Classical and Quantum Chaos from Foundations to Holography

Abstract: These notes develop the diagnostics of chaos from their classical definitions through to holography, in six lectures given at the ST4 Workshop, Chennai Mathematical Institute, in 2026. No prior exposure to chaos theory is assumed. Lectures 1 and 2 define classical chaos geometrically and build the instruments that measure it, every one of which reads a phase-space trajectory. Quantization removes that trajectory, and Lectures 3 and 4 rebuild the instruments from spectra and eigenstates instead: level statistics, random matrix theory, the spectral form factor and eigenstate thermalization, as much the working toolkit for thermalization in many-body physics as for holography. What they return is a classification and a count rather than a rate. Lecture 5 recovers the rate from the squared commutator, treats scrambling as operator growth, follows the commutator to the out-of-time-order correlator, marks where that diagnostic stops being reliable, and reaches the butterfly velocity and the chaos bound. Lecture 6 constructs the eternal black hole, the thermofield double and the shock-wave geometry behind that bound, then follows two probes whose exponents discriminate between backgrounds where the horizon result cannot, one tracking a thermodynamic phase transition, the other resolving an anisotropy induced by a background field. Every lecture closes with exercises. Claude Opus 5, an AI assistant developed by Anthropic, worked as a research assistant under the author's direction.

• The paper develops a unified framework from Hamiltonian dynamics and KAM theory to spectral statistics, ETH, OTOCs, and holographic shock waves, while distinguishing chaos, ergodicity, mixing, and instability.
• The paper shows that spectral statistics classify universality and count states but do not determine a dynamical rate, while OTOC growth measures operator spreading and can differ substantially from the classical Lyapunov exponent.
• The paper finds that holographic black holes saturate the bound λL=2π/β, whereas probe-dependent effects restore discrimination through a measured critical scaling exponent of 0.5024 and magnetic-field-dependent chaos in holographic QCD strings.

Overview and structure

"Classical and Quantum Chaos from Foundations to Holography" (2608.19131) is a set of six lecture notes by Bhaskar Shukla, delivered at the ST4 Workshop at the Chennai Mathematical Institute in July 2026. The notes are aimed at researchers in string theory and holography who want to follow the literature on scrambling, the chaos bound, and chaotic dynamics in black-hole and holographic QCD backgrounds. They assume only Hamiltonian mechanics, elementary statistical mechanics, and — for the last two lectures — the AdS/CFT dictionary and black-hole thermodynamics. The material is organized in three parts: classical chaos (Lectures 1–2), quantum chaos (Lectures 3–4), and scrambling with holographic applications (Lectures 5–6). Every lecture closes with exercises, and the figures are computed rather than sketched, with numerical checks quoted in the captions.

The pedagogical through-line is stated in the abstract and enforced structurally: every classical diagnostic reads a phase-space trajectory, quantization removes that trajectory, and the quantum diagnostics rebuild the instruments from spectra and eigenstates. What spectral statistics return is "a classification and a count rather than a rate"; recovering a rate requires the squared commutator, and the notes are explicit about where that diagnostic stops being trustworthy.

Lecture 1: determinism without predictability

The first lecture establishes the conceptual separation between determinism and predictability. Hamilton's equations define a unique flow $\Phi^t$ on phase space, but finite-precision initial data yield a predictability time

$$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$

so improving precision by orders of magnitude buys only a logarithmic improvement in forecast horizon. The historical arc runs from Poincaré's analysis of the three-body problem — where Bruns-type nonexistence results for additional integrals show the failure is structural — through Liouville integrability and the Liouville–Arnold theorem, to Lorenz's dissipative system and the strange attractor.

Several distinctions are drawn with unusual care. Recurrence is not chaos: the Poincaré recurrence theorem follows from volume conservation and boundedness alone, with no role for nonlinearity or sensitivity, and it supplies no timescale. Ergodicity is not mixing: the rigid rotation on the two-torus is ergodic (time averages converge, with the relative gap falling from $2.9\times10^{-2}$ at $10^3$ steps to $6.7\times10^{-7}$ at $10^7$) yet carries no mixing, since a disc of initial conditions is transported without deformation. Mixing, not ergodicity, is what licenses probabilistic language. Instability is not chaos: the inverted harmonic oscillator separates exponentially but has no folding and no return; the notes insist chaos requires local separation, boundedness, and repeated stretching-and-folding acting together, and shows each ingredient failing separately in identifiable examples.

