---
title: Researchers Certify a Quantum Advantage Result That No Classical Computer Could Verify
description: IBM and University of Chicago researchers built error detection into a quantum circuit, deriving a provable floor on the fidelity of a computation no classical machine can verify.
author: Darie Nani (Editor-in-Chief)
updated: 2026-07-30T13:22:36.316Z
canonical: https://www.sovereignmagazine.com/article/quantum-advantage-certified-logical-qubit-verification
image: https://cdn.nanimediahouse.com/ibm-quantum-system-two-verification-70867.webp
categories: Science & Tech
content_type: News
region: Global
publication: Sovereign Magazine
schema_type: Article
---

For years the hardest part of proving a quantum computer had beaten a classical one was not running the computation. It was checking the answer.

Researchers at IBM and the University of Chicago say they have closed that gap. In a preprint posted to arXiv on 28 July, a team of nine describes a way to build error detection into a quantum circuit so that the computation produces evidence of its own accuracy while it runs. The work was one of three quantum advantage demonstrations reported by IBM and its research partners on 30 July.

## The check broke down before the computers did

Quantum computing has always leaned on classical machines for validation. A classical computer can confirm a quantum result by working out the expected output, which is fine as long as the problem stays small enough to simulate. Once a quantum computer moves past that point, the check that certified it stops working.

The standard tool, cross-entropy benchmarking, requires computing ideal output probabilities, and that becomes prohibitively expensive on the largest circuits. So earlier experiments ran smaller or simplified versions, measured those, and extrapolated upward. IBM's own account calls this "a proxy of a proxy" rather than a direct certification of the hard computation, and notes that the extrapolation cannot account for how noise behaves or how errors spread in the harder regime, where both can be very different.

## The circuit carries its own certificate

The method the team describes, doped Clifford sampling, starts from a circuit a classical computer can still simulate. Researchers then insert non-Clifford T gates, the operations that make a circuit classically hard, while leaving the error-detecting structure intact.

That structure is a spacetime code, which distributes ancilla qubits across both the qubits and the circuit's evolution in time, so errors can be caught during the computation rather than inferred afterwards. Runs that fail the consistency checks are discarded. Combining the fidelity of the classically verifiable reference circuit with what the syndromes reveal about logical errors, the researchers derive a lower bound on the fidelity of the encoded computation.

"Rather than relying on an external proxy metric after the experiment, the computation carries enough information to certify its own quality," IBM's researchers write. The certificate is device dependent, but the paper argues it rests on substantially weaker assumptions about noise than the fidelity proxies used until now.

## Seventy logical qubits and a floor of 0.284

The demonstration ran a 70-qubit circuit at depth 70, doped with 468 T gates and encoded across a total of 97 physical qubits. Post-selecting on the syndrome checks cut effective gate error roughly tenfold. The certified figure that came out is a fidelity lower bound of 0.284 at 95 per cent confidence: not a measurement of how well the machine ran, but a floor the team can prove it cleared.

The company says the computation took about fifteen minutes, and that leading classical simulation methods faced prohibitive runtimes on the same task. It ran 2,415 logical two-qubit operations alongside the 468 T gates.

"Verification remains one of the biggest challenges in firmly establishing experimental quantum advantage," said Bill Fefferman, an associate professor at the University of Chicago and one of the paper's authors. "This experiment develops techniques to better characterize the fidelity of hard quantum states under noise, increasing confidence that the quantum computer is solving a computationally hard problem."

The other two papers took different routes to the same problem. Researchers at Qedma, working with RIKEN and BlueQubit, used circuits of up to 74 qubits to track a quantum system under repeated pulses of energy, and saw oscillations that two state-of-the-art classical methods running on one of the world's largest supercomputers did not produce. In the most demanding regime those classical methods disagreed with each other. A team at Algorithmiq ran 56-qubit experiments in a regime where at least three leading classical simulation groups produced inconsistent predictions.

## Now the classical-simulation groups get their turn

All three results have been posted to the Quantum Advantage Tracker, a tool a group of classical and quantum computing organisations set up last year to test advantage claims against the best available classical methods. Submissions currently on it include candidates from Q-CTRL, BlueQubit, and the Birla Institute of Technology and Science, Pilani.

That is the point of publishing a certificate rather than a headline number. As IBM puts it, announcing quantum advantage "does not close the case, it opens the results to a new level of scrutiny."

## FAQ

**Q: What is a logical qubit?**
A logical qubit is a single unit of quantum information spread across several physical qubits and protected by an error-correcting code, so that errors affecting the underlying hardware can be detected and corrected. In this experiment the logical computation reached effective error rates roughly ten times lower than the physical error rates beneath it.

**Q: How is a quantum computation verified?**
Traditionally by running the same problem on a classical computer and comparing, which only works while the problem remains small enough to simulate. Beyond that point researchers have relied on running smaller circuits and extrapolating. The approach described here instead builds error detection into the circuit, so measurements taken during the run yield a provable lower bound on the computation's fidelity.

**Q: Has this work been peer reviewed?**
Not yet. It was posted on 28 July 2026 as a preprint on arXiv, the repository where physicists routinely share results before journal review, and the circuits and results have also been released publicly on the Quantum Advantage Tracker.