Typicality versus thermality: An analytic distinction

The core of this paper is a short, correct calculation: if you average a local correlator over the ensemble of all normalized superpositions of e^S energy eigenstates, the variance is suppressed by e^−S relative to the variance over the basis states (Eq. 2.29). That follows directly from the uniform measure (Eq. 2.27) and is essentially the standard typicality story — Page, Popescu-Short-Winter — transplanted to black hole microstates. The genuinely new part is the application: the claim that individual microstates are indistinguishable from the thermal average in real time below the recurrence time, and that the D1-D5 crossover in imaginary time sits at τ ~ S² (Eq. 4.12). The free-boson section is the most convincing piece of work in the paper; the bounds in Eq. 3.23 are explicit and the F(n) > F̃(n) > 0 argument, while a bit quick, is plausible and checkable. The D1-D5 variance (Eq. 4.8) is written out and the scaling analysis is honest about which terms dominate in which regime. Credit where due: this is careful, and the authors do not hide the awkward parts — Sec. 5 and the free-boson summary both say plainly that at the imaginary time where the variance grows, the mean correlator grows too, so the separation is in-principle rather than operational. That is the right thing to say, and they say it. Two soft spots. First, the Haar measure is a modeling choice, not derived from the CFT. The physical claim needs black hole microstates to be typical draws from Msup, and the paper supports that with a citation to [22] plus the observation that generic states are random superpositions. The free-boson example shows how much rides on this: in Mbas the ratio is large for τ > β/2 (Eq. 3.25), in Msup it is crushed by e^−S (Eq. 3.26). If the fuzzball geometries are Mbas elements, the suppression does not apply. Second, the abstract's 'thereby distinguishing pure states from the thermal density matrix' is looser than the body. What actually distinguishes pure from thermal is imaginary-time periodicity — thermal correlators diverge at τ = β, microstates do not — and that is a much simpler and more robust point than variance amplification. The variance story does not obviously help, since everything grows in imaginary time. The D1-D5 computation is done canonically with the e^−S factor appended at the end; footnote 8 argues fractionation keeps the canonical variance under control, but that is an argument, not a check against fixed total twist. Worth a serious referee. The general result is right and the conceptual point — that the canonical CFT ensemble is not dual to a single eternal black hole — is worth putting in front of people who work on fuzzballs and information. I would cite it for the e^−S statement and the S² timescale, not for the abstract's framing.

The central result, Eq. (2.29), follows from the measure (2.27), which is the Haar measure on the sphere of superposition coefficients c_s. The physical claim — that black hole microstates are drawn from this ensemble — is supported by the assertion that 'generic foam microstates ... correspond to quantum superpositions of the chiral primary operators' [22]. But a superposition with fixed coefficients is a single state, not a Haar-random draw; the e^{-S} suppression requires averaging over the measure, and the paper does not derive the measure from the CFT dynamics. The paper's own free boson example makes the ensemble choice load-bearing: in Mbas the ratio sqrt(var)/<O> is large for tau > beta/2 (Eq. 3.25), while in Msup it is suppressed by e^{-S} (Eq. 3.26). If the physical microstates are Mbas elements (as the fuzzball program constructs specific geometries), the suppression does not apply and the crossover scale changes. The D1-D5 computation (Eq. 4.8) is done in the canonical ensemble of all twists, with the e^{-S} factor appended at the end (Eq. 4.12); the paper argues the canonical variance is under control due to fractionation (footnote 8), but this is not verified against the exact microcanonical ensemble of fixed total twist N. The abstract's 'thereby distinguishing' [abstract:s3] is also slightly stronger than the body, which shows the pure-vs-thermal distinction is via imaginary-time periodicity, not via variance amplification.

The paper's general result (Eq. 2.29) is a correct computation given the Haar measure (2.27), and the imaginary-time amplification of the variance is a real feature of the correlators. The two examples are worked carefully, and the D1-D5 crossover scale tau ~ S^2 (Eq. 4.12) is an explicit, checkable prediction. The load-bearing soft spot is the identification of physical microstates with typical elements of Msup, which is asserted via [22] rather than derived from the CFT. The free boson example shows the ensemble choice matters: Mbas gives large variance (Eq. 3.25), Msup suppressed (Eq. 3.26). The D1-D5 result uses the canonical ensemble, which the paper argues is under control due to fractionation, but this is not verified against the exact microcanonical ensemble. The abstract's 'thereby distinguishing' is slightly stronger than the body, which shows the pure-vs-thermal distinction is via periodicity, not variance amplification. These are modeling and presentation concerns, not errors in the computation. The paper is honest about the operational difficulty (Sec. 5). CONDITIONAL remains the appropriate verdict; the reader's identification of the Haar measure as the weakest assumption is correct, and the concern does not overturn the central computation but does limit its physical reach.

If correct, the result is a clean and general statement about the typicality of black-hole microstates: the e^{-S} suppression of correlator variance is derived from the measure alone and does not depend on coupling, and the paper is careful to distinguish the basis ensemble Mbas from the superposition ensemble Msup (Appendix C), which is a genuine strength and is what makes the ensemble dependence of the conclusion visible rather than hidden. The two worked examples are explicit and checkable, and the imaginary-time periodicity argument gives a sharp, falsifiable statement about how a single microstate differs from the thermal density matrix. The paper does not fit parameters to data and does not rely on unpublished results; the correlator in Sec. 4 is quoted from the published [17], which is legitimate input.

