{"id":"f0ec1c6f-acf6-444f-98da-64b20642ff4a","arxiv_id":"2504.13174","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Differential equations are solved on a quantum computer by preparing the ground state of a Hamiltonian constructed from Chebyshev differential and data constraints, bypassing variational search.","lead":"This paper presents a quantum algorithm that encodes differential equations into the ground state of an effective Hamiltonian built from Chebyshev polynomials, then solves the equation by preparing that ground state. It avoids variational training and claims to handle linear, nonlinear, and partial differential equations using data constraints.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The nonlinear section assumes imaginary-time evolution selects the product zero-energy state |ψ_g>⊗|ψ_g>, while H has exponentially many degenerate zero-energy states; no argument or proof is given for this selection, so the NDE results are not yet justified.","rationale":"The reader's weakest_assumption identifies exactly the same load-bearing concern: the nonlinear construction lives in a doubled Hilbert space whose effective Hamiltonian has a massively degenerate zero-energy subspace, and the paper assumes without proof that ground-state preparation selects the desired product state |ψ_g>⊗|ψ_g>. I agree this is the central gap. The linear, PDE, and inhomogeneous sections are more secure because those effective Hamiltonians can have a unique (or at least well-understood) null vector, and the numerical comparisons to analytical solutions are consistent. For the nonlinear cases, however, the algorithm's correctness depends entirely on the unproven selection rule. Imaginary-time evolution is not a selection rule: it projects onto the full degenerate eigenspace, and the final state is determined by the initial state's decomposition. The paper's own statement that the task becomes 'seeking a degenerate state' confirms that the nullspace is not one-dimensional, and no symmetry or initialization argument is given to rule out entangled null states. The advertised 'provable speed-up' is also unsupported because no spectral-gap or end-to-end complexity analysis is provided, and the Discussion explicitly defers gap analysis to future work; but that is secondary to the product-state problem. Since the reader already marked the paper CONDITIONAL and this concern is the same one, no verdict change is needed. A concrete numerical test can settle whether QITE from product initial states actually converges to the product solution.","tokens_in":41118,"tokens_out":8304,"duration_ms":78740,"concrete_test":"Classically implement Section V for Eq. (29) with n=3 (and n=4): build H exactly, diagonalize it, and record the dimension d of the zero-energy eigenspace. Let |ψ_g> be the n-qubit state whose overlap reproduces the analytical solution 1-x^2/8, and form |P>=|ψ_g>⊗|ψ_g>. Run imaginary-time evolution (exact matrix exponential e^(-τH)) from the all-zero state and from 100 random product states until convergence. For each final state compute the purity of the first n-qubit reduced density matrix and the fidelity with |P>. If the typical purity is far below 1 or the fidelity is not close to 1, the product-state selection assumption fails; if all converge to |P>, the concern is settled in the paper's favor.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is in Section V: for nonlinear DEs the solution is claimed to be the product state |ψ_g>⊗|ψ_g> in the zero-energy eigenspace of the doubled-space effective Hamiltonian H, which the paper itself notes has at least 2^(2n-1) degenerate zero-energy states. The only preparation method offered is QITE (or QSVT-based imaginary-time evolution). For any Hamiltonian with a degenerate nullspace, e^(-τH) maps an initial state to the projection onto the whole nullspace, weighted by the initial-state amplitudes; it contains no mechanism that favors product over entangled zero-energy states. The paper does not show that the nullspace is one-dimensional, that H has a symmetry protecting the symmetric subspace, or that the chosen initial state has zero overlap with all entangled null vectors. The numerical demonstrations simply assume the prepared state is product and read out the overlap; they do not verify purity or productness of the prepared state. The alternative workflow in Methods Sec. 6 says 'After discarding the 2^(2n) degenerate zero-energy states, we can find the ground state', but 'discarding' is not a physical operation available to QITE. Thus the nonlinear results, including Fig. 7, are not yet justified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a quantum algorithm for solving linear and nonlinear differential equations by encoding the problem into the ground state of a positive semi-definite 'effective Hamiltonian' H constructed in a Chebyshev latent space. The Hamiltonian is formed as a sum of Gram matrices of differential operators