The baker's map serves as the exactly solvable paradigm: uniformly hyperbolic with Lyapunov exponents exactly $\pm\log 2$, conjugate to a Bernoulli shift, with periodic orbits counted exactly ($2^k$ points fixed by $B^k$) and a predictability horizon that is exactly one bit of initial precision lost per iteration. The Lorenz system, by contrast, is dissipative with volume contracting at rate $-(\sigma+1+\beta)$, which is what permits an attractor at all — no Hamiltonian system can possess one.

Lecture 2: the classical instruments

Five diagnostics are developed in order: linear stability, bifurcations, KAM theory, Poincaré sections, and Lyapunov exponents. The trace–determinant diagram classifies fixed points of area-preserving maps along the line $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$0, with elliptic points ($$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$1) organizing regular islands and hyperbolic points ($$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$2) organizing chaotic layers; a transverse homoclinic intersection forces, via the Smale–Birkhoff theorem, a horseshoe conjugate to a two-symbol shift with positive topological entropy. The Poincaré–Birkhoff theorem supplies the origin of island chains, and the KAM theorem — with its Diophantine non-resonance condition and the measure statement that destroyed resonant tori are dense but thin — explains the structure of mixed phase space. Greene's residue criterion and the golden-circle conjecture are presented as conjectures with only partial rigorous justification, not results.

Lyapunov exponents receive a careful treatment: the Oseledec theorem guarantees the limit only almost everywhere on an ergodic component (a caveat that matters for the Hénon–Heiles section at $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$3, which is roughly half regular and half chaotic), symplectic pairing gives $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$4 for $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$5, and the finite-time exponent $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$6 separates regular orbits (decaying as $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$7) from chaotic ones (settling on a positive plateau). The numerical recipe is Benettin renormalization, and the Hénon–Heiles sections at $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$8 and $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$9 — with energy conserved to $2.9\times10^{-2}$0 and regular/chaotic classification done by carrying a tangent vector rather than by eye — make the KAM transition concrete: at $2.9\times10^{-2}$1 the regular set still holds about 48% of the section area, so the phase space is mixed rather than chaotic.

Lecture 3: why the classical definition cannot be copied

The central obstruction is kinematic, not dynamical. Unitarity fixes state overlaps for all time, but the same is true classically for Liouville-evolved densities, so linearity of the Schrödinger equation is a red herring. The genuine obstruction is that the classical definition requires $2.9\times10^{-2}$2, and the uncertainty relation $2.9\times10^{-2}$3 puts a floor under both the initial condition and the perturbation. Bounded quantum systems have discrete spectra and quasi-periodic evolution, so long-time classical chaos is unavailable; whatever survives must live in the window between the Ehrenfest time

$2.9\times10^{-2}$4

and the Heisenberg time $2.9\times10^{-2}$5. The limits $2.9\times10^{-2}$6 and $2.9\times10^{-2}$7 fail to commute, and the notes state plainly that the classical limit is singular in the way the short-wavelength limit of wave optics is singular.

Three locations for quantum signatures are identified. Spectra: integrable systems show Poisson statistics, chaotic ones show level repulsion (Bohigas–Giannoni–Schmit). Eigenstates: the quantum ergodicity theorem of Shnirelman, Zelditch and Colin de Verdière is highlighted as one of the few rigorous statements in the field, asserting equidistribution of a density-one subsequence; scarring is carefully distinguished from the exceptional density-zero set — a scarred eigenfunction generically sits inside the equidistributing subsequence, since the enhancement lives in an $2.9\times10^{-2}$8-scale tube. The quantum unique ergodicity conjecture of Rudnick and Sarnak remains open. Time-dependent observables: the squared commutator, whose growth registers non-commutativity between an early and a late operator.

The billiard discussion includes a pointed observation from Weyl's law: the smooth spectral staircase is fixed by area and perimeter, so a stadium and an ellipse built to match both ($2.9\times10^{-2}$9, $10^3$0) agree through the constant term, and every trace of chaos must live in the fluctuations — which is why unfolding is necessary.

Lecture 4: spectral diagnostics and eigenstate thermalization

The instruments are the nearest-neighbour spacing distribution, the spacing ratio $10^3$1, the long-range number variance $10^3$2 and $10^3$3, the spectral form factor, and the eigenstate thermalization hypothesis (ETH). The derivation of level repulsion from the codimension $10^3$4 of degeneracy is clean: the Dyson index counts real components of an off-diagonal element, and the same number appears three ways — as codimension, as the Vandermonde exponent in the joint eigenvalue density, and as the small-spacing exponent $10^3$5.

The spacing-ratio benchmarks are quoted with their surmise corrections: $10^3$6 (Poisson), $10^3$7 (GOE), $10^3$8 (GUE), $10^3$9 (GSE), each within a percent or less of the large-matrix values. The notes are careful that the surmises are exact for $6.7\times10^{-7}$0 (or $6.7\times10^{-7}$1) matrices and only approximate asymptotically.

