top of page

Research paper: Non-zero noise extrapolation

Accurately simulating noisy quantum circuits with tensor networks


We just released a new research paper describing a novel method for significantly improving the accuracy of emulations of noisy quantum computers.


Simulating noisy quantum computers is important for understanding the effects of noise on quantum computations. Modelling what happens when there is very little noise – i.e. when gate fidelities are very high – is crucial to the development of error correction and mitigation techniques, noise-resilient quantum algorithms, and more efficient decoders. Unfortunately, simulating such noisy computations turns out to be extremely challenging, especially when scaling up the number of qubits.


As quantum hardware advances, it is important that simulation tools remain capable of faithfully modelling computations with very low noise on larger numbers of qubits. With our quantum circuit emulator Ava providing the raw horsepower, we drew inspiration from quantum computing itself, namely the technique of zero-noise extrapolation, to develop a method that drastically improves the ability of Ava (and other tensor-network based emulators) to simulate noisy hardware in this highly relevant but challenging regime.


Key takeaways:

  • Simulating noisy quantum computations is difficult:

    • The most common approach – storing the full density matrix, cannot be utilised for more than around 17 qubits.

    • Other approaches that can scale to more qubits require unrealistic restrictions of the type of circuit and noise that can be simulated. 

  • Tensor networks can be used to simulate general types of circuits and noise on large numbers of qubits, but often struggle to obtain high accuracy in the regime of low-noise. 

  • Given the importance of the low-noise regime for quantum error correction and related topics, all of this means that there is currently a gap in the abilities of quantum emulation tools.

  • We developed a technique called non-zero noise extrapolation: a simple but powerful strategy that improves the accuracy of tensor-network based emulators in the low-noise regime.  


Non-zero noise extrapolation makes it possible to faithfully simulate noisy circuits with 60 qubits and beyond, without artificial restrictions on the type of circuit or noise model!

bottom of page