Asymptotic and Special-Function Methods

Use this chapter when a QFT calculation cannot be completed by exact evaluation alone: a parameter becomes large or small, an integral localizes, an oscillatory phase develops critical points, an ordinary differential equation has a slowly varying regime, endpoint scaling produces powers and logarithms, or spectral data must be reorganized through a heat trace. The organizing rule is to choose the method from the mathematical input—equation, contour, spectrum, boundary data, or canonical geometry—and then state the limit, domain, sector, branch, uniformity, and remainder before using the answer.

There are three independent entrances. The asymptotic-series route begins with scales, remainders, and uniformity, then branches to Laplace or steepest-descent integrals, stationary phase, or WKB. The mode-function route begins with linear ODEs and goes directly to special functions selected by boundary and normalization data; it does not require WKB or steepest descent. The spectral route joins Lebesgue integration and complex analysis at the Mellin transform, then combines Mellin information with elliptic heat-kernel theory and spectral calculus. Stationary phase also meets symplectic geometry in the advanced route through Lagrangian submanifolds and generating functions.

These methods extract controlled information from integrals and differential equations when exact evaluation is unavailable. They do not by themselves select a physical path-integral cycle, construct a continuum functional measure, define a vacuum globally, or provide a renormalization prescription. Full resurgence, transseries, and physical saddle sectors belong to Nonperturbative Dynamics; physical determinants and subtraction schemes belong to their physics volumes; and Riemann-surface or modular physics belongs to Conformal Field Theory.

Diagnose · Choose a route · Compare methods · Dependencies · Conventions · Page guide · Airy fold · Kernel thread · Review · Continue

Enter through the mathematical input

This overview has no prerequisite. The target pages do have prerequisites, but uncertainty in one branch need not block another. In particular, a reader who can solve and normalize a linear ODE may enter the special-function page without first learning saddle methods, while a reader interested in spectral asymptotics may enter through Mellin analysis without first learning WKB.

Observable readiness checks and exact repair routes

Check	Ready	Unsure	Repair
Can you distinguish a fixed-order asymptotic statement from convergence, and can you name the limiting variable and uniform set?	Enter asymptotic scales directly.	Write one remainder after truncation at a fixed order and say which quantities are held fixed.	Review Limits, Completeness, and Modes of Convergence, then use Asymptotic Scales, Remainders, Uniformity, and Optimal Truncation.
For a large-parameter integral, can you identify its original oriented contour, endpoints, accessible extrema or saddles, and local quadratic form?	Enter Laplace or steepest descent after asymptotic scales.	Check separately whether a critical point exists and whether its descent cycle can be reached legally.	Use Asymptotic Scales; review Holomorphic Functions and Cauchy Theory before Laplace Method and Steepest Descent.
For an oscillatory integral, can you compute the Hessian signature and recognize when critical points coalesce?	Enter stationary phase after asymptotic scales.	Compare the critical-point separation with the local large-parameter width; a vanishing Hessian requires a uniform model.	Use Asymptotic Scales; Laplace Method is recommended before Stationary Phase, Coalescing Saddles, and Stokes Geometry.
Can you solve a second-order linear ODE, propagate initial data, and use a Wronskian to test independence and normalization?	Enter exact special functions, or add asymptotic scales for WKB.	Write the equation in normal form and track how a change of variable transforms the Wronskian.	Repair Linear ODEs, Evolution Operators, and Wronskians. Then choose Special Functions or, with asymptotic preparation, WKB and Eikonal Methods.
Can you state the Mellin fundamental strip and justify inversion or a contour shift rather than only listing residues?	Enter Mellin transforms.	Check endpoint integrability, meromorphic continuation, vertical growth, crossed poles, and the final contour.	Repair Lebesgue Integration and Convergence Theorems and Cauchy Theory, then use Mellin Transforms and Scaling Asymptotics.
Given Bessel, Airy, Hankel, or hypergeometric solutions, can you select a basis from the equation, domain, boundary data, branch, and normalization?	Enter special functions after linear ODEs.	Do not choose by function name; test endpoint behavior and the Wronskian on the intended continuation path.	Use Linear ODEs, then Special Functions from Equations and Boundary Data.
Can you specify an operator domain and boundary condition, separate zero modes, and distinguish a heat trace from a zeta-regularized determinant?	Enter heat kernels and zeta functions after the three required branches meet.	List positivity, self-adjointness, compact-resolvent assumptions, boundary data, dimension, and the scale used to make the determinant dimensionless.	Combine Mellin Transforms, Elliptic Boundary Problems and Heat Kernels, and Spectra, Resolvents, Spectral Measures, and Functional Calculus.
Can you distinguish a smooth Lagrangian submanifold from a singular projection and say when a generating function is only local?	Enter Lagrangian phases after stationary phase and symplectic geometry.	Check isotropy, dimension, projection rank, and whether auxiliary phase variables are needed near a caustic.	Combine Stationary Phase with Symplectic Forms, Hamiltonian Flows, and Poisson Brackets, then use Lagrangian Submanifolds and Generating Functions.

These checks choose a route. The mathematical readiness diagnostic reports the relevant capabilities independently. The complex and asymptotic methods repair reviews contours, analytic continuation, saddle methods, and domain-aware asymptotics. It does not by itself prepare general WKB, special-function boundary problems, Mellin analysis, heat/zeta theory, or Lagrangian geometry.

Choose a route from the QFT question

In this chapter, requires means preparation used in the target page’s main argument. Recommended preparation improves fluency but does not block entry. Continue points to a later volume where the mathematical structure receives a developed physical role.

Goal-to-route choices through asymptotic and special-function methods

Reader goal	Route and preparation	Observable result
Control a large- or small-parameter approximation	Enter Asymptotic Scales. Branch to Laplace/Steepest Descent for decay integrals, Stationary Phase for oscillatory integrals, or WKB for slowly varying ODEs.	State the limit, fixed data, sector, uniform set, truncation order, branch, and remainder meaning; replace an uncontrolled formal approximation by a claim with a declared regime.
Choose and normalize an exact radial, curved-space, or bounded mode	Requires: Linear ODEs. Enter Special Functions from Equations and Boundary Data directly. Add WKB only for a slowly varying limit or turning-point match.	Derive the canonical equation, choose a basis from endpoint or frequency data, fix branches and continuation paths, and verify normalization with a Wronskian.
Extract powers and logarithms from endpoint scaling or a spectral sum	Requires: Lebesgue integration and Cauchy theory. Enter Mellin Transforms and Scaling Asymptotics; continue to Large Logarithms and RG Improvement only after the physical origin of the scale dependence is established.	Find a fundamental strip, translate powers and logarithms into pole location and order, and justify every contour shift with growth and remainder control.
Pass from short-time spectral data to an analytic determinant	Requires: Mellin transforms, elliptic heat-kernel theory, and spectral calculus. Enter Heat Kernels, Zeta Functions, and Spectral Determinants, then continue to Integrating Out Heavy Fields.	Specify the operator and domain, remove or retain zero modes explicitly, relate the primed heat trace to a spectral zeta function, and distinguish analytic regularization from physical renormalization.
Describe a semiclassical phase through a caustic	Requires: stationary phase and symplectic geometry. Enter Lagrangian Submanifolds, Generating Functions, and Semiclassical Phases, then continue to Negative Modes and Instability Indices.	Replace a singular graph phase by a smooth Lagrangian relation or generating family, track projection failure and phase correction, and keep the physical integration cycle as separate data.

