{"id":"82ddd373-58fd-4457-9f2b-fc295e835339","arxiv_id":"2411.15722","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the Doyle-Fuller-Newman battery model, the paper derives optimal-order error bounds for a backward Euler finite element scheme and demonstrates a twice-decoupled solver that is about twice as fast as existing solvers in 3D simulations.","lead":"This paper proves error estimates for a backward Euler finite element scheme applied to the Doyle-Fuller-Newman model of lithium-ion batteries. It also introduces a twice-decoupled solver that is about twice as fast as the fastest solver it compares against in 3D battery simulations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.1's O(τ²) rates require H² time regularity absent from Assumption 2.2; the τ² terms in Lemmas 2.3–2.4 are uncontrolled.","rationale":"The reader correctly identified Assumption 2.3 (uniform L∞ bounds on exact and discrete solutions) as a load-bearing unproven hypothesis; without it the Lipschitz estimates for J and the coefficient controls in Lemmas 2.3 and 2.4 fail, and no discrete existence is shown. I agree this is a serious gap. However, I find an even more directly fatal issue in the time regularity: the proof requires second time derivatives in L², but Assumption 2.2 provides only H¹ time regularity. This is not a matter of an unproven a priori bound — it is a mismatch between the hypotheses and the terms used in the derivation, so the O(τ²) claim does not follow from the stated assumptions. The theorem could likely be repaired either by adding H² time regularity to Assumption 2.2 or by downgrading the time rate to O(τ), which would weaken the advertised optimality. The paper's numerical experiments suggest first-order behavior in time (Figure 3 and Table B.4 show rates near 1.0–1.2), which is consistent with backward Euler but does not resolve the theoretical gap. Because the reader's verdict was already CONDITIONAL, and this concern sharpens the conditions under which the theorem can hold, I do not see the need to change the verdict; it should remain conditional pending a strengthened regularity assumption or a corrected time-rate statement.","tokens_in":37212,"tokens_out":4161,"duration_ms":38816,"concrete_test":"Re-derive the temporal error term in Lemma 2.3 using only Assumption 2.2: the term Cτ²∥∂²c1/∂t²∥²_{0,k;0,Ω} appears in (23); show that Assumption 2.2 (c1 ∈ H¹(0,T;H²_pw)) does not imply ∂²c1/∂t² ∈ L², e.g., by constructing a function in H¹ but not H² in time satisfying the other assumptions. If such a function exists, the proof of (23) fails and the theorem must either strengthen Assumption 2.2 to c1,c2 ∈ H²(0,T;L²) or reduce the time rate to O(τ).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central convergence result (Theorem 2.1) claims l² error bounds of order O(h² + τ² + (Δr)⁴). The τ² portion is derived in Lemma 2.3 and Lemma 2.4 through Taylor expansions that produce the terms Cτ²∥∂²c1/∂t²∥²_{0,k;0,Ω} (see (23)) and Cτ²∥∂²c2/∂t²∥²_{0,k;0,Ω₂;0,r} (see (29)). These terms are then carried into (37)–(38) and resummed to give the O(τ²) contribution in Theorem 2.1. However, Assumption 2.2 only postulates c1 ∈ H¹(0,T;H²_pw(Ω₁)) and c2 ∈ H¹(0,T;H¹(Ω₂;H¹_r)∩L²(Ω₂;H²_r)) — no second-order time differentiability. Under the stated hypotheses there is no guarantee that ∂²c1/∂t² or ∂²c2/∂t² belong to L², so the τ² terms may be infinite and the estimates in (23) and (29) are not legitimate. Separately, Assumption 2.3 postulates uniform L∞ bounds on the discrete solutions (ϕᵢh, ∇c₁h, c₁h, c̄₂h) and strict concentration bounds, without proof. These bounds are used in the Lipschitz estimates (26) and (31) for the nonlinear term J; if a discrete solution left this assumed region, the error recursion would break. The paper neither proves these discrete a priori bounds nor establishes existence of discrete solutions. Thus the theorem is conditional on unverified smoothness and boundedness assumptions that are not part of Hypotheses 1.2–2.3 as stated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes a backward Euler finite element discretization of the Doyle-Fuller-Newman (DFN) model of lithium-ion cells in the form of a coupled elliptic-parabolic system over the domains Ω1, Ω2, and the pseudo-(N+1)-dimensional domain Ω2r, for N=1,2,3. The main theoretical result, Theorem 2.1, claims l2(H1) and l2(L2(H^q_r)) error bounds of order O(h^2 + τ^2 + (Δr)^(4-2q)) for the potentials, electrolyte concentration, particle concentration, and surface concentration, under Assumptions 1.2, 2.1, 2.2, and 2.3. The proof follows the projection-based framework of the authors' prior work [18], adding fully discrete temporal error estimates via Lemmas 2.3 and 2.4. The paper also proposes a 'twice decoupled' solver that first eliminates the interior particle degrees of freedom by local inversion and