OTOC: Probing Quantum Chaos & Scrambling

• OTOC is a family of four-point dynamical correlators that quantify operator growth, quantum information scrambling, and sensitivity to perturbations.
• They capture early-time exponential behavior linked to classical Lyapunov exponents and spatial wavefront dynamics, bridging quantum and semiclassical analyses.
• OTOCs serve as versatile diagnostics for phase transitions, exceptional points, and measurement challenges, advancing both theoretical insights and experimental techniques.

Searching arXiv for recent and foundational OTOC papers to ground the article. Tool unavailable in this interface, so I will rely strictly on the provided arXiv-sourced data block and cite only those papers. The out-of-time-order correlator (OTOC) is a family of four-point dynamical correlators used to quantify operator growth, quantum information scrambling, and aspects of quantum chaos. In common notation,

$$\mathcal{F}(t)=\langle \hat{W}^\dagger(t)\hat{V}^\dagger(0)\hat{W}(t)\hat{V}(0)\rangle,$$

while an equivalent and often emphasized form is the squared commutator

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$

For unitary $W$ and $V$, the thermal OTOC is related to the squared commutator by

$C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$

and for Hermitian operators by $C_\beta(\tau)=2[1-F_\beta(\tau)]$. At infinite temperature one frequently uses

$$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$

so that the OTOC becomes an unweighted trace over the Hilbert space (PG et al., 2020, Emori et al., 27 Dec 2025).

1. Formal definitions and representations

The basic distinction within the OTOC literature is between the four-point function $\mathcal F(t)$ and the commutator-squared form $C(t)$. The former makes the out-of-time operator ordering explicit; the latter emphasizes sensitivity to the failure of initially commuting operators to remain commuting under Heisenberg evolution. In thermal settings one writes

$$C_T(t)\equiv-\langle [W(t),V(0)]^2\rangle_T, \qquad \langle O\rangle_T=\frac{1}{Z(T)}\mathrm{Tr}[e^{-\beta H}O],$$

while in an energy eigenbasis one may resolve the correlator into microcanonical contributions,

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$0

which is the form used in several single-particle and few-body calculations (Hashimoto et al., 2017, Akutagawa et al., 2020).

The infinite-temperature version is especially prominent because it is experimentally more accessible and naturally compatible with trace-estimation protocols. In the Rabi and Dicke models, for example, the infinite-temperature OTOC is written as

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$1

with $C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$2, the photon number operator (Sun et al., 2018). In DQC1-based measurement schemes, the same normalized-trace structure appears directly in the ancilla signal, which is one reason infinite-temperature OTOCs recur throughout the algorithmic and experimental literature (PG et al., 2020).

A further important distinction is between local, nonlocal, thermal, microcanonical, and spatially resolved OTOCs. In the one-dimensional XY model the choice of local operator $C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$3 versus nonlocal operators $C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$4 changes both the early-time power law and the late-time decay exponent, even though the butterfly velocity is operator-independent within a fixed Hamiltonian (Bao et al., 2019). In spatially extended classical systems such as the Kuramoto-Sivashinsky equation, the OTOC is formulated as a space-time dependent object,

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$5

with $C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$6 defined from the linear response of two initially nearby field configurations (Roy et al., 2023). These variants are not competing definitions so much as different representations of the same underlying notion: the sensitivity of evolved operators or trajectories to initially local perturbations.

2. Early-time growth, wavefronts, and semiclassical correspondence

A central semiclassical statement is that, for short evolution times, the commutator is replaced by a Poisson bracket,

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$7

so that the OTOC is quasiclassically described by

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$8

For chaotic dynamics this yields exponential growth,

$C(t)=-\langle[\hat{A}(t),\hat{B}]^2\rangle.$9

with $W$0 the classical Lyapunov exponent, and this agreement with the Wigner-Moyal formalism holds in the short-time regime before the Ehrenfest-time breakdown of the diagonal approximation (Michel et al., 2024).

The wavefront formulation makes the same physics spatially resolved. In the XY chain the OTOC outside the wavefront obeys the conjectured universal form

$W$1

and the butterfly speed $W$2 depends on the anisotropy $W$3 and magnetic field $W$4 but not on whether the chosen operators are local or nonlocal (Bao et al., 2019). In the Kuramoto-Sivashinsky equation one introduces the velocity-dependent Lyapunov exponent

$W$5

and the butterfly velocity is the largest $W$6 for which $W$7. The linearized equation already exhibits a sharp light-cone and admits a saddle-point analysis, while the full nonlinear equation retains a similar butterfly velocity and a reduced maximal Lyapunov exponent because nonlinear mode coupling dilutes the instability (Roy et al., 2023).

