Replica Wormholes

A replica wormhole is a connected gravitating bulk whose asymptotic or bath boundary conditions still describe several replicas. The topology change joins nominally separate gravitational interiors and thereby contributes a connected term to the replicated moment. Its importance is decided by the renormalized action, topology weight, determinants, zero modes, contour, and matter correlations—not by connectedness alone.

Required background. Quantum Information supplies cyclic moments and the distinction between correlated copies and an averaged observable.

Helpful background. Semiclassical Gravity supplies Euclidean actions and generalized entropy; Quantum Fields in Curved Spacetime supplies matter determinants and replica boundary conditions; Black-Hole Information supplies the radiation entropy and QES problem.

The topology that joins replicas

For $n$ copies of an evaporating state, the bath cut along $R$ is glued cyclically as required by $\operatorname{Tr}\rho_R^n$. In the replica-disconnected gravitational saddle, each copy has its own bulk component. A replica wormhole instead uses one bulk manifold with passages joining the gravitating regions of different copies. The asymptotic boundary data do not merge; only the integrated interior topology changes.

In a replica-symmetric example, the cyclic group permutes the sheets. The quotient has a codimension-two fixed locus, and lifting it reconstructs the connected $n$-sheet geometry. Near $n=1$ that locus becomes the QES bounding an island. The explicit construction in JT gravity coupled to baths is developed in Almheiri et al. 2020, §§2–5, with complementary models in Penington, Shenker, Stanford, and Yang 2022, §§2–5.

Calling the connection a “wormhole” does not imply a traversable Lorentzian tunnel. It is primarily a Euclidean or complex replica saddle. Its Lorentzian interpretation, if any, requires a separate continuation.

Action, topology weights, and zero modes

In JT gravity the topological term contributes a factor governed by the Euler characteristic,

$$I_{\mathrm{top}}=-S_0\chi, \qquad \chi=2-2g-b, \qquad e^{-I_{\mathrm{top}}}=e^{S_0\chi},$$

for an orientable surface of genus $g$ with $b$ asymptotic boundaries. $S_0$ is the extremal entropy. This formula assumes those are all boundary components; end-of-the-world branes and branch points require their own topology accounting. A topology with additional handles is suppressed by powers of $e^{-2S_0}$ before moduli volumes and matter factors are included. The systematic genus expansion and its matrix-integral organization are derived in Saad, Shenker, and Stanford 2019, §§2–5.

For $n$ disk-like boundaries, $n$ disconnected disks have total $\chi=n$, whereas a connected genus-zero surface with $n$ boundaries has $\chi=2-n$. The bare topological ratio is therefore

$$\frac{W_{\mathrm{conn}}}{W_{\mathrm{disc}}} \sim e^{-2S_0(n-1)}.$$

This is a topology-counting example for $n$ ordinary asymptotic boundaries, not a universal weight for every replica wormhole. In a branched cover with $m$ fully ramified fixed points, the corresponding Euler-characteristic relation is $\chi(M_n)=n\chi(\widehat M_n)-m(n-1)$ when there are no further branch or boundary contributions. One must first identify the actual replica topology. The dilaton action, matter partition function, normalization by $Z_1^n$, and time-dependent boundary data then determine the full competition.

Zero modes must be divided by gauge volume and integrated with the correct measure. Nonzero fluctuations produce determinants, while negative modes and complex phases bear on whether the saddle lies on the contour. A classical on-shell action without these factors is reliable only when its separation is parametrically larger than the omitted corrections.

Dominance and the island limit

Let $I_{\mathrm{wh}}(n)$ and $I_{\mathrm{disc}}(n)$ be the complete effective actions. At integer $n\gt1$, a schematic normalized moment contains

$$Z_n^{\mathrm{norm}} \simeq \frac{e^{-I_{\mathrm{disc}}(n)}+e^{-I_{\mathrm{wh}}(n)}+\cdots} {Z_1^n}.$$

One sums the distinct allowed configurations before taking the logarithm. The displayed expression is not a uniform two-term formula at $n=1$: the two families can reduce to the same one-copy geometry. Substituting two identical unit-weight terms there would incorrectly give $Z_1^{\mathrm{norm}}=2$. Correct continuation must retain the normalization, multiplicities, and additional contributions that prevent this double counting.

