betaFeedback

WorksPhysics and Astronomy

Virtual Autopurification in Translationally Invariant Quantum States

Abstract

We introduce virtual autopurification (VAP), a protocol for virtual purification that operates entirely within a single translationally invariant system. Virtual purification suppresses thermal fluctuations and decoherence noise by amplifying the dominant pure component of a mixed state, but its standard implementation requires multiple copies of the state and global entangled measurements. VAP avoids the multiple-copy overhead by constructing virtual copies from spatially separated subregions and performing entangled measurements only between them. Assuming clustering of correlations, we show that VAP reproduces the expectation values of the standard virtual purification protocol, with discrepancies that decay exponentially in both the buffer length and the inter-subregion distance. Numerical simulations support the theoretical predictions and demonstrate accurate implementation of virtual cooling with reduced resource requirements. Our results establish VAP as a resource-efficient and scalable approach to quantum-state purification in near-term devices.

Introduction

Introduction.— Among digital quantum simulation tasks, thermal state preparation is a central application of quantum computing with potential impact across condensed-matter physics, quantum chemistry, and high-energy physics. The goal is to prepare a Gibbs state ρβ∝e−βH\rho_\beta\propto e^{-\beta H} for a given Hamiltonian HH and temperature T=1/βT = 1/\beta on quantum devices, which is essential for faithfully reproducing realistic experimental conditions.

Despite substantial theoretical progress [1, 2, 3, 4] and experimental demonstrations [5, 6, 7, 8], reaching lower temperatures is still difficult, because thermal fluctuations and decoherence must be suppressed strongly.

A promising route to alleviating these difficulties is virtual purification, a post-processing technique that extracts low-temperature observables from noisy or higher-temperature states [9, 10, 11, 12, 13, 14]. By amplifying the dominant pure component of a mixed state using powers of ρ\rho as ρn\rho^n via swap techniques, virtual purification suppresses thermal noise and decoherence. For a Gibbs state ρβ\rho_\beta, this amounts to a virtual cooling ρβ↦ρnβ\rho_\beta\mapsto\rho_{n\beta}, increasing the inverse temperature by a factor of nn. However, the

standard formulation, or fully virtual purification (FVP), requires globally entangled measurements over multiple physical copies, and moreover incurs exponential sampling overhead with system size [15].

A first step toward reducing such an overhead was provided by localized virtual purification (LVP) [16], which exploits the exponential clustering property in geometrically local systems to replace the globally entangled measurement with the one supported on the vicinity of the target observable of interest. While such a technique lifts the need for exponential sampling overhead, it still leaves the common crucial property in condensed matter problems: translational symmetry. Under clustered correlations, each local window of support can be expected to exhibit nearly identical reduced density matrices, while it has remained theoretically unclear whether it can be used as an operational resource for virtual purification.

In this work, we formulate this idea as virtual autopurification (VAP), a copy-free virtual purification protocol for translationally invariant quantum states, and prove its validity by providing a performance guarantee. As illustrated in Figure 1, spatially separated subregions in a translationally invariant system act as approximately independent copies. Assuming exponential clustering, we show that VAP reproduces expectation values of the standard purification protocol up to a bias that decays exponentially in both the buffer length and the inter-subregion separation, while circumventing the need for exponential overhead with respect to the system size (see Table 1). Numerical simulations corroborate these bounds and demonstrate accurate virtual cooling with substantially reduced resources, establishing VAP as a scalable route to virtual purification.

Figure 1. Schematic illustration of the VAP protocol. Identical subregions {AiBi}i=1n\{A_iB_i\}_{i=1}^n are chosen in a translationally invariant Gibbs state ρT\rho_T, with AA supporting the observable and BB providing a correlation buffer. Entangled measurements between distant subregions realize virtual cooling to temperature T/nT/n.

FVPLVPVAP (ours)
# copies per shotnnnn1
Sampling overheadexp⁡O(nN)\exp O(nN)exp⁡O(n)\exp O(n)exp⁡O(n)\exp O(n)
Clustering assumptionnoneObservable clusteringTrace-norm clustering

Table I. Comparison of fully virtual purification (FVP), localized virtual purification (LVP), and virtual autopurification (VAP). Here, nn is an index for a virtually purified state and NN is the total number of lattice sites. Observable clustering and trace-norm clustering are given by Eqs. (1) and (6), respectively.

Virtual purification.— We begin by briefly reviewing the existing purification protocols for simulating the Gibbs state of a geometrically-local Hamiltonian HH. Given a quantum state ρ\rho and an integer nn, FVP estimates an observable oAo_A supported on a local region AA via ⟨oA⟩FVP=Tr⁡[ρnoA]/Tr⁡[ρn]\langle o_A\rangle_{\mathrm{FVP}}=\operatorname{Tr}[\rho^n o_A]/\operatorname{Tr}[\rho^n] as introduced in Refs. [9, 10, 11], which is designed to amplify the dominant component of ρ\rho. To implement FVP on a quantum device, one uses the cyclic shift operator DnD_n acting on nn copies, which gives ⟨oA⟩FVP=Tr⁡[ρ⊗nDnoA]/Tr⁡[ρ⊗nDn]\langle o_A\rangle_{\mathrm{FVP}}=\operatorname{Tr}[\rho^{\otimes n}D_n o_A]/\operatorname{Tr}[\rho^{\otimes n}D_n]. FVP therefore requires globally entangled operation for measuring DnoAD_n o_A and DnD_n, which can be performed, for example, via Hadamard tests [17].

LVP reduces these measurement resources by restricting the cyclic shift to a local subsystem ABAB, where BB is a buffer surrounding AA and CC is the remainder of the system [16]. The resulting estimator is ⟨oA⟩LVP=Tr⁡[(ρAB)noA]/Tr⁡[(ρAB)n]\langle o_A\rangle_{\mathrm{LVP}}=\operatorname{Tr}[(\rho_{AB})^n o_A]/\operatorname{Tr}[(\rho_{AB})^n], where ρAB=Tr⁡C[ρ]\rho_{AB}=\operatorname{Tr}_C[\rho]. Its accuracy follows from observable clustering: for normalized operators MRM_R and MR′M_{R'} supported on disjoint regions,

∣Tr⁡[ρMRMR′]−Tr⁡[ρMR]Tr⁡[ρMR′]∣≤O(e−dRR′).(1)\left|\operatorname{Tr}[\rho M_RM_{R'}]-\operatorname{Tr}[\rho M_R]\operatorname{Tr}[\rho M_{R'}]\right|\le O(e^{-d_{RR'}}). \tag*{(1)}

where dRR′d_{RR'} is the distance between the regions. This property holds, in particular, for Gibbs states of local Hamiltonians at sufficiently high temperatures [18, 19, 20, 21, 22]. Under this condition, the FVP–LVP error satisfies

∣⟨oA⟩FVP−⟨oA⟩LVP∣≤O(e−dbuf),(2)\left|\langle o_A\rangle_{\mathrm{FVP}}-\langle o_A\rangle_{\mathrm{LVP}}\right|\le O(e^{-d_{\mathrm{buf}}}), \tag*{(2)}

where dbufd_{\mathrm{buf}} is the distance between AA and CC, as shown in Figure 1. Thus, LVP reproduces FVP up to exponentially small corrections while avoiding global measurements and substantially reducing the sampling overhead.

Virtual autopurification.— The essential idea of VAP is to eliminate the multiple-copy requirement of LVP by constructing virtual copies within a single translationally invariant quantum state. When correlations decay exponentially, two spatially separated regions behave approximately independently; thus the LVP estimator can be reproduced by performing an entangled measurement between distant, isomorphic subregions of a single copy, as illustrated in Figure 1.

For clarity, we present the analytic calculation for the case n=2n=2. Let ABAB be the subsystem used in LVP and A′B′A'B' be a translated copy of ABAB separated by an inter-copy distance dsepd_{\mathrm{sep}}. The remainder of the system is denoted by CC. The VAP estimator is defined as

⟨oA⟩VAP=Tr⁡[ρABA′B′SAB,A′B′oA]Tr⁡[ρABA′B′SAB,A′B′].(3)\langle o_A\rangle_{\mathrm{VAP}}=\frac{\operatorname{Tr}[\rho_{ABA'B'}S_{AB,A'B'}o_A]}{\operatorname{Tr}[\rho_{ABA'B'}S_{AB,A'B'}]}. \tag*{(3)}

where SAB,A′B′S_{AB,A'B'} swaps the two subsystems ABAB and A′B′A'B'. Assuming that the state is translationally invariant as ρAB=ρA′B′\rho_{AB}=\rho_{A'B'}, the LVP estimator can be rewritten as

⟨oA⟩LVP=Tr⁡[ρAB⊗ρA′B′SAB,A′B′oA]Tr⁡[ρAB⊗ρA′B′SAB,A′B′].(4)\langle o_A\rangle_{\mathrm{LVP}}=\frac{\operatorname{Tr}[\rho_{AB}\otimes\rho_{A'B'}S_{AB,A'B'}o_A]}{\operatorname{Tr}[\rho_{AB}\otimes\rho_{A'B'}S_{AB,A'B'}]}. \tag*{(4)}

The difference between the two estimators is given by

⟨oA⟩VAP−⟨oA⟩LVP=Tr⁡[(ρABA′B′−ρAB⊗ρA′B′)SAB,A′B′o~A]Tr⁡[(ρAB)2],(5)\langle o_A\rangle_{\mathrm{VAP}}-\langle o_A\rangle_{\mathrm{LVP}} =\frac{\operatorname{Tr}[(\rho_{ABA'B'}-\rho_{AB}\otimes\rho_{A'B'})S_{AB,A'B'}\widetilde{o}_A]}{\operatorname{Tr}[(\rho_{AB})^2]}, \tag*{(5)}

where o~A=oA−⟨oA⟩VAP\widetilde{o}_A=o_A-\langle o_A\rangle_{\mathrm{VAP}}.

If the quantum state on ABAB and A′B′A'B' were completely separated, so that ρABA′B′=ρAB⊗ρA′B′\rho_{ABA'B'}=\rho_{AB}\otimes\rho_{A'B'}, the LVP and VAP estimators would coincide. In general, however, this factorization does not hold exactly. We therefore assume exponential clustering at the level of reduced density operators, which we refer to as trace-norm clustering:

∥ρRR′−ρR⊗ρR′∥1≤O(e−dRR′).(6)\left\|\rho_{RR'}-\rho_R\otimes\rho_{R'}\right\|_1\le O(e^{-d_{RR'}}). \tag*{(6)}

This is a stronger assumption than the clustering of observables in Eq. (1). This assumption exactly holds for one-dimensional systems at arbitrary temperatures [23]. In higher-dimensional systems, it is also known to hold under the assumption that the effective Hamiltonians associated with local reduced density operators of the Gibbs state remain local [24].

Building on this clustering property, we analyze the VAP–LVP discrepancy through second order in λ\lambda, a parameter that scales the boundary couplings between the virtual-copy regions and the remainder of the system. Concretely, together with the high-temperature condition specified in Lemma 1 of the Supplemental Material (SM) [25], we obtain the following bound.

Theorem 1. Assuming trace-norm clustering and sufficiently high temperature indicating quasi-locality of the imaginary-time evolution, the difference between the FVP and VAP estimators is suppressed with both the buffer length dbufd_{\mathrm{buf}} and the inter-copy separation dsepd_{\mathrm{sep}} as

∣⟨oA⟩FVP−⟨oA⟩VAP∣≤O(e−dbuf)+O(λ2e−dsep)+O(λ3).(7)\left|\langle o_A\rangle_{\mathrm{FVP}}-\langle o_A\rangle_{\mathrm{VAP}}\right|\le O(e^{-d_{\mathrm{buf}}})+O(\lambda^2e^{-d_{\mathrm{sep}}})+O(\lambda^3). \tag*{(7)}

This theorem follows by combining the previously established bound between FVP and LVP in Eq. (2) with the proposition below.

