Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2608.18592

Infrared Universality of Collective Dynamics across Transformer and State-Space Architectures

Byung Gyu Chae

Published Aug 19, 2026• 0 citations
arXivHTMLPDF

Abstract

Whether distinct neural architectures develop common collective dynamics remains an open question. Recent analysis of Transformer language models revealed a nearly flat, weakly infrared-enhanced time-scale density of states (TDOS) associated with near-marginal long-memory dynamics. Here we test whether a closely related organization emerges in Mamba, whose selective state-space dynamics provides a fundamentally different microscopic mechanism. Mamba allows relaxation dynamics to be resolved at three levels: the intrinsic spectrum of the learned state-space generator, its input-conditioned selective rescaling, and the collective TDOS of the complete block measured from its Jacobian. These spectra are not identical: selective dynamics and the remaining block transformations substantially reorganize the microscopic relaxation hierarchy. Nevertheless, the full block develops a reproducible slow-mode continuum whose infrared sector becomes progressively better resolved with increasing sequence length. Cumulative analysis yields $ρ(λ)\simλ^β$, with the long-sequence Mamba exponent stabilizing near $β_{\rm M}\simeq-0.17$. The corresponding memory dynamics follows $K(t)\sim t^{-(1+β)}$, close to the marginal $1/t$ regime. Despite fundamentally different microscopic dynamics, Transformer full-block spectra exhibit closely related infrared organization, with representative exponents of order $β_{\rm Tr}\sim-0.1$. These results separate explicit state-space memory from collective infrared organization and show that distinct sequence architectures can develop closely related near-marginal slow-mode dynamics. They extend infrared collective organization beyond Transformers and provide an independent test of the dynamical structure described by Cognitive Field Theory.

DOI
10.48550/arxiv.2608.18592

Related papers

Memory-Dominated Quantum Criticality as a Universal Route to High-Temperature Superconductivity

Byung Gyu Chae

Understanding the dynamical origin of high-temperature superconductivity remains a central challenge in strongly correlated quantum matter. Near quantum criticality, diverging correlation times reorganize the infrared dynamics into a scale-free continuum of collective relaxation processes. We show that the infrared behavior of interacting electrons is generically controlled by the relaxation-rate spectrum of the underlying many-body dynamics. Starting from a microscopic fermionic theory, we derive that the Cooper-channel kernel admits a universal spectral representation in terms of the time-scale density of states (TDOS) of collective decay modes, without invoking a specific bosonic mediator. The superconducting instability follows directly from the vanishing of the quadratic kernel via a standard ladder resummation and Thouless criterion, with the pairing interaction determined entirely by the infrared structure of the relaxation spectrum. A finite TDOS at vanishing relaxation rate produces a memory-dominated regime characterized by long-time kernels $K(t)\sim 1/t$ and logarithmic enhancement of the retarded pairing interaction, leading to a BCS-like exponential transition scale set by infrared spectral weight. More generally, infrared-singular spectra generate power-law response and algebraic enhancement of the transition scale. The same relaxation spectrum controls normal-state dynamics, giving rise to long-time correlations, non-Markovian response, and strange-metal behavior. These results identify the spectral organization of relaxation modes as a universal organizing principle of quantum critical matter and establish memory-dominated criticality as a natural mechanism for enhanced pairing.

arXiv:2602.22626•Feb 26, 2026•1 citation

Infrared Reorganization Patterns in Open-System Dynamics

Vite Cuauhtemoc T.

Local generators and time-local master equations provide a widely used framework for describing reduced dynamics in open systems. Infrared scaling analyses indicate that long-time behavior is controlled not only by coupling strength or memory time, but by the competition between analytic local contributions and nonanalytic spectral features of the environment.Here we investigate this mechanism across several physically motivated families of dynamical kernels, including sub-Ohmic environments, generalized Langevin dynamics, structured reservoirs, and critical spectra. For each class, we identify the analytic infrared order associated with local generators and the effective spectral exponent induced by the bath, and determine which contribution controls the limit omega goes to zero within a fixed reduced description.We show that a common infrared reorganization pattern emerges: whenever nonanalytic spectral scaling becomes infrared dominant, local generators cease to organize long-time dynamics, even though they may accurately describe intermediate-time evolution. These results establish the ubiquity of this reorganization mechanism and reveal a recurring physical pattern in the infrared structure of open-system dynamics across distinct environments.

•Feb 17, 2026•0 citations

Infrared Scaling Criterion for the Asymptotic Validity of Local Master Equations

Vite Cuauhtemoc T.

