For a Haar-random pure state of a balanced N×N bipartite system, the eigenvalues of the reduced density matrix (the entanglement spectrum), constrained to a fixed entropy deficit u = ln N − S_q, reorganize across two phase transitions in the large-N limit: a concentration (pushed-to-pulled) transition at u_C(q) and an evaporation transition at u_E(q). The 2013 source paper established the von Neumann case (q = 1), asserted without derivation that the evaporation transition disappears for Rényi exponent q < 1, and deferred the q = 1 crossover to follow-up work. The promised 2019 follow-up constructed the full (q, u) phase diagram — but only for q ≥ 1, and still in the large-N saddle-point limit. What remains open: derive the constrained typical spectrum and rate function for 0 < q < 1 and settle the fate of both transitions and the crossover mechanism at q = 1; derive finite-N corrections and the finite-size scaling of the critical lines, including the noted-but-underived O(1/ln N) derivative discontinuity; and characterize the entropy-independent (EIES/EISS) state classes.

Background

Setting. Consider a pure state |ψ⟩ of a bipartite quantum system with Hilbert space H_A ⊗ H_Ā, sampled uniformly with respect to the Haar (unitarily invariant) measure on the unit sphere, with a balanced bipartition dim H_A = dim H_Ā = N. The reduced density matrix ρ_A = tr_Ā |ψ⟩⟨ψ| is Hermitian, positive, and unit-trace; its eigenvalues λ_1 ≥ λ_2 ≥ ⋯ ≥ λ_N ≥ 0, Σ_k λ_k = 1, are the entanglement spectrum (also called Schmidt coefficients). Bipartite entanglement is quantified by entropies of this spectrum: the purity tr ρ_A²; the von Neumann entropy S_vN = −Σ_k λ_k ln λ_k; and the Rényi entropies S_q = (1/(1−q)) ln Σ_k λ_k^q for q > 0, q ≠ 1, with S_q → S_vN as q → 1 and S_2 = −ln tr ρ_A². (Notation warning: the 2013 source paper denotes the Rényi exponent by α, while its 2019 follow-up denotes it by q and reuses α for a spectral-support parameter; this record uses q throughout.)

Constrained typicality. For Haar-random states the joint density of the eigenvalues is p_N(λ⃗) ∝ Π_{j<k} (λ_j − λ_k)² on the simplex — the same Vandermonde-square factor that governs the eigenvalues of Wishart random matrices (covariance matrices W = X†X of rectangular Gaussian data matrices X, whose eigenvalue statistics are shared by reduced density matrices of Haar-random pure states). By Page's theorem the average von Neumann entropy is ⟨S_vN⟩ = ln N − 1/2 at large N, and the unconstrained typical spectrum follows the Marčenko–Pastur law with support [0, 4/N]. The constrained-typicality program asks for the most probable spectrum on isoentropic manifolds (manifolds of states at fixed entropy): fix the entropy deficit u = ln N − S ∈ [0, ln N] (u = 0 maximally entangled, u = ln N separable) and maximize p_N under this constraint. In the large-N limit this becomes a Coulomb-gas problem: the Vandermonde factor is the Gibbs weight of a two-dimensional logarithmic gas of N charges on a line, and the constrained maximizer solves saddle-point (singular integral) equations. The log-probability of the entropy, s(u) = lim_{N→∞} N^(−2) ln p_N({deficit u}), called the “entropy of the entropy”, is the associated rate function (s ≤ 0, maximal at the typical value). The same problem is equivalently cast in statistical-mechanics language, as in the 2019 paper: the constrained spectral statistics are generated by a Gibbs partition function Z_N(β) = ∫ dλ⃗ p_N(λ⃗) e^{−β N² E(λ⃗)} with energy density E(λ⃗) = ln N − S_q(λ⃗) (the deficit) and effective inverse temperature β conjugate to u; the unconstrained typical spectrum corresponds to β = 0, the entangled phase to β > 0, and the separable phase to β < 0, a “negative-temperature” construction in which one eigenvalue condenses out of the spectral sea.

