CoChat

Replica wormholes and the black hole interior

Open access
Penington, Geoff, Shenker, Stephen H., Stanford, Douglas, Yang, Zhenbin · 2022 · Journal of High Energy Physics

The DOI matches its Crossref record; 187 of the 187 references it cites were found in a registry or an index; 17 of the 21 sources its claims cite were checked for support.

Abstract

A bstract Recent work has shown how to obtain the Page curve of an evaporating black hole from holographic computations of entanglement entropy. We show how these computations can be justified using the replica trick, from geometries with a spacetime wormhole connecting the different replicas. In a simple model, we study the Page transition in detail by summing replica geometries with different topologies. We compute related quantities in less detail in more complicated models, including JT gravity coupled to conformal matter and the SYK model. Separately, we give a direct gravitational argument for entanglement wedge reconstruction using an explicit formula known as the Petz map; again, a spacetime wormhole plays an important role. We discuss an interpretation of the wormhole geometries as part of some ensemble average implicit in the gravity description.

Overview

CoChat Summary

TL;DR

The paper derives the Page curve and island formula for evaporating black holes from gravitational replica path integrals, showing that Euclidean wormholes connecting different replicas produce the Page transition. It also gives a direct gravitational argument for entanglement wedge reconstruction via the Petz map and interprets the wormhole geometries as part of an implicit ensemble average.

Key findings

  • In a simple JT gravity model with an end-of-the-world brane carrying \(k\) internal states entangled with radiation \(R\), the replica computation gives \(S(R)=\min(\log k, S_{\rm BH})\) from the island conjecture, with the Page transition arising from a competition between disconnected and connected replica topologies. The purity is schematically \(\operatorname{Tr}(\rho_R^2)=k^{-1}+e^{-S_0}\), so the connected wormhole dominates when \(k\gg e^{S_{\rm BH}}\).
  • The gravity path integral yields \(\langle\psi_i|\psi_j\rangle=\delta_{ij}\) but \(|\langle\psi_i|\psi_j\rangle|^2=\delta_{ij}+Z_2/Z_1^2\), which is inconsistent unless the path integral computes an ensemble average; an explicit dual random ensemble is constructed in appendix D.
  • Using the Petz map, the paper shows that after the Page time the connected topology dominates and reconstruction of interior operators from radiation succeeds, while before the Page time the disconnected topology dominates and reconstruction fails. Reconstruction requires \(k\gg d_{\rm code}e^{S_{\rm BH}}\), and the error increases linearly with the code-subspace dimension, implying state dependence.
  • Extensions include JT gravity coupled to conformal matter, where an \(n\to1\) pinwheel saddle reproduces the quantum extremal surface/island answer, and the SYK model, where a replica-nondiagonal wormhole saddle for the \(n=2\) Rényi entropy is found numerically with an action independent of time. A de Sitter toy model shows a Page-like transition from \(\log k\) to \(S_0\).

Methods

  • Euclidean gravitational path integral and replica trick: compute Rényi entropies and purity by summing over replica geometries, including connected Euclidean wormholes, then analytically continue in \(n\).
  • Simple model: JT gravity with an EOW brane behind the horizon, \(k\) orthogonal brane states entangled with auxiliary radiation \(R\); evaluate the full planar sum via a resolvent/Schwinger-Dyson equation inspired by free probability, in microcanonical and canonical ensembles.
  • Additional calculations: JT gravity coupled to conformal matter with two coupled thermofield-double black holes and a pinwheel geometry near \(n=1\); SYK model with two replicas and \(G,\Sigma\) collective fields solved numerically; de Sitter JT gravity with EOW branes; gravitational path-integral evaluation of Petz-map matrix elements.

