Pith. sign in

REVIEW 3 major objections 5 minor 47 references

This paper shows that neural quantum states, propagated by time-dependent variational methods, reproduce 1H NMR spectra of small molecules with spectral error below 10^-3, and that an exact interaction-frame transformation reduces the requi

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

2026-08-04 01:03 UTC pith:LNGNMZLL

load-bearing objection A solid proof-of-principle that NQS dynamics can match exact NMR spectra for small spins, with an exact interaction-frame trick that empirically cuts integration steps 8–40×; the theoretical justification is sloppier than the numerics, and the 14-spin demo is a benign disconnected case. the 3 major comments →

arxiv 2608.00178 v1 pith:LNGNMZLL submitted 2026-07-31 cond-mat.dis-nn

Neural Quantum States for Nuclear Magnetic Resonance Spectroscopy

Bharadwaj Chowdary Mummaneni , Bo Xing This is my paper
classification cond-mat.dis-nn PACS 76.60.-k
keywords neural quantum statesNMR spectroscopyrestricted Boltzmann machinetime-dependent variational principleprojected time-dependent variational Monte Carlointeraction framespin dynamicsspectral simulation
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.

The paper asks whether neural quantum states can turn computed NMR parameters into spectra without exponential Hilbert-space cost. For four molecules with 2 to 5 coupled protons, a restricted-Boltzmann-machine wavefunction propagated by TDVP reproduces all line positions and intensities with spectral mean-squared error below 10^-3; p-tVMC achieves the same features with larger, optimizer-controlled error. The paper identifies the real bottleneck as the steep growth in integration steps caused by chemical-shift precession, not network expressivity. An exact transformation to the chemical-shift interaction frame removes that stiffness, cutting steps roughly eightfold (3-spin) to tenfold or more (4- and 5-spin) at fixed accuracy, and makes a 14-spin sucrose simulation feasible via Monte Carlo sampling. If this scaling persists, it would connect routine DFT-computed shifts and couplings to experimental spectra for molecules beyond exact diagonalization.

Core claim

The central discovery is that the steep step-count growth in NQS-NMR simulation is numerical stiffness from the chemical-shift term H0 = Σ ω_i I_z,i, not a failure of the neural-network ansatz. Moving to the interaction frame of H0 — the Dirac picture, with |Ψ(t)⟩ = U0†(t)|Ψ_I(t)⟩ — removes the large one-body term from the generator exactly, leaving only coupling terms dressed by phases oscillating at chemical-shift differences. The transformation is exact and leaves the spectrum unchanged: for the RBM ansatz it amounts to a time-dependent drift of the visible biases, a_i(t) = a_i(0) − i(ω_i/2)t, and peak positions are restored analytically. In practice, this cuts the number of integration s

What carries the argument

The load-bearing object is the interaction-frame generator of Eq. (4), H_I(t) = 2π Σ_{i<j} J_{ij} [I_z,i I_z,j + (1/2)(e^{i(ω_i−ω_j)t} I_+^{(i)} I_-^{(j)} + h.c.)], obtained by moving into the rotating frame of the chemical-shift term H0 = Σ ω_i I_z,i. This transformation is exact: the spectrum is unchanged, and for the RBM the chemical-shift evolution becomes an analytic drift of the visible biases, a_i(t) = a_i(0) − i(ω_i/2)t. The generator removes the large one-body term that forces tiny integration steps, leaving a coupling-scale generator whose time dependence the state follows slowly. The RBM wavefunction of Eq. (6) supplies the variational manifold, and TDVP or p-tVMC supplies the pro

Load-bearing premise

The advantage rests on the assumption that the only numerically stiff part of the Hamiltonian is the one-body chemical-shift term, so that after applying it exactly the remaining time-dependent coupling generator can be integrated with large steps; if the oscillating phases between shifts still force tiny steps, or if strongly entangled states demand exponentially many hidden units, the 14-spin feasibility does not extend to realistic molecules.

