{"id":"73dda481-b134-4699-ba06-02709df578be","arxiv_id":"2507.05580","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The authors extend the Grassmann time-evolving matrix product operator method to polaron impurity problems by converting the phonon influence functional into a scalar reweighting of the fermionic path-integral tensor.","lead":"This paper extends a tensor network method to solve polaron impurity problems, where an electron drags a cloud of vibrations and also interacts with other electrons. The method works on imaginary and real time contours, avoids bath discretization error, and is checked against exact solutions, exact diagonalization, and quantum Monte Carlo, though its most advanced real-time results are not yet independently verified.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim requires that the first-order Trotter error be small for continuous-bath real-time Keldysh results, but those results are validated only by self-convergence; an independent benchmark is needed before they can serve as a baseline.","rationale":"I read the derivation in Sec. II.D carefully; the resolution of the finite-δτ issue in the phonon influence functional via the |a⟩⟨a| expansion (Eqs. 44-46) is technically sound and appears exact for the chosen Trotter splitting. The benchmarks against analytic solutions (Sec. III) and exact diagonalization (Sec. IV) provide strong evidence that the method is implemented correctly and that the only controllable errors are the time-step discretization and MPS bond truncation. However, the central claim of the paper is not just that the method works on toy models, but that it delivers accurate full-fledged real-time results on the Keldysh contour for continuous baths, which the authors explicitly propose as a benchmark baseline. Those results are supported only by self-convergence in χ and δt. Self-convergence is necessary but not sufficient: a consistent algorithm with an uncontrolled first-order Trotter error can converge to a wrong limit, and the magnitude of that error depends on non-commuting Hamiltonian terms that are not assessed a priori. The absence of an independent real-time benchmark for the continuous-bath case leaves the most novel part of the claim unverified. I agree with the reader's weakest assumption, and I do not think this rises to rejection because the imaginary-time CTQMC benchmarks and the ED benchmarks for discrete baths strongly support the method's correctness in other regimes, and the authors are transparent about the lack of a benchmark. Thus the reader's CONDITIONAL verdict remains appropriate and unchanged.","tokens_in":31091,"tokens_out":13967,"duration_ms":162926,"concrete_test":"Compute the continuous-bath Keldysh observables of Sec. V.B (e.g., single flavor, α=1, β=10, t=5, δt=0.0125, χ=200) with an independent real-time impurity solver, such as the inchworm Monte Carlo method (Refs. [56-62]) or a converged tensor-network influence-functional code (e.g., Ref. [81]) using the same continuous spectral functions; compare G^{neq,>}(t), G^{neq,<}(t), and X^{neq}(t) to the published GTEMPO curves. If the independent reference differs by more than the reported δt-convergence error (order 1e-3 to 1e-2), the first-order discretization error is not controlled and the central real-time claim is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing premise is that the first-order Trotter/QuAPI discretization (Eq. 36, e^{-δτH} ≈ e^{-δτH_imp} e^{-δτH_el} e^{-δτH_ph}) is the only time-step error and that it is controllable at practical step sizes for the key new capability: real-time evolution on the Keldysh contour for continuous baths. In Sec. V.B, the full-fledged real-time results are validated only by self-convergence in χ and δt (Figs. 11-14); there is no independent benchmark, even though the abstract claims these results could serve as a 'benchmarking baseline.' The first-order error is not bounded a priori, and its magnitude depends on commutators of H_imp, H_el, and H_ph, which can be large for strong electron-phonon coupling and long real times. If this error is not actually small for the parameters used, the central claim of accurate real-time polaron simulations is not established. The toy-model ED benchmarks in Sec. IV (Figs. 6-8) support consistency for discrete baths, but they use delta-function spectral functions and do not directly validate the continuous-bath real-time observables. Hence the accuracy of the most novel results is not independently established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript generalizes the GTEMPO method to polaron impurity problems, where an impurity is coupled to both a fermionic electron bath and a bosonic phonon bath. The central technical step is the representation of the phonon influence functional as a normal bosonic MPS and its subsequent use as a scalar reweighting of the Grassmann