{"id":"ba0707bf-e032-4f51-841b-22a0ca0e14dd","arxiv_id":"1908.08319","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"It derives two representation formulas for solutions of linear Caputo fractional differential equations with variable coefficients and intermediate-point initial conditions, using a modified measurable forcing term that encodes the memory of the initial interval.","lead":"This math paper proves new formulas that write the solution of a fractional-order linear differential equation as an explicit integral expression, valid when the initial data are given over a whole interval rather than at a single point. Such formulas are the key to later control and game theory problems for fractional systems that remember past states.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central representation formula (4.8) rests on the unproved integral identity in Lemma 4.1 and on an unverified Fubini–Tonelli step with the possibly unbounded term b*; both must be supplied before the theorem is established.","rationale":"The reader's weakest assumption is exactly the unproved Lemma 4.1 and the sketched Fubini step. My reading of the proof confirms that these are the only real weak points: Propositions 3.1 and 3.2 are proved in detail, the dual equation argument in Proposition 3.3 is essentially complete, and Theorem 4.1's Fubini step is justified because the involved functions are bounded. The novelty of the paper lies in extending formula (4.1) to intermediate initial conditions, and that extension passes exclusively through Lemma 4.1 and the forcing term b*. The lemma may be true, since it is attributed to [22, Theorem 13.10], but the identity in (4.7) and the companion formula for ψ are not reconstructable from the text alone, so formula (4.8) is not yet established to the standard claimed. The same holds for the assertion that the double integrals in the final step of Theorem 4.2 converge despite b* being unbounded: continuity of ψ does not by itself justify the interchange, and the paper supplies no explicit estimate. Thus the prudent disposition is conditional acceptance: the surrounding analysis is solid and the formulas are plausible, but these two supporting facts must be written out. I would not move the verdict away from conditional on the current evidence.","tokens_in":16426,"tokens_out":10846,"duration_ms":102707,"concrete_test":"Independently derive Lemma 4.1 in full from definition (4.6), without citing [22, Theorem 13.10], and in the same derivation prove a quantitative bound of the form |ψ(t)-ψ(t*)|≤C(t-t*)^α for t∈(t*,θ], or an L^p integrability statement for b* that makes the Fubini–Tonelli step in Theorem 4.2 valid. If such a bound cannot be obtained, compute the repeated integral in (4.5) for the concrete case α=1/2, t0=0, t*=1, A=0, φ≡1 and check whether both sides of (4.7) coincide and the double integral is finite; a failure would directly invalidate formula (4.8).","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 4.1 is the hinge of Theorem 4.2: its first equality in (4.7) is what converts the nonlocal initial segment [t0,t*] into the forcing term b*, and the second formula in (4.7) is used to define b* in (4.9). The proof is a single sentence delegating the argument to 'the scheme from [22, Theorem 13.10]' and no derivation is given in the paper. If that identity, or the displayed formula for ψ, has a hidden restriction, formula (4.8) simply does not follow. The second gap is in the Fubini–Tonelli step at the end of Theorem 4.2. The paper explicitly notes that b* is not essentially bounded in general and says only that 'equality (4.12) and the inclusion ψ*(·)∈C([t*,θ],R^n) should be used'. From (4.12), b*(t)=(ψ(t)-ψ(t*))/(t-t*)^α+b(t), so one must control the quotient by (t-t*)^α. Continuity of ψ alone yields no quantitative bound, and the integral in (4.8) has the singular factor (t-τ)^(α-1); without an estimate such as |ψ(t)-ψ(t*)|≤C(t-t*)^α (or an L^p bound for b*) the interchange of integrals used to verify (4.14) is not justified. These are not cosmetic omissions: both are load-bearing for the claimed representation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a linear fractional differential equation with a Caputo derivative of order α∈(0,1) and variable coefficients, subject to a Cauchy condition specified on an initial interval [t0,t*] rather than at a single point. The authors