Watson-Sommerfeld and Integral Representations

A special function usually arrives in a costume that fits only part of its body. A power series knows the function perfectly near one point and goes blind past its radius of convergence; a recurrence walks it along integers but cannot say what happens between them; a differential equation pins it down but hides its global shape. An integral representation is the costume that fits everywhere. Write the function as $f(z)=\int_C K(z,t)\,\mathrm{d}t$ with a single-valued kernel $K$ and a contour $C$ chosen so the integral converges far beyond the original domain. Differentiating under the sign proves $f$ still solves its defining equation; then complex analysis — deforming $C$ , collecting residues, sliding onto saddle points — delivers analytic continuation, connection formulae, and complete asymptotic expansions, all from one expression. Two templates run the field. The Hankel contour integral wraps a path around the negative real axis to make $1/\Gamma(z)$ an entire function, turning the poles of $\Gamma$ into the zeros of its reciprocal. The Mellin–Barnes integral writes a function as an inverse Mellin transform of a ratio of Gamma functions; closing its vertical contour left or right sums two different residue ladders, giving convergent series or divergent asymptotic ones on demand. The Bessel family adds the Poisson, Schläfli, and Sommerfeld integrals, whose saddle-point analysis yields the universal $\sqrt{2/\pi x}\,\cos(x-\nu\pi/2-\pi/4)$ tail. The capstone is the Watson–Sommerfeld transformation, which continues the angular-momentum label $\ell$ into a complex variable $\lambda$ and exposes the Regge poles that organise high-energy scattering.

Key takeaways

A series with finite radius of convergence can be re-expressed as a contour integral $\int_C K\,\mathrm{d}t$ with a single-valued, globally defined integrand — instantly delivering analytic continuation, connection formulae, and asymptotics.
The Hankel integral for $1/\Gamma(z)$ and the Mellin–Barnes integral are the two master templates: the first cures the poles of $\Gamma$ , the second writes a function as an inverse Mellin transform of products of Gamma functions.
Bessel integral representations (Poisson, Schläfli, Sommerfeld) turn defining ODEs into oscillatory integrals whose steepest-descent analysis gives $J_\nu(x)\sim\sqrt{2/\pi x}\,\cos(x-\nu\pi/2-\pi/4)$ directly.
The Watson–Sommerfeld transformation rewrites a partial-wave sum in complex angular momentum $\lambda$ , exposing Regge poles invisible term-by-term; these are measured — the ρ-meson family lines up on $\alpha(t)\approx 0.50+0.84\,t$ GeV, a straight line in the spin-versus-mass² plane.
Key idea: Replace a domain-limited series or recurrence by an integral over a cleverly chosen contour; deforming the contour and collecting residues simultaneously yields analytic continuation, connection formulae, and asymptotics, because the integrand is single-valued where the original series was not.

The core insight

Replace a domain-limited series or recurrence by an integral $\int_C K\,\mathrm{d}t$ over a cleverly chosen contour: deforming $C$ and collecting residues simultaneously yields analytic continuation, connection formulae, and asymptotics, because the integrand is single-valued where the original series was not.

Where this lives today

The Watson–Sommerfeld transform is not a museum piece — it is the language high-energy physics still uses for total cross-sections. When two protons collide at the LHC, the slow forward growth of the total cross-section is read as exchange of the Pomeron, the leading Regge trajectory with intercept $\alpha_0\approx1.08$ , and the Donnachie–Landshoff fit $\sigma_{\rm tot}\sim s^{\alpha_0-1}$ describes data across three decades of energy with a handful of parameters. The same machinery reappears in the AdS/CFT correspondence, where a graviton Regge trajectory controls deep-inelastic scattering at small Bjorken $x$ , and in the "BFKL Pomeron" of perturbative QCD. Mellin–Barnes representations are the standard weapon for multi-loop Feynman integrals, and Watson's original problem lives on in the "creeping-wave" fields engineers and seismologists compute for diffraction around the Earth and around aircraft. Every one is the same act: continue a discrete index, deform a contour, keep the poles that matter.

