Pith. sign in

REVIEW 4 major objections 3 minor 45 references

An unbiased measure over the matrix product state manifold

T0 review · 4 major / 3 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Sequential random matrix product states are not uniform on the MPS manifold; this paper derives and samples the unbiased Fubini-Study measure.

desk verdict A genuinely new and plausible correction to the standard random MPS ensemble, but the load-bearing metric determinant is asserted rather than proven; deserves serious refereeing with a request for a full derivation or a direct numerical check. read the letter →

arxiv 2505.00073 v1 pith:GT23LEBD submitted 2025-04-30 quant-ph cond-mat.stat-mech

Sebastian Leontica , Andrew G. Green This is my paper · ORCID
classification quant-phcond-mat.stat-mech
keywords matrixproductstatesFubini-StudymeasurerandomtensornetworksentanglementasymmetryHaarunitariesMetropolissamplingfreeprobabilityMarchenko-Pasturdistribution
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper establishes that the standard ensemble of random matrix product states, built by placing independent Haar-random unitaries on every site, is not uniform with respect to the Fubini-Study measure that the manifold inherits from the full Hilbert space. The bias is visible as an asymmetry in entanglement entropy when the chain is reflected, an artifact of the left-canonical construction. The paper derives the exact conversion factor between the two measures, $d\mu_{\mathrm{FS}} = d\mu_{\mathrm{RMPS}}\prod_i |\Gamma_i|^{D(d-1)}$, where $D$ is the bond dimension, $d$ the physical dimension, and $\Gamma_i$ the right-environment matrices that encode the Schmidt spectrum across each cut. It then constructs a Metropolis-Hastings sampler for the corrected measure and studies its properties, including a resolution of identity over MPS and a typical entanglement spectrum that differs from the sequentially generated ensemble. A sympathetic reader should care because statements about typical entanglement, chaos, and thermalization in tensor networks are only meaningful if the sampling measure is the natural unbiased one.

What carries the argument

The load-bearing object is the right-environment matrix $\Gamma_i$, defined recursively from the right edge of the chain; its determinant appears in the Fubini-Study volume element as $|g| = \prod_i |\Gamma_i|^{D(d-1)}$, so it directly controls the measure correction. The second ingredient is the exponential parameterization $U(x) = U_0 \exp\!\begin{pmatrix} 0 & -x^\dagger \\ x & 0 \end{pmatrix}$, which fixes the gauge and, as the paper asserts with a cited reference, covers the whole MPS manifold; the Haar measure over unitaries is then transformed into Lebesgue measure over the complex matrices $x$ with a Jacobian computed from random matrix theory. Comparing that Jacobian with the metric determinant gives the measure ratio, and makes the correction computable and samplable.

What would settle it

Compute the Fubini-Study metric determinant numerically at many points of the exponential coordinates for a small chain, including points where some Schmidt coefficient vanishes, and compare the local volume element with $\prod_i |\Gamma_i|^{D(d-1)}$; a mismatch on any set of positive measure would falsify Eq. (12). A complementary check is to generate Metropolis samples and test for exact left-right symmetry of the entanglement entropy profile.

Watch

Extended reading notes

Core claim

The central claim is the measure relation (12): the sequentially generated RMPS measure and the unbiased Fubini-Study measure on the matrix product state manifold differ by a state-dependent factor, $$d\mu_{\mathrm{FS}} = d\mu_{\mathrm{RMPS}}\prod_{i=1}^{N} |\Gamma_i|^{D(d-1)},$$ with $\Gamma_i$ the right environment at cut $i$, whose eigenvalues are the squared Schmidt coefficients of the bipartition across that cut. The derivation computes the metric determinant of the MPS manifold in a gauge-fixed exponential parameterization and compares it with the Jacobian of the Haar measure on the same coordinates; because the reference point can be chosen anywhere, the paper concludes the ratio holds globally. The direct consequence is that RMPS samples over-represent states with small right environments, producing a left-right asymmetry in entanglement entropy and a typical entanglement spectrum that differs from the unbiased ensemble. The paper also shows the corrected measure yields a resolution of identity proportional to the many-body identity, and provides an explicit Metropolis-Hastings algorithm with numerical verification.

