{"id":"c7e5e325-76b5-4175-828f-2886f547bd9b","arxiv_id":"2607.13993","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"iNORG is an open-source C++ package that solves zero-temperature, density-density Anderson impurity models for DMFT using natural-orbital-constrained Hilbert spaces and Lanczos-based Green's functions.","lead":"The paper presents iNORG, a free open-source C++ solver for the quantum impurity problems at the heart of dynamical mean-field theory calculations. It compresses the Hilbert space with natural-orbital occupancy constraints, but the paper itself contains no accuracy benchmarks, so its headline quality claim is not demonstrated here.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The accuracy claim rests on an unvalidated Hilbert-space truncation: NOOC discards configurations such as #6 of Table 1 with no error bound or cross-check against exact diagonalization.","rationale":"The reader identified the same load-bearing assumption: the NOOC truncation discards configurations that are asserted to contribute negligibly, with no error bound or cross-check. My independent reading of the manuscript confirms this. The paper is a software description and the code appears real and open-source, but the central accuracy/efficiency claim is not validated within the manuscript itself. The only quantitative plot is a hybridization-fit error, not an impurity-solver benchmark, and the example sections lack outputs. This is exactly the gap that makes the verdict CONDITIONAL: if the truncation were tested against exact diagonalization or a trusted solver (e.g., NRG/CT-HYB) and found to converge systematically, the concern would be resolved. No internal inconsistency in the algorithm itself was identified; the issue is a missing validation of the core approximation. Therefore I do not move the verdict.","tokens_in":23201,"tokens_out":3140,"duration_ms":35410,"concrete_test":"Run iNORG on a small single-orbital Anderson impurity model (e.g., N_bath=8-10, U=3, Delta=0.3, using the same parameter regime as Fig. 3) and compare the ground-state energy, natural-orbital occupations, and Green's function G(i omega_n) against full exact diagonalization in the complete Fock space with no NOOC truncation. Repeat with progressively looser NOOC constraints (1+/- in outer groups, then 2+/-, then 3+/-) and check whether results converge smoothly to the exact values. If the truncated results do not converge to the exact ones as constraints are relaxed, the NOOC truncation is uncontrolled and the central accuracy claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—'high accuracy with reduced computational cost'—depends on the natural-orbital occupancy constraints (NOOC) being a faithful compression of the full Hilbert space. In Section 3.3, the PHES rule explicitly removes configuration #6 (h* 2 1 i) on the assertion that it 'contributes little to the ground state,' but no discarded-weight estimate, convergence extrapolation, or comparison with an untruncated calculation is provided. Sections 2.2.1 and 3.3 assume that freezing orbitals with occupations near 0 or 1 introduces errors of order n or (1-n), yet this is stated without demonstration for the target models. Moreover, Section 3.6's convergence criteria (energy change and occupation-norm difference between consecutive iterations) are self-consistency checks, not error bounds relative to the exact solution; a self-consistent fixed point may differ from the true ground state. The paper reports no solver-level accuracy benchmark: Fig. 3 and Section 3.2.2 quantify only the hybridization-function fitting error, not the accuracy of calculated Green's functions or energies, and the examples in Section 5 present parameter files but no output or comparison. The method's validity is deferred to earlier publications (refs. [14,15,20-23]). If the NOOC truncation or frozen-orbital approximation omits sectors with small coefficients but significant spectral weight, the solver's claimed accuracy would fail even though the code runs and converges internally.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript describes iNORG, an open-source C++ implementation of the natural orbitals renormalization group (NORG) method for zero-temperature quantum impurity solvers in DMFT. The paper reviews the NORG workflow: iterative optimization of the bath representation in a natural-orbital basis, Hilbert-space compression via natural-orbital occupancy constraints (NOOC), Lanczos ground-state solution, continued-fraction evaluation of diagonal Green's functions, a Krylov-subspace method for matrix-valued Green's functions, and a Levenberg-Marquardt routine for fitting the hybridization function. Two example input files are provided, one for a one-shot three-orbital DMFT impurity calculation and one for a self-consistent two-orbital Bethe-lattice DMFT calculation. The main claims are that iNORG provides high accuracy with reduced computational cost and a polynomial scaling of about O(N_bath^3).","tokens_in":23612,"tokens_out":3092,"duration_ms":34307,"significance":"If the accuracy and scaling claims are supported, iNORG would be a valuable open-source zero-temperature impurity solver for multi-orbital density-density DMFT calculations, complementing existing ED, CT-QMC, and NRG solvers. The