Pith. sign in

REVIEW 3 major objections 7 minor 4 cited by

Solving and visualizing fractional quantum Hall wavefunctions with neural network

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A single self-attention neural network, solving the ν=1/3 Coulomb gas in real space without Landau-level projection, beats projected exact diagonalization in energy and reveals short-distance wavefunction structure plus a strong-mixing…

desk verdict LL-mixing energies are a real step forward; the 'unbiased discovery' of short-distance physics is pre-encoded in the Jastrow ansatz. read the letter →

arxiv 2412.00618 v2 pith:JFI6VEKW submitted 2024-11-30 cond-mat.str-el cond-mat.dis-nnquant-ph

Yi Teng , David D. Dai , Liang Fu This is my paper · ORCID
classification cond-mat.str-elcond-mat.dis-nnquant-ph PACS 73.43.-f71.15.-m
keywords fractionalquantumHalleffectneuralnetworkvariationalMonteCarloself-attentionwavefunctionLandaulevelmixingCoulombcuspLaughlinWignercrystalexactdiagonalization
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

The paper asks whether a single, physics-agnostic neural-network wavefunction can variationally solve the fractional quantum Hall problem directly in real space, without projecting onto the lowest Landau level. The authors adapt the self-attention network Psiformer to spin-polarized electrons on a disk at filling ν=1/3 with Coulomb interaction and a neutralizing background. They find that the network consistently reaches energies below lowest-Landau-level projected exact diagonalization, spontaneously learns circular symmetry and quantized angular momentum, and produces a wavefunction with short-distance Coulomb cusp and zero-splitting structure absent from the Laughlin ansatz. At strong Landau-level mixing the network finds a transition from the fractional quantum Hall droplet to a crystal-like state. If correct, this makes neural-network variational Monte Carlo a universal, unbiased solver for strongly correlated and topological states where Landau-level projection fails.

What carries the argument

The central object is Psiformer, a self-attention fermionic neural network whose permutation-equivariant complex generalized orbitals are built from one-electron features via attention layers, aggregated into a sum of Slater determinants, and multiplied by a Jastrow factor J = ∑ −β α²/(α+|ri−rj|). The Jastrow factor carries the two-electron Coulomb cusp: the short-distance wavefunction behaves as r^l[1 + r/(ϵ(2l+1))], and the initial β is set to 1/4, between the l=1 value 1/3 and the l=3 value 1/7, so the network can interpolate between generic Coulomb behavior and Laughlin-like correlation. Total angular momentum is measured during training as a convergence diagnostic, since low-lying states differ in angular momentum sector. This machinery lets the network represent unprojected wavefunctions with infinitely many Landau levels and allows direct visualization of phase and magnitude structure near electron collisions.

What would settle it

Run the same Psiformer at ν=1/3, λ=1/3 with the Jastrow factor's β fixed to the Laughlin value 1/7 (or with the Jastrow removed entirely) and check whether the optimized wavefunction still develops a β≈1/3 cusp and three split first-order zeros around each electron; if it does not, the short-distance structure was pre-imposed by the ansatz. Separately, repeat the λ=9 calculation on a torus or with a much softer confining potential for N=7,...,20 and check whether the angular-momentum drop and the one-center-plus-edge density survive in the thermodynamic limit.

Watch

Extended reading notes

Core claim

The central claim is that a self-attention fermionic neural network, with no Landau-level truncation and no FQH-specific design, variationally solves the ν=1/3 Coulomb electron gas in a magnetic field more accurately than LLL-projected exact diagonalization across a wide range of the Landau-level mixing parameter λ. At λ=1/3 the network's ground-state energy is lower than ED for 6–10 electrons, and the gap widens at larger λ; kinetic energy above ωc/2 directly evidences occupation of higher Landau levels. Microscopically, the learned wavefunction shows a Coulomb cusp as two electrons approach and, instead of the Laughlin m=3 zero, three first-order zeros whose combined phase winding around one electron is still 3, consistent with mixing of l=1 and l=3 relative angular momentum channels. At λ=9 the ground-state angular momentum drops from M=108 to M=100 and the charge density develops a one-electron-at-center, eight-on-the-edge shell structure, which the authors interpret as a transition from the FQH liquid to a rotating Wigner-molecule or crystal state on the disk.

Load-bearing premise

The short-distance 'discovery' rests on assuming the trial wavefunction's built-in two-electron factor—whose initial strength sits between the l=1 and l=3 cusp values—did not already force the zero-splitting and cusp reported as learned, and the transition claim rests on assuming the angular-momentum drop at the strongest mixing is the thermodynamic ground state rather than a finite-disk artifact.