and data constraints, and the solution is read out through overlap measurements with Chebyshev feature states. Ground state preparation is proposed via quantum imaginary-time evolution implemented with QSVT. The method is demonstrated on second-order ODEs with constant and variable coefficients, inhomogeneous ODEs, the Laplace, heat, and wave equations, and two nonlinear ODEs. The linear constructions are algebraically transparent and match the analytic solutions in the presented small-scale examples; the nonlinear construction relies on a doubled Hilbert space and on an unproven assumption that the relevant zero-energy eigenstate is the product state |ψ_g>⊗|ψ_g>.","tokens_in":41416,"tokens_out":8521,"duration_ms":81236,"significance":"If the construction were established for nonlinear equations, the approach would offer a non-variational quantum DE solver that combines compact spectral representations with ground-state preparation and function evaluation beyond grid points. The paper's strengths are the explicit construction of the Chebyshev latent-space operators, the block-encoding circuits for derivative and data-constraint matrices, and the clear demonstration that several canonical linear problems can be mapped to ground-state preparation. The manuscript also explicitly acknowledges in the Discussion that the spectral gaps of the effective Hamiltonians are problem-dependent and 'not easy to bound', which tempers the abstract's speed-up claim. However, the nonlinear extension, advertised as a central novelty, is not currently justified, and no complexity analysis or convergence data is provided. The derivations are explicit enough that the linear examples can be checked from the given matrices, but no machine-checkable proofs or reproducible code are supplied.","major_comments":[{"comment":"The central claim of the nonlinear section is that the solution of the NDE is encoded in the product state |ψ_g>⊗|ψ_g>, asserted to be the ground state of the doubled-space effective Hamiltonian H. This is not established. The text itself notes that H has at least 2^(2n−1) degenerate zero-energy states, and Methods Sec. 6 acknowledges 'discarding' the degenerate states. Quantum imaginary-time evolution with exp(−τH) maps an arbitrary initial state onto the projector onto the full zero-energy eigenspace; it contains no mechanism that favors product states over entangled zero-energy states. No symmetry argument, spectral analysis, or initialization analysis is provided. Consequently the numerical agreement in Figs. 7(b,c) only confirms that the product state is a zero-energy state, not that the algorithm prepares it, and the nonlinear results are not yet justified.","section":"Section V, text after Eq. (33)"},{"comment":"The abstract's claim of 'provable speed-up for state preparation' is unsupported. The only related statement in the paper is the generic near-optimal ground-state preparation result of Ref. [109], whose runtime depends on the spectral gap Δ, while the Discussion explicitly states that gaps of the effective Hamiltonians are problem-dependent and 'not easy to bound.' No complexity analysis of the overall algorithm, the block-encoding costs, or the gap of the constructed H is given. The phrase 'provable speed-up' should either be removed or substantiated with concrete bounds for the constructed Hamiltonians.","section":"Abstract and Discussion (fourth paragraph)"},{"comment":"The statement that H is 'a real symmetric positive-definite matrix with distinct eigenvalues' is internally inconsistent with Section V, where H is explicitly described as having at least 2^(2n−1) degenerate zero eigenvalues. Even in the linear constructions, H is positive semi-definite and can have degenerate eigenvalues whenever multiple functions satisfy the differential and invariant constraints. The sentence should be revised to acknowledge positive semi-definiteness and possible degeneracy.","section":"Core idea section, paragraph beginning 'For each latent space constrain...'"}],"minor_comments":[{"comment":"The notation changes from √η in Eq. (2) to η in Eq. (30) without explanation. Please clarify how the scaling of the doubled-space overlap relates to the normalization ⟨ψ|n|ψ⟩n = 1 and to the final rescaling factor ηe.","section":"Section V, Eq. (30) and surrounding text"},{"comment":"The phrase 'After discarding the 2^(2n) degenerate zero-energy states, we can find the ground state' is not an algorithmic instruction; no physical operation or circuit is proposed that would 'discard' a subspace. This should be rewritten to describe a concrete symmetry projection or an explicit preparation protocol.","section":"Methods, Sec. 6"},{"comment":"For Eq. (34), the regular constraint f(xs=0.5) is used to set the scale, but the value f(0.5) from the numerical solution is not given in the problem statement. The demonstration therefore relies on information about the unknown solution; the paper should state this