---
title: Balanced Partial Entanglement Entropy
url: https://www.emergentmind.com/topics/balanced-partial-entanglement-entropy
type: topic
---

# Balanced Partial Entanglement Entropy

Balanced partial entanglement entropy (BPE) is a mixed-state correlation measure built from partial entanglement entropy (PEE), the quasi-local contribution of a subset to an entanglement entropy. For a bipartite mixed state \(\rho_{AB}\), BPE is defined by introducing a purification \(AA'BB'\), partitioning the purifier, imposing balance conditions on the PEEs of \(A\) and \(B\), and then evaluating the balanced contribution of \(A\) inside \(AA'\); in the pure-state limit it reduces to \(S(A)=S(B)=I(A:B)/2\). In the holographic constructions studied so far, BPE coincides with the entanglement wedge cross section (EWCS), and later analyses extended this relation to covariant configurations, gravitational anomalies, asymptotically flat holography, island phases, and thread-based geometric formulations [2103.00415, 2205.10858, 2203.05491].

## 1. Definition from partial entanglement entropy

The starting point is PEE. For a subsystem \(A\) partitioned into three ordered, non-overlapping subsets \(A=\alpha_L\cup\alpha\cup\alpha_R\), the additive linear combination (ALC) proposal defines the PEE of \(\alpha\) by
$$
s_A(\alpha)=\frac{1}{2}\Big[S(\alpha_L\cup\alpha)+S(\alpha\cup\alpha_R)-S(\alpha_L)-S(\alpha_R)\Big].
$$
This quantity obeys positivity, additivity, normalization \(s_A(A)=S(A)\), and symmetry properties, and in the continuum defines an entanglement contour \(s(x)\) through \(s_A(\alpha)=\int_\alpha s(x)\,dx\). In 1+1 dimensions, and more generally in quasi-one-dimensional symmetric setups, this formula provides the operational input for BPE [2203.05491, 1910.10978].

For a bipartite mixed state \(\rho_{AB}\) on \(H_A\otimes H_B\), one introduces a purification \(|\psi\rangle\in H_A\otimes H_B\otimes H_{A'}\otimes H_{B'}\) such that \(\mathrm{Tr}_{A'B'}|\psi\rangle\langle\psi|=\rho_{AB}\). The purifier is partitioned into \(A'\) and \(B'\). The balance conditions, in the form used by Wen–Camargo, are
$$
s_{AA'}(A)=s_{BB'}(B),\qquad s_{AA'}(A')=s_{BB'}(B').
$$
BPE is then defined by minimizing the balanced PEE:
$$
\mathrm{BPE}(A:B)=\min_\psi\, s_{AA'}(A)\big|_{\text{balance}}.
$$
For adjacent intervals, the balance condition is equivalently \(I(A,B')=I(A',B)\), where \(I\) denotes the PEE interpreted as a mutual-like quasi-local measure. For disjoint intervals, the complement is itself disconnected, and the balance conditions split into pairwise constraints for the further decomposition \(A'=A_1'\cup A_2'\), \(B'=B_1'\cup B_2'\) [2103.00415, 2201.13362].

The pure-state limit is fixed. For bipartitions \(A\cup B\) of a pure state, the balanced point yields
$$
\mathrm{BPE}(A:B)=S(A)=S(B),
$$
and since \(I(A:B)=2S(A)\), one has
$$
\mathrm{BPE}(A:B)=\frac{1}{2}I(A:B).
$$
This fixes BPE as a genuine extension of half the mutual information away from the pure-state setting [2203.05491].

## 2. Mixed-state correlation structure and relation to other measures

BPE was proposed as a measure of the total intrinsic correlation between \(A\) and \(B\) in a mixed state. In the adjacent case, the PEE decomposition of the purified system separates a direct term \(I_{AB}\) from crossing terms such as \(I_{AB'}\) and \(I_{A'B}\). The balanced point minimizes the crossing contribution, and the decomposition takes the form
$$
C(A,B)=I_{AB}+I_{AB'}\big|_{\text{balance}}=\mathrm{BPE}(A:B),
$$
with the balanced crossing correlation identified as the excess over the direct \(I_{AB}\) contribution. In holographic CFT\(_2\), for adjacent intervals,
$$
[I_{AB'}+I_{A'B}]_{\text{balance}}=\frac{c}{3}\ln 2,\qquad
I_{AB'}\big|_{\text{balance}}=I_{A'B}\big|_{\text{balance}}=\frac{c}{6}\ln 2.
$$
For non-adjacent intervals, the corresponding crossing sums are extremized at the balance point but are not universal constants [2201.13362].