Editorial extensions

If this is right

  • The same Psiformer architecture solves the FQH droplet and the strong-mixing crystal state from one unbiased ansatz, without separate trial wavefunctions for each phase.
  • LL-projected ED energies are variational upper bounds that miss Landau-level mixing; at moderate to strong λ the FNN energies are lower and kinetic energy exceeds ωc/2, so unprojected treatment changes quantitative energetics.
  • Short-distance FQH pair correlations are not the Laughlin r^m form: the Coulomb cusp requires mixing of infinitely many Landau levels, and the m=3 zero splits into three first-order zeros with total phase winding 3.
  • At λ=9 the ground state on the disk changes from the FQH droplet (angular momentum M=108) to a crystal-like state (M=100) with one electron in the center and eight on the edge for N=9, with similar angular-momentum drops for 7–11 electrons.
  • The method reaches N=12 electrons on a disk without Landau-level truncation, beyond the 10-electron exact-diagonalization limit presented in the paper.

Reading between the lines

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

  • A clean test of whether the short-distance structure is genuinely learned would be to fix the Jastrow β at the Laughlin value 1/7 or remove the Jastrow entirely and check whether the optimized wavefunction still develops the l=1 cusp and the split first-order zeros.
  • The λ=9 transition could be an artifact of the disk's confining potential stabilizing the one-center-plus-edge shell; repeating the calculation on a torus or with a much softer boundary would show whether the angular-momentum drop and density profile persist in the thermodynamic limit.
  • The same unprojected approach should extend naturally to even-denominator and non-Abelian states such as ν=5/2, where Landau-level mixing is thought to be crucial, and to moiré fractional Chern insulators, where the angular-momentum diagnostic would need replacement by momentum or entanglement diagnostics.
  • The paper notes FermiNet struggles to reach the correct angular momentum in this problem; a controlled comparison with FermiNet at identical Jastrow factor and system size would isolate whether self-attention's all-to-all equivariant weighting is the key architectural advantage.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. This paper adapts the Psiformer self-attention fermionic neural network to the N-electron two-dimensional Coulomb gas in a magnetic field, working in the full Hilbert space of a disk without Landau-level (LL) truncation. The variational wavefunction is a sum of determinants of complex, permutation-equivariant generalized orbitals with isotropic Gaussian envelopes, multiplied by the real Jastrow factor exp(−β α²/(α+|ri−rj|)). The authors benchmark against LLL-projected exact diagonalization (ED) at fixed angular momentum M for N=6–10 at LL-mixing parameter λ=1/3, find slightly lower FNN energies at every N, and monitor convergence through the measured total angular momentum (converging to 197.97 for N=12, close to the Laughlin value M=198). Fixing N−1 electrons, they visualize the phase and magnitude of the learned 12-electron wavefunction and report a linear trimer of first-order zeros around each electron (splitting of the Laughlin third-order zero), Ψ∼r at short distance with a crossover to Ψ∼r³ near ℓB, and total phase winding 3. For λ=9 they find an angular-momentum drop (M=108→100 for N=9, 135→127 for N=10) and a radial density with one central and N−1 outer electrons, which they interpret, with explicit caveats, as a transition from the FQH droplet toward a crystal-like state (rotating Wigner molecule, Wigner crystal, or Hall crystal).

Significance. The paper's core methodological claim—that a single attention-based FNN can variationally solve the continuum FQH problem without LL projection, at N up to 12, and capture the l=1/l=3 short-distance crossover—is significant for the NN-VMC field if the benchmarks hold. Strengths to credit: the angular-momentum expectation value used as a convergence diagnostic is thoughtful; the SM derives the two-electron cusp condition analytically (Eq. S14); the confining-potential implementation is checked at the 0.1% level; and the hyperparameters and optimization details are reported transparently. The short-distance findings are, however, less novel than presented: the zero-splitting physics was predicted in the cited literature (Refs. [40–44]), and the cusp coefficient is an explicit input to the ansatz rather than a purely learned output. The phase-transition claim is a finite-size observation with the authors' own caveats. The demonstration is nevertheless credible and valuable: it provides the first full-Hilbert-space NN-VMC treatment of this canonical problem and opens a route to strong-LL-mixing regimes inaccessible to LLL-projected ED.

