Stokes Phenomenon and Anti-Stokes Lines

An asymptotic power series looks like an unambiguous fingerprint of a function, but it is not. Two different functions can share the very same divergent expansion $\sum a_n z^{-n}$ throughout a wedge of the complex plane, differing only by a term such as $e^{-z}$ that is smaller than every power of $1/z$ and therefore invisible to the series — and whether that hidden exponential is present depends on the direction $\arg z$ . The Stokes phenomenon is the precise statement of how it appears: as the phase sweeps across certain critical rays, the coefficient multiplying a subdominant exponential jumps from zero to a definite nonzero value. The power series never changes — what changes is the constant in front of the exponential the series cannot see. Two families of rays organise this geography. On an anti-Stokes line two competing exponentials $e^{S_1}$ and $e^{S_2}$ have equal modulus and trade the roles of dominant and recessive, producing oscillation. On a Stokes line one exponential is maximally larger than the other, and it is exactly there — where the small term is most deeply buried — that switching occurs. Stokes discovered this in 1857 while wrestling with the Airy function, and the same mechanism is the WKB connection problem at a turning point, the saddle reconnection of steepest descent, and the bridge between perturbative and nonperturbative physics in resurgence theory. This article builds all four faces from one idea.

Key takeaways

The Stokes phenomenon is the abrupt switching on or off of a subdominant exponential term in an asymptotic expansion as the phase $\arg z$ of the complex argument is varied.
An asymptotic series represents a function unambiguously only inside a Stokes sector; crossing a Stokes line changes which exponentials the true function requires.
A Stokes line is where one exponential maximally dominates another, $\mathrm{Im}(S_1-S_2)=0$ ; an anti-Stokes line is where they are equal in magnitude, $\mathrm{Re}(S_1-S_2)=0$ , and exchange dominance.
The prototype is the Airy function: its Stokes and anti-Stokes rays fall every $60°$ , and by $\arg z=\pi$ the decaying $e^{-\zeta}$ and switched-on $e^{+\zeta}$ combine into the oscillation $\mathrm{Ai}(-x)\sim \pi^{-1/2}x^{-1/4}\sin(\tfrac23 x^{3/2}+\tfrac{\pi}{4})$ .
The switch is not a true discontinuity: examined on the exponentially small scale near optimal truncation it rises smoothly as a complementary error function (Berry's smoothing).
The Stokes constant links late perturbative coefficients to nonperturbative sectors, the content of resurgence and Borel summation.
Key idea: an asymptotic power series is uniquely tied to a function only within a Stokes sector; the exponentially small terms that distinguish functions sharing that series are switched on and off as the argument's phase crosses Stokes lines, with strengths fixed by Stokes constants.

The core insight

An asymptotic power series is uniquely associated with a function only within a Stokes sector; the exponentially small terms that distinguish functions sharing that series are switched on and off as the argument's phase crosses Stokes lines, with strengths fixed by Stokes constants.

Where this lives today

The Stokes phenomenon has moved from a curiosity of special-function tables to a working tool at the frontier of theoretical physics. In the resurgence programme, the divergent perturbation series of quantum mechanics, quantum field theory, and matrix models are resummed into trans-series whose exponentially small terms are instanton and renormalon contributions; the Stokes constants are the physical data that stitch perturbative and nonperturbative sectors together, and much recent work computes them exactly for models from the double-well potential to topological string theory. Numerically, arbitrary-precision libraries for the Airy, Bessel, and confluent hypergeometric functions must track which sector the argument sits in and switch on the correct exponential combination, or return answers spoiled by catastrophic cancellation. In semiclassical physics the same switching governs above-barrier reflection, tunnelling rates, and the exponentially small splittings of nearly degenerate states. And in nonlinear waves, the Stokes structure of the Painlevé equations organises the connection problem for solitons, dispersive shocks, and random-matrix universality — one nineteenth-century sleepless night now underpinning a large slice of modern asymptotics.

