{"id":"61488a05-c83f-4ad5-a63b-899aaedb9ee2","arxiv_id":"2607.27150","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"CTRW scaling limits are semi-Markov time-changed processes whose laws solve general non-local evolution equations, with new pointwise theory for killed subordinate Brownian motion on domains.","lead":"These notes unify Continuous Time Random Walks, their scaling limits, and the non-local evolution equations those limits satisfy, including coupled and variable-order cases. They also give new pointwise well-posedness results for killed time-changed processes on bounded domains with initial data outside the generator domain.","discovery_kind":"review","skeptic_critique":{"model":"moonshotai/kimi-k3","headline":"No significant objection identified: the least secure chain (Theorems 5.6–5.9) survives line-by-line scrutiny, and the reader's flagged regularity/finite-measure restrictions are stated scope limits, not hidden gaps.","rationale":"The reader identified Section 5's reliance on A-regularity of D, finite Lebesgue measure, and the subordinate-Brownian-motion restriction as the weakest assumption. My independent scrutiny of the same section confirms this is the right soft spot, but finds it is a scope limitation rather than a correctness problem: every restriction appears explicitly in the theorem hypotheses, and the proof steps that could conceal a gap (the off-diagonal differentiation under Hunt's formula, the spectral weak formulation, the Laplace-transform commutations, τ_D = S(T_D) a.s.) each check out under the stated assumptions. Elsewhere the paper is a careful synthesis (Theorems 3.1, 3.6, 4.10 imported from [105; 87; 13; 27] with accurate statements), and the new proofs I examined in detail — the OCTRW embeddings (Theorems 2.4–2.5), the repair of [106, Theorem 5.1] via Theorem 4.3, and the pointwise killed theory — are argued at a level that can be verified line by line, with self-disclosed limitations (Proposition 5.4, Remark 5.10, footnote 9). Explicit example computations I spot-checked (the MSD values in (3.59), (3.63)–(3.64), the coupled scaling in (3.33)) are correct. The appropriate verdict therefore remains CONDITIONAL: the work is primarily expository notes with genuine incremental theorems whose strongest claims carry explicit regularity restrictions worth flagging to readers, but no load-bearing error was found and the proposed spectral/Monte-Carlo cross-check is a worth-running confirmation rather than a suspected failure.","tokens_in":61829,"tokens_out":12240,"duration_ms":428071,"concrete_test":"Cross-check Theorems 5.6 and 5.9 in the fully computable canonical case: D = unit ball, ψ(λ) = λ^{α/2} (fractional Laplacian), ϕ(λ) = λ^β, f ≡ 1. (i) Using Dirichlet eigenpairs {λ_n, e_n} of (−Δ)^{α/2} on D, form U(x,t) = Σ_n E_β(−λ_n t^β)⟨1,e_n⟩e_n(x); using |E_β(−λt^β)| ≤ C/(1+λt^β), Weyl asymptotics λ_n ~ n^{α/d}, and sup-norm eigenfunction bounds, verify the series and its termwise ψ(−Δ)-image converge in C² on compact subsets for t > 0, so the pointwise operator applies termwise and each mode satisfies the Caputo ODE — independently confirming ϕ(∂_t)(U−1) = −ψ(−Δ)U pointwise. (ii) Monte Carlo: simulate X^D = B(σ(L(t))) killed at exit using exact inverse-subordinator sampling (the authors' own algorithms in [25]) and compare P_x(τ_D > t) against the truncated series at several (x,t). Agreement within Monte Carlo error validates the weak→pointwise transfer and the Laplace-commutation;","verdict_should_be":"UNCHANGED","load_bearing_attack":"I treated Section 5 — the genuinely new content — as the place where the central claim is least secure, and stress-tested its two least mechanical steps.\n\n(1) The weak-to-pointwise transfer in Theorem 5.6. The proof establishes the L²-weak formulation (5.32) spectrally (legitimate: l(D)<∞ gives compactness of P^D_t by Theorem 5.5, and φ∈C^∞_c(D) lies in the L²-domain by Proposition 5.4), then swaps ψ(−Δ) from φ onto u via integration by parts, requiring u(·,s)=P^D_s 1 ∈ C²_loc(D). The C^k bounds come from decomposition (5.34) and Lemma 5.2's off-diagonal derivative bounds. This is exactly where a gap could hide, because p(t,·) may explode on the diagonal (geometric stable case) and the bounds depend on δ_D(x). But the decomposition keeps all differentiations off-diagonal: the first term uses translation invariance (∂^β P_t η = P_t ∂^β η), the second uses symmetry (5.16) plus |x−A(T_D)| ≥ δ_D(x) under P_y, the third differentiates p^D only for |x−y| ≥ dist(V,U^c) > 0. All bounds are uniform in t>0 via sup_{s>0} h^a s^{−b} e^{−h²/4s} < ∞. The upgrade of (5.35) from a.e. to everywhere uses continuity of both sides, which is justified. The boundary-vanishing of P^D_t u for merely bounded u (Lemma 5.1) survives the diagonal blow-up because ∫_{B(z,δ)} p(t,x,y)dy ≤ ∫_{|h|<2δ} p(t,h)dh → 0 as δ→0 by translation invariance.