{"id":"fa0ba8dc-514c-4af9-a0f8-e570a6c5ae7e","arxiv_id":"2411.09617","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"An energy-adaptive Riemannian gradient method provably converges to the unique ground state of multicomponent Bose-Einstein condensates, with fast local Newton-type completion.","lead":"This paper develops and analyzes curved-space optimization methods for computing ground states of multicomponent Bose-Einstein condensates, where each component is normalized to a fixed particle number. It proves an energy-adaptive gradient method converges globally to the unique ground state, and shows Newton-type variants accelerate convergence near it.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The FE implementation in Section 7 drops the positivity-preservation premise of Theorem 6.4, so the abstract's 'global convergence / discretization robustness' claim is not supported for the actual solver.","rationale":"I checked the infinite-dimensional core before focusing on the discretization gap. Theorem 2.4, Theorem 3.3, and Theorem 3.5 follow the cited single-component arguments; Lemma 6.2 is plausible, since the quadratic-form comparison F(|u|) <= F(u) is valid and Schroedinger resolvents are positivity preserving; and the energy-decay argument in Theorem 6.3 has the right structure. The paper is honest: Section 7 explicitly flags the missing discrete maximum principle. The weakness is that the central practical claim, global convergence and robustness across spatial discretizations, is made in the abstract for the method while the only rigorous global convergence holds in the infinite-dimensional setting and does not transfer to the implemented FEM. The numerical experiments use that FEM with constant step size tau = 1 and report no positivity monitoring; in the 2D random-potential case other methods find excited states, so the distinction is practically important. This does not invalidate the continuous theory, but it means the headline claim about the solver is conditional on a positivity-preserving discretization or a new discrete convergence proof. The reader identified the same premise, and I agree. Verdict remains conditional: the theoretical contribution is valuable, but the implemented method's advertised guarantee is not established and needs the concrete positivity check or a corrected claim.","tokens_in":35271,"tokens_out":10324,"duration_ms":114392,"concrete_test":"On a coarse FE mesh (for example h = 2^{-5}), assemble G = S + M V_j + M rho_j(Phi_0) + c_{Phi_0} M with the S2 shift and solve G u = 1, where 1 is the componentwise all-ones vector. If min_i u_i < 0, the discrete analogue of Lemma 6.2 fails for that mesh and nonlinearity. Then run one eaRGD update (5.2) from a normalized positive constant Phi_0 with tau = 1 and check whether the retracted iterate Phi_1 has any negative nodal value; if it does, Theorem 6.4's positivity induction is broken in the implemented discretization, and the global-convergence claim for the solver is unsupported. A positivity-preserving discretization from [38] should be used instead if the claim is to be retained.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that eaRGD converges globally to the unique ground state rests on Theorem 6.4, whose induction step uses Lemma 6.2 to conclude that bA^{-1}_{phi_k} v >= 0 for v >= 0 and hence that phi_{k+1} >= 0. This nonnegativity is what rules out convergence to sign-changing excited states. In the finite element discretization of Section 7, the discrete operator acting on coefficient vectors is M^{-1}(S + M V_j + M rho_j + c M), and standard FEM matrices are not inverse-positive: the paper itself states in Section 7 that 'the global convergence results for the eaRGD method no longer hold in the finite-dimensional case because our particular choice of the finite element discretisation does not satisfy a discrete maximum principle.' The numerical experiments use bi-quadratic FEM with tau = 1 and report only residual norms and energies, not the minimum nodal value of any iterate. Thus the implemented solver can leave the nonnegative cone; once it does, the only estimate excluding convergence to an excited state is gone, and no discrete analogue of Theorem 6.4 or of the quantitative local-rate statement is provided. The infinite-dimensional theorem may be correct, but the abstract's 'reliability and robustness with respect to the choice of the spatial discretisation' is stronger than what is proved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops Riemannian optimisation methods for computing ground states of multicomponent Bose-Einstein condensates, formulated as global minimisers of the energy functional (1.2) under component-wise mass constraints on the infinite-dimensional generalised oblique manifold (2.2). Under Assumptions A1 (nonnegative bounded potentials) and A2 (symmetric positive definite interaction matrix K), the authors prove existence of a ground state (Theorem 2.4), uniqueness up to component-wise sign (Theorem 3.3), and a second-order necessary optimality condition that is in fact positive definite on the tangent space (Theorem 3.5). Section 4 derives the manifold geometry, projections, and Riemannian gradients/Hessians for several metrics: the L2-metric, an energy-adaptive metric, and a Lagrangian-based metric. Section 5 presents the corresponding Riemannian gradient descent methods, an alternating variant, and Riemannian Newton methods. Section 6 contains the main convergence analysis: global convergence of the energy-adaptive Riemannian gradient descent (eaRGD) method to the unique nonnegative ground state (Theorem 6.4) and a local linear convergence rate (Theorem 6.7). Section 7 gives a finite element discretisation, and Section 8 reports 1D and 2D numerical experiments with reproducible Julia code. The infinite-dimensional results are carefully stated and the proofs are mostly self-contained, with some steps delegated to earlier works by the same authors and others.","tokens_in":35495,"tokens_out":16046,"duration_ms":150295,"significance":"If the central theorems are correct, the paper gives a substantial and useful framework: a rigorous infinite-dimensional Riemannian treatment of multicomponent BEC ground states, explicit convergence guarantees for a preconditioned gradient method, and a clean separation between the infinite-dimensional algorithm and its spatial discretisation. Strengths include the clearly stated assumptions, the absence of fitted parameters in the theoretical claims, the connection to the NLEVP formulation, and the availability of reproducible code. The advertised property of 'reliability and robustness with respect to the choice of the spatial discretisation' is, however, not supported for the implemented finite element method, because Section 7 explicitly concedes that the global convergence theorem does not transfer to the standard FEM. This gap is load-bearing for the paper's headline claim and should be fixed before publication.","major_comments":[{"comment":"The paper states in Section 7 that 'the global convergence results for the eaRGD method no longer hold in the finite-dimensional case because our particular choice of the finite element discretisation does not satisfy a discrete maximum principle.' This directly weakens the Abstract's claim of 'reliability and robustness with respect to the choice of the spatial discretisation.' The numerical experiments in Section 8 use exactly this FEM (bi-quadratic elements with tau_k = 1), so Theorem 6.4 does not apply to the implemented solver. In the infinite-dimensional proof, nonnegativity of the iterates is essential for ruling out convergence to sign-changing excited states; without a discrete maximum principle, no discrete analogue of that argument is available, and the experiments do not report any diagnostic such as minimum nodal values of the iterates. I request that the claims be restricted to the infinite-dimensional iteration, or that a positivity-preserving stabilised discretisation (e.g., the scheme of reference [38]) be analysed and used, or that the experiments explicitly monitor and report positivity of all iterates.","section":"Section 7; Abstract"},{"comment":"The final estimate of the proof, namely the bound on sum_j || |phi_{k,j}|^2 - |phi_{*,j}|^2 ||_{L^2}^2 tending to zero, is not by itself sufficient for the stated conclusion 'the whole sequence {phi_k} converges strongly to phi* in L and in H.' The L-convergence can be recovered from nonnegativity via the pointwise inequality |a-b|^2 <= |a^2-b^2| for a,b >= 0, and the H-convergence then follows by combining the energy identity with the Garding inequality (2.7) and the already established L^2 convergence of phi_k - phi*. Since Theorem 6.4 is the central global convergence result, these steps should be written out explicitly in the proof rather than left to the reader.","section":"Theorem 6.4 proof"},{"comment":"The numerical experiments impose Neumann boundary conditions while the theory is developed on H_0^1(Omega) with homogeneous Dirichlet conditions. The statement that 'all values on the boundary are sufficiently close to zero' due to the trapping potentials is not quantified in the paper. This is a mismatch between the analysed setting and the computed setting; please either use homogeneous Dirichlet conditions in the experiments or report quantitative evidence (e.g., boundary values or residuals) that the Neumann modification is negligible.","section":"Section 8, first paragraph"}],"minor_comments":[{"comment":"The sentence 'unless specifically triggered, a violation of the maximum principle and potential