{"id":"83274dac-bcfa-413d-9904-e2ff5dd2287b","arxiv_id":"2506.22731","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the planar surface diffusion flow, small V-shaped data produce genuine nonlinear forward self-similar graph solutions, while a growth condition forces any profile to be a line.","lead":"The paper resolves a contradiction in the surface diffusion flow literature between Koch and Lamm's existence theorem and Rybka and Wheeler's non-existence remark. It proves that small V-shaped data do generate genuine nonlinear forward self-similar graph solutions, and it gives a clean linearity condition under which only lines can occur.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'In particular' existence in Theorem 19 rests on an unproved transfer: Theorem 14 is quoted for the original equation with Lipschitz data phi_{A,B}, but the required auxiliary solution v to the differentiated equation (13) with step initial data and estimates (16) is never shown to exist.","rationale":"The paper's internal computations are mostly sound: the key identity in Lemma 2, the scaling calculation showing U is self-similar, and the equivalence of the differentiated equation with (12) are all consistent. The Duhamel step in Theorem 17 is formal but standard under the estimates (16). The real soft spot is the input hypothesis of Theorem 19: the paper takes as given a self-similar solution v to (13) with step initial data and estimates (16), but does not derive it from the quoted Koch-Lamm theorem. What must be true for the central claim is that differentiating the Koch-Lamm solution u with homogeneous data phi_{A,B} yields such a v satisfying the integral equation (14) and the right pointwise trace. This transfer is plausible because the linear semigroup regularizes a bounded step with exactly the t^{-ell/4} rates, and smallness of max(|A|,|B|) makes the constants small. Still, the fixed-point argument for the nonlinear equation with discontinuous v0 and the pointwise convergence of v to the step are not given. That is the load-bearing assumption behind the 'In particular' existence statement. The reader's conditional verdict already captures this weakness; I would keep the verdict unchanged and require the transfer to be made fully explicit. The reader's weakest_assumption also mentions uniqueness of (14), which is not needed for the direct self-similarity construction, so my agreement is only partial.","tokens_in":10633,"tokens_out":28163,"duration_ms":279329,"concrete_test":"Take A=epsilon, B=2epsilon so that phi_{A,B} has a genuine corner. Run the Koch-Lamm fixed-point construction directly on the derivative equation (14) with v0(x)=-B for x<0, A for x>0, in the weighted norm ||v|| = sup_{t>0}(||v(t)||_inf + t^{1/4}||v_x(t)||_inf + t^{1/2}||v_xx(t)||_inf). Check that the map v -> e^{-t d_x^4} v0 + integral_0^t d_x^2 e^{-(t-s)d_x^4}(alpha[v]v_xx + F[v]) ds is a contraction on {||v|| <= C epsilon} with C independent of epsilon, and that v(x,t) -> v0(x) for every x != 0 as t -> 0. If the contraction succeeds, Theorem 19's existence assertion is justified; if it fails for step data, the central claim lacks its auxiliary solution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is the final sentence of Theorem 19: there exists a forward self-similar, possibly nonlinear, solution with initial data phi_{A,B} for small A,B. The proof, however, assumes as a hypothesis a self-similar solution v to the differentiated equation (13) with v(x,0)=-B for x<0, v(x,0)=A for x>0, satisfying the pointwise estimates (16), and then constructs U from v by (17). The paper never proves that such a v exists. The quoted Koch-Lamm theorem (Theorem 14) is stated for the original surface diffusion flow (12), with Lipschitz homogeneous initial data u0=phi_{A,B}; it yields a solution u, not a solution v of the integral equation (14). The transfer v=u_x is plausible: for the linearized evolution of the step v0, |d_x^ell e^{-t d_x^4} v0| <= C t^{-ell/4} ||v0||_inf, and Theorem 14's estimates on derivatives of u give small C_ell for small max(|A|,|B|). But the fixed-point proof of (14) for discontinuous v0, the verification that the nonlinear terms retain the same decay, and the pointwise trace v(x,t)->v0(x) away from x=0 are not supplied. If that transfer fails or requires more regularity than a step, the construction of U collapses. Theorem 17 itself is conditional and does not fill this gap. This is the load-bearing input for the claimed reconciliation