introduce a nonsingular fundamental matrix F(t,s) through a regularized Volterra-type integral equation, establish its Hölder continuity with respect to both variables, prove a dual characterization, and then derive two representation formulas for the solution. The main result (Theorem 4.2, Eq. (4.8)) expresses the solution for t≥t* as an affine function of the terminal value w*(t*) plus a forced integral involving an explicitly defined function b* that encodes the memory of the initial segment. A second formula (Corollary 4.1, Eq. (4.15)) rewrites this in terms of w*(t0) and a double integral. The central derivation follows a contraction-mapping argument for F and a verification that the proposed formulas satisfy the equivalent integral equation from Proposition 2.1.","tokens_in":16777,"tokens_out":12477,"duration_ms":119069,"significance":"If the representation formulas are correct, they extend the variation-of-constants formula to nonlocal Caputo initial conditions, a setting relevant for control problems and differential games with fractional dynamics. The paper provides a self-contained study of the fundamental solution matrix with explicit constants (MF, HF in Propositions 3.1 and 3.2), which is a useful contribution in its own right. The main strategy is sound: the proofs of the fundamental matrix properties use standard contraction and Bellman-Gronwall techniques, and the representation formulas are verified by substitution into the integral equation. The chief unresolved points are two technical justifications in Section 4: the integral identity in Lemma 4.1 is delegated to an external source, and the Fubini-Tonelli interchange involving the singular function b* is only sketched. Both are load-bearing for the claimed representation, but they appear fixable with additional estimates.","major_comments":[{"comment":"Lemma 4.1 is the hinge of Theorem 4.2: its first equality in (4.7) converts the nonlocal memory term over [t0,t*] into a forcing term over [t*,t], and its second equality defines the singular part of b* in (4.9). The proof, however, consists of a single sentence referring to 'the scheme from [22, Theorem 13.10]' without stating the theorem or verifying its hypotheses. The exact result in [22] should be quoted with its conditions, or a complete proof should be given, because without (4.7) the representation formula (4.8) does not follow.","section":"Section 4, Lemma 4.1"},{"comment":"In the proof of Theorem 4.2, the author claims that the equality analogous to (4.3) is proved by the same steps as in Theorem 4.1, noting only that 'the function b* is not essentially bounded in general' and that 'equality (4.12) and the inclusion ψ*(·)∈C([t*,θ],Rn) should be used' when applying Fubini-Tonelli. This is insufficient. A rigorous proof must establish the absolute integrability of the double integral containing b*. For example, from (4.12) one obtains |b*(τ)| ≤ C(τ−t*)^{−α} + |b(τ)|, and the inner kernel integral ∫_{r}^{t}(t−τ)^{α−1}(τ−r)^{α−1}dτ equals B(α,α)(t−r)^{2α−1}, which is integrable against this bound. Alternatively, the explicit expression (4.11) for ψ* can be used. The authors should include such an estimate so that the interchange of the order of integration in the proof of (4.14) is fully justified.","section":"Section 4, Theorem 4.2"},{"comment":"The assertion that the function y defined by (4.13) is continuous on [t0,θ] is not proven. The formula for y involves b*, which near t* is a difference of two singular terms (see (4.9)); continuity at t* requires showing that these singularities cancel and that the resulting limit equals w*(t*). The paper states only that this follows from the inclusions and Proposition 3.2. A verification of the cancellation, or an alternative argument, is needed before applying Proposition 2.1 to conclude that y coincides with the solution.","section":"Section 4, Eq. (4.13)"}],"minor_comments":[{"comment":"In the second equality of (4.7), the factor (t−t*)^α is not clearly separated in the typeset display; a cleaner presentation would avoid possible confusion about which terms carry the singular factor.","section":"Section 4, Lemma 4.1"},{"comment":"The verification of the identity used in (4.10), namely ψ*(t*) = (w*(t*)−w*(t0))/Γ(1−α), relies on the reflection formula for the gamma function and the definition of the Riemann-Liouville integral; it would be helpful to mention this explicitly, as it is a key step.","section":"Section 4, Theorem 4.2"},{"comment":"In the proof of equality (4.16), the beta integral identity ∫_{t*}^{t}(t−τ)^{α−1}(τ−t*)^{−α}dτ = Γ(α)Γ(1−α) is used implicitly; citing this standard identity would improve readability.","section":"Section 4, Corollary 4.1"},{"comment":"The paper relies on Proposition 2.1 from the author's earlier work [11] for existence and uniqueness. This is acceptable, but since it is central to the method, a brief indication of the proof or a more precise theorem statement would make the paper more self-contained.