{"id":"0af244a2-030a-46a2-908d-364f47eb8031","arxiv_id":"2411.10758","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Optimal O(h+(Δr)^2) finite element convergence for the Doyle-Fuller-Newman battery model in dimensions N=2,3 is established with a new projection operator and verified numerically.","lead":"A new finite element analysis for the lithium-ion battery model proves optimal error rates of h plus radial step squared in two and three dimensions. It introduces a projection operator for the particle diffusion equation and reports the first 2D and 3D convergence tests.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.9 is conditional on the unproved discrete uniform bound of Assumption 3.4; without a proof that c1h and c̄2h/c2,max stay in the assumed intervals uniformly in h and Δr, the Lipschitz and coercivity steps in Lemmas 3.6–3.8 and the main theorem do not close.","rationale":"The reader's weakest assumption identifies exactly this point: Assumption 3.4 is the unproved discrete L∞/positivity bound needed for the nonlinear Lipschitz and coercivity arguments. My read agrees with that and with the conditional assessment. The projection analysis (Lemmas 3.1–3.5, Theorems 3.4–3.5) is internally coherent, and the conditioning of Theorem 3.9 is stated explicitly, so a conditional theorem can still be a valid contribution if the assumption is later supplied. The numerical experiments support O(h) and O((Δr)^2) rates, although the 3D h-rate is computed from a single usable mesh pair and the experiments do not stress Assumption 3.4. No internal contradiction is apparent, so REJECT is not warranted. The verdict should therefore remain CONDITIONAL: the advertised optimal convergence is not fully established until Assumption 3.4 is either proved from the scheme and data or replaced by a proven discrete invariant-region argument.","tokens_in":28950,"tokens_out":12723,"duration_ms":154911,"concrete_test":"Run the §4.1/4.2 experiments with an added monitor that records min_{x,t} c1h and max_{x,t} (c̄2h/c2,max) over all time steps and for each refinement level; repeat with initial data perturbed toward the singular limits (c10 near 0, c20 near c2,max). If any discrete solution leaves the interval required by Assumption 3.4 for h, Δr → 0, or exits faster than the bound allows, the theorem is conditional on a non-automatic bound. If none does across the sweep, the gap is reduced to a missing proof rather than a demonstrated failure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The single most load-bearing gap is Assumption 3.4. The central theorem quantifies over discrete solutions satisfying uniform L∞ bounds for φ1h, φ2h and uniform lower/upper bounds for c1h and c̄2h that are independent of h and Δr, but the paper gives no proof that the semi-discrete scheme (3.2)–(3.5) produces such solutions. These bounds are used essentially: in Lemma 3.7 they make κ1(c1h) ≥ C > 0 for coercivity and make Jm, Um, κi, f′ Lipschitz on the relevant range; Lemma 3.6 uses the resulting Lipschitz estimate (3.26); Lemma 3.8 uses the same estimate. Without Assumption 3.4, the constants in the proof could blow up as c1h → 0 or c̄2h → c2,max, and the error equation for (θc1, θc2) is not closed. Assumption 3.4 is not a consequence of Assumptions 3.1–3.3: those bound only the exact solution, and the c1 equation has a non-monotone source J(c1, c̄2, φ1, φ2) coupled to the potentials, so a standard maximum principle does not apply. The numerical experiments do not test the assumption; they use physical data that stay comfortably inside the bounds. Thus Theorem 3.9 is an a priori error estimate conditional on a discrete regularity statement, rather than a fully established optimal-convergence theorem. This does not invalidate the projection construction or the conditional estimates, but it is the place where the advertised claim is least secure.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a finite element semi-discrete error analysis for the Doyle-Fuller-Newman (DFN) lithium-ion battery model in spatial dimensions N=1,2,3. Its main technical contribution is a tensor-product projection operator P_hΔr on the pseudo-(N+1)-dimensional particle domain, for which the authors prove approximation estimates in L2(Ω;H^q_r), q=0,1, and a trace estimate at r=R_s. These projection estimates are combined with elliptic error estimates for the two potentials and parabolic estimates for the electrolyte and particle concentrations. Theorem 3.9 asserts that, under Assumptions 2.1-2.2 and 3.1-3.4, the semi-discrete errors satisfy the bound ||φ1-φ1h||_{L2(0,t;H1(Ω))} + ||φ2-φ2h||_{L2(0,t;H1(Ω2))} + ||c1-c1h||_{L2(0,t;H1(Ω))} + ||c̄2-c̄2h||_{L2(0,t;L2(Ω2))} ≤ C(h+(Δr)^2) plus initial-data errors, and ||c2-c2hΔr||_{L2(0,t;L2(Ω2;H^q_r))} ≤ C(h+(Δr)^{2-q}) for q=0,1. Numerical experiments in 2D+1D and 3D+1D settings with