Load-bearing premise

The whole argument stands or falls on whether every gauge-fixed matrix product state can be reached by the exponential coordinate patch used in the calculation, without missing regions or singularities, so that the local volume ratio computed at one reference point is valid everywhere.

Editorial extensions

If this is right

  • Entanglement profiles drawn from the unbiased FS ensemble are symmetric under spatial inversion, while RMPS profiles are not; shifting the orthogonality center to the middle does not fix this but introduces a discontinuity.
  • The determinant factor repels the eigenvalues of the right environment away from zero, so typical FS samples carry more entanglement entropy than RMPS samples at the same bond dimension.
  • The FS measure admits a resolution of identity, $\int d\mu_{\mathrm{FS}}(\Psi)|\Psi\rangle\langle\Psi| = Z\,\mathbb{I}_{d^N}$ with $Z = \int \prod_i dU_i\, e^{-D(d-1)\sum_i \log|\Gamma_i|^{-1}}$, turning the MPS manifold into a usable semiclassical phase space.
  • The typical right-environment spectrum in the FS ensemble is approximately Marchenko-Pastur with aspect ratio $1/(2d-1)$ instead of $1/d$, with deviations visible near the spectral edge.
  • A Metropolis-Hastings sweep over the chain mixes rapidly in the tested cases, making the unbiased ensemble practically samplable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the correction depends only on the Schmidt spectrum, any unbiased MPS sampling scheme in a different gauge must reproduce the same state-space volume; comparing central-gauge samples with FS samples would provide a direct numerical check.
  • The orientation bias found here should afflict other tensor-network ensembles built from independent local unitaries, and the same determinant-ratio strategy may extend naturally to loopless networks such as tree tensor networks, which the paper names as a plausible next step.
  • The paper leaves open three stated soft spots: its Marchenko-Pastur prediction for the FS spectrum is a first-order, non-self-consistent approximation; the Metropolis burn-in scaling with $N$ and $D$ is unknown; and an efficient sequential (rather than Markov-chain) sampler does not yet exist.
  • If the resolution of identity holds, the FS measure gives a bona fide classical phase space for the time-dependent variational principle, which would allow projected Lyapunov spectra and statistical-mechanics treatments of entanglement dynamics to be defined; this is an application the paper anticipates but does not carry out.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 3 minor

Summary. The manuscript considers the standard random matrix product state (RMPS) ensemble, formed by applying independent Haar-random unitaries at each site of a left-canonical MPS, and argues that this ensemble is biased with respect to the Fubini-Study volume on the MPS manifold. The central claim is Eq. (12): the Fubini-Study measure is related to the RMPS measure by dμ_FS = dμ_RMPS ∏_i |Γ_i|^{D(d-1)}, where Γ_i is the right environment at site i. The authors derive this from a local metric determinant, propose a Metropolis-Hastings sampler for the corrected measure, and study properties of the new ensemble, including a resolution of identity, entanglement asymmetry under spatial inversion, and an approximate Marchenko-Pastur spectrum with aspect ratio 1/(2d−1).

