Residue theorem

The formula

The core statement is a loop integral collapsing to a sum ^[1]:

∮_C f(z) dz = 2π·i · Σ Res(f, z_k)

Reading: walk once counterclockwise around a closed curve C, integrate f(z) as you go, and the answer is 2π·i times the sum of the residues at the poles z_k that sit inside the loop. Everything about the function away from its poles washes out. Only the singularities you enclose pay in.

A residue is the coefficient of the 1/(z - z_k) term in f's Laurent expansion around the pole. You almost never expand by hand. For a simple pole (order 1):

Res(f, c) = lim(z→c) (z - c)·f(z)

Reading: cancel the pole by multiplying it out, then plug in. For a pole of order p, you differentiate first:

Res(f, c) = (1/(p-1)!) · lim(z→c) d^(p-1)/dz^(p-1) [ (z - c)^p · f(z) ]

Reading: clear the pole to power p, differentiate p−1 times, take the limit, divide by the factorial. Order-1 is the p = 1 case of this — no derivative.

To turn a real integral over the whole line into a loop, you close the contour with a big semicircle and need its contribution to vanish. That is Jordan's lemma, the companion result ^[4]:

| ∫_arc  e^(i·a·z)·g(z) dz |  ≤  (π/a)·M_R ,   a > 0

Reading: for an oscillatory factor e^(i·a·z) with a > 0, the integral over the upper semicircle of radius R is bounded by (π/a) times the peak of |g| on that arc. If g dies off as R grows, the arc contributes zero, and the real-line integral equals 2π·i times the enclosed residues alone.

Where you meet it

You meet residues any time an answer lives in the frequency domain and you need it back in time, or any time a "compute this integral" problem has no elementary antiderivative.

Inverse Laplace on a control review. The Bromwich integral that inverts H(s) is a vertical contour in the s-plane. Close it to the left, and the time response is the sum of residues at the poles of H(s)·e^(s·t). Every pole hands you one mode: a real pole gives a decaying exponential, a complex pair gives a damped sinusoid. That is why a pole map on the review board screen is a time-response preview.
Inverse Fourier and filter impulse response, on the bench. Getting h(t) from H(jω) means integrating H(jω)·e^(jωt) over all ω. Jordan's lemma closes the contour in whichever half-plane makes e^(jωt) decay — upper for t > 0, lower for t < 0 — and the impulse response falls out of the enclosed poles. Causality (h(t) = 0 for t < 0) drops straight out of which poles you enclose.
RF noise and PSD integrals, on a link budget. Integrated noise power through a filter shape, or an autocorrelation from a spectrum, is often a rational-times-oscillatory integral over all frequency. Residues turn it into a finite sum instead of a numerical sweep.
Antenna and diffraction work. Radiation integrals and Sommerfeld-type integrals for fields over a ground plane get evaluated by deforming the contour and collecting the pole (guided-wave) and branch-cut (radiation) contributions separately.

How it works

The whole method rests on one fact from Cauchy: integrate an analytic function around a closed loop and you get exactly zero, unless the loop wraps a singularity. Poles are the only places a closed contour can accumulate a nonzero value, and each one contributes its residue, no more. So a nasty integral becomes bookkeeping: find the poles inside the loop, compute a residue at each, add them up, multiply by 2π·i.

The trick that makes it useful for real integrals is closing the contour. Take the standard case:

∫_{-∞}^{∞} 1/(x² + 1) dx = π

The integrand has poles at z = ±i. Close the real line with a large semicircle in the upper half-plane; only z = i is enclosed. Its residue is lim(z→i)(z - i)/((z-i)(z+i)) = 1/(2i). Multiply by 2π·i and the arc vanishes, leaving 2π·i · 1/(2i) = π. Verified numerically against direct quadrature to full double precision. The same contour handles the Fourier version:

∫_{-∞}^{∞} e^(i·t·x)/(x² + 1) dx = π·e^(-|t|)

Here Jordan's lemma earns its keep: e^(i·t·x) decays in the upper half-plane for t > 0, so the arc dies and only the pole at z = i counts. Checked at t = 2 against numerical integration — π·e^(-2) ≈ 0.42517, agreeing to four digits (the small residual is quadrature error on the oscillatory tail, not the formula). A quartic denominator works the same way:

∫_{-∞}^{∞} 1/(x⁴ + 1) dx = π/√2 ≈ 2.22144

confirmed to full precision. And angular integrals over 0 to 2π convert by the substitution z = e^(i·θ), turning the interval into the unit circle:

∫_0^{2π} 1/(2 + cos θ) dθ = 2π/√3 ≈ 3.62760

also verified numerically. Four different-looking integrals, one method.