What would settle it

Run interaction-frame TDVP on a synthetic 12-proton network with all spins coupled in one connected cluster (no isolated blocks), using α = 4, and compare the resulting spectrum to exact Krylov propagation at T = 1 s: if the step count needed for MSE < 10^-3 grows at the same steep rate as in the rotating frame, or if α must scale steeply with N to hold the error, the central claim fails.

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

If this is right

  • At the benchmarked sizes, accuracy is limited by integration and optimization error, not by the RBM's expressive capacity, shifting attention to integrators and update schemes rather than network architecture.
  • The interaction-frame transformation is exact and spectrum-preserving, so any future NQS or non-NQS propagator can adopt it as a cheap preconditioner for the chemical-shift term.
  • The 14-spin sucrose run is a feasibility point: it works because the coupling network splits into 7-, 5-, and 2-spin clusters, so the hardest part is the 7-spin block.
  • Spectral MSE in the 10^-4 to 10^-3 range is below typical experimental linewidths, so the accuracy is adequate for practical comparison with experiment if it persists at larger N.
  • For p-tVMC, the per-step infidelity provides an adaptive convergence diagnostic that TDVP lacks, potentially making optimization budgets self-tuning at larger scale.

Where Pith is reading between the lines

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

  • The interaction-frame idea is independent of the neural-network ansatz: tensor-network and restricted state-space NMR simulations that suffer from the same chemical-shift stiffness could absorb the one-body term analytically as well.
  • The sucrose result is a best-case large-system point; a molecule with 10–20 protons in a single connected coupling network is the natural next stress test, and volume-law entanglement could make the RBM fail exactly where the interaction frame has done its job.
  • Because the visible-bias drift is exact, the same trick should transfer directly to p-tVMC, combining the frame's step reduction with p-tVMC's error diagnostic.
  • One can test the entanglement-growth hypothesis directly by computing bipartite von Neumann entropy along the trajectory for a synthetic 10-spin connected network; if the worst-cut entropy approaches volume-law scaling, expect the RBM hidden-unit requirement to grow quickly.

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

3 major / 5 minor

Summary. The paper benchmarks Restricted Boltzmann Machine (RBM) neural quantum states (NQS) for simulating nuclear magnetic resonance spectra of small coupled proton spin systems. Using time-dependent variational principle (TDVP) and projected time-dependent variational Monte Carlo (p-tVMC), the authors compute 1H NMR spectra for four molecules with 2–5 spins and compare against exact diagonalization. They report spectral MSEs below 10^-3 for TDVP at optimized parameters, and find that the dominant cost is the number of integration steps, which grows with the chemical-shift bandwidth. An interaction-frame transformation is introduced that removes the one-body chemical-shift term exactly; the authors claim this reduces the required integration steps roughly eightfold (3 spins) to an order of magnitude or more (4–5 spins), and enables a Monte Carlo simulation of sucrose (14 spins). The paper also analyzes entanglement entropy and discusses limitations, including disconnected coupling clusters in sucrose.

Significance. If the results hold, the paper provides a credible proof-of-concept that NQS can reproduce exact NMR spectra at accuracy sufficient for practical comparison, and it identifies a potentially useful numerical trick (interaction-frame propagation) for reducing integration cost. The benchmarks are non-circular (compared with exact spectra), include multi-seed statistics, and the entanglement analysis is a valuable caveat. The claim of a general speed-up and the extension to a 14-spin molecule are, however, not fully supported by the theoretical argumentation and the specific choice of a factorizable test case. The paper is careful and honest in many of its stated limitations, which is a strength.