The physical picture

Picture two exponentials competing to represent the same function: one large and dominant, one tiny and subdominant. On the positive real axis you only ever see the large one — the small one is exponentially buried beneath the dominant exponential's divergent tail and leaves no algebraic trace. Now rotate $\arg z$ . The ratio of their sizes slides, and at a special ray the small term can slip in without contradicting anything: it appears precisely where it is most thoroughly hidden. That ray is the Stokes line, and the first figure maps the ray geometry for the Airy function, the canonical laboratory for the effect.

Figure

Stokes geometry of the Airy equation w′′ = z w. With ζ = (2/3)·z^(3/2) the exponents are ±ζ, so Stokes lines Im(2ζ) = 0 fall at arg z = 0, 120°, 240° and anti-Stokes lines Re(2ζ) = 0 at 60°, 180°, 300° — alternating every 60°. Only the decaying e^(−ζ) lives in the recessive wedge, and on the negative axis both exponentials are equal and oscillate.

Why must the hidden term appear at all? Because a single dominant exponential times a power series is generically not the exact function once you leave the starting sector; analytic continuation of the true function demands the second exponential, so reconstructing it means re-injecting the lost term with exactly the right coefficient — the Stokes constant. Yet the appearance is not really a jump. Treat the optimal-truncation remainder as a thin exponentially small mist; the Stokes constant is the strength of a new mist that condenses as you cross the line, and Berry showed the condensation is an error-function rise of finite width, sharp only in the strict large-argument limit. The second figure contrasts the naive switch with this smoothed reality.

Figure

Berry's smoothed Stokes multiplier. The naive picture jumps 0 → 1 at the Stokes line, but the true multiplier rises smoothly through exactly 1/2 on the line, with an angular width that shrinks as the two exponents grow more disparate. The step is recovered only in the extreme-dominance limit.

Geometrically, in the method of steepest descent the Stokes line is the ray where the descent path through one saddle suddenly connects to a second saddle, forcing its contribution into the answer — the concrete picture behind the abstract switching, and where the phenomenon first bites in real integral calculations.

The core equations

An asymptotic expansion is a statement attached to a sector, not to the whole plane. The wedge $\alpha<\arg z<\beta$ over which the same coefficients $a_n$ hold is the Stokes sector, and its edges are typically Stokes lines themselves.

f(z)\sim \sum_{n=0}^{\infty} a_n\,z^{-n}\quad (z\to\infty,\ \alpha<\arg z<\beta) \qquad (1)

where: $f(z)$ — the function, analytic in the sector; $a_n$ — asymptotic coefficients, typically giving a divergent series; $\alpha,\beta$ — edges of the Stokes sector of validity. Reading: the tilde means the truncated series approximates $f$ to any fixed order as $z\to\infty$ within the wedge — nothing is claimed outside it. The full function in a wider region is a sum of two exponentials, each carrying its own power series, with the subdominant one multiplied by a switch $K$ that is dead on one side of a Stokes line and alive on the other.

f(z)=\underbrace{e^{S_1(z)}\sum_n a_n^{(1)} z^{-n}}_{\text{dominant}}\;+\;K\,\underbrace{e^{S_2(z)}\sum_n a_n^{(2)} z^{-n}}_{\text{subdominant}} \qquad (2)

where: $S_{1,2}(z)$ — exponents of the two competing WKB or saddle terms; $K$ — the Stokes multiplier, $0$ on one side, the Stokes constant on the other; $a_n^{(1,2)}$ — coefficients attached to each exponential. Reading: away from the switching ray the subdominant term is beyond all orders, so the single series (1) suffices. Which ray is which is fixed by the exponent difference, never by the names alone.

\text{Stokes line:}\ \mathrm{Im}\big(S_1(z)-S_2(z)\big)=0;\qquad \text{anti-Stokes:}\ \mathrm{Re}\big(S_1-S_2\big)=0 \qquad (3)

where: $S_1-S_2$ — the exponent difference controlling the relative size of the two terms. Reading: on a Stokes line the ratio $|e^{S_1}/e^{S_2}|$ is extremal (maximal dominance, switching); on an anti-Stokes line the ratio is unity (equal magnitude, oscillation). The Airy function realises all of this. For $z>0$ it is the recessive solution and needs only the decaying exponential.