The gotchas are where people lose the answer:

You must enclose the right poles, and only those. Close the contour in the wrong half-plane and you either enclose nothing or the wrong set. For e^(i·a·z), a > 0 forces the upper half-plane; a < 0 forces the lower. Get the sign wrong and Jordan's lemma no longer kills the arc — the integral diverges instead of vanishing.
The arc has to actually vanish. Jordan's lemma needs the non-oscillatory factor g(z) to go to zero on the arc as R grows. For a plain rational function with no e^(i·a·z), the milder condition is that the integrand decays faster than 1/|z| — denominator degree at least two above the numerator. 1/(x²+1) qualifies; x/(x²+1) does not, and forcing it gives a wrong finite answer for a divergent integral.
Poles on the contour need a detour. A pole sitting on the real axis (common in sin(x)/x type integrals) is not inside or outside — you indent the contour with a small semicircle around it, which contributes π·i times the residue, half of a full loop. Skip the indentation and you drop a real term. This is the principal-value case.
Order matters for the residue formula. Use the simple-pole formula on a double pole and you get infinity or zero, both wrong. Count the multiplicity first, then apply the differentiate-p−1-times version.
Essential singularities are not poles. A residue is still defined (the 1/(z-c) Laurent coefficient), but the clean limit formulas do not apply — you expand the series. e^(1/z) at the origin is the classic example; no finite-order formula reaches it.

The limit of validity: the theorem needs f analytic on and inside C except at isolated singularities, and the contour must be a genuine closed loop that does not pass through a pole. Branch points (from roots or logs) are not isolated singularities — they carry cuts, and those cuts contribute their own integrals that you cannot ignore. That is the honest boundary of the shortcut.

History

The method is Cauchy's, and it grew out of his campaign through the 1820s to put complex integration on a firm footing. Augustin-Louis Cauchy (1789–1857) had stated his integral theorem — that a loop integral of an analytic function is zero — in full form in his 1825 memoir on definite integrals taken between imaginary limits ^[2]. The residue idea itself he pushed forward in 1826, in a paper with the mouthful title Sur un nouveau genre de calcul analogue au calcul infinitésimal, where he gave the residue its formal definition and began the calculus of residues as a systematic tool ^[2]^[3]. The name is apt: a residue is what remains of a loop integral after the analytic part cancels, leaving only the singular contribution. The residue theorem in its recognizable form Cauchy presented to the Academy of Turin in a memoir of 27 November 1831 — a companion memoir presented six weeks earlier, on 11 October, carried the integral formula that bears his name — and the Turin work reached print only a decade later, in the years of his Exercices d'analyse et de physique mathématique (1840–1847) ^[2]. Cauchy spent much of the following decade generalizing it and applying it to real integrals that had resisted every earlier method.

The contour-closing companion came two generations later. Marie Ennemond Camille Jordan (1838–1922), the French mathematician better remembered for group theory and the Jordan curve theorem, laid out the arc-vanishing bound that now carries his name ^[4] in his Cours d'analyse de l'École polytechnique, the analysis textbook first published in three volumes between 1882 and 1887 ^[6]; the lemma sits in the integral-calculus volume, Tome II ^[5]. That bound — the guarantee that the big semicircle contributes nothing when an oscillatory factor is present — is what lets Cauchy's closed-loop theorem attack open real-line integrals from Fourier and Laplace work. Cauchy supplied the machine; Jordan supplied the standard way to feed real integrals into it, which is why an RF or controls engineer meets both names on the same page.

Related tools

/tools/series-rlc-impedance
/tools/rc-filter
/tools/lc-resonance
/tools/frequency-period
/tools/vibration-natural-freq

Sources

Written by HE in our own words from the cited sources — engineering judgment included, your stamp still required. All entries →