Pith. sign in

REVIEW 2 major objections 5 minor 80 references

For any n-fermion state on m modes, the paper gives classical and quantum algorithms that return a Slater determinant within ε of the maximum fidelity in time m^{poly(n,1/ε)}, and proves a sharp 2/3 threshold above which stationary points a

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 09:50 UTC pith:P44UIHDW

load-bearing objection Solid, citable paper on closest-Slater learning with a clean 2/3 threshold; watch the classical access model and a couple of overclaims in the intro. the 2 major comments →

arxiv 2607.20623 v1 pith:P44UIHDW submitted 2026-07-22 quant-ph cond-mat.dis-nncond-mat.str-el

Learning the closest Slater determinant

Nisarga Paul , Haimeng Zhao , David D. Dai This is my paper
classification quant-ph cond-mat.dis-nncond-mat.str-el MSC 68Q1268Q1781P68
keywords Slater determinantfermionic fidelityagnostic tomographyoptimization landscape2/3 thresholdquantum threshold searchFermi-Hubbard modelnatural orbitals
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper asks how hard it is to find the single Slater determinant that best approximates a given fermionic many-body state, with closeness measured by fidelity. It answers with classical and quantum algorithms whose runtime is m^{poly(n,1/ε)} — polynomial in the number of modes for fixed particle number and accuracy — and a quantum protocol using only poly(m,n,1/ε) copies. It also proves matching hardness lower bounds along the particle-number and precision axes, assuming standard complexity conjectures, so the exponential dependence on n and 1/ε cannot be removed in general. On the geometry side, it shows that any stationary point of the fidelity above 2/3 is the unique globally closest Slater determinant, and that 2/3 is sharp: below it spurious stationary points exist. The same tools are applied to ground states of the Fermi-Hubbard model, including neural quantum state representations, where the closest-Slater fidelity tracks interaction strength.

Core claim

The central claim is that the closest-Slater problem is tractable in the number of modes m at fixed particle number n and accuracy ε. The classical algorithm estimates the single-particle reduced density matrix, truncates to an active space of dimension O(n²/ε), places an ε-covering net over the corresponding Slater manifold, and searches it; the quantum variant runs the same net search using quantum threshold search, needing poly(m,n,1/ε) copies. Hardness reductions show this is essentially optimal in 1/ε for both settings and in n for the classical setting. The second discovery is the optimization-landscape transition: if a Slater determinant is a stationary point of fidelity and has fidel

What carries the argument

The algorithmic engine is active-space truncation plus an ε-covering net: after diagonalizing an estimate of the 1-RDM, only orbitals with occupation above O(ε/n) are kept, shrinking the search from a manifold of dimension O(n(m−n)) to one of dimension O(n²/ε); a net of size exp[O((n³/ε) log(n/ε))] then makes exhaustive search possible, and in the quantum case a threshold-search routine turns the net search into poly(n,1/ε) sample complexity. The landscape proof relies on particle-hole decomposition: any Slater relative to a trial Slater splits into zero, one, and multi particle-hole components, and a bound q ≤ sqrt(2(1−s)) on the multi-particle-hole norm, together with the vanishing of sing

Load-bearing premise

The classical runtime guarantee depends on having a black box that can estimate the single-particle occupancy matrix to error O(ε/n) and evaluate fidelity for any explicitly given Slater determinant; for real tensor-network or neural states this sampling cost is not included in the advertised m^{poly(n,1/ε)} runtime.

