IBM added three entries to its IBM quantum advantage tracker on Thursday, saying each shows a quantum computer doing work that classical machines cannot practically match while also giving researchers ways to trust the answer. That second part is the rub. A result that cannot be checked by an ordinary computer is also a result that can be wrong in ways nobody sees.
Quantum computing uses qubits and quantum operations to exploit effects such as interference, which can make some calculations cheaper than they are on classical hardware. Today’s machines, however, are noisy and limited, so the field has spent years arguing over whether claimed advantages survive better classical algorithms and basic error checks.
IBM’s Jay Gambetta told Ars Technica that trusted computing matters most when classical simulation no longer works. The three new efforts, from IBM and outside collaborators, are not presented as useful applications. They are closer to stress tests for noisy quantum processors and for the methods used to decide whether those processors are telling the truth.
What did IBM claim about quantum advantage?
IBM said the three new manuscripts show quantum advantage using different strategies for handling errors and validating output. The company pointed readers to arXiv papers through an IBM blog post and listed the work on its public advantage tracker.
One project involved IBM, Japan’s RIKEN, and Qedma, a quantum software company focused on error mitigation. The team modeled a Floquet process in an Ising model, a simplified grid of magnet-like elements whose orientations influence their neighbors. As the model grows, tracking the intermediate states during periodic flips becomes harder for classical machines.
Qedma and RIKEN compared two classical algorithms on the Fugaku supercomputer. Those methods drifted apart over time, producing conflicting behavior for the model’s magnetism. On an IBM quantum processor using Qedma’s error-mitigation software, the result showed magnetism falling gradually with oscillations. The team then checked the output on a Quantinuum processor and said it matched. They also identified a weakness in at least one classical approach: it drops terms that are needed to capture the oscillations.
A second manuscript came from IBM and University of Chicago researchers. Their circuit mostly used Clifford gates, which classical computers can simulate relatively easily, but added selected T gates. The paper said those Z rotations, including T gates, are implemented on IBM hardware through virtual frame tracking and do not add extra noise. Gambetta told Ars Technica that the T gates also strengthen the complexity argument, because average sampling becomes exponentially hard for classical computers.
That experiment also used extra qubits around the main circuit. Gentle measurements on those qubits were used to detect errors during computation, with flagged runs discarded. The tradeoff is blunt: the method can throw away bad runs, but it can also discard valid runs if the error-detection readout itself misfires.
The third result came from Algorithmiq, using an approach similar to Google’s earlier “quantum echoes” work. The algorithm drives a quantum system forward, reverses it, and inserts operations during the reversal so the system produces an imperfect echo rather than returning to its starting state. Classical simulations require simplifications, and different simplifications give different answers.
Algorithmiq’s team selected a low-noise region of a quantum processor, used neighboring qubits to detect unwanted changes, optimized control signals, and applied other noise-suppression methods. The researchers then injected noise deliberately to measure its effect, estimated the remaining noise, and repeated the procedure on a separate IBM processor with a different noise pattern. That let them put a bound on the error rate rather than wave vaguely at “noise” and call it a day.
Are these quantum results useful yet?
No immediate application is claimed. The modeled systems are simplified, and previous quantum-advantage announcements have lost some force after classical algorithm developers improved their methods. These claims could face the same pressure.
The work is still useful as an engineering checkpoint. It tests how researchers suppress noise, certify fidelity, compare against classical methods, and avoid trusting a quantum processor just because it produced a hard-to-check pattern. Gambetta told Ars Technica that the field’s “holy grail” is comparing a real material or real experiment directly with a quantum computer. These results do not do that, but IBM argues they move the trust problem closer to the center of the work.
This story draws on original reporting from Ars Technica.
