Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Quantum embedding's bottleneck can be removed by learning a low-dimensional linear variational space for the embedding ground states, then solving a small K×K eigenvalue problem at each step; the same subspace transfers across lattices and

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

2026-08-03 14:02 UTC pith:7MZHITZT

load-bearing objection A clean Rayleigh-Ritz reduction for quantum embedding solves; the Pu cross-phase claim is real but is interpolation inside a hand-picked box, and speedups ignore training cost. the 3 major comments →

arxiv 2512.21666 v1 pith:7MZHITZT submitted 2025-12-25 cond-mat.str-el

Linear Foundation Model for Quantum Embedding: Data-Driven Compression of the Ghost Gutzwiller Variational Space

Samuele Giuli , Hasanat Hasan , Benedikt Kloss , Marius S. Frank , Tsung-Han Lee , Olivier Gingras , Yong-Xin Yao , Nicola Lanat\`a This is my paper
classification cond-mat.str-el
keywords quantum embeddingghost Gutzwiller approximationprincipal component analysisvariational subspaceactive learningstrongly correlated electronsplutoniumHubbard model
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 tries to show that the hardest step in quantum embedding—solving the large embedding Hamiltonian at every self-consistency iteration—can be replaced by a cheap, deterministic eigenvalue problem in a learned linear subspace. Its central hypothesis is that the ground states visited during a calculation lie on a low-dimensional manifold inside the exponentially large Hilbert space, and that principal component analysis on a modest set of exact ground states builds a subspace that covers this manifold. Once the subspace is fixed, every embedding solve reduces to diagonalizing a K×K matrix whose operator elements are precomputed, and physical constraints such as the Pauli principle are enforced by construction because the solution remains a variational ansatz. The benchmarks show that a Bethe-trained subspace works for square and cubic lattices without retraining, and that a single subspace trained on one parameter box reproduces the energy–volume curves of all six plutonium phases within a few meV per atom while cutting embedding time by roughly 4000×. If correct, this removes the main computational bottleneck of embedding methods and points toward correlated-materials simulations at near-DFT cost.

Core claim

The core claim is that the many-body ground states |Φ(X)> of the finite-bath embedding Hamiltonian, as the Hamiltonian parameters X are changed through a self-consistent loop, are not scattered across the full Fock space but concentrate near a low-dimensional linear subspace. The paper constructs that subspace by singular value decomposition of a data matrix of sampled ground states; the leading K singular vectors form a variational basis. Projecting the embedding Hamiltonian onto this basis gives a K×K matrix whose lowest eigenvalue is the variational energy, and its eigenvector yields all required density-matrix outputs through Hellmann-Feynman gradients. Because this is a variational ansa

What carries the argument

The central object is the learned linear variational space V^K: a K-dimensional subspace spanned by the leading principal components of a data matrix whose columns are exact ground states of the embedding Hamiltonian at sampled parameter points. The work it does is to convert each embedding solve into an affine K×K eigenvalue problem: since the Hamiltonian is linear in its parameters X, its matrix elements in this basis can be precomputed once, and the ground-state energy is the lowest eigenvalue of H0 + Σ X_r H_r. Two supporting mechanisms are the active-learning scheme, which creates local charts only when self-consistency visits uncovered parameters, and the method of snapshots, which bui

Load-bearing premise

The load-bearing assumption is that the embedding ground states visited during self-consistent calculations cluster near a low-dimensional linear subspace, and that a subspace built from a finite training set sampled inside a chosen parameter box covers every parameter point the calculation actually visits.

What would settle it

Take a target system whose embedding parameters lie outside the trained box—for example a d-electron or f-electron material at a stronger Hund's coupling or a wider hybridization range than used in training—and compute the squared overlap of the exact ground state with the learned subspace V^K; if that overlap falls below a threshold such as 1−10^-6, the compressed solver will fail. Alternatively, train on the same five-dimensional box but with bath size B=3 for the f-shell and inspect the singular-value spectrum: if σ_K/σ_1 fails to fall below the chosen tolerance before K approaches the samp

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • The embedding-Hamiltonian step, the dominant cost in ghost-GA and related quantum-embedding workflows, becomes a small deterministic diagonalization; the paper reports roughly 30× to 4000× reductions in embedding time.
  • A variational space trained on one lattice (Bethe) transfers without retraining to square and cubic lattices, so a single solver library can serve many lattice geometries.
  • For plutonium, a single subspace trained in one 5D parameter box reproduces the energetics of all six phases within a few meV per atom, meaning the local 5f environment is effectively shared across phases.
  • Because the ansatz is variational, outputs obey N-representability (occupations between 0 and 1) by construction, avoiding the unphysical predictions that can plague direct machine-learning of energies.
  • The method-of-snapshots reformulation lets the subspace be trained from overlaps supplied by MPS, neural quantum state, or quantum-assisted solvers, extending the compression to bath sizes where exact diagonalization is impossible.

Where Pith is reading between the lines

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

  • A testable extension is whether the steep singular-value decay persists when the bath size grows to B≥3 for d or f shells with cluster impurities; if it does, the method-of-snapshots route becomes the practical path, but if the manifold dimension grows rapidly with bath size, the compression factor will shrink accordingly.
  • The active-learning scheme has a built-in stress test: count charts as a function of system complexity. The paper finds 13 charts for the scanned Hubbard phase diagram and one box for plutonium; a material or model where chart count grows rapidly with volume or doping would indicate that the physically relevant parameters do not stay on a sparse manifold.
  • One could use the same PCA subspace as a preconditioner or initializer for full solvers rather than a replacement: the K-dimensional solution provides a good starting state, with exact diagonalization used only for error control at selected iterations.
  • The foundation-model framing suggests that a subspace trained for one impurity shell could be reused across many materials sharing that shell, since the embedding Hamiltonian structure is universal; the plutonium phase result and the Bethe-to-cubic transfer are two instances of this reuse.

