Pith. sign in

REVIEW 3 major objections 5 minor 142 references

This review chapter argues that Gaussian Process States, an exponentiated kernel model of many-body wavefunctions, unify strongly correlated quantum chemistry with machine learning on classical data.

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 ↗

A review of Bayesian Gaussian Process States for ab initio quantum chemistry, with new MNIST digit classification results reaching about 1.6% test error.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection A competent, honest review of the authors' own GPS framework; the MNIST result is modest and the 64-H 'state-of-the-art' claim overreaches, but the chapter is well written and worth refereeing. the 3 major comments →

arxiv 2508.21729 v1 pith:YJ77JN5F submitted 2025-08-29 cond-mat.str-el physics.chem-phphysics.comp-phquant-ph

Bayesian perspectives for quantum states and application to ab initio quantum chemistry

Yannic Rath , Massimo Bortone , George H. Booth This is my paper
classification cond-mat.str-el physics.chem-phphysics.comp-phquant-ph
keywords Gaussian Process Statesquantum chemistrysecond quantizationvariational Monte CarloBayesian inferencestrong correlationmachine learning wavefunctionstensor decomposition
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 reading

This chapter makes the case that a single family of Bayesian wavefunction models—Gaussian Process States (GPS) and their 'quantum' generalization qGPS—can serve as a common tool for strongly correlated quantum chemistry and for machine learning on classical data. The models represent the logarithm of a many-electron wavefunction amplitude as a weighted sum of kernel comparisons, or equivalently as the exponential of a low-rank tensor decomposition, so accuracy improves systematically as a support dimension grows. In second-quantized ab initio problems this representation captures strong correlation in hydrogen chains and lattices, and a Slater-Jastrow GPS yields a variational energy for 64 hydrogen atoms that the authors argue is a new state-of-the-art result. The same model, trained with Bayesian sweeping, classifies MNIST digits with roughly 1.6–1.7% test error, illustrating the reverse transfer from quantum states to classical pattern extraction. The reason this matters is that it offers a route to chemical accuracy in regimes where standard methods such as density functional theory fail, without requiring a hand-designed neural-network architecture.

Core claim

The Gaussian Process State is presented as a unified representation for quantum many-body states: wavefunction amplitudes are written as the exponential of a kernel regression over configurations, which expands into correlation features over all possible plaquettes, so the ansatz is product-separable and systematically improvable in the number of support configurations. Promoting the discrete support configurations to continuous product states gives the quantum GPS, whose log-amplitude is a sum of products of local tensors—the exponential of a CP decomposition—and can be trained either by Bayesian sweeping with a marginal-likelihood regularizer or variationally in Monte Carlo. In ab initio c

What carries the argument

The central object is the Gaussian Process State, Psi(n)=exp(Σ w_{n'} k(n,n')), with the exponential kernel k(n,n')=exp(-h(n,n')) where h is a weighted Hamming distance. Expanding the kernel generates correlation features over all possible plaquettes—single-site, two-site, many-site—so the ansatz is product-separable and can be systematically improved by adding support configurations. The 'quantum' variant qGPS replaces discrete support configurations with continuous product states and writes the log-amplitude as a sum over M products of local tensors, i.e. the exponential of a CP decomposition; the tensor's 4×L×M parameters are optimized either by Bayesian regression (relevance vector machi

Load-bearing premise

The framework's promise rests on the premise that the numerical optimization bottleneck—not the model's expressiveness—limits GPS/qGPS in signed, higher-dimensional wavefunctions; the chapter's own results show this is not yet solved.

What would settle it

Optimize a qGPS with support dimensions M = 8, 16, 32, 64, and 96 for the 4×4 hydrogen lattice in a localized orbital basis and record the variational energy per atom at each M; if the energy does not monotonically decrease toward the full configuration interaction value, the claimed systematic improvability fails. A complementary test: if increasing M in a canonical molecular-orbital basis eventually matches the local-basis accuracy, the basis-dependence claim is wrong, whereas if it plateaus orders of magnitude higher, the optimization-bottleneck premise is confirmed.

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

If this is right

  • Second-quantized machine-learned wavefunctions become a practical alternative to first-quantized neural-network ansatze in strongly correlated chemistry, especially when combined with explicit anti-symmetrization via a Slater determinant or backflow.
  • Localized orbital representations should be preferred over canonical molecular-orbital bases for this class of ansatze, because they avoid intricate sign structures and reduce the local-energy evaluation cost to O(N^2).
  • A single hyperparameter—the support dimension M—controls expressivity, giving a clean knob for systematically improvable, regularized wavefunction fits.
  • The same qGPS functional form can be trained by Bayesian sweeping for classical classification tasks, placing tensor-product-style models in the machine-learning toolbox without neural-network architecture search.
  • The variational energies on hydrogen systems provide references at geometries where standard coupled-cluster or density-matrix-renormalization-group methods struggle, useful for benchmarking other methods.

Where Pith is reading between the lines

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

  • The authors leave implicit that the Bayesian sweeping regularizer could be lifted out of quantum chemistry and applied to other multilinear models such as matrix product states or tensor trains used for generative modeling, transferring the overfitting resistance shown here.
  • If the optimization bottleneck is indeed the crux, improved optimizers for GPS—better stochastic estimators, curvature information, or annealing of the support dimension—would likely benefit the whole neural-quantum-state family, not just this ansatz.
  • A natural extension the paper does not pursue is learning the grayscale embedding used for MNIST inputs rather than fixing it to a linear function; the framework's own multilinear structure suggests this could substantially improve classification accuracy.
  • The basis-dependence result suggests that orbital choice should be treated as a core hyperparameter for any second-quantized machine-learned wavefunction, a conclusion that reaches beyond GPS to fermionic neural-network states generally.
Share X Bluesky LinkedIn Reddit HN

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 manuscript is a review-style chapter presenting the Gaussian Process State (GPS) family of second-quantized many-body wavefunction ansatzes. It derives the original kernel-based GPS (Eq. 7), the qGPS continuous-support generalization (Eq. 17), and connects these forms to Bayesian regression, relevance vector machines, correlator product states, tensor-network decompositions, and neural-network architectures. The authors then describe variational Monte Carlo applications to ab initio hydrogen chains and lattices, including autoregressive, Slater-Jastrow, and backflow variants, and report an MNIST digit-classification demonstration using qGPS in a one-vs-rest scheme. The chapter claims that GPS/qGPS unifies multiple paradigms and can serve as a novel tool for ab initio chemistry while also informing quantum-inspired machine learning.