Significance. If fully established, the paper provides a parameter-free, gauge-invariant correction to a commonly used random-MPS ensemble. The predicted spatial-inversion asymmetry is a sharp and falsifiable claim, and the correction factor depends only on the entanglement spectrum, which makes it easy to apply. The paper also gives a concrete sampling algorithm and makes an analytic prediction for the typical entanglement spectrum. The central formula is elegant and the numerical investigations are appropriate. However, several load-bearing derivations are either missing or contain inconsistencies, so the manuscript requires substantial revision before the claims are fully supported.

major comments (4)
  1. [MPS Measure, Eq. (8) and Supplement Eq. (26)] The Jacobian printed in Eq. (8) has Δ²(ˆx) in the denominator, whereas the supplement's Eq. (26) has Δ²(ˆx²). Since the text defines ˆx² = x†x, the main-text expression gives J(x,x)→0 as x→0 for D>1, which would make the coordinate comparison at the origin singular; the supplement version is the correct one. This inconsistency affects the derivation of Eq. (12) as written and must be fixed by correcting Eq. (8) and any statement that depends on it.
  2. [MPS Measure, Eq. (10)] The central determinant identity |g| = ∏_i |Γ_i|^{D(d−1)} is asserted after "standard MPS contraction calculations" with no derivation. Because Eq. (12) is the main result, this is a load-bearing gap. Please provide a full derivation in the supplement or an explicit reference, and ideally a direct numerical check of the local metric determinant against the formula at several reference points. The skeptic's requested numerical verification is appropriate and would substantially strengthen the paper.
  3. [Properties, Eqs. (14)–(16)] The resolution-of-identity calculation contains an algebraic error. For a fixed state, Haar averaging over all local unitaries gives I_{d^N}/d^N, so the equality in Eq. (15) should read (1/d^N)∫dμ_FS(Ψ) Tr(|Ψ⟩⟨Ψ|) I_{d^N} = (1/d^N) I_{d^N} ∫dμ_FS(Ψ), and the coefficient in Eq. (16) is Z/d^N rather than Z. As written, the statement that the volume is proportional to the identity with coefficient Z is dimensionally inconsistent.
  4. [MPS Measure, Eq. (6) and global extension] The pointwise argument "since we could have chosen any point as the reference" is sound only if the exponential parameterization Eq. (6) is a local chart at every point of the gauge-fixed manifold. The paper asserts this via Ref. [43] but does not state the gauge condition under which the chart is valid, nor the boundary condition on the right environment (Γ_N, which appears to be I_D from the recurrence but is never declared). Please make these assumptions explicit and state that the formula extends to degenerate points by continuity; otherwise the global statement of Eq. (12) is not fully supported.
minor comments (3)
  1. [Supplement Eq. (44)–(46)] The proof of free independence of A and B is sketched rather than completed; the crucial step "this guarantees..." is an assertion. If this is to stand as a proof, more details are needed. Since the FS spectrum is described as approximate, the heuristic character should also be made explicit for the RMPS fixed point, which is currently presented as proven.
  2. [Matrix product states, Eqs. (2)–(3)] Equation (2) is incomplete: A[i] = U[i] 0 should define A[i] as the first D columns of U[i], and Eq. (3) is garbled and should read Γ_{i−1} = ∑_σ A^{[i]}_σ Γ_i (A^{[i]}_σ)†. Please fix the notation.
  3. [Sampling, Eq. (17)] The acceptance ratio in Eq. (17) should specify that proposals are drawn from a symmetric distribution on the unitary U_j; if proposals are instead made in the x coordinates, the Jacobian J(x,x) must be included. Please clarify.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central measure correction is derived from the induced Fubini-Study metric with no fitted inputs or same-author uniqueness assumptions.

full rationale

The central claim, Eq. (12), is obtained by comparing two measures expressed in a common gauge-fixed coordinate system. The RMPS density is computed from the Haar measure through the Jacobian of the exponential parameterization (Eqs. 6-8), while the Fubini-Study density is computed from the determinant of the induced metric (Eqs. 9-11). The ratio is then extended to the whole manifold by the reference-point argument. None of these steps imports the target relation as an input, and no fitted parameter or empirical constant enters the derivation. The self-citations that appear, such as Refs. [25], [26], [38], [39], and [41], are used only as examples of applications or as motivational context, not as evidence for Eq. (12); the external citations used for the geometry and parameterization, Refs. [40] and [43], do not overlap with the present authors. The approximate entanglement-spectrum prediction c' = 1/(2d-1) is explicitly described in the paper as 'not a self-consistent solution' and 'a very crude approximation,' so it is not presented as a forced or fitted consequence. The unproven local metric-determinant formula, Eq. (10), is a verification and correctness concern rather than a circularity, because it is not identical by construction to the global measure relation being claimed.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on external geometric results, a sketched metric determinant calculation, and a large-D free-independence argument. No numerical constants are fitted; the correction factor and the c' = 1/(2d-1) spectrum are parameter-free predictions. No new entities are introduced.

