Three quantum runs come with their own error accounting

IBM's advantage tracker adds three results built to prove they are correct without a classical computer to check them.

IBM added three entries to its quantum advantage tracker on Thursday, each attacking the same awkward problem. If a quantum computer produces a result no classical machine can reproduce, nothing classical can confirm the result is right — and today's processors are noisy enough that being wrong is a live possibility.

The three take different routes. IBM, RIKEN and the error-mitigation firm Qedma modelled an oscillating Ising system and ran two classical algorithms against it on the Fugaku supercomputer; the classical runs diverged from each other, while the quantum run produced a steady decline in magnetism with periodic oscillations, reproduced on a Quantinuum processor to rule out a hardware-specific artefact. IBM and the University of Chicago built a circuit mostly from easily simulated Clifford gates, sprinkled with T gates chosen because IBM's hardware implements them without adding noise and because they make classical sampling provably, exponentially hard on average; extra qubits around the edge flagged errors so bad runs could be discarded. Algorithmiq took the noise head-on — picking the quietest region of a chip, injecting known noise to measure its effect, then bounding how much could remain.

Verification is the gate on this whole field. Compute you cannot trust is not compute, and techniques for measuring and bounding error outlive the hardware they were built for.

The caveat is large and the source states it plainly: none of these results is useful. The systems modelled are simplified past real-world relevance, earlier advantage claims have been erased by better classical algorithms, and this entire class of noisy-hardware work is expected to fade once error-corrected qubits arrive in numbers.

Source: Ars Technica