explicitly and discuss how such data would be obtained in practice.","section":"Results, Section V, Eq. (34)"},{"comment":"The hardware demonstration prepares simple two-qubit states that are not shown to be the ground states of any effective Hamiltonian for the Legendre equation under consideration. The connection between the pulse sequence and the proposed algorithm should be clarified or the section removed.","section":"Methods, Sec. 8 (Pulser demonstration)"}],"recommendation":"major_revision","confidential_remarks":"The linear part of the paper is a solid contribution, and the construction of effective Hamiltonians in the Chebyshev basis is presented with useful detail. The nonlinear extension, however, hinges on an unproven degeneracy argument, and the abstract overstates what is established. I would encourage the editor to request a revision that either proves that imaginary-time evolution selects the product zero-energy state (for example, by showing the nullspace is contained in the symmetric subspace under the specific construction, or by adding an explicit symmetry projection) or scales back the nonlinear claims. The 'provable speed-up' phrase in the abstract should also be moderated unless concrete complexity bounds are provided."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here's my take on 2504.13174. The core construction is real: building a positive-semidefinite effective Hamiltonian from Chebyshev differentiation and data constraints, then preparing its ground state, bypasses variational training. The linear ODE and PDE demos check out against analytical solutions, the block-encoding circuits are detailed, and the neutral-atom pulse demo is a nice sanity check. The paper is clearly written and the citations are appropriate. That is worth something.\n\nThe soft spot is exactly where the reader put it. Section V assumes the doubled-space ground state of the nonlinear Hamiltonian is the product state |ψ_g>⊗|ψ_g>. But the paper itself states there are at least 2^{2n-1} degenerate zero-energy states, and e^{-τH} projects onto the entire nullspace with weights set by the initial state. No symmetry, no unique-gap argument, no overlap bound is given to show imaginary-time evolution prefers the product state. The numerical examples just pick the prepared state and read out the overlap; they never verify purity or productness. The line in Methods Sec. 6 about \"discarding the degenerate zero-energy states\" papers over the fact that QITE cannot discard anything by itself. So the nonlinear claims, including Fig. 7, are not yet justified. That is a load-bearing gap, not a minor one.\n\nTwo smaller issues. First, the abstract's \"provable speed-up for state preparation\" is not substantiated anywhere; the Discussion correctly says gap scaling is problem-dependent and left for future work, so the abstract overreaches. Second, the scaling factor η_e is a post-hoc single-point fit; that is fine for the method but it means the reported agreements are partly normalized, not independent predictions. The paper should be explicit about that.\n\nIf I were refereeing, I'd recommend major revision: prove the product-state selection (or state explicit conditions), give gap or complexity estimates, release code/data, and quantify errors. The linear+PDE machinery is likely salvageable and interesting; the nonlinear part needs real work before the paper can stand as advertised. It deserves peer review, not a desk reject.","headline":"Genuinely useful non-variational construction for linear and PDE problems, but the nonlinear section rests on an unproven product-state assumption in a degenerate nullspace and the abstract oversells a speed-up.","tokens_in":41920,"tokens_out":1784,"would_cite":false,"duration_ms":17365,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","65N35"],"pacs":["03.67.Ac"],"model":"deepseek-v4-flash","headline":"Differential equations can be solved by preparing the ground state of an effective Hamiltonian built in a Chebyshev basis.","keywords":["quantum differential equation solver","effective Hamiltonian","Chebyshev feature map","ground state preparation","quantum imaginary time evolution","quantum singular value transformation","physics-informed quantum machine learning","nonlinear differential equations"],"falsifier":"Run the effective-Hamiltonian construction for a nonlinear example such as Eq. (29) or Eq. (34) on a small register, start imaginary-time evolution from several distinct initial states (the all-zero state, a random product state, and a superposition), and check whether the prepared state always lies in the product subspace $|\\psi\\rangle\\otimes|\\psi\\rangle$ and yields the same rescaled overlap solution. If any initial state converges to an entangled zero-energy state with a different overlap profile, or if the reported solution is not reproduced, the central claim fails.","tokens_in":40916,"feed_emoji":"⚛️","tokens_out":12046,"duration_ms":95630,"temperature":0.7,"pith_summary":"This