{"id":"63caab10-1bf8-4f68-83fe-900d1e155935","arxiv_id":"1908.08772","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For strictly monotone, piecewise spatially constant flux, monotone upwind finite volume schemes converge to the entropy solution at an O(sqrt(Delta x)) rate in L1.","lead":"This paper proves that monotone upwind finite volume schemes for scalar conservation laws with a discontinuous, nonlinear flux converge at the square-root mesh size in L1. It is the first proof of any convergence rate for this class of numerical methods.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The pivotal gap is Lemma 4.7: the exact boundary trace's temporal BV is asserted to pass from the discrete lemma to the limit, but trace continuity under L1+BV convergence is not automatic; J2 in Theorem 4.10 depends on it.","rationale":"The central theorem is proven by splitting at the interface and separately estimating errors on R- and R+. The R- estimate (Theorem 4.4) is self-contained. The R+ estimate (Theorem 4.10) is the coupling step: its boundary term C measures the difference between exact and numerical interface fluxes. H1 is controlled by the R- estimate (4.7) and Lemma 4.6; J2 is controlled solely by Lemma 4.7, which asserts spatial Lipschitz of f(u) in L1_t. If Lemma 4.7's hypothesis on the exact boundary trace's temporal BV is not proved, J2 is uncontrolled and the rate on R+ does not follow. The paper's own note in Lemma 4.7 flags precisely this missing passage. The reader's weakest_assumption identified the same spot. I do not see a counterexample to the theorem; the missing passage can likely be supplied via the two-dimensional BV argument sketched above, which is why the verdict stays CONDITIONAL rather than REJECT or UNVERDICTED. The numerical experiments are consistent but do not settle trace convergence. Thus the single most load-bearing concern is the unproved limit passage in Lemma 4.7.","tokens_in":20878,"tokens_out":24109,"duration_ms":252136,"concrete_test":"Fill the gap in Lemma 4.7 by proving (i) the exact entropy solution of (4.1) has u∈BV(R×[0,T]) (e.g. via TV_x(u(·,t))≤TV(u0) and u_t=-g(u)_x as a measure), hence TV_t(u(0-,·))≤C TV(u0); (ii) from the Rankine-Hugoniot condition, TV_t(f(u(0+,·)))≤C TV(u0); and (iii) the numerical traces uΔt(0-,·) converge to u(0-,·) in L1(0,T) using the rate estimate (4.7). If step (iii) cannot be established, the J2 bound in Theorem 4.10 is invalid and the proof of Theorem 5.1 must be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Lemma 4.7 the authors state that the temporal total variation bound of u(0+,t) \"follows from Lemma 4.6 on a discrete level and carries over in the limit.\" This is the load-bearing step for the R+ estimate. Lemma 4.7 is used in Theorem 4.10 to control the J2 term |f(u(0+,s))-f(u(y,s))| by C|y|, which after integration gives the Cε term. Without it, the boundary term C in the Kuznetsov inequality on R+ is not bounded and no sqrt(dx) rate on R+ (hence Theorem 5.1) follows. The limit passage is not automatic: uniform BV_x plus strong L1 convergence does not imply convergence of one-sided traces (e.g. v_n=1_{[0,1/n]} converges to 0 in L1 with BV=1, but trace at 0+ stays 1). The exact trace can be handled by a direct argument: for the entropy solution on R-, u_x and u_t=-g(u)_x are bounded measures, so u∈BV(R×(0,T)) and the trace u(0-,·) has BV_t; then f(u(0+,t)) has BV_t by Rankine-Hugoniot. But this argument is not supplied, and the paper instead relies on the numerical trace convergence. The self-referential note in Lemma 4.7 admits the missing support. This is a proof gap, not a demonstrated counterexample.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves an O(sqrt(Delta x)) L1 convergence rate for a class of monotone finite volume schemes approximating scalar conservation laws with discontinuous flux, under the assumptions that the flux is strictly monotone in the unknown and the spatial coefficient is piecewise constant with finitely many discontinuities. The proof decomposes the problem across the flux discontinuities into finitely many initial-boundary value problems, establishes Kuznetsov-type error estimates for the half-line and bounded-interval cases, and assembles these into Theorem 5.1. The appendix extends the technique to