The spectral form factor is developed with its slope–dip–ramp–plateau structure, the ramp slope $6.7\times10^{-7}$2 connected to Berry's diagonal approximation of the periodic-orbit sum, and the plateau onset at the Heisenberg time — exactly at $6.7\times10^{-7}$3 only in the unitary class ($6.7\times10^{-7}$4). The computed figures show measured ramp slopes $6.7\times10^{-7}$5, $6.7\times10^{-7}$6, $6.7\times10^{-7}$7 against the exact ratios $6.7\times10^{-7}$8, and plateau heights $6.7\times10^{-7}$9, $10^7$0, $10^7$1, $10^7$2 tracking the window dimension $10^7$3, $10^7$4, $10^7$5, $10^7$6. The form factor matters for holography because the plateau counts states, and black holes have $10^7$7 of them.

ETH is presented with its diagonal and off-diagonal ansätze, the fluctuation–dissipation relation following from a single envelope $10^7$8, and Page's theorem with its exact mean entropy $10^7$9 (deficit half a nat at $\pm\log 2$0, exponentially small for asymmetric cuts). The failure modes are named: integrability (generalized Gibbs ensemble), many-body localization, protected sectors, quantum many-body scars. Subsystem ETH is stated in trace distance, with the qualification that large-$\pm\log 2$1 factorization, protected sectors and finite-volume effects modify it.

The lecture closes with the limitation that motivates Lecture 5, stated plainly: spectral statistics yield a symmetry class and a state count; the Thouless time is the one system-specific scale they carry; and "level repulsion places a system in a chaotic universality class without saying how chaotic it is."

Lecture 5: scrambling, OTOCs, and the chaos bound

Three results anchor this lecture. First, the squared commutator measures operator size exactly: at infinite temperature, averaged over orientations, $\pm\log 2$2, so an OTOC is a census of operator weight at a site. Second, the exponential growth rate extracted from the correlator is not the Lyapunov exponent: by convexity, $\pm\log 2$3, and with large-deviation variance $\pm\log 2$4 the rate is $\pm\log 2$5. Numerically the excess is large — in the kicked rotor the correlator rate exceeds $\pm\log 2$6 at every kicking strength. The bound of Maldacena–Shenker–Stanford constrains this generalized order-two exponent; the typical exponent obeys the tighter $\pm\log 2$7 [Pappalardi–Kurchan]. Third, the window $\pm\log 2$8 with $\pm\log 2$9 requires both fast relaxation and many degrees of freedom — a property of having large $2^k$0, not of being chaotic.

The billiard warning is a strong and somewhat contrarian claim: the quantized stadium, classically chaotic, shows no clean exponential regime in its OTOC, which saturates to a temperature-linear value; the particle in a box grows by an order of magnitude and then returns exactly, with recurrence time $2^k$1 — half what spectral commensurability alone gives, the extra factor coming from the matrix elements. Conversely, exponential growth can occur without chaos: the inverted harmonic oscillator and the integrable Lipkin–Meshkov–Glick model both grow at rates set by a local saddle. The conclusion drawn is that an OTOC is a good diagnostic of operator growth and scrambling, while its relation to the classical $2^k$2 is model- and window-dependent and often invisible in few-body systems.

The holographic mechanism is assembled: near-horizon blueshift $2^k$3 gives $2^k$4 and $2^k$5; the butterfly velocity for the AdS–Schwarzschild brane is $2^k$6 (exactly $2^k$7 in $2^k$8, $2^k$9 in $B^k$0), so holographic chaos spreads strictly below light speed without any bound imposed by hand. The bound itself is proven from analyticity of the regularized correlator $B^k$1 in a strip of width $B^k$2 — the $B^k$3 being the conformal factor of the strip-to-half-plane map and nothing else. The notes insist the bound constrains $B^k$4, not the unregularized correlator an experiment or lattice calculation returns.

Lecture 6: geometry, probes, and the limits of universality

The first half builds what Lecture 5 described: the Kruskal extension of the eternal AdS black hole, the thermofield double state, the Dray–'t Hooft shock-wave solution (a screened Poisson equation for the shift $B^k$5, with screening mass $B^k$6), and the eikonal scattering phase $B^k$7 converting the geometry into the correlator $B^k$8. The prefactor acquires an identity: it is Newton's constant, so at strictly infinite $B^k$9 there is no chaos to see. Stringy corrections enter through $-(\sigma+1+\beta)$0, and the coefficient is half the squared screening mass — one number governing both the shock spread and the drop below the bound. A useful technical point: the $-(\sigma+1+\beta)$1 power in the shock profile means a threshold-fitted butterfly velocity can overshoot or undershoot $-(\sigma+1+\beta)$2 at finite time, with the drift vanishing only asymptotically.