with [RW, Remark 16].","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies graph-like forward self-similar solutions of the surface diffusion flow. Its main mathematical contributions are a geometric identity for profile surfaces (Lemma 2), several linearity criteria for graph-like profiles (Theorem 7, Theorem 11 and related corollaries), and a critique of the interpretation in [RW, Remark 16] that suggested no non-linear graph-like self-similar solutions exist. In Section 5, the authors recall the Koch--Lamm existence theorem for small homogeneous Lipschitz initial data and, under the extra hypothesis that a self-similar solution v of the differentiated equation exists with step initial data and pointwise decay estimates, they construct a function U by integrating v and prove that U is a self-similar solution with initial data phi_{A,B}. The paper concludes in Theorem 19 that, for small A and B, there exists a forward self-similar solution that may not be linear, thereby reconciling [RW] with [KL]. The central existence claim, however, is conditional on the existence of v, which is assumed rather than proved.","tokens_in":10928,"tokens_out":5161,"duration_ms":61782,"significance":"If the existence transfer from the Koch--Lamm theorem to the differentiated equation were supplied, the paper would resolve an interesting apparent contradiction in the literature: it would show that the small homogeneous-data solutions of Koch and Lamm are genuine graph-like self-similar solutions and not merely solutions modulo a time-dependent spatially constant function. The paper also gives a clean derivation of the key identity in general dimensions and convincingly demonstrates that the boundedness condition used in [RW, Remark 16] is unstable under natural notions of closeness. The main weakness is that the paper's headline existence result is conditional: Theorem 19 assumes the existence of v, and no proof is given that the solution u from Theorem 14 produces such a v via v = u_x. This gap is load-bearing because the final sentence of Theorem 19 is the central claim of the paper.","major_comments":[{"comment":"The asserted existence in the final sentence of Theorem 19 is not proved. The theorem's hypotheses include a self-similar solution v of (13) satisfying (16) with piecewise constant initial data, but the paper nowhere establishes that such a v exists. Theorem 14 is stated for the original equation (12) with Lipschitz homogeneous data phi_{A,B} and yields a solution u, not a solution v of the differentiated integral equation (14). The transfer v = u_x is plausible from the estimates of Theorem 14, but the paper does not supply the necessary verification: that u_x solves (14), satisfies the pointwise estimates (16), is self-similar, and has the pointwise initial trace v(x,0) = -B for x<0, v(x,0)=A for x>0. The 'In particular' claim therefore overstates what has been established. A proof of this transfer, or an explicit statement that Theorem 19 is conditional on it, is required.","section":null},{"comment":"Theorem 17 is conditional in the same way: it begins with 'Let v be a solution of (13)' and assumes the estimates (16). The proof that U is well-defined and solves (12) is given, but the existence of v with the required step initial data is not treated. Remark 18 invokes smallness of the constants C_l in (16) for uniqueness of (14) and says this is 'natural' if the L-infinity norm of v0 is small, but it does not prove that the smallness of A and B in Theorem 14 implies the smallness of the C_l for the transferred solution v = u_x. Since the paper's reconciliation with [RW] depends on the existence of this auxiliary v, the gap must be filled or the claims must be reformulated as conditional.","section":"Section 5, Theorem 17 and Remark 18"},{"comment":"The passage from the integral equation (18) to the PDE (12) is justified only by a formal Duhamel argument ('This is standard, so we only give a formal argument'). The estimates preceding it make the argument credible, but for a central construction the reader should either be given a rigorous justification or a precise reference covering the differentiated equation. This is secondary to the missing existence of v, but it is part of the same load-bearing construction.","section":"Section 5, proof of Theorem 17"}],"minor_comments":[{"comment":"The title contains typographical errors: 'for WARD' should be 'forward' and 'SURF ACE' should be 'surface'; the abstract also