{"id":"972e5711-cf44-4af4-82f6-7a59cc523b7b","arxiv_id":"2501.12679","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper finds an explicit formula, involving an integral of a Painlevé I Hamiltonian, for the previously unknown multiplicative constant in the large gap asymptotics of higher-order Tracy-Widom distributions.","lead":"This paper derives the exact constant term in the large gap asymptotics of higher-order analogues of the Tracy-Widom distribution, closing an open problem from previous work. The result gives a complete asymptotic formula for a family of random matrix edge distributions and introduces a method likely to apply to other Fredholm determinant constants.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The transition-region error in Lemma 2.3 is controlled only after integration; the missing uniformity of its O(|x|^{(k-1/3)/(2k+1)}) remainder as κ0→0 is the load-bearing gap.","rationale":"The reader identified Lemma 2.3's large transition-region error and its delicate integrated cancellation as the weakest assumption. I agree. The paper itself flags this in Remark 2.5 and performs the cancellation in (2.16), so the concern is not an internal inconsistency but a missing uniformity proof: the pointwise error estimate (2.4) must remain uniform over the whole transition interval, and the route via (5.71)–(5.78) and Lemmas B.1/B.4 is only sketched at the level of λ and κ0. The rest of the proof of Theorem 1.1—the steepest-descent analysis in Sections 4–6, the differential identities (3.14) and (3.21), and the explicit algebraic cancellation (2.19)–(2.22)—is long but standard in structure and can in principle be checked term-by-term. The final theorem is credible, and the missing uniformity is fillable; hence the reviewer's CONDITIONAL verdict is appropriate and no verdict change is warranted.","tokens_in":49881,"tokens_out":25848,"duration_ms":212709,"concrete_test":"Re-derive the remainder estimate in Lemma 2.3 by repeating the derivation of Θ2(r) and Θ3(r) in Section 5.3 while keeping explicit dependence on κ0 (or equivalently b0) in the constants of Lemmas 5.2 and B.4. Verify that the O(|x|^{(k-1/3)/(2k+1)}) term in (2.4) holds uniformly as κ0 ranges over [0, C1λ^{-k-1+δ}] with a constant independent of κ0 and r in the transition window; equivalently, show that the product length(s1,s2) × sup_{s∈(s1,s2)} O(|x|^{(k-1/3)/(2k+1)}) is O(|s|^{-1/6}) with the sup taken with an explicitly controlled constant. If such a tracking is impossible and the constant diverges as κ0→0, then (2.16)–(2.17) must be replaced by a direct estimate of the integrated transition contribution before Theorem 1.1's constant can be accepted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The final constant in Theorem 1.1 is obtained by integrating ∂F/∂s over the contour (1.41). In the transition region the integrand is approximated in Lemma 2.3 by ∂χ/∂s·H_PII(χ) with an additive error O(|x|^{(k-1/3)/(2k+1)}) that is not small as |x|→∞. The proof of Theorem 1.1 handles this in (2.16) by bounding the integral of this error over (s1,s2) by length·error = O(|s|^{-k+1/6})·O(|s|^{k-1/3}) = O(|s|^{-1/6}). For this to work the O-term in Lemma 2.3 must be uniform in s across the entire transition interval, including the point where α_k s^{2k+1}+x = 0 (i.e. κ0 = b0 = 0 in the scaled variables). Section 5.3 derives the error from (5.71) and (5.78), but the constants in Lemmas 5.2 and B.4 are not tracked as functions of κ0, and the parametrix estimates (B.20), (B.21) involve Φ0(-x) with x = λ^{(4k+3)/3}|κ0| tending to 0 possible as κ0→0. If the implicit constant diverges as |κ0|^{-α}, the interval-length bound would be multiplied by λ^{α(k+1-δ)}, potentially exceeding O(|s|^{-1/6}) and destroying the extracted constant −χ(0) − log(2k+1)/24 − log α_k/[24(2k+1)]. No inconsistency with known results is identified; rather, a crucial uniformity hypothesis is asserted without a displayed proof.