The method-selection principle is not merely pedagogical. A real nondegenerate minimum, an accessible complex saddle, an oscillatory critical point, two coalescing critical points, and a turning point have different local scales and different canonical models. Hunter develops these distinctions and their Gaussian or Airy reductions in Hunter 2004, Chapter 2 and §§3.1 and 3.3–3.6, pp. 19–31 and 35–47, PDF; maintained definitions, sector restrictions, and canonical formulas are collected in NIST DLMF 2026, § 2.1, NIST DLMF 2026, §§ 2.3(iii)–(iv), NIST DLMF 2026, §§ 2.4(iii)–(vi), NIST DLMF 2026, § 2.7(iii), and NIST DLMF 2026, § 2.8(iii).

Keep the mathematical objects distinct

The same symbol can conceal different levels of control. The comparison below records what each object licenses and what additional statement is still needed.

Asymptotic structures, warranted conclusions, and stop rules

Structure	Warranted conclusion	What does not follow automatically
Poincaré asymptotic expansion at fixed order	After truncation at each fixed order, the remainder is smaller than the last retained scale in the declared limit and region.	The infinite series need not converge, determine the exact function uniquely, or control a truncation order that grows with the limiting parameter.
Uniform asymptotic expansion	A named norm or supremum controls the remainder uniformly over a stated parameter set.	Pointwise control does not imply uniformity near endpoints, coalescing saddles, turning points, or singular parameter values.
Optimally truncated approximation	Under an order-dependent coefficient or remainder bound, truncating near the least term can expose an exponentially small error scale.	A decreasing finite list of terms does not prove a least-term error bound, identify a new saddle, or establish a transseries.
Critical point of an exponent	Local expansion gives a candidate Gaussian, Airy, or higher canonical contribution according to the degeneracy.	A critical point contributes only when the original oriented contour can be deformed to an appropriate descent cycle without crossing forbidden singularities or endpoints.
Stationary-phase expansion	An isolated nondegenerate critical point gives an algebraic expansion with a Hessian-signature phase for the declared exponential convention.	It is not uniform through coalescence, and a Stokes change in an asymptotic representation need not be a jump of the exact analytic object.
WKB or eikonal approximation	A phase-amplitude ansatz solves a slowly varying ODE or wave equation up to a calculable residual away from turning points and caustics.	A small local residual is not a global solution-error theorem, and an evanescent region does not by itself establish tunneling or particle creation.
Named special-function family	The canonical equation supplies standard bases, identities, Wronskians, and continuation formulas.	The family name does not select the solution; domain, endpoint data, branch, continuation path, and normalization still have to be supplied.
Meromorphically continued Mellin transform	With mapping-theorem hypotheses, poles encode asymptotic powers and higher-order poles encode logarithms.	The continuation is not the original integral outside its fundamental strip, and residue extraction is incomplete without vertical and final-contour estimates.
Short-time heat expansion and spectral zeta function	Under elliptic spectral hypotheses, local heat coefficients control candidate zeta poles and analytic continuation defines spectral invariants.	The heat series need not converge globally; it does not determine the full spectrum, and a zeta determinant is not an ordinary product or a renormalized observable.
Lagrangian submanifold and generating phase	A Lagrangian relation organizes local semiclassical phases, and auxiliary variables can keep the geometry smooth through a singular projection.	Half-dimension alone does not imply isotropy, one scalar generating function need not exist globally, and a real Lagrangian is not automatically an integration cycle.

The fixed-order, uniformity, and nonuniqueness distinctions follow the Poincaré definitions in Hunter 2004, Chapter 2, pp. 19–28, PDF and NIST DLMF 2026, §2.1(iii)–(v), especially Equations 2.1.13–2.1.20. NIST DLMF 2026, § 2.11(i) also explains why least-term estimates need hypotheses beyond the definition of an asymptotic expansion.

For special functions, the equation and its singular points determine possible local bases, while the actual solution still depends on branch, boundary, and normalization data; see NIST DLMF 2026, § 9.2, NIST DLMF 2026, § 9.7, NIST DLMF 2026, § 10.2, NIST DLMF 2026, § 10.5, NIST DLMF 2026, § 10.17, and NIST DLMF 2026, § 15.10. The corresponding geometric caution at caustics is developed in Bates and Weinstein 1997, §§4.2–4.3, pp. 41–55, and §5.2, pp. 76–78, PDF.

How the eight pages fit together

Chapter order is a guide, not a compulsory eight-page sequence. Across the leaf pages there are twelve direct required relationships—five within this chapter and seven from earlier chapters—and three recommended preparations. The table states the preparation actually used in each page’s main argument.

Required and recommended preparation for each page

Target page	Requires	Recommended, not required
Asymptotic Scales	No hard prerequisite.	Limits, Completeness, and Modes of Convergence.
Laplace Method and Steepest Descent	Asymptotic Scales.	Holomorphic Functions and Cauchy Theory.
Stationary Phase and Stokes Geometry	Asymptotic Scales.	Laplace Method and Steepest Descent.
WKB, Eikonal Methods, and Turning Points	Asymptotic Scales and Linear ODEs, Evolution Operators, and Wronskians.	None.
Mellin Transforms and Scaling Asymptotics	Lebesgue Integration and Convergence Theorems and Cauchy Theory.	None.
Special Functions from Equations and Boundary Data	Linear ODEs, Evolution Operators, and Wronskians.	None.
Heat Kernels, Zeta Functions, and Spectral Determinants	Mellin Transforms, Elliptic Boundary Problems and Heat Kernels, and Spectra, Resolvents, Spectral Measures, and Functional Calculus.	None.
Lagrangian Submanifolds and Generating Functions	Stationary Phase and Symplectic Forms, Hamiltonian Flows, and Poisson Brackets.	None.

