{"id":"2c9f451b-8dc7-439d-84da-5fe69ff58aca","arxiv_id":"2509.09248","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Cartan decomposition rewrites the time-evolution operator into fixed-depth quantum circuits, used here to compute time-domain Green's functions and spectral functions for a two-site Hubbard model and small Ising chains.","lead":"Quantum time evolution is usually expensive to simulate, but this paper uses Cartan decomposition to rewrite the evolution operator so the quantum circuit depth stays fixed no matter how long the simulation runs. The authors use the trick to compute Green's functions and spectral functions for two small model systems, matching exact results on the Hubbard model.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The fixed-depth claim depends on BFGS exactly solving a nonconvex Cartan decomposition; for TFIM no exact comparison or residual is reported, so the identity U(t)=K e^{-iht}K† is unverified for the model's headline demonstration.","rationale":"The reader identified the same weakest assumption: the entire fixed-depth construction rests on the classical BFGS optimization of Eq. (6) finding a K that exactly conjugates H into the Cartan subalgebra h. This is indeed the most load-bearing concern because Eq. (19) is the foundation of the claimed constant-depth time evolution; if the optimization fails, every subsequent Green's function and spectral function is wrong. For the Hubbard model, the exact benchmarks provide indirect evidence that the optimization succeeded, but for the TFIM no such evidence is shown. The paper's own conclusion acknowledges exponential classical scaling, but that is a scalability limitation, not a correctness gap. The concern is concrete: the paper reports no residual measure, no success statistics, and no exact comparison for the TFIM. A quick computational check—computing the residual or comparing against exact diagonalization for N=2,4,6—would settle the issue. If the residual is small, the concern is resolved; if not, the conditional verdict should be strengthened. The reader's verdict of CONDITIONAL is appropriate and our independent read does not change it. We agree with the reader's assessment and recommend no change to the verdict.","tokens_in":12411,"tokens_out":8350,"duration_ms":88213,"concrete_test":"For each TFIM chain length N∈{2,4,6}, after BFGS optimization compute (a) the Frobenius-norm residual r = ||K† H K - h||_F, with h projected onto the Cartan subalgebra; (b) the operator-norm error max_{t∈[0,35]} ||K e^{-iht} K† - e^{-iHt}||_2, or the state fidelity for several random initial states. If r is not below ~1e-8 or the operator error is not at machine precision, Eq. (19) fails for TFIM and the fixed-depth claim is unsupported. Additionally, rerun BFGS from at least 20 random starting points and report the distribution of residuals; if residuals vary substantially, the optimization does not reliably find the required K. Finally, overlay exact-diagonalization spectral functions (computable classically for N≤6) on Fig. 4 to confirm the Cartan curves.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central equality Eq. (19) is exact only if the classical BFGS optimization of Eq. (6) finds a critical point (or a high-quality approximation) of f(K)=Tr(K v K† H) on the product-of-exponentials manifold. The paper provides no convergence guarantee, no success statistics, and no residual measure ||K† H K - h||. For the two-site Hubbard model, the exact benchmarks in Figs. 1-3 indirectly validate the optimization. For the TFIM (Fig. 4), no exact spectral functions or time-domain comparisons are shown, so there is no evidence that the Cartan decomposition succeeded for the parameters used. If the optimizer terminates with a nonzero residual, U(t) is not e^{-iHt}, and the claimed 'fixed-depth for arbitrarily long times' fails; errors in U(t) propagate into all Green's function values and the spectral functions in Fig. 4. The paper's conclusion admits an exponential classical bottleneck, but that is a scalability limitation; the unresolved issue is whether the required K is actually found at all. This is especially acute because the parametrization in Eq. (9) is a fixed-order product of exponentials, which may not cover all of e^k; stationary points of f restricted to this submanifold need not satisfy [K v K†, H]=0, so even a 'successful' local optimum may fail to put H into h.