\n\n(2) Theorem 5.9's Laplace-transform commutations. The identity LU(x,λ) = ϕ(λ)/λ · Lu(x,ϕ(λ)) and the swap ψ(−Δ)L = Lψ(−Δ) require uniform-in-time C² bounds near x and global boundedness — inherited from Theorem 5.7's proof. The key probabilistic identity τ_D = S(T_D) a.s. uses \"S jumps at a fixed time with probability 0\" at the random time T_D; this is legitimate because S ⊥ A and T_D is A-measurable, so one conditions on T_D. The authors also disclose the sharp edges themselves: Proposition 5.4 (C²_c ⊄ C_0-domain, motivating the pointwise approach) and Remark 5.10 (non-uniqueness of pointwise solutions).\n\nConclusion: the new theorems' hypotheses (D A⊂R^","agreement_with_reader":"agree"},"referee_report":null,"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"These notes do what they claim: they give a self-contained probabilistic route from CTRWs through scaling limits to non-local evolution equations, and they do it cleanly. The bulk is synthesis of Meerschaert–Straka, Baeumer–Meerschaert, Kolokoltsov, Savov–Toaldo and related work, but the organization is useful and the technical care is high.\n\nWhat is actually new sits in three places. Theorems 2.4–2.5 give Markov embeddings for both the left-continuous and right-continuous versions of the overshooting CTRW; those statements look new and the proofs are direct. The “universal” Z± harmonic construction for coupled limits is a clear exposition of the method from their recent arXiv note, made usable. The real technical advance is Section 5: for subordinate Brownian motion killed on exit from a regular finite-measure domain, they obtain pointwise solutions of the non-local-in-time equation with initial data only in Bb(D) ∩ C²_loc(D). That class properly contains functions outside the C0-domain of the killed generator (they prove C²c is not in that domain when there is a jump measure), so the pointwise theory is not just a re-packaging of the abstract Cauchy problem. The Laplace-transform and weak-to-pointwise arguments survive line-by-line checks; the off-diagonal derivative bounds and the a.s. identity τD = S(TD) are handled correctly.\n\nSoft spots are mostly scope, not gaps. The killed theory is restricted to subordinate BM, regular D and finite Lebesgue measure (needed for compactness and boundary vanishing). They flag non-uniqueness of pure pointwise solutions themselves. The rest of the paper is notes, so novelty is moderate by design. Citations are appropriate; self-citations supply prior lemmas that are used, not circular.\n\nThis is for people who work on anomalous diffusion, time-changed Markov processes or non-local PDEs and want a single reliable reference that also contains the killed pointwise results. It deserves a serious referee. I would accept it for peer review and would cite the Section 5 statements and the OCTRW embeddings.","headline":"Solid, carefully written notes that cleanly synthesize CTRW scaling limits and non-local equations, with genuine incremental theorems on OCTRW embeddings and pointwise killed problems for subordinate BM.","tokens_in":64366,"tokens_out":542,"would_cite":true,"duration_ms":13727,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K50","60K15","35R11","35R09"],"pacs":[],"model":"grok-4.5","headline":"Scaling limits of continuous-time random walks are time-changed processes whose laws solve general non-local evolution equations for anomalous diffusion.","keywords":["anomalous diffusion","continuous time random walks","semi-Markov processes","time-changed Markov processes","non-local equations","fractional kinetics","subordinators","killed processes"],"falsifier":"Simulate a scaled CTRW whose joint jumps converge to a known Feller process with strictly increasing second coordinate, estimate the exit-time distribution or occupation measure of the limit on a regular bounded domain, and check whether it coincides with the numerical solution of the claimed non-local equation (or systematically fails on an irregular domain).","tokens_in":63868,"feed_emoji":"🎲","tokens_out":1066,"duration_ms":38598,"temperature":0.7,"pith_summary":"These notes give a self-contained probabilistic account of anomalous diffusion as the scaling limit of continuous-time random walks (CTRWs). Jumps and waiting times that form a Markov chain converge, after scaling, to a Feller process whose second coordinate is a strictly increasing clock; the position process time-changed by the inverse of that clock is the limit. The same construction yields both the ordinary CTRW limit and the overshooting version, preserves a semi-Markov embedding, and produces governing equations that are non-local in time and, when jumps and waits are coupled, jointly non-local in space and time. The theory covers abstract Cauchy problems for uncoupled cases, harmonic problems for auxiliary Markov processes in the coupled case, variable-order aggregation, and pointwise non-local