major comments (3)
  1. [Sec. II, Sec. IV, Eq. (3); SM 'Jastrow Factor and Cusp Conditions'] The claim in Sec. II that 'no information about FQH physics is put in by hand during NN initialization and training' is not consistent with the construction described in Sec. IV and the SM. The Jastrow factor in Eq. (3) is admitted to be motivated by the two-electron Coulomb cusp of Eq. (4), and β is initialized at 1/4, the midpoint of the window [1/7, 1/3] that presupposes the l=1/l=3 mixture whose 'discovery' is announced in Sec. VI; the SM states that 'the cusp information learned from FermiNet can then be put into our more expressive Psiformer ansatz.' I note a mitigating detail: the Jastrow is real and positive, so the zero-splitting and phase winding in Fig. 4 are produced by the learned determinants rather than literally inserted, and the optimized l=1-to-l=3 amplitude ratio and the trimer scale ξ are genuine variational outputs. Nevertheless, the qualitative short-distance content (Ψ∼r at r→0, winding l=1, crossover to winding 3 at ∼ℓB) is the input hypothesis rather than the network's discovery, and the same zero-splitting physics was already predicted and studied in Refs. [40–44]. The manuscript should (i) explicitly enumerate which elements are encoded inputs and which are learned; (ii) add a control (e.g., β initialized at 1/3 and 1/7, or the Jastrow omitted) showing that the converged energy and the short-distance structure are stable; and (iii) temper the 'no information... by hand' phrasing.
  2. [Sec. VI Fig. 3(a); Sec. VII Fig. 5(a); SM 'Exact Diagonalization Calculation'] The central benchmark claim that the FNN 'consistently attains energies lower than LL-projected ED' (Secs. VI–VII, Figs. 3(a) and 5(a)) is presented without any statistical error bars on the VMC energies, without the variance of the local energy that Sec. V itself identifies as the convergence criterion, and without a numeric table of energies, so at λ=1/3, where the advantage is described as 'slightly lower,' the reader cannot tell whether the difference is meaningful or within Monte Carlo noise. In addition, the SM section on ED specifies only the LLL basis and M=ML constraint; it does not state whether the disk confining potential Vc and the background term Vb are included in the ED Hamiltonian, which must be identical to the FNN Hamiltonian for the comparison to be valid. At large λ, 'outperforms ED significantly' is the expected behavior of any full-Hilbert-space method against a LLL-projected one, so an independent reference (multi-LL ED at N_LL=2–3, or fixed-phase DMC with the best trial state) is needed to support the accuracy claim in the strong-mixing regime. Please provide a table of energies with statistical errors and sample counts, clarify the ED Hamiltonian, and add an independent cross-check at λ≥3.
  3. [Sec. VII, Fig. 5(b)-(c); Abstract] The abstract's claim that 'a phase transition from FQH liquid to a crystal state is found at strong LL mixing' is materially stronger than what Sec. VII establishes, and the section itself concedes: 'it is difficult to draw a definite conclusion from the above NN-VMC studies' on a finite disk. The evidence consists of angular-momentum drops (M=108→100 for N=9; 135→127 for N=10) at λ=9 and a radial density with a single central electron; the asserted consistency across 7–11 electrons is not shown, there is no scaling of the crossover with N, no explicit comparison of converged FNN energies in the M=100 and M=108 sectors to show which sector wins, and no (rotating-frame) two-body correlation or order parameter that would discriminate a Wigner molecule, a Wigner crystal, or a Hall crystal. The 'unbiased solver' claim makes the absence of any independent cross-check at λ=9, where LLL-projected ED is no longer a useful reference, a further correctness risk. I recommend either softening the abstract and conclusion to a 'signature consistent with a crystal-like state on the disk,' or adding the scaling and order-parameter analysis required to justify the phase-transition claim.
minor comments (7)
  1. [Sec. VI] The word 'ciruclarly' appears twice in Sec. VI and should read 'circularly'; the affiliation line 'N¨ othnitzer Straße' on the first page also renders with a misplaced diacritic.
  2. [Title page] The running header title, 'Solving the fractional quantum Hall problem with self-attention neural network,' differs from the arXiv title 'Solving and visualizing fractional quantum Hall wavefunctions with neural network'; the submitted title should be unified.
  3. [Sec. VI] The angular-momentum monitor is a genuine strength, but the reported value 197.97 for N=12 is given without an error bar; report the statistical uncertainty, since the convergence interpretation rests on proximity to the integer 198.
  4. [Sec. IV and SM Eq. (S14)] The expression 1/ϵ(2|l|+1) in Eq. (4) and SM Eq. (S14) is ambiguous between [ϵ(2|l|+1)]^{-1} and (1/ϵ)(2|l|+1); add parentheses or state the intended reading in words.
  5. [Fig. 4 and surrounding text] The trimer-of-zeros structure is presented from a single 'typical Monte Carlo configuration' (the most probable of 5000 samples); state how the zero positions ξ and the winding structure vary across independently sampled configurations.
  6. [Sec. IV and SM] The SM says the cusp parameter was learned by an auxiliary FermiNet calculation and then transferred to Psiformer; this is an important methodological statement and should appear in the main text near Sec. IV, together with a note that β=1/4 is an effective cusp for the intermediate-distance window rather than the exact r→0 cusp 1/3 mentioned in the same section.
  7. [Data availability] No data-availability or code-release statement is given; for a method built on public repositories (FermiNet/JAX), releasing the evaluation and training scripts would aid reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Short-distance 'beyond Laughlin' features are partly pre-encoded: the Coulomb cusp and β=1/4 Jastrow factor are fitted inputs, so the claim of an unbiased discovery is not fully supported.

