{"id":"f4ed1c26-44ab-4360-94b7-83b8351c53c4","arxiv_id":"2506.04663","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"A two-step penalty and post-processing scheme cuts the gate complexity of non-variational spin-adapted ground state preparation from quartic to quadratic scaling for spin-rotationally symmetric Hamiltonians.","lead":"This paper shows how to prepare spin-adapted ground states on quantum computers with far fewer penalty terms than the standard approach. The trick is to prepare a state with maximum spin projection first, then rotate and project it to the desired spin component, reducing the penalty cost from quartic to quadratic scaling.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed O(n^2) gate reduction is per Trotter step; no bound is given on how penalty coefficients must scale with system size, so the total gate count advantage is unproven.","rationale":"The reader's weakest_assumption already identified the unquantified penalty-coefficient assumption as a source of conditionality. Our stress-test sharpens this: the linear penalty produces a much smaller spectral gap between spin sectors than the quartic penalty for the same CS, by a factor O(s*^2). This means the required CS for the linear penalty is likely larger, and the resulting Trotter step count may offset part of the per-step gate reduction. The paper gives no analysis of this trade-off, and Fig. 6 deliberately isolates a single Trotter step, so the advertised 'significant reduction in gate complexity' is not yet demonstrated for the full algorithm. We do not see an internal inconsistency or a fatal flaw: the mathematical core (condition (16), the two-step procedure, and the numerical demonstrations) is sound, and the term-count reduction is real. The missing piece is a resource model that accounts for step-count scaling. The reader's CONDITIONAL verdict is appropriate; we recommend no change. A concrete end-to-end resource estimate for a gapless model (the Heisenberg ring) would settle whether the asymptotic advantage survives the penalty-coefficient requirement.","tokens_in":16723,"tokens_out":14370,"duration_ms":179236,"concrete_test":"Estimate end-to-end resources for the spin-1/2 Heisenberg ring (n = 6, 8, ..., 20) preparing the s*=1 state via PITE with the linear penalty. For each n, choose minimal CS (Cz = CS(2s*+1)) so the gap between the s=1 and s=0 sectors of Hproblem equals a fixed ε (e.g., 0.1J); then choose the largest Δt meeting a fixed Trotter error budget. Compare total gates (steps × gates/step) with the quartic penalty under the same criteria. If the total-gate ratio scales as O(n^2), the headline holds; if it flattens or reverses, the claimed speedup is not end-to-end.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central complexity claim rests on the number of terms in the penalty Hamiltonian, and Fig. 6 reports transpiled gate counts for a single Trotter step at fixed CS and Δt. However, the total cost of ATE/PITE is (number of Trotter steps) × (gates per step), and the step count grows with the spectral norm of the total Hamiltonian. For the linear penalty H'_penalty in Eq. (12), the energy separation between the target spin sector s* and neighboring sectors is only CS (derived from Eq. (15) with Cz ≈ CS(2s*+1)), whereas the quartic penalty (Eq. (7)) gives a separation ~4s*^2 CS for the same CS. Thus, to dominate H_system, the linear penalty may require coefficients larger by O(s*^2), and if the spin gap of H_system closes with system size (e.g., the gapless spin-1/2 Heisenberg chain), CS must grow with n. The paper explicitly acknowledges the trade-off between large CS and small Δt but provides no quantitative bound (Sec. III A). Consequently, the asymptotic O(n^2) advantage in Fig. 6 is only for a single step; the end-to-end gate complexity is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-step method for preparing spin-adapted ground states of spin-rotationally symmetric Hamiltonians with non-variational algorithms such as ATE and PITE. In the first step, the usual quartic penalty in (S^2 - s*(s*+1))^2 is replaced by a linear penalty CS(S^2 - s*(s*+1)) - Cz(Sz - s*), which under the condition of Eq. (16) selects the desired total-spin sector s* while working in the maximal-Sz subspace. This reduces the number of swap-operator terms in the penalty Hamiltonian from O(n_spin^4) to O(n_spin^2) per Trotter step. In the second step, a global y-rotation and a Hamming-weight projection are used to reach the desired Sz value. Numerical experiments on spin-1/2 Heisenberg rings and a manganese trimer show convergence for both ATE and PITE, and Fig. 6 reports transpiled gate counts