pith. sign in

arxiv: 2606.22644 · v1 · pith:Y64RJGEDnew · submitted 2026-06-21 · ❄️ cond-mat.str-el · cond-mat.dis-nn

Neural Polaron: Learning Quasiparticle Operators in Quantum Many-Body Systems

Shang-Shun Zhang This is my paper

Pith reviewed 2026-06-26 09:30 UTC · model grok-4.3

classification ❄️ cond-mat.str-el cond-mat.dis-nn
keywords neural polaronquasiparticle operatorsneural quantum statesJ1-J2 Heisenberg modelmagnon dispersiondynamical responsefrustrated magnetsmany-body excitations
0
0 comments X

The pith

A neural polaron ansatz represents quasiparticle excitations as neural operators acting on a pretrained ground-state wave function.

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

The paper introduces a neural polaron ansatz to compute dynamical properties of quantum many-body systems by learning quasiparticle operators rather than independent excited-state wave functions. A compact neural head parameterizes a local dressing operator on the feature map of a pretrained ground-state network, enforcing translation symmetry, momentum resolution, and quasiparticle locality while separating ground-state correlations from excitation-specific effects. The method is benchmarked on the square-lattice J1-J2 Heisenberg model. It reproduces magnon dispersions and spectral weights across a range of frustration parameters. In particular, the construction captures the (π,0) anomaly and its progressive softening as J2/J1 increases.

Core claim

The neural polaron ansatz, which parameterizes a local dressing operator through a compact neural head on the feature map of a pretrained ground-state network, accurately reproduces magnon dispersions and spectral weights over a broad range of frustration in the square-lattice J1-J2 Heisenberg model, including the (π,0) anomaly and its softening with increasing J2/J1.

What carries the argument

The local dressing operator, a neural many-body operator parameterized by a compact neural head on the feature map of a pretrained ground-state network.

Load-bearing premise

A local dressing operator parameterized through a compact neural head on the feature map of a pretrained ground-state network is sufficient to capture the essential quasiparticle physics without independent excited-state wave functions.

What would settle it

If numerical results from the neural polaron ansatz on the J1-J2 model show that the spectral weight at momentum (π,0) does not soften progressively as J2/J1 is increased, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2606.22644 by Shang-Shun Zhang.

Figure 1
Figure 1. Figure 1: FIG. 1. Schematic of the neural polaron ansatz. A ground [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Exact diagonalization benchmark on a 4 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Momentum-resolved dynamical structure factor [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
read the original abstract

Understanding dynamical properties of quantum many-body systems remains a central challenge because excitations generally require information beyond a ground-state wave function. Here we introduce a neural polaron ansatz that represents quasiparticle excitations by neural many-body operators acting on a correlated ground state. Instead of learning an independent excited-state wave function, the method parameterizes a local dressing operator through a compact neural head defined on the feature map of a pretrained ground-state network. This operator-based construction builds in translation symmetry, momentum resolution, and quasiparticle locality, while separating ground-state correlations from excitation-specific dressing. We benchmark the method on the square-lattice $J_1$-$J_2$ Heisenberg model, where it accurately reproduces magnon dispersions and spectral weights over a broad range of frustration. In particular, it captures nontrivial many-body features including the $(\pi,0)$ anomaly and its progressive softening with increasing $J_2/J_1$. These results establish neural operators as a physically transparent route for extending neural quantum states from ground-state properties to dynamical response.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces a neural polaron ansatz that constructs quasiparticle states via a local dressing operator parameterized by a compact neural head on the feature map of a pretrained ground-state neural network. This operator-based approach is applied to the square-lattice J1-J2 Heisenberg model and is asserted to reproduce magnon dispersions, spectral weights, and the progressive softening of the (π,0) anomaly across a range of J2/J1 without requiring independent excited-state wave functions.

