Higher-Order Stokes Phenomenon

• Higher-Order Stokes Phenomenon is an advanced asymptotic effect where the classical switching of exponentially small terms is itself modified by additional switching processes.
• It is characterized by new Stokes curves, modified singulant conditions, and nonclassical multipliers that redefine how subdominant exponentials are activated.
• The phenomenon has broad applications spanning singular perturbation theory, hyperasymptotics, discrete systems, and multi-component WKBJ analyses.

Higher-order Stokes phenomenon denotes a class of beyond-all-orders switching effects in which the usual Stokes mechanism is itself modified by additional asymptotic structure. In the classical Stokes phenomenon, exponentially small contributions are switched on or off across curves where phases align, typically characterized by a singulant or action difference satisfying $\Im \chi = 0$ with $\Re \chi > 0$. In the higher-order setting, the ability of such exponentially small terms to trigger further switching can change across additional loci, new Stokes curves can be born or truncated, and the effective Stokes multipliers can depart from the standard half-jump picture associated with Dingle’s final main rule. Across the literature, this phenomenon appears in several technically distinct but closely related forms: late-term and late-late-term divergence in singular perturbation theory, sheet changes of Borel singularities in hyperasymptotics, cubic corrections near semiclassical caustics, infinite saddle families in discrete equations, multisummability in moment partial differential equations, and automorphisms of Stokes multipliers in multi-component WKBJ systems (Shelton et al., 2023).

1. Classical framework and the point of departure

Classical Stokes phenomenon concerns asymptotic representations that differ between sectors of the complex plane. For two competing exponentials with singulants $\chi_1,\chi_2$, Stokes curves satisfy

$\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$

while anti-Stokes curves satisfy

$\Re(\chi_1-\chi_2)=0.$

In Berry’s convention, an exponentially subdominant term changes multiplier from approximately $0$ to approximately $1$ across the Stokes curve, and the transition is smoothed by an error function rather than occurring as a literal discontinuity (1802.06407).

Higher-order Stokes phenomenon begins where this two-exponential picture ceases to be sufficient. In exponential asymptotics of singularly perturbed ODEs, an already switched-on exponentially small contribution can itself possess divergent late terms, generating further singulants and additional exponentials such as $e^{-(\chi_1+\chi_2)/\varepsilon}$ or $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$. The corresponding higher-order Stokes line is governed by Dingle-type conditions on singulant ratios rather than on a single singulant alone (Shelton et al., 2023). In Borel-plane language, the same effect occurs when singularities move between different Riemann sheets, so that a Borel singularity can be inactive on one side of a curve and active on the other (Howls et al., 2024).

This shift in viewpoint is decisive. The subject is not merely the switching of exponentials, but the switching of the switching mechanism itself. That is the sense in which the phenomenon is “higher-order” in the papers of Howls, Olde Daalhuis, Chapman, Berry, and later developments (Nemes, 2022).

2. Late-term divergence, Borel geometry, and higher-order lines

A standard starting point is a divergent asymptotic expansion

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$

with late coefficients of factorial-over-power type

$\Re \chi > 0$0

Optimal truncation near $\Re \chi > 0$1 leaves an exponentially small remainder of the form $\Re \chi > 0$2. Higher-order structure appears when the late coefficients themselves contain subsidiary divergent expansions. In the notation of the singular perturbation literature,

$\Re \chi > 0$3

and the inner coefficients can again diverge as

$\Re \chi > 0$4

This induces a new exponential scale controlled by $\Re \chi > 0$5, and the corresponding higher-order switching is detected by

$\Re \chi > 0$6

or, equivalently in geometric Borel-plane form,

$\Re \chi > 0$7

These are the diagnostic conditions for higher-order Stokes lines in the model singularly perturbed ODE analyzed by Chapman and coauthors (Shelton et al., 2023).

The model equation

$\Re \chi > 0$8

provides an explicit instance. Its primary singulant is

$\Re \chi > 0$9

while late-late divergence introduces

$\chi_1,\chi_2$0

The exact Borel transform

$\chi_1,\chi_2$1

shows directly that the branch point at $\chi_1,\chi_2$2 and the pole at $\chi_1,\chi_2$3 lie on different sheets depending on the region of the $\chi_1,\chi_2$4-plane. The higher-order Stokes line is precisely the locus where the pole becomes visible from the Laplace origin (Shelton et al., 2023).

This Borel-sheet mechanism also underlies hyperasymptotic treatments based on hyperterminants. The second hyperterminant

$\chi_1,\chi_2$5

contains two pole mechanisms, at $\chi_1,\chi_2$6 and $\chi_1,\chi_2$7, and each can be responsible for switching. When these switchings interact, one obtains higher-order Stokes curves, new local contributions, and line truncation phenomena (Howls et al., 2024).