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper determines the previously unknown multiplicative constant C(k) in the large gap asymptotics of the higher-order Tracy-Widom distributions det(I - K_s^{(k)}) for k = 1, 2, 3, \\ldots. Theorem 1.1 gives an explicit expansion of F(s;x) = log det(I - K_s^{(k)}) as s \\to -\\infty, with a constant term involving the integral I_h(x) of the Hamiltonian h of the special real pole-free solution of the even Painlev\\'e I hierarchy P_I^{2k}, the classical Airy constant \\chi(0), and explicit logarithmic terms, with error O(|s|^{-\\epsilon_0}) uniformly in x over the window [-c_1|s|^{2k+1}, \\alpha_k|s|^{2k+1} - c_2|s|^{2k/3+\\epsilon}]. The proof uses Riemann-Hilbert steepest descent within the framework of Claeys, Its and Krasovsky, together with a new two-contour strategy in the (s,x)-plane that exploits uniform asymptotics of both \\partial F/\\partial s and \\partial F/\\partial x and compares the contours ending at x_0 = \\pm|s|^{2k+1}. The paper also proves (Corollary 1.3) that the total integral of h - h_{Asy} vanishes for all k, and (Corollary 1.4) that the expansion reduces to the classical Tracy-Widom asymptotics near the right edge of the x-window.","tokens_in":50226,"tokens_out":23978,"duration_ms":215250,"significance":"If the main theorem is correct, this is a substantial result: it resolves an open problem from Claeys, Its and Krasovsky (2010) by giving an explicit formula for the previously unknown constant for every k, with no fitted parameters, and the appearance of the Painlev\\'e I Hamiltonian integral is a new structural feature. The paper is self-aware about the delicate step (Remark 2.5 explicitly acknowledges that the transition-region error in Lemma 2.3 is large and is controlled only after integration), and the cancellation in (2.19) is verified by the explicit expansions (2.20)-(2.22). The result passes internal consistency tests: for k = 1 it yields the form (1.20) with a concrete constant; Corollary 1.3 generalizes the known k = 1 total-integral identity of [21]; and Corollary 1.4 recovers the classical Tracy-Widom expansion (1.15) with the correct constant \\chi(0). The method is plausibly transferable to Pearcey-type and other Fredholm determinants. The main reservation is that the constant extraction relies on uniformity of the transition-region asymptotics in the parameter \\kappa_0, which is asserted rather than fully demonstrated; this is the subject of the major comments.","major_comments":[{"comment":"Lemma 2.3's estimate (2.4) has an additive error O(|x|^{(k-1/3)/(2k+1)}) which is not small as |x| \\to \\infty; as Remark 2.5 acknowledges, the estimate is used only through its integral over (s1, s2), which in (2.16) yields O(|s|^{-1/6}) because length x error = O(|s|^{-k+1/6}) x O(|s|^{k-1/3}). For this product to be a genuine error bound, the O-term in (2.4) must be uniform in s across the whole transition interval, including the point where \\alpha_k s^{2k+1} + x = 0, i.e. where \\kappa_0 = b_0 = 0 in the scaled variables of Section 5. The derivation in Section 5.3 obtains (5.71) and (5.78) from (5.69), (5.70), (B.20) and (B.21), but the passage is compressed ('readily seen'): Lemma B.4 is stated only for x \\to +\\infty, whereas the argument of \\Phi_0 and \\tilde{\\Phi}_0 in (5.67), (5.68), (5.75), (5.76) is X = \\lambda^{(4k+3)/3}\\kappa_0 (respectively X = \\lambda^{(4k+3)/3} f_3(r; r_0)), which takes values down to 0 at the point \\alpha_k s^{2k+1} + x = 0 inside the transition region; no displayed argument supplies a uniform bound for |X| = O(1) or X \\to 0 with constants independent of \\kappa_0. The manuscript should complete this uniformity argument (for example by a three-range estimate using analyticity of \\Phi_0 in X together with (B.20)), or restate Lemma B.4 as a uniform-in-X statement. Without this, the extraction of -\\chi(0) - \\log(2k+1)/24 - \\log \\alpha_k/(24(2k+1)) in (2.16)-(2.18) is not fully justified. This is a load-bearing gap, not a stylistic one.","section":"§5.3, Eqs. (5.71), (5.78); used at Eq. (2.16)"},{"comment":"There appears to be an internal inconsistency in the size of the local parametrix radius \\rho_2 in the case where \\lambda^{(4k+3)/3}\\kappa_0 is