tensor components of the bare impurity dynamics, which avoids the inaccurate replacement of the density operator by Grassmann bilinears in the discretized influence functional. The method is formulated on the imaginary, Keldysh, and Kadanoff contours. The paper benchmarks the method against analytic solutions of the independent bosons model, exact diagonalization of a two-bath toy model, and CTQMC for continuous-bath imaginary-time calculations. It also presents real-time Keldysh results for a full-fledged continuous-bath model, with convergence checks in bond dimension and time step.","tokens_in":31253,"tokens_out":16568,"duration_ms":183216,"significance":"If the method performs as claimed, it fills a clear gap: it provides real-time and multi-time impurity correlation functions for polaron problems with continuous baths, free of bath discretization error and sign problem, with essentially two controllable hyperparameters (time step and bond dimension). The central derivation (Eqs. 40-46) is clear and appears sound. The benchmarks against analytic, ED, and CTQMC references are appropriate and give strong support for the imaginary-time and the discrete-bath real-time results. The real-time continuous-bath results in Sec. V.B are the least independently verified part; because they are promoted in the abstract as a potential 'benchmarking baseline,' the lack of an independent reference is a load-bearing gap that should be addressed.","major_comments":[{"comment":"The real-time Keldysh results for the full-fledged model are validated only by self-convergence with respect to the bond dimension χ and time step δt, with the δt-convergence measured relative to a baseline at δt=0.0125 rather than against an independent solution. The abstract and Sec. VI present these results as a 'benchmarking baseline,' which is not yet supported by the evidence. The ED benchmarks in Sec. IV use delta-function spectral functions, and the authors argue that this is a harder case because the influence functional does not decay; nevertheless, those benchmarks do not directly establish the absolute accuracy of the continuous-bath real-time observables, where the first-order Trotter error in Eq. (36) is not bounded a priori and could in principle be larger for strong coupling. To support the central claim of accurate real-time polaron simulations, the authors should add an independent check—for example, a limiting case that reduces to a known solvable problem (such as α=0 or g=0), a comparison against an alternative real-time method in a restricted regime, or a careful statement that the results are unvalidated predictions rather than a benchmarking baseline.","section":"Sec. V.B (Figs. 11-14)"},{"comment":"The statement that 'the only two sources of errors in our method are the first-order discretization error of the impurity path integral... and the MPS bond truncation error' is somewhat imprecise: the discretization error also includes the QuAPI approximation of the influence-functional integrals (Eq. 27), and its magnitude for the continuous-bath Keldysh case is not quantified independently. Because the convergence plots in Sec. V.B compare one extended GTEMPO result against another (rather than against an exact or independent result), they demonstrate internal consistency but not absolute error control. The authors should either provide an external error scale for the real-time results or clarify that the claimed error control is only verified relative to the method's own convergence.","section":"Sec. VI and Eq. (27)"}],"minor_comments":[{"comment":"The inset axis labels in the figure panels read 'E δt', but the convergence variable analyzed in those insets is the imaginary-time step δτ; the figures should be relabeled to avoid confusion.","section":"Figs. 9 and 10"},{"comment":"The first factor in the expanded propagator is written as '⟨ak+1| ˆU ′ imp fN −1⟩' and should be typeset with an explicit ket, e.g., '⟨a_{k+1}| \\hat{U}'_{imp}|f_{N-1}\\rangle'.","section":"Eq. (43)"},{"comment":"The sentence 'The polaron problem is a very old problem... but still remain largely unsolved today' has a subject-verb disagreement; 'remain' should be 'remains'.","section":"Abstract"},{"comment":"The statement that 'for this model the choice of δτ is irrelevant for extended GTEMPO' would benefit from a one-sentence reminder that QuAPI is exact here because the relevant paths are piecewise constant on the time grid, as shown in Appendix A.","section":"Sec. III, first paragraph"},{"comment":"The phrase 'In the next we will consider' should be 'In this section we consider' or 'Next, we consider'.","section":"Sec. IV, first sentence"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid methodological extension of GTEMPO with a clear new ingredient (the phonon-IF scalar reweighting). The main obstacle is the validation of the new real-time continuous-bath