{"id":"39f9872c-5d42-4587-8bc3-6d0f831aed46","arxiv_id":"2411.18535","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A hardware-aware implementation of QETU prepares the ground state of the 2x2 Fermi-Hubbard model with over 99 percent fidelity in noiseless simulation, using a 9-qubit grid and native gates.","lead":"Scientists simulate a tiny 2x2 Fermi-Hubbard model on a nine-qubit grid using the QETU algorithm, simplifying the circuits with fermionic swap networks and testing noise tolerance. The noiseless version reaches high fidelity, but the required hardware quality is beyond today's quantum computers.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reported ground-state preparation rests on an empirically selected initial state with no scalable construction, so the 2x2 success does not establish the claimed QETU-based preparation for larger Fermi-Hubbard lattices.","rationale":"The paper is an honest engineering study of a 2x2 Hubbard toy model: it provides a concrete native-gate circuit, analyzes Trotter error in Fig. 5, shows noiseless and noisy simulations, and makes its main limitations explicit. The central construction, controlled forward and backward time evolution via fermionic swap networks and Pauli-string conjugation, is internally consistent for the 2x2 instance, and the reported 0.99 overlap is a genuine numerical demonstration. The reader's CONDITIONAL verdict is appropriate. My stress-test identifies the same weakest assumption as the reader: the initial-state overlap. I add that the filter also requires exact spectral data, namely mu, lambda_min, and lambda_max, making the toy demonstration dependent on classical pre-solution. Neither issue is hidden; both are stated in Section VIII, and the paper does not overclaim scalability. The typos in the fermionic-operator expressions for H2 and H3 are mechanical and do not propagate into the qubit Hamiltonians in Eqs. (18) and (19) that are actually implemented. Thus no verdict change is warranted, and a concrete test of initial-state behavior on a 4x4 lattice would tell whether the approach has any path beyond the toy model.","tokens_in":16981,"tokens_out":11170,"duration_ms":107281,"concrete_test":"On the 2x2 model, enumerate all 36 half-filling product states, compute gamma for each, and check whether the chosen state is near-optimal and how many states also reach gamma >= 0.091. Then move to the 4x4 model: prepare the QETU circuit with a Hartree-Fock Slater determinant and with the best computational-basis product state, compute gamma by exact diagonalization, and determine the polynomial degree d required to reach 0.99 overlap. If d increases as 1/gamma or worse while the swap-network depth grows polynomially, the total gate count for fixed fidelity grows quickly with L; if gamma itself drops exponentially with L, the method is not scalable without a separate initial-state construction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section VII initializes every circuit with |psi_init> = |1001>|--++>, selected 'empirically' to ensure gamma >= 0.09102, and Section VIII concedes that random initialization cannot guarantee this overlap and that scalable initial-state preparation is indispensable. This is the load-bearing step: QETU only amplifies the ground-state component already present in |psi_init>, so the 0.99 overlap reported in Fig. 6 is conditional on a favorable input state. For L sites in the half-filled sector, the Hilbert-space dimension grows exponentially, and no construction is supplied that would keep gamma from decaying exponentially; the empirically tuned 2x2 state does not generalize. The same section concedes that the central auxiliary qubit and the two-configuration fermionic swap network are specific to the 2x2 grid, so the 'efficiently realize ... on a 2D lattice' claim is verified only for that instance. A secondary but related limitation is that the filter parameters lambda_min, lambda_max, and mu, and hence c1 and c2, are taken from exact diagonalization, so the circuit is not self-contained as a ground-state solver.