bounded. Formula (5.19) sets \\rho_2 = 1/(\\lambda^{(4k+3)/2}|\\kappa_0|) + \\lambda^{2k/3+1/2}; for |\\kappa_0| \\le C\\lambda^{-(4k+3)/3} the first term is at least of order \\lambda^{(4k+3)/6} and the second term is exactly \\lambda^{(4k+3)/6}, so \\rho_2 \\to +\\infty. In the proof of Lemma 5.1, however, the same case is said to give \\rho_2 behaving like C(\\lambda)\\lambda^{-2k/3-1/2}, which tends to 0. These two statements are incompatible, and the estimates in (5.25)-(5.28) and in Lemma 5.2 depend on the actual size of \\rho_2 (through (5.23) and the factors (\\lambda\\rho_2)^{\\pm\\sigma_3/4}). Since this matching feeds into Lemma 5.2 and hence into the bound (5.70) used for Lemma 2.3, the authors should resolve the discrepancy and rerun the estimates with the correct \\rho_2.","section":"§5.1, Eq. (5.19) and Lemma 5.1"},{"comment":"Lemmas B.1 and B.2 are essential inputs for Lemma 2.3, yet their proofs are explicitly sketches, and the version of the P34 asymptotics actually needed in Section 5 is not literally the one stated: the applications require (B.1) and (B.12) to hold uniformly over a two-parameter regime in which the size of x = \\lambda^{(4k+3)/3}\\kappa_0 ranges from 0 to +\\infty and the ratio |\\zeta|/|x| ranges near the threshold L, with uniform constants; the manuscript neither proves nor cites such strengthened statements. Lemma B.4 similarly covers only x \\to +\\infty. The authors should provide full proofs, or precise reference results that cover the required uniformity, and should verify the hypotheses of Lemmas B.1 and B.2 on the actual contours of Section 5 (for example on \\partial U(r; \\rho_2) with \\rho_2 as in (5.19)).","section":"Appendix B, Lemmas B.1, B.2, B.4"},{"comment":"No additional major comments.","section":null}],"minor_comments":[{"comment":"The name is misspelled as 'Clayes' in the abstract (twice) and in the Introduction; it should be 'Claeys', Its and Krasovsky.","section":"Abstract and Section 1"},{"comment":"The notation 'large \\lambda behavior of \\Phi_0(-\\lambda)' is imprecise: the argument of \\Phi_0 in (5.67)-(5.68) is \\lambda^{(4k+3)/3}\\kappa_0, not \\lambda. Since the uniformity in \\kappa_0 is precisely the delicate point, the argument should be written out explicitly.","section":"§5.3, text before Eq. (5.71)"},{"comment":"The remark correctly flags the transition-region error as the delicate point; it would be helpful to point the reader to the specific estimates (5.71) and (5.78) and to state that uniformity in \\kappa_0 is claimed there, since that is what makes the integrated error small.","section":"Remark 2.5"}],"recommendation":"major_revision","confidential_remarks":"The reader's stress-test concern is legitimate and material: the uniformity of the transition-region asymptotics in \\kappa_0 near \\kappa_0 = 0 is asserted rather than proved, and the displayed derivation passes through Lemma B.4, which is stated only for x \\to +\\infty. I nevertheless judge the central claim likely correct and the gap fixable within the scope of a revision: what is needed is a completed Appendix B with explicit constant tracking, a resolution of the \\rho_2 inconsistency in Section 5.1, and a uniform-in-X statement for \\Phi_0 and \\tilde{\\Phi}_0. I see no citation-policy concerns; the self-citation [21] documents the k = 1 special case and is appropriate. The manuscript fits the journal's scope and, once the uniformity step is completed, would be a strong contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThis paper resolves the open constant problem in the large gap asymptotics of higher-order Tracy-Widom distributions. For every k, the previously unknown multiplicative constant C(k) is now explicit, expressed through an integral of the Hamiltonian of the special Painlevé I hierarchy solution. That is a genuine result, and the method is the real news. Instead of trying to control ∂F/∂s over an infinite interval, the authors integrate ∂F/∂x over x first, choosing x0 = ±|s|^{2k+1}. This gives a clean route through the transition region and extracts the Airy constant χ(0). The vanishing total integral of the Hamiltonian and the transition to the classical Tracy-Widom distribution are useful by-products. The paper is honest about what is new and what comes from Claeys–Its–Krasovsky.