Typical Entanglement of Superpositions

Technical Summary: Typical Entanglement of Superpositions

Problem Statement
The paper investigates the universal entanglement properties inherent to superpositions of randomized quantum states, specifically addressing whether superposing $m$ typical states $|\Psi_m\rangle = \sum_{i=1}^m \lambda_i |\psi_i\rangle$ can enhance total entanglement to the maximally allowed value. While random states drawn from a Haar measure are almost maximally entangled, structured ensembles such as random fermionic Gaussian states and random matrix-product states (MPS) typically exhibit "sub-maximal" entanglement, remaining significantly below the Page value regardless of system size. The central question is whether the superposition of $m$ such sub-maximally entangled components can bridge this gap to reach maximal entanglement, or if a distinct scaling law governs this regime.

Methodology
The authors analyze the entanglement of superpositions using the second Rényi entropy, $S_2 = -\ln \text{Tr}(\rho_A^2)$, where $\rho_A$ is the reduced density matrix (RDM) of a subsystem $A$. The core of the methodology involves analyzing the "purity tensor" $T_{ijk\ell} = \text{Tr}[\text{Tr}_B(|\psi_i\rangle\langle\psi_j|) \text{Tr}_B(|\psi_k\rangle\langle\psi_\ell|)]$, which determines the purity of the superposition.

The analysis relies on the thermodynamic limit ($|A| \to \infty$) and compares the self-overlap of component RDMs against their cross-overlap. The authors establish a logical chain connecting the entanglement density of the components ($s_2$) to the orthogonality of their RDMs:

1. Orthogonality Criterion: If the component entanglement density $s_2$ is strictly less than the Haar value $\ln(2)$, the cross-overlap between independent component RDMs becomes exponentially smaller than their self-overlap. This defines a "one-sided orthogonality" with a rate $c_\perp = \ln(2) - s_2 > 0$.
2. Two-Sided Orthogonality: At equal bipartitions, this implies two-sided orthogonality, where the RDMs of distinct components occupy disjoint subspaces and are incoherent mixtures.
3. Diagonal Dominance: This orthogonality forces the purity tensor to become diagonally dominant in the thermodynamic limit, collapsing the calculation of the superposition's purity to a sum of individual purities weighted by the superposition coefficients.

The authors explicitly calculate $s_2$ for two canonical sub-maximal ensembles:

• Random Fermionic Gaussian States: Utilizing the Wachter law (arcsine law) for the eigenvalues of the covariance matrix blocks.
• Random Matrix-Product States (MPS): Utilizing the Marchenko–Pastur law for the Schmidt spectrum of random tensors.

They also examine the "maximal" class, including random stabilizer states, where $s_2 \sim \ln(2)$.

Key Results
The study classifies superpositions of typical states into two distinct entanglement regimes based on the component entropy density $s_2$:

1. The Sub-Maximally Entangled Class ($s_2 < \ln(2)$):

   • In this regime, the components are forced to be two-sided orthogonal.
   • The entanglement of an $m$-component superposition with equal weights exhibits a universal logarithmic enhancement:
     $$\Delta S_2(m) \equiv \langle S_2(m) \rangle - \langle S_2(1) \rangle = \ln(m)$$
   • This result is a manifestation of the Linden–Popescu–Smolin (LPS) bound.
   • Specific Calculations:
     • For random fermionic Gaussian states at equal bipartition, $s_2 = \ln(2) - \ln\left(\frac{3+2\sqrt{2}}{4}\right) \approx 0.317$ (46% of the maximal value).
     • For random MPS with bond dimension $\chi$ and physical dimension $d$, $s_2 = \ln\left(\frac{d\chi}{d+1}\right)$.
   • Implication: Transitioning from the sub-maximal class to maximal entanglement requires an exponentially large number of superpositions ($m \sim e^N$). For Gaussian states, this logarithmic scaling corresponds to the "Gaussian extent" $\xi = m$, implying that states of polynomial extent carry only $O(\ln N)$ excess entanglement above the Gaussian baseline.

2. The Maximally Entangled Class ($s_2 \sim \ln(2)$):

   • This class includes Haar-typical states and random stabilizer states.
   • Here, the orthogonality condition $c_\perp > 0$ does not hold; the purity tensor is not diagonally dominant.
   • Superposition does not yield a $\ln(m)$ enhancement. Instead, the entanglement relaxes to the Haar limit via intensive, system-size ($N$) independent scaling laws.
   • For rank-deficit zero stabilizer states, the superposition causes a "one-bit collapse" from the maximal value at $m=1$ to the Haar value as $m \to \infty$, with a drop of $\Delta S_2(m) \to -\ln(2)$.
   • For generic stabilizer states, the relaxation is driven by fluctuations in rank-deficit, resulting in a negative $\Delta S_2$ that scales as $1/m^3$.

Significance and Claims
The paper claims to provide a unified framework for understanding entanglement in superpositions of random states, resolving the question of whether superposition can "fix" sub-maximal entanglement. The primary contribution is the identification of the logarithmic enhancement $\ln(m)$ as a universal upper bound for any system in the sub-maximal class, strictly constrained by the orthogonality of component RDMs.

The authors assert that:

• Superpositions of sub-maximally entangled states (like Gaussian states or low-bond-dimension MPS) cannot reach maximal entanglement without an exponential number of components.
• The $\ln(m)$ scaling is not merely a feature of specific models but a generic consequence of the "two-sided orthogonality" inherent to sub-maximal ensembles.
• This result has direct implications for quantum simulation and quantum chemistry: it suggests that truncated configuration-interaction (CI) expansions (superpositions of Slater determinants) are fundamentally limited in their ability to capture extensive entanglement without exponential resources.
• The work distinguishes between "extensive" growth (sub-maximal class) and "intensive" relaxation (maximal class), providing a quantitative hierarchy of state families based on the ratio $s_2 / \ln(2)$.

The paper does not propose new experimental setups but offers a theoretical classification and analytical derivation of entanglement scaling laws that serve as reference points for structured random ensembles in many-body physics.