What would settle it

Take the adversarial states constructed in the sharpness proof — superpositions of a Slater determinant and a two-particle-hole-excited Slater engineered so a stationary point sits just below 2/3. If a stationary Slater with fidelity above 2/3 is ever found not to be the global maximum, the landscape theorem fails. Alternatively, a classical algorithm solving the problem in m^{o(n)} time for all inputs would contradict the paper's lower bound, as would a poly(m,1/ε)-time algorithm for precision 1/ε.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • For any tensor network or neural quantum state with efficient amplitude access, the closest Slater approximation can be found in time m^{poly(n,1/ε)}, making extraction of a natural single-particle orbital basis systematic rather than heuristic.
  • A heuristic optimizer that reaches fidelity above 2/3 is guaranteed to be at the global optimum; below 2/3 the same heuristic can get stuck at spurious stationary points.
  • The quantum algorithm can benchmark a fermionic simulator with poly(m,n,1/ε) copies, independent of the simulation cost of preparing the state.
  • Under standard complexity conjectures, no classical algorithm can avoid exponential dependence in n, and no classical or quantum algorithm can achieve polynomial dependence in 1/ε, so the provided algorithms are essentially optimal along these axes.
  • For the Fermi-Hubbard model, OPT decreases monotonically with |U| and drops faster for attractive interactions, quantifying how correlations degrade the best single-determinant description.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The 2/3 threshold is a worst-case bound; the numerics suggest that for physical ground states spurious stationary points sit far below it, so a lower, state-dependent threshold may often certify optimality in practice.
  • Because the algorithm's active space and net search return both the optimal Slater and the natural orbitals defining it, the method doubles as a systematic way to define the optimal single-particle basis and particle-hole excitations for a strongly correlated state.
  • A natural extension, which the paper leaves open, is the same active-space-plus-covering strategy for the closest fermionic Gaussian state; if the manifold admits a similar covering, superconducting analogues would be accessible.
  • Testable extension: compute closest-Slater fidelity for larger repulsive and attractive Hubbard lattices to check whether the observed monotone dependence on U persists and, if so, use OPT as a cheap correlation measure.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper studies the problem of finding the Slater determinant with maximum fidelity to a given n-fermion state on m modes, in both classical and quantum access models. The main algorithmic results are a classical algorithm with runtime m^{poly(n,1/ε)} and a quantum algorithm with poly(m,n,1/ε) sample complexity, both returning a Slater determinant within ε of optimal fidelity. The paper also proves hardness lower bounds: exponential dependence on n is classically hard under an ETH-type assumption, and exponential dependence on 1/ε is hard for both classical and quantum algorithms (assuming NP is not in BPP/BQP) via a reduction from quantum separability. A separate landscape result shows a sharp threshold at fidelity 2/3: above it, any stationary Slater determinant is the unique global optimum; below it, spurious stationary points can exist, with an explicit adversarial construction. Numerical experiments on Fermi-Hubbard ground states (exact diagonalization and neural quantum states) and on various random/structured ensembles are used to benchmark gradient ascent and to probe the optimization landscape.

Significance. If the results hold as stated, they constitute a substantial contribution to agnostic tomography and fermionic learning. The 2/3 threshold is a clean, non-obvious landscape result with a constructive sharpness proof. The hardness reductions are carefully designed and connect the problem to well-studied conjectures (ETH, quantum separability). The paper is transparent about many of its assumptions, and the proofs are presented in enough detail that the core theorems (1, 2, 5, 6 and the separability reduction in App. A.2) can be independently verified; the algebra in these proofs checks out. The use of quantum threshold search to avoid a linear scan over the covering net is elegant. However, the advertised classical runtime is only proven under an oracle model that is not costed for the motivating state classes; this gap affects the scope of the central claim and needs to be addressed.