real battery parameters report observed rates consistent with O(h) and O((Δr)^2).","tokens_in":29308,"tokens_out":8212,"duration_ms":84460,"significance":"If the main theorem is accepted, this would be the first finite element convergence analysis for the DFN model beyond one spatial dimension, achieving rates that are optimal with respect to both the macroscopic mesh size h and the particle-radius mesh Δr. The projection operator construction is a genuine contribution: it is natural for the pseudo-(N+1)-dimensional structure, avoids the change of variables used in earlier works, and the approximation arguments in Section 3.1 are clean and internally coherent. The conditional error estimates are also presented carefully, with explicit tracking of where each assumption enters. The numerical experiments with realistic battery parameters are a useful step, and the paper explicitly identifies the added difficulty caused by avoiding the change of variables. However, the advertised optimal-convergence statement is conditional on Assumption 3.4, a uniform L∞ and positivity bound on the discrete solution that is neither proved nor derived from the other assumptions. This currently prevents the paper from fully delivering the claimed bridging of the gap for N=2,3.","major_comments":[{"comment":"Assumption 3.4 is a load-bearing hypothesis that is asserted without proof, and it is used essentially in the proof of the main theorem. In Lemma 3.7 it provides the coercivity κ1(c1h) ≥ C > 0 and the Lipschitz continuity of Jm, Um, κi, and f′; in Lemma 3.6 it provides the Lipschitz estimate (3.26); and Lemma 3.8 uses the same estimate. The manuscript gives no discrete maximum principle, no invariant-region argument, and no smallness/bootstrap argument showing that the semi-discrete solution of (3.2)-(3.5) satisfies these bounds independently of h and Δr. The exact-solution bounds in Assumption 3.3 do not imply them, since the c1 equation has a non-monotone source coupled to the potentials. Consequently Theorem 3.9 is an a priori error estimate conditional on discrete regularity, rather than an unconditional optimal-convergence theorem. The authors should either prove such a bound (for example, using the monotonicity of ∂Jm/∂η to control the coupling) or explicitly restate the theorem and the abstract as conditional.","section":"Assumption 3.4 and Theorem 3.9"},{"comment":"The coercivity step contains the sentence 'Selecting ε < 1/2 sufficiently large, we then have...', which is internally contradictory: for ε < 1/2 the coefficient (1 - 1/(2ε)) is negative, and its magnitude becomes small only as ε approaches 1/2 from below, not as ε becomes large. As written, the proof of coercivity is therefore not justified, although the intended argument (choose ε close to 1/2 so that the negative L2 coefficient can be absorbed by the Poincaré inequality) appears repairable. This must be corrected.","section":"Appendix A, proof of Lemma 3.7"},{"comment":"The reported L2(Ω2;L2_r) errors for c2 are on the order of 10^{-12} to 10^{-13} while the concentrations are O(1), so the observed rates around 2.04 are at the level of round-off and do not provide meaningful confirmation of the (Δr)^2 rate for that quantity. The H1_r errors and the errors for the other variables do show clean rates, but the claim of being the first detailed numerical validation should be limited to those more robust quantities, or the experiments should be redesigned so that the c2 L2 error is well above machine precision (for example, by using coarser radial meshes and a reference solution that is not excessively fine).","section":"Section 4, Tables 1-4"},{"comment":"Theorem 3.9 also assumes H2-regularity of the exact solution (Assumption 3.2) for N=2,3. As the introduction itself notes, well-posedness of the full P4D model in these dimensions is open. The theorem therefore does not unconditionally bridge the gap for N=2,3; it establishes the error estimate conditional on a regularity hypothesis that is currently unproved for the P4D case. This limitation should be stated explicitly when the contribution is summarized in the abstract and introduction, rather than only appearing as a technical assumption in Section 3.","section":"Assumption 3.2 and the claimed gap for N=2,3"}],"minor_comments":[{"comment":"The word 'convenent' should be 'convenient'.","section":"Section 2.1"},{"comment":"The text 'by Assumption 3.33' should read 'by Assumption 3.3'; the proof of Lemma 3.8 also relies on the Lipschitz estimate (3.26), which is derived only in Lemma 3.6, so the dependence should be stated explicitly.","section":"Appendix B, proof of Lemma 3.8"},{"comment":"The abstract states optimal convergence rates of h+(Δr)^2, but for c2 in the H1_r norm the theorem gives the rate h+Δr (i.e., (Δr)^{2-q} with q=1); the statement 'h+(Δr)^2' should be qualified as