Known results (large N, q ≥ 1). As u increases from 0, three phases appear: (i) entangled phase, 0 < u < u_C(q): a deformed semicircle spectrum bounded away from zero; (ii) typical phase, u_C(q) < u < u_E(q): Marčenko–Pastur; (iii) separable phase, u_E(q) < u < ln N: one O(1) eigenvalue splits off (“evaporates”) from the sea of O(1/N) eigenvalues. Two spectral phase transitions separate them: the concentration transition (also called pushed-to-pulled, in analogy with Coulomb gases pushed against a hard wall) at u = u_C(q), where the lower edge of the spectrum reaches the hard wall at λ = 0; and the evaporation transition at u = u_E(q). For Rényi exponent q ≥ 1 the critical lines are known explicitly (2019 follow-up, Eqs. (45) and (52)): uC(q)=(1/(q−1))ln[(4(q+1)/(3q))q⋅Γ(q+3/2)/(√π⋅Γ(q+2))]u_C(q) = (1/(q−1)) ln[ (4(q+1)/(3q))^q · Γ(q+3/2) / (√π · Γ(q+2)) ] and uE(q)=(1/(q−1))ln[2(2q)⋅Γ(q+1/2)/(√π⋅Γ(q+2))]u_E(q) = (1/(q−1)) ln[ 2^(2q) · Γ(q+1/2) / (√π · Γ(q+2)) ], with u_C(1) = ln(2/3) + 2/3 ≈ 0.26, u_E(1) = 1/2 (coinciding with the typical Page value), u_C(2) = ln(5/4), u_E(2) = ln 2, asymptotes u_C(∞) = ln(4/3) and u_E(∞) = 2 ln 2, and a minimum u_C ≈ 0.214 at q ≈ 3.733. The concentration transition is third order for q ≠ 1 and fourth order at q = 1; the evaporation transition is first order for q ≠ 1 and second order at q = 1 — both transitions soften by exactly one order at the von Neumann point. All of this is a saddle-point result: the 2013 paper states it “is valid when both subsystems are large”, the 2019 paper that it “is valid in the limit of large quantum systems”.

What lies outside the known picture. First, the Rényi range 0 < q < 1: the 2013 Letter asserts — without a derivation there — that “when the Renyi exponent α becomes smaller than 1, this phase transition disappears”, describing the von Neumann point as a crossover “between a violent (first-order) phase transition … for α > 1, and the absence of a phase transition when α < 1”. Neither the 2013 Letter, nor the Nadal–Majumdar–Vergassola analysis (which computes the distribution of S_q explicitly “for q > 1”, plus the q → 1 and q → ∞ limits, and remarks that the method is “also easily extendable to the case 0 < q < 1” without carrying out that extension), nor the 2019 phase diagram treats q < 1: the 2019 paper never addresses the q < 1 regime — its phase discussion, its explicit checks (q = 1, 2, 10 and q → ∞), and its Fig. 5 critical lines (which emanate from q = 1) all lie in q ≥ 1. The closed forms above are in fact finite for 0 < q < 1: their apparent 1/(q−1) singularity at q = 1 is removable, with limits u_C(1) and u_E(1), and the Γ(q − 1/2) factor — a potential pole at q = 1/2 — enters only the intermediate coefficients A_C(q), B_C(q) of the derivation (2019 Eqs. (42)–(43)), not the final critical lines. But these formulas were derived within the q ≥ 1 regime, and their continuation to q < 1 has neither derivation nor physical interpretation there; in particular whether the continued lines retain their meaning as phase boundaries (and keep their relative ordering) below q = 1 is unestablished. Second, finite N: the 2013 paper notes in the caption of its Fig. 4 that the derivative discontinuity of s(u) at u = 1/2 is O(1/ln N) — a logarithmic, non-power-law scaling that is noted but not derived; no finite-N correction or finite-size scaling theory for the critical lines exists in this literature line. In addition, the 2019 paper defines entropy-independent entangled states (EIES: states significantly entangled for every q, lying below the line tangent to the minimum of u_C(q)) and entropy-independent separable states (EISS: states with an O(1) separable component for every q, above u = 2 ln 2), remarking that “further investigation is needed to understand what EISS and EIES are and if they have some sort of characterization”.

Problem Statement

Goal: complete the characterization of the constrained-typical entanglement spectrum of Haar-random bipartite pure states — the (q, u) phase diagram — over the full Rényi range q > 0 and at finite N.

Setting (self-contained): balanced bipartition dim H_A = dim H_Ā = N; Haar-random pure state; entanglement spectrum λ_1 ≥ ⋯ ≥ λ_N of the reduced density matrix ρ_A; Rényi entropy Sq=(1/(1−q))lnΣkλkqS_q = (1/(1-q)) ln Σ_k λ_k^q; entropy deficit u=lnN−Squ = ln N − S_q; constrained typical spectrum = maximizer of the Vandermonde-square density p_N at fixed u; rate function sq(u)=limN→∞N−2lnpNs_q(u) = lim_{N→∞} N^{−2} ln p_N(states with deficit u). Known large-N results for q ≥ 1 are as in the Background: critical lines u_C(q), u_E(q) with the explicit formulas above; concentration transition of third order (fourth at q = 1); evaporation transition of first order (second at q = 1); three spectral phases (deformed semicircle, Marčenko–Pastur, evaporated).