Significance. The chapter provides a useful and generally clear synthesis of the GPS framework. The derivations in Sec. 2 are internally consistent, and the connection between exponential kernels, CP decompositions, and restricted matrix product states is pedagogically valuable. A concrete positive result is the Bayesian sweeping advantage over Adam in the supervised-learning setting (Fig. 5). However, the manuscript contains no new ab initio benchmark data; the flagship numerical results are imported from the authors' earlier papers. The paper's central ambition is constrained by its own admission that optimization of signed, higher-dimensional states remains an open problem, and several claims in the abstract and in Sec. 3.3.2 are stronger than the evidence presented. The chapter is best read as a perspective and review rather than as an established general solution for quantum chemistry.

major comments (3)
  1. [Sec. 5; Sec. 3.3.2, Fig. 8] The abstract's claim that these models 'can be used as novel tools to compute ab initio chemical properties' goes beyond what the variational results in this chapter demonstrate. In Fig. 8(left), the autoregressive GPS (Eq. 23) has relative energy errors orders of magnitude larger in a canonical Hartree-Fock basis than in a Boys-localized basis for the same 1D H16 chain, and the 2D 4x4 lattice is orders of magnitude worse than the 1D chain in both bases. Since the same functional form with complex parameters can formally represent signed states, this discrepancy points to optimization/learning failure rather than to a fundamental representational limit of Eq. (17). Sec. 5 explicitly concedes that the core optimization challenges 'have not yet seen general success in quantum chemistry.' Thus 'systematically improvable' is currently an expressivity statement, not a demonstrated practical p
  2. [Sec. 3.3.2, Fig. 9] The statement that the 64-hydrogen Slater-Jastrow GPS result 'represents a new state-of-the-art result' is not supported by evidence within this manuscript. The data shown in Fig. 9 are imported from the authors' Ref. [74]; no independent calculation or new validation is provided. The comparison consists of a single converged CCSD(T) point at a compressed geometry plus lower-bound variational-2RDM data at stretched geometries. Lower-bound data can rule out energies below the bound but cannot by itself confirm the quality of a variational energy. Please either present this as a review of prior work with the original reference's supporting analysis, or temper the 'new state-of-the-art' language accordingly.
  3. [Sec. 4, Eq. (27), Fig. 10] The quantum-inspired ML claim rests on a single MNIST demonstration whose benchmark comparison is not as meaningful as presented. The test errors of about 1.6-1.7% at M=200 are quoted as being 'within the range of other state-of-the-art results' because the shaded band in Fig. 10 spans from 0.09% to 7.53%. This range is two orders of magnitude wide and includes the best and worst results from a website list; merely falling inside it does not indicate competitiveness. In addition, the linear greyscale embedding in Eq. (27) is an ad hoc modelling choice, and no study of alternative embeddings is reported. Please label this as a proof-of-principle demonstration and soften the interpretation.
minor comments (5)
  1. [Sec. 2.6] 'achronym' should be 'acronym'.
  2. [Sec. 2.3, Eq. (9)] The Taylor expansion in Eq. (9) uses sums running to L-1, while the Hamming distance in Eq. (8) runs to L. Please clarify whether the upper limit is intentional or a typo.
  3. [Sec. 2.4, Eq. (14)] The likelihood noise variance in Eq. (14) involves division by |e^{phi(n)}|^2; if any training amplitude is exactly zero, the formula is singular. Please specify how zero amplitudes are handled in practice.
  4. [Fig. 8 caption] Unlike Figs. 7 and 9, the Fig. 8 caption does not indicate whether the data are 'adjusted' or directly reproduced from Ref. [68]. Please add a provenance note for consistency.
  5. [References] Reference [101] ('Block') is incomplete: no author names, journal, or year are given. Please complete the citation.

Circularity Check

0 steps flagged

No significant circularity: GPS/qGPS is a constructive ansatz (kernel feature resummation / exponentiated CP), and the claimed results are benchmarked against external references; self-citations are review content rather than load-bearing reductions.

full rationale

The core GPS construction is not circular. Equation (7) defines the GPS as exp(sum_i w_i k(n,n'_i)), with the exponential kernel of Eq. (8) obtained by Taylor-resumming all plaquette correlation features. The qGPS form in Eq. (17) is obtained by promoting classical support configurations to product states, and the paper explicitly identifies it as an exponentiated CP tensor decomposition. These are mathematical definitions and identities, not fitted quantities relabeled as predictions. The ab initio claims are tested against external references: autoregressive GPS energies are compared with FCI in Fig. 8; the Slater-Jastrow GPS and backflow CPD energies are compared with DMRG; and the 64-hydrogen-atom result is compared with CCSD(T) and lower-bound variational-2RDM data from Ref. [89]. The MNIST experiment is a fresh train/test transfer benchmark with external state-of-the-art error ranges. The paper does rely heavily on the authors' own prior work (Refs. [43,44,49,68,74,75]), but those works are externally falsifiable, and no argument in this chapter reduces to a self-citation for its mathematical content. The chapter also candidly flags its own limitation in Sec. 5 — optimization challenges "have not yet seen general success in quantum chemistry" — and in Sec. 3.3.2 notes that systematic improvability "can not always be guaranteed" in VMC. These are correctness and evidence caveats, not circularity. No equation was found in which an output is identical to an input by construction, no fitted parameter is renamed as a prediction, and no author-imported uniqueness theorem is used to force the ansatz.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

No new physical entities are introduced; the GPS and qGPS ansatzes are adopted from the authors' prior work (Refs [43,49]). The load-bearing choices are hyperparameters (support dimension, kernel weighting, noise level, linear embedding) and modeling assumptions about kernel validity, Hamiltonian sparsity, low-rank structure, and embedding sufficiency.