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 proposes a data-driven compression of the embedding Hamiltonian (EH) ground-state manifold in the ghost-Gutzwiller approximation. The authors collect exact-diagonalization ground states for sampled EH parameters, perform PCA, and use the leading K singular vectors as a linear variational basis; each EH solve is then replaced by a K×K projected eigenvalue problem. An active-learning scheme builds local charts only where needed. Benchmarks are reported for (i) a three-orbital Hubbard–Kanamori model, where a solver trained on the Bethe lattice transfers to square and cubic lattices with energy errors ~1e-4 W and speedups of one to two orders of magnitude, and (ii) plutonium, where a single K=250 subspace trained on a 5D parameter box reproduces the E(V) curves of six phases to a few meV/atom while reducing EH time by ~4000x. A 'method of snapshots' reformulation is proposed to interface with MPS/NQS solvers.

Significance. If the central claim holds, the method addresses a major computational bottleneck of quantum embedding: replacing a sequence of expensive many-body EH solves by deterministic low-dimensional diagonalizations, while preserving N-representability by construction. The PCA construction is a clean application of standard reduced-basis ideas, and the benchmarks are direct comparisons against ED. The Hubbard transferability result and the Pu multi-phase result are both impressive and would be of immediate interest to the quantum-embedding community. The main caveats are that the method relies on an empirical low-manifold hypothesis, the reported speedups exclude training costs, and the Pu transfer claim would benefit from quantitative coverage information relative to the training box. The manuscript is well written and the conceptual contribution is clear; the limitations are mostly about missing validation/reporting rather than a fundamental flaw.

major comments (3)
  1. [Sec. V D] The central Pu claim rests on a training box constructed from δ-Pu self-consistent trajectories ('range of X is chosen to be about twice the parameter value range in self-consistent LDA+ghost-GA calculation for δ-Pu'), but the paper does not report whether the X values encountered by the α, β, γ, δ′, and ε calculations remain inside the box, nor the PCA projection residual at those points. Since E_PCA is a variational upper bound, any trajectory that leaves the well-sampled region would silently give an uncontrolled error. The benchmark agreement is encouraging, but to substantiate the 'single variational space for all phases' claim, please provide per-phase ranges of X relative to the training box and the worst-case projection residuals or energy errors.
  2. [Secs. V C and V D] The reported speedups exclude training overhead. The active-learning Hubbard benchmark uses 13 charts, each requiring N_budget=600 ED solves (Sec. V B 1), i.e. roughly 7800 full ED diagonalizations, and the Pu benchmark requires 500 ED solves in a 2.8M-dimensional Fock space. The ~30x and ~4000x factors apply only to the inference phase. For a 'near-DFT cost' / high-throughput claim, please report the one-time training wall time, the number of EH solves amortized over, and the break-even number of solves against always-ED. Without this, the total-cost improvement is not quantified.
  3. [Eqs. (13)–(14) and Sec. V B 1] The optimization uses a uniform p(X) within each chart, and the truncation and chart-size hyperparameters (L_D, L_Λ, N_budget, ε_tol, K/N_budget<0.8) are fixed without sensitivity analysis. Since the method's central assumption is that a finite training set covers the physically relevant ground-state manifold, the dependence of K and of the resulting energy error on these choices should be characterized at least for one benchmark. This would clarify whether the favorable results are robust or rely on a specific tuning of the atlas construction.
minor comments (5)
  1. [Sec. V B 2] The statement that the atlas construction 'saturates after ~10^4 random trial centers' is not accompanied by a precise stopping rule. Please define the saturation criterion used.
  2. [Figs. 4 and 6] The insets show ΔE = E_ED − E_PCA, but the sign convention and whether absolute values are plotted should be stated explicitly. Since the projected solver is variational, one expects E_ED − E_PCA ≤ 0; the plots should be consistent with this.
  3. [Sec. V D] The training box for Pu is described in the text but the observed self-consistent parameter ranges for each phase are not tabulated. A supplementary table of the box bounds and the per-phase visited ranges would make the interpolation/extrapolation status transparent.
  4. [Sec. VI] The method-of-snapshots section is presented as a general integration route, but no numerical demonstration is given. It would help to mark this explicitly as a proposed extension or to include a small proof-of-concept.
  5. [General] There is no data/code availability statement. For a data-driven method, a reproducibility statement (even a minimal one) would be valuable.

Circularity Check

0 steps flagged

No circular step: the compressed solver is a genuine reduced-basis projection trained on ground states and benchmarked against independent ED at untrained parameters.

full rationale

The derivation chain is: define the embedding Hamiltonian H(X) (Eq. 3), sample ground states |Phi(X_s)> by ED (Eq. 14), build the PCA subspace V^K from those states (Eqs. 15-16), project H onto V^K (Eqs. 8-9), and evaluate E^K(X)=lambda_min(H^K(X)) (Eq. 10). The benchmark quantities (total energies, quasiparticle weights, E(V) curves) are then compared with full ED within self-consistent ghost-GA loops. Since the PCA target is the ground-state vector, not the energy or density matrix, and the tested X are distinct from the training samples (charts cover neighborhoods; active learning creates charts only from ED calls on uncovered X and then tests subsequent points), the agreement is a variational/interpolation check rather than a fitted reproduction. At a training point the equality E^K(X_s)=E(X_s) holds by construction because |Phi(X_s)> is in the subspace, but the paper's accuracy claims are not based on those points. Self-citations (notably Ref. [59]) motivate the low-dimensional-manifold hypothesis and a gauge convention, but the hypothesis is independently supported by the singular-value spectra (Fig. 3, Fig. 6(c)) and by the ED benchmarks, so the self-citations are not load-bearing. The Pu training box is chosen from delta-Pu self-consistent parameter ranges, so the six-phase result is interpolation inside a box and the paper does not report a coverage check for all phases; this is a generalization/correctness risk, not circularity.