general initial-boundary value problems with prescribed boundary data. If correct, this is the first convergence-rate proof for nonlinear schemes in the discontinuous-flux setting.","tokens_in":21144,"tokens_out":5434,"duration_ms":55332,"significance":"The main theorem is significant: for nonlinear conservation laws with discontinuous flux, no convergence-rate result for numerical schemes was previously available, and the decomposition strategy plus the discrete temporal-total-variation bound are potentially reusable tools. The paper also gives a self-contained Kuznetsov framework for half-line problems and extends the rate to general IBVPs, which is of independent interest. The authors are appropriately careful that the rate sqrt(Delta x) is optimal in the absence of extra assumptions, citing the classical Sabac example. However, the proof as written has a load-bearing gap in Lemma 4.7, and one further reduction in Section 2 is asserted rather than shown.","major_comments":[{"comment":"The Lipschitz bound ∫_0^T |f(u(x,t))-f(u(y,t))| dt ≤ C|x-y| is obtained by applying [31, Lem. 4] to w=f(u), and the required temporal total-variation bound of the boundary trace u(0+,t) is said to 'follow from Lemma 4.6 on a discrete level and carries over in the limit.' That limit passage is not demonstrated, and it is not automatic: uniform BV_x plus strong L1 convergence does not imply convergence of one-sided traces, as the sequence v_n = 1_{[0,1/n]} converges to 0 in L1 with BV=1 while the trace at 0+ stays 1. Since the J2 term in Theorem 4.10 is estimated by Cε exactly through Lemma 4.7, and J2 controls the boundary contribution on R+, this gap is load-bearing for Theorem 4.10 and hence for Theorem 5.1. The authors should supply a direct argument for the exact entropy-solution trace (for example, from BV regularity in (x,t) on R− combined with the Rankine–Hugoniot condition) or prove trace convergence of the numerical solutions by an explicit estimate.","section":"Section 4.2, Lemma 4.7"},{"comment":"The paragraph after (2.4) asserts that if u(0) and u(i) are entropy solutions of the respective IBVPs, then the composite u = Σ_i u(i) is the entropy solution of (1.1) 'by adding the entropy inequalities of u(i) and choosing the respective constant in each entropy inequality in accordance with (2.1).' This reduction is used in Theorem 5.1 to obtain the global rate from rates on each subdomain, so the summation must be shown explicitly. In particular, the trace terms at each interface ξ_i involve limits from both sides and must combine correctly under the discrete Rankine–Hugoniot condition; the authors should display this calculation so that no residual interface term remains.","section":"Section 2, composite solution assertion"}],"minor_comments":[{"comment":"There are numerous typographical errors, for example 'ra tes' in the abstract and 'conser v a tion' in the title; the manuscript needs a careful proofreading pass.","section":"Abstract and title"},{"comment":"The notation φ^n = φ(x,t_n,y,s) and φ_{j+1/2} = φ(x_{j+1/2},t,y,s) is introduced but the arguments of the mollifiers are not always written consistently in the sums; please clarify and check all occurrences.","section":"Lemma 4.9 proof"},{"comment":"The observed orders of convergence in both experiments exceed 1 for the finest grids (1.28 and 1.30), which is not predicted by Theorem 5.1; the authors should comment that these are pre-asymptotic effects or otherwise explain the discrepancy.","section":"Table 1"},{"comment":"The generalization to arbitrary BV boundary data is plausible, but the modified Lemma 4.6 is only described ('should then read') rather than proved, and the dependence of the constants on TV(a) is not tracked; please provide the proof or at least a precise statement with explicit constants.","section":"Appendix A"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the scope of the journal and the main result is significant. I do not see a circularity problem: the final choice ε = ε0 = sqrt(Delta x) is a standard balance of the explicit estimates, not a fitting of constants. The decisive issue is the unproved limit passage in Lemma 4.7; this is a proof gap that can likely be repaired within the manuscript's scope by supplying a direct trace-regularity argument. I recommend major revision, with acceptance conditional on closing that gap and on making the Section 2 gluing argument explicit."