free parameters (4)
  • Support dimension M for GPS/qGPS = M = 1, 50, 100, 200 (MNIST); M = 5, 16 (J1-J2); M = 96 (64 H atoms)
    Hyperparameter controlling model expressivity, chosen by hand; directly controls the MNIST error curves in Fig 10 and the qGPS benchmarks.
  • Kernel weighting function f(i) = not specified numerically in this chapter
    Defines the exponential kernel in Eq 8; controls the emphasis of correlation features of different order and range; a hyperparameter from Refs [43,44].
  • Likelihood noise hyperparameter sigma-tilde^2 = updated by marginal likelihood maximization
    Appears in Eq 14; sets the target accuracy for fitting wavefunction amplitudes and is a free hyperparameter in the Bayesian regression.
  • Linear embedding coefficients epsilon^(0) and epsilon^(1) = optimized during training
    Variational parameters for the MNIST greyscale embedding (Eq 27); the paper states that alternative embeddings require further investigation, so this is an ad hoc modeling choice.
axioms (5)
  • standard math Gaussian process regression assumptions: multivariate Gaussian prior and likelihood lead to closed-form posterior (Eqs 11-13).
    Invoked in Sec 2.4 for the Bayesian optimization of GPS weights; standard Gaussian conjugate analysis.
  • domain assumption The exponential kernel (Eq 8) is a valid positive semi-definite kernel and its Taylor expansion (Eq 9) resums all plaquette-based features.
    The kernel construction in Sec 2.3 underpins the GPS expressibility; the resummation over all plaquettes is asserted analytically.
  • domain assumption The two-electron integral tensor becomes sparse enough in a local orbital basis that screening with thresholds of 1e-5 to 1e-9 Eh preserves accuracy.
    Used in Sec 3.3.1 and Fig 7 to justify O(N^2) local energy evaluation; the variational error from screening is not quantified in this chapter.
  • ad hoc to paper The log-wavefunction amplitudes admit a low-rank CP decomposition with compact support dimension M for the target chemical states.
    Eqs 17-18 assume that M is sufficient; exact representation may require rank up to exponential in system size, so compactness is an empirical assumption.
  • ad hoc to paper A linear embedding of MNIST greyscale values is sufficient for the qGPS classifier.
    Eq 27 uses f(n_i) = eps^(0) + eps^(1) n_i; the paper explicitly calls for further investigation of alternative embeddings.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Bayesian perspectives for quantum states and application to ab initio quantum chemistry." pith.science (2026). https://pith.science/paper/YJ77JN5F

@misc{pith2026250821729,
  author       = {Pith},
  title        = {Pith review of: Bayesian perspectives for quantum states and application to ab initio quantum chemistry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YJ77JN5F}},
  note         = {Machine review of arXiv:2508.21729}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The quantum many-electron problem is not just at the heart of condensed matter phenomena, but also essential for first-principles simulation of chemical phenomena. Strong correlation in chemical systems are prevalent and present a formidable challenge in the simulation of these systems, while predictive phenomena in this domain often also requires a demanding level of accuracy to inform chemical behavior. Efficient representations of the many-electron states of chemical systems are therefore also being inspired by machine learning principles to provide an alternative to established approaches. In this chapter, we review recent progress in this endeavor for quantum chemical problems represented in second quantization, and the particular challenges present in this field. In particular, we focus on the application of Gaussian Process States emerging from efficient representations of the many-body wavefunction with rigorous Bayesian modeling frameworks, allowing for the unification of multiple paradigms under a common umbrella. We show how such models (and other representations derived from machine learning) can be used as novel tools to compute ab initio chemical properties, while in turn also informing the design of machine learning models to extract correlation patterns in classical data.

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

142 extracted references · 39 canonical work pages · 5 internal anchors

  1. [1]

    Challenges for Density Functional Theory

    Aron J. Cohen, Paula Mori-S ´anchez, and Weitao Yang. “Challenges for Density Functional Theory”. In:Chem. Rev.112.1 (Jan. 2012), pp. 289–320. issn: 0009-2665. doi: 10.1021/cr200107z. (Visited on 10/07/2024)

    doi:10.1021/cr200107z 2012

  2. [2]

    Why Is Quantum Chemistry So Complicated?

    Jack Simons. “Why Is Quantum Chemistry So Complicated?” In: JACS 145.8 (2023), pp. 4343–4354. doi: 10.1021/jacs.2c13042

    work page doi:10.1021/jacs.2c13042 2023

  3. [3]

    Spiers Memorial Lecture: Quantum chemistry, clas- sical heuristics, and quantum advantage

    Garnet Kin-Lic Chan. “Spiers Memorial Lecture: Quantum chemistry, clas- sical heuristics, and quantum advantage”. In:Faraday Discuss.254 (0 2024), pp. 11–52. doi: 10.1039/D4FD00141A

    doi:10.1039/d4fd00141a 2024

  4. [4]

    Elec- tronic Landscape of the P-cluster of Nitrogenase as Revealed through Many- Electron Quantum Wavefunction Simulations

    Zhendong Li, Sheng Guo, Qiming Sun, and Garnet Kin-Lic Chan. “Elec- tronic Landscape of the P-cluster of Nitrogenase as Revealed through Many- Electron Quantum Wavefunction Simulations”. In: Nat. Chem. 11.11 (Nov. 2019), pp. 1026–1033. issn: 1755-4349. doi: 10 . 1038 / s41557 - 019 - 0337-3. (Visited on 03/15/2025)

    2019

  5. [5]

    Low-Energy Spectrum of Iron–Sulfur Clusters Directly from Many-Particle Quantum Mechanics

    Sandeep Sharma, Kantharuban Sivalingam, Frank Neese, and Garnet Kin- Lic Chan. “Low-Energy Spectrum of Iron–Sulfur Clusters Directly from Many-Particle Quantum Mechanics”. In: Nat. Chem. 6.10 (Oct. 2014), pp. 927–933. issn: 1755-4349. doi: 10.1038/nchem.2041 . (Visited on 03/15/2025)

    doi:10.1038/nchem.2041 2014

  6. [6]