holding for the L2 norms of c2 and for the quantities in the first display of Theorem 3.9.","section":"Abstract and Theorem 3.9"},{"comment":"The space VhΔr is defined in (3.1) as an intersection of tensor-product spaces over Ωn and Ωp; since the two tensor-product spaces act on disjoint subdomains, the intended object is a piecewise-defined space rather than an intersection, and the notation should be clarified.","section":"Section 3, definition of VhΔr"}],"recommendation":"major_revision","confidential_remarks":"The core projection construction and the conditional estimates are credible and publishable in principle, but the main theorem currently rests on Assumption 3.4, a substantial unproved regularity/positivity condition on the discrete solution. I would like to see either a proof of that bound or an explicit reframing of the contribution as a conditional a priori estimate. The numerical verification also needs attention because some of the reported c2 L2 rates are at machine precision. With those changes, the paper could be suitable for publication in this journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague—\n\nThe thing worth knowing: this is the first finite element semi-discrete error analysis for the DFN battery model in 2D and 3D, and the core technical contribution—the tensor-product projection operator PhΔr of (3.6)–(3.8) plus the radial trace estimate in Lemma 3.2—is genuinely new and plausibly fixes the defects in Bermejo's 1D analysis. His composite barycentric-plus-radial interpolant is indeed questionable for N≥2, and his Δr rates were suboptimal; here the projection gives clean h+(Δr)^2 rates for potentials, c1, c̄2, and the expected h+(Δr)^(2−q) rates for c2 in the H^q_r pencil. The proof scaffold is coherent: Lemmas 3.1–3.5 establish the projection and trace bounds, Lemma 3.6 couples the c2 error, and the Gronwall step in Theorem 3.9 closes. One small slip in Lemma 3.2's displayed chain (the H1 interpolation error of the Green's function G is O(Δr), not O((Δr)^2)), but the product with the H1_r projection error still yields O((Δr)^2), so the conclusion survives.\n\nThe soft spot the stress-test note flags is the real one: Theorem 3.9 is conditional on Assumption 3.4, uniform L∞ bounds on the discrete solution independent of h and Δr, and the paper gives no proof that the semi-discrete scheme (3.2)–(3.5) produces such solutions. The bound is used essentially—for Lipschitz continuity of J, Um, κi, for coercivity of the elliptic system in Lemma 3.7, and to close the coupled error inequality. It is not a consequence of Assumptions 3.1–3.3, and a maximum principle does not apply because the c1 source is non-monotone and coupled to the potentials. So the headline result is an a priori estimate conditional on discrete regularity, not a fully established convergence theorem. To the authors' credit, the assumption is stated openly, and the gap is addressable—a discrete invariant-region argument or a truncation would likely close it.\n\nTwo minor points. The 2D/3D well-posedness and regularity are imported from P2D references (Kroener; Díaz et al.) even though the introduction acknowledges the P4D case is open; Assumptions 3.1–3.2 are assumed, not derived. And the 3D h-rate in Table 3 comes from a single mesh pair, which the authors disclose; suggestive, not conclusive. The numerics otherwise match the theory: ~1 in h and ~2 in Δr across components, and the striking size gap between c2 L2 errors (~1e-12) and H1_r errors (~1e-5) is consistent with radial interpolation error scaling at Δr ≈ 1e-6.\n\nThis paper is for anyone doing numerical analysis of battery models or nonlinear coupled elliptic-parabolic systems with a singular radial variable. It deserves a serious referee, with the instruction to push on Assumption 3.4; if that bound can be proved or relaxed, this becomes the standard reference for DFN finite element convergence.","headline":"New tensor-product projection for the pseudo-(N+1)-D particle equation genuinely fixes Bermejo's defects, but the optimal h+(Δr)^2 theorem is conditional on an unproved uniform L∞ bound on the discrete solution (Assumption 3.4).","tokens_in":29900,"tokens_out":7509,"would_cite":true,"duration_ms":75555,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M15","35K55","35Q80"],"pacs":[],"model":"deepseek-v4-flash","headline":"Optimal finite element error rates for the Doyle-Fuller-Newman battery model are proved in two and three spatial dimensions.","keywords":["lithium-ion batteries","Doyle-Fuller-Newman model","finite element error analysis","semi-discrete approximation","pseudo-(N+1)-dimensional equation","projection operator","optimal convergence rates","elliptic-parabolic system"],"falsifier":"Run the 2D or 3D discretization under an aggressive discharge that drives the discrete electrolyte concentration toward zero