  1. fitted input called prediction [Section IV 'Coulomb Cusp and Jastrow Factor' (Eq. 3); SM 'Jastrow Factor and Cusp Conditions' and 'FermiNet for Two Electron Cusp' (Eqs. S14, S17).]
    "To capture this crossover behavior, we choose the initial value for cusp parameter β to be 1/4, which lies between the value 1/3 for l = 1 and 1/7 for l = 3 (assuming ϵ = 1). ... The cusp information learned from FermiNet can then be put into our more expressive Psiformer ansatz in a Jastrow factor of the following form: J(r) = Σ_{i<j} -1/4 α²/(α + |ri - rj|)."

    The SM derives the short-distance cusp form ψ = e^{ilθ} ρ^{|l|} [1 + ρ/(ϵ(2|l|+1)) + O(ρ²)] and states that 'the Jastrow factor in Psiformer is recovered with β = 1/ϵ(2|l|+1)'. Since exp(-β α²/(α+ρ)) = const × (1 + βρ + O(ρ²)), the Jastrow factor with β initialized to 1/4 literally builds the linear-in-ρ Coulomb cusp into the ansatz before optimization. The paper's own construction therefore fixes the short-distance scaling Ψ ∼ r as r → 0 reported in Fig. 4(d), and the SM explicitly says the FermiNet-learned cusp was 'put into' Psiformer. This contradicts the paper's claim that 'no information about FQH physics is put in by hand during NN initialization and training'.

full rationale

The central energy benchmark is self-contained: FNN variational energies are evaluated on the full real-space Hamiltonian and compared against an independently implemented LL-projected ED, so the 'lower than ED' result is not circular. The phase-transition claim is explicitly qualified by the authors as a finite-disk statement ('it is difficult to draw a definite conclusion'), which is a finite-size caveat rather than a circularity. No load-bearing self-citation was identified: Psiformer is cited to von Glehn, Spencer, and Pfau, and the self-citation to Dai and Fu [74] is used only for ED implementation details. However, the headline claim that the FNN 'reveals microscopic features ... beyond the Laughlin ansatz' is substantially circular: the Coulomb cusp is not discovered from an unbiased network. The Jastrow factor in Eq. (3)/(S17) is expressly chosen to reproduce the cusp condition derived in Eq. (S14), with β initialized at 1/4 from a FermiNet fit; the short-distance Ψ ∼ r behavior shown in Fig. 4 is consequently an input of the ansatz. The zero-trimer splitting requires the network to learn a z¹ component in the complex determinant and therefore retains independent content, preventing a higher score. Overall, the variational method itself is credible, but the 'beyond Laughlin' discovery narrative is partly the return of an engineered ansatz, so the circularity score is 6.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The ledger shows the central short-distance claim is cosmetically supported by chosen parameters: the cusp is imposed by the Jastrow factor and the β initialization. The phase-transition claim additionally depends on a particular finite geometry. No new particles or mediators are introduced.

free parameters (5)
  • β (Jastrow cusp parameter) = 1/4 (initial value, then variational)
    Chosen by hand to lie between the Coulomb cusp values for l=1 (1/3) and l=3 (1/7), as explained in Sec. IV and SM; it encodes the very short-distance physics the paper claims to reveal.
  • α (Jastrow range parameter) = learned
    Variational parameter in Eq. (3); sets the spatial scale of the cusp region and is optimized during training with no independent benchmark.
  • Orbital envelope widths σ_i = learned
    Per-orbital Gaussian envelope widths in the Psiformer ansatz (SM Eq. S4); they are variational parameters that shape the orbitals.
  • Disk separation d = 0.1 ℓB
    Set by hand to avoid edge reconstruction (Sec. V); affects the confining potential and hence the density profile and phase-transition interpretation.
  • Disk radius a = set by flux condition Φ = NΦ0/ν
    Chosen so the average filling is ν=1/3; the finite radius and the soft edge influence the charge density and angular momentum results.
assumptions (6)
  • domain assumption Two spin-parallel electrons at short distance obey the 2D Coulomb cusp expansion ψ = z^{|l|}[1 + ρ/(ϵ(2|l|+1)) + O(ρ²)] (Eq. 4).
    Used to motivate the Jastrow factor and to choose β; derived in the SM from the 2D two-body Schrödinger equation under assumptions of finite confining potential and a separation of scales.
  • standard math LLL wavefunctions are holomorphic functions times a Gaussian (Eq. 5), hence cannot describe the linear-in-|z| cusp.
    Basis for the claim that LL mixing is required for the cusp; follows from the analytic structure of lowest-Landau-level states.
  • ad hoc to paper The Jastrow form exp(Σ −β α²/(α+|ri−rj|)) with β between 1/3 and 1/7 adequately captures the l=1/l=3 crossover.
    The authors state the choice 'works well in practice' (Sec. IV); no derivation establishes that this one-parameter form is sufficient to represent the true two-body short-range behavior.
  • domain assumption The total angular momentum M of the ν=1/3 Laughlin state is ML = mN(N−1)/2, and the ground state at ν=1/3 on the disk lies in the M=ML sector.
    Used both for the ED Hilbert space and as the FNN convergence criterion (Secs. V and VI); standard for FQH on a disk but an input assumption.
  • domain assumption Monitoring the convergence of total energy, local-energy variance, and angular momentum identifies the ground state rather than a superposition of edge excitations.
    Heuristic used in Sec. V to stop training; the paper notes low-lying gapless edge states can trap the optimization.
  • domain assumption The neutralizing charged disk with d=0.1ℓB is a faithful representation of the 2D Coulomb gas for FQH purposes.
    Choice of geometry (Sec. V) and the confining potential; edge effects and finite-size effects enter the density and phase-transition claims.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Solving and visualizing fractional quantum Hall wavefunctions with neural network." pith.science (2026). https://pith.science/paper/JFI6VEKW

