{"id":"e7b4264d-7577-4772-b859-9aac7864178d","arxiv_id":"2411.17175","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Exponential surface diffusion flows on R are globally well-posed for small initial slopes, and their large-time rescaled profiles converge to those of the classical linearized surface diffusion flow.","lead":"This paper proves global existence, uniqueness, and large-time self-similar asymptotics for a nonlinear exponential surface diffusion flow on the real line, under small initial slope data. It is the first rigorous result for this Mullins-type flow and shows that rescaled solutions converge to the known self-similar profiles of the linearized surface diffusion equation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof that every subsequential limit lies in Koch–Lamm's X∞ uniqueness class is asserted via a Hölder inequality that is misstated (t^{-1/4} instead of t^{1/4}) and never derived, so full convergence of u_σ is not yet established.","rationale":"I agree with the reader's weakest_assumption. The most load-bearing condition for Theorem 1.2 is the verification that every subsequential limit U belongs to the Koch–Lamm uniqueness class: without that, the compactness argument only gives convergence along subsequences, not the asserted full convergence of u_σ. The paper's one-line Hölder estimate is not merely terse; it contains a sign error in the time power (t^{-1/4} instead of t^{1/4}) and the resulting X∞ bound is not derived. This is a genuine proof gap, but it is fixable: (6.5) does control t^{1/4}||U_{xx}||_{L∞}, and the parabolic-cylinder computation shows ||U||_{X∞}≤Cε, so choosing ε* small relative to ρ* closes the gap. I also note a separate, concrete error in (6.6): the assertion lim_{σ→∞} f_σ(r)=r is false when f(0)≠0, which is the case for the main example f(r)=e^r; indeed f_σ(r)=σf(σ^{-1}r)→σf(0)+r. This is also fixable by observing that only the derivative of f_σ enters the PDE, and f_σ'(r)=f'(σ^{-1}r)→f'(0)=1, or by replacing f_σ with f_σ-σf(0). Because both issues are repair gaps and do not invalidate the main construction, the conditional verdict remains appropriate.","tokens_in":24491,"tokens_out":26407,"duration_ms":222229,"concrete_test":"Independently derive ||U||_{X∞} from (6.5) in the case n=1. For any R>0 and x0∈R, (6.5) with the correct scaling implies |U_{xx}(y,t)|≤Cε t^{-1/4} on (R^4/2,R^4)×B_R(x0), hence (∫∫|U_{xx}|^7)^{1/7}≤Cε R^{-2/7}, and after multiplication by R^{2/7} the X∞ contribution is ≤Cε. Verify this algebra, and then verify that choosing ε*=min(ε0,ρ*/C) makes every subsequential limit satisfy ||U||_{X∞}<ρ*. If the estimate yields a bound below ρ*, the uniqueness argument is sound; if not, Theorem 1.2 should be weakened to convergence along subsequences.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1.2 concludes that the whole family u_σ converges, not just subsequences. This requires every subsequential limit U to lie in the Koch–Lamm uniqueness class, i.e. ||U||_{X∞}<ρ* ([15, Theorem 3.4]). The proof attempts to obtain this from (6.5): ||U_x||'_{BC^{k+μ,(k+μ)/4}(R×(t/2,t))}≤C||(u0)_x||_{W^{1,∞}}. It then states, 'by the Hölder inequality', that ||U||_{X∞}≤C sup_{t>0}(||U_x||_{L∞}+t^{-1/4}||U_{xx}||_{L∞})(t). This displayed inequality is not correct as written: (6.5) gives t^{1/4}||U_{xx}||_{L∞}≤Cε, so t^{-1/4}||U_{xx}||_{L∞} is typically unbounded near t=0 (for a self-similar profile it grows like t^{-1/2}). The intended bound on ||U||_{X∞} can be recovered by using t^{1/4} and integrating over parabolic cylinders of size R^4, but that computation is absent. Moreover, the constant C is not tracked through the passage u_σ→U, and the choice ε*<ρ*/C is only asserted as 'sufficiently small'. Until this uniqueness-class verification is written out, the central claim of full convergence is not fully supported; if Cε* were not below ρ*, the argument would yield only subsequential convergence.