major comments (3)
  1. [§II.B, Eq. (4)] The claim that the interaction frame allows larger steps because 'the rate at which the state changes is bounded by the norm of the generator, not by how rapidly its phases oscillate' is not a correct criterion for step-size selection in a second-order integrator for a time-dependent generator. Local error of Heun's method involves derivatives of H_I(t), so the oscillatory phases at Δω must be resolved. For the reported 4-spin run (N_t≈200, h=5×10^-3, stiffness 36.9), the phase increment Δω h is O(1–10), which should make the trapezoidal approximation of the flip-flop terms unreliable. The paper needs either a local error analysis for this time-dependent Hamiltonian or an empirical convergence study (e.g., comparing with a smaller h or a higher-order integrator) to support the step-reduction claim.
  2. [§IV.A, Table I vs. p-tVMC paragraph] The TDVP/p-tVMC comparison is not apples-to-apples: TDVP used molecule-specific optimal α (Table I) while p-tVMC used α=4 and n_iter=50 (or 75) throughout. The reported p-tVMC MSEs (2.6×10^-3 to 3.7×10^-2) are substantially worse than TDVP's. If the paper aims to demonstrate that both methods 'reproduce the same features', the quantitative comparison is incomplete. An accuracy-matched comparison (e.g., tuning α and n_iter per molecule, or reporting MSE at matching wall time) would strengthen the claim.
  3. [§IV.B, Fig. 7] The sucrose (N=14) feasibility point is a weak demonstration of scalability because the J-coupling network splits into three disconnected clusters (7, 5, and 2 spins). As Fig. 7 shows, the state remains a product over clusters, so the effective problem size is 7 spins. The abstract's statement that the method 'enables a 14-spin molecule (sucrose) to be accurately propagated' is misleading without this caveat. A connected 14-spin system with genuine entanglement would be a more convincing test of the NQS approach at larger N.
minor comments (5)
  1. [§IV.B] The sentence 'The successful simulation of the 14-proton molecule points works in favor of NQ.' is malformed; it should read 'points work in favor of NQS' or similar.
  2. [§III.C] The dense propagator for p-tVMC is precomputed and cached, which limits the method to N≤12; this is mentioned but could be emphasized more clearly in the abstract and conclusions.
  3. [§II.B] The 'stiffness' ratio max_{ij}|ω_i−ω_j|/(2πJ_ij) is used to quantify the numerical difficulty, but for a multi-spin system the relevant frequency is the spectral gap of H0, not a single pair ratio. Consider clarifying or using a more conventional definition.
  4. [§I/References] The paper would benefit from a brief comparison with existing restricted state-space methods (e.g., Spinach) to contextualize the claimed advantages; the discussion section mentions this as future work, but not even a qualitative comparison is given.
  5. [§IV.A] The p-tVMC results are reported only for α=4 and n_iter=50 (and one n_iter=75 point for the 5-spin system). Since the paper argues p-tVMC is competitive, a more thorough parameter study (e.g., n_iter sweep for each molecule) would be useful.

Circularity Check

0 steps flagged

No significant circularity: spectra are benchmarked against independently computed exact references; the interaction-frame transformation is exact and standard, and the speed-up claim is empirical.

full rationale

The paper's central claim is that NQS (RBM) with TDVP and p-tVMC reproduces 1H NMR spectra for 2-5 spins, benchmarked against exact diagonalization / sparse Krylov references. This is an external, independent benchmark: the Hamiltonian parameters come from experimental databases, the reference spectra are computed by dense exponentiation or Krylov propagation, and the reported MSEs measure the variational/integration error against those references. Choosing the timestep or hidden-unit ratio by minimizing MSE is a standard numerical benchmark procedure, not fitting a parameter to a target and then calling it a prediction. The interaction-frame transformation of Eq. (4) is presented explicitly as the standard Dirac picture, with |Psi(t)> = U0^dag(t)|Psi_I(t)> reproducing the rotating-frame state exactly; the paper does not redefine any quantity in terms of the claimed result. The reported step-count reduction is an empirical observation from the same exact-reference comparison, not a conclusion forced by definition. The only self-citations are [23] and [27], co-authored by B. Xing, but they are background references for p-tVMC variants and ground-state search; the core p-tVMC method is attributed to [22] (Sinibaldi et al.), and no load-bearing conclusion rests on a self-citation or on an imported uniqueness theorem. The skeptical concern about the Heun integrator needing to resolve oscillatory phases in H_I(t) is a legitimate numerical-analysis and correctness concern, but it is not circularity: it challenges whether the stated error analysis is valid, not whether the derivation reduces to its inputs. The paper itself includes caveats (second-order integrator, sucrose as a feasibility point, open questions about strongly entangled systems) that further indicate the claims are presented as empirical benchmarks rather than as definitionally forced results.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities or exotic parameters. It does fit α, Δt, ϵ, and λ to the benchmark. The key unproved assumptions are that the RBM remains expressive for the systems considered and that the interaction-frame transformation actually removes the dominant cost without introducing new numerical stiffness, which is only shown on four molecules plus one easy 14-spin case.