Local (Markovian) master equations provide a standard framework for describing the dynamics of open quantum systems, yet their asymptotic range of validity is typically assessed through weak-coupling or short-memory assumptions rather than explicit infrared scaling properties. Here we formulate a model-independent infrared scaling criterion that delineates the regime in which local master equations organize asymptotic dynamics. The analysis is based on the infrared structure of the retarded kernel, where the analytic local contribution is characterized by an exponent n_IR, while the spectrally induced self-energy scales as (−iω)^γ. This formulation isolates the infrared scaling competition independently of microscopic details of the bath realization. A direct comparison of these exponents yields a simple diagnostic condition: local descriptions control the infrared scaling only when the analytic contribution remains dominant, whereas for γ < n_IR the non-analytic spectral term becomes infrared dominant and determines the long-time behavior. The criterion is illustrated with representative examples, including regimes where local equations remain asymptotically valid and the sub-Ohmic case where infrared spectral dominance leads to non-exponential relaxation. These results provide a compact infrared scaling diagnostic for the asymptotic validity of local master equations without introducing additional dynamical assumptions.

•Feb 16, 2026•0 citations

Transformer Learning of Chaotic Collective Dynamics in Many-Body Systems

Ho Jang, Gia-Wei Chern

Learning reduced descriptions of chaotic many-body dynamics is fundamentally challenging: although microscopic equations are Markovian, collective observables exhibit strong memory and exponential sensitivity to initial conditions and prediction errors. We show that a self-attention-based transformer framework provides an effective approach for modeling such chaotic collective dynamics directly from time-series data. By selectively reweighting long-range temporal correlations, the transformer learns a non-Markovian reduced description that overcomes intrinsic limitations of conventional recurrent architectures. As a concrete demonstration, we study the one-dimensional semiclassical Holstein model, where interaction quenches induce strongly nonlinear and chaotic dynamics of the charge-density-wave order parameter. While pointwise predictions inevitably diverge at long times, the transformer faithfully reproduces the statistical "climate" of the chaos, including temporal correlations and characteristic decay scales. Our results establish self-attention as a powerful mechanism for learning effective reduced dynamics in chaotic many-body systems.

arXiv:2601.19080•Jan 27, 2026•1 citation

Hierarchical separation of relaxation timescales from spectral localization bounds

Alex Stewart-Bozzo, Jakub Garwoła, Brett Min +1 more

We investigate the dissipative dynamics of multilevel quantum systems strongly coupled either to a lossy cavity mode or directly to a bosonic environment. By deriving spectral localization bounds, we establish conditions under which strong system-bath coupling gives rise to a hierarchy of population relaxation timescales. Our approach builds on the reaction-coordinate polaron-transform framework. By mapping the original strong-coupling problem onto an effective weakly dissipative model, we analyze the spectrum of the resulting Liouvillian superoperator through localization bounds. For the generalized V model, we find that strong system-bath coupling gives rise to a bright-dark structure in the effective system-bath coupling operator: a single collective mode remains strongly coupled to the environment, while the remaining modes become progressively dark. Consequently, the dynamics separate into fast and slow sectors and, at finite coupling strengths, develop a hierarchy of population relaxation timescales. Numerical simulations based on both secular and non-secular quantum master equations corroborate the emergence of timescale separation and the pronounced slowing down of dissipative dynamics at strong coupling. Our results reveal a general mechanism underlying anomalously slow relaxation in strongly coupled open quantum systems and provide a route for engineering long-lived states through system-environment interactions.

quant-pharXiv:2606.20860•New Journal of Physics•Jun 22, 2026•0 citations

Emergent universality in Kraus maps of quantum chaotic many-body dynamics

Qi Camm Huang, Wai-Keong Mok, Tobias Haug +1 more

Recent studies of "deep thermalization" have revealed universal physics in quantum many-body dynamics beyond equilibration towards Gibbs states: maximally random quantum state ensembles can emerge on local subsystems, generated by measurements on their complement. In this work, we further identify a new form of universality exhibited in the "finer fingerprints" of quantum dynamics for local subsystems, induced by global unitary time-evolution. Specifically, we consider the projected Kraus ensemble, an ensemble of Kraus operators obtained by unraveling the quantum channel on a small subsystem with respect to knowledge of the classical configurations of its complement. Our central result is a one-parameter random matrix Ansatz that captures the ensemble's emergent statistical behavior along particular spacetime scalings, valid for generic 1D circuit dynamics without conservation laws: the ensemble is described by the product of a complex Ginibre random matrix and an independent log-normal random real scalar. The former encodes scrambling within the subsystem, captured by rotational invariance of the Ginibre measure, whereas the latter encodes fluctuations in the Born probabilities, arising from locality of the underlying dynamics. Our Ansatz can be established in the special cases of dynamics generated by global Haar random unitaries and dual-unitary circuits, while we motivate it for generic circuits using arguments of spacetime duality and the multiplicative ergodic theorem on long products of spatial transfer matrices. Extensive numerical simulations for random and Floquet circuit models verify these predictions. This universality in Kraus operators also provides a microscopic mechanism for deep thermalization in generic 1D quantum circuit dynamics, and has implications for local quantum information recoverability involving classical side information tied to knowledge of the bath.