Proposition 1. Under the assumptions stated in Theorem 1, the difference between the LVP and VAP estimators decays exponentially with the inter-copy distance:

∣⟨oA⟩LVP−⟨oA⟩VAP∣≤O(λ2e−dsep)+O(λ3).(8)\left|\langle o_A\rangle_{\mathrm{LVP}}-\langle o_A\rangle_{\mathrm{VAP}}\right| \le O(\lambda^2 e^{-d_{\mathrm{sep}}})+O(\lambda^3). \tag*{(8)}

Proof sketch. We compare the two density operators ρABA′B′\rho_{ABA'B'} and ρAB⊗ρA′B′\rho_{AB}\otimes\rho_{A'B'} in Eq. (5) perturbatively around a decoupled Hamiltonian. Let H=H0+VH=H_0+V, where H0H_0 contains the interactions within ABAB, A′B′A'B', and CC, while V=VBC+VB′CV=V_{BC}+V_{B'C} contains the boundary couplings to CC. By introducing a coupling parameter λ\lambda for VV, we consider the perturbed thermal states ρ(λ)∝e−β(H0+λV)\rho(\lambda)\propto e^{-\beta(H_0+\lambda V)}. The VAP and LVP estimators coincide up to first order in λ\lambda, while the second-order deviation of the density operators is given by

ρABA′B′−ρAB⊗ρA′B′=λ2ρAB(0)⊗ρA′B′(0)∫0βds∫0sdt [ΔBB′(s,t)−ΔB(s,t)−ΔB′(s,t)+Δ(s,t)]+O(λ3).(9)\begin{aligned} \rho_{ABA'B'}-\rho_{AB}\otimes\rho_{A'B'}={}&\lambda^2\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\int_0^\beta ds\int_0^s dt\,[\Delta_{BB'}(s,t)\\ &-\Delta_B(s,t)-\Delta_{B'}(s,t)+\Delta(s,t)]+O(\lambda^3). \tag*{(9)} \end{aligned}

Here, ρ(0)\rho^{(0)} denotes the thermal state at λ=0\lambda=0, and ΔBB′(s,t)\Delta_{BB'}(s,t) encodes the two-copy correlations,

ΔBB′(s,t)=Tr⁡C[ρBC(0)VB′CI(s)VB′CI(t)]−V~B(s)V~B′(t)+Tr⁡C[ρB′C(0)VB′CI(s)VBCI(t)]−V~B′(s)V~B(t),(10)\begin{aligned} \Delta_{BB'}(s,t)={}&\operatorname{Tr}_C[\rho_{BC}^{(0)}V_{B'C}^{\mathrm{I}}(s)V_{B'C}^{\mathrm{I}}(t)]-\widetilde{V}_B(s)\widetilde{V}_{B'}(t)\\ &+\operatorname{Tr}_C[\rho_{B'C}^{(0)}V_{B'C}^{\mathrm{I}}(s)V_{BC}^{\mathrm{I}}(t)]-\widetilde{V}_{B'}(s)\widetilde{V}_B(t), \tag*{(10)} \end{aligned}

with VBCI(s)=esH0VBCe−sH0V_{BC}^{\mathrm{I}}(s)=e^{sH_0}V_{BC}e^{-sH_0} and V~B(s)=Tr⁡C[ρBC(0)VBCI(s)]\widetilde{V}_B(s)=\operatorname{Tr}_C[\rho_{BC}^{(0)}V_{BC}^{\mathrm{I}}(s)] denoting the corresponding effective operator associated with BB. We further define the reduced operator ΔB(s,t)=Tr⁡A′B′[ρA′B′(0)ΔBB′(s,t)]\Delta_B(s,t)=\operatorname{Tr}_{A'B'}[\rho_{A'B'}^{(0)}\Delta_{BB'}(s,t)], with ΔB′(s,t)\Delta_{B'}(s,t) obtained analogously by tracing out ABAB. The corresponding scalar function is defined as Δ(s,t)=Tr⁡[ρ(0)ΔBB′(s,t)]\Delta(s,t)=\operatorname{Tr}[\rho^{(0)}\Delta_{BB'}(s,t)]. Substituting Eq. (9) into the expression for the difference between the VAP and LVP estimators in Eq. (5), we obtain four classes of terms. The scalar term Δ(s,t)\Delta(s,t) vanishes because it multiplies the zeroth-order expectation value of the centered observable. For the terms involving ΔB(s,t)\Delta_B(s,t) and ΔB′(s,t)\Delta_{B'}(s,t), Lemma 1 of SM [25] approximates the imaginary-time-evolved boundary interactions by local operators with exponentially small errors. The local-observable clustering condition in Eq. (1) then gives an O(e−dsep)O(e^{-d_{\mathrm{sep}}}) bound. The ΔBB′(s,t)\Delta_{BB'}(s,t) term retains correlations between reduced density operators even after the imaginary-time-evolved boundary interactions are replaced by their local approximations, and therefore requires the stronger condition in Eq. (6), which gives the same exponential bound. Combining these estimates gives the exponential suppression stated in Proposition 1. The detailed derivation for n=2n=2 is presented in SM [25]. Since connected contributions at second order in λ\lambda involve at most two virtual copies, the same analysis applies to any fixed n≥3n\geq3.

Theorem 1 bounds the systematic bias of VAP relative to FVP. We quantify the required resources by the space–time overhead, defined as the measurement cost multiplied by the number of copies required per shot. For estimating the expectation value of a Pauli operator PAP_A, the FVP variance is bounded as

1Tr⁡[ρn]2≤Var⁡(PA)FVP≤1+⟨PA⟩FVP2Tr⁡[ρn]2.(11)\frac{1}{\operatorname{Tr}[\rho^n]^2}\leq\operatorname{Var}(P_A)_{\mathrm{FVP}}\leq\frac{1+\langle P_A\rangle_{\mathrm{FVP}}^2}{\operatorname{Tr}[\rho^n]^2}. \tag*{(11)}

This motivates defining the measurement cost of FVP as Tr⁡[ρn]−2\operatorname{Tr}[\rho^n]^{-2}, following the Supplemental Material of Ref. [16]. By the same reasoning, the measurement costs of LVP and VAP are given by Tr⁡[(ρAB)n]−2\operatorname{Tr}[(\rho_{AB})^n]^{-2} and ∣Tr⁡[ρDnAB]∣−2|\operatorname{Tr}[\rho D_n^{AB}]|^{-2}, respectively. Combining these measurement costs with the number of physical copies required per shot, we define the space–time overhead CstC_{\mathrm{st}} as CstFVP=n/Tr⁡[ρn]2C_{\mathrm{st}}^{\mathrm{FVP}}=n/\operatorname{Tr}[\rho^n]^2, CstLVP=n/Tr⁡[(ρAB)n]2C_{\mathrm{st}}^{\mathrm{LVP}}=n/\operatorname{Tr}[(\rho_{AB})^n]^2 and CstVAP=1/∣Tr⁡[ρDnAB]∣2C_{\mathrm{st}}^{\mathrm{VAP}}=1/|\operatorname{Tr}[\rho D_n^{AB}]|^2 where DnABD_n^{AB} denotes the cyclic shift among the nn spatially separated subregions used as virtual copies within a single system. Their leading scaling is

CstFVP∼nexp⁡[2(n−1)Nst],CstLVP∼nexp⁡[2(n−1)NABst],CstVAP∼exp⁡[2(n−1)NABst].(12)\begin{aligned} C_{\mathrm{st}}^{\mathrm{FVP}}&\sim n\exp[2(n-1)N_{\mathrm{st}}],\\ C_{\mathrm{st}}^{\mathrm{LVP}}&\sim n\exp[2(n-1)N_{AB_{\mathrm{st}}}],\\ C_{\mathrm{st}}^{\mathrm{VAP}}&\sim\exp[2(n-1)N_{AB_{\mathrm{st}}}]. \tag*{(12)} \end{aligned}

Here, NN and NABN_{AB} are the numbers of sites in the full system and region ABAB, respectively, and sths_{\mathrm{th}} is the thermal entropy density. Thus, LVP and VAP replace the dependence on the total system size NN with that on the local subsystem size NABN_{AB}, while VAP further removes the prefactor nn by requiring only one physical copy per shot.

Numerical simulations

We numerically test the theoretical predictions above and benchmark the performance of VAP. As a benchmark, we consider the one-dimensional transverse-field Ising model with open boundary conditions,

H=−∑i=1N−1σizσi+1z−g∑i=1Nσix.(13)H=-\sum_{i=1}^{N-1}\sigma_i^z\sigma_{i+1}^z-g\sum_{i=1}^{N}\sigma_i^x. \tag*{(13)}

where σiα\sigma_i^\alpha (α=x,y,z\alpha=x,y,z) are Pauli operators at site ii. In the following calculations, we focus on the critical point g=1g=1. To access large system sizes and directly manipulate thermal mixed states, we employ matrix-product density operators (MPDOs) [26, 27, 28] implemented using the ITensor library [29, 30]. Following the subsystem arrangement shown in Figure 2 (inset), we study how the distances dbufd_{\mathrm{buf}} and dsepd_{\mathrm{sep}} control the differences among the three estimators for n=2n=2. For each method, we define the effective reduced density operator on region AA as ρAFVP=Tr⁡Aˉ[ρ2]/Tr⁡[ρ2]\rho_A^{\mathrm{FVP}}=\operatorname{Tr}_{\bar{A}}[\rho^2]/\operatorname{Tr}[\rho^2], ρALVP=Tr⁡Aˉ[(ρAB)2]/Tr⁡[(ρAB)2]\rho_A^{\mathrm{LVP}}=\operatorname{Tr}_{\bar{A}}[(\rho_{AB})^2]/\operatorname{Tr}[(\rho_{AB})^2] and ρAVAP=Tr⁡Aˉ[SAB,A′B′ρ]/Tr⁡[SAB,A′B′ρ]\rho_A^{\mathrm{VAP}}=\operatorname{Tr}_{\bar{A}}[S_{AB,A'B'}\rho]/\operatorname{Tr}[S_{AB,A'B'}\rho] where Aˉ\bar{A} is a complement of the subsystem AA. We refer to these as effective reduced density operators because ρVAP\rho_{\mathrm{VAP}} is generally non-Hermitian, while still reproducing the correct VAP expectation values.

Figure 2. (a), (b) Trace distances between the effective reduced density operators of FVP, LVP, and VAP as functions of dbufd_{\mathrm{buf}} and dsepd_{\mathrm{sep}}, respectively. (c), (d) Corresponding space–time overheads CstC_{\mathrm{st}}. The subsystem configuration is shown in the inset. We set β=1\beta= 1, ∣A∣=1|A| = 1, and ∣B∣=2dbuf|B| = 2d_{\mathrm{buf}}, with five additional boundary sites at each end of the chain, giving N=12+4dbuf+dsepN = 12 + 4d_{\mathrm{buf}} + d_{\mathrm{sep}}.

Figure 2 shows the trace distances between the effective reduced density operators and the corresponding space–time overheads for the three protocols, as functions of dbufd_{\mathrm{buf}} and dsepd_{\mathrm{sep}}. The exponential decay of the FVP–LVP discrepancy with dbufd_{\mathrm{buf}} is consistent with the theoretical result established in Ref. [16]. Likewise, the exponential decay of the LVP–VAP discrepancy with dsepd_{\mathrm{sep}} supports the validity of Proposition 1. At the same time, the numerical results also suggest a stronger scaling than that stated in Proposition 1, since the LVP–VAP discrepancy appears to decay exponentially in both dbufd_{\mathrm{buf}} and dsepd_{\mathrm{sep}}:

∣⟨oA⟩LVP−⟨oA⟩VAP∣=O(e−dbuf)×O(e−dsep).(14)|\langle o_A\rangle_{\mathrm{LVP}}-\langle o_A\rangle_{\mathrm{VAP}}| = O(e^{-d_{\mathrm{buf}}}) \times O(e^{-d_{\mathrm{sep}}}). \tag*{(14)}

Consequently, when dsepd_{\mathrm{sep}} is sufficiently large for the LVP–VAP discrepancy to become negligible, the VAP approximation error relative to FVP is governed primarily by dbufd_{\mathrm{buf}}. The space–time overhead further highlights the practical advantage of LVP and VAP. For fixed dsepd_{\mathrm{sep}}, the suppression of the LVP overhead relative to FVP scales as CstFVP/CstLVP∼exp⁡[2(n−1)NCsth]C_{\mathrm{st}}^{\mathrm{FVP}}/C_{\mathrm{st}}^{\mathrm{LVP}} \sim\exp[2(n-1)N_Cs_{\mathrm{th}}], where NC=N−NABN_C = N-N_{AB}. Thus, the LVP overhead is exponentially smaller than the FVP overhead in the size of CC. For fixed dbufd_{\mathrm{buf}}, NABN_{AB} remains constant as dsepd_{\mathrm{sep}} increases, explaining the nearly constant LVP and VAP overheads. VAP further reduces the overhead relative to LVP by approximately a factor of nn, since it requires only one physical copy per shot.

Figure 3 plots the space–time overhead for the number of virtual copies varied from n=1n=1 to 44. As shown in

Figure 3. (a) Space–time overhead of the VAP protocol CstVAPC_{\mathrm{st}}^{\mathrm{VAP}} as a function of β\beta for several numbers of virtual copies nn. (b) Ratio of the LVP overhead to the VAP overhead CstLVP/CstVAPC_{\mathrm{st}}^{\mathrm{LVP}}/C_{\mathrm{st}}^{\mathrm{VAP}} as a function of nn for β=0\beta= 0, 0.60.6, 0.80.8, and 11. The calculations use N=60N = 60, ∣A∣=2|A| = 2, and dbuf=4d_{\mathrm{buf}} = 4. For each nn, the virtual copies are arranged to maximize their inter-copy separations; for n=4n=4, this gives dsep=4d_{\mathrm{sep}}=4.