repeats these errors.","section":"Title and Abstract"},{"comment":"The norm in (21) is written as ‖U - e^{-t\\partial_x^4}u_0‖_{L^\\infty} without specifying the domain; it should be L^\\infty(R) for consistency with the surrounding estimates.","section":"Section 5, equation (21)"},{"comment":"The phrase 'norm of homogeneous Lipschitz function is used' in Remark 16 is a little unclear; it would be clearer to say that the Koch--Lamm smallness condition in Theorem 14 is expressed in terms of the Lipschitz seminorm of u0, which is insensitive to additive constants, and this is what creates the apparent ambiguity.","section":"Section 5, Remark 15"},{"comment":"The reference [GGK] is cited for large-time behavior; if this paper is not yet published, a preprint identifier would help the reader, as is done for [RW].","section":"Section 1"}],"recommendation":"major_revision","confidential_remarks":"The central claim is likely correct and the gap is probably repairable by a lemma that transfers the Koch--Lamm estimates from u to v = u_x, or by a direct fixed-point argument for (14) with step initial data. The revision should make Theorem 19 non-conditional or explicitly state the missing hypothesis as an assumption; as written, the 'In particular' sentence is stronger than the proof supports. No concerns about novelty or attribution: the paper properly credits [RW] and [KL]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a remarks-style contribution that does two useful things. First, it re-derives, in a clean geometric form, the key identity that Rybka and Wheeler used, and then proves a linearity criterion (Theorem 7) under a growth condition slightly broader than RW's. Second, it correctly points out that RW's Remark 16 overreads their own non-existence results: the \"closeness\" condition on profiles is unstable, and inequality (8) in the old version of RW plainly fails for x1 < 0. Those parts are honest and correct.\n\nThe main new assertion is Theorem 19: if a self-similar solution v to the differentiated equation (13) exists with step initial data and the pointwise estimates (16), then U reconstructed via (17) is a genuine self-similar solution of the original surface diffusion equation with initial data phi_{A,B}, with no time-dependent constant ambiguity. That targets RW's conjecture directly.\n\nThe soft spot is the \"In particular\" at the end of Theorem 19. It claims existence of such U for small A,B. The proof is conditional on the existence of v. The paper never proves that such v exists. Theorem 14 from Koch and Lamm is stated for the original equation and gives a solution u with homogeneous Lipschitz data. The natural move is to set v = u_x. The KL estimates do give (16), and differentiability of u should give (13). But the paper does not make that transfer, and it does not show that v(x,t) tends to the step function pointwise away from x=0 as t→0. So the central reconciliation with RW rests on an unproven bridge. I think the bridge can be built from KL in a few lines, but a serious referee should ask for it.\n\nEverything else is clean. Theorem 17's final step is formal but standard; the uniqueness remark is careful. The citation practice is good: RW's identity and KL's existence are credited, and the authors' own papers are used for context, not to hide a gap.\n\nWho is this for? People working on surface diffusion and Willmore-type flows, and anyone puzzled by the apparent contradiction between RW's non-existence and KL's existence. It deserves a serious referee. I would recommend publication after the existence of v is made explicit — either derived from the KL solution or stated as a hypothesis with the \"In particular\" softened.","headline":"Useful clarification of the surface-diffusion self-similar literature, but the reconciliation with Rybka–Wheeler is conditional on an unproved transfer from Koch–Lamm.","tokens_in":11477,"tokens_out":6155,"would_cite":true,"duration_ms":68170,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35K55","53C44"],"pacs":[],"model":"deepseek-v4-flash","headline":"Small nonlinear graph-like forward self-similar solutions to planar surface diffusion exist, contrary to a recent conjecture.","keywords":["surface diffusion flow","forward self-similar solutions","graph-like solutions","nonlinear existence","convexity identity","soliton classification","geometric flows"],"falsifier":"For small $A$ and $B$ with $A \\neq B$, compute the unique small solution to the graph