Axiom & Free-Parameter Ledger

6 free parameters · 7 axioms · 0 invented entities

The central method adds no new physical constants or entities. Its free parameters are algorithmic hyperparameters (chart sizes, training budget, singular-value thresholds, training-box ranges) plus the chosen Pu training box. The main domain axiom is the empirical low-dimensional manifold hypothesis, validated only in the tested systems.

free parameters (6)
  • Chart linear sizes L_D=0.4, L_Λ=4.0 = L_D=0.4, L_Λ=4.0
    Hyperparameters defining chart hyperrectangles in X=(D1,D2,D3,Λ1,Λ2,Λ3) space; chosen by hand, not from data (Sec. V B 1).
  • Training budget N_budget per chart = 600
    Number of uniformly sampled EH ground states per chart for SVD; fixed by hand (Sec. V B 1).
  • Singular-value truncation threshold ε_tol = 1e-6
    Dimension K selected as first k with σ_k/σ_1 < ε_tol (Eq. 28). Controls accuracy vs cost; chosen by hand.
  • Buffer condition K/N_budget < 0.8 = 0.8
    Chart-splitting criterion to avoid undersampling; chosen by hand (Sec. V B 1).
  • Pu variational dimension K and singular threshold = K=250, σ_a ≥ 1e-4
    For plutonium, K=250 principal components retained with threshold 1e-4 (Sec. V D, Fig. 6c); chosen from the singular-value spectrum.
  • Pu training-box ranges = (E5/2-E7/2)/U ∈ [0.25,0.29], v/U ∈ [-1,0], ε5/2/U ∈ [3.38,6.10], ε7/2/U ∈ [2.92,5.48]
    Hyper-rectangular training domain chosen as ~2x the parameter range from δ-Pu LDA+ghost-GA at min/max volume; defines the interpolation domain (Sec. V D).
axioms (7)
  • standard math SVD/PCA provides the optimal rank-K linear subspace in mean squared projection error (Eckart-Young theorem).
    Used to justify using leading left singular vectors as variational basis (Eqs. 13-16).
  • domain assumption Ground states |Φ(X)> of the finite-bath EH lie close to a low-dimensional linear manifold.
    Central hypothesis: 'We hypothesize that the physically relevant ground states of the EH reside on a low-dimensional manifold' (Sec. IV). Supported empirically by singular-value decay, but if false the compressed basis cannot represent target ground states.
  • domain assumption The self-consistent EH trajectories of all tested lattices/phases stay within or near the trained charts/box.
    Transferability from Bethe to square/cubic and from one Pu box to all six phases depends on coverage of X values visited in the ghost-GA loop (Secs. V C-V D).
  • ad hoc to paper Uniform sampling of X inside a chart induces a suitable p(X) for the expected projection error in Eq. (13).
    The distribution p is arbitrary; uniform over the hyperrectangle is chosen for convenience. Different p would change the optimal subspace.
  • standard math Variational principle: diagonalizing the projected Hamiltonian in V^K gives an upper bound to the exact ground-state energy and N-representable density matrices.
    Used to claim Pauli constraints and N-representability are satisfied by construction (Sec. IV A).
  • domain assumption Ghost-GA with finite bath B is a valid embedding approximation; B=3 claims DMFT-level accuracy.
    The method is embedded in ghost-GA; accuracy relative to DMFT comes from prior work (Refs. [44,45,72-74]), not from this paper.
  • domain assumption For Pu, the CFS-averaged approximation treats the f shell as isotropic with SOC-split J=5/2 and J=7/2 manifolds.
    This defines the Pu EH structure in Eq. (32) and the training box; it is inherited from prior ghost-GA/GA treatment (Ref. [36]).

pith-pipeline@v1.3.0-alltime-deepseek · 22289 in / 16932 out tokens · 174386 ms · 2026-08-03T14:02:14.968025+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Linear Foundation Model for Quantum Embedding: Data-Driven Compression of the Ghost Gutzwiller Variational Space." pith.science (2026). https://pith.science/paper/7MZHITZT

@misc{pith2026251221666,
  author       = {Pith},
  title        = {Pith review of: Linear Foundation Model for Quantum Embedding: Data-Driven Compression of the Ghost Gutzwiller Variational Space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7MZHITZT}},
  note         = {Machine review of arXiv:2512.21666}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Simulations of quantum matter rely mainly on Kohn-Sham density functional theory (DFT), which often fails for strongly correlated systems. Quantum embedding (QE) theories address this limitation by mapping the system onto an auxiliary embedding Hamiltonian (EH) describing fragment-environment interactions, but the EH is typically large and its iterative solution is the primary computational bottleneck. We introduce a linear foundation model for QE that utilizes principal component analysis (PCA) to compress the space of quantum states needed to solve the EH within a small variational subspace. Using a data-driven active-learning scheme, we learn this subspace from EH ground states and reduce each embedding solve to a deterministic ground-state eigenvalue problem in the reduced space. Within the ghost Gutzwiller approximation (ghost-GA), we show for a three-orbital Hubbard model that a variational space learned on a Bethe lattice is transferable to square and cubic lattices without additional training, while substantially reducing the cost of the EH step. We further validate the approach on plutonium, where a single variational space reproduces the energetics of all six crystalline phases while reducing the cost of the EH solution by orders of magnitude. This provides a practical route to overcome the main computational bottleneck of QE frameworks, paving the way for high-throughput ab initio simulations of strongly correlated materials at a near-DFT cost.

Figures

Figures reproduced from arXiv: 2512.21666 by Benedikt Kloss, Hasanat Hasan, Marius S. Frank, Nicola Lanat\`a, Olivier Gingras, Samuele Giuli, Tsung-Han Lee, Yong-Xin Yao.