major comments (2)
  1. [Sec. II (Theorem 1) and App. B (Definition 1)] The classical runtime guarantee m^{poly(n,1/ε)} is conditional on an access model that provides (i) an estimate of the 1-RDM to operator-norm error O(ε/n) and (ii) a bounded unbiased estimator Y_S of ⟨S|ρ|S⟩ for every explicitly specified Slater determinant |S⟩. The paper does not establish that these oracles are efficiently implementable for the motivating input classes. For a generic circuit computing amplitudes, ⟨S|ψ⟩ = Σ_I det(U[I,:])⟨I|ψ⟩ is a #P-hard quantity; for a tensor network it may be efficient, but for a neural quantum state the estimator used in App. D is not Y_S. Without a proof or explicit cost analysis of the oracle, the abstract's unconditional phrasing ('given an n-fermion wavefunction ... we provide classical ... algorithms') overstates what is proven. The theorem itself is sound under the stated assumptions, but the central claim's applicability is narrower than adve
  2. [Sec. VI and App. D (numerics with NQS)] The numerical demonstration for neural quantum states uses the fidelity estimator \hat F = |r|^2/|r|^2 with r(x)=⟨x|S⟩/ψ(x), a ratio estimator that is not the bounded unbiased Y_S of Definition 1. No variance bound or bias correction is provided, and the estimator can diverge where ψ(x) is small on the support of S. Thus the NQS experiments do not instantiate the access model assumed in Theorem 1/Proposition 1, and cannot be taken as evidence that the algorithm runs in the advertised time for NQS inputs. The paper should either supply a correct sampling protocol with bounded variance for NQS, or explicitly state that the numerical results use a heuristic estimator and do not verify the theoretical runtime.
minor comments (5)
  1. [Abstract] 'matching hardness lower bounds' is slightly misleading: the n-axis lower bound is for the classical model, and the 1/ε lower bound is for mixed states with n=2; the quantum algorithm's exponential-in-n runtime is not matched by a lower bound, a gap the paper acknowledges. Consider qualifying the phrase.
  2. [Sec. II, Eq. (18)] The Lipschitz bound |F(U)-F(V)| ≤ 2n∥U-V∥_F is central to the net construction. It would help to state explicitly that the Frobenius norm is on the r×n matrix and that the constant 2n is sufficient for the subsequent δ choice; the current proof is correct but somewhat terse.
  3. [App. D, NQS estimator] The ratio estimator \hat F = |r|^2/|r|^2 is generally biased. The paper mentions the width of the Monte Carlo standard deviation but does not discuss bias. Since this estimator is used to claim agreement with exact diagonalization, a bias assessment would strengthen the numerics.
  4. [Sec. VI, Fig. 4(b)] The claim that 'a random Slater has fidelity of order one over the many body Hilbert space dimension' is heuristic; a precise statement of the expected fidelity of a random Slater versus a generic state would make the interpretation of the gradient-ascent stall more rigorous.
  5. [References] Reference [71] is listed as 'Work in progress' and is not a citable publication. If this is the concurrent work mentioned in the Note added, consider citing a stable version (arXiv preprint) or a more formal description.

Circularity Check

0 steps flagged

No significant circularity: the algorithms and landscape results are derived from explicit assumptions, not from their conclusions.

full rationale

The paper's derivation chain is self-contained under its stated access models. Theorem 1 assumes only a 1-RDM estimator and query access to Slater fidelities, then proves an additive-error guarantee by active-space truncation (Lemmas 4–5) and an exhaustive covering-net search; the target OPT is not an input, and the proof explicitly shows that the 1-RDM alone does not determine the closest Slater (Eq. (21) counterexample). Theorem 2 similarly reduces to fermionic classical shadow estimation plus agnostic quantum threshold search, neither of which presupposes the maximizing Slater. The 2/3 landscape result (Theorem 5) is a genuine analytical bound using a particle-hole decomposition and Lemma 8, which is proved in the appendix; the sharpness construction (Theorem 6) is an explicit counterexample, not a restatement of the theorem. Numerics use OPT as ground truth to benchmark gradient ascent, which is an application rather than a fit. The only citations involving a present author ([18], [33], [60]) are for related work or supporting technical tools; the load-bearing shadow bound is attributed to external Ref. [31] and re-derived in Lemma 5 via matrix Bernstein, so no central claim rests on an unverified self-citation. The skeptic's concern that the oracle in Definition 1 may be hard to realize for neural quantum states is a validity/implementation caveat about the assumption, not circularity: the theorems state their assumptions and do not smuggle in the target result.

Axiom & Free-Parameter Ledger

3 free parameters · 7 axioms · 0 invented entities

No free parameters are fitted to data; ε, δ, and the active-space cutoff τ̂ are algorithm tolerances selected in proofs, not model parameters. No new physical entities are invented. The main non-trivial inputs are external: fermionic classical shadows, quantum threshold search, and two complexity-theoretic conjectures used only for lower bounds.

