{"id":"3a371c95-fef7-4442-ac0c-4c05b7add6e0","arxiv_id":"1908.02888","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For a generalized CIR diffusion with degenerate noise X_t^h (1/2<h<1), the paper derives Harnack, log-Harnack, gradient bounds, and a super-Poincaré inequality with an optimal rate function.","lead":"This paper proves Harnack and super-Poincaré inequalities for a generalized Cox-Ingersoll-Ross process whose noise scales as X_t^h with h between 1/2 and 1. The result gives explicit quantitative control on how quickly this process forgets its starting point, useful for interest-rate and volatility models.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Super-Poincaré proof's reduction to half-line tails uses a boundary-additivity step that fails for sets with a neighborhood of x0","rationale":"The reader's weakest assumption already identified the isoperimetric reduction in Theorem 2.2 as the load-bearing step. My stress-test sharpens this into a concrete technical failure: the reduction from arbitrary open sets to two intervals requires boundary measure to behave additively over the pieces A1 and A2, but boundary measure is not additive when both pieces accumulate at the common point x0. A set containing a neighborhood of x0 has no boundary contribution at x0, whereas the proof's separate estimates for A1 and A2 both attribute boundary weight there. Consequently, the chain of inequalities leading to the claim that k(r) is minimized by a half-line tail is not justified. This is not a demonstrated counterexample to the theorem: the half-line tail may still be the true minimizer, and the numerical check I propose can confirm that the final bound survives. But as written, the proof of Theorem 2.2(2) has a gap in its central reduction, so the verdict should be conditional on supplying a correct isoperimetric argument rather than unconditional acceptance. The Harnack inequality proof does not exhibit a comparably serious issue; the ε→0 passages rely on standard dominated convergence that can be filled in.","tokens_in":12950,"tokens_out":37782,"duration_ms":366530,"concrete_test":"Take h = 3/4, α = δ = 1, let x0 be the maximizer of x^h η(x), and for small ε set A_ε = (x0−ε, x0+ε). Compute μ∂(∂A_ε)/μ(A_ε) directly from the definition and compare it with the proof's lower bound μ∂(∂(0,\\bar x1))/r + μ∂(∂(\\bar x2,∞))/r, where μ((0,\\bar x1)) = μ(A_ε ∩ (0,x0)) and μ((\\bar x2,∞)) = μ(A_ε ∩ (x0,∞)). If the two-tail sum exceeds μ∂(∂A_ε), the additivity step fails; then check whether the exact ratio still satisfies the claimed k(r) ≥ c√(−log r) to decide whether Theorem 2.2(2) needs only a proof repair or a substantive modification.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In Theorem 2.2(1), after decomposing an open set A into A1 = A ∩ (0,x0) and A2 = A ∩ (x0,∞), the proof bounds μ∂(∂A1) and μ∂(∂A2) separately and then asserts μ∂(∂A)/μ(A) ≥ μ∂(∂((0,\\bar x1) ∪ (\\bar x2,∞)))/μ(...). This implicitly assumes μ∂ is additive over A1 and A2. It is not: if A contains an open interval around x0, then x0 is an interior point of A and contributes nothing to ∂A, while the same point contributes to the boundaries of both A1 and A2 as they are taken in the half-lines (0,x0) and (x0,∞). Thus the sum μ∂(∂A1)+μ∂(∂A2) overcounts the boundary at x0, and the inequality μ∂(∂A) ≥ μ∂(∂A1)+μ∂(∂A2) is false; a concrete witness is A = (x0−ε, x0+ε). Since this reduction is the only step that eliminates arbitrary open sets and leads to k(r) ≥ c√(−log r), the proof of Theorem 2.2(2), and hence β(r) = e^{C(1+r^{-1})}, is not complete as written. The Harnack part, by contrast, appears sound modulo routine dominated-convergence details around (4.4)–(4.5).