Benchmarking ORCA PT-1 Boson Sampler in Simulation

TL;DR

The paper benchmarks the ORCA PT-1 time-bin interferometer boson sampler, using a noise-free simulation of a dominating set (surveillance coverage) problem to compare against classical algorithms. It employs a variational quantum algorithm with SPSA optimization to learn problem-specific angles on graphs generated via NetworkX, reporting results for moderate sizes ($n<250$). In simulations, the TBI solver can find near-minimal dominating sets with competitive performance to some classical methods, though wall-clock scaling is poorer than fast classical solvers; preliminary PT-2 two-loop results hint at potential gains that would require hardware to verify. The work illustrates a possible path for practical graph-optimization applications using boson-sampling hardware, while acknowledging current limitations and outlining future hardware-focused investigations to assess real-world advantages.

Abstract

Boson Sampling, a non-universal computing paradigm, has resulted in impressive claims of quantum supremacy. ORCA Computing have developed a time-bin interferometer (TBI) that claims to use the principles of boson sampling to solve a number of computational problems including optimisation and generative adversarial networks. We solve a dominating set problem with a surveillance use case on the ORCA TBI simulator to benchmark the use of these devices against classical algorithms. Simulation has been used to consider the optimal performance of the computing paradigm without having to factor in noise, errors and scaling limitations. We show that the ORCA TBI is capable of solving moderately sized (n<250) dominating set problems with comparable success to linear programming and greedy methods. Wall clock timing shows that the simulator has worse scaling than the classical methods, but this is unlikely to carry over to the physical device where the outputs are measured rather than calculated.

Figures (8)

• Figure 1: A simple boson sampling set up with 4 input photons, 5 output modes and 4 variable beam splitters each with angle $\theta_i$.
• Figure 2: Diagram showing time-bin interferometry. Taken from Bradler2021-xa
• Figure 3: Summary of test runs on the TBI Simulator. Markers are sized and labelled with the size of the dominating set found.
• Figure 4: Problem size fixed ($n=30$) with varying graph density ($p$).
• Figure 5: Fixed $p=0.5$ for a range of $n$ values.