Thermal OTOCs: Quantum Chaos & Scrambling

Updated 7 February 2026

Thermal OTOCs are four-point, non-time-ordered correlators that diagnose operator growth, information scrambling, and quantum chaos in systems at finite temperature.
They employ analytic continuation, KMS conditions, and spectral analysis to reveal key features like exponential Lyapunov growth and non-commutativity of operators.
Experimental and computational methods including thermofield double protocols and tensor-network techniques enable robust measurement and analysis of these correlators.

Thermal out-of-time-order correlators (OTOCs) are four-point, non-time-ordered correlation functions of the form

$F_\beta(t) = \frac{1}{Z} \mathrm{Tr}\big[e^{-\beta H} W(t) V(0) W(t) V(0)\big], \qquad Z=\mathrm{Tr}\,e^{-\beta H},$

where $W(t) = e^{iHt} W e^{-iHt}$ and the averaging is taken over a thermal (Gibbs) state at inverse temperature $\beta$ . OTOCs provide a diagnostic of operator growth, information scrambling, and the emergence of quantum chaos in both few-body and many-body quantum systems. The "thermal" label specifically refers to the inclusion of thermal weighting in the expectation, in contrast to the pure-state (zero temperature) or infinite-temperature ( $\beta\to0$ ) limits.

1. Formal Definition and Structural Properties

The canonical thermal OTOC is a four-point, out-of-time-ordered expectation value in a thermal ensemble. For two (not necessarily commuting) Hermitian operators $W, V$ , the most common variants are:

The unsymmetrized correlator:

$F_\beta(t) = \frac{1}{Z} \mathrm{Tr}[e^{-\beta H} W(t) V(0) W(t) V(0)]$

The commutator-squared form:

$C_\mathrm{comm}(t) = \langle [W(t), V(0)]^\dagger [W(t), V(0)] \rangle_\beta$

Symmetric and "bipartite" regularized versions, e.g.:

$F_{\mathrm{sym}}(t) = \mathrm{Tr}\,[\rho_\beta^{1/2} W(t) \rho_\beta^{1/2} V(0) W(t) V(0)],$

and higher generalized regularizations (Tsuji et al., 2016).

The Heisenberg time evolution $W(t)$ and the precise arrangement of operators encode sensitivity to operator growth and their non-commutativity at later times.

Thermal OTOCs generalize standard thermal correlation functions, but are not time-ordered and consequently do not admit an interpretation as simple response functions. Nevertheless, via application of the Kubo–Martin–Schwinger (KMS) condition, all thermal $n$ -point correlators (including OTOCs) exhibit relationships under analytic continuation of time arguments by $-i\beta$ and under cyclic permutations of operators (Haehl et al., 2017, Chaudhuri et al., 2018).

2. Physical Interpretation: Scrambling and Quantum Chaos

Thermal OTOCs probe the degree to which initially commuting or nearly commuting operators, $W$ and $V$ , fail to commute after time evolution. Their late-time behavior quantifies the sensitivity of the system to perturbations: rapid growth of $C_\mathrm{comm}(t)$ , or equivalently strong decay of $F_\beta(t)$ , is interpreted as a signature of information scrambling and quantum chaos (Haehl et al., 2017, Hunt, 29 Nov 2025).

In chaotic systems at finite temperature (including strongly interacting many-body Hamiltonians, holographic models, and large- $N$ matrix models), $C_\mathrm{comm}(t)$ often exhibits an intermediate-time exponential regime:

$C_\mathrm{comm}(t) \sim \epsilon\,e^{\lambda_L t}$

where $\lambda_L$ is called the thermal quantum Lyapunov exponent. The exponential window, when present, reflects the onset of quantum chaos and is bounded by the fundamental Maldacena–Shenker–Stanford (MSS) bound:

$\lambda_L \le 2\pi k_B T/\hbar,$

applicable for thermal OTOCs in quantum systems (Hunt, 29 Nov 2025).