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies the generalized Cox-Ingersoll-Ross diffusion dX_t = (α − δX_t)dt + X_t^h dB_t with 1/2 < h < 1 and α ≥ h/2. The main results are a dimension-free Harnack inequality and a log-Harnack inequality with explicit constants (Theorem 2.1), an intrinsic gradient estimate, and a super-Poincaré inequality for the associated Dirichlet form with rate function β(r) = e^{C(1+r^{-1})} (Theorem 2.2), together with an optimality statement showing that β(r) = e^{C(1+r^{-λ})} is impossible for λ < 1. The proofs combine coupling by change of measure with an isoperimetric analysis of the invariant measure.","tokens_in":13198,"tokens_out":26975,"duration_ms":267677,"significance":"If correct, the paper extends earlier functional-inequality results for the classical CIR model (h = 1/2) to the degenerate case h > 1/2, where the diffusion coefficient vanishes faster at zero. The explicit parameter-free Harnack constant and the identification of the correct exponential rate for the super-Poincaré inequality are useful and go beyond the known results. The optimality statement gives a sharp, falsifiable bound on the admissible rate functions. The main weakness is that the isoperimetric reduction, which is load-bearing for Theorem 2.2, is not justified as written, and the Harnack proof contains a small but real technical slip in a limiting step.","major_comments":[{"comment":"After defining A1 = A ∩ (0,x0) and A2 = A ∩ (x0,∞), the manuscript bounds μ∂(∂A1) and μ∂(∂A2) separately and then asserts μ∂(∂A)/μ(A) ≥ μ∂(∂((0,¯x1) ∪ (¯x2,∞)))/μ((0,¯x1) ∪ (¯x2,∞)). This step requires additivity of μ∂ over the partition, but μ∂ is not additive when A contains an open neighborhood of x0. In that case x0 is an interior point of A and contributes nothing to ∂A, while it contributes x0^h η(x0) to the boundary measure of each of A1 and A2 separately. A concrete witness is A = (x0−ε, x0+ε) with μ(A) = r: the two separate bounds effectively count 2x0^h η(x0), whereas the actual boundary measure is (x0−ε)^h η(x0−ε) + (x0+ε)^h η(x0+ε). Consequently the reduction of the isoperimetric minimization to half-line tails is not established by the written argument, and the lower bound k(r) ≥ c√(−log r) in Theorem 2.2(2) does not follow from the given proof. The argument needs an additional case distinction: when A contains x0, the component containing x0 should be compared directly with the tail competitor, for example by noting that its perimeter is bounded below by a positive constant as r → 0 while its mass tends to 0.","section":"Proof of Theorem 2.2(1), reduction to half-line tails"},{"comment":"The displayed bound I1 ≤ E∫_0^T e^{2(1−h)(δ−h/2)t} (I{Y_t≠0} − I{X_t≠0})^2 dt = 0 is false for fixed ε > 0: when 0 < X_t < Y_t, the left integrand is strictly positive while the right-hand integrand is zero. The desired conclusion I1 → 0 is nevertheless correct, because the integrand is bounded by 1 and converges pointwise to (I{Y_t≠0} − I{X_t≠0})^2, which is zero a.e. by Lemma 3.1. The proof should replace the inequality with an equality of limits after an explicit dominated-convergence step.","section":"Proof of Theorem 2.1(1), estimate of I1"},{"comment":"The statement of Theorem 2.1 allows δ = h/2, but the displayed Harnack constant contains the factor (δ − h/2) / (e^{2(1−h)(δ−h/2)T} − 1), which is undefined when δ = h/2. The coupling function ξ(t) in (4.1) is also undefined in this case. The theorem should either exclude δ = h/2 from its statement or provide a separate limiting argument for this endpoint.","section":"Theorem 2.1, parameter range"}],"minor_comments":[{"comment":"The abstract contains the typo 'isopermetric'; it should read 'isoperimetric'. Several other typos occur, e.g., 'Mover' for 'Moreover' in Section 3.","section":"Abstract and Introduction"},{"comment":"In the displayed bound for |φ_n''(x) x^{2h}|, the equality should be an upper bound: the expression is bounded by (2h+1)/n^{2h−1}, not equal to it for all x < 1/n.","section":"Lemma 3.1"},{"comment":"Reference [15] is spelled 'Wand, F.-Y.' in the bibliography; it should be 'Wang, F.-Y.' Also, reference [18] appears in the list but is not cited in the text.