free parameters (3)
  • active-space occupation threshold τ̂ = 3ε/4n (proof choice)
    Hand-chosen to balance the truncation error (τ̂+η ≤ ε/2n) and the active-space dimension bound (n/(τ̂−η) = O(n²/ε)); it is an algorithm constant, not fitted to data.
  • covering net radius δ = ε/4n (proof choice)
    Set to make the net fidelity error ε/2; algorithm constant, not fitted.
  • amplitude ratio α in Theorem 3 hardness reduction = >1, chosen so α²/(1+α²) > λ
    Part of the lower-bound construction to satisfy the promise OPT≥λ; not fitted to data.
axioms (7)
  • domain assumption Assumption 1: no randomized f(n)p^{o(n)} algorithm for n-Multicolored Clique with unique-clique promise
    Used in Theorem 3 to rule out classical subexponential-in-n algorithms; standard ETH-based parameterized hardness.
  • domain assumption NP-hardness of quantum separability with inverse-polynomial promise gap (Gharibian 2010)
    Used in Theorem 4 to rule out poly(m,1/ε) algorithms; external hardness result.
  • standard math Fermionic classical shadows yield an m×m 1-RDM estimator with operator-norm error η from O(m² log(m/δ)/η²) copies
    Relies on prior work [31,33]; used for quantum sample complexity.
  • standard math Quantum threshold search of Badescu-O'Donnell and its agnostic/binary-search extension (Lemma 9, App. C)
    Core subroutine for quantum algorithm; proof given in appendix.
  • standard math The Slater manifold admits a δ-covering net of size (c/δ)^{2n(r−n)} (Szarek)
    Basis of the covering-net search.
  • domain assumption Classical access model (Definition 1): efficient sampling estimators for few-body observables and for fidelities of specified Slaters
    Without this, the classical m^{poly(n,1/ε)} runtime claim is not grounded.
  • domain assumption Quantum access model: ability to implement two-outcome measurements {|S⟩⟨S|, I−|S⟩⟨S|} for arbitrary specified Slaters
    Needed to feed the net projectors into quantum threshold search.

pith-pipeline@v1.3.0-alltime-deepseek · 35708 in / 34589 out tokens · 250736 ms · 2026-08-01T09:50:11.036402+00:00 · methodology

0 comments
read the original abstract

Learning compact, interpretable descriptions of quantum many-body states is an important task in quantum science. We study the task of learning the Slater determinant with maximum fidelity to an arbitrary fermionic many-body state, with motivation from both Hartree-Fock methods and agnostic tomography. Given an $n$-fermion wavefunction built from $m$ fermionic modes, we provide classical and quantum algorithms returning a Slater determinant with fidelity within $\varepsilon$ of maximal in time $m^{\text{poly}(n,1/\varepsilon)}$. We prove matching hardness lower bounds, assuming standard complexity conjectures, along some parameter axes. Given access to quantum copies, we prove this can be accomplished with $\text{poly}(m,n,1/\varepsilon)$ copies of $\rho$. We also show that above a fidelity of $2/3$ any stationary point is the unique global maximum while below $2/3$ the optimization landscape can have spurious stationary points, and hence $2/3$ marks a transition point in the optimization landscape for this problem. We apply the algorithm to the Fermi-Hubbard model, extracting the closest Slater determinant from neural quantum state solutions. Together, our results provide algorithmic tools with provable guarantees in understanding fermionic many-body systems with classical or quantum simulation.

Figures

Figures reproduced from arXiv: 2607.20623 by David D. Dai, Haimeng Zhao, Nisarga Paul.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

80 extracted references · 4 linked inside Pith

  1. [1]

    J. C. Slater, Phys. Rev.34, 1293 (1929)

    1929

  2. [2]

    Bruus and K

    H. Bruus and K. Flensberg,Many-body quantum theory in condensed matter physics: an introduction(Oxford university press, 2004)

    2004

  3. [3]

    S. M. Girvin and K. Yang,Modern condensed matter physics(Cambridge University Press, 2019)

    2019

  4. [4]

    Lykos and G

    P. Lykos and G. W. Pratt, Rev. Mod. Phys.35, 496 (1963)

    1963

  5. [5]

    Echenique and J

    P. Echenique and J. L. Alonso, Molecular Physics105, 3057 (2007)

    2007

  6. [6]

    Kohn, Reviews of modern physics71, 1253 (1999)

    W. Kohn, Reviews of modern physics71, 1253 (1999)

    1999

  7. [7]

    Shavitt and R

    I. Shavitt and R. J. Bartlett,Many-body methods in chem- istry and physics: MBPT and coupled-cluster theory(Cam- bridge University Press, 2009)

    2009

  8. [8]

    Zhang and M

    J. Zhang and M. Kollar, Physical Review A89, 012504 (2014)

    2014

  9. [9]

    Zhang and N

    J.-M. Zhang and N. J. Mauser, Physical Review A94, 032513 (2016)

    2016

  10. [10]

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

    1992

  11. [11]