then eliminates the microscopic surface block in the Newton Jacobian via a Schur complement, and it reports 2D and 3D numerical convergence tests plus a large-scale solver comparison with realistic battery parameters.","tokens_in":37629,"tokens_out":6253,"duration_ms":58297,"significance":"If the central theorem is valid, this would be the first fully discrete optimal-order error analysis for the DFN model in two and three spatial dimensions, extending the semidiscrete analysis of [18] and improving on the limitations of [23]. The numerical experiments with realistic parameters and the systematic comparison of seven solvers are valuable, and the proposed two-step decoupling strategy appears genuinely useful for large 3D simulations. The paper is transparent about several limitations, including the open existence/uniqueness question for N=2,3 and the need for a discrete compatibility condition. However, the main convergence claim is conditional on unproved discrete a priori bounds and on temporal regularity that is stronger than what Assumption 2.2 states; these gaps affect the core theorem and require attention before the result can be accepted.","major_comments":[{"comment":"The claimed O(τ^2) temporal rate is not justified by the stated hypotheses. Assumption 2.2 only postulates c1 ∈ H1(0,T;H2_pw(Ω1)) and c2 ∈ H1(0,T;H1(Ω2;H1_r)∩L2(Ω2;H2_r)), with no second-order time differentiability. The proofs of Lemmas 2.3 and 2.4 use Taylor expansions whose remainder terms contain ∥∂2c1/∂t2∥_{0,k;0,Ω} and ∥∂2c2/∂t2∥_{0,k;0,Ω2;0,r}, and these norms are then carried into (37)–(38) and summed to give the τ^2 terms in Theorem 2.1. Under Assumption 2.2 these norms may be infinite, so the estimates are not legitimate. The authors should either strengthen Assumption 2.2 to include ∂2c1/∂t2 ∈ L2(0,T;L2(Ω1)) and ∂2c2/∂t2 ∈ L2(0,T;L2(Ω2;L2_r)), or relax the temporal rate to O(τ) throughout the theorem.","section":"§2.2–2.3, Lemmas 2.3–2.4, Eqs. (23) and (29)"},{"comment":"Assumption 2.3 postulates uniform L∞ bounds on the exact and discrete solutions and strict interior concentration bounds for both c1 and c̄2, but no proof is given that the discrete solutions satisfy these bounds or even exist. These bounds are load-bearing: they are used in the Lipschitz estimates (26) and (31) for the nonlinear term J, and without them the error recursion in Theorem 2.1 breaks. Moreover, the paper itself notes that existence and uniqueness of the exact solution for N=2,3 is open. Theorem 2.1 is therefore conditional on unverified smoothness and boundedness assumptions. The authors should either prove the discrete a priori bounds (for example by an induction argument with a smallness condition), or state the theorem explicitly as conditional and explain how Assumption 2.3 can be verified in applications.","section":"Assumption 2.3, used in Eqs. (26) and (31)"},{"comment":"The proof of Theorem 2.1 invokes Proposition 2.1, which gives error estimates for ϕ1 and ϕ2 in the full test space V_h(Ω). However, the discrete scheme (13) is posed in the subspace W_h(Ω), and the equivalence with the full test space is only stated in Remark 2.1 under the discrete compatibility condition (17). Condition (17) is not included among the hypotheses of Theorem 2.1, and Remark 4.1 only says it is enforced 'indirectly via tight residual control.' The authors should add (17) as an explicit hypothesis of the theorem, or show that it holds for their discretization and quadrature, otherwise the use of Proposition 2.1 in the coupled error argument is not justified.","section":"Remark 2.1 and Proposition 2.1 in Theorem 2.1"}],"minor_comments":[{"comment":"The reported errors for c2 in L2(Ω2;L2_r) are on the order of 10^{-12} for both h- and Δr-refinement, which is at machine precision; no meaningful convergence order can be inferred from these values. This part of the numerical validation does not support the theoretical q=0 estimate, and the authors should replace it with a manufactured-solution test or a reference solution that differs in the radial discretization.","section":"Tables 1(f) and 2(f)"},{"comment":"After adding (37) and (38), the discrete Gronwall inequality is applied without stating the required smallness of τ; the theorem should include the standard condition that Cτ is bounded by a constant less than 1, or the constants should be tracked explicitly.","section":"Proof of Theorem 2.1, Gronwall step"},{"comment":"The volume integral in (30) contains a spurious 'dt' in the spatial integral over Ω2×(0,Rs(x)); this is a typographical error that should be corrected.","section":"Eq. (30)"},{"comment":"The notation τ is used for the time step but the text sometimes writes Δt; the authors should use a single symbol and define it once in Section 2.