Out-of-Time-Order Correlators and Lyapunov Growth

The quantum chaos bound is a conditional theorem about a specially regulated thermal four-point function. It does not say that every out-of-time-order correlator (OTOC) grows exponentially, that every black hole saturates the bound, or that one fitted exponential proves a smooth horizon. Under a thermal KMS state, a particular operator ordering, analyticity and boundedness in a complex half-strip, an approximate factorization window, and a parametrically small connected correction, the Lyapunov exponent satisfies $\lambda_L\le 2\pi/\beta$. This page states those assumptions before deriving the complex-analytic step.

Required background. Quantum Information supplies density operators, norms, and operator inequalities; Gauge Theory and String-Theory Preparation supplies the large-system and holographic examples in which the bound is often applied.

Helpful background. Semiclassical Gravity supplies Hawking temperature and the near-horizon regime; Quantum Fields in Curved Spacetime supplies thermal correlation functions on a fixed background; and Black-Hole Information supplies the distinction between scrambling, recovery, and interior interpretation.

The regulated thermal correlator

Let $H$ be a Hamiltonian, $\rho=e^{-\beta H}/Z$ a thermal state, and $V,W$ bounded Hermitian operators with vanishing thermal one-point functions. Define $y$ by $y^4=\rho$ and use the regulated OTOC

$$F(t)=\operatorname{Tr} \left[yVyW(t)yVyW(t)\right], \qquad W(t)=e^{iHt}We^{-iHt}.$$

The four factors of $y$ fix the operator ordering or regularization. Moving all thermal factors together produces a different correlator whose analytic domain and transient behavior need not be the same. Let

$$F_d=\operatorname{Tr}(y^2Vy^2V)\, \operatorname{Tr}(y^2Wy^2W)$$

be the factorized value. The proposed exponential growth regime is a time window $t_d\ll t\ll t_*$ in which

$$F(t)=F_d-\varepsilon e^{\lambda_L t}+O(\varepsilon^2e^{2\lambda_L t}), \qquad 0\lt\varepsilon\ll1.$$

The early endpoint lies after ordinary dissipation; the late endpoint lies before saturation, finite-size noise, or recurrence. Without this window, $\lambda_L$ is not defined by the theorem.

Conditional statement of the chaos bound

Theorem. Assume that the analytically continued normalized correlator is holomorphic in the thermal half-strip, obeys the required reflection property, is bounded there by its factorized normalization up to controlled small corrections, and factorizes on the strip boundaries throughout the stated time window. Assume bounded or suitably normalized operators and a parametrically small connected part. Then its real-time departure from factorization cannot grow faster than

$$\lambda_L\le\frac{2\pi}{\beta}$$

at leading order in that window. Maldacena, Shenker, and Stanford give the precise analyticity, boundedness, and factorization hypotheses in Maldacena, Shenker, and Stanford 2016, §§ 2–4.

These chaos-bound assumptions are not automatic consequences of unitarity. Local quantum field theories require smearing or other control of unbounded operators. Large-$N$ factorization supplies a small parameter in common examples, but a finite system at arbitrary time has no indefinitely small connected term. Chemical potentials, additional conserved charges, or a different analytic strip can modify the applicable statement.

Proof architecture from the half-strip

Choose a time $t_0$ after dissipation and a factorization-error bound $\epsilon_{\rm fac}\ge0$, and set $f(z)=F(t_0+z)/(F_d+\epsilon_{\rm fac})$. The boundary and interior boundedness hypotheses ensure that $f$ maps the half-strip

$$\operatorname{Re}z\gt0, \qquad -\frac{\beta}{4}\lt\operatorname{Im}z\lt\frac{\beta}{4}$$

into the unit disk. A conformal map sends this half-strip to the disk. Schwarz–Pick contraction then bounds the derivative of $f$ by the hyperbolic metric. For real $f(t)$, its precise half-strip form is

$$\frac{1}{1-f(t)}\left|\frac{df}{dt}\right| \le\frac{2\pi}{\beta} \coth\!\left(\frac{2\pi t}{\beta}\right)\frac{1+f(t)}{2}.$$

