IBM and the University of Chicago say they have demonstrated quantum advantage while solving one of the field’s hardest problems: how do you trust an answer that classical computers cannot check?
Quantum advantage means performing a computation that leading classical methods cannot practically reproduce.
Previous demonstrations have faced a problem. Once a quantum calculation becomes too difficult for classical computers, independently verifying that the quantum machine produced the correct result also becomes difficult.
The researchers designed a new type of structured sampling circuit that remains extremely hard to simulate classically but carries information that can reveal errors during the calculation.
They encoded a computation involving 70 logical qubits using 97 physical qubits on an IBM quantum processor.
The circuit contained 2,415 logical two-qubit operations and 468 non-Clifford T gates, which increase the computational complexity dramatically.
By using spacetime error-detecting codes and rejecting runs that showed signs of errors, the researchers reduced effective gate error rates by about tenfold.
Crucially, they could also place a statistical lower bound on the fidelity of the final quantum state, giving them a way to assess whether the calculation remained trustworthy.
The quantum processor completed the experiment in roughly 15 minutes.
IBM says leading known classical simulation methods would require impractical computing resources to reproduce the hardest versions of the task.
But this does not mean quantum computers now outperform classical computers at everyday scientific or commercial problems.
The experiment is a specialized sampling benchmark, and the work is currently an arXiv preprint rather than a peer-reviewed paper. Future classical algorithms may also challenge the claimed advantage.
Still, combining beyond-classical complexity, logical encoding and built-in verification marks an important step toward quantum computers whose answers can be trusted even when classical machines can no longer check them directly.
📃 RESEARCH PAPER
📌 Martiel et al., “Sampling hard circuits with verifiably high fidelity,” arXiv preprint (2026)
0 Comments