{"id":"6ce80b0f-6d72-49f6-a83b-9abaada6ec4d","arxiv_id":"2506.14703","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An iPEPS study confirms the 2-zigzag and 3-zigzag antiferromagnetic phases of the square-lattice SU(3) Hubbard model at 1/3 filling, with transitions to the 3-sublattice stripe at U/t around 7.5 and 7.1.","lead":"Using tensor-network simulations, this paper maps out the magnetic ground states of the SU(3) Hubbard model on a square lattice at one particle per site, confirming a diagonal stripe phase at strong interactions and two zigzag antiferromagnetic phases at intermediate coupling. The results, which match prior quantum Monte Carlo and Hartree-Fock work, also explain which energy terms stabilize the zigzags.","discovery_kind":"replication","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Phase boundaries rest on simple-update iPEPS energy differences of ~0.001–0.006 t at D=24, with no extrapolation of the relative energies to D→∞; the AD cross-check only reaches D=6,8, so the 3-ZZ and 2-ZZ windows may not survive at infinite D.","rationale":"I read the paper in good faith. The main claim is credible in outline: the zigzag states were already reported by independent Hartree-Fock and CP-AFQMC studies, and the strong-coupling 3-SL state matches the known SU(3) Heisenberg result. The energy decomposition and the identification of corner bonds are useful and internally consistent. The single most load-bearing unproven premise is that the finite-D simple-update relative energy ordering at D=24 survives at D→∞. The reader's weakest-assumption assessment identifies the same point, and I agree with it. The paper's own statements confirm the difficulty: the D-dependence of the order parameter is unknown, the metal-insulator boundary is acknowledged as challenging for iPEPS, and no infinite-D extrapolation of the energy differences is attempted. The reported 0.7–2% shift of the transition values between D=21 and D=24 is suggestive of convergence but does not bound the systematic simple-update bias, which can be state-dependent. This does not warrant rejection, because the existence of zigzag order is supported by prior methods and the variational small-D data show consistent lowest-energy states. However, the quoted transition values U/t=7.1 and 7.5 should be regarded as approximate finite-D numbers rather than precisely determined thermodynamic-limit values. The reader's CONDITIONAL verdict is therefore the right one, and no adjustment is needed.","tokens_in":15202,"tokens_out":6140,"duration_ms":62074,"concrete_test":"At U/t=7.25 and U/t=6.0, compute full-update or automatic-differentiation iPEPS energies for the 2-ZZ, 3-ZZ, 4-ZZ, and 3-SL states at D=12 (and D=16 if feasible), with CTMRG χ converged so that the environment error is below about 2×10^-4 t, and extrapolate the energy differences ΔE(D) to 1/D→0 using both SU and variational data. If the variational or full-update data, or the joint extrapolation, does not keep 3-ZZ below 3-SL at U/t=7.25 and 2-ZZ lowest at U/t=6.0, the claimed phase boundaries and the intermediate zigzag windows are not established. A complementary check is a cylinder DMRG calculation (circumference 4 or 6) at the same couplings to verify the ordering independently.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central phase diagram (Fig. 1a, Fig. 4) is determined by comparing energies of the 2-ZZ, 3-ZZ, 4-ZZ, and 3-SL states from simple-update iPEPS at D=24. The energy differences that decide the ordering are extremely small, roughly 0.001–0.006 t per site (Fig. 4), and the phase boundaries are located by linear interpolation in U/t. The only variational cross-check (AD) reaches D=6,8, where the energy scale of the simple-update approximation error at fixed D is expected to be comparable to or larger than these physical differences. The paper reports that transition values shift by only 0.7–2% between D=21 and D=24, but this does not quantify the error at D→∞ nor the systematic simple-update bias relative to full-update or variational optimization. Since simple update truncates without full environment information, it can favor different states unequally; there is no controlled argument that the D=24 level ordering—and therefore the existence of the intermediate 3-ZZ phase and the width of the 2-ZZ window—survives in the thermodynamic limit. Previous HF/CP-AFQMC results support the existence of zigzag states, but they place transitions at substantially different U/t (5.65/4.75 and 8.3/7.7), so the specific boundaries here are not independently pinned. A quantitative χ-convergence statement is also missing; the statement that χ is 'sufficiently large' is not enough when energy differences are ~10^-3 t.