equation (12) with initial data $\\varphi_{A,B}$ using the integral representation (17), and check whether the profile $\\varphi(x)=U(x,1)$ has a nonzero second derivative at some point and whether the scaling identity $\\sigma^{-1/4}U(\\sigma^{1/4}x,\\sigma t)=U(x,t)$ holds for several $\\sigma$; a linear profile or a failure of scaling for any of these small data would contradict the conclusion of Theorem 19.","tokens_in":10395,"feed_emoji":"","tokens_out":13399,"duration_ms":123034,"temperature":0.7,"pith_summary":"The paper resolves an apparent contradiction in the theory of the planar surface diffusion flow. A recent classification result (see [RW]) suggested that, in two dimensions, the only graph-like forward self-similar profiles are straight lines, casting doubt on the existence theorem of [KL] which constructs small, possibly curved, self-similar graphs. The authors show that the earlier doubt is based on a misleading interpretation of what 'close to a homogeneous function' means, and that the solution produced by [KL] is a genuine graph-like self-similar solution, not merely one defined up to a time-dependent additive constant. Concretely, Theorem 19 demonstrates that for small slope parameters A and B, the reconstructed profile is a forward self-similar solution of the surface diffusion equation with V-shaped homogeneous initial data, and it may be nonlinear. The reconciliation rests on a key identity for profile surfaces and on reconstructing the solution of the original graph equation from a solution of the differentiated equation for the slope.","feed_headline":"Small nonlinear self-similar surface-diffusion graphs exist","feed_subtitle":"Reconciles two conflicting results and shows the construction has no time-dependent constant ambiguity.","key_machinery":"The central object is the identity of Lemma 2: for a profile surface $\\Gamma_*$ of a forward self-similar solution, $\\Delta_{\\Gamma_*}(H^2+|x|^2/4) = (d-1)/2 + 2|\\nabla_{\\Gamma_*}H|^2$. In the two-dimensional graph setting this becomes $\\partial_s^2(k^2+|x|^2/4) = 1/2+2(\\partial_s k)^2$, where $s$ is arc length and $k$ is curvature. This identity forces convexity of certain auxiliary functions $Q$, and combined with the geometric estimate $|x|^2 \\le s^2 - \\varphi(0)^2 + 2\\varphi(0)\\varphi(x_1)$ it implies that any profile satisfying the growth condition (5) must be linear. The second machinery is the integration-reconstruction step: from a self-similar solution $v$ of the slope equation (13) with estimates (16), the integral formula (17) builds the profile $U$ of the original equation, and the scaling property of $v$ transfers to $U$, proving self-similarity without needing uniqueness of the original equation.","core_discovery":"On its own terms, the paper establishes that the small forward self-similar graph solutions supplied by the existence theorem of [KL] are true solutions of the original surface diffusion equation, with no ambiguity from time-dependent constants. Writing the graph equation as (12) and letting $v = u_x$, the differentiated equation (13) is solved by a self-similar $v$ with piecewise constant initial data $v(x,0)=-B$ for $x<0$ and $v(x,0)=A$ for $x>0$. The paper proves (Theorem 17) that the formula $U(x,t)=e^{-t\\partial_x^4}u_0 + \\int_0^t \\partial_x e^{-(t-s)\\partial_x^4}(\\alpha[v]v_{xx}+F[v])(s)\\,ds$ reconstructs a solution to (12) whose $x$-derivative is exactly $v$, and that this $U$ is uniformly continuous up to $t=0$, attains the V-shaped initial data $\\varphi_{A,B}$, and is unique in the class of solutions of the integral equation. Since $v$ is self-similar, a scaling argument shows $U$ is self-similar (Theorem 19). In particular, for small $A$ and $B$ there exists a forward self-similar solution that may be nonlinear, which directly disproves the conjecture stated in [RW, Remark 16].","pith_inferences":["The reconciliation suggests that analogous 'apparent constant ambiguity' issues in other geometric flows with a differentiated-equation structure (e.g., the Willmore flow as noted in [KL]) can likely be resolved by the same integration-reconstruction argument, yielding true self-similar solutions rather than solutions modulo time-dependent constants.","The paper's convexity argument draws a sharp dichotomy: profiles satisfying one-sided growth bounds (5) are linear, while those governed by V-shaped homogeneous data at infinity escape the bound; this dichotomy might extend to higher dimensions or other