convergence to an excited state from a positive initial guess will typically not be observed' is an empirical claim that is not supported by data in the paper; either provide numerical evidence or remove the sentence.","section":"Section 7, last paragraph"},{"comment":"The AMS subject classification '66N25' appears to be a typo; the intended code is probably '65N25' for numerical methods for eigenvalue problems.","section":"Title page"},{"comment":"For the beta = 100 panel, the caption says the non-alternating versions are shown, but the legend entry does not identify which curves correspond to the non-alternating variants; please clarify the line styles or legend.","section":"Figure 8.2 caption"},{"comment":"The header row repeats 'outer iter.' and 'aver. matr.-vec. mult. per iter.' across the three beta columns; consider using a grouped header or a clearer table layout to avoid confusion.","section":"Table 8.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and the infinite-dimensional results appear sound and valuable. The main risk is the discrepancy between the advertised discretisation robustness and the explicit failure of the discrete maximum principle in Section 7; I would ask the authors to either prove a discrete convergence result or substantially qualify the abstract and conclusions. The stress-test concern about positivity preservation in the FEM implementation is valid and should be the primary focus of the revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe thing to know: this is a genuine advance, not a repackaging. It proves existence and uniqueness of ground states for multicomponent BECs on bounded domains, sets up an infinite-dimensional generalized oblique manifold, and gives a global convergence theorem for energy-adaptive Riemannian gradient descent plus a quantitative local rate. The multicomponent extension is nontrivial because the component Hamiltonians differ; the energy-adaptive metric acts as a preconditioner. The proofs are mostly careful. I checked the main steps: Garding inequality, projection formulas, energy decay, and the positivity argument in Lemma 6.2 all hang together. Delegating the compactness subsequence to earlier single-component work is acceptable, and those are independent published results. The paper also ships Julia code and data, and the experiments compare against prior benchmarks, so the numerical claims are reproducible.\n\nThe soft spot is exactly the one flagged in the stress test, and the authors themselves flag it: Theorem 6.4 uses positivity preservation to exclude convergence to excited states, but the finite element discretization in Section 7 does not satisfy a discrete maximum principle. The abstract says \"reliability and robustness with respect to the choice of the spatial discretisation,\" which is stronger than what is proved. Section 7 explicitly retracts the global convergence result for the FE method and says violations will \"typically not be observed\" in practice. That is honest, but it leaves the implemented solver without a convergence guarantee. A discrete convergence theorem or a positivity-preserving discretization would close the gap. This is a real limitation of the paper as a numerical methods paper, but it is not a flaw in the infinite-dimensional theory, and it is not hidden.\n\nMinor: the numerical experiments report residual norms and energies, not minimum nodal values, so we cannot see whether iterates ever leave the positive cone. The local rate theorem assumes K entrywise non-negative, which is fine but narrower than A2.\n\nWho this is for: anyone working on BEC ground states or nonlinear eigenvalue problems in infinite dimensions. The framework and the convergence analysis deserve a serious referee. I would send it out and ask the authors to either prove a discrete counterpart or soften the abstract.","headline":"Strong infinite-dimensional convergence theory for multicomponent BEC ground states, with an honest but real gap between the global theorem and the finite-element implementation.","tokens_in":36052,"tokens_out":2128,"would_cite":true,"duration_ms":22840,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65N25","81Q10","35Q55"],"pacs":[],"model":"deepseek-v4-flash","headline":"Energy-adaptive Riemannian descent is proven to converge to the unique ground state of multicomponent Bose-Einstein condensates.","keywords":["Riemannian optimisation","multicomponent Bose-Einstein condensates","coupled Gross-Pitaevskii equations","nonlinear eigenvector problem","generalised oblique manifold","energy-adaptive metric","global convergence","finite element discretisation"],"falsifier":"Run the exact eaRGD iteration (5.2) on a two-component example with a positive definite interaction matrix $K$ that has at least one negative off-diagonal entry and