{"id":"98531d56-6fe7-4168-a271-eb69c843c8fe","arxiv_id":"2607.17934","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":2,"one_line_summary":"A hybrid non-Gaussian variational and DMRG solver gives accurate Dicke and Dicke-Ising ground states without explicit photon truncation and with smaller MPS bond dimension.","lead":"This paper combines a variational description of light with DMRG for spins to compute ground states of spin-boson models without truncating the photon field. It benchmarks the method on the Dicke and Dicke-Ising models, reporting accurate energies with smaller tensor-network bond dimensions than direct simulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Ansatz manifold restricted to homogeneous single-mode; multi-mode/site-dependent spin-boson claims unvalidated","rationale":"The reader and I identify the same load-bearing concern: the benchmarks only exercise the homogeneous single-mode manifold (Eq. 13), while the title and abstract claim applicability to general interacting spin-boson Hamiltonians (Eq. 1). The paper is transparent about this scope limitation: Sec. III B states 'we focus on the homogeneous single-mode form,' and Sec. VI says expanding to multi-mode and spin-phonon systems is 'our next step.' Thus the concern is an external validity gap, not an internal inconsistency. The internal claims for Dicke/Dicke-Ising are well-supported: energy errors decay with N (Fig. 4), order parameters match (Figs. 2–3), and the self-consistent loop converges exponentially (Fig. 5b). The runtime comparison has a minor asymmetry—500 NGS realizations versus presumably single DMRG realizations—but this is secondary to the ansatz-expressiveness issue. A concrete test—implementing Eq. (14) on spin-Holstein—would resolve whether the framework's generality claim holds. Since the reader's CONDITIONAL verdict already reflects the scope restriction, no verdict change is needed; the condition should be explicitly retained.","tokens_in":24814,"tokens_out":13151,"duration_ms":88409,"concrete_test":"Implement the multi-mode, site-dependent dressing of Eq. (14) in the provided NGS-DMRG package and benchmark on the spin-Holstein chain (Eq. 5) with N=20, ω_j=1, ε=1, and g=0.5,1.0,2.0. Compare ground-state energy per site to converged reference phonon DMRG with phonon cutoff n_ph=100. If the multi-mode NGS-DMRG reproduces DMRG energies to within 10^-6 per site across couplings, the framework's general applicability is supported; if not, the homogeneous single-mode validation cannot be extrapolated to the claimed model class.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central numerical validation uses only the homogeneous single-mode manifold: the dressing U_λ = exp(i(g'/ω)λ S_x p) plus diagonal squeezing (Sec. III B, Eq. 13; Sec. IV A 2). All benchmarks (Figs. 2–5, Appendices E–F) target collective single-mode Dicke and Dicke-Ising models. The paper's stated target class is the general Hamiltonian Eq. (1), which includes site-dependent couplings and multiple modes—notably spin-Holstein models (Sec. II C, Eq. 5). The paper explicitly defers these: 'we focus on the homogeneous single-mode form relevant to the cavity-QED limit' (Sec. III B) and 'our next step is to expand the software to include these models' (Sec. VI). Thus the load-bearing assumption is that a collective λ S_x dressing plus scalar squeezing captures all relevant spin-boson correlations. If the true ground state has site-dependent polaron shifts or multi-mode correlations, the method's accuracy is untested and may fail. The benchmarks are internally valid for Dicke/Dicke-Ising, but the broader 'interacting spin-boson models' claim rests on an unvalidated extensibility assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a hybrid variational method for interacting spin-boson Hamiltonians. The bosonic sector is represented by a Gaussian state with displacement and squeezing, augmented by a variational spin-dependent displacement (dressing) transformation that captures spin-boson correlations. Tracing out the bosons leads to an effective spin Hamiltonian that is solved by DMRG, and the variational parameters are optimized in a self-consistent loop. The general formalism is derived for multi-mode, site-dependent couplings, but the numerical implementation is specialized to homogeneous single-mode cavity-QED models. The method is benchmarked on the Dicke and Dicke-Ising models against converged spin-boson DMRG with explicit photon truncation. The authors report accurate ground-state energies and order parameters, lower MPS bond dimensions, and reduced runtime compared with the reference solver, with energy errors decreasing with system size.","tokens_in":25078,"tokens_out":9237,"duration_ms":92380,"significance":"If the claims hold, the proposed NGS-DMRG scheme is a useful alternative to explicit bosonic truncation