quant-pharXiv:2609.12689•Sep 14, 2026•2 citations
Whether distinct neural architectures develop common collective dynamics remains an open question. Recent analysis of Transformer language models revealed a nearly flat, weakly infrared-enhanced time-scale density of states (TDOS) associated with near-marginal long-memory dynamics. Here we test whether a closely related organization emerges in Mamba, whose selective state-space dynamics provides a fundamentally different microscopic mechanism. Mamba allows relaxation dynamics to be resolved at three levels: the intrinsic spectrum of the learned state-space generator, its input-conditioned selective rescaling, and the collective TDOS of the complete block measured from its Jacobian. These spectra are not identical: selective dynamics and the remaining block transformations substantially reorganize the microscopic relaxation hierarchy. Nevertheless, the full block develops a reproducible slow-mode continuum whose infrared sector becomes progressively better resolved with increasing sequence length. Cumulative analysis yields ρ(λ)∼λβρ(λ)\simλ^βρ(λ)∼λβ, with the long-sequence Mamba exponent stabilizing near βM≃−0.17β_{\rm M}\simeq-0.17βM​≃−0.17. The corresponding memory dynamics follows K(t)∼t−(1+β)K(t)\sim t^{-(1+β)}K(t)∼t−(1+β), close to the marginal 1/t1/t1/t regime. Despite fundamentally different microscopic dynamics, Transformer full-block spectra exhibit closely related infrared organization, with representative exponents of order βTr∼−0.1β_{\rm Tr}\sim-0.1βTr​∼−0.1. These results separate explicit state-space memory from collective infrared organization and show that distinct sequence architectures can develop closely related near-marginal slow-mode dynamics. They extend infrared collective organization beyond Transformers and provide an independent test of the dynamical structure described by Cognitive Field Theory.
Understanding the dynamical origin of high-temperature superconductivity remains a central challenge in strongly correlated quantum matter. Near quantum criticality, diverging correlation times reorganize the infrared dynamics into a scale-free continuum of collective relaxation processes. We show that the infrared behavior of interacting electrons is generically controlled by the relaxation-rate spectrum of the underlying many-body dynamics. Starting from a microscopic fermionic theory, we derive that the Cooper-channel kernel admits a universal spectral representation in terms of the time-scale density of states (TDOS) of collective decay modes, without invoking a specific bosonic mediator. The superconducting instability follows directly from the vanishing of the quadratic kernel via a standard ladder resummation and Thouless criterion, with the pairing interaction determined entirely by the infrared structure of the relaxation spectrum. A finite TDOS at vanishing relaxation rate produces a memory-dominated regime characterized by long-time kernels K(t)∼1/tK(t)\sim 1/tK(t)∼1/t and logarithmic enhancement of the retarded pairing interaction, leading to a BCS-like exponential transition scale set by infrared spectral weight. More generally, infrared-singular spectra generate power-law response and algebraic enhancement of the transition scale. The same relaxation spectrum controls normal-state dynamics, giving rise to long-time correlations, non-Markovian response, and strange-metal behavior. These results identify the spectral organization of relaxation modes as a universal organizing principle of quantum critical matter and establish memory-dominated criticality as a natural mechanism for enhanced pairing.
Local generators and time-local master equations provide a widely used framework for describing reduced dynamics in open systems. Infrared scaling analyses indicate that long-time behavior is controlled not only by coupling strength or memory time, but by the competition between analytic local contributions and nonanalytic spectral features of the environment.Here we investigate this mechanism across several physically motivated families of dynamical kernels, including sub-Ohmic environments, generalized Langevin dynamics, structured reservoirs, and critical spectra. For each class, we identify the analytic infrared order associated with local generators and the effective spectral exponent induced by the bath, and determine which contribution controls the limit omega goes to zero within a fixed reduced description.We show that a common infrared reorganization pattern emerges: whenever nonanalytic spectral scaling becomes infrared dominant, local generators cease to organize long-time dynamics, even though they may accurately describe intermediate-time evolution. These results establish the ubiquity of this reorganization mechanism and reveal a recurring physical pattern in the infrared structure of open-system dynamics across distinct environments.