The physical picture

A power series is a microscope: it resolves a function flawlessly near one point and sees nothing beyond its circle of convergence, where the nearest singularity draws a hard wall. An integral representation is a satellite view — by integrating a single-valued kernel along a contour, one expression captures the entire analytic landscape, every pole and branch cut and saddle, in a single sweep. The residue theorem is the translator: read the integral by residues at integer poles and the discrete series reappears; read it by saddle points or a deformed contour and the asymptotic regime emerges. Both readings are exact, and both equal the one contour. The first figure shows the arena where this trade happens most cleanly — the complex $s$ -plane of a Mellin–Barnes integral, with two ladders of poles and a vertical contour between them. Which way you close the contour decides which physics you extract.

Figure

In the complex s-plane the Mellin–Barnes contour runs up the line Re s = c between two pole ladders: Γ(s) has poles at s = 0, −1, −2, … on the left and Γ(a − s) has poles at s = a, a+1, … on the right. Closing left for x ≤ 1 sums the convergent small-x series Γ(a)/(1+x)ᵃ; closing right gives the large-x asymptotic expansion

The same machinery, applied to a discrete sum, produces the deepest result in the field. The Watson–Sommerfeld transformation promotes the integer $\ell$ to a continuous complex variable $\lambda$ ; the original sum becomes the residue series of $\pi/\sin\pi\lambda$ at the integers, and the new physics lives off the lattice, at the Regge poles of the continued amplitude. This is not an abstraction — it is measured. Plotting the spin $J$ of a meson family against its mass-squared $m^2$ lines the states up on a straight Regge trajectory, as the second figure shows for the ρ family.

Figure

The Chew–Frautschi plot for the ρ-meson family: spin J against mass-squared m² places ρ(770), a₂(1320), ρ₃(1690) and a₄(2040) on one straight Regge trajectory α(t) ≈ 0.50 + 0.84 t, a slope of about 0.84 GeV⁻² and intercept near 0.50 — the linear law that the Watson–Sommerfeld continuation of angular momentum predicts

Choosing the contour is the whole art: it must thread between the poles kept on one side and those on the other, and close where the integrand decays. The physics dictates which poles matter; the mathematics, where the contour may legally go.

The core equations

The first master representation packages the reciprocal Gamma function. Starting from Euler's integral $\Gamma(z)=\int_0^\infty t^{z-1}e^{-t}\,\mathrm{d}t$ , valid only for $\Re z>0$ , one wraps the path around the branch cut on the negative real axis; the result is an entire function, analytic for every finite $z$ , with the poles of $\Gamma$ reappearing as its simple zeros.

\frac{1}{\Gamma(z)}=\frac{1}{2\pi i}\int_{C}e^{t}\,t^{-z}\,\mathrm{d}t \qquad (1)

where: $C$ — Hankel contour, running from $-\infty$ below the cut, around the origin, and back to $-\infty$ above it; $t^{-z}$ — principal branch with its cut along the negative real axis; $z$ — complex argument, with the integral entire in $z$ . Reading it: the integral converges for all finite $z$ because $e^{t}$ kills the integrand at both ends, so no series wall survives. The second template is the workhorse of the subject. A Mellin–Barnes integral writes a function as an inverse Mellin transform of a ratio of Gamma functions; almost every classical special function is a special case, making it the lingua franca connecting analytic structure to asymptotics.

f(x)=\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{\prod_{j}\Gamma(a_j+s)\,\prod_{k}\Gamma(b_k-s)}{\prod_{l}\Gamma(c_l+s)\,\prod_{m}\Gamma(d_m-s)}\,x^{-s}\,\mathrm{d}s \qquad (2)

where: $c$ — real abscissa chosen so the vertical line separates the increasing pole sequence (from $a_j+s$ ) from the decreasing one (from $b_k-s$ ); $x^{-s}=e^{-s\ln x}$ — the convergence controller, favouring closure left when $|x|<1$ and right when $|x|>1$ ; the Gamma products supply the poles whose residues become series terms. The closure direction selects which pole ladder to sum. For Bessel functions of the first kind, the Poisson integral makes the half-integer orders elementary and ties the normalisation to the Beta and Gamma functions.

J_\nu(x)=\frac{(x/2)^\nu}{\sqrt{\pi}\,\Gamma(\nu+\tfrac12)}\int_{-1}^{1}(1-t^2)^{\nu-1/2}\cos(xt)\,\mathrm{d}t \qquad (3)

where: $\nu$ — order, requiring $\Re\nu>-\tfrac12$ for the endpoints to converge; $x$ — argument, with the integral entire in $x$ ; $(1-t^2)^{\nu-1/2}$ — the weight whose normalisation is a Beta integral. At $\nu=\tfrac12$ the weight becomes unity and the integral collapses to $2\sin x/x$ , the spherical Bessel function. The Schläfli loop integral removes the integer restriction, defining $J_\nu$ for any complex order from the generating-function kernel.

