{"id":"11c1b0e9-9d55-4c22-b581-f5bbc9f52981","arxiv_id":"2501.00413","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Fermionic CAMPS, built by combining Clifford circuits with MPS via the Jordan-Wigner transformation, improves ground-state energy accuracy over plain MPS in benchmarks on the t-V and Hubbard models.","lead":"This paper applies the recently proposed Clifford-circuit-augmented matrix product state (CAMPS) method to fermionic lattice models by mapping them to spins with the Jordan-Wigner transformation. Benchmarks on the two-dimensional t-V model and the Hubbard model show that CAMPS reduces ground-state energy errors relative to plain MPS at the same bond dimension.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Hubbard benchmark may compare grand potentials, not fixed-N energies: μ, ⟨N⟩, and N variance are unreported, so the claimed physical-energy improvement over MPS is unverified.","rationale":"The reader identified the right vulnerability: the grand-canonical setup and particle-number control are not verified. My reading confirms that this is the most load-bearing issue, but I would sharpen it: the Hubbard reference values in Table I appear to be grand potentials per site, not canonical internal energies, so the problem may be even more direct than 'comparing grand-canonical CAMPS to canonical DMRG.' The paper explicitly states (in the CAMPS-for-fermions section and the Conclusion) that U(1) is not imposed and all calculations are grand-canonical, yet it does not report μ, ⟨N⟩, or N fluctuations, and it does not clarify whether the plotted energies are E or Ω. Without that information, the relative-error ratios for the Hubbard model cannot be interpreted as errors in the physical ground-state energy at fixed doping. For the t−V model at half-filling the issue is milder because μ=0 and particle-hole symmetry pins ⟨N⟩, but the Hubbard results are part of the central claim. I agree with the reader's CONDITIONAL verdict; the concern is not a demonstrated falsification but a missing verification that determines whether the headline claim transfers to fixed-particle-number physics. Reproducibility is also hindered by the absence of code and data, but the ensemble ambiguity is the primary technical gap.","tokens_in":9495,"tokens_out":21506,"duration_ms":237679,"concrete_test":"Recompute the Hubbard benchmarks for one geometry (e.g., 4×8) at D=100 and D=1000, reporting μ, the achieved ⟨N⟩, the variance ⟨(N−⟨N⟩)^2⟩, and the physical internal energy ⟨H⟩ = ⟨H_μ⟩ + μ⟨N⟩ for both CAMPS and MPS. Compare ⟨H⟩ against a canonical DMRG reference at exactly N=28 (the 1/8-hole-doped value). If the MPS/CAMPS error ratio in ⟨H⟩ remains close to the published ratio, with ⟨N⟩≈28 and small variance, the grand-canonical concern does not land. If the ratio drops or ⟨N⟩ deviates, the headline accuracy claim needs qualification.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central accuracy claim rests on the relative-error comparison in Figs. 2-4 against the Table I reference energies. For the Hubbard model this comparison is not well-defined as stated. The text says (after Eq. 3 and again in the Conclusion) that U(1) symmetry is not imposed and all calculations are grand-canonical, with μ tuned to target 1/8 doping. Yet Table I labels the Hubbard entries as 'ground state energy' without stating whether these are canonical internal energies E, grand potentials Ω = E − μN, per-site values, or total values. The magnitudes (≈ −1.7 to −2.1 for 32-site systems) are not consistent with physical internal energies of the doped Hubbard model at U=8, suggesting they are Ω per site. If the plotted 'relative error' is an error in Ω at some μ while the claim concerns ground-state energy at fixed doping, the benchmark can overstate accuracy: a variational state with particle-number fluctuations can have more negative Ω without having a more accurate physical energy at the target N. Neither μ nor the achieved ⟨N⟩ or ⟨(N−⟨N⟩)^2⟩ is reported for any bond dimension, so one cannot verify that CAMPS and MPS states are compared at the same chemical potential, at the same average particle number, and against a reference of the same quantity. Because the entire improvement ratio is computed from these energy differences, this missing verification is the load-bearing weakness.