in spin-boson DMRG, particularly for strongly coupled cavity-QED systems where the bosonic Hilbert space is large. The paper has clear strengths: explicit algebraic derivations of the effective spin Hamiltonian, a self-consistent variational loop, benchmarks against a well-converged external solver, checks of the photon-number cutoff in the reference calculations, and a publicly available code and data set. The manuscript also proposes a systematic manifold hierarchy for future accuracy improvements. The main gap is that the numerical validation is restricted to homogeneous single-mode collective models, whereas the general target class stated in Eq. (1) includes site-dependent and multi-mode couplings; this limits the scope of the broader claims.","major_comments":[{"comment":"The paper's stated target is the general Hamiltonian in Eq. (1), and Appendix B derives an effective Hamiltonian for the multi-mode, site-dependent dressing of Eq. (14). However, every numerical benchmark (Dicke and Dicke-Ising models, Figs. 2–5 and Appendices E–F) uses the homogeneous single-mode collective form. The text itself acknowledges this at Sec. III B ('we focus on the homogeneous single-mode form relevant to the cavity-QED limit') and Sec. VI ('our next step is to expand the software to include these models'). As a result, the broad claims about an 'interacting spin-boson' framework are validated only for a restricted subclass. I recommend either adding at least one proof-of-principle calculation for a site-dependent or multi-mode case (e.g., a spin-Holstein model, Eq. (5)), or explicitly restricting the title, abstract, and conclusions to 'single-mode cavity-QED models' and s","section":"Sec. III B, Eq. (14); Sec. VI; Figs. 2–5"},{"comment":"The accuracy claim is based on the absolute energy error |E0 − E_ref0| relative to reference DMRG. This quantity is not normalized by N or by the relevant energy scale, so the reported errors are difficult to interpret across system sizes. Additionally, the main text (Fig. 4b) states that the error decays 'algebraically' with N, while Appendix F says it decays 'roughly exponentially' with N. These statements are inconsistent and should be reconciled. The reference-DMRG photon-cutoff convergence check in Appendix E is shown only for N = 40, J = 0.8, while the benchmark uses nmax up to 350 for N = 100; a concise convergence summary for the largest system sizes and other J values would substantiate that the reference calculations are truly converged.","section":"Sec. V, Fig. 4; Appendix F"}],"minor_comments":[{"comment":"The caption lists only panels (a) and (b), but the text refers to 'Fig. 2(c)' for order parameters and to 'Fig. 2(b)' for non-Gaussian energies. The panel labels and caption should be corrected to match the actual figure.","section":"Fig. 2 caption and text"},{"comment":"The notation 'g′Sxx' is ambiguous; it should read g′ S_x x (or equivalent) to distinguish the spin operator S_x from the photon quadrature x.","section":"Eq. (2)"},{"comment":"The restriction of the covariance matrix to diagonal form, Γ = diag{e^{2ξ}/2, e^{−2ξ}/2}, assumes no x–p correlations. This is a variational restriction that is not justified in the text. A sentence explaining why this is expected to be optimal for Dicke-type coupling (or noting it as an additional ansatz constraint) would strengthen the presentation.","section":"Sec. IV A 2"},{"comment":"The convergence plot shows the energy difference per iteration, but the stopping criterion for the self-consistent outer loop is not stated in the main text or caption. The authors should specify the tolerance used.","section":"Sec. V, Fig. 5(b)"},{"comment":"The photon cutoffs nmax are listed with different values for the same N in different figures (e.g., N = 10 appears as 40 in Fig. 2 and 30 in Figs. 4–5). The notation should be unified or explained.","section":"Figs. 2, 4, 5"}],"recommendation":"major_revision","confidential_remarks":"The numerical work appears sound for the benchmarked homogeneous single-mode Dicke and Dicke-Ising models, and the method is a plausible contribution. The main issue is scope: the paper presents itself as a general spin-boson framework but validates only a restricted cavity-QED limit. This can likely be fixed by either adding one heterogeneous or multi-mode benchmark or by clearly restricting the claims in the title and abstract. I do not see a fatal technical error, but the current framing overstates the validated domain."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Solid, honest methods paper. The new piece is the DMRG backend: the bosonic mode and spin-boson correlations are folded into a compact non-Gaussian variational manifold (dressing plus displacement/squeezing), leaving a self-consistent effective spin Hamiltonian that DMRG solves. The benchmark on Dicke and Dicke-Ising models is the real contribution: it matches converged spin-boson DMRG across N=10-100, with energy error decaying in N and lower MPS bond dimension. Code and data are public. The derivations are explicit, particularly the general multi-mode effective Hamiltonian in Appendix B, and the variational hierarchy in Appendix D is a sensible roadmap even if not implemented.