paper proposes a quantum algorithm that solves linear and nonlinear differential equations without discretizing the domain and without variational training. The idea is to rewrite the differential equation, together with boundary and data constraints, as a real symmetric positive-semidefinite operator $H$ acting on the coefficients of a Chebyshev-polynomial expansion; the equation is solved when a ground state $|\\psi_g\\rangle$ of $H$ is found with minimum eigenvalue $\\lambda_{\\min}\\approx 0$. The solution function is then read out as a rescaled overlap $f_Q(x)=\\sqrt{\\eta}\\,\\langle\\tau(x)|\\psi_g\\rangle$, which can be evaluated at any point $x$, not just on a grid. Nonlinearities are handled by doubling the Hilbert space and encoding quadratic terms as entangling operators $G_n^T\\otimes G_n^T$ inside the same ground-state framework. If the construction works as claimed, it supplies a variational-free, physics-informed route to quantum scientific computing in the early fault-tolerant era, combining the compact basis of quantum machine learning approaches with the state-preparation guarantees of quantum linear algebra methods.","feed_headline":"Differential equations solved as ground-state problems","feed_subtitle":"A Chebyshev-space Hamiltonian turns linear and nonlinear DEs into state preparation, bypassing variational search.","key_machinery":"The load-bearing object is the effective Hamiltonian $H=\\sum_i T(O_i)$ with $T(O)=O^T O$, built from latent-space representations of the differential operator and the data constraints in the Chebyshev basis. Its ingredients are the Chebyshev differentiation matrix $G_n^T$ (a constant upper-triangular matrix that turns $d/dx$ into matrix multiplication), the quantum Chebyshev feature map $|\\tau(x)\\rangle$ that labels basis polynomials and makes overlaps measurable, the constant mapping matrices $M_1,M_x,\\ldots,N_1,N_x$ that promote products $x^p\\langle\\tau(x)|$ or tensor products $\\langle\\tau(x)|\\otimes\\langle\\tau(x)|$ to a higher-dimensional Chebyshev basis, and rank-one data-constraint operators $B_n(x)$, $D_n^{(0)}(x)$ that pin boundary values or nonzero anchors. These pieces convert every term of a linear or nonlinear differential equation into an $x$-independent operator, so the equation holds exactly when the state lies in the null space of the sum; the Gram construction turns that null-space condition into a ground-state condition. Ground-state preparation uses quantum imaginary-time evolution (QITE) implemented with quantum singular value transformation (QSVT) applied to a block-encoding of $H$, and overlap readout uses an interferometric measurement of $|\\langle 0_a\\tau(x)|\\psi_G\\rangle|^2$.","core_discovery":"The central claim is that a differential equation can be recast as an eigenvalue problem $H|\\psi_g\\rangle=\\lambda_{\\min}|\\psi_g\\rangle$ with $\\lambda_{\\min}\\approx 0$, where the effective Hamiltonian $H$ is formed by summing Gram matrices $T(A)=A^T A$ of latent-space operators that represent the equation, invariant boundary conditions, and regular data constraints. In the Chebyshev basis, differentiation becomes the constant matrix $G_n^T$, variable coefficients and multiplication by $x^p$ become constant mappings $M_{x^p}$, and data constraints become rank-one operators $B_n(x)$. Once $H$ is formed, the algorithm prepares its ground state by quantum imaginary-time evolution implemented through quantum singular value transformation, and evaluates the solution as an overlap $f_Q(x)=\\sqrt{\\eta_e}\\,\\langle\\tau(x)|\\psi_g\\rangle$; the scaling factor $\\sqrt{\\eta_e}$ is fixed by a single nonzero data point. The paper demonstrates this workflow on constant- and variable-coefficient ODEs, inhomogeneous ODEs, Laplace, heat, and wave equations, and two nonlinear ODEs, including one with no closed-form solution. The stated consequence is a variational-free DE solver whose cost is set by block-encoding the Hamiltonian and preparing its ground state.","pith_inferences":["If the degeneracy-selection issue is settled, the same construction could extend to systems of equations and higher-order nonlinearities via higher-order tensor products of the differentiation matrix, though the degenerate zero-energy subspace will grow correspondingly.","The overlap readout turns the solver into a quantum data source: a user can query $f_Q(x)$ at chosen points and use those samples to train classical surrogates, sidestepping full state tomography.","A testable consequence is that the quality of the nonlinear solution should depend on the initial state used for imaginary-time evolution; if different initial states converge to different zero-energy states with different overlap profiles, the product-state selection claim is not automatic.","The construction may transfer to other polynomial bases with real constant differentiation matrices and efficient feature maps, since