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript extends the Clifford-augmented matrix product state (CAMPS) method to fermionic systems by mapping fermions to spins via the Jordan-Wigner transformation. It benchmarks the resulting method on the spinless t-V model at half-filling on 6x6 and 8x8 lattices and on the Hubbard model at 1/8 hole doping on 1x32, 2x16, and 4x8 lattices, using large-bond-dimension DMRG energies as references. The reported results show that CAMPS reduces the relative energy error relative to plain MPS at the same bond dimension, with the improvement growing with interaction strength, and that the entanglement entropy of the MPS part is reduced. The central claim is that fermionic CAMPS significantly improves accuracy over MPS and may become a useful tool for strongly correlated fermion systems.","tokens_in":9765,"tokens_out":5328,"duration_ms":61277,"significance":"If the accuracy claim survives scrutiny, this is a useful and timely extension of an already published method: the Clifford augmentation strategy is established, and a fermionic version broadens its applicability to Hubbard-like models. The paper is clearly written and the benchmark data are extensive, including several system sizes and interaction strengths. The main weakness is that the comparison with the DMRG references is made in the grand-canonical ensemble while the references appear to be canonical fixed-particle-number energies; the manuscript does not report the chemical potentials, achieved particle numbers, or particle-number fluctuations. Because the improvement ratio is computed from precisely these energy differences, the quantitative claim is not yet fully supported.","major_comments":[{"comment":"The central benchmark is not well-defined as stated. The text says, after Eq. (3), \"Since U(1) symmetry is not imposed in the Clifford circuits, a chemical potential term is included, which is tuned to target the desired filling factor,\" and the Conclusion repeats that \"all the calculations in this work are in the grand-canonical ensemble.\" Table I, however, labels the Hubbard reference values as \"ground state energy\" without specifying whether they are internal energies E, grand potentials Omega = E - mu N, total values, or per-site values. The relative energy errors in Fig. 4 are computed against this reference. A grand-canonical variational state can lower Omega through particle-number fluctuations without having a more accurate fixed-N physical energy, so the claimed improvement ratio may be an artifact of comparing a variational Omega with a canonical E. To support the central claim, the authors must report, for every data point, the chemical potential mu used, the achieved average particle number <N>, and the variance <(N-<N>)^2>, and either impose U(1) symmetry in the variational state or compare against a grand-canonical DMRG reference at the same mu.","section":"Eq. (3), Hubbard model section, Conclusion"},{"comment":"The same grand-canonical issue affects the t-V benchmarks. The Hamiltonian in Eq. (5) has no chemical potential term and the reference energies in Table I are for half-filling, but the CAMPS calculations are performed without imposing U(1) symmetry. Even if mu=0 is used, the optimized CAMPS state can have particle-number fluctuations around half-filling, and the reference is a canonical half-filled DMRG state. The manuscript does not report <N> or its variance for the t-V calculations. Since the improvement ratio in Figs. 2-3 is defined as (E_MPS - E_ref)/(E_CAMPS - E_ref), any lowering of the CAMPS energy due to grand-canonical number fluctuations will directly inflate the reported ratio. The authors should either report the achieved particle-number statistics or justify that half-filling and mu=0 preclude a bias.","section":"t-V model section and Figs. 2-3"},{"comment":"The quantitative claim of a factor-of-four improvement rests on differences of very small energies, but no error bars or reference uncertainties are reported. For the largest bond dimensions the plotted relative errors approach 10^-6, and the reference energies in Table I are obtained by extrapolation in truncation error. The manuscript states that the listed significant digits are checked by extrapolation but does not give the extrapolation uncertainty or explain how it compares with the plotted variational errors. Without this information, the reader