{"id":"c308f9c0-6b6b-4cb9-84a7-ec69cb28edce","arxiv_id":"1908.03979","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A tensor-network optimized unitary transforms the Hubbard model into Heisenberg and t-J effective models, confirming the half-filling result J = 4t^2/U but with poor accuracy for doped systems.","lead":"This paper maps the Hubbard model to simpler Heisenberg and t-J models by numerically optimizing a tensor-network unitary that aligns the Hubbard ground state with a projected low-energy state. The method reproduces the known exchange coupling at half-filling, but for doped systems it requires very large bond dimensions and has limited accuracy.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim depends on an unsupported assumption that a unitary optimized from a single ground state also block-diagonalizes the Hamiltonian for all low-energy states; the doped-case results in Section 5 already show this assumption breaking down.","rationale":"The reader's weakest_assumption matches my read: a one-vector optimization cannot certify a whole-subspace canonical transformation. I agree with the CONDITIONAL verdict. The half-filling J = 4t^2/U recovery (Figure 4) is genuine but only checks one scalar in the ground-state sector; it does not test the Hamiltonian-level claim. The doped case (Figures 5 and 6) is the natural experiment, and the paper itself reports poor accuracy and large bond dimensions. The concrete test above would determine whether the concern is fatal or merely a matter of requiring more gates/layers. I also note that Equation (1) appears to mis-specify the Hubbard interaction as a nearest-neighbor density-density term; if the numerical implementation followed that equation rather than the standard on-site U, the results would not describe the Hubbard model at all. Because the surrounding text and references make the intended model clear, I treat this as a typo and not the primary concern.","tokens_in":2930,"tokens_out":4243,"duration_ms":44556,"concrete_test":"On a small cluster amenable to exact diagonalization (e.g. N=6, P=4, or a 2x3 lattice), construct U using the paper's ground-state-only loss. Then compute H_eff = P U H U^dagger P, where P projects onto the no-double-occupancy subspace, and compare its lowest several eigenpairs against the exact Hubbard spectrum. In addition, compute the norm of the residual coupling P_perp U H U^dagger P. If this norm is not small compared with the Hubbard gap, or if excited-state energy errors exceed a few percent, the ground-state-only optimization does not yield a valid Hamiltonian-level mapping.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The method optimizes U using only the ground-state fidelity, 1 - <psi(0)|psi> (Section 2). The central claim, however, is that this same U maps the full Hubbard Hamiltonian into an effective low-energy model. For that to hold, U must approximately block-diagonalize H: residual couplings between the no-double-occupancy sector and the high-energy sector must be small, and all low-lying eigenstates, not just the ground state, must be mapped accurately. No such condition is imposed or checked. Figure 5 compares only a few energy levels for one small doped cluster, and the text in Section 5 admits that the accuracy is 'much worse', the required MPO bond dimension is large, and high-order terms remain important. Thus the load-bearing assumption is exactly the one that the paper's own doped-sector results suggest is unsatisfied.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes a numerical method to construct effective low-energy Hamiltonians for the Hubbard model. The method optimizes a gate-based unitary transformation that maps the Hubbard ground state onto its projection into the no-double-occupancy sector (spin space at half-filling, t-J space when doped), then applies the same unitary to the Hubbard Hamiltonian to obtain an effective Hamiltonian represented as an MPO. At half-filling, the exchange coupling extracted from the effective Hamiltonian is compared with the perturbative result J = 4t^2/U. For a doped system with N=6 and P=4, low-energy levels of the effective model are compared with the exact Hubbard spectrum as a function of U.","tokens_in":3205,"tokens_out":5467,"duration_ms":56965,"significance":"If the method is made fully reproducible and its key assumption is verified, it would offer a non-perturbative numerical route to effective low-energy models of strongly correlated systems, complementing perturbation theory. The half-filling consistency check in Figure 4 is a clean, parameter-free comparison against the leading-order result, and the paper is candid about the difficulties in the doped case. However, the paper as written does not establish the central claim that the optimized unitary transforms the entire low-energy subspace correctly, and the numerical details are too under-specified to allow reproduction.","major_comments":[{"comment":"The loss function is ambiguous: the expression '1 - <psi(0)|psi>' is not meaningful unless both states are normalized, and the projected state is not normalized as defined; the paper also does not specify how the unitary gates are parameterized, initialized, or updated, which prevents reproduction of the method and is load-bearing for the central claim.","section":"Section 2"},{"comment":"The paper assumes that a unitary optimized using only the ground state also transforms the full Hamiltonian into an effective model for the low-energy subspace, but this assumption is not checked; the doped results in Section 5, including 'much worse' accuracy, large MPO bond dimensions, and the importance of high-order terms, indicate that the assumption may already be violated in the doped case.","section":"Sections 2 and 5"},{"comment":"The comparison between the effective model and the Hubbard model in the doped case is only qualitative; no numerical error metric