Written as prerequisite $\longrightarrow$ target, the five internal required links are

$$\begin{aligned} \text{asymptotic scales} &\longrightarrow \{\text{Laplace/steepest descent},\, \text{stationary phase},\, \text{WKB}\},\\ \text{Mellin transforms} &\longrightarrow \text{heat kernels/zeta/determinants},\\ \text{stationary phase} &\longrightarrow \text{Lagrangian phases}. \end{aligned}$$

Two absences are just as important as the arrows. Special Functions does not require WKB: an exactly reducible ODE with boundary data is already a complete problem. Heat/Zeta does not require the Asymptotic Scales page: its hard entry comes from Mellin analysis, elliptic heat kernels, and spectral calculus. Likewise, Laplace/Steepest Descent is recommended rather than required before Stationary Phase.

The outward links have different meanings. Every leaf has a first application, but only seven have a separate developed physical treatment: the Special Functions page points directly to its curved-spacetime benchmark. Following an application link therefore does not create an additional mathematical prerequisite.

Asymptotic statements and stop rules

Each formula below states the page-local Mellin strip, Fourier or Laplace phase, branch, sector, parameter set, and limiting path at the first point where it matters. Lorentzian-to-Euclidean changes are labeled explicitly.

An asymptotic statement has a complete address

Let $\varepsilon\to0^+$ and let $\phi_{n+1}(\varepsilon)=o(\phi_n(\varepsilon))$. A uniform Poincaré expansion on a named parameter set $K$ means that for every fixed $N$, with $N\ge1$,

$$f(\varepsilon,\lambda) = \sum_{n=0}^{N-1}a_n(\lambda)\phi_n(\varepsilon) +R_N(\varepsilon,\lambda), \qquad \sup_{\lambda\in K} \left| \frac{R_N(\varepsilon,\lambda)} {\phi_{N-1}(\varepsilon)} \right| \longrightarrow0.$$

The order $N$ is fixed before the limit is taken. Replacing it by $N=N(\varepsilon)$ requires an order-dependent bound. If such a bound has factorial growth, a least-term estimate may suggest exponentially small accuracy, but the conclusion belongs to the bound actually proved—not to the symbol $\sim$ alone. Nor does an all-orders algebraic expansion identify which exponentially small sectors occur.

Every asymptotic result in the chapter should therefore name:

• the limiting variable and direction;
• the quantities held fixed and the order of limits;
• the domain or complex sector and all branch choices;
• the parameter set and norm in which uniformity is claimed;
• the truncation order and the precise meaning of the remainder;
• any estimate that licenses differentiation, integration, or parameter-dependent truncation.

Decay and oscillation use different local data

The default decay convention is $\Lambda\to+\infty$ with $e^{-\Lambda S}$; the default oscillatory convention is $e^{+i\Lambda\Phi}$. For a real interior minimum $x_0$ with $S''(x_0)>0$ and $g(x_0)\ne0$,

$$\int_a^b g(x)e^{-\Lambda S(x)}\,\mathrm dx = e^{-\Lambda S(x_0)} g(x_0) \sqrt{\frac{2\pi}{\Lambda S''(x_0)}} \left[1+\mathcal O(\Lambda^{-1})\right]$$

under the stated localization and smoothness hypotheses. A nonstationary left-endpoint minimum with $S'(a)>0$ has width $\Lambda^{-1}$ and a different leading factor. A stationary nondegenerate endpoint instead retains the $\Lambda^{-1/2}$ scale but has a half-Gaussian factor. Neither case should be inserted into the interior formula without rederivation.

For an amplitude supported near a one-dimensional nondegenerate stationary point $x_\sigma$ of a real phase, with $a(x_\sigma)\ne0$,

$$\int a(x)e^{i\Lambda\Phi(x)}\,\mathrm dx \sim a(x_\sigma)e^{i\Lambda\Phi(x_\sigma)} e^{i\pi\,\operatorname{sgn}\Phi''(x_\sigma)/4} \sqrt{\frac{2\pi} {\Lambda|\Phi''(x_\sigma)|}}.$$

Replacing $+i\Lambda\Phi$ by $-i\Lambda\Phi$ conjugates the Fresnel phase. In several variables, the signature of the full Hessian replaces the one-dimensional sign. A complex critical point is only a candidate: orientation, endpoints, singularities, branch cuts, and accessibility from the original contour determine whether its descent cycle contributes. NIST DLMF 2026, §§ 2.3(iii)–(iv) and NIST DLMF 2026, §§ 2.4(iii)–(vi) give the corresponding real, contour, saddle, and coalescence hypotheses.

WKB stops at an ordinary turning point

For

$$\varepsilon^2y''(x)+q(x)y(x)=0, \qquad \varepsilon\to0^+,$$

the positive real branch $p=\sqrt q$ gives oscillatory branches

$$y_\pm(x)\simeq \frac{1}{\sqrt{p(x)}} \exp\left( \pm\frac{i}{\varepsilon} \int^x p(s)\,\mathrm ds \right)$$

where $q>0$. The positive real branch $\kappa=\sqrt{-q}$ gives exponential branches where $q<0$. Complex continuation requires an explicit root branch and path. At a simple zero of $q$, the amplitude above diverges and the separated approximation is nonuniform; an Airy-scaled local solution replaces it. A one-term connection arrow that has discarded a subdominant exponential is not reversible. Liouville–Green error control away from turning points and the uniform simple turning-point construction are stated in NIST DLMF 2026, § 2.7(iii) and NIST DLMF 2026, § 2.8(iii).

Mellin contour directions are part of the answer

This chapter uses

$$\mathcal M[f](s) = \int_0^\infty x^{s-1}f(x)\,\mathrm dx, \qquad f(x) = \frac{1}{2\pi i} \int_{c-i\infty}^{c+i\infty} x^{-s}\mathcal M[f](s)\,\mathrm ds,$$

with the inversion line oriented upward. For small $x$, shift the line to the left and add residues; for large $x$, shift it to the right and subtract residues. These signs are fixed by the contour orientation. The compact check

$$f(x)=\frac{1}{1+x}, \qquad \mathcal M[f](s)=\frac{\pi}{\sin\pi s}, \qquad 0<\operatorname{Re}s<1,$$

recovers the small- and large-$x$ geometric expansions with the stated directions. A pole of order $k$ can produce a polynomial in $\log x$ of degree $k-1$, but only after the mapping-theorem hypotheses control the fundamental strip, continuation, vertical growth, crossed poles, and final line. See NIST DLMF 2026, §2.5(i)–(ii), especially Equations 2.5.1–2.5.7 and 2.5.11 and Flajolet, Gourdon, and Dumas 1995, Part I, §§1–2, pp. 9–21, PDF and the Flajolet, Gourdon, and Dumas 1995 corrigenda, p. 1, PDF.