\mathrm{Ai}(z)\sim \frac{e^{-\zeta}}{2\sqrt{\pi}\,z^{1/4}}\sum_{n\ge0}\frac{(-1)^n c_n}{\zeta^{n}},\qquad \zeta=\tfrac{2}{3}z^{3/2} \qquad (4)

where: $\mathrm{Ai}(z)$ — the Airy function, prototype of the phenomenon; $\zeta=\tfrac23 z^{3/2}$ — the WKB exponent; $c_n$ — coefficients with $c_0=1,\ c_1=\tfrac{5}{72}$ . Reading: because $\zeta$ carries a $z^{3/2}$ , its phase rotates half again as fast as $z$ , so the Stokes and anti-Stokes rays land every $60°$ (Figure 1). Once the growing exponential has switched on and the two are equal on the negative axis, they combine into a single oscillation with a signature $\pi/4$ phase.

\mathrm{Ai}(-x)\sim \frac{1}{\sqrt{\pi}\,x^{1/4}}\sin\!\Big(\tfrac{2}{3}x^{3/2}+\tfrac{\pi}{4}\Big),\quad x\to+\infty \qquad (5)

where: $x=|z|$ along the negative real axis. Reading: the $\pi/4$ is a direct fingerprint of the Stokes constant and reappears as the Maslov phase in WKB quantization. Sharpen the crossing and the switch is smooth, governed by a complementary error function whose width shrinks as the terms become more disparate.

K(\theta)\;\approx\;\frac{1}{2}\,\mathrm{erfc}\!\left(-\theta\sqrt{\tfrac{|\Delta S|}{2}}\right)\cdot S \qquad (6)

where: $\theta$ — signed angular distance from the Stokes line; $|\Delta S|$ — magnitude of the exponent difference on the line; $S$ — the Stokes constant (the final jump); $\mathrm{erfc}$ — complementary error function. Reading: at $\theta=0$ the multiplier sits at exactly $S/2$ , approaching a step only as $|\Delta S|\to\infty$ . Assembling every exponential sector at once gives the trans-series of resurgence, whose striking consequence is that the late coefficients of the leading series encode the early data of the subdominant one.

f(z)=\sum_\sigma e^{S_\sigma(z)}\,z^{\beta_\sigma}\sum_{n\ge0} a_n^{(\sigma)} z^{-n},\qquad a_n^{(0)}\sim \frac{S_{01}}{2\pi i}\,\frac{(n-1)!}{(\Delta S_{01})^{n}}\big(a_0^{(1)}+\cdots\big) \qquad (7)

where: $\sigma$ — labels the saddles/WKB branches; $S_\sigma,\beta_\sigma$ — action and power-law exponent of sector $\sigma$ ; $S_{01}$ — Stokes constant linking sectors $0$ and $1$ ; $\Delta S_{01}$ — action gap. Reading: the factorial growth of the coefficients is controlled by the nonperturbative action $\Delta S_{01}$ — the divergence is not a defect but a message.

Derivation AStokes lines from steepest descent — switching a saddle

The goal here is to show that the discontinuous appearance of a saddle-point contribution in an exponential integral happens exactly along a Stokes line, defined by equal phase of the two exponents, so that the abstract switching of equation (2) is grounded in the concrete topology of steepest-descent contours, by watching the descent path through the dominant saddle reconnect to the other as the phase of $z$ turns. Consider $I(z)=\int_C e^{z\,\phi(t)}\,\mathrm{d}t$ with two saddles $t_1,t_2$ where $\phi'(t_j)=0$ .

Step 1. Each saddle lying on the steepest-descent contour contributes a Gaussian factor times its exponential, $I\sim \sum_j \sqrt{2\pi/(-z\phi''(t_j))}\;e^{z\phi(t_j)}$ . Why: expanding $\phi$ to second order gives a Gaussian, and only saddles the contour passes through count — that membership is the whole question.