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a hardware-oriented implementation of the QETU algorithm for ground-state preparation of the 2×2 Fermi-Hubbard model. The authors construct the controlled forward and backward time-evolution operator using a fermionic swap network tailored to a 9-qubit 3×3 grid, decompose all gates into a native gate set, and benchmark the resulting circuit in noiseless and noisy simulations. The main numerical result is that, with a single second-order Trotter step and a polynomial filter of degree around 45–50, the prepared state has overlap at least 0.99 with the exact ground state in the noiseless case. A noise model with depolarizing and measurement errors is used to estimate the required hardware quality, and the authors conclude that current superconducting devices are not yet accurate enough for this protocol.","tokens_in":17219,"tokens_out":12427,"duration_ms":107248,"significance":"If taken as an implementation study of a specific 2×2 instance, the paper is a useful and honest contribution. It provides explicit circuit decompositions into a realistic native gate set, makes the code available, and gives a candid noise-resource analysis. The main value is in showing how fermionic swap networks can reduce the cost of the controlled time evolution needed by QETU on a small toy model. However, the demonstrated protocol is conditional on two pieces of information that would not be available in an actual application: an empirically chosen initial state with a guaranteed overlap γ ≥ 0.09102, and spectral parameters (μ, λ_min, λ_max) obtained from exact diagonalization. The paper also does not provide a scaling analysis for larger lattices. These caveats are partly acknowledged in Section VIII, but the abstract and concluding claims should be scoped accordingly.","major_comments":[{"comment":"The reported overlap |⟨ψ_final|ψ_0⟩|² ≥ 0.99 is obtained with the initial state |ψ_init⟩ = |1001⟩|--++⟩, which is selected empirically to guarantee γ ≥ 0.09102. Section VIII concedes that random initialization cannot guarantee this overlap and that scalable initial-state preparation is indispensable. Since QETU is an eigenspace filter, its success is entirely conditional on the initial state containing a non-negligible ground-state component; the paper provides no argument that γ does not decay exponentially with system size. The abstract and concluding claims should therefore be framed as a proof-of-principle for the 2×2 instance rather than as a general ground-state preparation method for the Fermi-Hubbard model.","section":"Section VII, Fig. 6"},{"comment":"The circuit parameters c1, c2, and μ are computed from λ_min and λ_max obtained by exact diagonalization of the Hamiltonian. Consequently, the numerical results in Figs. 6 and 7 demonstrate the performance of the circuit conditional on spectral information that would not be available in an actual application. The paper should state this limitation prominently, e.g., in the abstract, and clarify that the reported experiments validate the circuit implementation rather than the complete QETU ground-state-solving protocol including the binary search over μ.","section":"Section IV E and Section VIII"},{"comment":"The sentence 'we showed that ... we can efficiently realize a controlled forward and backward time evolution operator of the Fermi-Hubbard model on a 2D lattice' is stronger than what the paper demonstrates. The construction exploits the 2×2-specific placement of the auxiliary qubit in the center of the grid and the fact that only two fermion-to-qubit mappings are needed; the paper itself notes that these optimizations do not generalize and that finding optimal Pauli strings for larger systems is nontrivial. The claim should be restricted to the 2×2 instance, or the paper should provide a concrete resource estimate (gate count and depth scaling) for general 2D lattices.","section":"Section VIII"}],"minor_comments":[{"comment":"There are apparent typos in the spin indices and duplicated hopping terms: in Eq. (16) the term (a†_{3,↓}a_{4,↓} + a†_{3,↓}a_{4,↓}) appears twice, and in Eq. (19) 'Y_{1,↑}Y_{2,↓}Z_{1,↓}Z_{2,↓}' should presumably read 'Y_{1,↑}Y_{2,↑}Z_{1,↓}Z_{2,↓}'. These should be corrected.","section":"Eqs. (16) and (19)"},{"comment":"The text contains a typo: 'The fist part of the kinetic hopping term' should be 'The first part'.","section":"Section IV C"},{"comment":"The phrase 'prior the the measurements' should be 'prior to the measurements'.","section":"Section V"},{"comment":"The sentence 'Both circuits were initialized with the same initial state' is ambiguous; Section V describes three measurement circuits, so the text should say 'all three circuits'.","section":"Section VII"},{"comment":"The axes of the magnified inset are not clearly labeled; please specify that the horizontal axis is the polynomial degree d and the vertical axis is the estimated ground-state energy ⟨ψ0|H|ψ0⟩.","section":"Fig. 7"},{"comment":"The matrix U is written with a bare 'H' in the