Grobe–Haake–Sommers in Open Quantum Chaos

• The Grobe–Haake–Sommers conjecture is an extension of BGS ideas to non-Hermitian quantum dynamics, relating spectral statistics to dissipative chaos.
• It distinguishes two regimes where regular dynamics show 2D Poisson statistics with linear repulsion, and chaotic dynamics exhibit Ginibre statistics with cubic repulsion.
• Recent studies reveal limitations through counterexamples and non-normality effects, prompting the use of dynamical and eigenvector analyses to refine chaos diagnostics.

The Grobe–Haake–Sommers conjecture is a proposal in dissipative quantum chaos that extends the Bohigas–Giannoni–Schmit logic from Hermitian Hamiltonians to non-Hermitian generators of open dynamics. In its original form, it asserts that the complex spectrum of a Liouvillian, or of an equivalent non-Hermitian evolution operator, distinguishes regular from chaotic dissipative dynamics: simple or regular classical attractors are associated with a two-dimensional Poisson process and linear nearest-neighbor repulsion in the complex plane, whereas chaotic attractors are associated with Ginibre statistics and cubic level repulsion. Subsequent work has both broadened this picture—by embedding it in Coulomb-gas and random-matrix universality arguments—and sharply limited it, by exhibiting models in which Ginibre-type spectral correlations arise without classical chaotic attractors and even without dynamical chaos on any relevant timescale (Akemann et al., 2019, Villaseñor et al., 2024, Naves et al., 31 Mar 2026).

1. Historical formulation

The conjecture originates in the work of Grobe, Haake, and Sommers on periodically kicked tops with damping. Their setting was explicitly dissipative and non-Hermitian: instead of real energy levels of a Hamiltonian, the relevant spectral object was a complex spectrum generated by a discrete quantum map or, more generally in later formulations, by a Liouvillian. In the historical formulation recalled in later analyses, the integrable limit yields nearest-neighbor spacings in radial distance $s$ that follow a 2D Poisson distribution, while the chaotic limit is described by the complex Ginibre ensemble, usually denoted GinUE (Akemann et al., 2019).

The conjecture has two closely related components. First, it posits a regular/chaotic dichotomy for dissipative systems that parallels Berry–Tabor and BGS for closed systems: regular classical dissipative dynamics correspond to uncorrelated complex eigenvalues, while chaotic classical dissipative dynamics correspond to random-matrix correlations in the complex plane. Second, in its more local form, it asserts a universal small-spacing law in the chaotic regime. The follow-up work summarized in later literature states that Grobe, Haake, and Sommers used symmetry-based perturbative arguments to show that, for small spacing $s$, the repulsion in generic chaotic Markovian open quantum systems is universally cubic, $P(s)\sim C s^3$ as $s\to 0$, with $C>0$ (Akemann et al., 2019).

Later authors explicitly framed this as the dissipative analogue of the BGS conjecture. In that reading, simple attractors in classical dissipative phase space correspond to 2D Poisson statistics with linear repulsion, while chaotic attractors correspond to Ginibre statistics with cubic repulsion in the complex plane (Villaseñor et al., 2024). That formulation became the standard meaning of the Grobe–Haake–Sommers conjecture in the open-systems literature.

2. Spectral framework and canonical statistics

The conjecture is formulated for generators of open quantum dynamics, typically Liouvillians of Lindblad form,

$\frac{d\rho}{dt}=\mathcal{L}[\rho].$

Because $\mathcal{L}$ is non-Hermitian, its eigenvalues $\{\mu_n\}$ or $\{z_n\}$ are complex. The relevant local statistic is the nearest-neighbor spacing in the complex plane, defined by Euclidean distance after unfolding to approximately uniform local density. In the notation used repeatedly in the literature, one considers the radial spacing $s\ge 0$, normalizes the first moment to unity after unfolding, and studies the probability density $s$0 or $s$1 for nearest-neighbor distances (Akemann et al., 2019, Villaseñor et al., 2023).

For the regular branch of the conjecture, the reference law is the 2D Poisson spacing distribution

$s$2

Its small-$s$3 behavior is linear,

$s$4

which later papers summarize as linear level repulsion in the complex plane (Akemann et al., 2019).

For the chaotic branch, the reference law is the Ginibre spacing distribution in the bulk of the complex Ginibre ensemble. The later literature emphasizes less the full closed expression than its defining local property: $s$5 for some constant $s$6. This cubic law is the canonical GHS signature of chaotic dissipative spectra (Akemann et al., 2019).

A second diagnostic, designed to avoid explicit unfolding, is the complex spacing ratio

$s$7

where $s$8 and $s$9 are the nearest and next-nearest neighbors in the complex plane. For a 2D Poisson process one has $P(s)\sim C s^3$0 and $P(s)\sim C s^3$1, while for GinUE one has $P(s)\sim C s^3$2 and $P(s)\sim C s^3$3 (Villaseñor et al., 2023).