Fig. 3(a), the overhead of VAP decreases toward lower temperatures. This behavior is consistent with the entropy scaling in Eq. (12): lowering the temperature reduces sths_{\mathrm{th}} and hence the space–time overhead of VAP. Figure 3(b) shows CstLVP/CstVAPC_{\mathrm{st}}^{\mathrm{LVP}}/C_{\mathrm{st}}^{\mathrm{VAP}}, i.e., the advantage of VAP over LVP. In the high-temperature regime, the expected factor-of-nn advantage arising from the smaller number of copies required by VAP is clearly observed. For the fixed subsystem geometries considered here, the LVP-to-VAP overhead ratio exceeds the copy-number factor nn at lower temperatures, indicating an additional reduction in sampling overhead. This comparison isolates the variance-related cost and does not account for possible increases in dbufd_{\mathrm{buf}} and dsepd_{\mathrm{sep}} required to maintain a fixed approximation bias as the correlation length grows.

Finally, we numerically examine the robustness of VAP to violations of translational invariance in the Gibbs state due to miscalibration or algorithmic errors. To this end, we consider the one-dimensional Heisenberg model with a random magnetic field,

H=∑i=1N−1Si⋅Si+1+∑i=1NhiSiz,(15)H = \sum_{i=1}^{N-1} S_i \cdot S_{i+1} + \sum_{i=1}^{N} h_i S_i^z, \tag*{(15)}

where Siα=σiα/2S_i^\alpha= \sigma_i^\alpha/2 (α=x,y,z\alpha= x,y,z) denotes the spin operator at site ii. Each hih_i is independently drawn from a uniform distribution on [−W,W][-W,W], where WW sets the strength of the random field. The random fields represent spatially varying calibration errors that cause the prepared state to be the Gibbs state of a Hamiltonian different from the target one. This setting allows us to test whether VAP suppresses the resulting deviation from the target Gibbs state. Figure 4 shows the numerical results for W=0.5W=0.5 and W=1.0W=1.0. In both cases, VAP substantially reduces the disorder-induced deviation of the energy, yielding values closer to the clean reference than to the disordered reference. This behavior suggests that VAP has an effective disorder-averaging effect and can extract the approximately translationally invariant component of the thermal state. A possible interpretation is that combining the density operators associated with multiple virtual copies acts more like averaging the underlying local Hamiltonians than directly averaging the density operators themselves. Unlike FVP and LVP, which purify copies of the same spatially inhomogeneous state, VAP combines distinct spatial regions and can partially average out their local deviations. These results therefore suggest that VAP may mitigate spatially inhomogeneous errors that are not directly addressed by conventional virtual-purification protocols.

Figure 4. VAP energy for the random-field Heisenberg model for (a) W=0.5W = 0.5 and (b) W=1.0W = 1.0. The data are averaged over 10 disorder realizations. The gray dashed line indicates the energy of the original disordered thermal state, while the black dotted line indicates the corresponding clean reference without disorder (N=50N = 50).

Summary and discussion.— We have presented virtual autopurification (VAP), a protocol that performs virtual purification entirely within a single translationally invariant quantum state. By using spatially separated subregions as approximate virtual replicas, VAP removes the need for multiple physical copies while retaining the purification mechanism of fully virtual purification (FVP) [9, 10, 11]. Under clustering assumptions, we showed that the VAP estimator reproduces the results of FVP and localized virtual purification (LVP) [16], with errors that decay exponentially with the buffer length and the inter-copy separation. Numerical simulations corroborated these predictions and demonstrated accurate virtual cooling with reduced space–time overhead. Numerical results in the End Matter further show that VAP can simultaneously realize virtual cooling and mitigate local depolarizing noise, suggesting robustness against this form of decoherence noise.

To implement virtual purification protocols on real quantum devices [12], it is crucial to estimate nonlinear quantities such as Tr⁡[ρn]\operatorname{Tr}[\rho^n] and Tr⁡[ρoAn]\operatorname{Tr}[\rho^{n}_{\mathrm{oA}}] efficiently. Beyond the naive generalized swap trick, several approaches have recently been developed for this purpose, including methods based on query access to a unitary that prepares a purification of ρ\rho [31], randomized measurements [32, 33], and classical shadows [34, 35]. However, such estimator-level approaches do not by themselves remove the exponential sampling overhead intrinsic to virtual purification. In contrast, LVP and VAP exploit physical properties of ρ\rho, such as clustering and translational invariance, to reduce this overhead at the level of the purification protocol itself. This protocol-level reduction is complementary to nonlinear-estimation techniques and to other error-mitigation or error-correction layers, including quantum subspace expansion [36, 37] and approaches that combine virtual purification with quantum error correction [38]. Related single-state approaches to virtual purification have also been proposed, such as applying the inverse of the state-preparation circuit [39] or approximating the SWAP operator in the eigenbasis of the target observable [40]. Compared with these approaches, VAP provides a rigorous error analysis controlled by physical properties of the state.

The idea that a single thermal state can contain substantial information about the underlying Hamiltonian has appeared in distinct contexts. In quantum information, Hamiltonian learning from Gibbs states has been actively studied [41, 42, 43]. In statistical mechanics, related questions have been discussed for reconstructing a Hamiltonian from a single thermal eigenstate [44]. These perspectives suggest that a single thermal state can encode more information than is directly accessible through conventional expectation values. From this viewpoint, VAP can be regarded as a concrete protocol for extracting lower-temperature information encoded in a single Gibbs state. Exploring what other physical information can be accessed from thermal states is an interesting direction for future work.

Acknowledgments.— A. I. was supported by the Center of Innovation for Sustainable Quantum AI (JST Grant No. JPMJPF2221) and JSPS KAKENHI (Grant No. 25K17308). H. H. was supported by the MEXT Quantum Leap Flagship Program (MEXT Q-LEAP) (Grant No. JPMXS0112019794), JST PRESTO (Grant No. JPMJPR25F7), and JSPS KAKENHI (Grant No. JP24K16979). N. Y. was supported by JST (Grant No. JPMJPF2221), JST CREST (Grant No. JPMJCR23I4), JST ASPIRE (Grant No. JPMJAP2316), JST ERATO (Grant No. JPMJER2302), JST Moonshot R&D (Grant No. JPMJMS256J), IBM Quantum, Google Quantum AI, and the Institute for AI and Beyond at the University of Tokyo.

Data availability.— The code used to generate the numerical results reported in this work is publicly available on GitHub [45].

End matter

Analysis under stochastic noise

We consider the effect of local depolarizing noise. The depolarizing noise acting on the iith qubit is defined by

E(i)(ρ)=(1−p)ρ+p3(XiρXi+YiρYi+ZiρZi),(16)\mathcal{E}^{(i)}(\rho)=(1-p)\rho+\frac{p}{3}\left(X_i\rho X_i+Y_i\rho Y_i+Z_i\rho Z_i\right), \tag*{(16)}

where pp denotes the error rate. The local depolarizing noise channel is given by applying this noise independently to all sites:

E=⨂i=1NE(i).(17)\mathcal{E}=\bigotimes_{i=1}^{N}\mathcal{E}^{(i)}. \tag*{(17)}

In the numerical calculations below, we consider a thermal state ρβ∝e−βH\rho_{\beta}\propto e^{-\beta H} subjected to local depolarizing noise, and study a setting in which VAP simultaneously performs virtual cooling and error mitigation.

We present numerical results for dbuf=4d_{\mathrm{buf}}=4 and dsep=4d_{\mathrm{sep}}=4, with the number of virtual copies varied from n=1n=1 to 44. Figure 5(a) shows the VAP energy plotted against nβn\beta for a fixed error rate p=0.01p=0.01. The dashed line denotes the exact result, and the VAP data collapse well onto it. Figure 5(b) shows the deviation of the VAP energy from its noiseless value at p=0p=0. For all nn, the noise-induced deviation grows polynomially with pp. Figure 5(c) shows the deviation of the VAP energy from the exact result. For all error rates considered, this deviation is suppressed exponentially with increasing nn. Although we have not developed a theory of error mitigation in VAP, these results numerically demonstrate that, under a realistic error model, VAP can simultaneously realize virtual cooling and error mitigation.

Figure 5. Thermal-state results in the presence of local depolarizing noise. (a) VAP energy rescaled by nβn\beta, plotted as a function of nβn\beta for p=0.01p = 0.01. The dashed line indicates the exact result. (b) Absolute deviation of the energy from its noiseless value at p=0p = 0, shown as a function of pp for β=0.5\beta= 0.5. (c) Absolute deviation of the energy from the exact result, shown as a function of nn for β=0.5\beta= 0.5.

To qualitatively understand Figure 5(b), consider a simple noninteracting toy model,

H=−∑iσiz.(18)H=-\sum_i \sigma_i^z. \tag*{(18)}

For this model, the thermal state factorizes into single-site mixed states with polarization m=⟨σz⟩=tanh⁡βm=\langle\sigma^z\rangle=\tanh\beta, and local depolarizing noise reduces the polarization as m↦ηmm\mapsto\eta m with η=1−4p/3\eta=1-4p/3. Virtual purification with nn copies then gives an effective polarization

mn=tanh⁡[narctanh⁡(ηm)],(19)m_n=\tanh\left[n\operatorname{arctanh}(\eta m)\right], \tag*{(19)}

so that, for small pp, the energy deviation behaves as

∣E(p)−E(0)∣N=Cn(β)p+O(p2),(20)\frac{\lvert E(p)-E(0)\rvert}{N}=C_n(\beta)p+O(p^2), \tag*{(20)}

where

Cn(β)=2n3sinh⁡(2β)sech⁡2(nβ).(21)C_n(\beta)=\frac{2n}{3}\sinh(2\beta)\operatorname{sech}^2(n\beta). \tag*{(21)}

At finite β\beta, the thermal state contains excited-state components whose weights are modified by stochastic noise. When these weights are raised to the nnth power, a first-order noise correction in one factor combines with the noiseless thermal weight in the remaining factors. As a result, a term linear in pp remains after virtual purification. For a pure ground state, incoherent errors can introduce weight in orthogonal excited states. Virtual purification raises these noise-induced weights to the nnth power, so their contribution to the error typically scales as pnp^n. In the toy model, the energy deviation in the pure ground-state limit β→∞\beta\to\infty is

∣E(p)−E(0)∣N=2(2p3)n+O(pn+1).(22)\frac{\lvert E(p)-E(0)\rvert}{N}=2\left(\frac{2p}{3}\right)^n+O(p^{n+1}). \tag*{(22)}

For nβ≫1n\beta\gg1, we have Cn(β)∼(8n/3)sinh⁡(2β)e−2nβC_n(\beta)\sim(8n/3)\sinh(2\beta)e^{-2n\beta}. Thus, at fixed β>0\beta>0, the coefficient decays exponentially with nn up to the prefactor nn, consistent with Figure 5(c).

Setup and Notation

Setup

We consider a quantum many-body system on a lattice equipped with the graph distance dd, with a local Hamiltonian

H=∑XhX,(23)H = \sum_X h_X, \tag*{(23)}

where each term hXh_X is supported on a subset XX of the lattice. We assume that every nonzero interaction term satisfies ∣X∣≤k|X| \le k and diam⁡(X)=max⁡x,y∈Xd(x,y)≤r\operatorname{diam}(X) = \max_{x,y\in X} d(x,y) \le r. We assume a uniform bound on the local interaction strength,

max⁡v∑X:v∈X∥hX∥≤g.(24)\max_v \sum_{X:v\in X} \lVert h_X\rVert\le g. \tag*{(24)}

Here, kk bounds the number of sites involved in each interaction, rr is the maximum spatial range of the interactions, and gg sets the typical local energy scale.

As illustrated schematically in Figure S1, we partition the system into disjoint regions AA, BB, A′A', B′B', CC. Region AA supports the local observable of interest oAo_A. We assume that its support contains at most kk sites:

Figure S1. Schematic picture of the spatial partition used in the analysis. Regions BB and B′B' form buffers around AA and A′A', respectively, while CC denotes the remainder of the system. The buffer length dbufd_{\mathrm{buf}} and the inter-copy separation dsepd_{\mathrm{sep}} are also indicated.

∣supp⁡(oA)∣≤k.(25)\lvert\operatorname{supp}(o_A)\rvert\le k. \tag*{(25)}

Region BB is a buffer surrounding AA, and we set

AB=A∪B.(26)AB = A \cup B. \tag*{(26)}

We further choose A′B′A'B' to be a translated copy of ABAB, spatially separated from it. The remainder of the system is denoted by CC.