@misc{pith2026241200618,
  author       = {Pith},
  title        = {Pith review of: Solving and visualizing fractional quantum Hall wavefunctions with neural network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFI6VEKW}},
  note         = {Machine review of arXiv:2412.00618}
}
abstract

We introduce an attention-based fermionic neural network (FNN) to variationally solve the problem of two-dimensional Coulomb electron gas in magnetic fields, a canonical platform for fractional quantum Hall (FQH) liquids, Wigner crystals and other unconventional electron states. Working directly with the full Hilbert space of $N$ electrons confined to a disk, our FNN consistently attains energies lower than LL-projected exact diagonalization (ED) and learns the ground state wavefunction to high accuracy. In low LL mixing regime, our FNN reveals microscopic features in the short-distance behavior of FQH wavefunction beyond the Laughlin ansatz. For moderate and strong LL mixing parameters, the FNN outperforms ED significantly. Moreover, a phase transition from FQH liquid to a crystal state is found at strong LL mixing. Our study demonstrates unprecedented power and universality of FNN based variational method for solving strong-coupling many-body problems with topological order and electron fractionalization.

Figures

Figures reproduced from arXiv: 2412.00618 by the authors.

Figure 1
Figure 1. FIG. 1. Psiformer architecture [ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. The geometry of our system: an infinite plane with [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. ED and FNN benchmark. (a) The ground state energy per particle (in the unit of 1 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Visualization of the many-body wavefunctions for Coulomb interaction (top panel) and for the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. The results for 9 electrons at various Landau mixing parameters. (a) Top panel shows the ground state reduced energy [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Hyperdeterminant wavefunctions

    cond-mat.str-el 2026-07 conditional novelty 7.5 of 10

    Hyperdeterminant wavefunctions with a locality structure, plus VMPI effective theories and projective expansion, give a practical variational framework for fractional Chern insulators and quantum spin liquids.

  2. Group Convolutional Neural Network for the Low-Energy Spectrum in the Quantum Dimer Model

    cond-mat.dis-nn 2025-05 conditional novelty 6.0 of 10

    Irrep-resolved GCNN variational energies indicate a 4-fold degenerate columnar ground state for V <= 0.4 in the square-lattice quantum dimer model, shifting possible plaquette or mixed ordering to 0.4 < V < 1.

  3. Is attention all you need to solve the correlated electron problem?

    cond-mat.str-el 2025-02 conditional novelty 6.0 of 10

    A self-attention neural network wavefunction gives lower variational energies than band-projected exact diagonalization for a moiré electron model and shows a roughly quadratic parameter scaling with electron number.

  4. Taming Landau level mixing in fractional quantum Hall states with deep learning

    cond-mat.str-el 2024-12 conditional novelty 5.0 of 10

    A real-space neural network wavefunction captures Landau level mixing in fractional quantum Hall systems and yields lower energies than lowest-Landau-level exact diagonalization at nu=1/3 and 2/5.