In the leading semiclassical continuation, a branch satisfies $I_\alpha(n)-nI_1=(n-1)S_{\mathrm{gen},\alpha}+O((n-1)^2)$. For $n\gt1$ the smaller generalized entropy therefore has the larger exponential weight. The wormhole branch’s fixed-locus derivative supplies the geometric term and its matter gluing supplies $S_{\mathrm{bulk}}(R\cup I)$, as demonstrated in Almheiri et al. 2020, §2.1 and §3.5, Eq. (3.35). This explains the leading minimum while leaving finite-entropy smoothing and the order of limits to the full replica calculation.

The ordering matters near a transition. Differentiating only the disconnected saddle forever reproduces the monotonically growing Hawking branch. Differentiating only the wormhole saddle at early times can select a branch that is present but subdominant.

Results and open questions as of 2026-08-19

This discussion reflects results available through 2026-08-19.

Evidence or result	Model and regime	Limitation or missing step
Replica wormhole Page transition, Almheiri et al. 2020, §§2–6	JT gravity with specified baths, matter states, and semiclassical saddle expansion	Does not establish the same topology or contour in generic four-dimensional gravity
Interior replica models, Penington et al. 2022, §§2–5	Solvable gravitational and random tensor-network constructions	Microscopic interpretation depends on the model and code subspace
Topology expansion and matrix integral, Saad, Shenker, and Stanford 2019, §§2–5	JT gravity’s perturbative genus expansion and a matrix-integral completion	Naturally ensemble-valued; a chosen nonperturbative completion is extra input
Entropy synthesis, Almheiri et al. 2021, §10	Semiclassical replica method across representative evaporation models	Analytic continuation and gravitational contour remain formulation dependent

One interpretation treats replica wormholes as ordinary semiclassical saddles of quantum gravity whose role should persist in a single theory. Another treats their connected correlations as evidence that the path integral computes an ensemble average. An operational interpretation instead emphasizes permutation observables across repeated experiments. These competing positions agree on the controlled saddle arithmetic but differ on factorization and the Hilbert space to which the result belongs.

Which interpretation applies to a generic evaporating black hole remains unresolved; the evidence does not establish a universal verdict beyond the stated semiclassical models.

The adverse check is to restrict the topology sum to disconnected interiors: all boundary replica gluing remains intact, yet the wormhole contribution vanishes. A second check compares the topological suppression above with the time-dependent matter enhancement; if the latter never exceeds it in the declared regime, claiming wormhole dominance is inconsistent.

The conclusions could change after a contour calculation fixes a wormhole’s thimble coefficient, after a higher-dimensional single-theory model reproduces the topology and Page transition, or after a symmetry-breaking or lower-action saddle displaces the replica-symmetric wormhole near $n=1$.

Exercises

For two disk boundaries in JT gravity, compare the topological weights of two disconnected disks and one connected genus-zero cylinder. If matter and boundary terms favor the cylinder by $e^{3S_0}$, what is the net ratio before determinants?

Solution

Two disks have total Euler characteristic $2$, while the cylinder has $\chi=0$. The topological ratio is therefore $e^{-2S_0}$. Multiplying by the assumed matter-and-boundary enhancement $e^{3S_0}$ gives a net connected-to-disconnected ratio $e^{S_0}$ before determinants and contour coefficients. The example shows why topology suppression alone does not decide dominance, while also making clear which additional factor reversed it.

References

• Almheiri et al. 2020, Ahmed Almheiri, Thomas Hartman, Juan Maldacena, Edgar Shaghoulian, and Amirhossein Tajdini, “Replica Wormholes and the Entropy of Hawking Radiation,” Journal of High Energy Physics 05 (2020) 013.
• Almheiri et al. 2021, Ahmed Almheiri, Thomas Hartman, Juan Maldacena, Edgar Shaghoulian, and Amirhossein Tajdini, “The Entropy of Hawking Radiation,” Reviews of Modern Physics 93 (2021) 035002.
• Penington et al. 2022, Geoffrey Penington, Stephen H. Shenker, Douglas Stanford, and Zhenbin Yang, “Replica Wormholes and the Black Hole Interior,” Journal of High Energy Physics 09 (2022) 002.
• Saad, Shenker, and Stanford 2019, Phil Saad, Stephen H. Shenker, and Douglas Stanford, “JT Gravity as a Matrix Integral,” arXiv:1903.11115 (2019).