Noise Flattens Quantum Landscapes, Compressing Span by Nearly 10 Percent

Dikran S Meliksetian at the University of New Haven, and colleagues investigated how hardware noise affects the performance of the Quantum Approximate Optimisation Algorithm (QAOA) for solving complex optimisation problems. Landscape Span Compression (LSC) is a new metric to quantify the distortion of the energy landscape caused by noise. The study, conducted on IBM’s ibm_fez quantum computer, shows that noise consistently compresses the landscape span by 24-30% but does not shift the optimal solution, offering insights into reliable parameter transfer. The research highlights a discrepancy between the IBM calibration model’s accuracy and the actual degradation observed, pinpointing crosstalk and coherent errors as key contributing factors to performance limitations in near-term quantum devices.

Hardware noise induces substantial field compression in quantum optimisation algorithms

Landscape Span Compression (LSC) revealed that hardware noise compresses the energy field of the Quantum Approximate Optimisation Algorithm (QAOA) by 24 to 30 per cent, a strong reduction from the ideal, uncompressed field. This compression consistently occurs without displacing the global minimum, a finding previously impossible to quantify with existing metrics like the approximation ratio. Consequently, classically optimised parameters can be directly applied to quantum hardware without requiring re-optimisation, simplifying the process of utilising near-term quantum devices for complex optimisation tasks.

Despite this compression, feasibility fractions at optimal parameters remained 1.5 to 1.7 times above random sampling, demonstrating a striking durability to noise-induced degradation. Measured across three constrained binary optimisation instances and a 156-qubit IBM quantum processor, hardware noise consistently compresses the energy field of the Quantum Approximate Optimisation Algorithm (QAOA) by 24 to 30 per cent. The location of optimal parameters remained unchanged during observation, meaning classically-derived solutions remain valid on the quantum hardware; an Optimal Parameter Shift of zero was recorded in all cases. Furthermore, analysis revealed the IBM calibration noise model captures only approximately 42 per cent of the degradation in solution quality, with crosstalk and coherent errors contributing sharply to the remaining performance loss; a consistent noise cost of 0.03 approximation-ratio units was observed across all tested scenarios.

Landscape Span Compression reveals limitations of current quantum error mitigation

Quantifying noise in quantum computers is important for realising the potential of these devices for complex calculations. Hardware imperfections consistently compress the ‘energy field’ explored by the Quantum Approximate Optimisation Algorithm (QAOA), a technique for finding approximate solutions to difficult problems. However, the analysis also revealed inconsistencies with Zero-Noise Extrapolation, a technique intended to counteract these errors, yielding unpredictable improvements and inflating uncertainty threefold.

Imperfections in quantum hardware distort calculations using the Quantum Approximate Optimisation Algorithm, and this work quantified those distortions, despite inconsistent results from Zero-Noise Extrapolation. Landscape Span Compression reliably measured noise levels and confirmed that existing calibration models only partially account for errors; crosstalk and coherent errors remain significant challenges. Now established as a reliable way to assess noise within the Quantum Approximate Optimisation Algorithm, Landscape Span Compression moves beyond simply detecting errors to understanding how they reshape problem-solving fields. By quantifying how noise consistently compresses these fields without shifting optimal solutions, the work validates transferring parameters from classical computers to quantum hardware, simplifying the utilisation of near-term devices for complex tasks. Analysis revealed existing calibration models only partially explain performance degradation, pinpointing crosstalk and coherent errors as key areas for improvement.

The research demonstrated that hardware noise compresses the energy landscape used in the Quantum Approximate Optimisation Algorithm by 24-30%, a phenomenon quantified by a new metric called Landscape Span Compression. This finding means that noise flattens the problem-solving field without necessarily changing the best solution, supporting the transfer of parameters from classical to quantum computers. Although current calibration models show structural agreement with hardware noise, they only explain approximately 42% of performance degradation, indicating that further improvements are needed to address crosstalk and coherent errors.