Quantinuum and JPMorganChase ran a 12-logical-qubit portfolio problem with error correction, and it beat random guessing
Error-corrected quantum computing just ran a small finance problem from start to finish. An X post from Quantum CPT on Saturday flagged new work from Quantinuum and JPMorganChase running full algorithms on error-corrected qubits.
## What they did
According to Quantum.world, the team published a paper in Physical Review X on 19 November. They ran two algorithms, QAOA, used for optimisation, and HHL, used for solving systems of linear equations, on Quantinuum's H2-1, H2-2 and Helios trapped-ion processors.
The qubits were protected with the [[7,1,3]] Steane code, which stores one logical qubit in seven physical qubits. The largest run was a 12-variable portfolio optimisation problem on Helios. It used 12 logical qubits, 97 physical qubits and, in its deeper circuit, 2,132 physical two-qubit gates.
## The results
Both versions of the largest circuit did better than random guessing. The deeper one did worse than the shallower one, a sign that adding more steps still adds more errors at this scale.
The paper also reports that circuits with up to eight logical qubits and nine logical T gates performed about as well as the same circuits without error correction. For five and six logical qubits, performance improved as the team added more algorithm layers, despite the extra complexity. The team reports a logical T gate with an infidelity of 2.7(4) × 10⁻³.
## The caveats
The authors are careful about the claim. They built the algorithms from fault-tolerant components, but say they did not establish fault tolerance for the whole computation. The portfolio problem was highly simplified, and its circuits were produced in a way that does not scale. For HHL, they found that compiling programs on the fly was a performance bottleneck.
## Why it matters
Running a complete algorithm on encoded qubits, rather than testing single error-correction tricks, is a step that useful machines will need. This is not quantum advantage, and a 12-variable portfolio problem is easy for ordinary computers. What it shows is the plumbing for error-corrected algorithms working on current hardware, at a size that is growing.