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere's my read of Badwaik–Ruf. The headline result is genuinely new: Theorem 5.1 gives the first O(sqrt(Delta x)) rate for monotone upwind schemes for scalar conservation laws with discontinuous, nonlinear flux, in the strictly monotone, piecewise-constant coefficient case. The proof strategy is sensible: split at flux discontinuities, treat each interval as an IBVP with flux-matched boundary data, glue via discrete Rankine–Hugoniot. The new ingredient is the temporal TV bound on the numerical solution at interface ghost cells (Lemma 4.6), which feeds into boundary terms in the Kuznetsov machinery. That part is detailed and I found no algebra issues.\n\nThe paper is not quite complete as written. Two passages are under-verified. The gluing in Section 2 is a sketch, but in this flux-monotone setting it is believable and likely works. More important is Lemma 4.7. The exact trace u(0+,t) is asserted to have bounded temporal TV because the discrete trace does and 'carries over in the limit.' That is not automatic from L1 convergence plus uniform BV_x; the stress-test example (v_n = 1_{[0,1/n]} -> 0) shows why. The paper itself flags the missing support. The good news is that a direct argument exists: on the left half-line, u is the entropy solution and lies in BV(R x (0,T)), so the trace u(0-,t) has bounded time variation; the Rankine–Hugoniot matching f(u(0+,t)) = g(u(0-,t)) transfers this to the right trace. That patch is short and does not disturb the rest. So the gap is real but repairable.\n\nMinor: numerical experiments have no code or data, and the scheme is non-conservative at the interface (the rate implies the defect vanishes). The citation pattern is honest: self-citations supply auxiliary facts, not the target rate.\n\nNet: the central claim is likely true and this is the first rate for this setting once Lemma 4.7 is patched. A serious referee should engage. It belongs in the literature after revision expanding the gluing and trace passages. I'd take it to reading group and would cite the revised version. Send it to peer review.","headline":"First rate proof for nonlinear discontinuous flux is plausible, but Lemma 4.7's trace-regularity passage needs a patch before the claim is fully certified.","tokens_in":21696,"tokens_out":2811,"would_cite":true,"duration_ms":25213,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35L65","65M08","65M12","35R05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Monotone finite volume schemes for conservation laws with a discontinuous, strictly monotone flux converge at the rate O(√Δx) in L1.","keywords":["scalar conservation laws","discontinuous flux","finite volume methods","convergence rate","monotone schemes","doubling-of-variables estimates","initial-boundary value problems","temporal total variation"],"falsifier":"Check the temporal total variation of $u(0+,t)$ for the two-flux problem with $g(u) = u$ and $f(u) = u^2/2$ and a Riemann-type initial datum that sends a shock into the interface. If this quantity is unbounded, Lemma 4.7's regularity premise fails and the proof of Theorem 4.10 collapses, so the rate bound is not established.","tokens_in":20634,"feed_emoji":"📉","tokens_out":15700,"duration_ms":126787,"temperature":0.7,"pith_summary":"This paper proves that monotone upwind finite volume schemes for scalar conservation laws with a flux that is strictly monotone in the conserved variable and a spatial coefficient that is piecewise constant with finitely many jumps converge to the unique entropy solution at the rate $O(\\sqrt{\\Delta x})$ in $L^1$. If true, this is the first convergence-rate proof for a numerical method in the nonlinear discontinuous-flux setting, and it matches the optimal rate known for conservation laws with a constant-in-space flux. The proof cuts the domain at the discontinuities of the coefficient, turning the problem into finitely many initial-boundary value problems with a flux that no longer depends on $x$, and controls the new interface terms with a discrete flux-matching condition and a temporal total variation bound. If the theorem holds, the error of the scheme is bounded