{"id":"aa0f48e7-3fee-4b83-a24b-63264198e80e","arxiv_id":"1908.05473","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For the anisotropic stable JCIR process, the heat kernel exists and obeys a weighted anisotropic Besov bound, the strong Feller property holds, and in the subcritical case convergence to the invariant measure is exponential in total variation.","lead":"This paper proves that a multidimensional jump-driven interest-rate process, the anisotropic stable JCIR process, has a well-behaved probability density, satisfies the strong Feller property, and converges exponentially to its long-run distribution. It appears to be the first exponential ergodicity result in total variation for a multidimensional affine process without any diffusion component.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the proof chain for Theorems 2.4 and 2.5 closes under condition (A); the only external dependency is the cited pathwise comparison [21, Prop. 4.2] used in Proposition 4.2.","rationale":"The reader identified condition (A) as the weakest assumption, and I agree that it is the load-bearing input for boundary non-attainment. My pass sharpens this to one specific step: the pathwise dominance X_k ≥ Y_k is delegated to [21, Prop. 4.2], an authors' preprint whose proof is not included. Everything downstream of that comparison is internally consistent: the one-dimensional density theorem is proved directly from the characteristic function decay in Proposition 3.1; the multidimensional density and Besov regularity are obtained via a short-time approximation and a discrete integration by parts; Proposition 4.2 converts this into the absence of boundary atoms; and the convolution trick cleanly removes the moment assumption. The ergodicity proof likewise closes once the strong Feller property and the cited Wasserstein contraction from [22] are granted. I therefore have no substantive mathematical objection. I did not change the verdict because the identified dependency is a standard type of citation to a preprint, and the paper's own arguments do not exhibit a circular step or an internal contradiction. The proposed concrete test would still be worth running as a guard against a possible flaw in the external comparison lemma.","tokens_in":28505,"tokens_out":45081,"duration_ms":434435,"concrete_test":"Independently verify the comparison lemma used in Proposition 4.2: for the system (1.3), with the same driving stable process Z and subordinator J, prove that X_k(t) ≥ Y_k(t) for all t ≥ 0 almost surely when X and Y start from the same x_k and Y omits only the nonnegative cross-drift terms. If the comparison is false, provide a counterexample; if it is true, give a self-contained proof or a precise citation to a peer-reviewed version. This single check determines whether condition (A) alone suffices to eliminate the boundary-singular part in Theorem 2.4.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the full argument in good faith and did not find a load-bearing flaw. The central claim depends on condition (A) through Proposition 4.2, which removes the boundary-singular part of the density in Theorem 4.1 and justifies inverting the volatility in Proposition C.3. The route is: condition (A) implies the one-dimensional condition (2.1) for each projected subordinator, Theorem 2.1 gives a continuous density for the projected processes, and the pathwise comparison X_k(t) ≥ Y_k(t) from [21, Prop. 4.2] extends this to the multivariate process. That comparison is not proved in this paper and is attributed to an authors' preprint; if it failed, the conclusion P_sing^t(x, ∂R^m_+) = 0 in Appendix C.3 would fail, and so would the density and strong Feller claim. However, the comparison is plausible and the rest of the proof closes: the anisotropic Besov estimate is obtained from a short-time approximation, Lemma 4.3 converts it to L1 continuity through Kolmogorov-Riesz compactness, the convolution argument in Section 4.3 removes the extra moment condition, and the Harris-type argument in Section 5.2 turns the local Dobrushin condition plus Lyapunov drift into exponential ergodicity. The residual risks—reliance on several author preprints and the literature-based novelty claim—are disclosed and do not amount to an internal inconsistency.