a nonnegative initial frame, using step sizes within the Theorem 6.3 bounds; if the iterates lose nonnegativity or converge to a sign-changing constrained critical point, Theorem 6.4 is false. Alternatively, in the standard finite element implementation where the discrete maximum principle fails, search for a positive initial vector and step size for which the residual tends to zero at an excited state rather than the ground state; such an example would confirm the paper's stated limitation and show that the guarantee is genuinely lost in the discretised method.","tokens_in":35031,"feed_emoji":"⚛️","tokens_out":7466,"duration_ms":63396,"temperature":0.7,"pith_summary":"This paper treats the ground state of a p-component Bose-Einstein condensate as the global minimiser of an energy functional on an infinite-dimensional generalised oblique manifold, the set of p-frames with prescribed component masses. It proves that under nonnegative external potentials and a positive definite interaction matrix, a ground state exists, is unique up to global signs of its components, and is the solution of the coupled Gross-Pitaevskii nonlinear eigenvector problem in which each component sits at the smallest eigenvalue of its shifted Hamiltonian. The central numerical claim is that a Riemannian gradient descent method using an energy-adaptive metric converges strongly in $H$ to that ground state from any nonnegative initial frame, with an energy decay uniform in the spatial discretisation, and converges linearly near the solution. This matters because it gives a globally convergent, discretisation-robust solver for the miscible regime, and the same manifold geometry supports efficient Newton-type accelerations.","feed_headline":"Energy-adaptive descent provably finds the BEC ground state","feed_subtitle":"A metric-aware gradient iteration converges globally to the unique nonnegative minimiser, with a rate set by the spectral gap.","key_machinery":"The central object is the infinite-dimensional generalised oblique manifold $OB_N(p,H)$, the set of p-frames $\\varphi\\in H=[H^1_0(\\Omega)]^p$ with prescribed component masses $\\langle\\langle\\varphi,\\varphi\\rangle\\rangle=N$. The argument is carried by the energy-adaptive metric $g_{\\varphi,a}(z,y)=\\tilde a_\\varphi(z,y)$, where $\\tilde a_\\varphi$ is the Gross-Pitaevskii bilinear form shifted to be coercive. Under this metric the Riemannian gradient of the energy is simply the projection $\\mathrm{grad}_a E(\\varphi)=P_{\\varphi,a}(\\varphi)$, so one eaRGD step is an inverse subspace iteration with adaptive damping. The machinery that converts energy decay into convergence to the ground state is the maximum-principle-type lemma (Lemma 6.2) that $\\tilde A_\\varphi^{-1}v\\ge 0$ for $v\\ge 0$, which preserves nonnegativity of all iterates.","core_discovery":"The central discovery is Theorem 6.4: the energy-adaptive Riemannian gradient descent iteration (5.2), with step sizes chosen as in Theorem 6.3 and a nonnegative starting frame, converges strongly in $H$ to the unique nonnegative ground state. The proof rests on preservation of positivity: the inverse of the shifted Gross-Pitaevskii operator maps nonnegative functions to nonnegative functions (Lemma 6.2), which prevents an energy-decreasing sequence from converging to a sign-changing excited state. Locally, the iteration is a contraction whose rate is controlled by the ratio of sums of the two smallest eigenvalues of the component Gross-Pitaevskii operators (Theorem 6.7). In the miscible regime, the paper also establishes existence, uniqueness up to global signs, and the characterisation of the ground state as the NLEVP eigenvector with minimal component-wise eigenvalues.","pith_inferences":["By the same positivity-preservation logic, the global convergence strategy should extend to other eigenvector nonlinearities such as Hartree-Fock and Kohn-Sham problems once a positivity-preserving discretisation is available.","The numerical failures of Newton-type methods on random potentials suggest that the practical bottleneck is the size of the basin of attraction rather than the local rate; a quantitative basin estimate for the regularised Newton method would be a natural next step.","All numerical experiments use interaction matrices with nonnegative entries, so the performance of eaRGD in the regime with negative off-diagonal couplings (where the shift becomes essential) remains an open test.","The alternating component-wise update offers easy parallelisation and, for strong interactions, appears more reliable than the simultaneous update; quantifying this advantage would be a direct experimental extension."],"forward_implications":["For any nonnegative initial frame in the miscible regime, the energy-adaptive Riemannian gradient descent method with the step-size