{"id":"403b1045-eccb-4d47-b4ce-4eeaddba2f63","arxiv_id":"1908.07386","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A numerical scheme is presented for a fractional tumor-growth free boundary problem with a claimed convergence and stability proof, yet the test case is internally inconsistent and the proof covers only a simplified homogeneous version.","lead":"This paper proposes a spectral-finite-difference method for a fractional parabolic-hyperbolic free boundary problem describing tumor growth with nutrient and drug. It claims unconditional convergence and stability, but the numerical example does not actually satisfy the model equations, so the validation for the stated tumor model fails.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The convergence proof applies only to the zero-boundary, zero-initial-data problem, because the 'without loss of generality' reduction discards nonzero c̄(t), w̄(t) and c0, w0; the nonlinear reactions f(c,p,q), g(w,p,q) make superposition invalid.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: the WLOG reduction to c = w = 0 and c0 = w0 = 0 after Eq. (29) removes exactly the nonzero nutrient and drug boundary supplies and initial profiles that distinguish the advertised model from a homogeneous test problem. This concern is more fundamental than the order inconsistencies because even a perfect repair of the rate estimates would not extend the theorem to the stated problem: the nonlinear coupling f(c,p,q), g(w,p,q) invalidates any additive superposition argument. The numerical example in Section 5 also uses zero boundary and initial data for c and w, so the reported validation is consistent only with the reduced problem. A computational or analytical test that exercises nonzero data would settle the issue directly. Since this concern supports the reader's REJECT verdict, no adjustment is needed.","tokens_in":24759,"tokens_out":8600,"duration_ms":83696,"concrete_test":"Implement the proposed scheme on the scalar analogue of (16)–(21) with nonlinear reaction f(c) = c^2 and nonzero data c(1,t) = 1, c0(ρ) = 1−ρ^2, using the same discretization as (30)–(31). Compare against a reference solution obtained by the same scheme after explicitly lifting the boundary/initial data, i.e. writing c = c̃ + φ with φ(1,t) = 1, φ(ρ,0) = 1−ρ^2 and retaining the resulting φ-dependent terms in the discrete equations. If the unmodified scheme fails to converge at the claimed order (Δt)^{2−α/2} to the reference solution, or if the lifting terms are indispensable, the 'without loss of generality' reduction is invalid and Theorem 3.1 does not cover the model (16)–(26).","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 3, immediately after Eq. (29), the paper states 'without loss of generality we can suppose that c = w = 0 and c0(ρ) = w0(ρ) = 0.' This is not a legitimate reduction for the nonlinear problem (16)–(26). The boundary supplies c̄(t), w̄(t) and the initial profiles c0, w0 are removed from the discretized equations (30)–(31), and all subsequent error estimates — Lemma 3.1, Theorem 3.1, and Theorem 4.1 — are proved for this homogeneous problem only. But since f(c,p,q) = K1(c)p + K2(c)q and g(w,p,q) = K3(w)p + K4(w)q depend nonlinearly on c and w, the solution map is not additive: solving the homogeneous problem does not determine the solution for nonzero boundary or initial data, and no lifting-function construction or fixed-point argument is provided to close this gap. Consequently, Theorem 3.1 does not establish convergence for the tumor model (16)–(26) with positive nutrient/drug supply, which is exactly the model advertised in the abstract. Section 5 confirms this limitation: Example 1 uses c(1,t) = w(1,t) = 0 and c0 = w0 = 0, so the numerical validation never exercises the general case. The abstract's claim to solve the stated problem is therefore unsupported even if all algebraic estimates are correct.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a combined finite-difference/spectral method for a fractional parabolic-hyperbolic free boundary system modeling radially symmetric tumor growth with nutrient and drug concentrations and three cell populations. The model (16)-(26) consists of two reaction-diffusion equations with a Riemann-Liouville fractional derivative applied to the diffusion term, three hyperbolic transport equations for the cell densities, and an ODE for the tumor radius. The time discretization is an L1-type scheme, the spatial discretization for the nutrient and drug is a Legendre spectral method, and the hyperbolic equations are treated along characteristics with midpoint-type