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies the SU(3) Hubbard model on the square lattice at unit filling using infinite projected entangled-pair states (iPEPS), optimized with simple update (SU) at bond dimensions up to D=24 and with automatic differentiation (AD) at D=6,8. The central claim is a zero-temperature phase diagram containing three ordered Mott phases: the 3-sublattice diagonal-stripe phase at large U/t, and 3-zigzag and 2-zigzag antiferromagnetic phases at intermediate U/t, with transitions around U/t=7.5 and 7.1. The paper also reports discontinuous color order parameter and energy anisotropy across the transitions, and analyzes kinetic and on-site energy contributions to identify low-energy bonds at the zigzag corners that stabilize the zigzag states.","tokens_in":15554,"tokens_out":6630,"duration_ms":68517,"significance":"If the phase diagram is correct, it provides independent tensor-network evidence for the zigzag states previously found in Hartree-Fock and CP-AFQMC calculations, and it connects the strong-coupling SU(3) Heisenberg 3-sublattice order to intermediate-coupling states in a way that may be testable in ultracold-atom experiments. The manuscript has clear strengths: it uses a direct Hamiltonian with no fitted parameters, benchmarks the large-U limit against the known Heisenberg result, cross-checks SU results with variational AD at D=6,8, and gives a transparent decomposition of the energy balance. The main weakness is that the phase boundaries rest on extremely small energy differences at finite bond dimension, without a controlled extrapolation of relative energies to infinite D. The qualitative phase sequence is plausible and consistent with prior work, but the quantitative transition locations are not yet established at the same level of rigor as the qualitative ordering.","major_comments":[{"comment":"The central phase diagram is determined by simple-update iPEPS energy differences of order 0.001–0.006 t per site at D=24, with the transition points located by linear interpolation in U/t. These energy differences are not extrapolated to infinite D, and the variational AD cross-check reaches only D=6 and D=8; at those bond dimensions the fixed-D truncation error of the simple update is expected to be comparable to or larger than the physical splittings that decide the ordering. The reported 0.7–2% shift of the transition values between D=21 and D=24 does not bound the systematic simple-update bias relative to full-update or variational optimization. Since the existence and width of the 3-ZZ and 2-ZZ windows are exactly what these small differences determine, the paper should provide a quantitative D→∞ extrapolation of the relative energies (or an independent full-update/AD comparison at larger D) and an uncertainty estimate for the transition locations. The qualitative agreement with Hartree-Fock and CP-AFQMC, whose transition values differ by several t, does not by itself pin these boundaries.","section":"§IV.B, Figs. 3–4"},{"comment":"The environment bond dimension is controlled only by the statement that χ is “sufficiently large” so that the contraction error is small compared to the symbol sizes in the plots. This is not a quantitative convergence statement. Because the energy differences in Fig. 4 are as small as 10^-3 t per site, the manuscript should report the χ-dependence of the total energies or energy differences at the largest D and demonstrate that the environment truncation error is below this scale; otherwise the numerical uncertainty of the phase boundaries is not established.","section":"§III (CTMRG)"}],"minor_comments":[{"comment":"The statement “the data on the x-axis shows the phase diagram” is confusing: the x-axis is U/t, while the phase diagram is the identity of the lowest-energy state as a function of U/t.","section":"Figs. 3–4"},{"comment":"The U/t sampling grid underlying the linear interpolation is not specified; please give the grid spacing and state how the quoted transition values (7.1 and 7.5) are obtained from the interpolation.","section":"Fig. 4"},{"comment":"The caption does not define σ or explain how the shaded error band is computed from the linear extrapolations of m and m^2; please define the quantity and the fitting procedure.","section":"Fig. 5"},{"comment":"The bars are offset from zero by the values printed above them, but this is easy to miss; please state explicitly in the caption that the bars are truncated and that the printed values are the offsets.","section":"Fig. 6"},{"comment":"The check for competing color-imbalanced states is described without specifying the bond dimensions or U/t values at which it was performed; please add this information.","section":"§IV.B"}],"recommendation":"major_revision","confidential_remarks":"No editor-only concerns. The manuscript cites the relevant prior literature, including the Hartree-Fock and CP-AFQMC studies with which it compares. The main issue is technical rather than conceptual: the phase boundaries need stronger finite-D and environment-convergence control before the quantitative claims can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, this is a careful tensor-network study of the SU(3) Hubbard model at 1/3 filling. The main thing to know: the qualitative phase diagram is credible and matches prior CP-AFQMC and Hartree-Fock work (Feng et al. 2023), but the precise phase boundaries at U/t≈7.5 and 7.1 rest on simple-update iPEPS energy differences of ~10^-3 t at bond dimension D=24, with no extrapolation of those energy differences to D→∞. That makes the boundaries real but soft.\n\nWhat's new: the independent iPEPS confirmation itself, plus an energy decomposition (Figs. 6–7) showing the zigzag states stabilize via non-Heisenberg kinetic processes, especially on bonds at the zigzag corners. That is a transparent, useful addition. The color order parameter and energy anisotropy discontinuities are also cleanly presented.