Local (Markovian) master equations provide a standard framework for describing the dynamics of open quantum systems, yet their asymptotic range of validity is typically assessed through weak-coupling or short-memory assumptions rather than explicit infrared scaling properties. Here we formulate a model-independent infrared scaling criterion that delineates the regime in which local master equations organize asymptotic dynamics. The analysis is based on the infrared structure of the retarded kernel, where the analytic local contribution is characterized by an exponent n_IR, while the spectrally induced self-energy scales as (−iω)^γ. This formulation isolates the infrared scaling competition independently of microscopic details of the bath realization. A direct comparison of these exponents yields a simple diagnostic condition: local descriptions control the infrared scaling only when the analytic contribution remains dominant, whereas for γ < n_IR the non-analytic spectral term becomes infrared dominant and determines the long-time behavior. The criterion is illustrated with representative examples, including regimes where local equations remain asymptotically valid and the sub-Ohmic case where infrared spectral dominance leads to non-exponential relaxation. These results provide a compact infrared scaling diagnostic for the asymptotic validity of local master equations without introducing additional dynamical assumptions.
Learning reduced descriptions of chaotic many-body dynamics is fundamentally challenging: although microscopic equations are Markovian, collective observables exhibit strong memory and exponential sensitivity to initial conditions and prediction errors. We show that a self-attention-based transformer framework provides an effective approach for modeling such chaotic collective dynamics directly from time-series data. By selectively reweighting long-range temporal correlations, the transformer learns a non-Markovian reduced description that overcomes intrinsic limitations of conventional recurrent architectures. As a concrete demonstration, we study the one-dimensional semiclassical Holstein model, where interaction quenches induce strongly nonlinear and chaotic dynamics of the charge-density-wave order parameter. While pointwise predictions inevitably diverge at long times, the transformer faithfully reproduces the statistical "climate" of the chaos, including temporal correlations and characteristic decay scales. Our results establish self-attention as a powerful mechanism for learning effective reduced dynamics in chaotic many-body systems.
We investigate the dissipative dynamics of multilevel quantum systems strongly coupled either to a lossy cavity mode or directly to a bosonic environment. By deriving spectral localization bounds, we establish conditions under which strong system-bath coupling gives rise to a hierarchy of population relaxation timescales. Our approach builds on the reaction-coordinate polaron-transform framework. By mapping the original strong-coupling problem onto an effective weakly dissipative model, we analyze the spectrum of the resulting Liouvillian superoperator through localization bounds. For the generalized V model, we find that strong system-bath coupling gives rise to a bright-dark structure in the effective system-bath coupling operator: a single collective mode remains strongly coupled to the environment, while the remaining modes become progressively dark. Consequently, the dynamics separate into fast and slow sectors and, at finite coupling strengths, develop a hierarchy of population relaxation timescales. Numerical simulations based on both secular and non-secular quantum master equations corroborate the emergence of timescale separation and the pronounced slowing down of dissipative dynamics at strong coupling. Our results reveal a general mechanism underlying anomalously slow relaxation in strongly coupled open quantum systems and provide a route for engineering long-lived states through system-environment interactions.
Recent studies of "deep thermalization" have revealed universal physics in quantum many-body dynamics beyond equilibration towards Gibbs states: maximally random quantum state ensembles can emerge on local subsystems, generated by measurements on their complement. In this work, we further identify a new form of universality exhibited in the "finer fingerprints" of quantum dynamics for local subsystems, induced by global unitary time-evolution. Specifically, we consider the projected Kraus ensemble, an ensemble of Kraus operators obtained by unraveling the quantum channel on a small subsystem with respect to knowledge of the classical configurations of its complement. Our central result is a one-parameter random matrix Ansatz that captures the ensemble's emergent statistical behavior along particular spacetime scalings, valid for generic 1D circuit dynamics without conservation laws: the ensemble is described by the product of a complex Ginibre random matrix and an independent log-normal random real scalar. The former encodes scrambling within the subsystem, captured by rotational invariance of the Ginibre measure, whereas the latter encodes fluctuations in the Born probabilities, arising from locality of the underlying dynamics. Our Ansatz can be established in the special cases of dynamics generated by global Haar random unitaries and dual-unitary circuits, while we motivate it for generic circuits using arguments of spacetime duality and the multiplicative ergodic theorem on long products of spatial transfer matrices. Extensive numerical simulations for random and Floquet circuit models verify these predictions. This universality in Kraus operators also provides a microscopic mechanism for deep thermalization in generic 1D quantum circuit dynamics, and has implications for local quantum information recoverability involving classical side information tied to knowledge of the bath.