results: the paper's strongest claim—that these results can serve as a benchmarking baseline—rests on self-convergence alone. I would be comfortable with publication after the authors either add an independent benchmark or explicitly temper the baseline claim. The reliance on the authors' prior GTEMPO papers is acceptable because the new step is clearly delineated. The minor issues are easily fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis paper delivers what it claims, with one significant caveat. The genuine novelty is the conversion of the phonon influence functional, which lives naturally in a bosonic Fock basis, into a scalar reweighting of the components of the fermionic Grassmann tensor (Eqs. 45-46). The authors correctly identify why a naive coherent-state replacement \\hat n -> \\bar a a fails at finite δτ (Eq. 42), and fix it by building the phonon IF as a bosonic MPS and multiplying it elementwise into the GMPS for the impurity. That is a clean, modular idea, and it extends GTEMPO to electron-phonon impurity problems on imaginary, Keldysh, and Kadanoff contours without bath discretization or sign problems, including multi-time correlations and non-diagonal couplings.\n\nThe numerical work is considerably more careful than is usual for a methods paper. Benchmarks cover the independent bosons model analytically, a two-bath toy model against exact diagonalization, and full-fledged continuous-bath imaginary-time results against CTQMC. Errors are shown to decay with bond dimension and time step, and no parameter is fitted to the target observable. The toy-model real-time ED comparisons are a real strength, and their observation that delta-function baths are the harder case for the IF tensor network (non-decaying long-range couplings) is a fair point.\n\nThe main soft spot is that the full-fledged real-time Keldysh results have no independent benchmark; they are validated only by self-convergence in χ and δt. The paper is honest about this - Section V.B says no reliable benchmark exists - but it then calls those results 'a benchmarking baseline for future method developments' in the abstract. That step is premature. Self-convergence plus the discrete-bath ED tests are suggestive, not establishing, and the first-order Trotter/QuAPI error is not bounded a priori; at strong coupling or long real times it could be the dominant error. The remedy is modest: either an independent real-time check when one becomes available (e.g. against inchworm for a continuous bath) or at least a sharper quantitative bound on the δt error over the simulated window, plus a release of code or data. The absence of code is a minor but real handicap for reproducibility.\n\nOverall the central mechanism is sound, the benchmarks are appropriate, and the limitations are disclosed rather than hidden. The citation pattern is appropriate; the self-citations are to the authors' own GTEMPO machinery, which this work directly extends. I'd send this to peer review and expect it to be accepted after the baseline claim is tempered and, ideally, after code/data release. It will be useful to anyone working on DMFT with electron-phonon or retarded interactions, and to the GTEMPO/TEMPO community.","headline":"This paper extends GTEMPO to electron-phonon impurity problems with a clean scalar-reweighting trick, and needs only a tempered real-time baseline claim and code release to be a solid contribution.","tokens_in":31909,"tokens_out":6045,"would_cite":true,"duration_ms":57564,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["71.38.-k"],"model":"deepseek-v4-flash","headline":"This paper claims an extended tensor-network path-integral method solves polaron impurity problems in real time with only two controllable sources of error.","keywords":["polaron impurity problem","tensor network","Grassmann time-evolving matrix product operator","Feynman-Vernon influence functional","path integral","Keldysh contour","Kadanoff contour","real-time Green's functions"],"falsifier":"A conclusive test would be a strong-coupling continuous-bath polaron calculation with a large bond dimension, comparing the real-time greater Green's function against an independent numerically exact short-time benchmark: if reducing $\\delta t$ does not produce the linear error decrease reported here, or if the error at fixed $\\delta t$ grows with final time $t$ instead of staying controlled, the claim that only two controllable error sources remain would be falsified. For Kadanoff-contour calculations, a direct check is to fix $\\delta\\tau$, lower $\\delta t$, and see whether errors to the analytic independent-bosons solution stay bounded and decreasing as the paper's note about keeping the ratio roughly the same suggests.","tokens_in":30781,"feed_emoji":"⚛️","tokens_out":7239,"duration_ms":75872,"temperature":0.7,"pith_summary":"The paper aims to solve the polaron