lower-right block, which is ambiguous; clarify that this denotes the Hadamard gate or write out the explicit matrix entries.","section":"Eq. (33)"},{"comment":"The notation 'hop_{1,3} · Z_2' is nonstandard and could be misread as a product of operators; please clarify that the resulting operator is the hopping term with a Z-string on the intermediate qubit.","section":"Appendix B, Eq. (B4)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a competent implementation study of QETU on a 2×2 Fermi-Hubbard toy model. The technical circuit construction appears sound, and the authors are transparent about several limitations in Section VIII. My main concern is the gap between the title/abstract and what is actually demonstrated: the numerical results rely on an empirically chosen initial state and on spectral parameters obtained from exact diagonalization, and the '2D lattice' claim is verified only for the 2×2 instance. I expect these can be fixed by appropriately scoping the claims and explicitly stating the conditional nature of the demonstration in the abstract. I would support publication after such a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it is the first end-to-end QETU implementation for a fermionic lattice model on a 2D grid, with a native-gate decomposition and a noise study. Second, the authors' own Discussion concedes almost every limitation a skeptic would raise, so the paper is more credible for being upfront about what it does not solve.\n\nWhat is genuinely new: the circuit simplifications for the 2x2 Fermi-Hubbard model, specifically placing the auxiliary qubit in the center of a 3x3 grid and using only two fermion-to-qubit mappings within a fermionic swap network. This lets them implement controlled time evolution with only nearest-neighbor gates, avoiding long Jordan-Wigner strings. The noiseless overlap exceeding 0.99 for polynomial degree around 45-50 is solid, and the noise study is honest: it shows that current two-qubit error rates are too high for practical ground-state energy estimation, even with post-selection. The source code is on GitHub, so the numerical experiments are reproducible.\n\nThe soft spots are real but mostly disclosed. The load-bearing assumption is the empirically chosen initial state |1001>|--++>, which guarantees overlap at least 0.091. QETU only amplifies the ground-state component already present, so without a scalable construction for a good initial state, the method does not extend to larger lattices. The stress-test note is right that the 2x2 success does not establish a general method, but the paper says this explicitly in Section VIII, so it is an acknowledged limitation rather than a hidden one. Similarly, the spectral parameters λmin, λmax, and μ come from exact diagonalization, so the circuit is not self-contained as a ground-state solver; the authors mention binary search as future work. The specific optimizations (central auxiliary qubit, two mappings) are also admitted to be toy-model-specific. Minor typos in Eqs. (15)-(16) and (19) are mechanical and do not affect the later derivation.\n\nWho is this for? Someone working on QETU implementations, early fault-tolerant resource estimation, or hardware-specific circuit optimization for fermionic simulations. It is not a physics breakthrough and does not claim to be. As an engineering contribution, it is careful and honest.\n\nRecommendation: send it to peer review. It deserves a serious referee, mainly to check the gate decompositions and the Trotter error analysis, and to suggest fixing the typos and softening the Section VIII claim about 'efficiently realize' on 2D lattices in general. But as a reproducible implementation study, it is a solid contribution.","headline":"A careful, honest QETU implementation study for a 2x2 Fermi-Hubbard toy model; the main limitations are disclosed by the authors themselves, so the result is a useful engineering benchmark rather than an overclaimed breakthrough.","tokens_in":17717,"tokens_out":2130,"would_cite":false,"duration_ms":19726,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["03.67.Ac","71.10.Fd"],"model":"deepseek-v4-flash","headline":"Fermionic swap networks let a 9-qubit grid prepare the 2x2 Fermi-Hubbard ground state with at least 0.99 overlap in noiseless simulation.","keywords":["Fermi-Hubbard model","QETU","ground-state preparation","fermionic swap networks","Trotter decomposition","2D qubit topology","noise simulation","quantum eigenvalue transformation"],"falsifier":"Run a noiseless statevector simulation of the published