finite differences. The main theoretical claim, stated in the abstract and Theorem 3.1, is unconditional convergence and stability with error O((Δt)^{2-α/2}+K(N)) for the cell-density errors, the radius error, and weighted H1 errors of the nutrient and drug. One numerical example is presented to justify the theoretical results.","tokens_in":25148,"tokens_out":8494,"duration_ms":78998,"significance":"If the main theorem were valid, the paper would provide a useful convergence and stability analysis for a coupled fractional moving-boundary problem, and the spectral basis (53) is a sensible construction adapted to the stated boundary conditions. The paper also makes a serious attempt at a full energy-estimate proof, and the stability analysis in Theorem 4.1 is a natural extension of the convergence argument. However, the central claim is not supported as written: the convergence proof is carried out only for a homogeneous version of the problem, and the numerical example used for validation does not satisfy the governing fractional equations or all of the stated initial and boundary conditions. These are load-bearing deficiencies, not presentation issues.","major_comments":[{"comment":"The reduction 'without loss of generality we can suppose that c = w = 0 and c0(ρ) = w0(ρ) = 0' is not valid for the nonlinear problem (16)-(26). Since f(c,p,q) = K1(c)p + K2(c)q and g(w,p,q) = K3(w)p + K4(w)q depend nonlinearly on c and w, the solution map is not additive: knowing the zero-data solution does not determine the solution with nonzero boundary supplies c̄(t), w̄(t) and nonzero initial profiles c0, w0, and no lifting or fixed-point argument is provided. All subsequent estimates, including Lemma 3.1, Theorem 3.1, and Theorem 4.1, are proved for the homogeneous discrete equations (46)-(47) with zero boundary and initial data. The abstract's claim to solve the original model (16)-(26) is therefore not established; this is confirmed by Example 1, which tests only c(1,t)=w(1,t)=0 and c0=w0=0.","section":"Section 3, after Eq. (29)"},{"comment":"The stated exact solutions do not satisfy the fractional diffusion equations. For example, w(t,ρ)=8t(1-ρ²) gives w_t=8(1-ρ²), while the diffusion term in (87) is a Riemann-Liouville fractional derivative of a quantity proportional to t R(t)^{-2}. Since D_t^α t = t^{1-α}/Γ(2-α), this term is not constant in time, so the two sides of the equation cannot agree for all t; the same obstruction occurs for c. Thus Tables 1-3 do not measure the error of the method against a true solution of the fractional PDE.","section":"Section 5, Eqs. (86)-(87), (94)-(95)"},{"comment":"The numerical example is internally inconsistent. The initial condition in (91) sets d(ρ,0)=4ρ, while the exact d in (95) has d(ρ,0)=((2ρ-1)²+1)+1=(2ρ-1)²+2; these agree only at isolated points such as ρ=1/2 and otherwise differ. The boundary condition in (92) sets v(0,t)=-ε e^{-1}/(t+1)², contradicting the required v(0,t)=0 in (26), and ε is never defined. In addition, p0(ρ)=-((2ρ-1)²+1) is negative, violating the model assumption p0≥0 in (27). These inconsistencies invalidate the numerical verification.","section":"Section 5, Eqs. (91)-(95) and Eq. (26)"}],"minor_comments":[{"comment":"The phrase 'where ∂α/∂ρ = 0D_t^α' appears to contain a typo; the symbol should be ∂α/∂t, and the definition of the Riemann-Liouville derivative should be stated explicitly and consistently throughout.","section":"Section 2, Eq. (16)"},{"comment":"The temporal truncation-error exponents are stated inconsistently: Eq. (29) gives O((Δt)^{2-α}), Eq. (33) gives O((Δt)^{3-α}), while Lemma 3.1 uses (Δt)^{2+(1-α)/2}; the derivation of the final rate in Theorem 3.1 should reconcile these estimates.","section":"Section 3, Eqs. (29), (33), (45), Lemma 3.1"},{"comment":"Equation (40) contains an apparent typo: d(ρ,t_{n+1}) = d(..., t_{n-1}), t_{n-1}) + ... has an extra argument, and the same formula is repeated in Eq. (50).","section":"Section 3, Eqs. (38)-(40) and (48)-(50)"},{"comment":"Several entries in Table 1 are garbled, for example '0.2435 2e-06' and '1.9342 9e-06', which makes the reported errors unreadable.","section":"Section 5, Table 1"},{"comment":"The paper does not specify how the coupled discrete system is solved at each time step, nor whether a nonlinear iteration is required; this should be stated for reproducibility.","section":"Section 3, Eqs. (46)-(52)"}],"recommendation":"reject","confidential_remarks":"To the editor: the two main problems are not local defects. The convergence theorem is proved only for a homogeneous version of the problem, while the advertised model has nonzero boundary supplies and initial profiles; and the numerical example used to validate the method is not a true solution of the fractional PDEs and violates several stated conditions. A substantially new convergence analysis for the nonhomogeneous case and a reliable numerical test would be needed, which goes beyond a revision of the present manuscript."