Significance. If validated, the separation of ground-state correlations from excitation-specific dressing via a compact neural head on pretrained features would provide a transparent route to extend neural quantum states to dynamical response, with built-in translation symmetry and momentum resolution. This could be particularly useful for frustrated magnets where full variational optimization of excited states is costly.

major comments (2)
  1. [Method (neural polaron construction)] The central claim that the pretrained ground-state feature map plus compact neural head suffices to encode the dressing for the (π,0) anomaly (without independent excited-state variational freedom) is load-bearing but untested; no ablation against jointly optimized excited-state networks or comparison to methods with explicit excited-state optimization is reported.
  2. [Results (J1-J2 benchmarks)] The assertion of accurate reproduction of dispersions and spectral weights over a broad range of frustration lacks quantitative support in the form of error bars, dataset sizes, or direct numerical comparisons to exact diagonalization or other benchmarks; this undermines evaluation of how well the (π,0) anomaly softening is captured.
minor comments (2)
  1. [Method] Notation for the neural head and feature map should be defined with explicit equations early in the method section to clarify how locality and momentum resolution are enforced.
  2. [Abstract] The abstract states 'accurately reproduces' without referencing any specific metrics or figures; this should be qualified or tied to a results table/figure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their detailed and constructive report. We address the major comments point by point below.

read point-by-point responses

  1. Referee: [Method (neural polaron construction)] The central claim that the pretrained ground-state feature map plus compact neural head suffices to encode the dressing for the (π,0) anomaly (without independent excited-state variational freedom) is load-bearing but untested; no ablation against jointly optimized excited-state networks or comparison to methods with explicit excited-state optimization is reported.

    Authors: The neural polaron construction is intentionally designed to use a pretrained ground-state feature map with a compact neural head to parameterize the local dressing operator, thereby separating ground-state correlations from excitation-specific effects. This approach builds in translation symmetry and momentum resolution by construction. While we did not perform an explicit ablation study against jointly optimized excited-state networks in the current manuscript, the successful reproduction of magnon dispersions, spectral weights, and the (π,0) anomaly across frustration values supports the sufficiency of this method. We will revise the manuscript to include a more detailed discussion of this design choice and its advantages, and consider adding a limited comparison if computational resources allow. revision: partial

  2. Referee: [Results (J1-J2 benchmarks)] The assertion of accurate reproduction of dispersions and spectral weights over a broad range of frustration lacks quantitative support in the form of error bars, dataset sizes, or direct numerical comparisons to exact diagonalization or other benchmarks; this undermines evaluation of how well the (π,0) anomaly softening is captured.

    Authors: We agree that additional quantitative details would strengthen the presentation. In the revised manuscript, we will include error bars on the reported dispersions and spectral weights, specify the sizes of the training datasets used, and provide direct comparisons to exact diagonalization results for small systems or other established benchmarks where available to better quantify the accuracy of the (π,0) anomaly softening. revision: yes

Circularity Check

0 steps flagged

No significant circularity; neural polaron ansatz is an independent parameterization

full rationale

The paper introduces the neural polaron ansatz as a new operator-based construction that applies a compact neural head (local dressing operator) to the feature map of a separately pretrained ground-state network. This explicitly separates ground-state correlations from excitation-specific dressing and builds in symmetries by design. The central claims (reproduction of magnon dispersions, spectral weights, and the (π,0) anomaly across J2/J1) rest on this parameterization plus numerical benchmarking on the J1-J2 model, with no quoted equations or steps that reduce the output to a fit of the target quantities, a self-citation chain, or a renaming of known results. The method is presented as an independent variational ansatz rather than a tautological re-expression of its inputs.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Based on abstract only; full details on parameters, assumptions, and training not available. The central claim rests on the domain assumption that quasiparticle excitations admit a local operator dressing representation.

free parameters (1)
  • neural head parameters
    Compact neural network weights for the dressing operator are learned from data or loss.
axioms (1)
  • domain assumption Quasiparticle excitations can be represented by local dressing operators acting on a correlated ground state
    Core premise of the neural polaron ansatz stated in the abstract.
invented entities (1)
  • neural polaron ansatz no independent evidence
    purpose: To represent quasiparticle excitations via neural many-body operators
    New method introduced in the paper.

