RT-like Entropy Decomposition in Holography

• RT-like entropy decomposition is a framework that reformulates entanglement measures as a dominant geometric or algebraic term plus systematic corrections.
• It utilizes min-cut graph models, tensor-network representations, and operator-algebra techniques to capture holographic, statistical, and microscopic structures.
• This approach clarifies how constraints like homology, modular data, and error-correcting codes combine to yield entropy formulas consistent with strong subadditivity and quantum corrections.

Ryu–Takayanagi-like entropy decomposition denotes a family of constructions in which entanglement entropy, Rényi entropy, or closely related measures are rewritten as a geometric, combinatorial, or algebraic object that plays the role of the Ryu–Takayanagi (RT) area term, together with additional sector, bulk, or finite-size contributions. In the static holographic setting the prototype is

$S(A)=\frac{\mathrm{area}(\tilde A)}{4G_N},$

with $\tilde A$ a minimal bulk surface homologous to $A$; the literature extends this structure to min-cut graph models, domain-wall partition functions in random tensor networks, operator-algebraic decompositions, Virasoro modular saddles, and analytically continued Lorentzian saddles (Bao et al., 2016, Chirco et al., 2017, Kamal et al., 2019, Nunez et al., 23 Jul 2025).

1. Geometric prototype and the meaning of “decomposition”

In AdS$_3$/CFT$_2$, the RT prescription assigns to a boundary interval or union of intervals a bulk minimal-length spacelike geodesic or union of geodesics. For a single interval $A=[a,b]$, $\tilde A$ is a single geodesic anchored at $a,b$; for a union of disjoint intervals,

$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$

the RT surface is a union

$\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$

with the homology condition and minimality selecting the admissible pairing pattern among candidate boundary-anchored geodesics (Bao et al., 2016).

The phrase “entropy decomposition” refers first to the fact that the RT functional can be cut into local geometric pieces without losing its variational meaning. In the static proof of strong subadditivity, one introduces bulk regions $\tilde A$0 with $\tilde A$1, forms $\tilde A$2 and $\tilde A$3, and obtains the exact cut-and-paste identity

$\tilde A$4

Area additivity on the right-hand side then implies

$\tilde A$5

so the entropy inequality is literally a rearrangement of bulk geometric pieces (Callan et al., 2012).

A common misconception is that RT-like decomposition means an unconstrained additive partition of entropy into independent local terms. The constructions surveyed here are more rigid. Homology, minimality or extremality, and in many cases a shared graph, algebra, or worldsheet, determine which local pieces are admissible. The decomposition is therefore structured rather than arbitrary.

2. Min-cut, graph models, and entropy cones

A particularly explicit RT-like decomposition arises in AdS$\tilde A$6/CFT$\tilde A$7 from a max-flow/min-cut reformulation. One constructs all admissible boundary-anchored geodesics $\tilde A$8 whose subtended intervals contain zero or an even number of endpoints, partitions the spatial slice $\tilde A$9 into regions $A$0, and defines a weighted graph $A$1 with vertices $A$2 and edge weights

$A$3

After merging boundary-adjacent vertices into $A$4 and $A$5, the minimal-weight $A$6-cut separating them is exactly the RT surface:

$A$7

The continuous area functional is thus discretized into a combinatorial sum over geodesic segments. For $A$8 boundary intervals, the resulting algorithm runs in polynomial time, with explicit scaling $A$9 at fixed numerical precision (Bao et al., 2016).

The same min-cut logic generalizes from a single entropy to whole entropy vectors. For $_3$0 disjoint boundary regions $_3$1, one collects

$_3$2

forming a holographic entropy vector. The set of all such vectors defines the holographic entropy cone $_3$3, which is a convex polyhedral cone. A graph model realizes each entropy as

$_3$4

so the cone is characterized by linear inequalities obeyed by all such cut functions. For $_3$5, strong subadditivity and monogamy of mutual information are complete; for $_3$6, there is an infinite family of new cyclic inequalities. The graph formulation makes the decomposition aspect manifest: different entropies are different minimal cuts on a common weighted network, and the cone facets encode the admissible cutting-and-gluing relations among them (Bao et al., 2015).

This shared-graph viewpoint is also the natural setting for “common entanglement” language. When different minimal cuts reuse the same weighted edges, the overlap records shared contributions; when cuts differ, the mismatch records conditional or exclusive contributions. The graph itself, not a single surface, is the organizing object.

