Read a holography paper through a small calculable model

Begin with a result whose parameters you can specify

Open section II of Ryu and Takayanagi’s 2006 paper. Equation (9) gives the distance between two cutoff endpoints in global AdS3. Equation (10) takes the large-cutoff limit and compares with interval entropy on a circle. Record the radius, radial coordinate and interval fraction before substituting. Its hyperbolic radial coordinate differs from the bounded radial angle of Chapter 1.

For a concrete check, choose the paper’s hyperbolic radius $r_0=2$ and angular separation $\Delta\phi=\pi/3$. Equation (9) gives $\cosh(\mathcal L/L)=1+2\sinh^2(r_0)\sin^2(\Delta\phi/2)$. Independently embed both endpoints in a hyperboloid and calculate the distance from their Minkowski inner product. The two methods use identical endpoints.

The Poincaré Lab fixes equal-height cutoff anchors instead. At finite cutoff these differ from boundary-anchored arcs clipped afterward, although both approach the same continuum logarithm. A reproduction should separate the source check, the Lab convention and their common limit. Match endpoint conditions before interpreting a discrepancy as integration error.

Source: Shinsei Ryu and Tadashi Takayanagi

Follow both interval candidates through their crossing

Figure 1 in Headrick’s section 4.2 shows two endpoint pairings. Equations (38)–(42) turn their total lengths into mutual information. Keep equal interval sizes and scan only their gap. Plotting both candidates before selecting the minimum helps reveal mistakes in endpoint ordering, cutoff conventions or branch choice.

Set each interval width to one. At gap $1/4$, the leading continuum result is $I/c=(1/3)\log(16/9)\simeq0.191788$. At gap $1/2$, it vanishes at leading order. Also scan near $d=\sqrt2-1$ to check the candidate crossing. A finite-cutoff geometric crossing shifts slightly and should converge toward this threshold as the cutoff shrinks.

This reproduces one analytic prediction, without solving replica saddles or finding the exact entropy of a CFT at arbitrary central charge. The discussion after equation (42) explains that subleading correlations can survive a vanishing leading mutual information. That observation determines what the zero on the plot means.

Source: Matthew Headrick

Test the code on coherent inputs

Appendix A.1 of Pastawski and colleagues gives the five-qubit stabilizers and logical operators. Construct two orthogonal codewords from their common eigenspace. For every erasure set, trace out the lost sites and apply recovery. Positions around the pentagon label tensor factors; their drawn distances do not enter the quantum channel.

Testing only logical zero and one is insufficient. A channel that discards all coherence can pass both tests. Check $\lvert+\rangle$ and $\lvert+i\rangle$ too, or entangle the logical input with an external reference qubit. Preserving the complete reference–logical joint state tests preservation of arbitrary superpositions and external correlations.

Save complex input amplitudes, erased sites, the retained density matrix and the reconstructed state. Recalculate $F=\langle\psi\rvert\rho_{\rm rec}\lvert\psi\rangle$ from those matrices. Assigning fidelity one from an erasure-count label would skip the recovery calculation. The exhaustive calculation is deterministic and needs no seed; additional sampled inputs need a saved seed and sampling rule.

Source: Fernando Pastawski · Ahmed Almheiri

Make the model fail where it should

For this five-qubit code, any retained pair contains no information about the logical state. Change the input from logical zero to one, then change its phase: the retained state is unchanged. Any recovery device given only that state must return the same output for these different inputs, so it cannot recover all of them. This failure follows directly from the channel.

Now redraw the five nodes at arbitrary positions without changing the encoding. Recovery performance stays identical. Display coordinates therefore do not define a continuous spatial metric. The code supplies isometry, secret sharing and redundant operator representations. It does not supply a local Hamiltonian, distance-dependent CFT correlations or Einstein dynamics. A full HaPPY network requires more tensors and specified contractions; continuous AdS/CFT needs further physical structure.

These failures suggest useful research questions. Add one structure, such as network size or approximate encoding error, define an observable and test whether it approaches a continuous result. Success in a finite example answers the channel question that was actually tested.

Source: Fernando Pastawski · Xi Dong · João Penedones

Match each question to its method

Integrals and causal equations test fixed-background geometry. Matrix identities and recovery errors test a finite encoding. A gravitational correspondence requires structure on both the field-theory and gravity sides. Our figures show calculated model quantities or explicitly labeled schematics; none are detector measurements. Black-hole microstates and gravitational path integrals remain paper-reading topics.

You can now start from a boundary interval, derive a curve, remove a cutoff and compare candidates. You can also encode a logical state, erase sites and reconstruct it. When a new paper changes one step of that chain, identify the step first. A precise question makes its new conclusion easier to connect to a calculation you already understand.