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The reliance on the authors' own prior projection estimates from [18] is acceptable because that paper is published and the present work genuinely extends it, but the unproved discrete a priori bound in Assumption 2.3 is the most serious gap and should be the main focus of the revision. The paper is otherwise well within the scope of math.NA and the numerical solver comparison is a strong practical contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper takes the authors' earlier semidiscrete projection framework for the DFN model and carries it to a fully discrete backward Euler scheme in 1–3 spatial dimensions. The claimed optimal rates (h², τ², Δr⁴) and the twice-decoupled solver are the two things worth knowing. The first is a real step past Bermejo's 1D non-optimal analysis; the second looks genuinely useful, with 2D/3D experiments on real parameters showing roughly a 2x speedup over the fastest existing solver and better memory behavior.\n\nWhat is good: the projection-based error decomposition from [18] is extended cleanly to the time-discrete case, and the error recursion for the coupled elliptic-parabolic system is handled carefully. The solver idea—local inversion of the tridiagonal particle systems to eliminate interior c2 unknowns, followed by a Schur-complement elimination that preserves sparsity—is well motivated, and Proposition 3.1 (sparsity of the Schur complement) is a nice observation. The numerical convergence rates in 2D match the stated orders.\n\nThe soft spots, in order of severity. First, the τ² claim is not justified by the stated hypotheses. Assumption 2.2 gives c1 and c2 only H¹ in time; Lemmas 2.3 and 2.4 and the proof of Theorem 2.1 freely use ∂²c1/∂t² and ∂²c2/∂t² in L². Either strengthen the assumption (e.g., H² in time) and say so, or downgrade the time rate to O(τ), which is the standard order for backward Euler. As written, the abstract and highlights overstate what the hypotheses support. Second, Assumption 2.3 postulates uniform L∞ bounds and interior concentration bounds on the discrete solutions with no indication of how they might be obtained. That is a common annoyance in nonlinear FE analysis, but here it is load-bearing for all the Lipschitz estimates; at minimum the paper should acknowledge this and state the result as conditional. For N=2,3 the authors themselves note existence/uniqueness is open, so the theorem holds for whatever regular solution is assumed to exist—a limitation worth stating plainly. Third, the validation has some artifacts: the c2 errors in L²(L²_r) in Tables 1 and 2 are at 1e-12 or smaller, which is round-off, and the measured order there is meaningless. Also the 3D results in Appendix B cover only h and τ; there is no 3D radial refinement test, so the claim that Δr rates are validated in 3D is unsubstantiated.\n\nNone of this sinks the paper. The central idea is sound and the solver contribution is likely to be useful even if the τ² detail is cleaned up. The paper deserves a serious referee and can be made publishable with a strengthened regularity assumption, an explicit conditional statement for Assumption 2.3, and a more careful presentation of the numerical convergence data.","headline":"Solid extension of the semidiscrete DFN analysis with a fast solver; the τ² rate needs stronger time regularity or a rate downgrade.","tokens_in":38151,"tokens_out":2682,"would_cite":true,"duration_ms":24345,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M15","65M60","65N15","65N30","78A57"],"pacs":[],"model":"deepseek-v4-flash","headline":"A backward Euler finite element scheme for the Doyle-Fuller-Newman lithium-ion cell model is proven to converge at optimal order in h, tau, and Delta r for spatial dimensions 1-3, and a twice-decoupled solver is shown to be the fastest…","keywords":["Doyle-Fuller-Newman model","backward Euler finite element","optimal-order error estimates","multiscale projection","lithium-ion battery simulation","nonlinear Gauss-Seidel solver","pseudo-two-dimensional model","error analysis"],"falsifier":"Run the scheme with a discharge rate high enough that the particle-surface concentration approaches its upper bound $c_{2,\\max}$ somewhere; if the error in $\\bar c_2$ or in $c_1$ then no longer decays like $h^2+\\tau^2+(\\Delta r)^4$ (or the discrete solution leaves the bounds of Assumption 2.3), the uniform convergence statement as proven would fail for that regime.","tokens_in":36961,"feed_emoji":"🔋","tokens_out":9581,"duration_ms":79245,"temperature":0.7,"pith_summary":"This paper claims that a backward Euler finite element discretization of the Doyle-Fuller-Newman (DFN) lithium-ion cell model converges at optimal order simultaneously in the spatial mesh size, the time step, and the particle-radius mesh size, in spatial dimensions N = 1, 2, 3. The result matters because the DFN model is the most widely used physics-based model