open-source release, modular C++ design, and the availability of input examples are concrete strengths. However, the manuscript as submitted does not itself demonstrate that the solver produces accurate Green's functions or energies for the target models: the only quantitative benchmark (Fig. 3) measures the hybridization-function fitting error, not the solver error, and the examples in Section 5 give no output or comparison. The central accuracy claim therefore rests on the unquantified faith of the NOOC truncation and on earlier NORG publications, which is a load-bearing gap for a software paper claiming 'high accuracy'.","major_comments":[{"comment":"The manuscript contains no solver-level accuracy benchmark. Fig. 3 and the text of Section 3.2.2 quantify only the error of the hybridization-function discretization (the fit of Γ_imp to Γ_loc); they do not report the accuracy of the computed Green's functions, self-energies, or ground-state energies. The two examples in Section 5 list parameter files and run commands but present no output or comparison with a reference. This is a load-bearing omission because the abstract and Program Summary claim 'high accuracy' for the solver. I request at least one benchmark for a small impurity model where full ED is possible, and ideally a comparison with CT-QMC or published results for a standard DMFT test case, reporting the error in a physical observable rather than only the fitting residual.","section":"Section 5, Fig. 3, Section 3.2.2"},{"comment":"The PHES constraint is justified by the assertion that configuration #6 (h* 2 1 i) 'contributes little to the ground state', but no discarded-weight estimate, convergence extrapolation, or comparison with an untruncated calculation is provided. Similarly, Section 2.2.1 states that freezing nearly empty/full orbitals introduces an error of approximately n_k or 1-n_k, but this is not demonstrated for the target impurity models. Since the claimed accuracy depends directly on this truncation being faithful, the paper needs a quantitative check, e.g., a plot of ground-state energy or a Green's function versus the NOOC excitation limit, or a comparison with full ED for a small system that demonstrates the discarded weight is negligible.","section":"Section 3.3.3, Table 1"},{"comment":"The convergence criteria used in the NORG self-consistent loop — the energy change between consecutive iterations (step G) and the occupation-norm difference d(n_last,n_new) (step I) — are self-consistency checks, not error bounds relative to the exact ground state of the untruncated impurity model. A fixed point of the truncated NORG equations may differ from the exact solution. The manuscript should either provide an error bound or demonstrate, for a representative case, that the converged NORG result approaches the exact ED result as the NOOC constraints are relaxed.","section":"Sections 3.1 and 3.6"}],"minor_comments":[{"comment":"The Program Summary states '~O(N_bath^3)' overall scaling, while Section 2.2.2 says 'typically O(N_bath^3) or O(N_bath^4)' and also mentions O(N^4) for full-orbital transformation. Please harmonize these statements and define precisely whether N_bath includes the impurity orbitals and whether the quoted scaling is for one NORG sweep or the full self-consistent DMFT cycle.","section":"Program Summary / Section 2.2.2"},{"comment":"In Eq. (14), the symbol V is used both for the impurity-bath coupling matrix and inside V(ε_bath−z_n)^{-1}V^†. Please define the dimensions of V, ε_bath, and the products, and explain how the second and third sum-rule terms are normalized relative to the first term (the weights w_n).","section":"Section 3.2.1, Eq. (14)"},{"comment":"The example parameter file contains a truncated list 'fit_points [0,1,2,...,3780]'. For reproducibility, either provide the complete list in the supplementary material or describe the rule that generates the points, since the discrete Lehmann representation parameters alone do not fully specify the selection as printed.","section":"Section 5.1"},{"comment":"The continued-fraction expressions for G< and G> are written with the tridiagonal coefficients α and β, but the notation is easy to misread because the earlier derivation in Eqs. (22)-(24) uses E_0 in different places. A brief definition of the sign convention (z+H−E_0 vs. z−H+E_0) immediately before Eqs. (26)-(27) would improve clarity.","section":"Section 3.7, Eqs. (26)-(27)"},{"comment":"Checkpoint/restart functionality is mentioned but not documented. If it is implemented, please give the relevant input parameters and file formats; otherwise remove the claim.","section":"Section 4.4"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern lands: the manuscript's central accuracy claim is not supported by any in-manuscript solver benchmark. The authors may intend the accuracy evidence to reside in the cited NORG papers (Refs. [14,15,20-23]), but a CPC software paper should contain at least one direct validation of the solver itself. I would also recommend that the editor check whether the GitHub repository actually contains runnable code and whether the provided examples produce the claimed output, since no output files are included in the submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: this is a software paper for a real, public AGPL C++ package that implements the same group's NORG impurity solver. The genuinely new content is implementation-level: the PHES constraint rule, the NORG-Table caching structure, combinatorial/mixed-radix Fock encoding, and a Krylov-based matrix-valued Green's function assembly. The NORG-Table idea is clever, and the Krylov Green's function trick—one Lanczos run per source vector instead of a full-space solve at every frequency—is legitimately useful. The paper is also refreshingly explicit about scope: zero-temperature, single-site, density-density only, with finite-temperature and cluster extensions deferred.