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a hybrid quantum–classical algorithm for computing retarded Green's functions on near-term quantum hardware. The key idea is to use a Cartan decomposition of the Hamiltonian Lie algebra: after a classical optimization step, the authors seek a unitary K and an Abelian subalgebra element h such that H = K h K†. Time evolution then takes the fixed-depth form U(t) = K e^{-i h t} K†, independent of t. The Green's function is assembled from Hadamard-test overlaps of time-evolved excited states, and spectral functions are obtained by Fourier transformation. The method is demonstrated on the two-site Fermi–Hubbard model with exact benchmarks in Figs. 1–3, and on transverse-field Ising chains of two, four, and six sites in Fig. 4. The conclusion explicitly acknowledges that the dimension of K scales exponentially and limits the method to small systems.","tokens_in":12664,"tokens_out":8017,"duration_ms":99171,"significance":"The central algebraic reduction is valuable and largely correct: Eq. (23) reduces the retarded Green's function to two time-evolution overlaps, Eq. (37) is the correct Jordan–Wigner two-site Hubbard Hamiltonian, and the basis h in Eq. (39) is pairwise commuting, as required for a Cartan subalgebra. For the Hubbard model, the exact comparisons in Figs. 1–3 give credible evidence that the Cartan decomposition succeeded for the parameters used. If the fixed-depth identity Eq. (19) can be reliably achieved, the algorithm offers a genuine advantage over Trotter-based simulation for small NISQ-era systems, and the analytical gradients in Eqs. (14)–(18) are a useful technical contribution. The paper is less convincing for the transverse-field Ising model, where no exact benchmark or optimization residual is provided, and the physical discussion of quantum criticality in Fig. 4 rests on unvalidated curves.","major_comments":[{"comment":"The TFIM demonstration does not validate the central claim. Unlike Figs. 1–3, Fig. 4 contains only Cartan-derived spectral functions: no exact (diagonalization or Trotter) reference curves, no time-domain overlap fidelity, and no residual measure ||K†HK − h|| or BFGS success statistics are reported. Since Eq. (19) is exact only if the optimization of Eq. (6) actually finds K satisfying Eq. (7), the TFIM results cannot be distinguished from an optimization failure. Please add exact benchmarks for at least one system size and report residual norms or success statistics for all Cartan decompositions used in the paper.","section":"One-dimensional Spin Chains / Fig. 4"},{"comment":"The statement after Eq. (8) that the exponential form K = e^{Σ ia_i k_i} and the product form K = ∏ e^{ia_i k_i} are 'equivalent' is not justified, and it is load-bearing. A local minimum of f(K) over the fixed-order product submanifold Eq. (9) is only a constrained stationary point; the tangent space of that submanifold is not obviously all of k at every point, so the critical-point condition that would imply K†HK ∈ h need not hold. BFGS has no convergence guarantee for this nonconvex problem. The paper should either prove the equivalence for the relevant Lie algebras or report residual norms and multiple-random-restart success rates; otherwise the identity U(t) = K e^{-iht} K† is an assumption rather than a derived result.","section":"Method, Eqs. (6)–(9)"},{"comment":"The abstract claims 'an efficient algorithm... requires only fixed-depth quantum circuits for arbitrarily long time simulations.' The conclusion correctly states that the dimension of K scales exponentially and limits the method to roughly 10 qubits. This is a significant qualification: the fixed-depth property holds only after a successful classical optimization whose cost is exponential and whose success is, for TFIM, unverified. Please state this limitation prominently in the abstract or introduction, and separate the quantum circuit depth from the classical preprocessing cost in the claims.","section":"Abstract / Conclusion"}],"minor_comments":[{"comment":"The last basis element of k is written as Z0Y1Z2X0, which repeats Z0 and is inconsistent with the pattern of the other elements; it should almost certainly be Z0Y1Z2X3.","section":"Eq. (38)"},{"comment":"The Jordan–Wigner expression is written as Q_a^† ⊗ Z_{a−1} ⊗ ... ⊗ Z_1; the ordering of the tensor factors should be specified explicitly to avoid ambiguity about which qubit is the least significant.","section":"Eq. (24)"},{"comment":"The notation k is used both for momentum labels in the Hubbard section and for the system-size index in Eq. (45), where 'different k means different sizes.' This is confusing; use N or n for the chain length.","section":"One-dimensional Spin Chains"},{"comment":"There are several typographical and grammatical issues: 'homomrophism', 'Cartan decomposition has been emerged', and 'We use [PiPj to represent' should be corrected.