J_\nu(x)=\frac{1}{2\pi i}\int_{C}e^{(x/2)(s-1/s)}\,s^{-\nu-1}\,\mathrm{d}s \qquad (4)

where: $C$ — a loop from $-\infty$ encircling the origin, like the Hankel contour; $e^{(x/2)(s-1/s)}$ — the generating kernel whose Laurent expansion reproduces the Bessel series; $\nu$ — arbitrary complex order. The loop lets a definition tied to integer $n$ (a Laurent coefficient) continue smoothly to complex $\nu$ . Deforming the loop into the complex plane converts $J_\nu$ into the Sommerfeld integral for the Hankel function, whose single saddle yields a pure outgoing exponential.

H_\nu^{(1)}(x)=\frac{1}{\pi i}\int_{-\infty}^{\infty+i\pi}e^{\,x\sinh w-\nu w}\,\mathrm{d}w \qquad (5)

where: $H_\nu^{(1)}$ — the first Hankel function, the outgoing-wave solution; $w$ — complex integration variable on a path from $-\infty$ to $+\infty+i\pi$ ; $x\sinh w$ — the phase whose stationary point drives the large- $x$ behaviour, isolating one saddle to give a single outgoing exponential rather than a cosine. The Legendre polynomials carry Laplace's integral — the door through which angular momentum is continued to complex degree.

P_\ell(\cos\theta)=\frac{1}{\pi}\int_0^{\pi}\big(\cos\theta+i\sin\theta\cos\phi\big)^{\ell}\,\mathrm{d}\phi \qquad (6)

where: $\ell$ — degree, which continues to complex $\lambda$ for the transform; $\theta$ — polar angle, $0\le\theta\le\pi$ ; $\phi$ — averaging variable encoding the addition theorem. Nothing requires $\ell$ to be an integer, so $P_\lambda(\cos\theta)$ is well-defined for complex $\lambda$ — the object the transform needs. Finally, the capstone: the Watson–Sommerfeld transform rewrites a partial-wave sum as a contour integral in complex angular momentum, with $1/\sin\pi\lambda$ supplying the poles that regenerate each term.

S(\theta)=\sum_{\ell=0}^{\infty}(2\ell+1)a_\ell P_\ell(\cos\theta)=\frac{i}{2}\int_{C}\frac{(2\lambda+1)\,a(\lambda)\,P_\lambda(-\cos\theta)}{\sin\pi\lambda}\,\mathrm{d}\lambda \qquad (7)

where: $a(\lambda)$ — partial-wave amplitude continued to complex angular momentum $\lambda$ ; $1/\sin\pi\lambda$ — meromorphic factor with unit-residue poles at every integer $\lambda=\ell$ ; $C$ — contour wrapping the positive real $\lambda$ -axis and enclosing all the integers; $P_\lambda(-\cos\theta)$ — Legendre function of complex degree. The contour form is identical to the sum until you deform $C$ ; only then do the Regge poles of $a(\lambda)$ enter.

Derivation AThe Hankel contour makes one over Gamma entire

The goal is to prove equation (1): that wrapping the Euler integral around its branch cut produces an entire function equal to $1/\Gamma(z)$ , with the poles of $\Gamma$ reappearing as the simple zeros of the reciprocal. We parametrise the two rays of the Hankel contour, add them with the correct branch phases, recognise the surviving integral as $\Gamma(1-z)$ , and close with the reflection formula. The sign of the phase difference is where errors hide, so we re-derive the factor $2i\sin\pi z$ from scratch.

Step 1. Split $C$ into the ray below the cut (argument $-\pi$ ), the small circle about the origin, and the ray above the cut (argument $+\pi$ ). On the lower ray put $t=u\,e^{-i\pi}$ with $u:\infty\to0$ ; on the upper ray $t=u\,e^{+i\pi}$ with $u:0\to\infty$ . On both, $e^{t}=e^{-u}$ and $|t|=u$ . Why: real positive $u$ on each ray reduces the contour to ordinary real integrals dressed only by branch phases.