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the exponential surface diffusion flow V = ∂_s^2 f(-κ) for graphs u(x,t) on R, where f is strictly increasing and f'(0)=1. It establishes global-in-time existence in Hölder spaces for initial data whose first two derivatives are small, with weighted decay estimates for all derivatives. It then studies the rescaling u_σ(x,t)=σ^{-1}u(σx,σ^4t) and claims convergence, uniformly on compact subsets of R×[0,∞), to the unique self-similar solution of the linearized surface diffusion equation (1.11), with derivatives converging on compact subsets of R×(0,∞). The proof rewrites the slope equation as a perturbed biharmonic heat equation, proves global linear decay estimates in weighted scaled Hölder norms, propagates smallness nonlinearly, and passes to the limit by Ascoli–Arzelà.","tokens_in":1597,"tokens_out":2103,"duration_ms":435089,"significance":"If correct, this is the first global well-posedness and large-time asymptotic profile theorem for the exponential surface diffusion flow on the whole line, going beyond the conventional linear case studied by Koch–Lamm and Du–Yip. The physical motivation, justifying Mullins grooving directly from the Gibbs–Thomson law without linearizing f near zero, is clearly explained. A particular strength is that the central decay estimate is a genuine a priori estimate against the biharmonic heat equation, with no fitting parameters and no circular use of the conclusion. The proof is detailed and the weighted Hölder framework is well adapted to the fourth-order quasilinear problem. The main caveat is a gap in the final uniqueness-class verification needed for full, rather than subsequential, convergence in Theorem 1.2.","major_comments":[{"comment":"The argument that every subsequential limit U lies in the Koch–Lamm uniqueness class is incomplete. The manuscript invokes the display ||U||_{X∞} ≤ C sup_{t>0}(||U_x||_{L∞} + t^{-1/4}||U_{xx}||_{L∞})(t) 'by the Hölder inequality', but this inequality is not correct as written: estimate (6.5) gives t^{1/4}||U_{xx}||_{L∞(R×(t/2,t))} ≤ C||(u0)_x||_{W^{1,∞}}, not t^{-1/4}, and for a self-similar profile the quantity t^{-1/4}||U_{xx}||_{L∞} is typically unbounded near t=0. The intended bound on ||U||_{X∞} can be recovered by using t^{1/4}||U_{xx}|| and integrating the pointwise bound over parabolic cylinders of radius R^4, but this computation is not written down. In addition, the constant C is not tracked through the passage u_σ→U, and the choice of ε* satisfying Cε*<ρ* is only asserted as 'sufficiently small'. Since this step upgrades subsequential convergence to convergence of the full family u_σ in Theorem 1.2, it is load-bearing and needs to be supplied.","section":"Section 6, proof of Theorem 1.2, after Eq. (6.5)"}],"minor_comments":[{"comment":"The sentence 'Hence lim_{σ→∞} ||u_σ−U||_{L∞(K)} = 0 for every compact set σ → ∞' is garbled; it should read 'for every compact set K ⊂ R×[0,∞)'.","section":"Section 6, proof of Theorem 1.2, last displayed line"},{"comment":"The displayed chain is missing a factor σ^3 on the middle right-hand side: from (5.10) one obtains σ^3||u_t||'... ≤ C||(u0)_x|| σ^3(σ^4t)^{-1/2}(1+σ^4t)^{-1/4}, which simplifies to the stated final bound ≤ C||(u0)_x|| t^{-3/4}; the intermediate inequality as written is not literally correct.","section":"Section 6, Eq. (6.2)"},{"comment":"The sentence 'By induction, we obtain v ∈ Z^k_T for all k ∈ Z^k_T for all k ∈ Z≥4' contains a typo and should read 'for all k ∈ Z with k ≥ 4'.","section":"Lemma 5.4, end of induction proof"},{"comment":"Under the stated fixed-k assumption, the proof yields U ∈ C^{k+1+μ,(k+1+μ)/4}_{time loc}(R×(0,∞)) rather than U ∈ C^∞(R×(0,∞)); the C^∞ conclusion requires either an additional smoothness argument allowing arbitrary k, or a weakened regularity statement.","section":"Theorem 1.2 statement and proof"},{"comment":"The symbol III is used both for the Duhamel integral ∫_{t̃}^{t} ∂_x^2 e^{-(t-s)∂_x^4}F(s)ds and, implicitly, for the function to which Proposition 2.6 is applied; since the two coincide only after identifying ∂_x^2 e^{-(t-s)∂_x^4}F with e^{-(t-s)∂_x^4}F_{xx}, a clarifying sentence would remove ambiguity.","section":"Lemma 4.2 proof"}],"recommendation":"major_revision","confidential_remarks":"The main theorem is plausible and the technical framework is sound, but the final uniqueness-class verification in Section 6 is a genuine gap in the proof of full convergence. The gap appears fixable by a short calculation using the correct t^{1/4} exponent and explicit constants, but it must be written out before the central claim of Theorem 1.2 is considered proved. I did not find evidence of circularity or parameter-fitting; the concern is purely about a missing estimate in the limit passage."