3. Tensor-network, domain-wall, and group-field-theoretic realizations

In symmetric random tensor networks, each tensor satisfies the local gauge-like constraint

$_3$7

so the effective random data live in dimension $_3$8 rather than $_3$9. For the second Rényi entropy, averaging over the random symmetric tensors turns the problem into a sum over $_2$0 domain configurations. The dominant contribution at large bond dimension $_2$1 is the minimal domain wall, yielding

$_2$2

or equivalently

$_2$3

The gauge symmetry changes the degeneracies and finite-$_2$4 corrections but does not change the leading geometric term (Chirco et al., 2017).

Group field theory (GFT) networks sharpen the same idea in a field-theoretic language. A boundary network state is built by placing group fields at graph vertices and maximally entangled link states on graph edges. Rényi partition functions become sums over stranded Feynman graphs, each amplitude scaling as

$_2$5

with $_2$6 the cutoff-dependent effective bond dimension. In the free symmetric GFT, the dominant diagrams are those whose local permutation pattern jumps across a unique minimal cut $_2$7, and one finds

$_2$8

For a broad class of networks, linear corrections from a polynomial perturbation of the Gaussian measure are negligible in the symmetric model, while in the non-symmetric model the same minimal-cut scaling survives but the prefactor is renormalized at linear order in the interaction strength (Chirco et al., 2019).

Outside holography proper, flow-equation holography provides an RT-like formula for the min-entropy of a weakly coupled region in a generic many-body system:

$_2$9

In one dimension this can be rewritten as the length of a minimal curve in an emergent geometry with metric

$A=[a,b]$0

so the decomposition becomes an integral of a disentangling density along an RG-like direction rather than a bulk area in AdS (Kehrein, 2017).

4. Algebraic entropy, edge terms, and error-correcting codes

A different decomposition starts from operator algebras rather than graphs. In finite-dimensional operator-algebra quantum error correction, a bulk von Neumann algebra $A=[a,b]$1 with center $A=[a,b]$2 induces a decomposition

$A=[a,b]$3

and the algebraic entropy of a state $A=[a,b]$4 with respect to $A=[a,b]$5 is

$A=[a,b]$6

For complementary recovery, the boundary algebra $A=[a,b]$7 obeys an exact RT-like identity

$A=[a,b]$8

with $A=[a,b]$9 a central “area operator”; the same $\tilde A$0 appears for the complementary algebra $\tilde A$1. The decomposition is therefore classical center entropy plus intra-sector bulk entropy, corrected by a linear central operator (Harlow, 2016, Kamal et al., 2019).

This framework clarifies which entropy notion is relevant. Algebraic entropy obeys the RT-like formula. Distillable entropy does not: removing the Shannon term over center sectors destroys the linear decomposition. By contrast, adding a boundary-local “$\tilde A$2” term still gives an RT-like identity, but with a shifted area operator. In the gauge-theory interpretation, that extra term is local to the cut and is therefore associated with regularization of the area at the bulk cutoff rather than with the state-dependent long-distance geometry (Kamal et al., 2019).

The proposal that the RT area is an entanglement edge term fits naturally into this picture. In lattice gauge theory, the extended Hilbert-space entropy contains a Shannon edge term and a $\tilde A$3 edge term. The gravitational claim is that

$\tilde A$4

should be understood as the gravitational analogue of the $\tilde A$5 contribution, with the bulk entanglement term playing the role of interior entropy. This identifies the area contribution with UV degrees of freedom associated with gauge constraints at the entangling surface (Lin, 2017).

A path-integral derivation pushes the algebraic interpretation further. Under general axioms for an asymptotically AdS Euclidean gravitational path integral, one obtains type I von Neumann algebras $\tilde A$6 and $\tilde A$7 acting on $\tilde A$8, with Hilbert-space decomposition

$\tilde A$9

The associated entropy takes the standard sector form

$a,b$0

projection entropies are quantized as $a,b$1, and in semiclassical limits the gravitational replica construction computes this algebraic entropy by RT/HRT with quantum corrections (Colafranceschi et al., 2023).

5. Covariant, Lorentzian, and worldsheet extensions

The static geometric cut-and-paste proof of strong subadditivity does not automatically survive in time-dependent settings because HRT uses extremal rather than globally minimal surfaces, and because the relevant bulk regions need not lie in a common spatial slice. In AdS$a,b$2-Vaidya, numerical and analytic tests show that strong subadditivity is obeyed for all tested interval configurations when the null energy condition holds, and can be violated when it is violated. In this sense, covariant RT-like decompositions are constrained not only by extensivity and homology but also by bulk focusing properties (Callan et al., 2012).

