Can a Refined Complexity Measure Avoid False Alarms About Scrambling?

When rapid growth can mislead

Physicists use operator growth to study how quantum systems evolve. Krylov complexity is one measure of that growth. The authors of this preprint focus on a potential complication: an unstable saddle—a feature of the mathematical analysis—can contribute exponential growth at early times. Such a contribution may complicate the interpretation of a scrambling signal.

The authors propose logarithmic Krylov complexity, or logK, defined through a replica approach. They present it as a refinement of conventional Krylov complexity and test whether it can distinguish early-time scrambling from behavior dominated by an unstable saddle.

Numerical tests in quantum models

The abstract reports numerical evidence from two finite-dimensional quantum systems: the Lipkin–Meshkov–Glick model and the mixed-field Ising model at a chaotic point. It says logK distinguishes genuine scrambling from saddle-dominated scrambling at early times. In the genuine-scrambling case, logK avoids the exponential contribution associated with the unstable saddle. In the saddle-dominated case, it closely tracks conventional Krylov complexity.

That distinction is important to state precisely: the abstract assigns avoiding the saddle’s exponential contribution to the genuine-scrambling case, and close tracking of conventional Krylov complexity to the saddle-dominated case. The abstract does not give the criteria used to classify these behaviors or the detailed numerical results, so readers cannot independently check that comparison from the available evidence alone.

The preprint also examines integrable quantum systems with infinite-dimensional Krylov subspaces: the SYK₂ model and the quantum inverted harmonic oscillator. For these examples, the authors modify the operator used to describe Krylov spreading, generalizing an analytic-continuation procedure in the replica trick. They report that this modified approach lets logK capture integrable properties. The abstract does not specify those properties or explain how they are identified.

Extending the framework to classical dynamics

The authors extend the Krylov formalism to classical dynamical systems and define classical versions of the operator-growth measures. They report that false positives arising from unstable saddles in classical phase space are absent in their analysis. This is a result reported for their framework and analysis; the abstract does not establish that such false positives are impossible in all classical systems.

Together, the examples suggest that the measure’s interpretation depends on the setting. The authors report a distinction between genuine and saddle-dominated scrambling in the finite-dimensional quantum models, while their integrable examples call for a modification to the spreading operator. The abstract does not establish that a single unchanged version of logK is a universal diagnostic.

What the abstract cannot tell us

This article is based on the paper’s abstract alone. The abstract describes numerical evidence but does not provide equations, model parameters, system sizes, comparison criteria, error estimates, or detailed outputs. Without those details, it is not possible here to judge the calculations’ sensitivity or how broadly the conclusions apply. The classical extension is also described only briefly.

The source is an arXiv preprint, not evidence here of peer review or experimental confirmation. Its findings should be understood as claims reported by the authors. A fuller assessment would require examining the paper’s definitions, calculations, and the evidence supporting its distinctions.