or the particle concentration toward its upper limit as $h$ and $\\Delta r$ shrink; if the observed convergence order drops below the predicted $O(h)+O((\\Delta r)^2)$, the unproved uniform-bound assumption is doing load-bearing work.","tokens_in":28672,"feed_emoji":"🔋","tokens_out":12732,"duration_ms":113010,"temperature":0.7,"pith_summary":"This paper proves that a semi-discrete finite element discretization of the Doyle-Fuller-Newman (DFN) lithium-ion battery model converges at optimal rates in spatial dimensions 2 and 3, not only in the one-dimensional setting analyzed before. The target is the full coupled elliptic-parabolic system for electrolyte potential $\\varphi_1$, electrode potential $\\varphi_2$, electrolyte concentration $c_1$, and particle concentration $c_2$, including the pseudo-($N+1$)-dimensional radial diffusion equation. Under the paper's regularity assumptions, the error is $O(h)+O((\\Delta r)^2)$ for the potentials, electrolyte concentration, and surface concentration, while the particle concentration itself obeys $O(h+(\\Delta r)^{2-q})$ in the radial $H^q_r$ norm for $q=0,1$. The proof runs through a new projection operator for the pseudo-($N+1$)-dimensional equation and deliberately avoids a change of variables, which lets the analysis extend to non-isothermal settings. Numerical experiments in 2D+1D and 3D+1D with real battery parameters report convergence rates matching the theory.","feed_headline":"Optimal error rates proven for 2D and 3D battery simulations","feed_subtitle":"All four DFN unknowns converge at first order in mesh size and second order in particle radius","key_machinery":"The load-bearing device is a new projection operator $P_{h\\Delta r}$ for the pseudo-($N+1$)-dimensional diffusion equation, projecting into the tensor-product space of piecewise-constant functions in $x$ and piecewise-linear functions in $r$ through the weighted inner product $\\int_\\Omega\\int_0^R (b\\,\\partial_r w\\,\\partial_r v+\\lambda wv)r^2\\,dr\\,dx$, with $b=k_2$ and $\\lambda=1$. The operator factorizes as $P_{h\\Delta r}=P_hP_{\\Delta r}=P_{\\Delta r}P_h$ and, through a piecewise-constant-in-$x$ averaging operator, commutes with the radial derivative $\\partial_r$. That factorization yields clean approximation errors $O(h+(\\Delta r)^{2-q})$ in the radial $H^q_r$ norm and, crucially, $O(h+(\\Delta r)^2)$ for the radial trace at $r=R_s(x)$, which is exactly the surface concentration $\\bar c_2$ entering the Butler-Volmer kinetics. All subsequent estimates for the coupled elliptic-parabolic system are organized around this decomposition.","core_discovery":"The paper's central claim is Theorem 3.9: under Assumptions 2.1-2.2 and 3.1-3.4, the semi-discrete finite element solution converges at optimal order in every unknown. For almost every $t\\in[0,T]$, $\\|\\varphi_1-\\varphi_{1h}\\|_{L^2(0,t;H^1(\\Omega))}+\\|\\varphi_2-\\varphi_{2h}\\|_{L^2(0,t;H^1(\\Omega_2))}+\\|c_1-c_{1h}\\|_{L^2(0,t;H^1(\\Omega))}+\\|\\bar c_2-\\bar c_{2h}\\|_{L^2(0,t;L^2(\\Omega_2))}$ is bounded by $C(h+(\\Delta r)^2)$ plus initial-data projection errors, and $\\|c_2-c_{2h\\Delta r}\\|_{L^2(0,t;L^2(\\Omega_2;H^q_r))}$ is bounded by $C(h+(\\Delta r)^{2-q})$ for $q=0,1$. In words, refining the spatial mesh and the radial particle mesh both deliver their full expected accuracy simultaneously, with the surface concentration losing nothing and the interior particle concentration losing exactly one radial power in the radial $H^1$ norm. This is the first convergence analysis of this kind for the genuine 2D+1D and 3D+1D cases, and the paper backs it with convergence tables computed from real battery parameters.","pith_inferences":["The paper leaves Assumption 3.4 as a hypothesis rather than deriving it, and it cites P2D analyses for the existence of regular weak solutions while noting the P4D case remains a gap; supplying either ingredient would make the theorem unconditional.","The projection operator construction is likely transferable to other multiscale equations coupling a macroscale coordinate with a radial microscale coordinate, such as thermal-electrochemical models, with the weight and shift chosen for the relevant operator.","A practical corollary not tested here is mesh balancing: choosing $\\Delta r\\approx\\sqrt{h}$ would equilibrate the spatial and radial error contributions, though the optimal constant would need numerical tuning.","A natural stress test is to push the model outside the stated assumptions, for example very high discharge, near-zero electrolyte concentration, or strong thermal gradients, and check whether the convergence rates degrade exactly where Assumption 3.4 fails."],"forward_implications":["If the theorem is correct, uniform refinement of the spatial mesh and the radial mesh in a DFN battery simulation yields the full expected first-order and second-order