Step 2. Evaluate the branch factor $t^{-z}$ . On the lower ray $t^{-z}=u^{-z}e^{+i\pi z}$ and $\mathrm{d}t=-\,\mathrm{d}u$ , giving $e^{i\pi z}\Gamma(1-z)$ ; on the upper ray $t^{-z}=u^{-z}e^{-i\pi z}$ , giving $-e^{-i\pi z}\Gamma(1-z)$ . The small circle contributes nothing for $\Re z<1$ . Why: the two rays carry opposite branch phases because the argument of $t$ jumps by $2\pi$ across the cut — that jump is the entire source of the result.

Step 3. Add the rays:

\int_C e^{t}t^{-z}\,\mathrm{d}t=\big(e^{i\pi z}-e^{-i\pi z}\big)\Gamma(1-z)=2i\sin(\pi z)\,\Gamma(1-z)

Why: the difference of the two conjugate phases is exactly $2i\sin\pi z$ , which cancels the $2\pi i$ prefactor and leaves a clean sine.

Step 4. Divide by $2\pi i$ and apply the reflection formula $\Gamma(z)\Gamma(1-z)=\pi/\sin\pi z$ :

\frac{1}{2\pi i}\int_C e^{t}t^{-z}\,\mathrm{d}t=\frac{\sin\pi z}{\pi}\,\Gamma(1-z)=\frac{1}{\Gamma(z)}

Why: the reflection formula converts $\sin\pi z\,\Gamma(1-z)$ into $\pi/\Gamma(z)$ , and the result — from an everywhere-convergent integral — is entire, with zeros at $z=0,-1,-2,\dots$ exactly where $\Gamma$ had poles.

Go deeper · Derivation B · 1 minBessel large-argument asymptotics by steepest descent

Derivation BBessel large-argument asymptotics by steepest descent

The goal is to extract the universal oscillatory tail $J_\nu(x)\sim\sqrt{2/\pi x}\,\cos(x-\nu\pi/2-\pi/4)$ from an integral representation as $x\to\infty$ , showing that the amplitude, the phase, and the mysterious $-\pi/4$ all fall out of the geometry of two saddle points. We start from Bessel's integral, locate the stationary points, expand quadratically about each, do the Gaussian integrals, and add the two conjugate contributions into a cosine. The same analysis on the Sommerfeld integral (5) keeps one saddle and gives the outgoing Hankel exponential.

Step 1. Write Bessel's integral $J_\nu(x)=\frac{1}{2\pi}\int_{-\pi}^{\pi}e^{i(x\sin\phi-\nu\phi)}\,\mathrm{d}\phi$ with phase $\Phi(\phi)=\sin\phi$ . Stationary phase demands $\Phi'(\phi)=\cos\phi=0$ , so $\phi=\pm\tfrac{\pi}{2}$ . Why: for large $x$ the integrand cancels everywhere except where the phase is stationary — only the saddles survive.

Step 2. Expand about each saddle: $\Phi(\phi)\approx\pm1-\tfrac12(\phi\mp\tfrac{\pi}{2})^2$ . Why: near a stationary point the phase is locally parabolic, making each neighbourhood a Gaussian integral of width $\sim x^{-1/2}$ whose curvature sign generates the Fresnel phase.

Step 3. Do the Gaussian integrals, $\int e^{ix(\pm1-\frac12u^2)}\,\mathrm{d}u=e^{\pm ix}\sqrt{2\pi/x}\,e^{\mp i\pi/4}$ , and attach the linear piece $-\nu\phi$ at $\phi=\pm\pi/2$ , which contributes $\mp\nu\pi/2$ :

J_\nu(x)\sim\frac{1}{2\pi}\sqrt{\frac{2\pi}{x}}\Big(e^{i(x-\nu\pi/2-\pi/4)}+e^{-i(x-\nu\pi/2-\pi/4)}\Big)

Why: each saddle contributes amplitude $\sqrt{2\pi/x}$ and a Fresnel phase $\mp\pi/4$ ; the two are complex conjugates.

