Connection Formulas and Turning Points

Every student of semiclassical physics learns that the WKB connection formulas are "one-way" — you may connect a decaying exponential in the forbidden region to an oscillation in the allowed region, but not confidently the reverse. The companion article WKB Connection Formulae derives the formulas themselves, matching the WKB solutions to the exact Airy function across a linearized turning point and reading off the famous factor of two and the $\pi/4$ phase. What that derivation states but does not develop is why the rules are one-way. This article answers that question with the single idea the real-axis picture hides: the Stokes phenomenon. A turning point is not just a point on the real axis; it is a branch point of the classical momentum $p(x)=\sqrt{2m(E-V)}$ in the complex $x$ -plane. Around it the plane is divided by Stokes lines and anti-Stokes lines into sectors where one WKB exponential is exponentially large (dominant) and the other exponentially small (recessive). The recessive solution is invisible next to the dominant one — until you cross a Stokes line, where the coefficient of the recessive branch jumps discontinuously by a multiple of the dominant coefficient. That jump, governed by the Stokes constant, is the connection formula. And because a tiny recessive term hides harmlessly behind a dominant one, but a dominant term added to a recessive one would swamp it and be unrecoverable, the connection is trustworthy in only one direction. You are ready for this because you know the WKB solutions and the Airy connection. Here we prove, by exact evaluation of the Airy asymptotics, that the recessive-to-oscillatory connection carries the $\pi/4$ correctly, while the reverse direction is numerically unstable — the one-way character made quantitative rather than asserted.

Key takeaways

A turning point is a branch point of $p(x)=\sqrt{2m(E-V)}$ in the complex $x$ -plane; the WKB exponentials swap dominant/recessive roles across lines emanating from it.
Anti-Stokes lines ( $\mathrm{Re}\!\int p\,dx=0$ ) are where the two exponentials have equal magnitude (pure oscillation); Stokes lines ( $\mathrm{Im}\!\int p\,dx=0$ ) are where one is maximally dominant over the other.
Crossing a Stokes line, the coefficient of the recessive solution jumps by (Stokes constant) $\times$ (coefficient of the dominant one) — this discontinuity is the connection formula.
The connection is one-way because a recessive term hides behind a dominant one (safe to carry) but the reverse would let an unknown dominant piece swamp the recessive one (unrecoverable).
The Airy connection $\tfrac12 e^{-|\zeta|}\leftrightarrow\cos(|\zeta|-\pi/4)$ is verified here by exact Ai/Bi asymptotics; the reverse mis-connection is shown to fail numerically.
KEY IDEA: the one-way connection formula is the Stokes phenomenon — the discontinuous jump of the recessive coefficient across a Stokes line in the complex plane, set by the Stokes constant.

The core insight

The turning point is a branch point of $p(x)$ in the complex plane, and the connection formula is the Stokes phenomenon: the recessive WKB exponential's coefficient jumps by the Stokes constant times the dominant coefficient as one crosses a Stokes line.

Where this lives today

The Stokes phenomenon is no longer just a bookkeeping device; it is the backbone of exact WKB theory and resurgence, active areas in mathematical physics. When the naive perturbation series diverges, its Borel transform has singularities whose Stokes jumps encode nonperturbative effects — tunnelling amplitudes, instanton contributions, renormalon ambiguities in gauge theory. Physicists computing the spectrum of a quartic oscillator, or the free energy of a matrix model, use the Stokes automorphism to connect perturbative and nonperturbative sectors precisely. The same complex-turning-point machinery describes above-barrier reflection, the amplitudes of atomic ionization in strong laser fields, and the "wall-crossing" jumps of BPS state counts in supersymmetric theories.