Limitations

  • Stated: the simple model’s interesting regime is artificial, requiring \(k\) of order \(e^{S_{\rm BH}}\); the radiating black hole analysis is more complicated, involves intrinsically Lorentzian physics and backreaction, and cannot explicitly find the saddle for integer \(n\) Rényi entropies in JT gravity coupled to conformal matter. Integer \(n>1\) saddles are difficult and involve conformal welding; the SYK wormhole solution required a source and exists only above a critical time; the de Sitter discussion is tentative, and section 7 is preliminary.
  • Stated: in the canonical ensemble the resolvent equation was not solved exactly, and parts of the entanglement spectrum near the cutoff are not under analytic control.
  • Evident from the text: wormholes are argued to be only an effective description in non-averaged systems, with no known bulk description of the off-diagonal microscopic effects needed for a systematically improvable calculation.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/jhep03(2022)205

Cites

187 of 187 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Euclidean and Lorentzian geometries for a black hole with an EOW brane behind the horizon.
Figure 1 Euclidean and Lorentzian geometries for a black hole with an EOW brane behind the horizon.From the PDF
Figure 2 We show the expected entanglement wedges for our simple model based on the 'island' conjecture. In the k ≪ e S BH phase, the entanglement wedge of the B system (boundary dual of the gravity theory) is the whole spacetime, shown hatched. In the k ≫ e S BH phase, an island develops. The entanglement wedge of the B system retreats to the exterior of the horizon, and the entanglement wedge of the auxiliary R system is the shaded blue island behind the horizon.
Figure 2 We show the expected entanglement wedges for our simple model based on the 'island' conjecture. In the k ≪ e S BH phase, the entanglement wedge of the B system (boundary dual of the gravity theory) is the whole spacetime, shown hatched. In the k ≫ e S BH phase, an island develops. The entanglement wedge of the B system retreats to the exterior of the horizon, and the entanglement wedge of the auxiliary R system is the shaded blue island behind the horizon.From the PDF
Image 3
From the PDF
Figure 3 The boundary conditions for |〈 ψ i | ψ j 〉| 2 are shown at left, and two ways of filling in the geometry are shown at right. To compute the purity, we want to sum over i, j by connecting the dashed lines together. For the disconnected geometry, this will lead to a single k index loop, and for the connected geometry it will lead to two loops.
Figure 3 The boundary conditions for |〈 ψ i | ψ j 〉| 2 are shown at left, and two ways of filling in the geometry are shown at right. To compute the purity, we want to sum over i, j by connecting the dashed lines together. For the disconnected geometry, this will lead to a single k index loop, and for the connected geometry it will lead to two loops.From the PDF
Figure 4 The boundary conditions for Tr ( ρ n R ) with n = 6 are shown at left, along with two extreme ways of filling in the geometry, corresponding to the completely connected and completely disconnected options. Note that the geometry at right contains a fixed point of the Z n symmetry that rotates the replicas.
Figure 4 The boundary conditions for Tr ( ρ n R ) with n = 6 are shown at left, along with two extreme ways of filling in the geometry, corresponding to the completely connected and completely disconnected options. Note that the geometry at right contains a fixed point of the Z n symmetry that rotates the replicas.From the PDF
Figure 5 The left figure is an example of planar geometry that we need to include in our analysis. The middle figure has an extra handle and is down by e -2 S 0 . The right figure involves a crossing, and is down by k -2 (it has two dashed index loops instead of four).
Figure 5 The left figure is an example of planar geometry that we need to include in our analysis. The middle figure has an extra handle and is down by e -2 S 0 . The right figure involves a crossing, and is down by k -2 (it has two dashed index loops instead of four).From the PDF
Image 7
From the PDF
Image 8
From the PDF
Image 9
From the PDF
Figure 6 Sketch (not to scale) of the entanglement spectrum in the canonical ensemble. Before the Page transition, the density of states has a very narrow distribution. Near the Page transition, it becomes a shifted thermal spectrum (2.51), with a cutoff at some energy. After the Page time, it becomes an ordinary thermal spectrum, again with a cutoff at some energy.
Figure 6 Sketch (not to scale) of the entanglement spectrum in the canonical ensemble. Before the Page transition, the density of states has a very narrow distribution. Near the Page transition, it becomes a shifted thermal spectrum (2.51), with a cutoff at some energy. After the Page time, it becomes an ordinary thermal spectrum, again with a cutoff at some energy.From the PDF