Step 4. Combine the conjugate exponentials into a cosine:

J_\nu(x)\sim\sqrt{\frac{2}{\pi x}}\,\cos\!\left(x-\frac{\nu\pi}{2}-\frac{\pi}{4}\right),\qquad x\to\infty

Why: a phase plus its conjugate gives twice a cosine, fixing the $\sqrt{2/\pi x}$ amplitude and the $-\nu\pi/2-\pi/4$ shift every Bessel function shares.

Go deeper · Derivation C · 1 minWatson–Sommerfeld: from partial-wave sum to background plus Regge poles

Derivation CWatson–Sommerfeld: from partial-wave sum to background plus Regge poles

The goal is to convert a slowly convergent partial-wave sum into a contour integral, then deform the contour to expose the Regge poles that govern high-energy scattering — showing how a discrete sum hides continuous information in the complex angular-momentum plane. We verify that the integer residues reproduce the sum, push the contour leftward, and read off the asymptotic law. The one subtle point is the double sign flip from $\sin\pi\lambda$ and $P_\lambda(-\cos\theta)$ , which must cancel for the sum to be reproduced.

Step 1. The factor $\pi/\sin\pi\lambda$ has simple poles at every integer $\lambda=\ell$ with residue $(-1)^\ell$ . The Legendre function carries a compensating $(-1)^\ell$ when $\lambda\to\ell$ , since $P_\ell(-\cos\theta)=(-1)^\ell P_\ell(\cos\theta)$ . Why: these two $(-1)^\ell$ factors must multiply to $+1$ so each residue is a genuine partial wave, not an alternating-sign impostor.

Step 2. The residue at $\lambda=\ell$ of the contour integrand is:

\mathrm{Res}_{\lambda=\ell}\frac{(2\lambda+1)\,a(\lambda)\,P_\lambda(-\cos\theta)}{\sin\pi\lambda}=\frac{(2\ell+1)\,a_\ell\,P_\ell(\cos\theta)}{\pi}\,(-1)^\ell(-1)^\ell

Why: the sign factors cancel and the $1/\pi$ is absorbed by the $i/2$ and $2\pi i$ of the residue theorem, returning exactly the summand $(2\ell+1)a_\ell P_\ell(\cos\theta)$ .

Step 3. With $C$ wrapping the non-negative integers, summing all residues rebuilds the sum, establishing equation (7). Now deform $C$ leftward to the symmetry line $\Re\lambda=-\tfrac12$ , where $P_\lambda=P_{-\lambda-1}$ . As it sweeps, the contour crosses the Regge poles $\lambda=\alpha_n(E)$ of the continued amplitude, $a(\lambda)\approx\beta_n/(\lambda-\alpha_n)$ :

S(\theta)=\underbrace{\frac{i}{2}\int_{-1/2-i\infty}^{-1/2+i\infty}\frac{(2\lambda+1)a(\lambda)P_\lambda(-\cos\theta)}{\sin\pi\lambda}\,\mathrm{d}\lambda}_{\text{background}}-\pi\sum_{n}\frac{(2\alpha_n+1)\,\beta_n\,P_{\alpha_n}(-\cos\theta)}{\sin\pi\alpha_n}

Why: the residue theorem makes the difference between the two contours equal $2\pi i$ times the enclosed pole residues — the Regge terms — while the background integral stays bounded.

Step 4. Read the asymptotics. Using $P_{\alpha_n}(-\cos\theta)\sim(\cos\theta)^{\alpha_n}$ as $\cos\theta\to\infty$ , the pole with the largest $\Re\alpha_n$ dominates, giving the amplitude the growth $s^{\alpha(t)}$ in the crossed channel. Why: a single Regge trajectory $\alpha(t)$ , continued in energy, controls the high-energy angular dependence — the step that turned a summation trick into Regge theory.

Limiting cases

Each master integral degenerates into a familiar special case in a controlled limit — the fastest way to check a representation is correct.