A notational subtlety enters in work that models the integrable-to-chaotic crossover by a 2D Coulomb gas. There, the symbol $P(s)\sim C s^3$4 denotes inverse temperature of the log-gas, not the small-$P(s)\sim C s^3$5 power-law exponent. In that parametrization, $P(s)\sim C s^3$6 yields the 2D Poisson limit and $P(s)\sim C s^3$7 yields the GinUE limit (Akemann et al., 2019).

3. Random-matrix interpolation and universality claims

A major extension of the original conjecture was provided by the study of boundary-driven open quantum spin chains in which the complex eigenvalues of the Liouvillian were modeled by a static two-dimensional Coulomb gas with harmonic confinement,

$P(s)\sim C s^3$8

Within this framework, $P(s)\sim C s^3$9 reproduces the non-interacting 2D Poisson limit, $s\to 0$0 reproduces GinUE, and intermediate $s\to 0$1 provide a one-parameter interpolation between integrability and chaos (Akemann et al., 2019).

This construction sharpened the original GHS picture in two ways. First, it promoted the regular/chaotic dichotomy into a crossover family: $s\to 0$2 functions as a continuous “chaoticity” parameter, with intermediate dissipative regimes well described by Coulomb-gas spacing distributions $s\to 0$3. Second, it generalized the universality claim from the small-$s\to 0$4 cubic law to the full spacing distribution in the fully chaotic limit. The same work states that the full bulk spacing distribution at $s\to 0$5 is universal across GinOE, GinUE, and GinSE, despite the fact that the real and quaternion Ginibre ensembles are not simple 2D log gases with Dyson indices $s\to 0$6 (Akemann et al., 2019).

The underlying argument is local. Away from the real axis and spectral edge, all three Ginibre ensembles reduce to the same bulk repulsion structure,

$s\to 0$7

so their nearest-neighbor spacing distributions in the bulk coincide. The same study further notes that Tao–Vu-type universality results for non-Gaussian perturbations support extension of GinUE bulk statistics beyond the Gaussian case. In this sense, the GHS conjecture was recast as a statement about a broad universality class of non-Hermitian random matrices and, by extension, of chaotic dissipative quantum systems (Akemann et al., 2019).

4. Model studies supporting the conjecture

Support for the conjecture has come from several explicit Liouvillian models. In the boundary-driven Heisenberg XXZ spin-$s\to 0$8 chain with nearest- and next-to-nearest-neighbor interactions, bulk dephasing, and boundary driving, the relevant spectrum is the complex spectrum of the Liouville super-operator acting on traceless density operators. In that model, local bulk statistics interpolate from 2D Poisson in integrable regimes to GinUE in chaotic regimes, with intermediate cases well fit by Coulomb-gas distributions $s\to 0$9. The reported best-fit values include an intermediate case near $C>0$0 and a nearly fully chaotic case near $C>0$1, while chaotic cases match GinUE with Kolmogorov distances of order $C>0$2 (Akemann et al., 2019).

A distinct line of support came from the open Dicke model with cavity losses. Because that Liouvillian acts on an infinite-dimensional Liouville space, the spectral analysis required a convergence criterion based on eigenstates rather than on direct matching of eigenvalues across truncations. The criterion introduces boundary weights

$C>0$3

with $C>0$4, to select eigenstates localized away from the Fock-space truncation edge. Spectral statistics were then computed in windows ordered by $C>0$5, using both unfolded nearest-neighbor spacings and complex spacing ratios (Villaseñor et al., 2023).

In that analysis, the coupling $C>0$6 corresponds to a classically chaotic regime of the isolated Dicke Hamiltonian, and the converged Liouvillian eigenvalues in the parity $C>0$7 sector exhibit GinUE statistics across essentially the whole spectrum examined. By contrast, at $C>0$8, where the isolated classical model is regular, low-$C>0$9 windows are consistent with 2D Poisson statistics, while higher-$\frac{d\rho}{dt}=\mathcal{L}[\rho].$0 windows cross over toward GinUE. Anderson–Darling tests and complex ratio diagnostics support both identifications. The authors therefore interpreted the open Dicke model as a verification of the GHS conjecture in an infinite-dimensional Markovian open quantum system, provided the spectrum is analyzed region by region rather than as a single undifferentiated set (Villaseñor et al., 2023).

Taken together, these model studies established a substantial empirical basis for GHS-type behavior. They also introduced an important qualification: the relevant statistics may depend on spectral location, especially in systems whose spectra mix regular and chaotic sectors or whose local density varies strongly with $\frac{d\rho}{dt}=\mathcal{L}[\rho].$1 (Akemann et al., 2019, Villaseñor et al., 2023).