is reported, so the claim that the effective model successfully reproduces the low-energy levels is not quantified and cannot be properly assessed.","section":"Section 5, Figure 5"},{"comment":"The convergence plot shows that the error at a fixed number of steps increases with system size, and for N=50 the error appears to saturate near 10^-1, which is inconsistent with the statement that three-site DMRG-like gates 'can reach very high accuracy'; the paper should quantify the achieved accuracy and discuss the scaling behavior.","section":"Section 3, Figure 3"},{"comment":"The Hubbard Hamiltonian is miswritten: the hopping term uses the same site index for both creation and annihilation operators, and the interaction term is written as a sum over bonds rather than an on-site term; this makes the central definition ambiguous even if it is only a typographical error.","section":"Eq. (1)"}],"minor_comments":[{"comment":"The MERA disentanglers are attributed to reference [4], which is the DMRG review by Schollwoeck; the correct citation is [5] (Vidal).","section":"Section 3"},{"comment":"The Heisenberg-model mapping is cited to reference [5], but the appropriate reference is [6] (Cleveland and Medina).","section":"Section 4"},{"comment":"The spin-spin interaction should be written as S_i · S_j; the dot product is missing.","section":"Eq. (2)"},{"comment":"'Matrix product state(MPO)' should read 'matrix product operator (MPO)'.","section":"Abstract"},{"comment":"There are several typos and formatting issues, for example 'change the of the doubly-occupied sites' in the text after Eq. (3) and 'excited statesexcited states' in Section 5.","section":"Throughout"},{"comment":"The reference list omits journal and volume information for several entries, including [5] and [6].","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reads like an early draft: the central idea is interesting, but the numerical method is underspecified and the main claim is supported only by a half-filling consistency check and a qualitative doped-case comparison. The authors should be asked to provide full details of the optimization, to validate the subspace-mapping assumption directly, and to add quantitative error measures. The citation mismatches should also be corrected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a short methods preprint with one new idea and one half-supported claim. The new thing is using a tensor-network unitary, optimized solely on the ground-state overlap, to map the Hubbard Hamiltonian into an effective spin or t-J model. The half-filling test is clean: the fitted J tracks 4t^2/U, and the paper does not oversell that part. The doped t-J section, admitted by the author, is not accurate enough to support the method's central promise.\n\nWhat is genuinely good: the numerical construction of effective Hamiltonians via disentanglers is a plausible non-perturbative alternative to canonical transformations, and I have not seen the exact ground-state-overlap optimization used this way. The paper is also honest about the limitations of the doped case — larger MPO bond dimensions, worse norm accuracy, and the need for high-order terms. The reference list is short but on-topic; there are no self-citation games.\n\nThe soft spots are real. Equation (1) is miswritten: the hopping term should connect different sites and the interaction should be on-site, so the notation as printed cannot be right. The projection, loss function, and optimization details are under-specified — no definition of the spin/t-J projector, no gate update rule, no stopping criteria. The stress-test concern lands: a unitary that maximizes overlap for a single ground state is not guaranteed to block-diagonalize the Hamiltonian, and the doped N=6, P=4 calculation already indicates the problem. Figure 5 compares a handful of levels but has no quantitative error measure, and the text concedes the effective model is poor. For a method paper, the lack of code or data is a significant barrier to checking.\n\nBottom line: this is an interesting proposal worth discussing, not yet a reliable method. I would not cite it as an established technique in the next year. But I would send it to a serious referee rather than desk reject: the idea is original, the half-filling check gives a concrete anchor, and a referee can push for the missing details. Expect a major-revision verdict.","headline":"A legitimate new numerical method for deriving effective models, with a clean half-filling check; the doped t-J case is not yet supported and the paper needs major revision.","tokens_in":3589,"tokens_out":2792,"would_cite":false,"duration_ms":31237,"reading_group":"yes","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 shows that a unitary transformation optimized from the Hubbard ground state maps the Hamiltonian into effective spin and t-J models, reproducing $J = 4t^2/U$ exchange at half-filling and approximating low-energy levels when…","keywords":["Hubbard model","t-J model","effective low-energy Hamiltonian","unitary transformation","tensor network","matrix product operator","DMRG","strongly correlated electrons"],"falsifier":"Take a small doped Hubbard cluster (for example $N=6$, $P=4$), build the unitary from the ground-state overlap, apply it to the full Hamiltonian, and compare the first several excitation energies with an exact diagonalization of the original Hubbard model; if the discrepancies do not shrink as the matrix-product-operator cutoff is lowered, the mapping is not a faithful low-energy model.","tokens_in":2687,"feed_emoji":"🧲","tokens_out":11245,"duration_ms":114337,"temperature":0.7,"pith_summary":"The paper tries to establish that effective low-energy models of the Hubbard model can be obtained without an analytic perturbation expansion, by numerically building a unitary transformation from the ground state. The unitary is chosen to send the Hubbard ground