Step 2. Compare their sizes: $|e^{z\phi(t_1)}/e^{z\phi(t_2)}|=e^{\mathrm{Re}[z(\phi_1-\phi_2)]}$ , so saddle $1$ dominates wherever $\mathrm{Re}[z(\phi_1-\phi_2)]>0$ . Why: dominance is set by the real part of the exponent difference, matching the anti-Stokes condition $\mathrm{Re}=0$ as the boundary of role-reversal.

Step 3. Turn $\arg z$ . The steepest-descent path through $t_1$ deforms continuously until, at a special ray, its constant-phase line aligns with that of $t_2$ ; this is exactly $\mathrm{Im}[z(\phi_1-\phi_2)]=0$ , the Stokes condition. Why: descent paths are curves of constant $\mathrm{Im}[z\phi]$ , so two saddles share a path only when their imaginary exponents match.

Step 4. At that ray the contour topology changes: $t_2$ 's contribution abruptly joins (or leaves) $C$ , and the subdominant exponential in (2) is switched on with its Stokes constant. Why: the leading-order saddle sum is blind to the smooth handover, registering the reconnection as a jump — the abruptness is an artefact of Gaussian-order truncation. This reproduces Olver's steepest-descent treatment and Bender and Orszag's saddle counting.

Go deeper · Derivation B · 1 minThe Airy connection and the value of the Stokes constant

Derivation BThe Airy connection and the value of the Stokes constant

The aim is to track how the recessive Airy expansion of equation (4), valid only in a $\pm\pi/3$ wedge, is forced to acquire a second growing exponential of a definite coefficient as $\arg z$ crosses the Stokes line at $2\pi/3$ , and to pin that coefficient down using an exact symmetry of the Airy function rather than a hand-wave. We start from the two formal WKB solutions, see which one Airy selects, rotate through the rays of Figure 1, and close the loop with the cyclic identity the three rotated Airy functions must satisfy. For $w''=zw$ with $\zeta=\tfrac23 z^{3/2}$ the formal solutions are $w_\pm=z^{-1/4}e^{\pm\zeta}\sum_n(\pm1)^n c_n\zeta^{-n}$ .

Step 1. In the sector $|\arg z|<\pi/3$ boundedness picks the recessive branch, $\mathrm{Ai}(z)\sim \tfrac{1}{2\sqrt\pi}z^{-1/4}e^{-\zeta}\sum_n(-1)^n c_n\zeta^{-n}$ , with only $e^{-\zeta}$ present. Why: this wedge is a single Stokes sector, so the subdominant $e^{+\zeta}$ is beyond all orders and genuinely absent.

Step 2. Rotate outward: at $\arg z=\pi/3$ an anti-Stokes line is met, then at $2\pi/3$ a Stokes line, where the growing solution $e^{+\zeta}$ switches on with a fixed multiplier. Why: equation (3) with $S_{1,2}=\pm\zeta$ places these rays every $60°$ , switching allowed only where $e^{+\zeta}$ is most deeply buried.

Step 3. Impose the exact identity $\mathrm{Ai}(z)+\omega\,\mathrm{Ai}(\omega z)+\omega^2\,\mathrm{Ai}(\omega^2 z)=0$ with $\omega=e^{2\pi i/3}$ , which relates the three rotated sectors. Why: this is a rigorous relation, not an asymptotic one, so demanding consistency of the expansions across the three sectors fixes the switching coefficient uniquely.

Step 4. Match to the negative axis: the two exponentials with the determined multiplier combine into $\sin(\tfrac23 x^{3/2}+\tfrac\pi4)/(\sqrt\pi x^{1/4})$ , fixing the relative phase $\pi/4$ . Why: the specific $\pi/4$ is the observable trace of the Stokes constant and could not come out right if the multiplier were wrong.