5. Breakdown and reinterpretation

The strongest challenge to the conjecture came from later studies of the open Dicke model. One analysis argued that the GHS conjecture does not hold in that system because Ginibre spectral correlations appear in broad parameter regions where the classical dissipative dynamics is not chaotic. In the isotropic open Dicke model, the classical dynamics exhibits only simple attractors and no chaotic attractors, yet for large couplings the Liouvillian spectrum displays GinUE statistics over the entire spectrum examined. In the anisotropic model, GinUE correlations occur both inside the classical chaotic region and outside it, including along the isotropic line $\frac{d\rho}{dt}=\mathcal{L}[\rho].$2, where the classical dynamics remains non-chaotic. Only along the open Tavis–Cummings limit $\frac{d\rho}{dt}=\mathcal{L}[\rho].$3 does the Liouvillian spectrum remain consistent with 2D Poisson statistics (Villaseñor et al., 2024).

This criticism was subsequently refined rather than simply rejected. A later work on the open anisotropic Dicke model distinguished three regimes: steady-state chaos, transient chaos, and regular dynamics. In that formulation, Ginibre spectral statistics appear in both steady-state chaos and transient chaos, while 2D Poisson appears only in the regular regime. The decisive observables are dynamical rather than purely spectral: bipartite entanglement entropy and OTOC-based fluctuations show rapid early-time growth in both chaotic regimes, but only steady-state chaos produces large long-time saturation values. On that basis, Ginibre statistics were reinterpreted as a signature of short-time chaotic dynamics or scrambling, not of the existence of a chaotic attractor in the steady state (Mondal et al., 5 Jun 2025).

An even more structural critique followed from the analysis of non-normality in open quantum systems. That work argues that the analogy with Hamiltonian level statistics is flawed because Lindbladians are typically non-normal, their eigenvalues have a different dynamical role from Hamiltonian energy levels, and late-time behavior depends only on eigenvalues with small real parts and on the corresponding eigenvectors. In explicit examples—a driven harmonic oscillator with thermal bath and an open tight-binding chain with incoherent hopping—boundary conditions or Hilbert-space truncations generate Ginibre-like bulk level repulsion even though the dynamics remain integrable or non-chaotic on all physically relevant timescales. The mechanism is tied to large eigenvalue condition numbers, pseudospectral sensitivity, and a non-Hermitian skin effect in Liouville space (Naves et al., 31 Mar 2026).

These developments change the status of the conjecture. They do not deny that chaotic dissipative dynamics can coexist with Ginibre spectra. Rather, they deny that bulk Ginibre statistics provide a universal and reliable diagnostic of dissipative chaos. In particular, the implication

$\frac{d\rho}{dt}=\mathcal{L}[\rho].$4

is explicitly shown to fail in the open Dicke model and in strongly non-normal Lindbladians more generally (Villaseñor et al., 2024, Naves et al., 31 Mar 2026).

6. Present status and open problems

The current literature presents the Grobe–Haake–Sommers conjecture less as a settled theorem than as a historically important organizing principle whose domain of validity is model-dependent. On the positive side, there is strong numerical evidence that many dissipative spectra exhibit the original 2D Poisson/GinUE dichotomy, at least in suitable bulk windows and in settings where non-normal effects do not overwhelm the spectral interpretation. The boundary-driven XXZ chain and the windowed spectral analysis of the open Dicke model remain central examples of that behavior (Akemann et al., 2019, Villaseñor et al., 2023).

On the negative side, the literature now contains explicit counterexamples to any universal reading of GHS. The open Dicke results show that Ginibre level repulsion can coexist with regular long-time classical dissipative dynamics, and the non-normality analysis shows that level statistics can be tuned almost arbitrarily without affecting short- or long-time dynamics in the relevant sectors. These findings imply that bulk level statistics cannot, in general, serve as a reliable standalone diagnostic of either transient or persistent quantum chaos in open systems (Villaseñor et al., 2024, Naves et al., 31 Mar 2026).

A revised research program has therefore emerged. One direction is dynamical: entanglement entropy, dissipative OTOCs or FOTOCs, and time-resolved fluctuations of observables such as $\frac{d\rho}{dt}=\mathcal{L}[\rho].$5 or photon number are proposed as diagnostics that distinguish transient from steady-state chaos and restore a closer quantum–classical correspondence. Another direction is spectral but more selective: attention shifts from bulk level spacings to the Liouvillian gap, near-gap modes, eigenvector non-orthogonality, Petermann factors, and pseudospectral structure (Mondal et al., 5 Jun 2025, Naves et al., 31 Mar 2026).

What remains open is the extent to which a restricted version of GHS survives. The spin-chain and random-matrix results suggest that a universality class based on 2D Poisson, Coulomb-gas interpolation, and Ginibre bulk statistics is mathematically natural for many non-Hermitian spectra (Akemann et al., 2019). The Dicke-model and non-normality critiques suggest that this universality need not coincide with classical dissipative chaos in any one-to-one sense (Villaseñor et al., 2024, Naves et al., 31 Mar 2026). A plausible implication is that the conjecture survives, if at all, as a statement about particular spectral sectors or about short-time scrambling, rather than as a universal equivalence between chaotic attractors and bulk Liouvillian Ginibre statistics.