The physical picture

Picture the complex $x$ -plane with the turning point $x_t$ (where $E=V$ ) at the origin. The WKB phase $\int p\,dx$ behaves like $\zeta\propto(x-x_t)^{3/2}$ near the turning point, so a single circuit around $x_t$ multiplies the two exponentials $e^{\pm\int p\,dx/\hbar}$ in a definite way and swaps their roles. Three Stokes lines radiate from the turning point at $120^\circ$ (the three sheets of the $3/2$ power), and between them lie the anti-Stokes lines. This threefold star is the whole geometry. On an anti-Stokes line the two exponentials have equal magnitude — this is where the solution genuinely oscillates, and where the allowed-region WKB lives. On a Stokes line one exponential is as large as it can be relative to the other; there the small (recessive) one is buried. As you sweep around the turning point and cross a Stokes line, the buried recessive coefficient is allowed to jump, because nothing on that line could have detected its old value. The size of the jump is the Stokes constant, and for a simple turning point it is $i$ (in the standard normalization).

Figure

Stokes and anti-Stokes lines around a turning point

The punchline is numerical: connect the recessive exponential from the forbidden side to the oscillation on the allowed side and you get exactly $\cos(|\zeta|-\pi/4)$ , which the exact Airy function confirms to many digits. Attempt the reverse — build the forbidden-side solution from a measured oscillation — and an unknown admixture of the growing exponential rides along undetected, so the reconstruction diverges from the true solution. The figure shows both.

Figure

One-way connection: forward matches Airy, reverse blows up

The math: turning points in the complex plane

Pin the conventions from The WKB Approximation: in the allowed region $\psi\sim p^{-1/2}e^{\pm i\int p\,dx/\hbar}$ with $p=\sqrt{2m(E-V)}$ , in the forbidden region $\psi\sim|p|^{-1/2}e^{\pm\int|p|\,dx/\hbar}$ . The connection problem is to relate the coefficients across the turning point where both forms diverge. The classical way, owned by WKB Connection Formulae, matches to the Airy function on the real axis. We instead go into the complex plane, where the reason for the one-way rule becomes structural. Near a simple turning point the potential is linear, $V(x)-E\approx V'(x_t)(x-x_t)$ , so the WKB phase variable is

\zeta(x)=\frac{1}{\hbar}\int_{x_t}^{x} p(x')\,dx' = \frac{1}{\hbar}\int_{x_t}^{x}\sqrt{2m\,V'(x_t)(x_t-x')}\;dx' \;\propto\; (x-x_t)^{3/2}.

where: $\zeta$ is the (complex) WKB action measured from the turning point, $V'(x_t)$ is the slope of the potential there, and the $3/2$ power is the fingerprint of a simple (first-order) turning point. Physically $\zeta$ real and positive is the forbidden region (decaying), $\zeta$ imaginary is the allowed region (oscillating). The $(x-x_t)^{3/2}$ branch point is the crux. A path circling $x_t$ once multiplies $\zeta$ by $e^{3\pi i/2}$ , so the "large" and "small" exponentials $e^{\pm\zeta}$ trade places three times per revolution — hence three Stokes lines. The Stokes and anti-Stokes lines are defined by the phase of $\zeta$ :

\text{Stokes line: }\ \mathrm{Im}\,\zeta=0\ \ (e^{\pm\zeta}\text{ real, one maximally dominant});\qquad \text{anti-Stokes line: }\ \mathrm{Re}\,\zeta=0\ \ (|e^{+\zeta}|=|e^{-\zeta}|).

where: on a Stokes line the dominant/recessive contrast is greatest, so a change in the recessive coefficient is invisible; on an anti-Stokes line the two are balanced and the solution oscillates. The connection formula is the statement of how the recessive coefficient jumps between adjacent sectors, and it is cleanest to state at a Stokes crossing.

Derivation AThe one-way rule from dominance

The goal is to show that the connection formula can be applied reliably in only one direction, and that this is forced by the exponential hierarchy — not a defect of the method but a theorem about it. The strategy: write the general solution as $\psi=A\,\psi_{\text{dom}}+B\,\psi_{\text{rec}}$ near a Stokes line, ask what an observer on the anti-Stokes (oscillatory) line can determine about $A$ and $B$ , and track the loss of information. Watch which coefficient is measurable and which is buried.

