arXiv:2610.05222
Efficient Fidelity Estimation of General Quantum Resource States via Clifford Circuits
Shunta Takahashi, Keisuke Fujii
Abstract
Fidelity estimation is a fundamental tool for assessing quantum resource states across quantum information science. For small infidelity $ε$, tomography and direct fidelity estimation (DFE) have worst-case sample complexity $O(1/ε^2)$ at fixed relative accuracy and confidence. Direct target projections achieve $O(1/ε)$ but generally require non-Clifford operations whose reliability depends on the resource states that are typically the target of benchmarking. Here, we introduce the Bell--coherence fidelity estimation (BCFE) with optimal sample complexity $Θ(1/ε)$ using only Clifford gates and Pauli measurements. We prove that BCFE is applicable to arbitrary pure targets, possibly multi-qubit, with the statistical error of the single-copy coherence part explicitly controlled. To evaluate BCFE in a practical setting, we apply BCFE to analog rotation resource states in the candidate early-FTQC architecture called the Space-Time Efficient Analog Rotations (STAR) architecture. In the simulations using surface codes, BCFE is estimated to require up to approximately $27$ times fewer resource states than DFE for the same standard error in the infidelity estimate, while its estimates remain close to reference values despite circuit-level noise in the additional Bell measurement operations. These results support efficient resource benchmarking within fault-tolerant quantum computing.
AI Summary
The paper asks whether small infidelity $\epsilon=1-\langle\psi|\rho|\psi\rangle$ to a known pure target can be estimated with optimal $1/\epsilon$ copy scaling using only Clifford gates and Pauli measurements, without trusted non-Clifford target projections. Bell–coherence fidelity estimation (BCFE) combines two-copy Bell measurements with single-copy Pauli data. For one qubit, the singlet probability is $b=\det\rho=F(1-F)-|c|^2$, where $c$ is coherence in the target basis. Positivity gives $b\le\epsilon$ and $|c|\le\sqrt\epsilon$, allowing both contributions to be resolved with $O(\epsilon^{-1})$ copies. Concentration bounds establish relative error $r$ with failure probability $\delta$ using $O(r^{-2}\epsilon^{-1}\log(1/\delta))$ copies for $0<\epsilon\le1/4$. A hypothesis-testing argument gives a matching lower bound, including against collective measurements. For $m$ qubits, an exact purity–distance identity yields an extension with copy bounds $O[(D/(r^2\epsilon)+D^2/(r\epsilon))\log(1/\delta)]$ using Pauli strings, or $O[(1/(r^2\epsilon)+D/(r\epsilon))\log(1/\delta)]$ using global Clifford shadows, where $D=2^m$. Thus optimality concerns estimation parameters at fixed dimension, not qubit scaling. Simulations of analog-rotation resource states in distance-three surface codes, with CNOT error probabilities $0.001$–$0.005$, report standard errors $1.3$–$5.2$ times smaller than direct fidelity estimation, corresponding to up to approximately $27$-fold lower copy requirements. Circuit noise produces residual bias, particularly at higher rates. Guarantees assume independent, identically prepared copies; correlated preparation and drift remain unresolved. Global-Clifford circuit and classical-processing costs can be substantial, and dimension dependence remains open.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 26 pages, many figures