pith-pipeline@v0.9.1-grok · 5711 in / 1299 out tokens · 18748 ms · 2026-06-26T09:30:10.945337+00:00 · methodology

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

69 extracted references · 3 linked inside Pith

  1. [1]

    P. W. Anderson, Science177, 393 (1972)

    1972

  2. [2]

    Preskill, arXiv:1203.5813 (2012)

    J. Preskill, arXiv:1203.5813 (2012)

    Pith/arXiv arXiv 2012

  3. [3]

    Eisert, M

    J. Eisert, M. Cramer, and M. B. Plenio, Rev. Mod. Phys. 82, 277 (2010)

    2010

  4. [4]

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

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

    2011

  5. [5]

    Or´ us, Annals of physics349, 117 (2014)

    R. Or´ us, Annals of physics349, 117 (2014)

    2014

  6. [6]

    Sachdev, Physics world12, 33 (1999)

    S. Sachdev, Physics world12, 33 (1999)

    1999

  7. [7]

    Sachdev,Quantum phases of matter(Cambridge Uni- versity Press, 2023)

    S. Sachdev,Quantum phases of matter(Cambridge Uni- versity Press, 2023)

    2023

  8. [8]

    Wen,Quantum field theory of many-body systems: From the origin of sound to an origin of light and elec- trons(Oxford university press, 2004)

    X.-G. Wen,Quantum field theory of many-body systems: From the origin of sound to an origin of light and elec- trons(Oxford university press, 2004)

    2004

  9. [9]

    Abanov, A

    A. Abanov, A. V. Chubukov, and J. Schmalian, Advances in Physics52, 119 (2003)

    2003

  10. [10]

    Abanov and A

    A. Abanov and A. V. Chubukov, Physical review letters 84, 5608 (2000)

    2000

  11. [11]

    A. W. Sandvik, inAIP Conference Proceedings, Vol. 1297 (American Institute of Physics, 2010) pp. 135–338

    2010

  12. [12]

    Troyer and U.-J

    M. Troyer and U.-J. Wiese, Physical review letters94, 170201 (2005)

    2005

  13. [13]

    Laurell, A

    P. Laurell, A. Scheie, E. Dagotto, and D. A. Tennant, Advanced Quantum Technologies8, 2400196 (2025)

    2025

  14. [14]

    P. P. Stavropoulos, Y. Yang, I. Rousochatzakis, and N. B. Perkins, Physical Review B110, 214406 (2024)

    2024

  15. [15]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Reviews of Modern Physics91, 045002 (2019)

    2019

  16. [16]

    Carrasquilla, Advances in Physics: X5, 1797528 (2020)

    J. Carrasquilla, Advances in Physics: X5, 1797528 (2020)

    2020

  17. [17]

    Y.-Z. You, Z. Yang, and X.-L. Qi, Physical Review B97, 045153 (2018)

    2018

  18. [18]

    Szab´ o and C

    A. Szab´ o and C. Castelnovo, Physical Review Research 2, 033075 (2020)

    2020

  19. [19]

    X.-Q. Sun, T. Nebabu, X. Han, M. O. Flynn, and X.-L. Qi, Physical Review B106, 115138 (2022)

    2022

  20. [20]

    Carleo and M

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

    2017

  21. [21]

    K. Choo, T. Neupert, and G. Carleo, Physical Review B 100, 125124 (2019)

    2019

  22. [22]

    L. L. Viteritti, R. Rende, S. Sachdev, and G. Carleo, arXiv preprint arXiv:2602.02665 (2026)

    arXiv 2026

  23. [23]

    Geier, K

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

    2025

  24. [24]

    W. T. Lou, H. Sutterud, G. Cassella, W. M. C. Foulkes, J. Knolle, D. Pfau, and J. S. Spencer, Physical Review X 14, 021030 (2024)

    2024

  25. [25]