Step 5. Read off the constant: the required jump equals the Airy Stokes constant, which is $i$ in the natural normalisation, equivalently encoding the WKB connection factor of $2$ and the $-\pi/4$ phase. Why: the same number will govern every linear turning point, so getting it once from Airy solves the general connection problem. The coefficients $c_0=1,\ c_1=5/72$ and the cyclic identity are standard (DLMF §9.2, Abramowitz and Stegun).

Go deeper · Derivation C · 1 minThe WKB connection formula as Stokes switching

Derivation CThe WKB connection formula as Stokes switching

The goal is to derive the one-dimensional turning-point connection formula as nothing more than the Airy Stokes geometry transplanted to the Schrödinger equation, so that its mysterious factor of $2$ and $-\pi/4$ phase become the Stokes constant of the previous derivation seen in physical variables — by linearising the potential, recognising the Airy equation, and carrying the recessive solution across the Stokes lines into the allowed region. Near a turning point $x=a$ where $E=V(a)$ , the WKB exponents are $S_\pm=\pm\tfrac1\hbar\int_a^x p\,\mathrm{d}x'$ with $p=\sqrt{2m(E-V)}$ .

Step 1. Linearise, $V(x)-E\approx V'(a)(x-a)$ , reducing the Schrödinger equation to Airy form with the turning point as the branch point of $\sqrt{E-V}$ . Why: every simple turning point is locally Airy, so its ray structure is the $120°$ fan of Figure 1.

Step 2. Three Stokes lines leave the turning point at $120°$ ; crossing them switches the growing exponential on in the classically forbidden region. Why: this is the Airy geometry, now oriented so that one Stokes line runs along the real axis separating allowed from forbidden motion.

Step 3. Physical boundedness selects the recessive (decaying) solution deep in the forbidden region, $\tfrac{1}{\sqrt{|p|}}\exp(-\tfrac1\hbar\int_a^x|p|\,\mathrm{d}x')$ — a single exponential. Why: a growing exponential in the barrier would blow up, so the state is uniquely recessive, exactly as Airy is for $z>0$ .

Step 4. Carry that solution across the Stokes lines into the allowed region; the switching generates $\tfrac{2}{\sqrt{|p|}}\cos(\tfrac1\hbar\int_x^a p\,\mathrm{d}x'-\tfrac{\pi}{4})$ , the connection formula. Why: the factor $2$ and $-\pi/4$ are the Airy Stokes constant in disguise — the same $\pi/4$ that appears in equation (5).

Step 5. Close the WKB solution between two turning points; the two $\pi/4$ phases sum to give Bohr–Sommerfeld quantization with its Maslov correction. Why: consistency of the switched solution around a classical orbit is the quantization condition, so Stokes switching is what quantizes energy semiclassically. This matches the connection formulae in Stone and Goldbart and Bender and Orszag.

Limiting cases

Classify any ray by the exponent difference $\Delta S=S_1-S_2$ between two competing terms $e^{S_1}$ and $e^{S_2}$ :

Case	Result	Meaning
Stokes line (switching)	$\mathrm{Im}\,\Delta S=0,\ \mathrm{Re}\,\Delta S\neq0$	one term maximally dominates; the subdominant term is switched on or off here, multiplied by the Stokes constant — maximal dominance hides it, permitting the apparent jump
Anti-Stokes line (equal magnitude)	$\mathrm{Re}\,\Delta S=0,\ \mathrm{Im}\,\Delta S\neq0$	$\vert e^{S_1}\vert=\vert e^{S_2}\vert$ ; dominant and subdominant exchange roles and the sum oscillates — physically where classically allowed and forbidden regions meet in WKB
Generic ray (sector interior)	$\mathrm{Re}\,\Delta S<0$ say	one exponential strictly dominates; the asymptotic series is unique and any subdominant term lies beyond all orders, invisible to the power series
Smoothed / superasymptotic regime	phase width $\sim\vert\Delta S\vert^{-1/2}$	resolving the crossing near optimal truncation, the multiplier rises smoothly as $\tfrac12\,\mathrm{erfc}(-\theta\sqrt{\vert\Delta S\vert/2})$ , recovering the step only as $\vert\Delta S\vert\to\infty$