Case	Result	Meaning
Mellin–Barnes, small argument	$\dfrac{1}{2\pi i}\displaystyle\int_{c-i\infty}^{c+i\infty}\Gamma(s)\Gamma(a-s)\,x^{-s}\,\mathrm{d}s=\dfrac{\Gamma(a)}{(1+x)^{a}}$	Closing left over the poles of $\Gamma(s)$ at $s=-n$ sums the convergent small- $x$ binomial series; closing right gives the large- $x$ expansion.
Bessel, large argument	$J_\nu(x)\sim\sqrt{\tfrac{2}{\pi x}}\,\cos\!\big(x-\tfrac{\nu\pi}{2}-\tfrac{\pi}{4}\big)$	Two real saddles of the Sommerfeld/Bessel integral combine into a cosine as $x\to\infty$ .
Half-integer order	$J_{1/2}(x)=\sqrt{\tfrac{2}{\pi x}}\,\sin x$	The Poisson integral with $\nu=\tfrac12$ has weight one and reduces to an elementary trig integral — the spherical Bessel function $j_0$ .
Hankel, one saddle	$H_\nu^{(1)}(x)\sim\sqrt{\tfrac{2}{\pi x}}\,e^{i(x-\nu\pi/2-\pi/4)}$	The Sommerfeld path keeps a single saddle, giving a pure outgoing wave rather than a standing cosine.
Watson–Sommerfeld, integer residues	$\displaystyle\sum_{\ell}(2\ell+1)a_\ell P_\ell(\cos\theta)=\frac{i}{2}\oint\frac{(2\lambda+1)a(\lambda)P_\lambda(-\cos\theta)}{\sin\pi\lambda}\,\mathrm{d}\lambda$	Residues of $1/\sin\pi\lambda$ at $\lambda=\ell$ reproduce the sum; deforming exposes the Regge-pole asymptotics.

Common trapDeforming a contour past a pole or a cut without accounting for it

Contour deformation is legitimate only where the integrand is analytic and the connecting arcs at infinity vanish. Sweep a Mellin–Barnes contour past a pole and you silently change the integral by $2\pi i$ times that residue — dropping or adding a whole series term or asymptotic contribution. The most common failure is choosing the abscissa $c$ so the vertical line does not separate the left pole ladder from the right; then neither closure is valid and both series are wrong. Equally fatal is ignoring a branch cut, or assuming an arc decays when the kernel grows on part of it. The rule is mechanical: confirm analyticity in the swept region, confirm the arc vanishes, and re-add every crossed pole as an explicit residue.

Worked examples

1 · A Regge trajectory from the meson spectrum (fully worked)

The ρ family has states ρ(770) with spin $J=1$ and $a_4$ (2040) with $J=4$ . Treating spin as the continued angular momentum $\alpha$ and mass-squared as $t=m^2$ , find the trajectory $\alpha(t)=\alpha_0+\alpha' t$ , predict the next recurrence $J=5$ , and compare with the observed ρ₅(2350).

Step 1. Convert masses to $m^2$ in GeV². For ρ(770), $m^2=0.770^2=\mathbf{0.593}$ ; for $a_4$ (2040), $m^2=2.040^2=\mathbf{4.162}$ . These are the anchor points $(0.593,\,1)$ and $(4.162,\,4)$ .

Step 2. The slope is the rise in spin over the rise in $m^2$ ,

\alpha'=\frac{4-1}{4.162-0.593}=\frac{3}{3.569}=\mathbf{0.841\ \text{GeV}^{-2}}.

This is the universal Regge slope, near $0.9$ GeV⁻² for essentially every meson family — a regularity string models later explained.

Step 3. The intercept follows from either point: $\alpha_0=1-\alpha'(0.593)=1-0.841\times0.593=\mathbf{0.50}$ . So $\alpha(t)\approx0.50+0.84\,t$ , the line drawn through the ρ family in the second figure — and $a_2$ (1320) and $ρ_3$ (1690) sit on it to within a percent.

Step 4. Predict $J=5$ : $t=(5-0.50)/0.841=\mathbf{5.35}$ GeV², so $m=\sqrt{5.35}=\mathbf{2.31\ \text{GeV}}$ . The observed ρ₅(2350) has $m=2.35$ GeV — agreement to about $2\%$ from a line fixed by only two states. The trajectory $\alpha(t)\approx0.50+0.84\,t$ predicts ρ₅ at 2.31 GeV against the measured 2.35 GeV. The discrete spins are samples of one analytic function $\alpha(t)$ — the Watson–Sommerfeld continuation made visible.

2 · Accuracy of the Bessel asymptotic (fully worked)

Estimate $J_0(10)$ from the leading steepest-descent asymptotic and compare with the exact value $J_0(10)=-0.24594$ .

