Source-linked AI summary
Resource-efficient Purification-based Quantum Error Mitigation
Zhenyu Cai
TL;DR
Noisy near-term hardware and limited qubit counts make full quantum error correction challenging, while purification-based proposals face limits in error suppression or long-range gate costs. The paper combines state verification with multi-copy purification and reports arbitrary-order error suppression with fewer qubits and long-range gates than multi-copy purification alone.
Problem
Purification-based approaches either require many long-range gates or, with state verification, suppress only leading-order state-preparation errors.
Source: §I · p. 1Method
The paper applies state verification on top of multi-copy purification, enabling 2Mth-degree purification with M copies.
Source: §V · p. 5, §IV · p. 5Results
The combined scheme suppresses state-preparation errors to any order while requiring half the qubits and less than half the additional long-range gates of multi-copy purification.
Source: §V · p. 5Takeaways & Limitations
For fourth-degree purification with two copies, the combined scheme uses one control-copy-swap instead of three, reducing qubit and additional-gate requirements.
Source: §IV · p. 5
Abstract
from arXiv · showhide
To achieve the practical applications of near-term noisy quantum devices, low-cost ways to mitigate the noise damages in the devices are essential. In many applications, the noiseless state we want to prepare is often a pure state, which has recently inspired a range of purification-based quantum error mitigation proposals. The existing proposals either are limited to the suppressions of only the leading-order state preparation errors, or require a large number of long-range gates that might be challenging to implement depending on the qubit architecture. This article will provide an overview of the different purification-based quantum error mitigation schemes and propose a resource-efficient scheme that can correct state preparation errors up to any order while requiring only half of the qubits and less than half of the long-range gates compared to before.
I. INTRODUCTION
Noise and limited qubit counts make full quantum error correction challenging, motivating quantum error mitigation. The section introduces purification-based approaches and explains how increasing purification degree can reduce state-preparation infidelity, subject to a coherent-mismatch floor.
I. INTRODUCTION
- Noise and limited qubit counts make full quantum error correction challenging, so quantum error mitigation seeks ideal measurement statistics through additional runs on noisy hardware.Examples include error extrapolation, quasi-probability methods, and symmetry verification [7].Source: §I · p. 1
- Multi-copy purification can suppress state-preparation noise exponentially but requires long-range gates, while state verification uses fewer gates but removes only leading-order errors.The introduction proposes combining the approaches to address these constraints.Source: §I · p. 1
II. PURIFICATION-BASED QUANTUM ERROR MITIGATION
- Purification-based quantum error mitigation extracts the pure-state component of a noisy mixed state; its general idea was discussed by Koczor [8] and Huggins et al. [9].The discussion assumes the ideal state is the noisy state's dominant eigenvector; similar states prepared by similar circuits can also be purified together [8] [12].Source: §II · p. 1, A. State Purification · pp. 1–2
A. State Purification
- Purification moves the input state toward its dominant eigenvector, and its infidelity is exponentially suppressed as degree M increases when pϵ is small.The infidelity does not increase with M when pϵ ≤ 0.5.Source: A. State Purification · p. 2
- If the dominant eigenvector differs from the ideal state, even infinite-degree purification leaves a floor set by their infidelity, called coherent mismatch [8] or noise floor [9].The article sets this effect aside when comparing schemes because it affects them largely the same way.Source: A. State Purification · p. 2
B. Purified Observable
Purification reduces the bias of an observable estimate as degree increases, but measurement-circuit noise is not itself mitigated. The section also relates sampling overhead to the strength of the purified-state signal in the input.
B. Purified Observable
- The bias in the purified expectation value is suppressed exponentially as purification degree M increases.The bias is linked to the purified state's infidelity against the ideal state.Source: B. Purified Observable · p. 2
- The variance and sampling overhead depend largely on the normalization signal in the input state, with weaker signal associated with larger variance.The overhead is also mostly determined by the input noisy state and purification degree M.Source: B. Purified Observable · p. 2
- Noise from additional circuit structures used to measure the purified observable is not mitigated, motivating minimal additional gate cost for a given purification degree.The purified observable is measured using the purified state rather than the original noisy state.Source: B. Purified Observable · p. 2
A. Purification using Multiple Noisy Copies
Multi-copy purification estimates a purified observable from M noisy copies using cyclic permutations. Its resource and noise costs arise from inter-copy long-range control-swaps, whose errors remain unsuppressed.
A. Purification using Multiple Noisy Copies
- M noisy-state copies are measured with the cyclic permutation operator C_M and the observable O acting on copy 1.The resulting expectation value is used to obtain the purified observable.Source: A. Purification using Multiple Noisy… · p. 2
- In typical 2D layouts, corresponding qubits in different copies are far apart, making the physical control-swaps long-range.The swap operations used to address connectivity, like the control-swap errors, are not suppressed and can limit bias reduction as M increases.Source: A. Purification using Multiple Noisy… · pp. 2–3
- Figure 1 depicts fourth-degree purification circuits for different schemes, with thick lines representing whole registers and register-level control-swaps.Figure 2 shows how control-swaps between registers decompose into control-swaps between qubits.Source: Figure 1 · p. 3, Figure 2 · p. 3, A. Purification using Multiple Noisy… · p. 3
B. Purification using State Verification
State verification measures a second-degree purified observable using the noisy state and a dual state. Its use of one state copy avoids inter-copy control-swaps, but it suppresses only leading-order state-preparation errors.