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the anisotropic stable JCIR process, an m-dimensional affine process on R^m_+ whose coordinates are driven by independent spectrally positive alpha_k-stable noises, a linear drift with nonnegative off-diagonal coefficients, and a subordinator J with jumps in R^m_+. The main results are: (i) Theorem 2.1, existence of a jointly continuous and smooth heat kernel for the one-dimensional stable JCIR process under condition (2.1); (ii) Theorem 2.4, existence of a density p_t(x,dy) whose dependence on the initial point x is L^1-continuous under condition (A), hence the strong Feller property; and (iii) Theorem 2.5, exponential ergodicity in total variation in the subcritical case under condition (A) and the log-moment condition (2.5). The proof combines the affine transform formula with a short-time approximation, anisotropic Besov regularity, boundary non-attainment estimates, a convolution trick to remove moment conditions on the big jumps, and a Harris-type theorem verified through a local Dobrushin condition and a Lyapunov function.","tokens_in":28609,"tokens_out":31622,"duration_ms":290678,"significance":"If the proofs are correct, this is the first exponential ergodicity result in total variation for multidimensional affine processes that does not rely on smoothing by a diffusion component. The method is not specific to the affine structure; the Besov-regularity and convolution arguments are applicable to other Markov processes with anisotropic jumps. The assumptions are stated directly in terms of the model primitives, with no free parameters, and the technical estimates are largely proved in full in the paper and its appendices. The one-dimensional smoothness theorem (Theorem 2.1) and the explicit exponential rate in Theorem 2.5 are concrete, falsifiable statements.","major_comments":[{"comment":"The proof of Proposition 4.2 uses the pathwise comparison result [21, Proposition 4.2] to conclude X^x_k(t) >= Y^{x_k}_k(t) for all k. This comparison is load-bearing: it is the mechanism by which condition (A) is transferred from the projected one-dimensional processes to the multidimensional process, and it is also used in Proposition C.3 to justify the invertibility of sigma(X(t-epsilon)). However, [21] is an authors' preprint and the comparison is neither stated nor proved in the present manuscript. Please provide the precise statement and either a proof in an appendix or a reference to a published version; without it the proof chain for Theorems 2.4 and 2.5 is not self-contained.","section":"Section 4.2, Proposition 4.2"}],"minor_comments":[{"comment":"The index set in 'k in {1,...,d}' should be {1,...,m}.","section":"Section 4.2, proof of Proposition 4.2"},{"comment":"The condition 'nu_k in (alpha_k-1, 1)' should read 'vartheta_k in (alpha_k-1, 1)'.","section":"Section 2.4, Example 2.7"},{"comment":"The Besov space B^{lambda,a}_{1,infty}(R^m_+) is used although the norm in (4.2) was defined for functions on R^m; please state explicitly that functions are extended by zero outside R^m_+.","section":"Section 4.1, Theorem 4.1"},{"comment":"The constants in Proposition 3.1 depend on t_0; for the claimed joint continuity in t, a sentence explaining how to make the estimate uniform on compact time intervals would be helpful.","section":"Section 3, proof of Theorem 2.1"},{"comment":"The step combining Proposition A.2 with weak convergence to conclude that integral |y| pi(dy) is finite is stated in one sentence; please spell out the truncation argument, for example using integral (|y| wedge R) pi(dy) <= liminf_t E|X^z(t)| <= C(1+|z|) and then R -> infinity.","section":"Section 5.1, Corollary 5.2"}],"recommendation":"major_revision","confidential_remarks":"The only substantive concern is the reliance on the authors' own preprints [18]-[22], especially [21, Proposition 4.2] for the pathwise comparison. If these are already published or accepted, the paper can likely be accepted after the requested clarifications; otherwise, the editors should ask the authors to include a proof of the comparison or to state its precise hypotheses. The novelty claim (first total-variation ergodicity for multidimensional affine processes without diffusion smoothing) appears fair relative to the cited literature, but it is worth checking against the recent references [2,27,38,45] during editorial handling."