    R. P. Nutakki, A. Shokry, and F. Vicentini, arXiv preprint arXiv:2505.03466 (2025)

    arXiv 2025

  26. [26]

    L. L. Viteritti, R. Rende, and F. Becca, Physical Review Letters130, 236401 (2023). 6

    2023

  27. [27]

    Rende, A

    R. Rende, A. Nikolaenko, L. L. Viteritti, S. Sachdev, and Y.-H. Zhang, arXiv preprint arXiv:2603.02316 (2026)

    arXiv 2026

  28. [28]

    Chatterji, inNeutron scattering from magnetic mate- rials(Elsevier, 2006) pp

    T. Chatterji, inNeutron scattering from magnetic mate- rials(Elsevier, 2006) pp. 1–24

    2006

  29. [29]

    Christianson, E

    A. Christianson, E. Goremychkin, R. Osborn, S. Rosenkranz, M. Lumsden, C. Malliakas, I. Todorov, H. Claus, D. Chung, M. Kanatzidis,et al., Nature456, 930 (2008)

    2008

  30. [30]

    D. A. Tennant, Journal of the Physical Society of Japan 88, 081009 (2019)

    2019

  31. [31]

    Damascelli, Z

    A. Damascelli, Z. Hussain, and Z.-X. Shen, Reviews of modern physics75, 473 (2003)

    2003

  32. [32]

    L. J. P. Ament, M. van Veenendaal, T. P. Devereaux, J. P. Hill, and J. van den Brink, Rev. Mod. Phys.83, 705 (2011)

    2011

  33. [33]

    Y. Xu, Y. Yang, J. Teyssier, T. Ohtsuki, Y. Qiu, S. Nakatsuji, D. van der Marel, N. B. Perkins, and N. Drichko, npj Quantum Materials10, 88 (2025)

    2025

  34. [34]

    Park, S.-S

    P. Park, S.-S. Zhang, P. M. Bonetti, A. A. Podlesnyak, D. M. Pajerowski, M. B. Stone, C. Petrovic, C. Stock, S. Sachdev, C. D. Batista,et al., arXiv preprint arXiv:2604.02481 (2026)

    Pith/arXiv arXiv 2026

  35. [35]

    Eisert, M

    J. Eisert, M. Friesdorf, and C. Gogolin, Nat. Phys.11, 124 (2015)

    2015

  36. [36]

    K. Choo, G. Carleo, N. Regnault, and T. Neupert, arXiv preprint arXiv:1807.03325 (2018)

    Pith/arXiv arXiv 2018

  37. [37]

    Mendes-Santos, M

    T. Mendes-Santos, M. Schmitt, and M. Heyl, Physical Review Letters131, 046501 (2023)

    2023

  38. [38]

    A. Ning, L. Yang, and G.-W. Chern, Physical Review E 113, 015307 (2026)

    2026

  39. [39]

    Dalla Piazza, M

    B. Dalla Piazza, M. Mourigal, N. B. Christensen, G. Nilsen, P. Tregenna-Piggott, T. Perring, M. Enderle, D. F. McMorrow, D. Ivanov, and H. M. Rønnow, Nature physics11, 62 (2015)

    2015

  40. [40]

    Powalski, K

    M. Powalski, K. P. Schmidt, and G. S. Uhrig, SciPost Physics4, 001 (2018)

    2018

  41. [41]

    H. Shao, Y. Q. Qin, S. Capponi, S. Chesi, Z. Y. Meng, and A. W. Sandvik, Physical Review X7, 041072 (2017)

    2017

  42. [42]

    Ferrari and F

    F. Ferrari and F. Becca, Physical Review B98, 100405 (2018)

    2018

  43. [43]

    A. W. Sandvik and R. R. Singh, Physical Review Letters 86, 528 (2001)

    2001

  44. [44]

    Powalski, G

    M. Powalski, G. Uhrig, and K. Schmidt, Physical review letters115, 207202 (2015)

    2015

  45. [45]