Task 1 (the q < 1 regime; core). For 0 < q < 1, determine the constrained typical entanglement spectrum and the rate function s_q(u) at large N. Specifically: (a) does the concentration (pushed-to-pulled) transition survive for q < 1, and if so at what u_C(q) and of what order? (b) Does the evaporation transition disappear for q < 1, as asserted without derivation in the 2013 source paper? Prove, refute, or precisely qualify this assertion: if the transition survives in some form, locate it and determine its order; if it disappears, characterize how the large-N non-analyticity of s_q(u) at the would-be u_E is smoothed. (c) Identify where and why the q ≥ 1 saddle-point analysis breaks down as q crosses 1 — the role of the 1/(q−1) prefactor in u_C and u_E (removable at q = 1, but with no established meaning below it), the Γ(q − 1/2) factors entering the intermediate coefficients A_C, B_C of the derivation, the convexity/concavity properties of the eigenvalue functional λ^q used to select the saddle, and the fate of the negative-temperature construction of the separable phase.

Task 2 (the q = 1 crossover; core). Characterize the mechanism by which both transitions soften by exactly one order at q = 1 (first → second order evaporation, third → fourth order concentration), and make explicit the crossover between the first-order evaporation transition for q > 1 and its asserted absence for q < 1.

Task 3 (finite-N corrections and finite-size scaling; core). Derive the leading finite-N corrections to (i) the constrained typical spectral densities in the three phases and (ii) the rate function s_q(u), and determine the finite-size scaling of the critical lines u_C(q, N) and u_E(q, N). In particular, derive — or refute — the O(1/ln N) scaling of the derivative discontinuity of s(u) at u = 1/2, q = 1, noted but not derived in the 2013 paper, and state the general form (power-law versus logarithmic) of the finite-N rounding of each transition.

Task 4 (EIES/EISS; included, lighter component). Provide an operational characterization of the entropy-independent entangled states (EIES) and entropy-independent separable states (EISS) introduced in the 2019 paper.

References

  1. P. Facchi, G. Florio, G. Parisi, S. Pascazio, K. Yuasa, “Entropy-driven phase transitions of entanglement”, Phys. Rev. A 87, 052324 (2013). DOI: 10.1103/PhysRevA.87.052324, https://doi.org/10.1103/PhysRevA.87.052324 (preprint arXiv:1302.3383, https://arxiv.org/abs/1302.3383).
  2. P. Facchi, G. Parisi, S. Pascazio, A. Scardicchio, K. Yuasa, “Phase diagram of bipartite entanglement”, J. Phys. A: Math. Theor. 52, 414002 (2019). DOI: 10.1088/1751-8121/ab3f4e, https://doi.org/10.1088/1751-8121/ab3f4e (preprint arXiv:1906.01269, https://arxiv.org/abs/1906.01269).
  3. C. Nadal, S. N. Majumdar, M. Vergassola, “Statistical distribution of quantum entanglement for a random bipartite state”, J. Stat. Phys. 142, 403–438 (2011). DOI: 10.1007/s10955-010-0108-4, https://doi.org/10.1007/s10955-010-0108-4 (preprint arXiv:1006.4091, https://arxiv.org/abs/1006.4091).
  4. A. De Pasquale, P. Facchi, G. Parisi, S. Pascazio, A. Scardicchio, “Phase transitions and metastability in the distribution of the bipartite entanglement of a large quantum system”, Phys. Rev. A 81, 052324 (2010). DOI: 10.1103/PhysRevA.81.052324, https://doi.org/10.1103/PhysRevA.81.052324.
  5. P. Facchi, U. Marzolino, G. Parisi, S. Pascazio, A. Scardicchio, “Phase transitions of bipartite entanglement”, Phys. Rev. Lett. 101, 050502 (2008). DOI: 10.1103/PhysRevLett.101.050502, https://doi.org/10.1103/PhysRevLett.101.050502.
  6. F. D. Cunden, P. Facchi, M. Ligabò, P. Vivo, “Third-order phase transition: random matrices and screened Coulomb gas with hard walls”, J. Stat. Phys. 175, 1262–1297 (2019). DOI: 10.1007/s10955-019-02281-9, https://doi.org/10.1007/s10955-019-02281-9.
  7. A. Grabsch, “General truncated linear statistics for the top eigenvalues of random matrices”, J. Phys. A: Math. Theor. 55, 124001 (2022). DOI: 10.1088/1751-8121/ac52e3, https://doi.org/10.1088/1751-8121/ac52e3.
  8. D. Bernard, L. Piroli, “Entanglement distribution in the quantum symmetric simple exclusion process”, Phys. Rev. E 104, 014146 (2021). DOI: 10.1103/PhysRevE.104.014146, https://doi.org/10.1103/PhysRevE.104.014146.