arXiv:2609.28623
Observation of Universal Quantum Chaos at Shallow Depths
Arman Sauliere, Guglielmo Lami, Hugo Lóio, Thomas J. Maldonado, Shashwat Kumar, Pedram Roushan, Andrea De Luca, Jacopo De Nardis
Abstract
Universal statistical laws are a hallmark of quantum chaos. A natural route to probing them in quantum circuits is offered by the statistics of output probabilities, which for sufficiently deep chaotic circuits approach the Porter--Thomas distribution. Here we show, theoretically and experimentally, that the approach to this limit is itself universal, and is organized into geometry-dependent universality classes. At shallow depths, growing only logarithmically with system size, we derive universal output-probability distributions that differ markedly from Porter--Thomas statistics, and we test them on a programmable superconducting processor. Two-dimensional lattices, open one-dimensional chains, and periodic rings exhibit distinct universal forms, retaining signatures of geometry and boundary topology within the same logarithmic-depth regime. Within each class, the distributions depend on only two parameters: an effective many-body Thouless scale and the accumulated number of incoherent errors. Extracting these directly from measured bitstring frequencies provides access to scrambling, entanglement growth of the target state, and global fidelity, without reconstructing or classically simulating the implemented circuit. We also identify a noise-driven transition between output statistics retaining system-wide quantum correlations and those reproducible by independent small patches. Our results establish universality beyond Porter--Thomas statistics as a quantitative tool for characterizing noisy quantum processors.
AI Summary
The paper asks whether output-probability statistics in local chaotic circuits have a universal approach to Porter–Thomas statistics, retaining geometry dependence at depths $t=O(\log N)$, and how noise modifies this crossover. A replica statistical-mechanics description yields distributions governed by two parameters: the relative many-body Thouless scale $\ell=L(t)/N$, with $L(t)=L_0e^{t/\tau}$, and accumulated incoherent noise $\eta\simeq\gamma Nt$. Dilute localized replica defects give a closed-form probability-of-probabilities distribution in 2D; permutation-domain-wall transfer matrices give distinct forms for open and periodic 1D chains. The timescale $\tau$ also controls ideal-state entanglement growth, while $e^{-\eta}$ acquires a fidelity interpretation in the fully scrambled, weak-noise limit. Tests on Willow use up to 20 qubits and $M=10\times2^N$ measurement shots per circuit. Two-parameter fits reproduce count distributions, with agreement improving with depth and size. Exact simulations independently check the inferred growth timescale. For a 20-qubit 2D circuit, the inferred incoherent error rate is $(1.16\pm0.01)\times10^{-3}$ per qubit per cycle, versus an XEB-derived $(2.5\pm0.2)\times10^{-3}$; the discrepancy is attributed to XEB’s additional sensitivity to coherent circuit mismatches. Theory and noisy simulations identify a crossover at $\gamma N=1/(2\tau)$ between patch-reproducible statistics and a surviving global contribution. This does not establish classical sampling hardness, and the global regime is a finite-depth window limited by fidelity decay. Experiments remain restricted to classically simulable sizes; benchmarking larger systems and resolving sufficiently small fidelities remain open challenges.
Central claim may not hold: The universal 2D integer moments are moment-indeterminate, so their calculation alone does not establish the claimed full output-probability distribution. The proposed universal circuit distributions rely on an uncontrolled random-tensor-network replacement, with the 1D distributions additionally reconstructed approximately from low-order moments. The identification of the fitted noise parameter with global fidelity is justified only in a fully scrambled, weak-noise regime and does not generally measure fidelity to the nominal circuit. The proposed noise threshold is supported as a candidate crossover in output-probability statistics, rather than established as a general boundary for system-wide quantum correlations or independent-patch reproducibility.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.