    Entangled Quan- tum Electronic Wavefunctions of the Mn4CaO5 Cluster in Photosystem II

    Yuki Kurashige, Garnet Kin-Lic Chan, and Takeshi Yanai. “Entangled Quan- tum Electronic Wavefunctions of the Mn4CaO5 Cluster in Photosystem II”. In: Nat. Chem. 5.8 (Aug. 2013), pp. 660–666. issn: 1755-4349. doi: 10.1038/nchem.1677. (Visited on 03/15/2025)

    doi:10.1038/nchem.1677 2013

  7. [7]

    Attila Szabo and Neil S. Ostlund. Modern Quantum Chemistry Introduc- tion to Advanced Electronic Structure Theory. Introduction to Advanced Electronic Structure Theory. Dover Publications, Incorporated, 2012. isbn: 9780486134598

    2012

  8. [8]

    Flexibilities of wavelets as a computational basis set for large-scale electronic structure calculations

    Laura E. Ratcliff et al. “Flexibilities of wavelets as a computational basis set for large-scale electronic structure calculations”. In: J. Chem. Phys. 152.19 (May 2020), p. 194110. issn: 0021-9606. doi: 10.1063/5.0004792

    work page doi:10.1063/5.0004792 2020

  9. [9]

    MADNESS: A Multiresolution, Adaptive Numer- ical Environment for Scientific Simulation

    Robert J. Harrison et al. “MADNESS: A Multiresolution, Adaptive Numer- ical Environment for Scientific Simulation”. In: SIAM J. Sci. Comput. 38.5 (2016), S123–S142. doi: 10.1137/15M1026171

    work page doi:10.1137/15m1026171 2016

  10. [10]

    Sliced Basis Density Matrix Renormalization Group for Electronic Structure

    E. Miles Stoudenmire and Steven R. White. “Sliced Basis Density Matrix Renormalization Group for Electronic Structure”. In:Phys. Rev. Lett.119 (4 July 2017), p. 046401. doi: 10.1103/PhysRevLett.119.046401

    work page doi:10.1103/physrevlett.119.046401 2017

  11. [11]

    From plane waves to local Gaussians for the simulation of corre- lated periodic systems

    George H. Booth, Theodoros Tsatsoulis, Garnet Kin-Lic Chan, and Andreas Gr¨ uneis. “From plane waves to local Gaussians for the simulation of corre- lated periodic systems”. In: J. Chem. Phys. 145.8 (Aug. 2016), p. 084111. issn: 0021-9606. doi: 10.1063/1.4961301

    work page doi:10.1063/1.4961301 2016

  12. [12]

    Gaussian basis sets for molecular applications

    J. Grant Hill. “Gaussian basis sets for molecular applications”. In: Int. J. Quantum Chem. 113.1 (2013), pp. 21–34. doi: 10.1002/qua.24355. 36 Yannic Rath, Massimo Bortone, George H. Booth

    doi:10.1002/qua.24355 2013

  13. [13]

    The updates in Libcint 6: More integrals, API refinements, and SIMD optimization techniques

    Qiming Sun. “The updates in Libcint 6: More integrals, API refinements, and SIMD optimization techniques”. In:J. Chem. Phys.160.17 (May 2024), p. 174116. issn: 0021-9606. doi: 10.1063/5.0200293

    work page doi:10.1063/5.0200293 2024

  14. [14]

    A priori identification of con- figurational deadwood

    Laimutis Bytautas and Klaus Ruedenberg. “A priori identification of con- figurational deadwood”. In: Chem. Phys. 356.1 (2009). Moving Frontiers in Quantum Chemistry: pp. 64–75. issn: 0301-0104. doi: 10 . 1016 / j . chemphys.2008.11.021 . url: https://www.sciencedirect.com/ science/article/pii/S0301010408005314

    2009

  15. [15]

    Ab-Initio Solution of the Many-Electron Schr ¨odinger Equation with Deep Neural Networks

    David Pfau, James S. Spencer, Alexander G. D. G. Matthews, and W. M. C. Foulkes. “Ab-Initio Solution of the Many-Electron Schr ¨odinger Equation with Deep Neural Networks”. In: Phys. Rev. Research 2.3 (Sept. 2020), p. 033429. doi: 10.1103/physrevresearch.2.033429

    doi:10.1103/physrevresearch.2.033429 2020

  16. [16]

    Deep-neural-network solution of the electronic Schr¨odinger equation

    Jan Hermann, Zeno Sch ¨atzle, and Frank No´e. “Deep-neural-network solution of the electronic Schr¨odinger equation”. In: Nat. Chem. 12.10 (Sept. 2020), pp. 891–897. doi: 10.1038/s41557-020-0544-y

    doi:10.1038/s41557-020-0544-y 2020

  17. [17]

    Variance extrapolation method for neural-network variational Monte Carlo

    Weizhong Fu, Weiluo Ren, and Ji Chen. “Variance extrapolation method for neural-network variational Monte Carlo”. In: Mach. Learn.: Sci. Technol. 5.1 (Jan. 2024), p. 015016. doi: 10.1088/2632-2153/ad1f75

    work page doi:10.1088/2632-2153/ad1f75 2024

  18. [18]

    Role of Backflow Correlations for the Nonmagnetic Phase of the T-t’ Hubbard Model

    Luca F. Tocchio, Federico Becca, Alberto Parola, and Sandro Sorella. “Role of Backflow Correlations for the Nonmagnetic Phase of the T-t’ Hubbard Model”. In: Phys. Rev. B 78.4 (July 2008), p. 041101. doi: 10 . 1103 / PhysRevB.78.041101. (Visited on 08/07/2023)

    2008

  19. [19]

    Backflow Correlations in the Hubbard Model: An Efficient Tool for the Study of the Metal-Insulator Transition and the Large-$U$ Limit

    Luca F. Tocchio, Federico Becca, and Claudius Gros. “Backflow Correlations in the Hubbard Model: An Efficient Tool for the Study of the Metal-Insulator Transition and the Large-$U$ Limit”. In: Phys. Rev. B 83.19 (May 2011), p. 195138. doi: 10.1103/PhysRevB.83.195138. (Visited on 10/19/2022)

    work page doi:10.1103/physrevb.83.195138 2011

  20. [20]

