The π/2 Heist

The function sin x / x looks broken at x = 0, but it is not: as x → 0, sin x / x → 1. It starts at height 1, then makes positive lobes and negative lobes that shrink forever.

The question: add up all the signed area, out to infinity. Drag the slider. The running total overshoots, undershoots, and squeezes toward a mystery number.

Try the second toggle too: counting all area as positive, the total never settles. It grows like 1 + ½ + ⅓ + ⋯, past every bound. The answer only exists because the lobes cancel. Hold that thought, it is why the proof must be careful.

The normal plan: find an antiderivative F, compute F(∞) − F(0). Watch each standard tool bounce off, then see why they all had to.

Define one new number: i, with i² = −1. A complex number a + b i is simply the point (a, b) in a plane. Nothing mystical: these numbers obey the usual algebra, and physics and engineering run on them daily.

Adding complex numbers is adding arrows, tip to tail. Multiplying is the beautiful part: multiplying by a number on the unit circle rotates the whole plane. Multiplying by i itself is exactly a quarter turn.

Negative numbers were once called absurd. Then they became bookkeeping for debt. i is the same story one century later: bookkeeping for rotation.

e^iθ is the point on the unit circle at angle θ. Where does that come from? Feed iθ into the power series for e^x, and the terms sort themselves: the even ones assemble cosine, the odd ones assemble i times sine. Exponentials and rotation are the same machine.

In the plane, an integral follows a path. Chop the path into tiny steps. Each step dz is itself a small complex number (a small arrow). At each spot multiply the function’s value f(z) by the step dz, then add everything up.

This is not area. It is a running total of complex products, and the result is a single complex number: where the chain of little arrows ends.

Watch f(z) = 1/z walk the top half of the unit circle. The chain of contributions lands at a number worth memorizing.

Intuition (offered as intuition, the proof lives in Chapter 2 of the book): such functions behave like conservative fields, and a round trip in a conservative world nets you nothing.

The power is the fine print. One bad point inside the loop, a singularity, and the zero guarantee breaks, in a precisely measurable way. Slide the loop and watch the integral snap between two values. Closed loops are singularity detectors.

We need a function that agrees with our problem on the real axis but behaves upstairs. Two auditions:

sin z / z fails the audition. Off the real axis, sine explodes: at height y, |sin(iy)| = sinh y ≈ e^y/2. At height 9 that is already about 4000.

e^iz/z gets the part. Upstairs it decays: |e^iz| = e^−y. And on the real axis, Im(e^ix/x) = sin x/x: exactly our integrand, waiting in the imaginary part.

The price: e^iz/z has a genuine pole at z = 0. The original sin x/x was innocent there. We installed the trap ourselves, on purpose, and next we route around it.

Build a closed loop out of four pieces, walked counterclockwise. It hugs the real axis (where our problem lives) but hops over the trap at 0, and closes through the upper half plane (where our hero decays).

Each piece has a destiny:

the two straight legs  become the target integral.
the small dodge C_ε  pays a fixed toll.
the big arc C_R  fades to nothing.

Because the loop never touches the pole, and e^iz/z is differentiable everywhere else, Cauchy’s theorem applies: the four pieces sum to exactly zero, for every R and every ε. That equation is the safe we are about to crack.

On the big arc, z = Re^iθ, and the hero’s magnitude is |e^iz| = e^{−R sin θ}: exponentially tiny at altitude, merely ordinary near the two ends.

The naive estimate fails. Bounding (size of function) × (length of path) gives (1/R) × (πR) = π. A constant. Not zero. The crude bound throws away the exponential decay.

The rescue is Jordan’s lemma (Lemma 4.2.2 in the book, page 222): accounting honestly for the decay, the arc integral is at most

Its proof is one clean inequality: on [0, π/2], the sine curve stays above its chord, sin θ ≥ 2θ/π. Full proof in the appendix.

Surely a vanishingly small detour contributes vanishingly little? No. The path shrinks like ε, but 1/z grows like 1/ε. The two effects cancel exactly, leaving a fixed toll.

Near 0, split the hero: e^iz/z = 1/z + (a part that stays bounded). The bounded part contributes at most (bound) × (πε) → 0. The 1/z part, walked clockwise over the top from angle π down to 0, gives exactly

Recognize it? It is the pocketed iπ from the walking slide, with a minus sign because we now walk the half-circle the other way. Orientation is not pedantry, it is the sign of the final answer.

The two straight legs live on the real axis, where the hero is e^ix/x. Pair each point x on the right leg with its mirror −x on the left leg.

Cauchy’s theorem said the four pieces always sum to zero. Send R → ∞ and ε → 0, and read the ledger:

Why no antiderivative exists

Liouville’s theory of integration in finite terms: if ∫f e^g is elementary with rational f, g, the antiderivative must equal R e^g for some rational R, forcing R′ + ig ′R = f. For e^ix/x this reads R′ + iR = 1/x, and a pole-order count shows no rational R works: any pole of R of order n makes R′ + iR have one of order n+1 ≥ 2, but 1/x has only a simple pole. Hence Si(x) is not elementary.