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a strong paper. It proves the multidimensional stable JCIR process has a density with L1-continuous dependence on the starting point, hence is strong Feller, and in the subcritical case converges exponentially fast in total variation — the first such result for a multidimensional affine process without any diffusion component. The price is an explicit boundary-non-attainment condition (A), and the main external dependencies are the authors' own preprints, not all of which are proved here.\n\nWhat is actually new: the anisotropic Besov regularity estimate with the boundary weight ρδ, the one-dimensional smoothness theorem via characteristic-function decay, and the Harris-type argument that turns a local Dobrushin condition into TV ergodicity. The proofs are written out in substantial detail — the Fourier argument in Section 3, the four-step compactness argument for Theorem 2.4, the convolution trick that removes the moment assumption, and the Lyapunov verification in Appendix B. The paper is honest about what it borrows: Section 4.1 credits [19,20] and the Debussche–Fournier framework, and the appendix states the anisotropic version of the density criterion is due to [20]. That is reproducible, parameter-free work, and the self-citations are appropriate.\n\nThe soft spots are in proportion. Condition (A) is genuinely load-bearing: it drives Proposition 4.2, which excludes a boundary atom and justifies inverting the volatility in the short-time approximation. The authors give examples where b_k = 0 but enough small jumps in direction k make (A) hold, so the condition is not vacuous. The larger concern is that Proposition 4.2 uses the pathwise comparison X_k(t) ≥ Y_k(t) from [21, Prop. 4.2], which is not proved in this paper. That comparison is plausible and likely true, but a referee should check it, because if it fails the density and strong Feller theorem lose their boundary control. The other preprints [18–20,22] are used for moment bounds and the Wasserstein ergodicity, and those are less delicate. Minor: the 'first result' claim rests on the authors' literature survey; I did not find a missing reference, but it is a claim about absence.\n\nFor whom: anyone working on affine processes, CBI processes, or ergodicity without diffusion smoothing. It deserves a serious referee. I would send it to review and ask the referee to verify the comparison principle and the moment assumption in Theorem 4.1. If those hold — and I expect they do — the paper is a solid advance.","headline":"Strong paper: multidimensional stable JCIR process gets a genuine density, strong Feller property, and exponential TV ergodicity without any diffusion component, under an explicit boundary condition that is honestly stated.","tokens_in":29320,"tokens_out":2552,"would_cite":true,"duration_ms":25070,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60J25","60J35","37A25"],"pacs":[],"model":"deepseek-v4-flash","headline":"The anisotropic stable JCIR process admits a density, is strong Feller, and mixes exponentially in total variation.","keywords":["stable JCIR process","affine process","heat kernel","anisotropic Besov space","strong Feller property","exponential ergodicity","total variation distance","spectrally positive stable process"],"falsifier":"Take $m=1$, $b=0$, and let the subordinator have Lévy measure $\\nu(dz)=z^{-1-\\vartheta}\\mathbf{1}_{z>1}\\,dz$ with $\\vartheta\\in(\\alpha-1,1)$. Then condition (2.1) fails because $\\int_0^\\infty(1-e^{-\\xi z})\\nu(dz)$ stays bounded as $\\xi\\to\\infty$ instead of growing like $\\xi^\\vartheta$. One can test the claimed conclusions directly: check whether $P[X^x(t)=0]>0$ for some $x,t$, or whether $x\\mapsto p_t(x,\\cdot)$ fails to be continuous in $L^1$ at $x=0$.","tokens_in":28102,"feed_emoji":"📈","tokens_out":7583,"duration_ms":72008,"temperature":0.7,"pith_summary":"This paper studies the anisotropic stable JCIR process, a multidimensional affine process whose coordinates are driven by independent spectrally positive stable noises and a common subordinator. It proves that, under a boundary-nonattainment condition, the transition semigroup has a genuine density, that the density depends