Research Cuts Quantum Simulation Costs by 61% with ArcTN

Optimising contraction orders for simulating quantum circuits has long been hampered by lengthy planning times or high computational costs; finding efficient solutions required considerable effort. Arclight Quantum Computing Inc. have developed ArcTN, a framework that achieves a sixty-one point thirty-two percent reduction in floating point operations (FLOPs) while simultaneously speeding up the planning process by one point six two times compared to existing tools like Cotengra Auto/CMA-ES on standard benchmarks. ArcTN is a new framework designed to optimise how calculations are performed during simulations of quantum circuits; it cuts down both computational demands and simulation time.

The team developed two approaches, ‘Light’ which prioritises quick initial setup, and ‘Heavy’, focusing on more thorough refinement for lower costs. Importantly, ArcTN also includes methods called “Fixed” and “Dynamic slicing”, reducing the size of intermediate data needed without substantially increasing overall calculation effort. Arclight Quantum Computing Inc. have unveiled ArcTN, a new framework designed to optimise calculations during simulations of quantum circuits; it aims to reduce both computational demands and simulation time.

The team developed two approaches, ‘Light’, which prioritises quick initial setup, and ‘Heavy’ focusing on more thorough refinement for lower costs. A key element is the concept of subtree reconfiguration: like rearranging puzzle pieces within a larger picture to find a better fit and simplify the overall structure, this process improves efficiency by optimising how parts of the calculation are organised.

Importantly, ArcTN also includes methods called “Fixed” and “Dynamic slicing”, reducing the size of intermediate data needed without substantially increasing effort; consider FLOPs, the number of basic math problems a computer needs to solve, lower FLOPs mean faster computation.

ArcTN accelerates quantum circuit simulations through optimised floating point operations

A sixty-one point thirty-two percent reduction in floating point operations (FLOPs) is delivered by ArcTN when contrasted with existing tools like Cotengra Auto/CMA-ES, overcoming a long-standing barrier where substantial gains demanded disproportionate planning time, previously limiting simulations of larger quantum circuits. The “Heavy” preset further improves performance, delivering speedups up to seven point two one times over configurations such as Cotengra AutoHQ/Optuna on completed QC32 networks and demonstrating strong efficiency improvements for complex calculations.

Nearly thirty-two percent fewer total sliced FLOPs are achieved through dynamic slicing relative to Fixed Slicing; this offers an effective method for managing memory constraints during tensor network simulation and enabling the exploration of more intricate systems.

ArcTN’s “Heavy” preset achieves speedups up to seven point two one times compared to configurations like Cotengra AutoHQ/Optuna on completed QC32 networks while also reducing FLOPs by eleven point eight five percent versus OMECO TreeSA-8, all with planning occurring one point three times faster. Though it comes at the expense of increased initial planning time, dynamic slicing enhances performance by cutting total sliced floating point operations by nearly thirty-two percent when contrasted with Fixed Slicing techniques; this technique successfully met size limits in all ninety runs performed on eighteen scalar-output networks.

Dynamic Slicing Balances Memory Reduction with Initial Processing Overhead

Arclight Quantum Computing has demonstrably improved tensor network simulation efficiency, an important step towards modelling increasingly complex quantum systems, but their work highlights an inherent trade-off between optimisation strategies concerning planning time versus computational cost reduction through ‘dynamic slicing’. While effectively managing memory constraints and lowering overall calculations, the technique demands more initial processing than its ‘Fixed counterpart. The team’s dynamic slicing reduces memory demands during complex calculations by breaking down large tasks; however, it requires increased initial processing time compared to the ‘Fixed alternative, a trade-off that doesn’t diminish ArcTN’s advance in tensor network simulation efficiency. Intelligent planning and execution strategies minimise calculations without sacrificing accuracy, sharply reducing the computational effort required to map complex circuits.

ArcTN is a new framework for optimizing quantum circuit simulations which reduced floating point operations (FLOPs) on completed QC32 networks by up to 61.32% while also speeding up the planning phase relative to existing methods like Cotengra Auto/CMA-ES and OMECO TreeSA-8. The authors demonstrated these results using datasets of up to 672 circuits from MQT Bench.

More like this

Quantum Research NewsResearchers Cut Simulation Errors Using Hamiltonian Norms

Quantum Research NewsResearchers Cut Quantum Computing Costs with New Scheduler

Quantum Research NewsPavia Team Finds Link Between Qubit Resources and State Overlap

Quantum Research NewsLoss Concentration Maximizes Response Strength in Third-Order Systems