for a single time-evolution operator showing the O(n^2) versus O(n^4) scaling per step.","tokens_in":17022,"tokens_out":9364,"duration_ms":102790,"significance":"If the claimed gate-complexity reduction holds end-to-end, the method would be a practically useful tool for spin-adapted state preparation in early fault-tolerant quantum simulation, particularly for spin-rotationally symmetric systems in condensed matter and quantum chemistry. The paper's analytic derivation of the penalty condition (Eq. 16), the Wigner-weight analysis for the rotation step, and the explicit numerical comparison against the quartic penalty are strengths. The authors also give a concrete circuit for Hamming-weight projection and provide numerical evidence that the modified penalty does not degrade convergence in the tested cases. However, the central complexity claim is only a per-step gate count; the end-to-end scaling, including the required growth of the penalty coefficient CS and the associated Trotter step count, is not established.","major_comments":[{"comment":"The central asymptotic claim is a per-step gate-count reduction, not an end-to-end complexity reduction. For the linear penalty in Eq. (12) with the choice Cz = CS(2s*+1) used throughout the numerics, Eq. (15) gives an energy separation of only CS between the target spin sector s* and the neighboring sectors s*±1; the quartic penalty in Eq. (7) gives a separation of order s*^2 CS for the same CS. To make the penalty dominate H_system, CS for the linear penalty may therefore need to be larger by a factor O(s*^2), and because the permissible Trotter step size decreases with the norm of the total Hamiltonian, the number of steps grows correspondingly. With s* = O(n_spin), the total gate count of the proposed method becomes O(n_spin^4), the same scaling as the naive quartic approach. Section III A only states qualitatively that CS must be large enough and Δt small enough; no bound is derived, so the claimed quadratic speedup is not established.","section":"Sec. III A, Eq. (15)"},{"comment":"Fig. 6 reports transpiled gate counts and depths for a single time-evolution operator exp(-i CS H'_S s Δt) at fixed CS=7.5 and Δt=0.015. The total cost of ATE and PITE is (number of Trotter steps) × (gates per step), and the required number of steps depends on the spectral gap of H_problem and on the norm of the penalty. No finite-size scaling of the step count, the total evolution time T, or the required CS is provided; the numerical demonstration is limited to n_spin=6 for the Heisenberg ring and to the fixed Mn trimer. Consequently, the numerical experiments show convergence for the tested cases but do not substantiate the asymptotic O(n^2) end-to-end gate-complexity claim.","section":"Sec. IV, Fig. 6"}],"minor_comments":[{"comment":"The title contains a typo: 'Heidenberg ring model' should be 'Heisenberg ring model'.","section":"Sec. IV A"},{"comment":"The phrase 'post processing' should be hyphenated as 'post-processing' for consistency with the rest of the text.","section":"Sec. III B"},{"comment":"The O(n^4) baseline is specifically for the swap-operator/Trotterized implementation of Eq. (11); block-encoding or LCU implementations of the same quartic penalty can have different scaling, so the comparison should be stated as a Trotterized-baseline comparison.","section":"Sec. II B"},{"comment":"The Wigner d-matrix element conventionally includes a sign factor (-1)^{s*-s*_z}; omitting it does not affect the squared weight but should be corrected for standard form.","section":"Eq. (18)"},{"comment":"The text says the binary-encoding modification of the Hamming-weight projection is 'shown in Appendix A', but Appendix A only provides the encoded spin operators; the modified projection circuit is not given.","section":"Sec. III B and Appendix A"},{"comment":"The notation switches between S, Sz, and s*, s*_z; please unify the symbols for total spin and its z-component.","section":"Fig. 3"},{"comment":"The statement that CS and Cz 'must be appropriately optimized' is not accompanied by any heuristic or numerical procedure for choosing them; a specific guideline would strengthen reproducibility.","section":"Sec. III A"}],"recommendation":"major_revision","confidential_remarks":"The core issue is the gap between the per-step gate-count reduction and an end-to-end complexity claim. The paper would be acceptable after the authors either provide a quantitative analysis of how CS and the number of Trotter steps scale with system size, or explicitly and consistently restrict the claims to per-step gate counts. The numerical experiments are small but clearly reported; no concerns about circularity or attribution were found."