uniformly in the grid size, which provides the quantitative error control needed for adaptive mesh refinement and multilevel Monte Carlo methods.","feed_headline":"Monotone schemes converge at √Δx for discontinuous fluxes","feed_subtitle":"The √Δx rate matches the optimal homogeneous-case rate and enables error-aware mesh and Monte Carlo design.","key_machinery":"The central object is the decomposition of the discontinuous-flux problem into a finite set of initial-boundary value problems on the intervals between the discontinuities of $k$, coupled by the discrete flux-matching condition $u^{n+1}_{P_i} = (f^{(i)})^{-1}\\bigl(f^{(i-1)}(u^{n+1}_{P_i-1})\\bigr)$ at each interface. The argument is carried by a comparison lemma using the doubling of variables with a mollified test kernel, which estimates the $L^1$ difference between the exact and numerical solutions in terms of moduli of continuity, the grid parameters, and an entropy-dissipation term. The new ingredient that makes the boundary terms manageable is the bound on the temporal total variation of the numerical solution, $\\sum_n \\lvert u^{n+1}_j - u^n_j \\rvert \\le C\\, \\mathrm{TV}(u_0)$, uniform in the cell index; together with the strict monotonicity of the flux this gives the Lipschitz-in-space regularity of the exact solution's flux (Lemma 4.7), which closes the estimate of the interface terms.","core_discovery":"The central claim is that the finite volume approximation defined by (3.1) satisfies $\\lVert u(\\cdot,T) - u_{\\Delta t}(\\cdot,T) \\rVert_{L^1(\\mathbb{R})} \\le C \\sqrt{\\Delta x}$ whenever the flux is strictly monotone, the coefficient $k$ is piecewise constant with finitely many discontinuities, and the initial datum lies in $L^1 \\cap BV(\\mathbb{R})$. The proof obtains this by decomposing the entropy solution and the numerical solution on the intervals between consecutive discontinuities of $k$, where the restriction of the exact solution solves an initial-boundary value problem with a spatially homogeneous flux. The discretization enforces the flux-matching condition (the Rankine–Hugoniot condition) across each discontinuity on the discrete level, which reduces the global error to a sum of errors on these subdomains. For each subdomain, a standard doubling-of-variables comparison estimate bounds the $L^1$ error, with the interface terms controlled by a new bound on the temporal total variation of the numerical solution (Lemma 4.6) and by a Lipschitz-in-space property of the flux of the exact solution (Lemma 4.7). Together these ingredients give the stated $\\sqrt{\\Delta x}$ rate on the whole line.","pith_inferences":["If the temporal total variation of the interface trace is not uniformly bounded for some admissible data, the rate may degrade or the scheme may fail to reach $\\sqrt{\\Delta x}$ order; this is a testable gap in the proof rather than a demonstrated counterexample.","A Wasserstein-distance analogue of this rate, analogous to the homogeneous case, may be provable with the same decomposition and temporal-total-variation control; the paper names this as a future direction.","The strict monotonicity assumption on the flux is probably not essential for the decomposition idea; non-monotone fluxes could be split into monotone branches, though the interface coupling would then require a new matching condition."],"forward_implications":["The scheme (3.1) converges to the entropy solution with an error of at most $C\\sqrt{\\Delta x}$, giving the first convergence-rate statement for nonlinear conservation laws with discontinuous flux.","For each fixed number of discontinuities, the global $L^1$ error on the whole real line obeys the same $\\sqrt{\\Delta x}$ bound as the error on any single subdomain, so the rate does not degrade when summing across interfaces.","The same proof strategy yields an $O(\\sqrt{\\Delta x})$ error bound for general initial-boundary value problems whose boundary data lie in $L^1\\cap BV(0,T)$, improving on the previously claimed $O(\\Delta x^{1/3})$ for that setting.","The $\\sqrt{\\Delta x}$ rate is optimal without extra assumptions on the initial datum, since the standard optimality construction for the homogeneous case can be placed away from the last discontinuity."],"supporting_citations":[{"why":"Supplies the