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the central advertised result is not what is proved, and the numerical example does not validate anything. I agree with the stress-test note: the \"without loss of generality\" in Section 3 is not harmless.\n\nWhat is new: the fractional generalization of Zhao's model is a legitimate modeling step, and the L1-in-time/spectral-in-space discretization with characteristic handling of the hyperbolic cell equations is a plausible combination. The energy estimates in the appendix are real work, and the literature is cited appropriately (Lin-Xu for L1, Shen-Tang-Wang for spectral tools, Zhao for the underlying model). If the target had been the homogeneous zero-data problem, Theorem 3.1 would be a reasonable contribution.\n\nThe soft spots are load-bearing. First, the reduction c = w = 0, c0 = w0 = 0 discards the nonzero boundary supplies and initial profiles. Because f(c,p,q) and g(w,p,q) are nonlinear in c and w, you cannot superpose the zero-data solution to recover the problem with data. No lifting argument is supplied. Theorems 3.1 and 4.1 therefore concern a different problem than (16)-(26), and the abstract's claim to solve the tumor model is unsupported. This is exactly the stress-test concern, and it survives reading the paper.\n\nSecond, Example 1 is internally inconsistent. The exact c and w are linear in t, but the Riemann-Liouville fractional derivative of t is t^{1-alpha}/Gamma(2-alpha), so the exact solutions cannot satisfy the fractional PDEs for 0 < alpha < 1. The initial condition d(rho,0) = 4rho contradicts the stated exact d at t = 0, and (92) imposes nonzero v(0,t) while the model (26) sets v(0,t) = 0. The tables and convergence-rate plots are therefore measuring residuals against a function that is not a solution. Third, the error-order statements drift: (33), Lemma 3.1, and Theorem 3.1 carry different powers of Delta t. That may be a minor issue if the proof is fixed, but it adds to the picture.\n\nWho this is for: a reader interested in energy estimates for the zero-data fractional subsystem might find the appendix useful. A reader who wants a numerical method for the tumor model in the abstract should not rely on it.\n\nRecommendation: not ready for peer review. The authors would need to prove or replace the WLOG reduction, construct a valid manufactured solution (or use a self-consistent discrete check), and reconcile the order estimates. I would reject the current version.","headline":"A substantial but fatally incomplete numerical analysis paper: the convergence proof covers only the zero-data subsystem and the validation example does not satisfy the model.","tokens_in":25600,"tokens_out":6089,"would_cite":false,"duration_ms":61903,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M70","65M12","35K20","35L03"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that a hybrid finite difference-spectral scheme for the fractional parabolic-hyperbolic tumor-growth model is unconditionally convergent and stable, with error $O((\\Delta t)^{2-\\alpha/2})$ plus a vanishing spectral term.","keywords":["fractional parabolic-hyperbolic system","free boundary problem","tumor growth model","drug application","spectral method","finite difference method","unconditional convergence","stability"],"falsifier":"Implement the proposed scheme on the original model (16)--(26) with nonzero constant boundary supplies $\\bar c(t),\\bar w(t)$ and nonzero initial profiles, using a manufactured exact solution or a fine-grid reference, and compare the observed temporal rate with the claimed $O((\\Delta t)^{2-\\alpha/2})$; if the error does not follow the bound or the proof cannot be rerun without the zero-data reduction, the 'without loss of generality' premise is what fails.","tokens_in":24614,"feed_emoji":"🧫","tokens_out":9036,"duration_ms":80636,"temperature":0.7,"pith_summary":"The paper takes a radially symmetric free-boundary model of tumor growth in which two diffusion equations describe nutrient and drug concentration and three hyperbolic equations track proliferative, quiescent, and dead cell densities, and replaces the diffusion equations with fractional time derivatives so that subdiffusion is represented. It then