Reference graph

Works this paper leans on

74 extracted references · 46 canonical work pages · cited by 4 Pith papers

  1. [1]

    Kohn and L

    W. Kohn and L. Sham, in Conference Proceedings-Italian Physical Society, Vol. 49 (Editrice Compositori, Bologna,

    work page

  2. [2]

    J. P. Perdew and S. Kurth, Density functionals for non- relativistic Coulomb systems in the new century, in A Primer in Density Functional Theory (Springer Berlin Heidelberg, 2003) p. 1–55

    work page 2003

  3. [3]

    A. D. Becke, The Journal of Chemical Physics 140, 10.1063/1.4869598 (2014)

    doi:10.1063/1.4869598 2014

  4. [4]

    ¨Ostlund and S

    S. ¨Ostlund and S. Rommer, Physical Review Letters 75, 3537–3540 (1995)

    work page 1995

  5. [5]

    Stoudenmire and S

    E. Stoudenmire and S. R. White, Annual Review of Con- densed Matter Physics 3, 111–128 (2012)

    work page 2012

  6. [6]

    Karrasch and J

    C. Karrasch and J. E. Moore, Phys. Rev. B 86, 155156 (2012)

    work page 2012

  7. [7]

    Davydova, Y

    M. Davydova, Y. Zhang, and L. Fu, Physical Review B 107, 10.1103/physrevb.107.224420 (2023)

    work page doi:10.1103/physrevb.107.224420 2023

  8. [8]

    Zhang and L

    Y. Zhang and L. Fu, SciPost Physics Core 6, 10.21468/scipostphyscore.6.2.038 (2023)

    work page doi:10.21468/scipostphyscore.6.2.038 2023

Show all 74 references
  1. [9]

    Z. Tao, W. Zhao, B. Shen, P. Kn¨ uppel, K. Watanabe, T. Taniguchi, J. Shan, and K. F. Mak, Observation of spin polarons in a frustrated moir´ e Hubbard system (2023), arXiv:2307.12205 [cond-mat.str-el]

    2023 arXiv

  2. [10]

    H. Li, U. Kumar, K. Sun, and S.-Z. Lin, Physical Review Research 3, 10.1103/physrevresearch.3.l032070 (2021)

    2021 doi

  3. [11]

    Cr´ epel and L

    V. Cr´ epel and L. Fu, Physical Review B 107, 10.1103/physrevb.107.l201109 (2023)

    2023 doi

  4. [12]

    J. Cai, E. Anderson, C. Wang, X. Zhang, X. Liu, W. Holtzmann, Y. Zhang, F. Fan, T. Taniguchi, K. Watanabe, Y. Ran, T. Cao, L. Fu, D. Xiao, W. Yao, and X. Xu, Nature 622, 63–68 (2023)

    2023

  5. [13]

    Y. Zeng, Z. Xia, K. Kang, J. Zhu, P. Kn¨ uppel, C. Vaswani, K. Watanabe, T. Taniguchi, K. F. Mak, and J. Shan, Nature 622, 69–73 (2023)

    2023

  6. [14]

    H. Park, J. Cai, E. Anderson, Y. Zhang, J. Zhu, X. Liu, C. Wang, W. Holtzmann, C. Hu, Z. Liu, T. Taniguchi, K. Watanabe, J.-H. Chu, T. Cao, L. Fu, W. Yao, C.-Z. Chang, D. Cobden, D. Xiao, and X. Xu, Nature 622, 74–79 (2023)

    2023

  7. [15]

    Z. Lu, T. Han, Y. Yao, A. P. Reddy, J. Yang, J. Seo, K. Watanabe, T. Taniguchi, L. Fu, and L. Ju, Nature 626, 759–764 (2024)

    2024

  8. [16]

    A. P. Reddy, F. Alsallom, Y. Zhang, T. Devakul, and L. Fu, Phys. Rev. B 108, 085117 (2023)

    2023

  9. [17]

    Abouelkomsan, A

    A. Abouelkomsan, A. P. Reddy, L. Fu, and E. J. Bergholtz, Physical Review B 109, 10.1103/phys- revb.109.l121107 (2024)

    2024 doi

  10. [18]

    J. Yu, J. Herzog-Arbeitman, M. Wang, O. Vafek, B. A. Bernevig, and N. Regnault, Phys. Rev. B 109, 045147 (2024)

    2024

  11. [19]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Reviews of Modern Physics 91, 10.1103/revmodphys.91.045002 (2019)

    2019 doi

  12. [20]

    Hermann, J

    J. Hermann, J. Spencer, K. Choo, A. Mezzacapo, W. M. C. Foulkes, D. Pfau, G. Carleo, and F. No´ e, Na- ture Reviews Chemistry 7, 692–709 (2023)

    2023

  13. [21]

    Carrasquilla and R

    J. Carrasquilla and R. G. Melko, Nature Physics 13, 431–434 (2017)

    2017

  14. [22]

    Luo and B

    D. Luo and B. K. Clark, Physical Review Letters 122, 10.1103/physrevlett.122.226401 (2019)

    2019 doi

  15. [23]

    Carleo, K

    G. Carleo, K. Choo, D. Hofmann, J. E. Smith, T. West- erhout, F. Alet, E. J. Davis, S. Efthymiou, I. Glasser, S.-H. Lin, M. Mauri, G. Mazzola, C. B. Mendl, E. van Nieuwenburg, O. O’Reilly, H. Th´ eveniaut, G. Torlai, F. Vicentini, and A. Wietek, SoftwareX 10, 100311 (2019)