continuously on the starting point in $L^1$, and hence that the process is strong Feller. In the subcritical case it further proves exponential convergence to the unique invariant measure in total variation. These are, the authors argue, the first total-variation ergodicity results for multidimensional affine processes that do not use smoothing from a diffusion component. A reader building models in finance or branching theory should care because density existence and exponential mixing are the statements that justify numerical pricing, filtering, and long-run simulation.","feed_headline":"Density and exponential mixing proven for stable JCIR process","feed_subtitle":"Multi-dimensional affine process is strong Feller and converges to equilibrium in total variation.","key_machinery":"The load-bearing object is the weighted anisotropic Besov space $B^{\\lambda,a}_{1,\\infty}(R^m)$ with anisotropy $a_i=\\alpha/\\alpha_i$ and weight $\\rho_\\delta(y)=\\min\\{\\delta,y_1^{1/\\alpha_1},\\dots,y_m^{1/\\alpha_m}\\}$. The proof builds a short-time approximation $X^\\varepsilon(t)$ of the process and applies a discrete integration by parts to test functions, obtaining the Besov estimate (4.3). Condition (A) enters through Proposition 4.2, giving $P[\\min_i X^x_i(t)\\le\\varepsilon]\\le C\\varepsilon$, so the process does not hit the boundary and the weight can be removed; a convolution argument removes the extra moment assumption on the large jumps. This machinery supplies both the density and its continuity in the starting point, and the local Dobrushin condition needed for Harris-type ergodicity.","core_discovery":"The central claim is Theorem 2.4: if condition (A) holds, then $P_t(x,dy)=p_t(x,y)dy$ and the map $R^m_+\\ni x\\mapsto p_t(x,\\cdot)\\in L^1(R^m_+)$ is continuous for every $t>0$, so the anisotropic stable JCIR process is strong Feller. In the subcritical case, under the log-moment condition (2.5), Theorem 2.5 gives the explicit bound $\\|P_t(x,\\cdot)-\\pi\\|_{TV}\\le C(1+\\log(1+|x|)+\\int\\log(1+|y|)\\pi(dy))e^{-\\delta t}$. In dimension one, Theorem 2.1 upgrades the density to a smooth, jointly continuous heat kernel. The authors obtain these as consequences of an a-priori bound on the heat kernel in a weighted anisotropic Besov norm, together with a boundary estimate showing that the process almost surely stays in the interior.","pith_inferences":["The multidimensional proof does not use the affine structure, so the same short-time approximation plus anisotropic Besov strategy should transfer to non-affine Markov processes with anisotropic jumps; testing it on an explicit non-affine example would be a direct extension.","Condition (A)'s exponent $\\vartheta_k$ is likely sharp: setting $\\vartheta_k$ at the lower endpoint $\\alpha_k-1$ should make the boundary-nonattainment rate fail or degrade, producing a threshold phenomenon that could be checked numerically.","The paper leaves aside coordinates with $\\sigma_i=0$; its own remark suggests a coordinatewise nondegeneracy condition on the subordinator should recover the results, so filling in that case is a concrete follow-up.","In interest-rate models, exponential convergence in total variation means the pricing kernel itself converges to equilibrium, not just moments, so the stated rate $\\delta$ could be used to give a quantitative burn-in time for simulations."],"forward_implications":["The transition kernel has a density, so probabilities and option prices for the model can be represented by integrals against $p_t(x,y)dy$ instead of abstract measures.","The strong Feller property follows, meaning the semigroup maps bounded measurable functions into continuous functions at positive times.","In the subcritical case the law converges to the invariant measure in total variation at rate $e^{-\\delta t}$, with the prefactor growing only logarithmically in the starting point.","The one-dimensional stable JCIR heat kernel is smooth in space with all derivatives bounded, so Fourier-based pricing and spectral methods are justified there."],"supporting_citations":[{"why":"supplies the unique strong solution and the affine representation used throughout the paper.","marker":"[3]"},{"why":"provides the affine process theory, Riccati equations, and Feller property underlying the heat kernel