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe genuinely new idea here is the two-step penalty: first prepare the spin-magnitude-adapted ground state at maximal Sz using a linear penalty (S^2 - s*(s*+1)) - Cz(Sz - s*), then rotate and Hamming-weight project to the desired Sz. This eliminates the quartic term (S^2 - s*(s*+1))^2 and cuts the penalty Hamiltonian from O(n^4) to O(n^2) terms. The angular momentum algebra in Eqs. (12)-(16) is correct, the condition on Cz/CS is properly derived, and the numerics on the Heisenberg ring and the Mn trimer support the method. The gate-count scaling in Fig. 6 also fits nicely: count ~ n^2.1 versus n^4.6 for the quartic penalty.\n\nThe stress-test note holds up on reading. The O(n^2) advantage is per Trotter step, not end-to-end. The linear penalty gives a gap of only CS between adjacent spin sectors, while the quartic penalty gives ~4(s*+1)^2 CS. To get the same leakage suppression, the linear penalty needs CS larger by O(s*^2), and the Trotter step must shrink as the Hamiltonian norm grows. The paper explicitly acknowledges the CS vs dt trade-off in Sec. IIIA but gives no quantitative bound. So the 'quadratic speedup' claim in the abstract and introduction is not yet established for total gate count. The numerical comparisons in Fig. 5 also use different CS values for the two penalties, so they don't equalize spectral gaps. For small s*, which is often the relevant case, the overhead may be modest, but that is not shown asymptotically. The absence of code is a minor issue; the math is standard enough to reproduce.\n\nWho benefits? Anyone estimating resources for spin-adapted ground-state preparation in fault-tolerant simulation. It is a solid incremental contribution with a clean core idea, and the self-citations are used as tools, not as circular support. A serious referee should be engaged; the per-step versus end-to-end complexity point needs to be addressed in revision.\n\nRecommendation: send to peer review, but expect the authors to clarify the scaling of penalty coefficients and provide an end-to-end complexity argument.","headline":"A clean two-step penalty decomposition gives a real reduction in penalty terms from O(n^4) to O(n^2) per Trotter step, but the end-to-end speedup is not yet proven because the linear penalty needs larger coefficients and the paper gives no quantitative bound.","tokens_in":17490,"tokens_out":4745,"would_cite":true,"duration_ms":55234,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","81V70"],"pacs":[],"model":"deepseek-v4-flash","headline":"A two-step penalty scheme prepares the spin-adapted ground state of rotationally symmetric Hamiltonians with quadratically fewer gates than the standard quartic penalty, and the paper verifies it on Heisenberg rings and a manganese trimer.","keywords":["spin-adapted ground state","non-variational quantum algorithm","penalty Hamiltonian","probabilistic imaginary-time evolution","adiabatic time evolution","Wigner D-matrix","Hamming-weight projection","Heisenberg model"],"falsifier":"Run the two-step PITE on a small Heisenberg ring with a perturbation that breaks spin-rotational symmetry, such as $\\varepsilon \\sum_i (-1)^i S_i^z$; if the fidelity of the prepared state to the exact spin-adapted ground state worsens at a rate linear in $\\varepsilon$, the central symmetry assumption is falsified.","tokens_in":16589,"feed_emoji":"⚛️","tokens_out":13893,"duration_ms":102209,"temperature":0.7,"pith_summary":"This paper proposes a two-step procedure that prepares the spin-adapted ground state (the lowest energy state of fixed total spin $s^*$ and spin projection $s_z^*$) with non-variational quantum algorithms at quadratically lower gate cost. Step one replaces the quartic penalty $(\\hat{S}^2 - s^*(s^*+1))^2$, which naively expands into $O(n_{\\mathrm{spin}}^4)$ terms, with a linear penalty $C_S(\\hat{S}^2 - s^*(s^*+1)) - C_z(\\hat{S}_z - s^*)$ that picks out the maximal-$S_z$ member of the target spin multiplet using only $O(n_{\\mathrm{spin}}^2)$ terms. Step two rotates the state by a global $y$-rotation to the desired $S_z$ and projects onto the correct Hamming weight with a circuit of depth $O(\\log^2 n)$. If correct, this removes the main gate-complexity gap between variational and non-variational spin-adapted state preparation for spin-rotationally symmetric Hamiltonians. The numerical experiments on Heisenberg rings and a manganese trimer support the practical usefulness of the scheme.","feed_headline":"Two-step scheme slashes spin-state penalty gate cost","feed_subtitle":"Non-variational ground-state preparation gets a quadratic