accuracy estimate for monotone schemes that the proof adapts to the initial-boundary setting.","marker":"[23]"},{"why":"Provides the discrete entropy inequality for monotone difference approximations used to control the numerical solution.","marker":"[10]"},{"why":"Standard reference for moduli of continuity, TVD properties, and the estimates for the A and B terms in the comparison lemma.","marker":"[15]"},{"why":"Alternative source for the A/B estimates in the comparison lemma on the half-line.","marker":"[8]"},{"why":"Gives the temporal-total-variation bound and the Lipschitz regularity of the flux for solutions on bounded domains, used in Lemma 4.7 and Theorem 4.10.","marker":"[31]"},{"why":"Establishes existence and uniqueness of the entropy solutions for strictly monotone discontinuous flux, giving the solution framework used in the decomposition.","marker":"[6]"},{"why":"Provides the optimality example for the homogeneous case that the paper adapts to argue the $\\sqrt{\\Delta x}$ rate cannot be improved.","marker":"[34]"}],"fun_headline_variants":["First proof of √Δx convergence for discontinuous flux","√Δx rate proven for monotone discontinuous-flux schemes","Optimal √Δx error for monotone schemes with discontinuous flux","First rate for nonlinear discontinuous flux: √Δx","Monotone schemes hit √Δx rate despite flux discontinuities"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes that the boundary trace $u(0+,t)$ of the exact solution on each interface has bounded temporal total variation, so that $w = f(u)$ is Lipschitz in $x$ with values in $L^1(0,T)$; the paper states this follows from the discrete bound of Lemma 4.6 'and carries over in the limit' without demonstrating the limit passage.","fun_headline_variants_meta":{"raw":{"variants":["First proof of √Δx convergence for discontinuous flux","√Δx rate proven for monotone discontinuous-flux schemes","Optimal √Δx error for monotone schemes with discontinuous flux","First rate for nonlinear discontinuous flux: √Δx","Monotone schemes hit √Δx rate despite flux discontinuities"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001302,"raw_usage":{"total_tokens":5302,"prompt_tokens":931,"completion_tokens":4371,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":4287}},"tokens_in":547,"tokens_out":4371,"duration_ms":28528,"temperature":1.0,"reasoning_tokens":4287,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:29:51.295139+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the temporal total variation of $u(0+,t)$ for the two-flux problem with $g(u) = u$ and $f(u) = u^2/2$ and a Riemann-type initial datum that sends a shock into the interface. If this quantity is unbounded, Lemma 4.7's regularity premise fails and the proof of Theorem 4.10 collapses, so the rate bound is not established.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the accuracy estimate for monotone schemes that the proof adapts to the initial-boundary setting."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the discrete entropy inequality for monotone difference approximations used to control the numerical solution."},{"cited_title":"Holden and N","cited_arxiv_id":null,"evidence_quote":"Standard reference for moduli of continuity, TVD properties, and the estimates for the A and B terms in the comparison lemma."},{"cited_title":"Coclite, J","cited_arxiv_id":null,"evidence_quote":"Alternative source for the A/B estimates in the comparison lemma on the half-line."},{"cited_title":"Ridder and A","cited_arxiv_id":null,"evidence_quote":"Gives the temporal-total-variation bound and the Lipschitz regularity of the flux for solutions on bounded domains, used in Lemma 4.7 and Theorem 4.10."},{"cited_title":"Baiti and H","cited_arxiv_id":null,"evidence_quote":"Establishes existence and uniqueness of the entropy solutions for strictly monotone discontinuous flux, giving the solution framework used in the decomposition."},{"cited_title":"S ¸abac, The optimal convergence rate of monotone finite difference me thods for hyperbolic conservation laws, SIAM Journal on Numerical Analysis, 34 (1997), pp","cited_arxiv_id":null,"evidence_quote":"Provides the optimality example for the homogeneous case that the paper adapts to argue the $\\sqrt{\\Delta x}$ rate cannot be improved."}],"review_version":1}