parabolic equations for killed subordinate Brownian motion on regular domains of finite measure, even for initial data outside the generator domain.","feed_headline":"CTRW limits yield non-local equations for anomalous diffusion","feed_subtitle":"Scaled jumps and waits become time-changed processes that solve equations far beyond classical fractional kinetics.","key_machinery":"The continuous-mapping theorem that sends the scaled joint space-time jump chain to a Feller process (A,S) and then, via the generalized inverse L of S, to the undershoot and overshoot limits; together with the age/remaining-lifetime Markov embeddings and the construction of the auxiliary processes Z± whose exit harmonic functions recover the evolutions of X and X+.","core_discovery":"Scaling limits of (overshooting) CTRWs are the time-changed processes X(t)=A(L(t)−)+ and X+(t)=A(L(t)). Under the stated Feller and Markov-additive assumptions their laws solve the corresponding non-local evolution equations—abstract Cauchy problems driven by generalized time derivatives when uncoupled, and harmonic problems for the auxiliary processes Z± when coupled—and, for killed subordinate Brownian motion on regular finite-measure domains, the map U(x,t)=E_x[f(X^D(t))] is a pointwise solution of ϕ(∂_t)(U−U(·,0))=−ψ(−Δ)U for merely bounded C²_loc initial data.","pith_inferences":["The auxiliary-process harmonic method should extend, with only notational changes, to other regenerative time changes beyond subordinators, giving a route to non-local equations for a wider class of semi-Markov limits.","Removing the subordinate-Brownian-motion restriction in the killed theory would let the same pointwise statements cover jump diffusions and other Feller processes used in applications with rough boundaries.","Explicit MSD formulae for general Markov-additive limits would turn the qualitative sub-/super-diffusion dichotomy into quantitative diagnostics for experimental trajectories."],"forward_implications":["Coupled jump-wait CTRWs produce governing operators that mix space and time and cannot be written as a sum of separate fractional derivatives.","Space-dependent waiting-time exponents produce asymptotic aggregation at the spatial minimum of the exponent whenever that minimum is sufficiently small relative to the values at infinity.","Exit-time tails of killed subdiffusions on regular finite-measure domains solve pointwise non-local parabolic equations for initial data outside the C0-domain of the killed generator.","The same continuous-mapping and semi-Markov framework recovers both ordinary and overshooting limits and distinguishes their mean-squared-displacement regimes (subdiffusive, diffusive, or infinite).","Variable-order and killed problems become analytically tractable once the limit process is identified as a time-changed Markov process."],"fun_headline_variants":["CTRW scaling limits yield non-local equations for anomalous diffusion","Scaled CTRWs become time-changed processes solving non-local equations","CTRW limits link semi-Markov processes to general non-local dynamics","Overshooting CTRW limits solve abstract Cauchy problems beyond fractional","CTRW scaling produces time-changed processes for non-local diffusion"],"cache_read_input_tokens":49280,"weakest_assumption_plain":"The pointwise killed theory needs the domain to be regular for the subordinate Brownian motion and to have finite volume, and it restricts the spatial motion to subordinate Brownian motion rather than a general Feller process.","fun_headline_variants_meta":{"raw":{"variants":["CTRW scaling limits yield non-local equations for anomalous diffusion","Scaled CTRWs become time-changed processes solving non-local equations","CTRW limits link semi-Markov processes to general non-local dynamics","Overshooting CTRW limits solve abstract Cauchy problems beyond fractional","CTRW scaling produces time-changed processes for non-local diffusion"]},"model":"grok-4.5","effort":"low","cost_usd":0.003667,"raw_usage":{"total_tokens":1189,"prompt_tokens":758,"num_sources_used":0,"completion_tokens":73,"cost_in_usd_ticks":36668000,"prompt_tokens_details":{"text_tokens":758,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":358,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":758,"tokens_out":73,"duration_ms":6670,"temperature":1.0,"reasoning_tokens":358,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-30T11:00:26.923992+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Simulate a scaled CTRW whose joint jumps converge to a known Feller process with strictly increasing second coordinate, estimate the exit-time distribution or occupation measure of the limit on a regular bounded domain, and check whether it coincides with the numerical solution of the claimed non-local equation (or systematically fails on an irregular domain).","supporting_citations":[],"review_version":1}