\n\nWhat the paper does well: the validation is external, not circular. The reference DMRG is checked for photon-cutoff convergence (Appendix E), the GS (Gaussian) limit provides a controlled upper-bound baseline, and the phase diagram reproduces known first- and second-order transitions, including the narrow AFM-SP coexistence window seen in QMC. The authors are also upfront about what they did not test. That honesty matters because the main caveat is scope.\n\nSoft spot: every benchmark is a homogeneous, single-mode, collective Dicke-type model, where the dressing ansatz is a single global lambda S_x p plus one squeezing parameter. The title and the general Hamiltonian (Eq. 1) promise site-dependent couplings and multiple modes - spin-Holstein models, for example - but those are explicitly deferred to future work (Sec. VI). So the general 'interacting spin-boson' claim is not yet demonstrated; the method is validated only for the collective cavity-QED limit. That is a scope limit, not a hidden flaw, as long as the authors keep the claim aligned with the evidence. A smaller concern: the runtime comparison averages NGS over 500 random initializations but does not report symmetric treatment for the reference DMRG. That could skew the wall-clock comparison, though not the accuracy results.\n\nWho it's for: anyone doing numerical ground-state calculations in strongly coupled cavity-QED or extended Dicke models. It deserves a serious referee. The referee should push for at least one multi-mode or site-dependent benchmark (or a title/abstract that matches the tested regime) and for clarity on the timing protocol. I'd send it to review.","headline":"A solid, reproducible methods paper that validates NGS+DMRG on collective single-mode Dicke models; the broader multi-mode/spin-Holstein claim is explicit future work, not demonstrated.","tokens_in":25579,"tokens_out":2821,"would_cite":true,"duration_ms":27791,"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":"A hybrid variational solver can match converged spin-boson DMRG accuracy in the Dicke and Dicke-Ising models while operating at a substantially reduced tensor-network bond dimension.","keywords":["spin-boson models","Dicke model","Dicke-Ising model","non-Gaussian variational states","dressing transformation","DMRG","cavity QED","superradiant phase transition"],"falsifier":"Run the same NGS-DMRG loop on a spin-Holstein chain with local phonon modes and site-resolved couplings, which the paper's general formalism permits but does not test, and compare against converged multi-mode DMRG for N=8–16: if the energy error does not decay with N, or the bond-dimension advantage disappears, the single-mode homogeneous manifold is missing correlations the method claims to capture.","tokens_in":24710,"feed_emoji":"⚛️","tokens_out":6455,"duration_ms":60640,"temperature":0.7,"pith_summary":"The paper tries to establish that strongly correlated spin-boson problems can be solved without truncating the bosonic Hilbert space by folding the boson mode and spin-boson entanglement into a compact non-Gaussian variational state, then solving only an effective spin Hamiltonian with DMRG. The central claim is that this hybrid loop is both accurate and cheaper: on the Dicke and Dicke-Ising models it reproduces converged spin-boson DMRG ground-state energies, order parameters, and phase boundaries with substantially reduced MPS bond dimension and runtime. If true, it offers a practical route to cavity-QED, spin-phonon, and related models where bosonic occupation is large and spin correlations are strong, without perturbatively eliminating the bosons. The key object is a variational dressing transformation that promotes the canonical polaron frame change into a tunable parameter, thereby generating photon-mediated all-to-all spin interactions that DMRG can handle.","feed_headline":"Photon dressing layer reproduces spin-boson DMRG at lower cost","feed_subtitle":"Dicke and Dicke-Ising ground states match converged reference DMRG with no photon cutoff and smaller tensors","key_machinery":"The load-bearing object is the non-Gaussian variational ansatz and its dressing transformation. U_λ = exp(i g'/ω λ S_x p) promotes the canonical polaron frame change from a fixed transformation to a variational layer controlled by λ: λ=0 gives a separable Gaussian state, while λ=1 corresponds to the polaron-decoupled frame. Applying U_λ to the Dicke Hamiltonian turns the linear spin-photon coupling into a photon-mediated all-to-all