constructs an approximate solver that discretizes the fractional time derivative with a finite difference formula, uses spectral (Legendre) polynomials in the radial variable, integrates the cell equations along characteristics, and updates the tumor radius from the cell-velocity relation. The paper's central claim is that this hybrid method is unconditionally convergent and stable: the maximum error in the cell densities and radius, and the weighted $H^1$ error in nutrient and drug concentrations, are bounded by $C e^{MT}((\\Delta t)^{2-\\alpha/2}+K(N_1))$, where $K(N_1)\\to 0$ as the polynomial degree grows. A sympathetic reader would care because a proven error bound is what turns simulations of the fractional model into reliable statements about how nutrient supply and drug application move the tumor boundary.","feed_headline":"Fractional tumor-growth scheme proven stable and convergent","feed_subtitle":"The method's error is bounded by (time step)^{2−α/2}, so simulated tumor dynamics can be trusted.","key_machinery":"The load-bearing object is the finite-difference approximation of the Riemann--Liouville fractional derivative: $$\\frac{\\partial^\\$\\alpha$ u}{\\partial t^\\$\\alpha$}(\\rho,t_n) \\approx u(\\rho,0)\\frac{$t_n^{{-\\alpha}}$}{\\Gamma(1-\\$\\alpha$)}+\\sum_{k=0}^{n-1} a_k\\big(u(\\rho,t_{n+1-k})-u(\\rho,t_{n-k})\\big),$$ with monotone weights $a_k=\\big((k+1)^{1-\\alpha}-k^{1-\\alpha}\\big)/(\\Delta t)^\\alpha\\Gamma(2-\\alpha)$, whose truncation error is $O((\\Delta t)^{2-\\alpha})$. This is combined with orthogonal polynomial trial functions that satisfy the zero Neumann condition at $\\rho=0$ and the zero Dirichlet condition at $\\rho=1$, the characteristic/midpoint treatment of the three cell equations, and the radius update $R^{n+1}=R^{n-1}\\exp(2\\Delta t \\int_0^1 \\rho^2 h(c^n,w^n,p^n,q^n,d^n)\\,d\\rho)$. The error analysis runs on weighted $L^2$ and $H^1$ norms, spectral interpolation and projection estimates, and the discrete Gronwall lemma.","core_discovery":"On the paper's own terms, the discovery is a convergence-and-stability theorem for a specific numerical recipe applied to the fractional model (16)--(26). Theorem 3.1 states that if the exact nutrient and drug concentrations are smooth enough, then for every time step $\\Delta t$ and every spectral degree $N_1$ the global error satisfies $$\\max_{0\\le k\\le n+1} E_k \\le C_4^* $e^{{MT}}$\\left((\\$\\Delta$ t)^{2-\\$\\alpha$/2}+K_4^*(N_1)\\right),$$ with an analogous bound for the weighted $H^1$ errors $e_k$, where $K_4^*(N_1)\\to 0$ as $N_1\\to\\infty$ and equals $1/N_1^m$ when the relevant second derivatives are $C^m$. Theorem 4.1 extends the same conclusion to the perturbed problem (69)--(79): if six small forcing or data perturbations of size $\\epsilon_1$ are added, the error contains an additional $\\epsilon_1$ term. The proof is inductive, closing the coupling between the parabolic unknowns, the hyperbolic cell densities, and the moving radius through a discrete Gronwall argument.","pith_inferences":["If the zero-data reduction is not harmless, the practical claim about the drug-supply model is stronger than the theorems establish; a separate proof that keeps $\\bar c$, $\\bar w$, $c_0$, and $w_0$ nonzero would be needed to cover the original problem.","The same L1-spectral template could be carried over to related fractional moving-boundary problems, for example nutrient-only spheroids or therapy models with different kinetic laws, whenever the reaction terms are Lipschitz and the boundary data are smooth.","The exponential factor $e^{MT}$ suggests that for long time horizons the guaranteed error bound may become pessimistic; the numerical tests shown use moderate $T$, so the theorem does not by itself certify very-long-time simulation.","Because the theorem controls concentrations in a weighted $H^1$ norm and cell densities in the $\\infty$-norm, the practical error seen at the moving boundary may be smaller than the worst-case bound, and a boundary-focused error analysis is a natural next step."],"forward_implications":["For the homogeneous problem the method needs no CFL restriction: the error bound holds for every time step, so simulations can use large $\\Delta t$ when accuracy permits.","The temporal order $2-\\alpha/2$ lies between $1.5$ and $2$ for $0<\\alpha<1$, so the fractional order visibly slows the convergence of the diffusion part, and the numerical tables confirm a rate near this value.","With $C^m$ smooth solutions the space error decays as $1/N_1^m$, so spectral accuracy in the radial direction improves rapidly as the polynomial