Short answer: When a quantum computer's output cannot be checked by classical simulation, one way to build trust is to compare the same computation across different machines, or across related circuits on one machine. Three published protocols do this, and were tried on devices from superconducting, trapped-ion, photonic and nuclear magnetic resonance platforms. They are research benchmarks on small processors, and they say nothing about which platform is better.
Why compare across platforms
A 2022 Nature Communications paper puts the problem this way: as quantum computers approach the era of quantum advantage, validating their performance is difficult. Algorithmic success, as in factoring or oracle problems, gives only pass or fail information for one specific task on one machine. The authors argue that comparing different machines preparing nominally the same arbitrary circuit gives generic validation, since a quantum computation is only as valid as the agreement between results on different machines. They compare this to how disparate atomic clocks are used to evaluate metrological standards.
Randomized and correlated measurements, 2022
The 2022 paper reports a cross-platform comparison using randomized and correlated measurements. The authors execute several quantum circuits on widely different physical platforms and analyse the cross-platform state fidelities, which they say yields a wealth of information on the systems.
Cross-verification with measurement-based computing, 2021
A 2021 Physical Review X paper presents a technique that uses the principles of measurement-based quantum computation to link quantum circuits of different input size, depth and structure. It allows consistency checks between independent devices and within a single device. The authors apply the protocol to five quantum processors on four architectures: nuclear magnetic resonance, superconducting circuits, trapped ions and photonics, with up to six qubits and up to 200 distinct circuits.
Comparing processes with local operations and classical communication, 2023
A 2023 npj Quantum Information paper presents a protocol for comparing arbitrary quantum processes run on spatially or temporally separate platforms using local operations and classical communication. Local unitary operators are sampled and sent to each platform by classical communication to build state-preparation and measurement circuits. Each platform implements them and produces probability distributions of measurement outcomes, from which the maximum process fidelity is estimated to quantify relative performance. The authors say it can also be adapted for process tomography. They apply it to five IBM devices and the Qianshi quantum computer from Baidu, accessed via the cloud. They report that it compares processes on different computers well, and that it scales, though still exponentially in the number of qubits, much more favourably than full quantum process tomography.
What these results do and do not show
All three experiments use small devices, and the figures describe agreement between machines on chosen circuits, not overall quality. Agreement between machines is evidence, not proof, of correctness, since correlated errors could affect both.
FAQ
Why not just check against a classical simulation?
For large enough circuits, classical simulation is not possible, which is the case the papers are concerned with.
What does cross-platform comparison show?
Whether independent devices agree on the same computation, which the 2022 authors treat as a basis for validation.
Which hardware types were compared?
The 2021 paper covered nuclear magnetic resonance, superconducting, trapped-ion and photonic processors.
Does agreement prove the result is right?
No. It is consistency evidence.
Does this rank the platforms?
No. The papers do not claim a ranking.
All field notes