    Schollw¨ ock, Annals of physics326, 96 (2011)

    U. Schollw¨ ock, Annals of physics326, 96 (2011)

    2011

  12. [12]

    Carleo and M

    G. Carleo and M. Troyer, Science355, 602 (2017)

    2017

  13. [13]

    L¨ owdin, Phys

    P.-O. L¨ owdin, Phys. Rev.97, 1474 (1955)

    1955

  14. [14]

    B. O. Roos, P. R. Taylor, and P. E. Sigbahn, Chemical Physics48, 157 (1980)

    1980

  15. [15]

    D. J. Thouless, Nuclear Physics21, 225 (1960)

    1960

  16. [16]

    Grewal, V

    S. Grewal, V. Iyer, W. Kretschmer, and D. Liang, Quan- tum10, 2027 (2026)

    2027

  17. [17]

    S. Chen, W. Gong, Q. Ye, and Z. Zhang, inProceed- ings of the 57th Annual ACM Symposium on Theory of Computing(2025) pp. 429–438

    2025

  18. [18]

    H. Zhao, L. Lewis, I. Kannan, Y. Quek, H.-Y. Huang, and M. C. Caro, PRX Quantum5, 040306 (2024)

    2024

  19. [19]

    Wadhwa, L

    C. Wadhwa, L. Lewis, E. Kashefi, and M. Doosti, PRX Quantum6, 040371 (2025)

    2025

  20. [20]

    Bakshi, J

    A. Bakshi, J. Bostanci, W. Kretschmer, Z. Landau, J. Li, A. Liu, R. O’Donnell, and E. Tang, inProceedings of the 57th Annual ACM Symposium on Theory of Computing (Association for Computing Machinery, 2025) pp. 1212– 1221

    2025

  21. [21]

    Wei and P

    T.-C. Wei and P. M. Goldbart, Physical Review A68, 042307 (2003)

    2003

  22. [22]

    Vedral, M

    V. Vedral, M. B. Plenio, M. A. Rippin, and P. L. Knight, Physical Review Letters78, 2275 (1997)

    1997

  23. [23]

    Badescu and R

    C. Badescu and R. O’Donnell, inProceedings of the 53rd Annual ACM SIGACT Symposium on Theory of Comput- ing(2021) pp. 1398–1411

    2021

  24. [24]

    Cygan, F

    M. Cygan, F. V. Fomin, /suppress L. Kowalik, D. Lokshtanov, D. Marx, M. Pilipczuk, M. Pilipczuk, and S. Saurabh, Parameterized Algorithms(Springer, 2015)

    2015

  25. [25]

    Gharibian, Quantum Information & Computation10, 343 (2010)

    S. Gharibian, Quantum Information & Computation10, 343 (2010)

    2010

  26. [26]

    Brillouin, Journal de Physique et le Radium4, 1 (1933)

    L. Brillouin, Journal de Physique et le Radium4, 1 (1933)

    1933

  27. [27]

    Y. A. Aoto and M. F. da Silva, Phys. Rev. A102, 052803 (2020)

    2020

  28. [28]

    Aaronson and S

    S. Aaronson and S. Grewal, in18th Conference on the Theory of Quantum Computation, Communication and Cryptography (TQC 2023), Leibniz International Proceed- ings in Informatics (LIPIcs), Vol. 266, edited by O. Fawzi and M. Walter (Schloss Dagstuhl – Leibniz-Zentrum f¨ ur Informatik, Dagstuhl, Germany, 2023) pp. 12:1–12:18

    2023

  29. [29]

    O’Gorman, arXiv preprint arXiv:2207.14787 (2022), arXiv:2207.14787 [quant-ph]

    B. O’Gorman, arXiv preprint arXiv:2207.14787 (2022), arXiv:2207.14787 [quant-ph]

    Pith/arXiv arXiv 2022

  30. [30]

    Huang, R

    H.-Y. Huang, R. Kueng, and J. Preskill, Nature Physics 16, 1050 (2020)

    2020

  31. [31]

    A. Zhao, N. C. Rubin, and A. Miyake, Physical Review Letters127, 110504 (2021)

    2021

  32. [32]

    Bittel, A

    L. Bittel, A. A. Mele, J. Eisert, and L. Leone, PRX Quantum6, 030341 (2025)

    2025

  33. [33]