Throughout the analysis, we use two characteristic distances: the buffer length dbufd_{\mathrm{buf}} between AA and CC, and the inter-copy separation dsepd_{\mathrm{sep}} between ABAB and A′B′A'B'. These two length scales, shown in Figure S1, govern the error of the LVP approximation and the additional error introduced in VAP, respectively. We assume dsep>rd_{\mathrm{sep}} > r to exclude direct interactions between ABAB and A′B′A'B'.

We denote the mixed state of the full system by ρ\rho. For any region RR, the reduced state on RR is written as

ρR=Tr⁡R‾[ρ],(27)\rho_R = \operatorname{Tr}_{\overline{R}}[\rho], \tag*{(27)}

where Rˉ\bar{R} is the complement of RR. When the global state is translationally invariant and A′B′A'B' is a translated copy of ABAB, we have

ρAB=ρA′B′.(28)\rho_{AB}=\rho_{A'B'}. \tag*{(28)}

Swap operator

Let SAB,A′B′S_{AB,A'B'} denote the swap operator exchanging the two isomorphic subsystems ABAB and A′B′A'B':

SAB,A′B′∣ψ⟩AB⊗∣ϕ⟩A′B′=∣ϕ⟩AB⊗∣ψ⟩A′B′,(29)S_{AB,A'B'}|\psi\rangle_{AB}\otimes|\phi\rangle_{A'B'}=|\phi\rangle_{AB}\otimes|\psi\rangle_{A'B'}, \tag*{(29)}

with trivial action on the rest of the system. We will repeatedly use the identity

Tr⁡ABA′B′[(XAB⊗YA′B′)SAB,A′B′]=Tr⁡AB[XABYAB],(30)\operatorname{Tr}_{ABA'B'}[(X_{AB}\otimes Y_{A'B'})S_{AB,A'B'}]=\operatorname{Tr}_{AB}[X_{AB}Y_{AB}], \tag*{(30)}

where XABX_{AB} and YA′B′Y_{A'B'} act on isomorphic subsystems.

LVP and VAP estimators (for n=2)

For the remainder of the Supplemental Material, we focus on the case n=2n=2. Given a local observable oAo_A supported on AA, we define the localized virtual purification (LVP) expectation value as

⟨oA⟩LVP=Tr⁡AB[(ρAB)2oA]Tr⁡AB[(ρAB)2]=Tr⁡ABA′B′[ρAB⊗ρA′B′SAB,A′B′oA]Tr⁡ABA′B′[ρAB⊗ρA′B′SAB,A′B′].(31)\begin{aligned} \langle o_A\rangle_{\mathrm{LVP}}=\frac{\operatorname{Tr}_{AB}[(\rho_{AB})^2o_A]}{\operatorname{Tr}_{AB}[(\rho_{AB})^2]} \\ &=\frac{\operatorname{Tr}_{ABA'B'}[\rho_{AB}\otimes\rho_{A'B'}S_{AB,A'B'}o_A]}{\operatorname{Tr}_{ABA'B'}[\rho_{AB}\otimes\rho_{A'B'}S_{AB,A'B'}]}. \tag*{(31)} \end{aligned}

The virtual autopurification (VAP) estimator is defined by

⟨oA⟩VAP=Tr⁡ABA′B′[ρABA′B′SAB,A′B′oA]Tr⁡ABA′B′[ρABA′B′SAB,A′B′].(32)\langle o_A\rangle_{\mathrm{VAP}}=\frac{\operatorname{Tr}_{ABA'B'}[\rho_{ABA'B'}S_{AB,A'B'}o_A]}{\operatorname{Tr}_{ABA'B'}[\rho_{ABA'B'}S_{AB,A'B'}]}. \tag*{(32)}

Using the identity in the previous subsection, one obtains the exact relation

⟨oA⟩VAP−⟨oA⟩LVP=Tr⁡ABA′B′[(ρABA′B′−ρAB⊗ρA′B′)SAB,A′B′(oA−⟨oA⟩VAP)]Tr⁡AB[(ρAB)2].(33)\langle o_A\rangle_{\mathrm{VAP}}-\langle o_A\rangle_{\mathrm{LVP}}= \frac{\operatorname{Tr}_{ABA'B'}[(\rho_{ABA'B'}-\rho_{AB}\otimes\rho_{A'B'})S_{AB,A'B'}(o_A-\langle o_A\rangle_{\mathrm{VAP}})]}{\operatorname{Tr}_{AB}[(\rho_{AB})^2]}. \tag*{(33)}

Perturbative setup

To control correlations between ABAB and A′B′A'B' through CC, we decompose the Hamiltonian into intra-region and boundary contributions:

H=HAB+HA′B′+HC+VBC+VB′C=H0+V,(34)\begin{aligned} H&=H_{AB}+H_{A'B'}+H_C+V_{BC}+V_{B'C}\\ &=H_0+V, \tag*{(34)} \end{aligned}

Here, H0=HAB+HA′B′+HCH_0=H_{AB}+H_{A'B'}+H_C and V=VBC+VB′CV=V_{BC}+V_{B'C}. For example, HAB=∑X⊂ABhXH_{AB}=\sum_{X\subset AB}h_X, and VBCV_{BC} collects interactions straddling the boundary ∂(BC)\partial(BC):

VBC=∑X⊂∂(BC)hX,(35)V_{BC}=\sum_{X\subset\partial(BC)}h_X, \tag*{(35)}

and similarly for VB′CV_{B'C}. We introduce a coupling parameter λ\lambda as

H(λ)=H0+λV,(36)H(\lambda)=H_0+\lambda V, \tag*{(36)}

and define the corresponding Gibbs state

ρ(λ)=e−βH(λ)Tr⁡[e−βH(λ)].(37)\rho(\lambda)=\frac{e^{-\beta H(\lambda)}}{\operatorname{Tr}[e^{-\beta H(\lambda)}]}. \tag*{(37)}

The original Hamiltonian and Gibbs state are recovered as H=H(λ=1)H=H(\lambda=1) and ρ=ρ(λ=1)\rho=\rho(\lambda=1). In § 2, we expand the relevant quantities around λ=0\lambda=0 and retain terms through second order in λ\lambda. When no confusion can arise, we leave the λ\lambda dependence of reduced states implicit.

Proof of Proposition 1

In this section, we prove Proposition 1 by a perturbative expansion in the coupling λ\lambda up to second order. The main goal is to bound the difference between the VAP and LVP estimators while keeping track of their normalization factors. Throughout the proof, we use the perturbative setup introduced in Sec. S0.

Clustering assumption

We begin by stating the clustering assumption used in the proof of Proposition 1. Throughout this section, we assume that the reduced density operators of distant regions satisfy exponential clustering in trace norm:

∥ρRR′−ρR⊗ρR′∥1≤const.×e−dRR′/ξ,(38)\lVert\rho_{RR'}-\rho_R\otimes\rho_{R'}\rVert_1 \leq\mathrm{const.}\times e^{-d_{RR'}/\xi}, \tag*{(38)}

for disjoint regions RR and R′R', with some correlation length ξ>0\xi>0.

This assumption immediately implies the standard exponential clustering of connected correlations. Indeed, for operators MRM_R and MR′M_{R'} supported on RR and R′R', respectively, we have

∣Tr⁡[ρMRMR′]−Tr⁡[ρMR]Tr⁡[ρMR′]∣=∣Tr⁡[(ρRR′−ρR⊗ρR′)(MR⊗MR′)]∣≤∥ρRR′−ρR⊗ρR′∥1∥MR∥∥MR′∥≤const.×∥MR∥∥MR′∥e−dRR′/ξ.(39)\begin{aligned} \left\lvert\operatorname{Tr}[\rho M_RM_{R'}]-\operatorname{Tr}[\rho M_R]\operatorname{Tr}[\rho M_{R'}]\right\rvert &=\left\lvert\operatorname{Tr}\left[(\rho_{RR'}-\rho_R\otimes\rho_{R'})(M_R\otimes M_{R'})\right]\right\rvert\\ &\leq\lVert\rho_{RR'}-\rho_R\otimes\rho_{R'}\rVert_1\lVert M_R\rVert\lVert M_{R'}\rVert\\ &\leq\mathrm{const.}\times\lVert M_R\rVert\lVert M_{R'}\rVert e^{-d_{RR'}/\xi}. \tag*{(39)} \end{aligned}

Exact identity for the VAP-LVP difference

We now rewrite the VAP-LVP difference in a form suitable for perturbative analysis. For this purpose, we first establish explicitly that the first-order contribution to the relevant density-operator difference vanishes. Let ρ0=ρ(0)=e−βH0/Tr⁡[e−βH0]\rho_0=\rho(0)=e^{-\beta H_0}/\operatorname{Tr}[e^{-\beta H_0}], and denote its reduced states on ABAB, A′B′A'B', and CC by ρAB(0)\rho_{AB}^{(0)}, ρA′B′(0)\rho_{A'B'}^{(0)}, and ρC(0)\rho_C^{(0)}, respectively. Since H0H_0 contains no couplings between these regions, the unperturbed Gibbs state factorizes as ρ0=ρAB(0)⊗ρA′B′(0)⊗ρC(0)\rho_0=\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\otimes\rho_C^{(0)}.

The first-order correction to the normalized Gibbs state is linear in the perturbation V=VBC+VB′CV=V_{BC}+V_{B'C}. Here and below, ρAB(1)\rho_{AB}^{(1)} and ρA′B′(1)\rho_{A'B'}^{(1)} denote the coefficients of λ\lambda in the expansions of ρAB(λ)\rho_{AB}(\lambda) and ρA′B′(λ)\rho_{A'B'}(\lambda), respectively. If only VBCV_{BC} is retained, A′B′A'B' remains decoupled from ABCABC for all λ\lambda, so this perturbation contributes only ρAB(1)⊗ρA′B′(0)\rho_{AB}^{(1)}\otimes\rho_{A'B'}^{(0)} to the first-order correction of ρABA′B′(λ)\rho_{ABA'B'}(\lambda). Similarly, VB′CV_{B'C} contributes only ρAB(0)⊗ρA′B′(1)\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(1)}. Therefore,

ρABA′B′(λ)=ρAB(0)⊗ρA′B′(0)+λ(ρAB(1)⊗ρA′B′(0)+ρAB(0)⊗ρA′B′(1))+O(λ2),(40)\rho_{ABA'B'}(\lambda)=\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}+\lambda\left(\rho_{AB}^{(1)}\otimes\rho_{A'B'}^{(0)}+\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(1)}\right)+O(\lambda^2), \tag*{(40)}

On the other hand, expanding the product of the reduced states gives

ρAB(λ)⊗ρA′B′(λ)=ρAB(0)⊗ρA′B′(0)+λ(ρAB(1)⊗ρA′B′(0)+ρAB(0)⊗ρA′B′(1))+O(λ2).(41)\rho_{AB}(\lambda)\otimes\rho_{A'B'}(\lambda)=\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}+\lambda\left(\rho_{AB}^{(1)}\otimes\rho_{A'B'}^{(0)}+\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(1)}\right)+O(\lambda^2). \tag*{(41)}

Thus, the zeroth- and first-order terms coincide, and hence

ρABA′B′(λ)−ρAB(λ)⊗ρA′B′(λ)=O(λ2).(42)\rho_{ABA'B'}(\lambda)-\rho_{AB}(\lambda)\otimes\rho_{A'B'}(\lambda)=O(\lambda^2). \tag*{(42)}

Consequently, the difference ⟨oA⟩VAP−⟨oA⟩LVP\langle o_A\rangle_{\mathrm{VAP}}-\langle o_A\rangle_{\mathrm{LVP}} starts at second order in λ\lambda. Therefore, up to O(λ2)O(\lambda^2), the normalization factors can be replaced by their zeroth-order values, which yields

⟨oA⟩VAP−⟨oA⟩LVP=Tr⁡ABA′B′[(ρABA′B′−ρAB⊗ρA′B′)SAB,A′B′(oA−⟨oA⟩ρ02)]Tr⁡AB[(ρAB(0))2]+O(λ3).(43)\langle o_A\rangle_{\mathrm{VAP}}-\langle o_A\rangle_{\mathrm{LVP}}= \frac{\operatorname{Tr}_{ABA'B'}\left[\left(\rho_{ABA'B'}-\rho_{AB}\otimes\rho_{A'B'}\right)S_{AB,A'B'}\left(o_A-\langle o_A\rangle_{\rho_0^2}\right)\right]}{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\right]}+O(\lambda^3). \tag*{(43)}