    Quantum Monte Carlo Approaches for Correlated Systems

    Federico Becca and Sandro Sorella. Quantum Monte Carlo Approaches for Correlated Systems. Cambridge University Press, Nov. 2017.doi: 10.1017/ 9781316417041

    2017

  21. [21]

    Explicitly Cor- related R12/F12 Methods for Electronic Structure

    Liguo Kong, Florian A. Bischoff, and Edward F. Valeev. “Explicitly Cor- related R12/F12 Methods for Electronic Structure”. In: Chem. Rev. 112.1 (2012), pp. 75–107. doi: 10.1021/cr200204r

    doi:10.1021/cr200204r 2012

  22. [22]

    Fermionic neural- network states for ab-initio electronic structure

    Kenny Choo, Antonio Mezzacapo, and Giuseppe Carleo. “Fermionic neural- network states for ab-initio electronic structure”. In: Nat. Commun. 11.1 (May 2020). doi: 10.1038/s41467-020-15724-9

    doi:10.1038/s41467-020-15724-9 2020

  23. [23]

    Matrix product operators, matrix product states, and ab initio density matrix renormalization group algorithms

    Garnet Kin-Lic Chan, Anna Keselman, Naoki Nakatani, Zhendong Li, and Steven R. White. “Matrix product operators, matrix product states, and ab initio density matrix renormalization group algorithms”. In: J. Chem. Phys. 145.1 (July 2016), p. 014102. issn: 0021-9606. doi: 10.1063/1.4955108

    doi:10.1063/1.4955108 2016

  24. [24]

    Unifying neural-network quantum states and correlator product states via tensor networks

    Stephen R Clark. “Unifying neural-network quantum states and correlator product states via tensor networks”. In:J. Phys. A: Math. Theor.51.13 (Feb. 2018), p. 135301. doi: 10.1088/1751-8121/aaaaf2

    work page doi:10.1088/1751-8121/aaaaf2 2018

  25. [25]

    Solving the quantum many-body problem with artificial neural networks

    Giuseppe Carleo and Matthias Troyer. “Solving the quantum many-body problem with artificial neural networks”. In: Science 355.6325 (Feb. 2017), pp. 602–606. doi: 10.1126/science.aag2302. Bayesian states for quantum chemistry 37

    doi:10.1126/science.aag2302 2017

  26. [26]

    Chapter 25 - Multiconfigurational quantum chemistry

    Bj ¨orn O. Roos. “Chapter 25 - Multiconfigurational quantum chemistry”. In: Theory and Applications of Computational Chemistry. Ed. by Clifford E. Dykstra, Gernot Frenking, Kwang S. Kim, and Gustavo E. Scuseria. Amster- dam: Elsevier, 2005, pp. 725–764.isbn: 978-0-444-51719-7.doi: 10.1016/ B978- 044451719- 7/50068- 8. url: https://www.sciencedirect. com/s...

    2005

  27. [27]

    Quantum Embedding Theories

    Qiming Sun and Garnet Kin-Lic Chan. “Quantum Embedding Theories”. In: Accounts Chem. Res.49.12 (2016), pp. 2705–2712. doi: 10.1021/acs. accounts.6b00356

    doi:10.1021/acs 2016

  28. [28]

    Quantum Embedding Theory for Strongly Correlated States in Materials

    He Ma, Nan Sheng, Marco Govoni, and Giulia Galli. “Quantum Embedding Theory for Strongly Correlated States in Materials”. In: J. Chem. Theory Comput. 17.4 (2021), pp. 2116–2125. doi: 10.1021/acs.jctc.0c01258

    doi:10.1021/acs.jctc.0c01258 2021

  29. [29]

    Systematic Improvability in Quan- tum Embedding for Real Materials

    Max Nusspickel and George H. Booth. “Systematic Improvability in Quan- tum Embedding for Real Materials”. In: Phys. Rev. X 12 (1 Mar. 2022), p. 011046. doi: 10.1103/PhysRevX.12.011046

    doi:10.1103/physrevx.12.011046 2022

  30. [30]

    Ab initio effective core potentials for molecular calculations. Potentials for K to Au including the outermost core orbitals

    P. Jeffrey Hay and Willard R. Wadt. “Ab initio effective core potentials for molecular calculations. Potentials for K to Au including the outermost core orbitals”. In:J. Chem. Phys.82.1 (Jan. 1985), pp. 299–310.issn: 0021-9606. doi: 10.1063/1.448975 . eprint: https://pubs.aip.org/aip/jcp/ article-pdf/82/1/299/18951300/299\_1\_online.pdf

    doi:10.1063/1.448975 1985

  31. [31]

    Canonical transcorrelated theory with projected Slater-type geminals

    Takeshi Yanai and Toru Shiozaki. “Canonical transcorrelated theory with projected Slater-type geminals”. In: J. Chem. Phys. 136.8 (Feb. 2012), p. 084107. issn: 0021-9606. doi: 10.1063/1.3688225

    work page doi:10.1063/1.3688225 2012

  32. [32]

    A first solution, for LiH, of a molecular transcorrelated wave equation by means of restricted numerical integration

    Samuel Francis Boys, Nicholas Charles Handy, and John Wilfrid Linnett. “A first solution, for LiH, of a molecular transcorrelated wave equation by means of restricted numerical integration”. In: Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences 311.1505 (1969), pp. 309–

    1969

  33. [33]

    Perspec- tive: Explicitly correlated electronic structure theory for complex systems

    Andreas Gr¨ uneis, So Hirata, Yu-ya Ohnishi, and Seiichiro Ten-no. “Perspec- tive: Explicitly correlated electronic structure theory for complex systems”. In: J. Chem. Phys. 146.8 (Feb. 2017), p. 080901. issn: 0021-9606. doi: 10.1063/1.4976974

    work page doi:10.1063/1.4976974 2017

  34. [34]

    Assessment of mul- tireference approaches to explicitly correlated full configuration interaction quantum Monte Carlo