assumptions (5)
  • standard math The exponential parameterization U(x) = U0 exp of the block matrix in Eq. (6) is surjective onto the gauge-fixed MPS manifold for every site.
    Invoked through citation [43]; it guarantees the coordinate chart covers every MPS, but the paper does not prove it.
  • domain assumption The induced Fubini-Study metric is block diagonal and its determinant is |g| = product over sites of |Gamma_i|^{D(d-1)}.
    The text says this follows from standard MPS contraction calculations but does not show the calculation; it is load-bearing for the central relation Eq. (12).
  • standard math Standard Haar measure and complex Wishart results give the Jacobian J(x,x) in Eq. (8).
    The supplement derives this using polar decompositions and Wishart measure theory from Mehta [44], but the proof is a sketch.
  • domain assumption The transfer operator on right-environment distributions has a finite correlation length and a non-degenerate leading eigenvalue.
    Used in Supplement Eq. (35) for the typical spectrum argument; no proof of the spectral gap is given.
  • domain assumption The matrices A and B in the free-probability argument are freely independent in the large-D limit.
    The supplement says this is 'easily checked' for d=2 and 'easily extended' to general d, but the proof is only sketched.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An unbiased measure over the matrix product state manifold." pith.science (2026). https://pith.science/paper/GT23LEBD

@misc{pith2026250500073,
  author       = {Pith},
  title        = {Pith review of: An unbiased measure over the matrix product state manifold},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GT23LEBD}},
  note         = {Machine review of arXiv:2505.00073}
}
read the original abstract

Matrix product states are useful representations for a large variety of naturally occurring quantum states. Studying their typical properties is important for understanding universal behavior, including quantum chaos and thermalization, as well as the limits of classical simulations of quantum devices. We show that the usual ensemble of sequentially generated random matrix product states (RMPS) using local Haar random unitaries is not uniform when viewed as a restriction of the full Hilbert space. As a result, the entanglement across the chain exhibits an anomalous asymmetry under spatial inversion. We show how to construct an unbiased measure starting from the left-canonical form and design a Metropolis algorithm for sampling random states. Some properties of this new ensemble are investigated both analytically and numerically, such as the resulting resolution of identity over matrix product states and the typical entanglement spectrum, which is found to differ from the sequentially generated case.

Figures

Figures reproduced from arXiv: 2505.00073 by the authors.

Figure 3
Figure 3. FIG. 3: Eigenvalue histogram of 1000 random right [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Cumulative distribution function of the entan [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 26 canonical work pages

  1. [43]

    Wouters, N

    S. Wouters, N. Nakatani, D. Van Neck, and G. K.-L. Chan, Phys. Rev. B 88, 075122 (2013)

    work page 2013

  2. [1]

    S. R. White, Phys. Rev. Lett. 69, 2863 (1992)

    1992

  3. [2]

    S. R. White, Phys. Rev. B 48, 10345 (1993)

    1993

  4. [3]

    Schollw¨ ock, Rev

    U. Schollw¨ ock, Rev. Mod. Phys.77, 259 (2005)

    2005

  5. [4]

    Schollwck, Annals of Physics 326, 96 (2011), january 2011 Special Issue

    U. Schollwck, Annals of Physics 326, 96 (2011), january 2011 Special Issue

    work page 2011

  6. [5]