3. Smoothing, atypical multipliers, and ghost contributions

Berry’s 1989 result established that ordinary Stokes switching is smooth: after optimal truncation, the multiplier varies with an error-function profile rather than jumping discontinuously. Recent work extends this principle to higher-order Stokes phenomena. The universal smoother is no longer the elementary complementary error function alone, but a new special function defined as a Gaussian convolution of $\chi_1,\chi_2$8,

$\chi_1,\chi_2$9

This function governs the simultaneous smoothing of the two higher-order switchings encoded in the second hyperterminant, and it reduces to classical Berry smoothing in special limits (Howls et al., 2024).

The higher-order setting also alters the on-line value of the multiplier. For the scaled gamma function

$\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$0

the transseries contains exponentials $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$1, but the corresponding Stokes multipliers on $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$2 are not always $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$3. Instead,

$\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$4

while for the reciprocal gamma function the sub-subdominant multipliers with $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$5 are nonzero only on the Stokes line itself (Nemes, 2022). At $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$6, the on-line values become $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$7 for $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$8 and $\Im(\chi_1-\chi_2)=0,\qquad \Re(\chi_1-\chi_2)>0,$9 for $\Re(\chi_1-\chi_2)=0.$0, realized through combinations of $\Re(\chi_1-\chi_2)=0.$1 and $\Re(\chi_1-\chi_2)=0.$2 in the level-2 hyperasymptotic remainder (Nemes, 2022).

In the collinear case for higher-order smoothing, a “ghost-like” contribution survives exactly on the Stokes line with multiplier

$\Re(\chi_1-\chi_2)=0.$3

rather than the classical $\Re(\chi_1-\chi_2)=0.$4. This ghost term decays rapidly on either side, so it is present only in the immediate vicinity of the line (Howls et al., 2024).

A plausible implication is that the phrase “higher-order Stokes phenomenon” should not be reserved for the birth of new lines alone. The modern hyperasymptotic usage encompasses the modified local law of switching, the emergence of nonclassical on-line multipliers, and the possibility that an exponentially small term can be active only on the Stokes line itself (Nemes, 2022).

4. Local normal forms, caustics, and semiclassical propagation

In semiclassical mechanics, higher-order Stokes phenomenon arises in a distinct but structurally related guise. Near a complex phase-space caustic, the projection from a complex trajectory manifold to the observable space becomes singular. For the projected coordinate

$\Re(\chi_1-\chi_2)=0.$5

a caustic occurs when

$\Re(\chi_1-\chi_2)=0.$6

The quadratic local expansion about $\Re(\chi_1-\chi_2)=0.$7,

$\Re(\chi_1-\chi_2)=0.$8

is symmetric under $\Re(\chi_1-\chi_2)=0.$9 and therefore produces a double covering. One branch is unphysical and yields the familiar Stokes divergence (1802.06407).

The key observation is that the symmetry is first broken by the cubic term. Denoting the conjugate point by $0$0, Koch and Tannor derive

$0$1

The Stokes variable

$0$2

then satisfies the phase-space caustic Stokes expansion

$0$3

Because $0$4, there are six rays in $0$5-space with $0$6, corresponding to three Stokes lines in $0$7-space separated by $0$8 (1802.06407).

This is a higher-order Stokes phenomenon in a very specific sense. The relevant novelty is not a higher catastrophe such as a cusp or swallowtail, although those belong to the same general circle of ideas. Rather, the cubic deviation beyond the quadratic normal form determines the actual Stokes geometry and identifies the sector associated with the unphysical sheet. Retaining the correction $0$9 is essential, since setting $1$0 gives incorrect Stokes directions (1802.06407).

The resulting elimination procedure combines Adachi’s principle of exponential dominance with Berry smoothing. The unphysical sector is removed where $1$1 diverges, and in the adjacent sectors the contribution is multiplied by

$1$2

Applied within the FINCO reconstruction framework, this systematic removal of the wrong sheet yields stable wavepacket reconstruction for the Morse, Quartic, Coulomb, and Eckart systems, with semi-quantitative agreement with exact quantum results over long times and in barrier transmission (1802.06407).

5. Discrete, $1$3-difference, and continuum-limit manifestations

Discrete and advance-delay equations exhibit especially rich higher-order Stokes structure because discretization can generate infinitely many saddles. For the discrete Airy equation

$1$4

Fourier analysis leads to saddle families

$1$5

so the steepest-descent geometry involves infinitely many pairwise interactions. This produces infinitely many Stokes and anti-Stokes curves, higher-order Stokes curves, Stokes-curve truncation at crossing points, and infinite accumulations of curves in the complex plane (Moston-Duggan et al., 7 Oct 2025).

For $1$6, the active Stokes curves $1$7 from $1$8 and $1$9 from $e^{-(\chi_1+\chi_2)/\varepsilon}$0 intersect at

$e^{-(\chi_1+\chi_2)/\varepsilon}$1

and these ordinary Stokes curves are truncated by higher-order curves

$e^{-(\chi_1+\chi_2)/\varepsilon}$2

A virtual turning point at $e^{-(\chi_1+\chi_2)/\varepsilon}$3 also appears, even though it is not a genuine turning point of the continuous Airy equation (Moston-Duggan et al., 7 Oct 2025). This makes the discrete problem a canonical example of higher-order Stokes structure created solely by discretization.