impurity problem, an old and still largely unsolved model in which an impurity is coupled to both an electron bath and a phonon bath, on imaginary-time, Keldysh, and L-shaped Kadanoff contours. Its central claim is that a tensor-network representation of the impurity path integral, called extended GTEMPO, makes both baths analytically integrable through the Feynman-Vernon influence functional, so the simulation has no bath discretization error. The method is claimed to handle non-diagonal impurity couplings and to compute multi-time correlations beyond single-particle Green's functions from the same compressed object. Extensive benchmarks against analytic solutions, exact diagonalization, and continuous-time quantum Monte Carlo support the claim that the only remaining error sources are time-step discretization and matrix-product-state bond truncation.","feed_headline":"Polaron impurity solver reaches real time without bath discretization","feed_subtitle":"Both electron and phonon baths are integrated out, leaving only time-step and tensor-bond errors to control.","key_machinery":"The load-bearing object is the augmented density tensor $A[\\bar a a] = K[\\bar a a] I_{\\mathrm{el}}[\\bar a a] I_{\\mathrm{ph}}[\\bar a a]$ discretized on a contour and represented as a Grassmann MPS. The key identity is the componentwise action of the phonon influence functional: after expanding in Fock states, $I_{\\mathrm{ph}}[\\hat n]$ multiplies each component $K[\\bar a a;\\bar n n] = \\eta[\\bar n n]\\langle \\bar n_0|\\hat U_{\\mathrm{imp}}|n_{M-1}\\rangle \\cdots \\langle \\bar n_1|\\hat U_{\\mathrm{imp}}|n_0\\rangle I_{\\mathrm{ph}}[n]$, so the phonon IF can be built as a normal MPS in the Fock basis with the partial-IF algorithm and applied before the electron IF. This ordering avoids the invalid replacement $\\hat n \\to \\bar a a$ at finite $\\delta\\tau$, which would introduce an extra first-order error.","core_discovery":"On its own terms, the paper establishes a construction: the augmented density tensor for the polaron impurity problem, $A[\\bar a a] = K[\\bar a a] I_{\\mathrm{el}}[\\bar a a] I_{\\mathrm{ph}}[\\bar a a]$, can be built entirely as a Grassmann matrix product state. The obstacle is that the phonon influence functional is naturally a bosonic object in the Fock basis, while the electron influence functional is a Grassmann object in the coherent-state basis, and replacing the density operator $\\hat n$ by $\\bar a a$ at finite time step is only first-order accurate. The paper's resolution is to apply the phonon influence functional to the Fock-state components of the impurity propagator first, forming $K I_{\\mathrm{ph}}$ by componentwise multiplication of a Grassmann MPS with a normal MPS, and only then multiply in the electron influence functional. With this ordering, all existing GTEMPO techniques apply, and the paper demonstrates accurate Matsubara, nonequilibrium, and equilibrium real-time Green's functions and density-density correlations, including full-fledged continuous-bath real-time calculations it states have not been done before.","pith_inferences":["If the claimed accuracy holds, the technique should provide reference real-time data for continuum-bath polaron models where imaginary-time quantum Monte Carlo data must be analytically continued; a natural check is to compare its real-time results with those obtained by real-time diagrammatic Monte Carlo at short times.","The first-order Trotter/QuAPI splitting is the likely bottleneck: the paper shows linear convergence in $\\delta t$ and notes that $\\delta\\tau$ and $\\delta t$ must be decreased together on Kadanoff contours, so a higher-order splitting or an error estimate for the splitting would be the natural next step.","Because the phonon influence functional is built once as a normal MPS in the Fock basis, the same $I_{\\mathrm{ph}}$ tensor may be reusable across contours and initial states, which would make thermal and quench calculations share the most expensive part of the construction."],"forward_implications":["Real-time greater and lesser Green's functions and density-density correlations for polaron impurity models can be computed directly on the Keldysh or L-shaped Kadanoff contour, without analytic continuation.","Bath discretization error is absent by construction, so only the time step and the bond dimension need to be converged.","Non-diagonal impurity-bath couplings and generic impurity Hamiltonians with off-diagonal flavor tunneling are treated on the same footing as diagonal ones.","The same augmented density tensor yields single-particle Green's functions and higher-order multi-time correlations at comparable cost.","The method can serve as an impurity solver in dynamical mean-field-type calculations with retarded interactions, where the phonon bath is the source of the retarded interaction."],"supporting_citations":[{"why":"Supplies