A determinant begins with an operator contract

Let $A$ be a specified self-adjoint nonnegative Laplace-type operator with compact resolvent on a compact smooth $d$-dimensional Riemannian manifold, with a stated strongly elliptic local boundary condition when a boundary is present. If $P_0$ projects onto $\ker A$, define

$$\Theta_A'(t) = \operatorname{Tr}\!\left(e^{-tA}-P_0\right) = \sum_{\lambda_j>0}e^{-t\lambda_j}.$$

The prime here means zero modes excluded. Initially in a right half-plane,

$$\zeta_A(s) = \frac{1}{\Gamma(s)} \int_0^\infty t^{s-1}\Theta_A'(t)\,\mathrm dt.$$

Short-time heat coefficients organize the meromorphic continuation, although zeros of $1/\Gamma(s)$ can cancel candidate poles. If the continuation is regular at zero, the dimensionless zeta determinant is

$$\log\det\nolimits_\zeta' \left(\frac{A}{\mu^2}\right) = -\left. \frac{\mathrm d}{\mathrm ds} \zeta_{A/\mu^2}(s) \right|_{s=0}.$$

The prime on $\det_\zeta'$ still removes zero modes, whereas the derivative in $s$ is written explicitly to avoid ambiguity. Negative modes require a spectral cut or phase convention rather than silent positivity. Gilkey gives the heat–zeta construction and zero-mode qualification in Gilkey 1984, § 1.10, printed pp. 78–80, PDF pp. 83–85, PDF. Vassilevich develops the determinant scale, zero/negative modes, and one-loop limitations in Vassilevich 2003, §2.2, pp. 14–17, Equations (2.23)–(2.35), PDF.

A phase may need auxiliary variables

On a cotangent bundle use the local canonical one-form $\vartheta=p_i\,\mathrm dq^i$ and $\omega=-\mathrm d\vartheta=\mathrm dq^i\wedge\mathrm dp_i$. A graph $p=\mathrm dS$ is Lagrangian, but a Lagrangian need not remain a graph under projection and need not admit one global scalar $S$. A generating family $\Phi(q,\theta)$ instead describes

$$\partial_\theta\Phi(q,\theta)=0, \qquad p=\partial_q\Phi(q,\theta).$$

The relevant nondegeneracy is the independence of the differentials of the components of $\partial_\theta\Phi$ on their common zero set; it is not the blanket condition $\det\Phi_{\theta\theta}\ne0$. Auxiliary variables can therefore survive precisely where graph phases fail. This finite-dimensional geometry does not choose a QFT integration cycle or regulate a functional determinant.

Exact page guide

Asymptotic Scales, Remainders, Uniformity, and Optimal Truncation

Question. What makes an expansion asymptotic, in which region is it uniform, and what controls truncation error?

Open the page. This is the root of the integral and WKB branches. It defines ordinary and generalized asymptotic scales, fixed-order remainders, pointwise versus uniform expansions, beyond-all-orders ambiguity, and the additional hypotheses needed for least-term truncation. Limits, Completeness, and Modes of Convergence is recommended but not required.

Use the page when the first question is whether a formal expansion has a declared regime. Its main output is not merely a coefficient list: it is a statement with a limiting direction, uniform set, norm, truncation order, and remainder. Continue to Saddles, Control Parameters, and Loop Counting for the developed physical role of large-parameter approximations.

Laplace Method and Steepest Descent

Question. How do dominant saddles and contour geometry determine the asymptotics of an integral?

Open the page. It begins with real Laplace localization, distinguishes interior and endpoint extrema, derives the Gaussian scale, and then moves to complex contour deformation. The page requires Asymptotic Scales; Cauchy theory is recommended because a complex saddle contributes only through a legal deformation of the original oriented contour.

Use it when the exponent and contour, rather than a named special function, are the primary data. The action ranking comes after the accessibility test, and phases must be summed before magnitudes are compared. Pinches, degenerate saddles, singular endpoints, and zero Hessians require another local model. The physical continuation is Negative Modes and Instability Indices.

Stationary Phase, Coalescing Saddles, and Stokes Geometry

Question. How do oscillatory phases, merging critical points, and Stokes changes alter an asymptotic approximation?

Open the page. It derives the signature phase for isolated real critical points, includes endpoint terms, and shows why separate Gaussian contributions fail when critical points merge. The Airy integral supplies the uniform two-saddle model. The page then separates local coalescence from global Stokes changes in a contour decomposition.

Asymptotic Scales is required and Laplace/Steepest Descent is recommended. The page defines Stokes and equal-magnitude sets by equations because naming conventions vary. It does not infer a multiplier from a Stokes equation alone, and it does not claim that the exact analytic function jumps. Physical saddle sectors, transseries, and resurgence continue in Saddles, Control Parameters, and Loop Counting.

WKB and Eikonal Methods and Turning-Point Matching

Question. How do slowly varying phase-amplitude ansatze approximate differential equations and connect across turning points?

Open the page. It develops one-dimensional WKB, higher-dimensional eikonal and transport equations, Wronskian normalization, error diagnostics away from turning points, and Airy matching at a simple zero. It requires both Asymptotic Scales and Linear ODEs, Evolution Operators, and Wronskians.

Use it when the equation has a slowly varying phase or ray structure. A simple Airy connection applies only to a transverse simple turning point; it does not cover higher degeneracy, a saddle–endpoint collision, or a true zero mode. An evanescent branch is mathematical data, not yet a tunneling probability. That physical step belongs to Quantum-Mechanical Instantons and Tunnel Splitting.

Mellin Transforms and Scaling Asymptotics

Question. How do Mellin singularities encode powers, logarithms, and asymptotic scaling?

Open the page. It fixes the Mellin and inversion conventions, derives the fundamental strip from endpoint behavior, proves the direction and sign of contour shifts, and translates pole locations and orders into powers and logarithms. It requires Lebesgue Integration and Convergence Theorems and Holomorphic Functions and Cauchy Theory.

Use it for scaling kernels, harmonic or spectral sums, and integrals whose two endpoints encode different regimes. Analytic continuation of the transform is not the original defining integral, and a transform pole is not automatically a particle pole, a dimensional-regularization pole, or an RG statement. Developed scale physics continues at Large Logarithms and RG Improvement.

Special Functions from Equations and Boundary Data

Question. Why do Bessel, Airy, Hankel, and hypergeometric functions recur, and how do equations and boundary data select the solution?

Open the page. It organizes the recurring families by canonical differential equations, regular or irregular singular points, endpoint behavior, branch cuts, Wronskians, connection formulas, and normalization. Its only required preparation is Linear ODEs, Evolution Operators, and Wronskians.

This is an independent entrance: basic Bessel, Airy, Hankel, and hypergeometric competence requires neither WKB nor steepest descent. Use WKB only when a slowly varying limit or turning-point match is part of the question. The first developed application is Parametric-Oscillator and Solvable Production Benchmarks, where the spacetime patch, state choice, and observable supply physical data that a function table cannot.

Heat Kernels, Zeta Functions, and Spectral Determinants

Question. How do heat traces and analytic continuation define spectral invariants and regularized determinants?