for lithium-ion cells, and prior convergence analyses either broke down for N = 2, 3 or produced rates that did not match numerical practice. The paper also develops a twice-decoupled nonlinear solver that removes the microscopic variable's interior unknowns and lowers the Newton Jacobian's order, and reports on large three-dimensional simulations with realistic battery parameters that the new solver is the fastest and keeps memory usage bounded relative to existing alternatives.","feed_headline":"Battery model scheme hits provably optimal error rates","feed_subtitle":"Rigorous error rates and a memory-light solver bring the standard lithium-ion cell model closer to real-time use.","key_machinery":"The argument is carried by two projection operators: a Ritz projection $P_h$ for the electrolyte concentration and a multiscale tensor-product projection $P_{h\\Delta r}$ acting on $V_h^{(0)}(\\bar\\Omega_2)\\otimes V_{\\Delta r}^{(1)}([0,R_s])$ for the particle concentration $c_2$, whose trace estimate (Lemma 2.2) bounds the surface error of the projection by $h+(\\Delta r)^2$ and is what produces the fourth-order $\\Delta r$ rate for the surface concentration $\\bar c_2$. Every error is split as $\\rho+\\theta$ (projection error plus discrete error), and the coupling through the nonlinear Butler-Volmer source $J$ is controlled under Assumption 2.3 so that a discrete Gronwall argument closes the recursion. For the solver, the first decoupling uses the tridiagonal structure of the radial stiffness-plus-mass matrix $A$ to solve a local nonlinear scalar equation for each particle's surface degree of freedom, and the second decoupling performs a Schur-complement elimination of the diagonal microscopic block $D_{\\rm micro}$ whose sparsity is preserved (Proposition 3.1), followed by an optional nonlinear Gauss-Seidel outer iteration that couples $(c_1)$ to $(\\phi_1,\\phi_2,\\bar c_2)$.","core_discovery":"The central claim is stated as Theorem 2.1: under Hypotheses 1.2-2.3, the fully discrete backward Euler finite element solution satisfies, with a constant independent of mesh size, time step, and radial mesh size, $\\sum_{k=1}^{K}\\tau\\big(\\|\\phi_1^k-\\phi_{1h}^k\\|_{1,\\Omega}^2+\\|\\phi_2^k-\\phi_{2h}^k\\|_{1,\\Omega_2}^2+\\|c_1^k-c_{1h}^k\\|_{1,\\Omega}^2+\\|\\bar c_2^k-\\bar c_{2h}^k\\|_{0,\\Omega_2}^2\\big)\\le C(h^2+\\tau^2+(\\Delta r)^4)$ plus initial discretization error, and $\\sum_{k=1}^{K}\\tau\\|c_2^k-c_{2h\\Delta r}^k\\|_{0,\\Omega_2;q,r}^2\\le C(h^2+\\tau^2+(\\Delta r)^{4-2q})$ for the radial Sobolev index $q=0,1$. These are optimal-order rates for the element choices used: piecewise-linear discrete spaces for the potentials and electrolyte concentration, piecewise-constant-in-x and piecewise-linear-in-r for the particle concentration, and first-order backward Euler in time. The proof extends the semidiscrete analysis of [18] to the fully discrete setting, closing the gap that the earlier analysis of [23] left for dimensions 2 and 3 and aligning the predicted rates with observed numerical convergence.","pith_inferences":["The same projection-plus-trace-error technique should transfer to other micro-macro parabolic systems with an embedded radial coordinate, such as thermal or stress evolution inside electrode particles; the paper does not draw that conclusion.","Assumption 2.3 is typically verified a posteriori in practice; a natural extension would be to prove invariant-region preservation or to add a limiter that keeps discrete concentrations inside the bounds, then test whether the proven rates persist.","The sparsity-preserving Schur complement elimination of the diagonal microscopic block suggests the second decoupling generalizes to any locally-supported block-diagonal microscopic coupling, not only tridiagonal radial matrices.","Runtimes in Table 3 indicate the fully coupled solver is fast but memory-hungry; for embedded or real-time battery estimation, the 2DS-Eta variant may offer a more practical speed-memory balance, a use the paper leaves implicit."],"forward_implications":["The fully discrete DFN scheme is justified as a reliable simulation tool at optimal order for problems in one, two, and three spatial dimensions with realistic battery parameters.","The surface concentration $\\bar c_2$, which enters the Butler-Volmer reaction rate, gains fourth-order accuracy in the radial mesh size, so moderate radial refinements reach high accuracy.","Memory use is kept under control because the interior radial unknowns are recovered by backward substitution and never enter the global Jacobian.","The solver's two decoupling levels are compatible with Newton's method with line search and with an optional nonlinear Gauss-Seidel outer loop, giving a speed-memory trade-off adjustable to available hardware.","Large 3D simulations with real parameters show the new solver is the