cannot judge whether the reported ratios are statistically meaningful, especially at large D where E_MPS - E_ref and E_CAMPS - E_ref are both close to the reference precision.","section":"Figs. 2-4 and Table I"},{"comment":"The target doping is 1/8 hole doping, but because the calculation is grand-canonical and U(1) symmetry is not imposed, the filling is controlled only indirectly through mu. For finite systems with open boundary conditions, the achieved average filling can differ from the target, and the difference can vary with bond dimension as the variational state improves. The manuscript does not report the achieved filling for any bond dimension. The authors should provide a table or plot of <N> versus D for each Hubbard system and explain how closely the target 1/8 doping is reached in both MPS and CAMPS calculations.","section":"Hubbard model section and Fig. 4"}],"minor_comments":[{"comment":"There is a typo: \"calcualtions\" should be \"calculations,\" and the phrase \"spin one\" should be \"spin system\" or \"spin-1/2 system.\"","section":"Conclusion"},{"comment":"Please state explicitly whether the listed reference energies are total energies or per-site energies, and whether they are internal energies or grand potentials. The magnitudes of the Hubbard entries—around -1.7 to -2.1 for 32-site systems—suggest per-site values, but the text calls them \"ground state energy\" without qualification.","section":"Table I"},{"comment":"The relative energy error is defined only in the captions of Figs. 2-4, and the axis label \"Relative error (Energy)\" is ambiguous. The definition should be given in the main text, and the figure axes should specify which energy (E or Omega) is used.","section":"Figure captions"},{"comment":"For reproducibility, it would be helpful to provide the explicit Jordan-Wigner transformed Hamiltonians for the 2D snake-like mapping used here, or at least to state that they are generated automatically from Eq. (2). The current text says the transformations are \"readily obtained,\" but a concrete example for a nearest-neighbor hopping term on a 2D lattice would remove ambiguity about boundary terms and string operators.","section":"Section 2 and Eq. (3)"}],"recommendation":"major_revision","confidential_remarks":"The novelty is incremental relative to the authors' earlier CAMPS papers (Refs. [21,22,27]), but the fermionic extension is a natural and useful step. The key technical risk is the grand-canonical versus canonical mismatch in the benchmarks; if the authors can report mu, <N>, and N-variance and show that the comparison is legitimate, the paper would be suitable for publication. I would also ask the editor to ensure that the reference energies in Table I are accompanied by a clear statement of what quantity they represent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Zhi—quick take on the CAMPS-fermion paper. The genuinely new thing here is the benchmark data: applying the authors' own CAMPS ansatz to fermionic models through a JW transformation, with t-V and Hubbard examples. The results consistently show CAMPS beating plain MPS at fixed bond dimension, and the improvement grows with interaction strength. That is a useful data point for anyone using DMRG-style methods in 2D. The paper is clearly written and the authors are upfront that the Clifford circuits do not impose U(1) symmetry, so they add a chemical potential.\n\nThe soft spot is the ensemble mismatch. The references in Table I are canonical fixed-N DMRG energies, while the CAMPS runs are grand-canonical with μ tuned to a target filling. The paper reports the resulting particle numbers nowhere, and it doesn't report N variance either. The stress-test note suggested the Table I magnitudes indicate the reference is actually Ω per site; I think that's a misread—the values look like per-site internal energies, not grand potentials. So the note's specific claim about Ω doesn't land. But the broader worry stands: a variational state with N fluctuations can have a lower ⟨H⟩ than the fixed-N ground state at the same average filling, which would inflate the improvement ratio. This affects the Hubbard benchmarks directly and the t-V benchmarks to the extent that N fluctuations are present at half-filling. Without ⟨N⟩ and variance, the central quantitative claim is not fully verified.