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The TFIM section appears to be an extended but unvalidated demonstration; the physical discussion about quantum criticality goes beyond the evidence shown. The manuscript is a special-issue contribution with modest scope, and the central Hubbard demonstration is sound. I would ask for the missing validation and a more careful statement of the optimization assumptions before publication. The self-contained limitation paragraph in the conclusion is a positive feature, but it should be reflected in the abstract's claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, this is a real but incremental step: it ports the Cartan-decomposition fixed-depth simulator from refs. [43,44] to the Green's function setting, with the genuinely new piece being the analytic gradient for the factorized K (Eqs. 14-18). Second, the two-site Hubbard demonstration is legitimate—the exact curves in Figs. 1-3 match, and I checked that Eq. (23), Eq. (37), and the Cartan h basis in Eq. (39) are correct. The puzzle is the TFIM section: no exact comparison is shown, so the central claim that U(t)=K e^{-iht}K† holds for those systems is unsupported.\n\nThe new contributions are the factorized gradient formulas and the explicit time-domain pipeline for the retarded Green's function (Eqs. 23-26). If the Cartan decomposition can be made exact, the fixed-depth property is nice, and the analytic gradient is reusable in other Cartan-based circuits. The conclusion is also refreshingly honest about the exponential classical overhead and the 10-qubit ceiling—that limitation belongs in the abstract.\n\nThe weak spot is the optimization. The whole construction rests on BFGS finding a K such that K† H K lies in h. For the two-site Hubbard model, the exact benchmarks indirectly validate it. For the TFIM, there is no residual measure ||K†HK-h|| and no exact spectral function to compare against; the optimization might have converged to a local minimum that does not actually put H into h. The paper even claims, without proof, that the product form in Eq. (9) is equivalent to the exponential sum in Eq. (8); for non-commuting generators that is not generally true, so the parameterization may not be expressive enough. That is a load-bearing gap for the 'fixed-depth for arbitrarily long times' headline. The interpretation of Mott crossover in a 2-site cluster and quantum criticality in a 6-site chain also outruns the systems, and the typos in Eqs. (38) and (42) are minor but avoidable.\n\nThis paper deserves a serious referee because the core equations are checkable and the gradient formula is reusable. The referee should ask for a residual report and at least one exact TFIM benchmark, and a statement about the coverage of the product ansatz. With those additions, the contribution is publishable, if modest.","headline":"Genuine but incremental: analytic gradients for Cartan decomposition plus a Green's-function pipeline; the two-site Hubbard test is solid, but the TFIM results and the fixed-depth claim are unverified.","tokens_in":13269,"tokens_out":3446,"would_cite":true,"duration_ms":35432,"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":"Constant-depth circuits compute Green's functions for any time","keywords":["Green's function","Cartan decomposition","Hamiltonian simulation","fixed-depth quantum circuits","spectral function","Fermi-Hubbard model","transverse-field Ising model","near-term quantum computers"],"falsifier":"Run the Cartan-BFGS procedure on a transverse-field Ising chain with six or more sites, compute the residual norm ‖K†HK−h‖ after optimization, and compare the resulting U(t) against exact diagonalization for a long time t. If the residual is not at machine precision, or if the Green's function deviates from the exact result beyond rounding, the fixed-depth claim fails for that Hamiltonian.","tokens_in":12190,"feed_emoji":"⚛️","tokens_out":5920,"duration_ms":67891,"temperature":0.7,"pith_summary":"This paper introduces a quantum algorithm that computes real-time Green's functions using quantum circuits whose depth does not grow with simulation time. The key move is a Cartan decomposition of the Hamiltonian's Lie algebra: the time-evolution operator e^{-iHt} is rewritten as K e^{-i h t} K†, where K is a fixed unitary and h is a commuting sum of Pauli terms, so the time-dependent part factorizes into a product of single-Pauli rotations. With that factorization, the retarded Green's function is assembled from Hadamard-test overlaps of the fixed circuit acting on simple states. The authors demonstrate the construction on the two-site Fermi-Hubbard model and on transverse-field Ising chains, reproducing exact spectral functions and capturing a metal-to-Mott-insulator crossover. A sympathetic reader would care because near-term quantum devices tolerate only shallow circuits, and this construction makes long-time spectral calculations compatible with that constraint.","feed_headline":"Constant-depth circuits compute Green's functions for any time","feed_subtitle":"Cartan decomposition turns arbitrary-time evolution into a fixed gate set, bringing spectral functions to near-term devices.","key_machinery":"The Cartan decomposition of the Hamiltonian's Lie algebra, a split g=k⊕m satisfying [k,k]⊂k, [m,m]⊂k, [k,m]⊂m, together