spin interaction ∼(S_x)^2 plus renormalized fields and squeezing-dependent exponentials, so that after averaging over the bosonic Gaussian layer the problem becomes an effective spin Hamiltonian. DMRG solves that spin problem, and the variational parameters (dress","core_discovery":"The paper claims that the ground state of a spin-boson Hamiltonian can be well approximated by writing it as |ψ_NGS⟩ = U_λ U_GS |0_b⟩ ⊗ |φ_s⟩, where U_GS is a Gaussian displacement-and-squeezing operation and U_λ = exp(i g'/ω λ S_x p) is a variational spin-dependent displacement. Holding the bosonic variational parameters fixed, the dressed problem reduces to an effective spin Hamiltonian with photon-mediated all-to-all (S_x)^2 interactions, whose ground state φ_s is computed by DMRG; the loop is iterated to self-consistency. Benchmarked on the Dicke and Dicke-Ising models from N=10 to 100, the resulting energies, photon numbers, magnetizations, staggered magnetization, and phase boundaries—","pith_inferences":["The single-mode homogeneous manifold is likely to underestimate correlations in models with local phonon modes or multiple cavities; a natural test is to allow site-dependent λ_im in the general dressing and compare against exact small-system results.","The optimized λ itself can be read as a quantitative measure of spin-boson entanglement: monitoring its finite-size flow could locate where the semi-classical λ→0 or polaron λ→1 limits become exact.","The runtime advantage may be specific to low-dimensional spin geometries, since the effective Hamiltonian carries a long-range (S_x)^2 term that becomes expensive for 2D and 3D backends; the paper's proposed extension to PEPS or neural quantum states would test this.","The claim that the thermodynamic limit of the Dicke-Ising model has vanishing spin-photon entanglement could be sharpened by computing the dressed correlator ⟨(S_x)^2⟩_NGS as a function of N and extracting its finite-size scaling exponent."],"forward_implications":["For the Dicke and Dicke-Ising models at N=10–100, the method reproduces converged spin-boson DMRG energies and order parameters while needing much smaller MPS bond dimensions, with the largest savings near the phase transition and in the AFM-superradiant coexistence window.","Because no photon cutoff is introduced on the variational side, the approach is designed for strong-coupling regimes where converged reference DMRG needs large photon truncation (up to n_max=350 for N=100 in the benchmarks).","The nested-manifold hierarchy implies that adding higher-order dressing generators (e.g., S_y x terms) yields a strictly lower variational energy, providing an in-principle systematic route toward controlled accuracy.","The framework is set up for multi-mode, site-dependent, and inhomogeneous couplings (e.g., spin-Holstein models), so the same solver architecture can be carried beyond the single-mode cavity limit."],"fun_headline_variants":["No photon cutoff in non-Gaussian spin-boson solver","Hybrid variational method matches DMRG for Dicke models","Non-Gaussian ansatz cuts spin-boson cost, no truncation","Spin-boson ground states via self-consistent variational loop","Reduced bond dimension with non-Gaussian spin-boson ansatz"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The ansatz is restricted to a homogeneous single-mode dressing G1 = λ S_x p plus diagonal squeezing (Sec. IV A 2; Appendix D), and its accuracy is benchmarked only on single-mode Dicke-type models; if real ground states require site-dependent dresses, multiple modes, or non-Gaussian bosonic fluctuations beyond squeezing in the targeted regimes, the claimed accuracy is not established.","fun_headline_variants_meta":{"raw":{"variants":["No photon cutoff in non-Gaussian spin-boson solver","Hybrid variational method matches DMRG for Dicke models","Non-Gaussian ansatz cuts spin-boson cost, no truncation","Spin-boson ground states via self-consistent variational loop","Reduced bond dimension with non-Gaussian spin-boson ansatz"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000169,"raw_usage":{"total_tokens":1076,"prompt_tokens":692,"completion_tokens":384,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":293}},"tokens_in":436,"tokens_out":384,"duration_ms":4394,"temperature":1.0,"reasoning_tokens":293,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T16:34:26.130815+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same NGS-DMRG loop on a spin-Holstein chain with local phonon modes and site-resolved couplings, which the paper's general formalism permits but does not test, and compare against converged multi-mode DMRG for N=8–16: if the energy error does not decay with N, or the bond-dimension advantage disappears, the single-mode homogeneous manifold is missing correlations the method claims to capture.","supporting_citations":[],"review_version":1}