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is the first global well-posedness and large-time self-similar convergence result for the exponential surface diffusion flow (1.1), and the main proof line is sound. The authors write v=u_x as a perturbed biharmonic heat equation, introduce weighted Hölder norms tailored to break the scaling invariance caused by the nonlinear curvature term, prove global linear decay estimates, close the nonlinear smallness, then pass to the limit. For f(r)=e^r this is genuinely new: earlier surface diffusion results (Koch–Lamm, Du–Yip) treated the linear flow f(r)=r, and the second-order exponential flow needed viscosity methods. The weighted Hölder framework is the paper's main technical contribution and I expect it to be reusable.\n\nSoft spots, in proportion. (1) The proof that the whole family u_σ converges, not just subsequences, depends on putting the limit U in Koch–Lamm's X∞ uniqueness class. The paper derives ∥U∥_{X∞} ≤ C sup_t(∥U_x∥_∞ + t^{-1/4}∥U_xx∥_∞) from (6.5) \"by the Hölder inequality\". That exponent is wrong as written: (6.5) gives t^{1/4}∥U_xx∥_∞ ≤ Cε, so t^{-1/4}∥U_xx∥_∞ generally blows up near t=0. The correct estimate is likely recoverable by integrating t^{1/4}∥U_xx∥_∞ over parabolic cylinders, but that computation plus constant tracking is absent. If Cε* were not below ρ*, the argument would only give subsequential convergence. This is a fixable gap, not a fatal one. (2) The display f_σ(r)=σ f(σ^{-1}r)→r in (6.6) is not literally true unless f(0)=0; for f(r)=e^r, f_σ diverges by a constant. Since only derivatives of f_σ matter, the limiting equation is still correct after subtracting f(0), but the statement should be repaired. (3) Minor: the honest comparison with Du–Yip in Remark 1.5 is good to see, and the convergence is only on compact sets, not uniformly in R.\n\nWho is this for: geometric analysts and anyone working on fourth-order quasilinear parabolic equations. The core theorem is new and the main estimates appear solid. I would send it to peer review; the referee should ask for a written proof of the uniqueness-class verification and a correction of the f_σ display, but not for a rewrite.","headline":"First global well-posedness and large-time self-similar convergence for exponential surface diffusion flows; the main proof is sound, with a fixable gap in the uniqueness-class verification.","tokens_in":25362,"tokens_out":6700,"would_cite":true,"duration_ms":55826,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35K55","35K35","35B40"],"pacs":[],"model":"deepseek-v4-flash","headline":"For small initial slopes, the exponential surface diffusion flow has a unique global classical solution, and its rescaled profiles converge to the unique self-similar profile of the linearized flow.","keywords":["exponential surface diffusion","surface diffusion flow","global well-posedness","self-similar solutions","large-time asymptotics","biharmonic heat equation","parabolic Hölder spaces","thermal grooving"],"falsifier":"A numerical experiment would settle the central claim: solve (1.1) with $f(r)=e^r$ and initial data $u_0(x)=a x_+ + b x_-$ for small $a,b$, evolve to large times, form $u_\\sigma(x,t)=\\sigma^{-1}u(\\sigma x,\\sigma^4 t)$, and compare on a fixed compact set with the unique self-similar solution $U$ of (1.11) with initial data $a x_+ + b x_-$; two different subsequential limits, or a limit failing (1.11), would refute the convergence claim.","tokens_in":24282,"feed_emoji":"🌊","tokens_out":15110,"duration_ms":131472,"temperature":0.7,"pith_summary":"The paper studies the exponential surface diffusion flow $V=\\partial_s^2 f(-\\kappa)$ for a curve given as a graph over the whole real line, with $f$ a strictly increasing smooth function, typically $f(r)=e^r$. It establishes that if the first and second derivatives of the initial profile are bounded and sufficiently small, a unique classical solution exists