the Feynman-Vernon influence functional that lets both baths be integrated out of the impurity path integral.","marker":"[63]"},{"why":"Provides the coherent-state expression of the fermionic influence functional used for the electron bath as a Grassmann tensor.","marker":"[64]"},{"why":"Introduces TEMPO, the bosonic MPS construction of the influence functional from partial influence functionals, reused here for the phonon bath.","marker":"[65]"},{"why":"Introduces GTEMPO, the Grassmann MPS framework in the coherent-state representation that the extended method builds on.","marker":"[67]"},{"why":"Supplies the refined first-order propagator treatment for the impurity dynamics that keeps the bare-impurity contribution accurate.","marker":"[68]"},{"why":"Extends GTEMPO to real-time Keldysh contour calculations, the basis for the nonequilibrium results.","marker":"[69]"},{"why":"Provides the efficient MPS construction of the influence functional used to keep the phonon and electron influence functionals manageable.","marker":"[70]"},{"why":"Extends GTEMPO to the L-shaped Kadanoff contour, the basis for equilibrium real-time results.","marker":"[71]"},{"why":"Supplies the QuAPI time-slicing discretization that produces the first-order time-step error controlled in the benchmarks.","marker":"[90]"}],"fun_headline_variants":["Tensor network polaron solver goes real-time with no bath cutoff","Bath-free real-time polaron impurity solver via tensor networks","Grassmann MPS polaron solver: real-time, no bath discretization","Polaron impurity: tensor network path integral in real time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central assumption is that the first-order splitting $e^{-\\delta\\tau H}\\approx e^{-\\delta\\tau H_{\\mathrm{imp}}}e^{-\\delta\\tau H_{\\mathrm{el}}}e^{-\\delta\\tau H_{\\mathrm{ph}}}$ makes the path integral accurate enough when the time step is small; the paper demonstrates convergence empirically, but does not bound this error a priori, so for strong coupling or long real times this discretization error could dominate.","fun_headline_variants_meta":{"raw":{"variants":["Tensor network polaron solver goes real-time with no bath cutoff","Bath-free real-time polaron impurity solver via tensor networks","Grassmann MPS polaron solver: real-time, no bath discretization","Polaron impurity: tensor network path integral in real time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00022,"raw_usage":{"total_tokens":1488,"prompt_tokens":1028,"completion_tokens":460,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":644,"completion_tokens_details":{"reasoning_tokens":385}},"tokens_in":644,"tokens_out":460,"duration_ms":5544,"temperature":1.0,"reasoning_tokens":385,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:23:30.624829+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A conclusive test would be a strong-coupling continuous-bath polaron calculation with a large bond dimension, comparing the real-time greater Green's function against an independent numerically exact short-time benchmark: if reducing $\\delta t$ does not produce the linear error decrease reported here, or if the error at fixed $\\delta t$ grows with final time $t$ instead of staying controlled, the claim that only two controllable error sources remain would be falsified. For Kadanoff-contour calculations, a direct check is to fix $\\delta\\tau$, lower $\\delta t$, and see whether errors to the analytic independent-bosons solution stay bounded and decreasing as the paper's note about keeping the ratio roughly the same suggests.","supporting_citations":[{"cited_title":"Bertrand, S","cited_arxiv_id":null,"evidence_quote":"Supplies the Feynman-Vernon influence functional that lets both baths be integrated out of the impurity path integral."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces TEMPO, the bosonic MPS construction of the influence functional from partial influence functionals, reused here for the phonon bath."},{"cited_title":"Thoenniss, M","cited_arxiv_id":null,"evidence_quote":"Supplies the refined first-order propagator treatment for the impurity dynamics that keeps the bare-impurity contribution accurate."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extends GTEMPO to real-time Keldysh contour calculations, the basis for the nonequilibrium results."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the efficient MPS construction of the influence functional used to keep the phonon and electron influence functionals manageable."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Extends GTEMPO to the L-shaped Kadanoff contour, the basis for equilibrium real-time results."},{"cited_title":"Jordan and E","cited_arxiv_id":null,"evidence_quote":"Supplies the QuAPI time-slicing discretization that produces the first-order time-step error controlled in the benchmarks."}],"review_version":1}