\n\nWhat the paper does well is describe the machinery clearly enough to reproduce. The code looks real (public repo, build instructions, MPI+MKL), and the algorithmic sections are concrete rather than hand-wavy. The claimed O(N_bath^3) scaling with the bath-only orbital transformation is plausible.\n\nThe soft spot is the one the stress-test note flags, and it is real. The central accuracy claim—'high accuracy with reduced computational cost'—is not demonstrated anywhere in the manuscript. Figure 3 shows only hybridization-function fitting error, not solver error. Section 5 gives parameter files but no outputs, no comparison, no runtime. Worse, the NOOC truncation, which is the method's scaling justification, is asserted to discard only negligible configurations. Table 1's configuration #6 is thrown out because it 'contributes little,' but no discarded-weight estimate, no convergence extrapolation, and no cross-check against exact diagonalization supports that assertion. The convergence criteria in Sections 3.6 and 3.7 are self-consistency checks, not error bounds relative to the exact ground state. That is a genuinely load-bearing gap: if the truncation is unfaithful, the polynomial scaling buys you the wrong answer.\n\nTo be fair, this is not a fatal flaw. The method has same-group publications behind it, and the package itself is a legitimate contribution. But a software paper claiming accuracy and efficiency needs at least one solver-level benchmark—a small ED comparison for energy and Green's functions, or a CT-QMC comparison at finite temperature where available—inside the paper, not in references 14, 15, and 20–23.\n\nWho gets value from this? Someone choosing an impurity solver, or writing their own, will find the implementation details instructive. I would not cite this paper as evidence that NORG is accurate; I would cite it as the implementation reference if I used the code.\n\nMy recommendation: send it to peer review, but make the authors add that missing validation before publication. A desk rejection would be too harsh for a real, open, carefully described solver; acceptance without the benchmark would be too generous.","headline":"A real open-source NORG solver with genuinely new implementation ideas, but the paper's accuracy claim rests on an unvalidated truncation and needs a benchmark before the package can be trusted.","tokens_in":24108,"tokens_out":2179,"would_cite":false,"duration_ms":26932,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper aims to establish that the natural orbitals renormalization group, as implemented in the open-source solver iNORG, offers a practical zero-temperature impurity solver for DMFT that is both accurate and polynomial in cost, achieve","keywords":["quantum impurity solver","dynamical mean-field theory","natural orbitals renormalization group","Anderson impurity model","Hilbert space truncation","Lanczos algorithm","Green's function","multi-orbital strongly correlated systems"],"falsifier":"Run iNORG on a small multi-orbital Anderson impurity model (for example, one or two impurity orbitals with a handful of bath sites) for which a full exact-diagonalization solution in the original basis is feasible, and compare ground-state energy and Green's functions. If the NORG results fail to approach the exact-diagonalization results as the NOOC constraints are relaxed, or if the PHES truncation changes the answer by more than the claimed accuracy in a regime where configuration #6 carries significant weight, the central accuracy claim would be falsified.","tokens_in":23101,"feed_emoji":"⚛️","tokens_out":4971,"duration_ms":52196,"temperature":0.7,"pith_summary":"The paper claims that the natural orbitals renormalization group (NORG), implemented in the open-source solver iNORG, can solve zero-temperature multi-orbital Anderson impurity models at polynomial cost while retaining high accuracy. The core idea is to rotate the non-interacting bath into the eigenbasis of the single-particle density matrix, where most orbitals are nearly full or nearly empty, and then freeze those inactive orbitals while allowing limited hole/electron excitations in the rest. The paper reports overall complexity scaling roughly as O(N_bath^3), a major improvement over the exponential cost of exact diagonalization. If correct, iNORG provides a DMFT impurity solver that avoids the sign problem and analytic continuation of quantum Monte Carlo while handling more orbitals than conventional numerical renormalization group. The paper documents the NOOC constraint hierarchy, Lanczos-based Green's functions, and two worked examples.","feed_headline":"Bath truncation solver claims polynomial