Open the page. It joins three required branches: Mellin transforms, Elliptic Boundary Problems and Heat Kernels, and Spectra, Resolvents, Spectral Measures, and Functional Calculus. It states an operator and boundary-condition contract, distinguishes local short-time coefficients from global spectrum, removes zero modes explicitly, and defines a dimensionless zeta determinant.

Use it for the mathematical input to a one-loop determinant, not for the renormalized observable itself. Boundary conditions, zero and negative modes, spectral cuts, the scale $\mu$, and local counterterms cannot be suppressed. The physical continuation is Integrating Out Heavy Fields.

Lagrangian Submanifolds, Generating Functions, and Semiclassical Phases

Question. How do Lagrangian submanifolds encode canonical relations, boundary data, and semiclassical phases?

Open the page. It combines Stationary Phase with Symplectic Forms, Hamiltonian Flows, and Poisson Brackets. The page develops graph phases, generating families, canonical relations, Van Vleck determinants, projection caustics, and Maslov phases in finite-dimensional models.

Use it when a scalar phase chart becomes singular even though the underlying canonical geometry remains smooth. A caustic can be a failure of projection, not a singularity of the Lagrangian itself. Global phases, composition of canonical relations, gauge constraints, functional measures, and physical cycles need additional hypotheses. Continue to Negative Modes and Instability Indices for the developed semiclassical application.

One Airy fold seen four ways

The Airy fold is a useful synthesis because it joins four genuine viewpoints without pretending to represent the whole chapter. Let $\hbar\to0^+$ and define the real-axis oscillatory integral by Abel damping:

$$\begin{aligned} I_\hbar(x) &= \lim_{\delta\downarrow0} \int_{\mathbb R} e^{-\delta\theta^2} \exp\left[ \frac{i}{\hbar} \left(x\theta-\frac{\theta^3}{3}\right) \right] \,\mathrm d\theta\\ &= 2\pi\hbar^{1/3} \operatorname{Ai}\!\left( -\frac{x}{\hbar^{2/3}} \right). \end{aligned}$$

The equality is exact for the prescribed integral. The real Airy integral and its sectorial asymptotics are fixed in NIST DLMF 2026, § 9.5(i) and NIST DLMF 2026, § 9.7(ii)–(iv). The same canonical function appears in the two-coalescing-saddle construction of NIST DLMF 2026, § 2.4(v) and the simple-turning-point construction of NIST DLMF 2026, § 2.8(iii).

Stationary phase away from the fold

For fixed $x>0$, the phase

$$\Phi(x,\theta)=x\theta-\frac{\theta^3}{3}$$

has two real critical points,

$$\theta_\pm=\pm\sqrt{x}, \qquad \Phi(x,\theta_\pm) = \pm\frac{2}{3}x^{3/2}, \qquad \Phi_{\theta\theta}(x,\theta_\pm) = \mp2\sqrt{x}.$$

The two signature phases interfere to give

$$\begin{aligned} I_\hbar(x) &= 2\sqrt{\pi\hbar}\,x^{-1/4} \cos\left( \frac{2x^{3/2}}{3\hbar}-\frac{\pi}{4} \right)\\ &\quad +\mathcal O\!\left( \hbar^{3/2}x^{-7/4} \right), \qquad \frac{\hbar}{x^{3/2}}\to0. \end{aligned}$$

The remainder statement is uniform only when $x$ stays away from zero in the scaled sense shown. Adding the two contributions is essential: ranking their equal magnitudes would erase the oscillation.

For fixed $x<0$, there are no real stationary points. Exact Airy asymptotics instead give the recessive behavior

$$I_\hbar(x) = \sqrt{\pi\hbar}\,|x|^{-1/4} \exp\left( -\frac{2|x|^{3/2}}{3\hbar} \right) \left[ 1+\mathcal O\!\left( \frac{\hbar}{|x|^{3/2}} \right) \right].$$

This formula does not follow by inserting an imaginary saddle and ignoring the contour. The Abel prescription and Airy continuation determine which combination is present.

WKB failure and exact special-function selection

Differentiating the prescribed integral twice and integrating a total $\theta$ derivative gives the exact equation

$$\hbar^2 I_\hbar''(x)+xI_\hbar(x)=0.$$

Thus $q(x)=x$ in the chapter’s WKB convention: $x>0$ is oscillatory, $x<0$ is evanescent, and $x=0$ is a simple turning point. The WKB amplitude $|x|^{-1/4}$ announces its own failure there. Balancing the two terms in the equation gives the transition layer

$$x=\mathcal O(\hbar^{2/3}),$$

where the exact Airy form is uniform. At the fold,

$$I_\hbar(0) = \frac{2\pi\hbar^{1/3}} {3^{2/3}\Gamma(2/3)},$$

which is finite even though both separated WKB amplitudes diverge. NIST DLMF 2026, Equation 9.2.3 supplies the exact value $\operatorname{Ai}(0)=1/[3^{2/3}\Gamma(2/3)]$ used here.

The same equation has a two-dimensional solution space. The equation alone does not choose $\operatorname{Ai}$ over $\operatorname{Bi}$ or another complex basis; here the real-axis Abel prescription, equivalently the recessive boundary behavior on the $x<0$ side plus normalization, performs that selection. This is why exact special-function competence is independent of having first derived the function by WKB or a saddle integral.

A smooth Lagrangian behind a singular projection

Treat $\Phi(x,\theta)$ as a generating family. Its critical set satisfies

$$\partial_\theta\Phi=x-\theta^2=0, \qquad p=\partial_x\Phi=\theta.$$

The resulting Lagrangian curve is the smooth parabola

$$x=p^2.$$

For $x>0$, its two graph charts have

$$p_\pm(x)=\pm\sqrt{x}, \qquad S_\pm(x)=\pm\frac{2}{3}x^{3/2}.$$

At $x=0$ the projection to the $x$ axis folds, so the graph phases and their stationary amplitudes cease to be regular. The Lagrangian curve and the generating family remain smooth. This is the local geometric content of a fold caustic, not evidence that every caustic is Airy or that every Lagrangian has a global generating family. The phase-function and Maslov framework is developed in Bates and Weinstein 1997, §§4.2–4.3, pp. 41–55, PDF.

The fold therefore supplies one controlled translation:

$$\begin{gathered} \text{two separated stationary points} \quad\longleftrightarrow\quad \text{two WKB graph branches},\\ \text{coalescing critical points} \quad\longleftrightarrow\quad \text{a turning point and projection caustic},\\ \text{one uniform Airy function} \quad\longleftrightarrow\quad \text{one smooth generating family}. \end{gathered}$$

It does not extend by analogy to higher catastrophes, a saddle–endpoint collision, a genuine zero mode, or the separate Mellin–heat spectral route.