    Christensen and A

    A. Christensen and A. Zhao, arXiv preprint arXiv:2602.05058 (2026)

    arXiv 2026

  34. [34]

    Bittel, A

    L. Bittel, A. A. Mele, J. Eisert, and L. Leone, Quantum 9, 1665 (2025)

    2025

  35. [35]

    A. D. Gottlieb and N. J. Mauser, Physical review letters 95, 123003 (2005)

    2005

  36. [36]

    A. D. Gottlieb and N. J. Mauser, International Journal of Quantum Information5, 815 (2007)

    2007

  37. [37]

    Benatti, R

    F. Benatti, R. Floreanini, and U. Marzolino, Physical Review A—Atomic, Molecular, and Optical Physics85, 042329 (2012). 13

    2012

  38. [38]

    C. J. Turner, K. Meichanetzidis, Z. Papi´ c, and J. K. Pachos, Nature communications8, 14926 (2017)

    2017

  39. [39]

    Hebenstreit, R

    M. Hebenstreit, R. Jozsa, B. Kraus, S. Strelchuk, and M. Yoganathan, Physical review letters123, 080503 (2019)

    2019

  40. [40]

    Dias and R

    B. Dias and R. Koenig, Quantum8, 1350 (2024)

    2024

  41. [41]

    Lyu and K

    X. Lyu and K. Bu, arXiv preprint arXiv:2409.08180 (2024)

    Pith/arXiv arXiv 2024

  42. [42]

    Coffman, G

    L. Coffman, G. Smith, and X. Gao, arXiv preprint arXiv:2501.06179 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    F. Ares, M. Mazzoni, S. Murciano, D. Sz´ asz- Schagrin, P. Calabrese, and L. Piroli, arXiv preprint arXiv:2603.16762 (2026)

    arXiv 2026

  44. [44]

    Sierant, P

    P. Sierant, P. Stornati, and X. Turkeshi, PRX Quantum 7, 010302 (2026)

    2026

  45. [45]

    Leone, S

    L. Leone, S. F. Oliviero, and A. Hamma, Physical Review Letters128, 050402 (2022)

    2022

  46. [46]

    Gigena and R

    N. Gigena and R. Rossignoli, Physical Review A92, 042326 (2015)

    2015

  47. [47]

    T. I. Vanhala and T. Ojanen, Physical Review Research 6, 023178 (2024)

    2024

  48. [48]

    P. S. Tarabunga, B. Jobst, R. Morral-Yepes, M. Langer, B. Kraus, F. Pollmann, and S.-H. Lin, Computable mea- sures of fermionic non-gaussianity from the covariance matrix (2026), arXiv:2607.02242 [quant-ph]

    Pith/arXiv arXiv 2026

  49. [49]

    Veryazov, P

    V. Veryazov, P. A. Malmqvist, and B. O. Roos, Interna- tional Journal of Quantum Chemistry111, 3329 (2011)

    2011

  50. [50]

    C. J. Stein and M. Reiher, Journal of chemical theory and computation12, 1760 (2016)

    2016

  51. [51]

    E. R. Sayfutyarova, Q. Sun, G. K.-L. Chan, and G. Knizia, Journal of chemical theory and computation13, 4063 (2017)

    2017

  52. [52]

    Ma and H

    Y. Ma and H. Ma, The Journal of Chemical Physics138 (2013)

    2013

  53. [53]

    S. J. Szarek, inProceedings of research workshop on Ba- nach space theory (Iowa City, Iowa, 1981), Vol. 169 (Uni- versity of Iowa Iowa City, IA, 1982) p. 185

    1981

  54. [54]

    K. Wan, W. J. Huggins, J. Lee, and R. Babbush, Com- munications in Mathematical Physics404, 629 (2023)

    2023

  55. [55]

    L. G. Valiant and V. V. Vazirani, Theoretical Computer Science47, 85 (1986)

    1986

  56. [56]

    Luo and B

    D. Luo and B. K. Clark, Physical review letters122, 226401 (2019)

    2019

  57. [57]

    Cassella, H

    G. Cassella, H. Sutterud, S. Azadi, N. D. Drummond, D. Pfau, J. S. Spencer, and W. M. C. Foulkes, Physical Review Letters130, 036401 (2023)

    2023

  58. [58]