free parameters (4)
  • hidden unit ratio α = α=4 for p-tVMC; α=10 (N=2,3,5), α=1 (N=4) for TDVP; α=2 for sucrose
    Network expressive capacity is tuned for each system; the 4-spin optimal at α=1 is an acknowledged anomaly.
  • timestep Δt = 0.005 s (N=2), 0.002 s (N=3), 2.5×10^-4 s (N=4), 1.0×10^-4 s (N=5)
    The timestep that minimizes spectral MSE is chosen per system; this is a fitted parameter for the benchmark.
  • QGT regularization ϵ = 0.01
    Chosen by hand for TDVP numerical stability; the paper notes it biases the equations of motion.
  • apodization decay λ = 3.0 s^-1
    Chosen to mimic ~1 Hz experimental linewidth; affects the spectral MSE comparison.
axioms (4)
  • domain assumption Variational Monte Carlo with 4096 samples estimates QGT/forces and infidelity gradients with small enough bias for the central accuracy claim.
    The TDVP equations (8-10) and p-tVMC estimator (13) rely on stochastic sampling; the paper notes that for N≤5 the 4096 samples exceed the Hilbert space dimension, which is effectively exact summation. For sucrose (N=14) the sampling is genuinely stochastic, but the result is only a feasibility point.
  • domain assumption The effective NMR Hamiltonian (1) is the correct model for the simulated spectra.
    The paper uses the standard high-temperature rotating-frame Hamiltonian, including only isotropic J-couplings and neglecting relaxation; acceptable for these benchmarks, but (as the paper notes) real experiments include T1/T2 relaxation.
  • domain assumption Spectral MSE over a normalized region is a meaningful error metric.
    The MSE is computed after normalizing to unit maximum, and the paper notes 4- and 5-spin systems achieve lower MSE than 3-spin, which they attribute to differing spectral complexity. This is a stated choice, not an external benchmark.
  • domain assumption The RBM can represent the time-evolved states with the chosen α (expressivity assumption).
    The paper explicitly says 'It is not known a priori which regime NMR time evolution falls into.' The 4-spin α=1 anomaly and the acknowledged volume-law risk make this a load-bearing assumption.

pith-pipeline@v1.3.0-alltime-deepseek · 15087 in / 7231 out tokens · 57396 ms · 2026-08-04T01:03:00.123011+00:00 · methodology

0 comments
read the original abstract

Predicting a nuclear magnetic resonance (NMR) spectrum from first principles requires propagating a quantum state of dimension $2^N$ for $N$ coupled spins, which becomes intractable beyond larger $N$. We benchmark Neural Quantum States (NQS), a class of variational quantum states expressed as an artificial neural network, as an alternative representation for this problem. Using two propagation methods, the Time-Dependent Variational Principle (TDVP) and projected time-dependent Variational Monte Carlo (p-tVMC), we compute the $^1$H spectra of four ($2 \to 5$ spins) experimentally parameterized molecules. TDVP reproduces all line positions and intensities with average spectral mean squared errors of $<10^{-3}$; p-tVMC reproduces the same features, with accuracy determined by its per-step optimization parameters. One dominant obstacle to larger systems is the steep growth of the number of integration steps with spectral bandwidth, which can be removed by propagating in the interaction frame of the chemical-shifted Hamiltonian. Retaining the same accuracy, this reduces the number of integration steps roughly eightfold for the 3-spin system and by at least an order of magnitude for the 4- and 5-spin systems, and it enables a 14-spin molecule (sucrose) to be accurately propagated via Monte Carlo sampling.