degree grows.","Small perturbations in the forcing terms or initial data change the solution by an amount controlled by the same exponential factor plus $\\epsilon_1$, giving stability in the sense the theorem states.","The same algorithm generates the full time history of the moving boundary radius $R(t)$ along with the cell densities, which is what a user needs to simulate the drug response of a radially symmetric tumor."],"supporting_citations":[{"why":"supplies the original integer-order parabolic-hyperbolic free boundary tumor model with drug application that the paper fractionalizes and solves.","marker":"[5]"},{"why":"supplies the finite-difference formula for the time-fractional derivative and its truncation bound, the core time discretization.","marker":"[21]"},{"why":"supplies the discrete Gronwall lemma used to close the inductive error estimate across coupled unknowns.","marker":"[17]"},{"why":"supplies the orthogonal polynomial spectral projection and interpolation error estimates controlling the space discretization.","marker":"[19]"},{"why":"supplies the weighted $L^p$ norm definitions used throughout the error analysis.","marker":"[18]"},{"why":"supplies the fractional Sobolev and H\\\"older function spaces used to state the regularity assumptions.","marker":"[20]"}],"fun_headline_variants":["Fractional tumor-drug solver proven stable and convergent","Tumor growth with drug: fractional numerical method proven stable","Fractional free-boundary tumor: convergence and stability proven","FD-spectral scheme for fractional tumor model proven convergent","Stable fractional tumor growth solver: proof of convergence"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the step in Section 3 that says 'without loss of generality' one may set $c=w=0$ and $c_0=w_0=0$: if that reduction is not actually lossless, because the reaction terms $f(c,p,q)$ and $g(w,p,q)$ depend nonlinearly on the nutrient and drug concentrations, then the convergence and stability theorems apply to the homogeneous version of the problem rather than to the full tumor model with nonzero boundary supplies.","fun_headline_variants_meta":{"raw":{"variants":["Fractional tumor-drug solver proven stable and convergent","Tumor growth with drug: fractional numerical method proven stable","Fractional free-boundary tumor: convergence and stability proven","FD-spectral scheme for fractional tumor model proven convergent","Stable fractional tumor growth solver: proof of convergence"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000353,"raw_usage":{"total_tokens":1957,"prompt_tokens":1017,"completion_tokens":940,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":633,"completion_tokens_details":{"reasoning_tokens":863}},"tokens_in":633,"tokens_out":940,"duration_ms":10479,"temperature":1.0,"reasoning_tokens":863,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:37:30.283074+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Implement the proposed scheme on the original model (16)--(26) with nonzero constant boundary supplies $\\bar c(t),\\bar w(t)$ and nonzero initial profiles, using a manufactured exact solution or a fine-grid reference, and compare the observed temporal rate with the claimed $O((\\Delta t)^{2-\\alpha/2})$; if the error does not follow the bound or the proof cannot be rerun without the zero-data reduction, the 'without loss of generality' premise is what fails.","supporting_citations":[{"cited_title":"Zhao, A parabolic-hyperbolic free boundary problem mo deling tumor growth with drug application, Electron","cited_arxiv_id":null,"evidence_quote":"supplies the original integer-order parabolic-hyperbolic free boundary tumor model with drug application that the paper fractionalizes and solves."},{"cited_title":"Lin, C.J","cited_arxiv_id":null,"evidence_quote":"supplies the finite-difference formula for the time-fractional derivative and its truncation bound, the core time discretization."},{"cited_title":"Quarteroni, A","cited_arxiv_id":null,"evidence_quote":"supplies the discrete Gronwall lemma used to close the inductive error estimate across coupled unknowns."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the orthogonal polynomial spectral projection and interpolation error estimates controlling the space discretization."},{"cited_title":"Canuto, M.Y","cited_arxiv_id":null,"evidence_quote":"supplies the weighted $L^p$ norm definitions used throughout the error analysis."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the fractional Sobolev and H\\\"older function spaces used to state the regularity assumptions."}],"review_version":1}