Analytically tractable quantum-chaotic maps supply a complementary perspective. In the quantum baker’s map, a semiquantum approximation valid up to the Ehrenfest time $W$8 yields OTOC growth at the exponential rate of the classical Lyapunov exponent $W$9, but modulated by slowly changing coefficients rather than by a pure $V$0 law. This model shows explicitly that the observable choice matters: using projectors, the OTOC reduces to a function of the singular values of a truncated unitary propagator, linking scrambling to subspace leakage under Heisenberg evolution (Lakshminarayan, 2018).

3. Late-time behavior, saturation, and limits of the chaos diagnosis

A recurring result across models is that exponential growth is neither universal nor sufficient as a criterion for quantum chaos. In one-dimensional van der Waals molecular potentials, the OTOC for high-lying vibrational states near the dissociation threshold grows exponentially over a finite interval even though the system is integrable. The transition occurs when the outer classical turning point enters a region of negative curvature $V$1, and the long-time behavior remains oscillatory rather than saturating; the growth is therefore tied to sensitive classical dynamics, not chaos (Li et al., 2023).

The converse failure also occurs. In the stadium billiard, a standard example of classical chaos, no clear exponential growth of the OTOC is found. Instead, the OTOC rises and saturates to a constant value at late times, while in integrable examples with commensurate spectra such as the harmonic oscillator and particle in a box the OTOC is exactly periodic. The same study further shows that the quantum OTOC can deviate from its classical value at a timescale much earlier than the Ehrenfest time, so naive quantum-classical correspondence is not reliable even in single-body systems (Hashimoto et al., 2017).

Late-time relaxation is model-dependent and often controlled by structures other than Lyapunov exponents. In mixed systems, the approach to equilibrium is governed by generalized resonances of the Perron-Frobenius operator. For the standard map one finds

$V$2

with the leading generalized resonance setting the decay rate of the OTOC toward equilibrium as regular islands appear and disappear in phase space (Notenson et al., 2023). In chaotic few-site Bose-Hubbard systems, quasiclassical diagonal approximations predict saturation to an ergodic average of the squared Poisson bracket, but this asymptotic value strongly underestimates the actual quantum plateau. The discrepancy indicates that nondiagonal, genuinely quantum interference terms dominate beyond the Ehrenfest time (Michel et al., 2024).

Coupling to an environment changes the structure again. For finite open systems with discrete spectra, OTOCs of the form

$V$3

do not exhibit chaotic growth; instead they decay and exponentially saturate,

$V$4

with different components decaying on dephasing and inelastic-relaxation time scales. In the classical-environment limit, the OTOC evolution can be mapped to the density-matrix evolution of two systems coupled to the same noise realization (Syzranov et al., 2017). Studies of single-body systems further report that instantons reduce OTOC growth rates and that ring polymer molecular dynamics is not sufficient to guarantee the Maldacena bound in general, while analytically continued Matsubara dynamics yields qualitatively different instability structure around the instanton (Hunt, 29 Nov 2025).

4. Measurement protocols and computational algorithms

Because OTOCs interleave forward and backward evolution, their measurement is algorithmically nontrivial. A particularly economical route is the DQC1 protocol, or Deterministic Quantum Computation with One Qubit. In this scheme the probe qubit is pure, the system register may be maximally mixed, and the controlled circuit implements the composite operator $V$5. Measuring $V$6 or $V$7 on the ancilla yields the real or imaginary part of the relevant matrix element,

$V$8

and with a maximally mixed register this becomes the normalized trace. To estimate the OTOC to accuracy $V$9, the protocol requires $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$0 repetitions, independent of system size, and it gives an exponential speedup over the best known classical algorithm provided the OTOC operator admits an efficient gate decomposition (PG et al., 2020).

The same work extends trace estimation to spectral characterization. By adding a second ancilla register and using quantum Fourier transforms around a controlled $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$1, one samples the spectral density

$C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$2

with resolution set by the number $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$3 of ancilla qubits. This gives access not only to a mean scrambling diagnostic but also to level statistics of the OTOC operator itself (PG et al., 2020).

A conceptually different route is the Jarzynski-like equality

$C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$4

in which the OTOC is expressed as derivatives of an average over a complex quasiprobability distribution $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$5. This formulation suggests indirect measurement through weak measurements or interference experiments and, in the interference version, does not require explicit time reversal in any single trial (Halpern, 2016).

Recent hardware-oriented work compares three protocols: the rewinding time method (RTM), the weak-measurement method (WMM), and the irreversibility-susceptibility method (ISM). In experiments on an XXZ spin-$C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$6 chain prepared in a thermal Gibbs state on the numerical emulator of the trapped-ion quantum computer reimei, RTM and WMM showed method-dependent late-time deviations, whereas ISM was reported as the first experimental demonstration of an approach that avoids explicit backward evolution at the cost of larger statistical error (Emori et al., 27 Dec 2025). Together, these methods show that OTOC measurement is not a single protocol but a family of trace-estimation, interferometric, weak-measurement, and susceptibility constructions.