Figures

Figures reproduced from arXiv: 2608.00178 by Bharadwaj Chowdary Mummaneni, Bo Xing.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p010_7.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

47 extracted references · 1 linked inside Pith

  1. [1]

    R. R. Ernst, G. Bodenhausen, and A. Wokaun,Prin- ciples of Nuclear Magnetic Resonance in One and Two Dimensions(Clarendon Press, Oxford, 1987)

    1987

  2. [2]

    M. H. Levitt,Spin Dynamics: Basics of Nuclear Mag- netic Resonance, 2nd ed. (John Wiley & Sons, 2008)

    2008

  3. [3]

    Helgaker, M

    T. Helgaker, M. Jaszu´ nski, and K. Ruud, Ab initio meth- ods for the calculation of NMR shielding and indirect spin–spin coupling constants, Chemical Reviews99, 293 (1999)

    1999

  4. [4]

    M. W. Lodewyk, M. R. Sieber, and D. J. Tantillo, Com- putational prediction of 1H and 13C chemical shifts: A useful tool for natural product, mechanistic, and syn- thetic organic chemistry, Chemical Reviews112, 1839 (2012)

    2012

  5. [5]

    Wolinski, J

    K. Wolinski, J. F. Hinton, and P. Pulay, Efficient im- plementation of the gauge-independent atomic orbital method for NMR chemical shift calculations, Journal of the American Chemical Society112, 8251 (1990)

    1990

  6. [6]

    M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, G. Scalmani, V. Barone, G. A. Petersson, H. Nakatsuji,et al., Gaussian 16, revi- sion C.01 (2016)

    2016

  7. [7]

    Neese, F

    F. Neese, F. Wennmohs, U. Becker, and C. Riplinger, The ORCA quantum chemistry program package, The Journal of Chemical Physics152, 224108 (2020)

    2020

  8. [8]

    D. V. Savostyanov, S. V. Dolgov, J. M. Werner, and I. Kuprov, Exact NMR simulation of protein-size spin systems using tensor train formalism, Physical Review B 90, 085139 (2014)

    2014

  9. [9]

    Kuprov, N

    I. Kuprov, N. Wagner-Rundell, and P. J. Hore, Polyno- mially scaling spin dynamics simulation algorithm based on adaptive state-space restriction, Journal of Magnetic Resonance189, 241 (2007)

    2007

  10. [10]

    H. J. Hogben, M. Krzystyniak, G. T. P. Charnock, P. J. Hore, and I. Kuprov, Spinach – a software library for sim- ulation of spin dynamics in large spin systems, Journal 12 of Magnetic Resonance208, 179 (2011)

    2011

  11. [11]

    J. A. Jones, Controlling NMR spin systems for quantum computation, Progress in Nuclear Magnetic Resonance Spectroscopy140–141, 49 (2024)

    2024

  12. [12]

    Lloyd, Universal quantum simulators, Science273, 1073 (1996)

    S. Lloyd, Universal quantum simulators, Science273, 1073 (1996)

    1996

  13. [13]

    Preskill, Quantum computing in the NISQ era and beyond, Quantum2, 79 (2018)

    J. Preskill, Quantum computing in the NISQ era and beyond, Quantum2, 79 (2018)

    2018

  14. [14]

    worst cut

    In this case, the coupling structure can further sim- plify molecules. More broadly, the appeal of NQS for larger systems is that, unlike tensor networks, they are not bounded by connectivity or entanglement across a cut [43]. They can remain efficient as that entanglement grows. Two caveats bound this comparison. First, both frames used the same second-o...