    J. A. F. Kersten, George H. Booth, and Ali Alavi. “Assessment of mul- tireference approaches to explicitly correlated full configuration interaction quantum Monte Carlo”. In: J. Chem. Phys. 145.5 (Aug. 2016), p. 054117. issn: 0021-9606. doi: 10.1063/1.4959245

    work page doi:10.1063/1.4959245 2016

  35. [35]

    An ex- plicitly correlated approach to basis set incompleteness in full configuration interaction quantum Monte Carlo

    George H. Booth, Deidre Cleland, Ali Alavi, and David P. Tew. “An ex- plicitly correlated approach to basis set incompleteness in full configuration interaction quantum Monte Carlo”. In: J. Chem. Phys. 137.16 (Oct. 2012), p. 164112. issn: 0021-9606. doi: 10.1063/1.4762445

    work page doi:10.1063/1.4762445 2012

  36. [36]

    Reduced scaling Hilbert space vari- ational Monte Carlo

    Haochuan Wei and Eric Neuscamman. “Reduced scaling Hilbert space vari- ational Monte Carlo”. In: J. Chem. Phys. 149.18 (Nov. 2018), p. 184106. doi: 10.1063/1.5047207. 38 Yannic Rath, Massimo Bortone, George H. Booth

    work page doi:10.1063/1.5047207 2018

  37. [37]

    Bayesian Modelling Approaches for Quantum States - The Ultimate Gaussian Process States Handbook

    Yannic Rath. “Bayesian Modelling Approaches for Quantum States - The Ultimate Gaussian Process States Handbook”. PhD thesis. King’s College London, 2023

    2023

  38. [38]

    Many-Body Perturbation Theory and Coupled Cluster Theory for Electron Correlation in Molecules

    R J Bartlett. “Many-Body Perturbation Theory and Coupled Cluster Theory for Electron Correlation in Molecules”. In:Annu. Rev. Phys. Chem.32.1 (Oct. 1981), pp. 359–401. doi: 10.1146/annurev.pc.32.100181.002043

    arXiv 1981

  39. [39]

    State-specific multireference coupled-cluster the- ory

    Andreas K ¨ohn, Matthias Hanauer, Leonie Anna M¨ uck, Thomas-Christian Jagau, and J¨ urgen Gauss. “State-specific multireference coupled-cluster the- ory”. In: WIREs Comput. Mol. Sci. 3.2 (2013), pp. 176–197. doi: 10.1002/ wcms.1120

    2013

  40. [40]

    Approximating strongly correlated wave functions with correlator product states

    Hitesh J. Changlani, Jesse M. Kinder, C. J. Umrigar, and Garnet Kin-Lic Chan. “Approximating strongly correlated wave functions with correlator product states”. In: Phys. Rev. B 80.24 (Dec. 2009), p. 245116. doi: 10. 1103/physrevb.80.245116

    2009

  41. [41]

    Ground-state prop- erties of quantum many-body systems: entangled-plaquette states and vari- ational Monte Carlo

    F Mezzacapo, N Schuch, M Boninsegni, and J I Cirac. “Ground-state prop- erties of quantum many-body systems: entangled-plaquette states and vari- ational Monte Carlo”. In: New J. Phys. 11.8 (Aug. 2009), p. 083026. doi: 10.1088/1367-2630/11/8/083026

    work page doi:10.1088/1367-2630/11/8/083026 2009

  42. [42]

    Gaussian processes for machine learning

    Carl Edward Rasmussen. Gaussian processes for machine learning . MIT Press, 2006. isbn: 026218253X. doi: 10.7551/mitpress/3206.001. 0001

    doi:10.7551/mitpress/3206.001 2006

  43. [43]

    Gaussian Process States: A Data-Driven Representation of Quan- tum Many-Body Physics

    Aldo Glielmo, Yannic Rath, G ´abor Cs´anyi, Alessandro De Vita, and George H. Booth. “Gaussian Process States: A Data-Driven Representation of Quan- tum Many-Body Physics”. In:Phys. Rev. X10.4 (Nov. 2020), p. 041026.doi: 10.1103/physrevx.10.041026

    work page doi:10.1103/physrevx.10.041026 2020

  44. [44]

    A Bayesian inference framework for compression and prediction of quantum states

    Yannic Rath, Aldo Glielmo, and George H. Booth. “A Bayesian inference framework for compression and prediction of quantum states”. In: J. Chem. Phys. 153.12 (Sept. 2020), p. 124108. doi: 10.1063/5.0024570

    work page doi:10.1063/5.0024570 2020

  45. [45]

    Learning ground states of gapped quantum Hamiltonians with Kernel Methods

    Clemens Giuliani, Filippo Vicentini, Riccardo Rossi, and Giuseppe Car- leo. “Learning ground states of gapped quantum Hamiltonians with Kernel Methods”. In: Quantum 7 (Aug. 2023), p. 1096. issn: 2521-327X. doi: 10.22331/q-2023-08-29-1096

    work page doi:10.22331/q-2023-08-29-1096 2023

  46. [46]

    Many-Body Problem with Strong Forces

    Robert Jastrow. “Many-Body Problem with Strong Forces”. In: Phys. Rev. 98.5 (June 1955), pp. 1479–1484. doi: 10.1103/physrev.98.1479

    doi:10.1103/physrev.98.1479 1955

  47. [47]

    Error Detecting and Error Correcting Codes

    R. W. Hamming. “Error Detecting and Error Correcting Codes”. In: Bell Syst. Tech. J. 29.2 (Apr. 1950), pp. 147–160. doi: 10 . 1002 / j . 1538 - 7305.1950.tb00463.x

    arXiv 1950

  48. [48]

    Additive Gaus- sian Processes

    David K Duvenaud, Hannes Nickisch, and Carl Rasmussen. “Additive Gaus- sian Processes”. In: Advances in Neural Information Processing Systems . Ed. by J. Shawe-Taylor, R. Zemel, P. Bartlett, F. Pereira, and K.Q. Wein- berger. Vol. 24. Curran Associates, Inc., 2011, pp. 226–234. url: https: / / proceedings . neurips . cc / paper _ files / paper / 2011 / fi...