for all time. It then proves that the rescaling $u_\\sigma(x,t)=\\sigma^{-1}u(\\sigma x,\\sigma^4 t)$ converges uniformly on compact sets to the unique self-similar solution of the linearized surface diffusion equation $V=-f'(0)\\kappa$, with derivatives converging on compact sets away from $t=0$. This matters because it justifies the thermal-grooving explanation derived from the Gibbs–Thomson law without first linearizing the curvature dependence of $f$ near $\\kappa=0$.","feed_headline":"Small-slope crystal surfaces flow to a linear self-similar profile","feed_subtitle":"For exponential surface diffusion, small initial slopes force convergence to the linearized flow's unique self-similar profile.","key_machinery":"The proof differentiates the equation once and works with $v=u_x$, rewriting the flow as $v_t=((1-\\alpha)v_{xx}+F)_{xx}$, where $\\alpha=\\alpha(v,v_x)$ and $F=F(v,v_x)$ are treated as small perturbations of the biharmonic heat equation $v_t=-v_{xxxx}$. The central technical objects are scaled parabolic Hölder norms $\\|\\cdot\\|'_{BC^{k+\\mu,(k+\\mu)/4}}$ and weighted norms $\\|\\cdot\\|_{Z^k_T}$, which respect the asymptotic scaling of the equation so that semigroup estimates for $e^{-t\\partial_x^4}$ and parabolic Schauder estimates remain uniform. The limit mechanism is the convergence $f_\\sigma(r)=\\sigma f(\\sigma^{-1}r)\\to f'(0)r$ as $\\sigma\\to\\infty$, which makes the explicit curvature nonlinearity disappear at the rescaled level and forces the limit $U$ to solve the linearized surface diffusion equation.","core_discovery":"The paper's central assertion is that the exponential surface diffusion flow is globally well-posed for small data and that its large-time behavior is exactly the self-similar profile of the linearized flow. More precisely, under the smallness condition $\\|(u_0)_x\\|_{W^{1,\\infty}}<\\varepsilon_*$ and the asymptotic slope conditions $u_0(x)=(a+o(1))x$ as $x\\to+\\infty$ and $u_0(x)=(b+o(1))x$ as $x\\to-\\infty$ with $|a|,|b|$ small, the rescaled solutions $u_\\sigma$ converge uniformly on compact subsets of $\\mathbb{R}\\times[0,\\infty)$ to the unique self-similar solution $U$ of the linearized equation (1.11), and derivatives up to order $k+1$ in the parabolic sense converge on compact subsets of $\\mathbb{R}\\times(0,\\infty)$. The curvature nonlinearity is genuinely washed out by the scaling, since $f_\\sigma(r)=\\sigma f(\\sigma^{-1}r)\\to f'(0)r$ as $\\sigma\\to\\infty$. The paper presents this as the first global existence and asymptotic-self-similarity theorem for this nonlinear flow, and as a direct justification of the thermal-grooving explanation based on the Gibbs–Thomson law without linearizing $f$ near $\\kappa=0$.","pith_inferences":["If the same scaled-Hölder machinery extends to $\\mathbb{R}^n$ as Remark 1.4 suggests, the exponential flow should admit self-similar solutions for small Lipschitz data on the sphere at infinity, paralleling the linearized theory.","The paper proves convergence only on compact sets; extending the equi-decay argument that Remark 1.5 attributes to [3] would upgrade the result to uniform convergence on all of $\\mathbb{R}$.","The Hölder inequality asserted in the proof of Theorem 1.2 could be written out with its constant; if the constant is not small enough, the theorem would still give subsequential convergence, and only the uniqueness of the limit would be open.","The theorem makes a concrete prediction for crystal-surface experiments: long-time groove profiles should match the linearized self-similar solution whenever the initial slope and its spatial derivative are small, even if curvature itself is not small initially."],"forward_implications":["The thermal-grooving profile of the linearized surface diffusion equation is recovered as the actual large-time attractor of the fully nonlinear exponential flow.","For small-gradient data, the exponential flow and the conventional linear flow share the same self-similar limit, so the exponential nonlinearity does not change the leading-order shape of grooves at large times.","The a priori decay estimate (1.8) converts local-in-time existence into global existence and supplies quantitative derivative bounds; the