    J. I. Cirac and F. Verstraete, Journal of Physics A: Math- ematical and Theoretical 42, 504004 (2009)

    2009

  7. [6]

    Ors, Annals of Physics 349, 117 (2014)

    R. Ors, Annals of Physics 349, 117 (2014)

    work page 2014

  8. [7]

    V. M. F. Verstraete and J. Cirac, Advances in Physics 57, 143 (2008)

    work page 2008

Show all 45 references
  1. [8]

    S. R. White and R. L. Martin, The Journal of Chemical Physics 110, 4127 (1999)

    1999

  2. [9]

    Baiardi and M

    A. Baiardi and M. Reiher, The Journal of Chemical Physics 152, 040903 (2020)

    2020

  3. [10]

    Szalay, M

    S. Szalay, M. Pfeffer, V. Murg, G. Barcza, F. Verstraete, R. Schneider, and . Legeza, International Journal of Quantum Chemistry 115, 1342 (2015)

    2015

  4. [11]

    I. L. Markov and Y. Shi, SIAM Journal on Computing 38, 963 (2008)

    2008

  5. [12]

    Pan and P

    F. Pan and P. Zhang, Phys. Rev. Lett. 128, 030501 (2022)

    2022

  6. [13]

    Stoudenmire and D

    E. Stoudenmire and D. J. Schwab, in Advances in Neu- ral Information Processing Systems , Vol. 29, edited by D. Lee, M. Sugiyama, U. Luxburg, I. Guyon, and R. Gar- nett (Curran Associates, Inc., 2016)

    2016

  7. [14]

    Rieser, F

    H.-M. Rieser, F. Kster, and A. P. Raulf, Proceedings of the Royal Society A: Mathematical, Physical and Engi- neering Sciences 479, 20230218 (2023)

    2023

  8. [15]

    T. Hao, X. Huang, C. Jia, and C. Peng, Frontiers in Physics 10 (2022), 10.3389/fphy.2022.906590

    2022

  9. [16]

    Tensor networks based quantum optimiza- tion algorithm,

    V. Akshay, A. Melnikov, A. Termanova, and M. R. Perelshtein, “Tensor networks based quantum optimiza- tion algorithm,” (2024), arXiv:2404.15048 [quant-ph]

    2024 arXiv

  10. [17]

    Verstraete and J

    F. Verstraete and J. I. Cirac, Phys. Rev. B 73, 094423 (2006)

    2006

  11. [18]

    Skinner, J

    B. Skinner, J. Ruhman, and A. Nahum, Physical Review X 9 (2019), 10.1103/physrevx.9.031009

    2019 doi

  12. [19]

    Y. Li, X. Chen, and M. P. A. Fisher, Physical Review B 100 (2019), 10.1103/physrevb.100.134306

    2019 doi

  13. [20]

    Constant-depth preparation of matrix prod- uct states with adaptive quantum circuits,

    K. C. Smith, A. Khan, B. K. Clark, S. M. Girvin, and T.-C. Wei, “Constant-depth preparation of matrix prod- uct states with adaptive quantum circuits,” (2024), arXiv:2404.16083 [quant-ph]

    2024 arXiv

  14. [21]

    E. V. Doggen, I. V. Gornyi, A. D. Mirlin, and D. G. Polyakov, Annals of Physics 435, 168437 (2021), special Issue on Localisation 2020

    2021

  15. [22]

    Serbyn, D

    M. Serbyn, D. A. Abanin, and Z. Papi´ c, Nature Physics 17, 675 (2021)

    2021

  16. [23]

    Chirco, D

    G. Chirco, D. Oriti, and M. Zhang, Phys. Rev. D 97, 126002 (2018)

    2018

  17. [25]

    Leontica and A

    S. Leontica and A. G. Green, Nature Communications 16, 1832 (2025)

    2025

  18. [26]