    2011

  49. [49]

    Quantum Gaussian process state: A kernel-inspired state with quantum support data

    Yannic Rath and George H. Booth. “Quantum Gaussian process state: A kernel-inspired state with quantum support data”. In: Phys. Rev. Research 4.2 (May 2022), p. 023126. doi: 10.1103/physrevresearch.4.023126

    work page doi:10.1103/physrevresearch.4.023126 2022

  50. [50]

    Bayesian Inference: An Introduction to Principles and Practice in Machine Learning

    Michael E. Tipping. “Bayesian Inference: An Introduction to Principles and Practice in Machine Learning”. In:Advanced Lectures on Machine Learning. Springer Berlin Heidelberg, 2004, pp. 41–62. doi: 10.1007/978-3-540- 28650-9_3

    doi:10.1007/978-3-540- 2004

  51. [51]

    The Relevance Vector Machine

    Michael E. Tipping. “The Relevance Vector Machine”. In: Advances in Neural Information Processing Systems. 2000, pp. 652–658

    2000

  52. [52]

    Fast Marginal Likelihood Maximi- sation for Sparse Bayesian Models

    Michael E Tipping, Anita C Faul, et al. “Fast Marginal Likelihood Maximi- sation for Sparse Bayesian Models.” In: AISTATS. 2003

    2003

  53. [53]

    Neural-network quantum state tomography

    Giacomo Torlai et al. “Neural-network quantum state tomography”. In: Nat. Phys. 14.5 (Feb. 2018), pp. 447–450.doi: 10.1038/s41567-018-0048-5

    doi:10.1038/s41567-018-0048-5 2018

  54. [54]

    Experimental Online Quantum Dots Charge Autotuning Using Neural Networks

    Victor Yon et al. “Experimental Online Quantum Dots Charge Autotuning Using Neural Networks”. In: Nano Lett. 25.10 (2025), pp. 3717–3725. doi: 10.1021/acs.nanolett.4c04889

    doi:10.1021/acs.nanolett.4c04889 2025

  55. [55]

    Neural-network quantum state tomography in a two-qubit experiment

    Marcel Neugebauer et al. “Neural-network quantum state tomography in a two-qubit experiment”. In: Phys. Rev. A 102.4 (Oct. 2020). doi: 10.1103/ physreva.102.042604

    2020

  56. [56]

    The density-matrix renormalization group in the age of matrix product states

    Ulrich Schollw ¨ock. “The density-matrix renormalization group in the age of matrix product states”. In: Ann. Phys. 326.1 (Jan. 2011), pp. 96–192. doi: 10.1016/j.aop.2010.09.012

    doi:10.1016/j.aop.2010.09.012 2011

  57. [57]

    Time-evolution methods for matrix-product states

    Sebastian Paeckel et al. “Time-evolution methods for matrix-product states”. In: Ann. Phys. 411 (Dec. 2019), p. 167998. doi: 10.1016/j.aop.2019. 167998

    doi:10.1016/j.aop.2019 2019

  58. [58]

    Recent developments in CANDECOMP/PARAFAC algorithms: a critical review

    Nicolaas (Klaas) M. Faber, Rasmus Bro, and Philip K. Hopke. “Recent developments in CANDECOMP/PARAFAC algorithms: a critical review”. In: Chemom. Intell. Lab. Syst.65.1 (Jan. 2003), pp. 119–137.doi: 10.1016/ s0169-7439(02)00089-8

    2003

  59. [59]

    Tensor Decompositions and Appli- cations

    Tamara G. Kolda and Brett W. Bader. “Tensor Decompositions and Appli- cations”. In: SIAM Rev. 51.3 (Aug. 2009), pp. 455–500. doi: 10 . 1137 / 07070111x

    2009

  60. [60]

    CP decom- position for tensors via alternating least squares with QR decomposition

    Rachel Minster, Irina Viviano, Xiaotian Liu, and Grey Ballard. “CP decom- position for tensors via alternating least squares with QR decomposition”. In: Numer. Linear Algebra Appl.30.6 (2023), e2511. doi: 10.1002/nla.2511

    work page doi:10.1002/nla.2511 2023

  61. [61]

    Generalization properties of neural net- work approximations to frustrated magnet ground states

    Tom Westerhout, Nikita Astrakhantsev, Konstantin S. Tikhonov, Mikhail I. Katsnelson, and Andrey A. Bagrov. “Generalization properties of neural net- work approximations to frustrated magnet ground states”. In:Nat. Commun. 11.1 (Mar. 2020). doi: 10.1038/s41467-020-15402-w

    doi:10.1038/s41467-020-15402-w 2020

  62. [62]

    Dirac-Type Nodal Spin Liquid Re- vealed by Refined Quantum Many-Body Solver Using Neural-Network Wave Function, Correlation Ratio, and Level Spectroscopy

    Yusuke Nomura and Masatoshi Imada. “Dirac-Type Nodal Spin Liquid Re- vealed by Refined Quantum Many-Body Solver Using Neural-Network Wave Function, Correlation Ratio, and Level Spectroscopy”. In:Phys. Rev. X 11.3 (Aug. 2021), p. 031034. doi: 10.1103/physrevx.11.031034

    doi:10.1103/physrevx.11.031034 2021

  63. [63]

    Adam: A Method for Stochastic Op- timization

    Diederik P. Kingma and Jimmy Ba. “Adam: A Method for Stochastic Op- timization”. In: 3rd International Conference on Learning Representations, 40 Yannic Rath, Massimo Bortone, George H. Booth ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Pro- ceedings. Ed. by Yoshua Bengio and Yann LeCun. 2015. doi: 10.48550/ arxiv.1412.6980. arXiv: 1412.6980

    Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2015

  64. [64]

    Neural-network quantum state tomography

    Dominik Koutn´ y, Libor Motka, Zden ˇek Hradil, Jaroslav ˇReh´aˇcek, and Luis L. S ´anchez-Soto. “Neural-network quantum state tomography”. In: Phys. Rev. A 106.1 (July 2022). doi: 10.1103/physreva.106.012409

    doi:10.1103/physreva.106.012409 2022

  65. [65]

