The Signal
IBM and Rutgers Steer Quantum Chaos Across 100 Qubits
IBM and Rutgers researchers used mid-circuit measurements and conditional resets to control quantum-chaotic dynamics on a processor with up to 100 qubits. Their Nature Physics paper reports a transition near a control probability of 0.5, wh
Quantum.world Team · Written with AI · fact-checked ·
Nearly 5,000 entangling gates and 5,000 measurement-and-reset operations: that is the workload IBM and Rutgers researchers report running on as many as 100 qubits to steer a quantum system between chaotic and controlled behavior. The study, published October 9 in Nature Physics, uses measurements taken during a computation to decide when to reset qubits. It is a demonstration of control over quantum dynamics, not a demonstration of quantum error correction.
Resets compete with a map that scrambles information
The team implemented a quantum version of the Bernoulli map, a model of chaotic dynamics that scrambles information across a system. Local mid-circuit measurements provide the feedback: depending on the measurement outcome, a qubit can be reset as the circuit runs. The researchers varied the probability of applying this control, setting scrambling and reset operations against one another.
When resets are infrequent, scrambling dominates and the average magnetization approaches zero. More frequent control drives the system toward an all-zero state. The researchers measured the change between those regimes on IBM’s 156-qubit ibm_fez Heron r2 processor, selecting a line of up to 100 qubits for each experiment.
A transition near an even split
The reported transition sits close to a control probability of one-half: finite-size scaling puts the experimental critical point at 0.495, with an uncertainty of 0.001. The paper reports a correlation-length exponent of 1.05, with an uncertainty of 0.02. Those estimates agree with the authors’ matrix-product-state calculations for systems up to 40 qubits and a statistical-mechanics model that reaches 100.
“By successfully applying up to nearly 5,000 entangling gates and 5,000 non-unitary mid-circuit operations on systems up to 100 qubits, the quantum processor yields accurate estimates of the universal critical properties,” write Bibek Pokharel, Haining Pan et al., Nature Physics. The team says it obtained its steady-state experimental results without error mitigation. The result shows that a large superconducting processor can run adaptive circuits at this scale and recover the predicted transition’s critical behavior.
Noise limits what the 100-qubit result establishes
The distinction between a quantum and classical system requires more than observing the average transition: the authors’ classical statistical-mechanics model reproduces that average. To probe quantum behavior, they also examined fluctuations in magnetization and compared their distributions with matrix-product-state simulations and a model that fully dephases the qubits after gates. On the chaotic side and near the transition, the measured distributions fit the quantum simulation better than the dephasing model, though the authors describe the difference as subtle.
On the controlled side, device noise disrupts the comparison: neither model captures the measured fluctuations well, and readout errors suppress a feature expected in the ideal all-zero state. The experiment therefore provides evidence for quantum dynamics in the tested regime, but does not establish a quantum computational advantage or demonstrate error correction. The paper also says its experimental data and simulation files are available from the corresponding author on request, rather than in a public repository.
Control is not correction
Mid-circuit measurement and conditional reset are techniques that can also appear in fault-tolerant quantum computing, where repeated checks help detect and correct errors. Here, however, the researchers control a model’s dynamics; they do not encode logical qubits, measure error syndromes or run an error-correction cycle. The distinction matters because thousands of feedback operations show the hardware can execute a demanding adaptive experiment, not that it can preserve useful computation against errors.
The result leaves a concrete technical question: how much of the observed fluctuation signal can future experiments distinguish from device noise, especially in the controlled regime? A second question is whether the same feedback approach can be measured more decisively at 100 qubits, where the paper’s quantum-versus-classical distribution comparison remains limited by noise.