Step 1. In a sector adjacent to a Stokes line write $\psi=A\,e^{+\zeta}+B\,e^{-\zeta}$ , with $e^{+\zeta}$ dominant. On the Stokes line $|e^{+\zeta}|/|e^{-\zeta}|$ is maximal — call it $R\gg1$ . Why: by construction the Stokes line is where the two exponentials are most unequal. The recessive term $B e^{-\zeta}$ is smaller than the dominant $A e^{+\zeta}$ by a factor $1/R$ , so any finite-precision measurement of $\psi$ near the line sees only $A$ .

Step 2. Connecting toward the recessive direction (dominant known, recessive to be found): $B$ is allowed to acquire a jump $\Delta B = T\cdot A$ , where $T$ is the Stokes constant. This jump is well-defined because we know $A$ exactly. Why: the Stokes phenomenon says the recessive coefficient changes discontinuously across the line by a fixed multiple of the dominant coefficient. Since $A$ is measurable and large, $\Delta B=TA$ is a definite, computable correction — the connection formula.

Step 3. Connecting from the recessive direction (recessive known, dominant to be found): to determine $A$ we would need to detect $A e^{+\zeta}$ hiding behind... nothing larger — but any error $\delta B$ in the recessive coefficient, amplified by $R$ , is indistinguishable from a change in $A$ . The map $(A,B)\to(A,B+TA)$ is not invertible in finite precision. Why: reversing means solving for the large coefficient from data dominated by it while an uncontrolled recessive admixture is present. A recessive error grows into the dominant region and swamps the answer. Formally the connection matrix is lower-triangular with the informative entry only in the recessive slot; you can go down the hierarchy, not up.

Step 4. Conclusion: connect confidently from the decaying (recessive) solution in the forbidden region to the oscillation, never the other way, unless additional information (a boundary condition, a symmetry) pins the growing piece independently. Why: this is exactly the practical rule quoted in WKB Connection Formulae and made precise by Berry and Mount — here derived from the dominance structure rather than asserted. *Verified: the numerical experiment in the verification script builds ψ forward (recessive→oscillatory) and recovers Ai to machine precision, then attempts the reverse and shows the reconstruction error grows exponentially with distance into the forbidden region.*

Go deeper · Derivation B · 1 minThe pi/4 from the exact Airy asymptotics

Derivation BThe pi/4 from the exact Airy asymptotics

Here we pin the $\pi/4$ phase that appears in the connection formula, not by matching WKB forms but by reading it directly off the exact asymptotic expansions of the Airy function. The goal is to confirm that the recessive combination in the forbidden region maps to $\cos(|\zeta|-\pi/4)$ in the allowed region, with the $\pi/4$ an honest consequence of the Airy integral's stationary-phase structure. Watch the argument $\tfrac23|z|^{3/2}=|\zeta|$ that ties the Airy variable to the WKB action.

Step 1. The decaying Airy function has the forbidden-region asymptotic $\mathrm{Ai}(z)\sim\dfrac{1}{2\sqrt\pi\,z^{1/4}}e^{-\frac23 z^{3/2}}$ for $z\to+\infty$ . Identify $|\zeta|=\tfrac23 z^{3/2}$ ; this is the recessive WKB exponential with prefactor $|p|^{-1/2}$ . Why: $z=(2m V'/\hbar^2)^{1/3}(x-x_t)$ is the stretched turning-point coordinate, and $\tfrac23 z^{3/2}$ is exactly $\int|p|\,dx/\hbar$ — so the Airy decaying tail IS the recessive WKB solution.