Here

⟨oA⟩ρ02=Tr⁡AB[(ρAB(0))2oA]Tr⁡AB[(ρAB(0))2].(44)\langle o_A\rangle_{\rho_0^2}=\frac{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2o_A\right]}{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\right]}. \tag*{(44)}

From this point on, we redefine oAo_A by shifting it as

oA→oA−⟨oA⟩ρ02.(45)o_A \to o_A-\langle o_A\rangle_{\rho_0^2}. \tag*{(45)}

With this convention,

Tr⁡AB[(ρAB(0))2oA]=0.(46)\operatorname{Tr}_{AB}\left[\left(\rho_{AB}^{(0)}\right)^2 o_A\right]=0. \tag*{(46)}

Second-order expansion

Having established the cancellation of the first-order contribution, we now evaluate the leading second-order term explicitly. We work in the interaction picture with respect to H0H_0 and define

VI(s)=esH0Ve−sH0,s∈[0,β].(47)V^{I}(s)=e^{sH_0}Ve^{-sH_0},\qquad s\in[0,\beta]. \tag*{(47)}

For any operator XX, we denote the thermal expectation value in the unperturbed state ρ0\rho_0 by

⟨X⟩0=Tr⁡[ρ0X].(48)\langle X\rangle_0=\operatorname{Tr}[\rho_0X]. \tag*{(48)}

It is convenient to introduce the fluctuation operators

δVI(s)=VI(s)−⟨V⟩0,(49)\delta V^{I}(s)=V^{I}(s)-\langle V\rangle_0, \tag*{(49)}

and

δ2VI(s)VI(t)=δVI(s)δVI(t)−⟨δVI(s)δVI(t)⟩0.(50)\delta^2V^{I}(s)V^{I}(t)=\delta V^{I}(s)\delta V^{I}(t)-\langle\delta V^{I}(s)\delta V^{I}(t)\rangle_0. \tag*{(50)}

With these definitions, the normalized Gibbs state admits the Duhamel expansion

ρ(λ)=ρ0−λρ0∫0βds δVI(s)+λ2ρ0∫0βds∫0sdt δ2VI(s)VI(t)+O(λ3).(51)\rho(\lambda)=\rho_0-\lambda\rho_0\int_0^\beta ds\,\delta V^{I}(s)+\lambda^2\rho_0\int_0^\beta ds\int_0^s dt\,\delta^2V^{I}(s)V^{I}(t)+O(\lambda^3). \tag*{(51)}

We next trace out the region CC in order to isolate the effective coupling between ABAB and A′B′A'B'. In the subscripts of the effective operators, AA and A′A' are omitted. The labels BB and B′B' indicate the boundaries from which these operators arise, although the operators may act on the full regions ABAB and A′B′A'B', respectively. The effective boundary operators are

V~B(s)=Tr⁡C[ρC(0)VBCI(s)],V~B′(s)=Tr⁡C[ρC(0)VB′CI(s)].(52)\widetilde V_B(s)=\operatorname{Tr}_C[\rho_C^{(0)}V^I_{BC}(s)],\qquad\widetilde V_{B'}(s)=\operatorname{Tr}_C[\rho_C^{(0)}V^I_{B'C}(s)]. \tag*{(52)}

where

ρC(0)=Tr⁡ABA′B′[ρ0].(53)\rho_C^{(0)}=\operatorname{Tr}_{ABA'B'}[\rho_0]. \tag*{(53)}

We then define

ΔBB′(s,t)=Tr⁡C[ρC(0)VBCI(s)VB′CI(t)]−V~B(s)V~B′(t)+Tr⁡C[ρC(0)VB′CI(s)VBCI(t)]−V~B′(s)V~B(t).(54)\begin{aligned} \Delta_{BB'}(s,t)={}&\operatorname{Tr}_C[\rho_C^{(0)}V^I_{BC}(s)V^I_{B'C}(t)]-\widetilde V_B(s)\widetilde V_{B'}(t)\\ &+\operatorname{Tr}_C[\rho_C^{(0)}V^I_{B'C}(s)V^I_{BC}(t)]-\widetilde V_{B'}(s)\widetilde V_B(t). \tag*{(54)} \end{aligned}

which captures the connected second-order contribution between the two boundary couplings through CC. From ΔBB′(s,t)\Delta_{BB'}(s,t), we further define

ΔB(s,t)=Tr⁡A′B′[ρA′B′(0)ΔBB′(s,t)],(55)\Delta_B(s,t)=\operatorname{Tr}_{A'B'}[\rho_{A'B'}^{(0)}\Delta_{BB'}(s,t)], \tag*{(55)}
ΔB′(s,t)=Tr⁡AB[ρAB(0)ΔBB′(s,t)],(56)\Delta_{B'}(s,t)=\operatorname{Tr}_{AB}[\rho_{AB}^{(0)}\Delta_{BB'}(s,t)], \tag*{(56)}
Δ(s,t)=⟨ΔBB′(s,t)⟩0.(57)\Delta(s,t)=\langle\Delta_{BB'}(s,t)\rangle_0. \tag*{(57)}

Collecting all second-order terms, we obtain

ρABA′B′−ρAB⊗ρA′B′=λ2ρAB(0)⊗ρA′B′(0)∫0βds∫0sdt[ΔBB′(s,t)−ΔB(s,t)−ΔB′(s,t)+Δ(s,t)]+O(λ3).(58)\rho_{ABA'B'}-\rho_{AB}\otimes\rho_{A'B'}=\lambda^2\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\int_0^\beta ds\int_0^s dt\left[\Delta_{BB'}(s,t)-\Delta_B(s,t)-\Delta_{B'}(s,t)+\Delta(s,t)\right]+O(\lambda^3). \tag*{(58)}

Equation (S36) reduces the problem to bounding the four second-order contributions separately, which we do in the next subsection.

Bounds on the second-order terms

Reduction of the numerator

Substituting Eq. (58) into Eq. (43), we find that the estimator difference is determined by four second-order contributions associated with ΔBB′(s,t)\Delta_{BB'}(s,t), ΔB(s,t)\Delta_B(s,t), ΔB′(s,t)\Delta_{B'}(s,t), and Δ(s,t)\Delta(s,t). To evaluate these terms, it is convenient to introduce the imaginary-time evolved operator

O(τ)=e−τHABOeτHAB.(59)O(\tau)=e^{-\tau H_{AB}}Oe^{\tau H_{AB}}. \tag*{(59)}

In particular, we will use oA(β)=e−βHABoAeβHABo_A(\beta)=e^{-\beta H_{AB}}o_Ae^{\beta H_{AB}}.

It is also convenient to rewrite the relevant traces in terms of the auxiliary state

ρ~=exp⁡(−2βHAB−βHA′B′−βHC)Tr⁡[exp⁡(−2βHAB−βHA′B′−βHC)].(60)\widetilde{\rho}=\frac{\exp(-2\beta H_{AB}-\beta H_{A'B'}-\beta H_C)}{\operatorname{Tr}[\exp(-2\beta H_{AB}-\beta H_{A'B'}-\beta H_C)]}. \tag*{(60)}

which naturally appears after applying the swap identity and collecting the thermal weights. We denote expectation values with respect to ρ~\widetilde{\rho} by ⟨⋯ ⟩ρ~\langle\cdots\rangle_{\widetilde{\rho}}.

Vanishing of the scalar term (s,t)

We first note that the contribution proportional to Δ(s,t)\Delta(s,t) vanishes. Indeed, since Δ(s,t)\Delta(s,t) is a scalar, it can be pulled out of the trace, and we obtain

Tr⁡[ρAB(0)⊗ρA′B′(0)Δ(s,t)SAB,A′B′oA]=Δ(s,t)Tr⁡[(ρAB(0))2oA].(61)\operatorname{Tr}\left[\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\Delta(s,t)S_{AB,A'B'}o_A\right]=\Delta(s,t)\operatorname{Tr}\left[(\rho_{AB}^{(0)})^2o_A\right]. \tag*{(61)}

By construction, the observable oAo_A has been centered so that

Tr⁡[(ρAB(0))2oA]=0.(62)\operatorname{Tr}\left[(\rho_{AB}^{(0)})^2o_A\right]=0. \tag*{(62)}

Therefore, the scalar contribution Δ(s,t)\Delta(s,t) does not contribute to the VAP–LVP difference at order λ2\lambda^2.

Bounds on the single-boundary terms B(s,t) and B'(s,t)

We first consider the contribution containing ΔB(s,t)\Delta_B(s,t). Applying the swap identity in Eq. (30) to the normalized trace in Eq. (43), and identifying the unperturbed Gibbs states of the translated copies ABAB and A′B′A'B' under the swap, we obtain

Tr⁡ABA′B′[ρAB(0)⊗ρA′B′(0)ΔB(s,t)SAB,A′B′oA]Tr⁡AB[(ρAB(0))2]=Tr⁡AB[ρAB(0)ΔB(s,t)ρAB(0)oA]Tr⁡AB[(ρAB(0))2]=Tr⁡AB[(ρAB(0))2ΔB(s,t)oA(β)]Tr⁡AB[(ρAB(0))2]=⟨ΔBB′(s,t)oA(β)⟩ρ~.\begin{aligned} \frac{\operatorname{Tr}_{ABA'B'}\left[\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\Delta_B(s,t)S_{AB,A'B'}o_A\right]}{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\right]} &=\frac{\operatorname{Tr}_{AB}\left[\rho_{AB}^{(0)}\Delta_B(s,t)\rho_{AB}^{(0)}o_A\right]}{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\right]}\\ &=\frac{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\Delta_B(s,t)o_A(\beta)\right]}{\operatorname{Tr}_{AB}\left[(\rho_{AB}^{(0)})^2\right]}\\ &=\langle\Delta_{BB'}(s,t)o_A(\beta)\rangle_{\widetilde{\rho}}. \end{aligned}

In the second equality, we used ρAB(0)oA=oA(β)ρAB(0)\rho_{AB}^{(0)}o_A=o_A(\beta)\rho_{AB}^{(0)} and the cyclicity of the trace. The last equality follows from the definition of ΔB(s,t)\Delta_B(s,t) as a trace over A′B′A'B' and the definition of ρ~\widetilde{\rho}. Expanding ΔBB′(s,t)\Delta_{BB'}(s,t), we obtain

⟨ΔBB′(s,t)oA(β)⟩ρ~=⟨VBCI(s)oA(β)VB′CI(t)⟩ρ~−⟨VBCI(s)oA(β)⟩ρ~⟨VB′CI(t)⟩ρ~+⟨VB′CI(s)VBCI(t)oA(β)⟩ρ~−⟨VB′CI(s)⟩ρ~⟨VBCI(t)oA(β)⟩ρ~.(63)\begin{aligned} \langle\Delta_{BB'}(s,t)o_A(\beta)\rangle_{\widetilde{\rho}} ={}&\langle V_{BC}^{\mathrm{I}}(s)o_A(\beta)V_{B'C}^{\mathrm{I}}(t)\rangle_{\widetilde{\rho}} -\langle V_{BC}^{\mathrm{I}}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{\mathrm{I}}(t)\rangle_{\widetilde{\rho}}\\ &+\langle V_{B'C}^{\mathrm{I}}(s)V_{BC}^{\mathrm{I}}(t)o_A(\beta)\rangle_{\widetilde{\rho}} -\langle V_{B'C}^{\mathrm{I}}(s)\rangle_{\widetilde{\rho}}\langle V_{BC}^{\mathrm{I}}(t)o_A(\beta)\rangle_{\widetilde{\rho}}. \tag*{(63)} \end{aligned}

Here, the products of expectation values arise from the factorization of ρ~\tilde{\rho} between ABAB, A′B′A'B', and CC. In the first term, we used [oA(β),VB′CI(t)]=0[o_A(\beta),V^I_{B'C}(t)]=0, since these operators act on disjoint regions. The contribution containing ΔB′(s,t)\Delta_{B'}(s,t) is treated in the same way.

To bound the connected correlators in Eq. (63), we use a quasi-locality property of the imaginary-time evolved boundary interactions. Roughly speaking, although VBC(s)V_{BC}(s) and VB′CI(t)V^I_{B'C}(t) are not strictly local, they can be approximated by operators supported near the corresponding boundaries, with errors decaying exponentially in the truncation distance. This allows us to combine quasi-locality with exponential clustering to obtain an exponential bound in the inter-copy distance dsepd_{\mathrm{sep}}.

The following lemma is closely related to standard quasi-locality estimates for imaginary-time evolution. For completeness, we provide a self-contained proof below. The argument follows the usual nested-commutator expansion together with standard bounds on multicommutators; see, e.g., Appendix E.1 and Lemma 13 of Ref. [46].