gate reduction for spin-rotationally symmetric Hamiltonians.","key_machinery":"The load-bearing object is the modified penalty Hamiltonian $H'_{\\mathrm{penalty}} = C_S(\\hat{S}^2 - s^*(s^*+1)) - C_z(\\hat{S}_z - s^*)$, whose eigenvalue on a $|s, s_z\\rangle$ state is $C_S(s(s+1)-s^*(s^*+1)) - C_z(s_z - s^*)$; minimizing over $s_z$ at $s_z = s$ turns the level spacing in $s$ into a linear function, so the inequality $2s^* < C_z/C_S < 2(s^*+1)$ makes the $s = s^*$ level the global minimum. This avoids the squared form $(\\hat{S}^2 - s^*(s^*+1))^2$, whose explicit expansion in swap operators $U^S_{ij}$ contains all pairwise products and hence $O(n_{\\mathrm{spin}}^4)$ terms. The second stage uses the global rotation $U_y(\\theta) = e^{-i\\theta\\hat{S}_y}$, whose Wigner D-matrix element $d^{s^*}_{s_z^* s^*}(\\theta)$ gives the weight of the desired $S_z$ component, and the optimal angle maximises that weight; the subsequent Hamming-weight projection is implemented by an $O(\\log^2 n)$-depth circuit in the encoding where each spin is a qubit. The same machinery works for higher spin via binary encoding, at the price of a modified projection circuit.","core_discovery":"For Hamiltonians that commute with total spin, the authors establish that the spin-adapted ground state can be prepared in two stages: first drive a non-variational evolution under $H_{\\mathrm{system}} + C_S(\\hat{S}^2 - s^*(s^*+1)) - C_z(\\hat{S}_z - s^*)$ toward the state with maximal $S_z$ in the $s^*$ sector, provided the coefficient ratio satisfies $2s^* < C_z/C_S < 2(s^*+1)$; then apply a global rotation about the $y$-axis through $\\theta_{\\mathrm{opt}} = 2\\arcsin\\sqrt{(s^* - s_z^*)/2s^*}$ and post-select on the computational-basis subspace of the desired Hamming weight. The first stage replaces the squared spin penalty, whose expansion in swap operators contains $O(n_{\\mathrm{spin}}^4)$ terms, with a linear combination containing only $O(n_{\\mathrm{spin}}^2)$ terms, and the second stage costs only $O(\\log^2 n)$ circuit depth with worst-case success probability $O(S^{-1/2})$. The claim is that this two-step procedure reaches the exact spin-adapted ground state of any spin-rotationally symmetric Hamiltonian, and the paper verifies it by numerical simulation of the adiabatic and probabilistic imaginary-time variants on spin-1/2 Heisenberg rings and the Mn(II)2Mn(III) trimer, including for excited spin sectors.","pith_inferences":["The same 'prepare an extremal-weight state with a linear penalty, then rotate and project' pattern should generalize to any symmetry group whose target multiplet has a degenerate extremal-weight state, such as particle-number or orbital-angular-momentum sectors, not just spin.","Because the first step pins $S_z$ to its maximal value, the post-selection step is the only probabilistic part; amplitude amplification on the Hamming-weight projection could lift the $O(S^{-1/2})$ success probability to near unity without changing the gate scaling.","A hybrid route not explored in the paper would use the linear-penalty Hamiltonian directly in a variational imaginary-time ansatz, reducing both measurement overhead and circuit depth relative to current variational penalty approaches.","The fitted exponents (gate count $\\propto n_{\\mathrm{spin}}^{2.07}$ for the new penalty versus $n_{\\mathrm{spin}}^{4.61}$ for the quartic one) suggest the asymptotic claims survive at practical system sizes, but the constant overhead of the second-stage projection will dominate on near-term hardware and is the natural target for the next benchmark."],"forward_implications":["With the linear-penalty first step, probabilistic imaginary-time evolution and adiabatic time evolution gain a quadratic reduction in the number of penalty gates for spin-rotationally symmetric Hamiltonians, closing the gate-count gap with variational methods that only needed $O(n_{\\mathrm{spin}}^2)$ measurements.","The two-step recipe is not limited to spin-1/2 Heisenberg models: the paper demonstrates it on a manganese trimer with $S = 5/2$ and $S = 2$ ions, and notes it extends to first-quantized Hamiltonians and to certain excited spin states.","The post-processing step's success probability scales as $O(S^{-1/2})$ in the worst case, so the method is most efficient for the low-spin sectors that are usually of interest in condensed-matter and quantum-chemistry problems.","The coefficient condition $2s^* < C_z/C_S < 2(s^*+1)$ gives a concrete tuning rule for the penalty strengths, subject to the requirement that they be large enough to dominate $H_{\\mathrm{system}}$ but not so large that they force impractically small Trotter