Figure 1
Figure 1. Figure 1: Geometric representation of the ground-state mani [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Pictorial representation of how different lattice [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of the number of principal components [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ghost-GA results for the three degenerate orbitals Hubbard-Kanamori model with [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ghost-GA results for the three degenerate-orbital Hubbard–Kanamori model with [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (a) Energy–volume curves for the α, β, δ, δ ′ , ε, and γ phases of Pu. Filled markers: LDA+ghost-GA with exact ED solver. Open markers: LDA+ghost-GA with the same PCA-compressed solver trained once on a fixed EH-parameter box and reused for all phases and volumes. (b) The energy difference between the calculations with ED and PCA solvers, ∆E = EED − EPCA, for each of the six phases. (c) Singular values σa … view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Magnetic phases of the anisotropic triangular Hubbard model from the ghost-Gutzwiller approximation in the rotating spin-frame

    cond-mat.str-el 2026-04 unverdicted novelty 7.0

    Ghost-GA provides an efficient way to compute the magnetic phase diagram of the anisotropic triangular Hubbard model, matching DMFT closely and ruling out the 1D AF phase.

Reference graph

Works this paper leans on

106 extracted references · 5 linked inside Pith · cited by 1 Pith paper

  1. [1]

    In ghost-GA, it is sufficient to solve an EH with a relatively small bath (B= 3) to obtain the total energy of the lattice model with an accuracy com- parable with DMFT, which in principle requires a bath with infinite degrees of freedom

  2. [2]

    sign problem

    While in DMFT, the outputYis the local Green’s function of the fragmenti, the ghost-GA requires only the iterative calculation of the entries of the ground-state single-particle density matrix. Because of these reasons, each ghost-GA iteration can be tackled with a wealth of efficient state-of-the-art meth- ods that do not suffer from the so-called contin...

  3. [3]

    For example, the occupation numbers are guaranteed to lie within physical bounds, 0≤ ⟨ΦK(X)|b† aba|ΦK(X)⟩ ≤1, strictly enforcing the Pauli principle

    Since this representation is based on the varia- tional principle, the resulting density matrices sat- isfy allN-representability conditions by construc- tion [86]. For example, the occupation numbers are guaranteed to lie within physical bounds, 0≤ ⟨ΦK(X)|b† aba|ΦK(X)⟩ ≤1, strictly enforcing the Pauli principle

  4. [4]

    Partitioning the parameter space in multiple charts reduces the complexity of the learning problem for each individual chart, and the cost of evaluating ob- servables decreases as a function of the variational- space truncationK

  5. [5]

    The partitioning strategy also provides us with the option of focusing only on training our model within physically-realizable regions of the parame- ters space, which, as we are going to show, is con- siderably more efficient and scalable. It is important to point out that the possibility of cal- culating an efficient linear variational approximation to ...

  6. [6]

    Lattice models To illustrate our approach, we first focus on solving model systems with 3 degenerate orbitals and with a lo- cal electron-electron Kanamori interaction [92, 93] and only nearest-neighbors hoppings. The Hamiltonian will generically read: ˆH= ˆH 0 + X R ˆH int R (U, J),(17) ˆH 0 =−t X ⟨RR′⟩ X ασ c† RασcR′ασ −µ X Rασ c† RασcRασ ,(18) wherec †...

  7. [7]

    Different lattices with the same point-group symmetry will share the structure of the embedding Hamiltonian as shown in Fig

    Structure of the EH in the case of three degenerate orbitals The global symmetries of the lattice Hamiltonian will be reflected as symmetries of the EH [34, 44]. Different lattices with the same point-group symmetry will share the structure of the embedding Hamiltonian as shown in Fig. 2 and imposing those symmetries will drastically re- duce the number o...

  8. [8]

    For each chart, we follow the PCA construction described in Sec

    Training procedure for a single chart Throughout this section we fixL D = 0.4 andL Λ = 4.0. For each chart, we follow the PCA construction described in Sec. IV using a finite local training set. Specifically, we drawN budget parameter vectors{X s}uniformly within the chart, compute the corresponding EH ground states, and perform the SVD of the resulting d...

  9. [9]

    V A, we take the pre-training domain to beD i ∈ [−1.0,0.0] and Λ i ∈[−10.0,10.0] fori= 1,2,3

    Pre-training construction and atlas statistics For the degenerate three-orbital problem introduced in Sec. V A, we take the pre-training domain to beD i ∈ [−1.0,0.0] and Λ i ∈[−10.0,10.0] fori= 1,2,3. Applying the procedure above across this pre-training domain yields an atlas consisting of 990 charts. In prac- tice, we find that the atlas construction sa...

  10. [10]

    method of snapshots

    Ghost-GA results with the pre-trained solver Having constructed the atlas and the corresponding database of projected operators, we next use the resulting compressed solver within full ghost-GA calculations for the three-orbital Hubbard–Kanamori model introduced in Sec. V A. We benchmark ghost-GA calculations in which the EH is solved using the pre-traine...

  11. [11]

    Ma- terials genome initiative for global competitiveness,

    National Science and Technology Council (U.S.), “Ma- terials genome initiative for global competitiveness,” (2011)

    2011

  12. [12]

    Design and discovery of materials guided by theory and computation,

    Long-Qing Chen, Li-Dong Chen, Sergei V. Kalinin, Gerhard Klimeck, Sanat K. Kumar, J¨ org Neugebauer, and Ichiro Terasaki, “Design and discovery of materials guided by theory and computation,” npj Computational Materials1, 15007 (2015)

    2015

  13. [13]

    Inhomogeneous electron gas,

    P. Hohenberg and W. Kohn, “Inhomogeneous electron gas,” Phys. Rev.136, B864 (1964)

    1964

  14. [14]

    Self-consistent equations in- cluding exchange and correlation effects,

    W. Kohn and L. J. Sham, “Self-consistent equations in- cluding exchange and correlation effects,” Phys. Rev. 140, A1133 (1965)