Lemma 1. Let hXh_X be a local Hamiltonian term supported on XX, and let RR be a region containing XX. Define

hX(R)(β)=e−βHRhXeβHR,(64)h_X^{(R)}(\beta)=e^{-\beta H_R}h_Xe^{\beta H_R}, \tag*{(64)}

where HRH_R denotes the restriction of the Hamiltonian to the region RR. Assume 2gkβ<12gk\beta<1, and let

D=⌈d(X,Rˉ)r⌉,(65)D=\left\lceil\frac{d(X,\bar{R})}{r}\right\rceil, \tag*{(65)}

where Rˉ\bar{R} is the complement of RR. Then

∥hX(β)−hX(R)(β)∥≤2g(2gkβ)D1−2gkβ,(66)\left\|h_X(\beta)-h_X^{(R)}(\beta)\right\|\leq2g\frac{(2gk\beta)^D}{1-2gk\beta}, \tag*{(66)}

where

hX(β)=e−βHhXeβH.(67)h_X(\beta)=e^{-\beta H}h_Xe^{\beta H}. \tag*{(67)}

Proof. We compare the full imaginary-time evolution generated by HH with the truncated one generated by HRH_R. Expanding both in nested commutators, we obtain

hX(β)−hX(R)(β)=∑n=0∞(−β)nn!(ad⁡H(n)[hX]−ad⁡HR(n)[hX]),(68)h_X(\beta)-h_X^{(R)}(\beta)=\sum_{n=0}^{\infty}\frac{(-\beta)^n}{n!}\left(\operatorname{ad}_H^{(n)}[h_X]-\operatorname{ad}_{H_R}^{(n)}[h_X]\right), \tag*{(68)}

where ad⁡K(O)=[K,O]\operatorname{ad}_K(O)=[K,O] and ad⁡K(n)=ad⁡K∘⋯∘ad⁡K\operatorname{ad}_K^{(n)}=\operatorname{ad}_K\circ\cdots\circ\operatorname{ad}_K.

Each nn-fold commutator in Eq. (68) can be written as a sum over sequences of local Hamiltonian terms whose supports overlap successively, starting from the original support XX of hXh_X. Since each Hamiltonian term has spatial diameter at most rr, the support of the operator can expand by at most distance rr at each commutator step. Therefore, if

n<D=⌈d(X,Rˉ)r⌉,(69)n<D=\left\lceil\frac{d(X,\bar{R})}{r}\right\rceil, \tag*{(69)}

no sequence of nn overlapping local terms can reach the complement Rˉ\bar{R}. In that case, every term contributing to ad⁡H(n)[hX]\operatorname{ad}_H^{(n)}[h_X] is already contained entirely inside RR, and hence coincides with the corresponding term in ad⁡HR(n)[hX]\operatorname{ad}_{H_R}^{(n)}[h_X]. It follows that

ad⁡H(n)[hX]=ad⁡HR(n)[hX](n<D).(70)\operatorname{ad}_H^{(n)}[h_X]=\operatorname{ad}_{H_R}^{(n)}[h_X]\qquad(n<D). \tag*{(70)}

Thus the difference in Eq. (68) receives contributions only from orders n≥Dn\geq D.

For the remaining terms with n≥Dn\geq D, we bound the norm of the difference naively by estimating each nested commutator separately. Using

∥ad⁡A(B)∥=∥[A,B]∥≤2∥A∥∥B∥,(71)\left\|\operatorname{ad}_A(B)\right\|=\left\|[A,B]\right\|\leq2\left\|A\right\|\left\|B\right\|, \tag*{(71)}

together with the assumption

max⁡v∑Y:v∈Y∥hY∥≤g,(72)\max_{v}\sum_{Y:v\in Y}\left\|h_Y\right\|\leq g, \tag*{(72)}

and the fact that each interaction term acts on at most kk sites, one obtains the standard estimate

∥ad⁡H(n)[hX]∥≤2g (2gk)nn!,(73)\left\|\operatorname{ad}_{H}^{(n)}[h_X]\right\| \leq2g\,(2gk)^n n!, \tag*{(73)}

and similarly for ad⁡HR(n)[hX]\operatorname{ad}_{H_R}^{(n)}[h_X]. Hence

∥ad⁡H(n)[hX]−ad⁡HR(n)[hX]∥≤2g (2gk)nn!.(74)\left\|\operatorname{ad}_{H}^{(n)}[h_X]-\operatorname{ad}_{H_R}^{(n)}[h_X]\right\| \leq2g\,(2gk)^n n!. \tag*{(74)}

Substituting this into Eq. (68), we obtain

∥hX(β)−hX(R)(β)∥≤∑n=D∞βnn!∥ad⁡H(n)[hX]−ad⁡HR(n)[hX]∥\left\|h_X(\beta)-h_X^{(R)}(\beta)\right\| \leq\sum_{n=D}^{\infty}\frac{\beta^n}{n!}\left\|\operatorname{ad}_{H}^{(n)}[h_X]-\operatorname{ad}_{H_R}^{(n)}[h_X]\right\|
≤2g∑n=D∞(2gkβ)n.(75)\leq2g\sum_{n=D}^{\infty}(2gk\beta)^n. \tag*{(75)}

Since 2gkβ<12gk\beta<1, the sum is geometric and converges:

2g∑n=D∞(2gkβ)n=2g(2gkβ)D1−2gkβ.(76)2g\sum_{n=D}^{\infty}(2gk\beta)^n=2g\frac{(2gk\beta)^D}{1-2gk\beta}. \tag*{(76)}

This proves

∥hX(β)−hX(R)(β)∥≤2g(2gkβ)D1−2gkβ.(77)\left\|h_X(\beta)-h_X^{(R)}(\beta)\right\|\leq2g\frac{(2gk\beta)^D}{1-2gk\beta}. \tag*{(77)}

The same estimate holds for the opposite convention eβHhXe−βHe^{\beta H}h_Xe^{-\beta H}, since replacing β\beta by −β-\beta changes only the signs of the nested-commutator coefficients and not their norm bounds.

Using this opposite-sign counterpart of Lemma 1, we truncate the imaginary-time evolved boundary operators VBCI(s)V^I_{BC}(s) and VB′CI(t)V^I_{B'C}(t) within CC. Let R1,R2⊂CR_1,R_2\subset C denote the regions retained when truncating at distances ℓ1\ell_1 and ℓ2\ell_2 from the BCBC and B′CB'C boundaries, respectively. The corresponding truncated operators are denoted by VBCI(R1)(s)V^{I(R_1)}_{BC}(s) and VB′CI(R2)(t)V^{I(R_2)}_{B'C}(t). Since

VBCI(s)=∑X⊂∂(BC)hXI(s),(78)V^I_{BC}(s)=\sum_{X\subset\partial(BC)}h_X^I(s), \tag*{(78)}

Lemma 1 implies

∥VBCI(s)−VBCI(R1)(s)∥≤∑X⊂∂(BC)∥hXI(s)−hXI(R1)(s)∥\left\|V^I_{BC}(s)-V^{I(R_1)}_{BC}(s)\right\|\leq\sum_{X\subset\partial(BC)}\left\|h_X^I(s)-h_X^{I(R_1)}(s)\right\|
≤2g∣∂B∣(2gks)ℓ1/r1−2gks.\leq2g|\partial B|\frac{(2gks)^{\ell_1/r}}{1-2gks}.

Therefore,

∥VBCI(s)−VBCI(R1)(s)∥≤const.×∣∂B∣e−ℓ1/ξs,(79)\left\|V^I_{BC}(s)-V^{I(R_1)}_{BC}(s)\right\|\leq\mathrm{const.}\times|\partial B|e^{-\ell_1/\xi_s}, \tag*{(79)}

where

ξs=rlog⁡(1/2gks).(80)\xi_s=\frac{r}{\log(1/2gks)}. \tag*{(80)}

Similarly,

∥VB′CI(t)−VB′CI(R2)(t)∥≤const.×∣∂B′∣e−ℓ2/ξt.(81)\left\|V^I_{B'C}(t)-V^{I(R_2)}_{B'C}(t)\right\|\leq\mathrm{const.}\times|\partial B'|e^{-\ell_2/\xi_t}. \tag*{(81)}

Moreover, using the support condition in Eq. (25) and Eq. (28) of Ref. [47], for 2gkβ<12gk\beta< 1 we have

∥oA(β)∥≤k∥oA∥11−2gkβ=const.⁡×∥oA∥.(82)\lVert o_A(\beta)\rVert\le k\lVert o_A\rVert\frac{1}{1-2gk\beta}=\operatorname{const.}\times\lVert o_A\rVert. \tag*{(82)}

Combining these bounds yields

∣⟨VBCI(s)oA(β)VB′CI(t)⟩ρ~−⟨VBCI(s)oA(β)⟩ρ~⟨VB′CI(t)⟩ρ~∣≤const.⁡×∣∂B∣∥oA∥∥VB′C∥e−ℓ1/ξs+const.⁡×∣∂B′∣∥oA∥∥VBC∥e−ℓ2/ξt+∣⟨VBCI(R1)(s)oA(β)VB′CI(R2)(t)⟩ρ~−⟨VBCI(R1)(s)oA(β)⟩ρ~⟨VB′CI(R2)(t)⟩ρ~∣.(83)\begin{aligned} \left|\langle V_{BC}^{I}(s)o_A(\beta)V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}-\langle V_{BC}^{I}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}\right| &\le\operatorname{const.}\times|\partial B|\lVert o_A\rVert\lVert V_{B'C}\rVert e^{-\ell_1/\xi_s}\\ &\quad+\operatorname{const.}\times|\partial B'|\lVert o_A\rVert\lVert V_{BC}\rVert e^{-\ell_2/\xi_t}\\ &\quad+\left|\langle V_{BC}^{I(R_1)}(s)o_A(\beta)V_{B'C}^{I(R_2)}(t)\rangle_{\widetilde{\rho}}-\langle V_{BC}^{I(R_1)}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{I(R_2)}(t)\rangle_{\widetilde{\rho}}\right|. \tag*{(83)} \end{aligned}

Although oA(β)o_A(\beta) need not be supported on AA, it remains supported within ABAB, because its imaginary-time evolution is generated by HABH_{AB} alone. At this stage, the weaker clustering assumption is sufficient. Using the exponential clustering of connected correlations introduced in Eq. (39), we obtain

∣⟨VBCI(R1)(s)oA(β)VB′CI(R2)(t)⟩ρ~−⟨VBCI(R1)(s)oA(β)⟩ρ~⟨VB′CI(R2)(t)⟩ρ~∣≤const.⁡×∥oA(β)∥∥VBCI(R1)(s)∥∥VB′CI(R2)(t)∥e−(dsep−ℓ1−ℓ2)/ξ~≤const.⁡×∥oA∥∥VBC∥∥VB′C∥e−(dsep−ℓ1−ℓ2)/ξ~.(84)\begin{aligned} \left|\langle V_{BC}^{I(R_1)}(s)o_A(\beta)V_{B'C}^{I(R_2)}(t)\rangle_{\widetilde{\rho}}-\langle V_{BC}^{I(R_1)}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{I(R_2)}(t)\rangle_{\widetilde{\rho}}\right| &\le\operatorname{const.}\times\lVert o_A(\beta)\rVert\lVert V_{BC}^{I(R_1)}(s)\rVert\lVert V_{B'C}^{I(R_2)}(t)\rVert e^{-(d_{\mathrm{sep}}-\ell_1-\ell_2)/\widetilde{\xi}}\\ &\le\operatorname{const.}\times\lVert o_A\rVert\lVert V_{BC}\rVert\lVert V_{B'C}\rVert e^{-(d_{\mathrm{sep}}-\ell_1-\ell_2)/\widetilde{\xi}}. \tag*{(84)} \end{aligned}

Substituting this into the previous bound, we obtain

∣⟨VBCI(s)oA(β)VB′CI(t)⟩ρ~−⟨VBCI(s)oA(β)⟩ρ~⟨VB′CI(t)⟩ρ~∣≤const.⁡×∣∂B∣∣∂B′∣∥oA∥(e−ℓ1/ξs+e−ℓ2/ξt)+const.⁡×∥oA∥∥VBC∥∥VB′C∥e−(dsep−ℓ1−ℓ2)/ξ~.(85)\begin{aligned} \left|\langle V_{BC}^{I}(s)o_A(\beta)V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}-\langle V_{BC}^{I}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}\right| &\le\operatorname{const.}\times|\partial B||\partial B'|\lVert o_A\rVert\left(e^{-\ell_1/\xi_s}+e^{-\ell_2/\xi_t}\right)\\ &\quad+\operatorname{const.}\times\lVert o_A\rVert\lVert V_{BC}\rVert\lVert V_{B'C}\rVert e^{-(d_{\mathrm{sep}}-\ell_1-\ell_2)/\widetilde{\xi}}. \tag*{(85)} \end{aligned}

Choosing

ℓ1=ξsξ~+ξs+ξtdsep,ℓ2=ξtξ~+ξs+ξtdsep,(86)\ell_1=\frac{\xi_s}{\widetilde{\xi}+\xi_s+\xi_t}d_{\mathrm{sep}},\qquad\ell_2=\frac{\xi_t}{\widetilde{\xi}+\xi_s+\xi_t}d_{\mathrm{sep}}, \tag*{(86)}

and defining

ξ′=ξ~+ξs+ξt,(87)\xi'=\widetilde{\xi}+\xi_s+\xi_t, \tag*{(87)}

we arrive at

∣⟨VBCI(s)oA(β)VB′CI(t)⟩ρ~−⟨VBCI(s)oA(β)⟩ρ~⟨VB′CI(t)⟩ρ~∣≤const.⁡×∣∂B∣∣∂B′∣∥oA∥e−dsep/ξ′.(88)\left|\langle V_{BC}^{I}(s)o_A(\beta)V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}-\langle V_{BC}^{I}(s)o_A(\beta)\rangle_{\widetilde{\rho}}\langle V_{B'C}^{I}(t)\rangle_{\widetilde{\rho}}\right| \le\operatorname{const.}\times|\partial B||\partial B'|\lVert o_A\rVert e^{-d_{\mathrm{sep}}/\xi'}. \tag*{(88)}

Applying the same reasoning to the other connected term in Eq. (63), we conclude that the contributions containing ΔB(s,t)\Delta_B(s,t) and ΔB′(s,t)\Delta_{B'}(s,t) are exponentially suppressed in dsepd_{\mathrm{sep}}.

