Quantum Benchmark Zoo

← All benchmarks

Mirror Benchmarking

System-level

Quantinuum's system-level benchmark that fits the exponential decay of mirrored random circuits' survival probability, with a decay rate that also gauges how coherent the noise is.

Measures
Exponential decay rate, with depth, of the survival probability of mirrored random circuits spanning the processor; the rate is quadratic in the error channel and equals the unitarity for certain noise
Introduced by
Mayer, Hall & colleagues (Honeywell Quantum Solutions, now Quantinuum) (2021)
Also known as
MB, Quantinuum mirror benchmarking, Honeywell mirror benchmarking

Mirror benchmarking (MB) is Honeywell Quantum Solutions’ (now Quantinuum) system-level benchmark, introduced by Mayer and colleagues in 2021 and built on self-inverting random circuits spanning the whole processor. Despite the near-identical name it is a distinct protocol from Sandia’s randomized mirror circuits (the paper credits the mirror-circuit construction to Proctor et al.) and from mirror RB. Its contribution is the theory: when the survival probability of such circuits must decay exponentially, and what the rate means.

How it works

Circuits are built from layers of random single-qubit Cliffords plus native two-qubit gates on random all-to-all qubit pairings (a natural fit for trapped ions), followed by the inverse of each layer in reverse order, with Pauli randomized compiling. The fraction of shots returning the ideal bitstring, the survival probability, is measured versus depth and fit to an exponential. The theory shows that if the twirling group forms a 2-design, then under a uniform-noise assumption the decay is exponential, with a rate quadratic in the error channel (equal to the unitarity for certain noise), so the same data also estimates how coherent the noise is.

Strengths and limitations

MB scales without classical simulation and extracts a noise-coherence estimate for free from the quadratic decay law. Its assumptions are real, though: uniform noise across layers, 2-design twirling, and a rate that equals the unitarity only for certain error types. The name causes chronic confusion: “mirror benchmarking” is often an umbrella term for the whole mirror-circuit family (NPL’s 2025 metrics review uses it that way); this entry is the Quantinuum protocol. The theory paper remains arXiv-only as of July 2026; the flagship deployments are the peer-reviewed record, and Quantinuum’s current documentation headlines Quantum Volume and GHZ state fidelity rather than MB.

Notable results

MB debuted on Honeywell’s H1 at 6–10 qubits (Mayer et al. 2021) and was one of four system-level tests in Quantinuum’s H2 launch benchmarking at 20–32 qubits (Phys. Rev. X 13, 041052 (2023)), with public data and analysis code. Quantinuum’s 98-qubit Helios random-circuit-sampling benchmark cites it for mirrored-circuit fidelity estimation.

Key papers

Reference implementations

  • Randomized Mirror Circuits : Sandia's scalable whole-processor test that runs self-inverting random circuits across the width × depth plane and maps where a device still returns the right bitstring.
  • Mirror randomized benchmarking : Scalable RB variant built from mirror circuits (motion-reversal sequences with random Pauli layers) that avoids compiling expensive inversion gates and extends layer-error estimates to many-qubit widths.
  • Quantum Volume : Single-number, full-system benchmark that scores a device by the largest random square circuit it can run while still generating heavy outputs reliably.