    1965

  15. [15]

    Toward a predic- tive theory of correlated materials,

    Paul R. C. Kent and Gabriel Kotliar, “Toward a predic- tive theory of correlated materials,” Science361, 348–354 (2018)

    2018

  16. [16]

    Eschrig,N

    H. Eschrig,N. F. Mott Metal-Insulator Transition(Tay- lor & Francis, London, 1990) p. 286, 166 figures

    1990

  17. [17]

    Phase stability and properties of manganese oxide polymorphs: Assessment and insights from diffusion Monte Carlo,

    Joshua A. Schiller, Lucas K. Wagner, and Elif Ertekin, “Phase stability and properties of manganese oxide polymorphs: Assessment and insights from diffusion Monte Carlo,” Phys. Rev. B92, 235209 (2015)

    2015

  18. [18]

    Polymorphic energy or- dering of MgO, ZnO, GaN, and MnO within the random phase approximation,

    Haowei Peng and Stephan Lany, “Polymorphic energy or- dering of MgO, ZnO, GaN, and MnO within the random phase approximation,” Phys. Rev. B87, 174113 (2013)

    2013

  19. [19]

    Energetic sta- bility and magnetic properties of MnO in the rocksalt, wurtzite, and zinc-blende structures: Influence of ex- change and correlation,

    A. Schr¨ on, C. R¨ odl, and F. Bechstedt, “Energetic sta- bility and magnetic properties of MnO in the rocksalt, wurtzite, and zinc-blende structures: Influence of ex- change and correlation,” Phys. Rev. B82, 165109 (2010). 15

    2010

  20. [20]

    Quantum Monte Carlo calculations of structural properties of FeO under pressure,

    Jind ˇ rich Kolorenˇ c and Lubos Mitas, “Quantum Monte Carlo calculations of structural properties of FeO under pressure,” Phys. Rev. Lett.101, 185502 (2008)

    2008

  21. [21]

    Magnetic collapse and the behavior of tran- sition metal oxides at high pressure,

    I. Leonov, L. Pourovskii, A. Georges, and I. A. Abrikosov, “Magnetic collapse and the behavior of tran- sition metal oxides at high pressure,” Phys. Rev. B94, 155135 (2016)

    2016

  22. [22]

    Structural predictions for correlated electron materials using the functional dynamical mean field theory approach,

    Kristjan Haule, “Structural predictions for correlated electron materials using the functional dynamical mean field theory approach,” Journal of the Physical Society of Japan87, 041005 (2018)

    2018

  23. [23]

    Connection between mott physics and crystal structure in a series of transition metal binary compounds,

    Nicola Lanat` a, Lee Tsung-Han, Vladan Stevanovi´ c Yong- Xin, Yao, and Vladimir Dobrosavljevi´ c, “Connection between mott physics and crystal structure in a series of transition metal binary compounds,” npj Comput. Mater.5, 30 (2019)

    2019

  24. [24]

    Origins of the odd optical observables in plutonium and americium tungstates,

    J. N. Cross, T.-H. Lee, C.-J. Kang, Y.-X. Yao, S. K. Cary, J. T. Stritzinger, M. J. Polinski, C. D. McKin- ley, T. E. Albrecht-Schmitt, and N. Lanat` a, “Origins of the odd optical observables in plutonium and americium tungstates,” Chemical Science10, 6508 (2019)

    2019

  25. [25]

    Un- covering the origin of divergence in the csm(cro 4)2 (m = la – sm; am) family through the chemical bonding in a molecular cluster and by band structure analysis,

    S. S. Galley, A. A. Arico, T.-H. Lee, X. Deng, Y.-X. Yao, J. M. Sperling, V. Proust, J. S. Storbeck, V. Dobrosavl- jevic, J. N. Neu, T. Siegrist, R. E. Baumbach, T. E. Albrecht-Schmitt, N. Kaltsoyannis, and N. Lanat` a, “Un- covering the origin of divergence in the csm(cro 4)2 (m = la – sm; am) family through the chemical bonding in a molecular cluster an...

    2018

  26. [26]

    Doping a mott insulator: Physics of high-temperature superconductivity,

    Patrick A. Lee, Naoto Nagaosa, and Xiao-Gang Wen, “Doping a mott insulator: Physics of high-temperature superconductivity,” Rev. Mod. Phys.78, 17–85 (2006)

    2006

  27. [27]

    High- temperature superconductivity in iron-based materials,

    Paglione Johnpierre and Greene Richard L., “High- temperature superconductivity in iron-based materials,” Nat. Phys.6, 645 (2010)

    2010

  28. [28]

    Role of the orbital degree of freedom in iron-based su- perconductors,

    Yi Ming, Zhang Yan, Shen Zhi-Xun, and Lu Donghui, “Role of the orbital degree of freedom in iron-based su- perconductors,” npj Comput. Mater.2, 57 (2017)

    2017

  29. [29]

    High temperature singlet-based mag- netism from hund’s rule correlations,

    Miao Lin, Basak Rourav, Ran Sheng, Xu Yishuai, Kotta Erica, He Haowei, Denlinger Jonathan D., Chuang Yi- De, Zhao Y., Xu Z., Lynn J. W., Jeffries J. R., Saha S. R., Giannakis Ioannis, Aynajian Pegor, Kang Chang-Jong, Wang Yilin, Kotliar Gabriel, Butch Nicholas P., and Wray L. Andrew, “High temperature singlet-based mag- netism from hund’s rule correlation...

