Lubkin-Page Bounds in Quantum Systems

• Lubkin-Page bounds are typicality statements showing that in composite quantum systems a small subsystem is almost maximally mixed when paired with a much larger environment.
• Recent extensions use trace methods and index theory to generalize the bounds from finite-dimensional (Type I) settings to Type II von Neumann factors, aligning with QFT and quantum gravity observables.
• In tripartite decompositions, the mutual information between two subsystems becomes parametrically suppressed as the environment size increases, offering insights into black hole Page curves.

Searching arXiv for relevant papers on Lubkin-Page bounds and related generalizations. Lubkin-Page bounds are typicality statements for random pure states in composite quantum systems. In the standard finite-dimensional setting, they show that a subsystem of a Haar-random pure state is almost maximally mixed when its complement is much larger, and that in a tripartite decomposition $A\otimes B\otimes E$ the mutual information between $A$ and $B$ is parametrically suppressed when $E$ is large. Recent work extends this paradigm from Type I von Neumann algebras to all Type II von Neumann factors, thereby addressing the fact that the observable algebras of quantum field theory and quantum gravity are generically Type II or Type III rather than finite-dimensional tensor products (Wang et al., 2 Jul 2026).

1. Finite-dimensional Lubkin-Page theorem

In the familiar Type I setting, one considers a pure state drawn uniformly with respect to Haar measure from a bipartite Hilbert space

$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$

For $d_B\ge d_A$, Page’s formula gives the average entanglement entropy of subsystem $A$ as

$$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$

Equivalently, the average purity is

$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$

so the reduced state is close to maximally mixed when $d_B\gg d_A$ (Wang et al., 2 Jul 2026).

The tripartite form of the result is central for typicality arguments. If

$A$0

with $A$1, then the mutual information

$A$2

obeys

$A$3

In particular, when the environment $A$4 is large, generic pure states have almost no correlations between $A$5 and $A$6 (Wang et al., 2 Jul 2026).

A complementary exact formulation writes the mean entropy for an $A$7-dimensional subsystem in a total Hilbert space of dimension $A$8 as

$A$9

where $B$0 is the $B$1-th harmonic number. In the unbalanced regime $B$2, one has $B$3, while in the balanced case $B$4,

$B$5

so the average entropy lies below the maximum by a half-nat (Gersdorff, 29 Jun 2026).

2. Proof techniques and moment formulas

One standard derivation expands a Haar-random vector as

$B$6

with $B$7 uniformly distributed, and then computes moments of the reduced density matrix from the second and fourth Haar moments of the matrix elements. This yields

$B$8

from which the entropy can be reconstructed through replica methods or related integral identities (Hotta et al., 2015).

A more recent derivation replaces the usual random-matrix-theoretic route by Schur-Weyl duality and the character theory of the symmetric group $B$9. For a random pure state in $E$0,

$E$1

where $E$2 is the rising Pochhammer, $E$3 is the product of hook lengths, $E$4 is the content polynomial, and $E$5 is the character of a $E$6-cycle in the $E$7-irrep $E$8. Differentiation at $E$9 reproduces the exact Page formula (Gersdorff, 29 Jun 2026).

These derivations make precise what is meant by “typical.” In the finite-dimensional theorem, typicality is defined by Haar measure on the full Hilbert-space sphere. The resulting concentration around near-maximal entropy is the mechanism behind the suppression of mutual information in the tripartite regime (Wang et al., 2 Jul 2026).

3. Type II von Neumann factors

The recent extension to Type II von Neumann factors begins by replacing ordinary dimension counting with traces, finite-dimensional approximants, and index theory. A Type II$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$0 factor $$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$1 is an infinite-dimensional von Neumann algebra with trivial center and a unique faithful normal trace

$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$2

The hyperfinite II$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$3 factor is built by an approximately finite-dimensional chain

$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$4

For a subfactor $$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$5, relative size is measured by the Jones index $$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$6; in the tensor-product case $$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$7, one has

$$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$8

mirroring the finite-dimensional relation $$\mathcal H=\mathcal H_A\otimes\mathcal H_B, \qquad \dim\mathcal H_A=d_A,\ \dim\mathcal H_B=d_B.$$9 with index $d_B\ge d_A$0 (Wang et al., 2 Jul 2026).

A Type II$d_B\ge d_A$1 factor has the form

$d_B\ge d_A$2

where $d_B\ge d_A$3 is II$d_B\ge d_A$4 and $d_B\ge d_A$5 is the algebra of bounded operators on a separable infinite-dimensional $d_B\ge d_A$6. It carries a semifinite trace $d_B\ge d_A$7 with $d_B\ge d_A$8. In both the II$d_B\ge d_A$9 and II$A$0 settings, one recovers effective dimensions by choosing increasing sequences of finite-dimensional subalgebras. If

$A$1

then

$A$2

and the conditional expectation onto $A$3 preserves the trace and induces finite-dimensional approximants of states in $A$4 (Wang et al., 2 Jul 2026).

This framework is the algebraic substitute for the tensor-product Hilbert-space assumptions used in the original Lubkin-Page theorem. The generalization is therefore not a cosmetic extension: it addresses precisely the mismatch between finite-dimensional typicality arguments and the operator-algebraic structures that arise in QFT and quantum gravity (Wang et al., 2 Jul 2026).