The physical content of thermal OTOCs is distinguished from standard correlators by the presence of operator non-commutativity at late times and the lack of simple interpretation as linear response; their exponential early-time growth in chaotic systems is directly connected to the butterfly effect in quantum mechanics.

3. Computational and Analytical Representation

The structure of thermal OTOCs is rooted in the Schwinger–Keldysh and Keldysh multi-fold contour formalism. Using the spectral representation, thermal $n$ -point OTOCs can be expanded in a basis of nested commutator structures—the "causal basis"—with structure constants given by generalized spectral functions derived from analytic continuation and KMS conditions (Chaudhuri et al., 2018, Haehl et al., 2017). Explicitly, for a general $n$ -point function:

$G^{(n)}(t_1,\ldots,t_n) = \int \frac{d^n\omega}{(2\pi)^n} \, (2\pi)\delta(\Sigma_i\omega_i) \, \rho_n(\omega_1,...,\omega_n) \, e^{-i\sum_j \omega_j t_j} f_\beta(\{\omega_j\})$

where $\rho_n$ are nested-commutator spectral densities, and $f_\beta$ encodes the thermal (KMS) factors (Chaudhuri et al., 2018).

All OTOC orderings can be written in terms of the (n−1)! causal (fully nested-commutator) structures, with exact transformation coefficients dictated by the KMS periodicity and analytic continuation. This structural redundancy is responsible for various "out-of-time-order fluctuation-dissipation theorems" (FDTs), generalizing the standard FDT to the OTOC setting (Tsuji et al., 2016).

4. Dynamical Behavior and Scaling: Model-Specific Findings

Thermal OTOCs have been computed for a wide variety of models, with diagnostics including the time window and scaling of exponential growth, operator spreading, and correspondence to phase transitions or universality classes.

Quantum Phase Transitions: In the Rabi and Dicke models, the long-time averaged infinite-temperature OTOC (for $W=V=a^\dag a$ ) exhibits a pronounced minimum as a function of coupling $g$ exactly at the quantum critical point, and its scaling with system parameters matches critical exponents and universality classes (Sun et al., 2018).

Temperature and Magnetic Field Dependence: In quantum billiard systems with a transverse magnetic field, $\lambda_L(T,B)$ extracted from the early-time exponential window of $C_T(t) = -\langle [x(t),p]^2 \rangle_\beta$ , shows a thermal-to-magnetic crossover: maximal at high $T$ , low $B$ , and vanishing as $B\to\infty$ due to cyclotron localization (Beetar et al., 5 Feb 2026).

Instanton Effects and the MSS Bound: In barrier models and double-well systems, instanton tunneling configurations dominate the late-time OTOC behavior below a characteristic temperature $T_c = \hbar\omega_b/2\pi k_B$ , quenching the OTOC growth rate and ensuring satisfaction of the MSS bound (Hunt, 29 Nov 2025).

Critical Quenches: In 1+1d CFT, global quantum quenches generate effective thermal behavior, and thermal OTOCs distinguish chaotic from integrable CFTs via exponential Lyapunov growth, with $\lambda_L$ saturating the $2\pi T$ upper bound at late times (Das et al., 2021).

Integrable and Nonintegrable Many-Body Systems: In exactly solvable models like the XY spin chain, OTOCs show ballistic wavefront propagation with a butterfly speed $v_B$ dependent on control parameters and universal functional forms for the light-cone and front exponents. Late-time decay shows model- and operator-dependent power-law scaling (Bao et al., 2019). In generic many-body non-integrable systems, OTOCs provide a probe of the fine structure and deviations from the ETH, with saturation behavior governed by operator-dependent random-matrix statistics (Brenes et al., 2021).