    Hallam, J

    A. Hallam, J. G. Morley, and A. G. Green, Nature Com- munications 10, 2708 (2019)

    2019

  19. [27]

    Arute, K

    F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, R. Barends, R. Biswas, S. Boixo, F. G. S. L. Brandao, D. A. Buell, et al. , Nature 574, 505 (2019)

    2019

  20. [28]

    Y. Zhou, E. M. Stoudenmire, and X. Waintal, Phys. Rev. X 10, 041038 (2020)

    2020

  21. [29]

    Ayral, T

    T. Ayral, T. Louvet, Y. Zhou, C. Lambert, E. M. Stoudenmire, and X. Waintal, PRX Quantum 4, 020304 6 (2023)

    2023

  22. [30]

    Garnerone, T

    S. Garnerone, T. R. de Oliveira, and P. Zanardi, Phys. Rev. A 81, 032336 (2010)

    2010

  23. [31]

    Garnerone, T

    S. Garnerone, T. R. de Oliveira, S. Haas, and P. Za- nardi, Physical Review A 82 (2010), 10.1103/phys- reva.82.052312

    2010 doi

  24. [32]

    Haferkamp, C

    J. Haferkamp, C. Bertoni, I. Roth, and J. Eisert, PRX Quantum 2, 040308 (2021)

    2021

  25. [33]

    D. Haag, F. Baccari, and G. Styliaris, PRX Quantum 4, 030330 (2023)

    2023

  26. [34]

    Lancien and D

    C. Lancien and D. Prez-Garca, Annales Henri Poincar 23, 141222 (2021)

    2021

  27. [35]

    L. Chen, R. J. Garcia, K. Bu, and A. Jaffe, Phys. Rev. B 109, 174207 (2024)

    2024

  28. [36]

    G. Lami, J. De Nardis, and X. Turkeshi, Phys. Rev. Lett. 134, 010401 (2025)

    2025

  29. [37]

    Universality in the anticoncentration of chaotic quantum circuits,

    A. Sauliere, B. Magni, G. Lami, X. Turkeshi, and J. D. Nardis, “Universality in the anticoncentration of chaotic quantum circuits,” (2025), arXiv:2503.00119 [quant-ph]

    2025 arXiv

  30. [38]

    Feynman path integrals over entangled states,

    A. G. Green, C. A. Hooley, J. Keeling, and S. H. Simon, “Feynman path integrals over entangled states,” (2016), arXiv:1607.01778 [cond-mat.str-el]

    2016 arXiv

  31. [39]

    F. Azad, A. Hallam, J. Morley, and A. G. Green, Scien- tific Reports 13, 8866 (2023)

    2023

  32. [40]

    Haegeman, T

    J. Haegeman, T. J. Osborne, and F. Verstraete, Phys. Rev. B 88, 075133 (2013)

    2013

  33. [41]

    Zero-point energy of tensor fluctuations on the mps manifold,

    S. Leontica and A. G. Green, “Zero-point energy of tensor fluctuations on the mps manifold,” (2024)

    2024

  34. [42]

    Movassagh and J

    R. Movassagh and J. Schenker, Communications in Mathematical Physics 395, 1175 (2022)

    2022

  35. [44]

    Mehta, Random Matrices , Pure and Applied Mathe- matics (Academic Press, 2004)

    M. Mehta, Random Matrices , Pure and Applied Mathe- matics (Academic Press, 2004)

    2004

  36. [45]

    V. A. Marchenko and L. A. Pastur, Mathematics of the USSR-Sbornik 1, 457 (1967)

    1967

  37. [46]

    Cecile, H

    G. Cecile, H. L´ oio, and J. De Nardis, Phys. Rev. Res. 6, 033220 (2024). I. JACOBIAN OF EXPONENTIAL P ARAMETERIZA TION In this section we consider the Jacobian of the mapping from the ( d− 1)D×D complex matrix x to the on-site unitaries U(x) under the exponential mapping give...

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.