Bound on the double-boundary term BB′(s,t)BB'(s,t)

We finally consider the contribution containing ΔBB′(s,t)\Delta_{BB'}(s,t). In contrast to the single-boundary terms, the estimate of this term requires the stronger trace-norm clustering assumption introduced in Eq. (38).

Including the normalization factor in Eq. (43), we define

ρ~AB=(ρAB(0))2Tr⁡AB[(ρAB(0))2].(89)\widetilde{\rho}_{AB}=\frac{\left(\rho_{AB}^{(0)}\right)^2}{\operatorname{Tr}_{AB}\left[\left(\rho_{AB}^{(0)}\right)^2\right]}. \tag*{(89)}

and V‾BC(s,t)\overline{V}_{BC}(s,t) such that it satisfies

IA⊗V‾BC⊗IA′B′=SAB,A′B′(IABA′⊗VB′C)SAB,A′B′,V‾BC(s,t)=e−sHAB−tHCV‾BCesHAB+tHC.(90)I_A\otimes\overline{V}_{BC}\otimes I_{A'B'}=S_{AB,A'B'}\left(I_{ABA'}\otimes V_{B'C}\right)S_{AB,A'B'},\qquad\overline{V}_{BC}(s,t)=e^{-sH_{AB}-tH_C}\overline{V}_{BC}e^{sH_{AB}+tH_C}. \tag*{(90)}

The swap identity in Eq. (30) places the ABAB factor of VBCIV_{BC}^{I} first in both cross terms. Moving a Gibbs factor through the swapped interaction and using ρAB(0)oA=oA(β)ρAB(0)\rho_{AB}^{(0)}o_A=o_A(\beta)\rho_{AB}^{(0)} gives, for the first ordering,

SAB,A′B′VB′CI(t)SAB,A′B′=V‾BC(−t,−t),S_{AB,A'B'}V_{B'C}^{I}(t)S_{AB,A'B'}=\overline{V}_{BC}(-t,-t),
ρAB(0)V‾BC(−t,−t)(ρAB(0))−1=V‾BC(β−t,−t).\rho_{AB}^{(0)}\overline{V}_{BC}(-t,-t)\left(\rho_{AB}^{(0)}\right)^{-1}=\overline{V}_{BC}(\beta-t,-t).

For the opposite ordering, the CC factors must also be reordered inside their thermal trace. Thermal cyclicity shifts their time argument, so that

ρC(0)VˉBC(−s,−s)=VˉBC(−s,β−s)ρC(0),ρAB(0)VˉBC(−s,β−s)(ρAB(0))−1=VˉBC(β−s,β−s).\rho_C^{(0)}\bar{V}_{BC}(-s,-s)=\bar{V}_{BC}(-s,\beta-s)\rho_C^{(0)}, \qquad \rho_{AB}^{(0)}\bar{V}_{BC}(-s,\beta-s)(\rho_{AB}^{(0)})^{-1}=\bar{V}_{BC}(\beta-s,\beta-s).

The factorizable subtraction terms transform in the same way. Combining the two orderings, we obtain

Tr⁡ABA′B′[ρAB(0)⊗ρA′B′(0)ΔBB′(s,t)SAB,A′B′oA]Tr⁡AB[(ρAB(0))2]=Tr⁡AB[ρ~AB{Tr⁡C[ρBC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρBC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]}oA(β)]+Tr⁡AB[ρ~AB{Tr⁡C[ρBC(0)VBCI(t)VˉBC(β−s,β−s)]−Tr⁡C[ρBC(0)VBCI(t)]Tr⁡C[ρC(0)VˉBC(β−s,β−s)]}oA(β)].(91)\begin{aligned} \frac{\operatorname{Tr}_{ABA'B'}[\rho_{AB}^{(0)}\otimes\rho_{A'B'}^{(0)}\Delta_{BB'}(s,t)S_{AB,A'B'}o_A]}{\operatorname{Tr}_{AB}[(\rho_{AB}^{(0)})^2]} ={}&\operatorname{Tr}_{AB}\bigl[\tilde{\rho}_{AB}\{\operatorname{Tr}_{C}[\rho_{BC}^{(0)}V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{(0)}V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\}o_A(\beta)\bigr]\\ &+\operatorname{Tr}_{AB}\bigl[\tilde{\rho}_{AB}\{\operatorname{Tr}_{C}[\rho_{BC}^{(0)}V_{BC}^{I}(t)\bar{V}_{BC}(\beta-s,\beta-s)]-\operatorname{Tr}_{C}[\rho_{BC}^{(0)}V_{BC}^{I}(t)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-s,\beta-s)]\}o_A(\beta)\bigr]. \tag*{(91)} \end{aligned}

Figure S2 schematically shows the supports of VBCV_{BC} and VˉBC\bar{V}_{BC}. In particular, note that the two operators have overlapping support in region BB.

For the first term in Eq. (91), we bound

∣Tr⁡AB[ρ~AB{Tr⁡C[ρC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]}oA(β)]∣≤const.⁡×∥oA∥∥Tr⁡C[ρC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]∥.(92)\begin{aligned} &\left|\operatorname{Tr}_{AB}\bigl[\tilde{\rho}_{AB}\{\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\}o_A(\beta)\bigr]\right|\\ &\leq\operatorname{const.}\times\|o_A\|\left\|\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\right\|. \tag*{(92)} \end{aligned}

Applying the same quasi-local truncation as above, we obtain

∥Tr⁡C[ρC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]∥≤const.⁡×∣∂B∣∣∂B′∣(e−ℓ1/ξs+e−ℓ2/ξt)+∥Tr⁡C[ρBCI(R1)(s)VˉBC(R2)(β−t,−t)]−Tr⁡C[ρBCI(R1)(s)]Tr⁡C[VˉBC(R2)(β−t,−t)]∥.(93)\begin{aligned} &\left\|\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{C}^{(0)}V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\right\|\\ &\leq\operatorname{const.}\times|\partial B||\partial B'|\left(e^{-\ell_1/\xi_s}+e^{-\ell_2/\xi_t}\right)\\ &\quad+\left\|\operatorname{Tr}_{C}[\rho_{BC}^{I(R_1)}(s)\bar{V}_{BC}^{(R_2)}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{I(R_1)}(s)]\operatorname{Tr}_{C}[\bar{V}_{BC}^{(R_2)}(\beta-t,-t)]\right\|. \tag*{(93)} \end{aligned}

For the last term, the stronger trace-norm clustering assumption in Eq. (6) is needed. Tracing over the truncation regions R1R_1 and R2R_2 shown in Figure S2, we have

Schematic illustration of the supports of $V_{BC}$ and $\bar{V}_{BC}$

Figure S2. Schematic illustration of the supports of VBCV_{BC} and VˉBC\bar{V}_{BC}. The operator VBCV_{BC} acts on the boundary region between BB and CC, shown in light red. The operator VˉBC\bar{V}_{BC} acts on the interior of BB and on the part of CC adjacent to B′B', shown in light blue. The regions R1R_1 and R2R_2 indicate the parts of CC retained by the corresponding truncations.

∥Tr⁡C[ρBCI(R1)(s)VˉBC(R2)(β−t,−t)]−Tr⁡C[ρBCI(R1)(s)]Tr⁡C[VˉBC(R2)(β−t,−t)]∥=∥Tr⁡R1R2[(ρR1R2(0)−ρR1(0)⊗ρR2(0))VBCI(R1)(s)VˉBC(R2)(β−t,−t)]∥.(94)\begin{aligned} &\left\|\operatorname{Tr}_{C}[\rho_{BC}^{I(R_1)}(s)\bar{V}_{BC}^{(R_2)}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{I(R_1)}(s)]\operatorname{Tr}_{C}[\bar{V}_{BC}^{(R_2)}(\beta-t,-t)]\right\|\\ &=\left\|\operatorname{Tr}_{R_1R_2}\left[(\rho_{R_1R_2}^{(0)}-\rho_{R_1}^{(0)}\otimes\rho_{R_2}^{(0)})V_{BC}^{I(R_1)}(s)\bar{V}_{BC}^{(R_2)}(\beta-t,-t)\right]\right\|. \tag*{(94)} \end{aligned}

Using Eq. (38), this is bounded by

∥Tr⁡R1R2[(ρR1R2(0)−ρR1(0)⊗ρR2(0))VBCI(R1)(s)VˉBC(R2)(β−t,−t)]∥≤const.⁡×∥VBCI(R1)(s)∥∥VˉBC(R2)(β−t,−t)∥e−(dsep−ℓ1−ℓ2)/ξ∗≤const.⁡×∥VBC∥∥VB′C∥e−(dsep−ℓ1−ℓ2)/ξ∗.(95)\begin{aligned} \left\|\operatorname{Tr}_{R_1R_2}\left[(\rho_{R_1R_2}^{(0)}-\rho_{R_1}^{(0)}\otimes\rho_{R_2}^{(0)})V_{BC}^{I(R_1)}(s)\bar{V}_{BC}^{(R_2)}(\beta-t,-t)\right]\right\| &\leq\operatorname{const.}\times\|V_{BC}^{I(R_1)}(s)\|\|\bar{V}_{BC}^{(R_2)}(\beta-t,-t)\|e^{-(d_{\mathrm{sep}}-\ell_1-\ell_2)/\xi^*}\\ &\leq\operatorname{const.}\times\|V_{BC}\|\|V_{B'C}\|e^{-(d_{\mathrm{sep}}-\ell_1-\ell_2)/\xi^*}. \tag*{(95)} \end{aligned}

Substituting this back into the previous estimate, we obtain

∥Tr⁡C[ρBC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρBC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]∥≤const.⁡×∣∂B∣∣∂B′∣(e−ℓ1/ξs+e−ℓ2/ξt)+const.⁡×∥VBC∥∥VB′C∥e−(dsep−ℓ1−ℓ2)/ξ∗.(96)\begin{aligned} \bigl\| \operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\bigr\| \\ \leq\operatorname{const.}\times|\partial B||\partial B'|\left(e^{-\ell_{1}/\xi_{s}}+e^{-\ell_{2}/\xi_{t}}\right) \\ \quad+\operatorname{const.}\times\|V_{BC}\|\|V_{B'C}\|e^{-(d_{\mathrm{sep}}-\ell_{1}-\ell_{2})/\xi^{*}}. \tag*{(96)} \end{aligned}

Choosing

ℓ1=ξsξ∗+ξs+ξtdsep,ℓ2=ξtξ∗+ξs+ξtdsep,(97)\ell_{1}=\frac{\xi_{s}}{\xi^{*}+\xi_{s}+\xi_{t}}d_{\mathrm{sep}},\qquad\ell_{2}=\frac{\xi_{t}}{\xi^{*}+\xi_{s}+\xi_{t}}d_{\mathrm{sep}}, \tag*{(97)}

and defining

ξ′′=ξ∗+ξs+ξt,(98)\xi''=\xi^{*}+\xi_{s}+\xi_{t}, \tag*{(98)}

we arrive at

∥Tr⁡C[ρBC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρBC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]∥≤const.⁡×∣∂B∣∣∂B′∣e−dsep/ξ′′.(99)\begin{aligned} \bigl\| \operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\bigr\| \\ \leq\operatorname{const.}\times|\partial B||\partial B'|e^{-d_{\mathrm{sep}}/\xi''}. \tag*{(99)} \end{aligned}

Therefore, the first term in Eq. (91) is bounded by

∣Tr⁡AB[ρ~AB{Tr⁡C[ρBC(0)VBCI(s)VˉBC(β−t,−t)]−Tr⁡C[ρBC(0)VBCI(s)]Tr⁡C[ρC(0)VˉBC(β−t,−t)]}oA(β)]∣≤const.⁡×∣∂B∣∣∂B′∣∥oA∥e−dsep/ξ′′.(100)\begin{aligned} \bigl|\operatorname{Tr}_{AB}\bigl[\tilde{\rho}_{AB}\{\operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)\bar{V}_{BC}(\beta-t,-t)]-\operatorname{Tr}_{C}[\rho_{BC}^{(0)} V_{BC}^{I}(s)]\operatorname{Tr}_{C}[\rho_{C}^{(0)}\bar{V}_{BC}(\beta-t,-t)]\}o_{A}(\beta)\bigr]\bigr| \\ \leq\operatorname{const.}\times|\partial B||\partial B'|\|o_{A}\|e^{-d_{\mathrm{sep}}/\xi''}. \tag*{(100)} \end{aligned}

The second term in Eq. (91) is treated in the same way. Hence the contribution containing ΔBB′(s,t)\Delta_{BB'}(s,t) is also exponentially suppressed in dsepd_{\mathrm{sep}}.