state into a subspace with no doubly occupied sites, and the same unitary is then applied to the full Hamiltonian. At half-filling the resulting spin Hamiltonian has an exchange coupling that tracks the textbook value $J = 4t^2/U$. For a doped 6-site, 4-particle cluster the resulting t-J model reproduces the low-lying Hubbard levels, though the paper reports that this doped mapping is much less accurate and needs a substantially larger tensor bond dimension.","feed_headline":"Ground-state unitary reproduces Hubbard exchange and t-J levels","feed_subtitle":"A numerical transformation from the Hubbard ground state matches the known exchange coupling and approximate t-J levels.","key_machinery":"The load-bearing object is a unitary built from layers of local unitary gates, arranged either in a sequential sweep across neighboring sites (DMRG-like gates) or in a hierarchical coarse-graining network (MERA-like gates). The gates are optimized to maximize the overlap between the Hubbard ground state and its projection into the no-double-occupancy subspace, equivalently minimizing the loss given by one minus that overlap. The Hamiltonian is represented as a matrix product operator, so applying the same unitary gives an effective Hamiltonian in tensor-network form. The optimization's accuracy is monitored by convergence of the overlap error with system size and with number of gate layers.","core_discovery":"The central discovery is that a single unitary, derived purely from ground-state overlap, carries the Hubbard Hamiltonian into an effective model of the same type perturbation theory produces. In the half-filled case the fitted exchange coupling follows $J = 4t^2/U$ over a range of $U$. In the doped case, effective-model eigenvalues track the exact Hubbard levels for the small system studied, but the author also finds that the effective matrix product operator needs large bond dimension and that high-order terms remain sizable, so a short effective Hamiltonian is hard to obtain.","pith_inferences":["A natural extension would be to optimize the unitary using several low-energy states rather than only the ground state; this would likely improve the doped case and could be checked by comparing more energy levels.","The doped-case difficulty suggests that, away from half-filling, the number of relevant high-order terms grows with doping, so the low-energy description may not localize into a few short-range operators on larger systems.","The same ground-state-derived unitary construction could be tried for multiband Hubbard or Kondo-lattice models where perturbation theory is uncontrolled, but the ground-state-only caveat would still apply."],"forward_implications":["At half-filling, the numerical unitary gives an effective spin Hamiltonian whose exchange coupling tracks $J=4t^2/U$, so the method reproduces the Heisenberg limit without classifying perturbation orders.","For doped clusters, the same construction yields a t-J-like effective model whose low-lying levels match the exact Hubbard spectrum on the systems tested.","The doped mapping is more demanding: useful accuracy requires a smaller cutoff and a larger matrix-product-operator bond dimension, and a few low-order terms do not suffice.","The convergence behavior with system size and gate layers gives a numerical handle for judging whether a proposed effective model is trustworthy."],"supporting_citations":[{"why":"Supplies the canonical-perturbation expansion of the Hubbard model that the numerical effective model is meant to reproduce.","marker":"[1]"},{"why":"Provides the ordered t-J effective-Hamiltonian expansion used to interpret the doped mapping.","marker":"[3]"},{"why":"Supplies the density-matrix renormalization-group method used to obtain the Hubbard ground state.","marker":"[4]"},{"why":"Cited in the text for the half-filling Heisenberg mapping with $J=4t^2/U$ that the numerical exchange coupling is checked against.","marker":"[5]"}],"fun_headline_variants":["Optimized unitary turns Hubbard into t-J model","Ground-state unitary recovers Hubbard's t-J limit","Numerical unitary maps Hubbard ground state to t-J","Hubbard to t-J via ground-state optimized unitary","Ground-state overlap yields effective Hubbard t-J"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that a unitary fixed by the ground state alone, without input from any excited state, also maps the Hamiltonian correctly on the whole low-energy subspace.","fun_headline_variants_meta":{"raw":{"variants":["Optimized unitary turns Hubbard into t-J model","Ground-state unitary recovers Hubbard's t-J limit","Numerical unitary maps Hubbard ground state to t-J","Hubbard to t-J via ground-state optimized unitary","Ground-state overlap yields effective Hubbard t-J"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00018,"raw_usage":{"total_tokens":1211,"prompt_tokens":762,"completion_tokens":449,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":378,"completion_tokens_details":{"reasoning_tokens":375}},"tokens_in":378,"tokens_out":449,"duration_ms":13981,"temperature":1.0,"reasoning_tokens":375,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:55:17.478242+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small doped Hubbard cluster (for example $N=6$, $P=4$), build the unitary from the ground-state overlap, apply it to the full Hamiltonian, and compare the first several excitation energies with an exact diagonalization of the original Hubbard model; if the discrepancies do not shrink as the matrix-product-operator cutoff is lowered, the mapping is not a faithful low-energy model.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the canonical-perturbation expansion of the Hubbard model that the numerical effective model is meant to reproduce."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the ordered t-J effective-Hamiltonian expansion used to interpret the doped mapping."}],"review_version":1}