This structure places BPE near reflected entropy and entanglement of purification, but not identically on the same footing in all formulations. On the canonical purification,
$$
\mathrm{BPE}(A:B)=\frac{1}{2}S_R(A:B)
$$
in the normalization used in the original 2021 construction. The covariant and anomalous analysis of 2022 adopts the convention \(S_R(A:B)\equiv S_{AA'}/2\), so that in that normalization one simply writes \(BPE=SR\). In the large-\(N\), semiclassical regime, these relations align with the holographic EWCS result [2103.00415, 2205.10858].

The entropy relations obeyed by BPE closely parallel those of entanglement of purification. The established bounds include
$$
\mathrm{BPE}(A:B)\le \min\{S(A),S(B)\},
$$
and, in holographic theories,
$$
\mathrm{BPE}(A:B)\ge \frac{1}{2}I(A:B).
$$
The 2021 construction also gives monotonicity under inclusion, \(B\subset BC\Rightarrow \mathrm{BPE}(A:BC)\ge \mathrm{BPE}(A:B)\), and the polygamy-type inequality \(\mathrm{BPE}(A:B)+\mathrm{BPE}(A:C)\ge \mathrm{BPE}(A:BC)\) for a global pure state on \(ABC\) [2103.00415].

A recurrent conceptual issue is purification dependence. The original definition describes BPE as purification-dependent and therefore not intrinsic. Later work conjectures purification independence and gives non-trivial checks in covariant anomalous CFT\(_2\), canonical purifications, and holographic constructions. The literature therefore treats purification dependence and purification independence as distinct but closely connected aspects of the quantity, depending on the regime and the strength of the evidence available [2103.00415, 2205.10858].

## 3. Holographic duality, covariance, and gravitational anomalies

In holography, the central equality is
$$
\mathrm{BPE}(A:B)=E_W(A:B)=\frac{\mathrm{Area}(\Sigma_{AB})}{4G_N},
$$
where \(\Sigma_{AB}\) is the entanglement wedge cross section. In AdS\(_3\)/CFT\(_2\), the derivation uses the holographic entanglement contour: \(\Sigma_{AB}\) is extended to a complete geodesic that becomes an RT or HRT surface for a suitable purified region, and the geodesic-chord picture identifies its length with the balanced PEE of \(A\) inside \(AA'\). This gives a contour-based interpretation of EWCS and explains why BPE reproduces it in the static and quasi-static configurations studied [2103.00415].

The covariant extension treats Lorentzian AdS\(_3\) and time-dependent intervals. In non-anomalous CFT\(_2\), one balance equation is insufficient to fix both coordinates of a partition point, but in anomalous theories with \(c_L\neq c_R\) the balance conditions split into geometric and anomalous sectors and fix the partition covariantly. Writing
$$
c_{\mathrm{geo}}\equiv c_L+c_R,\qquad c_{\mathrm{ano}}\equiv c_L-c_R,
$$
the adjacent-interval result is
$$
\mathrm{BPE}(A:B)=SR(A:B)=EW(A:B)
=\frac{c_{\mathrm{geo}}}{12}\log\!\left(\frac{2R_A R_B}{\delta R_{AB}}\right)
-\frac{c_{\mathrm{ano}}}{12}\left(\kappa_A+\kappa_B-\kappa_{AB}\right),
$$
where \(R_A,R_B,R_{AB}\) are proper lengths and \(\kappa_A,\kappa_B,\kappa_{AB}\) are boost angles. For non-adjacent intervals, the result is written in terms of Lorentzian cross-ratios \(\eta,\bar\eta\):
$$
\mathrm{BPE}(A:B)=SR(A:B)=EW(A:B)
=\frac{c_{\mathrm{geo}}}{24}\log\!\left[\frac{(\sqrt{\eta}+1)(\sqrt{\bar\eta}+1)}{(\sqrt{\eta}-1)(\sqrt{\bar\eta}-1)}\right]
+\frac{c_{\mathrm{ano}}}{24}\log\!\left[\frac{(\sqrt{\eta}+1)(\sqrt{\bar\eta}-1)}{(\sqrt{\eta}-1)(\sqrt{\bar\eta}+1)}\right].
$$
These formulas decompose cleanly into Einstein-Hilbert and anomalous sectors [2205.10858].