The second half deliberately gives up universality. The key observation is that $-(\sigma+1+\beta)$3 is a surface gravity and nothing besides — it takes the same value on every two-derivative black hole and therefore discriminates nothing. The discriminating quantities belong to probes:

• The dyonic AdS black hole: at fixed electric potential and magnetic charge $-(\sigma+1+\beta)$4, three branches coexist and the Gibbs free energy shows a swallowtail; $-(\sigma+1+\beta)$5 is a van der Waals critical point. The Lyapunov exponent of the unstable circular null geodesic folds at exactly the cusp temperatures, and the gap $-(\sigma+1+\beta)$6 between stable branches at the transition scales as $-(\sigma+1+\beta)$7 — measured as $-(\sigma+1+\beta)$8 over four and a half decades. The notes are candid that $-(\sigma+1+\beta)$9 inherits this exponent as a linear image of the horizon-radius order parameter rather than possessing independent critical information, and that the quasinormal reading of $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$00 (Cardoso et al.) does not apply here because those black holes are asymptotically AdS, not flat.
• The holographic QCD string: in an Einstein–Maxwell–dilaton background with warp factor $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$01, parameters fixed by the deconfinement temperature ($$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$02 GeV$$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$03) and heavy-meson spectrum ($$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$04 GeV$$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$05), a suspended Nambu–Goto string reduced to two degrees of freedom is chaotic on the near-horizon branch and absent on the far-tip branch. The exponent falls by about a third across the parameter grid (from $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$06 GeV to $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$07 GeV), and the anisotropy is the sharp result: at every nonzero magnetic field, the transverse string is less chaotic than the longitudinal one, the two coinciding only at $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$08 where $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$09 is unbroken. The frame dependence is flagged as unresolved and load-bearing — in the Einstein frame the signs differ — so nothing here is a frame-independent statement about QCD. The tension with inverse magnetic catalysis is raised and explicitly not claimed to be resolved.
• The bound comparisons: the notes separate three statements travelling under one name — the MSS theorem (about regularized OTOCs), the classical near-horizon result $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$10 for a held particle (Hashimoto–Tanahashi), and the mere habit of comparing classical exponents to $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$11. The string exponent never violates the bound, peaking at $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$12. Reported geodesic violations are diagnosed as artefacts of free angular momentum or as lying outside the theorem's regime; and the observation that a photon-ring probe saturates $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$13 against its own induced Unruh temperature (for geometries with a Killing–Yano tensor) is presented as an open alternative reading, with no $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$14 yet constructed for the charged-particle cases.

Limitations and open questions

The notes are unusually explicit about what their results do not establish. The main open questions they leave are: whether the quantum unique ergodicity conjecture holds on negatively curved manifolds; whether the string-frame suppression of chaos by $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$15 and $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$16 connects to the deconfinement crossover, given the Einstein-frame sign disagreement and the conjectural status of the link between string chaos and the transition; whether the induced-temperature substitution disposes of the charged-particle bound violations; whether the regularized-bound conclusion transfers to unregularized correlators; and whether an exponential OTOC window can be seen at all in few-body systems, given the stadium's negative answer. Methodological limitations are also stated: the geodesic and normal-mode "Lyapunov exponents" are local instability rates in integrable flows, not long-time averages along generic orbits; the ETH holds only for simple observables within resolved symmetry sectors; and the holographic large-$$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$17 limit hides level repulsion unless finite-$$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$18 effects are in play.

Conclusion

These notes trace a single arc with discipline: classical chaos is defined geometrically on trajectories; quantization removes trajectories and forces the diagnostics into spectra, eigenstates and commutators; spectral tools return a classification and a count but no rate; the squared commutator recovers a rate, which holography fixes at the surface gravity and bounds at $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$19; and the final lecture demonstrates that this universal answer identifies nothing, while probe exponents — tracking a swallowtail phase transition with a measured exponent $$t_{\rm p}\sim \frac{1}{\lambda}\log\!\left(\frac{\Delta}{|\delta X(0)|}\right),$$20, and resolving the direction of a magnetic field in a confining background — restore discriminative power at the cost of leaving the quantum theorem's protection. The recurring methodological insistence, that an exponential in a correlator certifies operator growth rather than a Lyapunov exponent and that the famous bound governs a regularized correlator of a specific exponent, is the notes' most valuable contribution for a reader entering the holographic-chaos literature.