Figure 7 We plot the exact entropy in the planar approximation (solid) and the approximation (2.52) (dashed) for the case β = 3 and large µ . The answer in the microcanonical ensemble with the same entropy is also shown (dotted). The difference between the curves is largest near the Page transition. For small β , the transition in the canonical curves gets smoothed out over a large region of size β -1 / 2 ∼ G -1 / 2 N in log( k ) , while the transition in the microcanonical curve always looks the same. The maximum error between the canonical curve and the simple approximation (2.52) will be of order β ∼ G N .
Figure 7 We plot the exact entropy in the planar approximation (solid) and the approximation (2.52) (dashed) for the case β = 3 and large µ . The answer in the microcanonical ensemble with the same entropy is also shown (dotted). The difference between the curves is largest near the Page transition. For small β , the transition in the canonical curves gets smoothed out over a large region of size β -1 / 2 ∼ G -1 / 2 N in log( k ) , while the transition in the microcanonical curve always looks the same. The maximum error between the canonical curve and the simple approximation (2.52) will be of order β ∼ G N .From the PDF
Figure 8 The two-replica geometry that contributes to the expectation value (3.11) for a = 1 , b = 0 of the Petz Lite operator. The wavy line represents the bulk particle. Note that gluing the right geometry to the left geometry has the same effect (from the perspective of the physical left system) as inserting a creation operator behind the horizon. Another sketch of the glued geometry is shown at right. The marked point is the point of fixed dilaton, and it has conical angle 4 π . The worldline of the particle created in the quantum computer (wiggly orange line) travels through the wormhole and into the physical black hole. The brane on the top half of each of the figures is a brane of type j and the brane on the bottom half is of type i .
Figure 8 The two-replica geometry that contributes to the expectation value (3.11) for a = 1 , b = 0 of the Petz Lite operator. The wavy line represents the bulk particle. Note that gluing the right geometry to the left geometry has the same effect (from the perspective of the physical left system) as inserting a creation operator behind the horizon. Another sketch of the glued geometry is shown at right. The marked point is the point of fixed dilaton, and it has conical angle 4 π . The worldline of the particle created in the quantum computer (wiggly orange line) travels through the wormhole and into the physical black hole. The brane on the top half of each of the figures is a brane of type j and the brane on the bottom half is of type i .From the PDF
Image 13
From the PDF
Image 14
From the PDF
Image 15
From the PDF
Image 16
From the PDF
Figure 10 For β = 3 and large µ , we plot the exact c 1 in the planar approximation (solid) and the simple approximation (3.45) (dashed). The answer in the microcanonical ensemble is also shown (dotted). For smaller β , the transition in the canonical ensemble takes place over a longer interval in log( k ) , of order ∼ β -1 / 2 .
Figure 10 For β = 3 and large µ , we plot the exact c 1 in the planar approximation (solid) and the simple approximation (3.45) (dashed). The answer in the microcanonical ensemble is also shown (dotted). For smaller β , the transition in the canonical ensemble takes place over a longer interval in log( k ) , of order ∼ β -1 / 2 .From the PDF
Image 18
From the PDF
Image 19
From the PDF
Image 20
From the PDF
Figure 11 The black dots on the boundaries correspond to the boundary twist operators, and the black dots in the center are the marked points corresponding to the Z n fi xed points. For large t , the entropy on the two intervals marked with wavy lines is approximately the same as the sum of the entropies on the two intervals taken separately.
Figure 11 The black dots on the boundaries correspond to the boundary twist operators, and the black dots in the center are the marked points corresponding to the Z n fi xed points. For large t , the entropy on the two intervals marked with wavy lines is approximately the same as the sum of the entropies on the two intervals taken separately.From the PDF
Figure 12 In (a) we sketch the boundary time contours for two copies of (4.36), where the long horizontal part is the Lorentzian portion and the 'caps' at the ends are Euclidean. In (b) we insert boundary twist operators (wiggly lines) in order to compute the purity Tr ( ρ 2 1 ) . In (c) we represent these twist operators explicitly by changing the pattern of connection of the contours. We can make a configuration with bounded action as t → ∞ by pasting the solution from the corresponding region of (a) into the shaded regions of (c). The configuration is thermofield-double-like in these two regions.
Figure 12 In (a) we sketch the boundary time contours for two copies of (4.36), where the long horizontal part is the Lorentzian portion and the 'caps' at the ends are Euclidean. In (b) we insert boundary twist operators (wiggly lines) in order to compute the purity Tr ( ρ 2 1 ) . In (c) we represent these twist operators explicitly by changing the pattern of connection of the contours. We can make a configuration with bounded action as t → ∞ by pasting the solution from the corresponding region of (a) into the shaded regions of (c). The configuration is thermofield-double-like in these two regions.From the PDF
