Out-of-Time-Order Correlators

A regulated thermal out-of-time-order correlator measures the failure of initially commuting simple operators to remain ordered under time evolution. In a semiclassical black-hole regime, its leading connected change is controlled by high-energy near-horizon scattering. A separate complex-analysis argument bounds the growth rate under explicit analyticity, boundedness, and factorization assumptions. Keeping those two arguments distinct prevents the Einstein value from being promoted into an unconditional theorem about every quantum system.

Required background. Gauge Theory and String-Theory Preparation provides thermal correlator and large-N normalization; Quantum Information provides operator growth and scrambling language.

Helpful background. Quantum Gravity supplies the bulk eikonal interpretation, while Black-Hole Perturbation Theory supplies the horizon scattering background.

The thermal chaos bound

Let $y$ satisfy $y^4=e^{-\beta H}/Z$ and take bounded, Hermitian, few-body operators $V$ and $W$ normalized so their disconnected thermal correlators are order one. Define

$$F(t)=\operatorname{Tr}\!\left[yVyW(t)yVyW(t)\right].$$

Subtract thermal one-point functions and define the positive disconnected normalization $F_d=\operatorname{Tr}(y^2Vy^2V)\operatorname{Tr}(y^2Wy^2W)$.

Theorem (conditional thermal chaos bound). Choose $t_0$ after dissipation. Assume $f(u)=F(t_0+u)/(F_d+\varepsilon)$ is holomorphic and bounded by one in $\operatorname{Re}u\gt0$, $|\operatorname{Im}u|\lt\beta/4$, and real on the positive real axis; $\varepsilon$ bounds the factorization error. If a window before scrambling has $\varepsilon\ll F_d-F(t)\ll F_d$ and

$$\frac{F_d-F(t)}{F_d}\sim a e^{\lambda_L t},$$

then

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

up to corrections controlled by the factorization error and the finite width of the window. The exact regulated ordering, not an unregulated symbol such as $\langle V(t)WV(t)W\rangle$, is part of the hypotheses. Maldacena, Shenker, and Stanford 2016, §§2–4.2, pp. 5–12 state the conjecture and give the analytic argument.

Proof architecture

The key step is to normalize the correlator to a function $f$ whose magnitude is bounded by one on the boundary of a half-strip. Thermal KMS relations and Cauchy–Schwarz estimates supply that boundary control, while factorization keeps $f$ near one in the physical window. Map the half-strip conformally to the unit disk. The Schwarz–Pick theorem bounds the derivative of any holomorphic map from the disk to itself.

For real $u\gt0$, the Schwarz–Pick step gives the explicit bound

$$\frac{|f'(u)|}{1-f(u)} \le\frac{\pi}{\beta}[1+f(u)] \coth\!\left(\frac{2\pi u}{\beta}\right).$$

For $u\gg\beta$ and $f$ close to one, the right side approaches $2\pi/\beta$. The connected deviation must exceed $\varepsilon$ before its logarithmic derivative can be identified with $\lambda_L$ Maldacena, Shenker, and Stanford 2016, §4.1, Eq. (17), and §4.2. The proof does not establish an exponential window or saturation.

The argument’s architecture can be checked independently: verify holomorphy from the spectral representation, derive the boundary norm estimate with the chosen $y$ insertions, perform the strip-to-disk map, and only then take the factorized growth window. Dropping any link changes the theorem.

Bulk scattering and saturation

In two-derivative Einstein gravity, an early perturbation is boosted by $e^{2\pi Tt}$ near the horizon. The eikonal phase of its collision with a late probe produces

$$1-\frac{F(t,\mathbf{x})}{F_d} \sim \varepsilon_{\rm grav}e^{2\pi T(t-\lvert\mathbf{x}\rvert/v_B)}$$

before saturation, where $\varepsilon_{\rm grav}$ is a dimensionless operator-dependent amplitude of order $G_{d+1}/L^{d-1}$ for fixed AdS-scaled data. Thus $\lambda_L=2\pi T$ and the conditional bound is saturated. Shenker and Stanford 2015, §§2–4, pp. 6–25 derive the scattering representation and show how string-scale effects alter the phase and transverse profile.

The bulk calculation controls the early connected term in a semiclassical, large-N, eikonal regime. Finite $N$ eventually enforces saturation and recurrences. String exchange can reduce the effective Regge intercept, and inelastic channels can change the simple single-exponential description. These corrections need not violate the theorem; they usually change the regime or reduce the exponent.

Failure modes and discriminating examples

An integrable or free theory can satisfy the analytic hypotheses yet have no sustained exponential growth, so its Lyapunov exponent is zero or inapplicable. This is a counterexample to universal saturation, not to the bound. Conversely, unbounded operators, an absent factorization window, or a correlator with inadequate strip control can fall outside the theorem.

The squared commutator and the regulated OTOC are related but not identical without assumptions about ordering and disconnected pieces. A decay in one normalized correlator should not be labeled $\lambda_L$ until its time window, prefactor, and operator normalization are stated. Spatially extended systems also require a position-dependent formulation; the front can broaden and need not have a single sharp velocity.

The proof only establishes a bound on an admitted exponential component. It does not prove ergodicity, diagnose all forms of quantum chaos, or show that every black hole saturates. Higher-spin theories and finite-coupling strings provide important adverse cases in which the Einstein approximation is not the correct bulk input.

Checks and physical scope

A practical calculation should compare several orderings, confirm the KMS shifts, and show that the connected term is larger than numerical or $1/N$ errors but smaller than one. The fit window must begin after ordinary dissipation and end before scrambling. Moving either endpoint should not change the inferred exponent within quoted uncertainty.

On the gravity side, compute both the shockwave exponent and the OTOC eikonal phase in the same boundary-time normalization. Agreement checks the horizon surface gravity. The amplitude should become order one near $t_*\sim(\beta/2\pi)\log N_{\rm eff}$, marking the failure of the leading connected expansion rather than a divergence of the exact correlator.

Exercises

Explain why the Schwarz–Pick step bounds an existing exponential but does not prove that one exists, and give a failure case that does not contradict the theorem.

Solution

Schwarz–Pick bounds the logarithmic rate at which a bounded holomorphic function can depart from its factorized value. To call that rate $\lambda_L$, one must separately show a time interval in which the connected part is proportional to a small constant times $e^{\lambda_Lt}$. An integrable system may obey thermal analyticity and boundedness while showing oscillatory or power-law behavior instead. It has no saturating exponential window, so it illustrates the theorem’s limited conclusion rather than violating the inequality.

References

• Maldacena, Juan, Stephen H. Shenker, and Douglas Stanford. “A Bound on Chaos.” Journal of High Energy Physics 08 (2016) 106. arXiv:1503.01409.
• Shenker, Stephen H., and Douglas Stanford. “Stringy Effects in Scrambling.” Journal of High Energy Physics 05 (2015) 132. arXiv:1412.6087.