Common trapThe Stokes phenomenon is a real discontinuity of the function

It is not. The function is analytic everywhere in its domain — only the truncated power-series representation appears to jump when you insist on describing the function with a single exponential across a Stokes line. Two further traps travel with this one. First, the names are not universal: some authors swap "Stokes" and "anti-Stokes", so always fix the ray by the defining condition $\mathrm{Im}(S_1-S_2)=0$ for switching versus $\mathrm{Re}(S_1-S_2)=0$ for equal magnitude, never by the label. Second, do not assume an asymptotic expansion is valid for all $\arg z$ : its coefficients are unambiguous only inside one Stokes sector, and reading them off in the wrong sector — forgetting the switched-on exponential — is the classic way to get an asymptotic answer that is quietly wrong beyond all algebraic orders.

Worked examples

1 · Optimal truncation meets the hidden exponential (fully worked)

Take the divergent series $f(z)\sim\sum_{n=0}^\infty(-1)^n\,n!\,z^{-n-1}$ , the asymptotic expansion of $f(z)=\int_0^\infty e^{-t}/(z+t)\,\mathrm{d}t=e^z E_1(z)$ . Find where the series is smallest at $z=10$ , and show that the least term is the same size as the exponentially small ambiguity $e^{-z}$ that the series cannot represent — the term the Stokes phenomenon switches on across the cut.

Step 1. The magnitude of the $n$ -th term is $|u_n|=n!/z^{n+1}$ , and the ratio $|u_{n+1}/u_n|=(n+1)/z$ crosses $1$ at $n\approx z$ . So the series is smallest — optimally truncated — at $n\approx z=\mathbf{10}$ , the hallmark of a factorially divergent series.

Step 2. Evaluate the least term at $n=10,\ z=10$ : $u_{10}=10!/10^{11}=3\,628\,800/10^{11}=\mathbf{3.63\times10^{-5}}$ . By Stirling, $u_z\approx\sqrt{2\pi/z}\,e^{-z}=\sqrt{2\pi/10}\times e^{-10}=0.7927\times4.54\times10^{-5}=\mathbf{3.60\times10^{-5}}$ , agreeing to under $1\%$ .

Step 3. Compare with the subdominant exponential: $e^{-z}=e^{-10}=\mathbf{4.54\times10^{-5}}$ . The least term ( $3.63\times10^{-5}$ ) and the hidden exponential ( $4.54\times10^{-5}$ ) are the same order — this is no accident. The optimal-truncation error of the dominant series is precisely the scale of the exponential the series omits.

Step 4. Across the branch cut of $E_1$ on the negative real axis, $f(z)$ jumps by $\mathrm{Im}\,f=\pm\pi e^{-z}$ — the Stokes phenomenon for this integral, its magnitude exactly the floor set in Step 3. The series can be trusted only down to $\sim e^{-z}$ , and the Stokes term supplies everything below. Optimal truncation, the least term, and the switched-on exponential are one story told three ways.

2 · The Airy oscillation from switched exponentials

Use the post-anti-Stokes form (5) to estimate $\mathrm{Ai}(-5)$ , and check it against the exact value to see the two switched-on exponentials at work.

Step 1. Compute the WKB phase: $\xi=\tfrac23 x^{3/2}=\tfrac23(5)^{3/2}=\tfrac23(11.180)=\mathbf{7.454}$ , so $\xi+\tfrac\pi4=7.454+0.785=8.239$ rad.

Step 2. The prefactor is $1/(\sqrt\pi\,x^{1/4})=1/(1.7725\times1.4953)=\mathbf{0.3773}$ , and $\sin(8.239)=0.9271$ , giving $\mathrm{Ai}(-5)\approx0.3773\times0.9271=\mathbf{0.3497}$ .

