Tensor-Network Models of Holographic Entanglement Geometry

Holographic tensor networks can realize RT-like minimal cuts, redundant bulk-to-boundary encoding, and sharp discrete reconstruction regions in exactly solvable finite-dimensional models. They do not derive the dynamics, diffeomorphism constraints, continuum operator algebra, or finite-$N$ corrections of AdS gravity. The right use is comparative: calculate what a specified network proves, then separately identify which geometric features it models. Here every leg has dimension $q$, logarithms are natural, and “area” means the number of cut bonds times $\log q$.

Required background. RT supplies the geometric target, while tensor-network entanglement ansätze supplies isometries, bond dimensions, and contraction rules.

Helpful background. Operator-algebra QEC states what reconstruction means, and bit threads supplies the distinct continuum flow duality.

Perfect tensors and a one-node code

A rank-$2m$ perfect tensor $T_{i_1\cdots i_{2m}}$ is proportional to an isometry from any set of at most $m$ legs to its complement. Equivalently, the normalized $2m$-party state defined by its components is absolutely maximally entangled: every subset of at most $m$ legs is maximally mixed.

Take a six-leg perfect tensor. Designate one leg $R$ as a reference purifying a logical bulk input and five legs $B_1,\ldots,B_5$ as boundary outputs. The normalized Choi state $|T\rangle_{RB_1\cdots B_5}$ obeys

$$\rho_X=\frac{I_X}{q^{|X|}}, \qquad |X|\leq3.$$

For a boundary region $A$ containing $k$ of the five outputs, purity of the six-leg state gives

$$S(A)= \begin{cases} k\log q,&k\leq3,\\ (6-k)\log q,&k\geq3. \end{cases}$$

At $k=3$, both expressions give $3\log q$. This equals the minimal number of external tensor legs that must be cut to separate $A$ from $A^cR$:

$$S(A)=|\gamma_A|\log q.$$

The equality is exact for this model. It resembles RT, with $\log q$ playing the bond-area unit, but it has no ultraviolet divergence, Newton coupling, or continuum extremal-surface equation.

Reconstruction region in the same example

Regard the tensor as an encoding isometry

$$V:\mathcal H_{\rm logical}\longrightarrow \bigotimes_{i=1}^{5}\mathcal H_{B_i}.$$

Perfectness implies that erasure of any two boundary legs is correctable. Therefore every set of three boundary legs can reconstruct any logical operator $O_L$: there exists $O_A$ such that

$$O_AV|\psi\rangle=VO_L|\psi\rangle$$

for every logical state $|\psi\rangle$. No generic two-leg region can do so, because its three-leg complement is also large enough to carry the logical information; simultaneous exact reconstruction on two disjoint unauthorized regions would violate no-cloning.

The discrete “entanglement wedge” therefore jumps when $k$ reaches three. The same minimal-cut transition that changes $|\gamma_A|$ changes which side of the cut contains the logical leg. This is the requested small-network application: boundary entropies and reconstruction regions are both calculated directly from one perfect tensor.

Gluing perfect tensors into a negatively curved tiling gives the HaPPY code. Greedy absorption of tensors identifies a discrete reconstruction wedge and minimal cuts reproduce entropy when bulk legs are in suitable states (Pastawski et al. 2015, §§2–4). Random tensor networks reproduce RT-like entropies with controlled large-bond-dimension corrections in a different ensemble (Hayden et al. 2016, §§2–3). These are model results with different assumptions.

What bit threads add—and do not add

A network cut is a discrete min cut. Assigning capacities $\log q_e$ to bonds permits a graph max-flow/min-cut dual, which resembles bit threads. In the continuum theorem, however, the capacity is a local norm bound derived from Riemannian area and flows are divergenceless vector fields. A tensor contraction graph has finite vertices and prescribed bond capacities. The shared optimization structure does not identify individual graph paths with gravitational flux lines or microscopic entanglement pairs.

Multiflows on a graph can prove entropy inequalities just as continuum multiflows do, but the transfer requires the graph to reproduce the relevant RT cut function. Tensor-network entropies can also have stabilizer-specific constraints absent in a generic holographic CFT.

Three adversarial deformations

Break the isometry. Let $T\to T+\varepsilon\Delta T$ with a generic perturbation. Then for a nominal input set $X$,

$$T_X^\dagger T_X=I+\varepsilon K_X+O(\varepsilon^2).$$

Reduced states are no longer exactly maximally mixed, the cut count is no longer the full entropy, and erasure recovery acquires an error controlled by $K_X$. The qualitative geometry may survive approximately, but exact RT and exact reconstruction do not.

Add gauge constraints. Gauge-invariant Hilbert spaces do not factor across a cut without specifying centers or edge modes. Bond counting omits flux-sector Shannon terms and representation-dependent edge contributions unless the network is enlarged to include them. A perfect tensor on factorized legs does not solve the gravitational Gauss-law problem.

Demand continuum finite-N locality. A finite network has a finite-dimensional algebra and a built-in lattice scale. Continuum local QFT algebras are type III, while gravitationally dressed observables have boundary tails. Refining the tiling or increasing $q$ is not by itself a proof of the required continuum limit, CFT spectrum, bulk equations, or nonperturbative precision.

The strongest surviving conclusion is that tensor networks furnish explicit quantum codes whose entropy and recovery geometry model selected kinematic features of holography. Their failures under these deformations are information about the model assumptions, not paradoxes for AdS/CFT.

Handoff

Wedge nesting derives inclusion from continuum extremal-surface hypotheses. Entropy cones explains which graph-cut results do transfer to the leading static holographic entropy vector. Full entanglement-wedge reconstruction belongs to the next chapter and retains code-subspace, algebra, and error data.

The chapter overview contains the structure diagram and validity and failure diagram. They are embedded there once so that their shared chapter-level context is not repeated on every article.

For the chapter-wide comparison of assumptions, counterevidence, falsifiers, and claim ceilings, see the claim-domain table.

References

• Hayden, P., Nezami, S., Qi, X.-L., Thomas, N., Walter, M., and Yang, Z. (2016). “Holographic duality from random tensor networks.” Journal of High Energy Physics 2016(11), 009. DOI.
• Pastawski, F., Yoshida, B., Harlow, D., and Preskill, J. (2015). “Holographic quantum error-correcting codes: Toy models for the bulk/boundary correspondence.” Journal of High Energy Physics 2015(6), 149. DOI.