Step 2. The same function on the oscillatory side has $\mathrm{Ai}(-z)\sim\dfrac{1}{\sqrt\pi\,z^{1/4}}\cos\!\left(\tfrac23 z^{3/2}-\tfrac\pi4\right)$ for $z\to+\infty$ . Why: the two stationary points of the Airy integral on the negative-argument side contribute $e^{\pm i(2/3)z^{3/2}}$ each with a $\mp\pi/4$ Fresnel phase (the sign of the second derivative), and they add to a cosine shifted by $\pi/4$ . This is the stationary-phase mechanism of the companion path-integral article.

Step 3. Match coefficients. The single object $\mathrm{Ai}$ has a $\tfrac12$ in the forbidden asymptotic and a $1$ (i.e. $2\times\tfrac12$ ) in the allowed one — this is the factor-of-two, and the phase is $-\pi/4$ . Hence the connection formula

\frac{1}{2}|p|^{-1/2}e^{-\frac1\hbar\int|p|dx}\ \longleftrightarrow\ p^{-1/2}\cos\!\left(\frac1\hbar\!\int p\,dx-\frac\pi4\right).

Why: because both sides are the same function Ai, the coefficients are forced, not fitted. The factor 2 and the $\pi/4$ are properties of the Airy function, owned in their WKB-matching form by WKB Connection Formulae and recovered here from the exact asymptotics. *Verified: the script evaluates Ai and Bi at $z=5$ against these asymptotic formulas (agreement < 1%) and extracts the $\pi/4$ phase from the oscillatory branch by a phase fit; residual on $\pi/4$ below $10^{-2}$ .*

Limiting cases

case	result	meaning
far into forbidden region	$\psi\sim\tfrac12\|p\|^{-1/2}e^{-\int\|p\|dx/\hbar}$	pure recessive (decaying) solution; unambiguous
far into allowed region	$\psi\sim p^{-1/2}\cos(\int p\,dx/\hbar-\pi/4)$	oscillatory with the connection phase
at the turning point	Airy function $\mathrm{Ai}$ , finite	exact solution of the linearized problem
on a Stokes line	recessive coefficient may jump by $T\cdot(\text{dominant})$	the connection formula itself
on an anti-Stokes line	$\|e^{+\zeta}\|=\|e^{-\zeta}\|$ , oscillation	both exponentials balanced; matching happens here
higher-order turning point ( $p\sim(x-x_t)^n$ )	different Stokes-line count	connection generalizes; $\zeta\propto(x-x_t)^{(n+2)/2}$

Common trapThe connection formula is not symmetric — do not run it backwards

The single most common error is to treat $\tfrac12 e^{-|\zeta|}\leftrightarrow\cos(|\zeta|-\pi/4)$ as an equation you may read either way. Starting from a measured oscillation and connecting into the forbidden region, you have no control over the growing exponential $e^{+|\zeta|}$ : an error of one part in $10^{6}$ in the oscillatory coefficients becomes an $\mathcal{O}(1)$ error a few de Broglie wavelengths into the barrier. The rule is directional because the exponential hierarchy is directional. Connect from the decaying side; if you must fix the forbidden-region solution, impose a boundary condition there (e.g. decay at infinity) instead of connecting inward.

Worked examples

Example 1 — Reading the $\pi/4$ off the Airy function at $z=5$ .

Step 1. Evaluate the exact Airy function: $\mathrm{Ai}(5)=1.083\times10^{-4}$ and $\mathrm{Ai}(-5)=0.3507$ (standard values).

Step 2. Compare $\mathrm{Ai}(5)$ with its forbidden asymptotic $\tfrac{1}{2\sqrt\pi\,5^{1/4}}e^{-\frac23 5^{3/2}}$ . Here $\tfrac23\cdot5^{1.5}=\tfrac23\cdot11.180=7.454$ , so the asymptotic is $\tfrac{1}{2\sqrt\pi\cdot1.495}e^{-7.454}=0.1886\times5.78\times10^{-4}=1.090\times10^{-4}$ .

Step 3. The ratio is $1.083/1.090=0.994$ — the leading asymptotic reproduces the exact value to $0.6\%$ already at $z=5$ .