  15. [15]

    Khedri, P

    A. Khedri, P. Stadler, K. Bark, M. Lodi, R. Reiner, N. Vogt, M. Marthaler, and J. Lepp¨ akangas, The im- pact of noise on the simulation of NMR spectroscopy on NISQ devices, arXiv preprint arXiv:2404.18903 10.48550/arXiv.2404.18903 (2024)

    doi:10.48550/arxiv.2404.18903 2024

  16. [16]

    K. R. Fratuset al., Can a quantum computer simulate nuclear magnetic resonance spectra better than a classical one?, arXiv preprint arXiv:2508.06448 10.48550/arXiv.2508.06448 (2025)

    doi:10.48550/arxiv.2508.06448 2025

  17. [17]

    Or´ us, A practical introduction to tensor networks: Matrix product states and projected entangled pair states, Annals of Physics349, 117 (2014)

    R. Or´ us, A practical introduction to tensor networks: Matrix product states and projected entangled pair states, Annals of Physics349, 117 (2014)

    2014

  18. [18]

    S. R. White and A. E. Feiguin, Real-time evolution us- ing the density matrix renormalization group, Physical Review Letters93, 076401 (2004)

    2004

  19. [19]

    Vidal, Efficient simulation of one-dimensional quan- tum many-body systems, Physical Review Letters93, 040502 (2004)

    G. Vidal, Efficient simulation of one-dimensional quan- tum many-body systems, Physical Review Letters93, 040502 (2004)

    2004

  20. [20]

    Carleo and M

    G. Carleo and M. Troyer, Solving the quantum many- body problem with artificial neural networks, Science 355, 602 (2017), seminal work introducing Neural Quan- tum States

    2017

  21. [21]

    Schmitt and M

    M. Schmitt and M. Heyl, Quantum many-body dynamics in two dimensions with artificial neural networks, Physi- cal Review Letters125, 100503 (2020)

    2020

  22. [22]

    X. Yuan, S. Endo, Q. Zhao, Y. Li, and S. C. Benjamin, Theory of variational quantum simulation, Quantum3, 191 (2019)

    2019

  23. [23]

    Sinibaldi, C

    A. Sinibaldi, C. Giuliani, G. Carleo, and F. Vicentini, Un- biasing time-dependent variational monte carlo by pro- jected quantum evolution, Quantum7, 1131 (2023)

    2023

  24. [24]

    Zhang, B

    W. Zhang, B. Xing, X. Xu, and D. Poletti, Paths to- wards time evolution with larger neural-network quan- tum states, Computer Physics Communications312, 109577 (2025)

    2025

  25. [25]

    Gao and L.-M

    X. Gao and L.-M. Duan, Efficient representation of quan- tum many-body states with deep neural networks, Na- ture Communications8, 662 (2017)

    2017

  26. [26]

    Hibat-Allah, M

    M. Hibat-Allah, M. Ganahl, L. E. Hayward, R. G. Melko, and J. Carrasquilla, Recurrent neural network wave func- tions, Physical Review Research2, 023358 (2020)

    2020

  27. [27]

    D. Pfau, J. S. Spencer, A. G. Matthews, and W. M. C. Foulkes, Ab initio solution of the many-electron Schr¨ odinger equation with deep neural networks, Physi- cal Review Research2, 033429 (2020)

    2020

  28. [28]

    Zhang, X

    W. Zhang, X. Xu, Z. Wu, V. Balachandran, and D. Po- letti, Ground state search by local and sequential updates of neural network quantum states, Phys. Rev. B107, 165149 (2023)

    2023

  29. [29]

    Dashti, W

    H. Dashti, W. M. Westler, M. Tonelli, J. R. Wedell, J. L. Markley, and H. R. Eghbalnia, Spin system modeling of nuclear magnetic resonance spectra for applications in metabolomics and small molecule screening, Analytical Chemistry89, 12201 (2017), gISSMO database