The paper asks a sharp question: if a black hole has $e^{S}$ microstates, how different are the answers that a local probe operator gets in a single microstate from the answer it gets in the thermal ensemble, and from the answers in other microstates? The authors argue, from statistics alone rather than from any specific dynamics, that the variance of any local correlation function across the ensemble of random superpositions of energy eigenstates is suppressed by a factor $e^{-S}$ relative to the variance across the basis states themselves. That suppression would make microstates practically indistinguishable to a semiclassical observer in real time, which is what one wants for the usual black hole geometry to be a good effective description. Their way out is analytic: correlation functions grow exponentially as time is continued to imaginary values, so the variance, though tiny, is amplified there, and the difference between a pure microstate and a thermal density matrix becomes visible. They check this in two models, a free chiral boson on a circle and the D1-D5 conformal field theory dual to the massless BTZ black hole, and locate the imaginary-time scale at which the distinction appears.

What carries the argument. The load-bearing object is the ensemble $M_{sup}$ of normalized random superpositions $|ps> = sum_s c_s |s>$ of energy eigenstates in a window of width $Delta E$, together with the uniform measure on the coefficients $c_s$. Averaging $|c_s|^2$ and $|c_s|^2 |c_t|^2$ over that measure (their Eq. 2.27) turns the generating functional $Z[J] = <ps| e^{int J O}|ps>$ into an average over basis states plus a term suppressed by $e^{-S}$, and that suppression propagates to the variance of every $k$-point function. The second piece of machinery is the analytic continuation $t -> -i tau$: the same mode sums that oscillate in real time become hyperbolic cosines $cosh(n tau)$ in imaginary time, so both the mean correlator and its spread grow exponentially, and the ratio of spread to mean is what the authors track to find when microstates become distinguishable. The two worked examples supply the mode sums: oscillator occupancies $N_n$ constrained by $sum n N_n = E$ for the free boson, and twist numbers $N_{n mu}$ with $sum n (N_{n mu} + N'_{n mu}) = N$ for the D1-D5 orbifold, the latter with fractional frequencies that mimic the dense spectrum of a black hole.

Core claim. The central claim is that the variance of a local correlator taken over the ensemble of all normalized superpositions of the $e^{S}$ energy eigenstates in a microcanonical window is smaller than the variance over the basis states by a factor of order $e^{-S}$ (their Eq. 2.29). The authors derive this from the uniform measure on the superposition coefficients, which gives $|c_s|^2$ averaging to $1/e^{S}$, and they note that the only escape from the suppression is a probe operator whose correlators are themselves exponentially large. Real-time correlators are not, but analytically continued ones are, because oscillatory factors $e^{-i n t}$ become growing factors $e^{n t}$ at imaginary time. In the free chiral boson the standard deviation to mean ratio exceeds one for imaginary time greater than about $3b/2$ (with $b$ the inverse temperature of the comparison thermal state), and the thermal correlator's periodicity in imaginary time, with its divergence at $t = b$, is simply absent in any individual microstate. In the D1-D5 system the corresponding crossover is pushed out to imaginary time of order $S^2$. The authors read this as evidence that a fixed-energy ensemble in a CFT is not dual to a single eternal black hole geometry, but to an ensemble of microstate spacetimes that a coarse-grained observer would still describe as the black hole.

The premise it rests on. The whole $e^{-S}$ suppression rests on the assumption that the microstates of a black hole are uniformly random superpositions of energy eigenstates, drawn with equal weight over the sphere of coefficients; the paper justifies this by typicality rather than deriving it from the dynamics of the dual field theory.

A semiclassical observer restricted to real-time, coarse-grained measurements cannot distinguish individual black hole microstates from the thermal ensemble before times of order the Poincare recurrence time $e^{S}$.

The microcanonical and canonical ensembles of a CFT are not dual to a single eternal black hole, since correlators in one asymptotic region of that geometry are periodic in imaginary time while correlators in any individual microstate are not.

The imaginary-time scale at which microstates separate from the thermal average is model dependent: about $3b/2$ for the free chiral boson and of order $S^2$ for the D1-D5 system.

Any proposal to distinguish microstates with local probes must either find operators with exponentially large correlators or work in the complex time plane, since the variance itself is suppressed by $e^{-S}$.

Horizon-free, singularity-free microstate geometries remain viable as an underlying description, because their departures from the black hole geometry are invisible to coarse-grained measurements.

The argument is purely kinematic: it uses the dimension of the energy window and the analytic structure of correlators, not the dynamics of the theory. That suggests the same $e^{-S}$ suppression should hold for any probe whose spectral function is bounded, which makes the imaginary-time escape route look like the only generic one.

If one could prepare and measure a single microstate, the natural experimental signature would be the failure of the fluctuation-dissipation relation at imaginary time, since the thermal answer is fixed by periodicity in $b$ while the microstate answer is not.

The crossover scales ($3b/2$ versus $S^2$) differ because the D1-D5 spectrum has fractional spacing, so the density of contributing modes near the relevant frequency is much larger. This hints that the imaginary-time scale is set by the density of states at the probe frequency rather than by the total entropy alone.

Nonlocal probes such as Wilson loops, which the authors mention only in passing, sum infinitely many local correlators and could in principle evade the suppression; testing whether their variance scales as $e^{-S}$ or more weakly would be a sharp check on how general the typicality argument is.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review