The anomalous contribution has a gravitational dual in topological massive gravity (TMG). For a geodesic chord \(\Sigma_{AB}\), the EWCS functional is
$$
EW(A:B)=\frac{\mathrm{Length}(\Sigma_{AB})}{4G}
+\frac{1}{4G\mu}\int_{\Sigma_{AB}} d\tau\,(\tilde n\cdot \nabla n),
$$
with the Chern–Simons term fixed by choosing endpoint normals from the HRT surfaces intersected by \(\Sigma_{AB}\). This is the first prescription, in the cited work, for the entropy quantity associated to EWCS beyond Einstein gravity. The balanced crossing PEE remains anomaly-free:
$$
I_C\big|_{\text{balance}}=I(A,B')+I(B,A')=\frac{c_{\mathrm{geo}}}{6}\log 2,
$$
so the Markov-gap-type constant depends only on the geometric sector [2205.10858].

## 4. Asymptotically flat holography and GCFT\(_2\)

A complete flat-space realization of BPE was developed in \((1+1)\)-dimensional Galilean conformal field theories dual to Einstein gravity and TMG in asymptotically flat spacetimes. The bulk geometry is written in Bondi gauge,
$$
ds^2=8G_N M\,du^2-2\,du\,dr+8G_N J\,du\,d\phi+r^2 d\phi^2,
$$
and the central charges are
$$
c_L=\frac{3}{\mu G_N},\qquad c_M=\frac{3}{G_N}.
$$
In the Einstein limit \(\mu\to\infty\), \(c_L\to 0\) while \(c_M=3/G_N\). In this setting, the field-theory BPE matches the geometric EWCS exactly in Minkowski vacuum, on the global Minkowski orbifold, and in the thermal flat-space cosmology (FSC) background, for both Einstein gravity and TMG [2203.05491].

For adjacent intervals on the plane,
$$
A=[(x_1,t_1),(x_2,t_2)],\qquad
B=[(x_2,t_2),(x_3,t_3)],
$$
the result is
$$
\mathrm{BPE}(A:B)=\frac{c_L}{12}\ln\!\left[\frac{2 t_{12} t_{23}}{\epsilon (t_{12}+t_{23})}\right]
+\frac{c_M}{12}\left[\frac{x_{12}}{t_{12}}+\frac{x_{23}}{t_{23}}-\frac{x_{13}}{t_{13}}\right].
$$
At the balance point, the crossing PEE behaves differently in the two bulk theories:
$$
\mathcal{I}(A,B')\big|_{\text{balance}}=0\qquad\text{for Einstein gravity }(c_L=0),
$$
while in TMG
$$
\mathcal{I}(A,B')\big|_{\text{balance}}=\frac{c_L}{12}\ln 2=\frac{\log 2}{4\mu G_N}.
$$
The vanishing Einstein result indicates perfect Markov recovery, whereas the TMG result realizes a topological Markov gap [2203.05491].

For disjoint intervals on the plane,
$$
A=[(x_1,t_1),(x_2,t_2)],\qquad B=[(x_3,t_3),(x_4,t_4)],
$$
one defines
$$
T=\frac{t_2(t_3-t_4)}{t_3(t_2-t_4)},
$$
$$
X=T\left[\frac{x_2}{t_2}+\frac{x_3-x_4}{t_3-t_4}-\frac{x_3}{t_3}-\frac{x_2-x_4}{t_2-t_4}\right],
$$
and obtains
$$
\mathrm{BPE}(A:B)=\frac{c_L}{12}\ln\!\left(\frac{1+\sqrt{T}}{1-\sqrt{T}}\right)
+\frac{c_M}{12}\left|\frac{X}{\sqrt{T}(1-T)}\right|.
$$
Analogous formulas hold on the global Minkowski orbifold, with \(\tilde T,\tilde X\), and in FSC, with \(\hat T,\hat X\). In all these backgrounds, the paper establishes
$$
\mathrm{BPE}(A:B)=E_W(A:B)=\frac{\mathrm{Area}(\Sigma_{AB})}{4G_N},
$$
and also notes that, up to OPE constants, the disjoint-interval expression matches half the reflected entropy at large \(c\) [2203.05491].