A more radical extension analytically continues RT across the light cone for slab-like regions. In holographic CFTs and confining backgrounds, the relevant extremal surfaces split into Type I saddles with real turning point $a,b$3 and Type II saddles with complex $a,b$4. For conformal backgrounds the entropy takes the spacelike form

$a,b$5

for $a,b$6, and the timelike continuation

$a,b$7

for $a,b$8. In the confining case, the scale $a,b$9 modifies the turning-point condition so that real Type I saddles can persist even for part of the timelike regime. The decomposition is then between real and complex branches of the same extremal-surface family rather than between two purely spacelike geometries (Nunez et al., 23 Jul 2025).

Worldsheet constructions provide yet another RT-like dictionary. In hyperbolic open-string vertices, the entanglement wedge cross section $A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$0 is identified with an open-string scattering distance, and reflected entropy $A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$1 with the circumference of the waist of a cylinder produced by disk-disk scattering. The critical length

$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$2

appears simultaneously as the minimal EWCS, the minimal reflected entropy divided by two, and the critical length selected by the geometric BV master equation. Here the decomposition is not only bulk geometric but also worldsheet geometric: boundary-anchored RT legs, cross-geodesics, and closed cylinder geodesics encode $A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$3, $A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$4, and $A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$5, respectively (Jiang et al., 2024).

These Lorentzian and worldsheet extensions show that “RT-like” need not mean a real static minimal surface. Depending on the setting, it can mean an extremal surface constrained by the null energy condition, a complex saddle across the light cone, or a hyperbolic worldsheet geodesic selected by string-field-theoretic consistency.

6. Microscopic realizations, statistical origins, and scope

Several recent constructions derive RT-like decompositions directly from boundary statistical data. In multi-boundary AdS$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$6/CFT$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$7 states associated with multi-boundary black holes, the norm of the state and the replica partition functions can be computed in a large-$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$8 CFT ensemble by replacing heavy OPE coefficients with their Gaussian moments. Each RT phase then arises from a distinct leading Gaussian contraction pattern, and phase transitions occur when the dominant contraction pattern changes. The same ensemble average reproduces the Liouville partition function with ZZ boundary conditions and the exact gravitational path integral on the corresponding multi-boundary geometry (Bao et al., 16 Apr 2025).

A finer microscopic decomposition comes from Virasoro modular data. For several holographic AdS$A=\bigcup_{i=1}^n A_i,\qquad A_i=[a_i,b_i],$9/CFT$\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$0 setups, the entropy can be rewritten as

$\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$1

where the $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$2 are Liouville momenta, $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$3 is the Virasoro modular $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$4-matrix element, and $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$5 is the entropy of the corresponding Virasoro-TQFT sector. At large $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$6, the integral is saddle-dominated and the $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$7 entropy becomes

$\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$8

This identifies the RT area with the Cardy density of heavy primaries in the dominant coarse-grained Virasoro sector, or equivalently with the multiplicity of coarse-grained Virasoro intertwiners across the cut (Lin, 29 Jun 2026).

A non-holographic confining example appears in two-dimensional QCD. There the vacuum Rényi entropy of a single interval is controlled at order $\tilde A=\bigcup_{k=1}^n \gamma_{i_k j_k},$9 by the rainbow-dressed quark propagator, while the $\tilde A$00 correction is governed by the off-diagonal and off mass-shell mesonic T-matrix and does not change the central charge. For small intervals the entropy behaves as

$\tilde A$01

while for large intervals it saturates. A soft-wall AdS$\tilde A$02 model reproduces this crossover through the transition from connected to disconnected RT surfaces, so the leading quark term plays the role of a classical RT contribution and the mesonic term plays the role of a subleading correction (Liu et al., 2022).

Taken together, these works show that RT-like entropy decomposition is not a single formal identity but a structural class of representations. The recurring pattern is a dominant geometric or algebraic term selected by minimality, extremality, saddle dominance, or sector multiplicity, plus corrections that encode bulk entanglement, gauge-edge data, finite bond dimension, interactions, or regulator dependence. The precise decomposition depends on the framework, but the central theme remains the same: entanglement is organized by a constrained geometric object whose local pieces can be recombined, counted, or coarse-grained in a way that reproduces the RT logic.