arXiv:2609.25992
Maximum-Likelihood Amplitude Estimation for Quantum Monte Carlo Integration on Trapped-Ion Hardware: Convergence, Noise-Floor Saturation, Depth-Dependent Bias
Aditi Lal, Alex Khan, Rut Lineswala, Abhishek Chopra
Abstract
Quantum Amplitude Estimation promises to accelerate Monte Carlo integration by replacing the classical $\varepsilon \propto N^{-1/2}$ sampling law with the Heisenberg-limited $\varepsilon \propto N^{-1}$ scaling, but whether any part of that advantage survives on present-day hardware is an empirical question. We report a systematic experimental benchmark of a Maximum-Likelihood Amplitude Estimation QMC pipeline on IonQ trapped-ion processors, comprising 60 hardware trials on IonQ Forte-1 across five settings(5 & 9 qubits, 100-200 shots per depth, linear & exponential amplification schedules, ~113 device-hours), together with 100 trials on IonQ Aria-1 & Forte-1 noise models at 5, 9 & 19 qubits. The pipeline combines controlled-$R_y$ rotational state preparation, Grover-based amplitude amplification and maximum-likelihood inference and is benchmarked against $b_{\max}^{-1}\!\int_0^{b_{\max}}\!\sin^2\!x\,\mathrm{d}x$ with $b_{\max}=π/5$ ,given results emerge. First, the noiseless limit it attains $\varepsilon \propto N^{-0.88}$, close to the Heisenberg-scaling reference, reducing the error 54-74x over a 133x increase in oracle calls. Second, scaling does not survive on hardware. Every noisy backend saturates at an error floor of (2-5)x 10^{-2}, with fitted exponents of 0.04-0.15 on Forte-1 linear schedules and $α\le\!0$ on the vendor noise models: beyond the optimal shallow depth, however, deeper amplification provides no sustained accuracy gain and instead returns to the noise floor. Forte-1 hardware also outperformed IonQ's own Aria-1 noise model at every depth $m \ge 2$, indicating that vendor noise models are bad predictors of MLAE accuracy. The practical implication for UQ is that on current devices the useful operating point is a shallow, schedule-tuned amplification depth chosen so that the deepest circuit lands near p~1/2 not the deepest schedule the coherence budget allows.
AI Summary
The paper asks whether maximum-likelihood amplitude estimation (MLAE) retains improved query scaling on current trapped-ion hardware, and how amplification depth affects accuracy. It benchmarks a Grover-based integration pipeline for $b^{-1}\int_0^b\sin^2x\,dx$ at $b=\pi/5$, using rotational state preparation and cumulative maximum-likelihood inference. Errors are measured against the discretised integral, excluding discretisation error. The campaign includes 60 IonQ Forte-1 hardware trials across five configurations (5 or 9 qubits, 100–200 shots per depth), approximately 113 recorded device-hours, and 100 vendor-noise-model trials at 5, 9 and 19 qubits. Linear schedules reach $m=6$ Grover iterations; exponential schedules reach $m=32$. Two noiseless simulation runs give fitted error exponents $\alpha=0.841$ and $0.923$ in $\varepsilon\propto N^{-\alpha}$, with errors decreasing 54–74-fold over a 133-fold increase in oracle calls. Noisy runs instead return to errors around $(2–5)\times10^{-2}$ after shallow-depth minima. Hardware linear-schedule exponents are $0.04–0.15$; exponential-schedule hardware and vendor-model fits have negative exponents. Linear hardware schedules reproducibly attain a minimum at $m=3$, reaching mean absolute error $1.61\times10^{-3}$ for 9 qubits and 200 shots. Increasing shots does not remove the subsequent error floor. Matched-width comparisons show Forte-1 hardware outperforming an Aria-1 noise model, but these represent different devices. Stated limitations include one integrand and amplitude, 10–20 hardware trials per configuration, unmonitored device drift, non-general state preparation, and no IonQ debiasing. The depth optimum is amplitude-dependent, and query counts do not establish a wall-clock advantage.
Central claim may not hold: The paper claims that a fitted single-parameter depolarizing model explains the depth dependence, but supplies no such calculation or comparison, leaving its general near-one-half scheduling prescription unestablished. Comparing Forte-1 hardware with an Aria-1 noise model does not establish that vendor noise models inaccurately predict their corresponding devices. The tested depths and 100–200-shot comparisons demonstrate late-depth degradation but do not establish a shot-independent limiting bias floor. The claimed approximately 13-fold quantum query saving uses Bernoulli variance instead of the ordinary integration estimator's variance, overstating the classical sample requirement by approximately a factor of ten.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 7 pages 3 figures