Figure 13 In (a) we sketch the boundary time contours for computing 〈 ψ A | ψ A 〉〈 ψ B | ψ B 〉 . In (b) we sketch the contours for computing |〈 ψ A | ψ B 〉| 2 , and in (c) we rearrange them. In the shaded regions, we glue the relevant part of the solution from (a).
Figure 13 In (a) we sketch the boundary time contours for computing 〈 ψ A | ψ A 〉〈 ψ B | ψ B 〉 . In (b) we sketch the contours for computing |〈 ψ A | ψ B 〉| 2 , and in (c) we rearrange them. In the shaded regions, we glue the relevant part of the solution from (a).From the PDF
Figure 14 The time contour C , divided into C 1 and C 2 . Note that τ = β is identified with τ = 0 , so C is closed. One swap operator is inserted at τ = 0 and another is inserted at τ = β/ 2 + it .
Figure 14 The time contour C , divided into C 1 and C 2 . Note that τ = β is identified with τ = 0 , so C is closed. One swap operator is inserted at τ = 0 and another is inserted at τ = β/ 2 + it .From the PDF
Figure 15 Numerical solutions to (5.11). The heatmaps at left represent G 1 ( t, t ′ ) and G 2 ( t, t ′ ) for two different types of solution discussed in the main text. We plot at infinite temperature so the solutions are real. The stripe along the main diagonal is somewhat trivial, representing large correlation between adjacent points along the contour. The other features in the heatmap represent nontrivial correlations. Contours of system 1 are interacting with the contours of system 2 with the same color due to the swap operator. The contours are drawn so that nearby points have large correlation in the solution. For this plot, Jt = 20 , βJ = 0 , ˆ J = J/ 2 , q = 4 , ˆ q = 2 .
Figure 15 Numerical solutions to (5.11). The heatmaps at left represent G 1 ( t, t ′ ) and G 2 ( t, t ′ ) for two different types of solution discussed in the main text. We plot at infinite temperature so the solutions are real. The stripe along the main diagonal is somewhat trivial, representing large correlation between adjacent points along the contour. The other features in the heatmap represent nontrivial correlations. Contours of system 1 are interacting with the contours of system 2 with the same color due to the swap operator. The contours are drawn so that nearby points have large correlation in the solution. For this plot, Jt = 20 , βJ = 0 , ˆ J = J/ 2 , q = 4 , ˆ q = 2 .From the PDF
Figure 16 The action of the replica-diagonal solution (dashed) and the wormhole solution (solid) as a function of t . The left plot is in the canonical ensemble at βJ = 2 . 4 . The right plot is in the microcanonical ensemble at the corresponding energy, E/ ( NJ ) = -0 . 03 per system. The black dots are data, and the curves are to guide the eye. The parameters were J = 1 , ˆ J = J/ 2 , q = 4 and ˆ q = 2 .
Figure 16 The action of the replica-diagonal solution (dashed) and the wormhole solution (solid) as a function of t . The left plot is in the canonical ensemble at βJ = 2 . 4 . The right plot is in the microcanonical ensemble at the corresponding energy, E/ ( NJ ) = -0 . 03 per system. The black dots are data, and the curves are to guide the eye. The parameters were J = 1 , ˆ J = J/ 2 , q = 4 and ˆ q = 2 .From the PDF
Image 27
From the PDF
Image 28
From the PDF
Figure 17 An example of a replicated geometry for two intervals in AdS, cut open.
Figure 17 An example of a replicated geometry for two intervals in AdS, cut open.From the PDF
Figure 18 A simple random tensor network, with a single random tensor | T 〉 and three bulk legs. One bulk leg flows into | T 〉 , while the other two are mapped directly into the boundary Hilbert spaces B and ¯ B .
Figure 18 A simple random tensor network, with a single random tensor | T 〉 and three bulk legs. One bulk leg flows into | T 〉 , while the other two are mapped directly into the boundary Hilbert spaces B and ¯ B .From the PDF
Figure 19 A plot of β log | δS | against β s k in the limit where µ ≫ 1 /β and β → 0 . Near the transition, the entropy correction decays as δS = -O (exp( -β ( s k -s (1) ) 2 / 2)) because the largest correction comes from energy fluctuations causing the transition to happen earlier or later than expected. Far from the transition, the entropy decays as δS = -exp( -O ( | s k -s (1) | )) , as for the microcanonical ensemble. Note that some important features of the correction, such as the maximal size δS = -O ( √ 1 /β ) , are not visible in this plot.
Figure 19 A plot of β log | δS | against β s k in the limit where µ ≫ 1 /β and β → 0 . Near the transition, the entropy correction decays as δS = -O (exp( -β ( s k -s (1) ) 2 / 2)) because the largest correction comes from energy fluctuations causing the transition to happen earlier or later than expected. Far from the transition, the entropy decays as δS = -exp( -O ( | s k -s (1) | )) , as for the microcanonical ensemble. Note that some important features of the correction, such as the maximal size δS = -O ( √ 1 /β ) , are not visible in this plot.From the PDF