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"To the editor: the Harnack part appears essentially sound apart from the DCT slip in (4.4), which is easy to repair. The super-Poincaré part has a genuine gap in the isoperimetric reduction; I expect it is patchable by a case distinction for sets containing x0. If the authors supply that argument and fix the δ = h/2 endpoint, the paper is likely acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Zhao and Huang extend the Zhang-Zheng CIR result from h=1/2 to the degenerate range h in (1/2,1). The Harnack and log-Harnack inequalities in Theorem 2.1 look right; the coupling by change of measure with the intrinsic metric is standard, and Lemma 3.2 is the genuinely new device that makes the residual drift term have the right sign. The gradient estimate is a routine corollary. That part is worth publishing.\n\nThe super-Poincaré part is where I part ways with the reader's ACCEPT. The proof of Theorem 2.2(1) reduces the isoperimetric minimization over arbitrary open sets to a pair of half-line tails. The step 'This yields μ∂(∂A)/μ(A) >= ...' assumes boundary content is additive across the split at x0. It is not: if A contains an interval around x0, then x0 is an interior point of A and contributes nothing to μ∂(∂A), while it contributes once to each of the relative sets A1 and A2. The sum overcounts by 2 x0^h η(x0). So the inequality goes the wrong way. The concrete example A=(x0-ε, x0+ε) kills the argument. Since this reduction is the only place that excludes other set geometries, the lower bound k(r) >= c√(-log r) and hence β(r)=e^{C(1+r^{-1})} are not established by the written proof. The optimality section (part 3) uses standard Lyapunov-type criteria and looks fine on its own, but it bounds from below; the positive part is missing.\n\nThere is also a minor paper-craft issue: [15] is cited as 'Wand' in the references, and the paper leans heavily on Wang's monograph for the bridge from isoperimetric constants to super-Poincaré inequalities. That is acceptable if the quoted theorems are exactly applicable, but the reduction gap above is not in the book.\n\nWho is this for? People working on functional inequalities for degenerate diffusions in finance. The Harnack result is solid and deserves referee time. The super-Poincaré claim needs a fix, maybe only a short one: one can try to show the infimum over sets not containing x0, or handle straddling sets by a different bound. As it stands, I would not accept. Send to a serious referee, with the expectation of major revision on Section 5.","headline":"Harnack part is a genuine, apparently correct extension; the super-Poincaré proof has a real gap in the isoperimetric reduction.","tokens_in":13806,"tokens_out":4441,"would_cite":true,"duration_ms":42420,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60H15"],"pacs":[],"model":"deepseek-v4-flash","headline":"For the generalized Cox-Ingersoll-Ross diffusion $dX_t=(\\alpha-\\delta X_t)dt+X_t^h dB_t$ with $1/2<h<1$ and $\\alpha\\ge h/2$, this paper proves Harnack and log-Harnack inequalities, an intrinsic-gradient contraction estimate, and a…","keywords":["Coupling by change of measure","Harnack inequality","Log-Harnack inequality","Isoperimetric constant","Super-Poincaré inequality","Generalized Cox-Ingersoll-Ross model","Intrinsic metric","Degenerate diffusion"],"falsifier":"Evaluate the left-hand side of (3.6) for a grid of values $h\\in(1/2,1)$, $x<y$, and $\\alpha\\ge h/2$; any positive value would invalidate the Harnack coupling argument. Separately, compute $k(r)$ numerically over two-interval sets $D=(0,x_1)\\cup(x_2,\\infty)$ with $\\mu(D)=r$ and check whether $k(r)/\\sqrt{-\\log r}$ tends to zero as $r\\to 0$; if it does, the super-Poincaré rate in Theorem 2.2(2) is false.","tokens_in":12690,"feed_emoji":"📈","tokens_out":10126,"duration_ms":98310,"temperature":0.7,"pith_summary":"This paper studies the generalized Cox-Ingersoll-Ross diffusion $dX_t=(\\alpha-\\delta X_t)dt+X_t^h dB_t$ for $1/2<h<1$, where the noise near the origin is even weaker than in the classical CIR model with $h=1/2$. It establishes dimension-free Harnack and log-Harnack inequalities for the associated semigroup, with