    2019

  16. [24]

    Kaubruegger, L

    R. Kaubruegger, L. Pastori, and J. C. Budich, Phys. Rev. B 97, 195136 (2018)

    2018

  17. [25]

    Glasser, N

    I. Glasser, N. Pancotti, M. August, I. D. Rodriguez, and J. I. Cirac, Phys. Rev. X 8, 011006 (2018)

    2018

  18. [26]

    C. Roth, A. Szab´ o, and A. H. MacDonald, Phys. Rev. B 108, 054410 (2023)

    2023

  19. [27]

    Hermann, Z

    J. Hermann, Z. Sch¨ atzle, and F. No´ e, Nature Chemistry 12, 891–897 (2020)

    2020

  20. [28]

    D. Pfau, J. S. Spencer, A. G. D. G. Matthews, and W. M. C. Foulkes, Phys. Rev. Res. 2, 033429 (2020)

    2020

  21. [29]

    J. S. Spencer, D. Pfau, A. Botev, and W. M. C. Foulkes, Better, faster fermionic neural networks (2020), arXiv:2011.07125 [physics.comp-ph]

    2020 arXiv

  22. [30]

    X. Li, Z. Li, and J. Chen, Nature Communications 13, 10.1038/s41467-022-35627-1 (2022)

    2022 doi

  23. [31]

    Cassella, H

    G. Cassella, H. Sutterud, S. Azadi, N. Drummond, D. Pfau, J. S. Spencer, and W. Foulkes, Physical Review Letters 130, 10.1103/physrevlett.130.036401 (2023)

    2023 doi

  24. [32]

    Pescia, J

    G. Pescia, J. Nys, J. Kim, A. Lovato, and G. Carleo, Message-passing neural quantum states for the homoge- neous electron gas (2023), arXiv:2305.07240 [quant-ph]

    2023 arXiv

  25. [33]

    D. Luo, D. D. Dai, and L. Fu, Pairing-based graph neu- ral network for simulating quantum materials (2023), arXiv:2311.02143 [cond-mat.str-el]

    2023 arXiv

  26. [34]

    X. Li, Y. Qian, W. Ren, Y. Xu, and J. Chen, Emergent wigner phases in moir´ e superlattice from deep learning (2024), arXiv:2406.11134 [physics.comp-ph]

    2024 arXiv

  27. [35]

    D. Luo, D. D. Dai, and L. Fu, Simulating moir´ e quantum matter with neural network (2024), arXiv:2406.17645 [cond-mat.str-el]

    2024 arXiv

  28. [36]

    von Glehn, J

    I. von Glehn, J. S. Spencer, and D. Pfau, A self- attention ansatz for Ab-initio quantum chemistry (2023), arXiv:2211.13672 [physics.chem-ph]

    2023 arXiv

  29. [37]

    R. Li, H. Ye, D. Jiang, X. Wen, C. Wang, Z. Li, X. Li, D. He, J. Chen, W. Ren, and L. Wang, For- ward Laplacian: A new computational framework for neural network-based variational Monte Carlo (2023), arXiv:2307.08214 [physics.comp-ph]

    2023 arXiv

  30. [38]

    R. B. Laughlin, Phys. Rev. Lett. 50, 1395 (1983)

    1983

  31. [39]

    Morf and B

    R. Morf and B. I. Halperin, Phys. Rev. B33, 2221 (1986)

    1986

  32. [40]

    M. B. Tavernier, E. Anisimovas, and F. M. Peeters, Phys. Rev. B 70, 155321 (2004)

    2004

  33. [41]

    M. B. Tavernier, E. Anisimovas, and F. M. Peeters, Phys. Rev. B 74, 125305 (2006)

    2006

  34. [42]

    Anisimovas, M

    E. Anisimovas, M. B. Tavernier, and F. M. Peeters, Phys. Rev. B 77, 045327 (2008)

    2008

  35. [43]

    T¨ ol¨ o, J

    E. T¨ ol¨ o, J. Suorsa, and A. Harju, Physica E: Low- dimensional Systems and Nanostructures 40, 1038–1041 (2008)

    2008

  36. [44]

    Wen and Z

    X.-G. Wen and Z. Wang, Pattern-of-zeros approach to fractional quantum Hall states and a classification of symmetric polynomial of infinite variables, in Conformal Field Theories and Tensor Categories (Springer Berlin Heidelberg, 2014) p. 33–65

    2014

  37. [45]

    Sodemann and A

    I. Sodemann and A. H. MacDonald, Phys. Rev. B 87, 245425 (2013)

    2013

  38. [46]

    Shi, E.-M

    Q. Shi, E.-M. Shih, M. V. Gustafsson, D. A. Rhodes, B. Kim, K. Watanabe, T. Taniguchi, Z. Papi´ c, J. Hone, and C. R. Dean, Nature Nanotechnology 15, 569–573 (2020). 10