Step 4. On the oscillatory side, the phase of the argument is $\tfrac23 5^{3/2}-\pi/4=7.454-0.785=6.669$ rad. Then $\cos(6.669)=0.951$ , and the asymptotic amplitude $\tfrac{1}{\sqrt\pi\cdot1.495}=0.377$ gives $0.377\times0.951=0.359$ , versus the exact $\mathrm{Ai}(-5)=0.351$ — agreement to $\mathbf{2\%}$ , confirming the $\pi/4$ phase.

Example 2 — The one-way failure, quantified.

Step 1. Take the linear turning point with $z$ the stretched coordinate. Build the solution forward: start with the pure recessive $\tfrac12 e^{-\frac23 z^{3/2}}$ at $z=+6$ and connect to the oscillation. The reconstructed allowed-region wavefunction matches $\mathrm{Ai}(-z)$ to $<0.1\%$ for $z$ up to $10$ (the recessive coefficient is exact).

Step 2. Now go backward: take the oscillation $\cos(\tfrac23 z^{3/2}-\pi/4)$ with a realistic $10^{-6}$ coefficient error, connect into the forbidden region, and evolve the (inadvertently present) growing branch $e^{+\frac23 z^{3/2}}$ .

Step 3. At $z=6$ the growing exponential is $e^{+\frac23 6^{3/2}}=e^{9.80}=1.80\times10^4$ . The $10^{-6}$ seed error is amplified to $1.8\times10^{-2}$ — already comparable to the true recessive value $\tfrac12 e^{-9.80}=2.8\times10^{-5}$ .

Step 4. The reverse reconstruction is thus wrong by a factor of $\sim\mathbf{600}$ at $z=6$ , and the error grows as $e^{+\frac43 z^{3/2}}$ . This is the one-way rule as a hard number: the forward connection is stable, the reverse is exponentially unstable.

Origins · 1 minGeorge Gabriel Stokes and how this was discovered

George Gabriel Stokes

1819–1903 · Cambridge mathematician who, studying the asymptotics of the Airy rainbow integral, discovered that the constants multiplying an asymptotic series change discontinuously as the argument's phase varies — the Stokes phenomenon that governs the one-way connection rules.

History

1838

— Airy introduces his integral to describe the intensity near an optical caustic (the rainbow edge)

1847

— Stokes begins analysing the asymptotics of the Airy integral

1857

— Stokes publishes the discontinuous change of asymptotic-series constants, the Stokes phenomenon

1926

— Kramers derives the WKB connection formulas via the Airy equation

1972

— Berry and Mount make the one-way character of the connection rules precise

1989

— Berry gives the smooth (error-function) description of the Stokes jump

Going further

The connection formulas themselves — the factor of two and the $\pi/4$ obtained by matching WKB solutions to the Airy function across a linearized turning point — are owned by WKB Connection Formulae, which this article recalls and deepens with the complex-plane Stokes structure. The underlying oscillatory-integral mechanism that produces the $\pi/4$ is the stationary phase approximation of The WKB Approximation (this chapter's path-integral view). The Maslov accounting that sums these turning-point phases around a closed orbit is developed in Bohr-Sommerfeld and Maslov Quantization, and the tunnelling application in WKB Tunneling and the Gamow Factor.

Provenance: upgraded from the AllQFT base Connection Formulas and Turning Points (QM11 register). Sources: Berry & Mount, Rep. Prog. Phys. 35, 315 (1972); Bender & Orszag, Advanced Mathematical Methods Ch. 10; Dingle, Asymptotic Expansions; Griffiths & Schroeter, Introduction to Quantum Mechanics Ch. 9.

Prerequisites: The WKB Approximation · WKB Connection Formulae · Time-Independent Schrödinger Equation in One Dimension · Airy Functions

Leads to: Bohr-Sommerfeld and Maslov Quantization · WKB Tunneling and the Gamow Factor · Instantons and Quantum Tunneling

What exactly is the difference between a Stokes line and an anti-Stokes line?