explicit exponential factors written in the intrinsic distance $\\rho(x,y)=\\int_x^y r^{-h}dr$, and a matching contraction estimate for the intrinsic gradient. It also proves a super-Poincaré inequality for the Dirichlet form with rate $\\beta(r)=e^{C(1+r^{-1})}$, and shows this rate is optimal in the sense that no rate $\\beta(r)=e^{C(1+r^{-\\lambda})}$ with $\\lambda<1$ can work. These inequalities quantify how quickly the degenerate process forgets its starting point, and they extend earlier results that were restricted to $h=1/2$.","feed_headline":"Degenerate CIR model gets sharp Harnack and super-Poincaré bounds","feed_subtitle":"Coupling by change of measure and isoperimetric constants fix the optimal rate $e^{C(1+1/r)}$ for the degenerate diffusion.","key_machinery":"The argument rests on three connected tools. First, the intrinsic metric $\\rho(s,t)=\\int_s^t r^{-h}dr$ and the intrinsic gradient $\\nabla_h f=x^h f'$ recast the degenerate diffusion so that distances between paths have a natural scale. Second, coupling by change of measure constructs two solutions $X_t$ and $Y_t$ starting from $x<y$, adding a carefully chosen drift $\\xi(t)Y_t^h$ to $Y_t$ until the coupling time $\\tau$; the key comparison inequality (3.6), controlling the residual drift term $M(X_t,Y_t,\\varepsilon)$ so that it is non-positive, forces $\\tau\\le T$, after which the change-of-measure theorem identifies the law of $Y_T$ with that of the original process started at $y$. Third, for the super-Poincaré inequality, the isoperimetric constant $k(r)=\\inf_{\\mu(D)\\le r}\\mu^\\partial(\\partial D)/\\mu(D)$ is shown, using boundary-measure comparison and the sign analysis of the function $x^h\\eta(x)$, to be minimized for small $r$ by half-line tails $(x,\\infty)$; this yields $k(r)\\ge c\\sqrt{-\\log r}$ and hence $\\beta(r)=e^{C(1+r^{-1})}$.","core_discovery":"The central claim is that the semigroup $(P_t)$ of the generalized CIR equation enjoys the Harnack inequality $$(P_T f)^p(y)\\le P_T f^p(x)\\exp\\left[\\frac{p(\\delta-h/2)($y^{{1-h}}$-$x^{{1-h}}$)^2}{(p-1)(1-h)($e^{{2(1-h)(\\delta-h/2)T}}$-1)}\\right]$$ for every $T>0$, $p>1$, $x,y\\ge 0$, together with the log-Harnack analogue and the intrinsic-gradient bound $|\\nabla_h P_T f|(x)\\le e^{-(1-h)(\\delta-h/2)T}P_T|\\nabla_h f|(x)$. For the Dirichlet form $E(f,f)=\\frac12\\int_0^\\infty x^{2h}(f')^2 d\\mu$, where $\\mu$ has density proportional to $x^{-2h}\\exp(\\frac{2\\alpha}{1-2h}x^{1-2h}-\\frac{\\delta}{1-h}x^{2-2h})$, the paper proves the super-Poincaré inequality $\\mu(f^2)\\le rE(f,f)+e^{C(1+r^{-1})}\\mu(|f|)^2$ and proves optimality of the exponent $1$ by excluding every rate of the form $e^{C(1+r^{-\\lambda})}$ with $\\lambda<1$.","pith_inferences":["The same coupling-by-change-of-measure construction likely extends to boundary cases such as $h=1$ or to multiplicative noise of the form $X_t^h g(X_t)$, with the intrinsic metric modified accordingly; the structure of the Harnack factor should remain similar.","A testable extension is to check numerically whether the optimal super-Poincaré rate $\\beta(r)=e^{C/r}$ reflects the volume growth of intrinsic metric balls near infinity; if so, compactly supported perturbations of the drift $(\\alpha-\\delta x)$ should preserve the exponent.","The condition $\\alpha\\ge h/2$ is probably sharp for the Harnack part, since the comparison inequality (3.6) fails when $\\alpha<h/2$; exploring that regime could reveal a slower coupling rate or no finite-time coupling at all."],"forward_implications":["The Harnack inequality makes the semigroup strong Feller, so $P_T$ maps bounded measurable functions to continuous functions on $[0,\\infty)$.","The log-Harnack inequality gives quantitative control over how far the laws from two different starting points can separate, with the intrinsic metric $\\rho(x,y)$ appearing directly in the exponent.","The intrinsic-gradient bound shows that the gradient contracts exponentially at rate $e^{-(1-h)(\\delta-h/2)T}$, a quantitative form of ergodicity for