Step 3. The exact value is $\mathrm{Ai}(-5)=\mathbf{0.35076}$ , so the leading two-exponential asymptotic is right to $0.3\%$ . The residual is the first correction $c_1/\xi=(5/72)/7.454=0.0093$ , about $0.9\%$ in the coefficient — exactly the size expected. Both the amplitude $x^{-1/4}$ and the $\pi/4$ phase are direct consequences of the growing exponential having been switched on by $\arg z=2\pi/3$ ; without Stokes switching there would be no oscillation at all, only a lone decaying tail.

Physics of the portrait

The Stokes phenomenon is named for the Victorian mathematician who first stared it in the face — computing an integral by hand, he noticed that a constant of integration would not stay constant.

Origins · 1 minGeorge Gabriel Stokes and how this was discovered

George Gabriel Stokes

1819–1903 · asymptotic analysis and mathematical physics

George Gabriel Stokes, born in Skreen, County Sligo, was one of the towering figures of nineteenth-century British physics, holding the Lucasian Chair at Cambridge for over half a century. His name is stamped across fluid dynamics (the Navier–Stokes equations, Stokes' law of drag), optics (fluorescence, which he named and explained), and vector calculus (Stokes' theorem). The phenomenon that bears his name grew out of a narrow problem: computing the asymptotics of the Airy integral, which describes the intensity of light near a caustic such as a rainbow. Around 1857 Stokes realised that the "arbitrary constant" multiplying the subdominant exponential could not be fixed once and for all — it took one value in one sector of the complex plane and another value elsewhere, changing discontinuously across certain rays. His 1864 paper, "On the discontinuity of arbitrary constants which appear in divergent developments," reported the effect with the caution of a man who knew he was touching something deep that the analysis of his day could not yet justify; he is said to have found it while unable to sleep, checking the Airy integral in the small hours. It took more than a century — through Dingle's systematisation and Berry's error-function smoothing — to see that Stokes' discontinuity is really a smooth, universal, computable feature of asymptotics, now central to resurgence and nonperturbative physics.

History

1837

— Airy introduces his integral to describe the intensity of light near a caustic

1857

— Stokes finds the constant multiplying a subdominant exponential jumps as arg z sweeps

1864

— Stokes publishes "On the discontinuity of arbitrary constants in divergent developments"

1926

— WKB connection formulae recast the same switching at a quantum turning point

1989

— Berry shows Stokes switching is error-function smooth, not a true jump

Going further

Texts. Stone and Goldbart, Mathematics for Physics (Cambridge, 2009), chapter on asymptotics and the Stokes phenomenon · Bender and Orszag, Advanced Mathematical Methods for Scientists and Engineers (Springer, 1999), asymptotic matching and turning points · Arfken, Weber and Harris, Mathematical Methods for Physicists, 7th ed., asymptotic series and Airy functions · R. B. Dingle, Asymptotic Expansions: Their Derivation and Interpretation (Academic Press, 1973) · M. V. Berry, "Uniform asymptotic smoothing of Stokes's discontinuities," Proc. R. Soc. Lond. A 422 (1989) 7–21 · Olver, Asymptotics and Special Functions (Academic Press, 1974). The Airy coefficients $c_0=1,\ c_1=5/72$ , the cyclic identity, and the oscillatory form with phase $\pi/4$ are standard results (DLMF §9.2, Abramowitz and Stegun), as are the WKB factor of $2$ and $-\pi/4$ from the linear-turning-point Airy reduction. The optimal-truncation example gives, at $z=10$ : least term $10!/10^{11}=3.63\times10^{-5}$ , Stirling estimate $3.60\times10^{-5}$ , and $e^{-10}=4.54\times10^{-5}$ . The Airy check yields $\mathrm{Ai}(-5)$ leading $0.3497$ versus exact $0.35076$ . Berry's error-function smoothing is from Proc. R. Soc. A 422 (1989); the resurgence large-order relation is the standard Bender–Wu / Écalle-type formula, with constants to be normalised per problem.

Prerequisites. Asymptotic Series and Poincaré Expansions · Method of Steepest Descent and Saddle Points · Order Symbols and Gauge Functions · WKB Approximation. Leads to. Connection Formulae and Turning Points · Optimal Truncation and Superasymptotics · Borel Summation and Resurgence · Bohr–Sommerfeld Quantization.

