arXiv:2610.04247
Q-MERGE: Parallelising Quantum State Preparation for Large-Scale Classical Data
Archie Butterworth, Jens Renders, Jingbo Wang
Abstract
Quantum processing of classical data fundamentally relies on efficiently mapping classical datasets onto the amplitudes of quantum states. Preparing large amplitude-encoded states, however, remains a major bottleneck in quantum computing. In this paper, we introduce a novel state-preparation framework we call Q-MERGE, that addresses this bottleneck by partitioning a large target state into $M$ $n$-qubit segments, preparing these segments independently and in parallel, and then coherently combining them into a single amplitude-encoded state using SELECT-SWAP operations and measurement. Q-MERGE is agnostic to the segment-level preparation method, allowing existing techniques to be applied to smaller subproblems while providing a tunable trade-off between circuit depth and qubit count. With mid-circuit measurement and preparation-register reuse, the required ancilla qubits can be reduced from O(Mn) to O(n). Applied to a real-world $32 x 1024$ ultrasound dataset, Q-MERGE achieves an infidelity of $3.5 x 10^{-9}$, compared with $6.2 x 10^{-3}$ for direct preparation using the same underlying method, a more than six orders of magnitude improvement at equal CNOT circuit depth. We demonstrate its feasibility experimentally with a 53 qubit circuit comprised of 954 1-qubit gates and 478 2-qubit gates, on the Quantinuum System Model H2 trapped-ion quantum computer. The prepared state is validated using shadow-overlap tomography. Numerical analysis of Haar-random states indicates consistent success probability for up to $M=10^7$ segments, supporting the scalability of Q-MERGE for encoding massive classical datasets.
AI Summary
Q-MERGE addresses amplitude encoding of classical data by splitting a target state into $M$ normalized $n$-qubit segments, preparing them independently, and combining them without controlled versions of the preparation circuits. SELECT-SWAP operations followed by measurement produce a coherent stack with measurement-dependent weights. A diagonal block encoding restores the target weights, succeeding with probability $1/\beta^2$, where $\beta=\max_i|\sigma_i/w_i|$. The framework accommodates arbitrary segment-preparation methods. Its stated compression depth is $O(M\log^3 m+Mn)$, with $m=\log_2 M$; mid-circuit measurement and register reuse reduce preparation ancillas from $O(Mn)$ to $O(n)$, trading away parallelism. Classically optimized shallow brickwall circuits reduce the inverse participation ratio before preparation, with their inverses applied afterward to recover the target. For a $32\times1024$ ultrasound dataset, numerical benchmarks using Schmidt Spectrum Optimisation obtain infidelity $3.5\times10^{-9}$ versus $6.2\times10^{-3}$ for direct preparation at the same CNOT depth, approximately $1.04\times10^4$. Three preconditioning layers raise mean reweighting success from $0.0098$ to $0.317$. Haar-random-state simulations report success probabilities above $0.05$ through $M=10^7$. A 53-qubit Quantinuum H2 experiment implements Stage 1 using 478 two-qubit gates. Branch-conditioned shadow-overlap measurements, converted through simulated calibration curves, yield a weighted average estimated branch fidelity of $0.7135\pm0.0720$; full reweighting is not demonstrated experimentally. Reweighting can become prohibitively unlikely for unfavorable amplitude distributions. The paper leaves dataset-dependent preconditioning effectiveness, alternative flattening objectives, and storage-register initialization for future investigation; the large-$M$ success claims are numerical rather than general guarantees.
Central claim may not hold: The hardware measurements certify only index-conditioned segment states and cannot distinguish coherent merging from an incoherent mixture, so they do not certify the merged-state fidelity or the final amplitude-encoded target. Register reuse achieves O(n) preparation ancillas, but retaining the advertised execution depth requires enough preparation throughput to hide preparation and reset latency, which is not guaranteed for arbitrary segment-preparation methods with a fixed number of registers. The Haar-random success-probability numerics do not establish general scalability under shallow preconditioning because no bound connects the proposed preconditioning circuits or their IPR objective to those success probabilities.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- Updated with improved results and expanded discussion