The flat-holographic analysis also identifies the phase structure. EWCS and BPE are nonzero only when the entanglement wedge of \(A\cup B\) is connected. For disjoint intervals, connectivity is controlled by the cross-ratios \(T,\tilde T,\hat T\); when the minimal surface channel flips, the entanglement wedge cross section disappears and \(E_W=BPE=0\). This reproduces the same connected/disconnected distinction familiar from holographic mixed-state observables, but now in the asymptotically flat setting [2203.05491].

## 5. Island phases, ownerless islands, and generalized balance

In island phases, the PEE/BPE construction must be modified because of self-encoding. If \(A\) admits an island \(\mathrm{Is}(A)\), then the state on \(\mathrm{Is}(A)\) is completely determined by the state on \(A\), and the normalization property of two-body PEE is altered accordingly. The generalized construction introduces the ownerless island region
$$
\mathrm{Is}_{\mathrm{ownerless}}
:=\mathrm{Is}(A\cup B)\setminus\big(\mathrm{Is}(A)\cup \mathrm{Is}(B)\big),
$$
which is partitioned as \(I_o(AB)=I_o(A)\cup I_o(B)\) and combined with the ordinary islands into generalized islands
$$
I_r(A):=\mathrm{Is}(A)\cup I_o(A),\qquad
I_r(B):=\mathrm{Is}(B)\cup I_o(B).
$$
The balanced contributions are then written using a generalized ALC prescription in terms of region-complement PEEs \(\tilde S\), and the generalized balance equations are imposed on the island-dressed regions [2305.04259].

This modification is not optional in the island phase. Different assignments of the ownerless island produce different BPEs, and these correspond exactly to different saddles of the EWCS in the entanglement wedge of \(A\cup B\). For adjacent intervals, the three canonical assignments are labeled \(A2a\), \(A2b\), and \(A2c\): assigning the ownerless island entirely to \(B\), splitting it, or assigning it entirely to \(A\). For disjoint intervals, the analogous saddles are \(D2a\), \(D2b\), and \(D2c\). The minimal BPE is obtained by minimizing over the ownerless-island assignment and coincides with the minimal EWCS saddle:
$$
\mathrm{BPE}(A:B)=E_W(A:B)=\frac{\mathrm{Area}[\Sigma_{AB}]}{4G_N}.
$$
In the disconnected-wedge phase \(D3\), one finds \(BPE(A:B)=0\), again matching EWCS [2305.04259].

For example, in adjacent island phases with \(A=[b_1,b_2]\) and \(B=[b_2,b_4]\), the three saddle values are
$$
A2a:\quad \mathrm{BPE}(A:B)=\frac{c}{6}\log\!\left[\frac{b_2^2-b_1^2}{b_1\delta}\right],
$$
$$
A2b:\quad \mathrm{BPE}(A:B)=\frac{c}{6}\log\!\left(\frac{2b_2}{\delta}\right)+\frac{c}{6}\kappa,
$$
$$
A2c:\quad \mathrm{BPE}(A:B)=\frac{c}{6}\log\!\left[\frac{b_4^2-b_2^2}{b_4\delta}\right].
$$
The split assignment \(A2b\) is fixed by minimizing over the internal split parameter \(q\), with the minimum at \(q=b_2\). Disjoint-interval island saddles have analogous closed forms, including the brane-anchored value
$$
D2b:\quad \mathrm{BPE}(A:B)=\frac{c}{6}\log\!\left(\frac{\sqrt{b_3}+\sqrt{b_2}}{\sqrt{b_3}-\sqrt{b_2}}\right)+\frac{c}{6}\kappa.
$$
These formulas are obtained entirely on the field-theory side from generalized PEE and match the geometric EWCS phase by phase [2305.04259].

