{"id":"a22935c2-3eb9-4aec-9f94-9cdd60428493","arxiv_id":"1908.03158","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a mixed fractional boundary value problem on the quadrant, the Green's function is shown to satisfy explicit two-sided bounds that depend on the distance to the boundary and the ratio of the fractional orders.","lead":"The paper derives two-sided estimates for the Green's function of a boundary value problem with two fractional derivatives in time-like directions and a diffusion in space. These bounds give quantitative control for fractional equations on the positive quadrant, with applications in insurance risk, barrier options, and credit portfolios.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Dominance ordering in Appendix B is false for Ω≥1, A≫Ω^{1/(2-β)} with β>γ: I3 dominates I4, so the lower bound in (3.6) fails.","rationale":"The reader correctly located the load-bearing issue in the Appendix B dominance ordering. I go further: the ordering is not merely unproved; it fails in a parameter regime covered by Proposition 3.1. For β>γ and A large compared with Ω^{1/(2-β)}, the I3 integral contains the interior Laplace saddle z* while I4 starts at z=A above z*, making I3 exponentially larger than I4. Because the proof uses I4 for both the upper and lower bounds in (3.6), the claimed two-sided estimate cannot hold: the actual Green's function is exponentially smaller than the stated lower bound. The numerical example β=0.8, γ=0.2, Ω=10^6, A=10^3 illustrates the gap. Secondary issues (the A^{-1-γ} versus A^{-1-1/γ} mismatch, the ambiguous integral limits) are real but would not by themselves overturn the result; the false dominance assertion does. This moves the verdict from CONDITIONAL to REJECT.","tokens_in":109,"tokens_out":24262,"duration_ms":298461,"concrete_test":"Evaluate I1–I4 as defined in Appendix B (using I3=∫_1^A for A>1, as the derivation requires; first correct the apparent limits typo if needed) numerically at β=0.8, γ=0.2, d=1, Ω=10^6, A=10^3, dropping the common factor t^{-β/γ-dβ/2}. Test whether I3/I4 is exponentially large (predicted ~e^{1.57e5}) and whether I3 lies far below the claimed lower bound of (3.6), whose exponent is -Ω^{1/(2-γ)}≈-2.15e3. If so, the dominance ordering and the Ω≥1, A≥1 two-sided estimate are refuted.","verdict_should_be":"REJECT","load_bearing_attack":"The reduction to I1/I4 in Appendix B asserts, for Ω≥1 and A≥1, the ordering 0=I2<I1≤I3≤I4, with I3=∫_1^A (as the derivation from §B requires). This ordering is false. Let β>γ, c=1/(1-β), and take A≫Ω^{1/(c+1)}. The Laplace saddle z*=(Ω/c)^{1/(c+1)} of h(z)=Ω/z+z^c lies inside [1,A], so I3≍A^{-1-1/γ}Ω^{q}exp{-KΩ^{1/(2-β)}} with K>0. I4 integrates only over [A,∞); since A>z*, its maximum is at the boundary, so I4 is exponentially smaller than I3. The claimed lower bound in (3.6), for A≥1, has exponent -Ω^{1/(2-γ)} with 1/(2-γ)<1/(2-β), so for large Ω it is exponentially larger than the actual I3 contribution. Concretely, β=0.8, γ=0.2, d=1, t=1, Ω=10^6, A=10^3 (r=10^{15}, t2>r) gives I3~e^{-1.57e5}, I4~e^{-1e15}, while the bound is ~e^{-2.15e3}; thus the lower bound in (3.6) is false. If the printed I3 limits are read literally as empty, the proof would omit the dominant term instead, which is equally fatal.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the boundary value problem on R_+ × R_+ × R^d obtained by combining two Caputo fractional derivatives in the time variables with a spatial diffusion operator. Using the stochastic representation of the solution and known asymptotic estimates for stable densities (Lemma 2.1), Aronson's Gaussian bounds (Lemma 2.2), and a Laplace-method appendix, the authors derive two-sided estimates for the boundary Green's function component G^{(β,γ)}. The main result is Proposition 3.1, which splits into a small-Ω regime (Ω = |x−y|^2 t^{−β} ≤ 1) and a large-Ω regime, with explicit powers and exponential factors depending on A = r^γ t^{−β}. Section 4 sketches an extension to k ≥ 2 fractional derivatives, stated as a conjecture. The central claim is that the Green's function is controlled by a single dominant integral I1 for Ω ≤ 1 and by I4 for Ω ≥ 1.","tokens_in":16623,"tokens_out":9277,"duration_ms":92700,"significance":"If Proposition 3.1 were correct, the paper would supply explicit two-sided kernel bounds for a class of fractional boundary value problems on orthants, with potential applications to multidimensional ruin probabilities, barrier options, and related models. The derivation is free of fitted parameters and relies only on established asymptotic estimates for stable densities and on Aronson's Gaussian bounds; the stochastic-representation route to the kernel decomposition is natural. However, the central estimates are not established as stated: the proof contains an exponent inconsistency and, more seriously, the dominance ordering on which the proof rests is false in an admissible parameter regime. Because the main theorem is the paper's principal