surface-diffusion variants, providing a general rigidity principle for graph-like profiles.","A testable numerical check is to compute the small solution for small $A \\neq B$ using the integral representation (17) and verify that the profile at $t=1$ has nonzero curvature and satisfies the scaling identity exactly; such a check would provide independent confirmation of the non-linearity claim.","If one could construct the auxiliary solution $v$ without the smallness restriction from [KL], the same reconstruction would produce large-amplitude forward self-similar graphs; the paper leaves the size threshold as the key open constraint."],"forward_implications":["The conjecture in [RW, Remark 16] is disproved: there exist forward self-similar graph-like solutions of the planar surface diffusion equation that are not linear, so the existence theorem of [KL] is not in conflict with the classification results.","For any self-similar solution $v$ of the differentiated slope equation satisfying (16), the reconstructed $U$ in (17) is a genuine self-similar solution of the original graph equation with the prescribed initial data, eliminating the possibility that the solution is only self-similar up to a time-dependent spatially constant function.","A sufficient condition (5) is established under which any graph-like forward self-similar profile must be linear; in particular, profiles that are globally Lipschitz and whose slope approaches a constant $a_0$ in $L^1(\\mathbb{R})$ are exactly the lines with slope $a_0$.","There is no smooth compact profile surface for any forward self-similar solution, since the key identity makes the maximum of $H^2+|x|^2/4$ impossible."],"supporting_citations":[{"why":"Supplies the existence theorem (Theorem 14) for small analytic self-similar graph solutions and the integral-equation framework (14)–(18) used throughout the reconstruction.","marker":"[KL]"},{"why":"Provides the non-existence and rigidity results for graph-like forward self-similar solutions, the conjecture in Remark 16 that the paper refutes, and the curvature blow-up lemma used in Theorem 7.","marker":"[RW]"},{"why":"Introduces the method of constructing self-similar solutions by solving the evolution equation with homogeneous initial data, the idea underlying the [KL] construction and the scaling argument in Theorem 19.","marker":"[GM]"}],"fun_headline_variants":["Small self-similar surface-diffusion graphs proven to exist","Self-similar surface-diffusion curves exist for small data","Nonlinear self-similar solutions to surface diffusion confirmed","Graph-like self-similar diffusion solutions exist, conjecture false","Small forward self-similar graphs exist for surface diffusion"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The existence of the small self-similar solution $v$ to the differentiated equation (13) with V-shaped initial data and the smoothing estimates (16) is assumed from the theorem of [KL] (Theorem 14); this paper does not reprove that existence, and if those estimates fail for some small V-shaped data, the reconstruction and the main theorem collapse.","fun_headline_variants_meta":{"raw":{"variants":["Small self-similar surface-diffusion graphs proven to exist","Self-similar surface-diffusion curves exist for small data","Nonlinear self-similar solutions to surface diffusion confirmed","Graph-like self-similar diffusion solutions exist, conjecture false","Small forward self-similar graphs exist for surface diffusion"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000241,"raw_usage":{"total_tokens":1465,"prompt_tokens":830,"completion_tokens":635,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":446,"completion_tokens_details":{"reasoning_tokens":554}},"tokens_in":446,"tokens_out":635,"duration_ms":6590,"temperature":1.0,"reasoning_tokens":554,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:59:52.906585+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For small $A$ and $B$ with $A \\neq B$, compute the unique small solution to the graph equation (12) with initial data $\\varphi_{A,B}$ using the integral representation (17), and check whether the profile $\\varphi(x)=U(x,1)$ has a nonzero second derivative at some point and whether the scaling identity $\\sigma^{-1/4}U(\\sigma^{1/4}x,\\sigma t)=U(x,t)$ holds for several $\\sigma$; a linear profile or a failure of scaling for any of these small data would contradict the conclusion of Theorem 19.","supporting_citations":[],"review_version":1}