A complementary thread formulation replaces each boundary point in the Weyl-transformed CFT\(_2\) model by a cutoff sphere in Poincaré AdS\(_3\), leaving the bulk PEE-thread network itself unchanged. Two-point and four-point twist correlators are then reproduced by minimizing the number of intersections between admissible homologous bulk surfaces and the PEE threads. In that picture, the island formula is recovered by allowing surfaces to anchor on any cutoff spheres, and the resulting geometric understanding provides the foundation for computing BPE in island phases [2408.13535].

## 6. Higher-dimensional extensions, PEE threads, and scope

Beyond 1+1 dimensions, the contour/PEE framework remains available but requires geometric regulators. For a ball \(A=\{r\le R\}\) in a \(d\)-dimensional CFT vacuum on a hyperplane, the entanglement contour is
$$
s_A(r)=\frac{c}{6}\left(\frac{2R}{R^2-r^2}\right)^{d-1},
$$
and the exact relation between the UV cutoff \(\delta\) and the geometric cutoff \(\epsilon\) is
$$
\epsilon=\frac{R\big(R+\delta-\sqrt{R^2-\delta^2}\big)}{R+\delta}
=\delta-\frac{\delta^2}{2R}+\frac{\delta^3}{2R^2}-\frac{3\delta^4}{8R^3}+\frac{3\delta^5}{8R^4}+O(\delta^6).
$$
In symmetric quasi-one-dimensional configurations, balanced partitions can be solved explicitly. For a spherical interface in a vacuum CFT, the balanced locus is the midpoint of the thin complement shell, and the resulting BPE equals the geometrically regulated entropy with half-width cutoff and holographically equals \(\mathrm{Area}(\Sigma_{AB})/(4G)\). The same framework extends to spherical shells and strips, but the paper emphasizes that naive insertion of UV-regulated entropies into ALC fails in higher dimensions; the subset entropies must be evaluated as PEE limits with a fixed geometric regulator [1905.05522].

A separate line of work geometrizes two-point PEE as bulk geodesics called PEE threads. For a fixed boundary point, the geodesics define a divergenceless PEE-thread flow, and for a static interval or sphere the superposition over boundary points produces a bit-thread flow saturating the standard norm bound on the RT surface. For disconnected intervals, however, a single locking vector flow is no longer adequate; instead one assigns each thread a weight equal to the number of times it intersects a homologous surface and minimizes the weighted sum. The paper does not explicitly define BPE. It nevertheless develops a weighted PEE-thread reformulation in which minimizing the flux across an internal separating surface yields a natural balanced quantity, and this suggests a direct thread-based route to \(BPE=EWCS\) in static holographic settings [2311.02301].

The PEE-network formulation sharpens this geometric picture. In static Poincaré AdS, the density of PEE threads through any bulk point is exactly \(1/(4G)\), and for any bulk codimension-2 surface \(\Sigma\) the number of intersections with the full PEE network equals \(\mathrm{Area}(\Sigma)/(4G)\). The RT formula is then reformulated as minimization of intersections, and the resulting area-counting statement is identified with the Crofton formula in Poincaré AdS. This does not by itself provide a standalone BPE definition, but it strongly supports the general interpretation of EWCS, and hence BPE, as an intersection count inside a fixed PEE network [2401.07471].

The current framework is technically powerful but not universal. The recurring assumptions are the large-central-charge, classical-gravity regime; the use of ALC formulas in 2D or quasi-one-dimensional symmetric setups; neglect of quantum and \(1/c\) corrections; and, in island phases, reliance on the generalized ALC plus the two basic proposals for region-complement PEEs. The literature therefore establishes BPE most sharply in semiclassical holography, while higher dimensions, general non-symmetric regions, fully covariant thread constructions, and quantum corrections remain open directions within the formulations surveyed here [2203.05491, 2305.04259, 2311.02301].

Source: https://www.emergentmind.com/topics/balanced-partial-entanglement-entropy