Frequent Resets Pushed a Quantum Computer From Chaos to Order

SHUFFLE a deck often enough and nobody, you included, can say where the ace went. A new experiment on a quantum processor from International Business Machines (IBM) did not correct anything that had gone wrong; instead it used repeated measurements and resets across a chain of up to 100 qubits to steer a deliberately chaotic circuit into order, the sort of machinery error correction will need. Scrambling was the shuffle, a random gate mixing information between neighboring qubits. Resetting, which means measuring a qubit and flipping it back to its starting state if needed, pulled the other way.

Qubits, the quantum cousins of ordinary computer bits, can sit in a blend of possibilities until somebody measures them, which is the source of their power and of their fragility: heat, stray electromagnetic signals and tiny hardware flaws all nudge their states, and the errors pile up until a calculation can no longer be trusted. Nobody has yet built a fault-tolerant quantum computer, meaning one that keeps computing correctly while its qubits keep slipping.

Researchers at Rutgers University, IBM and several other institutions ran the test on a 156-qubit IBM Heron processor, picking a connected chain of up to 100 qubits. At every step a weighted coin toss decided what happened next: either a random gate scrambled a pair of neighbors, or a single qubit was measured and reset. The scrambling follows a quantum version of a textbook chaotic rule (the Bernoulli map, which keeps doubling a number and discarding the whole part, so that tiny differences snowball), while the resets drag everything toward one orderly state in which every qubit sits at zero. A tug of war, in other words, between a shuffler and a janitor.

A Sharp Line Between Chaos and Control

Who wins? It depends on the weighting, and sharply. If scrambling takes more than half the steps, chaos wins and stays; if resets take more than half, control wins and the qubits can be guided to the orderly state, and at about an even split a small change in the balance flips the whole system, a phase transition of the sort that turns water to ice.

To pin the tipping point down, the team ran 50 random circuits at each setting, put each through 1,000 runs, and compared the results with classical simulations of the full quantum state for up to 40 qubits. In the data the switch sat at a reset rate of 0.495; in the simulations it sat at 0.4947, and no software cleanup of the hardware’s noise (error mitigation) was applied to the experimental results. “What was striking was that the transition became more clearly defined as we studied larger systems,” says Haining Pan, a Rutgers postdoctoral researcher at the time of the study and a co-first author.

Measuring one qubit is the easy part. “The challenge was not simply to measure a qubit, but to do it repeatedly while the rest of the processor continued to operate,” says Maika Takita, a principal research scientist at IBM who works on quantum error correction and dynamic circuits experiments, and a senior author of the study.

What a shuffle-only classical system cannot produce is quantum fluctuations, the shot-to-shot scatter that comes from the randomness of quantum measurement, so the team treated that scatter as a fingerprint. If the processor were merely a very noisy classical machine, its scatter should look like that of a simple classical noise model that wipes out quantum coherence after every gate. It did not quite. At 20 qubits or more, in the chaotic regime and at the tipping point, the data sat slightly closer to the quantum simulation than to the classical stand-in. Slightly. The authors call the distinction subtle, and on the controlled side read-out errors swamped the signal so thoroughly that neither model captured the experiment there, so a sober reading stops well short of declaring the machine provably quantum.

Not Error Correction, Yet

The university’s press release frames the result as a step toward quantum computers that can correct their own mistakes, and a step it may well be. The paper itself makes no such claim: it treats the coordination of gates, measurements and feedback as a needed ingredient on the road to fault tolerance, and it demonstrates no error-correcting code.

The release also says the processor handled systems far beyond the roughly two dozen qubits that could be simulated on conventional computers. The paper’s own comparison is less lopsided: full quantum-state simulations reached 40 qubits, and a cheaper classical model reproduced the average behavior around the tipping point out to 100, with the real classical difficulty concentrated in the heavily scrambled regime, where even single circuit runs can stump a simulation.

Why does it matter, then? Because a fault-tolerant machine has to do this sort of thing, and do it relentlessly. “To build a fault-tolerant quantum computer, we have to make these checks and corrections not just once or twice, but a huge number of times during a calculation,” says Jedediah Pixley, a physicist at Rutgers University and a senior author. Here the checks and resets came in their thousands: nearly 5,000 entangling gates alongside 5,000 mid-circuit measurements and resets in the largest runs. By the team’s account that is the largest successful demonstration of the approach so far.

None of that corrects an error. The project began, as it happens, with a bet at a 2021 birthday party over whether a puzzling quantum effect had a counterpart in ordinary physics, and Pixley paid up in blueberry ice cream when he lost. The harder test, which nobody has yet passed, is a machine that does all this to fix errors rather than to tame chaos.

FAQ