    2020

  39. [47]

    Yoshioka, B

    D. Yoshioka, B. I. Halperin, and P. A. Lee, Phys. Rev. Lett. 50, 1219 (1983)

    1983

  40. [48]

    B. I. Halperin, Phys. Rev. Lett. 52, 1583 (1984)

    1984

  41. [49]

    F. D. M. Haldane and E. H. Rezayi, Phys. Rev. Lett. 54, 237 (1985)

    1985

  42. [50]

    E. V. Tsiper and V. J. Goldman, Physical Review B 64, 10.1103/physrevb.64.165311 (2001)

    2001 doi

  43. [51]

    X. Wan, K. Yang, and E. H. Rezayi, Phys. Rev. Lett. 88, 056802 (2002)

    2002

  44. [52]

    X. Wan, E. H. Rezayi, and K. Yang, Phys. Rev. B 68, 125307 (2003)

    2003

  45. [53]

    M. P. Zaletel, R. S. K. Mong, F. Pollmann, and E. H. Rezayi, Phys. Rev. B 91, 045115 (2015)

    2015

  46. [54]

    Ortiz, D

    G. Ortiz, D. M. Ceperley, and R. M. Martin, Phys. Rev. Lett. 71, 2777 (1993)

    1993

  47. [55]

    A. D. G¨ u¸ cl¨ u, G. S. Jeon, C. J. Umrigar, and J. K. Jain, Phys. Rev. B 72, 205327 (2005)

    2005

  48. [56]

    J. Zhao, Y. Zhang, and J. K. Jain, Phys. Rev. Lett. 121, 116802 (2018)

    2018

  49. [57]

    T. Zhao, A. C. Balram, and J. K. Jain, Phys. Rev. Lett. 130, 186302 (2023)

    2023

  50. [58]

    Taut, Journal of Physics: Condensed Matter 12, 3689–3710 (2000)

    M. Taut, Journal of Physics: Condensed Matter 12, 3689–3710 (2000)

    2000

  51. [59]

    W. M. C. Foulkes, L. Mitas, R. J. Needs, and G. Ra- jagopal, Rev. Mod. Phys. 73, 33 (2001)

    2001

  52. [60]

    Kr´ alik, A

    B. Kr´ alik, A. M. Rappe, and S. G. Louie, Phys. Rev. B 56, 4760 (1997)

    1997

  53. [61]

    Price and S

    R. Price and S. Das Sarma, Physical Review B 54, 8033–8043 (1996)

    1996

  54. [62]

    Ciftja, Results in Physics 7, 1674–1675 (2017)

    O. Ciftja, Results in Physics 7, 1674–1675 (2017)

    2017

  55. [63]

    Martens and R

    J. Martens and R. Grosse, Optimizing neural networks with Kronecker-factored approximate curvature (2020), arXiv:1503.05671 [cs.LG]

    2020 arXiv

  56. [64]

    Stokes, J

    J. Stokes, J. Izaac, N. Killoran, and G. Carleo, Quantum 4, 269 (2020)

    2020

  57. [65]

    B. I. Halperin, Phys. Rev. B 25, 2185 (1982)

    1982

  58. [66]

    X. G. Wen, Physical Review B 41, 12838–12844 (1990)

    1990

  59. [67]

    S. M. Reimann and M. Manninen, Rev. Mod. Phys. 74, 1283 (2002)

    2002

  60. [68]

    Teˇ sanovi´ c, F

    Z. Teˇ sanovi´ c, F. m. c. Axel, and B. I. Halperin, Phys. Rev. B 39, 8525 (1989)

    1989

  61. [69]

    S. M. Girvin, A. H. MacDonald, and P. M. Platzman, Phys. Rev. B 33, 2481 (1986)

    1986

  62. [70]

    Ceperley, Phys

    D. Ceperley, Phys. Rev. B 18, 3126 (1978)

    1978

  63. [71]

    R. M. Martin, L. Reining, and D. M. Ceperley, Inter- acting Electrons: Theory and Computational Approaches (Cambridge University Press, Cambridge, 2016)

    2016

  64. [72]

    D. P. . C. James S. Spencer, FermiNet (Github Reposi- tory) (2020)

    2020

  65. [73]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. Van- derPlas, S. Wanderman-Milne, and Q. Zhang, JAX: com- posable transformations of Python+NumPy programs (Github Repository) (2018)

    2018

  66. [74]

    D. D. Dai and L. Fu, Electron bubbles in highly excited states of the lowest Landau level (2024), arXiv:2407.09204. 11 Supplementary Material Detailed Architecture of Psiformer Here we mostly follow the original Psiformer paper [36]. As an ab-initio Ansatz, we start with only ...

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.