circuit with $u=t=1$, the chosen initial state, one second-order Trotter step, and polynomial degree $d=50$; if the final overlap is materially below 0.99, the central efficiency claim is not reproducible as stated.","tokens_in":16774,"feed_emoji":"⚛️","tokens_out":11159,"duration_ms":95797,"temperature":0.7,"pith_summary":"The paper tries to show that QETU, a recently introduced algorithm for preparing ground states without expensive block-encoding, can be implemented on realistic 2D grid hardware for the 2x2 Fermi-Hubbard model. It does this by turning the controlled forward-and-backward time evolution operator into a circuit of nearest-neighbour gates using fermionic swap networks, on a 3x3 grid of nine qubits. In noiseless statevector simulations, a single second-order Trotter step plus a polynomial filter of degree about 45 to 50 produces a state whose overlap with the exact ground state is at least 0.99. The paper also quantifies noise sensitivity, showing that hardware fidelity requirements are the limiting resource, and identifies the need for an initial state with guaranteed overlap with the ground state as the key practical bottleneck. Because the Fermi-Hubbard model is a standard testbed for strongly correlated electrons, a hardware-native route to its ground state is a concrete step toward practical quantum simulation.","feed_headline":"Fermionic swaps bring 99% ground-state fidelity to a 9-qubit grid","feed_subtitle":"One Trotter step plus a 45-50 degree filter gives the 2x2 Hubbard ground state at 99 percent overlap","key_machinery":"The central object is the oracle $V = \\mathrm{diag}(e^{i\\Delta t H}, e^{-i\\Delta t H})$, the controlled forward and backward time evolution operator that QETU needs. Instead of controlling the time evolution directly, the paper builds V from a Trotter-Suzuki decomposition and anticommuting Pauli strings K1 and K2 that flip each term's sign, so only Pauli gates are controlled by the ancilla qubit. A fermionic swap network routes the fermionic modes so that every interacting orbital becomes a nearest neighbour at some point in the circuit; the 2x2 lattice needs only two distinct mappings, and the center-placed ancilla becomes adjacent to every orbital. The filter itself is a polynomial approximation F(x) to the shifted sign function, with rotation angles found by optimization, applied through the alternating QETU circuit.","core_discovery":"Using the 2x2 Fermi-Hubbard model as a toy system, the authors claim that QETU ground-state preparation can be made efficient on a 3x3 grid of qubits. The Hamiltonian is split into an onsite interaction H1 and two hopping halves H2 and H3; two fermion-to-qubit mappings, with the ancilla qubit placed in the center of the grid, bring every required pair of orbitals into adjacency so that all gates act between neighboring qubits. The controlled forward and backward time evolution operator V is realized without controlling the Trotter factors: Pauli strings K1 and K2 that anticommute with the Hamiltonian terms flip the sign of the evolution, and a single second-order Trotter step suffices because the spectrum shift multiplies the simulation time by a small constant c1. With a polynomial approximation of the shifted sign function of degree around 45 to 50, the final state has overlap at least 0.99 with the exact ground state in noiseless simulation; the same circuit estimates the ground-state energy, and noise simulations show the hardware accuracy needed for practical use.","pith_inferences":["The center-ancilla placement is a 2x2 special case; for larger lattices the ancilla will not be adjacent to every orbital in two mappings, so the controlled Pauli strings will require additional swap rounds and the gate-count advantage may not carry over.","The same swap-network construction for V could be reused for other two-dimensional fermionic Hamiltonians whose two-body terms have the same structure, since the local decomposition only relies on iSWAP and CPhase gates.","Because the polynomial filter is independent of the hardware layout, the paper's trade-off between Trotter depth and polynomial degree can be attacked from either side; adaptive or factorized polynomial filters are natural next steps.","A practical large-scale version would need a separate, cheap initial-state preparation method, since the overlap assumption is the only ingredient in the paper that is not backed by a general construction."],"forward_implications":["For the 2x2 Fermi-Hubbard model, one second-order Trotter step is enough: the Hamiltonian