5. OTOCs as probes of criticality, exceptional points, and driven dynamics

Although OTOCs were introduced as probes of chaos and scrambling, they are also effective dynamical diagnostics of phase structure. In the Rabi and few-body Dicke models, the long-time-averaged infinite-temperature OTOC

$C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$7

develops a local minimum at the quantum critical point separating the normal and superradiant phases. Finite-parameter scaling of the minimum position yields critical exponents $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$8 for the Rabi model and $C_\beta(\tau)=2\{1-\Re[F_\beta(\tau)]\},$9 for the Dicke model, indicating that the two transitions belong to the same universality class (Sun et al., 2018).

In non-Hermitian quantum Ising systems, the OTOC diagnoses both the ground-state exceptional point associated with the Yang-Lee edge singularity and the dynamical exceptional point in the excited spectrum. The time evolution separates into an oscillatory short-time regime, described by $C_\beta(\tau)=2[1-F_\beta(\tau)]$0-dimensional Yang-Lee and $C_\beta(\tau)=2[1-F_\beta(\tau)]$1-dimensional Ising scaling theories near the ground-state exceptional point, and a long-time regime with exponential growth

$C_\beta(\tau)=2[1-F_\beta(\tau)]$2

controlled by the dynamical exceptional point (Zhai et al., 2019).

Driven conformal field theories provide a nonequilibrium extension of the same theme. In large-$C_\beta(\tau)=2[1-F_\beta(\tau)]$3 CFTs subjected to periodic drive, OTOCs display exponential, oscillatory, and power-law behavior in the heating phase, the non-heating phase, and on the phase boundary, respectively. The corresponding butterfly velocity depends explicitly on operator position because the drive introduces spatial inhomogeneity. By contrast, in the driven Ising CFT, which is integrable, OTOCs do not display exponential growth in any regime, making the OTOC a discriminator between chaotic and integrable driven CFT dynamics (Das et al., 2022).

In holographic QCD with instantons, the OTOC of the holographic Skyrmion is used to detect baryonic phase structure. The quantum OTOC can exhibit an imaginary Lyapunov coefficient, interpreted in that work as a signature of a possibly metastable baryonic status in the presence of instantons, whereas sufficiently large instanton charge yields a real Lyapunov exponent and genuinely chaotic behavior. The same analysis stresses that the classical OTOC does not capture the theta-dependent features in the same way, implying that the instanton-induced phase structure is basically dominated by quantum properties (Li et al., 2024).

6. Large-$C_\beta(\tau)=2[1-F_\beta(\tau)]$4, holography, and higher-order generalizations

In large-$C_\beta(\tau)=2[1-F_\beta(\tau)]$5 systems the OTOC admits an explicitly two-way construction. On the thermofield-double background, one pair of operators is treated as a source and the other as a probe, and the correlator is assembled from forward- and backward-propagating scrambling modes on a double Keldysh contour. For large-$C_\beta(\tau)=2[1-F_\beta(\tau)]$6 SYK this yields a closed-form OTOC and a proof of the relation

$C_\beta(\tau)=2[1-F_\beta(\tau)]$7

linking the Lyapunov exponent $C_\beta(\tau)=2[1-F_\beta(\tau)]$8 to the high-frequency decay of the spectral function (Gu et al., 2021).

Holographic models provide an especially sharp realization of multi-point scrambling diagnostics. In holographic EPR pairs, the four-point and six-point OTOCs can be computed both from a holographic influence functional on shock-wave-perturbed worldsheets and from worldsheet scattering in the eikonal approximation, with agreement between the two approaches. The four-point function has the form

$C_\beta(\tau)=2[1-F_\beta(\tau)]$9

while the six-point function has a scrambling time larger by $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$0. In that setting, higher-point OTOCs probe more fine-grained chaos and disentanglement of the EPR pair (Kawamoto et al., 11 Mar 2025).

A more general algorithmic interpretation appears in higher-order OTOCs and quantum signal processing. For a spatially resolved truncated propagator with singular values $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$1, the $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$2-th order OTOC obeys

$$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$3

so each $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$4 is the $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$5-th Fourier component of the phase distribution or, equivalently, a Chebyshev polynomial transform of the singular values. Replacing fixed Pauli flips by tunable $$C_{W,V}(\tau)=1-\mathrm{Re}\left\langle W(\tau)^\dagger V^\dagger W(\tau)V\right\rangle, \qquad W(\tau)=U^\dagger(\tau)WU(\tau),$$6-rotations yields programmable polynomial filters and the scheme termed OTOC spectroscopy, which generalizes conventional OTOCs into a mode-resolved probe of scrambling and spectral structure (Fujii, 27 Nov 2025).

Taken together, these developments place the OTOC at the intersection of semiclassical chaos, many-body operator growth, nonequilibrium field theory, open-system dynamics, holography, and quantum algorithms. The resulting picture is technically rich but conceptually restrained: OTOCs are powerful precisely because they are sensitive not only to chaos, but also to integrability, criticality, exceptional points, environmental decoherence, spectral structure, and the detailed architecture of measurement protocols.