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is likely correct in its main claims, and the flaws are technical gaps in proof presentation rather than fundamental errors. The representation formulas are a useful extension of known results to nonlocal Caputo initial conditions. However, the proof of Lemma 4.1 is delegated to a vague reference, and the Fubini-Tonelli step in Theorem 4.2 is only sketched; both are load-bearing and should be supplied in full. The editor may also note the heavy reliance on the author's own prior papers for existence and uniqueness, which is acceptable but should be cited precisely. No issues of novelty or scope are apparent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper extends Duhamel-type representation formulas for linear Caputo fractional ODEs with variable coefficients to the case of nonlocal initial data on an interval [t0,t*]. As far as I can tell, that case is genuinely new; Bourdin and the standard texts only cover t*=t0. The main formula (4.8), with the modified forcing term b* in (4.9), is the real contribution, and the derivation is mostly detailed and standard. The fundamental solution matrix F is handled carefully, with Hölder continuity and a dual equation, and the comparison with Bourdin is fair. The reliance on the author's earlier existence results is fine; there's no circularity.\n\nThe soft spots are two. Lemma 4.1 is the hinge—it converts the nonlocal memory from the initial interval into the forcing term b*. Its proof is one sentence delegating to Samko-Kilbas-Marichev. That's thin for a lemma that carries the new idea. A referee should ask for a full derivation, or at least a precise statement of the theorem being used and a check of its hypotheses. This is fixable, not fatal.\n\nThe second concern is the Fubini-Tonelli interchange in Theorem 4.2. The paper notes that b* is not essentially bounded and says (4.12) and continuity of ψ suffice. The stress-test note worries that continuity alone gives no bound. I checked: continuity on a compact interval makes ψ bounded, and then (4.12) gives |b*(τ)| ≤ C (τ-t*)^(-α) + |b(τ)|. That's enough to make the repeated integral absolutely convergent, because the kernel (τ-ξ)^(α-1) is integrable against that singularity. So the stress-test's demand for a Hölder estimate is overstated. Still, the paper's sketch is too terse; the estimate should be written out.\n\nOverall the central argument holds up. This paper is for readers in fractional differential equations and control who need explicit representations for nonlocal initial data. It deserves a serious referee. I'd send it to review with a request for a full proof of Lemma 4.1 and a clear justification of the interchange.","headline":"A useful and likely correct extension of representation formulas to intermediate-point initial data; two proof gaps need filling, but neither looks fatal.","tokens_in":17269,"tokens_out":11913,"would_cite":true,"duration_ms":103005,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["26A33","34A08","34A30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves explicit representation formulas for solutions of linear Caputo fractional equations with variable coefficients when the initial condition is prescribed on an interval, not at a single point.","keywords":["fractional differential equations","Caputo derivative","fundamental solution matrix","representation formula","variation of constants","intermediate initial condition","Hölder continuity","Cauchy problem"],"falsifier":"Take $t_0=0$, $t_*=1$, $\\alpha=1/2$, $\\phi(\\tau)=1$ and evaluate the identity in Lemma 4.1 numerically at several $t>1$; both sides must agree. A disagreement at any one $t$, or failure of the resulting $b_*$ from (4.9) to be locally integrable on $(t_*,\\vartheta]$, would falsify the proof of Theorem 