cost scaling","feed_subtitle":"Open-source iNORG freezes nearly empty or full orbitals, enabling larger bath sizes in zero-temperature DMFT calculations.","key_machinery":"The central machinery is the iterative natural-orbital rotation: diagonalize the ground-state single-particle density matrix, transform the bath orbitals into natural orbitals, and classify each orbital as active or frozen according to its occupation number. The natural-orbital occupancy constraints (NOOC), with the particle-hole excitation state (PHES) rule as the most refined variant, define the truncated Hilbert space by allowing only a limited number of holes or electrons in nearly full or nearly empty orbital groups. The reduced Hamiltonian is solved by the Lanczos algorithm, and the NORG-Table data structure caches non-zero Hamiltonian entries so that only numerical coefficients need u","core_discovery":"The central claim is that natural orbitals provide a compact single-particle basis for the Anderson impurity model: most natural orbitals have occupations near 0 or 1, so freezing them loses little. iNORG iteratively solves the reduced Hamiltonian by Lanczos, computes the single-particle density matrix, rotates the bath into its natural-orbital basis, and uses nested occupancy constraints (NOOC, with the PHES rule as the refined variant) to define an active subspace. The paper's solution is that this self-consistent loop converges to a high-accuracy ground state and Green's functions at polynomial cost, roughly O(N_bath^3), for zero-temperature multi-orbital density-density impurity problems","pith_inferences":["A systematic extrapolation as NOOC constraints are loosened could turn the truncation into a controlled approximation; the paper does not demonstrate such a convergence study, but the constraint hierarchy makes it possible.","If natural-orbital compression works for density-density models, the same strategy may extend to cluster DMFT and finite temperature, where quantum Monte Carlo sign problems currently limit reach.","The PHES rule's explicit exclusion of configuration #6 suggests a practical diagnostic: monitoring the total weight of excluded configurations across self-consistency could give users an error estimate for the truncation.","Because only the bath block is transformed, the local interaction remains sparse, which is what supports the polynomial scaling; extending the method to off-diagonal hybridizations may require a generalized bath-fitting scheme to preserve this property."],"forward_implications":["DMFT calculations can use baths of tens to hundreds of sites per orbital with polynomial rather than exponential growth in cost.","Zero-temperature real-frequency spectral functions can be produced directly by the Lanczos continued-fraction and Krylov-projection methods, without analytic continuation.","Users can tune accuracy against cost by choosing among direct-product, joint, and PHES constraint rules.","As an open-source package, the solver can be embedded in DFT+DMFT workflows and reproduced by other groups.","The stated O(N_bath^3) complexity implies that the dominant scaling bottleneck is bath size, not the number of impurity orbitals, for the supported density-density models."],"fun_headline_variants":["Natural orbitals cut impurity solver cost to polynomial","Open-source iNORG freezes near-empty orbitals for speed","Bath truncation via natural orbitals yields polynomial scaling","iNORG: efficient quantum impurity solver with orbital freezing"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The central claim rests on the assumption that configurations discarded by the natural-orbital occupancy constraints—orbitals frozen when occupations sit near 0 or 1, and high-excitation sectors such as configuration #6 of Table 1—contribute negligibly to the ground state and to the computed Green's functions, an assumption stated without rigorous error bounds or convergence extrapolation.","fun_headline_variants_meta":{"raw":{"variants":["Natural orbitals cut impurity solver cost to polynomial","Open-source iNORG freezes near-empty orbitals for speed","Bath truncation via natural orbitals yields polynomial scaling","iNORG: efficient quantum impurity solver with orbital freezing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000178,"raw_usage":{"total_tokens":1106,"prompt_tokens":688,"completion_tokens":418,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":432,"completion_tokens_details":{"reasoning_tokens":354}},"tokens_in":432,"tokens_out":418,"duration_ms":4581,"temperature":1.0,"reasoning_tokens":354,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T03:04:33.585226+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run iNORG on a small multi-orbital Anderson impurity model (for example, one or two impurity orbitals with a handful of bath sites) for which a full exact-diagonalization solution in the original basis is feasible, and compare ground-state energy and Green's functions. If the NORG results fail to approach the exact-diagonalization results as the NOOC constraints are relaxed, or if the PHES truncation changes the answer by more than the claimed accuracy in a regime where configuration #6 carries significant weight, the central accuracy claim would be falsified.","supporting_citations":[],"review_version":1}