steps."],"supporting_citations":[{"why":"Supplies the single-ancilla probabilistic imaginary-time evolution scheme used in the numerical experiments.","marker":"[18]"},{"why":"Introduces the adiabatic time evolution method used as one of the two non-variational algorithms.","marker":"[14]"},{"why":"Prior non-variational spin-adapted penalty approach whose quartic gate count is the baseline the paper improves on.","marker":"[27]"},{"why":"Variational penalty method with $O(n_{\\mathrm{spin}}^2)$ measurement cost, the reference point for the comparison.","marker":"[29]"},{"why":"Shows that the $\\hat{S}^2$ penalty expectation value can be evaluated with $O(n_{\\mathrm{spin}}^2)$ measurements in variational settings.","marker":"[55]"},{"why":"Source for the algebraic expansion of the squared spin penalty used to count its quartic number of terms.","marker":"[56]"},{"why":"The Wigner D-matrix formula from which the optimal rotation angle and the success-weight expression are derived.","marker":"[63]"},{"why":"The Hamming-weight projection circuit used in the post-processing stage.","marker":"[64]"},{"why":"Transpilation tools used to produce the gate-count scaling data in the complexity plot.","marker":"[68]"}],"fun_headline_variants":["Two-step non-variational trick cuts spin penalty cost quadratically","Spin-adapted ground states prepared faster via two-step non-variational route","Quadratic gate reduction for spin-adapted ground state prep","Non-variational method slashes spin-state penalty gates from 4th to 2nd power","Two-stage algorithm speeds spin-adapted ground state preparation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The system Hamiltonian must commute with total spin, so that every member of the target spin multiplet is exactly degenerate and the rotation-and-projection post-processing cannot leak into states with different total spin.","fun_headline_variants_meta":{"raw":{"variants":["Two-step non-variational trick cuts spin penalty cost quadratically","Spin-adapted ground states prepared faster via two-step non-variational route","Quadratic gate reduction for spin-adapted ground state prep","Non-variational method slashes spin-state penalty gates from 4th to 2nd power","Two-stage algorithm speeds spin-adapted ground state preparation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00025,"raw_usage":{"total_tokens":1638,"prompt_tokens":1111,"completion_tokens":527,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":727,"completion_tokens_details":{"reasoning_tokens":431}},"tokens_in":727,"tokens_out":527,"duration_ms":5636,"temperature":1.0,"reasoning_tokens":431,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T10:37:36.540989+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the two-step PITE on a small Heisenberg ring with a perturbation that breaks spin-rotational symmetry, such as $\\varepsilon \\sum_i (-1)^i S_i^z$; if the fidelity of the prepared state to the exact spin-adapted ground state worsens at a rate linear in $\\varepsilon$, the central symmetry assumption is falsified.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the single-ancilla probabilistic imaginary-time evolution scheme used in the numerical experiments."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the adiabatic time evolution method used as one of the two non-variational algorithms."},{"cited_title":"Keller and M","cited_arxiv_id":null,"evidence_quote":"Prior non-variational spin-adapted penalty approach whose quartic gate count is the baseline the paper improves on."},{"cited_title":"Pauncz, Spin eigenfunctions: construction and use (Springer Science & Business Media, 2012)","cited_arxiv_id":null,"evidence_quote":"Variational penalty method with $O(n_{\\mathrm{spin}}^2)$ measurement cost, the reference point for the comparison."},{"cited_title":"McArdle, T","cited_arxiv_id":null,"evidence_quote":"Shows that the $\\hat{S}^2$ penalty expectation value can be evaluated with $O(n_{\\mathrm{spin}}^2)$ measurements in variational settings."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Source for the algebraic expansion of the squared spin penalty used to count its quartic number of terms."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Wigner D-matrix formula from which the optimal rotation angle and the success-weight expression are derived."},{"cited_title":"Moudgalya, N","cited_arxiv_id":null,"evidence_quote":"The Hamming-weight projection circuit used in the post-processing stage."},{"cited_title":"Wigner, Pure and applied physics (2012)","cited_arxiv_id":null,"evidence_quote":"Transpilation tools used to produce the gate-count scaling data in the complexity plot."}],"review_version":1}