the degenerate process.","The super-Poincaré inequality with $\\beta(r)=e^{C(1+r^{-1})}$ implies the compactness property of the Dirichlet form, so the generator has discrete spectrum.","The optimality statement rules out any polynomial improvement in the rate: as $r\\to 0$, the rate function must grow at least like $e^{C/r}$."],"supporting_citations":[{"why":"Supplies existence and uniqueness of a non-negative strong solution for the generalized CIR equation.","marker":"[11]"},{"why":"Provides the stochastic calculus background used in deriving the coupling dynamics and the boundary measure computations.","marker":"[12]"},{"why":"Supplies the general theorems converting lower bounds on the isoperimetric constant into super-Poincaré inequalities, and the criterion used to rule out slower rates.","marker":"[15]"},{"why":"Introduces the coupling-by-change-of-measure technique that the Harnack proof adapts to the degenerate diffusion.","marker":"[16]"},{"why":"Establishes the $h=1/2$ case whose Harnack and super-Poincaré results the present theorems extend.","marker":"[21]"}],"fun_headline_variants":["Optimal super-Poincaré rate for degenerate CIR","Sharp Harnack and super-Poincaré for degenerate CIR","Degenerate CIR: optimal super-Poincaré via coupling","Isoperimetric constants yield optimal super-Poincaré for CIR","Harnack and super-Poincaré bounds for generalized CIR model"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's load-bearing premise is that the comparison inequality (3.6) makes the residual drift in the coupling non-positive, and that among small sets the half-line tail $(x,\\infty)$ has the smallest boundary measure, so the isoperimetric constant grows at least like $\\sqrt{-\\log r}$; if either fails, the stated Harnack factor or the optimal rate $\\beta(r)=e^{C(1+r^{-1})}$ need not follow.","fun_headline_variants_meta":{"raw":{"variants":["Optimal super-Poincaré rate for degenerate CIR","Sharp Harnack and super-Poincaré for degenerate CIR","Degenerate CIR: optimal super-Poincaré via coupling","Isoperimetric constants yield optimal super-Poincaré for CIR","Harnack and super-Poincaré bounds for generalized CIR model"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000781,"raw_usage":{"total_tokens":3441,"prompt_tokens":928,"completion_tokens":2513,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":544,"completion_tokens_details":{"reasoning_tokens":2414}},"tokens_in":544,"tokens_out":2513,"duration_ms":17909,"temperature":1.0,"reasoning_tokens":2414,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:33:23.594201+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the left-hand side of (3.6) for a grid of values $h\\in(1/2,1)$, $x<y$, and $\\alpha\\ge h/2$; any positive value would invalidate the Harnack coupling argument. Separately, compute $k(r)$ numerically over two-interval sets $D=(0,x_1)\\cup(x_2,\\infty)$ with $\\mu(D)=r$ and check whether $k(r)/\\sqrt{-\\log r}$ tends to zero as $r\\to 0$; if it does, the super-Poincaré rate in Theorem 2.2(2) is false.","supporting_citations":[{"cited_title":"Amsterdam: North Holland, 1989","cited_arxiv_id":null,"evidence_quote":"Supplies existence and uniqueness of a non-negative strong solution for the generalized CIR equation."},{"cited_title":"E., Brownian motion and stochastic calculus, 2nd edition, cor- rected 6th printing","cited_arxiv_id":null,"evidence_quote":"Provides the stochastic calculus background used in deriving the coupling dynamics and the boundary measure computations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the general theorems converting lower bounds on the isoperimetric constant into super-Poincaré inequalities, and the criterion used to rule out slower rates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the coupling-by-change-of-measure technique that the Harnack proof adapts to the degenerate diffusion."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the $h=1/2$ case whose Harnack and super-Poincaré results the present theorems extend."}],"review_version":1}