Nine quantum processors, two very different pass marks

A cloud test of nine quantum processors produced two different answers to a basic question: did they pass? All nine met a permissive rule at a tested size, but only four met the stricter certification rule in the study’s main comparison.^[1]

The November 10 preprint examines an important capability: creating entanglement across a group of qubits, the information-carrying units of a quantum computer. Its results show why readers comparing machines need to look beyond both the qubit count and the word “success.” The experiments took place in July and September, not this week, and the paper has not been peer reviewed.^[1]

The researchers prepared GHZ states, named after physicists Greenberger, Horne and Zeilinger. These are a particular pattern of quantum correlations across the whole tested group. The benchmark estimates fidelity, a measure of how closely the prepared state matches the intended one. A fidelity above one-half is enough to establish that the group is genuinely entangled, although falling below that threshold does not prove entanglement is absent.^[1]

Measurements come with uncertainty. The team used a conservative allowance of 0.10 at 90% confidence. Its permissive rule accepts a result when the upper end of that allowance reaches the threshold. Certification requires the lower end to reach it instead.^[1]

An illustrative calculation makes the difference concrete. An estimated fidelity of 0.45 with a 0.10 allowance has an upper limit of 0.55 and a lower limit of 0.35. It clears the permissive rule but not the certification rule. This is an example, not a reported processor result: compatibility with success is weaker evidence than certification.

The hardware comparison has a similar trap. Under the permissive rule, IBM’s 120-qubit Phoenix processor reached a tested state size of 27, while AQT’s 12-qubit IBEX Q1 reached all 12 available qubits. AQT used a larger fraction of its machine, but IBM reached a larger absolute size. Neither percentage alone determines which machine suits an application. Under the stricter rule, the reported certified sizes were 16 for Phoenix and 12 for IBEX Q1.^[1]

Software also matters. The main comparison used provider cloud toolchains without error mitigation, the extra processing used to reduce the effects of errors. Those toolchains did not select physical qubits in identical ways. In a separate appendix test, manually choosing qubits enabled Rigetti’s processor to certify a two-qubit state, despite no tested size qualifying in its default-workflow run.^[1]

The search sampled selected sizes rather than every possible group, and limited repeated testing leaves uncertainty about day-to-day variability. The useful next step is repeated testing with a consistent certification rule and clearer separation of software choices from hardware capability. This is evidence about a specific cloud workflow, not a permanent league table of quantum computers.^[1]