with the KHK theorem: for H∈m there exists a fixed unitary K∈e^k and a Cartan subalgebra element h∈h such that H=KhK†. This turns time evolution into U(t)=K e^{-i h t} K†, and because h is Abelian, e^{-i h t} is a product of commuting Pauli rotations, giving fixed-depth circuits for arbitrary t. The paper finds K by minimizing the Killing form f(K)=Tr(K v K† H) over factorized unitaries K=∏ e^{i θ_i k_i}, with analytical gradients derived for efficient BFGS optimization; the gradients come from the factorized form, allowing each exponential to act on v and H as a Pauli rota","core_discovery":"The central claim is that for a Hamiltonian H whose Pauli-string Lie algebra admits a Cartan decomposition with H in the noncompact part m, the real-time evolution operator is exactly U(t)=K_0 e^{-i h t} K_0†, with K_0 time-independent and h a Cartan subalgebra element. Because h is Abelian, e^{-i h t} is a product of commuting rotations around single Pauli strings, so its circuit depth is constant in t. The retarded Green's function is then obtained from ground-state overlaps of the form ⟨Ψ| c_a K_0 e^{-i h t} K_0† c_b† |Ψ⟩, evaluated by Hadamard tests, plus the conjugate term. Numerical experiments on the two-site Fermi-Hubbard model show the fixed-depth circuits reproduce the exact Green'","pith_inferences":["Beyond the paper's examples, the same fixed-depth identity applies to any retarded or advanced correlation function of the form ⟨Ψ| O_1 e^{-iHt} O_2 |Ψ⟩, since the time-evolution operator is the only part that needs U(t).","If the classical optimization of K can be made locality-aware, the approach could move past the exponential scaling the authors note for the Lie algebra dimension; a natural test is a one-dimensional chain with local interactions where a block-structured K may suffice.","The method's advantage is likely to be most visible on devices where two-qubit gate errors dominate: once K is compiled, the same circuit is reused for every time step, so calibration overhead is amortized over the whole spectral function."],"forward_implications":["For any Hamiltonian admitting the Cartan decomposition, the quantum circuit needed to evolve a state to arbitrary time has constant depth, eliminating Trotter error accumulation and variational drift over long times.","The spectral function A_k(ω) is recovered by Fourier transforming the real-time Green's function, so the method gives direct access to Hubbard bands and quasiparticle features without summing over excited states.","The construction can be reused for any initial state: once K_0 and h are fixed classically, each Green's function entry is just a Hadamard-test overlap on the same short circuit.","The demonstrated crossover from a correlated metal at U=3 to a Mott insulator at U=6 in the two-site Fermi-Hubbard model shows the fixed-depth circuits preserve enough dynamical information to distinguish phases.","The method separates the expensive classical search for K_0 from the quantum sampling, which is the natural division for near-term hardware."],"fun_headline_variants":["Cartan trick: fixed-depth circuits for any-time Green's functions","Constant-depth circuits for Green's functions, any time","No deep circuits: Cartan decomposition for Green's functions","Fixed-depth quantum circuits compute Green's functions at any time"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The algorithm's correctness rests on the classical optimizer always reaching a global minimum of f(K)=Tr(KvK†H) so that K†HK exactly lands in the Cartan subalgebra; the paper demonstrates this on its benchmark Hamiltonians but provides no guarantee, and if the optimizer stalls the time-evolution operator is not e^{-iHt} at all.","fun_headline_variants_meta":{"raw":{"variants":["Cartan trick: fixed-depth circuits for any-time Green's functions","Constant-depth circuits for Green's functions, any time","No deep circuits: Cartan decomposition for Green's functions","Fixed-depth quantum circuits compute Green's functions at any time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000521,"raw_usage":{"total_tokens":2362,"prompt_tokens":749,"completion_tokens":1613,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":1554}},"tokens_in":493,"tokens_out":1613,"duration_ms":13883,"temperature":1.0,"reasoning_tokens":1554,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T19:27:33.872185+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the Cartan-BFGS procedure on a transverse-field Ising chain with six or more sites, compute the residual norm ‖K†HK−h‖ after optimization, and compare the resulting U(t) against exact diagonalization for a long time t. If the residual is not at machine precision, or if the Green's function deviates from the exact result beyond rounding, the fixed-depth claim fails for that Hamiltonian.","supporting_citations":[],"review_version":1}