    Chen and M

    A. Chen and M. Heyl, Nature Physics20, 1476 (2024)

    2024

  59. [59]

    Rende, L

    R. Rende, L. L. Viteritti, L. Bardone, F. Becca, and S. Goldt, Communications Physics7, 260 (2024)

    2024

  60. [60]

    H. Zhao, G. Carleo, and F. Vicentini, Quantum8, 1358 (2024)

    2024

  61. [61]

    Y. Teng, D. D. Dai, and L. Fu, Physical Review B111, 205117 (2025)

    2025

  62. [62]

    Geier, K

    M. Geier, K. Nazaryan, T. Zaklama, and L. Fu, Physical Review B112, 045119 (2025)

    2025

  63. [63]

    D. Pfau, J. S. Spencer, A. G. Matthews, and W. M. C. Foulkes, Physical review research2, 033429 (2020)

    2020

  64. [64]

    Hermann, Z

    J. Hermann, Z. Sch¨ atzle, and F. No´ e, Nature Chemistry 12, 891 (2020)

    2020

  65. [65]

    D. Pfau, S. Axelrod, H. Sutterud, I. von Glehn, and J. S. Spencer, Science385, eadn0137 (2024)

    2024

  66. [66]

    Yang and P

    Y. Yang and P. Zhao, Physical Review C107, 034320 (2023)

    2023

  67. [67]

    Adams, G

    C. Adams, G. Carleo, A. Lovato, and N. Rocco, Physical Review Letters127, 022502 (2021)

    2021

  68. [68]

    Gnech, B

    A. Gnech, B. Fore, A. J. Tropiano, and A. Lovato, Physical Review Letters133, 142501 (2024)

    2024

  69. [69]

    B. Fore, J. Kim, M. Hjorth-Jensen, and A. Lovato, Com- munications Physics8, 1 (2025)

    2025

  70. [70]

    A. A. Mele and Y. Herasymenko, PRX Quantum6, 010319 (2025)

    2025

  71. [71]

    Bakshi, J

    A. Bakshi, J. Bostanci, S. Grewal, N. Ju, J. Li, E. Tang, and A. Zhao, Work in progress (2026)

    2026

  72. [72]

    Gurvits, inProceedings of the thirty-fifth annual ACM symposium on Theory of computing(2003) pp

    L. Gurvits, inProceedings of the thirty-fifth annual ACM symposium on Theory of computing(2003) pp. 10–19

    2003

  73. [73]

    J. A. Tropp, Foundations of computational mathematics 12, 389 (2012)

    2012

  74. [74]

    J. A. Tropp, Foundations and trends in machine learning 8, 1 (2015)

    2015

  75. [75]

    Szabo and N

    A. Szabo and N. S. Ostlund,Modern Quantum Chem- istry: Introduction to Advanced Electronic Structure The- ory(Dover Publications, 1996)

    1996

  76. [76]

    Edelman, T

    A. Edelman, T. A. Arias, and S. T. Smith, SIAM journal on Matrix Analysis and Applications20, 303 (1998)

    1998

  77. [77]

    D. P. Arovas, E. Berg, S. A. Kivelson, and S. Raghu, Annual review of condensed matter physics13, 239 (2022)

    2022

  78. [78]

    Vicentini, D

    F. Vicentini, D. Hofmann, A. Szab´ o, D. Wu, C. Roth, C. Giuliani, G. Pescia, J. Nys, V. Vargas-Calder´ on, N. As- trakhantsev,et al., SciPost Physics Codebases , 007 (2022). 14 Appendix A: Computational hardness

    2022

  79. [79]

    no clique

    Classical hardness innfrom parameterized complexity Theorem 3(Classical complexity lower bound).Fix any λ< 1. Under Assumption 1, there exists a constant ελ > 0such that no randomized algorithm running in time f(n)mo(n) can solve the following task in general: given a succinct classical description of a nonzero n-fermion vector|Φ⟩onmmodes, output a Slater...

  80. [80]

    shadow norm

    Hardness in precision from the quantum separability problem The quantum learning algorithm described in The- orem 2 brings the sample complexity down to poly(m,n, 1/ε), but its runtime remains exponential. In this section, we show that this inefficiency is inevitable, by proving that no quantum algorithm can solve the problem to ε = 1/poly(m) accuracy in ...

    2000