4. The Type II$A$5 and Type II$A$6 bounds

For the hyperfinite Type II$A$7 factor, let

$A$8

with a compatible AFD tower

$A$9

and assume for large $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$0 that

$$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$1

If one draws a Haar-random unit vector in the GNS space $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$2, traces out $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$3, and denotes the resulting state on $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$4 by $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$5, then

$$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$6

Hence, as $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$7, the bound tends to zero, and any weak$$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$8 limit point $$\bigl\langle S(A)\bigr\rangle = \sum_{k=d_B+1}^{d_A d_B}\frac1k-\frac{d_A-1}{2d_B} = \ln d_A-\frac{d_A}{2d_B}+O\!\bigl((d_A/d_B)^2\bigr).$$9 satisfies $\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$0 (Wang et al., 2 Jul 2026).

The proof uses three ingredients. First, at each finite level $\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$1, the problem reduces to an honest finite-dimensional Haar-random state on

$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$2

so the standard Lubkin-Page bound applies. Second, lower semicontinuity of the Araki relative entropy under weak$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$3 limits transfers the estimate to the infinite-dimensional limit. Third, Markov’s inequality, or concentration of measure, shows that with high probability the mutual information is close to its mean (Wang et al., 2 Jul 2026).

For Type II$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$4, one writes

$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$5

and truncates the Type I$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$6 factors to dimensions $\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$7. At level $\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$8,

$\bigl\langle \Tr\,\rho_A^2\bigr\rangle =\frac{d_A+d_B}{d_A d_B+1} =\frac1{d_A}+\frac1{d_B}+O\!\bigl((d_A/d_B)^2\bigr),$9

with total GNS-dimension squared

$d_B\gg d_A$0

Assuming

$d_B\gg d_A$1

the Haar-random state $d_B\gg d_A$2 on $d_B\gg d_A$3 obtained by tracing out $d_B\gg d_A$4 satisfies

$d_B\gg d_A$5

This is the Type II$d_B\gg d_A$6 analogue of the finite-dimensional Lubkin-Page suppression (Wang et al., 2 Jul 2026).

5. Black holes, Page curves, and Hamiltonian refinements

In the gravitational crossed-product construction, the effective environment dimension satisfies

$d_B\gg d_A$7

so the Type II$d_B\gg d_A$8 bound becomes

$d_B\gg d_A$9

For macroscopic horizons this is an exponentially tiny correlation. The construction uses the modular flow, or equivalently the horizon area operator, to extend a QFT Type III algebra to Type II$A$00; the Bekenstein-Hawking bound cuts off the spectrum of the modular Hamiltonian and produces a finite effective dimension $A$01 (Wang et al., 2 Jul 2026).

Related black-hole applications appear in analyses of the Page curve. In a qubit-pair toy model of evaporation, the stepwise entropy change

$A$02

obeys

$A$03

with

$A$04

When $A$05, the bound collapses to $A$06, recovering the standard Lubkin-Page result for a two-dimensional subsystem. To obtain any negative $A$07, one must satisfy

$A$08

so small corrections do not make the Page curve turn over (Alvi et al., 2019).

A distinct refinement replaces Haar typicality on the full Hilbert space by canonical typicality in a microcanonical energy shell. For a nonzero Hamiltonian

$A$09

the typical reduced state of $A$10 is

$A$11

rather than the maximally mixed state. The corresponding entanglement entropy is approximately thermal entropy,

$A$12

which is in general strictly less than $A$13. In the black-hole context, one paper argues that this nonmaximal entanglement changes the standard firewall inference that was based on the zero-Hamiltonian Page limit (Hotta et al., 2015). This suggests that Lubkin-Page typicality is sensitive to the ensemble and dynamical constraints used to define “typical.”

6. Type III obstructions and open questions

The extension from Type II to Type III remains open. Type III factors, as in algebraic QFT on Minkowski space, admit no trace, no density matrices, and no Haar measure on state space. One instead has the Araki relative entropy

$A$14

and the mutual information

$A$15

Three obstacles are identified: there is no trace or maximally mixed state, there is no Haar measure, and there is no ordinary notion of dimension. The Jones index is replaced by the Kosaki-Longo index, but its relation to concentration of measure is not worked out (Wang et al., 2 Jul 2026).

Several open directions have been proposed. One is to use Tomita-Takesaki modular theory, including the Bisognano-Wichmann framework, to replace the partial-trace decomposition. Another is to invoke the eigenstate thermalization hypothesis for local algebras as a dynamical notion of typicality. A third is to identify the commutant in holographic settings, such as the black-hole interior, as a natural thermal bath; this may allow one to drop the explicit dimension constraint $A$16 and obtain tighter one-shot bounds (Wang et al., 2 Jul 2026).

A common misconception is that the finite-dimensional Lubkin-Page theorem automatically applies to local algebras in QFT. The Type II results show that a substantial generalization is possible, but they also clarify that Type III algebras require different tools and that the role of the ensemble, the trace, and the effective notion of dimension is structurally nontrivial. In that sense, the modern formulation of Lubkin-Page bounds is less a single theorem than a hierarchy of typicality statements whose validity depends on the operator-algebraic setting and on how randomness is defined.