contribution, these issues are load-bearing.","major_comments":[{"comment":"The displayed estimate (3.5) contains the factor A^{−1−γ}, while the integral I1 defined in the proof of Proposition 3.1 and in Appendix B has prefactor A^{−1−1/γ}. The asymptotic evaluation of I1 that follows in the text then silently replaces this with A^{−1−γ}. These two exponents are not the same, and the final two-sided estimate depends on which one is correct. The theorem statement is therefore not what the proof establishes. The mismatch must be resolved before the result can be assessed.","section":"§3.2, Proposition 3.1 and its proof"},{"comment":"The integrals I2 and I3 are written with limits that are empty or reversed in the regimes in which they are used: I2 is written as ∫_1^A ... 1_{A<1}, and I3 as ∫_A^1 ... 1_{A>1}. Taken literally, I3 vanishes for A>1, yet the domination orderings for Case 2b (A≥1, Ω≥1) treat I3 as a nonzero contribution that can dominate I1. The intended limits should presumably be ∫_A^1 for I2 and ∫_1^A for I3, but as printed the proof is not coherent.","section":"Appendix B, definitions of I2 and I3"},{"comment":"The assertions '0=I3<I4≤I2≤I1, A≤1', '0=I2<I4≤I3≤I1, A≥1', and the analogous orderings for Ω≥1 are stated without proof and are not correct in general. In particular, for Ω≥1 and A≥1 the claimed ordering 0=I2<I1≤I3≤I4 fails. Take β=0.8, γ=0.2, d=1, t=1, Ω=10^6, and A=10^3 (admissible, e.g. r=10^{15}, t2>r). Then I3 contains the factor exp{−Ω/z − c z^{1/(1−β)}} with β=0.8, whose Laplace saddle lies inside [1,A]; consequently I3 ≍ exp{−K Ω^{1/(2−β)}} with K>0. The integral I4 is integrated only over [A,∞), and since A lies beyond the saddle, I4 is exponentially smaller than I3. Thus I3, not I4, dominates the Green's function in this regime. The displayed lower bound in (3.6) uses α=min(β,γ)=0.2 and is of order exp{−C Ω^{1/(2−γ)}} = exp{−C Ω^{5/9}}, while the actual contribution is exp{−O(Ω^{5/6})}; for Ω=10^6 the former is exponentially larger than the latter, so the claimed lower bound is false. This is not a mere gap in the proof: the theorem as stated is contradicted in this regime.","section":"Appendix B, domination orderings before §B.1"}],"minor_comments":[{"comment":"After defining I1, the sentence 'Since the integral I1 is the main contributor to the estimate' presupposes the domination orderings that are asserted later in Appendix B; the main text should refer the reader to the proof of those orderings or prove them there.","section":"§3.2, proof of Proposition 3.1"},{"comment":"The k-dimensional extension is explicitly conjectural and the proof is omitted. This is acceptable as a discussion, but the paper should clearly state that the extension is not a theorem, especially because the k=2 case on which it is modelled is not fully proved.","section":"§4, Conjecture 4.1"},{"comment":"The symbol cβ appears in I3 and I4 without a precise definition; if it is the constant from Lemma 2.1, it should be stated explicitly, since the later Laplace estimates depend on its positivity.","section":"Appendix B, notation"},{"comment":"There are several typographical inconsistencies in exponents and indicators, e.g. the repeated use of A^{−1−γ} versus A^{−1−1/γ}, and the indicator notation 1_{A∈R+} is redundant. A careful re-editing of the displayed formulas is needed.","section":"Throughout"}],"recommendation":"reject","confidential_remarks":"The paper's main result, Proposition 3.1, is not merely under-proved: the dominance reduction to I1 and I4 is false in an admissible parameter regime, and the displayed lower bound in (3.6) is violated by the counterexample given in the major comments. This is a central, load-bearing error that cannot be fixed by local editing; a corrected theorem would need different regime-dependent estimates. The paper also has an exponent inconsistency between the theorem statement and the proof. I therefore recommend rejection. The stochastic-representation framework is reasonable, and the authors may be able to repair the results, but the current manuscript does not establish its advertised estimates."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nWhat's new: the paper tries to extend the authors' earlier single-fractional-derivative Green's function estimates [JK19] to a mixed problem with two Caputo derivatives in two time variables plus a spatial diffusion. The stochastic representation of the solution as an average over boundary data, with the exit times of two independent stable subordinators, is clean, and the decomposition into boundary components is a sensible path. The k=2 estimates in Proposition 3.1 would be a real extension if they were correct.\n\nWhat's good: the paper uses established machinery (stable density tail asymptotics, Aronson's Gaussian bounds) and lays out the calculation honestly, including the conjecture for k≥2. No curve-fitting; the constants are explicit in principle.