4.2.","tokens_in":16191,"feed_emoji":"📐","tokens_out":10652,"duration_ms":92153,"temperature":0.7,"pith_summary":"The paper aims to prove that solutions of a variable-coefficient linear system with a Caputo fractional derivative can be written in explicit closed form even when the initial condition is not a single value but a prescribed history on an interval $[t_0,t_*]$. Such nonlocal initial data arise naturally in control problems and differential games for fractional systems, where explicit representation formulas are the standard building block for feedback construction and numerical methods. The main result, Theorem 4.2, gives a Duhamel-type representation in which the whole pre-history affects the future only through a modified forcing term $b_*$ built from $w_*$. Along the way the paper develops a careful theory of the regularized fundamental solution matrix $F$: existence, boundedness, H\\\"older continuity in both arguments, and a dual definition.","feed_headline":"Caputo formula solves systems with nonlocal initial data","feed_subtitle":"The memory of the initial interval becomes a single singular forcing term in a Duhamel-type formula.","key_machinery":"The central object is the regularized fundamental solution matrix $F(t,s)$, defined as the continuous solution of integral equation (3.4) and related to the Riemann\\u2013Liouville fundamental matrix $Z$ by $F(t,s)=(t-s)^{1-\\alpha}Z(t,s)$; it is bounded and H\\\"older continuous in both variables on $\\Omega=\\{(t,s): t_0\\le s\\le t\\le\\vartheta\\}$, and it satisfies the dual equation $G=F$ (Proposition 3.3). The step that makes the general intermediate-point case work is Lemma 4.1's integral identity, which turns the left-sided fractional integral of the initial history into a right-sided integral over $[t_*,t]$: $\\int_{t_0}^{t_*}\\phi(\\tau)/(t-\\tau)^{1-\\alpha}\\,d\\tau=\\int_{t_*}^{t}\\psi(\\tau)/((t-\\tau)^{1-\\alpha}(\\tau-t_*)^\\alpha)\\,d\\tau$, with $\\psi$ defined by (4.6). This identity converts the nonlocal memory of $w_*$ into the singular but manageable forcing term $b_*$ and is what the representation formula (4.8) rests on.","core_discovery":"At the center is Theorem 4.2. For the Cauchy problem $({}^C\\!D^\\alpha_{t_0+}x)(t)=A(t)x(t)+b(t)$, $x(t)=w_*(t)$ on $[t_0,t_*]$, the solution is claimed to be $$x(t)=\\Bigl(\\operatorname{Id}+\\int_{t_*}^{t}\\frac{F(t,\\tau)A(\\tau)}{(t-\\tau)^{1-\\$\\alpha$}}\\,d\\tau\\Bigr)w_*(t_*)+\\int_{t_*}^{t}\\frac{F(t,\\tau)b_*(\\tau)}{(t-\\tau)^{1-\\$\\alpha$}}\\,d\\tau,\\quad t\\in[t_*,\\vartheta],$$ where $F$ is the regularized fundamental solution matrix defined by (3.4) and $b_*$ is the forcing term (4.9) that carries the memory of $w_*$. The proof rewrites the fractional integral of the initial history over $[t_0,t_*]$ as an integral over $[t_*,t]$ with a singular kernel (Lemma 4.1), so the nonlocal data become part of the forcing term and the standard Duhamel argument applies. Corollary 4.1 gives an equivalent form with the history appearing as an explicit double integral.","pith_inferences":["Because Lemma 4.1 only needs the fractional integral of the history to make sense, the memory-to-forcing conversion should extend to initial histories rougher than $AC^\\alpha$; testing this would require proving an $L^p$ or distributional version of the identity.","The singular $(\\tau-t_*)^{-\\alpha}$ factor in $b_*$ means any numerical discretization of (4.8) must resolve a weak endpoint singularity; the paper flags the analogous difficulty for (4.15) but does not propose a scheme, leaving adaptive or product-integration methods as a natural next step.","Since $F$ satisfies a dual equation, the same formulas should transpose to adjoint or backward-in-time problems in dynamic programming for fractional systems, though the paper does not develop that direction."],"forward_implications":["For $t_*=t_0$, formula (4.8) reduces to Theorem 4.1 and recovers the known Duhamel representation for standard Caputo initial conditions.","The whole pre-history of the solution enters the future only through $b_*$; once $w_*(t_*)$ and $b_*$ are known, the solution on $[t_*,\\vartheta]$ is determined by the same convolution as in the pointwise-initial-value case.","The fundamental matrix $F$ is bounded and H\\\"older continuous in both variables and satisfies the dual identity $G=F$, so the same object describes forward and backward propagation