fastest among the compared solvers and largely insensitive to finer microscopic discretization."],"supporting_citations":[{"why":"the semidiscrete error analysis and the multiscale projection operator whose trace estimate the fully discrete proof extends.","marker":"[18]"},{"why":"the first convergence analysis of the backward Euler FEM for this model, whose limitations (failure for N=2,3) the paper addresses.","marker":"[23]"},{"why":"supplies the approximation error bound for the Ritz projection used for the electrolyte concentration.","marker":"[50]"},{"why":"establishes the H^1_* weak formulation used for the potential subsystem in the analysis.","marker":"[19]"},{"why":"local existence and uniqueness of 1D solutions, cited to support the regularity assumptions.","marker":"[21]"},{"why":"the original DFN model formulation that defines the system of equations being discretized.","marker":"[2]"},{"why":"an existing decomposition solver used as a baseline in the performance comparison.","marker":"[13]"},{"why":"a fully decoupled solver benchmarked against the proposed solver.","marker":"[42]"},{"why":"supplies real battery parameters used in the three-dimensional numerical validation.","marker":"[3]"},{"why":"supplies the pouch-cell geometry and parameters used in the two-dimensional convergence tests.","marker":"[16]"}],"fun_headline_variants":["Twice-decoupled solver yields optimal error rates for Li-ion model","Optimal error bounds and a fast solver for lithium-ion simulations","Provably optimal errors for battery model with a decoupled solver","Li-ion finite element scheme: optimal rates, efficient solver"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes without proof that both the exact and the discrete solutions stay in a fixed bounded region (uniform $L^\\infty$ bounds on potentials, gradients, and concentrations, with concentrations bounded away from $0$ and $c_{2,\\max}$), and that a solution with the required regularity exists in dimensions 2 and 3, where existence is currently open.","fun_headline_variants_meta":{"raw":{"variants":["Twice-decoupled solver yields optimal error rates for Li-ion model","Optimal error bounds and a fast solver for lithium-ion simulations","Provably optimal errors for battery model with a decoupled solver","Li-ion finite element scheme: optimal rates, efficient solver"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000562,"raw_usage":{"total_tokens":2691,"prompt_tokens":990,"completion_tokens":1701,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":1630}},"tokens_in":606,"tokens_out":1701,"duration_ms":13218,"temperature":1.0,"reasoning_tokens":1630,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:58:58.980256+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the scheme with a discharge rate high enough that the particle-surface concentration approaches its upper bound $c_{2,\\max}$ somewhere; if the error in $\\bar c_2$ or in $c_1$ then no longer decays like $h^2+\\tau^2+(\\Delta r)^4$ (or the discrete solution leaves the bounds of Assumption 2.3), the uniform convergence statement as proven would fail for that regime.","supporting_citations":[{"cited_title":"Optimal convergence in finite element semi-discrete error analysis of the Doyle-Fuller-Newman model beyond 1D with a novel projection operator","cited_arxiv_id":"2411.10758","evidence_quote":"the semidiscrete error analysis and the multiscale projection operator whose trace estimate the fully discrete proof extends."},{"cited_title":"Bermejo, Numerical analysis of a finite element formulation of the P2D model for Lithium-ion cells, Numerische Mathematik 149 (2021) 463–505","cited_arxiv_id":null,"evidence_quote":"the first convergence analysis of the backward Euler FEM for this model, whose limitations (failure for N=2,3) the paper addresses."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the approximation error bound for the Ritz projection used for the electrolyte concentration."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"establishes the H^1_* weak formulation used for the potential subsystem in the analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"local existence and uniqueness of 1D solutions, cited to support the regularity assumptions."},{"cited_title":"Bermejo, P","cited_arxiv_id":null,"evidence_quote":"an existing decomposition solver used as a baseline in the performance comparison."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"a fully decoupled solver benchmarked against the proposed solver."},{"cited_title":"Smith, C.-Y","cited_arxiv_id":null,"evidence_quote":"supplies real battery parameters used in the three-dimensional numerical validation."},{"cited_title":"Timms, S","cited_arxiv_id":null,"evidence_quote":"supplies the pouch-cell geometry and parameters used in the two-dimensional convergence tests."}],"review_version":1}