Step 1. Take the boxed result with $\nu=0$ : $J_0(x)\sim\sqrt{2/\pi x}\,\cos(x-\pi/4)$ . The amplitude at $x=10$ is $\sqrt{2/(10\pi)}=\mathbf{0.25231}$ .

Step 2. The phase is $x-\pi/4=9.21460$ rad; reduced modulo $2\pi$ that is $\mathbf{2.93142}$ rad, and $\cos(2.93142)=\mathbf{-0.97800}$ .

Step 3. Multiply: $J_0(10)\approx0.25231\times(-0.97800)=\mathbf{-0.24676}$ . Against the exact $-0.24594$ this is an error of $\mathbf{0.34\%}$ — from the leading term alone, at only $x=10$ . The one-saddle-pair formula already nails $J_0(10)$ to three significant figures.

Step 4. Read the lesson. The error scales as the first neglected correction, $\sim1/(8x)\approx1.3\%$ here; the actual $0.34\%$ is better still because that correction enters the phase, not the amplitude. The steepest-descent series is asymptotic, not convergent, but its first term is astonishingly accurate once $x$ is moderately large.

Physics of the portrait

Integral representations are where nineteenth-century analysis and twentieth-century physics fuse, and no figure sits closer to that seam than the mathematician the transformation is named for.

Origins · 1 minGeorge Neville Watson and how this was discovered

George Neville Watson

1886–1965 · the Bessel treatise and the Watson transformation

George Neville Watson was an English mathematician who did more than anyone to make the special functions usable. With E. T. Whittaker he wrote A Course of Modern Analysis — the "Whittaker and Watson" that generations learned contour integration from — and alone he produced A Treatise on the Theory of Bessel Functions in 1922, still the definitive reference, in which almost every integral representation here appears cross-checked and generalised. His most consequential stroke came in 1918, on a concrete problem: the diffraction of radio waves around the curved Earth. The normal-mode sum there converged too slowly to use, so Watson converted it into a contour integral in complex angular momentum via the factor $\pi/\sin\pi\lambda$ , then deformed the contour to isolate the few modes that mattered. That "Watson transformation" lay dormant for four decades until Tullio Regge, in 1959, continued the angular momentum of a Schrödinger amplitude into the complex plane and found its poles moving along smooth trajectories with energy — Watson's summation trick had become the analytic backbone of high-energy scattering. He was a superb calculator who distrusted hand-waving, and it is exactly his rigour that lets modern physicists deform contours with confidence.

History

1864

— Hankel writes 1/Γ(z) as a contour integral wrapping the negative real axis

1873

— Schläfli's loop integral defines Jν for arbitrary complex order

1908

— Barnes and Mellin develop the Mellin–Barnes integral for hypergeometric functions

1918

— Watson converts a slowly convergent partial-wave sum into a contour integral in angular momentum

1959

— Regge continues angular momentum to the complex plane and discovers Regge poles

Going further

Texts. Stone & Goldbart, Mathematics for Physics, Ch. 18 · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Chs. 11–14 · Hassani, Mathematical Physics, 2nd ed., Chs. 11–12 · Byron & Fuller, Mathematics of Classical and Quantum Physics, Chs. 6–8 · Whittaker & Watson, A Course of Modern Analysis, Chs. 12–17 · Watson, A Treatise on the Theory of Bessel Functions, 2nd ed. · Regge, Nuovo Cimento 14 (1959) 951.

Provenance. Every formula is standard and was re-verified. The Hankel representation and reflection-formula reduction follow Whittaker & Watson §12.22 and Arfken Ch. 13 (the factor $2i\sin\pi z$ was re-derived ray by ray, fixing an easily-dropped sign). The Mellin–Barnes Beta evaluation $\Gamma(a)/(1+x)^a$ is Whittaker & Watson §14.5, checked at $a=2,\,x=0.5$ . The Bessel integrals and steepest-descent asymptotic are Watson §§6–7 and Abramowitz & Stegun 9.2.1; the checks $J_0(10)\approx-0.24676$ vs exact $-0.24594$ (0.34%) and $J_{1/2}(\pi/2)=2/\pi$ were computed directly. The Laplace integral for $P_\ell$ is Arfken Ch. 15. The Watson–Sommerfeld transform follows Regge (1959) and Whittaker & Watson §17; the ρ-family trajectory $\alpha(t)\approx0.50+0.84\,t$ was fitted to the masses of ρ(770), a₂(1320), ρ₃(1690), a₄(2040).