    2019

  30. [30]

    Orbital- selective pairing and superconductivity in iron selenides,

    Nica Emilian M., Yu Rong, and Si Qimiao, “Orbital- selective pairing and superconductivity in iron selenides,” npj Comput. Mater.2, 24 (2017)

    2017

  31. [31]

    Mag- netism and its microscopic origin in iron-based high- temperature superconductors,

    Dai Pengcheng, Hu Jiangping, and Dagotto Elbio, “Mag- netism and its microscopic origin in iron-based high- temperature superconductors,” Nat. Phys.8, 709 (2012)

    2012

  32. [32]

    Superconductivity above 100 K in single-layer FeSe films on doped SrTiO 3,

    Ge Jian-Feng, Liu Zhi-Long, Liu Canhua, Gao Chun-Lei, Qian Dong, Xue Qi-Kun, Liu Ying, and Jia Jin-Feng, “Superconductivity above 100 K in single-layer FeSe films on doped SrTiO 3,” Nat. Mater.14, 285 (2015)

    2015

  33. [33]

    Interfacial supercon- ductivity in a bi-collinear antiferromagnetically ordered FeTe monolayer on a topological insulator,

    Manna S., Kamlapure A., Cornils L., H¨ anke T., Hede- gaard E. M. J., Bremholm M., Iversen B. B., Hofmann P., Wiebe J., and R. Wiesendanger, “Interfacial supercon- ductivity in a bi-collinear antiferromagnetically ordered FeTe monolayer on a topological insulator,” Nat. Com- mun.8, 14074 (2017)

    2017

  34. [34]

    Anisimov and Y

    V. Anisimov and Y. Izyumov,Electronic Structure of Strongly Correlated Materials(Springer, 2010)

    2010

  35. [35]

    First-principles calculations of the electronic structure and spectra of strongly cor- related systems: dynamical mean-field theory,

    V. I. Anisimov, A. I. Oteryaev, M. A. Korotin, A. O. Anokhin, and G. Kotliar, “First-principles calculations of the electronic structure and spectra of strongly cor- related systems: dynamical mean-field theory,” J. Phys. Condens. Matter9, 7359 (1997)

    1997

  36. [36]

    Realistic investigations of correlated electron systems with LDA+DMFT,

    K. Held, A. Nekrasov, G. Keller, V. Eyert, N. Bl¨ umer, A. K. McMahan, R. T. Scalettar, Th. Pruschke, V. I. Anisimov, and D. Vollhardt, “Realistic investigations of correlated electron systems with LDA+DMFT,” Phys. Stat. Sol. (B)243, 2599 (2006)

    2006

  37. [37]

    Dynamical mean-field theory of strongly corre- lated fermion systems and the limit of infinite dimen- sions,

    A. Georges, G. Kotliar, W. Krauth, and M. J. Rozen- berg, “Dynamical mean-field theory of strongly corre- lated fermion systems and the limit of infinite dimen- sions,” Rev. Mod. Phys.68, 13 (1996)

    1996

  38. [38]

    Implementation of LDA+DMFT with the pseudo-potential-plane-wave method,

    J.-Z. Zhao, J.-N. Zhuang, X.-Y. Deng, Y. Bi, L.-C. Cai, Z. Fang, and X. Dai, “Implementation of LDA+DMFT with the pseudo-potential-plane-wave method,” Chin. Phys. B21, 057106 (2012)

    2012

  39. [39]

    Electronic structure calculations with dynamical mean-field theory,

    G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko, O. Parcollet, and C. A. Marianetti, “Electronic structure calculations with dynamical mean-field theory,” Rev. Mod. Phys.78, 865 (2006)

    2006

  40. [40]

    Density matrix embed- ding: A simple alternative to dynamical mean-field the- ory,

    G. Knizia and G. K.-L. Chan, “Density matrix embed- ding: A simple alternative to dynamical mean-field the- ory,” Phys. Rev. Lett.109, 186404 (2012)

    2012

  41. [41]

    Spectral functions of strongly correlated extended systems via an exact quan- tum embedding,

    G. H. Booth and G. K.-L. Chan, “Spectral functions of strongly correlated extended systems via an exact quan- tum embedding,” (2013), arXiv:cond-mat/1309.2320

    Pith/arXiv arXiv 2013

  42. [42]

    Rotationally invariant slave-boson formalism and momentum dependence of the quasiparticle weight,

    F. Lechermann, A. Georges, G. Kotliar, and O. Par- collet, “Rotationally invariant slave-boson formalism and momentum dependence of the quasiparticle weight,” Phys. Rev. B76, 155102 (2007)

    2007

  43. [43]

    Spin-rotation- invariant slave-boson approach to the Hubbard model,

    T. Li, P. W¨ olfle, and P. J. Hirschfeld, “Spin-rotation- invariant slave-boson approach to the Hubbard model,” Phys. Rev. B40, 6817 (1989)

    1989

  44. [44]

    Slave Boson Theory of Orbital Differentiation with Crystal Field Effects: Ap- plication to UO 2,

    Nicola Lanat` a, Yongxin Yao, Xiaoyu Deng, Vladimir Do- brosavljevi´ c, and Gabriel Kotliar, “Slave Boson Theory of Orbital Differentiation with Crystal Field Effects: Ap- plication to UO 2,” Phys. Rev. Lett.118, 126401 (2017)

    2017

  45. [45]

    Correlation of Electrons in a Narrow sBand,

    M. C. Gutzwiller, “Correlation of Electrons in a Narrow sBand,” Phys. Rev.137, A1726 (1965)

    1965

  46. [46]

    Phase diagram and elec- tronic structure of praseodymium and plutonium,

    Nicola Lanat` a, Yong Xin Yao, Cai-Zhuang Wang, Kai- Ming Ho, and Gabriel Kotliar, “Phase diagram and elec- tronic structure of praseodymium and plutonium,” Phys. Rev. X5, 011008 (2015)

    2015

  47. [47]

    Fermi-surface evolution across the magnetic phase transition in the Kondo lattice model,

    N. Lanat` a, P. Barone, and M. Fabrizio, “Fermi-surface evolution across the magnetic phase transition in the Kondo lattice model,” Phys. Rev. B78, 155127 (2008)