\n\nStrengths: the authors use two independent optimization schemes (simple update up to D=24, variational AD up to D=6,8), compare four competing states, and check color-imbalanced states. They are honest about D-dependence and cite the prior QMC/HF work squarely. There is no circularity: it's a direct simulation of Eq. (1).\n\nSoft spots: (1) The energy differences that decide the phase diagram are 0.001–0.006 t per site, comparable to the simple-update truncation error at finite D. The authors don't extrapolate ΔE to D→∞; they only quote 0.7–2% shifts in the transition values between D=21 and D=24. That is not the same as a controlled error bar. (2) The AD cross-check at D=6,8 is at a scale where simple-update errors can exceed the physical gaps, so it confirms trends but not the ordering at the transition. (3) The transition values disagree with both CP-AFQMC (8.3, 7.7) and HF (5.65, 4.75), so independent pinning is weak. (4) No code or data is provided, limiting reproducibility. These are not fatal; the qualitative sequence 3-SL → 3-ZZ → 2-ZZ is consistent across methods, but a careful referee should ask for a convergence statement on the energy differences.\n\nWho it's for: tensor-network practitioners and cold-atom SU(N) people. It deserves a serious referee: it's a well-executed numerical study with a new analytical angle. I'd accept it for review with requests for quantified uncertainty on the boundaries.\n\nRecommendation: send to peer review; conditional on a proper D-extrapolation or a clear statement of the systematic error.","headline":"Solid iPEPS confirmation of known zigzag phases in SU(3) Hubbard, with a useful energy decomposition; treat the quoted transition U's as indicative, not final.","tokens_in":16101,"tokens_out":2272,"would_cite":true,"duration_ms":21937,"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 establishes the zero-temperature phase diagram of the square-lattice SU(3) Hubbard model at one particle per site, in which the ground state passes from a 3-sublattice diagonal stripe antiferromagnet at strong coupling to a…","keywords":["SU(3) Hubbard model","square lattice","iPEPS","zigzag antiferromagnet","3-sublattice order","Mott insulator","ultracold fermions","phase diagram"],"falsifier":"An independent calculation—say, an unbiased auxiliary-field quantum Monte Carlo simulation or an iPEPS run at bond dimension beyond 24—that finds the 4-zigzag state lowest in energy anywhere in $U/t\\in[6,8]$, or finds no interval in which the 3-zigzag is the lowest state, would refute the claimed phase diagram.","tokens_in":14987,"feed_emoji":"🧲","tokens_out":9500,"duration_ms":88939,"temperature":0.7,"pith_summary":"The paper aims to pin down the zero-temperature ground states of the SU(3) Hubbard model on the square lattice at exactly one particle per site. Using infinite projected entangled-pair states, it finds three ordered Mott-insulating phases: a 3-sublattice diagonal-stripe antiferromagnet at strong coupling, and two zigzag antiferromagnets of segment length 3 and 2 at intermediate coupling, with transitions around $U/t=7.5$ and $U/t=7.1$. The zigzag states matter because they are a nontrivial color-ordering pattern for three-flavor fermions that cold-atom quantum simulators could in principle observe, and because the paper identifies the microscopic source of their stability: energy gains on bonds at the zigzag corners, mainly from non-Heisenberg kinetic processes. If the phase diagram is right, the model gives a concrete target for experiments and a benchmark for other numerical methods.","feed_headline":"SU(3) Hubbard model orders into zigzag antiferromagnets","feed_subtitle":"The 2- and 3-zigzag phases sit between stripe order and a metal, at interactions cold-atom gases can probe.","key_machinery":"The central object is the family of zigzag states—stripes of color order that turn by 90 degrees every $l$ sites along the diagonal—together with the 3-sublattice stripe as the $l\\to\\infty$ limit. The computational machinery is infinite projected entangled-pair states (iPEPS), a variational tensor-network ansatz for two-dimensional ground states in the thermodynamic limit, with a $U(1)^3$ flavor-conservation symmetry, simple-update optimization up to $D=24$, and automatic-differentiation cross-checks at smaller $D$. The argument is carried by energy comparisons between these candidate states and by a decomposition of the total energy into superexchange, other kinetic, and on-site contributions, which localizes the stabilizing gain on corner bonds.","core_discovery":"The central claim is that the square-lattice SU(3) Hubbard model at $n=1$ hosts two intermediate-coupling zigzag antiferromagnets, the 2-zigzag and 3-zigzag states, in addition to the known strong-coupling 3-sublattice diagonal stripe. All three states belong to one family: the 3-sublattice state is an infinite zigzag, and the finite zigzags are stripes that turn by 90 degrees every $l$ sites. Energy-versus-$1/D$ comparisons up to bond dimension $D=24$ place the first transition (3-sublattice to 3-zigzag) around $U/t=7.5$ and the second (3-zigzag to 2-zigzag) around $U/t=7.1$; the 4-zigzag competes closely but never wins. The color order parameter and the horizontal-vertical bond-energy