A bounded quartic integral checks saddle claims

A finite-dimensional decay integral can test both asymptotic remainders and contour accessibility. For $\Lambda>0$, define

$$Z(\Lambda) = \sqrt{\frac{\Lambda}{2\pi}} \int_{\mathbb R} \exp\left[ -\Lambda\left( \frac{\phi^2}{2}+\frac{\phi^4}{24} \right) \right] \,\mathrm d\phi.$$

After $X=\sqrt{\Lambda}\phi$, this is the Gaussian expectation

$$Z(\Lambda) = \mathbb E\!\left[ \exp\left( -\frac{X^4}{24\Lambda} \right) \right], \qquad X\sim N(0,1).$$

Taylor’s formula for $e^{-u}$ at nonnegative $u$, followed by the Gaussian moments, gives

$$Z(\Lambda) \sim \sum_{n=0}^{\infty} (-1)^n c_n\Lambda^{-n}, \qquad c_n = \frac{(4n)!} {96^n n!(2n)!},$$

and, more strongly, if

$$R_N(\Lambda) = Z(\Lambda) - \sum_{n=0}^{N-1} (-1)^n c_n\Lambda^{-n},$$

then the elementary alternating remainder obeys the exact bound

$$0 \le (-1)^N R_N(\Lambda) \le c_N\Lambda^{-N}.$$

The first terms are

$$Z(\Lambda) \sim 1-\frac{1}{8\Lambda} +\frac{35}{384\Lambda^2} -\frac{385}{3072\Lambda^3} +\cdots.$$

Here the order-dependent bound really does license a least-bound calculation:

$$\frac{c_{n+1}\Lambda^{-(n+1)}} {c_n\Lambda^{-n}} = \frac{(4n+1)(4n+3)} {24(n+1)\Lambda} \sim \frac{2n}{3\Lambda}.$$

The least available upper bound is therefore near $n=3\Lambda/2$ when that order is large. It remains an upper-bound scale; it is not automatically the exact error.

The exponent has critical points

$$\phi=0, \qquad \phi=\pm i\sqrt6.$$

The real saddle produces the algebraic coefficients above. At the complex critical points the action equals $-3/2$, so an isolated local saddle formula would contain the exponentially growing scale $e^{3\Lambda/2}$. But the original real contour is convergent and

$$0<Z(\Lambda)\le1.$$

The direct real contour does not introduce descent paths through those complex critical points, and the exact inequality rules out any uncancelled $e^{3\Lambda/2}$ contribution. On a Stokes boundary, alternative thimble bases can require a lateral convention and can reorganize cancellations; the critical-point list alone does not determine a contribution. This exposes the error in “find every saddle and rank its action.” It also marks the boundary of the example: a finite-dimensional contour calculation does not by itself specify a functional-integral cycle or a physical nonperturbative sector.

The free scalar kernel and a spectral circle

One volume-wide thread reaches this chapter:

$$\begin{aligned} \text{Fourier analysis} &\longrightarrow \text{tempered distributions} \longrightarrow \text{Green operators}\\ &\longrightarrow \text{pole and boundary prescriptions} \longrightarrow \text{spectra} \longrightarrow \text{Gaussian structure} \longrightarrow \text{asymptotics}. \end{aligned}$$

Each arrow adds data rather than replacing the earlier object. Fourier Series, Fourier Transforms, and Plancherel Theory diagonalizes translation-invariant problems. Tempered Distributions and Fourier Calculus gives singular kernels a test-function meaning. Fundamental Solutions and Green Operators states the operator equation and boundary data. Contour Deformation, Pinches, and Causal Prescriptions distinguishes Lorentzian boundary values. Spectra and Resolvents reorganizes the same operator by its spectral measure, while Gaussian Fields and Sources uses its inverse and determinant in a regulated free theory.

A one-dimensional Euclidean circle makes the last steps exact. Let $x\sim x+L$, let $m>0$, and let

$$A_m = -\frac{\mathrm d^2}{\mathrm dx^2}+m^2$$

act on periodic $H^2$ functions in $L^2(S_L^1)$. It is self-adjoint, positive, and has compact resolvent, with

$$\lambda_n = \left(\frac{2\pi n}{L}\right)^2+m^2, \qquad n\in\mathbb Z.$$

Its inverse kernel is the exact Fourier series

$$G_m(x,y) = \frac{1}{L} \sum_{n\in\mathbb Z} \frac{ e^{2\pi i n(x-y)/L} }{ (2\pi n/L)^2+m^2 }.$$

The same spectrum gives the heat trace

$$\Theta_m(t) = \sum_{n\in\mathbb Z} \exp\left[ -t\left( \frac{4\pi^2n^2}{L^2}+m^2 \right) \right].$$

Poisson summation changes momentum modes into winding sectors:

$$\Theta_m(t) = \frac{L}{\sqrt{4\pi t}}\, e^{-m^2t} \sum_{\ell\in\mathbb Z} \exp\left( -\frac{\ell^2L^2}{4t} \right).$$

This equality is exact. As $t\to0^+$, the $\ell=0$ term produces the local half-integer power series obtained by expanding $e^{-m^2t}$, while $\ell\ne0$ terms are exponentially small beyond every power of $t$. The short-time expansion therefore does not contain the full global information in the exact trace.

Because $m>0$, there is no zero mode, and initially $\operatorname{Re}s>1/2$,

$$\zeta_{A_m}(s) = \frac{1}{\Gamma(s)} \int_0^\infty t^{s-1}\Theta_m(t)\,\mathrm dt = \sum_{n\in\mathbb Z}\lambda_n^{-s}.$$

Analytic continuation gives $\zeta_{A_m}(0)=0$ and

$$\det\nolimits_\zeta \left(\frac{A_m}{\mu^2}\right) = 4\sinh^2\left(\frac{mL}{2}\right).$$

There is no $\mu$ dependence in this particular positive massive example because $\zeta_{A_m}(0)=0$. One direct check differentiates with respect to $m^2$:

$$\begin{aligned} \frac{\partial}{\partial m^2} \log\det\nolimits_\zeta \left(\frac{A_m}{\mu^2}\right) &= \operatorname{Tr}A_m^{-1}\\ &= \sum_{n\in\mathbb Z} \frac{1}{ (2\pi n/L)^2+m^2 } = \frac{L}{2m} \coth\left(\frac{mL}{2}\right). \end{aligned}$$

The last equality and the determinant normalization agree with the periodic spectral calculation in Boschi-Filho and Farina 1995, pp. 3–5, Equations (8)–(16), arXiv PDF. The small-mass behavior

$$4\sinh^2\left(\frac{mL}{2}\right) \sim m^2L^2$$

records the eigenvalue that approaches zero. At $m=0$, removing that zero mode defines a different spectral zeta function:

$$\zeta_{A_0}(s) = 2\left(\frac{L}{2\pi}\right)^{2s} \zeta_{\mathrm R}(2s),$$

where the zero eigenvalue is omitted from the left-hand side. Using $\zeta_{\mathrm R}(0)=-1/2$ and $\zeta_{\mathrm R}'(0)=-\tfrac12\log(2\pi)$ gives

$$\det\nolimits_\zeta' \left(\frac{A_0}{\mu^2}\right) = (\mu L)^2.$$

The two Riemann-zeta values are NIST DLMF 2026, Equation 25.6.1 and NIST DLMF 2026, Equation 25.6.11. The prime, the scale, and the limiting procedure cannot be inferred from the unprimed massive formula. This circle calculation demonstrates the spectral bridge—Fourier modes to heat trace to Mellin continuation to determinant—but it does not supply counterterms, a physical normalization condition, or a renormalized one-loop observable.

