arXiv:2610.03260
Compiling Together: High-Throughput Distributed Quantum Computing via Multi-Compilation
Yipei Liu, Sen Zhang, Zebo Yang, Lei Yang
Abstract
Quantum computing is a promising paradigm for problems that are challenging for classical machines, but realizing that promise requires far more qubits than a single processor can offer. Distributed quantum computing (DQC) scales out by connecting multiple quantum processing units (QPUs), at the cost of making entanglement the scarce resource: every remote gate consumes a Bell pair, and inter-QPU links generate Bell pairs at finite rates orders of magnitude slower than local gates. Since quantum programs are executed repeatedly, this rate bounds how fast shots complete and thus how fast results are obtained. Existing DQC compilers emit a single implementation per circuit, so throughput is capped by its busiest link while other links stay idle. We observe that alternative compilations of the same circuit are logically equivalent yet stress different links; executing them concurrently and pooling their samples converts idle Bell pairs into additional shots. We formulate the joint selection of compilations and allocation of shots under per-link Bell-pair capacities as Candidate-Constrained Max-Shot Allocation (CMA), prove it NP-hard, and solve it with a dynamic program and its approximate variant AppDP, a compact MILP, and a greedy heuristic Effi. In simulation on six-QPU networks, multi-compilation raises throughput by 2-4.5* over the single compilation and correspondingly improves output fidelity under equal Bell-pair budgets. On hardware, it increases measured fidelity by up to 93% and reaches the single-compilation fidelity in up to 10* less time. Scaling to 36 QPUs and 144-qubit circuits, MILP improves throughput over the single compilation by 76.8% on average across 48 configurations while Effi allocates in milliseconds.
AI Summary
Distributed quantum computing can be limited by the busiest entanglement-generation link used by a compiled circuit, even while other links remain underused. This paper asks whether executing alternative, logically equivalent compilations and pooling their measurement samples can increase shot throughput under per-link Bell-pair capacities. It formulates Candidate-Constrained Max-Shot Allocation: maximize integer shot rates subject to link capacities and a limit on active compilations. NP-completeness follows by reduction from set packing. Proposed solvers include an exact multidimensional dynamic program, a frontier-capped approximation (AppDP), a compact MILP, and a greedy allocator (Effi). The exact DP has a resource grid of size $\prod_e(R_e+1)$; AppDP retains at most 100 frontier states without an optimality guarantee. Simulations cover six 10-qubit workloads and five six-QPU topologies. For VQE, MILP increases throughput from 100 shots/s to 199–450 shots/s. Across 48 larger configurations, reaching 36 QPUs and 144-qubit circuits, its average improvement is 76.8%; Effi allocates in milliseconds. Hardware tests emulate nine QPUs on a monolithic IBM processor rather than using networked devices. Under modeled Bell-pair budgets, measured Hellinger fidelity improves by up to 93.3%, and a baseline fidelity target is reached with up to tenfold less modeled time. The model assumes independent link capacities and nonbinding local computation/control, and does not construct gate-level execution schedules. Candidate generation is outside the allocation problem and excluded from optimization runtimes. Additional shots need not improve accuracy when compilation-dependent noise dominates; some simulated workloads show fidelity saturation.
Error found: The claimed hardware time reduction compares modeled Bell-pair-generation budgets on a monolithic-device emulation, rather than measured end-to-end execution times on distributed quantum hardware. The reported throughput is aggregate resource-feasible throughput under explicitly idealized capacity assumptions, without a physical schedule demonstrating its concurrent achievability. The abstract's simulation summary omits reported 5.75× throughput gains and overgeneralizes fidelity improvement to workloads such as GHZ, whose fidelity is essentially unchanged.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.