\n\nMinor issues: no code or data, no explicit description of the Clifford circuit layout used in the benchmarks, no error bars on the ratios. The method itself is an extension of the authors' earlier CAMPS papers, which they cite properly; that's fine.\n\nBottom line: the paper likely deserves a rigorous referee, but the ensemble issue needs to be fixed before the numbers can be trusted. I'd send it to review with a request for the missing particle-number data, and ideally a canonical implementation or a like-for-like comparison.","headline":"Useful extension of CAMPS to fermions, but the benchmark comparisons are weakened by an unverified grand-canonical versus canonical mismatch.","tokens_in":10268,"tokens_out":5505,"would_cite":true,"duration_ms":55983,"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":"Adding Clifford circuits to matrix product states cuts fermionic ground-state energy error by up to a factor of four on the models tested.","keywords":["Clifford circuits","Matrix Product States","Jordan-Wigner transformation","fermionic systems","Hubbard model","t-V model","non-stabilizerness entanglement entropy","DMRG"],"falsifier":"Re-run the benchmarks with the particle number fixed exactly in the CAMPS optimization; if the roughly fourfold error reduction for the $8\\times8$, $V=3$ $t$-$V$ model disappears or shrinks, the reported gain is at least partly due to particle-number flexibility rather than to Clifford augmentation.","tokens_in":9304,"feed_emoji":"⚛️","tokens_out":10548,"duration_ms":90883,"temperature":0.7,"pith_summary":"Matrix product states are a widely used tool for quantum many-body simulations, but their accuracy is limited by entanglement; Clifford circuits can absorb a large part of that entanglement while remaining classically simulable. This paper extends the CAMPS ansatz, an MPS dressed by Clifford circuits, to fermionic systems by first mapping fermions to spins with the Jordan-Wigner transformation, and shows that it outperforms plain MPS on the spinless $t$-$V$ model and the Hubbard model. The improvement grows with interaction strength and with bond dimension, reaching a relative-error ratio of about four for the $8\\times8$, $V=3$ case. If the claim holds, fermionic CAMPS offers a more accurate route to strongly correlated fermion ground states at fixed computational cost.","feed_headline":"Adding Clifford circuits to MPS cuts fermion errors ~4x","feed_subtitle":"The same bond dimension gives several times smaller ground-state energy error on t-V and Hubbard models.","key_machinery":"The central object is the CAMPS wavefunction $|\\mathrm{CAMPS}\\rangle = C|\\mathrm{MPS}\\rangle$, in which a Clifford circuit $C$ acts on an ordinary matrix product state. Clifford circuits are the named machinery: they are generated by gates such as Hadamard, phase, and CNOT, and by the Gottesman-Knill theorem they can be simulated efficiently on a classical computer even though they create large amounts of entanglement. The Jordan-Wigner transformation carries the fermion problem into this spin language, turning fermionic creation and annihilation operators into strings of Pauli operators, and the 2D lattice is snaked into a 1D chain for the mapping. Because Clifford conjugation sends Pauli strings to Pauli strings, the transformed observable $O' = C^\\dagger O C$ has the same structure as $O$, so measurements remain efficient; the MPS then needs to represent only the non-stabilizerness part of the entanglement. A slightly modified two-site DMRG algorithm optimizes the MPS tensors, the local Clifford gates, and the circuit layout together.","core_discovery":"The paper establishes that the entanglement that makes fermionic tensor-network simulations difficult can be split into a part that Clifford circuits absorb and a residual part that the MPS must carry. After a Jordan-Wigner mapping, the ansatz is $|\\mathrm{CAMPS}\\rangle = C|\\mathrm{MPS}\\rangle$, where $C$ is a Clifford circuit; because Clifford circuits send Pauli strings to Pauli strings, expectation values of observables stay cheap to evaluate. Benchmarking against large-bond-dimension DMRG references, the paper finds that CAMPS reduces the relative ground-state energy error compared with MPS for every system and interaction strength tested, with the reduction growing at stronger interactions and larger bond dimension. For the $8\\times8$, $V=3$ $t$-$V$ model the error ratio is about four, and the entanglement entropy carried by the MPS part is consistently lower.","pith_inferences":["A fair check of the benchmarks would be to report the mean and variance of the particle number in the converged CAMPS state, since the comparison assumes the tuned chemical potential reproduces the reference filling exactly.","If the error reduction persists when particle number is fixed exactly, fermionic CAMPS could serve as a cross-check for other methods on doped Hubbard cylinders, where ground-state energies are difficult to converge.","Switching from the Jordan-Wigner mapping to a more local fermion-to-spin mapping, such as the Bravyi-Kitaev transformation, could reduce the nonlocal string operators and may make the residual MPS entanglement even smaller."],"forward_implications":["For a fixed bond dimension, CAMPS returns lower ground-state energy errors than MPS on the $t$-$V$ and Hubbard models, so a target accuracy can be reached with a smaller bond dimension and lower computational cost.","The accuracy gain grows with interaction strength, making the method most useful in the strongly correlated regime where plain MPS struggles.","The consistent reduction of entanglement entropy in the MPS part indicates the improvement comes from restructuring the ansatz rather than from a single favorable energy estimate.","Because the Clifford dressing is compatible with time-dependent variational principle and finite-temperature extensions of CAMPS, the same fermionic mapping can be carried into time evolution and finite-temperature simulations."],"supporting_citations":[{"why":"It introduces the CAMPS ansatz and its DMRG-style optimization for spin systems, which this paper extends to fermions.","marker":"[21]"},{"why":"It proposes augmenting MPS with disentanglers, the conceptual precursor that CAMPS makes practical with Clifford circuits.","marker":"[22]"},{"why":"It defines the non-stabilizerness entanglement entropy that the MPS part of CAMPS is left to represent.","marker":"[27]"},{"why":"It supplies the Jordan-Wigner transformation that maps fermionic Hamiltonians to spin Hamiltonians, the enabling device of the paper.","marker":"[39]"},{"why":"It provides the efficient classical simulation of Clifford circuits (Gottesman-Knill) that makes augmenting MPS with Clifford layers feasible.","marker":"[24]"},{"why":"It provides DMRG/MPS as the baseline method and the source of the reference energies used in the benchmarks.","marker":"[9]"}],"fun_headline_variants":["Fermion CAMPS: Clifford boosts MPS accuracy 4x","Clifford circuits cut fermion simulation errors 4x","Fermionic CAMPS: 4x lower error with same bond dimension","CAMPS for fermions: Clifford eats entanglement, errors drop 4x","Clifford circuits augment MPS for fermions: 4x fewer errors"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The main comparison assumes that adjusting a chemical potential puts the CAMPS calculation at exactly the same number of particles as the reference DMRG calculation, so the two energies describe the same physical state.","fun_headline_variants_meta":{"raw":{"variants":["Fermion CAMPS: Clifford boosts MPS accuracy 4x","Clifford circuits cut fermion simulation errors 4x","Fermionic CAMPS: 4x lower error with same bond dimension","CAMPS for fermions: Clifford eats entanglement, errors drop 4x","Clifford circuits augment MPS for fermions: 4x fewer errors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000512,"raw_usage":{"total_tokens":2465,"prompt_tokens":894,"completion_tokens":1571,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":1473}},"tokens_in":510,"tokens_out":1571,"duration_ms":41817,"temperature":1.0,"reasoning_tokens":1473,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:51:26.429249+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the benchmarks with the particle number fixed exactly in the CAMPS optimization; if the roughly fourfold error reduction for the $8\\times8$, $V=3$ $t$-$V$ model disappears or shrinks, the reported gain is at least partly due to particle-number flexibility rather than to Clifford augmentation.","supporting_citations":[{"cited_title":"Qian and M","cited_arxiv_id":null,"evidence_quote":"It proposes augmenting MPS with disentanglers, the conceptual precursor that CAMPS makes practical with Clifford circuits."}],"review_version":1}