Chapter-scale synthesis

The examples expose a common pattern:

1. Start from the exact object. Name the integral and oriented contour, the differential equation and domain, or the operator and boundary condition. An approximation cannot repair missing problem data.
2. Identify the controlling geometry. A minimum localizes a decay integral, a stationary point controls oscillation, a turning point degenerates a WKB chart, a Mellin pole records endpoint scaling, and a Lagrangian projection can fold while the relation remains smooth.
3. Choose a representation that stays regular. Gaussian expansions work at isolated nondegenerate points; Airy functions replace two separated branches at a generic fold; generating families replace singular graph phases; heat traces replace formal spectral products.
4. Translate conventions through an invariant. Check a Hessian-signature phase, Wronskian, exact ODE, residue sign, heat–zeta identity, or symplectic form after changing variables or imported conventions.
5. Attach the remainder and stop condition. State where the approximation is uniform, what happens at the boundary of its regime, and which later physical question remains unanswered.

This pattern also explains why the chapter cannot be compressed into a table of named functions. The Airy function is simultaneously an exact ODE solution, a uniform turning-point model, and a fold integral; its role is selected by the problem data. A Mellin pole is useful because a justified inverse transform connects it to an endpoint expansion. A zeta determinant is meaningful because an operator, domain, spectrum, zero-mode rule, and analytic continuation have been specified. The name is the end of the classification, not its beginning.

Review the chapter

1. Give an asymptotic claim its complete address

Review mode: retrieval and comparison.

Tested capability: distinguish equality, convergence, fixed-order asymptoticity, and uniform asymptoticity; state the limit, sector, fixed data, uniform set, norm, and remainder.

Pages needed: this overview and Asymptotic Scales, Remainders, Uniformity, and Optimal Truncation.

Expected response form: two compact claim cards. One should state a pointwise fixed-order expansion; the other should strengthen it to a uniform statement on a named compact parameter set. Each card must list the six data above and say whether its order is fixed or parameter-dependent.

Verification criterion: the response places the truncation before the limit, writes an explicit remainder quotient, and does not infer convergence or an $N(\varepsilon)$ estimate from $\sim$. A reader should be able to decide from the card whether a turning point or endpoint lies inside the uniform set.

Characteristic failure and repair: if “small parameter” is the only domain information, or if least-term truncation is asserted without an order-dependent bound, use Complex and asymptotic methods repair and reread the fixed-order and optimal-truncation sections of Asymptotic Scales.

2. Check a divergent expansion against an exact inequality

Review mode: derivation and proof check.

Tested capability: derive coefficients and a remainder bound, then separate critical-point existence from contour contribution.

Pages needed: Asymptotic Scales, Laplace Method and Steepest Descent, and the quartic benchmark in this overview.

Expected response form: a one-page derivation starting from the Gaussian expectation for $Z(\Lambda)$. Recover $c_n$, the first three corrections, the alternating remainder inequality, the term ratio, all three critical points, and the real-contour bound.

Verification criterion: the response obtains $0\le(-1)^N R_N\le c_N\Lambda^{-N}$ and $0<Z\le1$, and uses the latter to reject an uncancelled $e^{3\Lambda/2}$ contribution. It calls the least term an upper-bound scale rather than the proved exact error.

Characteristic failure and repair: if all critical points are summed without a contour-accessibility argument, repair at Laplace Method and Steepest Descent. If a growing truncation order is used without the displayed bound, repair at Asymptotic Scales.

3. Translate the Airy fold without crossing a failure boundary

Review mode: representation change and failure diagnosis.

Tested capability: translate one exact object among an oscillatory integral, a differential equation, separated WKB branches, and a Lagrangian generating family.

Pages needed: Stationary Phase, Coalescing Saddles, and Stokes Geometry, WKB and Eikonal Methods, Special Functions from Equations and Boundary Data, and Lagrangian Submanifolds and Generating Functions.

Expected response form: a four-panel derivation for $I_\hbar(x)$. Record the two stationary points and signature phases for $x>0$, derive the exact ODE, find the $\hbar^{2/3}$ transition scale, and show that the generated curve is $x=p^2$.

Verification criterion: the two saddle terms reproduce the cosine phase, the WKB amplitude is identified as nonuniform at $x=0$, and the response distinguishes a singular projection from a singular Lagrangian. It states the Abel prescription or equivalent Airy boundary data.

Characteristic failure and repair: if isolated stationary phase is used at $x=0$, repair at Coalescing Saddles. If the Airy formula is extended to a higher-order zero without checking the local normal form, repair at Turning-Point Matching.

4. Select a mode by data rather than by name

Review mode: explanation and comparison.

Tested capability: choose a special-function basis from a canonical equation, domain, endpoint or frequency condition, branch, and Wronskian normalization.

Pages needed: Linear ODEs, Evolution Operators, and Wronskians and Special Functions from Equations and Boundary Data.

Expected response form: a selection table for the Bessel equation on the positive real $z$ axis. Compare $J_\nu$, $Y_\nu$, $H_\nu^{(1)}$, and $H_\nu^{(2)}$ using origin behavior, large-$z$ phase under an explicitly chosen time convention, principal branches, and normalization.

Verification criterion: the table includes $W_z[J_\nu,Y_\nu]=2/(\pi z)$ and $W_z[H_\nu^{(1)},H_\nu^{(2)}]=-4i/(\pi z)$, and it applies the chain rule if $z$ is itself a time or radial coordinate. It never labels a solution “physical” without boundary or state data.

Characteristic failure and repair: if the answer says only “use a Hankel function,” repair at Special Functions from Equations and Boundary Data. If the Wronskian changes incorrectly under $z=z(\eta)$, repair at Linear ODEs.

5. Recover both sides of a Mellin strip

Review mode: transfer.

Tested capability: transfer contour orientation and residue signs to a new Mellin inversion.

Pages needed: Mellin Transforms and Scaling Asymptotics and Holomorphic Functions and Cauchy Theory.