on $\\Omega$.","The alternative formula (4.15) is simpler in shape but has a second term that need not vanish as $t\\downarrow t_*$; this is an explicit warning that numerical schemes based on (4.15) must treat the neighbourhood of $t_*$ specially."],"supporting_citations":[{"why":"Supplies the state-transition matrix and Duhamel formula for the $t_*=t_0$ case that this paper extends to intermediate initial points.","marker":"[3]"},{"why":"Provides weighted H\\\"older regularity of Riemann\\u2013Liouville fractional integrals, used to establish continuity of $F$.","marker":"[4]"},{"why":"Supplies the Bellman\\u2013Gronwall lemma behind the boundedness and H\\\"older continuity estimates for the fundamental matrix.","marker":"[7]"},{"why":"Gives the fractional integral and derivative facts (Propositions 1.1 and 1.2) used throughout, including the representation of $AC^\\alpha$ functions.","marker":"[9]"},{"why":"Quoted for existence, uniqueness, and the integral equation (2.4) characterizing the solution of the Cauchy problem.","marker":"[11]"},{"why":"Its Theorem 13.10 is the delegated source of Lemma 4.1's integral identity, which converts the initial memory into the forcing term $b_*$.","marker":"[22]"}],"fun_headline_variants":["Caputo nonlocal data recast as a singular forcing term","Singular kernel absorbs interval initial data in Caputo ODEs","New Duhamel formula handles Caputo systems with memory data","Caputo memory folded into a single forcing term via Duhamel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is Lemma 4.1's integral identity, whose proof is delegated to a cited theorem rather than carried out; if that identity fails, or if the resulting $b_*$ is not integrable, formula (4.8) collapses.","fun_headline_variants_meta":{"raw":{"variants":["Caputo nonlocal data recast as a singular forcing term","Singular kernel absorbs interval initial data in Caputo ODEs","New Duhamel formula handles Caputo systems with memory data","Caputo memory folded into a single forcing term via Duhamel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001046,"raw_usage":{"total_tokens":4378,"prompt_tokens":909,"completion_tokens":3469,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":525,"completion_tokens_details":{"reasoning_tokens":3396}},"tokens_in":525,"tokens_out":3469,"duration_ms":23730,"temperature":1.0,"reasoning_tokens":3396,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:42:25.787171+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $t_0=0$, $t_*=1$, $\\alpha=1/2$, $\\phi(\\tau)=1$ and evaluate the identity in Lemma 4.1 numerically at several $t>1$; both sides must agree. A disagreement at any one $t$, or failure of the resulting $b_*$ from (4.9) to be locally integrable on $(t_*,\\vartheta]$, would falsify the proof of Theorem 4.2.","supporting_citations":[{"cited_title":"Bourdin, Cauchy–Lipschitz theory for fractional mul ti-order dy- namics: State-transition matrices, Duhamel formulas and d ual- ity theorems","cited_arxiv_id":null,"evidence_quote":"Supplies the state-transition matrix and Duhamel formula for the $t_*=t_0$ case that this paper extends to intermediate initial points."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides weighted H\\\"older regularity of Riemann\\u2013Liouville fractional integrals, used to establish continuity of $F$."},{"cited_title":"Diethelm, The Analysis of Fractional Differential Equations","cited_arxiv_id":null,"evidence_quote":"Supplies the Bellman\\u2013Gronwall lemma behind the boundedness and H\\\"older continuity estimates for the fundamental matrix."},{"cited_title":"Gomoyunov, Fractional derivatives of convex Lyapu nov functions and control problems in fractional order systems","cited_arxiv_id":null,"evidence_quote":"Gives the fractional integral and derivative facts (Propositions 1.1 and 1.2) used throughout, including the representation of $AC^\\alpha$ functions."},{"cited_title":"Gomoyunov, Solution to a zero-sum differential game with frac- tional dynamics via approximations","cited_arxiv_id":null,"evidence_quote":"Quoted for existence, uniqueness, and the integral equation (2.4) characterizing the solution of the Cauchy problem."},{"cited_title":"Samko, A.A","cited_arxiv_id":null,"evidence_quote":"Its Theorem 13.10 is the delegated source of Lemma 4.1's integral identity, which converts the initial memory into the forcing term $b_*$."}],"review_version":1}