    Efficient quantum state tomography with convolutional neural networks

    Tobias Schmale, Moritz Reh, and Martin G ¨arttner. “Efficient quantum state tomography with convolutional neural networks”. In: npj Quantum Inf. 8.1 (Sept. 2022). doi: 10.1038/s41534-022-00621-4

    doi:10.1038/s41534-022-00621-4 2022

  66. [66]

    Attention-Based Transformer Networks for Quantum State Tomogra- phy

    Hailan Ma, Zhenhong Sun, Daoyi Dong, Chunlin Chen, and Herschel Ra- bitz. Attention-Based Transformer Networks for Quantum State Tomogra- phy. 2023. doi: 10 . 48550 / arxiv . 2305 . 05433. arXiv: 2305 . 05433 [quant-ph]

    2023

  67. [67]

    Towards a standardized notation and terminology in multiway analysis

    Henk A. L. Kiers. “Towards a standardized notation and terminology in multiway analysis”. In: J. Chemom. 14.3 (2000), pp. 105–122. doi: 10 . 1002/1099-128x(200005/06)14:3<105::aid-cem582>3.0.co;2-i

    2000

  68. [68]

    Impact of condi- tional modelling for a universal autoregressive quantum state

    Massimo Bortone, Yannic Rath, and George H. Booth. “Impact of condi- tional modelling for a universal autoregressive quantum state”. In:Quantum 8 (Feb. 2024), p. 1245.issn: 2521-327X. doi: 10.22331/q-2024-02-08- 1245

    work page doi:10.22331/q-2024-02-08- 2024

  69. [69]

    Helping restricted Boltzmann machines with quantum- state representation by restoring symmetry

    Yusuke Nomura. “Helping restricted Boltzmann machines with quantum- state representation by restoring symmetry”. In: J. Phys.: Condens. Matter 33.17 (Apr. 2021), p. 174003. doi: 10.1088/1361-648x/abe268

    work page doi:10.1088/1361-648x/abe268 2021

  70. [70]

    Deep Autoregressive Models for the Efficient Variational Simulation of Many-Body Quantum Systems

    Or Sharir, Yoav Levine, Noam Wies, Giuseppe Carleo, and Amnon Shashua. “Deep Autoregressive Models for the Efficient Variational Simulation of Many-Body Quantum Systems”. In: Phys. Rev. Lett. 124.2 (Jan. 2020), p. 020503. doi: 10.1103/physrevlett.124.020503

    doi:10.1103/physrevlett.124.020503 2020

  71. [71]

    Recurrent neural network wave functions

    Mohamed Hibat-Allah, Martin Ganahl, Lauren E. Hayward, Roger G. Melko, and Juan Carrasquilla. “Recurrent neural network wave functions”. In:Phys. Rev. Research 2.2 (June 2020). doi: 10 . 1103 / physrevresearch . 2 . 023358

    2020

  72. [72]

    Convolutional trans- former wave functions

    Ao Chen, Vighnesh Dattatraya Naik, and Markus Heyl. Convolutional trans- former wave functions. 2025. doi: 10.48550/arxiv.2503.10462. arXiv: 2503.10462 [cond-mat.dis-nn]

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.10462 2025

  73. [73]

    Pfaf- fian Pairing Wave Functions in Electronic-Structure Quantum Monte Carlo Simulations

    M. Bajdich, L. Mitas, G. Drobn´ y, L. K. Wagner, and K. E. Schmidt. “Pfaf- fian Pairing Wave Functions in Electronic-Structure Quantum Monte Carlo Simulations”. In: Phys. Rev. Lett. 96.13 (Apr. 2006), p. 130201. doi: 10. 1103/physrevlett.96.130201

    2006

  74. [74]

    Framework for efficient ab initio elec- tronic structure with Gaussian Process States

    Yannic Rath and George H. Booth. “Framework for efficient ab initio elec- tronic structure with Gaussian Process States”. In: Phys. Rev. B 107 (May 2023), p. 205119. doi: 10.1103/PhysRevB.107.205119

    doi:10.1103/physrevb.107.205119 2023

  75. [75]

    Simple Fermionic backflow states via a systematically improvable tensor decomposition

    Massimo Bortone, Yannic Rath, and George H Booth. “Simple Fermionic backflow states via a systematically improvable tensor decomposition”. In: Commun. Phys. 8.1 (2025), p. 169. Bayesian states for quantum chemistry 41

    2025

  76. [76]

    Gold-standard solutions to the Schr ¨odinger equation using deep learning: How much physics do we need?

    Leon Gerard, Michael Scherbela, Philipp Marquetand, and Philipp Grohs. “Gold-standard solutions to the Schr ¨odinger equation using deep learning: How much physics do we need?” In: Advances in Neural Information Pro- cessing Systems. Ed. by Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho. 2022. url: https://openreview.net/forum?id=nX- gReQ0OT

    2022

  77. [77]

    Solving many-electron Schr¨odinger equation using deep neural networks

    Jiequn Han, Linfeng Zhang, and Weinan E. “Solving many-electron Schr¨odinger equation using deep neural networks”. In:J. Comput. Phys.399 (Dec. 2019), p. 108929. doi: 10.1016/j.jcp.2019.108929

    arXiv 2019

  78. [78]

    Ab initio quantum chemistry with neural-network wave- functions

    Jan Hermann et al. “Ab initio quantum chemistry with neural-network wave- functions”. In: Nat. Rev. Chem. 7.10 (Aug. 2023), pp. 692–709. issn: 2397-

    2023

  79. [79]

    Ab initio calculation of real solids via neural network ansatz

    Xiang Li, Zhe Li, and Ji Chen. “Ab initio calculation of real solids via neural network ansatz”. In:Nat. Commun. 13.1 (Dec. 2022). doi: 10.1038/ s41467-022-35627-1

    2022

  80. [80]

    Backflow Transformations via Neural Networks for Quantum Many-Body Wave Functions

    Di Luo and Bryan K. Clark. “Backflow Transformations via Neural Networks for Quantum Many-Body Wave Functions”. In:Phys. Rev. Lett.122.22 (June 2019), p. 226401. doi: 10.1103/physrevlett.122.226401

    doi:10.1103/physrevlett.122.226401 2019

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.