A related but conceptually different discrete result concerns continuum limits of discrete Painlevé I. For the finite-difference discretization, the infinite-order differential representation yields the singulant condition

$e^{-(\chi_1+\chi_2)/\varepsilon}$4

so the limiting discrete Stokes pattern is an evenly spaced imaginary lattice. As the truncation order $e^{-(\chi_1+\chi_2)/\varepsilon}$5 increases in the associated finite-order ODE hierarchy, the roots of the singulant polynomial cluster near this lattice, and the higher-order continuous problems organize the orderly discrete Stokes switching seen in the full infinite-order limit (Lustri et al., 27 Mar 2025).

The integrable discrete Painlevé I case is exceptional. Although the same formal singulant condition $e^{-(\chi_1+\chi_2)/\varepsilon}$6 appears, the inner analysis gives a vanishing Stokes constant,

$e^{-(\chi_1+\chi_2)/\varepsilon}$7

so the subdominant exponentials are absent and Stokes phenomenon disappears in the continuum limit of the integrable equation (Lustri et al., 27 Mar 2025). By contrast, the non-integrable finite-difference case has nonzero prefactor and genuine Stokes switching.

In nonlinear $e^{-(\chi_1+\chi_2)/\varepsilon}$8-Painlevé I, the Stokes curves become $e^{-(\chi_1+\chi_2)/\varepsilon}$9-spirals rather than straight rays. After the rescaling $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$0, $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$1, and $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$2, the singulant equation takes the form

$e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$3

and Stokes curves are determined by

$e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$4

Mapping back to the $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$5-plane via $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$6 turns these into $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$7-spirals. For the Type B branch, three singulants $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$8 coexist, each with its own Stokes curves, and the geometry includes inter-sheet continuation across branch cuts, which the paper identifies as a genuine higher-order Stokes structure (Joshi et al., 2018).

These discrete and $e^{-(\chi_1+\chi_2+\chi_3)/\varepsilon}$9-difference examples establish that higher-order Stokes phenomenon is not confined to high-order differential equations. Discretization alone can induce the relevant complexity, including infinite saddle ladders, virtual turning points, and accumulations of switching curves (Moston-Duggan et al., 7 Oct 2025).

6. Multisummability, algebraic automorphisms, and scope

In moment partial differential equations with constant coefficients in two complex variables, higher-order Stokes structure is organized by multisummability. After factorizing the characteristic polynomial, the formal solution decomposes into simple components whose Gevrey orders are

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$0

and hence into summability levels

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$1

Each level has its own finite set of singular directions

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$2

together with corresponding Stokes and anti-Stokes lines. Crossing a level-$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$3 Stokes line affects only the level-$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$4 summand, and the jump is represented as the $$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$5-Laplace image of a hyperfunctional boundary value of the moment-Borel transform (Michalik et al., 2018).

In multi-component WKBJ problems, the same phenomenon can be expressed algebraically through automorphisms of Stokes multipliers. Regular Stokes lines are

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$6

and crossing $$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$7 updates the sector parameter by

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$8

Higher-order Stokes lines are

$$u(x;\varepsilon)\sim \sum_{n=0}^\infty u_n(x)\,\varepsilon^n,$$9

and crossing $\Re \chi > 0$00 updates the Stokes constant by

$\Re \chi > 0$01

This formalizes “Stokes-of-Stokes” behavior: the coefficients controlling ordinary Stokes jumps themselves jump across higher-order lines (Shelton et al., 5 Mar 2026).

The Swallowtail canonical integral,

$\Re \chi > 0$02

is the minimal four-component WKBJ example in which one sees intersections of higher-order lines and corresponding changes of higher-order automorphisms. The analysis shows that with four components the automorphism associated with higher-order Stokes phenomenon can itself change value across another higher-order Stokes line. The same paper argues that five or more WKBJ components introduce no qualitatively new mechanism beyond more instances of the same triple-sector update pattern (Shelton et al., 5 Mar 2026).

Taken together, these frameworks delimit the scope of the subject. In singular perturbation theory, the phenomenon is encoded in late-late divergence and Borel-sheet geometry; in semiclassics, in cubic deviations from symmetric local caustic forms; in discrete systems, in infinite saddle ladders and turning-point proliferation; in PDEs, in multiple Gevrey levels and multisummability; and in modern resurgent WKBJ theory, in automorphisms acting directly on Stokes multipliers. This suggests that “higher-order Stokes phenomenon” is best understood not as a single isolated effect, but as a family of structurally analogous mechanisms by which the geometry and algebra of exponentially small terms acquire a second layer of switching (Howls et al., 2024).