\n\nThe Riemann-Hilbert analysis in Sections 4–6 is detailed, and the cancellation proving (2.19) is explicit. The estimates behind Lemma 2.3 look plausible. The soft spots are exactly where the reader's report puts them. Appendix B gives sketches, not complete proofs, of the uniform P34 parametrix asymptotics. More seriously, the transition-region error in Lemma 2.3 is O(|x|^{(k-1/3)/(2k+1)}), which is not small pointwise; only its integrated contribution over (s1,s2) is controlled. The proof in Theorem 1.1 bounds length times error, and that only works if the O-term is uniform across the whole transition interval, including where κ0 = 0. The text does not display that uniformity argument. This is a genuinely delicate point. I do not think the result is wrong: the bounded-κ0 case should be covered by continuity of the model parametrix, and the shrinking disk ρ2 stays small. But the gap is real and should be closed in a revision. Lemma B.4's error terms also need x-dependence tracked.\n\nAll in all, this is credible, important work for the random matrix and Painlevé community. It deserves a serious referee, not a desk rejection. My own verdict would be: publish after the uniformity and parametrix estimates are filled in. I would cite it if I worked on these determinants.","headline":"Settles a real open constant problem with a clever contour-switching method, but the proof of two uniformity claims is not fully displayed.","tokens_in":50789,"tokens_out":5994,"would_cite":true,"duration_ms":57077,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","34M55","41A60","33E17"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves an explicit multiplicative constant in the large-gap asymptotics of higher-order Tracy-Widom distributions, expressed through the Hamiltonian of a special Painlevé I hierarchy solution.","keywords":["higher-order Tracy-Widom distribution","Painlevé I hierarchy","multiplicative constant","large gap asymptotics","Fredholm determinant","Riemann-Hilbert problem","Painlevé XXXIV parametrix","Airy kernel"],"falsifier":"Compute the left and right sides of the cancellation identity (2.19) numerically for large $|s|$ with $k=1$; if their difference is not $O(|s|^{-1/3})$, the extracted constant is wrong. Alternatively, evaluate $\\log\\det(I-K_s^{(1)})$ directly for $s=-40$ and fixed $x=0$, subtract the explicit $s$-dependent terms, and compare the remainder with the $k=1$ case of the constant formula in Corollary 1.2.","tokens_in":49646,"feed_emoji":"📐","tokens_out":4295,"duration_ms":47622,"temperature":0.7,"pith_summary":"This paper determines the previously unknown constant term in the large-gap asymptotics of higher-order Tracy-Widom distributions, which describe eigenvalue fluctuations at critical edge points of unitary random matrix models. The constant is expressed in closed form using an integral of the Hamiltonian associated with a special real pole-free solution of the even Painlevé I hierarchy, together with known constants from the classical Tracy-Widom distribution. A by-product of the proof is that the total integral of this Hamiltonian difference vanishes for every order, and the same uniform asymptotics reveal a transition from the higher-order distribution to the classical Tracy-Widom distribution in the appropriate scaling regime. The approach is presented as a template for evaluating similar multiplicative constants in other Fredholm determinants arising in mathematical physics.","feed_headline":"Gap constant solved for every higher-order Tracy-Widom law","feed_subtitle":"The long-open constant is now explicit, tied to a Painlevé I Hamiltonian, and recovers the classical Tracy-Widom limit.","key_machinery":"The central object is the $\\mathrm{P}_I^{2k}$ kernel, a higher-order analogue of the Airy kernel built from solutions of the even Painlevé I hierarchy, and the associated Fredholm