anisotropy both jump at the transitions, and the stabilization of the zigzag states is traced to low-energy bonds at the zigzag corners, where both superexchange and other kinetic processes gain energy despite increased double occupancy.","pith_inferences":["The corner-bond mechanism suggests a geometric rule: shortening the straight runs of a stripe lowers energy at weaker coupling by adding low-energy corner bonds, which predicts that even longer zigzags ($l=5,6$) could appear in a narrow window at higher numerical accuracy—something the paper does not claim.","If the zigzag family is not special to three flavors, SU(N) Hubbard models at filling $1/N$ may show analogous stripe-turn orders, with the stable segment lengths controlled by $N$; this is an extrapolation beyond the paper's scope.","The mismatch between finite-order $m$ and vanishing $m^2$ extrapolations near $U/t=6.5$ leaves open a weakly first-order magnetic-to-metal transition; a direct calculation of the single-particle gap would decide between that and a continuous crossover."],"forward_implications":["If correct, cold-atom experiments with three-flavor fermions in a square optical lattice should see color correlations switch from diagonal stripes to 3-zigzag to 2-zigzag order as the interaction strength is lowered toward the metal.","The discontinuous jumps in color order and energy anisotropy imply both transitions are first-order-like, so site-resolved measurements should show sharp switching rather than continuous growth of order.","The vanishing of the squared color order parameter near $U/t=6.5(4)$ places the loss of magnetism in the same interaction range as the previously estimated metal-insulator transition, making the ordered phases occupy a finite window above it.","The paper's energy hierarchy explains why only $l=2$ and $l=3$ appear: the ordering of zigzag energies reverses between large and small $U/t$, so the 4-zigzag, although competitive, is never the lowest."],"supporting_citations":[{"why":"Supplies the prior Hartree-Fock and constrained-path auxiliary-field quantum Monte Carlo study that reported the same zigzag phases and the transition values (8.3 and 7.7) this work compares against.","marker":"[67]"},{"why":"Establishes three-sublattice ordering in the SU(3) Heisenberg model, the strong-coupling limit the 3-SL phase reproduces.","marker":"[49]"},{"why":"Provides the square-lattice SU(3) Heisenberg reference for the 3-sublattice state and the competing two-color benchmark used at large $U/t$.","marker":"[52]"},{"why":"Gives determinant quantum Monte Carlo results for the metal-insulator transition and magnetism used as comparison for the low-$U/t$ behavior.","marker":"[68]"},{"why":"Experimental equation-of-state study locating the metal-insulator transition around $U/t=5.5$–$6$, used to place the ordered phases relative to the metal.","marker":"[19]"}],"fun_headline_variants":["Zigzag antiferromagnets predicted in SU(3) Hubbard model","Zigzag magnetic order emerges in SU(3) Hubbard model","SU(3) Hubbard model reveals zigzag antiferromagnets","Two zigzag phases found in SU(3) Hubbard model"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the ordering of the competing states' energies, computed at finite numerical accuracy (a bond dimension of 24) with gaps as small as about 0.001–0.006 $t$ per site, stays the same when the accuracy is increased without limit.","fun_headline_variants_meta":{"raw":{"variants":["Zigzag antiferromagnets predicted in SU(3) Hubbard model","Zigzag magnetic order emerges in SU(3) Hubbard model","SU(3) Hubbard model reveals zigzag antiferromagnets","Two zigzag phases found in SU(3) Hubbard model"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000684,"raw_usage":{"total_tokens":3107,"prompt_tokens":954,"completion_tokens":2153,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":570,"completion_tokens_details":{"reasoning_tokens":2075}},"tokens_in":570,"tokens_out":2153,"duration_ms":14798,"temperature":1.0,"reasoning_tokens":2075,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:48:03.701714+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"An independent calculation—say, an unbiased auxiliary-field quantum Monte Carlo simulation or an iPEPS run at bond dimension beyond 24—that finds the 4-zigzag state lowest in energy anywhere in $U/t\\in[6,8]$, or finds no interval in which the 3-zigzag is the lowest state, would refute the claimed phase diagram.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the prior Hartree-Fock and constrained-path auxiliary-field quantum Monte Carlo study that reported the same zigzag phases and the transition values (8.3 and 7.7) this work compares against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes three-sublattice ordering in the SU(3) Heisenberg model, the strong-coupling limit the 3-SL phase reproduces."},{"cited_title":"Bauer, P","cited_arxiv_id":null,"evidence_quote":"Provides the square-lattice SU(3) Heisenberg reference for the 3-sublattice state and the competing two-color benchmark used at large $U/t$."},{"cited_title":"Ibarra-Garc´ ıa-Padilla, C","cited_arxiv_id":null,"evidence_quote":"Gives determinant quantum Monte Carlo results for the metal-insulator transition and magnetism used as comparison for the low-$U/t$ behavior."}],"review_version":2}