Ask the tutor: Pick any ray in the complex plane and I will walk you through which exponentials are alive there, and why the Airy Stokes constant comes out to i.

What is the difference between a Stokes line and an anti-Stokes line, in one sentence each?

A Stokes line is a ray where one exponential is maximally larger than the other, $\mathrm{Im}(S_1-S_2)=0$ ; there the small subdominant exponential is most deeply buried and can be switched on or off without disturbing the power series. An anti-Stokes line is where the two have equal magnitude, $\mathrm{Re}(S_1-S_2)=0$ ; they trade the roles of dominant and recessive and their sum oscillates. The two families are perpendicular. Because some authors swap the names, always identify the ray by its defining condition rather than the label, or you will invert the whole picture.

If the function is analytic, in what sense does anything actually "jump"?

Nothing about the function jumps — it is smooth and single-valued throughout its domain. What jumps is your bookkeeping. When you try to describe the function in a new sector using the single dominant exponential and its power series, that description silently omits a second, exponentially small exponential. The coefficient you must attach to that omitted term is zero in one sector and nonzero in the next, and it changes across the Stokes line. So the "discontinuity" is a property of the truncated asymptotic representation, not of the function. Berry's smoothing makes this vivid: resolve the crossing on the exponentially small scale and even the representation changes smoothly, as an error function.

Why does the Airy function need two exponentials off the positive real axis but only one on it?

On the positive real axis the Airy function is the recessive solution of $w''=zw$ : of the two WKB solutions $e^{\pm\zeta}$ , boundedness selects only the decaying $e^{-\zeta}$ , and the growing solution is genuinely absent, lying beyond all algebraic orders. But rotating $\arg z$ crosses a Stokes line at $2\pi/3$ where the growing exponential switches on with a definite coefficient (the Stokes constant $i$ ), so by the negative axis both exponentials are present with equal magnitude and combine into $\sin(\tfrac23 x^{3/2}+\tfrac\pi4)$ . The single-exponential description was never wrong on the positive axis; it simply stopped being complete once you left that sector.

How is the Stokes phenomenon the same thing as a WKB connection formula?

Every simple turning point of a one-dimensional Schrödinger problem is, once you linearise the potential, an Airy equation, so the $120°$ fan of Stokes lines around it is the Airy fan. In the forbidden region boundedness picks the recessive (decaying) exponential — one term. Following that solution across the Stokes lines into the allowed region switches on the second exponential, and the two combine into a cosine with a factor of $2$ and a $-\pi/4$ phase: precisely the connection formula, the Airy Stokes constant in physical clothing. Close the loop between two turning points and the accumulated phases give Bohr–Sommerfeld quantization.

Why does the divergent perturbation series "know" about nonperturbative physics?

Because the late coefficients of an asymptotic series are controlled by the nearest singularity in the Borel plane, which sits at the action of the nonperturbative sector. Concretely, the perturbative coefficients grow like $a_n\sim S\,(n-1)!/(\Delta S)^n$ , where $\Delta S$ is the instanton action and $S$ the Stokes constant. So by studying how fast the perturbation series diverges you can read off the action, and even the fluctuation prefactor, of an instanton you never explicitly computed. This is the central insight of resurgence: divergence is not noise but structured information, and the Stokes constants are the numbers that connect the two worlds.

What does "optimal truncation" have to do with all of this?

A factorially divergent series has a smallest term, reached at an order $N$ roughly equal to the exponent difference $|\Delta S|$ . Truncating there — "optimal truncation" — gives the best accuracy the series alone can offer, with an error of order $e^{-|\Delta S|}$ , exactly the size of the subdominant exponential it omits. This is why the Stokes phenomenon lives at the exponentially small scale: the switched-on term is the same size as the irreducible error of the dominant series. Push beyond optimal truncation with Borel summation or Dingle's late-term methods and the switching resolves smoothly, entering the superasymptotic regime where Berry's error function appears.