spectrum shift shrinks the effective time step so the Trotter error stays around $5\\times 10^{-3}$, and higher polynomial degree rather than more Trotter steps is what raises the overlap.","The controlled time evolution operator V can be implemented with nearest-neighbour two-qubit gates on a 3x3 grid using native iSWAP and CPhase gates, so the algorithm maps directly onto currently available superconducting qubit topologies.","Ground-state energy estimation needs only three measurement circuits, one for the onsite terms and two for the hopping terms after basis rotations, which keeps the measurement overhead small.","Even with error mitigation based on fermion-number conservation and the ancilla state, today's depolarizing noise levels prevent accurate energy estimates for this toy model, meaning gate fidelity is the limiting resource.","The empirically chosen initial state $|1001\\rangle|--++\\rangle$ is necessary; random initialization does not reliably give the overlap that the filter needs."],"supporting_citations":[{"why":"Supplies the QETU algorithm, the control-free oracle V construction, and the eigenvalue-filtering theorem used to isolate the ground state.","marker":"[1]"},{"why":"Introduces fermionic swap networks, the technique the paper adapts to bring all interacting orbitals and the ancilla into adjacent positions on the 3x3 grid.","marker":"[8]"},{"why":"Provides the Jordan-Wigner transformation that maps fermionic operators to Pauli operators and produces the Z-string couplings that the swap network removes.","marker":"[9, 10]"},{"why":"Gives the symmetric second-order Trotter-Suzuki product formula used to approximate the time evolution operator with one step.","marker":"[16]"},{"why":"Provides the measurement strategy that diagonalizes (XX+YY)/2 terms so hopping energies can be extracted from three circuits.","marker":"[18]"},{"why":"Supplies the optimization-based phase-factor evaluation used to determine the symmetric rotation angles for the polynomial approximation.","marker":"[11]"},{"why":"Proposes the adaptive finer filtering scheme that the paper cites as the route to scale QETU to larger systems with increased spectral-gap demands.","marker":"[23]"},{"why":"Provides the noise simulator used for the numerical experiments that set the hardware-fidelity requirements.","marker":"[19]"}],"fun_headline_variants":["99% ground-state overlap for 2x2 Hubbard on 9-qubit grid","QETU on a grid: 2x2 Hubbard ground state with fermionic swaps","Fermionic swaps and QETU deliver 99% fidelity for Hubbard model","9 qubits, 2x2 Hubbard, 99% fidelity: QETU simplified","QETU simplifies to 9 qubits for Fermi-Hubbard ground state"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the initial state having a large enough overlap with the ground state; the authors pick $|1001\\rangle|--++\\rangle$ empirically because random initialization cannot guarantee that overlap, and they expect the requirement to become harder for larger systems.","fun_headline_variants_meta":{"raw":{"variants":["99% ground-state overlap for 2x2 Hubbard on 9-qubit grid","QETU on a grid: 2x2 Hubbard ground state with fermionic swaps","Fermionic swaps and QETU deliver 99% fidelity for Hubbard model","9 qubits, 2x2 Hubbard, 99% fidelity: QETU simplified","QETU simplifies to 9 qubits for Fermi-Hubbard ground state"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000966,"raw_usage":{"total_tokens":4125,"prompt_tokens":974,"completion_tokens":3151,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":3039}},"tokens_in":590,"tokens_out":3151,"duration_ms":20467,"temperature":1.0,"reasoning_tokens":3039,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:08:09.516529+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a noiseless statevector simulation of the published circuit with $u=t=1$, the chosen initial state, one second-order Trotter step, and polynomial degree $d=50$; if the final overlap is materially below 0.99, the central efficiency claim is not reproducible as stated.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the measurement strategy that diagonalizes (XX+YY)/2 terms so hopping energies can be extracted from three circuits."},{"cited_title":"Enhancing Scalability of Quantum Eigenvalue Transformation of Unitary Matrices for Ground State Preparation through Adaptive Finer Filtering","cited_arxiv_id":"2401.09091","evidence_quote":"Proposes the adaptive finer filtering scheme that the paper cites as the route to scale QETU to larger systems with increased spectral-gap demands."}],"review_version":1}