argument.","marker":"[13]"},{"why":"gives the Fourier-estimate method for affine jump-diffusions that the one-dimensional smoothness proof adapts.","marker":"[16]"},{"why":"is the source of the discrete integration-by-parts lemma (Lemma C.1) producing densities from Besov bounds.","marker":"[11]"},{"why":"supply the anisotropic short-time approximation and density method used to prove Theorem 4.1.","marker":"[19, 20]"},{"why":"provides the comparison result used to reduce boundary non-attainment to the one-dimensional estimate.","marker":"[21]"},{"why":"establishes existence and representation of the invariant measure in the subcritical case.","marker":"[28]"},{"why":"supplies the Wasserstein ergodicity and moment bounds used for the Lyapunov and smoothing estimates.","marker":"[22]"},{"why":"furnish the Harris-type theorem with local Dobrushin condition used to turn regularity into exponential TV convergence.","marker":"[24, 35]"},{"why":"gives the weak irreducibility input for verifying the local Dobrushin condition.","marker":"[40]"}],"fun_headline_variants":["Anisotropic stable JCIR: strong Feller and mixing","Exponential ergodicity for stable JCIR in any dimension","Smooth heat kernel in 1D for stable JCIR","Anisotropic stable JCIR: density and exponential mixing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Condition (A), which says that in every coordinate the inward drift plus the small jumps of the subordinator grow at least like $\\xi^{\\vartheta_k}$, so the process never hits the boundary; if it fails, the density and strong Feller conclusions need not hold.","fun_headline_variants_meta":{"raw":{"variants":["Anisotropic stable JCIR: strong Feller and mixing","Exponential ergodicity for stable JCIR in any dimension","Smooth heat kernel in 1D for stable JCIR","Anisotropic stable JCIR: density and exponential mixing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000513,"raw_usage":{"total_tokens":2432,"prompt_tokens":820,"completion_tokens":1612,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":436,"completion_tokens_details":{"reasoning_tokens":1542}},"tokens_in":436,"tokens_out":1612,"duration_ms":11306,"temperature":1.0,"reasoning_tokens":1542,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:14:49.161904+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $m=1$, $b=0$, and let the subordinator have Lévy measure $\\nu(dz)=z^{-1-\\vartheta}\\mathbf{1}_{z>1}\\,dz$ with $\\vartheta\\in(\\alpha-1,1)$. Then condition (2.1) fails because $\\int_0^\\infty(1-e^{-\\xi z})\\nu(dz)$ stays bounded as $\\xi\\to\\infty$ instead of growing like $\\xi^\\vartheta$. One can test the claimed conclusions directly: check whether $P[X^x(t)=0]>0$ for some $x,t$, or whether $x\\mapsto p_t(x,\\cdot)$ fails to be continuous in $L^1$ at $x=0$.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the unique strong solution and the affine representation used throughout the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the affine process theory, Riccati equations, and Feller property underlying the heat kernel argument."},{"cited_title":"Econometrics 176 (2013), no","cited_arxiv_id":null,"evidence_quote":"gives the Fourier-estimate method for affine jump-diffusions that the one-dimensional smoothness proof adapts."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"is the source of the discrete integration-by-parts lemma (Lemma C.1) producing densities from Besov bounds."},{"cited_title":"Boundary behavior of multi-type continuous-state branching processes with immigration","cited_arxiv_id":"1902.01162","evidence_quote":"provides the comparison result used to reduce boundary non-attainment to the one-dimensional estimate."},{"cited_title":"Existence of limiting distribution for affine processes","cited_arxiv_id":"1812.05402","evidence_quote":"establishes existence and representation of the invariant measure in the subcritical case."},{"cited_title":"Stochastic equation and exponential ergodicity in Wasserstein distances for affine processes","cited_arxiv_id":"1901.05815","evidence_quote":"supplies the Wasserstein ergodicity and moment bounds used for the Lyapunov and smoothing estimates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the weak irreducibility input for verifying the local Dobrushin condition."}],"review_version":1}