those are the only basis-specific requirements; the paper demonstrates the Chebyshev case only."],"forward_implications":["A broad family of linear ODEs and PDEs (constant- and variable-coefficient, inhomogeneous, Laplace, heat, wave) can be solved by the same workflow: build the effective Hamiltonian, prepare its ground state, and read out overlaps, with no grid and no variational loop.","Solutions are available at arbitrary points $x$ in the domain, because the Chebyshev feature map labels continuous positions rather than discrete grid points.","Nonlinear ODEs are reduced to ground-state preparation on a doubled Hilbert space, with the entangling term $G_n^T\\otimes G_n^T$ encoding the quadratic nonlinearity; this extends the effective-Hamiltonian construction beyond linear problems.","Because preparation uses QITE/QSVT rather than variational search, the algorithm inherits the scaling and state-preparation guarantees of quantum linear-algebra methods, with block-encoding circuits that are polynomial in the number of qubits.","The Chebyshev-space model can be mapped to real space on an extended register, yielding fine-grid solutions without requiring the grid during the solving stage."],"supporting_citations":[{"why":"Supplies the Chebyshev differentiation matrix and the overlap-based quantum model that the effective-Hamiltonian construction extends.","marker":"[91]"},{"why":"Supplies the quantum Chebyshev feature map and the orthonormal Chebyshev basis states used to encode points and read out solutions.","marker":"[106]"},{"why":"Supplies the quantum imaginary-time evolution subroutine used to prepare the ground state of the effective Hamiltonian.","marker":"[107]"},{"why":"Supplies the algorithmic thermalization alternative for preparing low-energy states of the effective Hamiltonian.","marker":"[108]"},{"why":"Supplies the near-optimal ground-state preparation result behind the claimed state-preparation speed-up and its spectral-gap dependence.","marker":"[109]"},{"why":"Supplies the quantum singular value transformation used to implement imaginary-time evolution and polynomial approximations to the matrix exponential.","marker":"[110]"},{"why":"Supplies the qubitization/linear-combination-of-unitaries framework used to block-encode the effective Hamiltonian terms.","marker":"[115]"},{"why":"Supplies explicit quantum circuits for block-encoding structured sparse matrices, used for the Chebyshev differentiation operator.","marker":"[116]"},{"why":"Supplies the companion explicit block-encoding method for sparse matrices used to implement the differentiation matrix efficiently.","marker":"[117]"}],"fun_headline_variants":["DEs as ground states: no variational search needed","Quantum algorithm: encode DEs into Hamiltonians' ground states","From DE to ground state: a Chebyshev-space quantum solver","Nonlinear DEs solved via ground-state preparation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that for nonlinear equations the ground state prepared by imaginary-time evolution is the product state $|\\psi_g\\rangle\\otimes|\\psi_g\\rangle$ whose two factors coincide with the solution, even though the doubled-space Hamiltonian has at least $2^{2n-1}$ degenerate zero-energy eigenstates and no argument is given that evolution selects the product state.","fun_headline_variants_meta":{"raw":{"variants":["DEs as ground states: no variational search needed","Quantum algorithm: encode DEs into Hamiltonians' ground states","From DE to ground state: a Chebyshev-space quantum solver","Nonlinear DEs solved via ground-state preparation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000272,"raw_usage":{"total_tokens":1640,"prompt_tokens":961,"completion_tokens":679,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":577,"completion_tokens_details":{"reasoning_tokens":612}},"tokens_in":577,"tokens_out":679,"duration_ms":6077,"temperature":1.0,"reasoning_tokens":612,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T12:13:22.522515+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the effective-Hamiltonian construction for a nonlinear example such as Eq. (29) or Eq. (34) on a small register, start imaginary-time evolution from several distinct initial states (the all-zero state, a random product state, and a superposition), and check whether the prepared state always lies in the product subspace $|\\psi\\rangle\\otimes|\\psi\\rangle$ and yields the same rescaled overlap solution. If any initial state converges to an entangled zero-energy state with a different overlap profile, or if the reported solution is not reproduced, the central claim fails.","supporting_citations":[{"cited_title":"patent application for the method described in this manuscript has been submitted by Pasqal with HYW and O","cited_arxiv_id":null,"evidence_quote":"Supplies explicit quantum circuits for block-encoding structured sparse matrices, used for the Chebyshev differentiation operator."}],"review_version":1}