Expected response form: a contour account for $F(s)=\pi/\sin(\pi s)$ on $0<\operatorname{Re}s<1$. Derive the first three small-$x$ and large-$x$ terms. Then state the direct inverse-residue contribution of $A/(s-s_0)^2+B/(s-s_0)$ before applying the contour-shift sign.

Verification criterion: the small-$x$ line moves left and residues are added, giving $1-x+x^2-\cdots$; the large-$x$ line moves right and residues are subtracted, giving $x^{-1}-x^{-2}+x^{-3}-\cdots$. The double-pole residue is $x^{-s_0}(B-A\log x)$ before the shift sign, and the response names the final-line estimate still required.

Characteristic failure and repair: if residues are listed without a fundamental strip, orientation, or final contour, use Complex and asymptotic methods repair and the contour-translation section of Mellin Transforms.

6. Reconstruct a primed spectral determinant

Review mode: derivation, proof check, and representation change.

Tested capability: move from operator data to heat trace, Mellin continuation, and a dimensionless zeta determinant while treating zero modes and scale dependence explicitly.

Pages needed: Mellin Transforms, Elliptic Boundary Problems and Heat Kernels, Spectra and Resolvents, and Heat Kernels, Zeta Functions, and Spectral Determinants.

Expected response form: an operator-to-determinant chain for the periodic circle. State the domain and spectrum, derive the Poisson-summed heat trace, write the initial zeta half-plane, and explain the difference between the $m>0$ determinant and the $m=0$ primed determinant.

Verification criterion: the response subtracts $P_0$ before the primed Mellin transform, retains $1/\Gamma(s)$, obtains $4\sinh^2(mL/2)$ for $m>0$, and explains why $(\mu L)^2$ is a different zero-mode-excluded object at $m=0$. It does not call either expression a renormalized QFT observable.

Characteristic failure and repair: if a zero eigenvalue is left inside a logarithm or the scale $\mu$ disappears without evaluating $\zeta_A(0)$, repair at Heat Kernels, Zeta Functions, and Spectral Determinants. If the heat trace is used without an operator domain or boundary condition, repair at Elliptic Boundary Problems and Heat Kernels.

7. Choose three routes without inventing a ladder

Review mode: synthesis.

Tested capability: choose the shortest valid route and distinguish mathematical preparation, first application, and developed physical treatment.

Pages needed: this overview plus the target pages selected for three cases: an oscillatory integral with merging critical points, an exactly reducible radial mode equation with boundary data, and a positive elliptic operator whose determinant is sought.

Expected response form: a three-row route map. For each case, name every hard prerequisite, any recommended preparation, the mathematical output, its failure boundary, and one correctly typed physical continuation.

Verification criterion: the first route uses Asymptotic Scales and Stationary Phase; the second goes from Linear ODEs directly to Special Functions; the third joins Mellin, elliptic heat kernels, and spectral calculus. The map does not invent WKB $\to$ Special Functions or Asymptotic Scales $\to$ Heat/Zeta as hard prerequisites.

Characteristic failure and repair: if all eight pages appear as one sequence, return to How the eight pages fit together. If the difficulty is broader than this dependency map, use Diagnose mathematical readiness before choosing a repair.

Continue from here

• For physical saddle sectors, negative modes, tunneling cycles, transseries, and resurgence, continue to Nonperturbative Dynamics. Mathematical accessibility of a finite-dimensional saddle is necessary information, not a physical cycle prescription.
• For Mellin-controlled logarithms, matching, subtraction scales, and renormalized determinants, continue to Renormalization and Effective Field Theory. A zeta regularization does not replace a renormalization condition.
• For exact time-dependent modes and particle-production questions, continue to Curved Spacetime QFT. Equation and Wronskian data do not by themselves define a global state or observable.
• For rigorous contour, ODE, elliptic, spectral, or symplectic hypotheses, return to Complex Analysis and Analytic Continuation, Differential Equations and Green Operators, Functional and Spectral Analysis, or Variational, Symplectic, and Constraint Methods.
• For Riemann-surface, monodromy, and modular structures as physical organizing principles, continue to Conformal Field Theory. This chapter uses analytic continuation without absorbing that separate physics.

References

• Sean Bates and Alan Weinstein, Lectures on the Geometry of Quantization, Berkeley Mathematics Lecture Notes 8, American Mathematical Society and University of California, Berkeley (1997), §§ 4.2–4.3, pp. 41–55, and § 5.2, pp. 76–78. AMS book record. Open author-hosted PDF. Phase functions, Lagrangian submanifolds, caustics, Maslov correction, and canonical relations.
• H. Boschi-Filho and C. Farina, “Generalized Thermal Zeta-Functions,” Physics Letters A 205 (1995), 255–260, arXiv:hep-th/9505154. Open arXiv PDF. Periodic spectrum, analytic zeta continuation, and the circle determinant.
• Philippe Flajolet, Xavier Gourdon, and Philippe Dumas, “Mellin Transforms and Asymptotics: Harmonic Sums,” Theoretical Computer Science 144 (1995), 3–58, Part I, §§ 1–2, pp. 9–21. DOI. Open PDF, author-hosted copy. Corrigenda PDF. Fundamental strips, inversion, direct and converse mapping theorems, and the corrected pole-order/log-degree relation.
• Peter B. Gilkey, Invariance Theory, the Heat Equation, and the Atiyah–Singer Index Theorem, Mathematics Lecture Series 11, Publish or Perish (1984), electronic reprint (1996), § 1.10, printed pp. 78–80 (PDF pp. 83–85). Open author-hosted PDF. Heat asymptotics, Mellin continuation of spectral zeta functions, and zero-mode treatment.
• John K. Hunter, Asymptotic Analysis and Singular Perturbation Theory, University of California, Davis lecture notes (February 2004), Chapter 2, pp. 19–28; § 3.1, pp. 29–31; §§ 3.3–3.6, pp. 35–47. Open author-hosted PDF. Asymptotic scales, nonuniformity, optimal truncation, stationary phase, Airy reduction, Laplace’s method, and steepest descent.
• NIST Digital Library of Mathematical Functions, F. W. J. Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller, B. V. Saunders, H. S. Cohl, and M. A. McClain, eds., National Institute of Standards and Technology, Version 1.2.7, released June 15, 2026, Chapters 2, 9, 10, 15, and 25. Maintained online reference. Definitions, canonical integral and ODE models, branches, Wronskians, asymptotic expansions, and exact constants.
• D. V. Vassilevich, “Heat Kernel Expansion: User’s Manual,” Physics Reports 388 (2003), 279–360, § 2.2, pp. 14–17, Equations (2.23)–(2.35). DOI. Open arXiv PDF. Heat–zeta relations, determinant scales, zero and negative modes, and the boundary between analytic regularization and physical renormalization.