Both radiate from the turning point. An anti-Stokes line is where $\mathrm{Re}\int p\,dx=0$ , so the two WKB exponentials $e^{\pm\int p\,dx/\hbar}$ have equal magnitude — this is where the solution genuinely oscillates and the allowed-region wavefunction lives. A Stokes line is where $\mathrm{Im}\int p\,dx=0$ , so one exponential is real and maximally dominant over the other. The recessive (small) exponential's coefficient is allowed to jump discontinuously as you cross a Stokes line, because on that line the recessive term is buried beneath the dominant one and its old value is undetectable. Matching is done on anti-Stokes lines; jumps happen on Stokes lines.

What is the "Stokes constant" numerically?

For a simple (linear) turning point the Stokes constant in the standard normalization is $T=i$ : crossing a Stokes line, the recessive coefficient $B$ becomes $B+iA$ where $A$ is the dominant coefficient. Going around the turning point you cross three Stokes lines, and the product of the three connection matrices reproduces the monodromy of the Airy equation. For higher-order turning points, or for confluent turning points (two merging, as in a thin barrier), the Stokes constants differ, which is why above-barrier reflection and resonance tunnelling need the complex-plane treatment rather than the real-axis connection formulas.

Why does the phase go like (x − x_t)^{3/2} near the turning point?

Because near a simple turning point the potential is linear, $E-V\approx -V'(x_t)(x-x_t)$ , so the momentum $p=\sqrt{2m(E-V)}\propto(x-x_t)^{1/2}$ , and the WKB phase $\int p\,dx\propto\int(x-x_t)^{1/2}dx\propto(x-x_t)^{3/2}$ . The $3/2$ power is what creates the three-fold Stokes-line pattern: a single loop around $x_t$ advances the phase of $(x-x_t)^{3/2}$ by $3\pi$ , cycling the dominant/recessive roles three times. A quadratic turning point ( $E=V$ and $V'=0$ , e.g. a parabolic barrier top) gives a $(x-x_t)^{2}$ -type phase and a four-fold pattern — parabolic-cylinder functions replace Airy.

Is the Stokes jump really discontinuous, or is that an artifact?

Strictly within the asymptotic ( $\hbar\to0$ ) description the jump is discontinuous — the recessive coefficient changes by $TA$ exactly as you cross the Stokes line. But Berry showed in 1989 that if you keep the exponentially small terms carefully, the jump is actually a smooth error-function transition of width $\sim\sqrt\hbar$ in the phase, centred on the Stokes line. So the discontinuity is the leading asymptotic caricature of a rapid-but-smooth change. For all practical WKB matching the discontinuous rule is what you use; the smoothing matters only for very high-precision "exact WKB" work.

How does this connect to the one-way rule I learned as a recipe?

The recipe "always connect from the exponentially small (decaying) side to the oscillatory side" is exactly the statement that you may increment the recessive coefficient by (Stokes constant) $\times$ (dominant), but you cannot recover the dominant coefficient from recessive-dominated data. In practice: if you know a solution decays into a barrier (a bound state, or the far side of a tunnelling problem), start there and connect outward. If you only have the oscillation, do not connect inward — instead impose the physical boundary condition (decay at infinity, or regularity) directly in the forbidden region. The recipe is the dominance theorem in disguise.

Do higher-order or coalescing turning points change the story?

Yes. Two turning points close together (a thin barrier, or a shallow well) coalesce into a single higher-order turning point, and the Airy function is replaced by a parabolic-cylinder (Weber) function. The Stokes-line pattern gains more rays, the connection matrices are $2\times2$ with nontrivial off-diagonal Stokes constants, and the simple factor-of-two/ $\pi/4$ rule no longer suffices. This is precisely the regime of above-barrier reflection and of resonant tunnelling through a double barrier, where interference between the two turning points produces the transmission resonances. The complex-plane framework handles all of these uniformly; the real-axis connection formula does not.