Collecting the bounds on the Δ(s,t)\Delta(s,t), ΔB(s,t)\Delta_{B}(s,t), ΔB′(s,t)\Delta_{B'}(s,t), and ΔBB′(s,t)\Delta_{BB'}(s,t) contributions, we conclude that each second-order term in Eq. (43) decays exponentially with dsepd_{\mathrm{sep}}. Therefore,

∣⟨oA⟩VAP−⟨oA⟩LVP∣≤const.⁡×λ2∣∂B∣∣∂B′∣∥oA∥e−dsep/ξeff+O(λ3),(101)|\langle o_{A}\rangle_{\mathrm{VAP}}-\langle o_{A}\rangle_{\mathrm{LVP}}|\leq\operatorname{const.}\times\lambda^{2}|\partial B||\partial B'|\|o_{A}\|e^{-d_{\mathrm{sep}}/\xi_{\mathrm{eff}}}+O(\lambda^{3}), \tag*{(101)}

for some effective correlation length ξeff\xi_{\mathrm{eff}}. This proves Proposition 1.

References

  1. [1]F. G. S. L. Brandão and M. J. Kastoryano, Finite correlation length implies efficient preparation of quantum thermal states, Communications in Mathematical Physics 365, 1–16 (2018).arxiv.org/abs/1609.07877
  2. [2]C.-F. Chen, M. Kastoryano, F. G. S. L. Brandão, and A. Gilyén, Efficient quantum thermal simulation, Nature 646, 561–566 (2025).DOI
  3. [3]C. Rouzé, D. Stilck França, and A. M. Alhambra, Optimal quantum algorithm for gibbs state preparation, Phys. Rev. Lett. 136, 060601 (2026).DOI
  4. [4]C. Rouzé, D. Stilck França, and A. M. Alhambra, Efficient thermalization and universal quantum computing with quantum gibbs samplers, Nature Physics (2026).
  5. [5]J. Wu and T. H. Hsieh, Variational thermal quantum simulation via thermofield double states, Phys. Rev. Lett. 123, 220502 (2019).DOI
  6. [6]M. Motta, C. Sun, A. T. K. Tan, M. J. O’Rourke, E. Ye, A. J. Minnich, F. G. S. L. Brandão, and G. K.-L. Chan, Determining eigenstates and thermal states on a quantum computer using quantum imaginary time evolution, Nature Physics 16, 205–210 (2019).DOI
  7. [7]D. Zhu, S. Johri, N. M. Linke, K. A. Landsman, C. Huerta Alderete, N. H. Nguyen, A. Y. Matsuura, T. H. Hsieh, and C. Monroe, Generation of thermofield double states and critical ground states with a quantum computer, Proceedings of the National Academy of Sciences 117, 25402–25406 (2020).
  8. [8]R. Sagastizabal, S. P. Premaratne, B. A. Klaver, M. A. Rol, V. Negîrneac, M. S. Moreira, X. Zou, S. Johri, N. Muthusubramanian, M. Beekman, C. Zachariadis, V. P. Ostroukh, N. Haider, A. Bruno, A. Y. Matsuura, and L. DiCarlo, Variational preparation of finite-temperature states on a quantum computer, npj Quantum Information 7, 130 (2021).DOI
  9. [9]J. Cotler, S. Choi, A. Lukin, H. Gharibyan, T. Grover, M. E. Tai, M. Rispoli, R. Schittko, P. M. Preiss, A. M. Kaufman, M. Greiner, H. Pichler, and P. Hayden, Quantum virtual cooling, Phys. Rev. X 9, 031013 (2019).
  10. [10]B. Koczor, Exponential error suppression for near-term quantum devices, Phys. Rev. X 11, 031057 (2021).
  11. [11]W. J. Huggins, S. McArdle, T. E. O’Brien, J. Lee, N. C. Rubin, S. Boixo, K. B. Whaley, R. Babbush, and J. R. McClean, Variational distillation for quantum error mitigation, Phys. Rev. X 11, 041036 (2021).DOI
  12. [12]T. E. O’Brien et al., Purification-based quantum error mitigation of pair-correlated electron simulations, Nature Physics 19, 1787–1792 (2023).DOI
  13. [13]Z. Cai, Resource-efficient purification-based quantum error mitigation, arXiv:2107.02779 (2021).
  14. [14]P. Czarnik, A. Arrasmith, L. Cincio, and P. J. Coles, Qubit-efficient exponential suppression of errors, arXiv:2102.06056 (2021).arxiv.org/abs/2102.06056
  15. [15]P. Vikstald, G. Ferrini, and S. Puri, Study of noise in virtual distillation circuits for quantum error mitigation, Quantum 8, 1441 (2024).
  16. [16]H. Hakoshima, S. Endo, K. Yamamoto, Y. Matsuzaki, and N. Yoshioka, Localized virtual purification, Phys. Rev. Lett. 133, 080601 (2024).DOI
  17. [17]A. K. Eckert, C. M. Alves, D. K. L. Oi, M. Horodecki, P. Horodecki, and L. C. Kwek, Direct estimations of linear and nonlinear functionals of a quantum state, Phys. Rev. Lett. 88, 217901 (2002).
  18. [18]H. Araki, Gibbs states of a one dimensional quantum lattice, Communications in Mathematical Physics 14, 120–157 (1969).DOI
  19. [19]Y. M. Park and H. J. Yoo, Uniqueness and clustering properties of gibbs states for classical and quantum unbounded spin systems, Journal of Statistical Physics 80, 223–271 (1995).DOI
  20. [20]D. Ueltschi, Cluster expansions and correlation functions, Moscow Mathematical Journal 4, 511–522 (2004).
  21. [21]M. Kliesch, C. Gogolin, M. J. Kastoryano, A. Riera, and J. Eisert, Locality of temperature, Phys. Rev. X 4, 031019 (2014).
  22. [22]J. Fröhlich and D. Ueltschi, Some properties of correlations of quantum lattice systems in thermal equilibrium, Journal of Mathematical Physics 56, 053302 (2015).DOI
  23. [23]A. Bluhm, A. Capel, and A. Pérez-Hernández, Exponential decay of mutual information for gibbs states of local hamiltonians, Quantum 6, 650 (2022).DOI
  24. [24]A. Bluhm, A. Capel, and A. Pérez-Hernández, Strong decay of correlations for gibbs states in any dimension, Journal of Statistical Physics 192, 134 (2025).
  25. [25]See Supplemental Material for details (URL to be added).
  26. [26]F. Verstraete, J. J. García-Ripoll, and J. I. Cirac, Matrix product density operators: Simulation of finite-temperature and dissipative systems, Phys. Rev. Lett. 93, 207204 (2004).pubmed.ncbi.nlm.nih.gov/15600964
  27. [27]M. Zwolak and G. Vidal, Mixed-state dynamics in one-dimensional quantum lattice systems: A time-dependent superoperator renormalization algorithm, Phys. Rev. Lett. 93, 207205 (2004).DOI
  28. [28]A. E. Feiguin and S. R. White, Finite-temperature density matrix renormalization using an enlarged hilbert space, Phys. Rev. B 72, 220401 (2005).DOI
  29. [29]M. Fishman, S. R. White, and E. M. Stoudenmire, The ITensor Software Library for Tensor Network Calculations, SciPost Phys. Codebases , 4 (2022).DOI
  30. [30]M. Fishman, S. R. White, and E. M. Stoudenmire, Codebase release 0.3 for ITensor, SciPost Phys. Codebases , 4 (2022).DOI
  31. [31]Y. Zhang, Y. Wu, Y. Zhou, and X. Yuan, Measuring less to learn more: Quadratic speedup in learning nonlinear properties of quantum states, Phys. Rev. Lett. 136, 130602 (2026).DOI
  32. [32]Z. Du, Y. Tang, A. Elben, I. Roth, J. Eisert, and Z. Liu, Optimal randomized measurements for a family of nonlinear quantum properties, PRX Quantum 7, 010360 (2026).arxiv.org/abs/2505.09206
  33. [33]Z. Li, D. Chen, D. Qin, Y. Yang, and Y. Zhou, Quantum nonlinear properties from a single measurement setting, arXiv:2605.09958 (2026).arxiv.org/abs/2605.09958
  34. [34]A. Seif, Z.-P. Cian, S. Zhou, S. Chen, and L. Jiang, Shadow distillation: Quantum error mitigation with classical shadows for near-term quantum processors, PRX Quantum 4, 010303 (2023).DOI
  35. [35]Q. Liu, Z. Li, X. Yuan, H. Zhu, and Y. Zhou, Auxiliary-free replica shadows: Efficient estimation of multiple nonlinear quantum properties, Phys. Rev. Lett. 136, 100602 (2026).DOI
  36. [36]N. Yoshioka, H. Hakoshima, Y. Matsuzaki, Y. Tokunaga, Y. Suzuki, and S. Endo, Generalized quantum subspace expansion, Phys. Rev. Lett. 129, 020502 (2022).DOI
  37. [37]B. Yang, N. Yoshioka, H. Harada, S. Hakkaku, Y. Tokunaga, H. Hakoshima, K. Yamamoto, and S. Endo, Resource-efficient generalized quantum subspace expansion, Phys. Rev. Appl. 23, 054021 (2025).DOI
  38. [38]H. Kwon, C. Oh, Y. Lim, H. Jeong, S-W. Lee, and L. Jiang, Virtual purification complements quantum error correction in quantum metrology, npj Quantum Information 12, 88 (2026).DOI
  39. [39]M. Huo and Y. Li, Dual-state purification for practical quantum error mitigation, Phys. Rev. A 105, 022427 (2022).DOI
  40. [40]A. Karim, H. J. Vallury, and M. Usman, Fictitious copy quantum error mitigation, arXiv:2603.09302 (2026).arxiv.org/abs/2603.09302
  41. [41]A. Anshu, S. Arunachalam, T. Kuwahara, and M. Soleimanifar, Sample-efficient learning of interacting quantum systems, Nature Physics 17, 931–935 (2021).DOI
  42. [42]J. Haah, R. Kothari, and E. Tang, Learning quantum Hamiltonians from high-temperature Gibbs states and real-time evolutions, Nature Physics 20, 1027–1031 (2024).DOI
  43. [43]A. Bakshi, A. Liu, A. Moitra, and E. Tang, Learning quantum Hamiltonians at any temperature in polynomial time, in Proceedings of the 56th Annual ACM Symposium on Theory of Computing, STOC ’24 (ACM, 2024), p. 1470–1477.
  44. [44]J. R. Garrison and T. Grover, Does a single eigenstate encode the full hamiltonian?, Phys. Rev. X 8, 021026 (2018).DOI
  45. [45]A. Iwaki, VAP Simulations, GitHub repository, https://github.com/AtshushiIwaki/vap-simulations (2026).
  46. [46]T. Kuwahara, A. M. Alhambra, and A. Anshu, Improved thermal area law and quasilinear time algorithm for quantum Gibbs states, Phys. Rev. X 11, 011047 (2021).DOI
  47. [47]A. M. Alhambra, Quantum many-body systems in thermal equilibrium, PRX Quantum 4, 040201 (2023).DOI

Paper details

vlnc:W_5p9yh55g
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