B. Purification using State Verification
- The implementable inverse is the noisy version of the inverse circuit, so verification measures the dual state rather than ρ.The dual state is positive semidefinite but may not be normalized.Source: B. Purification using State Verification · p. 4
- State verification measures Tr(ρOρ), equivalent to second-degree purification with one occurrence of ρ replaced by the dual state.The observable can be viewed as ρO measured on state ρ, using Hadamard tests or non-destructive measurements of O.Source: B. Purification using State Verification · pp. 3–4
- The dominant eigenvectors of ρ and its dual are expected to be very similar, so dual-state purification is expected to perform similarly to usual second-degree purification.The passage attributes this expectation to the circuit and its inverse using similar unitary gates.Source: B. Purification using State Verification · p. 4
- Using one state copy avoids control-swaps and connectivity workarounds, but state verification reaches only second-degree purification and suppresses only leading-order preparation errors.Its intermediate observable measurement is the remaining unaccounted-for error in the described scheme.Source: B. Purification using State Verification · p. 4
1. Multi-copy Purification with Recycled Qubits
Recycled-qubit multi-copy purification reduces the required number of copies by reusing two quantum registers, but trades that saving for a deeper circuit.
1. Multi-copy Purification with Recycled Qubits
- Only two quantum registers are needed for every purification degree: one serves as the master copy and the other prepares the remaining copies sequentially.The registers are reused to prepare the M−1 additional copies one by one.Source: §1 · p. 4
- The full circuit depth is M−1 times larger than in the original circuit, constraining use through qubit lifetime and initialization time.Source: §1 · p. 4
- Rearranging the circuit does not reduce the gate count, so the scheme trades circuit depth for fewer copies.Source: §1 · p. 4
2. Phase Estimation with State Verification
The paper describes phase estimation and linear combinations of purified states, then combines state verification with multi-copy purification to reach higher purification degrees using fewer resources.
2. Phase Estimation with State Verification
- Phase estimation with state verification fits the signal Tr(ρU(t)ρ) to estimate the expectation value of O, normalizing the fitted curve to 1 at t=0.Source: §2 · pp. 4–5
3. Linear Combination of Purified States
IV. RESOURCE-EFFICIENT PURIFICATION
- With M copies, the combined scheme achieves 2Mth-degree purification using M−1 rather than 2M−1 control-copy-swaps.Source: §IV · p. 5
- At M=2, fourth-degree purification uses 1 control-copy-swap instead of 3 and 2 rather than 4 copies, halving qubits and reducing additional gates threefold.Source: §IV · p. 5
- Applying state verification to only a subset of noisy copies can produce odd-degree purification and combine with extensions discussed earlier.Source: §IV · p. 5
V. CONCLUSION
The conclusion contrasts the resource demands of multi-copy purification with state verification’s degree limit, then presents their combination as a resource-efficient scheme for higher-order error suppression.
V. CONCLUSION
- The combined scheme suppresses state-preparation errors to any order while requiring half the qubits and less than half the additional long-range gates of multi-copy purification.Source: §V · p. 5
- Multi-copy purification is constrained by qubit requirements and additional long-range control-swaps, while state verification alone reaches only second-degree purification.Source: §V · p. 5
- For fourth-degree purification, the combined scheme uses half the qubits and one-third the additional long-range gates required by the earlier approach.Source: §V · p. 5
- The authors conclude that the combined scheme should be among the most practical choices for purification-based QEM in real experiments.Source: §V · p. 5
Appendix A: Measuring Products of Operators
The appendix explains how a Hadamard test measures products of operators when one factor is unitary, and how its ancilla readout yields real or imaginary parts.
1. Using Hadamard Test
- The Hadamard test measures SU when U is unitary and S is a directly measurable Hermitian operator, using controlled-U gates.Source: §1 · p. 6
- The circuit measures the product SU even when S and U do not commute, so it is not equivalent to measuring U non-destructively and then measuring S.Source: §1 · p. 6
- The X⊗S expectation gives Re(Tr(SUρ)), while Y⊗S gives Im(Tr(SUρ)).Source: Figure 6 · p. 6, §1 · p. 6
2. Using non-Destructive Measurement
The method measures the product of a directly measurable Hermitian operator S and an involution G, under conditions that allow their measurement results to be combined. Sequentially measuring G and then S removes the contribution from the component that anticommutes with G.
2. Using non-Destructive Measurement
- The method applies when the two operators commute or when one is an involution G, whose square is the identity.The passage identifies G as the involution to be measured non-destructively, while S is a directly measurable Hermitian operator.Source: §2 · p. 6
- G has eigenvalues ±1 and is therefore Hermitian and unitary, with projectors onto its two eigenspaces.The passage states that the projectors correspond to the ±1 eigensubspaces.Source: §2 · p. 6
- Measuring G first projects the input state into one of its eigenspaces; the subsequent S measurement has no expectation-value contribution from the component anticommuting with G.The anticommuting component has expectation value zero after measuring G, regardless of the measurement outcome.Source: §2 · p. 6
- The product of the G and S measurement results captures the contribution from the component of S that commutes with G; combining the stated equations yields Tr(SGρ).The passage says this method was introduced in Ref. and describes applying R+ and R− before combining the results.Source: §2 · pp. 6–7