accuracy in 2D and 3D without special radial meshing.","The simultaneous error control means coupled quantities such as electrode reaction current and surface concentration inherit the same $O(h+(\\Delta r)^2)$ accuracy in the norms stated.","Because the analysis avoids the change of variables used in earlier work, the same error estimates carry over to thermally coupled DFN models with non-uniform temperature.","The numerical verification using real battery parameters shows the predicted rates are observable in practice, supporting the use of piecewise-linear-in-$r$, piecewise-constant-in-$x$ discretizations in production codes."],"supporting_citations":[{"why":"previous finite element convergence analysis in one dimension whose auxiliary approximator and trace inequality the paper improves and extends.","marker":"Bermejo (2021)"},{"why":"original statement of the Doyle-Fuller-Newman model that defines the equations being discretized.","marker":"Doyle et al. (1993)"},{"why":"model derivation and simulation framework that supplies the battery parameter structure and nonlinear kinetics.","marker":"Fuller et al. (1994)"},{"why":"local well-posedness and Lipschitz lemmas for the P2D system used in the nonlinear error estimates.","marker":"Kroener (2016)"},{"why":"well-posedness of the P2D model cited for the regularity of weak solutions assumed in higher dimensions.","marker":"Díaz et al. (2019)"},{"why":"standard parabolic projection approximation theory used to prove the radial projection error estimates.","marker":"Thomée (2007)"},{"why":"radial trace and commutation results used to make the projection commute with the radial derivative.","marker":"Schreiber and Eisenstat (1981)"},{"why":"real battery parameters and asymptotic reduction used in the 2D and 3D numerical verification.","marker":"Timms et al. (2021)"},{"why":"piecewise $H^2$ projection estimate for discontinuous coefficients used for the electrolyte concentration projection.","marker":"Xu (1982)"}],"fun_headline_variants":["Optimal error rates proven for 2D and 3D battery simulations","2D/3D battery model hits optimal error bounds","First optimal convergence proof for 2D/3D DFN","Novel projector yields optimal FE rates for DFN in 2D/3D","Optimal error rates for full Doyle-Fuller-Newman in 2D/3D"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes the computed discrete solution stays bounded and away from the physical limits (zero concentration, full particle concentration, unbounded potentials) uniformly as the mesh is refined; this regularity is assumed, not proved, and the error bound collapses without it.","fun_headline_variants_meta":{"raw":{"variants":["Optimal error rates proven for 2D and 3D battery simulations","2D/3D battery model hits optimal error bounds","First optimal convergence proof for 2D/3D DFN","Novel projector yields optimal FE rates for DFN in 2D/3D","Optimal error rates for full Doyle-Fuller-Newman in 2D/3D"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000486,"raw_usage":{"total_tokens":2420,"prompt_tokens":992,"completion_tokens":1428,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":608,"completion_tokens_details":{"reasoning_tokens":1327}},"tokens_in":608,"tokens_out":1428,"duration_ms":10745,"temperature":1.0,"reasoning_tokens":1327,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:21:12.716087+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the 2D or 3D discretization under an aggressive discharge that drives the discrete electrolyte concentration toward zero or the particle concentration toward its upper limit as $h$ and $\\Delta r$ shrink; if the observed convergence order drops below the predicted $O(h)+O((\\Delta r)^2)$, the unproved uniform-bound assumption is doing load-bearing work.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"previous finite element convergence analysis in one dimension whose auxiliary approximator and trace inequality the paper improves and extends."},{"cited_title":"F., and Newman, J","cited_arxiv_id":null,"evidence_quote":"original statement of the Doyle-Fuller-Newman model that defines the equations being discretized."},{"cited_title":"F., Doyle, M., and Newman, J","cited_arxiv_id":null,"evidence_quote":"model derivation and simulation framework that supplies the battery parameter structure and nonlinear kinetics."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"local well-posedness and Lipschitz lemmas for the P2D system used in the nonlinear error estimates."},{"cited_title":"and Eisenstat, S","cited_arxiv_id":null,"evidence_quote":"radial trace and commutation results used to make the projection commute with the radial derivative."},{"cited_title":"G., Sulzer, V., Please, C","cited_arxiv_id":null,"evidence_quote":"real battery parameters and asymptotic reduction used in the 2D and 3D numerical verification."}],"review_version":1}