    2008

  48. [48]

    Correlated lattice Fermions ind=∞dimensions,

    W. Metzner and D. Vollhardt, “Correlated lattice Fermions ind=∞dimensions,” Phys. Rev. Lett.62, 324 (1989)

    1989

  49. [49]

    Local density approximation combined with Gutzwiller method for correlated electron systems: Formalism and applica- tions,

    X.-Y. Deng, L. Wang, X. Dai, and Z. Fang, “Local density approximation combined with Gutzwiller method for correlated electron systems: Formalism and applica- tions,” Phys. Rev. B79, 075114 (2009)

    2009

  50. [50]

    Gutzwiller density functional theory for correlated electron sys- tems,

    K. M. Ho, J. Schmalian, and C. Z. Wang, “Gutzwiller density functional theory for correlated electron sys- tems,” Phys. Rev. B77, 073101 (2008)

    2008

  51. [51]

    Efficient implementation of the Gutzwiller variational method,

    N. Lanat` a, H. U. R. Strand, X. Dai, and B. Hellsing, “Efficient implementation of the Gutzwiller variational method,” Phys. Rev. B85, 035133 (2012). 16

    2012

  52. [52]

    Gutzwiller-correlated wave functions: appli- cation to ferromagnetic nickel,

    J. B¨ unemann, F. Gebhard, T. Ohm, S. Weiser, and W. Weber, “Gutzwiller-correlated wave functions: appli- cation to ferromagnetic nickel,” inFrontiers in Magnetic Materials, edited by A.V. Narlikar (Springer, Berlin,

  53. [53]

    Properties of Gutzwiller wave functions for multiband models,

    Claudio Attaccalite and Michele Fabrizio, “Properties of Gutzwiller wave functions for multiband models,” Phys. Rev. B68, 155117 (2003)

    2003

  54. [54]

    Emergent Bloch excitations in Mott matter,

    Nicola Lanat` a, Tsung-Han Lee, Yong-Xin Yao, and Vladimir Dobrosavljevi´ c, “Emergent Bloch excitations in Mott matter,” Phys. Rev. B96, 195126 (2017)

    2017

  55. [55]

    Quantum embedding description of the Anderson lattice model with the ghost Gutzwiller approximation,

    Marius S. Frank, Tsung-Han Lee, Gargee Bhattacharyya, Pak Ki Henry Tsang, Victor L. Quito, Vladimir Do- brosavljevi´ c, Ove Christiansen, and Nicola Lanat` a, “Quantum embedding description of the Anderson lattice model with the ghost Gutzwiller approximation,” Phys. Rev. B104, L081103 (2021)

    2021

  56. [56]

    Operatorial formulation of the ghost ro- tationally invariant slave-boson theory,

    Nicola Lanat` a, “Operatorial formulation of the ghost ro- tationally invariant slave-boson theory,” Phys. Rev. B 105, 045111 (2022)

    2022

  57. [57]

    Derivation of the ghost gutzwiller ap- proximation from quantum embedding principles: Ghost density matrix embedding theory,

    Nicola Lanat` a, “Derivation of the ghost gutzwiller ap- proximation from quantum embedding principles: Ghost density matrix embedding theory,” Phys. Rev. B108, 235112 (2023)

    2023

  58. [58]

    Ma- chine learning for many-body physics: The case of the Anderson impurity model,

    Louis-Fran ¸ cois Arsenault, Alejandro Lopez-Bezanilla, O. Anatole von Lilienfeld, and Andrew J. Millis, “Ma- chine learning for many-body physics: The case of the Anderson impurity model,” Phys. Rev. B90, 155136 (2014)

    2014

  59. [59]

    Data-driven dynam- ical mean-field theory: An error-correction approach to solve the quantum many-body problem using machine learning,

    Evan Sheridan, Christopher Rhodes, Francois Jamet, Ivan Rungger, and Cedric Weber, “Data-driven dynam- ical mean-field theory: An error-correction approach to solve the quantum many-body problem using machine learning,” Phys. Rev. B104, 205120 (2021)

    2021

  60. [60]

    Predicting im- purity spectral functions using machine learning,

    Erica J. Sturm, Matthew R. Carbone, Deyu Lu, Andreas Weichselbaum, and Robert M. Konik, “Predicting im- purity spectral functions using machine learning,” Phys. Rev. B103, 245118 (2021)

    2021

  61. [61]

    Learning impurity spectral functions from density of states,

    Xing-Yuan Ren, Rong-Sheng Han, and Liang Chen, “Learning impurity spectral functions from density of states,” Journal of Physics: Condensed Matter33, 495601 (2021)

    2021

  62. [62]

    Machine learning many-body green’s functions for molecular excitation spectra,

    Christian Venturella, Christopher Hillenbrand, Jiachen Li, and Tianyu Zhu, “Machine learning many-body green’s functions for molecular excitation spectra,” Jour- nal of Chemical Theory and Computation20, 143–154 (2023)

    2023

  63. [63]

    Predicting inter- acting green’s functions with neural networks,

    Egor Agapov, Oriol Bertomeu, Andr´ es Carballo, Chris- tian B. Mendl, and Aaron Sander, “Predicting inter- acting green’s functions with neural networks,” (2024), arXiv:2411.13644 [cond-mat.str-el]

    Pith/arXiv arXiv 2024

  64. [64]

    Neural network impurity solver for real- frequency dynamical mean-field theory,

    Fenglin Deng, Yi Lu, Xiaodong Cao, and Zhicheng Zhong, “Neural network impurity solver for real- frequency dynamical mean-field theory,” (2025), arXiv:2511.14505 [cond-mat.str-el]

    arXiv 2025

  65. [65]

    Language-inspired machine learning approach for solving strongly correlated problems with dynamical mean-field theory,