Convergence status

Converges: near 0 the integrand extends continuously (limit 1). At infinity, integrate by parts: ∫^R₁ sin x/x dx = [−cos x/x]^R₁ − ∫^R₁ cos x/x² dx, and both pieces settle (the last is dominated by 1/x²).

Not absolutely: on the k-th arch [kπ, (k+1)π], the arch area of |sin| is 2 and x ≤ (k+1)π, so ∫|sin x|/x dx ≥ Σ 2/((k+1)π), a harmonic series. Divergent. All convergence is cancellation, so no dominated-convergence shortcuts.

The contour and the zero

Fix 0 < ε < R and set f(z) = e^iz/z. The closed contour Γ_R,ε, walked counterclockwise overall:

L₋: segment from −R to −ε.  C_ε: half-circle of radius ε over the origin, angle π down to 0 (clockwise).  L₊: segment from ε to R.  C_R: half-circle of radius R through the upper half plane, angle 0 up to π (counterclockwise).

f is analytic on ℂ ∖ {0}, and Γ_R,ε with its interior avoids 0 by construction. Cauchy’s integral theorem therefore gives, for every ε and R:

Proof

Parametrize z = Re^iθ, so |dz| = R dθ and |e^iaz| = e^{−aR sin θ}. Then the integral is bounded by M_R R ∫^π₀e^{−aR sin θ}dθ = 2M_R R ∫^π/2₀e^{−aR sin θ}dθ by the symmetry sin(π − θ) = sin θ.

Jordan’s inequality (concavity of sine): sin θ ≥ 2θ/π on [0, π/2]. Substitute:

2M_RR ∫ e^−2aRθ/π dθ = 2M_RR · (π/2aR)(1 − e^−aR) = (π/a) M_R (1 − e^−aR). ∎

Application to the big arc

Here a = 1 and g(z) = 1/z, so M_R = 1/R and

Why the crude ML bound cannot substitute: sup|f| · length = (1/R)(πR) = π, a constant. The exponential decay e^{−R sin θ} off the axis is the load-bearing fact, and Jordan’s chord inequality is the standard clean way to cash it in (any argument exploiting that decay also works).

Numerical spot check: at R = 100, the true ∫^π₀e^{−R sin θ}dθ ≈ 0.0200, against the bound π(1 − e⁻¹⁰⁰)/100 ≈ 0.0314. The bound holds with room.

Splitting off the pole

From the power series of the exponential, e^iz/z = 1/z + g(z) where g(z) = (e^iz − 1)/z extends to an entire function with g(0) = i. On the closed unit disk let M = max |g| (finite, by continuity on a compact set). For ε ≤ 1:

|∫_Cε g(z) dz| ≤ M · πε → 0.

The exact toll from 1/z

Parametrize z = εe^iθ with θ from π down to 0 (clockwise over the top). Then dz = iεe^iθdθ and

∫_Cε dz/z = ∫_π⁰ i dθ = −iπ, for every ε.

So ∫_Cε e^iz/z dz → −iπ as ε → 0 (the full integral at fixed ε is −iπ + O(ε); it is the limit that is exact).

The half-residue shortcut, and its warranty

The residue of e^iz/z at 0 is 1, so a full counterclockwise loop would collect 2πi. Our clockwise half-loop collected −½ · 2πi = −iπ. That proportional shortcut is guaranteed for simple poles only: for a pole of order 2 or higher it can fail outright (a c₋₂/z² term contributes on the order of 1/ε on the shrinking arc, which diverges), so at higher order the arc limit must be examined term by term rather than assumed.

Consistency check from below

Indent under the origin instead and the pole falls inside the contour: the total becomes 2πi (residue theorem) while the arc now contributes +iπ. The books balance to the same principal value, +iπ for the legs. Two different contours, one answer, as it must be.

Assembly

Take ε → 0 and R → ∞ in the Cauchy identity. Legs → 2i ∫^∞₀sin x/x dx (shown on slide 12), dodge → −iπ, big arc → 0. Hence 2iI − iπ = 0, so I = π/2. Equivalently, in principal-value form, PV∫^∞_−∞e^ix/x dx = iπ.

Why "principal value" appears at all

Split e^ix/x = cos x/x + i sin x/x. The cosine part behaves like 1/x near 0 and is not integrable there: it only makes sense with symmetric cutoffs ±ε (the PV), where its odd symmetry cancels it. The sine part is continuous at 0 (limit 1), so the imaginary part is a genuine improper integral, no crutch needed. The PV is scaffolding for the part we discard, not for the answer we keep.

Why this problem

The genuinely hardest material in Ablowitz & Fokas is the Chapter 7 Riemann–Hilbert theory (with the Wiener–Hopf method, the DBAR problem, and Painlevé connections), which the authors themselves place at advanced graduate level and which cannot be delivered to a calculus audience in 30 minutes. This talk instead takes the hardest teachable summit: the Dirichlet integral, worked in the book’s principal-value section via the indented contour, exercising the central theorems of Chapter 4 (Cauchy’s theorem, Jordan’s lemma, principal-value indentation).

Every mathematical step in this deck was independently re-derived and adversarially checked before publication. One caveat is flagged honestly below.