    2017

  30. [30]

    HQS Quantum Simulations GmbH, HQS quantum sim- ulations: Struqture and cheshire NMR parameter database (2023), accessed: 2024

    2023

  31. [31]

    Bloch, Nuclear induction, Physical Review70, 460 (1946)

    F. Bloch, Nuclear induction, Physical Review70, 460 (1946)

    1946

  32. [32]

    E. L. Hahn, Spin echoes, Physical Review80, 580 (1950)

    1950

  33. [33]

    Abragam,The Principles of Nuclear Magnetism(Ox- ford University Press, Oxford, 1961)

    A. Abragam,The Principles of Nuclear Magnetism(Ox- ford University Press, Oxford, 1961)

    1961

  34. [34]

    C. P. Slichter,Principles of Magnetic Resonance, 3rd ed. (Springer-Verlag, Berlin, 1990)

    1990

  35. [35]

    Kubo, Statistical-mechanical theory of irreversible processes

    R. Kubo, Statistical-mechanical theory of irreversible processes. I. general theory and simple applications to magnetic and conduction problems, Journal of the Phys- ical Society of Japan12, 570 (1957)

    1957

  36. [36]

    Becca and S

    F. Becca and S. Sorella,Quantum Monte Carlo Ap- proaches for Correlated Systems(Cambridge University Press, Cambridge, 2017)

    2017

  37. [37]

    A. D. McLachlan, A variational solution of the time- dependent Schr¨ odinger equation, Molecular Physics8, 39 (1964)

    1964

  38. [38]

    Haegeman, J

    J. Haegeman, J. I. Cirac, T. J. Osborne, I. Piˇ zorn, H. Ver- schelde, and F. Verstraete, Time-dependent variational principle for quantum lattices, Physical Review Letters 107, 070601 (2011)

    2011

  39. [39]

    Sorella, Green function Monte Carlo with stochastic reconfiguration, Physical Review Letters80, 4558 (1998)

    S. Sorella, Green function Monte Carlo with stochastic reconfiguration, Physical Review Letters80, 4558 (1998)

    1998

  40. [40]

    Stokes, J

    J. Stokes, J. Izaac, N. Killoran, and G. Carleo, Quantum natural gradient, Quantum4, 269 (2020)

    2020

  41. [41]

    Hackl, T

    L. Hackl, T. Guaita, T. Shi, J. Haegeman, E. Demler, and J. I. Cirac, Geometry of variational methods: dynamics of closed quantum systems, SciPost Physics9, 48 (2020)

    2020

  42. [42]

    D. P. Kingma and J. Ba, Adam: A method for stochas- tic optimization, arXiv preprint arXiv:1412.6980 (2015), presented at ICLR 2015

    Pith/arXiv arXiv 2015

  43. [43]

    Vicentini, D

    F. Vicentini, D. Hofmann, A. Szab´ o, D. Wu, C. Roth, C. Giuliani, G. Pescia, J. Nys, V. Vargas-Calder´ on, N. Astrakhantsev, and G. Carleo, NetKet 3: Machine learning toolbox for many-body quantum systems, Sci- Post Physics Codebases , 7 (2022)

    2022

  44. [44]

    D.-L. Deng, X. Li, and S. Das Sarma, Quantum entan- glement in neural network states, Physical Review X7, 021021 (2017)

    2017

  45. [45]

    Chen and M

    A. Chen and M. Heyl, Empowering deep neural quantum states through efficient optimization, Nature Physics20, 1476 (2024)

    2024

  46. [46]

    L. L. Viteritti, R. Rende, and F. Becca, Transformer vari- ational wave functions for frustrated quantum spin sys- tems, Physical Review Letters130, 236401 (2023)

    2023

  47. [47]

    Breuer and F

    H.-P. Breuer and F. Petruccione,The Theory of Open Quantum Systems(Oxford University Press, 2002)

    2002