5. Experimental Access and Measurement Protocols

The challenges of measuring thermal OTOCs stem from their non-time-ordered structure and the need for forward and backward evolution. Several protocols have been developed and experimentally demonstrated:

Thermofield Double State Encoding: Measurement of thermal OTOCs via two-copy protocols using the thermofield double (TFD) state, where a pure entangled state in a doubled Hilbert space is prepared with temperature encoded in the entanglement spectrum (Green et al., 2021, Sundar et al., 2021).
Decoherence and Error Robustness: Protocols using coupled spin chains and parent Hamiltonians have been shown to yield robust thermal OTOC signals in the presence of decoherence, via normalization and error mitigation strategies (Sundar et al., 2021).
No Time-Reversal Measurement: Schemes based on forward-only quantum evolution utilize interferometric or tomography protocols, extracting matrix elements and population probabilities in the thermal ensemble, without the need for time-reversal or negative Hamiltonians (Blocher et al., 2020).
Variational and Neural Network Approaches: For infinite-temperature OTOC computation in large-scale systems, variational machine-learning (RBM) techniques scale efficiently and accurately capture early-time OTOC evolution (Wu et al., 2019).

Numerical exact diagonalization, path-integral (RPMD/Matsubara), and other tensor-network approaches have been used for detailed model studies (Hunt, 29 Nov 2025).

6. Generalizations, Relations, and Theoretical Structure

A key structural result is the existence of general fluctuation-dissipation theorems for thermal OTOCs. By considering various forms of statistical averaging (bipartite, physical), higher-order KMS conditions, and operator-monotone extensions, one can derive an infinite hierarchy of exact relations between thermal OTOCs and nonlinear response functions, with corrections bounded by quantities such as the Wigner–Yanase skew information (Tsuji et al., 2016).

Thermal OTOCs are not just chaos diagnostics; they encode the interplay between operator complexity, integrability, many-body localization, entanglement growth, and quantum criticality.

Model Class / Protocol	OTOC Type	Key Thermal Property
Rabi/Dicke, few-body	$\langle a^\dag a(t) a^\dag a(0) a^\dag a(t) a^\dag a(0)\rangle_\beta$	Minimum at phase transition, scaling exponents (Sun et al., 2018)
Single-particle billiard	$C_T(t) = -\langle [x(t),p]^2 \rangle_\beta$	$\lambda_L(T,B)$ crossover (Beetar et al., 5 Feb 2026)
2D spin model, RBM	Infinite- $T$ , any $W,V$	Early-time propagation, polynomial compute (Wu et al., 2019)
CFT $_2$ after quench	$F_\beta(t)$	Emergent thermalization, maximal $\lambda_L$ (Das et al., 2021)

7. Limitations and Interpretation Caveats

Care is needed in interpreting exponential growth of thermal OTOCs. For example, in systems with local inverted-harmonic behavior, exponential growth can arise even in the absence of global chaos, reflecting local instability rather than true scrambling (Hashimoto et al., 2020). Similarly, in low-dimensional, single-body or integrable systems, OTOCs can flatten or show oscillatory behavior at late times due to coherence and finite Hilbert space.

Instanton effects, quantum tunneling, and Boltzmann suppression at low temperature can significantly reduce OTOC growth rates, and the naive classical Lyapunov exponent is typically not recovered unless the thermal population resolves the relevant saddle-point structure (Hunt, 29 Nov 2025, Akutagawa et al., 2020).

Thus, while the presence, scaling, and bounds of exponential OTOC growth are signatures of quantum chaos and scrambling, model details, operator choice, and temperature regime impose critical constraints on their diagnostic utility.

Key references: (Sun et al., 2018, Haehl et al., 2017, Chaudhuri et al., 2018, Hunt, 29 Nov 2025, Das et al., 2021, Beetar et al., 5 Feb 2026, Green et al., 2021, Sundar et al., 2021, Bao et al., 2019, Wu et al., 2019, Akutagawa et al., 2020, Tsuji et al., 2016, Hashimoto et al., 2020, Brenes et al., 2021, Blocher et al., 2020)