    Hovan Lee, Zelong Zhao, George H. Booth, Weifeng Ge, and Cedric Weber, “Language-inspired machine learning approach for solving strongly correlated problems with dynamical mean-field theory,” Phys. Rev. B112, 035165 (2025)

    2025

  66. [66]

    Equiv- ariant neural network for green’s functions of molecules and materials,

    Xinyang Dong, Emanuel Gull, and Lei Wang, “Equiv- ariant neural network for green’s functions of molecules and materials,” Phys. Rev. B109, 075112 (2024)

    2024

  67. [67]

    Deep learning- based prediction of self-energies from ab initio dynam- ical mean-field theory for real materials with minimal data sets,

    Pragati Mitra and Hrishit Banerjee, “Deep learning- based prediction of self-energies from ab initio dynam- ical mean-field theory for real materials with minimal data sets,” ChemRxiv (2025), 10.26434/chemrxiv-2025- dp7rd

    doi:10.26434/chemrxiv-2025- 2025

  68. [68]

    Bypassing the compu- tational bottleneck of quantum-embedding theories for strong electron correlations with machine learning,

    John Rogers, Tsung-Han Lee, Sahar Pakdel, Wenhu Xu, Vladimir Dobrosavljevi´ c, Yong-Xin Yao, Ove Chris- tiansen, and Nicola Lanat` a, “Bypassing the compu- tational bottleneck of quantum-embedding theories for strong electron correlations with machine learning,” Phys. Rev. Research3, 013101 (2021)

    2021

  69. [69]

    Active learning approach to simulations of strongly correlated matter with the ghost gutzwiller approximation,

    Marius S. Frank, Denis G. Artiukhin, Tsung-Han Lee, Yongxin Yao, Kipton Barros, Ove Christiansen, and Nicola Lanat` a, “Active learning approach to simulations of strongly correlated matter with the ghost gutzwiller approximation,” Phys. Rev. Res.6, 013242 (2024)

    2024

  70. [70]

    The itensor software library for tensor network calculations,

    M. Fishman, S. R. White, and E. M. Stoudenmire, “The itensor software library for tensor network calculations,” SciPost Physics Codebases4(2022), 10.21468/SciPost- PhysCodeb.4

    doi:10.21468/scipost- 2022

  71. [71]

    Block2: A comprehensive open source framework to develop and apply state-of-the- art dmrg algorithms in electronic structure and beyond,

    H. Zhai, H. R. Larsson, S. Lee, Z.-H. Cui, T. Zhu, C. Sung, L. Peng, R. Peng, K. Liao, J. T¨ olle, J. Yang, S. Li, and G. K.-L. Chan, “Block2: A comprehensive open source framework to develop and apply state-of-the- art dmrg algorithms in electronic structure and beyond,” Journal of Chemical Physics159, 234801 (2023)

    2023

  72. [72]

    The density-matrix renormalization group,

    U. Schollw¨ ock, “The density-matrix renormalization group,” Rev. Mod. Phys.77, 259–315 (2005)

    2005

  73. [73]

    Density matrix formulation for quan- tum renormalization groups,

    Steven R. White, “Density matrix formulation for quan- tum renormalization groups,” Phys. Rev. Lett.69, 2863– 2866 (1992)

    1992

  74. [74]

    Density-matrix algorithms for quan- tum renormalization groups,

    Steven R. White, “Density-matrix algorithms for quan- tum renormalization groups,” Phys. Rev. B48, 10345– 10356 (1993)

    1993

  75. [75]

    Tree tensor-network real-time multiorbital impurity solver: Spin-orbit coupling and correlation functions in sr2ruo4,

    X. Cao, Y. Lu, P. Hansmann, and M. W. Haverkort, “Tree tensor-network real-time multiorbital impurity solver: Spin-orbit coupling and correlation functions in sr2ruo4,” Phys. Rev. B104, 115119 (2021)

    2021

  76. [76]

    Quantum impurity models using superpositions of fermionic gaussian states: Practical methods and applications,

    Samuel Boutin and Bela Bauer, “Quantum impurity models using superpositions of fermionic gaussian states: Practical methods and applications,” Phys. Rev. Res.3, 033188 (2021)

    2021

  77. [77]

    Disentangling interacting systems with fermionic gaussian circuits: Application to quantum impurity mod- els,

    Ang-Kun Wu, Benedikt Kloss, Wladislaw Krinitsin, Matthew T. Fishman, J. H. Pixley, and E. M. Stouden- mire, “Disentangling interacting systems with fermionic gaussian circuits: Application to quantum impurity mod- els,” Phys. Rev. B111, 035119 (2025)

    2025

  78. [78]

    Quantum assisted ghost gutzwiller ansatz,

    P.V. Sriluckshmy, Fran¸ cois Jamet, and Fedor ˇSimkovic IV, “Quantum assisted ghost gutzwiller ansatz,” arXiv preprint arXiv:2506.21431 (2025), 20 pages, 14 figures

    Pith/arXiv arXiv 2025

  79. [79]

    Adaptive variational quantum imaginary time evolution approach for ground state preparation,

    Niladri Gomes, Anirban Mukherjee, Feng Zhang, Thomas Iadecola, Cai-Zhuang Wang, Kai-Ming Ho, Pe- ter P Orth, and Yong-Xin Yao, “Adaptive variational quantum imaginary time evolution approach for ground state preparation,” Adv. Quantum Technol.4, 2100114 (2021)

    2021

  80. [80]

    Error mit- igation in variational quantum eigensolvers using tai- lored probabilistic machine learning,

    Tao Jiang, John Rogers, Marius S. Frank, Ove Chris- tiansen, Yong-Xin Yao, and Nicola Lanat` a, “Error mit- igation in variational quantum eigensolvers using tai- lored probabilistic machine learning,” Phys. Rev. Res. 6, 033069 (2024)

    2024

Showing first 80 references.