For $t\gg\beta$ the right-hand side is at most $2\pi/\beta+O(\beta^{-1}e^{-4\pi t/\beta})$. Translating back to $F$ gives a bound on the growth of $F_d-F$ with an additive $\epsilon_{\rm fac}$ on the right-hand side. The exponential departure must therefore exceed that error floor while remaining small compared with $F_d$: $\epsilon_{\rm fac}\ll F_d-F\ll F_d$. The finite-time correction and error floor are explicit in Maldacena, Shenker, and Stanford 2016, § 4.1, equation (18), and § 4.2, equation (26).

In this regime, if $1-f(t)=\widetilde\varepsilon e^{\lambda_L t}$ with positive $\widetilde\varepsilon$, substitution gives the claimed upper bound. The factor $2\pi$ is geometric: it comes from the width $\beta/2$ of the analytic strip. This proof does not infer exponential growth; it limits the rate if the assumed exponential window exists.

As an independent check, a constant $f$ has zero logarithmic derivative and satisfies the inequality strictly. A trial form $f=1-\widetilde\varepsilon e^{(3\pi/\beta)t}$ violates it before saturation, showing exactly where the proposed rate conflicts with the analytic bound.

OTOCs, commutators, and gravitational saturation

For Hermitian unitary operators in a compatible thermal ordering, the squared commutator

$$C(t)=-\operatorname{Tr}\!\left(\rho[W(t),V]^2\right)$$

can be related to the real part of an OTOC plus time-ordered pieces. The relation changes for nonunitary operators and for another regularization. Thus commutator growth and decay of the chosen $F(t)$ must not be identified without checking normalization and ordering.

Large-$c$ holographic correlators provide examples with a rate approaching $2\pi/\beta$; the relevant continuation and Regge regime are exhibited in Roberts and Stanford 2015, § 3. Near-horizon shock-wave scattering gives the same scale in controlled geometries. These are saturation examples, not extra premises of the theorem and not evidence that all black-hole states share one interior.

Counterexamples and limitations

A free oscillator or integrable spin chain can have oscillatory OTOCs with no parametric exponential interval. It therefore supplies a counterexample to universal saturation while fully respecting the upper bound. Finite-dimensional systems eventually saturate and recur; fitting an exponent across that region does not test the theorem. At weak coupling, multiple kinetic timescales can also make a short apparent fit regulator dependent.

A numerical claim should therefore show a stable fit interval under changes of time step, precision, operator choice, and thermal regulator, then compare the inferred exponent with an independently measured dissipation time. A straight line on a short semilogarithmic plot is not by itself evidence for a Lyapunov regime.

The theorem does not prove rapid state preparation, decodability, or entropy production. Nor does it construct an interior. Generic quantum-information theory, stable thermodynamics, and explicit string constructions remain beyond this page; only the bounded OTOC statement and its black-hole use are retained.

Exercises

Assume $F_d=1$ and $F(t)=1-10^{-8}e^{\lambda t}$ throughout a controlled window at inverse temperature $\beta=2$. What maximum $\lambda$ is compatible with the bound, and why can the formula not be extrapolated indefinitely?

Solution

The theorem gives $\lambda\le2\pi/\beta=\pi$. The connected correction becomes order one at a time of order $\lambda^{-1}\log(10^8)$, where factorization and the linearized exponential form cease to hold. Beyond that scrambling time, saturation, finite-size fluctuations, and eventual recurrences replace the early growth regime, so extrapolating the exponential would violate both the bounded correlator and the hypotheses used in the proof.

References

• Maldacena, Juan, Stephen H. Shenker, and Douglas Stanford. 2016. “A Bound on Chaos.” Journal of High Energy Physics 08, 106. arXiv:1503.01409.
• Roberts, Daniel A., and Douglas Stanford. 2015. “Diagnosing Chaos Using Four-Point Functions in Two-Dimensional Conformal Field Theory.” Physical Review Letters 115, 131603. arXiv:1412.5123.