Verification

Claims the paper makes

12 Valid · 1 Not valid · 0 Not checked · 12 Own results and interpretations

20 more claims · 8 with their cited sources checked

Create a free account to see all 25 claims and every source verdict.

See all claims — free account

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

4 of 4

Citations

Cites

Show all 187 references

Cited by

Metadata

Authors
Penington, G., Shenker, S. H., Stanford, D., & Yang, Z.
Affiliations
Stanford University (US)
Corresponding author
Yang, Zhenbin
Venue
Journal of High Energy Physics
Published
2022-03
Volume
2022
Issue
3
ISSN
1029-8479 (ISSN-L)
OpenAlex
W2991561483
Cited by
807
References
134
Citations by year
2026: 62 · 2025: 139 · 2024: 127 · 2023: 148 · 2022: 131
FWCI
17.02
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of High Energy Physics (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • arXiv (Cornell University)
Funders
  • National Science Foundation (1720397)
  • U.S. Department of Energy (DESC0019380)
  • Simons Foundation
  • Air Force Office of Scientific Research (FA9550-, FA9550-16-1-0082, FA9550-16, -16-1-, FA9550-16-1)
Topics
  • Physical Sciences › Physics and Astronomy › Nuclear and High Energy Physics › Black Holes and Theoretical Physics
  • Physical Sciences › Physics and Astronomy › Nuclear and High Energy Physics › Noncommutative and Quantum Gravity Theories
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Quantum many-body systems
Keywords
replica wormholes, black hole evaporation, entanglement entropy, replica method, entanglement wedge reconstruction, JT gravity, conformal matter, SYK model, holography, black hole interior, ensemble average

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-09-27 (prompt extract-v2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.