    R. R. Singh and M. P. Gelfand, Physical Review B52, R15695 (1995)

    1995

  46. [46]

    L. D. Landau, Sov. Phys. JETP3, 920 (1956), see also Sov. Phys. JETP5, 101 (1957)

    1956

  47. [47]

    Tosatti, Quantum theory of the electron liquid (2006)

    E. Tosatti, Quantum theory of the electron liquid (2006)

    2006

  48. [48]

    J. T. Devreese and A. S. Alexandrov, Reports on Progress in Physics72, 066501 (2009)

    2009

  49. [49]

    R. P. Feynman, Physical Review94, 262 (1954)

    1954

  50. [50]

    R. P. Feynman and M. Cohen, Physical Review102, 1189 (1956)

    1956

  51. [51]

    A. J. Leggett,Quantum liquids: Bose condensation and Cooper pairing in condensed-matter systems(Oxford uni- versity press, 2006)

    2006

  52. [52]

    Read and S

    N. Read and S. Sachdev, Physical review letters66, 1773 (1991)

    1991

  53. [53]

    A. V. Chubukov and O. A. Starykh, Journal of Physics: Condensed Matter11, L169 (1999)

    1999

  54. [54]

    Zhang, E

    S.-S. Zhang, E. Ghioldi, L. O. Manuel, A. E. Trumper, and C. D. Batista, Physical Review B105, 224404 (2022)

    2022

  55. [55]

    Liang, W.-Y

    X. Liang, W.-Y. Liu, P.-Z. Lin, G.-C. Guo, Y.-S. Zhang, and L. He, Physical Review B98, 104426 (2018)

    2018

  56. [56]

    A. Chen, K. Choo, N. Astrakhantsev, and T. Neupert, Physical Review Research4, L022026 (2022)

    2022

  57. [57]

    C. Roth, A. Szab´ o, and A. H. MacDonald, Physical Re- view B108, 054410 (2023)

    2023

  58. [58]

    Wang and A

    L. Wang and A. W. Sandvik, Physical review letters121, 107202 (2018)

    2018

  59. [59]

    J. Yang, A. W. Sandvik, and L. Wang, Physical Review B105, L060409 (2022)

    2022

  60. [60]

    Maity, A

    A. Maity, A. Feuerpfeil, R. Thomale, S. Sachdev, and Y. Iqbal, arXiv preprint arXiv:2603.15745

    arXiv

  61. [61]

    K. Choo, A. Mezzacapo, and G. Carleo, Nature Commu- nications11, 2368 (2020)

    2020

  62. [62]

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

    2020

  63. [63]

    Hermann, Z

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

    2020

  64. [64]

    D. Luo, D. D. Dai, and L. Fu, Physical Review B113, 165107 (2026)

    2026

  65. [65]

    Marshall, Proceedings of the Royal Society of London

    W. Marshall, Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences232, 48 (1955)

    1955

  66. [66]

    Richter, N

    J. Richter, N. Ivanov, and K. Retzlaff, EPL (Europhysics Letters)25, 545 (1994)

    1994

  67. [67]

    Retzlaff, J

    K. Retzlaff, J. Richter, and N. Ivanov, Zeitschrift f¨ ur Physik B Condensed Matter93, 21 (1993)

    1993

  68. [68]

    S.-S. Gong, W. Zhu, D. Sheng, O. I. Motrunich, and M. P. Fisher, Physical review letters113, 027201 (2014)

    2014

  69. [69]

    Neural Polaron: Learning Quasiparticle Operators in Quantum Many-Body Systems

    S. Sorella, M. Casula, and D. Rocca, The Journal of chemical physics127(2007). 7 Supplemental Material for “Neural Polaron: Learning Quasiparticle Operators in Quantum Many-Body Systems” Shang-Shun Zhang Department of Physics and Astronomy, University of Tennessee, Knoxville, Tennessee 37996, USA NEURAL-NETWORK W A VEFUNCTION ARCHITECTURE In this section,...

    2007