\n\nWhere it breaks: the proof of Proposition 3.1 rests on an ordering of four integrals in Appendix B, stated without justification. That ordering is not merely unproved — it is false. For Ω≥1, A≥1, and β>γ, with A large compared to Ω^{(1-β)/(2-β)}, the Laplace saddle of I3 lies inside [1,A], so I3 dominates I4. The lower bound in (3.6) then claims a decay exp{ -Ω^{1/(2-min(β,γ))} } that is much larger than the true order, which is controlled by I3's saddle. This is load-bearing: (3.6) is the main theorem. There are also smaller defects: the theorem display says A^{-1-γ} while the proof derives A^{-1-1/γ}, and the printed limits of I3 look reversed (the integral would be empty for A>1). The k≥2 section is labelled a conjecture, which is honest, but the abstract's phrasing could mislead.\n\nIf the dominance analysis can be redone and the estimates adjusted to the correct saddle, a version of the result may survive. As it stands, the main theorem is not established.\n\nThis paper deserves a serious referee, but not acceptance in its present form. I would not cite it until the proof is fixed.\n\nBest,","headline":"A promising extension of the authors' single-derivative estimates, but the proof of the main two-sided bound has a false dominance ordering and needs major revision.","tokens_in":17244,"tokens_out":7435,"would_cite":false,"duration_ms":62779,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J35","60G52","26A33","35K08"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes explicit two-sided pointwise estimates for the Green's function of a diffusion driven by two Caputo fractional time derivatives on the positive quadrant, covering the boundary, initial-time, and large-separation…","keywords":["fractional boundary value problem","Green's function estimates","Caputo fractional derivative","stable subordinator","two-sided bounds","heat kernel estimates","orthant","ruin probability"],"falsifier":"Fix $\\beta=0.2$, $\\gamma=0.9$, $d=4$, $A=0.5$, and $\\Omega=0.5$, compute the four integrals $I_1$ through $I_4$ numerically, and check whether $I_1$ indeed dominates the sum within the asserted constants; alternatively, Monte-Carlo simulate the process to estimate $G^{(\\beta,\\gamma)}$ directly and compare with (3.5) at the crossovers $A\\approx 1$ and $\\Omega\\approx 1$.","tokens_in":16060,"feed_emoji":"⏱️","tokens_out":6567,"duration_ms":62091,"temperature":0.7,"pith_summary":"This paper sets out to prove explicit two-sided pointwise estimates for the Green's function of a boundary value problem whose time evolution is driven by two Caputo fractional derivatives, one in each of two time variables, together with a spatial diffusion. Such estimates matter because they give uniform control of the solution kernel in every relevant regime—near the boundary, near the initial time, and at large spatial separation—which is the kind of control needed to analyze fractional PDEs, ruin probabilities for multi-asset models, and heat-kernel-based approaches to boundary value problems. The main result, Proposition 3.1, provides the bounds in terms of two dimensionless quantities $\\Omega = |x-y|^2 t^{-\\beta}$ and $A = r^\\gamma t^{-\\beta}$, with dimension-dependent algebraic powers and a stretched-exponential decay whose rate is set by the smaller of the two fractional orders.","feed_headline":"Two-sided bounds pin the mixed fractional heat kernel","feed_subtitle":"Green's function of a two-Caputo problem is controlled up to constants in every regime, including the boundary layer.","key_machinery":"The carrying object is the stochastic representation of the solution as an expectation involving three independent processes: a diffusion $Y_x(s)$ with generator $L$, and two decreasing stable subordinators $X^\\beta$ and $X^\\gamma$ absorbed at zero. Conditioning on the exit times reduces the Green's function to an integral over $s$ of the diffusion density times the stable transition density of one coordinate and the exit-time density of the other; substituting the explicit exit-time density and using Aronson's Gaussian estimates together with the two-sided stable-density asymptotics splits the integral into four pieces, of which one dominates in each regime. The final step applies the Laplace method and the incomplete-gamma asymptotics to the dominant piece.","core_discovery":"The paper's central claim is Proposition 3.1: for the component of the Green's function coming from the first boundary, with $t, r, x, y$ as described, the two-sided estimate splits cleanly at $\\Omega = |x-y|^2 t^{-\\beta} = 1$. For $\\Omega \\le 1$, the kernel behaves like $t^{-\\beta/\\gamma - d\\beta/2} A^{-1-1/\\gamma}$ times a dimensional factor: constant in $d \\le 3$, logarithmic in $d = 4$, and $\\Omega^{(2-d)/2}$ for $d \\ge 5$, where $A = r^\\gamma t^{-\\beta}$. For $\\Omega \\ge 1$, the kernel decays as a power of $\\Omega$ and $A$ times $\\exp\\{-\\left(\\Omega \\max(A^{-1}, 1)\\right)^{1/(2-\\min(\\beta,\\gamma))}\\}$. The proof decomposes the kernel into four integrals $I_1$ through $I_4$, arguing that $I_1$ dominates for $\\Omega \\le 1$ and $I_4$ dominates for $\\Omega \\ge 1$; the stated dominance ordering is asserted without a full proof.","pith_inferences":["The dominance ordering of the four integrals is the step most likely to fail in extreme asymmetry $\\beta \\ll \\gamma$; a direct proof of those inequalities would be a natural complement, and numerical experiments in the $\\beta=0.2$, $\\gamma=0.9$ corner would settle whether the stated exponents hold.","The stretched-exponential exponent $1/(2-\\min(\\beta,\\gamma))$ suggests a general principle: in a mixture of fractional time derivatives, the smaller index controls the large-time and large-distance tail, just as the smaller drift controls large deviations in classical settings.","The logarithmic factor at $d=4$ (and $d=2k$ in the conjecture) is the fractional analogue of the classical heat kernel's critical dimension; this suggests that for fractional boundary problems the critical dimension equals twice the number of fractional coordinates, a pattern the authors do not explicitly name.","Because the full Green's function splits as a sum of boundary components, the estimates imply two-sided bounds on the harmonic measure of the orthant for the mixed fractional diffusion, which are exactly what barrier-option and default-time models require; the paper mentions these applications but does not develop them."],"forward_implications":["The two-sided bounds provide the kernel estimates needed to study well-posedness of mixed Caputo boundary value problems and to extend the results to spatial operators that generate stable-like Feller processes.","The uniform control in the boundary coordinate $r$, including $r \\to 0$, means the boundary potential inherits the same bounds, giving concrete control of the solution's dependence on boundary data.","The multidimensional extension (Conjecture 4.1) predicts analogous two-sided estimates for orthants in dimension $k \\ge 2$, with the critical-dimension logarithmic behavior appearing at $d = 2k$.","The estimates support the programme of replacing each fractional derivative by a generalised Caputo derivative with a Lévy-type kernel whose density is comparable to a $\\beta$-stable density, yielding estimates for the more general equation (1.3).","For ruin-probability and default-time applications, the bounds give explicit control of the probability that one of two stable processes hits zero first, with the decay governed by the slower index $\\min(\\beta,\\gamma)$."],"supporting_citations":[{"why":"Supplies the two-sided Gaussian bounds for the diffusion transition density used in Lemma 2.2.","marker":"[Aro67]"},{"why":"Supplies the asymptotic behavior of stable densities $w_\\alpha(r)$ that underlies the two-sided estimates of the integrands.","marker":"[UZ99]"},{"why":"Supplies the formula for the exit-time density $\\mu_0^\\alpha(s)$ used to write the Green's function as an integral over the diffusion time.","marker":"[MS04]"},{"why":"The predecessor whose single-time-fractional evolution estimates are extended here to the two-boundary setting.","marker":"[JK19]"},{"why":"Provides the Laplace asymptotics applied to reduce the dominant integral $I_4$ to the displayed stretched-exponential form.","marker":"[dB81]"}],"fun_headline_variants":["Two-sided bounds pin mixed fractional heat kernel","Sharp estimates for two-Caputo Green's function","Regime split controls fractional Green's function","Boundary layer sets Green's function asymptotics","Mixed fractional kernel: two-sided bounds in all regimes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, in each regime, one of the four constituent integrals dominates all the others—an ordering asserted without proof—so that if the ordering fails in any parameter range, the stated exponents do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Two-sided bounds pin mixed fractional heat kernel","Sharp estimates for two-Caputo Green's function","Regime split controls fractional Green's function","Boundary layer sets Green's function asymptotics","Mixed fractional kernel: two-sided bounds in all regimes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000583,"raw_usage":{"total_tokens":2788,"prompt_tokens":1032,"completion_tokens":1756,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":648,"completion_tokens_details":{"reasoning_tokens":1685}},"tokens_in":648,"tokens_out":1756,"duration_ms":17285,"temperature":1.0,"reasoning_tokens":1685,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:24:16.206675+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fix $\\beta=0.2$, $\\gamma=0.9$, $d=4$, $A=0.5$, and $\\Omega=0.5$, compute the four integrals $I_1$ through $I_4$ numerically, and check whether $I_1$ indeed dominates the sum within the asserted constants; alternatively, Monte-Carlo simulate the process to estimate $G^{(\\beta,\\gamma)}$ directly and compare with (3.5) at the crossovers $A\\approx 1$ and $\\Omega\\approx 1$.","supporting_citations":[],"review_version":1}