Builds on: Gamma Function · Analytic Continuation and Poles of the Gamma Function · Beta Function · Bessel Functions of the First and Second Kind. Leads to: Hankel Functions and Asymptotic Behaviour of Bessel Functions · Modified Bessel Functions · Legendre Polynomials · Recurrence Relations for Special Functions.

Ask the tutor: Ask me to deform a specific contour with you — pick the Gamma, Bessel, or partial-wave integral — and we will track every pole and saddle until the series and the asymptotics fall out of the same picture.

What actually makes an integral representation more powerful than the series it replaces?

A power series is defined only inside its circle of convergence, whose radius is fixed by the nearest singularity. An integral representation replaces it by $\int_C K(z,t)\,\mathrm{d}t$ with a single-valued integrand that often converges on a vastly larger domain. Wherever it converges it is analytic in $z$ , so it is the analytic continuation, produced automatically rather than by patching series. And the same integral reads two ways: residues at integer poles give the series back, saddle points give the asymptotics. One object, every regime.

How do I know which way to close a Mellin–Barnes contour?

Look at $x^{-s}=e^{-s\ln x}$ , the only factor that grows or decays as $\Re s\to\pm\infty$ . If $|x|<1$ then $\ln x<0$ and the integrand decays as $\Re s\to-\infty$ : close left, enclose the increasing pole ladder, and sum it into the convergent small- $x$ series. If $|x|>1$ , close right for the large- $x$ expansion. Two cautions: the vertical contour must already separate the two pole families, or neither closure is legitimate; and the added arc must vanish — guaranteed by the Gamma functions' Stirling decay along vertical lines, but only if the closure direction matches the sign of $\ln x$ .

Where do Regge poles physically come from?

They are poles of the scattering amplitude continued in angular momentum. At fixed energy the partial-wave amplitude $a_\ell$ is defined only at integer $\ell$ , but the potential problem admits a smooth continuation $a(\lambda)$ to complex $\lambda$ with poles at $\lambda=\alpha_n(E)$ . At integer $\lambda$ near a real $\alpha_n$ you see a resonance or bound state; as energy varies the pole traces a smooth curve $\alpha_n(E)$ — the Regge trajectory. The physical states of a family (like the ρ mesons) are where that trajectory crosses integer spin. So a bound state, a resonance, and the high-energy growth law are three views of one moving pole.

Why does the same steepest-descent calculation give a cosine for Jν but a single exponential for the Hankel function?

It comes down to how many saddle points the contour must pass through. Bessel's integral for $J_\nu$ straddles both stationary points $\phi=\pm\pi/2$ , so the result sums two complex-conjugate contributions into a real cosine — the standing wave. The Sommerfeld contour for $H_\nu^{(1)}$ passes through only one saddle, so a single exponential $e^{i(x-\nu\pi/2-\pi/4)}$ survives — the outgoing wave. Since $J_\nu=\tfrac12(H_\nu^{(1)}+H_\nu^{(2)})$ , adding the two Hankel exponentials reproduces the cosine, consistent with the two-saddle calculation.

Is the minus pi over four in the Bessel phase specific to Bessel functions?

No — it is universal to this class of oscillatory integrals, and it counts geometry, not physics. Each quadratic saddle of a stationary-phase integral contributes $e^{\pm i\pi/4}$ , the phase of the Fresnel integral $\int e^{\pm iu^2}\mathrm{d}u=\sqrt{\pi}\,e^{\pm i\pi/4}$ , with the sign set by the curvature at the saddle. Bessel, Airy, and the WKB connection formulae all inherit such fractions. In semiclassical language these are Maslov indices: each turning point a trajectory touches adds a fixed $-\pi/2$ to the phase.

Do these contour methods actually get used in computation, or are they just for proofs?

Very much used. Mellin–Barnes representations are a mainstream tool for multi-loop Feynman integrals: write each propagator as an MB integral, exchange orders, and sum residues to produce the Laurent expansion in the regularisation parameter — automated packages do exactly this. Steepest descent underlies fast asymptotic evaluation of special functions in numerical libraries, and the Watson transformation gives the "creeping-wave" diffraction fields of antenna engineering, where the raw modal sum is hopelessly slow. The contour is not decoration — it is the algorithm.