determinant $F(s;x)=\\log\\det(I-K_s^{(k)})$. The argument uses the two differential identities $\\partial F/\\partial x=-(Y_{-1})_{12}$ and $\\partial F/\\partial s=\\lim_{\\zeta\\to s}(2\\pi i)^{-1}(X(\\zeta)^{-1}X'(\\zeta))_{21}$, which connect the derivatives of the determinant to Riemann-Hilbert problems, and then applies nonlinear steepest descent analysis in three asymptotic regions: algebraic growth, transition, and exponential decay. The transition region uses a Painlevé XXXIV parametrix whose asymptotics, when integrated over the short transition interval, produce the classical Tracy-Widom constant $\\chi(0)$, while the Hamiltonian $h$, defined through $dh/dx=q$ for the special pole-free solution $q$ of $\\mathrm{P}_I^{2k}$, enters through the $x$-derivative identity and supplies the new constant $I_h(x)$.","core_discovery":"For $k=1,2,\\ldots$, the paper establishes that as $s\\to -\\infty$, with all auxiliary parameters $t_j$ set to zero, the logarithm of the higher-order Tracy-Widom determinant admits the expansion stated in Theorem 1.1, whose constant term is now explicit: it involves $-I_h(x)$, where $I_h(x)=\\int_{+\\infty}^{x}[h(\\mu)-h_{\\mathrm{Asy}}(\\mu)]d\\mu$, plus polynomial and logarithmic terms in $x$ and the classical Tracy-Widom constant $\\chi(0)$. This closes the open constant problem left by earlier work that derived the asymptotic expansion only up to an undetermined $s$-independent constant $C(k)$. The proof also yields the vanishing total integral $\\int_{-\\infty}^{+\\infty}[h(\\mu)-h_{\\mathrm{Asy}}(\\mu)]d\\mu=0$ for all $k$, and, by substituting a suitable $x$ depending on $s$, the higher-order distribution converges to the classical Tracy-Widom distribution in the large-gap regime.","pith_inferences":["Because the formula (1.29) is uniform in $x$ over an $s$-dependent window, one can likely read off subleading $x$-dependent corrections at the transition to the Airy kernel, beyond the leading transition recorded in Corollary 1.4.","The vanishing total integral of $h-h_{\\mathrm{Asy}}$ suggests that $h-h_{\\mathrm{Asy}}$ has equal tail contributions at $\\pm\\infty$, which may admit a tau-function interpretation and could be derived independently from isomonodromic properties of the Painlevé I hierarchy.","For $k=1$, the explicit constant in Corollary 1.2 can be tested numerically against direct evaluation of the Fredholm determinant for moderately large $|s|$; agreement would provide strong evidence for the general formula, and disagreement would point to the delicate transition-region error cancellation as the source of error.","The same two-derivative contour trick may work for determinants where only the asymptotics of derivatives are accessible, such as kernels built from the Painlevé II hierarchy, with the constant again expressed through a Hamiltonian integral."],"forward_implications":["The formerly open constant $C(k)$ in the known expansion (1.20) is now explicit for every $k$, completing the large-gap asymptotics of higher-order Tracy-Widom distributions.","The total integral of the Hamiltonian difference vanishes for all $k$, extending the known $\\mathrm{P}_I^{2}$ result and giving a global identity for the special solutions of the even Painlevé I hierarchy.","The uniform asymptotics in $x$ yield an explicit transition from the higher-order Tracy-Widom distribution to the classical Tracy-Widom distribution in the large-gap regime, matching a kernel-level transition established earlier.","The paper states that the same strategy, using uniform partial-derivative asymptotics and a two-path contour integration, can be adapted to compute analogous multiplicative constants in other problems from mathematical physics."],"supporting_citations":[{"why":"Derives the expansion (1.20) with the unknown constant $C(k)$ and establishes the connection between the higher-order Tracy-Widom determinant and the Painlevé II hierarchy.","marker":"[17]"},{"why":"Gives the classical Tracy-Widom determinant representation via a Painlevé II Hamiltonian and the leading asymptotics that define the constant $\\chi(0)$.","marker":"[62]"},{"why":"Proves the value of $\\chi(0)$, the classical Tracy-Widom constant that enters the new explicit formula.","marker":"[27]"},{"why":"Independently proves the Tracy-Widom constant $\\chi(0)$ and provides the total-integral identity for the Painlevé II Hamiltonian used in the transition-region computation.","marker":"[1]"},{"why":"Supplies the existence and asymptotics of the real pole-free solution of the even Painlevé I hierarchy and the Hamiltonian asymptotics used to define $h_{\\mathrm{Asy}}$.","marker":"[14]"},{"why":"Provides the nonlinear steepest descent method for Riemann-Hilbert problems that underlies the uniform asymptotic lemmas in all three regions.","marker":"[33]"}],"fun_headline_variants":["Gap constant solved for every higher-order Tracy-Widom law","Explicit TW constant from Painlevé I Hamiltonian integral","Higher-order Tracy-Widom asymptotics: missing constant found","TW hierarchy gap constant resolved, classical limit recovered","Painlevé I hierarchy yields elusive TW multiplicative constant"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole proof hinges on a delicate cancellation: the error term in the transition-region asymptotics of $\\partial F/\\partial s$ grows with $|x|$, and the derivation depends on its integrated contribution over the short transition interval being negligible when substituted into the contour identity (1.41).","fun_headline_variants_meta":{"raw":{"variants":["Gap constant solved for every higher-order Tracy-Widom law","Explicit TW constant from Painlevé I Hamiltonian integral","Higher-order Tracy-Widom asymptotics: missing constant found","TW hierarchy gap constant resolved, classical limit recovered","Painlevé I hierarchy yields elusive TW multiplicative constant"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000251,"raw_usage":{"total_tokens":1564,"prompt_tokens":961,"completion_tokens":603,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":577,"completion_tokens_details":{"reasoning_tokens":522}},"tokens_in":577,"tokens_out":603,"duration_ms":6415,"temperature":1.0,"reasoning_tokens":522,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T16:54:19.301278+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the left and right sides of the cancellation identity (2.19) numerically for large $|s|$ with $k=1$; if their difference is not $O(|s|^{-1/3})$, the extracted constant is wrong. Alternatively, evaluate $\\log\\det(I-K_s^{(1)})$ directly for $s=-40$ and fixed $x=0$, subtract the explicit $s$-dependent terms, and compare the remainder with the $k=1$ case of the constant formula in Corollary 1.2.","supporting_citations":[{"cited_title":"Claeys, A","cited_arxiv_id":null,"evidence_quote":"Derives the expansion (1.20) with the unknown constant $C(k)$ and establishes the connection between the higher-order Tracy-Widom determinant and the Painlevé II hierarchy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the classical Tracy-Widom determinant representation via a Painlevé II Hamiltonian and the leading asymptotics that define the constant $\\chi(0)$."},{"cited_title":"Deift, A","cited_arxiv_id":null,"evidence_quote":"Proves the value of $\\chi(0)$, the classical Tracy-Widom constant that enters the new explicit formula."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Independently proves the Tracy-Widom constant $\\chi(0)$ and provides the total-integral identity for the Painlevé II Hamiltonian used in the transition-region computation."},{"cited_title":"Claeys, Pole-free solutions of the first Painlev´ e hierarchy and non-generic critical be- havior for the KdV equation, Phys","cited_arxiv_id":null,"evidence_quote":"Supplies the existence and asymptotics of the real pole-free solution of the even Painlevé I hierarchy and the Hamiltonian asymptotics used to define $h_{\\mathrm{Asy}}$."},{"cited_title":"Deift and X","cited_arxiv_id":null,"evidence_quote":"Provides the nonlinear steepest descent method for Riemann-Hilbert problems that underlies the uniform asymptotic lemmas in all three regions."}],"review_version":1}