Research into quantum approximate optimisation algorithm (QAOA) now determines how much noise each QAOA depth can *tolerate* before becoming unreliable than focusing on finding the best circuit depth for a given level of device noise. Simulations up to ten layers identify an “effective noise threshold” that defines this limit, generally decreasing with increasing circuit depth. This differs from previous work seeking ideal depths for specific devices.

Simulations reveal how much error each QAOA layer can withstand before performance drops off, showing a general trend where deeper circuits require lower levels of noise to function effectively. GPU-accelerated simulations map performance, how well QAOA solves optimisation problems, against increasing levels of error, creating what is called an “effective noise threshold”. This threshold resembles reaching a tipping point: add just one more disturbance and it falls over.

The team found that deeper QAOA ‘recipes’, designed to solve complex calculations, generally require lower levels of noise than shallower ones. However, determining exactly where this limit lies, and whether any useful depth exists at all for certain noise levels, remains key; predicting when added complexity will hinder the search for solutions is vital.

Noise Tolerance Defines Reliable QAOA Circuit Depth

Previously, determining an optimal QAOA circuit depth required assessing multiple configurations. Instead, researchers at Calvin Institute of Technology discovered how much noise each layer could tolerate before becoming unreliable, an improvement over methods focused on ideal depths given device limitations.

Simulations achieved a ten-fold increase in scale compared to prior work by employing GPU acceleration and modelling up to ten layers with three distinct noise types for MaxCut problems involving graphs containing eighteen vertices; this also defined an “effective noise threshold” as the inflection point on a fitted logistic surface representing approximation ratios, delineating reliable from unreliable regimes.

MaxCut problems involving graphs with eighteen vertices were modelled using up to ten layers incorporating depolarising, bit-flip, and amplitude damping quantum noise. Analysis of data from the ibm_marrakesh processor showed that while simulated gate errors underestimated actual device noise, the modelling still accurately predicted qualitative trends in depth dependence and identified dominant error channels. This work introduces a new approach to assess noisy circuits by focusing on tolerance rather than optimal depths for fixed devices.

Simulated Noise Thresholds Guide Realistic Quantum Algorithm Optimisation

Progress beyond simply optimising circuit depth for specific devices is achieved when establishing how much noise a quantum algorithm can withstand; however, it’s important to remember these simulations rely on modelled gate errors which underestimate actual hardware performance. A key consideration as developers strive towards practical applications of near-term quantum computation is this discrepancy introducing uncertainty when extrapolating findings directly onto real systems. Still, acknowledging this limitation remains vital for practical quantum computing development. The team’s work establishes a valuable method, an ‘effective noise threshold’, for gauging error tolerance at each circuit depth during optimisation tasks like MaxCut. Even imperfect modelling informs useful strategies and helps refine algorithms by balancing complexity with reliability as researchers explore more powerful systems. At Calvin Institute of Technology, they redefined how quantum circuit performance is assessed; focusing not on optimal depth but on noise tolerance allowed them to establish an “effective noise threshold”, the point beyond which QAOA becomes unreliable, using GPU acceleration to simulate MaxCut problems up to ten layers deep while accounting for three distinct types of errors, charting success rates against increasing disturbance through fitting approximation ratios to a generalised logistic surface.

The research demonstrated that it’s possible to define an ‘effective noise threshold’, a level of error at which Quantum Approximate Optimisation Algorithm (QAOA) calculations become unreliable, for each layer in a quantum circuit. This matters because understanding how much noise a circuit can tolerate is valuable as developers work with noisy intermediate-scale quantum processors and attempt to balance computational depth with accuracy when solving optimisation problems like MaxCut on graphs containing between 6 and 18 vertices.

Using simulations incorporating depolarising, bit-flip, and amplitude damping errors up to ten layers deep, the authors showed this threshold generally decreases as circuit depth increases. Their modelling approach also proved useful for identifying dominant sources of noise even while acknowledging that simulated gate errors underestimate actual hardware performance.

👉 More information
🗞 Effective Noise Thresholds for Reliable Quantum Approximation Optimisation Algorithm under Noisy Environment
✍️ Yozef Tjandra and Andriyan Bayu Suksmono (Institut Teknologi Bandung); Hendrik Santoso Sugiarto, Celine Vania Setiadi and Joanna Gracia Tan (Calvin Institute of Technology)
🧠 ArXiv: https://arxiv.org/abs/2610.36597

More like this