argument is designed to extend to higher-dimensional graph-like surfaces as indicated in Remark 1.4.","Under the stated smallness condition, any two subsequential limits of the rescaled solutions coincide, so the large-time profile is unique rather than one of a family."],"supporting_citations":[{"why":"Supplies the uniqueness theorem for the limiting self-similar solution of the linearized flow, the uniqueness class that pins down the limit $U$.","marker":"[15]"},{"why":"Supplies the parabolic Schauder estimates used as the linear regularity engine for the perturbed biharmonic heat equation.","marker":"[23]"},{"why":"Supplies the Hölder-norm product, composition, and contractivity estimates used to control the nonlinear perturbation map.","marker":"[12]"},{"why":"Provides the thermal-grooving model whose linearized prediction the paper proves to be the large-time attractor.","marker":"[22]"},{"why":"Gives the corresponding self-similar stability result for the conventional linear surface diffusion flow, the baseline this paper extends to exponential $f$.","marker":"[3]"}],"fun_headline_variants":["Exponential surface diffusion converges to linear self-similar profile","Small slopes force exponential flow to a unique self-similar limit","Exponential surface flow: unique self-similar profile in large time","Small-data global well-posedness and self-similar limit for exponential flow","Exponential curvature flow reaches linear self-similar state"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that estimate (6.5) forces every subsequential limit $U$ into the uniqueness class used in [15, Theorem 3.4]; the paper asserts the needed Hölder inequality without proof, so if its constant does not fall below the smallness threshold the argument would only establish convergence along subsequences.","fun_headline_variants_meta":{"raw":{"variants":["Exponential surface diffusion converges to linear self-similar profile","Small slopes force exponential flow to a unique self-similar limit","Exponential surface flow: unique self-similar profile in large time","Small-data global well-posedness and self-similar limit for exponential flow","Exponential curvature flow reaches linear self-similar state"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001082,"raw_usage":{"total_tokens":4552,"prompt_tokens":998,"completion_tokens":3554,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":614,"completion_tokens_details":{"reasoning_tokens":3467}},"tokens_in":614,"tokens_out":3554,"duration_ms":30021,"temperature":1.0,"reasoning_tokens":3467,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:29:51.159397+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A numerical experiment would settle the central claim: solve (1.1) with $f(r)=e^r$ and initial data $u_0(x)=a x_+ + b x_-$ for small $a,b$, evolve to large times, form $u_\\sigma(x,t)=\\sigma^{-1}u(\\sigma x,\\sigma^4 t)$, and compare on a fixed compact set with the unique self-similar solution $U$ of (1.11) with initial data $a x_+ + b x_-$; two different subsequential limits, or a limit failing (1.11), would refute the convergence claim.","supporting_citations":[{"cited_title":"Koch and T","cited_arxiv_id":null,"evidence_quote":"Supplies the uniqueness theorem for the limiting self-similar solution of the linearized flow, the uniqueness class that pins down the limit $U$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the parabolic Schauder estimates used as the linear regularity engine for the perturbed biharmonic heat equation."},{"cited_title":"G¨ oßwein, Surface diffusion flow of triple junction clusters in higher sp ace dimensions , Ph.D","cited_arxiv_id":null,"evidence_quote":"Supplies the Hölder-norm product, composition, and contractivity estimates used to control the nonlinear perturbation map."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the thermal-grooving model whose linearized prediction the paper proves to be the large-time attractor."},{"cited_title":"Du and N","cited_arxiv_id":null,"evidence_quote":"Gives the corresponding self-similar stability result for the conventional linear surface diffusion flow, the baseline this paper extends to exponential $f$."}],"review_version":1}