bounds of Theorem 6.3 converges strongly in $H$ to the unique nonnegative ground state.","The energy decay estimate is independent of the spatial discretisation, so mesh refinement does not degrade the qualitative global convergence guarantee.","Near the ground state, the iteration converges linearly with a rate determined by the spectral gap of the component Gross-Pitaevskii operators.","The Riemannian Newton and regularised Newton methods show fast local convergence in the experiments, and the reliably convergent eaRGD iteration is a suitable globalisation strategy for them.","The positivity-preservation argument does not transfer to standard finite elements; a stabilised discretisation is needed to retain the global convergence guarantee."],"supporting_citations":[{"why":"Supplies the single-component existence, spectral characterisation, and minimum-eigenvalue arguments that Theorem 2.4 and Proposition 3.2 generalise to multicomponent BECs.","marker":"[24]"},{"why":"Provides the single-component Sobolev-gradient-flow global convergence analysis that Theorem 6.4 adapts to the multicomponent energy-adaptive setting.","marker":"[42]"},{"why":"Supplies the local convergence-rate technique, including the Ostrowski argument, that Theorem 6.7 generalises to multicomponent ground states.","marker":"[43]"},{"why":"Gives the two-component full-space existence result and the miscibility parameter that motivate the bounded-domain multicomponent existence theorem.","marker":"[15]"},{"why":"Provides the convex-density inequality on which the uniqueness proof of Theorem 3.3 relies.","marker":"[48]"},{"why":"Proposes the positivity-preserving finite element scheme that would restore the discrete global convergence guarantee that standard FEM lacks.","marker":"[38]"},{"why":"Supplies a recent convergence analysis of the RGD method with an $H^1$ metric for the single-component Gross-Pitaevskii problem, serving as a comparison baseline.","marker":"[28]"}],"fun_headline_variants":["Energy-adaptive metric proves global convergence to BEC ground state","Global convergence to unique nonnegative BEC ground state","Metric-adaptive Riemannian descent converges globally for BEC","Energy-adaptive iteration achieves provable global convergence","Provably convergent metric-adaptive method for BEC ground states"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that nonnegativity is preserved at every iteration: the inverse of the shifted Gross-Pitaevskii operator must map nonnegative functions to nonnegative functions, which the proof uses to ensure that an energy-decreasing sequence cannot drift to a sign-changing excited state.","fun_headline_variants_meta":{"raw":{"variants":["Energy-adaptive metric proves global convergence to BEC ground state","Global convergence to unique nonnegative BEC ground state","Metric-adaptive Riemannian descent converges globally for BEC","Energy-adaptive iteration achieves provable global convergence","Provably convergent metric-adaptive method for BEC ground states"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001293,"raw_usage":{"total_tokens":5245,"prompt_tokens":879,"completion_tokens":4366,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":495,"completion_tokens_details":{"reasoning_tokens":4286}},"tokens_in":495,"tokens_out":4366,"duration_ms":28826,"temperature":1.0,"reasoning_tokens":4286,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T20:27:59.938323+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the exact eaRGD iteration (5.2) on a two-component example with a positive definite interaction matrix $K$ that has at least one negative off-diagonal entry and a nonnegative initial frame, using step sizes within the Theorem 6.3 bounds; if the iterates lose nonnegativity or converge to a sign-changing constrained critical point, Theorem 6.4 is false. Alternatively, in the standard finite element implementation where the discrete maximum principle fails, search for a positive initial vector and step size for which the residual tends to zero at an excited state rather than the ground state; such an example would confirm the paper's stated limitation and show that the guarantee is genuinely lost in the discretised method.","supporting_citations":[{"cited_title":"Bao and Y","cited_arxiv_id":null,"evidence_quote":"Gives the two-component full-space existence result and the miscibility parameter that motivate the bounded-domain multicomponent existence theorem."},{"cited_title":"Positivity preserving finite element method for the Gross-Pitaevskii ground state: discrete uniqueness and global convergence","cited_arxiv_id":"2405.17090","evidence_quote":"Proposes the positivity-preserving finite element scheme that would restore the discrete global convergence guarantee that standard FEM lacks."}],"review_version":1}