of quantum error-correcting codes, Phys. Rev. A 55, 900 (1997).DOI
  10. [10]B. M. Terhal, Quantum error correction for quantum memories, Rev. Mod. Phys. 87, 307 (2015).DOI
  11. [11]Google Quantum AI and Collaborators, Quantum error correction below the surface code threshold, Nature 638, 920 (2025), arXiv:2408.13687.
  12. [12]P. Zanardi and M. Rasetti, Noiseless quantum codes, Phys. Rev. Lett. 79, 3306 (1997).DOI
  13. [13]D. A. Lidar, I. L. Chuang, and K. B. Whaley, Decoherence-free subspaces for quantum computation, Phys. Rev. Lett. 81, 2594 (1998).
  14. [14]D. Bacon, K. R. Brown, and K. B. Whaley, Coherence-preserving quantum bits, Phys. Rev. Lett. 87, 247902 (2001).pubmed.ncbi.nlm.nih.gov/11736540
  15. [15]S. P. Jordan, E. Farhi, and P. W. Shor, Error-correcting codes for adiabatic quantum computation, Phys. Rev. A 74, 052322 (2006).DOI
  16. [16]A. Gyenis, A. Di Paolo, J. Koch, A. Blais, A. A. Houck, and D. I. Schuster, Moving beyond the transmon: Noise-protected superconducting quantum circuits, PRX Quantum 2, 030101 (2021).DOI
  17. [17]A. Sur, Q. Guo, and R. Mondaini, From Bell products to Greenberger–Horne–Zeilinger states: Quantum memories via emergent Hamiltonians, Phys. Rev. A 113, 042453 (2026), arXiv:2510.01117.DOI
  18. [18]L. Vidmar, D. Iyer, and M. Rigol, Emergent eigenstate solution to quantum dynamics far from equilibrium, Phys. Rev. X 7, 021012 (2017).DOI
  19. [19]L. Vidmar, W. Xu, and M. Rigol, Emergent eigenstate solution and emergent Gibbs ensemble for expansion dynamics in optical lattices, Phys. Rev. A 96, 013608 (2017).DOI
  20. [20]R. Modak, L. Vidmar, and M. Rigol, Quantum adiabatic protocols using emergent local Hamiltonians, Phys. Rev. E 96, 042155 (2017).DOI
  21. [21]Y. Zhang, L. Vidmar, and M. Rigol, Emergent eigenstate solution for generalized thermalization, Phys. Rev. A 104, L031303 (2021).DOI
  22. [22]X. Li, Y. Ma, J. Han, T. Chen, Y. Xu, W. Cai, H. Wang, Y. P. Song, Z.-Y. Xue, Z.-q. Yin, and L. Sun, Perfect quantum state transfer in a superconducting qubit chain with parametrically tunable couplings, Phys. Rev. Appl. 10, 054009 (2018).DOI
  23. [23]L. Xiang, J. Chen, Z. Zhu, Z. Song, Z. Bao, X. Zhu, F. Jin, K. Wang, S. Xu, Y. Zou, H. Li, Z. Wang, C. Song, A. Yue, J. Partridge, Q. Guo, R. Mondaini, H. Wang, and R. T. Scalettar, Enhanced quantum state transfer by circumventing quantum chaotic behavior, Nat. Commun. 15, 4918 (2024).DOI
  24. [24]R. J. Schoelkopf, A. A. Clerk, S. M. Girvin, K. W. Lehnert, and M. H. Devoret, Qubits as spectrometers of quantum noise, in Quantum Noise in Mesoscopic Physics, edited by Y. V. Nazarov (Springer, Dordrecht, 2003) pp. 175–203, arXiv:cond-mat/0210247.DOI
  25. [25]A. A. Clerk, M. H. Devoret, S. M. Girvin, F. Marquardt, and R. J. Schoelkopf, Introduction to quantum noise, measurement, and amplification, Rev. Mod. Phys. 82, 1155 (2010).DOI
  26. [26]E. Paladino, Y. M. Galperin, G. Falci, and B. L. Altshuler, 1/f noise: Implications for solid-state quantum information, Rev. Mod. Phys. 86, 361 (2014).DOI
  27. [27]F. Yoshihara, K. Harrabi, A. O. Niskanen, Y. Nakamura, and J. S. Tsai, Decoherence of flux qubits due to 1/f flux noise, Phys. Rev. Lett. 97, 167001 (2006).DOI
  28. [28]J. Bylander, S. Gustavsson, F. Yan, F. Yoshihara, K. Harrabi, G. Fitch, D. G. Cory, Y. Nakamura, J.-S. Tsai, and W. D. Oliver, Noise spectroscopy through dynamical decoupling with a superconducting flux qubit, Nat. Phys. 7, 565 (2011).DOI
  29. [29]Ł. Cywiński, R. M. Lutchyn, C. P. Nave, and S. Das Sarma, How to enhance dephasing time in superconducting qubits, Phys. Rev. B 77, 174509 (2008).DOI
  30. [30]O. Astafiev, Y. A. Pashkin, Y. Nakamura, T. Yamamoto, and J. S. Tsai, Quantum noise in the Josephson charge qubit, Phys. Rev. Lett. 93, 267007 (2004).DOI
  31. [31]L. Viola, E. Knill, and S. Lloyd, Dynamical decoupling of open quantum systems, Phys. Rev. Lett. 82, 2417 (1999).DOI
  32. [32]A. G. Kofman and G. Kurizki, Universal dynamical control of quantum mechanical decay: Modulation of the coupling to the continuum, Phys. Rev. Lett. 87, 270405 (2001).DOI
  33. [33]F. Yan, S. Gustavsson, J. Bylander, X. Jin, F. Yoshihara, D. G. Cory, Y. Nakamura, T. P. Orlando, and W. D. Oliver, Rotating-frame relaxation as a noise spectrum analyser of a superconducting qubit undergoing driven evolution, Nat. Commun. 4, 2337 (2013).DOI
  34. [34]A. G. Kofman and G. Kurizki, Acceleration of quantum decay processes by frequent observations, Nature 405, 546 (2000).pubmed.ncbi.nlm.nih.gov/10850708
  35. [35]T. Matsubara and H. Matsuda, A lattice model of liquid helium, J. Prog. Theor. Phys. 16, 569 (1956).DOI
  36. [36]P. Jordan and E. Wigner, Über das Paulische Äquivalenzverbot, Z. Phys. 47, 631 (1928).
  37. [37]E. Lieb, T. Schultz, and D. Mattis, Two soluble models of an antiferromagnetic chain, Ann. Phys. 16, 407 (1961).
  38. [38]M. Christandl, N. Datta, A. Ekert, and A. J. Landahl, Perfect state transfer in quantum spin networks, Phys. Rev. Lett. 92, 187902 (2004).pubmed.ncbi.nlm.nih.gov/15169534
  39. [39]C. Albanese, M. Christandl, N. Datta, and A. Ekert, Mirror inversion of quantum states in linear registers, Phys. Rev. Lett. 93, 230502 (2004).pubmed.ncbi.nlm.nih.gov/15601134
  40. [40]A. Kay, Perfect, efficient, state transfer and its application as a constructive tool, Int. J. Quantum Inf. 8, 641 (2010).DOI
  41. [41]B. Alkurtass, L. Banchi, and S. Bose, Optimal quench for distance-independent entanglement and maximal block entropy, Phys. Rev. A 90, 042304 (2014).DOI
  42. [42]G. Vitagliano, A. Riera, and J. I. Latorre, Volume-law scaling for the entanglement entropy in spin-1/2 chains, New J. Phys. 12, 113049 (2010).
  43. [43]G. Ramírez, J. Rodríguez-Laguna, and G. Sierra, Entanglement over the rainbow, J. Stat. Mech. , P06002 (2015), arXiv:1503.02695.DOI
  44. [44]E. B. Davies, Markovian master equations, Commun. Math. Phys. 39, 91 (1974).DOI
  45. [45]V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, Completely positive dynamical semigroups of N-level systems, J. Math. Phys. 17, 821 (1976).DOI
  46. [46]G. Lindblad, On the generators of quantum dynamical semigroups, Commun. Math. Phys. 48, 119 (1976).DOI
  47. [47]T. Prosen, Third quantization: a general method to solve master equations for quadratic open Fermi systems, New J. Phys. 10, 043026 (2008).DOI
  48. [48]T. Barthel and Y. Zhang, Solving quasi-free and quadratic Lindblad master equations for open fermionic and bosonic systems, J. Stat. Mech. 2022, 113101 (2022).DOI
  49. [49]J. R. Garrison and T. Grover, Does a single eigenstate encode the full Hamiltonian?, Phys. Rev. X 8, 021026 (2018).DOI
  50. [50]E. Chertkov and B. K. Clark, Computational inverse method for constructing spaces of quantum models from wave functions, Phys. Rev. X 8, 031029 (2018).DOI
  51. [51]X.-L. Qi and D. Ranard, Determining a local Hamiltonian from a single eigenstate, Quantum 3, 159 (2019).DOI
  52. [52]N. Gjonbalaj, C. Kokail, S. F. Yelin, and S. Choi, Unified probe of quantum chaos and ergodicity from Hamil- tonian learning (2026), arXiv:2603.04486 [quant-ph].
  53. [53]A. Sur and R. Mondaini, Data for “Resilience of Emergent Hamiltonian quantum memories in the presence of noise”, Zenodo (2026), doi:10.5281/zenodo.2498576.arxiv.org/abs/2611.25286
  54. [54]A. Trushechkin, Unified Gorini–Kossakowski–Lindblad–Sudarshan quantum master equation beyond the secular approximation, Phys. Rev. A 103, 062226 (2021).DOI
  55. [55]S. A. Hill and W. K. Wootters, Entanglement of a pair of quantum bits, Phys. Rev. Lett. 78, 5022 (1997).DOI
  56. [56]W. K. Wootters, Entanglement of formation of an arbitrary state of two qubits, Phys. Rev. Lett. 80, 2245 (1998).DOI
  57. [57]M. Kitagawa and M. Ueda, Squeezed spin states, Phys. Rev. A 47, 5138 (1993).
  58. [58]J. Dalibard, Y. Castin, and K. Mølmer, Wave-function approach to dissipative processes in quantum optics, Phys. Rev. Lett. 68, 580 (1992).pubmed.ncbi.nlm.nih.gov/10045937
  59. [59]R. Dum, P. Zoller, and H. Ritsch, Monte carlo simulation of the atomic master equation for spontaneous emission, Phys. Rev. A 45, 4879 (1992).pubmed.ncbi.nlm.nih.gov/9907570
  60. [60]K. Mølmer, Y. Castin, and J. Dalibard, Monte carlo wave-function method in quantum optics, J. Opt. Soc. Am. B 10, 524 (1993).DOI
  61. [61]M. B. Plenio and P. L. Knight, The quantum-jump approach to dissipative dynamics in quantum optics, Rev. Mod. Phys. 70, 101 (1998).
  62. [62]A. J. Daley, Quantum trajectories and open many-body quantum systems, Adv. Phys. 63, 77 (2014).DOI

Paper details

vlnc:W_h3rfqyjx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents